1: \begin{abstract}

2: 

3: Max-product belief propagation is a local, iterative algorithm to find

4: the mode/MAP estimate of a probability distribution. While it has been

5: successfully employed in a wide variety of applications, there are

6: relatively few theoretical guarantees of convergence and correctness

7: for general loopy graphs that may have many short cycles. Of these,

8: even fewer provide exact ``necessary and sufficient''

9: characterizations.

10: 

11: In this paper we investigate the problem of using max-product to find the

12: maximum weight matching in an arbitrary graph with edge weights. This

13: is done by first constructing a probability distribution whose mode

14: corresponds to the optimal matching, and then running max-product.

15: Weighted matching can also be posed as an integer program, for which

16: there is an LP relaxation. This relaxation is not always tight. In

17: this paper we show that 

18: \begin{enumerate}

19: \item If the LP relaxation is tight, then max-product always

20:   converges, and that too to the correct answer.

21: \item If the LP relaxation is loose, then max-product does not

22:   converge.

23: \end{enumerate}

24: This provides an exact, data-dependent characterization of max-product

25: performance, and a precise connection to LP relaxation, which is a

26: well-studied optimization technique. Also, since LP relaxation is

27: known to be tight for bipartite graphs, our results generalize other

28: recent results on using max-product to find weighted matchings in

29: bipartite graphs.

30: \end{abstract}

31: