1: \begin{abstract}

2: We study the performance of different message passing algorithms in

3: the two dimensional Edwards Anderson model. We show that the standard

4: Belief Propagation (BP) algorithm converges only at high temperature

5: to a paramagnetic solution. Then, we test a Generalized Belief

6: Propagation (GBP) algorithm, derived from a Cluster Variational Method

7: (CVM) at the plaquette level. We compare its performance with BP and

8: with other algorithms derived under the same approximation: Double

9: Loop (DL) and a two-ways message passing algorithm (HAK). The

10: plaquette-CVM approximation improves BP in at least three ways: the

11: quality of the paramagnetic solution at high temperatures, a better

12: estimate (lower) for the critical temperature, and the fact that the

13: GBP message passing algorithm converges also to non paramagnetic

14: solutions. The lack of convergence of the standard GBP message passing

15: algorithm at low temperatures seems to be related to the

16: implementation details and not to the appearance of long range

17: order. In fact, we prove that a gauge invariance of the constrained

18: CVM free energy can be exploited to derive a new message passing

19: algorithm which converges at even lower temperatures. In all its

20: region of convergence this new algorithm is faster than HAK and DL by

21: some orders of magnitude.

22: \end{abstract}

23: