1: \begin{abstract}

2: 	Gaussian belief propagation (GBP) is a recursive computation method that is widely used in inference for computing marginal distributions efficiently.

3: 	Depending on how the factorization of the underlying joint Gaussian distribution is performed, GBP may exhibit different convergence properties as different factorizations may lead to fundamentally different recursive update structures.

4: 	In this paper,  we study the convergence of GBP derived from the factorization based on the distributed linear Gaussian model.

5: 	The motivation is twofold. From the factorization

6: 	viewpoint, i.e., by specifically employing a factorization based on the linear Gaussian model, in some cases, we are able to bypass difficulties that exist in other

7: 	convergence analysis methods that use a different (Gaussian Markov random field) factorization.

8: 	From the distributed inference viewpoint, the linear Gaussian model

9: 	readily conforms to the physical network topology

10: 	arising in large-scale networks, and, is practically useful.

11: 	For the distributed linear Gaussian model, under mild assumptions,

12: 	we show analytically three main results: the GBP message inverse variance converges exponentially fast to a unique positive  limit  for arbitrary nonnegative initialization;

13: 	we  provide a necessary and sufficient convergence condition for the belief mean  to converge to the optimal value;

14: 	and, when the underlying factor graph is given by the union of a forest and a single loop, we show that  GBP always converges.

15: \end{abstract}