d9c973cf828817c0.tex
1: \begin{abstract}
2: 
3: In this paper we introduce a new notion of convergence of
4: sparse graphs which we call Large Deviations or LD-convergence
5: and which is based on the theory of  large deviations. The
6: notion is introduced by ''decorating'' the nodes of the graph
7: with random uniform i.i.d. weights and constructing random
8: measures on $[0,1]$ and $[0,1]^2$ based on the decoration of
9: nodes and edges. A graph sequence is defined to be converging
10: if the corresponding sequence of random measures satisfies the
11: Large Deviations Principle with respect to the topology of weak
12: convergence on bounded measures on $[0,1]^d, d=1,2$. We then establish that
13: LD-convergence implies several previous notions of convergence,
14: namely so-called right-convergence, left-convergence, and
15: partition-convergence. The corresponding large deviation rate
16: function can be interpreted as the limit object of the sparse
17: graph sequence. In particular, we can express the limiting free
18: energies in terms of this limit object.
19: 
20: 
21: %We conjecture that the LD-convergence implies the
22: %colored-neighborhood-convergence, and we conjecture that the converse
23: %is not true. Finally,  we conjecture that the sequence of random regular
24: %graphs (or in general sparse graphs with fixed degree sequences) are
25: %LD-convergent. The partial support for these conjectures is given
26: %in~\cite{BayatiGamarnikTetali} and~\cite{GamarnikRightConvergence},
27: %but establishing this conjecture in general will have a far reaching
28: %implications for the theory of spin glasses (we will discuss this connection
29: %in detail in the body of the paper).
30: 
31: 
32: Finally, we establish several previously unknown relationships
33: between the formerly defined notions of convergence. In
34: particular, we show that partition-convergence does not imply
35: left or right-convergence, and that right-convergence does not
36: imply partition-convergence.
37: \end{abstract}