2018
DOI: 10.48550/arxiv.1804.07808
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

Spectral gap in random bipartite biregular graphs and applications

Abstract: We prove an analogue of Alon's spectral gap conjecture for random bipartite, biregular graphs. We use the Ihara-Bass formula to connect the non-backtracking spectrum to that of the adjacency matrix, employing the moment method to show there exists a spectral gap for the non-backtracking matrix. Finally, we give some applications in machine learning and coding theory.

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

0
28
0

Year Published

2018
2018
2022
2022

Publication Types

Select...
7
1

Relationship

0
8

Authors

Journals

citations
Cited by 16 publications
(28 citation statements)
references
References 43 publications
0
28
0
Order By: Relevance
“…goes to 1 as n → ∞. This was extended to (c, d)-biregular bipartite graphs in [3]. However, we do note that although one might expect almost-Ramanujan and Ramanujan graphs to have almost-identical properties, there are certain proof techniques (related to the Ihara Zeta function) for which Ramanujan graphs are distinctly better, for reasons that are beyond the scope of this article (see [2]).…”
Section: Previous Workmentioning
confidence: 82%
“…goes to 1 as n → ∞. This was extended to (c, d)-biregular bipartite graphs in [3]. However, we do note that although one might expect almost-Ramanujan and Ramanujan graphs to have almost-identical properties, there are certain proof techniques (related to the Ihara Zeta function) for which Ramanujan graphs are distinctly better, for reasons that are beyond the scope of this article (see [2]).…”
Section: Previous Workmentioning
confidence: 82%
“…Finally, we mention that the main technical challenge in our approach is to establish that in every gadget, with high probability, either every vertex of L is assigned "+" and every vertex of R is assigned "−" or vice versa; see Theorems 4.1 and 4.2. To show this, we require very precise bounds on the edge expansion of the random bipartite graph G. When d → ∞, these bounds can be derived in a fairly straightforward manner from the results in [10]. However, the case of d = O(1) is more difficult, and it requires for us to define the notion of edge expansion with respect to the ports of the gadget and extending some of the ideas in [39] (see Theorem 5.3).…”
Section: Main Results For the Ising Model: Proof Overviewmentioning
confidence: 99%
“…When G is d-regular and bipartite, the eigenvalues of the adjacency matrix of G are symmetric around 0. In other words, the eigenvalues of the adjacency matrix are either 0 or pairs of λ and −λ [34]. Thus, max{λ 2 , λ n } has the trivial value of d. In such cases, the spectral gap is defined as follows.…”
Section: B Spectral Expansion Of Incident Structurementioning
confidence: 99%
“…[34] Let G = (V, E) be an d-regular bipartite graph, where|V| = n. Let λ 1 λ 2 • • • −λ 2 −λ 1 bethe eigenvalues of the adjacency matrix of G. The spectral gap of the graph G is defined as ∆(G) = d − λ The associated graphs with round-robin and BIBD scheduling policies are d-regular bipartite, with spectral gap d − λ 2 . In the following, we derive the spectral gap of these two policies.…”
mentioning
confidence: 99%