arXiv++ Combinatorics

We study the combinatorial and algebraic properties of Nonnegative Matrices. Our results are divided into three different categories. 1. We show a quantitative generalization of the 100 year-old Perron-Frobenius theorem, a fundamental theorem which has been used within diverse areas of mathematics. The Perron-Frobenius theorem states that every irreducible nonnegative matrix $R$ has a largest positive eigenvalue $r$, and every other eigenvalue $λ$ of $R$ is such that $\text{Re}λ<r$ and $|λ|\leq r$. We capture the notion of irreducibility through the widely studied notion of edge expansion $φ$ of $R$ which intuitively measures how well-connected the underlying digraph of $R$ is, and show a quantitative relation between the spectral gap $Δ=1-\text{Re}λ/r$ (where $λ\not=r$ is the eigenvalue of $R$ with the largest real part) and the edge expansion $φ$, providing a more general result than the Cheeger-Buser inequalities, as follows.\[\dfrac{1}{15}\cdot\dfrac{Δ(R)}{n}\leqφ(R)\leq\sqrt{2\cdotΔ(R)}.\] 2. We study constructions of specific nonsymmetric matrices (or nonreversible Markov Chains) that have small edge expansion but large spectral gap, and provide a novel construction of a nonreversible chain for which\[φ(R)\leq\dfrac{Δ(R)}{\sqrt{n}},\] and we also present a candidate construction of matrices for which\[φ(R)\leq2\dfrac{Δ(R)}{n},\] which is the most beautiful contribution of this thesis. 3. We connect edge expansion and spectral gap to other combinatorial properties of nonsymmetric matrices, such as mixing time and capacity, and provide elementary proofs or unified views of known results and new results relating the different combinatorial/algebraic properties. Notably, we show the monotonicity of capacity for nonsymmetric nonnegative matrices.

The classic graphical Cheeger inequalities state that if $M$ is an $n\times n$ symmetric doubly stochastic matrix, then \[ \frac{1-λ_{2}(M)}{2}\leqφ(M)\leq\sqrt{2\cdot(1-λ_{2}(M))} \] where $φ(M)=\min_{S\subseteq[n],|S|\leq n/2}\left(\frac{1}{|S|}\sum_{i\in S,j\not\in S}M_{i,j}\right)$ is the edge expansion of $M$, and $λ_{2}(M)$ is the second largest eigenvalue of $M$. We study the relationship between $φ(A)$ and the spectral gap $1-\text{Re}λ_{2}(A)$ for any doubly stochastic matrix $A$ (not necessarily symmetric), where $λ_{2}(A)$ is a nontrivial eigenvalue of $A$ with maximum real part. Fiedler showed that the upper bound on $φ(A)$ is unaffected, i.e., $φ(A)\leq\sqrt{2\cdot(1-\text{Re}λ_{2}(A))}$. With regards to the lower bound on $φ(A)$, there are known constructions with \[ φ(A)\inΘ\left(\frac{1-\text{Re}λ_{2}(A)}{\log n}\right), \] indicating that at least a mild dependence on $n$ is necessary to lower bound $φ(A)$. In our first result, we provide an exponentially better construction of $n\times n$ doubly stochastic matrices $A_{n}$, for which \[φ(A_{n})\leq\frac{1-\text{Re}λ_{2}(A_{n})}{\sqrt{n}}.\] In fact, all nontrivial eigenvalues of our matrices are $0$, even though the matrices are highly nonexpanding. We further show that this bound is in the correct range (up to the exponent of $n$), by showing that for any doubly stochastic matrix $A$, \[φ(A)\geq\frac{1-\text{Re}λ_{2}(A)}{35\cdot n}.\] Our second result extends these bounds to general nonnegative matrices $R$, obtaining a two-sided quantitative refinement of the Perron-Frobenius theorem in which the edge expansion $φ(R)$ (appropriately defined), a quantitative measure of the irreducibility of $R$, controls the gap between the Perron-Frobenius eigenvalue and the next-largest real part of any eigenvalue.