Barrett-Johnson inequalities for totally nonnegative matrices

Abstract

Given a matrix $A$, let $A_{I,J}$ denote the submatrix of $A$ determined by rows $I$ and columns $J$. Fischer's Inequalities state that for each $n \times n$ Hermitian positive semidefinite matrix $A$, and each subset $I$ of $\{1,\dotsc,n\}$ and its complement $I^c$, we have $\det(A) \leq \det(A_{I,I})\det(A_{I^c,I^c})$. Barrett and Johnson (Linear Multilinear Algebra 34, 1993) extended these to state inequalities for sums of products of principal minors whose orders are given by nonincreasing integer sequences $(\lambda_1,\dotsc,\lambda_r)$, $(\mu_1,\dotsc,\mu_s)$ summing to $n$. Specifically, if $\lambda_1+\cdots+\lambda_i\leq \mu_1+\cdots+\mu_i$ for all $i$, then $$ \lambda_1!\cdots\lambda_r! \sum_{(I_1,\dotsc,I_r)} \det(A_{I_1,I_1}) \cdots \det(A_{I_r,I_r}) ~\geq~ \mu_1!\cdots\mu_s! \sum_{(J_1,\dotsc,J_s)} \det(A_{J_1,J_1}) \cdots \det(A_{J_s,J_s}), $$ where sums are over sequences of disjoint subsets of $\{1,\dotsc,n\}$ satisfying $|I_k| = \lambda_k$, $|J_k| = \mu_k$. We show that these inequalities hold for totally nonnegative matrices as well.