An analytic approach to a permanent conjecture (Q1940312)

The Hadamard product of two \(n\times n\) matrices \((a_{ij})\) and \((b_{ij})\) is defined to be \(A\circ B=(a_{ij}b_{ij})\). A permanent of an \(n\times n\) matrix \(A=(a_{ij})\) is \(\operatorname{per}A=\sum_{\sigma\in S_n}\prod_{t=1}^na_{t,\sigma(t)}\). \textit{J. Chollet} [Am. Math. Mon. 89, 57--58 (1982; Zbl 0507.15004)] asked if the following inequality holds for \(A\geq 0\) and \(B\geq 0\): \[ \operatorname{per}(A\circ B)\leq \operatorname{per}A \operatorname{per}B. \] Later, several stronger conjectures are formulated. In the present paper, the so-called maximizing matrices are defined and their properties are investigated. These matrices may be useful in investigation of the permanent conjecture. A matrix \(A\) is called a correlation matrix if \(A\geq 0\) and all main diagonal entries of \(A\) equal \(1\). For such \(A\) a matrix \(M_A\) is called maximizing if for it the function \(\operatorname{per}(A\circ X)\) takes the maximum value. It is proved in the paper that the maximizing matrix is singular, irreducible and invariant by row and column reduction. The latter means that \(\operatorname{per}(A(i|j))=\operatorname{per}(A)\) for all \(i,j\), where \(A(i|j)\) is the submatrix of \(A\) obtained by deleting row \(i\) and column \(j\).