Sufficient conditions for symmetric matrices to have exactly one positive eigenvalue (Q2175187)

Let \(M_{n}(\mathbb{R})\) be the set of all \(n\times n\) real matrices. If all entries of \(A=\left[ a_{ij}\right] \in M_{n}(\mathbb{R})\) are non-negative, then \(A^{\circ r}=\left[ a_{ij}^{r}\right] \) denotes the \(r\)-th Hadamard power of \(A\) for any \(r\geq 0\). We say that \(A\in M_{n}(\mathbb{R})\) is infinitely divisible if for every \(r\geq 0\), the matrix \(A^{\circ r}\) is positive semidefinite. For a matrix \(A=\left[ a_{ij}\right] \in M_{n}(\mathbb{R})\) with all its entries non-zero, \(A^{\circ -1}=\left[ 1/a_{ij}\right] \) is the Hadamard inverse of \(A\). If \(A\in M_{n}(\mathbb{R})\) is a symmetric matrix with positive entries and exactly one positive eigenvalue, then we write that \(A\in \mathcal{A}\). Let \(\mathbf{e}\in \mathbb{R}^{n}\) be the vector of all ones. We say that a symmetric matrix \(A\in M_{n}(\mathbb{R})\) is conditionally negative semidefinite if \textbf{x}\(^{t}A\)\textbf{x}\( \leq 0\) for all \textbf{x}\(\in \mathbb{R}^{n}\) such that \textbf{x}\(^{t}\)\textbf{e}\(=0\). A function \(f:(0,\infty )\rightarrow \mathbb{R}\) is said to be completely monotonic if \(a\) has derivatives of all orders and satisfies the inequality \((-1)^{n}f^{(n)}(x)\geq 0\) where \(x>0\) and \(n=0,1,2,\) \dots. A differentiable function \(f:(0,\infty )\rightarrow (0,\infty )\) is called a Bernstein function if its derivative \(f^{\prime }\) is completely monotonic. The authors first present and prove the following (main) result. Theorem. Let \(A=\left[ a_{ij}\right] \in M_{n}(\mathbb{R})\) be a (symmetric) conditionally negative semidefinite matrix with all entries positive and let \(f:(0,\infty )\rightarrow (0,\infty )\) be a Bernstein function. Then \(\left[ f(a_{ij})\right] \in \mathcal{A}\). Moreover, if \(A\) is invertible, then \(\left[ f(a_{ij})\right] \) is invertible. This theorem is then used to derive some new results about matrices that belong to the set \(\mathcal{A}\). The authors show that if \(A\in \mathcal{A}\), then \begin{itemize} \item[1.] \(A^{\circ \alpha }\in \mathcal{A}\) for every \(0<\alpha \leq 1\), and every such a Hadamard power of \(A\) is invertible if \(A\) is invertible; \item[2.] its Hadamard inverse \(A^{\circ -1}\) is infinitely divisible; \item[3.] for every \(r>0\), \(rA+\mathbf{vv}^{t}\in \mathcal{A}\) where \textbf{v } is the Perron eigenvector of \(A\). \end{itemize} The above result is also used to show that some (special) matrices that are derived from a symmetric conditionally negative semidefinite matrix \(A\) with positive entries are infinitely divisible. The authors conclude the paper by presenting the following two cases when the Hadamard power \(A^{\circ \alpha }\), \(0<\alpha <1\), of \(A\) has one positive eigenvalue and is invertible: \begin{itemize} \item[(i)] \(A=\left[ a_{ij}\right] \in M_{n}(\mathbb{R})\) is a symmetric conditionally negative semidefinite matrix with non-negative entries and \(2a_{ij}>a_{ii}+a_{jj}\) for all \(i\neq j\); \item[(ii)] \(A\in M_{n}(\mathbb{R})\) is a symmetric matrix with positive off-diagonal entries, zero diagonal entries, and just one positive eigenvalue. \end{itemize}