Some new bounds for singular values and eigenvalues of matrix products (Q5933814)

Suppose \(A\) and \(B\) are \(n\times n\) Hermitian matrices and that either \(A\) or \(B\) is positive semi-definite. Let \(\pi(A)\) and \(\nu(A)\) denote the number of positive eigenvalues and the number of negative eigenvalues of \(A\), respectively. Also let \(\lambda_1(\cdot)\geq\lambda_2(\cdot) \geq\ldots\geq\lambda_n(\cdot)\) denote the eigenvalues of a matrix arranged in non-increasing order. Based on Sylvester's law of inertia and on Weyl's bounds for the eigenvalues of sums of Hermitian matrices, the following is shown: (i) when \(1\leq k\leq \pi(A)\), \[ \min_{1\leq i\leq k} [\lambda_i(A)\lambda_{k+1-i}(B)] \geq\lambda_k(BA)\geq \max_{k\leq i\leq n} [\lambda_i(A)\lambda_{n+k-i}(B)]; \] (ii) when \(\pi(A) < k \leq n-\nu(A)\), \[ \lambda_k(BA)=0; \] (iii) when \(n-\nu(A)<k\leq n\), \[ \min_{1\leq i\leq k} [\lambda_i(A)\lambda_{n+i-k}(B)] \geq\lambda_k(BA)\geq \max_{k\leq i\leq n} [\lambda_i(A)\lambda_{i+1-k}(B)]. \] This result is applied to obtain bounds for the singular values of general matrix products.