Some remarks on a theorem of Gudkov (Q1899388)

Consider an \(n \times n\) complex matrix \(A = (a_{ij})\) with all diagonal entries nonzero and define: \[ \ell_1(A) = 0,\;\ell_i (A) = \sum^{i - 1}_{j = 1} |a_{ij}|{R_j(A) \over |a_{jj}|},\;2 \leq i \leq n, \] \[ R_i (A) = \ell_i (A) + \sum^n_{j = i + 1} |a_{ij}|,\;1 \leq i \leq n - 1,\;R_n(A) = \ell_n(A). \] The main result is the following. Let \(A\) satisfy: \(|a_{ii}|\geq R_i (A) \text{ and } |a_{ii}|> \ell_i(A), \quad 1 \leq i \leq n\). Then \(A\) is nonsingular (up to a simultaneous permutation of rows and columns). This is an improvement of a nonsingularity criterion of \textit{V. V. Gudkov} [Latv. mat. Ežegodnik, Riga 1965, 385-390 (1966; Zbl 0192.13303)] which is based on the inequalities \(|a_{ii} |> R_i(A)\), \(1 \leq i \leq n\), only. The new result covers a larger class of matrices and is used in order to bound an isolated eigenvalue and to investigate the corresponding eigenvector.