Spectra universally realizable by doubly stochastic matrices (Q1792443)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Spectra universally realizable by doubly stochastic matrices |
scientific article |
Statements
Spectra universally realizable by doubly stochastic matrices (English)
0 references
12 October 2018
0 references
Let \(\Lambda=\{\lambda_1,\lambda_2,\dots,\lambda_n\}\) be a list of complex numbers. If there exists a nonnegative matrix with spectrum \(\Lambda\), then \(\Lambda\) is said to be realizable. A list \(\Lambda\) is called universally realizable if it is the spectrum of a nonnegative matrix for every possible Jordan canonical form allowed by \(\Lambda\). Let \(\mathcal{CS}_\gamma\) be the set of all matrices with constant row sums \(\gamma\). A result of \textit{C. R. Johnson} [Linear Multilinear Algebra 10, 113--130 (1981; Zbl 0455.15019)] asserts that if a nonnegative matrix \(A\) has spectrum \(\Lambda=\{\lambda_1,\lambda_2,\dots,\lambda_n\}\) and Perron eigenvalue \(\lambda_1\), then \(A\) is cospectral to a nonnegative matrix \(B\in\mathcal{CS}_{\lambda_1}\), i.e., the spectrum of \(A\) and \(B\) coincide. Complete characterizations for realizable lists or universally realizable lists remain open. The authors extend Johnson's result from cospectrality to similarity by imposing irreducibility on the nonnegative matrix \(A\) and an additional condition that \(A\) has a positive row or column. This result is further applied to obtain sufficient conditions for similarity and cospectrality between nonnegative matrices and positive matrices. A main idea lies on rank one perturbation of matrices. More precisely, if the spectrum of \(A\) is given, one can characterize the spectrum of \(A+\mathbf{v}\mathbf{q}^T\) where \(\mathbf{v}\) is an eigenvector of \(A\) and \(\mathbf{q}\) is any vector. For nonnegative matrices \(A\in\mathcal{CS}_{\lambda_1}\), take \(\mathbf{v}\) to be the vector with all entries equal 1, and \(\mathbf{q}\) be a vector with \(\mathbf{q}^T\mathbf{v}=0\), then \(A+\mathbf{v}\mathbf{q}^T\) and \(A\) are similar. In particular, they have the same Jordan canonical form. The positive row or column of the nonnegative matrix \(A\in\mathcal{CS}_{\lambda_1}\) guarantees the existence of \(\mathbf{q}\) such that \(\mathbf{q}^T\mathbf{v}=0\) and \(A+\mathbf{v}\mathbf{q}^T\) is positive. The authors further consider nonnegative generalized doubly stochastic matrices, i.e., nonnegative matrices \(A\) with \(A,A^T\in \mathcal{CS}_\gamma\). They give a sufficient condition on a list \(\Lambda\) of real numbers in which \(\Lambda\) is universally realizable by nonnegative generalized doubly stochastic matrices. Moreover, if the list \(\Lambda\) is realizable by nonnegative generalized doubly stochastic matrices, then the authors construct a class of universally realizable lists of complex numbers from \(\Lambda\). However, for a weaker assumption that \(\Lambda\) is realizable by nonnegative matrices with a positive row or column, results weaker than universally realizable hold for the corresponding constructed class of lists. Some numerical examples are discussed to illustrate the results.
0 references
elementary divisor
0 references
nonnegative matrix
0 references
nonnegative inverse elementary divisors problem
0 references