Properties and preservers of the pseudospectrum (Q651204)

Let \(M_n\) denotes the set of \(n\times n\) complex matrices equipped with the operator norm \(\|\cdot\|\) induced by the usual vector norm \(\| x\|=(x^*x)^{1/2}\) on \({\mathbb C}^n\): \[ \|A\|=\max \{\|Ax\|: x\in {\mathbb C}^n, \;0<\|x\|\leq1\}. \] Let \(\varepsilon>0\). The \textit{pseudospectrum} of a matrix \(A\in M_n\) is defined by \[ \sigma_{\varepsilon} (A) =\{\mu\in {\mathbb C} : \text{ there are } x\in {\mathbb C}^n , E\in M_n \text{ such that } (A+E)x=\mu x \}. \] In the paper it is shown that the pseudospectrum can be used to study the algebraic and geometric properties of matrices. For example, it is shown that one can characterize Hermitian matrices, positive semidefinite matrices, orthogonal projections, unitary matrices, etc. in terms of the pseudospectrum. Moreover, the authors study maps \(\Phi: M_n\to M_n\) such that \(\sigma_{\varepsilon} (A*B)=\sigma_{\varepsilon}(\Phi(A)*\Phi(B))\) for all \(A,B\in M_n\), where \(A*B=A+B,\;A-B\) or \(AB\). It is shown that such maps are always unitary similarity transformations followed by some operations such as adding a constant matrix, taking the matrix transpose, or multiplying by a scalar from \(\{-1;1\}\).