Pseudospectra of operator polynomials

From MaRDI portal





Let \(L(H)\) be the \(C^*\)-algebra of all bounded linear operators on a separable Hilbert space \(H\). Let \(A\in L(H)\) have a matrix representation \((a_{ij})^\infty_{i,j=1}\) with respect to some fixed basis of \(H\), and denote the \(n\)th finite section \((a_{ij})^n_{i,j=1}\) of \(A\) by \(A_n\). The paper under review deals with the relation between the spectrum \(\sigma(A)\) and the set of all limit points of the eigenvalues of the matrix \(A_n\) as \(n\to\infty\), to be denoted by \(\lim_{n\to\infty} \sigma(A_n)\). It is known that if \(A\) is selfadjoint and has a connected spectrum, they coincide with each other, i.e. \(\lim_{n\to\infty} \sigma_{\mathbb{C}^n\times \mathbb{C}^n}(A_n)= \sigma_{L(H)}(A)\), while such convergence behaviour can fail unless \(A\) is normal. It is observed by \textit{L. Reichel} and \textit{L. N. Trefethen} [Linear Algebra Appl. 162-164, 153-185 (1992; Zbl 0748.15010); Pitman Res. Notes Math. Ser. 260, 234-266 (1992; Zbl 0798.15005)] that for some class of operators \(A\) in \(L(H)\), the convergence behaviour becomes better when the spectrum \(\sigma(A)\) of \(A\) is replaced by what is called the \(\varepsilon\)-pseudospectrum \(\sigma^{(\varepsilon)}(A)\) of \(A\). By definition, given \(\varepsilon> 0\), it is the set of all \(\lambda\in\mathbb{C}\) for which either \(A-\lambda I\) is not invertible or it is invertible but \(\|(A-\lambda I)^{-1}\|\geq 1/\varepsilon\). It is a non-empty compact subset of the complex plane \(\mathbb{C}\), as the usual spectrum is. As is mentioned, indeed, this notion is reflecting in some sense finite accuracy in numerical analysis, for computation by a computer cannot distinguish between a non-invertible matrix and an invertible matrix whose inverse has a very big norm. In this paper the author obtains such a result as above for the \(\varepsilon\)-pseudospectrum in place of the spectrum with \(A\) being an operator polynomial \(L\). Here an operator polynomial \(L(\lambda)\) is a polynomial of \(\lambda\) like \(b_0+ b_1\lambda+\cdots+ b_m\lambda^m\) with coefficients \(b_0,\dots, b_m\) being elements in \(L(H)\) (in fact, which need not be restricted to ones in \(L(H)\), but may be in an arbitrary \(C^*\)-algebra \({\mathcal B}\) with identity element \(e\)). The spectrum of \(L\) is defined as the set \(\sigma(L)\) of all \(\lambda\in\mathbb{C}\) for which \(L(\lambda)\) is not invertible, and for \(\varepsilon> 0\), the \(\varepsilon\)-pseudospectrum of \(L\) as the set \(\sigma^{(\varepsilon)}(L)\) of all \(\lambda\in\mathbb{C}\) for which either \(L(\lambda)\) is not invertible or it is invertible but \(\|L(\lambda)^{-1}\|\geq 1/\varepsilon\). In case \(L(\lambda)= b-\lambda e\), they reduce to the common spectrum and \(\varepsilon\)-pseudospectrum of \(b\in{\mathcal B}\). As for \(L_n(\lambda)\), they are polynomials \(a^{(0)}_n+ a^{(1)}_n\lambda+\cdots+ a^{(m)}_n \lambda^m\) with coefficients \((a^{(0)}_n), (a^{(1)}_n),\dots, (a^{(m)}_n)\) bounded sequences of elements in \(L(H)\) or \({\mathcal B}\). The paper is based on an approach, in an earlier work by the author and \textit{B. Silbermann} [J. Oper. Theory 35, No. 2, 241-280 (1996; Zbl 0865.65035)], to the study of asymptotic behaviour of pseudospectra, whose idea is essentially due to \textit{A. Böttcher} [J. Integral Equations Appl. 6, No. 3, 267-301 (1994; Zbl 0819.45002)]. For the entire collection see [Zbl 0966.00029].











This page was built for publication: Pseudospectra of operator polynomials

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2752833)