On spectral approximation of linear operators (Q1268750)

Let \(T\) be a bounded linear operator on a Banach space \(E\), and let \(\Lambda\) be a subset of the discrete spectrum of \(T\) which is separated by a simple closed curve from the rest of the spectrum and from \(0\), and suppose that a sequence \((T_n)\) of perturbations of \(T\) satisfies the conditions: (H1) spectralradius\((T- T_n)\to 0\) as \(n\to\infty\), (H2) \(\| (T- T_n)T_n\|\to 0\) as \(n\to\infty\). Let \(T= T^{(1)}\oplus T^{(2)}\) be the standard decomposition of \(T\) for which the spectrum of \(T^{(1)}\) is \(\Lambda\). Operators \({\mathbf T}\) and \(({\mathbf T}_n)\) on \({\mathbf X}= X\times X\) are constructed such that \(T\) and \({\mathbf T}\) have the same spectral structure, \(\|{\mathbf T}-{\mathbf T}_n\|\to 0\) and there are decompositions \({\mathbf T}= {\mathbf T}^{(1)}\oplus {\mathbf T}^{(2)}\) and \({\mathbf T}_n= {\mathbf T}^{(1)}\oplus {\mathbf T}^{(2)}_n\) such that the operators \(T^{(1)}\) and \({\mathbf T}^{(1)}\) are similar and, for all large \(n\), the operators \(T^{(1)}_n\) and \({\mathbf T}^{(1)}_n\) are similar. The construction allows refined error estimates for the approximation of the part \(\Lambda\) of the discrete spectrum of \(T\) by the discrete spectra of \((T_n)\) to be obtained from the known results for approximation by a norm convergent sequence of perturbations.