No label defined (Q4532113)

The author discusses some formulations of the spectral problem from linear algebra in connection with studying convergence of an iteration process which solve matrix equations. Complexity of many engineering problems is caused by the occurrence of a class of (in general) nonselfadjoint matrix operators whose eigenvalues of spectral cluster are ill-conditioned. In various applications, it is convenient to describe properties of such operators in terms of a criteria for spectral dichotomy. The spectral analysis can be reduced to the study of dichotomy of the spectrum by some curve \(\gamma\) of finite length \(l_{\gamma} = \oint_{\gamma}|d\lambda|\). The spectrum is divided into fibers by a series of plane curves depending on a single parameter. With such curves, a Hermitian matrix \(J_{\gamma(s)}\) is associated which is defined by the integral \[ J_{\gamma(s)} = \oint_{\gamma(s)}(\lambda I - A)^{-*}(\lambda I - A)^{-1} |d\lambda| \] and depends on a parameter \(s\) enumerating the curves. The matrix \(J_{\gamma(s)}\) is similar to an integral representation of the required solution to a matrix equation in the case of dichotomy by a circle. The graphical dependence of a criterion for dichotomy on this parameter is regarded as a spectral portrait of the matrix. The use of one-dimensional spectral portraits is based on the hypothesis of the existence of a partition of spectrum (linear, circular, or some other). For operators in an infinite-dimensional space, a criterion for an annular partition of spectrum was suggested by \textit{M. V. Keldysh} in [Dokl. Akad. Nauk SSSR, n. Ser. 77, 11-14 (1951; Zbl 0045.39402)] and justified by \textit{V. B. Lidskij} in [Tr. Moskov. Mat. Obshch. 11, 3-35 (1962; Zbl 0117.33103)]. In the finite-dimensional case, the criterion states that the so-called spectral conditioned number of class \(p\) is finite.