Spectra of Jacobi operators via connection coefficient matrices (Q2662845)

A Jacobi operator \(J\) on \(\ell_2\) has a real symmetric tridiagonal matrix representation with respect to the standard basis. Its spectral measure \(\mu\) is a probability measure supported on its spectrum \(\sigma(J)\subset\mathbb{R}\). The coefficients in the tridiagonal matrix are the three term recurrence coefficients of an orthogonal polynomial sequence (OPS) \(\{P_k\}\) forming a basis for \(L_\mu^2(\mathbb{R})\). Given two Jacobi operators \(J\) and \(D\) with OPS \(\{P_k\}\) and \(\{Q_k\}\) respectively, then there is an upper triangular connection matrix \(C\) such that \(P_k=\sum_{j=0}^k c_{j,k}Q_j\), \(k=0,1,\ldots\). A particular Jacobi operator \(\Delta\) with zero diagonal and constant 1/2 as sub and sup diagonal has a well known spectral measure that is a (scaled) semi-circle on \([-1,1]\) and as OPS \(\{U_k\}\) the Chebyshev polynomials of the second kind. The main results of this paper concern the spectra of Jacobi operators of the form \(J=\Delta+K\) with \(K\) compact. In particular the spectrum of \(J\) is computable using the connection matrix. If \(K\) has finite rank, then \(C=C_{\mathrm{Toep}}+C_{\mathrm{fin}}\) with \(C_{\mathrm{Toep}}\) Toeplitz, upper triangular whose symbol is a polynomial \(c\) of degree \(2n-1\), and \(C_{\mathrm{fin}}\) is zero outside its principle \((n-1)\times(2n-1)\) block. \(|c(e^{i\theta})|^2\) scales \(\mu_\Delta\) and the Joukowski transform of the \(r<n\) real zeros of \(c\) in \([-1,1]\) define an additional point spectrum of \(J\). As operators on a Banach space \(\ell_R^1\), \(R>1\) (with norm \(\|v\|_{1,R}=\sum|v_k|R^k<\infty\)), one may consider a perturbation \(K\) of trace class. In that case \(C=C_{\mathrm{Toep}}+C_K\) with \(C_{\mathrm{Toep}}\) Toeplitz with an symbol \(c\) analytic in the unit disk and \(C_K:\ell_R^1\to\ell_R^1\) is compact. The roots of \(c\) in the unit disk define the additional discrete spectrum as above via a Joukowski transform. Also the Solvability Complexity Index (SCI) (see for example [\textit{J. Ben-Artzi} et al., C. R., Math., Acad. Sci. Paris 353, No. 10, 931--936 (2015; Zbl 1343.68078)]) for these problems are investigated. For both cases mentioned above, it is guaranteed that any desired accuracy can be attained with guaranteed error control. In an appendix, numerical examples illustrate the result. They are obtained using the author's Julia software package \texttt{SpectralMeasures.jl} that they made available via github.