Spectral theory of the \(G\)-symmetric tridiagonal matrices related to Stahl's counterexample (Q2255300)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Spectral theory of the \(G\)-symmetric tridiagonal matrices related to Stahl's counterexample
scientific article

    Statements

    Spectral theory of the \(G\)-symmetric tridiagonal matrices related to Stahl's counterexample (English)
    0 references
    0 references
    9 February 2015
    0 references
    Let consider the nonpositive weight function \(w(t)= \frac{1}{\pi}(t-\cos \pi \alpha) (1- t^{2})^{-1/2}\) in the interval \([-1,1]\) with \(\alpha\in (0,1)\) an irrational number. From a result by \textit{H. Stahl} [J. Comput. Appl. Math. 99, No. 1--2, 511--527 (1998; Zbl 0928.41011)], every point of the real line located outside \((-1,1)\) is a spurious pole of the sequence of the diagonal Padé approximants to \(S(z)= \int_{-1}^{1} \frac{w(t)}{t- z} dt,\) the Cauchy transform of the weight function \(w\). Indeed, at most one zero of \(P_{n},\) the polynomial of degree \(n\) orthogonal with respect to \(w,\) is located outside the interval \([-1,1]\) and such a zero moves erratically in \(\mathbb {R} \setminus [-1,1]\) when \(n\) tends to infinity. Thus, the zeros of the above orthogonal polynomials are dense in the real line. Such a zero corresponds to a pole of the corresponding diagonal Padé approximant. Notice that for positive weight functions supported on \([-1,1]\) a classical result by Markov states that the zeros of the denominators of the diagonal Padé approximants to the corresponding Cauchy transform are dense in the interval \([-1,1]\). In the paper under review the author gives an operator theory approach to the above results based on the analysis of a tridiagonal matrix \(\tilde{J}\) associated with the weight \(w\). It is obtained by using a shifted Darboux transformation of the Jacobi matrix \(J\) associated with the Chebyshev polynomials of first kind together with a \(G\)-symmetric matrix that is a sign diagonal matrix. Indeed, \(\tilde{J}\) is not symmetric but it has symmetry properties. It generates an unbounded operator in \(\ell^{2}\) that is self-adjoint in the Krein space \(\ell^{2}(G),\) i.e. the linear space of elements of \(\ell^{2}\) with a bilinear form \([f, g]= \langle Gf, g\rangle _{ \ell^{2}}\). Furthermore, taking into account \([(\tilde{J}-\cos \pi\alpha I)f, f ]\geq 0\) then \(\tilde{J}\) is not definitizable in the sense of \textit{H. Langer} [Lect. Notes Math. 948, 1--46 (1982; Zbl 0511.47023)]. Indeed, its resolvent is an empty set. Finally, in a more general framework, for a measure \(d\mu\) supported in \([-1, 1]\) and \(x\in (-1,1),\) it is proved that the infinity is not a spurious pole for Padé approximants of the Cauchy transform of \((t-x) d \mu (t)\) if and only if the matrix \(\tilde{J}(x)\) (the shifted Darboux transformation at \(x\) of the Jacobi measure associated with \(d\mu\)) is a definitizable operator in \(\ell^{2}(G)\).
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    orthogonal polynomials
    0 references
    Jacobi matrix
    0 references
    \(G\)-symmetric tridiagonal matrix
    0 references
    \(G\)-non-negative matrix
    0 references
    definitizable operator
    0 references
    Darboux transformation
    0 references
    triangular factorization
    0 references
    definitizable function
    0 references
    Padé approximation
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references