Darboux transformations of Jacobi matrices and Padé approximation (Q716416)

The sequence \(\{P_j\}_{j=0}^\infty\) of monic orthogonal polynomials with respect to a probability measure \(\sigma\) on \({\mathbb R}\) (with infinite support and all moments finite) is known to satisfy the recurrence relation \[ \lambda P_j(\lambda) = P_{j+1}(\lambda) + b_j P_j(\lambda) +c_{j-1} P_{j-1}(\lambda) \] with initial conditions \[ P_{-1}(\lambda) = 0,\;P_0(\lambda) = 1, \] where \(b_j\in{\mathbb R}\) and \(c_j>0,\,j\in{\mathbb Z}_{+}\). The companions to these so-called polynomials of the first kind are the polynomials of the second kind \(Q_j\), satisfying the same recurrence relation, but with the initial conditions \[ Q_{-1}(\lambda) = -1,\;Q_0(\lambda) = 0. \] The semi-infinite tridiagonal matrix \(J\) satisfying \[ J p(\lambda) = \lambda p(\lambda), \] where \(p = (P_0,P_1,\dots,P_j,\dots)^{\text{T}}\), is said to be the monic Jacobi matrix associated with \(\sigma\). Using the unique \(LU\)-factorization \(J=LU\) (which exists if and only if \(P_j(0)\not= 0,\,j\in{\mathbb Z}_{+}\)), the Christoffel transformation of \(J\) is defined by \[ J = LU\mapsto J_C:= UL \] (\(J_C\) is the tridiagonal matrix associated with \(td\sigma(t)\)). If \(s_{-1}\) is a real number with \(Q_j(0)-s_{-1}P_j(0)\not= 0,\,j\in{\mathbb Z}_{+}\), then \(J\) has the \(UL\)-factorization \(J=UL\) and the Geronimus transformation of \(J\) with parameter \(s_{-1}\) is \[ J=UL\mapsto J_G:= LU. \] These two transformations are also called the Darboux transformations. In the paper under review, these transformations are extended beyond the conditions given: for an arbitrary monic Jacobi matrix, a pair of lower and upper triangular block matrices \(L\) and \(U\) is constructed such that \(J=LU\) and \[ {\mathcal J}_C=UL \] is a monic generalized Jacobi matrix, i.e., it is a tridiagonal block matrix with blocks as specified in their Definition 2.4. Similarly, with a free parameter \(s_{-1}\), the \(UL\)-factorization \(J=\mathcal{ U L}\) with lower and upper triangular block matrices is found and the Geronimus transform is also a tridiagonal block matrix \[ {\mathcal J}_G=\mathcal{ L U}. \] The associated Nevanlinna functions are \[ {\mathcal F}_C(\lambda)=\lambda\int_{{\mathbb R}}\,{d\sigma(t)\over t-\lambda}+1,\;{\mathcal F}_G(\lambda)=-{s_{-1}\over \lambda}+{1\over \lambda}\,\int_{{\mathbb R}}\,{d\sigma(t)\over t-\lambda}. \] After this introduction, the paper consists of the following sections: {\parindent=6mm \begin{itemize}\item[2.]Preliminaries. Several classes of Nevanlinna functions and normal indices of the sequence of coefficients of the asymptotic expansion, Schur transform, \(P\)-fraction, introduction of an indefinite inner product. \item[3.]The Christoffel transform and its inverse. Explicit forms of the relevant block structure for the matrices are given. \item[4.]The Geronimus transform and its inverse. Explicit forms of the relevant block structures, connection with normal indices. \item[5.]Generalized Cholesky decomposition. Explicit form of the block structure of the decomposition \(J=L\Lambda L^{\text{T}}\) using normal indices. \item[6.]Convergence of Padé approximants. Here, the probability measures are supposed to be supported on \([-1,1]\) and convergence results in terms of properties of the Darboux transform are given for sequences of diagonal approximants. \item[7.]Examples. Two examples are given, one exhibiting a bounded Jacobi matrix having an unbounded Christoffel transform. \end{itemize}}