Ratio asymptotics for orthogonal matrix polynomials with unbounded recurrence coefficients (Q5937252)

Consider orthogonal matrix polynomials \((R_{n,k})\), \(n \geq 1\), depending on a parameter \(k\), and given by a recurrence relation \[ t R_{n,k}(t)=A_{n+1, k} R_{n+1, k}(t) + B_{n,k} R_{n,k}(t)+ A_{n,k}^* R_{n-1, k}(t), \quad n \geq 0, \] with suitable initial conditions. Assuming convergence of certain sequences of coefficients \(\{A_{n,k} \}\), \(\{B_{n,k} \}\), the authors establish the ratio asymptotics of the corresponding sequences of \(R_{n,k}\), extending to the matrix setting an important result by \textit{A. B. J. Kuijlaars} and \textit{W. Van Assche} [J. Approximation Theory 99, No. 1, 167-197 (1999; Zbl 0967.33009)]. Assume now \(k\) fixed and \(\{A_{n,k} \}\), \(\{B_{n,k} \}\) increasing in a ``regular'' way: there exist positive definite matrices \(C_n\) such that \(\lim_n C_n^{-1/2} C_{n-1}^{1/2}=I\) and both \(C_n^{-1/2} A_{n,k} C_{n}^{1/2}\) and \(C_n^{-1/2} B_{n,k} C_{n}^{1/2}\) converge. In the scalar case this situation can be reduced to the one with varying coefficients by means of an appropriate change of variables. This is also true in the matrix setting, if we choose the correct definition of rescaled polynomials, issue thoroughly discussed in the paper. In this way, the authors apply their main theorem in order to derive the ratio asymptotics in the case of unbounded coefficients.