Nonnegative linearization for polynomials orthogonal with respect to discrete measures (Q5945736): Difference between revisions

Let \(\{P_n(x)\}\) denote a sequence of orthogonal polynomials and consider the linearization problem \[ P_n(x)P_m(x)=\sum_kg(n,m,k)P_k(x). \] For several families of orthogonal polynomials the linearization coefficients \(g(n,m,k)\) can be computed explicitly (for instance by using the orthogonality property), but for some families explicit formulas are not available. The problem of determining if all linearization coefficients are nonnegative is an important subject in the theory of orthogonal polynomials. Nonnegativity of the linearization coefficients gives rise to important properties such as a convolution structure for the polynomials \(\{P_n(x)\}\). All families of orthogonal polynomials satisfy a three term recurrence relation of the form \[ xP_n(x)=\alpha_nP_{n+1}(x)+\beta_nP_n(x)+\gamma_{n-1}P_{n-1}(x), \] where \(P_{-1}(x)=0\), \(P_0(x)=1\), \(\alpha_n>0\) and \(\gamma_n>0\). There exist general criteria in terms of the coefficients of this recurrence relation for which the sequence of polynomials admits such a nonnegative linearization. For instance, if the sequences \(\{\gamma_n\}\), \(\{\alpha_n+\gamma_{n-1}\}\) and \(\{\beta_n\}\) are all nondecreasing and \(\alpha_n\geq\gamma_{n-1}\) for every \(n\), then the polynomials \(\{P_n(x)\}\) admit a nonnegative linearization. This was shown in [\textit{R. Szwarc}, SIAM J. Math. Anal. 23, No. 4, 959-964 (1992; Zbl 0772.42013); ibid. 965-969 (1992; Zbl 0772.42014)]. This criterion is strong enough for many classical orthogonal polynomial systems, but it does not cover the case when the orthogonality measure is supported on a sequence of points accumulating at some point \(b\). In that case, which is considered in this paper, it can be shown that \(\alpha_n\gamma_n\rightarrow 0\) and \(\beta_n\rightarrow b\). By using two different methods, one based on Motzkin paths and the other one on the maximum principle, the following main result is obtained: if the sequence \(\{\beta_n\}\) is increasing and if for every \(n\) the sequence \[ \frac{\alpha_m\gamma_m}{(\beta_n-\beta_m)(\beta_n-\beta_{m+1})},\quad m=0,1,\ldots,n-2, \] is a chain sequence, then the polynomials \(\{P_n(x)\}\) admit a nonnegative linearization.