The Nevai condition (Q5962341): Difference between revisions

Let \(\text{d}\rho\) be a measure on \(\mathbb{R}\) of compact, but not finite, support and let \(\text{p}_n(x,\text{d}\rho)\) be the standard normalized orthogonal polynomial of degree \(n\) for \(\text{d}\rho\). The Jacobi parameters \(\{a_n,b_n\}\) are defined by \[ x \text{p}_n(x)=a_{n+1} \text{p}_{n+1}(x) + b_{n+1} \text{p}_{n}(x) + c_n \text{p}_{n-1}(x).\tag{1} \] Introduce the Christoffel-Darboux kernel by \[ K_n(x,y)=\sum_{j=0}n\, \text{p}_{j}(x) \text{p}_{j}(y), \] and \[ \text{d}\eta^{(x_0)}={|K_n(x,x_0)|^2\text{d}\rho(x)\over K_n(x_0,x_0)}, \] where \(x_0\) appears in the Christoffel variational principle: if \[ \lambda_n(x_0)=\min\{\int |Q_n(x)|^2\text{d}\rho(x)\mid \text{deg}Q_n\leq n,Q_n(x_0)=1\} \] then \[ \lambda_n(x_0)={1\over K_n(x_0,x_0)}. \] The authors then define that \(\text{d}\rho(x)\) obeys \textit{the Nevai condition} if \[ \text{w}-\lim_{n\rightarrow\infty}\,\text{d}\eta_n^{(x_0)}=\delta_{x_0}\;(\text{Dirac delta}) \] (where \(\text{w}-\lim_{n\rightarrow\infty}\,\text{d}\nu_n=\text{d}\nu\) means \(\lim_{n\rightarrow\infty}\int f(x)\text{d}\nu_n(x)= \int f(x)\text{d}\nu(x)\) for any compactly supported continuous \(f\)). This concept originated in [\textit{P. Nevai}, Mem. Am. Math. Soc. 18 (213), 1--183 (1979; Zbl 0405.33009)] and is connected with a limiting condition on the Jacobi parameters in (1). Under the condition \[ 0<A_{-}\equiv \inf_n a_n\leq\sup_n a_n\equiv A_{+}<\infty,\tag{2} \] Nevai showed that his condition holds at \(x_0\) if \[ \lim_{n\rightarrow\infty}\,{|\text{p}_n(x_0)|^2\over \sum_{j=0}^n\,|\text{p}_j(x_0)|^2}=0.\tag{3} \] In the paper under review the authors prove the converse: Theorem 2.2. Under condition (2), Nevai's condition is equivalent to \[ \int (x-x_0)^2\text{d}\eta_n^{(x_0)}(x)\rightarrow 0,\tag{4} \] and (4) holds if and only if (3) holds. Finally the authors exhibit an example of a measure (in the form of a Jacobi matrix) for which the Nevai condition fails for Lebesgue a.e. \(x_0 \in [-2,2]\setminus [-1,1]\). It must be remarked that the proofs in the paper exhibit many interesting and new intermediate results.