Gauss-Hermite interval quadrature rule (Q2460611)

Let \({\mathbf h}= (h_k)\in\mathbb{R}^n\), \(H\), \(M\), \(\varepsilon_0> 0\), \({\mathbf H}^H_n= \{{\mathbf h}\in\mathbb{R}^n\mid h_k\geq 0\), \(k=1,\dots, n\), \(\sum^n_{k=1} h_k< {\mathbf H}\}\), \({\mathbf X}_n({\mathbf h})= \{{\mathbf x}\in \mathbb{R}^n\mid -\infty< x_1- h_1\leq x_1+ h_1<\cdots< x_n- h_n\leq x_n+ h_n<+\infty\}\), \({\mathbf X}^{M,\varepsilon_0}_n({\mathbf h})= \{{\mathbf x}\in\mathbb{R}^n\mid -M< x_1- h_1\), \(x_{k+1}- h_{k+1}- x_k- h_k> \varepsilon_0\), \(k= 1,\dots, n-1\), \(x_n+ h_n< M\}\). Given \({\mathbf h}\in{\mathbf H}^H_n\) the Gauss-Hermite interval quadrature rule is an interpolation quadrature rule of the form \[ I_f({\mathbf h},{\mathbf x})= \sum^n_{k=1} {\mu_k\over 2h_k} \int^{x_k+ h_k}_{x_k- h_k} f(x) e^{-x^2} dx,\quad{\mathbf x}= (x_k)\in {\mathbf X}_n({\mathbf h}), \] designed as an approximation of the integral \(I_f= \int^b_a f(x) e^{-x^2} dx\) with coefficients \(\mu_k\) such that \(I_f({\mathbf h},{\mathbf x})= I_f\) if \(f\) is any polynomial of degree less or equal than \(2n- 1\). If \(a= -\infty\), \(b=\infty\) it is shown that for every \(H> 0\) there exist \(\varepsilon_0> 0\) and \(M> 0\) such that for every \({\mathbf h}\in{\mathbf H}^H_n\) the Gauss-Hermite quadrature rule with \({\mathbf x}\in{\mathbf X}^{M,\varepsilon_0}_n\) and positive weights \(\mu_k\), \(k= 1,\dots,n\), exists uniquely. This result is illustrated by numerical examples.