A Markov-Bernstein inequality for Gaussian networks (Q5316359)

Let \(g\) be a Gaussian network, i.e. a function on \({\mathbb R}^s\) of the form \[ g(x) = \sum_{k=1}^{N} a_k \exp{(-\| x-x_k\| ^2)}, \quad x,x_k \in {\mathbb R}^s, \;a_k \in {\mathbb R}, \;1 \leq k \leq N. \] Let \(g\) be such that \(\min\limits_{1 \leq k,j \leq N,\, k \neq j}{\| x_j-x_k\| } \geq \frac{1}{m}\) and \(N \leq C \exp{(Am^2)}\), where \(A,C,m>0\) are some constants. The following Markov-Bernstein inequality is proved in the paper: if \(g\) is a Gaussian network as described above and \({\mathcal D}\) is a partial derivative operator of order \(r\), then \[ \| {\mathcal D}g\| _{L_p({\mathbb R}^s)} \leq c m^r \| g\| _{L_p({\mathbb R}^s)}, \quad 1 \leq p \leq \infty, \] with a constant \(c\) depending only on \(A\), \(C\), \(p\), \(r\), and \(s\). For the proof of the inequality, the author established some new results on estimating the degree of polynomial approximation with exponential weights. For the entire collection see [Zbl 1064.41003].