Probabilistic and average linear widths of weighted Sobolev spaces on the ball equipped with a Gaussian measure (Q1734614)

Let $K$ be a bounded subset of a normed linear space $X$ with norm $\|\cdot\|_X$. Then $\lambda_n(K,X)=\inf_{L_n}\sup_{x\in K}\|x-L_nx\|_X$, where $L_n$ runs over all linear operators from $X$ to $X$ with rank at most $n$. \par Let $W$ be a separable Banach space and assume that $W$ contains the Borel field $\mathcal B$ consisting of open subsets of $W$ and is equiped with a probabilty measure $\gamma$ defined on $\mathcal B$. For $0<p<\infty$ we define \[ \lambda^{(a)}_n(W,\gamma,X)_p=\inf\limits_{L_n}\left(\int_W\|x-L_nx\|_X\gamma(dx)\right)^{1/p}, \quad \lambda_{n,\delta}(W,\gamma,X)_p=\inf\limits_{G_\delta}\lambda_n(W\setminus G_\delta, X), \] where $G_\delta$ runs over all measurable subsets in $W$ with measure $\gamma(G_\delta)\leq \delta$. \par Let $\mathbb B^d$ be the unit ball in euclidean space $\mathbb R^d$. For the weight function $w_\mu(x)=(1-\|x\|^2_{\mathbb R^n})^{\mu-1/2}$, $\mu\geq 0$, one can consider the weight space $L_{p,\mu}=L_p(\mathbb B^d, w_\mu(x)\,dx)$ and Sobolev space $W^r_{2,\mu}=W^r_{2,\mu}(\mathbb B^d)$. The following result is proved in paper. \par Theorem 1.2. Let $0<p<\infty$, $1\leq q\leq \infty$, $r>(d+2\mu)(1/2-1/q)_+$, $s>d$, $\rho=r+s/2>d/2+2\mu d(1/2-1/q)_+$. Then \[ \lambda^{(a)}_n(W^r_{2,\mu},\nu, L_{q,\mu})_p\asymp \begin{cases} n^{-\rho/d+1/2}, 1\leq q<\infty, \\ n^{-\rho/d+1/2}(\ln(en))^{1/2}; q=\infty, \end{cases} \] where $\nu$ is the Gaussian measure on $W^r_{2,\mu}$. \par A similar result is obtained for $\lambda_{n,\delta}(W^r_{2,\mu},\nu, L_{q,\mu})$.