Average \(n\)-width of point set in Hilbert space (Q1202819)

Let \(S\) be a bounded linear operator from a real separable Banach space \(F\) into a real separable Hilbert space \(H\), and let \(\mu\) be a centered Gaussian measure on \(F\) with covariance operator \(Q\). If \(\varphi: \mathbb{R}_ +\to \mathbb{R}_ +\) is a strictly increasing, continuous and convex function such that \(\int \varphi (\| Sx\|) \mu(dx)<\infty\), then the \((\varphi, \mu)\)-average deviation of \(K= S(F)\) from a subspace \(X\) of \(H\) is defined by \[ E(K,X)= \int_ F \varphi(\inf_{y\in X} \| Sx-y\|) \mu(dx). \] The \((\varphi, \mu)\)-average Kolmogorov \(n\)-width of \(K\) in \(H\) is defined by \[ d_ n(K,H, \mu)= \inf_ X E(K,X), \] where the infimum is taken over all subspaces \(X\) of \(H\) with \(\dim X\leq n\). A subspace \(X_ 0\) of \(H\) with \(\dim X_ 0\leq n\) is called optimal, if \(d_ n (K,H,\mu)= E(K,X_ 0)\). In the present paper the average \(n\)-width \(d_ n(K,H, \mu)\) is computed. Especially, it is proved that the space \(X_ 0\), spanned by the eigenvectors belonging to the \(n\) largest eigenvalues of \(SQS^ t\) (the covariance operator of \(S(\mu)\)), is optimal.