Covering an ellipsoid with equal balls (Q855867)

Consider the ball \(B^n_\varepsilon(y)\) of radius \(\varepsilon> 0\) centered at some point \(y\in\mathbb{R}^n\). For any set \(A\subset\mathbb{R}^n\), a subset \(M_\varepsilon(A)\subseteq\mathbb{R}^n\) is called its \(\varepsilon\)-covering if \(A\) is contained in the union of the balls of radius \(\varepsilon\) centered at points \(y\in M_\varepsilon(A)\). The so-called \(\varepsilon\)-entropy of a set \(A\subset\mathbb{R}^n\) is defined by \[ \chi_\varepsilon(A)= \log\min_{M_\varepsilon(A)}\, |M_\varepsilon(A)|, \] where the minimum is taken over all coverings and the logarithm has base \(e\). The author studies the \(\varepsilon\)-entropy of an arbitrary ellipsoid \[ E^n_a= \Biggl\{x\in \mathbb{R}^n: \sum^n_{i=1} {x^2_i\over a^2_i}\leq 1\Biggr\}, \] where \(a= (a_1,\dots, a_n)\) is assumed to satisfy, without loss of generality, \(0< a_1\leq\cdots\leq a_n\). Using the equivalent problem of covering an ellipsoid \(E^n_a\) with unit balls, the author removes the subscript \(\varepsilon\). His main goal is to find the asymptotic (unit) entropy \(\chi(E^n_a)\) as a function of \(n\) and \(a\), considering the subsets of ellipsoids such that \(\chi(E^n_a)\to \infty\). A tight asymptotic bound on the logarithm of the minimum number of unit balls necessary to cover ellipsoids is obtained.