Compact embeddings of Brézis--Wainger type (Q854553)

The paper deals with compact embeddings of type \[ \mathrm{id}_\Omega : B^{1+n/p}_{p,q}(\Omega) \rightarrow \mathrm{Lip}^{(1,-\alpha)}(\Omega), \] where \(\Omega\) is a bounded domain in \(\mathbb{R}^n\), and the spaces are defined by restriction from their \(\mathbb{R}^n\)-counterparts of the usual Besov spaces \(B^s_{p,q}\), \(0<p,q\leq\infty\), \(s\in\mathbb{R}\), and \( \mathrm{Lip}^{(1,-\alpha)}\), where the latter contain `almost' Lipschitz continuous functions \(f\), i.e., such that \[ \| f | \mathrm{Lip}^{(1,-\alpha)}(\mathbb{R}^n)\| = \| f| L_\infty( \mathbb{R}^n)\| + \sup_{0<| x-y| <1/2} \frac{| f(x)-f(y)| }{| x-y| | \log| x-y| | ^\alpha} \] is finite, \(\alpha\geq 0\). Embeddings of this type are usually called Brézis--Wainger embeddings nowadays, recalling the famous result on the almost Lipschitz continuity of functions from the Sobolev space \(H^{1+n/p}_p(\mathbb{R}^n)\), \(1<p<\infty\). It is well-known that the above embedding \(\mathrm{id}_\Omega\) is compact when \(\alpha> \max(1-1/q,0)\). Moreover, there are first results on the degree of compactness measured in terms of entropy (and approximation) numbers, respectively, but a complete characterisation in all possible cases (of the parameters) is still open. The present paper essentially contributes to closing this gap. As for the lower estimates, one obtains under very mild assumptions that \[ e_k(\mathrm{id}_\Omega) \geq c \max(k^{-1/p} (\log k)^{1/p-\alpha}, k^{-\alpha}) \] for \(k\to\infty\), whereas the outcome for the upper estimates is more tricky and depends, in addition, on the interplay between \(p\) and \(q\). Roughly speaking, the authors prove that \(e_k(\mathrm{id}_\Omega) \sim k^{-1/p}\) (up to a \(\log\)-term in \(k\)) when \(\alpha>\max(1+2/p-1/q, 1/p)\), and a two-sided (almost sharp) estimate in the case \(0<\alpha\leq 1/p\leq 1/q-1\). The main tools used in this interesting paper are the application of interpolation methods for entropy numbers, the (nowadays standard) transformation of the question from function spaces to corresponding sequence spaces, as well as refined estimates for embeddings of (weighted) sequence spaces.