Interpolation orbits and optimal Sobolev's embeddings (Q2464866)

The point of departure is the pair of extreme Sobolev embeddings, namely, \(W^{m}L_1\subset\Lambda_{1-\frac mn}\) and \(W^ m\Lambda_{\frac mn}\subset L^{\infty}\), where \(\Lambda_{\alpha}\) is the Lorentz endpoint space normed by \(\int_0^1f^*(t)\,dt^{\alpha}\) and \(f^*\) is the nonincreasing rearrangement of \(f\). Applying first an interpolation functor \(\mathcal F\) and then the Nilsson refinement of the DeVore-Scherer theorem, the authors obtain the inclusion \(W^ m\mathcal F(L_1,\Lambda_{\frac nm})\subset \mathcal F(\Lambda_{1-\frac mn}, L_{\infty})\). Restricting this to the \(K\)-method functors, that is, to those \(\mathcal F\) which can be obtained using some Banach lattice \(F\) from the identity \(\mathcal F(X_0,X_1)=(X_0,X_1)_F^ K\), one gets \(W^ m(L_1,\Lambda_{\frac nm})^ K_F\subset (\Lambda_{1-\frac mn}, L_{\infty})^ K_F\). Thus, for any r.i.~spaces \(E\) and \(G\) such that \(E\subset(L_1,\Lambda_{\frac nm})^ K_F\) and \((\Lambda_{1-\frac mn}, L_{\infty})^ K_F\subset G\), we arrive at the Sobolev embedding \(W^ mE\subset G\). In the main result of this interesting paper, the authors show that, in fact, this is the only way how to get such an embedding. In other words, for all pairs of interpolation r.i.~spaces \(E\) and \(G\) satisfying the above embedding, there is such an \(F\). The result has interesting consequences concerning optimality of Sobolev embeddings in the r.i.~environment. In particular, the result allows, by reiteration, to construct new optimal Sobolev embeddings from known ones. The second main result describes the optimal range space in a~Sobolev embedding in terms of interpolation orbits. Proofs are based on an exceptional role of the one-dimensional Hardy-type operator \(H_{\frac nm}\), defined by \(H_{\frac nm}f(t):=\int_t^1s^{\frac mn-1}f(s)\,ds\), acting on function spaces defined on \((0,1)\).