Sobolev inequalities: symmetrization and self-improvement via truncation (Q2461253)

Due to the Maz'ya truncation principle and iteration, all the first-order Sobolev--Lebesgue embeddings follow from the weak Gagliardo--Nirenberg inequality, or, equivalently, from the isoperimetric inequality. \textit{G.\,Talenti} [in: Nonlinear Analysis, Function spaces and Applications, Vol.~5, Prometheus, Prague, 177--230 (1994; Zbl 0872.46020)] obtained from the isoperimetric inequality and the coarea formula the estimate \[ t^{1-\frac1n}(-f^*)'(t)\leq C\int_{\{| f|>f^*(t)\}}|\nabla f(x)|\,dx, \] where \(C\) depends only on the dimension \(n\). Talenti's inequality is known to yield sharp Sobolev embeddings in the setting of rearrangement-invariant spaces. \textit{J.\,Bastero, M.\,Milman} and \textit{F.\,Ruiz Blasco} [Indiana Univ.\ Math.\ J.\ 52, No.\,5, 1215--1230 (2003; Zbl 1098.46023)] used, in order to study a~limiting Sobolev embedding, a slightly different rearrangement estimate, namely the Kolyada inequality \[ f^{**}(t)-f^*(t)\leq C_n t^{\frac1n}|\nabla f|^{**}(t), \] which is a good point of departure for Sobolev embeddings as long as the lower Boyd index of the domain space is kept away from~\(1\). To prove this inequality, symmetrization principles in the spirit of Pólya and Szegő have been used. The principal result of the paper under review shows that truncation implies symmetrization inequalities. The authors show that a~Sobolev inequality yields a~symmetrization principle from which all Sobolev inequalities follow. The main theorem states that the weak Gagliardo--Nirenberg embedding \(W_0^{1,1}(\mathbb R^ n)\subset L^{n/(n-1),\infty}(\mathbb R^ n)\) is self-improved to the strong one, \(W_0^{1,1}(\mathbb R^ n)\subset L^{n/(n-1),1}(\mathbb R^ n)\), and each of them is equivalent to the oscillation inequality \[ \int_0^ t s^{-\frac1n}\left[f^{**}(t)-f^*(t)\right]\,ds\leq C_n\int_0^ t| \nabla f| ^*(s)\,ds. \] The result applies to optimal Sobolev embeddings. It can also be used to simplify proofs of known results on sharp compact Sobolev embeddings, as long as the lower Boyd index of the domain space is kept away from the critical value \((k-1)/n\).