Characterization of the critical Sobolev space on the optimal singularity at the origin (Q971829): Difference between revisions

The authors study weighted inequalities with homogeneous weights for the critical Sobolev space \(H^{\frac{n}{p},p}\). This is the critical space in the sense that the embedding \(H^{\frac{n}{p},p} \hookrightarrow L^q\) holds for all \(p \leq q < \infty\), but fails for \(q = \infty\). Functions in the critical Sobolev space can have a singularity, such as a logarithmic singularity, suggesting that \(L^{\infty}\) should be replaced by a \(BMO\) norm. Some such results are known. \textit{T.\,Ozawa} [J.~Funct.\ Anal.\ 127, No.\,2, 259--269 (1995; Zbl 0846.46025)] gave a Gagliardo-Nirenberg type estimate near the critical case \[ \| u \|_{L^q(\mathbb R^n)} \leq Cq^{\frac{1}{p^{\prime} }} \| u \| _{L^p(\mathbb R^n)}^{\frac{p}{q}} \| (- \Delta)^{\frac{n}{2p}} u \|_{L^p(\mathbb R^n)}^{1 - \frac{p}{q}}. \] The authors study when the bound on the left hand side remains valid with a weight \[ \| u \|_{L^q(\mathbb R^n; \frac{dx}{w(x)})} = \left( \int_{\mathbb R^n} |u(x)|^q \frac{dx}{w(x)} \right)^{\frac{1}{q}}, \quad 1 < q < \infty, \] for a homogeneous weight \(w(x) = |x|^s\). Their main result is a bound of the above type which controls the type of singularity that can be present. Let \(n \in \mathbb N\), \(1 < p < \infty\). There are constants \(\tilde{p} \in (p, \infty)\), \(C>0\), which depend only on \(n\) and \(p\), such that the inequality \[ \| u \|_{L^q(\mathbb R^n); \frac{dx}{|x|^s}} \leq C\left( \frac{1}{n - s} \right)^{\frac{1}{q} + \frac{1}{p^{\prime}}} q^{\frac{1}{p^{\prime} }} \| u \| _{L^p(\mathbb R^n)}^{\theta} \| (- \Delta)^{\frac{n}{2p}} u \|_{L^p(\mathbb R^n)}^{1 - \theta} \] holds for all \(u \in H^{\frac{n}{p},p}\), \(0 \leq s < n\), \(\tilde{p} \leq q < \infty\), \(\theta = \frac{(n - s)p}{nq} \in (0,1)\). If \(n \geq 2\), \(\frac{n}{n - 1} \leq p < \infty\), the result is optimal in the sense that the powers of \(\frac{1}{n - s}\) and \(q\) cannot be replaced by smaller powers as \(s \uparrow n\) or as \(q \to \infty\). (For example, one cannot replace the power of \(q\) by \(q^{\frac{1}{p^{\prime}} - \varepsilon}\) for any \(\varepsilon > 0\) nor that of \(\frac{1}{n - s}\) by \(\frac{1}{q} + \frac{1}{p^{\prime}} - \varepsilon\).) The authors remark that their inequality, without the growth orders of \(s\) and \(q\), was proved by \textit{L.\,Caffarelli, R.\,Kohn} and \textit{L.\,Nirenberg} [Compos.\ Math.\ 53, 259--275 (1984; Zbl 0563.46024)] when \(\frac{n}{p} = 1\). They give a Trudinger type estimate for an Orlicz weight on the left hand side replacing the \(L^q\) norm, and a result in the critical case \(s = n\) with a power of a logarithm times the expected weight \(|x|^n\). The precise details of these two cases can be seen in Corollary 1.3 and Theorem 1.5 of the paper, which also contains an excellent survey of prior results.