Sharp gradient stability for the Sobolev inequality (Q2081378)

Suppose that \(n \geq 2\), \(1 < p < n\), \(Du = (\partial_{x_1}u, \ldots, \partial_{x_n}u)\). \(\dot{W}^{1,p}(\mathbb R^n)\) is the homogeneous Sobolev space of order \(1\) on \(L^p\) defined as the closure of \(C^{\infty}_c(\mathbb R^n)\) in the norm \[ \| u\|_{ \dot{W}^{1,p}(\mathbb R^n)} = \Bigl( \int_{\mathbb R^n} |Du|^p \, dx \Bigr)^{1/p}. \] Sobolev's inequality for \(u \in \dot{W}^{1,p}(\mathbb R^n)\) is \(\| Du\|_p \geq S\| u \|_{p^*}\). where \(p^{*} = \frac{np}{n - p}\) and the optimal value of \(S\) is known. \(\mathcal{M}\) is the \(n +2\) dimensional manifold of functions \[ v_{a, b, x_0} = \frac{a}{ (1 + b |x - x_0|^{\frac{p}{p - 1} } )^{ \frac{n - p}{np } } } , \quad a \in \mathbb R \setminus \{ 0\}, \ b>0, \ x_0 \in \mathbb R^n. \] \(\mathcal{M}\) is the space of all weak solutions to the equation \[ -\Delta_p v = S^p \| v\|^{p - p^*}_{L^{p^*}(\mathbb R^n)} |v|^{p^* - 2} v \ \text{ in } \mathbb R^n \] that do not change in sign, where \(S\) is the optimal Sobolev constant and \[ -\Delta_p v = \operatorname{div}(|Dv|^{p - 2} Dv). \] It has also been shown that \(\mathcal{M}\) coincides with the space of all the extremal functions in the Sobolev inequality, \[ \|D v \|_{L^p(\mathbb R^n)} = S \| v\|_{L^{p^*}(\mathbb R^n)} \text{ for all } v \in \mathcal{M}. \] The \(p\)-Sobolev deficit, which vanishes only on \(\mathcal{M}\), is \[ \delta(u) = \frac{ \|D u \|_{L^p(\mathbb R^n)} }{ \| u \|_{L^{p^*}(\mathbb R^n)} } -S \] \textit{H. Brézis} and \textit{E. H. Lieb} [J. Funct. Anal. 62, 73--86 (1985; Zbl 0577.46031)] asked if it was possible for \(p = 2\) to estimate the deficit from below by some measure of the distance to the manifold \(\mathcal{M}\). This was proved by \textit{G. Bianchi} and \textit{H. Egnell} [J. Funct. Anal. 100, No. 1, 18--24 (1991; Zbl 0755.46014)] who showed the existence of a constant \(c(n)>0\) such that \[ \delta (u) \geq c \inf_{v \in \mathcal{M} } \left( \frac{ \| Du - Dv\|^2_{L^2(\mathbb R^n)} }{ \| Du\|^2_{L^2(\mathbb R^n)} } \right)^2 \text{ for all } u \in \dot{W}^{1,2}(\mathbb R^n). \] The methods of Bianchi and Egnell do not apply for other values of \(p\) as they were heavily dependent on the Hilbert space structure. A number of partial results have been proved. A stability result was proved in [\textit{A. Cianchi} et al., J. Eur. Math. Soc. (JEMS) 11, No. 5, 1105--1139 (2009; Zbl 1185.46025)] with a distance of \[ \inf_{v \in \mathcal{M} } \left( \frac{ \| u - v\|_{L^{p^*}(\mathbb R^n)} }{ \| u \|_{L^{p^*}(\mathbb R^n)} } \right)^{\alpha} \text{ for all } u \in \dot{W}^{1,p}(\mathbb R^n). \] with an exponent \(\alpha_{CFMP}\) that is not sharp. (Its precise value is given in the paper.) This gave a result valid for all~\(p\). Eventually, for \(p > 2\), \textit{A. Figalli} and \textit{R. Neumayer} [J. Eur. Math. Soc. (JEMS) 21, No. 2, 319--354 (2019; Zbl 1417.46023)] gave a lower bound of \[ \delta(u) \geq \inf_{v \in \mathcal{M} } \left( \frac{ \| Du - Dv\|_{L^p(\mathbb R^n)} }{ \| Du \|_{L^p(\mathbb R^n)} } \right)^{\alpha^{\prime}} \text{ for all } \ u \in \dot{W}^{1,p}(\mathbb R^n), \] with \(\alpha^{\prime} = p \alpha_{CFMP}\). \textit{R. Neumayer} [Calc. Var. Partial Differ. Equ. 59, No. 1, Paper No. 25, 8 p. (2020; Zbl 1440.46033)] was able to extend this to all \(p\), \(1 < p < n\), and in the present paper, the authors show that for \(1 < p < n\), this lower bound on \(\delta(u)\) holds with exponent \(\alpha^{\prime} = \max\{2,p\}\), which is proved to be sharp. The authors also give a weighted Hardy-Sobolev-Poincaré inequality. For functions \(u \in \dot{W}^{1,p}(\mathbb R^n; |x|^{-\alpha})\), they prove that there is a constant \(C(n, p, \alpha)\) such that for any \(R > 1\), \[ \int_{\mathbb R^n \setminus B(0,R)} |u|^p |x|^{-\alpha} \, dx \leq C \int_{\mathbb R^n \setminus B(0,R)} | Du|^p |x|^{-\alpha} \, dx. \] (The reviewer has made an educated guess as to \(D\) since it is nowhere defined in the paper. In [Figalli and Neumayer, loc. cit.], \(\nabla\) was used where \(D\) is now used, (and a slightly different definition of \(\delta\) was made), but as defined by the reviewer, \(|Du| = | \nabla u |\).)