On a geometric inequality related to fractional integration (Q1746596)

The following inverse Hardy-Littlewood-Sobolev inequality is known: For \(0<p,q<1\), and all functions \(f\in L^{p}(\mathbb{R})\) and \(g\in L^{q}(\mathbb{R})\), one has \[ \|f\|_{p} \|g\|_{q}\leq C_{p,q }\int_{\mathbb{R}}\int_{\mathbb{R}} |f(s)g(t)| |s-t|^{\frac{1}{p}+\frac{1}{q}-2}\,ds\,dt, \] for some constant \(C_{p,q}\). In this paper, the author considers the variant of this problem where one replaces the integral form on the right-hand side by its ``sup'' form. The author investigates corresponding results in the setting of a general metric space and searches for multilinear analogues of this type of inequalities. Let \((M, d)\) be a metric space and \(\mu\) a \(\sigma\)-finite nonnegative Borel measure on~\(M\). Let us consider the following condition: (i) For any \(x\in M\), \(r > 0\), one has \[ \mu(B(x,r))\leq C_{\alpha}r^{\alpha} \] with a finite constant \(C_{\alpha}\), \(\alpha > 0\). And for \(f\), \(g\) nonnegative measurable functions on~\(M\), consider the condition \[ \text{(ii)}\qquad\qquad \|f\|_{L^{p}(d\mu)}\cdot \|g\|_{L^{q}(d\mu)}\leq C_{p,q,\gamma}\sup_{s,t}f(s)g(t)\,d(s,t)^{\gamma}, \] with a finite constant~\(C_{p,q,\gamma}\) independent of the functions~\(f\), \(g\). Then, the first main result is as follows \(\bullet\) Let \((M,d)\) be a metric space and \(\mu\) a \(\sigma\)-finite, nonnegative Borel measure on~\(M\). \text{(a)} If condition (i) holds, then (ii) holds for all nonnegative functions \(f\in L^{p} (d\mu)\), \(g\in L^{q}(d\mu)\) for all \(0<p,q<\infty\), \(\gamma\) such that \(\gamma=\alpha\left(\frac{1}{p}+\frac{1}{q}\right)\). \text{(b)} If condition (ii) holds for all nonnegative functions \(f\in L^{p} (d\mu)\), \(g\in L^{q}(d\mu)\) for some \(p,q > 0\), \(\gamma > 0\), then condition (i) holds for all \(\alpha\) such that \(\alpha = \gamma \left(\frac{1}{p}+\frac{1}{q}\right)\). For the special case of Euclidean space, the author obtaines \(\bullet\) Let \(f\), \(g\) be measurable functions defined on~\(\mathbb{R}^{n}\) with Lebesgue measure, then for all \(0< p,q <\infty\), \(\gamma >0\) such that \( \gamma =n\left(\frac{1}{p}+\frac{1}{q}\right)\), \[ \|f\|_{L^{p}(\mathbb{R}^{n}} \|g\|_{L^{q}(\mathbb{R}^{n}}\leq C_{p,q,n }\sup_{s,t} f(s)g(t) |s-t|^{\gamma}. \] On the other hand, it is not true that \[ \|f\|_{L^{p}(\mathbb{R}^{n})}\|g\|_{L^{\infty}(\mathbb{R}^{n})}\lesssim \sup_{s,t}f(s)g(t)|s-t|^{\frac{n}{p}} \] holds for all \(f\in L^{p}(\mathbb{R}^{n})\), \(g\in L^{\infty}(\mathbb{R}^{n})\). As it concerns the multilinear analogues of the previous results although they are true in the context of a real finite-dimensional Hilbert space we quote the determinant multilinear inequality in Euclidean spaces. \(\bullet\) Let \(f_{j}\in L^{p_{j}} (\mathbb{R}^{n})\) with Lebesgue measure, then \[ \prod^{n+1}_{j=1}\|f_{j}\|_{p_{j}} \leq C_{n,p_{j}}\sup_{y_{j}}\prod^{n+1}_{j=1}f_{j} (y_{j})\det (y_{1},\dotsc,y_{n+1})^{\gamma} \] holds, if and only if \(p_{j}\) satisfy \[ \frac{1}{p_{j}}< \frac{\gamma}{n}\text{ for all }1\leq j\leq n+1\text{ and }\gamma=\sum^{n+1}_{j=1}\frac{1}{p_{j}}. \] The author also considers a second class of multilinear inequalities where a product appears instead of a determinant. Furthermore, the following result is proved \(\bullet\) Let \(r_{ij}>0\) and \(r_{ij} = r_{ji}\). Let \(f_{j}\) be nonnegative measurable functions defined on \(\mathbb{R}^{n}\), \(1 \leq j \leq N\). Then \[ \prod^{N}_{j=1}\|f_{j}\|_{p_{j}} \leq C_{p_{j},r_{ij},n,N}\sup_{y_{j}}\prod^{N}_{j=1}f_{j}(y_{j})\prod_{1\leq i <j\leq N}|y_{i}-y_{j}|^{r_{ij}}, \] holds for any \(0<p_{j} <\infty\) satisfying \[ \frac{1}{p_{j}}< \frac{1}{2n}\sum_{i\neq j}r_{ij},\quad \sum^{N}_{j=1}\frac{1}{p_{j}}=\frac{1}{n}\sum_{1\leq i <j\leq N} r_{ij}. \] At the end, in the Euclidean space setting, the author establishes the existence of extremal functions for the geometric inequality bilinear form that appears in the first quoted result, when \(p= q\), and for the multilinear determinant form when \(p_{j}=p\), \(j=1,\dotsc,n+1\). Also, corresponding conformally equivalent formulations in the unit sphere space \(\mathbb{S}^{n}\) and in the hyperbolic space~\(\mathbb{H}^{n}\) are given.