Hardy’s inequality and the boundary size (Q4787501)

This paper establishes a self-improving property of the Hardy inequality and gives an estimate on the size of the boundary of a domain supporting a Hardy inequality. Let \(\Omega \) be a domain in \({\mathbb R}^{n}\) and suppose that the Hardy inequality \[ \int_{\Omega }\left|u(x)\right|^{p}d(x,\partial \Omega)^{-p}dx\leq C\int_{\Omega }\left|\nabla u(x)\right|^{p}dx, \] holds with a fixed \(p>1\) and \(C>0\) for all \(u\in C_{0}^{1}(\Omega).\) Theorem 1 shows the existence of a constant \(\varepsilon =\varepsilon (p,n,C)>0\) such that: i) either \(\dim _{H}(\partial \Omega)\geq n-p+\varepsilon \) and the Hardy inequality holds (with possibly a different constant) for all \(p-\varepsilon <q<p+\varepsilon \), or ii) \(\dim _{M}(\partial \Omega)\leq n-p-\varepsilon \) and the Hardy inequality holds (with possibly a different constant) for all \(1\leq q<p+\varepsilon .\) Here \( \dim _{H}\) refers to the usual Hausdorff dimension and \(\dim _{M}\) to the Minkowski dimension (arising from the use of covers with ball of equal radii). In what follows \(X\) denotes a proper metric space (i.e., each closed ball is compact) equipped with a regular Borel measure \(\mu \) such that \(\mu (2B)\leq C_{d}\mu (B)\) for all balls \(B\). Here \(2B\) means \(B(x,2r)\) provided that \(B=B(x,r).\) Theorem 2 concerns with the case when \(X\) supports a \(p_{0}\) -Poincaré inequality, i.e., there are constants \(C^{\prime }\) and \( \lambda \) such that \[ \mu (B)^{-1}\int_{\Omega }\left|u-u_{B}\right|d\mu \leq C^{\prime } \text{ diam} (B)\cdot \left( \mu (\lambda B)^{-1}\int_{\lambda B}g^{p}dx\right) ^{1/p} \] whenever \(B\) is a ball, \(u\) is continuous on \(\lambda B\) and \(g\) is a upper gradient of \(u\) on \(\lambda B.\) Here \(u_{B}\) is the integral average of \(u\) over \(B.\) These spaces cover the case of the Heisenberg and Carnot groups. Let \(\Omega \subset X\) be a proper subdomain and suppose that the Hardy inequality \[ \int_{\Omega }\left|u(x)\right|^{p}d(x,\partial \Omega)^{-p}dx\leq C_{p}\int_{\Omega }g^{p}dx \] holds with a fixed \(p>p_{0}.\) Theorem 2 asserts the existence of a constant \[ \varepsilon =\varepsilon (p,p_{0},C_{d},C^{\prime },\lambda ,C_p)>0 \] such that the last inequality holds (with possibly a different constant) for all \( p-\varepsilon <q<p+\varepsilon \).