ASYMPTOTIC BEHAVIOUR OF MINIMIZING SEQUENCES FOR HARDY'S INEQUALITY (Q4513341)

Let \(\Omega \) be a bounded domain in \(\mathbb{R}^{N}\) of class \(C^{2}\) and \(p\in (1,\infty).\) Let \(\eta \in \text{Lip}(\overline{\Omega })\) and \(\eta =0\) on \(\partial \Omega \), and let: \[ \mu _{p,\eta }(\Omega)=\inf_{u\in W_{0}^{1,p}(\Omega)}\frac{\int_{\Omega }|\nabla u|^{p} dx-\int_{\Omega }\eta |u/\delta |^{p} dx}{\int_{\Omega }|u/\delta |^{p} dx}\tag{1} \] where \(\delta (x)=\text{dist}(x,\partial \Omega)\). It is known that \(\mu _{p,\eta }(\Omega)\) for \(\eta =0\) is the best constant in the corresponding Hardy's \(N\)-dimensional inequality [see \textit{B. Opic} and \textit{A. Kufner}, Hardy-type inequalities, Pitman Research Notes in Mathematics, 219, Longman (1990; Zbl 0698.26007), and \textit{M. Marcus, V. J. Mizel} and \textit{Y. Pinchover}, Trans. Am. Math. Soc. 350, No. 8, 3237-3255 (1998; Zbl 0917.26016)]. Furthermore, the infimum in (1) is achieved if and only if \(\mu _{p,\eta }(\Omega)<c_{p}=(1-1/p)^{p}\) [see \textit{M. Marcus} and \textit{I. Shafrir}, Ann. Sc. Norm. Pisa (to appear)]. In the paper, the case when \(\mu _{p,\eta }(\Omega)=c_{p}\) is considered replacing (1) by the following one: \[ \upsilon _{\varepsilon ,p,\eta }(\Omega)=\inf_{u\in W_{0}^{1,p}(\Omega)} \frac{\int_{\Omega }|\nabla u|^{p} dx-\int_{\Omega }\eta |u/\delta |^{p} dx} {\int_{\Omega }|u|^{p}/\delta ^{p-\varepsilon } dx}.\tag{2} \] Denoting by \(u_{\varepsilon}\) (for each \(\varepsilon >0)\) the unique positive minimizer of (2) which is normalised by \[ \int_{\Omega }|u_{\varepsilon }|^{p}dx=1, \] it is proved the strong convergence of \(\{u_{\varepsilon }\}\) to a nontrivial limit \(u_{*}\) in \(W_{0}^{1,q}(\Omega)\), \(q\in [1,p),\) which satisfies the corresponding Euler-Lagrange equation: \[ -\text{div}(|\nabla u_{*}|^{p-2}\nabla u_{*})=(\mu _{p,\eta }(\Omega)+\eta) \frac{u_{*}^{p-1}}{\delta^{p}}\text{ in }\Omega , \] and the following pointwise estimate: \[ u_{*}(x)\leq C\delta (x)^{1-1/p}(1+|\log \delta (x)|^{2/p}). \] Moreover, in the case when \(\Omega \) is a bounded convex domain of class \(C^{2} \) and \(\eta =0\) the lower pointwise estimate for \(u_{*}\) is proved too: \[ C^{-1}\delta (x)^{1-1/p}(1+|\log \delta (x)|^{2/p})\leq u_{*}(x)\text{ in } \Omega . \]