On the fine structure of the free boundary for the classical obstacle problem (Q1717656)

In this interesting paper, the authors study the fine structure of the free boundary for the classical obstacle problem \[ \Delta u = \chi_{\{u>0\}}, \quad u\geq 0 \text{ in } B_1\subset \mathbb R^n. \] As shown by \textit{L. A. Caffarelli} [J. Fourier Anal. Appl. 4, No. 4--5, 383--402 (1998; Zbl 0928.49030)], free boundary points $\partial \{u>0\}$ are divided into regular and singular points depending on the types of the blow-ups: \[ x_0 \text{ is a regular point } \Leftrightarrow \frac{u(x_0+rx)}{r^2} \rightarrow \frac{1}{2}\max\{e\cdot x, 0\} \] for some direction $e=e_{x_0}\in \mathbb{S}^{n-1}$, and \[ x_0\text{ is a singular point } \Leftrightarrow \frac{u(x_0+rx)}{r^2}\rightarrow p_{x_0}(x)=\frac{1}{2}x\cdot Ax \] for some symmetric nonnegative definite matrix $A=A_{x_0}$ with $\text{tr}(A)=1$. The free boundary is a real analytic hypersurface near regular points. For the singular points the $m$-th stratum \[ \Sigma_m:=\{x_0: \text{singular point with } \dim \ker(A_{x_0})=m\}, \ m\in \{0,\dots, n-1\}, \] is locally contained in a $C^1$ $m$-dimensional manifold. This paper under review provides finer structure results for the singular set. More precisely, they show that $\ast$ for $n=2$ the singular set is locally contained in a $C^2$ curve; $\ast$ for $n\geq 3$ each $\Sigma_m$, $m=1,\dots, n-1$, is locally contained in a $C^{1,1}$ $m$-dimensional manifold up to the presence of some ``anomalous'' points of Hausdorff dimension less or equal to $m-1$. $\ast$ $\Sigma_{n-1}$ is locally contained in a $C^{1,\alpha_0}$ $(n-1)$-dimensional manifold, and $\Sigma_m$ for each $m=1,\dots, n-2$ is locally contained in a $C^{1,\log^{\epsilon_0}}$ $m$-dimensional manifold. They also construct example of solutions for $n\geq 3$, which prove that anomalous points may exist and their bound on the Hausdorff dimension is optimal. The classical results give $u(x_0+x)=p_{x_0}(x)+o(|x|^2)$ around the singular point $x_0$, and the goal would be to improve the convergence rate $o(|x|^2)$ into a quantitative bound. One key observation in the paper is that $w:=u(x_0+\cdot) -p_{x_0}$ satisfies an Almgren's monotonicity formula, which allows to carry out blow-up analysis and also to conclude $u(x_0+x)=p_{x_0}(x)+O(|x|^{\lambda_{\ast}})$, where $\lambda_\ast$ is the Almgren's frequency for $w$ at $0$. To characterize possible frequencies $\lambda_\ast$ and blow-up limits, they distinguish between the two cases $m=n-1$ and $m\leq n-2$. When $m=n-1$, they show that blow-up limits are solutions to the Signorini problem (or the \textit{thin} obstacle problem) and $\lambda_\ast \geq 2+\alpha_0$ for some $\alpha_0\in (0,1)$. When $m\leq n-2$, blow-up limits $q$ are homogeneous harmonic functions with homogeneity $\lambda_\ast \in \{2,3, \dots\}$. Furthermore, if $\lambda_\ast=2$ there is a strong relation between $p_{x_0}$ and $q$, which is a crucial ingredient in showing $\lambda_\ast=2$ cannot happen for ``too many'' free boundary points. The estimates on the Hausdorff dimension of the anomalous set $\Sigma_m^a:=\{x_0\in \Sigma_m: \lambda_\ast(x_0)<3\}$, $m\in\{1,\dots, n-1\}$, are based on a Federer-type dimension-reduction principle, the known knowledge of the structure of the free boundary for the Signorini problem (when $m=n-1$), and the finer characterization of the blow-up limits $q$ when $\lambda_\ast=2$ (when $m\leq n-2$).