Optimal regularity for the Signorini problem and its free boundary (Q284620): Difference between revisions

In this very interesting paper the author proves optimal regularity for minimizers of the Signorini problem for the Lame system and regularity of the free boundary close to points where \(\mathbf{u}\) is not degenerate. More precisely, if \(\mathbf{u}=(u^1,\dots, u^n)\in W^{1,2}(B_1^+: \mathbb R^n)\) minimizes \[ J(\mathbf{u}):=\int_{B_1^+}|\nabla\mathbf{u}+\nabla^{\perp}\mathbf{u}|^2+\lambda\operatorname{div}(\mathbf{u})^2 \] in the convex set \[ K=\{\mathbf{u}=(u^1,\dots, u^n)\in W^{1,2}(B_1^+:\mathbb R^n): u^n\geq 0 \text{ on }\prod, \mathbf{u}= f\in C^\infty(\partial B_1) \text{ on } (\partial B_1)^+\}, \] where \(\prod=\{x:x_n=0\}\) and \(\lambda \geq 0\), then \(u\in C^{1,1/2}(B_{1/2}^+)\). Moreover, the free boundary, which is defined as \[ \Gamma_{\mathbf{u}}:=\partial \{x:u^n(x)=0, \quad x_n=0\}\cap B_1 \] is a \(C^{1,\alpha}\) graph close to points where \(\mathbf{u}\) is not degenerate. Regarding the existing machinery developed to solve related problems, in [\textit{I. Athanasopoulos} and \textit{L. A. Caffarelli}, J. Math. Sci., New York 132, No. 3, 274--284 (2006; Zbl 1108.35038); translation from Zap. Nauchn. Semin. POMI 310, 49--66, 226 (2004)] the authors prove that minimizers of the thin obstacle problem for the Laplacian are in \(C^{1,1/2}\), which is the optimal regularity. Their proof is based on the Bernstein technique, a monotonicity formula and an iteration. As the maximum principle is not true for the Lame system, one cannot replicate this argument for the vectoral Signorini problem. Furthermore, the structure of the Lame system also does not allow to derive the essential monotonicity formula used. In [\textit{I. Athanasopoulos} et al., Am. J. Math. 130, No. 2, 485--498 (2008; Zbl 1185.35339)], the authors use comparison and boundary comparison principles to show that concerning the thin obstacle problem for the Laplacian, the free boundary \(\partial\{u>0\}\cap \prod\) is \(C^{1,\alpha}\) in a small neighborhood around free boundary points \(x^0\) where \(\sup_{B_r(x^0)}|u|\approx r^{1+\beta}\), for any \(\beta<1\). The use of comparison principles makes it impossible to apply their technique to the vectoral Signorini problem. In this paper, the author develops a regularity theory for free boundaries that is independent on maximum principles. Instead, the author relies on the blow-up method, the Liouville Theorem and linearization to tackle the problem at hand. The first part of the paper is dedicated to showing that, close to free boundary points where \(\mathbf{u}\) grows like \(r^{1+\alpha}\), for some \(\alpha<1\), the free boundary \(\Gamma_u\) is actually flat, i.e., in a small ball \(B_r\cap\prod\) the free boundary is contained in a strip of width \(\sigma(r)r\), for some modulus of continuity \(\sigma\). The proof uses a result by \textit{M. Benedicks} [Ark. Mat. 18, 53--72 (1980; Zbl 0455.31009)] (itself connected to a deep result by \textit{S. Friedland} and \textit{W. K. Hayman} [Comment. Math. Helv. 51, 133--161 (1976; Zbl 0339.31003)]) that states that the set of positive harmonic functions vanishing on a part of \(\prod\) that have zero Neumann boundary data on the rest of \(\prod\) is one dimensional. Using this result one deduces that the tangential derivatives of the blow-up of a solution \(\mathbf{u}\) are all multiples of each other, implying that the blow-up of \(\mathbf{u}\) depends only on two directions. Using the Liouville Theorem one can then classify such solutions and calculate the asymptotic profile of solutions close to points where \(\sup_{B_r}| \mathbf{u}| \approx r^{1+\alpha}\). Since the growth of the asymptotic profile is \(r^{3/2}\), the author can conclude hat \(u\in C^{1,\beta}(B_{1/2}^+)\) for each \(\beta<1/2\). It is also proved that the asymptotic of the solution is uniquely determined, at least at points of lowest regularity, which makes a linearization argument at such points possible. The second part of the paper includes the proof that \(u\in C^{1,1/2}\), i.e., \[ \|\mathbf{u}\|_{C^{1,1/2}(B_{1/2}^+)}\leq C \|\mathbf{u}\|_{L^2(B_1^+)}. \] It is also proved, using linearization and a flatness improvement argument, that the free boundary is \(C^{1,\alpha}\) close to points where \(\mathbf{u}\) has a certain given asymptotic profile.