Abstract
In the classical obstacle problem, the free boundary can be decomposed into “regular” and “singular” points. As shown by Caffarelli in his seminal papers (Caffarelli in Acta Math 139:155–184, 1977; J Fourier Anal Appl 4:383–402, 1998), regular points consist of smooth hypersurfaces, while singular points are contained in a stratified union of \(C^1\) manifolds of varying dimension. In two dimensions, this \(C^1\) result has been improved to \(C^{1,\alpha }\) by Weiss (Invent Math 138:23–50, 1999). In this paper we prove that, for \(n=2\) singular points are locally contained in a \(C^2\) curve. In higher dimension \(n\ge 3\), we show that the same result holds with \(C^{1,1}\) manifolds (or with countably many \(C^2\) manifolds), up to the presence of some “anomalous” points of higher codimension. In addition, we prove that the higher dimensional stratum is always contained in a \(C^{1,\alpha }\) manifold, thus extending to every dimension the result in Weiss (1999). We note that, in terms of density decay estimates for the contact set, our result is optimal. In addition, for \(n\ge 3\) we construct examples of very symmetric solutions exhibiting linear spaces of anomalous points, proving that our bound on their Hausdorff dimension is sharp.
Similar content being viewed by others
Notes
In many textbooks, the definition of \({\mathcal {H}}^{\beta }_\delta \) includes a normalization constant chosen so that the Hausdorff measure of dimension k coincides with the standard k-dimensional volume on smooth sets. However such normalization constant is irrelevant for our purposes, so we neglect it.
Otherwise there would be a sequence of points \(z_{\ell } \in r_{k_\ell }^{-1} E \cap \overline{ B_{1/2}}{\setminus } \bigcup _{i=1}^N \cup B_i\), and hence their limit z—up to a subsequence—would satisfy at the same time \(z\in {\mathcal {A}}\) and \(z\in \overline{B_{1/2}} {\setminus } \bigcup _{i=1}^N {\hat{B}}_i\), a contradiction.
References
Athanasopoulos, I., Caffarelli, L.: Optimal regularity of lower dimensional obstacle problems. Zap. Nauchn. Sem. S.-Peterburg. Otdel. Mat. Inst. Steklov. (POMI) 310 (2004), Kraev. Zadachi Mat. Fiz. i Smezh. Vopr. Teor. Funkts. 35 [34], 49–66, 226; translation in J. Math. Sci. (N. Y.) 132 (2006), no. 3, 274–284
Athanasopoulos, I., Caffarelli, L., Salsa, S.: The structure of the free boundary for lower dimensional obstacle problems. Am. J. Math. 130, 485–498 (2008)
Brézis, H., Kinderlehrer, D.: The smoothness of solutions to nonlinear variational inequalities. Indiana Univ. Math. J. 23, 831–844 (1973/74)
Caffarelli, L.: The regularity of free boundaries in higher dimensions. Acta Math. 139, 155–184 (1977)
Caffarelli, L.: The obstacle problem revisited. J. Fourier Anal. Appl. 4, 383–402 (1998)
Caffarelli, L., Cabré, X.: Fully Nonlinear Elliptic Equations. American Mathematical Society Colloquium Publications, vol. 43. American Mathematical Society, Providence, RI (1995)
Caffarelli, L., Rivière, N.: Smoothness and analyticity of free boundries in variational inequalities. Ann. Scuola Norm. Sup. Pisa Cl. Sci. 3, 289–310 (1976)
Caffarelli, L., Rivière, N.: Asymptotic behavior of free boundaries at their singular points. Ann. Math. 106, 309–317 (1977)
Colombo, M., Spolaor, L., Velichkov, B.: A logarithmic epiperimetric inequality for the obstacle problem (preprint). arXiv:1708.02045v1
Fefferman, C.: Extension of \(C^{m,\omega }\)-smooth functions by linear operators. Rev. Mat. Iberoam. 25(1), 1–48 (2009)
Focardi, M., Spadaro, E.: On the measure and structure of the free boundary of the lower dimensional obstacle problem (preprint). arXiv:1703.00678v2
Garofalo, N., Petrosyan, A.: Some new monotonicity formulas and the singular set in the lower dimensional obstacle problem. Invent. Math. 177, 415–461 (2009)
Gilbarg, D., Trudinger, N.S.: Elliptic Partial Differential Equations of Second Order. Reprint of the 1998 edition. Classics in Mathematics. Springer, Berlin (2001)
Kinderlehrer, D., Nirenberg, L.: Regularity in free boundary problems. Ann. Sc. Norm. Sup. Pisa 4, 373–391 (1977)
Monneau, R.: A brief overview on the obstacle problem. In: European Congress of Mathematics, Vol. II (Barcelona, 2000), volume 202 of Progr. Math., pp. 303–312. Birkhäuser, Basel (2001)
Monneau, R.: On the number of singularities for the obstacle problem in two dimensions. J. Geom. Anal. 13, 359–389 (2003)
Petrosyan, A., Shahgholian, H., Usaltseva, N.: Regularity of Free Boundaries in Obstacle-Type Problems. Graduate Studies in Mathematics, vol. 136. American Mathematical Society, Providence, RI (2012)
Sakai, M.: Regularity of a boundary having a Schwarz function. Acta Math. 166(3–4), 263–297 (1991)
Sakai, M.: Regularity of free boundaries in two dimensions. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) 20(3), 323–339 (1993)
Schaeffer, D.: Some examples of singularities in a free boundary. Ann. Scuola Norm. Sup. Pisa 4, 131–144 (1976)
Simon, L.: Lectures on geometric measure theory. In: Proceedings of the Centre for Mathematical Analysis, Australian National University, vol. 3. Australian National University, Centre for Mathematical Analysis, Canberra (1983)
Weiss, G.: A homogeneity improvement approach to the obstacle problem. Invent. Math. 138, 23–50 (1999)
Yeressian, K.: Obstacle problem with a degenerate force term. Anal. PDE 9(2), 397–437 (2016)
Acknowledgements
both authors are supported by ERC Grant “Regularity and Stability in Partial Differential Equations (RSPDE)”. The authors thank the anonymous referees for useful comments on a preliminary version of the manuscript.
Author information
Authors and Affiliations
Corresponding author
Appendix A: Examples of \((m-1)\)-dimensional anomalous sets \(\Sigma ^a_m\)
Appendix A: Examples of \((m-1)\)-dimensional anomalous sets \(\Sigma ^a_m\)
In this “Appendix”, for any \(1\le m \le n-2\) we construct an example of solution to the obstacle problem in \(\mathbb {R}^n\) for which the anomalous set \(\Sigma _m^a\) is \((m-1)\)-dimensional. The existence of such examples shows that the assertion \(\mathrm{dim}_\mathcal {H}(\Sigma ^a_m) \le m-1\) in Theorem 1.1(b) is optimal.
These solutions are constructed as follows: given \(1\le m \le n-2\) we consider functions
which are independent of the first \((m-1)\) variables and are axially symmetric with respect to the last \(m-n\) variables. Then our goal is to find solutions to the obstacle problem which are of the form (A.1) and for which all points in the \((m-1)\) dimensional affine space
are anomalous points in \(\Sigma _m^a\). To build these examples, we rely on some ideas introduced in [23].
Fix \(\phi :[-1,1] \rightarrow \mathbb {R}\) a nonnegative \(C^2\) function satisfying
Then, for any real number \(k>0\) we consider the solution \(u^k\ge 0\) to the obstacle problem in \((-1,1)\times (0,1)\subset \mathbb {R}^2\)
In other words, \(\frac{1}{k} u^k(x_{m}, \sqrt{x_{m+1}^2+\cdots + x_{n}^2})\) is a solution of the classical obstacle problem in the cylinder \(\mathbb {R}^{m-1} \times (-1,1)\times B_1^{(n-m)}\) satisfying the symmetries in (A.1).
Note that, when we think of this obstacle problem as a two dimensional problem (in the variables z, r), we do not prescribe “boundary” values at \(r=0\) since this line has zero capacity for the operator \(\mathrm{div}( r^{n-m-1}\nabla \,\cdot \,)\) when \(n-m-1\ge 1\) (this is equivalent to saying that the set \(\{x_{m+1} =\cdots =x_n=0 \}\) has zero harmonic capacity in \(\mathbb {R}^n\)).
We now claim that:
-
(i)
\(u^k\) is even z, namely \(u^k(z,r)=u^k(-z,r)\);
-
(ii)
\(\partial _z u^k \le 0\) in \((0,1)\times (0,1)\);
-
(iii)
the set \(\{u^k>0\}\) is convex in the direction z, and is symmetric with respect to \(z=0\).
Indeed, (i) follows from the symmetry of the boundary data (and the uniqueness of solution to the obstacle problem).
To show (ii) we consider the open set \(U := (0,1)\times (0,1) {\setminus } \{u^k=0\}\) and observe that \(\partial _z u^k \le 0\) on \(\partial U\). Indeed:
-
\(\partial _z u^k =0\) on \(\{z=0\}\), by symmetry;
-
\(\partial _z u^k =0\) on \(\partial \{u^k=0\}\), since \(u^k\) is nonnegative and of class \(C^{1,1}\);
-
\(\partial _z u^k \le 0\) on \(\{z=1\}\), since \(u^k(r,1) =0\) while \(u^k(r,z) \ge 0\) for \(z<1\);
-
\(\partial _z u^k =\phi ' \le 0\) on \(\{r=1\}\).
As a consequence, since \(\mathrm{div}( r^{n-m-1}\nabla (\partial _z u^k))=0\) inside U (this follows by differentiating the equation for u with respect to z), we deduce by the maximum principle that \(\partial _z u^k \le 0\) in U. Also, we note that \(\partial _z u^k = 0\) in \((0,1)\times (0,1) {\setminus } U\) since \(u^k\equiv 0\) there (recall that \(u^k\) is nonnegative and of class \(C^{1,1}\)), proving (ii).
Finally, (iii) is an immediate consequence of (i) and (ii).
We now observe that, for k sufficiently large, the contact set contains a neighborhood of the origin, and hence it must contain a cylindrical neighborhood of \(\{r=0\}\) [thanks to (iii)]. On the other hand, for \(k\ll 1\) we have \(u^k(0)>0\). Therefore, by continuity, there exists (a unique) \(k_{\star }>0\) such that \(\partial \{u^{k_\star }>0\}\) touches tangentially the line \(\{r=0\}\).
Set \(u^\star := u^{k_\star }\). Observe that that, with this definition, the function
has a full \((m-1)\)-dimenional space of singular points on \({\mathcal {Z}}=\{x_m=x_{m+1} = \cdots =x_n=0\}\). Also, by the given symmetry, these singular points belong to the stratum \(\Sigma _{m}\).
We now prove the following:
Proposition A.1
Let u be the symmetric solution defined in (A.2). Then the set \({\mathcal {Z}}\subset \Sigma _m\) consists of anomalous points, that is \({\mathcal {Z}}\subset \Sigma _m^a\).
Proof
The proof consists of three steps.
- Step 1. We show that u has no other singular points in a neighborhood of \({\mathcal {Z}}\) (except of course the points in \({\mathcal {Z}}\)).
To prove this, assume by contradiction that there exists a sequence of singular points \(x_\ell \rightarrow 0\). Note that, since u is invariant in the first \(m-1\) variable, we can assume that \(x_\ell \in \{x_1=\cdots =x_{m-1}=0\}\). Then, by the symmetries of u and Lemma 3.2, a blow-up q of \(u-p_*\) at 0 is a homogeneous harmonic polynomial that vanishes on the \(x_m\)-axis and that enjoys the same symmetries as u (note that \(p_*\) has the same symmetries as u and vanishes on \(r=0\)). Thus, \(q=q^\star (z,r)\), where \(q^\star \) solves
Let \(\ell \) be the degree of q. Since q is smooth and \(q^\star (z,0) = 0\), \(q^\star \) must be a polynomial of the form \(\sum _{1\le k\le \ell /2}a_kz^{\ell -2k}r^{2k}\). Let \(k_0\ge 1\) be the first index such that \(a_{k_0}\ne 0\). Then
where Q is a polynomial of degree \(\ell -2k_0-2\). In particular, the terms inside the parenthesis cannot be identically zero, giving the desired contradiction.
As a consequence, there is a neighborhood of \({\mathcal {Z}}\) which is free of singular points. (except the points in \({\mathcal {Z}}\)). In particular, as a consequence of [4, 5], there exists \(\varepsilon >0\) such that \((\partial \{u^{\star }>0\} {\setminus } \{0\})\cap B_\varepsilon \) is a smooth curve contained inside \((-\varepsilon ,\varepsilon )\times (0,\varepsilon ){\setminus }\{0\}\).
- Step 2. We show that, for any \(\alpha >0\) small, there exists \(R =R(\alpha )>0\) small such that the following holds:
where \(Y_{\varrho _\circ }\cdot {\varvec{e}}_r\) denotes the r-component of the point \(Y_{\varrho _\circ }\).
Let \(\rho := \sqrt{r^2+z^2}\) and \(\theta := \arctan (z/r) \in (-\frac{\pi }{2},\frac{\pi }{2})\) be polar coordinates in \({(0,1)\times (-1,1)}\). We now state the following fact, whose proof is postponed to the end of the Step 2.
Claim For any \(\alpha >0\) small, there exists \(\delta = \delta (\alpha )>0\), and a function \(\Theta _\alpha : [\delta ,\frac{\pi }{2} -\delta ] \rightarrow \mathbb {R}\) smooth, such that \(S_\alpha (r,z) := \rho ^{1+\alpha /4} \Theta _\alpha (\theta )\) satisfies
and
Note that, since 0 is a singular point and recalling the symmetries of u, \(u(x)=p_*(x)+o(|x|^2)\) where \(p_*(x)=\frac{1}{2(n-m)}\sum _{i=m+1}^nx_i^2\). This implies that \(u^\star (z,r) \ge \frac{k_\star }{4(n-m)}r^2+o(z^2)\) near the origin. Thus, given \(\alpha >0\), there exists \(\eta =\eta (\alpha )>0\) small enough such that the inclusion
holds. Note that, up to reducing \(\eta \), we can assume that \(\eta <\varepsilon ,\) so that (by Step 1) \(\partial \{u^\star >0\}{\setminus } \{0\}\) is a smooth curve inside \(B_\eta \).
Let us consider the function \(h :=-\partial _z u^{\star }\), and recall that \(h\ge 0\) (by property (iii) in the construction of \(u^\star \)). Also, differentiating the equation
with respect to z, we obtain
Since \(h>0\) inside \(\{u^\star >0\}\) (by the strong maximum principle), it follows that
and \({\mathcal {C}}_\delta \cap \partial B_\eta \subset \subset \{u^\star >0\}\). Thus there exists a constant \(c_0=c_0(\alpha )>0\) such that \(h\ge c_1S_\alpha \) on \({\mathcal {C}}_\delta \cap \partial B_\eta \). In addition \(h\ge 0=S_\alpha \) on \(\partial {\mathcal {C}}_\delta \cap B_\eta \). Hence, since h and \(S_\alpha \) solve the same equation, it follows by the maximum principle that \(h\ge S_\alpha \) inside \({\mathcal {C}}_\delta \cap B_\eta \). Recalling (A.4), this implies that
for some \(c_1= c_1(\alpha )>0\).
We now want to find a lower bound on the normal derivative of h at points on \(\partial \{u^\star >0\}\). For this we use a Hopf-type argument, constructing suitable barriers for our operator \(\mathrm{div}(r^{n-m-1}\nabla \,\cdot \,)\). These are given by the family of functions
Note that
and \(S^{H}_b<0\) outside of the parabolic region \({\mathcal {P}}_b:=\left\{ b +\frac{n-m-1}{2} (z-1)^2 \le r\right\} \).
Now, given \(0<\varrho _\circ \ll \eta \) we consider the rescaled function
and we note that [thanks to (A.8)]
Consider now our barriers \(S_b^H\). For \(b>1/2\) the parabolic region \({\mathcal {P}}_b\) is contained inside the set \(\left\{ \theta <\textstyle \frac{\pi }{2} -2\delta \right\} \). In particular \(S_b^H<0\) inside the cone \(\hat{{\mathcal {C}}}_{2\delta } := \left\{ \theta \in (\frac{\pi }{2}-2\delta ,\frac{\pi }{2}) \right\} \). Then, we start decreasing b until the first value \(b_\circ \) such that \(\partial {\mathcal {P}}_{b_\circ }\) touches \(\partial \{h_{\varrho _\circ } >0\}\). Note that, thanks to (A.7) and Step 1, \(b_\circ >0\) and the contact point will happen for some \({\widetilde{Y}}_{\varrho _\circ }\in \{r>0\}\).
Since \(h_{\varrho _\circ }\ge 0=S_{b_\circ }^H\) on \(\partial {\mathcal {P}}_{b_\circ }\cap \hat{{\mathcal {C}}}_{2\delta }\) and \(h_{\varrho _\circ }\ge c_2(\alpha )\varrho _\circ ^{-\alpha /4}\ge S_{b_\circ }^H\) on \({\mathcal {P}}_{b_\circ }\cap \partial \hat{{\mathcal {C}}}_{2\delta }\) for \(\varrho _\circ \) sufficiently small [see (A.9)], it follows by the maximum principle that \(h_{\varrho _\circ }\ge S_{b_\circ }^H\) inside \({\mathcal {P}}_{b_\circ }\cap \hat{{\mathcal {C}}}_{2\delta }\). Hence, since both \(h_{\varrho _\circ }\) and \(S_{b_\circ }^H\) vanish at \({\widetilde{Y}}_{\varrho _\circ }\), we deduce that
where \(\nu \) is the unit inwards normal to \(\partial \{h_{\varrho _\circ } >0\}\), and \(c_2=c_2(\alpha )\) is independent of \(\varrho _\circ \) (here we use \(b_\circ \le 1/2\)). Observe also that, provided \(\delta \) is small enough, \({\mathcal {P}}_{b_\circ }\cap \hat{{\mathcal {C}}}_{2\delta }\subset B_2\).
Rescaling (A.10), we obtain that for all \(\varrho _\circ \in (0,\eta )\) small enough there is a point \(Y_{\varrho _\circ } \in \partial \{u^\star >0\} \cap (0,1)\times (0,1)\) such that,
where \(\nu \) denotes the unit inwards normal to \(\{u^\star >0\}\) [see (A.7)].
To conclude the proof we observe that differentiating the equation
with respect to z we obtain \(-\mathrm{div}(r^{n-m-1}\nabla h) = k^\star \, r^{n-m-1} \nu _z\, {\mathcal {H}}^1|_{\partial \{u^\star >0\}}\), thus
where \(\nu _z=\nu \cdot {\varvec{e}}_z\) denotes the z-component of \(\nu \). Recalling (A.11), this proves that
Note that (A.12) already implies that \(\partial \{u^\star >0\}\) cannot be a \(C^{1,\alpha }\) curve at 0. We now prove the more precise estimate (A.3).
Since \(b_\circ >0\) and the rescaled parabolic cap
touches \(\partial \{u^\star =0\}\) at \(Y_{\varrho _\circ }=(r_{\varrho _\circ },z_{\varrho _\circ })\), it follows that
and that
Combining (A.13) with (A.12), we get
Recalling (A.13), this yields
Since \(\varrho _\circ ^{1+\alpha }\ll \varrho _\circ ^{1+\alpha /2}\), this proves (A.3).
We conclude Step 2 proving the claim.
Proof of Claim. The linear function \(z = \rho \sin \theta \) satisfies \(\mathrm{div}( r^{n-m-1} \nabla z)=0\). Also \(\Theta _0(\theta ) = \sin \theta \) is positive on \((0,\pi /2)\) and satisfies \(\Theta _0(0)=0\), hence \(\Theta _0\) must be the minimizer of the Rayleigh quotient
[equivalently, recalling (A.1), the restriction to the harmonic function \(x_m\) to \(\mathbb {R}^{m-1}\times {\mathbb {S}}^{n-m}\subset \mathbb {R}^{m-1}\times \mathbb {R}^{n-m+1}\) is a minimizer of the Rayleigh quotient for the classical Dirichlet integral on \({\mathbb {S}}^{n-m}\cap \{x_m\ge 0\}\)].
Since \(n-m-1\ge 1\) (this is where we crucially use this assumption), we have
(this is equivalent to saying that the north pole in \({\mathbb {S}}^{n-m}\cap \{x_m\ge 0\}\) has harmonic capacity 0, so the boundary condition \(\Theta (\pi /2-\delta )=0\) disappears in the limit \(\delta \rightarrow 0\)). Thus, by continuity, for any \(\alpha >0\) small there exists \(\delta =\delta (\alpha )>0\) small enough such that the previous Rayleigh quotient in \((\delta ,\pi /2-\delta )\) will give the value
This implies that if we denote by \(\Theta _\alpha \) the first eigenfunction (i.e., the function attaining the minimal quotient value \(\mu _\alpha \)), then \(-( (\cos \theta )^{n-m-1}\Theta _\alpha ')'=\mu _\alpha (\cos \theta )^{n-m-1}\Theta _\alpha \), and it follows by a direct computation (or by classical spectral theory) that \(S_\alpha (r,z) := \rho ^{1+\alpha /4} \Theta _\alpha (\theta )\) satisfies \(\mathrm{div}( r^{n-m-1} \nabla S_\alpha )=0\), as desired. Finally, the strict positivity of \(\Theta _\alpha \) inside \(\left( \delta ,\frac{\pi }{2}-\delta \right) \) is a classical property of the first eigenfunction.
- Step 3. We conclude the proof of the proposition by showing that (A.3) is incompatible with \(0\in \Sigma ^g_m\). Recall that, by definition, 0 belongs to \(\Sigma ^g_m\) (resp. \(\Sigma ^a_m\)) if \(\phi (0^+,u-p_*) \ge 3\) (resp. \(\phi (0^+,u-p_*)<3\)), see (3.13).
Assume by contradiction that \(\phi (0^+,u-p_*) \ge 3\). Then by Lemma 2.6 we have
We now note that
so it follows by the mean value formula for superharmonic functions that
Recalling that \(p_* = p_*(z,r) = \frac{1}{2(n-m)} r^2\) in the variables (z, r), it follows by (A.15) that
therefore
which clearly contradicts (A.3) if we choose \(\alpha <1/2\). As a consequence 0 (and by symmetry all points on \({\mathcal {Z}}\)) must belong to \(\Sigma ^a_m\). \(\square \)
Rights and permissions
About this article
Cite this article
Figalli, A., Serra, J. On the fine structure of the free boundary for the classical obstacle problem. Invent. math. 215, 311–366 (2019). https://doi.org/10.1007/s00222-018-0827-8
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00222-018-0827-8