betaFeedback

WorksMathematics

Overdetermined boundary problems for maximal graphs

Abstract

We study overdetermined boundary problems for the maximal surface equation with zero Dirichlet data and constant nonzero Neumann data . For the exterior of a bounded domain with connected boundary in Rn\mathbb{R}^n, n≥2n\geq2, we prove that such domain must be the complement of a ball and the solution is radial. In dimension two, we also show that a positive solution on a domain bounded by a properly embedded noncompact connected curve forces the domain to be a half-plane and the solution to be affine. The proofs combine the divergence structure of the equation with the induced geometry of maximal graphs.

Introduction

Let Ln+1\mathbb{L}^{n+1} denote the Lorentz–Minkowski space endowed with the metric

⟨⋅,⋅⟩L=dx12+⋯+dxn2−dxn+12.\langle\cdot,\cdot\rangle_{\mathrm{L}}=dx_{1}^{2}+\cdots+dx_{n}^{2}-dx_{n+1}^{2}.

The graph of a function u:Ω⊂Rn→Ru:\Omega\subset\mathbb{R}^{n}\to\mathbb{R} is spacelike if ∣Du∣<1|Du|<1, and its mean curvature with respect to the Lorentz metric vanishes precisely when uu satisfies the maximal surface equation

div⁡(Du1−∣Du∣2)=0in Ω.(1)\operatorname{div}\left(\frac{Du}{\sqrt{1-|Du|^{2}}}\right)=0 \quad\text{in }\Omega. \tag*{(1)}

This equation is the Euler–Lagrange equation of the Lorentzian area functional

A(u)=∫Ω1−∣Du∣2 dx,\mathcal{A}(u)=\int_{\Omega}\sqrt{1-|Du|^{2}}\,dx,

under compactly supported variations of the graph. The strict spacelikeness condition ∣Du∣<1|Du|<1 ensures that the induced metric gij=δij−uiujg_{ij}=\delta_{ij}-u_{i}u_{j} is Riemannian and that the equation is elliptic. Unlike minimal graphs in Euclidean space, spacelike solutions of (1.1) locally maximize the Lorentzian area among spacelike graph variations with fixed boundary values, owing to the strict concavity of the area integrand 1−∣Du∣2\sqrt{1-|Du|^{2}}. This variational property explains the terminology maximal.

The classical Bernstein theorem of Calabi and Cheng–Yau asserts that every smooth entire spacelike maximal graph must be a hyperplane [5, 6]. For domains with boundary, the geometry of the boundary and the prescribed boundary data allow a much richer family of solutions; see, for example, Bartnik–Simon [3].

On a bounded region KK, the functional

J(u)=∫K(1−1−∣Du∣2) dx\mathcal{J}(u)=\int_K\left(1-\sqrt{1-|Du|^2}\right)\,dx

differs from the negative Lorentzian area by a constant, and its Euler–Lagrange equation is (1.1).

We orient ∂K\partial K by the unit outward normal ν\nu. To describe domain variations without assuming that J(u)\mathcal{J}(u) is finite, fix a large ball BRB_R and consider deformations supported strictly inside BRB_R. Let

Kt=(Id⁡+tX)(K),JR(t)=∫BR∖Kt‾(1−1−∣Dut∣2) dx.K_t=(\operatorname{Id}+tX)(K),\qquad\mathcal{J}_R(t)=\int_{B_R\setminus\overline{K_t}}\left(1-\sqrt{1-|Du_t|^2}\right)\,dx.

where utu_t solves the maximal surface equation, vanishes on the moving boundary, and has fixed Dirichlet data on the outer boundary. For a smooth family of spacelike solutions, the shape derivative is

JR′(0)=∫∂K∩BR(11−∣Du∣2−1)X⋅ν dσ.\mathcal{J}_R'(0)=\int_{\partial K\cap B_R}\left(\frac{1}{\sqrt{1-|Du|^2}}-1\right)X\cdot\nu\,d\sigma.

The first variation of the enclosed volume is

ddt∣t=0∣Kt∣=∫∂KX⋅ν dσ.\left.\frac{d}{dt}\right|_{t=0}|K_t|=\int_{\partial K}X\cdot\nu\,d\sigma.

Thus, critical point under volume-preserving deformations implies there exists a Lagrange multiplier λ\lambda such that

11−∣Du∣2−1=λon ∂K.\frac{1}{\sqrt{1-|Du|^2}}-1=\lambda\quad\text{on }\partial K.

Hence we are led to the overdetermined problem

\begin{cases} \operatorname{div}\left(\dfrac{Du}{\sqrt{1-|Du|^2}}\right)=0 & \text{in }\Omega,\\ u=0,\qquad\partial_\nu u=\kappa& \text{on }\partial\Omega, \end{cases} \qquad0<\kappa<1. \tag*{(2)}\end{cases}

where Ω=Rn∖K\Omega=\mathbb{R}^n\setminus K and ν\nu denotes the unit normal pointing into Ω\Omega. Geometrically, the graph has boundary in the horizontal hyperplane and meets that hyperplane at a constant hyperbolic angle.

The study of overdetermined elliptic problems originates in Serrin’s symmetry theorem [20], which establishes spherical symmetry for positive solutions of a broad class of elliptic equations on bounded domains with constant Dirichlet and Neumann data. For exterior domains, Reichel [14, 15] developed a version of the moving plane method adapted to the unbounded setting. Sirakov [22] subsequently extended this theory, allowing, in particular, different constant boundary data on different components of a bounded obstacle. These results show that suitable overdetermined data, together with appropriate conditions at infinity, force the obstacle to be a ball.

For the corresponding Euclidean mean curvature equation, the planar problem admits a particularly strong classification. Cui [7] proved that a positive minimal graph with connected boundary, zero Dirichlet data, and constant normal derivative must be a half-plane or a portion of a vertical catenoid. Equivalently, its domain is a half-plane or the exterior of a disk. The proof uses analytic reflection and classification results for complete minimal surfaces.

Our first main result investigates (2) on an exterior domain of a bounded set.

Theorem 1.1. Let n≥2n \ge2 and let K⊂RnK \subset\mathbb{R}^{n} be a bounded domain with C3,αC^{3,\alpha} connected boundary, where 0<α<10 < \alpha< 1. Suppose that Ω=Rn∖K‾\Omega= \mathbb{R}^{n} \setminus\overline{K} is connected. Let u∈C3(Ω)∩C2(Ω‾)u \in C^{3}(\Omega) \cap C^{2}(\overline{\Omega}) satisfy the overdetermined problem (2). Then there exist x0∈Rnx_{0} \in\mathbb{R}^{n} and R>0R > 0 such that K=BR(x0)K = B_{R}(x_{0}) and

u(x)=∫R∣x−x0∣Dr2n−2+D2 dr,D=κRn−11−κ2.u(x) = \int_{R}^{|x-x_{0}|} \frac{D}{\sqrt{r^{2n-2}+D^{2}}}\,dr,\qquad D = \frac{\kappa R^{n-1}}{\sqrt{1-\kappa^{2}}}.

An important analytic input for this theorem is the asymptotic expansion of solutions to the maximal surface equation on exterior domains obtained by Hong and Yuan [10]. Let Ω\Omega be as above, and let uu be a smooth strictly spacelike solution of the maximal surface equation in Ω\Omega. The divergence structure implies that the quantity

Q[u]:=∫∂BRDu⋅νR1−∣Du∣2 dσQ[u] := \int_{\partial B_{R}} \frac{Du \cdot\nu_{R}}{\sqrt{1-|Du|^{2}}}\,d\sigma

is independent of RR whenever K‾⊂BR\overline{K} \subset B_{R}. Following the terminology of Hong and Yuan, the normalization quantity Res⁡[u]:=Q[u]γn\operatorname{Res}[u] := \frac{Q[u]}{\gamma_{n}} is called the residue of uu, where

γn:={2π,n=2,(n−2)∣Sn−1∣,n≥3.\gamma_{n} := \begin{cases} 2\pi, & n=2,\\ (n-2)|S^{n-1}|, & n\ge3. \end{cases}

Set

ρa(x):=∣x∣2−(a⋅x)2.\rho_{a}(x) := \sqrt{|x|^{2}-(a\cdot x)^{2}}.

the asymptotic expansion of Hong–Yuan asserts that there exists a∈Rna \in\mathbb{R}^n, ∣a∣<1|a|<1 and b∈Rb \in\mathbb{R} such that

u(x)={a⋅x+b+(1−∣a∣2)Res⁡[u]log⁡ρa(x)+Ok(∣x∣−1log⁡∣x∣),n=2,a⋅x+b−(1−∣a∣2)Res⁡[u]ρa(x)2−n+Ok(∣x∣1−n),n≥3,(3)u(x)=\begin{cases} a\cdot x+b+(1-|a|^2)\operatorname{Res}[u]\log\rho_a(x)+O_k\left(|x|^{-1}\log|x|\right), & n=2,\\ a\cdot x+b-(1-|a|^2)\operatorname{Res}[u]\rho_a(x)^{2-n}+O_k\left(|x|^{1-n}\right), & n\geq3, \end{cases} \tag*{(3)}

for every integer k≥0k\geq0. Here the notation R(x)=Ok(∣x∣μ(log⁡∣x∣)ℓ)R(x)=O_k\left(|x|^\mu(\log|x|)^\ell\right) means that, for every multi-index α\alpha with ∣α∣≤k|\alpha|\leq k,

∣DαR(x)∣≤Cα∣x∣μ−∣α∣(log⁡∣x∣)ℓ|D^\alpha R(x)|\leq C_\alpha|x|^{\mu-|\alpha|}(\log|x|)^\ell

for all sufficiently large ∣x∣|x|.

In dimension two, we have absorbed the explicit correction term of order ∣x∣−1log⁡∣x∣|x|^{-1}\log|x| in Hong–Yuan’s expansion into the remainder. When a=0a=0, their sharper expansion gives

u(x)=b+Res⁡[u]log⁡∣x∣+Ok(∣x∣−1).u(x)=b+\operatorname{Res}[u]\log|x|+O_k\left(|x|^{-1}\right).

We next turn to domains with noncompact boundary. The geometry of overdetermined problems in this setting is substantially richer. A central question was formulated by Berestycki, Caffarelli, and Nirenberg [4], who conjectured that a smooth domain Ω⊂Rn\Omega\subset\mathbb{R}^n with connected complement admitting a bounded positive solution of

Δv+f(v)=0in Ω,v=0,∂νv=constant≠0on ∂Ω,\Delta v+f(v)=0\quad\text{in }\Omega,\qquad v=0,\qquad\partial_\nu v=\text{constant}\ne0\quad\text{on }\partial\Omega,

for a Lipschitz nonlinearity ff, must be a ball, a half-space, a generalized cylinder, or the complement of one of these domains. The BCN conjecture is false in general. Sicbaldi [21] constructed counterexamples in dimensions n≥3n\geq3 by perturbing straight cylinders periodically. Ros, Ruiz, and Sicbaldi [18] subsequently constructed nonradial exterior domains supporting bounded positive solutions for suitable nonlinearities in every dimension n≥2n\geq2, thereby also disproving the conjecture in the plane.

For complements of unbounded obstacles, nontrivial examples exist even in the harmonic case f≡0f\equiv0. Fall, Minlend, and Weth [8] constructed exceptional domains in R3\mathbb{R}^3 as complements of nontrivial periodic perturbations of a straight cylinder. Minlend, Weth, and Wu [12] obtained analogous constructions in every dimension n≥4n\geq4. These domains have connected noncompact boundary and support positive harmonic functions with zero Dirichlet data and constant nonzero Neumann data. The corresponding functions have logarithmic growth in the three-dimensional construction and are bounded in dimensions n≥4n\geq4. Thus, in higher dimensions, connectedness of the boundary does not force even a harmonic overdetermined domain to be a half-space or the complement of a straight circular cylinder. These examples also suggest that a similarly unrestricted classification for maximal graphs should not be expected, although they do not themselves provide counterexamples for the maximal surface equation.

In dimension two, however, strong rigidity results remain available when the boundary is connected and noncompact. Ros, Ruiz, and Sicbaldi [19] proved that a planar domain with unbounded connected boundary supporting a bounded positive solution of the semilinear overdetermined problem above must be a half-plane, and the solution is one-dimensional.

Several rigidity results are also available for geometric quasilinear equations in dimension two. In addition to the aforementioned result of Cui [7], a recent contribution is due to Lian and Sicbaldi [11], who studied the capillary overdetermined problem

{div⁡(Dv1+∣Dv∣2)+f(v)=0in Ω,v>0in Ω,v=0,∂νv=κon ∂Ω.\begin{cases} \operatorname{div}\left(\frac{Dv}{\sqrt{1+\lvert Dv\rvert^2}}\right)+f(v)=0 & \text{in }\Omega,\\ v>0 & \text{in }\Omega,\\ v=0,\quad\partial_\nu v=\kappa& \text{on }\partial\Omega. \end{cases}

For planar domains with unbounded connected boundary, they obtained half-plane rigidity and one-dimensionality of the solution under boundedness of both vv and DvDv, provided that ff admits a nonpositive primitive FF satisfying F(0)≥11+κ2−1F(0)\ge\frac{1}{\sqrt{1+\kappa^2}}-1.

Our second result establishes a rigidity statement in the setting of noncompact domains with connected unbounded boundary in dimension two.

Theorem 1.2. Let Γ⊂R2\Gamma\subset\mathbb{R}^2 be a properly embedded C3,αC^{3,\alpha} curve diffeomorphic to R\mathbb{R}, where α∈(0,1)\alpha\in(0,1), and let Ω\Omega be one of the two connected components of R2∖Γ\mathbb{R}^2\setminus\Gamma. Let κ∈(0,1)\kappa\in(0,1) be a constant. Suppose that u∈C3(Ω)∩C2(Ω‾)u\in C^3(\Omega)\cap C^2(\overline{\Omega}) satisfies the overdetermined problem (1.2), then, after a rigid motion, Ω={(x1,x2)∈R2:x2>0}\Omega=\{(x_1,x_2)\in\mathbb{R}^2:x_2>0\}, and u(x1,x2)=κx2u(x_1,x_2)=\kappa x_2.

The proofs of both theorems combine the divergence structure of the equation with the geometry of maximal graphs, making essential use of the induced Riemannian metric. For Theorem 1.1, integral identities and geometric inequalities lead to rigidity through their equality cases. This strategy is reminiscent of Weinberger’s approach to Serrin’s problem [23] and of the integral methods underlying Alexandrov-type rigidity results [16, 17]. For Theorem 1.2, a key step is to establish the uniform spacelikeness (∣Du∣≤1−δ,δ>0\lvert Du\rvert\le1-\delta,\delta>0), which allows us to exploit global conformal coordinates and reduce the rigidity argument to the analysis of harmonic functions.

We present the proof of Theorem 1.1 in Section 2 and the proof of Theorem 1.2 in Section 3.

Rigidity for exterior domains of bounded obstacles

Throughout this section, we write

Ω=Rn∖K‾,Γ=∂K,W=1−∣Du∣2,\Omega=\mathbb{R}^n\setminus\overline{K},\qquad\Gamma=\partial K,\qquad W=\sqrt{1-\lvert Du\rvert^2},

and orient Γ\Gamma by the unit normal ν\nu pointing from KK into Ω\Omega. We denote the Euclidean area of Γ\Gamma by A=∣Γ∣A=\lvert\Gamma\rvert, the volume of KK by V=∣K∣V=\lvert K\rvert. The mean curvature HH is the sum of principal curvatures with respect to ν\nu, so that H=(n−1)/RH=(n-1)/R on ∂BR\partial B_R. νR\nu_R denotes the unit outward normal of BRB_R.

In view of the maximal surface equation (2), the divergence theorem in BR∖K‾B_R\setminus\overline{K} gives

γnRes⁡[u]=∫∂BRDu⋅νRW dσ=∫∂Kuν1−∣Du∣2 dσ=κ1−κ2A.(4)\gamma_n\operatorname{Res}[u]=\int_{\partial B_R}\frac{Du\cdot\nu_R}{W}\,d\sigma=\int_{\partial K}\frac{u_\nu}{\sqrt{1-|Du|^2}}\,d\sigma=\frac{\kappa}{\sqrt{1-\kappa^2}}A. \tag*{(4)}

For simplicity, we set

s0:=κ1−κ2.s_0:=\frac{\kappa}{\sqrt{1-\kappa^2}}.

The first step is to show that the overdetermined boundary conditions eliminate the linear part allowed by the general exterior asymptotic expansion (3).

Lemma 2.1. Under the assumptions of Theorem 1.1, the asymptotic slope aa of uu in the expansion (3) vanishes.

If n=2n=2, there exists b∈Rb\in\mathbb{R} such that

u(x)=Res⁡[u]log⁡∣x∣+b+Ok(∣x∣−1).(5)u(x)=\operatorname{Res}[u]\log|x|+b+O_k\left(|x|^{-1}\right). \tag*{(5)}

In this case, u>0u>0 in Ω\Omega.

If n≥3n\ge3, there exists c>0c>0 such that

u(x)=c−Res⁡[u]∣x∣2−n+Ok(∣x∣1−n).(6)u(x)=c-\operatorname{Res}[u]|x|^{2-n}+O_k\left(|x|^{1-n}\right). \tag*{(6)}

In this case, 0<u<c0<u<c in Ω\Omega. The expansions hold for every nonnegative integer kk.

Proof. Define

Tij=Wδij+uiujW.T_{ij}=W\delta_{ij}+\frac{u_i u_j}{W}.

Each row of TT has zero divergence, since

∑j=1n∂jTij=∂iW+1W∑j=1nuijuj+uidiv⁡(DuW)=0.\sum_{j=1}^{n}\partial_jT_{ij}=\partial_iW+\frac{1}{W}\sum_{j=1}^{n}u_{ij}u_j+u_i\operatorname{div}\left(\frac{Du}{W}\right)=0.

On ∂K\partial K,

Tν=(1−κ2+κ21−κ2)ν=ν1−κ2.T\nu=\left(\sqrt{1-\kappa^2}+\frac{\kappa^2}{\sqrt{1-\kappa^2}}\right)\nu=\frac{\nu}{\sqrt{1-\kappa^2}}.

The divergence theorem yields

∫∂BRTνR dσ=11−κ2∫∂Kν dσ=0.(7)\int_{\partial B_R}T\nu_R\,d\sigma=\frac{1}{\sqrt{1-\kappa^2}}\int_{\partial K}\nu\,d\sigma=0. \tag*{(7)}

By Hong and Yuan’s asymptotic expansion (3), there exists a∈Rna\in\mathbb{R}^n, ∣a∣<1|a|<1, such that

Du=a+h,h=O(∣x∣1−n).Du=a+h,\qquad h=O\left(|x|^{1-n}\right).

Write

Wa=1−∣a∣2,v=u−a⋅x,B=DuW.W_a=\sqrt{1-|a|^2},\qquad v=u-a\cdot x,\qquad B=\frac{Du}{W}.

Thus h=Dvh=Dv, and

T−a⊗B=WI+h⊗(a+h)W.T-a\otimes B=WI+\frac{h\otimes(a+h)}{W}.

Using

W=Wa−a⋅hWa+O(∣h∣2),1W=1Wa+O(∣h∣),W=W_a-\frac{a\cdot h}{W_a}+O(|h|^2),\qquad\frac{1}{W}=\frac{1}{W_a}+O(|h|),

we obtain

T−a⊗B=WaI+h⊗a−(a⋅h)IWa+O(∣h∣2).(8)T-a\otimes B=W_aI+\frac{h\otimes a-(a\cdot h)I}{W_a}+O(|h|^2). \tag*{(8)}

Choose a smooth function v~\widetilde v on Rn\mathbb{R}^n which agrees with vv outside a fixed large ball, using a cutoff near K‾\overline K. For each ii, the vector field Fj(i)=aj∂iv~−δij∑k=1nak∂kv~F_j^{(i)}=a_j\partial_i\widetilde v-\delta_{ij}\sum_{k=1}^n a_k\partial_k\widetilde v is divergence free. Since Dv~=hD\widetilde v=h on ∂BR\partial B_R for large RR, it follows that each row of h⊗a−(a⋅h)Ih\otimes a-(a\cdot h)I is also divergence free, and consequently

∫∂BR[h⊗a−(a⋅h)I]νR dσ=0.(9)\int_{\partial B_R}[h\otimes a-(a\cdot h)I]\nu_R\,d\sigma=0. \tag*{(9)}

Integrating (2.5) against νR\nu_R and using (2.1), (2.4) and (2.6), we find

−γnRes⁡[u]a=∫∂BRO(∣h∣2)νR dσ.-\gamma_n\operatorname{Res}[u]a=\int_{\partial B_R}O(|h|^2)\nu_R\,d\sigma.

Consequently,

γnRes⁡[u]∣a∣≤C∫∂BR∣h∣2 dσ≤CR1−n.\gamma_n\operatorname{Res}[u]|a|\le C\int_{\partial B_R}|h|^2\,d\sigma\le CR^{1-n}.

Letting R→∞R\to\infty and using Res⁡[u]>0\operatorname{Res}[u]>0, we conclude that a=0a=0.

The expansions (2.2) and (2.3) now follow from (1.3).

For n=2n=2, positivity of the residue implies u(x)→+∞u(x)\to+\infty as ∣x∣→∞|x|\to\infty. The maximum principle on bounded truncations, followed by the strong maximum principle, gives u>0u>0 in Ω\Omega.

For n≥3n\ge3, suppose that c≤0c\le0. Since u=0u=0 on ∂K\partial K and u(x)→cu(x)\to c at infinity, the maximum principle on bounded truncations would give u≤0u\le0 in Ω\Omega. This contradicts ∂νu=κ>0\partial_\nu u=\kappa>0 on ∂K\partial K. Hence c>0c>0, and the maximum principle gives 0<u<c0<u<c throughout Ω\Omega. □

Next, we shall utilize the geometric property of the maximal surface equation. We equip Ω\Omega with the metric induced by the maximal graph, gij=δij−uiujg_{ij}=\delta_{ij}-u_i u_j. We will reserve DD for Euclidean derivative, and all notions with a subscript gg such as ∇g\nabla_g are taken with respect to gg.

We list several geometric identities needed below.

gij=δij+uiujW2,det⁡g=W2,∇gu=DuW2.g^{ij}=\delta_{ij}+\frac{u_i u_j}{W^2},\qquad\det g=W^2,\qquad\nabla_g u=\frac{Du}{W^2}.

Setting s:=∣∇gu∣g=∣Du∣Ws:=|\nabla_g u|_g=\frac{|Du|}{W}, therefore

W=1−∣Du∣2=11+s2.W=\sqrt{1-|Du|^2}=\frac{1}{\sqrt{1+s^2}}.

From the maximal surface equation, it is easy to see that uu is harmonic with respect to gg:

Δgu=1Wdiv⁡(DuW)=0.(10)\Delta_g u=\frac{1}{W}\operatorname{div}\left(\frac{Du}{W}\right)=0. \tag*{(10)}

With the convention Πij=uijW\Pi_{ij}=\frac{u_{ij}}{W} for the second fundamental form in Lorentzian space, the Christoffel symbols and the Hessian of the height function are

Γijk=−ukuijW2,(∇g2u)ij=uijW2=ΠijW.\Gamma^k_{ij}=-\frac{u_k u_{ij}}{W^2},\qquad(\nabla_g^2u)_{ij}=\frac{u_{ij}}{W^2}=\frac{\Pi_{ij}}{W}.

Since the graph is maximal, tr⁡gΠ=0\operatorname{tr}_g\Pi=0. The Gauss equation therefore gives

Ric⁡g(X,Y)=∑j=1nΠ(X,ej)Π(Y,ej),(11)\operatorname{Ric}_g(X,Y)=\sum_{j=1}^{n}\Pi(X,e_j)\Pi(Y,e_j), \tag*{(11)}

where {ej}j=1n\{e_j\}_{j=1}^n is any gg-orthonormal frame.

For later use, let Σ\Sigma be a regular level set of uu, oriented by the Euclidean unit normal ν=Du∣Du∣\nu=\frac{Du}{|Du|}. Its gg-unit normal pointing toward increasing uu is η=∇gus=νW\eta=\frac{\nabla_g u}{s}=\frac{\nu}{W}. If H=div⁡νH=\operatorname{div}\nu denotes the sum of the Euclidean principal curvatures of Σ\Sigma, then its mean curvature in (Ω,g)(\Omega,g), with the same trace convention, is

Hg=div⁡gη=HW.H_g=\operatorname{div}_g\eta=\frac{H}{W}.

Consequently, (10) implies

0=div⁡g(∇gu)=div⁡g(sη)=η(s)+sHg.0=\operatorname{div}_g(\nabla_g u)=\operatorname{div}_g(s\eta)=\eta(s)+sH_g.

Therefore, we have

η(s)=−sHW.(12)\eta(s)=-\frac{sH}{W}. \tag*{(12)}

Lemma 2.2. For s>0s>0, define

Fn(s)={log⁡s1+1+s2,n=2,∫0sτ−1/(n−1)1+τ2 dτ,n≥3.F_n(s)= \begin{cases} \log\frac{s}{1+\sqrt{1+s^2}}, & n=2,\\ \int_0^s\frac{\tau^{-1/(n-1)}}{\sqrt{1+\tau^2}}\,d\tau, & n\ge3. \end{cases}

Plugging s:=∣∇gu∣g=∣Du∣Ws:=|\nabla_g u|_g=\frac{|Du|}{W}, then on {s>0}\{s>0\},

ΔgFn(s)=Fn′(s)s(∣∇g2u∣g2−nn−1∣∇gs∣g2)≥0.(13)\Delta_g F_n(s)=\frac{F_n'(s)}{s}\left(|\nabla_g^2u|_g^2-\frac{n}{n-1}|\nabla_gs|_g^2\right)\ge0. \tag*{(13)}

If n=2n=2, equality holds in (13). If n≥3n\ge3, setting Fn(0)=0F_n(0)=0 gives a continuous weakly subharmonic function on Ω\Omega. If n=2n=2, for every M>0M>0, the function max⁡{F2(s),−M}\max\{F_2(s),-M\}, assigned the value −M-M at s=0s=0, is weakly subharmonic on Ω\Omega.

Moreover, along every regular level set of uu,

η(Fn(s))=−Hs(n−2)/(n−1).(14)\eta(F_n(s))=-H s^{(n-2)/(n-1)}. \tag*{(14)}

Proof. We first work on {s>0}\{s>0\}. Since Π=W∇g2u\Pi=W\nabla_g^2u and ∇g2u(∇gu,⋅)=s ds\nabla_g^2u(\nabla_g u,\cdot)=s\,ds, equation (2.8) gives

Ric⁡g(∇gu,∇gu)=W2s2∣∇gs∣g2=s21+s2∣∇gs∣g2.\operatorname{Ric}_g(\nabla_g u,\nabla_g u)=W^2s^2|\nabla_gs|_g^2=\frac{s^2}{1+s^2}|\nabla_gs|_g^2.

Applying the Bochner formula to the harmonic function uu, we obtain

12Δgs2=∣∇g2u∣g2+Ric⁡g(∇gu,∇gu),\frac{1}{2}\Delta_gs^2=|\nabla_g^2u|_g^2+\operatorname{Ric}_g(\nabla_g u,\nabla_g u),
sΔgs=∣∇g2u∣g2−∣∇gs∣g21+s2.s\Delta_gs=|\nabla_g^2u|_g^2-\frac{|\nabla_gs|_g^2}{1+s^2}.

For both definitions of FnF_n,

Fn′(s)=s−1/(n−1)1+s2>0,Fn′′(s)Fn′(s)=−1(n−1)s−s1+s2.F_n'(s)=\frac{s^{-1/(n-1)}}{\sqrt{1+s^2}}>0,\qquad\frac{F_n''(s)}{F_n'(s)}=-\frac{1}{(n-1)s}-\frac{s}{1+s^2}.

The chain rule thus yields

ΔgFn(s)=Fn′(s)Δgs+Fn′′(s)∣∇gs∣g2=Fn′(s)s(∣∇g2u∣g2−nn−1∣∇gs∣g2).\begin{aligned} \Delta_gF_n(s)=F_n'(s)\Delta_gs+F_n''(s)|\nabla_gs|_g^2 \\ &=\frac{F_n'(s)}{s}\left(|\nabla_g^2u|_g^2-\frac{n}{n-1}|\nabla_gs|_g^2\right). \end{aligned}

By the refined Kato inequality, the term in the parentheses is nonnegative. When n=2n=2, it vanishes.

To handle critical points, suppose first that n≥3n\ge3. The defining integral converges at zero, and Fn(s)→0F_n(s)\to0 as s→0s\to0. For every ε>0\varepsilon>0, the function max⁡{Fn(s),ε}\max\{F_n(s),\varepsilon\} is locally constant near {s=0}\{s=0\} and weakly subharmonic on {s>0}\{s>0\}. It is therefore weakly subharmonic throughout Ω\Omega. Letting ε↓0\varepsilon\downarrow0, with locally uniform convergence, proves the assertion for Fn(s)F_n(s). For n=2n=2, the same argument applies directly to max⁡{F2(s),−M}\max\{F_2(s),-M\}, since F2(s)→−∞F_2(s)\to-\infty as s→0s\to0.

Finally, using (12) and Fn′(s)=Ws−1/(n−1)F_n'(s)=Ws^{-1/(n-1)}, we find

η(Fn(s))=Fn′(s)η(s)=Ws−1/(n−1)(−sHW)=−Hs(n−2)/(n−1).\eta(F_n(s))=F_n'(s)\eta(s)=Ws^{-1/(n-1)}\left(-\frac{sH}{W}\right)=-Hs^{(n-2)/(n-1)}.

□

Proof of Theorem 1.1. We divide the proof into two cases.

The planar case

Assume n=2n=2. By Lemma 2.1,

u=Res⁡[u]log⁡∣x∣+b+O1(∣x∣−1).u=\operatorname{Res}[u]\log|x|+b+O_1(|x|^{-1}).

It follows that s=Res⁡[u]∣x∣+O(∣x∣−2)s=\frac{\operatorname{Res}[u]}{|x|}+O(|x|^{-2}) and hence

F2(s)=−log⁡∣x∣+log⁡Res⁡[u]2+O(∣x∣−1).F_2(s)=-\log|x|+\log\frac{\operatorname{Res}[u]}{2}+O(|x|^{-1}).

Thus G:=F2(s)+uRes⁡[u]G:=F_2(s)+\frac{u}{\operatorname{Res}[u]} is bounded near infinity and equals F2(s0)F_2(s_0) on Γ\Gamma.

We claim that

G≤F2(s0)in Ω.(15)G\le F_2(s_0)\qquad\text{in }\Omega. \tag*{(15)}

Indeed fix ε>0\varepsilon>0. Since u→+∞u\to+\infty and GG is bounded near infinity, G−εu−F2(s0)<0G-\varepsilon u-F_2(s_0)<0 on every sufficiently large sphere. On Γ\Gamma, the same expression is zero.

To justify the comparison at critical points, fix such a sphere ∂BR\partial B_R and choose MM large enough that the truncation max⁡{F2(s),−M}\max\{F_2(s),-M\} agrees with F2(s)F_2(s) on both boundary components of Ω∩BR\Omega\cap B_R. This is possible because s>0s>0 on Γ\Gamma and on sufficiently large spheres. By Lemma 2.2,

max⁡{F2(s),−M}+(1Res⁡[u]−ε)u−F2(s0)\max\{F_2(s),-M\}+\left(\frac{1}{\operatorname{Res}[u]}-\varepsilon\right)u-F_2(s_0)

is weakly subharmonic. The maximum principle gives the desired comparison on Ω∩BR\Omega\cap B_R. Exhausting Ω\Omega and then letting ε↓0\varepsilon\downarrow0 proves (2.12).

It follows from (14) that

−H+s0Res⁡[u]≤0on Γ.-H+\frac{s_0}{\operatorname{Res}[u]}\le0\qquad\text{on }\Gamma.

Since Q=s0AQ=s_0A and Res⁡[u]=Q/(2π)\operatorname{Res}[u]=Q/(2\pi),

H≥2πA.(16)H\ge\frac{2\pi}{A}. \tag*{(16)}

The connected boundary Γ\Gamma is a simple closed curve, so the turning-angle formula gives

∫ΓH dS=2π.\int_\Gamma H\,dS=2\pi.

Equality must therefore hold everywhere in (16). Thus Γ\Gamma has constant positive curvature and KK is a disk.

The higher-dimensional case

Assume n≥3n \ge3 and abbreviate

β=1n−1,F=Fn,F0=F(s0).\beta= \frac{1}{n-1}, \qquad F = F_n, \qquad F_0 = F(s_0).

By Lemma 2.1, 0<u<c0 < u < c and u→cu \to c at infinity.

The function Z=F(s)−F0c(c−u)Z = F(s) - \frac{F_0}{c}(c-u) is weakly subharmonic, vanishes on Γ\Gamma, and tends to zero at infinity. The maximum principle on an exhaustion gives

F(s)≤F0c(c−u)in Ω.(17)F(s) \le\frac{F_0}{c}(c-u) \qquad\text{in } \Omega. \tag*{(17)}

Taking the inward derivative on Γ\Gamma yields

−Hs01−β+F0s0c≤0.-Hs_0^{1-\beta}+\frac{F_0s_0}{c}\le0.

Consequently,

H≥h0,h0:=F0s0βc>0.(18)H \ge h_0, \qquad h_0 := \frac{F_0s_0^\beta}{c} > 0. \tag*{(18)}

The asymptotic expansion implies that

J:=∫Ω(1−W) dx<∞.J := \int_\Omega(1-W)\,dx < \infty.

Multiplying the maximal surface equation by uu and integrating over Ω∩BR\Omega\cap B_R, we find

∫Ω∩BR∣Du∣2W dx=∫∂BRuDu⋅νRW dS.\int_{\Omega\cap B_R}\frac{|Du|^2}{W}\,dx = \int_{\partial B_R}u\frac{Du\cdot\nu_R}{W}\,dS.

Letting R→∞R \to\infty gives

∫Ω∣Du∣2W dx=cQ.(19)\int_\Omega\frac{|Du|^2}{W}\,dx = cQ. \tag*{(19)}

Consider the divergence-free tensor

S=I−T=(1−W)I−Du⊗DuW.S=I-T=(1-W)I-\frac{Du\otimes Du}{W}.

Its trace is

tr⁡S=n(1−W)−∣Du∣2W.\operatorname{tr}S=n(1-W)-\frac{|Du|^2}{W}.

On Γ\Gamma,

Sν=−(11−κ2−1)ν.S\nu=-\left(\frac{1}{\sqrt{1-\kappa^2}}-1\right)\nu.

Moreover, S=O(∣Du∣2)S=O(|Du|^2) at infinity, so

∫∂BRx⋅SνR dS=O(R2−n)⟶0.\int_{\partial B_R} x\cdot S\nu_R\,dS=O(R^{2-n})\longrightarrow0.

Applying the divergence theorem to SxSx and using (2.16), we obtain

nJ−cQ=(11−κ2−1)∫Γx⋅ν dσ=nV(1+s02−1).(20)nJ-cQ=\left(\frac{1}{\sqrt{1-\kappa^2}}-1\right)\int_{\Gamma}x\cdot\nu\,d\sigma=nV\left(\sqrt{1+s_0^2}-1\right). \tag*{(20)}

Define

j(s)=11+1+s2,I(s0)=∫0s0j(s)F′(s) ds.j(s)=\frac{1}{1+\sqrt{1+s^2}},\qquad I(s_0)=\int_0^{s_0}j(s)F'(s)\,ds.

The function jj is decreasing. For almost every t∈(0,c)t\in(0,c), the level set Σt={u=t}\Sigma_t=\{u=t\} is smooth and compact. The divergence theorem gives

∫Σts dσ=Q.(21)\int_{\Sigma_t}s\,d\sigma=Q. \tag*{(21)}

Since

1−W∣Du∣=s1+1+s2=sj(s),\frac{1-W}{|Du|}=\frac{s}{1+\sqrt{1+s^2}}=sj(s),

the Euclidean coarea formula yields

J=∫0c∫Σtsj(s) dσ dt.J=\int_0^c\int_{\Sigma_t}sj(s)\,d\sigma\,dt.

The critical set causes no additional contribution, because 1−W=01-W=0 wherever Du=0Du=0. For t∈[0,c]t\in[0,c], define sˉ(t)∈[0,s0]\bar{s}(t)\in[0,s_0] by

F(sˉ(t))=F0(1−tc).F(\bar{s}(t))=F_0\left(1-\frac{t}{c}\right).

The comparison (2.14) implies s≤sˉ(t)s\leq\bar{s}(t) on Σt\Sigma_t. Using the monotonicity of jj and (2.18), we obtain

J≥Q∫0c1j(sˉ(t)) dt=cQF0I(s0).(22)J\geq Q\int_0^c\frac{1}{j(\bar{s}(t))}\,dt=\frac{cQ}{F_0}I(s_0). \tag*{(22)}

The elementary identity

nI(s0)−F0=(n−1)(1+s02−1)s0−1−β(23)nI(s_0)-F_0=(n-1)\left(\sqrt{1+s_0^2}-1\right)s_0^{-1-\beta} \tag*{(23)}

follows by differentiating both sides with respect to s0s_0 and observing that both sides tend to zero as s0↓0s_0 \downarrow0. Combining (2.17), (2.19), and (2.20), we find

nV(1+s02−1)=nJ−cQ≥cQF0(nI(s0)−F0)=(n−1)cQF0(1+s02−1)s0−1−β.\begin{aligned} nV\left(\sqrt{1+s_0^2}-1\right)=nJ-cQ\geq\frac{cQ}{F_0}\left(nI(s_0)-F_0\right) \\ &=\frac{(n-1)cQ}{F_0}\left(\sqrt{1+s_0^2}-1\right)s_0^{-1-\beta}. \end{aligned}

Since Q=s0AQ=s_0A, cancellation gives

nV≥(n−1)Ah0.(24)nV\geq\frac{(n-1)A}{h_0}. \tag*{(24)}

The curvature bound (2.15) makes Γ\Gamma strictly mean convex. The Euclidean Heintze–Karcher inequality[9] states that

nV≤(n−1)∫Γ1H dσ.nV\leq(n-1)\int_{\Gamma}\frac{1}{H}\,d\sigma.

Together with (2.15) and (2.21), this gives

nV≤(n−1)∫Γ1H dσ≤(n−1)Ah0≤nV.nV\leq(n-1)\int_{\Gamma}\frac{1}{H}\,d\sigma\leq\frac{(n-1)A}{h_0}\leq nV.

All inequalities above are therefore equalities, and K=BR(x0)K=B_R(x_0) is a ball. Finally we show that uu is radial about x0x_0. For an orthogonal transformation OO, set

uO(x)=u(x0+O(x−x0)).u_O(x)=u\left(x_0+O(x-x_0)\right).

The functions uOu_O and uu solve the same equation and have the same zero boundary data. By Lemma 2.1,

uO−u⟶0as ∣x∣→∞.u_O-u\longrightarrow0\qquad\text{as }|x|\to\infty.

In dimension two, this follows from the common logarithmic coefficient and constant term; in higher dimensions, both functions tend to cc. The comparison principle on bounded exhaustions therefore gives uO=uu_O=u. Thus u(x)=U(∣x−x0∣)u(x)=U(|x-x_0|).

The radial equation is

(rn−1U′(r)1−U′(r)2)′=0.\left(r^{n-1}\frac{U'(r)}{\sqrt{1-U'(r)^2}}\right)'=0.

Since U′(R)=κU'(R)=\kappa, its first integral is

rn−1U′(r)1−U′(r)2=D,D=κRn−11−κ2.r^{n-1}\frac{U'(r)}{\sqrt{1-U'(r)^2}}=D,\qquad D=\frac{\kappa R^{n-1}}{\sqrt{1-\kappa^2}}.

Solving for U′U' and using U(R)=0U(R)=0, we obtain

U(r)=∫RrDρ2n−2+D2 dρ.U(r)=\int_R^r \frac{D}{\sqrt{\rho^{2n-2}+D^2}}\,d\rho.

This is the asserted formula and completes the proof. □

Rigidity for noncompact connected boundaries

In this section we prove Theorem 1.2. In the bounded-obstacle case considered above, the gradient vanishes at infinity; although such decay is unavailable in the present setting, the following lemma shows that the uniform spacelikeness still follows from the boundary conditions.

Lemma 3.1. Let Γ⊂R2\Gamma\subset\mathbb{R}^2 be a C3,αC^{3,\alpha} properly embedded curve diffeomorphic to R\mathbb{R}, and let Ω\Omega be one of the two components of R2∖Γ\mathbb{R}^2\setminus\Gamma. Let κ∈(0,1)\kappa\in(0,1) be a constant and we assume that uu satisfies (1.2), then

∣Du∣≤κthroughout Ω.|Du|\leq\kappa\qquad\text{throughout }\Omega.

Proof. Put

W=1−∣Du∣2,P=1−W1+W=∣Du∣2(1+W)2.W=\sqrt{1-|Du|^2},\qquad P=\frac{1-W}{1+W}=\frac{|Du|^2}{(1+W)^2}.

Notice that 0≤P<10\leq P<1. On the set {Du≠0}\{Du\ne0\}, put s=∣∇gu∣g=∣Du∣Ws=|\nabla_g u|_g=\frac{|Du|}{W}. The two-dimensional identity in Lemma 2.2 gives ΔgF2(s)=0\Delta_gF_2(s)=0, where

F2(s)=log⁡s1+1+s2.F_2(s)=\log\frac{s}{1+\sqrt{1+s^2}}.

Thus P=e2F2(s)P=e^{2F_2(s)} and

ΔgP=4P∣∇gF2(s)∣g2≥0on {Du≠0}.\Delta_gP=4P|\nabla_gF_2(s)|_g^2\geq0\qquad\text{on }\{Du\ne0\}.

The function PP is C2C^2 throughout Ω\Omega. At every critical point of uu, PP attains its minimum value zero, so ΔgP≥0\Delta_gP\geq0 there as well. Therefore PP is subharmonic on all of Ω\Omega.

Note

P=Pκ=1−1−κ21+1−κ2on ∂Ω.P=P_\kappa=\frac{1-\sqrt{1-\kappa^2}}{1+\sqrt{1-\kappa^2}}\qquad\text{on }\partial\Omega.

We shall show that P(x)≤Pκ(x)P(x)\leq P_\kappa(x) throughout Ω\Omega.

To this end, we need a superharmonic barrier as follows. For every x∈Ωx\in\Omega, choose a nearest boundary point y∈∂Ωy\in\partial\Omega. Since the segment [x,y)[x,y) lies in Ω\Omega, the conditions u(y)=0u(y)=0 and ∣Du∣<1|Du|<1 give

∣u(x)∣<∣x−y∣=dist⁡(x,∂Ω).|u(x)|<|x-y|=\operatorname{dist}(x,\partial\Omega).

Choose x∗∈R2∖Ω‾x_*\in\mathbb{R}^2\setminus\overline{\Omega} and δ>0\delta>0 such that Bδ(x∗)⊂R2∖Ω‾B_\delta(x_*)\subset\mathbb{R}^2\setminus\overline{\Omega}, and write r(x)=∣x−x∗∣r(x)=|x-x_*|. Then dist⁡(x,∂Ω)≤r(x)−δ\operatorname{dist}(x,\partial\Omega)\leq r(x)-\delta, and consequently

ρ(x):=r(x)2−u(x)2≥2δr(x)−δ2≥δ2.(25)\rho(x):=r(x)^2-u(x)^2\geq2\delta r(x)-\delta^2\geq\delta^2. \tag*{(25)}

Consider the translated graph immersion X(x)=(x−x∗,u(x))X(x)=(x-x_*,u(x)) into Lorentz–Minkowski space, so that ρ=⟨X,X⟩L\rho=\langle X,X\rangle_L. Let NN be its future-directed unit timelike normal. Maximality gives ΔgX=0\Delta_gX=0, and hence

Δgρ=4,∇gρ=2X⊤.\Delta_g\rho=4,\qquad\nabla_g\rho=2X^\top.

Since

X⊤=X+⟨X,N⟩LN,⟨N,N⟩L=−1,X^\top=X+\langle X,N\rangle_LN,\qquad\langle N,N\rangle_L=-1,

we obtain

∣∇gρ∣g2=4(ρ+⟨X,N⟩L2).|\nabla_g\rho|_g^2=4\left(\rho+\langle X,N\rangle_L^2\right).

It follows that

Δglog⁡ρ=Δgρρ−∣∇gρ∣g2ρ2=−4⟨X,N⟩L2ρ2≤0.\Delta_g\log\rho=\frac{\Delta_g\rho}{\rho}-\frac{|\nabla_g\rho|_g^2}{\rho^2} =-\frac{4\langle X,N\rangle_L^2}{\rho^2}\leq0.

Thus the function φ=log⁡ρδ2\varphi=\log\frac{\rho}{\delta^2} is nonnegative and superharmonic. Moreover, (3.1) implies that φ(x)→+∞\varphi(x)\to+\infty as ∣x∣→∞|x|\to\infty in Ω\Omega.

For ε>0\varepsilon>0, define Vε=P−Pκ−εφV_\varepsilon=P-P_\kappa-\varepsilon\varphi. It follows that

ΔgVε≥0in Ω,Vε≤0on ∂Ω.\Delta_gV_\varepsilon\geq0\quad\text{in }\Omega,\qquad V_\varepsilon\leq0\quad\text{on }\partial\Omega.

The maximum principle, applied on sufficiently large bounded truncations of Ω\Omega, gives

P−Pκ≤εφin Ω.P-P_\kappa\leq\varepsilon\varphi\quad\text{in }\Omega.

Letting ε↓0\varepsilon\downarrow0 at each fixed point, we conclude that P≤PκP\leq P_\kappa in Ω\Omega. Since the function

t⟼1−1−t21+1−t2t\longmapsto\frac{1-\sqrt{1-t^2}}{1+\sqrt{1-t^2}}

is strictly increasing on [0,1)[0,1), this implies ∣Du∣≤κ|Du|\leq\kappa.

This uniform spackelikeness enables one to build up a global conformal parametrization that respects the noncompact boundary.

Lemma 3.2. Under the assumptions of Theorem 1.2, there exists a conformal diffeomorphism

X:H⟶(Ω,g),H={z=x+iy:y>0},X:\mathbb{H}\longrightarrow(\Omega,g),\qquad\mathbb{H}=\{z=x+iy:y>0\},

where g=δ−du⊗dug=\delta-du\otimes du, which extends to a homeomorphism H∪R∪{∞}⟶Ω∪Γ∪{∞}\mathbb{H}\cup\mathbb{R}\cup\{\infty\}\longrightarrow\Omega\cup\Gamma\cup\{\infty\} in the spherical topology, with X(R)=ΓX(\mathbb{R})=\Gamma, and X(∞)=∞X(\infty)=\infty. At every finite boundary point, XX extends locally as a C1C^1 diffeomorphism up to the boundary.

Proof. Since Γ\Gamma is a properly embedded curve diffeomorphic to R\mathbb{R}, the component of R2∖Γ\mathbb{R}^2\setminus\Gamma other than Ω\Omega contains a closed ball. The boundary conditions give Du=κνDu=\kappa\nu on Γ\Gamma. Thus Lemma 3.1 yields

∣Du∣≤κ<1in Ω,|Du|\leq\kappa<1\quad\text{in }\Omega,

and consequently (1−κ2)δ≤g≤δ(1-\kappa^2)\delta\leq g\leq\delta. The conformal structure determined by gg has a Beltrami coefficient μ\mu satisfying

∣μ∣≤1−1−κ21+1−κ2<1.|\mu| \leq\frac{1-\sqrt{1-\kappa^2}}{1+\sqrt{1-\kappa^2}} < 1.

Extend μ\mu measurably to C\mathbb{C} by setting it equal to zero outside Ω\Omega. By the measurable Riemann mapping theorem [1], there exists a quasiconformal homeomorphism

Φ:C^⟶C^,Φ(∞)=∞,\Phi:\widehat{\mathbb{C}} \longrightarrow\widehat{\mathbb{C}}, \qquad\Phi(\infty)=\infty,

whose restriction to Ω\Omega is an isothermal coordinate map for gg.

Properness of Γ\Gamma implies that Γ^=Γ∪{∞}\widehat{\Gamma}=\Gamma\cup\{\infty\} is a Jordan curve in C^\widehat{\mathbb{C}}. Therefore Φ(Γ^)\Phi(\widehat{\Gamma}) is also a Jordan curve, and D=Φ(Ω)\mathcal{D}=\Phi(\Omega) is one of its complementary domains. The Riemann mapping theorem and the Carathéodory boundary extension theorem [13] provide a conformal map

ψ:H⟶D\psi:\mathbb{H}\longrightarrow\mathcal{D}

which extends homeomorphically to the closures in C^\widehat{\mathbb{C}}. After precomposing with an automorphism of H\mathbb{H}, we may arrange that ψ(∞)=∞\psi(\infty)=\infty. Hence X=Φ−1∘ψX=\Phi^{-1}\circ\psi has the asserted conformal and topological properties.

It remains to verify regularity at finite boundary points. This is a local issue. Fix p∈Γp\in\Gamma. Since u∈C2(Ω‾)u\in C^2(\overline{\Omega}) and ∣Du∣=κ<1|Du|=\kappa<1 on Γ\Gamma, the metric gg admits a positive definite C1C^1 extension to a neighborhood of pp. Local isothermal coordinates for this extension may be chosen of class C1,βC^{1,\beta} for some β∈(0,1)\beta\in(0,1). In these coordinates, Γ\Gamma becomes a C1,βC^{1,\beta} boundary arc, and XX becomes an ordinary conformal map.

The local boundary regularity theorem of Kellogg–Warschawski [13] then implies that this conformal map extends C1C^1 to the boundary arc, with nonvanishing derivative. Transforming back to the original coordinates shows that XX extends locally as a C1C^1 diffeomorphism up to Γ\Gamma. □

We also record a standard consequence of the representation of positive harmonic functions on the upper half-plane.

Lemma 3.3. Let hh be a nonnegative harmonic function on H\mathbb{H}, continuous on H∪R\mathbb{H}\cup\mathbb{R}. If

h(x,0)=bfor every x∈R,h(x,0)=b \qquad\text{for every }x\in\mathbb{R},

where b≥0b\geq0 is constant, then h(x,y)=b+ayh(x,y)=b+ay for some a≥0a\geq0.

Proof. The conclusion is immediate if h≡0h\equiv0. Otherwise, the strong maximum principle gives h>0h>0 in H\mathbb{H}. By the representation theorem for positive harmonic functions on the upper half-space [2 Theorem 7.26], there exist a≥0a\geq0 and a positive Borel measure μ\mu on R\mathbb{R} such that

h(x,y)=ay+1π∫Ry(x−t)2+y2 dμ(t),∫Rdμ(t)1+t2<∞.h(x,y)=ay+\frac{1}{\pi}\int_{\mathbb{R}}\frac{y}{(x-t)^2+y^2}\,d\mu(t), \qquad\int_{\mathbb{R}}\frac{d\mu(t)}{1+t^2}<\infty.

The approximate identity property of the Poisson kernel, together with this integrability condition, implies that

h(x,y) dx⇀μlocally as y↓0.h(x,y)\,dx \rightharpoonup\mu\qquad\text{locally as } y \downarrow0.

More precisely, for every φ∈Cc(R)\varphi\in C_c(\mathbb{R}),

lim⁡y↓0∫Rφ(x)h(x,y) dx=∫Rφ(t) dμ(t).\lim_{y\downarrow0}\int_{\mathbb{R}}\varphi(x)h(x,y)\,dx=\int_{\mathbb{R}}\varphi(t)\,d\mu(t).

On the other hand, continuity of hh on H∪R\mathbb{H}\cup\mathbb{R} and the boundary condition h(x,0)=bh(x,0)=b imply that h(x,y)→bh(x,y)\to b uniformly on each compact interval as y↓0y\downarrow0. Therefore

∫Rφ(t) dμ(t)=b∫Rφ(t) dt\int_{\mathbb{R}}\varphi(t)\,d\mu(t)=b\int_{\mathbb{R}}\varphi(t)\,dt

for every φ∈Cc(R)\varphi\in C_c(\mathbb{R}), and hence μ=b dt\mu=b\,dt.

Substituting this into the representation formula and using

1π∫Ry(x−t)2+y2 dt=1,\frac{1}{\pi}\int_{\mathbb{R}}\frac{y}{(x-t)^2+y^2}\,dt=1,

we obtain h(x,y)=b+ayh(x,y)=b+ay, with a≥0a\geq0. □\square

Now we are ready to complete the proof.

Proof of Theorem 1.2. Let XX be the conformal parametrization given by Lemma 3.2, and write

X∗g=e2φ(dx2+dy2).X^*g=e^{2\varphi}(dx^2+dy^2).

Since uu is gg-harmonic, the function U=u∘XU=u\circ X is harmonic on H\mathbb{H}. It is positive in H\mathbb{H} and continuous with zero boundary values on the entire real axis. Lemma 3.3 therefore gives

U(x,y)=ayfor some a>0.(26)U(x,y)=ay \qquad\text{for some } a>0. \tag*{(26)}

In particular,

s∘X=∣∇X∗gU∣=ae−φ>0.s\circ X=\lvert\nabla_{X^*g}U\rvert=ae^{-\varphi}>0.

Thus uu has no critical points.

Lemma 2.2 now applies without truncation and yields

ΔgF2(s)=0throughout Ω.\Delta_gF_2(s)=0 \qquad\text{throughout } \Omega.

Since F2(s)<0F_2(s)<0 for s>0s>0, the function V=−F2(s)∘XV=-F_2(s)\circ X is positive and harmonic on H\mathbb{H}. The boundary conditions imply

Du=κν,s=s0on Γ.Du=\kappa\nu,\qquad s=s_0\qquad\text{on } \Gamma.

Hence VV extends continuously to the real axis with the constant boundary value −F2(s0)>0-F_2(s_0)>0. A second application of Lemma 3.3 gives

V(x,y)=−F2(s0)+byfor some b≥0.V(x,y)=-F_2(s_0)+by \qquad\text{for some } b\geq0.

Combining this identity with (26), we obtain F2(s)∘X=F2(s0)−bau∘XF_2(s)\circ X=F_2(s_0)-\frac{b}{a}u\circ X. Since XX is onto,

F2(s)=F2(s0)−bauin Ω.(27)F_2(s)=F_2(s_0)-\frac{b}{a}u \qquad\text{in }\Omega. \tag*{(27)}

Let HH denote the signed Euclidean curvature of Γ\Gamma, with the convention H=div⁡ΓνH=\operatorname{div}_{\Gamma}\nu. Thus a circle has positive curvature when ν\nu points toward its exterior. The inward unit conormal in the graph metric is η=ν1−κ2\eta=\frac{\nu}{\sqrt{1-\kappa^2}}, On Γ\Gamma we have ∂ηu=s0\partial_{\eta}u=s_0. The boundary identity (14), specialized to dimension two, gives ∂ηF2(s)=−H\partial_{\eta}F_2(s)=-H. Taking the inward derivative of (27) therefore yields

H=bas0on Γ.(28)H=\frac{b}{a}s_0 \qquad\text{on }\Gamma. \tag*{(28)}

In particular, Γ\Gamma has constant curvature.

A properly embedded curve in R2\mathbb{R}^2 is complete in its induced Euclidean metric. A complete connected plane curve with constant nonzero curvature is a circle, whereas a complete connected curve with zero curvature is a line. Since Γ\Gamma is noncompact and diffeomorphic to R\mathbb{R}, the circle case is excluded. Thus H=0H=0. Because s0>0s_0>0, equation (28) implies b=0b=0. Consequently,

F2(s)=F2(s0)in Ω.F_2(s)=F_2(s_0) \qquad\text{in }\Omega.

The function F2F_2 is strictly increasing, so s≡s0s\equiv s_0, and ∣Du∣≡κ\lvert Du\rvert\equiv\kappa.

We finally show that uu is affine. Differentiating ∣Du∣2=κ2\lvert Du\rvert^2=\kappa^2 gives D2uDu=0D^2uDu=0. The nondivergence form of the maximal surface equation is

Δu+D2u(Du,Du)1−∣Du∣2=0,\Delta u+\frac{D^2u(Du,Du)}{1-\lvert Du\rvert^2}=0,

and therefore Δu=0\Delta u=0. The Euclidean Bochner identity gives

0=12Δ∣Du∣2=∣D2u∣2+Du⋅D(Δu)=∣D2u∣2.0=\frac{1}{2}\Delta\lvert Du\rvert^2=\lvert D^2u\rvert^2+Du\cdot D(\Delta u)=\lvert D^2u\rvert^2.

Hence D2u=0D^2u=0 and uu is affine.

We already know that Γ\Gamma is a line, and Ω\Omega is one of its complementary half-planes. After a Euclidean rigid motion we may write

Γ={x2=0},Ω={x2>0}.\Gamma=\{x_2=0\},\qquad\Omega=\{x_2>0\}.

The boundary conditions u=0u=0 and ∂νu=κ\partial_{\nu}u=\kappa then identify the affine function as u(x1,x2)=κx2u(x_1,x_2)=\kappa x_2. □

References

SCHOOL OF MATHEMATICAL SCIENCES, SHANGHAI JIAO TONG UNIVERSITY, SHANGHAI 200240, CHINA

Email address: [email protected]

References

  1. [1]L. V. Ahlfors, Lectures on Quasiconformal Mappings, 2nd ed., University Lecture Series, vol. 38, American Mathematical Society, Providence, RI, 2006.DOI
  2. [2]S. Axler, P. Bourdon, and W. Ramey, Harmonic Function Theory, 2nd ed., Graduate Texts in Mathematics, vol. 137, Springer-Verlag, New York, 2001.
  3. [3]R. Bartnik and L. Simon, Spacelike hypersurfaces with prescribed boundary values and mean curvature, Comm. Math. Phys. 87 (1982), no. 1, 131–152.DOI
  4. [4]H. Berestycki, L. A. Caffarelli, and L. Nirenberg, Monotonicity for elliptic equations in unbounded Lipschitz domains, Comm. Pure Appl. Math. 50 (1997), no. 11, 1089–1111.doi.org/10.1002/(sici)1097-0312(199711)50:11<1089::aid-cpa2>3.0.co;2-6
  5. [5]E. Calabi, Examples of Bernstein problems for some nonlinear equations, in Global Analysis, Proceedings of Symposia in Pure Mathematics, vol. 15, American Mathematical Society, Providence, RI, 1970, 223–230.DOI
  6. [6]S.-Y. Cheng and S.-T. Yau, Maximal space-like hypersurfaces in the Lorentz–Minkowski spaces, Ann. of Math. (2) 104 (1976), no. 3, 407–419.DOI
  7. [7]Q. Cui, A note on an overdetermined problem for minimal surface equation, Proc. Amer. Math. Soc. 150 (2022), no. 12, 5403–5409.DOI
  8. [8]M. M. Fall, I. A. Minlend, and T. Weth, On an electrostatic problem and a new class of exceptional subdomains of ℝ³, SIAM J. Math. Anal. 55 (2023), no. 3, 2347–2376. doi:10.1137/22M1487540.DOI
  9. [9]E. Heintze and H. Karcher, A general comparison theorem with applications to volume estimates for submanifolds, Ann. Sci. École Norm. Sup. (4) 11 (1978), no. 4, 451–470.DOI
  10. [10]G. Hong and Y. Yuan, Maximal hypersurfaces over exterior domains, Comm. Pure Appl. Math. 74 (2021), no. 3, 589–614.arxiv.org/abs/1903.00787
  11. [11]Y. Lian and P. Sicbaldi, Rigidity results for the capillary overdetermined problem, preprint, arXiv:2503.14215, 2025.arxiv.org/abs/2503.14215
  12. [12]I. A. Minlend, T. Weth, and J. Wu, Exceptional domains in higher dimensions, J. Geom. Anal. 36 (2026), Paper No. 6. doi:10.1007/s12220-025-02243-6.DOI
  13. [13]Ch. Pommerenke, Boundary Behaviour of Conformal Maps, Grundlehren der mathematischen Wissenschaften, vol. 299, Springer-Verlag, Berlin, 1992.DOI
  14. [14]W. Reichel, Radial symmetry for an electrostatic, a capillarity and some fully nonlinear overdetermined problems on exterior domains, Z. Anal. Anwendungen 15 (1996), no. 3, 619–635.DOI
  15. [15]W. Reichel, Radial symmetry for elliptic boundary-value problems on exterior domains, Arch. Rational Mech. Anal. 137 (1997), no. 4, 381–394.DOI
  16. [16]R. C. Reilly, Applications of the Hessian operator in a Riemannian manifold, Indiana Univ. Math. J. 26 (1977), no. 3, 459–472.
  17. [17]A. Ros, Compact hypersurfaces with constant higher order mean curvatures, Rev. Mat. Iberoamericana 3 (1987), 447–453.
  18. [18]A. Ros, D. Ruiz, and P. Sicbaldi, Solutions to overdetermined elliptic problems in nontrivial exterior domains, J. Eur. Math. Soc. (JEMS) 22 (2020), no. 1, 253–281. doi:10.4171/JEMS/921.
  19. [19]A. Ros, D. Ruiz, and P. Sicbaldi, A rigidity result for overdetermined elliptic problems in the plane, Comm. Pure Appl. Math. 70 (2017), no. 7, 1223–1252. doi:10.1002/cpa.21696.
  20. [20]J. Serrin, A symmetry problem in potential theory, Arch. Rational Mech. Anal. 43 (1971), 304–318.DOI
  21. [21]P. Sicbaldi, New extremal domains for the first eigenvalue of the Laplacian in flat tori, Calc. Var. Partial Differential Equations 37 (2010), nos. 3–4, 329–344. doi:10.1007/s00526-009-0264-z.DOI
  22. [22]B. Sirakov, Symmetry for exterior elliptic problems and two conjectures in potential theory, Ann. Inst. H. Poincaré C Anal. Non Linéaire 18 (2001), no. 2, 135–156.DOI
  23. [23]H. F. Weinberger, Remark on the preceding paper of Serrin, Arch. Rational Mech. Anal. 43 (1971), 319–320.DOI

Paper details

vlnc:W_7mexbn5m
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 8

downloads · 1

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents