betaFeedback

WorksMathematics

Uniqueness results for the exterior Dirichlet and Neumann problem in two dimensions for the Euler-Lagrange equations of variational integrals with linear growth

Abstract

We consider solutions u,v:Ω→RNu, v:Ω\rightarrow\mathbb{R}^{N} of the Euler-Lagrange equations associated to the variational integral ∫Ωf(∇u)dx\int_Ωf(\nabla u)dx, where ΩΩ is an exterior domain in R2\mathbb{R}^{2} e.g., the complement of a disk or a half plane, and the density ff is of linear growth. Among other things we show that u∣∂Ω=v∣∂Ωu|_{\partialΩ}=v|_{\partialΩ} together with the boundedness of u−vu-v yields that u=vu=v on the entire domain.

Introduction and statement of the results

In our note we investigate the exterior Dirichlet problem and in addition the exterior Neumann problem on planar domains associated with the Euler-Lagrange equations corresponding to variational integrals ∫Ωf(∇u) dx\int_{\Omega}f(\nabla u)\,dx with density ff being of linear growth providing some uniqueness results. A prominent example is given by the Dirichlet problem

{div⁡(∇u1+∣∇u∣2)=0on Ω,u=φon ∂Ω(1)\begin{cases} \operatorname{div}\left(\dfrac{\nabla u}{\sqrt{1+|\nabla u|^{2}}}\right)=0 & \text{on }\Omega,\\ u=\varphi& \text{on }\partial\Omega \tag*{(1)} \end{cases}

for functions u:Ω→Ru:\Omega\to\mathbb{R}, where Ω⊂R2\Omega\subset\mathbb{R}^{2} is a domain in R2\mathbb{R}^{2} such that R2−Ω\mathbb{R}^{2}-\Omega is bounded and φ\varphi denotes a given function. Note that the first line of (1) is the Euler-Lagrange equation for the area integral ∫Ω1+∣∇u∣2 dx\int_{\Omega}\sqrt{1+|\nabla u|^{2}}\,dx, but even in the case of problem (1) uniqueness of solutions can not be expected: for instance, if we let φ=0\varphi=0 and take Ω:=[∣x∣>1]:={x∈R2:∣x∣>1}\Omega:=[|x|>1]:=\{x\in\mathbb{R}^{2}:|x|>1\}, then problem (1) admits the three different solutions u=0u=0, k±(x):=±ln⁡(∣x∣+∣x∣2−1)k_{\pm}(x):=\pm\ln(|x|+\sqrt{|x|^{2}-1}), ∣x∣≥1|x|\geq1, which means that we have the trivial one together with the upper and lower catenoid. Therefore one can only hope for uniqueness under additional restrictions to be imposed on the solutions as it is stated in Theorem 11.3 of Osserman’s monograph [1]: if u,vu,v denote bounded solutions of (1), then it holds u=vu=v. This means that we have uniqueness for problem (1) in the space L∞(Ω)L^{\infty}(\Omega), however, as Theorem 11.4 in [1] shows, the existence of solutions to (1) in the space L∞(Ω)L^{\infty}(\Omega) can in general not be guaranteed. The delicate question of existence (and uniqueness) for the exterior Dirichlet problem (1) has been investigated by many authors leading to the result that, roughly speaking, a smallness condition for the oscillation of the boundary data φ\varphi is sufficient for the existence of a solution uu to (1), whereas prescribed asymptotic growth of uu at infinity or the existence of a constant asymptotic normal vector for the graph of uu provides uniqueness. We refer the reader to the contributions of for instance Kuwert [2], Kutev and Tomi [3], do Espírito-Santo and Ripoll [4], Ripoll and Tomi [5] and Aiolfi, Assmann and Ripoll [6], where one will find further references and also a careful analysis of the historical background. We now leave the minimal surface case and want to improve the uniqueness result from Theorem 11.3 in [1] even in a more general framework. Our assumptions are as follows: let G⊂R2G \subset\mathbb{R}^{2} denote a bounded Lipschitz domain and consider the exterior region

Ω:=R2−G‾.(2)\Omega:= \mathbb{R}^{2} - \overline{G}. \tag*{(2)}

For N≥1N \ge1 we denote by R2N\mathbb{R}^{2N} the space of (2×N)(2 \times N)-matrices. Suppose that f:R2N→Rf : \mathbb{R}^{2N} \to\mathbb{R} is of class C2C^{2} and in addition of linear growth in the sense that

a∣p∣−b≤f(p)≤A∣p∣+B,p∈R2N,(3)a|p| - b \le f(p) \le A|p| + B, \qquad p \in\mathbb{R}^{2N}, \tag*{(3)}

holds with positive constants a,b,A,Ba, b, A, B. We require further that

∣D2f∣∈L∞(R2N)(4)|D^{2}f| \in L^{\infty}(\mathbb{R}^{2N}) \tag*{(4)}

and that

D2f(p)(q,q)>0,p,q∈R2N−{0}(5)D^{2}f(p)(q,q) > 0, \qquad p,q \in\mathbb{R}^{2N} - \{0\} \tag*{(5)}

is satisfied. Due to (1.5) the density ff is a strictly convex function, which together with (1.3) yields ∣∇f∣∈L∞(R2N)|\nabla f| \in L^{\infty}(\mathbb{R}^{2N}) (compare, e.g., Lemma 2.2, p.156, in [7]). We then look at weak solutions u∈C1(Ω,RN)u \in C^{1}(\Omega,\mathbb{R}^{N}) of the equation

div⁡[∇f(∇u)]=0on Ω,(6)\operatorname{div}[\nabla f(\nabla u)] = 0 \quad\text{on } \Omega, \tag*{(6)}

which by definition means that we require

∫Ω∇f(∇u):∇φ dx=0(7)\int_{\Omega} \nabla f(\nabla u) : \nabla\varphi\, dx = 0 \tag*{(7)}

for all φ∈C01(Ω,RN)\varphi\in C^{1}_{0}(\Omega,\mathbb{R}^{N}), “:” denoting the scalar product of matrices. Our first result concerns the exterior Dirichlet problem for equation (1.6) giving an extension of Theorem 11.3 from [1]:

Theorem 1.1. Assume that we have (1.2)–(1.5) and suppose that u,v∈C1(Ω,RN)∩C0(Ω‾,RN)u,v \in C^{1}(\Omega,\mathbb{R}^{N}) \cap C^{0}(\overline{\Omega},\mathbb{R}^{N}) are weak solutions of (1.6) in the sense of (1.7). Assume further that

u=von ∂Ω.(8)u = v \quad\text{on } \partial\Omega. \tag*{(8)}

Then it follows u=vu = v on Ω\Omega, if

u−v∈L∞(Ω,RN)(9)u-v \in L^{\infty}(\Omega,\mathbb{R}^{N}) \tag*{(9)}

or

lim sup⁡R→∞∫[R<∣x∣<2R]∣u−v∣ dx<∞(10)\limsup_{R\to\infty}\int_{[R<|x|<2R]} |u-v|\,dx < \infty \tag*{(10)}

holds.

We note that in the case of bounded solutions u,vu,v condition (1.9) clearly holds, hence we have uniqueness in the class of bounded solutions of the exterior Dirichlet problem associated to equation (1.6).

From the proof of Theorem 1.1 it follows:

Corollary 1.2 (Uniqueness of entire solutions). Let (1.3)–(1.5) hold for the function ff and assume that u,v∈C1(R2,RN)u,v \in C^{1}(\mathbb{R}^{2},\mathbb{R}^{N}) are weak solutions of equation (1.6) on the whole plane, i.e. (1.7) is satisfied for all φ∈C01(R2,RN)\varphi\in C^{1}_{0}(\mathbb{R}^{2},\mathbb{R}^{N}). Then, in the case that u−vu-v is bounded, u=vu = v is true.

Next we apply Theorem 1.1 to a situation which is closer to the minimal surface setting introduced in problem (1.1): let N=1N=1 and consider a density ff with radial structure, i.e. we let f(p):=g(∣p∣)f(p):=g(|p|), p∈R2p\in\mathbb{R}^{2}, with g:[0,∞)→Rg:[0,\infty)\to\mathbb{R} of class C2C^{2} such that (compare (1.3)–(1.5))

{a~t−b~≤g(t)≤A~t+B~for t≥0,g′(0)=0,g′′∈L∞([0,∞)),and g′′(t)>0,t≥0(11)\begin{cases} \widetilde{a}t-\widetilde{b}\leq g(t)\leq\widetilde{A}t+\widetilde{B} & \text{for }t\geq0,\\ g'(0)=0,\quad g''\in L^{\infty}([0,\infty)), & \text{and }g''(t)>0,\quad t\geq0 \tag*{(11)} \end{cases}

holds with constants a~,b~,A~,B~>0\widetilde{a},\widetilde{b},\widetilde{A},\widetilde{B}>0. As domain Ω\Omega we take {∣x∣>1}\{|x|>1\} and observe that (1.6) now reads as

div⁡(g′(∣∇u∣)∣∇u∣∇u)=0on {∣x∣>1}.(12)\operatorname{div}\left(\frac{g'(|\nabla u|)}{|\nabla u|}\nabla u\right)=0\quad\text{on }\{|x|>1\}. \tag*{(12)}

In addition to (1.11) we require

∫0∞tg′′(t) dt<∞(13)\int_{0}^{\infty}tg''(t)\,dt<\infty \tag*{(13)}

and observe that (1.13) is true, if g′′(t)≤c(1+t)−μg''(t)\leq c(1+t)^{-\mu} with exponent μ>2\mu>2. We then introduce the gg-catenoids

kg±(x):=±∫1∣x∣(g′)−1(1t) dt,∣x∣≥1(14)k_{g}^{\pm}(x):=\pm\int_{1}^{|x|}(g')^{-1}\left(\frac{1}{t}\right)\,dt,\quad|x|\geq1 \tag*{(14)}

(compare formula (2.5) in [8]), which for the choice g(t):=1+t2g(t):=\sqrt{1+t^{2}} take the usual form. Note that by [8], Lemma 2, the gg-catenoids are solutions of (1.12) and in addition satisfy (u=kg±)(u=k_{g}^{\pm})

u∈C0({∣x∣≥1}),u(x)=0 for ∣x∣=1.(15)u\in C^{0}(\{|x|\geq1\}),\quad u(x)=0\text{ for }|x|=1. \tag*{(15)}

After these preparations we can deduce from Theorem 1.1:

Corollary 1.3. Let (1.11), (1.13) hold for the function gg and define kg±k_{g}^{\pm} as in (1.14). Suppose that u∈C1({∣x∣>1})u\in C^{1}(\{|x|>1\}) is a weak solution of (1.12) with the property (1.15). Then, if either u−kg+u-k_{g}^{+} or u−kg−u-k_{g}^{-} stays bounded, uu is equal to kg+k_{g}^{+} in the first case, whereas u=kg−u=k_{g}^{-} holds in the second case.

Next we present a result concerning solutions of equation (1.6) whose derivatives in all coordinate directions vanish on ∂Ω\partial\Omega in a rather strong sense.

Theorem 1.4. Let (1.2), (1.3) and (1.5) hold, whereas condition (1.4) is replaced by

∣D2f(p)∣≤c1+∣p∣,p∈R2N,(16)|D^{2}f(p)|\leq\frac{c}{1+|p|},\quad p\in\mathbb{R}^{2N}, \tag*{(16)}

cc denoting a positive constant. Let u∈C2(Ω,RN)∩C1(Ω‾,RN)u\in C^{2}(\Omega,\mathbb{R}^{N})\cap C^{1}(\overline{\Omega},\mathbb{R}^{N}) denote a solution of (1.6) with the property

∣∇u(x)∣≤c dist⁡(x,∂Ω)1/2(17)|\nabla u(x)|\leq c\,\operatorname{dist}(x,\partial\Omega)^{1/2} \tag*{(17)}

for x∈Ωx\in\Omega close to ∂Ω\partial\Omega for some other constant c>0c>0 (which for instance follows from ∂iu∈C0,1/2\partial_i u\in C^{0,1/2} near ∂Ω\partial\Omega together with ∂iu=0\partial_i u=0 on ∂Ω\partial\Omega, i=1,2i=1,2). Then, if in addition uu satisfies

lim⁡∣x∣→∞∣u(x)∣∣x∣<∞,(18)\lim_{|x|\to\infty}\frac{|u(x)|}{|x|}<\infty, \tag*{(18)}

the constancy of uu follows.

In order to proceed we discuss the situation for the case, when the domain Ω\Omega from (1.2) is replaced by a half-plane HH, for instance we may consider

H:=[x1>0]:={x=(x1,x2)∈R2:x1>0}.(19)H:=\left[x_{1}>0\right]:=\{x=(x_{1},x_{2})\in\mathbb{R}^{2}:x_{1}>0\}. \tag*{(19)}

Concerning problem (1.11) with HH in place of Ω\Omega the question of unique solvability has been analyzed by various authors, and without being complete we mention the papers of Hwang [9], [10] and of Collin and Krust [11] for further references. It turns out that two solutions u,vu,v must coincide, if u−vu-v satisfies certain growth restrictions as ∣x∣→∞|x|\to\infty. Concerning the question of the existence of solutions to (1.1) for the half-plane HH the interested should consult the paper of Jiang, Wang and Zhu [12].

Let us look at equation (1.6) for the half-plane. It is easy to prove:

Theorem 1.5. The results of Theorem 1.1 and of Theorem 1.4 remain true, if the domain Ω\Omega from (1.2) is replaced by the half-plane HH.

For the half-plane HH condition (1.17) seems to be somewhat artificial, since the gradient is required to vanish on ∂H\partial H. We finish the introduction with a more natural result.

Theorem 1.6. Let ff satisfy (1.3), (1.5), (1.16) and consider HH as in (1.19). Suppose that u∈C2(H,RN)∩C0(H‾,RN)u\in C^2(H,\mathbb{R}^N)\cap C^0(\overline{H},\mathbb{R}^N) is a solution of (1.7) on HH such that for a positive constant cc

∣∂1u(x)∣≤cdist⁡(x,∂H)=cx1(20)|\partial_1u(x)|\le c\operatorname{dist}(x,\partial H)=c\sqrt{x_1} \tag*{(20)}

holds for xx near ∂H=[x1=0]\partial H=[x_1=0] (which is true, if ∂1u=0\partial_1u=0 on ∂H\partial H and ∂1u∈C0,1/2\partial_1u\in C^{0,1/2} near ∂H\partial H). Then the validity of

lim⁡∣x∣→∞∣u(x)∣∣x∣<∞(21)\lim_{|x|\to\infty}\frac{|u(x)|}{|x|}<\infty \tag*{(21)}

implies the existence of points α,β∈RN\alpha,\beta\in\mathbb{R}^N with the property that u(x)=αx2+βu(x)=\alpha x_2+\beta, x∈Hx\in H.

Our paper is organized as follows: Section 2 contains the proofs of Theorem 1.1 and 1.4. In Section 3 we discuss the half-plane case presenting the proofs of Theorem 1.5 and 1.6.

Proofs of Theorem 1.1 and 1.4

During the proof of Theorem 1.1 we will benefit from:

Lemma 2.1. Let the density ff satisfy (1.3)–(1.5), let GG denote a bounded Lipschitz region in R2\mathbb{R}^2 and define Ω\Omega as in (1.2). Consider a function uu from the space C1(Ω,RN)∩C0(Ω‾,RN)C^1(\Omega,\mathbb{R}^N)\cap C^0(\overline{\Omega},\mathbb{R}^N) satisfying equation (1.7). Then it holds for any radius R>0R>0 such that G‾⊂{∣x∣<R}\overline{G}\subset\{|x|<R\}

u∈W1,1({∣x∣<R}−G‾,RN)(22)u\in W^{1,1}(\{|x|<R\}-\overline{G},\mathbb{R}^N) \tag*{(22)}

and the Sobolev trace of uu on ∂G=∂Ω\partial G=\partial\Omega coincides with the classical pointwise trace.

Remark 2.2. In the situation of Theorem 1.1 we can apply Lemma 2.1 to the functions uu and vv with the result that

Φ(u−v)∈W˚1,1({∣x∣<R}−G‾,RN)(23)\Phi(u-v)\in\mathring{W}^{1,1}(\{|x|<R\}-\overline{G},\mathbb{R}^N) \tag*{(23)}

holds for arbitrary Φ∈C01({∣x∣<R})\Phi\in C^1_0(\{|x|<R\}).

Proof of Lemma 2.1. We assume for notational simplicity that N=1N=1 and consider the case G=B1(0)={∣x∣<1}G=B_1(0)=\{|x|<1\}. Fix a radius R>1R>1 and choose a small number ε>0\varepsilon>0. We let η∈C01({∣x∣<2R})\eta\in C^1_0(\{|x|<2R\}) such that 0≤η≤10\le\eta\le1 and in addition

{η=1on [1+ε≤∣x∣≤R],η=0on [∣x∣≤1+ε/2],∣∇η∣≤c/εon [1+ε/2≤∣x∣≤1+ε],(24)\begin{cases} \eta=1 & \text{on } [1+\varepsilon\le|x|\le R],\\ \eta=0 & \text{on } [|x|\le1+\varepsilon/2],\\ |\nabla\eta|\le c/\varepsilon& \text{on } [1+\varepsilon/2\le|x|\le1+\varepsilon], \tag*{(24)} \end{cases}

cc denoting a positive constant independent of ϵ\epsilon. Noting that η2u\eta^2u belongs to C01(Ω)C^1_0(\Omega) with Ω=R2−G‾=[∣x∣>1]\Omega= \mathbb{R}^2-\overline{G}=[|x|>1] (recall (2)), we deduce from (1.7) combined with (2.3)

∫[1+ϵ/2<∣x∣<2R]∇f(∇u)⋅∇(η2u) dx=0.(25)\int_{[1+\epsilon/2<|x|<2R]} \nabla f(\nabla u)\cdot\nabla(\eta^2u)\,dx=0. \tag*{(25)}

Recalling the boundedness of ∇f\nabla f and using (24) one more time, we get from equation (25)

∫[1+ϵ<∣x∣<R]∇f(∇u)⋅∇u dx≤2∫[1+ϵ/2<∣x∣<2R]∣∇f(∇u)∣η∣∇η∣∣u∣ dx≤c(R)+c∫[1+ϵ/2<∣x∣<1+ϵ]∣∇η∣ dx\begin{aligned} \int_{[1+\epsilon<|x|<R]} \nabla f(\nabla u)\cdot\nabla u\,dx &\leq2\int_{[1+\epsilon/2<|x|<2R]}|\nabla f(\nabla u)|\eta|\nabla\eta||u|\,dx\\ &\leq c(R)+c\int_{[1+\epsilon/2<|x|<1+\epsilon]}|\nabla\eta|\,dx \end{aligned}

for a constant c(R)c(R) and another positive constant cc both being independent of ϵ\epsilon. Of course we have also used the boundedness of uu on the set [1≤∣x∣≤2R][1\leq|x|\leq2R]. This yields (recall the bound for ∣∇η∣|\nabla\eta| in (24))

∫[1+ϵ<∣x∣<R]∇f(∇u)⋅∇u dx≤K(26)\int_{[1+\epsilon<|x|<R]}\nabla f(\nabla u)\cdot\nabla u\,dx\leq K \tag*{(26)}

for all sufficiently small ϵ>0\epsilon>0 and with some number K∈(0,∞)K\in(0,\infty). Using the convexity of ff and condition (3) we finally observe that

∇f(∇u)⋅∇u≥f(∇u)−f(0)≥a∣∇u∣−b−f(0),\nabla f(\nabla u)\cdot\nabla u\geq f(\nabla u)-f(0)\geq a|\nabla u|-b-f(0),

and therefore (26) implies ∇u∈L1([1<∣x∣<R],R2)\nabla u\in L^1([1<|x|<R],\mathbb{R}^2), which proves (22). That the Sobolev trace on ∂Ω=[∣x∣=1]\partial\Omega=[|x|=1] of the function uu equals the classical trace then is a consequence of (22) combined with Theorem 3.6.2 and Theorem 3.6.3 (a) in the textbook [13] of Morrey.

Remark 2.3. Condition (4) is not needed during the proof of Lemma 2.1. Moreover, Lemma 2.1 is true for the nn-dimensional case, n≥2n\geq2.

Proof of Theorem 1.1. Let us assume that all the hypotheses hold, where for notational simplicity we let N=1N=1. We consider radii RR such that G‾⊂[∣x∣<R]\overline{G}\subset[|x|<R] and observe that according to Lemma 2.1 it holds u,v∈W1,1([∣x∣<2R]−G‾)u,v\in W^{1,1}([|x|<2R]-\overline{G}) and in addition

ϕ:=η2(u−v)∈W˚1,1([∣x∣<2R]−G‾),(27)\phi:=\eta^2(u-v)\in\mathring{W}^{1,1}([|x|<2R]-\overline{G}), \tag*{(27)}

where η∈C01([∣x∣<2R])\eta\in C^1_0([|x|<2R]) satisfies

{η=1 on [∣x∣≤R],0≤η≤1,∣∇η∣≤cR.(28)\begin{cases} \eta=1\text{ on }[|x|\leq R],\quad0\leq\eta\leq1,\\ |\nabla\eta|\leq\dfrac{c}{R}. \tag*{(28)} \end{cases}

For (27) we refer to Remark 2.2, in particular (23), with Φ\Phi being replaced by η2\eta^2. We have equation (7) for uu and vv with ϕ\phi being replaced by ϕ\phi defined in (27), which gives after subtracting the results

∫[∣x∣<2R]−G‾(∇f(∇u)−∇f(∇v))⋅∇(η2(u−v)) dx=0\int_{[|x|<2R]-\overline{G}}\left(\nabla f(\nabla u)-\nabla f(\nabla v)\right)\cdot\nabla\left(\eta^2(u-v)\right)\,dx=0

and in conclusion

∫[∣x∣<2R]−G‾[∇f(∇u)−∇f(∇v)]⋅(∇u−∇v)η2 dx=−2∫[∣x∣<2R]−G‾[∇f(∇u)−∇f(∇v)]⋅∇η η(u−v) dx.(29)\int_{[|x|<2R]-\overline{G}}\left[\nabla f(\nabla u)-\nabla f(\nabla v)\right]\cdot(\nabla u-\nabla v)\eta^2\,dx =-2\int_{[|x|<2R]-\overline{G}}\left[\nabla f(\nabla u)-\nabla f(\nabla v)\right]\cdot\nabla\eta\,\eta(u-v)\,dx. \tag*{(29)}

Consider the parameter dependent bilinear form

Bt,x(p,q):=D2f(∇v(x)+t(∇u(x)−∇v(x)))(p,q)B_{t,x}(p,q):=D^2f\bigl(\nabla v(x)+t(\nabla u(x)-\nabla v(x))\bigr)(p,q)

for p,q∈R2p,q\in\mathbb{R}^2 and parameters x,tx,t, and rewrite (29) in the form

∫{∣x∣<2R}∖G‾∫01Bt,x(∇u−∇v,∇u−∇v)η2 dt dx=−2∫{∣x∣<2R}∖G‾∫01Bt,x(η[∇u−∇v],(u−v)∇η) dt dx.(30)\int_{\{|x|<2R\}\setminus\overline{G}}\int_0^1 B_{t,x}(\nabla u-\nabla v,\nabla u-\nabla v)\eta^2\,dt\,dx = -2\int_{\{|x|<2R\}\setminus\overline{G}}\int_0^1 B_{t,x}(\eta[\nabla u-\nabla v],(u-v)\nabla\eta)\,dt\,dx. \tag*{(30)}

The Cauchy-Schwarz inequality applied to the positive (recall (5)) bilinear form Bt,xB_{t,x} shows

∣Bt,x(η[∇u−∇v],(u−v)∇η)∣≤ηBt,x(∇u−∇v,∇u−∇v)1/2∣u−v∣Bt,x(∇η,∇η)1/2,\left|B_{t,x}(\eta[\nabla u-\nabla v],(u-v)\nabla\eta)\right| \leq\eta B_{t,x}(\nabla u-\nabla v,\nabla u-\nabla v)^{1/2}|u-v|B_{t,x}(\nabla\eta,\nabla\eta)^{1/2},

hence by Young’s inequality we find that

∣Bt,x(η[∇u−∇v],(u−v)∇η)∣≤ϵη2Bt,x(∇u−∇v,∇u−∇v)+c(ϵ)(u−v)2Bt,x(∇η,∇η)\left|B_{t,x}(\eta[\nabla u-\nabla v],(u-v)\nabla\eta)\right| \leq\epsilon\eta^2B_{t,x}(\nabla u-\nabla v,\nabla u-\nabla v)+c(\epsilon)(u-v)^2B_{t,x}(\nabla\eta,\nabla\eta)

valid for all ϵ>0\epsilon>0. Using the latter inequality on the right-hand side of (30), we obtain after appropriate choice of ϵ\epsilon

∫{∣x∣<2R}∖G‾∫01η2Bt,x(∇u−∇v,∇u−∇v) dt dx≤c∫{∣x∣<2R}∖G‾∫01(u−v)2Bt,x(∇η,∇η) dt dx.\int_{\{|x|<2R\}\setminus\overline{G}}\int_0^1\eta^2B_{t,x}(\nabla u-\nabla v,\nabla u-\nabla v)\,dt\,dx \leq c\int_{\{|x|<2R\}\setminus\overline{G}}\int_0^1(u-v)^2B_{t,x}(\nabla\eta,\nabla\eta)\,dt\,dx.

Next we recall (4), use (28) together with assumption (9) or (10) in order to get for sufficiently large RR

∫{∣x∣<R}∖G‾∫01Bt,x(∇u−∇v,∇u−∇v) dt dx≤C(31)\int_{\{|x|<R\}\setminus\overline{G}}\int_0^1B_{t,x}(\nabla u-\nabla v,\nabla u-\nabla v)\,dt\,dx\leq C \tag*{(31)}

for some constant C∈(0,∞)C\in(0,\infty). With (31) we go back to (30) and recall that the Cauchy-Schwarz inequality applied to the right-hand side of (30) yields (also using (28))

∫{∣x∣<R}∖G‾∫01Bt,x(∇u−∇v,∇u−∇v) dt dx≤2∫{∣x∣<2R}∖G‾∫01Bt,x(∇u−∇v,∇u−∇v)1/2∣u−v∣Bt,x(∇η,∇η)1/2 dt dx,\int_{\{|x|<R\}\setminus\overline{G}}\int_0^1B_{t,x}(\nabla u-\nabla v,\nabla u-\nabla v)\,dt\,dx \leq 2\int_{\{|x|<2R\}\setminus\overline{G}}\int_0^1 B_{t,x}(\nabla u-\nabla v,\nabla u-\nabla v)^{1/2}|u-v|B_{t,x}(\nabla\eta,\nabla\eta)^{1/2}\,dt\,dx,

hence by Hölder’s inequality (and using spt⁡(∇η)⊂[R<∣x∣<2R]\operatorname{spt}(\nabla\eta)\subset[R<|x|<2R])

∫{∣x∣<R}∖G‾∫01Bt,x(∇u−∇v,∇u−∇v) dt dx≤(∫[R<∣x∣<2R]∖G‾∫01Bt,x(∇u−∇v,∇u−∇v) dt dx)1/2⋅(∫[R<∣x∣<2R]∖G‾∫01(u−v)2Bt,x(∇η,∇η) dt dx)1/2.(32)\int_{\{|x|<R\}\setminus\overline{G}}\int_0^1B_{t,x}(\nabla u-\nabla v,\nabla u-\nabla v)\,dt\,dx \leq \left(\int_{[R<|x|<2R]\setminus\overline{G}}\int_0^1B_{t,x}(\nabla u-\nabla v,\nabla u-\nabla v)\,dt\,dx\right)^{1/2} \cdot \left(\int_{[R<|x|<2R]\setminus\overline{G}}\int_0^1(u-v)^2B_{t,x}(\nabla\eta,\nabla\eta)\,dt\,dx\right)^{1/2}. \tag*{(32)}

From (31) it follows

∫R2∖G‾∫01Bt,x(∇u−∇v,∇u−∇v) dt dx<∞,\int_{\mathbb{R}^2\setminus\overline{G}}\int_0^1B_{t,x}(\nabla u-\nabla v,\nabla u-\nabla v)\,dt\,dx<\infty,

hence the first term on the right-hand side of (32) goes to zero as R→∞R\to\infty, whereas the second integral stays bounded according to the discussion after (30), which led us to (31). Altogether we conclude from (32) that

∫R2∖G‾∫01Bt,x(∇u−∇v,∇u−∇v) dt dx=0,\int_{\mathbb{R}^2\setminus\overline{G}}\int_0^1B_{t,x}(\nabla u-\nabla v,\nabla u-\nabla v)\,dt\,dx=0,

hence ∇u=∇v\nabla u=\nabla v according to (6), and (8) shows u=vu=v, which finishes the proof of Theorem 1.1.

For proving Corollary 1.2 we just note that in the case of entire solutions the previous calculations remain true, if we drop the set GG in all formulas.

Proof of Theorem 1.4. For simplicity we let N=1N=1 together with Ω=[∣x∣>1]\Omega=[|x|>1]. It is easy to see that under the assumptions and with the notation from Theorem 1.4 we can deduce from (27) the validity of (recall the weak form (7))

∫ΩD2f(∇u)(∇∂iu,∇φ) dx=0(33)\int_{\Omega}D^2f(\nabla u)(\nabla\partial_i u,\nabla\varphi)\,dx=0 \tag*{(33)}

for all φ∈C01(Ω)\varphi\in C^1_0(\Omega) and i∈{1,2}i\in\{1,2\} being fixed. Choosing η\eta as in the proof of Lemma 2.1 (see (24)) and letting φ=η2∂iu\varphi=\eta^2\partial_i u in equation (33), we obtain from (33), using the Cauchy-Schwarz inequality for the bilinear form D2f(∇u)D^2f(\nabla u) followed by an application of Young’s inequality, the estimate

∫Ωη2D2f(∇u)(∇∂iu,∇∂iu) dx≤c∫ΩD2f(∇u)(∇η,∇η)∣∂iu∣2 dx(34)\int_{\Omega}\eta^2D^2f(\nabla u)(\nabla\partial_i u,\nabla\partial_i u)\,dx\le c\int_{\Omega}D^2f(\nabla u)(\nabla\eta,\nabla\eta)|\partial_i u|^2\,dx \tag*{(34)}

with a uniform constant c∈(0,∞)c\in(0,\infty). Since we assume u∈C1(Ω‾)u\in C^1(\overline{\Omega}) and according to (24), it follows that we just have to find a bound independent of ϵ\epsilon for the quantity

I:=∫[1+ϵ/2<∣x∣<1+ϵ]D2f(∇u)(∇η,∇η)∣∂iu∣2 dx.I:=\int_{[1+\epsilon/2<|x|<1+\epsilon]}D^2f(\nabla u)(\nabla\eta,\nabla\eta)|\partial_i u|^2\,dx.

From assumption (16) we get (noting that here (4) is sufficient)

I≤ϵ−2∫[1+ϵ/2≤∣x∣≤1+ϵ]∣∇u∣2 dx,I\le\epsilon^{-2}\int_{[1+\epsilon/2\le|x|\le1+\epsilon]}|\nabla u|^2\,dx,

and (17) implies I≤CI\le C with C∈(0,∞)C\in(0,\infty) independent of ϵ\epsilon. Referring to (5) we find a number λ(R)∈(0,∞)\lambda(R)\in(0,\infty) such that on [1≤∣x∣≤R][1\le|x|\le R] it holds

D2f(∇u)(p,p)≥λ(R)∣p∣2,p∈R2;D^2f(\nabla u)(p,p)\ge\lambda(R)|p|^2,\qquad p\in\mathbb{R}^2;

the discussion concerning II turns (34) into

∫[1+ϵ<∣x∣≤R]D2f(∇u)(∇∂iu,∇∂iu) dx≤c(R)(35)\int_{[1+\epsilon<|x|\le R]}D^2f(\nabla u)(\nabla\partial_i u,\nabla\partial_i u)\,dx\le c(R) \tag*{(35)}

for any ϵ≪1\epsilon\ll1, and (35) clearly shows that

∂iu∈W1,2([1<∣x∣<R])(36)\partial_i u\in W^{1,2}([1<|x|<R]) \tag*{(36)}

for any R>1R>1 with vanishing Sobolev trace on [∣x∣=1][|x|=1] which follows from (17) (compare the remarks at the end of the proof of Lemma 2.1). In order to prove the constancy of uu, we again let R>0R>0 and choose η∈C01([∣x∣<2R])\eta\in C^1_0([|x|<2R]) with (28). The function φ=η2∂iu\varphi=\eta^2\partial_i u is admissible in equation (33): by the foregoing discussion (recall (36) together with ∂iu=0\partial_i u=0 on [∣x∣=1][|x|=1]) we have that η2∂iu∈W˚1,2([1<∣x∣<2R])\eta^2\partial_i u\in\mathring{W}^{1,2}([1<|x|<2R]), thus there is a sequence φm∈C0∞([1<∣x∣<2R])\varphi_m\in C^\infty_0([1<|x|<2R]) with the property φm→φ\varphi_m\to\varphi in the norm of W1,2([1<∣x∣<2R])W^{1,2}([1<|x|<2R]). Equation (33) holds for φm\varphi_m and by passing to the limit m→∞m\to\infty we get the desired result, from which we deduce (compare (29))

∫[1<∣x∣<2R]D2f(∇u)(∇∂iu,∇∂iu)η2 dx=−2∫[R<∣x∣<2R]D2f(∇u)(η∇∂iu,∂iu∇η) dx(37)\int_{[1<|x|<2R]}D^2f(\nabla u)(\nabla\partial_i u,\nabla\partial_i u)\eta^2\,dx=-2\int_{[R<|x|<2R]}D^2f(\nabla u)(\eta\nabla\partial_i u,\partial_i u\nabla\eta)\,dx \tag*{(37)}

where we have used the fact that spt⁡(∇η)⊂[R≤∣x∣≤2R]\operatorname{spt}(\nabla\eta) \subset[R \le|x| \le2R]. Applying the Cauchy-Schwarz inequality to D2f(∇u)D^2f(\nabla u) on the right-hand side of (37) and then using Hölder’s inequality we get from (37)

∫[1<∣x∣<2R]D2f(∇u)(∇∂iu,∇∂iu)η2 dx≤c(∫[R<∣x∣<2R]D2f(∇u)(∇∂iu,∇∂iu)η2 dx)1/2(∫[R<∣x∣<2R]D2f(∇u)(∇η,∇η)∣∂iu∣2 dx)1/2.(38)\int_{[1<|x|<2R]} D^2f(\nabla u)(\nabla\partial_i u,\nabla\partial_i u)\eta^2\,dx \le c\left(\int_{[R<|x|<2R]}D^2f(\nabla u)(\nabla\partial_i u,\nabla\partial_i u)\eta^2\,dx\right)^{1/2}\left(\int_{[R<|x|<2R]}D^2f(\nabla u)(\nabla\eta,\nabla\eta)|\partial_i u|^2\,dx\right)^{1/2}. \tag*{(38)}

with a constant c∈(0,∞)c\in(0,\infty) independent of RR. We claim

sup⁡R>1R−2∫[R<∣x∣<2R]∣∇u∣ dx<∞.(39)\sup_{R>1}R^{-2}\int_{[R<|x|<2R]}|\nabla u|\,dx<\infty. \tag*{(39)}

Assuming this for the moment, applying Young’s inequality to the right-hand side of (38) and using (16), we find that (39) implies (passing to R→∞R\to\infty)

∫[∣x∣>1]D2f(∇u)(∇∂iu,∇∂iu) dx<∞.(40)\int_{[|x|>1]}D^2f(\nabla u)(\nabla\partial_i u,\nabla\partial_i u)\,dx<\infty. \tag*{(40)}

Now, due to (39) and (40), the right-hand side of (38) actually vanishes as R→∞R\to\infty which means

∫[∣x∣>1]D2f(∇u)(∇∂iu,∇∂iu) dx=0,\int_{[|x|>1]}D^2f(\nabla u)(\nabla\partial_i u,\nabla\partial_i u)\,dx=0,

and ∇∂iu=0\nabla\partial_i u=0 is a consequence of (5). This yields the constancy of ∂iu\partial_i u, and from ∂iu=0\partial_i u=0 on [∣x∣=1][|x|=1] we get ∂iu=0\partial_i u=0 on Ω=[∣x∣>1]\Omega=[|x|>1]. Since ii was arbitrary, uu must be a constant function. It therefore remains to prove (39): let R>2R>2 and choose Ψ:R2→[0,1]\Psi:\mathbb{R}^2\to[0,1] such that Ψ∈C1\Psi\in C^1 and Ψ=1\Psi=1 on [R<∣x∣<2R][R<|x|<2R], Ψ=0\Psi=0 on [∣x∣≤34R]∪[∣x∣≥52R][|x|\le\frac{3}{4}R]\cup[|x|\ge\frac{5}{2}R], ∣∇Ψ∣≤c/R|\nabla\Psi|\le c/R. Then Ψ2u\Psi^2u is admissible in equation (7) and with the help of the boundedness of ∇f\nabla f we get

∫[R<∣x∣<2R]∇f(∇u)⋅∇u dx≤c∫spt⁡(∇Ψ)∣∇Ψ∣∣u∣ dx≤c∫[34R≤∣x∣≤52R]R−1∣u∣ dx≤cR2,\int_{[R<|x|<2R]}\nabla f(\nabla u)\cdot\nabla u\,dx\le c\int_{\operatorname{spt}(\nabla\Psi)}|\nabla\Psi||u|\,dx\le c\int_{[\frac{3}{4}R\le|x|\le\frac{5}{2}R]}R^{-1}|u|\,dx\le cR^2,

where the last inequality is a consequence of (18). Recalling that by the convexity of ff and assumption (3) it holds

∇f(∇u)⋅∇u≥f(∇u)−f(0)≥a∣∇u∣−b−f(0),\nabla f(\nabla u)\cdot\nabla u\ge f(\nabla u)-f(0)\ge a|\nabla u|-b-f(0),

we have shown (39), which completes the proof of Theorem 1.4. □\square

Remark 2.4. If GG is a bounded domain in R3\mathbb{R}^3, then an inspection of the proof shows, that the result of Theorem 1.4 still holds, if we replace (18) by the stronger condition that uu is a bounded function.

Proofs of Theorem 1.5 and 1.6

Just for notational simplicity we consider the scalar case N=1N=1 and recall definition (19) of the half-plane HH remarking that the particular choice H=[x1>0]H=[x_1>0] just enters through condition (20) from Theorem 1.6 and can easily be adjusted to other choices for $H`.

u∈W1,1(H∩BR(0)),BR(0):={∣x∣<R},(41)u \in W^{1,1}(H \cap B_R(0)), \qquad B_R(0) := \{ |x| < R \}, \tag*{(41)}

for any R∈(0,∞)R \in(0,\infty). For proving (41) we define (ϵ>0\epsilon> 0 sufficiently small)

η(x):=η1(x1)η2(x2),x∈R2,\eta(x) := \eta_1(x_1)\eta_2(x_2), \qquad x \in\mathbb{R}^2,

with functions ηi∈C01(R)\eta_i \in C^1_0(\mathbb{R}), 0≤ηi≤10 \leq\eta_i \leq1, i=1,2i = 1,2, such that

{spt⁡(η2)⊂[−2R,2R],η2=1 on [−R,R],∣η2′∣≤c/R,spt⁡(η1)⊂[ϵ/2,2R],η1=1 on [ϵ,R],∣η1′∣≤c/R on [R,2R],∣η1′∣≤c/ϵ on [ϵ/2,ϵ].(42)\begin{cases} \operatorname{spt}(\eta_2) \subset[-2R,2R], \quad\eta_2 = 1 \text{ on } [-R,R], \quad|\eta_2'| \leq c/R,\\ \operatorname{spt}(\eta_1) \subset[\epsilon/2,2R], \quad\eta_1 = 1 \text{ on } [\epsilon,R], \quad|\eta_1'| \leq c/R \text{ on } [R,2R], \quad|\eta_1'| \leq c/\epsilon\text{ on } [\epsilon/2,\epsilon]. \tag*{(42)} \end{cases}

From (42) we get η(x)=1\eta(x) = 1 if x∈[ϵ,R]×[−R,R]x \in[\epsilon,R] \times[-R,R], and η(x)≠0\eta(x) \ne0 implies x∈[ϵ/2,2R]×[−2R,2R]x \in[\epsilon/2,2R] \times[-2R,2R], hence spt⁡(η)\operatorname{spt}(\eta) is a compact subset of the open half-plane HH. Equation (25) is replaced by

∫[ϵ/2,2R]×[−2R,2R]∇f(∇u)⋅∇(η2u) dx=0\int_{[\epsilon/2,2R]\times[-2R,2R]} \nabla f(\nabla u) \cdot\nabla(\eta^2u)\,dx = 0

and following the lines after (25), we see that it remains to discuss the term

∫[ϵ/2,ϵ]×[−2R,2R]∣η1′(x1)∣ dx\int_{[\epsilon/2,\epsilon]\times[-2R,2R]} |\eta_1'(x_1)|\,dx

which however due to the bound imposed on η1′\eta_1' (see (42)) stays bounded independent of ϵ\epsilon. Thus we get u∈W1,1((0,R)×(−R,R))u \in W^{1,1}((0,R) \times(-R,R)) and in consequence (41) is true.

  • ii) From (41) it follows as before that the Sobolev trace of u−vu-v on ∂H\partial H vanishes, if we are in the situation of Theorem 1.1 with Ω=H\Omega= H and we deduce

ϕ:=η2(u−v)∈W˚1,1(H∩B2R(0)),(43)\phi:= \eta^2(u-v) \in\mathring{W}^{1,1}(H \cap B_{2R}(0)), \tag*{(43)}

where R>0R > 0 and η\eta now is defined as in (28). We have equation (6) for the functions uu and vv, which on account of (43) gives

∫H∩{∣x∣<2R}(∇f(∇u)−∇f(∇v))⋅∇ϕ dx=0.(44)\int_{H\cap\{|x|<2R\}}(\nabla f(\nabla u)-\nabla f(\nabla v))\cdot\nabla\phi\,dx=0. \tag*{(44)}

From (44) we obtain the half-plane version of (29), where one just replaces the domain of integration by the set H∩{∣x∣<2R}H \cap\{|x| < 2R\} and with obvious changes the proof of Theorem 1.1 can be finished as before.

  • iii) If we let Ω=H\Omega= H in the situation of Theorem 1.4, we first note that (36) has to be replaced by (i=1,2i = 1,2)

∂iu∈W1,2(H∩{∣x∣<R}).(45)\partial_i u \in W^{1,2}(H \cap\{|x| < R\}). \tag*{(45)}

For proving (45) we start from equation (33) by letting φ=η2∂iu\varphi= \eta^2\partial_i u, but now with the choice η(x):=η1(x1)η2(x2)\eta(x) := \eta_1(x_1)\eta_2(x_2) with η1,η2\eta_1,\eta_2 as in (42). The subsequent calculations are easily adjusted. This also concerns the second part of the proof of Theorem 1.4: we have equation (37) with H∩{∣x∣<2R}H \cap\{|x| < 2R\} as domain of integration and with η\eta exactly as in (28). Proceeding as in Theorem 1.4, we obtain the constancy of uu.

  • iv) We finally come to the proof of Theorem 1.6. We first observe that on account of (20) we obtain (45) just for the index i=1i = 1. For details we refer to the derivation of (35) (with i=1i = 1) noting that quantity II introduced after (34) involves ∣∂1u∣2|\partial_1u|^2, to which we

can apply (20). Of course, in (34) η\eta is defined according to (42). We then proceed as done after (36), letting i=1i = 1 and applying the adjustments from iii), and get that ∇∂1u=0\nabla\partial_1 u = 0, hence ∂1u=α\partial_1 u = \alpha for some α∈R\alpha\in\mathbb{R}. However, (20) shows α=0\alpha= 0 and uu just depends on x2x_2. From equation (6) it follows directly that we finally arrive at the claim of Theorem 1.6, for details we refer to the discussion after formula (27) of [14].

Michael Bildhauer

Saarland University Department of Mathematics P.O. Box 15 11 50 66041 Saarbrücken, Germany [email protected]

Martin Fuchs

Saarland University Department of Mathematics P.O. Box 15 11 50 66041 Saarbrücken, Germany [email protected]

References

  1. [1]R. Osserman, A survey of minimal surfaces. Second Edition, Dover Publications, New York 1986.
  2. [2]E. Kuwert, A bound for minimal graphs with a normal at infinity. Calc. Var. Partial Differential Equations 1 no.4 (1993), 407-416.DOI
  3. [3]N. Kutev, F. Tomi, Existence and nonexistence for the exterior Dirichlet problem for the minimal surface equation in the plane. Differential Integral Equations 11 no.6 (1998), 917-928.DOI
  4. [4]N. do Espírito-Santo, J. Ripoll, Some existence results on the exterior Dirichlet problem for the minimal hypersurface equation. Ann. Inst. H. Poincaré C Anal. Non Linéaire 28 no. 3 (2011), 385-393.
  5. [5]J. Ripoll, F. Tomi, On solutions to the exterior Dirichlet problem for the minimal surface equation with catenoidal ends. Adv. Calc. Var. 7 no. 2 (2014), 205-226.
  6. [6]A. Aiolfi, C. Assmann, J. Ripoll, The Dirichlet problem for the minimal surface equation on unbounded helicoidal domains of R³. arXiv 2023.
  7. [7]B. Dacorogna, Direct methods in the calculus of variations. Applied Mathematical Sciences 78, Springer Verlag, Berlin 1989.
  8. [8]M. Bildhauer, M. Fuchs, An extension of a theorem of Bers and Finn on the removability of isolated singularities to the Euler-Lagrange equations related to general linear growth problems. Calc. Var. Partial Differential Equations 61 no. 3 (2022), No. 99, 8 pp.arxiv.org/abs/2103.16133
  9. [9]J.-F. Hwang, Comparison principles and Liouville theorems for prescribed mean curvature equations in unbounded domains. Ann. Scuola Norm. Sup. Pisa Cl. Sci 15 no.4 (1988), 341-355.
  10. [10]J.-F. Hwang, A uniqueness theorem for the minimal surface equation. Pacific J. Math. 176 no.2 (1996), 357-364.DOI
  11. [11]P. Collin, R. Krust, Le problème de Dirichlet pour l’équation de surfaces minimales sur des domaines non bornés. Bull. Soc. Math. France 119 no.4 (1991), 443-462.
  12. [12]G. Jiang, Z. Wang, J. Zhu, Stability of Edelen-Wang’s Bernstein type theorem for the minimal surface equation. J. of Functional Analysis 284 (2023), 109821.DOI
  13. [13]C.B. Morrey, Multiple integrals in the calculus of variations. Springer Verlag, Berlin 2008.
  14. [14]M. Bildhauer, M. Fuchs, Variants of Bernstein’s theorem for variational integrals with linear and nearly linear growth. Ric. Mat. 73 no.5 (2024), 2911-2923.

Paper details

vlnc:W_gcwz889s
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
Usage

views · 0

downloads · 0

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