betaFeedback

WorksMathematics

Lehto-type growth conditions and a Picard theorem for the gradient of entire solutions to the Euler – Lagrange equations of variational integrals with linear growth in the plane including a small extension of Moser's Bernstein result

Abstract

We mainly consider entire solutions u:R2→Ru: \mathbb{R}^2 \rightarrow \mathbb{R} of the Euler – Lagrange equation for a variational integral with density g(∣∇u∣)g(|\nabla u|) of linear growth. Assuming a Lehto-type condition at infinity, we show that either ∇u\nabla u is a constant vector or the image of ∇u\nabla u is the whole plane with the possible exception of at most one point. This result is extended to certain energy densities of nearly linear growth. Moreover, we slightly generalize the Bernstein result of Moser.

Introduction

In our note we first discuss entire solutions u∈C2(R2)u \in C^{2}(\mathbb{R}^{2}) of the equation

div⁡(g′(∣∇u∣)∣∇u∣∇u)=0(1)\operatorname{div}\left(\frac{g'(|\nabla u|)}{|\nabla u|}\nabla u\right)=0 \tag*{(1)}

arising as the Euler equation of the variational integral ∫Ωg(∣∇u∣) dxdy\int_{\Omega} g(|\nabla u|)\,dxdy, where Ω\Omega is a domain in R2\mathbb{R}^{2}. Concerning the function g:[0,∞)→Rg : [0,\infty) \to\mathbb{R} we suppose that gg is of class C2([0,∞))C^{2}([0,\infty)) satisfying

g′(0)=0,g′′(t)>0,(2)g'(0)=0,\qquad g''(t)>0, \tag*{(2)}
m(t−1)≤g(t)≤M(t+1)(3)m(t-1)\leq g(t)\leq M(t+1) \tag*{(3)}

for all t≥0t\geq0 with numbers m,M>0m,M>0. We further assume (cc denoting a positive constant)

∃μ>1:g′′(t)≥c(1+t)−μfor all t≥0(4)\exists\mu>1:g''(t)\geq c(1+t)^{-\mu}\quad\text{for all }t\geq0 \tag*{(4)}

and finally add the technical condition

tg′′(t)≤g′(t),t≥0.(5)tg''(t)\leq g'(t),\qquad t\geq0. \tag*{(5)}

Evidently the second part of (1.2) is just a consequence of inequality (1.4), and (1.5) will follow, if for instance we require

g(3)(t)≤0for all t≥0.g^{(3)}(t)\leq0\quad\text{for all }t\geq0.

For the choice g(t)=1+t2g(t)=\sqrt{1+t^2} we have the nonparametric minimal surface case for which (4) holds with μ=3\mu=3, and as it was shown by Bernstein in his famous papers [1], [2], the only entire solutions are the affine functions R2→R\mathbb{R}^2\to\mathbb{R}. This Bernstein property of equation (1) under the assumptions (2), (3) combined with the reverse inequality in (4) was extended independently by Mikljukov [3] and by Farina, Sciunzi and Valdinoci [4]: in [3] the author considers the case μ≥2\mu\geq2 actually working with the weaker condition

∫0∞tg′′(t) dt<∞\int_{0}^{\infty}tg^{\prime\prime}(t)\,dt<\infty

replacing the opposite inequality in (4), whereas in [4] the authors can cover the case of exponents μ≥3\mu\geq3. The reader being in addition interested in the history of Bernstein’s problem is also referred to the papers [3], [4], and in our note [5] we summarized both approaches in the setting of equation (1), but – as we emphasize again – with (4) being replaced by the opposite inequality.

To our knowledge the validity of Bernstein’s theorem in the presence of condition (4) is an unsolved problem to which we can not present any substantial contribution. Quite the contrary, we start with the discussion of a rather strange behaviour of entire solutions uu, whose gradients do not grow too fast at infinity: in this case either uu is an affine function or the image of ∇u\nabla u is the whole plane with at most one exceptional point. More precisely, it holds

Theorem 1.1. Let the conditions (2) – (5) hold for the function gg, where in (4) we recall that μ>1\mu>1. Assume that u∈C2(R2)u\in C^2(\mathbb{R}^2) is an entire solution of (1) such that

∣∇u(z)∣≤c[ln⁡(2+∣z∣)]1μ−1,z∈R2,(6)|\nabla u(z)|\leq c[\ln(2+|z|)]^{\frac{1}{\mu-1}},\qquad z\in\mathbb{R}^2, \tag*{(6)}

with some constant c>0c>0. Then we have

either: ∇u≡const(“Bernstein case”)\text{either: }\nabla u\equiv\mathrm{const}\quad\text{(“Bernstein case”)}

or: ∇u(R2)⊃C−{p}\nabla u(\mathbb{R}^2)\supset\mathbb{C}-\{p\} for a point p∈R2p\in\mathbb{R}^2.

Remark 1.2. In the second case of Theorem 1.1 the situation is comparable to the so-called little Picard theorem from complex analysis (see, e.g. [6], Theorem 16.22, p. 357).

Remark 1.3. We have no idea, if the “or” case in Theorem 1.1 actually occurs. However, under rather strong assumptions, it can be excluded, we refer to Theorem 1.8 and Theorem 1.10 below.

Remark 1.4. As we will discuss later, condition (6) can be replaced by the weaker requirement

∫1∞1rK(r) dr=∞,(7)\int_{1}^{\infty}\frac{1}{rK(r)}\,dr=\infty, \tag*{(7)}

where K(r):=∫02π(1+∣∇u(reiθ)∣)μ−1 dθK(r):=\int_{0}^{2\pi}(1+|\nabla u(re^{i\theta})|)^{\mu-1}\,d\theta. The divergence of the integral guarantees the validity of the famous Lehto-condition for quasiconformal mappings (see, e.g., [7] or [8]) being of crucial importance for the proof of Theorem 1.1.

Remark 1.5. We note that in the situation of (6) the solution can not grow much faster than an affine function.

Let us now briefly look at a more general setting: instead of using the radial Euler–Lagrange equation (1), we consider the more general divergence equation

div⁡[Df(∇u)]=0on R2(8)\operatorname{div}[Df(\nabla u)]=0\quad\text{on }\mathbb{R}^2 \tag*{(8)}

with u∈C2(R2)u\in C^2(\mathbb{R}^2) and an integrand f∈C2(R2)f\in C^2(\mathbb{R}^2), f=f(P)f=f(P). For some μ>1\mu>1 and with constants c1,c2>0c_1,c_2>0, let the following ellipticity condition hold:

c1(1+∣P∣)−μ∣Q∣2≤D2f(P)(Q,Q)≤c2∣Q∣21+∣P∣(9)c_1(1+|P|)^{-\mu}|Q|^2\leq D^2f(P)(Q,Q)\leq c_2\frac{|Q|^2}{1+|P|} \tag*{(9)}

for all P,Q∈R2P,Q \in\mathbb{R}^{2}. Inequality (1.9) is just the condition of μ\mu-ellipticity introduced in the paper [9]. We have in extension of Theorem 1.1

Theorem 1.6. Let u∈C2(R2)u \in C^{2}(\mathbb{R}^{2}) denote an entire solution of equation (8) with ff satisfying (1.9). If in addition we have inequality (1.6), then the conclusions of Theorem 1.1 hold.

Remark 1.7. Clearly the conditions (1.2), (1.3), (1.4) and (1.5) imposed on the function gg yield inequality (1.9) for the density f(P):=g(t)f(P):=g(t), where tt is the norm of PP. We note that (1.9) inherits more flexibility, for instance, ff not necessarily is of linear growth, however, the case of integrands with radial structure seems to be of more geometric interest.

Coming back to Bernstein’s problem, we recall the famous result of Moser [10] stating that if in Theorem 1.1 and Theorem 1.6 condition (1.6) is replaced by the global boundedness of the gradient of the solution uu, then uu is an affine function. We have the following minor extension of Moser’s theorem:

Theorem 1.8. Let u∈C2(R2)u \in C^{2}(\mathbb{R}^{2}) denote an entire solution of equation (8) with ff satisfying (1.9). If inequality (1.6) is replaced by the condition

∣∇u(z)∣≤cln⁡(ln⁡∣z∣)(10)|\nabla u(z)| \leq c\ln(\ln|z|) \tag*{(10)}

with some constant c>0c>0 and for all points z∈R2z \in\mathbb{R}^{2} such that ∣z∣|z| is sufficiently large, then uu is an affine function.

Remark 1.9. Condition (1.10) is much more restrictive than the upper bound from (1.6), however, from the proof of Theorem 1.8 we get

Theorem 1.10. Let the Landau oo-condition for the gradient at infinity hold, i.e.

∣∇u(z)∣=o([ln⁡(2+∣z∣)]1μ−1)as ∣z∣→∞,(11)|\nabla u(z)|=o\left([\ln(2+|z|)]^{\frac{1}{\mu-1}}\right) \quad\text{as } |z| \to\infty, \tag*{(11)}

where u∈C2(R2)u \in C^{2}(\mathbb{R}^{2}) denotes an entire solution of equation (8) with ff satisfying (1.9). Then uu must be affine.

Our paper is organized as follows: in Section 2 we transform (1.1) into a generalized Beltrami equation and collect some technical tools. The proof of Theorem 1.1 is outlined in Section 3 making essential use of results due to Lehto and Virtanen presented in the monograph [7], and our line of arguments will mainly follow ideas taken from the work [8] of Astala, Iwaniec and Martin. In Section 4 we will add some short comments concerning the modifications necessary for the proof of Theorem 1.6, and Section 5 contains the proof of Theorem 1.8, the proof of Theorem 1.10 being presented in Section 6.

Transformation of equation (1.1) into a generalized Beltrami equation

Consider u∈C2(R2)u \in C^{2}(\mathbb{R}^{2}) satisfying (1.1) and let the function gg just denote an element of C2([0,∞))C^{2}([0,\infty)) satisfying the condition (1.2). We define the quantities

p:=∂∂xu=ux,q:=uy,t:=∣∇u∣=p2+q2,p:=\frac{\partial}{\partial x}u=u_x,\qquad q:=u_y,\qquad t:=|\nabla u|=\sqrt{p^2+q^2},

where z=(x,y)=x+iy∈Cz=(x,y)=x+iy \in\mathbb{C}, u=u(x,y)u=u(x,y). Clearly we have the identity

py−qx=0,(12)p_y-q_x=0, \tag*{(12)}

and (1.1) turns into

∂x(g′(t)tp)+∂y(g′(t)tq)=0.(13)\partial_x\left(\frac{g'(t)}{t}p\right)+\partial_y\left(\frac{g'(t)}{t}q\right)=0. \tag*{(13)}

From (2.1), (2.2) we infer

a11px+2a12py+a22qy=0(14)a_{11}p_x+2a_{12}p_y+a_{22}q_y=0 \tag*{(14)}

with coefficients defined through the formulas

{a11:=1tg′(t)+1t2p2[g′′(t)−1tg′(t)],a22:=1tg′(t)+1t2q2[g′′(t)−1tg′(t)],a12:=a21:=1t2pq[g′′(t)−1tg′(t)].(15)\begin{cases} a_{11}:=\frac{1}{t}g'(t)+\frac{1}{t^2}p^2\left[g''(t)-\frac{1}{t}g'(t)\right],\\ a_{22}:=\frac{1}{t}g'(t)+\frac{1}{t^2}q^2\left[g''(t)-\frac{1}{t}g'(t)\right],\\ a_{12}:=a_{21}:=\frac{1}{t^2}pq\left[g''(t)-\frac{1}{t}g'(t)\right]. \tag*{(15)} \end{cases}

Quoting (1.2) it follows from (2.4) that

a11+a22=g′′(t)+1tg′(t)>0(16)a_{11}+a_{22}=g''(t)+\frac{1}{t}g'(t)>0 \tag*{(16)}

holds, where in case t=0t=0 the quantity 1tg′(t)\frac{1}{t}g'(t) is just g′′(0)g''(0).

Let us consider the function

Φ(z):=Φ(x,y):=p(x,y)−iq(x,y)(17)\Phi(z):=\Phi(x,y):=p(x,y)-iq(x,y) \tag*{(17)}

and introduce the Wirtinger derivatives

∂z:=12(∂x−i∂y),∂zˉ:=12(∂x+i∂y).\partial_z:=\frac{1}{2}(\partial_x-i\partial_y),\qquad\partial_{\bar z}:=\frac{1}{2}(\partial_x+i\partial_y).

Then we have

Lemma 2.1. The function Φ\Phi from (2.6) is a solution of the generalized Beltrami equation

∂zˉΦ=K1∂zΦ+K2∂zΦ‾on C(18)\partial_{\bar z}\Phi=K_1\partial_z\Phi+K_2\overline{\partial_z\Phi}\quad\text{on }\mathbb{C} \tag*{(18)}

with coefficients (recall (2.5))

K1=12(a22−a11)−ia12a11+a22,K2=12(a22−a11)+ia12a11+a22.(19)K_1=\frac{\frac{1}{2}(a_{22}-a_{11})-ia_{12}}{a_{11}+a_{22}},\qquad K_2=\frac{\frac{1}{2}(a_{22}-a_{11})+ia_{12}}{a_{11}+a_{22}}. \tag*{(19)}

Proof of Lemma 2.1. We express the quantities pxp_x, qyq_y, pyp_y in terms of ∂zΦ\partial_z\Phi, ∂zˉΦ\partial_{\bar z}\Phi, ∂zΦ‾\overline{\partial_z\Phi} and return to (2.3) with the result

a11(∂zˉΦ+12∂zΦ+12∂zΦ‾)+2a12(i2∂zΦ−i2∂zΦ‾)+a22(∂zˉΦ−12∂zΦ−12∂zΦ‾)=0.a_{11}\left(\partial_{\bar z}\Phi+\frac{1}{2}\partial_z\Phi+\frac{1}{2}\overline{\partial_z\Phi}\right)+ 2a_{12}\left(\frac{i}{2}\partial_z\Phi-\frac{i}{2}\overline{\partial_z\Phi}\right)+ a_{22}\left(\partial_{\bar z}\Phi-\frac{1}{2}\partial_z\Phi-\frac{1}{2}\overline{\partial_z\Phi}\right)=0.

After collecting terms this immediately yields equation (2.7) with K1K_1 and K2K_2 according to (2.8).

For preparing the next auxiliary result we remark that the eigenvalues of the symmetric matrix (aij)1≤i,j≤2(a_{ij})_{1\leq i,j\leq2} are given by

λ=g′′(t)>0,Λ=g′(t)t>0(20)\lambda=g''(t)>0,\qquad\Lambda=\frac{g'(t)}{t}>0 \tag*{(20)}

being strictly positive on account of (1.2), hence the matrix of the coefficients aija_{ij} is positive definite.

Lemma 2.2. The determinant of the Jacobian matrix of Φ\Phi from (17) is nonnegative, i.e.

det⁡DΦ=det⁡(px−qxpy−qy)≥0.\det D\Phi= \det\begin{pmatrix} p_x & -q_x \\ p_y & -q_y \end{pmatrix} \ge0.

Proof of Lemma 2.2. It holds on account of (12)

det⁡DΦ=−pxqy+pyqx=py2−pxqy.\det D\Phi= -p_xq_y + p_yq_x = p_y^2 - p_xq_y.

Recalling (15) we have

a22=1t2q2g′′(t)+1tg′(t)(1−q2t2)>0,a_{22} = \frac{1}{t^2}q^2g^{\prime\prime}(t) + \frac{1}{t}g^\prime(t)\left(1-\frac{q^2}{t^2}\right) > 0,

and (14) implies

qy=−a11a22px−2a12a22py,q_y = -\frac{a_{11}}{a_{22}}p_x - 2\frac{a_{12}}{a_{22}}p_y,

hence

det⁡DΦ=py2+a11a22px2+2a12a22pxpy\det D\Phi= p_y^2 + \frac{a_{11}}{a_{22}}p_x^2 + 2\frac{a_{12}}{a_{22}}p_xp_y
=1a22(a22py2+a11px2+2a12pxpy)≥0= \frac{1}{a_{22}}\left(a_{22}p_y^2+a_{11}p_x^2+2a_{12}p_xp_y\right)\ge0

which proves our claim.

We finish this section by adding a few technical comments: returning to the defining equations (19) it is immediate that K2=K1‾K_2=\overline{K_1}, moreover, we have the identity

∣K1∣2=14(a22−a11)2+a122(a11+a22)2=14(a11+a22)2−a11a22+a122(a11+a22)2=14−a11a22−a122(a11+a22)2\begin{aligned} |K_1|^2 = \frac{\frac{1}{4}(a_{22}-a_{11})^2+a_{12}^2}{(a_{11}+a_{22})^2} = \frac{\frac{1}{4}(a_{11}+a_{22})^2-a_{11}a_{22}+a_{12}^2}{(a_{11}+a_{22})^2} \\ &= \frac{1}{4}-\frac{a_{11}a_{22}-a_{12}^2}{(a_{11}+a_{22})^2} \end{aligned}

hence

∣K1∣2=14−det⁡(aij)(trace⁡(aij))2=14−λΛ(λ+Λ)2<14|K_1|^2 = \frac{1}{4}-\frac{\det(a_{ij})}{(\operatorname{trace}(a_{ij}))^2} = \frac{1}{4}-\frac{\lambda\Lambda}{(\lambda+\Lambda)^2}<\frac{1}{4}

on account of (20), and we find

∣K1∣<12.(21)|K_1|<\frac{1}{2}. \tag*{(21)}

Proof of Theorem 1.1

Let the conditions (2)–(5) hold for the function gg and consider an entire solution u∈C2(R2)u\in C^2(\mathbb{R}^2) of equation (1) for which we have (6). We consider two cases:

Case 1: Φ\Phi is a constant function.

Then our claim is a direct consequence of equation (17).

Case 2: Φ\Phi is not constant.

In order to prove the Picard-property of the gradient mapping, we will apply techniques outlined in [8], Chapter 20.9, demonstrating the strength of the Lehto-condition . According to Lemma 2.1, the complex gradient mapping Φ(z)=p(x,y)−iq(x,y)\Phi(z)=p(x,y)-iq(x,y) is of class C1(R2)C^1(\mathbb{R}^2) and satisfies the generalized Beltrami-type system (recall (18))

∂zˉΦ(z)=K1(z)∂zΦ(z)+K2(z)∂zˉΦ(z)on C.\partial_{\bar z}\Phi(z)=K_1(z)\partial_z\Phi(z)+K_2(z)\partial_{\bar z}\Phi(z)\quad\text{on }\mathbb{C}.

Due to (21) we have the global uniform bound ∣K1(z)∣<12|K_1(z)|<\frac{1}{2} everywhere. By combining this with the nonnegativity of the Jacobian determinant, det⁡DΦ≥0\det D\Phi\ge0 (Lemma 2.2), we obtain the distortion inequality for the Wirtinger derivatives:

∣∂zˉΦ(z)∣≤(∣K1(z)∣+∣K2(z)∣)∣∂zΦ(z)∣.|\partial_{\bar z}\Phi(z)|\leq\left(|K_1(z)|+|K_2(z)|\right)|\partial_z\Phi(z)|.

We define the local maximal dilatation quotient DΦ(z)D_\Phi(z). In view of condition (1.5), it satisfies everywhere the upper bound:

DΦ(z):=∣∂zΦ∣+∣∂zˉΦ∣∣∂zΦ∣−∣∂zˉΦ∣≤Λλ=g′(∣∇u∣)∣∇u∣g′′(∣∇u∣).D_\Phi(z):=\frac{|\partial_z\Phi|+|\partial_{\bar z}\Phi|}{|\partial_z\Phi|-|\partial_{\bar z}\Phi|}\leq\frac{\Lambda}{\lambda}=\frac{g'(|\nabla u|)}{|\nabla u|g''(|\nabla u|)}.

Since gg is a strictly convex function of linear growth and u∈C2(R2)u\in C^2(\mathbb{R}^2), the maximal dilatation DΦ(z)D_\Phi(z) is locally bounded. Consequently, Φ\Phi is a classical mapping of bounded distortion. By Reshetnyak’s regularity theorem, any non-constant solution of the elliptic system inherits the qualitative topological properties of holomorphic functions. Namely, the gradient mapping Φ:C→C\Phi:\mathbb{C}\to\mathbb{C} is both open and discrete.

In order to show in more detail that the gradient mapping Φ:C→C\Phi:\mathbb{C}\to\mathbb{C} is both open and discrete, we can refer to Theorem 1.1 from the work [11] of Manfredi and Villamor or to the paper of Iwaniec and Sverak [12]. For quoting the second reference we just note that a mapping being locally open and discrete clearly inherits these properties globally.

Since Φ\Phi is discrete and open, we can apply the generalized topological factorization theorem of Stoilow as for instance it is stated on p. 305 in the work of Luisto and Pankka [13]: there exists a decomposition of the form

Φ(z)=Ψ(W(z)).\Phi(z)=\Psi(W(z)).

where:

  1. W:C→Ω⊂CW:\mathbb{C}\to\Omega\subset\mathbb{C} is a global quasiconformal homeomorphism satisfying the Beltrami equation ∂zˉW(z)=ν(z)∂zW(z)\partial_{\bar z}W(z)=\nu(z)\partial_zW(z) everywhere, with the complex dilatation defined by ν(z)=∂zˉΦ/∂zΦ\nu(z)=\partial_{\bar z}\Phi/\partial_z\Phi on the set where ∂zΦ≠0\partial_z\Phi\ne0.

  1. Ψ:Ω→C\Psi:\Omega\to\mathbb{C} is a holomorphic function defined on the image domain Ω=W(C)\Omega=W(\mathbb{C}).

In addition we refer to the analytic version of the factorization theorem presented in [8], which keeps the degree of regularity.

The maximal dilatation of the global homeomorphism WW is pointwise bounded by the dilatation of the gradient, i.e., DW(z)≤DΦ(z)D_W(z)\leq D_\Phi(z). For later purposes we recall the Lehto-type asymptotic growth condition: our assumption (1.6) implies the divergence of the integrated dilatation over concentric circles, more precisely we have

∫1∞1rK(r) dr=∞,where K(r):=∫02πDΦ(reiθ) dθ.(22)\int_1^\infty\frac{1}{rK(r)}\,dr=\infty,\qquad\text{where }K(r):=\int_0^{2\pi}D_\Phi(re^{i\theta})\,d\theta. \tag*{(22)}

In order to continue we need to know that the image domain of the quasiconformal homeomorphism WW is the complex plane:

Ω=W(C)=C.(23)\Omega=W(\mathbb{C})=\mathbb{C}. \tag*{(23)}

However, we can not immediately guarantee that (23) holds. We proceed by contradiction: assume that the image domain Ω=W(C)\Omega=W(\mathbb{C}) is not the entire complex plane, i.e., Ω≠C\Omega\ne\mathbb{C}. Since W:C→ΩW:\mathbb{C}\to\Omega is a global homeomorphism, Ω\Omega is a proper, simply connected domain in C\mathbb{C}. By the Riemann mapping theorem (see [6], Theorem 14.8, p. 302) there exists a biholomorphic (conformal) mapping h:Ω→Dh:\Omega\to\mathbb{D} onto the open unit disk D={w∈C:∣w∣<1}\mathbb{D}=\{w\in\mathbb{C}:|w|<1\}.

Let us define the composite mapping F:C→DF:\mathbb{C}\to\mathbb{D} as F(z)=(h∘W)(z)F(z)=(h\circ W)(z). By construction, FF is a homeomorphism from C\mathbb{C} onto D\mathbb{D}, meaning F(C)=DF(\mathbb{C})=\mathbb{D}. Since hh is conformal, the complex dilatation is preserved, and the maximal dilatation satisfies DF(z)=DW(z)≤DΦ(z)D_F(z)=D_W(z)\le D_\Phi(z) pointwise. Thus, FF satisfies the Lehto growth condition (22) on C\mathbb{C}.

For any radius R>1R>1, consider the annulus A(1,R)={z∈C:1<∣z∣<R}A(1,R)=\{z\in\mathbb{C}:1<|z|<R\}. Its image under FF is a topological ring domain ΩR=F(A(1,R))⊂D\Omega_R=F(A(1,R))\subset\mathbb{D}. The complement of ΩR\Omega_R in the entire complex plane, C∖ΩR\mathbb{C}\setminus\Omega_R, consists of exactly two disjoint connected components:

  • ER:=F(D‾1)E_R:=F(\overline{\mathbb{D}}_1), which is the image of the closed unit disk. Since D‾1\overline{\mathbb{D}}_1 is compact and FF is continuous, ERE_R is a compact, bounded set in D\mathbb{D}. Evidently, ERE_R is independent of RR. Since FF is a homeomorphism, d=diam⁡(ER)>0d=\operatorname{diam}(E_R)>0 is a fixed positive constant.

  • FR:=(C∖D)∪F({z∈C:∣z∣≥R})F_R:=(\mathbb{C}\setminus\mathbb{D})\cup F(\{z\in\mathbb{C}:|z|\ge R\}). This component is closed and unbounded because it contains the entire exterior of the unit disk, C∖D\mathbb{C}\setminus\mathbb{D}.

The geometric assumptions of Theorem 20.9.3 are satisfied for the ring domain ΩR=C∖(ER∪FR)\Omega_R=\mathbb{C}\setminus(E_R\cup F_R). Therefore, there exists an increasing homeomorphism ϕ:[0,∞]→[0,∞]\phi:[0,\infty]\to[0,\infty], independent of the continua, such that

Mod⁡(ΩR)≤ϕ(dist⁡(ER,FR)diam⁡(ER)).\operatorname{Mod}(\Omega_R)\le\phi\left(\frac{\operatorname{dist}(E_R,F_R)}{\operatorname{diam}(E_R)}\right).

Since ER⊂DE_R\subset\mathbb{D} and C∖D⊂FR\mathbb{C}\setminus\mathbb{D}\subset F_R, the Euclidean distance between ERE_R and FRF_R cannot exceed the distance from ERE_R to the boundary ∂D\partial\mathbb{D}. The unit disk has diameter 22, hence we have dist⁡(ER,FR)≤2\operatorname{dist}(E_R,F_R)\le2. By the monotonicity of ϕ\phi, we obtain a uniform upper bound independent of RR:

Mod⁡(ΩR)≤ϕ(2d)=C<∞for all R>1.(24)\operatorname{Mod}(\Omega_R)\le\phi\left(\frac{2}{d}\right)=C<\infty\quad\text{for all }R>1. \tag*{(24)}

On the other hand, the modulus estimate from below for mappings with integrable dilatation (compare [8], Corollary 20.9.2) provides the following lower bound for the conformal modulus of ΩR\Omega_R

Mod⁡(ΩR)=Mod⁡(F(A(1,R)))≥12π∫1R1rK(r) dr,(25)\begin{aligned} \operatorname{Mod}(\Omega_R)&=\operatorname{Mod}(F(A(1,R)))\\ &\ge\frac{1}{2\pi}\int_1^R\frac{1}{rK(r)}\,dr, \tag*{(25)} \end{aligned}

where the integrated maximal dilatation is defined as

K(r)=∫02πDΦ(reiθ) dθ.K(r)=\int_0^{2\pi}D_\Phi(re^{i\theta})\,d\theta.

By the Lehto condition (22), the integral on the right-hand side of (25) diverges to infinity as R→∞R\to\infty. Thus, by taking the limit, we obtain

lim⁡R→∞Mod⁡(ΩR)=∞.(26)\lim_{R\to\infty}\operatorname{Mod}(\Omega_R)=\infty. \tag*{(26)}

The divergence stated in (26) contradicts the uniform upper bound established in (24), which requires Mod⁡(ΩR)≤C\operatorname{Mod}(\Omega_R)\le C to hold globally for all R>1R>1.

Consequently, the initial assumption Ω≠C\Omega\ne\mathbb{C} was false, which completes the proof of

Ω=W(C)=C.\Omega=W(\mathbb{C})=\mathbb{C}.

Now, on account of (23), Ψ(z)=Φ(W−1(z))\Psi(z)=\Phi(W^{-1}(z)) is defined for any complex number zz and thereby a classical entire holomorphic function. In consequence, Picard’s little theorem applies to Ψ\Psi and we arrive at the ”or” case from Theorem 1.1.

Remarks concerning Theorem 1.6

The coefficient matrix in this new setting is just the Hessian matrix of ff: we define

aij(z):=∂2f∂Pi∂Pj(∇u(z))(i,j=1,2).a_{ij}(z) := \frac{\partial^{2}f}{\partial P_i\partial P_j}(\nabla u(z)) \quad(i,j=1,2).

Inequality (16) requires the strict positivity of the trace of the coefficient matrix, i.e., a11+a22>0a_{11}+a_{22}>0. Since aija_{ij} represents the components of the Hessian matrix D2f(∇u)D^{2}f(\nabla u), the sum a11+a22a_{11}+a_{22} corresponds exactly to trace⁡(D2f)\operatorname{trace}(D^{2}f). From condition (1.9), it follows for the eigenvalues λ1,λ2\lambda_1,\lambda_2 of D2fD^{2}f that:

a11+a22=λ1+λ2≥2c1(1+∣∇u∣)−μ>0a_{11}+a_{22}=\lambda_1+\lambda_2\ge2c_1(1+|\nabla u|)^{-\mu}>0

Since c1>0c_1>0, this sum is strictly positive everywhere. Consequently, (16) holds in the new setting.

The transformation of the Euler–Lagrange equation (1.8) into the generalized Beltrami equation

∂zΦ=K1∂z‾Φ+K2∂zΦ‾\partial_z\Phi=K_1\partial_{\overline{z}}\Phi+K_2\overline{\partial_z\Phi}

is of purely algebraic nature: expanding equation (1.8) div⁡[Df(∇u)]=0\operatorname{div}[Df(\nabla u)]=0 via the chain rule yields, due to the symmetry of the mixed derivatives (fP1P2=fP2P1f_{P_1P_2}=f_{P_2P_1}):

a11px+2a12py+a22qy=0a_{11}p_x+2a_{12}p_y+a_{22}q_y=0

This matches equation (14) exactly. The proof of Lemma 2.1 thus remains valid without modification, and the coefficients K1,K2K_1,K_2 possess the identical structure as in (2.8). We check the validity of Lemma 2.2 and inequality (2.10):

  • Lemma 2.2 (det⁡DΦ≥0\det D\Phi\ge0): The proof just uses relation (2.3) and the positive definiteness of the coefficient matrix. Since condition (1.9) guarantees the positive definiteness of D2fD^{2}f, the Jacobian determinant remains non-negative.

  • Equation (2.10) (∣K1∣<12|K_1|<\frac{1}{2}): Due to the real-valuedness of ff, the coefficients aija_{ij} are real, which implies K2=K1‾K_2=\overline{K_1}. For the absolute value, it holds that:

∣K1∣2=14−det⁡(aij)(trace⁡(aij))2=14−λ1λ2(λ1+λ2)2|K_1|^2=\frac{1}{4}-\frac{\det(a_{ij})}{(\operatorname{trace}(a_{ij}))^2}=\frac{1}{4}-\frac{\lambda_1\lambda_2}{(\lambda_1+\lambda_2)^2}

By the strict ellipticity, the eigenvalues satisfy λ1,λ2>0\lambda_1,\lambda_2>0, and consequently λ1λ2>0\lambda_1\lambda_2>0. Thus, a strictly positive value is subtracted from 14\frac{1}{4}, ensuring that ∣K1∣<12|K_1|<\frac{1}{2} is always satisfied.

Since det⁡DΦ≥0\det D\Phi\ge0 and ∣K1∣<12|K_1|<\frac{1}{2}, Φ\Phi is a mapping of bounded distortion, and as before it can be factored as Φ(z)=Ψ(W(z))\Phi(z)=\Psi(W(z)) with a quasiconformal homeomorphism WW and a holomorphic function Ψ\Psi.

The maximal dilatation satisfies:

DΦ(z)≤λmax⁡(D2f)λmin⁡(D2f)=c2(1+∣∇u∣)−1c1(1+∣∇u∣)−μ=c2c1(1+∣∇u∣)μ−1D_{\Phi}(z)\le\frac{\lambda_{\max}(D^{2}f)}{\lambda_{\min}(D^{2}f)} =\frac{c_2(1+|\nabla u|)^{-1}}{c_1(1+|\nabla u|)^{-\mu}} =\frac{c_2}{c_1}(1+|\nabla u|)^{\mu-1}

This corresponds precisely to the asymptotic growth behavior found in the radial case. Under the assumption of the growth condition (1.6) for the gradient, the integrated dilatation grows logarithmically, whereby the Lehto condition is satisfied and Ω=W(C)=C\Omega=W(\mathbb{C})=\mathbb{C} follows, provided we assume that Φ\Phi is not constant. Thus, the arguments in Section 3 carry over completely.

Proof of Theorem 1.8

As in § 3 we argue by contradiction assuming that the complex gradient mapping Φ\Phi is not constant. With the help of (1.10) we will derive a precise asymptotic lower bound for the global quasiconformal homeomorphism W(z)W(z) for large values of ∣z∣|z| leading to the desired contradiction.

From the former growth condition (1.6) for the gradient,

∣∇u(z)∣≤c[ln⁡(2+∣z∣)]1μ−1,z∈R2,|\nabla u(z)| \leq c\left[\ln(2+|z|)\right]^{\frac{1}{\mu-1}}, \qquad z \in\mathbb{R}^{2},

combined with the previous dilatation estimate, we deduce that the maximal dilatation quotient satisfies an asymptotic upper bound of the form DΦ(z)≤Cln⁡∣z∣D_{\Phi}(z) \leq C\ln|z| for sufficiently large ∣z∣|z|. Consequently, the integrated maximal dilatation K(r)K(r) defined in (3.1) grows at most logarithmically, more precisely it holds,

K(r)=∫02πDΦ(reiθ) dθ≤C0ln⁡ras r→∞,K(r)=\int_{0}^{2\pi}D_{\Phi}(re^{i\theta})\,d\theta\leq C_{0}\ln r \quad\text{as } r\to\infty,

where C0>0C_{0}>0 is a positive constant depending only on the data.

Using the distortion property under the global quasiconformal mapping WW, the conformal modulus of the image of the annulus A(1,R)={z∈C:1<∣z∣<R}A(1,R)=\{z\in\mathbb{C}:1<|z|<R\} satisfies the lower Lehto estimate (compare again [8], Corollary 20.9.2)

Mod⁡(W(A(1,R)))≥12π∫1R1rK(r) dr.(27)\operatorname{Mod}(W(A(1,R)))\geq\frac{1}{2\pi}\int_{1}^{R}\frac{1}{rK(r)}\,dr. \tag*{(27)}

Substituting the logarithmic upper bound for K(r)K(r) into (27) yields a double-logarithmic lower estimate for the modulus of the form

Mod⁡(W(A(1,R)))≥12π∫1R1C0rln⁡r dr=12πC0ln⁡(ln⁡R).(28)\begin{aligned} \operatorname{Mod}(W(A(1,R)))&\geq\frac{1}{2\pi}\int_{1}^{R}\frac{1}{C_{0}r\ln r}\,dr\\ &=\frac{1}{2\pi C_{0}}\ln(\ln R). \tag*{(28)} \end{aligned}

On the other hand, we can normalize W(0)=0W(0)=0 and choose a radius ρ>0\rho>0 such that B(0,ρ)‾⊂W(B(0,1))\overline{B(0,\rho)}\subset W(B(0,1)). We further abbreviate m(R)=min⁡∣z∣=R∣W(z)∣m(R)=\min_{|z|=R}|W(z)|. The Grötzsch ring estimate, applied after inversion, gives

Mod⁡(W(A(1,R)))≤12πln⁡4m(R)ρ.(29)\operatorname{Mod}(W(A(1,R)))\leq\frac{1}{2\pi}\ln\frac{4m(R)}{\rho}. \tag*{(29)}

For details the reader is referred to the monograph [14] of Gehring, Martin and Palka.

Combining the lower bound (28) with the upper bound (29) leads to

12πC0ln⁡(ln⁡R)≤12πln⁡(min⁡∣z∣=R∣W(z)∣),\frac{1}{2\pi C_{0}}\ln(\ln R)\leq\frac{1}{2\pi}\ln\left(\min_{|z|=R}|W(z)|\right),

where for notational simplicity we have just assumed that ρ=4\rho=4. Applying the exponential function to both sides, the following lower bound for ∣z∣=R|z|=R sufficiently large is established:

∣W(z)∣≥c1(ln⁡∣z∣)αfor ∣z∣≫1,(30)|W(z)|\geq c_{1}(\ln|z|)^{\alpha}\quad\text{for } |z|\gg1, \tag*{(30)}

where α=1C0>0\alpha=\frac{1}{C_{0}}>0 and c1>0c_{1}>0 are positive constants.

Inequality (30) immediately translates into an upper bound for the inverse mapping W−1W^{-1}, more precisely, it holds

∣W−1(z)∣≤exp⁡((∣z∣c1)1α)for ∣z∣≫1.(31)|W^{-1}(z)|\leq\exp\left(\left(\frac{|z|}{c_{1}}\right)^{\frac{1}{\alpha}}\right)\quad\text{for } |z|\gg1. \tag*{(31)}

Recalling the Stoilow factorization Φ(z)=Ψ(W(z))\Phi(z) = \Psi(W(z)), the entire holomorphic function Ψ:C→C\Psi: \mathbb{C} \to\mathbb{C} is given by Ψ=Φ∘W−1\Psi= \Phi\circ W^{-1}. Combining the gradient growth condition (10) with the upper bound (31), we obtain an estimate for the modulus of Ψ(z)\Psi(z):

∣Ψ(z)∣≤cln⁡[ln⁡∣W−1(z)∣]≤cln⁡([∣z∣c1]1α)=cαln⁡(∣z∣c1)≤c2ln⁡∣z∣(32)\begin{aligned} |\Psi(z)| &\le c\ln\left[\ln\left|W^{-1}(z)\right|\right] \le c\ln\left(\left[\frac{|z|}{c_1}\right]^{\frac{1}{\alpha}}\right) \\ &= \frac{c}{\alpha}\ln\left(\frac{|z|}{c_1}\right) \le c_2\ln|z| \tag*{(32)} \end{aligned}

being valid for sufficiently large values of zz. By a variant of Liouville’s theorem we deduce from (32) the constancy of Ψ\Psi and thereby of Φ\Phi, leading to the desired contradiction, which completes the proof of Theorem 1.8.

Proof of Theorem 1.10

Assuming that the conditions (2)–(5) hold for the energy density gg with a fixed exponent μ>1\mu> 1, we consider an entire solution u∈C2(R2)u \in C^2(\mathbb{R}^2) to the Euler–Lagrange equation (1) or its more general variant (8) under the μ\mu-ellipticity condition (9).

Instead of the Lehto growth condition (6) or the double-logarithmic bound (10) required in Theorem 1.8, we now impose the Landau oo-condition (11) for the gradient at infinity, i.e., we require

∣∇u(z)∣=o([ln⁡(2+∣z∣)]1μ−1)as ∣z∣→∞,(33)|\nabla u(z)| = o\left([\ln(2+|z|)]^{\frac{1}{\mu-1}}\right) \quad\text{as } |z| \to\infty, \tag*{(33)}

and claim that uu must be an affine function.

By contradiction, suppose that the complex gradient mapping Φ(z)=ux−iuy\Phi(z) = u_x - \mathrm{i}u_y is non-constant. As outlined in § 3 and 5 Φ\Phi is an open and discrete mapping, and using Stoilow’s factorization theorem, Φ\Phi decomposes into Φ=Ψ∘W\Phi= \Psi\circ W, where W:C→CW : \mathbb{C} \to\mathbb{C} is a global quasiconformal homeomorphism and Ψ:C→C\Psi: \mathbb{C} \to\mathbb{C} is an entire holomorphic function. We now repeat the calculations from § 5 having a more careful look at the constants:

Under the oo-condition (33), the local maximal dilatation satisfies DΦ(z)≤εCln⁡∣z∣D_\Phi(z) \le\varepsilon C\ln|z| for any choice of ε>0\varepsilon> 0 and sufficiently large ∣z∣|z|. Consequently, the integrated dilatation over concentric circles satisfies the asymptotic upper bound:

K(r)=∫02πDΦ(reiθ) dθ≤εC0ln⁡ras r→∞.(34)K(r) = \int_0^{2\pi} D_\Phi(re^{\mathrm{i}\theta})\,d\theta\le\varepsilon C_0\ln r \quad\text{as } r \to\infty. \tag*{(34)}

Inserting estimate (34) into the lower Lehto distortion inequality for the conformal modulus of the ring domain yields:

Mod⁡(W(A(1,R)))≥12πεC0ln⁡(ln⁡R).(35)\operatorname{Mod}(W(A(1,R))) \ge\frac{1}{2\pi\varepsilon C_0}\ln(\ln R). \tag*{(35)}

Combining (35) with the upper bound derived from the Grötzsch ring estimate (compare (29) and again choose ρ=4\rho= 4), we apply the exponential function to both sides. This implies a controlled asymptotic lower bound for the global homeomorphism W(z)W(z) for ∣z∣≫1|z| \gg1:

∣W(z)∣≥c1(ln⁡∣z∣)αεwith αε=1εC0.(36)|W(z)| \ge c_1(\ln|z|)^{\alpha_\varepsilon} \quad\text{with } \alpha_\varepsilon= \frac{1}{\varepsilon C_0}. \tag*{(36)}

Inequality (36) yields a sub-exponential growth rate for the inverse mapping W−1W^{-1} for ∣z∣≫1|z| \gg1, more precisely, it holds

∣W−1(z)∣≤exp⁡((∣z∣c1)1αε)=exp⁡((∣z∣c1)εC0).(37)|W^{-1}(z)| \le\exp\left(\left(\frac{|z|}{c_1}\right)^{\frac{1}{\alpha_\varepsilon}}\right) = \exp\left(\left(\frac{|z|}{c_1}\right)^{\varepsilon C_0}\right). \tag*{(37)}

We now estimate the modulus of the entire holomorphic function Ψ(z)=Φ(W−1(z))\Psi(z)=\Phi(W^{-1}(z)) by inserting the growth bound (37) into inequality (6):

∣Ψ(z)∣≤c[ln⁡(2+exp⁡((∣z∣c1)εC0))]1μ−1≤C3∣z∣εC0μ−1=C3∣z∣γ.(38)|\Psi(z)| \le c\left[\ln\left(2+\exp\left(\left(\frac{|z|}{c_1}\right)^{\varepsilon C_0}\right)\right)\right]^{\frac{1}{\mu-1}} \le C_3|z|^{\frac{\varepsilon C_0}{\mu-1}}=C_3|z|^\gamma. \tag*{(38)}

The resulting growth exponent is defined as γ:=εC0μ−1\gamma:=\frac{\varepsilon C_0}{\mu-1}. Since ε>0\varepsilon>0 can be chosen arbitrarily small from the very beginning, the requirement for a strictly sub-polynomial growth rate

γ<1\gamma<1

can evidently be fulfilled. By the extended variant of Liouville’s theorem for entire functions, the sub-polynomial growth stated in (38) shows that Ψ\Psi, and thereby Φ\Phi, is a constant vector. This provides the desired contradiction to our initial assumption, completing the proof that uu must be an affine function. □

Remark 6.1. We think that the proof of Theorem 1.10 can be adjusted to the case of densities ff satisfying an anisotropic (p,q)(p,q)-ellipticity condition with exponents 2≤p≤q2\le p\le q in place of (9). If p<qp<q, we then have to replace (11) by the condition

∣∇u(z)∣=o([ln⁡(2+∣z∣)]1q−p)for ∣z∣→∞|\nabla u(z)|=o\left([\ln(2+|z|)]^{\frac{1}{q-p}}\right)\quad\text{for }|z|\to\infty

implying that uu is an affine function, whereas in the isotropic setting q=pq=p this follows from the requirement ∣∇u(z)∣=O(∣z∣α)|\nabla u(z)|=O(|z|^\alpha) with the sharp bound for the exponent

α<1p−1.\alpha<\frac{1}{p-1}.

Acknowledgements

The author is very grateful to Kari Astala and Gaven Martin for their helpful comments.

Funding. Open Access funding enabled and organized by Projekt DEAL.

Data availability. No additional data are involved.

Declarations

Conflict of interest. There is no conflict of interest to be declared.

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

References

  1. [1]S. Bernstein, Sur un theoreme de geometrie et ses applications aux equations aux derivees partielles du type elliptique. Comm. de la Soc. Mat. de Kharkov (2eme ser. 15), 38–45 (1915–1917).
  2. [2]S. Bernstein, Über ein geometrisches Theorem und seine Anwendungen auf die partiellen Differentialgleichungen vom elliptischen Typus. Math. Z. 26, 551–558 (1927).DOI
  3. [3]V.M.Mikljukov, A new approach to the Bernstein theorem and to related questions of equations of minimal surface type. Mat. Sb. (N.S.) 108(150), no. 2, 268–289, 304 (1979).DOI
  4. [4]A.Farina, B.Sciunzi, E.Valdinoci, Bernstein and De Giorgi type problems: new results via a geometric approach. Ann. Sc. Norm. Super. Pisa Cl. Sci. (5) 7, no.4, 741–791 (2008).
  5. [5]M.Bildhauer, M.Fuchs, Bernstein’s theorem for variational integrals of linear growth and radial structure.(to appear)
  6. [6]W.Rudin, Real and complex analysis. McGraw-Hill (1970).
  7. [7]O.Lehto, K.I.Virtanen, Quasiconformal Mappings in the Plane. Grundlehren der mathematischen Wissenschaften, Band 126, Springer Verlag (1973).DOI
  8. [8]K.Astala, T.Iwaniec, G.Martin, Elliptic partial differential equations and quasiconformal mappings in the plane. Princeton Mathematical Series, 48. Princeton University Press, Princeton, NJ (2009).
  9. [9]M.Bildhauer, M.Fuchs, On a class of variational integrals with linear growth satisfying the condition of μ—ellipticity. Rendiconti di Matematica e delle sue applicazioni 22, 249-274 (2002).
  10. [10]J.Moser, On Harnack’s theorem for elliptic differential equations. Comm. Pure Appl. Math. 14, 577-591 (1961).DOI
  11. [11]J.Manfredi, E.Villamor, Mappings with integrable dilation in higher dimensions. Bulletin (New Series) of the American Mathematical Society Volume 32, Number 2 (1995).arxiv.org/abs/math/9504225
  12. [12]T.Iwaniec, V.Sverak, On mappings with integrable dilation. Proceedings of the American Mathematical Society , May, 1993, Vol. 118, No. 1(May, 1993), pp. 181-188.DOI
  13. [13]R.Luisto, P.Pankka, Stoilow’s theorem revisited. Expo.Math.38, 303-318(2020).DOI
  14. [14]F.Gehring, G.Martin, B.Palka, An introduction to the theory of higher-dimensional quasiconformal mappings. AMS Mathematical Surveys and Monographs Volume 216 (2017).DOI

Paper details

vlnc:W_u6mk283v
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