Abstract
H. Kneser (Jahresber Dt Math Vereinigung 35:123–124, 1926) showed by an ingenious method that plane harmonic mappings on the unit disc B, which attribute the circumference \(\partial B\) in a topological way to a convex curve \(\Gamma \), necessarily yield a diffeomorphism of B onto the interior G of the contour \(\Gamma \) and a homeomorphism between their closures. E. Heinz has generalized this method to solutions of nonlinear elliptic systems [see Chap. 13, Sect. 6 of Sauvigny (Partial differential equations. 1. Foundations and integral representations; 2. Functional analytic methods; with consideration of lectures by E. Heinz. Springer, London, 2012], however, this reasoning is restricted to the local situation and requires Lipschitz conditions for certain linear combinations of their coefficient functions. These Lewy-Heinz-systems comprise the equations for harmonic mappings with respect to a Riemannian metric and were utilized by Jost (J Reine Angew Math 342:141–153, 1981) to prove univalency for harmonic mappings between Riemannian surfaces. A global result is achieved by reconstruction of the solution for the Dirichlet problem, since this problem is uniquely determined by the uniqueness result of Jäger and Kaul (Manuscr Math 28:269–291, 1979). Here we shall adapt the original method of H. Kneser for harmonic mappings with respect to Riemannian metrics in order to receive harmonic diffeomorphisms from B onto stable Riemannian domains \(\Omega \). We construct a global nonlinear auxiliary function associated with an embedding into a field of geodesics. In the special case of planar harmonic mappings under semi-free boundary conditions, this procedure already appears in Proposition 3 of Hildebrandt and Sauvigny (J Reine Angew Math 422:69–89, 1991). By our present method to show univalency and to obtain a diffeomorphism between the domains, we can dispense of the uniqueness for the associate Dirichlet problem. The crucial idea consists of the notion stable Riemannian domains \(\Omega \), which possess a family of non-intersecting geodesic rays emanating from each boundary point and furnish a simple covering of the whole domain. Furthermore, we establish a convex hull property for harmonic mappings within \(\Omega \). On the basis of investigations by Hildebrandt et al. (Acta Math 138:1–16, 1977), we construct harmonic embeddings within the hemisphere by direct variational methods.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 The Dirichlet problem of harmonic mappings
For the coordinates \((x^1,x^2)\) we define the domain
and introduce the unit disc
with the parameters \(u+iv\cong (u,v)\). Now we prescribe the Riemannian metric
on the disc \(\Omega _1\). Here we require our coefficients to satisfy
with the Hölder constant \(\alpha \in (0,1)\) and the quantity \(\lambda \in (0,1]\).
By a continuity method the following profound result is established:
Theorem 1
(Conformal mappings w. r. t. Riemannian metrics) For the Riemannian metric (1.1), (1.2), (1.3) there exists a \(C^{2+\alpha }({\overline{B}},\overline{\Omega }_1)\)-diffeomorphic, positive-oriented mapping
satisfying the weighted conformality relations
Proof
See our uniformization theorem from [12] Chap. 12 the Theorem 8.2. \(\square \)
Due to Proposition 7.1 of [12] Chap. 12, the function X then satisfies the nonlinear elliptic system
Here we use the Christoffel symbols
with the inverse matrix \((g^{jk})_{j,k=1,2}:=(g_{jk})^{-1}_{j,k=1,2}\). Therefore, X represents a one-to-one harmonic mapping of the disc \(\{B,(\delta _{jk})\}\) with the Euclidean metric \((\delta _{jk})_{j,k=1,2}\) onto the disc \(\{\Omega _1,(g_{jk})\}\). On account of well-known regularity results, the associate boundary function
appearing within this approximation and selection procedure, yields a positive-oriented \(C^{2+\alpha }(\partial B,\partial \Omega _1)\)-diffeomorphism between the circumferences \(\partial B\) and \(\partial \Omega _1\). This weighted-conformal mapping is uniquely determined by a three-point-condition on the boundary. Of course, this boundary representation optimally appears for these weighted-conformal mappings and cannot be prescribed!
Remark 1
Starting with an analogous result to Theorem 1 above, Jost [7] has constructed harmonic diffeomorphisms, for arbitrary convex boundary data, by deformation of the boundary values via a topological method. This has been combined with a priori estimates for their Jacobian by E. Heinz. With the aid of the maximum principle by Jäger and Kaul [6], then Jost obtained the diffeomorphic character of harmonic maps by reconstruction.
In Sect. 4 we shall see directly the one-to-one character of our harmonic maps, established in Theorem 2 below, and may dispense of the uniqueness for the associate Dirichlet problem. Here we prescribe the Riemannian metric (1.1) on the whole plane \(\mathbb R^2\), which is Euclidean outside of the disc
of a fixed radius \(0<M<+\infty \). More precisely, we assume that our coefficients satisfy the following conditions with the Hölder constant \(\alpha \in (0,1)\) and a positive number \(\lambda \in (0,1]\) as follows:
and
Furthermore, we require that the metric \(ds^2\) possesses a moderate deviation in the disc \(\Omega _M\) from the Euclidean metric with the constant \(\displaystyle a\in (0,\frac{1}{2M})\) in the following sense: The associate Christoffel symbols (1.6) satisfy the estimate
By the Leray–Schauder degree of mapping we can establish the following
Theorem 2
(Dirichlet problem for moderate harmonic mappings) Let the Riemannian metric (1.1), (1.8), (1.9) be given with a moderate deviation (1.10) by the constant \(\displaystyle a\in (0,\frac{1}{2M})\) from the Euclidean metric. For each boundary function
there exists a solution
for the system (1.5), (1.6) of harmonic mappings under the boundary condition
Proof
From Theorem 4.4 of [12] Chap. 12 we deduce the existence of a harmonic mapping (1.11) under the boundary conditions (1.12). \(\square \)
Remark 2
Due to the geometric maximum principle by E. Heinz (see Theorem 1.4 in [12] Chap. 12), the solution X of Theorem 2 is subject to the inequality
When the boundary values satisfy \(|\Phi (u,v)|<M,\,\forall (u,v)\in \partial B\), then the estimate
follows, and \(X:{\overline{B}}\rightarrow \Omega _M\) represents an inner solution of the system (1.5), (1.6), briefly an inner harmonic mapping.
2 Geodesically stable Riemannian domains
We begin our considerations with the central
Definition 1
We call the disc \(\Omega _M\) of radius \(0<M<+\infty \) endowed with a Riemannian metric (1.1), (1.8), (1.9) a geodesically stable Riemannian domain or simply a stable Riemannian domain, if each geodesic—in unit velocity—emanating from an arbitrary boundary point \(X_0\in \partial \Omega _M\) into an interior direction \(\xi \in S^1\)—within the disc \(\Omega _M\) -
of the lenght \(\tau (X_0,\xi )>0\) does not contain conjugate points.
Remark 3
Let the Riemannian metric (1.1), (1.8), (1.9) of the regularity class \(C^3\) be given, such that their Gaussian curvature K satisfies
with the barrier \(\kappa \in [0,+\infty )\). Furthermore, let the diameter of the Riemannian domain be bounded by the constant \(\displaystyle \frac{\pi }{\sqrt{\kappa }}\in (0,+\infty ]\) as follows:
Then this Riemannian domain is necessarily stable.
In this context, we refer our readers to the comparison theorem of J. C. F. Sturm in Satz 3 of Kapitel VII, § 7 from our treatise Analysis [13].
Of central importance is the subsequent
Lemma 1
(Geodesic central fields) Let the domain \(\Omega _M\) be endowed with a stable Riemannian metric from Definition 1, and a boundary point \(X_0\in \partial \Omega _M\) be chosen arbitrarily. Then the family of geodesics
with their initial position
and their initial velocity
yields a simple covering of the pointed disc \(\overline{\Omega _M}{\setminus }\{X_0\}\).
Proof
For arbitrary \(\displaystyle -\frac{\pi }{2}<s_1<s_2<+\frac{\pi }{2}\) we consider the geodesic
the circular arc
and the geodesic
Since our geodesics do not contain conjugate points within \(\overline{\Omega _M}\), the arcs (2.7) and (2.8) and (2.9) consecutively constitute a Jordan contour \(\Gamma (s_1,s_2;X_0)\) for parameters \(s_1<s_2\) chosen sufficiently near. They form a Jordan curve \(\Gamma (s_1,s_2;X_0)\) for arbitrary parameters \(\displaystyle -\frac{\pi }{2}<s_1<s_2<+\frac{\pi }{2}\) as well, since the mapping
is strictly monotonic. Therefore, the interior of the Jordan curves \(\Gamma (s_1,s_2;X_0)\) exhausts the domain \(\Omega _M\) for \(\displaystyle s_1\rightarrow -\frac{\pi }{2}+\) and \(\displaystyle s_2\rightarrow +\frac{\pi }{2}-\). These contours \(\Gamma (s_1,s_2;X_0)\) cover \(\partial \Omega _M\) in the limit \(\displaystyle s_1= -\frac{\pi }{2}\) and \(\displaystyle s_2= +\frac{\pi }{2}\), where the singularity \(X_0\) remains fixed. \(\square \)
As in [13] Kap. VII, Sect. 5 we introduce
Definition 2
The Riemannian inner product of the planar vector fields \(Y(t)=(y_1(t),y_2(t))\) and \(Z(t)=(z_1(t),z_2(t))\) along the plane curve \(X(t)=(x_1(t),x_2(t))\) with the parameter \(a<t<b\) is determined as follows:
Remark 4
From the Gauß–Riemann-Lemma (see Satz 2 in [13] Kap. VII, Sect. 4), we realize the following identities for our geodesic central field in Lemma 1 above:
Definition 3
Let the stable Riemannian domain \(\Omega _M\) of Definition 1 with the geodesic central fields (2.4) of Lemma 1 be given. For all points \(X\in \Omega _M\) with their unique representation
we define the lifted vector fields
and the lifted Gaussian fundamental coefficient
We obtain with
the Gaussian geodesic frame, which constitutes an orthonormal, positive-oriented system of vectors—with respect to the Riemannian inner product (2.11)—on account of Remark 4 above.
In stable Riemannian domains, we can conveniently characterize the convex hull of arbitrary compact sets \(F\subset \Omega _M\) with the subsequent
Definition 4
Let the stable Riemannian domain \(\Omega _M\) of Definition 1 with the geodesic central fields (2.4) of Lemma 1 be given. For all \(\displaystyle \,-\frac{\pi }{2}<s_0<+\frac{\pi }{2}\,\) we introduce the geodesic region
This region is closed by the geodesic arc
and furthermore bounded by the circular arc
where these curves constitute with \(\partial \Theta (s_0;X_0)\) a positive-oriented Jordan contour. For an arbitrary compact set \(F\subset \Omega _M\), we define the convex hull \({{\mathcal {H}}}(F)\) of F within the stable Riemannian domain \(\Omega _M\) as follows:
Finally, we introduce the important geodesic function with
Definition 5
Let the stable Riemannian domain \(\Omega _M\) of Definition 1 with the geodesic central fields (2.4) of Lemma 1 be given. For all points \(X\in \overline{\Omega _M}{\setminus } \{X_0\}\) with their unique representation
we define the geodesic function
Remark 5
Obviously, the equation \(\,\Psi (X;X_0)=s_0,\,X\in \Omega _M\,\) describes the geodesic
for all \(\displaystyle -\frac{\pi }{2}<s_0<+\frac{\pi }{2}\), and the following characterization is valid:
3 Pseudoharmonic nonlinear combination for harmonic mappings
We refer to the covariant differentiation \(\displaystyle \,\frac{\nabla }{dt}\,\) from [13] Kap. VII, \(\S \) 5 and begin with
Definition 6
Let us consider the Riemannian metric (1.1), (1.8), (1.9) with its inverse tensor
Now we define the cogradient of the function \(\Psi (\,.,X_0)\) from Definition 5 as follows
where the boundary point \(\,X_0\in \partial \Omega _M\,\) is arbitrary.
Remark 6
Differentiation of the identity (3.1) yields
When an arbitrary mapping
is given, we consider the associate auxiliary function
We immediately comprehend the following covariant chain rule
With the aid of the covariant product rule (see Satz 2 in [13] Kap. VII, § 5) we calculate
Here \(\displaystyle \,\frac{\nabla }{du}\,\) denotes the covariant derivative of the respective vector field along the curve X(., v) due to Definition 1 in [13] Kap. VII, § 5. In order to evaluate the first bracket term in (3.7), we determine the partial derivative
Now we utilize the Christoffel symbols of the first kind (see (1.6) and the formula (10) in [13] Kap. VII, § 3):
With the aid of the identities (3.8) and (3.9) and the Remark 6, we determine the first bracket term in (3.7):
Here we use the modified Christoffel symbols of the first kind
Definition 7
We define the covariant Hessian bilinear form
The combination of (3.7) and (3.10) – (3.12) yields the identity
Analogously, we derive the identity
With the aid of Lemma 1 we shall see that the bilinear form in Definition 7 vanishes at each point on an appropriate one-dimensional space. More precisely, we have the
Lemma 2
(Covariant derivatives of the geodesic function)
For the geodesic function \(\Psi \) in Definition 5 the cogradient satisfies the equations
Moreover, the identity
holds true. Finally, the covariant Hessian form (3.12) of the second derivatives vanishes as follows:
Here the boundary point \(\,X_0\in \partial \Omega _M\,\) is chosen arbitrarily.
Proof
-
1.
We consider the auxiliary function
$$\begin{aligned} \psi (t,s):=\Psi (\mathbf Y(t,s;X_0);X_0)=s,\quad 0< t<\tau (X_0,s),\,-\frac{\pi }{2}<s<+\frac{\pi }{2}. \end{aligned}$$(3.18)With the covariant chain rule (3.6) we determine the derivatives
$$\begin{aligned} \displaystyle 0= & {} \psi _t(t,s)=\Big [{\varvec{\nabla }}\Psi (\mathbf Y(t,s;X_0);X_0),\,\mathbf Y_t(t,s;X_0)\Big ]_{\mathbf Y(t,s;X_0)},\nonumber \\ \displaystyle 1= & {} \psi _s(t,s)=\Big [{\varvec{\nabla }}\Psi (\mathbf Y(t,s;X_0);X_0),\,\mathbf Y_s(t,s;X_0)\Big ]_{\mathbf Y(t,s;X_0)},\nonumber \\&\displaystyle \text{ for } \text{ all }\quad 0< t<\tau (X_0,s),\quad -\frac{\pi }{2}<s<+\frac{\pi }{2}, \end{aligned}$$(3.19)which yields the Eq. (3.15). With the aid of the Gaussian geodesic frame (2.16), we deduce the identity (3.16) from the Eq. (3.15).
-
2.
Since the curve \(\mathbf Y(.,s;X_0)\) represents a geodesic, we have the identity
$$\begin{aligned} \frac{\nabla \,\mathbf Y_t(t,s;X_0)}{dt}=0,\quad 0< t<\tau (X_0,s),\quad -\frac{\pi }{2}<s<+\frac{\pi }{2}. \end{aligned}$$(3.20)Now the Eq. (3.13) yields
$$\begin{aligned} \displaystyle 0= & {} \psi _{tt}(t,s)=\Big [\mathbf Y_t(t,s;X_0),{\varvec{\nabla }}^2\,\Psi (\mathbf Y(t,s;X_0);X_0),\,\mathbf Y_t(t,s;X_0)\Big ]_{\mathbf Y(t,s;X_0)}\nonumber \\&\displaystyle +\Big [{\varvec{\nabla }}\Psi (\mathbf Y(t,s;X_0);X_0),\,\frac{\nabla \,\mathbf Y_t(t,s;X_0)}{dt}\Big ]_{\mathbf Y(t,s;X_0)}\nonumber \\ \displaystyle= & {} \Big [\mathbf Y_t(t,s;X_0)\,,{\varvec{\nabla }}^2\,\Psi (\mathbf Y(t,s;X_0);X_0),\,\mathbf Y_t(t,s;X_0)\Big ]_{\mathbf Y(t,s;X_0)} \nonumber \\&\displaystyle \text{ for } \text{ all }\quad 0< t<\tau (X_0,s)\,,\quad -\frac{\pi }{2}<s<+\frac{\pi }{2}\,, \end{aligned}$$(3.21)which implies the statement (3.17). \(\square \)
Now we present the principal device of our investigations within
Lemma 3
(Pseudoharmonic nonlinear combination for harmonic maps)
Let the mapping X(u, v) from (3.4) be harmonic, i.e. the Eqs. (1.5), (1.6) hold true. Then the geodesic auxiliary function \(\psi (u,v)\) in (3.5) satisfies the elliptic partial differential equation
with the continuous functions \(a=a(u,v):B\rightarrow \mathbb R\) and \(b=b(u,v):B\rightarrow \mathbb R\). The gradient \(\nabla \psi \) possesses only isolated zeroes in B and allows expansions of Hartman–Wintner-type (see Theorem 1.2 in [12] Chap. 9) there. Since this function \(\psi \) shares important properties with harmonic functions, we may address \(\psi \) as being pseudoharmonic.
Proof
-
1.
The mapping X is harmonic, and we have the identity
$$\begin{aligned} \frac{\nabla \,X_u(u,v)}{du}+ \frac{\nabla \,X_v(u,v)}{dv}=0\,,\quad (u,v)\in B. \end{aligned}$$(3.23)Now we add the Eqs. (3.13) and (3.14), and we obtain the following identity for our auxiliary function \(\psi (u,v),\,(u,v)\in B\) on account of (3.23):
$$\begin{aligned} \displaystyle \Delta \psi (u,v)= & {} \psi _{uu}(u,v)+\psi _{vv}(u,v)\nonumber \\ \displaystyle= & {} \Big [X_u(u,v)\,,{\varvec{\nabla }}^2\,\Psi (X(u,v);X_0),\,X_u(u,v)\Big ]_{X(u,v)}\nonumber \\&\displaystyle +\Big [X_v(u,v)\,,{\varvec{\nabla }}^2\,\Psi (X(u,v);X_0),\,X_v(u,v)\Big ]_{X(u,v)}\nonumber \\&\displaystyle +\Big [{\varvec{\nabla }}\Psi (X(u,v);X_0),\,\frac{\nabla \,X_u(u,v)}{du}+\frac{\nabla \,X_v(u,v)}{dv} \Big ]_{X(u,v)}\nonumber \\&\displaystyle =\Big [X_u(u,v)\,,{\varvec{\nabla }}^2\,\Psi (X(u,v);X_0),\,X_u(u,v)\Big ]_{X(u,v)}\nonumber \\&\displaystyle +\Big [X_v(u,v)\,,{\varvec{\nabla }}^2\,\Psi (X(u,v);X_0),\,X_v(u,v)\Big ]_{X(u,v)}\,,\quad (u,v)\in B.\nonumber \\ \end{aligned}$$(3.24) -
2.
With the aid of the Gaussian geodesic frame (2.16) and the identity (3.16), we expand the vector \(X_u(u,v)\) via the covariant chain rule (3.6) as follows:
$$\begin{aligned} \displaystyle X_u(u,v)= & {} \Big [X_u(u,v),\widehat{\mathbf Y_t}(X(u,v);X_0)\Big ]_{X(u,v)}\widehat{\mathbf Y_t}(X(u,v);X_0)\nonumber \\&\displaystyle +\Bigg [X_u(u,v),\frac{\widehat{\mathbf Y_s}(X(u,v);X_0)}{\sqrt{{\widehat{G}}(X(u,v);X_0)}}\Bigg ]_{X(u,v)}\,\frac{\widehat{\mathbf Y_s}(X(u,v);X_0)}{\sqrt{{\widehat{G}}(X(u,v);X_0)}}\nonumber \\ \displaystyle= & {} \Big [X_u(u,v),\widehat{\mathbf Y_t}(X(u,v);X_0)\Big ]_{X(u,v)}\widehat{\mathbf Y_t}(X(u,v);X_0)\nonumber \\&\displaystyle +\Bigg [X_u(u,v),\frac{\widehat{\mathbf Y_s}(X(u,v);X_0)}{{\widehat{G}}(X(u,v);X_0)}\Bigg ]_{X(u,v)}\,\widehat{\mathbf Y_s}(X(u,v);X_0)\nonumber \\ \displaystyle= & {} \Big [X_u(u,v),\widehat{\mathbf Y_t}(X(u,v);X_0)\Big ]_{X(u,v)}\widehat{\mathbf Y_t}(X(u,v);X_0)\nonumber \\ \displaystyle&+\Big [X_u(u,v),{\varvec{\nabla }}\Psi (X(u,v);X_0)\Big ]_{X(u,v)}\,\widehat{\mathbf Y_s}(X(u,v);X_0)\nonumber \\ \displaystyle= & {} \Big [X_u(u,v),\widehat{\mathbf Y_t}(X(u,v);X_0)\Big ]_{X(u,v)}\widehat{\mathbf Y_t}(X(u,v);X_0)\nonumber \\&\displaystyle +\psi _u(u,v)\,\widehat{\mathbf Y_s}(X(u,v);X_0)\,,\quad (u,v)\in B. \end{aligned}$$(3.25)Proceeding in the same way for the derivative with respect to v, we arrive at the following equations:
$$\begin{aligned} \displaystyle X_u(u,v)= & {} \Big [X_u(u,v),\widehat{\mathbf Y_t}(X(u,v);X_0)\Big ]_{X(u,v)}\widehat{\mathbf Y_t}(X(u,v);X_0)\nonumber \\&\displaystyle +\psi _u(u,v)\,\widehat{\mathbf Y_s}(X(u,v);X_0)\,,\quad (u,v)\in B\,;\nonumber \\ \displaystyle X_v(u,v)= & {} \Big [X_v(u,v),\widehat{\mathbf Y_t}(X(u,v);X_0)\Big ]_{X(u,v)}\widehat{\mathbf Y_t}(X(u,v);X_0)\nonumber \\&\displaystyle +\psi _v(u,v)\,\widehat{\mathbf Y_s}(X(u,v);X_0)\,,\quad (u,v)\in B. \end{aligned}$$(3.26) -
3.
When we insert the vectors \(X_u(u,v)\) and \(X_v(u,v)\) from (3.26) into the covariant Hessian forms within (3.24) and observe the property (3.17), we receive the representation (3.22) with continuous coefficient functions. \(\square \)
Remark 7
Similar arguments for the Euclidean situation under semi-free boundary conditions have been established in [5] Proposition 3 within my joint investigation with Hildebrandt.
4 Convex hull property, univalency and transversality for harmonic mappings
We start with the central definition and assume that setting throughout this section.
Definition 8
Let \(ds^2\) denote a stable Riemannian metric (1.1), (1.8), (1.9) on the disc \(\Omega _M\) of radius \(0<M<+\infty \) with a moderate deviation (1.10) by the constant \(\displaystyle a\in (0,\frac{1}{2M})\) from the Euclidean metric. For each continuous boundary function
we call the function
a solution of the Dirichlet problem \(\,{\mathcal {P}} (\Omega _M,ds^2;\Phi )\,\), when the function X satisfies the system (1.5), (1.6) of harmonic mappings and fulfills the boundary condition
Theorem 3
(Convex hull property for harmonic mappings)
Let the continuous function \(\Phi :\partial B \rightarrow \Omega _M\in C^0(\partial B)\) with the boundary point set \(\displaystyle F:=\Phi (\partial B)\subset \Omega _M\) and its convex hull \(\displaystyle {{\mathcal {H}}}(F)\subset \Omega _M\) due to Definition 4 be given. For each solution
of the Dirichlet problem we have the following inclusion:
Proof
-
1.
The boundary point set \(F:=\Phi (\partial B)\subset \Omega _M\) is compact in \(\Omega _M\), and the convex hull of the boundary values \({{\mathcal {H}}}(F)\subset \Omega _M\) as well. Therefore, we can find a unique number \(\displaystyle \,\sigma (X_0,F)\in (-\frac{\pi }{2},+\frac{\pi }{2})\,\), such that
$$\begin{aligned}&\displaystyle \bigcap \Big \{\Theta (s_0;X_0)\Big |\,s_0\in (-\frac{\pi }{2},+\frac{\pi }{2}):\,F\subset \Theta (s_0;X_0)\Big \} =\Theta (\sigma (X_0,F);X_0)\nonumber \\&\quad =\Big \{X\in \Omega _M\Big |\,\Psi (X;X_0)\le \sigma (X_0,F)\Big \}\quad \text{ for } \text{ each } \text{ point }\quad X_0\in \partial \Omega _M. \end{aligned}$$(4.4)Here we have utilized the characterization (2.23) for the last identity. Now we determine the convex hull of the boundary point set as follows:
$$\begin{aligned}&\displaystyle {\mathcal {H}}(F)=\bigcap \Big \{\Theta (s_0;X_0)\Big |\,X_0\in \partial \Omega _M,\,s_0\in (-\frac{\pi }{2},+\frac{\pi }{2}):\,F\subset \Theta (s_0;X_0)\Big \}\nonumber \\&\quad \displaystyle =\bigcap _{X_0\in \partial \Omega _M} \Theta (\sigma (X_0,F);X_0)= \bigcap _{X_0\in \partial \Omega _M}\Big \{X\in \Omega _M\Big |\,\Psi (X;X_0)\le \sigma (X_0,F)\Big \}.\nonumber \\ \end{aligned}$$(4.5) -
2.
With the aid of the geometric maximum principle by E. Heinz, we can see as in Remark 2 that the inclusion \(X(\partial B)\subset \Omega _M\) implies the property \(X(B)\subset \Omega _M\). For arbitrary points \(X_0\in \partial \Omega _M\) we consider the geodesic auxiliary function
$$\begin{aligned} \psi (u,v):=\Psi (X(u,v);X_0)\,,\quad (u,v)\in {\overline{B}}. \end{aligned}$$(4.6)Since the inclusion \(F\subset \Theta (\sigma (X_0,F);X_0)\) for all \(X_0\in \partial \Omega _M\) holds true, we receive
$$\begin{aligned} \psi (u,v)\le \sigma (X_0,F)\quad \text{ for } \text{ all }\quad (u,v)\in \partial B. \end{aligned}$$(4.7)Now Lemma 3 implies that the function \(\psi \) is subject to the maximum principle, which gives us the following statement:
$$\begin{aligned} \Psi (X(u,v);X_0)\le \sigma (X_0,F)\,,\quad (u,v)\in {\overline{B}}\quad \text{ for } \text{ all } \text{ points }\quad X_0\in \partial \Omega _M. \end{aligned}$$(4.8)On account of (4.5), we obtain that \(\displaystyle \,X({\overline{B}})\subset {\mathcal {H}}(F)\,\) holds true. \(\square \)
Definition 9
A Jordan contour \(\Gamma \subset \Omega _M\) is called convex in \(\Omega _M\), when the following properties are fulfilled:
-
(i)
The Jordan contour \(\Gamma \) coincides with the boundary \(\partial {\mathcal {H}}(\Gamma )\) of its convex hull, and the interior \(I(\Gamma )\) of the contour \(\Gamma \) corresponds to the open kernel of the convex hull \({\mathcal {H}}(\Gamma )\).
-
(ii)
A geodesic \(\mathbf Y(t,s_0;X_0),\,0\le t\le \tau (X_0,s_0)\) for the parameter \(\displaystyle s_0\in (-\frac{\pi }{2},+\frac{\pi }{2})\), such that \(\mathbf Y(.,s_0;X_0)\) meets the interior \(I(\Gamma )\) at an inner point \(Y_0\in I(\Gamma )\), shall decompose the Jordan curve into the closed Jordan arcs
$$\begin{aligned} \Gamma ^-(X_0,s_0):=\Gamma \cap \Theta (s_0,X_0)\quad \text{ and }\quad \Gamma ^+(X_0,s_0):=\overline{\Gamma {\setminus }\Theta (s_0,X_0)}. \end{aligned}$$These arcs meet at their end points on the geodesic \(\mathbf Y(.,s_0;X_0)\) above.
With the original method by Kneser [9] for the Euclidean plane, which we adapt to the Riemannian situation here, we shall establish the subsequent
Theorem 4
(Univalency for harmonic mappings)
Let the convex Jordan contour \(\Gamma \subset \Omega _M\) and the topological boundary function \(\Phi :\partial B\rightarrow \Gamma \in C^0(\partial B,\mathbb R^2)\) be given. Then each solution
of the Dirichlet problem furnishes a topological mapping of \({\overline{B}}\) onto \(\overline{I(\Gamma )}\) and a \(C^2\)-diffeomorphism of B onto \(I(\Gamma )\).
Proof
-
1.
From Theorem 3 and Definition 9, (i) we infer the inclusion
$$\begin{aligned} X({\overline{B}})\subset {\mathcal {H}}(\Gamma )=I(\Gamma )\cup \Gamma . \end{aligned}$$(4.9)Moreover, the strict inclusion
$$\begin{aligned} X(B)\subset I(\Gamma ) \end{aligned}$$(4.10)is valid, which we deduce as follows: If the statement (4.10) were violated, there exists a point \((u_0,v_0)\in B\) with \(Y_0=X(u_0,v_0)\in \Gamma \). On account of Definition 9 we can find a point \(X_0\in \partial \Omega _M\) and a value \(\displaystyle \,-\frac{\pi }{2}<s_0<+\frac{\pi }{2}\,\), such that
$$\begin{aligned} \Gamma \subset \Theta (s_0;X_0)\quad \text{ and }\quad Y_0\in \Gamma \cap \partial \Theta (s_0;X_0) \end{aligned}$$(4.11)holds true. Now we consider the auxiliary function
$$\begin{aligned} \psi (u,v):=\Psi (X(u,v);X_0)\le s_0,\,(u,v)\in {\overline{B}}\,\text{ with }\,\psi (u_0,v_0)=\Psi (Y_0;X_0)=s_0.\nonumber \\ \end{aligned}$$(4.12)From Lemma 3 we see that \(\psi \) is a pseudoharmonic function and cannot attain a strict maximum within B. Consequently, the equality
$$\begin{aligned} \psi (u,v)=s_0\quad \text{ for } \text{ all }\quad (u,v)\in {\overline{B}} \end{aligned}$$(4.13)holds true, which yields an evident contradiction. Therefore, the strict inclusion (4.10) is valid.
-
2.
Now we show indirectly that the Jacobian of the mapping X does not vanish:
$$\begin{aligned} J_X(u,v):=\frac{\partial (x^1(u,v),x^2(u,v))}{\partial (\,u\,,\,v\,)} =\left| \begin{array}{c} x^1_u(u,v),x^1_v(u,v)\\ x^2_u(u,v),x^2_v(u,v)\end{array}\right| \ne 0\,,\forall \,(u,v)\in B.\nonumber \\ \end{aligned}$$(4.14)If the statement (4.14) were violated, there exists a point
$$\begin{aligned} (u_0,v_0)\in B\quad \text{ with }\quad Y_0:=X(u_0,v_0)\in I(\Gamma )\,, \end{aligned}$$such that the vectors \(\{X_u(u_0,v_0),\,X_v(u_0,v_0)\}\) are linearly dependent. Consequently, we find a unit vector \(Z_0\) orthogonal to these vectors as follows:
$$\begin{aligned}&\displaystyle Z_0\in \mathbb R^2{\setminus }\{(0,0)\}\quad \text{ with }\quad \Big [Z_0\,,Z_0\Big ]_{X(u_0,v_0)}=1\,,\nonumber \\&\quad \displaystyle \Big [Z_0\,,X_u(u_0,v_0)\Big ]_{X(u_0,v_0)}=0=\Big [Z_0\,,X_v(u_0,v_0)\Big ]_{X(u_0,v_0)}. \end{aligned}$$(4.15) -
3.
Now the Gaussian geodesic frame at the fixed point \(Y_0\in \Omega _M\)
$$\begin{aligned} \Bigg \{ \widehat{\mathbf Y_t}(Y_0;X_0)\,,\,\frac{\widehat{\mathbf Y_s}(Y_0;X_0)}{\sqrt{{\widehat{G}}(Y_0;X_0)}}\Bigg \}\,,\quad X_0\in \partial \Omega _M \end{aligned}$$(4.16)performs one positive-oriented and continuous rotation, when \(X_0\) traverses the circumference \(\partial \Omega _M\) once in positive orientation. This results from the construction of the geodesic vector fields, which depend continuously on the point \(X_0\in \partial \Omega _M\) together with their nonvanishing derivatives. Therefore, we can choose a point \(X_0\in \partial \Omega _M\) such that
$$\begin{aligned} \frac{\widehat{\mathbf Y_s}(Y_0;X_0)}{\sqrt{{\widehat{G}}(Y_0;X_0)}}=Z_0 \end{aligned}$$(4.17)holds true. With the aid of (3.16) and (4.17), we obtain the following representation for the cogradient of the geodesic function \(\Psi \)
$$\begin{aligned} {\varvec{\nabla }}\Psi (Y_0;X_0)=\frac{\widehat{\mathbf Y_s}(Y_0;X_0)}{{\widehat{G}}(Y_0;X_0)}=\lambda \,Z_0\quad \text{ with }\quad \lambda :=\frac{1}{\sqrt{{\widehat{G}}(Y_0;X_0)}}. \end{aligned}$$(4.18) -
4.
Let us now consider the geodesic auxiliary function
$$\begin{aligned} \psi (u,v):=\Psi (X(u,v);X_0)\,,\quad (u,v)\in {\overline{B}}. \end{aligned}$$(4.19)With the aid of (4.15) and (4.18) we derive
$$\begin{aligned} \displaystyle \psi _u(u_0,v_0)= & {} \Big [{\varvec{\nabla }}\Psi (Y_0;X_0),\,X_u(u_0,v_0)\Big ]_{{X(u_0,v_0)}}\nonumber \\ \displaystyle= & {} \lambda \,\Big [Z_0,X_u(u_0,v_0)\Big ]_{{X(u_0,v_0)}}=0\,;\nonumber \\ \displaystyle \psi _v(u_0,v_0)= & {} \Big [{\varvec{\nabla }}\Psi (Y_0;X_0),\,X_v(u_0,v_0)\Big ]_{{X(u_0,v_0)}}\nonumber \\ \displaystyle= & {} \lambda \,\Big [Z_0,X_v(u_0,v_0)\Big ]_{{X(u_0,v_0)}}=0. \end{aligned}$$(4.20)Since the function \(\psi \) is pseudoharmonic due to Lemma 3 and \(\,\nabla \psi (u_0,v_0)=(0,0)\,\) holds true, now \(\psi \) represents a saddle point near \((u_0,v_0)\). This behavior propagates to the boundary \(\partial B\) on account of the maximum/minimum principle. This yields a contradiction to the behavior of the function \(\psi :\partial B\rightarrow \mathbb R\) on the boundary, which only possesses two points for the level \(s_0\) due to Definition 9, (ii) Consequently, the Jacobian \(J_X\) is not allowed to vanish within B, and the statement (4.14) holds true. For an exact proof, we can follow the arguments for harmonic functions in Lemma 2 and Lemma 3 of our book on Minimal Surfaces [2] within Section 4.9. These arguments remain valid for the pseudoharmonic function \(\psi \), due to the asymptotic expansions of P. Hartman and A. Wintner (see Theorem 1.2 in [12] Chap. 9.) at their critical points.
-
5.
With the monodromy principle (see Lemma 1 in [2], Sect. 4.9) we can infer the topological character of the mapping
$$\begin{aligned} X:{\overline{B}}\rightarrow \overline{I(\Gamma )}\subset \Omega _M \end{aligned}$$from (4.14) and the property that the boundary representation \(X:\partial B\rightarrow \Gamma \) is topological. Alternatively, we can use an index-argument from [11] Hilfssatz 7 in order to show that the mapping \(\,X:{\overline{B}}\rightarrow \overline{I(\Gamma )}\,\) is one-to-one. \(\square \)
Remark 8
In the Euclidean situation, we find this result by T. Radó and H. Kneser in § 398 of J. C. C. Nitsche’s monograph [10] Vorlesungen über Minimalflächen.
Furthermore, we refer to Proposition 4.2 in my joint treatise [4] with S. Hildebrandt.
The following statement contains the transversality of harmonic mappings to the boundary. More precisely, we shall establish
Theorem 5
(Existence of \(C^{2+\alpha }({\overline{B}},\overline{\Omega _M})\)-diffeomorphisms for \({{\mathcal {P}}}(\Omega _M,ds^2;\Phi )\)) Let the \(C^{2+\alpha }(\partial B,\partial \Omega _M)\)-diffeomorphic boundary function \(\Phi :\partial B \rightarrow \partial \Omega _M\) be given. Then there exists a \(C^{2+\alpha }({\overline{B}},\overline{\Omega _M})\)-diffeomorphism
which furnishes a solution of the Dirichlet Problem \({{\mathcal {P}}}(\Omega _M,ds^2;\Phi )\).
Proof
-
1.
We build upon our exsistence result in Theorem 2, and we receive a solution \(X=X(u,v)\in C^{2+\alpha }({\overline{B}},\overline{\Omega _M})\) for the Dirichlet problem \({{\mathcal {P}}}(\Omega _M,ds^2;\Phi )\). By the geometric maximum principle of E. Heinz the function
$$\begin{aligned} \chi (u,v):=|X(u,v)|^2\,,\quad (u,v)\in {\overline{B}}\quad \text{ satisfies }\quad \Delta \chi (u,v)\ge 0\,,\quad (u,v)\in B. \end{aligned}$$The boundary point lemma of E. Hopf implies the following inequality for the derivative w. r. t. the exterior normal \(\nu \) to B:
$$\begin{aligned} 0<\frac{d}{d\nu }\chi (u_1,v_1)=2\,X(u_1,v_1) \cdot \frac{d}{d\nu } X(u_1,v_1)\quad \text{ for } \text{ all } \text{ points }\quad (u_1,v_1)\in \partial B.\nonumber \\ \end{aligned}$$(4.21)This property (4.21) together with the arguments in [11] Satz 2 yield that our mapping X is transversal in the following sense:
$$\begin{aligned} J_X(u,v)\ne 0\quad \text{ for } \text{ all }\quad (u,v)\in \partial B. \end{aligned}$$(4.22) -
2.
Now we follow the parts (2)–(4) in the proof of Theorem 4, in order to exclude zeroes of the Jacobian \(J_X\) within B. When the geodesic field \(\mathbf Y(t,s;X_0)\) has the center \(X_0\in \partial \Omega _M\), we exempt from \(\Omega _M\) a disc about this singularity for a sufficiently small number \(\epsilon >0\). With the domain
$$\begin{aligned} \Omega _M^{\,\epsilon }(X_0):=\Big \{X\in \Omega _M\Big |\,|X-X_0|>\epsilon \Big \} \end{aligned}$$we modify the arguments in part (4) within the proof of Theorem 4, and we consider alternatively the auxiliary function
$$\begin{aligned} \psi (u,v):=\Psi (X(u,v);X_0),(u,v)\in {\overline{B}}_{\epsilon }:=\Big \{(u,v)\in {\overline{B}}\Big |X(u,v)\in \overline{\Omega _M^{\,\epsilon }(X_0)}\Big \}.\nonumber \\ \end{aligned}$$(4.23)Thus we can exclude each zero of the Jacobian in the interior of the disc B. With the part (5) in the proof of Theorem 4, we complete the derivation of our result above. \(\square \)
Remark 9
In order to show that a conformally parametrized H-surface represents a graph, one has to prove that the associate plane mapping is one-to-one. Here the investigation [11] contains as the decisive step that transversal mappings yield necessarily a diffeomorphism. There we need a stability condition in the sense that the second variation of the associate parametric integral is nonnegative.
5 Harmonic embeddings within the hemisphere
For all radii \(0<M<+\infty \) with their associate discs
we consider the upper hemisphere \(S^+_M\) in the following representation
Then we derive
and determine their first fundamental form (1.1) as follows
We denote the hemispherical metric (1.1), (5.3) by \(\,ds^2(M)\,\). This metric becomes singular near the boundary \(\partial \Omega _M\), and our Theorem 2 is not applicable globally. S. Hildebrandt, H. Kaul and K. Widman have constructed harmonic mappings into complete Riemannian manifolds with positive sectional curvature by direct variational methods (see [3]). Since this result is especially valid for hemispheres, we receive the following
Theorem 6
(Dirichlet problem for hemispherical harmonic mappings) Let a radius \(0<M<+\infty \) be chosen arbitrarily. For each boundary function
there exists a solution
for the Dirichlet problem \({\mathcal {P}}(\Omega _M,ds^2(M);\Phi )\) of the harmonic mapping associated with the hemispherical metric \(\,ds^2(M)\,\).
Proof
See the Theorems 1–4 in [3]. \(\square \)
We construct a field of geodesics, which emanates from an arbitrary equatorial point
and simply covers the hemisphere. We begin with the great circle on \(S^+_M\)
This circle represents a geodesic without interior conjugate points; it starts at the point \(Z_0=(M,0,0)\) and ends at the antipodal point \(Z_{\pi }=(-M,0,0)\), which is conjugate to \(Z_{0}\). We use the rotation by the angle s about the \(x^1\)-axis
and the rotation by the angle \(\vartheta \) about the \(x^3\)-axis
We obtain the field of geodesics
Via the projection from the Euclidean space onto the plane
we see from the construction above that the family of functions
constitutes a central field of geodesics for the hemispherical metric (1.1), (5.3). This central field \(\mathbf Y(.,.;\vartheta )\) simply covers the disc \(\Omega _M\) and emanates from the singular point \(\,X_0=\Pi ^3(X_{\vartheta })\in \partial \Omega _M\,\), where \(\,0\le \vartheta \le 2 \pi \,\) denotes an arbitrary angle.
With the aid of this central field of geodesics, we introduce geodesic regions and convex hulls for compact sets within \(\Omega _M\) as in Definition 4. Furthermore, we can define the geodesic function parallel to Definition 5 and receive the fundamental Lemma 3 for the hemispherical metric \(\,ds^2(M)\,\). Finally, we characterize convex Jordan contours \(\Gamma \subset \Omega _M\) as in Definition 9 with respect of the hemispherical metric. By the arguments in the proofs for Theorem 3 and Theorem 4, we can establish
Theorem 7
(Harmonic embeddings within the hemisphere) Let the convex Jordan contour \(\Gamma \subset \Omega _M\) and the topological boundary function \(\Phi :\partial B\rightarrow \Gamma \) as in (5.4) be given. Then each solution \(\,X(u,v)=(x^1(u,v),x^2(u,v))\,\) of the regularity (5.5) for the Dirichlet problem \({\mathcal {P}}(\Omega _M,ds^2(M);\Phi )\) of the hemispherical metric \(ds^2(M)\) furnishes a topological mapping of \({\overline{B}}\) onto \(\overline{I(\Gamma )}\) and a \(C^{2+\alpha }\)-diffeomorphism of B onto \(I(\Gamma )\).
Remark 10
Let \({\mathcal {M}}\) denote a 2-dimensional, geodesically complete, connected and oriented Riemannian manifold of the class \(C^3\) without boundary, whose Gaussian curvature is bounded from above by the constant \(\kappa \in [0,+\infty )\) as follows:
On this manifold we choose an arbitrary point \(P\in {\mathcal {M}}\) and a radius \(\displaystyle \,0<M<\frac{\pi }{2\sqrt{\kappa +}}\,\) such that the geodesic disc
satisfies a cut-locus-condition (see the treatise [3] by S. Hildebrandt, H. Kaul and K. Widman). Then we can solve the Dirichlet problem for harmonic mappings in the interior of \({\mathcal {B}}_M(P)\) by these investigations using direct variational methods. Here we can apply our methods from above in order to obtain harmonic diffeomorphisms.
Example 1
Harmonic diffeomorphisms in the Poincaré half-plane.
We consider the Poincaré half-plane (see [8] 5.1.3 in the Lectures by W. Klingenberg) with the following coefficients for their first fundamental form \(ds^2\) from (1.1):
Due to [8] Satz 5.1.7, the geodesics in the Poincaré half-plane with their Gaussian curvature \(\,K \equiv -1\,\) consist of all circular arcs within \(\mathbb R^2_+\) meeting the \(x_1\)-axis perpendicularly and the rays emanating orthogonally from the \(x_1\)-axis, which we address as orthocircles. From [1] §§ 81–84 we see that the geodesic discs \({\mathcal {B}}_M(P)\subset \mathbb R^2_+\) with their center P on the positive \(x_2\)-axis possess a convex circumference
with the constant geodesic curvature
Due to Figure 14 in [1] § 84 of the Grundlehren by W. Blaschke and K. Leichtweiß the circumferences for these geodesic discs constitute the orthogonal trajectories of the orthocircles. The geodesic discs \({\mathcal {B}}_M(P_M)\) exhaust the Poincaré half-plane for \(M\rightarrow +\infty \). Here we also refer to Abb. 5.1 in [8] 5.1.
Each boundary point \(X_0\in \partial {\mathcal {B}}_M(P_M)\) possesses a central field of geodesics, which emanates from \(X_0\) and foliates \({\mathcal {B}}_M(P_M)\). Consequently, the variational solution \(X(u,v),\,(u,v)\in {\overline{B}}\) of the Dirichlet problem for harmonic mappings by Hildebrandt, Kaul and Widman [3] exists within the geodesic discs of all radii \(M>0\). Then we can apply the methods from Sects. 2 to 4 above, and we see that this solution X shares the convex-hull property. Furthermore, this variational solution X yields a diffeomorphism in B and a topological mapping on \({\overline{B}}\) for topological boundary representations onto convex Jordan contours \(\Gamma \), which are contained in the interior of the disc \({\mathcal {B}}_M(P_M)\). Thus we receive an analogue of Theorem 7 within the Poincaré half-plane.
References
Blaschke, W., Leichtweiß, K.: Elementare Differentialgeometrie. Grundlehren der mathematischen Wissenschaften 1, 5. Auflage (vollständige Neubearbeitung). Springer, Berlin (1973)
Dierkes, U., Hildebrandt, S., Sauvigny, F.: Minimal Surfaces. Grundlehren der mathematischen Wissenschaften 339. Springer, Berlin (2010)
Hildebrandt, S., Kaul, H., Widman, K.: An existence theorem for harmonic mappings of Riemannian manifolds. Acta Math. 138, 1–16 (1977)
Hildebrandt, S., Sauvigny, F.: Embeddedness and uniquenes of minimal surfaces solving a partially free boundary value problem. J. Reine Angew. Math. 422, 69–89 (1991)
Hildebrandt, S., Sauvigny, F.: On one-to-one harmonic mappings and minimal surfaces. Nachrichten der Akademie der Wissenschaften in Göttingen, II. Math. Phys. Klasse, Jahrgang Nr.3 (1992)
Jäger, W., Kaul, H.: Uniqueness and stability of harmonic maps and their Jacobi fields. Manuscr. Math. 28, 269–291 (1979)
Jost, J.: Univalency of harmonic mappings between surfaces. J. Reine Angew. Math. 342, 141–153 (1981)
Klingenberg, W.: Eine Vorlesung über Differentialgeometrie. Heidelberger Taschenbücher. Springer, Berlin (1973)
Kneser, H.: Lösung der Aufgabe 41. Jahresber. Dt. Math. Vereinigung 35, 123–124 (1926)
Nitsche, J.C.C.: Vorlesungen über Minimalflächen. Grundlehren der mathematischen Wissenschaften 199. Springer, Berlin (1975)
Sauvigny, F.: Flächen vorgeschriebener mittlerer Krümmung mit eineindeutiger Projektion auf eine Ebene. Math. Z. 180, 41–67 (1982)
Sauvigny, F.: Partial Differential Equations. 1. Foundations and Integral Representations; 2. Functional Analytic Methods; With Consideration of Lectures by E. Heinz. Springer Universitext, 2nd edn. Springer, London (2012)
Sauvigny, F.: Analysis - Grundlagen, Differentiation, Integrationstheorie, Differentialgleichungen, Variationsmethoden. Springer, Berlin (2014)
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by J. Jost.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.