DOI 10.1515 / ADVGEOM.2009.021 de Gruyter 2009
Unstable minimal surfaces of annulus type in manifolds
Hwajeong Kim∗
(Communicated by F. Duzaar)
Abstract. Unstable minimal surfaces are the unstable stationary points of the Dirichlet integral.
In order to obtain unstable solutions, the method of the gradient flow together with the minimax- principle is generally used, an application of which was presented in [19] for minimal surfaces in Euclidean space. We extend this theory to obtain unstable minimal surfaces in Riemannian manifolds. In particular, we consider minimal surfaces of annulus type.
1 Introduction
For given curvesΓl ⊂ N,l = 1, . . . , m andΓ := Γ1 ∪ · · · ∪Γm, where(N, h)is a Riemannian manifold of dimensionn≥2 with metric(hαβ), we denote the generalized Plateau Problem byP(Γ). This deals with minimal surfaces bounded byΓ, in other words parametrizationsXdefined onΣ⊂R2with∂Σ = Γ, satisfying the following constraints:
(1) τh(X) =0,
(2) |Xu|2h− |Xv|2h=hXu, Xvih=0, (3) X|∂Σis weakly monotone and ontoΓ,
whereτh := ∆Xα−Γαβγ∇XβXγ =0 is the harmonic equation on(N, h)seen as the Euler–Lagrange equation of the energy functional.
A regular minimal surface is called unstable if its surface area is not a minimum among neighbouring surfaces with the same boundary. Extending the Ljusternik–
Schnirelmann theory on convex sets in Banach spaces, a variational approach to unstable minimal surfaces of disc or annulus type inRnwas proposed in 1983 ([21], see also [19], [20]). For the minimal surfaces of higher topological structure inRn, see [11].
Recently in [8], the existence of unstable minimal surfaces of higher topological struc- ture with one boundary in a nonpositively curved Riemannian manifold was studied by applying the method in [19]. In particular, the first part of that paper considers the Jacobi field extension operator as the derivative of the harmonic extension.
∗This paper is based on my thesis [12] supervised by Professor Michael Gr¨uter.
In this article, we study unstable minimal surfaces of annulus type in manifolds. The Euclidean case was tackled already in [20], and our aim is to generalize this result to manifolds satisfying appropriate conditions. Namely, we will consider two boundary curvesΓ1,Γ2in a Riemannian manifold(N, h)such that one of the following holds.
(C1) There existsp∈ N withΓ1,Γ2 ⊂ B(p, r), whereB(p, r)lies within the normal range of all its points. We assume thatr < π/(2√
κ), whereκis an upper bound of the sectional curvature of(N, h).
(C2) Nis compact with nonpositive sectional curvature.
These conditions are related to the existence and uniqueness of the harmonic extension for a given boundary parametrization.
First, we construct suitable spaces of functions, the boundary parametrizations, dis- tinguishing the Cases (C1) and (C2). We introduce a convex set which serves as a tangent space for the given boundary parametrization. Then we consider the following functional:
E(x) :=1 2
Z
|dF(x)|2h,
whereF(x) denotes the harmonic extension of annulus type or of two-disc type with boundary parametrizationx. We next discuss the differentiability ofE, in particular for the case in which the topology of the surfaces changes (from an annulus to two discs).
Defining critical points ofE, will show the equivalence between the harmonic extensions (inN) of critical points ofE and minimal surfaces inN. The H2,2-regularity of the harmonic extension of a critical point ofE(see the appendix or [13]) plays an important role in the argument.
In Section 4, we prove the Palais–Smale condition forE. In particular, we investigate carefully the behaviour of boundary mappings which are fixed at only one point. In order to deform level sets ofE, we also construct a suitable vector field and its corresponding flow. Roughly speaking, Lemma 4.3 shows that the energy of some annulus-type har- monic extensions is greater than that of two-disc type harmonic extensions by a uniformly positive constant. Although this result refers to Riemannian manifolds, it turns out to be more restrictive than that of Euclidean spaces, which holds uniformly on any bounded set of boundary parametrizations. This somewhat weaker result is anyhow enough for the present purposes.
Following the arguments set out in [21], we can prove the main theorem of this paper.
This states that if there exists a minimal surface (of annulus type) whose energy is a strict relative minimum inS(Γ1,Γ2)(suitably defined for each Case (C1) and (C2)), the exis- tence of an unstable minimal surface of annulus type is ensured under certain assumptions related to the solutions ofP(Γi). We eventually apply this result to the three-dimensional sphereS3 and the three-dimensional hyperbolic spaceH3, whose curvatures are 1 and
−1, respectively.
2 Preliminaries
2.1 Some definitions. Let(N, h)be a connected, oriented, complete Riemannian man- ifold of dimensionn≥2, embedded isometrically and properly into someRkas a closed
submanifold by means of the mapη([3]). Moreover,dωandd0denote the area elements inΩ⊂R2and in∂Ωrespectively.
Indicating
B:={w∈R2| |w|<1}
we define
H1,2∩C0(B, N) :={f ∈H1,2∩C0(B,Rk)|f(B)⊂N}
with normkfk1,2;0:=kdfkL2+kfkC0. Now set
TfH1,2∩C0(B, N)∼={V ∈H1,2∩C0(B,Rk)|V(·)∈Tf(·)N}
=:H1,2∩C0(B, f∗T N) with norm
kVk:=
Z
B
|∇fV|2hdω 1/2
+kVkC0 ∼= Z
B
|dV|2Rkdω 1/2
+kVkC0. (1) LetΓbe a Jordan curve in N diffeomorphic toS1 := ∂B. ThenN can be equipped with another metric ˜h such that Γ is a geodesic in (N,˜h). We observe that H1,2 ∩ C0 (B, ∂B),(N,Γ)h˜
andH1,2∩C0 (B, ∂B),(N,Γ)h
coincide as sets.
Using the exponential map in(N,˜h), we let
H12,2∩C0(∂B; Γ) :={u∈H12,2∩C0(∂B,Rk)|u(∂B) = Γ}, where the norm is given bykuk1
2,2;0 :=kdH(u)kL2+kukC0, andH(u)is the harmonic extension inRkwithH(u)|∂B(·) =u(·). In addition
TuH12,2∩C0(∂B; Γ)
:={ξ∈H12,2∩C0(∂B, u∗T N)|ξ(z)∈Tu(z)Γ,for allz∈∂B}
=H12,2∩C0(∂B, u∗TΓ).
Finally, the energy off ∈H1,2(Ω, N)is denoted by E(f) := 1
2 Z
Ω
|df|2hdw.
2.2 The setting. Let Γ1,Γ2 be two Jordan curves of classC3 inN with diffeomor- phismsγi:∂B→Γi, i=1,2, anddist(Γ1,Γ2)>0. Forρ∈(0,1)let
Aρ:={w∈B|ρ <|w|<1}
have boundaryC1:=∂BandCρ:=∂Bρ=:C2(ρfixed), and indicate
Xmoni :={xi∈H12,2∩C0(∂B; Γi)|xiis weakly monotone and ontoΓiwith degree 1}.
I. We first consider the following condition for(N, h)(⊃Γ1,Γ2).
(C1) There existsp∈ N withΓ1,Γ2 ⊂ B(p, r), whereB(p, r)lies within the normal range of all its points. We assumer < π/(2√
κ), whereκis an upper bound of the sectional curvature of(N, h).
Throughout the paper,B(p, r)denotes a geodesic ball with centerp∈Nas in (C1). We can easily observe the following property (see [13]).
Remark 2.1. IfΓ1,Γ2 ⊂ N satisfy (C1), then for each xi ∈ H12,2∩C0(∂B; Γi)and ρ∈(0,1)there existgρ∈H1,2∩C0(Aρ, B(p, r))andgi∈H1,2∩C0(B, B(p, r))with gρ|C1 =x1, gρ|Cρ(·) =x2(ρ·)andgi|∂B=xi,i=1,2.
From the results in [7], [9] and the above remark, we have a unique harmonic map of annulus and disc type in B(p, r) ⊂ N for a given boundary mapping in the class H12,2∩C0. Now we define
Mi:={xi∈ Xmoni |xipreserves the orientation}.
ThenMiis complete, since theC0-norm preserves monotonicity. Moreover, let S(Γ1,Γ2) ={X ∈H1,2∩C0(Aρ, B(p, r))|0< ρ <1, X|Ciis weakly monotone},
S(Γi) ={X ∈H1,2∩C0(B, B(p, r))|X|∂Bis weakly monotone}.
II. We now investigate another significant condition for(N, h).
(C2) Nis compact with nonpositive sectional curvature.
A compact Riemannian manifold is homogeneously regular and the condition of nonpos- itive sectional curvature impliesπ2(N) =0. In order to defineMi, we first consider for ρ∈(0,1)the following
G˜ρ:={f ∈H1,2∩C0(Aρ, N)|f|Ciis continuous, weakly monotone and ontoΓi}.
We may take a continuous homotopy class, denoted byF˜ρ⊂G˜ρ, so that every two el- ementsf, ginF˜ρare continuously homotopicf ∼g(not necessarily fixing the boundary parametrization). We further demand some relationF˜ρ∼F˜σto hold for anyρ, σ∈(0,1).
Precisely, for somef˜∈F˜σ,f ∈F˜ρand some diffeomorphismτσρ : [σ,1] →[ρ,1], we requiref˜(r, θ) =f(τσρ(r), θ). LetF˜ρbe fixed. Then for anyσ∈(0,1)we can findF˜σ
withF˜ρ∼F˜σ.
We now consider all possibleH1,2∩C0-extensions of disc type inN: S(Γi) :={X ∈H1,2∩C0(B, N)|X|∂Bis weakly monotone ontoΓi}, assuming that this set is not empty, for eachi=1,2.
Lemma 2.1. (i) ForX1∈ S(Γ1)andX2∈ S(Γ2)there existsfρ∈H1,2∩C0(Aρ, N) such thatfρ|C1(·) =X1|∂B(·)andfρ|Cρ(·) =X2|∂B(ρ·), forρ∈(0,1).
(ii) Moreover, there existsρ0 ∈(0,1)and a uniform positive constantCsuch that for somefρ∈H1,2∩C0(Aρ, N), withfρ|Cρ(·) =X2|∂B(ρ·)
E(fρ)≤C, for allρ≤ρ0. (2)
Proof. (i) For a given ε > 0, take σi > 0 with oscBσiXi < ε. Choose ρ > 0 with σρ
2 < σ1, and let H : Bσ1\Bρ
σ2
→ Rk be harmonic with X1|∂Bσ1 −X1(0) on ∂Bσ1 andX2|∂Bσ2 −X2(0) on ∂Bρ
σ2
. This implieskHkC0 < ε. Now letg ∈ H1,2∩C0(Bσ1\Bρ
σ2
, N)withX1(0)on∂Bσ1andX2(0)on∂Bρ
σ2
.
Considering coordinate neighbourhoods for the submanifoldN,→η Rk, we may take a finite covering offρ((Aρ)), and by projection we obtain a smooth mapr:Nδ(fρ(Aρ))→ N withr|N
δ(fρ(Aρ))∩N = Idfor someδ > 0, whereNδ(·)isδ-neighbourhood inRk. SettingT(s, θ) := (1sρ, θ)in polar coordinates, we can definefρwith the desired proper- ties:
fρ:=
X1|B\Bσ
1 onB\Bσ1,
r◦(g+H) onBσ1\Bρ
σ2
, X2(T−1(·)) onBρ
σ2
\Bρ.
(3)
(ii) The claim follows from the above construction, since σρ
2 < σ1,ρ≤ρ0for some
ρ0 >0. 2
Under the assumption thatS(Γi) 6=∅, for givenΓi ∈ N we have an annulus-type- extension like that of (3), and we take homotopy classes which contain such an extension.
From now on twiddles will be dropped.
Define
S(Γ1,Γ2) :={f ∈Fρ|0< ρ <1}, (4) as well as the two function spaces
M1:={x1(·) =f|C1(·), f ∈ S(Γ1,Γ2)|x1is orientation preserving with degree 1}, M2:={x2(·) =f|Cρ(·ρ), f∈ S(Γ1,Γ2)|x2is orientation preserving with degree 1}.
Forxi ∈ Xmoni ,Hρ(x1, x2)denotes the uniqueRk-harmonic extension onAρwithx1(·) onC1 andx2(ρ·) onCρ, while H(x)is the Rk-harmonic extension of disc type with boundaryx∈ Xmoni .
Lemma 2.2. (i) For eachxi0∈Mi,i=1,2, there existsε(xi0)>0such that ifxi∈ Xmoni withkxi−xi0k1
2,2;0< ε, thenxi∈Mi. (ii) Miis complete with respect tok · k1
2,2;0.
Proof. (i) Letfρ∈F˜ρwithfρ|C1=x10andfρ|Cρ(·) =y2(ρ·)for somey2∈M2. We consider the smooth retractionr:Nδ(fρ(Aρ))→Nas in the proof of Lemma 2.1.
Letkxi−xi0k1
2,2;0< ε < δ. Then by Lemma 4.2 from [20], Z
Aρ
|d(r(fρ+Hρ(x1−x10,0)))|2dω
≤C(kfρkC0, ε, N) Z
Aρ
|dfρ|2dω+ Z
B
|dH(x1−x10)|2dω
≤C(kfρk1,2;0, ε, N).
Now, let H(t,·) := (1 −t)Hρ(x1 −x10,0) : [0,1]×Aρ → Rk with kHkC0 < ε andG : [0,1]×Aρ → N withG(t,·) = fρ(·)for allt ∈ [0,1]. Sincer(G+H) : [0,1]×Aρ → N is a homotopy betweenfρ andr(fρ +Hρ(x1 −x10,0)), it follows r(fρ +Hρ(x1 −x10,0))(∼ fρ) ∈ F˜ρ, andx1 ∈ M1. Similarly, we can prove that x2∈M2ifkx2−x20k1
2,2;0< ε0for some smallε0 >0.
(ii) A Cauchy sequence{xin} ⊂Miconverges toxi ∈H12,2∩C0(∂B; Γi), and for somen,kxin−xikC0 < ε. ConsideringHρ(x1−x1n,0)andgρ ∈ Fρwithx1n onC1
and 0 onCρ, we can find a homotopy inN betweengρandr(gρ+Hρ(x1−x10,0))as in (i). We may also apply this argument tox2. Note thatxiis weakly monotone, and hence
xi∈Mi, concluding the proof. 2
From the proof we easily conclude that the set ofxi’s which possess annulus-type- extensions with uniform energy with respect toρ ≤ρ0is an open and closed subset of Xmoni . Thus, it is a non-empty connected component ofXmoni and must coincide withMi, sinceMiis a connected subset ofXmoni . Hence we obtain the following property.
Remark 2.2. For eachxi ∈ Mi, i =1,2, there existfρ ∈ S(Γ1,Γ2)andC > 0 with E(fρ) ≤ C for all ρ ≤ ρ0 and someρ0 ∈ (0,1). Clearly, this result also holds for xi∈Miif we assume (C1).
For disc-type extensions ofxi∈Mithe following lemmata will be useful.
Lemma 2.3. Let(N, h)be a homogeneously regular manifold anduan absolutely con- tinuous map of ∂Br(x0)into N 3 x0 withR2π
0 |u0(θ)|2hdθ ≤ Cπ0. Then there exists f ∈ H1,2(Br(x0), N) ∩ C0(Br(x0), N) with f|∂Br(x0) = u and EBr(x0)(f) ≤
C00 C0
R2π
0 |u0(θ)|2hdθ, whereC00, C0are the constants defined by homogeneous regularity.
Proof. See [17, Lemma 9.4.8 b)]. 2
Lemma 2.4. Let fρ ∈ H1,2(Aρ, N), 0 < ρ < 1. For each δ ∈ (ρ,1)there exists τ∈(δ,√
δ)withR2π 0
∂fρ(τ,θ)
∂θ
2
hdθ≤4E(fρ)
ln1δ .
Proof. Similar to the proof of the Courant–Lebesgue lemma. 2
Forxi ∈ Mi, and given the choice ofS(Γ1,Γ2), Remark 2.2 tells that we can find fρ ∈ H1,2(Aρ, N)with boundaryxi such thatE(fρ) ≤C for allρ ≤ρ0. Then from Lemma 2.4 and Lemma 2.3, we havegτ ∈H1,2(Bτ, N)with boundaryfρ|∂Bτ for some ρ. Together withgτandfρ|B\Bτ, we obtain a mapX ∈H1,2(B, N)with boundaryx1. Similarly, we haveX˜ ∈H1,2(B, N)with boundaryx2.
Moreover, the harmonic extension of disc type for eachxi ∈ Mi inN is unique, independently of the choice of homotopy classS(Γ1,Γ2), because of the following well- known fact.
Lemma 2.5. π2(N) =0⇔Anyh0, h1∈C0(B, N)withh0|∂B=h1|∂Bare homotopic.
On the other hand, using the Construction (3) and the previous lemma we can easily check that the traces of the elements inS(Γi)belong toMi. From [1], [15], [5], we then have the following.
Remark 2.3. (i) Forxi ∈ Mi, there exists a unique harmonic extension of disc type onBand of annulus type onAρ,ρ∈(0,1).
(ii) The elements ofMiare the traces of the elements ofS(Γi).
III. Now let(N, h)andΓi, i = 1,2 satisfy (C1) or (C2). Observing∂B ∼= R/2π, for a given orientedyi∈ Xmoni there exists a weakly monotone mapwi ∈C0(R,R)with wi(θ+2π) =wi(θ) +2πsuch thatyi(θ) =γi(cos(wi(θ)),sin(wi(θ))) =:γi◦wi(θ).
In additionwi= ˜wi+ Idfor somew˜i∈C0(∂B,R).
Denoting the Dirichlet integral byDand theRk-harmonic extension byH, let Wi
Rk :={wi∈C0(R,R)|wiis weakly monotone, wi(θ+2π) =wi(θ) +2π;
D(H(γi◦wi))<∞}.
Clearly,Wi
Rkis convex (for further details, refer to [21]).
Now takexi∈Mi. Consideringw−wias a tangent vector alongw˜i, let Txi =
dγi((w−wi)d
dθ ◦w˜i)|w∈Wi
Rkandγi◦wi=xi
.
Note thatTxiis convex inTxiH12,2∩C0(∂B; Γi), sinceWi
Rkis convex. Forξ=dγi((w−
wi)dθd ◦w˜i)∈ Txiwe have thatexpgxiξ=γi(w),expgdenoting the exponential map with respect to the metric˜h.
If (C1) holds, then clearlyexpgxiξ∈Miforξ∈ Txi. For the Case (C2), let us recall the proof of Lemma 2.2. SinceNis compact, there existsli >0, depending onγi, such that for anyxi∈Mi,expgxiξ∈Mi, provided thatkξkT
xi < li. The following set-up holds true in both Cases (C1) and (C2).
Definition. (i) Let M := M1×M2×(0,1)with the product topology and x :=
(x1, x2, ρ)∈ M. Then the setTxM:=Tx1× Tx2×Ris convex.
LetF(x) =F(x1, x2, ρ) =Fρ(x1, x2) :Aρ →N be the unique harmonic exten- sion withx1onC1andx2(ρ·)onC2, and define
E :M −→R: x7−→E(F(x)) := 1 2 Z
Aρ
|dFρ(x1, x2)|2hdω.
(ii) Define∂M:=M1×M2× {0},Tx∂M:=Tx1× Tx2andM:=M ∪∂M.
LetFi(xi) : Aρ → N be the unique harmonic extension with boundary xi, for x= (x1, x2,0)∈∂M, and define
E(x) :=E(F1(x1)) +E(F2(x2)).
2.3 Harmonic extension operators. LetΩ = AρorΩ = B. A weak Jacobi fieldJ with boundaryξalong a harmonic functionf is a weak solution of
Z
Ω
h∇J,∇Xi+htrR(J, df)df, Xidω=0,
for all X∈ H1,2(Ω, f∗T N)withX|∂Ω = ξ. Weak Jacobi fields are natural candidate derivatives of the harmonic operatorsFρandFi.
We have the following property of weak Jacobi fields, from [8].
Lemma 2.6. The weak Jacobi fieldJwith boundaryη∈TxiH12,2∩C0along a harmonic Fwith boundaryxiis well defined in the classH1,2and continuous up to the boundary.
It satisfies
kJFkC0 ≤ kJF|∂ΩkC0, kJFk1,2;0≤C(N,kfk1,2:0)kJF|∂Ωk1 2,2;0. Now we can discuss the differentiability of harmonic extension operators.
Lemma 2.7. The operatorsFρ,Fiare partially differentiable inx1(respectivelyx2)for variations inTx1H12,2∩C0(respectivelyTx2H12,2∩C0). Their derivatives are continuous Jacobi field operators with respect tox1, x2.
Proof. The proof reproduces an argument we shall explain in full detail in Lemma 3.1, Cases (B), (C), and as such will not be anticipated here. Alternatively, one can follow the
aforementioned [8]. 2
3 The variational problem
3.1 Differentiability ofEonM.
Lemma 3.1. The following hold:
(A) E is continuously partially differentiable inx1, x2with respect to variations inTx1, Tx2and the derivatives are continuous onM1×M2.
(B) Eis continuous with respect toρ∈[0,1), even uniformly onNε(xi0)for someε >0 independent ofxi0∈Mi,i=1,2.
(C) The partial derivatives inx1, x2are continuous with respect toρ∈[0,1), uniformly continuous onNε(xi0)for someε >0independent ofxi0∈Mi,i=1,2.
(D) Eis differentiable with respect toρ∈(0,1).
Proof. From now on, continuity will be understood in the sense of convergence of subse- quences.
(A) The Dirichlet integral functional is inC∞, so Lemma 2.7 guarantees thatE is continuously partially differentiable with continuous partial derivatives onM1×M2.
Computation of the derivatives:Letx= (x1, x2, ρ)∈ M,ξ1 ∈ Tx1. By Lemma 2.2 there is a smallt0>0 such thatexpgx1(tξ1)∈M1, 0≤t≤t0. Thus,
hδx1E, ξ1i:= d dt t=0
E(expgx1(tξ1), x2, ρ)
= Z
Aρ
hdFρ(x1, x2),∇Dx1Fρ(x1, x2)(ξ1)ihdω
= Z
Aρ
hdFρ(x1, x2),∇JFρ(ξ1,0)ihdω (by Lemma 2.7), (5)
since by computation we obtain, withFρ(t) :=Fρ(gexpx1(tξ1), x2),
∇d dt
Fρ,iα (t)dxi⊗ ∂
∂yα◦ Fρ(t)
=∇d
dtFρ expgx1(tξ1), x2
=∇ Dx1Fρ(x1, x2)(ξ1)
, t=0 .
Forξ2 ∈ Tx2 Lemma 2.7 yieldshδx2E, ξ2i = R
AρhdFρ(x1, x2),∇JFρ(0, ξ2(ρ·))ihdω.
Similarly, forx= (x1, x2,0)∈∂M,hδxiE, ξii=R
BhdFi(xi),∇JFi(ξi)ihdω, i=1,2.
For (B) we shall split the proof into three sub-steps B-I), B-II), B-III). Similarly for (C) we shall have C-I), C-II), C-III).
B-I)The set-up. The claim is thatEis continuous whenρ→ρ0. Fixingρ0 =0 is no great restriction, since the proof forρ0∈(0,1)carries over in an analogous, even easier, fashion. Takingρ0 =0 translates our claim into
Z
Aρ
dFρ(x1, x2)
2 hdω−→
Z
B
dF1(x1)
2 hdω+
Z
B
dF2(x2)
2
hdω (6)
uniformly onNε(xi0)for someε >0 independent ofxi0 ∈Mi, wheneverρ→0.
LetFρ :=Fρ(x1, x2)andFi := Fi(xi), i= 1,2. By Lemma 2.4, for eachδwith 0< ρ < δ <1 there existsν ∈(δ,√
δ)such that Z 2π
0
∂Fρ(ν, θ)
∂θ h
dθ≤√ 2π
Z 2π 0
∂Fρ(ν, θ)
∂θ
2
h
dθ 12
≤ C
p|lnδ|. (7) Due to Remark 2.2,Cis independent ofρ≤ρ0, for someρ0∈(0,1).
By means ofFρwe now construct two maps by setting
fν:Aν−→N with fν(reiθ) :=Fρ(reiθ), reiθ∈Aν,
gν0 :Aν0−→N withgν0(reiθ) :=Fρ(T(reiθ)), reiθ∈Aν0. (8) The constantsν0 := ρν,ν ∈ (δ,√
δ)andδ ∈ (ρ,1)satisfy the property (7) in the limit ν0, ν → 0 for ρ → 0. (One can take for instance δ = √
ρ). The map T(reiθ) =
ρ
reiθ goes fromAν0 toBν\Bρ surjectively. Then,fν andgν0 are harmonic maps into N withfν|∂B = x1,gν0|∂B = x2 andosc∂Bνfν → 0, osc∂Bν0gν0 → 0 asρ → 0.
Moreover, sinceTis conformal,E(Fρ) =E(Fρ|Aν) +E(Fρ|Bν\Bρ) =E(fν) +E(gν0) by conformal invariance of the Dirichlet integral.
B-II)The convergence of{fν},{gν0}toFi. We first investigate the modulus of con- tinuity of harmonic maps{hν}:Aν →N which converge uniformly (C0-norm) on∂B withE(hν)≤Lfor someL >0, independent ofν ≤ν0for someν0 ∈(0,1). We shall only deal with the assumption (C2), because the argument can clearly be applied to the Case (C1) as well.
LetGR := BR(z) ⊂ Aν for ν ≤ ν˜0. If z ∈ ∂B, considerGR := BR(z)∩Aν. Givenε > 0, by the Courant–Lebesgue lemma there existsδ > 0, independent ofν ≤ ν0, such that the length of hν|∂Gδ does not exceed min{ε4,i(N)4 }, i(N) > 0. Then hν|∂Gδ ⊂B(q, s)for someq∈N, s≤min{ε2,i(N2)}. Observe thathνis continuous on
∂Gδ, and there exists anH1,2-extensionX of disc type, whose image is inB(q, s)with X|∂Bδ = hν|∂Bδ, by the same argument of Remark 2.1. Thus there exists a harmonic extensionh0withh0(Gδ)⊂B(q, s)⊂B(q,ε2), by [7]. From Lemma 2.5,h0is homotopic tohonGδ, and from the energy minimizing property of harmonic maps,hν|Gδ = h0. Hence, the functionshνwithν ≤ν0have the same modulus of continuity. Furthermore, if these mappings have the same boundary image, they areC0-uniformly bounded on each relatively compact domain.
Now apply the above result to{Fρ, ρ ≤ ρ0}inRk. For someρ0 ∈ (0,1)then, the functionsfνrespectivelygν0 have the same modulus of continuity for allρ∈(0, ρ0), and some subsequences, denoted again byfν respectivelygν0 are locally uniformly conver- gent. Recall that our maps are continuous, so by localizing in both domain and image, harmonic functions, seen as solutions of Dirichlet problems, may be also regarded as weak solutionsfof the following elliptic systems in local coordinate charts ofN:
didifα=−Γαβγdifβdifγ=:Gα(·, f(·), df(·)). (9) We can take the same coordinate charts for the image of{fν}ν≤ν0and{gν0}ν0≤ν00, where ν0 := ν(ρ0),ν00 := ν0(ρ0), to the effect that we have the same weak solution system for (9). Moreover, sincehαβ andΓαβγ are smooth, the structure constants of the weak systems (see [10, Section 8.5]) are independent ofρ≤ρ0.
Now considerKσσ ={σ ≤ |z| ≤ 1−σ},σ∈(0,1). From the regularity theory of [14] and [10, Section 8.5] and by the covering argument, there existsC ∈ Rsuch that kfν|KσσkH4,2 ≤Cfor allν ∈ (0, ν0). Hence the Sobolev’s embedding theorem implies that for some sequence{ρi} ⊂ (0,1), limρi→0fν(ρi)|Kσσ = f0 inC2(Kσσ,Rn), with τh(f0) =0 inKσσ.
Forσ:= n1, we choose a sequence{fν(ρn,i)}as above such that{ρn+1,i}is a subse- quence of{ρn,i}. By diagonalizing we obtain a subsequence{fν(ρn,n)}, n≥n0 which converges locally tof0in theC2-norm, sof0is harmonic onB\(∂B∪ {0}).
On the other handfν|∂B =x1for allν, and thefν’s converge uniformly tof0in a compact neighbourhood of∂B. Thus,f0is continuous onB\{0}withf0|∂B=x1. Also observe thatosc∂Brf0→0 asr→0, by construction.
For each compactK ⊂ B\{0},R
K|df0|2dω = limρi→0R
K|dfν(ρi)|2 ≤ L, withL independent ofK. Thus, f0 ∈ H1,2(B\{0}, N), andf0 can be extended to a weakly harmonic map onB ([10, Lemma 8.4.5], see also [18], [4]). Thus,f0can be considered weakly harmonic andf0∈C0(B, N)∩C2(B, N)withf0|∂B =x1, so uniqueness forces f0=F1(x1).
Similar results hold forgν0.
B-III) The convergence of the energy. We consider η◦ f, and denote it again by f := (fa)a=1,...,k ∈ H1,2(Ω,Rk)for obvious reasons. Sinceη is isometric, forf :=
(fα)α=1,...,n ∈ H1,2(Ω, N)we haveR
Ω|d(fα)|2hdω = R
Ω|d(fa)|2
Rkdω. A harmonic mapf ∈H1,2(Ω, N)satisfies
Z
Ω
(hdf, dψi − hII◦f(df, df), ψi)dw=0 (10) for anyψ∈H01,2∩C0(Ω,Rk), whereIIis the second fundamental form ofη.
SetKσ={σ≤ |z| ≤1}, σ >0 and we considerRk-harmonic mapsHνandHeνon KσwithHν|∂Kσ =fν|∂KσandHeν|∂Kσ =F1|∂Kσ, whereν ∈(0, σ). LetH:B→Rk be harmonic withH|∂B = Hν|∂B = Heν|∂B = x1. Then{Hν},{Heν} have the same modulus of continuity up to∂B, and we havekHν−HkC0;Kσ →0, kHeν−HkC0;Kσ → 0 asν→0. Furthermore, forXν:= (fν− F1) + (Hν−Heν)∈H01,2∩C0(Kσ,Rk), we obtain
kXνk(C0;Kσ)≤ kfν− F1kC0;Kσ+kHν−HkC0;Kσ+kH−HeνkC0;Kσ →0 asν→0.
Now consider Z
Kσ
hd(fν− F1), d(fν− F1)idω
= Z
Kσ
hd(fν− F1), dXνidω
| {z }
=:I
− Z
Kσ
hd(fν− F1), d(Hν−Heν)idω
| {z }
=:II
.
Whenν →0
|I| ≤ Z
Kσ
hII◦fν(dfν, dfν), Xνidω
+ Z
Kσ
hII◦(dF1, dF1), Xνidω
=C(kfνk1,2;0),kF1k1,2;0)kXνk(C0;Kσ)→0 (11) from (10). Moreover, sinceHν−Heνis harmonic inRk,
|II| ≤ Z
∂Bσ
∂r(Hν−Heν)
dωkfν− F1kC0;Kσ →0 asν →0. (12)
ThusR
Kσ|d(fν− F1)|2dω→0, andR
Kσ|dfν|2dω→R
Kσ|dF1|2dω, for anyKσ. Since R
Bσ|dF1|2dω → 0 asσ → 0, we obtain R
Aν|dfν|2dω → R
B|dF1|2dω as ν → 0.
SimilarlyR
Aν0|dgν0|2dω→R
B|dF2|2dω asν0→0.
Now to the uniform convergence onNε(xi0). Replacef(Aρ)byB(p, r)(for (C1)) orN (for (C2)) in the proof of Lemma 2.2. Then,kFρ(x1, x2)kH1,2 ≤ Cuniformly on Nε(xi0), where the constantCdepends onxi0, whileεdoes not. The convergence in (11), (12) is uniform onNε(xi0). The proof of (B) is eventually completed.
C-I)The set-up. We must show that forxi∈Miandξi∈ Txi,
hδxiEρ, ξii −→ hδxiE, ξii uniformly onNε(xi0)⊂Mi, i=1,2 asρ→0.
It suffices to show the assertion fori=1. We know that hδx1Eρ, ξ1i
= Z
Aν(ρ)
hdFρ(x1, x2),∇JFρ(ξ1,0)ihdω+ Z
Bν(ρ)\Bρ
hdFρ(x1, x2),∇JFρ(ξ1,0)ihdω
= Z
Aν(ρ)
hdFρ(x1, x2),∇JFρ(ξ1,0)ihdω+ Z
Aν0
hdgν0,∇Jgν0(0, ζν0)idω,
wheregν0(·) =Fρ◦T(·)andζν0(ν0eiθ) = JFρ(ξ1,0)(νeiθ)withν0 := ν(ρ)ρ . Observe JFρ(ξ1,0)◦T is a Jacobi field alonggν0, by the conformal property ofT.
C-II)The convergence of Jacobi fields. First, letVν :=JFρ(ξ1,0)|Aν =vαν∂y∂α ◦fν, for which we will show the existence of aν0∈(0,1)giving
kDVνk22:=
Z
Aν
hαβ◦fνvαν,ivν,iβ dω≤Cfor allν∈(0, ν0). (13) By direct computation kDVνk22 ≤ CE(Vν) +C(N,kVνkC0,kfνkC0, E(fν)). Since Lemma 2.6 yieldskVνkC0 ≤ kξν1kC0, we only need to show that
E(Vν) :=
Z
Aν
|∇fνVν|2dω≤C, ν∈(0, ν0). (14)
LetXν := xαν∂y∂α ◦fν ∈ H1,2(Aν, fν∗T N), wherexαν(z) := vα2ν
0(τν2ν0
0 (z)),ν0 ≤
|z| ≤1 (see Section 2.2 for the definition ofτν2ν00) andxαν(z) :=0,ν≤ |z| ≤ν0. Clearly, kDXνk22≤C(ν0, N)kDV2ν0k22for allν≤ν0.
By the minimality property of Jacobi fields and Young’s inequality, Z
Aν
(|∇fν(Vν)|2− htrR(dfν, Vν)dfν, Vνi)dω
≤ Z
Aν
(|∇fν(Xν)|2− htrR(dfν, Xν)dfν, Xνi)dω
≤ Z
Aν
hαβ◦fνxαν,ixβν,idω+ε Z
Aν
|xαν,i ∂
∂yα◦fν|2hdω +ε−1
Z
Aν
|xγνf,iδΓβγδ◦fν
∂
∂yβ ◦fν|2hdω +
Z
Aν
hαβ◦fνxγνxλνf,iδf,iµΓαγδ◦fνΓβλµ◦fνdω− Z
Aν
htrR(dfν, Xν)dfν, Xνidω
≤C(N, ε,kfνkC0, E(fν),kV2ν0kC0,kDV2ν0k22).
ButE(Vν)≤C,ν∈(0, ν0), since Z
Aν
htrR(dfν, Vν)dfν, Vνidω≤C(N,kfνkC0, E(fν),kξ1kC0).
Therefore we have (13), and this means that {(vνα) | ν ≤ ν0}α=1,...,n has the same modulus of continuity, see the argument in B-III) and Lemma 2.6.
With the same charts as in (B),(vαν(ρ)) ∈ Rn, ν ≤ ν0 are weak solutions of the Jacobi fields system with uniformly bounded energy and same modulus of continuity on Kσ = {σ ≤ |z| ≤ 1}, withσ > 0 for small ρ, again by Lemma 2.6. Just as in (B), {Vν} converges to the Jacobi field alongF1|B\{0} with boundary ξ1, and for JF1(ξ1) =:wα ∂∂yβ ◦ F1, we have
k(vαν(z))−(wα(z))kC0;Kσ →0,
k(vαν(z))−(wα(z))kC2;K →0, asν(orρ)→0, on any compactK⊂B\{0}.
C-III)The convergence of derivatives. TakingKσ as above, we denotefν|Kσ and F1|Kσ byfν andF1, respectively. Note that expF1 : U(0) → H1,2∩C0(Kσ, N)is a diffeomorphism on some neighbourhoodU(0)∈ H1,2∩C0(Kσ,(F1)∗T N), because d(expF1)0 = Id. Moreover, kfν − F1|KσkH1,2∩C0 → 0 as ν → 0, so there exists ξν ∈H1,2∩C0(Kσ,(F1)∗T N)for smallν >0 withexpF1ξν =fν.
The mappingξ7→ dexpF1,ξ depends smoothly onξν ∈TF1H1,2∩C0(Kσ, N), so dexpF1,ξν →IdinH1,2∩C0(Kσ), sinceξν→0 inH1,2∩C0(Kσ,(F1)∗T N)asν →0.
ForWν := wαν∂y∂α ◦ F1 :=dexp−1F1,ξ
ν(Vν)we havekwνα(z)−wα(z)kC0;Kσ → 0 by C-II). Moreover,dF1→dfνinL2, thusR
Kσ|dexpF1,ξν(dF1)−dfν|2dω→0.
We next observe, for∇F1Wν= (wαν,i+wνγ(F1)β,iΓαβγ(F1))dzi⊗∂y∂α◦ F1, that Z
Kσ
|dexpF1,ξν(∇F1Wν)− ∇fνVν|2dω→0 asν →0, (15) sincekF1−fνk1,2;0→0,dexpF1,ξν →IdinC0,∂i(dexpF1,ξν)→∂i(Id) =0 inL2.
Thus, for Xν, Yν ∈ H1,2 ∩ C0(Kσ, T∗M ⊗ fν∗T N) with R
Kσ|Xν|2dω → 0, R
Kσ|Yν|2dω→0,
dexpF1,ξν(dF1) =dfν+Xν, dexpF1,ξν(∇F1Wν) =∇fνVν+Yν.