Regularity of n/2-harmonic maps into spheresVon der Fakult at fur Mathematik, Informatik und Naturwissenschaftender Rheinisch{Westf alischen Technischen Hochschule Aachenzur Erlangung des akademischen Grades eines Doktors der Naturwissenschaften genehmigteDissertationvorgelegt vonDiplom{MathematikerArmin Schikorraaus Altenkirchen (Westerwald)Berichter: Prof. Dr. Heiko von der Mosel (RWTH Aachen)Prof. Dr. Pawe l Strzelecki (Uniwersytet Warszawski)Tag der mundlic hen Prufung: 22. September 2010Diese Dissertation ist auf den Internetseiten der Hochschulbibliothek online verfugbar.AbstractIn the present thesis, we consider critical points of the functional n 2 4E (v) := vnnRn nn n n2 2evaluated at points v2 H (R ) with the constraintjvj = 1 on some domain DR . Here, H (R ) denotes2 nthe fractional Sobolev space of all functions v2L (R ) such thatn ^ 2 n2j j v ()2L (R );^where () is the Fourier transform. We extend earlier results for evenn andn = 1 to arbitrary dimensionn2Nby proving H older continuity of critical points u of E .nAs for the proof, we adapt an approach by L. Tartar, which was used originally to prove Wente’s inequality, inorder to ensure the existence of compensation phenoma appearing in the Euler-Lagrange equations.In order to localize this e ect, we establish several localization results for nonlocal operators comparable to thefractional laplacian.
VonderFakulta¨tfu¨rMathematik,InformatikundNaturwissenschaften derRheinisch–Westfa¨lischenTechnischenHochschuleAachen zur Erlangung des akademischen Grades eines Doktors der Naturwissenschaften genehmigte Dissertation
vorgelegt von
Diplom–Mathematiker Armin Schikorra aus Altenkirchen (Westerwald)
Berichter: Prof. Dr. Heiko von der Mosel (RWTH Aachen) Prof.Dr.PawelStrzelecki(UniwersytetWarszawski)
Abstract In the present thesis, we consider critical points of the functional En(v) :=ˆΔ4v n2 Rn evaluated at pointsv∈Hn2(Rn) with the constraint|v|= 1 on some domainD⊂Rn. Here,Hn2(Rn) denotes the fractional Sobolev space of all functionsv∈L2(Rn) such that |∙|n2v∧(∙)∈L2(Rn), where (∙)∧ extend earlier results for evenis the Fourier transform. Wenandn= 1 to arbitrary dimensionn∈N byprovingHo¨ldercontinuityofcriticalpointsuofEn. As for the proof, we adapt an approach by L. Tartar, which was used originally to prove Wente’s inequality, in order to ensure the existence of compensation phenoma appearing in the Euler-Lagrange equations. In order to localize this effect, we establish several localization results for nonlocal operators comparable to the fractional laplacian. Finally, in order to prove that the Euler-Lagrange equations govern the growth of Δ4nuwe develop a fractional version of Hodge decomposition with local estimates similar to Cauchy estimates. Then, a fractional yet localized version of the Dirichlet growth theorem which we establish in the appendix impliesH¨oldercontinuity.
1 Introduction In his seminal work [elH´90]F´e.Hko-dihetwromtapsfdtsiulinoianemsnlagureedovprinlemcinomrahrofytir B1(0)⊂R2into them-dimensional sphereSm−1⊂Rmfor arbitrarym∈N. These maps are critical points of the functional E2(u) :=ˆ|ru|2,whereu∈W1,2(B1(0),Sm−1). B1(0)⊂R2 The importance of this result is the fact that harmonic maps in two dimensions are special cases of critical points of conformally invariant variational functionals, which play an important role in physics and geometry and have been studied for a long time H´lein’s approach is based on the discovery of a compensation phenomenon : e appearing in the Euler-Lagrange equations ofE2, using a relation between div-curl expressions and the Hardy space.ThiskindofrelationhadbeendiscoveredshortlybeforeinthespecialcaseofdeterminantsbyS.Mu¨ller [l90M¨u] and was generalized by R. Coifman, P.L. Lions, Y. Meyer and S. Semmes [CLMS93hdsineedentxleie].H´ result to the case where the sphereSm−1general target manifold developing the so-called moving-is replaced by a frame technique which is used in order to enforce the compensation phenomenon in the Euler-Lagrange equations [H´el91inally,T].F[eR.vi`ireRiv07] was able to prove regularity for critical points of general conformally invariant functionals, thus solving a conjecture by S. Hildebrandt [Hil82]. He used an ingenious approach based on K. Uhlenbeck’s results in gauge theory [Uhl82] in order to implement div-curl expressions in the Euler-Lagrange equations,atechniquewhichcanbereinterpretedasanextensionofH´elein’smovingframemethod;see[Sch10]. FormoredetailsandreferenceswerefertoHe´lein’sbook[2l0´eH] and the extensive introduction in [Riv07] as well as [Riv09]. Naturally, it is interesting to see how these results extend to other dimensions: In the four-dimensional case, regularity can be proven for critical points of the following functional, the so-called extrinsic biharmonic maps: E4(u) :=ˆ|Δu|2,whereu∈W2,2(B1(0),Rm). B1(0)⊂R4 This was done by A. Chang, L. Wang, and P. Yang [CWY99a sphere as the target manifold,] in the case of and for more general targets by P. Strzelecki [Str03], C. Wang [Wan04] and C. Scheven [Sch08]; see also T. LammandT.Rivi`ere’spaper[LR08 generally, for all even]. Moren≥similar regularity results hold, and we6 refer to the work of A. Gastel and C. Scheven [GS09] as well as the article of P. Goldstein, P. Strzelecki and A. Zatorska-Goldstein [GSZG09]. In odd dimensions non-local operators appear, and only two results for dimensionn= 1 are available. In [DLR09aLioandT],F.Dnocredlofytiuniteri`iv.RH¨veroepfenufohtanltcoiiticorcrintsalpo EˆΔ14u2ributionsuwith finite energy andu∈Sm−1a.e. 1(u) =,defined on dist R1 In [DLR10] this is extended to the setting of general target manifolds. In general, we consider forn, m∈Nand some domainD⊂Rnthe regularity of critical points onDof the functional En(v) =ˆΔn4v2, v∈H2n(Rn,Rm), v∈Sm−1a.e. inD. (1.1) Rn Here, Δ4ndenotes the operator which acts on functionsv∈L2(Rn) according to n Δn4v∧(ξ) =|ξ|2v∧(ξ),for almost everyξ∈Rn, where ()∧denotes the application of the Fourier transform. The spaceHn2(Rn) is the space of all functions v∈L2(Rn) such that Δn4v∈L2(Rn us remark that the interest in energies of the type (). Let1.1) is not only motivated by above mentioned purely theoretic considerations. In fact, energies like that appear in physical models of, e.g., magnetostatic energies (cf. [KMM06 term “critical point” is defined as usual:]). The Definition 1.1(Critical Point).Letu∈Hn2(Rn,Rm),D⊂Rn. We say thatuis a critical point ofEn(∙)on Difu(x)∈Sm−1for almost everyx∈Dand d dtt0E(ut,ϕ) = 0 =
6
for anyϕ∈C0∞(D,Rm)whereut,ϕ∈Hn2(Rn)is defined as ut,ϕ=(uΠ(u+tϕ)ininRD,n\D. Here,Πdenotes the orthogonal projection from a tubular neighborhood ofSm−1intoSm−1defined asΠ(x) =|xx|. Ifnis an even number, the domain ofEn(∙) is just the classic Sobolev spaceHn2(Rn)≡Wn2,2(Rn), for odd dimensions this is a fractional Sobolev space (see Section2.3 in). FunctionsHn2(Rn) can contain logarithmic singularities (cf. [Fre73]) but this space embeds continuously intoBM O(Rn), and even only slightly improved integrability or more differentiability would imply continuity. In the light of the existing results in even dimensions and in the one-dimensional case, one may expect that similar regularity results should hold for any dimension. As a first step in that direction, we establish regularity ofn/2-harmonic maps into the sphere. Theorem 1.2.For anyn≥1, critical pointsu∈H2n(Rn)ofEnon a domainDllyH¨oldarelocantinercouous inD. Note that here – in contrast to [DLR09] – we work on general domainsD⊆Rn. This is motivated by the factsthatH¨oldercontinuityisalocalpropertyandthatΔn4(though it is a non-local operator) still behaves “pseudo-local”: We impose our conditions (here: being a critical point and mapping into the sphere) only in some domainD⊂Rn, and still get interior regularity withinD. Let us comment on the strategy of the proof. As said before, in all even dimensions the key tool for proving regularity is the discovery ofcompensation phenomenabuilt into the respective Euler-Lagrange equation. For example, critical pointsu∈W1,2(D,Sm−1) ofE2satisfy the following Euler-Lagrange equation [le09´H] Δui=ui|ru|2,weakly inD, for alli= 1. . . m. (1.2) For mappingsu∈W1,2(R2,Sm−1) this is a critical equation, as the right-hand side seems to lie only inL1: If we had no additional information, it would seem as if the equation admitted a logarithmic singularity (for examples see, e.g., [Riv07], [Fre73]). But, using the constraint|u| ≡1, one can rewrite the right-hand side of (1.2) as m m ui|ru|2=Xuiruj−ujrui∙ ruj=X∂1Bij∂2uj−∂2Bij∂1uj j=1j=1 where theBijare chosen such that∂1Bij=ui∂2uj−uj∂2ui, and−∂2Bij=ui∂1uj−uj∂1ui, a choice which is possibleduetoPoincar´e’sLemmaandbecause(1.2) implies divuiruj−ujrui= 0 for everyi, j= 1. . . m. Thus, (1.2) transforms into m Δui=X∂1Bij∂2uj−∂2Bij∂1uj,(1.3) j=1 a form whose right-hand side exhibits a compensation phenomenon which in a similar way already appeared in the so-called Wente inequality [Wen69], see also [BC84], [Tar85]. In fact, the right-hand side belongs to the Hardy space (cf. [9lu¨0M], [CLMS93]) which is a proper subspace ofL1with enhanced potential theoretic properties.Namely,membersoftheHardyspacebehavewellwithCalder´on-Zygmundoperators,andbythis one can conclude continuity ofu. An alternative and for our purpose more viable way to describe this can be found in L. Tartar’s proof [Tar85] of Wente’s inequality: Assume we have fora, b∈L2(R2) a solutionw∈H1(R2) of Δw=∂1a ∂2b−∂2a ∂1bweakly inR2.(1.4) Taking the Fourier-Transform on both sides, this is (formally) equivalent to |ξ|2w∧(ξ) =cˆa∧(x)b∧(ξ−x) (x1(ξ2−x2)−x2(ξ1−x1))dx,forξ∈R2.(1.5) R2 Now the compensation phenomena responsible for the higher regularity ofwcan be identified with the following inequality: 1 1 |x1(ξ2−x2)−x2(ξ1−x1)| ≤ |ξ||x|2|ξ−x|2.(1.6)
7
1 Introduction
Observe, that|x|as well as|ξ−x|appear to the power 1/ these factors as Fourier multi- Interpreting2, only. pliers, this means that only “half of the gradient”, more precisely Δ14, ofaandbenters the equation, which implies that the right-hand side is a “product of lower order” operators. In fact, plugging (1.6) into (1.5), one can concludew∧∈L1(R2dYanngould¨o’serytiloLnonis’auqeces–consrentzspanohesaqeeutnylbtHyj)su proven continuity ofw, because the inverse Fourier transform mapsL1intoC0. As explained earlier, (1.2) can be rewritten as (1.3) which has the form of (1.4), thus we have continuity for critical points ofE2, and by a bootstraping argument (see [Tom69]) one gets analyticity of these points. As in Theorem1.2we prove only interior regularity, it is natural to work with localized Euler-Lagrange equations which look as follows, see Section7: Lemma 1.3(Euler-Lagrange Equations).Letu∈H2n(Rn)be a critical point ofEnon a domainD⊂Rn. ˜ Then, for any cutoff functionη∈C0∞(D),η≡1on an open neighborhood of a ballD⊂Dandw:=ηu, we have −ˆwiΔn4wjΔ4nψij=ˆΔ4nwjH(wi, ψij)−ˆaijψij,for anyψij=−ψji∈C0∞(D)˜,(1.7) RnRnRn whereaij∈L2(Rn),i, j= 1, . . . , m, depend on the choice ofη. Here, we adopt Einstein’s summation convention. Moreover,H(∙,∙)is defined onHn2(Rn)×Hn2(Rn)as H(a, b) := Δn4(ab)−aΔn4b−bΔ4na,fora, b∈H2n(Rn).(1.8) Furthermore,u∈Sm−1onDimplies the following structure equation wi∙Δn4wi=−12H(wi, wi)1+Δ24nη2a.e. inRn.(1.9) Similar in its spirit to [DLR09] we use that (1.7) and (1.9) together control the full growth of Δn4w, though here we use a different argument applying an analogue of Hodge decomposition to show this, see below. Note moreover that as we have localized our Euler-Lagrange equation, we do not need further rewriting of the structure condition (1.9) as was done in [DLR09]. Whereas in (1.4) the compensation phenomenon stems from the structure of the right-hand side, here it comes from the leading order termH(∙,∙) appearing in (1.7) and (1.9be proved by Tartar’s approach [ can ). ThisTar85], using essentially only the following elementary “compensation inequality” similar in its spirit to (1.6) ||x−ξ|p− |ξ|p− |x|p| ≤Cp(||xx||p2p−|ξ1||2pξ|,+|ξ|p−1|x|,fiifpp∈>01(,,]1(..101) More precisely, we will prove in Section4 Theorem 1.4.ForHas in(1.8)andu, v∈H2n(Rn)one has ∧ kH(u, v)kL2(Rn)≤CkΔ4nu∧kL2(Rn)kΔn4vkL2,∞(Rn). An equivalent compensation phenomenon was observed in the casen= 1 in [DLR09]1 that interpreting. Note again the terms of (1.10) as Fourier multipliers, it seems as if this equation (and as a consequence Theorem1.4) estimates the operatorH(u, v) by products of lower order operators applied touandv by “products of. Here, lower order operators” we mean products of operators whose differential order is strictly between zero andn2and where the two operators together give an operator of order2n fact, this is exactly what happens in special. In cases, e.g. if we take the casen= 4 where Δn4= Δ: H(u, v) = 2ru∙ ruifn= 4. Let us remark, that by an interpolation argument similar results are also a consequence of the “Leibniz rule” for fractional order derivatives obtained by Kato and Ponce, see [KP88] and [Hof98 Another, Corollary 1.2]. case we will need to control is the case whereu=Pis a polynomial of degree less than2n. As (at least formally) Δn4P= 0 this is to estimate Δ4n(P v)−PΔ4nv. 1In fact, all compensation phenomena appearing in [DLR09] can be proven by our adaption of Tartar’s method using simple compensation inequalities, thus avoiding the use of paraproduct arguments (but at the expense of using the theory of Lorentz spaces).