You are on page 1of 22

Invent. math.

168, 1–22 (2007)
DOI: 10.1007/s00222-006-0023-0

Conservation laws for conformally invariant
variational problems
Tristan Rivière
Department of Mathematics, ETH Zentrum, CH-8093 Zürich, Switzerland

Oblatum 15-III-2006 & 9-X-2006
Published online: 7 December 2006 – © Springer-Verlag 2006

Abstract. We succeed in writing 2-dimensional conformally invariant non-
linear elliptic PDE (harmonic map equation, prescribed mean curvature
equations, . . . , etc.) in divergence form. These divergence-free quantities
generalize to target manifolds without symmetries the well known conser-
vation laws for weakly harmonic maps into homogeneous spaces. From
this form we can recover, without the use of moving frame, all the classical
regularity results known for 2-dimensional conformally invariant non-linear
elliptic PDE (see [Hel]). It enables us also to establish new results. In par-
ticular we solve a conjecture by E. Heinz asserting that the solutions to
the prescribed bounded mean curvature equation in arbitrary manifolds are
continuous and we solve a conjecture by S. Hildebrandt [Hil1] claiming that
critical points of continuously differentiable elliptic conformally invariant
Lagrangian in two dimensions are continuous.

I Introduction

The absence of possible applications of the maximum principle to solutions
to non-linear elliptic systems reduces drastically the tools available for an-
swering questions regarding the symmetry, the uniqueness or the regularity
of these solutions. In such an impoverishment of the available technics while
passing from scalar PDE to systems, the search for conservation laws is,
however, one of the remaining relevant strategy to adress these questions.
Weakly harmonic maps into spheres give a good illustration of the efficiency
of conservation laws in this setting. An weakly harmonic map u from the
n-dimensional unit ball B n into the unit sphere Sm−1 of Rm is a W 1,2 (B n , Rm )
map (maps in L 2 whose first derivatives are also in L 2 ) which takes values
almost everywhere in the sphere Sm−1 and which solves the following PDE

−∆u = u |∇u|2 (I.1)

2 T. Rivière

n ∂ 2
where ∆ is the negative laplacian in Rn : ∆ = i=1 ∂x 2 . They are the
critical points of the Dirichlet energy E(u) = B n |∇u|2 dx1 · · · dxn for all
perturbations of the form u t = u − tφ/|u − tφ| where φ is an arbitrary com-
pactly supported smooth map from B n into Rm . Because of the conformal
invariance of the Dirichlet energy E in 2 dimension (E(u ◦ϕ) = E(u) for ar-
bitrary u in W 1,2 (R2 , Rm ) and arbitrary conformal map ϕ from R2 into R2 ),
the harmonic map equation (I.1) is conformally invariant in 2 dimension: if
u is a solution to (I.1) in W 1,2 (B 2, Rm ), the composition with an arbitrary
conformal map ϕ: u ◦ ϕ is again a solution to (I.1). The conformal dimen-
sion 2 is also the critical dimension for (I.1): The left-hand-side of (I.1) for
a W 1,2 solution is in L 1 , therefore a solution has a laplacian in L 1 which
is the borderline case in 2 dimension which “almost” ensures that the first
derivatives are in L 2 (using standard estimates on Riesz potential [Ste]). So,
in some sense, by inserting the W 1,2 bound assumption in the non-linearity
we are almost back on our feet by bootstrapping this regularity information
in the linear part of the equation. None of the two sides, linear and non-
linear, of the equation is really dominant: the equation is critical. Solution
to this system enters in the family of solutions of systems of quadratic
growth and can be discontinuous (see for instance [Gia]). Hence, among the
fundamental analysis issues regarding (I.1) are 1) the regularity of solution
in conformal 2-dimension and 2) the passage to the limit in the equation
for sequences of solutions having bounded E energy. Both questions were
solved by the introduction of the following conservation laws discovered by
J. Shatah ([Sha]): u is a solution to (I.1) in W 1,2 if and only if the following
∀i, j ∈ {1, . . . , m} div(u i ∇u j − u j ∇u i ) = 0. (I.2)
The cancellation of these divergences can be interpreted by the mean of
Noether theorem using the symmetries of the target Sm−1 (see Hélein’s
book [Hel]). Using this form it becomes straightforward to answer to the
question 2) using the compactness of the embedding of W 1,2 into L 2
(Rellich Kondrachov embedding Theorem). The answer to question 1):
the fact that W 1,2 solutions to (I.1) are real analytic was established by
F. Hélein (see [Hel]) starting again from the conservation laws (I.2). The
main step was to prove the continuity of the solution since, by classical re-
sults in [HiW,LaU] and [Mo], continuous solutions
 are real analytic. Using
the conservation laws (I.2) and the fact that j u j ∇u j = 0, F. Hélein wrote
(I.1) in the following way:
−∆u =i
u ∇uj · ∇u =
i j
u i ∇uj − uj ∇u i · ∇u j
j=1 j=1

= ∇ Bji · ∇u ,


2 solving curlBji = u i ∇uj − uj ∇u i is given by the classical theory of elliptic operators. by R.1).2 (B n . by L. Y. are the weakly harmonic maps from B n into N k . It is now natural to try to understand to which extend the above re- sults are still valid when we are considering W 1.2 (B n . Lions. N k be a C 2 k-dimensional submanifold of Rm .1) are smooth in 2 dimension but also permits to establish the estimate  |∇ 2 u| dx1 · · · dxn < +∞. ∂x1 ). N k ) the subset of W 1.e. They are the maps in W 1. This special phenomenon that we recall in the appendix was first observed in a particular case in [We] by H.4) O for any solution to (I. for n arbitrary and where O is an arbitrary open subset with closure in B n .2 maps from B n into Rm which take values in N k almost everywhere. the product curl-grad was observed to be in the local Hardy space Hloc .Conservation laws for conformally invariant variational problems 3 where ∇ ⊥ is the ∇-operator rotated by π/2 (i. Coron in [BrC] extending Wente’s argument and independently. Observing that the non-linearity of the weakly harmonic 1 map equation is in Hloc gives not only another approach to conclude that solutions to (I. The existence of Bji in Wloc1.M. (I. ∇ ⊥ := (−∂x2 .1 which embeds in C 0 in 2 dimension.L. Coifman. N k ) of the Dirichlet energy E(u) = B n |∇u|2 for all perturbations of the form πN (u + tφ). Tartar in [Ta1]. For instance for k = m − 1 when N m−1 is an oriented codimen- . 2) or what about the validity of the estimate (I. This estimate happens to play a crucial role for establishing energy quantization results as described in [LiR].3) has in fact some additional regularity than being simply in L 1 and the inverse by the laplace operator of such a product is continuous. Among the special 1 features of distributions in Hloc is the “nice” behavior of this space with respect to Calderon–Zygmund operators. in particular the inverse of such a distribution by the Laplace operator are in W 2. ∇u) = A(u)(∂xl u. The product curl-grad in the right- hand-side of (I.  The critical points u in W 1.2 (B n . Meyer and S. Denote πN the C 1 orthogonal projection on N k defined in a small tubular neighborhood of N k in Rm which assigns to each point in this neighborhood the nearest point on N. Müller [Mu] where this result was obtained under some sign assumption on the product.5) l=1 where A(u) is the second fundamental form at u(x) for the submanifold N k in Rm . N k ) which solve the following Euler–Lagrange equation (in distributional sense)  n −∆u = A(u)(∇u. We denote by W 1. using a quite different approach. Semmes in [CLMS] following the work of S.2 weakly harmonic maps taking values in an arbitrary submanifold of Rm . Wente and was proved in its full generality by H. Later 1 on. 1 smaller than L . (I. P.4) in this general set- ting? Let then. Brezis and J. ∂xl u) = 0. What about questions 1). where φ is an arbitrary smooth compactly supported map from B n into Rm .

One can even find counterexamples for Ω non antisymmetric and for which u is not in L ∞ : for m = 2 the map u = (u 1 . m −∆u i = mj=1 Ωij · ∇u j .1 Let m ∈ N. .5) becomes −∆u = n ∇n · ∇u. This theorem is optimal because of the following counterexample of Frehse [Fre]: for m = 2 the map u = (u 1 . .4 T. (I.4) 1.9) (u 2 − u 1 )∇u 1 (u 2 − u 1 )∇u 2 but Ω is not antisymmetric and the solution is L ∞ but not continuous.7) for   (u 1 + u 2 )∇u 1 (u 1 + u 2 )∇u 2 Ω= ∈ L 2 D2 . the unit perpendicular vectorfields which generates the orientation of N m−1 . . so(m)⊗R2 ) j (i. u 2 ) from D2 into R2 given by 2 2 u 1 (x) = log log .2 for general W solutions to (I.5). For every Ω = (Ωij )1≤i.7) using coordinates stands for ∀i = 1.e. (I. Ωij ∈ L 2 (D2 . Hélein permitted to avoid the direct conservation law approach for proving questions like the regularity in dimension 2 of the weakly harmonic maps into general target (i. . (I.10) |x| |x| . u 2 ) from D2 into S1 ⊂ R2 defined by 2 2 u 1 (x) = sin log log .e.6).6)) – see again [Hel]. u 2 (x) = cos log log .5)? Can we write the equation (I. m}. This set of questions has motivated the following. if we denote by n(y) the Gauss map of N m−1 at y.6) where we keep denoting n the composition n ◦u. every u ∈ W 1. solutions to (I. it is then natural to look for conservation laws (divergence free quantities) generalizing (I.1) to solutions of (I. .2 solutions to (I.6) (or even 1 (I.5) in the local Hardy space Hloc ? Do we have estimates of the form (I. (I.6) or (I.7) is continuous where the contracted  notation in (I. . j ∈ {1. M2 ⊗ R2 . R2 ) solves (I.5) in divergence form? Until now the answers to these questions were open and only the introduction of the indirect but beautiful technic of moving frame by F. In order to try to extend the above described results established for W 1. which is one of the main result of the present paper: Theorem I. u 2 (x) = log log . Is for instance the non-linearity n ∇n · ∇u in the right-hand-side of (I. Rm ) solving −∆u = Ω · ∇u (I. or even more generally to solutions to (I.2 (D2 . . ∀i.2). R2 ) and Ωij = −Ωi ). (I. j≤m in L 2 (D2 .2 (D2 . Rivière sion 1 submanifold. .8) |x| |x| is in W 1.6) or even (I.

1 can be applied to them.12) j=1 Taking now Ωij := n i ∇n j − n j ∇n i we can apply Theorem I.13) D2 satisfies an equation of the form (I. N k ) of the Lagrangian    F(u) = |∇u|2 + ω(u)(∂x u. Precisely we have Theorem I.2 (D2 . One of the main observation of the present work is that what is important in (I.Conservation laws for conformally invariant variational problems 5 is in W 1. R2 ) and solves (I. Theorem I. ∂ y u). (I. so(m) ⊗ R2 ) and is therefore continuous.6) because of the following observation: every derivative of umsolving (I. Conversely.6) in the form: j  m   −∆u = i n i ∇n j − n j ∇n i · ∇u j . It was proved in [Gr1] that every conformally invari- ant elliptic Lagrangian.3) is not the divergence free structure of n i ∇n j − n j ∇n i .5) but every critical point of any elliptic conformally invariant Lagrangian in dimension 2 can be written in the form ( I. M2 ⊗ R2 . (I. Thus j=1 n j ∇u = 0 and we can rewrite (I. ∂x u. but it is the anti-symmetry of this quantity which is much more robust and which is the key point for the regularity of solution to (I. generates an .7) for   ∇u 1 0 Ω= ∈ L 2 D2 .7) and the regularity result obtained in Theorem I. satisfying some “natural conditions”. It is not difficult to see that the Lagrangian of the form (I. valid in the particular case of the round sphere and which disappears as soon as one perturbs the metric of the target.2 (D2 .13) are conformally invariant.7) for some Ω in L 2 (D2 .1 to get the continuity of u. Let ω be a C 1 2−form on N k such that the L ∞ norm of dω is bounded on N k .11) 0 ∇u 2 but Ω is symmetric and not antisymmetric.6) is tangent to N m−1 and is therefore perpendicular to n.6).2 Let N k be a C 2 submanifold of Rm (k and m being arbitrary integer satisfying 1 ≤ k ≤ m).3) except that in the general case there is no reason for Ωij := n i ∇n j − n j ∇n i to be divergence free.1 applies to (I. This is the so called prescribed mean curvature equation in a manifold N k . This is a new compensation phenomenon that we discovered. not only solutions to (I. In fact we observed that not only solutions to (I. This way of rewriting the equation has to be compared with the particular case (I. Critical points of F which are conformal are immersed discs in N k whose mean curvature in N k at u is given by |∇u|−2 dω(u)(·.6). which goes beyond the curl-grad structures although it is strongly linked to it as we will explain in the paper. Then every critical point in W 1. ∂ y u) dx ∧ dy (I.

so(3)⊗ R2 ).2 and B ∈ W 1. m} dAij − m k=1 Ak Ω j = i k −d ∗ Bji ). Hildebrandt claiming that the critical points of the second order C 1 elliptic conformally invariant lagrangian in 2 dimensions are continuous see [Hil1] and [Hil2] (3.15) ∇ ⊥u2 −∇ ⊥ u 1 0 Equation (I. . . we have: Theorem I. (I.2 solutions to (I.Cho]). Denoting Mm (R) the space of square m × m real matrices. By denoting ∇ ⊥ := (−∂ y .Hei2. It was conjectured by E. Heinz. that the weakest possible assumption H L ∞ (R3 ) < +∞ should suffices to ensure the continuity of W 1. .14) becomes of the form −∆u = Ω · ∇u. In [Hel1] the class of general conformally invariant lagrangian in 2 dimen- sion is analyzed. (I. Mm (R)) ∩ W 1.6 T.2). ∂x ) and introducing ⎛ ⎞ 0 ∇ ⊥u3 −∇ ⊥ u 2 ⎜ ⎟ Ω := H(u) ⎝−∇ ⊥ u 3 0 ∇ ⊥u1 ⎠ (I. A particular case of interest is the case k = m = 3 and N 3 = R3 .14) and that no control of any kind of the differentiability of the prescribed mean curvature H was needed.2 solves a conjecture by S. .1 is based on the discovery of conservation laws generaliz- ing (I.1 to (I.2 (B n . We can then apply Theorem I. so(m) ⊗ ∧1 Rn ) and let A ∈ L ∞ (B n .18) means ∀i. Theorem I.17) . Denote 2dω = H(z)dz 1 dz 2 dz 3 the Euler–Lagrange equation to F in that case is −∆u = −2H(u) ∂x u ∧ ∂ y u.3 Let m ∈ N. Let Ω = (Ωij )1≤i.BeG1.16)  (where explicitly (I. j≤m in L 2 (B n .BeG2.Bet1. Then every solution to (I. Mm (R) ⊗ ∧2 Rn ) satisfying dΩ A := dA − AΩ = −d ∗ B (I. where Ω ∈ L 2 (D2 . Rivière Euler–Lagrange equation corresponding to a prescribed mean curvature equation in a manifold.14) and we have then proved Heinz’s conjecture on prescribed mean curvature equa- tions.Gr2. j ∈ {1.7) on B n satisfies the following conser- vation law d ∗A du + (−1)n−1 (∗B) ∧ du = 0. In fact Theorem I.14) There has been several attempts to prove the continuity of solutions to (I. see [Hei3].15).14) under several assumptions on H like H ∞ + ∇ H ∞ < +∞ (see for instance [Hei1.

with these notations.22) D2 D2 iii) ∇Ω A := ∇ A − AΩ = ∇ ⊥ B.2 (D2 .Conservation laws for conformally invariant variational problems 7 For n = 2. so(m) ⊗ R2 ) satisfying  |Ω|2 ≤ εm . using different notations.2 and B ∈ W 1.19). for every Ω = (Ωij )1≤i. 2 SO(n)) 2∞ ≤ C(n) |Ω|2 . j≤m .2 (D2 . Mm (R)) satisfying i)   −1 2 |∇ A| + |∇ A | + dist(A. Using now the fact that div(Bji ∇ ⊥ u j ) = ∇ Bji · ∇ ⊥ u j = −∇ ⊥ Bji · ∇u j the harmonic map equation into Sm−1 .2 and B ∈ W 1. ∇ ⊥ Bji = u i ∇uj − uj ∇u i . Mm (R)) ∩ W 1. The question remains of finding A and B satisfying (I. the theorem says that given Ω = (Ωij )1≤i. j≤m in L 2 (D2 .20) D2 there exists A ∈ L ∞ (D2 . Glm (R)) ∩ W 1. so(m) ⊗ R2 ). We then have included the classical conservation law (I.18). (I. (I.4 There exists ε(m) > 0 and C(m) such that.23) A corresponding local existence result in higher dimension is still an open problem. We shall prove the following local existence result Theorem I.21) D2 D2 ii)   |∇ B| ≤ C(n) 2 |Ω|2 . A ∈ L ∞ (D2 . (I.7) satisfies the following conservation law div(A∇u + B∇ ⊥u) = 0. (I. (I. (I. is equivalent to (I. j≤m in L 2 (D2 . Mm (R)) satisfying ∇Ω A := ∇ A − AΩ = ∇ ⊥ B.18) Then every solution to (I.19). If the weakly harmonic map we are considering is stationary .19) For instance going back to the symmetric situation of weakly harmonic maps into Sm−1 we take A and B satisfying  j A = idm = δi 1≤i.2) into the larger set of conservation laws of the form (I.

Rivière (see [Hel]). so(m) ⊗ C ⊗ ∧1 R2 ) and let α ∈ L 2 (D2 . there exists a subsequence u n of u n which weakly converges in W 1. under the assumption that Ω M22 is below some positive constant depending only on n and m. so(m) ⊗ ∧1 R2 ) such that Ωn weakly con- verges in L 2 to some Ω.k (C)) such that α = Pβ. In both proofs a Lipschitz bound on the prescribed mean curvature was required.3 in higher di- mension). (I.26) ∂z then there exists P ∈ W 1.8 T.7). Rm ) solving −∆u n = Ωn · ∇u n + f n in D2 .5 where only an L ∞ bound on the prescribed mean curvature is needed. This program is partly realised in a work in preparation of the author together with Michael Struwe.16) becomes crit- ical and the existence of A and B solving (I. (I.16) should be looked for in the space M22 (following the search of a Coulomb gauge in Morrey spaces introduced in [MeR].2 (D2 . one has a replacement of Lemma A.1 can be extended in its spirit to first order elliptic complex valued PDE’s Theorem I. Ω is in the Morrey space given by  1 Ω M22 = supx.25) Then. This is no more the case in Theorem I. SOm (C)) and β ∈ C ∞ (D2 . . j≤m in L 2 (D2 .r n−2 |Ω|2 < +∞.19) we can prove the following result.2 (D2 . Theorem I. the elliptic linear system (I. Passage to the limit in the equation in 2 dimension for the prescribed mean curvature equation or for the harmonic map equation was established in [Bet2] using involved technics. Evans [Ev] for weakly harmonic maps into spheres following the same strategy that Evans introduced. Local existence of conservation law (I. A much simpler proof using moving frames was then given in [FMS].24) r Br (x) Then. (I.19) for stationary weakly har- monic maps permits to extend to general C 2 targets the partial regularity of L.6 Let m ∈ N and k ∈ N. Mm. Let Ω = (Ωij )1≤i. Following similar ideas Theorem I. where SOm (C) is the group of invertible matrices in Glm (C) satisfying P t P = idm . Mm.C. Let f n be a sequence in H −1 (D2 . Using conservation laws (I.2 to a solution of (I.5 Let Ωn ∈ L 2 (D2 .k (C)) solving ∂α = Ω α. Rm ) which converges to 0 in H −1 and u n be a bounded sequence in W 1.

(I.29) where |∇e|2 := |∇e1 |2 + |∇e2 |2 . however. 1 2 ∂x ∂y ∂y ∂x (I. ·) denotes the scalar product in Rm that we also sometime simply denote ·. e2 ) is called a Coulomb moving frame associated to u. e2 ) − sinh a (∇ ⊥ u. 2 D1/2 4π D2 (I.28) div cosh a (∇u.2 .30) B2n (0) . n. ∇e = ∂e1 · ∂e2 − ∂e1 · ∂e2 in D2 . weakly harmonic from D2 into S2 with uniformly bounded W 1.2 weakly harmonic map from D2 into N 2 . This is no more the case for N 2 = S2: one can find sequences of u n . ∇e1 )) = 0 on D2 (see [Hel]) where (·. we can choose e1 such that div((e2 . and for every C > 0 there exists δ(C.7 Let u be a W 1. N 2 ) where N 2 is a closed C 2 oriented 2-dimensional submanifold of Rm . When N 2 is not diffeomorphic to S2 one can estimate D2 |∇e|2 in terms of u W 1. (I. then there exists a map e1 in W 1. denoting e2 (x) the unit vector perpendicular to e1 such that e1 ∧e2 is the unit 2-vector giving the oriented tangent plane Tu(x) N 2 . because of Wente’s Lemma A. Moreover the following estimate holds     1 |∇ u| ≤ C exp 2 |∇e|2 ∇e L 2 (D2 ) + 1 ∇u L 2 (D2 ) . Sm−1 ) such that e1 (x) ∈ Tu(x) N 2 for almost every x in D2 .27) is bounded in L ∞ . Such a pair (e1 .Conservation laws for conformally invariant variational problems 9 Conservation laws and moving frames. e1 ) = 0.2 .2 (D2 . More- over. let a be the function solving ⎧ ⎨−∆a = ∇ ⊥ e .1 that we recall in the appendix. Existence of global conserva- tion laws can be obtained in the same spirit by the mean of moving frames. Considering a map u in W 1.2 norm but for which. Observe that. let (e1 . Nevertheless we still believe that the following holds true: Conjecture For every k ≤ m. e2 ) = 0. N k ) > 0 such that if u is a W 1. every sequence of Coulomb moving frame is not bounded in W 1. for every n ∈ N for every N k . the solution a of (I. e2 ) be a Coulomb moving frame associated to u. e1 ) + sinh a (∇ ⊥ u. k-dimensional closed submanifold of Rm . then the following conservation law holds  div cosh a (∇u.27) ⎩ a=0 on ∂D2 . We have then the following conservation law Theorem I.2 (D2 .2 weakly harmonic map from B2n (0) into N k satisfying  |∇u|2 ≤ C.

Let Φ ∈ L 4 (B n . Rk ⊗ ∧n−1 Rn ) given by Φi (du. Mk (R)) ∩ W 1. e) + (−1)n−1 (∗Ψ) ∧ (du. (e1 (x). The existence of such an estimate for arbitrary C would. j ∈ {1. k} d ∗ (ei . . . (I. Finally. .k and Φ(du.) Denote Ω = (Ωi ) ∈ j L (B . Rivière then  |∇ 2 u| ≤ δ(C. Then the following conservation laws are k satisfied d ∗Φ(du. de j ) = 0. (I.2 solving the linear equation ∗ dΩ Φ + dΩ Ψ = 0. e) and ∗Ψ ∧ (du. .32) becomes critical and . . a closed oriented C 2 k-dimensional submanifold of Rm .8 Let u be a W 1. the following conservation laws general- izing (I. e j ) j and ∗Ψi ∧ (du. under the assumption that Ω M 1 is below some positive constant depend- 2 ing only on n and k.33) generalizes (I. for almost every x. .. . the elliptic linear system (I. dei ). e j )} j=1.. . N k ). e) is the element in L 2 (B n . Mk (R) ⊗ ∧2 Rn ) ∩ W 1. ek ) be a Coulomb moving frame associated to u (the map x → (e1 . . e) = 0.28) to general dimension by taking     1 0 0 −1 Φ = cosh a and Ψ = sinh a dx ∧ dy. Observe that (I. ek ) is in W 1.34) 0 1 1 0 Existence of Coulomb moving frames is discussed in [Hel] and is proved under the assumption that N k is sufficiently regular and modulo some iso- metric embeddings in a submanifold diffeomorphic to a torus. in general dimension. . . e) denote respectively the elements in 4 4 j L 3 (B n . Let (e1 . .33) where (du. ek (x)) is an orthonormal basis of Tu(x) N k and ∀i. permit to extend the quantization result of [LiR] to general targets. so(k) ⊗ ∧ R ) the connection given by 2 n 1 n j Ωi := (e j .10 T.. in particular. . .32) j j ∗ j where (dΩ Φ)i := dΦi + Φki ∧ Ωk and dΩ is the adjoint of dΩ given by ∗ j ∗ j j (dΩ Ψ)i := d Ψi + ∗(∗Ψi ∧ Ωk ).2 . (I. .2 weakly harmonic map from B n into N k .31) B1n (0) Such an estimate is known when C is small enough. (I. . Rk ⊗ ∧1 Rn ) and in L 3 (B n .2 and Ψ ∈ L 4 (B n .28) should play an important role in the theory of weakly harmonic maps: Theorem I. n. . e j ). Rk ⊗∧1 Rn ) given by {(du. Again here.

II Proof of Theorems I. The paper is organised as follows: in Sect. 3 we prove Theorem I. j j j j .2) with ξ = 0 on ∂D2 and such that ξ W 1. (II. ⎪ ⎪ ⎪ ⎪ ⊥ −1 −1 −1 ⎨∆B = ∇  · ∇ P − div( Â∇ξP ) − div(∇ξP ) in D . Acknowledgements.2 ≤ C(m) Ω L 2 . (II. 2 ⎪ ∂  (II. Let ε(m) > 0 given by Lemma A. In Sect.1 to Theorem I. (II. 4 we prove Theorem I.2 + P W 1. 2 and 3. ⎪ ∂ν  ⎪ ⎪ ⎪ ⎪ ⎩ and  = 0 D2 for  ∈ Mm (R) and B ∈ Mm (R). He is also grateful to the referee for his careful reading and his contribution to improve the presentation of the present paper.32) should be looked for in the space M22 which embeds in L 4 in every dimension. This work was carried out while the author was visiting the Université de Bretagne Occidentale at Brest.3. so(m)) given by Lem- ma A.6 II.4.2 (D2 . Let Ω ∈ L 2 (D2 .3) We look for A and B solving (I.1 Proof of Theorem I.5) is made of jacobians: (∇  · ∇ ⊥ ξ)i = ∂ y Âki ∂x ξk − j j ∂x Âki ∂ y ξk and −(∇ ⊥ B ·∇ P)i = ∂x Bik ∂ y Pk −∂ y Bik ∂x Pk . using Lemma A.5) ⎪ ⎪ = 0 and B = 0 on ∂D2 . SO(m)) and ξ ∈ W 1.1. In Sect.1–I.1) D2 Let then P ∈ W 1. so(m) ⊗ ∧1 R2 ) satisfying  |Ω|2 ≤ ε(m). 2 we prove Theorem I. In the appendix we recall Wente’s result and establish several lemmas used in Sects. it means that we are looking for à and B solving ∇ à − ∇ ⊥ BP = Ã∇ ⊥ ξ.18) and introducing à := AP.3 satisfying ∇ ⊥ ξ = P −1 ∇ P + P −1 ΩP in D2 . The author would have like to thank the mathematics department of the UBO for it’s hospitality.5.6. (II.Conservation laws for conformally invariant variational problems 11 the existence of Φ and Ψ solving (I.2 + P −1 W 1.2 (D2 . Observing that the right-hand-side of the first equation of (II.4) First we aim to solve the following system ( à will be chosen to be  + idm later) ⎧ ⎪ ⎪ ∆  = ∇  · ∇ ⊥ ξ + ∇ ⊥ B · ∇ P in D2 .8.7 and Theorem I.

5) and  W 1.3 Proof of Theorem I. From the first equation of (II. using the jacobian structure of the right-hand-sides of the equations in (II.2 (D2 ) such that A ∇u = ∇ ⊥ E + ∇ D. Therefore ∇u = A−1 ∇ ⊥ E + A−1 ∇ D is in W on this disk.9) Moreover. (II.1 ≤ C Ω L 2 ). . using a standard fixed point argument.2  W 1.2 .2 B W 1. Theorem I.1 into C 0 in 2 dimension we 1.2 +  L ∞ + B W 1.2 .1 get the desired result and Theorem I. we even have  W 2. we have the a-priori estimates  W 1.1 2 on the disk of half radius D1/2 . C satisfies  div(∇C P −1 ) = 0 in D2 . we obtain the existence of  and B satisfying (II. (I.7) Thus for Ω L 2 small enough. II. (II. up to addition of a constant. (II.11) curl(A ∇u) = ∇ ⊥ A · ∇u. Using the embedding of W 2.2 + C ξ W 1.23).2 + C P W 1. using the result of [CLMS]. II.4. (II. (II. Moreover.8) (Observe that.2. (II.2 Proof of Theorem I.1. (II. the results in [CLMS] imply that E and D are in W 2.22) and (I. Let now à :=  + idm .12) Moreover. From Theo- rem I.2 +  L ∞ ≤ C ξ W 1. Using standard elliptic estimates.2  L ∞ + C ξ W 1. let A and B given by Theorem I.2 ≤ C Ω L 2 .10) C=0 on ∂D2 . we can always assume that D2 |Ω|2 ≤ ε(m) where ε(m) is given by Theorem I.1 is proved.4 is then proved.2 ≤ C P −1 W 1.2 satisfying ∇ à − à ∇ ⊥ ξ − ∇ ⊥ BP = ∇ ⊥ C.2  W 1.5) we obtain the existence of C in W 1.3 follows from a direct computa- tion.6) B W 1.11). we obtain that C is identically zero and A := à P −1 and B satisfy (I.4 they solve the following system  div(A ∇u) = −∇ B · ∇ ⊥ u. Using now Lemma A.4. we get the existence of E and D in W 1.12 T. Rivière standard elliptic estimates and the fact that P ∈ SO(m). Since  the desired result is local (continuity of u).21).3. Theorem I.

Let πN be the orthogonal projection on N k defined in a small tubular neighborhood of N k .l ) j=1.l ∇u l + λ(u)ij. Since (Ai.n r x. Following [Hel] Chap. .m = A(εi . so(m) ⊗ ∧1 R2 ) converging weakly to some Ω and f n and u n respectively converging to zero in H −1 (D2 . Theorem I.. so(m) ⊗ ∧1 R2 ).l ∇u j · ∇u l + λ(u)ij.l .. We can always assume that u n converges weakly to some u in W 1.5.. . Let N k be a k-dimensional submanifold of Rm . modulo extraction of a subsequence. 4. .l ∇ ⊥ u l · ∇u j = 0.Conservation laws for conformally invariant variational problems 13 II. N k ) satisfy the following Euler–Lagrange equation ∆u i + Ai (u) j. l ∈ {1. We extract a Vitali covering from it which ensures that every point in Bλ2(0) is covered by a number of balls bounded by a universal number. 4 j Introducing Ω := (Ωi )i. (II. Rm ) and bounded in W 1.2 (D2 .l = −λ(u)i. (II. .2.. the number of balls in each such a Vitali covering for each n is also uniformly bounded and.13) where λ(u)ij. Since D2 |Ωn |2 is uniformly bounded.4 Proof of Theorem I.2 is then proved. critical points of (I.l ∇ ⊥ u l .m is the canonical ba- j j sis of Rm ..2 critical points of lagrangian of the form (I.l − A j (u)i.13) in W 1. εl ) where (εl )l=1. j where   1 j  Ωij := Ai (u) j.16) 4 we have succeeded in writing W 1.4.l := d ω̃u (εi .n = 1 − |x| in case B1−|x| (x) |Ωn |2 < ε(m).. (II. Let ω be a 2-form on N k and let ω̃ be the pull back of ω by πN in this small tubular neighborhood: ω̃ := πN∗ ω. To every x in Bλ2(0) we assign r x. Let Ωn ∈ L 2 (D2 .13) becomes   ∆u i + Ai (u) j. Rm ). in particular.l ∇u j = 0. we have that  j ∀i. {Brx (x)} for every x in Bλ2 (0) realizes of course a covering of Bλ2(0). .2 (D2 . Rm ).15) + λ(u)ij. . m} Ai.l ∇u l · ∇u j 1 j  (II.7) for some Ω ∈ L 2 (D2 . εl ) is perpendicular to Tu N k for every i and l.l − A j (u)i.13) as solutions to PDE of the form (I.n ≤ 1 − |x| such that Br (x) |Ωn |2 = ε(m) or  x.l − λ(u)i.l ∂x u j ∂ y u l = 0. Let λ < 1 and let ε(m) given by Theorem I. Thus.l − λ(u)i.2 (D2 . we can assume that it . λ(u)ij. III Conservation laws and passage to the limit in PDEs The goal of this section is to prove Theorem I.14) j Thus finally (II. ε j .

n ∇u n + Bi.n satisfy ∇ Ai. Since this holds for every λ < 1 we have proved Theorem I.n (xi.n converges to a non negative number ri (which could be zero of course). Because of the weak convergences in W 1.4 in Bri. For convexity reason. Bi ) in every Bri (xi ).n Ωi.8) implies that ∆u + Ω · ∇u = 0 in Bri (xi ).1 or 2 points).n converges in B λ (0) to a limit xi and that each sequence ri.2 we have strong convergences in L 2 and then we have that Ai. We claim that equation (I.5) Combining then (III. . 2 different circles can intersect at only finitely many points (0. It seats then on the circle. Rivière is fixed and equal to N independent of n.2) We can extract a subsequence such that each of the couples (Ai. (III.7) is satisfied on each Bri (xi ).n ) for Ωn .4) and −Ai.N be this covering.7) Combining (III. it has to seat at the boundary of at least 2 different circles.n . boundary of one of the Bri (xi ).1).n f n → 0 in D  .n − Ai.n (xi.5) we obtain that div Ai ∇u + Bi ∇ ⊥ u = 0 in Bri (xi ). (III.7) we then have that Ai [∆u + Ω · ∇u] = 0 in Bri (xi ).n ∇ ⊥ u n −→ Ai ∇u + Bi ∇ ⊥ u in D  . (III. Let Ai.21).n . since there are finitely many circles.5.6) and that ∇ Ai − Ai Ω = ∇ ⊥ Bi in Bri (xi ).n and Bi.n −→ ∇ Ai − Ai Ω − ∇ ⊥ Bi in D  .. (III. We then have div Ai. .. Modulo extraction of a subsequence we can always assume that 2 each sequence xi. Since ∆u + Ω · ∇u ∈ H −1 + L 1 it is identically zero on Bλ2(0).1) where Ai.n given by Theorem I.n ) weakly converge in W 1.n − ∇ ⊥ Bi.n and Bi.3) ∇ Ai. . Thus the distribution ∆u + Ω · ∇u is supported at at mostly finitely many points.n f n in Bri.9) It is clear that every point in Bλ2(0) is in the closure of the union of the Bri (xi ). .8) From (I. (III. Let x be a point which is none of the Bri (xi ). because of the pointwise convergence of Ai.n ).n Ωi.n = ∇ ⊥ Bi. only finitely many points in Bλ2(0) can be outside the union of the Bri (xi ).6) and (III. (III.n (xi.n ∇u n + Bi. (III.n ∇ ⊥ u n = −Ai.14 T. Let {Bri. (III.. (III. Bi.2 to some limit (Ai .(III.n − Ai. . we get the invertibility of Ai and (III. .n )}i=1.n .

1 (D2 ) ≤ C e 1/2 ∇e L 2 ∇u L 2 + e a ∞ ∇a L 2 ∇u L 2 + e a ∞ ∇u L 2 .2) moreover E W 2. We rewrite (I.1 (D1/2 2 .4) Applying Lemma A. It remains then to prove (I.7 is proved. there exist E ∈ W 2. .1 (D2 ) ≤ C ∇(sinh ae) L 2 ∇u L 2 1/2 + ∇(cosh ae) L 2 ∇u L 2 (IV.2 Proof of Theorem I.7) we get (I. Theorem I.8 follows from a direct computa- tion.29).4).8. R2 ) and D ∈ W 2.28) is the result of a direct computation. ei )) = (−1)i (∇(sinh a ei+1 ). (IV. (IV.5) 2π and 1 ∇a L 2 (D2 ) ≤ √ ∇e1 L 2 ∇e2 L 2 .27) we have 1 a L ∞ (D2) ≤ ∇e1 L 2 ∇e2 L 2 . (IV.1 (with the optimal constants given in [Hel]) to (I.5) and (IV. (IV. IV.7. R2 ) such that cosh a(∇u. (IV. First (I. ∇ ⊥ u).1 (D1/2 2 ) + D W 2.7) Combining then (IV. Thus we have a ∞ E W 2. granting the fact that (∆u. e) = 0. ei )) = (∇u.1 Proof of Theorem I.6) 2π We rewrite (IV. ei ) = ∇ E i + ∇ ⊥ Di . (IV. e) L 2 .1.2) in the form ∇u = (cosh a)−1 ∇ E j e j + ∇ ⊥ D j e j .1 (D1/2 2 ) + D W 2.28) in the form (using Z2 indexation)  div(cosh a(∇u. ∇ ⊥ (cosh a ei )). and standard elliptic estimate.3) + cosh a(∇u.1 (D1/2 2 . (IV. Using then Lemma A.29) and Theorem I.Conservation laws for conformally invariant variational problems 15 IV Conservation laws and moving frames IV.1) curl(cosh a(∇u.

Thus C and D satisfy respectively  ∆C = ∇ ⊥ D · ∇ P in D2 .Ta1.2 From (A. (A. ∂x ∂y ∂y ∂x (A. Mm (R)) such that ∇ ⊥ D = ∇C P −1 and we can choose D such that D2 D = 0.2 (D2 .1 [We. for ε small enough. R) such that ∇a and ∇b are in L 2 (D2 ). Proof of Lemma A. ⎩ ∂D = 0 (A. C ≡ 0 is the unique solution in W 1.7) 1 ∇ D L 2 (D2 ) ≤ ∇C L 2 (D2 ) .4) there exists D ∈ W 1. .CLMS] Let a and b in L 1 (D2 . Let ϕ be the solution of ⎧ ⎪ ∂a ∂b ∂a ∂b ⎪ ⎨∆ϕ = − in D2 .16 T. Rivière A Appendix Lemma A.2 (D2 .2)  where we choose D2 ϕ = 0 for the Neumann boundary data.3) D2 then.6) on ∂D2 .BrC.2 is proved. (A.2 There exists ε > 0 such that for every P ∈ W 1.3).1 and (A. Gl(m)) satisfying  |∇ P|2 + |∇ P −1 |2 ≤ ε.4) C=0 on ∂D2 .2(D2 . Mm (R)) of the following problem  div(∇C P −1 ) = 0 in D2 . and ⎧ ⎨∆D = −∇ ⊥ C · ∇ P −1 in D2 . ∂ν Thus. we have 1 ∇C L 2 (D2 ) ≤ ∇ D L 2 (D2) and 2 (A. using Lemma A. Lemma A.1) ⎪ ⎪ ∂ϕ ⎩ϕ = 0 or = 0 on ∂D . (A.5) C=0 on ∂D2 . 2 ∂ν Then the following estimates hold ϕ L ∞ (D2) + ∇ϕ L 2 (D2) + ∇ 2 ϕ L 1 (D2 ) ≤ ∇a L 2 (D2) ∇b L 2 (D2 ) . 2 which implies that C ≡ 0 and D ≡ 0 and Lemma A. (A.

(A. .2 (D2 ) ≤ C(m) Ω W 1.2 (D2 . .2 (D2) + P W 2. Weak convergence in W 1.Conservation laws for conformally invariant variational problems 17 Lemma A.3 we follow the strategy of [Uhl] and we first prove the following result.8).4 ⇒ Lemma A.16) for Ωk .4 There exist ε(m) > 0 and C(m) > 0 such that for every Ω in W 1. so(m) ⊗ ∧1 R2 ) satisfying  |Ω|2 ≤ ε(m). (A. so(m)) and P ∈ W 2.9) ii) ξ=0 on ∂D2 . Let ξk and Pk satisfying (A. so(m) ⊗ ∧1 R2 ) satisfying  |Ω|2 ≤ ε(m).2 (D2 . so(m) ⊗ ∧1 R2 ) converging strongly in L 2 to Ω. so(m) ⊗ ∧1 R2 ) satisfying (A.2 (D2) . (A.2 (D2 ) ≤ C(m) Ω L 2 (D2) .13) ii) ξ=0 on ∂D2 . (A. (A. (A.2 (D2 ) ≤ C(m) Ω L 2 (D2) . SO(m)) such that i) ∇ ⊥ ξ = P −1 ∇ P + P −1 ΩP in D2 .16) Proof of Lemma A. this equation passes to the limit .12) D2 then there exist ξ ∈ W 2. .10) iii) ξ W 1. (A.11) In order to prove Lemma A. (A. (A. Because of (A.2 (D2 .2 (D2) + P W 1.2 to ξ and P.2 (D2 .15) iv) ξ W 2.2 (D2) + P W 1. Since Pkt Pk = idm .2 (D2 .15) there exists a subsequence ξk and Pk converging weakly in W 1. (A.8) D2 then there exist ξ ∈ W 1.3 There exist ε(m) > 0 and C(m) > 0 such that for every Ω in L 2 (D2 .3 Let Ω in L 2 (D2 . Lemma A. SO(m)) such that i) ∇ ⊥ ξ = P −1 ∇ P + P −1 ΩP in D2 .14) iii) ξ W 1.2 implies almost everywhere convergence of Pk to P.2 (D2 . so(m)) and P ∈ W 1.13). Introduce Ωk in W 1. .

so(m) ⊗ ∧1 R) satisfying λ W 1.18 T. there exists ξλ ∈ W 2. .2 (D2 .2 (D2 . SO(m)) and ν ∈ W 2. Therefore. (A.2 (D2 . by lower-semicontinuity of the W 1.2 2 1 2 |Ω| ≤ ε.C satisfying D2 |Ω|2 < ε.3 is proved.C there exists a neighborhood of Ω in W 1.19) From (A. for any Ω in Uε.17) The previous argument can be adapted to show that Uε.11) and Lemma A.2 (D2 .2 included in Uε.C is closed. SO(m)) satisfying  ∇ ⊥ ξλ = Q −1 −1 ⊥ λ ∇ Q λ + Q λ (∇ ξ + λ)Q λ in D2 . Therefore (A. for every λ ∈ W 1.C = and for which there exists ξ ∈ W 2. (A. (A.  Proof of the claim Let Ω ∈ Uε.2 embeds in W 1.8]. there exists R ∈ W 2. ⎪ 2 ⎪ ⎪ ⎨ ⎪ ⎬ D2 Uε.2 convergence. .2 (D2 .2 ≤ α. .20) Since P ∈ W 2. so(m) ⊗ ∧1 R) satisfying ζ W 1. Let ξ and P satisfying (A. and Q λ − Idm W 2.18) ξλ = 0 on ∂D2 .13) and (A. for every α > 0 we can find δ > 0 such that. for every β > 0.9) is satisfied at the limit. .2 (D2 . Finally.2 + ξλ − ξ W 2.2 ≤ η.13). We now establish the following assertion: Claim For any fixed C there exists ε small enough. .2 (D2 . SO(m)) solving (A.16) for Ω. Pk converges strongly in every L q (q < +∞) and Pkt Ωk Pk respectively Pkt ∇ Pk converge in distribution sense to P t ΩP and respec- tively P t ∇ P. so(m)). .2 norm with respect to the weak W 1.2 into W 1.4 in dimension 2 and that P ∈ SO(m)).21) ν=0 on ∂D2 . Proof of Lemma A. (A.4 We follow the strategy in [Uhl]. .7 and 2. ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ ⎪ ⎭ and P ∈ W 2.10) is also satisfied. Rivière and we have that P ∈ W 1. Moreover. . there exists η > 0 such that for every ζ ∈ W 1.2 (D2 . Lemmas 2. (A. such that.18) we then have ∇ ⊥ ξλ = (PQ λ )−1 ∇(PQ λ ) + (PQ λ )−1 (Ω + PλP −1 )PQ λ . (A. . so(m) ⊗ ∧ R ) satisfying 1.2 (D2 . By continuity of the trace (A.2 (using the fact that W 2.C . We introduce the set ⎧  ⎫ ⎪ ⎪ Ω ∈ W (D . SO(m)).16) (A. Mm (R)) such that  ∇ ⊥ ν = R−1 ∇ R + R−1 (Ω + ζ)R in D2 .2 the map λ → PλP −1 and its inverse ζ → P −1 ζP are continuous from W 1. . Following the arguments in [Uhl.2 ≤ δ.13). we also obtain (A. from Rellich compact embedding. so(m)) and Q λ ∈ W 2.

27) and (A. providing that ε has been chosen small enough.16).15) and (A.15) is satisfied for Ω ∈ Uε.14) for some Ω ∈ W 1. (A.5 There exist C(m) > 0 ε and δ > 0 such that for every P ∈ W 2.1 and standard elliptic PDE we have ∇ξ L 2 ≤ C ∇ P t L 2 ∇ P L 2 + C Ω L 2 .13) we have that ∇ P L 2 ≤ 2 ∇ξ L 2 + 2 Ω L 2 .2 (A.16).2 ≤ 2C δ ∇ξ L 2 + (C + 2C δ) Ω L 2 .29) Choosing then 2C δ < 1/2 we obtain inequality (A.C ).2 + ξ W 1.2 < β. (A.27) From (A.2 + ν − ξ W 2.23) In order to complete the proof of claim it remains to establish (A.25) ξ=0 on ∂D2 .26) Using the hypothesis that ∇ P L 2 ≤ δ we have that ∇ξ L 2 ≤ C δ ∇ P L 2 + C Ω L 2 . if P W 1.1 into L 2 in 2 dimensions. (A.2 + ν W 1. . It remains to establish (A.2 ≤ Cε 2 ((A. SO(m)) and ξ ∈ W 2. Proof of Lemma A. using standard elliptic estimates and the embedding of W 1.2 ≤ (C + 1)ε 2 .14) imply that ξ solves the following elliptic PDE  −∆ξ = ∇ P t · ∇ ⊥ P + div(P t ΩP) in D2 . (A.5 We first establish the critical estimate (A. (A.25) again.28) we have then ∇ξ W 1.13) and (A. (A.2 ≤ C ∇ P t · ∇ ⊥ P W 1. since P W 1.Conservation laws for conformally invariant variational problems 19 Moreover we have R − P W 2.2 ≤ δ.2 (D2 . This will be a consequence of the following lemma. (A. (A.2 (D2 . (A.24) then (A.30) + C ∇ P t Ω L 2 + C Ω∇ P L 2 . so(m)) satisfying D2 |Ω|2 ≤ ε. Lemma A.28) Combining (A.13) and (A. so(m))  satisfying (A. we have 1 R W 1.16) are satisfied.15) and (A.2 + ξ W 1.15). we have ξ W 2.15). From (A.22) 1 1 Considering now β < ε 2 .2 (D2 . Using Lemma A.1 + C Ω W 1.

C and Lemma A. Rassias. River Edge.C . Ghidaglia. Equ. 441–474 (1993) [BeG2] Bethuel.31) and (A.2 P W 1. R. for C(δ + ε 2 ) < 1/2 we obtain estimate (A.2 + C |∇ ⊥ ξ||∇ 2 P| L 1 + C |∇ 2 ξ||∇ P| L 1 + C Ω W 1.2 P W 1.1 ≤ C P W 2. F. Ghidaglia.1 + Ω∇ P W 1. F. J. End ofthe proof of Lemma A.2 . A.2 .) Geometry in Partial Differential Equations. Pures Appl. M...30). C.1 in L 2 and Cauchy-Schwarz we have ∇ P t Ω L 2 + Ω∇ P L 2 ≤ C ∇ P t Ω W 1. J. Rivière Using Cauchy-Schwarz we have first ∇ P t · ∇ ⊥ P W 1. Partial Differ. Publishing.2 ≤ C(δ + ε 2 ) P W 2. F.13).2 . pp. (A.1 1 ≤ C(1 + δ) ξ W 2. World Sci. 72(5). (A.2 + C(1 + δ) Ω W 1.2 + C(δ + ε 2 ) P W 2.20 T. Th. 267–310 (1993) [BeG1] Bethuel. References [Bet1] Bethuel.: Some applications of the coarea formula to partial differential equations. by the mean of a standard continuity argument we obtain that Ω is in Uε.: Weak limits of Palais-Smale sequences for a class of critical func- tionals.. (A. Var.4 is proved. Calc.5 is proved.2 .. Math. Sci. IX. In: Pràstaro.33) Using now (A. Sér. F. 1]. 1(3). (eds. we have ∇ P W 1.31) Using the embedding of W 1.16) and Lemma A.4 Let Ω in W 1.32) we have then that 1 ξ W 2.34) 1 Combining (A.33) and (A.-M.C .32) Combining (A.2 + C Ω W 1.2 + C ∇ PΩ W 1.34). Math. so(m) ⊗ R2 ) satisfying D2 |Ωt |2 ≤ ε connecting 0 and Ω. NJ (1994) .1 ≤ C P W 2.2 ≤ C ξ W 2. (A.-M. J. Paris. 1003– 1007 (1992) (French) [A regularity result for solutions to the equation of surfaces of prescribed mean curvature] [Bet2] Bethuel. Using the closeness of Uε. 1–17.2 (D2 .2 (D2 .: Improved regularity of solutions to elliptic equa- tions involving Jacobians and applications. 314(13). Sér.. I. the openness property given by claim and the fact that 0 ∈ Uε. Since D2 |Ωt |2 = D2 |Ω|2 t is an increasing function of t we have then a path among the elements in W 1. We  consider the  path Ωt = φt∗ Ω where φt (x) = tx and t ∈ [0. so(m) ⊗ ∧1 R2 ) satis- fying D2 |Ω|2 < ε for ε for which claim holds.2 Ω L 2 + C Ω W 1.: Un résultat de régularité pour les solutions de l’équation de surfaces courbure moyenne prescrite. (A. Acad.

. Pure Appl. (1). No. Pure Appl. 15–26 (1975) (German) [Hei3] Heinz. J. Mech. 1980. vol. Systems of non- linear partial differential equations.. 1–15 (1981) [Hei1] Heinz.. P. New York 1966 [Mu] Müller.. S. J. O. Pures Appl. 20–34 (1990) [Sha] Shatah. Iberoam. Ser C.-Phys. 247–286 (1993) [Ev] Evans. Nachr. Math. Reidel.: Compensated compactness and Hardy spaces. Princeton Mathematical Series. 41. F. Wiss.Conservation laws for conformally invariant variational problems 21 [BrC] Brezis.: Conformally invariant variational integrals and the removability of isolated singularities.: A partial regularity result for a class of stationary Yang- Mills fields in high dimension. K. S. Henri Poincaré 15(6). Dordrecht. 1983 [HiW] Hildebrandt.. Z. Wiss. Mat.: Higher integrability of determinants and weak convergence in L 1 . (1). Rivière..-L. M. Cambridge (2002) [Hel1] Hélein. 134. 1982 [Hil2] Hildebrandt. Gött. C. S. 269–296 (1995) [CLMS] Coifman..: Linear and Quasilinear Elliptic Equa- tions. (1970) .T. Math. New York London (1968) [LiR] Lin. Academic Press. Cambridge University Press. Ann. Beijing. Math.: Singular integrals and differentiability properties of functions. conservation laws and moving frames. Akad. vol. E. Y. E.: Energy quantization for harmonic maps. NATO ASI Ser.: Multiple integrals in the calculus of variations and nonlinear el- liptic systems.. M. http://fr. Rev. J. 1–15 (1986) (German) [On the regularity of weak solutions of nonlinear elliptic systems] [Hel] Hélein. Meyer. N. 116. 105. II. S.: Problèmes invariants par transformations conformes en dimension 2. Princeton. Reine Angew. (eds.. 19(1).: Multiple integrals in the calculus of variations.-Phys.: Multiple solutions of H-systems and Rellich’s conjec- ture. Reine Angew.-H. T. E. J. Math. 37(2). P. Die Grundlehren der mathematischen Wissenschaften. Anal. 195–219 (2003) [Mo] Morrey. Stud. NJ (1983) [Gr1] Grüter. 173–217. T. 1–13 (1975) (German) [Hei2] Heinz. E. Wu. IX. 177–193 (2002) [MeR] Meyer. W.: Partial regularity for stationary harmonic maps into spheres.: A regularity result for critical points of conformally invariant func- tionals. Duke Math. Math. Math. 1–3.M. Y. Princeton University Press. 67–86 (1975) [LaU] Ladyzhenskaya. Ration. Coron. Gött. 101–113 (1991) [Fre] Frehse. Beijing. Kl. 4(3).N.. Ann. II Math.) Proceedings of the 1980 Beijing Symposium on Differential Geometry and Differential Equations. 47(1–3). Math. 229–230 (1973) [FMS] Freire A.: Nonlinear elliptic systems and harmonic mappings. 72. Springer.B. Akad. Princeton. Sér.A.: Regularity of weak H-surfaces. J.: Weak solutions and development of singularities of the SU(2) σ- model. S. Z. 459–469 (1988) [Ste] Stein.-O.: Quasilinear elliptic systems in diagonal form. 30 Princeton University Press. Jr. J.. C. pp.. N. R.: Über die Regularität der Lösungen nichtlinearer Wellengleichungen. Wiss.. 111. vol. Science Press.: Über die Regularität schwacher Lösungen nichtlinearer elliptischer Systeme. J. Ural’tseva. 725–754 (1998) [Gia] Giaquinta. 111(1).: Weak compactness of wave maps and harmonic maps. Cambridge Tracts in Mathematics. F.. Struwe M. vol.. II Math. Kl. vol. Phys. Manuscr. Math. 130. Rivière..arxiv. Oxford 1982. Müller S. 412. Nachr.-Phys. Widman. pp.J. 149–187 (1984) [Cho] Choné.org/pdf/math/0101237? [Hil1] Hildebrandt. Sci.: A discontinuous solution of a mildly nonlinear elliptic system. M. S. H. Lions.: Some regularity results for quasilinear elliptic systems of second order. 481–615..S. Math. Arch.-M. Kl. Semmes. Nachr. 150. In: Chern. 142. 329. (2). Gött. Math. Akad. Potential Anal.. Inst.: Harmonic maps. Commun. Commun. Math.: Ein Regularitätssatz für schwache Lösungen nichtlinearer elliptischer Systeme. F. 85–104 (1984) [Gr2] Grüter.

Phys. Lecture Notes Phys.) Macroscopic Modelling of Turbulent Flows. Rivière [Ta1] Tartar.K.. J.: Connections with L p bounds on curvature.. In: Frisch.: Remarks on oscillations and Stokes’ equation. 31–42 (1982) [We] Wente. vol. H. K. Springer. 1984. Math. 230. Nice. 83(1). 24–31. (eds. pp. Commun.: An existence theorem for surfaces of constant mean curvature. 318–344 (1969) . U. 26.C. Berlin (1985) [Uhl] Uhlenbeck. Anal. L. et al. Math.22 T. Appl.