Low-energy α-harmonic maps into the round sphere Authors: Ben Sharp Published: 2026 ABSTRACT AbstractWe classify low-energy α-harmonic maps from a closed non-spherical Riemannian surface Σ of constant curvature to the round sphere via their bubble scales and centres. In particular we show that as 1<α↓1 and assuming Eα is close to |Σ|+4π then degree-one α-harmonic maps blow a bubble based at a critical point ac of a an explicit function J and at scale |J(ac)|-1(α-1). Up to a constant, J is the sum of the squares of any L2-orthonormal basis of holomorphic one-forms on the domain. Introduction One of the founding pillars of geometric analysis concerns the existence and regularity of harmonic maps. Given two Riemannian manifolds M, N, a harmonic map u∈W1,2(M,N) is a critical point (in some suitable sense) of the Dirichlet energy When M=Σ is two-dimensional this functional is invariant under conformal deformations of the domain which leads to a wonderfully rich theory from both an analytic and geometric viewpoint. In broad strokes this functional does not enjoy straightforward regularity properties, nor does it satisfy a Palais–Smale condition. These difficulties lead to both interesting and challenging mathematical problems. Remarkably critical points are always smooth in two dimensions thanks to Hélein’s moving frame approach [7, 8]; the resulting estimates hold only up to a potentially small regularity scale r(u,N,Σ)>0 whereupon the Dirichlet energy on any ball of radius r is below some threshold. Along a sequence of smooth maps uk (for instance) it is possible for rk to shrink to zero (i.e. for the Dirichlet energy to concentrate) and all the derivatives to blow up. Suitably re-scaling the domain to reverse the energy concentration leads to a sequence of harmonic maps defined on larger and larger regions of R2 which converge to a harmonic sphere ω:R2∪{∞}=S2→N. Starting with the famous work of Sacks–Uhlenbeck [24] α-harmonic maps have been used to study existence theory for harmonic maps in two-dimensions via a regularised functional Eα and α>1. Given a closed Riemannian surface Σ and N some closed Riemannian manifold, the α-energy Eα of a map u∈W1,2α(Σ,N) is defined as follows For α>1 fixed, Eα crucially satisfies the Palais–Smale condition, whilst all finite-energy critical points (α-harmonic maps) are uniformly regular, up to a regularity scale r=r(α,Eα(u),N,Σ) with smooth estimates which potentially degenerate (r→0), as α↓1, on regions where the Dirichlet energy concentrates. Nevertheless, it is well-known that harmonic maps appear as weak W1,2-limits of α-harmonic maps, as α↓1, modulo the appearance of singularities, again taking the form of harmonic spheres (bubbles) and geodesics of length ℓ∈[0,∞] (necks) which connect the weak limit with the bubbles. An ideal outcome would be that all the Dirichlet energy is captured by the limiting harmonic map and bubbles (an energy identity) or further that all necks have zero length (a no-neck result). In general both of these ideals are known to fail during the limiting process for α-harmonic maps: examples of (infinite-length) necks appearing and energy-loss can be found in [16]. Still, one can use analytical methods to find α-harmonic maps (direct minimisation, or min-max procedures) and in turn hope to recover relevant information on harmonic maps, as α↓1 see e.g. the seminal works of Sacks–Uhlenbeck [24] and Micallef–Moore [18]. Natural and important questions present themselves: how precisely does this singular convergence happen and can one recover all harmonic maps from surfaces as a limit of α-harmonic maps? When the target has positive Ricci curvature and the sequence of α-harmonic maps has finite Morse index, it is known that only finite-length geodesic necks appear and all energy is accounted for in the limiting process [15]. For arbitrary closed targets and certain sequences of α-harmonic maps obtained by min-max methods, similarly, all energy is accounted for in the limit by the work of Lamm [11], although it is not known whether or not necks might appear. If the target is a round sphere Sn, thanks to Jiayu Li and Xiangrong Zhu we know more: all energy is accounted for and the convergence happens in L∞ (no necks appear at all) [14]. See also [2] where this is proven for arbitrary homogeneous targets. When considering maps between round two-spheres u:S2→S2 it is well-known that, up to reversing orientations, the only harmonic maps are holomorphic and therefore rational maps from the extended complex plane to itself (see e.g. the Theorem of Wood–Lemaire [5, 11.3] for a more general result). Thus if we restrict to degree one harmonic maps, we are left only with the Möbius maps and their complex conjugates. Remarkably the question “are all harmonic maps attained as limits of α-harmonic maps” has only recently received attention. Lamm, Malchiodi and Micallef show that all degree ±1 α-harmonic maps u:S2→S2 with α-energy lower than 8α2π (Dirichlet energy lower than 12π), and α close to 1, must be rotations [12, 13]. Thus in particular the majority of degree-one harmonic maps between spheres are not reached by smooth limits of α-harmonic maps. Indeed they essentially classify all low-energy α-harmonic maps u:S2→S2 of degree 0 or ±1, see [13]. In this article we study α-harmonic maps u:Σ→S2, from non-spherical closed surfaces Σ, which have α-energy bounded above by 4π+|Σ|+δ for δ>0 small (in particular Dirichlet energy bounded above by 4π+δ) and α close to 1. The energy bound immediately tells us that the topological degrees of these maps are either 0 or ±1. Examples of such degree ±1 low-energy α-harmonic maps are guaranteed to exist by direct methods and we show that any such map must be quantitatively C1-close, with bounds purely in terms of α, to a finite-dimensional space of singularity models Z where each z∈Z approximates a single degree ±1 bubble and constant-body map configuration. In particular we obtain estimates on both the bubble scale and bubble proximity purely in terms of (α-1) and geometric data J:Σ→R. Definition 1.1 Let (Σ,g) be a closed Riemannian surface of genus γ≥1 equipped with a metric g of constant curvature zero when γ=1 and constant curvature -1 otherwise. In the flat case we also impose Areag(Σ)=1. Let {ϕj} be an arbitrary L2-orthonormal basis of holomorphic one-forms on Σ and define where c1=1 and when γ≥2, cγ=4. One may check directly that J is independent of the choice of basis. Remark 1.2 When γ=1 we have J≡-2π whilst in general we always know that J<0 by Riemann–Roch. J is directly related to the regular part of the Green’s function on Σ, see e.g. Appendix B and [17]. Without going into fine details, we let Z⊂W1,2(Σ,S2) denote the space of maps approximating a single degree ±1 bubble at small scale, so topologically Z≅Σ×[λ0,∞)×O(3) and given a∈Σ, λ≥λ0 and R∈O(3) the corresponding z=zλ,aR∈Z satisfies z∼Rπ(λ·) near a and in suitable local coordinates, otherwise z∼R(0,0,-1) on the rest of Σ. Here and throughout π(x)=2x1+|x|2,1-|x|21+|x|2 is the inverse stereographic projection from the South Pole and in particular for an almost-singular limit we have λ≫1, in which case π(λ·) maps a tiny disc (e.g of radius λ-12) into most of S2\(0,0,-1). Spaces closely related to Z have appeared elsewhere in [17, 22] where they are used in the study of H-surfaces in R3 and harmonic maps from surfaces into arbitrary analytic targets respectively; we recall the precise definition in Sect. 2.1. For now it is sufficient to note that any map which is (e.g. W1,2∩L∞) close to a single degree ±1 bubble and constant-body map configuration must be close to Z. In [17] the authors show that J(a) appears as the highest order term in an expansion in λ of the H-energy EH when restricted to Z. The space of singularity models Z appearing in [17] is slightly different to the one appearing here and the H-energy EH is the Dirichlet energy plus an enclosed volume term. In any case it is shown in that: On the other hand, if u is sufficiently close to Z and z∈Z is its closest-point projection (in an appropriate norm), one finds, after carefully analysing the higher-order derivatives of EH in a neighbourhood of Z, that for λ sufficiently large: These estimates in particular heavily rely on a suitable non-degeneracy of the second variation in normal (to Z) directions, along with some careful estimates on dEH and d2EH in a small neighbourhood of Z. Comparing the two facts above, one guarantees a “quantitative non-existence result” for almost-critical points close to Z in that setting, since crucially J<0, so ‖dEH(u)‖ must be bounded away from 0 by some suitable function of λ—see [17]. As appears in the work of Rupflin [22] this expansion remains morally true (with some serious additional complications) also for the Dirichlet energy when studying almost-harmonic maps into general closed analytic targets u:Σ→N which are sufficiently close to a zero-body one-bubble singularity based upon any un-branched bubble ω^:S2→N. The situation in [22] is more subtle as the space of nearby harmonic maps (to ω^) is not restricted to pre-composition by Möbius maps, moreover any space of singularity models Z must also take into account any non-integrable Jacobi fields. Nevertheless it is still the case that J(a) appears as the dominating term in the Dirichlet energy expansion (in λ), when restricted to Z. These facts yield deep implications for the harmonic map flow from non-spherical surfaces into analytical target manifolds, and in particular on discreteness of the low-end of the Dirichlet energy spectrum in this case. On the other hand recent progress concerning low-energy discreteness of the Dirichlet energy spectrum for maps from spherical domains to analytic targets can be found in [23]. In this article we restrict N=S2 and study the behaviour of Eα in a neighbourhood of Z - the restriction of the target to a sphere is necessary in order to have the favourable bubble-convergence results of Li and Zhu [14], which are unavailable (and untrue) for arbitrary targets. This then provides the technical benefit that the elements of Z are totally explicit. The flavour of the main results are rather different in the sense that α-harmonic maps are guaranteed to exist, so our focus is very much on gleaning their properties as opposed to a quantitative non-existence result. Following [22], it is necessary to define the following bubble-weighted inner product and norm for V,W∈W1,2(Σ,R3) when z∈Z where, for λ=λ(z) and ι is half the injectivity radius of Σ. Our results are guided by the properties of Eα when restricted to Z. When λ is bounded Eα(z) is comparable to the Dirichlet energy, so we expect to pick up similar terms in our expansion of Eα as for the Dirichlet energy, ∂λEα∼8πJλ-3<0. However, for each fixed α>1, it is straightforward to see that Eα(z)→∞ as λ→∞ (indeed we have λ2α-2≤C(Eα(z)+1) from Lemma 2.5) so for large λ we expect ∂λEα>0. These facts together suggest that ∂λEα must vanish somewhere on Z. Here, we show that (Proposition 4.1), for a=a(z) and λ=λ(z), This expansion indicates that ∂λEα(z) will vanish for a z∈Z when λ(z) and a(z) approximately satisfy J(a)λ-2+(α-1)=0. Thus for any α-harmonic map u which is close to z∈Z, we should also expect the first (highest-order) term in this expansion to be close to zero. This would force -J(a)λ-2∼(α-1) and suggest the appropriate bubble scale λ of u in terms of (α-1). Our first result confirms this line of reasoning, and since u is α-harmonic we are able to also prove quantitative-closeness to the space of singularity models Z in C1. We remind the reader that degree ±1 α-harmonic maps satisfying the hypotheses of the below Theorem are guaranteed to exist by direct methods. Theorem 1.3 Let (Σ,g) be as in Definition 1.1. There exist α0=α0(Σ)>1,δ=δ(Σ)>0,C=C(Σ)<∞ so that if 1<α≤α0 and u:Σ→S2 is α-harmonic with Eα(u)≤4π+|Σ|+δ then either u has degree zero and ‖∇u‖L∞(Σ)≤C, or u has degree zero and ‖∇u‖L∞(Σ)≤C, or u has degree ±1 and there is z(u)∈Z realising inf{‖u-z‖z:z∈Z}. Moreover for any such z(u), setting a=a(z(u)) and λ=λ(z(u)) we have |(α-1)+λ-2J(a)|≤C(α-1)32|log(α-1)|. Furthermore ‖u-z(u)‖W1,2α(Σ)≤C(α-1)|log(α-1)|. Finally, for any s∈[1,∞) there is K=K(s,Σ) so that ‖u-z(u)‖L∞(Σ)≤K(α-1)1-1s|log(α-1)|and‖∇u-∇z(u)‖L∞(Σ)≤K(α-1)12-1s|log(α-1)|. u has degree ±1 and there is z(u)∈Z realising inf{‖u-z‖z:z∈Z}. Moreover for any such z(u), setting a=a(z(u)) and λ=λ(z(u)) we have |(α-1)+λ-2J(a)|≤C(α-1)32|log(α-1)|. Furthermore ‖u-z(u)‖W1,2α(Σ)≤C(α-1)|log(α-1)|. Finally, for any s∈[1,∞) there is K=K(s,Σ) so that ‖u-z(u)‖L∞(Σ)≤K(α-1)1-1s|log(α-1)|and‖∇u-∇z(u)‖L∞(Σ)≤K(α-1)12-1s|log(α-1)|. We furthermore study the dependence of Eα|Z on the gluing point a. Of course a flat torus is invariant under Euclidean translations in the domain: given A∈TaΣ, in this case and there is no restriction on the location of a bubble. This is no longer true for higher-genus surfaces and we show that (see Proposition 4.2): Theorem 1.3 tells us that when u is α-harmonic and close to z∈Z then (α-1)∼λ-2, so the first term above is of highest order. Once again, our rule of thumb suggests that ∇AJ(a) must be close to zero, which we confirm next. Theorem 1.4 Let (Σ,g) be as in Definition 1.1 with genus γ≥2. There exist α0=α0(Σ)>1,δ=δ(Σ)>0,C=C(Σ)<∞ so that if 1<α≤α0 and u:Σ→S2 is α-harmonic of degree ±1 with Eα(u)≤4π+|Σ|+δ then if z(u)∈Z achieves inf{‖u-z‖z:z∈Z} (as guaranteed by Theorem 1.3 part 2) then for a=a(z): We immediately obtain the following compactness result: Corollary 1.5 Let (Σ,g) be as in Definition 1.1 and {uk:Σ→S2} a sequence of αk-harmonic maps with 1<αk↓1 and Eαk(uk)→Λ≤4π+|Σ| then, up to subsequence, we either have uk all have degree zero and they converge smoothly to a degree-zero harmonic map u:Σ→S2 or uk all have degree zero and they converge smoothly to a degree-zero harmonic map u:Σ→S2 or uk all have degree 1 or all have degree -1 and they bubble-converge to a single-bubble-constant-body configuration where the bubble is blown at a critical point ac of J and at scale rk=|J(ac)|-1(αk-1). uk all have degree 1 or all have degree -1 and they bubble-converge to a single-bubble-constant-body configuration where the bubble is blown at a critical point ac of J and at scale rk=|J(ac)|-1(αk-1). Remark 1.6 As stated previously J can be defined purely via mixed second-order derivatives of the regular part of the Green’s function of the Laplacian on Σ. For simply connected domains in the plane the equivalent function is known to be closely related to the Robin function—see [3]. The critical points of the Robin function, and related functions are known to be significant in other critical variational problems see e.g. [1, 4, 6, 9, 10, 20]. All the main results in this article have since been shown to have analogues in the setting of ϵ-harmonic maps, see [21]. Outline of the paper The proofs of the main Theorems (see Sect. 3) rely heavily upon a careful analysis of the behaviour of Eα when restricted to the manifold Z (and in small neighbourhoods thereof). In Sect. 2 we recall the space of singularity models Z and expand the α-energy along Z for λ large. We recall also the crucial background results concerning α-harmonic maps into round spheres [14] which we combine with suitable elliptic estimates to prove first that the α-harmonic maps satisfying our hypotheses are either analytically well-controlled or lie close to Z (Lemmata 2.10–2.12). We also recall the desired non-degeneracy of energy in normal directions which essentially follows from a similar estimate in [22] (cf [17]): this comparison relies heavily on the known theory for α-harmonic maps into spheres, see Lemma 2.13. In Sect. 4 we derive an expansion in λ and (α-1) for ∂λEα(z) and ∇A(Eα(z)) along the submanifold Z. The key point of these expansions are that properties of J determine the behaviour of the Eα-energy profile when restricted to Z. In Sect. 5 we obtain control on dEα and d2Eα when we are on and nearby Z—these estimates are crucially of “lower-order” than those appearing in Sect. 4 which allows us to pass information on the behaviour of Eα on Z to the behaviour of Eα in a neighbourhood of Z. We provide the proofs of our main Theorems in Sect. 3 assuming the technical expansions/estimates from Sects. 4 and 5. Preliminaries Definition of the set of adapted bubbles Let Σ be as in Definition 1.1. Here we will define the set of adapted bubbles, denoted by Z={Rzλ,a:Σ→S2} where λ≫1 is the bubble scale, a∈Σ is where the bubble is attached and R∈O(3). Notice this follows exactly along the lines of [17] and more relevantly [22]—indeed it corresponds to the same space in the latter case when N=S2 and ω^:S2→S2 is the identity. We recall the full details for our setting: Denote by π the inverse stereographic projection π(x)=2x1+|x|2,1-|x|21+|x|2 and we want to define zλ,a in a way that we have zλ,a≈πλ(x):=π(λx) in oriented isothermal coordinates x=Fa(p) on the geodesic ball Bι(a)⊂(Σ,g), ι:=12inj(Σ,g), which are as follows: Remark 2.1 If g is hyperbolic we set ρ=tanh(ι) and use that for any a∈Σ there exists an orientation preserving isometric isomorphism where Dρ:={x∈R2:|x|<ρ} and gE is the Euclidean metric. In the flat case we will always a-priori pick a tiling of R2 which represents Σ, set ρ=2ι and use the resulting euclidean translations Fa to the origin as coordinates. When γ≥2 we note that Fa is determined only up to a Euclidean rotation on Dρ, but that this will not affect the definition of the bubble set since this rotation can be undone by choosing an ambient rotation R∈O(3) (see Remark 2.3 and the following definition of Z). In the coordinates introduced in Remark 2.1 we write the Green’s function G via Ga(x,y)=-log|x-y|+Ja(x,y) for some smooth function J. Hence we have for x≠y The function J depends on the choice of coordinates used and is only defined when both arguments are close to a. We express Ja in terms which clearly show dependence on the choice of coordinates via (for p,q∈B2ι(a)): and setting ∂qi=(Fa-1)∗∂yi we have by definition Define r via 4r=tanh(ι) in the hyperbolic case and 4r=ι in the flat case. Let ϕ∈Cc∞(D2r,[0,1]) be radial with ϕ≡1 on Dr. Given any a∈Σ and Fa:B2ι(a)→Dρ a choice of isometry as in Remark 2.1, we define where z^λ,a is given in the local coordinates x=Fa(p) by (with ∇y=(∂y1,∂y2)) Letting j(x)=jλ,a(x):=2λ∇yJa(x,0)-2λ∇yJa(0,0),0, z~a,λ is described in these local coordinates on Bι(a) by The error in (2.5) follows from the definition of πλ and (2.2): Remark 2.2 The conformal covariance of the Laplacian implies that ∂yiJa(x,0) is harmonic in x, so in particular j is also harmonic. On an unrelated note, for symmetry reasons it is possible to check that ∇yJa(0,0)=0 when γ=1, however it is unclear whether this remains true for γ≥2. Remark 2.3 When γ≥2 we note the dependence on the choice of Fa via (for p close to a): with jFa(p)=jλa(p)=2λ∂q1Ja(p,a)-∂q1Ja(a,a),∂q2Ja(p,a)-∂q2Ja(a,a),0. Thus if we pick another isometry as per Remark 2.1Fˇa we have Rˇ=Fa∘Fˇa-1∈SO(2) and now let R be the obvious extension of Rˇ to O(3) (as the identity in the third component). Following the definitions above we have Thus a different choice of Fa in Remark 2.1 has the effect of an ambient rotation and does not change the eventual bubble set Z defined below. We then set See Appendix A for further expansions and estimates for z∈Z and their derivatives. Here are throughout we will also use the notation Remark 2.4 (Notation) The notation f=O(g) means that there exists some C<∞ so that |f|≤Cg and f≃g means there is C<∞ so that C-1|g|≤|f|≤C|g|. The constant C may change from one line to the next but in any given statement C is fixed and depends at most upon the injectivity radius of, and the Green’s function on, Σ (sometimes neither). In particular for any fixed Σ as in Definition 1.1, a-posteriori there is some uniform C which works for all statements in this paper. We will use the notation Δ, ∇, ·, dΣ to denote these objects with respect to the metric g on Σ unless we are in x=Fa(p) coordinates at which point we will use the notation ∇g, Δg, ·g, dxg etc to denote these objects. In these coordinate only, the same symbols without a g are the standard Euclidean ones (as in e.g. (2.4) above). In the first instance we will determine the behaviour of Eα on Z - whereas in Sects. 4 and 5 we will more carefully analyse the behaviour of dEα and d2Eα on and close to Z. Lemma 2.5 Let α≤2, δ≤1 and z∈Zα,δ, then for λ=λ(z), Corollary 2.6 Let 1<α≤2, δ≤1 and z∈Zα,δ, then there exists C=C(Σ)<∞ so that λ(z)2α-2≤C. Furthermore for any ε>0 there is α0=α0(Σ)>1 and δ=δ(Σ)>0 so that α≤α0 and z∈Zα,δ implies Proof where g denotes the metric in x-coordinates. With cγ∈{1,4} defined as in Definition 1.1 we have dxg=(cγ+O(|x|2))dx and |∇gz|g2=(cγ-1+O(|x|2))|∇z|2, noting that when γ=1 we have cγ=1 and no error (i.e. everything is Euclidean). On Dr we have from (A.1) and (A.3), Using that, for x∈Dρ, O(λ-2)+O(λ-1)∇πλ=O(|∇πλ|2(λ-2+|x|2)) we obtain where χ(λ,|x|)=O(λ-2+|x|2). By the mean value formula, we note that (given a,b>0) there is some c∈(0,a) so that Employing this twice, with the roles of a and b are reversed, we may write where a=2 and b=|∇πλ|2(cγ-1+χ) and e0≤(abα-1+aα-1b). This yields Using that (cγ-1+χ)α=cγ-α+χ for χ=O(λ-2+|x|2) we now have From e.g. (A.2) we have which gives from which the result follows. □ Analytical set up and further background results We crucially require the energy identity and no-neck result of Jiayu Li and Xiangrong Zhu for α-harmonic maps into round spheres [14, Theorem 1.1, Lemma 4.1]. In particular their results straightforwardly imply: Theorem 2.7 [14, Theorem 1.1, Lemma 4.1] If {uk}∈W1,2αk(Σ,S2) is a sequence of αk-harmonic maps with 1<αk↓1 and Eαk(uk)≤Λ then there exist a finite number of bounded-energy smooth harmonic maps u∞:Σ→S2, {ψj}:R2→S2 and accompanying distinct point-scale sequences (akj,rkj) so that akj→ak∈Σ, rkj→0 and : For all j we have (rkj)1-αk→1, and ‖(2+|∇uk|2)αk-1‖L∞(Σ)→1 For all j we have (rkj)1-αk→1, and ‖(2+|∇uk|2)αk-1‖L∞(Σ)→1 Eαk(uk)→|Σ|+E(u∞)+∑jE(ψj) Eαk(uk)→|Σ|+E(u∞)+∑jE(ψj) uk-u∞-∑jϕkj(·)ψj((rkj)-1Fakj(·))-ψj(∞)W1,2α(Σ)→0, and uk-u∞-∑jϕkj(·)ψj((rkj)-1Fakj(·))-ψj(∞)L∞(Σ)→0, where ϕkj∈C∞(Σ) is defined by ϕkj(p)=ϕ(Fakj(p)) on spt(ϕkj)⊂Bι(akj) for some fixed ϕ∈Cc∞(Dρ,[0,1]) with ϕ≡1 on Dρ/2 and ι, ρ are defined as in Remark 2.1. uk-u∞-∑jϕkj(·)ψj((rkj)-1Fakj(·))-ψj(∞)W1,2α(Σ)→0, and uk-u∞-∑jϕkj(·)ψj((rkj)-1Fakj(·))-ψj(∞)L∞(Σ)→0, where ϕkj∈C∞(Σ) is defined by ϕkj(p)=ϕ(Fakj(p)) on spt(ϕkj)⊂Bι(akj) for some fixed ϕ∈Cc∞(Dρ,[0,1]) with ϕ≡1 on Dρ/2 and ι, ρ are defined as in Remark 2.1. Remark 2.8 Notice that in the scenario that one bubble forms and u∞ is a constant, we must have ψ(∞)=u∞ and in the above convergence we may write for both norms. Weighted inner product: For v∈C∞(Σ,S2) we denote by Following [22], it is necessary to define the following inner product for V,W∈W1,2(Σ,R3) when z∈Z (so in particular also for V,W∈Γ2(z)) where so in particular |∇z|≤Cρz on Σ. Remark 2.9 We have the following useful estimate from [22, (2.28) and Appendix B]: For any s∈[1,∞) there exists K=K(s,Σ) so that Lemma 2.10 If u:Σ→S2 is an α-harmonic map with ‖u-z‖z2<ε04 (with ε0 from Theorem C.1), α>1 and z∈Z then there is some C=C(Σ)<∞ so that ‖∇u‖L∞≤Cλ(z) and we have : for any p∈Σ and s∈[1,∞) there exists K(s,Σ)<∞ Proof For z∈Z with ‖u-z‖z2<ε4 let λ and a be defined by z. Fix μ=12ι and notice first that for any p with dΣ(p,a)>μ there is some uniform radius r>0 so that ‖z‖C2(Br(p))≤Cλ-2. Thus, there is some uniform radius r so that ‖∇u‖L2(Br(p))2≤2‖u-z‖z2+Cr2λ-4<ε0 (ε0 from Corollary C.2) meaning that ‖∇u‖L∞(Br/2)≤C(‖u-z‖z+λ-2) by the regularity of α-harmonic maps. Thus the estimates trivially hold on Σ\Bι2(a). On Bι(a) our usual coordinates x are defined on DR where R=tanh(ι/2) when γ≥2 else R=ι. Let z^(x)=z(λ-1x), which satisfies ‖z^‖C2(DλR)≤C for some uniform constant C. In particular there is some uniform radius r so that for any Dr(x)⊂DλR. Defining u^(x)=u(λ-1(x)) in these coordinates, and the scale-invariance of ‖·‖z give that and since u is α-harmonic u^ solves, defining f(x)=4(1-|x|2)2 when γ≥2 else f≡1, and f^(x)=f(λ-1x): meaning that ∇u^ satisfies uniform L∞-estimates at this scale by Theorem C.1. In particular defining μ=tanh(ι4)tanh(ι2) when γ≥2 or μ=34 otherwise, ‖u^‖C1(DμλR)≤C and by undoing the scaling ‖∇u‖L∞(DμR)≤Cλ and ‖∇u‖L∞(Bι2)≤Cλ. Using Theorem C.1 again tells us that for any disc of radius 2 in DλR. Now, in our coordinates we have that -Δz^=z^|∇z^|2+O(λ-2) on DλR, as can be checked directly from the definition. Thus Standard elliptic estimates, using the scale-invariance of the ‖·‖z-norm, first give for all s≥1 (on balls with arbitrary centre x∈DμλR) the existence of a constant Ks so that and then by Sobolev embedding By undoing the scaling and using Remark 2.9 we have We are left with, in particular and therefore □ Lemma 2.11 There exist α0=α0(Σ)>1, δ=δ(Σ)>0 and C=C(Σ)<∞ so that if 1<α≤α0, and u:Σ→S2 is a degree zero α-harmonic map with Eα(u)≤4π+|Σ|+δ then ‖∇u‖L∞(Σ)≤C. Proof Suppose not, then there is a sequence uk of αk harmonic maps of degree zero with αk↓1, Eα(uk)→4π+|Σ| and uk bubble converges to a limit with at least one bubble. In fact it must have exactly one bubble of degree ±1 and energy 4π so u∞ is a constant map. Since bubble convergence preserves homotopy type in this case we have contradicted the fact that each uk has degree zero. □ Using the above, along with a direct application of Lemma 4.2 in [22] we are able to prove Lemma 2.12 For all ε1>0 there exist α0=α0(Σ)>1, δ=δ(Σ)>0 so that if 1<α≤α0 and u:Σ→S2 is a degree-one α-harmonic map with Eα(u)≤4π+|Σ|+δ, then ‖∇u‖L∞>ε1-1 ‖∇u‖L∞>ε1-1 1≤‖(2+|∇u|2)α-1‖L∞(Σ)<1+ε1 1≤‖(2+|∇u|2)α-1‖L∞(Σ)<1+ε1 There exists z∈Z so that ‖u-z‖W1,2α(Σ)+‖u-z‖L∞(Σ)<ε1. There exists z∈Z so that ‖u-z‖W1,2α(Σ)+‖u-z‖L∞(Σ)<ε1. (from [22, Lemma 4.2]) There exists z∈Z so that ‖u-z‖z=infζ∈Z{‖u-ζ‖ζ} and for this z we may still assume that ‖u-z‖L∞<ε1. (from [22, Lemma 4.2]) There exists z∈Z so that ‖u-z‖z=infζ∈Z{‖u-ζ‖ζ} and for this z we may still assume that ‖u-z‖L∞<ε1. Proof Suppose not, then in particular we know that there is some ε1>0 and sequences αk↓1, δk→0 and maps uk as in the statement for which one of 1-3 fails for infinitely many k and for this ε1>0. 2. cannot fail for infinitely many k since either ‖∇uk‖ remains uniformly bounded (in which case it is trivially true) or uk converges as in Theorem 2.7, for which we have that 2. holds eventually. If 1. fails for infinitely many k then the regularity theory of α-harmonic maps implies that uk must converge smoothly to a harmonic map u∞:Σ→S2 which itself has degree one and E(u∞)=E1(u∞)-|Σ|=4π. Since this map has degree one we know that the area of its image is at least 4π, thus its Dirichlet energy equals its area and it must be conformal. We have proved the existence of a degree one holomorphic map from Σ to S2, contradicting the fact that Σ is not a sphere. Thus in particular we cannot have smooth convergence to a limit, and we must have convergence to a non-trivial bubble configuration of total energy 4π and degree one. Since all harmonic spheres are conformal and therefore rational maps and their complex conjugates there is exactly one bubble ψ of degree one (or minus one) and u∞ must be a constant which we write in the form R(0,0,-1) for some appropriate R∈O(3). Thus ψ is of the form Rπλ(x-x0) for some fixed x0. By a translation in the domain and rotation in the target we may assume that ψ=πλ0 and we have that uk converges as in Theorem 2.7. Letting zk be defined via λ=rk-1λ0→∞ we see immediately that 3. cannot fail for infinitely many k (see also Remark 2.8). We have proven the required result for the first three points. For 4.: it follows directly from Lemma 4.2 in [22] that for any ν1>0 there is ν2>0 so that if there is z0∈Z with then there is z∈Z realising distz(u,Z)≤Cν2 so that thus by picking α0-1 and δ sufficiently small we can ensure 4. holds also. □ It is crucial that the energy is non-degenerate in normal directions (to Z) in our setting. The proof of this fact follows from the analogous results for the Dirichlet energy [22] (cf [17] for H-surfaces) - see the claim and its proof in the below. Lemma 2.13 There exist α0=α0(Σ)>1, δ=δ(Σ)>0, c0>0 so that for any α≤α0 and u:Σ→S2 α-harmonic, degree one, Eα(u)-4π-|Σ|<δ, and if z∈Z realises ‖u-z‖z=infζ∈Z{‖u-ζ‖ζ}, then setting W=dP(z)[u-z] and w=u-z we have Proof Write TzZ⊥z to mean the ⟨,⟩z-orthogonal complement of TzZ⊂W1,2(Σ,R3). We first Claim: If V∈Γ2(z)∩TzZ⊥z then there is c~0>0 so that Assuming the Claim, it follows from [22, Lemma 4.3], that if Y=W⊤TzZ∈Γ2(z)∩TzZ and V=W⊥TzZ∈Γ2(z)∩TzZ⊥z then there exists C<∞ so that and therefore Since W=dP(z)[u-z] we have (by checking directly, or see (5.5)) ‖W‖z≥12‖w‖z so that the above becomes, after choosing α0, δ appropriately and applying Lemma 2.12 For any ε>0, once again ensuring that α0, δ are sufficiently small, we may ensure from Lemmata 2.10 and 2.12λ(z)>Cε-1 and from Corollary 2.6 that which in turn, amongst other things yields that Putting the above threads together, along with some straightforward application of Young’s inequality with a ν and possibly after lowering α0 and δ we have some c0>0 so that as required. It remains to prove the Claim. From (2.9) We now appeal directly to Lemma 3.2 in [22] which tells us that all eigenvalues of d2E(z) are uniformly bounded away from zero on (TzZ)z⊥, for λ sufficiently large, which for us is ensured by Lemmata 2.10 and 2.12 implying λ(z)>Cε-1. In our case all eigenvalues must be non-negative since the limiting bubble (the identity map on the sphere) is stable, so all eigenvalues are uniformly positive and bounded away from zero: there exists c^0 so that By potentially lowering α0, δ to make ε sufficiently small, we have proved the claim. □ Proof of the main theorems assuming further technical details Proof of Theorem 1.3 Let u be an α-harmonic map as per the hypotheses. For energy reasons u must have degree zero or one. If it has degree zero then Lemma 2.11 proves the first part of the Theorem. If u has degree one, we may choose δ, α0 as in Lemma 2.12 associated with some ε<ε04 and let z be a resulting element in Z. Setting w=u-z, ut=P(z+t(u-z)) where P(q)=q|q| is the closest point projection to S2. We let Wt=∂tut∈Γ2(ut) with W0=W=dP(z)[w]. Then Now by potentially lowering α0, δ and utilising Lemma 2.13 there is some c0>0 so that By Lemmata 5.2 and 2.12 along with Corollary 2.6 we may lower α0, δ sufficiently so that Lemma 5.1 combined with (5.5) gives Putting the above estimates together gives Now, for T∈TzZ we set Tt=dP(ut)[T] and note that We will apply this formula for Tλ=∂λz which gives, first using Lemma 5.1, (5.5) and then (3.1): and Lemma 5.3 with (3.1): Therefore from (3.2) and (3.1) we have, By Proposition 4.1 we have which means Now Corollary 2.6 gives 1≤λ2α-2≤1+ε so in particular (α-1)logλ≤Cε and we can assume first which readily gives the existence of some C<∞ so that and then which is the first desired estimate in the degree one case. The final three estimates follow directly from this combined with (3.1), Lemma 2.10 and Remark 2.9. □ Proof of Theorem 1.4 Since we know that |J(a)|λ-2≃(α-1) applying (3.2) to T=TA and using Lemmata 5.1 and 5.3 combined with (3.1) we have first and second meaning that Now by Proposition 4.1 we have which means that as required. □ Energy expansions when restricted to Z Recalling the notation introduced in Remark 2.4: Proposition 4.1 Let α≤2, δ≤1 and z∈Zα,δ. For λ=λ(z) and a=a(z) we have Proof Using (A.11) and integrating (C.3) where we have used that Δgfdxg=Δfdx and furthermore ∇gf1·g∇gf2dxg=∇f1·∇f2dx. Dealing with I: For the first term above we have from (A.1), By (A.2) using (A.4), (A.6), (A.7) and (A.9). The first term above may be written and the second term, since j is harmonic (see Remark 2.2) using that πλ is conformal (see (A.2)) and ∂λπλ=λ-1x·∇πλ(x). Thus (4.1) becomes In particular using (2.7), (A.2) and Corollary 2.6: giving as can be checked directly. Recall again from (2.7) that |∇gz|g2=|∇πλ|2(cγ-1+χ) where χ=O(λ-2+|x|2) which gives (by a few applications of the mean value formula (2.8)), when |x|≤λ-12 since χ≤|∇πλ|-2. Thus where Now (4.3) and (4.4) give Recall that j(x)=jλ,a(x)=2λ∇yJa(x,0)-2λ∇yJa(0,0),0, and πλ1,2=2λx1+λ2|x|2 so, first by Taylor expansion about x=0 we have and then where f0 is the constant f0=∂x2∂y1J(0,0)+∂x1∂y2J(0,0). In particular and as can be checked directly. Using (4.7)–(4.9), (4.6) becomes Using the facts that, for radial f and a disc of any radius centred at the origin we see that, setting J(a):=(∂x1∂y1Ja(0,0)+∂x2∂y2Ja(0,0)), which coincides with Definition 1.1 (see Appendix B) Dealing with II: From (2.6) we have, on Dr and from (A.1), (A.6) From (A.1) and the definition of |∇gz|g2 it follows that After applying (A.2)–(A.4), Using (4.10)–(4.12), along with (A.2) and ∂λπλ=λ-1x·∇πλ we recover Similarly to (4.4) and since χ|∇πλ|2α-4≤C|∇πλ|2α-6 we have where Thus (4.14) gives and therefore (4.13) becomes By direct computation so by (4.14) again it can be checked that: and (4.15) becomes □ Proposition 4.2 Let α≤2, δ≤1 and z∈Zα,δ. Then for a surface of genus γ≥2, for λ=λ(z) and a=a(z) we have Proof From (A.11) and (A.18), similarly to the beginning of the proof of Proposition 4.1 we have Dealing with III: For the first term above we have and trivially giving since j is harmonic (Remark 2.2)and from (A.4), (A.9), (A.15), (A.17). The first term above may be written from (A.14), the definition of πλ and that |j|≤C|x|λ-1. Again using (A.14) the second term becomes since ∂1πλ·∂2πλ=0, |∂1πλ|2=|∂2πλ|2=12|∇πλ|2. Thus (4.16) may be written Using again (2+|∇gz|g2)α-1≤C (see (4.2)) we see that and therefore that Recall that j(x)=jλ,a(x)=2λ∇yJa(x,0)-2λ∇yJa(0,0),0, and πλ1,2=2λx1+λ2|x|2 so. Combining this with (4.4)–(4.5), after a short calculation we obtain By Taylor expansion about x=0 we have which gives for some suitable radial functions f0,f1,f2. Using also the facts: ∫Drxif(|x|)=0 for each i, ∫Drx1x2f(|x|)=0, ∫Dr(x1)2f(|x|)=∫Dr(x2)2f(|x|) and from Lemma B.1 we are left with (after a series of direct calculations from (4.17)) Dealing with IV: Combining (4.10) and (4.12) with we obtain since πλ is conformal. Using (4.14) we have that giving Now, since for some radial function f we note that again for any radial function F. Thus it remains to suitably estimate only the non-radial terms of (2+|∇gz|g2)α-2 in (4.18). Instead of (4.14) we again use |∇gz|g2=|∇πλ|2(14+χ) and the mean value theorem to give Since (2+|∇πλ|2)α-2 is radial we are left with as can be checked directly using e.g. (A.2). Thus we have IV=O(λ-1(α-1)) and we are done. □ Estimates for dEα and d2Eα close to Z Here we prove that the desired ‘error’ terms appearing in the comparisons of 0=dEα(u)[·] and dEα(z)[·] are lower order in relation to the expansions of Sect. 4. We first prove the required estimates on Z itself: Estimates on Z Lemma 5.1 Let α≤2, δ≤1 and z∈Zα,δ, then there exists C=C(Σ)<∞ so that for any V∈Γ2(z) we have Furthermore and Proof From (4.2) we know (2+|∇z|2)α-1≤C. Thus using (A.10) and (A.11) and Remark 2.9. Using (A.9) we note that which yields Finally since V∈Γ2(z) we need only estimate (Δπ+Δj⊤)⊤z which we do now. From (A.2), (A.8) and (A.5) we have and also, since πλ·j⊤=0 which gives We thus have Putting together (5.1), (5.2) and (5.3) gives the first estimate. For the second one we use (A.11) to give where the last term is bounded above by λ-2|Σ|‖V‖z by Cauchy-Schwartz. On Dr we have ∂λz=∂λπλ+O(|x|λ-2), |∇z|2=|∇πλ|2(1+χ) and so the above becomes, using also (A.9) and (A.10) We estimate using (A.2), (A.9) and (A.10) and also ∇∂λπλ|∂Dr=O(λ-2). Combining this with Remark 2.9) the above reduces to One checks, by direct calculation, that Thus the remaining term to estimate has an integrand of the form, using that V·z=0 a.e.: which finishes the second estimate. For the third estimate we do the same calculations for the second but using instead (A.13)–(A.18) to give first and then, on Dr we have ∇Az=-Aj∂jπλ+O(|x|2)|∇πλ|, |∇z|2=|∇πλ|2(1+χ) and meaning that (since |∇Aj∂jπλ|∼λ-1 on ∂Dr), Once again by direct calculation we have and using again that V·z=0 a.e.: We leave the final details to the reader since they are analogous to the previous case. □ Difference estimates for dEα and d2Eα near Z Here we will consider u∈W1,∞(Σ,S2) which are close to a specific region of Z in the sense that there is some z∈Z and Λ<∞ so that We will apply the below results in the setting that u is α-harmonic and satisfies the hypotheses of the main Theorems, so in particular Λ can be thought of as small. For future reference we first note, setting ut=P(z+tw)=z+tw|z+tw| and for A,B∈Γ2(z), C,D∈Γ2(ut): Lemma 5.2 Let Λ>0, u∈W1,∞(Σ,S2) and z∈Z. Suppose max{(2+|∇u|2)α-1,(2+|∇z|2)α-1}≤1+Λ and ‖u-z‖L∞≤12. Setting w:=u-z, ut=P(z+tw), Wt=dP(z+tw)[w] and W=W0, there exists C=C(Σ)<∞ so that and Proof First that since w⊥z=(w·z)z=-12|w|2z we have W=w+12|w|2z giving, using very rough estimates Thus there is some C so that Straight from the definition: so in particular using (5.5) We also have ∇ut=∇z(1+O(t|w|))+t∇w(1+O(|w|)) which gives and By the hypotheses on u and z, using (5.9) we have and then Combining this with (5.5)–(5.8) and (5.4) give for A=B=W and C=D=Wt, recalling that |∇z|≤Cρz on Σ: Now, Thus by (C.2), using that A(z)(V,W)=-z(V·W) for the round sphere, this gives which in particular yields This and (5.8) give as required. □ Lemma 5.3 Let u∈W1,∞(Σ,S2) and z∈Z. Suppose max{(2+|∇u|2)α-1,(2+|∇z|2)α-1}≤1+Λ and ‖u-z‖L∞≤Λ for some Λ<∞. Set w:=u-z, ut=P(z+tw) and Wt=dP(z+tw)[w] and given T∈Γ2(z) let Tt=dP(ut)[T]. Then there exists C=C(Σ)<∞ so that : When Tλ=∂λz|d2Eα(z)[Tλ,W]-d2Eα(ut)[Ttλ,Wt]|≤Cλ-1(1+Λ)‖w‖z2+(α-1)‖w‖z and |dEα(ut)[dP(ut)∂tTtλ]|≤Cλ-1(1+Λ)‖w‖z2. When Tλ=∂λz|d2Eα(z)[Tλ,W]-d2Eα(ut)[Ttλ,Wt]|≤Cλ-1(1+Λ)‖w‖z2+(α-1)‖w‖z and |dEα(ut)[dP(ut)∂tTtλ]|≤Cλ-1(1+Λ)‖w‖z2. When TA=∇Az,|d2Eα(z)[TA,W]-d2Eα(ut)[TtA,Wt]|≤Cλ(1+Λ)(‖w‖z2+(α-1)‖w‖z) and |dEα(ut)[dP(ut)∂tTtA]|≤Cλ(1+Λ)‖w‖z2. When TA=∇Az,|d2Eα(z)[TA,W]-d2Eα(ut)[TtA,Wt]|≤Cλ(1+Λ)(‖w‖z2+(α-1)‖w‖z) and |dEα(ut)[dP(ut)∂tTtA]|≤Cλ(1+Λ)‖w‖z2. Proof Given 0≤x≤y∈R assume that (2+y2)α-1≤1+Λ then the mean value formula straightforwardly gives the existence of some ξ∈[x,y] so that: Combining this with (5.8) gives whilst we trivially note for later that Directly from (5.11) and (5.8) we also see that Using |∇z|≃ρz on Bι(a) together with (5.4), (5.8), (5.10) and (5.12)–(5.14) gives, with U1={|∇u|≥12|∇z|} and U2=Σ\U1 We collect the following readily-checked facts concerning Tt which follow from the definition: Now for T=Tλ=∂λz we have |Tλ|≤Cλ-1 and by (5.15) we have the same bound for Ttλ. Also, Note that both Tλ and ∇Tλ are bounded by λ-2 on Σ\Bι(a). The above now becomes, using also (5.6), (5.13) and again with U1={|∇u|≥12|∇z|} and U2=Σ\U1: which readily gives the required estimate by (5.5) and (5.7). By definition ∂tTt=d2P(ut)[Tt,Wt], so by (C.1) and (C.2), dP(z)((d2P(z)[W,T]))=0 giving This means in particular using (5.6)–(5.8) and (5.15) and for T=Tλ When T=TA the proof follows exactly the same lines expect that we use |TA|≤Cλ and |∇TA|≤Cλρz, whilst both TA and ∇TA are bounded by λ-1 on Σ\Bι(a). These give precisely the same estimates with a factor of λ2 the only final difference. □ Acknowledgements I would like to thank Mario Micallef and Tobias Lamm for their interest, encouragement and inspiring conversations. I would also like to thank Melanie Rupflin and Andrea Malchiodi for their helpful comments on, and interest in, the final drafts of this work. The author was supported by the EPSRC grant EP/W026597/1. A Initial expansions In Ur(a)=Fa-1(Dr), z~λ,a=πλ(Fa(p))+jλ,a(Fa(p)) meaning that in Fa-coordinates x∈Dr we have where jλ,a⊤=dP(πλ)[jλ,a]=jλ,a-(jλ,a·πλ)πλ. Mostly we will drop the explicit λ,a dependencies from z, j and K, writing simply z(x)=πλ+j⊤+K. We will frequently use standard properties of πλ(x)=2λx1+λ2|x|2,1-λ2|x|21+λ2|x|2: On Ur(a)≅Dr we have the following rough estimates as can be checked directly: From the definition of j=jλ,a we have |∇j|=O(λ-1) and since j(0)=0 we have after a short calculation. Using (A.2) and the properties of P along S2 we also have Recall that the third component of j is zero, and πλ1,2=2λx1+λ2|x|2, which gives A.1 Expansions involving ∂λz After further estimates one can check that For our purposes later we note that Lemma A.1 Proof Since j is harmonic we have We also have These together give Using ∂λπλ=λ-1x·∇πλ and ∇j·∇πλ=O(1)(1+λ2|x|2) and the aforementioned estimates we see therefore that □ We finally have, on Ur(a): along with On the other hand, in Σ\Ur(a) it is straightforward to check, using (2.3), and (2.5) that we have the following estimates on zλ,a: A.2 Expansions involving ∇Az when γ≥2 Let A∈TaΣ≅(Fa)∗T0D and for 0≤s≪1, as:=ExpaΣ(sA) and we wish to compute Given our choice of local coordinates Fa at a we choose Fas:=(τsA)-1∘Fa where τsA is a Hyperbolic translation along the geodesic Exp0D(sA) by distance s|A|g(0). Specifically Exp0D(sA)=2Atanh(s/2):=γ(s) and τsA is given explicitly by (see e.g. [19]) where all norms and inner-products appearing above are Euclidean. In particular Thus for p close to a we have where we abuse notation by writing etc. In particular by (A.12) we see that for Fa coordinates x∈Dr, noting that we have ∇Aπλ·πλ≡0. For A a unit vector we collect the following rough estimates on Dr which can be checked by direct calculation - see also A.2.1. First of all |∇Ajλ,a|≤Cλ-1 which yields that where the final estimate above follows by suitably adapting the proof of Lemma A.1 where the only difference is that ∇Aj∼λ|x|-1∂λj and ∇Aπλ∼λ|x|-1∂λπλ. By appendix A.2.1 we also have |∇∇Aj|≤Cλ-1 which gives Thus we have on Ur(a). On the other hand, in Σ\Ur(a) it is straightforward to check, A.2.1 Computing ∇Aj Here we expand on the computations for ∇Aj - we note that the precise form of the derivation below is not crucial for our needs, but we nevertheless give full details for the sake of the reader. We have noting for later that, and By definition we have Fas-1=Fa-1∘τsA and in the below we set x=Fa(p) so that Fas(p)=τs-A(x). Thus we have, by (A.19) and (A.20): Thus in particular for a constant vector C∈R2. B Derivatives of J We work with the Green’s function which solves If x, y are local coordinates near a∈Σ as in (2.1), then for z=x1+ix2, ζ=y1+iy2 where, from [17, Proposition 6.2, Remark 6.3], in any holomorphic coordinates (z,ζ) we have and {ϕj}j is an L2-orthonormal basis of holomorphic one-forms on Σ. In particular when Σ has genus γ we have J(a)=-2πcγ∑j|ϕj(a)|g2 where c1=1 and cγ=4 for γ≥2. Lemma B.1 With the set-up as above we have Proof We first note that where ⟨,⟩g and ∇AΣ are the metric and connection on Σ extended complex linearly to TCΣ respectively. In our local coordinates x we write ϕj=ϕ~jdz and A=Ai∂xi. Since ∇Σdz(0)=0 we have that ∇AΣϕj(0)=(Ai∂xiϕ~j(0))dz. Noting also that Ai∂xi∂2Ga∂ζ¯∂z(x,0)=-π(Ai∂xiϕ~j(x))ϕ~j¯(0) and |dz(0)|g2=2cγ-1, gives since for x≠y, by (2.2) ∂2Ga∂ζ¯∂z(x,y)=14(∂x1-i∂x2)(∂y1+i∂y2)Ja(x,y). □ C First and second variation of Eα We briefly derive the first and second variations of Eα along maps u∈W1,2α(Σ,N) assuming N↪Rd is closed and isometrically embedded and Σ is any closed Riemannian manifold. Let Ων(N) denote a small ν-neighbourhood of N in Rd where ν>0 is small enough so that P:Ωδ(N)→N, the nearest point projection, is well-defined and smooth. We note the following easy to check facts concerning the behaviour of P on N. Specifically given any X,Y∈Rd and k∈N: where of course X⊥=X-X⊤ is the projection onto VkN and A(k)(X⊤,Y⊤):=∇⊤dY⊤⊥. For a round sphere A(k)(X⊤,Y⊤)=-k⟨X⊤,Y⊤⟩. C.1 First variation and regularity of solutions For u∈W1,2α(Σ,Sn), V∈W1,2α(Σ,Rd) so that V(x)∈Tu(x)N almost everywhere we have meaning that α-harmonic maps are weak solutions to divΣ((2+|∇u|2)α-1∇u)⊤=0 where ‘’ is projection onto Tu(x)N. Since α-harmonic maps are always smooth ( [24, Proposition 2.3]) in particular they are solutions to Let D1⊂R2 be a unit Euclidean ball equipped with a metric g=fg0 and g0 is the standard Euclidean metric. If u∈W1,2α(D,N) is α-harmonic then (with all operators below now being Euclidean) We first recall the standard ε-regularity estimate for systems like the above when α>1. For convenience we introduce an extra constant ν, reflecting that we will want to apply these estimates for a re-scaled version of an α-harmonic map, u^(x)=u(λ-1x), however this has no consequence on the result which can be easily adapted from [24, Section 3]: Theorem C.1 Let u∈W1,2α(D,N), ν>0, f∈C∞(D¯,R>0) and N↪Rd a smooth closed Riemannian manifold isometrically embedded in some Euclidean space. There exists ε0=ε0(N)>0, α0=α0>1 and K(N,‖logf‖C1)<∞ so that for all 1<α≤α0 if u weakly solves and ∫D|∇u|2<ε0 then In particular the above implies: Corollary C.2 Let Σ be as in Definition 1.1 and N↪Rd a smooth closed Riemannian manifold isometrically embedded in some Euclidean space. Let r0, α0>1 and K(N)<∞ so that if u:Σ→N is α-harmonic, 1<α<α0 and ∫Br(a)|∇u|2<ε0 then C.2 Second variation Given u∈W1,2α(Σ,N) and perturbations V,W∈W1,2α(Σ,Rd) so that V(x),W(x)∈Tu(x)N almost everywhere we consider us,t=P(u+sV+tW) and do a Taylor expansion about s,t=0 using (C.1)–(C.2): us,t=u+tW+sV+stA(u)(V,W)+O(t2)+O(s2). Thus Using ∇u·A(u)(V,W)=0 we see ∇u·∇(A(u)(V,W))=div(∇u·A(V,W))-Δu·A(u)(V,W) and since we have (Δu)⊥=A(u)(∇u,∇u) weakly, we have Of course when N=Sn is a round sphere and u∈W1,2α(Σ,Sn), V,W∈W1,2α(Σ,Rd) so that V(x),W(x)∈Tz(x)N almost everywhere: Data availability This manuscript has no associated data. Declarations Conflict of interest The author has no conflict of interest to declare that are relevant to the content of this article. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement This manuscript has no associated data.