Introduction and main theorem
Set S 1 = R / ( 2 π Z ) \mathbb{S}^1 = \mathbb{R}/(2\pi\mathbb{Z}) S 1 = R / ( 2 π Z ) , and equip S 3 = { z ∈ R 4 : ∣ z ∣ = 1 } \mathbb{S}^3 = \{z \in\mathbb{R}^4 : \lvert z\rvert= 1\} S 3 = { z ∈ R 4 : ∣ z ∣ = 1 } with the round metric g 0 g_0 g 0 . For an ordered pair Γ = ( γ 1 , γ 2 ) \Gamma= (\gamma_1,\gamma_2) Γ = ( γ 1 , γ 2 ) of rectifiable closed curves with disjoint component images, the Möbius cross energy is
E ( Γ ) = ∫ γ 1 ∫ γ 2 d ℓ 1 ( x ) d ℓ 2 ( y ) ∣ x − y ∣ 2 , (1) E(\Gamma) = \int_{\gamma_1}\int_{\gamma_2} \frac{\mathrm{d}\ell_1(x)\,\mathrm{d}\ell_2(y)}{\lvert x-y\rvert^2}, \tag*{(1)} E ( Γ ) = ∫ γ 1 ∫ γ 2 ∣ x − y ∣ 2 d ℓ 1 ( x ) d ℓ 2 ( y ) , (1)
where ∣ x − y ∣ \lvert x-y\rvert ∣ x − y ∣ is the Euclidean chordal distance in R 4 \mathbb{R}^4 R 4 . Integration along each parametrized curve counts multiplicity. The energy is invariant under separate reparametrizations and under Möbius transformations of S 3 \mathbb{S}^3 S 3 . A pair of disjoint compact connected sets is called split if a smoothly embedded 2 2 2 -sphere, disjoint from both sets, separates them; otherwise it is called non-split . For a curve pair, this terminology refers to its component images.
The standard Hopf link is
H = ( x , y ) , x ( s ) = ( cos s , sin s , 0 , 0 ) , y ( t ) = ( 0 , 0 , cos t , sin t ) . (2) H = (x,y), \qquad x(s) = (\cos s,\sin s,0,0), \qquad y(t) = (0,0,\cos t,\sin t). \tag*{(2)} H = ( x , y ) , x ( s ) = ( cos s , sin s , 0 , 0 ) , y ( t ) = ( 0 , 0 , cos t , sin t ) . (2)
Its components are unit-speed great circles in orthogonal 2 2 2 -planes. Since ∣ x ( s ) − y ( t ) ∣ 2 = 2 \lvert x(s)-y(t)\rvert^2 = 2 ∣ x ( s ) − y ( t ) ∣ 2 = 2 ,
E ( H ) = 2 π 2 . (3) E(H) = 2\pi^2. \tag*{(3)} E ( H ) = 2 π 2 . (3)
Freedman, He, and Wang [2 ] conjectured that this is the minimum cross energy among all non-split two-component links. An important step was taken by He [4 ] , who proved that a global minimizer exists and that every such minimizer is ambiently isotopic to the Hopf link. Agol, Marques, and Neves subsequently proved the sharp lower bound and classified all equality configurations [1 ] .
Theorem 1.1 (Agol–Marques–Neves). Every non-split two-component rectifiable link in S 3 \mathbb{S}^3 S 3 satisfies
E ( Γ ) ≥ 2 π 2 . E(\Gamma) \ge2\pi^2. E ( Γ ) ≥ 2 π 2 .
Equality holds if and only if a conformal transformation carries the pair to the standard Hopf link, up to separate reparametrizations and orientations.
2020 Mathematics Subject Classification. Primary 53A30, 57K10; Secondary 49Q10, 58E12.
Key words and phrases. Möbius cross energy, Hopf link, critical-value gap.
Choose a stereographic projection point outside the component images, and write Γ ^ = ( γ ^ 1 , γ ^ 2 ) \widehat{\Gamma}=(\widehat{\gamma}_1,\widehat{\gamma}_2) Γ = ( γ 1 , γ 2 ) for the Euclidean representative. The Gauss linking number of Γ \Gamma Γ is defined by
lk ( Γ ) = 1 4 π ∫ S 1 ∫ S 1 det ( γ ^ 1 ′ ( s ) , γ ^ 2 ′ ( t ) , γ ^ 1 ( s ) − γ ^ 2 ( t ) ) ∣ γ ^ 1 ( s ) − γ ^ 2 ( t ) ∣ 3 d s d t . (4) \operatorname{lk}(\Gamma)=\frac{1}{4\pi}\int_{\mathbb{S}^1}\int_{\mathbb{S}^1}\frac{\det\left(\widehat{\gamma}_1'(s),\widehat{\gamma}_2'(t),\widehat{\gamma}_1(s)-\widehat{\gamma}_2(t)\right)}{\left|\widehat{\gamma}_1(s)-\widehat{\gamma}_2(t)\right|^3}\,\mathrm{d}s\,\mathrm{d}t. \tag*{(4)} lk ( Γ ) = 4 π 1 ∫ S 1 ∫ S 1 ∣ γ 1 ( s ) − γ 2 ( t ) ∣ 3 det ( γ 1 ′ ( s ) , γ 2 ′ ( t ) , γ 1 ( s ) − γ 2 ( t ) ) d s d t . (4)
It is the sum of the signed over-crossings of the first component over the second in a regular oriented projection. Therefore a nonzero linking number implies non-splitness, but the converse fails, as illustrated by the Whitehead link.
A direct calculus implies
E ( Γ ) ≥ 4 π ∣ lk ( Γ ) ∣ . E(\Gamma)\geq4\pi\left|\operatorname{lk}(\Gamma)\right|. E ( Γ ) ≥ 4 π ∣ lk ( Γ ) ∣ .
Allowing a choice of component orientations, He [4 ] shows that a pair of non-split curves is link-homotopic to the Hopf link provided E ( Γ ) < 8 π E(\Gamma)<8\pi E ( Γ ) < 8 π .
The present paper concerns critical points rather than minimizers. The equality classification in (1) does not by itself exclude non-minimizing critical pairs with energies arbitrarily close to 2 π 2 2\pi^2 2 π 2 . We prove that no such sequence exists. To state the result precisely, consider the extrinsic Sobolev space
H 2 ( S 1 , S 3 ) = { γ ∈ H 2 ( S 1 , R 4 ) : ∣ γ ∣ = 1 a.e. } . H^2(\mathbb{S}^1,\mathbb{S}^3)=\left\{\gamma\in H^2(\mathbb{S}^1,\mathbb{R}^4):|\gamma|=1\ \text{a.e.}\right\}. H 2 ( S 1 , S 3 ) = { γ ∈ H 2 ( S 1 , R 4 ) : ∣ γ ∣ = 1 a.e. } .
The embedding H 2 ( S 1 ) ↪ C 1 , 1 / 2 ( S 1 ) H^2(\mathbb{S}^1)\hookrightarrow C^{1,1/2}(\mathbb{S}^1) H 2 ( S 1 ) ↪ C 1 , 1/2 ( S 1 ) provides a unique C 1 , 1 / 2 C^{1,1/2} C 1 , 1/2 representative, which we use throughout. A closed curve γ \gamma γ is regular if min S 1 ∣ γ ′ ∣ > 0 \min_{\mathbb{S}^1}|\gamma'|>0 min S 1 ∣ γ ′ ∣ > 0 . Define
A = { ( γ 1 , γ 2 ) : γ 1 , γ 2 are regular curves in H 2 ( S 1 , S 3 ) s.t. γ 1 ( S 1 ) ∩ γ 2 ( S 1 ) = ∅ } . (5) \mathcal{A}=\left\{(\gamma_1,\gamma_2):\ \gamma_1,\gamma_2\ \text{are regular curves in }H^2(\mathbb{S}^1,\mathbb{S}^3)\ \text{s.t. }\gamma_1(\mathbb{S}^1)\cap\gamma_2(\mathbb{S}^1)=\varnothing\right\}. \tag*{(5)} A = { ( γ 1 , γ 2 ) : γ 1 , γ 2 are regular curves in H 2 ( S 1 , S 3 ) s.t. γ 1 ( S 1 ) ∩ γ 2 ( S 1 ) = ∅ } . (5)
No injectivity of the individual parametrizations is assumed. The set A \mathcal{A} A is open in the H 2 H^2 H 2 mapping manifold, and E E E is smooth on A \mathcal{A} A . A pair is weakly critical if δ E Γ = 0 \delta E_\Gamma=0 δ E Γ = 0 on every H 2 H^2 H 2 tangent variation; explicitly, the admissible fields ξ i ∈ H 2 ( S 1 , R 4 ) \xi_i\in H^2(\mathbb{S}^1,\mathbb{R}^4) ξ i ∈ H 2 ( S 1 , R 4 ) satisfy ⟨ ξ i , γ i ⟩ = 0 \langle\xi_i,\gamma_i\rangle=0 ⟨ ξ i , γ i ⟩ = 0 everywhere.
Let Diff 2 ( S 1 ) \operatorname{Diff}^2(\mathbb{S}^1) Diff 2 ( S 1 ) denote the H 2 H^2 H 2 circle maps that are C 1 C^1 C 1 diffeomorphisms, allowing both orientations, and set
G = M o ¨ b + ( S 3 ) × Diff 2 ( S 1 ) × Diff 2 ( S 1 ) . (6) \mathcal{G}=\operatorname{M\ddot{o}b}^{+}(\mathbb{S}^3)\times\operatorname{Diff}^2(\mathbb{S}^1)\times\operatorname{Diff}^2(\mathbb{S}^1). \tag*{(6)} G = M o ¨ b + ( S 3 ) × Diff 2 ( S 1 ) × Diff 2 ( S 1 ) . (6)
We write G ⋅ H \mathcal{G}\cdot H G ⋅ H for the pairs obtained from H H H by a common orientation-preserving Möbius transformation and separate reparametrizations.2
Theorem 1.2 (Critical-value gap). There exists ε 0 > 0 \varepsilon_0>0 ε 0 > 0 such that every non-split, weakly critical pair Γ ∈ A \Gamma\in\mathcal{A} Γ ∈ A satisfying
E ( Γ ) ≤ 2 π 2 + ε 0 E(\Gamma)\leq2\pi^2+\varepsilon_0 E ( Γ ) ≤ 2 π 2 + ε 0
belongs to G ⋅ H \mathcal{G}\cdot H G ⋅ H . In particular, E ( Γ ) = 2 π 2 E(\Gamma)=2\pi^2 E ( Γ ) = 2 π 2 .
Remark 1.3. He [4 ] gave a geometric interpretation of critical points of the Möbius cross energy. With γ 2 \gamma_2 γ 2 fixed, define
ρ 2 ( x ) = ∫ γ 2 d ℓ 2 ( y ) ∣ x − y ∣ 2 . \rho_2(x)=\int_{\gamma_2}\frac{\mathrm{d}\ell_2(y)}{|x-y|^2}. ρ 2 ( x ) = ∫ γ 2 ∣ x − y ∣ 2 d ℓ 2 ( y ) .
Then the critical equation for γ 1 \gamma_1 γ 1 is precisely the geodesic equation for the conformal metric ρ 2 2 g 0 \rho_2^2g_0 ρ 2 2 g 0 on S 3 ∖ γ 2 \mathbb{S}^3\setminus\gamma_2 S 3 ∖ γ 2 , and similarly with the two components interchanged. Thus a critical pair can be viewed as a pair of mutually interacting closed geodesics in each other’s complement space equipped with a complete conformal metric.
Theorem 1.2 shows that 2 π 2 2\pi^2 2 π 2 is an isolated critical value of the Möbius cross energy among non-split pairs. Exact criticality is essential, since non-Möbius perturbations of the Hopf link can have energies arbitrarily close to 2 π 2 2\pi^2 2 π 2 . Although we work with regular H 2 H^2 H 2 pairs, the critical equation implies smoothness after constant-speed reparametrization.
Outline of the proof. The proof combines global compactness modulo Möbius transformations with local nondegeneracy of the Hopf orbit. Energy bounds and the critical equation yield smooth compactness, so any critical sequence with energy tending to 2 π 2 2\pi^2 2 π 2 converges, modulo symmetries, to a Hopf link. We then compute the Hopf Hessian, identify its kernel with the infinitesimal Möbius motions, and prove coercivity on the transverse space. A local Möbius slice and a nonlinear Taylor estimate give local uniqueness, which together with compactness proves the gap theorem.
Acknowledgments: The authors thank Professor Jian Ge for useful discussions. The first author is supported by the China Postdoctoral Science Foundation (2026M793350). The second author is supported by the Open Project Program of the Key Laboratory of Mathematics and Complex System at Beijing Normal University (Grant No. 202501). The principal arguments and the core of the paper were developed by the second author while visiting SIMIS.
AI assistance declaration: The main ideas and most proofs are solely the work of the authors. ChatGPT (version 5.6sol) was used only to assist in proving Proposition 3.1 and to aid in manuscript preparation. The authors have verified all contributions and assume complete accountability for the paper.
First variation and the critical equation
We derive the first variation directly in the spherical model, including its weak H 2 H^2 H 2 formulation. The corresponding Euclidean computation appears in the proof of [4 ] , Lemma 3.1, equations (3-2)-(3-3).
Analytic preliminaries
We denote D D D the Euclidean connection on R 4 \mathbb{R}^4 R 4 and ∇ \nabla ∇ the Levi-Civita connection of g 0 g_0 g 0 on S 3 \mathbb{S}^3 S 3 . Let X , Y X,Y X , Y be tangent vector fields on S 3 \mathbb{S}^3 S 3 , then for each x ∈ S 3 x \in\mathbb{S}^3 x ∈ S 3 , the Gauss formula gives
D X Y = ∇ X Y + I I ( X , Y ) , I I ( X , Y ) = − ⟨ X , Y ⟩ g 0 x , (7) D_XY = \nabla_XY + II(X,Y), \qquad II(X,Y) = -\langle X,Y\rangle_{g_0}x, \tag*{(7)} D X Y = ∇ X Y + I I ( X , Y ) , I I ( X , Y ) = − ⟨ X , Y ⟩ g 0 x , (7)
where I I II I I is the second fundamental form of S 3 ⊂ R 4 \mathbb{S}^3 \subset\mathbb{R}^4 S 3 ⊂ R 4 . Thus, for a unit-speed curve, γ ′ ′ = κ − γ \gamma'' = \kappa- \gamma γ ′′ = κ − γ , where κ = ∇ X X \kappa= \nabla_XX κ = ∇ X X is its curvature vector in S 3 \mathbb{S}^3 S 3 . S 1 = R / ( 2 π Z ) \mathbb{S}^1 = \mathbb{R}/(2\pi\mathbb{Z}) S 1 = R / ( 2 π Z ) , without averaging the integrals.
For a variation Γ τ \Gamma_\tau Γ τ with initial velocity ξ \xi ξ , we write δ E Γ [ ξ ] = d d τ E ( Γ τ ) ∣ τ = 0 \delta E_\Gamma[\xi] = \left.\frac{d}{d\tau}E(\Gamma_\tau)\right|_{\tau=0} δ E Γ [ ξ ] = d τ d E ( Γ τ ) τ = 0 and δ 2 \delta^2 δ 2 for the second differential in a fixed chart. Ordinary coordinate derivatives are denoted by ∂ \partial ∂ or d d d . The parameter (configuration) torus is T 2 = S s 1 × S t 1 \mathbb{T}^2 = \mathbb{S}^1_s \times\mathbb{S}^1_t T 2 = S s 1 × S t 1 . Throughout the paper,
∬ F ( s , t ) d s d t : = ∫ 0 2 π ∫ 0 2 π F ( s , t ) d s d t \iint F(s,t)\,ds\,dt := \int_0^{2\pi}\int_0^{2\pi} F(s,t)\,ds\,dt ∬ F ( s , t ) d s d t := ∫ 0 2 π ∫ 0 2 π F ( s , t ) d s d t
always refers to this fixed torus with the product parameter measure. Integrals over the component curves or other parameter intervals are written with explicit domains.
Lemma 2.1. Every regular closed H 2 H^2 H 2 curve admits a constant-speed parametrization obtained by an orientation-preserving H 2 H^2 H 2 circle diffeomorphism. Separate such reparametrizations preserve the cross energy, non-splitness, and weak criticality. Proof. For a lift of the parameter to [ 0 , 2 π ] [0,2\pi] [ 0 , 2 π ] , set
θ ( s ) = 2 π L ( γ ) ∫ 0 s ∣ γ ′ ( r ) ∣ d r , s ∈ [ 0 , 2 π ] . \theta(s)=\frac{2\pi}{L(\gamma)}\int_{0}^{s}|\gamma'(r)|\,\mathrm{d}r,\qquad s\in[0,2\pi]. θ ( s ) = L ( γ ) 2 π ∫ 0 s ∣ γ ′ ( r ) ∣ d r , s ∈ [ 0 , 2 π ] .
Since γ ′ ∈ H 1 ↪ C 0 \gamma'\in H^1\hookrightarrow C^0 γ ′ ∈ H 1 ↪ C 0 and its image stays away from the origin, composition with the Euclidean norm gives ∣ γ ′ ∣ ∈ H 1 |\gamma'|\in H^1 ∣ γ ′ ∣ ∈ H 1 . Consequently θ ∈ H 2 \theta\in H^2 θ ∈ H 2 , and θ ′ = 2 π ∣ γ ′ ∣ / L ( γ ) \theta'=2\pi|\gamma'|/L(\gamma) θ ′ = 2 π ∣ γ ′ ∣/ L ( γ ) is continuous and bounded below by a positive constant. The identity θ ( s + 2 π ) = θ ( s ) + 2 π \theta(s+2\pi)=\theta(s)+2\pi θ ( s + 2 π ) = θ ( s ) + 2 π makes θ \theta θ a degree-one C 1 C^1 C 1 circle diffeomorphism. Denote its inverse by ψ = θ − 1 \psi=\theta^{-1} ψ = θ − 1 . Almost everywhere,
ψ ′ = 1 θ ′ ∘ ψ , ψ ′ ′ = − θ ′ ′ ∘ ψ ( θ ′ ∘ ψ ) 3 . \psi'=\frac{1}{\theta'\circ\psi},\qquad\psi''=-\frac{\theta''\circ\psi}{(\theta'\circ\psi)^3}. ψ ′ = θ ′ ∘ ψ 1 , ψ ′′ = − ( θ ′ ∘ ψ ) 3 θ ′′ ∘ ψ .
Change of variables gives ψ ′ ′ ∈ L 2 \psi''\in L^2 ψ ′′ ∈ L 2 , and hence ψ ∈ H 2 \psi\in H^2 ψ ∈ H 2 . The same chain rule shows that γ ˉ = γ ∘ ψ ∈ H 2 \bar{\gamma}=\gamma\circ\psi\in H^2 γ ˉ = γ ∘ ψ ∈ H 2 and ∣ γ ˉ ′ ∣ = L ( γ ) / ( 2 π ) |\bar{\gamma}'|=L(\gamma)/(2\pi) ∣ γ ˉ ′ ∣ = L ( γ ) / ( 2 π ) .
Changing variables separately in each of the two integrals proves the energy invariance, since the curve images remain unchanged, the non-split property is also preserved. Composition by each fixed H 2 H^2 H 2 diffeomorphism preserves H 2 H^2 H 2 tangent fields. Pulling back a variation and differentiating energy invariance proves preservation of weak criticality in both directions. A fixed smooth Möbius transformation preserves weak criticality by the same argument. □
Lemma 2.2. The energy E : A → R E:\mathcal{A}\to\mathbb{R} E : A → R is C ∞ C^\infty C ∞ . Its first differential extends to a continuous linear functional on H 1 H^1 H 1 tangent fields. In fixed smooth pointwise charts, its second differential extends to a bounded bilinear form on H 1 × H 1 H^1\times H^1 H 1 × H 1 , uniformly when the curve values and first derivatives stay in a compact region with positive speed and positive component separation.
Proof. First work in the ambient space of curve pairs. The evaluation map
z Γ ( s , t ) = ( γ 1 ( s ) , γ 2 ( t ) , γ 1 ′ ( s ) , γ 2 ′ ( t ) ) z_{\Gamma}(s,t)=(\gamma_1(s),\gamma_2(t),\gamma_1'(s),\gamma_2'(t)) z Γ ( s , t ) = ( γ 1 ( s ) , γ 2 ( t ) , γ 1 ′ ( s ) , γ 2 ′ ( t ))
is continuous from H 2 ⊕ H 2 H^2\oplus H^2 H 2 ⊕ H 2 to C 0 C^0 C 0 on the parameter torus. Set
Ω = { ( p , q , a , b ) : p ≠ q , a ≠ 0 , b ≠ 0 } , Φ ( p , q , a , b ) = ∣ a ∣ ∣ b ∣ ∣ p − q ∣ 2 ∈ C ∞ ( Ω ) . \Omega=\{(p,q,a,b):p\ne q,\ a\ne0,\ b\ne0\},\qquad
\Phi(p,q,a,b)=\frac{|a|\,|b|}{|p-q|^2}\in C^\infty(\Omega). Ω = {( p , q , a , b ) : p = q , a = 0 , b = 0 } , Φ ( p , q , a , b ) = ∣ p − q ∣ 2 ∣ a ∣ ∣ b ∣ ∈ C ∞ ( Ω ) .
The tuple z Γ z_{\Gamma} z Γ takes values in a compact subset of Ω \Omega Ω , and this remains true on a sufficiently small H 2 H^2 H 2 neighborhood. Hence, the map u ↦ Φ ∘ u u\mapsto\Phi\circ u u ↦ Φ ∘ u is C ∞ C^\infty C ∞ on the corresponding C 0 C^0 C 0 neighborhood, with derivatives obtained pointwise from the finite-dimensional derivatives of Φ \Phi Φ . Therefore
E ( γ 1 , γ 2 ) = ∫ 0 2 π ∫ 0 2 π Φ ( z Γ ( s , t ) ) d s d t . E(\gamma_1,\gamma_2)=\int_{0}^{2\pi}\int_{0}^{2\pi}\Phi(z_{\Gamma}(s,t))\,\mathrm{d}s\,\mathrm{d}t. E ( γ 1 , γ 2 ) = ∫ 0 2 π ∫ 0 2 π Φ ( z Γ ( s , t )) d s d t .
is C ∞ C^\infty C ∞ . Every term in the first derivative δ E ( γ 1 , γ 2 ) [ w ] \delta E(\gamma_1,\gamma_2)[w] δ E ( γ 1 , γ 2 ) [ w ] contains at most one derivative of a variation field w = ( w 1 , w 2 ) w=(w_1,w_2) w = ( w 1 , w 2 ) , and every term in the second derivative δ 2 E ( γ 1 , γ 2 ) [ w , w ~ ] \delta^2E(\gamma_1,\gamma_2)[w,\widetilde{w}] δ 2 E ( γ 1 , γ 2 ) [ w , w ] contains at most one derivative of each of the two fields w = ( w 1 , w 2 ) w=(w_1,w_2) w = ( w 1 , w 2 ) , w ~ = ( w ~ 1 , w ~ 2 ) \widetilde{w}=(\widetilde{w}_1,\widetilde{w}_2) w = ( w 1 , w 2 ) . Then ∥ f ∥ L 1 ≤ ( 2 π ) 1 / 2 ∥ f ∥ L 2 \|f\|_{L^1}\le(2\pi)^{1/2}\|f\|_{L^2} ∥ f ∥ L 1 ≤ ( 2 π ) 1/2 ∥ f ∥ L 2 yields
∫ 0 2 π ∫ 0 2 π f w i w ~ i ′ d s d t ≤ ∥ f ∥ L ∞ ∥ w i ∥ L 2 ∥ w ~ i ′ ∥ L 2 , \int_{0}^{2\pi}\int_{0}^{2\pi}f w_i\widetilde{w}_i'\,\mathrm{d}s\,\mathrm{d}t
\le\|f\|_{L^\infty}\|w_i\|_{L^2}\|\widetilde{w}_i'\|_{L^2}, ∫ 0 2 π ∫ 0 2 π f w i w i ′ d s d t ≤ ∥ f ∥ L ∞ ∥ w i ∥ L 2 ∥ w i ′ ∥ L 2 ,
∫ 0 2 π ∫ 0 2 π f w i w ~ j ′ d s d t ≤ ∥ f ∥ L ∞ ∥ w i ∥ L 1 ∥ w ~ j ′ ∥ L 1 ≤ C ∥ w i ∥ L 2 ∥ w ~ j ∥ H 1 . \int_{0}^{2\pi}\int_{0}^{2\pi}f w_i\widetilde{w}_j'\,\mathrm{d}s\,\mathrm{d}t
\le\|f\|_{L^\infty}\|w_i\|_{L^1}\|\widetilde{w}_j'\|_{L^1}
\le C\|w_i\|_{L^2}\|\widetilde{w}_j\|_{H^1}. ∫ 0 2 π ∫ 0 2 π f w i w j ′ d s d t ≤ ∥ f ∥ L ∞ ∥ w i ∥ L 1 ∥ w j ′ ∥ L 1 ≤ C ∥ w i ∥ L 2 ∥ w j ∥ H 1 .
Here ∥ f ∥ L ∞ \|f\|_{L^\infty} ∥ f ∥ L ∞ bounds derivatives of Φ \Phi Φ on z Γ z_{\Gamma} z Γ and by the structure of Φ \Phi Φ its singular factors are exactly ∥ f ∥ L ∞ ≤ C \|f\|_{L^\infty}\le C ∥ f ∥ L ∞ ≤ C on a uniformly regular, uniformly separated neighborhood. □
First variation of the Möbius cross energy
Let Γ = ( γ 1 , γ 2 ) ∈ A \Gamma=(\gamma_1,\gamma_2)\in\mathcal{A} Γ = ( γ 1 , γ 2 ) ∈ A be a regular H 2 H^2 H 2 pair with disjoint component images in S 3 \mathbb{S}^3 S 3 . Write T i = ∂ ℓ i γ i T_i=\partial_{\ell_i}\gamma_i T i = ∂ ℓ i γ i and κ i = ∇ T i T i \kappa_i=\nabla_{T_i}T_i κ i = ∇ T i T i . The projection onto the curve-normal plane inside T γ i S 3 T_{\gamma_i}\mathbb{S}^3 T γ i S 3 is
P N i = Id − γ i ⊗ γ i − T i ⊗ T i . (8) P_{N_i}=\operatorname{Id}-\gamma_i\otimes\gamma_i-T_i\otimes T_i. \tag*{(8)} P N i = Id − γ i ⊗ γ i − T i ⊗ T i . (8)
Define the positive interaction potentials by
ρ 1 ( y ) = ∫ γ 1 d ℓ 1 ( x ) ∣ x − y ∣ 2 , ρ 2 ( x ) = ∫ γ 2 d ℓ 2 ( y ) ∣ x − y ∣ 2 . (9) \rho_1(y)=\int_{\gamma_1}\frac{\mathrm{d}\ell_1(x)}{\lvert x-y\rvert^2},\qquad
\rho_2(x)=\int_{\gamma_2}\frac{\mathrm{d}\ell_2(y)}{\lvert x-y\rvert^2}. \tag*{(9)} ρ 1 ( y ) = ∫ γ 1 ∣ x − y ∣ 2 d ℓ 1 ( x ) , ρ 2 ( x ) = ∫ γ 2 ∣ x − y ∣ 2 d ℓ 2 ( y ) . (9)
For K ( x , y ) = ∣ x − y ∣ − 2 K(x,y)=\lvert x-y\rvert^{-2} K ( x , y ) = ∣ x − y ∣ − 2 , spherical differentiation yields
grad x K = 2 ( y − ⟨ x , y ⟩ x ) r 4 , grad y K = 2 ( x − ⟨ x , y ⟩ y ) r 4 . (10) \operatorname{grad}_x K=\frac{2(y-\langle x,y\rangle x)}{r^4},\qquad
\operatorname{grad}_y K=\frac{2(x-\langle x,y\rangle y)}{r^4}. \tag*{(10)} grad x K = r 4 2 ( y − ⟨ x , y ⟩ x ) , grad y K = r 4 2 ( x − ⟨ x , y ⟩ y ) . (10)
In particular, P N 1 grad x K = 2 P N 1 y / r 4 P_{N_1}\operatorname{grad}_x K=2P_{N_1}y/r^4 P N 1 grad x K = 2 P N 1 y / r 4 , and similarly for the second component.
Proposition 2.3. For every Γ ∈ A \Gamma\in\mathcal{A} Γ ∈ A , define the L 2 L^2 L 2 normal fields
F 1 = P N 1 grad ρ 2 − ρ 2 κ 1 , F 2 = P N 2 grad ρ 1 − ρ 1 κ 2 . F_1=P_{N_1}\operatorname{grad}\rho_2-\rho_2\kappa_1,\qquad
F_2=P_{N_2}\operatorname{grad}\rho_1-\rho_1\kappa_2. F 1 = P N 1 grad ρ 2 − ρ 2 κ 1 , F 2 = P N 2 grad ρ 1 − ρ 1 κ 2 .
For every H 1 H^1 H 1 sphere-tangent field ξ = ( ξ 1 , ξ 2 ) \xi=(\xi_1,\xi_2) ξ = ( ξ 1 , ξ 2 ) , the extended first variation is
δ E Γ [ ξ ] = ∫ γ 1 ⟨ F 1 , ξ 1 ⟩ d ℓ 1 + ∫ γ 2 ⟨ F 2 , ξ 2 ⟩ d ℓ 2 . (11) \delta E_{\Gamma}[\xi]=\int_{\gamma_1}\langle F_1,\xi_1\rangle\,\mathrm{d}\ell_1+\int_{\gamma_2}\langle F_2,\xi_2\rangle\,\mathrm{d}\ell_2. \tag*{(11)} δ E Γ [ ξ ] = ∫ γ 1 ⟨ F 1 , ξ 1 ⟩ d ℓ 1 + ∫ γ 2 ⟨ F 2 , ξ 2 ⟩ d ℓ 2 . (11)
Consequently, weak criticality is equivalent to
κ 1 = P N 1 grad log ρ 2 , κ 2 = P N 2 grad log ρ 1 (12) \kappa_1=P_{N_1}\operatorname{grad}\log\rho_2,\qquad
\kappa_2=P_{N_2}\operatorname{grad}\log\rho_1 \tag*{(12)} κ 1 = P N 1 grad log ρ 2 , κ 2 = P N 2 grad log ρ 1 (12)
almost everywhere.
Proof. We write
E ( Γ ) = ∫ γ i ρ j d ℓ i , i , j = 1 , 2 , i ≠ j . E(\Gamma)=\int_{\gamma_i}\rho_j\,\mathrm{d}\ell_i,\qquad i,j=1,2,\quad i\ne j. E ( Γ ) = ∫ γ i ρ j d ℓ i , i , j = 1 , 2 , i = j .
For an H 2 H^2 H 2 sphere-tangent variation field ξ i \xi_i ξ i , differentiation and integration by parts along the closed curve give
δ E Γ [ ξ i ] = ∫ γ i ( ⟨ grad ρ j , ξ i ⟩ + ρ j ⟨ T i , ∇ T i ξ i ⟩ ) d ℓ i = ∫ γ i ⟨ grad ρ j − ∇ T i ( ρ j T i ) , ξ i ⟩ d ℓ i = ∫ γ i ⟨ F i , ξ i ⟩ d ℓ i . \begin{aligned}
\delta E_{\Gamma}[\xi_i]
&=\int_{\gamma_i}\left(\langle\operatorname{grad}\rho_j,\xi_i\rangle+\rho_j\langle T_i,\nabla_{T_i}\xi_i\rangle\right)\,\mathrm{d}\ell_i\\
&=\int_{\gamma_i}\left\langle\operatorname{grad}\rho_j-\nabla_{T_i}(\rho_jT_i),\xi_i\right\rangle\,\mathrm{d}\ell_i\\
&=\int_{\gamma_i}\langle F_i,\xi_i\rangle\,\mathrm{d}\ell_i.
\end{aligned} δ E Γ [ ξ i ] = ∫ γ i ( ⟨ grad ρ j , ξ i ⟩ + ρ j ⟨ T i , ∇ T i ξ i ⟩ ) d ℓ i = ∫ γ i ⟨ grad ρ j − ∇ T i ( ρ j T i ) , ξ i ⟩ d ℓ i = ∫ γ i ⟨ F i , ξ i ⟩ d ℓ i .
where the last equality uses
∇ T i ( ρ j T i ) = ⟨ grad ρ j , T i ⟩ T i + ρ j κ i . \nabla_{T_i}(\rho_jT_i)=\langle\operatorname{grad}\rho_j,T_i\rangle T_i+\rho_j\kappa_i. ∇ T i ( ρ j T i ) = ⟨ grad ρ j , T i ⟩ T i + ρ j κ i .
In arclength coordinates, T i ∈ H 1 T_i\in H^1 T i ∈ H 1 and κ i ∈ L 2 \kappa_i\in L^2 κ i ∈ L 2 , so the integration by parts is valid in the Sobolev sense. The same formula holds for H 1 H^1 H 1 test fields and agrees with the extension in Lemma 2.2. Summing the two partial variations proves (2.5).
Given an arbitrary smooth periodic R 4 \mathbb{R}^4 R 4 -valued field φ \varphi φ , take
ξ i = ( Id − γ i ⊗ γ i ) φ ∈ H 2 . \xi_i=(\operatorname{Id}-\gamma_i\otimes\gamma_i)\varphi\in H^2. ξ i = ( Id − γ i ⊗ γ i ) φ ∈ H 2 .
This is an admissible sphere-tangent field. Since F i ⊥ γ i F_i \perp\gamma_i F i ⊥ γ i , weak criticality gives ∫ ⟨ F i , φ ⟩ d ℓ i = 0 \int\langle F_i,\varphi\rangle\,\mathrm{d}\ell_i = 0 ∫ ⟨ F i , φ ⟩ d ℓ i = 0 . The speed is bounded away from zero, so the fundamental lemma implies F i = 0 F_i = 0 F i = 0 almost everywhere. This argument does not require the curve-normal projection P N i P_{N_i} P N i to preserve H 2 H^2 H 2 test fields. Conversely, F i = 0 F_i = 0 F i = 0 makes every first variation vanish. Dividing by the positive potentials proves (2.6).
Global normalization and compactness of critical pairs
We make the point–diameter normalization in [4 ] quantitative on the sphere. The critical equation then converts the geometric bounds into smooth compactness.
Global Möbius gauge
Let N = ( 0 , 0 , 0 , 1 ) N = (0,0,0,1) N = ( 0 , 0 , 0 , 1 ) and let σ : S 3 ∖ { N } → R 3 \sigma:\mathbb{S}^3 \setminus\{N\} \to\mathbb{R}^3 σ : S 3 ∖ { N } → R 3 be stereographic projection. Its inverse satisfies
σ − 1 ( z ) = ( 2 z 1 + ∣ z ∣ 2 , ∣ z ∣ 2 − 1 1 + ∣ z ∣ 2 ) , ∣ σ − 1 ( z ) − σ − 1 ( w ) ∣ = 2 ∣ z − w ∣ ( 1 + ∣ z ∣ 2 ) ( 1 + ∣ w ∣ 2 ) . (13) \sigma^{-1}(z)=\left(\frac{2z}{1+|z|^2},\frac{|z|^2-1}{1+|z|^2}\right),\qquad\left|\sigma^{-1}(z)-\sigma^{-1}(w)\right|=\frac{2|z-w|}{\sqrt{(1+|z|^2)(1+|w|^2)}}. \tag*{(13)} σ − 1 ( z ) = ( 1 + ∣ z ∣ 2 2 z , 1 + ∣ z ∣ 2 ∣ z ∣ 2 − 1 ) , σ − 1 ( z ) − σ − 1 ( w ) = ( 1 + ∣ z ∣ 2 ) ( 1 + ∣ w ∣ 2 ) 2∣ z − w ∣ . (13)
Proposition 3.1. Fix 0 < Λ < ∞ 0 < \Lambda< \infty 0 < Λ < ∞ . Every non-split pair Γ = ( γ 1 , γ 2 ) ∈ A \Gamma=(\gamma_1,\gamma_2)\in\mathcal{A} Γ = ( γ 1 , γ 2 ) ∈ A satisfying E ( Γ ) ≤ Λ E(\Gamma)\leq\Lambda E ( Γ ) ≤ Λ has an orientation-preserving Möbius image for which
2 π 3 ≤ L ( γ i ) ≤ 6 Λ π , i = 1 , 2 , (14) \frac{2\pi}{3}\leq L(\gamma_i)\leq\frac{6\Lambda}{\pi},\qquad i=1,2, \tag*{(14)} 3 2 π ≤ L ( γ i ) ≤ π 6Λ , i = 1 , 2 , (14)
and
dist R 4 ( γ 1 , γ 2 ) ≥ d Λ > 0. (15) \operatorname{dist}_{\mathbb{R}^4}(\gamma_1,\gamma_2)\geq d_\Lambda>0. \tag*{(15)} dist R 4 ( γ 1 , γ 2 ) ≥ d Λ > 0. (15)
The constant d Λ d_\Lambda d Λ depends only on Λ \Lambda Λ .
Proof. Step 1: normalize a point and a diameter. Choose p ∈ γ 1 ( S 1 ) p\in\gamma_1(\mathbb{S}^1) p ∈ γ 1 ( S 1 ) and send it to N = ( 0 , 0 , 0 , 1 ) N=(0,0,0,1) N = ( 0 , 0 , 0 , 1 ) by an orientation-preserving Möbius transformation. Since the second component is disjoint from the first, it avoids N N N , and hence σ ( γ 2 ( S 1 ) ) ⊂ R 3 \sigma(\gamma_2(\mathbb{S}^1))\subset\mathbb{R}^3 σ ( γ 2 ( S 1 )) ⊂ R 3 is compact. It is nonsingleton because γ 2 \gamma_2 γ 2 is regular.
Choose
z 0 ∈ σ ( γ 2 ( S 1 ) ) , D 2 : = diam R 3 σ ( γ 2 ( S 1 ) ) > 0. z_0 \in\sigma(\gamma_2(\mathbb{S}^1)), \qquad D_2 := \operatorname{diam}_{\mathbb{R}^3}\sigma(\gamma_2(\mathbb{S}^1)) > 0. z 0 ∈ σ ( γ 2 ( S 1 )) , D 2 := diam R 3 σ ( γ 2 ( S 1 )) > 0.
The Euclidean similarity
z ↦ D 2 − 1 ( z − z 0 ) z \mapsto D_2^{-1}(z-z_0) z ↦ D 2 − 1 ( z − z 0 )
extends over N N N to an orientation-preserving Möbius transformation fixing N N N . After composition, we may therefore assume that
N ∈ γ 1 ( S 1 ) , 0 ∈ σ ( γ 2 ( S 1 ) ) , diam R 3 σ ( γ 2 ( S 1 ) ) = 1. (16) N \in\gamma_1(\mathbb{S}^1), \qquad0 \in\sigma(\gamma_2(\mathbb{S}^1)), \qquad\operatorname{diam}_{\mathbb{R}^3}\sigma(\gamma_2(\mathbb{S}^1))=1. \tag*{(16)} N ∈ γ 1 ( S 1 ) , 0 ∈ σ ( γ 2 ( S 1 )) , diam R 3 σ ( γ 2 ( S 1 )) = 1. (16)
Since the second image contains 0 0 0 and has Euclidean diameter one,
σ ( γ 2 ( S 1 ) ) ⊂ B ‾ 1 ( 0 ) . \sigma(\gamma_2(\mathbb{S}^1)) \subset\overline{B}_1(0). σ ( γ 2 ( S 1 )) ⊂ B 1 ( 0 ) .
Step 2: use non-splitness to obtain lower length bounds. We claim that
σ ( γ 1 ( S 1 ) ∖ { N } ) ∩ B ‾ 1 ( 0 ) ≠ ∅ . (17) \sigma(\gamma_1(\mathbb{S}^1)\setminus\{N\})\cap\overline{B}_1(0)\ne\varnothing. \tag*{(17)} σ ( γ 1 ( S 1 ) ∖ { N }) ∩ B 1 ( 0 ) = ∅ . (17)
Otherwise the two compact sets γ 1 ( S 1 ) \gamma_1(\mathbb{S}^1) γ 1 ( S 1 ) and σ − 1 ( B ‾ 1 ( 0 ) ) \sigma^{-1}(\overline{B}_1(0)) σ − 1 ( B 1 ( 0 )) would be disjoint in S 3 \mathbb{S}^3 S 3 . For R > 1 R>1 R > 1 sufficiently close to 1 1 1 , the larger closed ball σ − 1 ( B ‾ R ( 0 ) ) \sigma^{-1}(\overline{B}_R(0)) σ − 1 ( B R ( 0 )) would still avoid γ 1 \gamma_1 γ 1 . Its boundary would then be a smooth sphere separating the entire second component from the entire first component, including N N N . This contradicts non-splitness; see Figure 1 . It proves (17) .
Figure 1. The topological step in the point–diameter gauge. After normalization, the second component lies in the closed unit ball of the stereographic chart. If the first component missed that ball, a slightly larger round sphere would separate the components. The drawings are schematic projections of spatial configurations; crossings in the left panel indicate projection crossings, not intersections of the components.
We also use the elementary estimate
L ( η ) ≥ 2 diam S 3 η ( S 1 ) (18) L(\eta)\ge2\operatorname{diam}_{\mathbb{S}^3}\eta(\mathbb{S}^1) \tag*{(18)} L ( η ) ≥ 2 diam S 3 η ( S 1 ) (18)
for every closed rectifiable curve η : S 1 → S 3 \eta:\mathbb{S}^1\to\mathbb{S}^3 η : S 1 → S 3 by considering the two arcs joining a diameter-realizing pair. By (17) , choose q ∈ γ 1 ( S 1 ) q\in\gamma_1(\mathbb{S}^1) q ∈ γ 1 ( S 1 ) with
∣ σ ( q ) ∣ ≤ 1. |\sigma(q)|\le1. ∣ σ ( q ) ∣ ≤ 1.
Formula eq:3.1 implies
d S 3 ( N , q ) ≥ π 2 . d_{\mathbb{S}^3}(N,q)\ge\frac{\pi}{2}. d S 3 ( N , q ) ≥ 2 π .
Applying (18) gives
L ( γ 1 ) ≥ π . (19) L(\gamma_1)\ge\pi. \tag*{(19)} L ( γ 1 ) ≥ π . (19)
The compact set σ ( γ 2 ( S 1 ) ) \sigma(\gamma_2(\mathbb{S}^1)) σ ( γ 2 ( S 1 )) realizes its diameter, so there exist z , w z,w z , w in this set such that
Because ∣ z ∣ , ∣ w ∣ ≤ 1 |z|,|w|\le1 ∣ z ∣ , ∣ w ∣ ≤ 1 , formula eq:3.1 gives
∣ σ − 1 ( z ) − σ − 1 ( w ) ∣ = 2 ∣ z − w ∣ 1 + ∣ z ∣ 2 1 + ∣ w ∣ 2 ≥ 1. |\sigma^{-1}(z)-\sigma^{-1}(w)|=\frac{2|z-w|}{\sqrt{1+|z|^2}\sqrt{1+|w|^2}}\ge1. ∣ σ − 1 ( z ) − σ − 1 ( w ) ∣ = 1 + ∣ z ∣ 2 1 + ∣ w ∣ 2 2∣ z − w ∣ ≥ 1.
If d d d denotes their spherical distance, then
2 sin d 2 = ∣ σ − 1 ( z ) − σ − 1 ( w ) ∣ ≥ 1 , 2\sin\frac{d}{2}=|\sigma^{-1}(z)-\sigma^{-1}(w)|\ge1, 2 sin 2 d = ∣ σ − 1 ( z ) − σ − 1 ( w ) ∣ ≥ 1 ,
and hence d ≥ π / 3 d\ge\pi/3 d ≥ π /3 . Another application of (18) yields
L ( γ 2 ) ≥ 2 π 3 . (20) L(\gamma_2)\ge\frac{2\pi}{3}. \tag*{(20)} L ( γ 2 ) ≥ 3 2 π . (20)
Step 3: obtain upper length bounds from the energy. Since ∣ x − y ∣ 2 ≤ 4 |x-y|^2\le4 ∣ x − y ∣ 2 ≤ 4 for x , y ∈ S 3 x,y\in\mathbb{S}^3 x , y ∈ S 3 ,
E ( Γ ) = ∫ γ 1 ∫ γ 2 d ℓ 1 d ℓ 2 ∣ x − y ∣ 2 ≥ 1 4 L ( γ 1 ) L ( γ 2 ) . (21) E(\Gamma)=\int_{\gamma_1}\int_{\gamma_2}\frac{\mathrm{d}\ell_1\,\mathrm{d}\ell_2}{|x-y|^2}\ge\frac{1}{4}L(\gamma_1)L(\gamma_2). \tag*{(21)} E ( Γ ) = ∫ γ 1 ∫ γ 2 ∣ x − y ∣ 2 d ℓ 1 d ℓ 2 ≥ 4 1 L ( γ 1 ) L ( γ 2 ) . (21)
Combining E ( Γ ) ≤ Λ E(\Gamma) \le\Lambda E ( Γ ) ≤ Λ with (19) –(20) gives the stronger individual estimates
L ( γ 1 ) ≤ 4 Λ L ( γ 2 ) ≤ 6 Λ π , L ( γ 2 ) ≤ 4 Λ L ( γ 1 ) ≤ 4 Λ π . L(\gamma_1) \le\frac{4\Lambda}{L(\gamma_2)} \le\frac{6\Lambda}{\pi}, \qquad L(\gamma_2) \le\frac{4\Lambda}{L(\gamma_1)} \le\frac{4\Lambda}{\pi}. L ( γ 1 ) ≤ L ( γ 2 ) 4Λ ≤ π 6Λ , L ( γ 2 ) ≤ L ( γ 1 ) 4Λ ≤ π 4Λ .
Together with (19) –(20) , these imply the symmetric bounds eq:3.2 .
Step 4: exclude collision by the energy bound. It remains to obtain a uniform lower bound for the distance between the two components. Set
d : = dist R 4 ( γ 1 , γ 2 ) = min x ∈ γ 1 ( S 1 ) , y ∈ γ 2 ( S 1 ) ∣ x − y ∣ . d := \operatorname{dist}_{\mathbb{R}^{4}}(\gamma_1,\gamma_2) = \min_{\substack{x \in\gamma_1(\mathbb{S}^{1}),\, y \in\gamma_2(\mathbb{S}^{1})}} \lvert x-y\rvert. d := dist R 4 ( γ 1 , γ 2 ) = x ∈ γ 1 ( S 1 ) , y ∈ γ 2 ( S 1 ) min ∣ x − y ∣ .
For this estimate, use unit-speed arclength coordinates on
R / L ( γ i ) Z , i = 1 , 2 , \mathbb{R}/L(\gamma_i)\mathbb{Z}, \qquad i=1,2, R / L ( γ i ) Z , i = 1 , 2 ,
with the two origins chosen at a closest pair, so that
∣ γ 1 ( 0 ) − γ 2 ( 0 ) ∣ = d . \lvert\gamma_1(0)-\gamma_2(0)\rvert=d. ∣ γ 1 ( 0 ) − γ 2 ( 0 )∣ = d .
Put
ℓ : = 2 π 3 . \ell:= \frac{2\pi}{3}. ℓ := 3 2 π .
The lower length bounds ensure that both curves contain arclength intervals [ 0 , ℓ ] [0,\ell] [ 0 , ℓ ] . For 0 ≤ s , t ≤ ℓ 0 \le s,t \le\ell 0 ≤ s , t ≤ ℓ , the triangle inequality and the fact that chordal distance is bounded above by arclength give
∣ γ 1 ( s ) − γ 2 ( t ) ∣ ≤ ∣ γ 1 ( s ) − γ 1 ( 0 ) ∣ + ∣ γ 1 ( 0 ) − γ 2 ( 0 ) ∣ + ∣ γ 2 ( 0 ) − γ 2 ( t ) ∣ ≤ s + d + t . \lvert\gamma_1(s)-\gamma_2(t)\rvert\le\lvert\gamma_1(s)-\gamma_1(0)\rvert+\lvert\gamma_1(0)-\gamma_2(0)\rvert+\lvert\gamma_2(0)-\gamma_2(t)\rvert
\le s+d+t. ∣ γ 1 ( s ) − γ 2 ( t )∣ ≤ ∣ γ 1 ( s ) − γ 1 ( 0 )∣ + ∣ γ 1 ( 0 ) − γ 2 ( 0 )∣ + ∣ γ 2 ( 0 ) − γ 2 ( t )∣ ≤ s + d + t .
Restricting the energy integral to this parameter rectangle, we obtain
Λ ≥ E ( Γ ) ≥ ∫ 0 ℓ ∫ 0 ℓ d s d t ( d + s + t ) 2 = log ( d + ℓ ) 2 d ( d + 2 ℓ ) . \Lambda\ge E(\Gamma) \ge\int_{0}^{\ell}\int_{0}^{\ell}\frac{\mathrm{d}s\,\mathrm{d}t}{(d+s+t)^2}
=\log\frac{(d+\ell)^2}{d(d+2\ell)}. Λ ≥ E ( Γ ) ≥ ∫ 0 ℓ ∫ 0 ℓ ( d + s + t ) 2 d s d t = log d ( d + 2 ℓ ) ( d + ℓ ) 2 .
Solving this inequality gives an explicit choice:
d ≥ d Λ : = 2 π 3 ( ( 1 − e − Λ ) − 1 / 2 − 1 ) > 0. d \ge d_{\Lambda}:=\frac{2\pi}{3}\left((1-e^{-\Lambda})^{-1/2}-1\right)>0. d ≥ d Λ := 3 2 π ( ( 1 − e − Λ ) − 1/2 − 1 ) > 0.
This is the quantitative version of the logarithmic collision estimate used in [4 ] , proof of Lemma 2.2. □
Remark 3.2. [3 ] proved that any two disjoint linked Jordan curves A , B ⊂ S 3 A,B \subset\mathbb{S}^{3} A , B ⊂ S 3 satisfy
dist S 3 ( A , B ) ≤ π 2 . (22) \operatorname{dist}_{\mathbb{S}^{3}}(A,B)\le\frac{\pi}{2}. \tag*{(22)} dist S 3 ( A , B ) ≤ 2 π . (22)
with equality precisely for two great circles in orthogonal 2-planes, or equivalently, a Hopf link up to an ambient metric isometry.
Remark 3.3. The proposition uses non-splitness only in the separating-sphere argument and does not use criticality. It selects one geometrically nondegenerate representative in each bounded-energy Möbius orbit.
In the rest of this section, separation means chordal distance in R 4 \mathbb{R}^{4} R 4 . We use Proposition 2.3 to obtain regularity from weak criticality.
Proposition 3.4. The constant-speed representatives of every weakly critical pair in A \mathcal{A} A are smooth.
Proof. By Lemma 2.1, take
∣ γ i ′ ∣ ≡ a i = L i 2 π > 0 , T i = γ i ′ a i . \lvert\gamma_i' \rvert\equiv a_i = \frac{L_i}{2\pi} > 0,\qquad T_i = \frac{\gamma_i'}{a_i}. ∣ γ i ′ ∣ ≡ a i = 2 π L i > 0 , T i = a i γ i ′ .
Thus the curvature and γ i ′ ′ \gamma_i'' γ i ′′ satisfy
γ i ′ ′ = a i 2 ( κ i − γ i ) . \gamma_i'' = a_i^2(\kappa_i-\gamma_i). γ i ′′ = a i 2 ( κ i − γ i ) .
Combining this identity with (12) gives the following Euler-Lagrange equation:
γ i ′ ′ = a i 2 ( P N i G j ( γ i ) − γ i ) , G j = grad log ρ j , j ≠ i . (23) \gamma_i'' = a_i^2\left(P_{N_i}G_j(\gamma_i)-\gamma_i\right),\qquad G_j=\operatorname{grad}\log\rho_j,\qquad j\ne i. \tag*{(23)} γ i ′′ = a i 2 ( P N i G j ( γ i ) − γ i ) , G j = grad log ρ j , j = i . (23)
Here P N i P_{N_i} P N i is the projection onto the normal space.
Let d = dist R 4 ( γ 1 , γ 2 ) > 0 d=\operatorname{dist}_{\mathbb{R}^{4}}(\gamma_1,\gamma_2)>0 d = dist R 4 ( γ 1 , γ 2 ) > 0 . For z z z near γ i \gamma_i γ i and y ∈ γ j y\in\gamma_j y ∈ γ j , then ∣ z − y ∣ ≥ d / 2 \lvert z-y\rvert\ge d/2 ∣ z − y ∣ ≥ d /2 . Every Euclidean derivative ∂ z m K ( z , y ) \partial_z^m K(z,y) ∂ z m K ( z , y ) is therefore uniformly bounded in y y y , and dominated convergence gives
∂ z m ρ ^ j ( z ) = ∫ γ j ∂ z m K ( z , y ) d ℓ j ( y ) . \partial_z^m\widehat{\rho}_j(z)=\int_{\gamma_j}\partial_z^mK(z,y)\,\mathrm{d}\ell_j(y). ∂ z m ρ j ( z ) = ∫ γ j ∂ z m K ( z , y ) d ℓ j ( y ) .
where ρ ^ j \widehat{\rho}_j ρ j is the ambient extension on a sufficiently small Euclidean neighborhood of the opposite component. On the sphere, its gradient is G j = ( Id − z ⊗ z ) D log ρ ^ j G_j=(\operatorname{Id}-z\otimes z)D\log\widehat{\rho}_j G j = ( Id − z ⊗ z ) D log ρ j . Thus G j G_j G j has a smooth ambient extension near the opposite component.
Initially H 2 ↪ C 1 , 1 / 2 H^2\hookrightarrow C^{1,1/2} H 2 ↪ C 1 , 1/2 , so γ i ∈ C 1 , 1 / 2 \gamma_i\in C^{1,1/2} γ i ∈ C 1 , 1/2 , which implies γ i ∈ C 2 , 1 / 2 \gamma_i\in C^{2,1/2} γ i ∈ C 2 , 1/2 . Once both curves are C m , α C^{m,\alpha} C m , α , the ambient potentials remain smooth and their compositions with γ i \gamma_i γ i are C m , α C^{m,\alpha} C m , α , while P N i P_{N_i} P N i is C m − 1 , α C^{m-1,\alpha} C m − 1 , α . Equation (23) then gives γ i ∈ C m + 1 , α \gamma_i\in C^{m+1,\alpha} γ i ∈ C m + 1 , α . Induction proves γ i ∈ C ∞ \gamma_i\in C^\infty γ i ∈ C ∞ . □ \square □
Lemma 3.5. Assume dist ( γ 1 , γ 2 ) ≥ d > 0 \operatorname{dist}(\gamma_1,\gamma_2)\ge d>0 dist ( γ 1 , γ 2 ) ≥ d > 0 . Then, for every integer m ≥ 0 m\ge0 m ≥ 0 , i , j = { 1 , 2 } i,j=\{1,2\} i , j = { 1 , 2 } , i ≠ j i\ne j i = j ,
∥ ∇ m ρ j ∥ L ∞ ( γ i ) ≤ C m L ( γ j ) d − m − 2 , ρ j ≥ L ( γ j ) 4 . \lVert\nabla^m\rho_j\rVert_{L^\infty(\gamma_i)}\le C_mL(\gamma_j)d^{-m-2},\qquad\rho_j\ge\frac{L(\gamma_j)}{4}. ∥ ∇ m ρ j ∥ L ∞ ( γ i ) ≤ C m L ( γ j ) d − m − 2 , ρ j ≥ 4 L ( γ j ) .
For m ≥ 1 m\ge1 m ≥ 1 ,
∥ ∇ m log ρ j ∥ L ∞ ( γ i ) ≤ C m ( d ) (24) \lVert\nabla^m\log\rho_j\rVert_{L^\infty(\gamma_i)}\le C_m(d) \tag*{(24)} ∥ ∇ m log ρ j ∥ L ∞ ( γ i ) ≤ C m ( d ) (24)
with C m ( d ) C_m(d) C m ( d ) independent of L ( γ j ) L(\gamma_j) L ( γ j ) .
Proof. We only prove the case i = 1 i=1 i = 1 , j = 2 j=2 j = 2 . Repeated Euclidean differentiation of K ( x , y ) = ∣ x − y ∣ − 2 K(x,y)=\lvert x-y\rvert^{-2} K ( x , y ) = ∣ x − y ∣ − 2 yields a finite sum of tensors
∂ x m K ( x , y ) = ∣ x − y ∣ − m − 2 P m ( x − y ∣ x − y ∣ ) , \partial_x^mK(x,y)=\lvert x-y\rvert^{-m-2}P_m\left(\frac{x-y}{\lvert x-y\rvert}\right), ∂ x m K ( x , y ) = ∣ x − y ∣ − m − 2 P m ( ∣ x − y ∣ x − y ) ,
where P m P_m P m is a universal tensor-valued polynomial. The Gauss formula (7) gives
∣ ∇ x m K ( x , y ) ∣ ≤ C m ∣ x − y ∣ − m − 2 ≤ C m d − m − 2 . \lvert\nabla_x^mK(x,y)\rvert\le C_m\lvert x-y\rvert^{-m-2}\le C_md^{-m-2}. ∣ ∇ x m K ( x , y )∣ ≤ C m ∣ x − y ∣ − m − 2 ≤ C m d − m − 2 .
Arclength parametrization and dominated convergence justify differentiation under the integral:
∇ m ρ 2 ( x ) = ∫ γ 2 ∇ x m K ( x , y ) d ℓ 2 ( y ) . \nabla^m\rho_2(x)=\int_{\gamma_2}\nabla_x^mK(x,y)\,\mathrm{d}\ell_2(y). ∇ m ρ 2 ( x ) = ∫ γ 2 ∇ x m K ( x , y ) d ℓ 2 ( y ) .
This proves the first estimate. Since ∣ x − y ∣ 2 ≤ 4 |x-y|^2 \le4 ∣ x − y ∣ 2 ≤ 4 , the kernel is at least 1 / 4 1/4 1/4 , giving the lower bound.
For the logarithmic estimates, write
ρ 2 = L 2 ρ ˉ 2 , ρ ˉ 2 ( x ) = 1 L 2 ∫ γ 2 K ( x , y ) d ℓ 2 ( y ) . \rho_2=L_2\bar{\rho}_2,\qquad\bar{\rho}_2(x)=\frac{1}{L_2}\int_{\gamma_2}K(x,y)\,\mathrm{d}\ell_2(y). ρ 2 = L 2 ρ ˉ 2 , ρ ˉ 2 ( x ) = L 2 1 ∫ γ 2 K ( x , y ) d ℓ 2 ( y ) .
The function ρ ˉ 2 \bar{\rho}_2 ρ ˉ 2 is bounded below by 1 / 4 1/4 1/4 , and all of its ambient derivatives are bounded by constants depending only on m m m and d d d . Since log ρ 2 = log L 2 + log ρ ˉ 2 \log\rho_2=\log L_2+\log\bar{\rho}_2 log ρ 2 = log L 2 + log ρ ˉ 2 , every derivative of positive order is independent of the constant factor L 2 L_2 L 2 . The ordinary chain rule for log \log log proves (24) .
Proposition 3.6. Let Γ n = ( γ 1 , n , γ 2 , n ) \Gamma_n=(\gamma_{1,n},\gamma_{2,n}) Γ n = ( γ 1 , n , γ 2 , n ) be weakly critical pairs in constant-speed parametrizations. Suppose
0 < L − ≤ L ( γ i , n ) ≤ L + < ∞ , dist ( γ 1 , n , γ 2 , n ) ≥ d 0 > 0. 0<L_-\le L(\gamma_{i,n})\le L_+<\infty,\qquad\operatorname{dist}(\gamma_{1,n},\gamma_{2,n})\ge d_0>0. 0 < L − ≤ L ( γ i , n ) ≤ L + < ∞ , dist ( γ 1 , n , γ 2 , n ) ≥ d 0 > 0.
Then, for every m ≥ 0 m\ge0 m ≥ 0 , the curves γ i , n \gamma_{i,n} γ i , n have a uniform C m C^m C m bound.
Proof. Let a i , n = L ( γ i , n ) / ( 2 π ) a_{i,n}=L(\gamma_{i,n})/(2\pi) a i , n = L ( γ i , n ) / ( 2 π ) and G j , n = grad log ρ j , n G_{j,n}=\operatorname{grad}\log\rho_{j,n} G j , n = grad log ρ j , n . The speeds have common positive upper and lower bounds, while (17) gives common bounds for every ambient derivative of G j , n G_{j,n} G j , n in a fixed neighborhood of the opposite component. The critical equation reads
γ i , n ′ ′ = a i , n 2 ( P i , n G j , n ( γ i , n ) − γ i , n ) , P i , n = Id − γ i , n ⊗ γ i , n − T i , n ⊗ T i , n . (25) \gamma_{i,n}^{\prime\prime}=a_{i,n}^2\left(P_{i,n}G_{j,n}(\gamma_{i,n})-\gamma_{i,n}\right),\qquad P_{i,n}=\operatorname{Id}-\gamma_{i,n}\otimes\gamma_{i,n}-T_{i,n}\otimes T_{i,n}. \tag*{(25)} γ i , n ′′ = a i , n 2 ( P i , n G j , n ( γ i , n ) − γ i , n ) , P i , n = Id − γ i , n ⊗ γ i , n − T i , n ⊗ T i , n . (25)
Since ∣ γ i , n ∣ = 1 |\gamma_{i,n}|=1 ∣ γ i , n ∣ = 1 and ∣ γ i , n ′ ∣ = a i , n |\gamma_{i,n}^{\prime}|=a_{i,n} ∣ γ i , n ′ ∣ = a i , n , the sequence is bounded in C 1 C^1 C 1 , then (25) gives the C 2 C^2 C 2 bound.
The higher-order estimates follow inductively. Indeed, if
∣ γ i , n ∣ C m ≤ C m |\gamma_{i,n}|_{C^m}\le C_m ∣ γ i , n ∣ C m ≤ C m
for some m ≥ 2 m\ge2 m ≥ 2 , then differentiating (25) m − 1 m-1 m − 1 times expresses γ i , n ( m + 1 ) \gamma_{i,n}^{(m+1)} γ i , n ( m + 1 ) as a finite sum involving only derivatives of γ i , n \gamma_{i,n} γ i , n of order at most m m m and ambient derivatives of G j , n G_{j,n} G j , n of order at most m − 1 m-1 m − 1 . All these terms are uniformly bounded by the induction hypothesis and (24) . Hence
∣ γ i , n ∣ C m + 1 ≤ C m + 1 . |\gamma_{i,n}|_{C^{m+1}}\le C_{m+1}. ∣ γ i , n ∣ C m + 1 ≤ C m + 1 .
Induction proves the result.
Compactness at the Hopf value
Lemma 3.7. Suppose that non-split curve pairs Γ n \Gamma_n Γ n converge uniformly to a curve pair Γ ∞ \Gamma_\infty Γ ∞ whose component images are disjoint. Then Γ ∞ \Gamma_\infty Γ ∞ is non-split.
Proof. If a smooth embedded sphere S S S separated the two limiting images, compactness would give a positive distance from S S S to each image. Uniform convergence then places each approximating image, for large n n n , on the same side of S S S as its limit. The same fixed sphere would split Γ n \Gamma_n Γ n , a contradiction.
Proposition 3.8. Let Γ n ∈ A \Gamma_n\in\mathcal{A} Γ n ∈ A be non-split and weakly critical, with E ( Γ n ) → 2 π 2 E(\Gamma_n)\to2\pi^2 E ( Γ n ) → 2 π 2 . Up to Möbius transformations and separate reparametrizations of the components, a subsequence converges to H H H in C m C^m C m for every m m m .
Proof. Discarding finitely many terms, fix Λ \Lambda Λ with E ( Γ n ) ≤ Λ E(\Gamma_n)\le\Lambda E ( Γ n ) ≤ Λ . Apply Proposition 3.1 and Lemma 2.1 to obtain two-sided length bounds, a uniform positive chordal separation, and constant-speed parametrizations. The curves are smooth by Proposition 3.4, and have uniform C m C^m C m bounds for every m m m by Proposition 3.6. From Arzelà–Ascoli theorem and a diagonal subsequence, we get a limit Γ ∞ \Gamma_\infty Γ ∞ in every C m C^m C m norm.
The lower speed bound persists under C 1 C^1 C 1 convergence, and the separation bound persists under uniform convergence. Thus Γ ∞ ∈ A \Gamma_\infty\in\mathcal{A} Γ ∞ ∈ A . It is non-split by Lemma 3.7. On the fixed parameter torus the energy densities converge uniformly, since their denominators stay away from zero. Hence
E ( Γ ∞ ) = lim n → ∞ E ( Γ n ) = 2 π 2 . E(\Gamma_\infty) = \lim_{n\to\infty} E(\Gamma_n) = 2\pi^2. E ( Γ ∞ ) = n → ∞ lim E ( Γ n ) = 2 π 2 .
The equality case of Theorem 1.1 implies that after a conformal map Φ ∈ M o ¨ b + ( S 3 ) \Phi\in\operatorname{Möb}^{+}(\mathbb{S}^3) Φ ∈ M o ¨ b + ( S 3 ) (labeling and orientation fixed as explained after (1.6)), each Φ ∘ γ i , ∞ \Phi\circ\gamma_{i,\infty} Φ ∘ γ i , ∞ is a regular covering of a standard Hopf circle of some degree m i ≠ 0 m_i \ne0 m i = 0 , and
2 π 2 = E ( Γ ∞ ) = E ( Φ Γ ∞ ) = 2 π 2 ∣ m 1 m 2 ∣ , 2\pi^2 = E(\Gamma_\infty) = E(\Phi\Gamma_\infty) = 2\pi^2 |m_1m_2|, 2 π 2 = E ( Γ ∞ ) = E ( Φ Γ ∞ ) = 2 π 2 ∣ m 1 m 2 ∣ ,
So equality forces ∣ m 1 ∣ = ∣ m 2 ∣ = 1 |m_1| = |m_2| = 1 ∣ m 1 ∣ = ∣ m 2 ∣ = 1 , both are circle diffeomorphisms. Thus, g ∞ ⋅ Γ ∞ = H g_\infty\cdot\Gamma_\infty= H g ∞ ⋅ Γ ∞ = H for some g ∞ ∈ G g_\infty\in\mathcal{G} g ∞ ∈ G . These preserve convergence in every C m C^m C m norm and give the stated conclusion. □
The Hopf Hessian and the light-cone description of its kernel
Put
P 1 = span { e 1 , e 2 } , P 2 = span { e 3 , e 4 } . P_1 = \operatorname{span}\{e_1,e_2\}, \qquad P_2 = \operatorname{span}\{e_3,e_4\}. P 1 = span { e 1 , e 2 } , P 2 = span { e 3 , e 4 } .
For the Hopf pair (1.2), the curve-normal bundles are the constant bundles N x x = P 2 N_xx = P_2 N x x = P 2 and N y y = P 1 N_yy = P_1 N y y = P 1 . Thus normal fields have the form
U : S 1 → P 2 , V : S 1 → P 1 . U : \mathbb{S}^1 \to P_2, \qquad V : \mathbb{S}^1 \to P_1. U : S 1 → P 2 , V : S 1 → P 1 .
For k = 0 , 1 , 2 k = 0,1,2 k = 0 , 1 , 2 , set
H k = H k ( S 1 , P 2 ) ⊕ H k ( S 1 , P 1 ) , H − 1 = ( H 1 ) ∗ , \mathcal{H}^k = H^k(\mathbb{S}^1,P_2) \oplus H^k(\mathbb{S}^1,P_1), \qquad\mathcal{H}^{-1} = (\mathcal{H}^1)^*, H k = H k ( S 1 , P 2 ) ⊕ H k ( S 1 , P 1 ) , H − 1 = ( H 1 ) ∗ ,
using the unnormalized L 2 L^2 L 2 pairing as pivot. In particular,
∥ W ∥ H 1 2 = ∫ 0 2 π ( ∣ U ∣ 2 + ∣ U ′ ∣ 2 ) d s + ∫ 0 2 π ( ∣ V ∣ 2 + ∣ V ′ ∣ 2 ) d t , W = ( U , V ) . (26) \|W\|_{H^1}^2 = \int_0^{2\pi} (|U|^2 + |U'|^2)\,ds + \int_0^{2\pi} (|V|^2 + |V'|^2)\,dt, \qquad W = (U,V). \tag*{(26)} ∥ W ∥ H 1 2 = ∫ 0 2 π ( ∣ U ∣ 2 + ∣ U ′ ∣ 2 ) d s + ∫ 0 2 π ( ∣ V ∣ 2 + ∣ V ′ ∣ 2 ) d t , W = ( U , V ) . (26)
The second variation at the Hopf link
Proposition 4.1. The Hopf link is critical. For W = ( U , V ) ∈ H 1 W = (U,V) \in\mathcal{H}^1 W = ( U , V ) ∈ H 1 its normal Hessian is given by
δ 2 E H ( W ) = π ∫ 0 2 π ∣ U ′ ∣ 2 d s + π ∫ 0 2 π ∣ V ′ ∣ 2 d t . \delta^2 E_{\mathrm{H}}(W) = \pi\int_0^{2\pi} |U'|^2\,ds + \pi\int_0^{2\pi} |V'|^2\,dt. δ 2 E H ( W ) = π ∫ 0 2 π ∣ U ′ ∣ 2 d s + π ∫ 0 2 π ∣ V ′ ∣ 2 d t .
+ 2 ∫ 0 2 π ∫ 0 2 π ⟨ y ( t ) , U ( s ) ⟩ ⟨ x ( s ) , V ( t ) ⟩ d s d t . (27) \qquad+ 2\int_0^{2\pi}\int_0^{2\pi} \langle y(t),U(s)\rangle\langle x(s),V(t)\rangle\,ds\,dt. \tag*{(27)} + 2 ∫ 0 2 π ∫ 0 2 π ⟨ y ( t ) , U ( s )⟩ ⟨ x ( s ) , V ( t )⟩ d s d t . (27)
Proof. It suffices first to take smooth fields and use the pointwise spherical exponential variations
x τ ( s ) = exp x ( s ) ( τ U ( s ) ) , y τ ( t ) = exp y ( t ) ( τ V ( t ) ) . x_\tau(s) = \exp_{x(s)}(\tau U(s)), \qquad y_\tau(t) = \exp_{y(t)}(\tau V(t)). x τ ( s ) = exp x ( s ) ( τ U ( s )) , y τ ( t ) = exp y ( t ) ( τ V ( t )) .
Since exp p ( ξ ) = cos ∣ ξ ∣ p + sin ∣ ξ ∣ ∣ ξ ∣ ξ \exp_p(\xi) = \cos|\xi|\,p + \frac{\sin|\xi|}{|\xi|}\xi exp p ( ξ ) = cos ∣ ξ ∣ p + ∣ ξ ∣ s i n ∣ ξ ∣ ξ ,
x τ = x + τ U − τ 2 2 ∣ U ∣ 2 x + O ( τ 3 ) , y τ = y + τ V − τ 2 2 ∣ V ∣ 2 y + O ( τ 3 ) . x_\tau= x + \tau U - \frac{\tau^2}{2}|U|^2x + O(\tau^3), \qquad y_\tau= y + \tau V - \frac{\tau^2}{2}|V|^2y + O(\tau^3). x τ = x + τ U − 2 τ 2 ∣ U ∣ 2 x + O ( τ 3 ) , y τ = y + τ V − 2 τ 2 ∣ V ∣ 2 y + O ( τ 3 ) .
Using x , x ′ ∈ P 1 x,x' \in P_1 x , x ′ ∈ P 1 and U , U ′ ∈ P 2 U,U' \in P_2 U , U ′ ∈ P 2 , and symmetrically for y , V y,V y , V , gives
∣ x τ ′ ∣ = 1 + τ 2 2 ( ∣ U ′ ∣ 2 − ∣ U ∣ 2 ) + O ( τ 3 ) , ∣ y τ ′ ∣ = 1 + τ 2 2 ( ∣ V ′ ∣ 2 − ∣ V ∣ 2 ) + O ( τ 3 ) . |x'_\tau| = 1 + \frac{\tau^2}{2}\left(|U'|^2-|U|^2\right) + O(\tau^3), \qquad|y'_\tau| = 1 + \frac{\tau^2}{2}\left(|V'|^2-|V|^2\right) + O(\tau^3). ∣ x τ ′ ∣ = 1 + 2 τ 2 ( ∣ U ′ ∣ 2 − ∣ U ∣ 2 ) + O ( τ 3 ) , ∣ y τ ′ ∣ = 1 + 2 τ 2 ( ∣ V ′ ∣ 2 − ∣ V ∣ 2 ) + O ( τ 3 ) .
Let
a ( s , t ) = ⟨ U ( s ) , y ( t ) ⟩ + ⟨ x ( s ) , V ( t ) ⟩ . a(s,t)=\langle U(s),y(t)\rangle+\langle x(s),V(t)\rangle. a ( s , t ) = ⟨ U ( s ) , y ( t )⟩ + ⟨ x ( s ) , V ( t )⟩ .
For c τ = ⟨ x τ , y τ ⟩ c_\tau=\langle x_\tau,y_\tau\rangle c τ = ⟨ x τ , y τ ⟩ , orthogonality of P 1 P_1 P 1 and P 2 P_2 P 2 gives c 0 = 0 c_0=0 c 0 = 0 , c ˙ 0 = a \dot{c}_0=a c ˙ 0 = a , and c ¨ 0 = 2 ⟨ U , V ⟩ = 0 \ddot{c}_0=2\langle U,V\rangle=0 c ¨ 0 = 2 ⟨ U , V ⟩ = 0 . Therefore
1 ∣ x τ − y τ ∣ 2 = 1 2 ( 1 − c τ ) = 1 2 + τ 2 a + τ 2 2 a 2 + O ( τ 3 ) . \frac{1}{|x_\tau-y_\tau|^2}=\frac{1}{2(1-c_\tau)}=\frac{1}{2}+\frac{\tau}{2}a+\frac{\tau^2}{2}a^2+O(\tau^3). ∣ x τ − y τ ∣ 2 1 = 2 ( 1 − c τ ) 1 = 2 1 + 2 τ a + 2 τ 2 a 2 + O ( τ 3 ) .
The first-order term is linear in x ( s ) x(s) x ( s ) and y ( t ) y(t) y ( t ) . Since both Hopf circles have vanishing mean, ∫ S 1 x ( s ) d s = ∫ S 1 y ( t ) d t = 0 \int_{\mathbb{S}^1}x(s)\,\mathrm{d}s=\int_{\mathbb{S}^1}y(t)\,\mathrm{d}t=0 ∫ S 1 x ( s ) d s = ∫ S 1 y ( t ) d t = 0 , every such term integrates to zero on the parameter torus, hence δ E H = 0 \delta E_H=0 δ E H = 0 and H H H is critical. Twice the coefficient of τ 2 \tau^2 τ 2 after integration is
∬ a 2 d s d t + π ∫ 0 2 π ( ∣ U ′ ∣ 2 − ∣ U ∣ 2 ) d s + π ∫ 0 2 π ( ∣ V ′ ∣ 2 − ∣ V ∣ 2 ) d t . \iint a^2\,\mathrm{d}s\,\mathrm{d}t+\pi\int_0^{2\pi}\left(|U'|^2-|U|^2\right)\,\mathrm{d}s+\pi\int_0^{2\pi}\left(|V'|^2-|V|^2\right)\,\mathrm{d}t. ∬ a 2 d s d t + π ∫ 0 2 π ( ∣ U ′ ∣ 2 − ∣ U ∣ 2 ) d s + π ∫ 0 2 π ( ∣ V ′ ∣ 2 − ∣ V ∣ 2 ) d t .
Finally,
∫ 0 2 π x ⊗ x d s = π Id P 1 , ∫ 0 2 π y ⊗ y d t = π Id P 2 . (28) \int_0^{2\pi}x\otimes x\,\mathrm{d}s=\pi\,\operatorname{Id}_{P_1},\qquad\int_0^{2\pi}y\otimes y\,\mathrm{d}t=\pi\,\operatorname{Id}_{P_2}. \tag*{(28)} ∫ 0 2 π x ⊗ x d s = π Id P 1 , ∫ 0 2 π y ⊗ y d t = π Id P 2 . (28)
Hence the pure squares in ∬ a 2 d s d t \iint a^2\,\mathrm{d}s\,\mathrm{d}t ∬ a 2 d s d t equal π ∫ ∣ U ∣ 2 \pi\int|U|^2 π ∫ ∣ U ∣ 2 and π ∫ ∣ V ∣ 2 \pi\int|V|^2 π ∫ ∣ V ∣ 2 and cancel the negative zeroth-order terms. The remaining cross term is exactly (27) . Continuity extends the formula to H 1 H^1 H 1 .
Fourier perfect square and the transverse space
Every P 2 P_2 P 2 -valued field has an orthogonal decomposition
U ( s ) = U 0 + A x ( s ) + U ≥ 2 ( s ) , A ∈ Hom ( P 1 , P 2 ) , U(s)=U_0+Ax(s)+U_{\geq2}(s),\qquad A\in\operatorname{Hom}(P_1,P_2), U ( s ) = U 0 + A x ( s ) + U ≥ 2 ( s ) , A ∈ Hom ( P 1 , P 2 ) ,
and similarly
V ( t ) = V 0 + B y ( t ) + V ≥ 2 ( t ) , B ∈ Hom ( P 2 , P 1 ) , V(t)=V_0+By(t)+V_{\geq2}(t),\qquad B\in\operatorname{Hom}(P_2,P_1), V ( t ) = V 0 + B y ( t ) + V ≥ 2 ( t ) , B ∈ Hom ( P 2 , P 1 ) ,
where the remainders contain only Fourier frequencies of absolute value at least two. The subscript ≥ 2 \geq2 ≥ 2 refers to Fourier frequency, not to a pointwise inequality. We now make the decomposition explicit.
Let e 1 , e 2 e_1,e_2 e 1 , e 2 be the standard orthonormal basis of P 1 P_1 P 1 , so that x ( s ) = e 1 cos s + e 2 sin s x(s)=e_1\cos s+e_2\sin s x ( s ) = e 1 cos s + e 2 sin s . For U ∈ L 2 ( S 1 , P 2 ) U\in L^2(\mathbb{S}^1,P_2) U ∈ L 2 ( S 1 , P 2 ) set
U 0 = 1 2 π ∫ 0 2 π U ( s ) d s , U c = 1 π ∫ 0 2 π U ( s ) cos s d s , U s = 1 π ∫ 0 2 π U ( s ) sin s d s . U_0=\frac{1}{2\pi}\int_0^{2\pi}U(s)\,\mathrm{d}s,\qquad U_c=\frac{1}{\pi}\int_0^{2\pi}U(s)\cos s\,\mathrm{d}s,\qquad U_s=\frac{1}{\pi}\int_0^{2\pi}U(s)\sin s\,\mathrm{d}s. U 0 = 2 π 1 ∫ 0 2 π U ( s ) d s , U c = π 1 ∫ 0 2 π U ( s ) cos s d s , U s = π 1 ∫ 0 2 π U ( s ) sin s d s .
There is a unique map A ∈ Hom ( P 1 , P 2 ) A\in\operatorname{Hom}(P_1,P_2) A ∈ Hom ( P 1 , P 2 ) such that
A e 1 = U c , A e 2 = U s . Ae_1=U_c,\qquad Ae_2=U_s. A e 1 = U c , A e 2 = U s .
Thus
A x ( s ) = U c cos s + U s sin s Ax(s)=U_c\cos s+U_s\sin s A x ( s ) = U c cos s + U s sin s
is precisely the L 2 L^2 L 2 -orthogonal projection of U U U onto the real first-harmonic subspace, and
U ≥ 2 ( s ) : = U ( s ) − U 0 − A x ( s ) . U_{\geq2}(s):=U(s)-U_0-Ax(s). U ≥ 2 ( s ) := U ( s ) − U 0 − A x ( s ) .
It satisfies
∫ 0 2 π U ≥ 2 ( s ) d s = 0 , ∫ 0 2 π U ≥ 2 ( s ) cos s d s = 0 , ∫ 0 2 π U ≥ 2 ( s ) sin s d s = 0. \int_0^{2\pi}U_{\geq2}(s)\,\mathrm{d}s=0,\qquad\int_0^{2\pi}U_{\geq2}(s)\cos s\,\mathrm{d}s=0,\qquad\int_0^{2\pi}U_{\geq2}(s)\sin s\,\mathrm{d}s=0. ∫ 0 2 π U ≥ 2 ( s ) d s = 0 , ∫ 0 2 π U ≥ 2 ( s ) cos s d s = 0 , ∫ 0 2 π U ≥ 2 ( s ) sin s d s = 0.
Equivalently, the Fourier expansion of U ≥ 2 U_{\geq2} U ≥ 2 contains only frequencies ∣ n ∣ ≥ 2 |n| \geq2 ∣ n ∣ ≥ 2 . If U ∈ H 1 U \in H^1 U ∈ H 1 , the same orthogonal decomposition holds in H 1 H^1 H 1 , because Fourier projection commutes with differentiation. The decomposition of V V V is defined identically from y ( t ) = e 3 cos t + e 4 sin t y(t) = e_3 \cos t + e_4 \sin t y ( t ) = e 3 cos t + e 4 sin t , producing V 0 V_0 V 0 , a unique B ∈ Hom ( P 2 , P 1 ) B \in\operatorname{Hom}(P_2,P_1) B ∈ Hom ( P 2 , P 1 ) , and V ≥ 2 V_{\geq2} V ≥ 2 .
Proposition 4.2. One has the exact identity
δ 2 E H ( W ) = π 2 ∣ B + A T ∣ F 2 + π ∫ ∣ U ≥ 2 ′ ∣ 2 + π ∫ ∣ V ≥ 2 ′ ∣ 2 . (29) \delta^2 E_{\mathrm{H}}(W) = \pi^2 \lvert B + A^T\rvert_F^2 + \pi\int\lvert U_{\geq2}'\rvert^2 + \pi\int\lvert V_{\geq2}'\rvert^2. \tag*{(29)} δ 2 E H ( W ) = π 2 ∣ B + A T ∣ F 2 + π ∫ ∣ U ≥ 2 ′ ∣ 2 + π ∫ ∣ V ≥ 2 ′ ∣ 2 . (29)
where ∣ A ∣ F = tr ( A T A ) \lvert A\rvert_F = \operatorname{tr}(A^T A) ∣ A ∣ F = tr ( A T A ) . Consequently, the nullspace K H \mathcal{K}_{\mathrm{H}} K H of the bilinear form is
K H = { ( U 0 , V 0 ) : U 0 ∈ P 2 , V 0 ∈ P 1 } ⊕ { ( A x , − A T y ) : A ∈ Hom ( P 1 , P 2 ) } , (30) \mathcal{K}_{\mathrm{H}} = \{(U_0,V_0) : U_0 \in P_2,\ V_0 \in P_1\} \oplus\{(Ax,-A^T y) : A \in\operatorname{Hom}(P_1,P_2)\}, \tag*{(30)} K H = {( U 0 , V 0 ) : U 0 ∈ P 2 , V 0 ∈ P 1 } ⊕ {( A x , − A T y ) : A ∈ Hom ( P 1 , P 2 )} , (30)
and dim K H = 8 \dim\mathcal{K}_{\mathrm{H}} = 8 dim K H = 8 . Its L 2 L^2 L 2 -orthogonal complement is
K H ⊥ = { ( A x + U ≥ 2 , A T y + V ≥ 2 ) : A ∈ Hom ( P 1 , P 2 ) } . (31) \mathcal{K}_{\mathrm{H}}^\perp= \{(Ax+U_{\geq2},A^T y+V_{\geq2}) : A \in\operatorname{Hom}(P_1,P_2)\}. \tag*{(31)} K H ⊥ = {( A x + U ≥ 2 , A T y + V ≥ 2 ) : A ∈ Hom ( P 1 , P 2 )} . (31)
For every W ∈ H 1 ∩ K H ⊥ W \in H^1 \cap\mathcal{K}_{\mathrm{H}}^\perp W ∈ H 1 ∩ K H ⊥ ,
δ 2 E H ( W ) ≥ 4 π 5 ∣ W ∣ H 1 2 . (32) \delta^2 E_{\mathrm{H}}(W) \geq\frac{4\pi}{5}\lvert W\rvert_{H^1}^2. \tag*{(32)} δ 2 E H ( W ) ≥ 5 4 π ∣ W ∣ H 1 2 . (32)
Proof. We first explain why the nonlocal term in (27) sees only first Fourier modes. Set
I ( U , V ) : = ∫ 0 2 π ∫ 0 2 π ⟨ y ( t ) , U ( s ) ⟩ ⟨ x ( s ) , V ( t ) ⟩ d s d t . I(U,V) := \int_0^{2\pi}\int_0^{2\pi} \langle y(t),U(s)\rangle\langle x(s),V(t)\rangle\,\mathrm{d}s\,\mathrm{d}t. I ( U , V ) := ∫ 0 2 π ∫ 0 2 π ⟨ y ( t ) , U ( s )⟩ ⟨ x ( s ) , V ( t )⟩ d s d t .
If U = U 0 U=U_0 U = U 0 is constant, integration in s s s and ∫ x ( s ) d s = 0 \int x(s)\,\mathrm{d}s=0 ∫ x ( s ) d s = 0 give I ( U 0 , V ) = 0 I(U_0,V)=0 I ( U 0 , V ) = 0 ; similarly I ( U , V 0 ) = 0 I(U,V_0)=0 I ( U , V 0 ) = 0 . If U = U ≥ 2 U=U_{\geq2} U = U ≥ 2 , then for each fixed t t t the function s ↦ ⟨ y ( t ) , U ≥ 2 ( s ) ⟩ s\mapsto\langle y(t),U_{\geq2}(s)\rangle s ↦ ⟨ y ( t ) , U ≥ 2 ( s )⟩ contains only frequencies ∣ n ∣ ≥ 2 |n|\geq2 ∣ n ∣ ≥ 2 , whereas s ↦ ⟨ x ( s ) , V ( t ) ⟩ s\mapsto\langle x(s),V(t)\rangle s ↦ ⟨ x ( s ) , V ( t )⟩ has frequency ∣ n ∣ = 1 |n|=1 ∣ n ∣ = 1 . Their integral therefore vanishes. Interchanging s s s and t t t shows that every term containing V ≥ 2 V_{\geq2} V ≥ 2 also vanishes. Hence
I ( U , V ) = I ( A x , B y ) . I(U,V)=I(Ax,By). I ( U , V ) = I ( A x , B y ) .
We now compute this remaining term. Since ⟨ y , A x ⟩ = ⟨ A T y , x ⟩ \langle y,Ax\rangle=\langle A^T y,x\rangle ⟨ y , A x ⟩ = ⟨ A T y , x ⟩ , the moment identity (28) gives, after first integrating in s s s ,
∫ 0 2 π ⟨ y , A x ⟩ ⟨ x , B y ⟩ d s = ⟨ A T y , ( ∫ 0 2 π x ⊗ x d s ) B y ⟩ = π ⟨ A T y , B y ⟩ = π ⟨ y , A B y ⟩ . \int_0^{2\pi} \langle y,Ax\rangle\langle x,By\rangle\,\mathrm{d}s
= \left\langle A^T y,\left(\int_0^{2\pi}x\otimes x\,\mathrm{d}s\right)By\right\rangle
= \pi\langle A^T y,By\rangle
= \pi\langle y,ABy\rangle. ∫ 0 2 π ⟨ y , A x ⟩ ⟨ x , B y ⟩ d s = ⟨ A T y , ( ∫ 0 2 π x ⊗ x d s ) B y ⟩ = π ⟨ A T y , B y ⟩ = π ⟨ y , A B y ⟩ .
A second use of (28) yields
I ( A x , B y ) = π tr ( A B ∫ 0 2 π y ⊗ y d t ) = π 2 tr ( A B ) . I(Ax,By)=\pi\operatorname{tr}\left(AB\int_0^{2\pi}y\otimes y\,\mathrm{d}t\right)=\pi^2\operatorname{tr}(AB). I ( A x , B y ) = π tr ( A B ∫ 0 2 π y ⊗ y d t ) = π 2 tr ( A B ) .
If ∣ A ∣ F 2 : = tr ( A T A ) \lvert A\rvert_F^2:=\operatorname{tr}(A^T A) ∣ A ∣ F 2 := tr ( A T A ) denotes the Frobenius norm, then
∫ 0 2 π ∣ ( A x ) ′ ∣ 2 d s = ∫ 0 2 π ⟨ A x ′ , A x ′ ⟩ d s = tr ( A T A ∫ 0 2 π x ′ ⊗ x ′ d s ) = π ∣ A ∣ F 2 , \int_0^{2\pi}\lvert(Ax)'\rvert^2\,\mathrm{d}s
=\int_0^{2\pi}\langle Ax',Ax'\rangle\,\mathrm{d}s
=\operatorname{tr}\left(A^T A\int_0^{2\pi}x'\otimes x'\,\mathrm{d}s\right)
=\pi\lvert A\rvert_F^2, ∫ 0 2 π ∣( A x ) ′ ∣ 2 d s = ∫ 0 2 π ⟨ A x ′ , A x ′ ⟩ d s = tr ( A T A ∫ 0 2 π x ′ ⊗ x ′ d s ) = π ∣ A ∣ F 2 ,
because ∫ x ′ ⊗ x ′ d s = π Id P 1 \int x'\otimes x'\,\mathrm{d}s=\pi\operatorname{Id}_{P_1} ∫ x ′ ⊗ x ′ d s = π Id P 1 . The second component gives ∫ ∣ ( B y ) ′ ∣ 2 d t = π ∣ B ∣ F 2 \int\lvert(By)'\rvert^2\,\mathrm{d}t=\pi\lvert B\rvert_F^2 ∫ ∣( B y ) ′ ∣ 2 d t = π ∣ B ∣ F 2 in the same way. Thus the first-mode contribution is
π 2 ( ∣ A ∣ F 2 + ∣ B ∣ F 2 + 2 tr ( A B ) ) = π 2 ∣ B + A T ∣ F 2 , \pi^2\left(\lvert A\rvert_F^2+\lvert B\rvert_F^2+2\operatorname{tr}(AB)\right)
=\pi^2\lvert B+A^T\rvert_F^2, π 2 ( ∣ A ∣ F 2 + ∣ B ∣ F 2 + 2 tr ( A B ) ) = π 2 ∣ B + A T ∣ F 2 ,
proving (29) and (30) . Orthogonality to the constant kernel removes U 0 U_0 U 0 , V 0 V_0 V 0 , while orthogonality to every ( C x , − C T y ) (Cx,-C^T y) ( C x , − C T y ) gives B = A T B=A^T B = A T , proving (31) .
On the transverse first modes,
δ 2 E H ( ( A x , A T y ) ) = 4 π 2 ∣ A ∣ F 2 and ∣ ( A x , A T y ) ∣ H 1 2 = 4 π ∣ A ∣ F 2 . \delta^2 E_H((Ax,A^T y)) = 4\pi^2 \lvert A\rvert_F^2 \quad\text{and}\quad\lvert(Ax,A^T y)\rvert_{H^1}^2 = 4\pi\lvert A\rvert_F^2. δ 2 E H (( A x , A T y )) = 4 π 2 ∣ A ∣ F 2 and ∣( A x , A T y ) ∣ H 1 2 = 4 π ∣ A ∣ F 2 .
On frequencies ∣ n ∣ ≥ 2 \lvert n\rvert\ge2 ∣ n ∣ ≥ 2 , Poincaré’s inequality gives ∫ ∣ f ′ ∣ 2 ≥ 4 ∫ ∣ f ∣ 2 \int\lvert f'\rvert^2 \ge4\int\lvert f\rvert^2 ∫ ∣ f ′ ∣ 2 ≥ 4 ∫ ∣ f ∣ 2 , and hence
π ∫ ∣ f ′ ∣ 2 ≥ ( 4 π / 5 ) ∣ f ∣ H 1 2 . \pi\int\lvert f'\rvert^2 \ge(4\pi/5)\lvert f\rvert_{H^1}^2. π ∫ ∣ f ′ ∣ 2 ≥ ( 4 π /5 ) ∣ f ∣ H 1 2 .
The Fourier pieces are orthogonal, which proves (32) .
The infinitesimal Möbius action
Equip R 4 , 1 = R 4 ⊕ R \mathbb{R}^{4,1}=\mathbb{R}^4\oplus\mathbb{R} R 4 , 1 = R 4 ⊕ R with the Lorentzian form
⟨ ( z , λ ) , ( w , μ ) ⟩ L = ⟨ z , w ⟩ − λ μ , \langle(z,\lambda),(w,\mu)\rangle_L=\langle z,w\rangle-\lambda\mu, ⟨( z , λ ) , ( w , μ ) ⟩ L = ⟨ z , w ⟩ − λ μ ,
we omit the transpose symbol here. Let
N + = { Z ≠ 0 : ⟨ Z , Z ⟩ L = 0 , Z 5 > 0 } \mathcal{N}^+=\{Z\ne0:\langle Z,Z\rangle_L=0,\ Z_5>0\} N + = { Z = 0 : ⟨ Z , Z ⟩ L = 0 , Z 5 > 0 }
be the future null cone. The affine section z ↦ ( z , 1 ) z\mapsto(z,1) z ↦ ( z , 1 ) identifies S 3 \mathbb{S}^3 S 3 with P ( N + ) \mathbb{P}(\mathcal{N}^+) P ( N + ) , and
− 2 ⟨ ( z , 1 ) , ( w , 1 ) ⟩ L = ∣ z − w ∣ 2 . (33) -2\langle(z,1),(w,1)\rangle_L=\lvert z-w\rvert^2. \tag*{(33)} − 2 ⟨( z , 1 ) , ( w , 1 ) ⟩ L = ∣ z − w ∣ 2 . (33)
The identity component S O + ( 4 , 1 ) SO^+(4,1) S O + ( 4 , 1 ) acts linearly on N + \mathcal{N}^+ N + and projectively on S 3 \mathbb{S}^3 S 3 as M o ¨ b + ( S 3 ) \operatorname{Möb}^+(\mathbb{S}^3) M o ¨ b + ( S 3 ) ; see [5 ] .
Let J = diag ( I 4 , − 1 ) J=\operatorname{diag}(I_4,-1) J = diag ( I 4 , − 1 ) . An element of the Lorentz Lie algebra satisfies A T J + J A = 0 A^T J+JA=0 A T J + J A = 0 . Starting from a general block matrix
A = ( B u v T α ) , \mathcal{A}=\begin{pmatrix} B & u \\ v^T & \alpha\end{pmatrix}, A = ( B v T u α ) ,
this equation is
B T + B = 0 , u = v , α = 0. B^T+B=0,\qquad u=v,\qquad\alpha=0. B T + B = 0 , u = v , α = 0.
Thus every A ∈ s o ( 4 , 1 ) \mathcal{A}\in\mathfrak{so}(4,1) A ∈ so ( 4 , 1 ) has the form
A = ( Ω a a T 0 ) , Ω ∈ s o ( 4 ) , a ∈ R 4 . (34) \mathcal{A}=\begin{pmatrix}\Omega&a\\a^T&0\end{pmatrix},\qquad\Omega\in\mathfrak{so}(4),\quad a\in\mathbb{R}^4. \tag*{(34)} A = ( Ω a T a 0 ) , Ω ∈ so ( 4 ) , a ∈ R 4 . (34)
To determine the induced vector field, apply e τ A e^{\tau\mathcal{A}} e τ A to the null lift ( z , 1 ) (z,1) ( z , 1 ) :
e τ A ( z , 1 ) = ( z , 1 ) + τ ( Ω z + a , ⟨ a , z ⟩ ) + O ( τ 2 ) . e^{\tau\mathcal{A}}(z,1)=(z,1)+\tau(\Omega z+a,\langle a,z\rangle)+O(\tau^2). e τ A ( z , 1 ) = ( z , 1 ) + τ ( Ω z + a , ⟨ a , z ⟩) + O ( τ 2 ) .
Returning to the affine section means dividing the first four coordinates by the last one. Therefore
z + τ ( Ω z + a ) + O ( τ 2 ) 1 + τ ⟨ a , z ⟩ + O ( τ 2 ) = z + τ ( Ω z + a − ⟨ a , z ⟩ z ) + O ( τ 2 ) , \frac{z+\tau(\Omega z+a)+O(\tau^2)}{1+\tau\langle a,z\rangle+O(\tau^2)}
=z+\tau(\Omega z+a-\langle a,z\rangle z)+O(\tau^2), 1 + τ ⟨ a , z ⟩ + O ( τ 2 ) z + τ ( Ω z + a ) + O ( τ 2 ) = z + τ ( Ω z + a − ⟨ a , z ⟩ z ) + O ( τ 2 ) ,
and the infinitesimal Möbius field is
X A ( z ) = Ω z + a − ⟨ a , z ⟩ z . (35) X_{\mathcal{A}}(z)=\Omega z+a-\langle a,z\rangle z. \tag*{(35)} X A ( z ) = Ω z + a − ⟨ a , z ⟩ z . (35)
The last expression is tangent to S 3 \mathbb{S}^3 S 3 because its inner product with z z z vanishes.
Relative to R 4 = P 1 ⊕ P 2 \mathbb{R}^4=P_1\oplus P_2 R 4 = P 1 ⊕ P 2 , decompose
Ω = ( Ω 1 − C T C Ω 2 ) , a = a 1 + a 2 , (36) \Omega=\begin{pmatrix}\Omega_1&-C^T\\C&\Omega_2\end{pmatrix},\qquad a=a_1+a_2, \tag*{(36)} Ω = ( Ω 1 C − C T Ω 2 ) , a = a 1 + a 2 , (36)
where Ω i ∈ s o ( P i ) \Omega_i\in\mathfrak{so}(P_i) Ω i ∈ so ( P i ) , C ∈ Hom ( P 1 , P 2 ) C\in\operatorname{Hom}(P_1,P_2) C ∈ Hom ( P 1 , P 2 ) , and a i ∈ P i a_i\in P_i a i ∈ P i . Along x ( s ) ∈ P 1 x(s)\in P_1 x ( s ) ∈ P 1 , formula (35) splits as
X A ( x ) = Ω 1 x + a 1 − ⟨ a 1 , x ⟩ x ⏟ ∈ P 1 ∩ x ⊥ = R x ′ + C x + a 2 ⏟ ∈ P 2 = N x x . X_{\mathcal{A}}(x)=\underbrace{\Omega_1x+a_1-\langle a_1,x\rangle x}_{\in P_1\cap x^\perp=\mathbb{R}x'}+Cx+\underbrace{a_2}_{\in P_2=N_xx}. X A ( x ) = ∈ P 1 ∩ x ⊥ = R x ′ Ω 1 x + a 1 − ⟨ a 1 , x ⟩ x + C x + ∈ P 2 = N x x a 2 .
Along y ( t ) ∈ P 2 y(t) \in P_2 y ( t ) ∈ P 2 it splits as
X A ( y ) = Ω 2 y + a 2 − ⟨ a 2 , y ⟩ y + a 1 − C T y . X_{\mathcal{A}}(y)=\Omega_2y+a_2-\langle a_2,y\rangle y+a_1-C^Ty. X A ( y ) = Ω 2 y + a 2 − ⟨ a 2 , y ⟩ y + a 1 − C T y .
Consequently the normal trace of the Lorentz infinitesimal action is
j ( A ) = ( a 2 + C x , a 1 − C T y ) . (37) j(\mathcal{A})=(a_2+Cx,a_1-C^Ty). \tag*{(37)} j ( A ) = ( a 2 + C x , a 1 − C T y ) . (37)
The block-diagonal rotations Ω 1 , Ω 2 \Omega_1,\Omega_2 Ω 1 , Ω 2 generate only tangential rotations of the two great circles, they form the two-dimensional infinitesimal stabilizer T = s o ( P 1 ) ⊕ s o ( P 2 ) \mathcal{T}=\mathfrak{so}(P_1)\oplus\mathfrak{so}(P_2) T = so ( P 1 ) ⊕ so ( P 2 ) . Choose the explicit complement
M = { ( 0 − C T a 1 C 0 a 2 a 1 T a 2 T 0 ) : C ∈ Hom ( P 1 , P 2 ) , a i ∈ P i } . (38) \mathcal{M}=\left\{\left(\begin{array}{ccc}
0 & -C^T & a_1\\
C & 0 & a_2\\
a_1^T & a_2^T & 0
\end{array}\right): C\in\operatorname{Hom}(P_1,P_2),\ a_i\in P_i\right\}. \tag*{(38)} M = ⎩ ⎨ ⎧ 0 C a 1 T − C T 0 a 2 T a 1 a 2 0 : C ∈ Hom ( P 1 , P 2 ) , a i ∈ P i ⎭ ⎬ ⎫ . (38)
It has dimension 4 + 2 + 2 = 8 4+2+2=8 4 + 2 + 2 = 8 . If the normal trace (37) vanishes for an element of M \mathcal{M} M , varying x x x and y y y around their circles gives C = a 1 = a 2 = 0 C=a_1=a_2=0 C = a 1 = a 2 = 0 ; hence j ∣ M j|_{\mathcal{M}} j ∣ M is injective. Its image is precisely
{ ( U 0 , V 0 ) } ⊕ { ( C x , − C T y ) } , \{(U_0,V_0)\}\oplus\{(Cx,-C^Ty)\}, {( U 0 , V 0 )} ⊕ {( C x , − C T y )} ,
which is the eight-dimensional space (30) . We have proved the following identification.
Proposition 4.3. The map
j : M ⟶ K H (39) j:\mathcal{M}\longrightarrow\mathcal{K}_{H} \tag*{(39)} j : M ⟶ K H (39)
is a linear isomorphism.
Möbius invariance already implies that these orbit directions lie in the Hessian kernel; the Fourier perfect-square identity proves conversely that the kernel has no other directions. Thus K H \mathcal{K}_{H} K H is the normal tangent space to the Möbius orbit of the unparametrized Hopf link. Tangential reparametrization directions have already been removed by working on normal fields. The space (31) is the transverse complement used below.
The linearized operator and its transverse inverse
We now derive the linearized operator from the quadratic form. For W = ( U , V ) W=(U,V) W = ( U , V ) and W ~ = ( U ~ , V ~ ) \widetilde{W}=(\widetilde{U},\widetilde{V}) W = ( U , V ) define the polarization
B H ( W , W ~ ) : = 1 2 ( δ 2 E H ( W + W ~ ) − δ 2 E H ( W ) − δ 2 E H ( W ~ ) ) . B_H(W,\widetilde{W}):=\frac{1}{2}\left(\delta^2E_H(W+\widetilde{W})-\delta^2E_H(W)-\delta^2E_H(\widetilde{W})\right). B H ( W , W ) := 2 1 ( δ 2 E H ( W + W ) − δ 2 E H ( W ) − δ 2 E H ( W ) ) .
Substitution of (27) gives
B H ( W , W ~ ) = π ∫ 0 2 π ⟨ U ′ , U ~ ′ ⟩ d s + π ∫ 0 2 π ⟨ V ′ , V ~ ′ ⟩ d t + ∫ 0 2 π ∫ 0 2 π ⟨ y ( t ) , U ( s ) ⟩ ⟨ x ( s ) , V ~ ( t ) ⟩ d s d t + ∫ 0 2 π ∫ 0 2 π ⟨ y ( t ) , U ~ ( s ) ⟩ ⟨ x ( s ) , V ( t ) ⟩ d s d t . \begin{aligned}
B_H(W,\widetilde{W})={}&\pi\int_0^{2\pi}\langle U',\widetilde{U}'\rangle\,ds+\pi\int_0^{2\pi}\langle V',\widetilde{V}'\rangle\,dt\\
&+\int_0^{2\pi}\int_0^{2\pi}\langle y(t),U(s)\rangle\langle x(s),\widetilde{V}(t)\rangle\,ds\,dt\\
&+\int_0^{2\pi}\int_0^{2\pi}\langle y(t),\widetilde{U}(s)\rangle\langle x(s),V(t)\rangle\,ds\,dt.
\end{aligned} B H ( W , W ) = π ∫ 0 2 π ⟨ U ′ , U ′ ⟩ d s + π ∫ 0 2 π ⟨ V ′ , V ′ ⟩ d t + ∫ 0 2 π ∫ 0 2 π ⟨ y ( t ) , U ( s )⟩ ⟨ x ( s ) , V ( t )⟩ d s d t + ∫ 0 2 π ∫ 0 2 π ⟨ y ( t ) , U ( s )⟩ ⟨ x ( s ) , V ( t )⟩ d s d t .
For smooth periodic fields, integration by parts on the closed parameter circles yields
π ∫ ⟨ U ′ , U ~ ′ ⟩ d s = − π ∫ ⟨ U ′ ′ , U ~ ⟩ d s , π ∫ ⟨ V ′ , V ~ ′ ⟩ d t = − π ∫ ⟨ V ′ ′ , V ~ ⟩ d t . \pi\int\langle U',\widetilde{U}'\rangle\,ds=-\pi\int\langle U'',\widetilde{U}\rangle\,ds,\qquad
\pi\int\langle V',\widetilde{V}'\rangle\,dt=-\pi\int\langle V'',\widetilde{V}\rangle\,dt. π ∫ ⟨ U ′ , U ′ ⟩ d s = − π ∫ ⟨ U ′′ , U ⟩ d s , π ∫ ⟨ V ′ , V ′ ⟩ d t = − π ∫ ⟨ V ′′ , V ⟩ d t .
Fubini’s theorem rewrites the nonlocal terms with the test fields factored out:
∬ ⟨ y ( t ) , U ~ ( s ) ⟩ ⟨ x ( s ) , V ( t ) ⟩ d s d t = ∫ 0 2 π ⟨ ∫ 0 2 π y ( t ) ⟨ x ( s ) , V ( t ) ⟩ d t , U ~ ( s ) ⟩ d s , \iint\langle y(t),\widetilde{U}(s)\rangle\langle x(s),V(t)\rangle\,ds\,dt
=\int_0^{2\pi}\left\langle\int_0^{2\pi}y(t)\langle x(s),V(t)\rangle\,dt,\widetilde{U}(s)\right\rangle ds, ∬ ⟨ y ( t ) , U ( s )⟩ ⟨ x ( s ) , V ( t )⟩ d s d t = ∫ 0 2 π ⟨ ∫ 0 2 π y ( t ) ⟨ x ( s ) , V ( t )⟩ d t , U ( s ) ⟩ d s ,
and
∬ ⟨ y ( t ) , U ( s ) ⟩ ⟨ x ( s ) , V ~ ( t ) ⟩ d s d t = ∫ 0 2 π ⟨ ∫ 0 2 π x ( s ) ⟨ y ( t ) , U ( s ) ⟩ d s , V ~ ( t ) ⟩ d t . \iint\langle y(t), U(s)\rangle\langle x(s), \widetilde{V}(t)\rangle\,\mathrm{d}s\,\mathrm{d}t
=
\int_{0}^{2\pi} \left\langle\int_{0}^{2\pi} x(s)\langle y(t),U(s)\rangle\,\mathrm{d}s,\widetilde{V}(t)\right\rangle\,\mathrm{d}t. ∬ ⟨ y ( t ) , U ( s )⟩ ⟨ x ( s ) , V ( t )⟩ d s d t = ∫ 0 2 π ⟨ ∫ 0 2 π x ( s ) ⟨ y ( t ) , U ( s )⟩ d s , V ( t ) ⟩ d t .
Consequently,
B H ( W , W ~ ) = ∫ ⟨ ( L H W ) 1 , U ~ ⟩ d s + ∫ ⟨ ( L H W ) 2 , V ~ ⟩ d t , B_H(W,\widetilde{W})=\int\langle(\mathcal{L}_H W)_1,\widetilde{U}\rangle\,\mathrm{d}s+\int\langle(\mathcal{L}_H W)_2,\widetilde{V}\rangle\,\mathrm{d}t, B H ( W , W ) = ∫ ⟨( L H W ) 1 , U ⟩ d s + ∫ ⟨( L H W ) 2 , V ⟩ d t ,
where the self-adjoint operator is
( L H W ) 1 ( s ) = − π U ′ ′ ( s ) + ∫ 0 2 π y ( t ) ⟨ x ( s ) , V ( t ) ⟩ d t , (\mathcal{L}_H W)_1(s)=-\pi U''(s)+\int_{0}^{2\pi} y(t)\langle x(s),V(t)\rangle\,\mathrm{d}t, ( L H W ) 1 ( s ) = − π U ′′ ( s ) + ∫ 0 2 π y ( t ) ⟨ x ( s ) , V ( t )⟩ d t ,
( L H W ) 2 ( t ) = − π V ′ ′ ( t ) + ∫ 0 2 π x ( s ) ⟨ y ( t ) , U ( s ) ⟩ d s . (40) (\mathcal{L}_H W)_2(t)=-\pi V''(t)+\int_{0}^{2\pi} x(s)\langle y(t),U(s)\rangle\,\mathrm{d}s. \tag*{(40)} ( L H W ) 2 ( t ) = − π V ′′ ( t ) + ∫ 0 2 π x ( s ) ⟨ y ( t ) , U ( s )⟩ d s . (40)
On H 0 \mathcal{H}^{0} H 0 with domain H 2 \mathcal{H}^{2} H 2 , L H \mathcal{L}_H L H is a self-adjoint operator. It has compact resolvent. Equivalently, (40) , with distributional second derivatives, defines a bounded map L H : H 1 → H − 1 \mathcal{L}_H:\mathcal{H}^{1}\to\mathcal{H}^{-1} L H : H 1 → H − 1 satisfying
⟨ L H W , W ~ ⟩ = B H ( W , W ~ ) . \langle\mathcal{L}_H W,\widetilde{W}\rangle=B_H(W,\widetilde{W}). ⟨ L H W , W ⟩ = B H ( W , W ) .
Its kernel is K H \mathcal{K}_H K H .
Since K H \mathcal{K}_H K H is finite-dimensional and consists of smooth fields, the L 2 L^2 L 2 -orthogonal projections
P K H , P ⊥ = I − P K H (41) P_{\mathcal{K}_H},\qquad P_{\perp}=I-P_{\mathcal{K}_H} \tag*{(41)} P K H , P ⊥ = I − P K H (41)
extend boundedly to H 1 \mathcal{H}^{1} H 1 and H − 1 \mathcal{H}^{-1} H − 1 .
On the closed space H 1 ∩ K H ⊥ \mathcal{H}^{1}\cap\mathcal{K}_H^{\perp} H 1 ∩ K H ⊥ , the form B H B_H B H is bounded and coercive by (4.7). The Lax-Milgram theorem identifies this space with its dual through L H \mathcal{L}_H L H . In particular,
∥ W ∥ H 1 ≤ C ∥ L H W ∥ H − 1 , W ∈ H 1 ∩ K H ⊥ . (42) \lVert W\rVert_{\mathcal{H}^{1}}\leq C\lVert\mathcal{L}_H W\rVert_{\mathcal{H}^{-1}},\qquad W\in\mathcal{H}^{1}\cap\mathcal{K}_H^{\perp}. \tag*{(42)} ∥ W ∥ H 1 ≤ C ∥ L H W ∥ H − 1 , W ∈ H 1 ∩ K H ⊥ . (42)
A local Möbius slice and isolation of the Hopf orbit
The kernel calculation in Proposition 4.3 and the coercivity estimate (4.7) give the local isolation of the Hopf orbit in Theorem 5.3.
Normal graphs and the first-variation estimate
Fix 0 < r 0 < π / 4 0<r_0<\pi/4 0 < r 0 < π /4 . The normal exponential maps
Ψ 1 ( s , ξ ) = exp x ( s ) ξ , ( s , ξ ) ∈ S 1 × B r 0 ( 0 ; P 2 ) , (43) \Psi_1(s,\xi)=\exp_{x(s)}\xi,\qquad(s,\xi)\in\mathbb{S}^{1}\times B_{r_0}(0;P_2), \tag*{(43)} Ψ 1 ( s , ξ ) = exp x ( s ) ξ , ( s , ξ ) ∈ S 1 × B r 0 ( 0 ; P 2 ) , (43)
Ψ 2 ( t , η ) = exp y ( t ) η , ( t , η ) ∈ S 1 × B r 0 ( 0 ; P 1 ) (44) \Psi_2(t,\eta)=\exp_{y(t)}\eta,\qquad(t,\eta)\in\mathbb{S}^{1}\times B_{r_0}(0;P_1) \tag*{(44)} Ψ 2 ( t , η ) = exp y ( t ) η , ( t , η ) ∈ S 1 × B r 0 ( 0 ; P 1 ) (44)
are diffeomorphisms onto disjoint tubular neighborhoods of the Hopf circles. For a small field v = ( U , V ) v=(U,V) v = ( U , V ) , set
Graph H ( v ) = ( X U , Y V ) , X U ( s ) = Ψ 1 ( s , U ( s ) ) , Y V ( t ) = Ψ 2 ( t , V ( t ) ) , \operatorname{Graph}_H(v)=(X_U,Y_V),\qquad X_U(s)=\Psi_1(s,U(s)),\qquad Y_V(t)=\Psi_2(t,V(t)), Graph H ( v ) = ( X U , Y V ) , X U ( s ) = Ψ 1 ( s , U ( s )) , Y V ( t ) = Ψ 2 ( t , V ( t )) ,
and define
E ( v ) = E ( Graph H ( v ) ) , ⟨ G ( v ) , ζ ⟩ = δ E ( v ) [ ζ ] . \mathcal{E}(v)=E(\operatorname{Graph}_H(v)),\qquad\langle\mathcal{G}(v),\zeta\rangle=\delta\mathcal{E}(v)[\zeta]. E ( v ) = E ( Graph H ( v )) , ⟨ G ( v ) , ζ ⟩ = δ E ( v ) [ ζ ] .
Here v ∈ H 2 v\in\mathcal{H}^{2} v ∈ H 2 has small C 1 C^{1} C 1 norm, and the extended first variation acts on ζ ∈ H 1 \zeta\in\mathcal{H}^{1} ζ ∈ H 1 , so G ( v ) ∈ H − 1 \mathcal{G}(v)\in\mathcal{H}^{-1} G ( v ) ∈ H − 1 . The notation δ G ( v ) [ ζ ] \delta\mathcal{G}(v)[\zeta] δ G ( v ) [ ζ ] denotes its linearization in the direction ζ \zeta ζ satisfies
⟨ δ G ( v ) [ ζ ] , η ⟩ = δ 2 E ( v ) [ ζ , η ] . \langle\delta\mathcal{G}(v)[\zeta],\eta\rangle=\delta^{2}\mathcal{E}(v)[\zeta,\eta]. ⟨ δ G ( v ) [ ζ ] , η ⟩ = δ 2 E ( v ) [ ζ , η ] .
The graph variation in the first component is
∂ ξ Ψ 1 ( s , U ( s ) ) [ ζ 1 ( s ) ] . \partial_{\xi}\Psi_1(s,U(s))[\zeta_1(s)]. ∂ ξ Ψ 1 ( s , U ( s )) [ ζ 1 ( s )] .
Its projection to the normal plane of the current curve is an isomorphism at v = 0 v=0 v = 0 , and remains so for small ∥ v ∥ C 1 \lVert v\rVert_{C^1} ∥ v ∥ C 1 . The second component has the same property. For ∥ v ∥ C 1 \lVert v\rVert_{C^1} ∥ v ∥ C 1 small, the normal projection of the graph variations onto the normal bundles of the current curves is pointwise invertible. Hence, Proposition 2.3 gives
G ( v ) = 0 ⟺ Graph H ( v ) is weakly critical . (45) \mathcal{G}(v)=0 \quad\Longleftrightarrow\quad\operatorname{Graph}_{\mathrm{H}}(v)\text{ is weakly critical}. \tag*{(45)} G ( v ) = 0 ⟺ Graph H ( v ) is weakly critical . (45)
Lemma 5.1. There are C , r 1 > 0 C,r_1>0 C , r 1 > 0 such that, for v , w ∈ H 2 v,w\in\mathcal{H}^2 v , w ∈ H 2 with ∥ v ∥ C 1 , ∥ w ∥ C 1 < r 1 \lVert v\rVert_{C^1},\lVert w\rVert_{C^1}<r_1 ∥ v ∥ C 1 , ∥ w ∥ C 1 < r 1 , the linearization δ G ( v ) \delta\mathcal{G}(v) δ G ( v ) extends to a uniformly bounded map H 1 → H − 1 \mathcal{H}^1\to\mathcal{H}^{-1} H 1 → H − 1 and
∥ δ G ( v ) − δ G ( w ) ∥ Lin ( H 1 , H − 1 ) ≤ C ∥ v − w ∥ C 1 . (46) \lVert\delta\mathcal{G}(v)-\delta\mathcal{G}(w)\rVert_{\operatorname{Lin}(\mathcal{H}^1,\mathcal{H}^{-1})}\leq C\lVert v-w\rVert_{C^1}. \tag*{(46)} ∥ δ G ( v ) − δ G ( w ) ∥ Lin ( H 1 , H − 1 ) ≤ C ∥ v − w ∥ C 1 . (46)
Here Lin ( X , Y ) \operatorname{Lin}(X,Y) Lin ( X , Y ) is the space of bounded linear maps with its operator norm. Moreover, G ( 0 ) = 0 \mathcal{G}(0)=0 G ( 0 ) = 0 , δ G ( 0 ) = L H \delta\mathcal{G}(0)=\mathcal{L}_{\mathrm{H}} δ G ( 0 ) = L H , and
∥ G ( v ) − L H v ∥ H − 1 ≤ C ∥ v ∥ C 1 ∥ v ∥ H 1 . (47) \lVert\mathcal{G}(v)-\mathcal{L}_{\mathrm{H}}v\rVert_{H^{-1}}\leq C\lVert v\rVert_{C^1}\lVert v\rVert_{H^1}. \tag*{(47)} ∥ G ( v ) − L H v ∥ H − 1 ≤ C ∥ v ∥ C 1 ∥ v ∥ H 1 . (47)
Proof. In the normal-graph coordinates, the energy density is a smooth function of
( s , t , U , U ′ , V , V ′ ) . (s,t,U,U',V,V'). ( s , t , U , U ′ , V , V ′ ) .
For ∥ v ∥ C 1 < r \lVert v\rVert_{C^1}<r ∥ v ∥ C 1 < r , the speeds and the separation of the two components are uniformly bounded away from zero. Hence D q k f D_q^k f D q k f is uniformly bounded for k ≤ 3 k\leq3 k ≤ 3 , where q = ( U , U ′ , V , V ′ ) q=(U,U',V,V') q = ( U , U ′ , V , V ′ ) . Differentiating twice and using the product estimates of Lemma 2.2 gives the uniform H 1 × H 1 H^1\times H^1 H 1 × H 1 bound for the Hessian, while the mean-value theorem gives (46) . The complete coefficient formula is recorded in Section A.3. At v = 0 v=0 v = 0 the chart Hessian is the normal Hopf Hessian, so δ G ( 0 ) = L H \delta\mathcal{G}(0)=\mathcal{L}_{\mathrm{H}} δ G ( 0 ) = L H . Integration along τ v \tau v τ v now gives
G ( v ) − L H v = ∫ 0 1 ( δ G ( τ v ) − δ G ( 0 ) ) [ v ] d τ , \mathcal{G}(v)-\mathcal{L}_{\mathrm{H}}v=\int_0^1\bigl(\delta\mathcal{G}(\tau v)-\delta\mathcal{G}(0)\bigr)[v]\,d\tau, G ( v ) − L H v = ∫ 0 1 ( δ G ( τ v ) − δ G ( 0 ) ) [ v ] d τ ,
and (46) proves (47) . □
In particular, writing R ( v ) = G ( v ) − L H v \mathcal{R}(v)=\mathcal{G}(v)-\mathcal{L}_{\mathrm{H}}v R ( v ) = G ( v ) − L H v , we have
G ( v ) = L H v + R ( v ) , ∥ R ( v ) ∥ H − 1 ≤ C ∥ v ∥ C 1 ∥ v ∥ H 1 . (48) \mathcal{G}(v)=\mathcal{L}_{\mathrm{H}}v+\mathcal{R}(v),\qquad\lVert\mathcal{R}(v)\rVert_{H^{-1}}\leq C\lVert v\rVert_{C^1}\lVert v\rVert_{H^1}. \tag*{(48)} G ( v ) = L H v + R ( v ) , ∥ R ( v ) ∥ H − 1 ≤ C ∥ v ∥ C 1 ∥ v ∥ H 1 . (48)
Eight orthogonality conditions and the local slice
Write Ψ i − 1 = ( π i , ν i ) \Psi_i^{-1}=(\pi_i,\nu_i) Ψ i − 1 = ( π i , ν i ) . For Z = ( Z 1 , Z 2 ) Z=(Z_1,Z_2) Z = ( Z 1 , Z 2 ) close to H in C 1 C^1 C 1 , let
θ i = π i ∘ Z i , η i = ν i ∘ Z i . \theta_i=\pi_i\circ Z_i,\qquad\eta_i=\nu_i\circ Z_i. θ i = π i ∘ Z i , η i = ν i ∘ Z i .
Then θ i \theta_i θ i is a degree-one circle diffeomorphism close to the identity. The field of its normal-graph representative is
ng H ( Z ) : = ( η 1 ∘ θ 1 − 1 , η 2 ∘ θ 2 − 1 ) . (49) \operatorname{ng}_{\mathrm{H}}(Z):=(\eta_1\circ\theta_1^{-1},\eta_2\circ\theta_2^{-1}). \tag*{(49)} ng H ( Z ) := ( η 1 ∘ θ 1 − 1 , η 2 ∘ θ 2 − 1 ) . (49)
By Lemma A.1, this operation is continuous in C 1 C^1 C 1 and preserves H 2 H^2 H 2 regularity. Let P K H P_{\mathcal{K}_{\mathrm{H}}} P K H denote the L 2 L^2 L 2 -orthogonal projection onto the kernel of the Hopf Hessian. For A ∈ M \mathcal{A}\in\mathcal{M} A ∈ M near 0 0 0 , let Φ A \Phi_{\mathcal{A}} Φ A be the Möbius transformation induced by e A e^{\mathcal{A}} e A , and set
F ( A , w ) = P K H ng H ( Φ A ( Graph H ( w ) ) ) , w ∈ X 1 , \mathcal{F}(\mathcal{A},w)=P_{\mathcal{K}_{\mathrm{H}}}\operatorname{ng}_{\mathrm{H}}\bigl(\Phi_{\mathcal{A}}(\operatorname{Graph}_{\mathrm{H}}(w))\bigr),\qquad w\in\mathcal{X}^1, F ( A , w ) = P K H ng H ( Φ A ( Graph H ( w )) ) , w ∈ X 1 ,
where
X 1 = C 1 ( S 1 , P 2 ) ⊕ C 1 ( S 1 , P 1 ) . \mathcal{X}^1=C^1(\mathbb{S}^1,P_2)\oplus C^1(\mathbb{S}^1,P_1). X 1 = C 1 ( S 1 , P 2 ) ⊕ C 1 ( S 1 , P 1 ) .
The following lemma, motivated by the renormalization lemma in [6 ] , provides a local Möbius gauge near H: after a small Möbius transformation and reparametrization, a nearby link can be written as a normal graph over H with graph field orthogonal to K H \mathcal{K}_{H} K H .
Lemma 5.2. For every sufficiently small w ∈ X 1 w \in\mathcal{X}^{1} w ∈ X 1 , there is a unique sufficiently small A ( w ) ∈ M \mathcal{A}(w) \in\mathcal{M} A ( w ) ∈ M such that
v = ng H ( Φ A ( w ) ( Graph H ( w ) ) ) satisfies P K H v = 0. v=\operatorname{ng}_{H}\left(\Phi_{\mathcal{A}(w)}\left(\operatorname{Graph}_{H}(w)\right)\right)\quad\text{satisfies}\quad P_{\mathcal{K}_{H}}v=0. v = ng H ( Φ A ( w ) ( Graph H ( w ) ) ) satisfies P K H v = 0.
Moreover, A ( w ) \mathcal{A}(w) A ( w ) depends smoothly on w w w , v → 0 v \to0 v → 0 in C 1 C^{1} C 1 as w → 0 w \to0 w → 0 , and H 2 H^{2} H 2 -regularity is preserved.
Proof. By Lemma A.2, F \mathcal{F} F is smooth near ( 0 , 0 ) (0,0) ( 0 , 0 ) , with
F ( 0 , 0 ) = 0. \mathcal{F}(0,0)=0. F ( 0 , 0 ) = 0.
Moreover,
D A F ( 0 , 0 ) [ B ] = j ( B ) , B ∈ M . (50) D_{\mathcal{A}}\mathcal{F}(0,0)[\mathcal{B}]=j(\mathcal{B}),\qquad\mathcal{B}\in\mathcal{M}. \tag*{(50)} D A F ( 0 , 0 ) [ B ] = j ( B ) , B ∈ M . (50)
By Proposition 4.3, this is a linear isomorphism M → K H \mathcal{M}\to\mathcal{K}_{H} M → K H . The implicit-function theorem, with w w w as a Banach-space parameter, gives A = A ( w ) \mathcal{A}=\mathcal{A}(w) A = A ( w ) and F ( A ( w ) , w ) = 0 \mathcal{F}(\mathcal{A}(w),w)=0 F ( A ( w ) , w ) = 0 . The remaining continuity and regularity statements follow from Lemma A.1.
Proof of local isolation
Theorem 5.3. For every 0 < α < 1 0<\alpha<1 0 < α < 1 there is a C 1 C^{1} C 1 neighborhood of the Hopf orbit such that every weakly critical pair in this neighborhood belongs to G ⋅ H \mathcal{G}\cdot H G ⋅ H .
Proof. By Lemma A.1, a pair Γ ∈ A \Gamma\in\mathcal{A} Γ ∈ A sufficiently close to H in C 1 C^{1} C 1 can be separately reparametrized as Graph H ( w ) \operatorname{Graph}_{H}(w) Graph H ( w ) , with w ∈ H 2 w\in\mathcal{H}^{2} w ∈ H 2 small in C 1 C^{1} C 1 . Apply Lemma 5.2 to obtain a Möbius-equivalent graph Graph H ( v ) \operatorname{Graph}_{H}(v) Graph H ( v ) with v ⊥ L 2 K H v\perp_{L^{2}}\mathcal{K}_{H} v ⊥ L 2 K H . The continuity statements in the two lemmas allow us to shrink the original neighborhood so that ∥ v ∥ C 1 < r ∗ \lVert v\rVert_{C^{1}}<r_{*} ∥ v ∥ C 1 < r ∗ , where r ∗ < r 1 r_{*}<r_{1} r ∗ < r 1 and C r ∗ < 4 π / 5 Cr_{*}<4\pi/5 C r ∗ < 4 π /5 for the constants in Lemma 5.1. Weak criticality is preserved by both operations, by Lemma 2.1. Hence (5.3) and (5.6) imply L H v = − R ( v ) \mathcal{L}_{H}v=-\mathcal{R}(v) L H v = − R ( v ) . Pairing with v v v gives
4 π 5 ∥ v ∥ H 1 2 ≤ δ 2 E H ( v ) = − ⟨ R ( v ) , v ⟩ ≤ C r ∗ ∥ v ∥ H 1 2 . (51) \frac{4\pi}{5}\lVert v\rVert_{H^{1}}^{2}\leq\delta^{2}E_{H}(v)=-\langle\mathcal{R}(v),v\rangle\leq Cr_{*}\lVert v\rVert_{H^{1}}^{2}. \tag*{(51)} 5 4 π ∥ v ∥ H 1 2 ≤ δ 2 E H ( v ) = − ⟨ R ( v ) , v ⟩ ≤ C r ∗ ∥ v ∥ H 1 2 . (51)
Since C r ∗ < 4 π / 5 Cr_{*}<4\pi/5 C r ∗ < 4 π /5 , one has v = 0 v=0 v = 0 . Undoing the transformations proves Γ ∈ G ⋅ H \Gamma\in\mathcal{G}\cdot H Γ ∈ G ⋅ H . Finally, each fixed element of G \mathcal{G} G acts by a C 1 C^{1} C 1 homeomorphism and preserves weak criticality, which proves the statement on the union of translates.
The critical-value gap and local consequences
Proof of the critical-value gap
Proof of Theorem 1.2. Suppose that no positive gap exists. Then, there is a sequence of non-split, weakly critical pairs Γ n ∈ A ∖ ( G ⋅ H ) \Gamma_{n}\in\mathcal{A}\setminus(\mathcal{G}\cdot H) Γ n ∈ A ∖ ( G ⋅ H ) such that
2 π 2 ≤ E ( Γ n ) ≤ 2 π 2 + 1 n . 2\pi^{2}\leq E(\Gamma_{n})\leq2\pi^{2}+\frac{1}{n}. 2 π 2 ≤ E ( Γ n ) ≤ 2 π 2 + n 1 .
By Proposition 3.8, a subsequence converges smoothly to H after common Möbius transformations and separate reparametrizations. For large n n n the resulting pairs lie in the neighborhood of Theorem 5.3, and hence belong to G ⋅ H \mathcal{G}\cdot H G ⋅ H . Invariance of this orbit gives a contradiction.
Quantitative local consequences
The Hessian calculation also gives the following quantitative local statements. They use the fixed parameter measure and the H 1 H^1 H 1 norm in (26) ; in particular, the L 2 L^2 L 2 spectral gap and the H 1 H^1 H 1 coercivity constant are different constants.
Corollary 6.1 (Normal spectrum). As a self-adjoint operator on H 0 \mathcal{H}^0 H 0 with domain H 2 \mathcal{H}^2 H 2 , L H \mathcal{L}_{H} L H has spectrum
spec ( L H ) = { 0 , 2 π } ∪ { π n 2 : n = 2 , 3 , … } . (52) \operatorname{spec}(\mathcal{L}_{H}) = \{0, 2\pi\} \cup\{\pi n^2 : n = 2, 3, \ldots\}. \tag*{(52)} spec ( L H ) = { 0 , 2 π } ∪ { π n 2 : n = 2 , 3 , … } . (52)
The real multiplicities are 8 for 0 0 0 , 4 for 2 π 2\pi 2 π , and 8 for each π n 2 \pi n^2 π n 2 , n ≥ 2 n \ge2 n ≥ 2 .
Proof. The kernel is (30) . On a transverse first mode W A = ( A x , A T y ) W_A=(Ax,A^{T}y) W A = ( A x , A T y ) , the identity (28) gives
∫ 0 2 π y ( t ) ⟨ x ( s ) , A T y ( t ) ⟩ d t = π A x ( s ) , \int_{0}^{2\pi} y(t)\left\langle x(s),A^{T}y(t)\right\rangle\,\mathrm{d}t = \pi A x(s), ∫ 0 2 π y ( t ) ⟨ x ( s ) , A T y ( t ) ⟩ d t = π A x ( s ) ,
and similarly on the other component. Thus (40) gives L H W A = 2 π W A \mathcal{L}_{H}W_A=2\pi W_A L H W A = 2 π W A . On every frequency n ≥ 2 n \ge2 n ≥ 2 , the finite-rank integral term vanishes and the operator is − π ∂ 2 -\pi\partial^2 − π ∂ 2 , with eigenvalue π n 2 \pi n^2 π n 2 . The Fourier decomposition is complete. At each such frequency there are two trigonometric modes, two normal directions, and two components, giving multiplicity eight.
Corollary 6.2 (Local quadratic orbital stability). There are c , C , r 2 > 0 c,C,r_2>0 c , C , r 2 > 0 such that every v ∈ H 2 v\in\mathcal{H}^2 v ∈ H 2 with v ⊥ K H v\perp\mathcal{K}_{H} v ⊥ K H and ∥ v ∥ C 1 < r 2 \lVert v\rVert_{C^1}<r_2 ∥ v ∥ C 1 < r 2 satisfies
c ∥ v ∥ H 1 2 ≤ E ( Graph H ( v ) ) − 2 π 2 ≤ C ∥ v ∥ H 1 2 . (53) c\lVert v\rVert_{H^1}^2 \le E(\operatorname{Graph}_{H}(v))-2\pi^2 \le C\lVert v\rVert_{H^1}^2. \tag*{(53)} c ∥ v ∥ H 1 2 ≤ E ( Graph H ( v )) − 2 π 2 ≤ C ∥ v ∥ H 1 2 . (53)
Proof. Since δ E ( 0 ) = 0 \delta\mathcal{E}(0)=0 δ E ( 0 ) = 0 , Taylor’s formula gives
E ( v ) − E ( 0 ) = ∫ 0 1 ( 1 − τ ) δ 2 E ( τ v ) [ v , v ] d τ . (54) \mathcal{E}(v)-\mathcal{E}(0)=\int_{0}^{1}(1-\tau)\delta^2\mathcal{E}(\tau v)[v,v]\,\mathrm{d}\tau. \tag*{(54)} E ( v ) − E ( 0 ) = ∫ 0 1 ( 1 − τ ) δ 2 E ( τ v ) [ v , v ] d τ . (54)
By (46) , decreasing r 2 r_2 r 2 makes
∣ δ 2 E ( τ v ) [ v , v ] − δ 2 E H ( v ) ∣ ≤ 2 π 5 ∥ v ∥ H 1 2 \left|\delta^2\mathcal{E}(\tau v)[v,v]-\delta^2E_H(v)\right|\le\frac{2\pi}{5}\lVert v\rVert_{H^1}^2 δ 2 E ( τ v ) [ v , v ] − δ 2 E H ( v ) ≤ 5 2 π ∥ v ∥ H 1 2
for 0 ≤ τ ≤ 1 0\le\tau\le1 0 ≤ τ ≤ 1 . Coercivity (32) gives the lower bound, and the uniform bilinear bound in Lemma 5.1 gives the upper bound.
Corollary 6.3 (Inverse first-variation estimate). After decreasing r 2 r_2 r 2 , every v ∈ H 2 v\in\mathcal{H}^2 v ∈ H 2 with v ⊥ K H v\perp\mathcal{K}_{H} v ⊥ K H and ∥ v ∥ C 1 < r 2 \lVert v\rVert_{C^1}<r_2 ∥ v ∥ C 1 < r 2 satisfies
∥ v ∥ H 1 ≤ C ∥ P ⊥ G ( v ) ∥ H − 1 , (55) \lVert v\rVert_{H^1}\le C\lVert P_{\perp}\mathcal{G}(v)\rVert_{H^{-1}}, \tag*{(55)} ∥ v ∥ H 1 ≤ C ∥ P ⊥ G ( v ) ∥ H − 1 , (55)
where P ⊥ P_{\perp} P ⊥ on the dual space is defined in (41) . Consequently, on this slice,
( E ( Graph H ( v ) ) − 2 π 2 ) 1 / 2 ≤ C ∥ P ⊥ G ( v ) ∥ H − 1 . (56) \left(E(\operatorname{Graph}_{H}(v))-2\pi^2\right)^{1/2}\le C\lVert P_{\perp}\mathcal{G}(v)\rVert_{H^{-1}}. \tag*{(56)} ( E ( Graph H ( v )) − 2 π 2 ) 1/2 ≤ C ∥ P ⊥ G ( v ) ∥ H − 1 . (56)
Proof. The distribution L H v \mathcal{L}_{H}v L H v annihilates K H \mathcal{K}_{H} K H , by symmetry of B H B_H B H . Thus (48) implies
L H v = P ⊥ G ( v ) − P ⊥ R ( v ) . \mathcal{L}_{H}v=P_{\perp}\mathcal{G}(v)-P_{\perp}\mathcal{R}(v). L H v = P ⊥ G ( v ) − P ⊥ R ( v ) .
Use (42) , boundedness of the dual projection, and (47) to obtain
∥ v ∥ H 1 ≤ C 0 ∥ P ⊥ G ( v ) ∥ H − 1 + C 1 ∥ v ∥ C 1 ∥ v ∥ H 1 . \lVert v\rVert_{H^1}\le C_0\lVert P_{\perp}\mathcal{G}(v)\rVert_{H^{-1}}+C_1\lVert v\rVert_{C^1}\lVert v\rVert_{H^1}. ∥ v ∥ H 1 ≤ C 0 ∥ P ⊥ G ( v ) ∥ H − 1 + C 1 ∥ v ∥ C 1 ∥ v ∥ H 1 .
The final term is absorbed for small r 2 r_2 r 2 , proving (55) . The upper bound in (53) then gives (56) .
These estimates describe the local Morse–Bott behavior in the H 1 H^1 H 1 –H − 1 H^{-1} H − 1 scale on the transverse slice.
Appendix A. Analytic details of the local slice
This appendix proves the coordinate assertions used in the local slice and gives the coefficient estimate underlying the Taylor remainder. All changes of parameter are between fixed circles; no differentiability of the operation that assigns a reparametrized curve is required.
Normal-graph coordinates and Sobolev reparametrization
The maps (43) –(44) have explicit inverses. For example, if z = z 1 + z 2 ∈ S 3 z=z_1+z_2\in\mathbb{S}^3 z = z 1 + z 2 ∈ S 3 , with z i ∈ P i z_i\in P_i z i ∈ P i , is close to the first Hopf circle, then
x ( π 1 ( z ) ) = z 1 ∣ z 1 ∣ , ν 1 ( z ) = arcsin ∣ z 2 ∣ ∣ z 2 ∣ z 2 . x(\pi_1(z))=\frac{z_1}{\lvert z_1\rvert},\qquad\nu_1(z)=\frac{\arcsin\lvert z_2\rvert}{\lvert z_2\rvert}z_2. x ( π 1 ( z )) = ∣ z 1 ∣ z 1 , ν 1 ( z ) = ∣ z 2 ∣ arcsin ∣ z 2 ∣ z 2 .
The quotient extends smoothly at z 2 = 0 z_2=0 z 2 = 0 by the value 1 1 1 . Indeed, arcsin r / r \arcsin r/r arcsin r / r is a smooth function of r 2 r^2 r 2 near zero. The second inverse is obtained by exchanging the planes. In particular, on the core circles,
d ν 1 ∣ x ( s ) = P P 2 , d ν 2 ∣ y ( t ) = P P 1 . (57) \mathrm{d}\nu_1\big|_{x(s)}=P_{P_2},\qquad\mathrm{d}\nu_2\big|_{y(t)}=P_{P_1}. \tag*{(57)} d ν 1 x ( s ) = P P 2 , d ν 2 y ( t ) = P P 1 . (57)
Lemma A.1 (Graph extraction). On a sufficiently small C 1 C^1 C 1 neighborhood of H \mathrm{H} H , the functions
θ i = π i ∘ Z i , η i = ν i ∘ Z i (58) \theta_i=\pi_i\circ Z_i,\qquad\eta_i=\nu_i\circ Z_i \tag*{(58)} θ i = π i ∘ Z i , η i = ν i ∘ Z i (58)
satisfy θ i ′ > 1 / 2 \theta_i'>1/2 θ i ′ > 1/2 , after taking lifts close to the identity. They define degree-one circle diffeomorphisms and
u i = η i ∘ θ i − 1 , Z i ( r ) = Ψ i ( θ i ( r ) , u i ( θ i ( r ) ) ) . u_i=\eta_i\circ\theta_i^{-1},\qquad Z_i(r)=\Psi_i\bigl(\theta_i(r),u_i(\theta_i(r))\bigr). u i = η i ∘ θ i − 1 , Z i ( r ) = Ψ i ( θ i ( r ) , u i ( θ i ( r )) ) .
The assignment Z ↦ ng H ( Z ) = ( u 1 , u 2 ) Z\mapsto\operatorname{ng}_{\mathrm{H}}(Z)=(u_1,u_2) Z ↦ ng H ( Z ) = ( u 1 , u 2 ) is continuous in C 1 C^1 C 1 , with ng H ( H ) = 0 \operatorname{ng}_{\mathrm{H}}(\mathrm{H})=0 ng H ( H ) = 0 . If Z ∈ H 2 Z\in H^2 Z ∈ H 2 , then θ i \theta_i θ i , θ i − 1 \theta_i^{-1} θ i − 1 , and u i u_i u i belong to H 2 H^2 H 2 .
Proof. At Z = H Z=\mathrm{H} Z = H , θ i ( r ) = r \theta_i(r)=r θ i ( r ) = r and η i = 0 \eta_i=0 η i = 0 . Smoothness of the tubular coordinates gives the lower derivative bound and the identity θ i ( r + 2 π ) = θ i ( r ) + 2 π \theta_i(r+2\pi)=\theta_i(r)+2\pi θ i ( r + 2 π ) = θ i ( r ) + 2 π on a small C 1 C^1 C 1 neighborhood. Thus each θ i \theta_i θ i is a degree-one diffeomorphism.
For completeness, suppose Z ( n ) → Z Z^{(n)}\to Z Z ( n ) → Z in C 1 C^1 C 1 . Smooth pointwise composition gives θ i ( n ) → θ i \theta_i^{(n)}\to\theta_i θ i ( n ) → θ i and η i ( n ) → η i \eta_i^{(n)}\to\eta_i η i ( n ) → η i in C 1 C^1 C 1 . The inverse lifts are uniformly Lipschitz, with constant at most 2 2 2 , so their uniform convergence follows from that of the original lifts. The formula
( ( θ i ( n ) ) − 1 ) ′ = 1 ( θ i ( n ) ) ′ ∘ ( θ i ( n ) ) − 1 \bigl((\theta_i^{(n)})^{-1}\bigr)'=\frac{1}{(\theta_i^{(n)})'\circ(\theta_i^{(n)})^{-1}} ( ( θ i ( n ) ) − 1 ) ′ = ( θ i ( n ) ) ′ ∘ ( θ i ( n ) ) − 1 1
and uniform continuity of the limiting derivative give convergence in C 1 C^1 C 1 . Applying the chain rule to u i ( n ) u_i^{(n)} u i ( n ) proves u i ( n ) → u i u_i^{(n)}\to u_i u i ( n ) → u i in C 1 C^1 C 1 .
If Z ∈ H 2 Z\in H^2 Z ∈ H 2 , the ordinary Sobolev chain rule gives θ i , η i ∈ H 2 \theta_i,\eta_i\in H^2 θ i , η i ∈ H 2 . Write ψ i = θ i − 1 \psi_i=\theta_i^{-1} ψ i = θ i − 1 . Almost everywhere,
ψ i ′ ′ = − θ i ′ ′ ∘ ψ i ( θ i ′ ∘ ψ i ) 3 , u i ′ ′ = ( η i ′ ′ ∘ ψ i ) ( ψ i ′ ) 2 + ( η i ′ ∘ ψ i ) ψ i ′ ′ . \psi_i''=-\frac{\theta_i''\circ\psi_i}{(\theta_i'\circ\psi_i)^3},\qquad
u_i''=(\eta_i''\circ\psi_i)(\psi_i')^2+(\eta_i'\circ\psi_i)\psi_i''. ψ i ′′ = − ( θ i ′ ∘ ψ i ) 3 θ i ′′ ∘ ψ i , u i ′′ = ( η i ′′ ∘ ψ i ) ( ψ i ′ ) 2 + ( η i ′ ∘ ψ i ) ψ i ′′ .
The derivative bounds for θ i \theta_i θ i and change of variables imply ψ i ′ ′ ∈ L 2 \psi_i''\in L^2 ψ i ′′ ∈ L 2 . The first term in u i ′ ′ u_i'' u i ′′ lies in L 2 L^2 L 2 by the same change of variables, and the second does because η i ′ \eta_i' η i ′ and ψ i ′ \psi_i' ψ i ′ are bounded. This proves the H 2 H^2 H 2 assertions. □
Smoothness of the eight scalar constraints
We define M ( Z ) M(Z) M ( Z ) by
M ( Z ) ≔ P K H ng H ( Z ) M(Z) \coloneqq P_{\mathcal{K}_H}\operatorname{ng}_H(Z) M ( Z ) : = P K H ng H ( Z )
for Z Z Z sufficiently C 1 C^1 C 1 -close to H H H into the finite-dimensional space K H \mathcal{K}_H K H .
Lemma A.2 (Smooth orthogonality constraints). The map M M M is smooth and satisfies
d M H [ ζ ] = P K H ( P 2 ζ 1 , P 1 ζ 2 ) . dM_H[\zeta] = P_{\mathcal{K}_H}\left(P_2\zeta_1,P_1\zeta_2\right). d M H [ ζ ] = P K H ( P 2 ζ 1 , P 1 ζ 2 ) .
Consequently, for
F ( A , w ) = M ( Φ A ( Graph H ( w ) ) ) , \mathcal{F}(\mathcal{A},w) = M\left(\Phi_{\mathcal{A}}(\operatorname{Graph}_H(w))\right), F ( A , w ) = M ( Φ A ( Graph H ( w )) ) ,
one has
D A F ( 0 , 0 ) [ B ] = j ( B ) . (59) D_{\mathcal{A}}\mathcal{F}(0,0)[\mathcal{B}] = j(\mathcal{B}). \tag*{(59)} D A F ( 0 , 0 ) [ B ] = j ( B ) . (59)
Proof. Let k 1 , … , k 8 k_1,\ldots,k_8 k 1 , … , k 8 be a smooth L 2 L^2 L 2 -orthonormal basis of K H \mathcal{K}_H K H with k a = ( k a , 1 , k a , 2 ) k_a=(k_{a,1},k_{a,2}) k a = ( k a , 1 , k a , 2 ) . Define
m a ( Z ) = ∑ i = 1 2 ∫ 0 2 π ⟨ η i ( r ) , k a , i ( θ i ( r ) ) ⟩ θ i ′ ( r ) d r . m_a(Z)=\sum_{i=1}^{2}\int_0^{2\pi}\left\langle\eta_i(r),k_{a,i}(\theta_i(r))\right\rangle\theta_i'(r)\,dr. m a ( Z ) = i = 1 ∑ 2 ∫ 0 2 π ⟨ η i ( r ) , k a , i ( θ i ( r )) ⟩ θ i ′ ( r ) d r .
From the definition of Ψ − 1 \Psi^{-1} Ψ − 1 , changing variables q = θ i ( r ) q=\theta_i(r) q = θ i ( r ) in the formula for m a m_a m a yields
m a ( Z ) = ∑ i = 1 2 ∫ 0 2 π ⟨ η i ( θ i − 1 ( q ) ) , k a , i ( q ) ⟩ d q = ( ng H ( Z ) , k a ) L 2 . m_a(Z)=\sum_{i=1}^{2}\int_0^{2\pi}\left\langle\eta_i\left(\theta_i^{-1}(q)\right),k_{a,i}(q)\right\rangle\,dq=(\operatorname{ng}_H(Z),k_a)_{L^2}. m a ( Z ) = i = 1 ∑ 2 ∫ 0 2 π ⟨ η i ( θ i − 1 ( q ) ) , k a , i ( q ) ⟩ d q = ( ng H ( Z ) , k a ) L 2 .
Thus, we have
M ( Z ) = ∑ a = 1 8 m a ( Z ) k a . M(Z)=\sum_{a=1}^{8}m_a(Z)k_a. M ( Z ) = a = 1 ∑ 8 m a ( Z ) k a .
Indeed, the integrand can be written as
f a , i ( z , p ) = ⟨ ν i ( z ) , k a , i ( π i ( z ) ) ⟩ d π i ∣ z [ p ] , f_{a,i}(z,p)=\left\langle\nu_i(z),k_{a,i}(\pi_i(z))\right\rangle d\pi_i|_z[p], f a , i ( z , p ) = ⟨ ν i ( z ) , k a , i ( π i ( z )) ⟩ d π i ∣ z [ p ] ,
which is smooth in ( z , p ) (z,p) ( z , p ) . Here angular coordinates are interpreted through local lifts; the expression is independent of the lift by periodicity. Thus each m a m_a m a , and hence M M M , is smooth in the C 1 C^1 C 1 topology.
Now let Z τ Z_\tau Z τ be a variation of H H H with velocity ζ \zeta ζ . At τ = 0 \tau=0 τ = 0 one has η i = 0 \eta_i=0 η i = 0 , θ i ( r ) = r \theta_i(r)=r θ i ( r ) = r , and θ i ′ = 1 \theta_i'=1 θ i ′ = 1 . In the product rule for m a m_a m a , the terms differentiating k a , i ( θ i ) k_{a,i}(\theta_i) k a , i ( θ i ) or θ i ′ \theta_i' θ i ′ therefore vanish. The remaining term is
d d τ m a ( Z τ ) ∣ τ = 0 = ∑ i ∫ 0 2 π ⟨ d ν i ∣ H i ( r ) [ ζ i ( r ) ] , k a , i ( r ) ⟩ d r . \left.\frac{d}{d\tau}m_a(Z_\tau)\right|_{\tau=0}=\sum_i\int_0^{2\pi}\left\langle d\nu_i|_{H_i(r)}[\zeta_i(r)],k_{a,i}(r)\right\rangle\,dr. d τ d m a ( Z τ ) τ = 0 = i ∑ ∫ 0 2 π ⟨ d ν i ∣ H i ( r ) [ ζ i ( r )] , k a , i ( r ) ⟩ d r .
(57) proves (59) . For Z τ = Φ τ B ( H ) Z_\tau=\Phi_{\tau\mathcal{B}}(H) Z τ = Φ τ B ( H ) , the pair of normal velocities is j ( B ) j(\mathcal{B}) j ( B ) by (4.12). Since j ( B ) ∈ K H j(\mathcal{B})\in\mathcal{K}_H j ( B ) ∈ K H , its projection onto K H \mathcal{K}_H K H is itself. This proves (5.8). The smoothness of F \mathcal{F} F follows from that of M M M , the graph map, and the finite-dimensional Möbius action. □
The implicit-function theorem in Lemma 5.2 yields a smooth parameter A ( w ) \mathcal{A}(w) A ( w ) in M \mathcal{M} M . It does not assert that the full field ng H ( Φ A ( w ) ( Graph H ( w ) ) ) \operatorname{ng}_H(\Phi_{\mathcal{A}(w)}(\operatorname{Graph}_H(w))) ng H ( Φ A ( w ) ( Graph H ( w ))) depends differentiably on w w w in C 1 C^1 C 1 . The continuity proved in Lemma A.1 is sufficient to make that field small in the proof of Theorem 5.3.
The coefficient estimate for the Hessian
The ordinary chain rule gives
X U ′ = ∂ s Ψ 1 ( s , U ) + ∂ ξ Ψ 1 ( s , U ) [ U ′ ] , X V ′ = ∂ t Ψ 2 ( t , V ) + ∂ η Ψ 2 ( t , V ) [ V ′ ] . X'_U = \partial_s\Psi_1(s,U) + \partial_\xi\Psi_1(s,U)[U'], \qquad X'_V = \partial_t\Psi_2(t,V) + \partial_\eta\Psi_2(t,V)[V']. X U ′ = ∂ s Ψ 1 ( s , U ) + ∂ ξ Ψ 1 ( s , U ) [ U ′ ] , X V ′ = ∂ t Ψ 2 ( t , V ) + ∂ η Ψ 2 ( t , V ) [ V ′ ] .
Hence
E ( v ) = ∬ f ( s , t , U ( s ) , U ′ ( s ) , V ( t ) , V ′ ( t ) ) d s d t , (60) \mathcal{E}(v) = \iint f(s,t,U(s),U'(s),V(t),V'(t))\,\mathrm{d}s\,\mathrm{d}t, \tag*{(60)} E ( v ) = ∬ f ( s , t , U ( s ) , U ′ ( s ) , V ( t ) , V ′ ( t )) d s d t , (60)
where
f ( s , t , ξ , p , η , q ) = ∣ ∂ s Ψ 1 ( s , ξ ) + ∂ ξ Ψ 1 ( s , ξ ) [ p ] ∣ ∣ ∂ t Ψ 2 ( t , η ) + ∂ η Ψ 2 ( t , η ) [ q ] ∣ ∣ Ψ 1 ( s , ξ ) − Ψ 2 ( t , η ) ∣ 2 . (61) f(s,t,\xi,p,\eta,q) = \frac{\left\lvert\partial_s\Psi_1(s,\xi) + \partial_\xi\Psi_1(s,\xi)[p]\right\rvert\left\lvert\partial_t\Psi_2(t,\eta) + \partial_\eta\Psi_2(t,\eta)[q]\right\rvert}{\left\lvert\Psi_1(s,\xi)-\Psi_2(t,\eta)\right\rvert^2}. \tag*{(61)} f ( s , t , ξ , p , η , q ) = ∣ Ψ 1 ( s , ξ ) − Ψ 2 ( t , η ) ∣ 2 ∣ ∂ s Ψ 1 ( s , ξ ) + ∂ ξ Ψ 1 ( s , ξ ) [ p ] ∣ ∣ ∂ t Ψ 2 ( t , η ) + ∂ η Ψ 2 ( t , η ) [ q ] ∣ . (61)
On a sufficiently small C 1 C^1 C 1 ball the speeds and separation are uniformly positive. Thus the first three derivatives of f f f with respect to q = ( ξ , p , η , q ) \mathbf{q}=(\xi,p,\eta,q) q = ( ξ , p , η , q ) are uniformly bounded on the compact set of values under consideration.
To display the differentiation, abbreviate the tuple of values and first derivatives by z v ( s , t ) = ( U ( s ) , U ′ ( s ) , V ( t ) , V ′ ( t ) ) z_v(s,t)=(U(s),U'(s),V(t),V'(t)) z v ( s , t ) = ( U ( s ) , U ′ ( s ) , V ( t ) , V ′ ( t )) , z ζ ( s , t ) = ( ζ 1 ( s ) , ζ 1 ′ ( s ) , ζ 2 ( t ) , ζ 2 ′ ( t ) ) z_\zeta(s,t)=(\zeta_1(s),\zeta'_1(s),\zeta_2(t),\zeta'_2(t)) z ζ ( s , t ) = ( ζ 1 ( s ) , ζ 1 ′ ( s ) , ζ 2 ( t ) , ζ 2 ′ ( t )) .
Twice differentiating the finite-dimensional integrand gives
⟨ δ G ( v ) [ ζ ] , η ⟩ = ∬ ∂ q 2 f ( s , t , z v ) [ z ζ , z η ] d s d t . (62) \langle\delta\mathcal{G}(v)[\zeta],\eta\rangle= \iint\partial_{\mathbf{q}}^2 f(s,t,z_v)[z_\zeta,z_\eta]\,\mathrm{d}s\,\mathrm{d}t. \tag*{(62)} ⟨ δ G ( v ) [ ζ ] , η ⟩ = ∬ ∂ q 2 f ( s , t , z v ) [ z ζ , z η ] d s d t . (62)
Each summand is a bounded coefficient times either two factors depending on the same parameter or one factor depending on each parameter. In the former case, Cauchy–Schwarz gives
∬ ∣ b ( s , t ) a ( s ) c ( s ) ∣ d s d t ≤ 2 π ∥ b ∥ ∞ ∥ a ∥ L 2 ∥ c ∥ L 2 ; \iint\lvert b(s,t)a(s)c(s)\rvert\,\mathrm{d}s\,\mathrm{d}t \le2\pi\lVert b\rVert_\infty\lVert a\rVert_{L^2}\lVert c\rVert_{L^2}; ∬ ∣ b ( s , t ) a ( s ) c ( s )∣ d s d t ≤ 2 π ∥ b ∥ ∞ ∥ a ∥ L 2 ∥ c ∥ L 2 ;
in the latter case,
∬ ∣ b ( s , t ) a ( s ) c ( t ) ∣ d s d t ≤ ∥ b ∥ ∞ ∥ a ∥ L 1 ∥ c ∥ L 1 ≤ 2 π ∥ b ∥ ∞ ∥ a ∥ L 2 ∥ c ∥ L 2 . \iint\lvert b(s,t)a(s)c(t)\rvert\,\mathrm{d}s\,\mathrm{d}t \le\lVert b\rVert_\infty\lVert a\rVert_{L^1}\lVert c\rVert_{L^1} \le2\pi\lVert b\rVert_\infty\lVert a\rVert_{L^2}\lVert c\rVert_{L^2}. ∬ ∣ b ( s , t ) a ( s ) c ( t )∣ d s d t ≤ ∥ b ∥ ∞ ∥ a ∥ L 1 ∥ c ∥ L 1 ≤ 2 π ∥ b ∥ ∞ ∥ a ∥ L 2 ∥ c ∥ L 2 .
The factors may be field values or first derivatives. Summing these estimates bounds (62) by C ∥ ζ ∥ H 1 ∥ η ∥ H 1 C\lVert\zeta\rVert_{H^1}\lVert\eta\rVert_{H^1} C ∥ ζ ∥ H 1 ∥ η ∥ H 1 . This proves the claimed extension to Lin ( H 1 , H − 1 ) \operatorname{Lin}(\mathcal{H}^1,\mathcal{H}^{-1}) Lin ( H 1 , H − 1 ) .
Finally, the bound for the third derivatives of f f f and the mean-value theorem give
sup s , t ∣ ∂ q 2 f ( s , t , z v ) − ∂ q 2 f ( s , t , z w ) ∣ ≤ C ∥ v − w ∥ C 1 . \sup_{s,t}\left\lvert\partial_{\mathbf{q}}^2f(s,t,z_v)-\partial_{\mathbf{q}}^2f(s,t,z_w)\right\rvert\le C\lVert v-w\rVert_{C^1}. s , t sup ∂ q 2 f ( s , t , z v ) − ∂ q 2 f ( s , t , z w ) ≤ C ∥ v − w ∥ C 1 .
Applying the same product estimates to this difference proves (5.4). In particular, its constant depends only on the chosen C 1 C^1 C 1 neighborhood, not on the H 2 H^2 H 2 norms of its base graphs.
References
Shanghai Institute for Mathematics and Interdisciplinary Sciences (SIMIS), Shanghai 200433, China Research Institute of Intelligent Complex Systems, Fudan University, Shanghai 200433, China Email address: chli@simis.cn
School of Mathematics and Statistics, Nanjing University of Information Science and Technology, Nanjing, China Email address: 003705@nuist.edu.cn