Introduction
The Sobolev inequality controls a function in terms of its gradient. For vector fields, replacing the gradient by the curl introduces an additional difficulty: the kernel of ∇ × \nabla\times ∇ × is infinite-dimensional, since it contains all gradients. A Sobolev–curl inequality can therefore control a field only modulo this kernel, and the denominator of the Sobolev quotient has to be minimized over a gauge class. For p = 2 p=2 p = 2 such a gauge-corrected inequality was first established by the author and Szulkin [21 ] .
We focus on the exponent p = 3 / 2 p=3/2 p = 3/2 , at which the quotient is conformally invariant, as in the work of Frank and Loss [9 , 10 , 11 ] . For vector fields u : R 3 → R 3 u:\mathbb{R}^3\to\mathbb{R}^3 u : R 3 → R 3 we consider the quotient
Q ( u ) = ∫ R 3 ∣ ∇ × u ∣ 3 / 2 d x ( inf ∇ × w = 0 , w ∈ L 3 ( R 3 , R 3 ) ∫ R 3 ∣ u + w ∣ 3 d x ) 1 / 2 ≥ S 3 / 2 , c u r l > 0 , (1) Q(u)=\frac{\int_{\mathbb{R}^3}|\nabla\times u|^{3/2}\,dx}{\left(\inf_{\substack{\nabla\times w=0,\;w\in L^3(\mathbb{R}^3,\mathbb{R}^3)}}\int_{\mathbb{R}^3}|u+w|^3\,dx\right)^{1/2}}\ge S_{3/2,\mathrm{curl}}>0, \tag*{(1)} Q ( u ) = ( inf ∇ × w = 0 , w ∈ L 3 ( R 3 , R 3 ) ∫ R 3 ∣ u + w ∣ 3 d x ) 1/2 ∫ R 3 ∣∇ × u ∣ 3/2 d x ≥ S 3/2 , curl > 0 , (1)
where S 3 / 2 , c u r l S_{3/2,\mathrm{curl}} S 3/2 , curl is the sharp Sobolev–curl constant and the inequality holds for every u ∈ L 3 ( R 3 ; R 3 ) u\in L^3(\mathbb{R}^3;\mathbb{R}^3) u ∈ L 3 ( R 3 ; R 3 ) with ∇ × u ∈ L 3 / 2 ( R 3 ; R 3 ∖ { 0 } ) \nabla\times u\in L^{3/2}(\mathbb{R}^3;\mathbb{R}^3\setminus\{0\}) ∇ × u ∈ L 3/2 ( R 3 ; R 3 ∖ { 0 }) ; see [9 , 10 , 11 , 22 ] . Fields in L 3 ( R 3 , R 3 ) L^3(\mathbb{R}^3,\mathbb{R}^3) L 3 ( R 3 , R 3 ) whose distributional curl belongs to L 3 / 2 ( R 3 , R 3 ) L^{3/2}(\mathbb{R}^3,\mathbb{R}^3) L 3/2 ( R 3 , R 3 ) are called finite-energy fields.
Symmetry classes play a central role in what follows. In cylindrical coordinates x = ( ρ cos θ , ρ sin θ , z ) x=(\rho\cos\theta,\rho\sin\theta,z) x = ( ρ cos θ , ρ sin θ , z ) , ρ > 0 \rho>0 ρ > 0 , we use the Euclidean orthonormal frame
( e ρ , e θ , e 3 ) = ( ( cos θ , sin θ , 0 ) , ( − sin θ , cos θ , 0 ) , ( 0 , 0 , 1 ) ) , (e_\rho,e_\theta,e_3)=((\cos\theta,\sin\theta,0),(-\sin\theta,\cos\theta,0),(0,0,1)), ( e ρ , e θ , e 3 ) = (( cos θ , sin θ , 0 ) , ( − sin θ , cos θ , 0 ) , ( 0 , 0 , 1 )) ,
pointing in the radial, azimuthal and axial directions. Let T \mathcal{T} T and S \mathcal{S} S denote, respectively, the toroidal and poloidal classes of finite-energy fields of the form
A ( x ) = a ( ρ , z ) e θ , U ( x ) = b ( ρ , z ) e ρ + c ( ρ , z ) e 3 A(x)=a(\rho,z)e_\theta,\qquad U(x)=b(\rho,z)e_\rho+c(\rho,z)e_3 A ( x ) = a ( ρ , z ) e θ , U ( x ) = b ( ρ , z ) e ρ + c ( ρ , z ) e 3
for scalar functions a , b , c a,b,c a , b , c . The restricted sharp constants are
S 3 / 2 , c u r l T : = inf A ∈ T ∇ × A ≠ 0 Q ( A ) , S 3 / 2 , c u r l S : = inf U ∈ S ∇ × U ≠ 0 Q ( U ) , S_{3/2,\mathrm{curl}}^{\mathcal{T}}:=\inf_{\substack{A\in\mathcal{T}\\ \nabla\times A\ne0}}Q(A),\qquad S_{3/2,\mathrm{curl}}^{\mathcal{S}}:=\inf_{\substack{U\in\mathcal{S}\\ \nabla\times U\ne0}}Q(U), S 3/2 , curl T := A ∈ T ∇ × A = 0 inf Q ( A ) , S 3/2 , curl S := U ∈ S ∇ × U = 0 inf Q ( U ) ,
and, since both classes are contained in the unrestricted admissible class,
0 < S 3 / 2 , c u r l ≤ min { S 3 / 2 , c u r l T , S 3 / 2 , c u r l S } . 0<S_{3/2,\mathrm{curl}}\le\min\left\{S_{3/2,\mathrm{curl}}^{\mathcal{T}},S_{3/2,\mathrm{curl}}^{\mathcal{S}}\right\}. 0 < S 3/2 , curl ≤ min { S 3/2 , curl T , S 3/2 , curl S } .
Whether equality holds is the analogue of the ground-state symmetry question raised for p = 2 p=2 p = 2 in [21 Section 7, (P3)] .
Frank and Loss posed the problem of determining the sharp constant and the optimizing fields in the critical vector inequality [10 Section 9, pp. 24–25] ; see also [11 p. 311, equation (8)] . Their attainment theorem [9 Theorem 1.1] does not identify the optimizing fields. The author and Szulkin [22 ] proved attainment for general 1 < p < 3 1<p<3 1 < p < 3 , also within the classes T \mathcal{T} T and S \mathcal{S} S . An explicit upper bound is furnished by the field
A T ( x ) = 3 ( 1 + ∣ x ∣ 2 ) 2 ( − x 2 , x 1 , 0 ) , A_{\mathcal{T}}(x)=\frac{3}{(1+|x|^2)^2}(-x_2,x_1,0), A T ( x ) = ( 1 + ∣ x ∣ 2 ) 2 3 ( − x 2 , x 1 , 0 ) ,
a combination of the Loss–Yau potential A L Y A_{\mathrm{LY}} A LY [18 ] and its reflection. Here A L Y = A L Y , e 3 A_{\mathrm{LY}}=A_{\mathrm{LY},e_3} A LY = A LY , e 3 belongs to the family
A L Y , w ( x ) = 3 ( 1 + ∣ x ∣ 2 ) 2 ( ( 1 − ∣ x ∣ 2 ) w + 2 ( w ⋅ x ) x + 2 w × x ) , w ∈ R 3 ∖ { 0 } , A_{\mathrm{LY},w}(x)=\frac{3}{(1+|x|^2)^2}\left((1-|x|^2)w+2(w\cdot x)x+2w\times x\right),\qquad w\in\mathbb{R}^3\setminus\{0\}, A LY , w ( x ) = ( 1 + ∣ x ∣ 2 ) 2 3 ( ( 1 − ∣ x ∣ 2 ) w + 2 ( w ⋅ x ) x + 2 w × x ) , w ∈ R 3 ∖ { 0 } ,
which solves equation (2) below for ∣ w ∣ = 4 / 3 |w|=4/3 ∣ w ∣ = 4/3 , as proved by the author and Szulkin [22 Theorem 1.3(d)] ; cf. [10 Section 9, (9.4)–(9.5)] . Bieganowski and the author [3 Theorem 1.1 and Section 6] computed
S 3 / 2 , c u r l T ≤ Q ( A T ) = 8 10 7 π < 4 π = Q ( A L Y ) , S_{3/2,\mathrm{curl}}^{\mathcal{T}}\le Q(A_{\mathcal{T}})=\frac{8\sqrt{10}}{7}\pi<4\pi=Q(A_{\mathrm{LY}}), S 3/2 , curl T ≤ Q ( A T ) = 7 8 10 π < 4 π = Q ( A LY ) ,
which gives a negative answer to the axisymmetric minimization question posed by Wang and Zhang [30 Section 6.2] . The sharp helicity inequality of Wang and Zhang [29 ] has a different denominator and does not determine the gauge-minimized constant considered here.
We identify the toroidal and poloidal extremals of the critical Sobolev–curl quotient through an exact hyperbolic representation, and construct them by a convergent gauge-canonical inverse iteration. We do not determine the unrestricted sharp constant: whether the restricted extremals are global minimizers remains open, and § 9 states the corresponding conjecture together with the available evidence.
Two structural strategies of this work
A minimizer, taken in its L 3 L^3 L 3 -minimal gauge and multiplied by a suitable positive constant, satisfies the Euler–Lagrange equation
∇ × ( ∣ ∇ × u ∣ − 1 / 2 ∇ × u ) = ∣ u ∣ u . (2) \nabla\times\left(\lvert\nabla\times u\rvert^{-1/2}\nabla\times u\right)=\lvert u\rvert u. \tag*{(2)} ∇ × ( ∣ ∇ × u ∣ − 1/2 ∇ × u ) = ∣ u ∣ u . (2)
The first fact is that (2) is self-dual : if u u u is a finite-energy weak solution, then its dual field
D ( u ) : = ∣ ∇ × u ∣ − 1 / 2 ∇ × u \mathcal{D}(u):=\lvert\nabla\times u\rvert^{-1/2}\nabla\times u D ( u ) := ∣ ∇ × u ∣ − 1/2 ∇ × u
is also a finite-energy weak solution, and D ( D ( u ) ) = u \mathcal{D}(\mathcal{D}(u))=u D ( D ( u )) = u ; the corresponding pair of dual equations already appears in [10 Section 9] . Thus D \mathcal{D} D is an involution on the set of finite-energy weak solutions; the new feature observed here is that it exchanges toroidal and poloidal solutions. § 2 shows that this is the self-dual case of a duality between the critical curl–curl equations with exponents p p p and p ^ = 3 p / ( 4 p − 3 ) \widehat{p}=3p/(4p-3) p = 3 p / ( 4 p − 3 ) , valid for all 1 < p < 3 1<p<3 1 < p < 3 , which interchanges the two cylindrical classes and relates the sharp constants by S p , c u r l = S p ^ , c u r l 1 / p ^ S_{p,\mathrm{curl}}=S_{\widehat{p},\mathrm{curl}}^{1/\widehat{p}} S p , curl = S p , curl 1/ p (Theorems 2.1 and 2.2 ). At p = 2 p=2 p = 2 it yields the new sharp inequality
∥ v ∥ 2 ≤ S 2 , c u r l − 1 / 2 ∥ ∇ × v ∥ 6 / 5 \lVert v\rVert_2\leq S_{2,\mathrm{curl}}^{-1/2}\lVert\nabla\times v\rVert_{6/5} ∥ v ∥ 2 ≤ S 2 , curl − 1/2 ∥ ∇ × v ∥ 6/5
for divergence-free fields, with optimizers obtained from p = 2 p=2 p = 2 ground states by taking curl and multiplying by a nonzero constant (Corollary 2.3).
The second new observation concerns the toroidal class. We write A ( ψ ) = ρ − 1 ψ ( ρ , z ) e θ A(\psi)=\rho^{-1}\psi(\rho,z)e_\theta A ( ψ ) = ρ − 1 ψ ( ρ , z ) e θ for the toroidal field associated with a scalar function ψ \psi ψ on the half-plane { ( ρ , z ) : ρ > 0 } \{(\rho,z):\rho>0\} {( ρ , z ) : ρ > 0 } , and we regard this half-plane as the half-plane model of H 2 \mathbb{H}^2 H 2 , with metric and area element
g = d ρ 2 + d z 2 ρ 2 , d V g = d ρ d z ρ 2 , ∣ ∇ g ψ ∣ g = ρ ∣ ∇ ψ ∣ . g=\frac{d\rho^2+dz^2}{\rho^2},\qquad dV_g=\frac{d\rho\,dz}{\rho^2},\qquad\lvert\nabla_g\psi\rvert_g=\rho\lvert\nabla\psi\rvert. g = ρ 2 d ρ 2 + d z 2 , d V g = ρ 2 d ρ d z , ∣ ∇ g ψ ∣ g = ρ ∣ ∇ ψ ∣ .
Integration in the azimuthal variable gives the exact identities
∫ R 3 ∣ A ( ψ ) ∣ 3 d x = 2 π ∫ H 2 ∣ ψ ∣ 3 d V g , ∫ R 3 ∣ ∇ × A ( ψ ) ∣ 3 / 2 d x = 2 π ∫ H 2 ∣ ∇ g ψ ∣ g 3 / 2 d V g . \int_{\mathbb{R}^3}\lvert A(\psi)\rvert^3\,dx=2\pi\int_{\mathbb{H}^2}\lvert\psi\rvert^3\,dV_g,\qquad\int_{\mathbb{R}^3}\lvert\nabla\times A(\psi)\rvert^{3/2}\,dx=2\pi\int_{\mathbb{H}^2}\lvert\nabla_g\psi\rvert_g^{3/2}\,dV_g. ∫ R 3 ∣ A ( ψ ) ∣ 3 d x = 2 π ∫ H 2 ∣ ψ ∣ 3 d V g , ∫ R 3 ∣ ∇ × A ( ψ ) ∣ 3/2 d x = 2 π ∫ H 2 ∣ ∇ g ψ ∣ g 3/2 d V g .
which Proposition 4.1 extends to the full energy spaces. For a general exponent the same computation produces the weights ρ 3 − 2 p \rho^{3-2p} ρ 3 − 2 p and ρ 3 − p ∗ \rho^{3-p^*} ρ 3 − p ∗ in front of the hyperbolic integrands; both are trivial exactly when p = 3 / 2 p=3/2 p = 3/2 . The toroidal critical quotient is therefore 2 π \sqrt{2\pi} 2 π times the quotient defining the subcritical problem
Λ H 2 : = inf ψ ∈ D 1 , 3 / 2 ( H 2 ) ∖ { 0 } ∫ H 2 ∣ ∇ g ψ ∣ g 3 / 2 d V g ( ∫ H 2 ∣ ψ ∣ 3 d V g ) 1 / 2 > 0. (3) \Lambda_{\mathbb{H}^2}:=\inf_{\psi\in D^{1,3/2}(\mathbb{H}^2)\setminus\{0\}}\frac{\int_{\mathbb{H}^2}\lvert\nabla_g\psi\rvert_g^{3/2}\,dV_g}{\left(\int_{\mathbb{H}^2}\lvert\psi\rvert^3\,dV_g\right)^{1/2}}>0. \tag*{(3)} Λ H 2 := ψ ∈ D 1 , 3/2 ( H 2 ) ∖ { 0 } inf ( ∫ H 2 ∣ ψ ∣ 3 d V g ) 1/2 ∫ H 2 ∣ ∇ g ψ ∣ g 3/2 d V g > 0. (3)
where D 1 , 3 / 2 ( H 2 ) D^{1,3/2}(\mathbb{H}^2) D 1 , 3/2 ( H 2 ) is the completion of C c ∞ ( H 2 ) C^\infty_c(\mathbb{H}^2) C c ∞ ( H 2 ) in the norm ∥ ∇ g ψ ∥ L 3 / 2 ( H 2 ) \lVert\nabla_g\psi\rVert_{L^{3/2}(\mathbb{H}^2)} ∥ ∇ g ψ ∥ L 3/2 ( H 2 ) ; the exponent 3 3 3 lies below the critical exponent ( 3 / 2 ) 2 ∗ = 6 (3/2)^*_2=6 ( 3/2 ) 2 ∗ = 6 . In the notation of [5 (1.2)] the constant is S 0 , 3 S_{0,3} S 0 , 3 with n = 2 n=2 n = 2 and gradient exponent 3 / 2 3/2 3/2 . We do not know a closed formula for it. Writing Δ 3 / 2 , g ψ : = div g ( ∣ ∇ g ψ ∣ g − 1 / 2 ∇ g ψ ) \Delta_{3/2,g}\psi:=\operatorname{div}_g(\lvert\nabla_g\psi\rvert_g^{-1/2}\nabla_g\psi) Δ 3/2 , g ψ := div g (∣ ∇ g ψ ∣ g − 1/2 ∇ g ψ ) , a positive minimizer for Λ H 2 \Lambda_{\mathbb{H}^2} Λ H 2 , after multiplication by a suitable positive constant, satisfies
− Δ 3 / 2 , g ψ = ψ 2 in H 2 (4) -\Delta_{3/2,g}\psi=\psi^2\qquad\text{in }\mathbb{H}^2 \tag*{(4)} − Δ 3/2 , g ψ = ψ 2 in H 2 (4)
in the weak sense. By [5 Theorem 1.1] every positive finite-energy solution is radial about some centre O ∈ H 2 O\in\mathbb{H}^2 O ∈ H 2 , and Theorem 3.2 below identifies it, via a weighted shooting classification, as the unique rapidly decaying member of a one-parameter family: all positive finite-energy solutions have the same radial profile ψ ∗ \psi_* ψ ∗ , with only the centre varying.
Main results
Hyperbolic isometries moving the centre correspond to dilations and axial translations of R 3 \mathbb{R}^{3} R 3 . For λ > 0 \lambda> 0 λ > 0 and z 0 ∈ R z_{0} \in\mathbb{R} z 0 ∈ R put
( T λ , z 0 u ) ( x ) : = λ u ( λ ( x − z 0 e 3 ) ) , (5) (T_{\lambda,z_{0}}u)(x) := \lambda u\bigl(\lambda(x-z_{0}e_{3})\bigr), \tag*{(5)} ( T λ , z 0 u ) ( x ) := λ u ( λ ( x − z 0 e 3 ) ) , (5)
the natural action on one-forms; Q Q Q , equation (2) and the two classes are invariant under these maps, and T λ , z 0 T_{\lambda,z_{0}} T λ , z 0 moves a toroidal field with profile radial about ( 1 , 0 ) (1,0) ( 1 , 0 ) to one with profile radial about ( 1 / λ , z 0 ) (1/\lambda,z_{0}) ( 1/ λ , z 0 ) . To state the iteration, let Γ ( u ) \Gamma(u) Γ ( u ) be the unique representative of the gauge class of u u u with minimal L 3 L^{3} L 3 norm, and let I ( u ) I(u) I ( u ) be the L 3 L^{3} L 3 -minimal field whose curl is ∣ Γ ( u ) ∣ Γ ( u ) |\Gamma(u)|\Gamma(u) ∣Γ ( u ) ∣Γ ( u ) ; equivalently,
I ( u ) = Γ ( ∇ × ( − Δ ) − 1 ( ∣ Γ ( u ) ∣ Γ ( u ) ) ) . I(u)=\Gamma\bigl(\nabla\times(-\Delta)^{-1}(|\Gamma(u)|\Gamma(u))\bigr). I ( u ) = Γ ( ∇ × ( − Δ ) − 1 ( ∣Γ ( u ) ∣Γ ( u )) ) .
The source is divergence-free, because the Euler–Lagrange condition for gauge minimization is div ( ∣ Γ ( u ) ∣ Γ ( u ) ) = 0 \operatorname{div}(|\Gamma(u)|\Gamma(u))=0 div ( ∣Γ ( u ) ∣Γ ( u )) = 0 . Thus D ( I ( u ) ) = Γ ( u ) \mathcal{D}(I(u))=\Gamma(u) D ( I ( u )) = Γ ( u ) : on solutions I I I agrees with D \mathcal{D} D , whereas for general fields with nonzero curl it satisfies Q ( I ( u ) ) ≤ Q ( u ) Q(I(u))\le Q(u) Q ( I ( u )) ≤ Q ( u ) (Lemma 6.1 ).
The first main result reads as follows.
Theorem 1.1. Let ψ ∗ \psi_{*} ψ ∗ be the positive minimizer for Λ H 2 \Lambda_{\mathbb{H}^{2}} Λ H 2 centred at ( 1 , 0 ) (1,0) ( 1 , 0 ) and normalized to solve (4) , and put
A ∗ : = ρ − 1 ψ ∗ e θ , U ∗ : = D ( A ∗ ) . A_{*}:=\rho^{-1}\psi_{*}e_{\theta},\qquad U_{*}:=\mathcal{D}(A_{*}). A ∗ := ρ − 1 ψ ∗ e θ , U ∗ := D ( A ∗ ) . (a) (Sharp constants.) 0 < S 3 / 2 , curl ≤ S 3 / 2 , curl T = S 3 / 2 , curl S = 2 π Λ H 2 0<S_{3/2,\operatorname{curl}}\le S^{T}_{3/2,\operatorname{curl}}=S^{S}_{3/2,\operatorname{curl}}=\sqrt{2\pi}\,\Lambda_{\mathbb{H}^{2}} 0 < S 3/2 , curl ≤ S 3/2 , curl T = S 3/2 , curl S = 2 π Λ H 2 .
(b) (Classification of the extremals.) A ∗ ∈ T A_{*}\in\mathcal{T} A ∗ ∈ T and U ∗ ∈ S U_{*}\in\mathcal{S} U ∗ ∈ S are extremals of the respective restricted constants, solve (2) , and satisfy D ( U ∗ ) = A ∗ \mathcal{D}(U_{*})=A_{*} D ( U ∗ ) = A ∗ . Every extremal of S 3 / 2 , curl T S^{T}_{3/2,\operatorname{curl}} S 3/2 , curl T is of the form c T λ , z 0 A ∗ cT_{\lambda,z_{0}}A_{*} c T λ , z 0 A ∗ , and every extremal of S 3 / 2 , curl S S^{S}_{3/2,\operatorname{curl}} S 3/2 , curl S is of the form c T λ , z 0 U ∗ + w cT_{\lambda,z_{0}}U_{*}+w c T λ , z 0 U ∗ + w , with c ∈ R ∖ { 0 } c\in\mathbb{R}\setminus\{0\} c ∈ R ∖ { 0 } , λ > 0 \lambda>0 λ > 0 , z 0 ∈ R z_{0}\in\mathbb{R} z 0 ∈ R and w ∈ S w\in\mathcal{S} w ∈ S with ∇ × w = 0 \nabla\times w=0 ∇ × w = 0 .
(c) (Convergence to the two-cycle.) Let ψ ∈ D 1 , 3 / 2 ( H 2 ) \psi\in D^{1,3/2}(\mathbb{H}^{2}) ψ ∈ D 1 , 3/2 ( H 2 ) be nonzero, of one sign and radial about a centre O ′ = ( 1 / λ , z 0 ) O'=(1/\lambda,z_{0}) O ′ = ( 1/ λ , z 0 ) , and let u 0 = A ( ψ ) u_{0}=A(\psi) u 0 = A ( ψ ) , u n + 1 = I ( u n ) u_{n+1}=I(u_{n}) u n + 1 = I ( u n ) , v n = u n / ∥ u n ∥ L 3 v_{n}=u_{n}/\lVert u_{n}\rVert_{L^{3}} v n = u n / ∥ u n ∥ L 3 . Then Q ( u n ) Q(u_{n}) Q ( u n ) is nonincreasing, Q ( u n ) → S 3 / 2 , curl T Q(u_{n})\to S^{T}_{3/2,\operatorname{curl}} Q ( u n ) → S 3/2 , curl T , and
v 2 k ⟶ σ T λ , z 0 A ∗ ∥ A ∗ ∥ L 3 , v 2 k + 1 ⟶ σ T λ , z 0 U ∗ ∥ U ∗ ∥ L 3 in L 3 ( R 3 ; R 3 ) , v_{2k}\longrightarrow\sigma\frac{T_{\lambda,z_{0}}A_{*}}{\lVert A_{*}\rVert_{L^{3}}},\qquad
v_{2k+1}\longrightarrow\sigma\frac{T_{\lambda,z_{0}}U_{*}}{\lVert U_{*}\rVert_{L^{3}}}
\quad\text{in }L^{3}(\mathbb{R}^{3};\mathbb{R}^{3}), v 2 k ⟶ σ ∥ A ∗ ∥ L 3 T λ , z 0 A ∗ , v 2 k + 1 ⟶ σ ∥ U ∗ ∥ L 3 T λ , z 0 U ∗ in L 3 ( R 3 ; R 3 ) , ∇ × v 2 k ⟶ σ ∇ × ( T λ , z 0 A ∗ ) ∥ A ∗ ∥ L 3 , ∇ × v 2 k + 1 ⟶ σ ∇ × ( T λ , z 0 U ∗ ) ∥ U ∗ ∥ L 3 in L 3 / 2 ( R 3 ; R 3 ) \nabla\times v_{2k}\longrightarrow\sigma\frac{\nabla\times(T_{\lambda,z_{0}}A_{*})}{\lVert A_{*}\rVert_{L^{3}}},\qquad
\nabla\times v_{2k+1}\longrightarrow\sigma\frac{\nabla\times(T_{\lambda,z_{0}}U_{*})}{\lVert U_{*}\rVert_{L^{3}}}
\quad\text{in }L^{3/2}(\mathbb{R}^{3};\mathbb{R}^{3}) ∇ × v 2 k ⟶ σ ∥ A ∗ ∥ L 3 ∇ × ( T λ , z 0 A ∗ ) , ∇ × v 2 k + 1 ⟶ σ ∥ U ∗ ∥ L 3 ∇ × ( T λ , z 0 U ∗ ) in L 3/2 ( R 3 ; R 3 ) as k → ∞ k\to\infty k → ∞ , where σ = 1 \sigma=1 σ = 1 if ψ ≥ 0 \psi\ge0 ψ ≥ 0 and σ = − 1 \sigma=-1 σ = − 1 if ψ ≤ 0 \psi\le0 ψ ≤ 0 .
The iteration is computable: for a nonnegative radial profile, one full cycle T → S → T \mathcal{T}\to\mathcal{S}\to\mathcal{T} T → S → T amounts to two one-dimensional quadratures, I 2 ( A ( ψ ) ) = A ( R ψ ) I^{2}(A(\psi))=A(R\psi) I 2 ( A ( ψ )) = A ( R ψ ) with R R R given in (59) , and each iterate is an admissible competitor, so that every Q ( u n ) Q(u_{n}) Q ( u n ) is a rigorous upper bound for the restricted constant. The first step from A T A_{\mathcal{T}} A T can be evaluated in closed form.
Corollary 1.2.
S 3 / 2 , curl T < Q ( I ( A T ) ) = 96 15 π 5 4455 π − 13568 < 8 10 7 π = Q ( A T ) . S^{T}_{3/2,\operatorname{curl}}<Q(I(A_{\mathcal{T}}))=\frac{96\sqrt{15}\,\pi}{5\sqrt{4455\pi-13568}}<\frac{8\sqrt{10}}{7}\pi=Q(A_{\mathcal{T}}). S 3/2 , curl T < Q ( I ( A T )) = 5 4455 π − 13568 96 15 π < 7 8 10 π = Q ( A T ) .
Numerically, S 3 / 2 , curl T ≈ 11.2615 S^{T}_{3/2,\operatorname{curl}}\approx11.2615 S 3/2 , curl T ≈ 11.2615 (§ 8 ), while the two bounds in Corollary 1.2 are 11.2948 11.2948 11.2948 and 11.3538 11.3538 11.3538 . The inverse I I I is iterated rather than D \mathcal{D} D , although the two maps agree on solutions: by Proposition 7.1, the forward orbit D j ( A T ) \mathcal{D}^{j}(A_{\mathcal{T}}) D j ( A T ) has increasing quotients and leaves the energy space at j = 3 j=3 j = 3 , whereas I I I lowers the quotient at every step.
The proof of Theorem 1.1 is given in § 7 . Part (a) combines the hyperbolic representation with the duality; part (b) rests on the uniqueness of the positive hyperbolic profile (Theorem 3.2 ), which we derive from the symmetry theorem of Dutta and Sandeep [5 ] and the weighted shooting classification of Kawano, Yanagida and Yotsutani [17 ] ; part (c) uses the exact descent formula for I I I , compactness at a fixed centre, and again the uniqueness. The profile ψ ∗ \psi_* ψ ∗ is smooth away from its centre but only C l o c 2 , 1 C^{2,1}_{\mathrm{loc}} C loc 2 , 1 there (Remark 3.4), so the extremals are not smooth across the circle { ρ = 1 , z = 0 } \{\rho= 1,\, z = 0\} { ρ = 1 , z = 0 } .
The reduction also determines the geometry of the poloidal extremal. With s = 1 + ∣ x ∣ 2 s = 1+|x|^2 s = 1 + ∣ x ∣ 2 and η = 2 ρ / s \eta= 2\rho/s η = 2 ρ / s , Theorem 5.1 gives
U ∗ ( x ) = s − 2 G ∗ ( η ) N ( x ) , N ( x ) = ( 2 x 1 z , 2 x 2 z , 1 + z 2 − ρ 2 ) , G ∗ ( η ) = 2 J ψ ∗ ( t ) 1 − η 2 , U_*(x) = s^{-2}G_*(\eta)N(x),\qquad N(x) = (2x_1z,\,2x_2z,\,1+z^2-\rho^2),\qquad G_*(\eta) = \frac{2\mathcal{J}_{\psi_*}(t)}{1-\eta^2}, U ∗ ( x ) = s − 2 G ∗ ( η ) N ( x ) , N ( x ) = ( 2 x 1 z , 2 x 2 z , 1 + z 2 − ρ 2 ) , G ∗ ( η ) = 1 − η 2 2 J ψ ∗ ( t ) ,
where t = arcosh ( 1 / η ) t = \operatorname{arcosh}(1/\eta) t = arcosh ( 1/ η ) and J ψ ∗ ( t ) = ∫ 0 t ψ ∗ ( s ) 2 sinh s d s \mathcal{J}_{\psi_*}(t) = \int_0^t \psi_*(s)^2 \sinh s\,\mathrm{d}s J ψ ∗ ( t ) = ∫ 0 t ψ ∗ ( s ) 2 sinh s d s is the radial flux of ψ ∗ \psi_* ψ ∗ . The formula is semi-explicit : N N N and η \eta η are explicit, and the only remaining factor is computed from the one-dimensional profile; G ∗ G_* G ∗ extends continuously to [ 0 , 1 ] [0,1] [ 0 , 1 ] with G ∗ ( 0 ) = 2 J ψ ∗ ( ∞ ) G_*(0) = 2\mathcal{J}_{\psi_*}(\infty) G ∗ ( 0 ) = 2 J ψ ∗ ( ∞ ) and G ∗ ( 1 ) = ψ ∗ ( 0 ) 2 G_*(1) = \psi_*(0)^2 G ∗ ( 1 ) = ψ ∗ ( 0 ) 2 .
Dirac zero modes
The constant S 3 / 2 , c u r l S_{3/2,\mathrm{curl}} S 3/2 , curl is also relevant to magnetic Weyl–Dirac operators D A = σ ⋅ ( − i ∇ − A ) D_A = \boldsymbol{\sigma}\cdot(-i\nabla-A) D A = σ ⋅ ( − i ∇ − A ) , where σ \boldsymbol{\sigma} σ denotes the Pauli matrices. A zero mode is a nonzero spinor Ψ ∈ L 2 ( R 3 ; C 2 ) \Psi\in L^2(\mathbb{R}^3;\mathbb{C}^2) Ψ ∈ L 2 ( R 3 ; C 2 ) satisfying D A Ψ = 0 D_A\Psi= 0 D A Ψ = 0 . The classical Loss–Yau potential A L Y A_{\mathrm{LY}} A LY admits an explicit L 2 L^2 L 2 zero mode, so ker L 2 D A L Y ≠ { 0 } \ker_{L^2}D_{A_{\mathrm{LY}}} \ne\{0\} ker L 2 D A LY = { 0 } [18 Section II] . In contrast, we prove that our semi-explicit extremals have trivial Dirac kernels at every coupling: ker L 2 D c A ∗ = ker L 2 D c U ∗ = { 0 } \ker_{L^2}D_{cA_*} = \ker_{L^2}D_{cU_*} = \{0\} ker L 2 D c A ∗ = ker L 2 D c U ∗ = { 0 } for all c ∈ R c \in\mathbb{R} c ∈ R ; see Theorem 1.3 . Frank and Loss [11 Theorem 1 and Remark 2] showed that a zero mode requires
inf ∇ χ ∈ L 3 ( R 3 ; R 3 ) ∥ A + ∇ χ ∥ L 3 2 ≥ 3 S S o b , 2 , S S o b , 2 = 3 ( π / 2 ) 4 / 3 (6) \inf_{\nabla\chi\in L^3(\mathbb{R}^3;\mathbb{R}^3)}\|A+\nabla\chi\|_{L^3}^2 \ge3S_{\mathrm{Sob},2},\qquad S_{\mathrm{Sob},2}=3(\pi/2)^{4/3} \tag*{(6)} ∇ χ ∈ L 3 ( R 3 ; R 3 ) inf ∥ A + ∇ χ ∥ L 3 2 ≥ 3 S Sob , 2 , S Sob , 2 = 3 ( π /2 ) 4/3 (6)
a sharp condition on the potential modulo gauge. Combined with the Sobolev–curl inequality ∥ ∇ × A ∥ L 3 / 2 ≥ S 3 / 2 , c u r l 2 / 3 inf ∇ χ ∥ A + ∇ χ ∥ L 3 \|\nabla\times A\|_{L^{3/2}}\ge S_{3/2,\mathrm{curl}}^{2/3}\inf_{\nabla\chi}\|A+\nabla\chi\|_{L^3} ∥∇ × A ∥ L 3/2 ≥ S 3/2 , curl 2/3 inf ∇ χ ∥ A + ∇ χ ∥ L 3 , it gives ∥ ∇ × A ∥ L 3 / 2 ≥ S 3 / 2 , c u r l 2 / 3 3 S S o b , 2 \|\nabla\times A\|_{L^{3/2}}\ge S_{3/2,\mathrm{curl}}^{2/3}\sqrt{3S_{\mathrm{Sob},2}} ∥∇ × A ∥ L 3/2 ≥ S 3/2 , curl 2/3 3 S Sob , 2 for a zero mode, which is strictly below the sharp field-norm threshold ∥ ∇ × A ∥ L 3 / 2 ≥ 4 S S o b , 2 \|\nabla\times A\|_{L^{3/2}}\ge4S_{\mathrm{Sob},2} ∥∇ × A ∥ L 3/2 ≥ 4 S Sob , 2 of Wang and Zhang [29 Theorem 1.6] , stated on S 3 S^3 S 3 and transferred to R 3 \mathbb{R}^3 R 3 by conformal invariance, because S 3 / 2 , c u r l < 4 π S_{3/2,\mathrm{curl}}<4\pi S 3/2 , curl < 4 π ; see also [31 ] and, for the related spinorial Sobolev inequalities on S n S^n S n and their stability, [28 ] . Their bound doubles the earlier field bound 2 S S o b , 2 2S_{\mathrm{Sob},2} 2 S Sob , 2 of [10 Theorem 1.1] and complements, rather than replaces, the gauge-invariant condition ((6) ) on the potential. Such norm thresholds say nothing at large coupling, where the following geometric exclusion takes over.
Theorem 1.3. (a) Let B = b ρ ( ρ , z ) e ρ + b z ( ρ , z ) e z ∈ L ∞ ( R 3 ; R 3 ) B=b_\rho(\rho,z)e_\rho+b_z(\rho,z)e_z\in L^\infty(\mathbb{R}^3;\mathbb{R}^3) B = b ρ ( ρ , z ) e ρ + b z ( ρ , z ) e z ∈ L ∞ ( R 3 ; R 3 ) be a bounded axisymmetric poloidal potential. Then ker L 2 D c B = { 0 } \ker_{L^2}D_{cB}=\{0\} ker L 2 D c B = { 0 } for every c ∈ R c\in\mathbb{R} c ∈ R .
(b) Let A ∗ A_* A ∗ and U ∗ U_* U ∗ be the extremals in Theorem 1.1 . For every c ∈ R c\in\mathbb{R} c ∈ R , λ > 0 \lambda>0 λ > 0 and z 0 ∈ R z_0\in\mathbb{R} z 0 ∈ R ,
ker L 2 D c T λ , z 0 A ∗ = ker L 2 D c T λ , z 0 U ∗ = { 0 } . \ker_{L^2}D_{cT_{\lambda,z_0}A_*}=\ker_{L^2}D_{cT_{\lambda,z_0}U_*}=\{0\}. ker L 2 D c T λ , z 0 A ∗ = ker L 2 D c T λ , z 0 U ∗ = { 0 } . The operators have domain H 1 ( R 3 ; C 2 ) H^1(\mathbb{R}^3;\mathbb{C}^2) H 1 ( R 3 ; C 2 ) in L 2 ( R 3 ; C 2 ) L^2(\mathbb{R}^3;\mathbb{C}^2) L 2 ( R 3 ; C 2 ) .
Both extremals are in the minimizing gauge, so the Nehari identity gives inf χ ∥ c B + ∇ χ ∥ L 3 2 = c 2 ( S 3 / 2 , c u r l ) 4 / 3 \inf_\chi\|cB+\nabla\chi\|_{L^3}^2=c^2(S_{3/2,\mathrm{curl}})^{4/3} inf χ ∥ c B + ∇ χ ∥ L 3 2 = c 2 ( S 3/2 , curl ) 4/3 for B ∈ { A ∗ , U ∗ } B\in\{A_*,U_*\} B ∈ { A ∗ , U ∗ } , and ((6) ) excludes zero modes only for ∣ c ∣ < 3 S S o b , 2 / ( S 3 / 2 , c u r l ) 2 / 3 ≈ 0.81 |c|<\sqrt{3S_{\mathrm{Sob},2}}/(S_{3/2,\mathrm{curl}})^{2/3}\approx0.81 ∣ c ∣ < 3 S Sob , 2 / ( S 3/2 , curl ) 2/3 ≈ 0.81 . Theorem 1.3 excludes them at every coupling: an arbitrarily large gauge-minimal norm does not ensure a zero mode. Elton [6 Theorem 1.1] proved a geometric exclusion of this kind for smooth potentials whose magnetic field is parallel to a conformal Killing field X X X with X ⋅ ∇ × X = 0 X\cdot\nabla\times X=0 X ⋅ ∇ × X = 0 ; part (a) is a version of his angular obstruction for bounded measurable poloidal potentials, which is what the non-smooth extremals require, and part (b) reduces A ∗ A_{*} A ∗ to part (a) by an explicit Möbius map (§ 5.2 ).
Fröhlich, Lieb and Loss [12 ] showed that zero modes with finite magnetic-field energy lead to instability of the one-electron atom in a self-generated magnetic field at sufficiently large nuclear charge. The absence of zero modes established in Theorem 1.3 may therefore provide a starting point for investigating stability within the corresponding classes of magnetic potentials.
Notation and organization
Unless another centre is specified, t t t denotes the hyperbolic distance from ( ρ , z ) = ( 1 , 0 ) (\rho,z)=(1,0) ( ρ , z ) = ( 1 , 0 ) , so that η = sech t \eta=\operatorname{sech}t η = sech t . The operators ∇ × \nabla\times ∇ × and div \operatorname{div} div act in the sense of distributions. For positive functions f , h f,h f , h , we write f ≍ h f\asymp h f ≍ h when f / h f/h f / h is bounded above and below by positive constants in the specified limit, and f ∼ h f\sim h f ∼ h when f / h → 1 f/h\to1 f / h → 1 .
Section 2 proves the general duality. Section 3 develops the hyperbolic problem and proves Theorem 3.2 . Section 4 establishes the exact toroidal reduction. Section 5 derives the semi-explicit poloidal extremal and proves Theorem 1.3 . Section 6 defines the gauge-canonical inverse and proves its descent property. Section 7 contains the proof of Theorem 1.1 and Corollary 1.2 , and the breakdown of the forward iteration. Numerical values and open problems conclude the paper.
Variational framework
In this section we recall the setting of [22 ] and fix the conventions used later. For r > 1 r>1 r > 1 , write r ′ = r / ( r − 1 ) r'=r/(r-1) r ′ = r / ( r − 1 ) for its conjugate exponent. Let 1 < p < 3 1<p<3 1 < p < 3 and
p ∗ : = 3 p 3 − p . p^{*}:=\frac{3p}{3-p}. p ∗ := 3 − p 3 p .
The critical p p p -curl–curl equation is
∇ × ( ∣ ∇ × u ∣ p − 2 ∇ × u ) = ∣ u ∣ p ∗ − 2 u in R 3 . (7) \nabla\times\left(\lvert\nabla\times u\rvert^{p-2}\nabla\times u\right)=\lvert u\rvert^{p^{*}-2}u\quad\text{in }\mathbb{R}^{3}. \tag*{(7)} ∇ × ( ∣ ∇ × u ∣ p − 2 ∇ × u ) = ∣ u ∣ p ∗ − 2 u in R 3 . (7)
The natural space is
D 1 , p ( curl ; R 3 ) : = { u ∈ L p ∗ ( R 3 , R 3 ) : ∇ × u ∈ L p ( R 3 , R 3 ) } , D^{1,p}(\operatorname{curl};\mathbb{R}^{3}):=\left\{u\in L^{p^{*}}(\mathbb{R}^{3},\mathbb{R}^{3}):\nabla\times u\in L^{p}(\mathbb{R}^{3},\mathbb{R}^{3})\right\}, D 1 , p ( curl ; R 3 ) := { u ∈ L p ∗ ( R 3 , R 3 ) : ∇ × u ∈ L p ( R 3 , R 3 ) } ,
and the kernel of curl is denoted by
W : = { w ∈ D 1 , p ( curl ; R 3 ) : ∇ × w = 0 } . \mathcal{W}:=\left\{w\in D^{1,p}(\operatorname{curl};\mathbb{R}^{3}):\nabla\times w=0\right\}. W := { w ∈ D 1 , p ( curl ; R 3 ) : ∇ × w = 0 } .
The gauge-corrected Sobolev–curl constant is
S p , curl : = inf ∇ × u ≠ 0 ∫ R 3 ∣ ∇ × u ∣ p d x ( inf w ∈ W ∫ R 3 ∣ u + w ∣ p ∗ d x ) p / p ∗ . (8) S_{p,\operatorname{curl}}:=\inf_{\nabla\times u\ne0}\frac{\int_{\mathbb{R}^{3}}\lvert\nabla\times u\rvert^{p}\,dx}{\left(\inf_{w\in\mathcal{W}}\int_{\mathbb{R}^{3}}\lvert u+w\rvert^{p^{*}}\,dx\right)^{p/p^{*}}}. \tag*{(8)} S p , curl := ∇ × u = 0 inf ( inf w ∈ W ∫ R 3 ∣ u + w ∣ p ∗ d x ) p / p ∗ ∫ R 3 ∣ ∇ × u ∣ p d x . (8)
Without this minimization, adding arbitrarily large curl-free fields to a fixed field with nonzero curl would force the quotient to zero.
For every u ∈ D 1 , p ( curl ; R 3 ) u\in D^{1,p}(\operatorname{curl};\mathbb{R}^{3}) u ∈ D 1 , p ( curl ; R 3 ) there is a unique w ( u ) ∈ W w(u)\in\mathcal{W} w ( u ) ∈ W satisfying
∫ R 3 ∣ u + w ( u ) ∣ p ∗ d x = inf w ∈ W ∫ R 3 ∣ u + w ∣ p ∗ d x . (9) \int_{\mathbb{R}^{3}}\lvert u+w(u)\rvert^{p^{*}}\,dx=\inf_{w\in\mathcal{W}}\int_{\mathbb{R}^{3}}\lvert u+w\rvert^{p^{*}}\,dx. \tag*{(9)} ∫ R 3 ∣ u + w ( u ) ∣ p ∗ d x = w ∈ W inf ∫ R 3 ∣ u + w ∣ p ∗ d x . (9)
By strict convexity, this is equivalent to
∫ R 3 ⟨ ∣ u + w ( u ) ∣ p ∗ − 2 ( u + w ( u ) ) , ζ ⟩ d x = 0 for every ζ ∈ W , (10) \int_{\mathbb{R}^{3}} \left\langle|u+w(u)|^{p^{*}-2}(u+w(u)),\zeta\right\rangle\, \mathrm{d}x=0 \qquad\text{for every }\zeta\in W, \tag*{(10)} ∫ R 3 ⟨ ∣ u + w ( u ) ∣ p ∗ − 2 ( u + w ( u )) , ζ ⟩ d x = 0 for every ζ ∈ W , (10)
and hence to the nonlinear Coulomb-type gauge condition
div ( ∣ u + w ( u ) ∣ p ∗ − 2 ( u + w ( u ) ) ) = 0 in D ′ ( R 3 ) . (11) \operatorname{div}\left(|u+w(u)|^{p^{*}-2}(u+w(u))\right)=0 \quad\text{in }\mathcal{D}'(\mathbb{R}^{3}). \tag*{(11)} div ( ∣ u + w ( u ) ∣ p ∗ − 2 ( u + w ( u )) ) = 0 in D ′ ( R 3 ) . (11)
Conversely, (11) implies (10) : for 1 < r < ∞ 1<r<\infty 1 < r < ∞ , every curl-free Z ∈ L r ( R 3 , R 3 ) Z\in L^{r}(\mathbb{R}^{3},\mathbb{R}^{3}) Z ∈ L r ( R 3 , R 3 ) is the L r L^{r} L r limit of gradients ∇ φ n \nabla\varphi_n ∇ φ n with φ n ∈ C ∞ ( R 3 ) \varphi_n\in C^{\infty}(\mathbb{R}^{3}) φ n ∈ C ∞ ( R 3 ) [21 Section 2.1] , so that ∫ R 3 ⟨ B , Z ⟩ d x = 0 \int_{\mathbb{R}^{3}}\langle B,Z\rangle\,\mathrm{d}x=0 ∫ R 3 ⟨ B , Z ⟩ d x = 0 for every divergence-free B ∈ L r ′ ( R 3 , R 3 ) B\in L^{r'}(\mathbb{R}^{3},\mathbb{R}^{3}) B ∈ L r ′ ( R 3 , R 3 ) . We refer to this fact as the annihilator identity; it will be used again in the duality arguments below. Condition (11) defines the set of gauge-minimal fields used in [22 ] ,
M : = { u ∈ D 1 , p ( curl ; R 3 ) : div ( ∣ u ∣ p ∗ − 2 u ) = 0 } , \mathcal{M}:=\left\{u\in\mathcal{D}^{1,p}(\operatorname{curl};\mathbb{R}^{3}):\operatorname{div}(|u|^{p^{*}-2}u)=0\right\}, M := { u ∈ D 1 , p ( curl ; R 3 ) : div ( ∣ u ∣ p ∗ − 2 u ) = 0 } ,
which contains exactly one representative of every gauge class.
The functional associated with (7) is
J p ( u ) : = 1 p ∫ R 3 ∣ ∇ × u ∣ p d x − 1 p ∗ ∫ R 3 ∣ u ∣ p ∗ d x . J_p(u):=\frac{1}{p}\int_{\mathbb{R}^{3}}|\nabla\times u|^{p}\,\mathrm{d}x-\frac{1}{p^{*}}\int_{\mathbb{R}^{3}}|u|^{p^{*}}\,\mathrm{d}x. J p ( u ) := p 1 ∫ R 3 ∣∇ × u ∣ p d x − p ∗ 1 ∫ R 3 ∣ u ∣ p ∗ d x .
By the main theorem of [22 ] , S p , c u r l S_{p,\mathrm{curl}} S p , curl is attained and (7) has a ground state for every 1 < p < 3 1<p<3 1 < p < 3 ; the same holds under the cylindrical symmetry restrictions recalled below.
The cylindrical symmetry classes
Write
x = ( ρ cos θ , ρ sin θ , z ) , ρ = ( x 1 2 + x 2 2 ) 1 / 2 , z = x 3 , x=(\rho\cos\theta,\rho\sin\theta,z),\qquad\rho=(x_1^2+x_2^2)^{1/2},\qquad z=x_3, x = ( ρ cos θ , ρ sin θ , z ) , ρ = ( x 1 2 + x 2 2 ) 1/2 , z = x 3 ,
and let e ρ , e θ , e 3 e_\rho,e_\theta,e_3 e ρ , e θ , e 3 be the cylindrical orthonormal frame. An S O ( 2 ) SO(2) S O ( 2 ) -equivariant field can be decomposed into radial, azimuthal and axial parts.
The two fixed-point classes introduced in [22 ] , viewed here as subspaces of D 1 , p ( curl ; R 3 ) \mathcal{D}^{1,p}(\operatorname{curl};\mathbb{R}^{3}) D 1 , p ( curl ; R 3 ) , are:
T : A ( x ) = a ( ρ , z ) e θ = a ( ρ , z ) ρ ( − x 2 , x 1 , 0 ) , S : U ( x ) = b ( ρ , z ) e ρ + c ( ρ , z ) e 3 = α ( ρ , z ) ( x 1 , x 2 , 0 ) + β ( ρ , z ) ( 0 , 0 , 1 ) . (12) \begin{aligned}
\mathcal{T}:&\quad A(x)=a(\rho,z)e_\theta=\frac{a(\rho,z)}{\rho}(-x_2,x_1,0),\\
\mathcal{S}:&\quad U(x)=b(\rho,z)e_\rho+c(\rho,z)e_3\\
&\quad=\alpha(\rho,z)(x_1,x_2,0)+\beta(\rho,z)(0,0,1). \tag*{(12)}
\end{aligned} T : S : A ( x ) = a ( ρ , z ) e θ = ρ a ( ρ , z ) ( − x 2 , x 1 , 0 ) , U ( x ) = b ( ρ , z ) e ρ + c ( ρ , z ) e 3 = α ( ρ , z ) ( x 1 , x 2 , 0 ) + β ( ρ , z ) ( 0 , 0 , 1 ) . (12)
Let S p , c u r l T S^{\mathcal{T}}_{p,\mathrm{curl}} S p , curl T and S p , c u r l S S^{\mathcal{S}}_{p,\mathrm{curl}} S p , curl S denote the constants obtained by restricting (8) to these classes. By [22 Theorem 1.4] both restricted constants are attained. After multiplication by a suitable positive constant, the gauge-minimal representatives of the minimizers solve the full equation (7) , not only its restriction to the symmetry subspace.
Since both classes are contained in the full energy space,
0 < S p , c u r l ≤ min { S p , c u r l T , S p , c u r l S } , 1 < p < 3. (13) 0<S_{p,\mathrm{curl}}\leq\min\left\{S^{\mathcal{T}}_{p,\mathrm{curl}},S^{\mathcal{S}}_{p,\mathrm{curl}}\right\},\qquad1<p<3. \tag*{(13)} 0 < S p , curl ≤ min { S p , curl T , S p , curl S } , 1 < p < 3. (13)
Whether the unrestricted constant equals either restricted constant remains open.
Dual exponents and nonlinear duality
The pair of dual equations written by Frank and Loss at p = 3 / 2 p=3/2 p = 3/2 [10 Section 9, pp. 24–25] is the self-dual case of a correspondence between different critical exponents. Inverting a nonlinear flux is also the mechanism of nonlinear Hodge duality [20 , 25 ] . Here we identify the paired critical exponents and the interchange of the cylindrical classes. For 1 < p < 3 1<p<3 1 < p < 3 , put
p ^ : = ( p ∗ ) ′ = 3 p 4 p − 3 . (14) \widehat{p}:=(p^*)'=\frac{3p}{4p-3}. \tag*{(14)} p := ( p ∗ ) ′ = 4 p − 3 3 p . (14)
Then 1 < p ^ < 3 1<\widehat{p}<3 1 < p < 3 and
p ^ ∗ = p ′ , p ^ = p , ( p ^ − 1 ) ( p ∗ − 1 ) = 1. \widehat{p}^*=p',\qquad\widehat{p}=p,\qquad(\widehat{p}-1)(p^*-1)=1. p ∗ = p ′ , p = p , ( p − 1 ) ( p ∗ − 1 ) = 1.
In particular, p ^ = p \widehat{p}=p p = p if and only if p = 3 / 2 p=3/2 p = 3/2 . We denote the quotient in (8) by
Q p ( u ) : = ∥ ∇ × u ∥ p p ∥ u + w ( u ) ∥ p ∗ p , ∇ × u ≠ 0 , Q_p(u):=\frac{\lVert\nabla\times u\rVert_p^p}{\lVert u+w(u)\rVert_{p^*}^p},\qquad\nabla\times u\ne0, Q p ( u ) := ∥ u + w ( u ) ∥ p ∗ p ∥ ∇ × u ∥ p p , ∇ × u = 0 ,
where the gauge space and the minimizing correction w ( u ) w(u) w ( u ) are understood at the exponent of the corresponding problem. Thus Q 3 / 2 = Q Q_{3/2}=Q Q 3/2 = Q . Define the nonlinear map
D p ( u ) : = ∣ ∇ × u ∣ p − 2 ∇ × u , \mathcal{D}_p(u):=\lvert\nabla\times u\rvert^{p-2}\nabla\times u, D p ( u ) := ∣ ∇ × u ∣ p − 2 ∇ × u ,
where the expression is defined to be zero when ∇ × u = 0 \nabla\times u=0 ∇ × u = 0 . For an arbitrary finite-energy field, D p ( u ) \mathcal{D}_p(u) D p ( u ) is merely an element of L p ′ L^{p'} L p ′ ; that it lies in the dual energy space is a consequence of the equation.
Theorem 2.1 (Duality of finite-energy solutions). Let 1 < p < 3 1<p<3 1 < p < 3 , let p ^ \widehat{p} p be given by (14) , and let u ∈ D 1 , p ( curl ; R 3 ) u\in D^{1,p}(\operatorname{curl};\mathbb{R}^3) u ∈ D 1 , p ( curl ; R 3 ) be a nonzero weak solution of (7) . Then v : = D p ( u ) ∈ D 1 , p ^ ( curl ; R 3 ) v:=\mathcal{D}_p(u)\in D^{1,\widehat{p}}(\operatorname{curl};\mathbb{R}^3) v := D p ( u ) ∈ D 1 , p ( curl ; R 3 ) solves
∇ × ( ∣ ∇ × v ∣ p ^ − 2 ∇ × v ) = ∣ v ∣ p ∗ − 2 v in R 3 , (15) \nabla\times\left(\lvert\nabla\times v\rvert^{\widehat{p}-2}\nabla\times v\right)=\lvert v\rvert^{p^*-2}v\qquad\text{in }\mathbb{R}^3, \tag*{(15)} ∇ × ( ∣ ∇ × v ∣ p − 2 ∇ × v ) = ∣ v ∣ p ∗ − 2 v in R 3 , (15) and D p ^ ( v ) = u \mathcal{D}_{\widehat{p}}(v)=u D p ( v ) = u almost everywhere. Thus D p \mathcal{D}_p D p is a bijection between the nonzero finite-energy solution sets at p p p and p ^ \widehat{p} p , with inverse D p ^ \mathcal{D}_{\widehat{p}} D p . It interchanges the cylindrical classes:
u ∈ T ⟹ v ∈ S , u ∈ S ⟹ v ∈ T , u\in\mathcal{T}\Longrightarrow v\in\mathcal{S},\qquad u\in\mathcal{S}\Longrightarrow v\in\mathcal{T}, u ∈ T ⟹ v ∈ S , u ∈ S ⟹ v ∈ T , where each class is taken in the energy space of its respective exponent. The two solutions satisfy
∫ R 3 ∣ ∇ × u ∣ p d x = ∫ R 3 ∣ u ∣ p ∗ d x = ∫ R 3 ∣ v ∣ p ^ ∗ d x = ∫ R 3 ∣ ∇ × v ∣ p ^ d x = : m , (16) \int_{\mathbb{R}^3}\lvert\nabla\times u\rvert^p\,dx=\int_{\mathbb{R}^3}\lvert u\rvert^{p^*}\,dx=\int_{\mathbb{R}^3}\lvert v\rvert^{\widehat{p}^*}\,dx=\int_{\mathbb{R}^3}\lvert\nabla\times v\rvert^{\widehat{p}}\,dx=:m, \tag*{(16)} ∫ R 3 ∣ ∇ × u ∣ p d x = ∫ R 3 ∣ u ∣ p ∗ d x = ∫ R 3 ∣ v ∣ p ∗ d x = ∫ R 3 ∣ ∇ × v ∣ p d x =: m , (16) and consequently
J p ( u ) = J p ^ ( v ) = m 3 , Q p ( u ) 1 / p = Q p ^ ( v ) 1 / p ^ = m 1 / 3 . (17) J_p(u)=J_{\widehat{p}}(v)=\frac{m}{3},\qquad Q_p(u)^{1/p}=Q_{\widehat{p}}(v)^{1/\widehat{p}}=m^{1/3}. \tag*{(17)} J p ( u ) = J p ( v ) = 3 m , Q p ( u ) 1/ p = Q p ( v ) 1/ p = m 1/3 . (17) In particular, the correspondence preserves ground states, both in the full space and between the paired cylindrical classes.
Proof. By definition, v ∈ L p ′ = L p ^ ∗ v\in L^{p'}=L^{\widehat{p}^*} v ∈ L p ′ = L p ∗ , and the equation for u u u gives
∇ × v = ∣ u ∣ p ∗ − 2 u ∈ L p ^ , \nabla\times v=\lvert u\rvert^{p^*-2}u\in L^{\widehat{p}}, ∇ × v = ∣ u ∣ p ∗ − 2 u ∈ L p ,
because ( p ∗ − 1 ) p ^ = p ∗ (p^*-1)\widehat{p}=p^* ( p ∗ − 1 ) p = p ∗ . The identity ( p ^ − 1 ) ( p ∗ − 1 ) = 1 (\widehat{p}-1)(p^*-1)=1 ( p − 1 ) ( p ∗ − 1 ) = 1 now yields
∣ ∇ × v ∣ p ^ − 2 ∇ × v = u . \lvert\nabla\times v\rvert^{\widehat{p}-2}\nabla\times v=u. ∣ ∇ × v ∣ p − 2 ∇ × v = u .
On the other hand, ∣ v ∣ p ^ ∗ − 2 v = ∇ × u \lvert v\rvert^{\widehat{p}^*-2}v=\nabla\times u ∣ v ∣ p ∗ − 2 v = ∇ × u . Taking the distributional curl of the preceding identity proves (15) . The same argument with p p p and p ^ \widehat{p} p interchanged gives the inverse correspondence. The cylindrical curl formulas interchange toroidal and poloidal fields, while multiplication by a scalar function of their length preserves each type. These statements remain valid distributionally, by rotational averaging and smooth approximation in the curl graph norm.
Testing the equation for u u u with u u u gives the first equality in (16) . This testing is legitimate by density of C c ∞ ( R 3 , R 3 ) C^\infty_c(\mathbb{R}^3,\mathbb{R}^3) C c ∞ ( R 3 , R 3 ) in the curl graph space: in a cutoff at radius R R R , the extra curl term is bounded in L p L^p L p by C ∥ u ∥ L p ∗ ( B 2 R 3 ∖ B R 3 ) → 0 C\lVert u\rVert_{L^{p^*}(B_{2R}^3\setminus B_R^3)}\to0 C ∥ u ∥ L p ∗ ( B 2 R 3 ∖ B R 3 ) → 0 as R → ∞ R\to\infty R → ∞ ; mollification then gives smooth test fields. The other equalities follow directly from the two pointwise power identities above. Taking divergence of both equations shows that u u u and v v v satisfy their respective nonlinear gauge conditions. By the annihilator identity, both are therefore the norm-minimizing representatives of their gauge classes. Since 1 / p − 1 / p ∗ = 1 / p ^ − 1 / p ^ ∗ = 1 / 3 1/p-1/p^*=1/\widehat p-1/\widehat p^*=1/3 1/ p − 1/ p ∗ = 1/ p − 1/ p ∗ = 1/3 , formula (17) follows. Finally, the bijection of solution sets preserves their energies, and hence also their least energies. □
Together with the attainment results of [22 ] , the solution correspondence gives the following identities for the sharp constants.
Theorem 2.2 (Duality of the sharp constants). For every 1 < p < 3 1<p<3 1 < p < 3 one has
S p , c u r l 1 / p = S p ^ , c u r l 1 / p ^ , (18) S_{p,\mathrm{curl}}^{1/p}=S_{\widehat p,\mathrm{curl}}^{1/\widehat p}, \tag*{(18)} S p , curl 1/ p = S p , curl 1/ p , (18) ( S p , c u r l T ) 1 / p = ( S p ^ , c u r l S ) 1 / p ^ , ( S p , c u r l S ) 1 / p = ( S p ^ , c u r l T ) 1 / p ^ . (19) \left(S_{p,\mathrm{curl}}^T\right)^{1/p}=\left(S_{\widehat p,\mathrm{curl}}^S\right)^{1/\widehat p},\qquad\left(S_{p,\mathrm{curl}}^S\right)^{1/p}=\left(S_{\widehat p,\mathrm{curl}}^T\right)^{1/\widehat p}. \tag*{(19)} ( S p , curl T ) 1/ p = ( S p , curl S ) 1/ p , ( S p , curl S ) 1/ p = ( S p , curl T ) 1/ p . (19) If a minimizer is rescaled to solve (7) , its image under D p \mathcal{D}_p D p attains the corresponding dual constant and solves the dual equation. Thus attainment transfers in both directions.
Proof. By [22 Theorems 1.1 and 1.4] , the unrestricted and the two cylindrical constants are attained, and their gauge-minimal minimizers can be rescaled to solve (7) . Let u u u be such a minimizer for S p , c u r l S_{p,\mathrm{curl}} S p , curl . Theorem 2.1 gives
S p ^ , c u r l 1 / p ^ ≤ Q p ^ ( D p u ) 1 / p ^ = Q p ( u ) 1 / p = S p , c u r l 1 / p . S_{\widehat p,\mathrm{curl}}^{1/\widehat p}\leq Q_{\widehat p}(\mathcal{D}_p u)^{1/\widehat p}=Q_p(u)^{1/p}=S_{p,\mathrm{curl}}^{1/p}. S p , curl 1/ p ≤ Q p ( D p u ) 1/ p = Q p ( u ) 1/ p = S p , curl 1/ p . Interchanging p p p and p ^ \widehat p p proves (18) . Starting instead with a toroidal minimizer yields
( S p ^ , c u r l S ) 1 / p ^ ≤ ( S p , c u r l T ) 1 / p , \left(S_{\widehat p,\mathrm{curl}}^S\right)^{1/\widehat p}\leq\left(S_{p,\mathrm{curl}}^T\right)^{1/p}, ( S p , curl S ) 1/ p ≤ ( S p , curl T ) 1/ p , because the dual solution is poloidal. Applying the inverse correspondence to a poloidal minimizer gives the reverse inequality. Interchanging the cylindrical classes proves the remaining identity. Equality in these comparisons also shows that the image of every normalized minimizer is a minimizer of the paired problem. □
Corollary 2.3 (The exponent dual to two). Let S S o b , 2 = 3 ( π / 2 ) 4 / 3 S_{\mathrm{Sob},2}=3(\pi/2)^{4/3} S Sob , 2 = 3 ( π /2 ) 4/3 be the classical scalar Sobolev constant in dimension three. Then
S 6 / 5 , c u r l = S 2 , c u r l 3 / 5 > S S o b , 2 3 / 5 = 3 3 / 5 ( π 2 ) 4 / 5 . S_{6/5,\mathrm{curl}}=S_{2,\mathrm{curl}}^{3/5}>S_{\mathrm{Sob},2}^{3/5}=3^{3/5}\left(\frac{\pi}{2}\right)^{4/5}. S 6/5 , curl = S 2 , curl 3/5 > S Sob , 2 3/5 = 3 3/5 ( 2 π ) 4/5 . The sharp inequality
∥ v ∥ 2 ≤ S 2 , c u r l − 1 / 2 ∥ ∇ × v ∥ 6 / 5 (20) \lVert v\rVert_2\leq S_{2,\mathrm{curl}}^{-1/2}\lVert\nabla\times v\rVert_{6/5} \tag*{(20)} ∥ v ∥ 2 ≤ S 2 , curl − 1/2 ∥ ∇ × v ∥ 6/5 (20) holds for every v ∈ L 2 ( R 3 , R 3 ) v\in L^2(\mathbb{R}^3,\mathbb{R}^3) v ∈ L 2 ( R 3 , R 3 ) satisfying div v = 0 \operatorname{div}v=0 div v = 0 and ∇ × v ∈ L 6 / 5 ( R 3 , R 3 ) \nabla\times v\in L^{6/5}(\mathbb{R}^3,\mathbb{R}^3) ∇ × v ∈ L 6/5 ( R 3 , R 3 ) , and is attained. If u u u is any ground state furnished by [21 Theorem 1.2] , then v = ∇ × u v=\nabla\times u v = ∇ × u is a ground state of
∇ × ( ∣ ∇ × v ∣ − 4 / 5 ∇ × v ) = v in R 3 , \nabla\times\left(\lvert\nabla\times v\rvert^{-4/5}\nabla\times v\right)=v\qquad\text{in }\mathbb{R}^3, ∇ × ( ∣ ∇ × v ∣ − 4/5 ∇ × v ) = v in R 3 ,
and the inverse formula is u = ∣ ∇ × v ∣ − 4 / 5 ∇ × v u = \lvert\nabla\times v\rvert^{-4/5}\nabla\times v u = ∣ ∇ × v ∣ − 4/5 ∇ × v . Their common ground-state energy is
J 6 / 5 ( v ) = J 2 ( u ) = 1 3 S 2 , c u r l 3 / 2 = 1 3 S 6 / 5 , c u r l 5 / 2 . J_{6/5}(v) = J_2(u) = \frac{1}{3}S_{2,\mathrm{curl}}^{3/2} = \frac{1}{3}S_{6/5,\mathrm{curl}}^{5/2}. J 6/5 ( v ) = J 2 ( u ) = 3 1 S 2 , curl 3/2 = 3 1 S 6/5 , curl 5/2 .
For the cylindrical constants one has
S 6 / 5 , c u r l S = ( S 2 , c u r l T ) 3 / 5 , S 6 / 5 , c u r l T = ( S 2 , c u r l S ) 3 / 5 . S_{6/5,\mathrm{curl}}^{S} = \left(S_{2,\mathrm{curl}}^{T}\right)^{3/5}, \qquad S_{6/5,\mathrm{curl}}^{T} = \left(S_{2,\mathrm{curl}}^{S}\right)^{3/5}. S 6/5 , curl S = ( S 2 , curl T ) 3/5 , S 6/5 , curl T = ( S 2 , curl S ) 3/5 .
If a toroidal p = 2 p = 2 p = 2 minimizer solves (7) , its curl is a poloidal p = 6 / 5 p = 6/5 p = 6/5 minimizer; the analogous assertion holds with the two classes interchanged.
Proof. For p = 2 p = 2 p = 2 one has p ∗ = 6 p^{*} = 6 p ∗ = 6 , p ^ = 6 / 5 \widehat{p} = 6/5 p = 6/5 , and p ^ ∗ = 2 \widehat{p}^{*} = 2 p ∗ = 2 . Theorem 2.2 , that is, (18) and (19) , gives the constant identities. The strict inequality S 2 , c u r l > S S o b , 2 S_{2,\mathrm{curl}} > S_{\mathrm{Sob},2} S 2 , curl > S Sob , 2 and attainment at p = 2 p = 2 p = 2 are precisely [21 Theorem 1.2(a)–(b)] . In L 2 L^{2} L 2 the gauge-minimizing representative is the divergence-free one, by the orthogonal Helmholtz decomposition. Taking the 5 / 6 5/6 5/6 power of the p = 6 / 5 p = 6/5 p = 6/5 Sobolev–curl inequality therefore gives (20) . Theorem 2.1 transfers the p = 2 p = 2 p = 2 ground state, proves attainment of this inequality, and gives both the equation and the inverse formula. The cylindrical attainment statements use [22 Theorem 1.4] at p = 2 p = 2 p = 2 .
The existence of ground states at p = 6 / 5 p = 6/5 p = 6/5 is contained in the general theorem of [22 ] : what the corollary adds is their explicit reconstruction from the p = 2 p = 2 p = 2 theory of [21 ] , the exact relation between the sharp constants, and the interchange of the two cylindrical classes. For p ≠ 3 / 2 p \ne3/2 p = 3/2 the identities (19) relate different exponents and say nothing about which of S p , c u r l T S_{p,\mathrm{curl}}^{T} S p , curl T and S p , c u r l S S_{p,\mathrm{curl}}^{S} S p , curl S is smaller. Note also that D p \mathcal{D}_{p} D p does not preserve the quotient of arbitrary fields; (17) relies on u u u being a solution.
The self-dual exponent
From now on we take p = 3 / 2 p = 3/2 p = 3/2 , so that p ∗ = 3 p^{*} = 3 p ∗ = 3 . The equation becomes
∇ × ( ∣ ∇ × u ∣ − 1 / 2 ∇ × u ) = ∣ u ∣ u . (21) \nabla\times\left(\lvert\nabla\times u\rvert^{-1/2}\nabla\times u\right) = \lvert u\rvert u. \tag*{(21)} ∇ × ( ∣ ∇ × u ∣ − 1/2 ∇ × u ) = ∣ u ∣ u . (21)
For a field satisfying the Nehari identity
∫ R 3 ∣ ∇ × u ∣ 3 / 2 d x = ∫ R 3 ∣ u ∣ 3 d x , \int_{\mathbb{R}^{3}} \lvert\nabla\times u\rvert^{3/2}\,dx = \int_{\mathbb{R}^{3}} \lvert u\rvert^{3}\,dx, ∫ R 3 ∣ ∇ × u ∣ 3/2 d x = ∫ R 3 ∣ u ∣ 3 d x ,
we have
J 3 / 2 ( u ) = 1 3 ∫ R 3 ∣ u ∣ 3 d x = 1 3 ∫ R 3 ∣ ∇ × u ∣ 3 / 2 d x . J_{3/2}(u) = \frac{1}{3}\int_{\mathbb{R}^{3}} \lvert u\rvert^{3}\,dx = \frac{1}{3}\int_{\mathbb{R}^{3}} \lvert\nabla\times u\rvert^{3/2}\,dx. J 3/2 ( u ) = 3 1 ∫ R 3 ∣ u ∣ 3 d x = 3 1 ∫ R 3 ∣ ∇ × u ∣ 3/2 d x .
At this exponent the gauge-invariant quotient is Q Q Q of (1) , the infimum in its denominator being taken over W \mathcal{W} W . If u ∈ M u \in\mathcal{M} u ∈ M , then w ( u ) = 0 w(u) = 0 w ( u ) = 0 and this denominator is simply ( ∫ R 3 ∣ u ∣ 3 d x ) 1 / 2 \left(\int_{\mathbb{R}^{3}} \lvert u\rvert^{3}\,dx\right)^{1/2} ( ∫ R 3 ∣ u ∣ 3 d x ) 1/2 .
We recall the following elementary gauge property, already used in [3 Section 1] . For the reader’s convenience, we include a short proof.
Lemma 2.4. If A = a ( ρ , z ) e θ ∈ L 3 ( R 3 , R 3 ) A = a(\rho,z)e_{\theta} \in L^{3}(\mathbb{R}^{3},\mathbb{R}^{3}) A = a ( ρ , z ) e θ ∈ L 3 ( R 3 , R 3 ) , then
div ( ∣ A ∣ A ) = 0 in D ′ ( R 3 ) . \operatorname{div}(\lvert A\rvert A) = 0 \quad\text{in } \mathcal{D}'(\mathbb{R}^{3}). div (∣ A ∣ A ) = 0 in D ′ ( R 3 ) . Consequently, the infimum of ∥ A + w ∥ 3 \lVert A+w\rVert_{3} ∥ A + w ∥ 3 over curl-free w ∈ L 3 w \in L^{3} w ∈ L 3 is attained at w = 0 w = 0 w = 0 . In particular, for a finite-energy field A A A , one has w ( A ) = 0 w(A) = 0 w ( A ) = 0 in (9) .
Proof. For φ ∈ C c ∞ ( R 3 ) \varphi\in C^\infty_c(\mathbb{R}^3) φ ∈ C c ∞ ( R 3 ) , cylindrical coordinates give
∫ R 3 ∣ A ∣ A ⋅ ∇ φ d x = ∫ R ∫ 0 ∞ ∣ a ∣ a ( ∫ 0 2 π ∂ θ φ ( ρ cos θ , ρ sin θ , z ) d θ ) d ρ d z = 0. \int_{\mathbb{R}^3} |A|A \cdot\nabla\varphi\,\mathrm{d}x
= \int_{\mathbb{R}}\int_0^\infty|a|a\left(\int_0^{2\pi}\partial_\theta\varphi(\rho\cos\theta,\rho\sin\theta,z)\,\mathrm{d}\theta\right)\,\mathrm{d}\rho\,\mathrm{d}z=0. ∫ R 3 ∣ A ∣ A ⋅ ∇ φ d x = ∫ R ∫ 0 ∞ ∣ a ∣ a ( ∫ 0 2 π ∂ θ φ ( ρ cos θ , ρ sin θ , z ) d θ ) d ρ d z = 0.
The integral is absolutely convergent because ∣ A ∣ A ∈ L 3 / 2 |A|A\in L^{3/2} ∣ A ∣ A ∈ L 3/2 and ∇ φ \nabla\varphi ∇ φ is bounded and compactly supported. This argument also rules out a distribution supported on the axis. The conclusion follows from (10) –(11) and uniqueness of the gauge minimizer. □
The hyperbolic problem
In this section we fix the hyperbolic notation and collect the properties of the reduced variational problem. For the hyperbolic geometry used below we refer to [24 ] .
Upper half-plane model
We use the upper half-plane model
H 2 : = { ( ρ , z ) ∈ R 2 : ρ > 0 } , g : = d ρ 2 + d z 2 ρ 2 . \mathbb{H}^2:=\{(\rho,z)\in\mathbb{R}^2:\rho>0\},\qquad g:=\frac{\mathrm{d}\rho^2+\mathrm{d}z^2}{\rho^2}. H 2 := {( ρ , z ) ∈ R 2 : ρ > 0 } , g := ρ 2 d ρ 2 + d z 2 .
It is the complete, simply connected two-dimensional Riemannian manifold of constant sectional curvature − 1 -1 − 1 . Its volume element is
d V g = d ρ d z ρ 2 . \mathrm{d}V_g=\frac{\mathrm{d}\rho\,\mathrm{d}z}{\rho^2}. d V g = ρ 2 d ρ d z .
For a scalar function ψ \psi ψ ,
∣ ∇ g ψ ∣ g = ρ ∣ ∇ ψ ∣ , |\nabla_g\psi|_g=\rho|\nabla\psi|, ∣ ∇ g ψ ∣ g = ρ ∣∇ ψ ∣ ,
where ∇ \nabla ∇ on the right is the ordinary Euclidean gradient in the ( ρ , z ) (\rho,z) ( ρ , z ) variables. The hyperbolic q q q -Laplacian is
Δ q , g ψ : = div g ( ∣ ∇ g ψ ∣ g q − 2 ∇ g ψ ) . (22) \Delta_{q,g}\psi:=\operatorname{div}_g\left(|\nabla_g\psi|_g^{q-2}\nabla_g\psi\right). \tag*{(22)} Δ q , g ψ := div g ( ∣ ∇ g ψ ∣ g q − 2 ∇ g ψ ) . (22)
In dimension two, formula (22) becomes
Δ q , g ψ = ρ 2 div ρ , z ( ∣ ∇ ψ ∣ q − 2 ∇ ψ ) . \Delta_{q,g}\psi=\rho^2\operatorname{div}_{\rho,z}\left(|\nabla\psi|^{q-2}\nabla\psi\right). Δ q , g ψ = ρ 2 div ρ , z ( ∣∇ ψ ∣ q − 2 ∇ ψ ) .
In particular,
Δ 3 / 2 , g ψ = ρ 2 div ρ , z ( ρ − 1 / 2 ∣ ∇ ψ ∣ − 1 / 2 ∇ ψ ) . (23) \Delta_{3/2,g}\psi=\rho^2\operatorname{div}_{\rho,z}\left(\rho^{-1/2}|\nabla\psi|^{-1/2}\nabla\psi\right). \tag*{(23)} Δ 3/2 , g ψ = ρ 2 div ρ , z ( ρ − 1/2 ∣∇ ψ ∣ − 1/2 ∇ ψ ) . (23)
Hyperbolic polar coordinates
Fix the point
O : = ( 1 , 0 ) ∈ H 2 O:=(1,0)\in\mathbb{H}^2 O := ( 1 , 0 ) ∈ H 2
and let
t : = dist g ( O , ( ρ , z ) ) . t:=\operatorname{dist}_g(O,(\rho,z)). t := dist g ( O , ( ρ , z )) .
The distance formula in the upper half-plane gives
cosh t = 1 + ( ρ − 1 ) 2 + z 2 2 ρ = 1 + ρ 2 + z 2 2 ρ . (24) \cosh t=1+\frac{(\rho-1)^2+z^2}{2\rho}=\frac{1+\rho^2+z^2}{2\rho}. \tag*{(24)} cosh t = 1 + 2 ρ ( ρ − 1 ) 2 + z 2 = 2 ρ 1 + ρ 2 + z 2 . (24)
Throughout we use the abbreviations
s : = 1 + ρ 2 + z 2 , η : = 2 ρ s , s:=1+\rho^2+z^2,\qquad\eta:=\frac{2\rho}{s}, s := 1 + ρ 2 + z 2 , η := s 2 ρ ,
so that
η = sech t , 0 < η ≤ 1. \eta=\operatorname{sech}t,\qquad0<\eta\leq1. η = sech t , 0 < η ≤ 1.
The value η = 1 \eta=1 η = 1 corresponds to the centre ( ρ , z ) = ( 1 , 0 ) (\rho,z)=(1,0) ( ρ , z ) = ( 1 , 0 ) , and η → 0 \eta\to0 η → 0 to hyperbolic infinity.
In geodesic polar coordinates ( t , φ ) (t,\varphi) ( t , φ ) centred at O O O ,
d V g = sinh t d t d φ . dV_g=\sinh t\,dt\,d\varphi. d V g = sinh t d t d φ .
For a radial function ψ = ψ ( t ) \psi=\psi(t) ψ = ψ ( t ) ,
Δ q , g ψ = 1 sinh t d d t ( sinh t ∣ ψ ′ ∣ q − 2 ψ ′ ) . \Delta_{q,g}\psi=\frac{1}{\sinh t}\frac{d}{dt}\left(\sinh t\lvert\psi'\rvert^{q-2}\psi'\right). Δ q , g ψ = sinh t 1 d t d ( sinh t ∣ ψ ′ ∣ q − 2 ψ ′ ) .
For a > 0 a>0 a > 0 , let ψ a ( t ) \psi_a(t) ψ a ( t ) be the radial solution about O O O of
− 1 sinh t d d t ( sinh t ∣ ψ a ′ ∣ − 1 / 2 ψ a ′ ) = ψ a 2 , t > 0 , (25) -\frac{1}{\sinh t}\frac{d}{dt}\left(\sinh t\lvert\psi_a'\rvert^{-1/2}\psi_a'\right)=\psi_a^2,\qquad t>0, \tag*{(25)} − sinh t 1 d t d ( sinh t ∣ ψ a ′ ∣ − 1/2 ψ a ′ ) = ψ a 2 , t > 0 , (25)
with ψ a ( 0 ) = a \psi_a(0)=a ψ a ( 0 ) = a and ψ a ′ ( 0 ) = 0 \psi_a'(0)=0 ψ a ′ ( 0 ) = 0 , considered up to its first zero.
A hyperbolic isometry is a bijection F : H 2 → H 2 F:\mathbb{H}^2\to\mathbb{H}^2 F : H 2 → H 2 preserving the distance dist g \operatorname{dist}_g dist g . For ρ 0 > 0 \rho_0>0 ρ 0 > 0 and z 0 ∈ R z_0\in\mathbb{R} z 0 ∈ R , the map
F ρ 0 , z 0 ( ρ , z ) = ( ρ 0 ρ , ρ 0 z + z 0 ) F_{\rho_0,z_0}(\rho,z)=(\rho_0\rho,\rho_0z+z_0) F ρ 0 , z 0 ( ρ , z ) = ( ρ 0 ρ , ρ 0 z + z 0 )
is an isometry sending O O O to ( ρ 0 , z 0 ) (\rho_0,z_0) ( ρ 0 , z 0 ) ; these maps act transitively on H 2 \mathbb{H}^2 H 2 , preserve the two integrals defining Λ H 2 \Lambda_{\mathbb{H}^2} Λ H 2 in (3) and equation (4) , and correspond to the Euclidean maps T 1 / ρ 0 , z 0 T_{1/\rho_0,z_0} T 1/ ρ 0 , z 0 of (5) under ψ ↦ A ( ψ ) \psi\mapsto A(\psi) ψ ↦ A ( ψ ) .
The reduced variational problem
The hyperbolic Poincaré inequality identifies the completion D 1 , 3 / 2 ( H 2 ) D^{1,3/2}(\mathbb{H}^2) D 1 , 3/2 ( H 2 ) of Section 1 with W 1 , 3 / 2 ( H 2 ) W^{1,3/2}(\mathbb{H}^2) W 1 , 3/2 ( H 2 ) , with equivalent norms. We study the constant Λ H 2 \Lambda_{\mathbb{H}^2} Λ H 2 defined in (3) ; since the critical Sobolev exponent of the 3 / 2 3/2 3/2 -Laplacian in dimension two is ( 3 / 2 ) 2 ∗ = 6 (3/2)_2^*=6 ( 3/2 ) 2 ∗ = 6 , the exponent 3 3 3 there is subcritical.
The positivity of Λ H 2 \Lambda_{\mathbb{H}^2} Λ H 2 follows by interpolation. Let S E = 2 ( 4 π 2 / ( 9 3 ) ) 3 / 4 S_E=2(4\pi^2/(9\sqrt{3}))^{3/4} S E = 2 ( 4 π 2 / ( 9 3 ) ) 3/4 be the sharp constant of the Euclidean Sobolev inequality in dimension two with gradient exponent 3 / 2 3/2 3/2 [1 , 26 ] . The hyperbolic Poincaré inequality ∥ ψ ∥ 3 / 2 ≤ 2 3 ∥ ∇ g ψ ∥ 3 / 2 \|\psi\|_{3/2}\leq\frac{2}{3}\|\nabla_g\psi\|_{3/2} ∥ ψ ∥ 3/2 ≤ 3 2 ∥ ∇ g ψ ∥ 3/2 and the critical embedding ∥ ψ ∥ 6 ≤ S E − 2 / 3 ∥ ∇ g ψ ∥ 3 / 2 \|\psi\|_6\leq S_E^{-2/3}\|\nabla_g\psi\|_{3/2} ∥ ψ ∥ 6 ≤ S E − 2/3 ∥ ∇ g ψ ∥ 3/2 of [5 (6.2)] , all norms taken with respect to d V g dV_g d V g , give
∥ ψ ∥ 3 ≤ ∥ ψ ∥ 3 / 2 1 / 3 ∥ ψ ∥ 6 2 / 3 ≤ ( 3 / 2 ) 1 / 3 S E − 4 / 9 ∥ ∇ g ψ ∥ 3 / 2 , (26) \|\psi\|_3\leq\|\psi\|_{3/2}^{1/3}\|\psi\|_6^{2/3}\leq(3/2)^{1/3}S_E^{-4/9}\|\nabla_g\psi\|_{3/2}, \tag*{(26)} ∥ ψ ∥ 3 ≤ ∥ ψ ∥ 3/2 1/3 ∥ ψ ∥ 6 2/3 ≤ ( 3/2 ) 1/3 S E − 4/9 ∥ ∇ g ψ ∥ 3/2 , (26)
hence Λ H 2 ≥ 2 / 3 S E 2 / 3 > 0 \Lambda_{\mathbb{H}^2}\geq\sqrt{2/3}\,S_E^{2/3}>0 Λ H 2 ≥ 2/3 S E 2/3 > 0 .
The following compactness statement is the case n = 2 n=2 n = 2 , p = 3 / 2 p=3/2 p = 3/2 , q = 3 q=3 q = 3 of [5 Lemma 6.1] .
Lemma 3.1 (Radial compactness at a fixed centre). Let D r a d , O 1 , 3 / 2 ( H 2 ) D_{\mathrm{rad},O}^{1,3/2}(\mathbb{H}^2) D rad , O 1 , 3/2 ( H 2 ) be the closure, in the homogeneous energy norm , of the smooth compactly supported functions that are radial about O O O , where
∥ f ∥ D 1 , 3 / 2 : = ( ∫ H 2 ∣ ∇ g f ∣ g 3 / 2 d V g ) 2 / 3 . \|f\|_{D^{1,3/2}}:=\left(\int_{\mathbb{H}^2}\lvert\nabla_g f\rvert_g^{3/2}\,dV_g\right)^{2/3}. ∥ f ∥ D 1 , 3/2 := ( ∫ H 2 ∣ ∇ g f ∣ g 3/2 d V g ) 2/3 . Then the following embedding is compact:
D r a d , O 1 , 3 / 2 ( H 2 ) ↪ L 3 ( H 2 ) . D_{\mathrm{rad},O}^{1,3/2}(\mathbb{H}^2)\hookrightarrow L^3(\mathbb{H}^2). D rad , O 1 , 3/2 ( H 2 ) ↪ L 3 ( H 2 ) .
For equation (4) , the symmetry theorem of Dutta and Sandeep [5 Theorem 1.1] states that every positive finite-energy solution is radial about some point of H 2 \mathbb{H}^{2} H 2 , strictly decreasing in the hyperbolic distance t t t , and satisfies ψ ( t ) ≍ ∣ ψ ′ ( t ) ∣ ≍ e − 2 t \psi(t) \asymp\lvert\psi'(t)\rvert\asymp e^{-2t} ψ ( t ) ≍ ∣ ψ ′ ( t )∣ ≍ e − 2 t as t → ∞ t \to\infty t → ∞ . An extremal for (3) exists by [5 Theorem 6.2] .
Every real minimizer has one sign: otherwise, applying (3) to its positive and negative parts and adding their gradient energies contradicts minimality, since m + + m − > m + + m − \sqrt{m_{+}}+\sqrt{m_{-}}>\sqrt{m_{+}+m_{-}} m + + m − > m + + m − , where m ± : = ∫ H 2 ∣ ψ ± ∣ 3 d V g > 0 m_{\pm}:=\int_{\mathbb{H}^{2}}\lvert\psi_{\pm}\rvert^{3}\,dV_{g}>0 m ± := ∫ H 2 ∣ ψ ± ∣ 3 d V g > 0 . After changing the sign if necessary, local regularity and the strong maximum principle [27 ] give strict positivity. After a positive rescaling we may therefore choose a positive extremal ψ ∗ \psi_{*} ψ ∗ solving (4) . By applying a hyperbolic isometry, we place its centre at O = ( 1 , 0 ) O=(1,0) O = ( 1 , 0 ) and write ψ ∗ = ψ ∗ ( t ) \psi_{*}=\psi_{*}(t) ψ ∗ = ψ ∗ ( t ) . It satisfies (25) with
ψ ∗ ′ ( 0 ) = 0 , ψ ∗ ( t ) > 0 , ψ ∗ ′ ( t ) < 0 ( t > 0 ) , ψ ∗ ( t ) → 0 ( t → ∞ ) . (27) \psi_{*}'(0)=0,\qquad\psi_{*}(t)>0,\qquad\psi_{*}'(t)<0\ (t>0),\qquad\psi_{*}(t)\to0\ (t\to\infty). \tag*{(27)} ψ ∗ ′ ( 0 ) = 0 , ψ ∗ ( t ) > 0 , ψ ∗ ′ ( t ) < 0 ( t > 0 ) , ψ ∗ ( t ) → 0 ( t → ∞ ) . (27)
We do not know a closed formula for ψ ∗ \psi_{*} ψ ∗ .
Uniqueness and the shooting trichotomy
The uniqueness of the positive finite-energy profile is the following classification of the initial-value solutions ψ a \psi_{a} ψ a of (25) .
Theorem 3.2. There exists a unique number a ∗ > 0 a_{*}>0 a ∗ > 0 with the following properties:
(i) if 0 < a < a ∗ 0<a<a_{*} 0 < a < a ∗ , then ψ a \psi_{a} ψ a remains positive, satisfies
ψ a ( t ) → 0 , e 2 t ψ a ( t ) → ∞ ( t → ∞ ) , \psi_{a}(t)\to0,\qquad e^{2t}\psi_{a}(t)\to\infty\qquad(t\to\infty), ψ a ( t ) → 0 , e 2 t ψ a ( t ) → ∞ ( t → ∞ ) , and does not belong to D 1 , 3 / 2 ( H 2 ) D^{1,3/2}(\mathbb{H}^{2}) D 1 , 3/2 ( H 2 ) ;
(ii) if a = a ∗ a=a_{*} a = a ∗ , then ψ a ∗ \psi_{a_{*}} ψ a ∗ remains positive and, for some C ∗ > 0 C_{*}>0 C ∗ > 0 ,
e 2 t ψ a ∗ ( t ) → C ∗ ( t → ∞ ) ; (28) e^{2t}\psi_{a_{*}}(t)\to C_{*}\qquad(t\to\infty); \tag*{(28)} e 2 t ψ a ∗ ( t ) → C ∗ ( t → ∞ ) ; (28) this is the unique positive radial solution of (4) in D 1 , 3 / 2 ( H 2 ) D^{1,3/2}(\mathbb{H}^{2}) D 1 , 3/2 ( H 2 ) with centre O O O ;
(iii) if a > a ∗ a>a_{*} a > a ∗ , then ψ a \psi_{a} ψ a has a zero at a finite hyperbolic radius.
Consequently every positive finite-energy solution of (4) is obtained from ψ a ∗ \psi_{a_{*}} ψ a ∗ by a hyperbolic isometry, and ψ ∗ = ψ a ∗ \psi_{*}=\psi_{a_{*}} ψ ∗ = ψ a ∗ .
The finite-energy assumption is essential: the boundary conditions (27) alone are satisfied by a continuum of positive radial solutions, which tend to zero too slowly to have finite energy. Dutta and Sandeep [5 Theorem 1.1] prove symmetry and two-sided decay but not uniqueness of the nonlinear profile; the substitution r = ( e 2 t − 1 ) / 2 r=(e^{2t}-1)/2 r = ( e 2 t − 1 ) /2 turns the radial equation into a weighted Euclidean problem covered by Kawano, Yanagida and Yotsutani [17 Theorem 1] , whose hypotheses we verify below.
Proof of Theorem 3.2. We transform the radial equation into a weighted Euclidean equation of the type studied by Kawano, Yanagida and Yotsutani [17 ] . Set
r = 1 coth t − 1 = e 2 t − 1 2 , v ( r ) = ψ ( t ) . r=\frac{1}{\coth t-1}=\frac{e^{2t}-1}{2},\qquad v(r)=\psi(t). r = coth t − 1 1 = 2 e 2 t − 1 , v ( r ) = ψ ( t ) .
Then
d r d t = 1 + 2 r , sinh t = r 1 + 2 r , \frac{dr}{dt}=1+2r,\qquad\sinh t=\frac{r}{\sqrt{1+2r}}, d t d r = 1 + 2 r , sinh t = 1 + 2 r r ,
and (25) becomes
− d d r ( r ∣ v ′ ∣ − 1 / 2 v ′ ) = r K ( r ) v 2 , K ( r ) : = ( 1 + 2 r ) − 3 / 2 . (29) -\frac{d}{dr}\left(r\lvert v'\rvert^{-1/2}v'\right)=rK(r)v^{2},\qquad K(r):=(1+2r)^{-3/2}. \tag*{(29)} − d r d ( r ∣ v ′ ∣ − 1/2 v ′ ) = r K ( r ) v 2 , K ( r ) := ( 1 + 2 r ) − 3/2 . (29)
The two radial integrals transform as
∫ 0 ∞ ∣ ψ ′ ∣ 3 / 2 sinh t d t = ∫ 0 ∞ r ∣ v ′ ∣ 3 / 2 d r , \int_{0}^{\infty} |\psi'|^{3/2}\sinh t\,dt = \int_{0}^{\infty} r|v'|^{3/2}\,dr, ∫ 0 ∞ ∣ ψ ′ ∣ 3/2 sinh t d t = ∫ 0 ∞ r ∣ v ′ ∣ 3/2 d r ,
∫ 0 ∞ ∣ ψ ∣ 3 sinh t d t = ∫ 0 ∞ r K ( r ) ∣ v ∣ 3 d r . \int_{0}^{\infty} |\psi|^3\sinh t\,dt = \int_{0}^{\infty} rK(r)|v|^3\,dr. ∫ 0 ∞ ∣ ψ ∣ 3 sinh t d t = ∫ 0 ∞ r K ( r ) ∣ v ∣ 3 d r .
Equation (29) is the radial equation in R n \mathbb{R}^n R n considered in [17 ] , with
n = 2 , m = 3 2 , q K Y Y = 2. n=2,\qquad m=\frac{3}{2},\qquad q_{\mathrm{KYY}}=2. n = 2 , m = 2 3 , q KYY = 2.
(In [17 ] the initial-value problem is continued with ( v + ) q K Y Y (v^+)^{q_{\mathrm{KYY}}} ( v + ) q KYY , which coincides with (29) up to the first zero of v v v , and this is all that the crossing classification requires.) We have n > m > 1 n>m>1 n > m > 1 and q K Y Y > m − 1 q_{\mathrm{KYY}}>m-1 q KYY > m − 1 ; the weight K K K is positive and smooth, with K ( r ) = 1 + O ( r ) K(r)=1+O(r) K ( r ) = 1 + O ( r ) at the origin, so the condition (K) of that paper holds with exponent ν = 0 > − m \nu=0>-m ν = 0 > − m , and the additional requirement imposed there for m > 2 m>2 m > 2 is void.
For these parameters the Pohozaev-type function of [17 ] is
G ( r ) : = 1 3 r 2 K ( r ) − 1 3 ∫ 0 r s K ( s ) d s . \mathcal{G}(r):=\frac{1}{3}r^2K(r)-\frac{1}{3}\int_0^r sK(s)\,ds. G ( r ) := 3 1 r 2 K ( r ) − 3 1 ∫ 0 r sK ( s ) d s .
A direct calculation gives
G ′ ( r ) = 1 3 ( r K ( r ) ) ′ = r ( 1 − r ) 3 ( 1 + 2 r ) 5 / 2 . \mathcal{G}'(r)=\frac{1}{3}\bigl(rK(r)\bigr)'=\frac{r(1-r)}{3(1+2r)^{5/2}}. G ′ ( r ) = 3 1 ( r K ( r ) ) ′ = 3 ( 1 + 2 r ) 5/2 r ( 1 − r ) .
Hence G ′ > 0 \mathcal{G}' > 0 G ′ > 0 on ( 0 , 1 ) (0,1) ( 0 , 1 ) and G ′ < 0 \mathcal{G}' < 0 G ′ < 0 on ( 1 , ∞ ) (1,\infty) ( 1 , ∞ ) . Since G ( 0 ) = 0 \mathcal{G}(0)=0 G ( 0 ) = 0 , we get G ≥ 0 \mathcal{G}\geq0 G ≥ 0 on ( 0 , 1 ) (0,1) ( 0 , 1 ) , G ′ ≤ 0 \mathcal{G}'\leq0 G ′ ≤ 0 on ( 1 , ∞ ) (1,\infty) ( 1 , ∞ ) and G ≢ 0 \mathcal{G}\not\equiv0 G ≡ 0 , which is hypothesis (G) of [17 Theorem 1] with R 1 = 1 R_1=1 R 1 = 1 . To determine which alternative in that theorem applies, we use the positive finite-energy solution provided by [5 Theorem 6.2] . Its symmetry and sharp decay [5 Theorem 1.1] allow us to place its centre at O O O and give
0 < lim inf r → ∞ r v ( r ) ≤ lim sup r → ∞ r v ( r ) < ∞ . 0<\liminf_{r\to\infty}rv(r)\leq\limsup_{r\to\infty}rv(r)<\infty. 0 < r → ∞ lim inf r v ( r ) ≤ r → ∞ lim sup r v ( r ) < ∞.
This solution is neither crossing nor slowly decaying, so the mixed case of the cited classification applies. Since ( n − m ) / ( m − 1 ) = 1 (n-m)/(m-1)=1 ( n − m ) / ( m − 1 ) = 1 , there is exactly one a ∗ > 0 a_*>0 a ∗ > 0 for which
r v ( r ; a ∗ ) → L ∗ ∈ ( 0 , ∞ ) ( r → ∞ ) , (30) rv(r;a_*)\to L_*\in(0,\infty)\qquad(r\to\infty), \tag*{(30)} r v ( r ; a ∗ ) → L ∗ ∈ ( 0 , ∞ ) ( r → ∞ ) , (30)
whereas r v ( r ; a ) → ∞ rv(r;a)\to\infty r v ( r ; a ) → ∞ as r → ∞ r\to\infty r → ∞ for 0 < a < a ∗ 0<a<a_* 0 < a < a ∗ , and the solutions with a > a ∗ a>a_* a > a ∗ cross zero. Since r ∼ 1 2 e 2 t r\sim\frac{1}{2}e^{2t} r ∼ 2 1 e 2 t as t → ∞ t\to\infty t → ∞ , this gives the claimed decay alternatives, with C ∗ = 2 L ∗ C_*=2L_* C ∗ = 2 L ∗ , once we verify that every global positive solution tends to zero.
Every global positive ψ a \psi_a ψ a is decreasing and therefore has a limit ℓ ≥ 0 \ell\geq0 ℓ ≥ 0 . Put z = − ψ a ′ z=\sqrt{-\psi_a'} z = − ψ a ′ . The radial equation gives
z ( t ) = 1 sinh t ∫ 0 t ψ a ( s ) 2 sinh s d s . (31) z(t)=\frac{1}{\sinh t}\int_0^t\psi_a(s)^2\sinh s\,ds. \tag*{(31)} z ( t ) = sinh t 1 ∫ 0 t ψ a ( s ) 2 sinh s d s . (31)
If ℓ > 0 \ell>0 ℓ > 0 , l’Hôpital’s rule in (31) yields z ( t ) → ℓ 2 z(t)\to\ell^2 z ( t ) → ℓ 2 , and hence ψ a ′ ( t ) → − ℓ 4 \psi_a'(t)\to-\ell^4 ψ a ′ ( t ) → − ℓ 4 , a contradiction. Thus ℓ = 0 \ell=0 ℓ = 0 .
Every positive finite-energy solution is radial and has the sharp decay used above, so, after moving its centre to O O O , it must correspond to the unique rapid branch. The previously constructed finite-energy solution belongs to this branch; hence the rapid solution itself has finite energy, and it coincides with ψ ∗ \psi_* ψ ∗ . In particular, no slowly decaying solution ψ a \psi_a ψ a with a < a ∗ a<a_* a < a ∗ has finite energy, which is the last assertion in (i). □
There are earlier hyperbolic versions for the Laplace–Beltrami operator. Mancini and Sandeep [19 Theorems 1.3–1.4] prove uniqueness, up to hyperbolic isometries, of positive finite-energy solutions of the semilinear problem, including − Δ g u = u 2 -\Delta_g u=u^2 − Δ g u = u 2 on H 2 \mathbb{H}^2 H 2 ; for H n \mathbb{H}^n H n , n ≥ 3 n\geq3 n ≥ 3 , Bonforte, Gazzola, Grillo and Vázquez [4 Theorem A] describe the analogous slow-decay, ground-state and zero-crossing alternatives. The quasilinear classification of Muratori and Soave [23 Theorems 1.3–1.5] concerns critical and supercritical source powers q ≥ p ∗ − 1 q\geq p^*-1 q ≥ p ∗ − 1 , which excludes the present source power 2.
Remark 3.3 (Fixing the coefficient in the equation). For λ > 0 \lambda>0 λ > 0 , the rescaling ψ ~ = λ 2 / 3 ψ \widetilde{\psi}=\lambda^{2/3}\psi ψ = λ 2/3 ψ transforms − Δ 3 / 2 , g ψ = λ ψ 2 -\Delta_{3/2,g}\psi=\lambda\psi^2 − Δ 3/2 , g ψ = λ ψ 2 into the coefficient-one equation. Uniqueness of positive extremals of the quotient is therefore understood up to a positive factor and hyperbolic isometries.
Remark 3.4 (Regularity at the hyperbolic centre). Let O O O be the centre of ψ ∗ \psi_* ψ ∗ and a ∗ = ψ ∗ ( O ) a_*=\psi_*(O) a ∗ = ψ ∗ ( O ) . We use the same symbol for the radial profile and the corresponding function on H 2 \mathbb{H}^2 H 2 . By [5 Theorem 1.1] , the profile is C 1 C^1 C 1 at t = 0 t=0 t = 0 , satisfies ψ ∗ ′ ( 0 ) = 0 \psi_*'(0)=0 ψ ∗ ′ ( 0 ) = 0 , and is smooth for t > 0 t>0 t > 0 . In the present case its precise spatial regularity is
ψ ∗ ∈ C l o c 2 , 1 ( H 2 ) ∩ C ∞ ( H 2 ∖ { O } ) , ψ ∗ ∉ C 3 ( H 2 ) . \psi_*\in C_{\mathrm{loc}}^{2,1}(\mathbb{H}^2)\cap C^\infty(\mathbb{H}^2\setminus\{O\}),\qquad\psi_*\notin C^3(\mathbb{H}^2). ψ ∗ ∈ C loc 2 , 1 ( H 2 ) ∩ C ∞ ( H 2 ∖ { O }) , ψ ∗ ∈ / C 3 ( H 2 ) .
Here C l o c 2 , 1 C_{\mathrm{loc}}^{2,1} C loc 2 , 1 means that the second derivatives are locally Lipschitz in smooth coordinates.
Indeed, integrating the radial equation from the centre gives
ψ ∗ ′ ( t ) = − ( ∫ 0 t ψ ∗ ( r ) 2 sinh r d r sinh t ) 2 = − t 2 ( ∫ 0 1 ψ ∗ ( t r ) 2 sinh ( t r ) sinh t d r ) 2 . \psi_*'(t)=-\left(\frac{\int_0^t \psi_*(r)^2\sinh r\,\mathrm{d}r}{\sinh t}\right)^2=-t^2\left(\int_0^1\psi_*(tr)^2\frac{\sinh(tr)}{\sinh t}\,\mathrm{d}r\right)^2. ψ ∗ ′ ( t ) = − ( sinh t ∫ 0 t ψ ∗ ( r ) 2 sinh r d r ) 2 = − t 2 ( ∫ 0 1 ψ ∗ ( t r ) 2 sinh t sinh ( t r ) d r ) 2 .
The quotient in the last integral extends smoothly to t = 0 t=0 t = 0 , with value r r r . Thus the integral preserves C k C^k C k regularity of the profile, and the identity bootstraps ψ ∗ \psi_* ψ ∗ to C ∞ ( [ 0 , ∞ ) ) C^\infty([0,\infty)) C ∞ ([ 0 , ∞ )) , with one-sided derivatives at zero. The integral at zero equals a ∗ 2 / 2 a_*^2/2 a ∗ 2 /2 . Consequently, as t → 0 + t\to0^+ t → 0 + ,
ψ ∗ ′ ( 0 ) = ψ ∗ ′ ′ ( 0 ) = 0 , ψ ∗ ′ ′ ′ ( 0 + ) = − a ∗ 4 2 , ψ ∗ ( t ) = a ∗ − a ∗ 4 12 t 3 + O ( t 4 ) . \psi_*'(0)=\psi_*''(0)=0,\qquad\psi_*'''(0+)=-\frac{a_*^4}{2},\qquad\psi_*(t)=a_*-\frac{a_*^4}{12}t^3+O(t^4). ψ ∗ ′ ( 0 ) = ψ ∗ ′′ ( 0 ) = 0 , ψ ∗ ′′′ ( 0 + ) = − 2 a ∗ 4 , ψ ∗ ( t ) = a ∗ − 12 a ∗ 4 t 3 + O ( t 4 ) .
In geodesic normal coordinates y y y centred at O O O , we have t = ∣ y ∣ t=|y| t = ∣ y ∣ . The radial Hessian formula, together with ψ ∗ ′ ( t ) = O ( t 2 ) \psi_*'(t)=O(t^2) ψ ∗ ′ ( t ) = O ( t 2 ) , ψ ∗ ′ ′ ( t ) = O ( t ) \psi_*''(t)=O(t) ψ ∗ ′′ ( t ) = O ( t ) and boundedness of ψ ∗ ′ ′ ′ \psi_*''' ψ ∗ ′′′ near zero, shows that the spatial second derivatives are locally Lipschitz. However, the restriction to a geodesic through O O O , parametrized by signed arclength s s s , is ψ ∗ ( ∣ s ∣ ) \psi_*(|s|) ψ ∗ ( ∣ s ∣ ) . Its third derivative has limits − a ∗ 4 / 2 -a_*^4/2 − a ∗ 4 /2 as s → 0 + s\to0^+ s → 0 + and a ∗ 4 / 2 a_*^4/2 a ∗ 4 /2 as s → 0 − s\to0^- s → 0 − . Thus the spatial function is not C 3 C^3 C 3 at its centre.
Reduction of the toroidal quotient
We now return to vector fields. A toroidal field a ( ρ , z ) e θ a(\rho,z)e_\theta a ( ρ , z ) e θ is determined by its stream function ψ = ρ a \psi=\rho a ψ = ρ a , and the two integrals entering the quotient Q Q Q are hyperbolic integrals of ψ \psi ψ . We make this correspondence precise on the level of the energy spaces and of the Euler–Lagrange equations.
Hyperbolic reductions already occur in gauge theory: Witten’s S O ( 3 ) SO(3) S O ( 3 ) -invariant self-dual Yang–Mills fields in R 4 \mathbb{R}^4 R 4 reduce to an abelian Higgs model on H 2 \mathbb{H}^2 H 2 [32 ] . This concerns a different equation and symmetry reduction from the present toroidal problem in R 3 \mathbb{R}^3 R 3 . For our ansatz the identities are exact for every 1 < p < 3 1<p<3 1 < p < 3 , with p ∗ = 3 p / ( 3 − p ) p^*=3p/(3-p) p ∗ = 3 p / ( 3 − p ) :
∫ R 3 ∣ ∇ × A ( ψ ) ∣ p d x = 2 π ∫ H 2 ρ 3 − 2 p ∣ ∇ g ψ ∣ g p d V g , ∫ R 3 ∣ A ( ψ ) ∣ p ∗ d x = 2 π ∫ H 2 ρ 3 − p ∗ ∣ ψ ∣ p ∗ d V g . \int_{\mathbb{R}^3}|\nabla\times A(\psi)|^p\,\mathrm{d}x=2\pi\int_{\mathbb{H}^2}\rho^{3-2p}|\nabla_g\psi|_g^p\,\mathrm{d}V_g,\qquad
\int_{\mathbb{R}^3}|A(\psi)|^{p^*}\,\mathrm{d}x=2\pi\int_{\mathbb{H}^2}\rho^{3-p^*}|\psi|^{p^*}\,\mathrm{d}V_g. ∫ R 3 ∣∇ × A ( ψ ) ∣ p d x = 2 π ∫ H 2 ρ 3 − 2 p ∣ ∇ g ψ ∣ g p d V g , ∫ R 3 ∣ A ( ψ ) ∣ p ∗ d x = 2 π ∫ H 2 ρ 3 − p ∗ ∣ ψ ∣ p ∗ d V g .
What is specific to p = 3 / 2 p=3/2 p = 3/2 is that both reduced integrals are unweighted hyperbolic integrals.
Proposition 4.1 (The T \mathcal{T} T –H 2 \mathbb{H}^2 H 2 correspondence). For ψ ∈ D 1 , 3 / 2 ( H 2 ) \psi\in D^{1,3/2}(\mathbb{H}^2) ψ ∈ D 1 , 3/2 ( H 2 ) define
A ( ρ , θ , z ) : = ψ ( ρ , z ) ρ e θ . (32) A(\rho,\theta,z):=\frac{\psi(\rho,z)}{\rho}e_\theta. \tag*{(32)} A ( ρ , θ , z ) := ρ ψ ( ρ , z ) e θ . (32) This defines a linear bijection from D 1 , 3 / 2 ( H 2 ) D^{1,3/2}(\mathbb{H}^2) D 1 , 3/2 ( H 2 ) onto T ∩ D 1 , 3 / 2 ( curl ; R 3 ) \mathcal{T}\cap D^{1,3/2}(\operatorname{curl};\mathbb{R}^3) T ∩ D 1 , 3/2 ( curl ; R 3 ) . In the weak sense,
∇ × A = − ψ z ρ e ρ + ψ ρ ρ e 3 . (33) \nabla\times A=-\frac{\psi_z}{\rho}e_\rho+\frac{\psi_\rho}{\rho}e_3. \tag*{(33)} ∇ × A = − ρ ψ z e ρ + ρ ψ ρ e 3 . (33) Moreover,
∫ R 3 ∣ A ∣ 3 d x = 2 π ∫ H 2 ∣ ψ ∣ 3 d V g , (34) \int_{\mathbb{R}^3}|A|^3\,\mathrm{d}x=2\pi\int_{\mathbb{H}^2}|\psi|^3\,\mathrm{d}V_g, \tag*{(34)} ∫ R 3 ∣ A ∣ 3 d x = 2 π ∫ H 2 ∣ ψ ∣ 3 d V g , (34) ∫ R 3 ∣ ∇ × A ∣ 3 / 2 d x = 2 π ∫ H 2 ∣ ∇ g ψ ∣ g 3 / 2 d V g . (35) \int_{\mathbb{R}^3}|\nabla\times A|^{3/2}\,\mathrm{d}x=2\pi\int_{\mathbb{H}^2}|\nabla_g\psi|_g^{3/2}\,\mathrm{d}V_g. \tag*{(35)} ∫ R 3 ∣∇ × A ∣ 3/2 d x = 2 π ∫ H 2 ∣ ∇ g ψ ∣ g 3/2 d V g . (35) Consequently,
S 3 / 2 , curl T = 2 π Λ H 2 . (36) S^{\mathcal{T}}_{3/2,\operatorname{curl}}=\sqrt{2\pi}\,\Lambda_{\mathbb{H}^2}. \tag*{(36)} S 3/2 , curl T = 2 π Λ H 2 . (36) Proof. For a toroidal field A = a ( ρ , z ) e θ A=a(\rho,z)e_\theta A = a ( ρ , z ) e θ ,
∇ × A = − a z e ρ + ( a ρ + a ρ ) e 3 . \nabla\times A=-a_z e_\rho+\left(a_\rho+\frac{a}{\rho}\right)e_3. ∇ × A = − a z e ρ + ( a ρ + ρ a ) e 3 . Taking a = ψ / ρ a=\psi/\rho a = ψ / ρ gives (33) . Since d x = ρ d ρ d θ d z \mathrm{d}x=\rho\,\mathrm{d}\rho\,\mathrm{d}\theta\,\mathrm{d}z d x = ρ d ρ d θ d z ,
∫ R 3 ∣ A ∣ 3 d x = 2 π ∫ 0 ∞ ∫ R ∣ ψ ∣ 3 ρ 3 ρ d z d ρ = 2 π ∫ H 2 ∣ ψ ∣ 3 d V g . \int_{\mathbb{R}^3}|A|^3\,\mathrm{d}x=2\pi\int_0^\infty\int_{\mathbb{R}}\frac{|\psi|^3}{\rho^3}\rho\,\mathrm{d}z\,\mathrm{d}\rho=2\pi\int_{\mathbb{H}^2}|\psi|^3\,\mathrm{d}V_g. ∫ R 3 ∣ A ∣ 3 d x = 2 π ∫ 0 ∞ ∫ R ρ 3 ∣ ψ ∣ 3 ρ d z d ρ = 2 π ∫ H 2 ∣ ψ ∣ 3 d V g . Similarly,
∣ ∇ × A ∣ = ∣ ∇ ψ ∣ ρ = ∣ ∇ g ψ ∣ g ρ 2 , |\nabla\times A|=\frac{|\nabla\psi|}{\rho}=\frac{|\nabla_g\psi|_g}{\rho^2}, ∣∇ × A ∣ = ρ ∣∇ ψ ∣ = ρ 2 ∣ ∇ g ψ ∣ g , which yields (35) . These computations apply first to ψ ∈ C c ∞ ( H 2 ) \psi\in C^\infty_c(\mathbb{H}^2) ψ ∈ C c ∞ ( H 2 ) , for which the field A A A is smooth and compactly supported away from the symmetry axis. By the continuous embedding (26) and the two norm identities, the map extends to the homogeneous completion, and the distributional curl of the limit field carries no contribution supported on the axis.
To prove that the map is onto, let A ∈ T ∩ D 1 , 3 / 2 ( curl ; R 3 ) A\in\mathcal{T}\cap D^{1,3/2}(\operatorname{curl};\mathbb{R}^3) A ∈ T ∩ D 1 , 3/2 ( curl ; R 3 ) . Cutoff at spatial infinity, followed by convolution with a radial smooth kernel, approximates A A A in L 3 L^3 L 3 and its curl in L 3 / 2 L^{3/2} L 3/2 by smooth, compactly supported toroidal fields A j = a j ( ρ , z ) e θ A_j=a_j(\rho,z)e_\theta A j = a j ( ρ , z ) e θ . The cutoff error is controlled by ∥ ∇ χ R × A ∥ 3 / 2 ≤ ∥ ∇ χ R ∥ 3 ∥ A ∥ L 3 ( { R < ∣ x ∣ < 2 R } ) \|\nabla\chi_R\times A\|_{3/2}\leq\|\nabla\chi_R\|_3\|A\|_{L^3(\{R<|x|<2R\})} ∥∇ χ R × A ∥ 3/2 ≤ ∥∇ χ R ∥ 3 ∥ A ∥ L 3 ({ R < ∣ x ∣ < 2 R }) , and radial convolution preserves the toroidal symmetry. Smoothness at the axis gives a j = O ( ρ ) a_j=O(\rho) a j = O ( ρ ) , so ψ j : = ρ a j = O ( ρ 2 ) \psi_j:=\rho a_j=O(\rho^2) ψ j := ρ a j = O ( ρ 2 ) and ∇ ψ j = O ( ρ ) \nabla\psi_j=O(\rho) ∇ ψ j = O ( ρ ) locally. Multiplying ψ j \psi_j ψ j by a smooth cutoff which is zero for ρ < ε \rho<\varepsilon ρ < ε and one for ρ > 2 ε \rho>2\varepsilon ρ > 2 ε makes it compactly supported in H 2 \mathbb{H}^2 H 2 . The lost hyperbolic L 3 L^3 L 3 mass is O ( ε 5 ) O(\varepsilon^5) O ( ε 5 ) and the additional gradient energy is O ( ε 2 ) O(\varepsilon^2) O ( ε 2 ) , using the weights ρ − 2 \rho^{-2} ρ − 2 and ρ − 1 / 2 \rho^{-1/2} ρ − 1/2 in (34) –(35) . Hence each ψ j \psi_j ψ j lies in D 1 , 3 / 2 ( H 2 ) D^{1,3/2}(\mathbb{H}^2) D 1 , 3/2 ( H 2 ) ; the norm identities make ( ψ j ) (\psi_j) ( ψ j ) Cauchy there, and its limit maps to A A A . By Lemma 2.4 , the gauge correction is zero for every admissible toroidal field. Taking the infimum over this bijection proves (36) . □
Proposition 4.2. Let ψ ∈ D 1 , 3 / 2 ( H 2 ) \psi\in\mathcal{D}^{1,3/2}(\mathbb{H}^{2}) ψ ∈ D 1 , 3/2 ( H 2 ) and let A A A be given by (32) . On { ρ > 0 } \{\rho> 0\} { ρ > 0 } , in the distributional sense,
∇ × ( ∣ ∇ × A ∣ − 1 / 2 ∇ × A ) = − 1 ρ 2 Δ 3 / 2 , g ψ e θ . (37) \nabla\times\left(\lvert\nabla\times A\rvert^{-1/2}\nabla\times A\right)=-\frac{1}{\rho^{2}}\Delta_{3/2,g}\psi e_{\theta}. \tag*{(37)} ∇ × ( ∣ ∇ × A ∣ − 1/2 ∇ × A ) = − ρ 2 1 Δ 3/2 , g ψ e θ . (37) Hence A A A solves (21) if and only if
− Δ 3 / 2 , g ψ = ∣ ψ ∣ ψ in H 2 . -\Delta_{3/2,g}\psi=\lvert\psi\rvert\psi\qquad\text{in }\mathbb{H}^{2}. − Δ 3/2 , g ψ = ∣ ψ ∣ ψ in H 2 . Proof. Put
m : = ρ − 1 / 2 ∣ ∇ ψ ∣ − 1 / 2 where ∇ ψ ≠ 0 , m : = 0 where ∇ ψ = 0. m:=\rho^{-1/2}\lvert\nabla\psi\rvert^{-1/2}\qquad\text{where }\nabla\psi\ne0,\qquad m:=0\qquad\text{where }\nabla\psi=0. m := ρ − 1/2 ∣ ∇ ψ ∣ − 1/2 where ∇ ψ = 0 , m := 0 where ∇ ψ = 0. Using (33) ,
∣ ∇ × A ∣ − 1 / 2 ∇ × A = − m ψ z e ρ + m ψ ρ e 3 . \lvert\nabla\times A\rvert^{-1/2}\nabla\times A=-m\psi_{z}e_{\rho}+m\psi_{\rho}e_{3}. ∣ ∇ × A ∣ − 1/2 ∇ × A = − m ψ z e ρ + m ψ ρ e 3 . The azimuthal component of its curl is
− ∂ z ( m ψ z ) − ∂ ρ ( m ψ ρ ) = − div ρ , z ( ρ − 1 / 2 ∣ ∇ ψ ∣ − 1 / 2 ∇ ψ ) , -\partial_{z}(m\psi_{z})-\partial_{\rho}(m\psi_{\rho})=-\operatorname{div}_{\rho,z}\left(\rho^{-1/2}\lvert\nabla\psi\rvert^{-1/2}\nabla\psi\right), − ∂ z ( m ψ z ) − ∂ ρ ( m ψ ρ ) = − div ρ , z ( ρ − 1/2 ∣ ∇ ψ ∣ − 1/2 ∇ ψ ) , which is (37) by (23) . At zeros of ∇ ψ \nabla\psi ∇ ψ the nonlinear vector expression is assigned its continuous value zero. Approximation in the hyperbolic energy norm proves the distributional identity on { ρ > 0 } \{\rho>0\} { ρ > 0 } for every admissible ψ \psi ψ . Since ∣ A ∣ A = ρ − 2 ∣ ψ ∣ ψ e θ \lvert A\rvert A=\rho^{-2}\lvert\psi\rvert\psi e_{\theta} ∣ A ∣ A = ρ − 2 ∣ ψ ∣ ψ e θ , the two weak equations are equivalent away from the axis; angular averaging reduces arbitrary vector tests there to their toroidal component.
If the scalar equation holds, put B = ∣ ∇ × A ∣ − 1 / 2 ∇ × A ∈ L 3 B=\lvert\nabla\times A\rvert^{-1/2}\nabla\times A\in L^{3} B = ∣ ∇ × A ∣ − 1/2 ∇ × A ∈ L 3 and F = ∣ A ∣ A ∈ L 3 / 2 F=\lvert A\rvert A\in L^{3/2} F = ∣ A ∣ A ∈ L 3/2 . To extend ∇ × B = F \nabla\times B=F ∇ × B = F across the axis, test with χ ε ( ρ ) Φ \chi_{\varepsilon}(\rho)\Phi χ ε ( ρ ) Φ , where Φ ∈ C c ∞ ( R 3 , R 3 ) \Phi\in C_{c}^{\infty}(\mathbb{R}^{3},\mathbb{R}^{3}) Φ ∈ C c ∞ ( R 3 , R 3 ) and χ ε \chi_{\varepsilon} χ ε vanishes for ρ < ε \rho<\varepsilon ρ < ε and equals one for ρ > 2 ε \rho>2\varepsilon ρ > 2 ε . On the fixed support of Φ \Phi Φ ,
∥ ∇ χ ε ∥ L 3 / 2 = O ( ε 1 / 3 ) . \lVert\nabla\chi_{\varepsilon}\rVert_{L^{3/2}}=O(\varepsilon^{1/3}). ∥ ∇ χ ε ∥ L 3/2 = O ( ε 1/3 ) . Hölder’s inequality therefore makes the boundary term involving B ⋅ ( ∇ χ ε × Φ ) B\cdot(\nabla\chi_{\varepsilon}\times\Phi) B ⋅ ( ∇ χ ε × Φ ) tend to zero. The other terms converge by local integrability. Thus the vector equation holds on all of R 3 \mathbb{R}^{3} R 3 , without an axial source. Conversely, testing the global vector equation by toroidal fields supported away from the axis gives the scalar weak equation; density extends its test class to D 1 , 3 / 2 ( H 2 ) \mathcal{D}^{1,3/2}(\mathbb{H}^{2}) D 1 , 3/2 ( H 2 ) .
Nonlinear self-duality and the poloidal minimizer
At p = 3 / 2 p=3/2 p = 3/2 , Theorem 2.1 makes D ( u ) = ∣ ∇ × u ∣ − 1 / 2 ∇ × u \mathcal{D}(u)=\lvert\nabla\times u\rvert^{-1/2}\nabla\times u D ( u ) = ∣ ∇ × u ∣ − 1/2 ∇ × u an involution on finite-energy solutions and interchanges the two cylindrical classes.
Let ψ ∗ \psi_{*} ψ ∗ be the positive hyperbolic minimizer centred at O O O and normalized by − Δ 3 / 2 , g ψ ∗ = ψ ∗ 2 -\Delta_{3/2,g}\psi_{*}=\psi_{*}^{2} − Δ 3/2 , g ψ ∗ = ψ ∗ 2 , and put A ∗ = ρ − 1 ψ ∗ e θ A_{*}=\rho^{-1}\psi_{*}e_{\theta} A ∗ = ρ − 1 ψ ∗ e θ . The correspondence and weak-equation identities in Propositions 4.1 and 4.2 show that A ∗ A_{*} A ∗ is a toroidal extremal solving (21) . Define
U ∗ : = D ( A ∗ ) = ∣ ∇ × A ∗ ∣ − 1 / 2 ∇ × A ∗ . (38) U_{*}:=\mathcal{D}(A_{*})=\lvert\nabla\times A_{*}\rvert^{-1/2}\nabla\times A_{*}. \tag*{(38)} U ∗ := D ( A ∗ ) = ∣ ∇ × A ∗ ∣ − 1/2 ∇ × A ∗ . (38)
Theorems 2.1 and 2.2 show directly that U ∗ U_{*} U ∗ solves the same equation and is a poloidal extremal. We use these facts here to describe its geometry and prove the zero-mode exclusion. Their role in the sharp-constant identities is detailed in § 7 .
The semi-explicit Euclidean formula
We now express U ∗ = D ( A ∗ ) U_*=\mathcal{D}(A_*) U ∗ = D ( A ∗ ) in terms of the radial hyperbolic profile. Define
N ( x ) : = ( 2 x 1 z 2 x 2 z 1 + z 2 − ρ 2 ) = 2 ρ z e ρ + ( 1 + z 2 − ρ 2 ) e 3 . (39) N(x):=\begin{pmatrix}2x_1z\\2x_2z\\1+z^2-\rho^2\end{pmatrix}=2\rho z e_\rho+(1+z^2-\rho^2)e_3. \tag*{(39)} N ( x ) := 2 x 1 z 2 x 2 z 1 + z 2 − ρ 2 = 2 ρ z e ρ + ( 1 + z 2 − ρ 2 ) e 3 . (39)
The elementary identities
∣ N ∣ 2 = ( 1 + z 2 − ρ 2 ) 2 + 4 ρ 2 z 2 = s 2 − 4 ρ 2 = s 2 ( 1 − η 2 ) |N|^2=(1+z^2-\rho^2)^2+4\rho^2z^2=s^2-4\rho^2=s^2(1-\eta^2) ∣ N ∣ 2 = ( 1 + z 2 − ρ 2 ) 2 + 4 ρ 2 z 2 = s 2 − 4 ρ 2 = s 2 ( 1 − η 2 )
and (24) imply
∣ N ∣ = 2 ρ sinh t . (40) |N|=2\rho\sinh t. \tag*{(40)} ∣ N ∣ = 2 ρ sinh t . (40)
For ρ > 0 \rho>0 ρ > 0 and t > 0 t>0 t > 0 , differentiating (24) gives
t ρ = − 1 + z 2 − ρ 2 2 ρ 2 sinh t , t z = z ρ sinh t . t_\rho=-\frac{1+z^2-\rho^2}{2\rho^2\sinh t},\qquad t_z=\frac{z}{\rho\sinh t}. t ρ = − 2 ρ 2 sinh t 1 + z 2 − ρ 2 , t z = ρ sinh t z .
Consequently,
− t z e ρ + t ρ e 3 = − N 2 ρ 2 sinh t . -t_z e_\rho+t_\rho e_3=-\frac{N}{2\rho^2\sinh t}. − t z e ρ + t ρ e 3 = − 2 ρ 2 sinh t N .
For A ∗ = ψ ∗ ( t ) ρ − 1 e θ A_*=\psi_*(t)\rho^{-1}e_\theta A ∗ = ψ ∗ ( t ) ρ − 1 e θ , formula (33) gives
∇ × A ∗ = − ψ ∗ ′ ( t ) 2 ρ 3 sinh t N . (41) \nabla\times A_*=-\frac{\psi_*'(t)}{2\rho^3\sinh t}N. \tag*{(41)} ∇ × A ∗ = − 2 ρ 3 sinh t ψ ∗ ′ ( t ) N . (41)
Since ψ ∗ ′ < 0 \psi_*'<0 ψ ∗ ′ < 0 , (40) yields
∣ ∇ × A ∗ ∣ = − ψ ∗ ′ ( t ) ρ 2 . (42) |\nabla\times A_*|=\frac{-\psi_*'(t)}{\rho^2}. \tag*{(42)} ∣∇ × A ∗ ∣ = ρ 2 − ψ ∗ ′ ( t ) . (42)
Inserting (41) –(42) into (38) gives
U ∗ ( x ) = − ψ ∗ ′ ( t ) 2 ρ 2 sinh t N ( x ) . (43) U_*(x)=\frac{\sqrt{-\psi_*'(t)}}{2\rho^2\sinh t}N(x). \tag*{(43)} U ∗ ( x ) = 2 ρ 2 sinh t − ψ ∗ ′ ( t ) N ( x ) . (43)
Using s = 2 ρ cosh t = 2 ρ / η s=2\rho\cosh t=2\rho/\eta s = 2 ρ cosh t = 2 ρ / η we arrive at the following formula. The continuous endpoint extensions established below define the field also on the axis and the circle { ρ = 1 , z = 0 } \{\rho=1,z=0\} { ρ = 1 , z = 0 } .
Theorem 5.1 (Semi-explicit form of the S S S -extremal). Let ψ ∗ \psi_* ψ ∗ be the positive finite-energy solution of (4) centred at O O O , as in Theorem 3.2 . Define
G ∗ ( η ) : = 2 − ψ ∗ ′ ( t ) η 1 − η 2 , t = arcosh 1 η , 0 < η < 1. (44) G_*(\eta):=\frac{2\sqrt{-\psi_*'(t)}}{\eta\sqrt{1-\eta^2}},\qquad t=\operatorname{arcosh}\frac{1}{\eta},\qquad0<\eta<1. \tag*{(44)} G ∗ ( η ) := η 1 − η 2 2 − ψ ∗ ′ ( t ) , t = arcosh η 1 , 0 < η < 1. (44) We use the continuous extension of G ∗ G_* G ∗ to [ 0 , 1 ] [0,1] [ 0 , 1 ] . Then
U ∗ ( x ) = G ∗ ( 2 ρ 1 + ρ 2 + z 2 ) ( 1 + ρ 2 + z 2 ) 2 ( 2 x 1 z 2 x 2 z 1 + z 2 − ρ 2 ) . (45) U_*(x)=\frac{G_*\left(\frac{2\rho}{1+\rho^2+z^2}\right)}{(1+\rho^2+z^2)^2}
\begin{pmatrix}2x_1z\\2x_2z\\1+z^2-\rho^2\end{pmatrix}. \tag*{(45)} U ∗ ( x ) = ( 1 + ρ 2 + z 2 ) 2 G ∗ ( 1 + ρ 2 + z 2 2 ρ ) 2 x 1 z 2 x 2 z 1 + z 2 − ρ 2 . (45) The field U ∗ U_* U ∗ has the following properties:
(i) U ∗ ∈ S ∩ M U_*\in\mathcal{S}\cap\mathcal{M} U ∗ ∈ S ∩ M and $w(U_*)=0;
(ii) U ∗ U_\ast U ∗ solves the full equation (21) ;
(iii) U ∗ U_\ast U ∗ realizes S 3 / 2 , curl S S^{S}_{3/2,\operatorname{curl}} S 3/2 , curl S ;
(iv) its coefficients in (12) are
α ∗ ( ρ , z ) = 2 z s 2 G ∗ ( 2 ρ s ) , \alpha_\ast(\rho,z)=\frac{2z}{s^2}G_\ast\left(\frac{2\rho}{s}\right), α ∗ ( ρ , z ) = s 2 2 z G ∗ ( s 2 ρ ) ,
β ∗ ( ρ , z ) = 1 + z 2 − ρ 2 s 2 G ∗ ( 2 ρ s ) , \beta_\ast(\rho,z)=\frac{1+z^2-\rho^2}{s^2}G_\ast\left(\frac{2\rho}{s}\right), β ∗ ( ρ , z ) = s 2 1 + z 2 − ρ 2 G ∗ ( s 2 ρ ) ,
and therefore
α ∗ ( ρ , − z ) = − α ∗ ( ρ , z ) , β ∗ ( ρ , − z ) = β ∗ ( ρ , z ) . \alpha_\ast(\rho,-z)=-\alpha_\ast(\rho,z),\qquad\beta_\ast(\rho,-z)=\beta_\ast(\rho,z). α ∗ ( ρ , − z ) = − α ∗ ( ρ , z ) , β ∗ ( ρ , − z ) = β ∗ ( ρ , z ) .
Proof. Formula (45) is equivalent to (43) . Propositions 4.1 and 4.2 make A ∗ A_\ast A ∗ a toroidal extremal solving (21) . Theorems 2.1 and 2.2 then give assertions (i)–(iii) for U ∗ = D ( A ∗ ) U_\ast=\mathcal{D}(A_\ast) U ∗ = D ( A ∗ ) ; the nonlinear gauge condition is part of the solution duality. Assertion (iv) follows from the decomposition of N N N in (39) .
Remark 5.2 (Endpoint values of the poloidal factor). The factor G ∗ G_\ast G ∗ extends continuously to the endpoints. Substituting the centre expansion from Remark 3.4 into (44) , with a ∗ = ψ ∗ ( 0 ) a_\ast=\psi_\ast(0) a ∗ = ψ ∗ ( 0 ) , gives
G ∗ ( 1 ) = a ∗ 2 . G_\ast(1)=a_\ast^2. G ∗ ( 1 ) = a ∗ 2 .
At hyperbolic infinity the value of G ∗ G_\ast G ∗ is determined by the flux
J ( t ) : = sinh t − ψ ∗ ′ ( t ) = ∫ 0 t ψ ∗ ( s ) 2 sinh s d s , \mathcal{J}(t):=\sinh t\sqrt{-\psi_\ast'(t)}=\int_0^t\psi_\ast(s)^2\sinh s\,\mathrm{d}s, J ( t ) := sinh t − ψ ∗ ′ ( t ) = ∫ 0 t ψ ∗ ( s ) 2 sinh s d s ,
which increases to a finite limit J ∞ \mathcal{J}_\infty J ∞ as t → ∞ t\to\infty t → ∞ because ψ ∗ ( t ) = O ( e − 2 t ) \psi_\ast(t)=O(e^{-2t}) ψ ∗ ( t ) = O ( e − 2 t ) . In the same limit, − ψ ∗ ′ ( t ) ∼ 4 J ∞ 2 e − 2 t -\psi_\ast'(t)\sim4\mathcal{J}_\infty^2e^{-2t} − ψ ∗ ′ ( t ) ∼ 4 J ∞ 2 e − 2 t and ψ ∗ ( t ) ∼ 2 J ∞ 2 e − 2 t \psi_\ast(t)\sim2\mathcal{J}_\infty^2e^{-2t} ψ ∗ ( t ) ∼ 2 J ∞ 2 e − 2 t , and (44) gives
G ∗ ( 0 ) = 2 J ∞ = 2 C ∗ = 2 L ∗ , G_\ast(0)=2\mathcal{J}_\infty=\sqrt{2C_\ast}=2\sqrt{L_\ast}, G ∗ ( 0 ) = 2 J ∞ = 2 C ∗ = 2 L ∗ ,
where C ∗ C_\ast C ∗ and L ∗ L_\ast L ∗ are the limits in (28) and (30) .
Absence of zero modes
We now prove Theorem 1.3, adapting the angular obstruction underlying [6 Theorem 1.1] . Part (a) is an angular identity for bounded axisymmetric poloidal potentials, which we isolate as a proposition; for part (b) an explicit conformal map reduces the toroidal extremal to that case, and we verify the L 2 L^2 L 2 integrability of the transformed spinor, which does not follow from conformal covariance alone.
The extremals are indeed not smooth. Along x = ( 1 , 0 , z ) x=(1,0,z) x = ( 1 , 0 , z ) , as z → 0 z\to0 z → 0 , one has t = 2 arsinh ( ∣ z ∣ / 2 ) = ∣ z ∣ + O ( ∣ z ∣ 3 ) t=2\operatorname{arsinh}(|z|/2)=|z|+O(|z|^3) t = 2 arsinh ( ∣ z ∣/2 ) = ∣ z ∣ + O ( ∣ z ∣ 3 ) , so Remark 3.4 gives
( A ∗ ) 2 ( 1 , 0 , z ) = a ∗ − a ∗ 4 12 ∣ z ∣ 3 + O ( ∣ z ∣ 4 ) . (A_\ast)_2(1,0,z)=a_\ast-\frac{a_\ast^4}{12}|z|^3+O(|z|^4). ( A ∗ ) 2 ( 1 , 0 , z ) = a ∗ − 12 a ∗ 4 ∣ z ∣ 3 + O ( ∣ z ∣ 4 ) .
Thus A ∗ A_\ast A ∗ is not C 3 C^3 C 3 across { ρ = 1 , z = 0 } \{\rho=1,z=0\} { ρ = 1 , z = 0 } . Moreover, ∇ × U ∗ = ∣ A ∗ ∣ A ∗ = ρ − 2 ψ ∗ ( t ) 2 e θ \nabla\times U_\ast=|A_\ast|A_\ast=\rho^{-2}\psi_\ast(t)^2e_\theta ∇ × U ∗ = ∣ A ∗ ∣ A ∗ = ρ − 2 ψ ∗ ( t ) 2 e θ , whose second Cartesian component on the same line is a ∗ 2 − a ∗ 5 ∣ z ∣ 3 / 6 + O ( ∣ z ∣ 4 ) a_\ast^2-a_\ast^5|z|^3/6+O(|z|^4) a ∗ 2 − a ∗ 5 ∣ z ∣ 3 /6 + O ( ∣ z ∣ 4 ) . Hence ∇ × U ∗ \nabla\times U_\ast ∇ × U ∗ is not C 3 C^3 C 3 , so U ∗ U_\ast U ∗ is not C 4 C^4 C 4 there. In particular, neither potential satisfies the smoothness assumption in Elton’s stated theorem.
Proposition 5.3 (Bounded poloidal potentials have no zero modes). Let B = b ρ ( ρ , z ) e ρ + b z ( ρ , z ) e 3 ∈ L ∞ ( R 3 ; R 3 ) B=b_\rho(\rho,z)e_\rho+b_z(\rho,z)e_3\in L^\infty(\mathbb{R}^3;\mathbb{R}^3) B = b ρ ( ρ , z ) e ρ + b z ( ρ , z ) e 3 ∈ L ∞ ( R 3 ; R 3 ) . Then ker L 2 D c B = { 0 } \operatorname{ker}_{L^2}D_{cB}=\{0\} ker L 2 D c B = { 0 } for every c ∈ R c\in\mathbb{R} c ∈ R , and D c B D_{cB} D c B is self-adjoint on H 1 ( R 3 ; C 2 ) H^1(\mathbb{R}^3;\mathbb{C}^2) H 1 ( R 3 ; C 2 ) .
Proof. Step 1: angular decomposition for a bounded poloidal potential. Fix c ∈ R c \in\mathbb{R} c ∈ R . The coefficients of B B B are independent of θ \theta θ , and there is no e θ e_\theta e θ component. If Ψ ∈ L 2 ( R 3 ; C 2 ) \Psi\in L^2(\mathbb{R}^3;\mathbb{C}^2) Ψ ∈ L 2 ( R 3 ; C 2 ) solves D c B Ψ = 0 \mathcal{D}_{cB}\Psi=0 D c B Ψ = 0 distributionally, then D 0 Ψ = c σ ⋅ B Ψ ∈ L 2 \mathcal{D}_0\Psi=c\boldsymbol{\sigma}\cdot B\Psi\in L^2 D 0 Ψ = c σ ⋅ B Ψ ∈ L 2 . The Pauli identity ∣ ( σ ⋅ ξ ) v ∣ 2 = ∣ ξ ∣ 2 ∣ v ∣ 2 \lvert(\boldsymbol{\sigma}\cdot\xi)v\rvert^2=\lvert\xi\rvert^2\lvert v\rvert^2 ∣( σ ⋅ ξ ) v ∣ 2 = ∣ ξ ∣ 2 ∣ v ∣ 2 and Fourier transformation give
∥ ∇ Ψ ∥ 2 = ∥ D 0 Ψ ∥ 2 ≤ ∣ c ∣ ∥ B ∥ ∞ ∥ Ψ ∥ 2 . \lVert\nabla\Psi\rVert_2=\lVert\mathcal{D}_0\Psi\rVert_2\leq\lvert c\rvert\lVert B\rVert_\infty\lVert\Psi\rVert_2. ∥ ∇Ψ ∥ 2 = ∥ D 0 Ψ ∥ 2 ≤ ∣ c ∣ ∥ B ∥ ∞ ∥ Ψ ∥ 2 .
Hence Ψ ∈ H 1 \Psi\in H^1 Ψ ∈ H 1 . Also, multiplication by c σ ⋅ B c\boldsymbol{\sigma}\cdot B c σ ⋅ B is bounded and Hermitian, so the self-adjoint operator D c B \mathcal{D}_{cB} D c B has domain H 1 H^1 H 1 .
To exploit the symmetry, use the rotating spin frame S ( θ ) = exp ( − i θ σ 3 / 2 ) S(\theta)=\exp(-i\theta\sigma_3/2) S ( θ ) = exp ( − i θ σ 3 /2 ) . Since S ( θ + 2 π ) = − S ( θ ) S(\theta+2\pi)=-S(\theta) S ( θ + 2 π ) = − S ( θ ) whereas Ψ \Psi Ψ is periodic, S ( θ ) ∗ Ψ S(\theta)^*\Psi S ( θ ) ∗ Ψ is antiperiodic. Its Fourier frequencies are therefore half-integers, and the angular modes have the form
Ψ m = S ( θ ) e i m θ u m ( ρ , z ) , m ∈ Z + 1 2 , u m : ( 0 , ∞ ) × R → C 2 . \Psi_m=S(\theta)e^{im\theta}u_m(\rho,z),\qquad m\in\mathbb{Z}+\frac{1}{2},\qquad u_m:(0,\infty)\times\mathbb{R}\to\mathbb{C}^2. Ψ m = S ( θ ) e im θ u m ( ρ , z ) , m ∈ Z + 2 1 , u m : ( 0 , ∞ ) × R → C 2 .
The projections onto these modes preserve H 1 H^1 H 1 and commute with D c B \mathcal{D}_{cB} D c B : they are averages of simultaneous rotations of space and spin, under which B B B is invariant. Consequently each Ψ m \Psi_m Ψ m is an H 1 H^1 H 1 solution of the zero-mode equation. In the rotating frame its equation is H m u m = 0 H_m u_m=0 H m u m = 0 , where
H m = − i σ 1 ( ∂ ρ + 1 2 ρ ) − i σ 3 ∂ z − c ( b ρ σ 1 + b z σ 3 ) + m ρ σ 2 . H_m=-i\sigma_1\left(\partial_\rho+\frac{1}{2\rho}\right)-i\sigma_3\partial_z-c(b_\rho\sigma_1+b_z\sigma_3)+\frac{m}{\rho}\sigma_2. H m = − i σ 1 ( ∂ ρ + 2 ρ 1 ) − i σ 3 ∂ z − c ( b ρ σ 1 + b z σ 3 ) + ρ m σ 2 .
The term 1 / ( 2 ρ ) 1/(2\rho) 1/ ( 2 ρ ) comes from differentiating the spin frame. It also makes the radial derivative formally skew-adjoint for the cylindrical measure ρ d ρ d z \rho\,d\rho\,dz ρ d ρ d z , so H m H_m H m is formally symmetric for that measure. Step 2: the angular identity and its boundary terms. First, the distributional identity div e ρ = 1 / ρ \operatorname{div}e_\rho=1/\rho div e ρ = 1/ ρ , integration by parts and Cauchy–Schwarz give
∫ R 3 ∣ Ψ m ∣ 2 ρ d x ≤ 2 ∥ Ψ m ∥ 2 ∥ ∇ Ψ m ∥ 2 < ∞ . \int_{\mathbb{R}^3}\frac{\lvert\Psi_m\rvert^2}{\rho}\,dx\leq2\lVert\Psi_m\rVert_2\lVert\nabla\Psi_m\rVert_2<\infty. ∫ R 3 ρ ∣ Ψ m ∣ 2 d x ≤ 2 ∥ Ψ m ∥ 2 ∥ ∇ Ψ m ∥ 2 < ∞.
This is initially proved for smooth compactly supported spinors and extends to H 1 H^1 H 1 by density. Since the left-hand side equals 2 π ∫ R ∫ 0 ∞ ∣ u m ∣ 2 d ρ d z 2\pi\int_{\mathbb{R}}\int_0^\infty\lvert u_m\rvert^2\,d\rho\,dz 2 π ∫ R ∫ 0 ∞ ∣ u m ∣ 2 d ρ d z , the unweighted integral of ∣ u m ∣ 2 \lvert u_m\rvert^2 ∣ u m ∣ 2 on the half-plane is finite.
Let χ ∈ C c ∞ ( ( 0 , ∞ ) × R ) \chi\in C_c^\infty((0,\infty)\times\mathbb{R}) χ ∈ C c ∞ (( 0 , ∞ ) × R ) be real. Testing H m u m = 0 H_m u_m=0 H m u m = 0 against χ σ 2 u m \chi\sigma_2u_m χ σ 2 u m and taking twice the real part gives
0 = 2 m ∫ R ∫ 0 ∞ χ ∣ u m ∣ 2 d ρ d z + ∫ R ∫ 0 ∞ ρ u m ∗ ( ( ∂ ρ χ ) σ 3 − ( ∂ z χ ) σ 1 ) u m d ρ d z . 0=2m\int_{\mathbb{R}}\int_0^\infty\chi\lvert u_m\rvert^2\,d\rho\,dz+\int_{\mathbb{R}}\int_0^\infty\rho\,u_m^*\bigl((\partial_\rho\chi)\sigma_3-(\partial_z\chi)\sigma_1\bigr)u_m\,d\rho\,dz. 0 = 2 m ∫ R ∫ 0 ∞ χ ∣ u m ∣ 2 d ρ d z + ∫ R ∫ 0 ∞ ρ u m ∗ ( ( ∂ ρ χ ) σ 3 − ( ∂ z χ ) σ 1 ) u m d ρ d z .
Here u m ∗ u_m^* u m ∗ denotes the conjugate transpose. To see the cancellation, all terms of H m H_m H m except ( m / ρ ) σ 2 (m/\rho)\sigma_2 ( m / ρ ) σ 2 anticommute with σ 2 \sigma_2 σ 2 . The surviving multiplication term is 2 m χ ∣ u m ∣ 2 / ρ 2m\chi\lvert u_m\rvert^2/\rho 2 m χ ∣ u m ∣ 2 / ρ ; differentiating χ \chi χ gives the second integral. The test is legitimate by local H 1 H^1 H 1 approximation away from ρ = 0 \rho=0 ρ = 0 , and no derivative of the bounded coefficients of B B B is used.
Choose 0 ≤ χ = χ ε , R ≤ 1 0\leq\chi=\chi_{\varepsilon,R}\leq1 0 ≤ χ = χ ε , R ≤ 1 which vanishes for ρ < ε \rho<\varepsilon ρ < ε and for ∣ x ∣ > 2 R \lvert x\rvert>2R ∣ x ∣ > 2 R , and equals one when ρ > 2 ε \rho>2\varepsilon ρ > 2 ε and ∣ x ∣ < R \lvert x\rvert<R ∣ x ∣ < R . Choose it as a product of an axis cutoff and a spatial cutoff. The derivative of the first factor is bounded by C / ρ C/\rho C / ρ on { ε < ρ < 2 ε } \{\varepsilon<\rho<2\varepsilon\} { ε < ρ < 2 ε } , and that of the second by C / R C/R C / R on the outer annulus. The absolute value of the second integral is then at most
C ∫ { ρ < 2 ε } ∣ Ψ m ∣ 2 ρ d x + C R ∥ Ψ m ∥ 2 2 . C\int_{\{\rho<2\varepsilon\}}\frac{\lvert\Psi_m\rvert^2}{\rho}\,dx+\frac{C}{R}\lVert\Psi_m\rVert_2^2. C ∫ { ρ < 2 ε } ρ ∣ Ψ m ∣ 2 d x + R C ∥ Ψ m ∥ 2 2 .
The first term tends to zero as ε → 0 \varepsilon\to0 ε → 0 by the weighted integrability proved above; the second tends to zero as R → ∞ R \to\infty R → ∞ . Dominated convergence in the first integral yields
0 = 2 m ∫ R ∞ ∫ 0 ∞ ∣ u m ∣ 2 d ρ d z . 0 = 2m \int_{R}^{\infty}\int_{0}^{\infty} |u_m|^2 \,d\rho\,dz. 0 = 2 m ∫ R ∞ ∫ 0 ∞ ∣ u m ∣ 2 d ρ d z .
There is no zero angular frequency, since m ∈ Z + 1 2 m \in\mathbb{Z} + \frac{1}{2} m ∈ Z + 2 1 . Thus every mode vanishes, and completeness of the angular Fourier decomposition gives Ψ = 0 \Psi= 0 Ψ = 0 . □
The absence of an azimuthal component is essential to this argument: a term b θ e θ b_\theta e_\theta b θ e θ would replace m / ρ m/\rho m / ρ by m / ρ − c b θ m/\rho- cb_\theta m / ρ − c b θ , whose sign need not be fixed; the Hopf contact geometry of the Loss–Yau potential [18 ] shows that zero modes do occur in that case.
Proof of Theorem 1.3 . Part (a) is Proposition 5.3 . For part (b) we first treat A ∗ A_* A ∗ and U ∗ U_* U ∗ themselves.
Step 3: boundedness of the extremals and the conformal pullback. Define
h ( q ) : = q − 2 ψ ∗ ( arcosh ( 1 / q ) ) , 0 < q ≤ 1. h(q) := q^{-2}\psi_*(\operatorname{arcosh}(1/q)), \qquad0 < q \le1. h ( q ) := q − 2 ψ ∗ ( arcosh ( 1/ q )) , 0 < q ≤ 1.
Since q = sech t ∼ 2 e − t q = \operatorname{sech} t \sim2e^{-t} q = sech t ∼ 2 e − t as t → ∞ t \to\infty t → ∞ , the endpoint estimates give h ( 0 ) : = C ∗ / 4 h(0) := C_*/4 h ( 0 ) := C ∗ /4 and h ( 1 ) = a ∗ h(1) = a_* h ( 1 ) = a ∗ . Thus h ∈ C ( [ 0 , 1 ] ) h \in C([0,1]) h ∈ C ([ 0 , 1 ]) , and the preceding endpoint analysis also gives G ∗ ∈ C ( [ 0 , 1 ] ) G_* \in C([0,1]) G ∗ ∈ C ([ 0 , 1 ]) . Recalling s = 1 + ∣ x ∣ 2 s = 1 + |x|^2 s = 1 + ∣ x ∣ 2 , η = 2 ρ / s \eta= 2\rho/s η = 2 ρ / s and ∣ N ∣ ≤ s |N| \le s ∣ N ∣ ≤ s , we have
∣ A ∗ ∣ = 4 ρ s 2 ∣ h ( η ) ∣ , ∣ U ∗ ∣ ≤ ∥ G ∗ ∥ ∞ s . |A_*| = \frac{4\rho}{s^2}|h(\eta)|, \qquad|U_*| \le\frac{\|G_*\|_\infty}{s}. ∣ A ∗ ∣ = s 2 4 ρ ∣ h ( η ) ∣ , ∣ U ∗ ∣ ≤ s ∥ G ∗ ∥ ∞ .
Both potentials are therefore bounded, which proves the operator-domain assertion in the theorem. Since U ∗ U_* U ∗ is axisymmetric and poloidal, Proposition 5.3 already excludes zero modes for c U ∗ cU_* c U ∗ .
For A ∗ A_* A ∗ , consider the Möbius map
M ( x ) = ( 2 z , 1 − ∣ x ∣ 2 , 2 x 1 ) ∣ x + e 2 ∣ 2 , λ ( x ) = 2 ∣ x + e 2 ∣ 2 , e 2 = ( 0 , 1 , 0 ) . M(x) = \frac{(2z,\,1-|x|^2,\,2x_1)}{|x+e_2|^2}, \qquad\lambda(x) = \frac{2}{|x+e_2|^2}, \qquad e_2 = (0,1,0). M ( x ) = ∣ x + e 2 ∣ 2 ( 2 z , 1 − ∣ x ∣ 2 , 2 x 1 ) , λ ( x ) = ∣ x + e 2 ∣ 2 2 , e 2 = ( 0 , 1 , 0 ) .
The identity
M ( x ) + e 2 = 2 ( z , x 2 + 1 , x 1 ) ∣ x + e 2 ∣ 2 M(x) + e_2 = \frac{2(z,\,x_2+1,\,x_1)}{|x+e_2|^2} M ( x ) + e 2 = ∣ x + e 2 ∣ 2 2 ( z , x 2 + 1 , x 1 )
expresses it as an inversion about − e 2 -e_2 − e 2 , followed by a reflection fixing − e 2 -e_2 − e 2 and a dilation about the same point. The two orientation reversals cancel. Directly,
M ∘ M = id , D M ( x ) T D M ( x ) = λ ( x ) 2 Id , det D M ( x ) = λ ( x ) 3 > 0 M \circ M = \operatorname{id}, \qquad DM(x)^{\mathsf T}DM(x) = \lambda(x)^2 \operatorname{Id}, \qquad\det DM(x) = \lambda(x)^3 > 0 M ∘ M = id , D M ( x ) T D M ( x ) = λ ( x ) 2 Id , det D M ( x ) = λ ( x ) 3 > 0
away from − e 2 -e_2 − e 2 .
A magnetic potential transforms as a one-form. Its pullback is therefore B M ( x ) = D M ( x ) T A ∗ ( M ( x ) ) B_M(x) = DM(x)^{\mathsf T}A_*(M(x)) B M ( x ) = D M ( x ) T A ∗ ( M ( x )) , rather than just A ∗ ( M ( x ) ) A_*(M(x)) A ∗ ( M ( x )) . The calculation can be checked from
s ( M ( x ) ) = λ ( x ) s ( x ) , η ( M ( x ) ) = 1 − η ( x ) 2 , s(M(x)) = \lambda(x)s(x), \qquad\eta(M(x)) = \sqrt{1-\eta(x)^2}, s ( M ( x )) = λ ( x ) s ( x ) , η ( M ( x )) = 1 − η ( x ) 2 ,
and
D M ( x ) T ( − M 2 ( x ) M 1 ( x ) 0 ) = − λ ( x ) 2 2 N ( x ) . DM(x)^{\mathsf T}
\begin{pmatrix}
-M_2(x)\\
M_1(x)\\
0
\end{pmatrix}
= -\frac{\lambda(x)^2}{2}N(x). D M ( x ) T − M 2 ( x ) M 1 ( x ) 0 = − 2 λ ( x ) 2 N ( x ) .
Indeed, A ∗ ( y ) = 4 h ( η ( y ) ) s ( y ) − 2 ( − y 2 , y 1 , 0 ) A_*(y) = 4h(\eta(y))s(y)^{-2}(-y_2,y_1,0) A ∗ ( y ) = 4 h ( η ( y )) s ( y ) − 2 ( − y 2 , y 1 , 0 ) , so
B M ( x ) = − 2 s ( x ) 2 h ( 1 − η ( x ) 2 ) N ( x ) , ∣ B M ( x ) ∣ ≤ 2 ∥ h ∥ ∞ s ( x ) . B_M(x) = -\frac{2}{s(x)^2}h\left(\sqrt{1-\eta(x)^2}\right)N(x), \qquad|B_M(x)| \le\frac{2\|h\|_\infty}{s(x)}. B M ( x ) = − s ( x ) 2 2 h ( 1 − η ( x ) 2 ) N ( x ) , ∣ B M ( x ) ∣ ≤ s ( x ) 2∥ h ∥ ∞ .
This formula defines a bounded axisymmetric poloidal field on all of R 3 \mathbb{R}^{3} R 3 . In particular, it extends across { ρ = 1 , z = 0 } \{\rho= 1, z = 0\} { ρ = 1 , z = 0 } , where N = 0 N = 0 N = 0 , using the continuous extension of h h h to zero.
Step 4: the transformed spinor remains a global zero mode. Suppose that Ψ ∈ L 2 \Psi\in L^{2} Ψ ∈ L 2 is a zero mode for c A ∗ cA_{*} c A ∗ . Boundedness of A ∗ A_{*} A ∗ gives Ψ ∈ H 1 \Psi\in H^{1} Ψ ∈ H 1 by the argument of Step 1. Conformal covariance of the magnetic Dirac operator [7 Section 4] , based on the classical spinorial transformation law [14 Section 1.4] , [13 ] and recalled for zero modes in [11 Remark 3] , gives
Φ ( x ) = λ ( x ) β ( x ) Ψ ( M ( x ) ) , D c B M Φ = 0 on R 3 ∖ { − e 2 } . \Phi(x) = \lambda(x)\beta(x)\Psi(M(x)), \qquad D_{cB_M}\Phi= 0 \quad\text{on } \mathbb{R}^{3}\setminus\{-e_{2}\}. Φ ( x ) = λ ( x ) β ( x ) Ψ ( M ( x )) , D c B M Φ = 0 on R 3 ∖ { − e 2 } .
Here β ( x ) \beta(x) β ( x ) is the unitary spinor identification associated with the orientation-preserving conformal map M M M . The factor λ \lambda λ is the spinor conformal weight in dimension three. On compact subsets away from − e 2 -e_{2} − e 2 the transformation preserves local H 1 H^{1} H 1 regularity, and the covariance identity extends by local H 1 H^{1} H 1 approximation. It requires no differentiability of the bounded potentials.
Conformal covariance alone does not guarantee that Φ ∈ L 2 \Phi\in L^{2} Φ ∈ L 2 . For this we use M ∘ M = i d M \circ M = \mathrm{id} M ∘ M = id , λ ( M ( x ) ) λ ( x ) = 1 \lambda(M(x))\lambda(x) = 1 λ ( M ( x )) λ ( x ) = 1 and the Jacobian computed above. Changing variables y = M ( x ) y = M(x) y = M ( x ) gives
∫ R 3 ∣ Φ ( x ) ∣ 2 d x = ∫ R 3 λ ( y ) ∣ Ψ ( y ) ∣ 2 d y = 2 ∫ R 3 ∣ Ψ ( y ) ∣ 2 ∣ y + e 2 ∣ d y ≤ 8 ∫ R 3 ∣ ∇ Ψ ( y ) ∣ 2 d y < ∞ . \begin{aligned}
\int_{\mathbb{R}^{3}} \lvert\Phi(x)\rvert^{2}\,\mathrm{d}x
&= \int_{\mathbb{R}^{3}} \lambda(y)\lvert\Psi(y)\rvert^{2}\,\mathrm{d}y \\
&= 2\int_{\mathbb{R}^{3}} \frac{\lvert\Psi(y)\rvert^{2}}{\lvert y+e_{2}\rvert}\,\mathrm{d}y
\le8\int_{\mathbb{R}^{3}} \lvert\nabla\Psi(y)\rvert^{2}\,\mathrm{d}y < \infty.
\end{aligned} ∫ R 3 ∣ Φ ( x ) ∣ 2 d x = ∫ R 3 λ ( y ) ∣ Ψ ( y ) ∣ 2 d y = 2 ∫ R 3 ∣ y + e 2 ∣ ∣ Ψ ( y ) ∣ 2 d y ≤ 8 ∫ R 3 ∣ ∇Ψ ( y ) ∣ 2 d y < ∞.
The final step is the three-dimensional Hardy inequality, applied componentwise.
It remains to remove the exceptional point from the weak equation. Let χ ε \chi_{\varepsilon} χ ε vanish on B ε ( − e 2 ) B_{\varepsilon}(-e_{2}) B ε ( − e 2 ) and equal one outside B 2 ε ( − e 2 ) B_{2\varepsilon}(-e_{2}) B 2 ε ( − e 2 ) , with ∣ ∇ χ ε ∣ ≤ C / ε \lvert\nabla\chi_{\varepsilon}\rvert\le C/\varepsilon ∣ ∇ χ ε ∣ ≤ C / ε . Then ∥ ∇ χ ε ∥ 2 = O ( ε 1 / 2 ) \lVert\nabla\chi_{\varepsilon}\rVert_{2} = O(\varepsilon^{1/2}) ∥ ∇ χ ε ∥ 2 = O ( ε 1/2 ) . For a compactly supported smooth test spinor ζ \zeta ζ , use χ ε ζ \chi_{\varepsilon}\zeta χ ε ζ in the equation away from − e 2 -e_{2} − e 2 . The additional term containing the cutoff derivative is bounded by
C ∥ ζ ∥ ∞ ∥ Φ ∥ L 2 ( B 2 ε ( − e 2 ) ) ∥ ∇ χ ε ∥ 2 , C\lVert\zeta\rVert_{\infty}\lVert\Phi\rVert_{L^{2}(B_{2\varepsilon}(-e_{2}))}\lVert\nabla\chi_{\varepsilon}\rVert_{2}, C ∥ ζ ∥ ∞ ∥ Φ ∥ L 2 ( B 2 ε ( − e 2 )) ∥ ∇ χ ε ∥ 2 ,
and tends to zero. The other terms converge by local integrability, since Φ ∈ L 2 \Phi\in L^{2} Φ ∈ L 2 and B M B_{M} B M is bounded. Thus D c B M Φ = 0 D_{cB_M}\Phi= 0 D c B M Φ = 0 on all of R 3 \mathbb{R}^{3} R 3 in distributions. The argument of Step 1 now gives Φ ∈ H 1 \Phi\in H^{1} Φ ∈ H 1 , and Proposition 5.3 yields Φ = 0 \Phi= 0 Φ = 0 . As λ > 0 \lambda> 0 λ > 0 , β \beta β is unitary and M M M is invertible away from one point, it follows that Ψ = 0 \Psi= 0 Ψ = 0 .
Finally, the fields T λ , z 0 A ∗ T_{\lambda,z_{0}}A_{*} T λ , z 0 A ∗ and T λ , z 0 U ∗ T_{\lambda,z_{0}}U_{*} T λ , z 0 U ∗ are obtained by axial translations and the dilations A ↦ λ A ( λ ⋅ ) A \mapsto\lambda A(\lambda\cdot) A ↦ λ A ( λ ⋅ ) , under which Ψ ↦ λ 3 / 2 Ψ ( λ ⋅ ) \Psi\mapsto\lambda^{3/2}\Psi(\lambda\cdot) Ψ ↦ λ 3/2 Ψ ( λ ⋅ ) maps zero modes to zero modes. Hence the conclusion holds for every centre. □
A gauge-canonical inverse
We prepare the proof of Theorem 1.1 by fixing the inverse map and establishing its descent and equality properties.
Away from solutions the symbol D − 1 \mathcal{D}^{-1} D − 1 is ambiguous, because a curl primitive is determined only modulo W \mathcal{W} W . We remove the ambiguity by imposing the nonlinear gauge after each inverse step. Throughout this section u ∈ D 1 , 3 / 2 ( curl ; R 3 ) u \in\mathcal{D}^{1,3/2}(\operatorname{curl};\mathbb{R}^{3}) u ∈ D 1 , 3/2 ( curl ; R 3 ) . Put
Γ ( u ) : = u + w ( u ) , \Gamma(u) := u + w(u), Γ ( u ) := u + w ( u ) ,
where w ( u ) w(u) w ( u ) is the unique correction from (9) . Thus Γ ( u ) ∈ M \Gamma(u) \in\mathcal{M} Γ ( u ) ∈ M and
div ( ∣ Γ ( u ) ∣ Γ ( u ) ) = 0. (46) \operatorname{div}(\lvert\Gamma(u)\rvert\Gamma(u)) = 0. \tag*{(46)} div (∣ Γ ( u )∣ Γ ( u )) = 0. (46)
Uniqueness in the homogeneous L 3 L^3 L 3 minimization also gives
Γ ( c u ) = c Γ ( u ) ( c ∈ R ) . (47) \Gamma(cu)=c\Gamma(u)\qquad(c\in\mathbb{R}). \tag*{(47)} Γ ( c u ) = c Γ ( u ) ( c ∈ R ) . (47)
For a divergence-free F ∈ L 3 / 2 ( R 3 , R 3 ) F\in L^{3/2}(\mathbb{R}^3,\mathbb{R}^3) F ∈ L 3/2 ( R 3 , R 3 ) , let
K F : = ∇ × ( − Δ ) − 1 F . (48) \mathcal{K}F:=\nabla\times(-\Delta)^{-1}F. \tag*{(48)} K F := ∇ × ( − Δ ) − 1 F . (48)
The Hardy–Littlewood–Sobolev inequality gives K F ∈ L 3 \mathcal{K}F\in L^3 K F ∈ L 3 , and ∇ × K F = F \nabla\times\mathcal{K}F=F ∇ × K F = F . Define the gauge-canonical inverse map by
I ( u ) : = Γ ( K ( ∣ Γ ( u ) ∣ Γ ( u ) ) ) . (49) I(u):=\Gamma\bigl(\mathcal{K}(|\Gamma(u)|\Gamma(u))\bigr). \tag*{(49)} I ( u ) := Γ ( K ( ∣Γ ( u ) ∣Γ ( u )) ) . (49)
The source ∣ Γ ( u ) ∣ Γ ( u ) |\Gamma(u)|\Gamma(u) ∣Γ ( u ) ∣Γ ( u ) belongs to L 3 / 2 L^{3/2} L 3/2 and is divergence-free by (46) , so I ( u ) ∈ D 1 , 3 / 2 ( curl ; R 3 ) I(u)\in D^{1,3/2}(\operatorname{curl};\mathbb{R}^3) I ( u ) ∈ D 1 , 3/2 ( curl ; R 3 ) . It depends only on the class of u u u modulo W \mathcal{W} W .
Lemma 6.1 (The canonical inverse and descent of the quotient). For u ∈ D 1 , 3 / 2 ( curl ; R 3 ) ∖ W u\in D^{1,3/2}(\operatorname{curl};\mathbb{R}^3)\setminus\mathcal{W} u ∈ D 1 , 3/2 ( curl ; R 3 ) ∖ W , the field I ( u ) I(u) I ( u ) defined in (49) is the unique element of M \mathcal{M} M satisfying
∇ × I ( u ) = ∣ Γ ( u ) ∣ Γ ( u ) . (50) \nabla\times I(u)=|\Gamma(u)|\Gamma(u). \tag*{(50)} ∇ × I ( u ) = ∣Γ ( u ) ∣Γ ( u ) . (50) Consequently,
D ( I ( u ) ) = Γ ( u ) , I ( c u ) = c ∣ c ∣ I ( u ) ( c ∈ R ) . (51) \mathcal{D}(I(u))=\Gamma(u),\qquad I(cu)=c|c|\,I(u)\qquad(c\in\mathbb{R}). \tag*{(51)} D ( I ( u )) = Γ ( u ) , I ( c u ) = c ∣ c ∣ I ( u ) ( c ∈ R ) . (51) Moreover,
Q ( I ( u ) ) ≤ Q ( u ) . (52) Q(I(u))\leq Q(u). \tag*{(52)} Q ( I ( u )) ≤ Q ( u ) . (52) and the decrease in one step is given by an exact formula. Define
E ( u ) : = ∫ R 3 ∣ ∇ × u ∣ 3 / 2 d x , M ( u ) : = ∫ R 3 ∣ Γ ( u ) ∣ 3 d x , v : = I ( u ) , (53) E(u):=\int_{\mathbb{R}^3}|\nabla\times u|^{3/2}\,dx,\qquad M(u):=\int_{\mathbb{R}^3}|\Gamma(u)|^3\,dx,\qquad v:=I(u), \tag*{(53)} E ( u ) := ∫ R 3 ∣∇ × u ∣ 3/2 d x , M ( u ) := ∫ R 3 ∣Γ ( u ) ∣ 3 d x , v := I ( u ) , (53) and
a ( u ) : = M ( u ) E ( u ) 2 / 3 M ( v ) 1 / 3 ∈ ( 0 , 1 ] . a(u):=\frac{M(u)}{E(u)^{2/3}M(v)^{1/3}}\in(0,1]. a ( u ) := E ( u ) 2/3 M ( v ) 1/3 M ( u ) ∈ ( 0 , 1 ] . Then
Q ( I ( u ) ) Q ( u ) = a ( u ) 3 / 2 . (54) \frac{Q(I(u))}{Q(u)}=a(u)^{3/2}. \tag*{(54)} Q ( u ) Q ( I ( u )) = a ( u ) 3/2 . (54) Equality in (52) holds if and only if, for some λ > 0 \lambda>0 λ > 0 ,
∇ × u = λ ∣ I ( u ) ∣ I ( u ) , I 2 ( u ) = λ − 1 Γ ( u ) . (55) \nabla\times u=\lambda|I(u)|I(u),\qquad I^2(u)=\lambda^{-1}\Gamma(u). \tag*{(55)} ∇ × u = λ ∣ I ( u ) ∣ I ( u ) , I 2 ( u ) = λ − 1 Γ ( u ) . (55) Equality thus holds exactly on the projective two-cycles of I I I : two inverse steps return the gauge-minimal field up to a positive factor. Finally, I I I interchanges the two cylindrical classes:
I ( T ) ⊂ S , I ( S ) ⊂ T . (56) I(\mathcal{T})\subset\mathcal{S},\qquad I(\mathcal{S})\subset\mathcal{T}. \tag*{(56)} I ( T ) ⊂ S , I ( S ) ⊂ T . (56) Proof. Equation (46) makes ∣ Γ ( u ) ∣ Γ ( u ) |\Gamma(u)|\Gamma(u) ∣Γ ( u ) ∣Γ ( u ) divergence-free, so (50) follows from (48) ; applying Γ \Gamma Γ does not change curl. If two elements of M \mathcal{M} M have this curl, their difference lies in W \mathcal{W} W , and both satisfy the Euler equation for the strictly convex L 3 L^3 L 3 -minimization in that affine class. Hence they coincide. The two identities in (51) follow from the definition, uniqueness, and (47) .
With the notation in (53) , one has Γ ( v ) = v \Gamma(v)=v Γ ( v ) = v and
E ( v ) = M ( u ) . E(v)=M(u). E ( v ) = M ( u ) .
Integration by parts and Hölder’s inequality give
M ( u ) = ∫ R 3 Γ ( u ) ⋅ ∇ × v d x = ∫ R 3 ∇ × u ⋅ v d x ≤ E ( u ) 2 / 3 M ( v ) 1 / 3 . (57) M(u)=\int_{\mathbb{R}^{3}}\Gamma(u)\cdot\nabla\times v\,dx=\int_{\mathbb{R}^{3}}\nabla\times u\cdot v\,dx
\le E(u)^{2/3}M(v)^{1/3}. \tag*{(57)} M ( u ) = ∫ R 3 Γ ( u ) ⋅ ∇ × v d x = ∫ R 3 ∇ × u ⋅ v d x ≤ E ( u ) 2/3 M ( v ) 1/3 . (57)
The integration-by-parts identity is first proved for smooth compactly supported fields and then extended by density in the natural L 3 L^{3} L 3 –L 3 / 2 L^{3/2} L 3/2 pairing. After raising (57) to the power 3 / 2 3/2 3/2 and rearranging, we obtain
Q ( v ) = M ( u ) M ( v ) 1 / 2 ≤ E ( u ) M ( u ) 1 / 2 = Q ( u ) . Q(v)=\frac{M(u)}{M(v)^{1/2}}\le\frac{E(u)}{M(u)^{1/2}}=Q(u). Q ( v ) = M ( v ) 1/2 M ( u ) ≤ M ( u ) 1/2 E ( u ) = Q ( u ) .
Division of the two endpoint expressions gives
Q ( v ) Q ( u ) = M ( u ) 3 / 2 E ( u ) M ( v ) 1 / 2 = α ( u ) 3 / 2 , \frac{Q(v)}{Q(u)}=\frac{M(u)^{3/2}}{E(u)M(v)^{1/2}}=\alpha(u)^{3/2}, Q ( u ) Q ( v ) = E ( u ) M ( v ) 1/2 M ( u ) 3/2 = α ( u ) 3/2 ,
which proves (54) . To identify the equality case, set
x : = v M ( v ) 1 / 3 , y : = D ( u ) E ( u ) 1 / 3 . x:=\frac{v}{M(v)^{1/3}},\qquad y:=\frac{\mathcal{D}(u)}{E(u)^{1/3}}. x := M ( v ) 1/3 v , y := E ( u ) 1/3 D ( u ) .
Then ∥ x ∥ L 3 = ∥ y ∥ L 3 = 1 \lVert x\rVert_{L^{3}}=\lVert y\rVert_{L^{3}}=1 ∥ x ∥ L 3 = ∥ y ∥ L 3 = 1 and
α ( u ) = ∫ R 3 x ⋅ ∣ y ∣ y d x . \alpha(u)=\int_{\mathbb{R}^{3}}x\cdot\lvert y\rvert y\,dx. α ( u ) = ∫ R 3 x ⋅ ∣ y ∣ y d x .
Equality in Hölder’s inequality is therefore equivalent to x = y x=y x = y , or, equivalently, to the first relation in (55) . If that relation holds, then λ − 1 Γ ( u ) \lambda^{-1}\Gamma(u) λ − 1 Γ ( u ) is the unique nonlinear-gauge primitive of ∣ v ∣ v \lvert v\rvert v ∣ v ∣ v , and hence equals I ( v ) I(v) I ( v ) . Conversely, the second relation in (55) implies the first one after taking curl. This proves the equality characterization.
Finally, K \mathcal{K} K interchanges toroidal and poloidal fields. To see that the gauge projection preserves each class, let R = diag ( 1 , − 1 , 1 ) R=\operatorname{diag}(1,-1,1) R = diag ( 1 , − 1 , 1 ) act by ( R ⋅ u ) ( x ) = R u ( R x ) (R\cdot u)(x)=Ru(Rx) ( R ⋅ u ) ( x ) = R u ( R x ) . Within the S O ( 2 ) SO(2) S O ( 2 ) -equivariant space, poloidal fields are fixed by this reflection and toroidal fields are anti-invariant. Uniqueness of the gauge minimizer shows that Γ \Gamma Γ commutes with this action and with rotations; it also commutes with negation by (47) . Thus Γ \Gamma Γ preserves both classes. The operator K \mathcal{K} K reverses reflection parity, by the transformation rule for curl under orientation-reversing orthogonal maps. This proves (56) . □
Proof of Theorem 1.1
Sharp constants, classification and the first inverse step
Proof of Theorem 1.1 (a) and (b). Reduction and duality. The positivity of the global constant and its comparison with the restricted constants follow from (13) . By Lemma 2.4, every admissible toroidal field A ( ψ ) A(\psi) A ( ψ ) is already in its L 3 L^{3} L 3 -minimal gauge. Proposition 4.1 therefore gives
Q ( A ( ψ ) ) = 2 π ∫ H 2 ∣ ∇ g ψ ∣ g 3 / 2 d V g ( ∫ H 2 ∣ ψ ∣ 3 d V g ) 1 / 2 ( ψ ≠ 0 ) . Q(A(\psi))=\sqrt{2\pi}\frac{\int_{\mathbb{H}^{2}}\lvert\nabla_{g}\psi\rvert_{g}^{3/2}\,dV_{g}}{\left(\int_{\mathbb{H}^{2}}\lvert\psi\rvert^{3}\,dV_{g}\right)^{1/2}}\qquad(\psi\ne0). Q ( A ( ψ )) = 2 π ( ∫ H 2 ∣ ψ ∣ 3 d V g ) 1/2 ∫ H 2 ∣ ∇ g ψ ∣ g 3/2 d V g ( ψ = 0 ) .
Taking the infimum over this bijection, and using (19) at p = p ^ = 3 / 2 p=\widehat{p}=3/2 p = p = 3/2 , we obtain
S 3 / 2 , c u r l T = S 3 / 2 , c u r l S = 2 π Λ H 2 . (58) S^{T}_{3/2,\mathrm{curl}}=S^{S}_{3/2,\mathrm{curl}}=\sqrt{2\pi}\,\Lambda_{\mathbb{H}^{2}}. \tag*{(58)} S 3/2 , curl T = S 3/2 , curl S = 2 π Λ H 2 . (58)
The hyperbolic infimum is attained, as recalled in § 3 . Choose a positive minimizer ψ ∗ \psi_* ψ ∗ normalized by − Δ 3 / 2 , g ψ ∗ = ψ ∗ 2 -\Delta_{3/2,g}\psi_*=\psi_*^2 − Δ 3/2 , g ψ ∗ = ψ ∗ 2 . The correspondence above makes A ∗ = ρ − 1 ψ ∗ e θ A_*=\rho^{-1}\psi_*e_\theta A ∗ = ρ − 1 ψ ∗ e θ a toroidal extremal, and Proposition 4.2 shows that it solves the full equation (2) , including across the symmetry axis. Theorems 2.1 and 2.2 make U ∗ = D ( A ∗ ) U_*=\mathcal{D}(A_*) U ∗ = D ( A ∗ ) a poloidal extremal solving the same equation, and give D ( U ∗ ) = A ∗ \mathcal{D}(U_*)=A_* D ( U ∗ ) = A ∗ . These conclusions hold for every positive hyperbolic minimizer, independently of its centre.
Classification. Let A = A ( ψ ) ∈ T A=A(\psi)\in\mathcal{T} A = A ( ψ ) ∈ T be an extremal of S 3 / 2 , curl T S^{\mathcal{T}}_{3/2,\operatorname{curl}} S 3/2 , curl T . By the correspondence, ψ \psi ψ minimizes Λ H 2 \Lambda_{\mathbb{H}^2} Λ H 2 , hence has one sign (§ 3 ), and after multiplication by a nonzero constant it is a positive solution of (4) . By Theorem 3.2 this solution is ψ ∗ ∘ F ρ 0 , z 0 − 1 \psi_*\circ F^{-1}_{\rho_0,z_0} ψ ∗ ∘ F ρ 0 , z 0 − 1 for some ρ 0 > 0 \rho_0>0 ρ 0 > 0 , z 0 ∈ R z_0\in\mathbb{R} z 0 ∈ R , so that A = c T 1 / ρ 0 , z 0 A ∗ A=cT_{1/\rho_0,z_0}A_* A = c T 1/ ρ 0 , z 0 A ∗ with c ≠ 0 c\ne0 c = 0 ; conversely every such field is an extremal, since Q Q Q is invariant under scaling and under the maps (5) . Now let U ∈ S U\in\mathcal{S} U ∈ S be an extremal of S 3 / 2 , curl S S^{\mathcal{S}}_{3/2,\operatorname{curl}} S 3/2 , curl S . Choose a > 0 a>0 a > 0 so that U ′ : = a Γ ( U ) U':=a\Gamma(U) U ′ := a Γ ( U ) solves (2) ; this is possible by [22 Theorem 1.4] . By Theorems 2.1 and 2.2 , D ( U ′ ) \mathcal{D}(U') D ( U ′ ) is a toroidal extremal solving (2) , hence of the form c T λ , z 0 A ∗ cT_{\lambda,z_0}A_* c T λ , z 0 A ∗ by the first part; since the maps (5) preserve solutions and c A ∗ cA_* c A ∗ solves (2) only for ∣ c ∣ = 1 |c|=1 ∣ c ∣ = 1 , we get D ( U ′ ) = ± T λ , z 0 A ∗ \mathcal{D}(U')=\pm T_{\lambda,z_0}A_* D ( U ′ ) = ± T λ , z 0 A ∗ and therefore U ′ = D ( D ( U ′ ) ) = ± T λ , z 0 U ∗ U'=\mathcal{D}(\mathcal{D}(U'))=\pm T_{\lambda,z_0}U_* U ′ = D ( D ( U ′ )) = ± T λ , z 0 U ∗ , because D \mathcal{D} D is odd and commutes with the maps (5) . Thus U = c T λ , z 0 U ∗ + w U=cT_{\lambda,z_0}U_*+w U = c T λ , z 0 U ∗ + w with c ≠ 0 c\ne0 c = 0 and w w w curl-free, which proves part (b). □
Radial quadrature for the inverse map. To evaluate the first inverse step explicitly and to prepare the convergence argument, we derive the radial formulas. Let ψ ≥ 0 \psi\ge0 ψ ≥ 0 be a nonzero profile in D r a d , O 1 , 3 / 2 ( H 2 ) D^{1,3/2}_{\mathrm{rad},O}(\mathbb{H}^2) D rad , O 1 , 3/2 ( H 2 ) . Since ψ 2 ∈ L 3 / 2 \psi^2\in L^{3/2} ψ 2 ∈ L 3/2 and D 1 , 3 / 2 ( H 2 ) ↪ L 3 D^{1,3/2}(\mathbb{H}^2)\hookrightarrow L^3 D 1 , 3/2 ( H 2 ) ↪ L 3 , strict monotonicity and coercivity give a unique weak solution χ ∈ D 1 , 3 / 2 ( H 2 ) \chi\in D^{1,3/2}(\mathbb{H}^2) χ ∈ D 1 , 3/2 ( H 2 ) of
− Δ 3 / 2 , g χ = ψ 2 . -\Delta_{3/2,g}\chi=\psi^2. − Δ 3/2 , g χ = ψ 2 .
Testing with the negative part gives χ ≥ 0 \chi\ge0 χ ≥ 0 , and uniqueness makes χ \chi χ radial about the same centre. Its radial flux F ( t ) = sinh t ∣ χ ′ ∣ − 1 / 2 χ ′ F(t)=\sinh t|\chi'|^{-1/2}\chi' F ( t ) = sinh t ∣ χ ′ ∣ − 1/2 χ ′ satisfies F ′ = − ψ 2 sinh t F'=-\psi^2\sinh t F ′ = − ψ 2 sinh t . Since the right-hand side is integrable near zero, F F F has a finite limit there. This integration constant is zero: a nonzero flux limit would make ∫ 0 1 ∣ F ( t ) ∣ 3 / sinh 2 t d t \int_0^1 |F(t)|^3/\sinh^2t\,dt ∫ 0 1 ∣ F ( t ) ∣ 3 / sinh 2 t d t infinite, contrary to finite energy. Consequently,
sinh t − χ ′ ( t ) = J ψ ( t ) , J ψ ( t ) : = ∫ 0 t ψ ( s ) 2 sinh s d s . \sinh t\sqrt{-\chi'(t)}=\mathcal{J}_\psi(t),\qquad\mathcal{J}_\psi(t):=\int_0^t\frac{\psi(s)^2}{\sinh s}\,ds. sinh t − χ ′ ( t ) = J ψ ( t ) , J ψ ( t ) := ∫ 0 t sinh s ψ ( s ) 2 d s .
Thus χ \chi χ is nonincreasing. Since χ ∈ L 3 ( H 2 ) \chi\in L^3(\mathbb{H}^2) χ ∈ L 3 ( H 2 ) , its limit at infinity is zero, and a second integration gives
χ ( t ) = ( R ψ ) ( t ) : = ∫ t ∞ J ψ ( s ) 2 sinh 2 s d s , t > 0. (59) \chi(t)=(\mathcal{R}\psi)(t):=\int_t^\infty\frac{\mathcal{J}_\psi(s)^2}{\sinh^2s}\,ds,\qquad t>0. \tag*{(59)} χ ( t ) = ( R ψ ) ( t ) := ∫ t ∞ sinh 2 s J ψ ( s ) 2 d s , t > 0. (59)
The identities D ( I 2 ( A ( ψ ) ) ) = I ( A ( ψ ) ) \mathcal{D}(\mathcal{I}^2(A(\psi)))=\mathcal{I}(A(\psi)) D ( I 2 ( A ( ψ ))) = I ( A ( ψ )) and ∇ × I ( A ( ψ ) ) = ∣ A ( ψ ) ∣ A ( ψ ) \nabla\times\mathcal{I}(A(\psi))=|A(\psi)|A(\psi) ∇ × I ( A ( ψ )) = ∣ A ( ψ ) ∣ A ( ψ ) , together with Proposition 4.2 , show that the toroidal profile of I 2 ( A ( ψ ) ) \mathcal{I}^2(A(\psi)) I 2 ( A ( ψ )) solves the same weak equation as χ \chi χ . By uniqueness,
I 2 ( A ( ψ ) ) = A ( R ψ ) . \mathcal{I}^2(A(\psi))=A(\mathcal{R}\psi). I 2 ( A ( ψ )) = A ( R ψ ) .
Write, as radial integrals without the angular factor,
e ( ψ ) : = ∫ 0 ∞ ∣ ψ ′ ∣ 3 / 2 sinh t d t , m ( ψ ) : = ∫ 0 ∞ ψ 3 sinh t d t . e(\psi):=\int_0^\infty|\psi'|^{3/2}\sinh t\,dt,\qquad m(\psi):=\int_0^\infty\psi^3\sinh t\,dt. e ( ψ ) := ∫ 0 ∞ ∣ ψ ′ ∣ 3/2 sinh t d t , m ( ψ ) := ∫ 0 ∞ ψ 3 sinh t d t .
Differentiating (59) gives
e ( R ψ ) = ∫ 0 ∞ J ψ ( t ) 3 sinh 2 t d t . e(\mathcal{R}\psi)=\int_0^\infty\frac{\mathcal{J}_\psi(t)^3}{\sinh^2t}\,dt. e ( R ψ ) = ∫ 0 ∞ sinh 2 t J ψ ( t ) 3 d t .
The hyperbolic polar-coordinate formula and the norm identities (4.3)–(4.4) give the first formula below. The second follows from E ( I ( A ( ψ ) ) ) = M ( A ( ψ ) ) E(I(A(\psi)))=M(A(\psi)) E ( I ( A ( ψ ))) = M ( A ( ψ )) and M ( I ( A ( ψ ) ) ) = E ( A ( R ψ ) ) M(I(A(\psi)))=E(A(\mathcal{R}\psi)) M ( I ( A ( ψ ))) = E ( A ( R ψ )) , with E E E and M M M as in (53) :
Q ( A ( ψ ) ) = 2 π e ( ψ ) m ( ψ ) 1 / 2 , Q ( I ( A ( ψ ) ) ) = 2 π m ( ψ ) e ( R ψ ) 1 / 2 . (60) Q(A(\psi))=2\pi\frac{e(\psi)}{m(\psi)^{1/2}},\qquad Q(I(A(\psi)))=2\pi\frac{m(\psi)}{e(\mathcal{R}\psi)^{1/2}}. \tag*{(60)} Q ( A ( ψ )) = 2 π m ( ψ ) 1/2 e ( ψ ) , Q ( I ( A ( ψ ))) = 2 π e ( R ψ ) 1/2 m ( ψ ) . (60)
Proof of Corollary 1.2. The trial field of [3 Section 6] is
A T ( x ) = 3 s 2 ( − x 2 , x 1 , 0 ) = A ( ψ 0 ) , ψ 0 ( t ) : = 3 4 sech 2 t . A_{\mathcal{T}}(x)=\frac{3}{s^2}(-x_2,x_1,0)=A(\psi_0),\qquad\psi_0(t):=\frac{3}{4}\operatorname{sech}^2t. A T ( x ) = s 2 3 ( − x 2 , x 1 , 0 ) = A ( ψ 0 ) , ψ 0 ( t ) := 4 3 sech 2 t .
It belongs to T ∩ M \mathcal{T}\cap\mathcal{M} T ∩ M , and
Q ( A T ) = 8 10 7 π . Q(A_{\mathcal{T}})=\frac{8\sqrt{10}}{7}\pi. Q ( A T ) = 7 8 10 π .
For this profile,
m ( ψ 0 ) = 27 320 , J ψ 0 ( t ) = 3 16 ( 1 − sech 3 t ) . m(\psi_0)=\frac{27}{320},\qquad J_{\psi_0}(t)=\frac{3}{16}(1-\operatorname{sech}^3t). m ( ψ 0 ) = 320 27 , J ψ 0 ( t ) = 16 3 ( 1 − sech 3 t ) .
Since I ( A T ) = D ( A ( R ψ 0 ) ) I(A_{\mathcal{T}})=\mathcal{D}(A(\mathcal{R}\psi_0)) I ( A T ) = D ( A ( R ψ 0 )) , the calculation leading to (43) gives
I ( A T ) = J ψ 0 ( t ) 2 ρ 2 sinh 2 t N = 3 8 s 2 1 − η 3 1 − η 2 N . I(A_{\mathcal{T}})=\frac{J_{\psi_0}(t)}{2\rho^2\sinh^2t}N=\frac{3}{8s^2}\frac{1-\eta^3}{1-\eta^2}N. I ( A T ) = 2 ρ 2 sinh 2 t J ψ 0 ( t ) N = 8 s 2 3 1 − η 2 1 − η 3 N .
After cancellation, this becomes
I ( A T ) ( x ) = 3 8 s 2 1 + η + η 2 1 + η N ( x ) , I(A_{\mathcal{T}})(x)=\frac{3}{8s^2}\frac{1+\eta+\eta^2}{1+\eta}N(x), I ( A T ) ( x ) = 8 s 2 3 1 + η 1 + η + η 2 N ( x ) ,
with the quotient extended continuously at η = 1 \eta=1 η = 1 . The substitution η = sech t \eta=\operatorname{sech}t η = sech t gives
e ( R ψ 0 ) = 27 4096 ∫ 0 1 η ( 1 − η 3 ) 3 ( 1 − η 2 ) 3 / 2 d η = 27 4096 ( 891 π 256 − 53 5 ) . \begin{aligned}
e(\mathcal{R}\psi_0)=\frac{27}{4096}\int_0^1\frac{\eta(1-\eta^3)^3}{(1-\eta^2)^{3/2}}\,d\eta\\
&=\frac{27}{4096}\left(\frac{891\pi}{256}-\frac{53}{5}\right).
\end{aligned} e ( R ψ 0 ) = 4096 27 ∫ 0 1 ( 1 − η 2 ) 3/2 η ( 1 − η 3 ) 3 d η = 4096 27 ( 256 891 π − 5 53 ) .
For the last equality, integrate by parts and use
∫ 0 1 η 2 1 − η 2 d η = π 4 , ∫ 0 1 η 5 1 − η 2 d η = 8 15 , ∫ 0 1 η 8 1 − η 2 d η = 35 π 256 . \int_0^1\frac{\eta^2}{\sqrt{1-\eta^2}}\,d\eta=\frac{\pi}{4},\qquad
\int_0^1\frac{\eta^5}{\sqrt{1-\eta^2}}\,d\eta=\frac{8}{15},\qquad
\int_0^1\frac{\eta^8}{\sqrt{1-\eta^2}}\,d\eta=\frac{35\pi}{256}. ∫ 0 1 1 − η 2 η 2 d η = 4 π , ∫ 0 1 1 − η 2 η 5 d η = 15 8 , ∫ 0 1 1 − η 2 η 8 d η = 256 35 π .
Substitution in (60) yields
Q ( I ( A T ) ) = 96 15 π 5 4455 π − 13568 . Q(I(A_{\mathcal{T}}))=\frac{96\sqrt{15}\pi}{5\sqrt{4455\pi-13568}}. Q ( I ( A T )) = 5 4455 π − 13568 96 15 π .
Its strict comparison with Q ( A T ) Q(A_{\mathcal{T}}) Q ( A T ) is equivalent to π > 349784 / 111375 \pi>349784/111375 π > 349784/111375 , which follows already from π > 223 / 71 \pi>223/71 π > 223/71 .
It remains to prove that this first iterate is not an extremal. Put u j = I j ( A T ) u_j=I^j(A_{\mathcal{T}}) u j = I j ( A T ) and q j = Q ( u j ) q_j=Q(u_j) q j = Q ( u j ) . If q 1 q_1 q 1 equalled the common restricted sharp constant, descent and (58) would give q 2 = q 1 q_2=q_1 q 2 = q 1 . The equality case in Lemma 6.1 would then give u 3 = c u 1 u_3=cu_1 u 3 = c u 1 for some c > 0 c>0 c > 0 . Applying D \mathcal{D} D and using D ( u j + 1 ) = u j \mathcal{D}(u_{j+1})=u_j D ( u j + 1 ) = u j yields u 2 = c u 0 u_2=\sqrt{c}u_0 u 2 = c u 0 , hence q 2 = q 0 q_2=q_0 q 2 = q 0 , contradicting q 1 < q 0 q_1<q_0 q 1 < q 0 . We have proved the full strict chain
S 3 / 2 , curl T < Q ( I ( A T ) ) = 96 15 π 5 4455 π − 13568 < 8 10 7 π = Q ( A T ) , S^{\mathcal{T}}_{3/2,\operatorname{curl}}<Q(I(A_{\mathcal{T}}))=\frac{96\sqrt{15}\pi}{5\sqrt{4455\pi-13568}}<\frac{8\sqrt{10}}{7}\pi=Q(A_{\mathcal{T}}), S 3/2 , curl T < Q ( I ( A T )) = 5 4455 π − 13568 96 15 π < 7 8 10 π = Q ( A T ) ,
which proves the corollary. □
Convergence from every one-signed radial profile
The proof uses descent, compactness at a fixed centre, and uniqueness of the positive hyperbolic profile. The scalar compactness mechanism is an instance of the abstract inverse-iteration theory of [8 ] , with p = 3 / 2 p=3/2 p = 3/2 , q = 3 q=3 q = 3 , X = D r a d , O 1 , 3 / 2 ( H 2 ) X=\mathcal{D}^{1,3/2}_{\mathrm{rad},O}(\mathbb{H}^2) X = D rad , O 1 , 3/2 ( H 2 ) and Y = L 3 ( H 2 ) Y=L^3(\mathbb{H}^2) Y = L 3 ( H 2 ) in the notation of that paper; on the scalar level the scheme is an inverse power method of the kind introduced for the p p p -Laplacian in [2 , 15 , 16 ] . We give the full argument here to track the two curl phases and their different normalizations.
Proof of Theorem 1.1 (c). Let 0 ≠ ψ ∈ D r a d , O 1 , 3 / 2 ( H 2 ) 0\ne\psi\in\mathcal{D}^{1,3/2}_{\mathrm{rad},O}(\mathbb{H}^2) 0 = ψ ∈ D rad , O 1 , 3/2 ( H 2 ) have one sign, where O O O is any fixed hyperbolic centre, and define
u 0 = A ( ψ ) , u n + 1 = I ( u n ) . u_0=A(\psi),\qquad u_{n+1}=I(u_n). u 0 = A ( ψ ) , u n + 1 = I ( u n ) .
The scalar equations, the fixed-centre compactness lemma and the uniqueness of the positive profile are invariant under hyperbolic isometries, so there is no restriction on O O O . Since I ( − u ) = − I ( u ) I(-u)=-I(u) I ( − u ) = − I ( u ) , it suffices to assume ψ ≥ 0 \psi\ge0 ψ ≥ 0 . The initial field lies in M \mathcal{M} M by Lemma 2.4. Lemma 6.1 shows inductively that every iterate lies in M \mathcal{M} M and is nonzero, because ∇ × u n + 1 = ∣ u n ∣ u n \nabla\times u_{n+1}=|u_n|u_n ∇ × u n + 1 = ∣ u n ∣ u n . In particular, every quotient below is defined. Put
M n : = ∫ R 3 ∣ u n ∣ 3 d x , q n : = Q ( u n ) . M_n:=\int_{\mathbb{R}^3}|u_n|^3\,\mathrm{d}x,\qquad q_n:=Q(u_n). M n := ∫ R 3 ∣ u n ∣ 3 d x , q n := Q ( u n ) .
Lemma 6.1 and the equality (58) proved in part (a) give
q n → L , L ≥ S 3 / 2 , c u r l T = S 3 / 2 , c u r l S > 0. (61) q_n\to L,\qquad L\ge S^{T}_{3/2,\mathrm{curl}}=S^{S}_{3/2,\mathrm{curl}}>0. \tag*{(61)} q n → L , L ≥ S 3/2 , curl T = S 3/2 , curl S > 0. (61)
Set v n = M n − 1 / 3 u n v_n=M_n^{-1/3}u_n v n = M n − 1/3 u n , so that ∥ v n ∥ 3 = 1 \lVert v_n\rVert_3=1 ∥ v n ∥ 3 = 1 and Q ( v n ) = q n Q(v_n)=q_n Q ( v n ) = q n . Since ∫ ∣ ∇ × u n + 1 ∣ 3 / 2 = M n \int|\nabla\times u_{n+1}|^{3/2}=M_n ∫ ∣∇ × u n + 1 ∣ 3/2 = M n , their recurrence is
∇ × v n + 1 = q n + 1 2 / 3 ∣ v n ∣ v n . (62) \nabla\times v_{n+1}=q_{n+1}^{2/3}|v_n|v_n. \tag*{(62)} ∇ × v n + 1 = q n + 1 2/3 ∣ v n ∣ v n . (62)
The even iterates are toroidal. Write
v 2 k = A ( ϕ k ) = ϕ k ρ e θ . v_{2k}=A(\phi_k)=\frac{\phi_k}{\rho}e_\theta. v 2 k = A ( ϕ k ) = ρ ϕ k e θ .
The rescaled initial profile ϕ 0 = M 0 − 1 / 3 ψ \phi_0=M_0^{-1/3}\psi ϕ 0 = M 0 − 1/3 ψ is nonnegative, nonzero, and radial about O O O . Suppose inductively that ϕ k \phi_k ϕ k has these properties. From (62) ,
D ( v 2 k + 2 ) = q 2 k + 2 1 / 3 v 2 k + 1 . \mathcal{D}(v_{2k+2})=q_{2k+2}^{1/3}v_{2k+1}. D ( v 2 k + 2 ) = q 2 k + 2 1/3 v 2 k + 1 .
Taking curl, using (62) once more, and applying Proposition 4.2 , we obtain
− Δ 3 / 2 , g ϕ k + 1 = q 2 k + 2 1 / 3 q 2 k + 1 2 / 3 ϕ k 2 in H 2 . (63) -\Delta_{3/2,g}\phi_{k+1}=q_{2k+2}^{1/3}q_{2k+1}^{2/3}\phi_k^2\qquad\text{in }\mathbb{H}^2. \tag*{(63)} − Δ 3/2 , g ϕ k + 1 = q 2 k + 2 1/3 q 2 k + 1 2/3 ϕ k 2 in H 2 . (63)
The 3 / 2 3/2 3/2 -Laplace operator on D 1 , 3 / 2 ( H 2 ) \mathcal{D}^{1,3/2}(\mathbb{H}^2) D 1 , 3/2 ( H 2 ) is strictly monotone and coercive, so (63) determines ϕ k + 1 \phi_{k+1} ϕ k + 1 uniquely. Testing with the negative part gives ϕ k + 1 ≥ 0 \phi_{k+1}\ge0 ϕ k + 1 ≥ 0 ; the right-hand side is nonnegative, nontrivial and invariant under the stabilizer of O O O , so by uniqueness ϕ k + 1 \phi_{k+1} ϕ k + 1 is radial about O O O . Formula (59) , applied to the nonzero nonnegative source in (63) , gives ϕ k + 1 ( t ) > 0 \phi_{k+1}(t)>0 ϕ k + 1 ( t ) > 0 for every t > 0 t>0 t > 0 : the corresponding flux is positive for all sufficiently large arguments. This completes the induction. Put
f k : = ( 2 π ) − 1 / 3 ϕ k . f_k:=(2\pi)^{-1/3}\phi_k. f k := ( 2 π ) − 1/3 ϕ k .
By (34) –(35) ,
∫ H 2 f k 3 d V g = 1 , ∫ H 2 ∣ ∇ g f k ∣ g 3 / 2 d V g = q 2 k 2 π . \int_{\mathbb{H}^2} f_k^3\,dV_g=1,\qquad\int_{\mathbb{H}^2}|\nabla_g f_k|_g^{3/2}\,dV_g=\frac{q_{2k}}{\sqrt{2\pi}}. ∫ H 2 f k 3 d V g = 1 , ∫ H 2 ∣ ∇ g f k ∣ g 3/2 d V g = 2 π q 2 k .
Moreover, (63) becomes
− Δ 3 / 2 , g f k + 1 = λ k f k 2 , λ k : = q 2 k + 2 1 / 3 q 2 k + 1 2 / 3 2 π . (64) -\Delta_{3/2,g}f_{k+1}=\lambda_k f_k^2,\qquad\lambda_k:=\frac{q_{2k+2}^{1/3}q_{2k+1}^{2/3}}{\sqrt{2\pi}}. \tag*{(64)} − Δ 3/2 , g f k + 1 = λ k f k 2 , λ k := 2 π q 2 k + 2 1/3 q 2 k + 1 2/3 . (64)
Here and below the equation is understood weakly in X : = D 1 , 3 / 2 ( H 2 ) X:=D^{1,3/2}(\mathbb{H}^2) X := D 1 , 3/2 ( H 2 ) :
∫ H 2 ∣ ∇ g f k + 1 ∣ g − 1 / 2 ⟨ ∇ g f k + 1 , ∇ g h ⟩ g d V g = λ k ∫ H 2 f k 2 h d V g ( h ∈ X ) . \int_{\mathbb{H}^2}|\nabla_g f_{k+1}|_g^{-1/2}\langle\nabla_g f_{k+1},\nabla_g h\rangle_g\,dV_g=\lambda_k\int_{\mathbb{H}^2}f_k^2h\,dV_g\qquad(h\in X). ∫ H 2 ∣ ∇ g f k + 1 ∣ g − 1/2 ⟨ ∇ g f k + 1 , ∇ g h ⟩ g d V g = λ k ∫ H 2 f k 2 h d V g ( h ∈ X ) .
The right-hand side belongs to X ∗ X^* X ∗ because X ↪ L 3 X\hookrightarrow L^3 X ↪ L 3 and f k 2 ∈ L 3 / 2 f_k^2\in L^{3/2} f k 2 ∈ L 3/2 . In view of (61) ,
λ k → ℓ : = L 2 π , ∫ H 2 ∣ ∇ g f k ∣ g 3 / 2 d V g → ℓ . (65) \lambda_k\to\ell:=\frac{L}{\sqrt{2\pi}},\qquad\int_{\mathbb{H}^2}|\nabla_g f_k|_g^{3/2}\,dV_g\to\ell. \tag*{(65)} λ k → ℓ := 2 π L , ∫ H 2 ∣ ∇ g f k ∣ g 3/2 d V g → ℓ . (65)
The sequence ( f k ) (f_k) ( f k ) is bounded in D r a d , O 1 , 3 / 2 ( H 2 ) D_{\mathrm{rad},O}^{1,3/2}(\mathbb{H}^2) D rad , O 1 , 3/2 ( H 2 ) . By Lemma 3.1, after passing to a subsequence, not relabelled, we may assume that, for some f , g f,g f , g ,
f k ⇀ f , f k + 1 ⇀ g in D 1 , 3 / 2 ( H 2 ) , f_k\rightharpoonup f,\qquad f_{k+1}\rightharpoonup g\qquad\text{in }D^{1,3/2}(\mathbb{H}^2), f k ⇀ f , f k + 1 ⇀ g in D 1 , 3/2 ( H 2 ) ,
f k → f , f k + 1 → g in L 3 ( H 2 ) . (66) f_k\to f,\qquad f_{k+1}\to g\qquad\text{in }L^3(\mathbb{H}^2). \tag*{(66)} f k → f , f k + 1 → g in L 3 ( H 2 ) . (66)
In particular, f , g ≥ 0 f,g\geq0 f , g ≥ 0 and ∥ f ∥ L 3 = ∥ g ∥ L 3 = 1 \|f\|_{L^3}=\|g\|_{L^3}=1 ∥ f ∥ L 3 = ∥ g ∥ L 3 = 1 . Testing (64) by f k + 1 f_{k+1} f k + 1 gives
∫ H 2 ∣ ∇ g f k + 1 ∣ g 3 / 2 d V g = λ k ∫ H 2 f k 2 f k + 1 d V g . \int_{\mathbb{H}^2}|\nabla_g f_{k+1}|_g^{3/2}\,dV_g=\lambda_k\int_{\mathbb{H}^2}f_k^2f_{k+1}\,dV_g. ∫ H 2 ∣ ∇ g f k + 1 ∣ g 3/2 d V g = λ k ∫ H 2 f k 2 f k + 1 d V g .
Using (65) and (66) , we obtain
∫ H 2 f 2 g d V g = 1. \int_{\mathbb{H}^2}f^2g\,dV_g=1. ∫ H 2 f 2 g d V g = 1.
Equality in Hölder’s inequality, together with the two unit L 3 L^3 L 3 norms, implies f = g f=g f = g .
Let A = − Δ 3 / 2 , g : X → X ∗ A=-\Delta_{3/2,g}:X\to X^* A = − Δ 3/2 , g : X → X ∗ . By Browder–Minty there is a unique h ∈ X h\in X h ∈ X such that A h = ℓ f 2 Ah=\ell f^2 A h = ℓ f 2 . On bounded subsets of X X X , the monotonicity inequality for p = 3 / 2 p=3/2 p = 3/2 gives
∥ f k + 1 − h ∥ X 2 ≤ C ⟨ A f k + 1 − A h , f k + 1 − h ⟩ ≤ C ∥ λ k f k 2 − ℓ f 2 ∥ X ∗ ∥ f k + 1 − h ∥ X , \begin{aligned}
\|f_{k+1}-h\|_X^2\leq C\langle Af_{k+1}-Ah,f_{k+1}-h\rangle\\
&\leq C\|\lambda_k f_k^2-\ell f^2\|_{X^*}\|f_{k+1}-h\|_X,
\end{aligned} ∥ f k + 1 − h ∥ X 2 ≤ C ⟨ A f k + 1 − A h , f k + 1 − h ⟩ ≤ C ∥ λ k f k 2 − ℓ f 2 ∥ X ∗ ∥ f k + 1 − h ∥ X ,
where ∥ h ∥ X = ∥ ∇ g h ∥ 3 / 2 \|h\|_X=\|\nabla_g h\|_{3/2} ∥ h ∥ X = ∥ ∇ g h ∥ 3/2 . The last dual norm tends to zero because f k → f f_k\to f f k → f in L 3 L^3 L 3 . Thus f k + 1 → h f_{k+1}\to h f k + 1 → h strongly in X X X . Its L 3 L^3 L 3 limit is g = f g=f g = f , so h = f h=f h = f and
− Δ 3 / 2 , g f = ℓ f 2 , ∥ f ∥ L 3 = 1. -\Delta_{3/2,g}f=\ell f^2,\qquad\|f\|_{L^3}=1. − Δ 3/2 , g f = ℓ f 2 , ∥ f ∥ L 3 = 1.
Testing with f f f yields ∥ ∇ g f ∥ g 3 / 2 = ℓ \|\nabla_gf\|_g^{3/2}=\ell ∥ ∇ g f ∥ g 3/2 = ℓ . The energies of f k f_k f k tend to the same value by (65) ; weak convergence and uniform convexity therefore imply f k → f f_k\to f f k → f strongly in X X X along this subsequence. The radial representation (59) , with source ℓ f 2 \ell f^2 ℓ f 2 , gives f ( t ) > 0 f(t)>0 f ( t ) > 0 for t > 0 t>0 t > 0 : since f ≢ 0 f\not\equiv0 f ≡ 0 , the source flux is positive for all sufficiently large radii, so every tail integral in that representation is positive. After the rescaling f ~ = ℓ 2 / 3 f \tilde f=\ell^{2/3}f f ~ = ℓ 2/3 f , Theorem 3.2 identifies f ~ \tilde f f ~ with the unique positive finite-energy profile centred at O O O . Let f ∗ f_* f ∗ be the positive hyperbolic minimizer centred at O O O with ∥ f ∗ ∥ L 3 = 1 \|f_*\|_{L^3}=1 ∥ f ∗ ∥ L 3 = 1 . Its Euler equation is
− Δ 3 / 2 , g f ∗ = Λ H 2 f ∗ 2 . -\Delta_{3/2,g}f_*=\Lambda_{\mathbb{H}^2}f_*^2. − Δ 3/2 , g f ∗ = Λ H 2 f ∗ 2 .
Thus Λ H 2 2 / 3 f ∗ \Lambda_{\mathbb{H}^{2}}^{2/3}f_* Λ H 2 2/3 f ∗ is another centred positive solution of the coefficient-one equation. Uniqueness gives
ℓ 2 / 3 f = Λ H 2 2 / 3 f ∗ . \ell^{2/3}f=\Lambda_{\mathbb{H}^{2}}^{2/3}f_*. ℓ 2/3 f = Λ H 2 2/3 f ∗ .
Taking L 3 L^3 L 3 norms therefore yields
(7.10) ℓ = Λ H 2 \ell=\Lambda_{\mathbb{H}^{2}} ℓ = Λ H 2 .
The same argument applies to every subsequence, and the limit is always the same profile of L 3 L^3 L 3 norm one; hence the whole sequence of rescaled even profiles converges strongly in D 1 , 3 / 2 ( H 2 ) D^{1,3/2}(\mathbb{H}^{2}) D 1 , 3/2 ( H 2 ) and in L 3 ( H 2 ) L^3(\mathbb{H}^{2}) L 3 ( H 2 ) . Combining (65) , (7.10) and (36) we conclude that
Q ( u n ) → L = 2 π Λ H 2 = S 3 / 2 , c u r l T . (67) Q(u_n)\to L=\sqrt{2\pi}\Lambda_{\mathbb{H}^{2}}=S^{T}_{3/2,\mathrm{curl}}. \tag*{(67)} Q ( u n ) → L = 2 π Λ H 2 = S 3/2 , curl T . (67)
Although only the even iterates entered the compactness argument, L L L was defined in (61) as the limit of the whole monotone sequence ( q n ) (q_n) ( q n ) , so (67) holds for the full sequence. If u 2 k = A ( ψ k ) u_{2k}=A(\psi_k) u 2 k = A ( ψ k ) , the profiles used above satisfy f k = ψ k / ∥ ψ k ∥ L 3 ( H 2 ) f_k=\psi_k/\|\psi_k\|_{L^3(\mathbb{H}^{2})} f k = ψ k /∥ ψ k ∥ L 3 ( H 2 ) , and (34) –(35) turn the strong convergence f k → f ∗ f_k\to f_* f k → f ∗ in X X X and in L 3 ( H 2 ) L^3(\mathbb{H}^{2}) L 3 ( H 2 ) into
v 2 k = ( 2 π ) − 1 / 3 A ( f k ) ⟶ ( 2 π ) − 1 / 3 A ( f ∗ ) = A ∗ ∥ A ∗ ∥ L 3 v_{2k}=(2\pi)^{-1/3}A(f_k)\longrightarrow(2\pi)^{-1/3}A(f_*)=\frac{A_*}{\|A_*\|_{L^3}} v 2 k = ( 2 π ) − 1/3 A ( f k ) ⟶ ( 2 π ) − 1/3 A ( f ∗ ) = ∥ A ∗ ∥ L 3 A ∗
in L 3 ( R 3 ) L^3(\mathbb{R}^3) L 3 ( R 3 ) , with curls converging in L 3 / 2 ( R 3 ) L^{3/2}(\mathbb{R}^3) L 3/2 ( R 3 ) ; here O = ( 1 , 0 ) O=(1,0) O = ( 1 , 0 ) , and for a general centre A ∗ A_* A ∗ is replaced by its image under (5) .
The odd iterates. The map I I I is continuous for the L 3 L^3 L 3 topology: u ↦ ∣ u ∣ u u\mapsto|u|u u ↦ ∣ u ∣ u is continuous from L 3 L^3 L 3 to L 3 / 2 L^{3/2} L 3/2 , K \mathcal{K} K is bounded from L 3 / 2 L^{3/2} L 3/2 to L 3 L^3 L 3 , and Γ = Id − P W \Gamma=\operatorname{Id}-P_{\mathcal{W}} Γ = Id − P W , where P W P_{\mathcal{W}} P W is the metric projection onto the closed subspace W \mathcal{W} W of the uniformly convex space L 3 L^3 L 3 , is continuous. With M ∗ : = ∫ R 3 ∣ A ∗ ∣ 3 d x M_*:=\int_{\mathbb{R}^3}|A_*|^3\,dx M ∗ := ∫ R 3 ∣ A ∗ ∣ 3 d x and v ∞ : = M ∗ − 1 / 3 A ∗ v_\infty:=M_*^{-1/3}A_* v ∞ := M ∗ − 1/3 A ∗ , the definition of I I I and (47) give
v 2 k + 1 = I ( u 2 k ) M 2 k + 1 1 / 3 = M 2 k 2 / 3 M 2 k + 1 1 / 3 Γ K ( ∣ v 2 k ∣ v 2 k ) = q 2 k + 1 2 / 3 Γ K ( ∣ v 2 k ∣ v 2 k ) ⟶ L 2 / 3 Γ K ( ∣ v ∞ ∣ v ∞ ) v_{2k+1}=\frac{I(u_{2k})}{M_{2k+1}^{1/3}}=\frac{M_{2k}^{2/3}}{M_{2k+1}^{1/3}}\Gamma\mathcal{K}(|v_{2k}|v_{2k})=q_{2k+1}^{2/3}\Gamma\mathcal{K}(|v_{2k}|v_{2k})\longrightarrow L^{2/3}\Gamma\mathcal{K}(|v_\infty|v_\infty) v 2 k + 1 = M 2 k + 1 1/3 I ( u 2 k ) = M 2 k + 1 1/3 M 2 k 2/3 Γ K ( ∣ v 2 k ∣ v 2 k ) = q 2 k + 1 2/3 Γ K ( ∣ v 2 k ∣ v 2 k ) ⟶ L 2/3 Γ K ( ∣ v ∞ ∣ v ∞ )
in L 3 ( R 3 ) L^3(\mathbb{R}^3) L 3 ( R 3 ) . Since A ∗ A_* A ∗ solves (2) , the gauge-minimal field with curl ∣ A ∗ ∣ A ∗ |A_*|A_* ∣ A ∗ ∣ A ∗ is U ∗ = D ( A ∗ ) U_*=\mathcal{D}(A_*) U ∗ = D ( A ∗ ) , so Γ K ( ∣ v ∞ ∣ v ∞ ) = M ∗ − 2 / 3 U ∗ \Gamma\mathcal{K}(|v_\infty|v_\infty)=M_*^{-2/3}U_* Γ K ( ∣ v ∞ ∣ v ∞ ) = M ∗ − 2/3 U ∗ ; moreover L = Q ( A ∗ ) = M ∗ 1 / 2 L=Q(A_*)=M_*^{1/2} L = Q ( A ∗ ) = M ∗ 1/2 and ∫ R 3 ∣ U ∗ ∣ 3 d x = M ∗ \int_{\mathbb{R}^3}|U_*|^3\,dx=M_* ∫ R 3 ∣ U ∗ ∣ 3 d x = M ∗ by the Nehari identity. Hence v 2 k + 1 → M ∗ − 1 / 3 U ∗ = U ∗ / ∥ U ∗ ∥ L 3 v_{2k+1}\to M_*^{-1/3}U_*=U_*/\|U_*\|_{L^3} v 2 k + 1 → M ∗ − 1/3 U ∗ = U ∗ /∥ U ∗ ∥ L 3 , and ∇ × v 2 k + 1 = q 2 k + 1 2 / 3 ∣ v 2 k ∣ v 2 k → L 2 / 3 ∣ v ∞ ∣ v ∞ = ∇ × ( M ∗ − 1 / 3 U ∗ ) \nabla\times v_{2k+1}=q_{2k+1}^{2/3}|v_{2k}|v_{2k}\to L^{2/3}|v_\infty|v_\infty=\nabla\times(M_*^{-1/3}U_*) ∇ × v 2 k + 1 = q 2 k + 1 2/3 ∣ v 2 k ∣ v 2 k → L 2/3 ∣ v ∞ ∣ v ∞ = ∇ × ( M ∗ − 1/3 U ∗ ) in L 3 / 2 ( R 3 ) L^{3/2}(\mathbb{R}^3) L 3/2 ( R 3 ) . For nonpositive initial data the oddness of I I I reverses the signs of both limits. □
Breakdown of the forward iteration
Proposition 7.1 (Breakdown of the forward iteration). Let u j : = D j ( A T ) u_j:=\mathcal{D}^j(A_T) u j := D j ( A T ) and q j : = Q ( u j ) q_j:=Q(u_j) q j := Q ( u j ) , as long as the forward image belongs to the natural domain. Then
(7.12) u j ∈ D 1 , 3 / 2 ( curl ; R 3 ) u_j\in D^{1,3/2}(\operatorname{curl};\mathbb{R}^3) u j ∈ D 1 , 3/2 ( curl ; R 3 ) ( j = 0 , 1 , 2 ) (j=0,1,2) ( j = 0 , 1 , 2 ) , u 3 ∈ L 3 ( R 3 , R 3 ) ∖ D 1 , 3 / 2 ( curl ; R 3 ) u_3\in L^3(\mathbb{R}^3,\mathbb{R}^3)\setminus D^{1,3/2}(\operatorname{curl};\mathbb{R}^3) u 3 ∈ L 3 ( R 3 , R 3 ) ∖ D 1 , 3/2 ( curl ; R 3 ) , and q 0 < q 1 < q 2 q_0<q_1<q_2 q 0 < q 1 < q 2 , with the values (68) –(69) below.
Proof. The involution in Theorem 2.1 at p = 3 / 2 p=3/2 p = 3/2 acts on solutions; it does not define a dynamical system on the energy space. For a toroidal field with a hyperbolically radial profile, Proposition 4.2 gives, whenever the expressions belong to the required spaces,
D 2 ( A ( ψ ) ) = A ( sgn ( F ) ∣ F ∣ ) , F : = − Δ 3 / 2 , g ψ . \mathcal{D}^{2}(A(\psi))=A\left(\operatorname{sgn}(F)\sqrt{|F|}\right),\qquad F:=-\Delta_{3/2,g}\psi. D 2 ( A ( ψ )) = A ( sgn ( F ) ∣ F ∣ ) , F := − Δ 3/2 , g ψ .
Indeed,
∇ × D ( A ( ψ ) ) = F ρ 2 e θ . \nabla\times\mathcal{D}(A(\psi)) = \frac{F}{\rho^2}e_\theta. ∇ × D ( A ( ψ )) = ρ 2 F e θ .
The first three finite quotients are
q 0 = 8 10 7 π ≈ 11.3538151609098 , (68) q_0 = \frac{8\sqrt{10}}{7}\pi\approx11.3538151609098, \tag*{(68)} q 0 = 7 8 10 π ≈ 11.3538151609098 , (68)
q 1 = π 7 2 ( 3 2 ) 3 / 2 B ( 5 8 , 7 4 ) ≈ 11.6468871437934 , q_1 = \pi\sqrt{\frac{7}{2}}\left(\frac{3}{2}\right)^{3/2} \mathrm{B}\left(\frac{5}{8},\frac{7}{4}\right) \approx11.6468871437934, q 1 = π 2 7 ( 2 3 ) 3/2 B ( 8 5 , 4 7 ) ≈ 11.6468871437934 ,
q 2 = π 4 2 B ( 5 8 , 7 4 ) 1 / 2 H 0 ≈ 18.5815053070815 , (69) q_2 = \frac{\pi}{4\sqrt{2}\,\mathrm{B}\left(\frac{5}{8},\frac{7}{4}\right)^{1/2}}\mathcal{H}_0 \approx18.5815053070815, \tag*{(69)} q 2 = 4 2 B ( 8 5 , 4 7 ) 1/2 π H 0 ≈ 18.5815053070815 , (69)
where
H 0 : = ∫ 0 1 x − 15 / 16 ( 1 − x ) − 3 / 8 ( 1 + 5 x ) 3 / 2 d x = B ( 1 16 , 5 8 ) 2 F 1 ( − 3 2 , 1 16 ; 11 16 ; − 5 ) . \begin{aligned}
\mathcal{H}_0 &:= \int_0^1 x^{-15/16}(1-x)^{-3/8}(1+5x)^{3/2}\,\mathrm{d}x \\
&= \mathrm{B}\left(\frac{1}{16},\frac{5}{8}\right){}_2F_1\left(-\frac{3}{2},\frac{1}{16};\frac{11}{16};-5\right).
\end{aligned} H 0 := ∫ 0 1 x − 15/16 ( 1 − x ) − 3/8 ( 1 + 5 x ) 3/2 d x = B ( 16 1 , 8 5 ) 2 F 1 ( − 2 3 , 16 1 ; 16 11 ; − 5 ) .
In particular q 0 < q 1 < q 2 q_0<q_1<q_2 q 0 < q 1 < q 2 , and q 3 q_3 q 3 is not defined. For ψ 0 = ( 3 / 4 ) sech 2 t \psi_0=(3/4)\operatorname{sech}^2t ψ 0 = ( 3/4 ) sech 2 t , a direct radial calculation gives
F 0 : = − Δ 3 / 2 , g ψ 0 = ( 3 2 ) 3 / 2 sech 3 t tanh − 1 / 2 t . F_0:=-\Delta_{3/2,g}\psi_0=\left(\frac{3}{2}\right)^{3/2}\operatorname{sech}^3t\tanh^{-1/2}t. F 0 := − Δ 3/2 , g ψ 0 = ( 2 3 ) 3/2 sech 3 t tanh − 1/2 t .
Since sinh t − ψ 0 ′ ( t ) = 3 / 2 tanh 3 / 2 t → 0 \sinh t\sqrt{-\psi_0'(t)}=\sqrt{3/2}\tanh^{3/2}t\to0 sinh t − ψ 0 ′ ( t ) = 3/2 tanh 3/2 t → 0 as t → 0 + t\to0^+ t → 0 + , this formula holds distributionally without a point mass at the hyperbolic centre. Consequently,
u 2 = A ( χ 0 ) , χ 0 ( t ) : = F 0 ( t ) = ( 3 2 ) 3 / 4 sech 3 / 2 t tanh − 1 / 4 t . u_2=A(\chi_0),\qquad\chi_0(t):=\sqrt{F_0(t)}=\left(\frac{3}{2}\right)^{3/4}\operatorname{sech}^{3/2}t\tanh^{-1/4}t. u 2 = A ( χ 0 ) , χ 0 ( t ) := F 0 ( t ) = ( 2 3 ) 3/4 sech 3/2 t tanh − 1/4 t .
Put M j : = ∫ R 3 ∣ u j ∣ 3 d x M_j:=\int_{\mathbb{R}^3}|u_j|^3\,\mathrm{d}x M j := ∫ R 3 ∣ u j ∣ 3 d x and E j : = ∫ R 3 ∣ ∇ × u j ∣ 3 / 2 d x E_j:=\int_{\mathbb{R}^3}|\nabla\times u_j|^{3/2}\,\mathrm{d}x E j := ∫ R 3 ∣∇ × u j ∣ 3/2 d x , so that q j = E j / M j 1 / 2 q_j=E_j/M_j^{1/2} q j = E j / M j 1/2 and M j + 1 = E j M_{j+1}=E_j M j + 1 = E j . In radial hyperbolic coordinates, the identity ∫ 0 ∞ sech a t tanh b t sinh t d t = 1 2 B ( 1 + b 2 , a − 1 2 ) \int_0^\infty\operatorname{sech}^a t\tanh^b t\sinh t\,\mathrm{d}t=\frac{1}{2}\mathrm{B}\left(1+\frac{b}{2},\frac{a-1}{2}\right) ∫ 0 ∞ sech a t tanh b t sinh t d t = 2 1 B ( 1 + 2 b , 2 a − 1 ) , valid for a > 1 a>1 a > 1 and b > − 2 b>-2 b > − 2 , gives M 0 = 27 π 2 / 80 M_0=27\pi^2/80 M 0 = 27 π 2 /80 , E 0 = 8 π 2 7 ( 3 / 2 ) 3 / 2 E_0=\frac{8\pi^2}{7}(3/2)^{3/2} E 0 = 7 8 π 2 ( 3/2 ) 3/2 and E 1 = 2 π 2 ( 3 / 2 ) 9 / 4 B ( 5 8 , 7 4 ) E_1=2\pi^2(3/2)^{9/4}\mathrm{B}\left(\frac{5}{8},\frac{7}{4}\right) E 1 = 2 π 2 ( 3/2 ) 9/4 B ( 8 5 , 4 7 ) , while − χ 0 ′ ( t ) = 1 4 ( 3 / 2 ) 3 / 4 sech 3 / 2 t tanh − 5 / 4 t ( 1 + 5 tanh 2 t ) -\chi_0'(t)=\frac{1}{4}(3/2)^{3/4}\operatorname{sech}^{3/2}t\tanh^{-5/4}t(1+5\tanh^2t) − χ 0 ′ ( t ) = 4 1 ( 3/2 ) 3/4 sech 3/2 t tanh − 5/4 t ( 1 + 5 tanh 2 t ) and the substitution x = tanh 2 t x=\tanh^2t x = tanh 2 t give E 2 = π 2 4 ( 3 / 2 ) 9 / 8 H 0 E_2=\frac{\pi^2}{4}(3/2)^{9/8}\mathcal{H}_0 E 2 = 4 π 2 ( 3/2 ) 9/8 H 0 . Formulas (68) –(69) follow.
Every forward image u j + 1 u_{j+1} u j + 1 which belongs to L 3 L^3 L 3 is automatically in the nonlinear gauge, because ∣ u j + 1 ∣ u j + 1 = ∇ × u j |u_{j+1}|u_{j+1}=\nabla\times u_j ∣ u j + 1 ∣ u j + 1 = ∇ × u j . Since u 0 = A T u_0=A_T u 0 = A T is also in that gauge, whenever two consecutive forward fields belong to the energy space,
I ( D ( u j ) ) = u j . I(\mathcal{D}(u_j))=u_j. I ( D ( u j )) = u j .
Lemma 6.1 gives Q ( u j ) ≤ Q ( u j + 1 ) Q(u_j)\leq Q(u_{j+1}) Q ( u j ) ≤ Q ( u j + 1 ) . Equality in the first step would require F 0 F_0 F 0 to be a constant multiple of ψ 0 2 \psi_0^2 ψ 0 2 , which is false. Equality in the second would imply, by taking curls and using uniqueness of the toroidal gauge representative, that χ 0 \chi_0 χ 0 is a constant multiple of ψ 0 \psi_0 ψ 0 , which is also false. Thus the two inequalities are strict.
It remains to determine the domain of the third image. As t → 0 + t \to0^{+} t → 0 + ,
χ 0 ( t ) = ( 3 2 ) 3 / 4 t − 1 / 4 ( 1 + O ( t 2 ) ) , − χ 0 ′ ( t ) = 1 4 ( 3 2 ) 3 / 4 t − 5 / 4 ( 1 + O ( t 2 ) ) . \chi_{0}(t)=\left(\frac{3}{2}\right)^{3/4}t^{-1/4}(1+O(t^{2})),\qquad-\chi_{0}'(t)=\frac{1}{4}\left(\frac{3}{2}\right)^{3/4}t^{-5/4}(1+O(t^{2})). χ 0 ( t ) = ( 2 3 ) 3/4 t − 1/4 ( 1 + O ( t 2 )) , − χ 0 ′ ( t ) = 4 1 ( 2 3 ) 3/4 t − 5/4 ( 1 + O ( t 2 )) .
Moreover,
sinh t − χ 0 ′ ( t ) ∼ 1 2 ( 3 2 ) 3 / 8 t 3 / 8 → 0. \sinh t\sqrt{-\chi_{0}'(t)}\sim\frac{1}{2}\left(\frac{3}{2}\right)^{3/8}t^{3/8}\to0. sinh t − χ 0 ′ ( t ) ∼ 2 1 ( 2 3 ) 3/8 t 3/8 → 0.
Thus the flux through small circles around the hyperbolic centre tends to zero and the distributional 3 / 2 3/2 3/2 -Laplacian of χ 0 \chi_{0} χ 0 carries no Dirac mass there. Moreover χ 0 ∈ L 3 ( H 2 ) \chi_{0}\in L^{3}(\mathbb{H}^{2}) χ 0 ∈ L 3 ( H 2 ) and
∣ χ 0 ′ ∣ 3 / 2 sinh t ≍ t − 7 / 8 |\chi_{0}'|^{3/2}\sinh t\asymp t^{-7/8} ∣ χ 0 ′ ∣ 3/2 sinh t ≍ t − 7/8
is integrable at zero, which confirms u 2 ∈ D 1 , 3 / 2 ( curl ; R 3 ) u_{2}\in D^{1,3/2}(\operatorname{curl};\mathbb{R}^{3}) u 2 ∈ D 1 , 3/2 ( curl ; R 3 ) . On the other hand, if F 2 : = − Δ 3 / 2 , g χ 0 F_{2}:=-\Delta_{3/2,g}\chi_{0} F 2 := − Δ 3/2 , g χ 0 , then
F 2 ( t ) = 1 sinh t d d t ( sinh t − χ 0 ′ ( t ) ) = 3 16 ( 3 2 ) 3 / 8 t − 13 / 8 ( 1 + o ( 1 ) ) . F_{2}(t)=\frac{1}{\sinh t}\frac{\mathrm{d}}{\mathrm{d}t}\left(\sinh t\sqrt{-\chi_{0}'(t)}\right)=\frac{3}{16}\left(\frac{3}{2}\right)^{3/8}t^{-13/8}(1+o(1)). F 2 ( t ) = sinh t 1 d t d ( sinh t − χ 0 ′ ( t ) ) = 16 3 ( 2 3 ) 3/8 t − 13/8 ( 1 + o ( 1 )) .
Consequently,
∣ F 2 ( t ) ∣ 3 / 2 sinh t ≍ t − 23 / 16 , |F_{2}(t)|^{3/2}\sinh t\asymp t^{-23/16}, ∣ F 2 ( t ) ∣ 3/2 sinh t ≍ t − 23/16 ,
which is not integrable at zero. At infinity, χ 0 ( t ) = O ( e − 3 t / 2 ) \chi_{0}(t)=O(e^{-3t/2}) χ 0 ( t ) = O ( e − 3 t /2 ) and F 2 ( t ) = O ( e − 3 t / 4 ) F_{2}(t)=O(e^{-3t/4}) F 2 ( t ) = O ( e − 3 t /4 ) , so the relevant L 3 L^{3} L 3 and curl-energy densities are integrable there. If ∇ × u 3 \nabla\times u_{3} ∇ × u 3 belonged to L 3 / 2 L^{3/2} L 3/2 , it would coincide almost everywhere with the pointwise curl ρ − 2 F 2 e θ \rho^{-2}F_{2}e_{\theta} ρ − 2 F 2 e θ , whose L 3 / 2 L^{3/2} L 3/2 norm is ( 2 π ∫ H 2 ∣ F 2 ∣ 3 / 2 d V g ) 2 / 3 = ∞ \left(2\pi\int_{\mathbb{H}^{2}}|F_{2}|^{3/2}\,\mathrm{d}V_{g}\right)^{2/3}=\infty ( 2 π ∫ H 2 ∣ F 2 ∣ 3/2 d V g ) 2/3 = ∞ . On the other hand, ∫ R 3 ∣ u 3 ∣ 3 d x = E 2 < ∞ \int_{\mathbb{R}^{3}}|u_{3}|^{3}\,\mathrm{d}x=E_{2}<\infty ∫ R 3 ∣ u 3 ∣ 3 d x = E 2 < ∞ . This proves (7.12). □
Numerical values
The numerical values in this section are given for orientation. They are not certified enclosures and are not used in any proof.
For a decreasing radial solution define the flux J ( t ) : = sinh t − ψ ∗ ′ ( t ) \mathcal{J}(t):=\sinh t\sqrt{-\psi_{*}'(t)} J ( t ) := sinh t − ψ ∗ ′ ( t ) . Equation (25) becomes the first-order flux system
ψ ∗ ′ ( t ) = − ( J ( t ) sinh t ) 2 , J ′ ( t ) = ψ ∗ ( t ) 2 sinh t . \psi_{*}'(t)=-\left(\frac{\mathcal{J}(t)}{\sinh t}\right)^{2},\qquad\mathcal{J}'(t)=\psi_{*}(t)^{2}\sinh t. ψ ∗ ′ ( t ) = − ( sinh t J ( t ) ) 2 , J ′ ( t ) = ψ ∗ ( t ) 2 sinh t .
with ψ ∗ ( 0 ) = a ∗ > 0 \psi_{*}(0)=a_{*}>0 ψ ∗ ( 0 ) = a ∗ > 0 and J ( 0 ) = 0 \mathcal{J}(0)=0 J ( 0 ) = 0 . The local expansion from Remark 3.4 gives J ( t ) = a ∗ 2 2 t 2 + o ( t 2 ) \mathcal{J}(t)=\frac{a_{*}^{2}}{2}t^{2}+o(t^{2}) J ( t ) = 2 a ∗ 2 t 2 + o ( t 2 ) and ψ ∗ ( t ) = a ∗ − a ∗ 4 12 t 3 + o ( t 3 ) \psi_{*}(t)=a_{*}-\frac{a_{*}^{4}}{12}t^{3}+o(t^{3}) ψ ∗ ( t ) = a ∗ − 12 a ∗ 4 t 3 + o ( t 3 ) as t → 0 + t\to0^{+} t → 0 + . Numerical shooting for the rapidly decaying branch, using fourth-order Runge–Kutta integration and bisection in the initial value, gives a ∗ = ψ ∗ ( 0 ) ≈ 2.18885994 a_{*}=\psi_{*}(0)\approx2.18885994 a ∗ = ψ ∗ ( 0 ) ≈ 2.18885994 .
Let I ∗ : = ∫ 0 ∞ ∣ ψ ∗ ′ ∣ 3 / 2 sinh t d t = ∫ 0 ∞ ψ ∗ 3 sinh t d t I_{*}:=\int_{0}^{\infty}|\psi_{*}'|^{3/2}\sinh t\,\mathrm{d}t=\int_{0}^{\infty}\psi_{*}^{3}\sinh t\,\mathrm{d}t I ∗ := ∫ 0 ∞ ∣ ψ ∗ ′ ∣ 3/2 sinh t d t = ∫ 0 ∞ ψ ∗ 3 sinh t d t , where the equality follows by testing (25) with ψ ∗ \psi_{*} ψ ∗ ; numerically, I ∗ ≈ 3.21240637 I_{*}\approx3.21240637 I ∗ ≈ 3.21240637 . The full hyperbolic integrals contain an additional angular factor 2 π 2\pi 2 π , so Proposition 4.1 gives S 3 / 2 , curl H = S 3 / 2 , curl R 3 = 2 π I ∗ ≈ 11.26147061 S^{\mathbb{H}}_{3/2,\operatorname{curl}}=S^{\mathbb{R}^{3}}_{3/2,\operatorname{curl}}=2\pi\sqrt{I_{*}}\approx11.26147061 S 3/2 , curl H = S 3/2 , curl R 3 = 2 π I ∗ ≈ 11.26147061 . The improvement over the original explicit bound is 8 10 π − 11.2614706054 ≈ 0.0923 \frac{8\sqrt{10}}{\pi}-11.2614706054\approx0.0923 π 8 10 − 11.2614706054 ≈ 0.0923 , that is, about 0.8 % 0.8\% 0.8% .
The inverse iteration of Theorem 1.1(c), started from A T A_{T} A T and computed by the quadratures (59) and (60) on a fine grid, gives the following values of q n = Q ( I n ( A T ) ) q_{n}=Q(I^{n}(A_{T})) q n = Q ( I n ( A T )) :
Table 1.
The differences q n − S 3 / 2 , curl R 3 q_{n}-S^{\mathbb{R}^{3}}_{3/2,\operatorname{curl}} q n − S 3/2 , curl R 3 decrease eventually by a factor of about 0.67 0.67 0.67 per step, which suggests geometric convergence of the projective iteration; the convergence theorem does not provide a rate.
Open problems
The main question left open is whether the restricted extremals are global minimizers.
Conjecture 9.1 (Symmetry of global minimizers). Every global minimizer of S 3 / 2 , c u r l S_{3/2,\mathrm{curl}} S 3/2 , curl is, after a conformal transformation of R 3 \mathbb{R}^{3} R 3 and a change of gauge, a field of class S \mathcal{S} S , equivalently of class T \mathcal{T} T ; in particular S 3 / 2 , c u r l = S 3 / 2 , c u r l S S_{3/2,\mathrm{curl}}=S^{\mathcal{S}}_{3/2,\mathrm{curl}} S 3/2 , curl = S 3/2 , curl S and U ∗ U_\ast U ∗ is a global ground state.
The evidence is as follows. The best values known to us are Q ( A L Y ) = 4 π ≈ 12.566 Q(A_{\mathrm{LY}})=4\pi\approx12.566 Q ( A LY ) = 4 π ≈ 12.566 , Q ( A T ) ≈ 11.354 Q(A_{\mathcal{T}})\approx11.354 Q ( A T ) ≈ 11.354 , and S 3 / 2 , c u r l T ≈ 11.261 S^{\mathcal{T}}_{3/2,\mathrm{curl}}\approx11.261 S 3/2 , curl T ≈ 11.261 , and no competitor outside the two classes with a smaller quotient is known; the exclusion of the Loss–Yau candidate within the axisymmetric class [3 ] does not decide the question. On S 3 \mathbb{S}^{3} S 3 the centred extremals A ∗ A_\ast A ∗ and U ∗ U_\ast U ∗ correspond to one-forms along the two circle factors of the Hopf tori, so the conjecture asks whether a global minimizer can be moved by conformal maps onto a single Hopf circle factor.
Question 9.2 (Symmetry and dual exponents). For p ≠ 3 / 2 p\ne3/2 p = 3/2 , which of S p , c u r l T S^{\mathcal{T}}_{p,\mathrm{curl}} S p , curl T and S p , c u r l S S^{\mathcal{S}}_{p,\mathrm{curl}} S p , curl S is smaller? The duality identities relate opposite classes at p p p and p ^ \hat{p} p ^ and do not determine their ordering at a fixed exponent. Does either restricted class attain the unrestricted constant at p = 2 p=2 p = 2 or at p = 6 / 5 p=6/5 p = 6/5 ?
Question 9.3 (Rate and basin of the inverse iteration). Theorem 1.1 (c) gives convergence of the normalized iterates without a rate. Is the convergence geometric, that is, are there C > 0 C>0 C > 0 and λ < 1 \lambda<1 λ < 1 such that ∥ v n − v ∞ ( n ) ∥ L 3 ≤ C λ n \lVert v_n-v_\infty^{(n)}\rVert_{L^3}\le C\lambda^n ∥ v n − v ∞ ( n ) ∥ L 3 ≤ C λ n and Q ( u n ) − S 3 / 2 , c u r l T ≤ C λ n Q(u_n)-S^{\mathcal{T}}_{3/2,\mathrm{curl}}\le C\lambda^n Q ( u n ) − S 3/2 , curl T ≤ C λ n , where v ∞ v_\infty v ∞ denotes the limit of the even or of the odd iterates according to the parity of n n n ? The values in § 8 suggest that Q ( u n ) − S 3 / 2 , c u r l T Q(u_n)-S^{\mathcal{T}}_{3/2,\mathrm{curl}} Q ( u n ) − S 3/2 , curl T decreases by a factor of about 0.67 per step. The second question concerns the basin of attraction of the two-cycle, that is, the set of initial fields u 0 u_0 u 0 whose normalized iterates v n v_n v n converge to ± T λ , z 0 A ∗ / ∥ A ∗ ∥ L 3 \pm T_{\lambda,z_0}A_\ast/\lVert A_\ast\rVert_{L^3} ± T λ , z 0 A ∗ / ∥ A ∗ ∥ L 3 and ± T λ , z 0 U ∗ / ∥ U ∗ ∥ L 3 \pm T_{\lambda,z_0}U_\ast/\lVert U_\ast\rVert_{L^3} ± T λ , z 0 U ∗ / ∥ U ∗ ∥ L 3 for some centre.
Acknowledgments
The author was partly supported by the National Science Centre, Poland, grant no. 2023/51/B/ST1/00968.
AI statement
The author used AI tools to assist with language editing, the presentation of mathematical arguments developed by the author, literature searches, and the numerical computations reported in § 7.3 and § 8 . All mathematical arguments and proofs in the final manuscript were written and verified by the author.
References
J. Mederski, Institute of Mathematics, Polish Academy of Sciences, ul. Śniadeckich 8, 00-656 Warsaw, Poland, and Faculty of Mathematics and Computer Science, Nicolaus Copernicus University, ul. Chopina 12/18, 87-100 Toruń, Poland. Email address: jmederski@impan.pl