Introduction and main theorems
The main results of this paper are a resonance-free strip and the construction of exponentially narrow whispering-gallery resonances for Maxwell’s equations for a transparent ball, although many of the spectral theory results are for Lipschitz domains. The permittivity ϵ \epsilon ϵ and permeability μ \mu μ are positive and piecewise constant, with subscripts 1 and 0 denoting the interior and exterior values. If m 0 = ϵ 1 μ 1 ϵ 0 μ 0 ≤ 1 m_0 = \sqrt{\frac{\epsilon_1\mu_1}{\epsilon_0\mu_0}} \le1 m 0 = ϵ 0 μ 0 ϵ 1 μ 1 ≤ 1 , the cutoff resolvent is holomorphic in a strip of fixed width below the real axis at high frequency. If m 0 > 1 m_0 > 1 m 0 > 1 , internal trapping produces resonances with exponentially small negative imaginary parts, so no such strip exists.
The main difficulty is to control the Maxwell transmission determinants uniformly in angular momentum and complex frequency. The explicit Mie formulas alone do not give this control: the interior and exterior Bessel arguments pass through different turning points, and both polarizations must be treated with their respective impedance factors. For m 0 ≤ 1 m_0 \le1 m 0 ≤ 1 , we exclude zeros throughout the subturning, turning and oscillatory regimes, treating the endpoint m 0 = 1 m_0 = 1 m 0 = 1 separately. For m 0 > 1 m_0 > 1 m 0 > 1 , the interior field can oscillate while the exterior field is evanescent; we construct real quasimodes in this regime and locate nearby resonances by Rouché’s theorem. This is the first time to the author’s knowledge that the case of precise resonances-free strips and whispering-gallery modes for piecewise constant coefficient Maxwell’s equations have been examined. The spectral theory and generalized eigenfunctions here are constructed for arbitrary Lipschitz domains, but for the main theorems we focus on the concrete example of the ball.
For scalar transmission problems, the corresponding distinction between resonance-free regions and resonances approaching the real axis was established by Popov and Vodev [1 , 2 , 17 ] and Cardoso, Popov and Vodev [16 ] . Related results include Galkowski’s quantum Sabine law [3 ] and the wavenumber-explicit estimates and resonance-free regions of Moiola and Spence [4 ] . For Maxwell’s equations, the interface conditions also depend on the impedance, and the two polarizations lead to different scalar radial transmission problems. High-frequency Maxwell estimates in heterogeneous media are obtained in [21 ] ; earlier asymptotic analyses of electromagnetic resonances in spheres and their perturbations include [18 , 19 , 20 ] .
To connect the determinant estimates to the resolvent, we construct normalized generalized eigenfunctions and derive the spectral resolution and resolvent jump. This extends the relation between scattering amplitudes and the T-matrix established for perfectly conducting spheres in [5 ] . Related spectral and scattering constructions appear in [6 , 7 , 8 , 9 , 10 ] . The self-adjoint realization by a quadratic form provides the operator framework for these constructions.
Boundary integral treatments of Maxwell transmission include [11 , 13 , 14 ] ; in particular, Helsing and Rosén use a Dirac formulation on Lipschitz domains. Costabel and Le Louër [15 ] study shape differentiability for dielectric scattering. For the scalar exterior Neumann problem, Galkowski, Marchand and Spence [12 ] give frequency-explicit bounds for a second-kind boundary integral operator and its inverse.
For the operator framework and the generalized eigenfunctions, let Ω ⊂ R 3 \Omega\subset\mathbb{R}^{3} Ω ⊂ R 3 be a bounded Lipschitz open set. The isotropic material coefficients are
ϵ ( x ) = { ϵ 1 , x ∈ Ω , ϵ 0 , x ∉ Ω ‾ , μ ( x ) = { μ 1 , x ∈ Ω , μ 0 , x ∉ Ω ‾ . (1) \epsilon(x)=
\begin{cases}
\epsilon_{1}, & x\in\Omega,\\
\epsilon_{0}, & x\notin\overline{\Omega},
\end{cases}
\qquad
\mu(x)=
\begin{cases}
\mu_{1}, & x\in\Omega,\\
\mu_{0}, & x\notin\overline{\Omega}.
\end{cases} \tag*{(1)} ϵ ( x ) = { ϵ 1 , ϵ 0 , x ∈ Ω , x ∈ / Ω , μ ( x ) = { μ 1 , μ 0 , x ∈ Ω , x ∈ / Ω . (1)
The values on ∂ Ω \partial\Omega ∂ Ω do not affect the operators. The two main resonance theorems and the explicit transmission coefficients below specialize to a ball.
We also assume that 0 < ϵ j < ∞ 0<\epsilon_j<\infty 0 < ϵ j < ∞ and 0 < μ j < ∞ 0<\mu_j<\infty 0 < μ j < ∞ , j = 0 , 1 j=0,1 j = 0 , 1 . We then consider the spectral version of Maxwell’s equations for the electric field E \mathcal{E} E and the magnetic field H \mathcal{H} H as:
curl E − i λ μ H = 0 in R 3 div ( ϵ ⋅ E ) = 0 curl H + i λ ϵ E = 0 div ( μ ⋅ H ) = 0 ν × E i n t = ν × E e x t on ∂ Ω ν × H i n t = ν × H e x t on ∂ Ω . (2) \begin{aligned}
\operatorname{curl}\mathcal{E}-i\lambda\mu\mathcal{H}&=0 &&\text{in }\mathbb{R}^{3}\\
\operatorname{div}(\epsilon\cdot\mathcal{E})&=0\\
\operatorname{curl}\mathcal{H}+i\lambda\epsilon\mathcal{E}&=0\\
\operatorname{div}(\mu\cdot\mathcal{H})&=0\\
\nu\times\mathcal{E}_{\mathrm{int}}&=\nu\times\mathcal{E}_{\mathrm{ext}} &&\text{on }\partial\Omega\\
\nu\times\mathcal{H}_{\mathrm{int}}&=\nu\times\mathcal{H}_{\mathrm{ext}} &&\text{on }\partial\Omega. \tag*{(2)}
\end{aligned} curl E − iλ μ H div ( ϵ ⋅ E ) curl H + iλ ϵ E div ( μ ⋅ H ) ν × E int ν × H int = 0 = 0 = 0 = 0 = ν × E ext = ν × H ext in R 3 on ∂ Ω on ∂ Ω. (2)
The Hilbert space for the electric field is
( L 2 ( R 3 ; ϵ ( x ) d x ) ) 3 , ⟨ u , v ⟩ ( L 2 ( R 3 ; ϵ ( x ) d x ) ) 3 = ∫ R 3 ϵ u ⋅ v ‾ d x . (3) \left(L^{2}(\mathbb{R}^{3};\epsilon(x)\,dx)\right)^{3},
\qquad
\langle u,v\rangle_{\left(L^{2}(\mathbb{R}^{3};\epsilon(x)\,dx)\right)^{3}}
=\int_{\mathbb{R}^{3}}\epsilon u\cdot\overline{v}\,dx. \tag*{(3)} ( L 2 ( R 3 ; ϵ ( x ) d x ) ) 3 , ⟨ u , v ⟩ ( L 2 ( R 3 ; ϵ ( x ) d x ) ) 3 = ∫ R 3 ϵ u ⋅ v d x . (3)
We abbreviate this as ( L 2 ( ϵ d x ) ) 3 \left(L^{2}(\epsilon\,dx)\right)^{3} ( L 2 ( ϵ d x ) ) 3 . Restricting to the divergence-free subspace we have
L ϵ ( div 0 ) = { u ∈ ( L 2 ( ϵ d x ) ) 3 : ∇ ⋅ ( ϵ u ) = 0 } . (4) L_{\epsilon}(\operatorname{div}0)=\left\{u\in\left(L^{2}(\epsilon\,dx)\right)^{3}:\nabla\cdot(\epsilon u)=0\right\}. \tag*{(4)} L ϵ ( div 0 ) = { u ∈ ( L 2 ( ϵ d x ) ) 3 : ∇ ⋅ ( ϵ u ) = 0 } . (4)
It follows that the operator P E P_{\mathcal{E}} P E associated to the Maxwell system electric field acting on 3 dimensional vectors u u u is
P E u = ϵ − 1 ∇ × μ − 1 ∇ × u . (5) P_{\mathcal{E}}u=\epsilon^{-1}\nabla\times\mu^{-1}\nabla\times u. \tag*{(5)} P E u = ϵ − 1 ∇ × μ − 1 ∇ × u . (5)
The boundary conditions for the electric field for this system can also be recast as:
ν × E i n t = ν × E e x t ν × μ 1 − 1 ∇ × E i n t = ν × μ 0 − 1 ∇ × E e x t on ∂ Ω . (6) \begin{aligned}
\nu\times\mathcal{E}_{\mathrm{int}} &= \nu\times\mathcal{E}_{\mathrm{ext}} \\
\nu\times\mu_1^{-1}\nabla\times\mathcal{E}_{\mathrm{int}} &= \nu\times\mu_0^{-1}\nabla\times\mathcal{E}_{\mathrm{ext}} \quad\text{on } \partial\Omega. \tag*{(6)}
\end{aligned} ν × E int ν × μ 1 − 1 ∇ × E int = ν × E ext = ν × μ 0 − 1 ∇ × E ext on ∂ Ω. (6)
We use H ( curl , D ) = { u ∈ L 2 ( D ; C 3 ) : curl u ∈ L 2 ( D ; C 3 ) } H(\operatorname{curl},D)=\{u\in L^2(D;\mathbb{C}^3):\operatorname{curl}u\in L^2(D;\mathbb{C}^3)\} H ( curl , D ) = { u ∈ L 2 ( D ; C 3 ) : curl u ∈ L 2 ( D ; C 3 )} , D = Ω , Ω c D=\Omega,\Omega^c D = Ω , Ω c or R 3 \mathbb{R}^3 R 3 , and the weighted orthogonal Helmholtz decomposition
( L 2 ( ϵ d x ) ) 3 = L ϵ ( div 0 ) ⊕ ∇ H ˙ 1 ( R 3 ) . (7) (L^2(\epsilon\,dx))^3=L_\epsilon(\operatorname{div}0)\oplus\nabla\dot{H}^1(\mathbb{R}^3). \tag*{(7)} ( L 2 ( ϵ d x ) ) 3 = L ϵ ( div 0 ) ⊕ ∇ H ˙ 1 ( R 3 ) . (7)
Here H ˙ 1 ( R 3 ) \dot{H}^1(\mathbb{R}^3) H ˙ 1 ( R 3 ) is the homogeneous Sobolev space with square-integrable gradient. The transmission conditions are encoded by the domain
D ( P E ) = { u ∈ L ϵ ( div 0 ) ∩ H ( curl , R 3 ) : μ − 1 curl u ∈ H ( curl , R 3 ) } . D(P_E)=\{u\in L_\epsilon(\operatorname{div}0)\cap H(\operatorname{curl},\mathbb{R}^3):\mu^{-1}\operatorname{curl}u\in H(\operatorname{curl},\mathbb{R}^3)\}. D ( P E ) = { u ∈ L ϵ ( div 0 ) ∩ H ( curl , R 3 ) : μ − 1 curl u ∈ H ( curl , R 3 )} .
Indeed, global H ( curl ) H(\operatorname{curl}) H ( curl ) regularity is equivalent to matching tangential traces across the interface; see [22 ] . Let ν \nu ν be the outward pointing normal vector to Ω \Omega Ω . We define the jump as [ [ F ] ] = ( F i n t − F e x t ) ∣ ∂ Ω [[F]]=(F^{\mathrm{int}}-F^{\mathrm{ext}})|_{\partial\Omega} [[ F ]] = ( F int − F ext ) ∣ ∂ Ω . For a piecewise smooth field G G G , with components G 0 , G 1 G_0,G_1 G 0 , G 1 , the curl Green identity with a compactly supported test field v v v reads
∫ Ω curl G 1 ⋅ v ‾ d x + ∫ R 3 ∖ Ω ‾ curl G 0 ⋅ v ‾ d x = ∫ R 3 G ⋅ curl v ‾ d x + ⟨ [ [ ν × G ] ] , v T ⟩ . (8) \int_{\Omega}\operatorname{curl}G_1\cdot\overline{v}\,dx+\int_{\mathbb{R}^3\setminus\overline{\Omega}}\operatorname{curl}G_0\cdot\overline{v}\,dx=\int_{\mathbb{R}^3}G\cdot\overline{\operatorname{curl}v}\,dx+\langle[[\nu\times G]],v_T\rangle. \tag*{(8)} ∫ Ω curl G 1 ⋅ v d x + ∫ R 3 ∖ Ω curl G 0 ⋅ v d x = ∫ R 3 G ⋅ curl v d x + ⟨[[ ν × G ]] , v T ⟩ . (8)
Here v T v_T v T is the tangential trace; the identity extends to piecewise H ( curl ) H(\operatorname{curl}) H ( curl ) fields by the H − 1 / 2 ( Div ∂ Ω ) H^{-1/2}(\operatorname{Div}_{\partial\Omega}) H − 1/2 ( Div ∂ Ω ) –H − 1 / 2 ( Curl ∂ Ω ) H^{-1/2}(\operatorname{Curl}_{\partial\Omega}) H − 1/2 ( Curl ∂ Ω ) trace duality, defined by this Green identity, [22 ] . All boundary pairings below use this duality, conjugate-linear in the second argument.
Proposition 1.1. The operator P E P_E P E is a non-negative self-adjoint operator on L ϵ ( div 0 ) L_\epsilon(\operatorname{div}0) L ϵ ( div 0 ) , i.e., let V E V_E V E be the space
V E = { u ∈ ( L 2 ( R 3 ; ϵ ( x ) d x ) ) 3 : curl u ∈ ( L 2 ( R 3 ) ) 3 div ( ϵ u ) = 0 } (9) V_E=\{u\in(L^2(\mathbb{R}^3;\epsilon(x)\,dx))^3:\operatorname{curl}u\in(L^2(\mathbb{R}^3))^3\ \operatorname{div}(\epsilon u)=0\} \tag*{(9)} V E = { u ∈ ( L 2 ( R 3 ; ϵ ( x ) d x ) ) 3 : curl u ∈ ( L 2 ( R 3 ) ) 3 div ( ϵ u ) = 0 } (9) and we define the corresponding quadratic form as:
q E ( u , v ) = ∫ R 3 μ − 1 ( curl u ) ⋅ ( curl v ) ‾ d x . (10) q_E(u,v)=\int_{\mathbb{R}^3}\mu^{-1}(\operatorname{curl}u)\cdot\overline{(\operatorname{curl}v)}\,dx. \tag*{(10)} q E ( u , v ) = ∫ R 3 μ − 1 ( curl u ) ⋅ ( curl v ) d x . (10) Then q E q_E q E is a densely defined, closed, non-negative symmetric form on L ϵ ( div 0 ) L_\epsilon(\operatorname{div}0) L ϵ ( div 0 ) and the self-adjoint operator P E P_E P E , is realized by q E ( u , u ) = ∥ P E 1 / 2 u ∥ ( L 2 ( ϵ d x ) ) 3 2 q_E(u,u)=\|P_E^{1/2}u\|^2_{(L^2(\epsilon\,dx))^3} q E ( u , u ) = ∥ P E 1/2 u ∥ ( L 2 ( ϵ d x ) ) 3 2 .
Proof. The form is symmetric and non-negative. The squared form norm q E ( u , u ) + ∥ u ∥ L 2 ( ϵ d x ) 3 2 q_E(u,u)+\|u\|^2_{L^2(\epsilon\,dx)^3} q E ( u , u ) + ∥ u ∥ L 2 ( ϵ d x ) 3 2 is equivalent to the squared H ( curl ) H(\operatorname{curl}) H ( curl ) norm, and the divergence constraint is closed, so q E q_E q E is closed. Let Π \Pi Π be the weighted orthogonal projection in (1.7). Since Π v = v − ∇ p \Pi v=v-\nabla p Π v = v − ∇ p and curl Π v = curl v \operatorname{curl}\Pi v=\operatorname{curl}v curl Π v = curl v , it maps C 0 ∞ C^\infty_0 C 0 ∞ into V E V_E V E . Boundedness of Π \Pi Π and density of C 0 ∞ C^\infty_0 C 0 ∞ in L 2 L^2 L 2 show that V E V_E V E is dense in L ϵ ( div 0 ) L_\epsilon(\operatorname{div}0) L ϵ ( div 0 ) .
The representation theorem [23 Theorem VIII.15] gives a non-negative self-adjoint operator A A A . For u ∈ D ( P E ) u\in D(P_E) u ∈ D ( P E ) , the global curl Green identity above gives q E ( u , v ) = ⟨ P E u , v ⟩ ( L 2 ( ϵ d x ) ) 3 q_E(u,v)=\langle P_Eu,v\rangle_{(L^2(\epsilon\,dx))^3} q E ( u , v ) = ⟨ P E u , v ⟩ ( L 2 ( ϵ d x ) ) 3 for all v ∈ V E v\in V_E v ∈ V E , hence P E ⊂ A P_E\subset A P E ⊂ A . Conversely, if u ∈ D ( A ) u\in D(A) u ∈ D ( A ) and g = A u g=Au g = A u , then
q E ( u , v ) = ⟨ g , v ⟩ ( L 2 ( ϵ d x ) ) 3 , v ∈ V E . q_E(u,v)=\langle g,v\rangle_{(L^2(\epsilon\,dx))^3},\qquad v\in V_E. q E ( u , v ) = ⟨ g , v ⟩ ( L 2 ( ϵ d x ) ) 3 , v ∈ V E .
Both sides vanish on gradients, so (1.7) extends this identity to all v ∈ H ( curl , R 3 ) v\in H(\operatorname{curl},\mathbb{R}^3) v ∈ H ( curl , R 3 ) . Thus curl ( μ − 1 curl u ) = ϵ g \operatorname{curl}(\mu^{-1}\operatorname{curl}u)=\epsilon g curl ( μ − 1 curl u ) = ϵ g distributionally on R 3 \mathbb{R}^3 R 3 . Since μ − 1 curl u ∈ L 2 \mu^{-1}\operatorname{curl}u\in L^2 μ − 1 curl u ∈ L 2 , it belongs to H ( curl ) H(\operatorname{curl}) H ( curl ) , so u ∈ D ( P E ) u\in D(P_E) u ∈ D ( P E ) and P E u = g P_Eu=g P E u = g . Therefore A = P E A=P_E A = P E .
Corollary 1.2. The magnetic operator P H = μ − 1 curl ϵ − 1 curl P_H=\mu^{-1}\operatorname{curl}\epsilon^{-1}\operatorname{curl} P H = μ − 1 curl ϵ − 1 curl on L μ ( div 0 ) L_\mu(\operatorname{div}0) L μ ( div 0 ) with corresponding quadratic form q H ( u , v ) = ∫ R 3 ϵ − 1 curl u ⋅ curl v d x q_H(u,v)=\int_{\mathbb{R}^3}\epsilon^{-1}\operatorname{curl}u\cdot\operatorname{curl}v\,\mathrm{d}x q H ( u , v ) = ∫ R 3 ϵ − 1 curl u ⋅ curl v d x is self adjoint.
Proof. Interchange ϵ \epsilon ϵ and μ \mu μ in the preceding proof. The form domain is V H = H ( curl , R 3 ) ∩ L μ ( div 0 ) V_H=H(\operatorname{curl},\mathbb{R}^3)\cap L_\mu(\operatorname{div}0) V H = H ( curl , R 3 ) ∩ L μ ( div 0 ) .
Let Π E \Pi_E Π E and Π H \Pi_H Π H be the weighted orthogonal projections onto L ϵ ( div 0 ) L_\epsilon(\operatorname{div}0) L ϵ ( div 0 ) and L μ ( div 0 ) L_\mu(\operatorname{div}0) L μ ( div 0 ) . Non-negativity and self-adjointness give
R λ , E = ( P E − λ 2 ) − 1 , R λ , H = ( P H − λ 2 ) − 1 , Im λ > 0. (11) R_{\lambda,E}=(P_E-\lambda^2)^{-1},\qquad R_{\lambda,H}=(P_H-\lambda^2)^{-1},\qquad\operatorname{Im}\lambda>0. \tag*{(11)} R λ , E = ( P E − λ 2 ) − 1 , R λ , H = ( P H − λ 2 ) − 1 , Im λ > 0. (11)
For real λ ≠ 0 \lambda\ne0 λ = 0 , we use their boundary values from the upper half-plane, established below, as operators from compactly supported data to L l o c 2 L^2_{\mathrm{loc}} L loc 2 . To allow scalar cutoffs, we extend the electric resolvent to all of ( L 2 ( ϵ d x ) ) 3 (L^2(\epsilon\,\mathrm{d}x))^3 ( L 2 ( ϵ d x ) ) 3 by
R ~ λ , E = R λ , E Π E − λ − 2 ( Id − Π E ) . (12) \widetilde{R}_{\lambda,E}=R_{\lambda,E}\Pi_E-\lambda^{-2}(\operatorname{Id}-\Pi_E). \tag*{(12)} R λ , E = R λ , E Π E − λ − 2 ( Id − Π E ) . (12)
The second term is the resolvent on the gradient summand, where the curl–curl operator vanishes. The magnetic extension is defined analogously. We make the following definition which is adapted from [29 ] for our setting:
Definition 1.3. A vector u ∈ H l o c ( curl ) u\in H_{\mathrm{loc}}(\operatorname{curl}) u ∈ H loc ( curl ) is a resonance state corresponding to λ ∈ C ∖ { 0 } \lambda\in\mathbb{C}\setminus\{0\} λ ∈ C ∖ { 0 } if and only if ( P E − λ 2 ) u = 0 (P_E-\lambda^2)u=0 ( P E − λ 2 ) u = 0 (resp. ( P H − λ 2 ) u = 0 (P_H-\lambda^2)u=0 ( P H − λ 2 ) u = 0 ) and there exists g ∈ ( L 0 2 ( R 3 ) ) 3 g\in(L^2_0(\mathbb{R}^3))^3 g ∈ ( L 0 2 ( R 3 ) ) 3 , R > 0 R>0 R > 0 with Ω ‾ ⊂ B R \overline{\Omega}\subset B_R Ω ⊂ B R such that
u ∣ R 3 ∖ B ( 0 , R ) = R 0 , λ g ∣ R 3 ∖ B ( 0 , R ) (13) u|_{\mathbb{R}^3\setminus B(0,R)}=R_{0,\lambda}g|_{\mathbb{R}^3\setminus B(0,R)} \tag*{(13)} u ∣ R 3 ∖ B ( 0 , R ) = R 0 , λ g ∣ R 3 ∖ B ( 0 , R ) (13) i.e. u u u is purely outgoing . Here R 0 , λ = ( P 0 − λ 2 ) − 1 R_{0,\lambda}=(P_0-\lambda^2)^{-1} R 0 , λ = ( P 0 − λ 2 ) − 1 denotes the free constant coefficient ϵ 1 = ϵ 0 \epsilon_1=\epsilon_0 ϵ 1 = ϵ 0 , μ 1 = μ 0 \mu_1=\mu_0 μ 1 = μ 0 outgoing resolvent.
For the following two theorems we take Ω = B r 0 = { x ∈ R 3 : ∣ x ∣ < r 0 } \Omega=B_{r_0}=\{x\in\mathbb{R}^3:|x|<r_0\} Ω = B r 0 = { x ∈ R 3 : ∣ x ∣ < r 0 } and m 0 = ϵ 1 μ 1 / ( ϵ 0 μ 0 ) m_0=\sqrt{\epsilon_1\mu_1/(\epsilon_0\mu_0)} m 0 = ϵ 1 μ 1 / ( ϵ 0 μ 0 ) . The two cases for m 0 m_0 m 0 give the following dichotomy.
Theorem 1.4. Let 0 < m 0 ≤ 1 0<m_0\le1 0 < m 0 ≤ 1 then there exist positive C 1 , C 2 C_1,C_2 C 1 , C 2 depending on μ 0 , μ 1 , ϵ 0 , ϵ 1 \mu_0,\mu_1,\epsilon_0,\epsilon_1 μ 0 , μ 1 , ϵ 0 , ϵ 1 and r 0 r_0 r 0 such that for every smooth compactly supported (scalar) function χ \chi χ which is one in a neighborhood of the ball, χ R ~ λ , E χ \chi\widetilde{R}_{\lambda,E}\chi χ R λ , E χ and χ R ~ λ , H χ \chi\widetilde{R}_{\lambda,H}\chi χ R λ , H χ are holomorphic in
{ λ ∈ C : ∣ Re ( λ ) ∣ ≥ C 1 , Im λ ≥ − C 2 } . (14) \{\lambda\in\mathbb{C}:|\operatorname{Re}(\lambda)|\ge C_1,\quad\operatorname{Im}\lambda\ge-C_2\}. \tag*{(14)} { λ ∈ C : ∣ Re ( λ ) ∣ ≥ C 1 , Im λ ≥ − C 2 } . (14)
Furthermore in the case in which ϵ 1 μ 1 > ϵ 0 μ 0 \epsilon_1\mu_1>\epsilon_0\mu_0 ϵ 1 μ 1 > ϵ 0 μ 0 we have resonances, which manifest themselves in the following theorem.
Theorem 1.5. Let 1 < m 0 1<m_0 1 < m 0 , then no resonance free strip exists. There exist constants C , c > 0 C,c>0 C , c > 0 which depend on the material parameters m 0 , μ 0 , μ 1 , ϵ 0 , ϵ 1 m_0,\mu_0,\mu_1,\epsilon_0,\epsilon_1 m 0 , μ 0 , μ 1 , ϵ 0 , ϵ 1 and r 0 r_0 r 0 as well as a sequence of resonances { λ j } j ∈ N \{\lambda_j\}_{j\in\mathbb{N}} { λ j } j ∈ N of the operator P E P_E P E such that
Re λ j → ∞ Im λ j < 0 ∣ Im λ j ∣ ≤ C e − c Re λ j . (15) \operatorname{Re}\lambda_j\to\infty\quad\operatorname{Im}\lambda_j<0\quad|\operatorname{Im}\lambda_j|\le Ce^{-c\operatorname{Re}\lambda_j}. \tag*{(15)} Re λ j → ∞ Im λ j < 0 ∣ Im λ j ∣ ≤ C e − c Re λ j . (15)
Vector spherical harmonics and the T-matrix
Let Y ℓ m Y_{\ell m} Y ℓ m , ℓ ≥ 0 \ell\ge0 ℓ ≥ 0 , ∣ m ∣ ≤ ℓ |m|\le\ell ∣ m ∣ ≤ ℓ , be a real orthonormal basis of scalar spherical harmonics, extended homogeneously of degree zero to x = r r ^ x=r\hat{r} x = r r ^ . For ℓ ≥ 1 \ell\ge1 ℓ ≥ 1 set
Ψ 1 , ℓ , m = ∇ Y ℓ m × x ℓ ( ℓ + 1 ) , Ψ 2 , ℓ , m = r ∇ Y ℓ m ℓ ( ℓ + 1 ) , Ψ 3 , ℓ , m = r ^ Y ℓ m . (16) \Psi_{1,\ell,m}=\frac{\nabla Y_{\ell m}\times x}{\sqrt{\ell(\ell+1)}},\qquad
\Psi_{2,\ell,m}=\frac{r\nabla Y_{\ell m}}{\sqrt{\ell(\ell+1)}},\qquad
\Psi_{3,\ell,m}=\hat{r}Y_{\ell m}. \tag*{(16)} Ψ 1 , ℓ , m = ℓ ( ℓ + 1 ) ∇ Y ℓ m × x , Ψ 2 , ℓ , m = ℓ ( ℓ + 1 ) r ∇ Y ℓ m , Ψ 3 , ℓ , m = r ^ Y ℓ m . (16)
The first two families form an orthonormal basis of
L t 2 ( S 2 ; C 3 ) = { F ∈ L 2 ( S 2 ; C 3 ) : r ^ ⋅ F = 0 } . (17) L^2_t(\mathbb{S}^2;\mathbb{C}^3)=\{F\in L^2(\mathbb{S}^2;\mathbb{C}^3):\hat{r}\cdot F=0\}. \tag*{(17)} L t 2 ( S 2 ; C 3 ) = { F ∈ L 2 ( S 2 ; C 3 ) : r ^ ⋅ F = 0 } . (17)
Together with the radial family, including Ψ 3 , 0 , 0 \Psi_{3,0,0} Ψ 3 , 0 , 0 , they form an orthonormal basis of L 2 ( S 2 ; C 3 ) L^2(\mathbb{S}^2;\mathbb{C}^3) L 2 ( S 2 ; C 3 ) . We write Y n = Ψ j , ℓ , m Y_n=\Psi_{j,\ell,m} Y n = Ψ j , ℓ , m with n = ( j , ℓ , m ) n=(j,\ell,m) n = ( j , ℓ , m ) on the unit sphere. The components of Ψ 1 , ℓ , m \Psi_{1,\ell,m} Ψ 1 , ℓ , m have harmonic degree ℓ \ell ℓ , whereas those of Ψ 2 , ℓ , m \Psi_{2,\ell,m} Ψ 2 , ℓ , m and Ψ 3 , ℓ , m \Psi_{3,\ell,m} Ψ 3 , ℓ , m combine degrees ℓ − 1 \ell-1 ℓ − 1 and ℓ + 1 \ell+1 ℓ + 1 ; see [22 Theorem 2.46 and Corollary 2.47] .
For z ℓ = j ℓ z_\ell=j_\ell z ℓ = j ℓ , h ℓ ( 1 ) h_\ell^{(1)} h ℓ ( 1 ) or h ℓ ( 2 ) h_\ell^{(2)} h ℓ ( 2 ) , the Hansen fields are
M ℓ m ( k r ) = curl ( x z ℓ ( k r ) Y ℓ m ) = ℓ ( ℓ + 1 ) z ℓ ( k r ) Ψ 1 , ℓ , m , N ℓ m ( k r ) = k − 1 curl M ℓ m ( k r ) = ℓ ( ℓ + 1 ) k r z ℓ ( k r ) Ψ 3 , ℓ , m + ℓ ( ℓ + 1 ) k r ( y z ℓ ( y ) ) ′ ∣ y = k r Ψ 2 , ℓ , m . \begin{aligned}
M_{\ell m}(kr)&=\operatorname{curl}(xz_\ell(kr)Y_{\ell m})=\sqrt{\ell(\ell+1)}z_\ell(kr)\Psi_{1,\ell,m},\\
N_{\ell m}(kr)&=k^{-1}\operatorname{curl}M_{\ell m}(kr)\\
&=\frac{\ell(\ell+1)}{kr}z_\ell(kr)\Psi_{3,\ell,m}+\frac{\sqrt{\ell(\ell+1)}}{kr}(yz_\ell(y))'\big|_{y=kr}\Psi_{2,\ell,m}.
\end{aligned} M ℓ m ( k r ) N ℓ m ( k r ) = curl ( x z ℓ ( k r ) Y ℓ m ) = ℓ ( ℓ + 1 ) z ℓ ( k r ) Ψ 1 , ℓ , m , = k − 1 curl M ℓ m ( k r ) = k r ℓ ( ℓ + 1 ) z ℓ ( k r ) Ψ 3 , ℓ , m + k r ℓ ( ℓ + 1 ) ( y z ℓ ( y ) ) ′ y = k r Ψ 2 , ℓ , m .
They satisfy curl M = k N \operatorname{curl}M=kN curl M = k N , curl N = k M \operatorname{curl}N=kM curl N = k M and div M = div N = 0 \operatorname{div}M=\operatorname{div}N=0 div M = div N = 0 . Set
k j = λ ϵ j μ j , η j = μ j ϵ j , j = 0 , 1. (18) k_j=\lambda\sqrt{\epsilon_j\mu_j},\qquad\eta_j=\sqrt{\frac{\mu_j}{\epsilon_j}},\qquad j=0,1. \tag*{(18)} k j = λ ϵ j μ j , η j = ϵ j μ j , j = 0 , 1. (18)
Either field with k = k j k=k_j k = k j gives an electric solution in medium j j j , with magnetic field H = ( i λ μ j ) − 1 curl E H=(i\lambda\mu_j)^{-1}\operatorname{curl}E H = ( iλ μ j ) − 1 curl E . These identities follow from the scalar Bessel equation; see [22 Lemma 2.42 and Theorem 2.43] and [24 ] . Unlike the PEC construction in [5 ] , the transmission eigenfunctions are defined on all of R 3 \mathbb{R}^3 R 3 .
We now define the generalized eigenfunctions piecewise for B r 0 B_{r_0} B r 0 . Here and below x j = k j r 0 x_j=k_jr_0 x j = k j r 0 for j = 0 , 1 j=0,1 j = 0 , 1 . The superscript 0 0 0 refers to j ℓ j_\ell j ℓ , the superscript − - − refers to h ℓ ( 1 ) h_\ell^{(1)} h ℓ ( 1 ) in the following:
E λ , 1 ( Y n ) = k 0 2 η 0 π ℓ ( ℓ + 1 ) { t ℓ 1 ( λ ) M ℓ m 0 ( k 1 r ) r < r 0 M ℓ m 0 ( k 0 r ) + A ℓ 1 ( λ ) M ℓ m − ( k 0 r ) r > r 0 (19) E_{\lambda,1}(Y_n)=k_0\sqrt{\frac{2\eta_0}{\pi\ell(\ell+1)}}\begin{cases}
t_\ell^1(\lambda)M_{\ell m}^0(k_1r)&r<r_0\\
M_{\ell m}^0(k_0r)+A_\ell^1(\lambda)M_{\ell m}^-(k_0r)&r>r_0
\end{cases} \tag*{(19)} E λ , 1 ( Y n ) = k 0 π ℓ ( ℓ + 1 ) 2 η 0 { t ℓ 1 ( λ ) M ℓ m 0 ( k 1 r ) M ℓ m 0 ( k 0 r ) + A ℓ 1 ( λ ) M ℓ m − ( k 0 r ) r < r 0 r > r 0 (19)
and
E λ , 2 ( Y n ) = k 0 2 η 0 π ℓ ( ℓ + 1 ) { t ℓ 2 ( λ ) N ℓ m 0 ( k 1 r ) r < r 0 N ℓ m 0 ( k 0 r ) + A ℓ 2 ( λ ) N ℓ m − ( k 0 r ) r > r 0 . (20) E_{\lambda,2}(Y_n)=k_0\sqrt{\frac{2\eta_0}{\pi\ell(\ell+1)}}\begin{cases}
t_\ell^2(\lambda)N_{\ell m}^0(k_1r)&r<r_0\\
N_{\ell m}^0(k_0r)+A_\ell^2(\lambda)N_{\ell m}^-(k_0r)&r>r_0.
\end{cases} \tag*{(20)} E λ , 2 ( Y n ) = k 0 π ℓ ( ℓ + 1 ) 2 η 0 { t ℓ 2 ( λ ) N ℓ m 0 ( k 1 r ) N ℓ m 0 ( k 0 r ) + A ℓ 2 ( λ ) N ℓ m − ( k 0 r ) r < r 0 r > r 0 . (20)
Here n = ( j , ℓ , m ) n=(j,\ell,m) n = ( j , ℓ , m ) in E λ , j ( Y n ) E_{\lambda,j}(Y_n) E λ , j ( Y n ) . The prefactors give delta normalization in L 2 ( ϵ d x ) 3 L^2(\epsilon\,\mathrm{d}x)^3 L 2 ( ϵ d x ) 3 , with orthogonal polarizations; this is established in Section 4. Set ν = ℓ + 1 2 \nu=\ell+\frac{1}{2} ν = ℓ + 2 1 and introduce the Riccati–Bessel functions
ψ ℓ ( y ) = y j ℓ ( y ) = π y 2 J ν ( y ) , \psi_\ell(y)=yj_\ell(y)=\sqrt{\frac{\pi y}{2}}J_\nu(y), ψ ℓ ( y ) = y j ℓ ( y ) = 2 π y J ν ( y ) ,
ξ ℓ ( y ) = y h ℓ ( 1 ) ( y ) = π y 2 H ν ( 1 ) ( y ) . (21) \xi_\ell(y)=yh_\ell^{(1)}(y)=\sqrt{\frac{\pi y}{2}}H_\nu^{(1)}(y). \tag*{(21)} ξ ℓ ( y ) = y h ℓ ( 1 ) ( y ) = 2 π y H ν ( 1 ) ( y ) . (21)
Primes denote differentiation with respect to the full argument.
The transverse electric coefficient A ℓ 1 A_\ell^1 A ℓ 1
Tangential continuity of E \mathcal{E} E and H \mathcal{H} H gives the TE matching system [24 ]
t ℓ 1 j ℓ ( x 1 ) = j ℓ ( x 0 ) + A ℓ 1 h ℓ ( 1 ) ( x 0 ) (22) t_\ell^1j_\ell(x_1)=j_\ell(x_0)+A_\ell^1h_\ell^{(1)}(x_0) \tag*{(22)} t ℓ 1 j ℓ ( x 1 ) = j ℓ ( x 0 ) + A ℓ 1 h ℓ ( 1 ) ( x 0 ) (22)
t ℓ 1 μ 1 ψ ℓ ′ ( x 1 ) = 1 μ 0 ( ψ ℓ ′ ( x 0 ) + A ℓ 1 ξ ℓ ′ ( x 0 ) ) . (23) \frac{t_\ell^1}{\mu_1}\psi_\ell'(x_1)=\frac{1}{\mu_0}\left(\psi_\ell'(x_0)+A_\ell^1\xi_\ell'(x_0)\right). \tag*{(23)} μ 1 t ℓ 1 ψ ℓ ′ ( x 1 ) = μ 0 1 ( ψ ℓ ′ ( x 0 ) + A ℓ 1 ξ ℓ ′ ( x 0 ) ) . (23)
It follows that
A ℓ 1 ( λ ) = η 1 ψ ℓ ( x 1 ) ψ ℓ ′ ( x 0 ) − η 0 ψ ℓ ( x 0 ) ψ ℓ ′ ( x 1 ) η 0 ξ ℓ ( x 0 ) ψ ℓ ′ ( x 1 ) − η 1 ψ ℓ ( x 1 ) ξ ℓ ′ ( x 0 ) . (24) A_{\ell}^{1}(\lambda)=\frac{\eta_{1}\psi_{\ell}(x_{1})\psi_{\ell}'(x_{0})-\eta_{0}\psi_{\ell}(x_{0})\psi_{\ell}'(x_{1})}{\eta_{0}\xi_{\ell}(x_{0})\psi_{\ell}'(x_{1})-\eta_{1}\psi_{\ell}(x_{1})\xi_{\ell}'(x_{0})}. \tag*{(24)} A ℓ 1 ( λ ) = η 0 ξ ℓ ( x 0 ) ψ ℓ ′ ( x 1 ) − η 1 ψ ℓ ( x 1 ) ξ ℓ ′ ( x 0 ) η 1 ψ ℓ ( x 1 ) ψ ℓ ′ ( x 0 ) − η 0 ψ ℓ ( x 0 ) ψ ℓ ′ ( x 1 ) . (24)
The interior amplitude is
t ℓ 1 ( λ ) = − i μ 1 x 0 μ 0 h ℓ ( 1 ) ( x 0 ) ψ ℓ ′ ( x 1 ) − μ 1 j ℓ ( x 1 ) ξ ℓ ′ ( x 0 ) . (25) t_{\ell}^{1}(\lambda)=-\frac{\mathrm{i}\mu_{1}}{x_{0}\mu_{0}h_{\ell}^{(1)}(x_{0})\psi_{\ell}'(x_{1})-\mu_{1}j_{\ell}(x_{1})\xi_{\ell}'(x_{0})}. \tag*{(25)} t ℓ 1 ( λ ) = − x 0 μ 0 h ℓ ( 1 ) ( x 0 ) ψ ℓ ′ ( x 1 ) − μ 1 j ℓ ( x 1 ) ξ ℓ ′ ( x 0 ) i μ 1 . (25)
The interior amplitude follows from the Wronskian (B.2).
The transverse magnetic coefficient A ℓ 2 A_{\ell}^{2} A ℓ 2
The TM matching system is
t ℓ 2 k 1 ψ ℓ ′ ( x 1 ) = 1 k 0 ( ψ ℓ ′ ( x 0 ) + A ℓ 2 ξ ℓ ′ ( x 0 ) ) . (26) \frac{t_{\ell}^{2}}{k_{1}}\psi_{\ell}'(x_{1})=\frac{1}{k_{0}}\left(\psi_{\ell}'(x_{0})+A_{\ell}^{2}\xi_{\ell}'(x_{0})\right). \tag*{(26)} k 1 t ℓ 2 ψ ℓ ′ ( x 1 ) = k 0 1 ( ψ ℓ ′ ( x 0 ) + A ℓ 2 ξ ℓ ′ ( x 0 ) ) . (26)
k 1 t ℓ 2 μ 1 j ℓ ( x 1 ) = k 0 μ 0 ( j ℓ ( x 0 ) + A ℓ 2 h ℓ ( 1 ) ( x 0 ) ) . (27) \frac{k_{1}t_{\ell}^{2}}{\mu_{1}}j_{\ell}(x_{1})=\frac{k_{0}}{\mu_{0}}\left(j_{\ell}(x_{0})+A_{\ell}^{2}h_{\ell}^{(1)}(x_{0})\right). \tag*{(27)} μ 1 k 1 t ℓ 2 j ℓ ( x 1 ) = μ 0 k 0 ( j ℓ ( x 0 ) + A ℓ 2 h ℓ ( 1 ) ( x 0 ) ) . (27)
We solve the system to obtain
A ℓ 2 ( λ ) = η 1 ψ ℓ ( x 0 ) ψ ℓ ′ ( x 1 ) − η 0 ψ ℓ ( x 1 ) ψ ℓ ′ ( x 0 ) η 0 ψ ℓ ( x 1 ) ξ ℓ ′ ( x 0 ) − η 1 ξ ℓ ( x 0 ) ψ ℓ ′ ( x 1 ) . (28) A_{\ell}^{2}(\lambda)=\frac{\eta_{1}\psi_{\ell}(x_{0})\psi_{\ell}'(x_{1})-\eta_{0}\psi_{\ell}(x_{1})\psi_{\ell}'(x_{0})}{\eta_{0}\psi_{\ell}(x_{1})\xi_{\ell}'(x_{0})-\eta_{1}\xi_{\ell}(x_{0})\psi_{\ell}'(x_{1})}. \tag*{(28)} A ℓ 2 ( λ ) = η 0 ψ ℓ ( x 1 ) ξ ℓ ′ ( x 0 ) − η 1 ξ ℓ ( x 0 ) ψ ℓ ′ ( x 1 ) η 1 ψ ℓ ( x 0 ) ψ ℓ ′ ( x 1 ) − η 0 ψ ℓ ( x 1 ) ψ ℓ ′ ( x 0 ) . (28)
and also
t ℓ 2 ( λ ) = i k 0 k 1 μ 1 / x 0 k 1 2 μ 0 j ℓ ( x 1 ) ξ ℓ ′ ( x 0 ) − k 0 2 μ 1 h ℓ ( 1 ) ( x 0 ) ψ ℓ ′ ( x 1 ) . (29) t_{\ell}^{2}(\lambda)=\frac{\mathrm{i}k_{0}k_{1}\mu_{1}/x_{0}}{k_{1}^{2}\mu_{0}j_{\ell}(x_{1})\xi_{\ell}'(x_{0})-k_{0}^{2}\mu_{1}h_{\ell}^{(1)}(x_{0})\psi_{\ell}'(x_{1})}. \tag*{(29)} t ℓ 2 ( λ ) = k 1 2 μ 0 j ℓ ( x 1 ) ξ ℓ ′ ( x 0 ) − k 0 2 μ 1 h ℓ ( 1 ) ( x 0 ) ψ ℓ ′ ( x 1 ) i k 0 k 1 μ 1 / x 0 . (29)
Lemma 2.1. For λ > 0 \lambda>0 λ > 0 , let D ℓ j ( λ ) ≠ 0 D_{\ell}^{j}(\lambda)\ne0 D ℓ j ( λ ) = 0 be the denominators of A ℓ j ( λ ) A_{\ell}^{j}(\lambda) A ℓ j ( λ ) . Define S ( λ ) S(\lambda) S ( λ ) diagonally in the vector spherical harmonic basis by
S ( λ ) Y j , ℓ , m = S ℓ j ( λ ) Y j , ℓ , m , S ℓ j ( λ ) = 1 + 2 A ℓ j ( λ ) = − D ℓ j ( λ ) ‾ D ℓ j ( λ ) . (30) S(\lambda)Y_{j,\ell,m}=S_{\ell}^{j}(\lambda)Y_{j,\ell,m},\qquad S_{\ell}^{j}(\lambda)=1+2A_{\ell}^{j}(\lambda)=-\frac{\overline{D_{\ell}^{j}(\lambda)}}{D_{\ell}^{j}(\lambda)}. \tag*{(30)} S ( λ ) Y j , ℓ , m = S ℓ j ( λ ) Y j , ℓ , m , S ℓ j ( λ ) = 1 + 2 A ℓ j ( λ ) = − D ℓ j ( λ ) D ℓ j ( λ ) . (30) In particular, S ( λ ) S(\lambda) S ( λ ) is unitary.
Proof. Write A ℓ j = N ℓ j / D ℓ j A_{\ell}^{j}=N_{\ell}^{j}/D_{\ell}^{j} A ℓ j = N ℓ j / D ℓ j using the preceding formulas. For real λ > 0 \lambda>0 λ > 0 , the numerators N ℓ j N_{\ell}^{j} N ℓ j are real and ξ ℓ + ξ ℓ ‾ = 2 ψ ℓ \xi_{\ell}+\overline{\xi_{\ell}}=2\psi_{\ell} ξ ℓ + ξ ℓ = 2 ψ ℓ , with the same identity for their derivatives. Hence D ℓ j ‾ + D ℓ j = − 2 N ℓ j \overline{D_{\ell}^{j}}+D_{\ell}^{j}=-2N_{\ell}^{j} D ℓ j + D ℓ j = − 2 N ℓ j , which gives the stated identity and unitarity.
We set
J ~ λ ( Y n ) ( k 0 r , θ , ϕ ) = k 0 2 η 0 π ℓ ( ℓ + 1 ) M ℓ m 0 ( k 0 r ) (31) \widetilde{J}_{\lambda}(Y_{n})(k_{0}r,\theta,\phi)=k_{0}\sqrt{\frac{2\eta_{0}}{\pi\ell(\ell+1)}}M_{\ell m}^{0}(k_{0}r) \tag*{(31)} J λ ( Y n ) ( k 0 r , θ , ϕ ) = k 0 π ℓ ( ℓ + 1 ) 2 η 0 M ℓ m 0 ( k 0 r ) (31)
and
H ~ λ ( Y n ) = k 0 i η 0 2 η 0 π ℓ ( ℓ + 1 ) M ℓ m 0 ( k 0 r ) . (32) \widetilde{H}_{\lambda}(Y_{n})=\frac{k_{0}}{\mathrm{i}\eta_{0}}\sqrt{\frac{2\eta_{0}}{\pi\ell(\ell+1)}}M_{\ell m}^{0}(k_{0}r). \tag*{(32)} H λ ( Y n ) = i η 0 k 0 π ℓ ( ℓ + 1 ) 2 η 0 M ℓ m 0 ( k 0 r ) . (32)
These free fields are tangential to spheres where Ω = B r 0 \Omega=B_{r_{0}} Ω = B r 0 and satisfy the weighted divergence constraints. We now give the construction for a general bounded Lipschitz Ω \Omega Ω , retaining R λ , E R_{\lambda,E} R λ , E on L ϵ ( div 0 ) L_{\epsilon}(\operatorname{div}0) L ϵ ( div 0 ) and R λ , H R_{\lambda,H} R λ , H on L μ ( div 0 ) L_{\mu}(\operatorname{div}0) L μ ( div 0 ) . We encode the coefficient contrast by compactly supported functionals on the form domains.
Definition 2.2. We define the compactly supported functional Θ λ , E ( Y n ) \Theta_{\lambda,E}(Y_n) Θ λ , E ( Y n ) on the space V E V_E V E by
Θ λ , E ( Y n ) [ φ ] = λ 2 ∫ Ω ( ϵ 0 μ 0 μ 1 − ϵ 1 ) j ~ λ ( Y n ) ⋅ φ ‾ d x − ⟨ [ [ ν × μ − 1 curl j ~ λ ( Y n ) ] ] , φ T ⟩ ∂ Ω (33) \Theta_{\lambda,E}(Y_n)[\varphi]=\lambda^2\int_{\Omega}\left(\frac{\epsilon_0\mu_0}{\mu_1}-\epsilon_1\right)\widetilde{j}_{\lambda}(Y_n)\cdot\overline{\varphi}\,\mathrm{d}x-\left\langle\left[\left[\nu\times\mu^{-1}\operatorname{curl}\widetilde{j}_{\lambda}(Y_n)\right]\right],\varphi_T\right\rangle_{\partial\Omega} \tag*{(33)} Θ λ , E ( Y n ) [ φ ] = λ 2 ∫ Ω ( μ 1 ϵ 0 μ 0 − ϵ 1 ) j λ ( Y n ) ⋅ φ d x − ⟨ [ [ ν × μ − 1 curl j λ ( Y n ) ] ] , φ T ⟩ ∂ Ω (33) and also the functional Θ λ , H ( Y n ) \Theta_{\lambda,H}(Y_n) Θ λ , H ( Y n ) on V H V_H V H by
Θ λ , H ( Y n ) [ φ ] = λ 2 ∫ Ω ( ϵ 0 μ 0 ϵ 1 − μ 1 ) H ~ λ ( Y n ) ⋅ φ ‾ d x − ⟨ [ [ ν × ϵ − 1 curl H ~ λ ( Y n ) ] ] , φ T ⟩ ∂ Ω . (34) \Theta_{\lambda,H}(Y_n)[\varphi]=\lambda^2\int_{\Omega}\left(\frac{\epsilon_0\mu_0}{\epsilon_1}-\mu_1\right)\widetilde{H}_{\lambda}(Y_n)\cdot\overline{\varphi}\,\mathrm{d}x-\left\langle\left[\left[\nu\times\epsilon^{-1}\operatorname{curl}\widetilde{H}_{\lambda}(Y_n)\right]\right],\varphi_T\right\rangle_{\partial\Omega}. \tag*{(34)} Θ λ , H ( Y n ) [ φ ] = λ 2 ∫ Ω ( ϵ 1 ϵ 0 μ 0 − μ 1 ) H λ ( Y n ) ⋅ φ d x − ⟨ [ [ ν × ϵ − 1 curl H λ ( Y n ) ] ] , φ T ⟩ ∂ Ω . (34) Give V E V_E V E the form norm ∥ u ∥ V E 2 = q E ( u , u ) + ∥ u ∥ L 2 ( ϵ d x ) 3 2 \lVert u\rVert_{V_E}^2=q_E(u,u)+\lVert u\rVert_{L^2(\epsilon\mathrm{d}x)^3}^2 ∥ u ∥ V E 2 = q E ( u , u ) + ∥ u ∥ L 2 ( ϵ d x ) 3 2 and write V E ∗ V_E^* V E ∗ for its continuous anti-dual. Then for Im λ > 0 \operatorname{Im}\lambda>0 Im λ > 0 , R λ , E : V E ∗ → V E R_{\lambda,E}:V_E^*\to V_E R λ , E : V E ∗ → V E . For Im λ > 0 \operatorname{Im}\lambda>0 Im λ > 0 , u = R λ , E Θ λ , E ( Y n ) ∈ V E u=R_{\lambda,E}\Theta_{\lambda,E}(Y_n)\in V_E u = R λ , E Θ λ , E ( Y n ) ∈ V E is the unique solution of
q E ( u , φ ) − λ 2 ⟨ u , φ ⟩ ( L 2 ( ϵ d x ) ) 3 = Θ λ , E ( Y n ) [ φ ] ∀ φ ∈ V E , q_E(u,\varphi)-\lambda^2\langle u,\varphi\rangle_{(L^2(\epsilon\mathrm{d}x))^3}=\Theta_{\lambda,E}(Y_n)[\varphi]\qquad\forall\varphi\in V_E, q E ( u , φ ) − λ 2 ⟨ u , φ ⟩ ( L 2 ( ϵ d x ) ) 3 = Θ λ , E ( Y n ) [ φ ] ∀ φ ∈ V E , that is, Θ λ , E ( Y n ) = ( P E − λ 2 ) u \Theta_{\lambda,E}(Y_n)=(P_E-\lambda^2)u Θ λ , E ( Y n ) = ( P E − λ 2 ) u in the weak sense. Equivalently,
⟨ R λ , E Θ λ , E ( Y n ) , f ⟩ ( L 2 ( ϵ d x ) ) 3 = Θ λ , E ( Y n ) [ ( P E − λ ‾ 2 ) − 1 f ] , f ∈ L ϵ ( div 0 ) . \left\langle R_{\lambda,E}\Theta_{\lambda,E}(Y_n),f\right\rangle_{(L^2(\epsilon\mathrm{d}x))^3}=\Theta_{\lambda,E}(Y_n)\left[(P_E-\overline{\lambda}^2)^{-1}f\right],\qquad f\in L_\epsilon(\operatorname{div}0). ⟨ R λ , E Θ λ , E ( Y n ) , f ⟩ ( L 2 ( ϵ d x ) ) 3 = Θ λ , E ( Y n ) [ ( P E − λ 2 ) − 1 f ] , f ∈ L ϵ ( div 0 ) . For λ > 0 \lambda>0 λ > 0 we define R λ , E Θ λ , E ( Y n ) R_{\lambda,E}\Theta_{\lambda,E}(Y_n) R λ , E Θ λ , E ( Y n ) by
⟨ R λ , E Θ λ , E ( Y n ) , f ⟩ ( L 2 ( ϵ d x ) ) 3 = Θ λ , E ( Y n ) [ R − λ , E f ] , f ∈ L c o m p 2 ∩ L ϵ ( div 0 ) ; \left\langle R_{\lambda,E}\Theta_{\lambda,E}(Y_n),f\right\rangle_{(L^2(\epsilon\mathrm{d}x))^3}=\Theta_{\lambda,E}(Y_n)[R_{-\lambda,E}f],\qquad f\in L^2_{\mathrm{comp}}\cap L_\epsilon(\operatorname{div}0); ⟨ R λ , E Θ λ , E ( Y n ) , f ⟩ ( L 2 ( ϵ d x ) ) 3 = Θ λ , E ( Y n ) [ R − λ , E f ] , f ∈ L comp 2 ∩ L ϵ ( div 0 ) ; Here R λ , E Θ λ , E R_{\lambda,E}\Theta_{\lambda,E} R λ , E Θ λ , E denotes the H l o c ( curl ) H_{\mathrm{loc}}(\operatorname{curl}) H loc ( curl ) limit from Im λ > 0 \operatorname{Im}\lambda>0 Im λ > 0 , justified in the proof below; the displayed pairing is its duality characterization.
For the general interface define p λ , E ∈ H ˙ 1 ( R 3 ) p_{\lambda,E}\in\dot{H}^1(\mathbb{R}^3) p λ , E ∈ H ˙ 1 ( R 3 ) by
∫ R 3 ϵ ∇ p λ , E ⋅ ∇ v ‾ d x = ( ϵ 1 − ϵ 0 ) ∫ Ω j ~ λ ( Y n ) ⋅ ∇ v ‾ d x , v ∈ H ˙ 1 ( R 3 ) . (35) \int_{\mathbb{R}^3}\epsilon\nabla p_{\lambda,E}\cdot\overline{\nabla v}\,\mathrm{d}x=(\epsilon_1-\epsilon_0)\int_{\Omega}\widetilde{j}_{\lambda}(Y_n)\cdot\overline{\nabla v}\,\mathrm{d}x,\qquad v\in\dot{H}^1(\mathbb{R}^3). \tag*{(35)} ∫ R 3 ϵ ∇ p λ , E ⋅ ∇ v d x = ( ϵ 1 − ϵ 0 ) ∫ Ω j λ ( Y n ) ⋅ ∇ v d x , v ∈ H ˙ 1 ( R 3 ) . (35) Uniform positivity gives a unique solution by Lax–Milgram. Define p λ , H p_{\lambda,H} p λ , H by interchanging ϵ , μ \epsilon,\mu ϵ , μ and replacing j ~ λ \widetilde{j}_{\lambda} j λ by H ~ λ \widetilde{H}_{\lambda} H λ . Then div ( ϵ ( j ~ λ − ∇ p λ , E ) ) = 0 \operatorname{div}(\epsilon(\widetilde{j}_{\lambda}-\nabla p_{\lambda,E}))=0 div ( ϵ ( j λ − ∇ p λ , E )) = 0 , and similarly for the magnetic field. For the ball the right-hand sides vanish by the fact that div ( j ~ λ ) = 0 \operatorname{div}(\widetilde{j}_{\lambda})=0 div ( j λ ) = 0 and j ~ λ ⋅ ν = 0 \widetilde{j}_{\lambda}\cdot\nu=0 j λ ⋅ ν = 0 , so correspondingly p λ , E = p λ , H = 0 p_{\lambda,E}=p_{\lambda,H}=0 p λ , E = p λ , H = 0 .
Lemma 2.3. For compactly supported test fields φ ∈ H ( curl , R 3 ) \varphi\in H(\operatorname{curl},\mathbb{R}^3) φ ∈ H ( curl , R 3 )
Θ λ , E [ φ ] = q E ( j ~ λ , φ ) − λ 2 ⟨ j ~ λ , φ ⟩ ( L 2 ( ϵ d x ) ) 3 . \Theta_{\lambda,E}[\varphi]=q_E(\widetilde{j}_{\lambda},\varphi)-\lambda^2\langle\widetilde{j}_{\lambda},\varphi\rangle_{(L^2(\epsilon\mathrm{d}x))^3}. Θ λ , E [ φ ] = q E ( j λ , φ ) − λ 2 ⟨ j λ , φ ⟩ ( L 2 ( ϵ d x ) ) 3 . For general φ ∈ V E \varphi\in V_E φ ∈ V E , the left-hand side is defined by the preceding volume-and-boundary formula; the magnetic identity below is interpreted in the same way on V H V_H V H . For general Ω \Omega Ω we define, for Im λ > 0 \operatorname{Im}\lambda>0 Im λ > 0 and by outgoing limits for λ > 0 \lambda>0 λ > 0 ,
E λ , 1 ( Y n ) = j ~ λ ( Y n ) − ∇ p λ , E − R λ , E Θ λ , E ( Y n ) . E_{\lambda,1}(Y_n)=\widetilde{j}_{\lambda}(Y_n)-\nabla p_{\lambda,E}-R_{\lambda,E}\Theta_{\lambda,E}(Y_n). E λ , 1 ( Y n ) = j λ ( Y n ) − ∇ p λ , E − R λ , E Θ λ , E ( Y n ) . This field solves the transmission problem with incident field j ~ λ ( Y n ) \widetilde{j}_{\lambda}(Y_n) j λ ( Y n ) and outgoing scattered part − ∇ p λ , E − R λ , E Θ λ , E ( Y n ) -\nabla p_{\lambda,E}-R_{\lambda,E}\Theta_{\lambda,E}(Y_n) − ∇ p λ , E − R λ , E Θ λ , E ( Y n ) . For Ω = B r 0 \Omega=B_{r_0} Ω = B r 0 it agrees with (2.4); explicitly,
R λ , E Θ λ , E ( Y n ) = { k 0 2 η 0 π ℓ ( ℓ + 1 ) ( M ℓ m 0 ( k 0 r ) − t ℓ 1 ( λ ) M ℓ m 0 ( k 1 r ) ) , r < r 0 , − k 0 2 η 0 π ℓ ( ℓ + 1 ) A ℓ 1 ( λ ) M ℓ m − ( k 0 r ) , r > r 0 . (36) R_{\lambda,E}\Theta_{\lambda,E}(Y_n)=
\begin{cases}
k_0\sqrt{\frac{2\eta_0}{\pi\ell(\ell+1)}}\left(M^0_{\ell m}(k_0r)-t^1_{\ell}(\lambda)M^0_{\ell m}(k_1r)\right), & r<r_0,\\
-k_0\sqrt{\frac{2\eta_0}{\pi\ell(\ell+1)}}A^1_{\ell}(\lambda)M^-_{\ell m}(k_0r), & r>r_0.
\end{cases} \tag*{(36)} R λ , E Θ λ , E ( Y n ) = ⎩ ⎨ ⎧ k 0 π ℓ ( ℓ + 1 ) 2 η 0 ( M ℓ m 0 ( k 0 r ) − t ℓ 1 ( λ ) M ℓ m 0 ( k 1 r ) ) , − k 0 π ℓ ( ℓ + 1 ) 2 η 0 A ℓ 1 ( λ ) M ℓ m − ( k 0 r ) , r < r 0 , r > r 0 . (36) Similarly, for general Ω \Omega Ω define
H λ ( Y n ) = H ~ λ ( Y n ) − ∇ p λ , H − R λ , H Θ λ , H ( Y n ) H_{\lambda}(Y_n)=\widetilde{H}_{\lambda}(Y_n)-\nabla p_{\lambda,H}-R_{\lambda,H}\Theta_{\lambda,H}(Y_n) H λ ( Y n ) = H λ ( Y n ) − ∇ p λ , H − R λ , H Θ λ , H ( Y n )
where
Θ λ , H [ φ ] = q H ( H ~ λ , φ ) − λ 2 ⟨ H ~ λ , φ ⟩ ( L 2 ( μ d x ) ) 3 (37) \Theta_{\lambda,H}[\varphi]=q_H(\widetilde{H}_{\lambda},\varphi)-\lambda^2\langle\widetilde{H}_{\lambda},\varphi\rangle_{(L^2(\mu\,dx))^3} \tag*{(37)} Θ λ , H [ φ ] = q H ( H λ , φ ) − λ 2 ⟨ H λ , φ ⟩ ( L 2 ( μ d x ) ) 3 (37)
and define the second electric family by
E λ , 2 ( Y n ) = i λ ϵ − 1 curl H λ ( Y n ) . (38) E_{\lambda,2}(Y_n)=\frac{\mathrm{i}}{\lambda}\epsilon^{-1}\operatorname{curl}H_{\lambda}(Y_n). \tag*{(38)} E λ , 2 ( Y n ) = λ i ϵ − 1 curl H λ ( Y n ) . (38)
These fields satisfy the corresponding transmission problems and agree with the ball construction, in particular (2.5), when Ω = B r 0 \Omega=B_{r_0} Ω = B r 0 .
Proof. Fix an incident mode Y n Y_n Y n and initially let Im λ > 0 \operatorname{Im}\lambda>0 Im λ > 0 . We first verify the electric construction for a general bounded Lipschitz inclusion. Integration by parts shows that Θ λ , E \Theta_{\lambda,E} Θ λ , E is the residual of the free incident field j ~ λ \widetilde{j}_{\lambda} j λ in the transmission equation. More explicitly, as a functional on H ( curl , R 3 ) H(\operatorname{curl},\mathbb{R}^3) H ( curl , R 3 ) it is
Θ λ , E [ v ] = ∫ R 3 F λ ⋅ v ‾ d x + ∫ R 3 G λ ⋅ curl v ‾ d x . \Theta_{\lambda,E}[v]=\int_{\mathbb{R}^3}F_{\lambda}\cdot\overline{v}\,dx+\int_{\mathbb{R}^3}G_{\lambda}\cdot\operatorname{curl}\overline{v}\,dx. Θ λ , E [ v ] = ∫ R 3 F λ ⋅ v d x + ∫ R 3 G λ ⋅ curl v d x .
where
F λ = − λ 2 ( ϵ 1 − ϵ 0 ) 1 Ω j ~ λ , G λ = ( μ 1 − 1 − μ 0 − 1 ) 1 Ω curl j ~ λ . F_{\lambda}=-\lambda^2(\epsilon_1-\epsilon_0)1_{\Omega}\widetilde{j}_{\lambda},\qquad G_{\lambda}=(\mu_1^{-1}-\mu_0^{-1})1_{\Omega}\operatorname{curl}\widetilde{j}_{\lambda}. F λ = − λ 2 ( ϵ 1 − ϵ 0 ) 1 Ω j λ , G λ = ( μ 1 − 1 − μ 0 − 1 ) 1 Ω curl j λ .
This agrees with the volume-and-boundary definition and gives the stated weak identity for compactly supported test fields.
Set u = R λ , E Θ λ , E ∈ V E u=R_{\lambda,E}\Theta_{\lambda,E}\in V_E u = R λ , E Θ λ , E ∈ V E . The resolvent equation is initially tested only on V E V_E V E , so we must account for the gradient part of an arbitrary test field. Since ( Id − Π E ) v (\operatorname{Id}-\Pi_E)v ( Id − Π E ) v is a gradient, the definition of p λ , E p_{\lambda,E} p λ , E in (2.20) gives
Θ λ , E [ ( Id − Π E ) v ] = − λ 2 ⟨ ∇ p λ , E , v ⟩ L 2 ( ϵ d x ) 3 . \Theta_{\lambda,E}[(\operatorname{Id}-\Pi_E)v]=-\lambda^2\langle\nabla p_{\lambda,E},v\rangle_{L^2(\epsilon\,dx)^3}. Θ λ , E [( Id − Π E ) v ] = − λ 2 ⟨ ∇ p λ , E , v ⟩ L 2 ( ϵ d x ) 3 .
Testing the resolvent equation on Π E v ∈ V E \Pi_Ev\in V_E Π E v ∈ V E therefore yields, for every v ∈ H ( curl , R 3 ) v\in H(\operatorname{curl},\mathbb{R}^3) v ∈ H ( curl , R 3 ) ,
q E ( u , v ) − λ 2 ⟨ u , v ⟩ L 2 ( ϵ d x ) 3 = Θ λ , E [ Π E v ] = Θ λ , E [ v ] + λ 2 ⟨ ∇ p λ , E , v ⟩ L 2 ( ϵ d x ) 3 . q_E(u,v)-\lambda^2\langle u,v\rangle_{L^2(\epsilon\,dx)^3}=\Theta_{\lambda,E}[\Pi_Ev]=\Theta_{\lambda,E}[v]+\lambda^2\langle\nabla p_{\lambda,E},v\rangle_{L^2(\epsilon\,dx)^3}. q E ( u , v ) − λ 2 ⟨ u , v ⟩ L 2 ( ϵ d x ) 3 = Θ λ , E [ Π E v ] = Θ λ , E [ v ] + λ 2 ⟨ ∇ p λ , E , v ⟩ L 2 ( ϵ d x ) 3 .
Equivalently, on all of R 3 \mathbb{R}^3 R 3 in the distributional sense,
curl μ − 1 curl u − λ 2 ϵ u = F λ + curl G λ + λ 2 ϵ ∇ p λ , E . \operatorname{curl}\mu^{-1}\operatorname{curl}u-\lambda^2\epsilon u=F_{\lambda}+\operatorname{curl}G_{\lambda}+\lambda^2\epsilon\nabla p_{\lambda,E}. curl μ − 1 curl u − λ 2 ϵ u = F λ + curl G λ + λ 2 ϵ ∇ p λ , E .
The residual of j ~ λ \widetilde{j}_{\lambda} j λ is F λ + curl G λ F_{\lambda}+\operatorname{curl}G_{\lambda} F λ + curl G λ , whereas that of ∇ p λ , E \nabla p_{\lambda,E} ∇ p λ , E is − λ 2 ϵ ∇ p λ , E -\lambda^2\epsilon\nabla p_{\lambda,E} − λ 2 ϵ ∇ p λ , E . Subtracting these identities proves that
E : = j ~ λ − ∇ p λ , E − u satisfies curl μ − 1 curl E − λ 2 ϵ E = 0. E:=\widetilde{j}_{\lambda}-\nabla p_{\lambda,E}-u\qquad\text{satisfies}\qquad\operatorname{curl}\mu^{-1}\operatorname{curl}E-\lambda^2\epsilon E=0. E := j λ − ∇ p λ , E − u satisfies curl μ − 1 curl E − λ 2 ϵ E = 0.
Moreover, (2.20) and u ∈ V E u\in V_E u ∈ V E give div ( ϵ E ) = 0 \operatorname{div}(\epsilon E)=0 div ( ϵ E ) = 0 . We have E ∈ H l o c ( curl ) E\in H_{\mathrm{loc}}(\operatorname{curl}) E ∈ H loc ( curl ) , and its equation gives μ − 1 curl E ∈ H l o c ( curl ) \mu^{-1}\operatorname{curl}E\in H_{\mathrm{loc}}(\operatorname{curl}) μ − 1 curl E ∈ H loc ( curl ) . These two global regularity statements imply the two tangential transmission conditions. The scattered field E − j ~ λ = − ∇ p λ , E − u E-\widetilde{j}_{\lambda}=-\nabla p_{\lambda,E}-u E − j λ = − ∇ p λ , E − u solves the free homogeneous equation outside Ω \Omega Ω and is square integrable there for Im λ > 0 \operatorname{Im}\lambda>0 Im λ > 0 , hence is outgoing.
It remains to justify the boundary value at real λ > 0 \lambda>0 λ > 0 . Introduce the auxiliary field
H u : = 1 i λ ( μ − 1 curl u − G λ ) . H_u:=\frac{1}{\mathrm{i}\lambda}\left(\mu^{-1}\operatorname{curl}u-G_{\lambda}\right). H u := i λ 1 ( μ − 1 curl u − G λ ) .
The equation for u u u is then equivalent to the first-order system
curl u − i λ μ H u = μ G λ , \operatorname{curl}u-\mathrm{i}\lambda\mu H_u=\mu G_{\lambda}, curl u − i λ μ H u = μ G λ ,
curl H u + i λ ϵ u = F λ i λ − i λ ϵ ∇ p λ , E . \operatorname{curl}H_u+\mathrm{i}\lambda\epsilon u=\frac{F_{\lambda}}{\mathrm{i}\lambda}-\mathrm{i}\lambda\epsilon\nabla p_{\lambda,E}. curl H u + i λ ϵ u = i λ F λ − i λ ϵ ∇ p λ , E .
The fields F λ F_\lambda F λ , G λ G_\lambda G λ are compactly supported and square integrable. The potential p λ , E p_{\lambda,E} p λ , E is harmonic outside a ball and has zero total flux, because the source in its divergence-form equation is the divergence of a compactly supported vector field. Its finite-energy exterior harmonic expansion therefore gives ∇ p λ , E = O ( r − 3 ) \nabla p_{\lambda,E}=O(r^{-3}) ∇ p λ , E = O ( r − 3 ) . Both sources in the first-order system consequently belong to L s 2 = { f : ( 1 + ∣ x ∣ 2 ) s / 2 f ∈ L 2 } L_s^2=\{f:(1+|x|^2)^{s/2}f\in L^2\} L s 2 = { f : ( 1 + ∣ x ∣ 2 ) s /2 f ∈ L 2 } for 1 / 2 < s < 3 / 2 1/2<s<3/2 1/2 < s < 3/2 .
We may now apply the whole-space limiting absorption principle [30 Theorem 3.10 and Corollary 3.12] , also summarized in [28 Theorem 4.3] . The coefficients are uniformly positive and bounded and are constant outside a ball; local Maxwell compactness follows by localization to balls. Theorem A.1 rules out the homogeneous radiating kernel at every real λ > 0 \lambda>0 λ > 0 . We obtain the outgoing H l o c ( curl ) H_{\mathrm{loc}}(\operatorname{curl}) H loc ( curl ) limit of u u u , with bounds locally uniform on compact subintervals of ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . This also gives the outgoing boundary value of E − j ~ λ E-\widetilde{j}_\lambda E − j λ . For compactly supported volume data f f f , the gradient part of its weighted Helmholtz decomposition likewise decays as O ( r − 3 ) O(r^{-3}) O ( r − 3 ) , so the same argument applies to R λ , E Π E f R_{\lambda,E}\Pi_Ef R λ , E Π E f .
Interchanging ε \varepsilon ε and μ \mu μ and using the incident field H ~ λ \widetilde{H}_\lambda H λ gives the total magnetic field
H λ = H ~ λ − ∇ p λ , H − R λ , H Θ λ , H . H_\lambda=\widetilde{H}_\lambda-\nabla p_{\lambda,H}-R_{\lambda,H}\Theta_{\lambda,H}. H λ = H λ − ∇ p λ , H − R λ , H Θ λ , H .
This is the field used to construct the second electric family; H u H_u H u above was only an auxiliary field for the limiting absorption argument. Since curl ε − 1 curl H λ = λ 2 μ H λ \operatorname{curl}\varepsilon^{-1}\operatorname{curl}H_\lambda=\lambda^2\mu H_\lambda curl ε − 1 curl H λ = λ 2 μ H λ , setting E λ , 2 = i ε − 1 curl H λ / λ E_{\lambda,2}=i\varepsilon^{-1}\operatorname{curl}H_\lambda/\lambda E λ , 2 = i ε − 1 curl H λ / λ gives both Maxwell curl equations and div ( ε E λ , 2 ) = 0 \operatorname{div}(\varepsilon E_{\lambda,2})=0 div ( ε E λ , 2 ) = 0 .
Finally, specialize to Ω = B r 0 \Omega=B_{r_0} Ω = B r 0 . The incident fields are divergence-free in the ordinary sense and have zero normal component on the spherical boundary. Thus p λ , E = p λ , H = 0 p_{\lambda,E}=p_{\lambda,H}=0 p λ , E = p λ , H = 0 . By uniqueness, the two electric fields just constructed coincide with (2.4) and (2.5), respectively. The identity u = j ~ λ − E λ , 1 u=\widetilde{j}_\lambda-E_{\lambda,1} u = j λ − E λ , 1 now gives (2.21). To identify the corresponding magnetic field explicitly, use H λ = ( i λ ) − 1 μ − 1 curl E λ , 2 H_\lambda=(i\lambda)^{-1}\mu^{-1}\operatorname{curl}E_{\lambda,2} H λ = ( iλ ) − 1 μ − 1 curl E λ , 2 . Applying curl N ℓ m ( k j r ) = k j M ℓ m ( k j r ) \operatorname{curl}N_{\ell m}(k_jr)=k_jM_{\ell m}(k_jr) curl N ℓ m ( k j r ) = k j M ℓ m ( k j r ) and k j / ( λ μ j ) = η j − 1 k_j/(\lambda\mu_j)=\eta_j^{-1} k j / ( λ μ j ) = η j − 1 in each medium gives
H λ ( Y n ) = k 0 i 2 π η 0 ℓ ( ℓ + 1 ) { η 0 η 1 t ℓ 2 ( λ ) M ℓ m 0 ( k 1 r ) , r < r 0 , M ℓ m 0 ( k 0 r ) + A ℓ 2 ( λ ) M ℓ m − ( k 0 r ) , r > r 0 . H_\lambda(Y_n)=\frac{k_0}{i}\sqrt{\frac{2}{\pi\eta_0\ell(\ell+1)}}\begin{cases}
\frac{\eta_0}{\eta_1}t_\ell^2(\lambda)M_{\ell m}^0(k_1r),&r<r_0,\\
M_{\ell m}^0(k_0r)+A_\ell^2(\lambda)M_{\ell m}^-(k_0r),&r>r_0.
\end{cases} H λ ( Y n ) = i k 0 π η 0 ℓ ( ℓ + 1 ) 2 { η 1 η 0 t ℓ 2 ( λ ) M ℓ m 0 ( k 1 r ) , M ℓ m 0 ( k 0 r ) + A ℓ 2 ( λ ) M ℓ m − ( k 0 r ) , r < r 0 , r > r 0 .
The factor η 0 / η 1 \eta_0/\eta_1 η 0 / η 1 comes from factoring out the exterior normalization 1 / η 0 1/\eta_0 1/ η 0 ; the interior magnetic field carries 1 / η 1 1/\eta_1 1/ η 1 . Conversely, applying i ε − 1 curl / λ i\varepsilon^{-1}\operatorname{curl}/\lambda i ε − 1 curl / λ to this formula recovers (2.5).
Proposition 2.4. For Ω = B r 0 \Omega=B_{r_0} Ω = B r 0 , the matrices A ℓ j ( λ ) A_\ell^j(\lambda) A ℓ j ( λ ) , j = 1 , 2 j=1,2 j = 1 , 2 coincide exactly with the T T T -matrix entries. The scattering amplitudes are obtained from these outgoing expansion coefficients by the phase and normalization factors displayed below.
Proof. Insert h ℓ ( 1 ) ( y ) = ( − i ) ℓ + 1 e i y y − 1 ( 1 + O ( y − 1 ) ) h_\ell^{(1)}(y)=(-i)^{\ell+1}e^{iy}y^{-1}(1+O(y^{-1})) h ℓ ( 1 ) ( y ) = ( − i ) ℓ + 1 e i y y − 1 ( 1 + O ( y − 1 )) from (B.14) into the Hansen fields. The recurrence (B.1) gives ( y h ℓ ( 1 ) ( y ) ) ′ / y = ( − i ) ℓ e i y y − 1 ( 1 + O ( y − 1 ) ) (yh_\ell^{(1)}(y))'/y=(-i)^\ell e^{iy}y^{-1}(1+O(y^{-1})) ( y h ℓ ( 1 ) ( y ) ) ′ / y = ( − i ) ℓ e i y y − 1 ( 1 + O ( y − 1 )) , while the radial component of N ℓ m − N_{\ell m}^- N ℓ m − is O ( e i k 0 r r − 2 ) O(e^{ik_0r}r^{-2}) O ( e i k 0 r r − 2 ) . Applying these asymptotics to (2.4)--(2.5) gives, for fixed ℓ \ell ℓ and k 0 k_0 k 0 in a compact subset of C ∖ { 0 } \mathbb{C}\setminus\{0\} C ∖ { 0 } ,
E λ , 1 ( Y n ) − j ~ λ ( Y n ) = ( − i ) ℓ + 1 2 η 0 π A ℓ 1 ( λ ) e i k 0 r r Ψ 1 ℓ m ( 1 + O ( r − 1 ) ) (39) E_{\lambda,1}(Y_n)-\widetilde{j}_\lambda(Y_n)=(-i)^{\ell+1}\sqrt{\frac{2\eta_0}{\pi}}A_\ell^1(\lambda)\frac{e^{ik_0r}}{r}\Psi_{1\ell m}(1+O(r^{-1})) \tag*{(39)} E λ , 1 ( Y n ) − j λ ( Y n ) = ( − i ) ℓ + 1 π 2 η 0 A ℓ 1 ( λ ) r e i k 0 r Ψ 1 ℓ m ( 1 + O ( r − 1 )) (39)
and
E λ , 2 ( Y n ) − 1 k 0 curl ( j ~ λ ( Y n ) ) = ( − i ) ℓ 2 η 0 π A ℓ 2 ( λ ) e i k 0 r r ( Ψ 2 ℓ m + O ( r − 1 ) ) . (40) E_{\lambda,2}(Y_n)-\frac{1}{k_0}\operatorname{curl}(\widetilde{j}_\lambda(Y_n))=(-i)^\ell\sqrt{\frac{2\eta_0}{\pi}}A_\ell^2(\lambda)\frac{e^{ik_0r}}{r}\bigl(\Psi_{2\ell m}+O(r^{-1})\bigr). \tag*{(40)} E λ , 2 ( Y n ) − k 0 1 curl ( j λ ( Y n )) = ( − i ) ℓ π 2 η 0 A ℓ 2 ( λ ) r e i k 0 r ( Ψ 2 ℓ m + O ( r − 1 ) ) . (40)
The following assertion holds for every bounded Lipschitz inclusion.
Proposition 2.5. There are no nonzero resonances for P E P_{\mathrm{E}} P E or P H P_{\mathrm{H}} P H with Im λ ≥ 0 \operatorname{Im}\lambda\ge0 Im λ ≥ 0 .
Proof. An outgoing solution with Im λ > 0 \operatorname{Im}\lambda> 0 Im λ > 0 decays exponentially and, by the transmission conditions, belongs to D ( P E ) D(P_{\mathrm{E}}) D ( P E ) . It would therefore be an L 2 L^2 L 2 eigenfunction with eigenvalue λ 2 ∉ [ 0 , ∞ ) \lambda^2 \notin[0,\infty) λ 2 ∈ / [ 0 , ∞ ) , contradicting non-negativity and self-adjointness. For real λ ≠ 0 \lambda\ne0 λ = 0 , Theorem A.1 gives uniqueness. The magnetic case is identical. □
Proposition 2.6. The poles of the scattering matrix are in the lower half-plane. For the ball, set z = x 0 = k 0 r 0 z=x_0=k_0r_0 z = x 0 = k 0 r 0 , so that x 1 = m 0 z x_1=m_0z x 1 = m 0 z . The TE resonance occurs when
ψ ℓ ′ ( m 0 z ) ψ ℓ ( m 0 z ) = η 1 η 0 ξ ℓ ′ ( z ) ξ ℓ ( z ) . (41) \frac{\psi_\ell'(m_0z)}{\psi_\ell(m_0z)}=\frac{\eta_1}{\eta_0}\frac{\xi_\ell'(z)}{\xi_\ell(z)}. \tag*{(41)} ψ ℓ ( m 0 z ) ψ ℓ ′ ( m 0 z ) = η 0 η 1 ξ ℓ ( z ) ξ ℓ ′ ( z ) . (41)
The TM resonance occurs when
ψ ℓ ′ ( m 0 z ) ψ ℓ ( m 0 z ) = η 0 η 1 ξ ℓ ′ ( z ) ξ ℓ ( z ) . (42) \frac{\psi_\ell'(m_0z)}{\psi_\ell(m_0z)}=\frac{\eta_0}{\eta_1}\frac{\xi_\ell'(z)}{\xi_\ell(z)}. \tag*{(42)} ψ ℓ ( m 0 z ) ψ ℓ ′ ( m 0 z ) = η 1 η 0 ξ ℓ ( z ) ξ ℓ ′ ( z ) . (42)
:::
Proof. A purely outgoing TE field has the form t M ℓ m 0 ( k 1 r ) tM^0_{\ell m}(k_1r) t M ℓ m 0 ( k 1 r ) inside and c M ℓ m − ( k 0 r ) cM^-_{\ell m}(k_0r) c M ℓ m − ( k 0 r ) outside. The transmission conditions are
( j ℓ ( x 1 ) − h ℓ ( 1 ) ( x 0 ) μ 1 − 1 ψ ℓ ′ ( x 1 ) − μ 0 − 1 ξ ℓ ′ ( x 0 ) ) ( t c ) = 0. \begin{pmatrix}
j_\ell(x_1) & -h_\ell^{(1)}(x_0) \\
\mu_1^{-1}\psi_\ell'(x_1) & -\mu_0^{-1}\xi_\ell'(x_0)
\end{pmatrix}
\begin{pmatrix}
t\\
c
\end{pmatrix}
=0. ( j ℓ ( x 1 ) μ 1 − 1 ψ ℓ ′ ( x 1 ) − h ℓ ( 1 ) ( x 0 ) − μ 0 − 1 ξ ℓ ′ ( x 0 ) ) ( t c ) = 0.
This system has a nonzero solution precisely when D ℓ 1 ( λ ) = 0 D^1_\ell(\lambda)=0 D ℓ 1 ( λ ) = 0 , which gives (2.26) when ψ ℓ ( m 0 z ) ξ ℓ ( z ) ≠ 0 \psi_\ell(m_0z)\xi_\ell(z)\ne0 ψ ℓ ( m 0 z ) ξ ℓ ( z ) = 0 . Interchanging μ \mu μ and ϵ \epsilon ϵ gives the TM condition. Proposition 2.5 locates the poles in the lower half-plane. □
Proof of Theorem 1.4
Proof of Theorem 1.4. The spectral representation below follows from Proposition 4.2 . We set for Im λ > 0 \operatorname{Im}\lambda> 0 Im λ > 0
G λ , E ( x , y ) = ∑ n ∫ 0 ∞ E σ ( Y n ) ( x ) ⊗ E σ ( Y n ) ( y ) ‾ σ 2 − λ 2 d σ . (43) G_{\lambda,\mathrm{E}}(x,y)=\sum_n\int_0^\infty\frac{E_\sigma(Y_n)(x)\otimes\overline{E_\sigma(Y_n)(y)}}{\sigma^2-\lambda^2}\,d\sigma. \tag*{(43)} G λ , E ( x , y ) = n ∑ ∫ 0 ∞ σ 2 − λ 2 E σ ( Y n ) ( x ) ⊗ E σ ( Y n ) ( y ) d σ . (43)
It follows that the resolvent can be interpreted as
( R λ , E f ) ( x ) = ∫ R 3 G λ , E ( x , y ) f ( y ) ϵ ( y ) d y div ( ϵ f ) = 0. (44) (R_{\lambda,\mathrm{E}}f)(x)=\int_{\mathbb{R}^3}G_{\lambda,\mathrm{E}}(x,y)f(y)\epsilon(y)\,dy
\qquad\operatorname{div}(\epsilon f)=0. \tag*{(44)} ( R λ , E f ) ( x ) = ∫ R 3 G λ , E ( x , y ) f ( y ) ϵ ( y ) d y div ( ϵ f ) = 0. (44)
Let E λ : L t 2 ( S 2 ; C 3 ) → L l o c 2 E_\lambda:L^2_t(S^2;\mathbb{C}^3)\to L^2_{\mathrm{loc}} E λ : L t 2 ( S 2 ; C 3 ) → L loc 2 , Y ↦ E λ ( Y ) Y\mapsto E_\lambda(Y) Y ↦ E λ ( Y ) , be as in Definition 4.1, and let ( χ E λ ‾ ) ∗ (\chi E_{\overline{\lambda}})^* ( χ E λ ) ∗ be the adjoint of χ E λ ‾ \chi E_{\overline{\lambda}} χ E λ with respect to L 2 ( ϵ d x ) L^2(\epsilon\,dx) L 2 ( ϵ d x ) and L 2 ( S 2 ) L^2(S^2) L 2 ( S 2 ) , so that
( χ E λ ) ( χ E λ ‾ ) ∗ f = ∑ n χ E λ ( Y n ) ⟨ χ f , E λ ‾ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 . (\chi E_\lambda)(\chi E_{\overline{\lambda}})^*f=\sum_n\chi E_\lambda(Y_n)\langle\chi f,E_{\overline{\lambda}}(Y_n)\rangle_{(L^2(\epsilon\,dx))^3}. ( χ E λ ) ( χ E λ ) ∗ f = n ∑ χ E λ ( Y n ) ⟨ χ f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 .
Since λ ↦ E λ ‾ ( Y ) ( x ) \lambda\mapsto E_{\overline{\lambda}}(Y)(x) λ ↦ E λ ( Y ) ( x ) is holomorphic wherever E λ ‾ E_{\overline{\lambda}} E λ is, so is ( χ E λ ‾ ) ∗ (\chi E_{\overline{\lambda}})^* ( χ E λ ) ∗ . After spatial localization, its mode expansion converges locally uniformly in operator norm away from zero and the zeros of D ℓ j D^j_\ell D ℓ j , by the fixed-argument large-order Bessel estimates: for fixed λ \lambda λ and large ℓ \ell ℓ , we have ∣ D ℓ j ∣ ≥ c ∣ ψ ℓ ( x 1 ) ξ ℓ ( x 0 ) ∣ |D^j_\ell|\ge c|\psi_\ell(x_1)\xi_\ell(x_0)| ∣ D ℓ j ∣ ≥ c ∣ ψ ℓ ( x 1 ) ξ ℓ ( x 0 ) ∣ which controls the matrix coefficients while j ℓ ( k r ) = O ( ( ∣ k R ∣ ) ℓ ( 2 ℓ + 1 ) ! ! ) j_\ell(kr)=O\left(\frac{(|kR|)^\ell}{(2\ell+1)!!}\right) j ℓ ( k r ) = O ( ( 2 ℓ + 1 )!! ( ∣ k R ∣ ) ℓ ) . For real χ \chi χ , the jump identity in Proposition 4.2 gives
χ R ~ λ , E χ = χ R ~ − λ , E χ + i π λ ( χ E λ ) ( χ E λ ‾ ) ∗ . \chi\widetilde{R}_{\lambda,\mathrm{E}}\chi
=
\chi\widetilde{R}_{-\lambda,\mathrm{E}}\chi
+
\frac{i\pi}{\lambda}(\chi E_\lambda)(\chi E_{\overline{\lambda}})^*. χ R λ , E χ = χ R − λ , E χ + λ iπ ( χ E λ ) ( χ E λ ) ∗ .
For Im λ < 0 \operatorname{Im}\lambda<0 Im λ < 0 , both R ~ − λ , E \widetilde{R}_{-\lambda,\mathrm{E}} R − λ , E and ( χ E λ ‾ ) ∗ (\chi E_{\overline{\lambda}})^* ( χ E λ ) ∗ are holomorphic. Thus the continuation has possible poles only at zeros of D ℓ j D^j_\ell D ℓ j . The gradient term − λ 2 ( I − Π E ) -\lambda^2(I-\Pi_{\mathrm{E}}) − λ 2 ( I − Π E ) has no jump; the same argument applies to P H P_H P H . Here the jump identity is used in the form for R ~ ± λ , E \widetilde{R}_{\pm\lambda,E} R ± λ , E and f ∈ L c o m p 2 f \in L^2_{\mathrm{comp}} f ∈ L comp 2 stated in Proposition 4.2. Both sides are holomorphic in their half planes and their boundary values agree on ( C 1 , ∞ ) (C_1,\infty) ( C 1 , ∞ ) , the boundary values of χ R ~ λ , E χ \chi\widetilde{R}_{\lambda,E}\chi χ R λ , E χ from Im λ > 0 \operatorname{Im}\lambda>0 Im λ > 0 being continuous (mode by mode, together with the uniform bound on the tail of the mode sum). Hence they define one holomorphic function across the real axis. By Propositions 2.5 and 2.6, it remains to exclude zeros of the transmission determinants in the lower half-plane. The denominators were originally defined as functions of the physical frequency λ \lambda λ . Therefore we set
D ℓ j ( z ) = D ℓ j ( z r 0 ϵ 0 μ 0 ) z = x 0 = k 0 r 0 x 1 = m 0 z . (45) D_{\ell}^{j}(z)=D_{\ell}^{j}\left(\frac{z}{r_0\sqrt{\epsilon_0\mu_0}}\right) \qquad z=x_0=k_0r_0 \quad x_1=m_0z. \tag*{(45)} D ℓ j ( z ) = D ℓ j ( r 0 ϵ 0 μ 0 z ) z = x 0 = k 0 r 0 x 1 = m 0 z . (45)
Thus
D ℓ 1 ( z ) = η 0 ξ ℓ ( z ) ψ ℓ ′ ( m 0 z ) − η 1 ψ ℓ ( m 0 z ) ξ ℓ ′ ( z ) (46) D_{\ell}^{1}(z)=\eta_0\xi_{\ell}(z)\psi_{\ell}'(m_0z)-\eta_1\psi_{\ell}(m_0z)\xi_{\ell}'(z) \tag*{(46)} D ℓ 1 ( z ) = η 0 ξ ℓ ( z ) ψ ℓ ′ ( m 0 z ) − η 1 ψ ℓ ( m 0 z ) ξ ℓ ′ ( z ) (46)
D ℓ 2 ( z ) = η 0 ψ ℓ ( m 0 z ) ξ ℓ ′ ( z ) − η 1 ξ ℓ ( z ) ψ ℓ ′ ( m 0 z ) . (47) D_{\ell}^{2}(z)=\eta_0\psi_{\ell}(m_0z)\xi_{\ell}'(z)-\eta_1\xi_{\ell}(z)\psi_{\ell}'(m_0z). \tag*{(47)} D ℓ 2 ( z ) = η 0 ψ ℓ ( m 0 z ) ξ ℓ ′ ( z ) − η 1 ξ ℓ ( z ) ψ ℓ ′ ( m 0 z ) . (47)
First we note that
ψ ℓ ′ ( z ) ψ ℓ ( z ) = 1 2 z + J ν ′ ( z ) J ν ( z ) and ξ ℓ ′ ( z ) ξ ℓ ( z ) = 1 2 z + ( H ν ( 1 ) ) ′ ( z ) H ν ( 1 ) ( z ) . (48) \frac{\psi_{\ell}'(z)}{\psi_{\ell}(z)}=\frac{1}{2z}+\frac{J_{\nu}'(z)}{J_{\nu}(z)}
\quad\text{and}\quad
\frac{\xi_{\ell}'(z)}{\xi_{\ell}(z)}=\frac{1}{2z}+\frac{(H_{\nu}^{(1)})'(z)}{H_{\nu}^{(1)}(z)}. \tag*{(48)} ψ ℓ ( z ) ψ ℓ ′ ( z ) = 2 z 1 + J ν ( z ) J ν ′ ( z ) and ξ ℓ ( z ) ξ ℓ ′ ( z ) = 2 z 1 + H ν ( 1 ) ( z ) ( H ν ( 1 ) ) ′ ( z ) . (48)
Assume first 0 < m 0 < 1 0<m_0<1 0 < m 0 < 1 ; the endpoint m 0 = 1 m_0=1 m 0 = 1 is treated at the end of the proof. Let M , K M,K M , K be large constants. Throughout the proof, ν 0 \nu_0 ν 0 is a natural number depending on M , K M,K M , K , and we assume that r 0 ϵ 0 μ 0 C 1 ≥ K ν 0 2 / m 0 r_0\sqrt{\epsilon_0\mu_0}C_1\ge K\nu_0^2/m_0 r 0 ϵ 0 μ 0 C 1 ≥ K ν 0 2 / m 0 (this is the condition Re z ≥ K ν 0 2 / m 0 \operatorname{Re}z\ge K\nu_0^2/m_0 Re z ≥ K ν 0 2 / m 0 in the variable z = r 0 ϵ 0 μ 0 λ z=r_0\sqrt{\epsilon_0\mu_0}\lambda z = r 0 ϵ 0 μ 0 λ ). This implies that for all ν , ν < ν 0 \nu,\nu<\nu_0 ν , ν < ν 0 only region 4 happens. In the rest of the proof C 2 C_2 C 2 denotes the width of the strip in the variable z z z , chosen smaller than both (3.35) and the Region 4 bound; the constant of Theorem 1.4 is then C 2 / ( r 0 ϵ 0 μ 0 ) C_2/(r_0\sqrt{\epsilon_0\mu_0}) C 2 / ( r 0 ϵ 0 μ 0 ) . We use the symmetry D ℓ j ( − z ) = − D ℓ j ( z ) D_{\ell}^{j}(-z)=-D_{\ell}^{j}(z) D ℓ j ( − z ) = − D ℓ j ( z ) so that we only consider Re z > 0 \operatorname{Re}z>0 Re z > 0 . It therefore suffices to consider − C 2 ≤ Im z < 0 -C_2\le\operatorname{Im}z<0 − C 2 ≤ Im z < 0 . We consider the regions:
Region 1: ∣ Re z ∣ ≤ ν − M ν 1 3 |\operatorname{Re}z|\le\nu-M\nu^{\frac{1}{3}} ∣ Re z ∣ ≤ ν − M ν 3 1
Region 2: ∣ Re z − ν ∣ < M ν 1 3 |\operatorname{Re}z-\nu|<M\nu^{\frac{1}{3}} ∣ Re z − ν ∣ < M ν 3 1
Region 3: Re z − ν ≥ M ν 1 3 \operatorname{Re}z-\nu\ge M\nu^{\frac{1}{3}} Re z − ν ≥ M ν 3 1 , ∣ Re z ∣ < K ν 2 m 0 |\operatorname{Re}z|<K\frac{\nu^2}{m_0} ∣ Re z ∣ < K m 0 ν 2
Region 4: ∣ Re z ∣ ≥ K ν 2 m 0 |\operatorname{Re}z|\ge K\frac{\nu^2}{m_0} ∣ Re z ∣ ≥ K m 0 ν 2
For region 1, both z z z and m 0 z m_0z m 0 z lie in the subturning range. Set
β 1 ( z ) = ν 2 − m 0 2 z 2 m 0 z β 0 ( z ) = ν 2 − z 2 z . (49) \beta_1(z)=\frac{\sqrt{\nu^2-m_0^2z^2}}{m_0z}
\qquad
\beta_0(z)=\frac{\sqrt{\nu^2-z^2}}{z}. \tag*{(49)} β 1 ( z ) = m 0 z ν 2 − m 0 2 z 2 β 0 ( z ) = z ν 2 − z 2 . (49)
The uniform estimates in Appendix B.2, together with (2.6), give
ψ ℓ ′ ( m 0 z ) ψ ℓ ( m 0 z ) = 1 2 m 0 z + β 1 ( z ) ( 1 + O ( M − 3 2 ) ) (50) \frac{\psi_{\ell}'(m_0z)}{\psi_{\ell}(m_0z)}
=\frac{1}{2m_0z}+\beta_1(z)\left(1+O\left(M^{-\frac{3}{2}}\right)\right) \tag*{(50)} ψ ℓ ( m 0 z ) ψ ℓ ′ ( m 0 z ) = 2 m 0 z 1 + β 1 ( z ) ( 1 + O ( M − 2 3 ) ) (50)
and
ξ ℓ ′ ( z ) ξ ℓ ( z ) = 1 2 z − β 0 ( z ) ( 1 + O ( M − 3 2 ) ) . (51) \frac{\xi_{\ell}'(z)}{\xi_{\ell}(z)}
=\frac{1}{2z}-\beta_0(z)\left(1+O\left(M^{-\frac{3}{2}}\right)\right). \tag*{(51)} ξ ℓ ( z ) ξ ℓ ′ ( z ) = 2 z 1 − β 0 ( z ) ( 1 + O ( M − 2 3 ) ) . (51)
The TE condition (2.26) would therefore force
β 1 ( 1 + O ( M − 3 2 ) ) + η 1 η 0 β 0 ( 1 + O ( M − 3 2 ) ) = ( η 1 η 0 − 1 m 0 ) 1 2 z . (52) \beta_1\left(1+O\left(M^{-\frac{3}{2}}\right)\right)+\frac{\eta_1}{\eta_0}\beta_0\left(1+O\left(M^{-\frac{3}{2}}\right)\right)=\left(\frac{\eta_1}{\eta_0}-\frac{1}{m_0}\right)\frac{1}{2z}. \tag*{(52)} β 1 ( 1 + O ( M − 2 3 ) ) + η 0 η 1 β 0 ( 1 + O ( M − 2 3 ) ) = ( η 0 η 1 − m 0 1 ) 2 z 1 . (52)
The fact that ∣ Im z ∣ ≤ C 2 |\operatorname{Im} z| \le C_2 ∣ Im z ∣ ≤ C 2 and Re z ≥ C 1 \operatorname{Re} z \ge C_1 Re z ≥ C 1 implies that ∣ arg ( β j ) ∣ ≤ π 4 |\arg(\beta_j)| \le\frac{\pi}{4} ∣ arg ( β j ) ∣ ≤ 4 π for all ν ≥ ν 0 \nu\ge\nu_0 ν ≥ ν 0 , then the real part of the left hand side is bigger than 1 2 Re β 1 \frac{1}{2}\operatorname{Re}\beta_1 2 1 Re β 1 . Because Re ( β 1 ) ≥ 1 − m 0 2 m 0 > 0 \operatorname{Re}(\beta_1) \ge\frac{\sqrt{1-m_0^2}}{m_0} > 0 Re ( β 1 ) ≥ m 0 1 − m 0 2 > 0 and Re β 0 ≥ C M 1 2 ν − 1 3 > 0 \operatorname{Re}\beta_0 \ge CM^{\frac{1}{2}}\nu^{-\frac{1}{3}} > 0 Re β 0 ≥ C M 2 1 ν − 3 1 > 0 . The right hand side is sufficiently smaller provided Re ( z ) \operatorname{Re}(z) Re ( z ) is chosen larger than a constant depending on m 0 m_0 m 0 , η 1 \eta_1 η 1 , η 0 \eta_0 η 0 and M M M . The resonance condition cannot be true for any ν ≥ ν 0 \nu\ge\nu_0 ν ≥ ν 0 , for some ν 0 > 0 \nu_0 > 0 ν 0 > 0 , i.e. D ℓ 1 ( z ) ≠ 0 \mathcal{D}^1_\ell(z) \ne0 D ℓ 1 ( z ) = 0 . The same computation with η 1 \eta_1 η 1 and η 0 \eta_0 η 0 swapped gives that D ℓ 2 ( z ) ≠ 0 \mathcal{D}^2_\ell(z) \ne0 D ℓ 2 ( z ) = 0 .
For region 2 , ∣ Re z − ν ∣ < M ν 1 / 3 |\operatorname{Re} z-\nu| < M\nu^{1/3} ∣ Re z − ν ∣ < M ν 1/3 and, for ν ≥ ν 0 ( m 0 , M ) \nu\ge\nu_0(m_0,M) ν ≥ ν 0 ( m 0 , M ) , m 0 Re z ≤ m 0 ( ν + M ν 1 / 3 ) ≤ ( 1 + m 0 ) ν / 2 m_0\operatorname{Re} z \le m_0(\nu+M\nu^{1/3}) \le(1+m_0)\nu/2 m 0 Re z ≤ m 0 ( ν + M ν 1/3 ) ≤ ( 1 + m 0 ) ν /2 . Thus the interior argument stays uniformly below its turning point, and the subturning estimates give
ψ ℓ ′ ( m 0 z ) ψ ℓ ( m 0 z ) = 1 2 m 0 z + β 1 ( z ) ( 1 + O ( ν − 1 ) ) = 1 − m 0 2 m 0 + O ( M ν − 2 / 3 ) . (53) \frac{\psi_\ell'(m_0z)}{\psi_\ell(m_0z)}=\frac{1}{2m_0z}+\beta_1(z)(1+O(\nu^{-1}))=\frac{\sqrt{1-m_0^2}}{m_0}+O(M\nu^{-2/3}). \tag*{(53)} ψ ℓ ( m 0 z ) ψ ℓ ′ ( m 0 z ) = 2 m 0 z 1 + β 1 ( z ) ( 1 + O ( ν − 1 )) = m 0 1 − m 0 2 + O ( M ν − 2/3 ) . (53)
and
Re ψ ℓ ′ ( m 0 z ) ψ ℓ ( m 0 z ) ≥ 1 2 1 − m 0 2 m 0 > 0 ν ≥ ν 0 . (54) \operatorname{Re}\frac{\psi_\ell'(m_0z)}{\psi_\ell(m_0z)}\ge\frac{1}{2}\frac{\sqrt{1-m_0^2}}{m_0}>0\qquad\nu\ge\nu_0. \tag*{(54)} Re ψ ℓ ( m 0 z ) ψ ℓ ′ ( m 0 z ) ≥ 2 1 m 0 1 − m 0 2 > 0 ν ≥ ν 0 . (54)
This is where the hypothesis that m 0 < 1 m_0<1 m 0 < 1 is used. With w = z w=z w = z and the notation of Proposition B.1, so that τ = ( z − ν ) ν − 1 / 3 \tau=(z-\nu)\nu^{-1/3} τ = ( z − ν ) ν − 1/3 , the additive expansions give
ξ ℓ ′ ( z ) ξ ℓ ( z ) = ν − 1 / 3 p A ^ ′ + O M ( ν − 2 / 3 ) A ^ + O M ( ν − 2 / 3 ) . \frac{\xi_\ell'(z)}{\xi_\ell(z)}=\nu^{-1/3}\frac{p\widehat{A}'+O_M(\nu^{-2/3})}{\widehat{A}+O_M(\nu^{-2/3})}. ξ ℓ ( z ) ξ ℓ ′ ( z ) = ν − 1/3 A + O M ( ν − 2/3 ) p A ′ + O M ( ν − 2/3 ) .
The compact turning set may be enlarged from M M M to M + 1 M+1 M + 1 to include the small imaginary part of τ \tau τ . For real τ \tau τ , σ \sigma σ lies on the rays of arguments − π / 3 -\pi/3 − π /3 and 2 π / 3 2\pi/3 2 π /3 , which avoid the zeros of Ai \operatorname{Ai} Ai . By continuity, ∣ A ^ ∣ ≥ c ( M ) > 0 |\widehat{A}|\ge c(M)>0 ∣ A ∣ ≥ c ( M ) > 0 on a sufficiently small complex neighbourhood of the compact real turning interval. Since ∣ Im τ ∣ ≤ C 2 ν − 1 / 3 |\operatorname{Im}\tau|\le C_2\nu^{-1/3} ∣ Im τ ∣ ≤ C 2 ν − 1/3 , it follows that
∣ ξ ℓ ′ ( z ) ξ ℓ ( z ) ∣ ≤ C ν − 1 / 3 . (55) \left|\frac{\xi_\ell'(z)}{\xi_\ell(z)}\right|\le C\nu^{-1/3}. \tag*{(55)} ξ ℓ ( z ) ξ ℓ ′ ( z ) ≤ C ν − 1/3 . (55)
These can also not be equal to each other for sufficiently large Re z \operatorname{Re} z Re z . Sufficiently large Re z \operatorname{Re} z Re z implies sufficiently large ν \nu ν .
For region 3 , There are all three sub-possibilities as, m 0 Re z < ν − M ν 1 / 3 m_0\operatorname{Re} z<\nu-M\nu^{1/3} m 0 Re z < ν − M ν 1/3 , ∣ m 0 Re z − ν ∣ < M ν 1 / 3 |m_0\operatorname{Re} z-\nu|<M\nu^{1/3} ∣ m 0 Re z − ν ∣ < M ν 1/3 , and m 0 Re z − ν > M ν 1 / 3 m_0\operatorname{Re} z-\nu>M\nu^{1/3} m 0 Re z − ν > M ν 1/3 are all possible. We set q = z ν q=\frac{z}{\nu} q = ν z , it then follows that Re q = Re z ν ≥ 1 + M ν − 2 / 3 \operatorname{Re}q=\frac{\operatorname{Re}z}{\nu}\ge1+M\nu^{-2/3} Re q = ν Re z ≥ 1 + M ν − 2/3 . We also have that ∣ Im z ∣ ≤ C 2 |\operatorname{Im}z|\le C_2 ∣ Im z ∣ ≤ C 2
∣ Im q ∣ ≤ ∣ Im z ∣ ν ≤ C 2 ν ≤ C 2 . (56) |\operatorname{Im}q|\le\frac{|\operatorname{Im}z|}{\nu}\le\frac{C_2}{\nu}\le C_2. \tag*{(56)} ∣ Im q ∣ ≤ ν ∣ Im z ∣ ≤ ν C 2 ≤ C 2 . (56)
Therefore q q q lies in the appropriate sector, ∣ arg ( q ) ∣ ≤ π − δ |\arg(q)|\le\pi-\delta ∣ arg ( q ) ∣ ≤ π − δ for sufficiently large ν \nu ν . The asymptotics in Appendix § 9 .§ 4 hold uniformly for the analytic continuation in q q q to this region of the complex plane. It follows that ζ ( q ) = ζ ( Re q ) + O ( C 2 / ν ) \zeta(q)=\zeta(\operatorname{Re}q)+O(C_2/\nu) ζ ( q ) = ζ ( Re q ) + O ( C 2 / ν ) . With ζ ( q ) \zeta(q) ζ ( q ) as in (B.10), set
W ν ( Re ( q ) ) = e 2 π i 3 ν 2 3 ζ ( Re ( q ) ) , Γ ( Re q ) = 1 Re q ( Re q ) 2 − 1 − ζ ( Re q ) . (57) W_\nu(\operatorname{Re}(q))=e^{\frac{2\pi i}{3}}\nu^{\frac{2}{3}}\zeta(\operatorname{Re}(q)),\qquad\Gamma(\operatorname{Re}q)=\frac{1}{\operatorname{Re}q}\sqrt{\frac{(\operatorname{Re}q)^2-1}{-\zeta(\operatorname{Re}q)}}. \tag*{(57)} W ν ( Re ( q )) = e 3 2 π i ν 3 2 ζ ( Re ( q )) , Γ ( Re q ) = Re q 1 − ζ ( Re q ) ( Re q ) 2 − 1 . (57)
Uniformly in Region 3,
ξ ℓ ′ ( z ) ξ ℓ ( z ) = ν − 1 / 3 e − π i 3 Γ ( q ) Ai ′ ( W ν ( q ) ) Ai ( W ν ( q ) ) + 1 2 ν q + O ( ν − 1 ) . (58) \frac{\xi_\ell'(z)}{\xi_\ell(z)}=\nu^{-1/3}e^{-\frac{\pi i}{3}}\Gamma(q)\frac{\operatorname{Ai}'(W_\nu(q))}{\operatorname{Ai}(W_\nu(q))}+\frac{1}{2\nu q}+O(\nu^{-1}). \tag*{(58)} ξ ℓ ( z ) ξ ℓ ′ ( z ) = ν − 1/3 e − 3 π i Γ ( q ) Ai ( W ν ( q )) Ai ′ ( W ν ( q )) + 2 ν q 1 + O ( ν − 1 ) . (58)
Since ζ ( Re ( q ) ) < 0 \zeta(\operatorname{Re}(q))<0 ζ ( Re ( q )) < 0 on the real ray and ζ \zeta ζ is conformal there, ∣ arg W ν ( Re ( q ) ) + π / 3 ∣ ≤ δ |\arg W_\nu(\operatorname{Re}(q))+\pi/3|\le\delta ∣ arg W ν ( Re ( q )) + π /3∣ ≤ δ for ν ≥ ν 0 \nu\ge\nu_0 ν ≥ ν 0 ; in particular W ν ( Re ( q ) ) W_\nu(\operatorname{Re}(q)) W ν ( Re ( q )) stays in a closed sector disjoint from the zeros of Ai \operatorname{Ai} Ai which are on the negative real axis, and ∣ W ν ( Re ( q ) ) ∣ ≥ c > 0 |W_\nu(\operatorname{Re}(q))| \ge c > 0 ∣ W ν ( Re ( q )) ∣ ≥ c > 0 . This is because Re q − 1 ≥ M ν − 2 3 \operatorname{Re}q-1 \ge M\nu^{-\frac{2}{3}} Re q − 1 ≥ M ν − 3 2 in this region. Along the real part of q q q we have W ν ( Re ( q ) ) W_\nu(\operatorname{Re}(q)) W ν ( Re ( q ))
Ai ′ ( W ν ( q ) ) Ai ( W ν ( q ) ) = − ( W ν ( q ) ) 1 2 ( 1 + O ( ∣ W ∣ − 3 2 ) ) , W ν ( Re q ) = e − π i 3 ν 2 3 ∣ ζ ( Re q ) ∣ . (59) \frac{\operatorname{Ai}'(W_\nu(q))}{\operatorname{Ai}(W_\nu(q))} = -(W_\nu(q))^{\frac{1}{2}}(1+O(|W|^{-\frac{3}{2}})),
\qquad
W_\nu(\operatorname{Re}q)=e^{-\frac{\pi i}{3}}\nu^{\frac{2}{3}}|\zeta(\operatorname{Re}q)|. \tag*{(59)} Ai ( W ν ( q )) Ai ′ ( W ν ( q )) = − ( W ν ( q ) ) 2 1 ( 1 + O ( ∣ W ∣ − 2 3 )) , W ν ( Re q ) = e − 3 π i ν 3 2 ∣ ζ ( Re q ) ∣. (59)
This expansion gives that
ξ ℓ ′ ( z ) ξ ℓ ( z ) = i 1 − q − 2 ( 1 + O ( ν − 1 ( − ζ ) − 3 2 ) ) + O ( ν − 1 ) . (60) \frac{\xi_\ell'(z)}{\xi_\ell(z)}
=i\sqrt{1-q^{-2}}\left(1+O\left(\nu^{-1}(-\zeta)^{-\frac{3}{2}}\right)\right)+O(\nu^{-1}). \tag*{(60)} ξ ℓ ( z ) ξ ℓ ′ ( z ) = i 1 − q − 2 ( 1 + O ( ν − 1 ( − ζ ) − 2 3 ) ) + O ( ν − 1 ) . (60)
At the lower edge since − ζ ∼ M ν − 2 3 -\zeta\sim M\nu^{-\frac{2}{3}} − ζ ∼ M ν − 3 2 there, the relative error O ( ν − 1 ( − ζ ) − 3 2 ) = O ( M − 3 2 ) O(\nu^{-1}(-\zeta)^{-\frac{3}{2}})=O(M^{-\frac{3}{2}}) O ( ν − 1 ( − ζ ) − 2 3 ) = O ( M − 2 3 ) . In the first subcase,
ψ ℓ ′ ( m 0 z ) ψ ℓ ( m 0 z ) = 1 2 m 0 z + β 1 ( z ) ( 1 + O ( M − 3 2 ) ) . (61) \frac{\psi_\ell'(m_0z)}{\psi_\ell(m_0z)}
=\frac{1}{2m_0z}+\beta_1(z)(1+O(M^{-\frac{3}{2}})). \tag*{(61)} ψ ℓ ( m 0 z ) ψ ℓ ′ ( m 0 z ) = 2 m 0 z 1 + β 1 ( z ) ( 1 + O ( M − 2 3 )) . (61)
One then has that ν 2 − m 0 2 ( Re z ) 2 ≥ c M ν 4 3 \nu^2-m_0^2(\operatorname{Re}z)^2\ge cM\nu^{\frac{4}{3}} ν 2 − m 0 2 ( Re z ) 2 ≥ c M ν 3 4 . Therefore Re β 1 ≥ c ′ M 1 2 ν − 1 3 \operatorname{Re}\beta_1\ge c'M^{\frac{1}{2}}\nu^{-\frac{1}{3}} Re β 1 ≥ c ′ M 2 1 ν − 3 1 . On the other side, by (60) , Re ( ξ ℓ ′ ( z ) / ξ ℓ ( z ) ) = − Im γ 0 + O ( ∣ γ 0 ∣ ν − 1 ( − ζ ) − 3 2 ) + O ( ν − 1 ) \operatorname{Re}(\xi_\ell'(z)/\xi_\ell(z))=-\operatorname{Im}\gamma_0+O(|\gamma_0|\nu^{-1}(-\zeta)^{-\frac{3}{2}})+O(\nu^{-1}) Re ( ξ ℓ ′ ( z ) / ξ ℓ ( z )) = − Im γ 0 + O ( ∣ γ 0 ∣ ν − 1 ( − ζ ) − 2 3 ) + O ( ν − 1 ) with γ 0 = 1 − q − 2 \gamma_0=\sqrt{1-q^{-2}} γ 0 = 1 − q − 2 . Since ∣ γ 0 ∣ ≃ ( q − 1 ) 1 2 |\gamma_0|\simeq(q-1)^{\frac{1}{2}} ∣ γ 0 ∣ ≃ ( q − 1 ) 2 1 and − ζ ≃ q − 1 -\zeta\simeq q-1 − ζ ≃ q − 1 for q q q near 1, the middle term is ≲ ν − 1 ( q − 1 ) − 1 ≤ M − 1 ν − 1 3 \lesssim\nu^{-1}(q-1)^{-1}\le M^{-1}\nu^{-\frac{1}{3}} ≲ ν − 1 ( q − 1 ) − 1 ≤ M − 1 ν − 3 1 on region 3, and ∣ Im γ 0 ∣ ≲ C 2 ν − 1 ∣ γ 0 ∣ − 1 ≤ C 2 M − 1 2 ν − 2 3 |\operatorname{Im}\gamma_0|\lesssim C_2\nu^{-1}|\gamma_0|^{-1}\le C_2M^{-\frac{1}{2}}\nu^{-\frac{2}{3}} ∣ Im γ 0 ∣ ≲ C 2 ν − 1 ∣ γ 0 ∣ − 1 ≤ C 2 M − 2 1 ν − 3 2 . Hence ∣ Re ( ξ ℓ ′ / ξ ℓ ) ( z ) ∣ ≤ C ( M − 1 ν − 1 3 + ν − 2 3 ) |\operatorname{Re}(\xi_\ell'/\xi_\ell)(z)|\le C(M^{-1}\nu^{-\frac{1}{3}}+\nu^{-\frac{2}{3}}) ∣ Re ( ξ ℓ ′ / ξ ℓ ) ( z ) ∣ ≤ C ( M − 1 ν − 3 1 + ν − 3 2 ) , which for M M M and ν \nu ν large is smaller than η 0 η 1 ⋅ c ′ M 1 2 ν − 1 3 2 \frac{\eta_0}{\eta_1}\cdot\frac{c'M^{\frac{1}{2}}\nu^{-\frac{1}{3}}}{2} η 1 η 0 ⋅ 2 c ′ M 2 1 ν − 3 1 . So the real part of the left hand side of (2.26) exceeds that of the right hand side (and likewise for (2.27)).
For the second subcase, let ∣ m 0 Re z − ν ∣ < M ν 1 3 |m_0\operatorname{Re}z-\nu|<M\nu^{\frac{1}{3}} ∣ m 0 Re z − ν ∣ < M ν 3 1 . Write m 0 z = ν + τ ν 1 3 m_0z=\nu+\tau\nu^{\frac{1}{3}} m 0 z = ν + τ ν 3 1 , so ∣ Re τ ∣ < M |\operatorname{Re}\tau|<M ∣ Re τ ∣ < M and ∣ Im τ ∣ ≤ m 0 C 2 ν − 1 3 |\operatorname{Im}\tau|\le m_0C_2\nu^{-\frac{1}{3}} ∣ Im τ ∣ ≤ m 0 C 2 ν − 3 1 . Use s , A , A ′ s,A,A' s , A , A ′ from Proposition B.1 with w = m 0 z w=m_0z w = m 0 z . The exterior variable satisfies Re q ≥ m 0 − 1 ( 1 − M ν − 2 3 ) ≥ 1 2 ( 1 + m 0 − 1 ) > 1 \operatorname{Re}q\ge m_0^{-1}(1-M\nu^{-\frac{2}{3}})\ge\frac{1}{2}(1+m_0^{-1})>1 Re q ≥ m 0 − 1 ( 1 − M ν − 3 2 ) ≥ 2 1 ( 1 + m 0 − 1 ) > 1 , so it is uniformly bounded below in this oscillatory regime. Since ξ ℓ ( z ) ≠ 0 \xi_\ell(z)\ne0 ξ ℓ ( z ) = 0 ,
L + ( z ) : = ξ ℓ ′ ( z ) ξ ℓ ( z ) = i γ 0 ( z ) + O ( ν − 1 ) , γ 0 ( z ) = 1 − m 0 2 + O ( M ν − 2 3 ) . L_+(z):=\frac{\xi_\ell'(z)}{\xi_\ell(z)}
=i\gamma_0(z)+O(\nu^{-1}),
\qquad
\gamma_0(z)=\sqrt{1-m_0^2}+O(M\nu^{-\frac{2}{3}}). L + ( z ) := ξ ℓ ( z ) ξ ℓ ′ ( z ) = i γ 0 ( z ) + O ( ν − 1 ) , γ 0 ( z ) = 1 − m 0 2 + O ( M ν − 3 2 ) .
Divide D ℓ 1 ( z ) \mathcal{D}_\ell^1(z) D ℓ 1 ( z ) by − π 2 1 6 ν − 1 6 ξ ℓ ( z ) -\sqrt{\pi}\,2^{\frac{1}{6}}\nu^{-\frac{1}{6}}\xi_\ell(z) − π 2 6 1 ν − 6 1 ξ ℓ ( z ) and insert the expansions from the Appendix (B.3), (B.4), (B.7). These are additive, so no division by ψ ℓ ( m 0 z ) \psi_\ell(m_0z) ψ ℓ ( m 0 z ) occurs. The result is
− D ℓ 1 ( z ) π 2 1 6 ν − 1 6 ξ ℓ ( z ) = η 0 A ′ + i η 1 γ 0 2 − 1 3 ν 1 3 A + F , ∣ F ∣ ≤ C M ν − 1 3 ( ∣ A ∣ + ∣ A ′ ∣ ) . (62) \frac{-\mathcal{D}_\ell^1(z)}{\sqrt{\pi}\,2^{\frac{1}{6}}\nu^{-\frac{1}{6}}\xi_\ell(z)}
=\eta_0A'+i\eta_1\gamma_0\,2^{-\frac{1}{3}}\nu^{\frac{1}{3}}A+\mathcal{F},
\qquad
|\mathcal{F}|\le CM\nu^{-\frac{1}{3}}(|A|+|A'|). \tag*{(62)} π 2 6 1 ν − 6 1 ξ ℓ ( z ) − D ℓ 1 ( z ) = η 0 A ′ + i η 1 γ 0 2 − 3 1 ν 3 1 A + F , ∣ F ∣ ≤ C M ν − 3 1 ( ∣ A ∣ + ∣ A ′ ∣ ) . (62)
where we used that ∣ A ∣ + ∣ A ′ ∣ ≥ c M > 0 |A|+|A'|\ge c_M>0 ∣ A ∣ + ∣ A ′ ∣ ≥ c M > 0 on the compact set ∣ s ∣ ≤ 2 1 3 ( M + 1 ) |s|\le2^{\frac{1}{3}}(M+1) ∣ s ∣ ≤ 2 3 1 ( M + 1 ) (Ai and Ai′ ' ′ have no common zero). We argue by contradiction that D ℓ 1 ( z ) = 0 \mathcal{D}_\ell^1(z)=0 D ℓ 1 ( z ) = 0 . Then ν 1 3 ∣ A ∣ ≤ C ( ∣ A ′ ∣ + ∣ F ∣ ) \nu^{\frac{1}{3}}|A|\le C(|A'|+|\mathcal{F}|) ν 3 1 ∣ A ∣ ≤ C ( ∣ A ′ ∣ + ∣ F ∣ ) , so ∣ A ∣ ≤ C M ν − 1 3 |A|\le CM\nu^{-\frac{1}{3}} ∣ A ∣ ≤ C M ν − 3 1 and ∣ A ′ ∣ ≥ c M / 2 |A'|\ge c_M/2 ∣ A ′ ∣ ≥ c M /2 . Thus s s s lies within C M ν − 1 3 CM\nu^{-\frac{1}{3}} C M ν − 3 1 of a zero a k < 0 a_k<0 a k < 0 of Ai. Since Ai ′ ′ ( a k ) = a k Ai ( a k ) = 0 \operatorname{Ai}''(a_k)=a_k\operatorname{Ai}(a_k)=0 Ai ′′ ( a k ) = a k Ai ( a k ) = 0 , we have A / A ′ = ( s − a k ) ( 1 + O ( ∣ s − a k ∣ 2 ) ) A/A'=(s-a_k)(1+O(|s-a_k|^2)) A / A ′ = ( s − a k ) ( 1 + O ( ∣ s − a k ∣ 2 )) . Solving (62) for A / A ′ A/A' A / A ′ gives
s − a k = i 2 1 3 η 0 η 1 γ 0 ν 1 3 ( 1 + O M ( ν − 1 3 ) ) , s-a_k=\frac{i\,2^{\frac{1}{3}}\eta_0}{\eta_1\gamma_0\nu^{\frac{1}{3}}}\left(1+O_M(\nu^{-\frac{1}{3}})\right), s − a k = η 1 γ 0 ν 3 1 i 2 3 1 η 0 ( 1 + O M ( ν − 3 1 ) ) ,
so that
Im s = 2 1 3 η 0 η 1 1 − m 0 2 ν − 1 3 ( 1 + O M ( ν − 1 3 ) ) . \operatorname{Im}s=\frac{2^{\frac{1}{3}}\eta_0}{\eta_1\sqrt{1-m_0^2}}\nu^{-\frac{1}{3}}\left(1+O_M(\nu^{-\frac{1}{3}})\right). Im s = η 1 1 − m 0 2 2 3 1 η 0 ν − 3 1 ( 1 + O M ( ν − 3 1 ) ) .
On the other hand Im s = − 2 1 3 m 0 Im z ν − 1 3 ≤ 2 1 3 m 0 C 2 ν − 1 3 \operatorname{Im}s=-2^{\frac{1}{3}}m_0\operatorname{Im}z\,\nu^{-\frac{1}{3}}\le2^{\frac{1}{3}}m_0C_2\nu^{-\frac{1}{3}} Im s = − 2 3 1 m 0 Im z ν − 3 1 ≤ 2 3 1 m 0 C 2 ν − 3 1 . This is a contradiction for ν ≥ ν 0 ( M ) \nu\ge\nu_0(M) ν ≥ ν 0 ( M ) as soon as
C 2 < η 0 2 m 0 η 1 1 − m 0 2 (TE) , C 2 < η 1 2 m 0 η 0 1 − m 0 2 (TM) . (63) C_2<\frac{\eta_0}{2m_0\eta_1\sqrt{1-m_0^2}}\quad\text{(TE)},\qquad C_2<\frac{\eta_1}{2m_0\eta_0\sqrt{1-m_0^2}}\quad\text{(TM)}. \tag*{(63)} C 2 < 2 m 0 η 1 1 − m 0 2 η 0 (TE) , C 2 < 2 m 0 η 0 1 − m 0 2 η 1 (TM) . (63)
Both conditions follow from (77) , because η 0 η 1 ( η 0 + η 1 ) 2 ≤ min ( η 0 η 1 , η 1 η 0 ) \frac{\eta_0\eta_1}{(\eta_0+\eta_1)^2}\le\min\left(\frac{\eta_0}{\eta_1},\frac{\eta_1}{\eta_0}\right) ( η 0 + η 1 ) 2 η 0 η 1 ≤ min ( η 1 η 0 , η 0 η 1 ) and 1 − m 0 2 ≤ ( 1 − m 0 2 ) − 1 / 2 \sqrt{1-m_0^2}\le(1-m_0^2)^{-1/2} 1 − m 0 2 ≤ ( 1 − m 0 2 ) − 1/2 .
For the third subcase, m 0 Re z ≥ ν + M ν 1 3 m_0\operatorname{Re}z\ge\nu+M\nu^{\frac{1}{3}} m 0 Re z ≥ ν + M ν 3 1 . Here ψ ℓ ( m 0 z ) \psi_\ell(m_0z) ψ ℓ ( m 0 z ) is oscillatory and has zeros, so ψ ℓ ′ ( m 0 z ) / ψ ℓ ( m 0 z ) \psi_\ell'(m_0z)/\psi_\ell(m_0z) ψ ℓ ′ ( m 0 z ) / ψ ℓ ( m 0 z ) is an unbounded function of z z z and cannot be compared termwise with (60) . Throughout this subcase set, with principal branches, we set
γ 0 ( z ) = 1 − ν 2 z 2 = 1 − q − 2 , γ 1 ( z ) = 1 − ν 2 m 0 2 z 2 , (64) \gamma_0(z)=\sqrt{1-\frac{\nu^2}{z^2}}=\sqrt{1-q^{-2}},\qquad\gamma_1(z)=\sqrt{1-\frac{\nu^2}{m_0^2z^2}}, \tag*{(64)} γ 0 ( z ) = 1 − z 2 ν 2 = 1 − q − 2 , γ 1 ( z ) = 1 − m 0 2 z 2 ν 2 , (64)
so that i γ 0 i\gamma_0 i γ 0 is precisely the leading term of (60) . Because m 0 < 1 m_0<1 m 0 < 1 , the subcase hypothesis m 0 Re z > ν m_0\operatorname{Re}z>\nu m 0 Re z > ν already forces Re z > ν / m 0 \operatorname{Re}z>\nu/m_0 Re z > ν / m 0 : the exterior variable is automatically far beyond its own turning point, and we do not sit near the lower edge of region 3. Consequently, for ν ≥ ν 0 \nu\ge\nu_0 ν ≥ ν 0 and ∣ Im z ∣ ≤ C 2 |\operatorname{Im}z|\le C_2 ∣ Im z ∣ ≤ C 2 ,
Re γ 0 ( z ) ≥ 1 − m 0 2 + O ( ν − 1 ) , ∣ γ 0 ( z ) ∣ ≤ 1 + O ( ν − 1 ) , (65) \operatorname{Re}\gamma_0(z)\ge\sqrt{1-m_0^2}+O(\nu^{-1}),\qquad|\gamma_0(z)|\le1+O(\nu^{-1}), \tag*{(65)} Re γ 0 ( z ) ≥ 1 − m 0 2 + O ( ν − 1 ) , ∣ γ 0 ( z ) ∣ ≤ 1 + O ( ν − 1 ) , (65)
while ( m 0 Re z ) 2 − ν 2 ≥ 2 M ν 4 3 (m_0\operatorname{Re}z)^2-\nu^2\ge2M\nu^{\frac{4}{3}} ( m 0 Re z ) 2 − ν 2 ≥ 2 M ν 3 4 gives
c M 1 2 ν − 1 3 ≤ Re γ 1 ( z ) ≤ ∣ γ 1 ( z ) ∣ ≤ 1 + O ( ν − 1 ) . (66) cM^{\frac{1}{2}}\nu^{-\frac{1}{3}}\le\operatorname{Re}\gamma_1(z)\le|\gamma_1(z)|\le1+O(\nu^{-1}). \tag*{(66)} c M 2 1 ν − 3 1 ≤ Re γ 1 ( z ) ≤ ∣ γ 1 ( z ) ∣ ≤ 1 + O ( ν − 1 ) . (66)
Put x = Re z x=\operatorname{Re}z x = Re z , y = Im z y=\operatorname{Im}z y = Im z and g j = γ j ( x ) > 0 g_j=\gamma_j(x)>0 g j = γ j ( x ) > 0 . Since ∣ γ j ′ ∣ ≤ C / ( ν g j ) |\gamma_j'|\le C/(\nu g_j) ∣ γ j ′ ∣ ≤ C / ( ν g j ) along x + i s x+is x + i s , ∣ s ∣ ≤ C 2 |s|\le C_2 ∣ s ∣ ≤ C 2 , and g 1 2 ≥ c M ν − 2 / 3 g_1^2\ge cM\nu^{-2/3} g 1 2 ≥ c M ν − 2/3 , g 0 ≥ 1 − m 0 2 g_0\ge\sqrt{1-m_0^2} g 0 ≥ 1 − m 0 2 , we have uniformly
γ j ( x + i s ) = g j ( 1 + O ( δ ν ) ) , δ ν = 1 + C 2 M ν 1 / 3 , j = 0 , 1. \gamma_j(x+is)=g_j(1+O(\delta_\nu)),\qquad\delta_\nu=\frac{1+C_2}{M\nu^{1/3}},\qquad j=0,1. γ j ( x + i s ) = g j ( 1 + O ( δ ν )) , δ ν = M ν 1/3 1 + C 2 , j = 0 , 1.
For fixed M M M , increase ν 0 \nu_0 ν 0 so that δ ν ≤ M − 3 / 2 \delta_\nu\le M^{-3/2} δ ν ≤ M − 3/2 . We use g j g_j g j in the real inequalities below. We write ε ( w ) = ν 2 ( ( Re w ) 2 − ν 2 ) − 3 / 2 \varepsilon(w)=\nu^2((\operatorname{Re}w)^2-\nu^2)^{-3/2} ε ( w ) = ν 2 (( Re w ) 2 − ν 2 ) − 3/2 for the relative error in the Debye expansions in the oscillatory regime, cf. §10.19(ii) of [27 ] and pp. 419–425 of [26 ] , which by the uniform complex-argument expansions in §10.20 of [27 ] are valid uniformly for ∣ arg w ∣ ≤ π − δ |\arg w|\le\pi-\delta ∣ arg w ∣ ≤ π − δ . In this subcase ε ( m 0 z ) = O ( M − 3 / 2 ) \varepsilon(m_0z)=O(M^{-3/2}) ε ( m 0 z ) = O ( M − 3/2 ) and ε ( z ) = O ( ν − 1 ) \varepsilon(z)=O(\nu^{-1}) ε ( z ) = O ( ν − 1 ) .
Step 1: an exact reflection identity. Put ξ ℓ ± ( w ) = w h ℓ ( 1 , 2 ) ( w ) \xi_\ell^\pm(w)=w h_\ell^{(1,2)}(w) ξ ℓ ± ( w ) = w h ℓ ( 1 , 2 ) ( w ) , so that ξ ℓ + = ξ ℓ \xi_\ell^+=\xi_\ell ξ ℓ + = ξ ℓ and ψ ℓ = 1 2 ( ξ ℓ + + ξ ℓ − ) \psi_\ell=\frac{1}{2}(\xi_\ell^++\xi_\ell^-) ψ ℓ = 2 1 ( ξ ℓ + + ξ ℓ − ) , and set
L ± ( w ) = ( ξ ℓ ± ) ′ ( w ) ξ ℓ ± ( w ) = 1 2 w + ( H ν ( 1 , 2 ) ) ′ ( w ) H ν ( 1 , 2 ) ( w ) , L_\pm(w)=\frac{(\xi_\ell^\pm)'(w)}{\xi_\ell^\pm(w)}=\frac{1}{2w}+\frac{(H_\nu^{(1,2)})'(w)}{H_\nu^{(1,2)}(w)}, L ± ( w ) = ξ ℓ ± ( w ) ( ξ ℓ ± ) ′ ( w ) = 2 w 1 + H ν ( 1 , 2 ) ( w ) ( H ν ( 1 , 2 ) ) ′ ( w ) ,
R ( w ) = ξ ℓ − ( w ) ξ ℓ + ( w ) = H ν ( 2 ) ( w ) H ν ( 1 ) ( w ) . (67) R(w)=\frac{\xi_\ell^-(w)}{\xi_\ell^+(w)}=\frac{H_\nu^{(2)}(w)}{H_\nu^{(1)}(w)}. \tag*{(67)} R ( w ) = ξ ℓ + ( w ) ξ ℓ − ( w ) = H ν ( 1 ) ( w ) H ν ( 2 ) ( w ) . (67)
Dividing D ℓ 1 ( z ) \mathcal{D}_\ell^1(z) D ℓ 1 ( z ) by 1 2 ξ ℓ ( z ) ξ ℓ + ( m 0 z ) \frac{1}{2}\xi_\ell(z)\xi_\ell^+(m_0z) 2 1 ξ ℓ ( z ) ξ ℓ + ( m 0 z ) , which is non-zero because the Debye expansions have relative error < 1 2 <\frac{1}{2} < 2 1 there, gives the identity
2 D ℓ 1 ( z ) ξ ℓ ( z ) ξ ℓ + ( m 0 z ) = [ η 0 L + ( m 0 z ) − η 1 L + ( z ) ] + R ( m 0 z ) [ η 0 L − ( m 0 z ) − η 1 L + ( z ) ] . (68) \frac{2\mathcal{D}_\ell^1(z)}{\xi_\ell(z)\xi_\ell^+(m_0z)}
=\left[\eta_0L_+(m_0z)-\eta_1L_+(z)\right]+R(m_0z)\left[\eta_0L_-(m_0z)-\eta_1L_+(z)\right]. \tag*{(68)} ξ ℓ ( z ) ξ ℓ + ( m 0 z ) 2 D ℓ 1 ( z ) = [ η 0 L + ( m 0 z ) − η 1 L + ( z ) ] + R ( m 0 z ) [ η 0 L − ( m 0 z ) − η 1 L + ( z ) ] . (68)
Hence, as long as the second bracket does not vanish,
D ℓ 1 ( z ) = 0 ⟺ R ( m 0 z ) = F ( z ) : = η 1 L + ( z ) − η 0 L + ( m 0 z ) η 0 L − ( m 0 z ) − η 1 L + ( z ) . (69) \mathcal{D}_\ell^1(z)=0\Longleftrightarrow R(m_0z)=\mathcal{F}(z):=\frac{\eta_1L_+(z)-\eta_0L_+(m_0z)}{\eta_0L_-(m_0z)-\eta_1L_+(z)}. \tag*{(69)} D ℓ 1 ( z ) = 0 ⟺ R ( m 0 z ) = F ( z ) := η 0 L − ( m 0 z ) − η 1 L + ( z ) η 1 L + ( z ) − η 0 L + ( m 0 z ) . (69)
The point of (3.27) is that the two sides are compared in modulus : the left side is forced to be almost unimodular by the strip, and the right side is a Fresnel reflection coefficient, which is strictly smaller than 1 exactly because m 0 < 1 m_0 < 1 m 0 < 1 .
Step 2: the interior Hankel quotient is almost unimodular. For real w > 0 w > 0 w > 0 and real order we have H ν ( 2 ) ( w ) = H ν ( 1 ) ( w ) ‾ H_\nu^{(2)}(w) = \overline{H_\nu^{(1)}(w)} H ν ( 2 ) ( w ) = H ν ( 1 ) ( w ) , so ∣ R ( w ) ∣ = 1 |\mathcal{R}(w)| = 1 ∣ R ( w ) ∣ = 1 exactly on the real axis. Since ∂ w log R = L − − L + = − 2 i γ ( w ) ( 1 + O ( ε ( w ) ) ) + O ( ∣ w ∣ − 1 ) \partial_w \log R = L_- - L_+ = -2i\gamma(w)(1 + O(\varepsilon(w))) + O(|w|^{-1}) ∂ w log R = L − − L + = − 2 iγ ( w ) ( 1 + O ( ε ( w ))) + O ( ∣ w ∣ − 1 ) with γ ( w ) = 1 − ν 2 w − 2 \gamma(w) = \sqrt{1-\nu^2w^{-2}} γ ( w ) = 1 − ν 2 w − 2 , integrating along the vertical segment from m 0 Re z m_0\operatorname{Re}z m 0 Re z to m 0 z m_0z m 0 z gives
log ∣ R ( m 0 z ) ∣ = Re ∫ 0 Im z i m 0 ( L − − L + ) ( m 0 ( Re z + i s ) ) d s = 2 m 0 ∫ 0 y Re γ 1 ( x + i s ) d s + O ( C 2 g 1 M − 3 / 2 + ν − 1 ) . (70) \log|\mathcal{R}(m_0z)| = \operatorname{Re}\int_0^{\operatorname{Im}z} im_0(L_- - L_+)(m_0(\operatorname{Re}z+is))\,ds
= 2m_0\int_0^y \operatorname{Re}\gamma_1(x+is)\,ds + O(C_2g_1M^{-3/2}+\nu^{-1}). \tag*{(70)} log ∣ R ( m 0 z ) ∣ = Re ∫ 0 Im z i m 0 ( L − − L + ) ( m 0 ( Re z + i s )) d s = 2 m 0 ∫ 0 y Re γ 1 ( x + i s ) d s + O ( C 2 g 1 M − 3/2 + ν − 1 ) . (70)
so that
∣ log ∣ R ( m 0 z ) ∣ ∣ ≤ 2 m 0 g 1 ∣ y ∣ ( 1 + C M − 3 / 2 ) + C ν − 1 , |\log|\mathcal{R}(m_0z)|| \le2m_0g_1|y|(1+CM^{-3/2})+C\nu^{-1}, ∣ log ∣ R ( m 0 z ) ∣∣ ≤ 2 m 0 g 1 ∣ y ∣ ( 1 + C M − 3/2 ) + C ν − 1 ,
∣ R ( m 0 z ) ∣ ≥ 1 − 2 m 0 C 2 g 1 ( 1 + C M − 3 / 2 ) − C ν − 1 . (71) |\mathcal{R}(m_0z)| \ge1-2m_0C_2g_1(1+CM^{-3/2})-C\nu^{-1}. \tag*{(71)} ∣ R ( m 0 z ) ∣ ≥ 1 − 2 m 0 C 2 g 1 ( 1 + C M − 3/2 ) − C ν − 1 . (71)
The factor g 1 g_1 g 1 in (3.29) is the whole point and must not be discarded: it is what compensates the degeneration of the right-hand side of (3.27) as the interior ray becomes grazing.
Step 3: the right-hand side of (3.27) is strictly sub-unimodular. The Debye expansions give
L + ( z ) = i γ 0 + O ( ν − 1 ) , L ± ( m 0 z ) = ± i γ 1 + O ( g 1 M − 3 / 2 ) + O ( ν − 1 ) . (72) L_+(z)=i\gamma_0+O(\nu^{-1}),\qquad L_\pm(m_0z)=\pm i\gamma_1+O(g_1M^{-3/2})+O(\nu^{-1}). \tag*{(72)} L + ( z ) = i γ 0 + O ( ν − 1 ) , L ± ( m 0 z ) = ± i γ 1 + O ( g 1 M − 3/2 ) + O ( ν − 1 ) . (72)
Put a = η 0 γ 1 a=\eta_0\gamma_1 a = η 0 γ 1 and b = η 1 γ 0 b=\eta_1\gamma_0 b = η 1 γ 0 . The denominator in (3.27) equals − i ( a + b ) + O ( g 1 M − 3 / 2 + ν − 1 ) -i(a+b)+O(g_1M^{-3/2}+\nu^{-1}) − i ( a + b ) + O ( g 1 M − 3/2 + ν − 1 ) and, by (3.23), has modulus at least 1 2 η 1 1 − m 0 2 > 0 \frac{1}{2}\eta_1\sqrt{1-m_0^2}>0 2 1 η 1 1 − m 0 2 > 0 ; in particular the second bracket of (3.26) does not vanish, and
∣ F ( z ) ∣ ≤ ∣ b − a b + a ∣ + C ( g 1 M − 3 / 2 + ν − 1 ) . (73) |\mathcal{F}(z)|\le\left|\frac{b-a}{b+a}\right|+C(g_1M^{-3/2}+\nu^{-1}). \tag*{(73)} ∣ F ( z ) ∣ ≤ b + a b − a + C ( g 1 M − 3/2 + ν − 1 ) . (73)
From ∣ b + a ∣ 2 − ∣ b − a ∣ 2 = 4 Re ( a b ‾ ) |b+a|^2-|b-a|^2=4\operatorname{Re}(a\overline{b}) ∣ b + a ∣ 2 − ∣ b − a ∣ 2 = 4 Re ( a b ) , together with ∣ a ∣ + ∣ b ∣ ≤ ( η 0 + η 1 ) ( 1 + C δ ν ) |a|+|b|\le(\eta_0+\eta_1)(1+C\delta\nu) ∣ a ∣ + ∣ b ∣ ≤ ( η 0 + η 1 ) ( 1 + C δ ν ) and Re ( a b ‾ ) ≥ η 0 η 1 g 1 g 0 ( 1 − C δ ν ) \operatorname{Re}(a\overline{b})\ge\eta_0\eta_1g_1g_0(1-C\delta\nu) Re ( a b ) ≥ η 0 η 1 g 1 g 0 ( 1 − C δ ν ) , we obtain, after increasing ν 0 \nu_0 ν 0 ,
1 − ∣ b − a b + a ∣ = 4 Re ( a b ‾ ) ∣ b + a ∣ ( ∣ b + a ∣ + ∣ b − a ∣ ) ≥ 2 Re ( a b ‾ ) ( ∣ a ∣ + ∣ b ∣ ) 2 ≥ c ∗ g 1 , 1-\left|\frac{b-a}{b+a}\right|
=\frac{4\operatorname{Re}(a\overline{b})}{|b+a|(|b+a|+|b-a|)}
\ge\frac{2\operatorname{Re}(a\overline{b})}{(|a|+|b|)^2}\ge c_*g_1, 1 − b + a b − a = ∣ b + a ∣ ( ∣ b + a ∣ + ∣ b − a ∣ ) 4 Re ( a b ) ≥ ( ∣ a ∣ + ∣ b ∣ ) 2 2 Re ( a b ) ≥ c ∗ g 1 ,
c ∗ : = η 0 η 1 1 − m 0 2 ( η 0 + η 1 ) 2 . (74) c_*:=\frac{\eta_0\eta_1\sqrt{1-m_0^2}}{(\eta_0+\eta_1)^2}. \tag*{(74)} c ∗ := ( η 0 + η 1 ) 2 η 0 η 1 1 − m 0 2 . (74)
It is exactly here that the hypothesis m 0 < 1 m_0 < 1 m 0 < 1 is used: it keeps Re γ 0 \operatorname{Re}\gamma_0 Re γ 0 bounded away from zero even when γ 1 → 0 \gamma_1\to0 γ 1 → 0 , i.e. it rules out total internal reflection at grazing interior incidence.
Step 4: conclusion. If D ℓ 1 ( z ) = 0 D_\ell^1(z)=0 D ℓ 1 ( z ) = 0 in this subcase, then (3.27), (3.29), (3.31) and (3.32) give
1 − 2 m 0 C 2 g 1 ( 1 + C M − 3 / 2 ) − C ν − 1 ≤ ∣ R ( m 0 z ) ∣ = ∣ F ( z ) ∣ 1-2m_0C_2g_1(1+CM^{-3/2})-C\nu^{-1}\le|\mathcal{R}(m_0z)|=|\mathcal{F}(z)| 1 − 2 m 0 C 2 g 1 ( 1 + C M − 3/2 ) − C ν − 1 ≤ ∣ R ( m 0 z ) ∣ = ∣ F ( z ) ∣
≤ 1 − c ∗ g 1 + C ( g 1 M − 3 / 2 + ν − 1 ) . (75) \le1-c_*g_1+C(g_1M^{-3/2}+\nu^{-1}). \tag*{(75)} ≤ 1 − c ∗ g 1 + C ( g 1 M − 3/2 + ν − 1 ) . (75)
Dividing by g 1 g_1 g 1 and using the lower bound (3.24) on the real axis in the terms C ν − 1 / g 1 C\nu^{-1}/g_1 C ν − 1 / g 1 , we arrive at
c ∗ ≤ 2 m 0 C 2 ( 1 + C M − 3 / 2 ) + C M − 3 / 2 + C M − 1 / 2 ν − 3 / 2 . (76) c_*\le2m_0C_2(1+CM^{-3/2})+CM^{-3/2}+CM^{-1/2}\nu^{-3/2}. \tag*{(76)} c ∗ ≤ 2 m 0 C 2 ( 1 + C M − 3/2 ) + C M − 3/2 + C M − 1/2 ν − 3/2 . (76)
This is false for all M ≥ M 0 ( m 0 , η 0 , η 1 ) M \ge M_0(m_0,\eta_0,\eta_1) M ≥ M 0 ( m 0 , η 0 , η 1 ) and ν ≥ ν 0 ( m 0 , η 0 , η 1 , M ) \nu\ge\nu_0(m_0,\eta_0,\eta_1,M) ν ≥ ν 0 ( m 0 , η 0 , η 1 , M ) as soon as
C 2 < c ∗ 4 m 0 = 1 − m 0 2 4 m 0 η 0 η 1 ( η 0 + η 1 ) 2 . (77) C_2 < \frac{c_*}{4m_0} = \frac{\sqrt{1-m_0^2}}{4m_0}\frac{\eta_0\eta_1}{(\eta_0+\eta_1)^2}. \tag*{(77)} C 2 < 4 m 0 c ∗ = 4 m 0 1 − m 0 2 ( η 0 + η 1 ) 2 η 0 η 1 . (77)
Hence D ℓ 1 ( z ) ≠ 0 D_\ell^1(z) \ne0 D ℓ 1 ( z ) = 0 in the third subcase. Since c ∗ c_* c ∗ is symmetric in η 0 , η 1 \eta_0,\eta_1 η 0 , η 1 , interchanging η 0 \eta_0 η 0 and η 1 \eta_1 η 1 gives D ℓ 2 ( z ) ≠ 0 D_\ell^2(z) \ne0 D ℓ 2 ( z ) = 0 under the same threshold (77) .
For region 4 , both arguments satisfy ∣ w ∣ ≥ K ν 2 |w| \ge K\nu^2 ∣ w ∣ ≥ K ν 2 . The leading-order estimates (B.12)–(B.15) therefore have uniform errors O ( K − 1 ) O(K^{-1}) O ( K − 1 ) at w = z , m 0 z w=z,m_0z w = z , m 0 z . With ω \omega ω as in (B.11) and θ = ω ( m 0 z ) \theta=\omega(m_0z) θ = ω ( m 0 z ) , they give
e − i ω ( z ) D ℓ 1 ( z ) = − η 0 sin θ − i η 1 cos θ + O ( K − 1 ) , (78) \mathrm{e}^{-\mathrm{i}\omega(z)}D_\ell^1(z)=-\eta_0\sin\theta-\mathrm{i}\eta_1\cos\theta+O(K^{-1}), \tag*{(78)} e − i ω ( z ) D ℓ 1 ( z ) = − η 0 sin θ − i η 1 cos θ + O ( K − 1 ) , (78)
e − i ω ( z ) D ℓ 2 ( z ) = i η 0 cos θ + η 1 sin θ + O ( K − 1 ) . (79) \mathrm{e}^{-\mathrm{i}\omega(z)}D_\ell^2(z)=\mathrm{i}\eta_0\cos\theta+\eta_1\sin\theta+O(K^{-1}). \tag*{(79)} e − i ω ( z ) D ℓ 2 ( z ) = i η 0 cos θ + η 1 sin θ + O ( K − 1 ) . (79)
For real θ \theta θ each leading line has modulus at least min ( η 0 , η 1 ) \min(\eta_0,\eta_1) min ( η 0 , η 1 ) . The leading line of (78) vanishes iff e 2 i θ = η 0 + η 1 η 0 − η 1 \mathrm{e}^{2\mathrm{i}\theta}=\frac{\eta_0+\eta_1}{\eta_0-\eta_1} e 2 i θ = η 0 − η 1 η 0 + η 1 , and that of (79) iff e 2 i θ = η 1 + η 0 η 1 − η 0 \mathrm{e}^{2\mathrm{i}\theta}=\frac{\eta_1+\eta_0}{\eta_1-\eta_0} e 2 i θ = η 1 − η 0 η 1 + η 0 ; in both cases this forces
Im θ = − y ∗ , y ∗ : = 1 2 log ∣ η 0 + η 1 η 0 − η 1 ∣ ∈ ( 0 , ∞ ] (80) \operatorname{Im}\theta=-y_*,\qquad y_*:=\frac{1}{2}\log\left|\frac{\eta_0+\eta_1}{\eta_0-\eta_1}\right|\in(0,\infty] \tag*{(80)} Im θ = − y ∗ , y ∗ := 2 1 log η 0 − η 1 η 0 + η 1 ∈ ( 0 , ∞ ] (80)
(y ∗ = ∞ y_*=\infty y ∗ = ∞ if η 0 = η 1 \eta_0=\eta_1 η 0 = η 1 ). Fix y 1 : = 1 2 y ∗ y_1:=\frac{1}{2}y_* y 1 := 2 1 y ∗ (any y 1 < y ∗ y_1<y_* y 1 < y ∗ works; y 1 : = 1 y_1:=1 y 1 := 1 if y ∗ = ∞ y_*=\infty y ∗ = ∞ ). The moduli of the leading lines are π \pi π -periodic in Re θ \operatorname{Re}\theta Re θ , and the leading lines have no zeros on { ∣ Im θ ∣ ≤ y 1 } \{|\operatorname{Im}\theta|\le y_1\} { ∣ Im θ ∣ ≤ y 1 } , so their modulus is ≥ c ( y 1 , η 0 , η 1 ) > 0 \ge c(y_1,\eta_0,\eta_1)>0 ≥ c ( y 1 , η 0 , η 1 ) > 0 there. Choosing K K K large makes the uniform O ( K − 1 ) O(K^{-1}) O ( K − 1 ) remainders in (78) –(79) smaller than c / 2 c/2 c /2 . Since Im θ = m 0 Im z \operatorname{Im}\theta=m_0\operatorname{Im}z Im θ = m 0 Im z , Region 4 contributes no zeros for
∣ Im z ∣ ≤ y 1 m 0 = 1 4 m 0 log ∣ η 0 + η 1 η 0 − η 1 ∣ ( η 0 ≠ η 1 ) , y 1 = 1 ( η 0 = η 1 ) . (81) |\operatorname{Im}z|\le\frac{y_1}{m_0}=\frac{1}{4m_0}\log\left|\frac{\eta_0+\eta_1}{\eta_0-\eta_1}\right|\quad(\eta_0\ne\eta_1),\qquad y_1=1\quad(\eta_0=\eta_1). \tag*{(81)} ∣ Im z ∣ ≤ m 0 y 1 = 4 m 0 1 log η 0 − η 1 η 0 + η 1 ( η 0 = η 1 ) , y 1 = 1 ( η 0 = η 1 ) . (81)
This completes the proof for m 0 < 1 m_0<1 m 0 < 1 .
The endpoint m 0 = 1 m_0=1 m 0 = 1 . For κ ∈ { η 1 / η 0 , η 0 / η 1 } \kappa\in\{\eta_1/\eta_0,\eta_0/\eta_1\} κ ∈ { η 1 / η 0 , η 0 / η 1 } put
d ℓ , κ ( z ) = ξ ℓ ( z ) ψ ℓ ′ ( z ) − κ ψ ℓ ( z ) ξ ℓ ′ ( z ) , d_{\ell,\kappa}(z)=\xi_\ell(z)\psi_\ell'(z)-\kappa\psi_\ell(z)\xi_\ell'(z), d ℓ , κ ( z ) = ξ ℓ ( z ) ψ ℓ ′ ( z ) − κ ψ ℓ ( z ) ξ ℓ ′ ( z ) ,
D ℓ 1 = η 0 d ℓ , η 1 / η 0 , D ℓ 2 = − η 1 d ℓ , η 0 / η 1 . D_\ell^1=\eta_0d_{\ell,\eta_1/\eta_0},\qquad D_\ell^2=-\eta_1d_{\ell,\eta_0/\eta_1}. D ℓ 1 = η 0 d ℓ , η 1 / η 0 , D ℓ 2 = − η 1 d ℓ , η 0 / η 1 .
If η 0 = η 1 \eta_0=\eta_1 η 0 = η 1 , the Wronskian gives d ℓ , 1 = − i d_{\ell,1}=-\mathrm{i} d ℓ , 1 = − i , so there are no zeros. Otherwise set
ρ = ∣ η 1 − η 0 ∣ η 1 + η 0 ∈ ( 0 , 1 ) , 0 < c < 1 4 log 1 ρ , \rho=\frac{|\eta_1-\eta_0|}{\eta_1+\eta_0}\in(0,1),\qquad0<c<\frac{1}{4}\log\frac{1}{\rho}, ρ = η 1 + η 0 ∣ η 1 − η 0 ∣ ∈ ( 0 , 1 ) , 0 < c < 4 1 log ρ 1 ,
and work in − c ≤ Im z ≤ 0 -c\le\operatorname{Im}z\le0 − c ≤ Im z ≤ 0 , Re z > 0 \operatorname{Re}z>0 Re z > 0 . Fix M M M large and then ν 0 \nu_0 ν 0 large; write e ν = M − 3 / 2 + M − 1 / 2 ν − 2 / 3 e_\nu=M^{-3/2}+M^{-1/2}\nu^{-2/3} e ν = M − 3/2 + M − 1/2 ν − 2/3 .
In Region 1 the subturning estimates give nonzero ψ ℓ , ξ ℓ \psi_\ell,\xi_\ell ψ ℓ , ξ ℓ and
d ℓ , κ ψ ℓ ξ ℓ = ( 1 + κ ) ν 2 − z 2 z ( 1 + O κ ( e ν ) ) ≠ 0. \frac{d_{\ell,\kappa}}{\psi_\ell\xi_\ell}=(1+\kappa)\frac{\sqrt{\nu^2-z^2}}{z}(1+O_\kappa(e_\nu))\ne0. ψ ℓ ξ ℓ d ℓ , κ = ( 1 + κ ) z ν 2 − z 2 ( 1 + O κ ( e ν )) = 0.
Here the 1 / ( 2 z ) 1/(2z) 1/ ( 2 z ) terms are absorbed using ∣ ν 2 − z 2 ∣ 1 / 2 ≳ M 1 / 2 ν 2 / 3 |\nu^2-z^2|^{1/2}\gtrsim M^{1/2}\nu^{2/3} ∣ ν 2 − z 2 ∣ 1/2 ≳ M 1/2 ν 2/3 . In Region 2 put s = − 2 1 / 3 ( z − ν ) ν − 1 / 3 s=-2^{1/3}(z-\nu)\nu^{-1/3} s = − 2 1/3 ( z − ν ) ν − 1/3 , A = Ai ( s ) A=\operatorname{Ai}(s) A = Ai ( s ) and B = Bi ( s ) B=\operatorname{Bi}(s) B = Bi ( s ) . The uniform Airy expansions in the appendix, with H ν ( 1 ) = J ν + i Y ν H_\nu^{(1)}=J_\nu+\mathrm{i}Y_\nu H ν ( 1 ) = J ν + i Y ν , yield
d ℓ , κ = π [ ( κ − 1 ) A A ′ + i ( B A ′ − κ A B ′ ) ] + O M , κ ( ν − 2 / 3 ) . d_{\ell,\kappa}=\pi\left[(\kappa-1)AA'+\mathrm{i}(BA'-\kappa AB')\right]+O_{M,\kappa}(\nu^{-2/3}). d ℓ , κ = π [ ( κ − 1 ) A A ′ + i ( B A ′ − κ A B ′ ) ] + O M , κ ( ν − 2/3 ) .
For real s s s , a zero of the bracket would force A A ′ = 0 AA'=0 A A ′ = 0 . The Wronskian A B ′ − A ′ B = 1 / π AB'-A'B=1/\pi A B ′ − A ′ B = 1/ π , (9.2.7) in [27 ] , makes its imaginary part − 1 / π -1/\pi − 1/ π when A = 0 A=0 A = 0 and − κ / π -\kappa/\pi − κ / π when A ′ = 0 A'=0 A ′ = 0 . Thus the bracket is bounded away from zero on the compact real turning interval. Since ∣ Im s ∣ ≤ 2 1 / 3 c ν − 1 / 3 |\operatorname{Im}s| \le2^{1/3}c\nu^{-1/3} ∣ Im s ∣ ≤ 2 1/3 c ν − 1/3 , continuity and the remainder bound exclude zeros throughout this region for large ν \nu ν .
In Region 3 use L ± , R L_{\pm},R L ± , R from (69) and put γ = 1 − ν 2 / z 2 \gamma=\sqrt{1-\nu^2/z^2} γ = 1 − ν 2 / z 2 . The Debye estimates give L ± = ± i γ + O ( ∣ γ ∣ e ν ) L_{\pm}=\pm i\gamma+O(|\gamma|e_{\nu}) L ± = ± iγ + O ( ∣ γ ∣ e ν ) . Thus L − − κ L + = − i ( 1 + κ ) γ ( 1 + O κ ( e ν ) ) ≠ 0 L_{-}-\kappa L_{+}=-i(1+\kappa)\gamma(1+O_{\kappa}(e_{\nu}))\ne0 L − − κ L + = − i ( 1 + κ ) γ ( 1 + O κ ( e ν )) = 0 for the chosen M , ν 0 M,\nu_{0} M , ν 0 , so the following division is valid. A zero of d ℓ , κ d_{\ell,\kappa} d ℓ , κ would therefore imply
∣ R ( z ) ∣ = ∣ ( κ − 1 ) L + ( z ) L − ( z ) − κ L + ( z ) ∣ ≤ ρ + O κ ( e ν ) . |R(z)|=\left|\frac{(\kappa-1)L_{+}(z)}{L_{-}(z)-\kappa L_{+}(z)}\right|\le\rho+O_{\kappa}(e_{\nu}). ∣ R ( z ) ∣ = L − ( z ) − κ L + ( z ) ( κ − 1 ) L + ( z ) ≤ ρ + O κ ( e ν ) .
But ∣ R ( x ) ∣ = 1 |R(x)|=1 ∣ R ( x ) ∣ = 1 for real x x x and ∣ ∂ y log ∣ R ( x + i y ) ∣ ∣ ≤ 3 |\partial_y\log|R(x+iy)||\le3 ∣ ∂ y log ∣ R ( x + i y ) ∣∣ ≤ 3 by the same estimates. Hence ∣ R ( z ) ∣ ≥ e − 3 c > ρ |R(z)|\ge e^{-3c}>\rho ∣ R ( z ) ∣ ≥ e − 3 c > ρ , a contradiction after increasing M , ν 0 M,\nu_{0} M , ν 0 . Region 4 applies unchanged at m 0 = 1 m_{0}=1 m 0 = 1 for this choice of c c c and large K K K . Taking C 1 C_{1} C 1 large enough to place all ν < ν 0 \nu<\nu_{0} ν < ν 0 in Region 4 completes the proof, with C 2 = c / ( r 0 ϵ 0 μ 0 ) C_{2}=c/(r_{0}\sqrt{\epsilon_{0}\mu_{0}}) C 2 = c / ( r 0 ϵ 0 μ 0 ) .
Generalized eigenfunctions and spectral resolution
Throughout this section Ω \Omega Ω may be any bounded Lipschitz open set, with coefficients (1.1). We use the two families from Lemma 2.3; spherical harmonics label the incident fields at infinity and do not impose a spherical interface.
Definition 4.1. We set a j , ℓ , m = ⟨ Y , Ψ j , ℓ , m ⟩ a_{j,\ell,m}=\langle Y,\Psi_{j,\ell,m}\rangle a j , ℓ , m = ⟨ Y , Ψ j , ℓ , m ⟩
E λ ( Y ) = ∑ ℓ , m ( − i ) ℓ + 1 a 1 , ℓ , m E λ , 1 ( Y 1 , ℓ , m ) + a 2 , ℓ , m ( − i ) ℓ E λ , 2 ( Y 2 , ℓ , m ) (82) E_{\lambda}(Y)=\sum_{\ell,m}(-i)^{\ell+1}a_{1,\ell,m}E_{\lambda,1}(Y_{1,\ell,m})+a_{2,\ell,m}(-i)^{\ell}E_{\lambda,2}(Y_{2,\ell,m}) \tag*{(82)} E λ ( Y ) = ℓ , m ∑ ( − i ) ℓ + 1 a 1 , ℓ , m E λ , 1 ( Y 1 , ℓ , m ) + a 2 , ℓ , m ( − i ) ℓ E λ , 2 ( Y 2 , ℓ , m ) (82)
This is the electric field generalized eigenfunction. For Y ∈ L t 2 ( S 2 ) Y\in L_{t}^{2}(\mathbb{S}^{2}) Y ∈ L t 2 ( S 2 ) the sum converges in H l o c ( curl ) H_{\mathrm{loc}}(\operatorname{curl}) H loc ( curl ) , by the local free-wave estimates, (35) , and the resolvent bounds in Lemma 2.3.
For smooth tangential Y Y Y , the free-wave asymptotics and the outgoing correction give
E λ ( Y ) = η 0 2 π ( e − i k 0 r r Y + e i k 0 r r Y ~ ) + O ( r − 2 ) . (83) E_{\lambda}(Y)=\sqrt{\frac{\eta_{0}}{2\pi}}\left(\frac{e^{-ik_{0}r}}{r}Y+\frac{e^{ik_{0}r}}{r}\widetilde{Y}\right)+O(r^{-2}). \tag*{(83)} E λ ( Y ) = 2 π η 0 ( r e − i k 0 r Y + r e i k 0 r Y ) + O ( r − 2 ) . (83)
Here Y ~ \widetilde{Y} Y is smooth and tangential. The generalized eigenfunctions E λ ( Y ) E_{\lambda}(Y) E λ ( Y ) can be viewed as tempered distributions in S ′ ( R 3 ; C 3 ) \mathcal{S}'(\mathbb{R}^{3};\mathbb{C}^{3}) S ′ ( R 3 ; C 3 ) . In particular g ∈ C 0 ∞ ( R + ) g\in C_{0}^{\infty}(\mathbb{R}_{+}) g ∈ C 0 ∞ ( R + ) then ∫ R g ( λ ) E λ ( Y ) d λ \int_{\mathbb{R}}g(\lambda)E_{\lambda}(Y)\,d\lambda ∫ R g ( λ ) E λ ( Y ) d λ is square integrable. Therefore the inner product
⟨ E λ ( Y ) , E λ ′ ( Y ′ ) ⟩ L 2 ( ϵ d x ) 3 (84) \langle E_{\lambda}(Y),E_{\lambda'}(Y')\rangle_{L^{2}(\epsilon\,dx)^{3}} \tag*{(84)} ⟨ E λ ( Y ) , E λ ′ ( Y ′ ) ⟩ L 2 ( ϵ d x ) 3 (84)
can be viewed as a bidistribution in D ′ ( R + × R + ) \mathcal{D}'(\mathbb{R}_{+}\times\mathbb{R}_{+}) D ′ ( R + × R + ) . The last inner product can be computed by taking the limit
lim R → ∞ 1 λ 2 − λ ′ 2 ( ⟨ P E E λ ( Y ) , χ R E λ ′ ( Y ′ ) ⟩ L 2 ( ϵ d x ) 3 − ⟨ χ R E λ ( Y ) , P E E λ ′ ( Y ′ ) ⟩ L 2 ( ϵ d x ) 3 ) \lim_{R\to\infty}\frac{1}{\lambda^{2}-\lambda'^{2}}\left(\langle P_{E}E_{\lambda}(Y),\chi_{R}E_{\lambda'}(Y')\rangle_{L^{2}(\epsilon\,dx)^{3}}-\langle\chi_{R}E_{\lambda}(Y),P_{E}E_{\lambda'}(Y')\rangle_{L^{2}(\epsilon\,dx)^{3}}\right) R → ∞ lim λ 2 − λ ′ 2 1 ( ⟨ P E E λ ( Y ) , χ R E λ ′ ( Y ′ ) ⟩ L 2 ( ϵ d x ) 3 − ⟨ χ R E λ ( Y ) , P E E λ ′ ( Y ′ ) ⟩ L 2 ( ϵ d x ) 3 )
and then using Green’s identity. Here χ R \chi_{R} χ R denotes the indicator function of a large ball B R B_{R} B R where Ω ‾ ⊂ B R \overline{\Omega}\subset B_{R} Ω ⊂ B R . Write a 2 = η 0 / ( 2 π ) a^{2}=\eta_{0}/(2\pi) a 2 = η 0 / ( 2 π ) , k = k 0 ( λ ) k=k_{0}(\lambda) k = k 0 ( λ ) , k ′ = k 0 ( λ ′ ) k'=k_{0}(\lambda') k ′ = k 0 ( λ ′ ) and C = ⟨ Y , Y ′ ⟩ C=\langle Y,Y'\rangle C = ⟨ Y , Y ′ ⟩ , D ( λ , λ ′ ) = ⟨ Y ~ λ , Y ~ λ ′ ′ ⟩ D(\lambda,\lambda')=\langle\widetilde{Y}_{\lambda},\widetilde{Y}'_{\lambda'}\rangle D ( λ , λ ′ ) = ⟨ Y λ , Y λ ′ ′ ⟩ , with angular inner products on S 2 \mathbb{S}^{2} S 2 . Green’s identity at equal frequencies gives, D ( λ , λ ) = C D(\lambda,\lambda)=C D ( λ , λ ) = C . Inserting (83) into the boundary term gives
⟨ E λ ( Y ) , χ R E λ ′ ( Y ′ ) ⟩ = ϵ 0 a 2 sin ( ( k − k ′ ) R ) k − k ′ ( C + D ( λ , λ ′ ) ) + o D ′ ( 1 ) . \langle E_{\lambda}(Y),\chi_{R}E_{\lambda'}(Y')\rangle=\epsilon_{0}a^{2}\frac{\sin((k-k')R)}{k-k'}(C+D(\lambda,\lambda'))+o_{\mathcal{D}'}(1). ⟨ E λ ( Y ) , χ R E λ ′ ( Y ′ )⟩ = ϵ 0 a 2 k − k ′ sin (( k − k ′ ) R ) ( C + D ( λ , λ ′ )) + o D ′ ( 1 ) .
Indeed, the mixed incoming–outgoing terms have phases ± ( k + k ′ ) R \pm(k+k')R ± ( k + k ′ ) R and vanish distributionally on R + 2 \mathbb{R}^{2}_{+} R + 2 . The remaining cosine term has coefficient ( C − D ) / ( k − k ′ ) (C-D)/(k-k') ( C − D ) / ( k − k ′ ) tends to zero distributionally. Since sin ( R x ) / x → π δ ( x ) \sin(Rx)/x \to\pi\delta(x) sin ( R x ) / x → π δ ( x ) and
2 π ϵ 0 a 2 δ ( k − k ′ ) = ϵ 0 η 0 ϵ 0 μ 0 δ ( λ − λ ′ ) = δ ( λ − λ ′ ) , 2\pi\epsilon_{0}a^{2}\delta(k-k')=\frac{\epsilon_{0}\eta_{0}}{\sqrt{\epsilon_{0}\mu_{0}}}\delta(\lambda-\lambda')=\delta(\lambda-\lambda'), 2 π ϵ 0 a 2 δ ( k − k ′ ) = ϵ 0 μ 0 ϵ 0 η 0 δ ( λ − λ ′ ) = δ ( λ − λ ′ ) ,
we obtain the distributional identity:
⟨ E λ ( Y ) , E λ ′ ( Y ′ ) ⟩ L 2 ( ϵ d x ) 3 = 2 λ δ ( λ 2 − λ ′ 2 ) ⟨ Y , Y ′ ⟩ L 2 ( S 2 ) = δ ( λ − λ ′ ) ⟨ Y , Y ′ ⟩ L 2 ( S 2 ) , λ , λ ′ > 0 \langle E_{\lambda}(Y),E_{\lambda'}(Y')\rangle_{L^{2}(\epsilon dx)^{3}}
=2\lambda\delta(\lambda^{2}-\lambda'^{2})\langle Y,Y'\rangle_{L^{2}(\mathbb{S}^{2})}
=\delta(\lambda-\lambda')\langle Y,Y'\rangle_{L^{2}(\mathbb{S}^{2})},\quad\lambda,\lambda'>0 ⟨ E λ ( Y ) , E λ ′ ( Y ′ ) ⟩ L 2 ( ϵ d x ) 3 = 2 λ δ ( λ 2 − λ ′ 2 ) ⟨ Y , Y ′ ⟩ L 2 ( S 2 ) = δ ( λ − λ ′ ) ⟨ Y , Y ′ ⟩ L 2 ( S 2 ) , λ , λ ′ > 0
The generalized eigenfunctions can be used to describe evolution operators using the Spectral theorem.
Proposition 4.2. If f ∈ L c o m p 2 ( R 3 ; C 3 ) ∩ L ϵ ( div 0 ) f\in L^{2}_{\mathrm{comp}}(\mathbb{R}^{3};\mathbb{C}^{3})\cap L_{\epsilon}(\operatorname{div}0) f ∈ L comp 2 ( R 3 ; C 3 ) ∩ L ϵ ( div 0 ) , then for real λ > 0 \lambda>0 λ > 0
( R λ , E − R − λ , E ) f = i π λ ∑ n E λ ( Y n ) ⟨ f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 (85) (R_{\lambda,E}-R_{-\lambda,E})f=\frac{\mathrm{i}\pi}{\lambda}\sum_{n}E_{\lambda}(Y_{n})\langle f,E_{\lambda}(Y_{n})\rangle_{(L^{2}(\epsilon dx))^{3}} \tag*{(85)} ( R λ , E − R − λ , E ) f = λ i π n ∑ E λ ( Y n ) ⟨ f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 (85) in L l o c 2 ( R 3 ) L^{2}_{\mathrm{loc}}(\mathbb{R}^{3}) L loc 2 ( R 3 ) . For the spectral measure d B λ dB_{\lambda} d B λ of P E \sqrt{P_{E}} P E , and compactly supported f , g ∈ L ϵ ( div 0 ) f,g\in L_{\epsilon}(\operatorname{div}0) f , g ∈ L ϵ ( div 0 ) , we have
⟨ d B λ f , g ⟩ ( L 2 ( ϵ d x ) ) 3 = χ [ 0 , ∞ ) ( λ ) ∑ n ⟨ f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 ⟨ E λ ( Y n ) , g ⟩ ( L 2 ( ϵ d x ) ) 3 d λ (86) \langle dB_{\lambda}f,g\rangle_{(L^{2}(\epsilon dx))^{3}}
=\chi_{[0,\infty)}(\lambda)\sum_{n}\langle f,E_{\lambda}(Y_{n})\rangle_{(L^{2}(\epsilon dx))^{3}}\langle E_{\lambda}(Y_{n}),g\rangle_{(L^{2}(\epsilon dx))^{3}}\,d\lambda \tag*{(86)} ⟨ d B λ f , g ⟩ ( L 2 ( ϵ d x ) ) 3 = χ [ 0 , ∞ ) ( λ ) n ∑ ⟨ f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 ⟨ E λ ( Y n ) , g ⟩ ( L 2 ( ϵ d x ) ) 3 d λ (86) For any bounded Borel function we then have that
⟨ h ( P E ) f , g ⟩ ( L 2 ( ϵ d x ) ) 3 = ∑ n ∫ 0 ∞ h ( λ 2 ) ⟨ f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 ⟨ E λ ( Y n ) , g ⟩ ( L 2 ( ϵ d x ) ) 3 d λ . (87) \langle h(P_{E})f,g\rangle_{(L^{2}(\epsilon dx))^{3}}
=\sum_{n}\int_{0}^{\infty}h(\lambda^{2})\langle f,E_{\lambda}(Y_{n})\rangle_{(L^{2}(\epsilon dx))^{3}}\langle E_{\lambda}(Y_{n}),g\rangle_{(L^{2}(\epsilon dx))^{3}}\,d\lambda. \tag*{(87)} ⟨ h ( P E ) f , g ⟩ ( L 2 ( ϵ d x ) ) 3 = n ∑ ∫ 0 ∞ h ( λ 2 ) ⟨ f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 ⟨ E λ ( Y n ) , g ⟩ ( L 2 ( ϵ d x ) ) 3 d λ . (87) The first identity (the resolvent jump) holds with R ± λ , E R_{\pm\lambda,E} R ± λ , E replaced by R ~ ± λ , E \widetilde{R}_{\pm\lambda,E} R ± λ , E for every f ∈ L c o m p 2 ( R 3 ; C 3 ) f\in L^{2}_{\mathrm{comp}}(\mathbb{R}^{3};\mathbb{C}^{3}) f ∈ L comp 2 ( R 3 ; C 3 ) , without the divergence condition.
Proof. The incoming counterpart of (A.1) gives
R − λ , E f = 2 η 0 π e − i k 0 r r ( Ψ + O ( r − 1 ) ) (88) R_{-\lambda,E}f=\sqrt{\frac{2\eta_{0}}{\pi}}\frac{e^{-\mathrm{i}k_{0}r}}{r}\left(\Psi+O(r^{-1})\right) \tag*{(88)} R − λ , E f = π 2 η 0 r e − i k 0 r ( Ψ + O ( r − 1 ) ) (88)
for some Ψ ∈ C ∞ ( S 2 ; C 3 ) \Psi\in C^{\infty}(\mathbb{S}^{2};\mathbb{C}^{3}) Ψ ∈ C ∞ ( S 2 ; C 3 ) . We then have that
( R λ , E − R − λ , E ) f = 2 η 0 π ( − e − i k 0 r r Ψ + e i k 0 r r Ψ ~ ) + O ( r − 2 ) (89) (R_{\lambda,E}-R_{-\lambda,E})f=\sqrt{\frac{2\eta_{0}}{\pi}}\left(-\frac{e^{-\mathrm{i}k_{0}r}}{r}\Psi+\frac{e^{\mathrm{i}k_{0}r}}{r}\widetilde{\Psi}\right)+O(r^{-2}) \tag*{(89)} ( R λ , E − R − λ , E ) f = π 2 η 0 ( − r e − i k 0 r Ψ + r e i k 0 r Ψ ) + O ( r − 2 ) (89)
as r → ∞ r\to\infty r → ∞ for some Ψ ~ ∈ C ∞ ( S 2 ; C 3 ) \widetilde{\Psi}\in C^{\infty}(\mathbb{S}^{2};\mathbb{C}^{3}) Ψ ∈ C ∞ ( S 2 ; C 3 ) and
( P E − λ 2 ) ( R λ , E − R − λ , E ) f = 0. (90) (P_{E}-\lambda^{2})(R_{\lambda,E}-R_{-\lambda,E})f=0. \tag*{(90)} ( P E − λ 2 ) ( R λ , E − R − λ , E ) f = 0. (90)
By (83) , E λ ( − 2 Ψ ) E_{\lambda}(-2\Psi) E λ ( − 2Ψ ) and ( R λ , E − R − λ , E ) f (R_{\lambda,E}-R_{-\lambda,E})f ( R λ , E − R − λ , E ) f have the same incoming leading asymptotic, so their difference w w w is purely outgoing. Since ( P E − λ 2 ) w = 0 (P_{E}-\lambda^{2})w=0 ( P E − λ 2 ) w = 0 on all of R 3 \mathbb{R}^{3} R 3 with the transmission conditions, Theorem A.1 gives w = 0 w=0 w = 0 . Integration by parts then gives that
⟨ f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 = ⟨ ( P E − λ 2 ) R − λ , E f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 = 2 i λ π ⟨ Ψ , Y n ⟩ L 2 ( S 2 ; C 3 ) (91) \langle f,E_{\lambda}(Y_{n})\rangle_{(L^{2}(\epsilon dx))^{3}}=\langle(P_{E}-\lambda^{2})R_{-\lambda,E}f,E_{\lambda}(Y_{n})\rangle_{(L^{2}(\epsilon dx))^{3}}=\frac{2\mathrm{i}\lambda}{\pi}\langle\Psi,Y_{n}\rangle_{L^{2}(\mathbb{S}^{2};\mathbb{C}^{3})} \tag*{(91)} ⟨ f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 = ⟨( P E − λ 2 ) R − λ , E f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 = π 2 i λ ⟨ Ψ , Y n ⟩ L 2 ( S 2 ; C 3 ) (91)
The boundary term at ∣ x ∣ = R |x|=R ∣ x ∣ = R is μ 0 − 1 ∫ ∂ B R [ ( r ^ × curl u ) ⋅ E ‾ − ( r ^ × curl E ‾ ) ⋅ u ] d S \mu_0^{-1}\int_{\partial B_R}\left[(\hat{r}\times\operatorname{curl}u)\cdot\overline{E}-(\hat{r}\times\operatorname{curl}\overline{E})\cdot u\right]\,dS μ 0 − 1 ∫ ∂ B R [ ( r ^ × curl u ) ⋅ E − ( r ^ × curl E ) ⋅ u ] d S with u = R − λ , E f u=R_{-\lambda,E}f u = R − λ , E f ; only the incoming part of E E E contributes in the limit, giving the result. Because the Y n Y_n Y n are a basis for L t 2 ( S 2 ; C 3 ) L^2_t(\mathbb{S}^2;\mathbb{C}^3) L t 2 ( S 2 ; C 3 ) , and Ψ \Psi Ψ is tangential, we can expand Ψ \Psi Ψ as
Ψ = ∑ n ⟨ Ψ , Y n ⟩ L 2 ( S 2 ; C 3 ) Y n (92) \Psi=\sum_n\langle\Psi,Y_n\rangle_{L^2(\mathbb{S}^2;\mathbb{C}^3)}Y_n \tag*{(92)} Ψ = n ∑ ⟨ Ψ , Y n ⟩ L 2 ( S 2 ; C 3 ) Y n (92)
and the sum converges in C ∞ ( S 2 ; C 3 ) C^\infty(\mathbb{S}^2;\mathbb{C}^3) C ∞ ( S 2 ; C 3 ) . Thus
Ψ = − i π 2 λ ∑ n ⟨ f , E − λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 Y n . (93) \Psi=-\frac{i\pi}{2\lambda}\sum_n\langle f,E_{-\lambda}(Y_n)\rangle_{(L^2(\epsilon\,dx))^3}Y_n. \tag*{(93)} Ψ = − 2 λ iπ n ∑ ⟨ f , E − λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 Y n . (93)
Consequently,
( R λ , E − R − λ , E ) f = E λ ( − 2 Ψ ) = i π λ ∑ n E λ ( Y n ) ⟨ f , E − λ ( Y n ) ⟩ . (94) (R_{\lambda,E}-R_{-\lambda,E})f=E_\lambda(-2\Psi)=\frac{i\pi}{\lambda}\sum_n E_\lambda(Y_n)\langle f,E_{-\lambda}(Y_n)\rangle. \tag*{(94)} ( R λ , E − R − λ , E ) f = E λ ( − 2Ψ ) = λ iπ n ∑ E λ ( Y n ) ⟨ f , E − λ ( Y n )⟩ . (94)
This agrees with Stone’s formula R λ , E − R − λ , E = 2 π i δ ( P E − λ 2 ) R_{\lambda,E}-R_{-\lambda,E}=2\pi i\delta(P_E-\lambda^2) R λ , E − R − λ , E = 2 π i δ ( P E − λ 2 ) for the δ ( λ − λ ′ ) \delta(\lambda-\lambda') δ ( λ − λ ′ ) -normalized family.
The jump formula also holds for R ~ λ , E \widetilde{R}_{\lambda,E} R λ , E ; its gradient term has no jump. The limiting absorption argument in Lemma 2.3 gives continuous, locally bounded matrix elements of R λ , E Π E R_{\lambda,E}\Pi_E R λ , E Π E for compactly supported L 2 L^2 L 2 data on compact subintervals of ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . The gradient term in R ~ λ , E \widetilde{R}_{\lambda,E} R λ , E is analytic there. Stone’s formula therefore gives absolutely continuous spectral measures for this dense class and hence for all data in L ϵ ( div 0 ) L_\epsilon(\operatorname{div}0) L ϵ ( div 0 ) . On L ϵ ( div 0 ) L_\epsilon(\operatorname{div}0) L ϵ ( div 0 ) , P E u = 0 P_Eu=0 P E u = 0 implies curl u = 0 \operatorname{curl}u=0 curl u = 0 , and (1.7) then gives u = 0 u=0 u = 0 . Thus
d B λ = λ π i ( R λ , E − R − λ , E ) d λ , λ > 0 , dB_\lambda=\frac{\lambda}{\pi i}(R_{\lambda,E}-R_{-\lambda,E})\,d\lambda,\qquad\lambda>0, d B λ = π i λ ( R λ , E − R − λ , E ) d λ , λ > 0 ,
accounts for the entire spectral measure of P E \sqrt{P_E} P E . Substitution of the jump formula proves the assertion, with extension to L ϵ ( div 0 ) L_\epsilon(\operatorname{div}0) L ϵ ( div 0 ) by the spectral theorem. For f ∈ L c o m p 2 f\in L^2_{\mathrm{comp}} f ∈ L comp 2 without the divergence condition, u = R ~ − λ , E f u=\widetilde{R}_{-\lambda,E}f u = R − λ , E f solves ( ϵ − 1 curl μ − 1 curl − λ 2 ) u = f (\epsilon^{-1}\operatorname{curl}\mu^{-1}\operatorname{curl}-\lambda^2)u=f ( ϵ − 1 curl μ − 1 curl − λ 2 ) u = f , so div ( ϵ u ) = − λ − 2 div ( ϵ f ) \operatorname{div}(\epsilon u)=-\lambda^{-2}\operatorname{div}(\epsilon f) div ( ϵ u ) = − λ − 2 div ( ϵ f ) is compactly supported and u u u is an incoming solution outside supp f \operatorname{supp}f supp f ; the argument above applies verbatim. Moreover ⟨ f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 = ⟨ Π E f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 \langle f,E_\lambda(Y_n)\rangle_{(L^2(\epsilon\,dx))^3}=\langle\Pi_Ef,E_\lambda(Y_n)\rangle_{(L^2(\epsilon\,dx))^3} ⟨ f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 = ⟨ Π E f , E λ ( Y n ) ⟩ ( L 2 ( ϵ d x ) ) 3 , since div ( ϵ E λ ( Y n ) ) = 0 \operatorname{div}(\epsilon E_\lambda(Y_n))=0 div ( ϵ E λ ( Y n )) = 0 and f f f has compact support, and the gradient term of R ~ ± λ , E \widetilde{R}_{\pm\lambda,E} R ± λ , E is even in λ \lambda λ .
Proof of Theorem 1.5
We assume that m 0 > 1 m_0>1 m 0 > 1 and use the notation of (2.6). We define
ϑ ℓ ( w ) = w y ℓ ( w ) ξ ℓ = ψ ℓ + i ϑ ℓ (95) \vartheta_\ell(w)=wy_\ell(w)\qquad\xi_\ell=\psi_\ell+i\vartheta_\ell \tag*{(95)} ϑ ℓ ( w ) = w y ℓ ( w ) ξ ℓ = ψ ℓ + i ϑ ℓ (95)
It follows that ψ ℓ \psi_\ell ψ ℓ is entire and ϑ ℓ \vartheta_\ell ϑ ℓ is holomorphic on C ∖ { 0 } \mathbb{C}\setminus\{0\} C ∖ { 0 } , these are both real on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) and we have that
v ′ ′ ( w ) + ( 1 − ℓ ( ℓ + 1 ) w 2 ) v ( w ) = 0 v = ψ ℓ , ϑ ℓ , ψ ℓ ϑ ℓ ′ − ψ ℓ ′ ϑ ℓ = 1. (96) v''(w)+\left(1-\frac{\ell(\ell+1)}{w^2}\right)v(w)=0\qquad v=\psi_\ell,\vartheta_\ell,\qquad\psi_\ell\vartheta_\ell'-\psi_\ell'\vartheta_\ell=1. \tag*{(96)} v ′′ ( w ) + ( 1 − w 2 ℓ ( ℓ + 1 ) ) v ( w ) = 0 v = ψ ℓ , ϑ ℓ , ψ ℓ ϑ ℓ ′ − ψ ℓ ′ ϑ ℓ = 1. (96)
The ψ ℓ , ψ ℓ ′ \psi_\ell,\psi_\ell' ψ ℓ , ψ ℓ ′ have no common zero and neither do ξ ℓ , ξ ℓ ′ \xi_\ell,\xi_\ell' ξ ℓ , ξ ℓ ′ because ψ ℓ ξ ℓ ′ − ψ ℓ ′ ξ ℓ = i \psi_\ell\xi_\ell'-\psi_\ell'\xi_\ell=i ψ ℓ ξ ℓ ′ − ψ ℓ ′ ξ ℓ = i . For κ > 0 \kappa>0 κ > 0 we put
D κ , ℓ ( z ) = ξ ℓ ( z ) ψ ℓ ′ ( m 0 z ) − κ ψ ℓ ( m 0 z ) ξ ℓ ′ ( z ) = P κ , ℓ ( z ) + i Q κ , ℓ ( z ) , (97) D_{\kappa,\ell}(z)=\xi_\ell(z)\psi_\ell'(m_0z)-\kappa\psi_\ell(m_0z)\xi_\ell'(z)=P_{\kappa,\ell}(z)+iQ_{\kappa,\ell}(z), \tag*{(97)} D κ , ℓ ( z ) = ξ ℓ ( z ) ψ ℓ ′ ( m 0 z ) − κ ψ ℓ ( m 0 z ) ξ ℓ ′ ( z ) = P κ , ℓ ( z ) + i Q κ , ℓ ( z ) , (97)
P κ , ℓ ( z ) = ψ ℓ ( z ) ψ ℓ ′ ( m 0 z ) − κ ψ ℓ ( m 0 z ) ψ ℓ ′ ( z ) , (98) P_{\kappa,\ell}(z)=\psi_\ell(z)\psi_\ell'(m_0z)-\kappa\psi_\ell(m_0z)\psi_\ell'(z), \tag*{(98)} P κ , ℓ ( z ) = ψ ℓ ( z ) ψ ℓ ′ ( m 0 z ) − κ ψ ℓ ( m 0 z ) ψ ℓ ′ ( z ) , (98)
Q κ , ℓ ( z ) = ϑ ℓ ( z ) ψ ℓ ′ ( m 0 z ) − κ ψ ℓ ( m 0 z ) ϑ ℓ ′ ( z ) (99) Q_{\kappa,\ell}(z)=\vartheta_\ell(z)\psi_\ell'(m_0z)-\kappa\psi_\ell(m_0z)\vartheta_\ell'(z) \tag*{(99)} Q κ , ℓ ( z ) = ϑ ℓ ( z ) ψ ℓ ′ ( m 0 z ) − κ ψ ℓ ( m 0 z ) ϑ ℓ ′ ( z ) (99)
The denominators of § 3 are
D ℓ 1 = η 0 D η 1 / η 0 , ℓ , D ℓ 2 = − η 1 D η 0 / η 1 , ℓ , (100) \mathcal{D}^{1}_{\ell}=\eta_{0}D_{\eta_{1}/\eta_{0},\ell},\qquad\mathcal{D}^{2}_{\ell}=-\eta_{1}D_{\eta_{0}/\eta_{1},\ell}, \tag*{(100)} D ℓ 1 = η 0 D η 1 / η 0 , ℓ , D ℓ 2 = − η 1 D η 0 / η 1 , ℓ , (100)
and P κ , ℓ , Q κ , ℓ \mathcal{P}_{\kappa,\ell},\mathcal{Q}_{\kappa,\ell} P κ , ℓ , Q κ , ℓ are real on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . Fix 1 m 0 < t 1 < t 2 < 1 \frac{1}{m_{0}}<t_{1}<t_{2}<1 m 0 1 < t 1 < t 2 < 1 and set
Λ ( t ) = arcsech t − 1 − t 2 0 < t < 1. (101) \Lambda(t)=\operatorname{arcsech}t-\sqrt{1-t^{2}}\qquad0<t<1. \tag*{(101)} Λ ( t ) = arcsech t − 1 − t 2 0 < t < 1. (101)
Then it follows that Λ > 0 \Lambda>0 Λ > 0 and also Λ ′ ( t ) = − 1 − t 2 t < 0 \Lambda'(t)=-\frac{\sqrt{1-t^{2}}}{t}<0 Λ ′ ( t ) = − t 1 − t 2 < 0 on ( 0 , 1 ) (0,1) ( 0 , 1 ) . The window ( 1 m 0 , 1 ) (\frac{1}{m_{0}},1) ( m 0 1 , 1 ) is non-empty exactly when m 0 > 1 m_{0}>1 m 0 > 1 ; for z = t ν z=t\nu z = t ν with t t t in this window the exterior argument z z z is below its turning point (tunnelling region) while the interior argument m 0 z m_{0}z m 0 z is past its turning point.
Lemma 5.1. Let 0 < τ 1 < τ 2 < 1 < s 1 < s 2 0<\tau_{1}<\tau_{2}<1<s_{1}<s_{2} 0 < τ 1 < τ 2 < 1 < s 1 < s 2 . We write t = sech α t=\operatorname{sech}\alpha t = sech α and s = sec β s=\sec\beta s = sec β and set Ω ν ( s ) = ν ( tan β − β ) − π 4 \Omega_{\nu}(s)=\nu(\tan\beta-\beta)-\frac{\pi}{4} Ω ν ( s ) = ν ( tan β − β ) − 4 π , so that sinh α = 1 − t 2 t \sinh\alpha=\frac{\sqrt{1-t^{2}}}{t} sinh α = t 1 − t 2 , sin β = s 2 − 1 s \sin\beta=\frac{\sqrt{s^{2}-1}}{s} sin β = s s 2 − 1 and α − tanh α = Λ ( t ) \alpha-\tanh\alpha=\Lambda(t) α − tanh α = Λ ( t ) . As ν → ∞ \nu\to\infty ν → ∞ , uniformly for t ∈ [ τ 1 , τ 2 ] t\in[\tau_{1},\tau_{2}] t ∈ [ τ 1 , τ 2 ] and s ∈ [ s 1 , s 2 ] s\in[s_{1},s_{2}] s ∈ [ s 1 , s 2 ] , we have
(i)
ϑ ℓ ( ν t ) = − ( sinh α ) − 1 2 e ν Λ ( t ) ( 1 + O ( ν − 1 ) ) (102) \vartheta_{\ell}(\nu t)=-(\sinh\alpha)^{-\frac{1}{2}}e^{\nu\Lambda(t)}(1+O(\nu^{-1})) \tag*{(102)} ϑ ℓ ( ν t ) = − ( sinh α ) − 2 1 e ν Λ ( t ) ( 1 + O ( ν − 1 )) (102) (ii)
ϑ ℓ ′ ( ν t ) ϑ ℓ ( ν t ) = − sinh α + O ( ν − 1 ) (103) \frac{\vartheta'_{\ell}(\nu t)}{\vartheta_{\ell}(\nu t)}=-\sinh\alpha+O(\nu^{-1}) \tag*{(103)} ϑ ℓ ( ν t ) ϑ ℓ ′ ( ν t ) = − sinh α + O ( ν − 1 ) (103) (iii)
ψ ℓ ( ν s ) = ( sin β ) − 1 2 ( cos Ω ν ( s ) + O ( ν − 1 ) ) , ψ ℓ ′ ( ν s ) = − ( sin β ) 1 2 ( sin Ω ν ( s ) + O ( ν − 1 ) ) . (104) \begin{aligned}
\psi_{\ell}(\nu s)&=(\sin\beta)^{-\frac{1}{2}}(\cos\Omega_{\nu}(s)+O(\nu^{-1})),\\
\psi'_{\ell}(\nu s)&=-(\sin\beta)^{\frac{1}{2}}(\sin\Omega_{\nu}(s)+O(\nu^{-1})). \tag*{(104)}
\end{aligned} ψ ℓ ( ν s ) ψ ℓ ′ ( ν s ) = ( sin β ) − 2 1 ( cos Ω ν ( s ) + O ( ν − 1 )) , = − ( sin β ) 2 1 ( sin Ω ν ( s ) + O ( ν − 1 )) . (104)
Proof. We use
ϑ ℓ ( w ) = π w 2 Y ν ( w ) , ϑ ℓ ′ = ϑ ℓ 2 w + π w 2 Y ν ′ , ψ ℓ ′ = ψ ℓ 2 w + π w 2 J ν ′ . (105) \vartheta_{\ell}(w)=\sqrt{\frac{\pi w}{2}}Y_{\nu}(w),\qquad\vartheta'_{\ell}=\frac{\vartheta_{\ell}}{2w}+\sqrt{\frac{\pi w}{2}}Y'_{\nu},\qquad\psi'_{\ell}=\frac{\psi_{\ell}}{2w}+\sqrt{\frac{\pi w}{2}}J'_{\nu}. \tag*{(105)} ϑ ℓ ( w ) = 2 π w Y ν ( w ) , ϑ ℓ ′ = 2 w ϑ ℓ + 2 π w Y ν ′ , ψ ℓ ′ = 2 w ψ ℓ + 2 π w J ν ′ . (105)
The points then follow using the Debye expansions (10.19.3), (10.19.4), (10.19.6), (10.19.7) in [27 ] : (i), (ii) from Y ν ( ν sech α ) Y_{\nu}(\nu\operatorname{sech}\alpha) Y ν ( ν sech α ) , Y ν ′ ( ν sech α ) Y'_{\nu}(\nu\operatorname{sech}\alpha) Y ν ′ ( ν sech α ) (note t / tanh α = ( sinh α ) − 1 / 2 \sqrt{t/\tanh\alpha}=(\sinh\alpha)^{-1/2} t / tanh α = ( sinh α ) − 1/2 ), (iii) from J ν ( ν sec β ) J_{\nu}(\nu\sec\beta) J ν ( ν sec β ) , J ν ′ ( ν sec β ) J'_{\nu}(\nu\sec\beta) J ν ′ ( ν sec β ) (note sec β / tan β = ( sin β ) − 1 / 2 \sqrt{\sec\beta/\tan\beta}=(\sin\beta)^{-1/2} sec β / tan β = ( sin β ) − 1/2 , sec β sin 2 β / 2 = ( sin β ) 1 / 2 \sqrt{\sec\beta\sin2\beta/2}=(\sin\beta)^{1/2} sec β sin 2 β /2 = ( sin β ) 1/2 ); the term ψ ℓ / ( 2 w ) \psi_{\ell}/(2w) ψ ℓ / ( 2 w ) is O ( ν − 1 ) O(\nu^{-1}) O ( ν − 1 ) .
Lemma 5.2. (i) ∣ ψ ℓ ( w ) ∣ ≤ ∣ w ∣ e ∣ Im w ∣ |\psi_{\ell}(w)|\leq|w|e^{|\operatorname{Im}w|} ∣ ψ ℓ ( w ) ∣ ≤ ∣ w ∣ e ∣ Im w ∣ for all w ∈ C w\in\mathbb{C} w ∈ C ;
(ii) We let 0 < τ − < τ + < 1 0<\tau_{-}<\tau_{+}<1 0 < τ − < τ + < 1 . Then there is a constant C C C such that for all ℓ ≥ 1 \ell\geq1 ℓ ≥ 1 and z = x + i y z=x+iy z = x + i y with x ν ∈ [ τ − , τ + ] \frac{x}{\nu}\in[\tau_{-},\tau_{+}] ν x ∈ [ τ − , τ + ] and ∣ y ∣ ≤ 3 |y|\leq3 ∣ y ∣ ≤ 3 we have
∣ ψ ℓ ( z ) ∣ ≤ C ν 1 2 e − ν Λ ( x ν ) ∣ ϑ ℓ ( z ) ∣ ≤ C ν 1 2 e ν Λ ( x ν ) (106) |\psi_{\ell}(z)|\leq C\nu^{\frac{1}{2}}e^{-\nu\Lambda(\frac{x}{\nu})}
\qquad
|\vartheta_{\ell}(z)|\leq C\nu^{\frac{1}{2}}e^{\nu\Lambda(\frac{x}{\nu})} \tag*{(106)} ∣ ψ ℓ ( z ) ∣ ≤ C ν 2 1 e − ν Λ ( ν x ) ∣ ϑ ℓ ( z ) ∣ ≤ C ν 2 1 e ν Λ ( ν x ) (106)
Proof. (i) follows from j ℓ ( w ) = ( − i ) ℓ 2 ∫ − 1 1 e i w s P ℓ ( s ) d s j_{\ell}(w)=\frac{(-i)^{\ell}}{2}\int_{-1}^{1}e^{iws}P_{\ell}(s)\,ds j ℓ ( w ) = 2 ( − i ) ℓ ∫ − 1 1 e i w s P ℓ ( s ) d s , see (10.54.2) in [27 ] , and ∣ P ℓ ∣ ≤ 1 |P_{\ell}|\leq1 ∣ P ℓ ∣ ≤ 1 on [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] .
(ii) Let τ = x / ν \tau=x/\nu τ = x / ν and α = arcsech τ \alpha=\operatorname{arcsech}\tau α = arcsech τ , so that τ cosh α = 1 \tau\cosh\alpha=1 τ cosh α = 1 and τ sinh α = tanh α \tau\sinh\alpha=\tanh\alpha τ sinh α = tanh α .
Bound for J ν J_{\nu} J ν . For Re z > 0 \operatorname{Re}z>0 Re z > 0 , J ν ( z ) = 1 2 π i ∫ ∞ − π i ∞ + π i e z sinh w − ν w d w J_{\nu}(z)=\frac{1}{2\pi i}\int_{\infty-\pi i}^{\infty+\pi i}e^{z\sinh w-\nu w}\,dw J ν ( z ) = 2 π i 1 ∫ ∞ − π i ∞ + π i e z s i n h w − ν w d w , see §10.9 in [27 ] . The integrand is entire and decays along Im w = ± π \operatorname{Im}w=\pm\pi Im w = ± π , so the path may be taken to be: Im w = − π \operatorname{Im}w=-\pi Im w = − π from + ∞ +\infty + ∞ to α − π i \alpha-\pi i α − π i , then the segment Re w = α \operatorname{Re}w=\alpha Re w = α , then Im w = π \operatorname{Im}w=\pi Im w = π back to $+\infty.
Hence ∣ J ν ( z ) ∣ ≤ ( e 3 / τ − + 1 ) e − ν Λ ( τ ) |J_\nu(z)|\le(e^{3/\tau_-}+1)e^{-\nu\Lambda(\tau)} ∣ J ν ( z ) ∣ ≤ ( e 3/ τ − + 1 ) e − ν Λ ( τ ) .
Bound for Y ν Y_\nu Y ν . Since cos ν π = 0 \cos\nu\pi=0 cos ν π = 0 , (10.9.7) in [27 ] gives
Y ν ( z ) = 1 π ∫ 0 π sin ( z sin θ − ν θ ) d θ − 1 π ∫ 0 ∞ e ν σ − z sinh σ d σ . Y_\nu(z)=\frac{1}{\pi}\int_0^\pi\sin(z\sin\theta-\nu\theta)\,d\theta-\frac{1}{\pi}\int_0^\infty e^{\nu\sigma-z\sinh\sigma}\,d\sigma. Y ν ( z ) = π 1 ∫ 0 π sin ( z sin θ − ν θ ) d θ − π 1 ∫ 0 ∞ e ν σ − z s i n h σ d σ .
The first term is at most cosh y ≤ e 3 \cosh y\le e^3 cosh y ≤ e 3 . In the second term, ∣ e ν σ − z sinh σ ∣ = e ν ( σ − τ sinh σ ) ≤ e ν Λ ( τ ) |e^{\nu\sigma-z\sinh\sigma}|=e^{\nu(\sigma-\tau\sinh\sigma)}\le e^{\nu\Lambda(\tau)} ∣ e ν σ − z s i n h σ ∣ = e ν ( σ − τ s i n h σ ) ≤ e ν Λ ( τ ) , the maximum being attained at σ = α \sigma=\alpha σ = α . Choose T = T ( τ − ) T=T(\tau_-) T = T ( τ − ) with τ − sinh σ ≥ 2 σ \tau_-\sinh\sigma\ge2\sigma τ − sinh σ ≥ 2 σ for σ ≥ T \sigma\ge T σ ≥ T ; then the integrand is bounded by e − ν σ e^{-\nu\sigma} e − ν σ for σ ≥ T \sigma\ge T σ ≥ T . Hence
∣ Y ν ( z ) ∣ ≤ e 3 + π − 1 ( T e ν Λ ( τ ) + ν − 1 ) ≤ C e ν Λ ( τ ) . |Y_\nu(z)|\le e^3+\pi^{-1}(Te^{\nu\Lambda(\tau)}+\nu^{-1})\le Ce^{\nu\Lambda(\tau)}. ∣ Y ν ( z ) ∣ ≤ e 3 + π − 1 ( T e ν Λ ( τ ) + ν − 1 ) ≤ C e ν Λ ( τ ) .
The factor π ∣ z ∣ / 2 ≤ C ν 1 / 2 \sqrt{\pi|z|/2}\le C\nu^{1/2} π ∣ z ∣/2 ≤ C ν 1/2 completes the proof.
Real quasimodes
Lemma 5.3. Let κ > 0 \kappa>0 κ > 0 . There are ℓ 0 \ell_0 ℓ 0 and c 1 > 0 c_1>0 c 1 > 0 such that for each ℓ ≥ ℓ 0 \ell\ge\ell_0 ℓ ≥ ℓ 0 there is x ℓ ∈ ( t 1 ν , t 2 ν ) x_\ell\in(t_1\nu,t_2\nu) x ℓ ∈ ( t 1 ν , t 2 ν ) with
Q κ , ℓ ( x ℓ ) = 0 , ∣ Q κ , ℓ ′ ( x ℓ ) ∣ ≥ c 1 e ν Λ ( x ℓ / ν ) . Q_{\kappa,\ell}(x_\ell)=0,\qquad|Q'_{\kappa,\ell}(x_\ell)|\ge c_1e^{\nu\Lambda(x_\ell/\nu)}. Q κ , ℓ ( x ℓ ) = 0 , ∣ Q κ , ℓ ′ ( x ℓ ) ∣ ≥ c 1 e ν Λ ( x ℓ / ν ) . Proof. Let I ν = [ t 1 ν , t 2 ν ] I_\nu=[t_1\nu,t_2\nu] I ν = [ t 1 ν , t 2 ν ] . For x ∈ I ν x\in I_\nu x ∈ I ν put t = x / ν t=x/\nu t = x / ν and s = m 0 t ∈ [ m 0 t 1 , m 0 t 2 ] ⊂ ( 1 , m 0 ) s=m_0t\in[m_0t_1,m_0t_2]\subset(1,m_0) s = m 0 t ∈ [ m 0 t 1 , m 0 t 2 ] ⊂ ( 1 , m 0 ) , define α , β \alpha,\beta α , β as in Lemma 5.1, and set
a = sin β , b = sinh α , A = a 2 + κ 2 b 2 , a=\sin\beta,\qquad b=\sinh\alpha,\qquad A=\sqrt{a^2+\kappa^2b^2}, a = sin β , b = sinh α , A = a 2 + κ 2 b 2 , ϖ ∈ ( 0 , π 2 ) with A cos ϖ = κ b , A sin ϖ = a , \varpi\in\left(0,\frac{\pi}{2}\right)\quad\text{with}\quad A\cos\varpi=\kappa b,\qquad A\sin\varpi=a, ϖ ∈ ( 0 , 2 π ) with A cos ϖ = κb , A sin ϖ = a , all smooth functions of t ∈ [ t 1 , t 2 ] t\in[t_1,t_2] t ∈ [ t 1 , t 2 ] . Let a 1 = a ( t 1 ) > 0 a_1=a(t_1)>0 a 1 = a ( t 1 ) > 0 . By Lemma 5.1(i), ϑ ℓ ≠ 0 \vartheta_\ell\ne0 ϑ ℓ = 0 on I ν I_\nu I ν for large ν \nu ν , so Q κ , ℓ = ϑ ℓ q Q_{\kappa,\ell}=\vartheta_\ell q Q κ , ℓ = ϑ ℓ q there, with
q ( x ) = ψ ℓ ′ ( m 0 x ) − κ L ( x ) ψ ℓ ( m 0 x ) , L = ϑ ℓ ′ ϑ ℓ , q(x)=\psi'_\ell(m_0x)-\kappa L(x)\psi_\ell(m_0x),\qquad L=\frac{\vartheta'_\ell}{\vartheta_\ell}, q ( x ) = ψ ℓ ′ ( m 0 x ) − κ L ( x ) ψ ℓ ( m 0 x ) , L = ϑ ℓ ϑ ℓ ′ , real valued on I ν I_\nu I ν . With Ω ( x ) = Ω ν ( m 0 x / ν ) + ϖ ( x / ν ) \Omega(x)=\Omega_\nu(m_0x/\nu)+\varpi(x/\nu) Ω ( x ) = Ω ν ( m 0 x / ν ) + ϖ ( x / ν ) , Lemma 5.1(ii), (iii) give
q ( x ) = a − 1 / 2 A cos Ω ( x ) + O ( ν − 1 ) . (107) q(x)=a^{-1/2}A\cos\Omega(x)+O(\nu^{-1}). \tag*{(107)} q ( x ) = a − 1/2 A cos Ω ( x ) + O ( ν − 1 ) . (107) For q ′ q' q ′ use (5.2) and ℓ ( ℓ + 1 ) = ν 2 − 1 4 \ell(\ell+1)=\nu^2-\frac14 ℓ ( ℓ + 1 ) = ν 2 − 4 1 :
d d x ψ ℓ ′ ( m 0 x ) = − m 0 ( a 2 + O ( ν − 2 ) ) ψ ℓ ( m 0 x ) \frac{d}{dx}\psi'_\ell(m_0x)=-m_0\left(a^2+O(\nu^{-2})\right)\psi_\ell(m_0x) d x d ψ ℓ ′ ( m 0 x ) = − m 0 ( a 2 + O ( ν − 2 ) ) ψ ℓ ( m 0 x ) , because 1 − ν 2 / ( m 0 x ) 2 = 1 − s − 2 = a 2 1-\nu^2/(m_0x)^2=1-s^{-2}=a^2 1 − ν 2 / ( m 0 x ) 2 = 1 − s − 2 = a 2 ;
L ′ = ℓ ( ℓ + 1 ) x 2 − 1 − L 2 = b 2 − L 2 + O ( ν − 2 ) = O ( ν − 1 ) L'=\frac{\ell(\ell+1)}{x^2}-1-L^2=b^2-L^2+O(\nu^{-2})=O(\nu^{-1}) L ′ = x 2 ℓ ( ℓ + 1 ) − 1 − L 2 = b 2 − L 2 + O ( ν − 2 ) = O ( ν − 1 ) , by Lemma 5.1(ii) and t − 2 − 1 = b 2 t^{-2}-1=b^2 t − 2 − 1 = b 2 .
Hence q ′ = − m 0 a 2 ψ ℓ ( m 0 x ) + m 0 κ b ψ ℓ ′ ( m 0 x ) + O ( ν − 1 ) q'=-m_0a^2\psi_\ell(m_0x)+m_0\kappa b\psi'_\ell(m_0x)+O(\nu^{-1}) q ′ = − m 0 a 2 ψ ℓ ( m 0 x ) + m 0 κb ψ ℓ ′ ( m 0 x ) + O ( ν − 1 ) , and by Lemma 5.1(iii)
q ′ ( x ) = − m 0 a − 1 / 2 A sin Ω ( x ) + O ( ν − 1 ) . (108) q'(x)=-m_0a^{-1/2}A\sin\Omega(x)+O(\nu^{-1}). \tag*{(108)} q ′ ( x ) = − m 0 a − 1/2 A sin Ω ( x ) + O ( ν − 1 ) . (108) The phase. Since
d d s ( s 2 − 1 − arcsec s ) = s 2 − 1 s , \frac{d}{ds}\left(\sqrt{s^2-1}-\operatorname{arcsec}s\right)=\frac{\sqrt{s^2-1}}{s}, d s d ( s 2 − 1 − arcsec s ) = s s 2 − 1 , we have d d x Ω ν ( m 0 x / ν ) = m 0 a ≥ m 0 a 1 \frac{d}{dx}\Omega_\nu(m_0x/\nu)=m_0a\ge m_0a_1 d x d Ω ν ( m 0 x / ν ) = m 0 a ≥ m 0 a 1 , while d d x ϖ ( x / ν ) = O ( ν − 1 ) \frac{d}{dx}\varpi(x/\nu)=O(\nu^{-1}) d x d ϖ ( x / ν ) = O ( ν − 1 ) . So Ω \Omega Ω is increasing on I ν I_\nu I ν , and Ω ( t 2 ν ) − Ω ( t 1 ν ) ≥ 1 2 m 0 a 1 ( t 2 − t 1 ) ν ≥ 2 π \Omega(t_2\nu)-\Omega(t_1\nu)\ge\frac12m_0a_1(t_2-t_1)\nu\ge2\pi Ω ( t 2 ν ) − Ω ( t 1 ν ) ≥ 2 1 m 0 a 1 ( t 2 − t 1 ) ν ≥ 2 π for ν ≥ ν 0 \nu\ge\nu_0 ν ≥ ν 0 . Choose k ∈ Z k\in\mathbb{Z} k ∈ Z and x − < x + x_-<x_+ x − < x + in I ν I_\nu I ν with Ω ( x − ) = k π \Omega(x_-)=k\pi Ω ( x − ) = k π and Ω ( x + ) = ( k + 1 ) π \Omega(x_+)=(k+1)\pi Ω ( x + ) = ( k + 1 ) π .
Sign change. As a ≤ 1 a\le1 a ≤ 1 , a − 1 / 2 A ≥ a 1 a^{-1/2}A\ge a_1 a − 1/2 A ≥ a 1 . So (5.13) gives ( − 1 ) k q ( x − ) ≥ a 1 − C ν − 1 > 0 > ( − 1 ) k q ( x + ) (-1)^kq(x_-)\ge a_1-C\nu^{-1}>0>(-1)^kq(x_+) ( − 1 ) k q ( x − ) ≥ a 1 − C ν − 1 > 0 > ( − 1 ) k q ( x + ) , and there is x ℓ ∈ ( x − , x + ) x_\ell\in(x_-,x_+) x ℓ ∈ ( x − , x + ) with q ( x ℓ ) = 0 q(x_\ell)=0 q ( x ℓ ) = 0 , i.e. Q κ , ℓ ( x ℓ ) = 0 Q_{\kappa,\ell}(x_\ell)=0 Q κ , ℓ ( x ℓ ) = 0 .
Derivative. At x ℓ x_\ell x ℓ , (5.13) gives ∣ cos Ω ( x ℓ ) ∣ ≤ C / ( a 1 ν ) |\cos\Omega(x_\ell)|\le C/(a_1\nu) ∣ cos Ω ( x ℓ ) ∣ ≤ C / ( a 1 ν ) , hence ∣ sin Ω ( x ℓ ) ∣ ≥ 1 2 |\sin\Omega(x_\ell)|\ge\frac12 ∣ sin Ω ( x ℓ ) ∣ ≥ 2 1 , and (5.14) gives ∣ q ′ ( x ℓ ) ∣ ≥ 1 4 m 0 a 1 3 / 2 |q'(x_\ell)|\ge\frac14m_0a_1^{3/2} ∣ q ′ ( x ℓ ) ∣ ≥ 4 1 m 0 a 1 3/2 . Finally Q κ , ℓ ′ ( x ℓ ) = ϑ ℓ ( x ℓ ) q ′ ( x ℓ ) Q'_{\kappa,\ell}(x_\ell)=\vartheta_\ell(x_\ell)q'(x_\ell) Q κ , ℓ ′ ( x ℓ ) = ϑ ℓ ( x ℓ ) q ′ ( x ℓ ) , and by Lemma 5.1(i), ∣ ϑ ℓ ( x ℓ ) ∣ ≥ 1 2 ( sinh α ( t 1 ) ) − 1 / 2 e ν Λ ( x ℓ / ν ) |\vartheta_\ell(x_\ell)|\ge\frac12(\sinh\alpha(t_1))^{-1/2}e^{\nu\Lambda(x_\ell/\nu)} ∣ ϑ ℓ ( x ℓ ) ∣ ≥ 2 1 ( sinh α ( t 1 ) ) − 1/2 e ν Λ ( x ℓ / ν ) . This gives the claim with c 1 = 1 8 m 0 a 1 3 / 2 ( sinh α ( t 1 ) ) − 1 / 2 c_1=\frac18m_0a_1^{3/2}(\sinh\alpha(t_1))^{-1/2} c 1 = 8 1 m 0 a 1 3/2 ( sinh α ( t 1 ) ) − 1/2 .
From quasimodes to resonances
Proposition 5.4. Let κ > 0 \kappa> 0 κ > 0 and let x ℓ x_\ell x ℓ be as in Lemma 5.3. There are ℓ 0 \ell_0 ℓ 0 and C > 0 C > 0 C > 0 such that for ℓ ≥ ℓ 0 \ell\ge\ell_0 ℓ ≥ ℓ 0 , D κ , ℓ D_{\kappa,\ell} D κ , ℓ has exactly one zero z ℓ z_\ell z ℓ in the disc { ∣ z − x ℓ ∣ < ρ ℓ } \{|z-x_\ell|<\rho_\ell\} { ∣ z − x ℓ ∣ < ρ ℓ } , where
ρ ℓ = C ν 3 2 e − 2 ν Λ ( x ℓ / ν ) . \rho_\ell=C\nu^{\frac{3}{2}}e^{-2\nu\Lambda(x_\ell/\nu)}. ρ ℓ = C ν 2 3 e − 2 ν Λ ( x ℓ / ν ) .
Proof. Let Λ ℓ = Λ ( x ℓ / ν ) ≥ Λ ( t 2 ) > 0 \Lambda_\ell=\Lambda(x_\ell/\nu)\ge\Lambda(t_2)>0 Λ ℓ = Λ ( x ℓ / ν ) ≥ Λ ( t 2 ) > 0 , and apply Lemma 5.2(ii) with τ − = t 1 2 \tau_-=\frac{t_1}{2} τ − = 2 t 1 , τ + = 1 + t 2 2 \tau_+=\frac{1+t_2}{2} τ + = 2 1 + t 2 . For ν \nu ν large, every z z z with ∣ z − x ℓ ∣ ≤ 2 |z-x_\ell|\le2 ∣ z − x ℓ ∣ ≤ 2 satisfies Re z / ν ∈ [ τ − , τ + ] \operatorname{Re}z/\nu\in[\tau_-,\tau_+] Re z / ν ∈ [ τ − , τ + ] , ∣ Im z ∣ ≤ 2 |\operatorname{Im}z|\le2 ∣ Im z ∣ ≤ 2 , and
ν ∣ Λ ( Re z / ν ) − Λ ℓ ∣ ≤ 2 ∥ Λ ′ ∥ L ∞ [ τ − , τ + ] . \nu|\Lambda(\operatorname{Re}z/\nu)-\Lambda_\ell|\le2\|\Lambda'\|_{L^\infty[\tau_-,\tau_+]}. ν ∣Λ ( Re z / ν ) − Λ ℓ ∣ ≤ 2∥ Λ ′ ∥ L ∞ [ τ − , τ + ] .
Upper bounds. On { ∣ z − x ℓ ∣ ≤ 2 } \{|z-x_\ell|\le2\} { ∣ z − x ℓ ∣ ≤ 2 } , Lemma 5.2 gives ∣ ψ ℓ ∣ ≤ C ν 1 / 2 e − ν Λ ℓ |\psi_\ell|\le C\nu^{1/2}e^{-\nu\Lambda_\ell} ∣ ψ ℓ ∣ ≤ C ν 1/2 e − ν Λ ℓ , ∣ ϑ ℓ ∣ ≤ C ν 1 / 2 e ν Λ ℓ |\vartheta_\ell|\le C\nu^{1/2}e^{\nu\Lambda_\ell} ∣ ϑ ℓ ∣ ≤ C ν 1/2 e ν Λ ℓ and ∣ ψ ℓ ( m 0 ⋅ ) ∣ ≤ C ν |\psi_\ell(m_0\cdot)|\le C\nu ∣ ψ ℓ ( m 0 ⋅ ) ∣ ≤ C ν . Cauchy’s estimates on discs of radius 1 give the same bounds, with new constants, for the derivatives of order ≤ 3 \le3 ≤ 3 of these three functions on { ∣ z − x ℓ ∣ ≤ 1 } \{|z-x_\ell|\le1\} { ∣ z − x ℓ ∣ ≤ 1 } . Since P κ , ℓ P_{\kappa,\ell} P κ , ℓ and Q κ , ℓ ′ ′ Q''_{\kappa,\ell} Q κ , ℓ ′′ are finite sums of products of such derivatives (one factor of ψ ℓ ( i ) \psi_\ell^{(i)} ψ ℓ ( i ) or ϑ ℓ ( i ) \vartheta_\ell^{(i)} ϑ ℓ ( i ) , one of ( ψ ℓ ( m 0 ⋅ ) ) ( j ) (\psi_\ell(m_0\cdot))^{(j)} ( ψ ℓ ( m 0 ⋅ ) ) ( j ) ),
sup ∣ z − x ℓ ∣ ≤ 1 ∣ P κ , ℓ ( z ) ∣ ≤ B 2 ν 3 2 e − ν Λ ℓ , sup ∣ z − x ℓ ∣ ≤ 1 ∣ Q κ , ℓ ′ ′ ( z ) ∣ ≤ B 3 ν 3 2 e ν Λ ℓ . (109) \sup_{|z-x_\ell|\le1}|P_{\kappa,\ell}(z)|\le B_2\nu^{\frac{3}{2}}e^{-\nu\Lambda_\ell},\qquad
\sup_{|z-x_\ell|\le1}|Q''_{\kappa,\ell}(z)|\le B_3\nu^{\frac{3}{2}}e^{\nu\Lambda_\ell}. \tag*{(109)} ∣ z − x ℓ ∣ ≤ 1 sup ∣ P κ , ℓ ( z ) ∣ ≤ B 2 ν 2 3 e − ν Λ ℓ , ∣ z − x ℓ ∣ ≤ 1 sup ∣ Q κ , ℓ ′′ ( z ) ∣ ≤ B 3 ν 2 3 e ν Λ ℓ . (109)
Choice of radius. Set ρ ℓ = 4 B 2 c 1 − 1 ν 3 2 e − 2 ν Λ ℓ \rho_\ell=4B_2c_1^{-1}\nu^{\frac{3}{2}}e^{-2\nu\Lambda_\ell} ρ ℓ = 4 B 2 c 1 − 1 ν 2 3 e − 2 ν Λ ℓ . For ℓ ≥ ℓ 0 \ell\ge\ell_0 ℓ ≥ ℓ 0 we have ρ ℓ ≤ 1 \rho_\ell\le1 ρ ℓ ≤ 1 and ρ ℓ B 3 ν 3 2 ≤ c 1 \rho_\ell B_3\nu^{\frac{3}{2}}\le c_1 ρ ℓ B 3 ν 2 3 ≤ c 1 , since Λ ℓ ≥ Λ ( t 2 ) > 0 \Lambda_\ell\ge\Lambda(t_2)>0 Λ ℓ ≥ Λ ( t 2 ) > 0 .
Lower bound for Q κ , ℓ Q_{\kappa,\ell} Q κ , ℓ . Since Q κ , ℓ ( x ℓ ) = 0 Q_{\kappa,\ell}(x_\ell)=0 Q κ , ℓ ( x ℓ ) = 0 , Taylor’s formula, (5.15) and Lemma 5.3 give, on ∣ z − x ℓ ∣ = ρ ℓ |z-x_\ell|=\rho_\ell ∣ z − x ℓ ∣ = ρ ℓ ,
∣ Q κ , ℓ ( z ) − Q κ , ℓ ′ ( x ℓ ) ( z − x ℓ ) ∣ ≤ 1 2 ρ ℓ 2 B 3 ν 3 2 e ν Λ ℓ ≤ 1 2 c 1 ρ ℓ e ν Λ ℓ ≤ 1 2 ∣ Q κ , ℓ ′ ( x ℓ ) ( z − x ℓ ) ∣ . |Q_{\kappa,\ell}(z)-Q'_{\kappa,\ell}(x_\ell)(z-x_\ell)|
\le\frac{1}{2}\rho_\ell^2B_3\nu^{\frac{3}{2}}e^{\nu\Lambda_\ell}
\le\frac{1}{2}c_1\rho_\ell e^{\nu\Lambda_\ell}
\le\frac{1}{2}|Q'_{\kappa,\ell}(x_\ell)(z-x_\ell)|. ∣ Q κ , ℓ ( z ) − Q κ , ℓ ′ ( x ℓ ) ( z − x ℓ ) ∣ ≤ 2 1 ρ ℓ 2 B 3 ν 2 3 e ν Λ ℓ ≤ 2 1 c 1 ρ ℓ e ν Λ ℓ ≤ 2 1 ∣ Q κ , ℓ ′ ( x ℓ ) ( z − x ℓ ) ∣.
Hence ∣ Q κ , ℓ ( z ) ∣ ≥ 1 2 c 1 ρ ℓ e ν Λ ℓ = 2 B 2 ν 3 2 e − ν Λ ℓ > ∣ P κ , ℓ ( z ) ∣ |Q_{\kappa,\ell}(z)|\ge\frac{1}{2}c_1\rho_\ell e^{\nu\Lambda_\ell}=2B_2\nu^{\frac{3}{2}}e^{-\nu\Lambda_\ell}>|P_{\kappa,\ell}(z)| ∣ Q κ , ℓ ( z ) ∣ ≥ 2 1 c 1 ρ ℓ e ν Λ ℓ = 2 B 2 ν 2 3 e − ν Λ ℓ > ∣ P κ , ℓ ( z ) ∣ on the circle.
Rouché, twice. By the first inequality, Q κ , ℓ Q_{\kappa,\ell} Q κ , ℓ has exactly one zero in the disc, like z ↦ Q κ , ℓ ′ ( x ℓ ) ( z − x ℓ ) z\mapsto Q'_{\kappa,\ell}(x_\ell)(z-x_\ell) z ↦ Q κ , ℓ ′ ( x ℓ ) ( z − x ℓ ) . By the second, ∣ D κ , ℓ − i Q κ , ℓ ∣ = ∣ P κ , ℓ ∣ < ∣ i Q κ , ℓ ∣ |D_{\kappa,\ell}-iQ_{\kappa,\ell}|=|P_{\kappa,\ell}|<|iQ_{\kappa,\ell}| ∣ D κ , ℓ − i Q κ , ℓ ∣ = ∣ P κ , ℓ ∣ < ∣ i Q κ , ℓ ∣ on the circle, so D κ , ℓ D_{\kappa,\ell} D κ , ℓ also has exactly one zero there. The constant in ρ ℓ \rho_\ell ρ ℓ is C = 4 B 2 / c 1 C=4B_2/c_1 C = 4 B 2 / c 1 .
Proof of Theorem 1.5. Let κ = η 1 / η 0 \kappa=\eta_1/\eta_0 κ = η 1 / η 0 and let z ℓ z_\ell z ℓ (ℓ ≥ ℓ 0 \ell\ge\ell_0 ℓ ≥ ℓ 0 ) be the zeros given by Proposition 5.4. By (5.6), D ℓ 1 ( z ℓ ) = 0 D_\ell^1(z_\ell)=0 D ℓ 1 ( z ℓ ) = 0 . Set λ ℓ = z ℓ / ( r 0 ϵ 0 μ 0 ) \lambda_\ell=z_\ell/(r_0\sqrt{\epsilon_0\mu_0}) λ ℓ = z ℓ / ( r 0 ϵ 0 μ 0 ) . Put Λ ℓ = Λ ( x ℓ / ν ) \Lambda_\ell=\Lambda(x_\ell/\nu) Λ ℓ = Λ ( x ℓ / ν ) . Lemma 5.1(i) and the Cauchy bounds in the proof of Proposition 5.4 give, on ∣ z − x ℓ ∣ ≤ ρ ℓ |z-x_\ell|\le\rho_\ell ∣ z − x ℓ ∣ ≤ ρ ℓ ,
∣ ϑ ℓ ( z ) ∣ ≥ ( c − C ρ ℓ ν 1 / 2 ) e ν Λ ℓ , ∣ ψ ℓ ( z ) ∣ ≤ C ν 1 / 2 e − ν Λ ℓ . |\vartheta_\ell(z)|\ge(c-C\rho_\ell\nu^{1/2})e^{\nu\Lambda_\ell},\qquad
|\psi_\ell(z)|\le C\nu^{1/2}e^{-\nu\Lambda_\ell}. ∣ ϑ ℓ ( z ) ∣ ≥ ( c − C ρ ℓ ν 1/2 ) e ν Λ ℓ , ∣ ψ ℓ ( z ) ∣ ≤ C ν 1/2 e − ν Λ ℓ .
For large ℓ \ell ℓ , the first bound exceeds the second, so ξ ℓ ( z ) = ψ ℓ ( z ) + i ϑ ℓ ( z ) ≠ 0 \xi_\ell(z)=\psi_\ell(z)+i\vartheta_\ell(z)\ne0 ξ ℓ ( z ) = ψ ℓ ( z ) + i ϑ ℓ ( z ) = 0 throughout this disc.
(i) By Proposition 2.6 and its proof, D ℓ 1 ( z ℓ ) = 0 D_\ell^1(z_\ell)=0 D ℓ 1 ( z ℓ ) = 0 gives the non-trivial purely outgoing solutions u ℓ m = t M ℓ m 0 ( k 1 r ) u_{\ell m}=tM_{\ell m}^0(k_1r) u ℓ m = t M ℓ m 0 ( k 1 r ) (r < r 0 r<r_0 r < r 0 ), u ℓ m = M ℓ m − ( k 0 r ) u_{\ell m}=M_{\ell m}^-(k_0r) u ℓ m = M ℓ m − ( k 0 r ) (r > r 0 r>r_0 r > r 0 ), with t = h ℓ ( 1 ) ( x 0 ) / j ℓ ( x 1 ) t=h_\ell^{(1)}(x_0)/j_\ell(x_1) t = h ℓ ( 1 ) ( x 0 ) / j ℓ ( x 1 ) (j ℓ ( x 1 ) ≠ 0 j_\ell(x_1)\ne0 j ℓ ( x 1 ) = 0 , since otherwise D ℓ 1 ( z ℓ ) = η 0 ξ ℓ ( z ℓ ) ψ ℓ ′ ( m 0 z ℓ ) ≠ 0 D_\ell^1(z_\ell)=\eta_0\xi_\ell(z_\ell)\psi_\ell'(m_0z_\ell)\ne0 D ℓ 1 ( z ℓ ) = η 0 ξ ℓ ( z ℓ ) ψ ℓ ′ ( m 0 z ℓ ) = 0 ). Hence λ ℓ \lambda_\ell λ ℓ is a resonance of P E P_E P E , with the 2 ℓ + 1 2\ell+1 2 ℓ + 1 resonance states u ℓ m u_{\ell m} u ℓ m , ∣ m ∣ ≤ ℓ |m|\le\ell ∣ m ∣ ≤ ℓ ; it is a genuine pole of A ℓ 1 A_\ell^1 A ℓ 1 , because N ℓ 1 N_\ell^1 N ℓ 1 and D ℓ 1 D_\ell^1 D ℓ 1 have no common zero (if both vanish, the Wronskian ψ ℓ ϑ ℓ ′ − ψ ℓ ′ ϑ ℓ = 1 \psi_\ell\vartheta_\ell'-\psi_\ell'\vartheta_\ell=1 ψ ℓ ϑ ℓ ′ − ψ ℓ ′ ϑ ℓ = 1 at x 0 x_0 x 0 forces ψ ℓ ( x 1 ) = ψ ℓ ′ ( x 1 ) = 0 \psi_\ell(x_1)=\psi_\ell'(x_1)=0 ψ ℓ ( x 1 ) = ψ ℓ ′ ( x 1 ) = 0 ).
(ii) Proposition 2.5 gives Im λ ℓ < 0 \operatorname{Im}\lambda_\ell<0 Im λ ℓ < 0 .
(iii) Since ∣ z ℓ − x ℓ ∣ < ρ ℓ ≤ 1 |z_\ell-x_\ell|<\rho_\ell\le1 ∣ z ℓ − x ℓ ∣ < ρ ℓ ≤ 1 and x ℓ ∈ ( t 1 ν , t 2 ν ) x_\ell\in(t_1\nu,t_2\nu) x ℓ ∈ ( t 1 ν , t 2 ν ) , we have t 1 ν − 1 < Re z ℓ < t 2 ν + 1 t_1\nu-1<\operatorname{Re}z_\ell<t_2\nu+1 t 1 ν − 1 < Re z ℓ < t 2 ν + 1 , so Re λ ℓ → ∞ \operatorname{Re}\lambda_\ell\to\infty Re λ ℓ → ∞ .
(iv) As Λ \Lambda Λ is decreasing, ∣ Im z ℓ ∣ < ρ ℓ ≤ C ν 3 2 e − 2 ν Λ ( t 2 ) ≤ C ′ e − ν Λ ( t 2 ) |\operatorname{Im} z_\ell| < \rho_\ell\le C\nu^{\frac{3}{2}}e^{-2\nu\Lambda(t_2)} \le C'e^{-\nu\Lambda(t_2)} ∣ Im z ℓ ∣ < ρ ℓ ≤ C ν 2 3 e − 2 ν Λ ( t 2 ) ≤ C ′ e − ν Λ ( t 2 ) . Using ν > ( Re z ℓ − 1 ) / t 2 \nu> (\operatorname{Re} z_\ell- 1)/t_2 ν > ( Re z ℓ − 1 ) / t 2 ,
∣ Im λ ℓ ∣ ≤ C e − c Re λ ℓ , c = r 0 ϵ 0 μ 0 Λ ( t 2 ) t 2 > 0. |\operatorname{Im}\lambda_\ell| \le Ce^{-c\operatorname{Re}\lambda_\ell}, \qquad c = \frac{r_0\sqrt{\epsilon_0\mu_0}\Lambda(t_2)}{t_2} > 0. ∣ Im λ ℓ ∣ ≤ C e − c Re λ ℓ , c = t 2 r 0 ϵ 0 μ 0 Λ ( t 2 ) > 0.
The constants depend only on m 0 , η 0 , η 1 , r 0 m_0,\eta_0,\eta_1,r_0 m 0 , η 0 , η 1 , r 0 and ϵ 0 μ 0 \epsilon_0\mu_0 ϵ 0 μ 0 .
(v) Given C 1 , C 2 > 0 C_1,C_2>0 C 1 , C 2 > 0 , (iii) and (iv) show that λ ℓ ∈ { ∣ Re λ ∣ ≥ C 1 , Im λ ≥ − C 2 } \lambda_\ell\in\{|\operatorname{Re}\lambda|\ge C_1,\ \operatorname{Im}\lambda\ge-C_2\} λ ℓ ∈ { ∣ Re λ ∣ ≥ C 1 , Im λ ≥ − C 2 } for all large ℓ \ell ℓ , so no resonance free strip exists.
Acknowledgements
The author A.W. wishes to thank Yan-Long Fang and Nelia Charalambous for useful discussions about Mie scattering that led to this paper. She also wishes to thank Andras Vasy for suggesting the topic of whispering gallery modes.
Declarations
Funding. None
Competing interests. None
Data availability. Data sharing is not applicable to this article, as no datasets were generated or analysed during the current study.
Ethics approval and consent to participate. Not applicable.
Consent for publication. Not applicable.
Materials availability. Not applicable.
Code availability. Not applicable.
Author contribution. Not applicable (single author).
Use of large language models. All ideas here are the original work of the author. Claude Opus 5.5 AI was used for proofreading/language organization and for assistance to develop the difficult Bessel function asymptotics specifically in Regions 2 and 3 of Theorem 1.4 and Theorem 1.5, after the various regimes were fixed first by the author.
Outgoing asymptotics and uniqueness
The outgoing condition in Definition 1.3 depends only on the field near infinity. For real λ ≠ 0 \lambda\ne0 λ = 0 , an outgoing field satisfying the free equation outside a ball has the expansion
f = e i k 0 r r ( Φ + O ( 1 r ) ) r → ∞ (110) f=\frac{e^{ik_0r}}{r}\left(\Phi+O\left(\frac{1}{r}\right)\right)\quad r\to\infty \tag*{(110)} f = r e i k 0 r ( Φ + O ( r 1 ) ) r → ∞ (110)
with Φ ∈ C ∞ ( S 2 ; C 3 ) \Phi\in C^\infty(\mathbb{S}^2;\mathbb{C}^3) Φ ∈ C ∞ ( S 2 ; C 3 ) the restriction of an entire function on C 3 \mathbb{C}^3 C 3 to the sphere. The expansion can be differentiated in r r r .
Theorem A.1 (Maxwell transmission uniqueness). Let Ω \Omega Ω be any bounded Lipschitz open set with the positive coefficients (1.1). Let λ ∈ R ∖ { 0 } \lambda\in\mathbb{R}\setminus\{0\} λ ∈ R ∖ { 0 } . An outgoing u ∈ H l o c ( curl ) u\in H_{\mathrm{loc}}(\operatorname{curl}) u ∈ H loc ( curl ) satisfying ( P E − λ 2 ) u = 0 (P_E-\lambda^2)u=0 ( P E − λ 2 ) u = 0 and the transmission conditions (1.6) vanishes. The same holds for P H P_H P H .
Sketch adapted from [29 ] . The equation gives div ( ϵ u ) = 0 \operatorname{div}(\epsilon u)=0 div ( ϵ u ) = 0 , so the outgoing coefficient Φ \Phi Φ in (A.1) is tangential. Green’s identity, with cancellation of the interface terms, yields
0 = lim R → ∞ Im ∫ ∂ B R ( r ^ × μ 0 − 1 curl u ) ⋅ u ‾ d S = − k 0 μ 0 ∥ Φ ∥ L 2 ( S 2 ) 2 . 0=\lim_{R\to\infty}\operatorname{Im}\int_{\partial B_R}(\hat{r}\times\mu_0^{-1}\operatorname{curl}u)\cdot\overline{u}\,dS=-\frac{k_0}{\mu_0}\|\Phi\|_{L^2(\mathbb{S}^2)}^2. 0 = R → ∞ lim Im ∫ ∂ B R ( r ^ × μ 0 − 1 curl u ) ⋅ u d S = − μ 0 k 0 ∥Φ ∥ L 2 ( S 2 ) 2 .
Thus Φ = 0 \Phi=0 Φ = 0 , and the scalar Rellich lemma applied componentwise gives u = 0 u=0 u = 0 in the unbounded component of R 3 ∖ Ω ‾ \mathbb{R}^3\setminus\overline{\Omega} R 3 ∖ Ω . Set H = ( i λ μ ) − 1 curl u H=(i\lambda\mu)^{-1}\operatorname{curl}u H = ( iλ μ ) − 1 curl u . At an interface neighborhood adjoining a region where both fields vanish, their tangential traces on the other side are zero, and the fields can be extended by zero. These extensions solve the constant-coefficient Maxwell system for that side in a full neighborhood of the patch. Their components solve the Helmholtz equation distributionally and are analytic there; vanishing on one side forces vanishing across the neighborhood. Interior unique continuation and repetition across the finitely many interface components give u = H = 0 u=H=0 u = H = 0 everywhere, including any bounded complementary components. Interchanging ϵ , μ \epsilon,\mu ϵ , μ for P H P_H P H , the result also holds for the magnetic operator. Conjugation also gives incoming uniqueness.
Bessel and Hankel estimates
We use ν \nu ν , ψ ℓ \psi_\ell ψ ℓ and ξ ℓ \xi_\ell ξ ℓ from (2.6); primes denote differentiation with respect to the full argument.
Basic identities
The Riccati equation is (96) . For f ℓ = j ℓ f_\ell=j_\ell f ℓ = j ℓ , h ℓ ( 1 ) h_\ell^{(1)} h ℓ ( 1 ) or h ℓ ( 2 ) h_\ell^{(2)} h ℓ ( 2 ) , the derivative recurrence is
f ℓ ′ ( z ) = f ℓ − 1 ( z ) − ℓ + 1 z f ℓ ( z ) . (111) f_\ell'(z)=f_{\ell-1}(z)-\frac{\ell+1}{z}f_\ell(z). \tag*{(111)} f ℓ ′ ( z ) = f ℓ − 1 ( z ) − z ℓ + 1 f ℓ ( z ) . (111)
The spherical Bessel Wronskian is
j ℓ ( z ) y ℓ ′ ( z ) − y ℓ ( z ) j ℓ ′ ( z ) = z − 2 . (112) j_\ell(z)y_\ell'(z)-y_\ell(z)j_\ell'(z)=z^{-2}. \tag*{(112)} j ℓ ( z ) y ℓ ′ ( z ) − y ℓ ( z ) j ℓ ′ ( z ) = z − 2 . (112)
Subturning estimates
The real Debye estimates away from the turning point are stated in Lemma 5.1. We record here the uniform logarithmic-derivative estimates needed up to the edge of the turning window.
For Region 1, fix a , C 0 > 0 a,C_0>0 a , C 0 > 0 and assume
a ≤ Re w ≤ ν − M ν 1 / 3 , ∣ Im w ∣ ≤ C 0 , M ≥ M 0 , ν ≥ ν 0 ( M , a , C 0 ) . a\leq\operatorname{Re}w\leq\nu-M\nu^{1/3},\qquad|\operatorname{Im}w|\leq C_0,\qquad M\geq M_0,\qquad\nu\geq\nu_0(M,a,C_0). a ≤ Re w ≤ ν − M ν 1/3 , ∣ Im w ∣ ≤ C 0 , M ≥ M 0 , ν ≥ ν 0 ( M , a , C 0 ) .
Set β ( w ) = ν 2 − w 2 / w \beta(w)=\sqrt{\nu^2-w^2}/w β ( w ) = ν 2 − w 2 / w , positive for 0 < w < ν 0<w<\nu 0 < w < ν , and ρ ν ( w ) = ν 2 / ∣ ν 2 − w 2 ∣ 3 / 2 \rho_\nu(w)=\nu^2/|\nu^2-w^2|^{3/2} ρ ν ( w ) = ν 2 /∣ ν 2 − w 2 ∣ 3/2 . The uniform Bessel expansions and their derivatives, [27 ] , (10.20.4), (10.20.5), (10.20.7), (10.20.8) and §10.20(ii), combined with the Airy expansions (9.7.5)–(9.7.8), give
J ν ′ ( w ) J ν ( w ) = β ( w ) ( 1 + O ( ρ ν ( w ) ) ) , \frac{J_\nu'(w)}{J_\nu(w)}=\beta(w)(1+O(\rho_\nu(w))), J ν ( w ) J ν ′ ( w ) = β ( w ) ( 1 + O ( ρ ν ( w ))) ,
Y ν ′ ( w ) Y ν ( w ) = − β ( w ) ( 1 + O ( ρ ν ( w ) ) ) , \frac{Y_\nu'(w)}{Y_\nu(w)}=-\beta(w)(1+O(\rho_\nu(w))), Y ν ( w ) Y ν ′ ( w ) = − β ( w ) ( 1 + O ( ρ ν ( w ))) ,
( H ν ( 1 ) ) ′ ( w ) H ν ( 1 ) ( w ) = − β ( w ) ( 1 + O ( ρ ν ( w ) ) ) . \frac{(H_\nu^{(1)})'(w)}{H_\nu^{(1)}(w)}=-\beta(w)(1+O(\rho_\nu(w))). H ν ( 1 ) ( w ) ( H ν ( 1 ) ) ′ ( w ) = − β ( w ) ( 1 + O ( ρ ν ( w ))) .
The constants depend only on a , C 0 a,C_0 a , C 0 . Since ∣ ν 2 − w 2 ∣ ≥ c M ν 4 / 3 |\nu^2-w^2|\geq cM\nu^{4/3} ∣ ν 2 − w 2 ∣ ≥ c M ν 4/3 , the relative errors are O ( M − 3 / 2 ) O(M^{-3/2}) O ( M − 3/2 ) , uniformly at both complex arguments w = z , m 0 z w=z,m_0z w = z , m 0 z used in Region 1.
Two-term Airy expansions in the turning windows
For the full complex argument w = ν + τ ν 1 / 3 w=\nu+\tau\nu^{1/3} w = ν + τ ν 1/3 , put
s = − 2 1 / 3 τ , p = 2 1 / 3 e − π i / 3 , σ = p τ , s=-2^{1/3}\tau,\qquad p=2^{1/3}e^{-\pi i/3},\qquad\sigma=p\tau, s = − 2 1/3 τ , p = 2 1/3 e − π i /3 , σ = p τ ,
A = Ai ( s ) , A ^ = Ai ( σ ) . A=\operatorname{Ai}(s),\qquad\widehat{A}=\operatorname{Ai}(\sigma). A = Ai ( s ) , A = Ai ( σ ) .
Here A ′ A' A ′ and A ^ ′ \widehat{A}' A ′ denote derivatives of the Airy function with respect to its argument. We use w = z w=z w = z in the exterior turning window and w = m 0 z w=m_0z w = m 0 z in the interior turning window.
Proposition B.1 (Uniform two-term Airy expansion). Fix M > 0 M>0 M > 0 and 0 < δ < π 0<\delta<\pi 0 < δ < π . There are constants C M , δ > 0 C_{M,\delta}>0 C M , δ > 0 and ν 0 = ν 0 ( M , δ ) \nu_0=\nu_0(M,\delta) ν 0 = ν 0 ( M , δ ) such that, for every positive half-integer ν ≥ ν 0 \nu\ge\nu_0 ν ≥ ν 0 and every ∣ τ ∣ ≤ M |\tau|\le M ∣ τ ∣ ≤ M , with ∣ ph ( w / ν ) ∣ ≤ π − δ |\operatorname{ph}(w/\nu)|\le\pi-\delta ∣ ph ( w / ν ) ∣ ≤ π − δ , the following formulas hold:
ψ ℓ ( w ) = π 2 − 1 / 6 ν 1 / 6 { A + 1 ν 2 / 3 ( 3 τ 10 A + 3 2 1 / 3 τ 2 10 A ′ ) } + r ψ , (113) \psi_\ell(w)=\sqrt{\pi}\,2^{-1/6}\nu^{1/6}\left\{A+\frac{1}{\nu^{2/3}}\left(\frac{3\tau}{10}A+\frac{3\,2^{1/3}\tau^2}{10}A'\right)\right\}+r_\psi, \tag*{(113)} ψ ℓ ( w ) = π 2 − 1/6 ν 1/6 { A + ν 2/3 1 ( 10 3 τ A + 10 3 2 1/3 τ 2 A ′ ) } + r ψ , (113) ψ ℓ ′ ( w ) = π 2 − 1 / 6 ν − 1 / 6 { − 2 1 / 3 A ′ + 1 ν 2 / 3 [ ( 3 10 + 3 τ 3 5 ) A + 3 2 1 / 3 τ 10 A ′ ] } + r ψ ′ , (114) \psi_\ell'(w)=\sqrt{\pi}\,2^{-1/6}\nu^{-1/6}\left\{-2^{1/3}A'+\frac{1}{\nu^{2/3}}\left[\left(\frac{3}{10}+\frac{3\tau^3}{5}\right)A+\frac{3\,2^{1/3}\tau}{10}A'\right]\right\}+r_{\psi'}, \tag*{(114)} ψ ℓ ′ ( w ) = π 2 − 1/6 ν − 1/6 { − 2 1/3 A ′ + ν 2/3 1 [ ( 10 3 + 5 3 τ 3 ) A + 10 3 2 1/3 τ A ′ ] } + r ψ ′ , (114) ξ ℓ ( w ) = π 2 5 / 6 e − π i / 3 ν 1 / 6 { A ^ + 1 ν 2 / 3 ( 3 τ 10 A ^ + 3 2 1 / 3 e 2 π i / 3 τ 2 10 A ^ ′ ) } + r ξ , (115) \xi_\ell(w)=\sqrt{\pi}\,2^{5/6}e^{-\pi i/3}\nu^{1/6}\left\{\widehat{A}+\frac{1}{\nu^{2/3}}\left(\frac{3\tau}{10}\widehat{A}+\frac{3\,2^{1/3}e^{2\pi i/3}\tau^2}{10}\widehat{A}'\right)\right\}+r_\xi, \tag*{(115)} ξ ℓ ( w ) = π 2 5/6 e − π i /3 ν 1/6 { A + ν 2/3 1 ( 10 3 τ A + 10 3 2 1/3 e 2 π i /3 τ 2 A ′ ) } + r ξ , (115) ξ ℓ ′ ( w ) = π 2 5 / 6 e − π i / 3 ν − 1 / 6 { p A ^ ′ + 1 ν 2 / 3 [ ( 3 10 + 3 τ 3 5 ) A ^ − 3 p τ 10 A ^ ′ ] } + r ξ ′ , (116) \xi_\ell'(w)=\sqrt{\pi}\,2^{5/6}e^{-\pi i/3}\nu^{-1/6}\left\{p\widehat{A}'+\frac{1}{\nu^{2/3}}\left[\left(\frac{3}{10}+\frac{3\tau^3}{5}\right)\widehat{A}-\frac{3p\tau}{10}\widehat{A}'\right]\right\}+r_{\xi'}, \tag*{(116)} ξ ℓ ′ ( w ) = π 2 5/6 e − π i /3 ν − 1/6 { p A ′ + ν 2/3 1 [ ( 10 3 + 5 3 τ 3 ) A − 10 3 p τ A ′ ] } + r ξ ′ , (116) The remainders satisfy the uniform absolute bounds
∣ r ψ ∣ + ∣ r ξ ∣ ≤ C M , δ ν − 7 / 6 , ∣ r ψ ′ ∣ + ∣ r ξ ′ ∣ ≤ C M , δ ν − 3 / 2 . (117) |r_\psi|+|r_\xi|\le C_{M,\delta}\nu^{-7/6},\qquad|r_{\psi'}|+|r_{\xi'}|\le C_{M,\delta}\nu^{-3/2}. \tag*{(117)} ∣ r ψ ∣ + ∣ r ξ ∣ ≤ C M , δ ν − 7/6 , ∣ r ψ ′ ∣ + ∣ r ξ ′ ∣ ≤ C M , δ ν − 3/2 . (117) Proof. Apply the transition expansions (10.19.8), (10.19.9), (10.19.12) and (10.19.13) of [27 ] to (2.6), and expand ( w / ν ) 1 / 2 = 1 + 1 2 τ ν − 2 / 3 + O M ( ν − 4 / 3 ) (w/\nu)^{1/2}=1+\frac{1}{2}\tau\nu^{-2/3}+O_M(\nu^{-4/3}) ( w / ν ) 1/2 = 1 + 2 1 τ ν − 2/3 + O M ( ν − 4/3 ) . This gives the displayed coefficients.
The complex uniform expansions in §10.20(ii) of [27 ] , with the remainder estimates on pp. 419–425 of [26 ] , give absolute errors O M , δ ( ν − 5 / 3 ) O_{M,\delta}(\nu^{-5/3}) O M , δ ( ν − 5/3 ) for J ν , H ν ( 1 ) J_\nu,H_\nu^{(1)} J ν , H ν ( 1 ) and O M , δ ( ν − 2 ) O_{M,\delta}(\nu^{-2}) O M , δ ( ν − 2 ) for their derivatives on ∣ τ ∣ ≤ M |\tau|\le M ∣ τ ∣ ≤ M . Multiplying by π w / 2 \sqrt{\pi w/2} π w /2 and including the derivative of this prefactor yields (B.7).
These additive estimates remain valid at zeros of A A A or A ′ A' A ′ ; no division by ψ ℓ ( m 0 z ) \psi_\ell(m_0z) ψ ℓ ( m 0 z ) is needed in the interior turning window.
Exterior Airy expansion
For q > 0 q>0 q > 0 we define ζ = ζ ( q ) \zeta=\zeta(q) ζ = ζ ( q ) by
( d ζ d q ) 2 = 1 − q 2 ζ q 2 . (118) \left(\frac{d\zeta}{dq}\right)^2=\frac{1-q^2}{\zeta q^2}. \tag*{(118)} ( d q d ζ ) 2 = ζ q 2 1 − q 2 . (118)
This is piecewise defined for real q q q as
2 3 ζ 3 / 2 = log ( 1 + 1 − q 2 q ) − 1 − q 2 , 0 < q ≤ 1 (119) \frac{2}{3}\zeta^{3/2}=\log\left(\frac{1+\sqrt{1-q^2}}{q}\right)-\sqrt{1-q^2},\qquad0<q\le1 \tag*{(119)} 3 2 ζ 3/2 = log ( q 1 + 1 − q 2 ) − 1 − q 2 , 0 < q ≤ 1 (119)
and
2 3 ( − ζ ) 3 / 2 = q 2 − 1 − arcsec q , 1 ≤ q < ∞ . (120) \frac{2}{3}(-\zeta)^{3/2}=\sqrt{q^2-1}-\operatorname{arcsec}q,\qquad1\le q<\infty. \tag*{(120)} 3 2 ( − ζ ) 3/2 = q 2 − 1 − arcsec q , 1 ≤ q < ∞. (120)
These are equations (10.20.1)–(10.20.3) of [27 ] . For q = z / ν q=z/\nu q = z / ν in Region 3, the uniform Hankel expansions [27 (10.20.6), (10.20.9), §10.20(ii)] give (3.16), with the uniform O ( ν − 1 ) O(\nu^{-1}) O ( ν − 1 ) remainder stated there. The range is
1 + M ν − 2 / 3 ≤ Re q < K ν / m 0 , ∣ Im q ∣ ≤ C 2 / ν , 1+M\nu^{-2/3}\le\operatorname{Re}q<K\nu/m_0,\qquad|\operatorname{Im}q|\le C_2/\nu, 1 + M ν − 2/3 ≤ Re q < K ν / m 0 , ∣ Im q ∣ ≤ C 2 / ν ,
with branches continued from real q > 1 q>1 q > 1 . Only this logarithmic-derivative estimate is needed. The Airy phase convention follows from Ai ( s ) − i Bi ( s ) = 2 e − π i / 3 Ai ( e 2 π i / 3 s ) \operatorname{Ai}(s)-i\operatorname{Bi}(s)=2e^{-\pi i/3}\operatorname{Ai}(e^{2\pi i/3}s) Ai ( s ) − i Bi ( s ) = 2 e − π i /3 Ai ( e 2 π i /3 s ) .
Large-argument estimates
Set
ω ( w ) = w − π ν 2 − π 4 . (121) \omega(w)=w-\frac{\pi\nu}{2}-\frac{\pi}{4}. \tag*{(121)} ω ( w ) = w − 2 π ν − 4 π . (121)
For ∣ w ∣ ≥ K ν 2 |w|\ge K\nu^2 ∣ w ∣ ≥ K ν 2 , K ≥ 1 K\ge1 K ≥ 1 , and ∣ Im w ∣ ≤ C 0 |\operatorname{Im}w|\le C_0 ∣ Im w ∣ ≤ C 0 , we have
ψ ℓ ( w ) = cos ω ( w ) + O ( ν 2 ∣ w ∣ ) , (122) \psi_\ell(w)=\cos\omega(w)+O\left(\frac{\nu^2}{|w|}\right), \tag*{(122)} ψ ℓ ( w ) = cos ω ( w ) + O ( ∣ w ∣ ν 2 ) , (122)
ψ ℓ ′ ( w ) = − sin ω ( w ) + O ( ν 2 ∣ w ∣ ) , (123) \psi'_\ell(w)=-\sin\omega(w)+O\left(\frac{\nu^2}{|w|}\right), \tag*{(123)} ψ ℓ ′ ( w ) = − sin ω ( w ) + O ( ∣ w ∣ ν 2 ) , (123)
ξ ℓ ( w ) = e i ω ( w ) ( 1 + O ( ν 2 ∣ w ∣ ) ) , (124) \xi_\ell(w)=e^{i\omega(w)}\left(1+O\left(\frac{\nu^2}{|w|}\right)\right), \tag*{(124)} ξ ℓ ( w ) = e iω ( w ) ( 1 + O ( ∣ w ∣ ν 2 ) ) , (124)
ξ ℓ ′ ( w ) = e i ω ( w ) ( i + O ( ν 2 ∣ w ∣ ) ) . (125) \xi'_\ell(w)=e^{i\omega(w)}\left(i+O\left(\frac{\nu^2}{|w|}\right)\right). \tag*{(125)} ξ ℓ ′ ( w ) = e iω ( w ) ( i + O ( ∣ w ∣ ν 2 ) ) . (125)
The constants are uniform in ℓ ≥ 1 \ell\ge1 ℓ ≥ 1 and K ≥ 1 K\ge1 K ≥ 1 and depend only on C 0 C_0 C 0 . The estimates for ξ ℓ , ξ ℓ ′ \xi_\ell,\xi'_\ell ξ ℓ , ξ ℓ ′ also hold without the restriction on Im w \operatorname{Im}w Im w ; for fixed ℓ \ell ℓ their remainders are O ℓ ( ∣ w ∣ − 1 ) O_\ell(|w|^{-1}) O ℓ ( ∣ w ∣ − 1 ) as ∣ w ∣ → ∞ |w|\to\infty ∣ w ∣ → ∞ .
Proof. Write ξ ℓ ( w ) = e i ω ( w ) ( 1 + r ℓ ( w ) ) \xi_\ell(w)=e^{i\omega(w)}(1+r_\ell(w)) ξ ℓ ( w ) = e iω ( w ) ( 1 + r ℓ ( w )) . In the finite spherical Hankel formula [27 (10.49.1), (10.49.6)] , the coefficient of w − k w^{-k} w − k has modulus at most [ ℓ ( ℓ + 1 ) / 2 ] k / k ! [\ell(\ell+1)/2]^k/k! [ ℓ ( ℓ + 1 ) /2 ] k / k ! . Thus, for w ≠ 0 w\ne0 w = 0 ,
∣ r ℓ ( w ) ∣ ≤ exp ( ℓ ( ℓ + 1 ) 2 ∣ w ∣ ) − 1 , |r_\ell(w)|\le\exp\left(\frac{\ell(\ell+1)}{2|w|}\right)-1, ∣ r ℓ ( w ) ∣ ≤ exp ( 2∣ w ∣ ℓ ( ℓ + 1 ) ) − 1 ,
∣ r ℓ ′ ( w ) ∣ ≤ ℓ ( ℓ + 1 ) 2 ∣ w ∣ 2 exp ( ℓ ( ℓ + 1 ) 2 ∣ w ∣ ) . |r'_\ell(w)|\le\frac{\ell(\ell+1)}{2|w|^2}\exp\left(\frac{\ell(\ell+1)}{2|w|}\right). ∣ r ℓ ′ ( w ) ∣ ≤ 2∣ w ∣ 2 ℓ ( ℓ + 1 ) exp ( 2∣ w ∣ ℓ ( ℓ + 1 ) ) .
Since ℓ ( ℓ + 1 ) < ν 2 \ell(\ell+1)<\nu^2 ℓ ( ℓ + 1 ) < ν 2 , these imply (B.14)–(B.15). The identity ψ ℓ ( w ) = 1 2 ( ξ ℓ ( w ) + ξ ℓ ( w ) ‾ ) \psi_\ell(w)=\frac{1}{2}(\xi_\ell(w)+\overline{\xi_\ell(w)}) ψ ℓ ( w ) = 2 1 ( ξ ℓ ( w ) + ξ ℓ ( w ) ) and its derivative give (B.12)–(B.13), using ∣ e ± i ω ( w ) ∣ ≤ e C 0 |e^{\pm i\omega(w)}|\le e^{C_0} ∣ e ± iω ( w ) ∣ ≤ e C 0 .
AI usage. Claude Opus 5.5 was used in the development of the estimates in Regions 2 and 3 in both main theorems.