Introduction
The model and the inverse problem
The isotropic elastic medium occupies a bounded domain Ω ⊂ R 3 \Omega\subset\mathbb{R}^{3} Ω ⊂ R 3 with C ∞ C^{\infty} C ∞ boundary. The Lamé parameters λ \lambda λ and μ \mu μ and the mass density ρ \rho ρ are real-valued and smooth up to the boundary. For a complex-valued displacement field u : Ω → C 3 u:\Omega\to\mathbb{C}^{3} u : Ω → C 3 , the linearized strain tensor and the Cauchy stress tensor are defined by
ε ( u ) = 1 2 ( ∇ u + ( ∇ u ) T ) , σ λ , μ ( u ) = 2 μ ε ( u ) + λ ( div u ) I 3 . (1) \varepsilon(u)=\frac{1}{2}\left(\nabla u+(\nabla u)^{T}\right), \qquad\sigma_{\lambda,\mu}(u)=2\mu\varepsilon(u)+\lambda(\operatorname{div}u)I_{3}. \tag*{(1)} ε ( u ) = 2 1 ( ∇ u + ( ∇ u ) T ) , σ λ , μ ( u ) = 2 μ ε ( u ) + λ ( div u ) I 3 . (1)
Here the superscript T T T denotes matrix transposition, and I 3 I_{3} I 3 is the 3 × 3 3\times3 3 × 3 identity matrix. For vector fields v v v , we use the gradient convention ( ∇ v ) i j = ∂ j v i (\nabla v)_{ij}=\partial_{j}v_{i} ( ∇ v ) ij = ∂ j v i . The corresponding elasticity operator is given by
L λ , μ u : = div σ λ , μ ( u ) . (2) L_{\lambda,\mu}u:=\operatorname{div}\sigma_{\lambda,\mu}(u). \tag*{(2)} L λ , μ u := div σ λ , μ ( u ) . (2)
At a prescribed angular frequency ω > 0 \omega>0 ω > 0 , the time-harmonic displacement field u u u satisfies
L λ , μ u + ω 2 ρ u = 0 in Ω . (3) L_{\lambda,\mu}u+\omega^{2}\rho u=0 \quad\text{in }\Omega. \tag*{(3)} L λ , μ u + ω 2 ρ u = 0 in Ω. (3)
The coefficients are assumed to satisfy the strong convexity and positivity conditions
μ > 0 , 3 λ + 2 μ > 0 , ρ > 0 on Ω ‾ . (4) \mu>0,\qquad3\lambda+2\mu>0,\qquad\rho>0 \quad\text{on }\overline{\Omega}. \tag*{(4)} μ > 0 , 3 λ + 2 μ > 0 , ρ > 0 on Ω . (4)
For the subsequent analysis, denote the auxiliary zeroth-order coefficient W W W by
W : = ω 2 ρ . (5) W:=\omega^{2}\rho. \tag*{(5)} W := ω 2 ρ . (5)
The subsequent reduction employs W W W , whereas the inverse boundary problem concerns the physical coefficient triple ( λ , μ , ρ ) (\lambda,\mu,\rho) ( λ , μ , ρ ) . For the weak Dirichlet realization, write
P D λ , μ , W : H 0 1 ( Ω ; C 3 ) ⟶ H − 1 ( Ω ; C 3 ) , P D λ , μ , W v : = ( L λ , μ + W I 3 ) v . P_{D}^{\lambda,\mu,W}:H_{0}^{1}(\Omega;\mathbb{C}^{3})\longrightarrow H^{-1}(\Omega;\mathbb{C}^{3}),\qquad P_{D}^{\lambda,\mu,W}v:=(L_{\lambda,\mu}+WI_{3})v. P D λ , μ , W : H 0 1 ( Ω ; C 3 ) ⟶ H − 1 ( Ω ; C 3 ) , P D λ , μ , W v := ( L λ , μ + W I 3 ) v .
The nonresonance condition is the triviality of its kernel:
ker P D λ , μ , W = { 0 } . (6) \ker P_{D}^{\lambda,\mu,W}=\{0\}. \tag*{(6)} ker P D λ , μ , W = { 0 } . (6)
Here the operator is understood in the weak sense, and H 0 1 ( Ω ; C 3 ) H_{0}^{1}(\Omega;\mathbb{C}^{3}) H 0 1 ( Ω ; C 3 ) incorporates the homogeneous Dirichlet boundary condition.
Let
γ 0 : H 1 ( Ω ; C 3 ) ⟶ H 1 / 2 ( ∂ Ω ; C 3 ) \gamma_{0}:H^{1}(\Omega;\mathbb{C}^{3})\longrightarrow H^{1/2}(\partial\Omega;\mathbb{C}^{3}) γ 0 : H 1 ( Ω ; C 3 ) ⟶ H 1/2 ( ∂ Ω ; C 3 )
be the Dirichlet trace operator. For each prescribed boundary value f ∈ H 1 / 2 ( ∂ Ω ; C 3 ) f\in H^{1/2}(\partial\Omega;\mathbb{C}^{3}) f ∈ H 1/2 ( ∂ Ω ; C 3 ) , consider
{ ( L λ , μ + W I 3 ) u = 0 in Ω , γ 0 u = f on ∂ Ω . (7) \begin{cases}
(L_{\lambda,\mu}+WI_{3})u=0 & \text{in }\Omega,\\
\gamma_{0}u=f & \text{on }\partial\Omega. \tag*{(7)}
\end{cases} { ( L λ , μ + W I 3 ) u = 0 γ 0 u = f in Ω , on ∂ Ω. (7)
Under the strong-convexity assumptions (4) and the nonresonance condition (6) , this problem is well posed: for every f ∈ H 1 / 2 ( ∂ Ω ; C 3 ) f\in H^{1/2}(\partial\Omega;\mathbb{C}^{3}) f ∈ H 1/2 ( ∂ Ω ; C 3 ) , there exists a unique weak solution u f ∈ H 1 ( Ω ; C 3 ) u_{f}\in H^{1}(\Omega;\mathbb{C}^{3}) u f ∈ H 1 ( Ω ; C 3 ) . The well-posedness is established in Subsection 1.7. The outward unit normal to ∂ Ω \partial\Omega ∂ Ω is denoted by ν \nu ν . The Dirichlet-to-Neumann operator is defined by
Λ λ , μ , ρ Ω ( ω ) : H 1 / 2 ( ∂ Ω ; C 3 ) ⟶ H − 1 / 2 ( ∂ Ω ; C 3 ) , (8) \Lambda_{\lambda,\mu,\rho}^{\Omega}(\omega):H^{1/2}(\partial\Omega;\mathbb{C}^{3})\longrightarrow H^{-1/2}(\partial\Omega;\mathbb{C}^{3}), \tag*{(8)} Λ λ , μ , ρ Ω ( ω ) : H 1/2 ( ∂ Ω ; C 3 ) ⟶ H − 1/2 ( ∂ Ω ; C 3 ) , (8)
Λ λ , μ , ρ Ω ( ω ) f = σ λ , μ ( u f ) ν ∣ ∂ Ω , \Lambda_{\lambda,\mu,\rho}^{\Omega}(\omega)f=\left.\sigma_{\lambda,\mu}(u_{f})\nu\right|_{\partial\Omega}, Λ λ , μ , ρ Ω ( ω ) f = σ λ , μ ( u f ) ν ∣ ∂ Ω ,
where the boundary traction is interpreted in the weak sense.
At the fixed frequency ω \omega ω , let A ω \mathcal{A}_{\omega} A ω denote the class of all triples ( λ , μ , ρ ) ∈ C ∞ ( Ω ‾ ; R ) 3 (\lambda,\mu,\rho)\in C^{\infty}(\overline{\Omega};\mathbb{R})^{3} ( λ , μ , ρ ) ∈ C ∞ ( Ω ; R ) 3 satisfying the positivity conditions (4) and the nonresonance condition (6) . The boundary map defines
F ω : A ω ⟶ L ( H 1 / 2 ( ∂ Ω ; C 3 ) , H − 1 / 2 ( ∂ Ω ; C 3 ) ) , (9) \mathcal{F}_{\omega}:\mathcal{A}_{\omega}\longrightarrow\mathcal{L}\left(H^{1/2}(\partial\Omega;\mathbb{C}^{3}),H^{-1/2}(\partial\Omega;\mathbb{C}^{3})\right), \tag*{(9)} F ω : A ω ⟶ L ( H 1/2 ( ∂ Ω ; C 3 ) , H − 1/2 ( ∂ Ω ; C 3 ) ) , (9)
( λ , μ , ρ ) ⟼ Λ λ , μ , ρ Ω ( ω ) . (\lambda,\mu,\rho)\longmapsto\Lambda_{\lambda,\mu,\rho}^{\Omega}(\omega). ( λ , μ , ρ ) ⟼ Λ λ , μ , ρ Ω ( ω ) .
The inverse boundary value problem is to determine the coefficient triple ( λ , μ , ρ ) (\lambda,\mu,\rho) ( λ , μ , ρ ) from the fixed-frequency Dirichlet-to-Neumann map Λ λ , μ , ρ Ω ( ω ) \Lambda^\Omega_{\lambda,\mu,\rho}(\omega) Λ λ , μ , ρ Ω ( ω ) . With F ω \mathcal{F}_\omega F ω defined in (9) , the corresponding uniqueness question is whether F ω \mathcal{F}_\omega F ω is injective on A ω \mathcal{A}_\omega A ω . More precisely, for any two admissible coefficient triples,
F ω ( λ 1 , μ 1 , ρ 1 ) = F ω ( λ 2 , μ 2 , ρ 2 ) only if ( λ 1 , μ 1 , ρ 1 ) = ( λ 2 , μ 2 , ρ 2 ) in Ω ‾ . \mathcal{F}_\omega(\lambda_1,\mu_1,\rho_1)=\mathcal{F}_\omega(\lambda_2,\mu_2,\rho_2)\quad\text{only if}\quad(\lambda_1,\mu_1,\rho_1)=(\lambda_2,\mu_2,\rho_2)\quad\text{in }\overline{\Omega}. F ω ( λ 1 , μ 1 , ρ 1 ) = F ω ( λ 2 , μ 2 , ρ 2 ) only if ( λ 1 , μ 1 , ρ 1 ) = ( λ 2 , μ 2 , ρ 2 ) in Ω .
For j = 1 , 2 j=1,2 j = 1 , 2 , we continue to write W j = ω 2 ρ j W_j=\omega^2\rho_j W j = ω 2 ρ j for the zeroth-order potential associated with the fixed frequency.
The uniqueness theorem
Theorem 1.1. Let Ω ⊂ R 3 \Omega\subset\mathbb{R}^3 Ω ⊂ R 3 be a bounded domain with C ∞ C^\infty C ∞ boundary, and fix ω > 0 \omega>0 ω > 0 . For j = 1 , 2 j=1,2 j = 1 , 2 , let λ j , μ j , ρ j ∈ C ∞ ( Ω ‾ ; R ) \lambda_j,\mu_j,\rho_j\in C^\infty(\overline{\Omega};\mathbb{R}) λ j , μ j , ρ j ∈ C ∞ ( Ω ; R ) satisfy the strong convexity and positivity conditions (4) . For j = 1 , 2 j=1,2 j = 1 , 2 , set
P D , j : = P D λ j , μ j , W j . P_{D,j}:=P_D^{\lambda_j,\mu_j,W_j}. P D , j := P D λ j , μ j , W j . Assume that both physical Dirichlet realizations are nonresonant:
ker P D , j = { 0 } , j = 1 , 2. \operatorname{ker}P_{D,j}=\{0\},\qquad j=1,2. ker P D , j = { 0 } , j = 1 , 2. If the Dirichlet-to-Neumann maps satisfy
Λ λ 1 , μ 1 , ρ 1 Ω ( ω ) = Λ λ 2 , μ 2 , ρ 2 Ω ( ω ) , (10) \Lambda^\Omega_{\lambda_1,\mu_1,\rho_1}(\omega)=\Lambda^\Omega_{\lambda_2,\mu_2,\rho_2}(\omega), \tag*{(10)} Λ λ 1 , μ 1 , ρ 1 Ω ( ω ) = Λ λ 2 , μ 2 , ρ 2 Ω ( ω ) , (10) then
λ 1 = λ 2 , μ 1 = μ 2 , ρ 1 = ρ 2 on Ω ‾ . \lambda_1=\lambda_2,\qquad\mu_1=\mu_2,\qquad\rho_1=\rho_2\quad\text{on }\overline{\Omega}. λ 1 = λ 2 , μ 1 = μ 2 , ρ 1 = ρ 2 on Ω .
Theorem 1.1 establishes the injectivity of F ω \mathcal{F}_\omega F ω on the entire admissible class A ω \mathcal{A}_\omega A ω . Its quantifiers range over every pair of smooth coefficient triples satisfying the stated positivity and nonresonance conditions. The boundary Taylor data used in the proof are recovered from the measured map through Theorem 2.5; they therefore enter the argument as consequences of the data. The nonresonance condition is imposed on the original physical Dirichlet realizations on Ω \Omega Ω .
Related work
Boundary determination for elastic systems has been studied at both the level of the full boundary Taylor series and through explicit reconstruction formulas. Nakamura and Uhlmann [15 ] determined the full Taylor series of the elastic tensor at the boundary. In the isotropic setting, explicit reconstruction formulas for the Lamé moduli and their higher-order normal derivatives were obtained in [12 ] . For the time-dependent isotropic elastic wave equation, the boundary values and derivatives of the Lamé parameters and density were recovered from the dynamical Dirichlet-to-Neumann map in [5 ] .
For the static inverse boundary value problem in isotropic elasticity, a global uniqueness result under a smallness condition on the gradients of the shear moduli was established by Nakamura and Uhlmann [14 , 16 ] . More precisely, the condition takes the form
∥ ∇ μ j ∥ C m ( Ω ) < ε , j = 1 , 2 , \lVert\nabla\mu_j\rVert_{C^m(\Omega)}<\varepsilon,\qquad j=1,2, ∥ ∇ μ j ∥ C m ( Ω ) < ε , j = 1 , 2 ,
where m m m is sufficiently large and ε > 0 \varepsilon>0 ε > 0 is sufficiently small. Complex geometrical optics methods and matrix-valued Cauchy–Riemann systems for isotropic elasticity were developed by Eskin and Ralston [9 , 10 ] . Uniqueness and Lipschitz stability for piecewise-constant Lamé parameters were established in [3 ] . Further uniqueness results for smooth coefficients under additional structural assumptions were obtained in [4 ] . In the static setting, two of the authors of the present paper established a global uniqueness result for smooth Lamé parameters using physical matrix moment identities and exact relative transport [6 ] . A subsequent work established logarithmic interior stability and sharp estimates for boundary derivatives [7 ] .
Density recovery for elastic systems has been studied under various assumptions on the elastic coefficients and the available boundary or scattering data. For constant Lamé parameters, single-frequency uniqueness for an inhomogeneous scalar density was established using far-field data [11 ] . For a homogeneous isotropic elastic background with an unknown symmetric matrix-valued density, uniqueness from boundary and scattering data was obtained in [1 ] . In the time-harmonic setting, simultaneous uniqueness and Lipschitz stability for the Lamé parameters and density were established in [2 ] for coefficients that are piecewise constant on a prescribed finite partition, under suitable geometric and frequency conditions. For the time-dependent elastic wave equation, global recovery of the Lamé parameters and density under a strictly convex foliation condition was established in [19 ] .
The present work establishes global uniqueness for arbitrary smooth admissible isotropic triples ( λ , μ , ρ ) (\lambda,\mu,\rho) ( λ , μ , ρ ) from the full displacement-to-traction map at one fixed known nonzero frequency. The static matrix-moment and relative-transport framework of [6 , 7 ] provides a methodological starting point, while the fixed-frequency problem requires a separate analysis. Equality of the stationary Dirichlet-to-Neumann maps does not identify the corresponding static maps, and the unknown potential W = ω 2 ρ W = \omega^{2}\rho W = ω 2 ρ together with its gradient ∇ W \nabla W ∇ W enters the four-component reduction. The reduced Green–Betti identity is therefore derived directly from the physical fixed-frequency boundary data, with all potential-dependent terms retained. Three physical polarizations of the rank-three characteristic quotient yield the complete 3 × 3 3 \times3 3 × 3 matrix moment identities. The resulting exact relative transport, its determinant identity, and its holomorphic dependence on the projective null direction lead to finite-type rigidity for the Lamé parameters. A further family of physical elastic solutions then recovers the density.
Main analytical challenges
The fixed-frequency inverse problem requires the Lamé parameters to be determined while the mass density is still unknown. In the time-harmonic system (3) , the density appears through the zeroth-order potential W = ω 2 ρ W = \omega^{2}\rho W = ω 2 ρ . Consequently, the principal-symbol information alone does not separate all three unknown coefficients. The proof addresses four related difficulties.
First, the four-component reduction (63) , obtained from the displacement representation (62) , has principal part Δ I 4 \Delta I_{4} Δ I 4 but contains both W W W and ∇ W \nabla W ∇ W in its zeroth-order coefficient matrix. The Green–Betti formula (75) for this reduced system produces additional boundary terms beyond the pairing of the physical Dirichlet-to-Neumann maps. The matching conditions (84) , which follow from the boundary determination theorem 2.5, eliminate these terms and yield the exact identity (85) . The boundary matching needed for this identity is thus derived from the measured map through the boundary determination theorem.
Second, a complete 3 × 3 3 \times3 3 × 3 matrix of moment identities must be realized by physical elastic solutions. The coefficient matrix in the leading four-component transport equation (151) has the common null vector ( θ , 0 ) T (\theta,0)^{T} ( θ , 0 ) T . Factoring out its span leads to the three-dimensional characteristic quotient described before (186) . The scaled lift (171) provides only two independent leading dual polarizations. A third polarization requires the unscaled lift (172) , for which the error in the unscaled physical lift involves both ∥ r h ∥ H 1 \lVert r_h\rVert_{H^{1}} ∥ r h ∥ H 1 and h − 1 ∥ r h ∥ L 2 h^{-1}\lVert r_h\rVert_{L^{2}} h − 1 ∥ r h ∥ L 2 . The same remainder is therefore constructed so that
∥ r h ∥ L 2 ( Ω 0 ) ≤ C h 2 , ∥ r h ∥ H 1 ( Ω 0 ) ≤ C h , \lVert r_h\rVert_{L^{2}(\Omega_0)} \le Ch^2,\qquad\lVert r_h\rVert_{H^{1}(\Omega_0)} \le Ch, ∥ r h ∥ L 2 ( Ω 0 ) ≤ C h 2 , ∥ r h ∥ H 1 ( Ω 0 ) ≤ C h ,
as in (153) . The transverse cutoff correction (187) is included without changing the prescribed leading frame near the support of the transverse test. The resulting quotient frames are selected before the tests, leading to the full matrix identities (189) . The potential and potential-gradient differences enter only through the zeroth-order matrix in (63) . Under the semiclassical scaling used to extract the matrix moments, their contributions carry an additional factor h h h and vanish in the leading-order limit.
Third, the matrix moments must determine a relative transport that is independent of local frame choices and normalized on a common exterior region. The Cauchy transform formula (199) produces such a matrix H H H without assuming invertibility. The transport equation and exterior normalization imply the determinant identity
det H = μ 2 / μ 1 > 0 , \det H = \sqrt{\mu_2/\mu_1} > 0, det H = μ 2 / μ 1 > 0 ,
proved in (202) . Holomorphic dependence on the projective null direction is established with the physical point x x x held fixed; differentiating in coordinates attached to a moving characteristic plane would not suffice. The resulting holomorphic block structure and transport equations yield the finite-type system (212) , which determines μ \mu μ and then λ \lambda λ throughout the admissible coefficient class.
Finally, after the Lamé parameters have been determined, the physical Green identity (216) isolates W 1 − W 2 W_1-W_2 W 1 − W 2 . A further family of exact elastic solutions is needed to obtain the product limit (219) . Using transverse tests to remove the resulting weight, Fourier uniqueness then gives W 1 = W 2 W_1=W_2 W 1 = W 2 and hence ρ 1 = ρ 2 \rho_1=\rho_2 ρ 1 = ρ 2 .
The proof strategy
Boundary determination first gives the complete Taylor data of λ \lambda λ , μ \mu μ , and W W W at ∂ Ω \partial\Omega ∂ Ω ; see Theorem 2.5 . The coefficients can therefore be extended smoothly so that the two triples agree outside Ω \Omega Ω , with constant Lamé parameters and vanishing potential sufficiently far away. The extensions are used to construct solutions, whereas the integral identity (85) continues to involve only the original physical Dirichlet-to-Neumann maps. In particular, the boundary values of W W W and the first normal derivative of μ \mu μ provide the matching conditions that cancel the additional boundary contributions.
On the extended domain, invertible transport frames are constructed on prescribed bounded characteristic disks, with smooth dependence on local direction and transverse parameters. The localized right inverse of Proposition 5.6 satisfies
h − 1 ∥ R h , θ ( 0 ) F ∥ L 2 ( Ω 0 ) + ∥ R h , θ ( 0 ) F ∥ H 1 ( Ω 0 ) ≤ C ∥ F ∥ L 2 ( Ω 0 ) . h^{-1}\lVert\mathcal{R}^{(0)}_{h,\theta}F\rVert_{L^2(\Omega_0)}+\lVert\mathcal{R}^{(0)}_{h,\theta}F\rVert_{H^1(\Omega_0)}\le C\lVert F\rVert_{L^2(\Omega_0)}. h − 1 ∥ R h , θ ( 0 ) F ∥ L 2 ( Ω 0 ) + ∥ R h , θ ( 0 ) F ∥ H 1 ( Ω 0 ) ≤ C ∥ F ∥ L 2 ( Ω 0 ) .
After the first transport correction, the simultaneous remainder bounds (153) allow all three dual polarizations to be realized by physical elastic solutions. The frames are selected independently of the transverse and polynomial tests, and Theorem 6.13 gives the full matrix moment identities (189) .
The moment identities yield a compactly supported Cauchy primitive of the coefficient-difference matrix on each characteristic plane. The formula (199) defines the relative transport H H H , and the determinant identity (202) proves its invertibility. Its normalization on the common exterior region and holomorphic dependence on projective null directions permit the block reduction leading to (212) . Uniqueness for the resulting homogeneous first-order system, together with the exterior agreement, determines μ \mu μ and then λ \lambda λ .
With the Lamé parameters known, the physical integral identity (216) and the product limit (219) recover ( W 1 − W 2 ) / μ (W_1-W_2)/\mu ( W 1 − W 2 ) / μ through its Fourier transform. The longitudinal parts of the leading elastic fields are retained in the product calculation; their cross terms have a vanishing contribution for the selected null phases. A finite family of transverse test functions removes the localization weight, yielding W 1 = W 2 W_1=W_2 W 1 = W 2 and hence ρ 1 = ρ 2 \rho_1=\rho_2 ρ 1 = ρ 2 because ω > 0 \omega>0 ω > 0 is known.
Conventions
All physical dot products, matrix transposes, and Green pairings are complex bilinear. Hermitian products are used for Hilbert-space estimates. The symbol C C C in estimates denotes a constant depending on fixed coefficients, domains, cutoffs, and chosen transport frames; it may change from line to line. A frame denoted by C C C is always a matrix-valued function. Uniform estimates over material classes are stated under the finite-norm and extension hypotheses of Corollary 5.9 .
Well-posedness and the Dirichlet-to-Neumann map
Hilbert-space arguments in this subsection use Hermitian pairings, whereas the Green identities elsewhere use complex bilinear pairings, as specified.
By Korn’s and Poincaré’s inequalities,
∥ v ∥ H 1 ( Ω ) ≤ C Ω ∥ ε ( v ) ∥ L 2 ( Ω ) , v ∈ H 0 1 ( Ω ; C 3 ) . \|v\|_{H^1(\Omega)} \le C_{\Omega}\|\varepsilon(v)\|_{L^2(\Omega)}, \qquad v \in H^1_0(\Omega;\mathbb{C}^3). ∥ v ∥ H 1 ( Ω ) ≤ C Ω ∥ ε ( v ) ∥ L 2 ( Ω ) , v ∈ H 0 1 ( Ω ; C 3 ) .
We now establish the well-posedness of the Dirichlet problem (7) . For brevity, write L = L λ , μ L=L_{\lambda,\mu} L = L λ , μ and P = L + W I 3 P=L+WI_3 P = L + W I 3 , and set P D : = P D λ , μ , W P_D:=P_D^{\lambda,\mu,W} P D := P D λ , μ , W , with domain and codomain as specified before (6) . Set
m ∗ = min Ω ‾ min { 2 μ , 3 λ + 2 μ } > 0. m_*=\min_{\overline{\Omega}}\min\{2\mu,3\lambda+2\mu\}>0. m ∗ = Ω min min { 2 μ , 3 λ + 2 μ } > 0.
The associated Hermitian sesquilinear form
a ( u , v ) : = ∫ Ω ( 2 μ ε ( u ) : ε ( v ) ‾ + λ div u div v ‾ ) d x = ⟨ − L u , v ⟩ a(u,v):=\int_{\Omega}\left(2\mu\varepsilon(u):\overline{\varepsilon(v)}+\lambda\operatorname{div}u\overline{\operatorname{div}v}\right)\,dx=\langle-Lu,v\rangle a ( u , v ) := ∫ Ω ( 2 μ ε ( u ) : ε ( v ) + λ div u div v ) d x = ⟨ − Lu , v ⟩
is bounded on H 0 1 ( Ω ; C 3 ) H^1_0(\Omega;\mathbb{C}^3) H 0 1 ( Ω ; C 3 ) . Here A : B = ∑ i , j = 1 3 A i j B i j A:B=\sum_{i,j=1}^3A_{ij}B_{ij} A : B = ∑ i , j = 1 3 A ij B ij denotes the bilinear contraction of 3 × 3 3\times3 3 × 3 matrices, without complex conjugation. In the following decomposition, dev A = A − 1 3 ( tr A ) I 3 \operatorname{dev}A=A-\frac{1}{3}(\operatorname{tr}A)I_3 dev A = A − 3 1 ( tr A ) I 3 denotes the deviatoric part of A A A . Moreover,
a ( v , v ) = ∫ Ω ( 2 μ ∣ ε ( v ) ∣ 2 + λ ∣ div v ∣ 2 ) d x = ∫ Ω ( 2 μ ∣ dev ε ( v ) ∣ 2 + ( λ + 2 μ / 3 ) ∣ div v ∣ 2 ) d x ≥ m ∗ ∥ ε ( v ) ∥ L 2 ( Ω ) 2 ≥ c Ω ∥ v ∥ H 1 ( Ω ) 2 . \begin{aligned}
a(v,v)&=\int_{\Omega}\left(2\mu|\varepsilon(v)|^2+\lambda|\operatorname{div}v|^2\right)\,dx\\
&=\int_{\Omega}\left(2\mu|\operatorname{dev}\varepsilon(v)|^2+\left(\lambda+2\mu/3\right)|\operatorname{div}v|^2\right)\,dx\\
&\ge m_*\|\varepsilon(v)\|_{L^2(\Omega)}^2\ge c_{\Omega}\|v\|_{H^1(\Omega)}^2.
\end{aligned} a ( v , v ) = ∫ Ω ( 2 μ ∣ ε ( v ) ∣ 2 + λ ∣ div v ∣ 2 ) d x = ∫ Ω ( 2 μ ∣ dev ε ( v ) ∣ 2 + ( λ + 2 μ /3 ) ∣ div v ∣ 2 ) d x ≥ m ∗ ∥ ε ( v ) ∥ L 2 ( Ω ) 2 ≥ c Ω ∥ v ∥ H 1 ( Ω ) 2 .
for v ∈ H 0 1 ( Ω ; C 3 ) v\in H^1_0(\Omega;\mathbb{C}^3) v ∈ H 0 1 ( Ω ; C 3 ) , with c Ω > 0 c_{\Omega}>0 c Ω > 0 depending on m ∗ m_* m ∗ and Ω \Omega Ω . By the Lax–Milgram theorem, − L -L − L , and hence L L L , is a bounded isomorphism from H 0 1 ( Ω ; C 3 ) H^1_0(\Omega;\mathbb{C}^3) H 0 1 ( Ω ; C 3 ) onto H − 1 ( Ω ; C 3 ) H^{-1}(\Omega;\mathbb{C}^3) H − 1 ( Ω ; C 3 ) . Multiplication by W ∈ L ∞ ( Ω ) W\in L^\infty(\Omega) W ∈ L ∞ ( Ω ) defines a compact operator from H 0 1 ( Ω ; C 3 ) H^1_0(\Omega;\mathbb{C}^3) H 0 1 ( Ω ; C 3 ) to H − 1 ( Ω ; C 3 ) H^{-1}(\Omega;\mathbb{C}^3) H − 1 ( Ω ; C 3 ) , since it factors through the compact embedding into L 2 L^2 L 2 . Thus,
P D = L ( I + L − 1 W ) P_D=L(I+L^{-1}W) P D = L ( I + L − 1 W )
is Fredholm of index zero. By the nonresonance condition (6) , we have ker P D = { 0 } \ker P_D=\{0\} ker P D = { 0 } . Consequently, P D P_D P D is a bounded isomorphism. For boundary data f ∈ H 1 / 2 ( ∂ Ω ; C 3 ) f\in H^{1/2}(\partial\Omega;\mathbb{C}^3) f ∈ H 1/2 ( ∂ Ω ; C 3 ) , choose a lifting v ∈ H 1 ( Ω ; C 3 ) v\in H^1(\Omega;\mathbb{C}^3) v ∈ H 1 ( Ω ; C 3 ) satisfying
γ 0 v = f , ∥ v ∥ H 1 ( Ω ) ≤ C ∥ f ∥ H 1 / 2 ( ∂ Ω ) . \gamma_0v=f,\qquad\|v\|_{H^1(\Omega)}\le C\|f\|_{H^{1/2}(\partial\Omega)}. γ 0 v = f , ∥ v ∥ H 1 ( Ω ) ≤ C ∥ f ∥ H 1/2 ( ∂ Ω ) .
Set
u = v − P D − 1 ( P v ) . u=v-P_D^{-1}(Pv). u = v − P D − 1 ( P v ) .
Then u u u is the unique weak solution of
P u = 0 in Ω , γ 0 u = f on ∂ Ω , Pu=0\quad\text{in }\Omega,\qquad\gamma_0u=f\quad\text{on }\partial\Omega, P u = 0 in Ω , γ 0 u = f on ∂ Ω ,
and satisfies
∥ u ∥ H 1 ( Ω ) ≤ C ∥ f ∥ H 1 / 2 ( ∂ Ω ) . \|u\|_{H^1(\Omega)}\le C\|f\|_{H^{1/2}(\partial\Omega)}. ∥ u ∥ H 1 ( Ω ) ≤ C ∥ f ∥ H 1/2 ( ∂ Ω ) .
The independence of u u u from the choice of lifting v v v follows from uniqueness. For g ∈ H 1 / 2 ( ∂ Ω ; C 3 ) g\in H^{1/2}(\partial\Omega;\mathbb{C}^3) g ∈ H 1/2 ( ∂ Ω ; C 3 ) , choose a lifting v g ∈ H 1 ( Ω ; C 3 ) v_g\in H^1(\Omega;\mathbb{C}^3) v g ∈ H 1 ( Ω ; C 3 ) such that
γ 0 v g = g , ∥ v g ∥ H 1 ( Ω ) ≤ C ∥ g ∥ H 1 / 2 ( ∂ Ω ) . \gamma_0v_g=g,\qquad\|v_g\|_{H^1(\Omega)}\le C\|g\|_{H^{1/2}(\partial\Omega)}. γ 0 v g = g , ∥ v g ∥ H 1 ( Ω ) ≤ C ∥ g ∥ H 1/2 ( ∂ Ω ) .
The map Λ \Lambda Λ defined in (8) admits the weak representation
⟨ Λ f , g ⟩ : = ∫ Ω [ λ div u div v g + 2 μ ε ( u ) : ε ( v g ) − W u ⋅ v g ] d x , (11) \langle\Lambda f,g\rangle:= \int_{\Omega} \left[\lambda\operatorname{div}u\operatorname{div}v_g + 2\mu\varepsilon(u):\varepsilon(v_g)-Wu\cdot v_g\right]\,dx, \tag*{(11)} ⟨ Λ f , g ⟩ := ∫ Ω [ λ div u div v g + 2 μ ε ( u ) : ε ( v g ) − W u ⋅ v g ] d x , (11)
where u u u is the unique weak solution corresponding to the boundary data f f f .
The weak formulation shows that this definition is independent of the choice of the lifting v g v_g v g . Moreover, the boundedness of the coefficients, together with the stability estimate for u u u and the lifting estimate for v g v_g v g , yields
∣ ⟨ Λ f , g ⟩ ∣ ≤ C ∥ f ∥ H 1 / 2 ( ∂ Ω ) ∥ g ∥ H 1 / 2 ( ∂ Ω ) . \left|\langle\Lambda f,g\rangle\right| \leq C\|f\|_{H^{1/2}(\partial\Omega)}\|g\|_{H^{1/2}(\partial\Omega)}. ∣ ⟨ Λ f , g ⟩ ∣ ≤ C ∥ f ∥ H 1/2 ( ∂ Ω ) ∥ g ∥ H 1/2 ( ∂ Ω ) .
Consequently,
Λ : H 1 / 2 ( ∂ Ω ; C 3 ) ⟶ H − 1 / 2 ( ∂ Ω ; C 3 ) \Lambda:H^{1/2}(\partial\Omega;\mathbb{C}^{3})\longrightarrow H^{-1/2}(\partial\Omega;\mathbb{C}^{3}) Λ : H 1/2 ( ∂ Ω ; C 3 ) ⟶ H − 1/2 ( ∂ Ω ; C 3 )
is bounded, consistently with the mapping property in (8) . These arguments require the nonresonance condition (6) only on the original physical domain Ω \Omega Ω .
Organization of the paper
Section 2 proves boundary determination and constructs the common extension. Section 3 derives the reduced system and its integral identity from the physical boundary map. Sections 4 and 5 construct transport frames and complex geometrical optics solutions. Section 6 proves the full matrix moment identities and identifies the freedom in choosing the third physical column. Section 7 establishes the independence of the moment construction from solution frames, constructs the relative transport matrix, and proves rigidity of the Lamé parameters. Section 8 recovers the density and completes the proof of Theorem 1.1. The appendices contain the detailed finite-order boundary layer estimates and the free inverse and Fourier operator bounds used in the solution construction.
Boundary determination and a common exterior extension
To prepare the coefficients for the subsequent CGO construction, we establish full boundary agreement and construct common exterior extensions. Theorem 2.5 shows that equality of the physical Dirichlet-to-Neumann maps defined in (8) implies equality of all boundary Taylor coefficients of λ \lambda λ , μ \mu μ , and W W W . The proof uses boundary-concentrating solutions constructed to arbitrary finite order. The lower-order boundary identities determine the boundary values and first normal derivatives of the Lamé parameters. At each elastic order j ≥ 2 j\geq2 j ≥ 2 , the leading boundary moment yields a finite-dimensional linear system that determines the j j j th normal derivatives of λ \lambda λ and μ \mu μ together with the ( j − 2 ) (j-2) ( j − 2 ) th normal derivative of W W W . Proposition 2.7 uses the resulting flatness of the coefficient differences at ∂ Ω \partial\Omega ∂ Ω to construct smooth extensions that agree on R 3 ∖ Ω ‾ \mathbb{R}^{3}\setminus\overline{\Omega} R 3 ∖ Ω . The extensions preserve strong convexity and coincide with a common constant Lamé pair and vanishing potentials outside a compact set.
Let Ω ⊂ R 3 \Omega\subset\mathbb{R}^{3} Ω ⊂ R 3 be a bounded domain with C ∞ C^{\infty} C ∞ boundary, possibly with several boundary components. The coefficients λ , μ , W \lambda,\mu,W λ , μ , W are real-valued and smooth on Ω ‾ \overline{\Omega} Ω , and the Lamé parameters satisfy the strong convexity conditions (4) . For a fixed coefficient triple, write
P : = L λ , μ + W I 3 , P D : = P D λ , μ , W . P:=L_{\lambda,\mu}+WI_{3},\qquad P_{D}:=P_{D}^{\lambda,\mu,W}. P := L λ , μ + W I 3 , P D := P D λ , μ , W .
When two coefficient triples are under consideration, we use
P j : = L λ j , μ j + W j I 3 , P D , j : = P D λ j , μ j , W j , j = 1 , 2. P_j:=L_{\lambda_j,\mu_j}+W_jI_{3},\qquad P_{D,j}:=P_{D}^{\lambda_j,\mu_j,W_j},\qquad j=1,2. P j := L λ j , μ j + W j I 3 , P D , j := P D λ j , μ j , W j , j = 1 , 2.
The nonresonance condition (6) is imposed on each physical coefficient triple. By the argument in Subsection 1.7, the Dirichlet realization associated with each such triple is a bounded isomorphism. In particular, for a pair of triples, P D , 1 P_{D,1} P D , 1 and P D , 2 P_{D,2} P D , 2 are bounded isomorphisms; for a construction involving one fixed triple, P D P_D P D denotes the corresponding realization.
Displacement components are expressed in a fixed Cartesian basis, and bilinear pairings follow the convention of § 1.6 . The boundary-layer construction and remainder estimates are formulated in the boundary normal coordinates introduced below. Uniqueness for the frozen half-line problem associated with (29) follows from a Hermitian energy identity. For each fixed finite expansion order, the constants in the estimates for a single material may depend on the coefficient seminorms, the fixed geometry and cutoffs, and ∥ P D − 1 ∥ H − 1 → H 0 1 \lVert P_D^{-1}\rVert_{H^{-1}\to H_0^1} ∥ P D − 1 ∥ H − 1 → H 0 1 . In estimates comparing two materials, the dependence on the physical Dirichlet inverses is through ∥ P D , 1 − 1 ∥ H − 1 → H 0 1 \lVert P_{D,1}^{-1}\rVert_{H^{-1}\to H_0^1} ∥ P D , 1 − 1 ∥ H − 1 → H 0 1 and ∥ P D , 2 − 1 ∥ H − 1 → H 0 1 \lVert P_{D,2}^{-1}\rVert_{H^{-1}\to H_0^1} ∥ P D , 2 − 1 ∥ H − 1 → H 0 1 . These constants are independent of the scaling parameter h h h .
Boundary coordinates and the scaled operator
Let Φ ∂ : V → ∂ Ω \Phi_{\partial}:V\to\partial\Omega Φ ∂ : V → ∂ Ω be a smooth boundary parametrization, with V ⊂ R 2 V\subset\mathbb{R}^2 V ⊂ R 2 open and y = ( y 1 , y 2 ) y=(y_1,y_2) y = ( y 1 , y 2 ) the tangential coordinates. Let ν \nu ν denote the outward unit normal to ∂ Ω \partial\Omega ∂ Ω . After restricting V V V , choose r 0 > 0 r_0>0 r 0 > 0 sufficiently small so that the map
X ( y , r ) = Φ ∂ ( y ) − r ν ( Φ ∂ ( y ) ) , ( y , r ) ∈ V × ( − r 0 , r 0 ) , (12) X(y,r)=\Phi_{\partial}(y)-r\nu(\Phi_{\partial}(y)),\qquad(y,r)\in V\times(-r_0,r_0), \tag*{(12)} X ( y , r ) = Φ ∂ ( y ) − r ν ( Φ ∂ ( y )) , ( y , r ) ∈ V × ( − r 0 , r 0 ) , (12)
is a diffeomorphism onto an open neighborhood U U U of Φ ∂ ( V ) \Phi_{\partial}(V) Φ ∂ ( V ) .
The inverse coordinates and the nearest boundary projection associated with (12) satisfy
X − 1 ( x ) = ( y 1 ( x ) , y 2 ( x ) , r ( x ) ) , X^{-1}(x)=(y_1(x),y_2(x),r(x)), X − 1 ( x ) = ( y 1 ( x ) , y 2 ( x ) , r ( x )) ,
π ∂ ( x ) = Φ ∂ ( y 1 ( x ) , y 2 ( x ) ) , \pi_{\partial}(x)=\Phi_{\partial}(y_1(x),y_2(x)), π ∂ ( x ) = Φ ∂ ( y 1 ( x ) , y 2 ( x )) ,
∣ x − π ∂ ( x ) ∣ = dist ( x , ∂ Ω ) = ∣ r ( x ) ∣ , x ∈ U . (13) |x-\pi_{\partial}(x)|=\operatorname{dist}(x,\partial\Omega)=|r(x)|,\qquad x\in U. \tag*{(13)} ∣ x − π ∂ ( x ) ∣ = dist ( x , ∂ Ω ) = ∣ r ( x ) ∣ , x ∈ U . (13)
The normal coordinate is the signed distance
r ( x ) = { dist ( x , ∂ Ω ) , x ∈ U ∩ Ω ‾ , − dist ( x , ∂ Ω ) , x ∈ U ∖ Ω ‾ . (14) r(x)=
\begin{cases}
\operatorname{dist}(x,\partial\Omega), & x\in U\cap\overline{\Omega},\\
-\operatorname{dist}(x,\partial\Omega), & x\in U\setminus\overline{\Omega}.
\end{cases} \tag*{(14)} r ( x ) = { dist ( x , ∂ Ω ) , − dist ( x , ∂ Ω ) , x ∈ U ∩ Ω , x ∈ U ∖ Ω . (14)
The coordinates in (13) , with the sign convention (14) , are boundary normal coordinates; see [17 Section 1] and [8 Section 8] . For the Euclidean metric, the normal geodesics are precisely the straight normal segments in (12) .
For a physical coefficient b b b , adopt the pullback convention and define the coordinate fields associated with (12) by
b ( y , r ) : = ( b ∘ X ) ( y , r ) , b(y,r):=(b\circ X)(y,r), b ( y , r ) := ( b ∘ X ) ( y , r ) ,
n i n : = ( ∇ x r ) ∘ X , w α : = ( ∇ x y α ) ∘ X , α = 1 , 2 , n_{\mathrm{in}}:=(\nabla_x r)\circ X,\qquad w^\alpha:=(\nabla_x y_\alpha)\circ X,\qquad\alpha=1,2, n in := ( ∇ x r ) ∘ X , w α := ( ∇ x y α ) ∘ X , α = 1 , 2 ,
X i : = n i n , i ∂ r + ∑ α = 1 2 w i α ∂ y α , i = 1 , 2 , 3 , \mathcal{X}_i:=n_{\mathrm{in},i}\partial_r+\sum_{\alpha=1}^{2}w_i^\alpha\partial_{y_\alpha},\qquad i=1,2,3, X i := n in , i ∂ r + α = 1 ∑ 2 w i α ∂ y α , i = 1 , 2 , 3 ,
( ∂ x i b ) ∘ X = X i ( b ∘ X ) . (15) (\partial_{x_i}b)\circ X=\mathcal{X}_i(b\circ X). \tag*{(15)} ( ∂ x i b ) ∘ X = X i ( b ∘ X ) . (15)
The signed distance convention (14) and the definitions in (15) imply
n i n ( y , 0 ) = − ν ( Φ ∂ ( y ) ) , n_{\mathrm{in}}(y,0)=-\nu(\Phi_{\partial}(y)), n in ( y , 0 ) = − ν ( Φ ∂ ( y )) ,
∣ n i n ( y , r ) ∣ = 1 , n i n ( y , r ) ⋅ w α ( y , r ) = 0 , α = 1 , 2. (16) |n_{\mathrm{in}}(y,r)|=1,\qquad n_{\mathrm{in}}(y,r)\cdot w^\alpha(y,r)=0,\qquad\alpha=1,2. \tag*{(16)} ∣ n in ( y , r ) ∣ = 1 , n in ( y , r ) ⋅ w α ( y , r ) = 0 , α = 1 , 2. (16)
Thus, the vector n i n n_{\mathrm{in}} n in in (16) is the inward unit normal to the parallel surfaces of constant r r r , and the coordinate gradients w α w^\alpha w α are tangent to the parallel surfaces.
For a fixed compact patch K ⋐ V K\Subset V K ⋐ V and 0 < r 1 < r 0 0<r_1<r_0 0 < r 1 < r 0 , the volume density satisfies
J ( y , r ) : = ∣ det D ( y , r ) X ( y , r ) ∣ , d x = J ( y , r ) d y d r , J(y,r):=\left|\det D_{(y,r)}X(y,r)\right|,\qquad dx=J(y,r)\,dy\,dr, J ( y , r ) := det D ( y , r ) X ( y , r ) , d x = J ( y , r ) d y d r ,
0 < c J ≤ J ( y , r ) ≤ C J < ∞ , ( y , r ) ∈ K × [ − r 1 , r 1 ] . (17) 0<c_J\le J(y,r)\le C_J<\infty,\qquad(y,r)\in K\times[-r_1,r_1]. \tag*{(17)} 0 < c J ≤ J ( y , r ) ≤ C J < ∞ , ( y , r ) ∈ K × [ − r 1 , r 1 ] . (17)
The constants c J c_J c J and C J C_J C J in (17) depend on the fixed coordinate patch and compact collar.
Introduce the oscillation parameter, normal scale, stretched normal coordinate, and phase sign by
k ≥ 1 , h = k − 1 , t = r / h , s ∈ { + 1 , − 1 } . (18) k \ge1,\qquad h=k^{-1},\qquad t=r/h,\qquad s\in\{+1,-1\}. \tag*{(18)} k ≥ 1 , h = k − 1 , t = r / h , s ∈ { + 1 , − 1 } . (18)
The real parameter k k k in (18) controls tangential oscillation in the boundary data, h h h is the scale of normal localization, and s s s determines the sign of the tangential phase. The measurement frequency ω \omega ω remains fixed as k → ∞ k\to\infty k → ∞ .
For a smooth vector profile U ( y , t ) U(y,t) U ( y , t ) , the chain rule in (15) gives
e − i s y 1 / h χ i ( e i s y 1 / h U ( y , r / h ) ) ∣ r = h t = h − 1 ( D i , h U ) ( y , t ) , e^{-isy_1/h}\chi_i\left(e^{isy_1/h}U(y,r/h)\right)\bigg|_{r=ht}=h^{-1}(\mathcal{D}_{i,h}U)(y,t), e − i s y 1 / h χ i ( e i s y 1 / h U ( y , r / h ) ) r = h t = h − 1 ( D i , h U ) ( y , t ) ,
D i , h : = n i n , i ( y , h t ) ∂ t + i s w i 1 ( y , h t ) + h ∑ α = 1 2 w i α ( y , h t ) ∂ y α , i = 1 , 2 , 3. (19) \mathcal{D}_{i,h}:=n_{\mathrm{in},i}(y,ht)\partial_t+isw_i^1(y,ht)+h\sum_{\alpha=1}^{2}w_i^\alpha(y,ht)\partial_{y_\alpha},\qquad i=1,2,3. \tag*{(19)} D i , h := n in , i ( y , h t ) ∂ t + i s w i 1 ( y , h t ) + h α = 1 ∑ 2 w i α ( y , h t ) ∂ y α , i = 1 , 2 , 3. (19)
The conjugation formula (19) extracts a factor h − 1 h^{-1} h − 1 from each Cartesian derivative. The second derivatives in the elasticity operator (2) therefore contribute a factor h − 2 h^{-2} h − 2 . The order of multiplication and differentiation is preserved in all compositions, retaining all derivatives of the variable coordinate coefficients.
All coefficients in the following identity are evaluated at ( y , h t ) (y,ht) ( y , h t ) . The ordered product rule gives
D i , h D j , h = n i n , i n i n , j ∂ t 2 + h ∑ α ( n i n , i w j α + w i α n i n , j ) ∂ t ∂ y α + h 2 ∑ α , β w i α w j β ∂ y α ∂ y β + [ i s ( n i n , i w j 1 + w i 1 n i n , j ) + h χ i ( n i n , j ) ] ∂ t + ∑ β [ i s h ( w i 1 w j β + w i β w j 1 ) + h 2 χ i ( w j β ) ] ∂ y β + [ − w i 1 w j 1 + i s h χ i ( w j 1 ) ] . (20) \begin{aligned}
\mathcal{D}_{i,h}\mathcal{D}_{j,h}
={}&n_{\mathrm{in},i}n_{\mathrm{in},j}\partial_t^2
+h\sum_\alpha(n_{\mathrm{in},i}w_j^\alpha+w_i^\alpha n_{\mathrm{in},j})\partial_t\partial_{y_\alpha}
+h^2\sum_{\alpha,\beta}w_i^\alpha w_j^\beta\partial_{y_\alpha}\partial_{y_\beta}\\
&+\left[is(n_{\mathrm{in},i}w_j^1+w_i^1n_{\mathrm{in},j})+h\chi_i(n_{\mathrm{in},j})\right]\partial_t\\
&+\sum_\beta\left[ish(w_i^1w_j^\beta+w_i^\beta w_j^1)+h^2\chi_i(w_j^\beta)\right]\partial_{y_\beta}
+\left[-w_i^1w_j^1+ish\chi_i(w_j^1)\right]. \tag*{(20)}
\end{aligned} D i , h D j , h = n in , i n in , j ∂ t 2 + h α ∑ ( n in , i w j α + w i α n in , j ) ∂ t ∂ y α + h 2 α , β ∑ w i α w j β ∂ y α ∂ y β + [ i s ( n in , i w j 1 + w i 1 n in , j ) + h χ i ( n in , j ) ] ∂ t + β ∑ [ i s h ( w i 1 w j β + w i β w j 1 ) + h 2 χ i ( w j β ) ] ∂ y β + [ − w i 1 w j 1 + i s h χ i ( w j 1 ) ] . (20)
Here χ i ( n i n , j ) \chi_i(n_{\mathrm{in},j}) χ i ( n in , j ) and χ i ( w j α ) \chi_i(w_j^\alpha) χ i ( w j α ) are derivatives of coordinate coefficients, with
χ i ( n i n , j ) = n i n , i ∂ r n i n , j + ∑ α = 1 2 w i α ∂ y α n i n , j = ( ∂ x i ∂ x j r ) ∘ X , \chi_i(n_{\mathrm{in},j})=n_{\mathrm{in},i}\partial_r n_{\mathrm{in},j}+\sum_{\alpha=1}^{2}w_i^\alpha\partial_{y_\alpha}n_{\mathrm{in},j}=(\partial_{x_i}\partial_{x_j}r)\circ X, χ i ( n in , j ) = n in , i ∂ r n in , j + α = 1 ∑ 2 w i α ∂ y α n in , j = ( ∂ x i ∂ x j r ) ∘ X ,
χ i ( w j α ) = n i n , i ∂ r w j α + ∑ β = 1 2 w i β ∂ y β w j α = ( ∂ x i ∂ x j y α ) ∘ X . (21) \chi_i(w_j^\alpha)=n_{\mathrm{in},i}\partial_r w_j^\alpha+\sum_{\beta=1}^{2}w_i^\beta\partial_{y_\beta}w_j^\alpha=(\partial_{x_i}\partial_{x_j}y_\alpha)\circ X. \tag*{(21)} χ i ( w j α ) = n in , i ∂ r w j α + β = 1 ∑ 2 w i β ∂ y β w j α = ( ∂ x i ∂ x j y α ) ∘ X . (21)
They arise from the product rule in (20) ; χ i ( n i n , j ) \chi_i(n_{\mathrm{in},j}) χ i ( n in , j ) denotes the action of the vector field, whereas n i n , i n i n , j n_{\mathrm{in},i}n_{\mathrm{in},j} n in , i n in , j denotes multiplication. These terms retain the variation of the normal and tangential coordinates along the curved boundary.
In Cartesian components, the physical operator is
( P u ) i = μ ∑ j ∂ x j 2 u i + ( λ + μ ) ∑ j ∂ x i ∂ x j u j + ( χ i ( λ ) ) ∑ j ∂ x j u j + ∑ j ( χ j ( μ ) ) ( ∂ x j u i + ∂ x i u j ) + W u i . (Pu)_i=\mu\sum_j\partial_{x_j}^2u_i+(\lambda+\mu)\sum_j\partial_{x_i}\partial_{x_j}u_j+(\chi_i(\lambda))\sum_j\partial_{x_j}u_j+\sum_j(\chi_j(\mu))(\partial_{x_j}u_i+\partial_{x_i}u_j)+Wu_i. ( P u ) i = μ j ∑ ∂ x j 2 u i + ( λ + μ ) j ∑ ∂ x i ∂ x j u j + ( χ i ( λ )) j ∑ ∂ x j u j + j ∑ ( χ j ( μ )) ( ∂ x j u i + ∂ x i u j ) + W u i .
To express the physical operator in the boundary-layer variables, we define the scaled operator P s , h \mathcal{P}_{s,h} P s , h by conjugating P P P with the oscillatory phase and rescaling the normal variable. For a smooth profile U = U ( y , t ) U=U(y,t) U = U ( y , t ) , define
( P s , h U ) ( y , t ) : = h 2 e − i s y 1 / h P ( e i s y 1 ( x ) / h U ( y ( x ) , r ( x ) h ) ) ∣ x = X ( y , h t ) . (\mathcal{P}_{s,h}U)(y,t):=h^2e^{-isy_1/h}P\left(e^{isy_1(x)/h}U\left(y(x),\frac{r(x)}{h}\right)\right)\bigg|_{x=X(y,ht)}. ( P s , h U ) ( y , t ) := h 2 e − i s y 1 / h P ( e i s y 1 ( x ) / h U ( y ( x ) , h r ( x ) ) ) x = X ( y , h t ) .
Here y ( x ) y(x) y ( x ) and r ( x ) r(x) r ( x ) are the boundary coordinate functions defined in (13) . Thus P P P acts on a function of the physical variable x x x , whereas P s , h \mathcal{P}_{s,h} P s , h acts on the profile U U U in the variables ( y , t ) (y,t) ( y , t ) . By definition,
P ( e i s y 1 ( x ) / h U ( y ( x ) , r ( x ) h ) ∣ x = X ( y , h t ) ) = h − 2 e i s y 1 / h ( P s , h U ) ( y , t ) . P\left(\left.e^{\mathrm{i}s y_{1}(x)/h}U\left(y(x),\frac{r(x)}{h}\right)\right|_{x=X(y,ht)}\right)=h^{-2}e^{\mathrm{i}s y_{1}/h}(\mathcal{P}_{s,h}U)(y,t). P ( e i s y 1 ( x ) / h U ( y ( x ) , h r ( x ) ) x = X ( y , h t ) ) = h − 2 e i s y 1 / h ( P s , h U ) ( y , t ) .
Using (19) , each physical derivative acting on the oscillatory profile can be written as h − 1 D i , h h^{-1}\mathcal{D}_{i,h} h − 1 D i , h after the phase is removed. The slow tangential derivatives are included in D i , h \mathcal{D}_{i,h} D i , h with an additional factor h h h . Substituting this identity into the Cartesian expression for P P P , and preserving the order of differentiation, gives the explicit formula
( P s , h U ) i = μ ∑ j D j , h D j , h U i + ( λ + μ ) ∑ j D i , h D j , h U j + h ( X i ( λ ) ) ∑ j D j , h U j + h ∑ j ( X j ( μ ) ) ( D j , h U i + D i , h U j ) + h 2 W U i . (22) (\mathcal{P}_{s,h}U)_i=\mu\sum_j\mathcal{D}_{j,h}\mathcal{D}_{j,h}U_i+(\lambda+\mu)\sum_j\mathcal{D}_{i,h}\mathcal{D}_{j,h}U_j+h(\mathcal{X}_i(\lambda))\sum_j\mathcal{D}_{j,h}U_j+h\sum_j(\mathcal{X}_j(\mu))(\mathcal{D}_{j,h}U_i+\mathcal{D}_{i,h}U_j)+h^2WU_i. \tag*{(22)} ( P s , h U ) i = μ j ∑ D j , h D j , h U i + ( λ + μ ) j ∑ D i , h D j , h U j + h ( X i ( λ )) j ∑ D j , h U j + h j ∑ ( X j ( μ )) ( D j , h U i + D i , h U j ) + h 2 W U i . (22)
All coefficients and their coordinate derivatives in this formula are evaluated at ( y , h t ) (y,ht) ( y , h t ) .
To describe the coefficients of the scaled operator, we write it in normal form, with all derivatives acting on the profile U U U placed to the right of their coefficient matrices. Expanding (22) by means of (20) , applying the product rule, and collecting equal powers of h h h defines the coefficient matrices A m , β ( y , r ) A_{m,\beta}(y,r) A m , β ( y , r ) uniquely through
P s , h = ∑ m = 0 2 ∑ ∣ β ∣ ≤ 2 h m A m , β ( y , h t ) ∂ ( y , t ) β . (23) \mathcal{P}_{s,h}=\sum_{m=0}^{2}\sum_{|\beta|\leq2}h^m A_{m,\beta}(y,ht)\partial_{(y,t)}^\beta. \tag*{(23)} P s , h = m = 0 ∑ 2 ∣ β ∣ ≤ 2 ∑ h m A m , β ( y , h t ) ∂ ( y , t ) β . (23)
Here m m m records the explicit power of h h h , β = ( β 1 , β 2 , β 3 ) ∈ N 0 3 \beta=(\beta_1,\beta_2,\beta_3)\in\mathbb{N}_0^3 β = ( β 1 , β 2 , β 3 ) ∈ N 0 3 , ∣ β ∣ = β 1 + β 2 + β 3 |\beta|=\beta_1+\beta_2+\beta_3 ∣ β ∣ = β 1 + β 2 + β 3 , with N 0 = { 0 , 1 , 2 , … } \mathbb{N}_0=\{0,1,2,\ldots\} N 0 = { 0 , 1 , 2 , … } , and
∂ ( y , t ) β = ∂ y 1 β 1 ∂ y 2 β 2 ∂ t β 3 . \partial_{(y,t)}^\beta=\partial_{y_1}^{\beta_1}\partial_{y_2}^{\beta_2}\partial_t^{\beta_3}. ∂ ( y , t ) β = ∂ y 1 β 1 ∂ y 2 β 2 ∂ t β 3 .
In (23) , each derivative acts only on U U U ; the coefficient matrices already include all terms produced when derivatives fall on the coefficients of the operators in (22) . Any coefficient that does not occur in the expansion is understood to be zero. The slow tangential order satisfies β 1 + β 2 ≤ m \beta_1+\beta_2\leq m β 1 + β 2 ≤ m whenever A m , β ≠ 0 A_{m,\beta}\ne0 A m , β = 0 , because each slow tangential derivative in (19) carries a factor h h h .
The matrices A m , β A_{m,\beta} A m , β are smooth 3 × 3 3\times3 3 × 3 matrix-valued functions formed from λ \lambda λ , μ \mu μ , W W W , n i n n_{\mathrm{in}} n in , w w w and their spatial derivatives of order at most one. Their dependence on the fixed phase sign s s s is suppressed. To expand P s , h \mathcal{P}_{s,h} P s , h in powers of h h h , we Taylor-expand each coefficient A m , β ( y , h t ) A_{m,\beta}(y,ht) A m , β ( y , h t ) at r = 0 r=0 r = 0 . The explicit factor h m h^m h m in (23) combines with the Taylor factor h p h^p h p , so the coefficient of h q h^q h q is obtained by setting p = q − m p=q-m p = q − m . We denote this coefficient operator by P s , q \mathcal{P}_{s,q} P s , q :
P s , q = ∑ m = 0 min ( 2 , q ) ∑ ∣ β ∣ ≤ 2 t q − m ( q − m ) ! ∂ r q − m A m , β ( y , 0 ) ∂ ( y , t ) β , q ≥ 0. (24) \mathcal{P}_{s,q}=\sum_{m=0}^{\min(2,q)}\sum_{|\beta|\leq2}\frac{t^{q-m}}{(q-m)!}\partial_r^{q-m}A_{m,\beta}(y,0)\partial_{(y,t)}^\beta,\qquad q\geq0. \tag*{(24)} P s , q = m = 0 ∑ m i n ( 2 , q ) ∣ β ∣ ≤ 2 ∑ ( q − m )! t q − m ∂ r q − m A m , β ( y , 0 ) ∂ ( y , t ) β , q ≥ 0. (24)
By construction, P s , q \mathcal{P}_{s,q} P s , q is the coefficient of h q h^q h q in the Taylor expansion of P s , h \mathcal{P}_{s,h} P s , h . In particular, P s , 0 \mathcal{P}_{s,0} P s , 0 is the frozen normal operator, while P s , q \mathcal{P}_{s,q} P s , q for q ≥ 1 q\geq1 q ≥ 1 gives the contribution at order h q h^q h q . These operators determine the successive profiles through the recursion (36) . The zeroth-order physical potential W W W first enters the scaled operator at order h 2 h^2 h 2 , as seen in (22) .
We now define the remainder operator explicitly. For a smooth scalar- or matrix-valued coefficient b = b ( y , r ) b=b(y,r) b = b ( y , r ) and an integer N ≥ 0 N\geq0 N ≥ 0 , let
b ( y , h t ) = ∑ q = 0 N h q t q q ! ∂ r q b ( y , 0 ) + h N + 1 R N , h [ b ] ( y , t ) , b(y,ht)=\sum_{q=0}^{N}\frac{h^qt^q}{q!}\partial_r^qb(y,0)+h^{N+1}R_{N,h}[b](y,t), b ( y , h t ) = q = 0 ∑ N q ! h q t q ∂ r q b ( y , 0 ) + h N + 1 R N , h [ b ] ( y , t ) ,
R N , h [ b ] ( y , t ) = t N + 1 N ! ∫ 0 1 ( 1 − s ) N ∂ r N + 1 b ( y , s h t ) d s . (25) R_{N,h}[b](y,t)=\frac{t^{N+1}}{N!}\int_0^1(1-s)^N\partial_r^{N+1}b(y,sht)\,ds. \tag*{(25)} R N , h [ b ] ( y , t ) = N ! t N + 1 ∫ 0 1 ( 1 − s ) N ∂ r N + 1 b ( y , s h t ) d s . (25)
The second line defines the normalized integral Taylor remainder, entry by entry for matrices. On 0 ≤ h t ≤ r 0 0 \le ht \le r_0 0 ≤ h t ≤ r 0 , for each finite S ≥ 0 S \ge0 S ≥ 0 ,
∣ ∂ ( y , t ) γ R N , h [ b ] ( y , t ) ∣ ≤ C N , S ( 1 + t ) N + 1 + S , ∣ γ ∣ ≤ S . (26) \left|\partial_{(y,t)}^\gamma R_{N,h}[b](y,t)\right| \le C_{N,S}(1+t)^{N+1+S}, \qquad|\gamma| \le S. \tag*{(26)} ∂ ( y , t ) γ R N , h [ b ] ( y , t ) ≤ C N , S ( 1 + t ) N + 1 + S , ∣ γ ∣ ≤ S . (26)
Indeed, differentiation of the integral in (25) uses only the seminorms of ∂ y α ∂ t p b \partial_y^\alpha\partial_t^p b ∂ y α ∂ t p b with ∣ α ∣ ≤ S |\alpha| \le S ∣ α ∣ ≤ S and p ≤ N + 1 + S p \le N+1+S p ≤ N + 1 + S on the compact collar. Each t t t derivative falling on b ( y , s h t ) b(y,sht) b ( y , s h t ) contributes an additional factor s h sh s h .
For a truncation order N b l ≥ 0 N_{\mathrm{bl}} \ge0 N bl ≥ 0 , define
B m , β , h ( N b l ) ( y , t ) = { R N b l − m , h [ A m , β ] ( y , t ) , 0 ≤ m ≤ min ( 2 , N b l ) , h m − N b l − 1 A m , β ( y , h t ) , N b l < m ≤ 2 , B_{m,\beta,h}^{(N_{\mathrm{bl}})}(y,t) =
\begin{cases}
R_{N_{\mathrm{bl}}-m,h}[A_{m,\beta}](y,t), & 0 \le m \le\min(2,N_{\mathrm{bl}}),\\
h^{m-N_{\mathrm{bl}}-1}A_{m,\beta}(y,ht), & N_{\mathrm{bl}} < m \le2,
\end{cases} B m , β , h ( N bl ) ( y , t ) = { R N bl − m , h [ A m , β ] ( y , t ) , h m − N bl − 1 A m , β ( y , h t ) , 0 ≤ m ≤ min ( 2 , N bl ) , N bl < m ≤ 2 ,
R s , N b l , h b l = ∑ m = 0 2 ∑ ∣ β ∣ ≤ 2 B m , β , h ( N b l ) ( y , t ) ∂ ( y , t ) β . (27) \mathfrak{R}_{s,N_{\mathrm{bl}},h}^{\mathrm{bl}}
= \sum_{m=0}^{2}\sum_{|\beta|\le2} B_{m,\beta,h}^{(N_{\mathrm{bl}})}(y,t)\partial_{(y,t)}^\beta. \tag*{(27)} R s , N bl , h bl = m = 0 ∑ 2 ∣ β ∣ ≤ 2 ∑ B m , β , h ( N bl ) ( y , t ) ∂ ( y , t ) β . (27)
The operator R s , N b l , h b l \mathfrak{R}_{s,N_{\mathrm{bl}},h}^{\mathrm{bl}} R s , N bl , h bl has order at most two and collects the Taylor remainders and all terms of explicit h h h -order greater than N b l N_{\mathrm{bl}} N bl . Substituting (25) into (23) yields the exact identity
P s , h = ∑ q = 0 N b l h q P s , q + h N b l + 1 R s , N b l , h b l . (28) \mathcal{P}_{s,h}=\sum_{q=0}^{N_{\mathrm{bl}}}h^q\mathcal{P}_{s,q}+h^{N_{\mathrm{bl}}+1}\mathfrak{R}_{s,N_{\mathrm{bl}},h}^{\mathrm{bl}}. \tag*{(28)} P s , h = q = 0 ∑ N bl h q P s , q + h N bl + 1 R s , N bl , h bl . (28)
By (26) , the coefficients of R s , N b l , h b l \mathfrak{R}_{s,N_{\mathrm{bl}},h}^{\mathrm{bl}} R s , N bl , h bl and their derivatives of any fixed finite order have at most polynomial growth in t t t , uniformly for 0 < h ≤ 1 0<h\le1 0 < h ≤ 1 and 0 ≤ h t ≤ r 0 0\le ht\le r_0 0 ≤ h t ≤ r 0 . Set τ t a n = w 1 ( y , 0 ) \tau_{\mathrm{tan}}=w^1(y,0) τ tan = w 1 ( y , 0 ) , c b l = ∣ τ t a n ∣ c_{\mathrm{bl}}=|\tau_{\mathrm{tan}}| c bl = ∣ τ tan ∣ , and e b l = τ t a n / c b l e_{\mathrm{bl}}=\tau_{\mathrm{tan}}/c_{\mathrm{bl}} e bl = τ tan / c bl . The tangential phase covector is identified with the vector τ t a n \tau_{\mathrm{tan}} τ tan , and e b l e_{\mathrm{bl}} e bl is its unit tangent direction. Since the profiles constructed below are polynomials in t t t multiplied by e − c b l ( y ) t e^{-c_{\mathrm{bl}}(y)t} e − c bl ( y ) t , with c b l ( y ) ≥ c 0 > 0 c_{\mathrm{bl}}(y)\ge c_0>0 c bl ( y ) ≥ c 0 > 0 , these polynomial factors can be absorbed into e − c t e^{-ct} e − c t for any 0 < c < c 0 0<c<c_0 0 < c < c 0 . This estimate is used to establish the residual bound (33) .
The frozen half-line operator
In the following formulas, n i n n_{\mathrm{in}} n in is evaluated at r = 0 r=0 r = 0 . By (16) , n i n ⋅ τ t a n = 0 n_{\mathrm{in}}\cdot\tau_{\mathrm{tan}}=0 n in ⋅ τ tan = 0 and ∣ n i n ∣ = 1 |n_{\mathrm{in}}|=1 ∣ n in ∣ = 1 , while c b l ≥ c 0 > 0 c_{\mathrm{bl}}\ge c_0>0 c bl ≥ c 0 > 0 on the compact patch. Writing λ b = λ ( y , 0 ) \lambda_b=\lambda(y,0) λ b = λ ( y , 0 ) and μ b = μ ( y , 0 ) \mu_b=\mu(y,0) μ b = μ ( y , 0 ) , the leading operator is
P s , 0 = μ b ( ∂ t 2 − c b l 2 ) I 3 + ( λ b + μ b ) ( i s τ t a n + n i n ∂ t ) ( i s τ t a n + n i n ∂ t ) T . (29) \mathcal{P}_{s,0}=\mu_b(\partial_t^2-c_{\mathrm{bl}}^2)I_3+(\lambda_b+\mu_b)(is\tau_{\mathrm{tan}}+n_{\mathrm{in}}\partial_t)(is\tau_{\mathrm{tan}}+n_{\mathrm{in}}\partial_t)^{T}. \tag*{(29)} P s , 0 = μ b ( ∂ t 2 − c bl 2 ) I 3 + ( λ b + μ b ) ( i s τ tan + n in ∂ t ) ( i s τ tan + n in ∂ t ) T . (29)
The boundary-normalized decaying matrix profile is given by
a b l = λ b + μ b λ b + 3 μ b ∈ ( 1 / 7 , 1 ) , a_{\mathrm{bl}}=\frac{\lambda_b+\mu_b}{\lambda_b+3\mu_b}\in(1/7,1), a bl = λ b + 3 μ b λ b + μ b ∈ ( 1/7 , 1 ) ,
N s = a b l [ e b l ⊗ e b l − n i n ⊗ n i n + i s ( e b l ⊗ n i n + n i n ⊗ e b l ) ] , N_s=a_{\mathrm{bl}}\left[e_{\mathrm{bl}}\otimes e_{\mathrm{bl}}-n_{\mathrm{in}}\otimes n_{\mathrm{in}}+is(e_{\mathrm{bl}}\otimes n_{\mathrm{in}}+n_{\mathrm{in}}\otimes e_{\mathrm{bl}})\right], N s = a bl [ e bl ⊗ e bl − n in ⊗ n in + i s ( e bl ⊗ n in + n in ⊗ e bl ) ] ,
F s ( y , t ) = e − c b l t ( I 3 − c b l t N s ) . (30) F_s(y,t)=e^{-c_{\mathrm{bl}}t}(I_3-c_{\mathrm{bl}}tN_s). \tag*{(30)} F s ( y , t ) = e − c bl t ( I 3 − c bl t N s ) . (30)
Here ( a ⊗ b ) i j = a i b j (a\otimes b)_{ij}=a_i b_j ( a ⊗ b ) ij = a i b j denotes the tensor product of vectors.
Lemma 2.1. Let d p ∈ N 0 d_p\in\mathbb{N}_0 d p ∈ N 0 and
p ( y , t ) = ∑ j = 0 d p p j ( y ) t j , p j ∈ C ∞ ( K ; C 3 ) , p(y,t)=\sum_{j=0}^{d_p}p_j(y)t^j,\qquad p_j\in C^\infty(K;\mathbb{C}^3), p ( y , t ) = j = 0 ∑ d p p j ( y ) t j , p j ∈ C ∞ ( K ; C 3 ) , where K ⋐ V K\Subset V K ⋐ V is the fixed compact parameter patch. For each y ∈ K y\in K y ∈ K , the problem
P s , 0 V ( y , t ) = e − c b l ( y ) t p ( y , t ) , t > 0 , V ( y , 0 ) = 0 , \mathcal{P}_{s,0}V(y,t)=e^{-c_{\mathrm{bl}}(y)t}p(y,t),\qquad t>0,\qquad V(y,0)=0, P s , 0 V ( y , t ) = e − c bl ( y ) t p ( y , t ) , t > 0 , V ( y , 0 ) = 0 , has a unique solution decaying as t → ∞ t\to\infty t → ∞ . Moreover,
V ( y , t ) = e − c b l ( y ) t q ( y , t ) , deg t q ( y , ⋅ ) ≤ d p + 2 , V(y,t)=e^{-c_{\mathrm{bl}}(y)t}q(y,t),\qquad\deg_t q(y,\cdot)\le d_p+2, V ( y , t ) = e − c bl ( y ) t q ( y , t ) , deg t q ( y , ⋅ ) ≤ d p + 2 ,
where q q q is a vector polynomial in t t t with smooth coefficients in y y y . For every multi-index α ∈ N 0 2 \alpha\in\mathbb{N}_0^2 α ∈ N 0 2 and integer j ≥ 0 j \ge0 j ≥ 0 , there exist constants C α , j > 0 C_{\alpha,j} > 0 C α , j > 0 and M α , j ∈ N 0 M_{\alpha,j} \in\mathbb{N}_0 M α , j ∈ N 0 such that
∣ ∂ y α ∂ t j V ( y , t ) ∣ ≤ C α , j ( 1 + t ) M α , j e − c 0 t , y ∈ K , t ≥ 0. \left|\partial_y^\alpha\partial_t^j V(y,t)\right| \le C_{\alpha,j}(1+t)^{M_{\alpha,j}}e^{-c_0t}, \qquad y \in K,\quad t \ge0. ∂ y α ∂ t j V ( y , t ) ≤ C α , j ( 1 + t ) M α , j e − c 0 t , y ∈ K , t ≥ 0.
The constants depend on finitely many seminorms of the data and the frozen coefficients on K K K .
Proof. Regard ∂ t \partial_t ∂ t as a commuting indeterminate D D D , and put d = D 2 − c b l 2 d = D^2-c_{\mathrm{bl}}^2 d = D 2 − c bl 2 and z = i s τ t a n + n i n D z = i s \tau_{\mathrm{tan}}+n_{\mathrm{in}}D z = i s τ tan + n in D . The identity z T z = d z^Tz=d z T z = d gives
det ( μ b d I 3 + ( λ b + μ b ) z z T ) = μ b 2 ( λ b + 2 μ b ) d 3 . \det\left(\mu_{\mathrm{b}}dI_3+(\lambda_{\mathrm{b}}+\mu_{\mathrm{b}})zz^T\right)=\mu_{\mathrm{b}}^2(\lambda_{\mathrm{b}}+2\mu_{\mathrm{b}})d^3. det ( μ b d I 3 + ( λ b + μ b ) z z T ) = μ b 2 ( λ b + 2 μ b ) d 3 .
This polynomial identity holds also at d = 0 d=0 d = 0 . To check the homogeneous profile, set v s = i s e b l − n i n v_s=i s e_{\mathrm{bl}}-n_{\mathrm{in}} v s = i s e bl − n in . Then z s T v s = 0 z_s^Tv_s=0 z s T v s = 0 , N s = − a b l v s v s T N_s=-a_{\mathrm{bl}}v_sv_s^T N s = − a bl v s v s T and n i n T N s = a b l v s T n_{\mathrm{in}}^TN_s=a_{\mathrm{bl}}v_s^T n in T N s = a bl v s T , so N s 2 = 0 N_s^2=0 N s 2 = 0 . Direct differentiation gives
( D 2 − c b l 2 ) F s = 2 c b l 2 e − c b l t N s , (D^2-c_{\mathrm{bl}}^2)F_s=2c_{\mathrm{bl}}^2e^{-c_{\mathrm{bl}}t}N_s, ( D 2 − c bl 2 ) F s = 2 c bl 2 e − c bl t N s ,
z ( D ) z ( D ) T F s = − c b l 2 ( 1 − a b l ) a b l − 1 e − c b l t N s . z(D)z(D)^TF_s=-c_{\mathrm{bl}}^2(1-a_{\mathrm{bl}})a_{\mathrm{bl}}^{-1}e^{-c_{\mathrm{bl}}t}N_s. z ( D ) z ( D ) T F s = − c bl 2 ( 1 − a bl ) a bl − 1 e − c bl t N s .
The identity 2 μ b a b l = ( λ b + μ b ) ( 1 − a b l ) 2\mu_{\mathrm{b}}a_{\mathrm{bl}}=(\lambda_{\mathrm{b}}+\mu_{\mathrm{b}})(1-a_{\mathrm{bl}}) 2 μ b a bl = ( λ b + μ b ) ( 1 − a bl ) gives P s , 0 F s = 0 P_{s,0}F_s=0 P s , 0 F s = 0 . Moreover, F s ( y , 0 ) = I 3 F_s(y,0)=I_3 F s ( y , 0 ) = I 3 .
The polynomial matrix
B s ( D ) = ( λ b + 2 μ b ) ( D 2 − c b l 2 ) I 3 − ( λ b + μ b ) z z T , (31) B_s(D)=(\lambda_{\mathrm{b}}+2\mu_{\mathrm{b}})(D^2-c_{\mathrm{bl}}^2)I_3-(\lambda_{\mathrm{b}}+\mu_{\mathrm{b}})zz^T, \tag*{(31)} B s ( D ) = ( λ b + 2 μ b ) ( D 2 − c bl 2 ) I 3 − ( λ b + μ b ) z z T , (31)
P s , 0 B s ( D ) = μ b ( λ b + 2 μ b ) ( D 2 − c b l 2 ) 2 I 3 P_{s,0}B_s(D)=\mu_{\mathrm{b}}(\lambda_{\mathrm{b}}+2\mu_{\mathrm{b}})(D^2-c_{\mathrm{bl}}^2)^2I_3 P s , 0 B s ( D ) = μ b ( λ b + 2 μ b ) ( D 2 − c bl 2 ) 2 I 3
provides a right inverse using only two scalar integrations. Indeed, ( z z T ) 2 = d z z T (zz^T)^2=dzz^T ( z z T ) 2 = d z z T , so the coefficient of d z z T dzz^T d z z T in the product is
( λ b + μ b ) [ − μ b + ( λ b + 2 μ b ) − ( λ b + μ b ) ] = 0. (\lambda_{\mathrm{b}}+\mu_{\mathrm{b}})\left[-\mu_{\mathrm{b}}+(\lambda_{\mathrm{b}}+2\mu_{\mathrm{b}})-(\lambda_{\mathrm{b}}+\mu_{\mathrm{b}})\right]=0. ( λ b + μ b ) [ − μ b + ( λ b + 2 μ b ) − ( λ b + μ b ) ] = 0.
Thus (31) is an identity of polynomial matrices; no division by a characteristic factor is made. Set C b = μ b ( λ b + 2 μ b ) > 0 C_{\mathrm{b}}=\mu_{\mathrm{b}}(\lambda_{\mathrm{b}}+2\mu_{\mathrm{b}})>0 C b = μ b ( λ b + 2 μ b ) > 0 , and take two componentwise antiderivatives with zero integration constants to obtain Q ∗ Q_* Q ∗ with ∂ t 2 Q ∗ = p / C b \partial_t^2Q_*=p/C_{\mathrm{b}} ∂ t 2 Q ∗ = p / C b . On polynomials of degree at most d p + 2 d_p+2 d p + 2 set
J c b l = − 1 2 c b l ∑ q = 0 d p + 2 ( ∂ t 2 c b l ) q , Q = J c b l 2 Q ∗ , V p a r t = B s ( ∂ t ) ( e − c b l t Q ) . (32) J_{c_{\mathrm{bl}}}=-\frac{1}{2c_{\mathrm{bl}}}\sum_{q=0}^{d_p+2}\left(\frac{\partial_t}{2c_{\mathrm{bl}}}\right)^q,\qquad Q=J_{c_{\mathrm{bl}}}^2Q_*,\qquad V_{\mathrm{part}}=B_s(\partial_t)(e^{-c_{\mathrm{bl}}t}Q). \tag*{(32)} J c bl = − 2 c bl 1 q = 0 ∑ d p + 2 ( 2 c bl ∂ t ) q , Q = J c bl 2 Q ∗ , V part = B s ( ∂ t ) ( e − c bl t Q ) . (32)
Differentiation is nilpotent on this space, and therefore ( ∂ t − 2 c b l ) J c b l = I (\partial_t-2c_{\mathrm{bl}})J_{c_{\mathrm{bl}}}=I ( ∂ t − 2 c bl ) J c bl = I . Its commutation with ∂ t \partial_t ∂ t gives C b ∂ t 2 ( ∂ t − 2 c b l ) 2 Q = p C_{\mathrm{b}}\partial_t^2(\partial_t-2c_{\mathrm{bl}})^2Q=p C b ∂ t 2 ( ∂ t − 2 c bl ) 2 Q = p . Conjugating ( ∂ t 2 − c b l 2 ) 2 (\partial_t^2-c_{\mathrm{bl}}^2)^2 ( ∂ t 2 − c bl 2 ) 2 by e − c b l t e^{-c_{\mathrm{bl}}t} e − c bl t in (31) now yields P s , 0 V p a r t = e − c b l t p P_{s,0}V_{\mathrm{part}}=e^{-c_{\mathrm{bl}}t}p P s , 0 V part = e − c bl t p . Hence
V = V p a r t − F s V p a r t ( 0 ) V=V_{\mathrm{part}}-F_sV_{\mathrm{part}}(0) V = V part − F s V part ( 0 )
is the required zero-trace solution. The polynomial degree of e c b l t V p a r t e^{c_{\mathrm{bl}}t}V_{\mathrm{part}} e c bl t V part is at most d p + 2 d_p+2 d p + 2 , and that of the subtracted term is at most one. All operations in (32) are finite, with denominators involving only c b l c_{\mathrm{bl}} c bl and C b = μ b ( λ b + 2 μ b ) C_{\mathrm{b}}=\mu_{\mathrm{b}}(\lambda_{\mathrm{b}}+2\mu_{\mathrm{b}}) C b = μ b ( λ b + 2 μ b ) , which are bounded away from zero on the compact parameter patch. The coefficients of the resulting polynomial are smooth functions of y y y . Differentiation in y y y preserves this structure, except that derivatives falling on e − c b l ( y ) t e^{-c_{\mathrm{bl}}(y)t} e − c bl ( y ) t introduce additional polynomial factors in t t t . Consequently, for every multi-index α ∈ N 0 2 \alpha\in\mathbb{N}_0^2 α ∈ N 0 2 and integer j ≥ 0 j \ge0 j ≥ 0 , there exist constants C α , j > 0 C_{\alpha,j}>0 C α , j > 0 and M α , j ∈ N 0 M_{\alpha,j}\in\mathbb{N}_0 M α , j ∈ N 0 such that
∣ ∂ y α ∂ t j V ( y , t ) ∣ ≤ C α , j ( 1 + t ) M α , j e − c 0 t , y ∈ K , t ≥ 0. \left|\partial_y^\alpha\partial_t^jV(y,t)\right|\le C_{\alpha,j}(1+t)^{M_{\alpha,j}}e^{-c_0t},\qquad y\in K,\quad t\ge0. ∂ y α ∂ t j V ( y , t ) ≤ C α , j ( 1 + t ) M α , j e − c 0 t , y ∈ K , t ≥ 0.
The estimates remain uniform when the degree of the forcing polynomial drops. For uniqueness, the adjugate identity applied to the homogeneous system shows that every component satisfies ( D 2 − c b l 2 ) 3 V i = 0 (D^2-c_{\mathrm{bl}}^2)^3V_i=0 ( D 2 − c bl 2 ) 3 V i = 0 . A solution tending to zero at infinity is consequently a polynomial times e − c b l t e^{-c_{\mathrm{bl}}t} e − c bl t and has square-integrable derivatives. If it has zero trace, integration by parts is justified and gives
− ∫ 0 ∞ ( P s , 0 V ) ⋅ V ‾ d t = μ b ∫ 0 ∞ ( ∣ V ′ ∣ 2 + c b l 2 ∣ V ∣ 2 ) d t + ( λ b + μ b ) ∫ 0 ∞ ∣ i s τ tan ⋅ V + n i n ⋅ V ′ ∣ 2 d t . -\int_{0}^{\infty}(P_{s,0}V)\cdot\overline{V}\,dt=\mu_b\int_{0}^{\infty}\left(\lvert V'\rvert^2+c_{\mathrm{bl}}^2\lvert V\rvert^2\right)\,dt+(\lambda_b+\mu_b)\int_{0}^{\infty}\left\lvert is\tau_{\tan}\cdot V+n_{\mathrm{in}}\cdot V'\right\rvert^2\,dt. − ∫ 0 ∞ ( P s , 0 V ) ⋅ V d t = μ b ∫ 0 ∞ ( ∣ V ′ ∣ 2 + c bl 2 ∣ V ∣ 2 ) d t + ( λ b + μ b ) ∫ 0 ∞ ∣ i s τ t a n ⋅ V + n in ⋅ V ′ ∣ 2 d t .
Strong convexity implies λ b + μ b > 0 \lambda_b+\mu_b>0 λ b + μ b > 0 , so V = 0 V=0 V = 0 . The same argument applies to homogeneous H 1 H^1 H 1 solutions with zero trace.
Finite boundary layers and exact correction
Proposition 2.2. Fix an integer N b l ≥ 0 N_{\mathrm{bl}}\ge0 N bl ≥ 0 , an amplitude A ∈ C ∞ ( V ; C 3 ) A\in C^\infty(V;\mathbb{C}^3) A ∈ C ∞ ( V ; C 3 ) , and a phase sign s ∈ { + 1 , − 1 } s\in\{+1,-1\} s ∈ { + 1 , − 1 } . Let η ( r ) \eta(r) η ( r ) be a fixed smooth normal cutoff, compactly supported in the coordinate collar and equal to one near r = 0 r=0 r = 0 . Then, for all sufficiently small h > 0 h>0 h > 0 , there exist profiles U 0 , … , U N b l U_0,\ldots,U_{N_{\mathrm{bl}}} U 0 , … , U N bl , independent of h h h , such that each U ℓ ( y , t ) U_\ell(y,t) U ℓ ( y , t ) is a polynomial in t t t multiplied by e − c b l ( y ) t e^{-c_{\mathrm{bl}}(y)t} e − c bl ( y ) t and is supported in a fixed compact set K ⋐ V K\Subset V K ⋐ V containing supp A \operatorname{supp}A supp A . The field
u h , N b l a p p ( X ( y , r ) ) = e i s y 1 / h η ( r ) ∑ ℓ = 0 N b l h ℓ U ℓ ( y , r / h ) u^{\mathrm{app}}_{h,N_{\mathrm{bl}}}(X(y,r))=e^{isy_1/h}\eta(r)\sum_{\ell=0}^{N_{\mathrm{bl}}}h^\ell U_\ell(y,r/h) u h , N bl app ( X ( y , r )) = e i s y 1 / h η ( r ) ℓ = 0 ∑ N bl h ℓ U ℓ ( y , r / h ) extends by zero to a smooth field on Ω \Omega Ω and has boundary trace e i s y 1 / h A ( y ) e^{isy_1/h}A(y) e i s y 1 / h A ( y ) . There exists an exact weak solution u h ∈ H 1 ( Ω ; C 3 ) u_h\in H^1(\Omega;\mathbb{C}^3) u h ∈ H 1 ( Ω ; C 3 ) of P u h = 0 Pu_h=0 P u h = 0 with the same boundary trace, and
∥ P u h , N b l a p p ∥ L 2 ( Ω ) ≤ C N b l h N b l − 1 / 2 , (33) \left\|Pu^{\mathrm{app}}_{h,N_{\mathrm{bl}}}\right\|_{L^2(\Omega)}\le C_{N_{\mathrm{bl}}}h^{N_{\mathrm{bl}}-1/2}, \tag*{(33)} P u h , N bl app L 2 ( Ω ) ≤ C N bl h N bl − 1/2 , (33) ∥ u h , N b l a p p ∥ L 2 ( Ω ) ≤ C N b l h 1 / 2 , ∥ u h , N b l a p p ∥ H 1 ( Ω ) ≤ C N b l h − 1 / 2 , (34) \left\|u^{\mathrm{app}}_{h,N_{\mathrm{bl}}}\right\|_{L^2(\Omega)}\le C_{N_{\mathrm{bl}}}h^{1/2},\qquad
\left\|u^{\mathrm{app}}_{h,N_{\mathrm{bl}}}\right\|_{H^1(\Omega)}\le C_{N_{\mathrm{bl}}}h^{-1/2}, \tag*{(34)} u h , N bl app L 2 ( Ω ) ≤ C N bl h 1/2 , u h , N bl app H 1 ( Ω ) ≤ C N bl h − 1/2 , (34) ∥ u h − u h , N b l a p p ∥ H 1 ( Ω ) ≤ C N b l h N b l − 1 / 2 . (35) \left\|u_h-u^{\mathrm{app}}_{h,N_{\mathrm{bl}}}\right\|_{H^1(\Omega)}\le C_{N_{\mathrm{bl}}}h^{N_{\mathrm{bl}}-1/2}. \tag*{(35)} u h − u h , N bl app H 1 ( Ω ) ≤ C N bl h N bl − 1/2 . (35) The constants are independent of h h h .
More precisely, for each fixed finite integer S ≥ 0 S\ge0 S ≥ 0 , the profiles and all derivatives required in the construction are controlled by the C 2 N b l + S + 4 C^{2N_{\mathrm{bl}}+S+4} C 2 N bl + S + 4 norm of A A A , the coefficient bounds
max ∣ α ∣ ≤ 2 N b l + S + 4 0 ≤ b ≤ N b l + S + 2 ∥ ∂ y α ∂ r b ( λ , μ , W , n i n , w ) ∥ L ∞ \max_{\substack{\lvert\alpha\rvert\le2N_{\mathrm{bl}}+S+4\\0\le b\le N_{\mathrm{bl}}+S+2}}\left\|\partial_y^\alpha\partial_r^b(\lambda,\mu,W,n_{\mathrm{in}},w)\right\|_{L^\infty} ∣ α ∣ ≤ 2 N bl + S + 4 0 ≤ b ≤ N bl + S + 2 max ∂ y α ∂ r b ( λ , μ , W , n in , w ) L ∞ on the fixed compact collar, finitely many cutoff norms, positive lower bounds for c b l , μ , λ + 2 μ , λ + 3 μ c_{\mathrm{bl}},\mu,\lambda+2\mu,\lambda+3\mu c bl , μ , λ + 2 μ , λ + 3 μ , and upper and positive lower bounds for the coordinate Jacobian. The exact correction also depends on ∥ P D − 1 ∥ H − 1 → H 1 \left\|P_D^{-1}\right\|_{H^{-1}\to H^1} P D − 1 H − 1 → H 1 .
Proof. Using the profile in (30) , set U 0 = F s A U_0=F_sA U 0 = F s A . By the defining properties of the frozen profile,
P s , 0 U 0 = 0 , U 0 ( y , 0 ) = A ( y ) . P_{s,0}U_0=0,\qquad U_0(y,0)=A(y). P s , 0 U 0 = 0 , U 0 ( y , 0 ) = A ( y ) .
For the finite sum ∑ ℓ = 0 N b l h ℓ U ℓ \sum_{\ell=0}^{N_{\mathrm{bl}}}h^\ell U_\ell ∑ ℓ = 0 N bl h ℓ U ℓ , the expansion (28) shows that the coefficient of h ℓ h^\ell h ℓ is ∑ q = 0 ℓ P s , q U ℓ − q \sum_{q=0}^{\ell}P_{s,q}U_{\ell-q} ∑ q = 0 ℓ P s , q U ℓ − q for ℓ ≤ N b l \ell\le N_{\mathrm{bl}} ℓ ≤ N bl . We cancel these coefficients successively by imposing
P s , 0 U ℓ = − ∑ q = 1 ℓ P s , q U ℓ − q , U ℓ ( y , 0 ) = 0 , 1 ≤ ℓ ≤ N b l . (36) P_{s,0}U_\ell=-\sum_{q=1}^{\ell}P_{s,q}U_{\ell-q},\qquad U_\ell(y,0)=0,\qquad1\le\ell\le N_{\mathrm{bl}}. \tag*{(36)} P s , 0 U ℓ = − q = 1 ∑ ℓ P s , q U ℓ − q , U ℓ ( y , 0 ) = 0 , 1 ≤ ℓ ≤ N bl . (36)
Every right-hand side is a polynomial in t t t multiplied by e − c b l t e^{-c_{\mathrm{bl}}t} e − c bl t . Hence Lemma 2.1 applies at each step and produces the required decaying profile while preserving the prescribed tangential support. The exact slow-derivative count after (23) gives the sharper bound
deg ( e c b l t U ℓ ) ≤ 1 + 3 ℓ . (37) \deg\left(e^{c_{\mathrm{bl}}t}U_\ell\right)\le1+3\ell. \tag*{(37)} deg ( e c bl t U ℓ ) ≤ 1 + 3 ℓ . (37)
Indeed, a term of P s , q \mathcal{P}_{s,q} P s , q contains t q − m t^{q-m} t q − m and at most m m m slow tangential derivatives. Its polynomial degree therefore increases by at most q q q , including derivatives falling on e − c b l ( y ) t e^{-c_{\mathrm{bl}}(y)t} e − c bl ( y ) t . The inverse in Lemma 2.1 adds at most two to the polynomial degree. Thus, assuming the estimate for lower indices, the q q q th source in the recursion produces degree at most
1 + 3 ( ℓ − q ) + q + 2 ≤ 1 + 3 ℓ , q ≥ 1 , 1+3(\ell-q)+q+2\leq1+3\ell,\qquad q\geq1, 1 + 3 ( ℓ − q ) + q + 2 ≤ 1 + 3 ℓ , q ≥ 1 ,
which proves (37) by induction.
The coefficients in (24) involve at most q + 1 q+1 q + 1 normal derivatives and one tangential derivative of the original coefficients, while each recursion applies at most two further tangential derivatives. The stated rectangular derivative bounds therefore control all profiles with ℓ ≤ N b l \ell\leq N_{\mathrm{bl}} ℓ ≤ N bl and the Taylor remainders. The additional S + 4 S+4 S + 4 derivatives cover the final operator application and the prescribed differentiations. The frozen inverse uses finite algebraic operations with positive denominators, so the normal derivative count remains within the rectangular bounds stated above.
On the region where η = 1 \eta=1 η = 1 , the recursion (36) cancels all powers h 0 , … , h N b l h^{0},\ldots,h^{N_{\mathrm{bl}}} h 0 , … , h N bl . Let p N b l ∈ N 0 p_{N_{\mathrm{bl}}\!}\in\mathbb{N}_{0} p N bl ∈ N 0 be a fixed polynomial-growth exponent bounding the finitely many Taylor remainders and profiles at this truncation order. Taylor’s formula (25) gives
∣ P s , h ∑ ℓ = 0 N b l h ℓ U ℓ ∣ ≤ C N b l h N b l + 1 ( 1 + t ) p N b l e − c 0 t . \left|\mathcal{P}_{s,h}\sum_{\ell=0}^{N_{\mathrm{bl}}}h^\ell U_\ell\right|
\leq C_{N_{\mathrm{bl}}}h^{N_{\mathrm{bl}}+1}(1+t)^{p_{N_{\mathrm{bl}}}}e^{-c_0t}. P s , h ℓ = 0 ∑ N bl h ℓ U ℓ ≤ C N bl h N bl + 1 ( 1 + t ) p N bl e − c 0 t .
Multiplication by h − 2 h^{-2} h − 2 shows that the physical residual is bounded by
C N b l h N b l − 1 ( 1 + t ) p N b l e − c 0 t . C_{N_{\mathrm{bl}}}h^{N_{\mathrm{bl}}-1}(1+t)^{p_{N_{\mathrm{bl}}}}e^{-c_0t}. C N bl h N bl − 1 ( 1 + t ) p N bl e − c 0 t .
Since the volume form is J ( y , r ) d y d r J(y,r)\,dy\,dr J ( y , r ) d y d r , with d r = h d t dr=h\,dt d r = h d t and J J J bounded above and below, integration in the stretched normal variable gives (33) .
On supp η \operatorname{supp}\eta supp η , one has r ≥ r ∗ > 0 r\geq r_* >0 r ≥ r ∗ > 0 . Consequently, every cutoff commutator is bounded by C h − M e − c 0 r ∗ / h Ch^{-M}e^{-c_0r_*/h} C h − M e − c 0 r ∗ / h for some fixed M ≥ 0 M\geq0 M ≥ 0 , and hence decays faster than any prescribed power of h h h . The same integration, with a factor at most h − 1 h^{-1} h − 1 for each physical derivative, proves (34) .
By § 1.7 , the physical Dirichlet realization P D P_D P D has a bounded inverse
P D − 1 : H − 1 ( Ω ; C 3 ) ⟶ H 0 1 ( Ω ; C 3 ) . P_D^{-1}:H^{-1}(\Omega;\mathbb{C}^3)\longrightarrow H_0^1(\Omega;\mathbb{C}^3). P D − 1 : H − 1 ( Ω ; C 3 ) ⟶ H 0 1 ( Ω ; C 3 ) .
Define
u h = u h , N b l a p p − P D − 1 ( P u h , N b l a p p ) . u_h=u_{h,N_{\mathrm{bl}}}^{\mathrm{app}}-P_D^{-1}(P u_{h,N_{\mathrm{bl}}}^{\mathrm{app}}). u h = u h , N bl app − P D − 1 ( P u h , N bl app ) .
The correction belongs to H 0 1 ( Ω ; C 3 ) H_0^1(\Omega;\mathbb{C}^3) H 0 1 ( Ω ; C 3 ) , so it preserves the boundary trace. Moreover, P u h = 0 Pu_h=0 P u h = 0 in the weak sense. Finally,
∥ u h − u h , N b l a p p ∥ H 1 ( Ω ) ≤ ∥ P D − 1 ∥ H − 1 → H 0 1 ∥ P u h , N b l a p p ∥ H − 1 ( Ω ) ≤ C ∥ P u h , N b l a p p ∥ L 2 ( Ω ) ≤ C N b l h N b l − 1 / 2 . \left\|u_h-u_{h,N_{\mathrm{bl}}}^{\mathrm{app}}\right\|_{H^1(\Omega)}
\leq\left\|P_D^{-1}\right\|_{H^{-1}\to H_0^1}\left\|Pu_{h,N_{\mathrm{bl}}}^{\mathrm{app}}\right\|_{H^{-1}(\Omega)}
\leq C\left\|Pu_{h,N_{\mathrm{bl}}}^{\mathrm{app}}\right\|_{L^2(\Omega)}
\leq C_{N_{\mathrm{bl}}}h^{N_{\mathrm{bl}}-1/2}. u h − u h , N bl app H 1 ( Ω ) ≤ P D − 1 H − 1 → H 0 1 P u h , N bl app H − 1 ( Ω ) ≤ C P u h , N bl app L 2 ( Ω ) ≤ C N bl h N bl − 1/2 .
This proves (35) and completes the proof. □
Recovery at the first two elastic orders
Consider two materials satisfying the preceding assumptions and having equal full boundary maps (8) . For either material, the weak representation (11) expresses the boundary pairing in terms of the corresponding interior solutions. We first use the leading boundary response to recover the boundary values of the Lamé parameters, and then use the bilinear identity between the two materials to recover their first normal derivatives.
At the boundary point under consideration, choose coordinates so that
e b l = e 1 , n i n = e 3 , ν = − e 3 , c b l = 1. e_{\mathrm{bl}}=e_1,\qquad n_{\mathrm{in}}=e_3,\qquad\nu=-e_3,\qquad c_{\mathrm{bl}}=1. e bl = e 1 , n in = e 3 , ν = − e 3 , c bl = 1.
In these frozen coordinates, write N ± = N s ∣ s = ± 1 N_{\pm}=N_s|_{s=\pm1} N ± = N s ∣ s = ± 1 , where N s N_s N s is defined in (30) . Thus
N ± = a b l ( 1 0 ± i 0 0 0 ± i 0 − 1 ) . N_{\pm}=a_{\mathrm{bl}}\begin{pmatrix}1&0&\pm i\\0&0&0\\\pm i&0&-1\end{pmatrix}. N ± = a bl 1 0 ± i 0 0 0 ± i 0 − 1 .
For each material separately, use the positive- and negative-phase boundary layers in (11) and divide the boundary pairing by k = h − 1 k=h^{-1} k = h − 1 . For N b l ≥ 1 N_{\mathrm{bl}}\geq1 N bl ≥ 1 , Proposition 2.2 allows the exact solutions to be replaced by the approximate layers in this pairing. Taking k → ∞ k\to\infty k → ∞ with a fixed tangential cutoff and then localizing at the chosen boundary point extracts the leading traction response. The potential term contributes O ( k − 2 ) O(k^{-2}) O ( k − 2 ) at this order.
Recall from (1) and (8) that the Dirichlet-to-Neumann map sends a boundary displacement f f f to the corresponding traction:
Λ λ , μ , ρ Ω ( ω ) f = σ λ , μ ( u f ) ν ∣ ∂ Ω . \Lambda_{\lambda,\mu,\rho}^{\Omega}(\omega)f=\left.\sigma_{\lambda,\mu}(u_f)\nu\right|_{\partial\Omega}. Λ λ , μ , ρ Ω ( ω ) f = σ λ , μ ( u f ) ν ∣ ∂ Ω .
where u f u_f u f is the solution with boundary value f f f .
At the fixed boundary point y 0 y_0 y 0 , freeze the coefficients and write
F + ( t ) = e − t ( I 3 − t N + ) . F_+(t)=e^{-t}(I_3-tN_+). F + ( t ) = e − t ( I 3 − t N + ) .
For a smooth boundary amplitude A ( y ) A(y) A ( y ) , consider the leading positive-phase profile e i y 1 / h F + ( r / h ) A ( y ) e^{iy_1/h}F_+(r/h)A(y) e i y 1 / h F + ( r / h ) A ( y ) . The leading outward traction is of order h − 1 h^{-1} h − 1 . We denote its coefficient matrix by Z + Z_+ Z + , so that
h e − i y 1 / h [ h σ λ , μ ( e i y 1 / h F + ( r / h ) A ( y ) ) ν ] ∣ ( y , r ) = ( y 0 , 0 ) = Z + A ( y 0 ) + O ( h ) . \left.he^{-iy_1/h}\left[h\sigma_{\lambda,\mu}\left(e^{iy_1/h}F_+(r/h)A(y)\right)\nu\right]\right|_{(y,r)=(y_0,0)}
=Z_+A(y_0)+O(h). h e − i y 1 / h [ h σ λ , μ ( e i y 1 / h F + ( r / h ) A ( y ) ) ν ] ( y , r ) = ( y 0 , 0 ) = Z + A ( y 0 ) + O ( h ) .
Thus Z + Z_+ Z + maps the boundary amplitude to the leading traction amplitude. Direct evaluation of the traction, using ν = − e 3 \nu=-e_3 ν = − e 3 , gives
Z + = ( d b d 0 − i b ∗ 0 μ 0 i b ∗ 0 d b d ) , d b d = 2 μ ( λ + 2 μ ) λ + 3 μ , b ∗ = 2 μ 2 λ + 3 μ . (38) Z_+=\begin{pmatrix}d_{\mathrm{bd}}&0&-ib_*\\0&\mu&0\\ib_*&0&d_{\mathrm{bd}}\end{pmatrix},
\qquad
d_{\mathrm{bd}}=\frac{2\mu(\lambda+2\mu)}{\lambda+3\mu},
\qquad
b_*=\frac{2\mu^2}{\lambda+3\mu}. \tag*{(38)} Z + = d bd 0 i b ∗ 0 μ 0 − i b ∗ 0 d bd , d bd = λ + 3 μ 2 μ ( λ + 2 μ ) , b ∗ = λ + 3 μ 2 μ 2 . (38)
Equivalently,
Z + = A 0 ( I 3 + N + ) − i R , A 0 = diag ( μ , μ , λ + 2 μ ) , R = ( 0 0 μ 0 0 0 λ 0 0 ) . Z_+=A_0(I_3+N_+)-iR,
\qquad
A_0=\operatorname{diag}(\mu,\mu,\lambda+2\mu),
\qquad
R=\begin{pmatrix}0&0&\mu\\0&0&0\\\lambda&0&0\end{pmatrix}. Z + = A 0 ( I 3 + N + ) − i R , A 0 = diag ( μ , μ , λ + 2 μ ) , R = 0 0 λ 0 0 0 μ 0 0 .
Reversing the phase gives Z − = Z + T Z_-=Z_+^{\mathsf{T}} Z − = Z + T . Since the two boundary maps agree, their leading traction matrices agree. The entries of Z + Z_+ Z + therefore determine the boundary values:
μ = ( Z + ) 22 , λ = 3 d b d μ − 4 μ 2 2 μ − d b d , 2 μ − d b d = b ∗ > 0. \mu=(Z_+)_{22},
\qquad
\lambda=\frac{3d_{\mathrm{bd}}\mu-4\mu^2}{2\mu-d_{\mathrm{bd}}},
\qquad
2\mu-d_{\mathrm{bd}}=b_*>0. μ = ( Z + ) 22 , λ = 2 μ − d bd 3 d bd μ − 4 μ 2 , 2 μ − d bd = b ∗ > 0.
Thus the boundary values of λ \lambda λ and μ \mu μ agree for the two materials. Define the boundary Taylor coefficients by
L j b d ( y ) : = ∂ r j ( λ 1 − λ 2 ) ( y , 0 ) , j ≥ 1 , M j b d ( y ) : = ∂ r j ( μ 1 − μ 2 ) ( y , 0 ) , j ≥ 1 , V j b d ( y ) : = ∂ r j − 2 ( W 1 − W 2 ) ( y , 0 ) , j ≥ 2. (39) \begin{aligned}
L_j^{\mathrm{bd}}(y)&:=\partial_r^j(\lambda_1-\lambda_2)(y,0), && j\geq1,\\
M_j^{\mathrm{bd}}(y)&:=\partial_r^j(\mu_1-\mu_2)(y,0), && j\geq1,\\
V_j^{\mathrm{bd}}(y)&:=\partial_r^{j-2}(W_1-W_2)(y,0), && j\geq2. \tag*{(39)}
\end{aligned} L j bd ( y ) M j bd ( y ) V j bd ( y ) := ∂ r j ( λ 1 − λ 2 ) ( y , 0 ) , := ∂ r j ( μ 1 − μ 2 ) ( y , 0 ) , := ∂ r j − 2 ( W 1 − W 2 ) ( y , 0 ) , j ≥ 1 , j ≥ 1 , j ≥ 2. (39)
The index j j j in (39) records the elastic Taylor order, rather than the material label; at j = 1 j=1 j = 1 only the Lamé coefficients occur. For a coefficient b b b carrying a material index, write δ 12 b : = b 1 − b 2 \delta_{12}b:=b_1-b_2 δ 12 b := b 1 − b 2 .
Define the bilinear form
B ( v 1 , v 2 ) : = ∫ Ω [ δ 12 λ div v 1 div v 2 + 2 δ 12 μ ε ( v 1 ) : ε ( v 2 ) − δ 12 W v 1 ⋅ v 2 ] d x . \mathcal{B}(v_1,v_2):=\int_{\Omega}\left[\delta_{12}\lambda\,\operatorname{div}v_1\operatorname{div}v_2+2\delta_{12}\mu\,\varepsilon(v_1):\varepsilon(v_2)-\delta_{12}W\,v_1\cdot v_2\right]\,dx. B ( v 1 , v 2 ) := ∫ Ω [ δ 12 λ div v 1 div v 2 + 2 δ 12 μ ε ( v 1 ) : ε ( v 2 ) − δ 12 W v 1 ⋅ v 2 ] d x .
Applying (11) to solutions u 1 , u 2 u_1,u_2 u 1 , u 2 for the two materials and using equality of the boundary maps gives the bilinear identity
B ( u 1 , u 2 ) = 0. (40) \mathcal{B}(u_1,u_2)=0. \tag*{(40)} B ( u 1 , u 2 ) = 0. (40)
Since the boundary values of the Lamé parameters agree, Taylor’s formula (25) gives
δ 12 λ = r L 1 b d + O ( r 2 ) , δ 12 μ = r M 1 b d + O ( r 2 ) . \delta_{12}\lambda=rL^{\mathrm{bd}}_1+O(r^2),\qquad\delta_{12}\mu=rM^{\mathrm{bd}}_1+O(r^2). δ 12 λ = r L 1 bd + O ( r 2 ) , δ 12 μ = r M 1 bd + O ( r 2 ) .
At this stage, the boundary value of δ 12 W \delta_{12}W δ 12 W remains an unknown in the moment system.
Fix the boundary point y 0 y_0 y 0 and suppress the dependence on this point in the following formulas. Use the boundary layers of Proposition 2.2 with N b l ≥ 3 N_{\mathrm{bl}}\geq3 N bl ≥ 3 . First take the high-frequency limit for a fixed tangential cutoff, and then localize at y 0 y_0 y 0 . At this order, the normalized contribution of the potential term is O ( k − 1 ) O(k^{-1}) O ( k − 1 ) and vanishes as k → ∞ k\to\infty k → ∞ . Rows of the moment matrix correspond to the negative-phase testing polarization, and columns to the positive-phase input polarization.
For p = 1 , 2 , 3 p=1,2,3 p = 1 , 2 , 3 , define the polynomial gradient matrices
G ± p = [ ± i F ~ ± p , 0 , ( ( t − 1 ) N ± − I 3 ) p ] , F ~ ± = I 3 − t N ± . (41) G_\pm^p=\left[\ \pm i\widetilde{F}_\pm^p,\ 0,\ \bigl((t-1)N_\pm-I_3\bigr)^p\right],\qquad\widetilde{F}_\pm=I_3-tN_\pm. \tag*{(41)} G ± p = [ ± i F ± p , 0 , ( ( t − 1 ) N ± − I 3 ) p ] , F ± = I 3 − t N ± . (41)
Here the superscript p p p denotes the p p p th column. We write sym A = 1 2 ( A + A T ) \operatorname{sym} A=\frac{1}{2}(A+A^T) sym A = 2 1 ( A + A T ) for the symmetric part of a matrix. For p , q ∈ { 1 , 2 , 3 } p,q\in\{1,2,3\} p , q ∈ { 1 , 2 , 3 } , introduce the bilinear products
D p q ( t ) : = tr G − p ( t ) tr G + q ( t ) , S p q ( t ) : = 2 sym G − p ( t ) : sym G + q ( t ) , Q p q ( t ) : = F ~ − p ( t ) ⋅ F ~ + q ( t ) . (42) \begin{aligned}
D_{pq}(t)&:=\operatorname{tr}G_-^p(t)\operatorname{tr}G_+^q(t),\\
S_{pq}(t)&:=2\operatorname{sym}G_-^p(t):\operatorname{sym}G_+^q(t),\\
Q_{pq}(t)&:=\widetilde{F}_-^p(t)\cdot\widetilde{F}_+^q(t). \tag*{(42)}
\end{aligned} D pq ( t ) S pq ( t ) Q pq ( t ) := tr G − p ( t ) tr G + q ( t ) , := 2 sym G − p ( t ) : sym G + q ( t ) , := F − p ( t ) ⋅ F + q ( t ) . (42)
All products are bilinear, not Hermitian. In particular, D p : = D p p D_p:=D_{pp} D p := D pp , S p : = S p p S_p:=S_{pp} S p := S pp , and Q p : = Q p p Q_p:=Q_{pp} Q p := Q pp will be used for the diagonal moments below.
To identify the first-order boundary moment, set r = h t r=ht r = h t . At the chosen boundary point y 0 y_0 y 0 , suppress the dependence on y 0 y_0 y 0 in the following formulas. Taylor’s formula gives
δ 12 λ ( y 0 , h t ) = h t L 1 b d + O ( h 2 t 2 ) , \delta_{12}\lambda(y_0,ht)=htL^{\mathrm{bd}}_1+O(h^2t^2), δ 12 λ ( y 0 , h t ) = h t L 1 bd + O ( h 2 t 2 ) ,
δ 12 μ ( y 0 , h t ) = h t M 1 b d + O ( h 2 t 2 ) . \delta_{12}\mu(y_0,ht)=htM^{\mathrm{bd}}_1+O(h^2t^2). δ 12 μ ( y 0 , h t ) = h t M 1 bd + O ( h 2 t 2 ) .
Substituting these expansions and the leading gradient matrices into the elastic terms of (40) , the principal integrand, before multiplication by the volume Jacobian and the common localization factors, is
h − 1 t e − 2 t [ L 1 b d D p q ( t ) + M 1 b d S p q ( t ) ] . h^{-1}te^{-2t}\left[L^{\mathrm{bd}}_1D_{pq}(t)+M^{\mathrm{bd}}_1S_{pq}(t)\right]. h − 1 t e − 2 t [ L 1 bd D pq ( t ) + M 1 bd S pq ( t ) ] .
The opposite phases cancel. Under the change of variables r = h t r=ht r = h t , the volume element is
d x = J ( y , h t ) d y h d t . dx=J(y,ht)\,dy\,h\,dt. d x = J ( y , h t ) d y h d t .
Thus the factor h − 1 h^{-1} h − 1 is canceled by the factor h h h in the normal integration measure. The potential contribution is O ( h ) O(h) O ( h ) and vanishes as h → 0 h\to0 h → 0 . The error estimates in Proposition 2.2 justify the high-frequency limit. After localization at y 0 y_0 y 0 and division by the positive Jacobian and localization factors, the remaining normal integral is the first-order moment.
Accordingly, define the first-order boundary moment matrix H 1 H_1 H 1 by
( H 1 ) p q = ∫ 0 ∞ t e − 2 t [ L 1 b d D p q ( t ) + M 1 b d S p q ( t ) ] d t , p , q = 1 , 2 , 3. (43) (H_1)_{pq}=\int_0^\infty te^{-2t}\left[L^{\mathrm{bd}}_1D_{pq}(t)+M^{\mathrm{bd}}_1S_{pq}(t)\right]\,dt,\qquad p,q=1,2,3. \tag*{(43)} ( H 1 ) pq = ∫ 0 ∞ t e − 2 t [ L 1 bd D pq ( t ) + M 1 bd S pq ( t ) ] d t , p , q = 1 , 2 , 3. (43)
Since (40) vanishes for every admissible choice of boundary data, its high-frequency limit and localization give
( H 1 ( y 0 ) ) p q = 0 , p , q = 1 , 2 , 3. \bigl(H_1(y_0)\bigr)_{pq}=0,\qquad p,q=1,2,3. ( H 1 ( y 0 ) ) pq = 0 , p , q = 1 , 2 , 3.
Hence H 1 ( y 0 ) = 0 H_1(y_0)=0 H 1 ( y 0 ) = 0 . We now compute its entries explicitly.
Using (41) and
∫ 0 ∞ t s e − 2 t d t = s ! 2 s + 1 , s ∈ N 0 , \int_{0}^{\infty} t^s e^{-2t}\,\mathrm{d}t = \frac{s!}{2^{s+1}}, \qquad s \in\mathbb{N}_0, ∫ 0 ∞ t s e − 2 t d t = 2 s + 1 s ! , s ∈ N 0 ,
gives
H 1 , 22 = 1 2 M 1 b d , H 1 , 31 = − H 1 , 13 , H_{1,22}=\frac{1}{2}M^{\mathrm{bd}}_1,\qquad H_{1,31}=-H_{1,13}, H 1 , 22 = 2 1 M 1 bd , H 1 , 31 = − H 1 , 13 ,
and
− i H 1 , 13 = 1 4 [ ( 1 − a b l ) 2 L 1 b d + ( 7 a b l 2 − 2 a b l − 1 ) M 1 b d ] , -iH_{1,13}=\frac{1}{4}\left[(1-a_{\mathrm{bl}})^2L^{\mathrm{bd}}_1+(7a_{\mathrm{bl}}^2-2a_{\mathrm{bl}}-1)M^{\mathrm{bd}}_1\right], − i H 1 , 13 = 4 1 [ ( 1 − a bl ) 2 L 1 bd + ( 7 a bl 2 − 2 a bl − 1 ) M 1 bd ] ,
H 1 , 11 = 1 4 [ ( 1 − a b l ) 2 L 1 b d + ( 7 a b l 2 − 6 a b l + 3 ) M 1 b d ] , H_{1,11}=\frac{1}{4}\left[(1-a_{\mathrm{bl}})^2L^{\mathrm{bd}}_1+(7a_{\mathrm{bl}}^2-6a_{\mathrm{bl}}+3)M^{\mathrm{bd}}_1\right], H 1 , 11 = 4 1 [ ( 1 − a bl ) 2 L 1 bd + ( 7 a bl 2 − 6 a bl + 3 ) M 1 bd ] ,
H 1 , 33 = 1 4 [ ( 1 − a b l ) 2 L 1 b d + ( 7 a b l 2 + 2 a b l + 3 ) M 1 b d ] . H_{1,33}=\frac{1}{4}\left[(1-a_{\mathrm{bl}})^2L^{\mathrm{bd}}_1+(7a_{\mathrm{bl}}^2+2a_{\mathrm{bl}}+3)M^{\mathrm{bd}}_1\right]. H 1 , 33 = 4 1 [ ( 1 − a bl ) 2 L 1 bd + ( 7 a bl 2 + 2 a bl + 3 ) M 1 bd ] .
The other entries are zero. Using
λ = μ 3 a b l − 1 1 − a b l , \lambda=\mu\frac{3a_{\mathrm{bl}}-1}{1-a_{\mathrm{bl}}}, λ = μ 1 − a bl 3 a bl − 1 ,
the expression for − i H 1 , 13 -iH_{1,13} − i H 1 , 13 reduces to
b = μ 2 L 1 b d + ( λ 2 − 2 μ 2 ) M 1 b d ( λ + 3 μ ) 2 . b=\frac{\mu^2L^{\mathrm{bd}}_1+(\lambda^2-2\mu^2)M^{\mathrm{bd}}_1}{(\lambda+3\mu)^2}. b = ( λ + 3 μ ) 2 μ 2 L 1 bd + ( λ 2 − 2 μ 2 ) M 1 bd .
The diagonal entries then give the explicit matrix
H 1 = ( M 1 b d ( 1 − a b l ) + b 0 i b 0 M 1 b d / 2 0 − i b 0 M 1 b d ( 1 + a b l ) + b ) (44) H_1=
\begin{pmatrix}
M^{\mathrm{bd}}_1(1-a_{\mathrm{bl}})+b & 0 & ib\\
0 & M^{\mathrm{bd}}_1/2 & 0\\
-ib & 0 & M^{\mathrm{bd}}_1(1+a_{\mathrm{bl}})+b
\end{pmatrix} \tag*{(44)} H 1 = M 1 bd ( 1 − a bl ) + b 0 − ib 0 M 1 bd /2 0 ib 0 M 1 bd ( 1 + a bl ) + b (44)
Since H 1 ( y 0 ) = 0 H_1(y_0)=0 H 1 ( y 0 ) = 0 , its ( 2 , 2 ) (2,2) ( 2 , 2 ) entry gives
1 2 M 1 b d = 0 , M 1 b d = 0. \frac{1}{2}M^{\mathrm{bd}}_1=0,\qquad M^{\mathrm{bd}}_1=0. 2 1 M 1 bd = 0 , M 1 bd = 0.
Then H 1 , 13 = 0 H_{1,13}=0 H 1 , 13 = 0 and the formula for − i H 1 , 13 -iH_{1,13} − i H 1 , 13 imply
( 1 − a b l ) 2 L 1 b d = 0. (1-a_{\mathrm{bl}})^2L^{\mathrm{bd}}_1=0. ( 1 − a bl ) 2 L 1 bd = 0.
Since a b l < 1 a_{\mathrm{bl}}<1 a bl < 1 , we conclude that L 1 b d = 0 L^{\mathrm{bd}}_1=0 L 1 bd = 0 .
Lemma 2.3. Fix a b l ∈ ( 0 , 1 ) a_{\mathrm{bl}}\in(0,1) a bl ∈ ( 0 , 1 ) and an integer j ≥ 2 j\geq2 j ≥ 2 . Use the frozen matrices N ± N_{\pm} N ± , F ± e F^{\mathrm{e}}_{\pm} F ± e , G ± p G^{\mathrm{p}}_{\pm} G ± p defined in the preceding subsection, with c b l = 1 c_{\mathrm{bl}}=1 c bl = 1 and e b l = e 1 e_{\mathrm{bl}}=e_1 e bl = e 1 , n i n = e 3 n_{\mathrm{in}}=e_3 n in = e 3 . For the diagonal products, write
D p : = D p p , S p : = S p p , Q p : = Q p p , p = 1 , 2 , 3 , D_p:=D_{pp},\qquad S_p:=S_{pp},\qquad Q_p:=Q_{pp},\qquad p=1,2,3, D p := D pp , S p := S pp , Q p := Q pp , p = 1 , 2 , 3 , where D p q D_{pq} D pq , S p q S_{pq} S pq , Q p q Q_{pq} Q pq are defined in (42) .
At a fixed boundary point y y y , regard L j b d ( y ) L^{\mathrm{bd}}_j(y) L j bd ( y ) , M j b d ( y ) M^{\mathrm{bd}}_j(y) M j bd ( y ) , and V j b d ( y ) V^{\mathrm{bd}}_j(y) V j bd ( y ) , defined in (39) , as scalars, and define
E p p = 2 j + 3 ∫ 0 ∞ e − 2 t [ t j j ! ( L j b d D p + M j b d S p ) − t j − 2 ( j − 2 ) ! V j b d Q p ] d t , p = 1 , 2 , 3. (45) E_{pp}=2^{j+3}\int_{0}^{\infty}e^{-2t}\left[\frac{t^j}{j!}\left(L^{\mathrm{bd}}_jD_p+M^{\mathrm{bd}}_jS_p\right)-\frac{t^{j-2}}{(j-2)!}V^{\mathrm{bd}}_jQ_p\right]\mathrm{d}t,\qquad p=1,2,3. \tag*{(45)} E pp = 2 j + 3 ∫ 0 ∞ e − 2 t [ j ! t j ( L j bd D p + M j bd S p ) − ( j − 2 )! t j − 2 V j bd Q p ] d t , p = 1 , 2 , 3. (45) Then
E 22 = 8 M j b d − 16 V j b d , E_{22}=8M^{\mathrm{bd}}_j-16V^{\mathrm{bd}}_j, E 22 = 8 M j bd − 16 V j bd , E 33 − E 11 = 32 a b l [ j M j b d − ( j − 1 ) V j b d ] , E_{33}-E_{11}=32a_{\mathrm{bl}}\left[jM^{\mathrm{bd}}_j-(j-1)V^{\mathrm{bd}}_j\right], E 33 − E 11 = 32 a bl [ j M j bd − ( j − 1 ) V j bd ] , E 11 = 4 ( 1 − a b l ) 2 L j b d + B j M j b d − C j V j b d , E_{11}=4(1-a_{\mathrm{bl}})^2L^{\mathrm{bd}}_j+B_jM^{\mathrm{bd}}_j-C_jV^{\mathrm{bd}}_j, E 11 = 4 ( 1 − a bl ) 2 L j bd + B j M j bd − C j V j bd , B j = 8 j ( j + 1 ) a b l 2 − 16 j a b l + 12 a b l 2 − 8 a b l + 12 , B_j=8j(j+1)a_{\mathrm{bl}}^2-16ja_{\mathrm{bl}}+12a_{\mathrm{bl}}^2-8a_{\mathrm{bl}}+12, B j = 8 j ( j + 1 ) a bl 2 − 16 j a bl + 12 a bl 2 − 8 a bl + 12 , C j = 8 j ( j − 1 ) a b l 2 − 16 ( j − 1 ) a b l + 16. (46) C_j=8j(j-1)a_{\mathrm{bl}}^2-16(j-1)a_{\mathrm{bl}}+16. \tag*{(46)} C j = 8 j ( j − 1 ) a bl 2 − 16 ( j − 1 ) a bl + 16. (46) In particular, E 11 = E 22 = E 33 = 0 E_{11}=E_{22}=E_{33}=0 E 11 = E 22 = E 33 = 0 implies L j b d = M j b d = V j b d = 0 L^{\mathrm{bd}}_j=M^{\mathrm{bd}}_j=V^{\mathrm{bd}}_j=0 L j bd = M j bd = V j bd = 0 .
Proof. Direct multiplication of the matrices defined in (41) , with bilinear rather than Hermitian products, yields
( D 1 , D 2 , D 3 ) = ( ( 1 − a b l ) 2 , 0 , ( 1 − a b l ) 2 ) . (D_1,D_2,D_3)=\left((1-a_{\mathrm{bl}})^2,0,(1-a_{\mathrm{bl}})^2\right). ( D 1 , D 2 , D 3 ) = ( ( 1 − a bl ) 2 , 0 , ( 1 − a bl ) 2 ) .
( S 1 , S 2 , S 3 ) = ( 8 a b l 2 t 2 − 8 a b l ( a b l + 1 ) t + 3 a b l 2 + 2 a b l + 3 , 2 , 8 a b l 2 t 2 − 8 a b l ( a b l − 1 ) t + 3 a b l 2 − 6 a b l + 3 ) . (S_1,S_2,S_3)=\left(8a_{\mathrm{bl}}^2t^2-8a_{\mathrm{bl}}(a_{\mathrm{bl}}+1)t+3a_{\mathrm{bl}}^2+2a_{\mathrm{bl}}+3,\ 2,\ 8a_{\mathrm{bl}}^2t^2-8a_{\mathrm{bl}}(a_{\mathrm{bl}}-1)t+3a_{\mathrm{bl}}^2-6a_{\mathrm{bl}}+3\right). ( S 1 , S 2 , S 3 ) = ( 8 a bl 2 t 2 − 8 a bl ( a bl + 1 ) t + 3 a bl 2 + 2 a bl + 3 , 2 , 8 a bl 2 t 2 − 8 a bl ( a bl − 1 ) t + 3 a bl 2 − 6 a bl + 3 ) .
( Q 1 , Q 2 , Q 3 ) = ( 1 − 2 a b l t + 2 a b l 2 t 2 , 1 , 1 + 2 a b l t + 2 a b l 2 t 2 ) . (Q_1,Q_2,Q_3)=\left(1-2a_{\mathrm{bl}}t+2a_{\mathrm{bl}}^2t^2,\ 1,\ 1+2a_{\mathrm{bl}}t+2a_{\mathrm{bl}}^2t^2\right). ( Q 1 , Q 2 , Q 3 ) = ( 1 − 2 a bl t + 2 a bl 2 t 2 , 1 , 1 + 2 a bl t + 2 a bl 2 t 2 ) .
For example,
tr G + 1 = i ( 1 − a b l ) , tr G − 1 = − i ( 1 − a b l ) , \operatorname{tr}G_+^1=i(1-a_{\mathrm{bl}}),\qquad\operatorname{tr}G_-^1=-i(1-a_{\mathrm{bl}}), tr G + 1 = i ( 1 − a bl ) , tr G − 1 = − i ( 1 − a bl ) ,
and both traces vanish for p = 2 p=2 p = 2 . For p = 3 p=3 p = 3 both traces equal a b l − 1 a_{\mathrm{bl}}-1 a bl − 1 . In particular,
S 3 − S 1 = 16 a b l t − 8 a b l , Q 3 − Q 1 = 4 a b l t . S_3-S_1=16a_{\mathrm{bl}}t-8a_{\mathrm{bl}},\qquad Q_3-Q_1=4a_{\mathrm{bl}}t. S 3 − S 1 = 16 a bl t − 8 a bl , Q 3 − Q 1 = 4 a bl t .
The factorial identity
1 n ! ∫ 0 ∞ t n + q e − 2 t d t = ( n + q ) ! n ! 2 n + q + 1 , n , q ≥ 0 (47) \frac{1}{n!}\int_0^\infty t^{n+q}e^{-2t}\,dt=\frac{(n+q)!}{n!2^{n+q+1}},\qquad n,q\geq0 \tag*{(47)} n ! 1 ∫ 0 ∞ t n + q e − 2 t d t = n ! 2 n + q + 1 ( n + q )! , n , q ≥ 0 (47)
now gives
E 22 = 2 j + 3 ( 2 M j b d 2 j + 1 − V j b d 2 j − 1 ) = 8 M j b d − 16 V j b d . E_{22}=2^{j+3}\left(\frac{2M_j^{\mathrm{bd}}}{2^{j+1}}-\frac{V_j^{\mathrm{bd}}}{2^{j-1}}\right)=8M_j^{\mathrm{bd}}-16V_j^{\mathrm{bd}}. E 22 = 2 j + 3 ( 2 j + 1 2 M j bd − 2 j − 1 V j bd ) = 8 M j bd − 16 V j bd .
and, by the two displayed differences,
E 33 − E 11 = 2 j + 3 [ ( 16 a b l ( j + 1 ) 2 j + 2 − 8 a b l 2 j + 1 ) M j b d − 4 a b l ( j − 1 ) 2 j V j b d ] = 32 a b l [ j M j b d − ( j − 1 ) V j b d ] . \begin{aligned}
E_{33}-E_{11}
&=2^{j+3}\left[\left(\frac{16a_{\mathrm{bl}}(j+1)}{2^{j+2}}-\frac{8a_{\mathrm{bl}}}{2^{j+1}}\right)M_j^{\mathrm{bd}}
-\frac{4a_{\mathrm{bl}}(j-1)}{2^j}V_j^{\mathrm{bd}}\right]\\
&=32a_{\mathrm{bl}}\left[jM_j^{\mathrm{bd}}-(j-1)V_j^{\mathrm{bd}}\right].
\end{aligned} E 33 − E 11 = 2 j + 3 [ ( 2 j + 2 16 a bl ( j + 1 ) − 2 j + 1 8 a bl ) M j bd − 2 j 4 a bl ( j − 1 ) V j bd ] = 32 a bl [ j M j bd − ( j − 1 ) V j bd ] .
The first diagonal polarization gives the separate coefficients. Here [ X ] E 11 [X]E_{11} [ X ] E 11 denotes the coefficient of the formal variable X X X in E 11 E_{11} E 11 , regarded as a linear expression in L j b d , M j b d , V j b d L_j^{\mathrm{bd}},M_j^{\mathrm{bd}},V_j^{\mathrm{bd}} L j bd , M j bd , V j bd :
[ L j b d ] E 11 = 4 ( 1 − a b l ) 2 , [L_j^{\mathrm{bd}}]E_{11}=4(1-a_{\mathrm{bl}})^2, [ L j bd ] E 11 = 4 ( 1 − a bl ) 2 ,
[ M j b d ] E 11 = 8 a b l 2 ( j + 1 ) ( j + 2 ) − 16 a b l ( a b l + 1 ) ( j + 1 ) + 4 ( 3 a b l 2 + 2 a b l + 3 ) = B j , [M_j^{\mathrm{bd}}]E_{11}=8a_{\mathrm{bl}}^2(j+1)(j+2)-16a_{\mathrm{bl}}(a_{\mathrm{bl}}+1)(j+1)+4(3a_{\mathrm{bl}}^2+2a_{\mathrm{bl}}+3)=B_j, [ M j bd ] E 11 = 8 a bl 2 ( j + 1 ) ( j + 2 ) − 16 a bl ( a bl + 1 ) ( j + 1 ) + 4 ( 3 a bl 2 + 2 a bl + 3 ) = B j ,
[ V j b d ] E 11 = 8 a b l 2 j ( j − 1 ) − 16 a b l ( j − 1 ) + 16 = C j . [V_j^{\mathrm{bd}}]E_{11}=8a_{\mathrm{bl}}^2j(j-1)-16a_{\mathrm{bl}}(j-1)+16=C_j. [ V j bd ] E 11 = 8 a bl 2 j ( j − 1 ) − 16 a bl ( j − 1 ) + 16 = C j .
This proves (46) by finite algebra for every integer j ≥ 2 j\geq2 j ≥ 2 , including j = 2 j=2 j = 2 . If all three moments vanish, the first identity yields
M j b d = 2 V j b d . M_j^{\mathrm{bd}}=2V_j^{\mathrm{bd}}. M j bd = 2 V j bd .
The second then gives
32 a b l ( j + 1 ) V j b d = 0 , 32a_{\mathrm{bl}}(j+1)V_j^{\mathrm{bd}}=0, 32 a bl ( j + 1 ) V j bd = 0 ,
so V j b d = M j b d = 0 V_j^{\mathrm{bd}}=M_j^{\mathrm{bd}}=0 V j bd = M j bd = 0 . The third identity now gives
4 ( 1 − a b l ) 2 L j b d = 0. 4(1-a_{\mathrm{bl}})^2L_j^{\mathrm{bd}}=0. 4 ( 1 − a bl ) 2 L j bd = 0.
Since a b l ∈ ( 0 , 1 ) a_{\mathrm{bl}}\in(0,1) a bl ∈ ( 0 , 1 ) , all three unknowns vanish. □
Recovery of the full boundary jets
Recall from (30) that the frozen decaying profile is
F s ( y , t ) = e − c b l ( y ) t ( I 3 − c b l ( y ) t N s ( y ) ) , s ∈ { + , − } . F_{\mathfrak{s}}(y,t)=e^{-c_{\mathrm{bl}}(y)t}\left(I_3-c_{\mathrm{bl}}(y)tN_{\mathfrak{s}}(y)\right),\qquad\mathfrak{s}\in\{+,-\}. F s ( y , t ) = e − c bl ( y ) t ( I 3 − c bl ( y ) t N s ( y ) ) , s ∈ { + , − } .
For fixed boundary amplitudes A ± ∈ C ∞ ( V ; C 3 ) A_{\pm}\in C^\infty(V;\mathbb{C}^3) A ± ∈ C ∞ ( V ; C 3 ) , define the frozen displacement profiles and their leading gradient matrices by
U s = F s A s , G s = i s U s ⊗ τ t a n + ( ∂ t U s ) ⊗ n i n , s ∈ { + , − } . (48) U_{\mathfrak{s}}=F_{\mathfrak{s}}A_{\mathfrak{s}},\qquad G_{\mathfrak{s}}=i\mathfrak{s}U_{\mathfrak{s}}\otimes\tau_{\mathrm{tan}}+(\partial_tU_{\mathfrak{s}})\otimes n_{\mathrm{in}},\qquad\mathfrak{s}\in\{+,-\}. \tag*{(48)} U s = F s A s , G s = i s U s ⊗ τ tan + ( ∂ t U s ) ⊗ n in , s ∈ { + , − } . (48)
The signs + + + and − - − are identified with + 1 +1 + 1 and − 1 -1 − 1 , respectively. The vectors τ t a n \tau_{\mathrm{tan}} τ tan and n i n n_{\mathrm{in}} n in are evaluated at ( y , 0 ) (y,0) ( y , 0 ) as in (29) .
At a fixed boundary point y 0 y_0 y 0 , use the normalized frame
c b l ( y 0 ) = 1 , τ t a n ( y 0 ) = e 1 , n i n ( y 0 ) = e 3 . c_{\mathrm{bl}}(y_0)=1,\qquad\tau_{\mathrm{tan}}(y_0)=e_1,\qquad n_{\mathrm{in}}(y_0)=e_3. c bl ( y 0 ) = 1 , τ tan ( y 0 ) = e 1 , n in ( y 0 ) = e 3 .
Then
F ± ( y 0 , t ) = e − t F ~ ± ( t ) , F ~ ± ( t ) = I 3 − t N ± . F_{\pm}(y_0,t)=e^{-t}\widetilde{F}_{\pm}(t),\qquad\widetilde{F}_{\pm}(t)=I_3-tN_{\pm}. F ± ( y 0 , t ) = e − t F ± ( t ) , F ± ( t ) = I 3 − t N ± .
Writing A ± ( y 0 ) = ∑ p = 1 3 A ± , p ( y 0 ) e p A_{\pm}(y_0)=\sum_{p=1}^{3}A_{\pm,p}(y_0)e_p A ± ( y 0 ) = ∑ p = 1 3 A ± , p ( y 0 ) e p , the general profiles satisfy
U ± ( y 0 , t ) = e − t ∑ p = 1 3 A ± , p ( y 0 ) F ~ ± p ( t ) , U_{\pm}(y_0,t)=e^{-t}\sum_{p=1}^{3}A_{\pm,p}(y_0)\widetilde{F}_{\pm}^{p}(t), U ± ( y 0 , t ) = e − t p = 1 ∑ 3 A ± , p ( y 0 ) F ± p ( t ) ,
G ± ( y 0 , t ) = e − t ∑ p = 1 3 A ± , p ( y 0 ) G ± p ( t ) . G_{\pm}(y_0,t)=e^{-t}\sum_{p=1}^{3}A_{\pm,p}(y_0)G_{\pm}^{p}(t). G ± ( y 0 , t ) = e − t p = 1 ∑ 3 A ± , p ( y 0 ) G ± p ( t ) .
Here F ~ ± p \widetilde{F}_{\pm}^{p} F ± p and G ± p G_{\pm}^{p} G ± p are the columns defined in (41) . These identities hold at the chosen point in the normalized frozen frame; the general profiles in (48) are used on the full boundary patch.
For each integer j ≥ 2 j\ge2 j ≥ 2 , define the leading boundary moment functional associated with A + A_+ A + and A − A_- A − by
M j = ∫ V ∫ 0 ∞ J ( y , 0 ) [ t j j ! { L j b d ( y ) tr G + tr G − + 2 M j b d ( y ) sym G + : sym G − } − t j − 2 ( j − 2 ) ! V j b d ( y ) U + ⋅ U − ] d t d y . (49) \mathcal{M}_j=\int_V\int_0^\infty J(y,0)\left[\frac{t^j}{j!}\left\{L_j^{\mathrm{bd}}(y)\operatorname{tr}G_+\operatorname{tr}G_-+2M_j^{\mathrm{bd}}(y)\operatorname{sym}G_+:\operatorname{sym}G_-\right\}-\frac{t^{j-2}}{(j-2)!}V_j^{\mathrm{bd}}(y)U_+\cdot U_-\right]\,dt\,dy. \tag*{(49)} M j = ∫ V ∫ 0 ∞ J ( y , 0 ) [ j ! t j { L j bd ( y ) tr G + tr G − + 2 M j bd ( y ) sym G + : sym G − } − ( j − 2 )! t j − 2 V j bd ( y ) U + ⋅ U − ] d t d y . (49)
Here U ± U_{\pm} U ± and G ± G_{\pm} G ± are the profiles defined in (48) and are evaluated at ( y , t ) (y,t) ( y , t ) .
Lemma 2.4. Fix an integer j ≥ 2 j\ge2 j ≥ 2 , a compact boundary patch, and amplitudes A + , A − ∈ C c ∞ ( V ; C 3 ) A_+,A_-\in C^\infty_c(V;\mathbb{C}^3) A + , A − ∈ C c ∞ ( V ; C 3 ) . Suppose the two materials have common boundary values of λ , μ \lambda,\mu λ , μ and, uniformly on the collar,
δ 12 λ = r j j ! L j b d + O ( r j + 1 ) , \delta_{12}\lambda=\frac{r^j}{j!}L_j^{\mathrm{bd}}+O(r^{j+1}), δ 12 λ = j ! r j L j bd + O ( r j + 1 ) , δ 12 μ = r j j ! M j b d + O ( r j + 1 ) , \delta_{12}\mu=\frac{r^j}{j!}M_j^{\mathrm{bd}}+O(r^{j+1}), δ 12 μ = j ! r j M j bd + O ( r j + 1 ) , δ 12 W = r j − 2 ( j − 2 ) ! V j b d + O ( r j − 1 ) , \delta_{12}W=\frac{r^{j-2}}{(j-2)!}V_j^{\mathrm{bd}}+O(r^{j-1}), δ 12 W = ( j − 2 )! r j − 2 V j bd + O ( r j − 1 ) , where the boundary Taylor coefficients are defined in (39) . Let N b l ≥ j + 1 N_{\mathrm{bl}}\ge j+1 N bl ≥ j + 1 , and let u 1 , h , u 2 , h u_{1,h},u_{2,h} u 1 , h , u 2 , h be the exact solutions of Proposition 2.2 with phases + , − +,- + , − , amplitudes A + , A − A_+,A_- A + , A − , and a common normal cutoff. Then, with B \mathcal{B} B and M j \mathcal{M}_j M j defined above,
∣ h 1 − j B ( u 1 , h , u 2 , h ) − M j ∣ ≤ C ( h + h N b l − j + h 2 N b l − j ) , 0 < h ≤ 1. (50) \left|h^{1-j}\mathcal{B}(u_{1,h},u_{2,h})-\mathcal{M}_j\right|\le C\left(h+h^{N_{\mathrm{bl}}-j}+h^{2N_{\mathrm{bl}}-j}\right),\qquad0<h\le1. \tag*{(50)} h 1 − j B ( u 1 , h , u 2 , h ) − M j ≤ C ( h + h N bl − j + h 2 N bl − j ) , 0 < h ≤ 1. (50)
For each fixed j j j , N b l N_{\mathrm{bl}} N bl , and pair of amplitudes, the constant depends on the corresponding finite coefficient seminorms, geometry, cutoffs, and the two physical Dirichlet inverse norms
∥ P D , 1 − 1 ∥ H − 1 → H 0 1 , ∥ P D , 2 − 1 ∥ H − 1 → H 0 1 , \left\|P_{D,1}^{-1}\right\|_{H^{-1}\to H_0^1},\qquad\left\|P_{D,2}^{-1}\right\|_{H^{-1}\to H_0^1}, P D , 1 − 1 H − 1 → H 0 1 , P D , 2 − 1 H − 1 → H 0 1 ,
and is independent of h h h .
Proof. Write v 1 , h , v 2 , h v_{1,h},v_{2,h} v 1 , h , v 2 , h for the corresponding approximate layers. For s = + \mathfrak{s}=+ s = + use v 1 , h v_{1,h} v 1 , h and for s = − \mathfrak{s}=- s = − use v 2 , h v_{2,h} v 2 , h , and set in the collar
v ~ s , h ( y , t ) = e − i s y 1 / h v s , h ( X ( y , h t ) ) , \widetilde{v}_{\mathfrak{s},h}(y,t)=e^{-i\mathfrak{s}y_1/h}v_{\mathfrak{s},h}(X(y,ht)), v s , h ( y , t ) = e − i s y 1 / h v s , h ( X ( y , h t )) ,
G ~ s , h ( y , t ) = h e − i s y 1 / h ∇ x v s , h ( X ( y , h t ) ) . \widetilde{G}_{\mathfrak{s},h}(y,t)=he^{-i\mathfrak{s}y_1/h}\nabla_xv_{\mathfrak{s},h}(X(y,ht)). G s , h ( y , t ) = h e − i s y 1 / h ∇ x v s , h ( X ( y , h t )) .
The notation v + , h = v 1 , h v_{+,h}=v_{1,h} v + , h = v 1 , h and v − , h = v 2 , h v_{-,h}=v_{2,h} v − , h = v 2 , h is used in these two definitions only. Formula (19) , the finite profiles, and Taylor’s formula for n i n , w n_{\mathrm{in}},w n in , w give, on 0 ≤ h t ≤ r 0 0\le ht\le r_0 0 ≤ h t ≤ r 0 ,
∣ v ~ s , h − U s ∣ + ∣ G ~ s , h − G s ∣ ≤ C h ( 1 + t ) N e − c 0 t , ∣ v ~ s , h ∣ + ∣ G ~ s , h ∣ + ∣ U s ∣ + ∣ G s ∣ ≤ C ( 1 + t ) N e − c 0 t . (51) \begin{aligned}
\left|\widetilde{v}_{\mathfrak{s},h}-U_{\mathfrak{s}}\right|+\left|\widetilde{G}_{\mathfrak{s},h}-G_{\mathfrak{s}}\right|&\le Ch(1+t)^N e^{-c_0t},\\
\left|\widetilde{v}_{\mathfrak{s},h}\right|+\left|\widetilde{G}_{\mathfrak{s},h}\right|+\left|U_{\mathfrak{s}}\right|+\left|G_{\mathfrak{s}}\right|&\le C(1+t)^N e^{-c_0t}. \tag*{(51)}
\end{aligned} ∣ v s , h − U s ∣ + G s , h − G s ∣ v s , h ∣ + G s , h + ∣ U s ∣ + ∣ G s ∣ ≤ C h ( 1 + t ) N e − c 0 t , ≤ C ( 1 + t ) N e − c 0 t . (51)
Here N N N is finite and may depend on N b l N_{\mathrm{bl}} N bl . To verify the first line, all profiles of index at least one have a factor h h h ; slow tangential derivatives in the scaled gradient have a factor h h h ; and
n i n ( y , h t ) − n i n ( y , 0 ) = O ( h t ) , n_{\mathrm{in}}(y,ht)-n_{\mathrm{in}}(y,0)=O(ht), n in ( y , h t ) − n in ( y , 0 ) = O ( h t ) ,
w 1 ( y , h t ) − τ t a n ( y ) = O ( h t ) . w^1(y,ht)-\tau_{\mathrm{tan}}(y)=O(ht). w 1 ( y , h t ) − τ tan ( y ) = O ( h t ) .
The normal cutoff satisfies ∣ η ( h t ) − 1 ∣ ≤ C h t |\eta(ht)-1|\le Cht ∣ η ( h t ) − 1∣ ≤ C h t , while its derivative in the scaled gradient appears as h η ′ ( h t ) h\eta'(ht) h η ′ ( h t ) . These bounds prove the first line also in the cutoff transition region. The second line follows from the same finite profile bounds.
After the substitution r = h t r=ht r = h t , the scaled elastic coefficient factors are
h − j δ 12 λ ( y , h t ) = t j j ! L j b d ( y ) + O ( h t j + 1 ) , h^{-j}\delta_{12}\lambda(y,ht)=\frac{t^j}{j!}L_j^{\mathrm{bd}}(y)+O(ht^{j+1}), h − j δ 12 λ ( y , h t ) = j ! t j L j bd ( y ) + O ( h t j + 1 ) ,
h − j δ 12 μ ( y , h t ) = t j j ! M j b d ( y ) + O ( h t j + 1 ) , h^{-j}\delta_{12}\mu(y,ht)=\frac{t^j}{j!}M_j^{\mathrm{bd}}(y)+O(ht^{j+1}), h − j δ 12 μ ( y , h t ) = j ! t j M j bd ( y ) + O ( h t j + 1 ) ,
and the potential factor is
h 2 − j δ 12 W ( y , h t ) = t j − 2 ( j − 2 ) ! V j b d ( y ) + O ( h t j − 1 ) . h^{2-j}\delta_{12}W(y,ht)=\frac{t^{j-2}}{(j-2)!}V_j^{\mathrm{bd}}(y)+O(ht^{j-1}). h 2 − j δ 12 W ( y , h t ) = ( j − 2 )! t j − 2 V j bd ( y ) + O ( h t j − 1 ) .
Moreover, J ( y , h t ) − J ( y , 0 ) = O ( h t ) J(y,ht)-J(y,0)=O(ht) J ( y , h t ) − J ( y , 0 ) = O ( h t ) . The signs of the two phases are opposite, so their product is exactly one. Using (51) in each bilinear product bounds the difference between the scaled approximate-layer integrand and the integrand in (49) by C h ( 1 + t ) N ′ e − 2 c 0 t Ch(1+t)^{N'}e^{-2c_0t} C h ( 1 + t ) N ′ e − 2 c 0 t , for a fixed finite N ′ N' N ′ . The leading integral over t > r 0 / h t>r_0/h t > r 0 / h is smaller than every prescribed power of h h h , since its integrand has polynomial-exponential decay. Integration yields
∣ h 1 − j B ( v 1 , h , v 2 , h ) − M j ∣ ≤ C h . \left|h^{1-j}\mathcal{B}(v_{1,h},v_{2,h})-\mathcal{M}_j\right|\le Ch. h 1 − j B ( v 1 , h , v 2 , h ) − M j ≤ C h .
Finally, boundedness of the coefficient differences gives
∣ B ( v 1 , v 2 ) ∣ ≤ C ∥ v 1 ∥ H 1 ∥ v 2 ∥ H 1 . |\mathcal{B}(v_1,v_2)|\le C\|v_1\|_{H^1}\|v_2\|_{H^1}. ∣ B ( v 1 , v 2 ) ∣ ≤ C ∥ v 1 ∥ H 1 ∥ v 2 ∥ H 1 .
Expanding the difference of the two bilinear forms and using (34) –(35) gives
h 1 − j ∣ B ( u 1 , h , u 2 , h ) − B ( v 1 , h , v 2 , h ) ∣ ≤ C h 1 − j ( h N b l − 1 / 2 h − 1 / 2 + h 2 N b l − 1 ) = C ( h N b l − j + h 2 N b l − j ) . \begin{aligned}
h^{1-j}\left|\mathcal{B}(u_{1,h},u_{2,h})-\mathcal{B}(v_{1,h},v_{2,h})\right|
&\le Ch^{1-j}\left(h^{N_{\mathrm{bl}}-1/2}h^{-1/2}+h^{2N_{\mathrm{bl}}-1}\right)\\
&=C\left(h^{N_{\mathrm{bl}}-j}+h^{2N_{\mathrm{bl}}-j}\right).
\end{aligned} h 1 − j ∣ B ( u 1 , h , u 2 , h ) − B ( v 1 , h , v 2 , h ) ∣ ≤ C h 1 − j ( h N bl − 1/2 h − 1/2 + h 2 N bl − 1 ) = C ( h N bl − j + h 2 N bl − j ) .
This estimate holds on all of Ω \Omega Ω , including the region outside the coordinate collar. Combining the last two bounds proves (50) . □
Theorem 2.5. Let the two coefficient triples satisfy the standing smoothness and strong-convexity assumptions, and suppose that the corresponding Dirichlet problems satisfy the nonresonance condition (6) . Assume that their full displacement-to-outward-traction maps defined in (8) agree on ∂ Ω \partial\Omega ∂ Ω . Then, in every inward normal coordinate chart, for every y y y on the boundary patch and every q ∈ N 0 q \in\mathbb{N}_0 q ∈ N 0 ,
∂ r q ( λ 1 − λ 2 ) ( y , 0 ) = ∂ r q ( μ 1 − μ 2 ) ( y , 0 ) = ∂ r q ( W 1 − W 2 ) ( y , 0 ) = 0. \partial_r^q(\lambda_1-\lambda_2)(y,0)=\partial_r^q(\mu_1-\mu_2)(y,0)=\partial_r^q(W_1-W_2)(y,0)=0. ∂ r q ( λ 1 − λ 2 ) ( y , 0 ) = ∂ r q ( μ 1 − μ 2 ) ( y , 0 ) = ∂ r q ( W 1 − W 2 ) ( y , 0 ) = 0. Consequently, for every tangential multi-index α ∈ N 0 2 \alpha\in\mathbb{N}_0^2 α ∈ N 0 2 ,
∂ y α ∂ r q ( λ 1 − λ 2 ) ( y , 0 ) = 0 , \partial_y^\alpha\partial_r^q(\lambda_1-\lambda_2)(y,0)=0, ∂ y α ∂ r q ( λ 1 − λ 2 ) ( y , 0 ) = 0 , ∂ y α ∂ r q ( μ 1 − μ 2 ) ( y , 0 ) = 0 , \partial_y^\alpha\partial_r^q(\mu_1-\mu_2)(y,0)=0, ∂ y α ∂ r q ( μ 1 − μ 2 ) ( y , 0 ) = 0 , ∂ y α ∂ r q ( W 1 − W 2 ) ( y , 0 ) = 0. \partial_y^\alpha\partial_r^q(W_1-W_2)(y,0)=0. ∂ y α ∂ r q ( W 1 − W 2 ) ( y , 0 ) = 0. If W j = ω 2 ρ j W_j=\omega^2\rho_j W j = ω 2 ρ j for a fixed known ω ≠ 0 \omega\ne0 ω = 0 , then the same boundary-jet equality holds for ρ 1 \rho_1 ρ 1 and ρ 2 \rho_2 ρ 2 .
Proof. Equations (38) and (44) recover the zeroth and first normal derivatives of the Lamé parameters. Let j ≥ 2 j\ge2 j ≥ 2 be the elastic Taylor order. Assume inductively that the Lamé derivatives agree for 0 ≤ q < j 0\le q<j 0 ≤ q < j and the potential derivatives agree for 0 ≤ q < j − 2 0\le q<j-2 0 ≤ q < j − 2 . Use the notation L j b d , M j b d , V j b d L_j^{\mathrm{bd}},M_j^{\mathrm{bd}},V_j^{\mathrm{bd}} L j bd , M j bd , V j bd from (39) for the current induction step. Taylor’s formula (25) then gives, uniformly on the fixed compact patch,
δ 12 λ ( y , r ) = r j j ! L j b d ( y ) + O ( r j + 1 ) , δ 12 μ ( y , r ) = r j j ! M j b d ( y ) + O ( r j + 1 ) , \delta_{12}\lambda(y,r)=\frac{r^j}{j!}L_j^{\mathrm{bd}}(y)+O(r^{j+1}),\qquad\delta_{12}\mu(y,r)=\frac{r^j}{j!}M_j^{\mathrm{bd}}(y)+O(r^{j+1}), δ 12 λ ( y , r ) = j ! r j L j bd ( y ) + O ( r j + 1 ) , δ 12 μ ( y , r ) = j ! r j M j bd ( y ) + O ( r j + 1 ) ,
δ 12 W ( y , r ) = r j − 2 ( j − 2 ) ! V j b d ( y ) + O ( r j − 1 ) . \delta_{12}W(y,r)=\frac{r^{j-2}}{(j-2)!}V_j^{\mathrm{bd}}(y)+O(r^{j-1}). δ 12 W ( y , r ) = ( j − 2 )! r j − 2 V j bd ( y ) + O ( r j − 1 ) .
For j = 2 j=2 j = 2 , the assumption on lower-order potential derivatives is empty. Fix N b l ≥ j + 1 N_{\mathrm{bl}}\ge j+1 N bl ≥ j + 1 . Lemma 2.4 applies to every pair of fixed tangential amplitudes. In particular, its global correction bound can still be written with k = h − 1 k=h^{-1} k = h − 1 as
C k j − 1 ( k 1 / 2 − N b l k 1 / 2 + k 1 − 2 N b l ) = C ( k j − N b l + k j − 2 N b l ) = o ( 1 ) . (52) Ck^{j-1}\left(k^{1/2-N_{\mathrm{bl}}}k^{1/2}+k^{1-2N_{\mathrm{bl}}}\right)=C\left(k^{j-N_{\mathrm{bl}}}+k^{j-2N_{\mathrm{bl}}}\right)=o(1). \tag*{(52)} C k j − 1 ( k 1/2 − N bl k 1/2 + k 1 − 2 N bl ) = C ( k j − N bl + k j − 2 N bl ) = o ( 1 ) . (52)
The additional approximate-layer error is O ( h ) O(h) O ( h ) by (50) . Thus the identity (40) gives M j = 0 \mathcal{M}_j=0 M j = 0 for every fixed pair of amplitudes.
Choose A + = A − = χ e p A_+=A_-=\chi e_p A + = A − = χ e p with a real χ ∈ C c ∞ ( V ) \chi\in C^\infty_c(V) χ ∈ C c ∞ ( V ) . The leading functional is the integral of χ 2 J ( y , 0 ) \chi^2J(y,0) χ 2 J ( y , 0 ) times a continuous pointwise diagonal moment. If that moment were nonzero at some point, multiplication by a fixed complex number would make its real part strictly positive on a smaller neighborhood. A nonzero real cutoff supported there would contradict M j = 0 \mathcal{M}_j=0 M j = 0 . Hence each pointwise moment is zero. For each fixed cutoff, the limit h → 0 h\to0 h → 0 is taken first; pointwise localization is then performed on the resulting moment identity.
At the chosen point set c b l = 1 c_{\mathrm{bl}}=1 c bl = 1 , e b l = e 1 e_{\mathrm{bl}}=e_1 e bl = e 1 , n i n = e 3 n_{\mathrm{in}}=e_3 n in = e 3 . The limiting gradient and displacement factors are precisely G ± G^\pm G ± and F e p ± F^\pm_{e_p} F e p ± in Lemma 2.3: the rescaled normal derivative is ( ∂ t − 1 ) F e p ± = ( ( t − 1 ) N ± − I 3 ) (\partial_t-1)F^\pm_{e_p}=((t-1)N_\pm-I_3) ( ∂ t − 1 ) F e p ± = (( t − 1 ) N ± − I 3 ) , and the rescaled tangential derivative is ± i F e p ± \pm iF^\pm_{e_p} ± i F e p ± . The localization argument therefore gives E 11 = E 22 = E 33 = 0 E_{11}=E_{22}=E_{33}=0 E 11 = E 22 = E 33 = 0 , where E p p E_{pp} E pp is defined in (45) using the boundary Taylor coefficients from (39) . Lemma 2.3 proves that this finite linear system has only the zero solution. This advances the Lamé jets through order j j j and the potential jet through order j − 2 j-2 j − 2 , closing the induction. In particular, the q q q th normal derivative of W W W is recovered at elastic order j = q + 2 j=q+2 j = q + 2 . For each fixed order j j j , the induction uses a finite truncation N b l ≥ j + 1 N_{\mathrm{bl}}\ge j+1 N bl ≥ j + 1 , and the limit h → 0 h\to0 h → 0 is taken at that order.
Tangential differentiation of the recovered boundary identities gives every mixed derivative and hence the full boundary jet. Applying the construction on every compact patch and every boundary component proves the assertion. The density follows by division by ω 2 \omega^2 ω 2 . □
A common exterior extension
By Theorem 2.5, the two coefficient triples have identical boundary Taylor data. In every inward normal coordinate chart,
∂ y α ∂ r q ( λ 1 − λ 2 ) ( y , 0 ) = 0 , \partial_y^\alpha\partial_r^q(\lambda_1-\lambda_2)(y,0)=0, ∂ y α ∂ r q ( λ 1 − λ 2 ) ( y , 0 ) = 0 ,
∂ y α ∂ r q ( μ 1 − μ 2 ) ( y , 0 ) = 0 , \partial_y^\alpha\partial_r^q(\mu_1-\mu_2)(y,0)=0, ∂ y α ∂ r q ( μ 1 − μ 2 ) ( y , 0 ) = 0 ,
∂ y α ∂ r q ( W 1 − W 2 ) ( y , 0 ) = 0 , α ∈ N 0 2 , q ∈ N 0 . \partial_y^\alpha\partial_r^q(W_1-W_2)(y,0)=0,\qquad\alpha\in\mathbb{N}_0^2,\quad q\in\mathbb{N}_0. ∂ y α ∂ r q ( W 1 − W 2 ) ( y , 0 ) = 0 , α ∈ N 0 2 , q ∈ N 0 .
We use these identities to construct smooth extensions that agree outside Ω ‾ \overline{\Omega} Ω . Common exterior extensions are also used in inverse boundary problems; see [8 Section 6] . We first establish a zero-extension result for functions with vanishing boundary jets.
Lemma 2.6. Let f ∈ C ∞ ( Ω ‾ ) f\in C^\infty(\overline{\Omega}) f ∈ C ∞ ( Ω ) satisfy
∂ α f ∣ ∂ Ω = 0 for every α ∈ N 0 3 . \partial^\alpha f|_{\partial\Omega}=0\qquad\text{for every }\alpha\in\mathbb{N}_0^3. ∂ α f ∣ ∂ Ω = 0 for every α ∈ N 0 3 . Define
( E 0 f ) ( x ) = { f ( x ) , x ∈ Ω , 0 , x ∉ Ω . (\mathcal{E}_0f)(x)=
\begin{cases}
f(x), & x\in\Omega,\\
0, & x\notin\Omega.
\end{cases} ( E 0 f ) ( x ) = { f ( x ) , 0 , x ∈ Ω , x ∈ / Ω. Then
E 0 f ∈ C c ∞ ( R 3 ) , supp ( E 0 f ) ⊂ Ω ‾ . \mathcal{E}_0f\in C^\infty_c(\mathbb{R}^3),\qquad\operatorname{supp}(\mathcal{E}_0f)\subset\overline{\Omega}. E 0 f ∈ C c ∞ ( R 3 ) , supp ( E 0 f ) ⊂ Ω .
Proof. In each chart (12) , Taylor’s formula in the normal variable gives, for every multi-index α \alpha α and integer N ≥ 1 N\ge1 N ≥ 1 ,
( ∂ α f ) ( X ( y , r ) ) = r N ( N − 1 ) ! ∫ 0 1 ( 1 − s ) N − 1 ∂ r N [ ( ∂ α f ) ∘ X ] ( y , s r ) d s . (\partial^\alpha f)(X(y,r))=\frac{r^N}{(N-1)!}\int_0^1(1-s)^{N-1}\partial_r^N[(\partial^\alpha f)\circ X](y,sr)\,\mathrm{d}s. ( ∂ α f ) ( X ( y , r )) = ( N − 1 )! r N ∫ 0 1 ( 1 − s ) N − 1 ∂ r N [( ∂ α f ) ∘ X ] ( y , sr ) d s .
Consequently, on every compact subpatch,
∣ ( ∂ α f ) ( X ( y , r ) ) ∣ ≤ C α , N r N . |(\partial^\alpha f)(X(y,r))|\le C_{\alpha,N}r^N. ∣ ( ∂ α f ) ( X ( y , r )) ∣ ≤ C α , N r N .
Thus every derivative has zero trace on the boundary. Integration by parts and induction on ∣ α ∣ |\alpha| ∣ α ∣ yield
∂ α ( E 0 f ) = E 0 ( ∂ α f ) in D ′ ( R 3 ) . \partial^\alpha(\mathcal{E}_0f)=\mathcal{E}_0(\partial^\alpha f)\quad\text{in }\mathcal{D}'(\mathbb{R}^3). ∂ α ( E 0 f ) = E 0 ( ∂ α f ) in D ′ ( R 3 ) .
The right-hand side is continuous across the boundary for every α \alpha α , so E 0 f \mathcal{E}_0f E 0 f is smooth. A finite chart cover handles all boundary components, and boundedness of Ω \Omega Ω gives compact support. □ \square □
Proposition 2.7. Let ( λ j , μ j , W j ) (\lambda_j,\mu_j,W_j) ( λ j , μ j , W j ) , j = 1 , 2 j=1,2 j = 1 , 2 , be the two smooth coefficient triples satisfying the standing strong-convexity assumptions. Suppose that they have identical boundary Taylor data on every boundary component. Then they admit smooth extensions to R 3 \mathbb{R}^3 R 3 that agree on R 3 ∖ Ω ‾ \mathbb{R}^3\setminus\overline{\Omega} R 3 ∖ Ω , preserve strong convexity of the Lamé pairs, and equal the same constant strongly convex Lamé pair with zero potentials outside a compact set.
Proof. We first construct a smooth extension operator using the classical reflection construction underlying Seeley’s theorem [18 ] . For extension operators with stronger mapping properties, see [13 ] .
Put b j = 2 j b_j=2^j b j = 2 j for j ≥ 0 j\ge0 j ≥ 0 . For 0 ≤ j ≤ N 0\le j\le N 0 ≤ j ≤ N , define
a j , N = ∏ 0 ≤ k ≤ N k ≠ j 1 + b k b k − b j , a j , N = 0 ( j > N ) . (53) a_{j,N}=\prod_{\substack{0\le k\le N\\k\ne j}}\frac{1+b_k}{b_k-b_j},\qquad a_{j,N}=0\quad(j>N). \tag*{(53)} a j , N = 0 ≤ k ≤ N k = j ∏ b k − b j 1 + b k , a j , N = 0 ( j > N ) . (53)
Lagrange interpolation at − b 0 , … , − b N -b_0,\ldots,-b_N − b 0 , … , − b N , evaluated at 1 1 1 , gives
∑ j = 0 N a j , N ( − b j ) p = 1 , 0 ≤ p ≤ N . \sum_{j=0}^{N}a_{j,N}(-b_j)^p=1,\qquad0\le p\le N. j = 0 ∑ N a j , N ( − b j ) p = 1 , 0 ≤ p ≤ N .
For each fixed j ≥ 0 j \ge0 j ≥ 0 , define
a j : = lim N → ∞ a j , N . a_j := \lim_{N \to\infty} a_{j,N}. a j := N → ∞ lim a j , N .
The limit exists because the factors with k > j k > j k > j are
1 + 2 − k 1 − 2 j − k . \frac{1+2^{-k}}{1-2^{j-k}}. 1 − 2 j − k 1 + 2 − k .
Splitting the product at j j j gives
∣ a j , N ∣ ≤ C 2 − j ( j + 1 ) / 2 , C = ∏ k ≥ 0 ( 1 + 2 − k ) ∏ ℓ ≥ 1 ( 1 − 2 − ℓ ) − 2 < ∞ . |a_{j,N}| \le C 2^{-j(j+1)/2}, \qquad C=\prod_{k\ge0}(1+2^{-k})\prod_{\ell\ge1}(1-2^{-\ell})^{-2}<\infty. ∣ a j , N ∣ ≤ C 2 − j ( j + 1 ) /2 , C = k ≥ 0 ∏ ( 1 + 2 − k ) ℓ ≥ 1 ∏ ( 1 − 2 − ℓ ) − 2 < ∞.
For every fixed integer p ≥ 0 p\ge0 p ≥ 0 , the bound is summable after multiplication by b j p b_j^p b j p . Dominated convergence therefore gives
∑ j ≥ 0 a j ( − b j ) p = 1 , ∑ j ≥ 0 ∣ a j ∣ b j p < ∞ ( p ≥ 0 ) . (54) \sum_{j\ge0}a_j(-b_j)^p=1,\qquad\sum_{j\ge0}|a_j|b_j^p<\infty\quad(p\ge0). \tag*{(54)} j ≥ 0 ∑ a j ( − b j ) p = 1 , j ≥ 0 ∑ ∣ a j ∣ b j p < ∞ ( p ≥ 0 ) . (54)
Use the inward coordinates (12) . Choose a finite smooth partition of unity near Ω ‾ \overline{\Omega} Ω , with its interior term compactly supported in Ω \Omega Ω and each boundary term η \eta η supported strictly inside a smaller coordinate box. For a boundary term, write
F ( y , r ) = ( η f ) ( X ( y , r ) ) , r ≥ 0. F(y,r)=(\eta f)(X(y,r)),\qquad r\ge0. F ( y , r ) = ( η f ) ( X ( y , r )) , r ≥ 0.
Choose a fixed smooth normal cutoff χ n \chi_n χ n , equal to one near 0 0 0 and supported in [ 0 , ε ) [0,\varepsilon) [ 0 , ε ) on the nonnegative half-axis, where ε \varepsilon ε is smaller than the collar width. Set
g ( y , s ) = χ n ( s ) F ( y , s ) , s ≥ 0 , g(y,s)=\chi_n(s)F(y,s),\qquad s\ge0, g ( y , s ) = χ n ( s ) F ( y , s ) , s ≥ 0 ,
and extend g g g by zero for s ≥ ε s\ge\varepsilon s ≥ ε . Define
( E l o c F ) ( y , r ) = { F ( y , r ) , r ≥ 0 , ∑ j ≥ 0 a j g ( y , − b j r ) , r < 0. (55) (\mathcal{E}_{\mathrm{loc}}F)(y,r)=
\begin{cases}
F(y,r), & r\ge0,\\
\sum_{j\ge0}a_jg(y,-b_jr), & r<0.
\end{cases} \tag*{(55)} ( E loc F ) ( y , r ) = { F ( y , r ) , ∑ j ≥ 0 a j g ( y , − b j r ) , r ≥ 0 , r < 0. (55)
For each fixed r < 0 r<0 r < 0 , the sum is finite. By (54) , every differentiated series converges uniformly up to r = 0 r=0 r = 0 on compact tangential patches, and
lim r ↑ 0 ∂ y α ∂ r p ( E l o c F ) ( y , r ) = ∑ j ≥ 0 a j ( − b j ) p ∂ y α ∂ r p F ( y , 0 ) = ∂ y α ∂ r p F ( y , 0 ) . \lim_{r\uparrow0}\partial_y^\alpha\partial_r^p(\mathcal{E}_{\mathrm{loc}}F)(y,r)
=\sum_{j\ge0}a_j(-b_j)^p\partial_y^\alpha\partial_r^pF(y,0)
=\partial_y^\alpha\partial_r^pF(y,0). r ↑ 0 lim ∂ y α ∂ r p ( E loc F ) ( y , r ) = j ≥ 0 ∑ a j ( − b j ) p ∂ y α ∂ r p F ( y , 0 ) = ∂ y α ∂ r p F ( y , 0 ) .
Thus the two pieces join smoothly. Since b j ≥ 1 b_j\ge1 b j ≥ 1 , the exterior piece vanishes for − r ≥ ε -r\ge\varepsilon − r ≥ ε . The fixed tangential supports and a larger coordinate box permit smooth extension by zero at all chart edges.
All partitions and cutoffs are fixed independently of f f f and of the derivative order. Adding the finitely many chartwise extensions and the zero extension of the interior term gives a real linear operator
E : C ∞ ( Ω ‾ ; R ) ⟶ C ∞ ( R 3 ; R ) , ( E f ) ∣ Ω ‾ = f . (56) E:C^\infty(\overline{\Omega};\mathbb{R})\longrightarrow C^\infty(\mathbb{R}^3;\mathbb{R}),\qquad(Ef)|_{\overline{\Omega}}=f. \tag*{(56)} E : C ∞ ( Ω ; R ) ⟶ C ∞ ( R 3 ; R ) , ( E f ) ∣ Ω = f . (56)
Put
λ 1 e = E λ 1 , μ 1 e = E μ 1 , W 1 e = E W 1 . \lambda_1^e=E\lambda_1,\qquad\mu_1^e=E\mu_1,\qquad W_1^e=EW_1. λ 1 e = E λ 1 , μ 1 e = E μ 1 , W 1 e = E W 1 .
Strong convexity and compactness of Ω ‾ \overline{\Omega} Ω give a relatively compact open neighborhood U ⊃ Ω ‾ U\supset\overline{\Omega} U ⊃ Ω and a constant c ∗ > 0 c_*>0 c ∗ > 0 such that
μ 1 e ≥ c ∗ , 3 λ 1 e + 2 μ 1 e ≥ c ∗ on U . \mu_1^e\ge c_*,\qquad3\lambda_1^e+2\mu_1^e\ge c_*\quad\text{on }U. μ 1 e ≥ c ∗ , 3 λ 1 e + 2 μ 1 e ≥ c ∗ on U .
Choose θ ∈ C c ∞ ( U ) \theta\in C^\infty_c(U) θ ∈ C c ∞ ( U ) with 0 ≤ θ ≤ 1 0 \le\theta\le1 0 ≤ θ ≤ 1 and θ = 1 \theta= 1 θ = 1 near Ω ‾ \overline{\Omega} Ω . Choose constants λ ∗ , μ ∗ \lambda_*,\mu_* λ ∗ , μ ∗ such that
μ ∗ > 0 , 3 λ ∗ + 2 μ ∗ > 0. \mu_* > 0,\qquad3\lambda_* + 2\mu_* > 0. μ ∗ > 0 , 3 λ ∗ + 2 μ ∗ > 0.
Define
λ ~ 1 = θ λ 1 e + ( 1 − θ ) λ ∗ , μ ~ 1 = θ μ 1 e + ( 1 − θ ) μ ∗ , W ~ 1 = θ W 1 e . (57) \begin{aligned}
\widetilde{\lambda}_1 &= \theta\lambda^e_1 + (1-\theta)\lambda_*,\\
\widetilde{\mu}_1 &= \theta\mu^e_1 + (1-\theta)\mu_*,\\
\widetilde{W}_1 &= \theta W^e_1. \tag*{(57)}
\end{aligned} λ 1 μ 1 W 1 = θ λ 1 e + ( 1 − θ ) λ ∗ , = θ μ 1 e + ( 1 − θ ) μ ∗ , = θ W 1 e . (57)
These functions agree with the original coefficients on Ω ‾ \overline{\Omega} Ω . Moreover,
μ ~ 1 ≥ min ( c ∗ , μ ∗ ) > 0 , \widetilde{\mu}_1 \ge\min(c_*,\mu_*) > 0, μ 1 ≥ min ( c ∗ , μ ∗ ) > 0 ,
3 λ ~ 1 + 2 μ ~ 1 ≥ min ( c ∗ , 3 λ ∗ + 2 μ ∗ ) > 0. 3\widetilde{\lambda}_1 + 2\widetilde{\mu}_1 \ge\min(c_*,3\lambda_* + 2\mu_*) > 0. 3 λ 1 + 2 μ 1 ≥ min ( c ∗ , 3 λ ∗ + 2 μ ∗ ) > 0.
Thus the cutoff interpolation preserves strong convexity, and
( λ ~ 1 , μ ~ 1 , W ~ 1 ) = ( λ ∗ , μ ∗ , 0 ) (\widetilde{\lambda}_1,\widetilde{\mu}_1,\widetilde{W}_1) = (\lambda_*,\mu_*,0) ( λ 1 , μ 1 , W 1 ) = ( λ ∗ , μ ∗ , 0 )
outside supp θ \operatorname{supp}\theta supp θ .
For f ∈ { λ , μ , W } f \in\{\lambda,\mu,W\} f ∈ { λ , μ , W } , the difference f 2 − f 1 f_2-f_1 f 2 − f 1 has zero boundary jets by Theorem 2.5 . Lemma 2.6 therefore permits the definition
f ~ 2 = f ~ 1 + E 0 ( f 2 − f 1 ) . (58) \widetilde{f}_2 = \widetilde{f}_1 + \mathcal{E}_0(f_2-f_1). \tag*{(58)} f 2 = f 1 + E 0 ( f 2 − f 1 ) . (58)
Consequently,
f ~ j ∣ Ω ‾ = f j , j = 1 , 2 , \left.\widetilde{f}_j\right|_{\overline{\Omega}} = f_j,\qquad j=1,2, f j Ω = f j , j = 1 , 2 ,
and
f ~ 2 = f ~ 1 on R 3 ∖ Ω , supp ( f ~ 2 − f ~ 1 ) ⊂ Ω ‾ . \widetilde{f}_2 = \widetilde{f}_1\quad\text{on }\mathbb{R}^3\setminus\Omega,\qquad
\operatorname{supp}(\widetilde{f}_2-\widetilde{f}_1)\subset\overline{\Omega}. f 2 = f 1 on R 3 ∖ Ω , supp ( f 2 − f 1 ) ⊂ Ω .
The second Lamé pair is strongly convex in Ω \Omega Ω by hypothesis, and outside Ω \Omega Ω it agrees with the strongly convex first extension. Smoothness across the boundary follows from (2.47). Hence both extended triples have the same exterior values and equal ( λ ∗ , μ ∗ , 0 ) (\lambda_*,\mu_*,0) ( λ ∗ , μ ∗ , 0 ) outside a compact set. Their differences, together with all derivatives of those differences, are supported in Ω ‾ \overline{\Omega} Ω . Thus integrals involving these differences may be evaluated on any larger construction domain. □
In the subsequent auxiliary constructions, we use λ j , μ j , W j \lambda_j,\mu_j,W_j λ j , μ j , W j also for the smooth extensions λ ~ j , μ ~ j , W ~ j \widetilde{\lambda}_j,\widetilde{\mu}_j,\widetilde{W}_j λ j , μ j , W j from Proposition 2.7 ; their restrictions to Ω \Omega Ω are the original physical coefficients. All boundary maps and Dirichlet problems continue to refer to the original domain Ω \Omega Ω .
The reduced Green–Betti identity
In this section, we derive a Green–Betti identity for the four-component system (63) using only the Dirichlet-to-Neumann maps. We first compute the boundary terms introduced by the reduction, including the contribution of the auxiliary scalar component, and then prove the cancellation of the corresponding boundary terms under the matching conditions (84) , which follow from the boundary determination in § 2 . The resulting identity (85) expresses an interior pairing involving the difference of the reduced operators in terms of the difference of the boundary maps. The common exterior extension allows the volume integral to be taken over a larger construction domain, while the boundary pairing remains on ∂ Ω \partial\Omega ∂ Ω .
For the reduced operators M j W j M_j^{W_j} M j W j defined in (63) , set
D W : = M 1 W 1 − M 2 W 2 . (59) D^W := M_1^{W_1} - M_2^{W_2}. \tag*{(59)} D W := M 1 W 1 − M 2 W 2 . (59)
The order of subtraction in (59) fixes the signs in the reduced Green–Betti identity (75) . Throughout the section, ν \nu ν denotes the outward unit normal to ∂ Ω \partial\Omega ∂ Ω , and we use the bilinear pairing and transpose convention fixed in § 1.6 .
The four-component reduction
Let Ω ⊂ R 3 \Omega\subset\mathbb{R}^{3} Ω ⊂ R 3 be a bounded domain with smooth boundary. For j = 1 , 2 j=1,2 j = 1 , 2 , consider smooth real coefficient triples ( λ j , μ j , W j ) (\lambda_j,\mu_j,W_j) ( λ j , μ j , W j ) on Ω ‾ \overline{\Omega} Ω satisfying μ j > 0 \mu_j>0 μ j > 0 and 3 λ j + 2 μ j > 0 3\lambda_j+2\mu_j>0 3 λ j + 2 μ j > 0 . Set
Σ j = λ j + 2 μ j , κ j = λ j + μ j Σ j , a j = μ j − 1 , c j = μ j − 1 / 2 , ℓ j = log μ j , m j = ∇ ℓ j . (60) \Sigma_j=\lambda_j+2\mu_j,\qquad\kappa_j=\frac{\lambda_j+\mu_j}{\Sigma_j},\qquad a_j=\mu_j^{-1},\qquad c_j=\mu_j^{-1/2},\qquad\ell_j=\log\mu_j,\qquad\mathbf{m}_j=\nabla\ell_j. \tag*{(60)} Σ j = λ j + 2 μ j , κ j = Σ j λ j + μ j , a j = μ j − 1 , c j = μ j − 1/2 , ℓ j = log μ j , m j = ∇ ℓ j . (60)
The strong convexity conditions imply Σ j > 0 \Sigma_j>0 Σ j > 0 , 1 / 4 < κ j < 1 1/4<\kappa_j<1 1/4 < κ j < 1 , and 1 − κ j = μ j / Σ j 1-\kappa_j=\mu_j/\Sigma_j 1 − κ j = μ j / Σ j . Equivalently, λ j = μ j ( 2 κ j − 1 ) / ( 1 − κ j ) \lambda_j=\mu_j(2\kappa_j-1)/(1-\kappa_j) λ j = μ j ( 2 κ j − 1 ) / ( 1 − κ j ) . Let ν \nu ν denote the outward unit normal to ∂ Ω \partial\Omega ∂ Ω . With the convention ( ∇ u ) i k = ∂ k u i (\nabla u)_{ik}=\partial_k u_i ( ∇ u ) ik = ∂ k u i , define the stress, elasticity operator, and traction by
σ j ( u ) = 2 μ j ε ( u ) + λ j div u I 3 , L j u : = L λ j , μ j u = div σ j ( u ) , t j ( u ) = σ j ( u ) ν . (61) \sigma_j(u)=2\mu_j\varepsilon(u)+\lambda_j\operatorname{div}u\,I_3,
\qquad
L_j u:=L_{\lambda_j,\mu_j}u=\operatorname{div}\sigma_j(u),\qquad
t_j(u)=\sigma_j(u)\nu. \tag*{(61)} σ j ( u ) = 2 μ j ε ( u ) + λ j div u I 3 , L j u := L λ j , μ j u = div σ j ( u ) , t j ( u ) = σ j ( u ) ν . (61)
The full elastic operator associated with the coefficient triple ( λ j , μ j , W j ) (\lambda_j,\mu_j,W_j) ( λ j , μ j , W j ) is L j + W j I 3 L_j+W_jI_3 L j + W j I 3 .
For three-component vector fields v , p v,p v , p and scalar fields f , s f,s f , s , write z = ( v , f ) T z=(v,f)^{T} z = ( v , f ) T and introduce the displacement map T j T_j T j and the source map S j S_j S j :
T j z = c j v + a j ∇ f − f ∇ a j , S j ( p , s ) = c j − 1 p + ∇ ( s 1 − κ j ) . (62) T_jz=c_jv+a_j\nabla f-f\nabla a_j,\qquad
S_j(p,s)=c_j^{-1}p+\nabla\left(\frac{s}{1-\kappa_j}\right). \tag*{(62)} T j z = c j v + a j ∇ f − f ∇ a j , S j ( p , s ) = c j − 1 p + ∇ ( 1 − κ j s ) . (62)
For j = 1 , 2 j=1,2 j = 1 , 2 , define the coefficients and the four-component reduced operator by
F μ j = 2 c j { ∇ 2 ℓ j − m j ⊗ m j + ( ∣ m j ∣ 2 − Δ ℓ j ) I 3 } , F_{\mu_j}=2c_j\left\{\nabla^2\ell_j-\mathbf{m}_j\otimes\mathbf{m}_j+\left(\lvert\mathbf{m}_j\rvert^2-\Delta\ell_j\right)I_3\right\}, F μ j = 2 c j { ∇ 2 ℓ j − m j ⊗ m j + ( ∣ m j ∣ 2 − Δ ℓ j ) I 3 } ,
V μ j = − ∇ 2 ℓ j − 1 2 m j ⊗ m j − ( 1 2 Δ ℓ j + 1 4 ∣ m j ∣ 2 ) I 3 , V_{\mu_j}=-\nabla^2\ell_j-\frac{1}{2}\mathbf{m}_j\otimes\mathbf{m}_j-\left(\frac{1}{2}\Delta\ell_j+\frac{1}{4}\lvert\mathbf{m}_j\rvert^2\right)I_3, V μ j = − ∇ 2 ℓ j − 2 1 m j ⊗ m j − ( 2 1 Δ ℓ j + 4 1 ∣ m j ∣ 2 ) I 3 ,
q μ j = Δ ℓ j − ∣ m j ∣ 2 , d j = κ j μ j , e j r e d = 1 2 ( 3 κ j − 2 ) μ j , q_{\mu_j}=\Delta\ell_j-\lvert\mathbf{m}_j\rvert^2,\qquad
d_j=\kappa_j\sqrt{\mu_j},\qquad
e_j^{\mathrm{red}}=\frac{1}{2}(3\kappa_j-2)\sqrt{\mu_j}, q μ j = Δ ℓ j − ∣ m j ∣ 2 , d j = κ j μ j , e j red = 2 1 ( 3 κ j − 2 ) μ j ,
M j 0 ( v , f ) = ( Δ v − m j div v + F μ j ∇ f + V μ j v + F μ j m j f Δ f + d j div v − e j r e d m j ⋅ v + q μ j f ) , M_j^0(v,f)=
\begin{pmatrix}
\Delta v-\mathbf{m}_j\operatorname{div}v+F_{\mu_j}\nabla f+V_{\mu_j}v+F_{\mu_j}\mathbf{m}_j f\\
\Delta f+d_j\operatorname{div}v-e_j^{\mathrm{red}}\mathbf{m}_j\cdot v+q_{\mu_j}f
\end{pmatrix}, M j 0 ( v , f ) = ( Δ v − m j div v + F μ j ∇ f + V μ j v + F μ j m j f Δ f + d j div v − e j red m j ⋅ v + q μ j f ) ,
W j = ( ( W j / μ j ) I 3 μ j − 3 / 2 ( 2 W j m j − ∇ W j ) 0 ( 1 − κ j ) W j / μ j ) , M j W j = M j 0 + W j . (63) \mathcal{W}_j=
\begin{pmatrix}
(W_j/\mu_j)I_3 & \mu_j^{-3/2}(2W_j\mathbf{m}_j-\nabla W_j)\\
0 & (1-\kappa_j)W_j/\mu_j
\end{pmatrix},
\qquad
M_j^{W_j}=M_j^0+\mathcal{W}_j. \tag*{(63)} W j = ( ( W j / μ j ) I 3 0 μ j − 3/2 ( 2 W j m j − ∇ W j ) ( 1 − κ j ) W j / μ j ) , M j W j = M j 0 + W j . (63)
The operator M j W j M_j^{W_j} M j W j has principal part Δ I 4 \Delta I_4 Δ I 4 . The potential W j W_j W j and the gradient ∇ W j \nabla W_j ∇ W j enter only through the 4 × 4 4\times4 4 × 4 zeroth-order multiplication matrix W j \mathcal{W}_j W j . For every integer k ≥ 0 k\geq0 k ≥ 0 , smoothness of the coefficients gives the bounded mappings
T j , S j : H k + 1 ( Ω ; C 4 ) ⟶ H k ( Ω ; C 3 ) , T_j,S_j:H^{k+1}(\Omega;\mathbb{C}^{4})\longrightarrow H^k(\Omega;\mathbb{C}^{3}), T j , S j : H k + 1 ( Ω ; C 4 ) ⟶ H k ( Ω ; C 3 ) ,
M j W j : H k + 2 ( Ω ; C 4 ) ⟶ H k ( Ω ; C 4 ) . M_j^{W_j}:H^{k+2}(\Omega;\mathbb{C}^{4})\longrightarrow H^k(\Omega;\mathbb{C}^{4}). M j W j : H k + 2 ( Ω ; C 4 ) ⟶ H k ( Ω ; C 4 ) .
Lemma 3.1 establishes the factorization (64) , allowing the CGO solutions constructed for M W M^{W} M W to generate elastic displacements through T T T . The scalar-row identity (65) identifies the fourth-component boundary contribution in (75) , including the density-dependent term. Under the matching conditions (84) , the fourth-component contribution cancels the traction discrepancy, yielding the reduced Green–Betti identity (85) in terms of the Dirichlet-to-Neumann maps.
Lemma 3.1. Suppose Ω ⊂ R 3 \Omega\subset\mathbb{R}^{3} Ω ⊂ R 3 is a bounded domain with C ∞ C^\infty C ∞ boundary and, for j = 1 , 2 j=1,2 j = 1 , 2 , λ j , μ j , W j ∈ C ∞ ( Ω ‾ ; R ) \lambda_j,\mu_j,W_j\in C^\infty(\overline{\Omega};\mathbb{R}) λ j , μ j , W j ∈ C ∞ ( Ω ; R ) satisfy μ j > 0 \mu_j>0 μ j > 0 and 3 λ j + 2 μ j > 0 3\lambda_j+2\mu_j>0 3 λ j + 2 μ j > 0 on Ω ‾ \overline{\Omega} Ω . The notation in (60) is used for each material. The elasticity operator L j = L λ j , μ j L_j=L_{\lambda_j,\mu_j} L j = L λ j , μ j , the maps T j , S j T_j,S_j T j , S j , and the reduced operator M j W j M_j^{W_j} M j W j are defined in (61) , (62) , and (63) , respectively. The formal transpose S j T S_j^T S j T is defined using the complex-bilinear L 2 L^2 L 2 pairing on compactly supported smooth fields.
For every z = ( v , f ) T ∈ C ∞ ( Ω ‾ ; C 4 ) z=(v,f)^{T}\in C^{\infty}(\overline{\Omega};\mathbb{C}^{4}) z = ( v , f ) T ∈ C ∞ ( Ω ; C 4 ) and u ∈ C ∞ ( Ω ‾ ; C 3 ) u\in C^{\infty}(\overline{\Omega};\mathbb{C}^{3}) u ∈ C ∞ ( Ω ; C 3 ) , the factorization and transpose identities hold pointwise in Ω \Omega Ω for j = 1 , 2 j=1,2 j = 1 , 2 :
( L j + W j I 3 ) ( T j z ) = S j ( M j W j z ) , S j T u = ( μ j u − ( 1 − κ j ) − 1 div u ) . (64) \begin{aligned}
(L_j+W_jI_3)(T_jz)&=S_j(M_j^{W_j}z),\\
S_j^{T}u&=\begin{pmatrix}\sqrt{\mu_j}u\\-(1-\kappa_j)^{-1}\operatorname{div}u\end{pmatrix}. \tag*{(64)}
\end{aligned} ( L j + W j I 3 ) ( T j z ) S j T u = S j ( M j W j z ) , = ( μ j u − ( 1 − κ j ) − 1 div u ) . (64)
The fourth component of M j W j z M_j^{W_j}z M j W j z satisfies
( M j W j z ) 4 1 − κ j = Σ j div ( T j z ) − μ j div v + 3 2 μ j m j ⋅ v + W j μ j f . (65) \frac{(M_j^{W_j}z)_4}{1-\kappa_j}=\Sigma_j\operatorname{div}(T_jz)-\sqrt{\mu_j}\operatorname{div}v+\frac{3}{2}\sqrt{\mu_j}\mathbf{m}_j\cdot v+\frac{W_j}{\mu_j}f. \tag*{(65)} 1 − κ j ( M j W j z ) 4 = Σ j div ( T j z ) − μ j div v + 2 3 μ j m j ⋅ v + μ j W j f . (65)
Consequently, the displacement map T j T_j T j sends every smooth solution z z z of M j W j z = 0 M_j^{W_j}z=0 M j W j z = 0 in Ω \Omega Ω to an elastic displacement u = T j z u=T_jz u = T j z satisfying ( L j + W j I 3 ) u = 0 (L_j+W_jI_3)u=0 ( L j + W j I 3 ) u = 0 in Ω \Omega Ω .
Proof. Fix j ∈ { 1 , 2 } j\in\{1,2\} j ∈ { 1 , 2 } . The spatial indices i , k i,k i , k range from 1 to 3, and repeated spatial indices are summed. We first verify the factorization (64) by treating the scalar and vector contributions to T j z T_jz T j z in (62) separately. For the scalar contribution, define b j = a j ∇ f − f ∇ a j b_j=a_j\nabla f-f\nabla a_j b j = a j ∇ f − f ∇ a j and p f , j = Δ f − ( Δ a j / a j ) f = Δ f + q μ j f p_{f,j}=\Delta f-(\Delta a_j/a_j)f=\Delta f+q_{\mu_j}f p f , j = Δ f − ( Δ a j / a j ) f = Δ f + q μ j f , where the last equality follows from (60) and (63) . The product rule gives ε ( b j ) = a j ∇ 2 f − f ∇ 2 a j \varepsilon(b_j)=a_j\nabla^2f-f\nabla^2a_j ε ( b j ) = a j ∇ 2 f − f ∇ 2 a j and div b j = a j p f , j \operatorname{div}b_j=a_jp_{f,j} div b j = a j p f , j . With ∂ i k = ∂ i ∂ k \partial_{ik}=\partial_i\partial_k ∂ ik = ∂ i ∂ k , expansion of ∂ k ( σ j ( b j ) ) i k \partial_k(\sigma_j(b_j))_{ik} ∂ k ( σ j ( b j ) ) ik according to (61) yields
( L j b j − ∇ ( a j Σ j p f , j ) ) i = 2 μ j ( Δ a j δ i k − ∂ i k a j ) ∂ k f + 2 ( ( ∂ i μ j ) Δ a j − ( ∂ k μ j ) ∂ i k a j ) f . (66) \begin{aligned}
(L_jb_j-\nabla(a_j\Sigma_jp_{f,j}))_i
&=2\mu_j(\Delta a_j\delta_{ik}-\partial_{ik}a_j)\partial_kf\\
&\quad+2\bigl((\partial_i\mu_j)\Delta a_j-(\partial_k\mu_j)\partial_{ik}a_j\bigr)f. \tag*{(66)}
\end{aligned} ( L j b j − ∇ ( a j Σ j p f , j ) ) i = 2 μ j ( Δ a j δ ik − ∂ ik a j ) ∂ k f + 2 ( ( ∂ i μ j ) Δ a j − ( ∂ k μ j ) ∂ ik a j ) f . (66)
The coefficient relations ∇ μ j = − μ j 2 ∇ a j \nabla\mu_j=-\mu_j^2\nabla a_j ∇ μ j = − μ j 2 ∇ a j , ∇ a j = − a j m j \nabla a_j=-a_j\mathbf{m}_j ∇ a j = − a j m j , Δ a j I 3 − ∇ 2 a j = ( c j / 2 ) F μ j \Delta a_jI_3-\nabla^2a_j=(c_j/2)F_{\mu_j} Δ a j I 3 − ∇ 2 a j = ( c j /2 ) F μ j , and a j Σ j = ( 1 − κ j ) − 1 a_j\Sigma_j=(1-\kappa_j)^{-1} a j Σ j = ( 1 − κ j ) − 1 , obtained from (60) and (63) , reduce (66) to
L j b j = ∇ ( Δ f + q μ j f 1 − κ j ) + c j − 1 F μ j ( ∇ f + m j f ) . (67) L_jb_j=\nabla\left(\frac{\Delta f+q_{\mu_j}f}{1-\kappa_j}\right)+c_j^{-1}F_{\mu_j}(\nabla f+\mathbf{m}_jf). \tag*{(67)} L j b j = ∇ ( 1 − κ j Δ f + q μ j f ) + c j − 1 F μ j ( ∇ f + m j f ) . (67)
For the vector contribution, define d v = div v d_v=\operatorname{div}v d v = div v and Ψ v , j = c j ( λ j + μ j ) d v + ( λ j − μ j ) ∇ c j ⋅ v \Psi_{v,j}=c_j(\lambda_j+\mu_j)d_v+(\lambda_j-\mu_j)\nabla c_j\cdot v Ψ v , j = c j ( λ j + μ j ) d v + ( λ j − μ j ) ∇ c j ⋅ v . Subtracting ∇ Ψ v , j \nabla\Psi_{v,j} ∇ Ψ v , j from L j ( c j v ) L_j(c_jv) L j ( c j v ) cancels the term c j ( λ j + μ j ) ∇ d v c_j(\lambda_j+\mu_j)\nabla d_v c j ( λ j + μ j ) ∇ d v and all terms involving ∇ λ j \nabla\lambda_j ∇ λ j . The relation 2 μ j ∇ c j + c j ∇ μ j = 0 2\mu_j\nabla c_j+c_j\nabla\mu_j=0 2 μ j ∇ c j + c j ∇ μ j = 0 , following from (60) , reduces the remaining terms to
( L j ( c j v ) − ∇ Ψ v , j ) i = c j − 1 ( Δ v i − ( m j ) i d v ) + ( Z j ) i k v k , ( Z j ) i k = 2 μ j ∂ i k c j + ( ∂ i μ j ) ( ∂ k c j ) + ( ∂ k μ j ) ( ∂ i c j ) + ( μ j Δ c j + ∇ μ j ⋅ ∇ c j ) δ i k . (68) \begin{aligned}
(L_j(c_jv)-\nabla\Psi_{v,j})_i
&=c_j^{-1}(\Delta v_i-(\mathbf{m}_j)_id_v)+(Z_j)_{ik}v_k,\\
(Z_j)_{ik}
&=2\mu_j\partial_{ik}c_j+(\partial_i\mu_j)(\partial_kc_j)+(\partial_k\mu_j)(\partial_ic_j)\\
&\quad+(\mu_j\Delta c_j+\nabla\mu_j\cdot\nabla c_j)\delta_{ik}. \tag*{(68)}
\end{aligned} ( L j ( c j v ) − ∇ Ψ v , j ) i ( Z j ) ik = c j − 1 ( Δ v i − ( m j ) i d v ) + ( Z j ) ik v k , = 2 μ j ∂ ik c j + ( ∂ i μ j ) ( ∂ k c j ) + ( ∂ k μ j ) ( ∂ i c j ) + ( μ j Δ c j + ∇ μ j ⋅ ∇ c j ) δ ik . (68)
The differentiation formulas ∂ i c j = − c j ( m j ) i / 2 \partial_ic_j=-c_j(\mathbf{m}_j)_i/2 ∂ i c j = − c j ( m j ) i /2 and ∂ i k c j = c j ( ( m j ) i ( m j ) k / 4 − ∂ i k ℓ j / 2 ) \partial_{ik}c_j=c_j((\mathbf{m}_j)_i(\mathbf{m}_j)_k/4-\partial_{ik}\ell_j/2) ∂ ik c j = c j (( m j ) i ( m j ) k /4 − ∂ ik ℓ j /2 ) , obtained from (60) , give c j Z j = V μ j c_jZ_j=V_{\mu_j} c j Z j = V μ j by (63) . The definitions of d j d_j d j and e j r e d e_j^{\mathrm{red}} e j red in (63) also give Ψ v , j = ( d j d v − e j r e d m j ⋅ v ) / ( 1 − κ j ) \Psi_{v,j}=(d_jd_v-e_j^{\mathrm{red}}\mathbf{m}_j\cdot v)/(1-\kappa_j) Ψ v , j = ( d j d v − e j red m j ⋅ v ) / ( 1 − κ j ) . Consequently, equation (68) becomes
L j ( c j v ) = c j − 1 ( Δ v − m j d v + V μ j v ) + ∇ ( d j d v − e j r e d m j ⋅ v 1 − κ j ) . (69) \begin{aligned}
L_j(c_jv)&=c_j^{-1}(\Delta v-\mathbf{m}_jd_v+V_{\mu_j}v)\\
&\quad+\nabla\left(\frac{d_jd_v-e_j^{\mathrm{red}}\mathbf{m}_j\cdot v}{1-\kappa_j}\right). \tag*{(69)}
\end{aligned} L j ( c j v ) = c j − 1 ( Δ v − m j d v + V μ j v ) + ∇ ( 1 − κ j d j d v − e j red m j ⋅ v ) . (69)
Adding (67) and (69) , with T j , S j T_j,S_j T j , S j defined in (62) and M j 0 M_j^0 M j 0 defined in (63) , yields L j T j z = S j M j 0 z L_jT_jz=S_jM_j^0z L j T j z = S j M j 0 z .
For the potential contribution, the relation ∇ a j = − a j m j \nabla a_j=-a_jm_j ∇ a j = − a j m j and the definitions in (62) and (63) give
S j ( W j z ) = W j c j v + a j ( 2 W j m j − ∇ W j ) f + ∇ ( W j a j f ) = W j ( c j v + a j ∇ f − f ∇ a j ) = W j T j z . (70) \begin{aligned}
S_j(W_jz)=W_jc_jv+a_j(2W_jm_j-\nabla W_j)f+\nabla(W_ja_jf) \\
&=W_j(c_jv+a_j\nabla f-f\nabla a_j)=W_jT_jz. \tag*{(70)}
\end{aligned} S j ( W j z ) = W j c j v + a j ( 2 W j m j − ∇ W j ) f + ∇ ( W j a j f ) = W j ( c j v + a j ∇ f − f ∇ a j ) = W j T j z . (70)
Hence ( L j + W j I 3 ) T j z = S j ( M j W j z ) (L_j+W_jI_3)T_jz=S_j(M_j^{W_j}z) ( L j + W j I 3 ) T j z = S j ( M j W j z ) . Integration by parts against compactly supported test fields in the source map (62) , with the complex-bilinear L 2 L^2 L 2 pairing, gives S j T u = ( c j − 1 u , − ( 1 − κ j ) − 1 div u ) T S_j^Tu=(c_j^{-1}u,-(1-\kappa_j)^{-1}\operatorname{div}u)^T S j T u = ( c j − 1 u , − ( 1 − κ j ) − 1 div u ) T , completing the verification of (64) .
To establish the scalar identity (65) , differentiation of the displacement map (62) gives
div ( T j z ) = c j d v − 1 2 c j m j ⋅ v + a j ( Δ f + q μ j f ) . (71) \operatorname{div}(T_jz)=c_jd_v-\frac{1}{2}c_jm_j\cdot v+a_j(\Delta f+q_{\mu_j}f). \tag*{(71)} div ( T j z ) = c j d v − 2 1 c j m j ⋅ v + a j ( Δ f + q μ j f ) . (71)
Substituting (71) into the right-hand side of (65) and using ∑ j a j = ( 1 − κ j ) − 1 \sum_j a_j=(1-\kappa_j)^{-1} ∑ j a j = ( 1 − κ j ) − 1 from (60) , together with the definitions of d j d_j d j and e j r e d e_j^{\mathrm{red}} e j red in (63) , we obtain
Δ f + d j d v − e j r e d m j ⋅ v + q μ j f 1 − κ j + W j μ j f = ( M j W j z ) 4 1 − κ j . (72) \frac{\Delta f+d_jd_v-e_j^{\mathrm{red}}m_j\cdot v+q_{\mu_j}f}{1-\kappa_j}+\frac{W_j}{\mu_j}f=\frac{(M_j^{W_j}z)_4}{1-\kappa_j}. \tag*{(72)} 1 − κ j Δ f + d j d v − e j red m j ⋅ v + q μ j f + μ j W j f = 1 − κ j ( M j W j z ) 4 . (72)
which proves the scalar identity (65) . □
Deriving the reduced Green–Betti identity from the Dirichlet-to-Neumann maps requires an explicit calculation of the boundary terms introduced by the maps T j T_j T j and S j S_j S j in (62) . Section 3.2 derives the complete identity (75) , retaining the displacement and traction discrepancies together with the auxiliary scalar boundary contribution. The explicit boundary formula provides the basis for verifying the cancellation under the matching conditions (84) .
Boundary terms in the reduced Green–Betti identity
To derive the reduced Green–Betti identity (75) , we first introduce a reduced solution z = ( v , f ) T ∈ C ∞ ( Ω ‾ ; C 4 ) z=(v,f)^T\in C^\infty(\overline{\Omega};\mathbb{C}^4) z = ( v , f ) T ∈ C ∞ ( Ω ; C 4 ) for the first material and an elastic displacement u 2 ∈ C ∞ ( Ω ‾ ; C 3 ) u_2\in C^\infty(\overline{\Omega};\mathbb{C}^3) u 2 ∈ C ∞ ( Ω ; C 3 ) for the second material, satisfying
M 1 W 1 z = 0 , ( L 2 + W 2 I 3 ) u 2 = 0 in Ω . M_1^{W_1}z=0,\qquad(L_2+W_2I_3)u_2=0\quad\text{in }\Omega. M 1 W 1 z = 0 , ( L 2 + W 2 I 3 ) u 2 = 0 in Ω.
With γ 0 \gamma_0 γ 0 denoting the Dirichlet trace on ∂ Ω \partial\Omega ∂ Ω , the associated fields and boundary traces are defined by
u 1 = T 1 z , V = T 2 z , w = M 2 W 2 z , u_1=T_1z,\qquad V=T_2z,\qquad w=M_2^{W_2}z, u 1 = T 1 z , V = T 2 z , w = M 2 W 2 z ,
y 2 = S 2 T u 2 , F i = γ 0 u i ( i = 1 , 2 ) . (73) y_2=S_2^Tu_2,\qquad F_i=\gamma_0u_i\quad(i=1,2). \tag*{(73)} y 2 = S 2 T u 2 , F i = γ 0 u i ( i = 1 , 2 ) . (73)
The following proposition relates the reduced volume pairing ∫ Ω y 2 T D W z d x \int_\Omega y_2^TD^Wz\,dx ∫ Ω y 2 T D W z d x to the physical boundary pairing associated with Λ 2 Ω − Λ 1 Ω \Lambda_2^\Omega-\Lambda_1^\Omega Λ 2 Ω − Λ 1 Ω . The identity (75) identifies the additional contributions from the displacement trace discrepancy, the traction discrepancy, and the scalar component w 4 w_4 w 4 . The formula is derived directly from the physical boundary maps and records all discrepancies before the matching conditions (84) are imposed.
Proposition 3.2. Assume that the coefficient triples ( λ j , μ j , W j ) ∈ C ∞ ( Ω ‾ ; R 3 ) (\lambda_j,\mu_j,W_j)\in C^\infty(\overline{\Omega};\mathbb{R}^3) ( λ j , μ j , W j ) ∈ C ∞ ( Ω ; R 3 ) , j = 1 , 2 j=1,2 j = 1 , 2 , satisfy μ j > 0 \mu_j>0 μ j > 0 and 3 λ j + 2 μ j > 0 3\lambda_j+2\mu_j>0 3 λ j + 2 μ j > 0 on Ω ‾ \overline{\Omega} Ω , and that the corresponding dirichlet problems satisfy the nonresonance condition (6) . For z = ( v , f ) T ∈ C ∞ ( Ω ‾ ; C 4 ) z=(v,f)^T\in C^\infty(\overline{\Omega};\mathbb{C}^4) z = ( v , f ) T ∈ C ∞ ( Ω ; C 4 ) and u 2 ∈ C ∞ ( Ω ‾ ; C 3 ) u_2\in C^\infty(\overline{\Omega};\mathbb{C}^3) u 2 ∈ C ∞ ( Ω ; C 3 ) satisfying
M 1 W 1 z = 0 , ( L 2 + W 2 I 3 ) u 2 = 0 in Ω , (74) M_1^{W_1}z=0,\qquad(L_2+W_2I_3)u_2=0\quad\text{in }\Omega, \tag*{(74)} M 1 W 1 z = 0 , ( L 2 + W 2 I 3 ) u 2 = 0 in Ω , (74) the fields and boundary traces defined in (73) satisfy the Green–Betti identity
∫ Ω y 2 T D W z d x = ⟨ ( Λ 2 Ω − Λ 1 Ω ) F 1 , F 2 ⟩ + ⟨ Λ 2 Ω F 2 , γ 0 V − F 1 ⟩ \int_\Omega y_2^TD^Wz\,dx=\langle(\Lambda_2^\Omega-\Lambda_1^\Omega)F_1,F_2\rangle+\langle\Lambda_2^\Omega F_2,\gamma_0V-F_1\rangle ∫ Ω y 2 T D W z d x = ⟨( Λ 2 Ω − Λ 1 Ω ) F 1 , F 2 ⟩ + ⟨ Λ 2 Ω F 2 , γ 0 V − F 1 ⟩
− ⟨ t 2 ( V ) − t 1 ( u 1 ) , F 2 ⟩ + ∫ ∂ Ω w 4 1 − κ 2 ( u 2 ⋅ ν ) d S . (75) -\langle t_2(V)-t_1(u_1),F_2\rangle+\int_{\partial\Omega}\frac{w_4}{1-\kappa_2}(u_2\cdot\nu)\,dS. \tag*{(75)} − ⟨ t 2 ( V ) − t 1 ( u 1 ) , F 2 ⟩ + ∫ ∂ Ω 1 − κ 2 w 4 ( u 2 ⋅ ν ) d S . (75)
In the above formulation, the scalar w 4 = ( M W 2 z ) 4 w_4=(M^{W_2}z)_4 w 4 = ( M W 2 z ) 4 represents the fourth component of w = M W 2 z w=M^{W_2}z w = M W 2 z . The traction operators t j t_j t j and the Dirichlet-to-Neumann maps Λ j Ω \Lambda_j^\Omega Λ j Ω for j = 1 , 2 j=1,2 j = 1 , 2 are defined in (61) and (8) , respectively. The operator difference D W D^W D W is introduced in (59) . All boundary pairings are understood as complex bilinear forms. Identity (75) holds under the stated coefficient assumptions, prior to imposing the boundary matching conditions (84) .
Proof. Integration by parts in the gradient term of the source map S 2 S_2 S 2 defined in (62) gives
∫ Ω u 2 T S 2 w d x − ∫ Ω ( S 2 T u 2 ) T w d x = ∫ ∂ Ω w 4 1 − κ 2 ( u 2 ⋅ ν ) d S . (76) \int_\Omega u_2^T S_2w\,dx-\int_\Omega(S_2^Tu_2)^Tw\,dx=\int_{\partial\Omega}\frac{w_4}{1-\kappa_2}(u_2\cdot\nu)\,dS. \tag*{(76)} ∫ Ω u 2 T S 2 w d x − ∫ Ω ( S 2 T u 2 ) T w d x = ∫ ∂ Ω 1 − κ 2 w 4 ( u 2 ⋅ ν ) d S . (76)
The constitutive law (61) implies
∇ u 2 : σ 2 ( V ) = 2 μ 2 ε ( u 2 ) : ε ( V ) + λ 2 ( div u 2 ) ( div V ) = ∇ V : σ 2 ( u 2 ) . (77) \nabla u_2:\sigma_2(V)=2\mu_2\varepsilon(u_2):\varepsilon(V)+\lambda_2(\operatorname{div}u_2)(\operatorname{div}V)
=\nabla V:\sigma_2(u_2). \tag*{(77)} ∇ u 2 : σ 2 ( V ) = 2 μ 2 ε ( u 2 ) : ε ( V ) + λ 2 ( div u 2 ) ( div V ) = ∇ V : σ 2 ( u 2 ) . (77)
Integration by parts, using (77) and the cancellation of the scalar potential terms, therefore yields the Green identity
∫ Ω [ u 2 T ( L 2 + W 2 I 3 ) V − V T ( L 2 + W 2 I 3 ) u 2 ] d x = ⟨ t 2 ( V ) , F 2 ⟩ − ⟨ t 2 ( u 2 ) , γ 0 V ⟩ . (78) \int_\Omega\left[u_2^T(L_2+W_2I_3)V-V^T(L_2+W_2I_3)u_2\right]\,dx
=\langle t_2(V),F_2\rangle-\langle t_2(u_2),\gamma_0V\rangle. \tag*{(78)} ∫ Ω [ u 2 T ( L 2 + W 2 I 3 ) V − V T ( L 2 + W 2 I 3 ) u 2 ] d x = ⟨ t 2 ( V ) , F 2 ⟩ − ⟨ t 2 ( u 2 ) , γ 0 V ⟩ . (78)
For each material, application of the Green identity to two homogeneous solutions with smooth boundary data φ , ψ \varphi,\psi φ , ψ gives
⟨ Λ j Ω φ , ψ ⟩ = ⟨ Λ j Ω ψ , φ ⟩ , j = 1 , 2. (79) \langle\Lambda_j^\Omega\varphi,\psi\rangle=\langle\Lambda_j^\Omega\psi,\varphi\rangle,\qquad j=1,2. \tag*{(79)} ⟨ Λ j Ω φ , ψ ⟩ = ⟨ Λ j Ω ψ , φ ⟩ , j = 1 , 2. (79)
The complex-bilinear reciprocity relation extends to all φ , ψ ∈ H 1 / 2 ( ∂ Ω ; C 3 ) \varphi,\psi\in H^{1/2}(\partial\Omega;\mathbb{C}^3) φ , ψ ∈ H 1/2 ( ∂ Ω ; C 3 ) by density of smooth boundary data and boundedness of Λ j Ω \Lambda_j^\Omega Λ j Ω .
The factorization identity (64) and the field definitions (73) give ( L 2 + W 2 I 3 ) V = S 2 w (L_2+W_2I_3)V=S_2w ( L 2 + W 2 I 3 ) V = S 2 w and y 2 = S 2 T u 2 y_2=S_2^Tu_2 y 2 = S 2 T u 2 , while M W 1 z = 0 M^{W_1}z=0 M W 1 z = 0 gives D W z = − w D^Wz=-w D W z = − w . Substitution of the field relations into (76) and (78) , together with ( L 2 + W 2 I 3 ) u 2 = 0 (L_2+W_2I_3)u_2=0 ( L 2 + W 2 I 3 ) u 2 = 0 and t 2 ( u 2 ) = Λ 2 Ω F 2 t_2(u_2)=\Lambda_2^\Omega F_2 t 2 ( u 2 ) = Λ 2 Ω F 2 , yields
∫ Ω y 2 T D W z d x = − ⟨ t 2 ( V ) , F 2 ⟩ + ⟨ Λ 2 Ω F 2 , γ 0 V ⟩ + ∫ ∂ Ω w 4 1 − κ 2 ( u 2 ⋅ ν ) d S . (80) \int_\Omega y_2^TD^Wz\,dx=-\langle t_2(V),F_2\rangle+\langle\Lambda_2^\Omega F_2,\gamma_0V\rangle
+\int_{\partial\Omega}\frac{w_4}{1-\kappa_2}(u_2\cdot\nu)\,dS. \tag*{(80)} ∫ Ω y 2 T D W z d x = − ⟨ t 2 ( V ) , F 2 ⟩ + ⟨ Λ 2 Ω F 2 , γ 0 V ⟩ + ∫ ∂ Ω 1 − κ 2 w 4 ( u 2 ⋅ ν ) d S . (80)
To isolate the physical Dirichlet-to-Neumann difference, we write γ 0 V = F 1 + ( γ 0 V − F 1 ) \gamma_0V=F_1+(\gamma_0V-F_1) γ 0 V = F 1 + ( γ 0 V − F 1 ) and t 2 ( V ) = t 1 ( u 1 ) + ( t 2 ( V ) − t 1 ( u 1 ) ) t_2(V)=t_1(u_1)+(t_2(V)-t_1(u_1)) t 2 ( V ) = t 1 ( u 1 ) + ( t 2 ( V ) − t 1 ( u 1 )) . Substituting the decompositions into (80) and using t 1 ( u 1 ) = Λ 1 Ω F 1 t_1(u_1)=\Lambda_1^\Omega F_1 t 1 ( u 1 ) = Λ 1 Ω F 1 together with the reciprocity relation (79) , we obtain (75) .
The discrepancy from the physical Dirichlet-to-Neumann pairing therefore admits the boundary representation
∫ Ω y 2 T D W z d x − ⟨ ( Λ 2 Ω − Λ 1 Ω ) F 1 , F 2 ⟩ = ∫ ∂ Ω [ t 2 ( u 2 ) ⋅ ( V − u 1 ) − ( t 2 ( V ) − t 1 ( u 1 ) ) ⋅ u 2 + w 4 1 − κ 2 ( u 2 ⋅ ν ) ] d S . (81) \int_\Omega y_2^TD^Wz\,dx-\langle(\Lambda_2^\Omega-\Lambda_1^\Omega)F_1,F_2\rangle
=\int_{\partial\Omega}\left[t_2(u_2)\cdot(V-u_1)-(t_2(V)-t_1(u_1))\cdot u_2+\frac{w_4}{1-\kappa_2}(u_2\cdot\nu)\right]\,dS. \tag*{(81)} ∫ Ω y 2 T D W z d x − ⟨( Λ 2 Ω − Λ 1 Ω ) F 1 , F 2 ⟩ = ∫ ∂ Ω [ t 2 ( u 2 ) ⋅ ( V − u 1 ) − ( t 2 ( V ) − t 1 ( u 1 )) ⋅ u 2 + 1 − κ 2 w 4 ( u 2 ⋅ ν ) ] d S . (81)
The boundary representation (81) is expressed through the physical traction operators and the source map S 2 S_2 S 2 , with the reduced-system contribution appearing as the explicit scalar term involving w 4 w_4 w 4 .
Elimination of the boundary terms
Section 3.2 derives the general Green–Betti identity (75) for arbitrary boundary traces. Section 3.3 then uses the matching conditions (84) to cancel the additional boundary contributions and reduce the general formula to (85) . The elimination of the boundary terms is needed to deduce a vanishing reduced volume pairing from equality of the Dirichlet-to-Neumann maps.
Lemma 3.3 establishes the normal form (83) for the boundary Hessian under the vanishing conditions (82) . The normal form is applied to the coefficient difference a 2 − a 1 a_2-a_1 a 2 − a 1 , with a j a_j a j defined in (60) , to identify the second-derivative contribution to the traction discrepancy (88) .
Lemma 3.3. Let Ω ⊂ R 3 \Omega\subset\mathbb{R}^3 Ω ⊂ R 3 be a domain with C 2 C^2 C 2 boundary, and let ν \nu ν denote the outward unit normal to ∂ Ω \partial\Omega ∂ Ω . Suppose that b ∈ C 2 ( Ω ‾ ; R ) b \in C^2(\overline{\Omega};\mathbb{R}) b ∈ C 2 ( Ω ; R ) satisfies
b = 0 , ∇ b = 0 on ∂ Ω . (82) b=0,\qquad\nabla b=0\quad\text{on }\partial\Omega. \tag*{(82)} b = 0 , ∇ b = 0 on ∂ Ω. (82) Then the boundary Hessian of b b b has only a normal–normal component:
∇ 2 b ( x ) = η ( x ) ν ( x ) ⊗ ν ( x ) , x ∈ ∂ Ω , (83) \nabla^2b(x)=\eta(x)\,\nu(x)\otimes\nu(x),\qquad x\in\partial\Omega, \tag*{(83)} ∇ 2 b ( x ) = η ( x ) ν ( x ) ⊗ ν ( x ) , x ∈ ∂ Ω , (83) where η ( x ) : = ν ( x ) T ∇ 2 b ( x ) ν ( x ) \eta(x):=\nu(x)^{T}\nabla^2b(x)\nu(x) η ( x ) := ν ( x ) T ∇ 2 b ( x ) ν ( x ) .
Proof. Fix x ∈ ∂ Ω x\in\partial\Omega x ∈ ∂ Ω and τ ∈ T x ( ∂ Ω ) \tau\in T_x(\partial\Omega) τ ∈ T x ( ∂ Ω ) . Tangential differentiation of the gradient condition in (82) gives
∇ 2 b ( x ) τ = 0. \nabla^2b(x)\tau=0. ∇ 2 b ( x ) τ = 0.
The symmetry of ∇ 2 b ( x ) \nabla^2b(x) ∇ 2 b ( x ) therefore implies (83) .
Proposition 3.4 identifies the cancellation of the boundary terms in (75) under (84) and computes the residual boundary contribution when the matching of W j W_j W j is relaxed.
Proposition 3.4. Under the coefficient assumptions of Proposition 3.2, let z = ( v , f ) T z=(v,f)^T z = ( v , f ) T and u 2 u_2 u 2 be smooth solutions of (74) , with the associated fields and boundary traces defined by (73) . Suppose that the boundary matching conditions
μ 1 = μ 2 , ∂ ν μ 1 = ∂ ν μ 2 , W 1 = W 2 on ∂ Ω (84) \mu_1=\mu_2,\qquad\partial_\nu\mu_1=\partial_\nu\mu_2,\qquad W_1=W_2\quad\text{on }\partial\Omega \tag*{(84)} μ 1 = μ 2 , ∂ ν μ 1 = ∂ ν μ 2 , W 1 = W 2 on ∂ Ω (84) hold. Then the reduced Green–Betti identity
∫ Ω u 2 T D W z d x = ⟨ ( Λ Ω 2 − Λ Ω 1 ) F 1 , F 2 ⟩ (85) \int_{\Omega}u_2^T D^W z\,dx=\langle(\Lambda_{\Omega}^{2}-\Lambda_{\Omega}^{1})F_1,F_2\rangle \tag*{(85)} ∫ Ω u 2 T D W z d x = ⟨( Λ Ω 2 − Λ Ω 1 ) F 1 , F 2 ⟩ (85) holds.
If the boundary condition W 1 = W 2 W_1=W_2 W 1 = W 2 on ∂ Ω \partial\Omega ∂ Ω is dropped while the two matching conditions for the shear modulus in (84) are retained, the right-hand side of (85) acquires the boundary defect term
∫ ∂ Ω W 2 − W 1 μ 1 f ( u 2 ⋅ ν ) d S . (86) \int_{\partial\Omega}\frac{W_2-W_1}{\mu_1}f(u_2\cdot\nu)\,dS. \tag*{(86)} ∫ ∂ Ω μ 1 W 2 − W 1 f ( u 2 ⋅ ν ) d S . (86) The boundary cancellation depends on the traces of μ j \mu_j μ j , ∂ ν μ j \partial_\nu\mu_j ∂ ν μ j , and W j W_j W j specified in (84) ; the boundary values of λ j \lambda_j λ j , the second derivatives of μ j \mu_j μ j , and derivatives of W j W_j W j do not enter the defect formula.
Proof. Under the shear-modulus matching conditions in (84) , differentiating μ 1 − μ 2 = 0 \mu_1-\mu_2=0 μ 1 − μ 2 = 0 along tangent directions and using the equality ∂ ν μ 1 = ∂ ν μ 2 \partial_\nu\mu_1=\partial_\nu\mu_2 ∂ ν μ 1 = ∂ ν μ 2 , we obtain ∇ μ 1 = ∇ μ 2 \nabla\mu_1=\nabla\mu_2 ∇ μ 1 = ∇ μ 2 on ∂ Ω \partial\Omega ∂ Ω . For a coefficient b b b carrying a material index, use δ 21 b : = b 2 − b 1 \delta_{21}b:=b_2-b_1 δ 21 b := b 2 − b 1 , the opposite subtraction order to δ 12 b \delta_{12}b δ 12 b . Consequently, the coefficient differences δ 21 a : = a 2 − a 1 \delta_{21}a := a_2-a_1 δ 21 a := a 2 − a 1 and δ 21 c : = c 2 − c 1 \delta_{21}c := c_2-c_1 δ 21 c := c 2 − c 1 , with a j , c j a_j,c_j a j , c j defined in (60) , satisfy
δ 21 a = δ 21 c = 0 , ∇ δ 21 a = ∇ δ 21 c = 0 on ∂ Ω . \delta_{21}a=\delta_{21}c=0,\qquad\nabla\delta_{21}a=\nabla\delta_{21}c=0\quad\text{on }\partial\Omega. δ 21 a = δ 21 c = 0 , ∇ δ 21 a = ∇ δ 21 c = 0 on ∂ Ω.
Applying Lemma 3.3, we find that the Hessian of δ 21 a \delta_{21}a δ 21 a assumes the normal form (83) :
∇ 2 δ 21 a = η μ ν ⊗ ν , η μ : = ν T ( ∇ 2 δ 21 a ) ν on ∂ Ω . \nabla^2\delta_{21}a=\eta_\mu\nu\otimes\nu,\qquad\eta_\mu:=\nu^{T}(\nabla^2\delta_{21}a)\nu\quad\text{on }\partial\Omega. ∇ 2 δ 21 a = η μ ν ⊗ ν , η μ := ν T ( ∇ 2 δ 21 a ) ν on ∂ Ω.
From the displacement representation (62) and the field definitions (73) , the difference of displacements satisfies
V − u 1 = δ 21 c v + δ 21 a ∇ f − f ∇ δ 21 a . (87) V-u_1=\delta_{21}c\,v+\delta_{21}a\nabla f-f\nabla\delta_{21}a. \tag*{(87)} V − u 1 = δ 21 c v + δ 21 a ∇ f − f ∇ δ 21 a . (87)
Since δ 21 a \delta_{21}a δ 21 a , δ 21 c \delta_{21}c δ 21 c , and the corresponding first-order derivatives vanish on ∂ Ω \partial\Omega ∂ Ω , formula (87) implies γ 0 V = γ 0 u 1 = F 1 \gamma_0V=\gamma_0u_1=F_1 γ 0 V = γ 0 u 1 = F 1 . Differentiating (87) and evaluating the result on ∂ Ω \partial\Omega ∂ Ω , we obtain
∇ ( V − u 1 ) = − f η μ ν ⊗ ν , div ( V − u 1 ) = − f η μ on ∂ Ω . \nabla(V-u_1)=-f\eta_\mu\nu\otimes\nu,\qquad\operatorname{div}(V-u_1)=-f\eta_\mu\quad\text{on }\partial\Omega. ∇ ( V − u 1 ) = − f η μ ν ⊗ ν , div ( V − u 1 ) = − f η μ on ∂ Ω.
By (84) , the equality μ 1 = μ 2 \mu_1=\mu_2 μ 1 = μ 2 holds on ∂ Ω \partial\Omega ∂ Ω . Evaluating the traction difference on ∂ Ω \partial\Omega ∂ Ω via (61) , we deduce
t 2 ( V ) − t 1 ( u 1 ) = { 2 μ 1 ε ( V − u 1 ) + ( λ 2 div V − λ 1 div u 1 ) I 3 } ν = { ( λ 2 − λ 1 ) div u 1 − ( λ 2 + 2 μ 1 ) f η μ } ν on ∂ Ω . (88) \begin{aligned}
t_2(V)-t_1(u_1)
&=\{2\mu_1\varepsilon(V-u_1)+(\lambda_2\operatorname{div}V-\lambda_1\operatorname{div}u_1)I_3\}\nu\\
&=\{(\lambda_2-\lambda_1)\operatorname{div}u_1-(\lambda_2+2\mu_1)f\eta_\mu\}\nu\quad\text{on }\partial\Omega. \tag*{(88)}
\end{aligned} t 2 ( V ) − t 1 ( u 1 ) = { 2 μ 1 ε ( V − u 1 ) + ( λ 2 div V − λ 1 div u 1 ) I 3 } ν = {( λ 2 − λ 1 ) div u 1 − ( λ 2 + 2 μ 1 ) f η μ } ν on ∂ Ω. (88)
The matching conditions for the first-order derivatives in (84) also imply that m 1 = m 2 \mathbf{m}_1=\mathbf{m}_2 m 1 = m 2 on ∂ Ω \partial\Omega ∂ Ω . Subtracting the scalar identities (65) for j = 1 , 2 j=1,2 j = 1 , 2 and using μ 1 = μ 2 \mu_1=\mu_2 μ 1 = μ 2 on ∂ Ω \partial\Omega ∂ Ω eliminates the terms involving div v \operatorname{div}v div v and m j ⋅ v \mathbf{m}_j\cdot v m j ⋅ v . Since ( M 1 W 1 z ) 4 = 0 (M_1^{W_1}z)_4=0 ( M 1 W 1 z ) 4 = 0 by (74) , evaluating the resulting difference on ∂ Ω \partial\Omega ∂ Ω yields
w 4 1 − κ 2 = Σ 2 div V − Σ 1 div u 1 + W 2 − W 1 μ 1 f = ( λ 2 − λ 1 ) div u 1 − ( λ 2 + 2 μ 1 ) f η μ + W 2 − W 1 μ 1 f on ∂ Ω . (89) \begin{aligned}
\frac{w_4}{1-\kappa_2}
&=\Sigma_2\operatorname{div}V-\Sigma_1\operatorname{div}u_1+\frac{W_2-W_1}{\mu_1}f\\
&=(\lambda_2-\lambda_1)\operatorname{div}u_1-(\lambda_2+2\mu_1)f\eta_\mu+\frac{W_2-W_1}{\mu_1}f\quad\text{on }\partial\Omega. \tag*{(89)}
\end{aligned} 1 − κ 2 w 4 = Σ 2 div V − Σ 1 div u 1 + μ 1 W 2 − W 1 f = ( λ 2 − λ 1 ) div u 1 − ( λ 2 + 2 μ 1 ) f η μ + μ 1 W 2 − W 1 f on ∂ Ω. (89)
Comparing (88) and (89) , the contributions involving ( λ 2 − λ 1 ) div u 1 (\lambda_2-\lambda_1)\operatorname{div}u_1 ( λ 2 − λ 1 ) div u 1 and η μ \eta_\mu η μ cancel identically, yielding
w 4 1 − κ 2 ( u 2 ⋅ ν ) − ( t 2 ( V ) − t 1 ( u 1 ) ) ⋅ u 2 = W 2 − W 1 μ 1 f ( u 2 ⋅ ν ) on ∂ Ω . (90) \frac{w_4}{1-\kappa_2}(u_2\cdot\nu)-(t_2(V)-t_1(u_1))\cdot u_2
=\frac{W_2-W_1}{\mu_1}f(u_2\cdot\nu)\quad\text{on }\partial\Omega. \tag*{(90)} 1 − κ 2 w 4 ( u 2 ⋅ ν ) − ( t 2 ( V ) − t 1 ( u 1 )) ⋅ u 2 = μ 1 W 2 − W 1 f ( u 2 ⋅ ν ) on ∂ Ω. (90)
Because γ 0 V = F 1 \gamma_0V=F_1 γ 0 V = F 1 , the boundary defect in (81) reduces, by (90) , to the boundary integral (86) . When W 1 = W 2 W_1=W_2 W 1 = W 2 on ∂ Ω \partial\Omega ∂ Ω , the boundary defect term vanishes and the reduced Green–Betti identity (85) follows. □
Integration over the common extension
The matching conditions (84) suffice for the identity on Ω \Omega Ω . The full Taylor matching of Theorem 2.5, together with Proposition 2.7, makes the extended difference operator smooth and supported in Ω ‾ \overline{\Omega} Ω . Its integral may therefore be taken over a larger construction domain, using only the original boundary data.
Transport frames on the plane
The construction of complex geometrical optics solutions requires invertible transport frames with smooth dependence on the direction and transverse variables. The relevant transport equations arise from the conjugated matrix operator
P δ = ( D + δ ) 2 + 2 ∑ k = 1 3 A k ( x ) ( D k + δ k ) + q ( x ) , (91) P_\delta=(D+\delta)^2+2\sum_{k=1}^{3}A_k(x)(D_k+\delta_k)+q(x), \tag*{(91)} P δ = ( D + δ ) 2 + 2 k = 1 ∑ 3 A k ( x ) ( D k + δ k ) + q ( x ) , (91)
A k , q ∈ C c ∞ ( R 3 ; M m ( C ) ) , k = 1 , 2 , 3. A_k,q\in C^\infty_c(\mathbb{R}^3;M_m(\mathbb{C})),\qquad k=1,2,3. A k , q ∈ C c ∞ ( R 3 ; M m ( C )) , k = 1 , 2 , 3.
In (91) , m m m denotes the matrix size, D = − i ∇ D=-i\nabla D = − i ∇ is the Fourier differential operator, with the pairing convention of § 1.6 , and the complex phase vector is δ = τ ( v 0 + i v 1 ) \delta=\tau(v_0+iv_1) δ = τ ( v 0 + i v 1 ) for real orthonormal vectors v 0 , v 1 v_0,v_1 v 0 , v 1 and τ ≥ 1 \tau\geq1 τ ≥ 1 . This phase vector is distinct from the ordered coefficient differences δ 12 b \delta_{12}b δ 12 b and δ 21 b \delta_{21}b δ 21 b . The construction applies to arbitrary m m m , with m = 2 , 3 , 4 m=2,3,4 m = 2 , 3 , 4 required in the elasticity applications. The relation between P δ P_\delta P δ and the reduced elastic operator is established in § 5.7 .
In planar coordinates, the leading transport equations take the form (94) , and a smooth solution matrix C C C with values in G L ( m , C ) GL(m,\mathbb{C}) G L ( m , C ) is called an invertible transport frame. Theorem 4.1 establishes global frames for compactly supported smooth planar coefficients and jointly smooth local families on prescribed disks. The global construction combines weighted Cauchy estimates with Fredholm theory. Polynomial corrections enforce compatibility with the finite-dimensional cokernel, producing a global solution matrix. Local holomorphic factorizations then permit the successive removal of determinant zeros, yielding an invertible frame without a smallness assumption on the coefficient.
The global frame supplies the reference solution for the local parameter construction. The right-inverse identity (100) yields the Neumann-series representation (101) , and interior regularity gives joint smoothness in the planar variables and real parameters. The mixed seminorms in (95) measure the finite-order derivatives of the frame and inverse needed in the subsequent operator estimates. Proposition 4.7 extends the construction to finitely differentiable coefficients, while Proposition 4.8 establishes the uniform mixed derivative bounds in (106) over bounded C 12 C^{12} C 12 coefficient families.
The planar construction supplies the two-, three-, and four-component transport frames in Proposition 4.10. For the subsequent local right inverse, Proposition 4.11 specifies the derivative orders required when the frames are expressed in moving plane coordinates. The geometry, coefficients, cutoffs, local parameter charts, and frames are fixed before the limit τ → ∞ \tau\to\infty τ → ∞ is taken. The estimate constants depend on the fixed data and the selected finite frame seminorms, and are independent of τ \tau τ .
The planar notation used in the construction is
z = x 1 + i x 2 , D R = { z ∈ C : ∣ z ∣ < R } , ⟨ z ⟩ = ( 1 + ∣ z ∣ 2 ) 1 / 2 , z=x_1+ix_2,\qquad D_R=\{z\in\mathbb{C}:|z|<R\},\qquad\langle z\rangle=(1+|z|^2)^{1/2}, z = x 1 + i x 2 , D R = { z ∈ C : ∣ z ∣ < R } , ⟨ z ⟩ = ( 1 + ∣ z ∣ 2 ) 1/2 ,
∂ z ‾ = 1 2 ( ∂ x 1 + i ∂ x 2 ) , ∂ z = 1 2 ( ∂ x 1 − i ∂ x 2 ) . (92) \partial_{\overline{z}}=\frac{1}{2}(\partial_{x_1}+i\partial_{x_2}),\qquad\partial_z=\frac{1}{2}(\partial_{x_1}-i\partial_{x_2}). \tag*{(92)} ∂ z = 2 1 ( ∂ x 1 + i ∂ x 2 ) , ∂ z = 2 1 ( ∂ x 1 − i ∂ x 2 ) . (92)
The auxiliary Hilbert inner products are linear in the first variable. With the planar notation in (92) , the Cauchy operators on compactly supported smooth scalar, vector, or matrix fields are defined componentwise by
C f ( z ) = 1 π ∫ C f ( w ) z − w d A ( w ) , C ∗ f ( z ) = − 1 π ∫ C f ( w ) z ‾ − w ‾ d A ( w ) . (93) \mathcal{C}f(z)=\frac{1}{\pi}\int_{\mathbb{C}}\frac{f(w)}{z-w}\,dA(w),\qquad\mathcal{C}^*f(z)=-\frac{1}{\pi}\int_{\mathbb{C}}\frac{f(w)}{\overline{z}-\overline{w}}\,dA(w). \tag*{(93)} C f ( z ) = π 1 ∫ C z − w f ( w ) d A ( w ) , C ∗ f ( z ) = − π 1 ∫ C z − w f ( w ) d A ( w ) . (93)
In (93) , d A dA d A denotes Euclidean area measure, and the superscript ∗ * ∗ denotes the unweighted formal Hermitian adjoint. In the weighted cokernel system (97) , B ∗ B^* B ∗ denotes the Hermitian transpose of the matrix coefficient B B B . The weighted Hilbert adjoint is defined separately in Lemma 4.5.
Theorem 4.1. Let m ≥ 1 m\geq1 m ≥ 1 be an integer.
(1) For every B ∈ C ∞ ( C ; M m ( C ) ) B \in C^\infty(\mathbb{C};M_m(\mathbb{C})) B ∈ C ∞ ( C ; M m ( C )) , there exists a frame C ∈ C ∞ ( C ; G L ( m , C ) ) C \in C^\infty(\mathbb{C};GL(m,\mathbb{C})) C ∈ C ∞ ( C ; G L ( m , C )) satisfying the transport equation
∂ z ˉ C = B C . (94) \partial_{\bar z} C = BC. \tag*{(94)} ∂ z ˉ C = B C . (94) Moreover, the frame C C C can be chosen to satisfy
det C = c exp ( C ( tr B ) ) \det C = c\exp(\mathcal{C}(\operatorname{tr} B)) det C = c exp ( C ( tr B )) for some constant c ∈ C ∖ { 0 } c \in\mathbb{C} \setminus\{0\} c ∈ C ∖ { 0 } , where C \mathcal{C} C is the Cauchy operator defined in (93) .
(2) Let 0 < L < R f r 0 < L < R_{\mathrm{fr}} 0 < L < R fr , let Y Y Y be a finite-dimensional smooth real manifold, and let
B ∈ C ∞ ( D R f r × Y ; M m ( C ) ) . B \in C^\infty(D_{R_{\mathrm{fr}}} \times Y;M_m(\mathbb{C})). B ∈ C ∞ ( D R fr × Y ; M m ( C )) . Here R f r R_{\mathrm{fr}} R fr is the outer radius of the planar disk on which the coefficient family is prescribed. For every y 0 ∈ Y y_0 \in Y y 0 ∈ Y , there exist a radius L ′ ∈ ( L , R f r ) L' \in(L,R_{\mathrm{fr}}) L ′ ∈ ( L , R fr ) , an open coordinate neighborhood U ⊂ Y U \subset Y U ⊂ Y of y 0 y_0 y 0 , and a frame
C ∈ C ∞ ( D L ′ × U ; G L ( m , C ) ) C \in C^\infty(D_{L'} \times U;GL(m,\mathbb{C})) C ∈ C ∞ ( D L ′ × U ; G L ( m , C )) such that C ( ⋅ , y ) C(\cdot,y) C ( ⋅ , y ) satisfies (94) in D L ′ D_{L'} D L ′ for every y ∈ U y \in U y ∈ U .
For every compact set K ⋐ U K \Subset U K ⋐ U and every integer J ≥ 0 J \ge0 J ≥ 0 , the frame C C C and the inverse matrix C − 1 C^{-1} C − 1 satisfy
max ∣ α ∣ + ∣ β ∣ ≤ J sup D L × K ( ∣ ∂ x α ∂ y β C ∣ + ∣ ∂ x α ∂ y β C − 1 ∣ ) < ∞ , (95) \max_{|\alpha|+|\beta|\le J}\sup_{D_L\times K}\left(\left|\partial_x^\alpha\partial_y^\beta C\right|+\left|\partial_x^\alpha\partial_y^\beta C^{-1}\right|\right)<\infty, \tag*{(95)} ∣ α ∣ + ∣ β ∣ ≤ J max D L × K sup ( ∂ x α ∂ y β C + ∂ x α ∂ y β C − 1 ) < ∞ , (95) where derivatives are taken with respect to x = ( Re z , Im z ) x=(\operatorname{Re}z,\operatorname{Im}z) x = ( Re z , Im z ) and the real coordinates on U U U . For every prescribed function
g ∈ C ∞ ( D L ′ × U ; C ∖ { 0 } ) g \in C^\infty(D_{L'} \times U;\mathbb{C}\setminus\{0\}) g ∈ C ∞ ( D L ′ × U ; C ∖ { 0 }) satisfying ∂ z ˉ g = ( tr B ) g \partial_{\bar z}g=(\operatorname{tr}B)g ∂ z ˉ g = ( tr B ) g , the local frame C C C can additionally be chosen to satisfy det C = g \det C=g det C = g on D L ′ × U D_{L'}\times U D L ′ × U . Under the determinant normalization det C = g \det C=g det C = g , the mixed derivative bound (95) continues to hold.
In particular, every compact subset of Y Y Y admits a finite cover by coordinate neighborhoods carrying local frame families.
The weighted space below allows a finite-dimensional cokernel. We first remove that cokernel by polynomial corrections, obtaining a global solution matrix, and then remove its determinant zeros. A perturbation argument on a fixed disk gives the parameter-dependent frames in part 2.
The Cauchy operator and local uniqueness
For the operators in (93) , integration by parts outside a disk of radius ε \varepsilon ε , followed by ε ↓ 0 \varepsilon\downarrow0 ε ↓ 0 , gives the distributional identity ∂ z ˉ ( 1 / ( π z ) ) = δ 0 \partial_{\bar z}(1/(\pi z))=\delta_0 ∂ z ˉ ( 1/ ( π z )) = δ 0 . Consequently,
∂ z ˉ C f = f , − ∂ z C ∗ f = f . (96) \partial_{\bar z}\mathcal{C}f=f,\qquad-\partial_z\mathcal{C}^*f=f. \tag*{(96)} ∂ z ˉ C f = f , − ∂ z C ∗ f = f . (96)
The Fourier multiplier of ∂ z C \partial_z\mathcal{C} ∂ z C is
i ξ 1 + ξ 2 i ξ 1 − ξ 2 ( ξ ≠ 0 ) , \frac{i\xi_1+\xi_2}{i\xi_1-\xi_2}\qquad(\xi\ne0), i ξ 1 − ξ 2 i ξ 1 + ξ 2 ( ξ = 0 ) ,
which has modulus one. Plancherel’s theorem therefore gives ∥ ∂ z C f ∥ 2 = ∥ f ∥ 2 \|\partial_z\mathcal{C}f\|_2=\|f\|_2 ∥ ∂ z C f ∥ 2 = ∥ f ∥ 2 . On fixed bounded disks, local integrability of ∣ z − w ∣ − 1 |z-w|^{-1} ∣ z − w ∣ − 1 and Young’s inequality control the local L 2 L^2 L 2 norm of C f \mathcal{C}f C f . Density extends these identities to compactly supported L 2 L^2 L 2 functions. In particular, C : L 2 ( D R ) → H 1 ( D R ) \mathcal{C}:L^2(D_R)\to H^1(D_R) C : L 2 ( D R ) → H 1 ( D R ) is bounded when the source is extended by zero.
Lemma 4.2. If u ∈ L l o c 2 ( V ) u\in L^2_{\mathrm{loc}}(V) u ∈ L loc 2 ( V ) satisfies ∂ z ˉ u = B u + f \partial_{\bar z}u=Bu+f ∂ z ˉ u = B u + f with smooth B , f B,f B , f , then u u u is smooth. The same holds for − ∂ z u = B ∗ u + f -\partial_z u=B^*u+f − ∂ z u = B ∗ u + f . On each relatively compact subdisk of V V V , every finite Sobolev norm is controlled by ∥ u ∥ 2 \|u\|_2 ∥ u ∥ 2 on a larger subdisk and finitely many seminorms of B B B and f f f .
Proof. For compactly supported v v v , the symbol of ∂ ‾ z \overline{\partial}_z ∂ z has modulus ∣ ξ ∣ / 2 |\xi|/2 ∣ ξ ∣/2 , so
∥ v ∥ H k + 1 ≤ C k ( ∥ ∂ ‾ z v ∥ H k + ∥ v ∥ 2 ) . \|v\|_{H^{k+1}} \le C_k\left(\|\overline{\partial}_zv\|_{H^k}+\|v\|_2\right). ∥ v ∥ H k + 1 ≤ C k ( ∥ ∂ z v ∥ H k + ∥ v ∥ 2 ) .
Apply this estimate to nested cutoffs of u u u . The equation first gives H l o c 1 H^1_{\mathrm{loc}} H loc 1 regularity; the product rule then gives higher Sobolev regularity by induction. Sobolev embedding proves the assertion and the estimates. Complex conjugation gives the result for the second operator. □ \square □
Lemma 4.3. Every smooth B B B admits a smooth invertible transport frame near each point. On a connected planar open set, a solution u ∈ L l o c 2 u\in L^2_{\mathrm{loc}} u ∈ L loc 2 of ∂ ‾ z u = B u \overline{\partial}_zu=Bu ∂ z u = B u that vanishes on a nonempty open subset vanishes identically. The same uniqueness statement holds for − ∂ z u = B ∗ u -\partial_zu=B^*u − ∂ z u = B ∗ u .
Proof. Translate the point to the origin and choose a smooth cutoff supported in D 2 r D_{2r} D 2 r . Let B c u t B_{\mathrm{cut}} B cut be the resulting coefficient, equal to B B B on D r D_r D r . Since
sup z 1 π ∫ D 2 r ∣ z − w ∣ − 1 d A ( w ) ≤ C r , \sup_z\frac{1}{\pi}\int_{D_{2r}}|z-w|^{-1}\,dA(w)\le Cr, z sup π 1 ∫ D 2 r ∣ z − w ∣ − 1 d A ( w ) ≤ C r ,
the equation d = I + C ( B c u t d ) d=I+\mathcal{C}(B_{\mathrm{cut}}d) d = I + C ( B cut d ) is a contraction on bounded continuous matrices for small r r r , and ∥ d − I ∥ ∞ < 1 / 2 \|d-I\|_\infty<1/2 ∥ d − I ∥ ∞ < 1/2 can be imposed. The Cauchy transform of a bounded compactly supported function is continuous, as follows from translation continuity of the locally integrable kernel. Thus d d d is invertible everywhere. Lemma 4.2 makes it smooth, and on D r D_r D r it satisfies ∂ ‾ z d = B d \overline{\partial}_zd=Bd ∂ z d = B d . For a local solution u u u , regularity and the equation imply ∂ ‾ z ( d − 1 u ) = 0 \overline{\partial}_z(d^{-1}u)=0 ∂ z ( d − 1 u ) = 0 . Hence d − 1 u d^{-1}u d − 1 u is holomorphic. The identity theorem propagates vanishing from one small disk to the next throughout the connected open set. Complex conjugation reduces the last equation to the same argument with coefficient − B T -B^T − B T . □ \square □
Weighted estimates and compactness
Fix 1 / 2 < s < 1 1/2<s<1 1/2 < s < 1 and set
L t 2 = { f : ⟨ z ⟩ t f ∈ L 2 } , H = L − s 2 ( C ; C m ) , ϖ = ⟨ z ⟩ − 2 s . L_t^2=\{f:\langle z\rangle^t f\in L^2\},\qquad\mathcal{H}=L_{-s}^2(\mathbb{C};\mathbb{C}^m),\qquad\varpi=\langle z\rangle^{-2s}. L t 2 = { f : ⟨ z ⟩ t f ∈ L 2 } , H = L − s 2 ( C ; C m ) , ϖ = ⟨ z ⟩ − 2 s .
We use ∥ f ∥ L t 2 = ∥ ⟨ z ⟩ t f ∥ L 2 ( C ) \|f\|_{L_t^2}=\|\langle z\rangle^t f\|_{L^2(\mathbb{C})} ∥ f ∥ L t 2 = ∥ ⟨ z ⟩ t f ∥ L 2 ( C ) and ( u , v ) H = ∫ C u ⋅ v ‾ ϖ d A (u,v)_{\mathcal{H}}=\int_{\mathbb{C}}u\cdot\overline{v}\,\varpi\,dA ( u , v ) H = ∫ C u ⋅ v ϖ d A . Multiplication by B B B followed by the Cauchy operator defines the compact operator used in the Fredholm argument.
Lemma 4.4. The operators C , C ∗ : L s 2 → L − s 2 \mathcal{C},\mathcal{C}^*:L_s^2\to L_{-s}^2 C , C ∗ : L s 2 → L − s 2 are bounded. For every compactly supported smooth matrix B B B , K = C B : H → H \mathcal{K}=\mathcal{C}B:\mathcal{H}\to\mathcal{H} K = C B : H → H is compact.
Proof. After conjugation by the weights, the absolute kernel is π − 1 ⟨ z ⟩ − s ∣ z − w ∣ − 1 ⟨ w ⟩ − s \pi^{-1}\langle z\rangle^{-s}|z-w|^{-1}\langle w\rangle^{-s} π − 1 ⟨ z ⟩ − s ∣ z − w ∣ − 1 ⟨ w ⟩ − s . Use a ( z ) = ⟨ z ⟩ − 1 a(z)=\langle z\rangle^{-1} a ( z ) = ⟨ z ⟩ − 1 in Schur’s test. For R = ∣ z ∣ ≥ 2 R=|z|\ge2 R = ∣ z ∣ ≥ 2 , split I ( z ) = R ∫ ∣ z − w ∣ − 1 ⟨ w ⟩ − 1 − s d A ( w ) I(z)=R\int|z-w|^{-1}\langle w\rangle^{-1-s}\,dA(w) I ( z ) = R ∫ ∣ z − w ∣ − 1 ⟨ w ⟩ − 1 − s d A ( w ) into the regions ∣ w ∣ ≤ R / 2 |w|\le R/2 ∣ w ∣ ≤ R /2 , ∣ z − w ∣ ≤ R / 2 |z-w|\le R/2 ∣ z − w ∣ ≤ R /2 , their complement with ∣ w ∣ ≤ 2 R |w|\le2R ∣ w ∣ ≤ 2 R , and ∣ w ∣ ≥ 2 R |w|\ge2R ∣ w ∣ ≥ 2 R . The respective bounds are
C R − 1 ∫ 0 R / 2 r ⟨ r ⟩ − 1 − s d r , C R − 1 − s ∫ 0 R / 2 d r , C R − 1 R − 1 − s R 2 , C ∫ 2 R ∞ r − 1 − s d r . CR^{-1}\int_0^{R/2}r\langle r\rangle^{-1-s}\,dr,\qquad
CR^{-1-s}\int_0^{R/2}dr,\qquad
CR^{-1}R^{-1-s}R^2,\qquad
C\int_{2R}^{\infty}r^{-1-s}\,dr. C R − 1 ∫ 0 R /2 r ⟨ r ⟩ − 1 − s d r , C R − 1 − s ∫ 0 R /2 d r , C R − 1 R − 1 − s R 2 , C ∫ 2 R ∞ r − 1 − s d r .
Each is at most C R − s CR^{-s} C R − s . The integral is uniformly bounded for bounded z z z . Both Schur quotients are thus bounded by C ⟨ z ⟩ 1 − 2 s C\langle z\rangle^{1-2s} C ⟨ z ⟩ 1 − 2 s , which is bounded because s > 1 / 2 s>1/2 s > 1/2 . The absolute kernel of C ∗ \mathcal{C}^* C ∗ is identical.
Suppose supp B ⊂ D R 0 \operatorname{supp}B\subset D_{R_0} supp B ⊂ D R 0 with R 0 ≥ 1 R_0\ge1 R 0 ≥ 1 . A bounded set in H \mathcal{H} H is mapped by multiplication by B B B to a bounded set in L 2 L^2 L 2 on a fixed compact set. The Fourier identity and the weighted estimate bound C B u \mathcal{C}Bu C B u in H 1 H^1 H 1 on every fixed disk. Rellich’s theorem gives local L 2 L^2 L 2 compactness. For the tail, when ∣ z ∣ > 2 R 0 |z|>2R_0 ∣ z ∣ > 2 R 0 we have
∣ C B u ( z ) ∣ ≤ C ∣ z ∣ − 1 ∥ u ∥ H , ∥ C B u ∥ H ( ∣ z ∣ > R ) 2 ≤ C R − 2 s ∥ u ∥ H 2 . |\mathcal{C}Bu(z)|\le C|z|^{-1}\|u\|_{\mathcal{H}},\qquad
\|\mathcal{C}Bu\|_{\mathcal{H}(|z|>R)}^2\le CR^{-2s}\|u\|_{\mathcal{H}}^2. ∣ C B u ( z ) ∣ ≤ C ∣ z ∣ − 1 ∥ u ∥ H , ∥ C B u ∥ H ( ∣ z ∣ > R ) 2 ≤ C R − 2 s ∥ u ∥ H 2 .
The second bound follows by integrating r − 1 − 2 s r^{-1-2s} r − 1 − 2 s from R R R to infinity. Uniform smallness of the tail and compactness on a sufficiently large fixed disk prove compactness in H \mathcal{H} H . □ \square □
The weighted cokernel and polynomial corrections
Lemma 4.5. Let ( I − K ) H ∗ (I-K)_{\mathcal H}^{*} ( I − K ) H ∗ denote the Hilbert adjoint in the weighted inner product just defined. If h ∈ ker ( I − K ) H ∗ h \in\ker(I-K)_{\mathcal H}^{*} h ∈ ker ( I − K ) H ∗ , then ϕ = C ∗ ( ϖ h ) \phi=\mathcal C^{*}(\varpi h) ϕ = C ∗ ( ϖ h ) is well defined, belongs to H \mathcal H H , and satisfies
ϖ h = B ∗ ϕ , ϕ = C ∗ ( B ∗ ϕ ) , − ∂ z ϕ = B ∗ ϕ . (97) \varpi h=B^{*}\phi,\qquad\phi=\mathcal C^{*}(B^{*}\phi),\qquad-\partial_z\phi=B^{*}\phi. \tag*{(97)} ϖ h = B ∗ ϕ , ϕ = C ∗ ( B ∗ ϕ ) , − ∂ z ϕ = B ∗ ϕ . (97) In particular, B ∗ ϕ B^{*}\phi B ∗ ϕ is compactly supported and ϕ \phi ϕ is smooth. For every compactly supported f ∈ L 2 f\in L^{2} f ∈ L 2 ,
( C f , h ) H = ( f , ϕ ) L 2 . (98) (\mathcal C f,h)_{\mathcal H}=(f,\phi)_{L^{2}}. \tag*{(98)} ( C f , h ) H = ( f , ϕ ) L 2 . (98)
Proof. The identity ∥ ϖ h ∥ L 2 = ∥ h ∥ H \|\varpi h\|_{L^{2}}=\|h\|_{\mathcal H} ∥ ϖ h ∥ L 2 = ∥ h ∥ H permits the weighted estimate. Interchanging the kernels gives (98) first for smooth compactly supported data; weighted boundedness and density give the general case. The cokernel condition yields, for every u ∈ H u\in\mathcal H u ∈ H ,
( u , h ) H = ( C B u , h ) H = ( B u , ϕ ) L 2 = ( u , B ∗ ϕ ) L 2 . (u,h)_{\mathcal H}=(\mathcal C Bu,h)_{\mathcal H}=(Bu,\phi)_{L^{2}}=(u,B^{*}\phi)_{L^{2}}. ( u , h ) H = ( C B u , h ) H = ( B u , ϕ ) L 2 = ( u , B ∗ ϕ ) L 2 .
It follows that ϖ h = B ∗ ϕ \varpi h=B^{*}\phi ϖ h = B ∗ ϕ . Substituting this into the definition of ϕ \phi ϕ and using (96) gives the remaining identities. Smoothness follows from Lemma 4.2. □
Lemma 4.6. Let h 1 , … , h r h_{1},\ldots,h_{r} h 1 , … , h r be a basis of ker ( I − K ) H ∗ \ker(I-K)_{\mathcal H}^{*} ker ( I − K ) H ∗ , the finite-dimensional cokernel supplied by Lemma 4.4 . For each j j j , set ϕ j = C ∗ ( ϖ h j ) \phi_{j}=\mathcal C^{*}(\varpi h_{j}) ϕ j = C ∗ ( ϖ h j ) as in Lemma 4.5 , and define
ℓ j ( P ) = ( B P , ϕ j ) L 2 , P ∈ C [ z ] m . \ell_{j}(P)=(BP,\phi_{j})_{L^{2}},\qquad P\in\mathbb C[z]^{m}. ℓ j ( P ) = ( B P , ϕ j ) L 2 , P ∈ C [ z ] m . Then P ↦ ( ℓ 1 ( P ) , … , ℓ r ( P ) ) P\mapsto(\ell_{1}(P),\ldots,\ell_{r}(P)) P ↦ ( ℓ 1 ( P ) , … , ℓ r ( P )) maps onto C r \mathbb C^{r} C r . Its restriction to polynomials of degree at most N N N is onto for some finite N ≥ 0 N\geq0 N ≥ 0 .
Proof. Otherwise, a nonzero cokernel element h h h would have a corresponding ϕ \phi ϕ such that ( B P , ϕ ) = 0 (BP,\phi)=0 ( B P , ϕ ) = 0 for every holomorphic vector polynomial P P P . The coefficients are conjugated according to the convention that the first variable is linear. Put f = B ∗ ϕ f=B^{*}\phi f = B ∗ ϕ . For the standard coordinate vectors e a ∈ C m e_{a}\in\mathbb C^{m} e a ∈ C m , choose P = z k e a P=z^{k}e_{a} P = z k e a and take complex conjugates to obtain
∫ C w ‾ k f ( w ) d A ( w ) = 0 ( k = 0 , 1 , … ) . \int_{\mathbb C}\overline{w}^{k}f(w)\,dA(w)=0\qquad(k=0,1,\ldots). ∫ C w k f ( w ) d A ( w ) = 0 ( k = 0 , 1 , … ) .
The function f f f is compactly supported and belongs to L 1 L^{1} L 1 . Outside a disk containing its support, the geometric series converges absolutely and uniformly, so
ϕ ( z ) = − 1 π ∑ k = 0 ∞ z ‾ − k − 1 ∫ C w ‾ k f ( w ) d A ( w ) = 0. \phi(z)=-\frac{1}{\pi}\sum_{k=0}^{\infty}\overline{z}^{-k-1}\int_{\mathbb C}\overline{w}^{k}f(w)\,dA(w)=0. ϕ ( z ) = − π 1 k = 0 ∑ ∞ z − k − 1 ∫ C w k f ( w ) d A ( w ) = 0.
Uniqueness in Lemma 4.3 , applied to − ∂ z ϕ = B ∗ ϕ -\partial_z\phi=B^{*}\phi − ∂ z ϕ = B ∗ ϕ , gives ϕ = 0 \phi=0 ϕ = 0 , and (97) then gives h = 0 h=0 h = 0 , a contradiction. Since the target is finite dimensional, finitely many polynomial images already span it; their maximum degree gives N N N . If r = 0 r=0 r = 0 , take N = 0 N=0 N = 0 . □
Proof of Theorem 4.1, part 1. By Lemma 4.4 , I − K I-K I − K is Fredholm of index zero. Choose the polynomial degree N N N supplied by Lemma 4.6 and put d = N + 1 d=N+1 d = N + 1 . That lemma gives, for each standard vector e i ∈ C m e_{i}\in\mathbb C^{m} e i ∈ C m , a polynomial Q i Q_{i} Q i of degree at most N N N such that
P i = z d e i + Q i , ℓ j ( P i ) = 0 ( 1 ≤ j ≤ r ) . P_{i}=z^{d}e_{i}+Q_{i},\qquad\ell_{j}(P_{i})=0\qquad(1\leq j\leq r). P i = z d e i + Q i , ℓ j ( P i ) = 0 ( 1 ≤ j ≤ r ) .
By (98) , C ( B P i ) \mathcal C(BP_{i}) C ( B P i ) is orthogonal to the weighted cokernel. The Fredholm alternative gives v i ∈ H v_{i}\in\mathcal H v i ∈ H satisfying
( I − K ) v i = C ( B P i ) . (I-K)v_{i}=\mathcal C(BP_{i}). ( I − K ) v i = C ( B P i ) .
Thus C i = P i + v i C_{i}=P_{i}+v_{i} C i = P i + v i solves ∂ z ‾ C i = B C i \partial_{\overline z}C_{i}=BC_{i} ∂ z C i = B C i , with v i = C ( B C i ) v_{i}=\mathcal C(BC_{i}) v i = C ( B C i ) . Local regularity makes C i C_{i} C i smooth, and expansion outside the source support gives v i = O ( 1 / z ) v_{i}=O(1/z) v i = O ( 1/ z ) . Hence C = ( C 1 , … , C m ) C=(C_{1},\ldots,C_{m}) C = ( C 1 , … , C m ) satisfies
C = z d I + O ( ∣ z ∣ d − 1 ) , det C = z m d ( 1 + O ( ∣ z ∣ − 1 ) ) ( ∣ z ∣ → ∞ ) . (99) C=z^{d}I+O(|z|^{d-1}),\qquad\det C=z^{md}(1+O(|z|^{-1}))\qquad(|z|\to\infty). \tag*{(99)} C = z d I + O ( ∣ z ∣ d − 1 ) , det C = z m d ( 1 + O ( ∣ z ∣ − 1 )) ( ∣ z ∣ → ∞ ) . (99)
In particular, its determinant is not identically zero.
Multilinearity of the determinant, including at singular matrices, gives
∂ z ˉ det C = ( tr B ) det C . \partial_{\bar z}\det C=(\operatorname{tr}B)\det C. ∂ z ˉ det C = ( tr B ) det C .
Set q = C ( tr B ) q=\mathcal{C}(\operatorname{tr}B) q = C ( tr B ) . Then F = e − q det C F=e^{-q}\det C F = e − q det C is entire. Since q = O ( 1 / z ) q=O(1/z) q = O ( 1/ z ) , formula (99) shows that F F F has polynomial growth with leading term z m d z^{md} z m d . Cauchy’s estimates make it a nonzero polynomial of degree m d md m d , whose zeros have finite total multiplicity.
At a zero z 0 z_0 z 0 , choose a local frame d 0 d_0 d 0 from Lemma 4.3 4.3 4.3 and write C = d 0 H C=d_0H C = d 0 H with H H H holomorphic. Extend a nonzero v ∈ ker H ( z 0 ) v\in\ker H(z_0) v ∈ ker H ( z 0 ) to the first column of a constant T ∈ G L ( m , C ) T\in GL(m,\mathbb{C}) T ∈ G L ( m , C ) . The first column of H T HT H T vanishes at z 0 z_0 z 0 . Divide the first column of the global matrix C T CT C T by z − z 0 z-z_0 z − z 0 and retain the other columns. The local expression d 0 H T d_0HT d 0 H T gives a smooth removable extension at z 0 z_0 z 0 , because every component of its holomorphic first column is divisible by z − z 0 z-z_0 z − z 0 . The extended matrix solves the same equation, while its modified determinant e − q det C e^{-q}\det C e − q det C is the previous one multiplied by a nonzero constant and divided by z − z 0 z-z_0 z − z 0 . The total zero multiplicity decreases by one. Repeating finitely many times yields det C = c e q \det C=ce^q det C = c e q with c ≠ 0 c\ne0 c = 0 , and hence a smooth invertible frame on the plane.
Frames on a fixed disk and smooth real parameters
Proof of Theorem 4.1, part 2. Choose L < L ′ < R < R f r L<L'<R<R_{\mathrm{fr}} L < L ′ < R < R fr , with finitely many intermediate radii as needed. Let χ ∈ C c ∞ ( D R f r ) \chi\in C^\infty_c(D_{R_{\mathrm{fr}}}) χ ∈ C c ∞ ( D R fr ) equal one near D R D_R D R , and set B ~ 0 = χ B ( ⋅ , y 0 ) \widetilde B_0=\chi B(\cdot,y_0) B 0 = χ B ( ⋅ , y 0 ) , extended by zero outside D R f r D_{R_{\mathrm{fr}}} D R fr .
Part 1 gives a global frame C 0 C_0 C 0 for B ~ 0 \widetilde B_0 B 0 . On D R D_R D R , define
S 0 f = C 0 C ( 1 D R C 0 − 1 f ) . S_0f=C_0\mathcal{C}(1_{D_R}C_0^{-1}f). S 0 f = C 0 C ( 1 D R C 0 − 1 f ) .
The local estimate following (96) gives a bounded map S 0 : L 2 ( D R ) → H 1 ( D R ) S_0:L^2(D_R)\to H^1(D_R) S 0 : L 2 ( D R ) → H 1 ( D R ) , applied columnwise to matrix fields. Since ∂ z ˉ C 0 = B ( ⋅ , y 0 ) C 0 \partial_{\bar z}C_0=B(\cdot,y_0)C_0 ∂ z ˉ C 0 = B ( ⋅ , y 0 ) C 0 on D R D_R D R , the product rule gives
( ∂ z ˉ − B ( ⋅ , y 0 ) ) S 0 f = f in D R . (100) (\partial_{\bar z}-B(\cdot,y_0))S_0f=f\quad\text{in }D_R. \tag*{(100)} ( ∂ z ˉ − B ( ⋅ , y 0 )) S 0 f = f in D R . (100)
Here the source is extended by zero as an L 2 L^2 L 2 function.
Write B y = B ( ⋅ , y ) B_y=B(\cdot,y) B y = B ( ⋅ , y ) and b y = B y − B y 0 b_y=B_y-B_{y_0} b y = B y − B y 0 , where y y y ranges over a local real coordinate chart near y 0 y_0 y 0 . Smoothness on the product of D R ‾ \overline{D_R} D R and a smaller parameter chart gives
A y = S 0 b y : H 1 ( D R ) → H 1 ( D R ) , ∥ A y ∥ ≤ C ∥ b y ∥ ∞ A_y=S_0b_y:H^1(D_R)\to H^1(D_R),\qquad\lVert A_y\rVert\le C\lVert b_y\rVert_\infty A y = S 0 b y : H 1 ( D R ) → H 1 ( D R ) , ∥ A y ∥ ≤ C ∥ b y ∥ ∞
with norm less than 1 / 2 1/2 1/2 near y 0 y_0 y 0 . Therefore
C y = C 0 + S 0 ( b y C y ) , C y = ( I − A y ) − 1 C 0 (101) C_y=C_0+S_0(b_yC_y),\qquad C_y=(I-A_y)^{-1}C_0 \tag*{(101)} C y = C 0 + S 0 ( b y C y ) , C y = ( I − A y ) − 1 C 0 (101)
defines an H 1 H^1 H 1 family. By (100) , it satisfies ∂ z ˉ C y = B y C y \partial_{\bar z}C_y=B_yC_y ∂ z ˉ C y = B y C y on all of D R D_R D R . The map y ↦ A y y\mapsto A_y y ↦ A y is smooth in operator norm; differentiation of the bounded inverse, or the Neumann series, gives all real parameter derivatives of C y C_y C y in H 1 ( D R ) H^1(D_R) H 1 ( D R ) , continuously in y y y . The Neumann bound is controlled by ∥ B y − B y 0 ∥ L ∞ \lVert B_y-B_{y_0}\rVert_{L^\infty} ∥ B y − B y 0 ∥ L ∞ ; the reference coefficient B ( ⋅ , y 0 ) B(\cdot,y_0) B ( ⋅ , y 0 ) is fixed within the given smooth coefficient class. More quantitatively, ∥ A y ∥ ≤ 1 / 2 \lVert A_y\rVert\le1/2 ∥ A y ∥ ≤ 1/2 and C y − C 0 = ( I − A y ) − 1 S 0 ( b y C 0 ) C_y-C_0=(I-A_y)^{-1}S_0(b_yC_0) C y − C 0 = ( I − A y ) − 1 S 0 ( b y C 0 ) give
∥ C y − C 0 ∥ H 1 ( D R ) ≤ 2 ∥ S 0 ∥ L 2 ( D R ) → H 1 ( D R ) ∥ B y − B y 0 ∥ L ∞ ( D R ) ∥ C 0 ∥ L 2 ( D R ) . (102) \lVert C_y-C_0\rVert_{H^1(D_R)}
\le2\lVert S_0\rVert_{L^2(D_R)\to H^1(D_R)}
\lVert B_y-B_{y_0}\rVert_{L^\infty(D_R)}\lVert C_0\rVert_{L^2(D_R)}. \tag*{(102)} ∥ C y − C 0 ∥ H 1 ( D R ) ≤ 2 ∥ S 0 ∥ L 2 ( D R ) → H 1 ( D R ) ∥ B y − B y 0 ∥ L ∞ ( D R ) ∥ C 0 ∥ L 2 ( D R ) . (102)
For each fixed reference coefficient B ( ⋅ , y 0 ) B(\cdot,y_0) B ( ⋅ , y 0 ) , (102) controls the dependence of C y C_y C y on B ( ⋅ , y ) B(\cdot,y) B ( ⋅ , y ) in a neighborhood of y 0 y_0 y 0 .
To obtain uniform invertibility, put W y = C y − C 0 W_y=C_y-C_0 W y = C y − C 0 . Estimate (102) gives W y → 0 W_y\to0 W y → 0 in H 1 ( D R ) H^1(D_R) H 1 ( D R ) and
∂ z ˉ W y = B y W y + b y C 0 . \partial_{\bar z}W_y=B_yW_y+b_yC_0. ∂ z ˉ W y = B y W y + b y C 0 .
On an intermediate disk D R 1 ⊆ D R D_{R_1}\subseteq D_R D R 1 ⊆ D R , localized Fourier estimates and the product rule give
∥ W y ∥ H 2 ( D R 1 ) ≤ C ( ∥ W y ∥ H 1 ( D R ) + ∥ b y ∥ C 1 ( D R ‾ ) ) ⟶ 0 , \lVert W_y\rVert_{H^2(D_{R_1})}
\le C\bigl(\lVert W_y\rVert_{H^1(D_R)}+\lVert b_y\rVert_{C^1(\overline{D_R})}\bigr)\longrightarrow0, ∥ W y ∥ H 2 ( D R 1 ) ≤ C ( ∥ W y ∥ H 1 ( D R ) + ∥ b y ∥ C 1 ( D R ) ) ⟶ 0 ,
where the C 1 C^1 C 1 norms of B y B_y B y are uniformly bounded after restricting the chart. For L ′ < R 1 L' < R_1 L ′ < R 1 , the two-dimensional Sobolev embedding gives C y → C 0 C_y \to C_0 C y → C 0 in C 0 ( D L ′ ‾ ) C^0(\overline{D_{L'}}) C 0 ( D L ′ ) . Let σ 0 = min D L ′ ‾ σ min ( C 0 ) > 0 \sigma_0 = \min_{\overline{D_{L'}}} \sigma_{\min}(C_0) > 0 σ 0 = min D L ′ σ m i n ( C 0 ) > 0 . Restrict the parameter chart so that sup D L ′ ‾ ∣ C y − C 0 ∣ < σ 0 / 2 \sup_{\overline{D_{L'}}} \lvert C_y-C_0\rvert< \sigma_0/2 sup D L ′ ∣ C y − C 0 ∣ < σ 0 /2 . All C y C_y C y are then invertible on this fixed disk. In the Euclidean operator norm the singular-value inequality σ min ( C y ) ≥ σ min ( C 0 ) − ∥ C y − C 0 ∥ \sigma_{\min}(C_y) \ge\sigma_{\min}(C_0)-\lVert C_y-C_0\rVert σ m i n ( C y ) ≥ σ m i n ( C 0 ) − ∥ C y − C 0 ∥ therefore yields the quantitative uniform bound
sup y ∈ U sup z ∈ D L ′ ∥ C y ( z ) − 1 ∥ ≤ 2 σ 0 − 1 , (103) \sup_{y\in U}\sup_{z\in D_{L'}} \lVert C_y(z)^{-1}\rVert\le2\sigma_0^{-1}, \tag*{(103)} y ∈ U sup z ∈ D L ′ sup ∥ C y ( z ) − 1 ∥ ≤ 2 σ 0 − 1 , (103)
after shrinking U U U if needed. This is the bound on the inverse frame used when compact parameter subcharts are selected later.
For every parameter multi-index β \beta β , the H 1 H^1 H 1 derivatives already constructed satisfy
∂ z ‾ ∂ y β C y = ∑ γ ≤ β ( β γ ) ( ∂ y γ B y ) ( ∂ y β − γ C y ) . \partial_{\overline{z}}\partial_y^\beta C_y = \sum_{\gamma\le\beta} \binom{\beta}{\gamma}(\partial_y^\gamma B_y)(\partial_y^{\beta-\gamma}C_y). ∂ z ∂ y β C y = γ ≤ β ∑ ( γ β ) ( ∂ y γ B y ) ( ∂ y β − γ C y ) .
Fix this parameter neighborhood after the invertibility step. Writing u β = ∂ y β C y u_\beta=\partial_y^\beta C_y u β = ∂ y β C y , an induction on the spatial Sobolev order on nested subdisks bounds all ∣ β ∣ ≤ q \lvert\beta\rvert\le q ∣ β ∣ ≤ q , for arbitrary finite q , k q,k q , k , in H k H^k H k by finitely many coefficient seminorms and the previously obtained norms on larger disks. Applying the same estimates to u β ( y ) − u β ( y ∗ ) u_\beta(y)-u_\beta(y_*) u β ( y ) − u β ( y ∗ ) gives continuity in the interior H k H^k H k spaces. The H 1 H^1 H 1 identity
u β ( y + t e j ) − u β ( y ) = ∫ 0 t u β + e j ( y + s e j ) d s u_\beta(y+te_j)-u_\beta(y)=\int_0^t u_{\beta+e_j}(y+se_j)\,ds u β ( y + t e j ) − u β ( y ) = ∫ 0 t u β + e j ( y + s e j ) d s
therefore holds in H k H^k H k as well, by continuous H k H^k H k -valued integration and injectivity of the embedding. Apply the preceding estimates through order q + 1 q+1 q + 1 to cover the derivative in the integrand. Thus the initially chosen family C y C_y C y is smooth with values in every interior H k H^k H k space. Sobolev embedding for sufficiently large k k k gives joint continuity of all mixed derivatives. Hence C C C is jointly smooth. The smooth map A ↦ A − 1 A\mapsto A^{-1} A ↦ A − 1 on G L ( m , C ) GL(m,\mathbb{C}) G L ( m , C ) gives joint smoothness of C − 1 C^{-1} C − 1 ; equivalently, differentiate ∂ C − 1 = − C − 1 ( ∂ C ) C − 1 \partial C^{-1}=-C^{-1}(\partial C)C^{-1} ∂ C − 1 = − C − 1 ( ∂ C ) C − 1 . Boundedness of continuous derivatives on D L ‾ × K \overline{D_L}\times K D L × K proves (95) . A finite subcover treats compact parameter sets, pairing each frame with its own inverse.
If g g g is prescribed, a = g / det C y a=g/\det C_y a = g / det C y is jointly smooth, nonvanishing, and holomorphic in z z z . Multiplying the first column by a a a gives determinant g g g . The mixed seminorm bounds for the new frame and its inverse follow from smoothness and compactness. □
Proposition 4.7. Let J ≥ 1 J\ge1 J ≥ 1 be an integer. If B ∈ C J + 3 ( C ; M m ( C ) ) B\in C^{J+3}(\mathbb{C};M_m(\mathbb{C})) B ∈ C J + 3 ( C ; M m ( C )) , part 1 of Theorem 4.1 holds with an invertible C J + 2 C^{J+2} C J + 2 frame. If B ∈ C J + 3 ( D R f r × Y ; M m ( C ) ) B\in C^{J+3}(D_{R_{\mathrm{fr}}}\times Y;M_m(\mathbb{C})) B ∈ C J + 3 ( D R fr × Y ; M m ( C )) , part 2 holds with a jointly C J C^J C J frame and inverse, and (95) holds through order J J J on compact subcharts. The frame equation holds classically at the stated finite regularity.
Proof. Put r = J + 3 r=J+3 r = J + 3 . We first record the finite-order version of the local regularity argument. For B ∈ C r B\in C^r B ∈ C r , a local L 2 L^2 L 2 solution of ∂ z ‾ u = B u + f \partial_{\overline{z}}u=Bu+f ∂ z u = B u + f , with f ∈ H l o c r f\in H^r_{\mathrm{loc}} f ∈ H loc r , belongs to H l o c r + 1 H^{r+1}_{\mathrm{loc}} H loc r + 1 . Indeed, the cutoff Fourier estimate in Lemma 4.2 gives H 1 H^1 H 1 first. At the k k k th step, 0 ≤ k ≤ r 0\le k\le r 0 ≤ k ≤ r , the product estimate on a larger compact subdisk is
∥ B u ∥ H k ≤ C k ∥ B ∥ C k ∥ u ∥ H k . (104) \lVert Bu\rVert_{H^k}\le C_k\lVert B\rVert_{C^k}\lVert u\rVert_{H^k}. \tag*{(104)} ∥ B u ∥ H k ≤ C k ∥ B ∥ C k ∥ u ∥ H k . (104)
Nested cutoffs and induction give the claim, with only the displayed finite coefficient norms. The same argument applies to the adjoint system. In two dimensions, H l o c r + 1 ⊂ C l o c r − 1 H^{r+1}_{\mathrm{loc}}\subset C^{r-1}_{\mathrm{loc}} H loc r + 1 ⊂ C loc r − 1 .
The small-disk contraction in Lemma 4.3 uses only boundedness of B B B . It therefore produces continuous invertible local frames, which the preceding argument makes C r − 1 C^{r-1} C r − 1 . Conjugation by such a frame reduces a homogeneous solution to a holomorphic function, so the local uniqueness statement remains valid. The weighted estimates, tail bound and compactness proof in Lemma 4.4 also use only boundedness and compact support of B B B . The weighted cokernel identity and polynomial elimination consequently hold for this B B B : the adjoint solution has H l o c r + 1 H_{\mathrm{loc}}^{r+1} H loc r + 1 regularity, its source is compactly supported, and local-frame uniqueness proves the same surjectivity of the polynomial moments.
Apply the Fredholm construction of part 1 to obtain the columns C i = P i + E ( B C i ) C_i=P_i+\mathcal{E}(BC_i) C i = P i + E ( B C i ) . They are H l o c r + 1 H_{\mathrm{loc}}^{r+1} H loc r + 1 , have the same polynomial behavior at infinity, and satisfy the determinant identity. The function exp ( − E tr B ) det C \exp(-\mathcal{E}\operatorname{tr}B)\det C exp ( − E tr B ) det C is therefore an entire polynomial. At each determinant zero write C = d 0 H C=d_0H C = d 0 H in a local invertible C r − 1 C^{r-1} C r − 1 frame. The matrix H H H is holomorphic. Dividing a vanishing column of H H H by z − z 0 z-z_0 z − z 0 preserves holomorphy; multiplying back by d 0 d_0 d 0 gives a C r − 1 C^{r-1} C r − 1 removable extension, not a division of an arbitrary finite-regularity function. The same finite zero-removal argument yields a global invertible C r − 1 = C J + 2 C^{r-1}=C^{J+2} C r − 1 = C J + 2 frame.
For the parameter statement, fix y 0 y_0 y 0 , choose the same nested disks as in part 2, and obtain the reference frame C 0 C_0 C 0 by this finite-order construction. The operator S 0 f = C 0 E ( χ D R C 0 − 1 f ) S_0f=C_0\mathcal{E}(\chi_{D_R}C_0^{-1}f) S 0 f = C 0 E ( χ D R C 0 − 1 f ) remains bounded L 2 ( D R ) → H 1 ( D R ) L^2(D_R)\to H^1(D_R) L 2 ( D R ) → H 1 ( D R ) . The family A y = S 0 ( B y − B y 0 ) A_y=S_0(B_y-B_{y_0}) A y = S 0 ( B y − B y 0 ) is C r C^r C r in operator norm on H 1 ( D R ) H^1(D_R) H 1 ( D R ) . After shrinking the parameter neighborhood, ∥ A y ∥ ≤ 1 / 2 \|A_y\|\leq1/2 ∥ A y ∥ ≤ 1/2 , and C y = ( I − A y ) − 1 C 0 C_y=(I-A_y)^{-1}C_0 C y = ( I − A y ) − 1 C 0 is C r C^r C r with values in H 1 ( D R ) H^1(D_R) H 1 ( D R ) . For parameter derivatives u β = ∂ y β C y u_\beta=\partial_y^\beta C_y u β = ∂ y β C y ,
∂ z ˉ u β = B y u β + ∑ 0 < γ ≤ β ( β γ ) ( ∂ y γ B y ) u β − γ . \partial_{\bar z}u_\beta=B_yu_\beta+\sum_{0<\gamma\leq\beta}\binom{\beta}{\gamma}(\partial_y^\gamma B_y)u_{\beta-\gamma}. ∂ z ˉ u β = B y u β + 0 < γ ≤ β ∑ ( γ β ) ( ∂ y γ B y ) u β − γ .
Using (104) , induction in the spatial Sobolev order, with nested disks, gives
u β ∈ C ( U ; H k ( D R 1 ) ) , k ≥ 1 , ∣ β ∣ + k ≤ r , R 1 < R . (105) u_\beta\in C(U;H^k(D_{R_1})),\qquad k\geq1,\quad|\beta|+k\leq r,\quad R_1<R. \tag*{(105)} u β ∈ C ( U ; H k ( D R 1 )) , k ≥ 1 , ∣ β ∣ + k ≤ r , R 1 < R . (105)
For continuity, apply the same estimates to the difference at two parameter points; all coefficient derivatives used have total order at most ∣ β ∣ + k − 1 |\beta|+k-1 ∣ β ∣ + k − 1 . The fundamental theorem of calculus in H 1 H^1 H 1 then identifies parameter derivatives in H k H^k H k whenever ∣ β ∣ + 1 + k ≤ r |\beta|+1+k\leq r ∣ β ∣ + 1 + k ≤ r , using continuous H k H^k H k -valued integration. For ∣ α ∣ + ∣ β ∣ ≤ J |\alpha|+|\beta|\leq J ∣ α ∣ + ∣ β ∣ ≤ J , take k = ∣ α ∣ + 2 k=|\alpha|+2 k = ∣ α ∣ + 2 and, when needed, one additional parameter derivative. The total order is at most J + 3 = r J+3=r J + 3 = r . Sobolev embedding on the two-dimensional disk therefore gives joint continuity of every mixed derivative through order J J J . The H 2 H^2 H 2 convergence to C 0 C_0 C 0 on a smaller disk gives uniform invertibility exactly as in (103) . Inversion is a smooth finite-dimensional matrix operation, so the inverse frame is also jointly C J C^J C J . Compactness gives the asserted finite seminorm bounds. For a prescribed smooth nonvanishing determinant satisfying the trace equation, the holomorphic column multiplication at the end of part 2 preserves this joint C J C^J C J regularity. □
Proposition 4.8. Fix 0 < L < R f r 0<L<R_{\mathrm{fr}} 0 < L < R fr , a compact parameter set Y ⊂ R d Y\subset\mathbb{R}^d Y ⊂ R d , a bounded open neighborhood Y + Y^+ Y + of Y Y Y with dist ( Y , ∂ Y + ) ≥ 0 \operatorname{dist}(Y,\partial Y^+)\geq0 dist ( Y , ∂ Y + ) ≥ 0 , a matrix size m m m , and M B < ∞ M_B<\infty M B < ∞ . Let B B B range over smooth matrix fields on a neighborhood of D R f r ‾ × Y + ‾ \overline{D_{R_{\mathrm{fr}}}}\times\overline{Y^+} D R fr × Y + with ∥ B ∥ C 12 ( D R f r ‾ × Y + ‾ ) ≤ M B \|B\|_{C^{12}(\overline{D_{R_{\mathrm{fr}}}}\times\overline{Y^+})}\leq M_B ∥ B ∥ C 12 ( D R fr × Y + ) ≤ M B . There is a fixed finite cover of Y Y Y by parameter neighborhoods such that, for each B B B , one can choose a smooth invertible frame on every member of this cover, satisfying ∂ z ˉ C = B C \partial_{\bar z}C=BC ∂ z ˉ C = B C near D L ‾ \overline{D_L} D L and
max ∣ α ∣ ≤ 6 , ∣ β ∣ ≤ 6 sup ( ∣ ∂ z α ∂ y β C ∣ + ∣ ∂ z α ∂ y β C − 1 ∣ ) ≤ C f r . (106) \max_{\lvert\alpha\rvert\leq6,\ \lvert\beta\rvert\leq6}\sup\left(\lvert\partial_z^\alpha\partial_y^\beta C\rvert+\lvert\partial_z^\alpha\partial_y^\beta C^{-1}\rvert\right)\leq C_{\mathrm{fr}}. \tag*{(106)} ∣ α ∣ ≤ 6 , ∣ β ∣ ≤ 6 max sup ( ∣ ∂ z α ∂ y β C ∣ + ∣ ∂ z α ∂ y β C − 1 ∣ ) ≤ C fr . (106)
Here ∂ z α \partial_z^\alpha ∂ z α denotes derivatives in the two real leaf coordinates, and the supremum is on the closed smaller parameter neighborhoods and D L ‾ \overline{D_L} D L . The constant and cover depend only on the fixed geometric and parameter data and the displayed C 12 C^{12} C 12 bound. When B = 0 B=0 B = 0 , the frames may be chosen with determinant one. The same assertion holds on a compact smooth parameter manifold after fixing finitely many coordinate charts.
Proof. Choose L < R < S < R f r L<R<S<R_{\mathrm{fr}} L < R < S < R fr and cutoffs χ o u t , χ \chi_{\mathrm{out}},\chi χ out , χ with χ o u t ∈ C c ∞ ( D R f r ) \chi_{\mathrm{out}}\in C^\infty_c(D_{R_{\mathrm{fr}}}) χ out ∈ C c ∞ ( D R fr ) equal to one near D S ‾ \overline{D_S} D S and χ ∈ C c ∞ ( D S ) \chi\in C^\infty_c(D_S) χ ∈ C c ∞ ( D S ) equal to one near D R ‾ \overline{D_R} D R . For $0\leq k\leq6$ and v ∈ C k ( D S ‾ ) v\in C^k(\overline{D_S}) v ∈ C k ( D S ) , the compact source $\chi v has a C k C^k C k zero extension. Distributional integration by parts gives
∂ z α C ( χ v ) = C ( ∂ z α ( χ v ) ) , ∥ C ( χ v ) ∥ C k ( D S ‾ ) ≤ C S , k , χ ∥ v ∥ C k ( D S ‾ ) . (107) \partial_z^\alpha\mathcal{C}(\chi v)=\mathcal{C}(\partial_z^\alpha(\chi v)),\qquad\|\mathcal{C}(\chi v)\|_{C^k(\overline{D_S})}\leq C_{S,k,\chi}\|v\|_{C^k(\overline{D_S})}. \tag*{(107)} ∂ z α C ( χ v ) = C ( ∂ z α ( χ v )) , ∥ C ( χ v ) ∥ C k ( D S ) ≤ C S , k , χ ∥ v ∥ C k ( D S ) . (107)
Indeed, the absolute integral of ∣ z − w ∣ − 1 |z-w|^{-1} ∣ z − w ∣ − 1 over D S D_S D S is uniformly bounded for z ∈ D S ‾ z\in\overline{D_S} z ∈ D S . Translation continuity of this locally integrable kernel proves continuity of each displayed convolution. These bounds justify the distributional derivative identities as classical identities through order k k k .
The frozen coefficients have the compact closure
R = { χ o u t B ( ⋅ , y ) : B as above , y ∈ Y + ‾ } C 9 ( C ) . (108) \mathcal{R}=\left\{\chi_{\mathrm{out}}B(\cdot,y):B\text{ as above},\ y\in\overline{Y^+}\right\}^{C^9(\mathbb{C})}. \tag*{(108)} R = { χ out B ( ⋅ , y ) : B as above , y ∈ Y + } C 9 ( C ) . (108)
To check compactness, their supports lie in the fixed compact support of χ o u t \chi_{\mathrm{out}} χ out , all derivatives through order twelve are bounded, and all derivatives through order nine are equicontinuous. Successive Arzelà–Ascoli extraction for the finitely many derivatives, together with the fundamental theorem of calculus, identifies every limit as a C 9 C^9 C 9 coefficient. This proves sequential compactness of the closure in the C 9 C^9 C 9 norm.
For any B ∘ ∈ R B^\circ\in\mathcal{R} B ∘ ∈ R , Proposition 4.7 with J = 6 J=6 J = 6 gives an invertible C 8 C^8 C 8 reference frame C ∘ ( z ) C^\circ(z) C ∘ ( z ) on the plane. It is independent of y y y . For a nearby coefficient define, on C k ( D S ‾ ) C^k(\overline{D_S}) C k ( D S ) ,
L y v = C ∘ C ( χ ( C ∘ ) − 1 ( B ( ⋅ , y ) − B ∘ ) v ) , \mathcal{L}_y v=C^\circ\mathcal{C}\left(\chi(C^\circ)^{-1}(B(\cdot,y)-B^\circ)v\right), L y v = C ∘ C ( χ ( C ∘ ) − 1 ( B ( ⋅ , y ) − B ∘ ) v ) ,
C y = ( I − L y ) − 1 C ∘ . (109) C_y=(I-\mathcal{L}_y)^{-1}C^\circ. \tag*{(109)} C y = ( I − L y ) − 1 C ∘ . (109)
By (107) , there is a reference-dependent c ∘ < ∞ c_\circ<\infty c ∘ < ∞ such that, simultaneously for 0 ≤ k ≤ 6 0\leq k\leq6 0 ≤ k ≤ 6 ,
∥ L y ∥ C k → C k ≤ c ∘ ∥ B ( ⋅ , y ) − B ∘ ∥ C 6 ( D S ‾ ) . \|\mathcal{L}_y\|_{C^k\to C^k}\leq c_\circ\|B(\cdot,y)-B^\circ\|_{C^6(\overline{D_S})}. ∥ L y ∥ C k → C k ≤ c ∘ ∥ B ( ⋅ , y ) − B ∘ ∥ C 6 ( D S ) .
Choose a positive radius δ ∘ \delta_\circ δ ∘ so that the latter norm is less than 1 / 2 1/2 1/2 and
2 c ∘ δ ∘ ∥ C ∘ ∥ C 6 ( D S ‾ ) < 1 2 min D S ‾ σ min ( C ∘ ) . 2c_\circ\delta_\circ\|C^\circ\|_{C^6(\overline{D_S})}<\frac{1}{2}\min_{\overline{D_S}}\sigma_{\min}(C^\circ). 2 c ∘ δ ∘ ∥ C ∘ ∥ C 6 ( D S ) < 2 1 D S min σ m i n ( C ∘ ) .
The Neumann series then defines one frame in all these spaces, ∥ C y − C ∘ ∥ C 6 ≤ 2 c ∘ δ ∘ ∥ C ∘ ∥ C 6 \|C_y-C^\circ\|_{C^6}\leq2c_\circ\delta_\circ\|C^\circ\|_{C^6} ∥ C y − C ∘ ∥ C 6 ≤ 2 c ∘ δ ∘ ∥ C ∘ ∥ C 6 , and C y C_y C y is invertible. Applying ∂ z ˉ − B ∘ \partial_{\bar z}-B^\circ ∂ z ˉ − B ∘ to C y = C ∘ + L y C y C_y=C^\circ+\mathcal{L}_yC_y C y = C ∘ + L y C y proves the frame equation on D R D_R D R .
Cover R \mathcal{R} R by the C 6 ( D S ‾ ) C^6(\overline{D_S}) C 6 ( D S ) balls of radii δ ∘ / 4 \delta_\circ/4 δ ∘ /4 and choose a finite subcover. These are open sets in the C 9 C^9 C 9 topology used above. All their references, radii and frame constants are now fixed. The coefficient bound gives, on sufficiently small parameter balls,
∥ B ( ⋅ , y ) − B ( ⋅ , y 0 ) ∥ C 6 ( D S ‾ ) ≤ C M B ∣ y − y 0 ∣ . \|B(\cdot,y)-B(\cdot,y_0)\|_{C^6(\overline{D_S})}\leq C M_B|y-y_0|. ∥ B ( ⋅ , y ) − B ( ⋅ , y 0 ) ∥ C 6 ( D S ) ≤ C M B ∣ y − y 0 ∣.
Consequently a parameter radius independent of B B B can be chosen so that a reference covering B ( ⋅ , y 0 ) B(\cdot,y_0) B ( ⋅ , y 0 ) remains within its δ ∘ \delta_\circ δ ∘ neighborhood for all y y y in this ball. Decrease the radius also to stay inside Y + Y^+ Y + . A fixed finite cover of Y Y Y by smaller such balls, each with a larger concentric ball still admissible, supplies the neighborhoods in the statement.
On any one larger ball, y ↦ L y y\mapsto\mathcal{L}_y y ↦ L y is smooth in the operator norm on C 6 ( D S ‾ ) C^6(\overline{D_S}) C 6 ( D S ) for each smooth B B B . Differentiating the inverse gives, for 0 < ∣ β ∣ ≤ 6 0<|\beta|\leq6 0 < ∣ β ∣ ≤ 6 ,
( I − L y ) ∂ y β C y = ∑ 0 < γ ≤ β ( β γ ) ( ∂ y γ L y ) ∂ y β − γ C y . (110) (I-\mathcal{L}_y)\partial_y^\beta C_y=\sum_{0<\gamma\leq\beta}\binom{\beta}{\gamma}(\partial_y^\gamma\mathcal{L}_y)\partial_y^{\beta-\gamma}C_y. \tag*{(110)} ( I − L y ) ∂ y β C y = 0 < γ ≤ β ∑ ( γ β ) ( ∂ y γ L y ) ∂ y β − γ C y . (110)
The operators on the right have C 6 → C 6 C^6\to C^6 C 6 → C 6 norms bounded using ∂ z ≤ 6 ∂ y ≤ 6 B \partial_z^{\leq6}\partial_y^{\leq6}B ∂ z ≤ 6 ∂ y ≤ 6 B , whose total order is at most twelve. Induction in ∣ β ∣ |\beta| ∣ β ∣ therefore bounds every mixed derivative in (106) , uniformly over the finite references. The lower singular-value bound and differentiated identities ∂ C y − 1 = − C y − 1 ( ∂ C y ) C y − 1 \partial C_y^{-1}=-C_y^{-1}(\partial C_y)C_y^{-1} ∂ C y − 1 = − C y − 1 ( ∂ C y ) C y − 1 give the same bounds for the inverse.
The reference coefficients have the stated finite regularity, while the resulting frames are smooth on D R D_R D R whenever B B B is smooth. In fact the operator formula is smooth in y y y with values in C 6 C^6 C 6 . Each parameter derivative satisfies the differentiated frame equation; Lemma 4.2, inductively in the parameter order, gives every leaf derivative on smaller disks. Difference estimates and the fundamental theorem of calculus, as in the proof of Theorem 4.1 , identify these as jointly smooth derivatives. If tr B = 0 \operatorname{tr} B=0 tr B = 0 , the nonvanishing function det C y \det C_y det C y is leaf-holomorphic. Multiplying the first column by its reciprocal normalizes the determinant; the inverse-frame bound controls all sixth-order derivatives of this normalization. Finally a fixed finite coordinate cover proves the manifold version. □
Remark 4.9. Choose smooth leaf coordinates in a local real chart of directions. The two-, three-, and four-component systems below then have the form above. For a transport field ∂ x 1 + i ∂ x 2 = 2 ∂ z ˉ \partial_{x_1}+i\partial_{x_2}=2\partial_{\bar z} ∂ x 1 + i ∂ x 2 = 2 ∂ z ˉ , divide its coefficient by two. Direction and transverse coordinates together form the real parameter y y y . For a prescribed disk and fixed material, choose the parameter charts and frames before the transverse test b b b , the leaf polynomial test g g g , and the CGO parameter used in § 6 . All finite mixed seminorm bounds apply to this single chosen frame and its inverse. A finite collection of local real parameter charts suffices.
Frames for the transport systems
We apply Theorem 4.1 to the reduced system, its dual and quotient systems, and the moving directions in the local right inverse. Use the first-order representation (147) and set A θ p h y s = ∑ k = 1 3 θ k A k p h y s A_\theta^{\mathrm{phys}}=\sum_{k=1}^{3}\theta_k A_k^{\mathrm{phys}} A θ phys = ∑ k = 1 3 θ k A k phys and T θ = 2 D θ + A θ p h y s T_\theta=2D_\theta+A_\theta^{\mathrm{phys}} T θ = 2 D θ + A θ phys , where D θ = θ ⋅ ∇ D_\theta=\theta\cdot\nabla D θ = θ ⋅ ∇ . Thus T θ T_\theta T θ is the leading four-component transport operator; D θ D_\theta D θ is the directional derivative without the Fourier factor − i -i − i .
Proposition 4.10. Fix a sufficiently large leaf disk D L D_L D L and a center for the direction and transverse parameters. The two-, three-, and four-component systems used in § 6 and 7 admit smooth invertible frames on this disk. Each frame and its inverse are jointly smooth in a local real parameter chart and have all prescribed finite mixed seminorms. They can be fixed before all tests b b b , g g g and all small parameters.
Proof. Choose a smooth local orthonormal frame θ = e 1 + i e 2 \theta=e_1+ie_2 θ = e 1 + i e 2 , ν θ = e 1 × e 2 \nu_\theta=e_1\times e_2 ν θ = e 1 × e 2 , and write x = z 1 e 1 + z 2 e 2 + t ν θ x=z_1e_1+z_2e_2+t\nu_\theta x = z 1 e 1 + z 2 e 2 + t ν θ , z = z 1 + i z 2 z=z_1+iz_2 z = z 1 + i z 2 . Here ν θ \nu_\theta ν θ is the real unit normal to the oriented θ \theta θ -plane, distinct from the outward boundary normal ν \nu ν . Then D θ = 2 ∂ z ˉ D_\theta=2\partial_{\bar z} D θ = 2 ∂ z ˉ . The four-component leading transport equation T θ C = 0 T_\theta C=0 T θ C = 0 becomes
∂ z ˉ C = − 1 4 A θ p h y s C . \partial_{\bar z}C=-\frac{1}{4}A_\theta^{\mathrm{phys}}C. ∂ z ˉ C = − 4 1 A θ phys C .
The dual equation ( 2 D θ − ( A θ p h y s ) T ) Y = 0 (2D_\theta-(A_\theta^{\mathrm{phys}})^T)Y=0 ( 2 D θ − ( A θ phys ) T ) Y = 0 has planar coefficient + ( A θ p h y s ) T / 4 +(A_\theta^{\mathrm{phys}})^T/4 + ( A θ phys ) T /4 . The three-component quotient coefficient P P P is the matrix induced by A θ p h y s / 2 A_\theta^{\mathrm{phys}}/2 A θ phys /2 on C 4 / C ( θ , 0 ) T \mathbb{C}^4/\mathbb{C}(\theta,0)^T C 4 / C ( θ , 0 ) T , in spatially constant local quotient coordinates; its explicit formula is given in Lemma 6.11 . Thus D θ X + P X = 0 D_\theta X+PX=0 D θ X + P X = 0 has planar coefficient − P / 2 -P/2 − P /2 . The two-component system in Lemma 6.7 is treated in the same way. All these coefficients are jointly smooth on any fixed larger disk and local real parameter chart. Choose L 1 < L R f L_1<L_{R_f} L 1 < L R f and a fixed cutoff in the leaf coordinates, equal to one near D L 1 D_{L_1} D L 1 and compactly supported in D R f D_{R_f} D R f . Multiplication by this cutoff and zero extension give smooth parameter families with common compact support, agreeing with the original systems on the required disk. Theorem 4.1 gives local frames and their inverses for these families. The frames are independent of b , g , h b,g,h b , g , h and apply to every fixed test whose transverse support lies in the chosen chart. A finite subcover treats compact parameter sets. Proposition 7.6 subsequently glues the relative fields by uniqueness on their common open shell. □
Proposition 4.11. For the near-characteristic symbols of the local right inverse, mixed derivatives of each frame and its inverse through order ten in the leaf and transverse variables and order two in the direction variables suffice for eight x x x derivatives and two ξ \xi ξ derivatives of the symbols. Theorem 4.1 with total order J = 12 J=12 J = 12 supplies these bounds.
Proof. On the fixed spatial support, write the moving leaf coordinates as Z ( x , η ) = ( z 1 , z 2 , t ) Z(x,\eta)=(z_1,z_2,t) Z ( x , η ) = ( z 1 , z 2 , t ) , with local real direction coordinates η \eta η . Frequency enters through η = η ( ξ , τ ) \eta=\eta(\xi,\tau) η = η ( ξ , τ ) and the cutoff functions, and in the near region ∂ η k η = O ( τ − k ) \partial_{\eta}^{k}\eta= O(\tau^{-k}) ∂ η k η = O ( τ − k ) , k = 1 , 2 k=1,2 k = 1 , 2 . A frequency derivative acts by the chain rule through
( ∂ Z C ) ∂ ξ Z + ( ∂ η C ) ∂ ξ η (\partial_Z C)\partial_{\xi}Z+(\partial_{\eta}C)\partial_{\xi}\eta ( ∂ Z C ) ∂ ξ Z + ( ∂ η C ) ∂ ξ η
and adds at most one spatial or direction derivative. After eight x x x derivatives and two ξ \xi ξ derivatives, ten spatial derivatives and two direction derivatives therefore suffice. Derivatives of the frequency cutoffs are estimated in the local right inverse construction. On the fixed compact subcharts and disk, J = 12 J=12 J = 12 in Theorem 4.1 bounds all the required terms. Compactness also controls the fixed derivatives of the transverse weights and coordinate changes. □
A local right inverse for variable coefficients
This section constructs a source-independent linear right inverse for the variable-coefficient conjugated operator P δ P_\delta P δ in (111) . For each sufficiently large τ \tau τ and each direction in the fixed compact parameter family, we select a linear operator R δ R_\delta R δ independently of the source. It satisfies
P δ R δ f = f in Ω 1 , τ ∥ R δ f ∥ L 2 ( Ω 0 ) + ∥ R δ f ∥ H 1 ( Ω 0 ) ≤ C ∥ f ∥ L 2 ( Ω 0 ) P_\delta R_\delta f=f\quad\text{in }\Omega_1,\qquad\tau\lVert R_\delta f\rVert_{L^2(\Omega_0)}+\lVert R_\delta f\rVert_{H^1(\Omega_0)}\leq C\lVert f\rVert_{L^2(\Omega_0)} P δ R δ f = f in Ω 1 , τ ∥ R δ f ∥ L 2 ( Ω 0 ) + ∥ R δ f ∥ H 1 ( Ω 0 ) ≤ C ∥ f ∥ L 2 ( Ω 0 )
for every f ∈ L 2 ( Ω 0 ; C m ) f\in L^2(\Omega_0;\mathbb{C}^m) f ∈ L 2 ( Ω 0 ; C m ) supported in K s r c K_{\mathrm{src}} K src . The construction combines the weighted free inverse E δ E_\delta E δ with local planar transport frames C α C_\alpha C α . A finite partition of the direction parameters and a near–far decomposition in frequency space produce a parametrix: in the near-frequency region the transport equations cancel the leading first-order contribution, whereas in the far-frequency region the free symbol is separated from its characteristic set. The resulting localized error satisfies
∥ T τ , γ ∥ L 2 → L 2 ≤ C ( τ − γ + τ γ − 1 ) , 0 < γ < 1 , \lVert T_{\tau,\gamma}\rVert_{L^2\to L^2}\leq C\left(\tau^{-\gamma}+\tau^{\gamma-1}\right),\qquad0<\gamma<1, ∥ T τ , γ ∥ L 2 → L 2 ≤ C ( τ − γ + τ γ − 1 ) , 0 < γ < 1 ,
as quantified in (134) . A Neumann-series correction gives the exact right inverse. The construction is then extended to finite-regularity coefficients, and perturbation and compactness yield uniform estimates on bounded coefficient classes. These results provide the right-inverse estimates used to construct exact CGO solutions in § 5.7 .
Geometry and coefficient bounds
5.1. Geometry and coefficient bounds. Use D = − i ∇ D=-i\nabla D = − i ∇ , with the pairing convention of § 1.6 ; let m m m be the matrix size, and write
P δ = ( D + δ ) 2 + 2 ∑ j = 1 3 A j ( x ) ( D j + δ j ) + q ( x ) , (111) P_\delta=(D+\delta)^2+2\sum_{j=1}^{3}A_j(x)(D_j+\delta_j)+q(x), \tag*{(111)} P δ = ( D + δ ) 2 + 2 j = 1 ∑ 3 A j ( x ) ( D j + δ j ) + q ( x ) , (111)
δ = τ ( v 0 + i v 1 ) , τ ≥ 1 , ∣ v 0 ∣ = ∣ v 1 ∣ = 1 , v 0 ⋅ v 1 = 0. \delta=\tau(v_0+iv_1),\qquad\tau\geq1,\qquad\lvert v_0\rvert=\lvert v_1\rvert=1,\qquad v_0\cdot v_1=0. δ = τ ( v 0 + i v 1 ) , τ ≥ 1 , ∣ v 0 ∣ = ∣ v 1 ∣ = 1 , v 0 ⋅ v 1 = 0.
The matrices A j , q A_j,q A j , q are fixed smooth m × m m\times m m × m coefficients. For w ∈ C 3 w\in\mathbb{C}^3 w ∈ C 3 , write A w = ∑ j w j A j A_w=\sum_j w_jA_j A w = ∑ j w j A j ; in expressions such as w ⋅ ( A b − i ∇ x b ) w\cdot(A b-i\nabla_xb) w ⋅ ( A b − i ∇ x b ) , this means ∑ j w j ( A j b − i ∂ j b ) \sum_j w_j(A_jb-i\partial_jb) ∑ j w j ( A j b − i ∂ j b ) . Extend and cut the coefficients off in Ω 0 \Omega_0 Ω 0 , preserving them near the spatial cutoffs used below. Their supports may extend beyond the source ball.
First fix nested domains, a source ball, and a source cutoff:
Ω ‾ ⋐ Ω C G O ⋐ int K s r c , K s r c ⋐ Ω 1 ⋐ Ω 0 , (112) \overline{\Omega}\Subset\Omega_{\mathrm{CGO}}\Subset\operatorname{int}K_{\mathrm{src}},\qquad K_{\mathrm{src}}\Subset\Omega_1\Subset\Omega_0, \tag*{(112)} Ω ⋐ Ω CGO ⋐ int K src , K src ⋐ Ω 1 ⋐ Ω 0 , (112)
χ 0 ∈ C c ∞ ( int K s r c ) , χ 0 = 1 near Ω ‾ C G O . \chi_0\in C^\infty_c(\operatorname{int}K_{\mathrm{src}}),\qquad\chi_0=1\quad\text{near }\overline{\Omega}_{\mathrm{CGO}}. χ 0 ∈ C c ∞ ( int K src ) , χ 0 = 1 near Ω CGO .
The source set K s r c K_{\mathrm{src}} K src may be chosen as a closed ball containing Ω ‾ C G O \overline{\Omega}_{\mathrm{CGO}} Ω CGO in its interior. The inclusion Ω ‾ C G O ⋐ int K s r c \overline{\Omega}_{\mathrm{CGO}}\Subset\operatorname{int}K_{\mathrm{src}} Ω CGO ⋐ int K src ensures that the required source cutoff exists; all geometric choices precede the right-inverse construction. The domains may be chosen as concentric balls. Fix ψ 2 , ψ 3 , χ s p ∈ C c ∞ ( Ω 0 ) \psi_2,\psi_3,\chi_{\mathrm{sp}}\in C^\infty_c(\Omega_0) ψ 2 , ψ 3 , χ sp ∈ C c ∞ ( Ω 0 ) equal to one on open neighborhoods of Ω ‾ 1 \overline{\Omega}_1 Ω 1 , supp ψ 2 \operatorname{supp}\psi_2 supp ψ 2 , and supp ψ 3 \operatorname{supp}\psi_3 supp ψ 3 , respectively. All choices precede τ \tau τ and the source.
All weighted spaces in this section are on R 3 \mathbb{R}^3 R 3 with values in C m \mathbb{C}^m C m ; pairings follow § 1.6 . Put ⟨ x ⟩ = ( 1 + ∣ x ∣ 2 ) 1 / 2 \langle x\rangle=(1+|x|^2)^{1/2} ⟨ x ⟩ = ( 1 + ∣ x ∣ 2 ) 1/2 and, for s ∈ R s\in\mathbb{R} s ∈ R and j = 0 , 1 j=0,1 j = 0 , 1 , define
∥ u ∥ H s j 2 = ∑ ∣ α ∣ ≤ j ∥ ⟨ x ⟩ s ∂ α u ∥ L 2 2 , L s 2 = H s 0 . \|u\|_{H^j_s}^2=\sum_{|\alpha|\leq j}\|\langle x\rangle^s\partial^\alpha u\|_{L^2}^2,\qquad L^2_s=H^0_s. ∥ u ∥ H s j 2 = ∣ α ∣ ≤ j ∑ ∥ ⟨ x ⟩ s ∂ α u ∥ L 2 2 , L s 2 = H s 0 .
The weight exponent s s s is distinct from the boundary-layer phase sign s \mathfrak{s} s . The free inverse and transport frames below are supplied by Theorem B.1 and Proposition 4.10 , respectively.
The free inverse E.: Let E δ E_\delta E δ be the selected free inverse from Theorem B.1 . Its defining identity and weighted estimate are (B.2) and (B.1); both hold uniformly on the compact direction families under consideration.
Transport frames F.: Choose R ∗ > sup x ∈ Ω 0 ∣ x ∣ R_*> \sup_{x\in\Omega_0}|x| R ∗ > sup x ∈ Ω 0 ∣ x ∣ and R ∗ < L < M R_*<L<M R ∗ < L < M . Near each parameter center for ordered orthonormal pairs ( v ^ , v 1 ) (\widehat v,v_1) ( v , v 1 ) and ∣ t ∣ ≤ R ∗ |t|\leq R_* ∣ t ∣ ≤ R ∗ , there is a smooth invertible frame C a C_a C a , defined near the fixed disk D L D_L D L , such that
i ( v ^ + i v 1 ) ⋅ ∇ x C a = A v ^ + i v 1 C a , x = z 1 v ^ + z 2 v 1 + t ( v ^ × v 1 ) , z = z 1 + i z 2 . (113) i(\widehat v+iv_1)\cdot\nabla_x C_a=A_{\widehat v+iv_1}C_a,\qquad x=z_1\widehat v+z_2v_1+t(\widehat v\times v_1),\qquad z=z_1+iz_2. \tag*{(113)} i ( v + i v 1 ) ⋅ ∇ x C a = A v + i v 1 C a , x = z 1 v + z 2 v 1 + t ( v × v 1 ) , z = z 1 + i z 2 . (113)
Each frame and its inverse are jointly smooth in the local real parameters and satisfy the finite mixed bounds below.
Cover the compact parameter set by finitely many charts, with weights supported strictly inside them. Let η \eta η be local real coordinates for the ordered orthonormal pair. The following bounds suffice:
max a max ∣ α ∣ ≤ 10 , ∣ β ∣ ≤ 2 sup ( ∥ ∂ η α ∂ ( z 1 , z 2 , t ) β C a ∥ + ∥ ∂ η α ∂ ( z 1 , z 2 , t ) β C a − 1 ∥ ) < ∞ (114) \max_a\max_{|\alpha|\leq10,\ |\beta|\leq2}\sup\left(\|\partial_\eta^\alpha\partial_{(z_1,z_2,t)}^\beta C_a\|+\|\partial_\eta^\alpha\partial_{(z_1,z_2,t)}^\beta C_a^{-1}\|\right)<\infty \tag*{(114)} a max ∣ α ∣ ≤ 10 , ∣ β ∣ ≤ 2 max sup ( ∥ ∂ η α ∂ ( z 1 , z 2 , t ) β C a ∥ + ∥ ∂ η α ∂ ( z 1 , z 2 , t ) β C a − 1 ∥ ) < ∞ (114)
on the chosen closed subcharts and ∣ z ∣ ≤ L |z|\leq L ∣ z ∣ ≤ L , together with the corresponding ∂ η ≤ 10 ∂ η ≤ 2 \partial_\eta^{\leq10}\partial_\eta^{\leq2} ∂ η ≤ 10 ∂ η ≤ 2 bounds on the weights, six spatial derivatives of the coefficients, eight derivatives of the spatial cutoffs, and two derivatives of the frequency cutoffs. Compactness controls coordinate changes. The two extra spatial derivatives accommodate frequency differentiation of the moving leaf coordinates.
Proposition 5.1. Fix the preceding data and 0 < γ < 1 0<\gamma<1 0 < γ < 1 . For each sufficiently large τ \tau τ and each chosen direction, there is a linear operator R δ R_\delta R δ , selected independently of the source, such that every f ∈ L 2 ( Ω 0 ; C m ) f\in L^2(\Omega_0;\mathbb{C}^m) f ∈ L 2 ( Ω 0 ; C m ) with supp f ⊂ K s r c \operatorname{supp}f\subset K_{\mathrm{src}} supp f ⊂ K src satisfies
P δ R δ f = f in Ω 1 , τ ∥ R δ f ∥ L 2 ( Ω 0 ) + ∥ R δ f ∥ H 1 ( Ω 0 ) ≤ C ∥ f ∥ 2 . (115) P_\delta R_\delta f=f\quad\text{in }\Omega_1,\qquad\tau\|R_\delta f\|_{L^2(\Omega_0)}+\|R_\delta f\|_{H^1(\Omega_0)}\leq C\|f\|_2. \tag*{(115)} P δ R δ f = f in Ω 1 , τ ∥ R δ f ∥ L 2 ( Ω 0 ) + ∥ R δ f ∥ H 1 ( Ω 0 ) ≤ C ∥ f ∥ 2 . (115) More precisely, the threshold τ 0 \tau_0 τ 0 and constant C C C may be chosen from γ \gamma γ , the matrix size, the fixed nested domains and spatial cutoffs in (112) , the free-inverse constant in (B.1), the coefficient and cutoff seminorms specified above, the bounds for the frames and their inverses in (114) , and the finitely many chart weights. Once these data are fixed, τ 0 \tau_0 τ 0 and C C C are independent of the source f f f , τ ≥ τ 0 \tau\geq\tau_0 τ ≥ τ 0 , and the direction in the selected compact family.
Fourier kernel estimates and weighted localization
For Fourier analysis we use u ^ ( ξ ) = ∫ R 3 e − i x ⋅ ξ u ( x ) d x \widehat u(\xi)=\int_{\mathbb{R}^3}e^{-ix\cdot\xi}u(x)\,dx u ( ξ ) = ∫ R 3 e − i x ⋅ ξ u ( x ) d x , with inverse-transform factor ( 2 π ) − 3 (2\pi)^{-3} ( 2 π ) − 3 . For a symbol, a ^ ( v , ξ ) \widehat a(v,\xi) a ( v , ξ ) denotes Fourier transformation in x x x alone. Left quantization and the finite symbol seminorms are
Op ( a ) u ( x ) = ( 2 π ) − 3 ∫ e i x ⋅ ξ a ( x , ξ ) u ^ ( ξ ) d ξ , s N , r ( a ) = max ∣ α ∣ ≤ N , ∣ β ∣ ≤ r ∥ ∂ x α ∂ ξ β a ∥ ∞ . \operatorname{Op}(a)u(x)=(2\pi)^{-3}\int e^{ix\cdot\xi}a(x,\xi)\widehat u(\xi)\,d\xi,\qquad s_{N,r}(a)=\max_{|\alpha|\leq N,\ |\beta|\leq r}\|\partial_x^\alpha\partial_\xi^\beta a\|_\infty. Op ( a ) u ( x ) = ( 2 π ) − 3 ∫ e i x ⋅ ξ a ( x , ξ ) u ( ξ ) d ξ , s N , r ( a ) = ∣ α ∣ ≤ N , ∣ β ∣ ≤ r max ∥ ∂ x α ∂ ξ β a ∥ ∞ .
Lemma B.2 gives the unweighted bounds needed here. For symbols with fixed compact spatial support, its proof by integration by parts with ( 1 − Δ x ) 3 (1-\Delta_x)^3 ( 1 − Δ x ) 3 gives ∣ a ^ ( η − ξ , ξ ) ∣ ≤ C s 6 , 0 ( a ) ⟨ η − ξ ⟩ − 6 |\widehat a(\eta-\xi,\xi)|\leq Cs_{6,0}(a)\langle\eta-\xi\rangle^{-6} ∣ a ( η − ξ , ξ ) ∣ ≤ C s 6 , 0 ( a ) ⟨ η − ξ ⟩ − 6 . Schur’s test and ⟨ η ⟩ / ⟨ ξ ⟩ ≤ C ⟨ η − ξ ⟩ \langle\eta\rangle/\langle\xi\rangle\leq C\langle\eta-\xi\rangle ⟨ η ⟩ / ⟨ ξ ⟩ ≤ C ⟨ η − ξ ⟩ imply Op ( a ) : H j → H j \operatorname{Op}(a):H^j\to H^j Op ( a ) : H j → H j , j = 0 , 1 j=0,1 j = 0 , 1 , with norm controlled by s 6 , 0 ( a ) s_{6,0}(a) s 6 , 0 ( a ) .
For compact spatial symbols a , c a,c a , c , Proposition B.4 gives the composition-defect estimate (B.12); the exact kernel and its matrix order are recorded in (B.13). Thus if s 6 , 0 ( ∂ ξ k a ) = O ( ϵ ) s_{6,0}(\partial_{\xi_k}a)=O(\epsilon) s 6 , 0 ( ∂ ξ k a ) = O ( ϵ ) for k = 1 , 2 , 3 k = 1, 2, 3 k = 1 , 2 , 3 and s 6 , 0 ( c ) = O ( 1 ) s_{6,0}(c) = O(1) s 6 , 0 ( c ) = O ( 1 ) , the defect is O ( ϵ ) O(\epsilon) O ( ϵ ) on H j H^{j} H j , j = 0 , 1 j = 0, 1 j = 0 , 1 . The pure Fourier-multiplier case is included in the same proposition.
The weighted localization estimate follows directly from Proposition B.3 : the exact identity is (B.11), and the resulting bound is (B.10). The proposition’s extension applies to the noncompact outputs of E δ E_{\delta} E δ as well.
Frequency cutoffs and paired symbols
Fix γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) . Choose real smooth weights w a ( v ^ , v 1 , t ) w_a(\widehat{\mathbf v},\mathbf v_1,t) w a ( v , v 1 , t ) supported inside the frame charts, with ∑ a w a 2 = 1 \sum_a w_a^2 = 1 ∑ a w a 2 = 1 on the compact parameter set. Since their only spatial variable is t t t , ( v ^ + i v 1 ) ⋅ ∇ x w a = 0 (\widehat{\mathbf v}+i\mathbf v_1)\cdot\nabla_xw_a=0 ( v + i v 1 ) ⋅ ∇ x w a = 0 . Each frame is paired with its own inverse.
Write
ξ v 1 = ξ ⋅ v 1 , ξ ′ = ξ − ξ v 1 v 1 , r = ∣ ξ ′ + τ v 0 ∣ , v ^ = ( ξ ′ + τ v 0 ) / r . \xi_{\mathbf v_1}=\xi\cdot\mathbf v_1,\qquad\xi'=\xi-\xi_{\mathbf v_1}\mathbf v_1,\qquad r=\lvert\xi'+\tau\mathbf v_0\rvert,\qquad\widehat{\mathbf v}=(\xi'+\tau\mathbf v_0)/r. ξ v 1 = ξ ⋅ v 1 , ξ ′ = ξ − ξ v 1 v 1 , r = ∣ ξ ′ + τ v 0 ∣ , v = ( ξ ′ + τ v 0 ) / r .
The real unit vector v ^ \widehat{\mathbf v} v is the moving direction used to select the near-frequency transport frame. Choose a fixed ϱ ∈ C c ∞ ( R ) \varrho\in C^\infty_c(\mathbb R) ϱ ∈ C c ∞ ( R ) with 0 ≤ ϱ ≤ 1 0\leq\varrho\leq1 0 ≤ ϱ ≤ 1 , equal to one on [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] and zero outside [ − 2 , 2 ] [-2,2] [ − 2 , 2 ] , and define
t τ , γ = ϱ ( ξ v 1 / τ γ ) ϱ ( ( r 2 − τ 2 ) / τ 1 + γ ) , χ N = sin ( π t τ , γ / 2 ) , χ F = cos ( π t τ , γ / 2 ) . (116) t_{\tau,\gamma}=\varrho(\xi_{\mathbf v_1}/\tau^\gamma)\varrho((r^2-\tau^2)/\tau^{1+\gamma}),\qquad\chi_N=\sin(\pi t_{\tau,\gamma}/2),\qquad\chi_F=\cos(\pi t_{\tau,\gamma}/2). \tag*{(116)} t τ , γ = ϱ ( ξ v 1 / τ γ ) ϱ (( r 2 − τ 2 ) / τ 1 + γ ) , χ N = sin ( π t τ , γ /2 ) , χ F = cos ( π t τ , γ /2 ) . (116)
For sufficiently large τ \tau τ , on supp χ N \operatorname{supp}\chi_N supp χ N ,
r ≍ τ , ∣ r − τ ∣ + ∣ ξ v 1 ∣ ≤ C τ γ , ∣ ξ ∣ ≤ 4 τ . (117) r\asymp\tau,\qquad\lvert r-\tau\rvert+\lvert\xi_{\mathbf v_1}\rvert\leq C\tau^\gamma,\qquad\lvert\xi\rvert\leq4\tau. \tag*{(117)} r ≍ τ , ∣ r − τ ∣ + ∣ ξ v 1 ∣ ≤ C τ γ , ∣ ξ ∣ ≤ 4 τ . (117)
The radial bound follows by division by r + τ ≍ τ r+\tau\asymp\tau r + τ ≍ τ . The near cutoff vanishes near r = 0 r=0 r = 0 , so the moving direction is smooth on its support. Cutoff derivatives of order k = 1 , 2 k=1,2 k = 1 , 2 are O ( τ − k γ ) O(\tau^{-k\gamma}) O ( τ − k γ ) . To verify the angular derivatives entering the frame symbols, set Π = I 3 − v 1 v 1 T \Pi=I_3-\mathbf v_1\mathbf v_1^T Π = I 3 − v 1 v 1 T and v ( ξ ) = Π ξ + τ v 0 = r v ^ v(\xi)=\Pi\xi+\tau\mathbf v_0=r\widehat{\mathbf v} v ( ξ ) = Π ξ + τ v 0 = r v . For r > 0 r>0 r > 0 direct differentiation gives
∂ ξ j r = v ^ ⋅ Π e j , ∂ ξ j v ^ = r − 1 ( I 3 − v ^ v ^ T ) Π e j . (118) \partial_{\xi_j}r=\widehat{\mathbf v}\cdot\Pi e_j,\qquad\partial_{\xi_j}\widehat{\mathbf v}=r^{-1}(I_3-\widehat{\mathbf v}\widehat{\mathbf v}^{T})\Pi e_j. \tag*{(118)} ∂ ξ j r = v ⋅ Π e j , ∂ ξ j v = r − 1 ( I 3 − v v T ) Π e j . (118)
Since r ≍ τ r\asymp\tau r ≍ τ on the near support, repeated differentiation of (118) gives ∂ ξ k v ^ = O ( τ − k ) \partial_\xi^k\widehat{\mathbf v}=O(\tau^{-k}) ∂ ξ k v = O ( τ − k ) for k = 1 , 2 k=1,2 k = 1 , 2 . The corresponding changes in the leaf coordinates are also O ( τ − k ) O(\tau^{-k}) O ( τ − k ) on the fixed compact spatial set. For the cutoff calculation, with s = ( r 2 − τ 2 ) / τ 1 + γ s=(r^2-\tau^2)/\tau^{1+\gamma} s = ( r 2 − τ 2 ) / τ 1 + γ , the full second radial derivative is
∂ r 2 ϱ ( s ) = 4 r 2 τ − 2 − 2 γ ϱ ′ ′ ( s ) + 2 τ − 1 − γ ϱ ′ ( s ) = O ( τ − 2 γ ) \partial_r^2\varrho(s)=4r^2\tau^{-2-2\gamma}\varrho''(s)+2\tau^{-1-\gamma}\varrho'(s)=O(\tau^{-2\gamma}) ∂ r 2 ϱ ( s ) = 4 r 2 τ − 2 − 2 γ ϱ ′′ ( s ) + 2 τ − 1 − γ ϱ ′ ( s ) = O ( τ − 2 γ )
on the transition region r ≍ τ r\asymp\tau r ≍ τ . The same bound holds for the sine and cosine cutoff profiles by the chain rule. Direction and moving-coordinate derivatives of order k k k are O ( τ − k ) O(\tau^{-k}) O ( τ − k ) .
Let the positive indices a a a run over the finitely many chosen frame charts, and reserve a = 0 a=0 a = 0 for the far-frequency term. For a ≥ 1 a\geq1 a ≥ 1 , set
b a = χ s p χ N w a C a , d a = χ s p χ N w a C a − 1 , b 0 = d 0 = χ F I m . (119) b_a=\chi_{\mathrm{sp}}\chi_Nw_aC_a,\qquad d_a=\chi_{\mathrm{sp}}\chi_Nw_aC_a^{-1},\qquad b_0=d_0=\chi_FI_m. \tag*{(119)} b a = χ sp χ N w a C a , d a = χ sp χ N w a C a − 1 , b 0 = d 0 = χ F I m . (119)
Extend the near symbols by zero outside their parameter charts; the strictly interior supports of the weights ensure smoothness. The chain rule and (114) give
∥ ∂ x α ∂ ξ β b a ∥ ∞ + ∥ ∂ x α ∂ ξ β d a ∥ ∞ ≤ C τ − γ ∣ β ∣ , ∣ α ∣ ≤ 8 , ∣ β ∣ ≤ 2. (120) \lVert\partial_x^\alpha\partial_\xi^\beta b_a\rVert_\infty+\lVert\partial_x^\alpha\partial_\xi^\beta d_a\rVert_\infty\leq C\tau^{-\gamma\lvert\beta\rvert},\qquad\lvert\alpha\rvert\leq8,\quad\lvert\beta\rvert\leq2. \tag*{(120)} ∥ ∂ x α ∂ ξ β b a ∥ ∞ + ∥ ∂ x α ∂ ξ β d a ∥ ∞ ≤ C τ − γ ∣ β ∣ , ∣ α ∣ ≤ 8 , ∣ β ∣ ≤ 2. (120)
The index-zero symbols are pure multipliers. On supp ψ 3 \operatorname{supp}\psi_3 supp ψ 3 we have the exact identities
∑ a ≥ 0 b a d a = I m , ( v ^ + i v 1 ) ⋅ ( A b a − i ∇ x b a ) = 0 ( a ≥ 1 ) . (121) \sum_{a\geq0}b_ad_a=I_m,\qquad(\widehat{\mathbf v}+i\mathbf v_1)\cdot(Ab_a-i\nabla_xb_a)=0\quad(a\geq1). \tag*{(121)} a ≥ 0 ∑ b a d a = I m , ( v + i v 1 ) ⋅ ( A b a − i ∇ x b a ) = 0 ( a ≥ 1 ) . (121)
The second identity uses χ s p = 1 \chi_{\mathrm{sp}}=1 χ sp = 1 on a neighborhood of this support. For any g ∈ L 2 ( R 3 ; C m ) g\in L^2(\mathbb R^3;\mathbb C^m) g ∈ L 2 ( R 3 ; C m ) , define
H a g = ψ 3 Op ( d a ) g , Q τ , γ = ∑ a ≥ 0 Op ( b a ) E δ H a g . (122) H_ag=\psi_3\operatorname{Op}(d_a)g,\qquad Q_{\tau,\gamma}=\sum_{a\geq0}\operatorname{Op}(b_a)E_\delta H_ag. \tag*{(122)} H a g = ψ 3 Op ( d a ) g , Q τ , γ = a ≥ 0 ∑ Op ( b a ) E δ H a g . (122)
The operators H a : L 2 → L 1 2 H_a:L^2\to L^2_1 H a : L 2 → L 1 2 are uniformly bounded and have output supported in supp ψ 3 \operatorname{supp}\psi_3 supp ψ 3 . Indeed, the ordinary Fourier–Schur estimate and sup supp ψ 3 ⟨ x ⟩ < ∞ \sup_{\operatorname{supp}\psi_3}\langle x\rangle<\infty sup supp ψ 3 ⟨ x ⟩ < ∞ give, for every near index,
∥ H a g ∥ L 1 2 ≤ sup supp ψ 3 ⟨ x ⟩ ∥ ψ 3 ∥ ∞ C s 6 , 0 ( d a ) ∥ g ∥ 2 ≤ C s r c ∥ g ∥ 2 . (123) \|H_ag\|_{L^2_1}\le\sup_{\operatorname{supp}\psi_3}\langle x\rangle\|\psi_3\|_\infty C s_{6,0}(d_a)\|g\|_2\le C_{\mathrm{src}}\|g\|_2. \tag*{(123)} ∥ H a g ∥ L 1 2 ≤ supp ψ 3 sup ⟨ x ⟩ ∥ ψ 3 ∥ ∞ C s 6 , 0 ( d a ) ∥ g ∥ 2 ≤ C src ∥ g ∥ 2 . (123)
For a = 0 a=0 a = 0 , the identical conclusion follows from Plancherel for the bounded multiplier d 0 = χ F I m d_0=\chi_F I_m d 0 = χ F I m , followed by multiplication by ψ 3 \psi_3 ψ 3 . These bounds hold for arbitrary L 2 L^2 L 2 sources g g g ; the spatial cutoff is applied to the output of Op ( d a ) \operatorname{Op}(d_a) Op ( d a ) . Proposition B.4 and its multiplier case give
S = ∑ a Op ( b a ) ψ 3 Op ( d a ) − ψ 3 I m , ∥ S ∥ L 2 → L 2 ≤ C τ − γ . (124) S=\sum_a\operatorname{Op}(b_a)\psi_3\operatorname{Op}(d_a)-\psi_3I_m,\qquad\|S\|_{L^2\to L^2}\le C\tau^{-\gamma}. \tag*{(124)} S = a ∑ Op ( b a ) ψ 3 Op ( d a ) − ψ 3 I m , ∥ S ∥ L 2 → L 2 ≤ C τ − γ . (124)
Here the right symbol is the compactly supported ψ 3 d a \psi_3d_a ψ 3 d a , and ∑ a b a ψ 3 d a = ψ 3 I m \sum_a b_a\psi_3d_a=\psi_3I_m ∑ a b a ψ 3 d a = ψ 3 I m .
The differential composition in the parametrix follows from the exact left-quantization identities
( D j + δ j ) Op ( b ) = Op ( ( ξ j + δ j ) b − i ∂ j b ) , (D_j+\delta_j)\operatorname{Op}(b)=\operatorname{Op}\bigl((\xi_j+\delta_j)b-i\partial_jb\bigr), ( D j + δ j ) Op ( b ) = Op ( ( ξ j + δ j ) b − i ∂ j b ) ,
( D + δ ) 2 Op ( b ) = Op ( p δ b − 2 i ( ξ + δ ) ⋅ ∇ x b − Δ x b ) . (125) (D+\delta)^2\operatorname{Op}(b)=\operatorname{Op}\bigl(p_\delta b-2i(\xi+\delta)\cdot\nabla_xb-\Delta_xb\bigr). \tag*{(125)} ( D + δ ) 2 Op ( b ) = Op ( p δ b − 2 i ( ξ + δ ) ⋅ ∇ x b − Δ x b ) . (125)
Here the differential operators act on the output variable x x x ; the symbols b b b are matrix-valued and are never commuted across the matrices A j A_j A j . In particular, multiplying the first identity by 2 A j ( x ) 2A_j(x) 2 A j ( x ) on the left and then adding q ( x ) Op ( b ) q(x)\operatorname{Op}(b) q ( x ) Op ( b ) gives, preserving matrix order,
P δ Op ( b a ) E δ = Op ( b a ) + Op ( r 0 , a ) E δ + Op ( r 1 , a ) E δ , P_\delta\operatorname{Op}(b_a)E_\delta=\operatorname{Op}(b_a)+\operatorname{Op}(r_{0,a})E_\delta+\operatorname{Op}(r_{1,a})E_\delta, P δ Op ( b a ) E δ = Op ( b a ) + Op ( r 0 , a ) E δ + Op ( r 1 , a ) E δ ,
r 0 , a = − Δ x b a − 2 i ∑ j A j ∂ j b a + q b a , r_{0,a}=-\Delta_xb_a-2i\sum_jA_j\partial_jb_a+qb_a, r 0 , a = − Δ x b a − 2 i j ∑ A j ∂ j b a + q b a ,
r 1 , a = 2 ∑ j ( ξ j + δ j ) ( A j b a − i ∂ j b a ) . (126) r_{1,a}=2\sum_j(\xi_j+\delta_j)(A_jb_a-i\partial_jb_a). \tag*{(126)} r 1 , a = 2 j ∑ ( ξ j + δ j ) ( A j b a − i ∂ j b a ) . (126)
Define the differential residual operator by
E e r r = ∑ a ≥ 0 [ Op ( r 0 , a ) E δ H a + Op ( r 1 , a ) E δ H a ] . E_{\mathrm{err}}=\sum_{a\ge0}\left[\operatorname{Op}(r_{0,a})E_\delta H_a+\operatorname{Op}(r_{1,a})E_\delta H_a\right]. E err = a ≥ 0 ∑ [ Op ( r 0 , a ) E δ H a + Op ( r 1 , a ) E δ H a ] .
Thus S S S in (124) is the composition defect and E e r r E_{\mathrm{err}} E err is the residual from differentiating the paired symbols. Substituting (126) into (122) gives
P δ Q τ , γ = ψ 3 I m + S + E e r r . (127) P_\delta Q_{\tau,\gamma}=\psi_3I_m+S+E_{\mathrm{err}}. \tag*{(127)} P δ Q τ , γ = ψ 3 I m + S + E err . (127)
For fixed τ \tau τ , the identity first holds on smooth sources and extends by the local estimates below and continuity in distributions.
Near and far residual estimates
The seminorms s 6 , 2 ( ψ 3 r 0 , a ) s_{6,2}(\psi_3r_{0,a}) s 6 , 2 ( ψ 3 r 0 , a ) are uniformly bounded; six derivatives of − Δ b a -\Delta b_a − Δ b a require eight spatial derivatives of b a b_a b a . Applying (B.1) followed by (B.10) gives
∥ Op ( ψ 3 r 0 , a ) E δ H a ∥ L 2 → L 2 ≤ C τ − 1 . (128) \|\operatorname{Op}(\psi_3r_{0,a})E_\delta H_a\|_{L^2\to L^2}\le C\tau^{-1}. \tag*{(128)} ∥ Op ( ψ 3 r 0 , a ) E δ H a ∥ L 2 → L 2 ≤ C τ − 1 . (128)
In the near region, the second identity in (121) cancels τ ( v ^ + i v 1 ) \tau(\widehat v+iv_1) τ ( v + i v 1 ) , leaving exactly
ξ + δ − τ ( v ^ + i v 1 ) = ( r − τ ) v ^ + ξ v 1 v 1 . (129) \xi+\delta-\tau(\widehat v+iv_1)=(r-\tau)\widehat v+\xi_{v_1}v_1. \tag*{(129)} ξ + δ − τ ( v + i v 1 ) = ( r − τ ) v + ξ v 1 v 1 . (129)
On the support of the near symbols, put R ( ξ ) = ( r − τ ) v ^ + ξ v 1 v 1 R(\xi)=(r-\tau)\widehat v+\xi_{v_1}v_1 R ( ξ ) = ( r − τ ) v + ξ v 1 v 1 . The estimates (117) , r ≍ τ r\asymp\tau r ≍ τ , and ∂ ξ v ^ = O ( τ − 1 ) \partial_\xi\widehat v=O(\tau^{-1}) ∂ ξ v = O ( τ − 1 ) give
∣ R ∣ ≤ C τ γ , ∣ ∂ ξ R ∣ ≤ C , ∣ ∂ ξ 2 R ∣ ≤ C τ − 1 . |R|\le C\tau^\gamma,\qquad|\partial_\xi R|\le C,\qquad|\partial_\xi^2R|\le C\tau^{-1}. ∣ R ∣ ≤ C τ γ , ∣ ∂ ξ R ∣ ≤ C , ∣ ∂ ξ 2 R ∣ ≤ C τ − 1 .
Here ∣ ∂ ξ q R ∣ |\partial_\xi^qR| ∣ ∂ ξ q R ∣ denotes the maximum of the component derivatives of exact order q q q . On supp ψ 3 \operatorname{supp}\psi_3 supp ψ 3 , where the transport cancellation is exact, the near residual is
r 1 , a = 2 R ( ξ ) ⋅ ( A b a − i ∇ x b a ) , a ≥ 1. r_{1,a}=2R(\xi)\cdot(A b_a-i\nabla_xb_a),\qquad a\ge1. r 1 , a = 2 R ( ξ ) ⋅ ( A b a − i ∇ x b a ) , a ≥ 1.
Every frequency derivative of order at most two falls on R R R or on a near symbol. Applying (120) and the product rule, for ∣ α ∣ ≤ 6 |\alpha| \le6 ∣ α ∣ ≤ 6 , yields
∥ ∂ x α ∂ ξ β ( ψ 3 r 1 , a ) ∥ ∞ ≤ C τ γ − γ ∣ β ∣ , ∣ α ∣ ≤ 6 , ∣ β ∣ ≤ 2 , a ≥ 1. \lVert\partial_x^\alpha\partial_\xi^\beta(\psi_3 r_{1,a}) \rVert_\infty\le C\tau^{\gamma-\gamma|\beta|}, \qquad|\alpha| \le6,\quad|\beta| \le2,\quad a \ge1. ∥ ∂ x α ∂ ξ β ( ψ 3 r 1 , a ) ∥ ∞ ≤ C τ γ − γ ∣ β ∣ , ∣ α ∣ ≤ 6 , ∣ β ∣ ≤ 2 , a ≥ 1.
Hence s 6 , 2 ( τ − γ ψ 3 r 1 , a ) = O ( 1 ) s_{6,2}(\tau^{-\gamma}\psi_3r_{1,a})=O(1) s 6 , 2 ( τ − γ ψ 3 r 1 , a ) = O ( 1 ) . The free estimate followed by weighted localization yields
∥ Op ( ψ 3 r 1 , a ) E δ H a ∥ L 2 → L 2 ≤ C τ γ − 1 , a ≥ 1. (130) \lVert\operatorname{Op}(\psi_3r_{1,a})E_\delta H_a\rVert_{L^2\to L^2} \le C\tau^{\gamma-1}, \qquad a \ge1. \tag*{(130)} ∥ Op ( ψ 3 r 1 , a ) E δ H a ∥ L 2 → L 2 ≤ C τ γ − 1 , a ≥ 1. (130)
This estimate differentiates only the smooth symbols.
For the far region, the identity
p δ = ξ v 1 2 + r 2 − τ 2 + 2 i τ ξ v 1 p_\delta=\xi_{v_1}^2+r^2-\tau^2+2i\tau\xi_{v_1} p δ = ξ v 1 2 + r 2 − τ 2 + 2 i τ ξ v 1
gives, on supp χ F \operatorname{supp}\chi_F supp χ F ,
∣ p δ ( ξ ) ∣ ≥ { 1 2 τ 1 + γ , ∣ ξ ∣ ≤ 4 τ , ( 1 − 1 / 2 ) ∣ ξ ∣ 2 , ∣ ξ ∣ ≥ 4 τ . (131) |p_\delta(\xi)| \ge
\begin{cases}
\frac{1}{2}\tau^{1+\gamma}, & |\xi|\le4\tau,\\
(1-1/\sqrt{2})|\xi|^2, & |\xi|\ge4\tau.
\end{cases} \tag*{(131)} ∣ p δ ( ξ ) ∣ ≥ { 2 1 τ 1 + γ , ( 1 − 1/ 2 ) ∣ ξ ∣ 2 , ∣ ξ ∣ ≤ 4 τ , ∣ ξ ∣ ≥ 4 τ . (131)
On the far support, either ∣ ξ v 1 ∣ ≥ τ γ |\xi_{v_1}|\ge\tau^\gamma ∣ ξ v 1 ∣ ≥ τ γ or ∣ r 2 − τ 2 ∣ ≥ τ 1 + γ |r^2-\tau^2|\ge\tau^{1+\gamma} ∣ r 2 − τ 2 ∣ ≥ τ 1 + γ . The first case follows from the imaginary part. Otherwise ξ v 1 2 ≤ τ 2 γ \xi_{v_1}^2\le\tau^{2\gamma} ξ v 1 2 ≤ τ 2 γ , so the real part has modulus at least τ 1 + γ − τ 2 γ ≥ 1 2 τ 1 + γ \tau^{1+\gamma}-\tau^{2\gamma}\ge\frac{1}{2}\tau^{1+\gamma} τ 1 + γ − τ 2 γ ≥ 2 1 τ 1 + γ . For high frequencies use ∣ 2 δ ⋅ ξ ∣ ≤ 2 2 τ ∣ ξ ∣ |2\delta\cdot\xi|\le2\sqrt{2}\tau|\xi| ∣2 δ ⋅ ξ ∣ ≤ 2 2 τ ∣ ξ ∣ .
Set
a F ( x , ξ ) = 2 ∑ j A j ( x ) ( ξ j + δ j ) χ F ( ξ ) p δ ( ξ ) . (132) a_F(x,\xi)=\frac{2\sum_j A_j(x)(\xi_j+\delta_j)\chi_F(\xi)}{p_\delta(\xi)}. \tag*{(132)} a F ( x , ξ ) = p δ ( ξ ) 2 ∑ j A j ( x ) ( ξ j + δ j ) χ F ( ξ ) . (132)
The far cutoff vanishes near the characteristic set, so the symbol extends smoothly by zero there. For ∣ ξ ∣ ≤ 4 τ |\xi|\le4\tau ∣ ξ ∣ ≤ 4 τ , direct differentiation yields
∂ ξ j p δ = 2 ( ξ j + δ j ) , ∂ ξ j ∂ ξ k p δ = 2 δ j k . \partial_{\xi_j}p_\delta=2(\xi_j+\delta_j), \qquad\partial_{\xi_j}\partial_{\xi_k}p_\delta=2\delta_{jk}. ∂ ξ j p δ = 2 ( ξ j + δ j ) , ∂ ξ j ∂ ξ k p δ = 2 δ j k .
On the far support in this bounded-frequency region, (131) and ∣ ∂ ξ p δ ∣ ≤ C τ |\partial_\xi p_\delta|\le C\tau ∣ ∂ ξ p δ ∣ ≤ C τ give, for ∣ β ∣ ≤ 2 |\beta|\le2 ∣ β ∣ ≤ 2 ,
∣ ∂ ξ β ( p δ − 1 ) ∣ ≤ C τ − 1 − γ − γ ∣ β ∣ . \left|\partial_\xi^\beta(p_\delta^{-1})\right|\le C\tau^{-1-\gamma-\gamma|\beta|}. ∂ ξ β ( p δ − 1 ) ≤ C τ − 1 − γ − γ ∣ β ∣ .
Indeed the first derivative is − ( ∂ ξ p δ ) p δ − 2 -(\partial_\xi p_\delta)p_\delta^{-2} − ( ∂ ξ p δ ) p δ − 2 , while the second is a sum of ( ∂ ξ p δ ) 2 p δ − 3 (\partial_\xi p_\delta)^2p_\delta^{-3} ( ∂ ξ p δ ) 2 p δ − 3 and ( ∂ ξ 2 p δ ) p δ − 2 (\partial_\xi^2p_\delta)p_\delta^{-2} ( ∂ ξ 2 p δ ) p δ − 2 ; the latter is bounded by the displayed power because 0 < γ < 1 0<\gamma<1 0 < γ < 1 . The numerator and its first two frequency derivatives, including the cutoff, have bounds C τ 1 − γ ∣ β ∣ C\tau^{1-\gamma|\beta|} C τ 1 − γ ∣ β ∣ . Consequently, the product rule gives
∥ ∂ x α ∂ ξ β ( ψ 3 a F ) ∥ ∞ ≤ C τ − γ ( 1 + ∣ β ∣ ) , ∣ α ∣ ≤ 6 , ∣ β ∣ ≤ 2. (133) \lVert\partial_x^\alpha\partial_\xi^\beta(\psi_3a_F)\rVert_\infty\le C\tau^{-\gamma(1+|\beta|)}, \qquad|\alpha|\le6,\quad|\beta|\le2. \tag*{(133)} ∥ ∂ x α ∂ ξ β ( ψ 3 a F ) ∥ ∞ ≤ C τ − γ ( 1 + ∣ β ∣ ) , ∣ α ∣ ≤ 6 , ∣ β ∣ ≤ 2. (133)
For ∣ ξ ∣ ≥ 4 τ |\xi|\ge4\tau ∣ ξ ∣ ≥ 4 τ , one has χ F = 1 \chi_F=1 χ F = 1 and ∣ p δ ∣ ≥ ( 1 − 1 / 2 ) ∣ ξ ∣ 2 |p_\delta|\ge(1-1/\sqrt{2})|\xi|^2 ∣ p δ ∣ ≥ ( 1 − 1/ 2 ) ∣ ξ ∣ 2 ; direct differentiation gives ∣ ∂ ξ β a F ∣ ≤ C ∣ ξ ∣ − 1 − ∣ β ∣ |\partial_\xi^\beta a_F|\le C|\xi|^{-1-|\beta|} ∣ ∂ ξ β a F ∣ ≤ C ∣ ξ ∣ − 1 − ∣ β ∣ , which is bounded by the right side of (133) for 0 < γ < 1 0<\gamma<1 0 < γ < 1 . The bounds hold in every transition region, and all frequency derivatives in the far region fall on the smooth symbol factors, while the singular multiplier is used at zeroth order on the characteristic set. Since the free inverse is a multiplier, Op ( r 1 , 0 ) E δ = Op ( a F ) \operatorname{Op}(r_{1,0})E_\delta=\operatorname{Op}(a_F) Op ( r 1 , 0 ) E δ = Op ( a F ) exactly. The ordinary kernel estimate bounds its localization composed with H 0 H_0 H 0 by O ( τ − γ ) O(\tau^{-\gamma}) O ( τ − γ ) .
Combining the estimates gives the localized error on L 2 ( R 3 ) L^2(\mathbb{R}^3) L 2 ( R 3 ) :
T τ , γ = ψ 3 ( S + E e r r ) , ∥ T τ , γ ∥ L 2 → L 2 ≤ C ( τ − γ + τ γ − 1 ) . (134) \mathcal{T}_{\tau,\gamma}=\psi_3(S+E_{\mathrm{err}}), \qquad\lVert\mathcal{T}_{\tau,\gamma}\rVert_{L^2\to L^2}\le C(\tau^{-\gamma}+\tau^{\gamma-1}). \tag*{(134)} T τ , γ = ψ 3 ( S + E err ) , ∥ T τ , γ ∥ L 2 → L 2 ≤ C ( τ − γ + τ γ − 1 ) . (134)
The derivatives in the transition region of χ s p \chi_{\mathrm{sp}} χ sp disappear after localization because χ s p = 1 \chi_{\mathrm{sp}}=1 χ sp = 1 near supp ψ 3 \operatorname{supp}\psi_3 supp ψ 3 . Both terms tend to zero for 0 < γ < 1 0<\gamma<1 0 < γ < 1 ; the choice γ = 1 / 2 \gamma=1/2 γ = 1/2 optimizes their common decay rate.
For each index put a 2 , a = ψ 2 b a a_{2,a}=\psi_2 b_a a 2 , a = ψ 2 b a . This is a compactly supported spatial symbol even for the far index a = 0 a=0 a = 0 , and sup τ s 6 , 2 ( a 2 , a ) < ∞ \sup_\tau s_{6,2}(a_{2,a})<\infty sup τ s 6 , 2 ( a 2 , a ) < ∞ by (120) and the cutoff estimates. Applying (B.10) to a 2 , a a_{2,a} a 2 , a , the same free field E δ H a g E_\delta H_a g E δ H a g satisfies
τ ∥ Op ( a 2 , a ) E δ H a g ∥ 2 + ∥ Op ( a 2 , a ) E δ H a g ∥ H 1 ≤ C ∥ H a g ∥ L 2 ≤ C ∥ g ∥ 2 . \tau\lVert\operatorname{Op}(a_{2,a})E_\delta H_a g\rVert_2+\lVert\operatorname{Op}(a_{2,a})E_\delta H_a g\rVert_{H^1}\le C\lVert H_a g\rVert_{L^2}\le C\lVert g\rVert_2. τ ∥ Op ( a 2 , a ) E δ H a g ∥ 2 + ∥ Op ( a 2 , a ) E δ H a g ∥ H 1 ≤ C ∥ H a g ∥ L 2 ≤ C ∥ g ∥ 2 .
The last step uses (123) (or Plancherel for a = 0 a=0 a = 0 ). As the parameter cover has a fixed finite number of indices and ψ 2 Op ( b a ) = Op ( ψ 2 b a ) \psi_2\operatorname{Op}(b_a)=\operatorname{Op}(\psi_2b_a) ψ 2 Op ( b a ) = Op ( ψ 2 b a ) , summing proves
τ ∥ ψ 2 Q τ , γ g ∥ 2 + ∥ ψ 2 Q τ , γ g ∥ H 1 ≤ C ∥ g ∥ 2 . (135) \tau\lVert\psi_2Q_{\tau,\gamma}g\rVert_2+\lVert\psi_2Q_{\tau,\gamma}g\rVert_{H^1}\le C\lVert g\rVert_2. \tag*{(135)} τ ∥ ψ 2 Q τ , γ g ∥ 2 + ∥ ψ 2 Q τ , γ g ∥ H 1 ≤ C ∥ g ∥ 2 . (135)
In particular, the H 1 H^1 H 1 bound uses the weighted derivative estimate for the selected free inverse and is controlled by ∥ g ∥ L 2 \lVert g\rVert_{L^2} ∥ g ∥ L 2 for arbitrary L 2 L^2 L 2 sources.
The Neumann inverse and support properties
Let C e r r C_{\mathrm{err}} C err be a constant for which the right side of (134) is at most C e r r ( τ − γ + τ γ − 1 ) C_{\mathrm{err}}(\tau^{-\gamma}+\tau^{\gamma-1}) C err ( τ − γ + τ γ − 1 ) on the selected compact direction family. Choose τ 0 \tau_0 τ 0 so that
C e r r ( τ 0 − γ + τ 0 γ − 1 ) ≤ 1 2 . (136) C_{\mathrm{err}}(\tau_0^{-\gamma}+\tau_0^{\gamma-1})\le\frac{1}{2}. \tag*{(136)} C err ( τ 0 − γ + τ 0 γ − 1 ) ≤ 2 1 . (136)
Both exponents are negative, so the same inequality holds for every τ ≥ τ 0 \tau\ge\tau_0 τ ≥ τ 0 . Define on L 2 ( R 3 ) L^2(\mathbb{R}^3) L 2 ( R 3 )
G τ , γ = ( I + T τ , γ ) − 1 = ∑ k = 0 ∞ ( − T τ , γ ) k , R δ = ψ 2 Q τ , γ G τ , γ . (137) G_{\tau,\gamma}=(I+T_{\tau,\gamma})^{-1}=\sum_{k=0}^{\infty}(-T_{\tau,\gamma})^k,\qquad R_\delta=\psi_2Q_{\tau,\gamma}G_{\tau,\gamma}. \tag*{(137)} G τ , γ = ( I + T τ , γ ) − 1 = k = 0 ∑ ∞ ( − T τ , γ ) k , R δ = ψ 2 Q τ , γ G τ , γ . (137)
Here G = G τ , γ G=G_{\tau,\gamma} G = G τ , γ and T = T τ , γ T=T_{\tau,\gamma} T = T τ , γ . Estimate (136) makes the series converge in operator norm, with ∥ G ∥ 2 → 2 ≤ 2 \lVert G\rVert_{2\to2}\le2 ∥ G ∥ 2 → 2 ≤ 2 . For a source f f f , extended by zero, set g = G f g=Gf g = G f . Then ∥ g ∥ 2 ≤ 2 ∥ f ∥ 2 \lVert g\rVert_2\le2\lVert f\rVert_2 ∥ g ∥ 2 ≤ 2 ∥ f ∥ 2 . Because T T T has output supported in supp ψ 3 \operatorname{supp}\psi_3 supp ψ 3 and supp f ⊂ K s r c ⊂ { ψ 3 = 1 } \operatorname{supp}f\subset K_{\mathrm{src}}\subset\{\psi_3=1\} supp f ⊂ K src ⊂ { ψ 3 = 1 } , the identity g = f − T g g=f-Tg g = f − T g implies
supp f ⊂ K s r c , supp g ⊂ supp ψ 3 , supp H a g ⊂ supp ψ 3 , \operatorname{supp}f\subset K_{\mathrm{src}},\qquad\operatorname{supp}g\subset\operatorname{supp}\psi_3,\qquad\operatorname{supp}H_a g\subset\operatorname{supp}\psi_3, supp f ⊂ K src , supp g ⊂ supp ψ 3 , supp H a g ⊂ supp ψ 3 ,
E δ H a g may have a noncompact tail , supp R δ f ⊂ supp ψ 2 ⋐ Ω 0 . (138) E_\delta H_a g\ \text{may have a noncompact tail},\qquad\operatorname{supp}R_\delta f\subset\operatorname{supp}\psi_2\Subset\Omega_0. \tag*{(138)} E δ H a g may have a noncompact tail , supp R δ f ⊂ supp ψ 2 ⋐ Ω 0 . (138)
The construction localizes the final output while retaining the free tail in the preceding estimates. Since ψ 2 = ψ 3 = 1 \psi_2=\psi_3=1 ψ 2 = ψ 3 = 1 near all of Ω 1 \Omega_1 Ω 1 , differentiation of the final cutoff contributes no terms there. Formula (127) and g + T g = f g+Tg=f g + T g = f give
P δ R δ f = g + ( S + E e r r ) g = g + T g = f in Ω 1 . P_\delta R_\delta f=g+(S+E_{\mathrm{err}})g=g+Tg=f\qquad\text{in }\Omega_1. P δ R δ f = g + ( S + E err ) g = g + T g = f in Ω 1 .
In particular, with C p a r a C_{\mathrm{para}} C para denoting the constant in (135) , the same field R δ f = ψ 2 Q τ , γ g R_\delta f=\psi_2Q_{\tau,\gamma}g R δ f = ψ 2 Q τ , γ g satisfies
τ ∥ R δ f ∥ 2 + ∥ R δ f ∥ H 1 ≤ C p a r a ∥ g ∥ 2 ≤ 2 C p a r a ∥ f ∥ 2 . \tau\lVert R_\delta f\rVert_2+\lVert R_\delta f\rVert_{H^1}\le C_{\mathrm{para}}\lVert g\rVert_2\le2C_{\mathrm{para}}\lVert f\rVert_2. τ ∥ R δ f ∥ 2 + ∥ R δ f ∥ H 1 ≤ C para ∥ g ∥ 2 ≤ 2 C para ∥ f ∥ 2 .
The displayed equation holds on all of Ω 1 \Omega_1 Ω 1 , whereas the norms are taken on Ω 0 \Omega_0 Ω 0 . The free outputs E δ H a g E_\delta H_a g E δ H a g retain their possible noncompact tails, while the final cutoff ψ 2 \psi_2 ψ 2 localizes the right-inverse output inside Ω 0 \Omega_0 Ω 0 . Since the frames, compact subcharts, and cutoffs were fixed before f f f and τ \tau τ were chosen, these constants depend only on the finite data stated in Proposition 5.1 .
Perturbation and uniform bounds on finite-regularity classes
The estimates above also permit uniformity on an explicitly bounded coefficient class. To apply compactness while retaining the hypotheses at a limit coefficient, we first extend the fixed-coefficient construction to the finite regularity needed by its Fourier kernels.
Proposition 5.2. For matrix coefficients
A j ∈ C 15 ( R 3 ; M m ( C ) ) , q ∈ C c 6 ( R 3 ; M m ( C ) ) , A_j\in C^{15}(\mathbb{R}^3;M_m(\mathbb{C})),\qquad q\in C^6_{\mathrm{c}}(\mathbb{R}^3;M_m(\mathbb{C})), A j ∈ C 15 ( R 3 ; M m ( C )) , q ∈ C c 6 ( R 3 ; M m ( C )) , the conclusions of Proposition 5.1 remain valid, with the equation interpreted distributionally. The source and output spaces, support properties, and simultaneous two-norm estimate are unchanged.
Proof. The moving-coordinate coefficient − i A v ^ + i v 1 / 2 -iA_{\hat{v}+iv_{1}}/2 − i A v ^ + i v 1 /2 is jointly C 15 C^{15} C 15 in leaf, transverse and local real direction variables. Apply Proposition 4.7 with J = 12 J=12 J = 12 . It gives the frame and inverse derivatives with ∣ α ∣ ≤ 10 |\alpha|\le10 ∣ α ∣ ≤ 10 , ∣ β ∣ ≤ 2 |\beta|\le2 ∣ β ∣ ≤ 2 in (114) , because their total order is at most twelve. A finite cover of the fixed compact parameter set gives the required bounds and weights. The symbol estimates use at most eight spatial and two frequency derivatives of b a , d a b_a,d_a b a , d a ; after forming r 0 , a , r 1 , a r_{0,a},r_{1,a} r 0 , a , r 1 , a they use at most six spatial derivatives of A j , q A_j,q A j , q . These are within the stated regularity.
The Fourier–Schur proofs require precisely these finite derivatives: application of ( 1 − Δ x ) 3 (1-\Delta_x)^3 ( 1 − Δ x ) 3 to compact spatial symbols and two frequency derivatives for weight exchange. Their absolutely integrable kernel bounds, the mean-value formula for composition, and the removal of frequency cutoffs remain valid. The differential identities (125) –(126) require only two spatial derivatives of the symbols and hold in distributions. The displayed finite derivative bounds therefore suffice for (134) and (135) . The same Neumann series and source-support argument give the selected right inverse on all of Ω 1 \Omega_1 Ω 1 . Extension from smooth sources uses these bounds and continuity of P δ : H l o c 1 → H l o c − 1 P_\delta:H^1_{\mathrm{loc}}\to H^{-1}_{\mathrm{loc}} P δ : H loc 1 → H loc − 1 , not smoothness of the resulting solution. □
In the perturbation estimates below, matrix norms are Euclidean operator norms. Use the fixed cutoff χ 0 \chi_0 χ 0 of (112) , and set
X = { f ∈ L 2 ( Ω 0 ; C m ) : supp f ⊂ K s r c } , ∥ u ∥ τ = τ ∥ u ∥ L 2 ( Ω 0 ) + ∥ u ∥ H 1 ( Ω 0 ) , c χ = ∥ χ 0 ∥ ∞ . (139) \mathcal{X}=\{f\in L^2(\Omega_0;\mathbb{C}^m):\operatorname{supp}f\subset K_{\mathrm{src}}\},\qquad
\|u\|_\tau=\tau\|u\|_{L^2(\Omega_0)}+\|u\|_{H^1(\Omega_0)},\qquad
c_\chi=\|\chi_0\|_\infty. \tag*{(139)} X = { f ∈ L 2 ( Ω 0 ; C m ) : supp f ⊂ K src } , ∥ u ∥ τ = τ ∥ u ∥ L 2 ( Ω 0 ) + ∥ u ∥ H 1 ( Ω 0 ) , c χ = ∥ χ 0 ∥ ∞ . (139)
The space X \mathcal{X} X carries the L 2 L^2 L 2 norm; support here is essential support after zero extension. In particular X \mathcal{X} X is a closed Hilbert subspace and c χ ≥ 1 c_\chi\ge1 c χ ≥ 1 .
Lemma 5.3. Let P δ ∘ P^\circ_\delta P δ ∘ be a fixed reference operator of (111) with coefficients A j ∘ , q ∘ A_j^\circ,q^\circ A j ∘ , q ∘ . Suppose its selected inverse R δ ∘ : X → H 1 ( Ω 0 ) R^\circ_\delta:\mathcal{X}\to H^1(\Omega_0) R δ ∘ : X → H 1 ( Ω 0 ) satisfies P δ ∘ R δ ∘ f = f P^\circ_\delta R^\circ_\delta f=f P δ ∘ R δ ∘ f = f in Ω 1 \Omega_1 Ω 1 and ∥ R δ ∘ f ∥ τ ≤ C ∘ ∥ f ∥ 2 \|R^\circ_\delta f\|_\tau\le C^\circ\|f\|_2 ∥ R δ ∘ f ∥ τ ≤ C ∘ ∥ f ∥ 2 for τ ≥ τ ∘ \tau\ge\tau_\circ τ ≥ τ ∘ , uniformly in the chosen direction family. For another operator with bounded coefficients, put
η A = ∑ j = 1 3 ∥ A j − A j ∘ ∥ L ∞ ( Ω 0 ) , η q = ∥ q − q ∘ ∥ L ∞ ( Ω 0 ) . \eta_A=\sum_{j=1}^{3}\|A_j-A_j^\circ\|_{L^\infty(\Omega_0)},\qquad
\eta_q=\|q-q^\circ\|_{L^\infty(\Omega_0)}. η A = j = 1 ∑ 3 ∥ A j − A j ∘ ∥ L ∞ ( Ω 0 ) , η q = ∥ q − q ∘ ∥ L ∞ ( Ω 0 ) . If
C ∘ c χ ( 4 η A + τ − 1 η q ) ≤ 1 2 , τ ≥ τ ∘ , (140) C^\circ c_\chi(4\eta_A+\tau^{-1}\eta_q)\le\frac{1}{2},\qquad\tau\ge\tau_\circ, \tag*{(140)} C ∘ c χ ( 4 η A + τ − 1 η q ) ≤ 2 1 , τ ≥ τ ∘ , (140) then the source-independent linear operator
T δ p = χ 0 ( P δ − P δ ∘ ) R δ ∘ , R δ p = R δ ∘ ( I X + T δ p ) − 1 (141) T^\mathrm{p}_\delta=\chi_0(P_\delta-P^\circ_\delta)R^\circ_\delta,\qquad
R^\mathrm{p}_\delta=R^\circ_\delta(I_{\mathcal{X}}+T^\mathrm{p}_\delta)^{-1} \tag*{(141)} T δ p = χ 0 ( P δ − P δ ∘ ) R δ ∘ , R δ p = R δ ∘ ( I X + T δ p ) − 1 (141) is well defined and satisfies
P δ R δ p f = f in Ω C G O , ∥ R δ p f ∥ τ ≤ 2 C ∘ ∥ f ∥ 2 . P_\delta R^\mathrm{p}_\delta f=f\quad\text{in }\Omega_{\mathrm{CGO}},\qquad
\|R^\mathrm{p}_\delta f\|_\tau\le2C^\circ\|f\|_2. P δ R δ p f = f in Ω CGO , ∥ R δ p f ∥ τ ≤ 2 C ∘ ∥ f ∥ 2 . If all outputs of R δ ∘ R^\circ_\delta R δ ∘ are supported in a fixed closed set, the same support bound holds for all outputs of R δ p R^\mathrm{p}_\delta R δ p . Two such operators, selected using the same reference, satisfy
R δ p − R δ p ′ = − R δ p χ 0 ( P δ − P δ ′ ) R δ p ′ . (142) R^\mathrm{p}_\delta-R^{\mathrm{p}\prime}_\delta
=-R^\mathrm{p}_\delta\chi_0(P_\delta-P^\prime_\delta)R^{\mathrm{p}\prime}_\delta. \tag*{(142)} R δ p − R δ p ′ = − R δ p χ 0 ( P δ − P δ ′ ) R δ p ′ . (142) In particular their difference, as an operator from X \mathcal{X} X to the norm ∥ ⋅ ∥ τ \|\cdot\|_\tau ∥ ⋅ ∥ τ , is at most
4 ( C ∘ ) 2 c χ ( 4 ∑ j ∥ A j − A j ′ ∥ ∞ + τ − 1 ∥ q − q ′ ∥ ∞ ) . 4(C^\circ)^2c_\chi\left(4\sum_j\|A_j-A'_j\|_\infty+\tau^{-1}\|q-q'\|_\infty\right). 4 ( C ∘ ) 2 c χ ( 4 j ∑ ∥ A j − A j ′ ∥ ∞ + τ − 1 ∥ q − q ′ ∥ ∞ ) .
Proof. The difference has order at most one, so it acts on H 1 H^1 H 1 without requiring a second derivative of the inverse output. Since ∣ δ j ∣ ≤ 2 τ |\delta_j|\le\sqrt{2\tau} ∣ δ j ∣ ≤ 2 τ , the triangle inequality gives
∥ ( P δ − P δ ∘ ) u ∥ 2 ≤ ( 4 η A + τ − 1 η q ) ∥ u ∥ τ . (143) \|(P_\delta-P^\circ_\delta)u\|_2\le(4\eta_A+\tau^{-1}\eta_q)\|u\|_\tau. \tag*{(143)} ∥ ( P δ − P δ ∘ ) u ∥ 2 ≤ ( 4 η A + τ − 1 η q ) ∥ u ∥ τ . (143)
Multiplication by χ 0 \chi_0 χ 0 puts the source inside K s r c K_{\mathrm{src}} K src . Thus T δ ρ T_\delta^\rho T δ ρ is an operator on X \mathcal{X} X with norm at most the left side of (140) . Its Neumann inverse has norm at most two. For g = ( I + T δ ρ ) − 1 f g=(I+T_\delta^\rho)^{-1}f g = ( I + T δ ρ ) − 1 f , the field u = R δ ρ g u=R_\delta^\rho g u = R δ ρ g satisfies, on Ω 1 \Omega_1 Ω 1 ,
P δ u = g + ( P δ − P δ ∘ ) R δ ρ g , g + χ 0 ( P δ − P δ ∘ ) R δ ρ g = f . P_\delta u=g+(P_\delta-P_\delta^\circ)R_\delta^\rho g,\qquad g+\chi_0(P_\delta-P_\delta^\circ)R_\delta^\rho g=f. P δ u = g + ( P δ − P δ ∘ ) R δ ρ g , g + χ 0 ( P δ − P δ ∘ ) R δ ρ g = f .
These equations give P δ u = f P_\delta u=f P δ u = f where χ 0 = 1 \chi_0=1 χ 0 = 1 , in particular on Ω C G O \Omega_{\mathrm{CGO}} Ω CGO . The cutoff multiplies the residual rather than the output; hence the localized equation contains only the original residual and no cutoff derivatives. The reference estimate gives the two-norm bound and output support. Finally apply ( I + T ) − 1 − ( I + T ′ ) − 1 = ( I + T ) − 1 ( T ′ − T ) ( I + T ′ ) − 1 (I+T)^{-1}-(I+T')^{-1}=(I+T)^{-1}(T'-T)(I+T')^{-1} ( I + T ) − 1 − ( I + T ′ ) − 1 = ( I + T ) − 1 ( T ′ − T ) ( I + T ′ ) − 1 on X \mathcal{X} X to obtain (142) . Estimate (143) between the two selected inverse bounds proves the last assertion. □
Proposition 5.4. Fix the geometry in (112) , a compact set K ⋐ Ω 0 K\Subset\Omega_0 K ⋐ Ω 0 , the matrix size, a compact family of normalized phase directions, and finite numbers M A , M q ≥ 0 M_A,M_q\ge0 M A , M q ≥ 0 . For every smooth first-order coefficient triple satisfying
supp A j ⊂ K , ∑ j = 1 3 ∥ A j ∥ C 16 ( R 3 ) ≤ M A , (144) \operatorname{supp} A_j\subset K,\qquad\sum_{j=1}^{3}\lVert A_j\rVert_{C^{16}(\mathbb{R}^3)}\le M_A, \tag*{(144)} supp A j ⊂ K , j = 1 ∑ 3 ∥ A j ∥ C 16 ( R 3 ) ≤ M A , (144) and every matrix q ∈ L ∞ ( Ω 0 ) q\in L^\infty(\Omega_0) q ∈ L ∞ ( Ω 0 ) with ∥ q ∥ ∞ ≤ M q \lVert q\rVert_\infty\le M_q ∥ q ∥ ∞ ≤ M q , there is a selected linear right inverse R δ , A , q : X → H 1 ( Ω 0 ) R_{\delta,A,q}:\mathcal{X}\to H^1(\Omega_0) R δ , A , q : X → H 1 ( Ω 0 ) such that
P δ R δ , A , q f = f in Ω C G O , ∥ R δ , A , q f ∥ τ ≤ C ∥ f ∥ 2 ( τ ≥ τ ∗ ) . (145) P_\delta R_{\delta,A,q}f=f\quad\text{in }\Omega_{\mathrm{CGO}},\qquad\lVert R_{\delta,A,q}f\rVert_\tau\le C\lVert f\rVert_2\quad(\tau\ge\tau_*). \tag*{(145)} P δ R δ , A , q f = f in Ω CGO , ∥ R δ , A , q f ∥ τ ≤ C ∥ f ∥ 2 ( τ ≥ τ ∗ ) . (145) The constants C , τ ∗ C,\tau_* C , τ ∗ depend only on the fixed data and cutoffs; the source enters the estimates through its L 2 L^2 L 2 norm.
Proof. Let A \mathcal{A} A be the closure of the triples in (144) in C 15 ( R 3 ) C^{15}(\mathbb{R}^3) C 15 ( R 3 ) . It is compact. Indeed all derivatives through order fifteen are uniformly bounded and equicontinuous on a fixed compact ball, using the sixteenth derivative bounds. The Arzelà–Ascoli theorem, that a bounded equicontinuous family on a compact set is relatively compact in the uniform norm, gives a subsequence for each of the finitely many derivatives. Their compatibility, obtained by integration along coordinate segments, identifies the limit as a C 15 C^{15} C 15 function. All limits remain supported in K K K , so convergence holds globally.
Every A ∘ ∈ A A^\circ\in\mathcal{A} A ∘ ∈ A , including nonsmooth limits, has a reference inverse for ( A ∘ , 0 ) (A^\circ,0) ( A ∘ , 0 ) by Proposition 5.2. Denote its constants by C ∘ , τ ∘ C_\circ,\tau_\circ C ∘ , τ ∘ . The neighborhoods
{ A ∈ A : ∑ j ∥ A j − A j ∘ ∥ ∞ < ( 16 c χ C ∘ ) − 1 } \left\{A\in\mathcal{A}:\sum_j\lVert A_j-A_j^\circ\rVert_\infty<(16c_\chi C_\circ)^{-1}\right\} { A ∈ A : j ∑ ∥ A j − A j ∘ ∥ ∞ < ( 16 c χ C ∘ ) − 1 }
are open in C 15 C^{15} C 15 and cover A \mathcal{A} A . Choose a finite subcover with reference indices a = 1 , … , N a=1,\ldots,N a = 1 , … , N , and put
C = 2 max a C a , τ ∗ ≥ max a { 1 , τ a , 4 c χ C a M q } . (146) C=2\max_a C_a,\qquad\tau_*\ge\max_a\{1,\tau_a,4c_\chi C_aM_q\}. \tag*{(146)} C = 2 a max C a , τ ∗ ≥ a max { 1 , τ a , 4 c χ C a M q } . (146)
Select a reference neighborhood containing A A A , independently of the source. Its first-order perturbation contributes less than 1 / 4 1/4 1/4 to (140) ; the bounded zeroth-order term contributes at most 1 / 4 1/4 1/4 when τ ≥ τ ∗ \tau\ge\tau_* τ ≥ τ ∗ . Lemma 5.3 therefore gives (145) . The constants of the finite references are fixed by the stated class and geometry. The finite-regularity reference construction is essential: density of smooth coefficients alone would not supply inverse neighborhoods covering their nonsmooth limits. □
Complex geometrical optics solutions for the reduced elastic system
Set
c = μ − 1 / 2 , κ = λ + μ λ + 2 μ , m = ∇ log μ , d = κ μ , e r e d = 1 2 ( 3 κ − 2 ) μ . c=\mu^{-1/2},\qquad\kappa=\frac{\lambda+\mu}{\lambda+2\mu},\qquad\mathbf{m}=\nabla\log\mu,\qquad d=\kappa\sqrt{\mu},\qquad e^{\mathrm{red}}=\frac{1}{2}(3\kappa-2)\sqrt{\mu}. c = μ − 1/2 , κ = λ + 2 μ λ + μ , m = ∇ log μ , d = κ μ , e red = 2 1 ( 3 κ − 2 ) μ .
The reduced system is
M W = Δ I 4 + ∑ k = 1 3 A k p h y s ∂ k + V W , A k p h y s = ( − m e k T F μ e k d e k T 0 ) (147) M^{W}=\Delta I_{4}+\sum_{k=1}^{3}A_{k}^{\mathrm{phys}}\partial_{k}+V^{W},\qquad A_{k}^{\mathrm{phys}}=\begin{pmatrix}-\mathbf{m}e_{k}^{T}&F_{\mu}e_{k}\\ de_{k}^{T}&0\end{pmatrix} \tag*{(147)} M W = Δ I 4 + k = 1 ∑ 3 A k phys ∂ k + V W , A k phys = ( − m e k T d e k T F μ e k 0 ) (147)
Here e 1 , e 2 , e 3 e_{1},e_{2},e_{3} e 1 , e 2 , e 3 are the standard coordinate vectors in R 3 \mathbb{R}^{3} R 3 . The matrices F μ , V μ F_{\mu},V_{\mu} F μ , V μ and the scalars e r e d , q μ e^{\mathrm{red}},q_{\mu} e red , q μ are defined in (63) . Collecting its zeroth-order terms gives
V W = ( V μ + ( W / μ ) I 3 F μ m + μ − 3 / 2 ( 2 W m − ∇ W ) − e r e d m T q μ + ( 1 − κ ) W / μ ) . V^{W}=\begin{pmatrix}V_{\mu}+(W/\mu)I_{3}&F_{\mu}\mathbf{m}+\mu^{-3/2}(2W\mathbf{m}-\nabla W)\\-e^{\mathrm{red}}\mathbf{m}^{T}&q_{\mu}+(1-\kappa)W/\mu\end{pmatrix}. V W = ( V μ + ( W / μ ) I 3 − e red m T F μ m + μ − 3/2 ( 2 W m − ∇ W ) q μ + ( 1 − κ ) W / μ ) .
Thus V W V^{W} V W is a 4 × 4 4\times4 4 × 4 multiplication matrix, including the potential and its gradient. Proposition 2.7 supplies the common extension, with constant exterior Lamé coefficients and W = 0 W=0 W = 0 sufficiently far away.
Lemma 5.5. Let λ ∗ , μ ∗ \lambda_{*},\mu_{*} λ ∗ , μ ∗ be the constant exterior Lamé coefficients from Proposition 2.7, and put κ ∗ = ( λ ∗ + μ ∗ ) / ( λ ∗ + 2 μ ∗ ) \kappa_{*}=(\lambda_{*}+\mu_{*})/(\lambda_{*}+2\mu_{*}) κ ∗ = ( λ ∗ + μ ∗ ) / ( λ ∗ + 2 μ ∗ ) and d ∗ = κ ∗ μ ∗ d_{*}=\kappa_{*}\sqrt{\mu_{*}} d ∗ = κ ∗ μ ∗ . With e 4 = ( 0 , 0 , 0 , 1 ) T ∈ C 4 e_{4}=(0,0,0,1)^{T}\in\mathbb{C}^{4} e 4 = ( 0 , 0 , 0 , 1 ) T ∈ C 4 , define
A k , ∗ = d ∗ e 4 ( e k T , 0 ) , N ( x ) = ∑ k x k A k , ∗ , B b g ( x ) = I 4 − 1 2 N ( x ) . A_{k,*}=d_{*}e_{4}(e_{k}^{T},0),\qquad N(x)=\sum_{k}x_{k}A_{k,*},\qquad\mathcal{B}_{\mathrm{bg}}(x)=I_{4}-\frac{1}{2}N(x). A k , ∗ = d ∗ e 4 ( e k T , 0 ) , N ( x ) = k ∑ x k A k , ∗ , B bg ( x ) = I 4 − 2 1 N ( x ) . Then N 2 = 0 N^{2}=0 N 2 = 0 , det B b g = 1 \det\mathcal{B}_{\mathrm{bg}}=1 det B bg = 1 , B b g − 1 = I 4 + N / 2 \mathcal{B}_{\mathrm{bg}}^{-1}=I_{4}+N/2 B bg − 1 = I 4 + N /2 , and
M ∗ : = Δ I 4 + ∑ k A k , ∗ ∂ k , B b g − 1 M ∗ B b g = Δ I 4 . M_{*}:=\Delta I_{4}+\sum_{k}A_{k,*}\partial_{k},\qquad\mathcal{B}_{\mathrm{bg}}^{-1}M_{*}\mathcal{B}_{\mathrm{bg}}=\Delta I_{4}. M ∗ := Δ I 4 + k ∑ A k , ∗ ∂ k , B bg − 1 M ∗ B bg = Δ I 4 . If A k p h y s − A k , ∗ A_{k}^{\mathrm{phys}}-A_{k,*} A k phys − A k , ∗ and V W V^{W} V W in (147) are smooth and compactly supported, then all lower-order coefficients of
M ~ W = B b g − 1 M W B b g = Δ I 4 + ∑ k A ~ k ∂ k + V ~ W (148) \widetilde{M}^{W}=\mathcal{B}_{\mathrm{bg}}^{-1}M^{W}\mathcal{B}_{\mathrm{bg}}=\Delta I_{4}+\sum_{k}\widetilde{A}_{k}\partial_{k}+\widetilde{V}^{W} \tag*{(148)} M W = B bg − 1 M W B bg = Δ I 4 + k ∑ A k ∂ k + V W (148) are smooth and compactly supported. On each fixed working ball, B b g ± 1 \mathcal{B}_{\mathrm{bg}}^{\pm1} B bg ± 1 and all their finite-order derivatives are bounded independently of h h h .
Proof. Since A j , ∗ A k , ∗ = 0 A_{j,*}A_{k,*}=0 A j , ∗ A k , ∗ = 0 for all j , k j,k j , k , the assertions about N N N and the inverse follow. The product rule gives
A ~ k = B b g − 1 ( 2 ∂ k B b g + A k p h y s B b g ) , V ~ W = B b g − 1 ( Δ B b g + ∑ k A k p h y s ∂ k B b g + V W B b g ) . \widetilde{A}_{k}=\mathcal{B}_{\mathrm{bg}}^{-1}(2\partial_{k}\mathcal{B}_{\mathrm{bg}}+A_{k}^{\mathrm{phys}}\mathcal{B}_{\mathrm{bg}}),\qquad\widetilde{V}^{W}=\mathcal{B}_{\mathrm{bg}}^{-1}\left(\Delta\mathcal{B}_{\mathrm{bg}}+\sum_{k}A_{k}^{\mathrm{phys}}\partial_{k}\mathcal{B}_{\mathrm{bg}}+V^{W}\mathcal{B}_{\mathrm{bg}}\right). A k = B bg − 1 ( 2 ∂ k B bg + A k phys B bg ) , V W = B bg − 1 ( Δ B bg + k ∑ A k phys ∂ k B bg + V W B bg ) .
In the background region, ∂ k B b g = − A k , ∗ / 2 \partial_{k}\mathcal{B}_{\mathrm{bg}}=-A_{k,*}/2 ∂ k B bg = − A k , ∗ /2 and Δ B b g = 0 \Delta\mathcal{B}_{\mathrm{bg}}=0 Δ B bg = 0 , and both parentheses vanish. Thus the conjugated coefficients are supported in the original nonconstant region. The remaining bounds follow from the affine formulas for B b g ± 1 \mathcal{B}_{\mathrm{bg}}^{\pm1} B bg ± 1 .
Proposition 5.6. For the operator in (91) , let R δ R_{\delta} R δ be the source-independent linear right inverse selected in Proposition 5.1; for sources supported in K s r c K_{\mathrm{src}} K src it satisfies (115) . For θ = a + i b \theta=a+ib θ = a + ib with real orthonormal a , b a,b a , b , make the substitution
A k = A ~ k 2 i , q = − V ~ W , δ = i θ h = − b + i a h , τ = h − 1 . (149) A_{k}=\frac{\widetilde{A}_{k}}{2i},\qquad q=-\widetilde{V}^{W},\qquad\delta=\frac{i\theta}{h}=\frac{-b+ia}{h},\qquad\tau=h^{-1}. \tag*{(149)} A k = 2 i A k , q = − V W , δ = h i θ = h − b + ia , τ = h − 1 . (149) Then
R h , θ ( 0 ) F : = − B b g R δ ( B b g − 1 F ) \mathcal{R}_{h,\theta}^{(0)}F:=-\mathcal{B}_{\mathrm{bg}}R_{\delta}\left(\mathcal{B}_{\mathrm{bg}}^{-1}F\right) R h , θ ( 0 ) F := − B bg R δ ( B bg − 1 F )
defines a linear map from L 2 ( Ω 0 ; C 4 ) L^2(\Omega_0;\mathbb{C}^4) L 2 ( Ω 0 ; C 4 ) sources supported in K s r c K_{\mathrm{src}} K src to H 1 ( Ω 0 ; C 4 ) H^1(\Omega_0;\mathbb{C}^4) H 1 ( Ω 0 ; C 4 ) , for 0 < h ≤ τ 0 − 1 0<h\leq\tau_0^{-1} 0 < h ≤ τ 0 − 1 . For every such F F F ,
( e θ ⋅ x / h M W e − θ ⋅ x / h ) R h , θ ( 0 ) F = F in Ω 1 , h − 1 ∥ R h , θ ( 0 ) F ∥ L 2 ( Ω 0 ) + ∥ R h , θ ( 0 ) F ∥ H 1 ( Ω 0 ) ≤ C ′ ∥ F ∥ L 2 ( Ω 0 ) . (150) \left(e^{\theta\cdot x/h}M^W e^{-\theta\cdot x/h}\right)R_{h,\theta}^{(0)}F=F\ \text{ in }\Omega_1,\qquad h^{-1}\left\|R_{h,\theta}^{(0)}F\right\|_{L^2(\Omega_0)}+\left\|R_{h,\theta}^{(0)}F\right\|_{H^1(\Omega_0)}\leq C'\left\|F\right\|_{L^2(\Omega_0)}. \tag*{(150)} ( e θ ⋅ x / h M W e − θ ⋅ x / h ) R h , θ ( 0 ) F = F in Ω 1 , h − 1 R h , θ ( 0 ) F L 2 ( Ω 0 ) + R h , θ ( 0 ) F H 1 ( Ω 0 ) ≤ C ′ ∥ F ∥ L 2 ( Ω 0 ) . (150)
Proof. Since θ ⋅ θ = 0 \theta\cdot\theta=0 θ ⋅ θ = 0 , the convention D = − i ∇ D=-i\nabla D = − i ∇ and the bilinear pairing of § 1.6 give
P δ = − e θ ⋅ x / h M ~ W e − θ ⋅ x / h . P_\delta=-e^{\theta\cdot x/h}\widetilde{M}^W e^{-\theta\cdot x/h}. P δ = − e θ ⋅ x / h M W e − θ ⋅ x / h .
The scalar exponential commutes with B b g \mathcal{B}_{\mathrm{bg}} B bg . Multiplication by B b g − 1 \mathcal{B}_{\mathrm{bg}}^{-1} B bg − 1 preserves source support, and multiplication by B b g ± 1 \mathcal{B}_{\mathrm{bg}}^{\pm1} B bg ± 1 is bounded on the fixed L 2 L^2 L 2 and H 1 H^1 H 1 domains independently of h h h . Both estimates therefore hold for the same output. Replacing θ \theta θ by − θ -\theta − θ gives the opposite phase, with uniform constants on compact direction subcharts. □
For the moving directions in the local right inverse, write w = v ^ + i v 1 w=\widehat{v}+iv_1 w = v + i v 1 and let z g z_g z g be its leaf coordinate. For any complex direction w w w , use the contractions A w = ∑ k w k A k A_w=\sum_k w_kA_k A w = ∑ k w k A k and A ~ w = ∑ k w k A ~ k \widetilde{A}_w=\sum_k w_k\widetilde{A}_k A w = ∑ k w k A k . Here A k A_k A k is from (91) and A ~ k \widetilde{A}_k A k from (148) . The frame equation is
i ( v ^ + i v 1 ) ⋅ ∇ C = A v ^ + i v 1 C ⟺ ∂ z ‾ g C = − i 2 A v ^ + i v 1 C . i(\widehat{v}+iv_1)\cdot\nabla C=A_{\widehat{v}+iv_1}C\qquad\Longleftrightarrow\qquad\partial_{\overline{z}_g}C=-\frac{i}{2}A_{\widehat{v}+iv_1}C. i ( v + i v 1 ) ⋅ ∇ C = A v + i v 1 C ⟺ ∂ z g C = − 2 i A v + i v 1 C .
The substitution (149) , in these coordinates, gives
B g = − i 2 A w = − 1 4 A ~ w . B_g=-\frac{i}{2}A_w=-\frac{1}{4}\widetilde{A}_w. B g = − 2 i A w = − 4 1 A w .
At a fixed phase θ = a + i b \theta=a+ib θ = a + ib , with real orthonormal a , b a,b a , b , let z p h y s = a ⋅ x + i b ⋅ x z_{\mathrm{phys}}=a\cdot x+ib\cdot x z phys = a ⋅ x + ib ⋅ x . The central moving direction is w 0 = i θ = − b + i a w_0=i\theta=-b+ia w 0 = i θ = − b + ia . Thus z g = − b ⋅ x + i a ⋅ x = i z p h y s z_g=-b\cdot x+ia\cdot x=iz_{\mathrm{phys}} z g = − b ⋅ x + ia ⋅ x = i z phys , so ∂ z ‾ g = i ∂ z ‾ p h y s \partial_{\overline{z}_g}=i\partial_{\overline{z}_{\mathrm{phys}}} ∂ z g = i ∂ z phys , while B g = − i A ~ θ / 4 B_g=-i\widetilde{A}_\theta/4 B g = − i A θ /4 . Dividing the equation by i i i gives, in the phase coordinates, ∂ z ‾ p h y s C = − A ~ θ C / 4 \partial_{\overline{z}_{\mathrm{phys}}}C=-\widetilde{A}_\theta C/4 ∂ z phys C = − A θ C /4 . Thus the moving-direction frame is a leading transport frame for the conjugated operator M ~ W \widetilde{M}^W M W . Multiplication by B b g \mathcal{B}_{\mathrm{bg}} B bg gives a leading transport frame for M W M^W M W , since
( 2 D θ + A θ p h y s ) ( B b g C ) = B b g ( 2 D θ + A ~ θ ) C = 0. (2D_\theta+A_\theta^{\mathrm{phys}})(\mathcal{B}_{\mathrm{bg}}C)=\mathcal{B}_{\mathrm{bg}}(2D_\theta+\widetilde{A}_\theta)C=0. ( 2 D θ + A θ phys ) ( B bg C ) = B bg ( 2 D θ + A θ ) C = 0.
Remark 5.7. If a null path θ ( h ) \theta(h) θ ( h ) has real and imaginary parts of common length a ( h ) > 0 a(h)>0 a ( h ) > 0 , bounded above and below, apply the result with θ ^ ( h ) = θ ( h ) / a ( h ) \widehat{\theta}(h)=\theta(h)/a(h) θ ( h ) = θ ( h ) / a ( h ) and h ^ = h / a ( h ) \widehat{h}=h/a(h) h = h / a ( h ) . The estimates retain their stated orders in h h h . The paths ϑ j ( h ) \vartheta_j(h) ϑ j ( h ) in § 8 are normalized and lie in fixed compact subcharts near their limiting directions; the uniform constants above apply to them.
Proposition 5.8. Fix χ 0 ∈ C c ∞ ( int K s r c ) \chi_0\in C^\infty_c(\operatorname{int}K_{\mathrm{src}}) χ 0 ∈ C c ∞ ( int K src ) equal to one near Ω ‾ C G O \overline{\Omega}_{\mathrm{CGO}} Ω CGO . Choose a fixed open set U U U such that supp χ 0 ∪ Ω ‾ C G O ⊂ U ⋐ Ω 1 \operatorname{supp}\chi_0\cup\overline{\Omega}_{\mathrm{CGO}}\subset U\Subset\Omega_1 supp χ 0 ∪ Ω CGO ⊂ U ⋐ Ω 1 . Let s ∈ { + 1 , − 1 } s\in\{+1,-1\} s ∈ { + 1 , − 1 } and let θ \theta θ be a normalized null direction in one of the fixed compact direction families. Suppose smooth amplitudes near U U U satisfy
T θ a 0 = 0 , T θ a 1 = s M W a 0 , T θ = 2 θ ⋅ ∇ + A θ p h y s . (151) T_\theta a_0=0,\qquad T_\theta a_1=sM^W a_0,\qquad T_\theta=2\theta\cdot\nabla+A_\theta^{\mathrm{phys}}. \tag*{(151)} T θ a 0 = 0 , T θ a 1 = s M W a 0 , T θ = 2 θ ⋅ ∇ + A θ phys . (151)
Assume ∥ a 1 ∥ H 2 ( U ) \|a_1\|_{H^2(U)} ∥ a 1 ∥ H 2 ( U ) is uniformly bounded along the selected parameter family. Put
F = χ 0 M W a 1 , r h , s = − R h , s θ ( 0 ) ( h F ) , Z h , s = e − s θ ⋅ x / h ( a 0 + h a 1 + r h , s ) . (152) F=\chi_0M^W a_1,\qquad r_{h,s}=-R_{h,s\theta}^{(0)}(hF),\qquad Z_{h,s}=e^{-s\theta\cdot x/h}(a_0+ha_1+r_{h,s}). \tag*{(152)} F = χ 0 M W a 1 , r h , s = − R h , s θ ( 0 ) ( h F ) , Z h , s = e − s θ ⋅ x / h ( a 0 + h a 1 + r h , s ) . (152)
Then M W Z h , s = 0 M^WZ_{h,s}=0 M W Z h , s = 0 in Ω C G O \Omega_{\mathrm{CGO}} Ω CGO , and this single remainder satisfies
∥ r h , s ∥ L 2 ( Ω 0 ) ≤ C h 2 , ∥ r h , s ∥ H 1 ( Ω 0 ) ≤ C h . (153) \|r_{h,s}\|_{L^2(\Omega_0)}\leq Ch^2,\qquad\|r_{h,s}\|_{H^1(\Omega_0)}\leq Ch. \tag*{(153)} ∥ r h , s ∥ L 2 ( Ω 0 ) ≤ C h 2 , ∥ r h , s ∥ H 1 ( Ω 0 ) ≤ C h . (153)
The constants are uniform for either sign on the fixed compact parameter family. The case s = + 1 s=+1 s = + 1 is the negative physical phase e − θ ⋅ x / h e^{-\theta\cdot x/h} e − θ ⋅ x / h ; the case s = − 1 s=-1 s = − 1 is the positive phase e θ ⋅ x / h e^{\theta\cdot x/h} e θ ⋅ x / h .
Proof. For s = ± 1 s=\pm1 s = ± 1 , the null identity θ ⋅ θ = 0 \theta\cdot\theta=0 θ ⋅ θ = 0 and the product rule give the exact conjugation formula
e s θ ⋅ x / h M W e − s θ ⋅ x / h = M W − s h T θ . (154) e^{s\theta\cdot x/h}M^W e^{-s\theta\cdot x/h}=M^W-\frac{s}{h}T_\theta. \tag*{(154)} e s θ ⋅ x / h M W e − s θ ⋅ x / h = M W − h s T θ . (154)
The transport conditions (151) imply
( M W − s h T θ ) ( a 0 + h a 1 ) = M W a 0 + h M W a 1 − s T θ a 1 = h M W a 1 . (155) \left(M^{W}-\frac{\mathfrak{s}}{h}T_{\theta}\right)(a_{0}+ha_{1})=M^{W}a_{0}+hM^{W}a_{1}-\mathfrak{s}T_{\theta}a_{1}=hM^{W}a_{1}. \tag*{(155)} ( M W − h s T θ ) ( a 0 + h a 1 ) = M W a 0 + h M W a 1 − s T θ a 1 = h M W a 1 . (155)
The compact support of χ 0 \chi_{0} χ 0 in U U U permits extension of F F F by zero. Since M W : H 2 ( U ) → L 2 ( U ) M^{W}:H^{2}(U)\to L^{2}(U) M W : H 2 ( U ) → L 2 ( U ) is bounded,
∥ F ∥ 2 ≤ C ∥ a 1 ∥ H 2 ( U ) . \lVert F\rVert_{2}\leq C\lVert a_{1}\rVert_{H^{2}(U)}. ∥ F ∥ 2 ≤ C ∥ a 1 ∥ H 2 ( U ) .
Proposition 5.6, applied with the normalized direction s θ \mathfrak{s}\theta s θ , shows that
( M W − s h T θ ) r h , s = − h χ 0 M W a 1 in Ω 1 . \left(M^{W}-\frac{\mathfrak{s}}{h}T_{\theta}\right)r_{h,\mathfrak{s}}=-h\chi_{0}M^{W}a_{1}\quad\text{in }\Omega_{1}. ( M W − h s T θ ) r h , s = − h χ 0 M W a 1 in Ω 1 .
Together with (155) , the total conjugated residual is h ( 1 − χ 0 ) M W a 1 h(1-\chi_{0})M^{W}a_{1} h ( 1 − χ 0 ) M W a 1 , which vanishes near Ω ‾ C G O \overline{\Omega}_{\mathrm{CGO}} Ω CGO . This proves the asserted exact equation on the target. The same right-inverse output in (152) obeys, by (150) ,
h − 1 ∥ r h , s ∥ L 2 ( Ω 0 ) + ∥ r h , s ∥ H 1 ( Ω 0 ) ≤ C h ∥ F ∥ 2 , h^{-1}\lVert r_{h,\mathfrak{s}}\rVert_{L^{2}(\Omega_{0})}+\lVert r_{h,\mathfrak{s}}\rVert_{H^{1}(\Omega_{0})}\leq Ch\lVert F\rVert_{2}, h − 1 ∥ r h , s ∥ L 2 ( Ω 0 ) + ∥ r h , s ∥ H 1 ( Ω 0 ) ≤ C h ∥ F ∥ 2 ,
which implies both estimates in (153) . The norm of the inverse and the bound for F F F are uniform over the chosen compact families, including their negatives. For each fixed h h h , local elliptic regularity makes Z h , s Z_{h,\mathfrak{s}} Z h , s smooth in the target whenever the coefficients and amplitudes are smooth. □
Corollary 5.9. Fix the working domains and cutoffs, a compact set K ⋐ Ω 0 K\Subset\Omega_{0} K ⋐ Ω 0 , a constant strongly convex exterior pair ( λ ∗ , μ ∗ ) (\lambda_{*},\mu_{*}) ( λ ∗ , μ ∗ ) , and numbers m 0 > 0 m_{0}>0 m 0 > 0 , M m a t , M W < ∞ M_{\mathrm{mat}},M_{W}<\infty M mat , M W < ∞ . Consider smooth extended materials with
supp ( λ − λ ∗ ) ∪ supp ( μ − μ ∗ ) ⊂ K , μ ≥ m 0 , 3 λ + 2 μ ≥ m 0 on Ω 0 , (156) \operatorname{supp}(\lambda-\lambda_{*})\cup\operatorname{supp}(\mu-\mu_{*})\subset K,\qquad\mu\geq m_{0},\qquad3\lambda+2\mu\geq m_{0}\quad\text{on }\Omega_{0}, \tag*{(156)} supp ( λ − λ ∗ ) ∪ supp ( μ − μ ∗ ) ⊂ K , μ ≥ m 0 , 3 λ + 2 μ ≥ m 0 on Ω 0 , (156) ∥ λ ∥ C 18 ( Ω ‾ 0 ) + ∥ μ ∥ C 18 ( Ω ‾ 0 ) ≤ M m a t , ∥ W ∥ W 1 , ∞ ( Ω 0 ) ≤ M W . \lVert\lambda\rVert_{C^{18}(\overline{\Omega}_{0})}+\lVert\mu\rVert_{C^{18}(\overline{\Omega}_{0})}\leq M_{\mathrm{mat}},\qquad\lVert W\rVert_{W^{1,\infty}(\Omega_{0})}\leq M_{W}. ∥ λ ∥ C 18 ( Ω 0 ) + ∥ μ ∥ C 18 ( Ω 0 ) ≤ M mat , ∥ W ∥ W 1 , ∞ ( Ω 0 ) ≤ M W . There are h ∗ > 0 h_{*}>0 h ∗ > 0 and C < ∞ C<\infty C < ∞ , depending only on these data, such that for every normalized null direction and 0 < h ≤ h ∗ 0<h\leq h_{*} 0 < h ≤ h ∗ a selected linear map R h , θ u R^{\mathrm{u}}_{h,\theta} R h , θ u satisfies
( e θ ⋅ x / h M W e − θ ⋅ x / h ) R h , θ u F = F in Ω C G O , (157) \left(e^{\theta\cdot x/h}M^{W}e^{-\theta\cdot x/h}\right)R^{\mathrm{u}}_{h,\theta}F=F\quad\text{in }\Omega_{\mathrm{CGO}}, \tag*{(157)} ( e θ ⋅ x / h M W e − θ ⋅ x / h ) R h , θ u F = F in Ω CGO , (157) h − 1 ∥ R h , θ u F ∥ L 2 ( Ω 0 ) + ∥ R h , θ u F ∥ H 1 ( Ω 0 ) ≤ C ∥ F ∥ 2 , F ∈ X . h^{-1}\lVert R^{\mathrm{u}}_{h,\theta}F\rVert_{L^{2}(\Omega_{0})}+\lVert R^{\mathrm{u}}_{h,\theta}F\rVert_{H^{1}(\Omega_{0})}\leq C\lVert F\rVert_{2},\qquad F\in\mathcal{X}. h − 1 ∥ R h , θ u F ∥ L 2 ( Ω 0 ) + ∥ R h , θ u F ∥ H 1 ( Ω 0 ) ≤ C ∥ F ∥ 2 , F ∈ X . The superscript u \mathrm{u} u distinguishes this separately selected material-uniform right inverse from the fixed-material selection R h , θ ( 0 ) R^{(0)}_{h,\theta} R h , θ ( 0 ) . The constants are uniform over the coefficient class in (156) and depend on each source only through its L 2 L^{2} L 2 norm. For any family of amplitudes satisfying (151) and a uniform H 2 ( U ) H^{2}(U) H 2 ( U ) bound on a 1 a_{1} a 1 , the remainders obtained with this inverse satisfy (153) uniformly for both phase signs.
Proof. The lower bounds give Σ = ( 3 λ + 2 μ + 4 μ ) / 3 ≥ 5 m 0 / 3 \Sigma=(3\lambda+2\mu+4\mu)/3\geq5m_{0}/3 Σ = ( 3 λ + 2 μ + 4 μ ) /3 ≥ 5 m 0 /3 . In (147) , the first-order matrices involve at most two spatial derivatives of μ \mu μ ; λ \lambda λ enters through its values. Repeated product and chain rules, with denominators bounded below by m 0 m_{0} m 0 and 5 m 0 / 3 5m_{0}/3 5 m 0 /3 , therefore bound their C 16 C^{16} C 16 norms from (156) . The zeroth-order matrix uses at most two derivatives of μ \mu μ , the values of λ \lambda λ , and W W W , ∇ W \nabla W ∇ W ; its L ∞ L^{\infty} L ∞ norm has a bound from the same data.
The affine matrix B b g \mathcal{B}_{\mathrm{bg}} B bg in Lemma 5.5 is fixed because the exterior pair is fixed. Its formulas show that A ~ j \widetilde{A}_{j} A j have support in K K K and uniform C 16 C^{16} C 16 norms, while V ~ W \widetilde{V}^{W} V W has a uniform L ∞ ( Ω 0 ) L^{\infty}(\Omega_{0}) L ∞ ( Ω 0 ) norm. Proposition 5.4 uses this norm bound for V ~ W \widetilde{V}^{W} V W , with its support unrestricted. Apply that proposition with A j = A ~ j / ( 2 i ) A_{j}=\widetilde{A}_{j}/(2i) A j = A j / ( 2 i ) , q = − V ~ W q=-\widetilde{V}^{W} q = − V W and δ = i θ / h \delta=i\theta/h δ = i θ / h . Then
R h , θ u F = − B b g R δ , A , q ( B b g − 1 F ) R^{\mathrm{u}}_{h,\theta}F=-\mathcal{B}_{\mathrm{bg}}R_{\delta,A,q}\left(\mathcal{B}_{\mathrm{bg}}^{-1}F\right) R h , θ u F = − B bg R δ , A , q ( B bg − 1 F )
has the asserted equation and estimates. Multiplication by B b g ± 1 \mathcal{B}_{\mathrm{bg}}^{\pm1} B bg ± 1 preserves source support and is uniformly bounded on the fixed L 2 L^{2} L 2 and H 1 H^{1} H 1 spaces. The set of normalized null directions is compact and contains the negatives, so the constants cover both signs. Finally the coefficients of M W : H 2 ( U ) → L 2 ( U ) M^W:H^2(U)\to L^2(U) M W : H 2 ( U ) → L 2 ( U ) are uniformly bounded under (156) . Thus ∥ χ 0 M W a 1 ∥ 2 ≤ C ∥ a 1 ∥ H 2 ( U ) \lVert\chi_0M^W a_1\rVert_2\le C\lVert a_1\rVert_{H^2(U)} ∥ χ 0 M W a 1 ∥ 2 ≤ C ∥ a 1 ∥ H 2 ( U ) uniformly. Use r h , s = − R u h , s θ ( h χ 0 M W a 1 ) r_{h,s}=-\mathcal{R}_{u_h,s\theta}(h\chi_0M^W a_1) r h , s = − R u h , s θ ( h χ 0 M W a 1 ) . The exact residual identity (155) only requires the right-inverse identity on Ω C G O \Omega_{\mathrm{CGO}} Ω CGO , where χ 0 = 1 \chi_0=1 χ 0 = 1 . It gives an exact solution there and the same uniform L 2 = O ( h 2 ) L^2=O(h^2) L 2 = O ( h 2 ) , H 1 = O ( h ) H^1=O(h) H 1 = O ( h ) bounds for this single remainder.
The bounds in (156) concern the extended coefficients on the construction region. Uniform first-correction amplitudes under explicit potential and test bounds are established in Lemma 6.4 and Proposition 6.14 below. The right-inverse construction on the extended region uses these finite-regularity coefficient bounds; the physical nonresonance hypothesis enters in the definition of the measured map and in boundary determination. Theorem 1.1 is stated for smooth material coefficients.
Consequently, the error in the first-order representation of the unscaled third column is bounded by
C ( h ∥ a 1 ∥ H 1 + ∥ r h ∥ H 1 + h − 1 ∥ r h ∥ 2 ) = O ( h ) . C\left(h\lVert a_1\rVert_{H^1}+\lVert r_h\rVert_{H^1}+h^{-1}\lVert r_h\rVert_2\right)=O(h). C ( h ∥ a 1 ∥ H 1 + ∥ r h ∥ H 1 + h − 1 ∥ r h ∥ 2 ) = O ( h ) .
This estimate holds throughout the target domain, including all transverse cutoff transitions. The identity used to replace the second-order lift applies to the complete exact solution.
Corollary 5.10. Under the Lamé bounds and common exterior background of (156) , the two-, three- and four-component frames used for the transport amplitudes may be chosen on fixed local direction and transverse charts with uniform mixed derivatives through total order six, as may their inverses. The two-component frame may have determinant one. The constants and chart margins are uniform over the material class and all subsequent tests. These transport frames are selected independently of the frames used in the Fourier parametrix.
Proof. Outside K ⋐ Ω 0 K\Subset\Omega_0 K ⋐ Ω 0 the Lamé coefficients have the fixed constant values, so their C 18 C^{18} C 18 bounds and positive lower bounds extend to every fixed larger construction ball, with constants also depending on the exterior pair. The planar coefficients of the systems in Proposition 4.10 use at most two material derivatives. On compact spinor charts, all coordinate matrices and their right inverses have fixed derivative bounds. The chain and product rules in x = z 1 e 1 + z 2 e 2 + t ν θ x=z_1e_1+z_2e_2+t\nu_\theta x = z 1 e 1 + z 2 e 2 + t ν θ therefore bound the full C 12 C^{12} C 12 norm of each planar coefficient by derivatives of the materials through order fourteen, under the derivative bounds in (156) . The direction manifold and the fixed closed transverse interval are compact. Proposition 4.8 gives the assertion after taking a common finite refinement for the three matrix sizes and the spinor charts. Its normalization applies to the trace-zero two-component system. The frame equations depend only on the Lamé coefficients, so the uniform frame estimates use only their stated regularity bounds.
The full matrix moment identities
This section derives the complete 3 × 3 3\times3 3 × 3 matrix of polynomial moment identities from the reduced Green–Betti identity (158) . The main task is to realize the primal and dual quotient frames by physical elastic solutions with compatible high-frequency normalizations. We proceed in four stages. First, we fix the leafwise notation and construct the first transport corrections, together with the simultaneous remainder bounds needed for the physical lifts (Subsection 6.1), we then identify the canonical lift operator and derive its high-frequency asymptotics (Subsection 6.2). Next, we construct two leading polarization columns using scaled lifts and a third column using an unscaled lift whose leading amplitude satisfies J θ A S = 0 J_\theta A^S=0 J θ A S = 0 ; the freedom in the first correction is used to preserve the invertibility of the resulting quotient frame (Subsection 6.3). Finally, we fix the quotient frames before choosing the polynomial and transverse tests, extract the leafwise moment identities, and obtain their uniform realization on bounded test classes (Subsection 6.4). The resulting moment identities are stated in Theorem 6.13.
The first two physical testing columns are normalized by − h -h − h , whereas the third is unscaled. Both limits therefore rely on the same remainder satisfying
∥ r h ∥ L 2 = O ( h 2 ) , ∥ r h ∥ H 1 = O ( h ) , \lVert r_h\rVert_{L^2}=O(h^2),\qquad\lVert r_h\rVert_{H^1}=O(h), ∥ r h ∥ L 2 = O ( h 2 ) , ∥ r h ∥ H 1 = O ( h ) ,
as stated in (153) . All pairings and transposes below are complex-bilinear, with the convention of Section 1.6.
Preliminaries and first-correction estimates
Definition 6.1. The two materials are described by the coefficient triples
( λ j , μ j , W j ) , j = 1 , 2 , (\lambda_j,\mu_j,W_j),\qquad j=1,2, ( λ j , μ j , W j ) , j = 1 , 2 , on the common physical domain Ω \Omega Ω . For formulas involving one fixed material, we suppress the index j j j and write
λ , μ , W , κ , a , c , ℓ , m , d , e r e d , q μ , F μ , V μ , M W , W \lambda,\mu,W,\kappa,a,c,\ell,\mathbf{m},d,e^{\mathrm{red}},q_\mu,F_\mu,V_\mu,M^W,\mathcal{W} λ , μ , W , κ , a , c , ℓ , m , d , e red , q μ , F μ , V μ , M W , W for the corresponding coefficients and derived quantities, with the meanings fixed in (5) , (60) , and (63) . Write M = M 0 M=M^0 M = M 0 for the zero-potential part of the reduced operator.
Use the leaf coordinates z , x ⊥ z,x_\perp z , x ⊥ , the directional derivative D θ D_\theta D θ , and the leading transport operator T θ T_\theta T θ fixed in Section 4.5. The coefficient contraction A θ p h y s A^{\mathrm{phys}}_\theta A θ phys is determined by (147) . The leafwise right inverse C θ C_\theta C θ of D θ D_\theta D θ is defined by the Cauchy formula
( C θ H ) ( z , t ) = 1 2 π ∫ C H ( w , t ) z − w d A ( w ) , t = x ⊥ . (C_\theta H)(z,t)=\frac{1}{2\pi}\int_{\mathcal{C}}\frac{H(w,t)}{z-w}\,dA(w),\qquad t=x_\perp. ( C θ H ) ( z , t ) = 2 π 1 ∫ C z − w H ( w , t ) d A ( w ) , t = x ⊥ . Here H H H has compact support in the leaf variable and d A dA d A is planar area measure. In applications a leaf cutoff is included in H H H . Thus D θ C θ H = H D_\theta C_\theta H=H D θ C θ H = H on the working disk, and the operator leaves the transverse parameter t t t fixed.
Proposition 6.2. Under the assumptions of Section 1, the following constructions and identities are available.
(1) Theorem 4.1 and Proposition 4.10 supply smooth invertible transport frames on the prescribed leaf disk, with all required finite derivatives of each frame and its inverse bounded on compact parameter subcharts.
(2) Fix the nested domains and source set in (112) . For each material, use the selected right inverse R h , θ ( 0 ) R_{h,\theta}^{(0)} R h , θ ( 0 ) of Proposition 5.6; its equation and simultaneous estimate are given in (150) for sources supported in K s r c K_{\mathrm{src}} K src . The phase e − θ ⋅ x / h e^{-\theta\cdot x/h} e − θ ⋅ x / h uses this operator, and the opposite phase uses the operator selected at − θ -\theta − θ , as in Proposition 5.8. The exact solutions are used on Ω C G O \Omega_{\mathrm{CGO}} Ω CGO , with Ω ‾ ⊂ Ω C G O ⊂ Ω 1 \overline{\Omega}\subset\Omega_{\mathrm{CGO}}\subset\Omega_1 Ω ⊂ Ω CGO ⊂ Ω 1 .
(3) Theorem 2.5 and Proposition 2.7 give smooth common extensions of the two coefficient sets, with differences supported in Ω \Omega Ω . Their Dirichlet-to-Neumann maps on the original domain agree. For every exact reduced solution M 1 W 1 z 1 = 0 M_1^{W_1}z_1=0 M 1 W 1 z 1 = 0 and elastic solution ( L 2 + W 2 I ) u 2 = 0 (L_2+W_2I)u_2=0 ( L 2 + W 2 I ) u 2 = 0 , the field y 2 = S 2 T u 2 y_2=S_2^Tu_2 y 2 = S 2 T u 2 satisfies
∫ Ω y 2 T ( M 1 W 1 − M 2 W 2 ) z 1 d x = 0. (158) \int_{\Omega}y_2^T(M_1^{W_1}-M_2^{W_2})z_1\,dx=0. \tag*{(158)} ∫ Ω y 2 T ( M 1 W 1 − M 2 W 2 ) z 1 d x = 0. (158)
The source cutoff, target domain, and transport region are fixed nested neighborhoods. Constants may depend on each fixed test amplitude and its polynomial degree.
Proof. Theorem 2.5 and Proposition 2.7 give the common smooth extension and compact support of all coefficient differences. The boundary value of W W W is recovered at j = 2 j=2 j = 2 in Theorem 2.5. Together with the value and first normal derivative of μ \mu μ , this gives the matching conditions of Proposition 3.4, and hence (158) from the original boundary data.
Theorem 4.1 and Proposition 4.10 supply the required frames and their inverses, jointly smoothly in local real parameters. The leaf primitive is C θ C_\theta C θ from Definition 6.1 ; its local bounds are used in Lemma 6.3 . Fix the disk, parameter chart, frames, and primitive before choosing b , g b,g b , g . The transverse cutoffs, amplitudes, and corrections may depend on these tests; choose h h h last. The transverse tests are supported in the chosen local chart. Use the selected right inverse of Proposition 5.6. Choose the first correction on the common leaf disk by Lemma 6.3 . Proposition 5.8, applied to h χ 0 M W A 1 h\chi_0M^W A_1 h χ 0 M W A 1 , gives a single remainder with L 2 = O ( h 2 ) L^2=O(h^2) L 2 = O ( h 2 ) and H 1 = O ( h ) H^1=O(h) H 1 = O ( h ) . Since χ 0 = 1 \chi_0=1 χ 0 = 1 near all of Ω C G O \Omega_{\mathrm{CGO}} Ω CGO and Ω ‾ ⋐ Ω C G O \overline{\Omega}\Subset\Omega_{\mathrm{CGO}} Ω ⋐ Ω CGO , elliptic regularity makes the resulting exact field smooth near both sides of ∂ Ω \partial\Omega ∂ Ω . Its restriction to Ω \Omega Ω therefore satisfies the Betti identity. Use − θ -\theta − θ for the positive phase. All constants may depend on the fixed test. □
By Proposition 5.8, an exact CGO solution of the form (152) is obtained once the transport conditions (151) are satisfied. The purpose of the following lemma is to construct the first correction a 1 a_1 a 1 for a prescribed leading amplitude a 0 a_0 a 0 , using the leafwise Cauchy right inverse, and to establish the H 2 ( U ) H^2(U) H 2 ( U ) bound required in Proposition 5.8.
Lemma 6.3. Let U U U be the fixed open set in Proposition 5.8. Let π θ ( x ) = z \pi_\theta(x)=z π θ ( x ) = z and π ⊥ ( x ) = x ⊥ \pi_\perp(x)=x_\perp π ⊥ ( x ) = x ⊥ be the leaf and transverse coordinates from Definition 6.1 . Choose a relatively compact transverse interval I ⊥ I_\perp I ⊥ inside the selected compact parameter chart and containing π ⊥ ( U ‾ ) \pi_\perp(\overline{U}) π ⊥ ( U ) , and radii L s r c < L f r L_{\mathrm{src}}<L_{\mathrm{fr}} L src < L fr such that
π θ ( U ‾ ) ⋐ D L s r c ⋐ D L f r . \pi_\theta(\overline{U})\Subset D_{L_{\mathrm{src}}}\Subset D_{L_{\mathrm{fr}}}. π θ ( U ) ⋐ D L src ⋐ D L fr . Define the source cylinder
Q s r c : = { x ∈ R 3 : z ( x ) ∈ D L s r c , x ⊥ ∈ I ⊥ } . (159) Q_{\mathrm{src}}:=\{x\in\mathbb{R}^3:z(x)\in D_{L_{\mathrm{src}}},\ x_\perp\in I_\perp\}. \tag*{(159)} Q src := { x ∈ R 3 : z ( x ) ∈ D L src , x ⊥ ∈ I ⊥ } . (159) Let a 0 a_0 a 0 be smooth on a neighborhood of Q ‾ s r c \overline{Q}_{\mathrm{src}} Q src and satisfy the first transport condition in (151) on a neighborhood of U ‾ \overline{U} U .
Choose the invertible four-component transport frame C θ C_\theta C θ supplied by Theorem 4.1 and Proposition 4.10 , smooth on a neighborhood of D ‾ L f r × I ‾ ⊥ \overline{D}_{L_{\mathrm{fr}}}\times\overline{I}_\perp D L fr × I ⊥ , with
T θ C θ = 0 , det C θ ≠ 0. T_\theta C_\theta=0,\qquad\det C_\theta\ne0. T θ C θ = 0 , det C θ = 0. Choose
χ l e a f ∈ C c ∞ ( D L s r c ) , χ l e a f = 1 near π θ ( U ‾ ) , \chi_{\mathrm{leaf}}\in C^\infty_c(D_{L_{\mathrm{src}}}),\qquad\chi_{\mathrm{leaf}}=1\ \text{near }\pi_\theta(\overline{U}), χ leaf ∈ C c ∞ ( D L src ) , χ leaf = 1 near π θ ( U ) , and let C θ C_\theta C θ be the leafwise Cauchy right inverse from Definition 6.1 . Choose χ ⊥ ∈ C c ∞ ( I ⊥ ) \chi_\perp\in C^\infty_c(I_\perp) χ ⊥ ∈ C c ∞ ( I ⊥ ) such that χ ⊥ = 1 \chi_\perp=1 χ ⊥ = 1 near π ⊥ ( U ‾ ) \pi_\perp(\overline{U}) π ⊥ ( U ) , and set
a ~ 0 ( x ) = χ ⊥ ( x ⊥ ) a 0 ( x ) , \widetilde{a}_0(x)=\chi_\perp(x_\perp)a_0(x), a 0 ( x ) = χ ⊥ ( x ⊥ ) a 0 ( x ) , with zero extension outside I ⊥ I_\perp I ⊥ . For s ∈ { + 1 , − 1 } s\in\{+1,-1\} s ∈ { + 1 , − 1 } , define
a 1 = s 2 C θ C θ ( χ l e a f C θ − 1 M W a ~ 0 ) . (160) a_1=\frac{s}{2}C_\theta C_\theta\left(\chi_{\mathrm{leaf}}C_\theta^{-1}M^W\widetilde{a}_0\right). \tag*{(160)} a 1 = 2 s C θ C θ ( χ leaf C θ − 1 M W a 0 ) . (160) where s s s has the phase convention of Proposition 5.8. Then
T θ a 1 = s M W a 0 on U , T_\theta a_1=sM^W a_0\quad\text{on }U, T θ a 1 = s M W a 0 on U , so ( a 0 , a 1 ) (a_0,a_1) ( a 0 , a 1 ) satisfies (151) on U U U . For each fixed smooth amplitude family, a 1 a_1 a 1 has an H 2 ( U ) H^2(U) H 2 ( U ) bound uniform on the chosen compact parameter subchart; the bound may depend on the fixed coefficients, amplitudes, frames, disks, and cutoffs, but not on h h h .
The source in (160) is the full M W a ~ 0 M^W\widetilde{a}_0 M W a 0 , including all derivatives of χ ⊥ \chi_\perp χ ⊥ . Since χ ⊥ = 1 \chi_\perp=1 χ ⊥ = 1 near π ⊥ ( U ‾ ) \pi_\perp(\overline{U}) π ⊥ ( U ) , M W a ~ 0 = M W a 0 M^W\widetilde{a}_0=M^W a_0 M W a 0 = M W a 0 on U U U . The Cauchy primitive leaves x ⊥ x_\perp x ⊥ fixed, so a 1 a_1 a 1 extends smoothly by zero outside its compact transverse support.
Proof. The product rule and T θ C θ = 0 T_\theta C_\theta=0 T θ C θ = 0 give
T θ ( C θ v ) = 2 C θ D θ v T_\theta(C_\theta v)=2C_\theta D_\theta v T θ ( C θ v ) = 2 C θ D θ v
for every smooth four-component field v v v . Using the right-inverse identity D θ C θ H = H D_\theta C_\theta H=H D θ C θ H = H from Definition 6.1 , we obtain
T θ a 1 = s C θ D θ C θ ( χ l e a f C θ − 1 M W a ~ 0 ) = s C θ χ l e a f C θ − 1 M W a ~ 0 = s M W a ~ 0 = s M W a 0 on U , \begin{aligned}
T_\theta a_1&=sC_\theta D_\theta C_\theta\left(\chi_{\mathrm{leaf}}C_\theta^{-1}M^W\widetilde a_0\right)\\
&=sC_\theta\chi_{\mathrm{leaf}}C_\theta^{-1}M^W\widetilde a_0\\
&=sM^W\widetilde a_0=sM^W a_0\quad\text{on }U,
\end{aligned} T θ a 1 = s C θ D θ C θ ( χ leaf C θ − 1 M W a 0 ) = s C θ χ leaf C θ − 1 M W a 0 = s M W a 0 = s M W a 0 on U ,
because both cutoffs equal one on the respective projections of U U U . The source is smooth and compactly supported in the leaf variable. The derivative-transfer estimate (107) , together with the smooth bounds for C θ C_\theta C θ and C θ − 1 C_\theta^{-1} C θ − 1 , gives the asserted H 2 ( U ) H^2(U) H 2 ( U ) bound. Finally, the primitive acts separately on each leaf, so the source and a 1 a_1 a 1 retain compact transverse support and admit smooth zero extension. This completes the proof. □
By Lemma 6.3 , the first correction can be constructed for each fixed leading amplitude. For the uniform realization of the matrix moment identities, we also need bounds uniform over the prescribed material class and compact parameter charts. The following lemma establishes these bounds for the first correction and for the single remainder in the CGO construction (152) .
Lemma 6.4. Assume the extended Lamé coefficients satisfy (156) . Fix the two leaf disks, transverse interval I ⊥ I_\perp I ⊥ , and source cylinder Q s r c Q_{\mathrm{src}} Q src from Lemma 6.3 . Let J ⋐ I ⊥ J\Subset I_\perp J ⋐ I ⊥ be a compact transverse set contained in one of the uniform parameter charts. Choose the transverse cutoff from Lemma 6.3 to equal one near π ⊥ ( U ‾ ) ∪ J \pi_\perp(\overline U)\cup J π ⊥ ( U ) ∪ J . Let a 0 a_0 a 0 be the leading amplitude in (152) , already localized by the transverse cutoff of Lemma 6.3 , and assume
T θ a 0 = 0 on a neighborhood of U ‾ , supp x ⊥ a 0 ⊂ J . T_\theta a_0=0\quad\text{on a neighborhood of }\overline U,\qquad\operatorname{supp}_{x_\perp}a_0\subset J. T θ a 0 = 0 on a neighborhood of U , supp x ⊥ a 0 ⊂ J . With this choice of cutoff, a ~ 0 = a 0 \widetilde a_0=a_0 a 0 = a 0 for the present amplitude family. Set
N 0 : = ∥ a 0 ∥ C 4 ( Q s r c ) . \mathcal{N}_0:=\lVert a_0\rVert_{C^4(Q_{\mathrm{src}})}. N 0 := ∥ a 0 ∥ C 4 ( Q src ) . Suppose either
∥ W ∥ C 3 ( Ω ‾ 0 ) ≤ M 3 , \lVert W\rVert_{C^3(\overline{\Omega}_0)}\le M_3, ∥ W ∥ C 3 ( Ω 0 ) ≤ M 3 , or the scalar component of a 0 a_0 a 0 vanishes identically on Q s r c Q_{\mathrm{src}} Q src and
∥ W ∥ C 2 ( Ω ‾ 0 ) ≤ M 2 . \lVert W\rVert_{C^2(\overline{\Omega}_0)}\le M_2. ∥ W ∥ C 2 ( Ω 0 ) ≤ M 2 . Then, for either phase sign s ∈ { + 1 , − 1 } s\in\{+1,-1\} s ∈ { + 1 , − 1 } , the first correction a 1 a_1 a 1 and the exact CGO solution furnished by Proposition 5.8 can be chosen so that
M W z h , s = 0 in Ω C G O , (161) M^W z_{h,s}=0\quad\text{in }\Omega_{\mathrm{CGO}}, \tag*{(161)} M W z h , s = 0 in Ω CGO , (161) and
∥ a 1 ∥ H 2 ( U ) + ∥ χ 0 M W a 1 ∥ L 2 ( Ω 0 ) ≤ C N 0 , \lVert a_1\rVert_{H^2(U)}+\lVert\chi_0M^W a_1\rVert_{L^2(\Omega_0)}\le C\mathcal{N}_0, ∥ a 1 ∥ H 2 ( U ) + ∥ χ 0 M W a 1 ∥ L 2 ( Ω 0 ) ≤ C N 0 , h − 1 ∥ r h , s ∥ L 2 ( Ω 0 ) + ∥ r h , s ∥ H 1 ( Ω 0 ) ≤ C h N 0 . h^{-1}\lVert r_{h,s}\rVert_{L^2(\Omega_0)}+\lVert r_{h,s}\rVert_{H^1(\Omega_0)}\le Ch\mathcal{N}_0. h − 1 ∥ r h , s ∥ L 2 ( Ω 0 ) + ∥ r h , s ∥ H 1 ( Ω 0 ) ≤ C h N 0 . The remainder r h , s r_{h,s} r h , s is the same in both estimates. The constant C C C and the admissible upper bound for h h h depend only on the prescribed material bounds, fixed domains, cutoffs, compact parameter chart, and transverse support, not on h h h or on a 0 a_0 a 0 beyond the displayed C 4 C^4 C 4 norm. In the second case, the dependence on the potential is through M 2 M_2 M 2 .
Proof. Let χ s p \chi_{\mathrm{sp}} χ sp be the fixed spatial cutoff in (112) , and define
W c ( x ) = { χ s p ( x ) W ( x ) , x ∈ Ω 0 , 0 , x ∉ Ω 0 . (162) W^{\mathrm{c}}(x)=
\begin{cases}
\chi_{\mathrm{sp}}(x)W(x), & x\in\Omega_0,\\
0, & x\notin\Omega_0.
\end{cases} \tag*{(162)} W c ( x ) = { χ sp ( x ) W ( x ) , 0 , x ∈ Ω 0 , x ∈ / Ω 0 . (162)
Since χ s p ∈ C c ∞ ( Ω 0 ) \chi_{\mathrm{sp}}\in C^\infty_{\mathrm{c}}(\Omega_0) χ sp ∈ C c ∞ ( Ω 0 ) , this gives a smooth compactly supported extension and W c = W W^{\mathrm{c}}=W W c = W near Ω ‾ 1 \overline{\Omega}_1 Ω 1 , ∥ W c ∥ C k ( R 3 ) ≤ C k ∥ W ∥ C k ( Ω ‾ 0 ) \lVert W^{\mathrm{c}}\rVert_{C^k(\mathbb{R}^3)}\leq C_k\lVert W\rVert_{C^k(\overline{\Omega}_0)} ∥ W c ∥ C k ( R 3 ) ≤ C k ∥ W ∥ C k ( Ω 0 ) , k = 2 , 3 k=2,3 k = 2 , 3 .
Write M W c M^{W^{\mathrm{c}}} M W c for the operator obtained by replacing W W W with W c W^{\mathrm{c}} W c in (63) . Let C θ C_\theta C θ be the uniformly controlled frame supplied by Corollary 5.10 . With the leaf cutoff fixed in Lemma 6.3 , set
a 1 = 5 2 C θ C θ ( χ l e a f C θ − 1 M W c a 0 ) . a_1=\frac{5}{2}C_\theta C_\theta\left(\chi_{\mathrm{leaf}}C_\theta^{-1}M^{W^{\mathrm{c}}}a_0\right). a 1 = 2 5 C θ C θ ( χ leaf C θ − 1 M W c a 0 ) .
Here a 0 a_0 a 0 already includes the transverse cutoff, so M W c M^{W^{\mathrm{c}}} M W c acts on the full localized amplitude and retains all cutoff-derivative terms. The product rule and the Cauchy right-inverse identity give
T θ a 1 = s M W c a 0 = s M W a 0 on U , T_\theta a_1=\mathfrak{s}M^{W^{\mathrm{c}}}a_0=\mathfrak{s}M^W a_0\quad\text{on }U, T θ a 1 = s M W c a 0 = s M W a 0 on U ,
since W c = W W^{\mathrm{c}}=W W c = W near Ω ‾ 1 \overline{\Omega}_1 Ω 1 .
The explicit operator formula (63) gives
∥ M W c a 0 ∥ C 2 ( Q s r c ) ≤ C ( 1 + ∥ W c ∥ C 3 ( R 3 ) ) N 0 . (163) \lVert M^{W^{\mathrm{c}}}a_0\rVert_{C^2(Q_{\mathrm{src}})}
\leq C\left(1+\lVert W^{\mathrm{c}}\rVert_{C^3(\mathbb{R}^3)}\right)\mathcal{N}_0. \tag*{(163)} ∥ M W c a 0 ∥ C 2 ( Q src ) ≤ C ( 1 + ∥ W c ∥ C 3 ( R 3 ) ) N 0 . (163)
Indeed, two derivatives of the Laplacian term require four derivatives of a 0 a_0 a 0 ; differentiating the first-order and static zeroth-order coefficients twice uses at most four derivatives of the Lamé coefficients; and two derivatives of ∇ W c \nabla W^{\mathrm{c}} ∇ W c require three derivatives of W c W^{\mathrm{c}} W c . If a 0 = ( v 0 , 0 ) T a_0=(v_0,0)^T a 0 = ( v 0 , 0 ) T throughout Q s r c Q_{\mathrm{src}} Q src , then
( M W c a 0 ) ( x ) = ( ( W c / μ ) v 0 0 ) , \left(M^{W^{\mathrm{c}}}a_0\right)(x)=
\begin{pmatrix}
(W^{\mathrm{c}}/\mu)v_0\\
0
\end{pmatrix}, ( M W c a 0 ) ( x ) = ( ( W c / μ ) v 0 0 ) ,
so the potential contribution requires only a C 2 C^2 C 2 bound on W c W^{\mathrm{c}} W c .
The uniform Cauchy estimate (107) , including transverse derivatives, and the uniform frame bounds imply
∥ a 1 ∥ H 2 ( U ) ≤ C N 0 . \lVert a_1\rVert_{H^2(U)}\leq C\mathcal{N}_0. ∥ a 1 ∥ H 2 ( U ) ≤ C N 0 .
Since supp χ 0 ⋐ U \operatorname{supp}\chi_0\Subset U supp χ 0 ⋐ U and the coefficients of M W M^W M W are uniformly bounded under either potential assumption,
∥ χ 0 M W a 1 ∥ L 2 ( Ω 0 ) ≤ C ∥ a 1 ∥ H 2 ( U ) ≤ C N 0 . \lVert\chi_0M^W a_1\rVert_{L^2(\Omega_0)}
\leq C\lVert a_1\rVert_{H^2(U)}
\leq C\mathcal{N}_0. ∥ χ 0 M W a 1 ∥ L 2 ( Ω 0 ) ≤ C ∥ a 1 ∥ H 2 ( U ) ≤ C N 0 .
Apply Corollary 5.9 to the source − h χ 0 M W a 1 -h\chi_0M^W a_1 − h χ 0 M W a 1 . The two-sign conjugation and residual identities (154) –(155) show that the resulting single remainder cancels the residual on Ω C G O \Omega_{\mathrm{CGO}} Ω CGO , since χ 0 = 1 \chi_0=1 χ 0 = 1 near Ω ‾ C G O \overline{\Omega}_{\mathrm{CGO}} Ω CGO . The simultaneous estimate (157) gives
h − 1 ∥ r h , s ∥ L 2 ( Ω 0 ) + ∥ r h , s ∥ H 1 ( Ω 0 ) ≤ C h ∥ χ 0 M W a 1 ∥ L 2 ( Ω 0 ) ≤ C h N 0 . h^{-1}\lVert r_{h,\mathfrak{s}}\rVert_{L^2(\Omega_0)}
+\lVert r_{h,\mathfrak{s}}\rVert_{H^1(\Omega_0)}
\leq Ch\lVert\chi_0M^W a_1\rVert_{L^2(\Omega_0)}
\leq Ch\mathcal{N}_0. h − 1 ∥ r h , s ∥ L 2 ( Ω 0 ) + ∥ r h , s ∥ H 1 ( Ω 0 ) ≤ C h ∥ χ 0 M W a 1 ∥ L 2 ( Ω 0 ) ≤ C h N 0 .
This proves (161) . □
Canonical lifts and high-frequency asymptotics
The Green–Betti identity (158) involves the dual testing field y 2 = S 2 T u 2 y_2=S_2^T u_2 y 2 = S 2 T u 2 , where u 2 u_2 u 2 is an elastic solution. To construct this field from an exact CGO solution of the reduced system, let z h − z_h^- z h − be the solution constructed in (152) , and set
u h : = T z h − , y h : = S T u h = S T T z h − . u_h:=Tz_h^-,
\qquad
y_h:=S^T u_h=S^T Tz_h^-. u h := T z h − , y h := S T u h = S T T z h − .
Thus u h u_h u h is the associated elastic displacement and y h y_h y h is the corresponding dual field. This motivates the definition
K l i f t : = S T T , y h = K l i f t z h − . K_{\mathrm{lift}}:=S^T T,
\qquad
y_h=K_{\mathrm{lift}}z_h^-. K lift := S T T , y h = K lift z h − .
The operator K l i f t K_{\mathrm{lift}} K lift contains second derivatives, whereas the remainder estimates in (153) are stated in L 2 L^2 L 2 and H 1 H^1 H 1 . The following lemma rewrites it as K l i f t = K ^ l i f t W − E l i f t M W K_{\mathrm{lift}}=\widehat{K}_{\mathrm{lift}}^W-E_{\mathrm{lift}}M^W K lift = K lift W − E lift M W . On exact solutions of the reduced system, the term involving M W M^W M W vanishes, and the first-order operator K ^ l i f t W \widehat{K}_{\mathrm{lift}}^W K lift W yields the required high-frequency asymptotics of y h y_h y h .
Lemma 6.5. With K l i f t K_{\mathrm{lift}} K lift defined above, let e 4 = ( 0 , 0 , 0 , 1 ) T e_4=(0,0,0,1)^T e 4 = ( 0 , 0 , 0 , 1 ) T . For z = ( v , f ) T z=(v,f)^T z = ( v , f ) T , define
K l i f t ( v , f ) = ( v + c ∇ f + c f m c div v − 1 2 c m ⋅ v + a ( Δ f + q μ f ) 1 − κ ) , (164) K_{\mathrm{lift}}(v,f)=\begin{pmatrix} v+c\nabla f+cf\mathbf{m}\\ \dfrac{c\operatorname{div}v-\frac{1}{2}c\mathbf{m}\cdot v+a(\Delta f+q_\mu f)}{1-\kappa} \end{pmatrix}, \tag*{(164)} K lift ( v , f ) = v + c ∇ f + c f m 1 − κ c div v − 2 1 c m ⋅ v + a ( Δ f + q μ f ) , (164) K ^ l i f t W ( v , f ) = ( v + c ∇ f + c f m − c div v + 3 2 c m ⋅ v + W μ − 2 f ) , E l i f t = e 4 e 4 T μ ( 1 − κ ) , \widehat{K}_{\mathrm{lift}}^W(v,f)=\begin{pmatrix} v+c\nabla f+cf\mathbf{m}\\ -c\operatorname{div}v+\frac{3}{2}c\mathbf{m}\cdot v+W\mu^{-2}f \end{pmatrix},\qquad E_{\mathrm{lift}}=\frac{e_4e_4^T}{\mu(1-\kappa)}, K lift W ( v , f ) = ( v + c ∇ f + c f m − c div v + 2 3 c m ⋅ v + W μ − 2 f ) , E lift = μ ( 1 − κ ) e 4 e 4 T , K l i f t = K ^ l i f t W − E l i f t M W , (165) K_{\mathrm{lift}}=\widehat{K}_{\mathrm{lift}}^W-E_{\mathrm{lift}}M^W, \tag*{(165)} K lift = K lift W − E lift M W , (165) N θ ( v , f ) = ( − c f θ , c θ ⋅ v ) T , J θ = − N θ − E l i f t T θ , (166) N_\theta(v,f)=(-cf\theta,c\theta\cdot v)^T,\qquad J_\theta=-N_\theta-E_{\mathrm{lift}}T_\theta, \tag*{(166)} N θ ( v , f ) = ( − c f θ , c θ ⋅ v ) T , J θ = − N θ − E lift T θ , (166) J θ ( v , f ) = ( c f θ , − c θ ⋅ v + 2 a D θ f 1 − κ ) T . J_\theta(v,f)=\left(cf\theta,-\frac{c\theta\cdot v+2aD_\theta f}{1-\kappa}\right)^T. J θ ( v , f ) = ( c f θ , − 1 − κ c θ ⋅ v + 2 a D θ f ) T . For every exact solution z z z satisfying M W z = 0 M^Wz=0 M W z = 0 in Ω C G O \Omega_{\mathrm{CGO}} Ω CGO ,
K l i f t z = K ^ l i f t W z in Ω C G O . (167) K_{\mathrm{lift}}z=\widehat{K}_{\mathrm{lift}}^Wz\quad\text{in }\Omega_{\mathrm{CGO}}. \tag*{(167)} K lift z = K lift W z in Ω CGO . (167) Moreover, the conjugated lift satisfies
e θ ⋅ x / h K l i f t e − θ ⋅ x / h = K l i f t − h − 1 J θ . (168) e^{\theta\cdot x/h}K_{\mathrm{lift}}e^{-\theta\cdot x/h}=K_{\mathrm{lift}}-h^{-1}J_\theta. \tag*{(168)} e θ ⋅ x / h K lift e − θ ⋅ x / h = K lift − h − 1 J θ . (168) For amplitudes a 0 , a 1 a_0,a_1 a 0 , a 1 satisfying (151) with s = + 1 s=+1 s = + 1 ,
− N θ a 0 = J θ a 0 , K ^ l i f t W a 0 + N θ a 1 = K l i f t a 0 − J θ a 1 . (169) -N_\theta a_0=J_\theta a_0,\qquad\widehat{K}_{\mathrm{lift}}^W a_0+N_\theta a_1=K_{\mathrm{lift}}a_0-J_\theta a_1. \tag*{(169)} − N θ a 0 = J θ a 0 , K lift W a 0 + N θ a 1 = K lift a 0 − J θ a 1 . (169) Proof. The reduction maps T T T , S S S and the formal bilinear transpose S T S^T S T are given in (62) and Lemma 3.1. Differentiating T ( v , f ) T(v,f) T ( v , f ) gives
div T ( v , f ) = c div v − 1 2 c m ⋅ v + a ( Δ f + q μ f ) . \operatorname{div}T(v,f)=c\operatorname{div}v-\frac{1}{2}c\mathbf{m}\cdot v+a(\Delta f+q_\mu f). div T ( v , f ) = c div v − 2 1 c m ⋅ v + a ( Δ f + q μ f ) . Applying S T S^T S T yields (164) .
To prove (165) , compare the fourth components of the operators. The coefficient of div v \operatorname{div}v div v in K ^ l i f t W − E l i f t M W \widehat{K}_{\mathrm{lift}}^W-E_{\mathrm{lift}}M^W K lift W − E lift M W is
− c − κ c 1 − κ = − c 1 − κ , -c-\frac{\kappa c}{1-\kappa}=-\frac{c}{1-\kappa}, − c − 1 − κ κ c = − 1 − κ c , and that of m ⋅ v \mathbf{m}\cdot v m ⋅ v is
3 2 c + ( 3 κ − 2 ) c 2 ( 1 − κ ) = c 2 ( 1 − κ ) . \frac{3}{2}c+\frac{(3\kappa-2)c}{2(1-\kappa)}=\frac{c}{2(1-\kappa)}. 2 3 c + 2 ( 1 − κ ) ( 3 κ − 2 ) c = 2 ( 1 − κ ) c . The terms involving W μ − 2 f W\mu^{-2}f W μ − 2 f cancel. The remaining coefficients agree with (164) , proving the operator identity (165) . Equation (167) then follows from M W z = 0 M^Wz=0 M W z = 0 .
Direct differentiation of the first-order operator gives
e θ ⋅ x / h K ^ l i f t W e − θ ⋅ x / h = K ^ l i f t W + h − 1 N θ . (170) e^{\theta\cdot x/h}\widehat{K}_{\mathrm{lift}}^We^{-\theta\cdot x/h}=\widehat{K}_{\mathrm{lift}}^W+h^{-1}N_\theta. \tag*{(170)} e θ ⋅ x / h K lift W e − θ ⋅ x / h = K lift W + h − 1 N θ . (170) The conjugation identity for M W M^W M W is (154) with s = + 1 s=+1 s = + 1 . Consequently, using (165) ,
e θ ⋅ x / h K l i f t e − θ ⋅ x / h = K ^ l i f t W + h − 1 N θ − E l i f t ( M W − h − 1 T θ ) e^{\theta\cdot x/h}K_{\mathrm{lift}}e^{-\theta\cdot x/h}
=\widehat{K}_{\mathrm{lift}}^W+h^{-1}N_\theta-E_{\mathrm{lift}}(M^W-h^{-1}T_\theta) e θ ⋅ x / h K lift e − θ ⋅ x / h = K lift W + h − 1 N θ − E lift ( M W − h − 1 T θ ) = K l i f t + h − 1 ( N θ + E l i f t T θ ) =K_{\mathrm{lift}}+h^{-1}(N_\theta+E_{\mathrm{lift}}T_\theta) = K lift + h − 1 ( N θ + E lift T θ ) = K l i f t − h − 1 J θ . =K_{\mathrm{lift}}-h^{-1}J_\theta. = K lift − h − 1 J θ .
which proves (168) .
Finally, the transport conditions (151) with s = + 1 \mathfrak{s}=+1 s = + 1 give
T θ a 0 = 0 , T θ a 1 = M W a 0 . T_\theta a_0=0,\qquad T_\theta a_1=M^W a_0. T θ a 0 = 0 , T θ a 1 = M W a 0 .
By the definition of J θ J_\theta J θ ,
J θ a 0 = − N θ a 0 − E l i f t T θ a 0 = − N θ a 0 . J_\theta a_0=-N_\theta a_0-E_{\mathrm{lift}}T_\theta a_0=-N_\theta a_0. J θ a 0 = − N θ a 0 − E lift T θ a 0 = − N θ a 0 .
Similarly, (165) gives
K l i f t a 0 − J θ a 1 = K ^ l i f t W a 0 − E l i f t M W a 0 + N θ a 1 + E l i f t T θ a 1 = K ^ l i f t W a 0 + N θ a 1 . \begin{aligned}
K_{\mathrm{lift}}a_0-J_\theta a_1
&=\widehat{K}_{\mathrm{lift}}^W a_0-E_{\mathrm{lift}}M^W a_0+N_\theta a_1+E_{\mathrm{lift}}T_\theta a_1\\
&=\widehat{K}_{\mathrm{lift}}^W a_0+N_\theta a_1.
\end{aligned} K lift a 0 − J θ a 1 = K lift W a 0 − E lift M W a 0 + N θ a 1 + E lift T θ a 1 = K lift W a 0 + N θ a 1 .
This proves (169) .
The exact-solution identity (167) , the conjugation formula (170) , and the amplitude identities (169) from Lemma 6.5, together with the remainder estimates (153) , yield the high-frequency asymptotics of y h = K l i f t z ~ h y_h=K_{\mathrm{lift}}\widetilde{z}_h y h = K lift z h .
Lemma 6.6. Let a 0 , a 1 a_0,a_1 a 0 , a 1 satisfy (151) with s = + 1 \mathfrak{s}=+1 s = + 1 , and let z ~ h \widetilde{z}_h z h be the exact CGO solution furnished by Proposition 5.8, with remainder r h r_h r h . Then the scaled lift satisfies
e θ ⋅ x / h ( − h K l i f t z ~ h ) = J θ a 0 + O L 2 ( Ω C G O ) ( h ) . (171) e^{\theta\cdot x/h}(-hK_{\mathrm{lift}}\widetilde{z}_h)=J_\theta a_0+O_{L^2(\Omega_{\mathrm{CGO}})}(h). \tag*{(171)} e θ ⋅ x / h ( − h K lift z h ) = J θ a 0 + O L 2 ( Ω CGO ) ( h ) . (171) If J θ a 0 = 0 J_\theta a_0=0 J θ a 0 = 0 , then the unscaled lift satisfies
e θ ⋅ x / h K l i f t z ~ h = K l i f t a 0 − J θ a 1 + O L 2 ( Ω C G O ) ( h ) . (172) e^{\theta\cdot x/h}K_{\mathrm{lift}}\widetilde{z}_h=K_{\mathrm{lift}}a_0-J_\theta a_1+O_{L^2(\Omega_{\mathrm{CGO}})}(h). \tag*{(172)} e θ ⋅ x / h K lift z h = K lift a 0 − J θ a 1 + O L 2 ( Ω CGO ) ( h ) . (172) Both estimates hold for the same remainder r h r_h r h .
Proof. By (167) and the first-order conjugation identity (170) ,
e θ ⋅ x / h K l i f t z ~ h = h − 1 N θ a 0 + K ^ l i f t W a 0 + N θ a 1 + h K ^ l i f t W a 1 + K ^ l i f t W r h + h − 1 N θ r h . e^{\theta\cdot x/h}K_{\mathrm{lift}}\widetilde{z}_h
=h^{-1}N_\theta a_0+\widehat{K}_{\mathrm{lift}}^W a_0+N_\theta a_1
+h\widehat{K}_{\mathrm{lift}}^W a_1+\widehat{K}_{\mathrm{lift}}^W r_h+h^{-1}N_\theta r_h. e θ ⋅ x / h K lift z h = h − 1 N θ a 0 + K lift W a 0 + N θ a 1 + h K lift W a 1 + K lift W r h + h − 1 N θ r h . The amplitude identities (169) give
− N θ a 0 = J θ a 0 , K ^ l i f t W a 0 + N θ a 1 = K l i f t a 0 − J θ a 1 . -N_\theta a_0=J_\theta a_0,\qquad\widehat{K}_{\mathrm{lift}}^W a_0+N_\theta a_1=K_{\mathrm{lift}}a_0-J_\theta a_1. − N θ a 0 = J θ a 0 , K lift W a 0 + N θ a 1 = K lift a 0 − J θ a 1 . Consequently,
e θ ⋅ x / h K l i f t z ~ h = − h − 1 J θ a 0 + K l i f t a 0 − J θ a 1 + h K ^ l i f t W a 1 + K ^ l i f t W r h + h − 1 N θ r h . e^{\theta\cdot x/h}K_{\mathrm{lift}}\widetilde{z}_h
=-h^{-1}J_\theta a_0+K_{\mathrm{lift}}a_0-J_\theta a_1
+h\widehat{K}_{\mathrm{lift}}^W a_1+\widehat{K}_{\mathrm{lift}}^W r_h+h^{-1}N_\theta r_h. e θ ⋅ x / h K lift z h = − h − 1 J θ a 0 + K lift a 0 − J θ a 1 + h K lift W a 1 + K lift W r h + h − 1 N θ r h . The remainder estimates (153) and the fact that K ^ l i f t W \widehat{K}_{\mathrm{lift}}^W K lift W is first order imply
∥ h K ^ l i f t W a 1 ∥ L 2 = O ( h ) , \left\|h\widehat{K}_{\mathrm{lift}}^W a_1\right\|_{L^2}=O(h), h K lift W a 1 L 2 = O ( h ) , ∥ K ^ l i f t W r h ∥ L 2 ≤ C ∥ r h ∥ H 1 = O ( h ) , \left\|\widehat{K}_{\mathrm{lift}}^W r_h\right\|_{L^2}\leq C\left\|r_h\right\|_{H^1}=O(h), K lift W r h L 2 ≤ C ∥ r h ∥ H 1 = O ( h ) , ∥ h − 1 N θ r h ∥ L 2 ≤ C h − 1 ∥ r h ∥ L 2 = O ( h ) . \left\|h^{-1}N_\theta r_h\right\|_{L^2}\leq Ch^{-1}\left\|r_h\right\|_{L^2}=O(h). h − 1 N θ r h L 2 ≤ C h − 1 ∥ r h ∥ L 2 = O ( h ) . Thus
e θ ⋅ x / h K l i f t z ~ h = − h − 1 J θ a 0 + K l i f t a 0 − J θ a 1 + O L 2 ( Ω C G O ) ( h ) . e^{\theta\cdot x/h}K_{\mathrm{lift}}\widetilde{z}_h
=-h^{-1}J_\theta a_0+K_{\mathrm{lift}}a_0-J_\theta a_1+O_{L^2(\Omega_{\mathrm{CGO}})}(h). e θ ⋅ x / h K lift z h = − h − 1 J θ a 0 + K lift a 0 − J θ a 1 + O L 2 ( Ω CGO ) ( h ) . Multiplication by − h -h − h proves (171) . If J θ a 0 = 0 J_\theta a_0=0 J θ a 0 = 0 , the leading h − 1 h^{-1} h − 1 term vanishes, which gives (172) . Both formulas use the same exact solution and hence the same remainder r h r_h r h . □
Construction of the three polarization columns
We construct the first two leading polarization columns from the scaled physical lifts and then construct a third column from an unscaled lift. The latter is chosen so that its leading amplitude lies in the kernel of J θ J_\theta J θ , allowing the second asymptotic formula in (172) to be used.
Lemma 6.7. Consider the two-component system
D θ f = − 1 2 μ κ R , D θ R = f D θ 2 a , a = μ − 1 . (173) D_\theta f=-\frac{1}{2}\mu\kappa R,\qquad D_\theta R=fD_\theta^2a,\qquad a=\mu^{-1}. \tag*{(173)} D θ f = − 2 1 μ κ R , D θ R = f D θ 2 a , a = μ − 1 . (173) By the first assertion of Theorem 4.1 and Proposition 4.10 , there exists a smooth invertible matrix
Φ = ( f 1 f 2 R 1 R 2 ) \Phi=\begin{pmatrix}f_1&f_2\\R_1&R_2\end{pmatrix} Φ = ( f 1 R 1 f 2 R 2 ) satisfying
D θ Φ = ( 0 − 1 2 μ κ D θ 2 a 0 ) Φ , det Φ = 1. D_\theta\Phi=
\begin{pmatrix}
0&-\frac{1}{2}\mu\kappa\\
D_\theta^2a&0
\end{pmatrix}\Phi,\qquad\det\Phi=1. D θ Φ = ( 0 D θ 2 a − 2 1 μ κ 0 ) Φ , det Φ = 1. The normalization det Φ = 1 \det\Phi=1 det Φ = 1 is obtained in the proof.
For j = 1 , 2 j=1,2 j = 1 , 2 , define
Y j F = ( c f j θ − R j ) . (174) Y_j^F=\begin{pmatrix}cf_j\theta\\-R_j\end{pmatrix}. \tag*{(174)} Y j F = ( c f j θ − R j ) . (174) Then
( 2 D θ − ( A θ p h y s ) T ) Y j F = 0. (2D_\theta-(A_\theta^{\mathrm{phys}})^T)Y_j^F=0. ( 2 D θ − ( A θ phys ) T ) Y j F = 0. Moreover, there exist smooth amplitudes A 0 , j F A_{0,j}^F A 0 , j F and A 1 , j F A_{1,j}^F A 1 , j F satisfying
T θ A 0 , j F = 0 , J θ A 0 , j F = Y j F , T θ A 1 , j F = M W A 0 , j F in U . T_\theta A_{0,j}^F=0,\qquad J_\theta A_{0,j}^F=Y_j^F,\qquad T_\theta A_{1,j}^F=M^W A_{0,j}^F\quad\text{in }U. T θ A 0 , j F = 0 , J θ A 0 , j F = Y j F , T θ A 1 , j F = M W A 0 , j F in U . Let z ~ h , j \widetilde z_{h,j} z h , j be the exact CGO solution furnished by Proposition 5.8, with s = + 1 s=+1 s = + 1 and the generic amplitudes there chosen as a 0 = A 0 , j F a_0=A_{0,j}^F a 0 = A 0 , j F and a 1 = A 1 , j F a_1=A_{1,j}^F a 1 = A 1 , j F . Then
e θ ⋅ x / h ( − h K l i f t z ~ h , j ) = Y j F + O L 2 ( Ω C G O ) ( h ) , j = 1 , 2. (175) e^{\theta\cdot x/h}(-hK_{\mathrm{lift}}\widetilde z_{h,j})=Y_j^F+O_{L^2(\Omega_{\mathrm{CGO}})}(h),\qquad j=1,2. \tag*{(175)} e θ ⋅ x / h ( − h K lift z h , j ) = Y j F + O L 2 ( Ω CGO ) ( h ) , j = 1 , 2. (175)
Proof. The coefficient matrix in (173) has trace zero. Hence the frame equation gives
D θ det Φ = tr ( 0 − 1 2 μ κ D θ 2 a 0 ) det Φ = 0. D_\theta\det\Phi=\operatorname{tr}\begin{pmatrix}0&-\frac{1}{2}\mu\kappa\\D_\theta^2a&0\end{pmatrix}\det\Phi=0. D θ det Φ = tr ( 0 D θ 2 a − 2 1 μ κ 0 ) det Φ = 0.
The frame supplied by Theorem 4.1 is invertible. Thus det Φ \det\Phi det Φ is nonzero and holomorphic in the leaf variable. Multiplying its first column by ( det Φ ) − 1 (\det\Phi)^{-1} ( det Φ ) − 1 preserves the frame equation and normalizes det Φ = 1 \det\Phi=1 det Φ = 1 .
For the normalized null direction, put
a θ = θ ˉ 2 , θ ⋅ a θ = 1. a_\theta=\frac{\bar\theta}{2},\qquad\theta\cdot a_\theta=1. a θ = 2 θ ˉ , θ ⋅ a θ = 1.
For j = 1 , 2 j=1,2 j = 1 , 2 , define
H j = 2 D θ ( c − 1 R j a θ ) − m c − 1 R j + F μ θ f j . H_j=2D_\theta(c^{-1}R_ja_\theta)-mc^{-1}R_j+F_\mu\theta f_j. H j = 2 D θ ( c − 1 R j a θ ) − m c − 1 R j + F μ θ f j .
Using the coefficient identity
D θ 2 a = − c 2 θ T F μ θ (176) D_\theta^2a=-\frac{c}{2}\theta^T F_{\mu}\theta \tag*{(176)} D θ 2 a = − 2 c θ T F μ θ (176)
together with D θ c = − c 2 θ ⋅ m D_\theta c=-\frac{c}{2}\theta\cdot\mathbf m D θ c = − 2 c θ ⋅ m and (173) , we obtain
θ ⋅ H j = 2 D θ ( c − 1 R j ) − c − 1 R j θ ⋅ m + f j θ T F μ θ = 2 c − 1 D θ R j + f j θ T F μ θ = 2 c − 1 f j D θ 2 a + f j θ T F μ θ = − f j θ T F μ θ + f j θ T F μ θ = 0. \begin{aligned}
\theta\cdot H_j=2D_\theta(c^{-1}R_j)-c^{-1}R_j\,\theta\cdot\mathbf m+f_j\theta^T F_\mu\theta
=2c^{-1}D_\theta R_j+f_j\theta^T F_\mu\theta\\
&= 2c^{-1}f_jD_\theta^2a+f_j\theta^TF_\mu\theta
\\
=-f_j\theta^TF_\mu\theta+f_j\theta^TF_\mu\theta
\\
=0.
\end{aligned} θ ⋅ H j = 2 D θ ( c − 1 R j ) − c − 1 R j θ ⋅ m + f j θ T F μ θ = 2 c − 1 D θ R j + f j θ T F μ θ = − f j θ T F μ θ + f j θ T F μ θ = 0. = 2 c − 1 f j D θ 2 a + f j θ T F μ θ
Using the leafwise Cauchy right inverse and cutoff fixed in Lemma 6.3 , set
w j = − 1 2 C θ ( χ l e a f H j ) , A 0 , j F = ( c − 1 R j a θ + w j f j ) . w_j=-\frac{1}{2}\mathcal{C}_\theta(\chi_{\mathrm{leaf}}H_j),\qquad
A_{0,j}^{F}=\begin{pmatrix}c^{-1}R_ja\theta+w_j\\ f_j\end{pmatrix}. w j = − 2 1 C θ ( χ leaf H j ) , A 0 , j F = ( c − 1 R j a θ + w j f j ) .
Since θ ⋅ H j = 0 \theta\cdot H_j=0 θ ⋅ H j = 0 on the source of the Cauchy integral, θ ⋅ w j = 0 \theta\cdot w_j=0 θ ⋅ w j = 0 . Moreover,
2 D θ w j = − H j 2D_\theta w_j=-H_j 2 D θ w j = − H j
on the working disk, where χ l e a f = 1 \chi_{\mathrm{leaf}}=1 χ leaf = 1 . Consequently,
θ ⋅ ( c − 1 R j a θ + w j ) = c − 1 R j . \theta\cdot(c^{-1}R_ja\theta+w_j)=c^{-1}R_j. θ ⋅ ( c − 1 R j a θ + w j ) = c − 1 R j .
The vector and scalar components of T θ A 0 , j F T_\theta A_{0,j}^{F} T θ A 0 , j F therefore reduce to
H j + 2 D θ w j = 0 , H_j+2D_\theta w_j=0, H j + 2 D θ w j = 0 ,
2 D θ f j + d c − 1 R j = − μ κ R j + κ μ R j = 0. 2D_\theta f_j+dc^{-1}R_j=-\mu\kappa R_j+\kappa\mu R_j=0. 2 D θ f j + d c − 1 R j = − μ κ R j + κ μ R j = 0.
Thus T θ A 0 , j F = 0 T_\theta A_{0,j}^{F}=0 T θ A 0 , j F = 0 .
The explicit formula for J θ J_\theta J θ in (166) and the first equation of (173) give
J θ A 0 , j F = ( c f j θ − c θ ⋅ ( c − 1 R j a θ + w j ) + 2 a D θ f j 1 − κ ) = ( c f j θ − R j − κ R j 1 − κ ) = Y j F . J_\theta A_{0,j}^{F}
=
\begin{pmatrix}
cf_j\theta\\
-\dfrac{c\theta\cdot(c^{-1}R_ja\theta+w_j)+2aD_\theta f_j}{1-\kappa}
\end{pmatrix}
=
\begin{pmatrix}
cf_j\theta\\
-\dfrac{R_j-\kappa R_j}{1-\kappa}
\end{pmatrix}
=Y_j^{F}. J θ A 0 , j F = c f j θ − 1 − κ c θ ⋅ ( c − 1 R j a θ + w j ) + 2 a D θ f j = ( c f j θ − 1 − κ R j − κ R j ) = Y j F .
It remains to verify the dual transport equation. For Y = ( p , s ) T Y=(p,s)^T Y = ( p , s ) T , the equation ( 2 D θ − ( A θ p h y s ) T ) Y = 0 (2D_\theta-(A_\theta^{\mathrm{phys}})^T)Y=0 ( 2 D θ − ( A θ phys ) T ) Y = 0 is equivalent to
2 D θ p + θ ( m ⋅ p ) − d θ s = 0 , 2 D θ s − θ T F μ p = 0. (177) 2D_\theta p+\theta(m\cdot p)-d\theta s=0,
\\
2D_\theta s-\theta^TF_\mu p=0. \tag*{(177)} 2 D θ p + θ ( m ⋅ p ) − d θ s = 0 , 2 D θ s − θ T F μ p = 0. (177)
Substituting p = c f j θ p=cf_j\theta p = c f j θ and s = − R j s=-R_j s = − R j reduces the two rows to
θ ( 2 c D θ f j + d R j ) = 0 , − 2 ( D θ R j − f j D θ 2 a ) = 0 , \theta(2cD_\theta f_j+dR_j)=0,\qquad
-2(D_\theta R_j-f_jD_\theta^2a)=0, θ ( 2 c D θ f j + d R j ) = 0 , − 2 ( D θ R j − f j D θ 2 a ) = 0 ,
respectively. Both vanish by (173) . Hence ( 2 D θ − ( A θ p h y s ) T ) Y j F = 0 (2D_\theta-(A_\theta^{\mathrm{phys}})^T)Y_j^F=0 ( 2 D θ − ( A θ phys ) T ) Y j F = 0 .
Finally, apply Lemma 6.3 with the leading amplitude a 0 = A 0 , j F a_0=A_{0,j}^{F} a 0 = A 0 , j F and s = + 1 s=+1 s = + 1 , and denote the resulting first correction by A 1 , j F A_{1,j}^{F} A 1 , j F . In particular,
T θ A 1 , j F = M W A 0 , j F in U , A 1 , j F ∈ H 2 ( U ) . T_\theta A_{1,j}^{F}=M^{W}A_{0,j}^{F}\ \text{in }U,\qquad A_{1,j}^{F}\in H^2(U). T θ A 1 , j F = M W A 0 , j F in U , A 1 , j F ∈ H 2 ( U ) .
Proposition 5.8, with a 0 = A 0 , j F a_0=A_{0,j}^{F} a 0 = A 0 , j F and a 1 = A 1 , j F a_1=A_{1,j}^{F} a 1 = A 1 , j F , then gives the exact CGO solution z h , j z_{h,j} z h , j . Applying the scaled lift formula (171) and using J θ A 0 , j F = Y j F J_\theta A_{0,j}^{F}=Y_j^F J θ A 0 , j F = Y j F proves (175) . □
The first two columns Y 1 F , Y 2 F Y_1^F,Y_2^F Y 1 F , Y 2 F are constructed in Lemma 6.7 . We now construct a third column Y S Y^S Y S by choosing a spatially constant leading amplitude A 0 S A_0^S A 0 S satisfying
T θ A 0 S = 0 , J θ A 0 S = 0. T_\theta A_0^S=0,\qquad J_\theta A_0^S=0. T θ A 0 S = 0 , J θ A 0 S = 0.
The second identity allows us to use the unscaled lift formula (172) . We will also show that the resulting leading column Y S Y^S Y S is independent of the potential W W W .
Proposition 6.8. Represent the fixed null direction θ \theta θ by a nonzero spinor ( s 0 , s 1 ) ∈ C 2 (s_0,s_1) \in\mathbb{C}^2 ( s 0 , s 1 ) ∈ C 2 :
θ = ( s 0 2 − s 1 2 i ( s 0 2 + s 1 2 ) 2 s 0 s 1 ) , Q = 1 2 ( − s 1 − s 0 i s 1 − i s 0 s 0 − s 1 ) , n s p = ( − s 1 s 0 ) . \theta=
\begin{pmatrix}
s_0^2-s_1^2 \\
i(s_0^2+s_1^2) \\
2s_0s_1
\end{pmatrix},
\qquad
Q=\frac{1}{2}
\begin{pmatrix}
-s_1 & -s_0 \\
is_1 & -is_0 \\
s_0 & -s_1
\end{pmatrix},
\qquad
n_{\mathrm{sp}}=
\begin{pmatrix}
-s_1 \\
s_0
\end{pmatrix}. θ = s 0 2 − s 1 2 i ( s 0 2 + s 1 2 ) 2 s 0 s 1 , Q = 2 1 − s 1 i s 1 s 0 − s 0 − i s 0 − s 1 , n sp = ( − s 1 s 0 ) . Then
Q n s p = − θ 2 , ran Q = θ ⊥ . Qn_{\mathrm{sp}}=-\frac{\theta}{2},
\qquad
\operatorname{ran}Q=\theta^\perp. Q n sp = − 2 θ , ran Q = θ ⊥ . Choose
ξ q = { ( s 0 − 1 , 0 ) T , s 0 ≠ 0 , ( 0 , s 1 − 1 ) T , s 1 ≠ 0 , q θ : = Q ξ q . \xi_q=
\begin{cases}
(s_0^{-1},0)^T, & s_0\ne0, \\
(0,s_1^{-1})^T, & s_1\ne0,
\end{cases}
\qquad
q_\theta:=Q\xi_q. ξ q = { ( s 0 − 1 , 0 ) T , ( 0 , s 1 − 1 ) T , s 0 = 0 , s 1 = 0 , q θ := Q ξ q . Thus
θ ⋅ q θ = 0 , det ( n s p , ξ q ) = − 1. \theta\cdot q_\theta=0,
\qquad
\det(n_{\mathrm{sp}},\xi_q)=-1. θ ⋅ q θ = 0 , det ( n sp , ξ q ) = − 1. The vector q θ q_\theta q θ is spatially constant. Define
A 0 S : = ( q θ 0 ) . A_0^S:=
\begin{pmatrix}
q_\theta\\
0
\end{pmatrix}. A 0 S := ( q θ 0 ) . Then
T θ A 0 S = 0 , J θ A 0 S = 0. T_\theta A_0^S=0,
\qquad
J_\theta A_0^S=0. T θ A 0 S = 0 , J θ A 0 S = 0. Choose a smooth first correction
A 1 , 0 S = ( v S f S ) A_{1,0}^S=
\begin{pmatrix}
v_S \\
f_S
\end{pmatrix} A 1 , 0 S = ( v S f S ) satisfying
T θ A 1 , 0 S = M A 0 S , T_\theta A_{1,0}^S=MA_0^S, T θ A 1 , 0 S = M A 0 S , where M M M is the zero-potential operator defined by M = M 0 M=M^0 M = M 0 in (63) . Set
R S : = c θ ⋅ v S . R_S:=c\theta\cdot v_S. R S := c θ ⋅ v S . For the full operator M W M^W M W , there exists a first correction A 1 , W S A_{1,W}^S A 1 , W S satisfying
T θ A 1 , W S = M W A 0 S . T_\theta A_{1,W}^S=M^W A_0^S. T θ A 1 , W S = M W A 0 S . The third column is
Y S = ( q θ − c f S θ R S + 3 2 c m ⋅ q θ ) , ( 2 D θ − ( A θ p h y s ) T ) Y S = 0. (178) Y^S=
\begin{pmatrix}
q_\theta-cf_S\theta\\
R_S+\frac{3}{2}c\,m\cdot q_\theta
\end{pmatrix},
\qquad
(2D_\theta-(A_\theta^{\mathrm{phys}})^T)Y^S=0. \tag*{(178)} Y S = ( q θ − c f S θ R S + 2 3 c m ⋅ q θ ) , ( 2 D θ − ( A θ phys ) T ) Y S = 0. (178) Moreover, for the exact CGO solution z h − z_h^- z h − constructed from A 0 S , A 1 , W S A_0^S,A_{1,W}^S A 0 S , A 1 , W S by Proposition 5.8,
e θ ⋅ x / h K l i f t z h − = Y S + O L 2 ( Ω C G O ) ( h ) . e^{\theta\cdot x/h}K_{\mathrm{lift}}z_h^-=Y^S+O_{L^2(\Omega_{\mathrm{CGO}})}(h). e θ ⋅ x / h K lift z h − = Y S + O L 2 ( Ω CGO ) ( h ) . The leading column Y S Y^S Y S is independent of W W W .
Proof. Let C C C be the invertible four-component transport frame supplied by Theorem 4.1 and Proposition 4.10 , so that
T θ C = 0 , det C ≠ 0. T_\theta C=0,
\qquad
\det C\ne0. T θ C = 0 , det C = 0. Using the leafwise Cauchy right inverse from Definition 6.1 , choose
A 1 , 0 S = 1 2 C C θ ( χ l e a f C − 1 M A 0 S ) . (179) A_{1,0}^S=\frac{1}{2}CC_\theta\left(\chi_{\mathrm{leaf}}C^{-1}MA_0^S\right). \tag*{(179)} A 1 , 0 S = 2 1 C C θ ( χ leaf C − 1 M A 0 S ) . (179)
where χ l e a f \chi_{\mathrm{leaf}} χ leaf is the cutoff fixed on a leaf disk containing the working disk. The identity D θ C θ H = H D_\theta C_\theta H=H D θ C θ H = H and the product rule give
T θ A 1 , 0 S = C D θ C θ ( χ l e a f C − 1 M A 0 S ) = M A 0 S T_\theta A^S_{1,0}=CD_\theta C_\theta\left(\chi_{\mathrm{leaf}}C^{-1}MA^S_0\right)
=MA^S_0 T θ A 1 , 0 S = C D θ C θ ( χ leaf C − 1 M A 0 S ) = M A 0 S
on the working disk.
Writing A 1 , 0 S = ( v S , f S ) T A^S_{1,0}=(v_S,f_S)^T A 1 , 0 S = ( v S , f S ) T and R S = c θ ⋅ v S R_S=c\theta\cdot v_S R S = c θ ⋅ v S , the scalar row and the contraction of the vector row with θ \theta θ give
D θ f S = − 1 2 μ κ R S − 1 2 e r e d m ⋅ q θ , (180) D_\theta f_S=-\frac{1}{2}\mu\kappa R_S-\frac{1}{2}e^{\mathrm{red}}\mathbf{m}\cdot q_\theta, \tag*{(180)} D θ f S = − 2 1 μ κ R S − 2 1 e red m ⋅ q θ , (180)
D θ R S = f S D θ 2 a + c 2 θ T ∇ V q θ . (181) D_\theta R_S=f_SD_\theta^2a+\frac{c}{2}\theta^T\nabla Vq_\theta. \tag*{(181)} D θ R S = f S D θ 2 a + 2 c θ T ∇ V q θ . (181)
In the second identity, differentiating c c c in D θ ( c θ ⋅ v S ) D_\theta(c\theta\cdot v_S) D θ ( c θ ⋅ v S ) cancels the term involving m \mathbf{m} m from the contracted vector equation.
We next account for the potential. By (63) ,
M W = M + W , W = ( ( W / μ ) I 3 μ − 3 / 2 ( 2 W m − ∇ W ) 0 ( 1 − κ ) W / μ ) . M^W=M+\mathcal{W},\qquad
\mathcal{W}=
\begin{pmatrix}
(W/\mu)I_3 & \mu^{-3/2}(2W\mathbf{m}-\nabla W)\\
0 & (1-\kappa)W/\mu
\end{pmatrix}. M W = M + W , W = ( ( W / μ ) I 3 0 μ − 3/2 ( 2 W m − ∇ W ) ( 1 − κ ) W / μ ) .
Since A 0 S = ( q θ , 0 ) T A^S_0=(q_\theta,0)^T A 0 S = ( q θ , 0 ) T , it follows that
W A 0 S = ( ( W / μ ) q θ 0 ) . \mathcal{W}A^S_0=
\begin{pmatrix}
(W/\mu)q_\theta\\
0
\end{pmatrix}. W A 0 S = ( ( W / μ ) q θ 0 ) .
Define
β = 1 2 c θ ( χ l e a f W μ ) , A 1 , W S = A 1 , 0 S + ( β q θ 0 ) . \beta=\frac{1}{2}c_\theta\left(\chi_{\mathrm{leaf}}\frac{W}{\mu}\right),\qquad
A^S_{1,W}=A^S_{1,0}+
\begin{pmatrix}
\beta q_\theta\\
0
\end{pmatrix}. β = 2 1 c θ ( χ leaf μ W ) , A 1 , W S = A 1 , 0 S + ( β q θ 0 ) .
On the working disk,
D θ β = W 2 μ . D_\theta\beta=\frac{W}{2\mu}. D θ β = 2 μ W .
Since D θ q θ = 0 D_\theta q_\theta=0 D θ q θ = 0 and θ ⋅ q θ = 0 \theta\cdot q_\theta=0 θ ⋅ q θ = 0 , the definition T θ = 2 D θ + A θ p h y s T_\theta=2D_\theta+A^{\mathrm{phys}}_\theta T θ = 2 D θ + A θ phys gives
T θ ( β q θ 0 ) = ( 2 D θ β q θ 0 ) = ( ( W / μ ) q θ 0 ) = W A 0 S . T_\theta
\begin{pmatrix}
\beta q_\theta\\
0
\end{pmatrix}
=
\begin{pmatrix}
2D_\theta\beta q_\theta\\
0
\end{pmatrix}
=
\begin{pmatrix}
(W/\mu)q_\theta\\
0
\end{pmatrix}
=\mathcal{W}A^S_0. T θ ( β q θ 0 ) = ( 2 D θ β q θ 0 ) = ( ( W / μ ) q θ 0 ) = W A 0 S .
Consequently,
T θ A 1 , W S = M A 0 S + W A 0 S = M W A 0 S . T_\theta A^S_{1,W}=MA^S_0+\mathcal{W}A^S_0=M^WA^S_0. T θ A 1 , W S = M A 0 S + W A 0 S = M W A 0 S .
Furthermore, the explicit formula for J θ J_\theta J θ in (166) and θ ⋅ q θ = 0 \theta\cdot q_\theta=0 θ ⋅ q θ = 0 imply
J θ ( β q θ 0 ) = 0 , J θ A 1 , W S = J θ A 1 , 0 S . J_\theta
\begin{pmatrix}
\beta q_\theta\\
0
\end{pmatrix}
=0,\qquad
J_\theta A^S_{1,W}=J_\theta A^S_{1,0}. J θ ( β q θ 0 ) = 0 , J θ A 1 , W S = J θ A 1 , 0 S .
We now compute the unscaled lift. Formula (164) gives
K l i f t A 0 S = ( q θ c m ⋅ q θ 2 ( 1 − κ ) ) . K_{\mathrm{lift}}A^S_0=
\begin{pmatrix}
q_\theta\\
\dfrac{c\,\mathbf{m}\cdot q_\theta}{2(1-\kappa)}
\end{pmatrix}. K lift A 0 S = ( q θ 2 ( 1 − κ ) c m ⋅ q θ ) .
By (166) ,
J θ A 1 , 0 S = ( c f S θ − R S + 2 a D θ f S 1 − κ ) . J_\theta A^S_{1,0}=
\begin{pmatrix}
c f_S\theta\\
-\dfrac{R_S+2aD_\theta f_S}{1-\kappa}
\end{pmatrix}. J θ A 1 , 0 S = c f S θ − 1 − κ R S + 2 a D θ f S .
Thus
K l i f t A 0 S − J θ A 1 , W S = K l i f t A 0 S − J θ A 1 , 0 S = ( q θ − c f S θ 1 2 c m ⋅ q θ + R S + 2 a D θ f S 1 − κ ) . K_{\mathrm{lift}}A_0^S-J_\theta A_{1,W}^S=K_{\mathrm{lift}}A_0^S-J_\theta A_{1,0}^S
=\left(
\begin{array}{c}
q_\theta-cf_S\theta\\
\dfrac{\frac{1}{2}c\,\mathbf m\cdot q_\theta+R_S+2aD_\theta f_S}{1-\kappa}
\end{array}
\right). K lift A 0 S − J θ A 1 , W S = K lift A 0 S − J θ A 1 , 0 S = q θ − c f S θ 1 − κ 2 1 c m ⋅ q θ + R S + 2 a D θ f S .
Substituting (6.22) and the coefficient identity a e r e d = ( 3 κ − 2 ) c / 2 a^{\mathrm{ered}}=(3\kappa-2)c/2 a ered = ( 3 κ − 2 ) c /2 from (60) , the last component simplifies to
1 2 c m ⋅ q θ + R S + 2 a D θ f S 1 − κ = R S + 3 2 c m ⋅ q θ . \frac{\frac{1}{2}c\,\mathbf m\cdot q_\theta+R_S+2aD_\theta f_S}{1-\kappa}
=R_S+\frac{3}{2}c\,\mathbf m\cdot q_\theta. 1 − κ 2 1 c m ⋅ q θ + R S + 2 a D θ f S = R S + 2 3 c m ⋅ q θ .
Hence
K l i f t A 0 S − J θ A 1 , W S = Y S . K_{\mathrm{lift}}A_0^S-J_\theta A_{1,W}^S=Y^S. K lift A 0 S − J θ A 1 , W S = Y S .
Since J θ A 0 S = 0 J_\theta A_0^S=0 J θ A 0 S = 0 , the unscaled lift formula (172) yields
e θ ⋅ x / h K l i f t z h = Y S + O L 2 ( Ω C G O ) ( h ) . e^{\theta\cdot x/h}K_{\mathrm{lift}}z_h=Y^S+O_{L^2(\Omega_{\mathrm{CGO}})}(h). e θ ⋅ x / h K lift z h = Y S + O L 2 ( Ω CGO ) ( h ) .
The expression for Y S Y^S Y S depends on q θ q_\theta q θ , f S f_S f S , R S R_S R S and the Lamé coefficients, so its leading term is independent of W W W .
It remains to verify the equation in (178) . Using the two rows of the dual equation (177) , substitute
p = q θ − c f S θ , s = R S + 3 2 c m ⋅ q θ . p=q_\theta-cf_S\theta,\qquad s=R_S+\frac{3}{2}c\mathbf m\cdot q_\theta. p = q θ − c f S θ , s = R S + 2 3 c m ⋅ q θ .
Because D θ q θ = 0 D_\theta q_\theta=0 D θ q θ = 0 and θ ⋅ q θ = 0 \theta\cdot q_\theta=0 θ ⋅ q θ = 0 , the vector row becomes
θ [ − 2 c D θ f S + ( 1 − 3 2 κ ) m ⋅ q θ − d R S ] . \theta\left[-2cD_\theta f_S+\left(1-\frac{3}{2}\kappa\right)\mathbf m\cdot q_\theta-dR_S\right]. θ [ − 2 c D θ f S + ( 1 − 2 3 κ ) m ⋅ q θ − d R S ] .
This vanishes by (6.22) and the coefficient identities c e = ( 3 κ − 2 ) / 2 ce=(3\kappa-2)/2 ce = ( 3 κ − 2 ) /2 and d = κ μ d=\kappa\sqrt{\mu} d = κ μ with c = μ − 1 / 2 c=\mu^{-1/2} c = μ − 1/2 . The scalar row becomes
2 D θ R S + 3 D θ ( c m ⋅ q θ ) − θ T F μ q θ + c f S θ T F μ θ . 2D_\theta R_S+3D_\theta(c\mathbf m\cdot q_\theta)-\theta^T F_\mu q_\theta+cf_S\theta^T F_\mu\theta. 2 D θ R S + 3 D θ ( c m ⋅ q θ ) − θ T F μ q θ + c f S θ T F μ θ .
Using (181) and (176) , the terms containing f S f_S f S cancel, leaving
θ T [ c V μ + 3 ( ∇ ( c m ) ) T − F μ ] q θ . \theta^T\left[cV_\mu+3(\nabla(c\mathbf m))^T-F_\mu\right]q_\theta. θ T [ c V μ + 3 ( ∇ ( c m ) ) T − F μ ] q θ .
With m = − 2 ∇ c / c \mathbf m=-2\nabla c/c m = − 2∇ c / c , direct expansion gives
F μ = − 4 ∇ 2 c − 4 c − 1 ∇ c ⊗ ∇ c + 4 ( Δ c + c − 1 ∣ ∇ c ∣ 2 ) I 3 , F_\mu=-4\nabla^2c-4c^{-1}\nabla c\otimes\nabla c+4(\Delta c+c^{-1}|\nabla c|^2)I_3, F μ = − 4 ∇ 2 c − 4 c − 1 ∇ c ⊗ ∇ c + 4 ( Δ c + c − 1 ∣∇ c ∣ 2 ) I 3 ,
c V μ = 2 ∇ 2 c − 4 c − 1 ∇ c ⊗ ∇ c + ( Δ c − 2 c − 1 ∣ ∇ c ∣ 2 ) I 3 , cV_\mu=2\nabla^2c-4c^{-1}\nabla c\otimes\nabla c+(\Delta c-2c^{-1}|\nabla c|^2)I_3, c V μ = 2 ∇ 2 c − 4 c − 1 ∇ c ⊗ ∇ c + ( Δ c − 2 c − 1 ∣∇ c ∣ 2 ) I 3 ,
3 ( ∇ ( c m ) ) T = − 6 ∇ 2 c . 3(\nabla(c\mathbf m))^T=-6\nabla^2c. 3 ( ∇ ( c m ) ) T = − 6 ∇ 2 c .
Consequently,
F μ − c V μ − 3 ( ∇ ( c m ) ) T = ( 3 Δ c + 6 c − 1 ∣ ∇ c ∣ 2 ) I 3 . (182) F_\mu-cV_\mu-3(\nabla(c\mathbf m))^T=(3\Delta c+6c^{-1}|\nabla c|^2)I_3. \tag*{(182)} F μ − c V μ − 3 ( ∇ ( c m ) ) T = ( 3Δ c + 6 c − 1 ∣∇ c ∣ 2 ) I 3 . (182)
The scalar row is therefore a scalar multiple of θ ⋅ q θ \theta\cdot q_\theta θ ⋅ q θ , which is zero. Both rows of the dual equation hold, completing the proof. □
Proposition 6.8 constructs the third column from a first correction that depends on the potential. We now compare the third columns obtained from two potentials W W W and W ~ \widetilde W W . The following lemma shows that their difference is a linear combination of the first two columns, so the determinant of the three-column matrix is unchanged.
Lemma 6.9. Fix the Lamé coefficients and the normalized direction θ \theta θ . Let A 0 S = ( q θ , 0 ) T A^{S}_{0}=(q_{\theta},0)^{T} A 0 S = ( q θ , 0 ) T be the leading amplitude from Proposition 6.8 , and let A 0 , 1 F , A 0 , 2 F A^{F}_{0,1},A^{F}_{0,2} A 0 , 1 F , A 0 , 2 F and Y 1 F , Y 2 F Y^{F}_{1},Y^{F}_{2} Y 1 F , Y 2 F be the amplitudes and columns constructed in Lemma 6.7 . Let U U U be the working set fixed in Lemma 6.3 .
Let W , W ~ W,\widetilde{W} W , W be two smooth potentials for the same Lamé coefficients. In the CGO transport conditions (151) with s = + 1 s=+1 s = + 1 , take the common leading amplitude a 0 = A 0 S a_{0}=A^{S}_{0} a 0 = A 0 S and denote the respective first corrections a 1 a_{1} a 1 by A 1 A_{1} A 1 and A ~ 1 \widetilde{A}_{1} A 1 . Thus,
T θ A 1 = M W A 0 S , T θ A ~ 1 = M W ~ A 0 S in U . T_{\theta}A_{1}=M^{W}A^{S}_{0},\qquad T_{\theta}\widetilde{A}_{1}=M^{\widetilde{W}}A^{S}_{0}\quad\text{in }U. T θ A 1 = M W A 0 S , T θ A 1 = M W A 0 S in U . Define
Y S = K l i f t A 0 S − J θ A 1 , Y ~ S = K l i f t A 0 S − J θ A ~ 1 . Y^{S}=K_{\mathrm{lift}}A^{S}_{0}-J_{\theta}A_{1},\qquad\widetilde{Y}^{S}=K_{\mathrm{lift}}A^{S}_{0}-J_{\theta}\widetilde{A}_{1}. Y S = K lift A 0 S − J θ A 1 , Y S = K lift A 0 S − J θ A 1 . Then there exists a smooth function η = ( η 1 , η 2 ) T \eta=(\eta_{1},\eta_{2})^{T} η = ( η 1 , η 2 ) T such that
Y ~ S = Y S − η 1 Y 1 F − η 2 Y 2 F , D θ η = 0 in U . (183) \widetilde{Y}^{S}=Y^{S}-\eta_{1}Y^{F}_{1}-\eta_{2}Y^{F}_{2},\qquad D_{\theta}\eta=0\quad\text{in }U. \tag*{(183)} Y S = Y S − η 1 Y 1 F − η 2 Y 2 F , D θ η = 0 in U . (183) To express these columns in three-dimensional coordinates, set U θ = diag ( Q , 1 ) ∈ C 4 × 3 U_{\theta}=\operatorname{diag}(Q,1)\in\mathbb{C}^{4\times3} U θ = diag ( Q , 1 ) ∈ C 4 × 3 , where Q Q Q is given in Proposition 6.8 , and define the 3 × 3 3\times3 3 × 3 matrices Y 2 , Y ~ 2 Y_{2},\widetilde{Y}_{2} Y 2 , Y 2 by
U θ Y 2 = ( Y 1 F , Y 2 F , Y S ) , U θ Y ~ 2 = ( Y 1 F , Y 2 F , Y ~ S ) . U_{\theta}Y_{2}=(Y^{F}_{1},Y^{F}_{2},Y^{S}),\qquad U_{\theta}\widetilde{Y}_{2}=(Y^{F}_{1},Y^{F}_{2},\widetilde{Y}^{S}). U θ Y 2 = ( Y 1 F , Y 2 F , Y S ) , U θ Y 2 = ( Y 1 F , Y 2 F , Y S ) . Then
Y ~ 2 = Y 2 ( 1 0 − η 1 0 1 − η 2 0 0 1 ) , (184) \widetilde{Y}_{2}=Y_{2}\begin{pmatrix}1&0&-\eta_{1}\\0&1&-\eta_{2}\\0&0&1\end{pmatrix}, \tag*{(184)} Y 2 = Y 2 1 0 0 0 1 0 − η 1 − η 2 1 , (184) and hence det Y ~ 2 = det Y 2 \det\widetilde{Y}_{2}=\det Y_{2} det Y 2 = det Y 2 . Conversely, every smooth η \eta η on U U U satisfying D θ η = 0 D_{\theta}\eta=0 D θ η = 0 can be realized by changing A 1 A_{1} A 1 while keeping W W W and A 0 S A^{S}_{0} A 0 S fixed.
Proof. Use the cutoff χ l e a f \chi_{\mathrm{leaf}} χ leaf fixed in Lemma 6.3 , which is compactly supported in the prescribed source disk and equals one near π θ ( U ) \pi_{\theta}(U) π θ ( U ) .
By (63) ,
M W = M + W , M W ~ = M + W ~ . M^{W}=M+W,\qquad M^{\widetilde{W}}=M+\widetilde{W}. M W = M + W , M W = M + W .
Since A 0 S = ( q θ , 0 ) T A^{S}_{0}=(q_{\theta},0)^{T} A 0 S = ( q θ , 0 ) T , the explicit potential matrices in (63) give
( M W ~ − M W ) A 0 S = ( W ~ − W μ q θ 0 ) . \left(M^{\widetilde{W}}-M^{W}\right)A^{S}_{0}
=
\begin{pmatrix}
\dfrac{\widetilde{W}-W}{\mu}q_{\theta}\\
0
\end{pmatrix}. ( M W − M W ) A 0 S = μ W − W q θ 0 .
Define
β Δ = 1 2 C θ ( χ l e a f W ~ − W μ ) , B = A ~ 1 − A 1 − ( β Δ q θ 0 ) . (185) \beta_{\Delta}=\frac{1}{2}C_{\theta}\left(\chi_{\mathrm{leaf}}\frac{\widetilde{W}-W}{\mu}\right),\qquad
B=\widetilde{A}_{1}-A_{1}-\begin{pmatrix}\beta_{\Delta}q_{\theta}\\0\end{pmatrix}. \tag*{(185)} β Δ = 2 1 C θ ( χ leaf μ W − W ) , B = A 1 − A 1 − ( β Δ q θ 0 ) . (185)
The Cauchy right-inverse identity from Definition 6.1 yields
D θ β Δ = W ~ − W 2 μ in U . D_{\theta}\beta_{\Delta}=\frac{\widetilde{W}-W}{2\mu}\quad\text{in }U. D θ β Δ = 2 μ W − W in U .
Since q θ q_{\theta} q θ is spatially constant and θ ⋅ q θ = 0 \theta\cdot q_{\theta}=0 θ ⋅ q θ = 0 ,
T θ ( β Δ q θ 0 ) = ( 2 D θ β Δ q θ 0 ) = ( W ~ − W μ q θ 0 ) = ( M W ~ − M W ) A 0 S . T_{\theta}\begin{pmatrix}\beta_{\Delta}q_{\theta}\\0\end{pmatrix}
=
\begin{pmatrix}2D_{\theta}\beta_{\Delta}\,q_{\theta}\\0\end{pmatrix}
=
\begin{pmatrix}\dfrac{\widetilde{W}-W}{\mu}q_{\theta}\\0\end{pmatrix}
=
\left(M^{\widetilde{W}}-M^{W}\right)A^{S}_{0}. T θ ( β Δ q θ 0 ) = ( 2 D θ β Δ q θ 0 ) = μ W − W q θ 0 = ( M W − M W ) A 0 S .
Subtracting the two first-correction equations therefore gives
T θ B = 0 in U . T_\theta B = 0 \quad\text{in } U. T θ B = 0 in U .
Write
B = ( v B , f B ) T , R B : = c θ ⋅ v B . B = (v_B, f_B)^T,\qquad R_B := c\theta\cdot v_B. B = ( v B , f B ) T , R B := c θ ⋅ v B .
Expanding T θ B = 0 T_\theta B = 0 T θ B = 0 into its scalar row and the contraction of its vector row with θ \theta θ , and using (176) , gives
D θ f B = − 1 2 μ κ R B , D_\theta f_B = -\frac{1}{2}\mu\kappa R_B, D θ f B = − 2 1 μ κ R B ,
D θ R B = f B D θ 2 a . D_\theta R_B = f_B D_\theta^2 a. D θ R B = f B D θ 2 a .
Thus ( f B , R B ) T (f_B, R_B)^T ( f B , R B ) T satisfies the same two-component system as the columns of Φ \Phi Φ in Lemma 6.7 .
Define
η : = Φ − 1 ( f B R B ) , ( f B R B ) = Φ η . \eta:= \Phi^{-1}\begin{pmatrix}f_B\\R_B\end{pmatrix},\qquad
\begin{pmatrix}f_B\\R_B\end{pmatrix}=\Phi\eta. η := Φ − 1 ( f B R B ) , ( f B R B ) = Φ η .
Differentiating the second identity and using the two identical system equations gives
Φ D θ η = 0. \Phi D_\theta\eta= 0. Φ D θ η = 0.
Since det Φ = 1 \det\Phi=1 det Φ = 1 , we obtain
D θ η = 0 in U . D_\theta\eta= 0 \quad\text{in } U. D θ η = 0 in U .
By the explicit formula for J θ J_\theta J θ in (166) ,
J θ B = ( c f B θ − R B + 2 a D θ f B 1 − κ ) = ( c f B θ − R B ) = Y F η . J_\theta B =
\begin{pmatrix}
cf_B\theta\\
\dfrac{-R_B+2aD_\theta f_B}{1-\kappa}
\end{pmatrix}
=
\begin{pmatrix}
cf_B\theta\\
-R_B
\end{pmatrix}
=Y^F\eta. J θ B = c f B θ 1 − κ − R B + 2 a D θ f B = ( c f B θ − R B ) = Y F η .
Here the second equality uses a = μ − 1 a=\mu^{-1} a = μ − 1 and D θ f B = − μ κ R B / 2 D_\theta f_B=-\mu\kappa R_B/2 D θ f B = − μ κ R B /2 ; the last equality uses ( f B , R B ) T = Φ η (f_B,R_B)^T=\Phi\eta ( f B , R B ) T = Φ η and the definition of Y F Y^F Y F in Lemma 6.7 .
Moreover, (166) and θ ⋅ q θ = 0 \theta\cdot q_\theta=0 θ ⋅ q θ = 0 give
J θ ( β Δ q θ 0 ) = 0. J_\theta\begin{pmatrix}\beta\Delta q_\theta\\0\end{pmatrix}=0. J θ ( β Δ q θ 0 ) = 0.
Consequently,
Y ~ S − Y S = − J θ ( A ~ 1 − A 1 ) = − J θ B = − Y F η . \begin{aligned}
\widetilde{Y}^S-Y^S=-J_\theta(\widetilde{A}_1-A_1) \\
&=-J_\theta B \\
&=-Y^F\eta.
\end{aligned} Y S − Y S = − J θ ( A 1 − A 1 ) = − J θ B = − Y F η .
This proves (183) . The matrix identity (184) follows by expressing the same column relation in the fixed dual quotient coordinates. Since the triangular matrix has determinant one, det Y ~ 2 = det Y 2 \det\widetilde{Y}_2=\det Y_2 det Y 2 = det Y 2 .
For the converse, let η \eta η be any smooth function on U U U satisfying D θ η = 0 D_\theta\eta=0 D θ η = 0 , and set
B = A 0 , 1 F η 1 + A 0 , 2 F η 2 . B=A^F_{0,1}\eta_1+A^F_{0,2}\eta_2. B = A 0 , 1 F η 1 + A 0 , 2 F η 2 .
By the product rule and T θ A 0 , j F = 0 T_\theta A^F_{0,j}=0 T θ A 0 , j F = 0 from Lemma 6.7 ,
T θ B = 2 ( D θ η 1 ) A 0 , 1 F + 2 ( D θ η 2 ) A 0 , 2 F + η 1 T θ A 0 , 1 F + η 2 T θ A 0 , 2 F = 0. \begin{aligned}
T_\theta B=2(D_\theta\eta_1)A^F_{0,1}+2(D_\theta\eta_2)A^F_{0,2}
+\eta_1T_\theta A^F_{0,1}+\eta_2T_\theta A^F_{0,2} \\
&=0.
\end{aligned} T θ B = 2 ( D θ η 1 ) A 0 , 1 F + 2 ( D θ η 2 ) A 0 , 2 F + η 1 T θ A 0 , 1 F + η 2 T θ A 0 , 2 F = 0.
Also, J θ A 0 , j F = Y j F J_\theta A^F_{0,j}=Y^F_j J θ A 0 , j F = Y j F by Lemma 6.7 , so
J θ B = Y F η . J_\theta B=Y^F\eta. J θ B = Y F η .
Replacing A 1 A_1 A 1 by A 1 + B A_1+B A 1 + B preserves T θ A 1 = M W A 0 S T_{\theta}A_1=M^{W}A^{S}_0 T θ A 1 = M W A 0 S and changes the third column to Y S − Y F η Y^{S}-Y_{F}\eta Y S − Y F η , as claimed. This correction is performed at the transport-amplitude level, before the construction of exact physical solutions. □
Lemma 6.10. Let q = q θ q=q_{\theta} q = q θ , spatially constant for a fixed direction, and let Φ \Phi Φ be the frame fixed in Lemma 6.7 . The static first correction in Proposition 6.8 can be constructed from Φ \Phi Φ and the leaf primitive.
Proof. Set
B 2 = ( 0 − μ κ / 2 D θ 2 a 0 ) , G 2 = ( − e r e d ( m ⋅ q ) / 2 c θ T V μ q / 2 ) , ( f S R S ) = Φ C θ ( χ l e a f Φ − 1 G 2 ) . B_2=\begin{pmatrix} 0 & -\mu\kappa/2 \\ D_{\theta}^{2}a & 0 \end{pmatrix},\qquad
G_2=\begin{pmatrix} -e^{\mathrm{red}}(\mathbf{m}\cdot q)/2 \\ c\theta^{T}V_{\mu}q/2 \end{pmatrix},\qquad
\begin{pmatrix} f_S \\ R_S \end{pmatrix}=\Phi C_{\theta}(\chi_{\mathrm{leaf}}\Phi^{-1}G_2). B 2 = ( 0 D θ 2 a − μ κ /2 0 ) , G 2 = ( − e red ( m ⋅ q ) /2 c θ T V μ q /2 ) , ( f S R S ) = Φ C θ ( χ leaf Φ − 1 G 2 ) .
Since D θ Φ = B 2 Φ D_{\theta}\Phi=B_2\Phi D θ Φ = B 2 Φ and D θ C θ = I D_{\theta}C_{\theta}=I D θ C θ = I , the pair ( f S , R S ) (f_S,R_S) ( f S , R S ) satisfies (6.22)–(181) where χ l e a f = 1 \chi_{\mathrm{leaf}}=1 χ leaf = 1 . Put a θ = θ ˉ / 2 a_{\theta}=\bar{\theta}/2 a θ = θ ˉ /2 , v 0 = c − 1 R S a θ v_0=c^{-1}R_Sa_{\theta} v 0 = c − 1 R S a θ , and
H S = 2 D θ v 0 − m c − 1 R S + F μ θ f S − V μ q . H_S=2D_{\theta}v_0-\mathbf{m}c^{-1}R_S+F_{\mu\theta}f_S-V_{\mu}q. H S = 2 D θ v 0 − m c − 1 R S + F μ θ f S − V μ q .
The identities θ ⋅ a θ = 1 \theta\cdot a_{\theta}=1 θ ⋅ a θ = 1 , D θ c = − c θ ⋅ m / 2 D_{\theta}c=-c\theta\cdot\mathbf{m}/2 D θ c = − c θ ⋅ m /2 , and (176) imply
θ ⋅ H S = 2 c − 1 D θ R S + θ T F μ θ f S − θ T V μ q = 0. \theta\cdot H_S=2c^{-1}D_{\theta}R_S+\theta^{T}F_{\mu\theta}f_S-\theta^{T}V_{\mu}q=0. θ ⋅ H S = 2 c − 1 D θ R S + θ T F μ θ f S − θ T V μ q = 0.
Choose a smaller source cutoff χ 1 \chi_1 χ 1 within this disk, equal to one on the final working disk, and define
w S = − 1 2 C θ ( χ 1 H S ) , v S = v 0 + w S . w_S=-\frac{1}{2}C_{\theta}(\chi_1H_S),\qquad v_S=v_0+w_S. w S = − 2 1 C θ ( χ 1 H S ) , v S = v 0 + w S .
Choose the disks and cutoffs so that
D ‾ w o r k ⊂ { χ 1 = 1 } , supp χ 1 ⋐ D e f f ⊂ { χ l e a f = 1 } , \overline{D}_{\mathrm{work}}\subset\{\chi_1=1\},\qquad\operatorname{supp}\chi_1\Subset D_{\mathrm{eff}}\subset\{\chi_{\mathrm{leaf}}=1\}, D work ⊂ { χ 1 = 1 } , supp χ 1 ⋐ D eff ⊂ { χ leaf = 1 } ,
where D e f f D_{\mathrm{eff}} D eff is a disk on which the two-component equations hold. Hence θ ⋅ H S = 0 \theta\cdot H_S=0 θ ⋅ H S = 0 on the entire primitive source, so θ ⋅ w S = 0 \theta\cdot w_S=0 θ ⋅ w S = 0 and 2 D θ w S = − H S 2D_{\theta}w_S=-H_S 2 D θ w S = − H S on the working disk. It follows that c θ ⋅ v S = R S c\theta\cdot v_S=R_S c θ ⋅ v S = R S and all four rows hold:
2 D θ v S − m ( θ ⋅ v S ) + F μ θ f S = V μ q , 2 D θ f S + d ( θ ⋅ v S ) = − e r e d m ⋅ q . 2D_{\theta}v_S-\mathbf{m}(\theta\cdot v_S)+F_{\mu\theta}f_S=V_{\mu}q,\qquad
2D_{\theta}f_S+d(\theta\cdot v_S)=-e^{\mathrm{red}}\mathbf{m}\cdot q. 2 D θ v S − m ( θ ⋅ v S ) + F μ θ f S = V μ q , 2 D θ f S + d ( θ ⋅ v S ) = − e red m ⋅ q .
This proves sufficiency of the two-component equations for the four-component reconstruction. Smooth dependence on the transverse parameter and finite norm bounds follow from the fixed frame and the leaf primitive. Add the full W W W correction of Proposition 6.8 and the transverse transition correction of Lemma 6.11 , then apply the same right inverse. □
Extraction and uniform realization of the matrix moments
The three leading columns are now assembled into fixed primal and dual quotient frames. We next show that these frames can be chosen before the polynomial and transverse tests, so that the same matrix field occurs in every moment identity.
Lemma 6.11. Fix a local chart of directions and transverse parameters, together with a relatively compact open interval I I I in its transverse-parameter range. For material j j j , let A j , θ p h y s A^{\mathrm{phys}}_{j,\theta} A j , θ phys denote the matrix A θ p h y s A^{\mathrm{phys}}_{\theta} A θ phys evaluated at its coefficients. On this fixed chart, choose invertible 3 × 3 3\times3 3 × 3 quotient frames X 1 , Y 2 X_1,Y_2 X 1 , Y 2 satisfying
D θ X 1 + P 1 X 1 = 0 , D θ Y 2 − P 2 T Y 2 = 0 , det Y 2 = 2 c 2 ≠ 0. (186) D_{\theta}X_1+P_1X_1=0,\qquad D_{\theta}Y_2-P_2^{T}Y_2=0,\qquad\det Y_2=2c_2\ne0. \tag*{(186)} D θ X 1 + P 1 X 1 = 0 , D θ Y 2 − P 2 T Y 2 = 0 , det Y 2 = 2 c 2 = 0. (186) Here U θ = diag ( Q , 1 ) U_{\theta}=\operatorname{diag}(Q,1) U θ = diag ( Q , 1 ) is a 4 × 3 4\times3 4 × 3 matrix. The map U θ T : C 4 → C 3 U_{\theta}^{T}:\mathbb{C}^{4}\to\mathbb{C}^{3} U θ T : C 4 → C 3 has kernel C ( θ , 0 ) T C(\theta,0)^{T} C ( θ , 0 ) T and gives coordinates on the characteristic quotient. Choose a spatially constant right inverse V θ ∈ C 4 × 3 V_{\theta}\in\mathbb{C}^{4\times3} V θ ∈ C 4 × 3 , so that U θ T V θ = I 3 U_{\theta}^{T}V_{\theta}=I_3 U θ T V θ = I 3 , and set P j = U θ T A j , θ p h y s V θ / 2 P_j=U_{\theta}^{T}A^{\mathrm{phys}}_{j,\theta}V_{\theta}/2 P j = U θ T A j , θ phys V θ /2 . Since A j , θ p h y s ( θ , 0 ) T = 0 A^{\mathrm{phys}}_{j,\theta}(\theta,0)^{T}=0 A j , θ phys ( θ , 0 ) T = 0 , this 3 × 3 3\times3 3 × 3 quotient coefficient is independent of the choice of V θ V_{\theta} V θ . The frame X 1 X_1 X 1 is primal and Y 2 Y_2 Y 2 is dual; the transpose in the second equation of (186) expresses their bilinear duality. For every subsequently chosen
b ∈ C ∞ ( I ) b \in C^\infty(I) b ∈ C ∞ ( I ) , all three columns can be realized by elastic solutions with O L 2 ( h ) O_{L^2}(h) O L 2 ( h ) error throughout the target, and with the same leading matrix U θ Y 2 S U_\theta Y^S_2 U θ Y 2 S near supp b \operatorname{supp} b supp b .
Proof. The large-disk frame theorem first gives X 1 X_1 X 1 . Put Z 1 = V θ X 1 + e θ α Z_1=V_\theta X_1+e_\theta\alpha Z 1 = V θ X 1 + e θ α , where e θ = ( θ , 0 ) T e_\theta=(\theta,0)^{\mathsf T} e θ = ( θ , 0 ) T . Since A 1 , θ p h y s e θ = 0 A^{\mathrm{phys}}_{1,\theta}e_\theta=0 A 1 , θ phys e θ = 0 , each column of T 1 , θ ( V θ X 1 ) T_{1,\theta}(V_\theta X_1) T 1 , θ ( V θ X 1 ) has the form e θ f e_\theta f e θ f . Choosing α = − 1 2 C θ ( χ l e a f f ) \alpha=-\frac{1}{2}\mathcal{C}_\theta(\chi_{\mathrm{leaf}}f) α = − 2 1 C θ ( χ leaf f ) gives T 1 , θ Z 1 = 0 T_{1,\theta}Z_1=0 T 1 , θ Z 1 = 0 , U θ T Z 1 = X 1 U_\theta^{\mathsf T}Z_1=X_1 U θ T Z 1 = X 1 . Lemma 6.7 and Proposition 6.8 give three dual columns in the four-component space. Since Q n s p = − θ / 2 Q_{n_{\mathrm{sp}}}=-\theta/2 Q n sp = − θ /2 , their quotient coordinates are
Y 2 = ( − 2 c f 1 n s p − 2 c f 2 n s p ξ q + 2 c f S n s p − R 1 − R 2 R S + 3 2 c m ⋅ q θ ) . Y_2=\begin{pmatrix}
-2cf_1n_{\mathrm{sp}} & -2cf_2n_{\mathrm{sp}} & \xi_q+2cf_Sn_{\mathrm{sp}}\\
-R_1 & -R_2 & R_S+\frac{3}{2}\mathbf{c}_m\cdot q_\theta
\end{pmatrix}. Y 2 = ( − 2 c f 1 n sp − R 1 − 2 c f 2 n sp − R 2 ξ q + 2 c f S n sp R S + 2 3 c m ⋅ q θ ) .
Put B θ = ( n s p , ξ q ) ∈ G L ( 2 , C ) B_\theta=(n_{\mathrm{sp}},\xi_q)\in GL(2,\mathbb{C}) B θ = ( n sp , ξ q ) ∈ G L ( 2 , C ) and s S = R S + 3 c m ⋅ q θ / 2 s_S=R_S+3\mathbf{c}_m\cdot q_\theta/2 s S = R S + 3 c m ⋅ q θ /2 . The displayed matrix factors as
Y 2 = diag ( B θ , 1 ) ( − 2 c f 1 − 2 c f 2 2 c f S 0 0 1 − R 1 − R 2 s S ) . Y_2=\operatorname{diag}(B_\theta,1)
\begin{pmatrix}
-2cf_1 & -2cf_2 & 2cf_S\\
0 & 0 & 1\\
-R_1 & -R_2 & s_S
\end{pmatrix}. Y 2 = diag ( B θ , 1 ) − 2 c f 1 0 − R 1 − 2 c f 2 0 − R 2 2 c f S 1 s S .
Expanding the second determinant in its second row gives det Y 2 = − 2 c det Φ det ( n s p , ξ q ) = 2 c \det Y_2=-2c\det\Phi\det(n_{\mathrm{sp}},\xi_q)=2c det Y 2 = − 2 c det Φ det ( n sp , ξ q ) = 2 c . The determinant is independent of the values of f S f_S f S and R S R_S R S . The frame choices are made before b b b and every subsequent polynomial g g g are selected.
For a given b b b , choose χ ∈ C c ∞ ( I ) \chi\in C^\infty_c(I) χ ∈ C c ∞ ( I ) equal to one near supp b \operatorname{supp}b supp b , and use χ ( x ⊥ ) \chi(x_\perp) χ ( x ⊥ ) as a spatial cutoff. Primes on χ \chi χ below denote differentiation with respect to the real transverse variable x ⊥ x_\perp x ⊥ . The commutator [ M , χ ] = M ∘ χ − χ M [M,\chi]=M\circ\chi-\chi M [ M , χ ] = M ∘ χ − χ M records all source terms created by this cutoff. For the third column, set
A 0 , χ S = χ A 0 S , A 1 , χ S = χ A 1 , W S + w χ , T θ w χ = [ M , χ ] A 0 S . A^S_{0,\chi}=\chi A^S_0,\qquad A^S_{1,\chi}=\chi A^S_{1,W}+w_\chi,\qquad T_\theta w_\chi=[M,\chi]A^S_0. A 0 , χ S = χ A 0 S , A 1 , χ S = χ A 1 , W S + w χ , T θ w χ = [ M , χ ] A 0 S .
Since D θ χ = 0 D_\theta\chi=0 D θ χ = 0 ,
[ M , χ ] A 0 S = ( χ ′ ′ q θ − χ ′ m ( ν θ ⋅ q θ ) d χ ′ ( ν θ ⋅ q θ ) ) (187) [M,\chi]A^S_0=
\begin{pmatrix}
\chi''q_\theta-\chi'\mathbf{m}(\nu_\theta\cdot q_\theta)\\
d\chi'(\nu_\theta\cdot q_\theta)
\end{pmatrix} \tag*{(187)} [ M , χ ] A 0 S = ( χ ′′ q θ − χ ′ m ( ν θ ⋅ q θ ) d χ ′ ( ν θ ⋅ q θ ) ) (187)
Construct w χ w_\chi w χ from the fixed four-component frame and leaf primitive. The primitive preserves x ⊥ x_\perp x ⊥ , so w χ w_\chi w χ and all its spatial derivatives vanish on the open interval where χ = 1 \chi=1 χ = 1 . Moreover, [ W , χ ] = 0 [W,\chi]=0 [ W , χ ] = 0 , and these amplitudes satisfy the full transport relations. The potential correction A 1 , W S − A 1 , 0 S = ( β q θ , 0 ) T A^S_{1,W}-A^S_{1,0}=(\beta q_\theta,0)^{\mathsf T} A 1 , W S − A 1 , 0 S = ( β q θ , 0 ) T changes neither J θ A 1 , W S J_\theta A^S_{1,W} J θ A 1 , W S nor the unscaled leading physical column, since J θ ( β q θ , 0 ) T = 0 J_\theta(\beta q_\theta,0)^{\mathsf T}=0 J θ ( β q θ , 0 ) T = 0 by θ ⋅ q θ = 0 \theta\cdot q_\theta=0 θ ⋅ q θ = 0 . The transition correction w χ w_\chi w χ vanishes with its derivatives on a neighborhood of supp b \operatorname{supp}b supp b : for such transverse parameters its entire Cauchy source is zero, and the primitive acts separately on each leaf. Thus the leading physical column there is exactly the previously fixed Y S Y^S Y S . For the first two columns, take χ A 0 F \chi A^F_0 χ A 0 F and solve the first correction equation. The amplitudes and sources have compact transverse support and extend smoothly by zero. The chosen leaf disks contain all leaf sections of the working domains. Lemma 6.6 gives the error estimate throughout the target, including the transition region of χ \chi χ .
Lemma 6.12. Let I ⊂ R I\subset\mathbb{R} I ⊂ R be an interval, D ⊂ C D\subset\mathbb{C} D ⊂ C a bounded disk, and let F ∈ C ( I ; L 1 ( D ; C 3 × 3 ) ) F\in C(I;L^1(D;\mathbb{C}^{3\times3})) F ∈ C ( I ; L 1 ( D ; C 3 × 3 )) . Suppose that for every fixed g ∈ C [ z ] g\in\mathbb{C}[z] g ∈ C [ z ] and every b ∈ C c ∞ ( I ) b\in C^\infty_c(I) b ∈ C c ∞ ( I ) ,
∫ I b ( t ) ∫ D g ( z ) F ( t , z ) d A ( z ) d t = 0. (188) \int_I b(t)\int_D g(z)F(t,z)\,dA(z)\,dt=0. \tag*{(188)} ∫ I b ( t ) ∫ D g ( z ) F ( t , z ) d A ( z ) d t = 0. (188) Then ∫ D g ( z ) F ( t , z ) d A ( z ) = 0 \int_D g(z)F(t,z)\,dA(z)=0 ∫ D g ( z ) F ( t , z ) d A ( z ) = 0 for every t ∈ I t\in I t ∈ I and every g ∈ C [ z ] g\in\mathbb{C}[z] g ∈ C [ z ] . In particular, the matrix F F F is held fixed while all the polynomial tests are varied.
Proof. Fix g g g and put J g ( t ) = ∫ D g ( z ) F ( t , z ) d A ( z ) J_g(t)=\int_D g(z)F(t,z)\,dA(z) J g ( t ) = ∫ D g ( z ) F ( t , z ) d A ( z ) . The boundedness of g g g on D D D gives ∥ J g ( t ) − J g ( t ′ ) ∥ ≤ ∥ g ∥ L ∞ ( D ) ∥ F ( t , ⋅ ) − F ( t ′ , ⋅ ) ∥ L 1 ( D ) \lVert J_g(t)-J_g(t')\rVert\leq\lVert g\rVert_{L^\infty(D)}\lVert F(t,\cdot)-F(t',\cdot)\rVert_{L^1(D)} ∥ J g ( t ) − J g ( t ′ )∥ ≤ ∥ g ∥ L ∞ ( D ) ∥ F ( t , ⋅ ) − F ( t ′ , ⋅ ) ∥ L 1 ( D ) , so J g J_g J g is continuous. Applied entrywise, the fundamental lemma for distributions and (188) imply that J g = 0 J_g=0 J g = 0 almost everywhere in I I I . Continuity makes this identity pointwise. The argument applies to every polynomial using the same F F F .
Theorem 6.13. Under the notation of Definition 6.1 and the constructions in Proposition 6.2 , choose Ω ‾ ⊂ B R 0 \overline{\Omega} \subset B_{R_0} Ω ⊂ B R 0 , R 0 < R ⊥ < R 1 R_0 < R_{\perp} < R_1 R 0 < R ⊥ < R 1 , and I ⊂ ( − R ⊥ , R ⊥ ) I \subset(-R_{\perp},R_{\perp}) I ⊂ ( − R ⊥ , R ⊥ ) with I ‾ \overline{I} I contained in a preselected local transverse chart. For each fixed compact direction and transverse-parameter chart, there exist frames X 1 , Y 2 X_1,Y_2 X 1 , Y 2 supplied by Lemma 6.11 , selected independently of all subsequent polynomial and transverse tests, such that
∫ ∣ z ∣ < R 1 2 − x ⊥ 2 g ( z ) Y 2 T ( P 1 − P 2 ) X 1 d A = 0 , for every x ⊥ ∈ I and g ∈ C [ z ] . (189) \int_{|z|<\sqrt{R_1^2-x_{\perp}^2}} g(z)Y_2^{T}(P_1-P_2)X_1\,dA=0,\qquad\text{for every }x_{\perp}\in I\text{ and }g\in\mathbb{C}[z]. \tag*{(189)} ∫ ∣ z ∣ < R 1 2 − x ⊥ 2 g ( z ) Y 2 T ( P 1 − P 2 ) X 1 d A = 0 , for every x ⊥ ∈ I and g ∈ C [ z ] . (189) This is a full 3 × 3 3 \times3 3 × 3 matrix identity on each leaf, for every polynomial.
Proof. First fix X 1 , Y 2 X_1,Y_2 X 1 , Y 2 on the entire parameter chart, as in Lemma 6.11 ; neither frame is changed during this proof. Let Z 1 Z_1 Z 1 denote the four-component lift constructed in the proof of Lemma 6.11 . With U θ = diag ( Q , 1 ) U_\theta=\operatorname{diag}(Q,1) U θ = diag ( Q , 1 ) , it satisfies
T 1 , θ Z 1 = 0 , U θ T Z 1 = X 1 . T_{1,\theta}Z_1=0,\qquad U_\theta^{T}Z_1=X_1. T 1 , θ Z 1 = 0 , U θ T Z 1 = X 1 .
Now fix b ∈ C c ∞ ( I ) b\in C^\infty_c(I) b ∈ C c ∞ ( I ) and g ∈ C [ z ] g\in\mathbb{C}[z] g ∈ C [ z ] . Since b = b ( x ⊥ ) b=b(x_{\perp}) b = b ( x ⊥ ) and g = g ( z ) g=g(z) g = g ( z ) is holomorphic,
D θ b = 0 , D θ g = 0. D_\theta b=0,\qquad D_\theta g=0. D θ b = 0 , D θ g = 0.
Consequently, the positive-phase leading amplitude
satisfies
T 1 , θ A = 2 D θ ( b g ) Z 1 + b g T 1 , θ Z 1 = 0. T_{1,\theta}A=2D_\theta(bg)Z_1+bgT_{1,\theta}Z_1=0. T 1 , θ A = 2 D θ ( b g ) Z 1 + b g T 1 , θ Z 1 = 0.
Choose the first correction A 1 A_1 A 1 to satisfy
T 1 , θ A 1 = − M 1 W 1 A . T_{1,\theta}A_1=-M_1^{W_1}A. T 1 , θ A 1 = − M 1 W 1 A .
Applying the selected right inverse at − θ -\theta − θ gives
z 1 , h = e θ ⋅ x / h ( A + h A 1 + r h ) , z_{1,h}=e^{\theta\cdot x/h}(A+hA_1+r_h), z 1 , h = e θ ⋅ x / h ( A + h A 1 + r h ) ,
with the estimates in (153) . Apply the reduced Betti identity columnwise to the three columns of Z 1 Z_1 Z 1 and the three physical testing columns. The resulting nine scalar identities assemble into the full matrix identity used below. Write u 2 , h ( 1 ) , u 2 , h ( 2 ) , u 2 , h ( 3 ) u_{2,h}^{(1)},u_{2,h}^{(2)},u_{2,h}^{(3)} u 2 , h ( 1 ) , u 2 , h ( 2 ) , u 2 , h ( 3 ) for the three elastic displacements of Lemma 6.11 , with the first two rescaled by − h -h − h and the third unscaled, and set Y 2 , h = ( S 2 T u 2 , h ( 1 ) , S 2 T u 2 , h ( 2 ) , S 2 T u 2 , h ( 3 ) ) \mathcal{Y}_{2,h}=(S_2^T u_{2,h}^{(1)},S_2^T u_{2,h}^{(2)},S_2^T u_{2,h}^{(3)}) Y 2 , h = ( S 2 T u 2 , h ( 1 ) , S 2 T u 2 , h ( 2 ) , S 2 T u 2 , h ( 3 ) ) . The three negative-phase lifts then form
Y 2 , h = e − θ ⋅ x / h ( Y χ ( 0 ) + q h ) , ∥ q h ∥ 2 = O ( h ) , Y χ ( 0 ) = U θ Y 2 near supp b . \mathcal{Y}_{2,h}=e^{-\theta\cdot x/h}(\mathcal{Y}_{\chi}^{(0)}+q_h),\qquad\|q_h\|_2=O(h),\qquad\mathcal{Y}_{\chi}^{(0)}=U_\theta Y_2\text{ near }\operatorname{supp}b. Y 2 , h = e − θ ⋅ x / h ( Y χ ( 0 ) + q h ) , ∥ q h ∥ 2 = O ( h ) , Y χ ( 0 ) = U θ Y 2 near supp b .
Here Y χ ( 0 ) \mathcal{Y}_{\chi}^{(0)} Y χ ( 0 ) denotes the fixed leading matrix amplitude of the negative-phase tests. Write D W = M 1 W 1 − M 2 W 2 = ∑ k Δ A k ∂ k + Δ B W D^W=M_1^{W_1}-M_2^{W_2}=\sum_k\Delta A_k\partial_k+\Delta B^W D W = M 1 W 1 − M 2 W 2 = ∑ k Δ A k ∂ k + Δ B W , where Δ A k = A 1 , k p h y s − A 2 , k p h y s \Delta A_k=A_{1,k}^{\mathrm{phys}}-A_{2,k}^{\mathrm{phys}} Δ A k = A 1 , k phys − A 2 , k phys , Δ A θ = ∑ k θ k Δ A k \Delta A_\theta=\sum_k\theta_k\Delta A_k Δ A θ = ∑ k θ k Δ A k , and Δ B W \Delta B^W Δ B W is the difference of the zeroth-order matrices in (63) . The notation D W D^W D W denotes this difference operator. The factor h h h in the following expansion isolates the first-order coefficient difference in the limit: Throughout the target, the exact expansion is
h e − θ ⋅ x / h D W z 1 , h = Δ A θ A + h ( Δ A θ A 1 + D W A ) + Δ A θ r h + h 2 D W A 1 + h D W r h = Δ A θ A + e h , ∥ e h ∥ 2 = O ( h ) . he^{-\theta\cdot x/h}D^Wz_{1,h}=\Delta A_\theta A+h(\Delta A_\theta A_1+D^W A)+\Delta A_\theta r_h+h^2D^W A_1+hD^W r_h
=\Delta A_\theta A+e_h,\qquad\|e_h\|_2=O(h). h e − θ ⋅ x / h D W z 1 , h = Δ A θ A + h ( Δ A θ A 1 + D W A ) + Δ A θ r h + h 2 D W A 1 + h D W r h = Δ A θ A + e h , ∥ e h ∥ 2 = O ( h ) .
In particular, ∥ Δ A θ r h ∥ 2 = O ( h 2 ) \|\Delta A_\theta r_h\|_2=O(h^2) ∥Δ A θ r h ∥ 2 = O ( h 2 ) and ∥ h D W r h ∥ 2 = O ( h 2 ) \|hD^W r_h\|_2=O(h^2) ∥ h D W r h ∥ 2 = O ( h 2 ) . The potential and potential-gradient differences occur in the zeroth-order term Δ B W \Delta B^W Δ B W , carrying the factor h h h . The phases cancel, and Cauchy–Schwarz gives
∥ h Y 2 , h T D W z 1 , h − ( Y χ ( 0 ) ) T Δ A θ A ∥ 1 ≤ ∥ q h ∥ 2 ∥ Δ A θ A ∥ 2 + ( ∥ Y χ ( 0 ) ∥ 2 + ∥ q h ∥ 2 ) ∥ e h ∥ 2 = O ( h ) . \left\|h\mathcal{Y}_{2,h}^{T}D^Wz_{1,h}-(\mathcal{Y}_{\chi}^{(0)})^{T}\Delta A_\theta A\right\|_1
\leq\|q_h\|_2\|\Delta A_\theta A\|_2+\left(\|\mathcal{Y}_{\chi}^{(0)}\|_2+\|q_h\|_2\right)\|e_h\|_2
=O(h). h Y 2 , h T D W z 1 , h − ( Y χ ( 0 ) ) T Δ A θ A 1 ≤ ∥ q h ∥ 2 ∥Δ A θ A ∥ 2 + ( ∥ Y χ ( 0 ) ∥ 2 + ∥ q h ∥ 2 ) ∥ e h ∥ 2 = O ( h ) .
The first two elastic solutions are multiplied by the constant − h -h − h ; the third is unscaled. Thus every column satisfies the original-domain identity (158) . Since the coefficient differences vanish outside Ω \Omega Ω , extend the integral to the working domain. Passing to the limit gives
∫ b g Y 2 T U θ T Δ A θ Z 1 d x = 0. \int bgY_2^T U_\theta^T \Delta A_\theta Z_1\,dx=0. ∫ b g Y 2 T U θ T Δ A θ Z 1 d x = 0.
Using U θ T Z 1 = X 1 U_\theta^T Z_1=X_1 U θ T Z 1 = X 1 and A j , θ p h y s e θ = 0 A_{j,\theta}^{\mathrm{phys}}e_\theta=0 A j , θ phys e θ = 0 , U θ T Δ A θ Z 1 = 2 ( P 1 − P 2 ) X 1 U_\theta^T\Delta A_\theta Z_1=2(\mathrm{P}_1-\mathrm{P}_2)X_1 U θ T Δ A θ Z 1 = 2 ( P 1 − P 2 ) X 1 . In leaf coordinates d x = d A d x ⊥ dx=dA\,dx_\perp d x = d A d x ⊥ . The matrix F ( x ⊥ , z ) = Y 2 T ( P 1 − P 2 ) X 1 F(x_\perp,z)=Y_2^T(\mathrm{P}_1-\mathrm{P}_2)X_1 F ( x ⊥ , z ) = Y 2 T ( P 1 − P 2 ) X 1 is continuous in x ⊥ x_\perp x ⊥ with values in L 1 L^1 L 1 on any fixed leaf disk containing B R 0 B_{R_0} B R 0 ; the coefficient difference is supported in B R 0 B_{R_0} B R 0 , and all frames were chosen smoothly in the parameters. Fubini therefore gives precisely hypothesis (188) of Lemma 6.12, for every fixed g g g and every b ∈ C c ∞ ( I ) b\in C^\infty_c(I) b ∈ C c ∞ ( I ) . The lemma yields (189) for every x ⊥ ∈ I x_\perp\in I x ⊥ ∈ I and every polynomial g g g , with one and the same choice of X 1 , Y 2 X_1,Y_2 X 1 , Y 2 . Neither the estimates nor the smallness threshold for h h h are required to be uniform over the polynomial tests.
Proposition 6.14. Fix the extended Lamé class in (156) and a bound ∥ W j ∥ C 3 ( Ω 0 ) ≤ M 3 \|W_j\|_{C^3(\Omega_0)}\le M_3 ∥ W j ∥ C 3 ( Ω 0 ) ≤ M 3 , j = 1 , 2 j=1,2 j = 1 , 2 . There are fixed local direction and transverse charts on which the frames X 1 , Y 2 X_1,Y_2 X 1 , Y 2 of Theorem 6.13 can be selected before the tests, with
∥ X 1 ∥ C 4 + ∥ X 1 − 1 ∥ C 4 + ∥ Y 2 ∥ C 4 + ∥ Y 2 − 1 ∥ C 4 ≤ C . (190) \|X_1\|_{C^4}+\|X_1^{-1}\|_{C^4}+\|Y_2\|_{C^4}+\|Y_2^{-1}\|_{C^4}\le C. \tag*{(190)} ∥ X 1 ∥ C 4 + ∥ X 1 − 1 ∥ C 4 + ∥ Y 2 ∥ C 4 + ∥ Y 2 − 1 ∥ C 4 ≤ C . (190) The norms refer to the working leaf cylinder and a compact direction subchart. Choose a fixed compact transverse set J ⋐ I J\Subset I J ⋐ I and χ ∈ C c ∞ ( I ) \chi\in C^\infty_c(I) χ ∈ C c ∞ ( I ) equal to one near J J J . For tests satisfying
b ∈ C c ∞ ( I ) , supp b ⊂ J , g ∈ C [ z ] , ∥ b ∥ C 4 ≤ B t e s t , ∥ g ∥ C 4 ( D L s r c ‾ ) ≤ G t e s t , (191) b\in C^\infty_c(I),\qquad\operatorname{supp}b\subset J,\qquad g\in\mathbb{C}[z],\qquad\|b\|_{C^4}\le B_{\mathrm{test}},\qquad\|g\|_{C^4(\overline{D_{L_{\mathrm{src}}}})}\le G_{\mathrm{test}}, \tag*{(191)} b ∈ C c ∞ ( I ) , supp b ⊂ J , g ∈ C [ z ] , ∥ b ∥ C 4 ≤ B test , ∥ g ∥ C 4 ( D L src ) ≤ G test , (191) the exact solutions in the proof of Theorem 6.13 may be chosen so that all first corrections have uniform H 2 H^2 H 2 bounds, all remainders have uniform L 2 = O ( h 2 ) L^2=O(h^2) L 2 = O ( h 2 ) and H 1 = O ( h ) H^1=O(h) H 1 = O ( h ) bounds, and
∥ h Y 2 , h T D W z 1 , h − 2 b ( x ⊥ ) g ( z ) Y 2 T ( P 1 − P 2 ) X 1 ∥ L 1 ( Ω ) ≤ C h . (192) \left\|hY_{2,h}^T D^W z_{1,h}-2b(x_\perp)g(z)Y_2^T(\mathrm{P}_1-\mathrm{P}_2)X_1\right\|_{L^1(\Omega)}\le Ch. \tag*{(192)} h Y 2 , h T D W z 1 , h − 2 b ( x ⊥ ) g ( z ) Y 2 T ( P 1 − P 2 ) X 1 L 1 ( Ω ) ≤ C h . (192) Here D W = M 1 W 1 − M 2 W 2 D^W=M_1^{W_1}-M_2^{W_2} D W = M 1 W 1 − M 2 W 2 and the columns of Y 2 , h Y_{2,h} Y 2 , h have exactly the two scaled and one unscaled normalizations in that theorem. The constants and threshold depend only on the coefficient bounds in (156) , the fixed geometry and chart margins, B t e s t B_{\mathrm{test}} B test , G t e s t G_{\mathrm{test}} G test , and the fixed cutoffs.
Proof. Use Corollary 5.10 to select the four-component frame, the primal quotient frame X 1 X_1 X 1 , and the trace-normalized frame Φ \Phi Φ . Take a common finite parameter refinement for both materials. Choose a fixed finite chain of nested leaf disks: each new primitive has its source strictly inside a disk on which the preceding transport equation holds, and the final disk contains every leaf projection of U ‾ \overline{U} U . The radii and cutoff derivatives are independent of the materials and tests. All potential sources outside Ω 1 \Omega_1 Ω 1 use the cutoff (162) .
The uniform leading-amplitude bounds follow from the frame and primitive estimates, as follows. In Lemma 6.11, the source for the characteristic correction to ∇ θ X 1 \nabla_\theta X_1 ∇ θ X 1 is obtained by one derivative of X 1 X_1 X 1 and multiplication by A 1 , θ p h y s A_{1,\theta}^{\mathrm{phys}} A 1 , θ phys . The uniform sixth-order frame bounds therefore give a uniform C 5 C^5 C 5 source bound. The primitive preserves this bound by (107) . Thus the homogeneous lift Z 1 Z_1 Z 1 with U θ T Z 1 = X 1 U_\theta^T Z_1=X_1 U θ T Z 1 = X 1 is uniformly C 5 C^5 C 5 on its working disk. The locally fixed matrices U θ U_\theta U θ , V θ V_\theta V θ , e θ e_\theta e θ and their derivatives are bounded on the chosen spinor subcharts.
For the first two dual columns, the source H H H in Lemma 6.7 contains one derivative of c − 1 R c^{-1}R c − 1 R and otherwise only products of f f f , R R R and coefficients with at most two material derivatives. It is uniformly C 5 C^5 C 5 . The resulting A 0 , j F A_{0,j}^F A 0 , j F are uniformly C 5 C^5 C 5 . The normalization det Φ = 1 \det\Phi=1 det Φ = 1 was made before any test was chosen and preserves these bounds.
For the third column choose the static correction exactly as in Proposition 6.8. The source M A 0 S MA_{0}^{S} M A 0 S has a uniform C 6 C^6 C 6 bound, since A 0 S A_{0}^{S} A 0 S is spatially constant and the static coefficients use at most two material derivatives. The frame and primitive give a uniform C 6 C^6 C 6 bound on A 1 , 0 S A_{1,0}^{S} A 1 , 0 S , hence on f S f_S f S , R S R_S R S and the entries of the prescribed Y 2 Y_2 Y 2 . The identity det Y 2 = 2 c 2 \det Y_2 = 2c_2 det Y 2 = 2 c 2 , the upper bound for μ 2 \mu_2 μ 2 , and the adjugate formula bound Y 2 − 1 Y_2^{-1} Y 2 − 1 . Differentiating the inverse identity through order four gives (190) .
For each test, b g Z 1 bgZ_1 b g Z 1 and χ A 0 , j F \chi A^F_{0,j} χ A 0 , j F have uniform C 4 C^4 C 4 bounds. Lemma 6.4 applies to their first corrections. For the third column one must preserve the already selected Y 2 Y_2 Y 2 , not replace its correction by an unrelated one. Use instead
A 1 , χ S = χ ( A 1 , 0 S + ( β q θ , 0 ) T ) + w χ , β = 1 2 C θ ( χ l e a f W c / μ ) . A^S_{1,\chi}=\chi\left(A^S_{1,0}+(\beta q_\theta,0)^{\mathsf T}\right)+w_\chi,\qquad\beta=\frac{1}{2}C_\theta(\chi_{\mathrm{leaf}}W^c/\mu). A 1 , χ S = χ ( A 1 , 0 S + ( β q θ , 0 ) T ) + w χ , β = 2 1 C θ ( χ leaf W c / μ ) .
with w χ w_\chi w χ constructed as in Lemma 6.11. The C 2 C^2 C 2 norm of β \beta β uses only a C 2 C^2 C 2 bound on W W W . The C 2 C^2 C 2 norm of the source (187) uses derivatives of χ \chi χ through order four and bounded material derivatives. Its primitive is therefore uniformly C 2 C^2 C 2 . These estimates prove the required uniform H 2 H^2 H 2 bound for this specific A 1 , χ S A^S_{1,\chi} A 1 , χ S , including the full transition region. Its leading column is unchanged near J J J by the exact cancellations in Lemma 6.11.
Corollary 5.9 now supplies the simultaneous two-norm estimates for each selected remainder. All coefficients of the first-order representative K ^ l i f t W \widehat K^W_{\mathrm{lift}} K lift W and of N θ N_\theta N θ are uniformly bounded, so Lemma 6.6 yields ∥ q h ∥ 2 ≤ C h \lVert q_h\rVert_2 \leq Ch ∥ q h ∥ 2 ≤ C h in the notation of the matrix moment proof. Also the first-order operator D W : H 1 → L 2 D^W:H^1\to L^2 D W : H 1 → L 2 and its phase coefficient Δ A θ \Delta A_\theta Δ A θ have uniform bounds. The exact expansion in that proof consequently gives ∥ e h ∥ 2 ≤ C h \lVert e_h\rVert_2 \leq Ch ∥ e h ∥ 2 ≤ C h , while ∥ V χ ( 0 ) ∥ 2 + ∥ Δ A θ A ∥ 2 ≤ C \lVert\mathcal V^{(0)}_\chi\rVert_2+\lVert\Delta A_\theta A\rVert_2\leq C ∥ V χ ( 0 ) ∥ 2 + ∥ Δ A θ A ∥ 2 ≤ C . Cauchy–Schwarz proves (192) , using V χ ( 0 ) = U θ Y 2 \mathcal V^{(0)}_\chi=U_\theta Y_2 V χ ( 0 ) = U θ Y 2 near supp b \operatorname{supp} b supp b and U θ T Δ A θ Z 1 = 2 ( P 1 − P 2 ) X 1 U_\theta^{\mathsf T}\Delta A_\theta Z_1=2(P_1-P_2)X_1 U θ T Δ A θ Z 1 = 2 ( P 1 − P 2 ) X 1 . The construction and (192) hold under the stated coefficient bounds. Equality of the measured maps and the boundary matching conditions enter when the physical integral identity is applied as in Theorem 6.13. □
Remark 6.15. The frames are fixed on each local parameter chart before the tests b , g b,g b , g ; the cutoffs, amplitudes, fields, constants, and threshold h 0 h_0 h 0 may depend on each fixed test. The same remainder satisfies L 2 = O ( h 2 ) L^2=O(h^2) L 2 = O ( h 2 ) and H 1 = O ( h ) H^1=O(h) H 1 = O ( h ) throughout the target, including the transverse cutoff transitions.
Relative transport and rigidity of the Lamé parameters
This section turns the complete matrix moment identities of Theorem 6.13 into a global relative transport on the characteristic quotient and then uses its holomorphic structure to prove rigidity of the Lamé parameters. The construction has two stages. We first identify the primal and dual quotient connections and their transformation laws between normalized leaf coordinates and the two affine spinor charts of C P 1 \mathbb{CP}^{1} CP 1 . The moment identities then yield a compactly supported primitive through the Cauchy transform. In dual quotient coordinates this primitive defines a relative matrix H H H ; the chart transition laws show that its local representatives agree on overlaps. The relative transport equation and the common exterior normalization give the determinant identity
det H = μ 2 μ 1 > 0. \det H = \sqrt{\frac{\mu_{2}}{\mu_{1}}} > 0. det H = μ 1 μ 2 > 0.
Thus the resulting global endomorphism is an automorphism of the characteristic quotient bundle. Its holomorphic dependence on the projective null direction is established by differentiating the transport equation while keeping the physical point x x x fixed.
The second stage derives the rigidity of the material coefficients from this global holomorphic structure. The holomorphic block classification, together with the relative transport equations, identifies the common parameter κ 1 = κ 2 \kappa_{1}=\kappa_{2} κ 1 = κ 2 and reduces the remaining Lamé parameter difference to a trace-free Hessian equation for ℓ 12 = log ( μ 1 / μ 2 ) \ell_{12}=\log(\mu_{1}/\mu_{2}) ℓ 12 = log ( μ 1 / μ 2 ) . Introducing the four-component state
U = ( ∇ ℓ 12 , Δ ℓ 12 ) T U=(\nabla\ell_{12},\Delta\ell_{12})^{T} U = ( ∇ ℓ 12 , Δ ℓ 12 ) T
converts this overdetermined second-order equation into a homogeneous first-order system with a 4 × 4 4\times4 4 × 4 coefficient matrix. The common exterior normalization gives U = 0 U=0 U = 0 and ℓ 12 = 0 \ell_{12}=0 ℓ 12 = 0 on a nonempty open set. Uniqueness along paths then yields U = 0 U=0 U = 0 throughout the connected domain. Thus ∇ ℓ 12 = 0 \nabla\ell_{12}=0 ∇ ℓ 12 = 0 , and the exterior value fixes the constant to be zero. Consequently, μ 1 = μ 2 \mu_{1}=\mu_{2} μ 1 = μ 2 , and the relation between λ \lambda λ , μ \mu μ , and κ \kappa κ gives λ 1 = λ 2 \lambda_{1}=\lambda_{2} λ 1 = λ 2 .
The induced dual transport connection
We use ℓ \ell ℓ , m m m , c c c , d d d , F μ F_{\mu} F μ with the meanings fixed in (60) –(63) . In particular, F μ F_{\mu} F μ is real and symmetric, and A θ p h y s A_{\theta}^{\mathrm{phys}} A θ phys is the contraction introduced in § 4.5 . A nonzero spinor ( s 0 , s 1 ) (s_{0},s_{1}) ( s 0 , s 1 ) determines
θ = ( s 0 2 − s 1 2 , i ( s 0 2 + s 1 2 ) , 2 s 0 s 1 ) T , Q = 1 2 ( − s 1 − s 0 i s 1 − i s 0 s 0 − s 1 ) , n s p = ( − s 1 , s 0 ) T , U θ = diag ( Q , 1 ) . \theta=\left(s_{0}^{2}-s_{1}^{2},i(s_{0}^{2}+s_{1}^{2}),2s_{0}s_{1}\right)^{T},\quad
Q=\frac{1}{2}\begin{pmatrix}-s_{1}&-s_{0}\\is_{1}&-is_{0}\\s_{0}&-s_{1}\end{pmatrix},\quad
n_{\mathrm{sp}}=(-s_{1},s_{0})^{T},\quad U_{\theta}=\operatorname{diag}(Q,1). θ = ( s 0 2 − s 1 2 , i ( s 0 2 + s 1 2 ) , 2 s 0 s 1 ) T , Q = 2 1 − s 1 i s 1 s 0 − s 0 − i s 0 − s 1 , n sp = ( − s 1 , s 0 ) T , U θ = diag ( Q , 1 ) .
Direct multiplication gives θ T Q = 0 \theta^{T}Q=0 θ T Q = 0 , Q n s p = − θ / 2 Qn_{\mathrm{sp}}=-\theta/2 Q n sp = − θ /2 and rank Q = 2 \operatorname{rank}Q=2 rank Q = 2 . Thus U θ U_{\theta} U θ identifies C 3 \mathbb{C}^{3} C 3 with the dual of the characteristic quotient,
E θ = C 4 / C ( θ , 0 ) T , E θ ∗ = { ( p , η ) : θ T p = 0 } . E_{\theta}=\mathbb{C}^{4}/\mathbb{C}(\theta,0)^{T},\qquad
E_{\theta}^{*}=\{(p,\eta):\theta^{T}p=0\}. E θ = C 4 / C ( θ , 0 ) T , E θ ∗ = {( p , η ) : θ T p = 0 } .
The duality is the bilinear pairing fixed in § 1.6 . In particular, E θ ∗ E_{\theta}^{*} E θ ∗ is the annihilator of the characteristic line.
Proposition 7.1. Put r m = − 2 m T Q r_{m}=-2m^{T}Q r m = − 2 m T Q . The dual connection induced by the physical reduction is
Γ = ( n s p r m 2 d n s p − θ T F μ Q 0 ) , − ( A θ p h y s ) T U θ = U θ Γ . (193) \Gamma=
\begin{pmatrix}
n_{\mathrm{sp}}r_{m} & 2dn_{\mathrm{sp}}\\
-\theta^{T}F_{\mu}Q & 0
\end{pmatrix},
\qquad
-\left(A_{\theta}^{\mathrm{phys}}\right)^{T}U_{\theta}=U_{\theta}\Gamma. \tag*{(193)} Γ = ( n sp r m − θ T F μ Q 2 d n sp 0 ) , − ( A θ phys ) T U θ = U θ Γ. (193) For every spatial field Y Y Y ,
( 2 D θ − ( A θ p h y s ) T ) ( U θ Y ) = U θ ( 2 D θ + Γ ) Y . (194) \left(2D_{\theta}-\left(A_{\theta}^{\mathrm{phys}}\right)^{T}\right)(U_{\theta}Y)
=U_{\theta}(2D_{\theta}+\Gamma)Y. \tag*{(194)} ( 2 D θ − ( A θ phys ) T ) ( U θ Y ) = U θ ( 2 D θ + Γ ) Y . (194) If V V V is any spatially constant right inverse of U θ T U_{\theta}^{T} U θ T , the quotient connection matrix A = U θ T A θ p h y s V \mathcal{A}=U_{\theta}^{T}A_{\theta}^{\mathrm{phys}}V A = U θ T A θ phys V is independent of V V V and satisfies Γ = − A T \Gamma=-\mathcal{A}^{T} Γ = − A T .
Proof. The identity A θ p h y s ( θ , 0 ) T = 0 A_{\theta}^{\mathrm{phys}}(\theta,0)^{T}=0 A θ phys ( θ , 0 ) T = 0 shows that A θ p h y s A_{\theta}^{\mathrm{phys}} A θ phys induces a well-defined operator on the quotient. Using Q n s p = − θ / 2 Q_{\mathrm{nsp}}=-\theta/2 Q nsp = − θ /2 and F μ T = F μ F_{\mu}^{T}=F_{\mu} F μ T = F μ , we compute
U θ Γ = ( θ m T Q − d θ − θ T F μ Q 0 ) = − ( A θ p h y s ) T U θ . U_{\theta}\Gamma=
\begin{pmatrix}
\theta m^{T}Q & -d\theta\\
-\theta^{T}F_{\mu}Q & 0
\end{pmatrix}
=-(A_{\theta}^{\mathrm{phys}})^{T}U_{\theta}. U θ Γ = ( θ m T Q − θ T F μ Q − d θ 0 ) = − ( A θ phys ) T U θ .
Since U θ U_{\theta} U θ is spatially constant, this also proves (194) . If V ~ \widetilde{V} V is another right inverse, then ran ( V − V ~ ) \operatorname{ran}(V-\widetilde{V}) ran ( V − V ) lies in ker U θ T = C ( θ , 0 ) T \ker U_{\theta}^{T}=\mathbb{C}(\theta,0)^{T} ker U θ T = C ( θ , 0 ) T , which A θ p h y s A_{\theta}^{\mathrm{phys}} A θ phys annihilates. Hence A \mathcal{A} A is independent of the right inverse. Multiplying (193) on the left by V T V^{T} V T gives Γ = − V T ( A θ p h y s ) T U θ = − A T \Gamma=-V^{T}(A_{\theta}^{\mathrm{phys}})^{T}U_{\theta}=-\mathcal{A}^{T} Γ = − V T ( A θ phys ) T U θ = − A T .
The matrix P = A / 2 P=\mathcal{A}/2 P = A /2 in (186) is the primal quotient coefficient, whereas Γ \Gamma Γ acts on dual quotient coordinates. Proposition 7.1 therefore identifies them by Γ = − 2 P T \Gamma=-2P^{T} Γ = − 2 P T . This identity converts the matrix moments into the relative dual transport equation used below. Using the leaf-coordinate identity D θ = 2 ∂ z ‾ D_{\theta}=2\partial_{\overline{z}} D θ = 2 ∂ z from Section 4.5, we obtain
( 2 D θ + A ) X = 0 ⟺ 2 ∂ z ‾ X + P X = 0 , ( 2 D θ + Γ ) Y = 0 ⟺ 2 ∂ z ‾ Y − P T Y = 0. (2D_{\theta}+\mathcal{A})X=0\ \Longleftrightarrow\ 2\partial_{\overline{z}}X+PX=0,\qquad
(2D_{\theta}+\Gamma)Y=0\ \Longleftrightarrow\ 2\partial_{\overline{z}}Y-P^{T}Y=0. ( 2 D θ + A ) X = 0 ⟺ 2 ∂ z X + P X = 0 , ( 2 D θ + Γ ) Y = 0 ⟺ 2 ∂ z Y − P T Y = 0.
Normalization and transition between the two affine charts
The moment construction uses normalized real leaf coordinates, whereas holomorphic dependence on direction is expressed in spinor coordinates. We record their change of frame before constructing the relative field. For an invertible relative matrix H H H in dual quotient coordinates, write G = H − T G=H^{-T} G = H − T for the corresponding primal matrix; Proposition 7.5 will construct such a matrix H H H and prove its invertibility. Let θ h o l , Q h o l , n s p , h o l , U h o l \theta_{\mathrm{hol}},Q_{\mathrm{hol}},n_{\mathrm{sp,hol}},U_{\mathrm{hol}} θ hol , Q hol , n sp , hol , U hol denote the quantities associated with a holomorphic spinor representative. The subscripts hol and nor distinguish this representative from its real-normalized version. Direct calculation gives
θ h o l ⋅ θ h o l ‾ = 2 ( ∣ s 0 ∣ 2 + ∣ s 1 ∣ 2 ) 2 . \theta_{\mathrm{hol}}\cdot\overline{\theta_{\mathrm{hol}}}=2\left(\lvert s_{0}\rvert^{2}+\lvert s_{1}\rvert^{2}\right)^{2}. θ hol ⋅ θ hol = 2 ( ∣ s 0 ∣ 2 + ∣ s 1 ∣ 2 ) 2 .
Set
a n o r = ( ∣ s 0 ∣ 2 + ∣ s 1 ∣ 2 ) − 1 > 0 , ς = a n o r , S = diag ( ς I 2 , 1 ) , θ n o r = a n o r θ h o l . a_{\mathrm{nor}}=\left(\lvert s_{0}\rvert^{2}+\lvert s_{1}\rvert^{2}\right)^{-1}>0,\qquad
\varsigma=\sqrt{a_{\mathrm{nor}}},\qquad
S=\operatorname{diag}(\varsigma I_{2},1),\qquad
\theta_{\mathrm{nor}}=a_{\mathrm{nor}}\theta_{\mathrm{hol}}. a nor = ( ∣ s 0 ∣ 2 + ∣ s 1 ∣ 2 ) − 1 > 0 , ς = a nor , S = diag ( ς I 2 , 1 ) , θ nor = a nor θ hol .
The normalized spinor is ς ( s 0 , s 1 ) \varsigma(s_{0},s_{1}) ς ( s 0 , s 1 ) and U n o r = U h o l S U_{\mathrm{nor}}=U_{\mathrm{hol}}S U nor = U hol S . Coordinates of the same ambient dual vector satisfy v h o l = S v n o r v_{\mathrm{hol}}=Sv_{\mathrm{nor}} v hol = S v nor , while primal quotient coordinates satisfy u n o r = S T u h o l u_{\mathrm{nor}}=S^{T}u_{\mathrm{hol}} u nor = S T u hol . Accordingly,
Y h o l = S Y n o r , X h o l = S − T X n o r , Y_{\mathrm{hol}}=SY_{\mathrm{nor}},\qquad
X_{\mathrm{hol}}=S^{-T}X_{\mathrm{nor}}, Y hol = S Y nor , X hol = S − T X nor ,
Γ n o r = a n o r S − 1 Γ h o l S , A n o r = a n o r S T A h o l S − T , \Gamma_{\mathrm{nor}}=a_{\mathrm{nor}}S^{-1}\Gamma_{\mathrm{hol}}S,\qquad
\mathcal{A}_{\mathrm{nor}}=a_{\mathrm{nor}}S^{T}\mathcal{A}_{\mathrm{hol}}S^{-T}, Γ nor = a nor S − 1 Γ hol S , A nor = a nor S T A hol S − T ,
{#eq:7.3}
H_{\mathrm{hol}}=SH_{\mathrm{nor}}S^{-1},\qquad
G_{\mathrm{hol}}=S^{-T}G_{\mathrm{nor}}S^{T}.
These change-of-frame matrices depend only on the direction and hence are constant under spatial differentiation. Real normalization involves ζ ‾ \overline{\zeta} ζ ; holomorphic dependence will be established after passing back to the holomorphic spinor representatives.
The two affine charts of C P 1 \mathbb{CP}^{1} CP 1 use the spinor representatives ( 1 , ζ ) (1,\zeta) ( 1 , ζ ) with ζ = s 1 / s 0 \zeta=s_{1}/s_{0} ζ = s 1 / s 0 and ( υ , 1 ) (\upsilon,1) ( υ , 1 ) with υ = s 0 / s 1 \upsilon=s_{0}/s_{1} υ = s 0 / s 1 . On their overlap υ = ζ − 1 \upsilon=\zeta^{-1} υ = ζ − 1 . Put S ζ = diag ( ζ I 2 , 1 ) S_{\zeta}=\operatorname{diag}(\zeta I_{2},1) S ζ = diag ( ζ I 2 , 1 ) . Homogeneity gives
θ 0 = ζ 2 θ ∞ , U 0 = U ∞ S ζ , Γ 0 = ζ 2 S ζ − 1 Γ ∞ S ζ , A 0 = ζ 2 S ζ T A ∞ S ζ − T . \theta_{0}=\zeta^{2}\theta_{\infty},\qquad
U_{0}=U_{\infty}S_{\zeta},\qquad
\Gamma_{0}=\zeta^{2}S_{\zeta}^{-1}\Gamma_{\infty}S_{\zeta},\qquad
\mathcal{A}_{0}=\zeta^{2}S_{\zeta}^{T}\mathcal{A}_{\infty}S_{\zeta}^{-T}. θ 0 = ζ 2 θ ∞ , U 0 = U ∞ S ζ , Γ 0 = ζ 2 S ζ − 1 Γ ∞ S ζ , A 0 = ζ 2 S ζ T A ∞ S ζ − T .
The relative matrices therefore obey
{#eq:7.4}
H_{0}=S_{\zeta}^{-1}H_{\infty}S_{\zeta},\qquad
G_{0}=S_{\zeta}^{T}G_{\infty}S_{\zeta}^{-T}.
These are the coordinate expressions of the corresponding quotient-bundle automorphisms.
Lemma 7.2. Let D ⊂ C D\subset\mathbb{C} D ⊂ C be an arbitrary connected open set, and let P 1 , P 2 P_{1},P_{2} P 1 , P 2 be continuous 3 × 3 3\times3 3 × 3 matrix fields on D D D . Suppose that invertible C 1 C^{1} C 1 frames X 1 , Y 2 X_{1},Y_{2} X 1 , Y 2 on D D D satisfy
2 ∂ z ‾ X 1 = − P 1 X 1 , 2 ∂ z ‾ Y 2 = P 2 T Y 2 . 2\partial_{\overline{z}}X_{1}=-P_{1}X_{1},\qquad
2\partial_{\overline{z}}Y_{2}=P_{2}^{T}Y_{2}. 2 ∂ z X 1 = − P 1 X 1 , 2 ∂ z Y 2 = P 2 T Y 2 .
If H H H and H ~ \widetilde{H} H are C 1 C^1 C 1 solutions of
2 ∂ z ‾ H − P 1 T H + H P 2 T = 0 (195) 2\partial_{\overline{z}}H-P_1^{\mathsf{T}}H+HP_2^{\mathsf{T}}=0 \tag*{(195)} 2 ∂ z H − P 1 T H + H P 2 T = 0 (195)
on D D D and agree on a nonempty open subset, then H = H ~ H=\widetilde{H} H = H throughout D D D . The conclusion is unaffected by a change of quotient or dual-quotient frame that is constant in the spatial variable, provided both connections and both relative fields are transformed accordingly.
Proof. Set V = H − H ~ V=H-\widetilde{H} V = H − H . By the three displayed transport equations,
2 ∂ z ‾ ( X 1 T V Y 2 ) = − X 1 T P 1 T V Y 2 + X 1 T ( P 1 T V − V P 2 T ) Y 2 + X 1 T V P 2 T Y 2 = 0. 2\partial_{\overline{z}}(X_1^{\mathsf{T}} VY_2)=-X_1^{\mathsf{T}}P_1^{\mathsf{T}}VY_2+X_1^{\mathsf{T}}(P_1^{\mathsf{T}}V-VP_2^{\mathsf{T}})Y_2+X_1^{\mathsf{T}}VP_2^{\mathsf{T}}Y_2=0. 2 ∂ z ( X 1 T V Y 2 ) = − X 1 T P 1 T V Y 2 + X 1 T ( P 1 T V − V P 2 T ) Y 2 + X 1 T V P 2 T Y 2 = 0.
Thus each entry of X 1 T V Y 2 X_1^{\mathsf{T}}VY_2 X 1 T V Y 2 is holomorphic. It vanishes on the prescribed open set and hence on all of connected D D D by the identity theorem. Since both frames are invertible, V = 0 V=0 V = 0 on D D D . An x x x -independent invertible frame change conjugates the transformed relative equation to (195) ; it also takes equality on the open set to equality there. This proves the last assertion. □
Construction from the matrix moment identities
Lemma 7.3. Let 0 < R 0 < L 0<R_0<L 0 < R 0 < L and let F ∈ C ∞ ( D L ‾ ; C 3 × 3 ) \mathbf{F}\in C^\infty(\overline{D_L};\mathbb{C}^{3\times3}) F ∈ C ∞ ( D L ; C 3 × 3 ) be supported in D ‾ R 0 \overline{D}_{R_0} D R 0 , extended by zero to C \mathbb{C} C . Define its moments and Cauchy transform by
M q = ∫ C w q F ( w ) d A ( w ) , ( C F ) ( z ) = 1 π ∫ C F ( w ) z − w d A ( w ) , q ≥ 0. (196) M_q=\int_{\mathbb{C}}w^q\mathbf{F}(w)\,dA(w),\qquad(\mathcal{C}\mathbf{F})(z)=\frac{1}{\pi}\int_{\mathbb{C}}\frac{\mathbf{F}(w)}{z-w}\,dA(w),\qquad q\geq0. \tag*{(196)} M q = ∫ C w q F ( w ) d A ( w ) , ( C F ) ( z ) = π 1 ∫ C z − w F ( w ) d A ( w ) , q ≥ 0. (196)
The following conditions are equivalent: all M q M_q M q vanish; C F = 0 \mathcal{C}\mathbf{F}=0 C F = 0 for ∣ z ∣ > R 0 |z|>R_0 ∣ z ∣ > R 0 ; and there exists a smooth matrix Ψ \Psi Ψ supported in D ‾ R 0 \overline{D}_{R_0} D R 0 with ∂ z ‾ Ψ = F \partial_{\overline{z}}\Psi=\mathbf{F} ∂ z Ψ = F . Such a compactly supported primitive is unique and equals C F \mathcal{C}\mathbf{F} C F . For any fixed matrix norm and its associated L 1 L^1 L 1 norm,
∥ C F ( z ) − 1 π ∑ q = 0 N M q z q + 1 ∥ ≤ ∥ F ∥ L 1 π ( ∣ z ∣ − R 0 ) ( R 0 ∣ z ∣ ) N + 1 , ∣ z ∣ > R 0 , N ≥ 0. (197) \left\|\mathcal{C}\mathbf{F}(z)-\frac{1}{\pi}\sum_{q=0}^{N}\frac{M_q}{z^{q+1}}\right\|\leq\frac{\|\mathbf{F}\|_{L^1}}{\pi(|z|-R_0)}\left(\frac{R_0}{|z|}\right)^{N+1},\qquad|z|>R_0,\quad N\geq0. \tag*{(197)} C F ( z ) − π 1 q = 0 ∑ N z q + 1 M q ≤ π ( ∣ z ∣ − R 0 ) ∥ F ∥ L 1 ( ∣ z ∣ R 0 ) N + 1 , ∣ z ∣ > R 0 , N ≥ 0. (197)
Under the equivalent conditions, the tests extend from polynomials to all scalar holomorphic functions on D L D_L D L :
∫ D L φ ( z ) F ( z ) d A ( z ) = 0 , φ ∈ O ( D L ) . (198) \int_{D_L}\varphi(z)\mathbf{F}(z)\,dA(z)=0,\qquad\varphi\in\mathcal{O}(D_L). \tag*{(198)} ∫ D L φ ( z ) F ( z ) d A ( z ) = 0 , φ ∈ O ( D L ) . (198)
Here O ( D L ) \mathcal{O}(D_L) O ( D L ) denotes the holomorphic functions on the disk.
Proof. The distributional identity (96) gives ∂ z ‾ C F = F \partial_{\overline{z}}\mathcal{C}\mathbf{F}=\mathbf{F} ∂ z C F = F . For smooth compactly supported sources every real derivative may be transferred to the source; local integrability of ∣ z − w ∣ − 1 |z-w|^{-1} ∣ z − w ∣ − 1 then makes the transform smooth. The finite geometric identity
1 z − w = ∑ q = 0 N w q z q + 1 + w N + 1 z N + 1 ( z − w ) \frac{1}{z-w}=\sum_{q=0}^{N}\frac{w^q}{z^{q+1}}+\frac{w^{N+1}}{z^{N+1}(z-w)} z − w 1 = q = 0 ∑ N z q + 1 w q + z N + 1 ( z − w ) w N + 1
gives (197) by integration, without an interchange of infinite sums and integrals. If all moments vanish, letting N → ∞ N\to\infty N → ∞ at each ∣ z ∣ > R 0 |z|>R_0 ∣ z ∣ > R 0 proves exterior vanishing and hence the asserted compact primitive.
Conversely, suppose ∂ z ‾ Ψ = F \partial_{\overline{z}}\Psi=\mathbf{F} ∂ z Ψ = F with the specified support. Given φ ∈ O ( D L ) \varphi\in\mathcal{O}(D_L) φ ∈ O ( D L ) , choose χ ∈ C c ∞ ( D L ) \chi\in C^\infty_c(D_L) χ ∈ C c ∞ ( D L ) equal to one near D ‾ R 0 \overline{D}_{R_0} D R 0 . Integration by parts gives
∫ D L φ F d A = − ∫ D L Ψ ∂ z ‾ ( χ φ ) d A = 0 , \int_{D_L}\varphi\mathbf{F}\,dA=-\int_{D_L}\Psi\,\partial_{\overline{z}}(\chi\varphi)\,dA=0, ∫ D L φ F d A = − ∫ D L Ψ ∂ z ( χ φ ) d A = 0 ,
since ∂ z ‾ φ = 0 \partial_{\overline{z}}\varphi=0 ∂ z φ = 0 and ∇ χ \nabla\chi ∇ χ vanishes on supp Ψ \operatorname{supp}\Psi supp Ψ . This proves (198) and, in particular, all polynomial moments. The difference of two smooth compact primitives is an entire holomorphic matrix and vanishes outside a compact set; the identity theorem makes it zero. Applying this to Ψ \Psi Ψ and the already compact C F \mathcal{C}F C F proves uniqueness and the formula.
Proposition 7.4. Let P 1 P_1 P 1 , P 2 P_2 P 2 be smooth matrix fields on D L D_L D L , with supp ( P 1 − P 2 ) ⊂ D ‾ R 0 \operatorname{supp}(P_1-P_2)\subset\overline{D}_{R_0} supp ( P 1 − P 2 ) ⊂ D R 0 , where 0 < R 0 < L 0<R_0<L 0 < R 0 < L . Let X 1 , Y 2 X_1,Y_2 X 1 , Y 2 be smooth invertible frames satisfying
2 ∂ z ˉ X 1 = − P 1 X 1 , 2 ∂ z ˉ Y 2 = P 2 T Y 2 . 2\partial_{\bar z}X_1=-P_1X_1,\qquad2\partial_{\bar z}Y_2=P_2^{T}Y_2. 2 ∂ z ˉ X 1 = − P 1 X 1 , 2 ∂ z ˉ Y 2 = P 2 T Y 2 . Put F m o m = Y 2 T ( P 1 − P 2 ) X 1 F_{\mathrm{mom}}=Y_2^{T}(P_1-P_2)X_1 F mom = Y 2 T ( P 1 − P 2 ) X 1 and W = Y 2 T X 1 W=Y_2^{T}X_1 W = Y 2 T X 1 . Vanishing of every polynomial moment of F m o m F_{\mathrm{mom}} F mom is equivalent to the existence of a smooth solution H H H of (195) with H = I 3 H=I_3 H = I 3 on D L ∖ D ‾ R 0 D_L\setminus\overline{D}_{R_0} D L ∖ D R 0 . This solution is unique and is explicitly
H = I 3 + 1 2 X 1 − T ( C F m o m ) T Y 2 − 1 . (199) H=I_3+\frac{1}{2}X_1^{-T}(\mathcal{C}F_{\mathrm{mom}})^TY_2^{-1}. \tag*{(199)} H = I 3 + 2 1 X 1 − T ( C F mom ) T Y 2 − 1 . (199) Invertibility is established after the global construction by (202) . The field is independent of the choices of the two solution frames, while holding the quotient coordinates fixed.
Proof. The frame equations give 2 ∂ z ˉ W = − F m o m 2\partial_{\bar z}W=-F_{\mathrm{mom}} 2 ∂ z ˉ W = − F mom . If all moments vanish, Lemma 7.3 gives the compact primitive C F m o m \mathcal{C}F_{\mathrm{mom}} C F mom , so
R h o l = W + 1 2 C F m o m R_{\mathrm{hol}}=W+\frac{1}{2}\mathcal{C}F_{\mathrm{mom}} R hol = W + 2 1 C F mom
is holomorphic. Then H = X 1 − T R h o l T Y 2 − 1 H=X_1^{-T}R_{\mathrm{hol}}^TY_2^{-1} H = X 1 − T R hol T Y 2 − 1 equals (199) . Differentiating this expression gives (195) ; it equals I 3 I_3 I 3 in the outer annulus. Uniqueness follows from Lemma 7.2. Conversely, for any such H H H the frame equations imply ∂ z ˉ ( Y 2 T H T X 1 ) = 0 \partial_{\bar z}(Y_2^TH^TX_1)=0 ∂ z ˉ ( Y 2 T H T X 1 ) = 0 . Hence
Ψ = 2 Y 2 T ( H T − I 3 ) X 1 \Psi=2Y_2^T(H^T-I_3)X_1 Ψ = 2 Y 2 T ( H T − I 3 ) X 1
is smooth, supported in D ‾ R 0 \overline{D}_{R_0} D R 0 , and satisfies ∂ z ˉ Ψ = F m o m \partial_{\bar z}\Psi=F_{\mathrm{mom}} ∂ z ˉ Ψ = F mom after zero extension. Lemma 7.3 proves all the moments.
To establish independence directly, let X ~ 1 , Y ~ 2 \widetilde{X}_1,\widetilde{Y}_2 X 1 , Y 2 be any other invertible frames for these same equations. Define
G X = X 1 − 1 X ~ 1 , G Y = Y 2 − 1 Y ~ 2 , ∂ z ˉ G X = ∂ z ˉ G Y = 0. (200) G_X=X_1^{-1}\widetilde{X}_1,\qquad G_Y=Y_2^{-1}\widetilde{Y}_2,\qquad\partial_{\bar z}G_X=\partial_{\bar z}G_Y=0. \tag*{(200)} G X = X 1 − 1 X 1 , G Y = Y 2 − 1 Y 2 , ∂ z ˉ G X = ∂ z ˉ G Y = 0. (200)
The last two equalities follow by differentiating the first two expressions and canceling the identical connection terms. For the new source, F ~ m o m = G Y T F m o m G X \widetilde{F}_{\mathrm{mom}}=G_Y^TF_{\mathrm{mom}}G_X F mom = G Y T F mom G X . The smooth compact matrix G Y T ( C F m o m ) G X G_Y^T(\mathcal{C}F_{\mathrm{mom}})G_X G Y T ( C F mom ) G X , extended by zero outside D L D_L D L , has ∂ z ˉ \partial_{\bar z} ∂ z ˉ derivative F ~ m o m \widetilde{F}_{\mathrm{mom}} F mom . The lemma therefore gives all its moments and the exact identity
C F ~ m o m = G Y T ( C F m o m ) G X on D L . (201) \mathcal{C}\widetilde{F}_{\mathrm{mom}}=G_Y^T(\mathcal{C}F_{\mathrm{mom}})G_X\quad\text{on }D_L. \tag*{(201)} C F mom = G Y T ( C F mom ) G X on D L . (201)
Inserting this into (199) , with X ~ 1 = X 1 G X \widetilde{X}_1=X_1G_X X 1 = X 1 G X and Y ~ 2 = Y 2 G Y \widetilde{Y}_2=Y_2G_Y Y 2 = Y 2 G Y , cancels both frame factors in their stated order and gives exactly the same H H H . The inverse factors are also holomorphic, so the covariance identity holds for arbitrary holomorphic frame changes arising from the same transport equations.
The changes in (200) act on the solution frames while the quotient coordinates remain fixed. They therefore leave H H H unchanged; a spinor coordinate change instead conjugates H H H according to (7.4). In particular, the correction freedom in Lemma 6.9 leaves H H H invariant. The moment transfer is carried out through compactly supported primitives, with the exact physical solutions left unchanged.
Proposition 7.5. Suppose the two extended materials agree on B R ∖ Ω ‾ B_R\setminus\overline{\Omega} B R ∖ Ω and all coefficient differences are supported in B R 0 B_{R_0} B R 0 , with R 0 < R R_0<R R 0 < R . For each normalized direction and each centre of the transverse
parameter interval, suppose the moment construction provides invertible, smoothly parameter-dependent frames X 1 , Y 2 X_1,Y_2 X 1 , Y 2 on a fixed leaf disk D L D_L D L , L > R L>R L > R , over a local transverse and direction parameter chart, such that
2 ∂ z ˉ X 1 = − P 1 X 1 , 2 ∂ z ˉ Y 2 = P 2 T Y 2 , ∫ D L z q Y 2 T ( P 1 − P 2 ) X 1 d A = 0 ( q ≥ 0 ) . 2\partial_{\bar z}X_1=-P_1X_1,\qquad2\partial_{\bar z}Y_2=P_2^{T}Y_2,\qquad\int_{D_L}z^qY_2^{T}(P_1-P_2)X_1\,\mathrm{d}A=0\quad(q\geq0). 2 ∂ z ˉ X 1 = − P 1 X 1 , 2 ∂ z ˉ Y 2 = P 2 T Y 2 , ∫ D L z q Y 2 T ( P 1 − P 2 ) X 1 d A = 0 ( q ≥ 0 ) .
The frames are fixed before the tests are chosen, and the identities hold for every leaf parameter in the local chart. Then there is a unique jointly smooth relative bundle automorphism H H H over B R × C P 1 B_R\times\mathbb{CP}^1 B R × CP 1 with the properties in Proposition 7.6. Its determinant is
det H = μ 2 μ 1 > 0. (202) \det H=\sqrt{\frac{\mu_2}{\mu_1}}>0. \tag*{(202)} det H = μ 1 μ 2 > 0. (202)
Proof. On each fixed leaf of a local parameter cylinder, apply Proposition 7.4 with F m o m = Y 2 T ( P 1 − P 2 ) X 1 F_{\mathrm{mom}}=Y_2^{T}(P_1-P_2)X_1 F mom = Y 2 T ( P 1 − P 2 ) X 1 . Its hypotheses hold because the coefficient difference is supported inside B R 0 B_{R_0} B R 0 , and hence inside D R 0 D_{R_0} D R 0 on every leaf. Formula (199) gives H H H satisfying (195) and equal to I 3 I_3 I 3 for ∣ z ∣ > R 0 |z|>R_0 ∣ z ∣ > R 0 . The Cauchy transform C \mathcal{C} C in (196) has twice the normalization of C θ C_\theta C θ defined in Definition 6.1. Proposition 7.1 converts the leaf equation to 2 D θ H + Γ 1 H − H Γ 2 = 0 2D_\theta H+\Gamma_1H-H\Gamma_2=0 2 D θ H + Γ 1 H − H Γ 2 = 0 .
The determinant identity now establishes invertibility. Formula (193) gives tr Γ j = m j ⋅ θ \operatorname{tr}\Gamma_j=\mathbf{m}_j\cdot\theta tr Γ j = m j ⋅ θ . Let adj ( H ) \operatorname{adj}(H) adj ( H ) denote the classical adjugate, the transpose of the cofactor matrix. It is defined for every square matrix and satisfies H adj ( H ) = det ( H ) I H\operatorname{adj}(H)=\det(H)I H adj ( H ) = det ( H ) I . The columnwise determinant identity reads
D θ det H = tr ( adj ( H ) D θ H ) . D_\theta\det H=\operatorname{tr}\bigl(\operatorname{adj}(H)D_\theta H\bigr). D θ det H = tr ( adj ( H ) D θ H ) .
Together with H adj ( H ) = adj ( H ) H = det ( H ) I H\operatorname{adj}(H)=\operatorname{adj}(H)H=\det(H)I H adj ( H ) = adj ( H ) H = det ( H ) I , the relative equation yields
2 D θ det H + ( D θ log μ 1 − D θ log μ 2 ) det H = 0. 2D_\theta\det H+(D_\theta\log\mu_1-D_\theta\log\mu_2)\det H=0. 2 D θ det H + ( D θ log μ 1 − D θ log μ 2 ) det H = 0.
Thus μ 1 / μ 2 det H \sqrt{\mu_1/\mu_2}\det H μ 1 / μ 2 det H is holomorphic on each leaf and equals one on an exterior open subset. The identity theorem on the connected leaf disk proves (202) . In particular, H H H is invertible.
The Cauchy transform is jointly smooth in the fixed leaf coordinates and the parameters. Its source has compact support in the interior of the fixed disk, uniformly on a smaller parameter chart. Thus no moving-boundary term appears when differentiating the integral. Derivatives in the leaf variable can be transferred to this compactly supported source by integration by parts, while parameter derivatives act on the source and the fixed-coordinate pullback. Local integrability of ∣ z − w ∣ − 1 |z-w|^{-1} ∣ z − w ∣ − 1 then bounds every prescribed finite number of these derivatives. The full chain rule gives joint smoothness in physical coordinates.
For a fixed transverse coordinate t t t , the exterior region
E t e x t = { z ∈ D L : ∣ z ∣ 2 + t 2 > R 0 2 } E_t^{\mathrm{ext}}=\{z\in D_L:|z|^2+t^2>R_0^2\} E t ext = { z ∈ D L : ∣ z ∣ 2 + t 2 > R 0 2 }
is connected: it is an annulus with inner radius R 0 2 − t 2 \sqrt{R_0^2-t^2} R 0 2 − t 2 if ∣ t ∣ < R 0 |t|<R_0 ∣ t ∣ < R 0 , a punctured disk if ∣ t ∣ = R 0 |t|=R_0 ∣ t ∣ = R 0 , and the full disk if ∣ t ∣ > R 0 |t|>R_0 ∣ t ∣ > R 0 . This distinction at ∣ t ∣ = R 0 |t|=R_0 ∣ t ∣ = R 0 is needed because the defining inequality is strict. Every E t e x t E_t^{\mathrm{ext}} E t ext intersects the open set { ∣ z ∣ > R 0 } \{|z|>R_0\} { ∣ z ∣ > R 0 } , since L > R > R 0 L>R>R_0 L > R > R 0 . On E t e x t E_t^{\mathrm{ext}} E t ext the material connections agree; thus I I I solves the relative transport equation there. The frames from the full disk restrict to invertible frames on E t e x t E_t^{\mathrm{ext}} E t ext . Lemma 7.2, now stated for arbitrary connected planar domains, applies to H H H and I I I on E t e x t E_t^{\mathrm{ext}} E t ext and gives H = I H=I H = I throughout that region. For ∣ t ∣ ≥ R 0 |t|\geq R_0 ∣ t ∣ ≥ R 0 , the moment source is supported in the empty intersection of the leaf with the open ball B R 0 B_{R_0} B R 0 , so it vanishes on the entire leaf and the construction gives H = I H=I H = I directly, including at equality.
To check independence of the local data, take two constructions over a common direction-and-transverse-parameter neighborhood. Express both matrices in the same dual quotient coordinates. The changes from real-normalized coordinates and between the two spinor charts are given by (7.3) and (7.4); their factors are constant under spatial differentiation. In particular, on a chart overlap, if S ζ = diag ( ζ I 2 , 1 ) S_\zeta=\operatorname{diag}(\zeta I_2,1) S ζ = diag ( ζ I 2 , 1 ) and H ~ 0 = S ζ − 1 H ∞ S ζ \widetilde H_0=S_\zeta^{-1}H_\infty S_\zeta H 0 = S ζ − 1 H ∞ S ζ , then
2 D θ 0 H ~ 0 + Γ 1 , 0 H ~ 0 − H ~ 0 Γ 2 , 0 = ζ 2 S ζ − 1 ( 2 D θ ∞ H ∞ + Γ 1 , ∞ H ∞ − H ∞ Γ 2 , ∞ ) S ζ = 0. 2D_{\theta_0}\widetilde H_0+\Gamma_{1,0}\widetilde H_0-\widetilde H_0\Gamma_{2,0}
=\zeta^2S_\zeta^{-1}(2D_{\theta_\infty}H_\infty+\Gamma_{1,\infty}H_\infty-H_\infty\Gamma_{2,\infty})S_\zeta=0. 2 D θ 0 H 0 + Γ 1 , 0 H 0 − H 0 Γ 2 , 0 = ζ 2 S ζ − 1 ( 2 D θ ∞ H ∞ + Γ 1 , ∞ H ∞ − H ∞ Γ 2 , ∞ ) S ζ = 0.
Both local constructions therefore solve the same equation in these coordinates and equal I I I on the same exterior shell. Each connected section B R ∩ { ν θ ⋅ x = t } B_R \cap\{\nu_\theta\cdot x=t\} B R ∩ { ν θ ⋅ x = t } is a disk and meets the common exterior normalization shell E e x t = { R 0 < ∣ x ∣ < R } E_{\mathrm{ext}}=\{R_0<|x|<R\} E ext = { R 0 < ∣ x ∣ < R } in a nonempty relatively open set. Applying Lemma 7.2 on that section shows that the two constructions agree wherever both are defined. Independence of the solution frames on a common disk is also the explicit covariance (201) ; this is distinct from the spinor coordinate conjugation used above.
Choose R 0 < R ⊥ < R R_0<R_\perp<R R 0 < R ⊥ < R , use the local construction for ∣ ν θ ⋅ x ∣ < R ⊥ |\nu_\theta\cdot x|<R_\perp ∣ ν θ ⋅ x ∣ < R ⊥ , and set H = I H=I H = I where ∣ ν θ ⋅ x ∣ > R 0 |\nu_\theta\cdot x|>R_0 ∣ ν θ ⋅ x ∣ > R 0 . These sets cover B R B_R B R , and on their overlap H = I H=I H = I by the exterior argument. The local constructions are smooth in spatial and real direction parameters; agreement on all intersections hence patches them to a jointly smooth bundle field over B R × C P 1 B_R\times\mathbb{CP}^1 B R × CP 1 . Formula (202) is invariant under these coordinate conjugations and makes the patched field an automorphism. The hypotheses of Proposition 7.6 now hold, so its independent fixed-point argument establishes holomorphic dependence on direction and the transition law (7.4). □
Theorem 6.13 , together with the frame and right-inverse constructions, supplies the frames and moment identities used in this proposition. The resulting field is constructed entirely from the identities on the original physical domain.
Uniqueness and holomorphic dependence on direction
Proposition 7.6. Suppose H H H is a jointly smooth invertible relative field on the fixed physical ball B R B_R B R , with
2 D θ H + Γ 1 H − H Γ 2 = 0 , H = I on E e x t = { R 0 < ∣ x ∣ < R } 2D_\theta H+\Gamma_1H-H\Gamma_2=0,\qquad H=I\quad\text{on }E_{\mathrm{ext}}=\{R_0<|x|<R\} 2 D θ H + Γ 1 H − H Γ 2 = 0 , H = I on E ext = { R 0 < ∣ x ∣ < R } for every direction. If Γ j \Gamma_j Γ j are given by (193) , then H H H is holomorphic in the two affine spinor charts for each fixed physical point x x x and satisfies (7.4).
Proof. After the changes of frame in (7.3) and (7.4), all candidates satisfy the same spatial relative equation and equal I I I on the same shell. Their difference satisfies a homogeneous relative ∂ ˉ \bar\partial ∂ ˉ system on each connected leaf section. The frames required in Lemma 7.2 exist by the planar frame construction. Every such section meets E e x t E_{\mathrm{ext}} E ext in a nonempty relatively open set, so Lemma 7.2 proves uniqueness and the stated transition law.
In a holomorphic spinor chart, the three nonzero blocks of (193) have respective direction degrees 2 , 1 , 3 2,1,3 2 , 1 , 3 , and the material coefficients depend only on x x x . Hence ∂ ζ ˉ Γ j = 0 \partial_{\bar\zeta}\Gamma_j=0 ∂ ζ ˉ Γ j = 0 and ∂ ζ ˉ θ = 0 \partial_{\bar\zeta}\theta=0 ∂ ζ ˉ θ = 0 at fixed x x x . Joint smoothness allows us to differentiate the spatial equation at a fixed physical point x x x , after returning to the holomorphic spinor representative by (7.3). The derivative is taken after returning to the holomorphic spinor representative, so the normalization factor is absent from the holomorphic-chart equation. Since partial derivatives in x x x and ζ ˉ \bar\zeta ζ ˉ commute in this chart, the matrix V = ∂ ζ ˉ H ∣ x V=\partial_{\bar\zeta}H|_x V = ∂ ζ ˉ H ∣ x satisfies
2 D θ V + Γ 1 V − V Γ 2 = 0 , V ∣ E e x t = 0. 2D_\theta V+\Gamma_1V-V\Gamma_2=0,\qquad V\big|_{E_{\mathrm{ext}}}=0. 2 D θ V + Γ 1 V − V Γ 2 = 0 , V E ext = 0.
Let X 1 X_1 X 1 and Y 2 Y_2 Y 2 be invertible primal and dual frames on a leaf, expressed in the current holomorphic spinor representative. Their equations and the equation for V V V give D θ ( X 1 T V Y 2 ) = 0 D_\theta(X_1^{T}VY_2)=0 D θ ( X 1 T V Y 2 ) = 0 . This matrix is therefore holomorphic in the leaf variable and vanishes on the nonempty relatively open intersection with E e x t E_{\mathrm{ext}} E ext . The identity theorem gives V = 0 V=0 V = 0 on the connected leaf section, in each full affine chart. □
Finite-type rigidity for the Lamé parameters
Propositions 7.5 and 7.6 give a relative transport matrix with a common exterior normalization and holomorphic dependence on direction at fixed physical points. We now prove that these properties determine the Lamé parameters. Lemma 7.8 first restricts the matrix blocks by their chart transition laws. The transport equation then determines all blocks in terms of ℓ 12 = log ( μ 1 / μ 2 ) \ell_{12}=\log(\mu_1/\mu_2) ℓ 12 = log ( μ 1 / μ 2 ) and yields the trace-free Hessian identity (211) . Differentiating this identity once gives a closed homogeneous first-order system for the three components of ∇ ℓ 12 \nabla\ell_{12} ∇ ℓ 12 and the scalar Δ ℓ 12 \Delta\ell_{12} Δ ℓ 12 . Zero data on the normalization set propagate through this system.
Let O ⊂ R 3 \mathcal{O} \subset\mathbb{R}^{3} O ⊂ R 3 be connected and open and let E e q ⊂ O E_{\mathrm{eq}} \subset\mathcal{O} E eq ⊂ O be nonempty and open. For j = 1 , 2 j=1,2 j = 1 , 2 , let μ j , κ j \mu_j,\kappa_j μ j , κ j be real smooth functions, with μ j > 0 \mu_j>0 μ j > 0 and 1 / 4 < κ j < 1 1/4<\kappa_j<1 1/4 < κ j < 1 , and assume ( μ 1 , κ 1 ) = ( μ 2 , κ 2 ) (\mu_1,\kappa_1)=(\mu_2,\kappa_2) ( μ 1 , κ 1 ) = ( μ 2 , κ 2 ) on E e q E_{\mathrm{eq}} E eq . Thus E e q E_{\mathrm{eq}} E eq is the open equality set used in the rigidity argument. Use ℓ j , m j , c j , F μ j \ell_j,m_j,c_j,F_{\mu_j} ℓ j , m j , c j , F μ j as in (60) and (63) , and put d j = κ j μ j d_j=\kappa_j\sqrt{\mu_j} d j = κ j μ j . We retain the tensor and gradient conventions used in the reduction and the bilinear pairing convention of § 1.6 . For every nonzero spinor ( s 0 , s 1 ) ∈ C 2 (s_0,s_1)\in\mathbb{C}^{2} ( s 0 , s 1 ) ∈ C 2 , define
θ = ( s 0 2 − s 1 2 , i ( s 0 2 + s 1 2 ) , 2 s 0 s 1 ) T , Q = 1 2 ( − s 1 − s 0 i s 1 − i s 0 s 0 − s 1 ) , n s p = ( − s 1 , s 0 ) T , R a = − 2 a T Q . (203) \begin{aligned}
\theta&=\left(s_0^2-s_1^2,\mathrm{i}(s_0^2+s_1^2),2s_0s_1\right)^{T},\\
Q&=\frac{1}{2}\begin{pmatrix}
-s_1 & -s_0\\
\mathrm{i}s_1 & -\mathrm{i}s_0\\
s_0 & -s_1
\end{pmatrix},\\
n_{\mathrm{sp}}&=(-s_1,s_0)^{T},\qquad R_a=-2a^{T}Q. \tag*{(203)}
\end{aligned} θ Q n sp = ( s 0 2 − s 1 2 , i ( s 0 2 + s 1 2 ) , 2 s 0 s 1 ) T , = 2 1 − s 1 i s 1 s 0 − s 0 − i s 0 − s 1 , = ( − s 1 , s 0 ) T , R a = − 2 a T Q . (203)
Direct multiplication gives θ T θ = 0 \theta^{T}\theta=0 θ T θ = 0 , θ T Q = 0 \theta^{T}Q=0 θ T Q = 0 , Q n s p = − θ / 2 Qn_{\mathrm{sp}}=-\theta/2 Q n sp = − θ /2 and R a n s p = a ⋅ θ R_an_{\mathrm{sp}}=a\cdot\theta R a n sp = a ⋅ θ . The three minors of Q Q Q are i s 0 s 1 / 2 \mathrm{i}s_0s_1/2 i s 0 s 1 /2 , ( s 0 2 + s 1 2 ) / 4 (s_0^2+s_1^2)/4 ( s 0 2 + s 1 2 ) /4 , i ( s 0 2 − s 1 2 ) / 4 \mathrm{i}(s_0^2-s_1^2)/4 i ( s 0 2 − s 1 2 ) /4 ; they do not vanish simultaneously. Thus Q Q Q has rank two, including at s 0 = 0 s_0=0 s 0 = 0 and at s 1 = 0 s_1=0 s 1 = 0 .
Theorem 7.7. Suppose there is an invertible dual quotient automorphism H H H on O × C P 1 \mathcal{O}\times\mathbb{CP}^{1} O × CP 1 , jointly smooth in local spatial and real direction coordinates, holomorphic in projective direction for every fixed physical x x x , and equal to I 3 I_3 I 3 on E e q E_{\mathrm{eq}} E eq for every direction. Suppose its matrices in the two full affine spinor charts ( 1 , ζ ) (1,\zeta) ( 1 , ζ ) and ( υ , 1 ) (\upsilon,1) ( υ , 1 ) obey
H 0 ( ζ ) = S ζ − 1 H ∞ ( ζ − 1 ) S ζ , S ζ = diag ( ζ I 2 , 1 ) , H_0(\zeta)=S_\zeta^{-1}H_\infty(\zeta^{-1})S_\zeta,\qquad S_\zeta=\operatorname{diag}(\zeta I_2,1), H 0 ( ζ ) = S ζ − 1 H ∞ ( ζ − 1 ) S ζ , S ζ = diag ( ζ I 2 , 1 ) , 2 D θ H + Γ 1 H − H Γ 2 = 0 , Γ j = ( n s p R m j 2 d j n s p − θ T F μ j Q 0 ) , D θ = θ ⋅ ∇ x . (204) 2D_\theta H+\Gamma_1H-H\Gamma_2=0,\qquad
\Gamma_j=\begin{pmatrix}
n_{\mathrm{sp}}R_{m_j} & 2d_jn_{\mathrm{sp}}\\
-\theta^{T}F_{\mu_j}Q & 0
\end{pmatrix},\qquad D_\theta=\theta\cdot\nabla_x. \tag*{(204)} 2 D θ H + Γ 1 H − H Γ 2 = 0 , Γ j = ( n sp R m j − θ T F μ j Q 2 d j n sp 0 ) , D θ = θ ⋅ ∇ x . (204) Then μ 1 = μ 2 \mu_1=\mu_2 μ 1 = μ 2 , κ 1 = κ 2 \kappa_1=\kappa_2 κ 1 = κ 2 and H = I 3 H=I_3 H = I 3 on O \mathcal{O} O . Consequently, if λ j = μ j ( 2 κ j − 1 ) / ( 1 − κ j ) \lambda_j=\mu_j(2\kappa_j-1)/(1-\kappa_j) λ j = μ j ( 2 κ j − 1 ) / ( 1 − κ j ) , then λ 1 = λ 2 \lambda_1=\lambda_2 λ 1 = λ 2 .
Lemma 7.8. Under the direction and chart hypotheses of Theorem 7.7 ,
H = ( A 0 C δ H ) , C = s 0 C s 0 + s 1 C s 1 , (205) H=\begin{pmatrix} A & 0\\ C & \delta_H \end{pmatrix},\qquad C=s_0C_{s_0}+s_1C_{s_1}, \tag*{(205)} H = ( A C 0 δ H ) , C = s 0 C s 0 + s 1 C s 1 , (205) Here A A A is a 2 × 2 2\times2 2 × 2 invertible matrix, δ H \delta_H δ H is the nonzero lower-right scalar block of H H H , and C s 0 , C s 1 C_{s_0},C_{s_1} C s 0 , C s 1 are 1 × 2 1\times2 1 × 2 row vectors. All are smooth functions of the spatial variable; A A A and δ H \delta_H δ H are independent of direction, while the row C C C is linear in the homogeneous spinor coordinates.
Proof. Write the chart blocks as ( A , B ; C , δ H ) (A,B;C,\delta_H) ( A , B ; C , δ H ) . Their transition laws are A 0 = A ∞ A_0=A_\infty A 0 = A ∞ , δ H , 0 = δ H , ∞ \delta_{H,0}=\delta_{H,\infty} δ H , 0 = δ H , ∞ , B 0 = ζ − 1 B ∞ B_0=\zeta^{-1}B_\infty B 0 = ζ − 1 B ∞ , C 0 = ζ C ∞ C_0=\zeta C_\infty C 0 = ζ C ∞ , with infinity entries evaluated at ζ − 1 \zeta^{-1} ζ − 1 . Holomorphy on each full affine chart implies, for large ∣ ζ ∣ |\zeta| ∣ ζ ∣ , bounds A 0 , δ H , 0 = O ( 1 ) A_0,\delta_{H,0}=O(1) A 0 , δ H , 0 = O ( 1 ) , B 0 = O ( ∣ ζ ∣ − 1 ) B_0=O(|\zeta|^{-1}) B 0 = O ( ∣ ζ ∣ − 1 ) and C 0 = O ( 1 + ∣ ζ ∣ ) C_0=O(1+|\zeta|) C 0 = O ( 1 + ∣ ζ ∣ ) . For an entire scalar entry f f f satisfying ∣ f ( ζ ) ∣ ≤ C ( 1 + ∣ ζ ∣ ) m |f(\zeta)|\leq C(1+|\zeta|)^m ∣ f ( ζ ) ∣ ≤ C ( 1 + ∣ ζ ∣ ) m with m = 0 m=0 m = 0 or 1 1 1 , the Cauchy formula on the circle of radius R R R centered at any fixed ζ ∗ \zeta_* ζ ∗ gives
∣ f ( m + 1 ) ( ζ ∗ ) ∣ ≤ ( m + 1 ) ! C ( 1 + ∣ ζ ∗ ∣ + R ) m R − m − 1 ⟶ 0. \left|f^{(m+1)}(\zeta_*)\right|\leq(m+1)!C(1+|\zeta_*|+R)^mR^{-m-1}\longrightarrow0. f ( m + 1 ) ( ζ ∗ ) ≤ ( m + 1 )! C ( 1 + ∣ ζ ∗ ∣ + R ) m R − m − 1 ⟶ 0.
Thus A 0 , δ H , 0 A_0,\delta_{H,0} A 0 , δ H , 0 are constant and C 0 C_0 C 0 is affine. The entire entries of B 0 B_0 B 0 are bounded and tend to zero at infinity, so the same argument with m = 0 m=0 m = 0 makes them identically zero. Consequently B 0 = 0 B_0=0 B 0 = 0 and C 0 = C s 0 + ζ C s 1 C_0=C_{s_0}+\zeta C_{s_1} C 0 = C s 0 + ζ C s 1 . The infinity expression is C ∞ = υ C s 0 + C s 1 C_\infty=\upsilon C_{s_0}+C_{s_1} C ∞ = υ C s 0 + C s 1 . The spatial coefficients are smooth, as is seen by evaluation at fixed chart points. The determinant is ( det A ) δ H (\det A)\delta_H ( det A ) δ H , so invertibility gives the asserted nonvanishing.
The dual ambient frame is U θ = diag ( Q , 1 ) U_\theta=\operatorname{diag}(Q,1) U θ = diag ( Q , 1 ) . For b ∈ C ∖ { 0 } b\in\mathbb{C}\setminus\{0\} b ∈ C ∖ { 0 } , put S b = diag ( b I 2 , 1 ) S_b=\operatorname{diag}(bI_2,1) S b = diag ( b I 2 , 1 ) . Replacing the spinor ( s 0 , s 1 ) (s_0,s_1) ( s 0 , s 1 ) by b ( s 0 , s 1 ) b(s_0,s_1) b ( s 0 , s 1 ) changes the frame to U θ S b U_\theta S_b U θ S b , so the matrix of the same dual quotient map changes to S b − 1 H S b S_b^{-1}HS_b S b − 1 H S b . The resulting homogeneities are
( Q , n s p , C ) ↦ b ( Q , n s p , C ) , θ ↦ b 2 θ , ( A , δ H ) ↦ ( A , δ H ) , (Q,n_{\mathrm{sp}},C)\mapsto b(Q,n_{\mathrm{sp}},C),\qquad\theta\mapsto b^2\theta,\qquad(A,\delta_H)\mapsto(A,\delta_H), ( Q , n sp , C ) ↦ b ( Q , n sp , C ) , θ ↦ b 2 θ , ( A , δ H ) ↦ ( A , δ H ) ,
( Γ j 11 , Γ j 12 , Γ j 21 ) ↦ ( b 2 Γ j 11 , b Γ j 12 , b 3 Γ j 21 ) . (206) (\Gamma^{11}_j,\Gamma^{12}_j,\Gamma^{21}_j)\mapsto(b^2\Gamma^{11}_j,b\Gamma^{12}_j,b^3\Gamma^{21}_j). \tag*{(206)} ( Γ j 11 , Γ j 12 , Γ j 21 ) ↦ ( b 2 Γ j 11 , b Γ j 12 , b 3 Γ j 21 ) . (206)
where Γ j 11 = n s p R m j \Gamma^{11}_j=n_{\mathrm{sp}}R_{m_j} Γ j 11 = n sp R m j , Γ j 12 = 2 d j n s p \Gamma^{12}_j=2d_jn_{\mathrm{sp}} Γ j 12 = 2 d j n sp and Γ j 21 = − θ T F μ j Q \Gamma^{21}_j=-\theta^TF_{\mu_j}Q Γ j 21 = − θ T F μ j Q . Thus the block equations have the induced spinor homogeneities.
Lemma 7.9. Set ℓ 12 = log ( μ 1 / μ 2 ) \ell_{12}=\log(\mu_1/\mu_2) ℓ 12 = log ( μ 1 / μ 2 ) and r = ∇ ℓ 12 r=\nabla\ell_{12} r = ∇ ℓ 12 . Then
A = I 2 , δ H = e − ℓ 12 / 2 , κ 1 = κ 2 = : κ , C = − R r 2 d 1 = g T Q , g = c 1 κ r . (207) A=I_2,\qquad\delta_H=e^{-\ell_{12}/2},\qquad\kappa_1=\kappa_2=:\kappa,\qquad C=-\frac{R_r}{2d_1}=g^TQ,\qquad\mathbf{g}=\frac{c_1}{\kappa}r. \tag*{(207)} A = I 2 , δ H = e − ℓ 12 /2 , κ 1 = κ 2 =: κ , C = − 2 d 1 R r = g T Q , g = κ c 1 r . (207)
Proof. Multiplication of the matrices in their given material order yields
2 D θ A + Γ 1 11 A + Γ 1 12 C − A Γ 2 11 = 0 , Γ 1 12 δ H − A Γ 2 12 = 0 , 2 D θ C + Γ 1 21 A − C Γ 2 11 − δ H Γ 2 21 = 0 , 2 D θ δ H − C Γ 2 12 = 0. (208) \begin{aligned}
2D_\theta A+\Gamma^{11}_1A+\Gamma^{12}_1C-A\Gamma^{11}_2&=0,\\
\Gamma^{12}_1\delta_H-A\Gamma^{12}_2&=0,\\
2D_\theta C+\Gamma^{21}_1A-C\Gamma^{11}_2-\delta_H\Gamma^{21}_2&=0,\\
2D_\theta\delta_H-C\Gamma^{12}_2&=0. \tag*{(208)}
\end{aligned} 2 D θ A + Γ 1 11 A + Γ 1 12 C − A Γ 2 11 Γ 1 12 δ H − A Γ 2 12 2 D θ C + Γ 1 21 A − C Γ 2 11 − δ H Γ 2 21 2 D θ δ H − C Γ 2 12 = 0 , = 0 , = 0 , = 0. (208)
Their homogeneous spinor degrees are respectively 2, 1, 3, 2, so the equations in both affine charts hold for every homogeneous nonzero spinor. These spinors cover every complex null ray: when θ 1 − i θ 2 ≠ 0 \theta_1-i\theta_2\ne0 θ 1 − i θ 2 = 0 the chart parameter is θ 3 / ( θ 1 − i θ 2 ) \theta_3/(\theta_1-i\theta_2) θ 3 / ( θ 1 − i θ 2 ) , and the exceptional ray is represented by ( 0 , 1 ) (0,1) ( 0 , 1 ) . The second equation gives d 1 δ H n s p = d 2 A n s p d_1\delta_Hn_{\mathrm{sp}}=d_2An_{\mathrm{sp}} d 1 δ H n sp = d 2 A n sp . Since n s p ( 1 , 0 ) = ( 0 , 1 ) T n_{\mathrm{sp}}(1,0)=(0,1)^T n sp ( 1 , 0 ) = ( 0 , 1 ) T and n s p ( 0 , 1 ) = ( − 1 , 0 ) T n_{\mathrm{sp}}(0,1)=(-1,0)^T n sp ( 0 , 1 ) = ( − 1 , 0 ) T , it follows that A = α I 2 A=\alpha I_2 A = α I 2 with α = d 1 δ H / d 2 \alpha=d_1\delta_H/d_2 α = d 1 δ H / d 2 . The first equation becomes
2 ( D θ α ) I 2 + n s p { α ( R m 1 − R m 2 ) + 2 d 1 C } = 0. 2(D_\theta\alpha)I_2+n_{\mathrm{sp}}\{\alpha(R_{m_1}-R_{m_2})+2d_1C\}=0. 2 ( D θ α ) I 2 + n sp { α ( R m 1 − R m 2 ) + 2 d 1 C } = 0.
The second term has rank at most one, whereas a nonzero multiple of I 2 I_2 I 2 has rank two. Hence D θ α = 0 D_\theta\alpha=0 D θ α = 0 for every null direction. Since 2 e i = ( e i + i e j ) + ( e i − i e j ) 2e_i=(e_i+ie_j)+(e_i-ie_j) 2 e i = ( e i + i e j ) + ( e i − i e j ) for j ≠ i j\ne i j = i , ∇ α = 0 \nabla\alpha=0 ∇ α = 0 . Connectedness and the data on E e q E_{\mathrm{eq}} E eq give α = 1 \alpha=1 α = 1 . The nonzero column n s p n_{\mathrm{sp}} n sp then gives C = − R r / ( 2 d 1 ) C=-R_r/(2d_1) C = − R r / ( 2 d 1 ) and δ H = d 2 / d 1 > 0 \delta_H=d_2/d_1>0 δ H = d 2 / d 1 > 0 .
The fourth equation, using R r n s p = D θ ℓ 12 R_rn_{\mathrm{sp}}=D_\theta\ell_{12} R r n sp = D θ ℓ 12 , now reads 2 D θ δ H + δ H D θ ℓ 12 = 0 2D_\theta\delta_H+\delta_HD_\theta\ell_{12}=0 2 D θ δ H + δ H D θ ℓ 12 = 0 . Therefore δ H e ℓ 12 / 2 \delta_He^{\ell_{12}/2} δ H e ℓ 12 /2 is constant on O \mathcal{O} O , equal to one on E e q E_{\mathrm{eq}} E eq . Thus δ H = e − ℓ 12 / 2 \delta_H=e^{-\ell_{12}/2} δ H = e − ℓ 12 /2 . Since d 2 / d 1 = ( κ 2 / κ 1 ) e − ℓ 12 / 2 d_2/d_1=(\kappa_2/\kappa_1)e^{-\ell_{12}/2} d 2 / d 1 = ( κ 2 / κ 1 ) e − ℓ 12 /2 , cancellation gives the common κ \kappa κ . Finally − R r / ( 2 d 1 ) = ( c 1 / κ ) r T Q -R_r/(2d_1)=(c_1/\kappa)r^TQ − R r / ( 2 d 1 ) = ( c 1 / κ ) r T Q .
Lemma 7.10. The third equation of (208) is equivalent to Q T K r i g θ = 0 Q^TK_{\mathrm{rig}}\theta=0 Q T K rig θ = 0 , where
K r i g = 2 ∇ g − F 1 + e − ℓ 12 / 2 F 2 − m 2 g T . (209) K_{\mathrm{rig}}=2\nabla\mathbf{g}-F_1+e^{-\ell_{12}/2}F_2-\mathbf{m}_2\mathbf{g}^T. \tag*{(209)} K rig = 2∇ g − F 1 + e − ℓ 12 /2 F 2 − m 2 g T . (209) Here K r i g K_{\mathrm{rig}} K rig is the 3 × 3 3\times3 3 × 3 matrix used in the rigidity argument, distinct from the four-component differential operator K l i f t K_{\mathrm{lift}} K lift . For every x x x , K r i g ( x ) K_{\mathrm{rig}}(x) K rig ( x ) is a scalar multiple of I 3 I_3 I 3 .
Proof. With C = g T Q C=\mathbf{g}^TQ C = g T Q , the third equation becomes
θ T { 2 ( ∇ g ) T − F 1 + e − ℓ 12 / 2 F 2 − g m 2 T } Q = 0. \theta^T\{2(\nabla\mathbf{g})^T-F_1+e^{-\ell_{12}/2}F_2-\mathbf{g}\mathbf{m}_2^T\}Q=0. θ T { 2 ( ∇ g ) T − F 1 + e − ℓ 12 /2 F 2 − g m 2 T } Q = 0. Indeed, D θ C = θ T ( ∇ g ) T Q D_\theta C=\theta^T(\nabla\mathbf{g})^TQ D θ C = θ T ( ∇ g ) T Q and − C n s p R m 2 = − ( g ⋅ θ ) m 2 T Q -Cn_{\mathrm{sp}}R_{m_2}=-(\mathbf{g}\cdot\theta)\mathbf{m}_2^TQ − C n sp R m 2 = − ( g ⋅ θ ) m 2 T Q . Transposition gives (209) . Since ran Q = θ ⊥ \operatorname{ran}Q=\theta^\perp ran Q = θ ⊥ and the ambient complex bilinear form is nondegenerate, ker Q T = C θ \ker Q^T=\mathbb{C}\theta ker Q T = C θ , even for a null θ \theta θ . Thus K r i g θ ∈ C θ K_{\mathrm{rig}}\theta\in\mathbb{C}\theta K rig θ ∈ C θ for all null θ \theta θ . At each fixed x x x , K r i g K_{\mathrm{rig}} K rig is independent of the direction variable. For distinct i , j , k i,j,k i , j , k , the k k k components of K r i g ( e i ± i e j ) K_{\mathrm{rig}}(e_i\pm ie_j) K rig ( e i ± i e j ) give ( K r i g ) k i ± i ( K r i g ) k j = 0 (K_{\mathrm{rig}})_{ki}\pm i(K_{\mathrm{rig}})_{kj}=0 ( K rig ) k i ± i ( K rig ) k j = 0 , hence both entries vanish. Choosing each remaining index removes all six off-diagonal entries. Preservation of C ( e i + i e j ) C(e_i+ie_j) C ( e i + i e j ) then makes the diagonal entries equal. Therefore K r i g = q I 3 K_{\mathrm{rig}}=qI_3 K rig = q I 3 , with real smooth q = tr K r i g / 3 q=\operatorname{tr}K_{\mathrm{rig}}/3 q = tr K rig /3 .
Proposition 7.11. Let k κ = ∇ log κ \mathbf{k}_{\kappa}=\nabla\log\kappa k κ = ∇ log κ , S = m 1 + m 2 S=\mathbf{m}_{1}+\mathbf{m}_{2} S = m 1 + m 2 and Ξ = Δ ℓ 12 − S ⋅ r \Xi=\Delta\ell_{12}-S\cdot r Ξ = Δ ℓ 12 − S ⋅ r . Then the full matrix identity is
K r i g c 1 = 2 ( κ − 1 − 1 ) ∇ 2 ℓ 12 + ( 1 − ( 2 κ ) − 1 ) ( S ⊗ r + r ⊗ S ) − 2 κ − 1 r ⊗ k + 2 Ξ I 3 . (210) \frac{K_{\mathrm{rig}}}{c_{1}}=2(\kappa^{-1}-1)\nabla^{2}\ell_{12}+\left(1-(2\kappa)^{-1}\right)(S\otimes r+r\otimes S)-2\kappa^{-1}r\otimes\mathbf{k}+2\Xi I_{3}. \tag*{(210)} c 1 K rig = 2 ( κ − 1 − 1 ) ∇ 2 ℓ 12 + ( 1 − ( 2 κ ) − 1 ) ( S ⊗ r + r ⊗ S ) − 2 κ − 1 r ⊗ k + 2Ξ I 3 . (210) Consequently, with tf B = B − ( tr B ) I 3 / 3 \operatorname{tf}B=B-(\operatorname{tr}B)I_{3}/3 tf B = B − ( tr B ) I 3 /3 ,
tf ∇ 2 ℓ 12 = tf ( β ⊗ r + r ⊗ β ) , β = 2 k − ( 2 κ − 1 ) ( m 1 + m 2 ) 4 ( 1 − κ ) . (211) \operatorname{tf}\nabla^{2}\ell_{12}=\operatorname{tf}(\beta\otimes r+r\otimes\beta),\qquad\beta=\frac{2\mathbf{k}-(2\kappa-1)(\mathbf{m}_{1}+\mathbf{m}_{2})}{4(1-\kappa)}. \tag*{(211)} tf ∇ 2 ℓ 12 = tf ( β ⊗ r + r ⊗ β ) , β = 4 ( 1 − κ ) 2 k − ( 2 κ − 1 ) ( m 1 + m 2 ) . (211) Proof. Since ∇ log ( c 1 / κ ) = − m 1 / 2 − k \nabla\log(c_{1}/\kappa)=-\mathbf{m}_{1}/2-\mathbf{k} ∇ log ( c 1 / κ ) = − m 1 /2 − k and e − ℓ 12 / 2 c 2 = c 1 e^{-\ell_{12}/2}c_{2}=c_{1} e − ℓ 12 /2 c 2 = c 1 , direct expansion gives
2 ∇ g = 2 c 1 κ ∇ 2 ℓ 12 − c 1 κ r ⊗ m 1 − 2 c 1 κ r ⊗ k , 2\nabla g=\frac{2c_{1}}{\kappa}\nabla^{2}\ell_{12}-\frac{c_{1}}{\kappa}r\otimes\mathbf{m}_{1}-\frac{2c_{1}}{\kappa}r\otimes\mathbf{k}, 2∇ g = κ 2 c 1 ∇ 2 ℓ 12 − κ c 1 r ⊗ m 1 − κ 2 c 1 r ⊗ k , − F 1 + e − ℓ 12 / 2 F 2 = 2 c 1 { − ∇ 2 ℓ 12 + m 1 ⊗ m 1 − m 2 ⊗ m 2 + Ξ I 3 } . -F_{1}+e^{-\ell_{12}/2}F_{2}=2c_{1}\{-\nabla^{2}\ell_{12}+\mathbf{m}_{1}\otimes\mathbf{m}_{1}-\mathbf{m}_{2}\otimes\mathbf{m}_{2}+\Xi I_{3}\}. − F 1 + e − ℓ 12 /2 F 2 = 2 c 1 { − ∇ 2 ℓ 12 + m 1 ⊗ m 1 − m 2 ⊗ m 2 + Ξ I 3 } . Because r = m 1 − m 2 r=\mathbf{m}_{1}-\mathbf{m}_{2} r = m 1 − m 2 , both m 1 ⊗ m 1 − m 2 ⊗ m 2 \mathbf{m}_{1}\otimes\mathbf{m}_{1}-\mathbf{m}_{2}\otimes\mathbf{m}_{2} m 1 ⊗ m 1 − m 2 ⊗ m 2 and r ⊗ m 1 + m 2 ⊗ r r\otimes\mathbf{m}_{1}+\mathbf{m}_{2}\otimes r r ⊗ m 1 + m 2 ⊗ r equal ( S ⊗ r + r ⊗ S ) / 2 (S\otimes r+r\otimes S)/2 ( S ⊗ r + r ⊗ S ) /2 . Substitution in (209) proves (210) , including its scalar and skew terms. The skew part is − ( c 1 / κ ) ( r ⊗ k − k ⊗ r ) -(c_{1}/\kappa)(r\otimes\mathbf{k}-\mathbf{k}\otimes r) − ( c 1 / κ ) ( r ⊗ k − k ⊗ r ) and is zero because K r i g = q I 3 K_{\mathrm{rig}}=qI_{3} K rig = q I 3 . The symmetric trace-free part is
0 = 2 ( κ − 1 − 1 ) tf ∇ 2 ℓ 12 + ( 1 − ( 2 κ ) − 1 ) tf ( S ⊗ r + r ⊗ S ) − κ − 1 tf ( r ⊗ k + k ⊗ r ) . 0=2(\kappa^{-1}-1)\operatorname{tf}\nabla^{2}\ell_{12}+\left(1-(2\kappa)^{-1}\right)\operatorname{tf}(S\otimes r+r\otimes S)-\kappa^{-1}\operatorname{tf}(r\otimes\mathbf{k}+\mathbf{k}\otimes r). 0 = 2 ( κ − 1 − 1 ) tf ∇ 2 ℓ 12 + ( 1 − ( 2 κ ) − 1 ) tf ( S ⊗ r + r ⊗ S ) − κ − 1 tf ( r ⊗ k + k ⊗ r ) . Division by 2 ( 1 − κ ) / κ > 0 2(1-\kappa)/\kappa>0 2 ( 1 − κ ) / κ > 0 gives (211) . □
Lemma 7.12. Suppose ℓ 12 ∈ C 3 ( O ; R ) \ell_{12}\in C^{3}(\mathcal{O};\mathbb{R}) ℓ 12 ∈ C 3 ( O ; R ) , β ∈ C 1 ( O ; R 3 ) \beta\in C^{1}(\mathcal{O};\mathbb{R}^{3}) β ∈ C 1 ( O ; R 3 ) and (211) holds. Put r = ∇ ℓ 12 r=\nabla\ell_{12} r = ∇ ℓ 12 and w = Δ ℓ 12 w=\Delta\ell_{12} w = Δ ℓ 12 . The column U = ( r 1 , r 2 , r 3 , w ) T U=(r_{1},r_{2},r_{3},w)^{T} U = ( r 1 , r 2 , r 3 , w ) T collects the four scalar unknowns. In this lemma ∂ j = ∂ / ∂ x j \partial_{j}=\partial/\partial x_{j} ∂ j = ∂ / ∂ x j , δ i j \delta_{ij} δ ij is the Kronecker delta, and a repeated index is summed from 1 to 3. Define
T i j k = β i δ j k + β j δ i k − 2 3 δ i j β k . T_{ij}^{k}=\beta_{i}\delta_{jk}+\beta_{j}\delta_{ik}-\frac{2}{3}\delta_{ij}\beta_{k}. T ij k = β i δ j k + β j δ ik − 3 2 δ ij β k . The trace-free Hessian identity determines the derivatives of r r r up to the scalar w w w . Its differentiated equation determines the derivatives of w w w , giving U ∈ C 1 U\in C^{1} U ∈ C 1 and ∂ j U = C j U \partial_{j}U=C_{j}U ∂ j U = C j U with continuous 4 × 4 4\times4 4 × 4 coefficient matrices
C j = ( ( T i j k ) i , k = 1 3 ( δ i j / 3 ) i = 1 3 ( B j k ) k = 1 3 5 β j / 3 ) , B j k = 3 2 [ ( div β ) δ j k + ∂ k β j − 2 3 ∂ j β k + 1 3 ∣ β ∣ 2 δ j k + 1 9 β j β k ] . (212) C_{j}=\begin{pmatrix}(T_{ij}^{k})_{i,k=1}^{3}&(\delta_{ij}/3)_{i=1}^{3}\\(B_{j}^{k})_{k=1}^{3}&5\beta_{j}/3\end{pmatrix},\qquad B_{j}^{k}=\frac{3}{2}\left[(\operatorname{div}\beta)\delta_{jk}+\partial_{k}\beta_{j}-\frac{2}{3}\partial_{j}\beta_{k}+\frac{1}{3}|\beta|^{2}\delta_{jk}+\frac{1}{9}\beta_{j}\beta_{k}\right]. \tag*{(212)} C j = ( ( T ij k ) i , k = 1 3 ( B j k ) k = 1 3 ( δ ij /3 ) i = 1 3 5 β j /3 ) , B j k = 2 3 [ ( div β ) δ j k + ∂ k β j − 3 2 ∂ j β k + 3 1 ∣ β ∣ 2 δ j k + 9 1 β j β k ] . (212) Proof. Restoring the trace in (211) gives ∂ j r i = δ i j w / 3 + T i j k r k \partial_{j}r_{i}=\delta_{ij}w/3+T_{ij}^{k}r_{k} ∂ j r i = δ ij w /3 + T ij k r k . Differentiate in x i x_{i} x i and sum. Commutation of third derivatives gives
2 3 ∂ j w = ( ∂ i T i j k ) r k + T i j k ( 1 3 δ k i w + T k i ℓ r ℓ ) . \frac{2}{3}\partial_{j}w=(\partial_{i}T_{ij}^{k})r_{k}+T_{ij}^{k}\left(\frac{1}{3}\delta_{ki}w+T_{ki}^{\ell}r_{\ell}\right). 3 2 ∂ j w = ( ∂ i T ij k ) r k + T ij k ( 3 1 δ k i w + T k i ℓ r ℓ ) . The contractions are
∑ i T i j i = 10 3 β j , \sum_{i}T_{ij}^{i}=\frac{10}{3}\beta_{j}, i ∑ T ij i = 3 10 β j , ∑ i ∂ i T i j k = ( div β ) δ j k + ∂ k β j − 2 3 ∂ j β k , \sum_{i}\partial_{i}T_{ij}^{k}=(\operatorname{div}\beta)\delta_{jk}+\partial_{k}\beta_{j}-\frac{2}{3}\partial_{j}\beta_{k}, i ∑ ∂ i T ij k = ( div β ) δ j k + ∂ k β j − 3 2 ∂ j β k , ∑ i , ℓ T i j ℓ T ℓ i k = 1 3 ∑ i β i T i j i = 1 3 ∣ β ∣ 2 δ j k + 1 9 β j β k . \sum_{i,\ell}T_{ij}^{\ell}T_{\ell i}^{k}=\frac{1}{3}\sum_{i}\beta_{i}T_{ij}^{i}=\frac{1}{3}|\beta|^{2}\delta_{jk}+\frac{1}{9}\beta_{j}\beta_{k}. i , ℓ ∑ T ij ℓ T ℓ i k = 3 1 i ∑ β i T ij i = 3 1 ∣ β ∣ 2 δ j k + 9 1 β j β k . For the last equality, insert the three terms in T i j ℓ T_{ij}^{\ell} T ij ℓ ; the middle term vanishes by ∑ i T i i k = 0 \sum_{i}T_{ii}^{k}=0 ∑ i T ii k = 0 , and the first and last combine with coefficients 1 − 2 / 3 = 1 / 3 1-2/3=1/3 1 − 2/3 = 1/3 . Multiplication by 3 / 2 3/2 3/2 yields (212) . The coefficient matrices involve only β \beta β and its first derivatives, and the system is homogeneous. □
Proposition 7.13. Let ℓ 12 ∈ C 3 ( O ; R ) \ell_{12} \in C^{3}(\mathcal O;\mathbb R) ℓ 12 ∈ C 3 ( O ; R ) and β ∈ C 1 ( O ; R 3 ) \beta\in C^{1}(\mathcal O;\mathbb R^{3}) β ∈ C 1 ( O ; R 3 ) . For a general ℓ 12 \ell_{12} ℓ 12 , define the symmetric trace-free defect by
E = tf ∇ 2 ℓ 12 − tf ( β ⊗ r + r ⊗ β ) , r = ∇ ℓ 12 , U = ( r 1 , r 2 , r 3 , Δ ℓ 12 ) T . (213) \mathcal E=\operatorname{tf}\nabla^{2}\ell_{12}-\operatorname{tf}(\beta\otimes r+r\otimes\beta),\qquad r=\nabla\ell_{12},\qquad U=(r_{1},r_{2},r_{3},\Delta\ell_{12})^{T}. \tag*{(213)} E = tf ∇ 2 ℓ 12 − tf ( β ⊗ r + r ⊗ β ) , r = ∇ ℓ 12 , U = ( r 1 , r 2 , r 3 , Δ ℓ 12 ) T . (213) With the matrices C j C_{j} C j in (212) , one has
∂ j U = C j U + F j , F j = ( 3 2 ∑ i ∂ i E i j + 1 2 ∑ i β i E i j ) . (214) \partial_{j}U=C_{j}U+F_{j},\qquad F_{j}=\left(\frac{3}{2}\sum_{i}\partial_{i}\mathcal E_{ij}+\frac{1}{2}\sum_{i}\beta_{i}\mathcal E_{ij}\right). \tag*{(214)} ∂ j U = C j U + F j , F j = ( 2 3 i ∑ ∂ i E ij + 2 1 i ∑ β i E ij ) . (214) For a piecewise C 1 C^{1} C 1 path p : [ 0 , 1 ] → O \mathbf p:[0,1]\to\mathcal O p : [ 0 , 1 ] → O , write
a p ( t ) = ∫ 0 t ∑ j ∣ p ˙ j ( s ) ∣ ∥ C j ( p ( s ) ) ∥ d s , b p ( t ) = ∑ j ∣ p ˙ j ( t ) ∣ ∣ F j ( p ( t ) ) ∣ , a_{\mathbf p}(t)=\int_{0}^{t}\sum_{j}|\dot p_{j}(s)|\lVert C_{j}(\mathbf p(s))\rVert\,\mathrm ds,\qquad b_{\mathbf p}(t)=\sum_{j}|\dot p_{j}(t)||F_{j}(\mathbf p(t))|, a p ( t ) = ∫ 0 t j ∑ ∣ p ˙ j ( s ) ∣ ∥ C j ( p ( s ))∥ d s , b p ( t ) = j ∑ ∣ p ˙ j ( t ) ∣∣ F j ( p ( t )) ∣ , where vector norms are Euclidean and matrix norms are the induced operator norms. Then
∣ U ( p ( t ) ) ∣ ≤ e a p ( t ) ∣ U ( p ( 0 ) ) ∣ + ∫ 0 t e a p ( t ) − a p ( s ) b p ( s ) d s , 0 ≤ t ≤ 1. (215) |U(\mathbf p(t))|\leq e^{a_{\mathbf p}(t)}|U(\mathbf p(0))|+\int_{0}^{t}e^{a_{\mathbf p}(t)-a_{\mathbf p}(s)}b_{\mathbf p}(s)\,\mathrm ds,\qquad0\leq t\leq1. \tag*{(215)} ∣ U ( p ( t )) ∣ ≤ e a p ( t ) ∣ U ( p ( 0 )) ∣ + ∫ 0 t e a p ( t ) − a p ( s ) b p ( s ) d s , 0 ≤ t ≤ 1. (215) In particular, if O \mathcal O O is connected, E = 0 \mathcal E=0 E = 0 , and U ( x 0 ) = 0 U(x_{0})=0 U ( x 0 ) = 0 at one point, then U = 0 U=0 U = 0 on O \mathcal O O .
Proof. Put w = Δ ℓ 12 w=\Delta\ell_{12} w = Δ ℓ 12 . Restoring the trace in (213) gives
∂ j r i = δ i j w / 3 + T i j k r k + E i j . \partial_{j}r_{i}=\delta_{ij}w/3+T_{ij}^{k}r_{k}+\mathcal E_{ij}. ∂ j r i = δ ij w /3 + T ij k r k + E ij .
Differentiate in x i x_{i} x i , sum, and commute the third derivatives of ℓ 12 \ell_{12} ℓ 12 . The additional terms compared with the proof of Lemma 7.12 are ∂ i E i j + T i j k E k i \partial_{i}\mathcal E_{ij}+T_{ij}^{k}\mathcal E_{ki} ∂ i E ij + T ij k E k i on the right-hand side of the equation for 2 ∂ j w / 3 2\partial_{j}w/3 2 ∂ j w /3 . Symmetry and zero trace of E \mathcal E E give the exact contraction
∑ i , k T i j k E k i = ∑ i β i E j i + β j tr E − 2 3 ∑ k β k E k j = 1 3 ∑ i β i E i j . \sum_{i,k}T_{ij}^{k}\mathcal E_{ki}=\sum_{i}\beta_{i}\mathcal E_{ji}+\beta_{j}\operatorname{tr}\mathcal E-\frac{2}{3}\sum_{k}\beta_{k}\mathcal E_{kj}=\frac{1}{3}\sum_{i}\beta_{i}\mathcal E_{ij}. i , k ∑ T ij k E k i = i ∑ β i E j i + β j tr E − 3 2 k ∑ β k E k j = 3 1 i ∑ β i E ij .
Together with the contractions already proved in Lemma 7.12 , this yields (214) .
The compact image of p \mathbf p p lies in O \mathcal O O , so all coefficient and source terms along it are bounded. For V = U ∘ p V=U\circ\mathbf p V = U ∘ p , the chain rule gives, almost everywhere, V ′ = ∑ j p ˙ j ( C j V + F j ) V'=\sum_{j}\dot p_{j}(C_{j}V+F_{j}) V ′ = ∑ j p ˙ j ( C j V + F j ) . For ε > 0 \varepsilon>0 ε > 0 let v ε = ( ∣ V ∣ 2 + ε 2 ) 1 / 2 v_{\varepsilon}=(|V|^{2}+\varepsilon^{2})^{1/2} v ε = ( ∣ V ∣ 2 + ε 2 ) 1/2 . Then
v ε ′ ≤ a p ′ v ε + b p v_{\varepsilon}'\leq a_{\mathbf p}'v_{\varepsilon}+b_{\mathbf p} v ε ′ ≤ a p ′ v ε + b p
almost everywhere. Multiplying by e − a p e^{-a_{\mathbf p}} e − a p , integrating over the finitely many smooth path pieces, and using continuity at their endpoints proves (215) upon letting ε ↓ 0 \varepsilon\downarrow0 ε ↓ 0 . Every two points of a connected open subset of Euclidean space can be joined by a polygonal path: the points reachable from x 0 x_{0} x 0 by such paths form a nonempty relatively open set whose complement is also relatively open, by the existence of small interior balls. If E = 0 \mathcal E=0 E = 0 and U ( x 0 ) = 0 U(x_{0})=0 U ( x 0 ) = 0 , the path estimate gives the last assertion. In that case ℓ 12 \ell_{12} ℓ 12 is constant as well. □
The Laplacian component of the initial state cannot in general be omitted from this scalar propagation statement. For example, β = 0 \beta=0 β = 0 and ℓ 12 ( x ) = ∣ x ∣ 2 \ell_{12}(x)=|x|^{2} ℓ 12 ( x ) = ∣ x ∣ 2 satisfy E = 0 \mathcal E=0 E = 0 and ∇ ℓ 12 ( 0 ) = 0 \nabla\ell_{12}(0)=0 ∇ ℓ 12 ( 0 ) = 0 , but Δ ℓ 12 ( 0 ) = 6 \Delta\ell_{12}(0)=6 Δ ℓ 12 ( 0 ) = 6 and ℓ 12 \ell_{12} ℓ 12 is not constant. In Theorem 7.7 , normalization on the open set E e q E_{\mathrm{eq}} E eq supplies both components of the zero initial state.
Proof of Theorem 7.7 . Lemma 7.9 gives the common κ \kappa κ and expresses the remaining blocks in terms of ℓ 12 \ell_{12} ℓ 12 . Lemma 7.10 and Proposition 7.11 yield (211) ; Lemma 7.12 then gives the homogeneous system (212) . For this fixed pair of materials, β \beta β and C j C_{j} C j are fixed spatial coefficient fields, so ordinary uniqueness along paths applies to the unknown column U U U . Choose x 0 ∈ E e q x_0 \in E_{\mathrm{eq}} x 0 ∈ E eq . Since ℓ 12 \ell_{12} ℓ 12 vanishes on the open set E e q E_{\mathrm{eq}} E eq , its first and second derivatives vanish there. Thus U ( x 0 ) = 0 U(x_0)=0 U ( x 0 ) = 0 and ℓ 12 ( x 0 ) = 0 \ell_{12}(x_0)=0 ℓ 12 ( x 0 ) = 0 . Equation (211) says that the defect E \mathcal{E} E of Proposition 7.13 is zero. Its path estimate therefore gives U = 0 U=0 U = 0 throughout the connected set O \mathcal{O} O . It follows that ℓ 12 \ell_{12} ℓ 12 is constant, and ℓ 12 ( x 0 ) = 0 \ell_{12}(x_0)=0 ℓ 12 ( x 0 ) = 0 makes it zero. Lemma 7.9 now gives μ 1 = μ 2 \mu_1=\mu_2 μ 1 = μ 2 , the common κ \kappa κ , and H = I 3 H=I_3 H = I 3 , and the inverse formula gives the common λ \lambda λ . □ \square □
Propositions 7.5 and 7.6 provide the relative matrix on the common ball with the hypotheses of Theorem 7.7 . In primal quotient coordinates the corresponding matrix G G G satisfies 2 D θ G + A 1 G − G A 2 = 0 2D_\theta G+A_1G-GA_2=0 2 D θ G + A 1 G − G A 2 = 0 . Since Γ j = − A j T \Gamma_j=-A_j^T Γ j = − A j T , differentiation gives 2 D θ ( G − T ) = − Γ 1 G − T + G − T Γ 2 2D_\theta(G^{-T})=-\Gamma_1G^{-T}+G^{-T}\Gamma_2 2 D θ ( G − T ) = − Γ 1 G − T + G − T Γ 2 , so H = G − T H=G^{-T} H = G − T has the material order used throughout this section.
Density recovery and completion of the proof
This section completes the uniqueness argument after the Lamé parameters have been determined in § 7.5 . With the common Lamé operator fixed, the physical Green–Betti identity (216) isolates the difference of the two zeroth-order coefficients W j = ω 2 ρ j W_j=\omega^2\rho_j W j = ω 2 ρ j in a bilinear integral of products of elastic displacements. The remaining task is to construct exact physical solutions whose products have a controlled high-frequency limit and detect this difference.
For each nonzero Fourier direction ξ \xi ξ , we choose null phases whose sum is i ξ i\xi i ξ and construct the corresponding reduced CGO solutions using the local invertible transport frames of Theorem 4.1 and the right inverse with simultaneous bounds in Proposition 5.6. The simultaneous L 2 L^2 L 2 and H 1 H^1 H 1 remainder estimates yield an L 1 L^1 L 1 limit for the products of the physical displacements. This limit first gives a Fourier identity with a localized transverse weight. A finite family of local tests whose squares sum to one on the projection of Ω \Omega Ω removes that weight and recovers the full Fourier transform of ( W 1 − W 2 ) / μ (W_1-W_2)/\mu ( W 1 − W 2 ) / μ . Fourier uniqueness then implies W 1 = W 2 W_1=W_2 W 1 = W 2 in Ω \Omega Ω , and the fixed known frequency ω > 0 \omega>0 ω > 0 gives ρ 1 = ρ 2 \rho_1=\rho_2 ρ 1 = ρ 2 . A uniform version of the product estimate on a growing frequency window is also established.
The physical integral identity and the potential reduction
Since the Lamé coefficients have been recovered, let L = L 1 = L 2 L=L_1=L_2 L = L 1 = L 2 and t = t 1 = t 2 t=t_1=t_2 t = t 1 = t 2 be the common elastic operator and traction from (61) . Let ν \nu ν be the outward unit normal. In the bilinear setting of § 1.6 , the trace γ 0 \gamma_0 γ 0 is understood as the bounded map H 1 ( Ω ; C 3 ) → H 1 / 2 ( ∂ Ω ; C 3 ) H^1(\Omega;\mathbb{C}^3)\to H^{1/2}(\partial\Omega;\mathbb{C}^3) H 1 ( Ω ; C 3 ) → H 1/2 ( ∂ Ω ; C 3 ) . In this section ( L + W j ) u j = 0 (L+W_j)u_j=0 ( L + W j ) u j = 0 abbreviates ( L + W j I 3 ) u j = 0 (L+W_jI_3)u_j=0 ( L + W j I 3 ) u j = 0 . For f j = γ 0 u j f_j=\gamma_0u_j f j = γ 0 u j , Green’s formula and symmetry of the stress give
∫ Ω ( W 1 − W 2 ) u 1 ⋅ u 2 d x = ⟨ ( Λ 2 Ω − Λ 1 Ω ) f 1 , f 2 ⟩ = 0. (216) \int_{\Omega}(W_1-W_2)u_1\cdot u_2\,dx=\langle(\Lambda_2^\Omega-\Lambda_1^\Omega)f_1,f_2\rangle=0. \tag*{(216)} ∫ Ω ( W 1 − W 2 ) u 1 ⋅ u 2 d x = ⟨( Λ 2 Ω − Λ 1 Ω ) f 1 , f 2 ⟩ = 0. (216)
We use the bilinear pairing convention fixed in § 1.6 . Here Λ j Ω = Λ λ j , μ j , ρ j Ω ( ω ) \Lambda_j^\Omega=\Lambda_{\lambda_j,\mu_j,\rho_j}^\Omega(\omega) Λ j Ω = Λ λ j , μ j , ρ j Ω ( ω ) denotes the physical Dirichlet-to-Neumann map on Ω \Omega Ω . To obtain the displayed sign, apply Green’s formula first to two solutions of the same system, which gives the bilinear symmetry of Λ j Ω \Lambda_j^\Omega Λ j Ω , and then to u 1 , u 2 u_1,u_2 u 1 , u 2 .
Use the common reduction coefficients and maps a , c , m , κ , T , S a,c,m,\kappa,T,S a , c , m , κ , T , S from (60) –(62) . For each j = 1 , 2 j=1,2 j = 1 , 2 , write M W j = M j W j M^{W_j}=M_j^{W_j} M W j = M j W j for the reduced operator with these common Lamé coefficients and its own potential W j W_j W j . Lemma 3.1 already establishes the factorization ( L + W j I 3 ) T = S M W j (L+W_jI_3)T=SM^{W_j} ( L + W j I 3 ) T = S M W j . The potential matrix W \mathcal{W} W is defined in (63) ; although its upper-right entry contains ∇ W j \nabla W_j ∇ W j , it is a multiplication matrix and therefore changes only zeroth-order terms, leaving the leading transport operator unchanged.
Null phases and the leading displacement amplitudes
Fix ξ ≠ 0 \xi\ne0 ξ = 0 , put k ξ = ∣ ξ ∣ k_\xi=|\xi| k ξ = ∣ ξ ∣ and ξ ^ = ξ / k ξ \widehat{\xi}=\xi/k_\xi ξ = ξ / k ξ , and choose a positively oriented real orthonormal basis ( e 1 , e 2 , ξ ^ ) (e_1,e_2,\widehat{\xi}) ( e 1 , e 2 , ξ ) . Let θ = e 1 + i e 2 \theta=e_1+ie_2 θ = e 1 + i e 2 and set
d h = h 2 k ξ 2 16 , α = 1 + d h , ϵ = h α , ζ 1 , 2 = ± θ h + i ξ 2 ± h k ξ 2 16 θ ˉ , ϑ j = − ϵ ζ j . d_h=\frac{h^2k_\xi^2}{16},\qquad\alpha=1+d_h,\qquad\epsilon=\frac{h}{\alpha},\qquad\zeta_{1,2}=\pm\frac{\theta}{h}+\frac{i\xi}{2}\pm\frac{hk_\xi^2}{16}\bar{\theta},\qquad\vartheta_j=-\epsilon\zeta_j. d h = 16 h 2 k ξ 2 , α = 1 + d h , ϵ = α h , ζ 1 , 2 = ± h θ + 2 i ξ ± 16 h k ξ 2 θ ˉ , ϑ j = − ϵ ζ j .
Here ξ ^ \hat{\xi} ξ ^ is the unit Fourier direction; k ξ k_\xi k ξ is distinct from the boundary-layer oscillation parameter k k k , and d h d_h d h is a scalar phase correction distinct from the reduced coefficient d d d . The vectors ζ j \zeta_j ζ j are the exponential phases, with ζ j 2 = ζ j ⋅ ζ j = 0 \zeta_j^2=\zeta_j\cdot\zeta_j=0 ζ j 2 = ζ j ⋅ ζ j = 0 and ζ 1 + ζ 2 = i ξ \zeta_1+\zeta_2=i\xi ζ 1 + ζ 2 = i ξ . For fixed ξ \xi ξ , the auxiliary parameter ϵ = h / α \epsilon=h/\alpha ϵ = h / α satisfies ϵ ∼ h \epsilon\sim h ϵ ∼ h as h ↓ 0 h\downarrow0 h ↓ 0 . The vectors ϑ j = − ϵ ζ j \vartheta_j=-\epsilon\zeta_j ϑ j = − ϵ ζ j are normalized transport directions, so e ϵ ζ j ⋅ x = e − ϑ j ⋅ x / ϵ e^{\epsilon\zeta_j\cdot x}=e^{-\vartheta_j\cdot x/\epsilon} e ϵ ζ j ⋅ x = e − ϑ j ⋅ x / ϵ has the form used in Proposition 5.8. Define the leading polarization vectors
q 1 = ξ ^ − i h k ξ 4 θ ˉ , q 2 = ξ ^ + i h k ξ 4 θ ˉ . q_1=\hat{\xi}-\frac{ihk_\xi}{4}\bar{\theta},\qquad q_2=\hat{\xi}+\frac{ihk_\xi}{4}\bar{\theta}. q 1 = ξ ^ − 4 ih k ξ θ ˉ , q 2 = ξ ^ + 4 ih k ξ θ ˉ .
Their bilinear products are
ϑ j ⋅ q j = 0 , q 1 ⋅ q 2 = 1 , q 1 ⋅ ϑ 2 = ϑ 1 ⋅ q 2 = − i ϵ k ξ , ϑ 1 ⋅ ϑ 2 = − ϵ 2 k ξ 2 2 . (217) \vartheta_j\cdot q_j=0,\qquad q_1\cdot q_2=1,\qquad q_1\cdot\vartheta_2=\vartheta_1\cdot q_2=-i\epsilon k_\xi,\qquad\vartheta_1\cdot\vartheta_2=-\frac{\epsilon^2k_\xi^2}{2}. \tag*{(217)} ϑ j ⋅ q j = 0 , q 1 ⋅ q 2 = 1 , q 1 ⋅ ϑ 2 = ϑ 1 ⋅ q 2 = − i ϵ k ξ , ϑ 1 ⋅ ϑ 2 = − 2 ϵ 2 k ξ 2 . (217)
The real and imaginary parts of ϑ j \vartheta_j ϑ j are orthogonal unit vectors. The corresponding leaf normals are
ν 1 , 2 = ( 1 − d h ) ξ ^ ∓ ( h k ξ / 2 ) e 2 1 + d h . \nu_{1,2}=\frac{(1-d_h)\hat{\xi}\mp(hk_\xi/2)e_2}{1+d_h}. ν 1 , 2 = 1 + d h ( 1 − d h ) ξ ^ ∓ ( h k ξ /2 ) e 2 .
Thus ν j → ξ ^ \nu_j\to\hat{\xi} ν j → ξ ^ and ∣ ν j − ξ ^ ∣ = O ξ ( ϵ ) |\nu_j-\hat{\xi}|=O_\xi(\epsilon) ∣ ν j − ξ ^ ∣ = O ξ ( ϵ ) . For fixed ξ \xi ξ , each direction path lies in a fixed compact chart for sufficiently small h h h .
The limiting directions are ϑ 1 → − θ \vartheta_1\to-\theta ϑ 1 → − θ and ϑ 2 → θ \vartheta_2\to\theta ϑ 2 → θ . At each prescribed transverse centre, choose local frame families for these two directions separately and intersect their transverse parameter intervals. Shrinking this common interval I I I and the two direction neighborhoods leaves a fixed compact parameter neighborhood for each path when h h h is sufficiently small. Choose a fixed real b ∈ C c ∞ ( I ) b\in C^\infty_c(I) b ∈ C c ∞ ( I ) . Since ϑ j ⋅ q j = 0 \vartheta_j\cdot q_j=0 ϑ j ⋅ q j = 0 and ϑ j ⋅ ν j = 0 \vartheta_j\cdot\nu_j=0 ϑ j ⋅ ν j = 0 , the leading amplitude a 0 , j a_{0,j} a 0 , j below solves T ϑ j a 0 , j = 0 T_{\vartheta_j}a_{0,j}=0 T ϑ j a 0 , j = 0 , where T ϑ = 2 D ϑ + A p h y s T_{\vartheta}=2D_{\vartheta}+A^{\mathrm{phys}} T ϑ = 2 D ϑ + A phys uses the common Lamé coefficients. Lemma 6.3 supplies its first correction, and Proposition 5.8 supplies an exact solution throughout Ω C G O \Omega_{\mathrm{CGO}} Ω CGO :
Z j = e ζ j ⋅ x ( a 0 , j + ϵ a 1 , j + r j ) , a 0 , j = ( b ( ν j ⋅ x ) q j , 0 ) T , M W j Z j = 0 , Z_j=e^{\zeta_j\cdot x}(a_{0,j}+\epsilon a_{1,j}+r_j),\qquad a_{0,j}=(b(\nu_j\cdot x)q_j,0)^T,\qquad M^{W_j}Z_j=0, Z j = e ζ j ⋅ x ( a 0 , j + ϵ a 1 , j + r j ) , a 0 , j = ( b ( ν j ⋅ x ) q j , 0 ) T , M W j Z j = 0 ,
with a 1 , j a_{1,j} a 1 , j uniformly bounded in H 2 H^2 H 2 and
∥ r j ∥ L 2 = O ( ϵ 2 ) , ∥ r j ∥ H 1 = O ( ϵ ) . (218) \|r_j\|_{L^2}=O(\epsilon^2),\qquad\|r_j\|_{H^1}=O(\epsilon). \tag*{(218)} ∥ r j ∥ L 2 = O ( ϵ 2 ) , ∥ r j ∥ H 1 = O ( ϵ ) . (218)
Lemma 8.1. Fix ξ ∈ R 3 ∖ { 0 } \xi\in\mathbb{R}^3\setminus\{0\} ξ ∈ R 3 ∖ { 0 } and a real test b ∈ C c ∞ ( I ) b\in C^\infty_c(I) b ∈ C c ∞ ( I ) on an interval common to the two limiting direction charts. Let Z j Z_j Z j be the exact reduced solutions constructed above, with phases ζ j \zeta_j ζ j , leading vectors q j q_j q j , uniformly bounded first corrections a 1 , j a_{1,j} a 1 , j , and remainders satisfying (218) . The physical elastic displacements u j = T Z j u_j=T Z_j u j = T Z j satisfy
e − i ξ ⋅ x u 1 ⋅ u 2 ⟶ μ − 1 b ( ξ ^ ⋅ x ) 2 in L 1 ( Ω ) . (219) e^{-i\xi\cdot x}u_1\cdot u_2\longrightarrow\mu^{-1}b(\hat{\xi}\cdot x)^2\quad\text{in }L^1(\Omega). \tag*{(219)} e − i ξ ⋅ x u 1 ⋅ u 2 ⟶ μ − 1 b ( ξ ^ ⋅ x ) 2 in L 1 ( Ω ) . (219) For each fixed ξ ≠ 0 \xi\ne0 ξ = 0 and b b b , the convergence holds along the two compact direction paths selected above.
Proof. Write a 1 , j = ( v 1 , j , f 1 , j ) a_{1,j}=(v_{1,j},f_{1,j}) a 1 , j = ( v 1 , j , f 1 , j ) and r j = ( r v , j , r f , j ) r_j=(r_{v,j},r_{f,j}) r j = ( r v , j , r f , j ) , separating their vector and scalar components. Set u j = T Z j u_j=T Z_j u j = T Z j . The factorization ( L + W j ) T = S M W j (L+W_j)T=SM^{W_j} ( L + W j ) T = S M W j proved above gives ( L + W j ) u j = 0 (L+W_j)u_j=0 ( L + W j ) u j = 0 , so these displacements may be used in (216) . Expanding the lift gives
e − ζ j ⋅ x u j = U j + E j , e r r , U j = c b ( ν j ⋅ x ) q j − a v j f 1 , j . e^{-\zeta_j\cdot x}u_j=U_j+E_{j,\mathrm{err}},\qquad U_j=cb(\nu_j\cdot x)q_j-av_jf_{1,j}. e − ζ j ⋅ x u j = U j + E j , err , U j = c b ( ν j ⋅ x ) q j − a v j f 1 , j .
The longitudinal term in U j U_j U j is of order one. The remaining terms satisfy
∥ E j , e r r ∥ L 2 ≤ C [ ϵ ∥ a 1 , j ∥ H 1 + ∥ r j ∥ H 1 + ϵ − 1 ∥ r j ∥ L 2 ] = O ( ϵ ) . \|E_{j,\mathrm{err}}\|_{L^2}\le C\left[\epsilon\|a_{1,j}\|_{H^1}+\|r_j\|_{H^1}+\epsilon^{-1}\|r_j\|_{L^2}\right]=O(\epsilon). ∥ E j , err ∥ L 2 ≤ C [ ϵ ∥ a 1 , j ∥ H 1 + ∥ r j ∥ H 1 + ϵ − 1 ∥ r j ∥ L 2 ] = O ( ϵ ) .
Using (217) , we obtain
U 1 ⋅ U 2 = μ − 1 b 1 b 2 + i ϵ k ξ c a ( b 1 f 1 , 2 + b 2 f 1 , 1 ) − 1 2 ϵ 2 k ξ 2 a 2 f 1 , 1 f 1 , 2 , b j = b ( ν j ⋅ x ) . U_1 \cdot U_2=\mu^{-1}b_1b_2+i\epsilon k_\xi ca(b_1f_{1,2}+b_2f_{1,1})-\frac{1}{2}\epsilon^2k_\xi^2a^2f_{1,1}f_{1,2},\qquad b_j=b(\nu_j\cdot x). U 1 ⋅ U 2 = μ − 1 b 1 b 2 + i ϵ k ξ c a ( b 1 f 1 , 2 + b 2 f 1 , 1 ) − 2 1 ϵ 2 k ξ 2 a 2 f 1 , 1 f 1 , 2 , b j = b ( ν j ⋅ x ) .
Hence ∥ U 1 ⋅ U 2 − μ − 1 b 1 b 2 ∥ L 1 = O ( ϵ ) \lVert U_1\cdot U_2-\mu^{-1}b_1b_2\rVert_{L^1}=O(\epsilon) ∥ U 1 ⋅ U 2 − μ − 1 b 1 b 2 ∥ L 1 = O ( ϵ ) . The H 2 H^2 H 2 bounds for a 1 , j a_{1,j} a 1 , j give ∥ U j ∥ 2 = O ( 1 ) \lVert U_j\rVert_2=O(1) ∥ U j ∥ 2 = O ( 1 ) . Therefore Cauchy–Schwarz and ∥ E j , e r r ∥ 2 = O ( ϵ ) \lVert E_{j,\mathrm{err}}\rVert_2=O(\epsilon) ∥ E j , err ∥ 2 = O ( ϵ ) bound E 1 , e r r ⋅ U 2 E_{1,\mathrm{err}}\cdot U_2 E 1 , err ⋅ U 2 , U 1 ⋅ E 2 , e r r U_1\cdot E_{2,\mathrm{err}} U 1 ⋅ E 2 , err , and E 1 , e r r ⋅ E 2 , e r r E_{1,\mathrm{err}}\cdot E_{2,\mathrm{err}} E 1 , err ⋅ E 2 , err in L 1 L^1 L 1 by O ( ϵ ) O(\epsilon) O ( ϵ ) . Since ν j → ξ ^ \nu_j\to\hat{\xi} ν j → ξ ^ , smoothness of the fixed compactly supported b b b gives b 1 b 2 → b ( ξ ^ ⋅ x ) 2 b_1b_2\to b(\hat{\xi}\cdot x)^2 b 1 b 2 → b ( ξ ^ ⋅ x ) 2 uniformly on the bounded working region. Finally ζ 1 + ζ 2 = i ξ \zeta_1+\zeta_2=i\xi ζ 1 + ζ 2 = i ξ , so the product of the exponential phases is exactly e i ξ ⋅ x e^{i\xi\cdot x} e i ξ ⋅ x . These facts prove (219) . Each potential has its own first correction; the estimates apply to this pair of exact elastic displacement fields. □
Corollary 8.2. Assume the common extended Lamé parameters satisfy (156) , and let ∥ W j ∥ C 2 ( Ω ‾ 0 ) ≤ M 2 \lVert W_j\rVert_{C^2(\overline{\Omega}_0)}\le M_2 ∥ W j ∥ C 2 ( Ω 0 ) ≤ M 2 , j = 1 , 2 j=1,2 j = 1 , 2 . Use the uniform local parameter charts of Corollary 5.10, with a fixed margin around the limiting directions and transverse support. For real b b b supported in a fixed compact subinterval of the common transverse chart with ∥ b ∥ C 4 ≤ B t e s t \lVert b\rVert_{C^4}\le B_{\mathrm{test}} ∥ b ∥ C 4 ≤ B test , there are constants c 0 , h ∗ , C > 0 c_0,h_*,C>0 c 0 , h ∗ , C > 0 such that, whenever
0 < h ≤ h ∗ , 0 < k ξ = ∣ ξ ∣ , h k ξ ≤ c 0 , 0<h\le h_*,\qquad0<k_\xi=|\xi|,\qquad hk_\xi\le c_0, 0 < h ≤ h ∗ , 0 < k ξ = ∣ ξ ∣ , h k ξ ≤ c 0 , the solutions in Lemma 8.1 can be chosen with
∥ e − i ξ ⋅ x u 1 ⋅ u 2 − μ − 1 b ( ξ ^ ⋅ x ) 2 ∥ L 1 ( Ω ) ≤ C ϵ ( 1 + k ξ ) ≤ C h ( 1 + k ξ ) . (220) \left\lVert e^{-i\xi\cdot x}u_1\cdot u_2-\mu^{-1}b(\hat{\xi}\cdot x)^2\right\rVert_{L^1(\Omega)}
\le C\epsilon(1+k_\xi)\le Ch(1+k_\xi). \tag*{(220)} e − i ξ ⋅ x u 1 ⋅ u 2 − μ − 1 b ( ξ ^ ⋅ x ) 2 L 1 ( Ω ) ≤ C ϵ ( 1 + k ξ ) ≤ C h ( 1 + k ξ ) . (220) The constants are uniform in the materials and in the normalized directions, on the fixed finite cover and with the stated test bounds. For a fixed frequency-window exponent 0 < γ w i n < 1 0<\gamma_{\mathrm{win}}<1 0 < γ win < 1 , this implies
sup 0 < ∣ ξ ∣ ≤ h − γ w i n ∥ e − i ξ ⋅ x u 1 ⋅ u 2 − μ − 1 b ( ξ ^ ⋅ x ) 2 ∥ L 1 ( Ω ) ≤ C h 1 − γ w i n (221) \sup_{0<|\xi|\le h^{-\gamma_{\mathrm{win}}}}
\left\lVert e^{-i\xi\cdot x}u_1\cdot u_2-\mu^{-1}b(\hat{\xi}\cdot x)^2\right\rVert_{L^1(\Omega)}
\le Ch^{1-\gamma_{\mathrm{win}}} \tag*{(221)} 0 < ∣ ξ ∣ ≤ h − γ win sup e − i ξ ⋅ x u 1 ⋅ u 2 − μ − 1 b ( ξ ^ ⋅ x ) 2 L 1 ( Ω ) ≤ C h 1 − γ win (221) for sufficiently small h h h , where the solutions and admissible local tests may be selected separately for each direction with the same bounds. The estimate controls interior products of the selected elastic solutions over the stated frequency window.
Proof. Decrease c 0 c_0 c 0 to at most one. With d h = h 2 k ξ 2 / 16 d_h=h^2k_\xi^2/16 d h = h 2 k ξ 2 /16 as above,
∣ q j ∣ 2 = 1 + h 2 k ξ 2 8 ≤ 9 8 , ϑ 1 = − θ − i h ξ / 2 − d h θ ‾ 1 + d h , ϑ 2 = θ − i h ξ / 2 + d h θ ‾ 1 + d h . |q_j|^2=1+\frac{h^2k_\xi^2}{8}\le\frac{9}{8},\qquad
\vartheta_1=\frac{-\theta-ih\xi/2-d_h\overline{\theta}}{1+d_h},\qquad
\vartheta_2=\frac{\theta-ih\xi/2+d_h\overline{\theta}}{1+d_h}. ∣ q j ∣ 2 = 1 + 8 h 2 k ξ 2 ≤ 8 9 , ϑ 1 = 1 + d h − θ − ih ξ /2 − d h θ , ϑ 2 = 1 + d h θ − ih ξ /2 + d h θ .
Thus ∣ ϑ 1 + θ ∣ + ∣ ϑ 2 − θ ∣ ≤ C h k ξ |\vartheta_1+\theta|+|\vartheta_2-\theta|\le Chk_\xi ∣ ϑ 1 + θ ∣ + ∣ ϑ 2 − θ ∣ ≤ C h k ξ . The uniform chart margins allow a further fixed decrease of c 0 c_0 c 0 so that each path remains in its selected neighborhood. The leading amplitudes ( b ( ν j ⋅ x ) q j , 0 ) T (b(\nu_j\cdot x)q_j,0)^{T} ( b ( ν j ⋅ x ) q j , 0 ) T then have uniform C 4 C^4 C 4 bounds on the fixed source cylinders. Their scalar components vanish identically. The C 2 C^2 C 2 version of Lemma 6.4 therefore gives uniform H 2 H^2 H 2 first corrections and simultaneous ∥ r j ∥ 2 ≤ C ϵ 2 \lVert r_j\rVert_2\le C\epsilon^2 ∥ r j ∥ 2 ≤ C ϵ 2 , ∥ r j ∥ H 1 ≤ C ϵ \lVert r_j\rVert_{H^1}\le C\epsilon ∥ r j ∥ H 1 ≤ C ϵ . The construction uses the uniform within-chart frame estimates for each of the two selected direction families.
In the notation of Lemma 8.1, it follows that ∥ U j ∥ 2 ≤ C \lVert U_j\rVert_2\le C ∥ U j ∥ 2 ≤ C and ∥ E j , e r r ∥ 2 ≤ C ϵ \lVert E_{j,\mathrm{err}}\rVert_2\le C\epsilon ∥ E j , err ∥ 2 ≤ C ϵ . The exact contractions (217) imply
∥ U 1 ⋅ U 2 − μ − 1 b 1 b 2 ∥ 1 ≤ C ( ϵ k ξ + ϵ 2 k ξ 2 ) . \lVert U_1\cdot U_2-\mu^{-1}b_1b_2\rVert_1\le C(\epsilon k_\xi+\epsilon^2k_\xi^2). ∥ U 1 ⋅ U 2 − μ − 1 b 1 b 2 ∥ 1 ≤ C ( ϵ k ξ + ϵ 2 k ξ 2 ) .
The explicit normal vectors give the more precise identity
∣ ν j − ξ ^ ∣ = h k ξ 2 1 + d h = 1 2 ϵ k ξ 1 + d h . (222) |\nu_j-\hat{\xi}|=\frac{hk_\xi}{2\sqrt{1+d_h}}=\frac{1}{2}\epsilon k_\xi\sqrt{1+d_h}. \tag*{(222)} ∣ ν j − ξ ^ ∣ = 2 1 + d h h k ξ = 2 1 ϵ k ξ 1 + d h . (222)
Indeed, the normal and tangential components of ν j − ξ ^ \nu_j-\hat{\xi} ν j − ξ ^ have lengths 2 d h / ( 1 + d h ) 2d_h/(1+d_h) 2 d h / ( 1 + d h ) and h k ξ / [ 2 ( 1 + d h ) ] hk_\xi/[2(1+d_h)] h k ξ / [ 2 ( 1 + d h )] , and their squared sum is h 2 k ξ 2 / [ 4 ( 1 + d h ) ] h^2k_\xi^2/[4(1+d_h)] h 2 k ξ 2 / [ 4 ( 1 + d h )] . Since Ω \Omega Ω is bounded, the mean value theorem for the zero-extended b b b now yields ∥ b 1 b 2 − b ( ξ ^ ⋅ x ) 2 ∥ L ∞ ( Ω ) ≤ C ϵ k ξ \lVert b_1b_2-b(\hat{\xi}\cdot x)^2\rVert_{L^\infty(\Omega)}\le C\epsilon k_\xi ∥ b 1 b 2 − b ( ξ ^ ⋅ x ) 2 ∥ L ∞ ( Ω ) ≤ C ϵ k ξ . The products involving E j , e r r E_{j,\mathrm{err}} E j , err contribute at most C ϵ C\epsilon C ϵ by Cauchy–Schwarz. Finally ϵ k ξ ≤ h k ξ ≤ 1 \epsilon k_\xi\le hk_\xi\le1 ϵ k ξ ≤ h k ξ ≤ 1 , so ϵ 2 k ξ 2 ≤ ϵ k ξ \epsilon^2k_\xi^2\le\epsilon k_\xi ϵ 2 k ξ 2 ≤ ϵ k ξ . Combining these bounds with ζ 1 + ζ 2 = i ξ \zeta_1+\zeta_2=i\xi ζ 1 + ζ 2 = i ξ proves (220) . For the stated frequency window, h k ξ ≤ h 1 − γ w i n → 0 h k_\xi\le h^{1-\gamma_{\mathrm{win}}}\to0 h k ξ ≤ h 1 − γ win → 0 and h ( 1 + k ξ ) ≤ 2 h 1 − γ w i n h(1+k_\xi)\le2h^{1-\gamma_{\mathrm{win}}} h ( 1 + k ξ ) ≤ 2 h 1 − γ win when 0 < h ≤ 1 0<h\le1 0 < h ≤ 1 .
Removal of the local weights and Fourier uniqueness
Proposition 8.3. Suppose the Lamé parameters are common, smooth, and strongly convex, both Dirichlet problems on Ω \Omega Ω are nonresonant, and the physical DN maps agree. If (219) is available for every fixed ξ ≠ 0 \xi\ne0 ξ = 0 and every transverse centre, on a local interval and for each fixed real test function b b b , then W 1 = W 2 W_1=W_2 W 1 = W 2 on Ω ‾ \overline{\Omega} Ω .
Proof. Equations (216) and (219) give
∫ Ω e i ξ ⋅ x W 1 − W 2 μ b ( ξ ^ ⋅ x ) d x = 0. \int_{\Omega} e^{i\xi\cdot x}\frac{W_1-W_2}{\mu}b(\widehat{\xi}\cdot x)\,dx=0. ∫ Ω e i ξ ⋅ x μ W 1 − W 2 b ( ξ ⋅ x ) d x = 0.
The compact set J = ξ ^ ⋅ Ω ‾ J=\widehat{\xi}\cdot\overline{\Omega} J = ξ ⋅ Ω has a finite cover by local intervals I 1 , … , I N I_1,\ldots,I_N I 1 , … , I N . Choose χ ℓ ∈ C c ∞ ( I ℓ ; R ) \chi_\ell\in C^\infty_c(I_\ell;\mathbb{R}) χ ℓ ∈ C c ∞ ( I ℓ ; R ) so that S = ∑ ℓ χ ℓ 2 > 0 S=\sum_\ell\chi_\ell^2>0 S = ∑ ℓ χ ℓ 2 > 0 near J J J , and take η ∈ C c ∞ ( { S > 0 } ) \eta\in C^\infty_c(\{S>0\}) η ∈ C c ∞ ({ S > 0 }) equal to one near J J J . Put b ℓ = η χ ℓ / S b_\ell=\eta\chi_\ell/\sqrt{S} b ℓ = η χ ℓ / S on { S > 0 } \{S>0\} { S > 0 } and extend by zero. The denominator is bounded below on supp η \operatorname{supp}\eta supp η , so these are smooth local tests and ∑ ℓ b ℓ 2 = 1 \sum_\ell b_\ell^2=1 ∑ ℓ b ℓ 2 = 1 on J J J . Applying the local identity to each b ℓ b_\ell b ℓ and adding the resulting integrals gives
∫ Ω e i ξ ⋅ x ( W 1 − W 2 ) / μ d x = 0 ( ξ ≠ 0 ) . \int_{\Omega}e^{i\xi\cdot x}(W_1-W_2)/\mu\,dx=0\qquad(\xi\ne0). ∫ Ω e i ξ ⋅ x ( W 1 − W 2 ) / μ d x = 0 ( ξ = 0 ) .
Let 1 Ω \mathbf{1}_{\Omega} 1 Ω denote the indicator of Ω \Omega Ω and set F = 1 Ω ( W 1 − W 2 ) / μ ∈ L c 1 ( R 3 ) ∩ L 2 ( R 3 ) F=\mathbf{1}_{\Omega}(W_1-W_2)/\mu\in L^1_c(\mathbb{R}^3)\cap L^2(\mathbb{R}^3) F = 1 Ω ( W 1 − W 2 ) / μ ∈ L c 1 ( R 3 ) ∩ L 2 ( R 3 ) . In the bilinear convention of Section 1.6, use the Fourier transform
F ^ ( ξ ) = ∫ R 3 e − i ξ ⋅ x F ( x ) d x . \widehat{F}(\xi)=\int_{\mathbb{R}^3}e^{-i\xi\cdot x}F(x)\,dx. F ( ξ ) = ∫ R 3 e − i ξ ⋅ x F ( x ) d x .
Replacing ξ \xi ξ by − ξ -\xi − ξ in the preceding identity gives F ^ ( ξ ) = 0 \widehat{F}(\xi)=0 F ( ξ ) = 0 for ξ ≠ 0 \xi\ne0 ξ = 0 . Compact support and integrability make this transform continuous, so it vanishes at ξ = 0 \xi=0 ξ = 0 as well. For G t ( x ) = ( 4 π t ) − 3 / 2 e − ∣ x ∣ 2 / ( 4 t ) G_t(x)=(4\pi t)^{-3/2}e^{-|x|^2/(4t)} G t ( x ) = ( 4 π t ) − 3/2 e − ∣ x ∣ 2 / ( 4 t ) , absolute integrability and Fubini’s theorem give
G t ∗ F ( x ) = ( 2 π ) − 3 ∫ e i ξ ⋅ x − t ∣ ξ ∣ 2 F ^ ( ξ ) d ξ = 0. G_t*F(x)=(2\pi)^{-3}\int e^{i\xi\cdot x-t|\xi|^2}\widehat{F}(\xi)\,d\xi=0. G t ∗ F ( x ) = ( 2 π ) − 3 ∫ e i ξ ⋅ x − t ∣ ξ ∣ 2 F ( ξ ) d ξ = 0.
Since the Gaussian kernels form an approximate identity, G t ∗ F → F G_t*F\to F G t ∗ F → F in L 1 L^1 L 1 , and hence F = 0 F=0 F = 0 almost everywhere. Continuity of the physical coefficients gives W 1 = W 2 W_1=W_2 W 1 = W 2 on Ω ‾ \overline{\Omega} Ω .
Lemma 8.1, using the fixed-frame construction, first corrections, and the same two-norm remainder estimate, supplies the local product limits required in Proposition 8.3. Since the common measurement frequency is known and nonzero, ρ 1 − ρ 2 = ( W 1 − W 2 ) / ω 2 = 0 \rho_1-\rho_2=(W_1-W_2)/\omega^2=0 ρ 1 − ρ 2 = ( W 1 − W 2 ) / ω 2 = 0 .
Completion of the uniqueness proof
Proof of Theorem 1.1. The boundary determination theorem, Theorem 2.5, recovers all boundary Taylor data of λ j , μ j , W j \lambda_j,\mu_j,W_j λ j , μ j , W j from the original boundary maps. Proposition 2.7 supplies smooth extensions with common exterior coefficients, constant Lamé parameters and zero potential far from Ω \Omega Ω . The boundary values of W j W_j W j and the first boundary jets of μ j \mu_j μ j give the reduced integral identity in Proposition 3.4.
Theorem 4.1 constructs the local families of transport frames. Theorem B.1 and Proposition 5.1 give the local right inverse. Propositions 5.6 and 5.8 apply it to the four-component elastic reduction, and Proposition 6.2 fixes the domains, cutoffs, and frames for the moment construction. Lemmas 6.6 and 6.11 realize all three columns by elastic solutions. Theorem 6.13 therefore gives every polynomial moment of the full matrix for the chosen frames.
The coefficient difference is supported strictly inside B R 0 B_{R_0} B R 0 , so its plane integrals may be taken over the chosen larger disk D L D_L D L . Proposition 7.1 identifies the dual quotient connection. Lemma 7.3 turns the moments into a compact Cauchy primitive. Proposition 7.4 gives the unique normalized relative matrix, independent of solution frames; Proposition 7.5 constructs its parameter-dependent global version and proves det H = μ 2 / μ 1 > 0 \det H = \sqrt{\mu_2/\mu_1} > 0 det H = μ 2 / μ 1 > 0 . The common exterior material and uniqueness on each plane give its normalization on a common physical shell. Proposition 7.6 proves holomorphic dependence on direction at each fixed physical point in both affine charts. The hypotheses of Theorem 7.7 are thus satisfied. It follows that μ 1 = μ 2 \mu_1 = \mu_2 μ 1 = μ 2 and κ 1 = κ 2 \kappa_1 = \kappa_2 κ 1 = κ 2 , hence λ 1 = λ 2 \lambda_1 = \lambda_2 λ 1 = λ 2 .
For these common Lamé parameters, the exact null phases in (217) and the remainder estimates give the product limit (219) . Proposition 8.3 removes the transverse weights by a finite local square decomposition and recovers W 1 = W 2 W_1 = W_2 W 1 = W 2 by Fourier uniqueness. Finally W j = ω 2 ρ j W_j = \omega^2\rho_j W j = ω 2 ρ j and the known ω > 0 \omega> 0 ω > 0 imply ρ 1 = ρ 2 \rho_1 = \rho_2 ρ 1 = ρ 2 . □
Appendix A. Finite derivative bounds for the boundary layers
We give the finite derivative bounds and global remainder estimates used in § 2 .
Let Ω ⊂ R 3 \Omega\subset\mathbb{R}^3 Ω ⊂ R 3 be bounded with smooth boundary, and let
P = div ( 2 μ ε ( ⋅ ) + λ div ( ⋅ ) I 3 ) + W I 3 , μ > 0 , 3 λ + 2 μ > 0 , P = \operatorname{div}(2\mu\varepsilon(\cdot) + \lambda\operatorname{div}(\cdot)I_3) + WI_3,\qquad\mu> 0,\quad3\lambda+ 2\mu> 0, P = div ( 2 μ ε ( ⋅ ) + λ div ( ⋅ ) I 3 ) + W I 3 , μ > 0 , 3 λ + 2 μ > 0 ,
have real smooth coefficients on Ω ‾ \overline{\Omega} Ω and trivial Dirichlet kernel. All components are taken in a fixed Cartesian basis. We use the inward boundary normal coordinates and coordinate differential fields defined in (12) and (15) . Let y = ( y 1 , y 2 ) ∈ V ⊂ R 2 y = (y_1,y_2) \in V \subset\mathbb{R}^2 y = ( y 1 , y 2 ) ∈ V ⊂ R 2 , choose a compact patch K ⋐ V K \Subset V K ⋐ V , and take 0 < h ≤ 1 0 < h \le1 0 < h ≤ 1 , t = r / h t = r/h t = r / h and s ∈ { + 1 , − 1 } s \in\{+1,-1\} s ∈ { + 1 , − 1 } . The normal coordinate r r r is positive inside Ω \Omega Ω ; thus the outward normal at r = 0 r = 0 r = 0 is − ∇ r -\nabla r − ∇ r . The action of X i \mathcal{X}_i X i on coordinate coefficients is as in (15) and (21) ; in particular, X i ( n i n , j ) \mathcal{X}_i(n_{\mathrm{in},j}) X i ( n in , j ) is a derivative of the coefficient n i n , j n_{\mathrm{in},j} n in , j , not a multiplication operator. We use the operators D i , h D_{i,h} D i , h , P s , h P_{s,h} P s , h and P s , q P_{s,q} P s , q defined in (19) , (22) and (24) , respectively. The dependence of A m , β A_{m,\beta} A m , β on the fixed sign s s s is suppressed. Repeated tangential indices run from one to two and are summed. The volume form is J ( y , r ) d y d r J(y,r)\,dy\,dr J ( y , r ) d y d r , where J > 0 J > 0 J > 0 . Constants may depend on the fixed coordinate collar, cutoffs and global Dirichlet inverse, including upper and positive lower bounds for J J J and bounds for coordinate nondegeneracy.
Coordinate coefficients and ordered products
Lemma A.1. With A m , β A_{m,\beta} A m , β as defined in (23) , the coefficient matrices satisfy the following derivative bounds. The matrices A 0 , β A_{0,\beta} A 0 , β depend only on the values of the original coefficients. The matrices A 1 , β , A 2 , β A_{1,\beta}, A_{2,\beta} A 1 , β , A 2 , β involve spatial derivatives of λ , μ , n i n , w \lambda,\mu,n_{\mathrm{in}},w λ , μ , n in , w of order at most one; W W W occurs only for m = 2 m = 2 m = 2 . Moreover, A m , β = 0 A_{m,\beta} = 0 A m , β = 0 if β 1 + β 2 > m \beta_1 + \beta_2 > m β 1 + β 2 > m . For q ≥ m q \ge m q ≥ m , the largest normal derivative orders needed in ∂ r q − m A m , β \partial_r^{q-m} A_{m,\beta} ∂ r q − m A m , β are q q q , q q q , q − 1 q-1 q − 1 for m = 0 , 1 , 2 m = 0,1,2 m = 0 , 1 , 2 , respectively. The largest tangential order is at most one.
Proof. For the fixed sign s s s , write D i , h = n i n , i ∂ t + i s w i 1 + h w i α ∂ y α D_{i,h} = n_{\mathrm{in},i}\partial_t + isw_i^1 + hw_i^\alpha\partial_{y_\alpha} D i , h = n in , i ∂ t + i s w i 1 + h w i α ∂ y α , with its coefficients evaluated at ( y , h t ) (y,ht) ( y , h t ) . Let B i j B_{ij} B ij be the nine terms obtained by multiplying the three summands in each of D i , h D_{i,h} D i , h and D j , h D_{j,h} D j , h while holding their coefficient functions fixed. When a derivative acts on a coefficient b ( y , r ) b(y,r) b ( y , r ) , the chain rule gives
( n i n , i ∂ t + h w i α ∂ y α ) b ( y , h t ) = h ( X i ( b ) ) ( y , h t ) . (n_{\mathrm{in},i}\partial_t + hw_i^\alpha\partial_{y_\alpha})b(y,ht) = h(\mathcal{X}_i(b))(y,ht). ( n in , i ∂ t + h w i α ∂ y α ) b ( y , h t ) = h ( X i ( b )) ( y , h t ) .
The derivatives X i ( b ) \mathcal{X}_i(b) X i ( b ) are taken at the physical scale before evaluation at r = h t r = ht r = h t . Applying the ordinary product rule to the fields in (15) therefore gives
D i , h D j , h = B i j + h ( X i ( n i n , j ) ) ∂ t + h 2 ( X i ( w j β ) ) ∂ y β + i s h X i ( w j 1 ) . (223) D_{i,h}D_{j,h} = B_{ij} + h(\mathcal{X}_i(n_{\mathrm{in},j}))\partial_t + h^2(\mathcal{X}_i(w_j^\beta))\partial_{y_\beta} + ish\mathcal{X}_i(w_j^1). \tag*{(223)} D i , h D j , h = B ij + h ( X i ( n in , j )) ∂ t + h 2 ( X i ( w j β )) ∂ y β + i s h X i ( w j 1 ) . (223)
Thus all derivatives of the normal, tangential and phase coefficients are included. Substitution into
( P s , h U ) i = μ ∑ j D j , h D j , h U i + ( λ + μ ) ∑ j D i , h D j , h U j + h ( X i ( λ ) ) ∑ j D j , h U j + h ∑ j ( X j ( μ ) ) ( D j , h U i + D i , h U j ) + h 2 W U i . (\mathcal{P}_{s,h}U)_i=\mu\sum_j D_{j,h}D_{j,h}U_i+(\lambda+\mu)\sum_j D_{i,h}D_{j,h}U_j+h(\mathcal{X}_i(\lambda))\sum_jD_{j,h}U_j+h\sum_j(\mathcal{X}_j(\mu))(D_{j,h}U_i+D_{i,h}U_j)+h^2WU_i. ( P s , h U ) i = μ j ∑ D j , h D j , h U i + ( λ + μ ) j ∑ D i , h D j , h U j + h ( X i ( λ )) j ∑ D j , h U j + h j ∑ ( X j ( μ )) ( D j , h U i + D i , h U j ) + h 2 W U i .
gives the coefficient structure in (23) and the stated derivative bounds. The two first-order factors produce powers of h h h no higher than two, and every coefficient derivative has at least one factor h h h . For m = 1 m=1 m = 1 , taking q − 1 q-1 q − 1 further normal derivatives of a first coefficient derivative requires at most order q q q ; for m = 2 m=2 m = 2 , the bound is q − 1 q-1 q − 1 . The dependence of X i \mathcal{X}_i X i on n i n , w n_{\mathrm{in}},w n in , w obeys the same bounds under the product rule. □
For 0 ≤ q ≤ N b l 0\le q\le N_{\mathrm{bl}} 0 ≤ q ≤ N bl , define
P s , q = ∑ m = 0 min ( 2 , q ) ∑ ∣ β ∣ ≤ 2 t q − m ( q − m ) ! ∂ t q − m A m , β ( y , 0 ) ∂ ( y , t ) β . \mathcal{P}_{s,q}=\sum_{m=0}^{\min(2,q)}\sum_{|\beta|\le2}\frac{t^{q-m}}{(q-m)!}\partial_t^{q-m}A_{m,\beta}(y,0)\partial_{(y,t)}^\beta. P s , q = m = 0 ∑ m i n ( 2 , q ) ∣ β ∣ ≤ 2 ∑ ( q − m )! t q − m ∂ t q − m A m , β ( y , 0 ) ∂ ( y , t ) β .
For m ≤ N b l m\le N_{\mathrm{bl}} m ≤ N bl , expand A m , β A_{m,\beta} A m , β to Taylor order N = N b l − m N=N_{\mathrm{bl}}-m N = N bl − m . After ∣ α ∣ + b ≤ S |\alpha|+b\le S ∣ α ∣ + b ≤ S derivatives in ( y , t ) (y,t) ( y , t ) , the integral remainder uses at most
N b l − m + 1 ⏟ Taylor remainder + b ⏟ ∂ t on coefficients + 1 m ≥ 1 ⏟ existing coefficient derivative ≤ N b l + S + 1 (224) \underbrace{N_{\mathrm{bl}}-m+1}_{\text{Taylor remainder}}+\underbrace{b}_{\partial_t\text{ on coefficients}}+\underbrace{\mathbf{1}_{m\ge1}}_{\text{existing coefficient derivative}}\le N_{\mathrm{bl}}+S+1 \tag*{(224)} Taylor remainder N bl − m + 1 + ∂ t on coefficients b + existing coefficient derivative 1 m ≥ 1 ≤ N bl + S + 1 (224)
normal derivatives of the original coefficients. In Taylor’s integral remainder, s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] is the integration variable. Every ∂ t \partial_t ∂ t acting on a coefficient evaluated at ( y , s h t ) (y,sht) ( y , s h t ) contributes a factor s h sh s h . We use the remainder and coefficient formulas (28) and (27) ; the bound h N b l + 1 ( 1 + t ) N b l + 1 + S h^{N_{\mathrm{bl}}+1}(1+t)^{N_{\mathrm{bl}}+1+S} h N bl + 1 ( 1 + t ) N bl + 1 + S suffices. For m > N b l m>N_{\mathrm{bl}} m > N bl , retain the original term in the remainder. Since h m ≤ h N b l + 1 h^m\le h^{N_{\mathrm{bl}}+1} h m ≤ h N bl + 1 , this requires at most S + 1 S+1 S + 1 normal derivatives. In particular, for N b l = 0 N_{\mathrm{bl}}=0 N bl = 0 only A 0 A_0 A 0 is expanded to order zero, while h A 1 hA_1 h A 1 and h 2 A 2 h^2A_2 h 2 A 2 remain in the remainder.
The frozen inverse and rectangular derivative bounds
We use the frozen operator and the decaying matrix profile defined in (29) and (30) . Lemma 2.1 provides the unique decaying solution for polynomial-exponential forcing, with a polynomial degree increase of at most two and uniform parameter-derivative bounds on the compact patch. The construction involves only operations in the normal variable and smooth functions of the boundary coefficients; in particular, it introduces no additional normal derivatives of the material coefficients.
Theorem A.2. Fix N b l , S ≥ 0 N_{\mathrm{bl}},S\ge0 N bl , S ≥ 0 , A ∈ C c ∞ ( V ; C 3 ) A\in C^\infty_c(V;\mathbb{C}^3) A ∈ C c ∞ ( V ; C 3 ) , s = ± 1 \mathfrak{s}=\pm1 s = ± 1 , and a normal cutoff η \eta η as in Proposition 2.2. Let U 0 , … , U N b l U_0,\ldots,U_{N_{\mathrm{bl}}} U 0 , … , U N bl be the profiles constructed there. The parameter derivatives of the profiles through order S + 2 S+2 S + 2 and of the scaled residual through order S S S are controlled by the C 2 N b l + S + 2 C^{2N_{\mathrm{bl}}+S+2} C 2 N bl + S + 2 norm of A A A and the rectangular coefficient bounds
max ∣ α ∣ ≤ 2 N b l + S + 2 , 0 ≤ b ≤ N b l + S + 1 ∥ ∂ y α ∂ t b ( λ , μ , W , n i n , w ) ∥ ∞ (225) \max_{\substack{|\alpha|\le2N_{\mathrm{bl}}+S+2,\,0\le b\le N_{\mathrm{bl}}+S+1}}\left\|\partial_y^\alpha\partial_t^b(\lambda,\mu,W,n_{\mathrm{in}},w)\right\|_\infty \tag*{(225)} ∣ α ∣ ≤ 2 N bl + S + 2 , 0 ≤ b ≤ N bl + S + 1 max ∂ y α ∂ t b ( λ , μ , W , n in , w ) ∞ (225) together with the fixed cutoff norms, Jacobian and coordinate nondegeneracy bounds, and lower bounds for the positive denominators.
Proof. The profile recursion is (36) , and the support and polynomial-degree bounds have already been established in the proof of Proposition 2.2, in particular (37) . The additional derivative bounds follow by induction: for every s ≥ 0 s\ge0 s ≥ 0 , that ∂ y α ∂ t s + 2 U ℓ \partial_y^{\alpha}\partial_t^{s+2}U_\ell ∂ y α ∂ t s + 2 U ℓ requires coefficient derivatives of tangential order at most 2 ℓ + s 2\ell+s 2 ℓ + s and normal order at most ℓ \ell ℓ , and amplitude derivatives of tangential order at most 2 ℓ + s 2\ell+s 2 ℓ + s . The assertion is immediate for ℓ = 0 \ell=0 ℓ = 0 . The q q q th source term adds at most two tangential derivatives to those required at a lower level, while its coefficients require at most s + 1 s+1 s + 1 tangential and q q q normal derivatives. Since s + 2 ( ℓ − q ) + 1 ≤ s + 2 ℓ s + 2(\ell- q) + 1 \le s + 2\ell s + 2 ( ℓ − q ) + 1 ≤ s + 2 ℓ , the induction closes. The frozen inverse acts algebraically on boundary parameters and adds no normal derivatives. Taking s = S + 2 s = S + 2 s = S + 2 and using (A.2) gives (A.3). Differentiation in t t t of a profile adds no normal coefficient derivatives; derivatives of unfrozen coefficients are counted by (A.2).
Where η = 1 \eta= 1 η = 1 , the recursion cancels h 0 , … , h N b l h^0,\ldots,h^{N_{\mathrm{bl}}} h 0 , … , h N bl . Taylor's integral remainder for m ≤ N b l m \le N_{\mathrm{bl}} m ≤ N bl , with the terms m > N b l m > N_{\mathrm{bl}} m > N bl retained in the remainder, gives, after any ∣ γ ∣ ≤ S \lvert\gamma\rvert\le S ∣ γ ∣ ≤ S derivatives in ( y , t ) (y,t) ( y , t ) ,
∣ ∂ ( y , t ) γ P S , h ∑ ℓ = 0 N b l h ℓ U ℓ ∣ ≤ C h N b l + 1 ( 1 + t ) M b l , S e − c 0 t . \left\lvert\partial_{(y,t)}^\gamma\mathcal{P}_{S,h}\sum_{\ell=0}^{N_{\mathrm{bl}}} h^\ell U_\ell\right\rvert\le C h^{N_{\mathrm{bl}}+1}(1+t)^{M_{\mathrm{bl}},S}e^{-c_0t}. ∂ ( y , t ) γ P S , h ℓ = 0 ∑ N bl h ℓ U ℓ ≤ C h N bl + 1 ( 1 + t ) M bl , S e − c 0 t .
The finite products with q + ℓ > N b l q+\ell>N_{\mathrm{bl}} q + ℓ > N bl also contain at least h N b l + 1 h^{N_{\mathrm{bl}}+1} h N bl + 1 . In the cutoff-transition region, all differentiated profile and coefficient terms, including cutoff commutators, are bounded by C h − M e − c 0 r ∗ / h C h^{-M}e^{-c_0r_*/h} C h − M e − c 0 r ∗ / h for some finite M = M ( N b l , S ) M=M(N_{\mathrm{bl}},S) M = M ( N bl , S ) and a fixed r ∗ > 0 r_*>0 r ∗ > 0 . They are therefore O ( h B ) O(h^B) O ( h B ) for every prescribed B B B . Taking γ = 0 \gamma=0 γ = 0 is consistent with the residual estimate (33) ; the L 2 L^2 L 2 , H 1 H^1 H 1 , and exact-correction estimates (34) –(35) follow from Proposition 2.2. □
Relation to the boundary determination theorem
Theorem 2.5 establishes the agreement of the full boundary Taylor coefficients. The present appendix supplies the finite-order coefficient-derivative bounds and differentiated residual estimates in Theorem A.2. These quantitative bounds complement the boundary-layer construction in Proposition 2.2 and the separation estimate in Lemma 2.4.
Appendix B. The free inverse and Fourier operator estimates
We construct a scalar free inverse, acting componentwise on vectors. We use the bilinear convention of Section 1.6, the weighted Sobolev spaces of Section 5.1, and the Fourier convention of Section 5.2. In this section the spatial weight exponent is written s w t s_{\mathrm{wt}} s wt to distinguish it from the boundary-layer phase sign used in Appendix A. The derivatives are distributional; equivalent unweighted Bessel norms may be used in the Fourier-kernel arguments below.
Theorem B.1. Let δ = τ ( v 0 + i v 1 ) \delta=\tau(v_0+iv_1) δ = τ ( v 0 + i v 1 ) , where τ ≥ 1 \tau\ge1 τ ≥ 1 and v 0 , v 1 ∈ R 3 v_0,v_1\in\mathbb{R}^3 v 0 , v 1 ∈ R 3 are orthonormal, and put
p δ ( ξ ) = ∣ ξ ∣ 2 + 2 δ ⋅ ξ . p_\delta(\xi)=\lvert\xi\rvert^2+2\delta\cdot\xi. p δ ( ξ ) = ∣ ξ ∣ 2 + 2 δ ⋅ ξ . The ordinary locally integrable function m δ = 1 / p δ m_\delta=1/p_\delta m δ = 1/ p δ defines on Schwartz functions the selected operator E δ f = F − 1 ( m δ f ^ ) E_\delta f=\mathcal{F}^{-1}(m_\delta\widehat{f}) E δ f = F − 1 ( m δ f ) , where F \mathcal{F} F denotes the Fourier transform and F − 1 \mathcal{F}^{-1} F − 1 its inverse. It has a unique bounded extension
E δ : L 1 2 ( R 3 ) ⟶ H − 1 1 ( R 3 ) E_\delta:L^2_1(\mathbb{R}^3)\longrightarrow H^1_{-1}(\mathbb{R}^3) E δ : L 1 2 ( R 3 ) ⟶ H − 1 1 ( R 3 ) with
τ ∥ E δ f ∥ L − 1 2 + ∥ E δ f ∥ H − 1 1 ≤ C ∥ f ∥ L 1 2 . (226) \tau\lVert E_\delta f\rVert_{L^2_{-1}}+\lVert E_\delta f\rVert_{H^1_{-1}}\le C\lVert f\rVert_{L^2_1}. \tag*{(226)} τ ∥ E δ f ∥ L − 1 2 + ∥ E δ f ∥ H − 1 1 ≤ C ∥ f ∥ L 1 2 . (226) The same output satisfies both estimates, and
( D + δ ) 2 E δ f = f in S ′ ( R 3 ) . (227) (D+\delta)^2E_\delta f=f\quad\text{in }S'(\mathbb{R}^3). \tag*{(227)} ( D + δ ) 2 E δ f = f in S ′ ( R 3 ) . (227) The constant is independent of τ \tau τ and of the orthonormal pair.
Proof. Rotate to ( v 0 , v 1 ) = ( e 1 , e 2 ) (v_0,v_1)=(e_1,e_2) ( v 0 , v 1 ) = ( e 1 , e 2 ) ; radial weighted derivative norms are unchanged up to fixed dimension constants under this rotation. Write x ′ = ( x 1 , x 3 ) x'=(x_1,x_3) x ′ = ( x 1 , x 3 ) , t = x 2 t=x_2 t = x 2 , u = e − i x 1 τ u u=e^{-ix_1\tau}u u = e − i x 1 τ u , and F = e i x 1 τ f F=e^{ix_1\tau}f F = e i x 1 τ f . A tilde denotes Fourier transformation in x ′ = ( x 1 , x 3 ) x'=(x_1,x_3) x ′ = ( x 1 , x 3 ) alone, with dual variable η ′ = ( η 1 , η 3 ) \eta'=(\eta_1,\eta_3) η ′ = ( η 1 , η 3 ) ; the normal variable t = x 2 t=x_2 t = x 2 is retained. Tangential Fourier transformation of the conjugated equation gives
L b = − ∂ t 2 + 2 τ ∂ t + b 2 − τ 2 = − ( ∂ t − a + ) ( ∂ t − a − ) , b = ∣ η ′ ∣ , a ± = τ ± b . (228) L_b=-\partial_t^2+2\tau\partial_t+b^2-\tau^2=-(\partial_t-a_+)(\partial_t-a_-),\qquad b=\lvert\eta'\rvert,\quad a_\pm=\tau\pm b. \tag*{(228)} L b = − ∂ t 2 + 2 τ ∂ t + b 2 − τ 2 = − ( ∂ t − a + ) ( ∂ t − a − ) , b = ∣ η ′ ∣ , a ± = τ ± b . (228)
For a nonzero real number a a a set
G a ( t ) = { − 1 t < 0 e a t , a > 0 , 1 t > 0 e a t , a < 0. G_a(t)=
\begin{cases}
-1_{t<0}e^{at}, & a>0,\\
1_{t>0}e^{at}, & a<0.
\end{cases} G a ( t ) = { − 1 t < 0 e a t , 1 t > 0 e a t , a > 0 , a < 0.
Its value has jump one at zero, so ( ∂ t − a ) G a = δ 0 (\partial_t-a)G_a=\delta_0 ( ∂ t − a ) G a = δ 0 distributionally, where δ 0 \delta_0 δ 0 is the Dirac mass at t = 0 t=0 t = 0 . For b ≠ τ b\ne\tau b = τ put K b = − G a + ∗ G a − K_b=-G_{a+}*G_{a-} K b = − G a + ∗ G a − . Both factors are integrable for these fibers. Direct convolution gives
K b ( t ) = { 1 t < 0 e ( τ + b ) t − e ( τ − b ) t 2 b , 0 < b < τ , 1 2 b { e ( τ + b ) t , t < 0 , e ( τ − b ) t , t > 0 , b > τ , 1 t < 0 t e τ t , b = 0. (229) K_b(t)=
\begin{cases}
1_{t<0}\dfrac{e^{(\tau+b)t}-e^{(\tau-b)t}}{2b}, & 0<b<\tau,\\
\dfrac{1}{2b}
\begin{cases}
e^{(\tau+b)t}, & t<0,\\
e^{(\tau-b)t}, & t>0,
\end{cases}
& b>\tau,\\
1_{t<0}t e^{\tau t}, & b=0.
\end{cases} \tag*{(229)} K b ( t ) = ⎩ ⎨ ⎧ 1 t < 0 2 b e ( τ + b ) t − e ( τ − b ) t , 2 b 1 { e ( τ + b ) t , e ( τ − b ) t , t < 0 , t > 0 , 1 t < 0 t e τ t , 0 < b < τ , b > τ , b = 0. (229)
In each case K b K_b K b is continuous at zero and K b ′ ( 0 + ) − K b ′ ( 0 − ) = − 1 K_b'(0+)-K_b'(0-)=-1 K b ′ ( 0 + ) − K b ′ ( 0 − ) = − 1 . The regular part solves L b K b = 0 L_bK_b=0 L b K b = 0 away from zero; the distributional second derivative therefore gives L b K b = δ 0 L_bK_b=\delta_0 L b K b = δ 0 , including the double-root fiber b = 0 b=0 b = 0 .
The characteristic fiber b = τ b=\tau b = τ has zero measure in the two tangential frequency variables. The construction uses only the zeroth-order frequency multiplier, since radial-frequency derivatives may fail to be uniformly bounded across the critical circle b = τ b=\tau b = τ . On this fiber one may take either of the following representatives:
K τ − ( t ) = 1 t < 0 ( e 2 τ t − 1 ) 2 τ , K τ + ( t ) = 1 t < 0 e 2 τ t + 1 t > 0 2 τ . (230) K_\tau^-(t)=\frac{1_{t<0}(e^{2\tau t}-1)}{2\tau},
\qquad
K_\tau^+(t)=\frac{1_{t<0}e^{2\tau t}+1_{t>0}}{2\tau}. \tag*{(230)} K τ − ( t ) = 2 τ 1 t < 0 ( e 2 τ t − 1 ) , K τ + ( t ) = 2 τ 1 t < 0 e 2 τ t + 1 t > 0 . (230)
Both are continuous representatives at zero, both have derivative jump − 1 -1 − 1 , and both solve L τ K = δ 0 L_\tau K=\delta_0 L τ K = δ 0 . Their difference is the constant 1 / ( 2 τ ) 1/(2\tau) 1/ ( 2 τ ) . They are the respective one-sided limits of (B.4). Choosing values on the circle b = τ b=\tau b = τ does not change the direct integral with respect to d η ′ d\eta' d η ′ ; it does not authorize an additional distribution in the b b b variable.
The factorization also proves bounds uniform through the critical circle:
∥ G a + ∥ 1 = ( τ + b ) − 1 , ∥ G a − ∥ ∞ ≤ 1 , K b ′ = a − K b − G a + , (231) \|G_{a+}\|_1=(\tau+b)^{-1},
\qquad
\|G_{a-}\|_\infty\le1,
\qquad
K_b'=a_-K_b-G_{a+}, \tag*{(231)} ∥ G a + ∥ 1 = ( τ + b ) − 1 , ∥ G a − ∥ ∞ ≤ 1 , K b ′ = a − K b − G a + , (231)
and hence
∣ K b ( t ) ∣ ≤ ( τ + b ) − 1 , b ∣ K b ( t ) ∣ ≤ 1 , ∣ K b ′ ( t ) ∣ ≤ 2. |K_b(t)|\le(\tau+b)^{-1},
\qquad
b|K_b(t)|\le1,
\qquad
|K_b'(t)|\le2. ∣ K b ( t ) ∣ ≤ ( τ + b ) − 1 , b ∣ K b ( t ) ∣ ≤ 1 , ∣ K b ′ ( t ) ∣ ≤ 2.
for almost every tangential frequency. The same inequalities hold for the chosen critical representatives. In particular the apparent factor 1 / ( τ − b ) 1/(\tau-b) 1/ ( τ − b ) never enters these bounds.
Let w ( t ) = ⟨ t ⟩ − 1 w(t)=\langle t\rangle^{-1} w ( t ) = ⟨ t ⟩ − 1 , and for a fiber source h h h write h = w g h=wg h = w g , so ∥ g ∥ 2 = ∥ ⟨ t ⟩ h ∥ 2 \|g\|_2=\|\langle t\rangle h\|_2 ∥ g ∥ 2 = ∥ ⟨ t ⟩ h ∥ 2 . The kernel from g g g to w ( K b ∗ h ) w(K_b*h) w ( K b ∗ h ) is w ( t ) K b ( t − s ) w ( s ) w(t)K_b(t-s)w(s) w ( t ) K b ( t − s ) w ( s ) . Its Hilbert–Schmidt norm is at most
∥ w ∥ 2 2 τ + b . \frac{\|w\|_2^2}{\tau+b}. τ + b ∥ w ∥ 2 2 .
For the differentiated kernels b K b bK_b b K b and K b ′ K_b' K b ′ the corresponding bounds are ∥ w ∥ 2 2 \|w\|_2^2 ∥ w ∥ 2 2 and 2 ∥ w ∥ 2 2 2\|w\|_2^2 2∥ w ∥ 2 2 . Indeed these statements follow by squaring the pointwise kernel bounds and integrating w ( t ) 2 w ( s ) 2 w(t)^2w(s)^2 w ( t ) 2 w ( s ) 2 . Tangential Plancherel, applied to v ~ ( η ′ , t ) = K ∣ η ′ ∣ ∗ F ~ ( η ′ , ⋅ ) ( t ) \widetilde v(\eta',t)=K_{|\eta'|}*\widetilde F(\eta',\cdot)(t) v ( η ′ , t ) = K ∣ η ′ ∣ ∗ F ( η ′ , ⋅ ) ( t ) , now gives
τ ∥ w v ∥ 2 ≤ C ∥ ⟨ t ⟩ F ∥ 2 , \tau\|wv\|_2\le C\|\langle t\rangle F\|_2, τ ∥ w v ∥ 2 ≤ C ∥ ⟨ t ⟩ F ∥ 2 ,
∥ w ∂ t v ∥ 2 + ∥ w ∂ x 1 v ∥ 2 + ∥ w ∂ x 3 v ∥ 2 ≤ C ∥ ⟨ t ⟩ F ∥ 2 . \|w\partial_t v\|_2+\|w\partial_{x_1}v\|_2+\|w\partial_{x_3}v\|_2
\le C\|\langle t\rangle F\|_2. ∥ w ∂ t v ∥ 2 + ∥ w ∂ x 1 v ∥ 2 + ∥ w ∂ x 3 v ∥ 2 ≤ C ∥ ⟨ t ⟩ F ∥ 2 .
Undoing the modulation replaces η 1 \eta_1 η 1 by η 1 − τ \eta_1-\tau η 1 − τ ; its extra τ K b \tau K_b τ K b term is bounded by one. Finally
⟨ x ⟩ − 1 ≤ ⟨ t ⟩ − 1 , ⟨ t ⟩ ≤ ⟨ x ⟩ \langle x\rangle^{-1}\le\langle t\rangle^{-1},
\qquad
\langle t\rangle\le\langle x\rangle ⟨ x ⟩ − 1 ≤ ⟨ t ⟩ − 1 , ⟨ t ⟩ ≤ ⟨ x ⟩
prove (B.1) for this direct-integral operator.
It remains to identify this operator with exactly the stated multiplier. Put s = η 2 s=\eta_2 s = η 2 . On every noncritical fiber, direct Fourier transformation of the integrable kernels gives
K b ^ ( s ) = 1 s 2 + 2 i τ s + b 2 − τ 2 . (232) \widehat{K_b}(s)=\frac{1}{s^2+2i\tau s+b^2-\tau^2}. \tag*{(232)} K b ( s ) = s 2 + 2 i τ s + b 2 − τ 2 1 . (232)
The joint characteristic set in ( η ′ , s ) (\eta',s) ( η ′ , s ) is { b = τ , s = 0 } \{b=\tau,s=0\} { b = τ , s = 0 } . Near it, the map
( b , s ) ⟼ ( u , v ) = ( b 2 + s 2 − τ 2 , 2 τ s ) (b,s)\longmapsto(u,v)=(b^2+s^2-\tau^2,2\tau s) ( b , s ) ⟼ ( u , v ) = ( b 2 + s 2 − τ 2 , 2 τ s )
has Jacobian 4 b τ ≠ 0 4b\tau\ne0 4 b τ = 0 . In tubular charts the absolute reciprocal is therefore bounded by a constant times ( u 2 + v 2 ) − 1 / 2 (u^2+v^2)^{-1/2} ( u 2 + v 2 ) − 1/2 . Its normal integral is bounded by C ∫ 0 ρ ρ − 1 d ρ < ∞ C\int_0^\rho\rho^{-1}\,d\rho<\infty C ∫ 0 ρ ρ − 1 d ρ < ∞ . The measure in tangential polar coordinates is b d b d θ b\,db\,d\theta b d b d θ ; b = τ b=\tau b = τ causes no degeneracy. At b = 0 b=0 b = 0 the denominator has no zero for real s s s . Outside a ball depending on the fixed τ \tau τ , the denominator has size at least c ∣ η ′ ∣ 2 c|\eta'|^2 c ∣ η ′ ∣ 2 . Consequently the reciprocal is a locally integrable tempered function, and its product with any Schwartz test is absolutely integrable.
The direct-integral distribution pairs against a Schwartz function Φ ( η ′ , t ) \Phi(\eta',t) Φ ( η ′ , t ) by an absolutely convergent integral:
∬ ∣ K b ( t ) Φ ( η ′ , t ) ∣ d t d η ′ ≤ ∫ 1 τ + b ∫ ∣ Φ ( η ′ , t ) ∣ d t d η ′ < ∞ . \iint|K_b(t)\Phi(\eta',t)|\,dt\,d\eta'\le\int\frac{1}{\tau+b}\int|\Phi(\eta',t)|\,dt\,d\eta'<\infty. ∬ ∣ K b ( t ) Φ ( η ′ , t ) ∣ d t d η ′ ≤ ∫ τ + b 1 ∫ ∣Φ ( η ′ , t ) ∣ d t d η ′ < ∞.
For almost every η ′ \eta' η ′ , apply (B.7) to the one-dimensional Schwartz test. The integral of the resulting frequency pairing is absolutely convergent by the preceding local-integrability and large-frequency bounds. Fubini therefore identifies the full Fourier transform of this direct integral with the ordinary function in (B.7); there is no residual distribution supported on b = τ , s = 0 b=\tau,s=0 b = τ , s = 0 . Undoing the modulation shifts η 1 = ξ 1 + τ \eta_1=\xi_1+\tau η 1 = ξ 1 + τ , giving precisely m δ = 1 / p δ m_\delta=1/p_\delta m δ = 1/ p δ . Since p δ m δ = 1 p_\delta m_\delta=1 p δ m δ = 1 almost everywhere, the same identity holds as distributions. This proves (B.2) on Schwartz inputs.
Smooth compactly supported functions are dense in L 1 2 L^2_1 L 1 2 : first truncate with a smooth cutoff at radius R R R , then mollify on the resulting compact set. The bound extends the same operator to all of L 1 2 L^2_1 L 1 2 . The limiting weighted derivative norms define the claimed H − 1 1 H^1_{-1} H − 1 1 output, and local L 2 L^2 L 2 convergence implies distributional convergence; hence the right identity also extends. One may still interpret the extension as the multiplier product: f ∈ L 1 2 f\in L^2_1 f ∈ L 1 2 implies f ^ ∈ H 1 \widehat f\in H^1 f ∈ H 1 , so locally f ^ ∈ L 6 \widehat f\in L^6 f ∈ L 6 , while m δ ∈ L l o c 6 / 5 m_\delta\in L^{6/5}_{\mathrm{loc}} m δ ∈ L loc 6/5 by its two normal-dimensional singularity. Hölder’s inequality controls the product and its approximation by Schwartz inputs. At infinity the quadratic denominator and Cauchy–Schwarz suffice. This also identifies the extension with the multiplier product. □
Fourier kernel estimates with finite derivatives
For a fixed compact set K ⊂ R 3 K\subset\mathbb{R}^3 K ⊂ R 3 and an m × m m\times m m × m matrix symbol supported in K K K in the x x x variable, define for nonnegative integers N , r N,r N , r
s N , r ( a ) = max ∣ α ∣ ≤ N , ∣ β ∣ ≤ r sup x , ξ ∥ ∂ x α ∂ ξ β a ( x , ξ ) ∥ . s_{N,r}(a)=\max_{|\alpha|\le N,\,|\beta|\le r}\sup_{x,\xi}\|\partial_x^\alpha\partial_\xi^\beta a(x,\xi)\|. s N , r ( a ) = ∣ α ∣ ≤ N , ∣ β ∣ ≤ r max x , ξ sup ∥ ∂ x α ∂ ξ β a ( x , ξ ) ∥.
The symbols are assumed smooth whenever an exact Schwartz identity is written; the bounds and extensions below require only the displayed finite derivatives. Constants depend on K K K , the dimension, and the selected matrix norm, but not on the input support or an unlisted frequency cutoff. Left quantization is
Op ( a ) u ( x ) = ( 2 π ) − 3 ∫ e i x ⋅ ξ a ( x , ξ ) u ^ ( ξ ) d ξ . \operatorname{Op}(a)u(x)=(2\pi)^{-3}\int e^{ix\cdot\xi}a(x,\xi)\widehat{u}(\xi)\,d\xi. Op ( a ) u ( x ) = ( 2 π ) − 3 ∫ e i x ⋅ ξ a ( x , ξ ) u ( ξ ) d ξ .
Lemma B.2. For j = 0 , 1 j=0,1 j = 0 , 1 ,
∥ Op ( a ) ∥ H j → H j ≤ C K s 6 , 0 ( a ) . (233) \|\operatorname{Op}(a)\|_{H^j\to H^j}\le C_Ks_{6,0}(a). \tag*{(233)} ∥ Op ( a ) ∥ H j → H j ≤ C K s 6 , 0 ( a ) . (233)
Proof. As in § 5.2 , a ^ ( v , ξ ) \widehat{a}(v,\xi) a ( v , ξ ) denotes Fourier transformation in x x x alone. Applying ( 1 − Δ x ) 3 (1-\Delta_x)^3 ( 1 − Δ x ) 3 before integration gives
∥ a ^ ( v , ξ ) ∥ ≤ C K s 6 , 0 ( a ) ⟨ v ⟩ − 6 . (234) \|\widehat{a}(v,\xi)\| \le C_K s_{6,0}(a)\langle v\rangle^{-6}. \tag*{(234)} ∥ a ( v , ξ ) ∥ ≤ C K s 6 , 0 ( a ) ⟨ v ⟩ − 6 . (234)
Here compact support bounds the L x 1 L_x^1 L x 1 norms of all derivatives in that sixth-order expression. The full Fourier kernel is ( 2 π ) − 3 a ^ ( η − ξ , ξ ) (2\pi)^{-3}\widehat{a}(\eta-\xi,\xi) ( 2 π ) − 3 a ( η − ξ , ξ ) . After conjugation by the Bessel weights of H j H^j H j , its norm is at most
C K s 6 , 0 ( a ) ⟨ η − ξ ⟩ j − 6 , C_Ks_{6,0}(a)\langle\eta-\xi\rangle^{j-6}, C K s 6 , 0 ( a ) ⟨ η − ξ ⟩ j − 6 ,
because ⟨ η ⟩ j / ⟨ ξ ⟩ j ≤ C j ⟨ η − ξ ⟩ j \langle\eta\rangle^j/\langle\xi\rangle^j\le C_j\langle\eta-\xi\rangle^j ⟨ η ⟩ j / ⟨ ξ ⟩ j ≤ C j ⟨ η − ξ ⟩ j . Both Schur integrals are bounded by C K s 6 , 0 ( a ) ∫ ⟨ v ⟩ j − 6 d v < ∞ C_Ks_{6,0}(a)\int\langle v\rangle^{j-6}\,dv<\infty C K s 6 , 0 ( a ) ∫ ⟨ v ⟩ j − 6 d v < ∞ . This proves the claim first for Schwartz inputs and then by density in H j H^j H j . □
Proposition B.3. For a compact spatial symbol with s 6 , 2 ( a ) < ∞ s_{6,2}(a)<\infty s 6 , 2 ( a ) < ∞ and j = 0 , 1 j=0,1 j = 0 , 1 ,
∥ Op ( a ) u ∥ H j ≤ C K s 6 , 2 ( a ) ∥ u ∥ H − 1 j . (235) \|\operatorname{Op}(a)u\|_{H^j}\le C_Ks_{6,2}(a)\|u\|_{H^j_{-1}}. \tag*{(235)} ∥ Op ( a ) u ∥ H j ≤ C K s 6 , 2 ( a ) ∥ u ∥ H − 1 j . (235) The extension from Schwartz inputs is unique, has spatial support in K K K , and retains all contributions of the input outside K K K . On Schwartz inputs the exact identity is
Op ( a ) ⟨ x ⟩ 2 = ⟨ x ⟩ 2 Op ( a ) − 2 i ∑ k = 1 3 x k Op ( ∂ ξ k a ) − Op ( Δ ξ a ) . (236) \operatorname{Op}(a)\langle x\rangle^2=\langle x\rangle^2\operatorname{Op}(a)-2i\sum_{k=1}^{3}x_k\operatorname{Op}(\partial_{\xi_k}a)-\operatorname{Op}(\Delta_\xi a). \tag*{(236)} Op ( a ) ⟨ x ⟩ 2 = ⟨ x ⟩ 2 Op ( a ) − 2 i k = 1 ∑ 3 x k Op ( ∂ ξ k a ) − Op ( Δ ξ a ) . (236) Proof. Since x k u ^ = i ∂ ξ k u ^ \widehat{x_k u}=i\partial_{\xi_k}\widehat{u} x k u = i ∂ ξ k u , integration by parts gives Op ( a ) x k = x k Op ( a ) − i Op ( ∂ ξ k a ) \operatorname{Op}(a)x_k=x_k\operatorname{Op}(a)-i\operatorname{Op}(\partial_{\xi_k}a) Op ( a ) x k = x k Op ( a ) − i Op ( ∂ ξ k a ) . Apply this identity twice and sum over k k k to obtain (B.11), with the displayed signs and no remainder. All integrations are justified by Schwartz decay and bounded symbol derivatives.
For u ∈ H − 1 j u\in H^j_{-1} u ∈ H − 1 j set v = ⟨ x ⟩ − 2 u v=\langle x\rangle^{-2}u v = ⟨ x ⟩ − 2 u . For j = 0 j=0 j = 0 , ∥ v ∥ 2 ≤ ∥ ⟨ x ⟩ − 1 u ∥ 2 \|v\|_2\le\|\langle x\rangle^{-1}u\|_2 ∥ v ∥ 2 ≤ ∥ ⟨ x ⟩ − 1 u ∥ 2 . For j = 1 j=1 j = 1 use
∂ k v = ⟨ x ⟩ − 2 ∂ k u − 2 x k ⟨ x ⟩ − 4 u \partial_kv=\langle x\rangle^{-2}\partial_ku-2x_k\langle x\rangle^{-4}u ∂ k v = ⟨ x ⟩ − 2 ∂ k u − 2 x k ⟨ x ⟩ − 4 u to get ∥ v ∥ H j ≤ C ∥ u ∥ H − 1 j \|v\|_{H^j}\le C\|u\|_{H^j_{-1}} ∥ v ∥ H j ≤ C ∥ u ∥ H − 1 j . Apply Lemma B.2 to a a a , each ∂ ξ k a \partial_{\xi_k}a ∂ ξ k a , and Δ ξ a \Delta_\xi a Δ ξ a in (B.11). Their outputs are all supported in K K K . The factors x k x_k x k and ⟨ x ⟩ 2 \langle x\rangle^2 ⟨ x ⟩ 2 are therefore bounded multipliers on these H j H^j H j outputs, including their first derivatives. This proves (B.10) without differentiating the symbol seven times in x x x .
To justify the extension, C c ∞ C^\infty_c C c ∞ is dense also in H − 1 1 H^1_{-1} H − 1 1 . For a smooth radial cutoff χ ( x / R ) \chi(x/R) χ ( x / R ) , the derivative error R − 1 ( ∂ χ ) ( x / R ) u R^{-1}(\partial\chi)(x/R)u R − 1 ( ∂ χ ) ( x / R ) u has weighted L 2 L^2 L 2 norm bounded by C R − 1 ∥ 1 R ≲ ∣ x ∣ ≲ 2 R ⟨ x ⟩ − 1 u ∥ 2 → 0 CR^{-1}\|1_{R\lesssim|x|\lesssim2R}\langle x\rangle^{-1}u\|_2\to0 C R − 1 ∥ 1 R ≲ ∣ x ∣ ≲ 2 R ⟨ x ⟩ − 1 u ∥ 2 → 0 ; the other terms follow by tail convergence. Mollification after truncation completes the density proof. The weighted estimate makes the operator limit independent of the approximating sequence. Distributional limits of fields supported in the fixed closed set K K K remain supported there. The estimate thus applies to the full weighted input. □
Localized Fourier multipliers. For ψ ∈ C c ∞ \psi\in C^\infty_c ψ ∈ C c ∞ and a Fourier multiplier a 0 ( ξ ) a_0(\xi) a 0 ( ξ ) , ψ Op ( a 0 ) = Op ( ψ ( x ) a 0 ( ξ ) ) \psi\operatorname{Op}(a_0)=\operatorname{Op}(\psi(x)a_0(\xi)) ψ Op ( a 0 ) = Op ( ψ ( x ) a 0 ( ξ )) . Proposition B.3 therefore applies whenever the stated seminorms of this compact spatial symbol are finite.
Proposition B.4. Let a , c a,c a , c be matrix symbols whose spatial supports lie in one fixed compact set K K K , and let
A 1 = max k = 1 , 2 , 3 s 6 , 0 ( ∂ ξ k a ) , C 0 = s 6 , 0 ( c ) . A_1=\max_{k=1,2,3}s_{6,0}(\partial_{\xi_k}a),\qquad C_0=s_{6,0}(c). A 1 = k = 1 , 2 , 3 max s 6 , 0 ( ∂ ξ k a ) , C 0 = s 6 , 0 ( c ) . For j = 0 , 1 j=0,1 j = 0 , 1 ,
∥ Op ( a ) Op ( c ) − Op ( a c ) ∥ H j → H j ≤ C K A 1 C 0 . (237) \|\operatorname{Op}(a)\operatorname{Op}(c)-\operatorname{Op}(ac)\|_{H^j\to H^j}\le C_KA_1C_0. \tag*{(237)} ∥ Op ( a ) Op ( c ) − Op ( a c ) ∥ H j → H j ≤ C K A 1 C 0 . (237) In particular a uniform bound A 1 ≤ C ε A_1\le C\varepsilon A 1 ≤ C ε gives a defect O ( ε ) O(\varepsilon) O ( ε ) when C 0 C_0 C 0 is uniformly bounded. The same conclusion holds if the left symbol is a Fourier multiplier a 0 ( ξ ) a_0(\xi) a 0 ( ξ ) , with A 1 A_1 A 1 replaced by ∥ ∇ ξ a 0 ∥ ∞ \|\nabla_\xi a_0\|_\infty ∥ ∇ ξ a 0 ∥ ∞ . A right Fourier multiplier has zero composition defect in left quantization.
Proof. On Schwartz inputs the exact full Fourier kernel of the defect, including the Fourier normalization, is
( 2 π ) − 6 ∫ [ a ^ ( η − ζ , ζ ) − a ^ ( η − ζ , ξ ) ] c ^ ( ζ − ξ , ξ ) d ζ . (238) (2\pi)^{-6}\int[\widehat{a}(\eta-\zeta,\zeta)-\widehat{a}(\eta-\zeta,\xi)]\widehat{c}(\zeta-\xi,\xi)\,d\zeta. \tag*{(238)} ( 2 π ) − 6 ∫ [ a ( η − ζ , ζ ) − a ( η − ζ , ξ )] c ( ζ − ξ , ξ ) d ζ . (238)
The product a c ac a c retains the matrix order in (B.13). The identity is first established with compact frequency cutoffs, which are removed using the integrable bounds below. The mean value theorem at fixed first Fourier variable, followed by (B.9) for ∂ ξ k a \partial_{\xi_k}a ∂ ξ k a , bounds (B.13) by
C K A 1 C 0 ∫ ∣ ζ − ξ ∣ ⟨ η − ζ ⟩ − 6 ⟨ ζ − ξ ⟩ − 6 d ζ . C_K A_1 C_0\int|\zeta-\xi|\langle\eta-\zeta\rangle^{-6}\langle\zeta-\xi\rangle^{-6}\,d\zeta. C K A 1 C 0 ∫ ∣ ζ − ξ ∣ ⟨ η − ζ ⟩ − 6 ⟨ ζ − ξ ⟩ − 6 d ζ .
For the H j H^j H j conjugation use
⟨ η ⟩ j ⟨ ξ ⟩ j ≤ C j ⟨ η − ζ ⟩ j ⟨ ζ − ξ ⟩ j . \frac{\langle\eta\rangle^j}{\langle\xi\rangle^j}\le C_j\langle\eta-\zeta\rangle^j\langle\zeta-\xi\rangle^j. ⟨ ξ ⟩ j ⟨ η ⟩ j ≤ C j ⟨ η − ζ ⟩ j ⟨ ζ − ξ ⟩ j .
The weighted kernel is thus bounded by C K A 1 C 0 ( F j ∗ G j ) ( η − ξ ) C_K A_1 C_0(F_j*G_j)(\eta-\xi) C K A 1 C 0 ( F j ∗ G j ) ( η − ξ ) , where
F j ( v ) = ⟨ v ⟩ j − 6 , G j ( v ) = ∣ v ∣ ⟨ v ⟩ j − 6 . F_j(v)=\langle v\rangle^{j-6},\qquad G_j(v)=|v|\langle v\rangle^{j-6}. F j ( v ) = ⟨ v ⟩ j − 6 , G j ( v ) = ∣ v ∣ ⟨ v ⟩ j − 6 .
Both functions are integrable in dimension three for j = 0 , 1 j=0,1 j = 0 , 1 ; for j = 1 j=1 j = 1 their large-radius powers are − 5 -5 − 5 and − 4 -4 − 4 . The two Schur integrals are at most C K A 1 C 0 ∥ F j ∥ 1 ∥ G j ∥ 1 C_K A_1 C_0\|F_j\|_1\|G_j\|_1 C K A 1 C 0 ∥ F j ∥ 1 ∥ G j ∥ 1 . This proves (B.12). If the left symbol is a multiplier, the kernel is instead
( 2 π ) − 3 [ a 0 ( η ) − a 0 ( ξ ) ] c ^ ( η − ξ , ξ ) . (2\pi)^{-3}[a_0(\eta)-a_0(\xi)]\widehat{c}(\eta-\xi,\xi). ( 2 π ) − 3 [ a 0 ( η ) − a 0 ( ξ )] c ( η − ξ , ξ ) .
Its weighted bound is C K ∥ ∇ a 0 ∥ ∞ C 0 ∣ η − ξ ∣ ⟨ η − ξ ⟩ j − 6 C_K\|\nabla a_0\|_\infty C_0|\eta-\xi|\langle\eta-\xi\rangle^{j-6} C K ∥∇ a 0 ∥ ∞ C 0 ∣ η − ξ ∣ ⟨ η − ξ ⟩ j − 6 , which has finite Schur integrals by the same powers. If the right symbol is a multiplier c 0 ( ξ ) c_0(\xi) c 0 ( ξ ) , multiplication of u ^ \widehat{u} u by c 0 c_0 c 0 before left quantization gives Op ( a ) Op ( c 0 ) = Op ( a c 0 ) \operatorname{Op}(a)\operatorname{Op}(c_0)=\operatorname{Op}(ac_0) Op ( a ) Op ( c 0 ) = Op ( a c 0 ) exactly. These statements also handle symbols that are finite sums of compact spatial symbols and Fourier multipliers, by bilinearity; only the left-multiplier/right-compact and compact/compact defects remain. □
Corollary B.5. Let a τ a_\tau a τ have a fixed compact spatial support and sup τ ≥ 1 s 6 , 2 ( a τ ) < ∞ \sup_{\tau\ge1}s_{6,2}(a_\tau)<\infty sup τ ≥ 1 s 6 , 2 ( a τ ) < ∞ . Then the same composed operator satisfies
τ ∥ Op ( a τ ) E δ f ∥ 2 + ∥ Op ( a τ ) E δ f ∥ H 1 ≤ C ∥ f ∥ L 2 . \tau\|\operatorname{Op}(a_\tau)E_\delta f\|_2+\|\operatorname{Op}(a_\tau)E_\delta f\|_{H^1}\le C\|f\|_{L^2}. τ ∥ Op ( a τ ) E δ f ∥ 2 + ∥ Op ( a τ ) E δ f ∥ H 1 ≤ C ∥ f ∥ L 2 . If f f f has support in a fixed compact source set, its L 1 L^1 L 1 norm can be replaced on the right by a fixed constant times ∥ f ∥ 2 \|f\|_2 ∥ f ∥ 2 . Moreover, let r τ ( x , ξ ) r_\tau(x,\xi) r τ ( x , ξ ) be a family of residual matrix symbols with spatial support in a fixed compact set. If, for some γ ∈ R \gamma\in\mathbb{R} γ ∈ R , s 6 , 2 ( r τ ) ≤ C τ γ s_{6,2}(r_\tau)\le C\tau^\gamma s 6 , 2 ( r τ ) ≤ C τ γ , then
∥ Op ( r τ ) E δ ∥ L 1 2 → L 2 ≤ C τ γ − 1 . \|\operatorname{Op}(r_\tau)E_\delta\|_{L^2_1\to L^2}\le C\tau^{\gamma-1}. ∥ Op ( r τ ) E δ ∥ L 1 2 → L 2 ≤ C τ γ − 1 .
Proof. Apply Proposition B.3 at j = 0 j=0 j = 0 to the first free bound and at j = 1 j=1 j = 1 to the second free bound. For the residual use only j = 0 j=0 j = 0 , so the growth τ γ \tau^\gamma τ γ is multiplied by the free gain τ − 1 \tau^{-1} τ − 1 . The compact-source comparison is ∥ f ∥ L 1 2 ≤ ( sup K s r c ⟨ x ⟩ ) ∥ f ∥ 2 \|f\|_{L^2_1}\le(\sup_{K_{\rm src}}\langle x\rangle)\|f\|_2 ∥ f ∥ L 1 2 ≤ ( sup K src ⟨ x ⟩) ∥ f ∥ 2 . □
Remark B.6. Six spatial derivatives suffice in all ordinary, weighted-to-local, and composition bounds at j = 0 , 1 j=0,1 j = 0 , 1 . Two frequency derivatives suffice for the weight exchange. The composition estimate uses six spatial derivatives of one frequency derivative of its left symbol and six spatial derivatives of its right symbol; it does not differentiate the singular multiplier 1 / p δ 1/p_\delta 1/ p δ . Applying the six-derivative bound to a residual containing − Δ x b -\Delta_x b − Δ x b requires eight spatial derivatives of b b b ; mixed derivatives such as ∂ x α ∂ ξ β b \partial_x^\alpha\partial_\xi^\beta b ∂ x α ∂ ξ β b , ∣ α ∣ ≤ 8 |\alpha|\le8 ∣ α ∣ ≤ 8 , ∣ β ∣ ≤ 2 |\beta|\le2 ∣ β ∣ ≤ 2 , are the frame bounds used in the localized construction.
Acknowledgments
The work of H. Diao was supported by the National Natural Science Foundation of China (Grant No. 12371422) and the Fundamental Research Funds for the Central Universities, Jilin University.