Mathematics Subject Classification (2020). Primary 35R30; Secondary 74B05, 35J57.
Introduction
The inverse boundary value problem for isotropic elasticity concerns the recovery of the two spatially varying Lamé parameters from boundary displacement and traction measurements. We investigate the stability of this recovery in three dimensions, establishing logarithmic stability for the coefficients in the interior and power-type stability estimates for their boundary derivatives. Both estimates are expressed in terms of the discrepancy between the full Dirichlet-to-Neumann maps on the original boundary.
The elasticity system and the inverse boundary problem
Let Ω ⊂ R 3 \Omega\subset\mathbb{R}^{3} Ω ⊂ R 3 be a bounded domain with smooth boundary, and let λ \lambda λ and μ \mu μ be real-valued Lamé parameters. For a displacement field u : Ω → C 3 u:\Omega\to\mathbb{C}^{3} u : Ω → C 3 , the strain tensor, stress tensor, and elasticity operator are given by
ε ( u ) = 1 2 ( ∇ u + ∇ u T ) , σ λ , μ ( u ) = λ ( div u ) I 3 + 2 μ ε ( u ) , L λ , μ u = div σ λ , μ ( u ) . (1) \begin{aligned}
\varepsilon(u) &= \frac{1}{2}\left(\nabla u+\nabla u^{T}\right),\\
\sigma_{\lambda,\mu}(u) &= \lambda(\operatorname{div}u)I_{3}+2\mu\varepsilon(u),\\
L_{\lambda,\mu}u &= \operatorname{div}\sigma_{\lambda,\mu}(u). \tag*{(1)}
\end{aligned} ε ( u ) σ λ , μ ( u ) L λ , μ u = 2 1 ( ∇ u + ∇ u T ) , = λ ( div u ) I 3 + 2 μ ε ( u ) , = div σ λ , μ ( u ) . (1)
where I 3 I_{3} I 3 denotes the 3 × 3 3 \times3 3 × 3 identity matrix. We consider the static isotropic elasticity system
{ div ( λ ( div u ) I 3 + μ ( ∇ u + ∇ u T ) ) = 0 in Ω , u = f on ∂ Ω . (2) \begin{cases}
\operatorname{div}\left(\lambda(\operatorname{div}u)I_{3}+\mu(\nabla u+\nabla u^{T})\right)=0 & \text{in }\Omega,\\
u=f & \text{on }\partial\Omega. \tag*{(2)}
\end{cases} { div ( λ ( div u ) I 3 + μ ( ∇ u + ∇ u T ) ) = 0 u = f in Ω , on ∂ Ω. (2)
Throughout the paper, we use the component conventions
( ∇ u ) i j = ∂ j u i , ( a ⊗ b ) i j = a i b j , A : B = ∑ i , j A i j B i j . (3) (\nabla u)_{ij}=\partial_{j}u_{i},\qquad(a\otimes b)_{ij}=a_{i}b_{j},\qquad A:B=\sum_{i,j}A_{ij}B_{ij}. \tag*{(3)} ( ∇ u ) ij = ∂ j u i , ( a ⊗ b ) ij = a i b j , A : B = i , j ∑ A ij B ij . (3)
For an integer s ≥ 2 s \ge2 s ≥ 2 and fixed constants M , c 0 > 0 M,c_{0}>0 M , c 0 > 0 , define the admissible class
C s ( Ω ; M , c 0 ) = { ( λ , μ ) ∈ C s ( Ω ‾ ; R ) 2 : μ ≥ c 0 , 3 λ + 2 μ ≥ c 0 , ∥ λ ∥ C s ( Ω ‾ ) + ∥ μ ∥ C s ( Ω ‾ ) ≤ M } . (4) \mathcal{C}^{s}(\Omega;M,c_{0})=\left\{(\lambda,\mu)\in C^{s}(\overline{\Omega};\mathbb{R})^{2}:\ \mu\ge c_{0},\quad3\lambda+2\mu\ge c_{0},\quad\|\lambda\|_{C^{s}(\overline{\Omega})}+\|\mu\|_{C^{s}(\overline{\Omega})}\le M\right\}. \tag*{(4)} C s ( Ω ; M , c 0 ) = { ( λ , μ ) ∈ C s ( Ω ; R ) 2 : μ ≥ c 0 , 3 λ + 2 μ ≥ c 0 , ∥ λ ∥ C s ( Ω ) + ∥ μ ∥ C s ( Ω ) ≤ M } . (4)
The inequalities in (4) hold pointwise on Ω ‾ \overline{\Omega} Ω . The strong-convexity conditions μ ≥ c 0 \mu\ge c_{0} μ ≥ c 0 and 3 λ + 2 μ ≥ c 0 3\lambda+2\mu\ge c_{0} 3 λ + 2 μ ≥ c 0 in (4) , together with Korn’s inequality [14 ] , ensure that (2) has 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 ) for each f ∈ H 1 / 2 ( ∂ Ω ; C 3 ) f\in H^{1/2}(\partial\Omega;\mathbb{C}^{3}) f ∈ H 1/2 ( ∂ Ω ; C 3 ) (cf. [14 ] ).
Let ν \nu ν denote the outward unit normal to ∂ Ω \partial\Omega ∂ Ω . For each f ∈ H 1 / 2 ( ∂ Ω ; C 3 ) f\in H^{1/2}(\partial\Omega;\mathbb{C}^{3}) f ∈ H 1/2 ( ∂ Ω ; C 3 ) , let u f u_{f} u f be the unique weak solution of (2) . The Dirichlet-to-Neumann (DN) map associates the prescribed boundary displacement with the corresponding elastic traction:
Λ λ , μ Ω : H 1 / 2 ( ∂ Ω ; C 3 ) ⟶ H − 1 / 2 ( ∂ Ω ; C 3 ) , \Lambda_{\lambda,\mu}^{\Omega}:H^{1/2}(\partial\Omega;\mathbb{C}^{3})\longrightarrow H^{-1/2}(\partial\Omega;\mathbb{C}^{3}), Λ λ , μ Ω : H 1/2 ( ∂ Ω ; C 3 ) ⟶ H − 1/2 ( ∂ Ω ; C 3 ) ,
Λ λ , μ Ω f = σ λ , μ ( u f ) ν . (5) \Lambda_{\lambda,\mu}^{\Omega}f=\sigma_{\lambda,\mu}(u_{f})\nu. \tag*{(5)} Λ λ , μ Ω f = σ λ , μ ( u f ) ν . (5)
The traction is understood in the weak sense defined by (19) .
For two admissible coefficient pairs ( λ j , μ j ) ∈ C s ( Ω ; M , c 0 ) (\lambda_{j},\mu_{j})\in\mathcal{C}^{s}(\Omega;M,c_{0}) ( λ j , μ j ) ∈ C s ( Ω ; M , c 0 ) , j = 1 , 2 j=1,2 j = 1 , 2 , we define their boundary measurement discrepancy by
δ : = ∥ Λ λ 1 , μ 1 Ω − Λ λ 2 , μ 2 Ω ∥ H 1 / 2 → H − 1 / 2 . (6) \delta:=\left\|\Lambda_{\lambda_{1},\mu_{1}}^{\Omega}-\Lambda_{\lambda_{2},\mu_{2}}^{\Omega}\right\|_{H^{1/2}\to H^{-1/2}}. \tag*{(6)} δ := Λ λ 1 , μ 1 Ω − Λ λ 2 , μ 2 Ω H 1/2 → H − 1/2 . (6)
The operator norm in this definition is taken from H 1 / 2 ( ∂ Ω ; C 3 ) H^{1/2}(\partial\Omega;\mathbb{C}^{3}) H 1/2 ( ∂ Ω ; C 3 ) to H − 1 / 2 ( ∂ Ω ; C 3 ) H^{-1/2}(\partial\Omega;\mathbb{C}^{3}) H − 1/2 ( ∂ Ω ; C 3 ) . In particular, δ = 0 \delta=0 δ = 0 if and only if the two full DN maps coincide.
For any smooth domain D D D on which the coefficient pairs are defined, we write
Λ j D = Λ λ j , μ j D , j = 1 , 2. (7) \Lambda_{j}^{D}=\Lambda_{\lambda_{j},\mu_{j}}^{D},\qquad j=1,2. \tag*{(7)} Λ j D = Λ λ j , μ j D , j = 1 , 2. (7)
The inverse boundary value problem is to determine the two Lamé parameters λ \lambda λ and μ \mu μ from the full DN map,
Λ λ , μ Ω ⟼ ( λ , μ ) ∣ Ω , \Lambda_{\lambda,\mu}^{\Omega} \longmapsto(\lambda,\mu)|_{\Omega}, Λ λ , μ Ω ⟼ ( λ , μ ) ∣ Ω ,
and to quantify their recovery in terms of the measurement discrepancy δ \delta δ . In particular, we seek stability estimates for both coefficients in L ∞ ( Ω ) L^\infty(\Omega) L ∞ ( Ω ) and for their boundary derivatives. The measurements are made on the original boundary ∂ Ω \partial\Omega ∂ Ω ; no exterior coefficients are prescribed, and the candidate materials need not have matching boundary values or derivatives.
To formulate the boundary stability estimates, we use the multi-index notation
a ∈ N 0 3 , ∣ a ∣ = a 1 + a 2 + a 3 , ∂ a = ∂ 1 a 1 ∂ 2 a 2 ∂ 3 a 3 . (8) a \in\mathbb{N}_0^3,\qquad|a|=a_1+a_2+a_3,\qquad\partial^a=\partial_1^{a_1}\partial_2^{a_2}\partial_3^{a_3}. \tag*{(8)} a ∈ N 0 3 , ∣ a ∣ = a 1 + a 2 + a 3 , ∂ a = ∂ 1 a 1 ∂ 2 a 2 ∂ 3 a 3 . (8)
For each integer 0 ≤ q ≤ s 0\leq q\leq s 0 ≤ q ≤ s , define
J q = ∑ ∣ a ∣ ≤ q ( ∥ ∂ a ( λ 1 − λ 2 ) ∥ C 0 ( ∂ Ω ) + ∥ ∂ a ( μ 1 − μ 2 ) ∥ C 0 ( ∂ Ω ) ) . (9) J_q=\sum_{|a|\leq q}\left(\|\partial^a(\lambda_1-\lambda_2)\|_{C^0(\partial\Omega)}+\|\partial^a(\mu_1-\mu_2)\|_{C^0(\partial\Omega)}\right). \tag*{(9)} J q = ∣ a ∣ ≤ q ∑ ( ∥ ∂ a ( λ 1 − λ 2 ) ∥ C 0 ( ∂ Ω ) + ∥ ∂ a ( μ 1 − μ 2 ) ∥ C 0 ( ∂ Ω ) ) . (9)
Here the derivatives are the boundary traces of the corresponding Cartesian derivatives of the coefficients in C s ( Ω ‾ ) C^s(\overline{\Omega}) C s ( Ω ) .
Previous results and contributions
The inverse boundary value problem for isotropic elasticity is a system analogue of Calderón’s conductivity problem [10 ] . Sylvester and Uhlmann [32 ] established global uniqueness for smooth positive conductivities in dimensions n ≥ 3 n\geq3 n ≥ 3 using complex geometrical optics (CGO) solutions. In two dimensions, Nachman [27 ] established global uniqueness and reconstruction for smooth conductivities, while Astala and Päivärinta [6 ] studied uniqueness for bounded measurable conductivities. On the quantitative side, Alessandrini [3 ] established logarithmic stability under suitable a priori assumptions, whereas Mandache [25 ] proved exponential instability for the inverse Schrödinger problem, with corresponding consequences for conductivity. The elasticity system introduces additional difficulties through its coupled principal part and the simultaneous variation of two Lamé parameters.
Boundary identification of the Lamé parameters, including stability estimates for their boundary values, was established in [2 ] . Higher-order boundary determination and reconstruction from localized measurements were subsequently developed in [29 , 24 ] . For smooth Lamé parameters on a smooth Riemannian manifold, on a smooth Riemannian manifold, the full symbol of the elastic DN map was computed, and the determination of all boundary derivatives was established in [33 ] . The boundary estimates in the present paper quantify derivative recovery under finite regularity assumptions. Quantitative boundary reconstruction in the presence of random measurement errors was studied in [11 ] . The latter result concerns stochastic perturbations of boundary measurements, rather than the deterministic operator-norm discrepancy considered here.
In two dimensions, Imanuvilov and Yamamoto [23 ] established full-data global uniqueness for the isotropic Lamé system without a smallness assumption on the coefficients. For three-dimensional interior recovery, Nakamura and Uhlmann developed a CGO approach to the elasticity system [28 , 30 ] . Their corrected uniqueness theorem requires sufficiently small gradients of both candidate shear moduli in a sufficiently high regularity norm, without an analogous smallness assumption on the first Lamé parameter. Related CGO constructions and perturbative uniqueness results were obtained in [17 , 19 , 18 ] . More recent uniqueness results for three-dimensional isotropic elasticity impose axisymmetry, variable separation, or directional quasianalyticity assumptions on the shear modulus [13 ] . For partial Cauchy data on an arbitrary open boundary subset, Imanuvilov, Uhlmann, and Yamamoto [22 ] , Theorem 0.1 proved uniqueness when the shear moduli are constant and the two first Lamé parameters agree on the inaccessible part of the boundary.
Quantitative recovery has also been studied under more restrictive a priori assumptions. In [9 ] , uniqueness and Lipschitz stability were established for piecewise constant Lamé parameters on a prescribed partition, under certain geometric assumptions. The analysis was extended to non-flat interfaces in [8 ] . For time-harmonic elastic waves, uniqueness and Lipschitz stability from local DN data were established for piecewise constant Lamé parameters and density on a known partition, under geometric and frequency assumptions [7 ] . Lipschitz stability has also been obtained from Neumann-to-Dirichlet measurements using monotonicity and localized potentials, for finite-dimensional coefficient classes satisfying additional regularity and ordering assumptions [16 ] . These results rely on restricted admissible classes, in contrast to the full bounded regularity classes considered in the present paper.
A related but distinct problem is the determination of an elastic inclusion. Logarithmic stability for recovering an unknown inclusion in a piecewise homogeneous isotropic body was established in [4 ] . In that setting the unknown is the inclusion geometry, whereas the present problem concerns the simultaneous recovery of two spatially varying Lamé parameters.
In our preceding work [15 ] , developed independently of the OpenAI preprint [31 ] , we established full-data global uniqueness for smooth, strongly convex Lamé parameters in three dimensions without smallness or structural assumptions. A central feature of our approach is the derivation of complete, holomorphically weighted 3 × 3 3 \times3 3 × 3 matrix identities from the Betti identity for physical elastic solutions. The missing third physical adjoint polarization is obtained from the first nonvanishing correction to a reduced amplitude annihilated by the leading lift. These matrix identities then provide the compatibility needed to construct an invertible, exterior-normalized relative transport through Cauchy–Plemelj factorization. The present paper develops this physical-solution and matrix-moment approach into a quantitative stability theory.
The main result of this paper is logarithmic stability for the simultaneous recovery of both Lamé parameters from the full Dirichlet-to-Neumann map defined by (5) . Theorem 1.1 establishes the L ∞ ( Ω ) L^\infty(\Omega) L ∞ ( Ω ) estimate (11) uniformly on every fixed admissible class C 22 ( Ω ; M , c 0 ) \mathcal{C}^{22}(\Omega;M,c_0) C 22 ( Ω ; M , c 0 ) , without smallness assumptions on the material gradients, prescribed boundary agreement, or known exterior coefficients. A central ingredient is the quantitative recovery of complete matrix moments from physical elastic solutions (Theorem 4.2). The exact reduction alone does not identify the unrestricted boundary map of the auxiliary system with the measured elastic Dirichlet-to-Neumann map. Quantitative control of the relative transport, together with boundary determination and paired extensions, then yields stability from measurements on the original boundary.
Theorem 1.2 gives a separate family of boundary stability estimates. For every integer s ≥ 2 s \ge2 s ≥ 2 and 0 ≤ q < s 0 \le q < s 0 ≤ q < s , the Cartesian boundary derivatives satisfy (12) , with the power δ ( s − q ) / s \delta^{(s-q)/s} δ ( s − q ) / s . These estimates hold uniformly on each fixed admissible C s \mathcal{C}^{s} C s class, without additional prior-slack assumptions. Proposition 1.3 shows that the powers are optimal on classes containing a constant material satisfying the a priori inequalities strictly. Under the same hypothesis, no uniform vanishing modulus exists for derivatives of order s s s . The sharpness assertion concerns the boundary estimates, not the interior logarithmic exponent.
Finally, the zero-error assertion of Proposition 10.1, with m = 4 m=4 m = 4 , yields global uniqueness for C 22 ( Ω ‾ ) \mathcal{C}^{22}(\overline{\Omega}) C 22 ( Ω ) Lamé parameters (Corollary 1.4). This relaxes the C ∞ \mathcal{C}^{\infty} C ∞ regularity assumptions in the independent uniqueness results [15 , 31 ] . The conclusion follows from the present quantitative argument without invoking either of the preceding uniqueness theorems.
Main results
Theorem 1.1. Fix the bounded smooth domain Ω ⊂ R 3 \Omega\subset\mathbb{R}^{3} Ω ⊂ R 3 and constants M , c 0 > 0 M,c_0>0 M , c 0 > 0 . Let C 22 ( Ω ; M , c 0 ) \mathcal{C}^{22}(\Omega;M,c_0) C 22 ( Ω ; M , c 0 ) be the class defined in (4) with s = 22 s=22 s = 22 . There exists γ ∗ , e x t ∈ ( 0 , 1 ) \gamma_{*,\mathrm{ext}} \in(0,1) γ ∗ , ext ∈ ( 0 , 1 ) , depending only on Ω \Omega Ω , M M M , and c 0 c_0 c 0 , with the following property: for every α \alpha α satisfying
0 < α < γ ∗ , e x t 112 , (10) 0 < \alpha< \frac{\gamma_{*,\mathrm{ext}}}{112}, \tag*{(10)} 0 < α < 112 γ ∗ , ext , (10) there exist constants C α , δ A , 0 > 0 C_\alpha,\delta_{A,0}>0 C α , δ A , 0 > 0 such that
∥ λ 1 − λ 2 ∥ L ∞ ( Ω ) + ∥ μ 1 − μ 2 ∥ L ∞ ( Ω ) ≤ C α ∣ log δ ∣ − α (11) \lVert\lambda_1-\lambda_2\rVert_{L^\infty(\Omega)}+\lVert\mu_1-\mu_2\rVert_{L^\infty(\Omega)} \le C_\alpha\lvert\log\delta\rvert^{-\alpha} \tag*{(11)} ∥ λ 1 − λ 2 ∥ L ∞ ( Ω ) + ∥ μ 1 − μ 2 ∥ L ∞ ( Ω ) ≤ C α ∣ log δ ∣ − α (11)
for every two pairs ( λ j , μ j ) ∈ C 22 ( Ω ; M , c 0 ) (\lambda_j,\mu_j) \in\mathcal{C}^{22}(\Omega;M,c_0) ( λ j , μ j ) ∈ C 22 ( Ω ; M , c 0 ) , j = 1 , 2 j=1,2 j = 1 , 2 , with 0 < δ < δ A , 0 0<\delta<\delta_{A,0} 0 < δ < δ A , 0 . Here δ \delta δ is defined in (6) , and the constants C α , δ A , 0 C_\alpha,\delta_{A,0} C α , δ A , 0 depend only on Ω \Omega Ω , M M M , c 0 c_0 c 0 , and α \alpha α .
Remark. The estimate (11) holds uniformly over the full admissible class C 22 ( Ω ; M , c 0 ) \mathcal{C}^{22}(\Omega;M,c_0) C 22 ( Ω ; M , c 0 ) . No exterior coefficients are prescribed, the boundary traces and derivatives of the two materials need not agree, and no smallness condition is imposed on their gradients.
The C 22 C^{22} C 22 regularity assumption is sufficient for the requirements of the present quantitative construction. Indeed, the C x 3 C_x^3 C x 3 projection estimate (195) , obtained by interpolation, yields a positive power of the error when m > 3 m>3 m > 3 . The mixed-derivative count in (173) , together with the regularity requirements in (94) , gives the sufficient material regularity order s = 2 m + 14 s=2m+14 s = 2 m + 14 . The smallest admissible integer m = 4 m=4 m = 4 therefore yields s = 22 s=22 s = 22 . We explain this derivative count in Subsection 1.4.
Theorem 1.1 follows from the positive-error assertion of Proposition 10.1 with m = 4 m=4 m = 4 . The auxiliary enlarged coefficient class is constructed from C 22 ( Ω ; M , c 0 ) \mathcal{C}^{22}(\Omega;M,c_0) C 22 ( Ω ; M , c 0 ) in Subsection 10.1. Its bounds, geometry, and exponent parameter γ ∗ , e x t \gamma_{*,\mathrm{ext}} γ ∗ , ext depend only on the original a priori data and are independent of the measurement error. In particular, the auxiliary construction imposes no additional assumptions on the original materials.
Boundary derivatives satisfy the following order-dependent estimates.
Theorem 1.2. Fix an integer s ≥ 2 s\ge2 s ≥ 2 and the class C s ( Ω ; M , c 0 ) \mathcal{C}^{s}(\Omega;M,c_0) C s ( Ω ; M , c 0 ) in (4) . There exist constants C , δ B , 0 > 0 C,\delta_{B,0}>0 C , δ B , 0 > 0 , depending only on Ω \Omega Ω , s s s , M M M , c 0 c_0 c 0 , such that for every two members of this class, simultaneously for all integers 0 ≤ q < s 0\le q<s 0 ≤ q < s ,
J q ≤ C δ ( s − q ) / s , q ∈ { 0 , … , s − 1 } , 0 < δ < δ B , 0 . (12) J_q \le C\delta^{(s-q)/s}, \qquad q\in\{0,\ldots,s-1\}, \qquad0<\delta<\delta_{B,0}. \tag*{(12)} J q ≤ C δ ( s − q ) / s , q ∈ { 0 , … , s − 1 } , 0 < δ < δ B , 0 . (12) Here J q J_q J q and δ \delta δ are defined in (9) and (6) , respectively. If δ = 0 \delta=0 δ = 0 , then J q = 0 J_q=0 J q = 0 for every 0 ≤ q < s 0\le q<s 0 ≤ q < s .
Remark. The boundary values satisfy a Lipschitz estimate, while the powers for higher-order derivatives arise from balancing the measurement error against the Taylor remainder in (312) . In particular, the estimate J s − 1 ≤ C δ 1 / s J_{s-1}\le C\delta^{1/s} J s − 1 ≤ C δ 1/ s provides the boundary control required for the paired extension in Section 9. This construction uses the uniform C s C^s C s a priori bound but requires no stability estimate for J s J_s J s .
The exponents in Theorem 1.2 are optimal on admissible classes C s ( Ω ; M , c 0 ) \mathcal{C}^{s}(\Omega;M,c_0) C s ( Ω ; M , c 0 ) containing a constant material satisfying the a priori inequalities strictly.
Proposition 1.3. Let s ≥ 2 s\ge2 s ≥ 2 be an integer, and suppose that C s ( Ω ; M , c 0 ) \mathcal{C}^{s}(\Omega;M,c_0) C s ( Ω ; M , c 0 ) contains a constant pair ( λ b , μ b ) (\lambda_b,\mu_b) ( λ b , μ b ) satisfying
μ b > c 0 , 3 λ b + 2 μ b > c 0 , ∥ λ b ∥ C s ( Ω ‾ ) + ∥ μ b ∥ C s ( Ω ‾ ) < M . (13) \mu_b>c_0,\qquad3\lambda_b+2\mu_b>c_0,\qquad\|\lambda_b\|_{C^s(\overline{\Omega})}+\|\mu_b\|_{C^s(\overline{\Omega})}<M. \tag*{(13)} μ b > c 0 , 3 λ b + 2 μ b > c 0 , ∥ λ b ∥ C s ( Ω ) + ∥ μ b ∥ C s ( Ω ) < M . (13) For each q ∈ { 0 , … , s } q\in\{0,\ldots,s\} q ∈ { 0 , … , s } , there exist constants c q , C q > 0 c_q,C_q>0 c q , C q > 0 and a family of smooth coefficient pairs ( λ a , q , μ a , q ) ∈ C s ( Ω ; M , c 0 ) (\lambda_{a,q},\mu_{a,q})\in\mathcal{C}^{s}(\Omega;M,c_0) ( λ a , q , μ a , q ) ∈ C s ( Ω ; M , c 0 ) , defined for all sufficiently small a > 0 a>0 a > 0 , such that
0 < δ a , q ≤ C q a s , J q , a ≥ c q a s − q . (14) 0<\delta_{a,q}\le C_q a^s,\qquad J_{q,a}\ge c_q a^{s-q}. \tag*{(14)} 0 < δ a , q ≤ C q a s , J q , a ≥ c q a s − q . (14) Here δ a , q \delta_{a,q} δ a , q and J q , a J_{q,a} J q , a denote the discrepancies defined in (6) and (9) , respectively, between ( λ a , q , μ b ) (\lambda_{a,q},\mu_b) ( λ a , q , μ b ) and ( λ b , μ b ) (\lambda_b,\mu_b) ( λ b , μ b ) .
Consequently, for 0 ≤ q ≤ s 0\le q\le s 0 ≤ q ≤ s , no estimate
J q ≤ ω ( δ ) , ω ( t ) = o ( t ( s − q ) / s ) as t ↓ 0 , J_q\le\omega(\delta),\qquad\omega(t)=o(t^{(s-q)/s})\ \text{as }t\downarrow0, J q ≤ ω ( δ ) , ω ( t ) = o ( t ( s − q ) / s ) as t ↓ 0 , holds uniformly on this class. For q = s q=s q = s , no uniform estimate J s ≤ ω ( δ ) J_s\le\omega(\delta) J s ≤ ω ( δ ) is possible with ω ( t ) → 0 \omega(t)\to0 ω ( t ) → 0 as $t\downarrow 0.
The zero-error conclusion of Proposition 10.1 also yields uniqueness under finite regularity.
Corollary 1.4. Let ( λ j , μ j ) ∈ C 22 ( Ω ‾ ; R ) 2 (\lambda_j,\mu_j) \in C^{22}(\overline{\Omega};\mathbb{R})^2 ( λ j , μ j ) ∈ C 22 ( Ω ; R ) 2 , j = 1 , 2 j=1,2 j = 1 , 2 , satisfy
μ j > 0 , 3 λ j + 2 μ j > 0 on Ω ‾ . \mu_j>0,\qquad3\lambda_j+2\mu_j>0\quad\text{on }\overline{\Omega}. μ j > 0 , 3 λ j + 2 μ j > 0 on Ω . If
Λ λ 1 , μ 1 Ω = Λ λ 2 , μ 2 Ω , \Lambda^\Omega_{\lambda_1,\mu_1}=\Lambda^\Omega_{\lambda_2,\mu_2}, Λ λ 1 , μ 1 Ω = Λ λ 2 , μ 2 Ω , then
λ 1 = λ 2 , μ 1 = μ 2 in Ω . \lambda_1=\lambda_2,\qquad\mu_1=\mu_2\quad\text{in }\Omega. λ 1 = λ 2 , μ 1 = μ 2 in Ω.
The boundary statements and the uniqueness corollary are proved in Subsection 10.2.
Regularity requirements
The C 22 C^{22} C 22 regularity assumption in Theorem 1.1 arises from the quantitative transport construction. The coefficient-rigidity theorem (Theorem 7.3) requires the first-order transport residual to be small in C x 2 C_x^2 C x 2 . By (6.21), this requires C x 3 C_x^3 C x 3 control of the projection error. Proposition 6.3 yields the interpolation estimate (6.18),
∥ H − H 0 ∥ C x 3 C ζ 0 ≤ C ω F β m , β m = m − 3 2 m . \lVert H-H_0\rVert_{C_x^3 C_\zeta^0}\leq C\omega_F^{\beta_m},\qquad\beta_m=\frac{m-3}{2m}. ∥ H − H 0 ∥ C x 3 C ζ 0 ≤ C ω F β m , β m = 2 m m − 3 .
Thus, a positive power of the error is obtained precisely when m > 3 m>3 m > 3 .
The remaining regularity requirements arise from mixed spatial and complex-directional derivatives of the transport frames. With N = m + 2 N=m+2 N = m + 2 , the derivative count (5.39) gives
N + ( N + 3 ) = 2 m + 7 < r f r = 2 m + 8. N+(N+3)=2m+7<r_{\mathrm{fr}}=2m+8. N + ( N + 3 ) = 2 m + 7 < r fr = 2 m + 8.
Applying Theorem 3.1 with r = r f r + 2 r=r_{\mathrm{fr}}+2 r = r fr + 2 requires C r f r + 4 C^{r_{\mathrm{fr}}+4} C r fr + 4 bounds for the transport coefficients. Proposition 2.2 shows that the elastic reduction involves derivatives of the Lamé parameters through order two; see (2.22). Consequently, Corollary 3.4 gives the sufficient material regularity
s = r f r + 6 = 2 m + 14 , s=r_{\mathrm{fr}}+6=2m+14, s = r fr + 6 = 2 m + 14 ,
as specified in (3.14). The smallest admissible choice m = 4 m=4 m = 4 therefore gives r f r = 16 r_{\mathrm{fr}}=16 r fr = 16 and s = 22 s=22 s = 22 .
This regularity threshold is sufficient for the present construction. It also suffices for global uniqueness, as stated in Corollary 1.4. The higher-regularity result is given in Proposition 10.1, with the admissible logarithmic exponents specified in (10.4).
Proof strategy
The argument first treats a fixed common-exterior class and then transfers the resulting estimate to the original domain by boundary recovery and paired extensions.
For the auxiliary problem, fix an integer m ≥ 4 m\geq4 m ≥ 4 and set s = 2 m + 14 s=2m+14 s = 2 m + 14 . Let D ⊂ R 3 D\subset\mathbb{R}^3 D ⊂ R 3 be a bounded smooth domain, and let ( λ b g , μ b g ) (\lambda_{\mathrm{bg}},\mu_{\mathrm{bg}}) ( λ bg , μ bg ) be a fixed, uniformly strongly convex C s C^s C s pair on R 3 \mathbb{R}^3 R 3 , constant outside a fixed ball. We consider global coefficient pairs with uniform C s C^s C s bounds satisfying
( λ j , μ j ) = ( λ b g , μ b g ) on R 3 ∖ D , j = 1 , 2. (15) (\lambda_j,\mu_j)=(\lambda_{\mathrm{bg}},\mu_{\mathrm{bg}})\quad\text{on }\mathbb{R}^3\setminus D,\qquad j=1,2. \tag*{(15)} ( λ j , μ j ) = ( λ bg , μ bg ) on R 3 ∖ D , j = 1 , 2. (15)
For a fixed measurement domain D D D , write
δ D = ∥ Λ 1 D − Λ 2 D ∥ H 1 / 2 ( ∂ D ; C 3 ) → H − 1 / 2 ( ∂ D ; C 3 ) , δ Ω = δ , δ B R e = Δ e . (16) \delta_D=\left\|\Lambda_1^D-\Lambda_2^D\right\|_{H^{1/2}(\partial D;\mathbb{C}^3)\to H^{-1/2}(\partial D;\mathbb{C}^3)},
\qquad
\delta_\Omega=\delta,\qquad\delta_{B_{R_e}}=\Delta_e. \tag*{(16)} δ D = Λ 1 D − Λ 2 D H 1/2 ( ∂ D ; C 3 ) → H − 1/2 ( ∂ D ; C 3 ) , δ Ω = δ , δ B R e = Δ e . (16)
Here B R e B_{R_e} B R e is the enlarged measurement domain introduced in the second part of the argument. The common-exterior geometry is fixed in Subsection 7.3. In particular, the coefficients agree in a neighborhood of the larger working sphere ∂ B R \partial B_R ∂ B R . Lemma 2.4 gives δ B R ≤ C δ D \delta_{B_R} \le C\delta_D δ B R ≤ C δ D . Thus the Betti identity for the elastic system (Lemma 2.3 ) may be applied on B R B_R B R with the boundary pairing bounded in terms of the discrepancy measured on ∂ D \partial D ∂ D .
The exact four-component reduction (Proposition 2.2), together with the finite-order transport frames and right inverse of Theorems 3.1 and 3.2 , provides the required CGO solutions with uniform remainder estimates. A central difficulty is that the leading physical adjoint lift has a two-dimensional image in the rank-three characteristic quotient. Proposition 4.1 supplies the missing third polarization through the first nonvanishing correction to an amplitude annihilated by the leading lift. Applying the Betti identity to these solutions relates the polynomially weighted matrix moments in the fixed primal and dual quotient frames to DN pairings of elastic solutions. Theorem 4.2 bounds all nine entries in terms of the DN discrepancy δ D \delta_D δ D and the CGO remainders. The dependence on the test degree and localization width is given explicitly by K n , ς K_{n,\varsigma} K n , ς in (4.33), while the constants and the threshold h ∗ h_* h ∗ are independent of these test parameters.
The moment estimates control the negative Fourier modes of the boundary traces of the frame products W a = Y 2 , a T X 1 , a W_a = Y_{2,a}^{T}X_{1,a} W a = Y 2 , a T X 1 , a ; see (5.4) and Lemma 5.1 . Lemma 5.2 shows that the corresponding Cauchy extensions R C , a R_{C,a} R C , a are pointwise invertible on the closed leaf disks. The primal and dual quotient frames then define G a = X 1 , a R C , a − 1 Y 2 , a T G_a = X_{1,a}R_{C,a}^{-1}Y_{2,a}^{T} G a = X 1 , a R C , a − 1 Y 2 , a T , which satisfies
2 ∂ z ˉ G a + P 1 G a − G a P 2 = 0 ; 2\partial_{\bar z}G_a + P_1G_a - G_aP_2 = 0; 2 ∂ z ˉ G a + P 1 G a − G a P 2 = 0 ;
see (5.27). Quantitative comparison on overlaps yields the global invertible quotient endomorphisms G G G and H = G − T H = G^{-T} H = G − T , with uniform regularity bounds and controlled deviation from the identity on ∂ B R \partial B_R ∂ B R ; see Proposition 5.4 . At each fixed physical point, the projection H 0 = Π H H_0 = \Pi H H 0 = Π H is holomorphic in the complex null direction. Proposition 6.3 controls the projection error. The first-order transport equation then yields the C x 2 C^2_x C x 2 residual estimate (6.24). The trace-free Hessian estimate and the coefficient-rigidity theorem (Theorem 7.3 ) convert this residual into estimates for both Lamé parameters.
More precisely, the estimates assembled in (7.70) give
ϑ D ≤ C ( h + δ D h − 2 e C 0 / h ) γ ′ β m / 14 , β m = m − 3 2 m , 0 < γ ′ < γ ∗ , \vartheta_D \le C\left(h+\delta_Dh^{-2}e^{C_0/h}\right)^{\gamma'\beta_m/14}, \qquad\beta_m = \frac{m-3}{2m}, \qquad0 < \gamma' < \gamma_*, ϑ D ≤ C ( h + δ D h − 2 e C 0 / h ) γ ′ β m /14 , β m = 2 m m − 3 , 0 < γ ′ < γ ∗ ,
whenever the auxiliary parameters satisfy the fixed smallness conditions. Here h > 0 h>0 h > 0 is the CGO parameter, ϑ D \vartheta_D ϑ D is the L ∞ L^\infty L ∞ coefficient discrepancy defined in (1.17), and γ ∗ \gamma_* γ ∗ is given by (5.8). Logarithmic optimization yields the common-exterior stability estimate in Proposition 7.5 .
To remove the common-exterior assumption, we first recover the boundary derivatives directly from the physical DN map. The uniform boundary energy expansion (Theorem 8.2 ) and the triangular recovery of successive normal derivatives yield, after tangential interpolation, the estimates
J q ≤ C δ ( s − q ) / s , 0 ≤ q < s , J_q \le C\delta^{(s-q)/s}, \qquad0 \le q < s, J q ≤ C δ ( s − q ) / s , 0 ≤ q < s ,
of Theorem 8.5 . The perturbation construction in Proposition 8.6 establishes the sharpness of these powers under the stated strict-slack assumptions.
The highest-order estimate needed for the extension argument is
J s − 1 ≤ C δ 1 / s . J_{s-1} \le C\delta^{1/s}. J s − 1 ≤ C δ 1/ s .
Theorem 9.2 uses this boundary control and an arbitrary positive tolerance ϵ t o l \epsilon_{\mathrm{tol}} ϵ tol to construct paired C s C^s C s extensions with fixed strong-convexity bounds and a common exterior background. The extensions agree with the original coefficients on Ω ‾ \overline{\Omega} Ω , but need not agree with each other in B R e ∖ Ω ‾ B_{R_e}\setminus\overline{\Omega} B R e ∖ Ω . Writing ϵ e x t \epsilon_{\mathrm{ext}} ϵ ext for their exterior L ∞ L^\infty L ∞ discrepancy, the extension theorem and Proposition 2.1 give
ϵ e x t ≤ C ( J s − 1 + ϵ t o l ) , \epsilon_{\mathrm{ext}} \le C(J_{s-1}+\epsilon_{\mathrm{tol}}), ϵ ext ≤ C ( J s − 1 + ϵ tol ) ,
Δ e ≤ C ( δ + ϵ e x t ) ≤ C ( δ + J s − 1 + ϵ t o l ) . \Delta_e \le C(\delta+\epsilon_{\mathrm{ext}}) \le C(\delta+J_{s-1}+\epsilon_{\mathrm{tol}}). Δ e ≤ C ( δ + ϵ ext ) ≤ C ( δ + J s − 1 + ϵ tol ) .
These are the estimates (353) and (366) , respectively. In particular, Δ e \Delta_e Δ e is controlled by the original measurements and is not an additional datum.
For 0 < δ 0 < \delta 0 < δ sufficiently small, choosing ϵ t o l = δ + J s − 1 \epsilon_{\mathrm{tol}} = \delta+ J_{s-1} ϵ tol = δ + J s − 1 gives
Δ e ≤ C δ 1 / s , \Delta_e \le C\delta^{1/s}, Δ e ≤ C δ 1/ s ,
as established in Corollary 9.3 . All extensions belong to one fixed enlarged coefficient class, whose bounds and geometry are independent of δ \delta δ and ϵ t o l \epsilon_{\mathrm{tol}} ϵ tol . The common-exterior stability estimate can therefore be applied with uniform constants. The power 1 / s 1/s 1/ s changes only the multiplicative constant in the logarithmic comparison, not the admissible logarithmic exponent.
Finally, when δ = 0 \delta= 0 δ = 0 , the boundary estimates give J s − 1 = 0 J_{s-1} = 0 J s − 1 = 0 . For every sufficiently small t > 0 t > 0 t > 0 , the extension construction remains valid with ϵ t o l = t \epsilon_{\mathrm{tol}} = t ϵ tol = t , and (373) gives
Δ e ( t ) ≤ C t . \Delta_e(t) \le Ct. Δ e ( t ) ≤ C t .
If Δ e ( t ) = 0 \Delta_e(t) = 0 Δ e ( t ) = 0 , the common-exterior exact-data result applies directly. Otherwise, its positive-error estimate provides a numerical upper bound for the unchanged original coefficient discrepancy that tends to zero as t ↓ 0 t \downarrow0 t ↓ 0 . This proves the exact-data conclusion without taking a limit of the extensions themselves. Proposition 10.1 completes both parts of the argument.
Notation and conventions
The tensor conventions are fixed in (3) . A superscript T T T denotes ordinary matrix transpose or formal transpose for the complex-bilinear pairing, without complex conjugation. The three-component physical displacement u u u is distinguished from the four-component reduced unknown z z z in (35) . The elastic traction uses the outward normal, whereas boundary-layer coordinates are directed inward.
For a normalized null direction θ = e 1 + i e 2 \theta= e_1 + i e_2 θ = e 1 + i e 2 , the complex leaf coordinate z z z and transverse coordinate t ⊥ t_\perp t ⊥ are defined in (54) . The affine spinor coordinates ζ \zeta ζ and υ \upsilon υ are defined in (57) . The Fourier projections P + P_+ P + and P − P_- P − act on the boundaries of the leaf disks, as in (140) . The fixed projection Π \Pi Π in (182) acts on endomorphism sections over the projective space of complex null directions at each fixed physical point x x x . Thus the projected field H 0 = Π H H_0 = \Pi H H 0 = Π H , defined in (186) , is holomorphic in the null direction, whereas the Cauchy extensions R C , a R_{C,a} R C , a are holomorphic in the leaf coordinate z z z .
The CGO parameter is h > 0 h > 0 h > 0 , with τ = h − 1 \tau= h^{-1} τ = h − 1 . We use ς \varsigma ς for the transverse localization width, ϵ \epsilon ϵ for the boundary-layer scale, and ϵ t o l \epsilon_{\mathrm{tol}} ϵ tol for the positive tolerance in the paired extension construction. Their respective roles are specified in (125) , (288) , and (351) . The integer m ≥ 4 m \ge4 m ≥ 4 indexes the regularity in the interior estimates.
For the common-exterior problem, set
η D = h + δ D h − 2 e C 0 / h , β m = m − 3 2 m , \eta_D = h + \delta_D h^{-2} e^{C_0/h}, \qquad\beta_m = \frac{m-3}{2m}, η D = h + δ D h − 2 e C 0 / h , β m = 2 m m − 3 ,
ϑ D = ∥ λ 1 − λ 2 ∥ L ∞ ( D ) + ∥ μ 1 − μ 2 ∥ L ∞ ( D ) . (17) \vartheta_D = \|\lambda_1-\lambda_2\|_{L^\infty(D)} + \|\mu_1-\mu_2\|_{L^\infty(D)}. \tag*{(17)} ϑ D = ∥ λ 1 − λ 2 ∥ L ∞ ( D ) + ∥ μ 1 − μ 2 ∥ L ∞ ( D ) . (17)
Here δ D \delta_D δ D is defined in (16) , and C 0 C_0 C 0 is the fixed constant from Theorem 4.2 . The Fourier error ω F \omega_F ω F and rigidity error ε r i g \varepsilon_{\mathrm{rig}} ε rig are specified in (177) and (236) , respectively. Unless otherwise indicated, C C C denotes a positive constant depending on the fixed geometry, a priori bounds, and explicitly fixed auxiliary parameters, and may change from line to line.
Physical identities and the common-exterior framework
This section fixes the physical variational pairing, compares the elastic DN maps defined by (5) on nested domains, and establishes the exact reduction. Under the common-collar hypothesis, the Betti identity expresses a reduced volume pairing in terms of displacement and traction data.
Physical variational identities and uniform energy bounds
For any fixed smooth domain D D D , let ε ( u ) \varepsilon(u) ε ( u ) be the strain in (1) . Define the bilinear and Hermitian energies by
a D ( u , v ) = ∫ D { λ div u div v + 2 μ ε ( u ) : ε ( v ) } d x , a D ( u , v ) = a D ( u , v ‾ ) . (18) a_D(u,v)=\int_D\{\lambda\operatorname{div}u\operatorname{div}v+2\mu\varepsilon(u):\varepsilon(v)\}\,\mathrm{d}x,\qquad\mathbf{a}_D(u,v)=a_D(u,\overline{v}). \tag*{(18)} a D ( u , v ) = ∫ D { λ div u div v + 2 μ ε ( u ) : ε ( v )} d x , a D ( u , v ) = a D ( u , v ) . (18)
The form a D a_D a D in (18) is complex bilinear, and a D \mathbf{a}_D a D is Hermitian. The elastic Betti identities use a D ( u , v ) a_D(u,v) a D ( u , v ) , whereas coercive energy estimates use a D ( u , u ) = a D ( u , u ‾ ) \mathbf{a}_D(u,u)=a_D(u,\overline{u}) a D ( u , u ) = a D ( u , u ) . Both forms act on H 1 ( D ; C 3 ) H^1(D;\mathbb{C}^3) H 1 ( D ; C 3 ) . The equation is the static elasticity equation div σ λ , μ ( u ) = 0 \operatorname{div}\sigma_{\lambda,\mu}(u)=0 div σ λ , μ ( u ) = 0 . For the solution u f u_f u f with trace f f f , the DN map defined as in (5) on D D D satisfies
⟨ Λ D f , g ⟩ = a D ( u f , v g ) , f , g ∈ H 1 / 2 ( ∂ D ; C 3 ) . (19) \langle\Lambda^D f,g\rangle=a_D(u_f,v_g),\qquad f,g\in H^{1/2}(\partial D;\mathbb{C}^3). \tag*{(19)} ⟨ Λ D f , g ⟩ = a D ( u f , v g ) , f , g ∈ H 1/2 ( ∂ D ; C 3 ) . (19)
Here ⟨ ⋅ , ⋅ ⟩ \langle\cdot,\cdot\rangle ⟨ ⋅ , ⋅ ⟩ denotes the complex-bilinear duality between H − 1 / 2 ( ∂ D ; C 3 ) H^{-1/2}(\partial D;\mathbb{C}^3) H − 1/2 ( ∂ D ; C 3 ) and H 1 / 2 ( ∂ D ; C 3 ) H^{1/2}(\partial D;\mathbb{C}^3) H 1/2 ( ∂ D ; C 3 ) , and v g v_g v g is any H 1 H^1 H 1 lifting of g g g . The weak equation makes it independent of the lifting and agrees with traction for the outward unit normal. The original-domain discrepancy remains δ \delta δ from (6) ; the domain-dependent discrepancy is δ D \delta_D δ D from (16) .
For a symmetric complex matrix E E E , define its trace-free part by
E 0 = E − 1 3 tr ( E ) I 3 . (20) E_0=E-\frac{1}{3}\operatorname{tr}(E)I_3. \tag*{(20)} E 0 = E − 3 1 tr ( E ) I 3 . (20)
The priors (4) give
λ ∣ tr E ∣ 2 + 2 μ ∣ E ∣ 2 = 2 μ ∣ E 0 ∣ 2 + 3 λ + 2 μ 3 ∣ tr E ∣ 2 ≥ c 0 ∣ E ∣ 2 . (21) \lambda|\operatorname{tr}E|^2+2\mu|E|^2=2\mu|E_0|^2+\frac{3\lambda+2\mu}{3}|\operatorname{tr}E|^2\ge c_0|E|^2. \tag*{(21)} λ ∣ tr E ∣ 2 + 2 μ ∣ E ∣ 2 = 2 μ ∣ E 0 ∣ 2 + 3 3 λ + 2 μ ∣ tr E ∣ 2 ≥ c 0 ∣ E ∣ 2 . (21)
Zero extension, Plancherel’s identity and Poincaré’s inequality yield
2 ∥ ε ( w ) ∥ L 2 ( D ) 2 = ∥ ∇ w ∥ L 2 ( D ) 2 + ∥ div w ∥ L 2 ( D ) 2 , 2\|\varepsilon(w)\|_{L^2(D)}^2=\|\nabla w\|_{L^2(D)}^2+\|\operatorname{div}w\|_{L^2(D)}^2, 2∥ ε ( w ) ∥ L 2 ( D ) 2 = ∥∇ w ∥ L 2 ( D ) 2 + ∥ div w ∥ L 2 ( D ) 2 ,
a D ( w , w ) ≥ c D ∥ w ∥ H 1 ( D ) 2 , w ∈ H 0 1 ( D ; C 3 ) . (22) \mathbf{a}_D(w,w)\ge c_D\|w\|_{H^1(D)}^2,\qquad w\in H_0^1(D;\mathbb{C}^3). \tag*{(22)} a D ( w , w ) ≥ c D ∥ w ∥ H 1 ( D ) 2 , w ∈ H 0 1 ( D ; C 3 ) . (22)
Applying (22) to the zero-trace difference between the solution and a bounded lifting of its trace, we obtain
∥ u f ∥ H 1 ( D ) ≤ C D ∥ f ∥ H 1 / 2 ( ∂ D ) . (23) \|u_f\|_{H^1(D)}\le C_D\|f\|_{H^{1/2}(\partial D)}. \tag*{(23)} ∥ u f ∥ H 1 ( D ) ≤ C D ∥ f ∥ H 1/2 ( ∂ D ) . (23)
The trace lifting can equivalently be defined using the quotient norm of the trace of H 1 ( D ) H^1(D) H 1 ( D ) ; on the fixed smooth boundary that norm is equivalent to the usual H 1 / 2 H^{1/2} H 1/2 norm. This observation fixes the norm used in the physical estimates of Sections 2, 8, and 9. On Ω \Omega Ω we abbreviate a Ω a_\Omega a Ω by
a = a Ω . (24) a=a_\Omega. \tag*{(24)} a = a Ω . (24)
A single material is understood when this abbreviation is used.
Comparison of DN discrepancies on nested domains
To transfer measurements to an enlarged domain, we compare the DN discrepancies on ∂ Ω \partial\Omega ∂ Ω and ∂ B R e \partial B_{R_e} ∂ B R e , defined in (16) , while retaining the exterior material mismatch. For the scalar conductivity equation, a formula relating the original and enlarged-domain DN maps is given in [21 Section 2] . Fix a ball B R e B_{R_e} B R e with Ω ‾ ⋐ B R e \overline{\Omega}\Subset B_{R_e} Ω ⋐ B R e . The comparison uses only bounded strongly convex material values and allows an exterior mismatch at the inner boundary.
Proposition 2.1. Let two material pairs on the fixed B R e B_{R_e} B R e be uniformly bounded and strongly convex, and let δ \delta δ be their DN discrepancy on Ω \Omega Ω . Put
δ e x t B = ∥ λ 1 − λ 2 ∥ L ∞ ( B R e ∖ Ω ‾ ) + ∥ μ 1 − μ 2 ∥ L ∞ ( B R e ∖ Ω ‾ ) . (25) \delta_{\mathrm{ext}}^B=\|\lambda_1-\lambda_2\|_{L^\infty(B_{R_e}\setminus\overline{\Omega})}+\|\mu_1-\mu_2\|_{L^\infty(B_{R_e}\setminus\overline{\Omega})}. \tag*{(25)} δ ext B = ∥ λ 1 − λ 2 ∥ L ∞ ( B R e ∖ Ω ) + ∥ μ 1 − μ 2 ∥ L ∞ ( B R e ∖ Ω ) . (25)
Then the enlarged-domain discrepancy satisfies
Δ e ≤ C ( δ + ϵ e x t B ) . (26) \Delta_e \le C(\delta+\epsilon_{\mathrm{ext}}^B). \tag*{(26)} Δ e ≤ C ( δ + ϵ ext B ) . (26)
Here δ \delta δ and Δ e \Delta_e Δ e are defined in (6) and (16) , respectively. The constant depends only on the two fixed domains, the upper material bounds and the strong-convexity lower bound.
Proof. Let u 1 , u 2 u_1,u_2 u 1 , u 2 solve the two problems on B R e B_{R_e} B R e with outer traces F , G F,G F , G , respectively. By the definition in (19) , symmetry of the bilinear energy and the weak equations,
⟨ ( Λ 1 B R e − Λ 2 B R e ) F , G ⟩ = ∫ B R e ( λ 1 − λ 2 ) div u 1 div u 2 d x + 2 ∫ B R e ( μ 1 − μ 2 ) ε ( u 1 ) : ε ( u 2 ) d x . (27) \begin{aligned}
\langle(\Lambda_1^{B_{R_e}}-\Lambda_2^{B_{R_e}})F,G\rangle
&=\int_{B_{R_e}}(\lambda_1-\lambda_2)\operatorname{div}u_1\operatorname{div}u_2\,dx\\
&\quad+2\int_{B_{R_e}}(\mu_1-\mu_2)\varepsilon(u_1):\varepsilon(u_2)\,dx. \tag*{(27)}
\end{aligned} ⟨( Λ 1 B R e − Λ 2 B R e ) F , G ⟩ = ∫ B R e ( λ 1 − λ 2 ) div u 1 div u 2 d x + 2 ∫ B R e ( μ 1 − μ 2 ) ε ( u 1 ) : ε ( u 2 ) d x . (27)
More explicitly, a 1 ( u 1 , u 2 ) = ⟨ Λ 1 F , G ⟩ a_1(u_1,u_2)=\langle\Lambda_1F,G\rangle a 1 ( u 1 , u 2 ) = ⟨ Λ 1 F , G ⟩ . In a 2 ( u 1 , u 2 ) a_2(u_1,u_2) a 2 ( u 1 , u 2 ) the first argument may be replaced by the material-two solution with trace F F F , because its difference from u 1 u_1 u 1 has zero trace and u 2 u_2 u 2 solves the material-two equation. This gives the second DN pairing and proves the identity without a boundary-matching assumption.
The restrictions u j ∣ Ω u_j|_{\Omega} u j ∣ Ω solve their respective equations. Define
f = Tr ∂ Ω u 1 , g = Tr ∂ Ω u 2 . (28) f=\operatorname{Tr}_{\partial\Omega}u_1,\qquad g=\operatorname{Tr}_{\partial\Omega}u_2. \tag*{(28)} f = Tr ∂ Ω u 1 , g = Tr ∂ Ω u 2 . (28)
Then the integral over Ω \Omega Ω in (27) is ⟨ ( Λ 1 Ω − Λ 2 Ω ) f , g ⟩ \langle(\Lambda_1^\Omega-\Lambda_2^\Omega)f,g\rangle ⟨( Λ 1 Ω − Λ 2 Ω ) f , g ⟩ . We therefore have the exact split identity
⟨ ( Λ 1 B R e − Λ 2 B R e ) F , G ⟩ = ⟨ ( Λ 1 Ω − Λ 2 Ω ) f , g ⟩ + ∫ B R e ∖ Ω ‾ ( λ 1 − λ 2 ) div u 1 div u 2 d x + 2 ∫ B R e ∖ Ω ‾ ( μ 1 − μ 2 ) ε ( u 1 ) : ε ( u 2 ) d x . (29) \begin{aligned}
\langle(\Lambda_1^{B_{R_e}}-\Lambda_2^{B_{R_e}})F,G\rangle
&=\langle(\Lambda_1^\Omega-\Lambda_2^\Omega)f,g\rangle\\
&\quad+\int_{B_{R_e}\setminus\overline{\Omega}}(\lambda_1-\lambda_2)\operatorname{div}u_1\operatorname{div}u_2\,dx\\
&\quad+2\int_{B_{R_e}\setminus\overline{\Omega}}(\mu_1-\mu_2)\varepsilon(u_1):\varepsilon(u_2)\,dx. \tag*{(29)}
\end{aligned} ⟨( Λ 1 B R e − Λ 2 B R e ) F , G ⟩ = ⟨( Λ 1 Ω − Λ 2 Ω ) f , g ⟩ + ∫ B R e ∖ Ω ( λ 1 − λ 2 ) div u 1 div u 2 d x + 2 ∫ B R e ∖ Ω ( μ 1 − μ 2 ) ε ( u 1 ) : ε ( u 2 ) d x . (29)
The uniform Dirichlet estimate (23) and the trace bound on the fixed inner domain give ∥ u 1 ∥ H 1 ( B R e ) + ∥ f ∥ H 1 / 2 ( ∂ Ω ) ≤ C ∥ F ∥ H 1 / 2 ( ∂ B R e ) \lVert u_1\rVert_{H^1(B_{R_e})}+\lVert f\rVert_{H^{1/2}(\partial\Omega)}\le C\lVert F\rVert_{H^{1/2}(\partial B_{R_e})} ∥ u 1 ∥ H 1 ( B R e ) + ∥ f ∥ H 1/2 ( ∂ Ω ) ≤ C ∥ F ∥ H 1/2 ( ∂ B R e ) , and the analogous estimate for u 2 , g , G u_2,g,G u 2 , g , G . Cauchy–Schwarz bounds the two exterior terms by C ϵ e x t B ∥ F ∥ H 1 / 2 ∥ G ∥ H 1 / 2 C\epsilon_{\mathrm{ext}}^B\lVert F\rVert_{H^{1/2}}\lVert G\rVert_{H^{1/2}} C ϵ ext B ∥ F ∥ H 1/2 ∥ G ∥ H 1/2 . The inner term is bounded by the same expression with δ \delta δ . All pairings in this proof are bilinear; Hermitian energy was used only for the solution norm estimates. Taking the supremum over F , G ∈ H 1 / 2 ( ∂ B R e ; C 3 ) F,G\in H^{1/2}(\partial B_{R_e};\mathbb{C}^3) F , G ∈ H 1/2 ( ∂ B R e ; C 3 ) with ∥ F ∥ H 1 / 2 = ∥ G ∥ H 1 / 2 = 1 \lVert F\rVert_{H^{1/2}}=\lVert G\rVert_{H^{1/2}}=1 ∥ F ∥ H 1/2 = ∥ G ∥ H 1/2 = 1 , we obtain (26) by the definition of Δ e \Delta_e Δ e in (16) . This completes the proof.
Remark. This comparison separates two sources of enlarged-domain error: the measured discrepancy inside Ω \Omega Ω and the material mismatch outside it. Thus the paired extension needs to make only the exterior coefficient values close; the comparison itself does not require their boundary derivatives to agree.
The identity (27) , applied on a single fixed domain D D D , gives the forward bound used by the sharpness examples:
∥ Λ 1 D − Λ 2 D ∥ H 1 / 2 → H − 1 / 2 ≤ C D ( ∥ λ 1 − λ 2 ∥ ∞ + ∥ μ 1 − μ 2 ∥ ∞ ) . (30) \lVert\Lambda_1^D-\Lambda_2^D\rVert_{H^{1/2}\to H^{-1/2}}
\le C_D\left(\lVert\lambda_1-\lambda_2\rVert_\infty+\lVert\mu_1-\mu_2\rVert_\infty\right). \tag*{(30)} ∥ Λ 1 D − Λ 2 D ∥ H 1/2 → H − 1/2 ≤ C D ( ∥ λ 1 − λ 2 ∥ ∞ + ∥ μ 1 − μ 2 ∥ ∞ ) . (30)
Indeed, (23) bounds each solution, and Cauchy–Schwarz bounds the two terms in (27) .
Conventions, coefficient bounds, and operator types
The four-component reduction has Laplacian principal part. A related reduction obtained by adjoining the divergence of the displacement appears in [5 ] . The reduced unknown is defined in (35) ; the material variables and operator mapping properties are specified below before the factorization is stated.
For an open set O ⊂ R 3 O \subset\mathbb{R}^{3} O ⊂ R 3 , use the physical operator, stress and strain from (1) , with u : O → C 3 u: O \to\mathbb{C}^{3} u : O → C 3 and the component conventions (3) . Superscript T T T denotes the ordinary matrix or formal differential transpose for the complex-bilinear pairing. The Hermitian energy is a D a_{\mathcal{D}} a D from (18) .
For an integer r ≥ 2 r \ge2 r ≥ 2 , assume real coefficients with ∥ λ ∥ C r ( O ) + ∥ μ ∥ C r ( O ) ≤ M \|\lambda\|_{C^{r}(O)}+\|\mu\|_{C^{r}(O)}\le M ∥ λ ∥ C r ( O ) + ∥ μ ∥ C r ( O ) ≤ M , M ≥ 1 M\ge1 M ≥ 1 , and μ ≥ c 0 > 0 \mu\ge c_{0}>0 μ ≥ c 0 > 0 , 3 λ + 2 μ ≥ c 0 3\lambda+2\mu\ge c_{0} 3 λ + 2 μ ≥ c 0 . Define
κ = λ + μ λ + 2 μ , c = μ − 1 / 2 , ℓ μ = log μ , g μ = ∇ ℓ μ , d = κ μ , e = 1 2 ( 3 κ − 2 ) μ . (31) \kappa=\frac{\lambda+\mu}{\lambda+2\mu},\qquad c=\mu^{-1/2},\qquad\ell_{\mu}=\log\mu,\qquad g_{\mu}=\nabla\ell_{\mu},\qquad d=\kappa\sqrt{\mu},\qquad e=\frac{1}{2}(3\kappa-2)\sqrt{\mu}. \tag*{(31)} κ = λ + 2 μ λ + μ , c = μ − 1/2 , ℓ μ = log μ , g μ = ∇ ℓ μ , d = κ μ , e = 2 1 ( 3 κ − 2 ) μ . (31)
All these are scalar fields except g μ : O → R 3 g_{\mu}: O\to\mathbb{R}^{3} g μ : O → R 3 . The following bounds explicitly control the denominators:
5 c 0 3 ≤ λ + 2 μ ≤ 3 M , 1 − κ ≥ c 0 3 M , κ − 1 4 ≥ c 0 12 M , \frac{5c_{0}}{3}\le\lambda+2\mu\le3M,\qquad1-\kappa\ge\frac{c_{0}}{3M},\qquad\kappa-\frac{1}{4}\ge\frac{c_{0}}{12M}, 3 5 c 0 ≤ λ + 2 μ ≤ 3 M , 1 − κ ≥ 3 M c 0 , κ − 4 1 ≥ 12 M c 0 ,
M − 1 / 2 ≤ c ≤ c 0 − 1 / 2 , λ = μ 2 κ − 1 1 − κ . (32) M^{-1/2}\le c\le c_{0}^{-1/2},\qquad\lambda=\mu\frac{2\kappa-1}{1-\kappa}. \tag*{(32)} M − 1/2 ≤ c ≤ c 0 − 1/2 , λ = μ 1 − κ 2 κ − 1 . (32)
Indeed, λ + 2 μ = ( ( 3 λ + 2 μ ) + 4 μ ) / 3 \lambda+2\mu=((3\lambda+2\mu)+4\mu)/3 λ + 2 μ = (( 3 λ + 2 μ ) + 4 μ ) /3 , 1 − κ = μ / ( λ + 2 μ ) 1-\kappa=\mu/(\lambda+2\mu) 1 − κ = μ / ( λ + 2 μ ) , and 4 κ − 1 = ( 3 λ + 2 μ ) / ( λ + 2 μ ) 4\kappa-1=(3\lambda+2\mu)/(\lambda+2\mu) 4 κ − 1 = ( 3 λ + 2 μ ) / ( λ + 2 μ ) .
Set
F μ = 2 c [ ∇ 2 ℓ μ − g μ ⊗ g μ + ( ∣ g μ ∣ 2 − Δ ℓ μ ) I 3 ] , F_{\mu}=2c[\nabla^{2}\ell_{\mu}-g_{\mu}\otimes g_{\mu}+(|g_{\mu}|^{2}-\Delta\ell_{\mu})I_{3}], F μ = 2 c [ ∇ 2 ℓ μ − g μ ⊗ g μ + ( ∣ g μ ∣ 2 − Δ ℓ μ ) I 3 ] ,
V μ = − ∇ 2 ℓ μ − 1 2 g μ ⊗ g μ − ( 1 2 Δ ℓ μ + 1 4 ∣ g μ ∣ 2 ) I 3 , V_{\mu}=-\nabla^{2}\ell_{\mu}-\frac{1}{2}g_{\mu}\otimes g_{\mu}-\left(\frac{1}{2}\Delta\ell_{\mu}+\frac{1}{4}|g_{\mu}|^{2}\right)I_{3}, V μ = − ∇ 2 ℓ μ − 2 1 g μ ⊗ g μ − ( 2 1 Δ ℓ μ + 4 1 ∣ g μ ∣ 2 ) I 3 ,
q μ = Δ ℓ μ − ∣ g μ ∣ 2 . (33) q_{\mu}=\Delta\ell_{\mu}-|g_{\mu}|^{2}. \tag*{(33)} q μ = Δ ℓ μ − ∣ g μ ∣ 2 . (33)
F μ F_{\mu} F μ , V μ V_{\mu} V μ are symmetric 3 × 3 3\times3 3 × 3 matrix fields, and q μ q_{\mu} q μ is scalar. We use the physical transformation in [18 Section 3, p. 6] in the following normalization:
T μ ( v , f ) = c v + μ − 1 ∇ f − f ∇ ( μ − 1 ) , T_{\mu}(v,f)=cv+\mu^{-1}\nabla f-f\nabla(\mu^{-1}), T μ ( v , f ) = c v + μ − 1 ∇ f − f ∇ ( μ − 1 ) ,
S μ , κ ( p , q ) = μ p + ∇ ( q 1 − κ ) , S_{\mu,\kappa}(p,q)=\sqrt{\mu}p+\nabla\left(\frac{q}{1-\kappa}\right), S μ , κ ( p , q ) = μ p + ∇ ( 1 − κ q ) ,
M μ , κ ( v f ) = ( Δ v − g μ div v + F μ ∇ f + V μ v + F μ g μ f Δ f + d div v − e g μ ⋅ v + q μ f ) . (34) \mathcal{M}_{\mu,\kappa}\begin{pmatrix}v\\f\end{pmatrix}
=
\begin{pmatrix}
\Delta v-g_{\mu}\operatorname{div}v+F_{\mu}\nabla f+V_{\mu}v+F_{\mu}g_{\mu}f\\
\Delta f+d\operatorname{div}v-e g_{\mu}\cdot v+q_{\mu}f
\end{pmatrix}. \tag*{(34)} M μ , κ ( v f ) = ( Δ v − g μ div v + F μ ∇ f + V μ v + F μ g μ f Δ f + d div v − e g μ ⋅ v + q μ f ) . (34)
Here v : O → C 3 v: O\to\mathbb{C}^{3} v : O → C 3 , f : O → C f: O\to\mathbb{C} f : O → C , and
z = ( v f ) : O → C 4 , T = T μ , S = S μ , κ , M = M μ , κ , L = L λ , μ . (35) z=\begin{pmatrix}v\\f\end{pmatrix}:O\to\mathbb{C}^{4},\qquad T=T_{\mu},\qquad S=S_{\mu,\kappa},\qquad\mathcal{M}=\mathcal{M}_{\mu,\kappa},\qquad L=L_{\lambda,\mu}. \tag*{(35)} z = ( v f ) : O → C 4 , T = T μ , S = S μ , κ , M = M μ , κ , L = L λ , μ . (35)
The operators T T T and S S S in (34) are 3 × 4 3\times4 3 × 4 first-order differential operators. On any fixed bounded smooth O ′ ⋐ O O'\Subset O O ′ ⋐ O , the coefficient bounds (32) and the product rule give bounded maps
T , S : H 2 ( O ′ ; C 4 ) ⟶ H 1 ( O ′ ; C 3 ) , T,S:H^{2}(O';\mathbb{C}^{4})\longrightarrow H^{1}(O';\mathbb{C}^{3}), T , S : H 2 ( O ′ ; C 4 ) ⟶ H 1 ( O ′ ; C 3 ) ,
M : H 2 ( O ′ ; C 4 ) ⟶ L 2 ( O ′ ; C 4 ) , S : L 2 ( O ′ ; C 4 ) ⟶ H − 1 ( O ′ ; C 3 ) . (36) \mathcal{M}:H^{2}(O';\mathbb{C}^{4})\longrightarrow L^{2}(O';\mathbb{C}^{4}),\qquad S:L^{2}(O';\mathbb{C}^{4})\longrightarrow H^{-1}(O';\mathbb{C}^{3}). \tag*{(36)} M : H 2 ( O ′ ; C 4 ) ⟶ L 2 ( O ′ ; C 4 ) , S : L 2 ( O ′ ; C 4 ) ⟶ H − 1 ( O ′ ; C 3 ) . (36)
Let e 1 , e 2 , e 3 e_{1},e_{2},e_{3} e 1 , e 2 , e 3 denote the coordinate vectors in R 3 \mathbb{R}^{3} R 3 . For h > 0 h>0 h > 0 and a nonzero constant null vector θ ∈ C 3 \theta\in\mathbb{C}^{3} θ ∈ C 3 , set
θ ⋅ θ = 0 , φ = θ ⋅ x , D θ = θ ⋅ ∇ x , e θ = ( θ , 0 ) T , σ p h ∈ { + 1 , − 1 } . (37) \theta\cdot\theta=0,\qquad\varphi=\theta\cdot x,\qquad D_{\theta}=\theta\cdot\nabla_{x},\qquad e_{\theta}=(\theta,0)^{T},\qquad\sigma_{\mathrm{ph}}\in\{+1,-1\}. \tag*{(37)} θ ⋅ θ = 0 , φ = θ ⋅ x , D θ = θ ⋅ ∇ x , e θ = ( θ , 0 ) T , σ ph ∈ { + 1 , − 1 } . (37)
We use the exponential e σ p h φ / h e^{\sigma_{\mathrm{ph}}\varphi/h} e σ ph φ / h , with no additional factor of i i i .
Proposition 2.2. Under the coefficient assumptions of this subsection, let L , T , S , M L,T,S,M L , T , S , M be defined by (34) and (35) , and fix a bounded smooth O ′ ⋐ O O' \Subset O O ′ ⋐ O . The following identities hold in the sense of distributions.
(i) The elastic operator admits the factorization
L T = S M , M = Δ I 4 + ∑ k = 1 3 A k ∂ k + B , LT=SM,\qquad M=\Delta I_4+\sum_{k=1}^{3} A_k\partial_k+B, L T = S M , M = Δ I 4 + k = 1 ∑ 3 A k ∂ k + B , A k = ( − g μ e k T F μ e k d e k T 0 ) , B = ( V μ F μ g μ − e g μ T q μ ) . (38) A_k=
\begin{pmatrix}
-g\mu e_k^T & F_\mu e_k\\
d e_k^T & 0
\end{pmatrix},
\qquad
B=
\begin{pmatrix}
V_\mu& F_\mu g_\mu\\
-e g_\mu^T & q_\mu
\end{pmatrix}. \tag*{(38)} A k = ( − g μ e k T d e k T F μ e k 0 ) , B = ( V μ − e g μ T F μ g μ q μ ) . (38) For C 2 C^2 C 2 Lamé parameters, this factorization holds in H l o c − 1 ( O ; C 3 ) H_{\mathrm{loc}}^{-1}(O;\mathbb{C}^3) H loc − 1 ( O ; C 3 ) for every z ∈ H l o c 2 ( O ; C 4 ) z\in H_{\mathrm{loc}}^2(O;\mathbb{C}^4) z ∈ H loc 2 ( O ; C 4 ) . For every integer r ≥ 2 r\ge2 r ≥ 2 under the stated C r C^r C r bounds,
∑ k = 1 3 ∥ A k ∥ C r − 2 ( O ′ ) + ∥ B ∥ C r − 2 ( O ′ ) ≤ C ( r , M , c 0 , O ′ ) . (39) \sum_{k=1}^{3}\|A_k\|_{C^{r-2}(O')}+\|B\|_{C^{r-2}(O')}\le C(r,M,c_0,O'). \tag*{(39)} k = 1 ∑ 3 ∥ A k ∥ C r − 2 ( O ′ ) + ∥ B ∥ C r − 2 ( O ′ ) ≤ C ( r , M , c 0 , O ′ ) . (39) (ii) The formal transpose S T S^T S T and the composite operator
K = S T T : H 2 ( O ′ ; C 4 ) ⟶ L 2 ( O ′ ; C 4 ) (40) K=S^T T:H^2(O';\mathbb{C}^4)\longrightarrow L^2(O';\mathbb{C}^4) \tag*{(40)} K = S T T : H 2 ( O ′ ; C 4 ) ⟶ L 2 ( O ′ ; C 4 ) (40) are given by
S T u = ( μ u − div u / ( 1 − κ ) ) , K ( v , f ) = ( v + c ∇ f + c f g μ − [ c div v − 1 2 c g μ ⋅ v + μ − 1 Δ f + μ − 1 q μ f ] / ( 1 − κ ) ) . (41) S^T u=
\begin{pmatrix}
\sqrt{\mu}\,u\\
-\operatorname{div}u/(1-\kappa)
\end{pmatrix},
\qquad
K(v,f)=
\begin{pmatrix}
v+c\nabla f+cf g_\mu\\
-\left[c\operatorname{div}v-\frac{1}{2}c g_\mu\cdot v+\mu^{-1}\Delta f+\mu^{-1}q_\mu f\right]/(1-\kappa)
\end{pmatrix}. \tag*{(41)} S T u = ( μ u − div u / ( 1 − κ ) ) , K ( v , f ) = ( v + c ∇ f + c f g μ − [ c div v − 2 1 c g μ ⋅ v + μ − 1 Δ f + μ − 1 q μ f ] / ( 1 − κ ) ) . (41) They satisfy the formal transpose identity
M T S T = T T L . (42) M^T S^T=T^T L. \tag*{(42)} M T S T = T T L . (42) (iii) With the phase conventions in (37) , one has
e − σ p h ϕ / h M e σ p h ϕ / h = M + σ p h h − 1 ( 2 D θ + A θ ) , A θ = ∑ k = 1 3 θ k A k , e^{-\sigma_{ph}\phi/h}Me^{\sigma_{ph}\phi/h}
=M+\sigma_{ph}h^{-1}(2D_\theta+A_\theta),
\qquad
A_\theta=\sum_{k=1}^{3}\theta_kA_k, e − σ p h ϕ / h M e σ p h ϕ / h = M + σ p h h − 1 ( 2 D θ + A θ ) , A θ = k = 1 ∑ 3 θ k A k , e − σ p h ϕ / h K e σ p h ϕ / h = K + σ p h h − 1 J θ , J θ ( v , f ) = ( c f θ − ( c θ ⋅ v + 2 μ − 1 D θ f ) / ( 1 − κ ) ) . (43) e^{-\sigma_{ph}\phi/h}Ke^{\sigma_{ph}\phi/h}
=K+\sigma_{ph}h^{-1}J_\theta,
\qquad
J_\theta(v,f)=
\begin{pmatrix}
c f\theta\\
-(c\theta\cdot v+2\mu^{-1}D_\theta f)/(1-\kappa)
\end{pmatrix}. \tag*{(43)} e − σ p h ϕ / h K e σ p h ϕ / h = K + σ p h h − 1 J θ , J θ ( v , f ) = ( c f θ − ( c θ ⋅ v + 2 μ − 1 D θ f ) / ( 1 − κ ) ) . (43) The first-order operators S T S^T S T and J θ J_\theta J θ define bounded maps
S T : H 1 ( O ′ ; C 3 ) ⟶ L 2 ( O ′ ; C 4 ) , J θ : H 1 ( O ′ ; C 4 ) ⟶ L 2 ( O ′ ; C 4 ) . (44) S^T:H^1(O';\mathbb{C}^3)\longrightarrow L^2(O';\mathbb{C}^4),
\qquad
J_\theta:H^1(O';\mathbb{C}^4)\longrightarrow L^2(O';\mathbb{C}^4). \tag*{(44)} S T : H 1 ( O ′ ; C 3 ) ⟶ L 2 ( O ′ ; C 4 ) , J θ : H 1 ( O ′ ; C 4 ) ⟶ L 2 ( O ′ ; C 4 ) . (44)
Proof. The scalar and vector identities (A.5) and (A.11) give L T = S M LT=SM L T = S M with the coefficients in (38) . For compactly supported smooth ( p , q ) (p,q) ( p , q ) and u u u ,
∫ O S ( p , q ) ⋅ u d x = ∫ O ( μ p ⋅ u − q div u 1 − κ ) d x = ∫ O ( p , q ) ⋅ S T u d x . (45) \begin{aligned}
\int_O S(p,q)\cdot u\,dx
=\int_O\left(\sqrt{\mu}\,p\cdot u-q\frac{\operatorname{div}u}{1-\kappa}\right)\,dx \\
&=\int_O(p,q)\cdot S^T u\,dx. \tag*{(45)}
\end{aligned} ∫ O S ( p , q ) ⋅ u d x = ∫ O ( μ p ⋅ u − q 1 − κ div u ) d x = ∫ O ( p , q ) ⋅ S T u d x . (45)
This proves the formula for S T S^{T} S T in (41) . Substitution of T T T from (34) gives the formula for K \mathcal{K} K . Transposing L T = S M LT=SM L T = S M and using L T = L L^{T}=L L T = L yields (42) .
The substitution
e − σ p h ϕ / h ∂ k e σ p h ϕ / h = ∂ k + σ p h h − 1 θ k , θ ⋅ θ = 0 , (46) e^{-\sigma_{\mathrm{ph}}\phi/h}\partial_k e^{\sigma_{\mathrm{ph}}\phi/h}=\partial_k+\sigma_{\mathrm{ph}}h^{-1}\theta_k,\qquad\theta\cdot\theta=0, \tag*{(46)} e − σ ph ϕ / h ∂ k e σ ph ϕ / h = ∂ k + σ ph h − 1 θ k , θ ⋅ θ = 0 , (46)
applied to (38) and (41) gives (43) . The product rule gives the boundedness in (44) . The coefficient formulas (33) involve at most two derivatives of the Lamé parameters. The denominator bounds (32) and Leibniz’ rule therefore give (39) . For C 2 C^{2} C 2 Lamé parameters and z ∈ H l o c 2 ( O ; C 4 ) z\in H^{2}_{\mathrm{loc}}(O;\mathbb{C}^{4}) z ∈ H loc 2 ( O ; C 4 ) , (36) and the boundedness of L : H 1 ( O ′ ; C 3 ) → H − 1 ( O ′ ; C 3 ) L:H^{1}(O';\mathbb{C}^{3})\to H^{-1}(O';\mathbb{C}^{3}) L : H 1 ( O ′ ; C 3 ) → H − 1 ( O ′ ; C 3 ) imply
T z ∈ H l o c 1 , M z ∈ L l o c 2 , Tz\in H^{1}_{\mathrm{loc}},\qquad Mz\in L^{2}_{\mathrm{loc}}, T z ∈ H loc 1 , M z ∈ L loc 2 ,
L T z − S M z ∈ H l o c − 1 . (47) LTz-SMz\in H^{-1}_{\mathrm{loc}}. \tag*{(47)} L T z − S M z ∈ H loc − 1 . (47)
Approximation of the Lamé parameters in C 2 C^{2} C 2 and of z z z in H l o c 2 H^{2}_{\mathrm{loc}} H loc 2 extends the identities to the stated distributional setting.
Remark. The factorization maps a reduced solution z ∈ H l o c 2 ( O ; C 4 ) z\in H^{2}_{\mathrm{loc}}(O;\mathbb{C}^{4}) z ∈ H loc 2 ( O ; C 4 ) to an elastic displacement u = T z u=Tz u = T z satisfying L u = 0 Lu=0 Lu = 0 . For a weak elastic solution u ∈ H l o c 1 ( O ; C 3 ) u\in H^{1}_{\mathrm{loc}}(O;\mathbb{C}^{3}) u ∈ H loc 1 ( O ; C 3 ) , the formal transpose identity gives the lift S T u ∈ L l o c 2 ( O ; C 4 ) S^{T}u\in L^{2}_{\mathrm{loc}}(O;\mathbb{C}^{4}) S T u ∈ L loc 2 ( O ; C 4 ) satisfying M T ( S T u ) = 0 M^{T}(S^{T}u)=0 M T ( S T u ) = 0 in D ′ ( O ; C 4 ) D'(O;\mathbb{C}^{4}) D ′ ( O ; C 4 ) . These identities relate solutions of the two systems; they do not, by themselves, identify the unrestricted boundary map of the reduced system with the measured elastic Dirichlet-to-Neumann map.
A common background and the paired quotient coordinates
The constant-background first-order term is removed before constructing paired coordinates on the characteristic quotient and its dual.
Suppose the materials equal a fixed constant pair outside a fixed ball, and use the material variables in (31) . The common background multiplier and transformed operators are
d ∗ = κ ∗ μ ∗ , R ∗ ( x ) = ( I 3 0 − d ∗ x T / 2 1 ) , M ~ = R ∗ − 1 M R ∗ , T ~ = T R ∗ , S ~ = S R ∗ . (48) d_{*}=\kappa_{*}\sqrt{\mu_{*}},\qquad
R_{*}(x)=
\begin{pmatrix}
I_{3} & 0\\
-d_{*}x^{T}/2 & 1
\end{pmatrix},\qquad
\widetilde{M}=R_{*}^{-1}MR_{*},\qquad
\widetilde{T}=TR_{*},\qquad
\widetilde{S}=SR_{*}. \tag*{(48)} d ∗ = κ ∗ μ ∗ , R ∗ ( x ) = ( I 3 − d ∗ x T /2 0 1 ) , M = R ∗ − 1 M R ∗ , T = T R ∗ , S = S R ∗ . (48)
R ∗ R_{*} R ∗ is a 4 × 4 4\times4 4 × 4 multiplication operator. Since Δ ( x ⋅ v ) = x ⋅ Δ v + 2 div v \Delta(x\cdot v)=x\cdot\Delta v+2\operatorname{div}v Δ ( x ⋅ v ) = x ⋅ Δ v + 2 div v , the constant background satisfies M ∗ R ∗ = R ∗ Δ M_{*}R_{*}=R_{*}\Delta M ∗ R ∗ = R ∗ Δ . The coefficients of M ~ − Δ I 4 \widetilde{M}-\Delta I_{4} M − Δ I 4 thus have fixed compact support, and their formulas involve at most two derivatives of the Lamé parameters. Explicitly,
A ~ k = R ∗ − 1 ( 2 ∂ k R ∗ + A k R ∗ ) , \widetilde{A}_{k}=R_{*}^{-1}(2\partial_kR_{*}+A_kR_{*}), A k = R ∗ − 1 ( 2 ∂ k R ∗ + A k R ∗ ) ,
B ~ = R ∗ − 1 ( Δ R ∗ + ∑ k A k ∂ k R ∗ + B R ∗ ) , (49) \widetilde{B}=R_{*}^{-1}\left(\Delta R_{*}+\sum_k A_k\partial_kR_{*}+BR_{*}\right), \tag*{(49)} B = R ∗ − 1 ( Δ R ∗ + k ∑ A k ∂ k R ∗ + B R ∗ ) , (49)
S ~ T = R ∗ T S T , K ~ = R ∗ T K R ∗ , J ~ θ = R ∗ T J θ R ∗ . \widetilde{S}^{T}=R_{*}^{T}S^{T},\qquad
\widetilde{\mathcal{K}}=R_{*}^{T}\mathcal{K}R_{*},\qquad
\widetilde{J}_{\theta}=R_{*}^{T}J_{\theta}R_{*}. S T = R ∗ T S T , K = R ∗ T K R ∗ , J θ = R ∗ T J θ R ∗ .
For a material pair, define
A Δ , θ = A 1 , θ − A 2 , θ , A ~ Δ , θ = A ~ 1 , θ − A ~ 2 , θ , (50) A_{\Delta,\theta}=A_{1,\theta}-A_{2,\theta},\qquad
\widetilde{A}_{\Delta,\theta}=\widetilde{A}_{1,\theta}-\widetilde{A}_{2,\theta}, \tag*{(50)} A Δ , θ = A 1 , θ − A 2 , θ , A Δ , θ = A 1 , θ − A 2 , θ , (50)
where A j , θ A_{j,\theta} A j , θ is the contraction in (43) . These are differential compositions; in the last identity J θ J_{\theta} J θ still differentiates R ∗ R_{*} R ∗ . For the transformed primal and dual fields
z ~ = R ∗ − 1 z , y ~ = R ∗ T y , (51) \widetilde{z}=R_{*}^{-1}z,\qquad\widetilde{y}=R_{*}^{T}y, \tag*{(51)} z = R ∗ − 1 z , y = R ∗ T y , (51)
with R ∗ R_{*} R ∗ from (48) , one has
y ~ T ( M ~ 1 − M ~ 2 ) z ~ = y T ( M 1 − M 2 ) z , A ~ Δ , θ = R ∗ − 1 A Δ , θ R ∗ . (52) \widetilde{y}^{T}(\widetilde{M}_{1}-\widetilde{M}_{2})\widetilde{z}
=y^{T}(M_{1}-M_{2})z,\qquad
\widetilde{A}_{\Delta,\theta}=R_{*}^{-1}A_{\Delta,\theta}R_{*}. \tag*{(52)} y T ( M 1 − M 2 ) z = y T ( M 1 − M 2 ) z , A Δ , θ = R ∗ − 1 A Δ , θ R ∗ . (52)
The first equality includes all derivatives of R ∗ R_* R ∗ . The leading transport satisfies the exact operator identity
2 D θ + A ~ θ = R ∗ − 1 ( 2 D θ + A θ ) R ∗ , e ~ θ = R ∗ − 1 e θ = ( θ d ∗ x ⋅ θ / 2 ) . (53) 2D_\theta+\widetilde{A}_\theta=R_*^{-1}(2D_\theta+A_\theta)R_*,\qquad\widetilde{e}_\theta=R_*^{-1}e_\theta=\begin{pmatrix}\theta\\ d_*x\cdot\theta/2\end{pmatrix}. \tag*{(53)} 2 D θ + A θ = R ∗ − 1 ( 2 D θ + A θ ) R ∗ , e θ = R ∗ − 1 e θ = ( θ d ∗ x ⋅ θ /2 ) . (53)
Indeed the first coefficient formula contributes 2 R ∗ − 1 ( D θ R ∗ ) 2R_*^{-1}(D_\theta R_*) 2 R ∗ − 1 ( D θ R ∗ ) as well as R ∗ − 1 A θ R ∗ R_*^{-1}A_\theta R_* R ∗ − 1 A θ R ∗ . Thus the transformed characteristic line is preserved. If the paired lifts are U ~ = R ∗ − 1 U \widetilde{U}=R_*^{-1}U U = R ∗ − 1 U and V ~ = R ∗ − 1 V \widetilde{V}=R_*^{-1}V V = R ∗ − 1 V , their pairing remains the identity. The quotient connection is unchanged when the spatial derivatives of the transported lifts U ~ \widetilde{U} U and V ~ \widetilde{V} V are included, by (2.36). In particular Y ~ 2 T A ~ Δ , θ Z ~ 1 = Y 2 T A Δ , θ Z 1 \widetilde{Y}_2^T\widetilde{A}_{\Delta,\theta}\widetilde{Z}_1=Y_2^TA_{\Delta,\theta}Z_1 Y 2 T A Δ , θ Z 1 = Y 2 T A Δ , θ Z 1 .
For a normalized null vector θ = e 1 + i e 2 \theta=e_1+ie_2 θ = e 1 + i e 2 , where e 1 , e 2 e_1,e_2 e 1 , e 2 are real orthonormal vectors, let
ν l e a f = e 1 × e 2 , z = e 1 ⋅ x + i e 2 ⋅ x , t ⊥ = ν l e a f ⋅ x , \nu_{\mathrm{leaf}}=e_1\times e_2,\qquad z=e_1\cdot x+ie_2\cdot x,\qquad t_\perp=\nu_{\mathrm{leaf}}\cdot x, ν leaf = e 1 × e 2 , z = e 1 ⋅ x + i e 2 ⋅ x , t ⊥ = ν leaf ⋅ x ,
∂ z ‾ = 1 2 ( ∂ ℜ z + i ∂ ℑ z ) , D θ = 2 ∂ z ‾ . (54) \partial_{\overline{z}}=\frac{1}{2}(\partial_{\Re z}+i\partial_{\Im z}),\qquad D_\theta=2\partial_{\overline{z}}. \tag*{(54)} ∂ z = 2 1 ( ∂ ℜ z + i ∂ ℑ z ) , D θ = 2 ∂ z . (54)
The coordinate vectors in this display are local to the normalized frame, not the coordinate vectors used in (2.21). For e θ e_\theta e θ from (2.20), one has A θ e θ = 0 A_\theta e_\theta=0 A θ e θ = 0 . Define the primal quotient and its bilinear dual by
E θ = C 4 / C e θ , E θ ∗ = { ( p , η ) ∈ C 3 × C : θ T p = 0 } . (55) E_\theta=\mathbb{C}^4/\mathbb{C}e_\theta,\qquad E_\theta^*=\{(p,\eta)\in\mathbb{C}^3\times\mathbb{C}:\theta^Tp=0\}. \tag*{(55)} E θ = C 4 / C e θ , E θ ∗ = {( p , η ) ∈ C 3 × C : θ T p = 0 } . (55)
Introduce homogeneous spinor coordinates [ s 0 : s 1 ] ∈ C P 1 [s_0:s_1]\in\mathbb{CP}^1 [ s 0 : s 1 ] ∈ CP 1 and
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 θ = ( − s 1 , s 0 ) T , \theta=(s_0^2-s_1^2,i(s_0^2+s_1^2),2s_0s_1)^T,\qquad Q=\frac{1}{2}\begin{pmatrix}-s_1&-s_0\\ is_1&-is_0\\ s_0&-s_1\end{pmatrix},\qquad n_\theta=(-s_1,s_0)^T, 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 θ = ( − s 1 , s 0 ) T ,
θ T Q = 0 , Q n θ = − θ / 2 , r a = − 2 a T Q ( a ∈ C 3 ) , U θ = ( Q 0 0 1 ) . (56) \theta^TQ=0,\qquad Qn_\theta=-\theta/2,\qquad r_a=-2a^TQ\ (a\in\mathbb{C}^3),\qquad U_\theta=\begin{pmatrix}Q&0\\0&1\end{pmatrix}. \tag*{(56)} θ T Q = 0 , Q n θ = − θ /2 , r a = − 2 a T Q ( a ∈ C 3 ) , U θ = ( Q 0 0 1 ) . (56)
The matrix Q Q Q has size 3 × 2 3\times2 3 × 2 , r a r_a r a has size 1 × 2 1\times2 1 × 2 , and U θ U_\theta U θ is a 4 × 3 4\times3 4 × 3 embedding of E θ ∗ E_\theta^* E θ ∗ from (2.38). On the two affine spinor charts, set
ζ = s 1 / s 0 , υ = s 0 / s 1 = 1 / ζ , D t r ( ζ ) = diag ( ζ I 2 , 1 ) . (57) \zeta=s_1/s_0,\qquad\upsilon=s_0/s_1=1/\zeta,\qquad D_{\mathrm{tr}}(\zeta)=\operatorname{diag}(\zeta I_2,1). \tag*{(57)} ζ = s 1 / s 0 , υ = s 0 / s 1 = 1/ ζ , D tr ( ζ ) = diag ( ζ I 2 , 1 ) . (57)
Here n θ n_\theta n θ is a two-vector, never the polynomial degree. Q Q Q has rank two: Q ( a , b ) T = 0 Q(a,b)^T=0 Q ( a , b ) T = 0 implies s 1 a = s 0 b = s 0 a − s 1 b = 0 s_1a=s_0b=s_0a-s_1b=0 s 1 a = s 0 b = s 0 a − s 1 b = 0 , including the chart axes. On a finite refinement of compact spinor charts select three rows of U θ U_\theta U θ with a nonzero minor. We invert the selected nonzero minor of U θ U_\theta U θ and obtain a holomorphic 4 × 3 4\times3 4 × 3 matrix V θ V_\theta V θ satisfying
U θ T V θ = I 3 , ker U θ T = C e θ . (58) U_\theta^TV_\theta=I_3,\qquad\ker U_\theta^T=\mathbb{C}e_\theta. \tag*{(58)} U θ T V θ = I 3 , ker U θ T = C e θ . (58)
Each selected determinant is bounded away from zero after shrinking the compact subchart; the minimum over the finite cover is positive. This is the additional geometric denominator, independent of materials. More explicitly, let q 12 , q 13 , q 23 q_{12},q_{13},q_{23} q 12 , q 13 , q 23 be the two-row minors of Q Q Q . Direct substitution in (2.39) gives
∣ q 12 ∣ 2 + ∣ q 13 ∣ 2 + ∣ q 23 ∣ 2 = ( ∣ s 0 ∣ 2 + ∣ s 1 ∣ 2 ) 2 8 . (59) |q_{12}|^2+|q_{13}|^2+|q_{23}|^2=\frac{(|s_0|^2+|s_1|^2)^2}{8}. \tag*{(59)} ∣ q 12 ∣ 2 + ∣ q 13 ∣ 2 + ∣ q 23 ∣ 2 = 8 ( ∣ s 0 ∣ 2 + ∣ s 1 ∣ 2 ) 2 . (59)
At a unit spinor at least one has modulus at least 1 / 24 1/\sqrt{24} 1/ 24 ; the corresponding three-row minor of U θ U_\theta U θ has that same modulus. Fixed smaller chart neighborhoods therefore give an explicit geometric lower bound before any material is chosen. On a holomorphic chart overlap the paired local choices obey
V 0 = V ∞ D t r − T + e θ 0 L , (60) V_0=V_\infty D_{\mathrm{tr}}^{-T}+e_{\theta_0}L, \tag*{(60)} V 0 = V ∞ D tr − T + e θ 0 L , (60)
where D t r D_{\mathrm{tr}} D tr is defined in (2.40) and L L L is a holomorphic 1 × 3 1\times3 1 × 3 row. In fact U 0 = U ∞ D t r U_0=U_\infty D_{\mathrm{tr}} U 0 = U ∞ D tr , so applying U 0 T U_0^T U 0 T to the difference gives zero. Its kernel is the characteristic line. A fixed nonzero component of e θ 0 e_{\theta_0} e θ 0 on a smaller chart extracts the holomorphic row L L L .
Substitution of y = U θ ( ξ , η ) T y=U_\theta(\xi,\eta)^T y = U θ ( ξ , η ) T into 2 D θ y − A θ T y = 0 2D_\theta y-A_\theta^T y=0 2 D θ y − A θ T y = 0 gives
Γ = ( n θ r g μ 2 d n θ − θ T F μ Q 0 ) , A = U θ T A θ V θ = − Γ T , P = A / 2. (61) \Gamma=\begin{pmatrix} n_\theta r_{g_\mu} & 2d n_\theta\\ -\theta^T F_\mu Q & 0 \end{pmatrix},\qquad\mathcal{A}=U_\theta^T A_\theta V_\theta=-\Gamma^T,\qquad P=\mathcal{A}/2. \tag*{(61)} Γ = ( n θ r g μ − θ T F μ Q 2 d n θ 0 ) , A = U θ T A θ V θ = − Γ T , P = A /2. (61)
Here Γ , A , P \Gamma,\mathcal{A},P Γ , A , P are 3 × 3 3\times3 3 × 3 connection matrices; Q , n θ , r g μ Q,n_\theta,r_{g_\mu} Q , n θ , r g μ are defined in (56) , and d , F μ d,F_\mu d , F μ in (31) and (33) . The substitution gives A θ T U θ = − U θ Γ A_\theta^T U_\theta=-U_\theta\Gamma A θ T U θ = − U θ Γ ; transpose this equality and multiply by V θ V_\theta V θ . Thus the primal and dual frame equations are
( 2 D θ + A ) X = 0 , ( 2 D θ + Γ ) Y = 0 , or 2 ∂ z ˉ X + P X = 0 , 2 ∂ z ˉ Y − P T Y = 0. (62) (2D_\theta+\mathcal{A})X=0,\qquad(2D_\theta+\Gamma)Y=0,\qquad\text{or}\qquad2\partial_{\bar z}X+PX=0,\qquad2\partial_{\bar z}Y-P^T Y=0. \tag*{(62)} ( 2 D θ + A ) X = 0 , ( 2 D θ + Γ ) Y = 0 , or 2 ∂ z ˉ X + P X = 0 , 2 ∂ z ˉ Y − P T Y = 0. (62)
The matrices X , Y X,Y X , Y are 3 × 3 3\times3 3 × 3 primal and dual quotient frames, respectively; D θ D_\theta D θ is defined in (37) . For any given primal solution X X X , the defect of V θ X V_\theta X V θ X lies in C e θ \mathbb{C}e_\theta C e θ . Choose a spatially constant row on the chart satisfying
ℓ θ ∈ C 1 × 4 , ℓ θ e θ = 1. (63) \ell_\theta\in\mathbb{C}^{1\times4},\qquad\ell_\theta e_\theta=1. \tag*{(63)} ℓ θ ∈ C 1 × 4 , ℓ θ e θ = 1. (63)
Solve
2 D θ α l i f t = − ℓ θ ( 2 D θ + A θ ) V θ X , Z = V θ X + e θ α l i f t , 2D_\theta\alpha_{\mathrm{lift}}=-\ell_\theta(2D_\theta+A_\theta)V_\theta X,\qquad Z=V_\theta X+e_\theta\alpha_{\mathrm{lift}}, 2 D θ α lift = − ℓ θ ( 2 D θ + A θ ) V θ X , Z = V θ X + e θ α lift ,
( 2 D θ + A θ ) Z = 0 , U θ T Z = X , U θ T A Δ , θ Z 1 = 2 ( P 1 − P 2 ) X 1 . (64) (2D_\theta+A_\theta)Z=0,\qquad U_\theta^T Z=X,\qquad U_\theta^T A_{\Delta,\theta}Z_1=2(P_1-P_2)X_1. \tag*{(64)} ( 2 D θ + A θ ) Z = 0 , U θ T Z = X , U θ T A Δ , θ Z 1 = 2 ( P 1 − P 2 ) X 1 . (64)
Here Z Z Z is a 4 × 3 4\times3 4 × 3 ambient amplitude, α l i f t \alpha_{\mathrm{lift}} α lift is a 1 × 3 1\times3 1 × 3 row, and A Δ , θ A_{\Delta,\theta} A Δ , θ is defined in (50) . The construction of X X X with uniform finite-order bounds uses Theorem 3.1 . The row α l i f t \alpha_{\mathrm{lift}} α lift has size 1 × 3 1\times3 1 × 3 . Its three scalar equations are solved on fixed leaf disks as described in § 4 .
With υ \upsilon υ and D t r D_{\mathrm{tr}} D tr as in (57) , the representatives ( 1 , ζ ) (1,\zeta) ( 1 , ζ ) and ( υ , 1 ) (\upsilon,1) ( υ , 1 ) give
Q 0 = ζ Q ∞ , θ 0 = ζ 2 θ ∞ , Γ 0 = ζ 2 D t r − 1 Γ ∞ D t r , Q_0=\zeta Q_\infty,\qquad\theta_0=\zeta^2\theta_\infty,\qquad\Gamma_0=\zeta^2D_{\mathrm{tr}}^{-1}\Gamma_\infty D_{\mathrm{tr}}, Q 0 = ζ Q ∞ , θ 0 = ζ 2 θ ∞ , Γ 0 = ζ 2 D tr − 1 Γ ∞ D tr ,
H ( 0 ) = D t r − 1 H ( ∞ ) D t r , G ( 0 ) = D t r T G ( ∞ ) D t r − T , H = G − T . (65) H^{(0)}=D_{\mathrm{tr}}^{-1}H^{(\infty)}D_{\mathrm{tr}},\qquad G^{(0)}=D_{\mathrm{tr}}^T G^{(\infty)}D_{\mathrm{tr}}^{-T},\qquad H=G^{-T}. \tag*{(65)} H ( 0 ) = D tr − 1 H ( ∞ ) D tr , G ( 0 ) = D tr T G ( ∞ ) D tr − T , H = G − T . (65)
These are coordinate laws for a given endomorphism, not an assertion that separately constructed transports agree. When θ n = a θ h \theta_n=a\theta_h θ n = a θ h for a positive, spatially constant normalization factor, define
S a = diag ( a I 2 , 1 ) , a > 0. (66) S_a=\operatorname{diag}(\sqrt{a}I_2,1),\qquad a>0. \tag*{(66)} S a = diag ( a I 2 , 1 ) , a > 0. (66)
Then
U n = U h S a , Γ n = a S a − 1 Γ h S a , H h = S a H n S a − 1 . (67) U_n=U_hS_a,\qquad\Gamma_n=aS_a^{-1}\Gamma_hS_a,\qquad H_h=S_aH_nS_a^{-1}. \tag*{(67)} U n = U h S a , Γ n = a S a − 1 Γ h S a , H h = S a H n S a − 1 . (67)
The factor a a a need not be holomorphic. Antiholomorphic derivatives must be taken after returning to holomorphic coordinates and at fixed physical x x x . In particular,
∂ ζ ˉ H ( 0 ) = − ζ ˉ − 2 D t r − 1 ( ∂ υ ˉ H ( ∞ ) ) D t r . (68) \partial_{\bar\zeta}H^{(0)}=-\bar\zeta^{-2}D_{\mathrm{tr}}^{-1}(\partial_{\bar\upsilon}H^{(\infty)})D_{\mathrm{tr}}. \tag*{(68)} ∂ ζ ˉ H ( 0 ) = − ζ ˉ − 2 D tr − 1 ( ∂ υ ˉ H ( ∞ ) ) D tr . (68)
For a spatially constant b ∈ C ∗ b\in\mathbb{C}^* b ∈ C ∗ , the characteristic line and affine complex leaves of θ \theta θ and b θ b\theta b θ agree. Choose a local square root of b b b and define
S b = diag ( b I 2 , 1 ) , b ∈ C ∗ . (69) S_b=\operatorname{diag}(\sqrt{b}I_2,1),\qquad b\in\mathbb{C}^*. \tag*{(69)} S b = diag ( b I 2 , 1 ) , b ∈ C ∗ . (69)
Then
Γ b = b S b − 1 Γ S b , H b = S b − 1 H S b , 2 D b θ H b + Γ 1 , b H b − H b Γ 2 , b = b S b − 1 ( 2 D θ H + Γ 1 H − H Γ 2 ) S b . (70) \Gamma_b=bS_b^{-1}\Gamma S_b,\qquad H_b=S_b^{-1}HS_b,\qquad2D_{b\theta}H_b+\Gamma_{1,b}H_b-H_b\Gamma_{2,b}=bS_b^{-1}(2D_\theta H+\Gamma_1H-H\Gamma_2)S_b. \tag*{(70)} Γ b = b S b − 1 Γ S b , H b = S b − 1 H S b , 2 D b θ H b + Γ 1 , b H b − H b Γ 2 , b = b S b − 1 ( 2 D θ H + Γ 1 H − H Γ 2 ) S b . (70)
This proves representative invariance of the equation and its identity boundary normalization. Lemma 5.3 compares independently constructed local transports after these coordinate changes.
The common-collar Betti identity and its exact measurement domain
Under agreement near the working sphere, the Betti identity relates the reduced volume pairing to the elastic displacement–traction pairing. When the materials also agree in the region between the measurement boundary and the working sphere, their DN discrepancies can be compared without an exterior mismatch term.
We use the elastic DN maps Λ j D \Lambda_j^D Λ j D from (7) , with the weak pairing (19) and discrepancy (16) . In particular, δ B R \delta_{B_R} δ B R is obtained by taking D = B R D=B_R D = B R .
Under agreement near the working sphere, the factorization transfers the reduced volume identity to a pairing of genuine elastic solutions. The next lemma identifies the corresponding DN pairing on ∂ B R \partial B_R ∂ B R .
Lemma 2.3. Let the real C 4 C^4 C 4 materials be uniformly strongly convex on a neighborhood O O O of B R ‾ \overline{B_R} B R . Suppose that there is an open set N ⊂ O N\subset O N ⊂ O , with ∂ B R ⊂ N \partial B_R\subset N ∂ B R ⊂ N , on which both λ 1 = λ 2 \lambda_1=\lambda_2 λ 1 = λ 2 and μ 1 = μ 2 \mu_1=\mu_2 μ 1 = μ 2 . Let z 1 ∈ H 4 ( O ; C 4 ) z_1\in H^4(O;\mathbb{C}^4) z 1 ∈ H 4 ( O ; C 4 ) solve M 1 z 1 = 0 \mathcal{M}_1z_1=0 M 1 z 1 = 0 on O O O , and let u 2 ∈ H 1 ( B R ; C 3 ) u_2\in H^1(B_R;\mathbb{C}^3) u 2 ∈ H 1 ( B R ; C 3 ) solve L 2 u 2 = 0 L_2u_2=0 L 2 u 2 = 0 . Set
u 1 = T 1 z 1 , y 2 = S 2 T u 2 ∈ L 2 ( B R ; C 4 ) , (71) u_1=T_1z_1,\qquad y_2=S_2^T u_2\in L^2(B_R;\mathbb{C}^4), \tag*{(71)} u 1 = T 1 z 1 , y 2 = S 2 T u 2 ∈ L 2 ( B R ; C 4 ) , (71) where T j T_j T j , S j T S_j^T S j T are defined in (34) and (41) . Then
∫ B R y 2 T ( M 1 − M 2 ) z 1 d x = ⟨ ( Λ 2 B R − Λ 1 B R ) Tr u 1 , Tr u 2 ⟩ . (72) \int_{B_R}y_2^T(\mathcal{M}_1-\mathcal{M}_2)z_1\,dx
=\left\langle(\Lambda_2^{B_R}-\Lambda_1^{B_R})\operatorname{Tr}u_1,\operatorname{Tr}u_2\right\rangle. \tag*{(72)} ∫ B R y 2 T ( M 1 − M 2 ) z 1 d x = ⟨ ( Λ 2 B R − Λ 1 B R ) Tr u 1 , Tr u 2 ⟩ . (72) The boundary duality is complex bilinear and has traction in its first slot.
Proof. Use the bilinear energy a B R a_{B_R} a B R from (18) . For u ∈ H 1 ( B R ; C 3 ) u\in H^1(B_R;\mathbb{C}^3) u ∈ H 1 ( B R ; C 3 ) with L u ∈ L 2 Lu\in L^2 Lu ∈ L 2 , define its weak traction by
⟨ t u , g ⟩ = a B R ( u , E tr g ) + ∫ B R ( L u ) T E tr g d x , (73) \langle t_u,g\rangle=a_{B_R}(u,\mathcal{E}_{\operatorname{tr}}g)+\int_{B_R}(Lu)^T\mathcal{E}_{\operatorname{tr}}g\,dx, \tag*{(73)} ⟨ t u , g ⟩ = a B R ( u , E tr g ) + ∫ B R ( Lu ) T E tr g d x , (73)
where E tr : H 1 / 2 ( ∂ B R ; C 3 ) → H 1 ( B R ; C 3 ) \mathcal{E}_{\operatorname{tr}}:H^{1/2}(\partial B_R;\mathbb{C}^3)\to H^1(B_R;\mathbb{C}^3) E tr : H 1/2 ( ∂ B R ; C 3 ) → H 1 ( B R ; C 3 ) is a bounded right inverse of the trace. Independence of E tr \mathcal{E}_{\operatorname{tr}} E tr follows by testing the distributional equation against the zero-trace difference. Hence ∥ t u ∥ H − 1 / 2 ≤ C ( ∥ u ∥ H 1 + ∥ L u ∥ 2 ) \|t_u\|_{H^{-1/2}}\leq C(\|u\|_{H^1}+\|Lu\|_2) ∥ t u ∥ H − 1/2 ≤ C ( ∥ u ∥ H 1 + ∥ Lu ∥ 2 ) , and subtraction gives
∫ B R ( u T L v − v T L u ) = ⟨ t v , Tr u ⟩ − ⟨ t u , Tr v ⟩ . (74) \int_{B_R}(u^TLv-v^TLu)=\langle t_v,\operatorname{Tr}u\rangle-\langle t_u,\operatorname{Tr}v\rangle. \tag*{(74)} ∫ B R ( u T Lv − v T Lu ) = ⟨ t v , Tr u ⟩ − ⟨ t u , Tr v ⟩ . (74)
For L u = L v = 0 Lu=Lv=0 Lu = Lv = 0 , identity (74) gives ⟨ Λ Tr u , Tr v ⟩ = ⟨ Λ Tr v , Tr u ⟩ \langle\Lambda\operatorname{Tr}u,\operatorname{Tr}v\rangle=\langle\Lambda\operatorname{Tr}v,\operatorname{Tr}u\rangle ⟨ Λ Tr u , Tr v ⟩ = ⟨ Λ Tr v , Tr u ⟩ . Thus Λ \Lambda Λ is symmetric for the bilinear pairing. Uniform solvability follows from (22) and (23) .
Define
w = M 2 z 1 , v = T 2 z 1 . (75) w=\mathcal{M}_2z_1,\qquad v=T_2z_1. \tag*{(75)} w = M 2 z 1 , v = T 2 z 1 . (75)
On the common collar N N N of Lemma 2.3,
w = 0 , v = u 1 , t v ( 2 ) = t u 1 ( 1 ) . (76) w=0,\qquad v=u_1,\qquad t_v^{(2)}=t_{u_1}^{(1)}. \tag*{(76)} w = 0 , v = u 1 , t v ( 2 ) = t u 1 ( 1 ) . (76)
The complete transpose transfer is
∫ B R ( S 2 T u 2 ) T w = ∫ B R u 2 T S 2 w − ∫ ∂ B R ( u 2 ⋅ ν ) w 4 1 − κ 2 d S . (77) \int_{B_R}(S_2^Tu_2)^Tw=\int_{B_R}u_2^TS_2w-\int_{\partial B_R}(u_2\cdot\nu)\frac{w_4}{1-\kappa_2}\,dS. \tag*{(77)} ∫ B R ( S 2 T u 2 ) T w = ∫ B R u 2 T S 2 w − ∫ ∂ B R ( u 2 ⋅ ν ) 1 − κ 2 w 4 d S . (77)
The last term is zero, precisely because w w w vanishes on N N N . Since M 1 z 1 = 0 \mathcal{M}_1z_1=0 M 1 z 1 = 0 and S 2 M 2 z 1 = L 2 v S_2\mathcal{M}_2z_1=L_2v S 2 M 2 z 1 = L 2 v , the left side of (72) is − ∫ B R u 2 T L 2 v -\int_{B_R}u_2^TL_2v − ∫ B R u 2 T L 2 v . By (74) , equality of the collar tractions, and the bilinear symmetry of Λ 2 \Lambda_2 Λ 2 , we obtain − ⟨ Λ 1 Tr u 1 , Tr u 2 ⟩ + ⟨ Λ 2 Tr u 1 , Tr u 2 ⟩ -\langle\Lambda_1\operatorname{Tr}u_1,\operatorname{Tr}u_2\rangle+\langle\Lambda_2\operatorname{Tr}u_1,\operatorname{Tr}u_2\rangle − ⟨ Λ 1 Tr u 1 , Tr u 2 ⟩ + ⟨ Λ 2 Tr u 1 , Tr u 2 ⟩ . There is no assumption that the comparison field v v v is homogeneous. □
When the materials agree between an inner measurement boundary and the working sphere, their DN discrepancies satisfy the following comparison.
Lemma 2.4. Let D ‾ ⊂ B R \overline{D} \subset B_R D ⊂ B R be fixed smooth domains and let the two uniformly strongly convex bounded materials coincide on B R ∖ D ‾ B_R \setminus\overline{D} B R ∖ D . Then
δ B R ≤ C D , R δ D . (78) \delta_{B_R} \le C_{D,R}\delta_D. \tag*{(78)} δ B R ≤ C D , R δ D . (78) Here both discrepancies are defined in (16) . The constant depends only on the fixed domains and the uniform upper and strong-convexity bounds for the coefficients.
Proof. Let L 1 u = L 2 v = 0 L_1u=L_2v=0 L 1 u = L 2 v = 0 on B R B_R B R , with traces f , g f,g f , g . Applying (27) on B R B_R B R and D D D , and using equality of the coefficients outside D ‾ \overline{D} D , we obtain
⟨ ( Λ 1 B R − Λ 2 B R ) f , g ⟩ = ∫ D [ ( λ 1 − λ 2 ) div u div v + 2 ( μ 1 − μ 2 ) ε ( u ) : ε ( v ) ] d x = ⟨ ( Λ 1 D − Λ 2 D ) Tr D u , Tr D v ⟩ . (79) \begin{aligned}
\langle(\Lambda^{B_R}_1-\Lambda^{B_R}_2)f,g\rangle
&=\int_D\left[(\lambda_1-\lambda_2)\operatorname{div}u\operatorname{div}v+2(\mu_1-\mu_2)\varepsilon(u):\varepsilon(v)\right]dx\\
&=\langle(\Lambda^D_1-\Lambda^D_2)\operatorname{Tr}_D u,\operatorname{Tr}_D v\rangle. \tag*{(79)}
\end{aligned} ⟨( Λ 1 B R − Λ 2 B R ) f , g ⟩ = ∫ D [ ( λ 1 − λ 2 ) div u div v + 2 ( μ 1 − μ 2 ) ε ( u ) : ε ( v ) ] d x = ⟨( Λ 1 D − Λ 2 D ) Tr D u , Tr D v ⟩ . (79) The restrictions are solutions of their respective equations on D D D . The fixed-domain trace theorem and (23) imply
∥ Tr D u ∥ H 1 / 2 ≤ C ∥ f ∥ H 1 / 2 ( ∂ B R ) , ∥ Tr D v ∥ H 1 / 2 ≤ C ∥ g ∥ H 1 / 2 ( ∂ B R ) . (80) \|\operatorname{Tr}_D u\|_{H^{1/2}}\le C\|f\|_{H^{1/2}(\partial B_R)},\qquad\|\operatorname{Tr}_D v\|_{H^{1/2}}\le C\|g\|_{H^{1/2}(\partial B_R)}. \tag*{(80)} ∥ Tr D u ∥ H 1/2 ≤ C ∥ f ∥ H 1/2 ( ∂ B R ) , ∥ Tr D v ∥ H 1/2 ≤ C ∥ g ∥ H 1/2 ( ∂ B R ) . (80) Taking the supremum in (79) over f , g ∈ H 1 / 2 ( ∂ B R ; C 3 ) f,g\in H^{1/2}(\partial B_R;\mathbb{C}^3) f , g ∈ H 1/2 ( ∂ B R ; C 3 ) with ∥ f ∥ H 1 / 2 = ∥ g ∥ H 1 / 2 = 1 \|f\|_{H^{1/2}}=\|g\|_{H^{1/2}}=1 ∥ f ∥ H 1/2 = ∥ g ∥ H 1/2 = 1 , and using (80) , we obtain (78) . Both discrepancies are defined in (16) .
When the measurements are made directly on ∂ B R \partial B_R ∂ B R , no domain comparison is needed. Otherwise, Lemma 2.4 transfers the discrepancy from ∂ D \partial D ∂ D to ∂ B R \partial B_R ∂ B R under its stated exterior agreement assumption. In the prescribed-exterior problem, D = Ω D=\Omega D = Ω and δ D = δ \delta_D=\delta δ D = δ . For the paired extensions constructed in § 9 , we take D = B R e D=B_{R_e} D = B R e and δ D = Δ e \delta_D=\Delta_e δ D = Δ e ; the comparison with the original error δ \delta δ retains the exterior mismatch through Proposition 2.1. The collar assumption of Lemma 2.3 is imposed at the working sphere, not on the original materials at ∂ Ω \partial\Omega ∂ Ω .
Uniform finite-regularity frames and a local right inverse
Two analytic results underlie the CGO construction: finite-order leafwise transport frames and a local right inverse for the conjugated elliptic system. Their regularity requirements are distinct. The matrix-transport construction for conjugated systems is described in [18 Sections 1–2] . The finite-order uniform estimates used here are proved in Subsections B.2 and B.7. The frame existence argument uses [20 Theorem 1.3] ; the quantitative bounds are obtained in the proof.
Coefficient class and fixed geometry
The coefficient support and nested domains below are fixed independently of the source and phase direction.
In this section d s y s ≥ 1 d_{\mathrm{sys}}\ge1 d sys ≥ 1 is the system dimension, not the scalar material coefficient denoted by d d d elsewhere. The index k ≥ 0 k\ge0 k ≥ 0 is an integer Sobolev order. Fix balls Ω 1 ⋐ Ω 0 ⊂ R 3 \Omega_1\Subset\Omega_0\subset\mathbb{R}^3 Ω 1 ⋐ Ω 0 ⊂ R 3 , a compact source ball K s r c ⋐ Ω 1 K_{\mathrm{src}}\Subset\Omega_1 K src ⋐ Ω 1 , and a compact set K c o e f ⋐ int K s r c K_{\mathrm{coef}}\Subset\operatorname{int}K_{\mathrm{src}} K coef ⋐ int K src . All containments mean positive separation of the corresponding closed sets. Set
M ~ = Δ I d s y s + ∑ j = 1 3 A ~ j ( x ) ∂ j + B ~ ( x ) , A ~ j , B ~ ∈ C c k + 14 ( R 3 ; C d s y s × d s y s ) , supp ( A ~ j ) , supp ( B ~ ) ⊂ K c o e f , j = 1 , 2 , 3 , ∑ j = 1 3 ∥ A ~ j ∥ C k + 14 + ∥ B ~ ∥ C k + 14 ≤ M k . (81) \begin{aligned}
\widetilde{\mathcal{M}}&=\Delta I_{d_{\mathrm{sys}}}+\sum_{j=1}^{3}\widetilde{A}_j(x)\partial_j+\widetilde{B}(x),\\
\widetilde{A}_j,\widetilde{B}&\in C_c^{k+14}(\mathbb{R}^3;\mathbb{C}^{d_{\mathrm{sys}}\times d_{\mathrm{sys}}}),\\
\operatorname{supp}(\widetilde{A}_j),\operatorname{supp}(\widetilde{B})&\subset K_{\mathrm{coef}},\qquad j=1,2,3,\\
\sum_{j=1}^{3}\|\widetilde{A}_j\|_{C^{k+14}}+\|\widetilde{B}\|_{C^{k+14}}&\le M_k. \tag*{(81)}
\end{aligned} M A j , B supp ( A j ) , supp ( B ) j = 1 ∑ 3 ∥ A j ∥ C k + 14 + ∥ B ∥ C k + 14 = Δ I d sys + j = 1 ∑ 3 A j ( x ) ∂ j + B ( x ) , ∈ C c k + 14 ( R 3 ; C d sys × d sys ) , ⊂ K coef , j = 1 , 2 , 3 , ≤ M k . (81)
The class in (3.1) includes all complex matrix coefficients with this prior; ellipticity here comes from the fixed Laplacian principal part. Define the closed source space
E k = { f ∈ H k ( Ω 0 ; C d s y s ) : supp f ⊂ K s r c } . (82) E_k=\{f\in H^k(\Omega_0;\mathbb{C}^{d_{\mathrm{sys}}}):\operatorname{supp}f\subset K_{\mathrm{src}}\}. \tag*{(82)} E k = { f ∈ H k ( Ω 0 ; C d sys ) : supp f ⊂ K src } . (82)
Support is distributional. Extension by zero identifies this space with a closed subspace of H k ( R 3 ; C d s y s ) H^k(\mathbb{R}^3;\mathbb{C}^{d_{\mathrm{sys}}}) H k ( R 3 ; C d sys ) ; it is bounded because the support stays a fixed distance from ∂ Ω 0 \partial\Omega_0 ∂ Ω 0 .
The normalized direction set and the conjugated operator are
Θ = { e 1 + i e 2 : e 1 , e 2 ∈ R 3 , ∣ e 1 ∣ = ∣ e 2 ∣ = 1 , e 1 ⋅ e 2 = 0 } , \Theta=\{e_1+ie_2:e_1,e_2\in\mathbb{R}^3,\ |e_1|=|e_2|=1,\ e_1\cdot e_2=0\}, Θ = { e 1 + i e 2 : e 1 , e 2 ∈ R 3 , ∣ e 1 ∣ = ∣ e 2 ∣ = 1 , e 1 ⋅ e 2 = 0 } ,
L h , θ = e θ ⋅ x / h M ~ e − θ ⋅ x / h = M ~ − h − 1 ( 2 θ ⋅ ∇ + ∑ j θ j A ~ j ) . (83) L_{h,\theta}=e^{\theta\cdot x/h}\widetilde{\mathcal{M}}e^{-\theta\cdot x/h}
=\widetilde{\mathcal{M}}-h^{-1}\left(2\theta\cdot\nabla+\sum_j\theta_j\widetilde{A}_j\right). \tag*{(83)} L h , θ = e θ ⋅ x / h M e − θ ⋅ x / h = M − h − 1 ( 2 θ ⋅ ∇ + j ∑ θ j A j ) . (83)
The dot product in the null condition θ ⋅ θ = 0 \theta\cdot\theta=0 θ ⋅ θ = 0 is bilinear. No data discrepancy is used in this section; in particular δ \delta δ is not reused as a Fourier parameter.
Finite-order leaf frames
Uniform leafwise frames are obtained on complete disks from a finite atlas over the compact coefficient–parameter class.
Let D ⋐ U ⊂ C D\Subset U\subset\mathbb{C} D ⋐ U ⊂ C be fixed disks, r ≥ 0 r\geq0 r ≥ 0 an integer, and 0 < α H < 1 0<\alpha_H<1 0 < α H < 1 . The operator ∂ ‾ z \overline{\partial}_z ∂ z is defined in (2.37). For the uniform assertion, fix a compact finite-dimensional smooth real parameter manifold P \mathfrak{P} P and larger spatial and parameter neighborhoods. Compact charts with boundary are allowed. A jointly bounded C r + 2 C^{r+2} C r + 2 family { B α ( z , p ) } \{\mathrm{B}^{\alpha}(z,p)\} { B α ( z , p )} has the weaker compact closure
B r = { B α } C r + 1 , α H , ( α , p ) ∈ B r × P , (84) \mathfrak{B}_r=\{\mathrm{B}^{\alpha}\}^{C^{r+1,\alpha_H}},\qquad(\alpha,p)\in\mathfrak{B}_r\times\mathfrak{P}, \tag*{(84)} B r = { B α } C r + 1 , α H , ( α , p ) ∈ B r × P , (84)
where each B α \mathrm{B}^{\alpha} B α takes values in C d s y s × d s y s \mathbb{C}^{d_{\mathrm{sys}}\times d_{\mathrm{sys}}} C d sys × d sys . The coefficient label has the topology in (3.4); it is not a differentiation variable.
Theorem 3.1. Let D ⋐ U ⊂ C D\Subset U\subset\mathbb{C} D ⋐ U ⊂ C be the fixed disks, r ≥ 0 r\geq0 r ≥ 0 an integer, and 0 < α H < 1 0<\alpha_H<1 0 < α H < 1 . For every B ∈ C r , α H ( U ; C d s y s × d s y s ) \mathrm{B}\in C^{r,\alpha_H}(U;\mathbb{C}^{d_{\mathrm{sys}}\times d_{\mathrm{sys}}}) B ∈ C r , α H ( U ; C d sys × d sys ) , there exist a neighborhood V ⊂ U V\subset U V ⊂ U of D ‾ \overline{D} D and an invertible frame satisfying
F ∈ C r + 1 , α H ( V ; G L ( d s y s , C ) ) , ∂ ‾ z F = B F in V . (85) F\in C^{r+1,\alpha_H}(V;GL(d_{\mathrm{sys}},\mathbb{C})),\qquad\overline{\partial}_zF=\mathrm{B}F\quad\text{in }V. \tag*{(85)} F ∈ C r + 1 , α H ( V ; G L ( d sys , C )) , ∂ z F = B F in V . (85) For the compact coefficient class B r \mathfrak{B}_r B r in (3.4) and the fixed parameter space P \mathfrak{P} P , there exists a finite cover by coefficient–parameter neighborhoods carrying invertible frames F ℓ α ( z , p ) F_\ell^\alpha(z,p) F ℓ α ( z , p ) , defined near D ‾ \overline{D} D , satisfying
∂ ‾ z F ℓ α = B α F ℓ α . \overline{\partial}_zF_\ell^\alpha=\mathrm{B}^\alpha F_\ell^\alpha. ∂ z F ℓ α = B α F ℓ α . On smaller neighborhoods still covering B r × P \mathfrak{B}_r\times\mathfrak{P} B r × P , these frames and their inverses satisfy
max ∣ γ ∣ + ∣ β ∣ ≤ r ( ∥ ∂ z γ ∂ p β F ℓ α ∥ ∞ + ∥ ∂ z γ ∂ p β ( F ℓ α ) − 1 ∥ ∞ ) ≤ C r , (86) \max_{|\gamma|+|\beta|\leq r}\left(\left\|\partial_z^\gamma\partial_p^\beta F_\ell^\alpha\right\|_\infty+\left\|\partial_z^\gamma\partial_p^\beta(F_\ell^\alpha)^{-1}\right\|_\infty\right)\leq C_r, \tag*{(86)} ∣ γ ∣ + ∣ β ∣ ≤ r max ( ∂ z γ ∂ p β F ℓ α ∞ + ∂ z γ ∂ p β ( F ℓ α ) − 1 ∞ ) ≤ C r , (86) where γ \gamma γ denotes differentiation in the two real leaf variables. Both the frames and their inverses depend continuously on α \alpha α in the mixed norms of total order at most r r r .
The cover admits subordinate weights w ℓ ( α , p ) w_\ell(\alpha,p) w ℓ ( α , p ) , continuous in α \alpha α and smooth in p p p , such that
w ℓ ≥ 0 , ∑ ℓ w ℓ 2 = 1 , max ∣ β ∣ ≤ r ∥ ∂ p β w ℓ ∥ ∞ ≤ C r . (87) w_\ell\geq0,\qquad\sum_\ell w_\ell^2=1,\qquad\max_{|\beta|\leq r}\|\partial_p^\beta w_\ell\|_\infty\leq C_r. \tag*{(87)} w ℓ ≥ 0 , ℓ ∑ w ℓ 2 = 1 , ∣ β ∣ ≤ r max ∥ ∂ p β w ℓ ∥ ∞ ≤ C r . (87) When p p p consists of an oriented direction and a transverse coordinate, the weights are constant along each complex leaf.
The finite cover and the constants depend only on the fixed coefficient class, disks, neighborhood margins, and parameter geometry. Finitely many systems admit a common finite product refinement with the same properties.
Remark. The uniform estimates are obtained from a finite atlas of local frames; no globally defined frame over the parameter space is required. The coefficient label a a a enters only continuously and is never differentiated. This distinction allows the construction to be uniform over the compact coefficient class.
The proof is given in Subsection B.2.
The local right inverse and residual estimates
Theorem 3.2. For the full coefficient class (3.1) there exist h ∗ > 0 h_* > 0 h ∗ > 0 and C k < ∞ C_k < \infty C k < ∞ , depending only on k k k , d s y s d_{\mathrm{sys}} d sys , M k M_k M k , K c o e f K_{\mathrm{coef}} K coef , K s r c K_{\mathrm{src}} K src , Ω 1 \Omega_1 Ω 1 , Ω 0 \Omega_0 Ω 0 and the fixed cutoff geometry, with the following property. For every member of the class, every θ ∈ Θ \theta\in\Theta θ ∈ Θ , and every 0 < h < h ∗ 0 < h < h_* 0 < h < h ∗ , there is a linear operator
R h , θ : E k ⟶ H k + 1 ( Ω 0 ; C d s y s ) , R_{h,\theta}: E_k \longrightarrow H^{k+1}(\Omega_0;\mathbb{C}^{d_{\mathrm{sys}}}), R h , θ : E k ⟶ H k + 1 ( Ω 0 ; C d sys ) , L h , θ R h , θ f = f in D ′ ( Ω 1 ) , L_{h,\theta}R_{h,\theta}f=f \quad\text{in } \mathcal{D}'(\Omega_1), L h , θ R h , θ f = f in D ′ ( Ω 1 ) , h − 1 ∥ R h , θ f ∥ H k ( Ω 0 ) + ∥ R h , θ f ∥ H k + 1 ( Ω 0 ) ≤ C k ∥ f ∥ H k ( Ω 0 ) . (88) h^{-1}\lVert R_{h,\theta}f\rVert_{H^k(\Omega_0)}+\lVert R_{h,\theta}f\rVert_{H^{k+1}(\Omega_0)}\leq C_k\lVert f\rVert_{H^k(\Omega_0)}. \tag*{(88)} h − 1 ∥ R h , θ f ∥ H k ( Ω 0 ) + ∥ R h , θ f ∥ H k + 1 ( Ω 0 ) ≤ C k ∥ f ∥ H k ( Ω 0 ) . (88) Here E k E_k E k , Θ \Theta Θ , and L h , θ L_{h,\theta} L h , θ are defined in (3.2) and (3.3). The identity and both estimates hold for every f ∈ E k f \in E_k f ∈ E k . The constants and threshold are uniform over the coefficient class and all θ ∈ Θ \theta\in\Theta θ ∈ Θ , including both phase signs.
Remark. The same correction must satisfy both estimates: the H k H^k H k bound controls terms multiplied by h − 1 h^{-1} h − 1 , whereas the H k + 1 H^{k+1} H k + 1 bound controls the second-order composite operator in the physical lift. The right-inverse identity is asserted only on Ω 1 \Omega_1 Ω 1 ; no support preservation or differentiability of the exact inverse with respect to the direction is asserted. The operator estimate imposes no smallness condition on the source norm.
The proof in Subsection B.7 specifies
N = k + 6 , r p a r = N + 4 = k + 10 , C Σ = ∑ j = 1 5 C j , N=k+6,\qquad r_{\mathrm{par}}=N+4=k+10,\qquad C_{\Sigma}=\sum_{j=1}^{5}C_j, N = k + 6 , r par = N + 4 = k + 10 , C Σ = j = 1 ∑ 5 C j ,
τ 0 = max { 16 , ( 4 C Σ ) 2 } , h ∗ = τ 0 − 1 . (89) \tau_0=\max\{16,(4C_{\Sigma})^2\},\qquad h_*=\tau_0^{-1}. \tag*{(89)} τ 0 = max { 16 , ( 4 C Σ ) 2 } , h ∗ = τ 0 − 1 . (89)
where C 1 , … , C 5 C_1,\ldots,C_5 C 1 , … , C 5 are the uniform constants of the five separately proved operator errors in (B.56). Any enlargement of the fixed geometric lower threshold can be included in the maximum. These constants are fixed by the displayed coefficient bounds and geometry.
Multiplying the right inverse by the residual scale gives the correction used in the CGO ansatz.
Corollary 3.3. For the inverse of Theorem 3.2, define
r h = − h R h , θ F , F ∈ E k , (90) r_h=-hR_{h,\theta}F,\qquad F\in E_k, \tag*{(90)} r h = − h R h , θ F , F ∈ E k , (90) where E k E_k E k is the source space (3.2). Then
∥ r h ∥ H k ( Ω 0 ) ≤ C k h 2 ∥ F ∥ H k , ∥ r h ∥ H k + 1 ( Ω 0 ) ≤ C k h ∥ F ∥ H k . (91) \lVert r_h\rVert_{H^k(\Omega_0)}\leq C_kh^2\lVert F\rVert_{H^k},\qquad
\lVert r_h\rVert_{H^{k+1}(\Omega_0)}\leq C_kh\lVert F\rVert_{H^k}. \tag*{(91)} ∥ r h ∥ H k ( Ω 0 ) ≤ C k h 2 ∥ F ∥ H k , ∥ r h ∥ H k + 1 ( Ω 0 ) ≤ C k h ∥ F ∥ H k . (91) Define the leading transport operator by
T θ = 2 D θ + ∑ j θ j A ~ j . (92) T_\theta=2D_\theta+\sum_j\theta_j\widetilde{A}_j. \tag*{(92)} T θ = 2 D θ + j ∑ θ j A j . (92) If T θ a 0 = 0 T_\theta a_0=0 T θ a 0 = 0 and T θ a 1 = M ~ a 0 T_\theta a_1=\widetilde{\mathcal{M}}a_0 T θ a 1 = M a 0 on a fixed neighborhood, and if F = χ 0 M ~ a 1 ∈ E k F=\chi_0\widetilde{\mathcal{M}}a_1\in E_k F = χ 0 M a 1 ∈ E k with χ 0 = 1 \chi_0=1 χ 0 = 1 on a fixed working subdomain Ω C G O ⋐ int K s r c \Omega_{\mathrm{CGO}}\Subset\operatorname{int}K_{\mathrm{src}} Ω CGO ⋐ int K src , then
M ~ ( e − θ ⋅ x / h ( a 0 + h a 1 + r h ) ) = 0 in Ω C G O . \widetilde{\mathcal{M}}\left(e^{-\theta\cdot x/h}(a_0+ha_1+r_h)\right)=0
\quad\text{in }\Omega_{\mathrm{CGO}}. M ( e − θ ⋅ x / h ( a 0 + h a 1 + r h ) ) = 0 in Ω CGO . The threshold is independent of the size and auxiliary test parameters of F F F .
Proof. Since r h = − h R h , θ F r_h=-h\mathcal{R}_{h,\theta}F r h = − h R h , θ F , the two bounds in (3.8) give
h − 2 ∥ r h ∥ H k ( Ω 0 ) + h − 1 ∥ r h ∥ H k + 1 ( Ω 0 ) ≤ C k ∥ F ∥ H k . h^{-2}\lVert r_h\rVert_{H^k(\Omega_0)}+h^{-1}\lVert r_h\rVert_{H^{k+1}(\Omega_0)}\le C_k\lVert F\rVert_{H^k}. h − 2 ∥ r h ∥ H k ( Ω 0 ) + h − 1 ∥ r h ∥ H k + 1 ( Ω 0 ) ≤ C k ∥ F ∥ H k .
This proves (3.11).
To verify the equation, use the exact conjugation formula (3.3) and the two transport equations to obtain
L h , θ ( a 0 + h a 1 ) = h M ~ a 1 . (93) L_{h,\theta}(a_0+ha_1)=h\widetilde{\mathcal{M}}a_1. \tag*{(93)} L h , θ ( a 0 + h a 1 ) = h M a 1 . (93)
On the working subdomain F = M ~ a 1 F=\widetilde{\mathcal{M}}a_1 F = M a 1 , and L h , θ r h = − h F L_{h,\theta}r_h=-hF L h , θ r h = − h F by (3.8). Therefore
L h , θ ( a 0 + h a 1 + r h ) = h M ~ a 1 − h F = 0 on Ω C G O . L_{h,\theta}(a_0+ha_1+r_h)=h\widetilde{\mathcal{M}}a_1-hF=0\quad\text{on }\Omega_{\mathrm{CGO}}. L h , θ ( a 0 + h a 1 + r h ) = h M a 1 − h F = 0 on Ω CGO .
Conjugating back gives the asserted solution. The threshold h ∗ h_* h ∗ is fixed by Theorem 3.2 on the whole source space E k E_k E k and is independent of the norm or auxiliary parameters of F F F .
The following corollary gives the regularity requirements for the frame construction and the k = 1 k=1 k = 1 inverse in elasticity.
Corollary 3.4. Fix the regularity orders
m ≥ 4 , r f r = 2 m + 8 , s = r f r + 6 = 2 m + 14. (94) m\ge4,\qquad r_{\mathrm{fr}}=2m+8,\qquad s=r_{\mathrm{fr}}+6=2m+14. \tag*{(94)} m ≥ 4 , r fr = 2 m + 8 , s = r fr + 6 = 2 m + 14. (94) Suppose that the fixed-support reduced coefficients and the pulled-back transport coefficients of sizes 2, 3 and 4 are bounded in C s − 2 C^{s-2} C s − 2 on the fixed neighborhoods of (3.1) and (3.4). Then one may choose their finite frame atlas once, with total leaf–transverse–direction derivatives through order r f r + 2 r_{\mathrm{fr}}+2 r fr + 2 , before choosing a source or h h h . For the four-component right inverse, take k = 1 k=1 k = 1 in Theorem 3.2. For any family F ν ∈ E 1 F_\nu\in E_1 F ν ∈ E 1 with ∥ F ν ∥ H 1 ≤ C s r c K ν \lVert F_\nu\rVert_{H^1}\le C_{\mathrm{src}}K_\nu ∥ F ν ∥ H 1 ≤ C src K ν , K ν > 0 K_\nu>0 K ν > 0 , its corrections satisfy simultaneously, for either phase and every 0 < h < h ∗ 0<h<h_* 0 < h < h ∗ ,
∥ r h , ν ∥ H 1 ( Ω 0 ) ≤ C C s r c h 2 K ν , ∥ r h , ν ∥ H 2 ( Ω 0 ) ≤ C C s r c h K ν , r h , ν = − h R h , ± θ F ν . (95) \lVert r_{h,\nu}\rVert_{H^1(\Omega_0)}\le CC_{\mathrm{src}}h^2K_\nu,\qquad
\lVert r_{h,\nu}\rVert_{H^2(\Omega_0)}\le CC_{\mathrm{src}}hK_\nu,\qquad
r_{h,\nu}=-h\mathcal{R}_{h,\pm\theta}F_\nu. \tag*{(95)} ∥ r h , ν ∥ H 1 ( Ω 0 ) ≤ C C src h 2 K ν , ∥ r h , ν ∥ H 2 ( Ω 0 ) ≤ C C src h K ν , r h , ν = − h R h , ± θ F ν . (95) The atlas, C C C and h ∗ h_* h ∗ are independent of ν \nu ν and K ν K_\nu K ν . No differentiability of F ν F_\nu F ν in its index or of the exact inverse in the direction is required.
Proof. First apply Theorem 3.1 with r = r f r + 2 r=r_{\mathrm{fr}}+2 r = r fr + 2 . By (3.14),
r + 2 = r f r + 4 = s − 2 , r + 1 = s − 3. r+2=r_{\mathrm{fr}}+4=s-2,\qquad r+1=s-3. r + 2 = r fr + 4 = s − 2 , r + 1 = s − 3.
The assumed joint C s − 2 C^{s-2} C s − 2 bounds therefore give the C r + 2 C^{r+2} C r + 2 family required by that theorem, with compact coefficient closure in C H s − 3 , α C^{s-3,\alpha}_H C H s − 3 , α as in (3.4). Its finite-product assertion, applied to the systems of sizes 2, 3 and 4, gives one finite atlas with the bounds (3.6) through total order r f r + 2 r_{\mathrm{fr}}+2 r fr + 2 . The atlas is fixed by the coefficient prior and the neighborhoods, before any source or h h h is chosen.
For the four-component right inverse, take k = 1 k=1 k = 1 in Theorem 3.2. The coefficient requirement in (3.1) is satisfied because
k + 14 = 15 ≤ s − 2 = 2 m + 12 , m ≥ 4. k+14=15\le s-2=2m+12,\qquad m\ge4. k + 14 = 15 ≤ s − 2 = 2 m + 12 , m ≥ 4.
Consequently (3.8) holds for both phases with one constant C C C and one threshold h ∗ h_* h ∗ . For each F ν ∈ E 1 F_\nu\in E_1 F ν ∈ E 1 , set r h , ν = − h R h , ± θ F ν r_{h,\nu}=-h\mathcal{R}_{h,\pm\theta}F_\nu r h , ν = − h R h , ± θ F ν as in Corollary 3.3. The operator estimate gives
h − 2 ∥ r h , ν ∥ H 1 ( Ω 0 ) + h − 1 ∥ r h , ν ∥ H 2 ( Ω 0 ) ≤ C ∥ F ν ∥ H 1 ≤ C C s r c K ν , h^{-2}\lVert r_{h,\nu}\rVert_{H^1(\Omega_0)}+h^{-1}\lVert r_{h,\nu}\rVert_{H^2(\Omega_0)}
\le C\lVert F_\nu\rVert_{H^1}\le CC_{\mathrm{src}}K_\nu, h − 2 ∥ r h , ν ∥ H 1 ( Ω 0 ) + h − 1 ∥ r h , ν ∥ H 2 ( Ω 0 ) ≤ C ∥ F ν ∥ H 1 ≤ C C src K ν ,
which implies (3.15). The constant and threshold belong to an operator bound on all of E 1 E_1 E 1 ; hence they are independent of ν \nu ν and K ν K_\nu K ν . The argument applies separately to each source and phase and takes no derivative in the source index or in the direction of the exact inverse. In the physical application, the assumed coefficient bounds follow from Proposition 2.2, specifically (2.22), with the stated loss of two material derivatives.
Elastic CGO solutions and matrix-valued moment estimates
Three physical dual polarizations and a fixed primal frame provide the nine matrix moments. The frames are chosen independently of the polynomial degree, localization width, and CGO parameter.
Fixed domains and the analytic inputs
Use the regularity orders in (94) . The two real materials have a uniform C s C^{s} C s bound and the strong-convexity bounds of Section 2. They satisfy (15) for the input domain D D D and the fixed background of Subsection 7.3. Choose, before all tests,
B ‾ R ∪ B ‾ R b ⋐ Ω C G O ⋐ int K s r c ⋐ Ω 1 ⋐ Ω 0 , K c o e f ⋐ int K s r c . (96) \overline{B}_{R} \cup\overline{B}_{R_b} \Subset\Omega_{\mathrm{CGO}} \Subset\operatorname{int} K_{\mathrm{src}} \Subset\Omega_{1} \Subset\Omega_{0}, \qquad K_{\mathrm{coef}} \Subset\operatorname{int} K_{\mathrm{src}}. \tag*{(96)} B R ∪ B R b ⋐ Ω CGO ⋐ int K src ⋐ Ω 1 ⋐ Ω 0 , K coef ⋐ int K src . (96)
Here K c o e f K_{\mathrm{coef}} K coef contains the conjugated coefficient supports; K s r c K_{\mathrm{src}} K src is a closed ball, and the other working sets can be fixed balls. Fix χ 0 ∈ C ∞ ( int K s r c ) \chi_{0} \in C^{\infty}(\operatorname{int} K_{\mathrm{src}}) χ 0 ∈ C ∞ ( int K src ) with χ 0 = 1 \chi_{0}=1 χ 0 = 1 near Ω ‾ C G O \overline{\Omega}_{\mathrm{CGO}} Ω CGO . All amplitudes and their first corrections are defined on a fixed neighborhood U χ 0 U_{\chi_{0}} U χ 0 of supp χ 0 \operatorname{supp}\chi_{0} supp χ 0 . Use a fixed disk of radius L L L larger than the working-ball radii and slightly larger source disks for every leafwise construction. All transport frames and their equations are defined on a neighborhood of the support of every leaf cutoff used with them. Choose the target, cutoff-support, and frame-existence disks nested with fixed positive gaps; their radii never change with a test.
Uniform frames for the physical systems
For the 2, 3, and 4 dimensional systems in (101) , (62) , and (98) , respectively, a uniform joint C r f r + 4 C^{r_{\mathrm{fr}}+4} C r fr + 4 coefficient bound on the fixed larger cylinders supplies local invertible frames and their inverses with derivatives of total order at most r f r + 2 r_{\mathrm{fr}}+2 r fr + 2 bounded uniformly on finitely many coefficient–direction–transverse neighborhoods. The coefficient label is not differentiated. Apply Theorem 3.1 with frame order r = r f r + 2 r=r_{\mathrm{fr}}+2 r = r fr + 2 and input order r + 2 = r f r + 4 r+2=r_{\mathrm{fr}}+4 r + 2 = r fr + 4 , at (98) , (101) , and (112) . The construction (B.9) also supplies continuity in the material label, with values in the stated total-order frame topology on fixed smaller cylinders. Its coefficient closure and finite-cover argument are proved in Appendix B; no differentiability in the material label is used.
The physical right inverse
For j = 1 , 2 j=1,2 j = 1 , 2 and each normalized null direction ω \omega ω , a linear map R h , ω ( j ) : E 1 → H 2 ( Ω 0 ; C 4 ) \mathcal{R}^{(j)}_{h,\omega}: E_{1} \to H^{2}(\Omega_{0};\mathbb{C}^{4}) R h , ω ( j ) : E 1 → H 2 ( Ω 0 ; C 4 ) , with E 1 E_{1} E 1 defined by (82) at k = 1 k=1 k = 1 and d s y s = 4 d_{\mathrm{sys}}=4 d sys = 4 , satisfies, for 0 < h < h ∗ 0<h<h_{*} 0 < h < h ∗ ,
e ω ⋅ x / h M ~ j e − ω ⋅ x / h R h , ω ( j ) F = F on Ω 1 , e^{\omega\cdot x/h}\widetilde{\mathcal{M}}_{j}e^{-\omega\cdot x/h}\mathcal{R}^{(j)}_{h,\omega}F=F \quad\text{on }\Omega_{1}, e ω ⋅ x / h M j e − ω ⋅ x / h R h , ω ( j ) F = F on Ω 1 ,
h − 1 ∥ R h , ω ( j ) F ∥ H 1 ( Ω 0 ) + ∥ R h , ω ( j ) F ∥ H 2 ( Ω 0 ) ≤ C ∥ F ∥ H 1 ( Ω 0 ) . (97) h^{-1}\left\|\mathcal{R}^{(j)}_{h,\omega}F\right\|_{H^{1}(\Omega_{0})}+\left\|\mathcal{R}^{(j)}_{h,\omega}F\right\|_{H^{2}(\Omega_{0})}\leq C\|F\|_{H^{1}(\Omega_{0})}. \tag*{(97)} h − 1 R h , ω ( j ) F H 1 ( Ω 0 ) + R h , ω ( j ) F H 2 ( Ω 0 ) ≤ C ∥ F ∥ H 1 ( Ω 0 ) . (97)
The map in (97) is the k = 1 k=1 k = 1 , d s y s = 4 d_{\mathrm{sys}}=4 d sys = 4 specialization of (88) . The same map gives both estimates; h ∗ , C h_{*},C h ∗ , C are uniform over the class and the directions. No support preservation or direction differentiability of this exact inverse is assumed. The material-to-coefficient bound gives C s − 2 C^{s-2} C s − 2 coefficients, and s − 2 = 2 m + 12 ≥ 20 > 15 = 1 + 14 s-2=2m+12\geq20>15=1+14 s − 2 = 2 m + 12 ≥ 20 > 15 = 1 + 14 , so the k = 1 k=1 k = 1 specialization of Theorem 3.2 applies. The high-order frame construction and this low-order inverse have different regularity requirements, summarized by Corollary 3.4. In the present notation its source index is ( n , t 0 , ζ ) (n,t_{0},\zeta) ( n , t 0 , ζ ) on the positive side and the fixed physical column label on the negative side.
Finite-order leaf inverses and the physical dual columns
The two principal physical dual columns span only a two-dimensional subspace of the rank-three quotient. A supplementary column supplies the remaining direction.
For compactly supported leaf data on the larger disk use
C θ F ( z , t ⊥ ) = 1 2 π ∫ F ( w , t ⊥ ) z − w d A ( w ) , V θ F = 1 2 F θ C θ ( χ l e a f F θ − 1 F ) . (98) C_{\theta}F(z,t_{\perp})=\frac{1}{2\pi}\int\frac{F(w,t_{\perp})}{z-w}\,dA(w),\qquad\mathcal{V}_{\theta}F=\frac{1}{2}F_{\theta}C_{\theta}(\chi_{\mathrm{leaf}}F_{\theta}^{-1}F). \tag*{(98)} C θ F ( z , t ⊥ ) = 2 π 1 ∫ z − w F ( w , t ⊥ ) d A ( w ) , V θ F = 2 1 F θ C θ ( χ leaf F θ − 1 F ) . (98)
where ( 2 D θ + A θ ) F θ = 0 (2D_{\theta}+A_{\theta})F_{\theta}=0 ( 2 D θ + A θ ) F θ = 0 is the four-component frame from Theorem 3.1 , and χ l e a f = 1 \chi_{\mathrm{leaf}}=1 χ leaf = 1 on the required target disk. Then D θ C θ F = F D_{\theta}C_{\theta}F=F D θ C θ F = F and ( 2 D θ + A θ ) V θ F = F (2D_{\theta}+A_{\theta})\mathcal{V}_{\theta}F=F ( 2 D θ + A θ ) V θ F = F there. The factor 1 / ( 2 π ) 1/(2\pi) 1/ ( 2 π ) is forced by D θ = 2 ∂ z ˉ D_{\theta}=2\partial_{\bar z} D θ = 2 ∂ z ˉ . For data extended by zero from the larger disk, spatial derivatives commute with the convolution; transverse derivatives commute because the integral is at fixed t ⊥ t_{\perp} t ⊥ . In fixed pulled-back coordinates its kernel is independent of direction parameters. The integral of ∣ z − w ∣ − 1 \lvert z-w\rvert^{-1} ∣ z − w ∣ − 1 on the fixed disk is uniformly finite. Leibniz’ rule, including derivatives of the fixed cutoff, therefore bounds the joint C a C^{a} C a norm of this inverse by finitely many joint source and frame norms. There is no claimed gain of a transverse derivative. For the scalar primal lift of (64) one may take
α l i f t = − 1 2 C θ ( χ l e a f ℓ θ ( 2 D θ + A θ ) V θ X ) . (99) \alpha_{\mathrm{lift}}=-\frac{1}{2}C_{\theta}\bigl(\chi_{\mathrm{leaf}}\ell_{\theta}(2D_{\theta}+A_{\theta})\mathcal{V}_{\theta}X\bigr). \tag*{(99)} α lift = − 2 1 C θ ( χ leaf ℓ θ ( 2 D θ + A θ ) V θ X ) . (99)
The cutoff is one on a neighborhood of the target disk. Applying 2 D θ 2D_{\theta} 2 D θ to (99) , and using D θ C θ F = F D_{\theta}C_{\theta}F=F D θ C θ F = F with χ l e a f = 1 \chi_{\mathrm{leaf}}=1 χ leaf = 1 near the target disk, we obtain the scalar equation in (64) . The same fixed-support convolution bounds give the finite total-order parameter estimates. This scalar correction uses the finite-frame theorem for X X X but no elliptic right inverse and no new fourth-dimensional frame.
For a fixed normalized spinor representative, define
a θ = θ ˉ / 2 ∈ C 3 , θ ⋅ a θ = 1 , Φ = ( f 1 f 2 R 1 R 2 ) . (100) a_{\theta}=\bar{\theta}/2\in\mathbb{C}^{3},\qquad\theta\cdot a_{\theta}=1,\qquad\Phi=\begin{pmatrix}f_{1}&f_{2}\\R_{1}&R_{2}\end{pmatrix}. \tag*{(100)} a θ = θ ˉ /2 ∈ C 3 , θ ⋅ a θ = 1 , Φ = ( f 1 R 1 f 2 R 2 ) . (100)
Choose the 2 × 2 2\times2 2 × 2 matrix Φ \Phi Φ so that each column solves
D θ ( f R ) = ( 0 − μ κ / 2 α μ 0 ) ( f R ) , α μ = D θ 2 ( μ − 1 ) = − c 2 θ T F μ θ . (101) D_{\theta}\begin{pmatrix}f\\R\end{pmatrix}
=
\begin{pmatrix}0&-\mu\kappa/2\\\alpha_{\mu}&0\end{pmatrix}
\begin{pmatrix}f\\R\end{pmatrix},
\qquad\alpha_{\mu}=D_{\theta}^{2}(\mu^{-1})=-\frac{c}{2}\theta^{T}F_{\mu}\theta. \tag*{(101)} D θ ( f R ) = ( 0 α μ − μ κ /2 0 ) ( f R ) , α μ = D θ 2 ( μ − 1 ) = − 2 c θ T F μ θ . (101)
Existence and finite-order bounds for Φ ± 1 \Phi^{\pm1} Φ ± 1 use Theorem 3.1 . The determinant has D θ det Φ = 0 D_{\theta}\det\Phi=0 D θ det Φ = 0 ; multiplication of its first column by ( det Φ ) − 1 (\det\Phi)^{-1} ( det Φ ) − 1 gives det Φ = 1 \det\Phi=1 det Φ = 1 without changing the equation or the finite-order bounds. For either column define
H F = 2 D θ ( c − 1 R a θ ) − g μ c − 1 R + F μ θ f , w F = − 1 2 C θ ( χ l e a f H F ) , H_{F}=2D_{\theta}(c^{-1}Ra_{\theta})-g_{\mu}c^{-1}R+F_{\mu}\theta f,\qquad w_{F}=-\frac{1}{2}C_{\theta}(\chi_{\mathrm{leaf}}H_{F}), H F = 2 D θ ( c − 1 R a θ ) − g μ c − 1 R + F μ θ f , w F = − 2 1 C θ ( χ leaf H F ) ,
z 0 F = ( c − 1 R a θ + w F , f ) T , Y F = ( c f θ , − R ) T . (102) z_{0}^{F}=(c^{-1}Ra_{\theta}+w_{F},f)^{T},\qquad\mathcal{Y}_{F}=(cf\theta,-R)^{T}. \tag*{(102)} z 0 F = ( c − 1 R a θ + w F , f ) T , Y F = ( c f θ , − R ) T . (102)
The source satisfies θ ⋅ H F = 0 \theta\cdot H_{F}=0 θ ⋅ H F = 0 : its first two terms contract to 2 c − 1 D θ R 2c^{-1}D_{\theta}R 2 c − 1 D θ R , which is canceled by the last term using (101) . Hence θ ⋅ w F = 0 \theta\cdot w_{F}=0 θ ⋅ w F = 0 and 2 D θ w F = − H F 2D_{\theta}w_{F}=-H_{F} 2 D θ w F = − H F . Both blocks of ( 2 D θ + A θ ) z 0 F = 0 (2D_{\theta}+A_{\theta})z_{0}^{F}=0 ( 2 D θ + A θ ) z 0 F = 0 follow, and
J θ z 0 F = Y F , ( 2 D θ − A θ T ) Y F = ( 2 c θ ( D θ f + μ κ R / 2 ) − 2 ( D θ R − α μ f ) ) = 0. (103) J_{\theta}z_{0}^{F}=\mathcal{Y}_{F},\qquad
(2D_{\theta}-A_{\theta}^{T})\mathcal{Y}_{F}
=
\begin{pmatrix}
2c\theta(D_{\theta}f+\mu\kappa R/2)\\
-2(D_{\theta}R-\alpha_{\mu}f)
\end{pmatrix}
=0. \tag*{(103)} J θ z 0 F = Y F , ( 2 D θ − A θ T ) Y F = ( 2 c θ ( D θ f + μ κ R /2 ) − 2 ( D θ R − α μ f ) ) = 0. (103)
On each compact spinor chart, choose spatially constant vectors satisfying
ξ q ∈ C 2 , q θ = Q ξ q ∈ C 3 , det ( n θ , ξ q ) = − 1 , (104) \xi_{q}\in\mathbb{C}^{2},\qquad q_{\theta}=Q\xi_{q}\in\mathbb{C}^{3},\qquad\det(n_{\theta},\xi_{q})=-1, \tag*{(104)} ξ q ∈ C 2 , q θ = Q ξ q ∈ C 3 , det ( n θ , ξ q ) = − 1 , (104)
where Q , n θ Q,n_{\theta} Q , n θ are defined in (56) . For example, take ξ q = ( 1 / s 0 , 0 ) T \xi_{q}=(1/s_{0},0)^{T} ξ q = ( 1/ s 0 , 0 ) T where s 0 s_{0} s 0 is bounded away from zero, and ξ q = ( 0 , 1 / s 1 ) T \xi_{q}=(0,1/s_{1})^{T} ξ q = ( 0 , 1/ s 1 ) T on the other chart. All resulting normalization denominators have fixed positive lower bounds on the chosen subcharts. The supplementary amplitudes are
z 0 s u p = ( q θ , 0 ) T , J θ z 0 s u p = 0 , M z 0 s u p = ( V μ q θ , − e g μ ⋅ q θ ) T , z_{0}^{\mathrm{sup}}=(q_{\theta},0)^{T},\qquad
J_{\theta}z_{0}^{\mathrm{sup}}=0,\qquad
\mathcal{M}z_{0}^{\mathrm{sup}}=(V_{\mu}q_{\theta},-eg_{\mu}\cdot q_{\theta})^{T}, z 0 sup = ( q θ , 0 ) T , J θ z 0 sup = 0 , M z 0 sup = ( V μ q θ , − e g μ ⋅ q θ ) T ,
z 1 s u p = V θ ( M z 0 s u p ) = ( v s u p , f s u p ) T , R c o l = c θ ⋅ v s u p , ( 2 D θ + A θ ) z 1 s u p = M z 0 s u p . (105) z_{1}^{\mathrm{sup}}=\mathcal{V}_{\theta}(\mathcal{M}z_{0}^{\mathrm{sup}})=(v_{\mathrm{sup}},f_{\mathrm{sup}})^{T},\qquad
R_{\mathrm{col}}=c\theta\cdot v_{\mathrm{sup}},\qquad
(2D_{\theta}+A_{\theta})z_{1}^{\mathrm{sup}}=\mathcal{M}z_{0}^{\mathrm{sup}}. \tag*{(105)} z 1 sup = V θ ( M z 0 sup ) = ( v sup , f sup ) T , R col = c θ ⋅ v sup , ( 2 D θ + A θ ) z 1 sup = M z 0 sup . (105)
Here V θ V_\theta V θ is the fixed leaf inverse in (98) . The scalar equation and the contracted vector equation give separately
D θ f s u p = − μ κ R c o l / 2 − e g μ ⋅ q θ / 2 , D θ R c o l = α μ f s u p + ( c / 2 ) θ T V μ q θ . (106) D_\theta f_{\mathrm{sup}}=-\mu\kappa R_{\mathrm{col}}/2-e\,g_\mu\cdot q_\theta/2,\qquad D_\theta R_{\mathrm{col}}=\alpha_\mu f_{\mathrm{sup}}+(c/2)\theta^{T}V_\mu q_\theta. \tag*{(106)} D θ f sup = − μ κ R col /2 − e g μ ⋅ q θ /2 , D θ R col = α μ f sup + ( c /2 ) θ T V μ q θ . (106)
The first nonzero physical lift defines the ambient dual column:
Y s u p = K z 0 s u p − J θ z 1 s u p = ( q θ − c f s u p θ R c o l + 3 2 c g μ ⋅ q θ ) . (107) \mathcal{Y}_{\mathrm{sup}}=Kz_0^{\mathrm{sup}}-J_\theta z_1^{\mathrm{sup}}=\begin{pmatrix}q_\theta-cf_{\mathrm{sup}}\theta\\ R_{\mathrm{col}}+\frac{3}{2}c\,g_\mu\cdot q_\theta\end{pmatrix}. \tag*{(107)} Y sup = K z 0 sup − J θ z 1 sup = ( q θ − c f sup θ R col + 2 3 c g μ ⋅ q θ ) . (107)
The contraction identity (A.14) is proved in Appendix A. Consequently the two blocks of its dual-transport defect are
[ ( 2 D θ − A θ T ) Y s u p ] 1 : 3 = − 2 c θ [ D θ f s u p + μ κ R c o l / 2 + e g μ ⋅ q θ / 2 ] = 0 , [ ( 2 D θ − A θ T ) Y s u p ] 4 = 2 [ D θ R c o l − α μ f s u p − ( c / 2 ) θ T V μ q θ ] = 0. (108) \begin{aligned}
[(2D_\theta-A_\theta^{T})\mathcal{Y}_{\mathrm{sup}}]_{1:3}&=-2c\theta[D_\theta f_{\mathrm{sup}}+\mu\kappa R_{\mathrm{col}}/2+e\,g_\mu\cdot q_\theta/2]=0,\\
[(2D_\theta-A_\theta^{T})\mathcal{Y}_{\mathrm{sup}}]_4&=2[D_\theta R_{\mathrm{col}}-\alpha_\mu f_{\mathrm{sup}}-(c/2)\theta^{T}V_\mu q_\theta]=0. \tag*{(108)}
\end{aligned} [( 2 D θ − A θ T ) Y sup ] 1 : 3 [( 2 D θ − A θ T ) Y sup ] 4 = − 2 c θ [ D θ f sup + μ κ R col /2 + e g μ ⋅ q θ /2 ] = 0 , = 2 [ D θ R col − α μ f sup − ( c /2 ) θ T V μ q θ ] = 0. (108)
Define the 4 × 3 4\times3 4 × 3 ambient dual matrix and its quotient frame by
Y = ( Y F 1 , Y F 2 , Y s u p ) = U θ Y . (109) \mathcal{Y}=(\mathcal{Y}^{1}_{F},\mathcal{Y}^{2}_{F},\mathcal{Y}_{\mathrm{sup}})=U_\theta Y. \tag*{(109)} Y = ( Y F 1 , Y F 2 , Y sup ) = U θ Y . (109)
Here the columns are defined in (102) and (107) , and U θ U_\theta U θ in (56) . The 3 × 3 3\times3 3 × 3 quotient frame is
Y = ( − 2 c f 1 n θ − 2 c f 2 n θ ξ q + 2 c f s u p n θ − R 1 − R 2 R c o l + 3 2 c g μ ⋅ q θ ) , det Y = − 2 c det Φ det ( n θ , ξ q ) = 2 c . (110) Y=\begin{pmatrix}
-2cf_{1}n_\theta& -2cf_{2}n_\theta& \xi_q+2cf_{\mathrm{sup}}n_\theta\\
-R_{1} & -R_{2} & R_{\mathrm{col}}+\frac{3}{2}c\,g_\mu\cdot q_\theta
\end{pmatrix},\qquad\det Y=-2c\det\Phi\det(n_\theta,\xi_q)=2c. \tag*{(110)} Y = ( − 2 c f 1 n θ − R 1 − 2 c f 2 n θ − R 2 ξ q + 2 c f sup n θ R col + 2 3 c g μ ⋅ q θ ) , det Y = − 2 c det Φ det ( n θ , ξ q ) = 2 c . (110)
The top block has two rows. Expansion of the determinant in the basis ( n θ , ξ q ) (n_\theta,\xi_q) ( n θ , ξ q ) proves the formula; the third column’s multiple of n θ n_\theta n θ and its bottom entry do not affect it. Thus the ambient matrix is 4 × 3 4\times3 4 × 3 , whereas Y Y Y and its determinant are 3 × 3 3\times3 3 × 3 .
Transverse localization
Choose χ ( t ⊥ ) \chi(t_\perp) χ ( t ⊥ ) supported in the frame chart and equal to one on an open interval containing every permitted test support on its smaller subchart. For principal columns use a 0 F , i = χ z 0 F , i a^{F,i}_0=\chi z^{F,i}_0 a 0 F , i = χ z 0 F , i and a 1 F , i = V θ ( M a 0 F , i ) a^{F,i}_1=V_\theta(\mathcal{M}a^{F,i}_0) a 1 F , i = V θ ( M a 0 F , i ) . For the supplementary column,
a 0 s u p = χ z 0 s u p , a 1 s u p = χ z 1 s u p + w χ , w χ = V θ ( [ M , χ ] z 0 s u p ) , a^{\mathrm{sup}}_0=\chi z^{\mathrm{sup}}_0,\qquad a^{\mathrm{sup}}_1=\chi z^{\mathrm{sup}}_1+w_\chi,\qquad w_\chi=V_\theta([\mathcal{M},\chi]z^{\mathrm{sup}}_0), a 0 sup = χ z 0 sup , a 1 sup = χ z 1 sup + w χ , w χ = V θ ([ M , χ ] z 0 sup ) ,
[ M , χ ] z 0 s u p = ( χ ′ ′ q θ − χ ′ g μ ( ν l e a f ⋅ q θ ) d χ ′ ( ν l e a f ⋅ q θ ) ) . (111) [\mathcal{M},\chi]z^{\mathrm{sup}}_0=
\begin{pmatrix}
\chi''q_\theta-\chi'g_\mu(\nu_{\mathrm{leaf}}\cdot q_\theta)\\
d\chi'(\nu_{\mathrm{leaf}}\cdot q_\theta)
\end{pmatrix}. \tag*{(111)} [ M , χ ] z 0 sup = ( χ ′′ q θ − χ ′ g μ ( ν leaf ⋅ q θ ) d χ ′ ( ν leaf ⋅ q θ ) ) . (111)
Here ν l e a f \nu_{\mathrm{leaf}} ν leaf is defined in (54) , and V θ V_\theta V θ in (98) . Because this inverse integrates on each leaf separately, w χ w_\chi w χ and its available derivatives are zero on the open interval where χ = 1 \chi=1 χ = 1 . Also J θ a 0 s u p = 0 J_\theta a^{\mathrm{sup}}_0=0 J θ a 0 sup = 0 and its scalar component is zero throughout the working domain, not merely on the test support. The equations ( 2 D θ + A θ ) a 1 ℓ = M a 0 ℓ (2D_\theta+A_\theta)a^{\ell}_1=\mathcal{M}a^{\ell}_0 ( 2 D θ + A θ ) a 1 ℓ = M a 0 ℓ hold on all of U χ 0 U_{\chi_0} U χ 0 . Extension by zero transversely is legitimate because the cutoff support is compact in its chart. None of these assertions applies to the elliptic remainder, whose transverse support is unrestricted.
The constructions of the primal and dual quotient frames, their lifts, and the localized supplementary amplitudes give the following uniform estimates.
Proposition 4.1. For each fixed 0 < R 0 < R 0<R_0<R 0 < R 0 < R , the frames F θ F_\theta F θ , Φ \Phi Φ , X j X_j X j , Y j Y_j Y j , Z j Z_j Z j and the supplementary amplitudes constructed in (98) , (100) , (62) , (110) , (64) , and (105) –(111) can be chosen, together with their leaf and transverse cutoffs, on a fixed finite product cover of the compact material-pair parameter space and C P 1 × [ − R 0 , R 0 ] \mathbb{CP}^{1}\times[-R_0,R_0] CP 1 × [ − R 0 , R 0 ] . Each smaller chart is compactly contained in its frame-existence chart. These choices are independent of h h h , the polynomial degree, and the transverse localization center and width.
On the corresponding fixed cylinders, containing the entire closed observation disks and all residual supports,
max ∣ β ∣ + ∣ γ ∣ ≤ r f r ∥ ∂ p β ∂ x γ ( X j , X j − 1 , Y j , Y j − 1 , Z j , z 1 s u p , z 0 s u p , w χ ) ∥ C 0 ≤ C m . (112) \max_{|\beta|+|\gamma|\le r_{\mathrm{fr}}}\left\|\partial_p^\beta\partial_x^\gamma\left(X_j,X_j^{-1},Y_j,Y_j^{-1},Z_j,z_1^{\mathrm{sup}},z_0^{\mathrm{sup}},w_\chi\right)\right\|_{C^0}\le C_m. \tag*{(112)} ∣ β ∣ + ∣ γ ∣ ≤ r fr max ∂ p β ∂ x γ ( X j , X j − 1 , Y j , Y j − 1 , Z j , z 1 sup , z 0 sup , w χ ) C 0 ≤ C m . (112)
Here p p p denotes local direction parameters, and transverse differentiation is counted among the spatial derivatives. Equivalently, in the fixed pulled-back coordinates, the estimate controls mixed derivatives of total leaf–transverse–direction order at most r f r r_{\mathrm{fr}} r fr . The displayed fields depend continuously on the material label in this topology. The cover and C m C_m C m depend only on the fixed a priori class and geometry.
For the two materials, the matrix W = Y 2 T X 1 W=Y_2^{T}X_1 W = Y 2 T X 1 is invertible on the full closed observation disks, with W − 1 = X 1 − 1 Y 2 − T W^{-1}=X_1^{-1}Y_2^{-T} W − 1 = X 1 − 1 Y 2 − T , and satisfies
2 ∂ z ‾ W = − Y 2 T ( P 1 − P 2 ) X 1 . (113) 2\partial_{\overline{z}}W=-Y_2^{T}(P_1-P_2)X_1. \tag*{(113)} 2 ∂ z W = − Y 2 T ( P 1 − P 2 ) X 1 . (113)
In particular, W W W is holomorphic on every common-exterior leaf annulus.
Proof. Put r = r f r + 2 r=r_{\mathrm{fr}}+2 r = r fr + 2 . Proposition 2.2 and (39) give the derivative count
C r f r + 6 materials ⟼ C r f r + 4 = C r + 2 transport coefficients . C^{r_{\mathrm{fr}}+6}\text{ materials}\longmapsto C^{r_{\mathrm{fr}}+4}=C^{r+2}\text{ transport coefficients}. C r fr + 6 materials ⟼ C r fr + 4 = C r + 2 transport coefficients .
Theorem 3.1 and Corollary 3.4 therefore supply a common finite atlas with full-disk frame and inverse bounds through total order r f r + 2 r_{\mathrm{fr}}+2 r fr + 2 .
The amplitudes in (64) , (102) , (105) , and (111) are obtained by products, differentiation, and the fixed-support leaf inverses (98) . In the fixed pulled-back coordinates, the Cauchy kernel is parameter independent and has uniformly bounded L 1 L^1 L 1 norm on the integration disks. Differentiation and Leibniz’ rule thus bound each integral by the corresponding source and frame norms. The explicit formulas use at most two of the reserved frame derivatives, and their material coefficients use at most two material derivatives. All the displayed fields are therefore bounded through total order r f r r_{\mathrm{fr}} r fr . For the inverse of the physical quotient frame, use (110) :
Y − 1 = adj Y det Y , ∂ j Y − 1 = − Y − 1 ( ∂ j Y ) Y − 1 . Y^{-1}=\frac{\operatorname{adj}Y}{\det Y},\qquad\partial_jY^{-1}=-Y^{-1}(\partial_jY)Y^{-1}. Y − 1 = det Y adj Y , ∂ j Y − 1 = − Y − 1 ( ∂ j Y ) Y − 1 .
The lower bounds in (32) first bound c − 1 c^{-1} c − 1 and Y − 1 Y^{-1} Y − 1 . Differentiating Y Y − 1 = I YY^{-1}=I Y Y − 1 = I then bounds each higher spatial or parameter derivative in terms of already controlled derivatives. This proves (112) .
To check the material-label assertion, a bounded C s C^s C s material family has compact closure in C H s − 1 , α C^{s-1,\alpha}_H C H s − 1 , α on the fixed smaller neighborhoods. The coefficient formulas with two derivatives define continuous maps into
C H s − 3 , α = C H r f r + 3 , α = C H r + 1 , α . C^{s-3,\alpha}_H=C^{r_{\mathrm{fr}}+3,\alpha}_H=C^{r+1,\alpha}_H. C H s − 3 , α = C H r fr + 3 , α = C H r + 1 , α .
This is the topology of (84) . The coefficient-label continuity in Theorem 3.1 therefore passes through the fixed integrals and algebraic operations above. The normalization denominators stay uniformly separated from zero. The theorem’s finite-product refinement gives the asserted common cover with fixed margins for both materials and all systems.
Finally, (62) and the product rule yield
2 ∂ z ‾ ( Y 2 T X 1 ) = ( P 2 T Y 2 ) T X 1 − Y 2 T P 1 X 1 = − Y 2 T ( P 1 − P 2 ) X 1 . 2\partial_{\overline{z}}(Y_2^TX_1)=(P_2^TY_2)^TX_1-Y_2^TP_1X_1=-Y_2^T(P_1-P_2)X_1. 2 ∂ z ( Y 2 T X 1 ) = ( P 2 T Y 2 ) T X 1 − Y 2 T P 1 X 1 = − Y 2 T ( P 1 − P 2 ) X 1 .
This is (113) . Wherever the materials agree, P 1 − P 2 = 0 P_1-P_2=0 P 1 − P 2 = 0 , so W W W is holomorphic there. □
Remark. The supplementary physical polarization makes the dual quotient frame invertible, as shown by the determinant identity (110) . Since the frames are fixed independently of the moment tests, all nine matrix moments in Theorem 4.2 are obtained, on each chart, from the same frame product W W W , with uniform bounds throughout the subsequent Fourier and transport arguments.
The construction uses local normalized representatives of the projective null directions and does not assert a global trivialization over C P 1 \mathbb{CP}^1 CP 1 . Only continuous dependence on the material label is required.
Exact physical solutions and Sobolev remainder estimates
The finite-order amplitudes are corrected to exact reduced solutions and then transformed into physical elastic solutions. Both Sobolev remainder estimates enter the physical trace and dual-lift bounds.
For a positive-phase leading amplitude a 0 + a_0^+ a 0 + , use
a 1 + = − V 1 , θ ( M 1 a 0 + ) , a 1 − , ℓ = + V 2 , θ ( M 2 a 0 − , ℓ ) , ℓ = F 1 , F 2 , s u p . (114) a_1^+ = -\mathcal{V}_{1,\theta}(\mathcal{M}_1a_0^+), \qquad a_1^{-,\ell} = +\mathcal{V}_{2,\theta}(\mathcal{M}_2a_0^{-,\ell}), \qquad\ell= F1,F2,\mathrm{sup}. \tag*{(114)} a 1 + = − V 1 , θ ( M 1 a 0 + ) , a 1 − , ℓ = + V 2 , θ ( M 2 a 0 − , ℓ ) , ℓ = F 1 , F 2 , sup . (114)
For the supplementary negative column take exactly the localized a 1 s u p a_1^{\mathrm{sup}} a 1 sup in (111) , which is the same linear construction by expansion of M ( χ 0 z s u p ) \mathcal{M}(\chi_0 z^{\mathrm{sup}}) M ( χ 0 z sup ) . Assume ∥ a 0 + ∥ C 12 ≤ C K \|a_0^+\|_{C^{12}} \le CK ∥ a 0 + ∥ C 12 ≤ C K ; the negative amplitudes are fixed and have the corresponding bound with K = 1 K=1 K = 1 . The fixed leaf inverse and Leibniz’ rule give
∥ a 1 + ∥ H 3 ( U χ 0 ) + ∥ χ 0 R ∗ − 1 M 1 a 1 + ∥ H 1 ( Ω 0 ) ≤ C K , ∥ a 1 − , ℓ ∥ H 3 + ∥ χ 0 R ∗ − 1 M 2 a 1 − , ℓ ∥ H 1 ≤ C . (115) \|a_1^+\|_{H^3(U_{\chi_0})}+\|\chi_0R_*^{-1}\mathcal{M}_1a_1^+\|_{H^1(\Omega_0)}\le CK,\qquad\|a_1^{-,\ell}\|_{H^3}+\|\chi_0R_*^{-1}\mathcal{M}_2a_1^{-,\ell}\|_{H^1}\le C. \tag*{(115)} ∥ a 1 + ∥ H 3 ( U χ 0 ) + ∥ χ 0 R ∗ − 1 M 1 a 1 + ∥ H 1 ( Ω 0 ) ≤ C K , ∥ a 1 − , ℓ ∥ H 3 + ∥ χ 0 R ∗ − 1 M 2 a 1 − , ℓ ∥ H 1 ≤ C . (115)
For clarity, derivatives of the source M a 0 \mathcal{M}a_0 M a 0 through order four cost at most six derivatives of a 0 a_0 a 0 , well within the C 12 C^{12} C 12 bound. The H 1 H^1 H 1 residual uses at most three derivatives of a 1 a_1 a 1 .
Write
a ~ = R ∗ − 1 a , (116) \widetilde{a}=R_*^{-1}a, \tag*{(116)} a = R ∗ − 1 a , (116)
where R ∗ R_* R ∗ is defined in (48) . Apply the source operators of Theorem 3.2 at k = 1 k=1 k = 1 , and define
r 1 , h + = − R ∗ R h , − θ ( 1 ) ( h χ 0 M 1 a 1 + ~ ) , r_{1,h}^+=-R_*\mathcal{R}_{h,-\theta}^{(1)}(h\chi_0\widetilde{\mathcal{M}_1a_1^+}), r 1 , h + = − R ∗ R h , − θ ( 1 ) ( h χ 0 M 1 a 1 + ) ,
{#eq:4.22}
r_{2,h}^{-,\ell}=-R_*\mathcal{R}_{h,\theta}^{(2)}(h\chi_0\widetilde{\mathcal{M}_2a_1^{-,\ell}}).
E a c h r e m a i n d e r i n x r e f ‘ ( 4.22 ) ‘ s a t i s f i e s b o t h S o b o l e v e s t i m a t e s : Each remainder in {xref}`(4.22)` satisfies both Sobolev estimates: E a c h r e main d er in x r e f ‘ ( 4.22 ) ‘ s a t i s f i es b o t h S o b o l e v es t ima t es :
|r_{1,h}^+|_{H^1(\Omega_{\mathrm{CGO}})}\le Ch^2K,\qquad |r_{1,h}^+|_{H^2(\Omega_{\mathrm{CGO}})}\le ChK,
|r_{2,h}^{-,\ell}|_{H^1(\Omega_{\mathrm{CGO}})}\le Ch^2,\qquad |r_{2,h}^{-,\ell}|_{H^2(\Omega_{\mathrm{CGO}})}\le Ch.
{#eq:4.23}
Here the remainders are defined in {xref}`(4.22)`. The reduced fields
z_{1,h}^+=e^{\varphi/h}(a_0^+ + ha_1^+ + r_{1,h}^+),\qquad z_{2,h}^{-,\ell}=e^{-\varphi/h}(a_0^{-,\ell}+ha_1^{-,\ell}+r_{2,h}^{-,\ell})
{#eq:4.24}
solve the corresponding reduced equations on Ω C G O \Omega_{\mathrm{CGO}} Ω CGO by (43) : after the first transport cancellation, the residual is h M j a 1 h\mathcal{M}_ja_1 h M j a 1 , and the remainder cancels it where χ 0 = 1 \chi_0=1 χ 0 = 1 . It is not necessary that the equation hold on all of Ω 0 \Omega_0 Ω 0 .
The genuine physical displacements and their canonical lifts are
u 1 , h = T 1 z 1 , h + , u 2 , h F i = − h T 2 z 2 , h − , F i , u 2 , h s u p = T 2 z 2 , h − , s u p , y 2 , h ℓ = S 2 T u 2 , h ℓ . (117) u_{1,h}=T_1z_{1,h}^+,\qquad u_{2,h}^{F_i}=-hT_2z_{2,h}^{-,F_i},\qquad u_{2,h}^{\mathrm{sup}}=T_2z_{2,h}^{-,\mathrm{sup}},\qquad y_{2,h}^{\ell}=S_2^Tu_{2,h}^{\ell}. \tag*{(117)} u 1 , h = T 1 z 1 , h + , u 2 , h F i = − h T 2 z 2 , h − , F i , u 2 , h sup = T 2 z 2 , h − , sup , y 2 , h ℓ = S 2 T u 2 , h ℓ . (117)
Each u u u solves the homogeneous physical equation by the exact factorization, not by prescribing reduced boundary observations. With the zeroth-order 3 × 4 3\times4 3 × 4 phase symbol
t μ , θ ( v , f ) = μ − 1 f θ , e − σ p h φ / h T μ e σ p h φ / h = T μ + σ p h h − 1 t μ , θ , (118) t_{\mu,\theta}(v,f)=\mu^{-1}f\theta,\qquad e^{-\sigma_{\mathrm{ph}}\varphi/h}T_\mu e^{\sigma_{\mathrm{ph}}\varphi/h}=T_\mu+\sigma_{\mathrm{ph}}h^{-1}t_{\mu,\theta}, \tag*{(118)} t μ , θ ( v , f ) = μ − 1 f θ , e − σ ph φ / h T μ e σ ph φ / h = T μ + σ ph h − 1 t μ , θ , (118)
the supplementary displacement has the exact expansion
e φ / h u 2 , h s u p = T 2 a 0 s u p − t 2 , θ a 1 s u p + h T 2 a 1 s u p + T 2 r h s u p − h − 1 t 2 , θ r h s u p . (119) e^{\varphi/h}u_{2,h}^{\mathrm{sup}}=T_2a_0^{\mathrm{sup}}-t_{2,\theta}a_1^{\mathrm{sup}}+hT_2a_1^{\mathrm{sup}}+T_2r_h^{\mathrm{sup}}-h^{-1}t_{2,\theta}r_h^{\mathrm{sup}}. \tag*{(119)} e φ / h u 2 , h sup = T 2 a 0 sup − t 2 , θ a 1 sup + h T 2 a 1 sup + T 2 r h sup − h − 1 t 2 , θ r h sup . (119)
The absent h − 1 t a 0 s u p h^{-1}t_{a_0}^{\mathrm{sup}} h − 1 t a 0 sup term is zero on the whole working domain. Since T : H 2 → H 1 T:H^2\to H^1 T : H 2 → H 1 and t : H 1 → H 1 t:H^1\to H^1 t : H 1 → H 1 , the last three terms are O ( h ) O(h) O ( h ) in phase-removed H 1 H^1 H 1 by (4.23). For a principal column the corresponding expansion is t ( a 0 + h a 1 + r ) − h T ( a 0 + h a 1 + r ) t(a_0+ha_1+r)-hT(a_0+ha_1+r) t ( a 0 + h a 1 + r ) − h T ( a 0 + h a 1 + r ) , bounded in H 1 H^1 H 1 . For the positive column it is ( T + h − 1 t ) ( a 0 + h a 1 + r ) (T+h^{-1}t)(a_0+ha_1+r) ( T + h − 1 t ) ( a 0 + h a 1 + r ) , with H 1 H^1 H 1 bound C h − 1 K Ch^{-1}K C h − 1 K . By the product rule for the exponential factors, the amplitude bounds in (126) , and the remainder estimates in (4.23), we obtain the physical trace bounds
∥ u 1 , h ∥ H 1 ( B R ) ≤ C h − 2 e R / h K , ∥ u 2 , h F , i ∥ H 1 ( B R ) + ∥ u 2 , h s u p ∥ H 1 ( B R ) ≤ C h − 1 e R / h . (120) \|u_{1,h}\|_{H^1(B_R)} \le Ch^{-2}e^{R/h}K,\qquad\|u_{2,h}^{F,i}\|_{H^1(B_R)}+\|u_{2,h}^{\mathrm{sup}}\|_{H^1(B_R)}\le Ch^{-1}e^{R/h}. \tag*{(120)} ∥ u 1 , h ∥ H 1 ( B R ) ≤ C h − 2 e R / h K , ∥ u 2 , h F , i ∥ H 1 ( B R ) + ∥ u 2 , h sup ∥ H 1 ( B R ) ≤ C h − 1 e R / h . (120)
With K \mathcal{K} K from (40) for material two, the canonical lifts (117) satisfy
e φ / h y 2 , h F , i = J θ a 0 + h ( J θ a 1 − K a 0 ) + J θ r − h 2 K a 1 − h K r , e^{\varphi/h}y_{2,h}^{F,i}=J_\theta a_0+h(J_\theta a_1-\mathcal{K}a_0)+J_\theta r-h^2\mathcal{K}a_1-h\mathcal{K}r, e φ / h y 2 , h F , i = J θ a 0 + h ( J θ a 1 − K a 0 ) + J θ r − h 2 K a 1 − h K r ,
e φ / h y 2 , h s u p = K a 0 s u p − J θ a 1 s u p + h K a 1 s u p − h − 1 J θ r s u p + K r s u p . (121) e^{\varphi/h}y_{2,h}^{\mathrm{sup}}=\mathcal{K}a_0^{\mathrm{sup}}-J_\theta a_1^{\mathrm{sup}}+h\mathcal{K}a_1^{\mathrm{sup}}-h^{-1}J_\theta r^{\mathrm{sup}}+\mathcal{K}r^{\mathrm{sup}}. \tag*{(121)} e φ / h y 2 , h sup = K a 0 sup − J θ a 1 sup + h K a 1 sup − h − 1 J θ r sup + K r sup . (121)
The mapping properties (40) and (44) , together with (4.23), give
∥ h ( J θ a 1 − K a 0 ) ∥ 2 ≤ C h , ∥ J θ r ∥ 2 + ∥ h 2 K a 1 ∥ 2 + ∥ h K r ∥ 2 ≤ C h 2 , \|h(J_\theta a_1-\mathcal{K}a_0)\|_2\le Ch,\qquad\|J_\theta r\|_2+\|h^2\mathcal{K}a_1\|_2+\|h\mathcal{K}r\|_2\le Ch^2, ∥ h ( J θ a 1 − K a 0 ) ∥ 2 ≤ C h , ∥ J θ r ∥ 2 + ∥ h 2 K a 1 ∥ 2 + ∥ h K r ∥ 2 ≤ C h 2 ,
∥ h K a 1 s u p ∥ 2 ≤ C h , ∥ h − 1 J θ r s u p ∥ 2 + ∥ K r s u p ∥ 2 ≤ C h . (122) \|h\mathcal{K}a_1^{\mathrm{sup}}\|_2\le Ch,\qquad\|h^{-1}J_\theta r^{\mathrm{sup}}\|_2+\|\mathcal{K}r^{\mathrm{sup}}\|_2\le Ch. \tag*{(122)} ∥ h K a 1 sup ∥ 2 ≤ C h , ∥ h − 1 J θ r sup ∥ 2 + ∥ K r sup ∥ 2 ≤ C h . (122)
The first row applies to either principal column and the second to the supplementary column; all norms are in L 2 ( B R ) L^2(B_R) L 2 ( B R ) . Thus every term in (121) is controlled by the same remainder. The leading matrix on the whole integration ball is
Y χ = ( χ Y F 1 , χ Y F 2 , Y χ , s u p ) , \mathcal{Y}_\chi=(\chi\mathcal{Y}_F^1,\chi\mathcal{Y}_F^2,\mathcal{Y}_{\chi,\mathrm{sup}}), Y χ = ( χ Y F 1 , χ Y F 2 , Y χ , sup ) ,
Y χ , s u p = χ Y s u p − J θ w χ − ( 0 C 3 c χ ′ ( ν l e a f ⋅ q θ ) / ( 1 − κ ) ) . (123) \mathcal{Y}_{\chi,\mathrm{sup}}=\chi\mathcal{Y}_{\mathrm{sup}}-J_\theta w_\chi-\begin{pmatrix}0_{\mathbb{C}^3}\\ c\chi'(\nu_{\mathrm{leaf}}\cdot q_\theta)/(1-\kappa)\end{pmatrix}. \tag*{(123)} Y χ , sup = χ Y sup − J θ w χ − ( 0 C 3 c χ ′ ( ν leaf ⋅ q θ ) / ( 1 − κ ) ) . (123)
The last term is the commutator of K \mathcal{K} K with the transverse cutoff on ( q θ , 0 ) T (q_\theta,0)^T ( q θ , 0 ) T ; its calculation is given in (A.18). All terms are bounded independently of the positive test, and both correction terms vanish wherever χ = 1 \chi=1 χ = 1 . In particular,
e φ / h y 2 , h = Y χ + q h , ∥ q h ∥ L 2 ( B R ) ≤ C h , Y χ = U θ Y 2 where χ = 1. (124) e^{\varphi/h}y_{2,h}=\mathcal{Y}_\chi+q_h,\qquad\|q_h\|_{L^2(B_R)}\le Ch,\qquad\mathcal{Y}_\chi=U_\theta\mathcal{Y}_2\ \text{where }\chi=1. \tag*{(124)} e φ / h y 2 , h = Y χ + q h , ∥ q h ∥ L 2 ( B R ) ≤ C h , Y χ = U θ Y 2 where χ = 1. (124)
The factor h − 1 h^{-1} h − 1 is controlled by the H 1 H^1 H 1 estimate O ( h 2 ) O(h^2) O ( h 2 ) ; the other second-order term requires the simultaneous H 2 H^2 H 2 estimate O ( h ) O(h) O ( h ) . No uniform H 5 H^5 H 5 remainder and no pointwise rank of the exact canonical matrix is used. At fixed h h h , interior Laplacian regularity upgrades the exact reduced fields from H 2 H^2 H 2 to H 4 H^4 H 4 near B ‾ R \overline{B}_R B R , so Lemma 2.3 applies. The upgrade in Appendix A is only an identity-justification step; its constants may depend on h h h .
All nine moments in fixed quotient frames
The matrix moments are estimated using δ D \delta_D δ D in (16) and the previously fixed frames. The polynomial degree and transverse localization parameters may vary after the frames have been chosen.
On a fixed smaller transverse subchart let t 0 t_0 t 0 be any permitted center and choose 0 < ς ≤ ς ∗ 0<\varsigma\le\varsigma_* 0 < ς ≤ ς ∗ such that the support below lies in the open interval χ = 1 \chi=1 χ = 1 . Fix ψ ⊥ ∈ C c ∞ ( ( − 1 , 1 ) ) \psi_\perp\in C^\infty_c((-1,1)) ψ ⊥ ∈ C c ∞ (( − 1 , 1 )) , ψ ⊥ ≥ 0 \psi_\perp\ge0 ψ ⊥ ≥ 0 , ∫ ψ ⊥ = 1 \int\psi_\perp=1 ∫ ψ ⊥ = 1 , and put
r ( t ) = R 2 − t 2 , b ς , t 0 ( t ) = ς − 1 ψ ⊥ ( ( t − t 0 ) / ς ) , g n ( z , t ) = ( z / r ( t ) ) n , n ∈ N 0 , r(t)=\sqrt{R^2-t^2},\qquad b_{\varsigma,t_0}(t)=\varsigma^{-1}\psi_\perp((t-t_0)/\varsigma),\qquad g_n(z,t)=(z/r(t))^n,\quad n\in\mathbb{N}_0, r ( t ) = R 2 − t 2 , b ς , t 0 ( t ) = ς − 1 ψ ⊥ (( t − t 0 ) / ς ) , g n ( z , t ) = ( z / r ( t ) ) n , n ∈ N 0 ,
a 0 + = b ς , t 0 g n Z 1 , K n , ς = A n ( n + 1 ) 12 ς − 13 . (125) a_0^+=b_{\varsigma,t_0}g_n Z_1,\qquad K_{n,\varsigma}=A^n(n+1)^{12}\varsigma^{-13}. \tag*{(125)} a 0 + = b ς , t 0 g n Z 1 , K n , ς = A n ( n + 1 ) 12 ς − 13 . (125)
The admissible transverse range is compact in ( − R , R ) (-R,R) ( − R , R ) , so r r r has a positive lower bound there. Choose A > 1 A>1 A > 1 larger than the ratio of the large source-disk radius to this lower bound, increasing it for fixed geometric factors. Define r r r on a fixed neighborhood of the compact test supports, and extend b ς , t 0 g n b_{\varsigma,t_0}g_n b ς , t 0 g n by zero transversely. Because D θ z = D θ t = 0 D_\theta z=D_\theta t=0 D θ z = D θ t = 0 , the leading amplitude solves transport. Leibniz’ rule through order 12 gives, on the entire large working cylinder,
∥ a 0 + ∥ C 12 ≤ C K n , ς . (126) \|a_0^+\|_{C^{12}}\le CK_{n,\varsigma}. \tag*{(126)} ∥ a 0 + ∥ C 12 ≤ C K n , ς . (126)
Each differentiation of a power costs at most a factor n + 1 n+1 n + 1 ; the largest derivative of b ς , t 0 b_{\varsigma,t_0} b ς , t 0 is of order ς − 13 \varsigma^{-13} ς − 13 . Since r f r ≥ 16 r_{\mathrm{fr}} \ge16 r fr ≥ 16 , the one chosen high-order frame supplies all these bounds. The weighted matrix integrals in (127) are obtained from elastic solutions through the DN pairing (19) .
Theorem 4.2. Under the fixed C 2 m + 14 C^{2m+14} C 2 m + 14 prior and geometry of this section, select the r f r = 2 m + 8 r_{\mathrm{fr}}=2m+8 r fr = 2 m + 8 local family of Proposition 4.1 for each pair before any tests. Assume the common-collar condition of Lemma 2.3. The measurement domain is either D = B R D=B_R D = B R , or a fixed smooth domain with D ‾ ⋐ B R \overline{D}\Subset B_R D ⋐ B R such that the materials agree on B R ∖ D ‾ B_R\setminus\overline{D} B R ∖ D . There exist constants A > 1 A>1 A > 1 , C , C 0 > 0 C,C_0>0 C , C 0 > 0 and 0 < h ∗ ≤ 1 0<h_* \le1 0 < h ∗ ≤ 1 , uniform over all pairs and active charts, such that for every n ≥ 0 n\ge0 n ≥ 0 , every permitted center and width 0 < ς ≤ ς ∗ 0<\varsigma\le\varsigma_* 0 < ς ≤ ς ∗ , and every 0 < h < h ∗ 0<h<h_* 0 < h < h ∗ ,
∥ ∫ B R b ς , t 0 g n Y 2 T ( P 1 − P 2 ) X 1 d x ∥ ≤ C K n , ς ( h + δ D h − 2 e C 0 / h ) . (127) \left\|\int_{B_R} b_{\varsigma,t_0}g_nY_2^{T}(P_1-P_2)X_1\,dx\right\|\le CK_{n,\varsigma}\left(h+\delta_Dh^{-2}e^{C_0/h}\right). \tag*{(127)} ∫ B R b ς , t 0 g n Y 2 T ( P 1 − P 2 ) X 1 d x ≤ C K n , ς ( h + δ D h − 2 e C 0 / h ) . (127) Here b ς , t 0 b_{\varsigma,t_0} b ς , t 0 , g n g_n g n , K n , ς K_{n,\varsigma} K n , ς are defined in (125) , P j P_j P j in (61) , and X 1 , Y 2 X_1,Y_2 X 1 , Y 2 are the frames of Proposition 4.1. The discrepancy δ D \delta_D δ D is defined in (16) . The inverse threshold is independent of n n n , ς \varsigma ς and of the linear amplitude size. All nine entries use the same X 1 , Y 2 X_1,Y_2 X 1 , Y 2 and physical displacements (117) . Constants may depend on m m m , the full C 2 m + 14 C^{2m+14} C 2 m + 14 prior, the fixed geometry, cutoffs, and compact atlas, but not on the chosen materials, n n n , t 0 t_0 t 0 , ς \varsigma ς , h h h , or δ D \delta_D δ D .
Proof. Construct the positive columns with (126) , and the three fixed negative columns with the uniform amplitude bounds corresponding to K = 1 K=1 K = 1 . Apply Theorems 3.1 and 3.2 as in (4.19)–(4.22), so the bounds (4.23) and (124) hold simultaneously with K = K n , ς K=K_{n,\varsigma} K = K n , ς .
Define the first-order 4 × 4 4\times4 4 × 4 difference operator by
D = M 1 − M 2 , (128) \mathcal{D}=\mathcal{M}_1-\mathcal{M}_2, \tag*{(128)} D = M 1 − M 2 , (128)
and use A Δ , θ A_{\Delta,\theta} A Δ , θ from (50) . The exact phase expansion is
h e − φ / h D z 1 , h + = A Δ , θ a 0 + + h A Δ , θ a 1 + + h D a 0 + + A Δ , θ r 1 , h + + h 2 D a 1 + + h D r 1 , h + . (129) he^{-\varphi/h}\mathcal{D}z_{1,h}^{+}
=A_{\Delta,\theta}a_0^{+}+hA_{\Delta,\theta}a_1^{+}+h\mathcal{D}a_0^{+}+A_{\Delta,\theta}r_{1,h}^{+}
+h^2\mathcal{D}a_1^{+}+h\mathcal{D}r_{1,h}^{+}. \tag*{(129)} h e − φ / h D z 1 , h + = A Δ , θ a 0 + + h A Δ , θ a 1 + + h D a 0 + + A Δ , θ r 1 , h + + h 2 D a 1 + + h D r 1 , h + . (129)
In L 2 ( B R ) L^2(B_R) L 2 ( B R ) , the five nonleading terms have bounds respectively C h K ChK C h K , C h K ChK C h K , C h 2 K Ch^2K C h 2 K , C h 2 K Ch^2K C h 2 K , C h 3 K Ch^3K C h 3 K . Pair the expansion with Y χ + q h \mathcal{Y}_{\chi}+q_h Y χ + q h from (124) ; the exponential factors cancel exactly. If E h E_h E h denotes the sum of the five nonleading terms in (129) , then ∥ E h ∥ 2 ≤ C h K \|E_h\|_2\le ChK ∥ E h ∥ 2 ≤ C h K and ∥ A Δ , θ a 0 + ∥ 2 ≤ C K \|A_{\Delta,\theta}a_0^{+}\|_2\le CK ∥ A Δ , θ a 0 + ∥ 2 ≤ C K . Hence the entire integrated error is bounded by
∥ q h ∥ 2 ∥ A Δ , θ a 0 + ∥ 2 + ( ∥ Y χ ∥ 2 + ∥ q h ∥ 2 ) ∥ E h ∥ 2 ≤ C h K . (130) \|q_h\|_2\|A_{\Delta,\theta}a_0^{+}\|_2+\left(\|\mathcal{Y}_{\chi}\|_2+\|q_h\|_2\right)\|E_h\|_2\le ChK. \tag*{(130)} ∥ q h ∥ 2 ∥ A Δ , θ a 0 + ∥ 2 + ( ∥ Y χ ∥ 2 + ∥ q h ∥ 2 ) ∥ E h ∥ 2 ≤ C h K . (130)
Only the positive side has size K K K . Thus the full error is linear in K K K , including the product of the two remainder terms. No product estimate in L ∞ L^\infty L ∞ is needed. On the support of b ς , t 0 b_{\varsigma,t_0} b ς , t 0 the exact quotient contraction in (64) gives
∥ h ∫ B R Y 2 , h T D z 1 , h + d x − 2 ∫ B R b ς , t 0 g n Y 2 T ( P 1 − P 2 ) X 1 d x ∥ ≤ C h K . (131) \left\|h\int_{B_R}Y_{2,h}^{T}\mathcal{D}z_{1,h}^{+}\,dx-2\int_{B_R}b_{\varsigma,t_0}g_nY_2^{T}(P_1-P_2)X_1\,dx\right\|\le ChK. \tag*{(131)} h ∫ B R Y 2 , h T D z 1 , h + d x − 2 ∫ B R b ς , t 0 g n Y 2 T ( P 1 − P 2 ) X 1 d x ≤ C h K . (131)
For a , b ∈ { 1 , 2 , 3 } a,b\in\{1,2,3\} a , b ∈ { 1 , 2 , 3 } the ( a , b ) (a,b) ( a , b ) entry uses the physical pair ( u 2 , h ℓ ( a ) , u 1 , h ℓ ( b ) ) (u_{2,h}^{\ell(a)},u_{1,h}^{\ell(b)}) ( u 2 , h ℓ ( a ) , u 1 , h ℓ ( b ) ) , with the polarization indexing
ℓ ( 1 ) = F 1 , ℓ ( 2 ) = F 2 , ℓ ( 3 ) = s u p . (132) \ell(1)=F1,\qquad\ell(2)=F2,\qquad\ell(3)=\mathrm{sup}. \tag*{(132)} ℓ ( 1 ) = F 1 , ℓ ( 2 ) = F 2 , ℓ ( 3 ) = sup . (132)
This specifies the three-by-three array of nine pairings, without replacing any row by an auxiliary solution. The estimates are on the whole ball, including where the leading positive amplitude vanishes and the exact remainder need not vanish.
The materials coincide on an open neighborhood of ∂ B R \partial B_R ∂ B R ; apply Lemma 2.3 to each of these nine pairs:
∫ B R ( Y 2 , h ℓ ( a ) ) T D Z 1 , h + , ( b ) d x = ⟨ ( Λ 2 B R − Λ 1 B R ) Tr u 1 , h ( b ) , Tr u 2 , h ℓ ( a ) ⟩ . (133) \int_{B_R} \left(Y_{2,h}^{\ell(a)}\right)^{T}\mathcal{D} Z_{1,h}^{+,(b)}\,dx
=
\left\langle\left(\Lambda_{2}^{B_R}-\Lambda_{1}^{B_R}\right)\operatorname{Tr}u_{1,h}^{(b)},\operatorname{Tr}u_{2,h}^{\ell(a)}\right\rangle. \tag*{(133)} ∫ B R ( Y 2 , h ℓ ( a ) ) T D Z 1 , h + , ( b ) d x = ⟨ ( Λ 2 B R − Λ 1 B R ) Tr u 1 , h ( b ) , Tr u 2 , h ℓ ( a ) ⟩ . (133)
By (120) and the trace theorem, multiplication by h h h bounds each entry by
C δ B R h ( h − 2 K ) ( h − 1 ) e 2 R / h ≤ C δ D h − 2 e C 0 / h K , (134) C\delta_{B_R}h\left(h^{-2}K\right)\left(h^{-1}\right)e^{2R/h}
\leq C\delta_D h^{-2}e^{C_0/h}K, \tag*{(134)} C δ B R h ( h − 2 K ) ( h − 1 ) e 2 R / h ≤ C δ D h − 2 e C 0 / h K , (134)
using Lemma 2.4 only on its stated domain and exterior-equality hypotheses. All norms of a 3 × 3 3 \times3 3 × 3 matrix are equivalent with fixed constants. Combining (131) with (134) and dividing by two, we obtain (127) . The inverses in (98) and (88) are linear on arbitrary sources; the only smallness requirement is its operator threshold h < h ∗ h<h_* h < h ∗ . No argument uses smallness of h K hK h K or invertibility of the exact canonical field.
Remark. For δ D = 0 \delta_D=0 δ D = 0 , fix an admissible ( n , t 0 , ς ) (n,t_0,\varsigma) ( n , t 0 , ς ) and let h → 0 h\to0 h → 0 in (127) to obtain the exact zero moment. For positive error the inequality holds with simultaneous quantifiers; tests may later depend on the error without changing the family or imposing h K hK h K smallness. For the nonmatching extensions in the original-domain problem, we apply Theorem 4.2 with D = B R e D=B_{R_e} D = B R e and δ D = Δ e \delta_D=\Delta_e δ D = Δ e . The comparison to original measurements then uses Proposition 2.1 with its exterior-mismatch term, as developed in § 9 .
Global solutions of the relative transport equations
The matrix moments control the negative boundary Fourier modes of W a = Y 2 , a T X 1 , a W_a=Y_{2,a}^{T}X_{1,a} W a = Y 2 , a T X 1 , a . Invertible Cauchy extensions on the closed leaf disks and quantitative comparison on overlaps then yield global relative-transport endomorphisms G G G and H = G − T H=G^{-T} H = G − T .
The fixed input and the measurement domain
Use the quotient coordinates of § 2 , the frames of Proposition 4.1 , and the moment estimates of Theorem 4.2 , with the regularity orders in (94) and the corresponding common-exterior material prior. Corollary 3.4 supplies the finite atlas and residual estimates.
The measurement domain D D D need not equal Ω \Omega Ω . The discrepancies δ D \delta_D δ D , Δ e \Delta_e Δ e are defined in (16) , and η D \eta_D η D in (17) . When the materials agree outside D D D , Lemma 2.4 bounds the working-sphere discrepancy by δ D \delta_D δ D ; its constant is absorbed in the moment estimate. The enlarged-domain application uses Δ e \Delta_e Δ e , and the enlarged-domain map requires no additional observation.
Choose fixed radii
D ‾ ⊂ B R s u p , R s u p < R 0 < R s < R , r ∗ = ( R 2 − R 0 2 ) 1 / 2 > 0. (135) \overline{D}\subset B_{R_{\mathrm{sup}}},\qquad R_{\mathrm{sup}}<R_0<R_s<R,\qquad r_*=(R^2-R_0^2)^{1/2}>0. \tag*{(135)} D ⊂ B R sup , R sup < R 0 < R s < R , r ∗ = ( R 2 − R 0 2 ) 1/2 > 0. (135)
The two connections agree outside B ‾ R s u p \overline{B}_{R_{\mathrm{sup}}} B R sup . The common-exterior material input also says that the material pairs themselves agree there. Define the logarithmic material ratio by
ρ = log ( μ 1 / μ 2 ) : B R ⟶ R . (136) \rho=\log(\mu_1/\mu_2):B_R\longrightarrow\mathbb{R}. \tag*{(136)} ρ = log ( μ 1 / μ 2 ) : B R ⟶ R . (136)
The common-exterior assumption gives ρ = 0 \rho=0 ρ = 0 on the fixed shell E E E in (203) . Use the normalized coordinates (54) and radius (125) , writing t = t ⊥ t=t_\perp t = t ⊥ . Define the complete leaf disks by
B θ , t = B R ∩ { x : ν l e a f ⋅ x = t } , ∣ t ∣ ≤ R 0 . (137) B_{\theta,t}=B_R\cap\{x:\nu_{\mathrm{leaf}}\cdot x=t\},\qquad|t|\leq R_0. \tag*{(137)} B θ , t = B R ∩ { x : ν leaf ⋅ x = t } , ∣ t ∣ ≤ R 0 . (137)
The frames on the disks (137) are given on neighborhoods of the complete closed disks and the annuli used in the proof of Lemma 5.1 .
For each material pair, fix one finite family of 3 × 3 3\times3 3 × 3 frames X j , a X_{j,a} X j , a , Y j , a Y_{j,a} Y j , a , j = 1 , 2 j=1,2 j = 1 , 2 , on direction-transverse patches. The family is held fixed for every degree n n n , width s s s , and parameter h h h . After pullback to fixed cylinders, frames and square-frame inverses have uniform derivatives of total order at most r f r r_{\mathrm{fr}} r fr . The frames satisfy (62) , with A j A_j A j , Γ j \Gamma_j Γ j , P j P_j P j defined in (61) . The transpose here is the ordinary bilinear transpose. Define the frame products W a W_a W a and the matrices F ~ a \widetilde{F}_a F a by
W a = Y 2 , a T X 1 , a , F ~ a = Y 2 , a T ( P 1 − P 2 ) X 1 , a , 2 ∂ z ‾ W a = − F ~ a . (138) W_a = Y_{2,a}^{T}X_{1,a}, \qquad\widetilde{F}_a = Y_{2,a}^{T}(P_1-P_2)X_{1,a}, \qquad2\partial_{\overline{z}}W_a=-\widetilde{F}_a. \tag*{(138)} W a = Y 2 , a T X 1 , a , F a = Y 2 , a T ( P 1 − P 2 ) X 1 , a , 2 ∂ z W a = − F a . (138)
In particular F ~ a \widetilde{F}_a F a is supported in B ‾ R s u p \overline{B}_{R_{\mathrm{sup}}} B R sup and W a − 1 = X 1 , a − 1 Y 2 , a − T W_a^{-1}=X_{1,a}^{-1}Y_{2,a}^{-T} W a − 1 = X 1 , a − 1 Y 2 , a − T exists on the whole closed disk. Theorem 4.2 supplies the following estimate for this fixed frame family. For every n ≥ 0 n\ge0 n ≥ 0 , every admissible center t 0 t_0 t 0 and width 0 < ς ≤ ς ∗ 0<\varsigma\le\varsigma_* 0 < ς ≤ ς ∗ , and every 0 < h < h ∗ 0<h<h_* 0 < h < h ∗ ,
∥ ∫ b ς , t 0 ( t ) ( z r ( t ) ) n F ~ a ( x , θ ) d x ∥ ≤ C A n ( n + 1 ) 12 ς − 13 η D , A > 1. (139) \left\lVert\int b_{\varsigma,t_0}(t)\left(\frac{z}{r(t)}\right)^n\widetilde{F}_a(x,\theta)\,dx \right\rVert\le CA^n(n+1)^{12}\varsigma^{-13}\eta_D,\qquad A>1. \tag*{(139)} ∫ b ς , t 0 ( t ) ( r ( t ) z ) n F a ( x , θ ) d x ≤ C A n ( n + 1 ) 12 ς − 13 η D , A > 1. (139)
Here b ς , t 0 b_{\varsigma,t_0} b ς , t 0 and the normalized profile are defined in (125) ; F ~ a \widetilde{F}_a F a is defined in (138) . The constants C C C , A A A and the upper bound h ∗ h_* h ∗ are independent of n n n , t 0 t_0 t 0 , ς \varsigma ς and the amplitude size; the dependence of the bound on degree and width is explicit in (139) . The cutoffs have fixed parameter margins, and the frame family is held fixed for every test.
For the global sum we also use a finite product cover of a compact material-pair parameter space and C P 1 × [ − R 0 , R 0 ] \mathbb{CP}^{1}\times[-R_0,R_0] CP 1 × [ − R 0 , R 0 ] , with the same local frame family available on larger patches. Uniform bounds, a fixed finite cover with margins, and continuity in the material label are input properties. No derivative with respect to that label, and no parameter derivative of the exact CGO right inverse, is required. Here the required continuity means local availability of the same frame construction on open coefficient neighborhoods, so that the fixed cover and its material cutoffs can be chosen. It is supplied by Theorem 3.1 and the physical construction in Proposition 4.1 ; no differentiability of the selected frame with respect to material is used below. The matrices used here are 3 × 3 3\times3 3 × 3 quotient frames, not the 4 × 3 4\times3 4 × 3 ambient lifts in the proof of Theorem 4.2 .
Moments, radii, and negative Fourier modes
Localized moment estimates and annular decay yield bounds for the negative Fourier coefficients of the boundary traces of W a W_a W a ; see (138) .
For a sufficiently regular matrix-valued function V V V on the unit circle, set
V ^ ( k ) = 1 2 π ∫ 0 2 π V ( e i ϕ ) e − i k ϕ d ϕ , P − V = ∑ k < 0 V ^ ( k ) e i k ϕ , P + V = ∑ k ≥ 0 V ^ ( k ) e i k ϕ , V − = P − V . (140) \widehat{V}(k)=\frac{1}{2\pi}\int_{0}^{2\pi}V(e^{i\phi})e^{-ik\phi}\,d\phi,
\qquad
P_-V=\sum_{k<0}\widehat{V}(k)e^{ik\phi},\qquad
P_+V=\sum_{k\ge0}\widehat{V}(k)e^{ik\phi},\qquad
V_-=P_-V. \tag*{(140)} V ( k ) = 2 π 1 ∫ 0 2 π V ( e i ϕ ) e − ik ϕ d ϕ , P − V = k < 0 ∑ V ( k ) e ik ϕ , P + V = k ≥ 0 ∑ V ( k ) e ik ϕ , V − = P − V . (140)
The Cauchy integral in (154) , applied to V V V , is holomorphic in the disk and has boundary trace P + V P_+V P + V .
The next lemma removes the transverse averaging from the moment estimates and bounds the negative Fourier coefficients of the boundary traces of W a W_a W a . It combines the estimates for low-degree moments with annular decay for the high modes. The resulting boundary error is used to construct the invertible Cauchy extensions R C , a R_{C,a} R C , a and the local solutions G a G_a G a .
Lemma 5.1. Under the hypotheses of Subsection 5.1, take the negative Fourier part after pulling each leaf boundary back to the unit circle, as in (140) . Define
I n ( θ , t ) = ∫ B θ , t q n ( z , t ) F ~ a d A , ρ = R s / R < 1. (141) I_n(\theta,t)=\int_{B_{\theta,t}}q_n(z,t)\widetilde{F}_a\,dA,\qquad\rho=R_s/R<1. \tag*{(141)} I n ( θ , t ) = ∫ B θ , t q n ( z , t ) F a d A , ρ = R s / R < 1. (141) Here B θ , t B_{\theta,t} B θ , t , q n q_n q n , F ~ a \widetilde{F}_a F a are defined in (137) , (125) , and (138) . Choose a fixed A 1 > A A_1>A A 1 > A and put
γ ∗ = log ( 1 / ρ ) log A 1 + log ( 1 / ρ ) . (142) \gamma_*=\frac{\log(1/\rho)}{\log A_1+\log(1/\rho)}. \tag*{(142)} γ ∗ = log A 1 + log ( 1/ ρ ) log ( 1/ ρ ) . (142)
There is a uniform η m o m > 0 \eta_{\mathrm{mom}}>0 η mom > 0 such that, for 0 < η D < η m o m 0<\eta_D<\eta_{\mathrm{mom}} 0 < η D < η mom ,
∥ I n ( θ , t ) ∥ ≤ C A n ( n + 1 ) 12 η D 1 / 14 , ∥ ( W a , ∂ ) − ∥ C 0 ≤ C γ ′ η D γ ′ / 14 ( 0 < γ ′ < γ ∗ ) . (143) \lVert I_n(\theta,t)\rVert\le C A^n(n+1)^{12}\eta_D^{1/14}, \qquad\lVert(W_{a,\partial})^{-}\rVert_{C^0}\le C\gamma'\eta_D^{\gamma'/14}\quad(0<\gamma'<\gamma_*). \tag*{(143)} ∥ I n ( θ , t )∥ ≤ C A n ( n + 1 ) 12 η D 1/14 , ∥( W a , ∂ ) − ∥ C 0 ≤ C γ ′ η D γ ′ /14 ( 0 < γ ′ < γ ∗ ) . (143)
Here η D \eta_D η D is defined in (17) . The constants depend only on the fixed frame bounds, atlas, cutoffs, radii, and γ ′ \gamma' γ ′ .
Proof. Use z = r ( t ) w z=r(t)w z = r ( t ) w to write
I n ( θ , t ) = ∫ ∣ w ∣ < 1 w n r ( t ) 2 ς a ( t ν l e a f + r ( t ) ( ℜ w e 1 + ℑ w e 2 ) , θ ) d A ( w ) . (144) I_n(\theta,t)=\int_{|w|<1}w^n r(t)^2\varsigma_a(t\nu_{\mathrm{leaf}}+r(t)(\Re w\,e_1+\Im w\,e_2),\theta)\,dA(w). \tag*{(144)} I n ( θ , t ) = ∫ ∣ w ∣ < 1 w n r ( t ) 2 ς a ( t ν leaf + r ( t ) ( ℜ w e 1 + ℑ w e 2 ) , θ ) d A ( w ) . (144)
The positive lower bound r ∗ r_* r ∗ and the full-disk frame priors justify one transverse derivative under the integral and give ∥ ∂ t I n ∥ ≤ C A n ( n + 1 ) 12 \lVert\partial_t I_n\rVert\le CA^n(n+1)^{12} ∥ ∂ t I n ∥ ≤ C A n ( n + 1 ) 12 on larger parameter patches. The fixed source support also prevents a boundary support issue. By the transverse derivative bound, the normalization ∫ b ς , t 0 = 1 \int b_{\varsigma,t_0}=1 ∫ b ς , t 0 = 1 , and (5.5), we obtain
∥ I n ( θ , t 0 ) ∥ ≤ C A n ( n + 1 ) 12 ( ς + η D ς − 13 ) . (145) \lVert I_n(\theta,t_0)\rVert\le CA^n(n+1)^{12}(\varsigma+\eta_D\varsigma^{-13}). \tag*{(145)} ∥ I n ( θ , t 0 )∥ ≤ C A n ( n + 1 ) 12 ( ς + η D ς − 13 ) . (145)
Take ς = η D 1 / 14 \varsigma=\eta_D^{1/14} ς = η D 1/14 . For a fixed legal width ς 0 > 0 \varsigma_0>0 ς 0 > 0 smaller than all atlas margins, one may take η m o m ≤ min { 1 , ς 0 14 } \eta_{\mathrm{mom}}\le\min\{1,\varsigma_0^{14}\} η mom ≤ min { 1 , ς 0 14 } . This proves the first bound.
For a negative-mode index k ≥ 1 k\ge1 k ≥ 1 , define
c − k = 1 2 π ∫ 0 2 π W a ( r e i ϕ ) e i k ϕ d ϕ . (146) c_{-k}=\frac{1}{2\pi}\int_0^{2\pi}W_a(re^{i\phi})e^{ik\phi}\,d\phi. \tag*{(146)} c − k = 2 π 1 ∫ 0 2 π W a ( r e i ϕ ) e ik ϕ d ϕ . (146)
Here W a W_a W a and r ( t ) r(t) r ( t ) are defined in (138) and (125) , respectively. Counterclockwise orientation and Cauchy–Green give the exact identity
2 π i r k c − k = ∫ ∣ z ∣ = r z k − 1 W a ( z ) d z = 2 i ∫ ∣ z ∣ < r z k − 1 ∂ z ˉ W a d A = − i r k − 1 I k − 1 , c − k = − I k − 1 2 π r . (147) \begin{aligned}
2\pi i r^k c_{-k}
&=\int_{|z|=r}z^{k-1}W_a(z)\,dz
=2i\int_{|z|<r}z^{k-1}\partial_{\bar z}W_a\,dA\\
&=-ir^{k-1}I_{k-1},\qquad c_{-k}=-\frac{I_{k-1}}{2\pi r}. \tag*{(147)}
\end{aligned} 2 π i r k c − k = ∫ ∣ z ∣ = r z k − 1 W a ( z ) d z = 2 i ∫ ∣ z ∣ < r z k − 1 ∂ z ˉ W a d A = − i r k − 1 I k − 1 , c − k = − 2 π r I k − 1 . (147)
Thus degree n n n controls mode − ( n + 1 ) -(n+1) − ( n + 1 ) ; the sign and r − 1 r^{-1} r − 1 factor are not absorbed before this equality is established.
Set
r s ( t ) = ( R s 2 − t 2 ) 1 / 2 . (148) r_s(t)=(R_s^2-t^2)^{1/2}. \tag*{(148)} r s ( t ) = ( R s 2 − t 2 ) 1/2 . (148)
where R s R_s R s is the fixed radius in (135) . The fixed annulus r s ( t ) ≤ ∣ z ∣ ≤ r ( t ) r_s(t)\le|z|\le r(t) r s ( t ) ≤ ∣ z ∣ ≤ r ( t ) lies outside the source and has a neighborhood of holomorphy for W a W_a W a . Laurent contour deformation and the uniform inner-circle bound imply
∥ c − k ∥ ≤ C ( r s ( t ) / r ( t ) ) k ≤ C ϱ k . (149) \lVert c_{-k}\rVert\le C\left(r_s(t)/r(t)\right)^k\le C\varrho^k. \tag*{(149)} ∥ c − k ∥ ≤ C ( r s ( t ) / r ( t ) ) k ≤ C ϱ k . (149)
The last inequality follows by cross-multiplying ( R s 2 − t 2 ) / ( R 2 − t 2 ) ≤ R s 2 / R 2 (R_s^2-t^2)/(R^2-t^2)\le R_s^2/R^2 ( R s 2 − t 2 ) / ( R 2 − t 2 ) ≤ R s 2 / R 2 . Both inner and outer radii are uniformly nondegenerate. The moment estimate and sup k ≥ 1 k 12 ( A / A 1 ) k < ∞ \sup_{k\ge1}k^{12}(A/A_1)^k<\infty sup k ≥ 1 k 12 ( A / A 1 ) k < ∞ now give
∥ c − k ∥ ≤ C min { A 1 k ε , ϱ k } , ε = η D 1 / 14 . (150) \lVert c_{-k}\rVert\le C\min\{A_1^k\varepsilon,\varrho^k\},\qquad\varepsilon=\eta_D^{1/14}. \tag*{(150)} ∥ c − k ∥ ≤ C min { A 1 k ε , ϱ k } , ε = η D 1/14 . (150)
Choose the Fourier splitting index by
L = log ( A 1 / ϱ ) , K = ⌊ log ( 1 / ε ) L ⌋ . (151) L=\log(A_1/\varrho),\qquad K=\left\lfloor\frac{\log(1/\varepsilon)}{L}\right\rfloor. \tag*{(151)} L = log ( A 1 / ϱ ) , K = ⌊ L log ( 1/ ε ) ⌋ . (151)
The sums for k ≤ K k\le K k ≤ K and k > K k>K k > K are bounded respectively by C ε A 1 K C\varepsilon A_1^K C ε A 1 K and C ϱ K + 1 C\varrho^{K+1} C ϱ K + 1 ; both are at most C ε γ ∗ C\varepsilon^{\gamma_*} C ε γ ∗ . The case K = 0 K=0 K = 0 satisfies the same bound. Absolute convergence bounds the negative Fourier series in C 0 C^0 C 0 . For η D ≤ 1 \eta_D\le1 η D ≤ 1 this implies the stated weaker exponent γ ′ \gamma' γ ′ . No small high-order norm of that series is inferred. All later high-order bounds come from the full-boundary frame input.
An invertible Cauchy extension on the closed disk
The Cauchy integral of the boundary trace of W a W_a W a gives a holomorphic matrix on the leaf disk. The small negative Fourier part and the continuous invertible extension of the boundary trace ensure invertibility up to the boundary, with the mixed derivative bounds needed for the local relative transport. Related constructions for connection Laplacians appear in [12 Section 3.3] .
Lemma 5.2. Let p p p vary in a fixed compact parameter chart and N ≥ 2 N \ge2 N ≥ 2 . Suppose W p ♯ ∈ C 0 ( D ‾ ; G L ( 3 , C ) ) W^\sharp_p \in C^0(\overline{\mathbb{D}}; GL(3,\mathbb{C})) W p ♯ ∈ C 0 ( D ; G L ( 3 , C )) , ∥ ( W p ♯ ) − 1 ∥ C 0 ≤ B \|(W^\sharp_p)^{-1}\|_{C^0} \le B ∥ ( W p ♯ ) − 1 ∥ C 0 ≤ B , and define its boundary trace by
W ∂ , p = W p ♯ ∣ ∂ D . (152) W_{\partial,p}=W^\sharp_p|_{\partial\mathbb{D}}. \tag*{(152)} W ∂ , p = W p ♯ ∣ ∂ D . (152) Assume
sup ∣ α ∣ ≤ N ∥ ∂ p α W ∂ , p ∥ C ϕ N + 3 ≤ M N , ∥ ( W ∂ , p ) − ∥ C 0 ≤ ω F . (153) \sup_{|\alpha|\le N}\|\partial_p^\alpha W_{\partial,p}\|_{C_\phi^{N+3}}\le M_N,\qquad\|(W_{\partial,p})_-\|_{C^0}\le\omega_F. \tag*{(153)} ∣ α ∣ ≤ N sup ∥ ∂ p α W ∂ , p ∥ C ϕ N + 3 ≤ M N , ∥ ( W ∂ , p ) − ∥ C 0 ≤ ω F . (153) If B ω F < 1 / 2 B\omega_F<1/2 B ω F < 1/2 , then
R C , p ♯ ( w ) = 1 2 π i ∫ ∣ ξ ∣ = 1 W ∂ , p ( ξ ) ξ − w d ξ (154) R^\sharp_{C,p}(w)=\frac{1}{2\pi i}\int_{|\xi|=1}\frac{W_{\partial,p}(\xi)}{\xi-w}\,d\xi \tag*{(154)} R C , p ♯ ( w ) = 2 π i 1 ∫ ∣ ξ ∣ = 1 ξ − w W ∂ , p ( ξ ) d ξ (154) extends continuously to the closed disk, is pointwise invertible there, and
∥ ( R C ♯ ) − 1 ∥ C 0 ≤ 2 B , ∥ R C ♯ ∥ C p N C w N + ∥ ( R C ♯ ) − 1 ∥ C p N C w N ≤ C ( N , M N , B ) . (155) \|(R^\sharp_C)^{-1}\|_{C^0}\le2B,\qquad\|R^\sharp_C\|_{C_p^N C_w^N}+\|(R^\sharp_C)^{-1}\|_{C_p^N C_w^N}\le C(N,M_N,B). \tag*{(155)} ∥ ( R C ♯ ) − 1 ∥ C 0 ≤ 2 B , ∥ R C ♯ ∥ C p N C w N + ∥ ( R C ♯ ) − 1 ∥ C p N C w N ≤ C ( N , M N , B ) . (155) Here the mixed norm permits parameter and leaf derivatives separately through order N N N ; it is stronger than a total-order-N N N bound.
Proof. For k ≥ 0 k\ge0 k ≥ 0 , write the Fourier coefficients as
c k ( p ) = 1 2 π ∫ 0 2 π W ∂ , p ( e i ϕ ) e − i k ϕ d ϕ . c_k(p)=\frac{1}{2\pi}\int_0^{2\pi}W_{\partial,p}(e^{i\phi})e^{-ik\phi}\,d\phi. c k ( p ) = 2 π 1 ∫ 0 2 π W ∂ , p ( e i ϕ ) e − ik ϕ d ϕ .
Integration by parts N + 3 N+3 N + 3 times, using (5.19), gives
∥ ∂ p α c k ∥ ≤ C M N ( 1 + k ) − N − 3 , ∣ α ∣ ≤ N , \|\partial_p^\alpha c_k\|\le CM_N(1+k)^{-N-3},\qquad|\alpha|\le N, ∥ ∂ p α c k ∥ ≤ C M N ( 1 + k ) − N − 3 , ∣ α ∣ ≤ N ,
where k = 0 k=0 k = 0 is bounded directly. The Cauchy integral (5.20) equals R C , p ♯ ( w ) = ∑ k ≥ 0 c k ( p ) w k R^\sharp_{C,p}(w)=\sum_{k\ge0}c_k(p)w^k R C , p ♯ ( w ) = ∑ k ≥ 0 c k ( p ) w k for ∣ w ∣ < 1 |w|<1 ∣ w ∣ < 1 . After b ≤ N b\le N b ≤ N leaf derivatives, the series is bounded by C M N ∑ k ≥ 0 ( 1 + k ) b − N − 3 < ∞ CM_N\sum_{k\ge0}(1+k)^{b-N-3}<\infty C M N ∑ k ≥ 0 ( 1 + k ) b − N − 3 < ∞ . It therefore extends with the stated mixed derivatives to the closed disk, and R C ♯ = W ∂ − ( W ∂ ) − R^\sharp_C=W_{\partial}-(W_{\partial})_- R C ♯ = W ∂ − ( W ∂ ) − on its boundary.
Fix p p p and set E = ( W ∂ ) − E=(W_{\partial})_- E = ( W ∂ ) − . The boundary homotopy satisfies
M t = W ∂ − t E = W ∂ ( I − t W ∂ − 1 E ) , M_t=W_{\partial}-tE=W_{\partial}(I-tW_{\partial}^{-1}E), M t = W ∂ − tE = W ∂ ( I − t W ∂ − 1 E ) ,
∥ M t − 1 ∥ ≤ B 1 − B ω F ≤ 2 B , 0 ≤ t ≤ 1. \|M_t^{-1}\|\le\frac{B}{1-B\omega_F}\le2B,\qquad0\le t\le1. ∥ M t − 1 ∥ ≤ 1 − B ω F B ≤ 2 B , 0 ≤ t ≤ 1.
Indeed ∥ t W ∂ − 1 E ∥ ≤ B ω F < 1 / 2 \|tW_{\partial}^{-1}E\|\le B\omega_F<1/2 ∥ t W ∂ − 1 E ∥ ≤ B ω F < 1/2 , so the Neumann series applies uniformly in t t t . The loop det W ∂ \det W_{\partial} det W ∂ has winding number zero because it extends to the nonvanishing function det W ♯ \det W^\sharp det W ♯ on the disk. Since M 1 = R C ♯ M_1=R^\sharp_C M 1 = R C ♯ on the circle, homotopy invariance gives
wind ( det R C ♯ ∣ ∂ D , 0 ) = wind ( det W ∂ , 0 ) = 0. \operatorname{wind}(\det R^\sharp_C|_{\partial\mathbb{D}},0)=\operatorname{wind}(\det W_{\partial},0)=0. wind ( det R C ♯ ∣ ∂ D , 0 ) = wind ( det W ∂ , 0 ) = 0.
The argument principle now excludes zeros of det R C ♯ \det R^\sharp_C det R C ♯ in the disk; see [1 Chapter 4, §5.2, Theorem 18, pp. 152–153] . It may be applied on circles ∣ w ∣ = ρ < 1 |w|=\rho<1 ∣ w ∣ = ρ < 1 sufficiently close to one: continuity and boundary nonvanishing give a zero-free outer annulus, and radial deformation preserves the winding number.
Thus ( R C ♯ ) − 1 (R_C^\sharp)^{-1} ( R C ♯ ) − 1 is holomorphic inside and continuous on the closed disk. Apply the scalar maximum principle to v ∗ ( R C ♯ ) − 1 u v^*(R_C^\sharp)^{-1}u v ∗ ( R C ♯ ) − 1 u for constant unit vectors u u u , v v v , and take their supremum. The boundary bound above yields
sup w ∈ D ∥ ( R C ♯ ( w ) ) − 1 ∥ ≤ sup ∣ w ∣ = 1 ∥ ( R C ♯ ( w ) ) − 1 ∥ ≤ 2 B . \sup_{w\in\mathbb{D}}\left\|(R_C^\sharp(w))^{-1}\right\| \le\sup_{|w|=1}\left\|(R_C^\sharp(w))^{-1}\right\| \le2B. w ∈ D sup ( R C ♯ ( w ) ) − 1 ≤ ∣ w ∣ = 1 sup ( R C ♯ ( w ) ) − 1 ≤ 2 B .
The star is used here only for the matrix operator norm. For any real leaf or parameter derivative, differentiation gives
∂ j ( R C ♯ ) − 1 = − ( R C ♯ ) − 1 ( ∂ j R C ♯ ) ( R C ♯ ) − 1 . \partial_j(R_C^\sharp)^{-1}=-(R_C^\sharp)^{-1}(\partial_jR_C^\sharp)(R_C^\sharp)^{-1}. ∂ j ( R C ♯ ) − 1 = − ( R C ♯ ) − 1 ( ∂ j R C ♯ ) ( R C ♯ ) − 1 .
Induction on the mixed derivative order, using the direct derivative bounds and this zero-order inverse bound, proves (155) .
Remark. Boundary invertibility alone would not exclude zeros of the holomorphic determinant inside the disk. The existing nonvanishing extension of W 0 W_0 W 0 fixes its winding number, which supplies that interior conclusion. The pointwise invertibility of R C , a R_{C,a} R C , a on the closed disk permits G a G_a G a in (161) to be defined up to the leaf boundary.
Common coordinates and quantitative comparison
Local transports constructed in different charts must first be expressed in the same quotient coordinates before their boundary differences can be compared.
We use the normalization law (67) , with S a S_a S a defined in (66) , and the spinor transition law (65) , with D t r D_{\mathrm{tr}} D tr from (57) . For the primal transport, inversion and transposition give the additional normalization formula
G h = S a − T G n S a − T . (156) G_h=S_a^{-T}G_nS_a^{-T}. \tag*{(156)} G h = S a − T G n S a − T . (156)
The coordinate laws apply to a given endomorphism. The local solutions of the relative transport equation constructed from different Cauchy extensions are compared by Lemma 5.3. On compact overlaps all these changes and their required derivatives are uniformly bounded. After removing the direction-scaling factor and applying the coordinate transformations (156) and (65) , we express both local equations in the same coordinates at the same physical point.
Once both equations are in these common coordinates, the matrix X 1 − 1 ( G a − G b ) X 2 X_1^{-1}(G_a-G_b)X_2 X 1 − 1 ( G a − G b ) X 2 is holomorphic in the leaf coordinate. The following maximum-principle estimate controls G a − G b G_a-G_b G a − G b throughout the disk.
Lemma 5.3. On a complete closed leaf disk let G a , G b G_a,G_b G a , G b solve 2 ∂ z ˉ G + P 1 G − G P 2 = 0 2\partial_{\bar z}G+P_1G-GP_2=0 2 ∂ z ˉ G + P 1 G − G P 2 = 0 in the same coordinates. Suppose invertible frames X 1 , X 2 X_1,X_2 X 1 , X 2 solve 2 ∂ z ˉ X j + P j X j = 0 2\partial_{\bar z}X_j+P_jX_j=0 2 ∂ z ˉ X j + P j X j = 0 , with each frame and inverse bounded by M 0 M_0 M 0 . Then
∥ G a − G b ∥ C 0 ( B ‾ ) ≤ M 0 4 ∥ G a − G b ∥ C 0 ( ∂ B ) . (157) \|G_a-G_b\|_{C^0(\overline{B})}\le M_0^4\|G_a-G_b\|_{C^0(\partial B)}. \tag*{(157)} ∥ G a − G b ∥ C 0 ( B ) ≤ M 0 4 ∥ G a − G b ∥ C 0 ( ∂ B ) . (157)
Proof. Set
V = X 1 − 1 ( G a − G b ) X 2 , (158) V=X_1^{-1}(G_a-G_b)X_2, \tag*{(158)} V = X 1 − 1 ( G a − G b ) X 2 , (158)
where X j X_j X j are the quotient frames in (62) . The equations (62) imply
∂ z ˉ V = 0 , ∥ V ∥ C 0 ( B ‾ ) ≤ ∥ V ∥ C 0 ( ∂ B ) . (159) \partial_{\bar z}V=0,\qquad\|V\|_{C^0(\overline{B})}\le\|V\|_{C^0(\partial B)}. \tag*{(159)} ∂ z ˉ V = 0 , ∥ V ∥ C 0 ( B ) ≤ ∥ V ∥ C 0 ( ∂ B ) . (159)
The second inequality follows by applying the maximum principle to constant-vector matrix elements. Multiplying by the common frame factors on the boundary and in the interior, we obtain
∥ G a − G b ∥ C 0 ( B ‾ ) ≤ M 0 2 ∥ V ∥ C 0 ( B ‾ ) ≤ M 0 4 ∥ G a − G b ∥ C 0 ( ∂ B ) , (160) \|G_a-G_b\|_{C^0(\overline{B})}\le M_0^2\|V\|_{C^0(\overline{B})}\le M_0^4\|G_a-G_b\|_{C^0(\partial B)}, \tag*{(160)} ∥ G a − G b ∥ C 0 ( B ) ≤ M 0 2 ∥ V ∥ C 0 ( B ) ≤ M 0 4 ∥ G a − G b ∥ C 0 ( ∂ B ) , (160)
which is (157) .
Define the local solution of the relative transport equation in its initial quotient coordinates by
G a = X 1 , a R C , a − 1 Y 2 , a T , G a − 1 = Y 2 , a − T R C , a X 1 , a − 1 . (161) G_a = X_{1,a}R_{C,a}^{-1}Y_{2,a}^{T}, \qquad G_a^{-1} = Y_{2,a}^{-T}R_{C,a}X_{1,a}^{-1}. \tag*{(161)} G a = X 1 , a R C , a − 1 Y 2 , a T , G a − 1 = Y 2 , a − T R C , a X 1 , a − 1 . (161)
Here R C , a R_{C,a} R C , a is the Cauchy extension (154) pulled back from the unit disk to B θ , t B_{\theta,t} B θ , t , and X j , a , Y j , a X_{j,a}, Y_{j,a} X j , a , Y j , a are the frames of Proposition 4.1 . The frame equations (62) and ∂ z ˉ R C , a = 0 \partial_{\bar z}R_{C,a}=0 ∂ z ˉ R C , a = 0 give 2 ∂ z ˉ G a + P 1 G a − G a P 2 = 0 2\partial_{\bar z}G_a+P_1G_a-G_aP_2=0 2 ∂ z ˉ G a + P 1 G a − G a P 2 = 0 . On the leaf boundary ∂ B θ , t ⊂ ∂ B R \partial B_{\theta,t}\subset\partial B_R ∂ B θ , t ⊂ ∂ B R ,
G a − I = X 1 , a ( R C , a − 1 − W a − 1 ) Y 2 , a T , G_a-I=X_{1,a}(R_{C,a}^{-1}-W_a^{-1})Y_{2,a}^{T}, G a − I = X 1 , a ( R C , a − 1 − W a − 1 ) Y 2 , a T ,
R C , a − 1 − W a − 1 = R C , a − 1 ( W a − R C , a ) W a − 1 . (162) R_{C,a}^{-1}-W_a^{-1}=R_{C,a}^{-1}(W_a-R_{C,a})W_a^{-1}. \tag*{(162)} R C , a − 1 − W a − 1 = R C , a − 1 ( W a − R C , a ) W a − 1 . (162)
Consequently, after conversion to a common chart, ∥ G a − I ∥ ∂ B ≤ C b ω F \|G_a-I\|_{\partial B}\le C_b\omega_F ∥ G a − I ∥ ∂ B ≤ C b ω F and ∥ G a − G b ∥ B ≤ C o v ω F \|G_a-G_b\|_B\le C_{\mathrm{ov}}\omega_F ∥ G a − G b ∥ B ≤ C ov ω F . The constants are fixed by the atlas, frame bounds and Lemma 5.2 . In particular, G a G_a G a and G a − 1 G_a^{-1} G a − 1 have a common zero-order bound M l o c M_{\mathrm{loc}} M loc independent of the actual size of ω F \omega_F ω F below its Cauchy threshold.
Projective weights, an invertible sum, and one threshold
Leaf-constant weights combine the local transports without introducing additional ∂ z ˉ \partial_{\bar z} ∂ z ˉ terms. The overlap estimate supplies a uniform invertibility threshold.
For a nonzero complex null vector, define
ν l e a f ( [ θ ] ) = ℜ θ × ℑ θ ∣ ℜ θ ∣ 2 , π ( x , [ θ ] ) = ( [ θ ] , ν l e a f ( [ θ ] ) ⋅ x ) . (163) \nu_{\mathrm{leaf}}([\theta])=\frac{\Re\theta\times\Im\theta}{|\Re\theta|^2}, \qquad\pi(x,[\theta])=([\theta],\nu_{\mathrm{leaf}}([\theta])\cdot x). \tag*{(163)} ν leaf ([ θ ]) = ∣ℜ θ ∣ 2 ℜ θ × ℑ θ , π ( x , [ θ ]) = ([ θ ] , ν leaf ([ θ ]) ⋅ x ) . (163)
Nullity makes the two real vectors orthogonal with equal nonzero length. Multiplication by b ∈ C ∗ b\in\mathbb{C}^* b ∈ C ∗ multiplies both the numerator and denominator by ∣ b ∣ 2 |b|^2 ∣ b ∣ 2 ; hence ν l e a f \nu_{\mathrm{leaf}} ν leaf is projectively well defined and D θ ( ν l e a f ⋅ x ) = 0 D_\theta(\nu_{\mathrm{leaf}}\cdot x)=0 D θ ( ν leaf ⋅ x ) = 0 .
Refine the finite product cover to smaller sets with closures in the original patches. To obtain this cover on projective directions, choose a smooth normalized null representative on each larger compact spinor chart and pull back the coefficient–oriented-direction frame atlas of Theorem 3.1 along that local section. Its continuity and finite mixed bounds persist under this fixed smooth pullback. Refine the resulting cover on the compact material-pair space times C P 1 × [ − R 0 , R 0 ] \mathbb{C}\mathbb{P}^1\times[-R_0,R_0] C P 1 × [ − R 0 , R 0 ] once. Local constructions obtained from different sections need not coincide: they are compared only after conversion by (156) and (65) . Thus the following weights are functions on projective directions themselves. On the material factor choose continuous cutoffs η a ( a ) \eta_a(a) η a ( a ) , and on its direction–transverse factor choose smooth cutoffs ψ a ( p ) \psi_a(p) ψ a ( p ) , nonnegative and positive on the smaller sets. They may be bounded by one. Compactness gives a positive lower bound s ∗ s_* s ∗ for ∑ a η a ( a ) 2 ψ a ( p ) 2 \sum_a\eta_a(a)^2\psi_a(p)^2 ∑ a η a ( a ) 2 ψ a ( p ) 2 . Thus
χ a ( a , p ) = η a ( a ) 2 ψ a ( p ) 2 ∑ b η b ( a ) 2 ψ b ( p ) 2 , ∑ a χ a = 1 , χ a ≥ 0 , D θ ( χ a ∘ π ) = 0. (164) \chi_a(a,p)=\frac{\eta_a(a)^2\psi_a(p)^2}{\sum_b\eta_b(a)^2\psi_b(p)^2}, \qquad\sum_a\chi_a=1,\qquad\chi_a\ge0,\qquad D_\theta(\chi_a\circ\pi)=0. \tag*{(164)} χ a ( a , p ) = ∑ b η b ( a ) 2 ψ b ( p ) 2 η a ( a ) 2 ψ a ( p ) 2 , a ∑ χ a = 1 , χ a ≥ 0 , D θ ( χ a ∘ π ) = 0. (164)
Every required parameter derivative is bounded using the fixed cutoffs and s ∗ s_* s ∗ . The material label is held fixed. This does not reuse weights depending on a choice of oriented frame.
Fix B ≥ 1 B\ge1 B ≥ 1 bounding the full-disk inverse of W a W_a W a from (138) , and set
ω C = ( 4 B ) − 1 . (165) \omega_C=(4B)^{-1}. \tag*{(165)} ω C = ( 4 B ) − 1 . (165)
For ω F < ω C \omega_F<\omega_C ω F < ω C , Lemmas 5.2 and 5.3 , together with (162) , fix M l o c ≥ 1 M_{\mathrm{loc}}\ge1 M loc ≥ 1 , C b ≥ 1 C_b\ge1 C b ≥ 1 , and C o v ≥ 1 C_{\mathrm{ov}}\ge1 C ov ≥ 1 in the same compact quotient charts. Include all chart-norm comparison constants in them. Put
ω G = ( 2 M l o c C o v ) − 1 , ω t r = min { 1 , ω C , ω G } . (166) \omega_G=(2M_{\mathrm{loc}}C_{\mathrm{ov}})^{-1}, \qquad\omega_{\mathrm{tr}}=\min\{1,\omega_C,\omega_G\}. \tag*{(166)} ω G = ( 2 M loc C ov ) − 1 , ω tr = min { 1 , ω C , ω G } . (166)
Let C H ≥ 1 C_H\ge1 C H ≥ 1 be the boundary constant constructed in the proof of Proposition 5.4 . For transport and projection, set
ω ∗ = min { ω t r , C H − 1 } . (167) \omega_*=\min\{\omega_{\mathrm{tr}},C_H^{-1}\}. \tag*{(167)} ω ∗ = min { ω tr , C H − 1 } . (167)
This last fixed restriction only ensures that the boundary error parameter used in the projection is at most one.
On the fixed compact holomorphic spinor charts K a K_a K a , use the mixed norm
∥ U ∥ C x q C ζ r = max a max ∣ α ∣ ≤ q , ∣ β ∣ ≤ r sup x ∈ B R , ζ ∈ K a ∣ ∂ x α ∂ ( ℜ ζ , ℑ ζ ) β U a ( x , ζ ) ∣ F , (168) \|U\|_{C_x^q C_\zeta^r}=\max_a\max_{\lvert\alpha\rvert\leq q,\lvert\beta\rvert\leq r}\sup_{x\in B_R,\zeta\in K_a}\left\lvert\partial_x^\alpha\partial_{(\Re\zeta,\Im\zeta)}^\beta U_a(x,\zeta)\right\rvert_F, \tag*{(168)} ∥ U ∥ C x q C ζ r = a max ∣ α ∣ ≤ q , ∣ β ∣ ≤ r max x ∈ B R , ζ ∈ K a sup ∂ x α ∂ ( ℜ ζ , ℑ ζ ) β U a ( x , ζ ) F , (168)
where U a U_a U a is the local matrix and ∣ ⋅ ∣ F \lvert\cdot\rvert_F ∣ ⋅ ∣ F is the Frobenius norm. The spatial derivatives in (168) hold the spinor representative fixed.
The overlap estimate makes the leaf-constant weighted sum invertible under a fixed threshold, with mixed bounds for the subsequent direction projection.
Proposition 5.4. Under the fixed input of Subsection 5.1, suppose ∥ ( W a , ∂ ) − ∥ C 0 ≤ ω F < ω t r \|(W_{a,\partial})_-\|_{C^0}\leq\omega_F<\omega_{\mathrm{tr}} ∥ ( W a , ∂ ) − ∥ C 0 ≤ ω F < ω tr on the same family. Then there are global invertible quotient endomorphisms G G G and H = G − T H=G^{-T} H = G − T over B ‾ R × C P 1 \overline{B}_R\times\mathbb{CP}^1 B R × CP 1 satisfying
2 D θ G + A 1 G − G A 2 = 0 , 2 D θ H + Γ 1 H − H Γ 2 = 0 , 2D_\theta G+A_1G-GA_2=0,\qquad2D_\theta H+\Gamma_1H-H\Gamma_2=0, 2 D θ G + A 1 G − G A 2 = 0 , 2 D θ H + Γ 1 H − H Γ 2 = 0 , ∥ G − I ∥ C 0 ( ∂ B R × C P 1 ) + ∥ H − I ∥ C 0 ( ∂ B R × C P 1 ) ≤ C H ω F , \|G-I\|_{C^0(\partial B_R\times\mathbb{CP}^1)}+\|H-I\|_{C^0(\partial B_R\times\mathbb{CP}^1)}\leq C_H\omega_F, ∥ G − I ∥ C 0 ( ∂ B R × CP 1 ) + ∥ H − I ∥ C 0 ( ∂ B R × CP 1 ) ≤ C H ω F , ∥ G ∥ C x m C ζ 2 + ∥ G − 1 ∥ C x m C ζ 2 + ∥ H ∥ C x m C ζ 2 + ∥ H − 1 ∥ C x m C ζ 2 ≤ C m . (169) \|G\|_{C_x^mC_\zeta^2}+\|G^{-1}\|_{C_x^mC_\zeta^2}+\|H\|_{C_x^mC_\zeta^2}+\|H^{-1}\|_{C_x^mC_\zeta^2}\leq C_m. \tag*{(169)} ∥ G ∥ C x m C ζ 2 + ∥ G − 1 ∥ C x m C ζ 2 + ∥ H ∥ C x m C ζ 2 + ∥ H − 1 ∥ C x m C ζ 2 ≤ C m . (169)
Here the connections are defined in (61) , the threshold in (166) , and the norms in (168) . The norms use fixed compact holomorphic spinor charts and physical x x x derivatives. Every constant is uniform over the material class and independent of n n n , ς \varsigma ς , h h h , δ D \delta_D δ D .
Proof. Express each G a G_a G a in the same quotient chart before forming
G ( x , [ θ ] ) = ∑ a χ a ( a , π ( x , [ θ ] ) ) G a ( x , [ θ ] ) . (170) G(x,[\theta])=\sum_a\chi_a(a,\pi(x,[\theta]))G_a(x,[\theta]). \tag*{(170)} G ( x , [ θ ]) = a ∑ χ a ( a , π ( x , [ θ ])) G a ( x , [ θ ]) . (170)
Cutoff supports have margins, so each summand extends by zero off its patch with all required derivatives. Scalar weights commute with coordinate transitions; hence the sum is an endomorphism. The exact relative equation is preserved by (164) and linearity. At any parameter choose an active index a 0 a_0 a 0 . No differentiation of this choice is made. The full-disk comparison yields
∥ G − G a 0 ∥ ≤ C o v ω F , ∥ G a 0 − 1 ( G − G a 0 ) ∥ < 1 / 2 , \|G-G_{a_0}\|\leq C_{\mathrm{ov}}\omega_F,\qquad\|G_{a_0}^{-1}(G-G_{a_0})\|<1/2, ∥ G − G a 0 ∥ ≤ C ov ω F , ∥ G a 0 − 1 ( G − G a 0 ) ∥ < 1/2 ,
G − 1 = [ I + G a 0 − 1 ( G − G a 0 ) ] − 1 G a 0 − 1 , ∥ G − 1 ∥ ≤ 2 M l o c . (171) G^{-1}=[I+G_{a_0}^{-1}(G-G_{a_0})]^{-1}G_{a_0}^{-1},\qquad\|G^{-1}\|\leq2M_{\mathrm{loc}}. \tag*{(171)} G − 1 = [ I + G a 0 − 1 ( G − G a 0 ) ] − 1 G a 0 − 1 , ∥ G − 1 ∥ ≤ 2 M loc . (171)
It is this estimate, not convexity of invertible matrices, that proves invertibility. The boundary bound for G − I G-I G − I follows by averaging. The identity H − I = G − T ( I − G T ) H-I=G^{-T}(I-G^T) H − I = G − T ( I − G T ) gives the claimed boundary constant C H C_H C H , which can be fixed using 2 M l o c 2M_{\mathrm{loc}} 2 M loc before the actual error is chosen.
If R s u p < ∣ t ∣ < R 0 R_{\mathrm{sup}}<\lvert t\rvert<R_0 R sup < ∣ t ∣ < R 0 , the whole leaf has F ~ a = 0 \widetilde{F}_a=0 F a = 0 ; W a W_a W a is holomorphic on the disk, R C , a = W a R_{C,a}=W_a R C , a = W a , and G a = I G_a=I G a = I . Thus the sum joins exactly to G = I G=I G = I for ∣ t ∣ ≥ R 0 \lvert t\rvert\geq R_0 ∣ t ∣ ≥ R 0 on an open band. Near tangent leaves it is already the identity. No uniform estimate on a shrinking disk is asserted or needed. This joining band is a band of leaves; at nonzero noise it is not a claim that G = I G=I G = I at every point of the exterior spatial shell.
For the derivative estimates, set N = m + 2 N=m+2 N = m + 2 and use
x p ( w ) = t ν l e a f + r ( t ) ( ℜ w e 1 + ℑ w e 2 ) . (172) x_p(w)=t\nu_{\mathrm{leaf}}+r(t)(\Re w\,e_1+\Im w\,e_2). \tag*{(172)} x p ( w ) = t ν leaf + r ( t ) ( ℜ w e 1 + ℑ w e 2 ) . (172)
The boundary product has parameter derivatives through N N N and circle derivatives through N + 3 N+3 N + 3 . Each mixed input therefore costs at most
N + ( N + 3 ) = 2 m + 7 < r f r = 2 m + 8 , r f r + 6 = 2 m + 14. (173) N+(N+3)=2m+7<r_{\mathrm{fr}}=2m+8,\qquad r_{\mathrm{fr}}+6=2m+14. \tag*{(173)} N + ( N + 3 ) = 2 m + 7 < r fr = 2 m + 8 , r fr + 6 = 2 m + 14. (173)
Leibniz distributes these orders among the two frames; it does not require this many derivatives of each factor in addition. Lemma 5.2 supplies the separate orders N , N N,N N , N for R C R_C R C and its inverse.
To obtain the same rectangular bound for G a ♯ G^\sharp_a G a ♯ , the frame factors need at most 2 N = 2 m + 4 2N=2m+4 2 N = 2 m + 4 derivatives, also available.
At a fixed physical point,
p ( x , ζ ) = ( [ θ ( ζ ) ] , ν l e a f ( ζ ) ⋅ x ) , w ( x , ζ ) = e 1 ( ζ ) ⋅ x + i e 2 ( ζ ) ⋅ x r ( ν l e a f ( ζ ) ⋅ x ) . (174) p(x,\zeta)=([\theta(\zeta)],\nu_{\mathrm{leaf}}(\zeta)\cdot x),\qquad w(x,\zeta)=\frac{e_1(\zeta)\cdot x+ie_2(\zeta)\cdot x}{r(\nu_{\mathrm{leaf}}(\zeta)\cdot x)}. \tag*{(174)} p ( x , ζ ) = ([ θ ( ζ )] , ν leaf ( ζ ) ⋅ x ) , w ( x , ζ ) = r ( ν leaf ( ζ ) ⋅ x ) e 1 ( ζ ) ⋅ x + i e 2 ( ζ ) ⋅ x . (174)
The denominator is at least r ∗ r_* r ∗ . For ∣ α ∣ ≤ m |\alpha|\le m ∣ α ∣ ≤ m and β 1 + β 2 ≤ 2 \beta_1+\beta_2\le2 β 1 + β 2 ≤ 2 , the real multivariable chain rule bounds
∥ ∂ x α ∂ ζ 1 β 1 ∂ ζ 2 β 2 [ χ a ( a , p ( x , ζ ) ) G a ♯ ( p ( x , ζ ) , w ( x , ζ ) ) ] ∥ C 0 ≤ C m , g e o m max ∣ γ ∣ + j ≤ ∣ α ∣ + β 1 + β 2 ∥ ∂ ρ γ ∇ w j ( χ a G a ♯ ) ∥ C 0 . (175) \left\|\partial_x^\alpha\partial_{\zeta_1}^{\beta_1}\partial_{\zeta_2}^{\beta_2}\left[\chi_a(a,p(x,\zeta))G^\sharp_a(p(x,\zeta),w(x,\zeta))\right]\right\|_{C^0}
\le C_{m,\mathrm{geom}}\max_{|\gamma|+j\le|\alpha|+\beta_1+\beta_2}\left\|\partial_\rho^\gamma\nabla_w^j(\chi_aG^\sharp_a)\right\|_{C^0}. \tag*{(175)} ∂ x α ∂ ζ 1 β 1 ∂ ζ 2 β 2 [ χ a ( a , p ( x , ζ )) G a ♯ ( p ( x , ζ ) , w ( x , ζ )) ] C 0 ≤ C m , geom ∣ γ ∣ + j ≤ ∣ α ∣ + β 1 + β 2 max ∂ ρ γ ∇ w j ( χ a G a ♯ ) C 0 . (175)
The order on the right is at most m + 2 = N m+2=N m + 2 = N . Coordinate conversions and weights consume at most that order and have fixed smooth bounds. The finite sum therefore has the asserted mixed regularity. Repeatedly differentiating G G − 1 = I GG^{-1}=I G G − 1 = I uses only derivatives within the same rectangle and the zero-order bound (5.37); it controls G − 1 G^{-1} G − 1 , then H H H , H − 1 H^{-1} H − 1 . Finally, differentiating H = G − T H=G^{-T} H = G − T in the equation for G G G and using A j = − Γ j T A_j=-\Gamma_j^T A j = − Γ j T in the original matrix order, we obtain the equation for H H H in (5.35).
Remark. The global sum preserves the relative equation because its scalar weights are constant along the leaves. Its invertibility follows from the quantitative agreement of the local solutions.
For a fixed 0 < γ ′ < γ ∗ 0<\gamma'<\gamma_* 0 < γ ′ < γ ∗ , choose the Fourier constant C F ≥ 1 C_F\ge1 C F ≥ 1 in Lemma 5.1. One common smallness choice for the physical input, transport, and projection is
0 < η D < η ∗ : = min { η m o m , 1 , ( ω ∗ 2 C F ) 14 / γ ′ } , 0 < h < h ∗ . (176) 0<\eta_D<\eta_*:=\min\left\{\eta_{\mathrm{mom}},1,\left(\frac{\omega_*}{2C_F}\right)^{14/\gamma'}\right\},\qquad0<h<h_*. \tag*{(176)} 0 < η D < η ∗ := min { η mom , 1 , ( 2 C F ω ∗ ) 14/ γ ′ } , 0 < h < h ∗ . (176)
Define the Fourier error bound by
ω F = C F η D γ ′ / 14 < ω ∗ . (177) \omega_F=C_F\eta_D^{\gamma'/14}<\omega_*. \tag*{(177)} ω F = C F η D γ ′ /14 < ω ∗ . (177)
Here η D \eta_D η D and ω ∗ \omega_* ω ∗ are defined in (1.17) and (5.33). The number γ ′ \gamma' γ ′ and then all constants and thresholds are fixed before varying the measured error or h h h . An additional rigidity threshold is imposed only after the transport and projection estimates have been proved; see (6.25). At δ D = 0 \delta_D=0 δ D = 0 the same fixed family is used for all small h h h ; η D = h ↓ 0 \eta_D=h\downarrow0 η D = h ↓ 0 . This observation supplies numerical error bounds tending to zero and does not use a uniqueness theorem or log 0 \log0 log 0 .
Holomorphic projection in the null direction
At each fixed physical point x x x , we project the global endomorphism H = G − T H=G^{-T} H = G − T onto endomorphisms holomorphic in the complex null direction. The projected field H 0 = Π H H_0=\Pi H H 0 = Π H has a finite-dimensional block form. Direction differentiation and spatial interpolation bound H − H 0 H-H_0 H − H 0 ; the first-order transport equation then gives the residual required for material recovery. No invertibility of H 0 H_0 H 0 is required.
A fixed nine-dimensional endomorphism space
The fixed quotient bundle has a finite-dimensional space of holomorphic endomorphisms. Its explicit block structure determines the projection used for coefficient recovery.
Use homogeneous spinor coordinates [ s 0 : s 1 ] [s_0:s_1] [ s 0 : s 1 ] , with representatives ( 1 , ζ ) (1,\zeta) ( 1 , ζ ) and ( ν , 1 ) (\nu,1) ( ν , 1 ) in the two affine charts. The dual quotient coordinates have the transition D t r D_{\mathrm{tr}} D tr defined in (2.40). With the convention that the tautological frames satisfy e 0 = ζ e ∞ e_0=\zeta e_\infty e 0 = ζ e ∞ , this gives
E ∗ = O ( − 1 ) ⊕ 2 ⊕ O , E = End ( E ∗ ) , E^*=\mathcal{O}(-1)^{\oplus2}\oplus\mathcal{O},\qquad\mathcal{E}=\operatorname{End}(E^*), E ∗ = O ( − 1 ) ⊕ 2 ⊕ O , E = End ( E ∗ ) ,
E = O ⊕ 5 ⊕ O ( − 1 ) ⊕ 2 ⊕ O ( 1 ) ⊕ 2 . (178) \mathcal{E}=\mathcal{O}^{\oplus5}\oplus\mathcal{O}(-1)^{\oplus2}\oplus\mathcal{O}(1)^{\oplus2}. \tag*{(178)} E = O ⊕ 5 ⊕ O ( − 1 ) ⊕ 2 ⊕ O ( 1 ) ⊕ 2 . (178)
This fixed bundle is independent of the material pair, x x x , and noise. The transition law restricts the degrees of the holomorphic blocks.
Lemma 6.1. For the bundle E \mathcal{E} E in (6.1), define
H = H 0 ( C P 1 , E ) . (179) \mathcal{H}=H^{0}(\mathbb{CP}^{1},\mathcal{E}). \tag*{(179)} H = H 0 ( CP 1 , E ) . (179) This space is exactly
{ ( A H 0 s 0 C H , 0 + s 1 C H , 1 d H ) : A H ∈ C 2 × 2 , d H ∈ C , C H , 0 , C H , 1 ∈ C 1 × 2 } , dim C H = 9. (180) \left\{\left(\begin{array}{cc}
A_{H} & 0 \\
s_{0}C_{H,0}+s_{1}C_{H,1} & d_{H}
\end{array}\right): A_{H}\in\mathbb{C}^{2\times2},\ d_{H}\in\mathbb{C},\ C_{H,0},C_{H,1}\in\mathbb{C}^{1\times2}\right\},\qquad\dim_{\mathbb{C}}\mathcal{H}=9. \tag*{(180)} { ( A H s 0 C H , 0 + s 1 C H , 1 0 d H ) : A H ∈ C 2 × 2 , d H ∈ C , C H , 0 , C H , 1 ∈ C 1 × 2 } , dim C H = 9. (180) Here the matrix is evaluated in the local spinor representative; it is not a single trivialization over projective space. Neither A H A_{H} A H nor d H d_{H} d H is assumed invertible.
Proof. For an arbitrary holomorphic endomorphism write its chart blocks as A 0 A_{0} A 0 , B 0 B_{0} B 0 , C 0 a f f C_{0}^{\mathrm{aff}} C 0 aff , d 0 d_{0} d 0 and A ∞ A_{\infty} A ∞ , B ∞ B_{\infty} B ∞ , C ∞ a f f C_{\infty}^{\mathrm{aff}} C ∞ aff , d ∞ d_{\infty} d ∞ . The transition law H ( 0 ) = D t r − 1 H ( ∞ ) D t r H^{(0)}=D_{\mathrm{tr}}^{-1}H^{(\infty)}D_{\mathrm{tr}} H ( 0 ) = D tr − 1 H ( ∞ ) D tr reads
A 0 ( ζ ) = A ∞ ( 1 / ζ ) , d 0 ( ζ ) = d ∞ ( 1 / ζ ) , A_{0}(\zeta)=A_{\infty}(1/\zeta),\qquad d_{0}(\zeta)=d_{\infty}(1/\zeta), A 0 ( ζ ) = A ∞ ( 1/ ζ ) , d 0 ( ζ ) = d ∞ ( 1/ ζ ) ,
B 0 ( ζ ) = ζ − 1 B ∞ ( 1 / ζ ) , C 0 a f f ( ζ ) = ζ C ∞ a f f ( 1 / ζ ) . (181) B_{0}(\zeta)=\zeta^{-1}B_{\infty}(1/\zeta),\qquad C_{0}^{\mathrm{aff}}(\zeta)=\zeta C_{\infty}^{\mathrm{aff}}(1/\zeta). \tag*{(181)} B 0 ( ζ ) = ζ − 1 B ∞ ( 1/ ζ ) , C 0 aff ( ζ ) = ζ C ∞ aff ( 1/ ζ ) . (181)
The first and last diagonal blocks are entire and bounded at infinity, hence constant. The entire upper-right column is O ( ∣ ζ ∣ − 1 ) O(|\zeta|^{-1}) O ( ∣ ζ ∣ − 1 ) and vanishes. The lower-left row is entire with at most linear growth. Cauchy’s estimates on expanding circles give C 0 a f f ( ζ ) = C H , 0 + ζ C H , 1 C_{0}^{\mathrm{aff}}(\zeta)=C_{H,0}+\zeta C_{H,1} C 0 aff ( ζ ) = C H , 0 + ζ C H , 1 and C ∞ a f f ( υ ) = υ C H , 0 + C H , 1 C_{\infty}^{\mathrm{aff}}(\upsilon)=\upsilon C_{H,0}+C_{H,1} C ∞ aff ( υ ) = υ C H , 0 + C H , 1 . These formulas are precisely the homogeneous row in (6.3). Conversely these matrices satisfy the transitions and are holomorphic on both charts. Counting their free coefficients gives 4 + 1 + 2 + 2 = 9 4+1+2+2=9 4 + 1 + 2 + 2 = 9 .
Fix smooth Hermitian metrics on E \mathcal{E} E and C P 1 \mathbb{CP}^{1} CP 1 and let Π \Pi Π be the L 2 L^{2} L 2 orthogonal projection onto H \mathcal{H} H . For a fixed orthonormal basis e 1 , … , e 9 e_{1},\ldots,e_{9} e 1 , … , e 9 ,
Π u = ∑ a = 1 9 ( ∫ C P 1 ⟨ u , e a ⟩ d V ) e a , Π I = I , (182) \Pi u=\sum_{a=1}^{9}\left(\int_{\mathbb{CP}^{1}}\langle u,e_{a}\rangle\,dV\right)e_{a},\qquad\Pi I=I, \tag*{(182)} Π u = a = 1 ∑ 9 ( ∫ CP 1 ⟨ u , e a ⟩ d V ) e a , Π I = I , (182)
with inner product linear in its first entry. This is a fixed linear operator, not a projection depending on x x x or on the material pair.
Lemma 6.2. Let Σ = C P 1 \Sigma=\mathbb{CP}^{1} Σ = CP 1 and let E \mathcal{E} E be the bundle in (6.1). For every integer N ≥ 0 N\geq0 N ≥ 0 , every p > 2 p>2 p > 2 , and every u ∈ W 1 , p ( Σ ; E ) u\in W^{1,p}(\Sigma;\mathcal{E}) u ∈ W 1 , p ( Σ ; E ) , one has
∥ Π u ∥ C N ≤ C N ∥ u ∥ C 0 , ∥ u − Π u ∥ W 1 , p ≤ C p ∥ ∂ ‾ u ∥ L p . (183) \|\Pi u\|_{C^{N}}\leq C_{N}\|u\|_{C^{0}},\qquad\|u-\Pi u\|_{W^{1,p}}\leq C_{p}\|\overline{\partial}u\|_{L^{p}}. \tag*{(183)} ∥Π u ∥ C N ≤ C N ∥ u ∥ C 0 , ∥ u − Π u ∥ W 1 , p ≤ C p ∥ ∂ u ∥ L p . (183) Here Π \Pi Π is the projection in (6.5). For a family u = u ( x , ⋅ ) u=u(x,\cdot) u = u ( x , ⋅ ) , the projection commutes with physical-space derivatives whenever these derivatives exist in the indicated spaces.
Proof. Set Σ = C P 1 \Sigma=\mathbb{CP}^{1} Σ = CP 1 and H = H 0 ( Σ , E ) \mathcal{H}=H^{0}(\Sigma,\mathcal{E}) H = H 0 ( Σ , E ) . By the finite-rank representation (6.5), the smoothness of the fixed basis { e a } a = 1 9 \{e_{a}\}_{a=1}^{9} { e a } a = 1 9 , and Hölder’s inequality,
∥ Π u ∥ C N ≤ C N ∑ a = 1 9 ∣ ∫ Σ ⟨ u , e a ⟩ d V ∣ ≤ C N ∥ u ∥ C 0 , \|\Pi u\|_{C^{N}}\leq C_{N}\sum_{a=1}^{9}\left|\int_{\Sigma}\langle u,e_{a}\rangle\,dV\right|\leq C_{N}\|u\|_{C^{0}}, ∥Π u ∥ C N ≤ C N a = 1 ∑ 9 ∫ Σ ⟨ u , e a ⟩ d V ≤ C N ∥ u ∥ C 0 ,
∥ Π v ∥ L p ≤ C p ∥ v ∥ L p , v ∈ L p ( Σ ; E ) . \|\Pi v\|_{L^{p}}\leq C_{p}\|v\|_{L^{p}},\qquad v\in L^{p}(\Sigma;\mathcal{E}). ∥Π v ∥ L p ≤ C p ∥ v ∥ L p , v ∈ L p ( Σ ; E ) .
Since Π \Pi Π is independent of the physical variable x x x , it also satisfies
∂ x γ ( Π u ) = Π ( ∂ x γ u ) \partial_{x}^{\gamma}(\Pi u)=\Pi(\partial_{x}^{\gamma}u) ∂ x γ ( Π u ) = Π ( ∂ x γ u )
whenever the indicated derivatives exist.
The elliptic estimate for the Dolbeault operator on the fixed holomorphic bundle E \mathcal{E} E over Σ \Sigma Σ [26 Lemma C.2.1] gives
∥ v ∥ W 1 , p ≤ C p ( ∥ ∂ ˉ v ∥ L p + ∥ v ∥ L p ) . (184) \lVert v\rVert_{W^{1,p}} \leq C_p\left(\lVert\bar{\partial}v\rVert_{L^p}+\lVert v\rVert_{L^p}\right). \tag*{(184)} ∥ v ∥ W 1 , p ≤ C p ( ∥ ∂ ˉ v ∥ L p + ∥ v ∥ L p ) . (184)
Initially valid for smooth sections, this estimate extends to W 1 , p ( Σ ; E ) W^{1,p}(\Sigma;\mathcal{E}) W 1 , p ( Σ ; E ) by density and the continuity of ∂ ˉ : W 1 , p → L p \bar{\partial}:W^{1,p}\to L^p ∂ ˉ : W 1 , p → L p .
We claim that
∥ v ∥ W 1 , p ≤ C p ∥ ∂ ˉ v ∥ L p , v ∈ W 1 , p ( Σ ; E ) , Π v = 0. \lVert v\rVert_{W^{1,p}}\leq C_p\lVert\bar{\partial}v\rVert_{L^p},\qquad v\in W^{1,p}(\Sigma;\mathcal{E}),\quad\Pi v=0. ∥ v ∥ W 1 , p ≤ C p ∥ ∂ ˉ v ∥ L p , v ∈ W 1 , p ( Σ ; E ) , Π v = 0.
Otherwise, there exists a sequence v k ∈ W 1 , p ( Σ ; E ) v_k\in W^{1,p}(\Sigma;\mathcal{E}) v k ∈ W 1 , p ( Σ ; E ) such that
Π v k = 0 , ∥ v k ∥ W 1 , p = 1 , ∥ ∂ ˉ v k ∥ L p ⟶ 0. \Pi v_k=0,\qquad\lVert v_k\rVert_{W^{1,p}}=1,\qquad\lVert\bar{\partial}v_k\rVert_{L^p}\longrightarrow0. Π v k = 0 , ∥ v k ∥ W 1 , p = 1 , ∥ ∂ ˉ v k ∥ L p ⟶ 0.
By the compact embedding W 1 , p ( Σ ; E ) ↪ L p ( Σ ; E ) W^{1,p}(\Sigma;\mathcal{E})\hookrightarrow L^p(\Sigma;\mathcal{E}) W 1 , p ( Σ ; E ) ↪ L p ( Σ ; E ) , after passing to a subsequence,
v k ⇀ v in W 1 , p , v k ⟶ v in L p . v_k\rightharpoonup v\quad\text{in }W^{1,p},\qquad v_k\longrightarrow v\quad\text{in }L^p. v k ⇀ v in W 1 , p , v k ⟶ v in L p .
Consequently,
∂ ˉ v = 0 , Π v = 0. \bar{\partial}v=0,\qquad\Pi v=0. ∂ ˉ v = 0 , Π v = 0.
The first identity holds distributionally. Elliptic regularity for ∂ ˉ \bar{\partial} ∂ ˉ implies v ∈ H v\in\mathcal{H} v ∈ H , and hence
It follows from (184) that
1 = ∥ v k ∥ W 1 , p ≤ C p ( ∥ ∂ ˉ v k ∥ L p + ∥ v k ∥ L p ) ⟶ 0 , 1=\lVert v_k\rVert_{W^{1,p}}\leq C_p\left(\lVert\bar{\partial}v_k\rVert_{L^p}+\lVert v_k\rVert_{L^p}\right)\longrightarrow0, 1 = ∥ v k ∥ W 1 , p ≤ C p ( ∥ ∂ ˉ v k ∥ L p + ∥ v k ∥ L p ) ⟶ 0 ,
a contradiction.
Finally, for w = u − Π u w=u-\Pi u w = u − Π u we have
Π w = 0 , ∂ ˉ w = ∂ ˉ u , \Pi w=0,\qquad\bar{\partial}w=\bar{\partial}u, Π w = 0 , ∂ ˉ w = ∂ ˉ u ,
because Π u ∈ H \Pi u\in\mathcal{H} Π u ∈ H . Therefore,
∥ u − Π u ∥ W 1 , p = ∥ w ∥ W 1 , p ≤ C p ∥ ∂ ˉ w ∥ L p = C p ∥ ∂ ˉ u ∥ L p . \lVert u-\Pi u\rVert_{W^{1,p}}=\lVert w\rVert_{W^{1,p}}\leq C_p\lVert\bar{\partial}w\rVert_{L^p}=C_p\lVert\bar{\partial}u\rVert_{L^p}. ∥ u − Π u ∥ W 1 , p = ∥ w ∥ W 1 , p ≤ C p ∥ ∂ ˉ w ∥ L p = C p ∥ ∂ ˉ u ∥ L p .
This proves (183) .
Direction differentiation at a fixed physical point
Direction differentiation is taken with the physical point x x x fixed. The boundary deviation of H H H from the identity controls its difference from the holomorphic projection H 0 = Π H H_0=\Pi H H 0 = Π H .
The formulas (56) and (61) are polynomial in the homogeneous spinor representative. Their material coefficients are independent of direction. Therefore, in the holomorphic charts (57) ,
∂ ζ ˉ θ = 0 , ∂ ζ ˉ Γ j = 0. (185) \partial_{\bar{\zeta}}\theta=0,\qquad\partial_{\bar{\zeta}}\Gamma_j=0. \tag*{(185)} ∂ ζ ˉ θ = 0 , ∂ ζ ˉ Γ j = 0. (185)
These derivatives hold physical x x x fixed. Define the relative transport operator and the projected field by
T ( U ) = 2 D θ U + Γ 1 U − U Γ 2 , H 0 = Π H , (186) \mathcal{T}(U)=2D_\theta U+\Gamma_1U-U\Gamma_2,\qquad H_0=\Pi H, \tag*{(186)} T ( U ) = 2 D θ U + Γ 1 U − U Γ 2 , H 0 = Π H , (186)
where Π \Pi Π is defined in (182) .
We propagate the boundary direction defect through the leaf disks and use spatial interpolation to estimate H − H 0 H-H_0 H − H 0 .
Proposition 6.3. Let m > 3 m>3 m > 3 be an integer. Suppose a jointly C 2 C^2 C 2 endomorphism H H H on B R ‾ × C P 1 \overline{B_R}\times\mathbb{CP}^1 B R × CP 1 satisfies
T ( H ) = 0 (187) \mathcal{T}(H)=0 \tag*{(187)} T ( H ) = 0 (187) with T \mathcal{T} T defined in (6.9), and the fixed-physical-point bounds
∥ H ∥ C x m C ζ 0 + ∥ H ∥ C x 0 C ζ 2 ≤ M H , ∥ H − I ∥ C x 0 ( ∂ B R × C P 1 ) ≤ ω ≤ 1. (188) \lVert H\rVert_{C_x^m C_\zeta^0}+\lVert H\rVert_{C_x^0 C_\zeta^2}\leq M_H,\qquad\lVert H-I\rVert_{C_x^0(\partial B_R\times\mathbb{CP}^1)}\leq\omega\leq1. \tag*{(188)} ∥ H ∥ C x m C ζ 0 + ∥ H ∥ C x 0 C ζ 2 ≤ M H , ∥ H − I ∥ C x 0 ( ∂ B R × CP 1 ) ≤ ω ≤ 1. (188) On each leaf used to propagate a nonzero defect suppose the dual frames Y 1 , Y 2 Y_1,Y_2 Y 1 , Y 2 and their inverses exist and have uniform bounds on its whole closed disk. Then H 0 H_0 H 0 from (6.9) is a holomorphic endomorphism in direction and
∥ ∂ ˉ H ∥ C x 0 C ζ 0 + ∥ H − H 0 ∥ C x 0 C ζ 0 ≤ C ω 1 / 2 , \lVert\bar{\partial}H\rVert_{C_x^0 C_\zeta^0}+\lVert H-H_0\rVert_{C_x^0 C_\zeta^0}\leq C\omega^{1/2}, ∥ ∂ ˉ H ∥ C x 0 C ζ 0 + ∥ H − H 0 ∥ C x 0 C ζ 0 ≤ C ω 1/2 , ∥ H − H 0 ∥ C x ℓ C ζ 0 ≤ C m , ℓ ω ( m − ℓ ) / ( 2 m ) ( 0 ≤ ℓ < m ) , \lVert H-H_0\rVert_{C_x^\ell C_\zeta^0}\leq C_{m,\ell}\omega^{(m-\ell)/(2m)}\qquad(0\leq\ell<m), ∥ H − H 0 ∥ C x ℓ C ζ 0 ≤ C m , ℓ ω ( m − ℓ ) / ( 2 m ) ( 0 ≤ ℓ < m ) , ∥ H 0 − I ∥ C x 0 ( ∂ B R × C P 1 ) ≤ C ω , ∥ H 0 ∥ C x m C ζ 0 ≤ C M H . (189) \lVert H_0-I\rVert_{C_x^0(\partial B_R\times\mathbb{CP}^1)}\leq C\omega,\qquad\lVert H_0\rVert_{C_x^m C_\zeta^0}\leq CM_H. \tag*{(189)} ∥ H 0 − I ∥ C x 0 ( ∂ B R × CP 1 ) ≤ C ω , ∥ H 0 ∥ C x m C ζ 0 ≤ C M H . (189) The constants depend only on the fixed atlas, bundle, ball, frame bounds and M H M_H M H . No invertibility of H 0 H_0 H 0 is assumed or concluded.
Proof. Take compact direction charts with fixed larger charts around them. For fixed x ∈ ∂ B R x\in\partial B_R x ∈ ∂ B R , a real first difference with step a a a and its second-order Taylor remainder yield
∥ ∂ ˉ ζ ( H − I ) ∥ C ζ 0 ≤ C ( ω / a + M H a ) . (190) \lVert\bar{\partial}_\zeta(H-I)\rVert_{C_\zeta^0}\leq C(\omega/a+M_Ha). \tag*{(190)} ∥ ∂ ˉ ζ ( H − I ) ∥ C ζ 0 ≤ C ( ω / a + M H a ) . (190)
For small positive ω \omega ω , use a a a comparable to ω / M H \sqrt{\omega/M_H} ω / M H within the fixed chart margin. For the remaining ω ≤ 1 \omega\leq1 ω ≤ 1 , enlarge a constant depending on that margin and M H M_H M H . If ω = 0 \omega=0 ω = 0 , the boundary section is identically I I I in direction and its direction derivative is zero. Thus define the direction defect by
V = ∂ ˉ ζ H . (191) V=\bar{\partial}_\zeta H. \tag*{(191)} V = ∂ ˉ ζ H . (191)
Its boundary norm is at most C ω 1 / 2 C\omega^{1/2} C ω 1/2 . Differentiate (6.10) at fixed physical x x x . By (6.8), no derivative of θ \theta θ or Γ j \Gamma_j Γ j appears, and
2 D θ V + Γ 1 V − V Γ 2 = 0. (192) 2D_\theta V+\Gamma_1V-V\Gamma_2=0. \tag*{(192)} 2 D θ V + Γ 1 V − V Γ 2 = 0. (192)
This is not differentiation at fixed moving leaf coordinates. The transition formula (2.51) applies to this derivative. Consequently these coefficients are the components of a well-defined bundle-valued ( 0 , 1 ) (0,1) ( 0 , 1 ) -form. By the dual-frame equation (2.45), direct multiplication yields
D θ ( Y 1 − 1 V Y 2 ) = 0. (193) D_\theta(Y_1^{-1}VY_2)=0. \tag*{(193)} D θ ( Y 1 − 1 V Y 2 ) = 0. (193)
The holomorphic representative of θ \theta θ is a nonzero complex multiple of the normalized direction of the same leaf. Hence this is a leaf Cauchy–Riemann equation. The scalar maximum principle applied to constant-vector matrix elements, and the four frame factors, give ∥ V ∥ B θ , t ≤ C ∥ V ∥ ∂ B θ , t \lVert V\rVert_{B_{\theta,t}}\leq C\lVert V\rVert_{\partial B_{\theta,t}} ∥ V ∥ B θ , t ≤ C ∥ V ∥ ∂ B θ , t . The boundary circle lies on ∂ B R \partial B_R ∂ B R , so the first defect bound follows. Where the construction has H = I H=I H = I on an open neighborhood, the defect is already zero.
For each x x x , Lemma 6.2 and the fixed embedding W 1 , p ( C P 1 ) ⊂ C 0 W^{1,p}(\mathbb{CP}^1)\subset C^0 W 1 , p ( CP 1 ) ⊂ C 0 , p > 2 p>2 p > 2 , imply ∥ H − H 0 ∥ C x 0 C ζ 0 ≤ C ω 1 / 2 \lVert H-H_0\rVert_{C_x^0 C_\zeta^0}\leq C\omega^{1/2} ∥ H − H 0 ∥ C x 0 C ζ 0 ≤ C ω 1/2 . Since Π \Pi Π commutes with ∂ x \partial_x ∂ x , the independent spatial prior gives ∥ H − H 0 ∥ C x m C ζ 0 ≤ C M H \lVert H-H_0\rVert_{C_x^m C_\zeta^0}\leq CM_H ∥ H − H 0 ∥ C x m C ζ 0 ≤ C M H . On the fixed ball, a simultaneous bounded extension in C 0 C^0 C 0 and C m C^m C m , smoothing, and optimization of the smoothing scale give
∥ f ∥ C ℓ ≤ C ( ∥ f ∥ C 0 1 − ℓ / m ∥ f ∥ C m ℓ / m + ∥ f ∥ C 0 ) , 0 ≤ ℓ < m . (194) \lVert f\rVert_{C^\ell}\leq C\left(\lVert f\rVert_{C^0}^{1-\ell/m}\lVert f\rVert_{C^m}^{\ell/m}+\lVert f\rVert_{C^0}\right),\qquad0\leq\ell<m. \tag*{(194)} ∥ f ∥ C ℓ ≤ C ( ∥ f ∥ C 0 1 − ℓ / m ∥ f ∥ C m ℓ / m + ∥ f ∥ C 0 ) , 0 ≤ ℓ < m . (194)
One may obtain the extension in boundary charts by a fixed finite linear combination of reflected values, matching derivatives up to m m m , followed by fixed cutoffs. Its coefficients and charts are independent of f f f ; no extra derivative of f f f is used. Apply this inequality to each local matrix entry uniformly in direction. This proves the spatial exponent in (189) without ever estimating ∂ x m ∂ ˉ ζ H \partial_x^m\bar{\partial}_\zeta H ∂ x m ∂ ˉ ζ H . Finally H 0 − I = Π ( H − I ) H_0-I=\Pi(H-I) H 0 − I = Π ( H − I ) on the sphere gives the boundary estimate without a square-root loss.
Remark. For the construction in Proposition 5.4, the full-disk frame bounds are required only for ∣ t ∣ ≤ R 0 |t|\le R_0 ∣ t ∣ ≤ R 0 ; on the remaining open region H = I H=I H = I . Taking three spatial derivatives of H − H 0 H-H_0 H − H 0 gives a positive error power when m > 3 m>3 m > 3 and a small C x 2 C_x^2 C x 2 transport residual, as quantified in Subsection 6.3.
The projection-to-rigidity estimate
The projection estimate is converted into a transport residual in the C x 2 C_x^2 C x 2 norm required by material rigidity.
Use β m \beta_m β m from (1.17) and apply Proposition 6.3 to the endomorphism H = G − T H=G^{-T} H = G − T constructed in Proposition 5.4, with ω = C H ω F ≤ 1 \omega=C_H\omega_F\le1 ω = C H ω F ≤ 1 and a fixed high-order bound M H M_H M H . For all spatial matrix norms use the maximum supremum Frobenius norm over the same compact spinor charts and the displayed spatial multi-indices, as in Theorem 7.3. Fix C Π ≥ 1 C_\Pi\ge1 C Π ≥ 1 bounding Π : C ζ 0 → C ζ 0 \Pi:C_\zeta^0\to C_\zeta^0 Π : C ζ 0 → C ζ 0 in these norms. If C m , 3 C_{m,3} C m , 3 denotes the corresponding constant in (189) , one may choose C P = C m , 3 C H β m C_P=C_{m,3}C_H^{\beta_m} C P = C m , 3 C H β m . Then
∥ H − H 0 ∥ C x 3 C ζ 0 ≤ C P ω F β m , ∥ H 0 − I ∥ C 0 ( ∂ B R × C P 1 ) ≤ C Π C H ω F . (195) \lVert H-H_0\rVert_{C_x^3C_\zeta^0}\le C_P\omega_F^{\beta_m},\qquad\lVert H_0-I\rVert_{C^0(\partial B_R\times\mathbb{CP}^1)}\le C_\Pi C_H\omega_F. \tag*{(195)} ∥ H − H 0 ∥ C x 3 C ζ 0 ≤ C P ω F β m , ∥ H 0 − I ∥ C 0 ( ∂ B R × CP 1 ) ≤ C Π C H ω F . (195)
The projected object is a holomorphic endomorphism with the block form of Lemma 6.1. Projection alone is not used to claim that it is an automorphism.
Define the projection error by
E 0 = H 0 − H , (196) E_0=H_0-H, \tag*{(196)} E 0 = H 0 − H , (196)
where H 0 H_0 H 0 is defined in (186) . By (187) ,
T ( H 0 ) = 2 θ ⋅ ∇ x E 0 + Γ 1 E 0 − E 0 Γ 2 . (197) T(H_0)=2\theta\cdot\nabla_xE_0+\Gamma_1E_0-E_0\Gamma_2. \tag*{(197)} T ( H 0 ) = 2 θ ⋅ ∇ x E 0 + Γ 1 E 0 − E 0 Γ 2 . (197)
For every ∣ α ∣ ≤ 2 |\alpha|\le2 ∣ α ∣ ≤ 2 , differentiation gives
∂ x α T ( H 0 ) = 2 θ ⋅ ∇ x ∂ x α E 0 + ∑ σ ≤ α ( α σ ) [ ( ∂ x σ Γ 1 ) ( ∂ x α − σ E 0 ) − ( ∂ x σ E 0 ) ( ∂ x α − σ Γ 2 ) ] . (198) \partial_x^\alpha T(H_0)=2\theta\cdot\nabla_x\partial_x^\alpha E_0+
\sum_{\sigma\le\alpha}\binom{\alpha}{\sigma}\left[(\partial_x^\sigma\Gamma_1)(\partial_x^{\alpha-\sigma}E_0)-(\partial_x^\sigma E_0)(\partial_x^{\alpha-\sigma}\Gamma_2)\right]. \tag*{(198)} ∂ x α T ( H 0 ) = 2 θ ⋅ ∇ x ∂ x α E 0 + σ ≤ α ∑ ( σ α ) [ ( ∂ x σ Γ 1 ) ( ∂ x α − σ E 0 ) − ( ∂ x σ E 0 ) ( ∂ x α − σ Γ 2 ) ] . (198)
Thus only three spatial derivatives of E 0 E_0 E 0 and two of the connections are needed. Set
Θ 1 = max charts , ζ ∑ i = 1 3 ∣ θ i ( ζ ) ∣ , B Γ , j ≥ ∥ Γ j ∥ C x 2 C ζ 0 , C T = 2 Θ 1 + 4 ( B Γ , 1 + B Γ , 2 ) . (199) \Theta_1=\max_{\text{charts},\zeta}\sum_{i=1}^{3}|\theta_i(\zeta)|,\qquad B_{\Gamma,j}\ge\lVert\Gamma_j\rVert_{C_x^2C_\zeta^0},\qquad C_T=2\Theta_1+4(B_{\Gamma,1}+B_{\Gamma,2}). \tag*{(199)} Θ 1 = charts , ζ max i = 1 ∑ 3 ∣ θ i ( ζ ) ∣ , B Γ , j ≥ ∥ Γ j ∥ C x 2 C ζ 0 , C T = 2 Θ 1 + 4 ( B Γ , 1 + B Γ , 2 ) . (199)
Choose B Γ , j B_{\Gamma,j} B Γ , j uniformly over the full material class. The factor 4 4 4 is max ∣ α ∣ ≤ 2 ∑ σ ≤ α ( α σ ) \max_{|\alpha|\le2}\sum_{\sigma\le\alpha}\binom{\alpha}{\sigma} max ∣ α ∣ ≤ 2 ∑ σ ≤ α ( σ α ) ; submultiplicativity of the Frobenius norm in (198) therefore gives ∥ T ( E 0 ) ∥ C x 2 C ζ 0 ≤ C T ∥ E 0 ∥ C x 3 C ζ 0 \lVert T(E_0)\rVert_{C_x^2C_\zeta^0}\le C_T\lVert E_0\rVert_{C_x^3C_\zeta^0} ∥ T ( E 0 ) ∥ C x 2 C ζ 0 ≤ C T ∥ E 0 ∥ C x 3 C ζ 0 . In particular all constants can be fixed before the error by
B H = C Π M H , C l i n k = max { 1 , C T C P + C Π C H } . (200) B_H=C_\Pi M_H,\qquad C_{\mathrm{link}}=\max\{1,C_TC_P+C_\Pi C_H\}. \tag*{(200)} B H = C Π M H , C link = max { 1 , C T C P + C Π C H } . (200)
Here ∥ H 0 ∥ C x 3 C ζ 0 ≤ B H \lVert H_0\rVert_{C_x^3C_\zeta^0}\le B_H ∥ H 0 ∥ C x 3 C ζ 0 ≤ B H , since m ≥ 4 m\ge4 m ≥ 4 and Π \Pi Π commutes with the spatial derivatives. These explicit choices give
∥ T ( H 0 ) ∥ C x 2 C ζ 0 + ∥ H 0 − I ∥ C 0 ( ∂ B R × C P 1 ) ≤ C l i n k ω F β m ≤ C l i n k C F β m η D γ ′ ( m − 3 ) / ( 28 m ) . (201) \lVert T(H_0)\rVert_{C_x^2C_\zeta^0}+\lVert H_0-I\rVert_{C^0(\partial B_R\times\mathbb{CP}^1)}
\le C_{\mathrm{link}}\omega_F^{\beta_m}
\le C_{\mathrm{link}}C_F^{\beta_m}\eta_D^{\gamma'(m-3)/(28m)}. \tag*{(201)} ∥ T ( H 0 ) ∥ C x 2 C ζ 0 + ∥ H 0 − I ∥ C 0 ( ∂ B R × CP 1 ) ≤ C link ω F β m ≤ C link C F β m η D γ ′ ( m − 3 ) / ( 28 m ) . (201)
Here ω F ≤ 1 \omega_F \le1 ω F ≤ 1 was used to absorb its linear contribution. With the rigidity threshold for this fixed B H B_H B H , one sufficient simultaneous choice is
ω a l l = min { 1 , ω C , ω G , C H − 1 , ( ε r i g , 0 2 C l i n k ) 1 / β m } , η a l l = min { η m o m , 1 , ( ω a l l 2 C F ) 14 / γ ′ } , 0 < η D < η a l l , 0 < h < h ∗ . (202) \begin{aligned}
\omega_{\mathrm{all}} &= \min\left\{1,\omega_C,\omega_G,C_H^{-1},\left(\frac{\varepsilon_{\mathrm{rig},0}}{2C_{\mathrm{link}}}\right)^{1/\beta_m}\right\},\\
\eta_{\mathrm{all}} &= \min\left\{\eta_{\mathrm{mom}},1,\left(\frac{\omega_{\mathrm{all}}}{2C_F}\right)^{14/\gamma'}\right\}, \qquad0<\eta_D<\eta_{\mathrm{all}},\qquad0<h<h_*. \tag*{(202)}
\end{aligned} ω all η all = min { 1 , ω C , ω G , C H − 1 , ( 2 C link ε rig , 0 ) 1/ β m } , = min { η mom , 1 , ( 2 C F ω all ) 14/ γ ′ } , 0 < η D < η all , 0 < h < h ∗ . (202)
Then ω F < ω a l l \omega_F<\omega_{\mathrm{all}} ω F < ω all and the sum of the residual and sphere errors is less than ε r i g , 0 / 2 \varepsilon_{\mathrm{rig},0}/2 ε rig , 0 /2 . Theorem 7.3 therefore applies. The shell condition is imposed on the materials; the projected endomorphism H 0 H_0 H 0 is normalized by the sphere estimate in (195) .
Quantitative material rigidity and common-exterior stability
The holomorphic projection reduces coefficient recovery to a finite system of block equations. We first establish the scalar coercive estimate used in those equations, then prove material rigidity and complete the common-exterior stability argument.
Uniform coercivity for the trace-free Hessian
A fixed shell normalization gives a coercive estimate for the trace-free Hessian with bounded first-order drift. This estimate controls the logarithmic shear-modulus ratio in the block recovery argument.
Fix 0 < R 0 < R 0<R_0<R 0 < R 0 < R and set
E = { x ∈ R 3 : R 0 < ∣ x ∣ < R } , V R 0 = { v ∈ H 2 ( B R ; R ) : v = 0 a.e. on E } . (203) E=\{x\in\mathbb{R}^3:R_0<|x|<R\},\qquad
V_{R_0}=\{v\in H^2(B_R;\mathbb{R}):v=0\text{ a.e. on }E\}. \tag*{(203)} E = { x ∈ R 3 : R 0 < ∣ x ∣ < R } , V R 0 = { v ∈ H 2 ( B R ; R ) : v = 0 a.e. on E } . (203)
The shell and both radii are fixed before varying coefficients or errors. For a real vector field β ∈ W 1 , ∞ ( B R ; R 3 ) \beta\in W^{1,\infty}(B_R;\mathbb{R}^3) β ∈ W 1 , ∞ ( B R ; R 3 ) define
P β v = tf ( ∇ 2 v ) − tf ( β ⊗ ∇ v + ∇ v ⊗ β ) , tf K = K − 1 3 ( tr K ) I 3 . (204) P_\beta v=\operatorname{tf}(\nabla^2v)-\operatorname{tf}(\beta\otimes\nabla v+\nabla v\otimes\beta),\qquad
\operatorname{tf}K=K-\frac{1}{3}(\operatorname{tr}K)I_3. \tag*{(204)} P β v = tf ( ∇ 2 v ) − tf ( β ⊗ ∇ v + ∇ v ⊗ β ) , tf K = K − 3 1 ( tr K ) I 3 . (204)
The tensor convention is (3) ; matrix norms are Frobenius norms. Theorem 7.1 is independent of the transport construction.
The shell normalization removes the homogeneous kernel, giving coercivity uniformly over the bounded drift class.
Theorem 7.1. For every finite B β B_\beta B β there is a constant C = C ( R , R 0 , B β ) C=C(R,R_0,B_\beta) C = C ( R , R 0 , B β ) such that, for every real β ∈ W 1 , ∞ ( B R ; R 3 ) \beta\in W^{1,\infty}(B_R;\mathbb{R}^3) β ∈ W 1 , ∞ ( B R ; R 3 ) with ∥ β ∥ W 1 , ∞ ≤ B β \|\beta\|_{W^{1,\infty}}\le B_\beta ∥ β ∥ W 1 , ∞ ≤ B β and every v ∈ V R 0 v\in V_{R_0} v ∈ V R 0 ,
∥ v ∥ H 2 ( B R ) + ∥ v ∥ L ∞ ( B R ) ≤ C ∥ P β v ∥ L 2 ( B R ) . (205) \|v\|_{H^2(B_R)}+\|v\|_{L^\infty(B_R)}\le C\|P_\beta v\|_{L^2(B_R)}. \tag*{(205)} ∥ v ∥ H 2 ( B R ) + ∥ v ∥ L ∞ ( B R ) ≤ C ∥ P β v ∥ L 2 ( B R ) . (205) Here V R 0 V_{R_0} V R 0 and P β P_\beta P β are defined in (203) and (204) .
Proof. The principal estimate. A function in V R 0 V_{R_0} V R 0 vanishes in an open collar of ∂ B R \partial B_R ∂ B R . Its zero extension v ~ \widetilde v v therefore belongs to H 2 ( R 3 ) H^2(\mathbb{R}^3) H 2 ( R 3 ) , has support in B ‾ R 0 \overline{B}_{R_0} B R 0 , and has no interface distribution at ∂ B R \partial B_R ∂ B R . For example, multiply by a smooth cutoff equal to one on B ‾ R 0 \overline{B}_{R_0} B R 0 and compactly supported in B R B_R B R before extending. For real ξ \xi ξ ,
∣ ξ ⊗ ξ − 1 3 ∣ ξ ∣ 2 I 3 ∣ 2 = 2 3 ∣ ξ ∣ 4 . (206) \left|\xi\otimes\xi-\frac{1}{3}|\xi|^2I_3\right|^2=\frac{2}{3}|\xi|^4. \tag*{(206)} ξ ⊗ ξ − 3 1 ∣ ξ ∣ 2 I 3 2 = 3 2 ∣ ξ ∣ 4 . (206)
Plancherel consequently gives the exact identity
∥ tf ( ∇ 2 v ~ ) ∥ 2 2 = 2 3 ∥ ∇ 2 v ~ ∥ 2 2 = 2 3 ∥ Δ v ~ ∥ 2 2 . (207) \|\operatorname{tf}(\nabla^2\widetilde v)\|_2^2=\frac{2}{3}\|\nabla^2\widetilde v\|_2^2=\frac{2}{3}\|\Delta\widetilde v\|_2^2. \tag*{(207)} ∥ tf ( ∇ 2 v ) ∥ 2 2 = 3 2 ∥ ∇ 2 v ∥ 2 2 = 3 2 ∥Δ v ∥ 2 2 . (207)
The tensor term in (7.2) has norm at most 2 B β ∣ ∇ v ∣ 2B_{\beta}|\nabla v| 2 B β ∣∇ v ∣ . Also, for every η > 0 \eta>0 η > 0 , the Fourier inequality ∣ ξ ∣ 2 ≤ η 2 ∣ ξ ∣ 4 + ( 4 η 2 ) − 1 |\xi|^2\leq\eta^2|\xi|^4+(4\eta^2)^{-1} ∣ ξ ∣ 2 ≤ η 2 ∣ ξ ∣ 4 + ( 4 η 2 ) − 1 yields
∥ ∇ v ∥ 2 ≤ η ∥ ∇ 2 v ∥ 2 + ( 2 η ) − 1 ∥ v ∥ 2 . (208) \|\nabla v\|_2\leq\eta\|\nabla^2v\|_2+(2\eta)^{-1}\|v\|_2. \tag*{(208)} ∥∇ v ∥ 2 ≤ η ∥ ∇ 2 v ∥ 2 + ( 2 η ) − 1 ∥ v ∥ 2 . (208)
Choose η \eta η using the fixed number B β B_{\beta} B β , not by restricting B β B_{\beta} B β . Absorption in (207) gives
∥ v ∥ H 2 ( B R ) ≤ C ( R , B β ) ( ∥ P β v ∥ L 2 ( B R ) + ∥ v ∥ L 2 ( B R ) ) . (209) \|v\|_{H^2(B_R)}\leq C(R,B_{\beta})\bigl(\|P_{\beta}v\|_{L^2(B_R)}+\|v\|_{L^2(B_R)}\bigr). \tag*{(209)} ∥ v ∥ H 2 ( B R ) ≤ C ( R , B β ) ( ∥ P β v ∥ L 2 ( B R ) + ∥ v ∥ L 2 ( B R ) ) . (209)
The homogeneous weak kernel. Suppose v ∈ V R 0 v\in\mathcal{V}_{R_0} v ∈ V R 0 and P β v = 0 P_{\beta}v=0 P β v = 0 , with V R 0 \mathcal{V}_{R_0} V R 0 and P β P_{\beta} P β defined in (203) and (204) . Set
r = ∇ v ∈ H 1 ( B R ; R 3 ) , w = Δ v ∈ L 2 ( B R ) . (210) r=\nabla v\in H^1(B_R;\mathbb{R}^3),\qquad w=\Delta v\in L^2(B_R). \tag*{(210)} r = ∇ v ∈ H 1 ( B R ; R 3 ) , w = Δ v ∈ L 2 ( B R ) . (210)
Define
T i j k = β i δ j k + β j δ i k − 2 3 δ i j β k . (211) T_{ij}^{k}=\beta_i\delta_{jk}+\beta_j\delta_{ik}-\frac{2}{3}\delta_{ij}\beta_k. \tag*{(211)} T ij k = β i δ j k + β j δ ik − 3 2 δ ij β k . (211)
Repeated component indices run from one to three. Restoring the trace in the weak equation, we obtain, in the sense of distributions,
∂ j r i = 1 3 δ i j w + T i j k r k . (212) \partial_jr_i=\frac{1}{3}\delta_{ij}w+T_{ij}^{k}r_k. \tag*{(212)} ∂ j r i = 3 1 δ ij w + T ij k r k . (212)
Apply ∂ i \partial_i ∂ i and sum in i i i . Distributional derivatives of v v v commute, and the product rule is valid for T ∈ W 1 , ∞ T\in W^{1,\infty} T ∈ W 1 , ∞ and r ∈ H 1 r\in H^1 r ∈ H 1 . Thus
2 3 ∂ j w = ∑ i , k [ ( ∂ i T i j k ) r k + T i j k ∂ i r k ] . (213) \frac{2}{3}\partial_jw=\sum_{i,k}\left[(\partial_iT_{ij}^{k})r_k+T_{ij}^{k}\partial_ir_k\right]. \tag*{(213)} 3 2 ∂ j w = i , k ∑ [ ( ∂ i T ij k ) r k + T ij k ∂ i r k ] . (213)
The right side already lies in L 2 L^2 L 2 ; this proves w ∈ H 1 w\in H^1 w ∈ H 1 before substituting (212) again. The substitution then gives
∂ j w = 5 3 β j w + B j ℓ r ℓ , B j ℓ = 3 2 ( ∑ i ∂ i T i j ℓ + ∑ i , k T i j k T k i ℓ ) . (214) \partial_jw=\frac{5}{3}\beta_jw+B_{j\ell}r_\ell,\qquad B_{j\ell}=\frac{3}{2}\left(\sum_i\partial_iT_{ij}^{\ell}+\sum_{i,k}T_{ij}^{k}T_{ki}^{\ell}\right). \tag*{(214)} ∂ j w = 3 5 β j w + B j ℓ r ℓ , B j ℓ = 2 3 i ∑ ∂ i T ij ℓ + i , k ∑ T ij k T k i ℓ . (214)
For clarity, all contractions used here are
∑ i T i j i = 10 3 β j , \sum_iT_{ij}^{i}=\frac{10}{3}\beta_j, i ∑ T ij i = 3 10 β j ,
∑ i ∂ i T i j ℓ = ( div β ) δ j ℓ + ∂ ℓ β j − 2 3 ∂ j β ℓ , \sum_i\partial_iT_{ij}^{\ell}=(\operatorname{div}\beta)\delta_{j\ell}+\partial_\ell\beta_j-\frac{2}{3}\partial_j\beta_\ell, i ∑ ∂ i T ij ℓ = ( div β ) δ j ℓ + ∂ ℓ β j − 3 2 ∂ j β ℓ ,
∑ i , k T i j k T k i ℓ = 1 3 ∣ β ∣ 2 δ j ℓ + 1 9 β j β ℓ . (215) \sum_{i,k}T_{ij}^{k}T_{ki}^{\ell}=\frac{1}{3}|\beta|^2\delta_{j\ell}+\frac{1}{9}\beta_j\beta_\ell. \tag*{(215)} i , k ∑ T ij k T k i ℓ = 3 1 ∣ β ∣ 2 δ j ℓ + 9 1 β j β ℓ . (215)
The last identity follows by inserting (211) : the trace ∑ i T i i ℓ \sum_iT_{ii}^{\ell} ∑ i T ii ℓ vanishes and the remaining two sums combine into 1 3 ∑ i β i T i j ℓ \frac{1}{3}\sum_i\beta_iT_{ij}^{\ell} 3 1 ∑ i β i T ij ℓ . Consequently,
U = ( r 1 , r 2 , r 3 , w ) T ∈ H 1 ( B R ; R 4 ) , (216) U=(r_1,r_2,r_3,w)^T\in H^1(B_R;\mathbb{R}^4), \tag*{(216)} U = ( r 1 , r 2 , r 3 , w ) T ∈ H 1 ( B R ; R 4 ) , (216)
and it satisfies
∂ j U = C j U a.e. , ∥ C j ∥ ∞ ≤ C ( 1 + B β + B β 2 ) , \partial_jU=C_jU\quad\text{a.e.},\qquad\|C_j\|_\infty\leq C(1+B_\beta+B_\beta^2), ∂ j U = C j U a.e. , ∥ C j ∥ ∞ ≤ C ( 1 + B β + B β 2 ) ,
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 ) . (217) C_j=
\begin{pmatrix}
(T_{ij}^{k})_{i,k=1}^{3} & (\delta_{ij}/3)_{i=1}^{3}\\
(B_{jk})_{k=1}^{3} & 5\beta_j/3
\end{pmatrix}. \tag*{(217)} 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 ) . (217)
Equivalently, the only lower-row coefficient not already explicit is
B j ℓ = 3 2 [ ( div β ) δ j ℓ + ∂ ℓ β j − 2 3 ∂ j β ℓ + 1 3 ∣ β ∣ 2 δ j ℓ + 1 9 β j β ℓ ] . (218) B_{j\ell}=\frac{3}{2}\left[(\operatorname{div}\beta)\delta_{j\ell}+\partial_\ell\beta_j-\frac{2}{3}\partial_j\beta_\ell+\frac{1}{3}|\beta|^2\delta_{j\ell}+\frac{1}{9}\beta_j\beta_\ell\right]. \tag*{(218)} B j ℓ = 2 3 [ ( div β ) δ j ℓ + ∂ ℓ β j − 3 2 ∂ j β ℓ + 3 1 ∣ β ∣ 2 δ j ℓ + 9 1 β j β ℓ ] . (218)
Only first weak derivatives of β \beta β occur; they need not be continuous. On the three-dimensional ball, H 1 ↪ L 6 H^1 \hookrightarrow L^6 H 1 ↪ L 6 . Equation (217) first implies U ∈ W 1 , 6 U \in W^{1,6} U ∈ W 1 , 6 and then, by W 1 , 6 ↪ C 0 ∩ L ∞ W^{1,6} \hookrightarrow C^0 \cap L^\infty W 1 , 6 ↪ C 0 ∩ L ∞ , U ∈ W 1 , ∞ U \in W^{1,\infty} U ∈ W 1 , ∞ . The Euclidean vector norm and the bounded coefficient matrices give ∣ ∇ U ∣ ≤ C ∗ ∣ U ∣ |\nabla U| \le C_*|U| ∣∇ U ∣ ≤ C ∗ ∣ U ∣ a.e., where
C ∗ = ( ∑ j ∥ C j ∥ ∞ 2 ) 1 / 2 , q a = log ( a + ∣ U ∣ ) , a > 0. (219) C_* = \left(\sum_j \|C_j\|_\infty^2\right)^{1/2}, \qquad q_a = \log(a+|U|), \qquad a>0. \tag*{(219)} C ∗ = ( j ∑ ∥ C j ∥ ∞ 2 ) 1/2 , q a = log ( a + ∣ U ∣ ) , a > 0. (219)
The constant is fixed by the drift class. Apply the Sobolev chain rule to q a q_a q a . Its Lipschitz representative satisfies
∣ ∇ q a ∣ ≤ C ( 1 + B β + B β 2 ) a.e. , (220) |\nabla q_a| \le C(1+B_\beta+B_\beta^2) \quad\text{a.e.}, \tag*{(220)} ∣∇ q a ∣ ≤ C ( 1 + B β + B β 2 ) a.e. , (220)
with the same constant for all a a a . Convexity of B R B_R B R therefore gives
a + ∣ U ( x ) ∣ ≤ exp ( C ( 1 + B β + B β 2 ) ∣ x − x 0 ∣ ) ( a + ∣ U ( x 0 ) ∣ ) . (221) a+|U(x)| \le\exp\bigl(C(1+B_\beta+B_\beta^2)|x-x_0|\bigr)(a+|U(x_0)|). \tag*{(221)} a + ∣ U ( x ) ∣ ≤ exp ( C ( 1 + B β + B β 2 ) ∣ x − x 0 ∣ ) ( a + ∣ U ( x 0 ) ∣ ) . (221)
Since v = 0 v=0 v = 0 on the open shell, its weak derivatives r , w r,w r , w vanish there a.e., and the continuous representative of U U U vanishes at every point of E E E . Choose x 0 ∈ E x_0 \in E x 0 ∈ E in (221) and let a ↓ 0 a \downarrow0 a ↓ 0 . Then U = 0 U=0 U = 0 , and v = 0 v=0 v = 0 follows from ∇ v = 0 \nabla v=0 ∇ v = 0 and the shell normalization. This argument does not restrict an L ∞ L^\infty L ∞ coefficient matrix to an arbitrary path.
Uniformity over the Lipschitz closure. If the H 2 H^2 H 2 estimate in (205) failed uniformly, there would exist β n \beta_n β n and v n ∈ V R 0 v_n \in\mathcal{V}_{R_0} v n ∈ V R 0 such that
∥ β n ∥ W 1 , ∞ ≤ B β , ∥ v n ∥ H 2 = 1 , ∥ P β n v n ∥ 2 ⟶ 0. (222) \|\beta_n\|_{W^{1,\infty}} \le B_\beta, \qquad\|v_n\|_{H^2}=1, \qquad\|P_{\beta_n}v_n\|_2 \longrightarrow0. \tag*{(222)} ∥ β n ∥ W 1 , ∞ ≤ B β , ∥ v n ∥ H 2 = 1 , ∥ P β n v n ∥ 2 ⟶ 0. (222)
The Lipschitz representatives on B R ‾ \overline{B_R} B R are equibounded and equicontinuous. Arzelà–Ascoli, weak-star compactness, and Rellich compactness give, after passage to a subsequence,
β n → β uniformly , β ∈ W 1 , ∞ , ∥ β ∥ W 1 , ∞ ≤ B β , \beta_n \to\beta\text{ uniformly}, \qquad\beta\in W^{1,\infty}, \qquad\|\beta\|_{W^{1,\infty}} \le B_\beta, β n → β uniformly , β ∈ W 1 , ∞ , ∥ β ∥ W 1 , ∞ ≤ B β ,
v n ⇀ v in H 2 , v n → v in H 1 . (223) v_n \rightharpoonup v \text{ in } H^2, \qquad v_n \to v \text{ in } H^1. \tag*{(223)} v n ⇀ v in H 2 , v n → v in H 1 . (223)
In particular β n ⊗ ∇ v n → β ⊗ ∇ v \beta_n \otimes\nabla v_n \to\beta\otimes\nabla v β n ⊗ ∇ v n → β ⊗ ∇ v in L 2 L^2 L 2 . More explicitly, the operator difference obeys
∥ ( P β n − P β ) v n ∥ 2 ≤ 2 ∥ β n − β ∥ ∞ ∥ ∇ v n ∥ 2 ⟶ 0. (224) \|(P_{\beta_n}-P_\beta)v_n\|_2 \le2\|\beta_n-\beta\|_\infty\|\nabla v_n\|_2 \longrightarrow0. \tag*{(224)} ∥ ( P β n − P β ) v n ∥ 2 ≤ 2∥ β n − β ∥ ∞ ∥∇ v n ∥ 2 ⟶ 0. (224)
Thus P β v n → 0 P_\beta v_n \to0 P β v n → 0 strongly in L 2 L^2 L 2 , while its principal part converges weakly to that of P β v P_\beta v P β v . Passing to the limit against smooth compactly supported matrix fields, we obtain P β v = 0 P_\beta v=0 P β v = 0 in distributions. Strong L 2 L^2 L 2 convergence also gives ∥ v ∥ L 2 ( E ) = 0 \|v\|_{L^2(E)}=0 ∥ v ∥ L 2 ( E ) = 0 . The weak-kernel argument therefore applies to this same Lipschitz limit and gives v = 0 v=0 v = 0 . Notice that weak-star convergence of ∇ β n \nabla\beta_n ∇ β n suffices to retain the drift bound; no convergence of the derived matrices B j ℓ B_{j\ell} B j ℓ is needed to pass to the second-order equation. The uniform estimate (209) , applied to v n v_n v n , now contradicts ∥ v n ∥ H 2 = 1 \|v_n\|_{H^2}=1 ∥ v n ∥ H 2 = 1 . Finally, H 2 ( B R ) ↪ L ∞ ( B R ) H^2(B_R) \hookrightarrow L^\infty(B_R) H 2 ( B R ) ↪ L ∞ ( B R ) proves the stated estimate.
Remark. The estimate holds for every finite bound B β B_\beta B β ; no smallness condition is imposed on the drift. The finite-type system (217) is used only for a homogeneous weak kernel. For an inhomogeneous equation P β v = F ∈ L 2 P_\beta v=F \in L^2 P β v = F ∈ L 2 , the proof of (205) never differentiates F F F . It does not assert L ∞ L^\infty L ∞ estimates for ∇ v \nabla v ∇ v or Δ v \Delta v Δ v .
Quantitative recovery from the block equations
The transport residual of the holomorphic endomorphism H 0 H_0 H 0 controls the Lamé parameters through the connection blocks. The scalar and matrix blocks are estimated in order before applying the trace-free Hessian estimate.
Use the tensor convention (1.3) and the mixed norms (5.34), on the same compact holomorphic charts as Proposition 5.4. All spatial derivatives hold the spinor representative fixed.
Material priors and denominator bounds
Fix the ball and shell in (203) . Let both real material pairs satisfy, for j = 1 , 2 j=1,2 j = 1 , 2 ,
∥ λ j ∥ C 4 ( B R ‾ ) + ∥ μ j ∥ C 4 ( B R ‾ ) ≤ M 4 , μ j ≥ c 0 > 0 , 3 λ j + 2 μ j ≥ c 0 , (225) \lVert\lambda_j\rVert_{C^4(\overline{B_R})}+\lVert\mu_j\rVert_{C^4(\overline{B_R})}\le M_4,\qquad\mu_j\ge c_0>0,\qquad3\lambda_j+2\mu_j\ge c_0, \tag*{(225)} ∥ λ j ∥ C 4 ( B R ) + ∥ μ j ∥ C 4 ( B R ) ≤ M 4 , μ j ≥ c 0 > 0 , 3 λ j + 2 μ j ≥ c 0 , (225)
and suppose the pairs agree on E E E from (203) . For each material use κ j , c j , d j , g μ , j \kappa_j,c_j,d_j,g_{\mu,j} κ j , c j , d j , g μ , j from (31) , F μ j F_{\mu_j} F μ j from (33) , and ρ \rho ρ from (136) . Thus F μ j T = F μ j F_{\mu_j}^{T}=F_{\mu_j} F μ j T = F μ j , ρ \rho ρ is real, and ρ = 0 \rho=0 ρ = 0 on E E E . Continuity supplies the same material equalities on ∂ B R \partial B_R ∂ B R ; no identity for the transport throughout E E E is assumed. Define fixed positive numbers
κ − : = 1 4 + c 0 12 M 4 , ν − : = c 0 3 M 4 . (226) \kappa_-:=\frac{1}{4}+\frac{c_0}{12M_4},\qquad\nu_-:=\frac{c_0}{3M_4}. \tag*{(226)} κ − := 4 1 + 12 M 4 c 0 , ν − := 3 M 4 c 0 . (226)
The following nonoptimal bounds make every material division explicit:
λ j + 2 μ j ≥ 5 c 0 3 , λ j + μ j ≥ 2 c 0 3 , λ j + 2 μ j ≤ 3 M 4 , \lambda_j+2\mu_j\ge\frac{5c_0}{3},\qquad\lambda_j+\mu_j\ge\frac{2c_0}{3},\qquad\lambda_j+2\mu_j\le3M_4, λ j + 2 μ j ≥ 3 5 c 0 , λ j + μ j ≥ 3 2 c 0 , λ j + 2 μ j ≤ 3 M 4 ,
κ j − 1 4 = 3 λ j + 2 μ j 4 ( λ j + 2 μ j ) ≥ c 0 12 M 4 , 1 − κ j = μ j λ j + 2 μ j ≥ ν − , (227) \kappa_j-\frac{1}{4}=\frac{3\lambda_j+2\mu_j}{4(\lambda_j+2\mu_j)}\ge\frac{c_0}{12M_4},\qquad1-\kappa_j=\frac{\mu_j}{\lambda_j+2\mu_j}\ge\nu_-, \tag*{(227)} κ j − 4 1 = 4 ( λ j + 2 μ j ) 3 λ j + 2 μ j ≥ 12 M 4 c 0 , 1 − κ j = λ j + 2 μ j μ j ≥ ν − , (227)
d j ≥ κ − c 0 , M 4 − 1 / 2 ≤ c j ≤ c 0 − 1 / 2 , 2 c 1 ( κ 1 − 1 − 1 ) ≥ 2 ν − M 4 − 1 / 2 . d_j\ge\kappa_-\sqrt{c_0},\qquad M_4^{-1/2}\le c_j\le c_0^{-1/2},\qquad2c_1(\kappa_1^{-1}-1)\ge2\nu_-M_4^{-1/2}. d j ≥ κ − c 0 , M 4 − 1/2 ≤ c j ≤ c 0 − 1/2 , 2 c 1 ( κ 1 − 1 − 1 ) ≥ 2 ν − M 4 − 1/2 .
In particular κ − ≤ κ j ≤ 1 − ν − \kappa_-\le\kappa_j\le1-\nu_- κ − ≤ κ j ≤ 1 − ν − . The chain and product rules give fixed finite C 4 C^4 C 4 bounds for the scalar material functions and their reciprocals whenever used; F μ j F_{\mu_j} F μ j is uniformly C 2 C^2 C 2 .
Connection blocks and holomorphic endomorphisms
Use θ , Q , n θ , r a \theta,Q,n_\theta,r_a θ , Q , n θ , r a from (56) and D θ D_\theta D θ from (37) . The rank calculation following (56) gives ker Q T = C θ \ker Q^T=\mathbb{C}\theta ker Q T = C θ . For the connection (61) , introduce the block notation
Γ j = ( L j N j P j 0 ) , L j = n θ r g μ , j , N j = 2 d j n θ , P j = − θ T F μ j Q . (228) \Gamma_j=\begin{pmatrix}L_j&N_j\\P_j&0\end{pmatrix},\qquad L_j=n_\theta r g_{\mu,j},\qquad N_j=2d_jn_\theta,\qquad P_j=-\theta^T F_{\mu_j}Q. \tag*{(228)} Γ j = ( L j P j N j 0 ) , L j = n θ r g μ , j , N j = 2 d j n θ , P j = − θ T F μ j Q . (228)
These three blocks have sizes 2 × 2 2\times2 2 × 2 , 2 × 1 2\times1 2 × 1 , and 1 × 2 1\times2 1 × 2 , respectively. The reduced 4 × 4 4\times4 4 × 4 operator M j \mathcal{M}_j M j and the leaf connection P j P_j P j are not renamed as any of these blocks.
Let H 0 ( x , ⋅ ) H_0(x,\cdot) H 0 ( x , ⋅ ) be a holomorphic endomorphism of the bundle (178) . By Lemma 6.1, its block form is (180) . Abbreviate its lower row by
H 0 = ( A H 0 C H d H ) , C H = s 0 C H , 0 + s 1 C H , 1 . (229) H_0=\begin{pmatrix}A_H&0\\C_H&d_H\end{pmatrix},\qquad C_H=s_0C_{H,0}+s_1C_{H,1}. \tag*{(229)} H 0 = ( A H C H 0 d H ) , C H = s 0 C H , 0 + s 1 C H , 1 . (229)
Here A H , d H , C H , 0 , C H , 1 A_H,d_H,C_{H,0},C_{H,1} A H , d H , C H , 0 , C H , 1 depend only on x x x and may be complex. Their C x 3 C_x^3 C x 3 bounds follow from fixed chart evaluations of H 0 H_0 H 0 . Theorem 7.3 does not require H 0 H_0 H 0 to be invertible.
The following algebraic identity resolves the null contractions; scalar matrices form the undetected subspace.
Lemma 7.2. Set
a = 1 − i 2 . (230) a=\frac{1-i}{2}. \tag*{(230)} a = 2 1 − i . (230) Choose the following six normalized spinors:
( 1 , 0 ) , ( 0 , i ) , ( 2 − 1 / 2 , i 2 − 1 / 2 ) , (1,0),\qquad(0,i),\qquad(2^{-1/2},i2^{-1/2}), ( 1 , 0 ) , ( 0 , i ) , ( 2 − 1/2 , i 2 − 1/2 ) , ( 2 − 1 / 2 , − i 2 − 1 / 2 ) , ( a , − a ) , ( a , a ) . (231) (2^{-1/2},-i2^{-1/2}),\qquad(a,-a),\qquad(a,a). \tag*{(231)} ( 2 − 1/2 , − i 2 − 1/2 ) , ( a , − a ) , ( a , a ) . (231) Their vectors θ ℓ \theta_\ell θ ℓ are, in order, e 1 ± i e 2 e_1\pm ie_2 e 1 ± i e 2 , e 1 ± i e 3 e_1\pm ie_3 e 1 ± i e 3 , and e 2 ± i e 3 e_2\pm ie_3 e 2 ± i e 3 . Let Q ℓ Q_\ell Q ℓ be their matrices in (56) . For every K ∈ C 3 × 3 K\in\mathbb{C}^{3\times3} K ∈ C 3 × 3 ,
∑ ℓ = 1 6 ∣ Q ℓ T K θ ℓ ∣ 2 = 3 2 ∣ tf ( sym K ) ∣ 2 + 1 2 ∣ skew K ∣ 2 . (232) \sum_{\ell=1}^{6}\lvert Q_\ell^T K\theta_\ell\rvert^2=\frac{3}{2}\lvert\operatorname{tf}(\operatorname{sym}K)\rvert^2+\frac{1}{2}\lvert\operatorname{skew}K\rvert^2. \tag*{(232)} ℓ = 1 ∑ 6 ∣ Q ℓ T K θ ℓ ∣ 2 = 2 3 ∣ tf ( sym K ) ∣ 2 + 2 1 ∣ skew K ∣ 2 . (232)
Here tf is defined in (204) , and
sym K = ( K + K T ) / 2 , skew K = ( K − K T ) / 2. (233) \operatorname{sym} K = (K + K^{T})/2,\qquad\operatorname{skew} K = (K - K^{T})/2. \tag*{(233)} sym K = ( K + K T ) /2 , skew K = ( K − K T ) /2. (233)
In particular,
∣ tf K ∣ ≤ 2 ( ∑ ℓ = 1 6 ∣ Q ℓ T K θ ℓ ∣ 2 ) 1 / 2 . (234) |\operatorname{tf} K| \le\sqrt{2}\left(\sum_{\ell=1}^{6}|Q_{\ell}^{T}K\theta_{\ell}|^{2}\right)^{1/2}. \tag*{(234)} ∣ tf K ∣ ≤ 2 ( ℓ = 1 ∑ 6 ∣ Q ℓ T K θ ℓ ∣ 2 ) 1/2 . (234)
The kernel of this linear map is exactly C I 3 \mathbb{C}I_{3} C I 3 .
Proof. For i < j i<j i < j and the remaining index k k k , substitution of the two spinors with θ = e i ± i e j \theta=e_i\pm ie_j θ = e i ± i e j gives
∑ ± ∣ Q i j , ± T K ( e i ± i e j ) ∣ 2 = 1 2 ( ∣ K i i − K j j ∣ 2 + ∣ K i j + K j i ∣ 2 + ∣ K k i ∣ 2 + ∣ K k j ∣ 2 ) . (235) \sum_{\pm}|Q_{ij,\pm}^{T}K(e_i\pm ie_j)|^{2}=\frac{1}{2}\left(|K_{ii}-K_{jj}|^{2}+|K_{ij}+K_{ji}|^{2}+|K_{ki}|^{2}+|K_{kj}|^{2}\right). \tag*{(235)} ± ∑ ∣ Q ij , ± T K ( e i ± i e j ) ∣ 2 = 2 1 ( ∣ K ii − K j j ∣ 2 + ∣ K ij + K j i ∣ 2 + ∣ K k i ∣ 2 + ∣ K k j ∣ 2 ) . (235)
Sum over the three pairs. The diagonal contribution is ∑ i < j ∣ K i i − K j j ∣ 2 = 3 ∑ i ∣ K i i − tr K / 3 ∣ 2 \sum_{i<j}|K_{ii}-K_{jj}|^{2}=3\sum_i|K_{ii}-\operatorname{tr}K/3|^{2} ∑ i < j ∣ K ii − K j j ∣ 2 = 3 ∑ i ∣ K ii − tr K /3 ∣ 2 . The other terms split into the symmetric and skew-symmetric parts of the off-diagonal entries, which proves (232) . These two parts are orthogonal for the Hermitian Frobenius norm, although their definition uses ordinary transpose.
We use the four transport blocks to derive the material equations and then apply the scalar coercive estimate to recover the coefficients.
Theorem 7.3. Assume (225) , agreement of the two materials on E E E , and the fixed compact direction charts used in (168) . Let H 0 H_0 H 0 be as in (229) , with ∥ H 0 ∥ C x 3 C ζ 0 ≤ B H \lVert H_0\rVert_{C_x^3C_\zeta^0}\le B_H ∥ H 0 ∥ C x 3 C ζ 0 ≤ B H . Use the transport operator T \mathcal{T} T defined in (186) . There are fixed C < ∞ C<\infty C < ∞ and ε r i g , 0 > 0 \varepsilon_{\mathrm{rig},0}>0 ε rig , 0 > 0 , depending only on R R R , R 0 R_0 R 0 , M 4 M_4 M 4 , c 0 c_0 c 0 , B H B_H B H and the chosen charts, such that
∥ T ( H 0 ) ∥ C x 2 C ζ 0 + ∥ H 0 − I ∥ C 0 ( ∂ B R × C P 1 ) ≤ ε r i g < ε r i g , 0 (236) \lVert\mathcal{T}(H_0)\rVert_{C_x^2C_\zeta^0}+\lVert H_0-I\rVert_{C^0(\partial B_R\times\mathbb{CP}^1)}\le\varepsilon_{\mathrm{rig}}<\varepsilon_{\mathrm{rig},0} \tag*{(236)} ∥ T ( H 0 ) ∥ C x 2 C ζ 0 + ∥ H 0 − I ∥ C 0 ( ∂ B R × CP 1 ) ≤ ε rig < ε rig , 0 (236) implies
∥ κ 1 − κ 2 ∥ C 2 ( B R ‾ ) + ∥ ρ ∥ H 2 ( B R ) + ∥ λ 1 − λ 2 ∥ L ∞ ( B R ) + ∥ μ 1 − μ 2 ∥ L ∞ ( B R ) ≤ C ε r i g . (237) \lVert\kappa_1-\kappa_2\rVert_{C^2(\overline{B_R})}+\lVert\rho\rVert_{H^2(B_R)}+\lVert\lambda_1-\lambda_2\rVert_{L^\infty(B_R)}+\lVert\mu_1-\mu_2\rVert_{L^\infty(B_R)}\le C\varepsilon_{\mathrm{rig}}. \tag*{(237)} ∥ κ 1 − κ 2 ∥ C 2 ( B R ) + ∥ ρ ∥ H 2 ( B R ) + ∥ λ 1 − λ 2 ∥ L ∞ ( B R ) + ∥ μ 1 − μ 2 ∥ L ∞ ( B R ) ≤ C ε rig . (237) Here κ j \kappa_j κ j and ρ \rho ρ are defined in (31) and (136) ; ε r i g \varepsilon_{\mathrm{rig}} ε rig is the error bound in (236) .
Proof. Write the four blocks of T ( H 0 ) \mathcal{T}(H_0) T ( H 0 ) as R u l , R u r , R l l , R l r R_{\mathrm{ul}},R_{\mathrm{ur}},R_{\mathrm{ll}},R_{\mathrm{lr}} R ul , R ur , R ll , R lr . Ordinary matrix multiplication gives
2 D θ A H + L 1 A H + N 1 C H − A H L 2 = R u l , N 1 d H − A H N 2 = R u r , 2 D θ C H + P 1 A H − C H L 2 − d H P 2 = R l l , 2 D θ d H − C H N 2 = R l r . (238) \begin{aligned}
2D_\theta A_H+L_1A_H+N_1C_H-A_HL_2&=R_{\mathrm{ul}},\\
N_1d_H-A_HN_2&=R_{\mathrm{ur}},\\
2D_\theta C_H+P_1A_H-C_HL_2-d_HP_2&=R_{\mathrm{ll}},\\
2D_\theta d_H-C_HN_2&=R_{\mathrm{lr}}. \tag*{(238)}
\end{aligned} 2 D θ A H + L 1 A H + N 1 C H − A H L 2 N 1 d H − A H N 2 2 D θ C H + P 1 A H − C H L 2 − d H P 2 2 D θ d H − C H N 2 = R ul , = R ur , = R ll , = R lr . (238)
Write O q ( ε r i g ) O_q(\varepsilon_{\mathrm{rig}}) O q ( ε rig ) only for a specified error bounded in C x q C ζ 0 C_x^qC_\zeta^0 C x q C ζ 0 by a fixed multiple of ε r i g \varepsilon_{\mathrm{rig}} ε rig . Under spinor rescaling by b ≠ 0 b\ne0 b = 0 , Q Q Q , n θ n_\theta n θ , C H C_H C H have degree one, θ \theta θ degree two, and A H A_H A H , d H d_H d H degree zero. The four equations have degrees two, one, three, and two. Thus fixed representative changes at the finitely many directions below alter error constants only, not their spatial derivatives.
Upper right: no spatial derivative is lost. Set
a H = d 1 d H d 2 , E A = A H − a H I 2 . (239) a_H=\frac{d_1d_H}{d_2},\qquad\mathcal{E}_A=A_H-a_HI_2. \tag*{(239)} a H = d 2 d 1 d H , E A = A H − a H I 2 . (239)
The upper-right equation is − 2 d 2 E A n θ = R u r -2d_2\mathcal{E}_A n_\theta=R_{\mathrm{ur}} − 2 d 2 E A n θ = R ur . The spinors ( 1 , 0 ) (1,0) ( 1 , 0 ) and ( 0 , 1 ) (0,1) ( 0 , 1 ) give n θ = ( 0 , 1 ) T n_\theta=(0,1)^T n θ = ( 0 , 1 ) T and n θ = ( − 1 , 0 ) T n_\theta=(-1,0)^T n θ = ( − 1 , 0 ) T , so they determine both columns of E A \mathcal{E}_A E A . Let e 1 , e 2 e_1,e_2 e 1 , e 2 be the coordinate columns of C 2 \mathbb{C}^2 C 2 . Evaluation at the spinors ( 1 , 0 ) (1,0) ( 1 , 0 ) and ( 0 , 1 ) (0,1) ( 0 , 1 ) gives
E A e 1 = R u r ( 0 , 1 ) 2 d 2 , E A e 2 = − R u r ( 1 , 0 ) 2 d 2 . (240) \mathcal{E}_A e_1=\frac{R_{\mathrm{ur}}(0,1)}{2d_2},\qquad\mathcal{E}_A e_2=-\frac{R_{\mathrm{ur}}(1,0)}{2d_2}. \tag*{(240)} E A e 1 = 2 d 2 R ur ( 0 , 1 ) , E A e 2 = − 2 d 2 R ur ( 1 , 0 ) . (240)
The material bounds for d 2 − 1 d_2^{-1} d 2 − 1 and its first two derivatives give
∥ E A ∥ C x 2 ≤ C ε r i g . (241) \|\mathcal{E}_A\|_{C^2_x} \le C\varepsilon_{\mathrm{rig}}. \tag*{(241)} ∥ E A ∥ C x 2 ≤ C ε rig . (241)
2. Upper left: one derivative is spent, then the scalar is normalized. Put
Q u l = R u l − 2 D θ E A − L 1 E A + E A L 2 . (242) Q_{\mathrm{ul}}=R_{\mathrm{ul}}-2D_\theta\mathcal{E}_A-L_1\mathcal{E}_A+\mathcal{E}_A L_2. \tag*{(242)} Q ul = R ul − 2 D θ E A − L 1 E A + E A L 2 . (242)
Then Q u l = O 1 ( ε r i g ) Q_{\mathrm{ul}}=O_1(\varepsilon_{\mathrm{rig}}) Q ul = O 1 ( ε rig ) , and
2 ( D θ a H ) I 2 + n θ [ a H T ∇ ρ + 2 d 1 C H ] = Q u l . (243) 2(D_\theta a_H)I_2+n_\theta[a_H{}^T\nabla\rho+2d_1C_H]=Q_{\mathrm{ul}}. \tag*{(243)} 2 ( D θ a H ) I 2 + n θ [ a H T ∇ ρ + 2 d 1 C H ] = Q ul . (243)
For a fixed nonzero spinor define
ℓ = ( s 0 , s 1 ) , v = ( s 0 ‾ , s 1 ‾ ) T ∣ s 0 ∣ 2 + ∣ s 1 ∣ 2 . (244) \ell=(s_0,s_1),\qquad v=\frac{(\overline{s_0},\overline{s_1})^T}{|s_0|^2+|s_1|^2}. \tag*{(244)} ℓ = ( s 0 , s 1 ) , v = ∣ s 0 ∣ 2 + ∣ s 1 ∣ 2 ( s 0 , s 1 ) T . (244)
Then ℓ n θ = 0 \ell n_\theta=0 ℓ n θ = 0 and ℓ v = 1 \ell v=1 ℓ v = 1 , whence 2 D θ a H = ℓ Q u l v 2D_\theta a_H=\ell Q_{\mathrm{ul}}v 2 D θ a H = ℓ Q ul v . The six directions in Lemma 7.2 span C 3 \mathbb{C}^3 C 3 : sums and differences of each plus/minus pair recover the coordinate derivatives. Apply this observation also after one spatial derivative. It gives ∥ ∇ a H ∥ C 1 ≤ C ε r i g \|\nabla a_H\|_{C^1}\le C\varepsilon_{\mathrm{rig}} ∥∇ a H ∥ C 1 ≤ C ε rig . At ∂ B R \partial B_R ∂ B R , d 1 = d 2 d_1=d_2 d 1 = d 2 and d H = 1 + O 0 ( ε r i g ) d_H=1+O_0(\varepsilon_{\mathrm{rig}}) d H = 1 + O 0 ( ε rig ) , so a H = 1 + O 0 ( ε r i g ) a_H=1+O_0(\varepsilon_{\mathrm{rig}}) a H = 1 + O 0 ( ε rig ) there. Integration along segments in the closed ball, of length at most 2 R 2R 2 R , therefore gives
∥ a H − 1 ∥ C 2 ≤ C ε r i g . (245) \|a_H-1\|_{C^2}\le C\varepsilon_{\mathrm{rig}}. \tag*{(245)} ∥ a H − 1 ∥ C 2 ≤ C ε rig . (245)
Only the zero-order boundary value, not a boundary derivative, was used. Define the real vector field
b ρ = ∇ ρ d 1 . (246) b_\rho=\frac{\nabla\rho}{d_1}. \tag*{(246)} b ρ = d 1 ∇ ρ . (246)
Since r ∇ ρ = − 2 ( ∇ ρ ) T Q r\nabla\rho=-2(\nabla\rho)^TQ r ∇ ρ = − 2 ( ∇ ρ ) T Q , extraction of the remaining outer product in (243) gives
C H = a H b ρ T Q + E C , E C = O 1 ( ε r i g ) . (247) C_H=a_Hb_\rho^TQ+\mathcal{E}_C,\qquad\mathcal{E}_C=O_1(\varepsilon_{\mathrm{rig}}). \tag*{(247)} C H = a H b ρ T Q + E C , E C = O 1 ( ε rig ) . (247)
For this extraction use n θ ∗ / ∣ n θ ∣ 2 n_\theta^*/|n_\theta|^2 n θ ∗ /∣ n θ ∣ 2 and then divide by 2 d 1 2d_1 2 d 1 . The star appears only in this bounded auxiliary left inverse; the transport identities still use ordinary transpose. All spinor normalizing denominators are bounded below on the fixed charts.
3. Lower right: control the integrating factor and recover κ \kappa κ . Use Q n θ = − θ / 2 Qn_\theta=-\theta/2 Q n θ = − θ /2 and the exact scalar identity a H d 2 / d 1 = d H a_Hd_2/d_1=d_H a H d 2 / d 1 = d H in the lower-right equation to obtain
2 D θ d H + d H D θ ρ = R l r + 2 d 2 E C n θ = O 1 ( ε r i g ) . (248) 2D_\theta d_H+d_HD_\theta\rho=R_{\mathrm{lr}}+2d_2\mathcal{E}_Cn_\theta=O_1(\varepsilon_{\mathrm{rig}}). \tag*{(248)} 2 D θ d H + d H D θ ρ = R lr + 2 d 2 E C n θ = O 1 ( ε rig ) . (248)
Define
b H = d H e ρ / 2 , (249) b_H=d_He^{\rho/2}, \tag*{(249)} b H = d H e ρ /2 , (249)
where ρ \rho ρ is given by (136) . Multiplication by e ρ / 2 / 2 e^{\rho/2}/2 e ρ /2 /2 gives D θ b H = O 1 ( ε r i g ) D_\theta b_H=O_1(\varepsilon_{\mathrm{rig}}) D θ b H = O 1 ( ε rig ) . The material prior already bounds ρ \rho ρ and its needed derivatives; smallness of ρ \rho ρ is not being used. Since the chosen null directions span C 3 \mathbb{C}^3 C 3 , the bounds for D θ b H D_\theta b_H D θ b H control ∇ b H \nabla b_H ∇ b H in C 1 C^1 C 1 . Together with b H = 1 + O 0 ( ε r i g ) b_H=1+O_0(\varepsilon_{\mathrm{rig}}) b H = 1 + O 0 ( ε rig ) on the sphere, these bounds give a fixed constant C b ≥ 1 C_b\ge1 C b ≥ 1 such that
∥ b H − 1 ∥ C 2 ≤ C b ε r i g , ∥ d H − e − ρ / 2 ∥ C 2 ≤ C ε r i g . (250) \|b_H-1\|_{C^2}\le C_b\varepsilon_{\mathrm{rig}},\qquad\|d_H-e^{-\rho/2}\|_{C^2}\le C\varepsilon_{\mathrm{rig}}. \tag*{(250)} ∥ b H − 1 ∥ C 2 ≤ C b ε rig , ∥ d H − e − ρ /2 ∥ C 2 ≤ C ε rig . (250)
Choose, once and for all,
ε r i g , 0 : = min { 1 , ( 2 C b ) − 1 } . (251) \varepsilon_{\mathrm{rig},0}:=\min\{1,(2C_b)^{-1}\}. \tag*{(251)} ε rig , 0 := min { 1 , ( 2 C b ) − 1 } . (251)
Then ∣ b H ∣ ≥ 1 / 2 |b_H|\ge1/2 ∣ b H ∣ ≥ 1/2 , including when H 0 H_0 H 0 has complex entries. Only now is the new denominator b H b_H b H introduced. Its first and second reciprocal derivatives are bounded, as follows from ∂ b H − 1 = − b H − 2 ∂ b H \partial b_H^{-1}=-b_H^{-2}\partial b_H ∂ b H − 1 = − b H − 2 ∂ b H and ∂ i ∂ j b H − 1 = 2 b H − 3 ∂ i b H ∂ j b H − b H − 2 ∂ i ∂ j b H \partial_i\partial_jb_H^{-1}=2b_H^{-3}\partial_i b_H\partial_j b_H-b_H^{-2}\partial_i\partial_jb_H ∂ i ∂ j b H − 1 = 2 b H − 3 ∂ i b H ∂ j b H − b H − 2 ∂ i ∂ j b H . The exact material identities
a H = κ 1 κ 2 b H , κ 1 − κ 2 = κ 2 b H ( a H − b H ) (252) a_H=\frac{\kappa_1}{\kappa_2}b_H,\qquad\kappa_1-\kappa_2=\frac{\kappa_2}{b_H}(a_H-b_H) \tag*{(252)} a H = κ 2 κ 1 b H , κ 1 − κ 2 = b H κ 2 ( a H − b H ) (252)
now imply ∥ κ 1 − κ 2 ∥ C 2 ≤ C ε r i g \lVert\kappa_1-\kappa_2\rVert_{C^2}\le C\varepsilon_{\mathrm{rig}} ∥ κ 1 − κ 2 ∥ C 2 ≤ C ε rig .
4. Lower left: a second derivative is spent, producing only a value error. Equations (241) , (247) , (245) , and (250) yield
A H = I 2 + E A 0 , C H = b ρ T Q + E C 0 , d H = e − ρ / 2 + E d , (253) A_H=I_2+\mathcal{E}_A^0,\qquad C_H=b_\rho^TQ+\mathcal{E}_C^0,\qquad d_H=e^{-\rho/2}+\mathcal{E}_d, \tag*{(253)} A H = I 2 + E A 0 , C H = b ρ T Q + E C 0 , d H = e − ρ /2 + E d , (253)
The fields b ρ b_\rho b ρ and ρ \rho ρ are defined in (246) and (136) . Here E A 0 , E d = O 2 ( ε r i g ) \mathcal{E}_A^0,\mathcal{E}_d=O_2(\varepsilon_{\mathrm{rig}}) E A 0 , E d = O 2 ( ε rig ) and E C 0 = O 1 ( ε r i g ) \mathcal{E}_C^0=O_1(\varepsilon_{\mathrm{rig}}) E C 0 = O 1 ( ε rig ) . Indeed E C 0 = E C + ( a H − 1 ) b ρ T Q \mathcal{E}_C^0=\mathcal{E}_C+(a_H-1)b_\rho^TQ E C 0 = E C + ( a H − 1 ) b ρ T Q and the C 1 C^1 C 1 norm of b ρ b_\rho b ρ was already bounded by the material prior. Define the real, direction-independent matrix
K ρ = 2 ∇ b ρ − F μ 1 + e − ρ / 2 F μ 2 − g μ , 2 b ρ T . (254) K_\rho=2\nabla b_\rho-F_{\mu1}+e^{-\rho/2}F_{\mu2}-g_{\mu,2}b_\rho^T. \tag*{(254)} K ρ = 2∇ b ρ − F μ 1 + e − ρ /2 F μ 2 − g μ , 2 b ρ T . (254)
The lower-left equation gives exactly
( Q T K ρ θ ) T = R l l − E l l , (Q^TK_\rho\theta)^T=R_{\mathrm{ll}}-\mathcal{E}_{\mathrm{ll}}, ( Q T K ρ θ ) T = R ll − E ll ,
E l l = 2 D θ E C 0 + P 1 E A 0 − E C 0 L 2 − E d P 2 = O 0 ( ε r i g ) . (255) \mathcal{E}_{\mathrm{ll}}=2D_\theta\mathcal{E}_C^0+P_1\mathcal{E}_A^0-\mathcal{E}_C^0L_2-\mathcal{E}_dP_2=O_0(\varepsilon_{\mathrm{rig}}). \tag*{(255)} E ll = 2 D θ E C 0 + P 1 E A 0 − E C 0 L 2 − E d P 2 = O 0 ( ε rig ) . (255)
The transpose convention gives
D θ ( b ρ T Q ) = θ T ( ∇ b ρ ) T Q , − ( b ρ T Q ) n θ r g μ , 2 = − ( b ρ ⋅ θ ) g μ , 2 T Q . (256) D_\theta(b_\rho^TQ)=\theta^T(\nabla b_\rho)TQ,\qquad-(b_\rho^TQ)n_\theta r g_{\mu,2}=-(b_\rho\cdot\theta)g_{\mu,2}^TQ. \tag*{(256)} D θ ( b ρ T Q ) = θ T ( ∇ b ρ ) T Q , − ( b ρ T Q ) n θ r g μ , 2 = − ( b ρ ⋅ θ ) g μ , 2 T Q . (256)
After transposing the second identity in (256) , we obtain the order g μ , 2 b ρ T g_{\mu,2}b_\rho^T g μ , 2 b ρ T in (254) . Symmetry of both F μ j F_{\mu j} F μ j handles the other two terms. Lemma 7.2 therefore implies
∥ tf K ρ ∥ C 0 ≤ C ε r i g . (257) \lVert\operatorname{tf}K_\rho\rVert_{C^0}\le C\varepsilon_{\mathrm{rig}}. \tag*{(257)} ∥ tf K ρ ∥ C 0 ≤ C ε rig . (257)
For the material expansion define
r = ∇ ρ , k 1 = ∇ log κ 1 , g Σ = g μ , 1 + g μ , 2 . (258) r=\nabla\rho,\qquad k_1=\nabla\log\kappa_1,\qquad g_\Sigma=g_{\mu,1}+g_{\mu,2}. \tag*{(258)} r = ∇ ρ , k 1 = ∇ log κ 1 , g Σ = g μ , 1 + g μ , 2 . (258)
Here ρ \rho ρ , κ j \kappa_j κ j , and g μ , j g_{\mu,j} g μ , j are defined in (136) and (31) . Since b ρ = ( c 1 / κ 1 ) r b_\rho=(c_1/\kappa_1)r b ρ = ( c 1 / κ 1 ) r , the gradient convention gives
2 ∇ b ρ = 2 c 1 κ 1 ∇ 2 ρ − c 1 κ 1 r ⊗ g μ , 1 − 2 c 1 κ 1 r ⊗ k 1 . (259) 2\nabla b_\rho=\frac{2c_1}{\kappa_1}\nabla^2\rho-\frac{c_1}{\kappa_1}r\otimes g_{\mu,1}-\frac{2c_1}{\kappa_1}r\otimes k_1. \tag*{(259)} 2∇ b ρ = κ 1 2 c 1 ∇ 2 ρ − κ 1 c 1 r ⊗ g μ , 1 − κ 1 2 c 1 r ⊗ k 1 . (259)
Next e − ρ / 2 c 2 = c 1 e^{-\rho/2}c_2=c_1 e − ρ /2 c 2 = c 1 and g μ , 1 − g μ , 2 = r g_{\mu,1}-g_{\mu,2}=r g μ , 1 − g μ , 2 = r imply
− F μ 1 + e − ρ / 2 F μ 2 = 2 c 1 [ − ∇ 2 ρ + g μ , 1 ⊗ g μ , 1 − g μ , 2 ⊗ g μ , 2 + Ξ ρ I 3 ] , -F_{\mu1}+e^{-\rho/2}F_{\mu2}=2c_1\left[-\nabla^2\rho+g_{\mu,1}\otimes g_{\mu,1}-g_{\mu,2}\otimes g_{\mu,2}+\Xi_\rho I_3\right], − F μ 1 + e − ρ /2 F μ 2 = 2 c 1 [ − ∇ 2 ρ + g μ , 1 ⊗ g μ , 1 − g μ , 2 ⊗ g μ , 2 + Ξ ρ I 3 ] ,
Ξ ρ = Δ ρ − g Σ ⋅ r . (260) \Xi_\rho=\Delta\rho-g_\Sigma\cdot r. \tag*{(260)} Ξ ρ = Δ ρ − g Σ ⋅ r . (260)
Both identities
g μ , 1 ⊗ g μ , 1 − g μ , 2 ⊗ g μ , 2 = 1 2 ( g Σ ⊗ r + r ⊗ g Σ ) , g_{\mu,1}\otimes g_{\mu,1}-g_{\mu,2}\otimes g_{\mu,2}=\frac{1}{2}(g_\Sigma\otimes r+r\otimes g_\Sigma), g μ , 1 ⊗ g μ , 1 − g μ , 2 ⊗ g μ , 2 = 2 1 ( g Σ ⊗ r + r ⊗ g Σ ) ,
r ⊗ g μ , 1 + g μ , 2 ⊗ r = 1 2 ( g Σ ⊗ r + r ⊗ g Σ ) (261) r\otimes g_{\mu,1}+g_{\mu,2}\otimes r=\frac{1}{2}(g_\Sigma\otimes r+r\otimes g_\Sigma) \tag*{(261)} r ⊗ g μ , 1 + g μ , 2 ⊗ r = 2 1 ( g Σ ⊗ r + r ⊗ g Σ ) (261)
follow by writing g μ , 1 = g μ , 2 + r g_{\mu,1}=g_{\mu,2}+r g μ , 1 = g μ , 2 + r . Together with − g μ , 2 b ρ T = − ( c 1 / κ 1 ) g μ , 2 ⊗ r -g_{\mu,2}b_\rho^T=-(c_1/\kappa_1)g_{\mu,2}\otimes r − g μ , 2 b ρ T = − ( c 1 / κ 1 ) g μ , 2 ⊗ r , they give the exact expansion
K ρ / c 1 = 2 ( κ 1 − 1 − 1 ) ∇ 2 ρ + ( 1 − 1 2 κ 1 ) ( g Σ ⊗ r + r ⊗ g Σ ) − 2 κ 1 r ⊗ k 1 + 2 Ξ ρ I 3 . (262) K_\rho/c_1=2(\kappa_1^{-1}-1)\nabla^2\rho+\left(1-\frac{1}{2\kappa_1}\right)(g_\Sigma\otimes r+r\otimes g_\Sigma)-\frac{2}{\kappa_1}r\otimes k_1+2\Xi_\rho I_3. \tag*{(262)} K ρ / c 1 = 2 ( κ 1 − 1 − 1 ) ∇ 2 ρ + ( 1 − 2 κ 1 1 ) ( g Σ ⊗ r + r ⊗ g Σ ) − κ 1 2 r ⊗ k 1 + 2 Ξ ρ I 3 . (262)
No κ 2 \kappa_2 κ 2 appears in this identity. Define
β = 2 k 1 − ( 2 κ 1 − 1 ) g Σ 4 ( 1 − κ 1 ) , E ρ = tf ( sym K ρ ) 2 c 1 ( κ 1 − 1 − 1 ) . (263) \beta=\frac{2k_1-(2\kappa_1-1)g_\Sigma}{4(1-\kappa_1)},\qquad E_\rho=\frac{\operatorname{tf}(\operatorname{sym}K_\rho)}{2c_1(\kappa_1^{-1}-1)}. \tag*{(263)} β = 4 ( 1 − κ 1 ) 2 k 1 − ( 2 κ 1 − 1 ) g Σ , E ρ = 2 c 1 ( κ 1 − 1 − 1 ) tf ( sym K ρ ) . (263)
The tensors tf and sym are defined in (204) and (233) ; k 1 k_1 k 1 , g Σ g_\Sigma g Σ are defined in (258) . We apply the symmetric trace-free projection to (262) . With
S g = g Σ ⊗ r + r ⊗ g Σ , S k = k 1 ⊗ r + r ⊗ k 1 , (264) S_g = g_\Sigma\otimes r + r \otimes g_\Sigma,\qquad S_k = k_1 \otimes r + r \otimes k_1, \tag*{(264)} S g = g Σ ⊗ r + r ⊗ g Σ , S k = k 1 ⊗ r + r ⊗ k 1 , (264)
the first-order terms after division by 2 ( κ 1 − 1 − 1 ) 2(\kappa_1^{-1}-1) 2 ( κ 1 − 1 − 1 ) are
2 κ 1 − 1 4 ( 1 − κ 1 ) S g − 1 2 ( 1 − κ 1 ) S k = − ( β ⊗ r + r ⊗ β ) . (265) \frac{2\kappa_1-1}{4(1-\kappa_1)}S_g-\frac{1}{2(1-\kappa_1)}S_k=-(\beta\otimes r+r\otimes\beta). \tag*{(265)} 4 ( 1 − κ 1 ) 2 κ 1 − 1 S g − 2 ( 1 − κ 1 ) 1 S k = − ( β ⊗ r + r ⊗ β ) . (265)
The scalar term 2 Ξ ρ I 3 2\Xi_\rho I_3 2 Ξ ρ I 3 vanishes under the trace-free projection. This proves, without any discarded error,
P β ρ = E ρ , ∥ E ρ ∥ C 0 ≤ C ε r i g . (266) P_\beta\rho=E_\rho,\qquad\lVert E_\rho\rVert_{C^0}\leq C\varepsilon_{\mathrm{rig}}. \tag*{(266)} P β ρ = E ρ , ∥ E ρ ∥ C 0 ≤ C ε rig . (266)
The source is real, because it is defined from the real material tensor, even if the auxiliary residual is complex.
Here is an explicit way to fix the drift class before using coercivity. Let K a K_a K a , G a G_a G a (a = 0 , 1 a=0,1 a = 0 , 1 ) bound ∥ ∇ a k 1 ∥ ∞ \lVert\nabla^a k_1\rVert_\infty ∥ ∇ a k 1 ∥ ∞ , ∥ ∇ a g Σ ∥ ∞ \lVert\nabla^a g_\Sigma\rVert_\infty ∥ ∇ a g Σ ∥ ∞ , respectively, and let L 1 L_1 L 1 bound ∥ ∇ κ 1 ∥ ∞ \lVert\nabla\kappa_1\rVert_\infty ∥ ∇ κ 1 ∥ ∞ , all over the entire material class (225) . They are finite functions of M 4 M_4 M 4 , c 0 c_0 c 0 by the chain rule. With the convention ∥ β ∥ W 1 , ∞ = ∥ β ∥ ∞ + ∥ ∇ β ∥ ∞ \lVert\beta\rVert_{W^{1,\infty}}=\lVert\beta\rVert_\infty+\lVert\nabla\beta\rVert_\infty ∥ β ∥ W 1 , ∞ = ∥ β ∥ ∞ + ∥ ∇ β ∥ ∞ , a permissible fixed bound is
B β = 2 K 0 + G 0 + 2 K 1 + 2 L 1 G 0 + G 1 4 ν − + L 1 ( 2 K 0 + G 0 ) 4 ν − 2 . (267) B_\beta=\frac{2K_0+G_0+2K_1+2L_1G_0+G_1}{4\nu_-}+\frac{L_1(2K_0+G_0)}{4\nu_-^2}. \tag*{(267)} B β = 4 ν − 2 K 0 + G 0 + 2 K 1 + 2 L 1 G 0 + G 1 + 4 ν − 2 L 1 ( 2 K 0 + G 0 ) . (267)
Indeed ∣ 2 κ 1 − 1 ∣ ≤ 1 |2\kappa_1-1|\leq1 ∣2 κ 1 − 1∣ ≤ 1 and differentiating the numerator and ( 1 − κ 1 ) − 1 (1-\kappa_1)^{-1} ( 1 − κ 1 ) − 1 proves this bound. In particular β \beta β is real and uniformly W 1 , ∞ W^{1,\infty} W 1 , ∞ before any small estimate for ρ \rho ρ is known.
Apply the scalar estimate to the undifferentiated source. Since ρ ∈ V R 0 \rho\in V_{R_0} ρ ∈ V R 0 , Theorem 7.1 gives
∥ ρ ∥ H 2 + ∥ ρ ∥ ∞ ≤ C ∥ E ρ ∥ 2 ≤ C ∣ B R ∣ 1 / 2 ∥ E ρ ∥ C 0 ≤ C ε r i g . (268) \lVert\rho\rVert_{H^2}+\lVert\rho\rVert_\infty\leq C\lVert E_\rho\rVert_2\leq C|B_R|^{1/2}\lVert E_\rho\rVert_{C^0}\leq C\varepsilon_{\mathrm{rig}}. \tag*{(268)} ∥ ρ ∥ H 2 + ∥ ρ ∥ ∞ ≤ C ∥ E ρ ∥ 2 ≤ C ∣ B R ∣ 1/2 ∥ E ρ ∥ C 0 ≤ C ε rig . (268)
No derivative of E ρ E_\rho E ρ occurs. The a priori bound ∣ ρ ∣ ≤ log ( M 4 / c 0 ) |\rho|\leq\log(M_4/c_0) ∣ ρ ∣ ≤ log ( M 4 / c 0 ) and μ 1 − μ 2 = μ 2 ( e ρ − 1 ) \mu_1-\mu_2=\mu_2(e^\rho-1) μ 1 − μ 2 = μ 2 ( e ρ − 1 ) give ∥ μ 1 − μ 2 ∥ ∞ ≤ C ε r i g \lVert\mu_1-\mu_2\rVert_\infty\leq C\varepsilon_{\mathrm{rig}} ∥ μ 1 − μ 2 ∥ ∞ ≤ C ε rig . Finally, the recovery formula for λ \lambda λ in (32) is Lipschitz on [ c 0 , M 4 ] × [ κ − , 1 − ν − ] [c_0,M_4]\times[\kappa_-,1-\nu_-] [ c 0 , M 4 ] × [ κ − , 1 − ν − ] : its partial derivatives have absolute values at most ν − − 1 \nu_-^{-1} ν − − 1 and M 4 ν − − 1 M_4\nu_-^{-1} M 4 ν − − 1 . Combine this fact with (252) to conclude (237) . □
Remark. The estimate is linear in the transport residual and boundary normalization error, so this step causes no further loss of their power. Smallness is required only for these errors, not for the material gradients or ρ \rho ρ . Neither invertibility of H 0 H_0 H 0 nor equality of κ 1 \kappa_1 κ 1 and κ 2 \kappa_2 κ 2 is assumed.
The bounds (241) , (247) , (245) , and (257) use at most two spatial derivatives of the transport residual. The C 4 C^4 C 4 material prior (225) supplies the required C x 2 C_x^2 C x 2 connection bounds, while (198) uses three spatial derivatives of the projection error. Finally, Theorem 7.1 controls ρ \rho ρ without differentiating its L 2 L^2 L 2 source.
Applied to the projected transport, the rigidity estimate bounds the coefficient difference in terms of the Fourier boundary error.
Corollary 7.4. Assume the material hypotheses of Theorem 7.3 . Let H H H be the global invertible endomorphism constructed in Proposition 5.4 ; it satisfies the second equation in (169) . For m ≥ 4 m\geq4 m ≥ 4 , use β m \beta_m β m from (17) and H 0 H_0 H 0 from (186) . With the fixed constants and thresholds of (200) and (202) ,
∥ λ 1 − λ 2 ∥ L ∞ ( B R ) + ∥ μ 1 − μ 2 ∥ L ∞ ( B R ) ≤ C ω F β m . (269) \lVert\lambda_1-\lambda_2\rVert_{L^\infty(B_R)}+\lVert\mu_1-\mu_2\rVert_{L^\infty(B_R)}\leq C\omega_F^{\beta_m}. \tag*{(269)} ∥ λ 1 − λ 2 ∥ L ∞ ( B R ) + ∥ μ 1 − μ 2 ∥ L ∞ ( B R ) ≤ C ω F β m . (269)
Proof. By Lemma 6.1 , the field H 0 = I H H_0=\mathbb{I}H H 0 = I H has the block form (229) . The projection bound in (189) , with m ≥ 4 m\geq4 m ≥ 4 , gives
∥ H 0 ∥ C x 3 C σ 0 ≤ C Π M H = B H . \lVert H_0\rVert_{C_x^3C_\sigma^0}\leq C_\Pi M_H=B_H. ∥ H 0 ∥ C x 3 C σ 0 ≤ C Π M H = B H .
Thus the projected field satisfies the algebraic and a priori hypotheses of Theorem 7.3 .
Choose its error bound to be ε r i g = C l i n k ω F β m \varepsilon_{\mathrm{rig}}=C_{\mathrm{link}}\omega_{\mathrm{F}}^{\beta_m} ε rig = C link ω F β m . Equations (201) and (202) give
∥ T ( H 0 ) ∥ C x 2 C ξ 0 + ∥ H 0 − I ∥ C 0 ( ∂ B R × C P 1 ) ≤ ε r i g , \left\|\mathcal{T}(H_0)\right\|_{C^2_xC^0_\xi}+\left\|H_0-I\right\|_{C^0(\partial B_R\times\mathbb{CP}^1)}\leq\varepsilon_{\mathrm{rig}}, ∥ T ( H 0 ) ∥ C x 2 C ξ 0 + ∥ H 0 − I ∥ C 0 ( ∂ B R × CP 1 ) ≤ ε rig ,
ε r i g < ε r i g , 0 / 2 \varepsilon_{\mathrm{rig}}<\varepsilon_{\mathrm{rig},0}/2 ε rig < ε rig , 0 /2 .
Apply Theorem 7.3 and retain the two material terms in (237) . Their sum is at most C C l i n k ω F β m CC_{\mathrm{link}}\omega_{\mathrm{F}}^{\beta_m} C C link ω F β m , which is (269) after adjusting the fixed constant. If ω F = 0 \omega_{\mathrm{F}}=0 ω F = 0 , both errors vanish and the same theorem applies with ε r i g = 0 \varepsilon_{\mathrm{rig}}=0 ε rig = 0 .
Common-exterior geometry and quantitative estimates
On a fixed common-exterior class, the moment, transport, projection, and rigidity bounds yield a single coefficient estimate. The positive-error and exact-data cases are treated separately.
Fix m ≥ 4 m\geq4 m ≥ 4 , s = 2 m + 14 s=2m+14 s = 2 m + 14 , and a bounded smooth measurement domain D D D . Choose fixed concentric balls with
D ‾ ⋐ B R s u p , R s u p < R 0 < R s < R , E = { x : R 0 < ∣ x ∣ < R } . (270) \overline{D}\Subset B_{R_{\mathrm{sup}}},\qquad R_{\mathrm{sup}}<R_0<R_s<R,\qquad E=\{x:R_0<|x|<R\}. \tag*{(270)} D ⋐ B R sup , R sup < R 0 < R s < R , E = { x : R 0 < ∣ x ∣ < R } . (270)
Fix the background ( λ b g , μ b g ) (\lambda_{\mathrm{bg}},\mu_{\mathrm{bg}}) ( λ bg , μ bg ) of (15) , constant beyond a fixed ball. Each admissible material is a real C s C^s C s pair on R 3 \mathbb{R}^3 R 3 satisfying (15) , with fixed C s C^s C s bounds and fixed positive lower bounds for μ \mu μ and 3 λ + 2 μ 3\lambda+2\mu 3 λ + 2 μ . All working balls, leaf cylinders, support margins, and background bounds are fixed before the noise is chosen. In particular the two materials agree on E E E and in a neighborhood of the working sphere ∂ B R \partial B_R ∂ B R . These are assumptions of the auxiliary common-exterior problem; the original-domain problem supplies them only after the extension step in § 9 . The high-order physical frame is chosen for each material pair before varying degree, width, or h h h .
Use the coefficient distance d D \mathfrak{d}_D d D and exponent β m \beta_m β m from (17) . Define the remaining exponents by
b F = γ ′ 14 , a c h a i n = b F β m , 0 < γ ′ < γ ∗ < 1. (271) b_{\mathrm{F}}=\frac{\gamma'}{14},\qquad a_{\mathrm{chain}}=b_{\mathrm{F}}\beta_m,\qquad0<\gamma'<\gamma_*<1. \tag*{(271)} b F = 14 γ ′ , a chain = b F β m , 0 < γ ′ < γ ∗ < 1. (271)
The number γ ∗ \gamma_* γ ∗ is the fixed exponent in (142) . For each fixed γ ′ \gamma' γ ′ , Theorem 4.2 , Lemma 5.1 , and Corollary 7.4 give
ς = η D 1 / 14 , ω F ≤ C F η D b F , \varsigma=\eta_D^{1/14},\qquad\omega_{\mathrm{F}}\leq C_{\mathrm{F}}\eta_D^{b_{\mathrm{F}}}, ς = η D 1/14 , ω F ≤ C F η D b F ,
d D ≤ C 6 ε r i g ≤ C 6 C R η D a c h a i n , C R = C l i n k C F β m . (272) \mathfrak{d}_D\leq C_6\varepsilon_{\mathrm{rig}}\leq C_6C_R\eta_D^{a_{\mathrm{chain}}},\qquad C_R=C_{\mathrm{link}}C_{\mathrm{F}}^{\beta_m}. \tag*{(272)} d D ≤ C 6 ε rig ≤ C 6 C R η D a chain , C R = C link C F β m . (272)
Here η D \eta_D η D is defined in (17) , ε r i g \varepsilon_{\mathrm{rig}} ε rig satisfies (236) , and (201) supplies its bound. All constants are fixed by the class, geometry and γ ′ \gamma' γ ′ ; enlarge C 0 C_0 C 0 so that C 0 ≥ 1 C_0\geq1 C 0 ≥ 1 . Choose η † > 0 \eta_\dagger>0 η † > 0 smaller than the joint threshold in (202) and the width threshold ς ∗ 14 \varsigma_*^{14} ς ∗ 14 :
η † = 1 2 min { 1 , η a l l , ς ∗ 14 } . (273) \eta_\dagger=\frac{1}{2}\min\{1,\eta_{\mathrm{all}},\varsigma_*^{14}\}. \tag*{(273)} η † = 2 1 min { 1 , η all , ς ∗ 14 } . (273)
Then (272) holds for 0 < h < h ∗ 0<h<h_* 0 < h < h ∗ and 0 < η D < η † 0<\eta_D<\eta_\dagger 0 < η D < η † . The common h ∗ h_* h ∗ is also smaller than the thresholds in Theorem 4.2 and Theorem 3.2 .
Under the fixed thresholds above, the common-exterior estimates yield the following logarithmic bound and its exact-data conclusion.
Proposition 7.5. Fix the common-exterior class and geometry of Subsection 7.3. For every 0 < α < γ ∗ β m / 14 0<\alpha<\gamma_*\beta_m/14 0 < α < γ ∗ β m /14 , there exist C c o m , δ c o m , 0 > 0 C_{\mathrm{com}},\delta_{\mathrm{com},0}>0 C com , δ com , 0 > 0 , depending only on this fixed class, its geometry, and α \alpha α , such that
d D ≤ C c o m ∣ log δ D ∣ − α , 0 < δ D < δ c o m , 0 . (274) \mathfrak{d}_D\leq C_{\mathrm{com}}|\log\delta_D|^{-\alpha},\qquad0<\delta_D<\delta_{\mathrm{com},0}. \tag*{(274)} d D ≤ C com ∣ log δ D ∣ − α , 0 < δ D < δ com , 0 . (274) Here d D \mathfrak{d}_D d D and δ D \delta_D δ D are defined in (17) and (16) , and γ ∗ \gamma_* γ ∗ in (142) . At δ D = 0 \delta_D=0 δ D = 0 , the two pairs coincide on D D D .
Proof. First choose
γ ′ = 1 2 ( γ ∗ + 14 α β m ) . (275) \gamma'=\frac{1}{2}\left(\gamma_*+\frac{14\alpha}{\beta_m}\right). \tag*{(275)} γ ′ = 2 1 ( γ ∗ + β m 14 α ) . (275)
Then 0 < γ ′ < γ ∗ 0<\gamma'<\gamma_* 0 < γ ′ < γ ∗ and a c h a i n > α a_{\mathrm{chain}}>\alpha a chain > α . All constants in (272) , including those depending on this fixed γ ′ \gamma' γ ′ , are fixed before choosing the noise.
For positive noise put L = ∣ log δ D ∣ L=\lvert\log\delta_D\rvert L = ∣ log δ D ∣ and
h = 2 C 0 L , η D = 2 C 0 L + L 2 e − L / 2 4 C 0 2 , ς = η D 1 / 14 . (276) h=\frac{2C_0}{L},\qquad\eta_D=\frac{2C_0}{L}+\frac{L^2e^{-L/2}}{4C_0^2},\qquad\varsigma=\eta_D^{1/14}. \tag*{(276)} h = L 2 C 0 , η D = L 2 C 0 + 4 C 0 2 L 2 e − L /2 , ς = η D 1/14 . (276)
The elementary bound L 3 e − L / 2 ≤ 216 L^3e^{-L/2}\le216 L 3 e − L /2 ≤ 216 for L > 0 L>0 L > 0 gives
η D ≤ C η L , C η = 2 C 0 + 54 C 0 2 . (277) \eta_D\le\frac{C_\eta}{L},\qquad C_\eta=2C_0+\frac{54}{C_0^2}. \tag*{(277)} η D ≤ L C η , C η = 2 C 0 + C 0 2 54 . (277)
One explicit sufficient choice is
L 0 = max { 1 , 4 C 0 h ∗ , 2 C η η † } , δ c o m , 0 = e − L 0 . (278) L_0=\max\left\{1,\frac{4C_0}{h_*},\frac{2C_\eta}{\eta^\dagger}\right\},\qquad\delta_{\mathrm{com},0}=e^{-L_0}. \tag*{(278)} L 0 = max { 1 , h ∗ 4 C 0 , η † 2 C η } , δ com , 0 = e − L 0 . (278)
For 0 < δ D < δ c o m , 0 0<\delta_D<\delta_{\mathrm{com},0} 0 < δ D < δ com , 0 , this gives h < h ∗ / 2 h<h_*/2 h < h ∗ /2 and η D < η † / 2 \eta_D<\eta^\dagger/2 η D < η † /2 . Using Theorem 7.3 and (272) , we obtain
∂ D ≤ C 6 C R η D a c h a i n L − a c h a i n ≤ C c o m L − α . (279) \boldsymbol{\partial}_D\le C_6C_R\eta_D^{a_{\mathrm{chain}}}L^{-a_{\mathrm{chain}}}\le C_{\mathrm{com}}L^{-\alpha}. \tag*{(279)} ∂ D ≤ C 6 C R η D a chain L − a chain ≤ C com L − α . (279)
The threshold contains neither the polynomial degree nor the linear amplitude size.
For δ D = 0 \delta_D=0 δ D = 0 , use every 0 < h < min { h ∗ , η † } 0<h<\min\{h_*,\eta^\dagger\} 0 < h < min { h ∗ , η † } and set η D = h \eta_D=h η D = h , ς = h 1 / 14 \varsigma=h^{1/14} ς = h 1/14 . The same estimates give
∂ D ≤ C 6 C R h a c h a i n . (280) \boldsymbol{\partial}_D\le C_6C_Rh^{a_{\mathrm{chain}}}. \tag*{(280)} ∂ D ≤ C 6 C R h a chain . (280)
The material pair and the left side are fixed, while the constants are uniform and a c h a i n > 0 a_{\mathrm{chain}}>0 a chain > 0 . Letting h ↓ 0 h\downarrow0 h ↓ 0 in (280) , we obtain ∂ D = 0 \boldsymbol{\partial}_D=0 ∂ D = 0 . This argument uses no limit of the transports and never evaluates log 0 \log0 log 0 . □
Remark. At zero original-domain error, the paired extension construction gives a family within one fixed class. This proposition applies separately at each positive tolerance, with constants independent of the tolerance.
Sharp stability of boundary derivatives
Boundary stability follows from localized elastic Dirichlet energies through a finite Taylor expansion, coefficient extraction, and recovery of successive normal derivatives. Boundary-localized perturbations establish sharpness and the top-order obstruction under strict prior slack.
Local hypotheses and the measurement norm
Fix an integer s ≥ 2 s\ge2 s ≥ 2 and the material class C s ( Ω ; M , c 0 ) \mathcal{C}^s(\Omega;M,c_0) C s ( Ω ; M , c 0 ) of (4) . All boundary estimates use only this class, not prescribed exterior coefficients. Theorem 1.1 uses s = 22 s=22 s = 22 , while Proposition 10.1 retains the general orders of (94) ; the present section allows every integer s ≥ 2 s\ge2 s ≥ 2 . Constants depend on the fixed class, the domain, the boundary atlas (282) , and the tests (287) . There is no uniform modulus of continuity assumption for the derivatives of order s s s .
We use the physical variational forms and the uniform Dirichlet estimate (2.1)–(2.6) from § 2.1 . The abbreviation a a a is defined in (24) .
Use J q J_q J q from (9) ; all its derivatives are Cartesian traces from the interior. For the outward normal ν \nu ν fix the tubular map and its normal derivative convention
Φ ( p , t ) = p + t ν ( p ) , ∂ ν a f ( p ) = ∂ t a ( f ∘ Φ ) ( p , t ) ∣ t = 0 − . (281) \Phi(p,t)=p+t\nu(p),\qquad\partial_\nu^a f(p)=\left.\partial_t^a(f\circ\Phi)(p,t)\right|_{t=0^-}. \tag*{(281)} Φ ( p , t ) = p + t ν ( p ) , ∂ ν a f ( p ) = ∂ t a ( f ∘ Φ ) ( p , t ) ∣ t = 0 − . (281)
The derivative follows a straight normal segment; no unspecified extension of the normal vector field is used.
Fixed boundary coordinates and annular tests
The boundary tests use fixed normal coordinates and annular tangential frequencies, avoiding the zero-frequency singularity while retaining uniform trace norms.
We use only (1.4). The atlas and its cutoffs are fixed before either candidate and before the scale is chosen. The geometric C s + 1 C^{s+1} C s + 1 bounds needed for the coordinate tensor are fixed bounds on the domain, not additional material derivatives.
For each p ∈ ∂ Ω p \in\partial\Omega p ∈ ∂ Ω choose the boundary normal chart Ψ p \Psi_p Ψ p in (8.2), with z 3 ≥ 0 z_3 \ge0 z 3 ≥ 0 inward. The sizes, inverse bounds and every required finite geometric derivative are uniform after subdividing a finite atlas. More precisely, choose a surface parametrization P p P_p P p and set
Ψ p ( z ′ , z 3 ) = P p ( z ′ ) − z 3 ν ( P p ( z ′ ) ) , P p ( 0 ) = p , R p = D Ψ p ( 0 ) ∈ O ( 3 ) . (282) \Psi_p(z',z_3)=P_p(z')-z_3\nu(P_p(z')), \qquad P_p(0)=p, \qquad R_p=D\Psi_p(0)\in O(3). \tag*{(282)} Ψ p ( z ′ , z 3 ) = P p ( z ′ ) − z 3 ν ( P p ( z ′ )) , P p ( 0 ) = p , R p = D Ψ p ( 0 ) ∈ O ( 3 ) . (282)
Tangential coordinates are linearly adjusted at their center to obtain this orthogonality. Throughout the chart, not merely at its center,
∂ z ′ γ ∂ z 3 a ( f ∘ Ψ p ) ∣ z 3 = 0 = ( − 1 ) a ∂ z ′ γ [ ∂ ν a f ∘ P p ] , ∣ γ ∣ + a ≤ s . (283) \left.\partial_{z'}^\gamma\partial_{z_3}^a(f\circ\Psi_p)\right|_{z_3=0}=(-1)^a\partial_{z'}^\gamma[\partial_\nu^a f\circ P_p], \qquad|\gamma|+a\le s. \tag*{(283)} ∂ z ′ γ ∂ z 3 a ( f ∘ Ψ p ) z 3 = 0 = ( − 1 ) a ∂ z ′ γ [ ∂ ν a f ∘ P p ] , ∣ γ ∣ + a ≤ s . (283)
The equality follows by holding P p ( z ′ ) P_p(z') P p ( z ′ ) and its normal fixed while performing the a a a normal derivatives, and then differentiating in z ′ z' z ′ . Rotate the physical displacement by the constant matrix R p R_p R p , not by a moving frame. The physical Hermitian energy in the chart then has the form
a ( v , w ) = ∫ A p ( z ) ∇ z v : ∇ z w ‾ d z , A p ( 0 ) ∇ v : ∇ w ‾ = λ ( p ) div v div w ‾ + 2 μ ( p ) ε ( v ) : ε ( w ) ‾ . (284) \begin{aligned}
a(v,w)&=\int A_p(z)\nabla_zv:\overline{\nabla_zw}\,dz,\\
A_p(0)\nabla v:\overline{\nabla w}&=\lambda(p)\operatorname{div}v\overline{\operatorname{div}w}+2\mu(p)\varepsilon(v):\overline{\varepsilon(w)}. \tag*{(284)}
\end{aligned} a ( v , w ) A p ( 0 ) ∇ v : ∇ w = ∫ A p ( z ) ∇ z v : ∇ z w d z , = λ ( p ) div v div w + 2 μ ( p ) ε ( v ) : ε ( w ) . (284)
Here A p A_p A p includes the Jacobian and the inverse coordinate matrices. The transformed tensor A p A_p A p is uniformly bounded in C s C^s C s . Away from z = 0 z=0 z = 0 it need not be an isotropic tensor in these coordinates. There are no lower-order terms, because the component rotation is constant.
For boundary profiles use the unitary two-dimensional Fourier transform
ϕ ^ ( ξ ) = ( 2 π ) − 1 ∫ R 2 e − i y ′ ⋅ ξ ϕ ( y ′ ) d y ′ . (285) \widehat{\phi}(\xi)=(2\pi)^{-1}\int_{\mathbb{R}^2}e^{-iy'\cdot\xi}\phi(y')\,dy'. \tag*{(285)} ϕ ( ξ ) = ( 2 π ) − 1 ∫ R 2 e − i y ′ ⋅ ξ ϕ ( y ′ ) d y ′ . (285)
Choose once and for all a nonzero scalar ψ ∂ ∈ C ∞ ( { 1 < ∣ ξ ∣ < 2 } ) \psi_{\partial}\in C^\infty(\{1<|\xi|<2\}) ψ ∂ ∈ C ∞ ({ 1 < ∣ ξ ∣ < 2 }) , normalized by the following conventions:
∫ R 2 ∣ ψ ∂ ( ξ ) ∣ 2 d ξ = 1 , ϖ = ∣ ξ ∣ , ξ ∈ R 2 ∖ { 0 } , \int_{\mathbb{R}^2}|\psi_{\partial}(\xi)|^2\,d\xi=1,\qquad\varpi=|\xi|,\qquad\xi\in\mathbb{R}^2\setminus\{0\}, ∫ R 2 ∣ ψ ∂ ( ξ ) ∣ 2 d ξ = 1 , ϖ = ∣ ξ ∣ , ξ ∈ R 2 ∖ { 0 } ,
ξ ^ = ( ξ 1 , ξ 2 , 0 ) / ϖ , ξ ^ ⊥ = ( − ξ 2 , ξ 1 , 0 ) / ϖ . (286) \widehat{\xi}=(\xi_1,\xi_2,0)/\varpi,\qquad\widehat{\xi}^{\perp}=(-\xi_2,\xi_1,0)/\varpi. \tag*{(286)} ξ = ( ξ 1 , ξ 2 , 0 ) / ϖ , ξ ⊥ = ( − ξ 2 , ξ 1 , 0 ) / ϖ . (286)
The two fixed Schwartz boundary tests are
ϕ S ^ ( ξ ) = ψ ∂ ( ξ ) ξ ^ ⊥ , ϕ N ^ ( ξ ) = ψ ∂ ( ξ ) e 3 . (287) \widehat{\phi^S}(\xi)=\psi_{\partial}(\xi)\widehat{\xi}^{\perp},\qquad\widehat{\phi^N}(\xi)=\psi_{\partial}(\xi)e_3. \tag*{(287)} ϕ S ( ξ ) = ψ ∂ ( ξ ) ξ ⊥ , ϕ N ( ξ ) = ψ ∂ ( ξ ) e 3 . (287)
Their Fourier supports avoid zero. Superscripts S , N S,N S , N refer to these shear and normal boundary tests. Let χ p \chi_p χ p be a fixed real cutoff in the chart, equal to one on a smaller half-ball. For either test ϕ \phi ϕ prescribe
f p , ϵ ϕ ( Ψ p ( z ′ , 0 ) ) = ϵ − 1 / 2 χ p ( z ′ , 0 ) R p ϕ ( z ′ / ϵ ) , f p , ϵ ϕ = 0 off the chart . (288) f^\phi_{p,\epsilon}(\Psi_p(z',0))=\epsilon^{-1/2}\chi_p(z',0)R_p\phi(z'/\epsilon),\qquad f^\phi_{p,\epsilon}=0\ \text{off the chart}. \tag*{(288)} f p , ϵ ϕ ( Ψ p ( z ′ , 0 )) = ϵ − 1/2 χ p ( z ′ , 0 ) R p ϕ ( z ′ / ϵ ) , f p , ϵ ϕ = 0 off the chart . (288)
The scale ϵ > 0 \epsilon>0 ϵ > 0 in this section is a boundary dilation scale, not the CGO parameter h h h . Although ϕ \phi ϕ is not compactly supported, the physical trace is. Multiplication by the fixed cutoff and chart norm equivalence give
sup p , 0 < ϵ < ϵ ∗ ∥ f p , ϵ ϕ ∥ H 1 / 2 ( ∂ Ω ) ≤ C ϕ . (289) \sup_{p,\ 0<\epsilon<\epsilon_*}\left\|f^\phi_{p,\epsilon}\right\|_{H^{1/2}(\partial\Omega)}\le C_\phi. \tag*{(289)} p , 0 < ϵ < ϵ ∗ sup f p , ϵ ϕ H 1/2 ( ∂ Ω ) ≤ C ϕ . (289)
Indeed the squared flat norm before cutoff is ∫ ( ϵ 2 + ∣ ξ ∣ 2 ) 1 / 2 ∣ ϕ ^ ( ξ ) ∣ 2 d ξ \int(\epsilon^2+|\xi|^2)^{1/2}|\widehat{\phi}(\xi)|^2d\xi ∫ ( ϵ 2 + ∣ ξ ∣ 2 ) 1/2 ∣ ϕ ( ξ ) ∣ 2 d ξ . The trace normalization in (8.9) is the one used in the DN pairing (2.2).
Frozen physical impedance and its inverse
The frozen elastic impedance gives the order-zero boundary response, whose dependence on the two Lamé parameters can be inverted. For classical boundary determination, see [24 ] . Let ( λ 0 , μ 0 ) (\lambda_0,\mu_0) ( λ 0 , μ 0 ) be frozen boundary values. In the ordered basis ( ξ ^ , ξ ^ ⊥ , e 3 ) (\hat{\xi},\hat{\xi}^{\perp},e_3) ( ξ ^ , ξ ^ ⊥ , e 3 ) , with e 3 e_3 e 3 inward, the physical traction uses − e 3 -e_3 − e 3 . The corresponding 3 × 3 3 \times3 3 × 3 boundary impedance multiplier, distinct from any interior amplitude, is
Z ∂ ( ξ ) = ϖ ( a Z 0 − i b Z 0 μ 0 0 i b Z 0 a Z ) , a Z = 2 μ 0 ( λ 0 + 2 μ 0 ) λ 0 + 3 μ 0 , b Z = 2 μ 0 2 λ 0 + 3 μ 0 . (290) Z_{\partial}(\xi)=\varpi\begin{pmatrix} a_Z & 0 & -ib_Z \\ 0 & \mu_0 & 0 \\ ib_Z & 0 & a_Z \end{pmatrix},\qquad a_Z=\frac{2\mu_0(\lambda_0+2\mu_0)}{\lambda_0+3\mu_0},\qquad b_Z=\frac{2\mu_0^2}{\lambda_0+3\mu_0}. \tag*{(290)} Z ∂ ( ξ ) = ϖ a Z 0 i b Z 0 μ 0 0 − i b Z 0 a Z , a Z = λ 0 + 3 μ 0 2 μ 0 ( λ 0 + 2 μ 0 ) , b Z = λ 0 + 3 μ 0 2 μ 0 2 . (290)
Indeed the profile in (297) has boundary derivative ϖ ( b − a ) \varpi(b-a) ϖ ( b − a ) and boundary divergence ϖ ( 1 − g 0 ) ( i a ∥ − a 3 ) \varpi(1-g_0)(ia_{\parallel}-a_3) ϖ ( 1 − g 0 ) ( i a ∥ − a 3 ) . The three traction components are respectively − μ 0 ( ∂ t u ∥ + i ϖ u 3 ) -\mu_0(\partial_tu_{\parallel}+i\varpi u_3) − μ 0 ( ∂ t u ∥ + iϖ u 3 ) , − μ 0 ∂ t u ⊥ -\mu_0\partial_tu_{\perp} − μ 0 ∂ t u ⊥ , and − λ 0 div u − 2 μ 0 ∂ t u 3 -\lambda_0\operatorname{div}u-2\mu_0\partial_tu_3 − λ 0 div u − 2 μ 0 ∂ t u 3 . Substituting these boundary derivatives into the three traction components, we obtain (290) . In particular the sign of each off-diagonal entry is tied to the outward traction, not the inward coordinate. The algebraic inverse needed at order zero is
λ 0 = μ 0 3 a Z − 4 μ 0 2 μ 0 − a Z , 2 μ 0 − a Z = 2 μ 0 2 λ 0 + 3 μ 0 > 0. (291) \lambda_0=\mu_0\frac{3a_Z-4\mu_0}{2\mu_0-a_Z},\qquad2\mu_0-a_Z=\frac{2\mu_0^2}{\lambda_0+3\mu_0}>0. \tag*{(291)} λ 0 = μ 0 2 μ 0 − a Z 3 a Z − 4 μ 0 , 2 μ 0 − a Z = λ 0 + 3 μ 0 2 μ 0 2 > 0. (291)
The class bounds keep this denominator uniformly away from zero, so this inverse is uniformly Lipschitz on the admissible value range.
Decaying half-space corrections without low frequencies
Half-space corrections retain annular tangential frequency support and admit weighted decay estimates uniform in the material class.
Freeze the energy form (284) at the boundary point p p p :
( λ 0 , μ 0 ) = ( λ ( p ) , μ ( p ) ) , (\lambda_0,\mu_0)=(\lambda(p),\mu(p)), ( λ 0 , μ 0 ) = ( λ ( p ) , μ ( p )) ,
a 0 ( V , W ) = ∫ R + 3 A p ( 0 ) ∇ V : ∇ W d y , a_0(V,W)=\int_{\mathbb{R}^3_+} A_p(0)\nabla V:\nabla W\,dy, a 0 ( V , W ) = ∫ R + 3 A p ( 0 ) ∇ V : ∇ W d y ,
L 0 + = − μ 0 Δ − ( λ 0 + μ 0 ) ∇ div . (292) L_0^+=-\mu_0\Delta-(\lambda_0+\mu_0)\nabla\operatorname{div}. \tag*{(292)} L 0 + = − μ 0 Δ − ( λ 0 + μ 0 ) ∇ div . (292)
Here L 0 + L_0^+ L 0 + is the positive 3 × 3 3 \times3 3 × 3 second-order elastic operator associated with a 0 a_0 a 0 . For tangential frequency ξ \xi ξ and real normal frequency ξ 3 \xi_3 ξ 3 , its full-space inverse symbol is
[ μ 0 ( ϖ 2 + ξ 3 2 ) I 3 + ( λ 0 + μ 0 ) ξ f u l l ξ f u l l T ] − 1 = I 3 μ 0 ( ϖ 2 + ξ 3 2 ) − λ 0 + μ 0 μ 0 ( λ 0 + 2 μ 0 ) ξ f u l l ξ f u l l T ( ϖ 2 + ξ 3 2 ) 2 , \left[\mu_0(\varpi^2+\xi_3^2)I_3+(\lambda_0+\mu_0)\xi_{\mathrm{full}}\xi_{\mathrm{full}}^T\right]^{-1}
=\frac{I_3}{\mu_0(\varpi^2+\xi_3^2)}-\frac{\lambda_0+\mu_0}{\mu_0(\lambda_0+2\mu_0)}\frac{\xi_{\mathrm{full}}\xi_{\mathrm{full}}^T}{(\varpi^2+\xi_3^2)^2}, [ μ 0 ( ϖ 2 + ξ 3 2 ) I 3 + ( λ 0 + μ 0 ) ξ full ξ full T ] − 1 = μ 0 ( ϖ 2 + ξ 3 2 ) I 3 − μ 0 ( λ 0 + 2 μ 0 ) λ 0 + μ 0 ( ϖ 2 + ξ 3 2 ) 2 ξ full ξ full T ,
ξ f u l l = ( ξ 1 , ξ 2 , ξ 3 ) T . (293) \xi_{\mathrm{full}}=(\xi_1,\xi_2,\xi_3)^T. \tag*{(293)} ξ full = ( ξ 1 , ξ 2 , ξ 3 ) T . (293)
The material denominators in (293) and (297) are uniformly separated from zero by (1.4). Restricting the tangential Fourier support to a fixed annulus avoids the low-frequency singularity and preserves exponential normal decay through the correction equations.
Lemma 8.1. Use the finite-energy spaces
H ˙ 1 ( R + 3 ; C 3 ) = { V ∈ L 6 ( R + 3 ; C 3 ) : ∇ V ∈ L 2 } , \dot{H}^1(\mathbb{R}^3_+;\mathbb{C}^3)=\{V\in L^6(\mathbb{R}^3_+;\mathbb{C}^3):\nabla V\in L^2\}, H ˙ 1 ( R + 3 ; C 3 ) = { V ∈ L 6 ( R + 3 ; C 3 ) : ∇ V ∈ L 2 } , H ˙ 0 1 ( R + 3 ; C 3 ) = { V ∈ H ˙ 1 ( R + 3 ; C 3 ) : Tr V = 0 } . (294) \dot{H}^1_0(\mathbb{R}^3_+;\mathbb{C}^3)=\{V\in\dot{H}^1(\mathbb{R}^3_+;\mathbb{C}^3):\operatorname{Tr}V=0\}. \tag*{(294)} H ˙ 0 1 ( R + 3 ; C 3 ) = { V ∈ H ˙ 1 ( R + 3 ; C 3 ) : Tr V = 0 } . (294) Consider profiles with tangential Fourier transforms given by finite sums
V ^ ( ξ , t ) = e − ϖ t ∑ n = 0 N p o l t n v n ( ξ ) , t ≥ 0 , v n ∈ C c ∞ ( { 1 < ϖ < 2 } ; C 3 ) . (295) \widehat{V}(\xi,t)=e^{-\varpi t}\sum_{n=0}^{N_{\mathrm{pol}}}t^n v_n(\xi),\qquad t\geq0,\qquad v_n\in C^\infty_c(\{1<\varpi<2\};\mathbb{C}^3). \tag*{(295)} V ( ξ , t ) = e − ϖ t n = 0 ∑ N pol t n v n ( ξ ) , t ≥ 0 , v n ∈ C c ∞ ({ 1 < ϖ < 2 } ; C 3 ) . (295)
The class (295) is preserved by spatial differentiation and multiplication by polynomials in ( y ′ , t ) (y',t) ( y ′ , t ) . If F F F belongs to this class, the zero-trace finite-energy solution of L 0 + V = F L_0^+ V=F L 0 + V = F also belongs to it. The constant half-space solution with any annular smooth boundary Fourier data belongs to the same class. For every fixed pair of integers B , N ≥ 0 B,N\geq0 B , N ≥ 0 , the seminorms
∑ ∣ α ∣ ≤ B ∥ ( 1 + ∣ y ∣ ) N ∂ α V ∥ L 2 ( R + 3 ) (296) \sum_{|\alpha|\leq B}\left\|(1+|y|)^N\partial^\alpha V\right\|_{L^2(\mathbb{R}^3_+)} \tag*{(296)} ∣ α ∣ ≤ B ∑ ( 1 + ∣ y ∣ ) N ∂ α V L 2 ( R + 3 ) (296)
are bounded by finitely many seminorms of the input coefficients, uniformly on the frozen material range. The solution maps for L 0 + L_0^+ L 0 + in (292) depend smoothly and locally Lipschitz-continuously, in the seminorms (296) , on the frozen material values and the finitely many input coefficients.
Proof. Multiplication by y ′ y' y ′ acts by differentiation in tangential Fourier variables. Differentiating e − ϖ t e^{-\varpi t} e − ϖ t only creates further powers of t t t ; it neither introduces small frequencies nor leaves the fixed compact annulus. Normal multiplication and differentiation are immediate.
For the inverse, take the inverse Fourier transform of (293) in ξ 3 \xi_3 ξ 3 . The scalar kernels ( ϖ 2 + ξ 3 2 ) − 1 (\varpi^2+\xi_3^2)^{-1} ( ϖ 2 + ξ 3 2 ) − 1 and ( ϖ 2 + ξ 3 2 ) − 2 (\varpi^2+\xi_3^2)^{-2} ( ϖ 2 + ξ 3 2 ) − 2 are constant multiples of e − ϖ ∣ t ∣ / ϖ e^{-\varpi|t|}/\varpi e − ϖ ∣ t ∣ / ϖ and ( 1 + ϖ ∣ t ∣ ) e − ϖ ∣ t ∣ / ϖ 3 (1+\varpi|t|)e^{-\varpi|t|}/\varpi^3 ( 1 + ϖ ∣ t ∣ ) e − ϖ ∣ t ∣ / ϖ 3 . The factors ξ 3 \xi_3 ξ 3 and ξ 3 2 \xi_3^2 ξ 3 2 produce their first and second derivatives. Thus the matrix Green kernel on the full line is, on each side of zero, a polynomial of degree at most one times e − ϖ ∣ t ∣ e^{-\varpi|t|} e − ϖ ∣ t ∣ , with smooth coefficients on the annulus and the frozen material range. Extend F ′ F' F ′ by zero to t < 0 t<0 t < 0 and convolve this kernel with F F F . For t > 0 t>0 t > 0 , split the convolution at t t t . The integrals over [ 0 , t ] [0,t] [ 0 , t ] and [ t , ∞ ) [t,\infty) [ t , ∞ ) are elementary integrals of polynomials against exponentials; their sum is e − ϖ t e^{-\varpi t} e − ϖ t times a polynomial in t t t . Its coefficients remain smooth and compactly supported in ξ \xi ξ . The full-line convolution solution may have nonzero boundary trace.
For an arbitrary boundary vector a = a ( ξ ) a=a(\xi) a = a ( ξ ) the decaying homogeneous solution is
P ^ 0 a ( ξ , t ) = e − ϖ t ( a + ϖ t b ( a ) ) , \widehat{P}_{0a}(\xi,t)=e^{-\varpi t}(a+\varpi t b(a)), P 0 a ( ξ , t ) = e − ϖ t ( a + ϖ t b ( a )) ,
b ( a ) = g 0 [ − ( a ∥ + i a 3 ) ξ ^ + ( − i a ∥ + a 3 ) e 3 ] , b(a)=g_0\left[-(a_{\parallel}+ia_3)\widehat{\xi}+(-ia_{\parallel}+a_3)e_3\right], b ( a ) = g 0 [ − ( a ∥ + i a 3 ) ξ + ( − i a ∥ + a 3 ) e 3 ] ,
a ∥ = ξ ^ ⋅ a , g 0 = λ 0 + μ 0 λ 0 + 3 μ 0 , d 0 = 1 − g 0 = 2 μ 0 λ 0 + 3 μ 0 . (297) a_{\parallel}=\widehat{\xi}\cdot a,\qquad g_0=\frac{\lambda_0+\mu_0}{\lambda_0+3\mu_0},\qquad d_0=1-g_0=\frac{2\mu_0}{\lambda_0+3\mu_0}. \tag*{(297)} a ∥ = ξ ⋅ a , g 0 = λ 0 + 3 μ 0 λ 0 + μ 0 , d 0 = 1 − g 0 = λ 0 + 3 μ 0 2 μ 0 . (297)
Set
c a = i a ∥ − a 3 , b = g 0 c a ( i ξ ^ − e 3 ) , i b ∥ − b 3 = 0 , c_a=ia_{\parallel}-a_3,\qquad b=g_0c_a(i\widehat{\xi}-e_3),\qquad ib_{\parallel}-b_3=0, c a = i a ∥ − a 3 , b = g 0 c a ( i ξ − e 3 ) , i b ∥ − b 3 = 0 ,
div ^ P 0 a = ϖ d 0 c a e − ϖ t , Δ ^ P 0 a = − 2 ϖ 2 b e − ϖ t . (298) \widehat{\operatorname{div}}P_{0a}=\varpi d_0c_a e^{-\varpi t},\qquad\widehat{\Delta}P_{0a}=-2\varpi^2 b e^{-\varpi t}. \tag*{(298)} div P 0 a = ϖ d 0 c a e − ϖ t , Δ P 0 a = − 2 ϖ 2 b e − ϖ t . (298)
Substitution in (292) gives
L 0 + P 0 a ^ = ϖ 2 c a e − ϖ t [ 2 μ 0 g 0 − ( λ 0 + μ 0 ) d 0 ] ( i ξ ^ − e 3 ) = 0. (299) \widehat{L_0^+P_{0a}}=\varpi^2c_a e^{-\varpi t}\left[2\mu_0g_0-(\lambda_0+\mu_0)d_0\right](i\widehat{\xi}-e_3)=0. \tag*{(299)} L 0 + P 0 a = ϖ 2 c a e − ϖ t [ 2 μ 0 g 0 − ( λ 0 + μ 0 ) d 0 ] ( i ξ − e 3 ) = 0. (299)
We subtract the Poisson solution (297) with boundary data equal to the trace of the convolution solution. The resulting field has zero trace and remains in the class (295) . Uniqueness in the zero-trace finite-energy space follows from strong convexity and Korn’s inequality; one may extend a zero-trace field by zero and apply the full-space Fourier Korn inequality.
Plancherel turns tangential weights into frequency derivatives. All such derivatives are supported in the fixed annulus and produce only finitely many additional powers of t t t . Normal weighted integrals of polynomials times e − ϖ t e^{-\varpi t} e − ϖ t are uniformly finite because ϖ ≥ 1 \varpi\geq1 ϖ ≥ 1 . This proves (296) . The convolution formulas are uniformly nonvanishing material denominators and positive powers of ϖ \varpi ϖ in denominators. Differentiation in any finite collection of material or source parameters is consequently legitimate under the integrals and has the same bounds. Differentiation under the integral therefore establishes the asserted Lipschitz dependence of the half-space profile operator. ◻
Lemma 8.1 uses arbitrary fixed spatial weights of a profile built from finitely many derivative values. It does not require arbitrarily many spatial derivatives of the original material. The distinction between profile polynomial degree and coefficient regularity controls the remote cutoff tails without strengthening the C s C^s C s prior.
Uniform energy expansion for the exact Dirichlet solution
The transplanted half-space profiles are compared with the exact Dirichlet solution to obtain a boundary-energy expansion with a uniform order-s s s remainder.
Define the homogeneous Taylor tensors and their forms by
A p , r ( y ) = ∑ ∣ α ∣ = r ∂ α A p ( 0 ) α ! y α , a r ( V , W ) = ∫ R + 3 A p , r ( y ) ∇ V : ∇ W ‾ d y . (300) A_{p,r}(y)=\sum_{|\alpha|=r}\frac{\partial^\alpha A_p(0)}{\alpha!}y^\alpha,\qquad a_r(V,W)=\int_{\mathbb{R}^3_+}A_{p,r}(y)\nabla V:\overline{\nabla W}\,dy. \tag*{(300)} A p , r ( y ) = ∣ α ∣ = r ∑ α ! ∂ α A p ( 0 ) y α , a r ( V , W ) = ∫ R + 3 A p , r ( y ) ∇ V : ∇ W d y . (300)
For r = 0 r=0 r = 0 , this is the frozen form (292) . On the fixed chart, Taylor’s integral remainder gives the uniform bound
∣ A p ( ϵ y ) − ∑ r = 0 s − 1 ϵ r A p , r ( y ) ∣ ≤ C ϵ s ∣ y ∣ s . (301) \left|A_p(\epsilon y)-\sum_{r=0}^{s-1}\epsilon^r A_{p,r}(y)\right|\leq C\epsilon^s|y|^s. \tag*{(301)} A p ( ϵy ) − r = 0 ∑ s − 1 ϵ r A p , r ( y ) ≤ C ϵ s ∣ y ∣ s . (301)
Only the bounded order-s s s derivatives are used; their modulus of continuity need not be uniform across the material class.
Let V ^ 0 ( ξ , t ) \widehat V_0(\xi,t) V 0 ( ξ , t ) be the right-hand side of (297) with a = ϕ ^ ( ξ ) a=\widehat\phi(\xi) a = ϕ ( ξ ) . For 1 ≤ r ≤ s − 1 1\leq r\leq s-1 1 ≤ r ≤ s − 1 define the zero-trace corrections by
a 0 ( V r , W ) = − ∑ ℓ = 1 r a ℓ ( V r − ℓ , W ) , Tr V r = 0 , W ∈ H ˙ 0 1 ( R + 3 ) . (302) a_0(V_r,W)=-\sum_{\ell=1}^{r}a_\ell(V_{r-\ell},W),\qquad\operatorname{Tr}V_r=0,\qquad W\in\dot H^1_0(\mathbb{R}^3_+). \tag*{(302)} a 0 ( V r , W ) = − ℓ = 1 ∑ r a ℓ ( V r − ℓ , W ) , Tr V r = 0 , W ∈ H ˙ 0 1 ( R + 3 ) . (302)
Equivalently, the positive operator satisfies L + 0 V r = div ( ∑ ℓ = 1 r A p , ℓ ∇ V r − ℓ ) L_+^0V_r=\operatorname{div}\left(\sum_{\ell=1}^{r}A_{p,\ell}\nabla V_{r-\ell}\right) L + 0 V r = div ( ∑ ℓ = 1 r A p , ℓ ∇ V r − ℓ ) in the half-space; this fixes the source sign. It is a divergence source with polynomial coefficients times previous profiles. Lemma 8.1 constructs V r V_r V r successively. Every V r V_r V r has the weighted bounds (296) , uniformly in p p p and the material, and depends smoothly on the tensor derivatives through order r r r .
Theorem 8.2. Let (4) hold, and let ϕ \phi ϕ be either of the fixed tests in (287) . If u p , ϵ ϕ u^\phi_{p,\epsilon} u p , ϵ ϕ is the actual elastic Dirichlet solution with trace (288) , put
E p , ϕ ( ϵ ) = ⟨ Λ λ , μ f p , ϵ ϕ , f p , ϵ ϕ ⟩ . (303) \mathcal{E}_{p,\phi}(\epsilon)=\langle\Lambda_{\lambda,\mu}f^\phi_{p,\epsilon},f^\phi_{p,\epsilon}\rangle. \tag*{(303)} E p , ϕ ( ϵ ) = ⟨ Λ λ , μ f p , ϵ ϕ , f p , ϵ ϕ ⟩ . (303) There are uniform C , ϵ ∗ > 0 C,\epsilon_*>0 C , ϵ ∗ > 0 such that
∣ E p , ϕ ( ϵ ) − ∑ r = 0 s − 1 ϵ r E p , r ϕ ∣ ≤ C ϵ s , E p , r ϕ = ∑ ℓ + b + c = r a ℓ ( V b , V c ) , 0 < ϵ < ϵ ∗ . (304) \left|\mathcal{E}_{p,\phi}(\epsilon)-\sum_{r=0}^{s-1}\epsilon^r E^\phi_{p,r}\right|\leq C\epsilon^s,\qquad E^\phi_{p,r}=\sum_{\ell+b+c=r}a_\ell(V_b,V_c),\qquad0<\epsilon<\epsilon_*. \tag*{(304)} E p , ϕ ( ϵ ) − r = 0 ∑ s − 1 ϵ r E p , r ϕ ≤ C ϵ s , E p , r ϕ = ℓ + b + c = r ∑ a ℓ ( V b , V c ) , 0 < ϵ < ϵ ∗ . (304) Here a ℓ a_\ell a ℓ is defined in (300) , and V b , V c V_b,V_c V b , V c are the profiles defined by (297) and (302) . The coefficients are smooth functions of the finite material and geometric derivatives of total order at most r r r . Their first derivatives with respect to those derivative values are uniformly bounded on the indicated bounded, strongly convex parameter range.
Proof. Set
V ϵ [ s − 1 ] = ∑ r = 0 s − 1 ϵ r V r , (305) V_\epsilon^{[s-1]}=\sum_{r=0}^{s-1}\epsilon^rV_r, \tag*{(305)} V ϵ [ s − 1 ] = r = 0 ∑ s − 1 ϵ r V r , (305)
where the profiles are defined by (297) and (302) . Define the physical approximation by
U p , ϵ ( Ψ p ( z ) ) = ϵ − 1 / 2 R p χ p ( z ) V ϵ [ s − 1 ] ( z / ϵ ) , U p , ϵ = 0 outside the chart . (306) U_{p,\epsilon}(\Psi_p(z))=\epsilon^{-1/2}R_p\chi_p(z)V_\epsilon^{[s-1]}(z/\epsilon),\qquad U_{p,\epsilon}=0\text{ outside the chart}. \tag*{(306)} U p , ϵ ( Ψ p ( z )) = ϵ − 1/2 R p χ p ( z ) V ϵ [ s − 1 ] ( z / ϵ ) , U p , ϵ = 0 outside the chart . (306)
All positive-order correctors have zero trace, so this is an exact lifting of the prescribed Dirichlet data (288) . Lemma C.1 gives uniform weighted bounds for all the profiles. The four-term calculation in Proposition C.2, which uses only the recursion and these weighted bounds, proves
∣ a ( U p , ϵ , w ) ∣ ≤ C ϵ s ∥ w ∥ H 1 ( Ω ) , w ∈ H 0 1 ( Ω ; C 3 ) . (307) |a(U_{p,\epsilon},w)|\leq C\epsilon^s\|w\|_{H^1(\Omega)},\qquad w\in H^1_0(\Omega;\mathbb{C}^3). \tag*{(307)} ∣ a ( U p , ϵ , w ) ∣ ≤ C ϵ s ∥ w ∥ H 1 ( Ω ) , w ∈ H 0 1 ( Ω ; C 3 ) . (307)
By (2.5), the zero-trace error
z = u p , ε ϕ − U p , ε ∈ H 0 1 ( Ω ; C 3 ) (308) z=u_{p,\varepsilon}^{\phi}-U_{p,\varepsilon}\in H_0^1(\Omega;\mathbb{C}^3) \tag*{(308)} z = u p , ε ϕ − U p , ε ∈ H 0 1 ( Ω ; C 3 ) (308)
satisfies
∥ z ∥ H 1 ( Ω ) ≤ C ε s . (309) \lVert z\rVert_{H^1(\Omega)}\le C\varepsilon^s. \tag*{(309)} ∥ z ∥ H 1 ( Ω ) ≤ C ε s . (309)
Since a ( u p , ε ϕ , z ) = 0 a(u_{p,\varepsilon}^{\phi},z)=0 a ( u p , ε ϕ , z ) = 0 , Hermitian symmetry gives the exact identity
E p , ϕ ( ε ) = a ( U p , ε , U p , ε ) − a ( z , z ) . (310) \mathcal{E}_{p,\phi}(\varepsilon)=a(U_{p,\varepsilon},U_{p,\varepsilon})-a(z,z). \tag*{(310)} E p , ϕ ( ε ) = a ( U p , ε , U p , ε ) − a ( z , z ) . (310)
The last term is O ( ε 2 s ) O(\varepsilon^{2s}) O ( ε 2 s ) . Applying (C.12) to the first term proves the expansion in (8.24).
For each fixed order, the profile recursion uses finitely many tensor jets, sums and products, and the constant Green operator from Lemma 8.1. Its frequency coefficients are rational smooth functions of the frozen moduli and polynomial functions of the remaining jets; all material denominators have fixed positive lower bounds. Weighted integrals and their parameter derivatives are dominated by the same weighted profile bounds. The coefficients and their first jet parameter derivatives are therefore uniformly bounded. The argument uses order-s s s material derivatives only in the Taylor remainder.
Remark. The expansion concerns energies of the exact physical Dirichlet solutions, so its coefficients can be estimated using the measured DN norm. Its uniform order-s s s remainder is also what permits the scale choice used to recover boundary derivatives of orders below s s s , without a common modulus of continuity for the top material derivatives.
Vandermonde extraction at fixed scale ratios
Lemma 8.3. For 0 ≤ q ≤ s − 1 0\le q\le s-1 0 ≤ q ≤ s − 1 choose the real weights w q ℓ w_{q\ell} w q ℓ , 1 ≤ ℓ ≤ s 1\le\ell\le s 1 ≤ ℓ ≤ s , uniquely determined by
∑ ℓ = 1 s w q ℓ ℓ r = { 1 , r = q , 0 , r ≠ q , 0 ≤ r ≤ s − 1. (311) \sum_{\ell=1}^{s}w_{q\ell}\ell^r=
\begin{cases}
1,&r=q,\\
0,&r\ne q,
\end{cases}
\qquad0\le r\le s-1. \tag*{(311)} ℓ = 1 ∑ s w q ℓ ℓ r = { 1 , 0 , r = q , r = q , 0 ≤ r ≤ s − 1. (311) For two materials in (1.4), use δ \delta δ from (1.6). For either fixed test and all 0 < ε < ε ∗ / s 0<\varepsilon<\varepsilon_*/s 0 < ε < ε ∗ / s ,
∣ E 1 , p , q ϕ − E 2 , p , q ϕ ∣ ≤ C s ( δ ε − q + ε s − q ) . (312) \left|E_{1,p,q}^{\phi}-E_{2,p,q}^{\phi}\right|\le C_s\left(\delta\varepsilon^{-q}+\varepsilon^{s-q}\right). \tag*{(312)} E 1 , p , q ϕ − E 2 , p , q ϕ ≤ C s ( δ ε − q + ε s − q ) . (312) Here E j , p , q ϕ E_{j,p,q}^{\phi} E j , p , q ϕ is the coefficient in (8.24) for material j j j , and ε ∗ \varepsilon_* ε ∗ is the uniform scale threshold of Theorem 8.2. In particular, for δ \delta δ sufficiently small and 0 ≤ q < s 0\le q<s 0 ≤ q < s ,
sup p ∣ E 1 , p , q ϕ − E 2 , p , q ϕ ∣ ≤ C s δ ( s − q ) / s . (313) \sup_p\left|E_{1,p,q}^{\phi}-E_{2,p,q}^{\phi}\right|\le C_s\delta^{(s-q)/s}. \tag*{(313)} p sup E 1 , p , q ϕ − E 2 , p , q ϕ ≤ C s δ ( s − q ) / s . (313) The weights depend only on s s s , and the constants are uniform over the fixed admissible coefficient class.
Proof. The nodes 1 , … , s 1,\ldots,s 1 , … , s are distinct, so the Vandermonde matrix in (8.31) is invertible. Apply (8.24) at ℓ ε \ell\varepsilon ℓ ε and form
ε − q ∑ ℓ = 1 s w q ℓ E p , ϕ ( ℓ ε ) . (314) \varepsilon^{-q}\sum_{\ell=1}^{s}w_{q\ell}\mathcal{E}_{p,\phi}(\ell\varepsilon). \tag*{(314)} ε − q ℓ = 1 ∑ s w q ℓ E p , ϕ ( ℓ ε ) . (314)
The moment identities (8.31) cancel every polynomial coefficient except the one of order q q q . Thus, for material j = 1 , 2 j=1,2 j = 1 , 2 ,
∣ ε − q ∑ ℓ = 1 s w q ℓ E j , p , ϕ ( ℓ ε ) − E j , p , q ϕ ∣ ≤ C s ε s − q . \left|\varepsilon^{-q}\sum_{\ell=1}^{s}w_{q\ell}\mathcal{E}_{j,p,\phi}(\ell\varepsilon)-E_{j,p,q}^{\phi}\right|\le C_s\varepsilon^{s-q}. ε − q ℓ = 1 ∑ s w q ℓ E j , p , ϕ ( ℓ ε ) − E j , p , q ϕ ≤ C s ε s − q .
The trace bound (8.9) and the measured DN norm (1.6) give
∣ E 1 , p , ϕ ( ℓ ε ) − E 2 , p , ϕ ( ℓ ε ) ∣ ≤ C δ . (315) \left|\mathcal{E}_{1,p,\phi}(\ell\varepsilon)-\mathcal{E}_{2,p,\phi}(\ell\varepsilon)\right|\le C\delta. \tag*{(315)} ∣ E 1 , p , ϕ ( ℓ ε ) − E 2 , p , ϕ ( ℓ ε ) ∣ ≤ C δ . (315)
Subtracting the two extraction identities and using (315) therefore yields
∣ E 1 , p , q ϕ − E 2 , p , q ϕ ∣ ≤ C δ ϵ − q ∑ ℓ = 1 s ∣ w q ℓ ∣ + C s ϵ s − q ≤ C s ( δ ϵ − q + ϵ s − q ) . \left|E^\phi_{1,p,q}-E^\phi_{2,p,q}\right| \le C\delta\epsilon^{-q}\sum_{\ell=1}^{s}|w_{q\ell}|+C_s\epsilon^{s-q}
\le C_s\left(\delta\epsilon^{-q}+\epsilon^{s-q}\right). E 1 , p , q ϕ − E 2 , p , q ϕ ≤ C δ ϵ − q ℓ = 1 ∑ s ∣ w q ℓ ∣ + C s ϵ s − q ≤ C s ( δ ϵ − q + ϵ s − q ) .
which proves (312) .
For δ > 0 \delta>0 δ > 0 , take ϵ = δ 1 / s \epsilon=\delta^{1/s} ϵ = δ 1/ s and choose a uniform noise threshold ensuring s ϵ < ϵ ∗ s\epsilon<\epsilon_* sϵ < ϵ ∗ . Both terms then have order δ ( s − q ) / s \delta^{(s-q)/s} δ ( s − q ) / s , proving (313) . For q = 0 q=0 q = 0 , letting ϵ ↓ 0 \epsilon\downarrow0 ϵ ↓ 0 at fixed δ \delta δ already gives the Lipschitz bound ∣ E 1 , p , 0 ϕ − E 2 , p , 0 ϕ ∣ ≤ C s δ \left|E^\phi_{1,p,0}-E^\phi_{2,p,0}\right|\le C_s\delta E 1 , p , 0 ϕ − E 2 , p , 0 ϕ ≤ C s δ . If δ = 0 \delta=0 δ = 0 , keep the nodes and weights fixed and let their common scale ϵ ↓ 0 \epsilon\downarrow0 ϵ ↓ 0 . Since s − q > 0 s-q>0 s − q > 0 , (312) gives equality of the two coefficients. Only the measured DN maps and the expansion have been used.
A triangular matrix for the new normal derivative
At each step of the normal-derivative induction, two extracted energy coefficients determine the new pair of Lamé derivatives through a uniformly invertible 2 × 2 2\times2 2 × 2 matrix.
Let
j q + ( p ) = ( λ q + ( p ) μ q + ( p ) ) , λ q + ( p ) = ∂ z 3 q ( λ ∘ Ψ p ) ( 0 ) , μ q + ( p ) = ∂ z 3 q ( μ ∘ Ψ p ) ( 0 ) . (316) j_q^+(p)=\begin{pmatrix}\lambda_q^+(p)\\ \mu_q^+(p)\end{pmatrix},\qquad
\lambda_q^+(p)=\partial_{z_3}^q(\lambda\circ\Psi_p)(0),\quad
\mu_q^+(p)=\partial_{z_3}^q(\mu\circ\Psi_p)(0). \tag*{(316)} j q + ( p ) = ( λ q + ( p ) μ q + ( p ) ) , λ q + ( p ) = ∂ z 3 q ( λ ∘ Ψ p ) ( 0 ) , μ q + ( p ) = ∂ z 3 q ( μ ∘ Ψ p ) ( 0 ) . (316)
The inward normal derivative differs from the outward derivative in (281) by ( − 1 ) q (-1)^q ( − 1 ) q without changing any stability norm. For q ≥ 1 q\ge1 q ≥ 1 , define the finite jet list
K p , q < = ( ∂ z α ( f ∘ Ψ p ) ( 0 ) : f ∈ { λ , μ } , ∣ α ∣ ≤ q , α 3 < q ) . (317) \mathcal{K}_{p,q}^{<}=\left(\partial_z^\alpha(f\circ\Psi_p)(0):f\in\{\lambda,\mu\},\ |\alpha|\le q,\ \alpha_3<q\right). \tag*{(317)} K p , q < = ( ∂ z α ( f ∘ Ψ p ) ( 0 ) : f ∈ { λ , μ } , ∣ α ∣ ≤ q , α 3 < q ) . (317)
The collection K p , q < \mathcal{K}_{p,q}^{<} K p , q < includes boundary values and all lower normal derivatives and their required tangential derivatives. The derivatives of the coordinate geometry are fixed and implicit in this notation.
The extracted coefficients still have to be related to the material derivatives. At each normal order, the two fixed polarizations produce a triangular system with a leading matrix acting on the two new Lamé normal derivatives.
Proposition 8.4. For every 1 ≤ q < s 1\le q<s 1 ≤ q < s the two coefficients in (304) have the form
\begin{pmatrix}E^S_{p,q}\\ E^N_{p,q}\end{pmatrix}
=M_q(\lambda(p),\mu(p))j_q^+(p)+\mathcal{F}_{p,q}(\mathcal{K}_{p,q}^{<}), \tag*{(318)}\end{pmatrix}
where j q + j_q^+ j q + is defined in (316) , K p , q < \mathcal{K}_{p,q}^{<} K p , q < in (317) , and
E p , q S = E p , q ϕ S , E p , q N = E p , q ϕ N . (319) E^S_{p,q}=E^{\phi^S}_{p,q},\qquad E^N_{p,q}=E^{\phi^N}_{p,q}. \tag*{(319)} E p , q S = E p , q ϕ S , E p , q N = E p , q ϕ N . (319) The coefficients on the right are from (304) , with the tests (287) . The function F p , q \mathcal{F}_{p,q} F p , q is uniformly Lipschitz on the finite-order derivative range. Put
I q = ∫ R 2 ∣ ψ ∂ ( ξ ) ∣ 2 ∣ ξ ∣ 1 − q d ξ > 0 , B q = 3 d 0 2 + 4 d 0 g 0 ( q + 1 ) + 2 g 0 2 ( q + 1 ) ( q + 2 ) . (320) \begin{aligned}
I_q&=\int_{\mathbb{R}^2}|\psi_{\partial}(\xi)|^2|\xi|^{1-q}\,d\xi>0,\\
B_q&=3d_0^2+4d_0g_0(q+1)+2g_0^2(q+1)(q+2). \tag*{(320)}
\end{aligned} I q B q = ∫ R 2 ∣ ψ ∂ ( ξ ) ∣ 2 ∣ ξ ∣ 1 − q d ξ > 0 , = 3 d 0 2 + 4 d 0 g 0 ( q + 1 ) + 2 g 0 2 ( q + 1 ) ( q + 2 ) . (320) Here g 0 , d 0 g_0,d_0 g 0 , d 0 are defined in (297) , and ψ ∂ \psi_{\partial} ψ ∂ is the fixed annular test profile in (287) . Then
M q = I q ( 0 2 − q 2 − q − 1 d 0 2 2 − q − 1 B q ) , det M q = − 2 − 2 q − 1 I q 2 d 0 2 ≠ 0. (321) M_q=I_q\begin{pmatrix}0&2^{-q}\\2^{-q-1}d_0^2&2^{-q-1}B_q\end{pmatrix},\qquad
\det M_q=-2^{-2q-1}I_q^2d_0^2\ne0. \tag*{(321)} M q = I q ( 0 2 − q − 1 d 0 2 2 − q 2 − q − 1 B q ) , det M q = − 2 − 2 q − 1 I q 2 d 0 2 = 0. (321) The matrices and their inverses are uniformly bounded for each of the finitely many orders q < s q<s q < s under (4) .
Proof. In E p , q ϕ = ∑ ℓ + b + c = q a ℓ ( V b , V c ) E_{p,q}^{\phi}=\sum_{\ell+b+c=q}a_{\ell}(V_b,V_c) E p , q ϕ = ∑ ℓ + b + c = q a ℓ ( V b , V c ) , the only terms containing V q V_q V q are a 0 ( V q , V 0 ) a_0(V_q,V_0) a 0 ( V q , V 0 ) and a 0 ( V 0 , V q ) a_0(V_0,V_q) a 0 ( V 0 , V q ) . Both vanish: V q V_q V q has zero trace and V 0 V_0 V 0 solves the homogeneous constant problem. Hence
E p , q ϕ = a q ( V 0 , V 0 ) + a smooth function of tensor derivatives of order < q . (322) E_{p,q}^{\phi}=a_q(V_0,V_0)+\text{a smooth function of tensor derivatives of order}<q. \tag*{(322)} E p , q ϕ = a q ( V 0 , V 0 ) + a smooth function of tensor derivatives of order < q . (322)
The pure-normal part of the order-q q q Taylor tensor in a q ( V 0 , V 0 ) a_q(V_0,V_0) a q ( V 0 , V 0 ) contributes exactly
λ q + q ! ∫ R + 3 t q ∣ div V 0 ∣ 2 d y + 2 μ q + q ! ∫ R + 3 t q ∣ ε ( V 0 ) ∣ 2 d y . (323) \frac{\lambda_q^+}{q!}\int_{\mathbb{R}_+^3}t^q|\operatorname{div}V_0|^2\,dy+\frac{2\mu_q^+}{q!}\int_{\mathbb{R}_+^3}t^q|\varepsilon(V_0)|^2\,dy. \tag*{(323)} q ! λ q + ∫ R + 3 t q ∣ div V 0 ∣ 2 d y + q ! 2 μ q + ∫ R + 3 t q ∣ ε ( V 0 ) ∣ 2 d y . (323)
Indeed the coordinate Jacobian is orthogonal at the base point. In the q q q th tensor derivative, derivatives falling on coordinate factors multiply lower-order material-parameter derivatives. All other order-q q q material monomials have smaller normal order and belong to K p , q < K_{p,q}^{<} K p , q < . For clarity, the tensor in Appendix C is a sum of fixed smooth coordinate factors times λ ∘ Ψ p \lambda\circ\Psi_p λ ∘ Ψ p or μ ∘ Ψ p \mu\circ\Psi_p μ ∘ Ψ p . After applying ∂ z α \partial_z^\alpha ∂ z α , ∣ α ∣ = q |\alpha|=q ∣ α ∣ = q , to the transformed tensor, we separate the terms according to whether a derivative falls on a coordinate factor. The terms with all q q q derivatives on a material coefficient are linear in its order-q q q coordinate derivatives. Every term with a differentiated coordinate factor uses a material derivative of order at most q − 1 q-1 q − 1 . Among the terms with all q q q derivatives on a material component, α = ( 0 , 0 , q ) \alpha=(0,0,q) α = ( 0 , 0 , q ) gives exactly (323) ; every α \alpha α with α 3 < q \alpha_3<q α 3 < q is in K p , q < K_{p,q}^{<} K p , q < by (283) . Also V r V_r V r for r < q r<q r < q depends only on tensor derivatives through order r r r . Thus the remainder in (322) contains no order-q q q pure normal material derivative, even nonlinearly. This proves the structural decomposition including all mixed and geometric terms. Uniform Lipschitz dependence follows from Theorem 8.2 and Lemma 8.1 on the bounded finite-order derivative set.
For ϕ S \phi^S ϕ S , (297) gives V ^ 0 S = ψ ∂ e − ϖ t ξ ⊥ \widehat V_0^S=\psi\partial e^{-\varpi t}\xi^\perp V 0 S = ψ ∂ e − ϖ t ξ ⊥ . It is divergence free, and at each frequency 2 ∣ ε ( V ^ 0 S ) ∣ 2 = 2 ϖ 2 ∣ ψ ∂ ∣ 2 e − 2 ϖ t 2|\varepsilon(\widehat V_0^S)|^2=2\varpi^2|\psi_\partial|^2e^{-2\varpi t} 2∣ ε ( V 0 S ) ∣ 2 = 2 ϖ 2 ∣ ψ ∂ ∣ 2 e − 2 ϖ t , and the strain includes the tangential Fourier derivatives. Since ∫ 0 ∞ t q e − 2 ϖ t d t / q ! = ( 2 ϖ ) − q − 1 \int_0^\infty t^q e^{-2\varpi t}\,dt/q!=(2\varpi)^{-q-1} ∫ 0 ∞ t q e − 2 ϖ t d t / q ! = ( 2 ϖ ) − q − 1 , its two weights are 0 0 0 and 2 − q I q 2^{-q}I_q 2 − q I q .
For ϕ N \phi^N ϕ N one has
V ^ 0 N ( ξ , t ) = ψ ∂ e − ϖ t [ e 3 + g 0 ϖ t ( − i ξ ^ + e 3 ) ] , div V 0 N ^ = − d 0 ϖ ψ ∂ e − ϖ t . (324) \widehat V_0^N(\xi,t)=\psi_\partial e^{-\varpi t}\left[e_3+g_0\varpi t(-i\widehat\xi+e_3)\right],\qquad\widehat{\operatorname{div}V_0^N}=-d_0\varpi\psi_\partial e^{-\varpi t}. \tag*{(324)} V 0 N ( ξ , t ) = ψ ∂ e − ϖ t [ e 3 + g 0 ϖ t ( − i ξ + e 3 ) ] , div V 0 N = − d 0 ϖ ψ ∂ e − ϖ t . (324)
At ϖ = 1 \varpi=1 ϖ = 1 , ξ ^ = e 1 \widehat\xi=e_1 ξ = e 1 , the three nonzero strain types are
ε 11 = g 0 t e − t , ε 33 = − ( d 0 + g 0 t ) e − t , ε 13 = ε 31 = i 2 ( d 0 + 2 g 0 t ) e − t . (325) \varepsilon_{11}=g_0te^{-t},\qquad\varepsilon_{33}=-(d_0+g_0t)e^{-t},\qquad\varepsilon_{13}=\varepsilon_{31}=\frac{i}{2}(d_0+2g_0t)e^{-t}. \tag*{(325)} ε 11 = g 0 t e − t , ε 33 = − ( d 0 + g 0 t ) e − t , ε 13 = ε 31 = 2 i ( d 0 + 2 g 0 t ) e − t . (325)
Restoring the frequency ϖ \varpi ϖ and the profile ψ ∂ \psi_\partial ψ ∂ in (325) , we obtain
2 ∣ ε ( V ^ 0 N ) ∣ 2 = ϖ 2 ∣ ψ ∂ ∣ 2 e − 2 ϖ t [ 3 d 0 2 + 8 d 0 g 0 ϖ t + 8 g 0 2 ( ϖ t ) 2 ] . (326) 2|\varepsilon(\widehat V_0^N)|^2=\varpi^2|\psi_\partial|^2e^{-2\varpi t}\left[3d_0^2+8d_0g_0\varpi t+8g_0^2(\varpi t)^2\right]. \tag*{(326)} 2∣ ε ( V 0 N ) ∣ 2 = ϖ 2 ∣ ψ ∂ ∣ 2 e − 2 ϖ t [ 3 d 0 2 + 8 d 0 g 0 ϖ t + 8 g 0 2 ( ϖ t ) 2 ] . (326)
Integrate as in the shear case. The resulting weights are 2 − q − 1 d 0 2 I q 2^{-q-1}d_0^2I_q 2 − q − 1 d 0 2 I q and 2 − q − 1 B q I q 2^{-q-1}B_qI_q 2 − q − 1 B q I q , proving (321) . Its determinant is the displayed nonzero quantity; d 0 d_0 d 0 has a uniform positive lower bound and I q I_q I q is a fixed positive number. ◻
Remark. At each order the shear test isolates the new normal derivative of μ \mu μ , and the normal test then determines that of λ \lambda λ . The remainder contains only lower normal orders and their tangential derivatives. This triangular structure makes an induction possible without assuming that any boundary derivatives agree beforehand.
The full induction with tangential derivatives
The triangular recovery system and tangential interpolation yield estimates for all Cartesian boundary derivatives below order s s s , including the exact-data case.
For two candidates put
n a = ∥ ∂ ν a ( λ 1 − λ 2 ) ∥ C 0 ( ∂ Ω ) + ∥ ∂ ν a ( μ 1 − μ 2 ) ∥ C 0 ( ∂ Ω ) , 0 ≤ a ≤ s − 1. (327) n_a=\|\partial_\nu^a(\lambda_1-\lambda_2)\|_{C^0(\partial\Omega)}+\|\partial_\nu^a(\mu_1-\mu_2)\|_{C^0(\partial\Omega)},\qquad0\leq a\leq s-1. \tag*{(327)} n a = ∥ ∂ ν a ( λ 1 − λ 2 ) ∥ C 0 ( ∂ Ω ) + ∥ ∂ ν a ( μ 1 − μ 2 ) ∥ C 0 ( ∂ Ω ) , 0 ≤ a ≤ s − 1. (327)
Normal derivatives mean derivatives along the fixed straight normal segments, as in (8.1). This definition avoids unspecified extensions of the normal vector field. Cartesian boundary derivative discrepancies J q J_q J q retain their definition in (1.9), truncated at order q q q .
Proposition 8.4 and the tangential interpolation estimate (8.50) yield the following boundary result, independently of the global CGO construction.
Theorem 8.5. For the fixed class S s ( Ω ; M , c 0 ) \mathcal{S}^{s}(\Omega;M,c_0) S s ( Ω ; M , c 0 ) in (1.4), there are C , δ 0 > 0 C,\delta_0>0 C , δ 0 > 0 such that every two members with 0 < δ < δ 0 0<\delta<\delta_0 0 < δ < δ 0 satisfy, simultaneously for every integer 0 ≤ q < s 0\leq q<s 0 ≤ q < s ,
J q ≤ C δ ( s − q ) / s . (328) J_q \leq C\delta^{(s-q)/s}. \tag*{(328)} J q ≤ C δ ( s − q ) / s . (328) Here J q J_q J q and δ \delta δ are defined in (1.9) and (1.6). The constant depends only on the fixed class and geometry. If δ = 0 \delta=0 δ = 0 , all these boundary derivatives through order s − 1 s-1 s − 1 are equal.
Proof. Step 1: the order-zero anchor. For the two annular tests the zeroth coefficients are
E p , 0 S = μ ( p ) I 0 , E p , 0 N = a Z ( p ) I 0 . (329) E^{S}_{p,0}=\mu(p)I_0,\qquad E^{N}_{p,0}=a_Z(p)I_0. \tag*{(329)} E p , 0 S = μ ( p ) I 0 , E p , 0 N = a Z ( p ) I 0 . (329)
Here I 0 I_0 I 0 and a Z a_Z a Z are defined in (8.40) and (8.10). Use Lemma 8.3 at q = 0 q=0 q = 0 and the uniformly Lipschitz inverse (8.11). It follows that n 0 ≤ C δ n_0\leq C\delta n 0 ≤ C δ . This is a direct Lipschitz boundary-value estimate in the DN operator norm of (1.6).
Step 2: estimate mixed derivatives of lower normal order. Suppose n a ≤ C δ ( s − a ) / s n_a\leq C\delta^{(s-a)/s} n a ≤ C δ ( s − a ) / s for 0 ≤ a < q 0\leq a<q 0 ≤ a < q . The a a a th normal boundary derivative is uniformly C s − a C^{s-a} C s − a in surface coordinates. Fixed surface-chart interpolation, using cutoffs in slightly larger charts, gives for 0 ≤ b ≤ q − a 0\leq b\leq q-a 0 ≤ b ≤ q − a
∥ ∂ y β ∂ ν a ( f 1 − f 2 ) ∥ C 0 ≤ C n a 1 − b / ( s − a ) ≤ C δ ( s − q ) / s , f = λ , μ . (330) \left\|\partial_y^\beta\partial_\nu^a(f_1-f_2)\right\|_{C^0}\leq Cn_a^{1-b/(s-a)}\leq C\delta^{(s-q)/s},\qquad f=\lambda,\mu. \tag*{(330)} ∂ y β ∂ ν a ( f 1 − f 2 ) C 0 ≤ C n a 1 − b / ( s − a ) ≤ C δ ( s − q ) / s , f = λ , μ . (330)
The usual additional C n a Cn_a C n a term is absorbed for n a ≤ 1 n_a\leq1 n a ≤ 1 . The bound uses only s − a s-a s − a tangential derivatives, already furnished by the C s C^s C s material prior. In particular the exponents are not replaced at every stage by one fixed worst-case exponent: the identity
s − a s ( 1 − q − a s − a ) = s − q s (331) \frac{s-a}{s}\left(1-\frac{q-a}{s-a}\right)=\frac{s-q}{s} \tag*{(331)} s s − a ( 1 − s − a q − a ) = s s − q (331)
is retained. By (8.3), a coordinate derivative of lower normal order ∂ z γ ∂ z a ( f ∘ Ψ p ) \partial_z^\gamma\partial_z^a(f\circ\Psi_p) ∂ z γ ∂ z a ( f ∘ Ψ p ) has a < q a<q a < q and ∣ γ ∣ + a ≤ q |\gamma|+a\leq q ∣ γ ∣ + a ≤ q . It is a tangential derivative of the already recovered a a a th normal boundary function. Hence every difference in K p , q < K^{<}_{p,q} K p , q < is O ( δ ( s − q ) / s ) O(\delta^{(s-q)/s}) O ( δ ( s − q ) / s ) , uniformly in p p p . This is the order of the induction: lower normal functions, their tangential derivatives through total order q q q , then the new normal function of order q q q .
Here is an elementary justification of the interpolation inequality used in Step 2. On a larger fixed surface chart, multiply the relevant trace by a cutoff equal to one on the smaller chart. Let F F F be the resulting compactly supported function, and set
N = s − a , A = ∥ F ∥ ∞ . (332) N=s-a,\qquad A=\|F\|_\infty. \tag*{(332)} N = s − a , A = ∥ F ∥ ∞ . (332)
Its C N C^N C N norm is uniformly bounded. Choose a smooth compactly supported, possibly signed convolution kernel K K K with integral one and moments zero through order N − 1 N-1 N − 1 . Such a kernel is obtained from a fixed bump by a finite linear combination of its distinct dilates: a Vandermonde system sets each moment order to zero. Define
K η = η − 2 K ( ⋅ / η ) . (333) K_\eta=\eta^{-2}K(\cdot/\eta). \tag*{(333)} K η = η − 2 K ( ⋅ / η ) . (333)
Taylor’s formula applied to ∂ β F \partial^\beta F ∂ β F , ∣ β ∣ = b < N |\beta|=b<N ∣ β ∣ = b < N , gives
∥ ∂ β ( F − K η ∗ F ) ∥ ∞ ≤ C η N − b ∥ F ∥ C N , ∥ ∂ β K η ∗ F ∥ ∞ ≤ C η − b A . (334) \left\|\partial^\beta(F-K_\eta*F)\right\|_\infty\leq C\eta^{N-b}\|F\|_{C^N},\qquad\left\|\partial^\beta K_\eta*F\right\|_\infty\leq C\eta^{-b}A. \tag*{(334)} ∂ β ( F − K η ∗ F ) ∞ ≤ C η N − b ∥ F ∥ C N , ∂ β K η ∗ F ∞ ≤ C η − b A . (334)
For small A > 0 A>0 A > 0 take η = A 1 / N \eta=A^{1/N} η = A 1/ N ; for larger A A A enlarge the fixed constant. If A = 0 A=0 A = 0 , F = 0 F=0 F = 0 . This proves ∥ ∂ β F ∥ ∞ ≤ C ( A 1 − b / N + A ) \lVert\partial^\beta F\rVert_\infty\le C(A^{1-b/N}+A) ∥ ∂ β F ∥ ∞ ≤ C ( A 1 − b / N + A ) . Only N N N derivatives, not N + 1 N+1 N + 1 , were used. In the induction b ≤ q − a < N b\le q-a<N b ≤ q − a < N .
Step 3: recover the new normal derivative. Subtract (318) for the two materials, using M q M_q M q of the first material on the left. The difference of the two matrices is O ( n 0 ) O(n_0) O ( n 0 ) because it depends smoothly only on the boundary values. The second material’s q q q th normal derivative is uniformly bounded. Consequently Lemma 8.3, Step 2 and the uniform inverse of (321) give
n q ≤ C [ δ ( s − q ) / s + n 0 + ∑ a = 0 q − 1 n a ( s − q ) / ( s − a ) ] ≤ C δ ( s − q ) / s . (335) \begin{aligned}
n_q\le C\left[\delta^{(s-q)/s}+n_0+\sum_{a=0}^{q-1}n_a^{(s-q)/(s-a)}\right] \\
&\le C\delta^{(s-q)/s}. \tag*{(335)}
\end{aligned} n q ≤ C [ δ ( s − q ) / s + n 0 + a = 0 ∑ q − 1 n a ( s − q ) / ( s − a ) ] ≤ C δ ( s − q ) / s . (335)
Construct the finitely many induction constants before choosing the noise level, and let C ∗ C_* C ∗ bound them. Then one may shrink a single threshold to ensure both s δ 1 / s < ϵ ∗ s\delta^{1/s}<\epsilon_* s δ 1/ s < ϵ ∗ and C ∗ δ 1 / s ≤ 1 C_*\delta^{1/s}\le1 C ∗ δ 1/ s ≤ 1 . This makes every interpolation absorption valid for all q < s q<s q < s in the same class. No new class or threshold is chosen after q q q or a material pair is fixed.
Step 4: return to full Cartesian boundary derivatives and handle zero noise. For each a ≤ q a\le q a ≤ q apply the same interpolation up to tangential order q − a q-a q − a . For the coordinate conversion, define
B ( z ) = D Ψ p ( z ) − 1 , (336) B(z)=D\Psi_p(z)^{-1}, \tag*{(336)} B ( z ) = D Ψ p ( z ) − 1 , (336)
where Ψ p \Psi_p Ψ p is given by (282) . For a Cartesian multi-index α \alpha α with 1 ≤ ∣ α ∣ ≤ q 1\le|\alpha|\le q 1 ≤ ∣ α ∣ ≤ q , repeatedly applying ∂ x i = ∑ b B b i ( z ) ∂ z b \partial_{x_i}=\sum_bB_{bi}(z)\partial_{z_b} ∂ x i = ∑ b B bi ( z ) ∂ z b gives
( ∂ x α f ) ∘ Ψ p = ∑ 1 ≤ ∣ β ∣ ≤ ∣ α ∣ c α β , p ( z ) ∂ z β ( f ∘ Ψ p ) , ∥ c α β , p ∥ ∞ ≤ C q , Ω . (337) (\partial_x^\alpha f)\circ\Psi_p
=\sum_{1\le|\beta|\le|\alpha|}c_{\alpha\beta,p}(z)\partial_z^\beta(f\circ\Psi_p),
\qquad\lVert c_{\alpha\beta,p}\rVert_\infty\le C_{q,\Omega}. \tag*{(337)} ( ∂ x α f ) ∘ Ψ p = 1 ≤ ∣ β ∣ ≤ ∣ α ∣ ∑ c α β , p ( z ) ∂ z β ( f ∘ Ψ p ) , ∥ c α β , p ∥ ∞ ≤ C q , Ω . (337)
The coefficients are fixed functions of the coordinate geometry. At z 3 = 0 z_3=0 z 3 = 0 each term on the right is, by (283) , a tangential derivative of a normal boundary function with total order at most q q q . Every one has just been bounded; the order-zero term is the anchor. Summing the estimates over the finite atlas and over all ∣ α ∣ ≤ q |\alpha|\le q ∣ α ∣ ≤ q , we obtain (328) . If δ = 0 \delta=0 δ = 0 , coefficient extraction follows by letting ϵ ↓ 0 \epsilon\downarrow0 ϵ ↓ 0 . We repeat the induction from order zero and obtain equality of all the normal derivatives under consideration. Their surface functions agree, hence so do their existing tangential derivatives; thus J q = 0 J_q=0 J q = 0 for q < s q<s q < s .
Remark. The order-dependent interpolation estimates preserve the exponent ( s − q ) / s (s-q)/s ( s − q ) / s through the induction. The estimate for J s − 1 J_{s-1} J s − 1 supplies the boundary control used in Theorem 9.2 . Sharpness requires the additional strict-slack hypothesis of Proposition 8.6.
Boundary-localized perturbations with bounded C s C^s C s norm give the following obstruction to improving the exponents.
Proposition 8.6. Fix the class C s ( Ω ; M , c 0 ) \mathcal{C}^s(\Omega;M,c_0) C s ( Ω ; M , c 0 ) in (4) . Suppose there are real constants ( λ b , μ b ) (\lambda_b,\mu_b) ( λ b , μ b ) and a number σ b > 0 \sigma_b>0 σ b > 0 such that
μ b ≥ c 0 + σ b , 3 λ b + 2 μ b ≥ c 0 + σ b , \mu_b\ge c_0+\sigma_b,\qquad3\lambda_b+2\mu_b\ge c_0+\sigma_b, μ b ≥ c 0 + σ b , 3 λ b + 2 μ b ≥ c 0 + σ b , N b : = ∥ λ b ∥ C s ( Ω ‾ ) + ∥ μ b ∥ C s ( Ω ‾ ) ≤ M − σ b . (338) N_b:=\lVert\lambda_b\rVert_{C^s(\overline{\Omega})}+\lVert\mu_b\rVert_{C^s(\overline{\Omega})}\le M-\sigma_b. \tag*{(338)} N b := ∥ λ b ∥ C s ( Ω ) + ∥ μ b ∥ C s ( Ω ) ≤ M − σ b . (338) Choose one boundary patch, one point p p p in it and the profiles (340) . There are fixed a 0 , a a m p > 0 a_0,a_{\mathrm{amp}}>0 a 0 , a amp > 0 such that for each integer q ∈ { 0 , … , s } q\in\{0,\ldots,s\} q ∈ { 0 , … , s } the entire family ( λ a , q , μ b ) (\lambda_{a,q},\mu_b) ( λ a , q , μ b ) , 0 < a < a 0 0<a<a_0 0 < a < a 0 , belongs to this fixed class. Compared with ( λ b , μ b ) (\lambda_b,\mu_b) ( λ b , μ b ) , its DN discrepancy and Cartesian jet discrepancy satisfy
0 < δ a , q ≤ C q a a m p a s , J q ≥ c q a a m p a s − q . (339) 0<\delta_{a,q}\le C_q a_{\mathrm{amp}}a^s,\qquad J_q\ge c_q a_{\mathrm{amp}}a^{s-q}. \tag*{(339)} 0 < δ a , q ≤ C q a amp a s , J q ≥ c q a amp a s − q . (339) The constants and the amplitude are independent of a a a . Thus for q < s q<s q < s no estimate J q ≤ C δ b J_q\le C\delta^b J q ≤ C δ b with b > ( s − q ) / s b>(s-q)/s b > ( s − q ) / s can hold uniformly on this fixed class. For q = s q=s q = s no uniform modulus tending to zero with δ \delta δ can bound J s J_s J s on this class.
Proof. Use an outward tubular chart ( y ′ , t ) (y',t) ( y ′ , t ) on a fixed patch, with t ≤ 0 t \le0 t ≤ 0 inside. Fix 0 ≤ χ ≤ 1 0 \le\chi\le1 0 ≤ χ ≤ 1 supported compactly in the surface patch and equal to one near p p p . Choose χ n ∈ C c ∞ ( ( − 1 , 1 ) ) \chi_n \in C^\infty_c((-1,1)) χ n ∈ C c ∞ (( − 1 , 1 )) , 0 ≤ χ n ≤ 1 0 \le\chi_n \le1 0 ≤ χ n ≤ 1 , equal to one near zero, and put
ϑ q ( τ ) = χ n ( τ ) ( − τ ) q q ! , q = 0 , … , s . (340) \vartheta_q(\tau)=\chi_n(\tau)\frac{(-\tau)^q}{q!},\qquad q=0,\ldots,s. \tag*{(340)} ϑ q ( τ ) = χ n ( τ ) q ! ( − τ ) q , q = 0 , … , s . (340)
The cutoff χ n \chi_n χ n is fixed independently of a a a ; these profiles enter the perturbations (341) . These are fixed profiles, nonnegative for τ ≤ 0 \tau\le0 τ ≤ 0 , positive on a nonempty interior interval, and equal to ( − τ ) q / q ! (-\tau)^q/q! ( − τ ) q / q ! near zero. Choose a 0 ≤ 1 a_0 \le1 a 0 ≤ 1 so small that − a 0 < t ≤ 0 -a_0 < t \le0 − a 0 < t ≤ 0 remains within the tubular chart, and define
λ a , q − λ b = a a m p a s χ ( y ′ ) ϑ q ( t / a ) , μ a , q = μ b , 0 < a < a 0 , (341) \lambda_{a,q}-\lambda_b=a_{\mathrm{amp}}a^s\chi(y')\vartheta_q(t/a),\qquad\mu_{a,q}=\mu_b,\qquad0<a<a_0, \tag*{(341)} λ a , q − λ b = a amp a s χ ( y ′ ) ϑ q ( t / a ) , μ a , q = μ b , 0 < a < a 0 , (341)
with zero perturbation off this collar patch. The compact support of χ \chi χ and χ n \chi_n χ n makes this a smooth function on Ω ‾ \overline{\Omega} Ω .
We verify the a priori bounds uniformly for these families. In chart coordinates a derivative with b b b normal derivatives and tangential multi-index γ \gamma γ is
a a m p a s − b ( ∂ y ′ γ χ ) ϑ q ( b ) ( t / a ) , b + ∣ γ ∣ ≤ s . (342) a_{\mathrm{amp}}a^{s-b}(\partial_{y'}^\gamma\chi)\vartheta_q^{(b)}(t/a),\qquad b+|\gamma|\le s. \tag*{(342)} a amp a s − b ( ∂ y ′ γ χ ) ϑ q ( b ) ( t / a ) , b + ∣ γ ∣ ≤ s . (342)
The fixed coordinate chain rule expresses every Cartesian derivative through order s s s as a finite sum of these terms with fixed bounded coefficients. Since a ≤ 1 a \le1 a ≤ 1 and b ≤ s b \le s b ≤ s , there is a fixed K s ≥ 1 K_s \ge1 K s ≥ 1 , the maximum over the finitely many profiles q = 0 , … , s q=0,\ldots,s q = 0 , … , s , such that
∥ λ a , q − λ b ∥ C s ( Ω ‾ ) ≤ K s a a m p . (343) \|\lambda_{a,q}-\lambda_b\|_{C^s(\overline{\Omega})}\le K_s a_{\mathrm{amp}}. \tag*{(343)} ∥ λ a , q − λ b ∥ C s ( Ω ) ≤ K s a amp . (343)
Fix 0 < a a m p ≤ σ b / ( 2 K s ) 0<a_{\mathrm{amp}}\le\sigma_b/(2K_s) 0 < a amp ≤ σ b / ( 2 K s ) before taking any a ↓ 0 a\downarrow0 a ↓ 0 . The triangle inequality and (338) give ∥ λ a , q ∥ C s + ∥ μ b ∥ C s ≤ M − σ b / 2 \|\lambda_{a,q}\|_{C^s}+\|\mu_b\|_{C^s}\le M-\sigma_b/2 ∥ λ a , q ∥ C s + ∥ μ b ∥ C s ≤ M − σ b /2 . Moreover λ a , q − λ b ≥ 0 \lambda_{a,q}-\lambda_b\ge0 λ a , q − λ b ≥ 0 and μ a , q = μ b \mu_{a,q}=\mu_b μ a , q = μ b , so both strong-convexity inequalities retain at least the original lower bound c 0 + σ b c_0+\sigma_b c 0 + σ b . The amplitude therefore works for the whole family in the one prescribed class; it is not selected separately for each a a a .
At p p p , the straight-normal convention (281) gives
∣ ∂ ν q ( λ a , q − λ b ) ( p ) ∣ = a a m p a s − q . (344) \left|\partial_\nu^q(\lambda_{a,q}-\lambda_b)(p)\right|=a_{\mathrm{amp}}a^{s-q}. \tag*{(344)} ∣ ∂ ν q ( λ a , q − λ b ) ( p ) ∣ = a amp a s − q . (344)
Expansion of ( ν ( p ) ⋅ ∇ ) q (\nu(p)\cdot\nabla)^q ( ν ( p ) ⋅ ∇ ) q in Cartesian derivatives shows ∣ ∂ ν q f ( p ) ∣ ≤ 3 q / 2 ∑ ∣ α ∣ = q ∣ ∂ α f ( p ) ∣ |\partial_\nu^q f(p)|\le3^{q/2}\sum_{|\alpha|=q}|\partial^\alpha f(p)| ∣ ∂ ν q f ( p ) ∣ ≤ 3 q /2 ∑ ∣ α ∣ = q ∣ ∂ α f ( p ) ∣ . Hence the asserted lower bound for J q J_q J q holds with c q = 3 − q / 2 c_q=3^{-q/2} c q = 3 − q /2 . The same argument includes q = 0 q=0 q = 0 . The undifferentiated perturbation is at most a a m p a s ∥ ϑ q ∥ ∞ a_{\mathrm{amp}}a^s\|\vartheta_q\|_\infty a amp a s ∥ ϑ q ∥ ∞ . The physical forward estimate (30) therefore gives the upper bound for δ a , q \delta_{a,q} δ a , q .
Strict positivity of that discrepancy does not require inverse uniqueness. Prescribe the real affine trace f ( x ) = x f(x)=x f ( x ) = x on every boundary component. The constant-medium solution is u b ( x ) = x u_b(x)=x u b ( x ) = x . Write u a = u b + w u_a=u_b+w u a = u b + w with w ∈ H 1 ( Ω ; R 3 ) w\in H^1(\Omega;\mathbb{R}^3) w ∈ H 1 ( Ω ; R 3 ) , and denote the baseline Hermitian form by a b a_b a b . Its weak equation gives a b ( u b , w ) = 0 a_b(u_b,w)=0 a b ( u b , w ) = 0 . Consequently the exact energy identity is
⟨ ( Λ a , q − Λ b ) f , f ⟩ = a b ( w , w ) + ∫ Ω ( λ a , q − λ b ) ∣ div u a ∣ 2 d x > 0. (345) \left\langle(\Lambda_{a,q}-\Lambda_b)f,f\right\rangle=a_b(w,w)+\int_\Omega(\lambda_{a,q}-\lambda_b)|\operatorname{div}u_a|^2\,dx>0. \tag*{(345)} ⟨ ( Λ a , q − Λ b ) f , f ⟩ = a b ( w , w ) + ∫ Ω ( λ a , q − λ b ) ∣ div u a ∣ 2 d x > 0. (345)
Both terms are nonnegative. If their sum were zero, coercivity on H 0 1 H^1_0 H 0 1 would force w = 0 w=0 w = 0 ; the second term would then equal 9 ∫ Ω ( λ a , q − λ b ) > 0 9\int_\Omega(\lambda_{a,q}-\lambda_b)>0 9 ∫ Ω ( λ a , q − λ b ) > 0 , a contradiction. Since the trace f f f has finite H 1 / 2 H^{1/2} H 1/2 norm, its strictly positive pairing implies δ a , q > 0 \delta_{a,q}>0 δ a , q > 0 .
For q < s q<s q < s , an improved power would give c q a a m p a s − q ≤ C ( C q a a m p a s ) b c_q a_{\mathrm{amp}}a^{s-q}\le C(C_q a_{\mathrm{amp}}a^s)^b c q a amp a s − q ≤ C ( C q a amp a s ) b for all small a a a . This is impossible when b > s − q b>s-q b > s − q . For q = s q=s q = s , the positive data errors tend to zero while J s ≥ c s a a m p > 0 J_s\ge c_s a_{\mathrm{amp}}>0 J s ≥ c s a amp > 0 , excluding every modulus that tends to zero. This proposition concerns boundary sharpness; it makes no claim about optimality of the interior logarithmic exponent. ▫
A strict inequality in each of the three prior conditions for a constant background is equivalent to the existence of a σ b > 0 \sigma_b>0 σ b > 0 in (8.58).
For smooth materials with equal data, the finite-order boundary result can instead be applied separately at every derivative order. This yields the following boundary consequence without a uniform estimate over all orders.
Corollary 8.7. If both materials are smooth up to the boundary, are strongly convex, and have identical full DN maps, all their Cartesian boundary derivatives agree, on every boundary component.
Proof. Fix q ≥ 0 q\geq0 q ≥ 0 and choose an integer s > q s>q s > q , s ≥ 2 s\geq2 s ≥ 2 . Smoothness on Ω ‾ \overline{\Omega} Ω and strong convexity give finite M s M_s M s and c 0 > 0 c_0>0 c 0 > 0 such that both materials belong to C s ( Ω ‾ ; M s , c 0 ) C^s(\overline{\Omega};M_s,c_0) C s ( Ω ; M s , c 0 ) in (1.4). Equality of the DN maps means δ = 0 \delta=0 δ = 0 . The exact-data conclusion of Theorem 8.5 therefore gives J q = 0 J_q=0 J q = 0 , and hence
∂ α ( λ 1 − λ 2 ) ∣ ∂ Ω = 0 , ∂ α ( μ 1 − μ 2 ) ∣ ∂ Ω = 0 , ∣ α ∣ ≤ q . \partial^\alpha(\lambda_1-\lambda_2)|_{\partial\Omega}=0,\qquad\partial^\alpha(\mu_1-\mu_2)|_{\partial\Omega}=0,\qquad|\alpha|\leq q. ∂ α ( λ 1 − λ 2 ) ∣ ∂ Ω = 0 , ∂ α ( μ 1 − μ 2 ) ∣ ∂ Ω = 0 , ∣ α ∣ ≤ q .
The boundary norm defining J q J_q J q covers every boundary component. Since q q q was arbitrary, all finite-order boundary derivatives agree. The bound M s M_s M s may depend on s s s ; no uniform estimate as q → ∞ q\to\infty q → ∞ is needed.
At one fixed point and scale the coefficient extraction uses s s s scales and two polarizations. The point p p p still ranges over the entire measurement boundary and the scale is tied to the noise level. This is an estimate from the full DN operator, not a uniqueness or stability theorem for a fixed finite set of globally prescribed displacements.
The top-order endpoint and extension
The case q = s q=s q = s of Proposition 8.6 shows that a bounded C s C^s C s class with strict prior slack has no uniform vanishing modulus for the top-order boundary jet. The extension in Section 9 therefore uses only J s − 1 J_{s-1} J s − 1 , together with the uniform C s C^s C s bound and a positive tolerance. Its construction preserves the original coefficients without requiring their order-s s s boundary derivatives to match.
Uniform extensions and DN comparison on an enlarged domain
Uniform paired extensions reduce the original-domain problem to a common-exterior class; the enlargement estimate then controls the DN discrepancy on the fixed larger domain.
A fixed strongly convex initial extension
The initial extension to a fixed constant background depends only on the original admissible class (1.4), independently of boundary determination.
Lemma 9.1. There are a fixed ball B R e ⊃ ⊃ Ω ‾ B_{R_e}\supset\supset\overline{\Omega} B R e ⊃⊃ Ω , a fixed constant pair ( λ ∗ , μ ∗ ) (\lambda_*,\mu_*) ( λ ∗ , μ ∗ ) , a fixed number η e > 0 \eta_e>0 η e > 0 and M 0 M_0 M 0 , depending only on s , Ω , M , c 0 s,\Omega,M,c_0 s , Ω , M , c 0 , such that every admissible material has an extension ( λ 0 , μ 0 ) ∈ C s ( R 3 ; R ) 2 (\lambda^0,\mu^0)\in C^s(\mathbb{R}^3;\mathbb{R})^2 ( λ 0 , μ 0 ) ∈ C s ( R 3 ; R ) 2 satisfying
( λ 0 , μ 0 ) = ( λ , μ ) on Ω ‾ , ∥ λ 0 ∥ C s + ∥ μ 0 ∥ C s ≤ M 0 , μ 0 ≥ c 0 / 2 , 3 λ 0 + 2 μ 0 ≥ c 0 / 2 , ( λ 0 , μ 0 ) = ( λ ∗ , μ ∗ ) on R 3 ∖ B R e − η e ‾ . (346) \begin{aligned}
(\lambda^0,\mu^0)&=(\lambda,\mu)\ \text{on }\overline{\Omega},\qquad\|\lambda^0\|_{C^s}+\|\mu^0\|_{C^s}\leq M_0,\\
\mu^0&\geq c_0/2,\qquad3\lambda^0+2\mu^0\geq c_0/2,\\
(\lambda^0,\mu^0)&=(\lambda_*,\mu_*)\ \text{on }\mathbb{R}^3\setminus\overline{B_{R_e-\eta_e}}. \tag*{(346)}
\end{aligned} ( λ 0 , μ 0 ) μ 0 ( λ 0 , μ 0 ) = ( λ , μ ) on Ω , ∥ λ 0 ∥ C s + ∥ μ 0 ∥ C s ≤ M 0 , ≥ c 0 /2 , 3 λ 0 + 2 μ 0 ≥ c 0 /2 , = ( λ ∗ , μ ∗ ) on R 3 ∖ B R e − η e . (346)
Proof. Use the tubular map (8.1) for ∣ t ∣ < a |t|<a ∣ t ∣ < a , with a fixed a > 0 a>0 a > 0 . Its inverse and derivatives through the orders used in (9.2)–(9.5) have fixed bounds on every boundary component. The disjoint fixed tubular collars are chosen before a material is selected; the remaining exterior, including bounded complementary components, will be assigned the same background. For a scalar f ∈ C s ( Ω ‾ ) f \in C^{s}(\overline{\Omega}) f ∈ C s ( Ω ) and 0 < t < a / ( s + 1 ) 0 < t < a/(s+1) 0 < t < a / ( s + 1 ) define a preliminary reflection by
F f ( p , t ) = ∑ ℓ = 1 s + 1 c ℓ f ( Φ ( p , − ℓ t ) ) , c ℓ = ( − 1 ) ℓ − 1 ℓ ( s + 2 ℓ + 1 ) . (347) F_f(p,t)=\sum_{\ell=1}^{s+1}c_{\ell}f(\Phi(p,-\ell t)),\qquad c_{\ell}=(-1)^{\ell-1}\ell\binom{s+2}{\ell+1}. \tag*{(347)} F f ( p , t ) = ℓ = 1 ∑ s + 1 c ℓ f ( Φ ( p , − ℓ t )) , c ℓ = ( − 1 ) ℓ − 1 ℓ ( ℓ + 1 s + 2 ) . (347)
The coefficients are the Lagrange evaluation weights at 1 1 1 for the nodes − 1 , … , − ( s + 1 ) -1,\ldots,-(s+1) − 1 , … , − ( s + 1 ) . Consequently
∑ ℓ = 1 s + 1 c ℓ ( − ℓ ) b = 1 , 0 ≤ b ≤ s . (348) \sum_{\ell=1}^{s+1}c_{\ell}(-\ell)^b=1,\qquad0\le b\le s. \tag*{(348)} ℓ = 1 ∑ s + 1 c ℓ ( − ℓ ) b = 1 , 0 ≤ b ≤ s . (348)
One can verify the weight formula directly by evaluating ∏ 1 ≤ ℓ ′ ≤ s + 1 ℓ ′ ≠ ℓ x + ℓ ′ ℓ ′ − ℓ \prod_{\substack{1\le\ell'\le s+1\\\ell'\ne\ell}}\frac{x+\ell'}{\ell'-\ell} ∏ 1 ≤ ℓ ′ ≤ s + 1 ℓ ′ = ℓ ℓ ′ − ℓ x + ℓ ′ at x = 1 x=1 x = 1 . For any surface multi-index γ \gamma γ and normal order b b b with ∣ γ ∣ + b ≤ s |\gamma|+b\le s ∣ γ ∣ + b ≤ s , differentiation of (347) gives the interior trace ∂ p γ ∂ t b ( f ∘ Φ ) ( p , 0 ) \partial_p^\gamma\partial_t^b(f\circ\Phi)(p,0) ∂ p γ ∂ t b ( f ∘ Φ ) ( p , 0 ) by (348) . Each such derivative is continuous up to t = 0 t=0 t = 0 . The interior function f f f and the exterior function F f F_f F f have matching traces of all derivatives through order s s s , by (348) . They therefore define a C s C^s C s function across the boundary. The chain rule on fixed charts gives its norm bound C s , Ω ∥ f ∥ C s C_{s,\Omega}\|f\|_{C^s} C s , Ω ∥ f ∥ C s .
The reflection weights need not be nonnegative, so positivity is not inferred from their signs. Instead, their sum is one and the mean-value formula along the interior normal segments gives
∣ F f ( p , t ) − f ( p ) ∣ ≤ t ∥ ∇ f ∥ ∞ ∑ ℓ = 1 s + 1 ∣ c ℓ ∣ ℓ . (349) |F_f(p,t)-f(p)|\le t\|\nabla f\|_{\infty}\sum_{\ell=1}^{s+1}|c_{\ell}|\ell. \tag*{(349)} ∣ F f ( p , t ) − f ( p ) ∣ ≤ t ∥∇ f ∥ ∞ ℓ = 1 ∑ s + 1 ∣ c ℓ ∣ ℓ . (349)
Apply this inequality to f = μ f=\mu f = μ and f = 3 λ + 2 μ f=3\lambda+2\mu f = 3 λ + 2 μ . Choose a fixed b 0 < a / ( s + 1 ) b_0<a/(s+1) b 0 < a / ( s + 1 ) , also b 0 ≤ 1 b_0\le1 b 0 ≤ 1 , so that the right-hand side is at most c 0 / 2 c_0/2 c 0 /2 for all the materials when 0 < t < b 0 0<t<b_0 0 < t < b 0 . Both elastic inequalities then hold with lower bound c 0 / 2 c_0/2 c 0 /2 for the preliminary reflected pair on this collar.
Fix ( λ ∗ , μ ∗ ) = ( c 0 , c 0 ) (\lambda_*,\mu_*)=(c_0,c_0) ( λ ∗ , μ ∗ ) = ( c 0 , c 0 ) and a smooth scalar function ϑ \vartheta ϑ with values in [ 0 , 1 ] [0,1] [ 0 , 1 ] , equal to one for t ≤ b 0 / 3 t\le b_0/3 t ≤ b 0 /3 and zero for t ≥ 2 b 0 / 3 t\ge2b_0/3 t ≥ 2 b 0 /3 . On the exterior collar put
( λ 0 , μ 0 ) ( Φ ( p , t ) ) = ϑ ( t ) ( F λ , F μ ) ( p , t ) + ( 1 − ϑ ( t ) ) ( λ ∗ , μ ∗ ) . (350) (\lambda^0,\mu^0)(\Phi(p,t))=\vartheta(t)(F_\lambda,F_\mu)(p,t)+(1-\vartheta(t))(\lambda_*,\mu_*). \tag*{(350)} ( λ 0 , μ 0 ) ( Φ ( p , t )) = ϑ ( t ) ( F λ , F μ ) ( p , t ) + ( 1 − ϑ ( t )) ( λ ∗ , μ ∗ ) . (350)
Use the original material inside and the constant pair everywhere else. Near t = 0 t=0 t = 0 this is the reflected extension, so all derivatives through order s s s still match. Near the far edge of the collar it is constant. Convexity of the two linear elastic inequalities preserves the lower bound c 0 / 2 c_0/2 c 0 /2 . All cutoff derivatives have fixed bounds, so the C s C^s C s norm is bounded by a fixed M 0 M_0 M 0 . Finally choose R e , η e R_e,\eta_e R e , η e so that the closure of the fixed collar lies strictly inside B R e − η e B_{R_e-\eta_e} B R e − η e . These choices were made before considering either a discrepancy or a noise level. This proves the lemma. □ \square □
Applied separately to two materials, this construction gives uniform extensions but does not control their exterior difference by the boundary discrepancy. We obtain that control in the following paired construction.
All derivative orders in a compensated collar
A compensated collar preserves the full C s C^s C s bounds of the extensions while reducing their exterior discrepancy using boundary derivatives below order s s s .
Before selecting a pair or a noise level, fix the extension in Lemma 9.1, its background, B R e B_{R_e} B R e , η e \eta_e η e , M 0 M_0 M 0 , and the cutoff used in (9.11). Fix also one number M e M_e M e large enough for the finite bounds in (9.15) after replacing e r e x t − s e^{r_{\mathrm{ext}}-s} e r ext − s by 1 1 1 and r e x t s − n r e x t − b r_{\mathrm{ext}}^{s-n}r_{\mathrm{ext}}^{-b} r ext s − n r ext − b by 1 1 1 . This defines M e M_e M e from s , Ω , M 0 s,\Omega,M_0 s , Ω , M 0 alone; Theorem 9.2 verifies the uniform bound. In particular M e M_e M e is fixed before any later common-exterior exponent γ ∗ , e x t \gamma_{*,\mathrm{ext}} γ ∗ , ext is chosen. For two materials set
e ∂ = J s − 1 , e t o l > 0 , r e x t = ( e ∂ + e t o l ) 1 / s . (351) e_{\partial}=J_{s-1},\qquad e_{\mathrm{tol}}>0,\qquad r_{\mathrm{ext}}=(e_{\partial}+e_{\mathrm{tol}})^{1/s}. \tag*{(351)} e ∂ = J s − 1 , e tol > 0 , r ext = ( e ∂ + e tol ) 1/ s . (351)
Here J s − 1 J_{s-1} J s − 1 is defined in (9) . The symbols ε t o l \varepsilon_{\mathrm{tol}} ε tol and r e x t r_{\mathrm{ext}} r ext denote a free positive tolerance and an exterior collar width; neither is a CGO parameter or the boundary test scale ε \varepsilon ε .
The positive collar width permits exact boundary matching while suppressing the exterior difference, with derivative bounds uniform as the collar shrinks.
Theorem 9.2. There exist fixed constants M e , C , t e x t , 0 > 0 M_{\mathrm{e}}, C, t_{\mathrm{ext},0} > 0 M e , C , t ext , 0 > 0 and the ball and background from Lemma 9.1 such that, if
0 < e ∂ + ε t o l < t e x t , 0 , (352) 0 < e_{\partial} + \varepsilon_{\mathrm{tol}} < t_{\mathrm{ext},0}, \tag*{(352)} 0 < e ∂ + ε tol < t ext , 0 , (352) the materials have extensions ( λ ~ j , μ ~ j ) ∈ C s ( R 3 ; R ) 2 (\widetilde{\lambda}_{j},\widetilde{\mu}_{j}) \in C^{s}(\mathbb{R}^{3};\mathbb{R})^{2} ( λ j , μ j ) ∈ C s ( R 3 ; R ) 2 , j = 1 , 2 j=1,2 j = 1 , 2 , with exact original values on Ω ‾ \overline{\Omega} Ω , common fixed background outside B R e − η e B_{R_{\mathrm{e}}-\eta_{\mathrm{e}}} B R e − η e , C s C^{s} C s norm at most M e M_{\mathrm{e}} M e , and strong-convexity lower bounds c 0 / 2 c_{0}/2 c 0 /2 . They satisfy
ε e x t : = ∥ λ ~ 1 − λ ~ 2 ∥ L ∞ ( R 3 ∖ Ω ‾ ) + ∥ μ ~ 1 − μ ~ 2 ∥ L ∞ ( R 3 ∖ Ω ‾ ) ≤ C ( e ∂ + ε t o l ) . (353) \varepsilon_{\mathrm{ext}} := \lVert\widetilde{\lambda}_{1} - \widetilde{\lambda}_{2} \rVert_{L^{\infty}(\mathbb{R}^{3}\setminus\overline{\Omega})} + \lVert\widetilde{\mu}_{1} - \widetilde{\mu}_{2} \rVert_{L^{\infty}(\mathbb{R}^{3}\setminus\overline{\Omega})} \leq C(e_{\partial} + \varepsilon_{\mathrm{tol}}). \tag*{(353)} ε ext := ∥ λ 1 − λ 2 ∥ L ∞ ( R 3 ∖ Ω ) + ∥ μ 1 − μ 2 ∥ L ∞ ( R 3 ∖ Ω ) ≤ C ( e ∂ + ε tol ) . (353) The input discrepancy e ∂ e_{\partial} e ∂ , tolerance ε t o l \varepsilon_{\mathrm{tol}} ε tol , and width r e x t r_{\mathrm{ext}} r ext are defined in (351) . All constants and the outer support geometry are independent of the two materials, e ∂ e_{\partial} e ∂ and ε t o l \varepsilon_{\mathrm{tol}} ε tol within the stated class.
Proof. Use Lemma 9.1 for each member, with the same fixed construction. Define, for either scalar component,
D f = f 2 0 − f 1 0 , f ∈ { λ , μ } . (354) D_f = f^{0}_{2} - f^{0}_{1}, \qquad f \in\{\lambda,\mu\}. \tag*{(354)} D f = f 2 0 − f 1 0 , f ∈ { λ , μ } . (354)
This is a C s C^{s} C s scalar function on R 3 \mathbb{R}^{3} R 3 with norm at most 2 M 0 2M_{0} 2 M 0 , and its Cartesian boundary derivatives through order s − 1 s-1 s − 1 have norm bounded by e ∂ e_{\partial} e ∂ .
For a multi-index α \alpha α , n = ∣ α ∣ ≤ s − 1 n = |\alpha| \leq s-1 n = ∣ α ∣ ≤ s − 1 , apply the one-dimensional Taylor formula to t ↦ ∂ α D f ( p + t ν ( p ) ) t \mapsto\partial^{\alpha}D_f(p+t\nu(p)) t ↦ ∂ α D f ( p + t ν ( p )) through order s − 1 − n s-1-n s − 1 − n . The vector ν ( p ) \nu(p) ν ( p ) is held fixed in this differentiation. Every Taylor coefficient is a linear combination of Cartesian boundary derivatives of total order at most s − 1 s-1 s − 1 . The integral remainder uses only derivatives of total order s s s . For 0 < t < b 0 ≤ 1 0<t<b_{0}\leq1 0 < t < b 0 ≤ 1 this proves
∣ ∂ α D f ( p + t ν ( p ) ) ∣ ≤ C e ∂ + C M 0 t s − ∣ α ∣ , ∣ α ∣ ≤ s − 1 , \left|\partial^{\alpha}D_f(p+t\nu(p))\right| \leq Ce_{\partial}+CM_{0}t^{s-|\alpha|}, \qquad|\alpha|\leq s-1, ∣ ∂ α D f ( p + t ν ( p )) ∣ ≤ C e ∂ + C M 0 t s − ∣ α ∣ , ∣ α ∣ ≤ s − 1 ,
∣ ∂ α D f ∣ ≤ 2 M 0 , ∣ α ∣ = s . (355) \left|\partial^{\alpha}D_f\right|\leq2M_{0}, \qquad|\alpha|=s. \tag*{(355)} ∣ ∂ α D f ∣ ≤ 2 M 0 , ∣ α ∣ = s . (355)
No derivative of order s + 1 s+1 s + 1 occurs in this argument.
Fix a smooth function χ : R → [ 0 , 1 ] \chi:\mathbb{R}\to[0,1] χ : R → [ 0 , 1 ] equal to one on ( − ∞ , 1 / 3 ] (-\infty,1/3] ( − ∞ , 1/3 ] and zero on [ 2 / 3 , ∞ ) [2/3,\infty) [ 2/3 , ∞ ) . Set
d ∂ ( x ) = dist ( x , ∂ Ω ) , χ r e x t ( x ) = χ ( d ∂ ( x ) / r e x t ) (356) d_{\partial}(x)=\operatorname{dist}(x,\partial\Omega), \qquad\chi_{r_{\mathrm{ext}}}(x)=\chi(d_{\partial}(x)/r_{\mathrm{ext}}) \tag*{(356)} d ∂ ( x ) = dist ( x , ∂ Ω ) , χ r ext ( x ) = χ ( d ∂ ( x ) / r ext ) (356)
in the exterior collar, and extend χ r e x t \chi_{r_{\mathrm{ext}}} χ r ext by zero beyond that collar. Choose t e x t , 0 ≤ min { 1 , ( b 0 / 4 ) s } t_{\mathrm{ext},0}\leq\min\{1,(b_{0}/4)^{s}\} t ext , 0 ≤ min { 1 , ( b 0 /4 ) s } so that 0 < r e x t < b 0 / 4 0<r_{\mathrm{ext}}<b_{0}/4 0 < r ext < b 0 /4 . The distance function is smooth here with fixed derivatives. For b = ∣ β ∣ ≥ 1 b=|\beta|\geq1 b = ∣ β ∣ ≥ 1 , the chain rule is a finite sum of terms
r e x t − j χ ( j ) ( d ∂ / r e x t ) ∏ i = 1 j ∂ γ i d ∂ , 1 ≤ j ≤ b , ∣ γ i ∣ ≥ 1 , ∑ i γ i = β . (357) r_{\mathrm{ext}}^{-j}\chi^{(j)}(d_{\partial}/r_{\mathrm{ext}})\prod_{i=1}^{j}\partial^{\gamma_i}d_{\partial}, \qquad1\leq j\leq b,\quad|\gamma_i|\geq1,\quad\sum_i\gamma_i=\beta. \tag*{(357)} r ext − j χ ( j ) ( d ∂ / r ext ) i = 1 ∏ j ∂ γ i d ∂ , 1 ≤ j ≤ b , ∣ γ i ∣ ≥ 1 , i ∑ γ i = β . (357)
All derivatives of d ∂ d_{\partial} d ∂ are fixed and bounded. Since r e x t ≤ 1 r_{\mathrm{ext}}\leq1 r ext ≤ 1 , each r e x t − j ≤ r e x t − b r_{\mathrm{ext}}^{-j}\leq r_{\mathrm{ext}}^{-b} r ext − j ≤ r ext − b . For b = 0 b=0 b = 0 use 0 ≤ χ r e x t ≤ 1 0\leq\chi_{r_{\mathrm{ext}}}\leq1 0 ≤ χ r ext ≤ 1 . Thus every multi-index satisfies
∣ ∂ β χ r e x t ∣ ≤ C β r e x t − ∣ β ∣ , 0 ≤ ∣ β ∣ ≤ s , supp χ r e x t ⊂ { d ∂ < 2 r e x t / 3 } . (358) \left|\partial^{\beta}\chi_{r_{\mathrm{ext}}}\right|\leq C_{\beta}r_{\mathrm{ext}}^{-|\beta|}, \qquad0\leq|\beta|\leq s,\qquad\operatorname{supp}\chi_{r_{\mathrm{ext}}}\subset\{d_{\partial}<2r_{\mathrm{ext}}/3\}. \tag*{(358)} ∂ β χ r ext ≤ C β r ext − ∣ β ∣ , 0 ≤ ∣ β ∣ ≤ s , supp χ r ext ⊂ { d ∂ < 2 r ext /3 } . (358)
For the order-zero bound use 0 ≤ χ r e x t ≤ 1 0\leq\chi_{r_{\mathrm{ext}}}\leq1 0 ≤ χ r ext ≤ 1 . Keep the first extension and, outside Ω \Omega Ω , define
f ~ 1 = f 1 0 , f ~ 2 = f 1 0 + χ r e x t D f , f = λ , μ . (359) \widetilde{f}_{1}=f^{0}_{1}, \qquad\widetilde{f}_{2}=f^{0}_{1}+\chi_{r_{\mathrm{ext}}}D_f, \qquad f=\lambda,\mu. \tag*{(359)} f 1 = f 1 0 , f 2 = f 1 0 + χ r ext D f , f = λ , μ . (359)
Inside Ω \Omega Ω keep each original member unchanged. Since r e x t > 0 r_{\mathrm{ext}}>0 r ext > 0 and χ r e x t = 1 \chi_{r_{\mathrm{ext}}}=1 χ r ext = 1 on a neighborhood of the boundary, the second extension there is exactly its initial extension. This proves C s C^{s} C s matching, including the top derivative, without assuming that the two top jets agree. The second pair is a convex combination of the two initial pairs, so it has the same lower strong-convexity bounds.
For the derivative estimates, let ∣ α ∣ = n ≤ s |\alpha|=n\le s ∣ α ∣ = n ≤ s and β ≤ α \beta\le\alpha β ≤ α componentwise, and write b = ∣ β ∣ b=|\beta| b = ∣ β ∣ . For b ≥ 1 b\ge1 b ≥ 1 , n − b ≤ s − 1 n-b\le s-1 n − b ≤ s − 1 and the support is in d ≤ r e x t d\le r_{\mathrm{ext}} d ≤ r ext , so
∣ ∂ β χ r e x t ∂ α − β D f ∣ ≤ C r e x t − b ( e ∂ + M 0 r e x t s − n + b ) = C { ( e ∂ r e x t − s ) r e x t s − b + M 0 r e x t s − n } ≤ C ( 1 + M 0 ) , e ∂ r e x t − s = e ∂ e ∂ + ε t o l ≤ 1. (360) \begin{aligned}
|\partial^\beta\chi_{r_{\mathrm{ext}}}\partial^{\alpha-\beta}D_f|
&\le C r_{\mathrm{ext}}^{-b}\left(e_{\partial}+M_0r_{\mathrm{ext}}^{s-n+b}\right)\\
&=C\left\{(e_{\partial}r_{\mathrm{ext}}^{-s})r_{\mathrm{ext}}^{s-b}+M_0r_{\mathrm{ext}}^{s-n}\right\}\le C(1+M_0),\\
e_{\partial}r_{\mathrm{ext}}^{-s}
&=\frac{e_{\partial}}{e_{\partial}+\varepsilon_{\mathrm{tol}}}\le1. \tag*{(360)}
\end{aligned} ∣ ∂ β χ r ext ∂ α − β D f ∣ e ∂ r ext − s ≤ C r ext − b ( e ∂ + M 0 r ext s − n + b ) = C { ( e ∂ r ext − s ) r ext s − b + M 0 r ext s − n } ≤ C ( 1 + M 0 ) , = e ∂ + ε tol e ∂ ≤ 1. (360)
Both exponents s − b s-b s − b and s − n s-n s − n are nonnegative. The exact rule is ∂ α ( χ r e x t D f ) = ∑ β ≤ α ( α β ) ∂ β χ r e x t ∂ α − β D f \partial^\alpha(\chi_{r_{\mathrm{ext}}}D_f)=\sum_{\beta\le\alpha}\binom{\alpha}{\beta}\partial^\beta\chi_{r_{\mathrm{ext}}}\partial^{\alpha-\beta}D_f ∂ α ( χ r ext D f ) = ∑ β ≤ α ( β α ) ∂ β χ r ext ∂ α − β D f . Its binomial factors and number of terms depend only on s s s . The sum of these estimates gives the preselected C s C^s C s bound M e M_e M e , uniformly in r e x t r_{\mathrm{ext}} r ext . The undifferentiated-cutoff term uses the initial C s C^s C s norm and does not require a small top jet.
Finally, the order-zero line in (9.10) gives ∣ χ r e x t D f ∣ ≤ C ( e ∂ + M 0 r e x t s ) ≤ C ′ ( e ∂ + ε t o l ) |\chi_{r_{\mathrm{ext}}}D_f|\le C(e_{\partial}+M_0r_{\mathrm{ext}}^s)\le C'(e_{\partial}+\varepsilon_{\mathrm{tol}}) ∣ χ r ext D f ∣ ≤ C ( e ∂ + M 0 r ext s ) ≤ C ′ ( e ∂ + ε tol ) . This proves (9.8). The background and the fixed outer collar from the initial extension have not moved.
Remark. The extensions belong to one fixed C s C^s C s class, independently of the boundary discrepancy and the positive tolerance. This uniformity permits the subsequent common-exterior estimates to be applied with fixed constants. When e ∂ = 0 e_{\partial}=0 e ∂ = 0 , the construction remains valid for every sufficiently small positive ε t o l \varepsilon_{\mathrm{tol}} ε tol ; it does not require a zero-width collar or convergence of the extensions as the tolerance tends to zero.
Noise transfer on one fixed enlarged class
Theorem 8.5 and Proposition 2.1 give a bound for the enlarged-domain error from the measured original-domain error. The order of choices is important: first fix the original class, the boundary atlas, the initial extension and the compensated bound M e M_e M e ; then fix any constants of a common-exterior theorem on that entire enlarged class; only afterwards choose the noise and positive tolerance.
Corollary 9.3. For two members of C s ( Ω ; M , c 0 ) C^s(\Omega;M,c_0) C s ( Ω ; M , c 0 ) in (1.4), let C J C_J C J be the class-uniform constant of Theorem 8.5 at order s − 1 s-1 s − 1 . For 0 < δ < min { 1 , δ 0 } 0<\delta<\min\{1,\delta_0\} 0 < δ < min { 1 , δ 0 } , that theorem gives
e ∂ = J s − 1 ≤ C J δ 1 / s (361) e_{\partial}=J_{s-1}\le C_J\delta^{1/s} \tag*{(361)} e ∂ = J s − 1 ≤ C J δ 1/ s (361) with the original-domain DN discrepancy δ \delta δ . There is a class-uniform threshold δ t r , 0 ≤ min { 1 , δ 0 } \delta_{\mathrm{tr},0}\le\min\{1,\delta_0\} δ tr , 0 ≤ min { 1 , δ 0 } such that, for 0 < δ < δ t r , 0 0<\delta<\delta_{\mathrm{tr},0} 0 < δ < δ tr , 0 , one may choose
ε t o l = δ + e ∂ > 0 , r e x t s = e ∂ + ε t o l = δ + 2 e ∂ . (362) \varepsilon_{\mathrm{tol}}=\delta+e_{\partial}>0,\qquad r_{\mathrm{ext}}^s=e_{\partial}+\varepsilon_{\mathrm{tol}}=\delta+2e_{\partial}. \tag*{(362)} ε tol = δ + e ∂ > 0 , r ext s = e ∂ + ε tol = δ + 2 e ∂ . (362) Then the extensions of Theorem 9.2 belong to one fixed enlarged-domain class and satisfy
Δ e ≤ C ( δ + e ∂ ) ≤ C ′ δ 1 / s . (363) \Delta_e\le C(\delta+e_{\partial})\le C'\delta^{1/s}. \tag*{(363)} Δ e ≤ C ( δ + e ∂ ) ≤ C ′ δ 1/ s . (363) Here δ \delta δ , e ∂ e_{\partial} e ∂ , and Δ e \Delta_e Δ e are defined in (1.6), (9.6), and (1.16), respectively. If δ = 0 \delta=0 δ = 0 , then e ∂ = 0 e_{\partial}=0 e ∂ = 0 , and every 0 < ε t o l < t e x t , 0 0<\varepsilon_{\mathrm{tol}}<t_{\mathrm{ext},0} 0 < ε tol < t ext , 0 gives extensions in the same fixed class satisfying Δ e ≤ C ε t o l \Delta_e\le C\varepsilon_{\mathrm{tol}} Δ e ≤ C ε tol .
Proof. For positive noise, (9.16) and 0 < δ < 1 0<\delta<1 0 < δ < 1 give
0 < e ∂ + ε t o l = δ + 2 e ∂ ≤ ( 1 + 2 C J ) δ 1 / s . 0<e_{\partial}+\varepsilon_{\mathrm{tol}}=\delta+2e_{\partial}\le(1+2C_J)\delta^{1/s}. 0 < e ∂ + ε tol = δ + 2 e ∂ ≤ ( 1 + 2 C J ) δ 1/ s .
Choose
0 < δ t r , 0 < min { 1 , δ 0 , ( t e x t , 0 2 ( 1 + 2 C J ) ) s } . 0<\delta_{\mathrm{tr},0}<\min\left\{1,\delta_0,\left(\frac{t_{\mathrm{ext},0}}{2(1+2C_J)}\right)^s\right\}. 0 < δ tr , 0 < min { 1 , δ 0 , ( 2 ( 1 + 2 C J ) t ext , 0 ) s } .
Then (352) holds for 0 < δ < δ t r , 0 0 < \delta< \delta_{\mathrm{tr},0} 0 < δ < δ tr , 0 . Apply Theorem 9.2 , followed by Proposition 2.1, to obtain
ϵ e x t ≤ C E ( δ + 2 e ∂ ) , \epsilon_{\mathrm{ext}} \le C_E(\delta+ 2e_{\partial}), ϵ ext ≤ C E ( δ + 2 e ∂ ) ,
Δ e ≤ C A ( δ + ϵ e x t ) ≤ C A { δ + C E ( δ + 2 e ∂ ) } ≤ C ′ δ 1 / s . \Delta_e \le C_A(\delta+ \epsilon_{\mathrm{ext}}) \le C_A\{\delta+ C_E(\delta+ 2e_{\partial})\} \le C'\delta^{1/s}. Δ e ≤ C A ( δ + ϵ ext ) ≤ C A { δ + C E ( δ + 2 e ∂ )} ≤ C ′ δ 1/ s .
Here C E C_E C E is the extension constant and C A C_A C A is the enlargement constant in the cited results. This also gives Δ e ≤ C ( δ + e ∂ ) \Delta_e \le C(\delta+ e_{\partial}) Δ e ≤ C ( δ + e ∂ ) and proves (363) . If δ = 0 \delta= 0 δ = 0 , the exact-data conclusion of Theorem 8.5 gives e ∂ = 0 e_{\partial} = 0 e ∂ = 0 . For every 0 < ϵ t o l < t e x t , 0 0 < \epsilon_{\mathrm{tol}} < t_{\mathrm{ext},0} 0 < ϵ tol < t ext , 0 , Theorem 9.2 and Proposition 2.1 give
ϵ e x t ≤ C E ϵ t o l , Δ e ≤ C A C E ϵ t o l . \epsilon_{\mathrm{ext}} \le C_E\epsilon_{\mathrm{tol}}, \qquad\Delta_e \le C_A C_E\epsilon_{\mathrm{tol}}. ϵ ext ≤ C E ϵ tol , Δ e ≤ C A C E ϵ tol .
The ball, background, C s C^s C s bound M e M_e M e , and strong-convexity bound are those fixed in Theorem 9.2 . They are independent of the positive tolerance, as required.
The enlarged DN maps are proof quantities, not additional measurements. If a larger working ball B R B_R B R is required by a common-exterior CGO/Betti argument, fix it with B R e ⋐ B R B_{R_e} \Subset B_R B R e ⋐ B R before choosing noise. Both extensions already equal the same background near and outside ∂ B R e \partial B_{R_e} ∂ B R e ; extending them by that background changes no class bound. Proposition 2.1 on B R e ⊂ B R B_{R_e} \subset B_R B R e ⊂ B R gives ∥ Λ 1 B R − Λ 2 B R ∥ ≤ C Δ e \lVert\Lambda_1^{B_R}-\Lambda_2^{B_R}\rVert\le C\Delta_e ∥ Λ 1 B R − Λ 2 B R ∥ ≤ C Δ e . Thus B R e B_{R_e} B R e remains the DN comparison domain and B R B_R B R the larger working domain. Choose any later exponent γ ∗ , e x t \gamma_{*,\mathrm{ext}} γ ∗ , ext for this fixed class before selecting the noise level.
Original-domain stability and exact-data uniqueness
Boundary recovery and paired extensions transfer Proposition 7.5 to the original domain. The enlarged prior class is fixed before the measurement error or tolerance is chosen.
Boundary control and the fixed enlarged class
Fix an integer m ≥ 4 m \ge4 m ≥ 4 , put s = 2 m + 14 s = 2m + 14 s = 2 m + 14 , and fix G s ( Ω ; M , c 0 ) \mathcal{G}^s(\Omega; M, c_0) G s ( Ω ; M , c 0 ) as in (4) . This section proves the general regularity-dependent estimate; Theorem 1.1 is its specialization to m = 4 m = 4 m = 4 . Theorem 8.5 , Theorem 9.2 , and Proposition 2.1 give the following original-domain estimates independently of the common-exterior theorem.
By Theorem 8.5 with q = s − 1 q = s - 1 q = s − 1 , (361) holds for fixed C J C_J C J , δ B , 0 > 0 \delta_{B,0} > 0 δ B , 0 > 0 , and e ∂ = 0 e_{\partial} = 0 e ∂ = 0 for δ = 0 \delta= 0 δ = 0 . Here e ∂ e_{\partial} e ∂ is defined in (351) .
Theorem 9.2 supplies fixed B R e B_{R_e} B R e , M e M_e M e , strong-convexity bounds, a common constant exterior, and C E C_E C E , t e x t , 0 > 0 t_{\mathrm{ext},0} > 0 t ext , 0 > 0 such that, whenever
ϵ t o l > 0 , 0 < e ∂ + ϵ t o l < t e x t , 0 , (364) \epsilon_{\mathrm{tol}} > 0, \qquad0 < e_{\partial} + \epsilon_{\mathrm{tol}} < t_{\mathrm{ext},0}, \tag*{(364)} ϵ tol > 0 , 0 < e ∂ + ϵ tol < t ext , 0 , (364)
the original coefficients have exact interior C s C^s C s extensions in this fixed enlarged class, with
ϵ e x t ≤ C E ( e ∂ + ϵ t o l ) . (365) \epsilon_{\mathrm{ext}} \le C_E(e_{\partial} + \epsilon_{\mathrm{tol}}). \tag*{(365)} ϵ ext ≤ C E ( e ∂ + ϵ tol ) . (365)
Here ϵ e x t \epsilon_{\mathrm{ext}} ϵ ext is defined in (353) and r e x t r_{\mathrm{ext}} r ext in (351) . Only the collar width changes. Theorem 9.2 fixes the enlarged norm and convexity bounds and the common background. The moment and annulus constants for that class, hence γ ∗ , e x t \gamma_{*,\mathrm{ext}} γ ∗ , ext , are fixed before choosing the tolerance.
Proposition 2.1 gives, without exterior agreement on ∂ Ω \partial\Omega ∂ Ω ,
Δ e ≤ C A ( δ + ϵ e x t ) . (366) \Delta_e \le C_A(\delta+ \epsilon_{\mathrm{ext}}). \tag*{(366)} Δ e ≤ C A ( δ + ϵ ext ) . (366)
Here Δ e \Delta_e Δ e is defined in (16) . This comparison follows from the elastic Alessandrini identity (27) . The reduced Betti identity with a common sphere-neighborhood condition is used in the separate common-exterior construction, on its larger working sphere.
The extension bound (365) and DN comparison (366) allow application of Proposition 7.5 with uniform constants.
Proposition 10.1. Fix an integer m ≥ 4 m \ge4 m ≥ 4 , put s = 2 m + 14 s = 2m + 14 s = 2 m + 14 , and let C S ( Ω ; M , c 0 ) \mathcal{C}^{\mathrm{S}}(\Omega; M, c_0) C S ( Ω ; M , c 0 ) be defined by (4) . There exists γ ∗ , e x t ∈ ( 0 , 1 ) \gamma_{*,\mathrm{ext}} \in(0,1) γ ∗ , ext ∈ ( 0 , 1 ) , depending only on Ω \Omega Ω , s s s , M M M , c 0 c_0 c 0 , such that, for every
0 < α < γ ∗ , e x t ( m − 3 ) 28 m , (367) 0 < \alpha< \frac{\gamma_{*,\mathrm{ext}}(m-3)}{28m}, \tag*{(367)} 0 < α < 28 m γ ∗ , ext ( m − 3 ) , (367) there are constants C α , δ A , 0 > 0 C_\alpha,\delta_{A,0}>0 C α , δ A , 0 > 0 , depending only on Ω \Omega Ω , s s s , M M M , c 0 c_0 c 0 , α \alpha α , for which (11) holds for every two pairs ( λ j , μ j ) ∈ C S ( Ω ; M , c 0 ) (\lambda_j,\mu_j) \in\mathcal{C}^{\mathrm{S}}(\Omega;M,c_0) ( λ j , μ j ) ∈ C S ( Ω ; M , c 0 ) , j = 1 , 2 j=1,2 j = 1 , 2 , with 0 < δ < δ A , 0 0 < \delta< \delta_{A,0} 0 < δ < δ A , 0 . The discrepancy δ \delta δ is defined in (6) . If δ = 0 \delta=0 δ = 0 , the two pairs coincide in Ω \Omega Ω .
Proof. Fix α \alpha α satisfying (367) and use the constants of Proposition 7.5 for the fixed enlarged class constructed in Subsection 10.1. Its exterior condition is (15) with D = B R e D=B_{R_e} D = B R e and with the constant background from Lemma 9.1. For positive noise choose the tolerance and width in (362) . For 0 < δ < 1 0<\delta<1 0 < δ < 1 , the boundary estimate gives
δ + 2 e ∂ ≤ ( 1 + 2 C J ) δ 1 / s . (368) \delta+2e_{\partial}\le(1+2C_J)\delta^{1/s}. \tag*{(368)} δ + 2 e ∂ ≤ ( 1 + 2 C J ) δ 1/ s . (368)
The extension and physical enlargement estimates therefore imply
Δ e ≤ K e δ 1 / s , K e = max { 1 , C A ( 1 + C E + 2 C E C J ) } . (369) \Delta_e\le K_e\delta^{1/s},\qquad K_e=\max\{1,C_A(1+C_E+2C_EC_J)\}. \tag*{(369)} Δ e ≤ K e δ 1/ s , K e = max { 1 , C A ( 1 + C E + 2 C E C J )} . (369)
An explicit sufficient original-domain threshold is
δ A , 0 = min { 1 , δ B , 0 , ( t e x t , 0 2 ( 1 + 2 C J ) ) s , ( δ c o m , 0 2 K e ) s , ( 2 K e ) − 2 s } > 0. (370) \delta_{A,0}=\min\left\{1,\delta_{B,0},\left(\frac{t_{\mathrm{ext},0}}{2(1+2C_J)}\right)^s,\left(\frac{\delta_{\mathrm{com},0}}{2K_e}\right)^s,(2K_e)^{-2s}\right\}>0. \tag*{(370)} δ A , 0 = min { 1 , δ B , 0 , ( 2 ( 1 + 2 C J ) t ext , 0 ) s , ( 2 K e δ com , 0 ) s , ( 2 K e ) − 2 s } > 0. (370)
For 0 < δ < δ A , 0 0<\delta<\delta_{A,0} 0 < δ < δ A , 0 , the extension range is admissible and Δ e < δ c o m , 0 \Delta_e<\delta_{\mathrm{com},0} Δ e < δ com , 0 . If Δ e = 0 \Delta_e=0 Δ e = 0 , the exact-data conclusion of Proposition 7.5 gives equality on B R e B_{R_e} B R e , hence on Ω \Omega Ω .
If Δ e > 0 \Delta_e>0 Δ e > 0 , let L = ∣ log δ ∣ L=|\log\delta| L = ∣ log δ ∣ . From (369) and (370) ,
∣ log Δ e ∣ ≥ L s − log K e ≥ L 2 s . (371) |\log\Delta_e|\ge\frac{L}{s}-\log K_e\ge\frac{L}{2s}. \tag*{(371)} ∣ log Δ e ∣ ≥ s L − log K e ≥ 2 s L . (371)
The common-exterior estimate, with D = B R e D=B_{R_e} D = B R e and δ D = Δ e \delta_D=\Delta_e δ D = Δ e , then gives
∥ λ 1 − λ 2 ∥ L ∞ ( Ω ) + ∥ μ 1 − μ 2 ∥ L ∞ ( Ω ) ≤ C c o m ( 2 s ) α ∣ log δ ∣ − α . (372) \|\lambda_1-\lambda_2\|_{L^\infty(\Omega)}+\|\mu_1-\mu_2\|_{L^\infty(\Omega)}
\le C_{\mathrm{com}}(2s)^\alpha|\log\delta|^{-\alpha}. \tag*{(372)} ∥ λ 1 − λ 2 ∥ L ∞ ( Ω ) + ∥ μ 1 − μ 2 ∥ L ∞ ( Ω ) ≤ C com ( 2 s ) α ∣ log δ ∣ − α . (372)
The factor 1 / s 1/s 1/ s in the outer-noise power changes the multiplicative constant in this comparison, not the logarithmic exponent.
If the original δ = 0 \delta=0 δ = 0 , the exact-data boundary estimate gives e ∂ = 0 e_{\partial}=0 e ∂ = 0 . For every sufficiently small t > 0 t>0 t > 0 , use ϵ t o l = t \epsilon_{\mathrm{tol}}=t ϵ tol = t in Theorem 9.2. The enlarged discrepancy satisfies
Δ e ( t ) ≤ C A C E t . (373) \Delta_e(t)\le C_A C_E t. \tag*{(373)} Δ e ( t ) ≤ C A C E t . (373)
If it vanishes for any such t t t , equality follows from the common-exterior zero-error assertion. Otherwise put K t = max { 1 , C A C E } K_t=\max\{1,C_A C_E\} K t = max { 1 , C A C E } and take t t t small enough that K t t < min { 1 , δ c o m , 0 } K_t t<\min\{1,\delta_{\mathrm{com},0}\} K t t < min { 1 , δ com , 0 } . For the fixed original pair, the same uniform common-exterior estimate yields
∥ λ 1 − λ 2 ∥ L ∞ ( Ω ) + ∥ μ 1 − μ 2 ∥ L ∞ ( Ω ) ≤ C c o m ∣ log ( K t t ) ∣ − α ⟶ 0. (374) \|\lambda_1-\lambda_2\|_{L^\infty(\Omega)}+\|\mu_1-\mu_2\|_{L^\infty(\Omega)}
\le C_{\mathrm{com}}|\log(K_t t)|^{-\alpha}\longrightarrow0. \tag*{(374)} ∥ λ 1 − λ 2 ∥ L ∞ ( Ω ) + ∥ μ 1 − μ 2 ∥ L ∞ ( Ω ) ≤ C com ∣ log ( K t t ) ∣ − α ⟶ 0. (374)
Only this numerical upper bound is sent to zero. Each extension is used at a positive tolerance; neither a zero-width extension nor convergence of the extensions in C S C^{\mathrm{S}} C S is assumed. □ \square □
Taking m = 4 m=4 m = 4 in Proposition 10.1 yields Theorem 1.1.
Boundary stability and finite-regularity uniqueness
The boundary upper bounds and finite-regularity uniqueness require no strict prior slack; that additional hypothesis enters only the sharpness assertion.
Proof of Theorem 1.2. For each q = 0 , … , s − 1 q=0,\ldots,s-1 q = 0 , … , s − 1 , Theorem 8.5 on Ω \Omega Ω gives fixed constants C q , δ q > 0 C_q,\delta_q>0 C q , δ q > 0 such that
J q ≤ C q δ ( s − q ) / s , 0 < δ < δ q . J_q \le C_q\delta^{(s-q)/s}, \qquad0<\delta<\delta_q. J q ≤ C q δ ( s − q ) / s , 0 < δ < δ q .
Set C = max 0 ≤ q < s C q C=\max_{0\le q<s}C_q C = max 0 ≤ q < s C q and δ B , 0 = min 0 ≤ q < s δ q > 0 \delta_{B,0}=\min_{0\le q<s}\delta_q>0 δ B , 0 = min 0 ≤ q < s δ q > 0 . These choices prove (12) simultaneously for all the stated orders. At δ = 0 \delta=0 δ = 0 , the exact-data conclusion of the cited theorem gives J q = 0 J_q=0 J q = 0 for every q < s q<s q < s .
Proof of Proposition 1.3. Proposition 8.6 constructs the smooth families in the prescribed class using the strict slack (13) . Absorb its fixed positive amplitude into c q , C q c_q,C_q c q , C q . Its estimates (339) give
0 < δ a , q ≤ C q a s , J q ≥ c q a s − q . 0<\delta_{a,q}\le C_qa^s,\qquad J_q\ge c_qa^{s-q}. 0 < δ a , q ≤ C q a s , J q ≥ c q a s − q .
For q < s q<s q < s , set ϑ q = ( s − q ) / s \vartheta_q=(s-q)/s ϑ q = ( s − q ) / s . The inequalities above imply
J q δ a , q ϑ q ≥ c q C q ϑ q > 0. \frac{J_q}{\delta_{a,q}^{\vartheta_q}}\ge\frac{c_q}{C_q^{\vartheta_q}}>0. δ a , q ϑ q J q ≥ C q ϑ q c q > 0.
Since 0 < δ a , q ≤ C q a s → 0 0<\delta_{a,q}\le C_qa^s\to0 0 < δ a , q ≤ C q a s → 0 , a uniform estimate J q ≤ ω ( δ ) J_q\le\omega(\delta) J q ≤ ω ( δ ) with ω ( t ) = o ( t ϑ q ) \omega(t)=o(t^{\vartheta_q}) ω ( t ) = o ( t ϑ q ) would imply
0 < c q C q ϑ q ≤ ω ( δ a , q ) δ a , q ϑ q ⟶ 0 , 0<\frac{c_q}{C_q^{\vartheta_q}}\le\frac{\omega(\delta_{a,q})}{\delta_{a,q}^{\vartheta_q}}\longrightarrow0, 0 < C q ϑ q c q ≤ δ a , q ϑ q ω ( δ a , q ) ⟶ 0 ,
a contradiction. For q = s q=s q = s , the same construction gives J s ≥ c s > 0 J_s\ge c_s>0 J s ≥ c s > 0 while δ a , s → 0 \delta_{a,s}\to0 δ a , s → 0 through positive values. An estimate J s ≤ ω ( δ a , s ) J_s\le\omega(\delta_{a,s}) J s ≤ ω ( δ a , s ) with ω ( t ) → 0 \omega(t)\to0 ω ( t ) → 0 as t ↓ 0 t\downarrow0 t ↓ 0 is therefore impossible.
Proof of Corollary 1.4. For the fixed C 22 C^{22} C 22 strongly convex materials, choose
c 0 = 1 2 min j = 1 , 2 x ∈ Ω ‾ { μ j ( x ) , 3 λ j ( x ) + 2 μ j ( x ) } > 0 , c_0=\frac{1}{2}\min_{\substack{j=1,2\\x\in\overline{\Omega}}}\{\mu_j(x),3\lambda_j(x)+2\mu_j(x)\}>0, c 0 = 2 1 j = 1 , 2 x ∈ Ω min { μ j ( x ) , 3 λ j ( x ) + 2 μ j ( x )} > 0 ,
M = 1 + max j = 1 , 2 ( ∥ λ j ∥ C 22 ( Ω ‾ ) + ∥ μ j ∥ C 22 ( Ω ‾ ) ) < ∞ . M=1+\max_{j=1,2}\left(\|\lambda_j\|_{C^{22}(\overline{\Omega})}+\|\mu_j\|_{C^{22}(\overline{\Omega})}\right)<\infty. M = 1 + j = 1 , 2 max ( ∥ λ j ∥ C 22 ( Ω ) + ∥ μ j ∥ C 22 ( Ω ) ) < ∞.
The strict strong-convexity inequalities on the compact set Ω ‾ \overline{\Omega} Ω give c 0 > 0 c_0>0 c 0 > 0 , and the C 22 C^{22} C 22 regularity makes M M M finite. Both pairs then belong to C 22 ( Ω ; M , c 0 ) C^{22}(\Omega;M,c_0) C 22 ( Ω ; M , c 0 ) in (4) . Equality of their DN maps gives δ = 0 \delta=0 δ = 0 , so the exact-data conclusion of Proposition 10.1 with m = 4 m=4 m = 4 yields
∥ λ 1 − λ 2 ∥ L ∞ ( Ω ) + ∥ μ 1 − μ 2 ∥ L ∞ ( Ω ) = 0. \|\lambda_1-\lambda_2\|_{L^\infty(\Omega)}+\|\mu_1-\mu_2\|_{L^\infty(\Omega)}=0. ∥ λ 1 − λ 2 ∥ L ∞ ( Ω ) + ∥ μ 1 − μ 2 ∥ L ∞ ( Ω ) = 0.
The coefficients are continuous, hence agree on Ω ‾ \overline{\Omega} Ω . The constants M , c 0 M,c_0 M , c 0 are chosen for this fixed pair and need not be prescribed in advance.
Algebraic identities and local regularity
The computations below verify the exact reduction, its formal transpose, and the supplementary physical polarization. Interior regularity for each fixed h h h then justifies the Betti identity for the exact reduced solutions.
Verification of the exact reduction
All indices below range from one to three, and repeated indices are summed. For smooth fields, set
a = μ − 1 = c 2 , Σ = λ + 2 μ , p f = Δ f − ( Δ a / a ) f , u ( v ) = c v , u ( f ) = a ∇ f − f ∇ a , T ( v , f ) = u ( v ) + u ( f ) . (375) \begin{aligned}
a &= \mu^{-1}=c^2,\qquad\Sigma=\lambda+2\mu,\qquad p_f=\Delta f-(\Delta a/a)f,\\
u^{(v)}&=cv,\qquad u^{(f)}=a\nabla f-f\nabla a,\qquad T(v,f)=u^{(v)}+u^{(f)}. \tag*{(375)}
\end{aligned} a u ( v ) = μ − 1 = c 2 , Σ = λ + 2 μ , p f = Δ f − ( Δ a / a ) f , = c v , u ( f ) = a ∇ f − f ∇ a , T ( v , f ) = u ( v ) + u ( f ) . (375)
Here c c c and T T T are defined in (31) and (34) .
Scalar contribution. By the product rule,
div u ( f ) = a Δ f − f Δ a = a p f , ε ( u ( f ) ) = a ∇ 2 f − f ∇ 2 a . (376) \operatorname{div}u^{(f)}=a\Delta f-f\Delta a=ap_f,\qquad\varepsilon(u^{(f)})=a\nabla^2f-f\nabla^2a. \tag*{(376)} div u ( f ) = a Δ f − f Δ a = a p f , ε ( u ( f ) ) = a ∇ 2 f − f ∇ 2 a . (376)
Consequently
σ λ , μ ( u ( f ) ) i j = a λ p f δ i j + 2 f i j − 2 μ f a i j . (377) \sigma_{\lambda,\mu}(u^{(f)})_{ij}=a\lambda p_f\delta_{ij}+2f_{ij}-2\mu f a_{ij}. \tag*{(377)} σ λ , μ ( u ( f ) ) ij = aλ p f δ ij + 2 f ij − 2 μ f a ij . (377)
Taking the divergence of the stress tensor in (A.3), we obtain the component identities
( L u ( f ) ) i = ∂ i ( a λ p f ) + 2 ∂ i Δ f − 2 μ a i j f j − 2 f μ j a i j − 2 μ f ∂ i Δ a . (378) (Lu^{(f)})_i=\partial_i(a\lambda p_f)+2\partial_i\Delta f-2\mu a_{ij}f_j-2f\mu_j a_{ij}-2\mu f\partial_i\Delta a. \tag*{(378)} ( L u ( f ) ) i = ∂ i ( aλ p f ) + 2 ∂ i Δ f − 2 μ a ij f j − 2 f μ j a ij − 2 μ f ∂ i Δ a . (378)
Since a Σ p f = a λ p f + 2 Δ f − 2 μ ( Δ a ) f a\Sigma p_f=a\lambda p_f+2\Delta f-2\mu(\Delta a)f a Σ p f = aλ p f + 2Δ f − 2 μ ( Δ a ) f , subtract its gradient from this equality. The derivatives ∂ i Δ a \partial_i\Delta a ∂ i Δ a cancel, leaving
L u ( f ) = ∇ ( a Σ p f ) + 2 μ ( Δ a I 3 − ∇ 2 a ) ∇ f + 2 μ 2 ( ∇ 2 a − Δ a I 3 ) ∇ a f . (379) Lu^{(f)}=\nabla(a\Sigma p_f)+2\mu(\Delta aI_3-\nabla^2a)\nabla f+2\mu^2(\nabla^2a-\Delta aI_3)\nabla a\,f. \tag*{(379)} L u ( f ) = ∇ ( a Σ p f ) + 2 μ ( Δ a I 3 − ∇ 2 a ) ∇ f + 2 μ 2 ( ∇ 2 a − Δ a I 3 ) ∇ a f . (379)
Now ∇ a = − a g μ \nabla a=-ag_\mu ∇ a = − a g μ , ∇ 2 a = a ( g μ ⊗ g μ − ∇ 2 ℓ μ ) \nabla^2a=a(g_\mu\otimes g_\mu-\nabla^2\ell_\mu) ∇ 2 a = a ( g μ ⊗ g μ − ∇ 2 ℓ μ ) , − Δ a / a = q μ -\Delta a/a=q_\mu − Δ a / a = q μ , and Δ a I 3 − ∇ 2 a = ( c / 2 ) F μ \Delta aI_3-\nabla^2a=(c/2)F_\mu Δ a I 3 − ∇ 2 a = ( c /2 ) F μ . Thus the nongradient terms of (A.5) are c − 1 F μ ∇ f + c − 1 F μ g μ f c^{-1}F_\mu\nabla f+c^{-1}F_\mu g_\mu f c − 1 F μ ∇ f + c − 1 F μ g μ f , and its gradient is ∇ [ ( Δ f + q μ f ) / ( 1 − κ ) ] \nabla[(\Delta f+q_\mu f)/(1-\kappa)] ∇ [( Δ f + q μ f ) / ( 1 − κ )] .
Vector contribution. The expanded Lamé operator is
( L u ) i = μ Δ u i + ( λ + μ ) ∂ i div u + λ i div u + μ j ( ∂ j u i + ∂ i u j ) . (380) (Lu)_i=\mu\Delta u_i+(\lambda+\mu)\partial_i\operatorname{div}u+\lambda_i\operatorname{div}u+\mu_j(\partial_j u_i+\partial_i u_j). \tag*{(380)} ( Lu ) i = μ Δ u i + ( λ + μ ) ∂ i div u + λ i div u + μ j ( ∂ j u i + ∂ i u j ) . (380)
Define
Ψ v = c ( λ + μ ) div v + ( λ − μ ) c j v j . (381) \Psi_v=c(\lambda+\mu)\operatorname{div}v+(\lambda-\mu)c_jv_j. \tag*{(381)} Ψ v = c ( λ + μ ) div v + ( λ − μ ) c j v j . (381)
To make the first-order cancellation explicit, before subtracting ∂ i Ψ v \partial_i\Psi_v ∂ i Ψ v , the coefficients of the indicated terms are
[ ∂ j v i ] : 2 μ c j + c μ j = 0 , [ ∂ i v j ] : ( λ + μ ) c j + c μ j , [ div v ] : ( λ + μ ) c i + c λ i . (382) \begin{aligned}
[\partial_jv_i]&:\quad2\mu c_j+c\mu_j=0,\\
[\partial_iv_j]&:\quad(\lambda+\mu)c_j+c\mu_j,\\
[\operatorname{div}v]&:\quad(\lambda+\mu)c_i+c\lambda_i. \tag*{(382)}
\end{aligned} [ ∂ j v i ] [ ∂ i v j ] [ div v ] : 2 μ c j + c μ j = 0 , : ( λ + μ ) c j + c μ j , : ( λ + μ ) c i + c λ i . (382)
We subtract the displayed gradient contribution. The term ( λ − μ ) c j ∂ i v j (\lambda-\mu)c_j\partial_i v_j ( λ − μ ) c j ∂ i v j cancels, and 2 μ c j + c μ j = 0 2\mu c_j+c\mu_j=0 2 μ c j + c μ j = 0 . We also subtract ∂ i [ c ( λ + μ ) ] div v \partial_i[c(\lambda+\mu)]\operatorname{div}v ∂ i [ c ( λ + μ )] div v , which leaves − c μ i div v = − c ( g μ ) i div v -c\mu_i\operatorname{div}v=-c(g_\mu)_i\operatorname{div}v − c μ i div v = − c ( g μ ) i div v . The mixed second derivatives cancel as well. The remaining zeroth-order matrix is
Z i j = 2 μ c i j + μ i c j + μ j c i + ( μ Δ c + μ k c k ) δ i j . (383) Z_{ij}=2\mu c_{ij}+\mu_i c_j+\mu_j c_i+(\mu\Delta c+\mu_kc_k)\delta_{ij}. \tag*{(383)} Z ij = 2 μ c ij + μ i c j + μ j c i + ( μ Δ c + μ k c k ) δ ij . (383)
Since μ = c − 2 \mu=c^{-2} μ = c − 2 , we obtain
c Z = 2 c − 1 ∇ 2 c − 4 c − 2 ∇ c ⊗ ∇ c + ( c − 1 Δ c − 2 c − 2 ∣ ∇ c ∣ 2 ) I 3 = V μ . (384) cZ=2c^{-1}\nabla^2c-4c^{-2}\nabla c\otimes\nabla c+(c^{-1}\Delta c-2c^{-2}|\nabla c|^2)I_3=V_\mu. \tag*{(384)} c Z = 2 c − 1 ∇ 2 c − 4 c − 2 ∇ c ⊗ ∇ c + ( c − 1 Δ c − 2 c − 2 ∣∇ c ∣ 2 ) I 3 = V μ . (384)
The final equality follows by substituting ∇ c = − c g μ / 2 \nabla c=-cg_\mu/2 ∇ c = − c g μ /2 and ∇ 2 c = c ( − ∇ 2 ℓ μ / 2 + g μ ⊗ g μ / 4 ) \nabla^2c=c(-\nabla^2\ell_\mu/2+g_\mu\otimes g_\mu/4) ∇ 2 c = c ( − ∇ 2 ℓ μ /2 + g μ ⊗ g μ /4 ) . We have therefore proved
L ( c v ) = c − 1 ( Δ v − g μ div v + V μ v ) + ∇ [ c ( λ + μ ) div v + ( λ − μ ) ∇ c ⋅ v ] . (385) L(cv)=c^{-1}(\Delta v-g_\mu\operatorname{div}v+V_\mu v)+\nabla[c(\lambda+\mu)\operatorname{div}v+(\lambda-\mu)\nabla c\cdot v]. \tag*{(385)} L ( c v ) = c − 1 ( Δ v − g μ div v + V μ v ) + ∇ [ c ( λ + μ ) div v + ( λ − μ ) ∇ c ⋅ v ] . (385)
The identities c ( λ + μ ) = d / ( 1 − κ ) c(\lambda+\mu)=d/(1-\kappa) c ( λ + μ ) = d / ( 1 − κ ) and ( λ − μ ) ∇ c = − e g μ / ( 1 − κ ) (\lambda-\mu)\nabla c=-eg_\mu/(1-\kappa) ( λ − μ ) ∇ c = − e g μ / ( 1 − κ ) convert the last gradient into the scalar block of S M SM S M . The scalar and vector product-rule identities (A.5) and (A.11) together imply L T = S M LT=SM L T = S M , with all coefficients given in (38) .
Formal transpose and constant background. The formal transpose is taken with respect to the complex-bilinear pairing. For compactly supported tests, ∫ u T ∇ ( q / ( 1 − κ ) ) = − ∫ q div u / ( 1 − κ ) \int u^{T}\nabla(q/(1-\kappa))=-\int q\operatorname{div}u/(1-\kappa) ∫ u T ∇ ( q / ( 1 − κ )) = − ∫ q div u / ( 1 − κ ) . There is no term − ∇ ( 1 − κ ) − 1 ⋅ u -\nabla(1-\kappa)^{-1}\cdot u − ∇ ( 1 − κ ) − 1 ⋅ u in this transpose. The divergence of the physical transform is
div T ( v , f ) = c div v − 1 2 c g μ ⋅ v + μ − 1 ( Δ f + q μ f ) . (386) \operatorname{div}T(v,f)=c\operatorname{div}v-\frac{1}{2}cg_{\mu}\cdot v+\mu^{-1}(\Delta f+q_{\mu}f). \tag*{(386)} div T ( v , f ) = c div v − 2 1 c g μ ⋅ v + μ − 1 ( Δ f + q μ f ) . (386)
These two identities prove (41) directly. For constant coefficients g μ = F μ = V μ = q μ = 0 g_{\mu}=F_{\mu}=V_{\mu}=q_{\mu}=0 g μ = F μ = V μ = q μ = 0 and M ( v , f ) = ( Δ v , Δ f + d div v ) T \mathcal{M}(v,f)=(\Delta v,\Delta f+d\operatorname{div}v)^{T} M ( v , f ) = ( Δ v , Δ f + d div v ) T . Applying the constant-background formula to ( v , f − d ∗ x ⋅ v / 2 ) (v,f-d_{*}x\cdot v/2) ( v , f − d ∗ x ⋅ v /2 ) with d = d ∗ d=d_{*} d = d ∗ , we obtain ( Δ v , Δ f − d ∗ x ⋅ Δ v / 2 ) T = R ∗ Δ ( v , f ) (\Delta v,\Delta f-d_{*}x\cdot\Delta v/2)^{T}=R_{*}\Delta(v,f) ( Δ v , Δ f − d ∗ x ⋅ Δ v /2 ) T = R ∗ Δ ( v , f ) , as used in (48) .
Local regularity of reduced solutions. We justify the local H 4 H^{4} H 4 regularity required by the Betti identity of Lemma 2.3 using interior Laplacian estimates at fixed h h h . This regularity argument is separate from the uniform two-norm remainder estimates.
If an exact reduced field z h ∈ H 2 z_{h}\in H^{2} z h ∈ H 2 solves M z h = 0 \mathcal{M}z_{h}=0 M z h = 0 , then Δ z h = − ∑ A k ∂ k z h − B z h ∈ H 1 \Delta z_{h}=-\sum A_{k}\partial_{k}z_{h}-Bz_{h}\in H^{1} Δ z h = − ∑ A k ∂ k z h − B z h ∈ H 1 locally. An interior Laplacian estimate on nested fixed neighborhoods gives z h ∈ H 3 z_{h}\in H^{3} z h ∈ H 3 ; with C 2 C^{2} C 2 bounds for A k , B A_{k},B A k , B , the right side is then in H 2 H^{2} H 2 and gives z h ∈ H 4 z_{h}\in H^{4} z h ∈ H 4 on the smaller neighborhood. The interior estimate follows by cutting off and using the Fourier estimate for the scalar Laplacian componentwise, with the cutoff commutators controlled by the H 2 H^{2} H 2 estimate for the H 3 H^{3} H 3 step, and by the H 3 H^{3} H 3 estimate for the H 4 H^{4} H 4 step. Thus no system-specific regularity theorem or higher-order uniform CGO inverse is needed. For fixed h h h this proves the regularity used by Lemma 2.3; it makes no h h h -uniform H 4 H^{4} H 4 or H 5 H^{5} H 5 claim.
The supplementary physical polarization
The supplementary lift is checked in both transport blocks, including the transverse cutoff commutators on the entire working domain.
Substitution of g μ = − 2 ∇ c / c g_{\mu}=-2\nabla c/c g μ = − 2∇ c / c in the coefficients gives
F μ = − 4 ∇ 2 c − 4 c − 1 ∇ c ⊗ ∇ c + ( 4 Δ c + 4 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 , F μ − c V μ − 3 ( ∇ ( c g μ ) ) T = ( 3 Δ c + 6 c − 1 ∣ ∇ c ∣ 2 ) I 3 . (387) \begin{aligned}
F_{\mu}&=-4\nabla^{2}c-4c^{-1}\nabla c\otimes\nabla c+(4\Delta c+4c^{-1}|\nabla c|^{2})I_{3},\\
cV_{\mu}&=2\nabla^{2}c-4c^{-1}\nabla c\otimes\nabla c+(\Delta c-2c^{-1}|\nabla c|^{2})I_{3},\\
F_{\mu}-cV_{\mu}-3(\nabla(cg_{\mu}))^{T}&=(3\Delta c+6c^{-1}|\nabla c|^{2})I_{3}. \tag*{(387)}
\end{aligned} F μ c V μ F μ − c V μ − 3 ( ∇ ( c g μ ) ) T = − 4 ∇ 2 c − 4 c − 1 ∇ c ⊗ ∇ c + ( 4Δ c + 4 c − 1 ∣∇ c ∣ 2 ) I 3 , = 2 ∇ 2 c − 4 c − 1 ∇ c ⊗ ∇ c + ( Δ c − 2 c − 1 ∣∇ c ∣ 2 ) I 3 , = ( 3Δ c + 6 c − 1 ∣∇ c ∣ 2 ) I 3 . (387)
Indeed c g μ = − 2 ∇ c cg_{\mu}=-2\nabla c c g μ = − 2∇ c , so the third matrix is symmetric and all Hessian terms cancel. Contracting the last identity in (A.13) with θ T \theta^{T} θ T and q θ q\theta q θ , and using θ ⋅ q θ = 0 \theta\cdot q\theta=0 θ ⋅ q θ = 0 , we obtain
θ T F μ q θ = c θ T V μ q θ + 3 D θ ( c g μ ⋅ q θ ) . (388) \theta^{T}F_{\mu}q\theta=c\theta^{T}V_{\mu}q\theta+3D_{\theta}(cg_{\mu}\cdot q\theta). \tag*{(388)} θ T F μ q θ = c θ T V μ q θ + 3 D θ ( c g μ ⋅ q θ ) . (388)
Contracting the first identity in (A.13) with θ \theta θ in both slots, we obtain − c θ T F μ θ / 2 = D θ 2 ( c 2 ) = α μ -c\theta^{T}F_{\mu}\theta/2=D_{\theta}^{2}(c^{2})=\alpha_{\mu} − c θ T F μ θ /2 = D θ 2 ( c 2 ) = α μ . The contraction uses nullness to remove the scalar multiple of I 3 I_{3} I 3 .
For z 0 s u p = ( q θ , 0 ) T z_{0}^{\mathrm{sup}}=(q\theta,0)^{T} z 0 sup = ( q θ , 0 ) T , z 1 s u p = ( v s u p , f s u p ) T z_{1}^{\mathrm{sup}}=(v_{\mathrm{sup}},f_{\mathrm{sup}})^{T} z 1 sup = ( v sup , f sup ) T and R c o l = c θ ⋅ v s u p R_{\mathrm{col}}=c\theta\cdot v_{\mathrm{sup}} R col = c θ ⋅ v sup , the scalar component of K z 0 s u p − J θ z 1 s u p \mathcal{K}z_{0}^{\mathrm{sup}}-J_{\theta}z_{1}^{\mathrm{sup}} K z 0 sup − J θ z 1 sup is
1 2 c g μ ⋅ q θ + R c o l + 2 c 2 D θ f s u p 1 − κ . (389) \frac{\frac{1}{2}cg_{\mu}\cdot q\theta+R_{\mathrm{col}}+2c^{2}D_{\theta}f_{\mathrm{sup}}}{1-\kappa}. \tag*{(389)} 1 − κ 2 1 c g μ ⋅ q θ + R col + 2 c 2 D θ f sup . (389)
The scalar equation in (4.11) replaces 2 c 2 D θ f s u p 2c^{2}D_{\theta}f_{\mathrm{sup}} 2 c 2 D θ f sup by − κ R c o l − c 2 e g μ ⋅ q θ -\kappa R_{\mathrm{col}}-c^{2}e\,g_{\mu}\cdot q\theta − κ R col − c 2 e g μ ⋅ q θ . Since c / 2 − c 2 e = 3 c ( 1 − κ ) / 2 c/2-c^{2}e=3c(1-\kappa)/2 c /2 − c 2 e = 3 c ( 1 − κ ) /2 , this is precisely − 3 c g μ ⋅ q θ / 2 -3cg_{\mu}\cdot q\theta/2 − 3 c g μ ⋅ q θ /2 . For a general ( p , η ) (p,\eta) ( p , η ) the ambient dual equation has blocks 2 D θ p + θ ( g μ ⋅ p − d η ) 2D_{\theta}p+\theta(g_{\mu}\cdot p-d\eta) 2 D θ p + θ ( g μ ⋅ p − d η ) and 2 D θ η − θ T F μ p 2D_{\theta}\eta-\theta^{T}F_{\mu}p 2 D θ η − θ T F μ p . Insert p = q θ − c f s u p θ p=q\theta-cf_{\mathrm{sup}}\theta p = q θ − c f sup θ and the scalar just computed. The D θ c = − c g μ ⋅ θ / 2 D_{\theta}c=-cg_{\mu}\cdot\theta/2 D θ c = − c g μ ⋅ θ /2 terms cancel in the first block, leaving exactly the first line of (4.13). For the second block, use (A.14) and c θ T F μ θ = − 2 α μ c\theta^{T}F_{\mu}\theta=-2\alpha_{\mu} c θ T F μ θ = − 2 α μ to obtain exactly its second line. This proves the supplementary lift and both transport equations.
In the basis ( n θ , ξ q ) (n_{\theta},\xi_{q}) ( n θ , ξ q ) of the upper two coordinates, Y Y Y has coordinate matrix
\begin{pmatrix}
-2cf_{1} & -2cf_{2} & 2cf_{\mathrm{sup}}\\
0 & 0 & 1\\
-R_{1} & -R_{2} & \eta_{\mathrm{sup}}
\end{pmatrix}. \tag*{(390)}\end{pmatrix}
Its determinant is − 2 c ( f 1 R 2 − f 2 R 1 ) -2c(f_1R_2-f_2R_1) − 2 c ( f 1 R 2 − f 2 R 1 ) . Multiplication by det ( n θ , ξ q ) \det(n_\theta,\xi_q) det ( n θ , ξ q ) proves (4.15) for every value of f s u p , η s u p f_{\mathrm{sup}},\eta_{\mathrm{sup}} f sup , η sup . The geometric minor bound used for the paired lift is also explicit:
q 12 = i s 0 s 1 / 2 , q 13 = ( s 0 2 + s 1 2 ) / 4 , q 23 = i ( s 0 2 − s 1 2 ) / 4. (391) q_{12}=is_0s_1/2,\qquad q_{13}=(s_0^2+s_1^2)/4,\qquad q_{23}=i(s_0^2-s_1^2)/4. \tag*{(391)} q 12 = i s 0 s 1 /2 , q 13 = ( s 0 2 + s 1 2 ) /4 , q 23 = i ( s 0 2 − s 1 2 ) /4. (391)
After taking squared moduli and adding the expressions, we obtain (2.42); the mixed phase terms cancel. The formula also holds on both chart axes.
For the localized supplementary column one must also keep the cutoff term away from the test support. Since D θ χ = 0 D_\theta\chi=0 D θ χ = 0 , div ( χ q θ ) = χ ′ ( ν l e a f ⋅ q θ ) \operatorname{div}(\chi q_\theta)=\chi'(\nu_{\mathrm{leaf}}\cdot q_\theta) div ( χ q θ ) = χ ′ ( ν leaf ⋅ q θ ) , and the scalar component of z 0 s u p z_0^{\mathrm{sup}} z 0 sup is zero, (2.24) gives
[ K , χ ] z 0 s u p = − ( 0 C 3 c χ ′ ( ν l e a f ⋅ q θ ) / ( 1 − κ ) ) , [ J θ , χ ] z 1 s u p = 0 , K a 0 s u p − J θ a 1 s u p = χ Y s u p − J θ w χ + [ K , χ ] z 0 s u p . (392) \begin{aligned}
[K,\chi]z_0^{\mathrm{sup}}&=-\begin{pmatrix}0_{\mathbb{C}^3}\\ c\chi'(\nu_{\mathrm{leaf}}\cdot q_\theta)/(1-\kappa)\end{pmatrix},\\
[J_\theta,\chi]z_1^{\mathrm{sup}}&=0,\\
\mathcal{K}a_0^{\mathrm{sup}}-J_\theta a_1^{\mathrm{sup}}&=\chi\mathcal{Y}_{\mathrm{sup}}-J_\theta w_\chi+[K,\chi]z_0^{\mathrm{sup}}. \tag*{(392)}
\end{aligned} [ K , χ ] z 0 sup [ J θ , χ ] z 1 sup K a 0 sup − J θ a 1 sup = − ( 0 C 3 c χ ′ ( ν leaf ⋅ q θ ) / ( 1 − κ ) ) , = 0 , = χ Y sup − J θ w χ + [ K , χ ] z 0 sup . (392)
This proves (4.31) on the whole working domain. The fixed leaf inverse defining w χ w_\chi w χ preserves transverse support, so w χ w_\chi w χ and its available derivatives vanish on every open interval where χ \chi χ is constant. In particular the localized leading column is exactly Y s u p \mathcal{Y}_{\mathrm{sup}} Y sup near the test support, while remaining uniformly bounded off that support.
Finite-order frames, free inversion, and the parametrix
This appendix proves the frame and local right-inverse theorems of Section 3, with explicit derivative requirements. Both Sobolev bounds are obtained from one Neumann inverse.
Throughout this appendix, norms of finite matrices can be any fixed submultiplicative norm. Constants absorb the fixed equivalence constants. All products involving matrix symbols are kept in their displayed order. The coefficient label a a a is a parameter in a compact metric space; no coefficient-label derivatives will be taken.
The finite-order planar input
Lemma B.1. For compactly supported functions in a fixed disk of the complex plane, set
( C F ) ( z ) = 1 π ∫ C F ( w ) z − w d A ( w ) . (393) (CF)(z)=\frac{1}{\pi}\int_{\mathbb{C}}\frac{F(w)}{z-w}\,dA(w). \tag*{(393)} ( C F ) ( z ) = π 1 ∫ C z − w F ( w ) d A ( w ) . (393) It acts entrywise on matrices. If F ∈ C r , α H F\in C^{r,\alpha_H} F ∈ C r , α H with that support, then on any fixed bounded disk
∂ z ˉ C F = F , ∥ C F ∥ C r + 1 , α H ≤ C ∥ F ∥ C r , α H . (394) \partial_{\bar z}CF=F,\qquad\|CF\|_{C^{r+1,\alpha_H}}\le C\|F\|_{C^{r,\alpha_H}}. \tag*{(394)} ∂ z ˉ C F = F , ∥ C F ∥ C r + 1 , α H ≤ C ∥ F ∥ C r , α H . (394) There is also a bound from compactly supported L ∞ L^\infty L ∞ into C 0 , α H C^{0,\alpha_H} C 0 , α H on fixed disks, for every α H < 1 \alpha_H<1 α H < 1 . Derivatives in additional real parameters commute with this integral; consequently the estimate applies separately to every mixed derivative whose indicated input norm is available. It gains a leaf derivative, not a transverse derivative.
Proof. 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 follows by integration by parts outside a disk of radius ε \varepsilon ε and passage to the limit in its boundary integral. The normalization is thus fixed. The integral of ∣ z − w ∣ − 1 |z-w|^{-1} ∣ z − w ∣ − 1 over a disk of radius a a a is at most C a Ca C a , uniformly in z z z . For two points a distance t t t apart, split the difference of the kernels into the union of their 2 t 2t 2 t -disks and its complement. The first part is bounded by C ∥ F ∥ ∞ t C\|F\|_\infty t C ∥ F ∥ ∞ t ; the second, the kernel difference is bounded by C t ∣ z − w ∣ − 2 Ct|z-w|^{-2} C t ∣ z − w ∣ − 2 . Radial integration bounds this part by C ∥ F ∥ ∞ t ( 1 + ∣ log t ∣ ) C\|F\|_\infty t(1+|\log t|) C ∥ F ∥ ∞ t ( 1 + ∣ log t ∣ ) on fixed disks. Since t ( 1 + ∣ log t ∣ ) ≤ C α H t α H t(1+|\log t|)\le C_{\alpha_H}t^{\alpha_H} t ( 1 + ∣ log t ∣ ) ≤ C α H t α H , this proves the L ∞ L^\infty L ∞ assertion.
For the first derivative, ∂ z C F \partial_z CF ∂ z C F is the principal-value integral with kernel − 1 / ( π z 2 ) -1/(\pi z^2) − 1/ ( π z 2 ) . Since the kernel has zero integral on centered annuli, we may replace the local integrand F ( w ) F(w) F ( w ) by F ( w ) − F ( z ) F(w)-F(z) F ( w ) − F ( z ) . The resulting integral is bounded by C [ F ] α H ∫ 0 R s α H − 1 d s C[F]_{\alpha_H}\int_0^R s^{\alpha_H-1}\,ds C [ F ] α H ∫ 0 R s α H − 1 d s . For its difference at two points separated by t t t , split at radius 4 t 4t 4 t . The near part is bounded by C [ F ] α H t α H C[F]_{\alpha_H}t^{\alpha_H} C [ F ] α H t α H . In the common far region, subtract the value at the first point and shift the second integration variable to put the two kernels on the same domain. The kernel difference is bounded by C t s − 3 Ct s^{-3} C t s − 3 and the subtracted function by [ F ] α H s α H [F]_{\alpha_H}s^{\alpha_H} [ F ] α H s α H . The resulting bound is
C [ F ] α H t ∫ 4 t R s α H − 2 d s ≤ C α H [ F ] α H t α H . (395) C[F]_{\alpha_H}t\int_{4t}^{R}s^{\alpha_H-2}\,ds\leq C_{\alpha_H}[F]_{\alpha_H}t^{\alpha_H}. \tag*{(395)} C [ F ] α H t ∫ 4 t R s α H − 2 d s ≤ C α H [ F ] α H t α H . (395)
The annular pieces introduced by the shift have area O ( t s ) O(ts) O ( t s ) at radius s s s ; the same subtraction and annular cancellation give the identical bound at the inner boundary and C ∥ F ∥ C 0 , α H t C\lVert F\rVert_{C^{0,\alpha_H}}t C ∥ F ∥ C 0 , α H t at the fixed outer boundary. The nonsingular part outside a fixed disk is bounded and Lipschitz on the observation disk. This proves the C 0 , α H C^{0,\alpha_H} C 0 , α H estimate for ∂ z C F \partial_zCF ∂ z C F . Together with ∂ z ˉ C F = F \partial_{\bar z}CF=F ∂ z ˉ C F = F , it proves (B.2) for r = 0 r=0 r = 0 . For higher r r r , real distributional derivatives commute with convolution; apply the r = 0 r=0 r = 0 estimate to ∂ γ F \partial^\gamma F ∂ γ F , ∣ γ ∣ ≤ r |\gamma|\leq r ∣ γ ∣ ≤ r . A fixed cutoff reduces interior variants to this compact-support version. Difference quotients in the extra parameters and the same estimates justify differentiation under the integral and then its iteration.
Proof of Theorem 3.1 . Local existence without a smoothness assumption. On a disk of radius 2 a 2a 2 a choose a cutoff equal to one on a smaller disk and solve
F i = I + C ( χ i B F i ) . (396) F_i=I+C(\chi_iBF_i). \tag*{(396)} F i = I + C ( χ i B F i ) . (396)
On bounded continuous matrices the linear part has norm at most C a ∥ B ∥ ∞ Ca\lVert B\rVert_\infty C a ∥ B ∥ ∞ . Choose a a a so that this is less than 1 / 4 1/4 1/4 . Its Neumann series gives ∥ F i − I ∥ ∞ ≤ 1 / 3 \lVert F_i-I\rVert_\infty\leq1/3 ∥ F i − I ∥ ∞ ≤ 1/3 , hence invertibility. The L ∞ L^\infty L ∞ part of Lemma B.1 first gives F i ∈ C 0 , α H F_i\in C^{0,\alpha_H} F i ∈ C 0 , α H ; its Schauder part then gives F i ∈ C 1 , α H F_i\in C^{1,\alpha_H} F i ∈ C 1 , α H . Iterating the Schauder estimate of Lemma B.1 in (B.4), we obtain F i ∈ C r + 1 , α H F_i\in C^{r+1,\alpha_H} F i ∈ C r + 1 , α H on smaller disks. The construction applies to finite-Hölder-regularity coefficients themselves, not only to their smooth approximations.
Existence on a full disk. Choose an open disk X X X with D ‾ ⋐ X ⋐ U \overline{D}\Subset X\Subset U D ⋐ X ⋐ U , and cover X X X by finitely many of the smaller disks just constructed; this can be done by first covering X ‾ \overline{X} X with disks lying in U U U . On overlaps define
g i j = F i − 1 F j . (397) g_{ij}=F_i^{-1}F_j. \tag*{(397)} g ij = F i − 1 F j . (397)
Differentiation of the inverse gives
∂ z ˉ g i j = − F i − 1 B F i F i − 1 F j + F i − 1 B F j = 0. (398) \partial_{\bar z}g_{ij}=-F_i^{-1}BF_iF_i^{-1}F_j+F_i^{-1}BF_j=0. \tag*{(398)} ∂ z ˉ g ij = − F i − 1 B F i F i − 1 F j + F i − 1 B F j = 0. (398)
These C 1 C^1 C 1 functions are holomorphic, and g i j g j k = g i k g_{ij}g_{jk}=g_{ik} g ij g j k = g ik . Glue the trivial complex manifolds U i × G L ( d s y s , C ) U_i\times GL(d_{\mathrm{sys}},\mathbb{C}) U i × G L ( d sys , C ) by the holomorphic maps ( z , h ) j ↦ ( z , g i j ( z ) h ) i (z,h)_j\mapsto(z,g_{ij}(z)h)_i ( z , h ) j ↦ ( z , g ij ( z ) h ) i . This is a locally trivial holomorphic fiber bundle over the open disk X X X ; the transition maps are biholomorphic left multiplications on its fiber. The local continuous maps
h i = F i − 1 , h i = g i j h j , (399) h_i=F_i^{-1},\qquad h_i=g_{ij}h_j, \tag*{(399)} h i = F i − 1 , h i = g ij h j , (399)
define a continuous section. The precise external input is the existence consequence of Forstnerič–Prezelj [20 Definitions 1.1–1.2 and Theorem 1.3] : a continuous section of a locally trivial holomorphic fiber bundle over a Stein manifold is homotopic to a holomorphic section with its fiber admits a dominating holomorphic spray. Here the open disk X X X is Stein. The trivial vector bundle over the fiber with fiber C d s y s × d s y s \mathbb{C}^{d_{\mathrm{sys}}\times d_{\mathrm{sys}}} C d sys × d sys has the spray ( g , Z ) ↦ g exp Z (g,Z)\mapsto g\exp Z ( g , Z ) ↦ g exp Z ; its vertical differential at Z = 0 Z=0 Z = 0 is the surjective map Z ↦ g Z Z\mapsto gZ Z ↦ g Z . All the hypotheses therefore hold. Write the resulting holomorphic section in the local coordinates as h ~ i \tilde h_i h ~ i . Then F i h ~ i F_i\tilde h_i F i h ~ i agree on overlaps, are invertible, and solve ∂ z ˉ F = B F \partial_{\bar z}F=BF ∂ z ˉ F = B F on X X X . On a neighborhood of the compact disk D ‾ \overline{D} D they have the stated finite regularity. This external input is solely qualitative. Uniform bounds and parameter derivatives will follow from the local perturbation construction, without a parametric Oka theorem.
Perturbation around every finite-regularity center. For clarity choose fixed disks D ⋐ D 1 ⋐ D 2 ⋐ U D \Subset D_1 \Subset D_2 \Subset U D ⋐ D 1 ⋐ D 2 ⋐ U within the original margin. A bounded joint C r + 2 C^{r+2} C r + 2 family, restricted to fixed slightly smaller neighborhoods, has compact closure in C r + 1 , α H C^{r+1,\alpha_H} C r + 1 , α H . Indeed, in each fixed coordinate chart Arzelà–Ascoli gives a subsequence converging uniformly through order r + 1 r+1 r + 1 . For the differences v v v of derivatives of order r + 1 r+1 r + 1 , their common Lipschitz bound and the interpolation inequality
[ v ] α H ≤ C ∥ v ∥ ∞ 1 − α H Lip ( v ) α H (400) [v]_{\alpha_H} \le C\lVert v\rVert_{\infty}^{1-\alpha_H}\operatorname{Lip}(v)^{\alpha_H} \tag*{(400)} [ v ] α H ≤ C ∥ v ∥ ∞ 1 − α H Lip ( v ) α H (400)
give convergence in that Hölder topology. Finitely many charts and the fixed neighborhood margins give the joint assertion, including parameter variables. Work on that entire closure and the compact parameter set, not just on smooth members. More explicitly, the coefficient labels are the joint functions on the fixed spatial and parameter neighborhoods, with the C r + 1 , α H C^{r+1,\alpha_H} C r + 1 , α H topology. Evaluation at p p p and all parameter derivatives through order r r r are continuous into the corresponding leaf Hölder spaces. An original material label enters through its continuous coefficient map. At a center ( a 0 , p 0 ) (a_0,p_0) ( a 0 , p 0 ) , the small-disk construction (B.4) and the holomorphic transitions (B.6) give an invertible frame F 0 F_0 F 0 on a neighborhood of D ‾ 2 \overline{D}_2 D 2 . Take one fixed cutoff χ = 1 \chi=1 χ = 1 near D ‾ 1 \overline{D}_1 D 1 and supported in D 2 D_2 D 2 . Put
V a , p = F 0 − 1 ( B a ( ⋅ , p ) − B a 0 ( ⋅ , p 0 ) ) F 0 , K a , p u = C ( χ V a , p u ) , Ψ = ( I − K a , p ) − 1 I , F = F 0 Ψ . (401) \begin{aligned}
V_{a,p}&=F_0^{-1}\bigl(B^a(\mathord{\cdot},p)-B^{a_0}(\mathord{\cdot},p_0)\bigr)F_0,\\
K_{a,p}u&=C(\chi V_{a,p}u),\qquad\Psi=(I-K_{a,p})^{-1}I,\qquad F=F_0\Psi. \tag*{(401)}
\end{aligned} V a , p K a , p u = F 0 − 1 ( B a ( ⋅ , p ) − B a 0 ( ⋅ , p 0 ) ) F 0 , = C ( χ V a , p u ) , Ψ = ( I − K a , p ) − 1 I , F = F 0 Ψ. (401)
Solve on C 0 , α H ( D ‾ 2 ) C^{0,\alpha_H}(\overline{D}_2) C 0 , α H ( D 2 ) , with the integral extended outside by its formula. Continuity of B B B into that space allows a coefficient–parameter neighborhood on which ∥ K a , p ∥ < 1 / 4 \lVert K_{a,p}\rVert<1/4 ∥ K a , p ∥ < 1/4 and ∥ Ψ − I ∥ ∞ < 1 / 2 \lVert\Psi-I\rVert_\infty<1/2 ∥ Ψ − I ∥ ∞ < 1/2 . Then F F F is invertible and ∂ z ‾ F = B a F \partial_{\overline{z}}F=B^aF ∂ z F = B a F on D 1 D_1 D 1 .
For a nonzero parameter multi-index β \beta β , difference quotients in (B.9), followed by induction, give
( I − K ) ∂ p β Ψ = ∑ γ < β ( β γ ) ( ∂ p β − γ K ) ∂ p γ Ψ . (402) (I-K)\partial_p^\beta\Psi
=
\sum_{\gamma<\beta}
\binom{\beta}{\gamma}
(\partial_p^{\beta-\gamma}K)\partial_p^\gamma\Psi. \tag*{(402)} ( I − K ) ∂ p β Ψ = γ < β ∑ ( γ β ) ( ∂ p β − γ K ) ∂ p γ Ψ. (402)
The inequality γ < β \gamma<\beta γ < β means componentwise γ ≤ β \gamma\le\beta γ ≤ β and γ ≠ β \gamma\ne\beta γ = β . The inverse on the left is the same bounded C 0 , α H C^{0,\alpha_H} C 0 , α H inverse for every β \beta β . Thus all parameter derivatives through order r r r are bounded in that space. Differentiate the equation ∂ z ‾ Ψ = V Ψ \partial_{\overline{z}}\Psi=V\Psi ∂ z Ψ = V Ψ in p p p , and use Lemma B.1 on fixed nested disks. Induction on the leaf order then bounds ∂ p β Ψ \partial_p^\beta\Psi ∂ p β Ψ in C r − ∣ β ∣ + 1 , α H C^{r-|\beta|+1,\alpha_H} C r − ∣ β ∣ + 1 , α H near D ‾ \overline{D} D . At each step the product on the right has total input order at most $r+1; the joint C r + 1 , α H C^{r+1,\alpha_H} C r + 1 , α H closure supplies it. No rectangular requirement of r r r derivatives in each variable is used. Only finitely many nested disks are needed for these inductions: for a fixed requested r r r , choose them between D D D and D 1 D_1 D 1 before any parameter is varied. Their inverse separation constants are therefore fixed. This prevents the local Schauder estimates from introducing a parameter-dependent boundary margin. The identity F F − 1 = I FF^{-1}=I F F − 1 = I and its differentiated versions give the inverse bounds; for example ∂ j F − 1 = − F − 1 ( ∂ j F ) F − 1 \partial_jF^{-1}=-F^{-1}(\partial_jF)F^{-1} ∂ j F − 1 = − F − 1 ( ∂ j F ) F − 1 , with ordered products at every higher order.
The Neumann construction (B.9) also proves the coefficient-label continuity used in the physical construction. If a n → a a_n\to a a n → a within this neighborhood, then K n → K K_n\to K K n → K in operator norm on C 0 , α H C^{0,\alpha_H} C 0 , α H , uniformly on a compact smaller parameter chart. The resolvent identity reads
( I − K n ) − 1 − ( I − K ) − 1 = ( I − K n ) − 1 ( K n − K ) ( I − K ) − 1 . (403) (I-K_n)^{-1}-(I-K)^{-1}=(I-K_n)^{-1}(K_n-K)(I-K)^{-1}. \tag*{(403)} ( I − K n ) − 1 − ( I − K ) − 1 = ( I − K n ) − 1 ( K n − K ) ( I − K ) − 1 . (403)
Its two inverse factors have norm at most 4 / 3 4/3 4/3 . Thus Ψ n → Ψ \Psi_n\to\Psi Ψ n → Ψ in that space. Apply (B.10) successively to obtain the same convergence for every parameter derivative through order r r r . Applying the fixed-disk Cauchy estimate of Lemma B.1 to the difference of the two frame equations, we obtain convergence for all total mixed orders through r r r on the smaller cylinders.
The inverse identities give the same continuity for F − 1 F^{-1} F − 1 . No derivative of a coefficient label has been used, and no neighborhood depends on h h h .
One atlas for the full prior. Apply the coefficient–parameter neighborhoods from (B.9) to the compact closure of the whole prior-bounded class, times P \mathfrak{P} P , and take a finite subcover. The maximum of the finitely many bounds is C r C_r C r . Refine to nested product neighborhoods. Choose continuous coefficient bumps η ℓ \eta_\ell η ℓ supported in the larger coefficient neighborhoods and smooth parameter bumps φ ℓ \varphi_\ell φ ℓ supported compactly in the larger parameter charts, positive on their smaller covering sets. Set
w ℓ ( a , p ) = η ℓ ( a ) φ ℓ ( p ) ( ∑ j η j ( a ) 2 φ j ( p ) 2 ) 1 / 2 . (404) w_\ell(a,p)=\frac{\eta_\ell(a)\varphi_\ell(p)}{\left(\sum_j\eta_j(a)^2\varphi_j(p)^2\right)^{1/2}}. \tag*{(404)} w ℓ ( a , p ) = ( ∑ j η j ( a ) 2 φ j ( p ) 2 ) 1/2 η ℓ ( a ) φ ℓ ( p ) . (404)
The denominator is bounded below on the compact set. Its parameter derivatives, and therefore those of every w ℓ w_\ell w ℓ , are uniformly bounded. The products w ℓ F ℓ w_\ell F_\ell w ℓ F ℓ and w ℓ F ℓ − 1 w_\ell F_\ell^{-1} w ℓ F ℓ − 1 extend by zero outside their charts because their supports lie strictly inside the larger chart domains. A weight depending only on direction and transverse position has zero derivative along the corresponding leaf. For finitely many transport systems, take the product of their compact coefficient closures with the same compact parameter set. Pull back each system’s cover and take all finite intersections, then choose nested product neighborhoods and the coefficient and parameter cutoffs in (B.12) on this common refinement. The refinement is finite, and its largest seminorm and smallest fixed margin are still prior-dependent constants. In particular it is fixed before selecting any polynomial test, width, or h h h . This proves all assertions. These oriented-pair weights are not asserted to descend to the projectivized direction space.
Free inverse and weighted estimates
The free inverse satisfies a small same-order bound and a bound gaining one derivative in weighted Sobolev spaces. Its full spatial output is retained for the variable-coefficient parametrix.
For the rest of the appendix use the local notation
D x = − i ∇ , A j = A ~ j / ( 2 i ) , V = − B ~ , τ = h − 1 , ξ c = i τ θ = τ ( ω 0 + i ν ) , ω 0 = − e 2 , ν = e 1 , P ξ c = ( D x + ξ c ) 2 I d s y s + 2 ∑ j A j ( D x , j + ξ c , j ) + V = − L h , θ . (405) \begin{aligned}
D_x&=-i\nabla,\qquad A_j=\widetilde{A}_j/(2i),\qquad V=-\widetilde{B},\qquad\tau=h^{-1},\\
\xi_c&=i\tau\theta=\tau(\omega_0+i\nu),\qquad\omega_0=-e_2,\qquad\nu=e_1,\\
P_{\xi_c}&=(D_x+\xi_c)^2 I_{d_{\mathrm{sys}}}+2\sum_j A_j(D_{x,j}+\xi_{c,j})+V=-L_{h,\theta}. \tag*{(405)}
\end{aligned} D x ξ c P ξ c = − i ∇ , A j = A j / ( 2 i ) , V = − B , τ = h − 1 , = i τ θ = τ ( ω 0 + i ν ) , ω 0 = − e 2 , ν = e 1 , = ( D x + ξ c ) 2 I d sys + 2 j ∑ A j ( D x , j + ξ c , j ) + V = − L h , θ . (405)
Here A = ( A 1 , A 2 , A 3 ) A=(A_1,A_2,A_3) A = ( A 1 , A 2 , A 3 ) is a triple of d s y s × d s y s d_{\mathrm{sys}}\times d_{\mathrm{sys}} d sys × d sys matrices, and V V V is a matrix potential.
For a Euclidean vector x x x , set
⟨ x ⟩ = ( 1 + ∣ x ∣ 2 ) 1 / 2 . (406) \langle x\rangle=(1+|x|^2)^{1/2}. \tag*{(406)} ⟨ x ⟩ = ( 1 + ∣ x ∣ 2 ) 1/2 . (406)
For an integer j ≥ 0 j\ge0 j ≥ 0 and a real number σ \sigma σ , define
∥ u ∥ H σ j 2 = ∑ ∣ α ∣ ≤ j ∫ R 3 ⟨ x ⟩ 2 σ ∣ ∂ α u ( x ) ∣ 2 d x . (407) \|u\|_{H^j_\sigma}^2=\sum_{|\alpha|\le j}\int_{\mathbb{R}^3}\langle x\rangle^{2\sigma}|\partial^\alpha u(x)|^2\,dx. \tag*{(407)} ∥ u ∥ H σ j 2 = ∣ α ∣ ≤ j ∑ ∫ R 3 ⟨ x ⟩ 2 σ ∣ ∂ α u ( x ) ∣ 2 d x . (407)
Here σ \sigma σ is only a weight exponent. The free operator and its inverse are
P ξ c 0 = ( D x + ξ c ) 2 , p ξ c ( ξ ) = ∣ ξ ∣ 2 + 2 ξ c ⋅ ξ , E ξ c = p ξ c ( D x ) − 1 . (408) P_{\xi_c}^0=(D_x+\xi_c)^2,\qquad p_{\xi_c}(\xi)=|\xi|^2+2\xi_c\cdot\xi,\qquad E_{\xi_c}=p_{\xi_c}(D_x)^{-1}. \tag*{(408)} P ξ c 0 = ( D x + ξ c ) 2 , p ξ c ( ξ ) = ∣ ξ ∣ 2 + 2 ξ c ⋅ ξ , E ξ c = p ξ c ( D x ) − 1 . (408)
The reciprocal is its locally integrable distribution, not a symbol obtained by differentiating across its characteristic set.
Lemma B.2. For τ ≥ 1 \tau\ge1 τ ≥ 1 and every integer j ≥ 0 j\ge0 j ≥ 0 , the multiplier in (B.16) has the two bounds
∥ E ξ c f ∥ H σ − 1 j ≤ C j τ − 1 ∥ f ∥ H σ j , ∥ E ξ c f ∥ H σ − 1 j + 1 ≤ C j ∥ f ∥ H σ j . (409) \|E_{\xi_c}f\|_{H^{j}_{\sigma-1}}\le C_j\tau^{-1}\|f\|_{H^j_\sigma},\qquad\|E_{\xi_c}f\|_{H^{j+1}_{\sigma-1}}\le C_j\|f\|_{H^j_\sigma}. \tag*{(409)} ∥ E ξ c f ∥ H σ − 1 j ≤ C j τ − 1 ∥ f ∥ H σ j , ∥ E ξ c f ∥ H σ − 1 j + 1 ≤ C j ∥ f ∥ H σ j . (409) It satisfies P ξ c 0 E ξ c f = f P_{\xi_c}^0E_{\xi_c}f=f P ξ c 0 E ξ c f = f in distributions. Constants are uniform over all oriented orthonormal pairs ( ω 0 , ν ) (\omega_0,\nu) ( ω 0 , ν ) . Its output is not assumed compactly supported.
Proof. Rotations reduce the proof to ω 0 = ( 1 , 0 , 0 ) \omega_0=(1,0,0) ω 0 = ( 1 , 0 , 0 ) and ν = ( 0 , 1 , 0 ) \nu=(0,1,0) ν = ( 0 , 1 , 0 ) . Write u = e − i τ x 1 v u=e^{-i\tau x_1}v u = e − i τ x 1 v , let t = x 2 t=x_2 t = x 2 , and take the Fourier transform in y = ( x 1 , x 3 ) y=(x_1,x_3) y = ( x 1 , x 3 ) . If η ∈ R 2 \eta\in\mathbb{R}^2 η ∈ R 2 and b = ∣ η ∣ b=|\eta| b = ∣ η ∣ , then
e i τ x 1 P ξ 0 e − i τ x 1 ⟷ − ∂ t 2 + 2 τ ∂ t + b 2 − τ 2 = − ( ∂ t − τ − b ) ( ∂ t − τ + b ) . (410) e^{i\tau x_1}P_\xi^0e^{-i\tau x_1}\ \longleftrightarrow\ -\partial_t^2+2\tau\partial_t+b^2-\tau^2=-(\partial_t-\tau-b)(\partial_t-\tau+b). \tag*{(410)} e i τ x 1 P ξ 0 e − i τ x 1 ⟷ − ∂ t 2 + 2 τ ∂ t + b 2 − τ 2 = − ( ∂ t − τ − b ) ( ∂ t − τ + b ) . (410)
For real a ≠ 0 a\ne0 a = 0 , put
G a ( t ) = { − 1 t < 0 e a t , a > 0 , 1 t > 0 e a t , a < 0 , K b = − G τ + b ∗ G τ − b . (411) G_a(t)=
\begin{cases}
-1_{t<0}e^{at}, & a>0,\\
1_{t>0}e^{at}, & a<0,
\end{cases}
\qquad K_b=-G_{\tau+b}*G_{\tau-b}. \tag*{(411)} G a ( t ) = { − 1 t < 0 e a t , 1 t > 0 e a t , a > 0 , a < 0 , K b = − G τ + b ∗ G τ − b . (411)
The jump at zero proves ( ∂ t − a ) G a = δ 0 (\partial_t-a)G_a=\delta_0 ( ∂ t − a ) G a = δ 0 . Since ∥ G τ + b ∥ L 1 = ( τ + b ) − 1 \lVert G_{\tau+b}\rVert_{L^1}=(\tau+b)^{-1} ∥ G τ + b ∥ L 1 = ( τ + b ) − 1 and ∥ G τ − b ∥ ∞ = 1 \lVert G_{\tau-b}\rVert_\infty=1 ∥ G τ − b ∥ ∞ = 1 , convolution gives
∣ K b ( t ) ∣ ≤ ( τ + b ) − 1 , b ∣ K b ( t ) ∣ ≤ 1 , ∣ K b ′ ( t ) ∣ ≤ 2. (412) |K_b(t)|\le(\tau+b)^{-1},\qquad b|K_b(t)|\le1,\qquad|K_b'(t)|\le2. \tag*{(412)} ∣ K b ( t ) ∣ ≤ ( τ + b ) − 1 , b ∣ K b ( t ) ∣ ≤ 1 , ∣ K b ′ ( t ) ∣ ≤ 2. (412)
For the last inequality, differentiate distributionally: K b ′ = − G τ + b + ( τ − b ) K b K_b'=-G_{\tau+b}+(\tau-b)K_b K b ′ = − G τ + b + ( τ − b ) K b and use ∣ τ − b ∣ ≤ τ + b |\tau-b|\le\tau+b ∣ τ − b ∣ ≤ τ + b . The convolution is continuous, so no additional Dirac mass occurs in this first derivative. The exceptional set b = τ b=\tau b = τ has measure zero in the partial Fourier variables.
If a convolution kernel H ( t − s ) H(t-s) H ( t − s ) is bounded by B B B , multiplication on both sides by ⟨ t ⟩ − 1 \langle t\rangle^{-1} ⟨ t ⟩ − 1 gives the Hilbert–Schmidt bound
∥ ⟨ t ⟩ − 1 H ∗ ⟨ t ⟩ − 1 ∥ L 2 → L 2 ≤ B ( ∫ R ⟨ t ⟩ − 2 d t ) = π B . (413) \left\|\langle t\rangle^{-1}H*\langle t\rangle^{-1}\right\|_{L^2\to L^2}\le B\left(\int_{\mathbb{R}}\langle t\rangle^{-2}\,dt\right)=\pi B. \tag*{(413)} ⟨ t ⟩ − 1 H ∗ ⟨ t ⟩ − 1 L 2 → L 2 ≤ B ( ∫ R ⟨ t ⟩ − 2 d t ) = π B . (413)
Apply it to K b K_b K b , η 1 K b \eta_1K_b η 1 K b , η 2 K b \eta_2K_b η 2 K b , and K b ′ K_b' K b ′ , and integrate in η \eta η by Plancherel. When we undo the modulation, the x 1 x_1 x 1 derivative acquires the term τ K b \tau K_b τ K b , which satisfies the same unit bound. Hence
∥ ⟨ t ⟩ − 1 u ∥ 2 ≤ C τ − 1 ∥ ⟨ t ⟩ f ∥ 2 , ∑ ∣ α ∣ ≤ 1 ∥ ⟨ t ⟩ − 1 ∂ α u ∥ 2 ≤ C ∥ ⟨ t ⟩ f ∥ 2 . (414) \|\langle t\rangle^{-1}u\|_2\le C\tau^{-1}\|\langle t\rangle f\|_2,\qquad
\sum_{|\alpha|\le1}\|\langle t\rangle^{-1}\partial^\alpha u\|_2\le C\|\langle t\rangle f\|_2. \tag*{(414)} ∥ ⟨ t ⟩ − 1 u ∥ 2 ≤ C τ − 1 ∥ ⟨ t ⟩ f ∥ 2 , ∣ α ∣ ≤ 1 ∑ ∥ ⟨ t ⟩ − 1 ∂ α u ∥ 2 ≤ C ∥ ⟨ t ⟩ f ∥ 2 . (414)
The inequalities ⟨ x ⟩ − 1 ≤ ⟨ t ⟩ − 1 \langle x\rangle^{-1}\le\langle t\rangle^{-1} ⟨ x ⟩ − 1 ≤ ⟨ t ⟩ − 1 and ⟨ t ⟩ ≤ ⟨ x ⟩ \langle t\rangle\le\langle x\rangle ⟨ t ⟩ ≤ ⟨ x ⟩ give the desired radial-weight bounds for j = 0 j=0 j = 0 .
For completeness the inverse just constructed is the specified multiplier. Fourier transformation of K b K_b K b in t t t gives, for b ≠ τ b\ne\tau b = τ ,
K b ^ ( s ) = − 1 ( i s − τ − b ) ( i s − τ + b ) = 1 s 2 + 2 i τ s + b 2 − τ 2 . (415) \widehat{K_b}(s)=-\frac{1}{(is-\tau-b)(is-\tau+b)}=\frac{1}{s^2+2i\tau s+b^2-\tau^2}. \tag*{(415)} K b ( s ) = − ( i s − τ − b ) ( i s − τ + b ) 1 = s 2 + 2 i τ s + b 2 − τ 2 1 . (415)
Near its zero set b = τ b=\tau b = τ , s = 0 s=0 s = 0 , the real variables ( b , s ) ↦ ( b 2 − τ 2 + s 2 , 2 τ s ) (b,s)\mapsto(b^2-\tau^2+s^2,2\tau s) ( b , s ) ↦ ( b 2 − τ 2 + s 2 , 2 τ s ) have nonzero Jacobian. The reciprocal is therefore locally integrable, like ( u + i v ) − 1 (u+iv)^{-1} ( u + i v ) − 1 in its two transverse variables. Fubini against Schwartz test functions identifies the partial Fourier inverse with this locally integrable reciprocal; the uniform kernel bound permits passage through b = τ b=\tau b = τ . No distribution supported on the characteristic set is added. Multiplication by p ξ c p_{\xi^c} p ξ c then proves the inverse identity.
The original multiplier commutes with every constant-coefficient spatial derivative. Apply the j = 0 j=0 j = 0 estimates to ∂ α f \partial^\alpha f ∂ α f , ∣ α ∣ ≤ j |\alpha|\le j ∣ α ∣ ≤ j , and sum. This proves both estimates at order j j j without differentiating the modulation of the input. Density in the weighted spaces completes the extension. Rotations preserve radial weights and give uniformly equivalent integer Sobolev norms, proving the directional uniformity.
Finite seminorm operator estimates
We use left quantization
Op ( a ) u ( x ) = ( 2 π ) − 3 ∫ R 3 e i x ⋅ ξ a ( x , ξ ) u ^ ( ξ ) d ξ , u ^ ( ξ ) = ∫ e − i x ⋅ ξ u ( x ) d x . (416) \operatorname{Op}(a)u(x)=(2\pi)^{-3}\int_{\mathbb{R}^3}e^{ix\cdot\xi}a(x,\xi)\widehat{u}(\xi)\,d\xi,\qquad
\widehat{u}(\xi)=\int e^{-ix\cdot\xi}u(x)\,dx. \tag*{(416)} Op ( a ) u ( x ) = ( 2 π ) − 3 ∫ R 3 e i x ⋅ ξ a ( x , ξ ) u ( ξ ) d ξ , u ( ξ ) = ∫ e − i x ⋅ ξ u ( x ) d x . (416)
We use the derivative order N N N in (3.9) for Lemma B.3. A common compact spatial support for a symbol is denoted by K a K_a K a .
The following estimates apply to the left-quantized symbols and their compositions, including weighted inputs.
Lemma B.3. Suppose a ( x , ξ ) a(x,\xi) a ( x , ξ ) has fixed compact x x x -support and ∣ ∂ x α a ∣ ≤ A |\partial_x^\alpha a| \le A ∣ ∂ x α a ∣ ≤ A for ∣ α ∣ ≤ N |\alpha| \le N ∣ α ∣ ≤ N , uniformly in ξ \xi ξ . Then, for 0 ≤ j ≤ k + 1 0 \le j \le k+1 0 ≤ j ≤ k + 1 ,
∥ Op ( a ) ∥ H j → H j ≤ C A . (417) \|\operatorname{Op}(a)\|_{H^j\to H^j} \le CA. \tag*{(417)} ∥ Op ( a ) ∥ H j → H j ≤ C A . (417) If c c c is likewise compactly supported in x x x with these derivatives bounded by B B B , and ∣ ∂ x α ∂ ξ β a ∣ ≤ A 1 |\partial_x^\alpha\partial_\xi^\beta a| \le A_1 ∣ ∂ x α ∂ ξ β a ∣ ≤ A 1 for ∣ α ∣ ≤ N |\alpha| \le N ∣ α ∣ ≤ N , then
∥ Op ( a ) Op ( c ) − Op ( a c ) ∥ H k → H k ≤ C A 1 B . (418) \|\operatorname{Op}(a)\operatorname{Op}(c)-\operatorname{Op}(ac)\|_{H^k\to H^k} \le CA_1B. \tag*{(418)} ∥ Op ( a ) Op ( c ) − Op ( a c ) ∥ H k → H k ≤ C A 1 B . (418) This also holds when a a a is a pure Fourier multiplier with ∣ ∇ ξ a ∣ ≤ A 1 |\nabla_\xi a| \le A_1 ∣ ∇ ξ a ∣ ≤ A 1 ; its unweighted operator norm is bounded by sup ∣ a ∣ \sup|a| sup ∣ a ∣ . Finally, if ∣ ∂ x α ∂ ξ β a ∣ ≤ A 2 |\partial_x^\alpha\partial_\xi^\beta a| \le A_2 ∣ ∂ x α ∂ ξ β a ∣ ≤ A 2 for ∣ α ∣ ≤ N |\alpha| \le N ∣ α ∣ ≤ N , ∣ β ∣ ≤ 2 |\beta| \le2 ∣ β ∣ ≤ 2 , then
∥ Op ( a ) u ∥ H j ≤ C A 2 ∥ u ∥ H j − 1 ( 0 ≤ j ≤ k + 1 ) . (419) \|\operatorname{Op}(a)u\|_{H^j} \le CA_2\|u\|_{H^{j-1}}\qquad(0 \le j \le k+1). \tag*{(419)} ∥ Op ( a ) u ∥ H j ≤ C A 2 ∥ u ∥ H j − 1 ( 0 ≤ j ≤ k + 1 ) . (419) The constants depend only on the orders, dimension and fixed spatial supports. In (B.27) no spatial truncation of u u u is performed.
Proof. Write a ^ ( ζ , ξ ) \widehat{a}(\zeta,\xi) a ( ζ , ξ ) for the partial Fourier transform in x x x . For ∣ ζ ∣ ≥ 1 |\zeta| \ge1 ∣ ζ ∣ ≥ 1 , integrate by parts N N N times in the unit direction ζ / ∣ ζ ∣ \zeta/|\zeta| ζ /∣ ζ ∣ . Compact support and the assumed derivatives give
∣ a ^ ( ζ , ξ ) ∣ ≤ C A ⟨ ζ ⟩ − N . (420) |\widehat{a}(\zeta,\xi)| \le CA\langle\zeta\rangle^{-N}. \tag*{(420)} ∣ a ( ζ , ξ ) ∣ ≤ C A ⟨ ζ ⟩ − N . (420)
This uses N N N derivatives even when N N N is odd; no even-order rounding is needed. The weighted Fourier kernel for the H j H^j H j estimate is bounded by C A ⟨ η − ξ ⟩ − N + j CA\langle\eta-\xi\rangle^{-N+j} C A ⟨ η − ξ ⟩ − N + j , 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 . Its integrals in either variable are finite when N > j + 3 N>j+3 N > j + 3 . Schur’s test proves (B.25).
Up to fixed Fourier normalization factors, the Fourier kernel of the composition difference is exactly
∫ [ a ^ ( η − ζ , ζ ) − a ^ ( η − ζ , ξ ) ] c ^ ( ζ − ξ , ξ ) d ζ . (421) \int[\widehat{a}(\eta-\zeta,\zeta)-\widehat{a}(\eta-\zeta,\xi)]\widehat{c}(\zeta-\xi,\xi)\,d\zeta. \tag*{(421)} ∫ [ a ( η − ζ , ζ ) − a ( η − ζ , ξ )] c ( ζ − ξ , ξ ) d ζ . (421)
The mean value theorem in the second slot, applied after x x x Fourier transformation, bounds it by
C A 1 B ∫ ∣ ζ − ξ ∣ ⟨ η − ζ ⟩ − N ⟨ ζ − ξ ⟩ − N d ζ . (422) CA_1B\int|\zeta-\xi|\langle\eta-\zeta\rangle^{-N}\langle\zeta-\xi\rangle^{-N}\,d\zeta. \tag*{(422)} C A 1 B ∫ ∣ ζ − ξ ∣ ⟨ η − ζ ⟩ − N ⟨ ζ − ξ ⟩ − N d ζ . (422)
Insert the Sobolev weight using ⟨ η ⟩ k / ⟨ ξ ⟩ k ≤ C ⟨ η − ζ ⟩ k ⟨ ζ − ξ ⟩ k \langle\eta\rangle^k/\langle\xi\rangle^k \le C\langle\eta-\zeta\rangle^k\langle\zeta-\xi\rangle^k ⟨ η ⟩ k / ⟨ ξ ⟩ k ≤ C ⟨ η − ζ ⟩ k ⟨ ζ − ξ ⟩ k . Both kernels are integrable since N > k + 4 N>k+4 N > k + 4 . The two Schur integrals are bounded by the product of their L 1 L^1 L 1 norms. For a pure left multiplier, the kernel is ( a ( η ) − a ( ξ ) ) c ^ ( η − ξ , ξ ) (a(\eta)-a(\xi))\widehat{c}(\eta-\xi,\xi) ( a ( η ) − a ( ξ )) c ( η − ξ , ξ ) ; the same estimate applies directly. This proves the composition assertion without commuting any matrix factors.
To prove the weighted assertion, direct integration by parts in ξ \xi ξ gives the exact identity
Op ( a ) ⟨ x ⟩ 2 = ⟨ x ⟩ 2 Op ( a ) − 2 i ∑ j = 1 3 x j Op ( ∂ ξ j a ) − Op ( Δ ξ a ) . (423) \operatorname{Op}(a)\langle x\rangle^2=\langle x\rangle^2\operatorname{Op}(a)-2i\sum_{j=1}^{3}x_j\operatorname{Op}(\partial_{\xi_j}a)-\operatorname{Op}(\Delta_\xi a). \tag*{(423)} Op ( a ) ⟨ x ⟩ 2 = ⟨ x ⟩ 2 Op ( a ) − 2 i j = 1 ∑ 3 x j Op ( ∂ ξ j a ) − Op ( Δ ξ a ) . (423)
Define
v = ⟨ x ⟩ − 2 u , (424) v=\langle x\rangle^{-2}u, \tag*{(424)} v = ⟨ x ⟩ − 2 u , (424)
using the weight in (B.14). The product rule gives ∥ v ∥ H j ≤ C j ∥ u ∥ H j − 1 \|v\|_{H^j}\le C_j\|u\|_{H^{j-1}} ∥ v ∥ H j ≤ C j ∥ u ∥ H j − 1 , since every derivative of ⟨ x ⟩ − 2 \langle x\rangle^{-2} ⟨ x ⟩ − 2 is bounded by C ⟨ x ⟩ − 1 C\langle x\rangle^{-1} C ⟨ x ⟩ − 1 . Apply the unweighted estimate to each term on the right of (B.31). All polynomial x x x factors are bounded, with all needed derivatives, on K a K_a K a ; each output there is compactly supported. This proves (B.27) first for compactly supported smooth u u u and then by weighted density. In particular the bound applies directly to u = E ξ c F u=E_{\xi_c}F u = E ξ c F :
∥ Op ( a ) E ξ c F ∥ H j ≤ C A 2 ∥ E ξ c F ∥ H j − 1 . (425) \|\operatorname{Op}(a)E_{\xi_c}F\|_{H^j}\le CA_2\|E_{\xi_c}F\|_{H^{j-1}}. \tag*{(425)} ∥ Op ( a ) E ξ c F ∥ H j ≤ C A 2 ∥ E ξ c F ∥ H j − 1 . (425)
The weighted identity applies to the full input u u u , including its tail outside K a K_a K a .
Fixed frame geometry and near/far symbols
The symbol construction separates a neighborhood of the characteristic set, where transport frames are used, from its complement, where division by the free symbol is permitted.
Choose once and for all smooth real spatial cutoffs satisfying
ψ 2 , ψ 3 , χ s p ∈ C ∞ ( Ω ‾ 0 ) , 0 ≤ ψ 2 , ψ 3 , χ s p ≤ 1 , \psi_2,\psi_3,\chi_{\mathrm{sp}} \in C^\infty(\overline{\Omega}_0),\qquad0\leq\psi_2,\psi_3,\chi_{\mathrm{sp}}\leq1, ψ 2 , ψ 3 , χ sp ∈ C ∞ ( Ω 0 ) , 0 ≤ ψ 2 , ψ 3 , χ sp ≤ 1 ,
ψ 2 = 1 near Ω ‾ 1 , ψ 3 = 1 near supp ψ 2 , χ s p = 1 near supp ψ 3 . (426) \psi_2=1\ \text{near }\overline{\Omega}_1,\qquad\psi_3=1\ \text{near }\operatorname{supp}\psi_2,\qquad\chi_{\mathrm{sp}}=1\ \text{near }\operatorname{supp}\psi_3. \tag*{(426)} ψ 2 = 1 near Ω 1 , ψ 3 = 1 near supp ψ 2 , χ sp = 1 near supp ψ 3 . (426)
They are chosen before h h h , coefficients, amplitudes or test parameters.
Compact coefficient families
The entire coefficient ball in (81) , with fixed support, has compact closure in C k + 12 , 1 / 2 C^{k+12,1/2} C k + 12 , 1/2 on a fixed large ball. Indeed derivatives through order k + 13 k+13 k + 13 are uniformly Lipschitz, and Arzelà–Ascoli followed by Hölder interpolation gives the asserted compactness at the lower order. Limits remain supported in K c o e f K_{\mathrm{coef}} K coef . They need not be C k + 14 C^{k+14} C k + 14 , so existence is used at their actual finite regularity.
Use r p a r r_{\mathrm{par}} r par from (89) . For an oriented pair ( ω , ν ) (\omega,\nu) ( ω , ν ) set
n = ω × ν . (427) n=\omega\times\nu. \tag*{(427)} n = ω × ν . (427)
Use the leaf coordinates
x = ( ℜ z ) ω + ( ℑ z ) ν + t n , z = x ⋅ ω + i x ⋅ ν , t = x ⋅ n . (428) x=(\Re z)\omega+(\Im z)\nu+tn,\qquad z=x\cdot\omega+ix\cdot\nu,\qquad t=x\cdot n. \tag*{(428)} x = ( ℜ z ) ω + ( ℑ z ) ν + t n , z = x ⋅ ω + i x ⋅ ν , t = x ⋅ n . (428)
The parameter p = ( ω , ν , t ) p=(\omega,\nu,t) p = ( ω , ν , t ) belongs to the compact oriented-pair manifold times a fixed closed interval containing all transverse coordinates of supp χ s p \operatorname{supp}\chi_{\mathrm{sp}} supp χ sp , with a fixed margin. Take fixed large leaf disks containing all its leaf projections, also with fixed margins. Pulled-back coefficients have joint C r p a r + 2 C^{r_{\mathrm{par}}+2} C r par + 2 bounds there, including at the C k + 12 , 1 / 2 C^{k+12,1/2} C k + 12 , 1/2 closure points. Apply Theorem 3.1 to the matrix coefficient
A ω + i ν = ∑ j ( ω j + i ν j ) A j , B = A ω + i ν / ( 2 i ) . (429) A_{\omega+i\nu}=\sum_j(\omega_j+i\nu_j)A_j,\qquad B=A_{\omega+i\nu}/(2i). \tag*{(429)} A ω + i ν = j ∑ ( ω j + i ν j ) A j , B = A ω + i ν / ( 2 i ) . (429)
This produces a finite atlas F ℓ F_\ell F ℓ and weights w ℓ w_\ell w ℓ , 1 ≤ ℓ ≤ J 1\leq\ell\leq J 1 ≤ ℓ ≤ J , such that
i ( ω + i ν ) ⋅ ∇ F ℓ = A ω + i ν F ℓ , ( ω + i ν ) ⋅ ∇ w ℓ = 0 , ∑ ℓ = 1 J w ℓ 2 = 1. (430) i(\omega+i\nu)\cdot\nabla F_\ell=A_{\omega+i\nu}F_\ell,\qquad(\omega+i\nu)\cdot\nabla w_\ell=0,\qquad\sum_{\ell=1}^{J}w_\ell^2=1. \tag*{(430)} i ( ω + i ν ) ⋅ ∇ F ℓ = A ω + i ν F ℓ , ( ω + i ν ) ⋅ ∇ w ℓ = 0 , ℓ = 1 ∑ J w ℓ 2 = 1. (430)
Every needed derivative of the frames, inverses, weights and coordinates through total order r p a r r_{\mathrm{par}} r par is uniformly bounded. For a particular coefficient, charts with zero weight contribute zero. Products are extended by zero through inactive chart boundaries. The same finite atlas is fixed over the entire prior; it is not reselected as h h h changes.
Frequency geometry
Fix a smooth ϑ : R → [ 0 , 1 ] \vartheta:\mathbb{R}\to[0,1] ϑ : R → [ 0 , 1 ] equal to one on [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] and zero outside [ − 2 , 2 ] [-2,2] [ − 2 , 2 ] . For ξ ∈ R 3 \xi\in\mathbb{R}^3 ξ ∈ R 3 let
ξ ν = ξ ⋅ ν , ξ ′ = ξ − ξ ν ν , r = ∣ ξ ′ + τ ω 0 ∣ , ω = ξ ′ + τ ω 0 r , \xi_\nu=\xi\cdot\nu,\qquad\xi'=\xi-\xi_\nu\nu,\qquad r=\lvert\xi'+\tau\omega_0\rvert,\qquad\omega=\frac{\xi'+\tau\omega_0}{r}, ξ ν = ξ ⋅ ν , ξ ′ = ξ − ξ ν ν , r = ∣ ξ ′ + τ ω 0 ∣ , ω = r ξ ′ + τ ω 0 ,
t τ = ϑ ( ξ ν τ ) ϑ ( r 2 − τ 2 τ 3 / 2 ) , χ N = sin ( π t τ / 2 ) , χ F = cos ( π t τ / 2 ) . (431) t_\tau=\vartheta\left(\frac{\xi_\nu}{\sqrt{\tau}}\right)\vartheta\left(\frac{r^2-\tau^2}{\tau^{3/2}}\right),\qquad\chi_N=\sin(\pi t_\tau/2),\qquad\chi_F=\cos(\pi t_\tau/2). \tag*{(431)} t τ = ϑ ( τ ξ ν ) ϑ ( τ 3/2 r 2 − τ 2 ) , χ N = sin ( π t τ /2 ) , χ F = cos ( π t τ /2 ) . (431)
The scalar r r r here is a frequency radius, not the order in the frame theorem; t τ t_\tau t τ is a frequency cutoff, not the transverse coordinate t t t of (B.36). The products in (B.42) vanish near r = 0 r=0 r = 0 , so ω \omega ω is only used where it is defined. Set
Ξ N = supp χ N , Ξ F = supp χ F . (432) \Xi_N=\operatorname{supp}\chi_N,\qquad\Xi_F=\operatorname{supp}\chi_F. \tag*{(432)} Ξ N = supp χ N , Ξ F = supp χ F . (432)
For τ ≥ 16 \tau\ge16 τ ≥ 16 , on Ξ N \Xi_N Ξ N ,
∣ ξ ℓ ∣ ≤ 2 τ , ∣ r 2 − τ 2 ∣ ≤ 2 τ 3 / 2 , τ / 2 ≤ r ≤ 3 / 2 τ . (433) |\xi_\ell| \le2\sqrt{\tau}, \qquad|r^2-\tau^2| \le2\tau^{3/2}, \qquad\tau/\sqrt{2} \le r \le\sqrt{3/2}\tau. \tag*{(433)} ∣ ξ ℓ ∣ ≤ 2 τ , ∣ r 2 − τ 2 ∣ ≤ 2 τ 3/2 , τ / 2 ≤ r ≤ 3/2 τ . (433)
The characteristic circle is { ξ ℓ = 0 , r = τ } \{\xi_\ell=0,\ r=\tau\} { ξ ℓ = 0 , r = τ } ; a full neighborhood of it has χ F = 0 \chi_F=0 χ F = 0 . Since χ N 2 + χ F 2 = 1 \chi_N^2+\chi_F^2=1 χ N 2 + χ F 2 = 1 , define the paired matrix symbols
b ℓ ( x , ξ ) = χ s p ( x ) χ N ( ξ ) w ℓ ( a , p ) F ℓ ( z , p ) , b_\ell(x,\xi)=\chi_{\mathrm{sp}}(x)\chi_N(\xi)w_\ell(a,p)F_\ell(z,p), b ℓ ( x , ξ ) = χ sp ( x ) χ N ( ξ ) w ℓ ( a , p ) F ℓ ( z , p ) ,
d ℓ ( x , ξ ) = χ s p ( x ) χ N ( ξ ) w ℓ ( a , p ) F ℓ ( z , p ) − 1 , 1 ≤ ℓ ≤ J , d_\ell(x,\xi)=\chi_{\mathrm{sp}}(x)\chi_N(\xi)w_\ell(a,p)F_\ell(z,p)^{-1}, \qquad1\le\ell\le J, d ℓ ( x , ξ ) = χ sp ( x ) χ N ( ξ ) w ℓ ( a , p ) F ℓ ( z , p ) − 1 , 1 ≤ ℓ ≤ J ,
b 0 ( ξ ) = d 0 ( ξ ) = χ F ( ξ ) I d s y s . (434) b_0(\xi)=d_0(\xi)=\chi_F(\xi)I_{\mathrm{dsys}}. \tag*{(434)} b 0 ( ξ ) = d 0 ( ξ ) = χ F ( ξ ) I dsys . (434)
Here d ℓ d_\ell d ℓ denotes the paired matrix symbol. The leading identity, with matrix order preserved, is
∑ ℓ = 0 J b ℓ ψ 3 d ℓ = ψ 3 I d s y s on all R x 3 × R ξ 3 . (435) \sum_{\ell=0}^{J}b_\ell\psi_3d_\ell=\psi_3I_{\mathrm{dsys}} \quad\text{on all }\mathbb{R}_x^3\times\mathbb{R}_\xi^3. \tag*{(435)} ℓ = 0 ∑ J b ℓ ψ 3 d ℓ = ψ 3 I dsys on all R x 3 × R ξ 3 . (435)
Lemma B.4. For τ ≥ 16 \tau\ge16 τ ≥ 16 the following bounds have constants fixed by the prior and cutoffs. First, for ∣ α ∣ ≤ N + 2 |\alpha|\le N+2 ∣ α ∣ ≤ N + 2 and ∣ β ∣ ≤ 2 |\beta|\le2 ∣ β ∣ ≤ 2 ,
∣ ∂ x α ∂ ξ β b ℓ ∣ + ∣ ∂ x α ∂ ξ β d ℓ ∣ ≤ C τ − ∣ β ∣ / 2 , ℓ ≥ 1. (436) |\partial_x^\alpha\partial_\xi^\beta b_\ell|+|\partial_x^\alpha\partial_\xi^\beta d_\ell|\le C\tau^{-|\beta|/2}, \qquad\ell\ge1. \tag*{(436)} ∣ ∂ x α ∂ ξ β b ℓ ∣ + ∣ ∂ x α ∂ ξ β d ℓ ∣ ≤ C τ − ∣ β ∣/2 , ℓ ≥ 1. (436) Their supports in ( x , ξ ) (x,\xi) ( x , ξ ) lie in supp χ s p × Ξ N \operatorname{supp}\chi_{\mathrm{sp}}\times\Xi_N supp χ sp × Ξ N and the corresponding parameter charts. The pure multipliers b 0 , d 0 b_0,d_0 b 0 , d 0 obey the frequency bounds in (B.44), but have no compact spatial support.
Define the differential remainder symbols by
r 0 , ℓ = − Δ x b ℓ − 2 i ∑ j A j ∂ x j b ℓ + V b ℓ , r_{0,\ell}=-\Delta_xb_\ell-2i\sum_j A_j\partial_{x_j}b_\ell+Vb_\ell, r 0 , ℓ = − Δ x b ℓ − 2 i j ∑ A j ∂ x j b ℓ + V b ℓ , r 1 , ℓ = 2 ∑ j ( ξ j + ξ c , j ) ( A j b ℓ − i ∂ x j b ℓ ) . (437) r_{1,\ell}=2\sum_j(\xi_j+\xi_{c,j})(A_jb_\ell-i\partial_{x_j}b_\ell). \tag*{(437)} r 1 , ℓ = 2 j ∑ ( ξ j + ξ c , j ) ( A j b ℓ − i ∂ x j b ℓ ) . (437) For ∣ α ∣ ≤ N |\alpha|\le N ∣ α ∣ ≤ N , ∣ β ∣ ≤ 2 |\beta|\le2 ∣ β ∣ ≤ 2 , they satisfy
∣ ∂ x α ∂ ξ β ( ψ 3 r 0 , ℓ ) ∣ ≤ C τ − ∣ β ∣ / 2 , ℓ ≥ 1 , |\partial_x^\alpha\partial_\xi^\beta(\psi_3r_{0,\ell})|\le C\tau^{-|\beta|/2}, \qquad\ell\ge1, ∣ ∂ x α ∂ ξ β ( ψ 3 r 0 , ℓ ) ∣ ≤ C τ − ∣ β ∣/2 , ℓ ≥ 1 , ∣ ∂ x α ∂ ξ β ( ψ 3 r 1 , ℓ ) ∣ ≤ C τ ( 1 − ∣ β ∣ ) / 2 , ℓ ≥ 1 , |\partial_x^\alpha\partial_\xi^\beta(\psi_3r_{1,\ell})|\le C\tau^{(1-|\beta|)/2}, \qquad\ell\ge1, ∣ ∂ x α ∂ ξ β ( ψ 3 r 1 , ℓ ) ∣ ≤ C τ ( 1 − ∣ β ∣ ) /2 , ℓ ≥ 1 , ∣ ∂ x α ∂ ξ β ( ψ 3 V χ F ) ∣ ≤ C τ − ∣ β ∣ / 2 , |\partial_x^\alpha\partial_\xi^\beta(\psi_3V\chi_F)|\le C\tau^{-|\beta|/2}, ∣ ∂ x α ∂ ξ β ( ψ 3 V χ F ) ∣ ≤ C τ − ∣ β ∣/2 , ∣ ∂ x α ∂ ξ β a F ∣ ≤ C τ − ( 1 + ∣ β ∣ ) / 2 . (438) |\partial_x^\alpha\partial_\xi^\beta a_F|\le C\tau^{-(1+|\beta|)/2}. \tag*{(438)} ∣ ∂ x α ∂ ξ β a F ∣ ≤ C τ − ( 1 + ∣ β ∣ ) /2 . (438) where
a F ( x , ξ ) = 2 ∑ j A j ( x ) ( ξ j + ξ c , j ) χ F ( ξ ) p ξ c ( ξ ) (439) a_F(x,\xi)=\frac{2\sum_j A_j(x)(\xi_j+\xi_{c,j})\chi_F(\xi)}{p_{\xi_c}(\xi)} \tag*{(439)} a F ( x , ξ ) = p ξ c ( ξ ) 2 ∑ j A j ( x ) ( ξ j + ξ c , j ) χ F ( ξ ) (439) is zero in a neighborhood of the characteristic set. The two near symbols in (B.46) are supported in supp ψ 3 × Ξ N \operatorname{supp}\psi_3\times\Xi_N supp ψ 3 × Ξ N ; the two far symbols are supported in K c o e f × Ξ F K_{\mathrm{coef}}\times\Xi_F K coef × Ξ F . The maximum frame order used is N + 4 N+4 N + 4 , and the maximum coefficient x x x order in the remainder products is N N N .
Proof. On Ξ N \Xi_N Ξ N , direct differentiation gives ∂ ξ r = O ( 1 ) \partial_\xi r=O(1) ∂ ξ r = O ( 1 ) , ∂ ξ 2 r = O ( τ − 1 ) \partial_\xi^2r=O(\tau^{-1}) ∂ ξ 2 r = O ( τ − 1 ) , ∂ ξ ω = O ( τ − 1 ) \partial_\xi\omega=O(\tau^{-1}) ∂ ξ ω = O ( τ − 1 ) , and ∂ ξ 2 ω = O ( τ − 2 ) \partial_\xi^2\omega=O(\tau^{-2}) ∂ ξ 2 ω = O ( τ − 2 ) . Where a derivative of either cutoff factor is nonzero, the first derivatives of its scaled argument are O ( τ − 1 / 2 ) O(\tau^{-1/2}) O ( τ − 1/2 ) ; its second derivatives are O ( τ − 3 / 2 ) O(\tau^{-3/2}) O ( τ − 3/2 ) or zero. Outside these transition regions the corresponding cutoff derivatives vanish. The chain rule therefore bounds the first and second derivatives of χ N , χ F \chi_N,\chi_F χ N , χ F by C τ − 1 / 2 C\tau^{-1/2} C τ − 1/2 and C τ − 1 C\tau^{-1} C τ − 1 . On the compact spatial support, (B.36) shows that one ξ \xi ξ derivative of a pulled-back frame or weight costs at most one additional leaf/parameter derivative and a factor O ( τ − 1 ) O(\tau^{-1}) O ( τ − 1 ) ; two cost at most two and a factor O ( τ − 2 ) O(\tau^{-2}) O ( τ − 2 ) . Thus a mixed derivative with ∣ α ∣ ≤ N + 2 |\alpha|\le N+2 ∣ α ∣ ≤ N + 2 , ∣ β ∣ ≤ 2 |\beta|\le2 ∣ β ∣ ≤ 2 costs at most N + 4 N+4 N + 4 total frame derivatives. Leibniz’ rule, allowing derivatives to fall on the wider frequency cutoffs, gives (B.44). All assertions extend through r = 0 r=0 r = 0 by the zero products.
The r 0 , ℓ r_{0,\ell} r 0 , ℓ formula costs two additional x x x derivatives of the base frame product, so its maximal mixed input order is N + 2 + 2 = N + 4 N+2+2=N+4 N + 2 + 2 = N + 4 . Derivatives falling on A j A_j A j or V V V have order at most N N N . This proves the first bound in (B.46). On a neighborhood of supp ψ 3 \operatorname{supp}\psi_3 supp ψ 3 the spatial cutoff is constant. Equation (B.38) then gives the exact ordered cancellation
ψ 3 r 1 , ℓ = 2 ψ 3 ( ( r − τ ) ω + ξ ν ν ) ⋅ ( A b ℓ − i ∇ x b ℓ ) , ℓ ≥ 1. (440) \psi_3r_{1,\ell}=2\psi_3\bigl((r-\tau)\omega+\xi_\nu\nu\bigr)\cdot(A b_\ell-i\nabla_xb_\ell),\qquad\ell\ge1. \tag*{(440)} ψ 3 r 1 , ℓ = 2 ψ 3 ( ( r − τ ) ω + ξ ν ν ) ⋅ ( A b ℓ − i ∇ x b ℓ ) , ℓ ≥ 1. (440)
The vector in parentheses and its first and second frequency derivatives have orders O ( τ ) O(\sqrt{\tau}) O ( τ ) , O ( 1 ) O(1) O ( 1 ) and O ( τ − 1 ) O(\tau^{-1}) O ( τ − 1 ) , respectively. Combining these frequency-derivative estimates with (B.44), we obtain the second estimate in (B.46). Its maximum frame order is only N + 1 + 2 = N + 3 N+1+2=N+3 N + 1 + 2 = N + 3 . There are no cutoff-gradient terms from χ s p \chi_{\mathrm{sp}} χ sp on this support.
For ℓ = 0 \ell=0 ℓ = 0 the spatial derivatives of b 0 b_0 b 0 vanish and r 0 , 0 = V χ F r_{0,0}=V\chi_F r 0 , 0 = V χ F . This proves the third bound. To treat r 1 , 0 r_{1,0} r 1 , 0 , write
p ξ c = r 2 + ξ ν 2 − τ 2 + 2 i τ ξ ν . (441) p_{\xi_c}=r^2+\xi_\nu^2-\tau^2+2i\tau\xi_\nu. \tag*{(441)} p ξ c = r 2 + ξ ν 2 − τ 2 + 2 i τ ξ ν . (441)
On Ξ F \Xi_F Ξ F , either ∣ ξ ν ∣ ≥ τ |\xi_\nu|\ge\sqrt{\tau} ∣ ξ ν ∣ ≥ τ and the imaginary part has modulus at least 2 τ 3 / 2 2\tau^{3/2} 2 τ 3/2 , or ∣ r 2 − τ 2 ∣ ≥ τ 3 / 2 |r^2-\tau^2|\ge\tau^{3/2} ∣ r 2 − τ 2 ∣ ≥ τ 3/2 with ∣ ξ ν ∣ ≤ τ |\xi_\nu|\le\sqrt{\tau} ∣ ξ ν ∣ ≤ τ . In the second case the real part has modulus at least τ 3 / 2 − τ ≥ ( 3 / 4 ) τ 3 / 2 \tau^{3/2}-\tau\ge(3/4)\tau^{3/2} τ 3/2 − τ ≥ ( 3/4 ) τ 3/2 . For ∣ ξ ∣ ≤ 4 τ |\xi|\le4\tau ∣ ξ ∣ ≤ 4 τ we also have ∣ ∂ ξ p ∣ ≤ C τ |\partial_\xi p|\le C\tau ∣ ∂ ξ p ∣ ≤ C τ and ∂ ξ 2 p = 2 I \partial_\xi^2p=2I ∂ ξ 2 p = 2 I . The formulae
∂ l ( p − 1 ) = − p − 2 ∂ l p , ∂ l ∂ j ( p − 1 ) = 2 p − 3 ( ∂ l p ) ( ∂ j p ) − p − 2 ∂ l ∂ j p (442) \partial_l(p^{-1})=-p^{-2}\partial_lp,\qquad\partial_l\partial_j(p^{-1})=2p^{-3}(\partial_lp)(\partial_jp)-p^{-2}\partial_l\partial_jp \tag*{(442)} ∂ l ( p − 1 ) = − p − 2 ∂ l p , ∂ l ∂ j ( p − 1 ) = 2 p − 3 ( ∂ l p ) ( ∂ j p ) − p − 2 ∂ l ∂ j p (442)
give the first two derivatives of the reciprocal on this far set. The uncut numerator of (B.47) has frequency orders O ( τ ) O(\tau) O ( τ ) , O ( 1 ) O(1) O ( 1 ) , and zero. Applying the product rule also to χ F \chi_F χ F , we obtain the factor τ − ( 1 + ∣ β ∣ ) / 2 \tau^{-(1+|\beta|)/2} τ − ( 1 + ∣ β ∣ ) /2 for ∣ β ∣ ≤ 2 |\beta|\le2 ∣ β ∣ ≤ 2 .
For ∣ ξ ∣ ≥ 4 τ |\xi|\ge4\tau ∣ ξ ∣ ≥ 4 τ , we have ∣ p ∣ ≥ ∣ ξ ∣ 2 / 2 |p|\ge|\xi|^2/2 ∣ p ∣ ≥ ∣ ξ ∣ 2 /2 and ∣ ∂ ξ p ∣ ≤ C ∣ ξ ∣ |\partial_\xi p|\le C|\xi| ∣ ∂ ξ p ∣ ≤ C ∣ ξ ∣ . The near support in (B.41) lies inside ∣ ξ ∣ < 4 τ |\xi|<4\tau ∣ ξ ∣ < 4 τ , so χ F = 1 \chi_F=1 χ F = 1 in this exterior region. Differentiation gives C ∣ ξ ∣ − 1 − ∣ β ∣ C|\xi|^{-1-|\beta|} C ∣ ξ ∣ − 1 − ∣ β ∣ , which is stronger than the displayed bounds. The cutoff vanishes on an open neighborhood of the poles, so these reciprocal derivatives never cross the characteristic circle. All spatial derivatives cost only the indicated coefficient derivatives, and support in K c o e f K_{\mathrm{coef}} K coef follows from the factor A A A .
Ordered parametrix errors and two Sobolev bounds
The parametrix error is decomposed into five terms with the matrix factors kept in their stated order. Both Sobolev bounds are proved for this parametrix before the Neumann correction.
Define, initially on Schwartz functions and then by the bounds in Lemma B.5 ,
Q τ = ∑ ℓ = 0 J Op ( b ℓ ) E ξ c ψ 3 Op ( d ℓ ) . (443) Q_\tau=\sum_{\ell=0}^{J}\operatorname{Op}(b_\ell)E_{\xi_c}\psi_3\operatorname{Op}(d_\ell). \tag*{(443)} Q τ = ℓ = 0 ∑ J Op ( b ℓ ) E ξ c ψ 3 Op ( d ℓ ) . (443)
The full output of E ξ c E_{\xi_c} E ξ c is used by Op ( b ℓ ) \operatorname{Op}(b_\ell) Op ( b ℓ ) . The inner source F ℓ = ψ 3 Op ( d ℓ ) g F_\ell=\psi_3\operatorname{Op}(d_\ell)g F ℓ = ψ 3 Op ( d ℓ ) g is supported in supp ψ 3 \operatorname{supp}\psi_3 supp ψ 3 , and
∥ F ℓ ∥ H 1 k ≤ C ∥ g ∥ H k . (444) \|F_\ell\|_{H_1^k}\le C\|g\|_{H^k}. \tag*{(444)} ∥ F ℓ ∥ H 1 k ≤ C ∥ g ∥ H k . (444)
For ℓ = 0 \ell=0 ℓ = 0 this uses the multiplier bound before multiplication by ψ 3 \psi_3 ψ 3 . For ℓ ≥ 1 \ell\ge1 ℓ ≥ 1 it uses the compact-symbol bound. Compact spatial support then bounds the weight. Notice that F ℓ F_\ell F ℓ need not be frequency-supported in Ξ F \Xi_F Ξ F or Ξ ‾ F \overline{\Xi}_F Ξ F , because multiplication by ψ 3 \psi_3 ψ 3 mixes frequencies.
A direct product-rule differentiation in left quantization gives
P ξ c Op ( b ℓ ) E ξ c = Op ( b ℓ ) + Op ( r 0 , ℓ ) E ξ c + Op ( r 1 , ℓ ) E ξ c . (445) P_{\xi_c}\operatorname{Op}(b_\ell)E_{\xi_c}=\operatorname{Op}(b_\ell)+\operatorname{Op}(r_{0,\ell})E_{\xi_c}+\operatorname{Op}(r_{1,\ell})E_{\xi_c}. \tag*{(445)} P ξ c Op ( b ℓ ) E ξ c = Op ( b ℓ ) + Op ( r 0 , ℓ ) E ξ c + Op ( r 1 , ℓ ) E ξ c . (445)
For instance D x , j ( b e i x ⋅ ξ ) = ( ξ j b − i ∂ x j b ) e i x ⋅ ξ D_{x,j}\left(b e^{ix\cdot\xi}\right)=(\xi_j b-i\partial_{x_j}b)e^{ix\cdot\xi} D x , j ( b e i x ⋅ ξ ) = ( ξ j b − i ∂ x j b ) e i x ⋅ ξ ; applying it twice produces precisely (B.45). There is no derivative of A j A_j A j here because it multiplies on the left in the nondivergence operator. Write
S τ = ∑ ℓ = 0 J Op ( b ℓ ) ψ 3 Op ( d ℓ ) − ψ 3 I , S_\tau=\sum_{\ell=0}^{J}\operatorname{Op}(b_\ell)\psi_3\operatorname{Op}(d_\ell)-\psi_3I, S τ = ℓ = 0 ∑ J Op ( b ℓ ) ψ 3 Op ( d ℓ ) − ψ 3 I ,
E τ = ∑ ℓ = 0 J ∑ j = 0 1 Op ( r j , ℓ ) E ξ c ψ 3 Op ( d ℓ ) , \mathcal{E}_\tau=\sum_{\ell=0}^{J}\sum_{j=0}^{1}\operatorname{Op}(r_{j,\ell})E_{\xi_c}\psi_3\operatorname{Op}(d_\ell), E τ = ℓ = 0 ∑ J j = 0 ∑ 1 Op ( r j , ℓ ) E ξ c ψ 3 Op ( d ℓ ) ,
P ξ c Q τ = ψ 3 I + S τ + E τ . (446) P_{\xi_c}Q_\tau=\psi_3I+S_\tau+\mathcal{E}_\tau. \tag*{(446)} P ξ c Q τ = ψ 3 I + S τ + E τ . (446)
The last identity is distributional. Its localization is bounded on H k H^k H k by the following five estimates.
Lemma B.5. Let
T c o m p , τ = ψ 3 S τ , T_{\mathrm{comp},\tau}=\psi_3S_\tau, T comp , τ = ψ 3 S τ , T n e a r , j , τ = ∑ ℓ = 1 J Op ( ψ 3 r j , ℓ ) E ξ c ψ 3 Op ( d ℓ ) , j = 0 , 1 , T_{\mathrm{near},j,\tau}=\sum_{\ell=1}^{J}\operatorname{Op}(\psi_3r_{j,\ell})E_{\xi_c}\psi_3\operatorname{Op}(d_\ell),\qquad j=0,1, T near , j , τ = ℓ = 1 ∑ J Op ( ψ 3 r j , ℓ ) E ξ c ψ 3 Op ( d ℓ ) , j = 0 , 1 , T f a r , j , τ = Op ( ψ 3 r j , 0 ) E ξ c ψ 3 Op ( d 0 ) , j = 0 , 1 , T_{\mathrm{far},j,\tau}=\operatorname{Op}(\psi_3r_{j,0})E_{\xi_c}\psi_3\operatorname{Op}(d_0),\qquad j=0,1, T far , j , τ = Op ( ψ 3 r j , 0 ) E ξ c ψ 3 Op ( d 0 ) , j = 0 , 1 , T τ = T c o m p , τ + T n e a r , 0 , τ + T n e a r , 1 , τ + T f a r , 0 , τ + T f a r , 1 , τ = ψ 3 ( S τ + E τ ) . (447) T_\tau=T_{\mathrm{comp},\tau}+T_{\mathrm{near},0,\tau}+T_{\mathrm{near},1,\tau}+T_{\mathrm{far},0,\tau}+T_{\mathrm{far},1,\tau}=\psi_3(S_\tau+\mathcal{E}_\tau). \tag*{(447)} T τ = T comp , τ + T near , 0 , τ + T near , 1 , τ + T far , 0 , τ + T far , 1 , τ = ψ 3 ( S τ + E τ ) . (447) There are five common constants C j C_j C j , independent of the coefficient, normalized direction and τ ≥ 16 \tau\geq16 τ ≥ 16 , such that
∥ T c o m p , τ ∥ H k → H k ≤ C 1 τ − 1 / 2 ∥ T n e a r , 0 , τ ∥ H k → H k ≤ C 2 τ − 1 ∥ T n e a r , 1 , τ ∥ H k → H k ≤ C 3 τ − 1 / 2 ∥ T f a r , 0 , τ ∥ H k → H k ≤ C 4 τ − 1 ∥ T f a r , 1 , τ ∥ H k → H k ≤ C 5 τ − 1 / 2 . (448) \begin{aligned}
\|T_{\mathrm{comp},\tau}\|_{H^k\to H^k}&\leq C_1\tau^{-1/2}\\
\|T_{\mathrm{near},0,\tau}\|_{H^k\to H^k}&\leq C_2\tau^{-1}\\
\|T_{\mathrm{near},1,\tau}\|_{H^k\to H^k}&\leq C_3\tau^{-1/2}\\
\|T_{\mathrm{far},0,\tau}\|_{H^k\to H^k}&\leq C_4\tau^{-1}\\
\|T_{\mathrm{far},1,\tau}\|_{H^k\to H^k}&\leq C_5\tau^{-1/2}. \tag*{(448)}
\end{aligned} ∥ T comp , τ ∥ H k → H k ∥ T near , 0 , τ ∥ H k → H k ∥ T near , 1 , τ ∥ H k → H k ∥ T far , 0 , τ ∥ H k → H k ∥ T far , 1 , τ ∥ H k → H k ≤ C 1 τ − 1/2 ≤ C 2 τ − 1 ≤ C 3 τ − 1/2 ≤ C 4 τ − 1 ≤ C 5 τ − 1/2 . (448) Every error has output supported in supp ψ 3 \operatorname{supp}\psi_3 supp ψ 3 ; the two far errors in fact have output in K c o e f K_{\mathrm{coef}} K coef . The symbol frequency supports are those of Lemma B.4 ; no preservation of the Fourier support of an input or output function is asserted.
Proof. Composition. For each ℓ \ell ℓ , use the compact right symbol c ℓ = ψ 3 d ℓ c_\ell=\psi_3d_\ell c ℓ = ψ 3 d ℓ , since ψ 3 Op ( d ℓ ) = Op ( c ℓ ) \psi_3\operatorname{Op}(d_\ell)=\operatorname{Op}(c_\ell) ψ 3 Op ( d ℓ ) = Op ( c ℓ ) exactly. For ℓ ≥ 1 \ell\geq1 ℓ ≥ 1 , the left-symbol first frequency derivative, with N N N spatial derivatives, has size C τ − 1 / 2 C\tau^{-1/2} C τ − 1/2 . For ℓ = 0 \ell=0 ℓ = 0 apply the pure-multiplier version of Lemma B.3 . The principal products sum to ψ 3 I \psi_3I ψ 3 I by (B.43). Summing the finitely many composition remainders and multiplying on the left by ψ 3 \psi_3 ψ 3 , we obtain the first estimate in (B.56).
Near zeroth order. The inner source satisfies (B.52). The first free estimate maps it to H − 1 k H^k_{-1} H − 1 k with factor C τ − 1 C\tau^{-1} C τ − 1 . The symbol ψ 3 r 0 , ℓ \psi_3r_{0,\ell} ψ 3 r 0 , ℓ has all seminorms required by the weighted estimate bounded by a constant, by (B.46). Apply that weighted estimate after the free inverse and sum in ℓ \ell ℓ . This gives the second bound.
Near first order. The same ordered composition applies. The seminorms of ψ 3 r 1 , ℓ \psi_3r_{1,\ell} ψ 3 r 1 , ℓ are now bounded by C τ C\sqrt{\tau} C τ ; the τ − 1 \tau^{-1} τ − 1 factor of the free inverse gives C τ − 1 / 2 C\tau^{-1/2} C τ − 1/2 . The transport cancellation used to obtain this factor was proved in (B.48), including the spatial-cutoff condition. This proves the third bound.
Far zeroth order. Here r 0 , 0 = V χ F r_{0,0}=V\chi_F r 0 , 0 = V χ F . Its compactly supported left symbol has uniformly bounded weighted seminorms. Apply (B.52), the first free bound and the weighted PDO bound in that order. This proves the fourth bound.
Far first order. Since the right-hand free inverse is a Fourier multiplier, there is the exact, not asymptotic, identity
Op ( r 1 , 0 ) E ξ c = Op ( a F ) . (449) \operatorname{Op}(r_{1,0})E_{\xi_c}=\operatorname{Op}(a_F). \tag*{(449)} Op ( r 1 , 0 ) E ξ c = Op ( a F ) . (449)
It is valid because the numerator vanishes near the singular set of p − 1 p^{-1} p − 1 . The unweighted PDO bound and the last bound in (B.46) give C τ − 1 / 2 C\tau^{-1/2} C τ − 1/2 . The remaining factor ψ 3 Op ( d 0 ) \psi_3\operatorname{Op}(d_0) ψ 3 Op ( d 0 ) is uniformly bounded on H k H^k H k . This proves the fifth bound.
All these constants are products and finite sums of the constants in Lemmas B.2 , B.3 and B.4 , with the fixed cutoff multiplier norms. The frame atlas was taken over the full compact prior closure, so its number of charts and its seminorms are already common constants. Thus all five bounds are uniform over the fixed compact coefficient class and frame atlas. The output support assertions follow from the left spatial symbols. They do not restrict the nonlocal input of those symbols.
The symbols b ℓ , d ℓ b_\ell,d_\ell b ℓ , d ℓ in (B.42) require at most N + 2 N+2 N + 2 spatial and two frequency derivatives; the frame order is at most N + 4 N+4 N + 4 . For the near symbols in (B.48), the bounds in (B.46) use at most N N N spatial and two frequency derivatives. Their largest frame orders are N + 4 N+4 N + 4 for the zeroth-order symbol and N + 3 N+3 N + 3 for the first-order symbol. The far symbols contain no frame. The orders fixed in (89) satisfy
N > k + 4 , r p a r ≥ N + 4. (450) N > k+4,\qquad r_{\mathrm{par}} \ge N+4. \tag*{(450)} N > k + 4 , r par ≥ N + 4. (450)
The composition estimate actually needs only N N N spatial derivatives and one frequency derivative of its left symbol. The bounds for b ℓ , d ℓ b_\ell,d_\ell b ℓ , d ℓ through spatial order N + 2 N+2 N + 2 and frequency order two also control the differential remainders. For an H k + 1 H^{k+1} H k + 1 bound, Schur integrability needs N > k + 4 N>k+4 N > k + 4 ; for the composition kernel its extra frequency-difference factor gives the same strict inequality. Both are met by N = k + 6 N=k+6 N = k + 6 .
The free estimates (B.17) and weighted bound (B.27) yield both Sobolev bounds for the same parametrix, in the stated operator order.
Lemma B.6. For τ ≥ 16 \tau\ge16 τ ≥ 16 and arbitrary g ∈ H k ( R 3 ) g \in H^k(\mathbb{R}^3) g ∈ H k ( R 3 ) ,
∥ ψ 2 Q τ g ∥ H k ≤ C Q , 0 τ − 1 ∥ g ∥ H k , ∥ ψ 2 Q τ g ∥ H k + 1 ≤ C Q , 1 ∥ g ∥ H k . (451) \lVert\psi_2 Q_\tau g\rVert_{H^k} \le C_{Q,0}\tau^{-1}\lVert g\rVert_{H^k},\qquad
\lVert\psi_2 Q_\tau g\rVert_{H^{k+1}} \le C_{Q,1}\lVert g\rVert_{H^k}. \tag*{(451)} ∥ ψ 2 Q τ g ∥ H k ≤ C Q , 0 τ − 1 ∥ g ∥ H k , ∥ ψ 2 Q τ g ∥ H k + 1 ≤ C Q , 1 ∥ g ∥ H k . (451) Both constants are uniform over the full class and all directions.
Proof. For each summand of (B.51), put F ℓ = ψ 3 Op ( d ℓ ) g F_\ell=\psi_3\operatorname{Op}(d_\ell)g F ℓ = ψ 3 Op ( d ℓ ) g . By (B.52), ∥ F ℓ ∥ H k ≤ C ∥ g ∥ H k \lVert F_\ell\rVert_{H^k}\le C\lVert g\rVert_{H^k} ∥ F ℓ ∥ H k ≤ C ∥ g ∥ H k uniformly in ℓ \ell ℓ . Apply the two free estimates (B.17) to obtain
∥ E ξ c F ℓ ∥ H k − 1 ≤ C τ − 1 ∥ g ∥ H k , ∥ E ξ c F ℓ ∥ H k + 1 ≤ C ∥ g ∥ H k . \lVert E_{\xi_c}F_\ell\rVert_{H^{k-1}} \le C\tau^{-1}\lVert g\rVert_{H^k},\qquad
\lVert E_{\xi_c}F_\ell\rVert_{H^{k+1}} \le C\lVert g\rVert_{H^k}. ∥ E ξ c F ℓ ∥ H k − 1 ≤ C τ − 1 ∥ g ∥ H k , ∥ E ξ c F ℓ ∥ H k + 1 ≤ C ∥ g ∥ H k . The symbol ψ 2 b ℓ \psi_2b_\ell ψ 2 b ℓ has fixed compact spatial support and the mixed seminorm bounds of Lemma B.4 . This also holds for ℓ = 0 \ell=0 ℓ = 0 after multiplication of the Fourier multiplier b 0 b_0 b 0 by ψ 2 \psi_2 ψ 2 . The derivative condition (B.58) permits use of (B.27) at both orders k k k and k + 1 k+1 k + 1 . Consequently
∥ Op ( ψ 2 b ℓ ) E ξ c F ℓ ∥ H k ≤ C τ − 1 ∥ g ∥ H k , \lVert\operatorname{Op}(\psi_2b_\ell)E_{\xi_c}F_\ell\rVert_{H^k} \le C\tau^{-1}\lVert g\rVert_{H^k}, ∥ Op ( ψ 2 b ℓ ) E ξ c F ℓ ∥ H k ≤ C τ − 1 ∥ g ∥ H k , ∥ Op ( ψ 2 b ℓ ) E ξ c F ℓ ∥ H k + 1 ≤ C ∥ g ∥ H k . \lVert\operatorname{Op}(\psi_2b_\ell)E_{\xi_c}F_\ell\rVert_{H^{k+1}} \le C\lVert g\rVert_{H^k}. ∥ Op ( ψ 2 b ℓ ) E ξ c F ℓ ∥ H k + 1 ≤ C ∥ g ∥ H k . Since Op ( ψ 2 b ℓ ) = ψ 2 Op ( b ℓ ) \operatorname{Op}(\psi_2b_\ell)=\psi_2\operatorname{Op}(b_\ell) Op ( ψ 2 b ℓ ) = ψ 2 Op ( b ℓ ) , summing over the fixed finite atlas proves (B.59). The weighted estimate acts on the full output of the free inverse; no spatial cutoff has been inserted between the two operators.
One Neumann inverse and the local equation Proof of Theorem 3.2 . The frame construction of Theorem 3.1 , the free inverse of Lemma B.2 , and the operator estimates of Lemma B.3 apply to every finite-regularity point of the weaker compact coefficient closure. Lemma B.5 gives five common constants. Use (89) . For τ ≥ τ 0 \tau\ge\tau_0 τ ≥ τ 0 , since τ − 1 ≤ τ − 1 / 2 \tau^{-1}\le\tau^{-1/2} τ − 1 ≤ τ − 1/2 ,
∥ T τ ∥ H k → H k ≤ C Σ τ − 1 / 2 ≤ 1 4 . (452) \lVert T_\tau\rVert_{H^k\to H^k} \le C_\Sigma\tau^{-1/2} \le\frac{1}{4}. \tag*{(452)} ∥ T τ ∥ H k → H k ≤ C Σ τ − 1/2 ≤ 4 1 . (452)
Construct exactly one inverse, on H k ( R 3 ; C d s y s ) H^k(\mathbb{R}^3;\mathbb{C}^{d_{\mathrm{sys}}}) H k ( R 3 ; C d sys ) :
J τ = ( I + T τ ) − 1 = ∑ n = 0 ∞ ( − T τ ) n , ∥ J τ ∥ H k → H k ≤ 4 / 3. (453) J_\tau=(I+T_\tau)^{-1}=\sum_{n=0}^{\infty}(-T_\tau)^n,\qquad\|J_\tau\|_{H^k\to H^k}\leq4/3. \tag*{(453)} J τ = ( I + T τ ) − 1 = n = 0 ∑ ∞ ( − T τ ) n , ∥ J τ ∥ H k → H k ≤ 4/3. (453)
For f ∈ E k f\in E_k f ∈ E k extend by zero, put g = J τ f g=J_\tau f g = J τ f , and define
R h , θ f = − ψ 2 Q τ J τ f ∣ Ω 0 . (454) \mathcal{R}_{h,\theta}f=-\left.\psi_2Q_\tau J_\tau f\right|_{\Omega_0}. \tag*{(454)} R h , θ f = − ψ 2 Q τ J τ f ∣ Ω 0 . (454)
The negative sign is forced by P ξ c = − L h , θ P_{\xi_c}=-L_{h,\theta} P ξ c = − L h , θ . The relation g = f − T τ g g=f-T_\tau g g = f − T τ g shows only
supp g ⊂ K s r c ∪ supp ψ 3 ; (455) \operatorname{supp}g\subset K_{\mathrm{src}}\cup\operatorname{supp}\psi_3; \tag*{(455)} supp g ⊂ K src ∪ supp ψ 3 ; (455)
it does not say g ∈ E k g\in E_k g ∈ E k . This larger fixed support is harmless: the actual inputs to the free inverse are always ψ 3 Op ( d ℓ ) g \psi_3\operatorname{Op}(d_\ell)g ψ 3 Op ( d ℓ ) g as in (B.52).
On Ω 1 \Omega_1 Ω 1 both spatial cutoffs equal one on a neighborhood. Thus no derivatives of ψ 2 \psi_2 ψ 2 occur there, and (B.54) gives
L h , θ R h , θ f = P ξ c Q τ g = g + ( S τ + E τ ) g = g + ψ 3 ( S τ + E τ ) g = ( I + T τ ) g = f in D ′ ( Ω 1 ) . (456) \begin{aligned}
L_{h,\theta}\mathcal{R}_{h,\theta}f&=P_{\xi_c}Q_\tau g=g+(S_\tau+E_\tau)g\\
&=g+\psi_3(S_\tau+E_\tau)g=(I+T_\tau)g=f\quad\text{in }D'(\Omega_1). \tag*{(456)}
\end{aligned} L h , θ R h , θ f = P ξ c Q τ g = g + ( S τ + E τ ) g = g + ψ 3 ( S τ + E τ ) g = ( I + T τ ) g = f in D ′ ( Ω 1 ) . (456)
In the middle equality the comparison is restricted to Ω 1 \Omega_1 Ω 1 ; it is not an identity on all of R 3 \mathbb{R}^3 R 3 . Lemma B.6 and (B.61) now give
∥ ψ 2 Q τ J τ f ∥ H k ≤ 4 3 C Q , 0 τ − 1 ∥ f ∥ H k , ∥ ψ 2 Q τ J τ f ∥ H k + 1 ≤ 4 3 C Q , 1 ∥ f ∥ H k . (457) \begin{aligned}
\|\psi_2Q_\tau J_\tau f\|_{H^k}&\leq\frac{4}{3}C_{Q,0}\tau^{-1}\|f\|_{H^k},\\
\|\psi_2Q_\tau J_\tau f\|_{H^{k+1}}&\leq\frac{4}{3}C_{Q,1}\|f\|_{H^k}. \tag*{(457)}
\end{aligned} ∥ ψ 2 Q τ J τ f ∥ H k ∥ ψ 2 Q τ J τ f ∥ H k + 1 ≤ 3 4 C Q , 0 τ − 1 ∥ f ∥ H k , ≤ 3 4 C Q , 1 ∥ f ∥ H k . (457)
Choose C k = 4 3 ( C Q , 0 + C Q , 1 ) C_k=\frac{4}{3}(C_{Q,0}+C_{Q,1}) C k = 3 4 ( C Q , 0 + C Q , 1 ) multiplied by the fixed extension norm. Then (B.65) gives the required sum of norms in (88) . No Neumann series on H k + 1 H^{k+1} H k + 1 has been constructed or used. The atlas includes all oriented pairs, hence both phases have the same h ∗ h_* h ∗ . Their respective Q τ Q_\tau Q τ , T τ T_\tau T τ , J τ J_\tau J τ are allowed to differ. The geometric cutoff of the final output limits the inverse identity to Ω 1 \Omega_1 Ω 1 , but never deletes a tail inside the parametrix. There is no amplitude-size restriction.
Finite-order physical boundary residual
This appendix proves the physical residual and energy-remainder estimates from the fixed charts, the constant half-space inverse and the recursion of § 8 .
The full transformed elasticity tensor and weighted half-space profiles control the Taylor remainder and the remote cutoff terms at order s s s , without additional material derivatives.
For the chart (282) define
J p ( z ) = D Ψ p ( z ) , C i j k l = λ δ i j δ k l + μ ( δ i k δ j l + δ i l δ j k ) . (458) J_p(z)=D\Psi_p(z),\qquad C_{ijkl}=\lambda\delta_{ij}\delta_{kl}+\mu(\delta_{ik}\delta_{jl}+\delta_{il}\delta_{jk}). \tag*{(458)} J p ( z ) = D Ψ p ( z ) , C ij k l = λ δ ij δ k l + μ ( δ ik δ j l + δ i l δ j k ) . (458)
Here δ i j \delta_{ij} δ ij is the Kronecker symbol, J p J_p J p is a 3 × 3 3\times3 3 × 3 Jacobian matrix, and R p R_p R p is the fixed component rotation in (282) . The transformed tensor in (284) is
A a b c d , p ( z ) = ∣ det J p ( z ) ∣ ∑ i 1 , i 2 , i 3 , i 4 = 1 3 C i 1 i 2 i 3 i 4 ( Ψ p ( z ) ) × ( R p ) i 1 a ( R p ) i 3 c ( J p − 1 ( z ) ) b i 2 ( J p − 1 ( z ) ) d i 4 . (459) \begin{aligned}
A_{abcd,p}(z)
&=|\det J_p(z)|\sum_{i_1,i_2,i_3,i_4=1}^{3}
C_{i_1i_2i_3i_4}(\Psi_p(z))\\
&\quad\times(R_p)_{i_1a}(R_p)_{i_3c}(J_p^{-1}(z))_{bi_2}(J_p^{-1}(z))_{di_4}. \tag*{(459)}
\end{aligned} A ab c d , p ( z ) = ∣ det J p ( z ) ∣ i 1 , i 2 , i 3 , i 4 = 1 ∑ 3 C i 1 i 2 i 3 i 4 ( Ψ p ( z )) × ( R p ) i 1 a ( R p ) i 3 c ( J p − 1 ( z ) ) b i 2 ( J p − 1 ( z ) ) d i 4 . (459)
Indeed, writing u ( Ψ p ( z ) ) = R p v ( z ) u(\Psi_p(z)) = R_pv(z) u ( Ψ p ( z )) = R p v ( z ) , the chain rule gives
( ∂ x i 2 u i 1 ) ( Ψ p ( z ) ) = ∑ a , b = 1 3 ( R p ) i 1 a ( ∂ z b v a ) ( z ) ( J p − 1 ( z ) ) b i 2 . (\partial_{x_{i_2}}u_{i_1})(\Psi_p(z)) = \sum_{a,b=1}^{3}(R_p)_{i_1a}(\partial_{z_b}v_a)(z)(J_p^{-1}(z))_{bi_2}. ( ∂ x i 2 u i 1 ) ( Ψ p ( z )) = a , b = 1 ∑ 3 ( R p ) i 1 a ( ∂ z b v a ) ( z ) ( J p − 1 ( z ) ) b i 2 .
Substitution in the physical energy gives (C.2) without differentiating R p R_p R p . The absolute determinant handles either orientation of the orthogonal base matrix. At z = 0 z=0 z = 0 , J p = R p J_p=R_p J p = R p and isotropy gives the frozen tensor in (284) . Away from the base point one must use the entire tensor (C.2), not its frozen isotropic formula. The fixed smooth charts give a uniform C s C^s C s tensor bound from the stated C s C^s C s material bound. Taylor expansion is performed only along segments contained in a fixed coordinate half-ball.
For this subsection s ≥ 2 s\ge2 s ≥ 2 is the fixed integer of (4) . Let V r V_r V r , 0 ≤ r < s 0\le r<s 0 ≤ r < s , be the profiles in (302) . Define the following single finite collection of seminorms:
B s : = max 0 ≤ r < s ( ∥ ⟨ y ⟩ s V r ∥ L 2 ( R + 3 ) + ∥ ⟨ y ⟩ s ∇ V r ∥ L 2 ( R + 3 ) ) . (460) \mathcal{B}_s := \max_{0\le r<s}\left(\|\langle y\rangle^s V_r\|_{L^2(\mathbb{R}^3_+)}+\|\langle y\rangle^s\nabla V_r\|_{L^2(\mathbb{R}^3_+)}\right). \tag*{(460)} B s := 0 ≤ r < s max ( ∥ ⟨ y ⟩ s V r ∥ L 2 ( R + 3 ) + ∥ ⟨ y ⟩ s ∇ V r ∥ L 2 ( R + 3 ) ) . (460)
The weight (B.14) is equivalent to the weight in (296) .
Lemma C.1. The profiles in (302) can be written in (295) with normal polynomial degree at most 3 r + 1 3r+1 3 r + 1 . The number B s \mathcal{B}_s B s is uniformly bounded by the fixed C s C^s C s material class, the fixed charts, and the fixed annular tests. Moreover, for any N > 1 N>1 N > 1 , L ≥ 1 L\ge1 L ≥ 1 , and any scalar, vector or matrix field F F F on R + 3 \mathbb{R}^3_+ R + 3 ,
∥ 1 ∣ y ∣ > L F ∥ 2 ≤ C L − N ∥ ⟨ y ⟩ N F ∥ 2 , \|1_{|y|>L}F\|_2\le CL^{-N}\|\langle y\rangle^N F\|_2, ∥ 1 ∣ y ∣ > L F ∥ 2 ≤ C L − N ∥ ⟨ y ⟩ N F ∥ 2 , ∥ 1 ∣ y ∣ > L F ∥ 6 / 5 ≤ C N L 1 − N ∥ ⟨ y ⟩ N F ∥ 2 . (461) \|1_{|y|>L}F\|_{6/5}\le C_NL^{1-N}\|\langle y\rangle^N F\|_2. \tag*{(461)} ∥ 1 ∣ y ∣ > L F ∥ 6/5 ≤ C N L 1 − N ∥ ⟨ y ⟩ N F ∥ 2 . (461)
Proof. The homogeneous Poisson profile has degree at most one. If a source has normal polynomial degree D D D , convolution with the Green kernel in Lemma 8.1 has degree at most D + 2 D+2 D + 2 : on 0 < u < t 0<u<t 0 < u < t one integrates a polynomial of degree at most D D D against a polynomial of degree at most one in t − u t-u t − u , and on u > t u>t u > t the exponentially decaying integral has no larger degree. We subtract the homogeneous Poisson solution with boundary data equal to the trace of the particular solution. This correction increases the normal polynomial degree by at most one. The interior source is first calculated for t > 0 t>0 t > 0 and only then extended by zero for this convolution. Extending a divergence-form flux before differentiating would create a boundary distribution and is not the construction used here.
Multiplication by a physical polynomial of degree ℓ \ell ℓ increases the normal polynomial degree by at most ℓ \ell ℓ : each tangential frequency derivative can create at most one additional power of t t t in e − ϖ t e^{-\varpi t} e − ϖ t . Spatial differentiation does not increase this degree. If D r D_r D r denotes a degree bound for V r V_r V r , the recursion therefore gives
D 0 ≤ 1 , D r ≤ max 1 ≤ ℓ ≤ r ( D r − ℓ + ℓ + 2 ) ≤ 3 r + 1. (462) D_0\le1,\qquad D_r\le\max_{1\le\ell\le r}(D_{r-\ell}+\ell+2)\le3r+1. \tag*{(462)} D 0 ≤ 1 , D r ≤ 1 ≤ ℓ ≤ r max ( D r − ℓ + ℓ + 2 ) ≤ 3 r + 1. (462)
All frequency coefficients have support in the same compact subannulus. Their finitely many seminorms depend smoothly on the finite-order tensor derivatives and on the frozen moduli, with uniformly nonzero denominators. The integral formulas in Lemma 8.1 then bound (C.3). The profile bounds use derivatives of the Lamé parameters only through the finite Taylor orders, not derivatives of order 3 r + 1 3r+1 3 r + 1 of the original material. Polynomial degree of a constructed profile and material regularity are different quantities.
One may take as a concrete input seminorm for the fixed annular test max ∣ γ ∣ ≤ 2 s + 2 ∥ ∂ γ ψ a ∥ ∞ \max_{|\gamma|\le2s+2}\|\partial^\gamma\psi_a\|_\infty max ∣ γ ∣ ≤ 2 s + 2 ∥ ∂ γ ψ a ∥ ∞ . Indeed, to bound N N N frequency derivatives of a coefficient of V r V_r V r , multiplication by a Taylor polynomial of degree ℓ \ell ℓ uses at most ℓ \ell ℓ additional frequency derivatives of V r − ℓ V_{r-\ell} V r − ℓ ; the Green and Poisson operations introduce none. Induction needs at most N + r N+r N + r derivatives of the initial test. The weights in B s \mathcal{B}_s B s use N ≤ s N\le s N ≤ s , and r ≤ s − 1 r\le s-1 r ≤ s − 1 . Thus derivatives of the fixed annular test through order 2 s + 2 2s + 2 2 s + 2 suffice. More explicitly, if T r ( N ) T_r(N) T r ( N ) is the largest derivative order of the initial annular test needed for N N N tangential frequency derivatives of the coefficients of V r V_r V r , then
T 0 ( N ) ≤ N , T r ( N ) ≤ max 1 ≤ ℓ ≤ r T r − ℓ ( N + ℓ ) ≤ N + r . (463) T_0(N) \le N,\qquad T_r(N) \le\max_{1\le\ell\le r} T_{r-\ell}(N+\ell) \le N+r. \tag*{(463)} T 0 ( N ) ≤ N , T r ( N ) ≤ 1 ≤ ℓ ≤ r max T r − ℓ ( N + ℓ ) ≤ N + r . (463)
Spatial differentiation of a profile only multiplies its frequency coefficients or differentiates the finite normal polynomial; it creates no derivatives of the test. The Green and trace-subtraction operations are finite smooth multipliers and integrals on the fixed annulus. Their own frequency derivatives cost fixed multiplier constants. The weight ⟨ y ⟩ s \langle y\rangle^s ⟨ y ⟩ s requires at most s s s tangential frequency derivatives and at most s s s normal powers. Thus r ≤ s − 1 r\le s-1 r ≤ s − 1 uses at most 2 s − 1 2s-1 2 s − 1 derivatives of the fixed test. In particular, the seminorm involving test derivatives through order 2 s + 2 2s+2 2 s + 2 controls both polarizations and their gradients. Tensor jets are still used only through order r r r , and the tensor Taylor remainder only through order s s s .
The first tail inequality is immediate. For the second, apply Hölder with 1 / ( 6 / 5 ) = 1 / 2 + 1 / 3 1/(6/5)=1/2+1/3 1/ ( 6/5 ) = 1/2 + 1/3 to ⟨ y ⟩ N F 1 ∣ y ∣ > L ⟨ y ⟩ − N \langle y\rangle^N F\,1_{|y|>L}\langle y\rangle^{-N} ⟨ y ⟩ N F 1 ∣ y ∣ > L ⟨ y ⟩ − N . The L 3 L^3 L 3 norm of the second factor is at most C N L 1 − N C_NL^{1-N} C N L 1 − N in three dimensions, by radial integration. The weighted Hölder argument proves (C.4). Choosing N = s N=s N = s is legitimate since s ≥ 2 s\ge2 s ≥ 2 .
The physical residual and energy expansion
The weak residual splits into Taylor interactions, a tensor remainder, and two cutoff commutators. Each term is estimated against arbitrary zero-trace physical tests.
For this proof, abbreviate
V ϵ = ∑ r = 0 s − 1 ϵ r V r , χ ϵ ( y ) = χ p ( ϵ y ) , A ϵ ( y ) = A p ( ϵ y ) , V_\epsilon=\sum_{r=0}^{s-1}\epsilon^rV_r,\qquad\chi_\epsilon(y)=\chi_p(\epsilon y),\qquad A_\epsilon(y)=A_p(\epsilon y), V ϵ = r = 0 ∑ s − 1 ϵ r V r , χ ϵ ( y ) = χ p ( ϵy ) , A ϵ ( y ) = A p ( ϵy ) ,
T ϵ ( y ) = ∑ ℓ = 0 s − 1 ϵ ℓ A p , ℓ ( y ) , R ϵ = A ϵ − T ϵ . (464) T_\epsilon(y)=\sum_{\ell=0}^{s-1}\epsilon^\ell A_{p,\ell}(y),\qquad R_\epsilon=A_\epsilon-T_\epsilon. \tag*{(464)} T ϵ ( y ) = ℓ = 0 ∑ s − 1 ϵ ℓ A p , ℓ ( y ) , R ϵ = A ϵ − T ϵ . (464)
A ϵ A_\epsilon A ϵ is used only on the scaled chart. The polynomial T ϵ T_\epsilon T ϵ is defined throughout the half-space. Here R ϵ R_\epsilon R ϵ denotes the Taylor remainder defined in (C.7). Let w ϵ w_\epsilon w ϵ be the scaled physical test defined on the scaled chart by
w ϵ ( y ) = ϵ 1 / 2 R p T w ( Ψ p ( ϵ y ) ) . (465) w_\epsilon(y)=\epsilon^{1/2}R_p^T w(\Psi_p(\epsilon y)). \tag*{(465)} w ϵ ( y ) = ϵ 1/2 R p T w ( Ψ p ( ϵy )) . (465)
All terms containing it in (C.9) have support inside that chart.
Proposition C.2. For U p , ϵ U_{p,\epsilon} U p , ϵ in (8.26) and every w ∈ H 0 1 ( Ω ; C 3 ) w\in H^1_0(\Omega;\mathbb{C}^3) w ∈ H 0 1 ( Ω ; C 3 ) , one has the exact identity
a ( U p , ϵ , w ) = H ϵ ( w ) + D ϵ ( w ) + C 1 , ϵ ( w ) − C 2 , ϵ ( w ) , (466) a(U_{p,\epsilon},w)=H_\epsilon(w)+D_\epsilon(w)+C_{1,\epsilon}(w)-C_{2,\epsilon}(w), \tag*{(466)} a ( U p , ϵ , w ) = H ϵ ( w ) + D ϵ ( w ) + C 1 , ϵ ( w ) − C 2 , ϵ ( w ) , (466) with
H ϵ ( w ) = ∑ 0 ≤ ℓ , r < s ℓ + r ≥ s ϵ ℓ + r a ℓ ( V r , χ ϵ w ϵ ) , H_\epsilon(w)=\sum_{\substack{0\le\ell,r<s\\ \ell+r\ge s}}\epsilon^{\ell+r}a_\ell(V_r,\chi_\epsilon w_\epsilon), H ϵ ( w ) = 0 ≤ ℓ , r < s ℓ + r ≥ s ∑ ϵ ℓ + r a ℓ ( V r , χ ϵ w ϵ ) , D ϵ ( w ) = ∫ χ ϵ R ϵ ∇ V ϵ : ∇ w ϵ , D_\epsilon(w)=\int\chi_\epsilon R_\epsilon\nabla V_\epsilon:\nabla w_\epsilon, D ϵ ( w ) = ∫ χ ϵ R ϵ ∇ V ϵ : ∇ w ϵ , C 1 , ϵ ( w ) = ∫ A ϵ ( V ϵ ⊗ ∇ χ ϵ ) : ∇ w ϵ , C_{1,\epsilon}(w)=\int A_\epsilon(V_\epsilon\otimes\nabla\chi_\epsilon):\nabla w_\epsilon, C 1 , ϵ ( w ) = ∫ A ϵ ( V ϵ ⊗ ∇ χ ϵ ) : ∇ w ϵ , C 2 , ϵ ( w ) = ∫ T ϵ ∇ V ϵ : ( w ϵ ⊗ ∇ χ ϵ ) . (467) C_{2,\epsilon}(w)=\int T_\epsilon\nabla V_\epsilon:(w_\epsilon\otimes\nabla\chi_\epsilon). \tag*{(467)} C 2 , ϵ ( w ) = ∫ T ϵ ∇ V ϵ : ( w ϵ ⊗ ∇ χ ϵ ) . (467) The integrals are over the scaled coordinate half-ball, with zero extension on the region of vanishing cutoffs. Uniformly in the material, p p p and 0 < ϵ < ϵ ∗ 0<\epsilon<\epsilon_* 0 < ϵ < ϵ ∗ ,
∣ H ϵ ( w ) ∣ + ∣ D ϵ ( w ) ∣ ≤ C ϵ s ∥ w ∥ H 1 , |H_\epsilon(w)|+|D_\epsilon(w)|\le C\epsilon^s\|w\|_{H^1}, ∣ H ϵ ( w ) ∣ + ∣ D ϵ ( w ) ∣ ≤ C ϵ s ∥ w ∥ H 1 , ∣ C 1 , ϵ ( w ) ∣ ≤ C ϵ s + 1 ∥ w ∥ H 1 , |C_{1,\epsilon}(w)|\le C\epsilon^{s+1}\|w\|_{H^1}, ∣ C 1 , ϵ ( w ) ∣ ≤ C ϵ s + 1 ∥ w ∥ H 1 , ∣ C 2 , ϵ ( w ) ∣ ≤ C ϵ s ∥ w ∥ H 1 . (468) |C_{2,\epsilon}(w)|\le C\epsilon^s\|w\|_{H^1}. \tag*{(468)} ∣ C 2 , ϵ ( w ) ∣ ≤ C ϵ s ∥ w ∥ H 1 . (468)
Moreover, for the fixed test ϕ \phi ϕ used to construct U p , ϵ U_{p,\epsilon} U p , ϵ ,
∣ a ( U p , ϵ , U p , ϵ ) − ∑ r = 0 s − 1 ϵ r E p , r ϕ ∣ ≤ C ϵ s , (469) \left|a(U_{p,\epsilon},U_{p,\epsilon})-\sum_{r=0}^{s-1}\epsilon^r E_{p,r}^{\phi}\right|\leq C\epsilon^s, \tag*{(469)} a ( U p , ϵ , U p , ϵ ) − r = 0 ∑ s − 1 ϵ r E p , r ϕ ≤ C ϵ s , (469)
where E p , r ϕ = ∑ ℓ + b + c = r a ℓ ( V b , V c ) E_{p,r}^{\phi}=\sum_{\ell+b+c=r}a_{\ell}(V_b,V_c) E p , r ϕ = ∑ ℓ + b + c = r a ℓ ( V b , V c ) , with a ℓ a_{\ell} a ℓ defined in (300) and the profiles defined in (297) and (302) .
Proof. Physical scaling and (C.2) give exactly
a ( U p , ϵ , w ) = ∫ A ϵ ∇ ( χ ϵ V ϵ ) : ∇ w ϵ . (470) a(U_{p,\epsilon},w)=\int A_{\epsilon}\nabla(\chi_{\epsilon}V_{\epsilon}):\nabla w_{\epsilon}. \tag*{(470)} a ( U p , ϵ , w ) = ∫ A ϵ ∇ ( χ ϵ V ϵ ) : ∇ w ϵ . (470)
Apply the product rule, insert A ϵ = T ϵ + R ϵ A_{\epsilon}=T_{\epsilon}+R_{\epsilon} A ϵ = T ϵ + R ϵ only in the term with χ ϵ ∇ V ϵ \chi_{\epsilon}\nabla V_{\epsilon} χ ϵ ∇ V ϵ , and move χ ϵ \chi_{\epsilon} χ ϵ onto the test in the T ϵ T_{\epsilon} T ϵ term. The last operation contributes − C 2 , ϵ -C_{2,\epsilon} − C 2 , ϵ . The remaining Taylor-polynomial integral is
∑ ℓ , r < s ϵ ℓ + r a ℓ ( V r , χ ϵ w ϵ ) . (471) \sum_{\ell,r<s}\epsilon^{\ell+r}a_{\ell}(V_r,\chi_{\epsilon}w_{\epsilon}). \tag*{(471)} ℓ , r < s ∑ ϵ ℓ + r a ℓ ( V r , χ ϵ w ϵ ) . (471)
For every n < s n<s n < s , its coefficient of ϵ n \epsilon^n ϵ n is
∑ ℓ + r = n a ℓ ( V r , χ ϵ w ϵ ) = 0. (472) \sum_{\ell+r=n}a_{\ell}(V_r,\chi_{\epsilon}w_{\epsilon})=0. \tag*{(472)} ℓ + r = n ∑ a ℓ ( V r , χ ϵ w ϵ ) = 0. (472)
For n = 0 n=0 n = 0 this is the frozen homogeneous equation. For n ≥ 1 n\geq1 n ≥ 1 it is (302) . The test is in H ˙ 0 1 ( R + 3 ) \dot{H}^{1}_{0}(\mathbb{R}^{3}_{+}) H ˙ 0 1 ( R + 3 ) because w w w has zero physical boundary trace and the cutoff vanishes before the edge of the chart. This gives the identity (C.9).
We record the scale factors. The volume scaling is ϵ 3 \epsilon^3 ϵ 3 and each gradient scaling is ϵ − 3 / 2 \epsilon^{-3/2} ϵ − 3/2 , so there is no extra prefactor in the identity. Also
∥ ∇ w ϵ ∥ 2 + ∥ w ϵ ∥ 6 + ∥ ∇ ( χ ϵ w ϵ ) ∥ 2 ≤ C ∥ w ∥ H 1 ( Ω ) , ∥ ∇ χ ϵ ∥ 3 ≤ C . (473) \|\nabla w_{\epsilon}\|_{2}+\|w_{\epsilon}\|_{6}+\|\nabla(\chi_{\epsilon}w_{\epsilon})\|_{2}\leq C\|w\|_{H^{1}(\Omega)},\qquad\|\nabla\chi_{\epsilon}\|_{3}\leq C. \tag*{(473)} ∥∇ w ϵ ∥ 2 + ∥ w ϵ ∥ 6 + ∥∇ ( χ ϵ w ϵ ) ∥ 2 ≤ C ∥ w ∥ H 1 ( Ω ) , ∥∇ χ ϵ ∥ 3 ≤ C . (473)
These bounds are local on the scaled chart; its radius is of order ϵ − 1 \epsilon^{-1} ϵ − 1 . No Poincare constant for that expanding chart is used. The L 6 L^{6} L 6 bound follows from the physical three-dimensional Sobolev inequality; ϵ 1 / 2 \epsilon^{1/2} ϵ 1/2 cancels the change of volume in that norm. The last gradient bound follows by L 3 L^{3} L 3 –L 6 L^{6} L 6 Hölder.
Each polynomial A p , ℓ A_{p,\ell} A p , ℓ grows at most as ⟨ y ⟩ ℓ \langle y\rangle^{\ell} ⟨ y ⟩ ℓ , and ℓ ≤ s − 1 \ell\leq s-1 ℓ ≤ s − 1 . The uniform weighted half-space profile bounds therefore give
∣ a ℓ ( V r , χ ϵ w ϵ ) ∣ ≤ C B s ∥ w ∥ H 1 . (474) |a_{\ell}(V_r,\chi_{\epsilon}w_{\epsilon})|\leq C\mathcal{B}_{s}\|w\|_{H^{1}}. \tag*{(474)} ∣ a ℓ ( V r , χ ϵ w ϵ ) ∣ ≤ C B s ∥ w ∥ H 1 . (474)
Since ℓ + r ≥ s \ell+r\geq s ℓ + r ≥ s in H ϵ H_{\epsilon} H ϵ and ϵ ≤ 1 \epsilon\leq1 ϵ ≤ 1 , the asserted bound for H ϵ H_{\epsilon} H ϵ follows. Similarly, ∣ R ϵ ( y ) ∣ ≤ C ϵ s ∣ y ∣ s |R_{\epsilon}(y)|\leq C\epsilon^{s}|y|^{s} ∣ R ϵ ( y ) ∣ ≤ C ϵ s ∣ y ∣ s and the first gradient bound in (C.16) control D ϵ D_{\epsilon} D ϵ . There is no derivative of R ϵ R_{\epsilon} R ϵ in this calculation.
There are fixed a , A > 0 a,A>0 a , A > 0 such that
supp ∇ χ ϵ ⊂ { a / ϵ ≤ ∣ y ∣ ≤ A / ϵ } , ∣ ∇ χ ϵ ∣ ≤ C ϵ . (475) \operatorname{supp}\nabla\chi_{\epsilon}\subset\{a/\epsilon\leq|y|\leq A/\epsilon\},\qquad|\nabla\chi_{\epsilon}|\leq C\epsilon. \tag*{(475)} supp ∇ χ ϵ ⊂ { a / ϵ ≤ ∣ y ∣ ≤ A / ϵ } , ∣∇ χ ϵ ∣ ≤ C ϵ . (475)
On this set A ϵ A_{\epsilon} A ϵ is bounded by the physical coefficient bound, and T ϵ T_{\epsilon} T ϵ is also bounded because ϵ ℓ ∣ A p , ℓ ( y ) ∣ ≤ C ∣ ϵ y ∣ ℓ \epsilon^{\ell}|A_{p,\ell}(y)|\leq C|\epsilon y|^{\ell} ϵ ℓ ∣ A p , ℓ ( y ) ∣ ≤ C ∣ ϵy ∣ ℓ . The first tail inequality with N = s N=s N = s yields
∥ A ϵ ( V ϵ ⊗ ∇ χ ϵ ) ∥ 2 ≤ C ϵ ϵ s B s . (476) \|A_{\epsilon}(V_{\epsilon}\otimes\nabla\chi_{\epsilon})\|_{2}\leq C\epsilon\,\epsilon^{s}\mathcal{B}_{s}. \tag*{(476)} ∥ A ϵ ( V ϵ ⊗ ∇ χ ϵ ) ∥ 2 ≤ C ϵ ϵ s B s . (476)
By Cauchy–Schwarz, (C.19), and the gradient bound in (C.16), we obtain the required estimate for C 1 , ϵ C_{1,\epsilon} C 1 , ϵ . For the other commutator, the second tail inequality gives
∥ 1 supp ∇ χ ϵ T ϵ ∇ V ϵ ∇ χ ϵ ∥ 6 / 5 ≤ C ϵ ϵ s − 1 B s . (477) \|\mathbf{1}_{\operatorname{supp}\nabla\chi_{\epsilon}}T_{\epsilon}\nabla V_{\epsilon}\nabla\chi_{\epsilon}\|_{6/5}\leq C\epsilon\,\epsilon^{s-1}\mathcal{B}_{s}. \tag*{(477)} ∥ 1 supp ∇ χ ϵ T ϵ ∇ V ϵ ∇ χ ϵ ∥ 6/5 ≤ C ϵ ϵ s − 1 B s . (477)
By Hölder’s inequality, (C.20), and the L 6 L^6 L 6 bound for w ε w_\varepsilon w ε in (C.16), we obtain the required estimate for C 2 , ε C_{2,\varepsilon} C 2 , ε . Equivalently, with R ε \mathcal{R}_\varepsilon R ε the physical functional in (C.9),
∥ R ε ∥ H − 1 ( Ω ) : = sup w ∈ H 0 1 ( Ω ) , ∥ w ∥ H 1 = 1 ∣ a ( U p , ε , w ) ∣ ≤ C ε s . (478) \|\mathcal{R}_\varepsilon\|_{H^{-1}(\Omega)} := \sup_{\substack{w\in H^1_0(\Omega),\ \|w\|_{H^1}=1}} |a(U_{p,\varepsilon},w)| \le C\varepsilon^s. \tag*{(478)} ∥ R ε ∥ H − 1 ( Ω ) := w ∈ H 0 1 ( Ω ) , ∥ w ∥ H 1 = 1 sup ∣ a ( U p , ε , w ) ∣ ≤ C ε s . (478)
For a scalar scaled density F F F , the corresponding physical density F p h y s ( Ψ p ( ε y ) ) = ε − 5 / 2 F ( y ) F^{\mathrm{phys}}(\Psi_p(\varepsilon y))=\varepsilon^{-5/2}F(y) F phys ( Ψ p ( ε y )) = ε − 5/2 F ( y ) (with the bounded tensor/Jacobian factors retained) pairs against w ε w_\varepsilon w ε with no extra power. Its L 6 / 5 L^{6/5} L 6/5 norm is invariant, since 3 / ( 6 / 5 ) − 5 / 2 = 0 3/(6/5)-5/2=0 3/ ( 6/5 ) − 5/2 = 0 . A scaled flux becomes a physical flux of size ε − 3 / 2 \varepsilon^{-3/2} ε − 3/2 and its L 2 L^2 L 2 norm is invariant, since 3 / 2 − 3 / 2 = 0 3/2-3/2=0 3/2 − 3/2 = 0 . These are exactly the density pairing used for C 2 , ε C_{2,\varepsilon} C 2 , ε and the flux pairings used for D ε D_\varepsilon D ε , C 1 , ε C_{1,\varepsilon} C 1 , ε . The high-interaction term H ε H_\varepsilon H ε uses ∇ ( χ ε w ε ) \nabla(\chi_\varepsilon w_\varepsilon) ∇ ( χ ε w ε ) , whose bound is already in (C.16). Thus the dual estimate is on the fixed physical domain, with no inhomogeneous L 2 L^2 L 2 bound for w ε w_\varepsilon w ε on the expanding half-ball.
For the approximate energy a ( U p , ε , U p , ε ) a(U_{p,\varepsilon},U_{p,\varepsilon}) a ( U p , ε , U p , ε ) , the Taylor error is controlled by ε s ∫ ∣ y ∣ s ∣ ∇ V ε ∣ 2 \varepsilon^s\int|y|^s|\nabla V_\varepsilon|^2 ε s ∫ ∣ y ∣ s ∣∇ V ε ∣ 2 . To make the energy-cutoff terms explicit, set
Q ε = V ε ⊗ ∇ χ ε , (479) Q_\varepsilon=V_\varepsilon\otimes\nabla\chi_\varepsilon, \tag*{(479)} Q ε = V ε ⊗ ∇ χ ε , (479)
where V ε V_\varepsilon V ε and χ ε \chi_\varepsilon χ ε are defined in (C.7). The already proved tail estimate gives ∥ Q ε ∥ 2 ≤ C ε s + 1 B s \|Q_\varepsilon\|_2\le C\varepsilon^{s+1}\mathcal{B}_s ∥ Q ε ∥ 2 ≤ C ε s + 1 B s , whereas ∥ χ ε ∇ V ε ∥ 2 ≤ C B s \|\chi_\varepsilon\nabla V_\varepsilon\|_2\le C\mathcal{B}_s ∥ χ ε ∇ V ε ∥ 2 ≤ C B s . Boundedness of A ε A_\varepsilon A ε therefore bounds the cross terms involving one Q ε Q_\varepsilon Q ε by C ε s + 1 B s 2 C\varepsilon^{s+1}\mathcal{B}_s^2 C ε s + 1 B s 2 and the quadratic term by C ε 2 s + 2 B s 2 C\varepsilon^{2s+2}\mathcal{B}_s^2 C ε 2 s + 2 B s 2 . No differentiation of the material remainder occurs in this calculation. The remaining cutoff factor χ ε 2 \chi_\varepsilon^2 χ ε 2 differs from one only on the remote tail. Extending a Taylor-polynomial integral to the full half-space adds at most
ε ℓ ∫ ∣ y ∣ > a / ε ∣ y ∣ ℓ ∣ ∇ V b ∣ ∣ ∇ V c ∣ d y ≤ C ε 2 s B s 2 , 0 ≤ ℓ < s . (480) \varepsilon^\ell\int_{|y|>a/\varepsilon}|y|^\ell|\nabla V_b|\,|\nabla V_c|\,dy\le C\varepsilon^{2s}\mathcal{B}_s^2,\qquad0\le\ell<s. \tag*{(480)} ε ℓ ∫ ∣ y ∣ > a / ε ∣ y ∣ ℓ ∣∇ V b ∣ ∣∇ V c ∣ d y ≤ C ε 2 s B s 2 , 0 ≤ ℓ < s . (480)
Indeed insert two weights ⟨ y ⟩ s \langle y\rangle^s ⟨ y ⟩ s and bound the remaining power on the tail. All polynomial products of total degree at least s s s are O ( ε s ) O(\varepsilon^s) O ( ε s ) by the finite weighted bounds. Thus the retained coefficients are exactly
∑ ℓ + b + c = r a ℓ ( V b , V c ) , (481) \sum_{\ell+b+c=r}a_\ell(V_b,V_c), \tag*{(481)} ℓ + b + c = r ∑ a ℓ ( V b , V c ) , (481)
which proves (C.12). □
Acknowledgments
This work was supported by the National Natural Science Foundation of China (Grant No. 12371422) and the Fundamental Research Funds for the Central Universities (Jilin University). The authors used ChatGPT (OpenAI) for exploratory mathematical discussions and editorial assistance during the preparation of this manuscript. The authors take full responsibility for all mathematical statements, arguments, and proofs.