Source-linked AI summary

Curvature-Domain Wireless Communications: Gauge-Fixed Signal Spaces, Fredholm Capacity, and Differentiation-Limited Scaling for Continuous Apertures

Yasser Al Eryani

arXiv:2609.28363v1eess.SPeess.SY

TL;DR

Continuous-aperture wireless signaling needs a signal-space account that handles extended apertures and affine piston-and-tilt phase freedom. This paper uses curvature as a gauge-fixed coordinate to formulate compact synthesis and capacity, and finds that receiver differentiation limits modal growth, with fourth-root high-SNR capacity scaling in that regime.

  • Problem

    The paper addresses how to represent continuous-aperture signals while accounting for extended apertures’ infinite-dimensional nature and affine piston-and-tilt phase freedom.

    Source: §I · p. 1, §XIV · p. 19
  • Method

    The paper quotients phase profiles by affine piston-and-tilt freedom, uses curvature as the gauge-fixed coordinate, and derives compact synthesis and capacity formulations under curvature and phase-excursion budgets.

    Source: §XIV · p. 19
  • Results

    The receiver analysis finds fourth-root high-SNR capacity scaling in the differentiation-limited regime, while diagnostics match the closed-form modal spectrum to 3.78 × 10^-15 maximum relative deviation.

    Source: §XIV · p. 20
  • Takeaways & Limitations

    Regularized curvature reconstruction is essential for practical architectures because estimating curvature from phase requires two spatial differentiations and amplifies unregularized noise quartically with sampling resolution.

    Source: §XIV · p. 20

    The receiver regularizer is analytically specified as L = D2; replacing it with a learned operator trained on measured hardware phase-noise data is deferred future work.

    Source: D. Limitations · p. 16

Abstract

from arXiv · show

We develop a curvature-domain formulation for continuous-aperture signaling, in which the transmit phase is represented through its second spatial derivative after quotienting out affine piston-and-tilt gauge freedom. The resulting gauge-fixed synthesis operator is bounded and compact, with sharp Poincare-Wirtinger constant $C_L=L^2/β_1^2$, and its modal Gram spectrum is available in closed form, $ρ_m=(L/β_m)^4$, with $β_m$ the roots of $\cosβ\coshβ=1$. Under a bounded-support square-integrable propagation kernel the tangent operator is Hilbert-Schmidt, so the infinite-dimensional capacity is a well-defined Fredholm-determinant supremum. The optimal signaling law is a dual-budget generalized water-filling with one Lagrange multiplier for curvature power and one for phase excursion. From the exact nonlinear phase-only aperture law we derive the coherent tangent channel with an explicit Frechet remainder bound and a multi-chart atlas for large excursions. At the receiver, curvature inferred from noisy phase samples by second differences has a pentadiagonal noise covariance with spectral norm $Θ(Δx^{-4})$. A deterministic diagnostic suite measures each mechanism against its closed form, with a null run beside every claim: the computed spectrum matches $(L/β_m)^4$ to relative error $3.78\times10^{-15}$; the tangent remainder has fitted slope 1.0000; the dual-budget law is solved with both multipliers strictly active to a KKT residual of $3.90\times10^{-15}$; the derivative-noise bound is approached to 0.999981; and the differentiation-limited branch is observed at exponent 0.2175, then collapses to 0.0576 once the mode count saturates at the Shannon number, while a flat-propagation null run holds at 0.2019. Curvature is thus a well-posed, gauge-invariant coordinate, and below the Shannon number the gauge, rather than the medium, governs the scaling.

I. INTRODUCTION … B. Gauge-Fixed Synthesis Operator

The paper treats affine-invariant phase curvature as a complete signaling coordinate for continuous apertures and proves that synthesis from curvature is bounded and compact with a sharp norm. It places this coordinate within classical propagation and derives associated tangent-channel, spectral, capacity, and receiver-noise results.

I. INTRODUCTION; B. Main Contributions

  • The exact phase-only aperture law yields a Fréchet tangent channel with an explicit O(ϵ) remainder and a multi-chart continuation for large modulation.The contribution also gives a chart-size bound and characterizes the signed second-order cost of linearization relative to the tangent channel.Source: §I · p. 1, B. Main Contributions · p. 2

B. Main Contributions; F. Information geometry

  • The curvature modes have exact Gram eigenvalues ρ_m=(L/β_m)^4, and the capacity framework combines curvature-power and phase-excursion budgets in a two-multiplier water-filling law.The work also establishes Fredholm capacity attainment, modal convergence, and a two-dimensional gauge-fixed extension.Source: B. Main Contributions · p. 2, F. Information geometry · p. 4

B. Main Contributions; D. Inverse problems and regularized differentiation; E. Electromagnetic information theory

II. RELATED WORK AND POSITIONING; A. Classical Shannon information theory and signal-space MIMO; B. Holographic and continuous-aperture MIMO

C. Adaptive optics and wavefront theory; D. Inverse problems and regularized differentiation; E. Electromagnetic information theory

F. Information geometry; C. Notation; D. Organization

  • The curvature-manifold perspective shares operator-geometric methods with information geometry without claiming to derive ICEL from it.The paper uses real Hilbert spaces for physical phases and complex-equivalent Gaussian notation for capacity, and its sections proceed from physical modeling through spectral, capacity, estimation, scaling, and extension results.Source: C. Notation · p. 2, F. Information geometry · pp. 3–4, D. Organization · p. 3

I. INTRODUCTION; A. What This Paper Is—and Is Not; A. Aperture Geometry and Phase Gauge

B. Gauge-Fixed Synthesis Operator

  • The gauge-fixed synthesis operator D2:V→X is a bounded bijection with bounded inverse; viewed as S:X→X, it is compact, and sharpness makes C_L the operator norm of S.Sharpness holds on the minimizing eigenfunction.Source: B. Gauge-Fixed Synthesis Operator · pp. 4–5

C. Exact Nonlinear Phase-Only Manifold Channel · D. Coherent Local Linearization

The exact phase-only aperture law is nonlinear, while its coherent local channel is the Fréchet linearization around a bias phase with a deterministic remainder bound. Under the stated compactness assumptions, the tangent channel is Hilbert–Schmidt and its Fredholm determinant is finite for trace-class covariance.

C. Exact Nonlinear Phase-Only Manifold Channel

  • Bounded compact apertures and a square-integrable propagation kernel make the propagation operator Hilbert–Schmidt; without bounded receive support, the pure scalar Green kernel is only bounded absent added integrability.The assumption holds for paraxial Fresnel propagation over compact apertures at finite separation and for scalar Green kernels on compact receive domains.Source: C. Exact Nonlinear Phase-Only Manifold… · p. 5
  • Equation (9) gives the exact physical phase-only law before linearization, with uniform transmit amplitude treated as a fixed hardware property rather than a signaling variable.This convention reflects the phase-only realizability constraint of physical apertures.Source: C. Exact Nonlinear Phase-Only Manifold… · pp. 5–6
  • For a nonuniform fixed amplitude a, the tangent operator becomes T(a), and compactness, capacity, and KKT statements carry over because multiplication by a is bounded.The passage gives ∥M_ja∥ = ∥a∥L∞(Ω).Source: C. Exact Nonlinear Phase-Only Manifold… · p. 6

D. Coherent Local Linearization

  • The exact nonlinear phase-only manifold has a coherent local operator channel as its Fréchet linearization around the bias phase, with a deterministic remainder bound on bounded curvature balls.The theorem assumes bounded propagation and reconstruction operators.Source: D. Coherent Local Linearization · p. 6
  • Under Assumption 1, the tangent channel is Hilbert–Schmidt, and its trace-class covariance product has a finite Fredholm determinant.The tangent channel is compact when the propagation and reconstruction operators are bounded.Source: D. Coherent Local Linearization · p. 6
  • For L = 0.1 m, P/L = 102 V/m, ∥R∥∥P∥ = 1, and ∥ϕb∥L∞ ≤ π, the bound gives KR ≈ 1.3 × 10−3 and ϵmax ≈ 7.5 for 1% tolerance on a unit-radius curvature ball.Larger modulation ranges are handled by switching across a finite atlas of bias phases.Source: D. Coherent Local Linearization · pp. 6–7

E. Multi-Chart Continuation for Large Modulation · F. The Exact Cost of Linearization

Large excursions are handled by overlapping bias-phase charts whose switching cost can be amortized, while the exact second-order cost of linearization is determined by measurable aperture powers and can change sign. Numerical verification reports close agreement with the predicted deficit, including its sign reversal.

E. Multi-Chart Continuation for Large Modulation

  • An admissible atlas of overlapping charts covers the operational curvature range, and re-centering on the nearest centroid preserves tangent-channel tolerance δ.Each chart has radius Rmax = δ/(KRϵ); consecutive chart centers are separated by at most Rmax/2.Source: E. Multi-Chart Continuation for Large… · p. 7
  • Θ(1/(ϵNcoh)) bits per use bounds aggregate rate loss when charts are redetermined at most once per Ncoh coherent uses and ϵNcoh ≫ 1.The bound amortizes the bias-phase-index retransmission cost across the coherence block while using single-chart Fredholm capacity within each block.Source: E. Multi-Chart Continuation for Large… · p. 7
  • J ≈ 30 charts cover a 2π operational curvature range in the reference near-field configuration with a 1% tangent-channel tolerance.The per-chart validity radius is Rmax ≈ 7.5, and the rate loss is bounded by a fraction of a bit per use when switches occur less than once per 10^3 coherent uses.Source: E. Multi-Chart Continuation for Large… · p. 7

F. The Exact Cost of Linearization

  • The coefficient α ranges from (2 − N)/(4(N + 2)) to 1/(2(N + 2)); it equals 1/(4(N + 2)) when the radiated fields are mutually orthogonal.The coefficient depends on coherent received power and summed element power, rather than being a universal constant.Source: F. The Exact Cost of Linearization · p. 8
  • The exact mean deficit coefficient is α = (2 tr g − ∥s(ϕ)∥²)/(4(N + 2) tr g), and it changes sign when coherent power crosses twice the summed element power.Below that threshold the exact aperture separates signals less than the tangent model predicts; above it, the tangent channel is conservative. The coherent-gain identity gives 1ᵀg(ϕ)1 = ∥s(ϕ)∥².Source: F. The Exact Cost of Linearization · pp. 7–8
  • The deficit theorem averages over isotropic modulation directions: the cubic term can remain direction-dependent, and it does not establish that nonlinearity adds capacity missed by the tangent channel.The companion foresight essay’s Information–Curvature Efficiency Law [1] has the same functional form, but here α is fixed by the aperture Gram matrix and operating-point coherent gain and can change sign.Source: F. The Exact Cost of Linearization · p. 8
  • 0.13%–2.2% agreement with direct Monte Carlo measurements validates the predicted deficit across excursions where its coefficient changes sign and varies fourfold.The propagator is an exact spherical-wave Green’s function; the sign reversal occurs at the predicted crossing, and the geometry is independently verified to relative error 5.7 × 10^-16.Source: F. The Exact Cost of Linearization · p. 8

G. Noise Model … B. Galerkin Reduction

The section establishes an exact curvature-mode spectrum and uses it to define consistent finite Galerkin channels, under a Gaussian observation-noise model that permits whitening.

G. Noise Model

  • Assumption 2 models observation noise as proper complex Gaussian with positive covariance, permitting whitening in the standard boundedly invertible Gaussian model.Source: G. Noise Model · p. 8

A. Spectral Curvature Basis

  • Compactness of S makes Q positive, self-adjoint, and compact, with an orthonormal curvature-mode basis and positive eigenvalues tending to zero.Source: A. Spectral Curvature Basis · p. 8
  • ρ_m=(L/β_m)^4, where β_m are the positive roots of cos β cosh β=1, and the curvature modes are second derivatives of the corresponding free–free eigenfunctions.Source: A. Spectral Curvature Basis · p. 9
  • The modal spectrum has a +1/2 asymptotic offset and quartic decay; its fitted log–log slope is -3.9575 over m∈[16,128].Source: A. Spectral Curvature Basis · p. 9

B. Galerkin Reduction

  • Projection onto M transmit curvature modes and an M-dimensional receive subspace yields a finite-dimensional curvature channel, with phase-synthesis penalty diagonal in the u_m basis.Source: B. Galerkin Reduction · p. 9
  • If T is compact and the transmit and receive projections converge strongly to identity, Theorem 4 establishes Galerkin consistency and bounds the approximation error.Source: B. Galerkin Reduction · p. 9

V. FINITE-MODE CAPACITY UNDER DUAL BUDGETS … C. Commuting Case: Generalized Water-Filling

The finite-mode Gaussian channel is optimized under two quadratic budgets, with capacity attained by proper complex Gaussian signaling; when the relevant matrices commute, the optimal covariance follows generalized water-filling, alongside a rank-sensitive upper bound.

A. Feasible Covariances

B. Exact Finite-Mode Capacity

C. Commuting Case: Generalized Water-Filling

VI. INFINITE-DIMENSIONAL CAPACITY AND THE MODAL LIMIT … A. Discrete Phase Sampling

The section defines an attained Fredholm capacity for the whitened compact operator channel, derives its operator optimality and modal limits, and shows how gauge-fixed discrete phase sampling yields curvature estimates with quartically amplified noise.

A. Whitened Operator Channel

  • Under Assumption 2, the whitened channel operator and its Gram operator are compact, while admissible covariance operators are positive trace-class and obey two budgets.The budgets are tr(K) ≤ Pc and tr(QK) ≤ Pϕ.Source: A. Whitened Operator Channel · pp. 10–11

B. Fredholm Capacity Formula

C. Capacity Attainment and Operator KKT Conditions

  • The capacity supremum is attained under positive budgets and a nonzero channel, with operator KKT conditions extending finite-dimensional water-filling beyond a finite-rank artifact.The theorem assumes Pc > 0, Pϕ > 0, and T̃ ≠ 0.Source: C. Capacity Attainment and Operator… · p. 11

D. Infinite-Dimensional Generalized Water-Filling

  • When B and Q commute, an orthonormal common eigenbasis supports the infinite-dimensional generalized water-filling allocation.The theorem also specifies zero allocation whenever β_m = 0.Source: D. Infinite-Dimensional Generalized… · p. 11

E. Convergence of Modal Capacities

  • Modal capacities converge to the infinite-dimensional limit, and receive-side Galerkin capacities converge to the same limit when the receive projections converge strongly to identity.The theorem gives an exact justification for finite-mode curvature-domain channels as a convergent approximation scheme.Source: E. Convergence of Modal Capacities · p. 11

A. Discrete Phase Sampling

  • On the gauge-fixed subspace, the full-row-rank interior second-difference operator yields an unbiased curvature estimate with pentadiagonal noise covariance, whose noise grows quartically with finer sampling.The discrete null space is the sampled affine gauge; restricting to the gauge-fixed subspace makes the operator injective and removes boundary-condition ambiguity.Source: A. Discrete Phase Sampling · p. 12

B. Projected Noise Covariance … A. Useful-Mode Counting

The section links aperture-sampled noise and regularized curvature reconstruction to detection, codebook and system design, and useful-mode counting. It specifies design criteria and analytical results while distinguishing curvature coherence from received power alone.

B. Projected Noise Covariance

  • The finite-dimensional information-theoretic noise matrix is directly computable from aperture sampling geometry.The projected observation uses samples of M discrete curvature basis functions.Source: B. Projected Noise Covariance · p. 12

C. Regularized Curvature Reconstruction

  • The regularizer is the interior second-difference operator D2, yielding a discrete H2-Sobolev penalty consistent with transmit regularization and the receiver’s m^-4 modal decay.Its bias–variance decomposition depends on the regularizer’s sparsity pattern, allowing weighted or graph-Laplacian variants; the operator may also be learned without changing the stated MSE-bound form.Source: C. Regularized Curvature Reconstruction · p. 12

A. Whitened Geometric Distance

  • Proposition 9 gives the pairwise error exponent for maximum-likelihood detection under the stated whitened-noise model.The numerical diagnostic uses a real-scalar normalization, while the passage also identifies a complex-equivalent formula.Source: A. Whitened Geometric Distance · pp. 12–13

B. Finite-Codebook Packing 

  • Finite-codebook packing is shaped jointly by the propagation operator and the phase-synthesis penalty.The natural design problem is identified as equation (94).Source: B. Finite-Codebook Packing  · p. 13

C. System Design Problems

P2: Finite-codebook packing:

P3: Programmable-environment design:

  • P3’s structure-preserving programmable-environment objective measures curvature coherence rather than received power alone.The environment is parameterized by θ and yields H_M(θ).Source: P3: Programmable-environment design · p. 13

A. Useful-Mode Counting

  • Proposition 10 gives a useful-mode law for a fixed reliability threshold Γ, with differentiation-induced loss represented in the diagonal modal model.The supplied passage does not include the law’s explicit formula.Source: A. Useful-Mode Counting · p. 13

B. Differentiation-Limited Root-Law … C. The crossover

The sections distinguish the diagonal-model differentiation law from aperture capacity: total-power water-filling changes the exponent, and tangent-channel scaling follows curvature decay only below the geometry’s Shannon number. Beyond that crossover, propagation limits the active modes and the local capacity exponent tends to zero.

B. Differentiation-Limited Root-Law

  • The differentiation-limited diagonal law is Θ(Γ^1/4), with a sharp leading coefficient depending only on gauge strength β, not other problem parameters.The corollary identifies the asymptotic constants in Theorem 11’s bounds.Source: B. Differentiation-Limited Root-Law · p. 13

X. SPECTRAL COMPOSITION AND THE CURVATURE CROSSOVER

  • Theorem 11 concerns a diagonal model; applying it to an aperture requires specifying the budget variable and checking whether the model gains match the tangent channel’s gains.The resulting aperture law has exponent 1/5 under a power budget and holds only below the geometry’s Shannon number.Source: §X · p. 14

A. The exponent depends on the budget

  • Equal per-mode power gives Θ(Γ^1/4), whereas optimizing under total power P gives C(P)=Θ(P^1/5); reaching per-mode SNR Γ costs Θ(Γ^5/4) total power.The exponents describe different resource conventions, so the total-power result is the relevant high-SNR scaling when power is the budget.Source: A. The exponent depends on the budget · p. 14

B. The composition law

  • The tangent-channel singular values do not factor mode by mode; a two-sided spectral envelope shows that decay is controlled by whichever of gauge or propagation decays faster.Physical propagation has near-constant singular values through the Shannon number and a super-exponential tail beyond it, consistent with prolate-spectrum behavior,,.Source: B. The composition law · p. 14

C. The crossover

  • On the tangent channel, active gains track curvature gains up to a bounded multiplicative factor, and the capacity exponent falls from 0.23 over the first ten modes to 1/5 asymptotically.Source: C. The crossover · p. 14
  • Above the Shannon number, M(P)=NS+O(log P), capacity is Θ(NS log P + log^2 P), and the local exponent tends to zero.The plunge modes contribute the log^2 P term; the crossover power is where M(P)=NS and is computable from the gain sequence.Source: C. The crossover · pp. 14–15
  • The falsifiable prediction is that the polynomial branch appears only below the geometry-computed Shannon number and terminates there, while flat propagation does not produce that termination.A diagonal model with flat propagation remains in the polynomial branch by construction.Source: C. The crossover · p. 15
  • The m^-4 gauge decay is not a capacity ceiling: below the Shannon number it constrains rate, whereas above it the medium binds and shared signal spaces have capacity within a bounded factor.Above the crossover, curvature’s advantages must instead be argued through conditioning, realizability, and estimation cost.Source: C. The crossover · p. 15

XI. TWO-DIMENSIONAL GAUGE-FIXED EXTENSION … C. Falsifiable Predictions

The framework extends to two-dimensional apertures through Hessian-compatible curvature, with affine phase freedom removed. Its falsifiable predictions include distinct rate-scaling laws and a geometry-determined termination of the polynomial branch at the Shannon number.

XI. TWO-DIMENSIONAL GAUGE-FIXED EXTENSION

  • In two dimensions, the Hessian map is a bounded bijection with bounded inverse, and the gauge-fixed synthesis operator is compact; equal Hessians identify phases differing only by an affine function.The curvature space consists of Hessian-compatible tensors, not all symmetric L2 tensor fields; compatibility is enforced intrinsically to exclude curvature fields that cannot arise from a phase.Source: §XI · pp. 15–16
  • Finite-mode, Fredholm-capacity, attainment, KKT, Galerkin-convergence, and water-filling results extend to two dimensions by replacing the one-dimensional spaces and operators with their two-dimensional counterparts.The replacements are (X, V, S, Q, T) with (C2, V2, S2, Q2, T2).Source: §XI · p. 16

XII. DISCUSSION: WHY THE FRAMEWORK IS PHYSICALLY CONSERVATIVE

  • The framework is physically conservative: propagation remains classical and linear in the field, and the transmitter assumes only the ordinary phase-only aperture law.The curvature-domain channel is the coherent tangent model of that law.Source: §XII · p. 16, A. No New Electromagnetic Law · p. 16

A. No New Electromagnetic Law

  • The curvature-domain channel adds no new electromagnetic law: its transmitter uses the ordinary phase-only aperture field, and its channel is the coherent tangent model.Propagation remains classical and linear in the field.Source: A. No New Electromagnetic Law · p. 16

B. No Degrees of Freedom Created Ex Nihilo

  • Curvature-domain coding does not exceed full aperture-field channel capacity under the same physical constraints; it supplies a geometric coordinate for ordinary phase-coded apertures.Insufficient sensing resolution to estimate curvature reliably is predicted by Theorem 10.Source: B. No Degrees of Freedom Created Ex… · p. 16

C. Falsifiable Predictions

  • Without regularization, second-difference curvature-estimation noise scales as Δx^-4; programmable environments can improve curvature coherence even when received power changes modestly.The noise scaling is a consistency check on the receiver model, not a discriminating test of the curvature coordinate; both are measurement-level predictions testable without nonstandard physics.Source: C. Falsifiable Predictions · p. 16
  • Useful curvature modes follow a quarterpower law in per-mode SNR while propagation supplies modes; under differentiation-limited sensing, rate scales as Θ(Γ^1/4) or Θ(P^1/5) under a total-power budget.The per-mode-SNR and total-power gain laws must not be interchanged.Source: C. Falsifiable Predictions · p. 16
  • The polynomial branch exists only below the geometry-determined Shannon number NS = LtLr/(λd) and terminates there, with the local exponent falling from ≈1/5 to 0 as active modes cross NS.The cutoff is a discriminating prediction with no free parameter (Theorem 13).Source: C. Falsifiable Predictions · p. 16

D. Limitations … XIV. CONCLUSION

The paper’s diagnostics test its curvature-domain theory, from the closed-form modal spectrum through receiver noise and scaling limits. They support the reported mechanisms while leaving calibrated hardware and several electromagnetic extensions for future work.

D. Limitations; H. What These Diagnostics Do and Do Not Establish

  • The diagnostics do not establish calibrated-device performance; practical extensions include hardware-in-the-loop evaluation, learned regularization, wideband propagation, and vector electromagnetic modeling.The present diagnostics use a scalar paraxial propagation model; two-dimensional numerical support and effects such as polarization, mutual coupling, phase resolution, and calibration error remain outside the reported results.Source: D. Limitations · p. 16, H. What These Diagnostics Do and Do… · p. 19, §XIV · p. 20

XIII. NUMERICAL DIAGNOSTICS AND REPRODUCIBILITY; A. Discretisation and its verification

  • The shifted-Legendre discretization deletes the two coefficients spanning the affine gauge and constructs the synthesis operator by exact integration, avoiding a projector, pseudo-inverse, boundary stencil, and quadrature.The section reports deterministic diagnostics computed from the operators in the theorems, without fitting spectra or solving backward for target values.Source: A. Discretisation and its verification · p. 17, §XIII · p. 17

B. Modal Spectrum Against Its Closed Form

  • 3.78 × 10^-15 is the maximum relative deviation between the computed modal spectrum and ρ_m = (L/β_m)^4 over 128 modes.The computed Poincare–Wirtinger constant is 0.044696, agreeing with L^2/β_1^2 to relative error 6.21 × 10^-16; the Dirichlet sine null overstates C_L by a factor of 2.26689.Source: B. Modal Spectrum Against Its Closed… · p. 17, Figure 1 · p. 17

C. Tangent-Channel Remainder; D. Galerkin Truncation

E. Dual-Budget Generalized Water-Filling

  • 3.90 × 10^-15 is the maximum measured KKT violation for dual-budget water-filling with both multipliers strictly positive.At the tightest excursion budget, the rate given up is 9.69%; at equal capacity, the classical single-budget allocation spends 4.342 times as much phase excursion.Source: E. Dual-Budget Generalized Water-Filling · p. 18, Figure 4 · p. 18

F. Derivative-Noise Amplification; XIV. CONCLUSION

  • 4.0670 is the fitted sample-count exponent for derivative-noise amplification, whose measured norm approaches the analytic bound to 0.999981123 from below.The diagnostic supports the conclusion that curvature estimation’s two spatial differentiations amplify unregularized curvature noise quartically with sampling resolution.Source: F. Derivative-Noise Amplification · p. 18, Figure 5 · p. 19, §XIV · p. 20

G. The Two Scaling Exponents, and the Crossover

  • 0.2175 is the physical-kernel local exponent while active modes lie between 8 and NS; it falls to at most 0.0576 at saturation, versus 0.2019 for the flat-propagation null.The per-mode SNR exponent is 0.2520 and the total-power water-filling exponent is 0.2045; the crossover is observed at NS = 32.0000.Source: G. The Two Scaling Exponents, and the… · p. 19, Figure 6 · p. 19

XIV. CONCLUSION

  • The framework uses curvature as a complete coordinate for phase profiles modulo affine piston-and-tilt freedom, yielding compact synthesis and finite- and infinite-dimensional capacity formulations.The infinite-dimensional formulation uses a Fredholm determinant, admits an optimal covariance, and is recovered as a monotone limit of modal truncations.Source: §XIV · pp. 19–20

APPENDIX A PROOF OF PROPOSITION 1 AND THEOREM 1 … APPENDIX D PROOF OF THEOREM 5, PROPOSITION 6, COROLLARY 3,

The appendices establish bijectivity and compactness for the curvature-coordinate synthesis map, then justify modal approximation and finite-dimensional covariance optimization using compact-operator and KKT arguments.

APPENDIX A PROOF OF PROPOSITION 1 AND THEOREM 1

  • The curvature map is bijective: zero curvature corresponds to the affine gauge, and each prescribed curvature has a representative in V.The inverse synthesis operator is bounded and compact, using the open mapping theorem and the compact embedding H2(Ω) into L2(Ω).Source: Appendix A · p. 20

APPENDIX B PROOF OF THEOREM 2 AND COROLLARY 1

  • The composition of compact synthesis with bounded phase, propagation, and receiver operators is compact; square-integrable propagation kernels are Hilbert–Schmidt.Hilbert–Schmidt operators are compact.Source: Appendix B · p. 21

Proof of Proposition 3

  • The compact, self-adjoint, positive operator Q has an orthonormal eigenbasis with positive eigenvalues decreasing to zero.Its kernel is trivial because the curvature map is onto.Source: Proof of Proposition 3 · p. 21
  • The approximation error splits into output- and input-projection terms, which vanish under compactness and strong convergence of the projections.The passage also gives an operator-norm error bound and a covariance formula.Source: Proof of Proposition 3 · p. 21

APPENDIX D PROOF OF THEOREM 5, PROPOSITION 6, COROLLARY 3,

  • The finite-dimensional covariance optimization attains a maximum because its feasible set is compact and its continuous, concave objective is maximized over that set.After whitening, Gaussian input maximizes differential entropy for fixed covariance.Source: Appendix D · p. 21
  • Lagrangian stationarity yields the stated optimality relations, while joint diagonalization reduces the problem to scalar KKT conditions with equality on active modes.The scalar conditions give the stated expression for Corollary 3.Source: Appendix D · p. 21
  • Feasibility and the rank constraint, combined with Jensen’s inequality, establish the upper bound in (58).The proof concludes by combining the derived inequalities.Source: Appendix D · p. 21

APPENDIX E PROOF OF THEOREMS 6, 7, 8, AND 9 … APPENDIX F PROOF OF THEOREM 10 AND PROPOSITION 8

The appendices establish that the Fredholm capacity is finite and attained, derive its diagonal water-filling structure and truncation behavior, and prove error properties for second-difference curvature estimation.

APPENDIX E PROOF OF THEOREMS 6, 7, 8, AND 9

  • For positive trace-class inputs, the Fredholm determinant is well defined and the capacity is finite by the log-determinant trace bound.The proof uses positivity and boundedness of the synthesis operator, then bounds log2 det_F(I + A) by tr(A)/ln 2 for A ⪰ 0.Source: Appendix E · p. 22
  • Finite-rank spectral truncations converge in trace norm, allowing the finite-dimensional mutual-information expression to pass to the limit and yield the stated supremum.The operator images also converge in trace norm before the limit is taken.Source: Appendix E · p. 22

Proof of Theorem 7

  • The feasible trace-class covariance set yields an attained capacity supremum, and concavity and differentiability give KKT conditions with nonnegative multipliers and a positive normal-cone operator.Complementary slackness supplies the stated KKT relations.Source: Proof of Theorem 7 · p. 22

Proof of Theorem 8

  • An optimal covariance can be chosen diagonal in the common eigenbasis of B and Q, reducing the problem to scalar optimization whose KKT conditions give the stated law.The determinant inequality becomes equality when the covariance matrix is diagonal in that basis.Source: Proof of Theorem 8 · p. 22

Proof of Theorem 9

  • The truncated capacities C_M increase monotonically and are bounded above by C_∞, so their limit exists and is at most C_∞.Source: Proof of Theorem 9 · p. 22

APPENDIX F PROOF OF THEOREM 10 AND PROPOSITION 8

  • The second-difference estimator is unbiased, and its covariance follows from shifted [1, −2, 1] stencils with spectral norm bounded by 16σ_ϕ^2/Δx^4.Non-overlapping stencil rows have zero covariance overlap.Source: Appendix F · p. 23
  • The regularized estimator’s normal equations establish its bias and covariance expressions, whose squared-bias and variance terms yield the stated mean-squared-error result.Source: Appendix F · p. 23

APPENDIX G PROOF OF PROPOSITION 9 … APPENDIX I PROOF OF THEOREM 14

These appendices establish the stated exponential bound, derive the threshold and integral asymptotics, and prove that the gauge-fixed Hessian map is invertible with compact synthesis into L2.

APPENDIX G PROOF OF PROPOSITION 9

  • The proof establishes (92) and the exponential bound (93).This closes the proof of Proposition 9.Source: Appendix G · p. 23

APPENDIX H PROOF OF PROPOSITION 10, THEOREM 11, AND

  • The passage states that solving for m yields Equation (99), with a positive-floor convention accounting for a case involving αp/Γ − σ².Source: Appendix H · p. 23
  • The proof establishes a positive constant c1(β) bounding C(Γ)/Γ^1/4 below globally.It combines a sufficiently-large-Γ bound with a positive minimum on the remaining compact interval [1, Γ0].Source: Appendix H · p. 24

Proof of Corollary 4

  • The Corollary 4 proof evaluates the limiting integral and uses it to derive asymptotic result (103).It applies convergence to the rescaled integrand and evaluates the remaining integral using the n=4 case of the stated integral formula.Source: Proof of Corollary 4 · p. 24

APPENDIX I PROOF OF THEOREM 14

  • On the gauge-fixed space V2, the Hessian map ∇2 is injective and surjective onto C2.Zero Hessian implies an affine function, and orthogonality to ψ0, ψ1, ψ2 removes its affine components; surjectivity follows from the definition of C2.Source: Appendix I · p. 24
  • The inverse S2 is bounded and compact into L2(Ω2), and equality of Hessians identifies functions up to an affine difference.Boundedness follows from the second-order Poincare–Bramble–Hilbert inequality, while compactness uses the Rellich embedding.Source: Appendix I · p. 24
Loading 2609.28363v1…