Not R Kurvature: Beating Large-Scale White Noise

The paper Not R Kurvature: Beating Large-Scale White Noise by Wayne Hu investigates the claim that kurvature, a curvature invariant, generically acquires superhorizon (large-scale) white noise from hard-hard momentum mode coupling, and that this white noise leads to an infrared-divergent and ultraviolet-sensitive contribution to the observable cosmological curvature perturbation R.

The paper's central finding is that while kurvature does acquire large-scale white noise, this white noise does not translate into an infrared divergence in the curvature perturbation R, contradicting the BIS-II conjecture. The abstract states: Kurvature is not an intrinsic 3-curvature: on comoving slices, it contains extrinsic-curvature terms, and the intrinsic curvature itself is not related to R by a Poisson equation beyond linear order.

The paper performs a second-order perturbation theory calculation in radiation domination to test the Poisson relation (called the BIS relation) between kurvature and curvature. The key results are:

1. Kurvature white noise is real but extrinsic: The paper verifies that quadratic hard-hard composites do indeed cause large-scale white noise in the kurvature density, but the Hamiltonian constraint separates these contributions into intrinsic curvature and extrinsic shear (or alternately density and extrinsic expansion). The growing contribution to the dimensionless kurvature density δK comes exclusively from the extrinsic terms. Specifically, the paper states: "The growing contribution to the fractional kurvature density δK, which mimics that of the matter density δ above the horizon and provides its LSWN, comes exclusively from the shear of nearly antiparallel acoustic beat modes."

2. No infrared divergence in curvature: The hard-hard modes produce "negligible and ultraviolet-convergent contributions to the curvature power spectrum, leaving no IR relic in R from purely ultraviolet modes, with dominant contributions from the horizon-sized modes at the epoch of evaluation. The paper explicitly shows that the Laplacian of the second-order curvature perturbation has identically zero white contribution: ∇2R(2) LSWN = 0."

3. The BIS-II conjecture fails: The paper states: By contrast, the Poisson construction of a conjectured curvature potential from kurvature is indeed infrared divergent and ultraviolet cutoff sensitive. The kurvature white noise employed in that construction arises from the extrinsic curvature associated with acoustic beat modes of a radiation fluid.

The paper provides detailed calculations showing that the second-order curvature response R2 is sourced by an acoustic propagation equation rather than a Poisson-like constraint. The master equation (Eq. 40) shows that the source is regular in the soft limit, and the solution for R2 has no inverse powers of the soft wavenumber. The paper states: This already establishes the LSWN kurvature mechanism at second order: the composite first-order shear modes of high wavenumber beat couple to a low wavenumber in their quadratic combination.

The paper also shows that the intrinsic 3-curvature contribution to kurvature is bounded and does not grow as a2, while the shear and expansion terms grow as x2 (where x is acoustic time). The paper explains: "The intrinsic curvature and comoving density contribute much smaller bounded terms to δK. Moreover the relativistic constraint system keeps even this small but white density term from becoming the kL−2 curvature branch that the BIS-II conjecture implies."

The power spectrum calculations show that the hard-hard contribution to the curvature power spectrum ∆222 is strongly IR convergent, scaling as (kη)3 on superhorizon scales, with the dominant contribution coming from modes that cross the sound horizon at the evaluation epoch (x ≈ 2.51). In contrast, the BIS-II inferred spectrum ∆2 BIS is UV sensitive and IR divergent, scaling as q max/k.

The paper concludes: "We have shown by explicit calculation of the second-order evolution of acoustic waves during radiation domination: where LSWN lives and where it does not, and what contributes to kurvature and what does not. Beyond linear theory, kurvature contains both intrinsic 3-curvature and extrinsic curvature through the quadratic composite of first-order shear."

The paper includes several appendices providing additional checks: Appendix A develops the extrinsic-curvature interpretation and directly evaluates the expansion contribution; Appendix B details the curvature and density multipoles; Appendix C gives an independent synchronous-gauge construction; and Appendix D reconciles the comoving curvature result with the exact Langlois–Vernizzi curvature conservation law.

Improvements to AI Systems Based on This Paper:

1. Physics-constrained generative models for cosmological perturbations: Train AI to distinguish between intrinsic geometric quantities (e.g., 3-curvature) and extrinsic ones (e.g., shear, expansion) when generating or predicting higher-order perturbation fields. The AI should learn that naive Poisson-like reconstructions (BIS relation) fail beyond linear order, preventing hallucinated infrared divergences in curvature power spectra.

2. IR/UV divergence-aware simulation surrogates: Build AI emulators for second-order cosmological perturbation theory that explicitly enforce the paper’s finding that hard-hard mode coupling yields no IR divergence in the curvature perturbation (∇2R(2) LSWN = 0). The AI can be trained to output power spectra that scale as (kη)3 on superhorizon scales, rejecting any spurious q max/k divergence from extrinsic-curvature white noise.

3. Hierarchical mode-coupling classifiers: Develop an AI system that automatically classifies quadratic composite terms (e.g., shear vs. density vs. expansion) in arbitrary field theories, flagging which composites contribute to growing, white-noise, or bounded behavior. This prevents AI from misattributing extrinsic shear noise to intrinsic curvature—a core error the paper corrects.

4. Gauge-invariant feature extraction for inference: Improve AI-based parameter inference from large-scale structure by teaching it to use comoving-slice curvature rather than kurvature as the observable. The AI should learn to ignore extrinsic-curvature white noise when reconstructing primordial non-Gaussianity, avoiding false detections from acoustic beat modes.

5. Conservative AI for theoretical conjecture testing: Implement a meta-reasoning module that, given a conjecture (e.g., BIS-II), automatically checks whether the proposed relation holds beyond linear order by testing for (a) Poisson-equation validity, (b) gauge dependence, and (c) UV sensitivity. This prevents AI from propagating conjectures that fail under explicit second-order calculations, as demonstrated here.

6. Acoustic-time-aware neural solvers: Train neural PDE solvers to incorporate the acoustic time variable x (η) and the master equation (Eq. 40) for curvature response, ensuring that soft-limit regularity is preserved. The AI can then accurately propagate second-order curvature without spurious inverse-wavenumber growth, matching the paper’s result that sources are regular in the soft limit.

7. White-noise source separation in data pipelines: For AI analyzing CMB or LSS data, add a preprocessing layer that separates intrinsic curvature signals from extrinsic shear-induced white noise using the paper’s decomposition (intrinsic curvature vs. shear vs. expansion). This improves signal-to-noise for primordial curvature detection by filtering out the acoustic beat-mode contamination.

8. Automated cross-check against conservation laws: Enhance AI verification systems to automatically reconcile any computed curvature perturbation with exact conservation laws (e.g., Langlois–Vernizzi) as done in Appendix D. This catches AI-generated solutions that violate gauge-invariant conservation, ensuring physical consistency in generated cosmological scenarios.

Kurvature, a recently identified curvature invariant, has been argued to acquire superhorizon, or large-scale, white noise from hard-hard momentum coupling even when matter nonlinearities are small. If kurvature were related directly to cosmological curvature perturbations R through a Poisson equation, this white noise would cause an infrared-divergent variance sensitive to ultraviolet hard-mode physics. However, this relation does not generically hold. Kurvature is not intrinsic 3-curvature: on comoving slices it contains extrinsic-curvature terms, and intrinsic curvature is not related to R by a Poisson equation beyond linear order. We test this inference with second-order perturbation theory in radiation domination, relevant to CMB observables. Quadratic hard-hard composites do generate large-scale white noise in the kurvature density, but the Hamiltonian constraint separates it into intrinsic curvature and extrinsic shear, or equivalently density and expansion. Only the extrinsic terms carry the growing dimensionless kurvature density that mimics an ordinary density fluctuation above the horizon. The direct hard-hard curvature power is ultraviolet convergent, dominated by horizon-scale modes at evaluation, and leaves no IR relic in R from purely ultraviolet modes. By contrast, the Poisson construction of a curvature potential from kurvature is infrared divergent and cutoff sensitive; its white noise arises from extrinsic curvature associated with nonlinear acoustic beat modes in a radiation fluid.