arxiv: 2605.02453 · v1 · submitted 2026-05-04 · 🌀 gr-qc · astro-ph.HE· cs.LG· physics.data-an· stat.ML

Recognition: 3 theorem links

· Lean Theorem

Testing General Relativity Through Gravitational Wave Classification: A Convolutional Neural Network Framework

Lavinia Heisenberg , Shayan Hemmatyar , Hector Villarrubia-Rojo

Authors on Pith no claims yet

Pith reviewed 2026-05-08 18:45 UTC · model grok-4.3

classification 🌀 gr-qc astro-ph.HEcs.LGphysics.data-anstat.ML

keywords gravitational wavestests of general relativitymachine learningconvolutional neural networksbinary black hole mergersresponse functionsmassive gravityparameterized post-Einsteinian formalism

0 comments

The pith

Convolutional neural networks using response functions classify gravitational wave signals to test general relativity with 33 times higher sensitivity than raw waveforms.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper develops a machine learning method to check whether gravitational waves from black hole mergers follow general relativity or show modifications. It uses parameters from 173 real events in the GWTC catalog to create realistic simulated signals and then builds modified versions by applying controlled changes to the wave phase. A response function derived from waveform mismatch serves as input to convolutional neural networks, isolating phase effects rather than feeding the full signal. This input choice raises classification performance by a factor of about 33 over standard whitened waveforms. The same setup applied to massive gravity models detects effects from graviton masses around 10^{-23} eV/c^{2} at aLIGO design sensitivity.

Core claim

We introduce a response function formalism that provides a systematic framework for quantifying how any observable responds to modifications of GR. We train convolutional neural networks on two input representations: whitened waveforms and a response function type observable derived from the waveform mismatch, which isolates the effect of phase deviations from the bulk signal. Using response functions as the CNN input improves the classification sensitivity by a factor of approximately 33 compared to whitened waveforms. We extend the framework to physically motivated theories using the parameterized post Einsteinian formalism and apply it to massive gravity, where the classifier detects devi

What carries the argument

The response function, a quantity derived from waveform mismatch that isolates phase deviations from the overall gravitational wave signal and supplies the CNN input.

If this is right

The CNN outperforms the best single-feature classifier at every deformation scale tested.
The framework extends directly to other modified gravity theories through the parameterized post-Einsteinian formalism.
Bayes optimal error analysis sets the fundamental performance limit for the classification task.
Averaging techniques reveal coherent patterns in the noise that aid detection.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The response function approach could be applied to real LIGO-Virgo-KAGRA data to place new bounds on alternative gravity models.
Future detectors with improved sensitivity could adopt similar representations to probe even smaller deviations from GR.
Data representation choices may prove as decisive as network architecture in other scientific machine learning tasks involving noisy time series.

Load-bearing premise

Controlled phase deformations applied to GR waveforms accurately model the signatures produced by physically motivated alternative theories of gravity.

What would settle it

A test set of independent simulated signals where phase deformations do not correspond to consistent beyond-GR theories shows no sensitivity gain when response functions replace whitened waveforms as CNN input.

read the original abstract

We present a machine learning framework for testing general relativity (GR) with gravitational wave signals from binary black hole mergers. Using the source parameters of 173 BBH events from the GWTC catalog as a realistic astrophysical population, we generate simulated GR waveforms and construct beyond GR (BGR) waveforms by applying controlled phase deformations. We introduce a response function formalism that provides a systematic framework for quantifying how any observable responds to modifications of GR. We train convolutional neural networks (CNNs) on two input representations: whitened waveforms and a response function type observable derived from the waveform mismatch, which isolates the effect of phase deviations from the bulk signal. Using response functions as the CNN input improves the classification sensitivity by a factor of approximately 33 compared to whitened waveforms, demonstrating that the choice of observable representation is as important as the classifier architecture. We study the fundamental limits of this classification through Bayes optimal error analysis, averaging methods that reveal coherent patterns hidden in noise, and a comparison between CNN accuracy and a single feature classifier as a proxy for human performance. At all deformation scales, the CNN outperforms the best single feature approach. We extend the framework to physically motivated theories using the parameterized post Einsteinian (ppE) formalism and apply it to massive gravity, where the classifier detects deviations for graviton masses of order $m_g \sim 10^{-23}\;\mathrm{eV}/c^2$ with aLIGO design sensitivity.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Response-function input gives CNNs a 33x sensitivity boost on controlled phase deformations in simulated GW signals, but the paper does not check whether that advantage carries over to the actual frequency-dependent shifts in massive gravity.

read the letter

The paper's central move is to replace raw whitened waveforms with a response function built from the mismatch between GR and deformed signals, then feed that to a CNN for classifying binary black hole events as GR or beyond-GR. On waveforms drawn from the 173 GWTC events and deformed by controlled phase shifts, this input improves classification sensitivity by roughly a factor of 33 over the standard whitened case. They also run Bayes-optimal error calculations, averaging tests, and a single-feature baseline, all of which the CNN beats at every deformation scale. The extension to the ppE parametrization for massive gravity then claims detectable deviations at graviton masses around 10^{-23} eV/c² with aLIGO design sensitivity. Those quantitative comparisons and the use of catalog parameters are the parts that hold up cleanly. The soft spot is exactly where the stress-test note flags it. The large performance gain and the claim that observable choice matters as much as architecture are shown only on the artificial controlled deformations. The paper switches to the ppE form for the massive-gravity result but does not report a parallel test confirming that the response-function representation still delivers comparable gains when the phase modification comes from the actual massive-gravity dispersion relation accumulated over luminosity distance. Without that check, the physically motivated claim rests on an unverified modeling assumption. This work sits at the ML-plus-GW-fundamental-physics intersection. It has enough new formalism and concrete benchmarks to merit a serious referee, even if the generalization step needs tightening. I would send it to peer review.

Referee Report

2 major / 2 minor

Summary. The paper presents a machine-learning framework for testing GR using CNNs to classify binary black hole gravitational-wave signals as GR or beyond-GR. Simulated waveforms are generated from the 173 GWTC BBH events; BGR versions are created via controlled phase deformations. A response-function observable is defined from the waveform mismatch to isolate phase deviations. CNNs trained on response functions achieve approximately 33 times higher classification sensitivity than those trained on whitened waveforms. The work includes Bayes-optimal error analysis, comparisons to single-feature classifiers, and an extension to the ppE formalism applied to massive gravity, claiming detectable deviations for graviton masses m_g ∼ 10^{-23} eV/c² at aLIGO design sensitivity.

Significance. If the central results hold, the work usefully demonstrates that input representation can be as consequential as network architecture for GW classification tasks and supplies a concrete, quantitative benchmark (the factor-of-33 gain) together with a comparison to the Bayes-optimal limit. The use of a realistic astrophysical population drawn from GWTC parameters and the systematic inclusion of both artificial deformations and the ppE parametrization are strengths. The framework is falsifiable in principle once real data are analyzed, and the explicit comparison to a single-feature baseline provides a useful proxy for human-level performance.

major comments (2)

[Abstract] Abstract and the section describing the massive-gravity application: the factor-of-33 sensitivity improvement and the associated claim that 'the choice of observable representation is as important as the classifier architecture' are demonstrated exclusively on controlled phase deformations. No quantitative verification is reported that the same response-function representation produces comparable gains when the phase shift is generated by the actual massive-gravity dispersion relation (frequency-dependent propagation delay accumulated over luminosity distance) under the ppE parametrization.
[Abstract] The central claim that the classifier 'detects deviations for graviton masses of order m_g ∼ 10^{-23} eV/c²' therefore rests on an unverified modeling assumption that the performance gains observed for artificial deformations transfer to the theory-specific waveform modification; this assumption is load-bearing for the physically motivated application.

minor comments (2)

The precise mathematical definition of the response function (how the mismatch is normalized, windowed, and formatted as CNN input) should be stated explicitly with an equation, as the current description leaves the construction ambiguous for reproduction.
Clarify whether the reported 33× factor is an average over the 173 events or a best-case value, and include error bars or a distribution of the improvement across the catalog.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment of the work's significance and for the constructive comments on the transferability of our results. We address each major comment below and will incorporate explicit verification in the revised manuscript.

read point-by-point responses

Referee: [Abstract] Abstract and the section describing the massive-gravity application: the factor-of-33 sensitivity improvement and the associated claim that 'the choice of observable representation is as important as the classifier architecture' are demonstrated exclusively on controlled phase deformations. No quantitative verification is reported that the same response-function representation produces comparable gains when the phase shift is generated by the actual massive-gravity dispersion relation (frequency-dependent propagation delay accumulated over luminosity distance) under the ppE parametrization.

Authors: We agree that the reported factor-of-33 improvement is quantified exclusively for controlled phase deformations. The response function is constructed from the normalized waveform mismatch and is designed to isolate phase deviations irrespective of their physical origin. In the massive-gravity section we generate waveforms using the ppE parametrization with the frequency-dependent dispersion relation appropriate to a massive graviton and feed the resulting response functions into the CNN trained on the deformation ensemble. The detection threshold is obtained from this procedure. To supply the requested quantitative verification, the revised manuscript will contain a new subsection that recomputes classification accuracy and sensitivity gain using response functions versus whitened waveforms for the specific massive-gravity phase shifts, thereby demonstrating that the performance advantage persists under the theory-specific modification. revision: yes
Referee: [Abstract] The central claim that the classifier 'detects deviations for graviton masses of order m_g ∼ 10^{-23} eV/c²' therefore rests on an unverified modeling assumption that the performance gains observed for artificial deformations transfer to the theory-specific waveform modification; this assumption is load-bearing for the physically motivated application.

Authors: The m_g ∼ 10^{-23} eV/c² threshold is obtained by applying the response-function CNN directly to ppE-modified waveforms that incorporate the massive-gravity dispersion relation. While the numerical factor of 33 was not recomputed for this specific case, the underlying mechanism—enhanced sensitivity to accumulated phase shifts—remains the same. We acknowledge that an explicit side-by-side comparison for the ppE waveforms would remove any reliance on transferability. The revision will therefore include this direct comparison, confirming that the response-function representation yields a comparable sensitivity gain when the phase modification is generated by the massive-gravity model. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper generates GR waveforms from GWTC parameters, applies controlled phase deformations to produce BGR versions, constructs a response function from the mismatch between each pair, and trains CNNs to classify the two classes. The reported 33x sensitivity gain, Bayes-optimal comparison, and single-feature baseline are all direct empirical outcomes on this simulated dataset. The subsequent application to massive gravity uses the ppE parametrization on the same trained classifier without re-deriving or fitting the response function to the target theory's dispersion relation. No equation or claim reduces a prediction to its own input by construction, no self-citation chain is load-bearing, and the response-function representation is an independently defined observable rather than a tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

Review based solely on abstract; limited details available on parameters or assumptions. The framework relies on standard GR waveform generation and mismatch calculations as background.

axioms (2)

domain assumption General relativity waveforms can be accurately simulated using standard methods from source parameters.
Invoked when generating GR waveforms from GWTC catalog parameters.
ad hoc to paper Controlled phase deformations represent possible beyond-GR effects.
Used to construct BGR waveforms for training and testing.

invented entities (1)

Response function observable no independent evidence
purpose: To isolate the effect of phase deviations from the bulk signal for improved classification.
Introduced as a new formalism in the paper.

pith-pipeline@v0.9.0 · 5572 in / 1597 out tokens · 49280 ms · 2026-05-08T18:45:52.926729+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Unification.SpacetimeEmergence / Foundation.RealityFromDistinction reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We introduce a response function formalism that provides a systematic framework for quantifying how any observable responds to modifications of GR. ... R(f) = 4i h22(f)/(∥h∥ S_n(f)) [−ŝ*(f) + M ĥ*(f)]
Constants (c, ℏ, G as φ-powers); Gravity zero-parameter certificate reality_from_one_distinction (G = φ-power) unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

the classifier detects deviations for graviton masses of order m_g ∼ 10^{-23} eV/c² with aLIGO design sensitivity
Cost.FunctionalEquation washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

δψ_22(f) = exp[−(f−50)²/100] ... a deliberately simple, localized deformation used only to benchmark the pipeline

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

36 extracted references · 31 canonical work pages · 1 internal anchor

[1]

Fundamental Theoretical Bias in Gravitational Wave Astrophysics and the Parameterized Post-Einsteinian Framework

N. Yunes and F. Pretorius,Fundamental Theoretical Bias in Gravitational Wave Astrophysics and the Parameterized Post-Einsteinian Framework,Phys. Rev. D80(2009) 122003 [0909.3328]

work page Pith review arXiv 2009
[2]

A systematic approach to generalisations of General Relativity and their cosmological implications,

L. Heisenberg,A systematic approach to generalisations of General Relativity and their cosmological implications,Phys. Rept.796(2019) 1 [1807.01725]. [8]LIGO Scientific, Virgocollaboration,Tests of General Relativity with GW150914,Phys. Rev. Lett.116(2016) 221101 [1602.03841]

work page arXiv 2019
[3]

Tahura and K

S. Tahura and K. Yagi,Parameterized Post-Einsteinian Gravitational Waveforms in Various Modified Theories of Gravity,Phys. Rev. D98(2018) 084042 [1809.00259]

work page arXiv 2018
[4]

Maggio, H

E. Maggio, H.O. Silva, A. Buonanno and A. Ghosh,Tests of general relativity in the nonlinear regime: A parametrized plunge-merger-ringdown gravitational waveform model, Phys. Rev. D108(2023) 024043 [2212.09655]. – 35 –

work page arXiv 2023
[5]

Pompili, E

L. Pompili, E. Maggio, H.O. Silva and A. Buonanno,Parametrized spin-precessing inspiral-merger-ringdown waveform model for tests of general relativity,Phys. Rev. D111 (2025) 124040 [2504.10130]

work page arXiv 2025
[6]

Piarulli, S

M. Piarulli, S. Marsat, E.M. S¨ anger, A. Buonanno, J. Steinhoff and N. Tamanini, Parametrized test of general relativity for LISA massive black hole binary inspirals,Phys. Rev. D112(2025) 124044 [2510.06330]

work page arXiv 2025
[7]

Cuoco et al.,Enhancing Gravitational-Wave Science with Machine Learning,Mach

E. Cuoco et al.,Enhancing Gravitational-Wave Science with Machine Learning,Mach. Learn.: Sci. Technol.2(2021) 011002 [2005.03745]

work page arXiv 2021
[8]

Huerta et al.,Accelerated, Scalable and Reproducible AI-driven Gravitational Wave Detection,Nature Astron.5(2021) 1062 [2012.08545]

E.A. Huerta et al.,Accelerated, Scalable and Reproducible AI-driven Gravitational Wave Detection,Nature Astron.5(2021) 1062 [2012.08545]

work page arXiv 2021
[9]

Xie et al.,Neural Post-Einsteinian Framework for Efficient Theory-Agnostic Tests of General Relativity with Gravitational Waves,Phys

Y. Xie et al.,Neural Post-Einsteinian Framework for Efficient Theory-Agnostic Tests of General Relativity with Gravitational Waves,Phys. Rev. D110(2024) 024036 [2403.18936]

work page arXiv 2024
[10]

Heisenberg, H

L. Heisenberg, H. Villarrubia-Rojo and J. Zosso,Simultaneously solving theH 0 andσ 8 tensions with late dark energy,Phys. Dark Univ.39(2023) 101163 [2201.11623]

work page arXiv 2023
[11]

Heisenberg, H

L. Heisenberg, H. Villarrubia-Rojo and J. Zosso,Can late-time extensions solve theH 0 and σ8 tensions?,Phys. Rev. D106(2022) 043503 [2202.01202]

work page arXiv 2022
[12]

doi: 10.1007/978-1-4612-0711-5

L. Devroye, L. Gy¨ orfi and G. Lugosi,A Probabilistic Theory of Pattern Recognition, vol. 31 ofStochastic Modelling and Applied Probability, Springer (1996), 10.1007/978-1-4612-0711-5

work page doi:10.1007/978-1-4612-0711-5 1996
[13]

Bishop,Pattern Recognition and Machine Learning, Springer (2006)

C.M. Bishop,Pattern Recognition and Machine Learning, Springer (2006)

2006
[14]

de Rham, Living Rev

C. de Rham,Massive Gravity,Living Rev. Rel.17(2014) 7 [1401.4173]

work page arXiv 2014
[15]

de Rham, G

C. de Rham, G. Gabadadze and A.J. Tolley,Resummation of Massive Gravity,Phys. Rev. Lett.106(2011) 231101 [1011.1232]

work page arXiv 2011
[16]

Bounding the mass of the graviton using gravitional-wave observations of inspiralling compact binaries

C.M. Will,Bounding the mass of the graviton using gravitational-wave observations of inspiralling compact binaries,Phys. Rev. D57(1998) 2061 [gr-qc/9709011]

work page Pith review arXiv 1998
[17]

Rezzolla and O

M. Maggiore,Gravitational Waves: Theory and Experiments, Oxford University Press (2007), 10.1093/acprof:oso/9780198570745.001.0001

work page doi:10.1093/acprof:oso/9780198570745.001.0001 2007
[18]

Mehta et al.,Accurate inspiral-merger-ringdown gravitational waveforms for nonspinning black-hole binaries including the effect of subdominant modes,Phys

A.K. Mehta et al.,Accurate inspiral-merger-ringdown gravitational waveforms for nonspinning black-hole binaries including the effect of subdominant modes,Phys. Rev. D96 (2017) 124010 [1708.03501]

work page arXiv 2017
[19]

Frequency-domain gravitational waves from non-precessing black-hole binaries. II. A phenomenological model for the advanced detector era

S. Khan et al.,Frequency-domain gravitational waves from nonprecessing black-hole binaries. II. A phenomenological model for the advanced detector era,Phys. Rev. D93(2016) 044007 [1508.07253]

work page Pith review arXiv 2016
[20]

Inspiral-merger-ringdown waveforms of spinning, precessing black-hole binaries in the effective-one-body formalism

Y. Pan et al.,Inspiral-merger-ringdown waveforms of spinning, precessing black-hole binaries in the effective-one-body formalism,Phys. Rev. D89(2014) 084006 [1307.6232]

work page Pith review arXiv 2014
[21]

Van Den Broeck and A.S

C. Van Den Broeck and A.S. Sengupta,Phenomenology of amplitude-corrected post-Newtonian gravitational waveforms for compact binary inspiral: Signal-to-noise ratios, Class. Quant. Grav.24(2007) 155 [gr-qc/0607092]. [28]LIGO Scientific, Virgocollaboration,GWTC-2: Compact Binary Coalescences Observed by LIGO and Virgo during the First Half of the Third Obser...

work page arXiv 2007
[22]

Nitz et al.,gwastro/pycbc: v2.3.3 release of PyCBC, 2024

A.H. Nitz et al.,gwastro/pycbc: v2.3.3 release of PyCBC, 2024. 10.5281/zenodo.10473621

work page doi:10.5281/zenodo.10473621 2024
[23]

García-Quirós, M

C. Garc´ ıa-Quir´ os et al.,Multimode frequency-domain model for the gravitational wave signal from nonprecessing black-hole binaries,Phys. Rev. D102(2020) 064002 [2001.10914]

work page arXiv 2020
[24]

Shorten and T.M

C. Shorten and T.M. Khoshgoftaar,A Survey on Image Data Augmentation for Deep Learning,J. Big Data6(2019) 60

2019
[25]

Gravitational-wave sensitivity curves

C.J. Moore, R.H. Cole and C.P.L. Berry,Gravitational-wave sensitivity curves,Class. Quant. Grav.32(2015) 015014 [1408.0740]. [34]LIGO Scientificcollaboration,Advanced LIGO,Class. Quant. Grav.32(2015) 074001 [1411.4547]

work page Pith review arXiv 2015
[26]

Mehta et al.,Tests of general relativity with gravitational-wave observations using a flexible theory-independent method,Phys

A.K. Mehta et al.,Tests of general relativity with gravitational-wave observations using a flexible theory-independent method,Phys. Rev. D107(2023) 044020 [2203.13937]

work page arXiv 2023
[27]

Brans and R.H

C. Brans and R.H. Dicke,Mach’s Principle and a Relativistic Theory of Gravitation,Phys. Rev.124(1961) 925

1961
[28]

Hawking,Black Holes in the Brans-Dicke Theory of Gravitation,Commun

S.W. Hawking,Black Holes in the Brans-Dicke Theory of Gravitation,Commun. Math. Phys.25(1972) 167

1972
[29]

Black holes in scalar-tensor gravity

T.P. Sotiriou and V. Faraoni,Black Holes in Scalar-Tensor Gravity,Phys. Rev. Lett.108 (2012) 081103 [1109.6324]

work page Pith review arXiv 2012
[30]

Hinterbichler, Rev

K. Hinterbichler,Theoretical Aspects of Massive Gravity,Rev. Mod. Phys.84(2012) 671 [1105.3735]

work page arXiv 2012
[31]

Jackiw and S.Y

R. Jackiw and S.-Y. Pi,Chern-Simons modification of general relativity,Phys. Rev. D68 (2003) 104012 [gr-qc/0308071]

work page arXiv 2003
[32]

Chern-Simons Modified General Relativity

S. Alexander and N. Yunes,Chern-Simons Modified General Relativity,Phys. Rept.480 (2009) 1 [0907.2562]

work page Pith review arXiv 2009
[33]

Canizares, J

P. Canizares, J.R. Gair and C.F. Sopuerta,Testing Chern-Simons Modified Gravity with Gravitational-Wave Detections of Extreme-Mass-Ratio Binaries,Phys. Rev. D86(2012) 044010 [1205.1253]

work page arXiv 2012
[34]

Graviton Mass Bounds

C. de Rham, J.T. Deskins, A.J. Tolley and S.-Y. Zhou,Graviton Mass Bounds,Rev. Mod. Phys.89(2017) 025004 [1606.08462]

work page Pith review arXiv 2017
[35]

Adam: A Method for Stochastic Optimization

D.P. Kingma and J. Ba,Adam: A Method for Stochastic Optimization,Proceedings of the 3rd International Conference on Learning Representations (ICLR)(2015) [1412.6980]

work page internal anchor Pith review arXiv 2015
[36]

Szegedy, V

C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens and Z. Wojna,Rethinking the Inception Architecture for Computer Vision, inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2818–2826, 2016, DOI. – 37 –

2016