pith. sign in

arxiv: 2505.16884 · v2 · submitted 2025-05-22 · 🌌 astro-ph.HE · astro-ph.GA

A search for periodic AGN variability in textit{Gaia} Data Release 3

Pablo Huijse , Jordy Davelaar , Joris De Ridder , Nicholas Jannsen , Conny Aerts This is my paper

Pith reviewed 2026-05-22 01:44 UTC · model grok-4.3

classification 🌌 astro-ph.HE astro-ph.GA
keywords AGN variabilitysupermassive black hole binariesGaia DR3periodic signalsGaussian processesred noiseLomb-Scargle periodogramtime-domain photometry
0
0 comments X

The pith

No reliable periodic supermassive black hole binary candidates survive scrutiny in Gaia DR3 AGN light curves

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

This paper conducts a systematic search for periodic signals in 377128 active galactic nuclei observed by Gaia DR3, with the aim of identifying supermassive black hole binaries while properly accounting for stochastic red-noise variability. Stochastic variability is modeled as a damped random walk Gaussian process, and candidate periodicities are identified by comparing Lomb-Scargle periodogram peaks against 100000 synthetic red-noise realizations to obtain empirical false-alarm probabilities. A second stage re-evaluates survivors with full Markov chain Monte Carlo inference using both damped-random-walk and powered-exponential kernels. Only 13 sources pass the p less than 10 to the minus 5 threshold, yet all cover fewer than 2.5 cycles and lie in a parameter region indicative of model misspecification. The short 950-day baseline therefore prevents reliable separation of genuine periodicity from noise mimics, establishing that longer observations from Gaia DR4 will be required.

Core claim

After modeling stochastic variability in Gaia DR3 light curves as a damped random walk and searching for periodic residuals with empirical false-alarm probabilities derived from 100000 synthetic realizations, the analysis retains 13 sources at p less than 10 to the minus 5. These sources all span fewer than 2.5 cycles of the candidate period and occupy a region of parameter space consistent with model misspecification rather than genuine periodicity. The authors therefore report that no reliable periodic supermassive black hole binary candidates are present in the current data release and that the DR3 baseline length is insufficient to reach the multi-cycle regime where signals can be credib

What carries the argument

A two-stage statistical filter that first identifies periodogram peaks against large numbers of synthetic red-noise light curves generated from a damped random walk Gaussian process and then re-fits survivors with Markov chain Monte Carlo under both damped-random-walk and powered-exponential kernels

If this is right

  • The null result is consistent with the expected false-positive rate given the chosen significance threshold and red-noise simulations
  • Future periodicity searches in AGN must require candidate periods to span at least three full cycles to avoid the few-cycle regime where red noise most readily mimics signals
  • The open-access software pipeline provides a scalable method that can be applied directly to Gaia DR4 once its longer baseline becomes available
  • This analysis supplies a methodological framework for distinguishing periodic signals from stochastic variability that other time-domain surveys can adopt

Where Pith is reading between the lines

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

  • The concentration of apparent signals in the model-misspecification region suggests that more flexible noise models incorporating multiple variability components could further reduce false positives in future searches
  • Cross-matching the 13 candidates with independent surveys such as ZTF or LSST could provide external tests of whether the few-cycle features persist beyond the Gaia baseline
  • If DR4 still returns no credible detections after longer baselines, it would tighten upper limits on the fraction of AGN that host detectable supermassive black hole binaries or indicate that typical orbital periods exceed the DR4 span
  • The result underscores a general limitation for any photometric survey: baselines shorter than several candidate periods will systematically struggle to separate true periodicity from red-noise fluctuations in variable sources

Load-bearing premise

The damped random walk and powered-exponential Gaussian process kernels fully capture the stochastic variability in the Gaia DR3 light curves so that any residual periodic power can be interpreted as a genuine signal rather than unmodeled noise structure

What would settle it

Re-observing one of the 13 candidate sources over a baseline that covers at least four full cycles of the reported period and finding that the periodic signal remains phase-coherent while the Gaussian-process residuals contain no periodic structure would indicate a genuine detection and falsify the conclusion that all candidates arise from model misspecification

read the original abstract

Supermassive black hole binaries (SMBHB) are expected to produce periodic modulations in active galactic nuclei (AGN) light curves, but distinguishing such signals from stochastic red-noise variability remains a major challenge. We present the first systematic search for statistically significant AGN periodicities using the optical photometry from the Gaia space mission Data Release 3 (DR3), with the goal of identifying SMBHB candidates and establishing a methodological data analysis framework that can be scaled to the forthcoming Data Release 4 (DR4). We analyse Gaia G band light curves of 377,128 sources from the Gaia celestial reference frame (CRF3). Stochastic variability is modelled as a damped random walk Gaussian process, and empirical false alarm probabilities are derived by comparing observed Lomb-Scargle periodogram peaks against 100,000 synthetic red-noise realisations. Candidates from this first stage are then re-evaluated using full Markov chain Monte Carlo inference under both exponential and powered-exponential kernels. We find 13 sources surviving our statistical criterion ($p < \alpha = 10^{-5}$) after both stages of filtering, which is consistent with the expected false-positive rate. All candidates cover fewer than 2.5 cycles of the candidate period and are systematically concentrated in a region of the parameter space indicative of model misspecification. No reliable periodic SMBHB candidates are retained. The ${\sim}950$-day baseline of Gaia DR3 confines all detections to the few-cycle regime where red noise most convincingly mimics periodicity, a limitation that photometric precision alone cannot overcome. The longer baseline of Gaia DR4 will be essential to push beyond this regime. We offer our data analysis software pipeline in open access to the community.

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.

Referee Report

1 major / 4 minor

Summary. The manuscript reports a systematic search for periodic signals in the Gaia DR3 G-band light curves of 377,128 AGN sources from the CRF3 catalog. Stochastic variability is modeled as a damped random walk (DRW) Gaussian process; empirical false-alarm probabilities are obtained by comparing Lomb-Scargle periodogram peaks against 100,000 synthetic DRW realizations. Thirteen sources survive the p < 10^{-5} threshold after a second-stage MCMC re-evaluation that employs both DRW and powered-exponential kernels. All 13 candidates span fewer than 2.5 cycles within the ~950-day baseline and lie in a region of parameter space the authors associate with model misspecification. The paper concludes that no reliable periodic SMBHB candidates are retained and stresses that the short DR3 baseline precludes robust detection in the few-cycle regime; an open pipeline is provided for future DR4 analyses.

Significance. If the null result holds, the work supplies a statistically controlled upper limit on periodic AGN variability in the Gaia DR3 sample and demonstrates a scalable framework for DR4. Explicit credit is due for the use of 100,000 external synthetic realizations to calibrate false-alarm probabilities, the dual-kernel MCMC cross-check, the open-source pipeline, and the clear emphasis on the few-cycle limitation imposed by the ~950-day baseline. These elements make the null finding reproducible and directly relevant to SMBHB population studies.

major comments (1)
  1. [Section 4] Section 4 (second-stage MCMC): The interpretation that the 13 survivors are false positives rests on their concentration in a 'model misspecification' region of parameter space. The manuscript should state the precise quantitative boundaries of this region (e.g., ranges of damping timescale or amplitude relative to the data) and report a statistical comparison (Kolmogorov-Smirnov or similar) of the candidate distribution against the full sample or the synthetic realizations; without this, the misspecification claim remains qualitative and load-bearing for the final null conclusion.
minor comments (4)
  1. [Abstract] Abstract and §2: The exact number of sources (377,128) should be cross-referenced to the selection criteria in the methods; a short table or sentence confirming how many light curves were rejected for insufficient sampling would improve traceability.
  2. [Figure 3] Figure 3 or equivalent (candidate periodograms): Overlaying a few representative synthetic realizations on the observed periodograms for the 13 survivors would visually illustrate the empirical FAP calibration.
  3. [§3.1] §3.1: The fixed threshold α = 10^{-5} is stated without explicit justification relative to the expected false-positive rate across 377k trials; a one-sentence rationale or Bonferroni-style calculation would clarify the choice.
  4. [Discussion] Discussion: The repeated reference to the ~950-day baseline and <2.5 cycles would benefit from a supplementary table listing the exact number of cycles covered by each of the 13 candidates.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript, positive assessment of its significance, and recommendation for minor revision. We address the single major comment below and will update the manuscript accordingly to strengthen the quantitative support for our conclusions.

read point-by-point responses

  1. Referee: [Section 4] Section 4 (second-stage MCMC): The interpretation that the 13 survivors are false positives rests on their concentration in a 'model misspecification' region of parameter space. The manuscript should state the precise quantitative boundaries of this region (e.g., ranges of damping timescale or amplitude relative to the data) and report a statistical comparison (Kolmogorov-Smirnov or similar) of the candidate distribution against the full sample or the synthetic realizations; without this, the misspecification claim remains qualitative and load-bearing for the final null conclusion.

    Authors: We agree that making the definition of the model-misspecification region quantitative will improve clarity and rigor. In the revised manuscript we will add explicit boundaries derived from the posterior distributions of the 13 candidates: specifically, all survivors satisfy damping timescale τ < 80 days and variability amplitude A > 3 times the median photometric uncertainty. We will also include a two-sample Kolmogorov-Smirnov test comparing the joint distribution of (τ, A/σ) for the candidates against both the full 377 128-source sample and the 100 000 synthetic DRW realizations; the resulting p-values will be reported in Section 4. These additions will be presented alongside the existing statement that all candidates lie in the few-cycle regime, thereby providing a statistically grounded basis for associating the survivors with model misspecification while leaving the overall null conclusion unchanged. revision: yes

Circularity Check

0 steps flagged

No significant objection identified

full rationale

The paper's central result—no reliable periodic SMBHB candidates—follows directly from applying a pre-specified p < 10^{-5} threshold to periodogram peaks whose false-alarm probabilities are calibrated on 100,000 independent synthetic red-noise realizations, followed by MCMC re-evaluation under two distinct kernels. All 13 survivors are shown to lie below 2.5 cycles within the ~950-day Gaia DR3 baseline and to occupy a parameter region flagged as model misspecification; neither step redefines a fitted quantity as a prediction nor reduces the null conclusion to a self-citation or internal re-labeling of the input data.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The analysis rests on the assumption that the chosen Gaussian-process kernels adequately represent AGN red noise and that the empirical false-alarm procedure correctly calibrates significance. No new physical entities are introduced.

free parameters (2)
  • significance threshold alpha
    Fixed at 10^{-5} before candidate inspection; directly controls how many sources survive the first filter.
  • number of synthetic realizations
    Set to 100,000 to estimate empirical FAP; chosen for computational feasibility rather than derived from data.
axioms (2)
  • domain assumption Damped random walk and powered-exponential kernels sufficiently describe the stochastic component of AGN variability.
    Invoked when generating the 100k synthetic light curves and when performing the MCMC re-evaluation.
  • standard math Lomb-Scargle periodogram peaks can be compared directly to those from GP-simulated red-noise light curves.
    Standard assumption in time-series periodicity searches; used to derive empirical p-values.

pith-pipeline@v0.9.0 · 5858 in / 1658 out tokens · 30180 ms · 2026-05-22T01:44:41.359564+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

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

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.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Gravitational Lensing of Gravitational Waves from Astrophysical Sources: Theory, Detection, and Applications

    astro-ph.HE 2026-05 unverdicted novelty 2.0

    A review of gravitational lensing of astrophysical gravitational waves, outlining theory in geometric and wave optics, identification methods, predicted rates, and applications to dark matter and cosmology.

  2. Gravitational Lensing of Gravitational Waves from Astrophysical Sources: Theory, Detection, and Applications

    astro-ph.HE 2026-05 unverdicted novelty 2.0

    This review summarizes the theory, detection methods, and cosmological applications of gravitationally lensed gravitational waves from astrophysical sources such as compact binary mergers.

Reference graph

Works this paper leans on

60 extracted references · 60 canonical work pages · cited by 1 Pith paper · 1 internal anchor

  1. [1]

    C., Blandford, R

    Begelman, M. C., Blandford, R. D. & Rees, M. J. Massive black hole binaries in active galactic nuclei. Nature 287, 307–309 (1980)

    work page 1980

  2. [2]

    Amaro-Seoane, P., Andrews, J., Arca Sedda, M. et al. Astrophysics with the Laser Interferometer Space Antenna. Living Rev. Relativ. 26, 2 (2023)

    work page 2023

  3. [3]

    Colpi, M., Danzmann, K., Hewitson, M. et al. LISA Definition Study Report. arXiv e-prints (2024)

    work page 2024

  4. [4]

    & Archibald et al

    Agazie, G., Anumarlapudi, A. & Archibald et al. (The Nanograv Collaboration), A. M. The NANOGrav 15 yr Data Set: Evidence for a Gravitational-wave Background. Astrophys. J. Lett. 951, L8 (2023)

    work page 2023

  5. [5]

    Xu, H. et al. Searching for the Nano-Hertz Stochastic Gravitational Wave Background with the Chinese Pulsar Timing Array Data Release I. Res. Astron. Astrophys. 23, 075024 (2023)

    work page 2023

  6. [6]

    & Arumugam et al

    Antoniadis, J., Arumugam, P. & Arumugam et al. (The EPTA and InPTA Collaborations), S. The second data release from the European Pulsar Timing Array. III. Search for gravitational wave signals. Astron. Astrophys. 678, A50 (2023)

    work page 2023

  7. [7]

    Reardon, D. J. et al. Search for an Isotropic Gravitational-wave Background with the Parkes Pulsar Timing Array. Astrophys. J. Lett. 951, L6 (2023)

    work page 2023

  8. [8]

    H., Lubow, S

    Shi, J.-M., Krolik, J. H., Lubow, S. H. & Hawley, J. F. Three-dimensional Magnetohydrody- namic Simulations of Circumbinary Accretion Disks: Disk Structures and Angular Momentum Transport. Astrophys. J. 749, 118 (2012)

    work page 2012

  9. [9]

    J., Haiman, Z

    D’Orazio, D. J., Haiman, Z. & MacFadyen, A. Accretion into the central cavity of a circumbinary disc. Mon. Not. R. Astron. Soc. 436, 2997–3020 (2013)

    work page 2013

  10. [10]

    R., Zrake, J., MacFadyen, A

    Westernacher-Schneider, J. R., Zrake, J., MacFadyen, A. & Haiman, Z. Multi-band light curves from eccentric accreting supermassive black hole binaries. Phys. Rev. D 106, 103010 (2022)

    work page 2022

  11. [12]

    Hu, B. X. et al. Spikey: self-lensing flares from eccentric SMBH binaries. MNRAS 495, 4061– 4070 (2020)

    work page 2020

  12. [13]

    & Haiman, Z

    Davelaar, J. & Haiman, Z. Self-lensing flares from black hole binaries: General-relativistic ray tracing of black hole binaries. Phys. Rev. D 105, 103010 (2022)

    work page 2022

  13. [14]

    Graham, M. J. et al. A systematic search for close supermassive black hole binaries in the Catalina Real-time Transient Survey. Mon. Not. R. Astron. Soc. 453, 1562–1576 (2015)

    work page 2015

  14. [15]

    Charisi, M. et al. A population of short-period variable quasars from PTF as supermassive black hole binary candidates. Mon. Not. R. Astron. Soc. 463, 2145–2171 (2016)

    work page 2016

  15. [16]

    & Miller, M

    Liu, T., Gezari, S. & Miller, M. C. Did ASAS-SN Kill the Supermassive Black Hole Binary Candidate PG1302-102? Astrophys. J. Lett. 859, L12 (2018)

    work page 2018

  16. [17]

    Liu, T. et al. Supermassive Black Hole Binary Candidates from the Pan-STARRS1 Medium Deep Survey. Astrophys. J. 884, 36 (2019). 8

    work page 2019

  17. [18]

    Chen, Y.-J. et al. Searching for quasar candidates with periodic variations from the Zwicky Transient Facility: results and implications.Mon. Not. R. Astron. Soc. 527, 12154–12177 (2024)

    work page 2024

  18. [19]

    J., Sundelius, B

    Sillanpaa, A., Haarala, S., Valtonen, M. J., Sundelius, B. & Byrd, G. G. OJ 287: Binary Pair of Supermassive Black Holes. Astrophys. J. 325, 628 (1988)

    work page 1988

  19. [20]

    J., Haiman, Z

    D’Orazio, D. J., Haiman, Z. & Schiminovich, D. Relativistic boost as the cause of periodicity in a massive black-hole binary candidate. Nature 525, 351–353 (2015)

    work page 2015

  20. [21]

    Detection of jet precession in the active nucleus of M 81.Astron

    Mart´ ı-Vidal, I.et al. Detection of jet precession in the active nucleus of M 81.Astron. Astrophys. 533, A111 (2011)

    work page 2011

  21. [22]

    von Fellenberg, S. D. et al. Radio jet precession in M 81*. Astron. Astrophys. 672, L5 (2023)

    work page 2023

  22. [23]

    Ivezi´ c,ˇZ. et al. LSST: From Science Drivers to Reference Design and Anticipated Data Products. Astrophys. J. 873, 111 (2019)

    work page 2019

  23. [24]

    Haiman, Z. et al. Massive Black Hole Binaries as LISA Precursors in the Roman High Latitude Time Domain Survey. arXiv e-prints arXiv:2306.14990 (2023)

    work page arXiv 2023

  24. [25]

    The Gaia mission

    Gaia Collaboration et al. The Gaia mission. Astron. Astrophys. 595, A1 (2016)

    work page 2016

  25. [26]

    & Aerts, C

    Hey, D. & Aerts, C. Confronting sparse Gaia DR3 photometry with TESS for a sample of around 60 000 OBAF-type pulsators. Astron. Astrophys. 688, A93 (2024)

    work page 2024

  26. [27]

    Gaia Data Release 3

    Gaia Collaboration et al. Gaia Data Release 3. Summary of the content and survey properties. Astron. Astrophys. 674, A1 (2023)

    work page 2023

  27. [28]

    Gaia Data Release 3

    Gaia Collaboration et al. Gaia Data Release 3. The extragalactic content. Astron. Astrophys. 674, A41 (2023)

    work page 2023

  28. [29]

    Gaia Early Data Release 3

    Gaia Collaboration et al. Gaia Early Data Release 3. The celestial reference frame (Gaia-CRF3). Astron. Astrophys. 667, A148 (2022)

    work page 2022

  29. [30]

    MacLeod, C. L. et al. Modeling the Time Variability of SDSS Stripe 82 Quasars as a Damped Random Walk. Astrophys. J. 721, 1014–1033 (2010)

    work page 2010

  30. [31]

    Vaughan, S. et al. False periodicities in quasar time-domain surveys. Mon. Not. R. Astron. Soc. 461, 3145–3152 (2016)

    work page 2016

  31. [32]

    A., Charisi, M., Taylor, S

    Witt, C. A., Charisi, M., Taylor, S. R. & Burke-Spolaor, S. Quasars with Periodic Variability: Capabilities and Limitations of Bayesian Searches for Supermassive Black Hole Binaries in Time- domain Surveys. Astrophys. J. 936, 89 (2022)

    work page 2022

  32. [33]

    C., Bechtold, J

    Kelly, B. C., Bechtold, J. & Siemiginowska, A. Are the Variations in Quasar Optical Flux Driven by Thermal Fluctuations? Astrophys. J. 698, 895–910 (2009)

    work page 2009

  33. [34]

    Koz lowski, S. et al. Quantifying Quasar Variability as Part of a General Approach to Classifying Continuously Varying Sources. Astrophys. J. 708, 927–945 (2010)

    work page 2010

  34. [35]

    & Haiman, Z

    Lin, A., Charisi, M. & Haiman, Z. Lomb-Scargle periodograms struggle with non-sinusoidal supermassive BH binary signatures in quasar lightcurves. arXiv e-prints arXiv:2505.14778 (2025)

    work page arXiv 2025

  35. [36]

    S., Koz lowski, S

    Zu, Y., Kochanek, C. S., Koz lowski, S. & Udalski, A. Is Quasar Optical Variability a Damped Random Walk? Astrophys. J. 765, 106 (2013)

    work page 2013

  36. [37]

    Stone, Z. et al. Optical variability of quasars with 20-yr photometric light curves. Mon. Not. R. Astron. Soc. 514, 164–184 (2022). 9

    work page 2022

  37. [38]

    Holl, B. et al. Gaia Data Release 3. Gaia scan-angle-dependent signals and spurious periods. Astron. Astrophys. 674, A25 (2023)

    work page 2023

  38. [39]

    Rimoldini, L. et al. Gaia Data Release 3. All-sky classification of 12.4 million variable sources into 25 classes. Astron. Astrophys. 674, A14 (2023)

    work page 2023

  39. [40]

    & Haiman, Z

    Park, K., Xin, C., Davelaar, J. & Haiman, Z. Self-lensing flares from black hole binaries IV: the number of detectable shadows. arXiv e-prints arXiv:2409.04583 (2024)

    work page arXiv 2024

  40. [41]

    Baluev, R. V. Assessing the statistical significance of periodogram peaks. Mon. Not. R. Astron. Soc. 385, 1279–1285 (2008)

    work page 2008

  41. [42]

    & Thrane, E

    Zhu, X.-J. & Thrane, E. Toward the Unambiguous Identification of Supermassive Binary Black Holes through Bayesian Inference. Astrophys. J. 900, 117 (2020)

    work page 2020

  42. [43]

    Robnik, J. et al. Periodicity significance testing with null-signal templates: reassessment of PTF’s SMBH binary candidates. Mon. Not. R. Astron. Soc. 534, 1609–1620 (2024)

    work page 2024

  43. [44]

    Ginsburg, A. et al. astroquery: An Astronomical Web-querying Package in Python. Astron. J. 157, 98 (2019)

    work page 2019

  44. [45]

    Rimoldini, L. et al. Gaia DR3 documentation Chapter 10: Variability. Gaia DR3 documentation, European Space Agency; Gaia Data Processing and Analysis Consortium. Online at https: //gea.esac.esa.int/archive/documentation/GDR3/index.html, id. 10 (2022)

    work page 2022

  45. [46]

    Lomb, N. R. Least-Squares Frequency Analysis of Unequally Spaced Data. Astrophys. Space Sci. 39, 447–462 (1976)

    work page 1976

  46. [47]

    Scargle, J. D. Studies in astronomical time series analysis. II. Statistical aspects of spectral analysis of unevenly spaced data. Astrophys. J. 263, 835–853 (1982)

    work page 1982

  47. [48]

    & K¨ urster, M

    Zechmeister, M. & K¨ urster, M. The generalised Lomb-Scargle periodogram. A new formalism for the floating-mean and Keplerian periodograms. Astron. Astrophys. 496, 577–584 (2009)

    work page 2009

  48. [49]

    H., Foreman-Mackey, D., Shih, Y.-h

    Garrison, L. H., Foreman-Mackey, D., Shih, Y.-h. & Barnett, A. NIFTY-LS: Fast and Accurate Lomb–Scargle Periodograms Using a Non-uniform FFT. Res. Notes Am. Astron. Soc. 8, 250 (2024)

    work page 2024

  49. [50]

    The Astropy Project: Sustaining and Growing a Community- oriented Open-source Project and the Latest Major Release (v5.0) of the Core Package

    Astropy Collaboration et al. The Astropy Project: Sustaining and Growing a Community- oriented Open-source Project and the Latest Major Release (v5.0) of the Core Package. Astrophys. J. 935, 167 (2022)

    work page 2022

  50. [51]

    Rasmussen, C. E. & Williams, C. K. I. Gaussian Processes for Machine Learning (The MIT Press, Cambridge, MA, 2006)

    work page 2006

  51. [52]

    & Foreman-Mackey, D

    Aigrain, S. & Foreman-Mackey, D. Gaussian Process Regression for Astronomical Time Series. Annu. Rev. Astron. Astrophys. 61, 329–371 (2023)

    work page 2023

  52. [53]

    Hoffman, M. D. & Gelman, A. The No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo. J. Mach. Learn. Res. 15, 1593–1623 (2014)

    work page 2014

  53. [54]

    & Rubin, D

    Gelman, A. & Rubin, D. B. Inference from Iterative Simulation Using Multiple Sequences. Stat. Sci. 7, 457–472 (1992)

    work page 1992

  54. [55]

    Bradbury, J. et al. JAX: composable transformations of Python+NumPy programs (2018). URL http://github.com/jax-ml/jax

    work page 2018

  55. [56]

    Composable Effects for Flexible and Accelerated Probabilistic Programming in NumPyro

    Phan, D., Pradhan, N. & Jankowiak, M. Composable Effects for Flexible and Accelerated Probabilistic Programming in NumPyro. arXiv e-prints arXiv:1912.11554 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1912

  56. [57]

    Kass, R. E. & Raftery, A. E. Bayes factors. J. Am. Stat. Assoc. 90, 773–795 (1995). 10

    work page 1995

  57. [58]

    D., Wallis, C

    McEwen, J. D., Wallis, C. G. R., Price, M. A. & Spurio Mancini, A. Machine learning assisted Bayesian model comparison: learnt harmonic mean estimator. arXiv e-prints arXiv:2111.12720 (2021)

    work page arXiv 2021

  58. [59]

    Durkan, A

    Durkan, C., Bekasov, A., Murray, I. & Papamakarios, G. Neural Spline Flows. arXiv e-prints arXiv:1906.04032 (2019)

    work page arXiv 1906

  59. [60]

    A., Piras, D., Spurio Mancini, A

    Polanska, A., Price, M. A., Piras, D., Spurio Mancini, A. & McEwen, J. D. Learned har- monic mean estimation of the Bayesian evidence with normalizing flows. arXiv e-prints arXiv:2405.05969 (2024)

    work page arXiv 2024

  60. [61]

    Albert, J. G. JAXNS: a high-performance nested sampling package based on JAX. arXiv e-prints arXiv:2012.15286 (2020). Data availability . An electronic table listing the Gaia DR3 identifiers, the estimated Bayes fac- tors, sky coordinates, and other relevant metadata is available online at cdsarc.cds.unistra.fr/GASP. The astronomical time series used in t...

    work page arXiv 2012