pith. sign in

arxiv: 2606.30536 · v1 · pith:XOHZPVEBnew · submitted 2026-06-29 · 🌀 gr-qc · astro-ph.HE

Evaluating the Fourier Approximation in Pulsar Timing Array Analysis

Yongqi Zhang , Hayden Scholz , Ken D. Olum , Lucas Steinberger , Gabriella Agazie , Akash Anumarlapudi , Anne M. Archibald , Zaven Arzoumanian
show 48 more authors
Paul T. Baker Paul R. Brook H. Thankful Cromartie Kathryn Crowter Megan E. DeCesar Paul B. Demorest Timothy Dolch Justin A. Ellis Elizabeth C. Ferrara William Fiore Emmanuel Fonseca Gabriel E. Freedman Nate Garver-Daniels Peter A. Gentile Joseph Glaser Deborah C. Good Jeffrey S. Hazboun Ross J. Jennings Megan L. Jones David L. Kaplan Matthew Kerr Michael T. Lam Duncan R. Lorimer Jing Luo Ryan S. Lynch Alexander McEwen Maura A. McLaughlin Natasha McMann Bradley W. Meyers Cherry Ng David J. Nice Timothy T. Pennucci Benetge B. P. Perera Nihan S. Pol Henri A. Radovan Scott M. Ransom Paul S. Ray Ann Schmiedekamp Carl Schmiedekamp Brent J. Shapiro-Albert Ingrid H. Stairs Kevin Stovall Abhimanyu Susobhanan Joseph K. Swiggum Stephen R. Taylor Michele Vallisneri Rutger van Haasteren Haley M. Wahl
This is my paper

Pith reviewed 2026-06-30 04:43 UTC · model grok-4.3

classification 🌀 gr-qc astro-ph.HE
keywords pulsar timing arraysFourier approximationgravitational wave backgroundpower spectral densitymarginal likelihoodHellings-Downs correlationNANOGrav
0
0 comments X

The pith

The Fourier approximation overestimates marginal likelihoods for power-law spectra in pulsar timing arrays by a factor of about two on average.

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

Pulsar timing arrays detect stochastic signals such as gravitational wave backgrounds by comparing observed pulse arrival times to models with a specified power spectral density. Standard Bayesian analyses approximate the signal as a sum of Fourier modes spaced to the total observation length, even though the true signal is not periodic. This paper compares those approximate likelihoods to results obtained with finer frequency spacing on the NANOGrav 15-year dataset. The true marginal likelihoods for power-law spectra average about half the size of the approximated values. The factor-of-two shift could affect model comparisons, yet the same correction applies to both uncorrelated and Hellings-Downs correlated models, leaving their relative ranking unchanged, while parameter estimates remain similar.

Core claim

The true marginal likelihoods for power-law PSDs are on average about half as large as the likelihoods computed using the Fourier approximation. This could lead to an error of a factor of two in model comparison. However, in the important comparison of uncorrelated vs. Hellings-Downs correlated models, a very similar correction appears in both, so the model comparison is essentially unaffected. Parameter estimation results for power law PSDs show little difference between the methods. Spectra with sharper features could see larger discrepancies.

What carries the argument

The Fourier approximation, which models the stochastic signal as a sum of discrete Fourier modes whose frequencies are spaced by the inverse of the total observation time even though the actual signal is not periodic.

If this is right

  • Comparisons between uncorrelated noise models and Hellings-Downs correlated models remain reliable because the approximation error affects both similarly.
  • Inferred parameters such as the power-law amplitude and spectral index change little when switching from the standard Fourier method to finer frequencies.
  • Analyses of spectra containing sharp features may require exact or finer-frequency methods to avoid substantially larger errors.

Where Pith is reading between the lines

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

  • Analyses of longer or denser datasets could reveal whether the factor-of-two offset grows or shrinks with observation span.
  • Similar non-periodicity effects may appear in other time-domain analyses that rely on Fourier bases for aperiodic processes.
  • Adopting the finer-spacing approach as a default would increase computational cost but could be tested first on subsets of existing data.

Load-bearing premise

That calculations with more-closely spaced frequencies serve as a reliable proxy for the exact result and that the observed difference arises primarily from the non-periodicity of the true signal.

What would settle it

Direct evaluation of the exact marginal likelihood for a power-law PSD in a small simulated dataset without frequency discretization, then checking whether it matches the finer-spacing proxy.

Figures

Figures reproduced from arXiv: 2606.30536 by Abhimanyu Susobhanan, Akash Anumarlapudi, Alexander McEwen, Anne M. Archibald, Ann Schmiedekamp, Benetge B. P. Perera, Bradley W. Meyers, Brent J. Shapiro-Albert, Carl Schmiedekamp, Cherry Ng, David J. Nice, David L. Kaplan, Deborah C. Good, Duncan R. Lorimer, Elizabeth C. Ferrara, Emmanuel Fonseca, Gabriel E. Freedman, Gabriella Agazie, Haley M. Wahl, Hayden Scholz, Henri A. Radovan, H. Thankful Cromartie, Ingrid H. Stairs, Jeffrey S. Hazboun, Jing Luo, Joseph Glaser, Joseph K. Swiggum, Justin A. Ellis, Kathryn Crowter, Ken D. Olum, Kevin Stovall, Lucas Steinberger, Matthew Kerr, Maura A. McLaughlin, Megan E. DeCesar, Megan L. Jones, Michael T. Lam, Michele Vallisneri, Natasha McMann, Nate Garver-Daniels, Nihan S. Pol, Paul B. Demorest, Paul R. Brook, Paul S. Ray, Paul T. Baker, Peter A. Gentile, Ross J. Jennings, Rutger van Haasteren, Ryan S. Lynch, Scott M. Ransom, Stephen R. Taylor, Timothy Dolch, Timothy T. Pennucci, William Fiore, Yongqi Zhang, Zaven Arzoumanian.

Figure 1
Figure 1. Figure 1: FIG. 1. Histogram of Frobenius norms of the difference between the exact and 1/T, 1/10T, and [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. The difference between the 1 [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. As in Fig. 2 for HD likelihoods with 1548 samples. [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Comparison of parameter estimations using 1 [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. As in Fig. 4 but for the HD model [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
read the original abstract

Pulsar timing arrays search for stochastic processes such as gravitational waves by comparing pulse time of arrival data for millisecond pulsars to expectations from a background with a given power spectral density (PSD). To make the analysis computationally tractable, the Bayesian likelihood is usually computed using an approximation in which the signal is taken to be a sum of Fourier modes appropriate to the total time of observation, even though the true signal is not periodic. We study the difference between likelihoods computed with this Fourier approximation method for power law spectra and those computed exactly (or using more-closely spaced frequencies as a proxy for the exact result) in the NANOGrav 15-year dataset. We find that the true marginal likelihoods for power-law PSDs are on average about half as large as the likelihoods computed using the Fourier approximation. This could lead to an error of a factor of two in model comparison. However, in the important comparison of uncorrelated vs. Hellings-Downs correlated models, a very similar correction appears in both, so the model comparison is essentially unaffected. We also compare parameter estimation results for power law PSDs, finding little difference between the methods. We briefly discuss spectra with sharper features, for which the approximation could be much worse.

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

2 major / 2 minor

Summary. The manuscript evaluates the Fourier approximation commonly used in pulsar timing array (PTA) likelihood calculations for stochastic signals with power-law power spectral densities (PSDs). On the NANOGrav 15-year dataset, it compares marginal likelihoods obtained with the standard Fourier-mode sum (periodic over the observation span) against those computed with a finer frequency grid serving as a proxy for the exact non-periodic covariance integral. The central numerical result is that the proxy 'true' marginal likelihoods are on average roughly half as large as the Fourier-approximation values, implying a potential factor-of-two bias in model evidence; however, this bias is nearly identical for uncorrelated and Hellings-Downs correlated models, leaving their Bayes factor essentially unchanged. Parameter posteriors for power-law PSDs show little difference between the two methods, while the approximation is flagged as potentially worse for spectra with sharp features.

Significance. If the reported factor-of-two offset is robust, the work supplies a concrete, data-driven calibration of a systematic error in standard PTA pipelines. The direct numerical comparison on a public dataset is a clear strength that supports reproducibility. At the same time, the finding that the dominant science result (model comparison between uncorrelated and correlated backgrounds) is insensitive to the approximation limits the immediate practical impact on existing PTA conclusions, though it remains relevant for analyses involving non-power-law spectra or future higher-precision data.

major comments (2)
  1. [Methods] Methods section: The headline claim that true marginal likelihoods are 'on average about half' those from the Fourier approximation rests on treating the finer-spaced frequency grid as a faithful proxy for the exact non-periodic covariance. No explicit convergence test is presented (e.g., results for successively halved frequency spacings demonstrating that the marginal likelihood stabilizes to within a small fraction of the observed difference). Without this, the discrepancy could arise from residual discretization error rather than non-periodicity itself.
  2. [Results] Results section: The statement that 'a very similar correction appears in both' the uncorrelated and Hellings-Downs models (leaving model comparison unaffected) is load-bearing for the practical conclusion. The manuscript should report the actual ratio of the two Bayes factors (or the change in log-evidence) with quantitative uncertainty rather than the qualitative phrase 'essentially unaffected'.
minor comments (2)
  1. [Abstract] Abstract: the phrase 'on average about half as large' would be more precise if the actual mean ratio (with its standard deviation across pulsars or realizations) were stated numerically.
  2. Notation: the distinction between the Fourier approximation and the 'exact' (proxy) covariance should be introduced with an explicit equation for the covariance integral being approximated, to make the source of the non-periodicity clear.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading and constructive feedback on our manuscript. We address each of the major comments below and will incorporate the suggested improvements in a revised version.

read point-by-point responses

  1. Referee: [Methods] Methods section: The headline claim that true marginal likelihoods are 'on average about half' those from the Fourier approximation rests on treating the finer-spaced frequency grid as a faithful proxy for the exact non-periodic covariance. No explicit convergence test is presented (e.g., results for successively halved frequency spacings demonstrating that the marginal likelihood stabilizes to within a small fraction of the observed difference). Without this, the discrepancy could arise from residual discretization error rather than non-periodicity itself.

    Authors: We agree that demonstrating convergence of the finer frequency grid is important to confirm it serves as a reliable proxy for the exact covariance. In the revised manuscript, we will add an explicit convergence test by computing the marginal likelihoods for successively halved frequency spacings and showing that the values stabilize, thereby confirming that the factor-of-two difference arises from the non-periodic nature of the signal rather than discretization effects. revision: yes

  2. Referee: [Results] Results section: The statement that 'a very similar correction appears in both' the uncorrelated and Hellings-Downs models (leaving model comparison unaffected) is load-bearing for the practical conclusion. The manuscript should report the actual ratio of the two Bayes factors (or the change in log-evidence) with quantitative uncertainty rather than the qualitative phrase 'essentially unaffected'.

    Authors: We concur that a quantitative assessment strengthens the conclusion. In the revised version, we will report the actual ratios of the Bayes factors (or differences in log-evidence) between the Fourier approximation and the finer-grid method for both the uncorrelated and Hellings-Downs models, including any associated uncertainties. This will demonstrate quantitatively that the model comparison remains essentially unaffected. revision: yes

Circularity Check

0 steps flagged

No significant circularity in numerical comparison study

full rationale

The paper performs a direct numerical comparison of likelihoods computed via the Fourier approximation versus a finer-frequency proxy on external NANOGrav 15-year observational data. No mathematical derivation chain exists that reduces predictions or results to fitted inputs, self-definitions, or self-citation load-bearing premises by construction. The reported factor-of-two difference is an empirical output of the computation itself, with the study self-contained against external benchmarks and no load-bearing steps matching the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

This is a computational validation study. It relies on standard domain assumptions in Bayesian time-series analysis for stochastic gravitational wave backgrounds but introduces no new free parameters, axioms beyond those, or invented entities.

axioms (2)
  • domain assumption The gravitational wave background can be modeled as a stationary Gaussian stochastic process characterized by a power spectral density.
    Standard modeling choice in PTA analyses for the stochastic background.
  • domain assumption Marginal likelihoods computed via Bayesian methods are the appropriate quantity for model comparison.
    Common framework in the field for comparing signal models.

pith-pipeline@v0.9.1-grok · 6040 in / 1356 out tokens · 61857 ms · 2026-06-30T04:43:51.822610+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

17 extracted references · 12 canonical work pages · 7 internal anchors

  1. [1]

    R. W. Hellings and G. S. Downs, Upper limits on the isotropic gravitational radiation back- ground from pulsar timing analysis, Astrophys. J.265, L39 (1983)

    1983

  2. [2]

    Agazieet al.(NANOGrav), The NANOGrav 15 yr Data Set: Observations and Timing of 68 Millisecond Pulsars, Astrophys

    G. Agazieet al.(NANOGrav), The NANOGrav 15 yr Data Set: Observations and Timing of 68 Millisecond Pulsars, Astrophys. J. Lett.951, L9 (2023), arXiv:2306.16217 [astro-ph.HE]

    work page arXiv 2023

  3. [3]

    Antoniadiset al.(EPTA), The second data release from the European Pulsar Timing Array - I

    J. Antoniadiset al.(EPTA), The second data release from the European Pulsar Timing Array - I. The dataset and timing analysis, Astron. Astrophys.678, A48 (2023), arXiv:2306.16224 [astro-ph.HE]

    work page arXiv 2023

  4. [4]

    Zicet al., The Parkes Pulsar Timing Array third data release, Publ

    A. Zicet al., The Parkes Pulsar Timing Array third data release, Publ. Astron. Soc. Austral. 40, e049 (2023), arXiv:2306.16230 [astro-ph.HE]

    work page arXiv 2023

  5. [5]

    Searching for the nano-Hertz stochastic gravitational wave background with the Chinese Pulsar Timing Array Data Release I

    H. Xuet 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), arXiv:2306.16216 [astro-ph.HE]

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  6. [6]

    The NANOGrav 15-year Data Set: Evidence for a Gravitational-Wave Background

    G. Agazieet al.(NANOGrav), The NANOGrav 15 yr Data Set: Evidence for a Gravitational- wave Background, Astrophys. J. Lett.951, L8 (2023), arXiv:2306.16213 [astro-ph.HE]. 15

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  7. [7]

    The second data release from the European Pulsar Timing Array III. Search for gravitational wave signals

    J. Antoniadiset al.(EPTA, InPTA), The second data release from the European Pulsar Timing Array - III. Search for gravitational wave signals, Astron. Astrophys.678, A50 (2023), arXiv:2306.16214 [astro-ph.HE]

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  8. [8]

    D. J. Reardonet al., Search for an Isotropic Gravitational-wave Background with the Parkes Pulsar Timing Array, Astrophys. J. Lett.951, L6 (2023), arXiv:2306.16215 [astro-ph.HE]

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  9. [9]

    Hyper-efficient model-independent Bayesian method for the analysis of pulsar timing data

    L. Lentati, P. Alexander, M. P. Hobson, S. Taylor, J. Gair, S. T. Balan, and R. van Haasteren, Hyper-efficient model-independent Bayesian method for the analysis of pulsar timing data, Phys. Rev. D87, 104021 (2013), arXiv:1210.3578 [astro-ph.IM]

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  10. [10]

    van Haasteren and M

    R. van Haasteren and M. Vallisneri, Low-rank approximations for large stationary covariance matrices, as used in the bayesian and generalized-least-squares analysis of pulsar-timing data, Monthly Notices of the Royal Astronomical Society446, 1170 (2014)

    2014

  11. [11]

    On measuring the gravitational-wave background using Pulsar Timing Arrays

    R. van Haasteren, Y. Levin, P. McDonald, and T. Lu, On measuring the gravitational-wave background using Pulsar Timing Arrays, Mon. Not. Roy. Astron. Soc.395, 1005 (2009), arXiv:0809.0791 [astro-ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  12. [12]

    van Haasteren and Y

    R. van Haasteren and Y. Levin, Understanding and analysing time-correlated stochastic signals in pulsar timing, Monthly Notices of the Royal Astronomical Society428, 1147 (2012)

    2012

  13. [13]

    R. T. Edwards, G. B. Hobbs, and R. N. Manchester, Tempo2, a new pulsar timing package

  14. [14]

    The timing model and precision estimates, Mon. Not. Roy. Astron. Soc.372, 1549 (2006), arXiv:astro-ph/0607664

    work page internal anchor Pith review Pith/arXiv arXiv 2006

  15. [15]

    Payne, C

    E. Payne, C. Talbot, and E. Thrane, Higher order gravitational-wave modes with likelihood reweighting, Physical Review D100, 10.1103/physrevd.100.123017 (2019)

    work page doi:10.1103/physrevd.100.123017 2019

  16. [16]

    Hourihane, P

    S. Hourihane, P. Meyers, A. Johnson, K. Chatziioannou, and M. Vallisneri, Accurate char- acterization of the stochastic gravitational-wave background with pulsar timing arrays by likelihood reweighting, Physical Review D107, 10.1103/physrevd.107.084045 (2023)

    work page doi:10.1103/physrevd.107.084045 2023

  17. [17]

    W. Xie, P. O. Lewis, Y. Fan, L. Kuo, and M.-H. Chen, Improving marginal likelihood estimation for bayesian phylogenetic model selection, Systematic Biology60, 150 (2011), https://academic.oup.com/sysbio/article-pdf/60/2/150/24552358/syq085.pdf. 16

    2011