Pith. sign in

REVIEW 2 major objections 5 minor 7 cited by

Beyond diagonal approximations: improved covariance modeling for pulsar timing array data analysis

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The standard diagonal Fourier-space covariance used in PTA analyses biases the inferred parameters of the gravitational-wave background; the paper's FFT-plus-interpolation method recovers the injected values in simulation.

desk verdict Solid methods paper with a real bias finding for common red processes; the GWB-specific claim needs a Hellings-Downs test. read the letter →

arxiv 2506.13866 v1 pith:JGSBQ4TV submitted 2025-06-16 astro-ph.IM astro-ph.COastro-ph.GAastro-ph.HEgr-qc

Marco Crisostomi , Rutger van Haasteren , Patrick M. Meyers , Michele Vallisneri This is my paper
classification astro-ph.IMastro-ph.COastro-ph.GAastro-ph.HEgr-qc
keywords pulsartimingarraysgravitational-wavebackgroundcovariancematrixlow-rankapproximationfastFouriertransformdiagonalbiasspectralparameterestimationWiener-Khinchintheorem
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that the usual shortcut in pulsar timing array (PTA) analyses—modeling time-correlated noise as diagonal in a Fourier basis, with no correlations between frequency bins—is not accurate enough. It shows analytically and in simulation that the shortcut biases the recovered spectral parameters of the array-wide red process that would carry the gravitational-wave background: the amplitude comes out too high, the spectral index too low, and the break frequency too high. The paper proposes a way to keep the low-rank computational speed while using the correct covariance: compute the process's autocorrelation on a coarse time grid with a fast Fourier transform, interpolate it to the unevenly spaced observation times, and feed the result into a low-rank likelihood. In a 67-pulsar simulation patterned on a 15-year pulsar-timing dataset, the new method recovers the injected parameters that the diagonal method misses. If this is right, PTA gravitational-wave-background parameter estimates may need revisiting as datasets grow more sensitive.

What carries the argument

The engine is the Wiener–Khinchin relation evaluated on a coarse grid. For a stationary process with one-sided power spectral density $S(f)$, the autocorrelation is $C(\tau)=\int_0^\infty df\,S(f)\cos(2\pi f\tau)$, and the method evaluates this integral by an inverse fast cosine transform with an oversampling factor $\omega$ and frequency cutoff $\zeta$, obtaining $\hat c_a=C(a\Delta\hat t)$ on $\hat N$ regularly spaced times. The Toeplitz property of stationary covariances turns that autocorrelation vector into the coarse covariance matrix $\hat C$, and a localized linear-interpolation matrix $B$ carries $\hat C$ to the unevenly spaced times of arrival, giving $C\approx B\hat C B^T$ inside the low-rank Sherman–Morrison–Woodbury likelihood. This avoids the expensive double-sinc integral of the exact finite-window covariance while preserving the frequency correlations that the diagonal Fourier prior throws away.

What would settle it

Re-run the end-to-end analysis injecting a common process with the quadrupolar inter-pulsar correlation expected for a gravitational-wave background instead of the spatially uncorrelated common process; if the diagonal and FFT-interpolation posteriors for amplitude, spectral index, and break frequency then agree with the injected values, the paper's central bias claim would not transfer to real background searches.

Watch

Extended reading notes

Core claim

The central claim is that the standard PTA approximation $\Phi_{jk}=S(f_j)\delta_{jk}$, which treats the Fourier-domain covariance as diagonal, neglects the inter-frequency correlations that a finite observation window necessarily introduces through sinc-shaped spectral leakage. Those correlations are not a detail: for the array-wide common red process, the diagonal approximation systematically prefers a larger break frequency, a smaller spectral index, and a larger amplitude than the true injected values in an end-to-end simulation. The paper's method, FFTInt, replaces the Fourier low-rank basis with a time-domain construction: it evaluates the autocorrelation function on a coarse regular grid by an inverse fast cosine transform of the power spectral density, assembles the coarse Toeplitz covariance matrix, and interpolates it to the actual unevenly sampled observation times with a localized linear-interpolation matrix. This captures the frequency correlations faithfully, avoids the Gibbs ringing that afflicts non-diagonal Fourier reconstructions, and recovers unbiased spectral parameters at modest computational cost.

Load-bearing premise

The analysis assumes that an array-wide correlated signal with no spatial pattern behaves like the real gravitational-wave background; if the real background's directional correlation changes how the diagonal approximation distorts parameters, the claimed bias could differ for actual detections.

Editorial extensions

If this is right

  • With the diagonal approximation, the common-process amplitude is overestimated, the spectral index underestimated, and the break frequency overestimated; the new method recovers the injected values in the full-array simulation.
  • Individual-pulsar spin-noise parameters are only mildly affected, and the common-process bias becomes apparent only when many pulsars are analyzed together.
  • The new method's likelihood converges at a modest number of coarse grid points and an oversampling factor around five, so the unbiased covariance is attainable at acceptable computational cost.
  • Because the biased parameters are the same ones that recent PTA results have found in tension with theoretical expectations, using the full covariance could change how those tensions are interpreted.
  • The method transfers straightforwardly to larger arrays and combined multi-telescope datasets, where the tolerance for covariance mismodeling is smaller.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • We infer that the same diagonal-covariance shortcut in other stationary-process analyses built on truncated Fourier bases could produce analogous parameter biases, since the mechanism is general to finite-window Fourier representations.
  • The paper's simulation uses a spatially uncorrelated common process as a stand-in for the gravitational-wave background; the size and direction of the bias for a quadrupolar-correlated background remain an open test.
  • If the simulation transfers to real data, switching from the diagonal to the full-covariance treatment should move the inferred background amplitude downward, the spectral index upward, and the break frequency downward relative to published diagonal-based results.
  • The coarse-grid FFT plus localized interpolation scheme could also serve other stationary, unevenly sampled Gaussian processes in pulsar-timing analysis, such as clock or solar-system-ephemeris noise, wherever full-rank covariance is too expensive.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper challenges the standard PTA practice of approximating the covariance of stationary red-noise processes as diagonal in a low-rank Fourier basis. It derives the exact finite-window frequency correlations (Eq. 10), reviews the standard Fourier-sum reconstruction (Sec. II.B), and proposes a new method, FFTInt, which evaluates the Wiener-Khinchin autocorrelation on a coarse uniform grid via a fast cosine transform and linearly interpolates it to the unevenly sampled TOAs, using the Woodbury identity for the likelihood. The paper compares FFTInt, the diagonal approximation, and a non-diagonal 'SincPhi' Fourier approximation against exact Matérn and Gaussian covariance matrices, and runs NANOGrav-15yr-inspired simulations comparing posterior distributions under DiagonalPhi and FFTInt. It finds that DiagonalPhi is the least accurate covariance approximation and yields visibly biased common red-noise parameters (γgw, log10 Agw, log10 fb), while FFTInt recovers the injection.

Significance. If correct, the proposed FFTInt method is a valuable, practical improvement for PTA covariance modeling: it is derived from first principles, is implemented in Enterprise and Discovery, converges with tunable parameters (ω, ζ, and N_hat), and avoids the Gibbs ringing that affects Fourier-based reconstructions. The analytic covariance comparisons (Table I, Figs. 1-3) are clean and support the method's accuracy. The end-to-end simulations provide a clear demonstration that the diagonal approximation can bias common-process spectral parameters, and the npsr scaling in Fig. 8 is a nice illustration that the bias emerges with statistical power. The main caveat is that the demonstrative 'common process' is spatially uncorrelated and therefore does not directly test the Hellings-Downs-correlated GWB that motivates the paper's headline claim.

major comments (2)
  1. [Sec. IV, second paragraph] The simulation uses a 'common spatially uncorrelated red-noise process (which represents the GWB)' rather than a Hellings-Downs-correlated background. The bias mechanism in Eqs. (10)-(11) is derived for each pulsar's autocovariance, and the paper does not test whether the diagonal approximation's finite-window leakage produces the same bias when the cross-pulsar blocks of the GWB covariance are included. Because the real GWB likelihood has off-diagonal cross-pulsar correlations that add information, the magnitude, direction, and even existence of the DiagonalPhi bias could differ. Please add an end-to-end test with an HD-correlated common process, or provide an analytic/Fisher-matrix argument showing that the uncorrelated proxy is representative, before claiming in the abstract that the bias applies to the GWB. In addition, Sec. II.C describes FFTInt only for single-pulsar autocovariances; the extension to cross-pulsar HD blocks (B_i C B_j^T times the HD overlap) should be described and tested to support the GWB claim.
  2. [Sec. IV, simulation setup] The simulation description does not state whether timing-model parameters are marginalized in the likelihood. The covariance comparisons in Sec. III show that quadratic projection substantially reduces the differences between covariance approximations (Table I and Fig. 2), so the end-to-end bias could depend on this modeling choice. Please specify the treatment of the timing model in the simulation, and if it was omitted, rerun the analysis with timing-model marginalization or justify why the projection results imply the bias survives.
minor comments (5)
  1. [Sec. II.C, Eq. (20)] The notation around n, nmax, and N_hat is inconsistent; please clarify the relationships among the number of FFT frequencies, the coarse-grid size, ω, and ζ.
  2. [Sec. IV] The statement that a 67-pulsar dataset gives 137 parameters is correct only if the timing model is fixed or marginalized analytically, which should be stated explicitly in the setup.
  3. [Sec. IV / Abstract] The paper claims 'modest computational cost' but does not report wall-clock times or operation counts for DiagonalPhi and FFTInt; a timing benchmark would strengthen the practical-relevance claim.
  4. [Fig. 3] The horizontal axis is labeled N while the text uses N_hat for the number of coarse nodes; please unify the notation in the figure and caption.
  5. [Appendix B / Fig. 9] The caption uses 'PhiDiag' while the main text uses 'DiagonalPhi'; please use a single name throughout.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the method is benchmarked against exact analytic covariances, and the diagonal-bias result is an emergent simulation outcome.

full rationale

I walked the derivation chain from the Wiener–Khinchin relation (Eq. 3), through the exact finite-window Fourier covariance (Eq. 10), to the new FFTInt construction (Eq. 20) and its low-rank likelihood via the Woodbury identity. The method's accuracy is validated directly against analytic covariance matrices (Matérn 3/2 in Sec. IIIA, Gaussian PSDs in Appendix B), so FFTInt is not defined in terms of the quantities it is later used to estimate. The central claim that DiagonalPhi biases common-process parameters is an emergent result of the Sec. IV simulation, not a fitted parameter renamed as a prediction. The simulation does generate data using FFTInt, but the paper supports FFTInt's fidelity against exact covariances and uses high oversampling; this is a consistency check rather than a circular reduction. Self-citations such as [10], [22], [24], [25], and [26] provide methodological background and are not load-bearing premises that force the paper's conclusions. The uncorrelated common-process proxy for the GWB is a potential external-validity limitation, not a circularity. I find no specific equation or inference step that reduces to its own input, so the circularity score is low.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The method introduces no new physical entities. The listed free parameters are numerical accuracy controls that determine how finely the autocorrelation is represented and integrated; they are not fitted to the data and do not affect the qualitative conclusion about the diagonal approximation bias.

free parameters (3)
  • oversampling factor omega = 5-50
    Chosen to make the FFT evaluation of the Wiener-Khinchin integral accurate; convergence in Fig. 5 shows little change for omega >= 5.
  • Nyquist cutoff factor zeta = 1
    Set to 1 throughout; Fig. 3 shows zeta has very small effect when N is adequate.
  • number of coarse nodes N = 61-1001 depending on test
    Coarse-grid resolution controls interpolation accuracy; larger N improves accuracy at higher computational cost.
assumptions (6)
  • standard math Wiener-Khinchin theorem relates autocorrelation to PSD
    Used in Eq. (3) to define C(tau) from S(f).
  • domain assumption Relevant noise processes are Gaussian and stationary
    Assumed in Sec. IIA for spin noise and GWB models.
  • domain assumption Power-law PSDs require low- or high-frequency regularization
    Sec. IIA notes the integral in Eq. (13) diverges for power-law indices; the analysis uses cutoffs or breaks.
  • domain assumption Diagonal Fourier covariance is the standard PTA approximation
    Built into the comparison; the paper aims to show it is inaccurate.
  • domain assumption Linear interpolation of the autocorrelation is accurate on the coarse grid
    The method in Sec. IIC relies on smoothness of C(tau) for low-frequency-dominated processes.
  • domain assumption Quadratic projection mimics timing-model marginalization
    Sec. IIIA projects out constant, linear, and quadratic functions to approximate the effect of fitting the timing model.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond diagonal approximations: improved covariance modeling for pulsar timing array data analysis." pith.science (2026). https://pith.science/paper/JGSBQ4TV

@misc{pith2026250613866,
  author       = {Pith},
  title        = {Pith review of: Beyond diagonal approximations: improved covariance modeling for pulsar timing array data analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JGSBQ4TV}},
  note         = {Machine review of arXiv:2506.13866}
}
read the original abstract

Pulsar Timing Array (PTA) searches for nHz gravitational-wave backgrounds (GWBs) typically model time-correlated noise by assuming a diagonal covariance in Fourier space, neglecting inter-frequency correlations introduced by the finite observation window. We show that this diagonal approximation can lead to biased estimates of spectral parameters, especially for the common red process that represents the GWB. To address these limitations, we present a method that (i) computes the time-domain autocorrelation on a coarse grid using a fast Fourier transform (FFT), (ii) interpolates it accurately to the unevenly sampled observation times, and (iii) incorporates it into a low-rank likelihood via the Sherman--Morrison--Woodbury identity. Using both analytic covariance comparisons and end-to-end simulations inspired by the NANOGrav 15-year dataset, we demonstrate that our method captures frequency correlations faithfully, avoids Gibbs ringing, and recovers unbiased spectral parameters with modest computational cost. As PTA datasets increase in sensitivity and complexity, our approach offers a practical and scalable path to fully accurate covariance modeling for current and future analyses.

Figures

Figures reproduced from arXiv: 2506.13866 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Element-wise absolute difference between the exact and low-rank–approximated Matérn [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Mean absolute element-wise difference of the approximated and true covariance matrix as a function of the number [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Comparison of signal reconstruction with time- and frequency-domain bases. The blue curve depicts the “true” [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Change [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. Intrinsic timing-noise posteriors ( [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. Common red-noise posteriors ( [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8. Common red-noise posteriors ( [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: FIG. 9. Mean absolute element-wise difference [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 7 Pith papers

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

  1. A new framework for lightning-fast gravitational wave analysis of pulsar timing data

    gr-qc 2026-07 accept novelty 7.0 of 10

    A standardizing coordinate transform turns PTA Fourier coefficients into near-standard normals so HMC/NUTS on GPU recovers NANOGrav-scale posteriors in ~15 minutes.

  2. Modeling non-stationary noise: applications in gravitational wave astronomy

    gr-qc 2026-07 conditional novelty 6.0 of 10

    A positive dynamic spectrum S(f,t) generalizes the stationary power spectrum by defining Gramian closed-form noise covariances in Fourier and Wilson-Daubechies wavelet bases for gravitational wave data.

  3. Comparing gravitational wave background predictions from cosmological simulations to pulsar timing observations

    astro-ph.GA 2026-07 conditional novelty 6.0 of 10

    FABLE simulation predictions for the nanohertz gravitational wave background are statistically consistent with NANOGrav 15-year data at 1–2.5σ tension, with physically motivated population modifications further improv...

  4. Data span and frequency coverage requirements for robust detection and inference in PTAs: A case study with EPTA DR2

    astro-ph.HE 2025-11 conditional novelty 6.0 of 10

    Sparse early frequency coverage and noise fluctuations explain why EPTA DR2's shorter subset reports higher GWB significance than the full 25-year dataset.

  5. Finite Populations & Finite Time: The Non-Gaussianity of a Gravitational Wave Background

    gr-qc 2025-11 unverdicted novelty 6.0 of 10

    Finite source number and finite observation time make PTA gravitational-wave-background Fourier coefficients non-Gaussian, with an analytically computed excess kurtosis and a Cauchy-distributed argument probe.

  6. Mapping the Gravitational-wave Background Across the Spectrum with a Next-Generation Anisotropic Per-frequency Optimal Statistic

    astro-ph.IM 2025-09 conditional novelty 6.0 of 10

    A new pulsar-timing-array pipeline maps the gravitational-wave sky per frequency, folds cosmic variance into significance estimates, and detects a simulated loud source at p=0.01 versus 0.2 broadband.

  7. Directional Anisotropic Sensitivity Curves for Pulsar Timing Arrays

    astro-ph.IM 2026-07 conditional novelty 4.0 of 10

    PTA gravitational-wave sensitivity is mapped direction-by-direction, with radiometer and full-Fisher estimators bracketing the usual isotropic curve.

Reference graph

Works this paper leans on

35 extracted references · 14 canonical work pages · cited by 7 Pith papers

  1. [1]

    Timing a millisecond pulsar array,

    R. W. Romani, “Timing a millisecond pulsar array,” in Timing Neutron Stars , H. Ögelman and E. P. J. Heuvel, eds., pp. 113–117. Springer, 1989

    work page 1989

  2. [2]

    Constructing a Pulsar Timing Array,

    R. S. Foster and D. C. Backer, “Constructing a Pulsar Timing Array,” Astrophys. J. 361 (Sept., 1990) 300

    work page 1990

  3. [3]

    Upper limits on the isotropic gravitational radiation background from pulsar timing analysis,

    R. W. Hellings and G. S. Downs, “Upper limits on the isotropic gravitational radiation background from pulsar timing analysis,” Astrophys. J. Lett. 265 (Feb., 1983) L39–L42

    work page 1983

  4. [4]

    Expected properties of the first gravitational wave signal detected with pulsar timing arrays,

    P. A. Rosado, A. Sesana, and J. Gair, “Expected properties of the first gravitational wave signal detected with pulsar timing arrays,”Mon. Not. Roy. Astron. Soc. 451 no. 3, (06, 2015) 2417–2433

    work page 2015

  5. [5]

    Optimal strategies for gravitational wave stochastic background searches in pulsar timing data,

    M. Anholm, S. Ballmer, J. D. E. Creighton, L. R. Price, and X. Siemens, “Optimal strategies for gravitational wave stochastic background searches in pulsar timing data,” Phys. Rev. D 79 no. 8, (Apr., 2009) 084030, arXiv:0809.0701 [gr-qc]

    arXiv 2009

  6. [6]

    Time-domain implementation of the optimal cross-correlation statistic for stochastic gravitational-wave background searches in pulsar timing data,

    S. J. Chamberlin, J. D. E. Creighton, X. Siemens, P. Demorest, J. Ellis, L. R. Price, and J. D. Romano, “Time-domain implementation of the optimal cross-correlation statistic for stochastic gravitational-wave background searches in pulsar timing data,” Phys. Rev. D 91 no. 4, (Feb., 2015) 044048, 13 10.0 9.5 9.0 8.5 8.0 7.5 7.0 6.5 log10 f 11.0 10.5 10.0 9...

    arXiv 2015

  7. [7]

    Noise-marginalized optimal statistic: A robust hybrid frequentist-Bayesian statistic for the stochastic gravitational-wave background in pulsar timing arrays,

    S. J. Vigeland, K. Islo, S. R. Taylor, and J. A. Ellis, “Noise-marginalized optimal statistic: A robust hybrid frequentist-Bayesian statistic for the stochastic gravitational-wave background in pulsar timing arrays,” Phys. Rev. D 98 no. 4, (Aug., 2018) 044003, arXiv:1805.12188 [astro-ph.IM]

    arXiv 2018

  8. [8]

    Detection methods for stochastic gravitational-wave backgrounds: a unified treatment,

    J. D. Romano and N. J. Cornish, “Detection methods for stochastic gravitational-wave backgrounds: a unified treatment,” Living Rev. Rel. 20 no. 1, (2017) 2, 14 arXiv:1608.06889 [gr-qc]

    arXiv 2017

Show all 35 references
  1. [9]

    S. R. Taylor,Nanohertz Gravitational Wave Astronomy . CRC Press, Boca Raton, FL, 2021

    2021

  2. [10]

    Low-rank approximations for large stationary covariance matrices, as used in the Bayesian and generalized-least-squares analysis of pulsar-timing data,

    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,”Mon. Not. Roy. Astron. Soc.446 no. 2, (Jan., 2015) 1170–1174, arXiv:1407.6710 [astro-ph.IM]

    2015 arXiv

  3. [11]

    Realistic sensitivity curves for pulsar timing arrays,

    J. S. Hazboun, J. D. Romano, and T. L. Smith, “Realistic sensitivity curves for pulsar timing arrays,” Phys. Rev. D 100 no. 10, (2019) 104028, arXiv:1907.04341 [gr-qc]

    2019 arXiv

  4. [12]

    Pulsar timing array sensitivity to anisotropies in the gravitational wave background,

    P. F. Depta, V. Domcke, G. Franciolini, and M. Pieroni, “Pulsar timing array sensitivity to anisotropies in the gravitational wave background,”arXiv:2407.14460 [astro-ph.CO]

    arXiv

  5. [13]

    Mitigating cosmic variance in the Hellings-Downs curve: a Cosmic Microwave Background analogy,

    C. Pitrou and G. Cusin, “Mitigating cosmic variance in the Hellings-Downs curve: a Cosmic Microwave Background analogy,” arXiv:2412.12073 [gr-qc]

    arXiv

  6. [14]

    Accurate pulsar timing array residual variances and correlation of the stochastic gravitational wave background,

    R. C. Bernardo and K.-W. Ng, “Accurate pulsar timing array residual variances and correlation of the stochastic gravitational wave background,” arXiv:2409.01218 [astro-ph.CO]

  7. [15]

    ENTERPRISE: Enhanced Numerical Toolbox Enabling a Robust PulsaR Inference SuitE,

    J. A. Ellis, M. Vallisneri, S. R. Taylor, and P. T. Baker, “ENTERPRISE: Enhanced Numerical Toolbox Enabling a Robust PulsaR Inference SuitE,” Dec., 2019

    2019

  8. [16]

    DISCOVERY: the next-generation pulsar-timing-array data-analysis package,

    M. Vallisneriet al., “DISCOVERY: the next-generation pulsar-timing-array data-analysis package,” 2025. in preparation

    2025

  9. [17]

    JAX: composable transformations of Python+NumPy programs,

    J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. VanderPlas, S. Wanderman-Milne, and Q. Zhang, “JAX: composable transformations of Python+NumPy programs,” 2018. http://github.com/google/jax

    2018

  10. [18]

    Automatic differentiation in machine learning: a survey,

    A. G. Baydin, B. A. Pearlmutter, A. A. Radul, and J. M. Siskind, “Automatic differentiation in machine learning: a survey,”Journal of machine learning research18 no. 153, (2018) 1–43

    2018

  11. [19]

    The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo,

    M. D. Hoffman, A. Gelman,et al., “The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo,” J. Mach. Learn. Res. 15 no. 1, (2014) 1593–1623

    2014

  12. [20]

    Composable effects for flexible and accelerated probabilistic programming in NumPyro,

    D. Phan, N. Pradhan, and M. Jankowiak, “Composable effects for flexible and accelerated probabilistic programming in NumPyro,” (2019) ,arXiv:1912.11554

    2019 arXiv

  13. [21]

    Pyro: Deep universal probabilistic programming,

    E. Bingham, J. P. Chen, M. Jankowiak, F. Obermeyer, N. Pradhan, T. Karaletsos, R. Singh, P. A. Szerlip, P. Horsfall, and N. D. Goodman, “Pyro: Deep universal probabilistic programming,” J. Mach. Learn. Res. 20 (2019) 28:1–28:6. http://jmlr.org/papers/v20/18-403.html

    2019

  14. [22]

    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 no. 2, (May, 2009) 1005–1014,arXiv:0809.0791 [astro-ph]

    2009 arXiv

  15. [23]

    The NANOGrav 15 yr Data Set: Observations and Timing of 68 Millisecond Pulsars,

    G. Agazie et al., “The NANOGrav 15 yr Data Set: Observations and Timing of 68 Millisecond Pulsars,” Astrophys. J. Lett. 951 no. 1, (July, 2023) L9, arXiv:2306.16217 [astro-ph.HE]

    2023 arXiv

  16. [24]

    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. D 87 no. 10, (2013) 104021,arXiv:1210.3578 [astro-ph.IM]

    2013 arXiv

  17. [25]

    Understanding and analysing time-correlated stochastic signals in pulsar timing,

    R. van Haasteren and Y. Levin, “Understanding and analysing time-correlated stochastic signals in pulsar timing,” Mon. Not. Roy. Astron. Soc. 428 (2013) 1147, arXiv:1202.5932 [astro-ph.IM]

    2013 arXiv

  18. [26]

    New advances in the Gaussian-process approach to pulsar-timing data analysis,

    R. van Haasteren and M. Vallisneri, “New advances in the Gaussian-process approach to pulsar-timing data analysis,” Phys. Rev. D 90 no. 10, (Nov., 2014) 104012, arXiv:1407.1838 [gr-qc]

    2014 arXiv

  19. [27]

    The NANOGrav 15 yr Data Set: Evidence for a Gravitational-wave Background,

    NANOGrav Collaboration, G. Agazieet al., “The NANOGrav 15 yr Data Set: Evidence for a Gravitational-wave Background,” Astrophys. J. Lett. 951 no. 1, (2023) L8,arXiv:2306.16213 [astro-ph.HE]

    2023 arXiv

  20. [28]

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

    EPTACollaboration, J. Antoniadiset al., “The second data release from the European Pulsar Timing Array III. Search for gravitational wave signals,” (6, 2023) , arXiv:2306.16214 [astro-ph.HE]

    2023 arXiv

  21. [29]

    Search for an Isotropic Gravitational-wave Background with the Parkes Pulsar Timing Array,

    D. J. Reardonet al., “Search for an Isotropic Gravitational-wave Background with the Parkes Pulsar Timing Array,” Astrophys. J. Lett. 951 no. 1, (2023) L6, arXiv:2306.16215 [astro-ph.HE]

    2023 arXiv

  22. [30]

    The MeerKAT Pulsar Timing Array: the first search for gravitational waves with the MeerKAT radio telescope,

    M. T. Miles, R. M. Shannon, D. J. Reardon, M. Bailes, D. J. Champion, M. Geyer, P. Gitika, K. Grunthal, M. J. Keith, M. Kramer, and et al., “The MeerKAT Pulsar Timing Array: the first search for gravitational waves with the MeerKAT radio telescope,”Mon. Not. Roy. Astron. Soc. ...

    2025 arXiv

  23. [31]

    Searching for the Nano-Hertz Stochastic Gravitational Wave Background with the Chinese Pulsar Timing Array Data Release I,

    H. Xu et al., “Searching for the Nano-Hertz Stochastic Gravitational Wave Background with the Chinese Pulsar Timing Array Data Release I,”Res. Astron. Astrophys. 23 no. 7, (2023) 075024,arXiv:2306.16216 [astro-ph.HE]

    2023 arXiv

  24. [32]

    Fourier’s Series,

    J. W. Gibbs, “Fourier’s Series,”Nature (London) 59 no. 1522, (Dec., 1898) 200

  25. [33]

    On a certain periodic function,

    H. Wilbraham, “On a certain periodic function,”The Cambridge and Dublin Mathematical Journal 3 (1848) 198–201

  26. [34]

    Strang, Introduction to Applied Mathematics

    G. Strang, Introduction to Applied Mathematics . Wellesley-Cambridge Press, Wellesley, MA, 1986

    1986

  27. [35]

    NANOGrav 15-year gravitational-wave background methods,

    A. D. Johnsonet al., “NANOGrav 15-year gravitational-wave background methods,”Phys. Rev. D 109 no. 10, (May, 2024) 103012,arXiv:2306.16223 [astro-ph.HE]

    2024 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.