Pith. sign in

REVIEW 3 major objections 5 minor 50 references

The full spectral density matrix of correlated detector noise—diagonals and cross-terms—can be estimated nonparametrically from TDI channels, positive definite at every frequency and with posterior uncertainty.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 12:10 UTC pith:CLLVZ7CJ

load-bearing objection A coherent Bayesian nonparametric spectral-matrix estimator with a real misspecification in the LISA demo; the method deserves a serious referee, but the LISA uncertainty claims need a redo. the 3 major comments →

arxiv 2607.26619 v1 pith:CLLVZ7CJ submitted 2026-07-29 gr-qc astro-ph.HEastro-ph.IMstat.AP

Bayesian nonparametric estimation of correlated gravitational wave detector network noise using matrix-gamma process priors

Yixuan Liu , Renate Meyer , Nelson Christensen , Jeung Eun Lee , Jianan Liu , Patricio Maturana-Russel , Avi Vajpeyi This is my paper
classification gr-qc astro-ph.HEastro-ph.IMstat.AP MSC 62M1562F1562G05 PACS 04.80.Nn95.55.Ym
keywords matrix-gamma processBernstein polynomial priorspectral density matrixcorrelated noisegravitational-wave detectorsLISAEinstein Telescopeblocked Whittle likelihood
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Next-generation gravitational-wave detectors like LISA and the Einstein Telescope will produce multiple correlated data streams, and ignoring or fixing the noise correlations biases signal recovery. This paper develops a Bayesian nonparametric method, VNP, that estimates the entire spectral density matrix—each channel's PSD and the frequency-dependent cross-correlations—directly from the TDI XYZ channels, without assuming a parametric shape or relying on single-link noise models. A companion variant, VNP-P, builds in a rough parametric template and nonparametrically corrects its residual errors, recovering correlated noise peaks that the pure nonparametric version oversmooths. The prior guarantees Hermitian positive definite estimates at every frequency, and the sampler targets the exact posterior, so the resulting uncertainty can be propagated downstream. Simulations plus realistic LISA and ET noise injections support the claim, and the paper positions the exact MCMC sampler as a reference against which faster approximate methods can be assessed.

Core claim

The paper's central claim is that a matrix-gamma process prior placed on the matrix-valued weights of a Bernstein polynomial basis supplies a nonparametric prior over full spectral density matrices that is Hermitian positive definite at every frequency and, when updated with a blocked multivariate Whittle likelihood, yields a posterior that concentrates around the true spectrum as data accumulate. The blocked likelihood reduces to a product of complex Wishart densities for the averaged periodogram matrix, which is what makes long LISA-like records tractable. In the VNP-P extension, the spectral density is modeled as the matrix geometric mean of a fitted parametric spectrum and a nonparametri

What carries the argument

Matrix-gamma process prior on Bernstein polynomial coefficients: matrix-valued weights are increments of a matrix-gamma process, so every linear combination of Beta basis densities is automatically a Hermitian positive definite spectral matrix at each frequency. Blocked multivariate Whittle likelihood: segmenting the time series and averaging periodogram matrices turns the likelihood into a product of complex Wishart densities with Nb degrees of freedom, enabling long-record applications. VNP-P geometric-mean correction: the spectral density is parametrized as the matrix geometric mean of a fixed parametric fit Sp and a nonparametric spectrum Snp, so the nonparametric prior is placed on the

Load-bearing premise

The load-bearing premise is that the blocked Whittle likelihood with averaged, sometimes Hann-windowed periodogram matrices remains a complex Wishart distribution with the assumed degrees of freedom, and that the posterior-consistency guarantee proven for unsegmented data carries over to the many-block regime used in all applications; the paper does not prove this, and its own VAR(2) coverage numbers fall below nominal.

What would settle it

A concrete check: simulate many realizations of a bivariate process with a known spectral density matrix, apply the exact blocked likelihood with Hann windows and the same number of blocks as the LISA analysis (384 blocks of 16,384 points), and compute the empirical frequentist coverage of the reported 90% credible intervals across frequencies. If coverage is persistently far from 90%, the Wishart approximation—and therefore the posterior intervals—is invalid. A quicker distributional test compares the eigenvalues of the block-averaged periodogram with the expected Wishart law via a goodness-o

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the central claim holds, LISA and ET noise can be described by a full spectral density matrix estimated from the TDI XYZ channels alone, bypassing single-link test-mass and optical-metrology noise decomposition.
  • Because the estimate is Hermitian positive definite at every frequency by construction, it can be inverted inside the signal likelihood, and posterior samples can propagate noise uncertainty into gravitational-wave parameter estimates.
  • An adequately chosen parametric working model combined with the nonparametric correction resolves correlated spectral peaks that a purely nonparametric estimate oversmooths, at a modest computational cost.
  • The exact MCMC sampler can serve as a reference to calibrate the accuracy and uncertainty of faster approximation methods such as variational, coarsened-likelihood, and deep-learning point estimators.
  • The simulation study indicates insensitivity to block length within the tested range, which is what makes the method practical for very long records.

Where Pith is reading between the lines

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

  • If the blocked-likelihood Wishart assumption holds, the method is a natural plug-in for joint on-source signal-plus-noise inference in LISA, where no dedicated pure-noise stretches exist; the paper flags this as future work but does not demonstrate it.
  • The geometric-mean correction trick is generic: any parametric spectral template, not only VAR, could seed VNP-P, and the choice of template likely controls the bias-variance trade-off; the paper only explores VAR working models.
  • A direct stress test would be to run the sampler on long time series while increasing the number of blocks and check empirical coverage of credible intervals; the paper's own VAR(2) simulations show below-nominal coverage, suggesting the Wishart approximation may be the binding constraint.
  • The same matrix-gamma machinery transfers to any multichannel time-series problem where cross-spectral uncertainty matters, such as geophysical or neuroimaging data, though the paper is written for the gravitational-wave context.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper develops a Bayesian nonparametric method (VNP) for estimating the d×d spectral density matrix of stationary multivariate noise in gravitational-wave detector networks. The likelihood is a blocked Whittle/Wishart likelihood based on averaged periodogram matrices; the prior is a matrix-gamma process over Bernstein-polynomial coefficients, truncated at finite L in computation, and is designed to guarantee Hermitian positive definiteness at every frequency. A semiparametric variant VNP-P models a matrix geometric-mean correction around a fitted parametric (VAR) spectral template. The method is demonstrated on a 500-replicate simulation study for VAR(2) and VMA(1) processes, on simulated LISA TDI noise using Hann-windowed segments, and on ET-like three-channel noise with injected correlated Gaussian peaks. The central claims are flexibility over the full spectral matrix, positive definiteness by construction, sampling from the posterior via adaptive MCMC, and improved accuracy of VNP-P relative to VNP when the parametric template captures broad spectral structure.

Significance. If the claims hold, VNP would provide a useful benchmark for next-generation GW noise estimation, and the paper's code/data availability on GitHub and Zenodo is a genuine strength. The paper correctly identifies the need for full spectral-matrix, uncertainty-quantified noise models and provides a substantial simulation study. However, the evidence is currently weaker than the central claims: the LISA application feeds Hann-tapered periodograms into an unmodified complex-Wishart likelihood; the finite-block posterior-calibration properties are not established and the reported simulation coverage is below nominal for VAR(2); and the ET VNP-P comparison is in-sample. These issues are load-bearing for the claims of exactness and honest uncertainty, but they appear fixable within the manuscript's scope.

major comments (3)
  1. [§IV.A, Eq. (8)–(9)] The likelihood used for the LISA analysis is derived under the assumption that the per-block periodogram I^(i)(f_k) is CW_d(S(f_k),1), so that Y = N_b * mean(I) is CW_d(S,N_b). In the LISA application, however, a Hann window is applied to each segment before computing the DFT. A Hann-tapered periodogram is not complex Wishart with one degree of freedom: tapering reduces the effective degrees of freedom below 1, correlates adjacent frequencies, and biases the periodogram through the window main lobe. Since Eq. (9) is used unchanged, the sampled posterior is not the exact posterior for the actual tapered data. The paper's own observation of low-frequency underestimation 'caused by spectral leakage' is consistent with this misspecification. This needs to be addressed, e.g., by using untapered blocks, by modifying the likelihood to account for the taper, or by a simulation study demonstratin
  2. [§III and §II.A] The posterior-consistency guarantee is imported from [26] for unsegmented data (N_b=1), while every numerical experiment uses the blocked Wishart likelihood with N_b>1 (block lengths 256–1024 in the simulations, N_b=125 for ET, N_b=384 for LISA). No theorem is supplied for contraction of the blocked posterior, and finite-block behavior is exactly the regime used. Table I compounds this: for the VAR(2) case, pointwise coverage of the 90% credible intervals is only 0.66–0.67 for VNP and 0.75–0.76 for VNP-P, well below nominal. The text's appeal to [37] concerns univariate/unblocked settings. Since 'honest posterior uncertainty' is a central selling point, the authors should either provide a contraction result for the blocked likelihood or, at minimum, report coverage under the exact blocking and tapering protocols used in the applications.
  3. [§IV.B, Figs. 5–7] The VNP-P advantage is demonstrated in-sample. The autoregressive order p=303 is selected from the elbow plot in Fig. 5 using the same 2000-s realization on which the L2 errors (0.068 vs 0.078) and coherence peaks are then reported. The paper notes that the elbow criterion suggests p=7 but that p=303 is chosen because it captures the injected peaks; this is a data-dependent model selection. A VAR(303) for d=3 contains roughly d^2 p = 2727 coefficient parameters, so overfitting is a real concern. To support the claim that VNP-P improves accuracy, order selection should be performed on a training set (or by cross-validation) and the comparison should be reported on independent test data or repeated over many noise realizations.
minor comments (5)
  1. [Table I caption] The caption says 'median computation time (in seconds)', but the table row is labeled 'Time [min]' and the values are on the order of tens of minutes. Correct the inconsistency.
  2. [Fig. 4] The legend 'Analytical fit' is not defined in the text. State the analytic formula for the squared coherence and how it is computed.
  3. [§II.B, Eq. (15)] The prior is truncated at finite L, so the sampler targets the posterior of the truncated process, not the full matrix-gamma process. The Introduction's phrase 'sampled from the exact posterior rather than an approximation' should be qualified accordingly.
  4. [§II.B] The hyperparameters α0, β0, Σ0, and c are fixed without sensitivity analysis. A brief robustness check would strengthen the claim that the prior is noninformative in practice.
  5. [§IV.A] The suggestion that low-frequency underestimation 'could be suppressed by more aggressive tapering' is unclear, since a Hann window was already applied and stronger tapering would further reduce the effective degrees of freedom. Clarify the intended remedy.

Circularity Check

0 steps flagged

No circular derivation: estimation is data-driven and the main guarantees are external published results; flagged issues are in-sample model selection and likelihood misspecification, not circularity.

full rationale

The derivation chain is not circular. The blocked Whittle/Wishart likelihood (Eqs. 5-9) is a standard approximation derived from the DFT/periodogram distribution; the matrix-gamma Bernstein prior is adopted from Meier et al. [26], a published, parameter-free consistency result for Nb=1, and the paper explicitly flags that the blocked-case transfer is an illustration rather than a proven theorem: 'For unsegmented multivariate time series (Nb=1), [26] established posterior consistency ... Here, we illustrate the performance ... with increasing block lengths.' That is an unproven assumption (a correctness risk), not a definitional identity. VNP-P's parametric template S_p is fitted, but the final estimate is a geometric-mean-corrected posterior median, not equal to S_p by construction; no fitted parameter is renamed as a prediction. The ET order p=303 is selected from the same data used to report L2 error ('We therefore select order 303 for the parametric VAR fit'), making that accuracy comparison in-sample; this is a validation/overfitting caveat, not a circular reduction. The LISA application applies a Hann window ('A Hann window was applied to each segment to reduce spectral leakage') while retaining the Wishart likelihood, so the 'exact posterior' claim is misspecified there; again this is internal-model inconsistency, not circularity. Self-citations [26,30,37] overlap with the authors, but the load-bearing consistency result is a published external theorem with stated assumptions, and no uniqueness theorem or ansatz is smuggled in via citation. Hence no step reduces to its own inputs; score 2 reflects minor self-citation and in-sample-selection caveats rather than substantive circularity.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 0 invented entities

No new physical entities are introduced. The method rests on standard spectral-analysis approximations and on imported matrix-gamma-process constructions from prior work. The strongest non-standard imports are the complex-Wishart approximation for block-averaged periodograms and the posterior-consistency theorem from [26], which is cited for Nb=1 while the paper's applications all use blocked data.

free parameters (6)
  • VNP prior hyperparameters (α0, β0, Σ0) = α0=β0=d=3, Σ0=10^4 I_d
    Chosen as a noninformative prior setting in §II.B; not fitted to data but set by hand and controls prior spread.
  • Polynomial-order penalty c in p(m) ∝ exp(-c m log m) = 0.01
    Chosen from [26,30] in §II.B; controls the prior on the Bernstein polynomial order and hence smoothness.
  • Truncation level L of the matrix-gamma process series representation = max{20, (n/Nb)^(1/3)}
    Chosen in §II.C as a conservative truncation from [26,30]; affects the fidelity of the infinite-series prior approximation.
  • ET VNP-P VAR order p = 303
    Selected by an elbow criterion on the same ET data in §IV.B; p=7 is first suggested but rejected because it does not capture the injected correlated peaks.
  • VAR(p) coefficient matrices for the VNP-P working model = Maximum blocked-Whittle-likelihood estimates
    Fitted from the ET data in §IV.B before applying the nonparametric correction; these form the parametric template S_p.
  • Blocking choices (number and length of blocks) = LISA: 384 blocks of 16384; ET: 125 blocks of 32768; simulations: n/Nb = 256, 512, 1024
    Chosen by hand to balance frequency resolution, Wishart degrees of freedom, and MCMC cost (§III, §IV).
axioms (5)
  • domain assumption DFT coefficients are asymptotically independent complex Gaussian with covariance S(f_k) (multivariate Whittle approximation)
    Invoked in Eqs. (3)–(4), §II.A; standard asymptotic approximation for stationary time series.
  • domain assumption Non-overlapping blocks are independent and their averaged periodogram is complex Wishart CW_d(S(f_k), Nb)
    Eqs. (5)–(8), §II.A; needed for the closed-form blocked likelihood. Not exact for finite windows, and the Hann window used in §IV.A is not accounted for.
  • domain assumption Posterior consistency of the matrix-gamma Bernstein prior under the multivariate Whittle likelihood (Meier et al. 2020) carries over to the blocked setting
    Imported in §III: '[26] established posterior consistency' for Nb=1, but all applications use Nb>1 with no proof of contraction in that setting.
  • standard math Bernstein polynomials with matrix-gamma increments can represent smooth Hermitian positive definite spectral density matrices
    Eqs. (12)–(14), §II.B; standard density-approximation construction from [26].
  • standard math The matrix geometric mean of positive definite matrices is well-defined and preserves positive definiteness
    Eqs. (10)–(11), §II.B, using the Ando–Li–Mathias geometric mean [31].

pith-pipeline@v1.3.0-daily-deepseek · 15400 in / 16271 out tokens · 169452 ms · 2026-08-01T12:10:46.905918+00:00 · methodology

0 comments
read the original abstract

This paper addresses the important problem of estimating the noise spectral density of next-generation gravitational-wave detectors, such as LISA and the Einstein Telescope (ET), where cross-channel correlations must be accounted for to avoid biased parameter estimation of gravitational-wave signals. Unlike approaches that estimate test-mass and optical-metrology-system noise separately at the single-link level and then map them to the Time-Delay Interferometry (TDI) channels through known transfer functions, we develop a Bayesian nonparametric method that directly estimates the spectral density matrix of the XYZ channels, thereby accommodating additional sources of uncertainty. Our approach combines a flexible matrix-gamma process prior on the matrix-valued coefficients of a Bernstein polynomial basis expansion with a blocked multivariate Whittle likelihood. The prior guarantees Hermitian positive definiteness of the spectral estimate at every frequency. To avoid reversible-jump methods, we use an adaptive Markov chain Monte Carlo (MCMC) algorithm for posterior sampling. The proposed framework can also be used to correct misspecified parametric noise models. Results from a simulation study and simulated correlated-noise data for both LISA and ET demonstrate the effectiveness of the proposed method.

Figures

Figures reproduced from arXiv: 2607.26619 by Avi Vajpeyi, Jeung Eun Lee, Jianan Liu, Nelson Christensen, Patricio Maturana-Russel, Renate Meyer, Yixuan Liu.

Figure 1
Figure 1. Figure 1: The top panel displays the Bernstein polynomial basis of order m = 10, the middle panel a draw from a matrix-gamma process prior, and the bottom panel the corresponding mixture representation of the bivariate PSD matrix with auto-spectra on the diagonal and real and imaginary parts of the cross-spectrum on the off-diagonals. III. SIMULATION STUDY For unsegmented multivariate time series (Nb = 1), [26] esta… view at source ↗
Figure 2
Figure 2. Figure 2: Median estimates for the VAR(2) model from [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Median VNP estimates of the PSD of the simulated LISA noise are shown in red. The true PSD is [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 7
Figure 7. Figure 7: Although VNP accurately reconstructs the co- [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 4
Figure 4. Figure 4: Median VNP estimates of the coherence of [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Negative log-Whittle-likelihood for prefitted VAR models with different autoregressive orders. The inset [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Median estimates of the PSD of the ET data from VNP (red) and VNP-P(303) (blue). The periodogram [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Median estimates of the squared coherence of [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗

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

50 extracted references · 10 linked inside Pith

  1. [1]

    T. B. Littenberg and N. J. Cornish, Bayesian inference for spectral estimation of gravitational wave detector noise, Physical review. D, Particles, fields, gravitation, and cos- mology91(2015)

    2015

  2. [2]

    This is a special multivariate generalization of the univariate gamma distribution

    so that the incrementsG(dx) have an independent Hermitian positive definite matrix-gamma distribution [26]. This is a special multivariate generalization of the univariate gamma distribution. This construction is il- lustrated in Figure 1 in the bivariate case for a fixed poly- nomial orderm= 10. Using different colours, at the top are the ten normalized ...

  3. [3]

    The theoretical results were complemented by simulation studies that empiri- cally confirmed this asymptotic behaviour for increasing time series lengths

    established posterior consistency of VNP by prov- ing that, as the sample size increases, the posterior dis- tribution under the multivariate Whittle likelihood and matrix-Gamma process prior concentrates around the true spectral density matrix. The theoretical results were complemented by simulation studies that empiri- cally confirmed this asymptotic be...

  4. [4]

    P. M. Meyers, K. Martinovic, N. Christensen, and M. Sakellariadou, Detecting a stochastic gravitational- wave background in the presence of correlated mag- netic noise, Physical Review D102, 102005 (2020), arXiv:2008.00789 [gr-qc]

    Pith/arXiv arXiv 2020

  5. [5]

    Throughout this paper, these realizations are collectively referred to asET noise

    The simulated data were sampled at 2048 Hz, produc- ing multichannel time series containing 4096 k samples in each detector channel. Throughout this paper, these realizations are collectively referred to asET noise. To model correlated instrumental disturbances, we fol- low the approach of Janssenset al.[43]. Additional colored Gaussian noise is injected ...

    2048

  6. [6]

    Janssens, G

    K. Janssens, G. Boileau, N. Christensen, F. Badaracco, and N. van Remortel, Impact of correlated seismic and correlated Newtonian noise on the Einstein Telescope, Physical Review D106, 042008 (2022), arXiv:2206.06809 [astro-ph.IM]

    Pith/arXiv arXiv 2022

  7. [7]

    Chatziioannou, C.-J

    K. Chatziioannou, C.-J. Haster, T. B. Littenberg, W. M. Farr, S. Ghonge, M. Millhouse, J. A. Clark, and N. Cor- nish, Noise spectral estimation methods and their impact on gravitational wave measurement of compact binary mergers, arXiv.org (2019)

    2019

  8. [8]

    Thrane, N

    E. Thrane, N. Christensen, and R. Schofield, Correlated magnetic noise in global networks of gravitational-wave interferometers: observations and implications, Physi- cal Review D87, 123009 (2013), arXiv:1303.2613 [astro- ph.IM]

    Pith/arXiv arXiv 2013

  9. [9]

    Christensen and R

    N. Christensen and R. Meyer, Parameter estimation with gravitational waves, Rev. Mod. Phys.94, 025001 (2022)

    2022

  10. [10]

    K. Janssenset al., Correlated 1–1000 Hz magnetic field fluctuations from lightning over Earth-scale distances and their impact on gravitational wave searches, Physical Review D107, 022004 (2023), arXiv:2209.00284 [gr-qc]

    Pith/arXiv arXiv 2023

  11. [11]

    Janssens, K

    K. Janssens, K. Martinovic, N. Christensen, P. M. Mey- ers, and M. Sakellariadou, Impact of schumann reso- nances on the einstein telescope and projections for the magnetic coupling function, Physical Review D104, 122006 (2021)

    2021

  12. [12]

    K. Janssenset al., Correlated 0.01–40 Hz seismic and Newtonian noise and its impact on future gravitational- wave detectors, Physical Review D109, 102002 (2024), arXiv:2402.17320 [gr-qc]

    Pith/arXiv arXiv 2024

  13. [13]

    Schillings, S

    P. Schillings, S. Yao, J. Erdmann, and A. Rietbrock, A numerical framework for newtonian-noise estimation at the einstein telescope: 2-d simulations beyond the plane- wave approximation (2026), arXiv:2603.15424 [astro- ph.IM]

    arXiv 2026

  14. [14]

    Cireddu, M

    F. Cireddu, M. Wils, I. C. F. Wong, P. T. H. Pang, T. G. F. Li, and W. Del Pozzo, Likelihood for a network of gravitational-wave detectors with correlated noise, Physical Review D110, 104060 (2024)

    2024

  15. [15]

    Badaracco and J

    F. Badaracco and J. Harms, Optimization of seismome- ter arrays for the cancellation of newtonian noise from seismic body waves, Classical and Quantum Gravity36, 145006 (2019)

    2019

  16. [16]

    Baghi, N

    Q. Baghi, N. Karnesis, J.-B. Bayle, M. Besan¸ con, and H. Inchausp´ e, Uncovering gravitational-wave back- grounds from noises of unknown shape with lisa, Journal of cosmology and astroparticle physics2023, 66 (2023)

    2023

  17. [17]

    Janssens, G

    K. Janssens, G. Boileau, N. Christensen, F. Badaracco, and N. van Remortel, Impact of correlated seismic and correlated newtonian noise on the einstein telescope, Physical Review D106, 042008 (2022)

    2022

  18. [18]

    Janssens, G

    K. Janssens, G. Boileau, M.-A. Bizouard, N. Christensen, T. Regimbau, and N. v. Remortel, Formalism for power 12 spectral density estimation for non-identical and corre- lated noise using the null channel in einstein telescope, European physical journal plus138(2023)

    2023

  19. [19]

    Hartig, S

    M.-S. Hartig, S. Paczkowski, M. Hewitson, G. Heinzel, and G. Wanner, Postprocessing subtraction of tilt-to- length noise in lisa in the presence of gravitational wave signals, Physical Review D111, 043048 (2025)

    2025

  20. [20]

    Quang Nam, J

    D. Quang Nam, J. Martino, Y. Lemi` ere, A. Petiteau, J.- B. Bayle, O. Hartwig, and M. Staab, Time-delay interfer- ometry noise transfer functions for lisa, Physical Review D108, 082004 (2023)

    2023

  21. [21]

    Boileau, N

    G. Boileau, N. Christensen, R. Meyer, and N. J. Cor- nish, Spectral separation of the stochastic gravitational- wave background for LISA: Observing both cosmological and astrophysical backgrounds, Physical Review D103, 103529 (2021)

    2021

  22. [22]

    Santini, M

    A. Santini, M. Muratore, J. Gair, and O. Hartwig, Flex- ible, gpu-accelerated approach for the joint characteriza- tion of LISA instrumental noise and stochastic gravita- tional wave backgrounds, Physical Review D112, 084050 (2025)

    2025

  23. [23]

    Muratore, J

    M. Muratore, J. Gair, and L. Speri, Impact of the noise knowledge uncertainty for the science exploitation of cos- mological and astrophysical stochastic gravitational wave background with lisa, Physical Review D109, 042001 (2024)

    2024

  24. [24]

    J. Liu, A. Vajpeyi, R. Meyer, K. Janssens, J. E. Lee, P. Maturana-Russel, N. Christensen, and Y. Liu, Varia- tional inference for correlated gravitational wave detector network noise (2024), arXiv:2409.13224 [gr-qc]

    Pith/arXiv arXiv 2024

  25. [25]

    P. Welch, The use of fast fourier transform for the estima- tion of power spectra: A method based on time averaging over short, modified periodograms, IEEE transactions on audio and electroacoustics15, 70 (1967)

    1967

  26. [26]

    We will first describe the nonparametric prior for the univariate case and then its generalization to multivariate spectral densities

    showed good asymptotic properties such as poste- rior consistency which guarantees that the posterior dis- tribution will contract around the true spectral density with increasing sample size. We will first describe the nonparametric prior for the univariate case and then its generalization to multivariate spectral densities. A univariate spectral density...

  27. [27]

    P. Tang, R. Meyer, and J. J. Eldridge, Gravitational wave energy spectral density properties from bpass galactic binary population in the milky way galaxy, Monthly Notices of the Royal Astronomical Society543, 2855 (2025), https://academic.oup.com/mnras/article- pdf/543/3/2855/64328492/staf1558.pdf

    2025

  28. [28]

    D. M. Blei, A. Kucukelbir, and J. D. McAuliffe, Vari- ational inference: A review for statisticians, Journal of the American Statistical Association112, 859 (2017), https://doi.org/10.1080/01621459.2017.1285773

    arXiv 2017

  29. [29]

    M. Dax, S. R. Green, J. Gair, M. P¨ urrer, J. Wildberger, J. H. Macke, A. Buonanno, and B. Sch¨ olkopf, Neural importance sampling for rapid and reliable gravitational- wave inference, Phys. Rev. Lett.130, 171403 (2023)

    2023

  30. [30]

    Vajpeyi, R

    A. Vajpeyi, R. Meyer, P. Maturana-Russel, and J. Liu, Multivariate Bayesian P-spline estimation of spectral density matrices, with application to LISA TDI noise (2026), arXiv:2607.04833 [stat.ME]

    Pith/arXiv arXiv 2026

  31. [31]

    Meier, C

    A. Meier, C. Kirch, and R. Meyer, Bayesian nonparamet- ric analysis of multivariate time series: A matrix gamma process approach, Journal of multivariate analysis175, 104560 (2020)

    2020

  32. [32]

    Diaconis and D

    P. Diaconis and D. Freedman, On the consistency of Bayes estimates, The Annals of Statistics14, 1 (1986)

    1986

  33. [33]

    Flauger, N

    R. Flauger, N. Karnesis, G. Nardini, M. Pieroni, A. Ric- ciardone, and J. Torrado, Improved reconstruction of a stochastic gravitational wave background with LISA, Journal of cosmology and astroparticle physics2021, 59 (2021)

    2021

  34. [34]

    Sethuraman, A constructive definition of Dirichlet pri- ors, Statistica Sinica , 639 (1994)

    J. Sethuraman, A constructive definition of Dirichlet pri- ors, Statistica Sinica , 639 (1994)

    1994

  35. [35]

    Y. Liu, C. Kirch, J. E. Lee, and R. Meyer, A nonparamet- rically corrected likelihood for Bayesian spectral analysis of multivariate time series, Computational Statistics & Data Analysis199, 108010 (2024)

    2024

  36. [36]

    Ando, C.-K

    T. Ando, C.-K. Li, and R. Mathias, Geometric means, Linear algebra and its applications385, 305 (2004)

    2004

  37. [37]

    R. T. Krafty and W. O. Collinge, Penalized multivari- ate Whittle likelihood for power spectrum estimation, Biometrika100, 447 (2013)

    2013

  38. [38]

    Choudhuri, S

    N. Choudhuri, S. Ghosal, and A. Roy, Bayesian estima- tion of the spectral density of a time series, Journal of the American Statistical Association99, 1050 (2004)

    2004

  39. [39]

    Tinto and S

    M. Tinto and S. V. Dhurandhar, Time-delay interferom- etry, Living Reviews in Relativity24(2021)

    2021

  40. [40]

    Freedman, On the bernstein-von mises theorem with infinite-dimensional parameters, The Annals of Statistics 27, 1119 (1999)

    D. Freedman, On the bernstein-von mises theorem with infinite-dimensional parameters, The Annals of Statistics 27, 1119 (1999)

    1999

  41. [41]

    B. T. Knapik, A. W. van der Vaart, and J. H. van Zan- ten, Bayesian inverse problems with gaussian priors, The Annals of statistics39, 2626 (2011)

    2011

  42. [42]

    Kirch, M

    C. Kirch, M. C. Edwards, A. Meier, and R. Meyer, Be- yond whittle: Nonparametric correction of a parametric likelihood with a focus on bayesian time series analysis, Bayesian analysis14, 1037 (2019)

    2019

  43. [44]

    Janssens, M

    K. Janssens, M. Ball, R. M. S. Schofield, N. Christensen, R. Frey, N. van Remortel,et al., Correlated 1–1000 hz magnetic field fluctuations from lightning over earth- scale distances and their impact on gravitational wave searches, Physical Review D107, 022004 (2023)

    2023

  44. [45]

    S. Hild, S. Chelkowski, A. Freise, J. Franc, N. Morgado, R. Flaminio, and R. DeSalvo, A xylophone configuration for a third-generation gravitational wave detector, Clas- sical and Quantum Gravity27, 015003 (2009)

    2009

  45. [46]

    Hildet al., Sensitivity Studies for Third-Generation Gravitational Wave Observatories, Class

    S. Hildet al., Sensitivity Studies for Third-Generation Gravitational Wave Observatories, Class. Quant. Grav. 28, 094013 (2011), arXiv:1012.0908 [gr-qc]

    Pith/arXiv arXiv 2011

  46. [47]

    Branchesi, M

    M. Branchesi, M. Maggiore, D. Alonso, C. Badger, B. Banerjee, F. Beirnaert, E. Belgacem, S. Bhagwat, G. Boileau, S. Borhanian,et al., Science with the Einstein Telescope: a comparison of different designs, J. Cosmol. Astropart. Phys.2023(7), 068, arXiv:2303.15923 [gr-qc]

    Pith/arXiv arXiv 2023

  47. [48]

    Janssens, G

    K. Janssens, G. Boileau, M.-A. Bizouard, N. Christensen, T. Regimbau, and N. v. Remortel, Formalism for power spectral density estimation for non-identical and cor- related noise using the null channel in Einstein Tele- scope, European Physical Journal Plus138, 352 (2023), arXiv:2205.00416 [gr-qc]

    Pith/arXiv arXiv 2023

  48. [50]

    Janssens, G

    K. Janssens, G. Boileau, N. Christensen, N. van Re- mortel, F. Badaracco, B. Canuel, A. Cardini, A. Contu, M. W. Coughlin, J.-B. Decitre, R. De Rosa, M. Di Gio- vanni, D. D’Urso, S. Gaffet, C. Giunchi, J. Harms, S. Ko- ley, V. Mangano, L. Naticchioni, M. Olivieri, F. Pao- letti, D. Rozza, D. O. Sabulsky, S. Shani-Kadmiel, 13 and L. Trozzo, Correlated 0.01...

    doi:10.1103/phys- 2024

  49. [51]

    Bayesian nonparametric estima- tion of correlated gravitational wave detector network noise using matrix-gamma process priors

    Y. Liu, R. Meyer, N. Christensen, J. E. Lee, , J. Liu, P. Maturana-Russel, and A. Vajpeyi, Dataset and R- package accompaning “Bayesian nonparametric estima- tion of correlated gravitational wave detector network noise using matrix-gamma process priors”, 10.5281/zen- odo.21281721 (2026)

    doi:10.5281/zen- 2026

  50. [1024]

    The 90% pointwise credible interval of VNP-P is shaded in blue

    The true PSD is represented by a black solid line. The 90% pointwise credible interval of VNP-P is shaded in blue. The residual plot for each PSD component is stacked below its corresponding estimate plot. wrong rate and lead to poor coverage. Only for paramet- ric models under certain regularity conditions, the Bern- stein von Mises theorem guarantees th...

    2000