Pith. sign in

REVIEW 5 major objections 4 minor 38 references

A Kalman-smoother based data imputation strategy to data gaps in spaceborne gravitational wave detectors

T0 review · 5 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A Kalman smoother can fill short data gaps in spaceborne gravitational wave observations without biasing the recovered binary parameters.

desk verdict A plausible proof-of-concept for Kalman-smoother gap imputation in MBHB TDI data, but the single-demonstration evidence base and a misspecified likelihood for colored noise mean the stronger claims are not yet established. read the letter →

arxiv 2507.02458 v1 pith:SIWTMAJ4 submitted 2025-07-03 astro-ph.IM gr-qc

Tingyang Shen , He Wang , Jibo He This is my paper
classification astro-ph.IMgr-qc
keywords KalmansmootherdatagapsgravitationalwaveanalysismassiveblackholebinariesparameterestimationimputationLISATaiji
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

The paper proposes using a Kalman filter and smoother to fill in missing stretches of time-series data from spaceborne gravitational wave observatories such as LISA and Taiji, so that parameter estimation of massive black hole binary mergers is not biased by gaps. In a simulated one-year observation with many short random gaps totaling about 1% of the data, the repaired posteriors for chirp mass, mass ratio, and spins match the no-gap optimal posteriors, whereas conventional windowing shifts mass ratio and spins by roughly 3 standard deviations. The imputation is computationally lighter than Bayesian noise inpainting because it only needs local linear predictions and noise covariances learned by expectation-maximization, not full signal and noise sampling. The authors also identify a boundary: a single 3-day gap placed 0.2 days before coalescence is too much for the linear model, widening and slightly biasing the recovered posterior. If correct, this gives a fast, training-free default gap treatment for massive-black-hole-binary data analysis.

What carries the argument

The central object is a linear-Gaussian state-space model whose three-dimensional state carries the local signal value and its first two time derivatives, $(h, \dot h, \ddot h)$, with a transition matrix $A$ built from a second-order Taylor expansion. The Kalman filter propagates this state forward through observed points, the Rauch-Tung-Striebel smoother runs backward to refine the estimates, and an expectation-maximization routine learns the transition and observation noise covariances $Q$ and $R$ from the observed segments. Smoothed samples replace the zeroed gap points, producing a repaired time series that can be fed into standard Bayesian parameter estimation without discarding neighboring data or assuming the noise is nonstationary.

What would settle it

Take an injected massive-black-hole-binary with known parameters and replace a contiguous 3-day stretch 0.2 days before coalescence; if the Kalman-repaired posterior's credible interval does not contain the injected values while the no-gap posterior does, the method's bias-mitigation claim is limited to the short-gap regime, and a sweep of gap length would pinpoint where the repair breaks down.

Watch

Extended reading notes

Core claim

The paper's central claim is that Kalman filtering and smoothing provide a practical way to impute missing stretches of massive-black-hole-binary data: for randomized gaps of about ten minutes each, totaling roughly one percent of a one-year observation, the repaired time series yields posterior distributions for chirp mass, mass ratio, and spins that are unbiased and close to the no-gap case, while windowing shifts mass ratio and both spins by about three standard deviations. The same framework is claimed to be computationally cheaper than noise inpainting because the state dimension is only three, giving per-iteration complexity $O(N n^3)$ that is effectively linear in the number of data points, versus noise inpainting's $O(N \log N) + O(M^2)$ with $M$ the number of missing points. The authors also explicitly delimit the method: for a single three-day gap placed 0.2 days before coalescence, the repaired posterior widens and remains slightly biased, which they attribute to the breakdown of the local linear approximation.

Load-bearing premise

The load-bearing premise is that across a gap the signal is locally smooth enough for a second-order Taylor expansion in a linear-Gaussian model to predict the missing samples; for long gaps or gaps near coalescence this fails, as the paper's own 3-day-gap test shows.

Editorial extensions

If this is right

  • For randomized short-gap patterns with up to about 1% data loss, Kalman imputation can replace windowing as the default gap treatment for massive-black-hole-binary sources.
  • Because the method scales linearly with the number of time steps and cubically only with the small state dimension, it becomes increasingly favorable over noise inpainting as observations get longer or more finely sampled.
  • The imputation requires no labeled training data and only a chirping-signal assumption plus EM-learned covariances, so it can be applied before precise source parameters are known.
  • The demonstrated failure on a long gap near coalescence defines a clear boundary, suggesting a hybrid pipeline: Kalman imputation for benign gaps and separate treatment for long gaps close to merger.
  • Since the imputation takes only a fraction of the total parameter-estimation time and was run on a laptop GPU, it is practical for production-scale analysis.
  • Repaired-data posteriors reproduce the same extrinsic-parameter structure as the optimal no-gap case, so gap repair does not introduce new multimodality or range inflation.

Reading between the lines

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

  • Inference: the same local linear predictor could be upgraded to nonlinear Kalman variants such as the unscented or extended Kalman filter, which the authors mention as future work; a testable extension is whether these variants prolong the range of gap lengths that can be repaired without bias.
  • Inference: the method processes the A, E, and T channels separately, but a joint multi-channel state-space model might exploit correlations between TDI channels to carry information across longer gaps.
  • Inference: because EM learns noise covariances from observed segments, nonstationary or mismodeled noise could be partially absorbed into the learned $R$; comparing the learned covariance with the known PSD would quantify how much noise mismatch leaks into the imputation.
  • Inference: for other long-duration sources such as extreme-mass-ratio inspirals, the linear-Gaussian assumption may be less valid; testing the method on EMRI gap patterns would map its boundary beyond the heavy MBHB case studied here.
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

5 major / 4 minor

Summary. This paper proposes to impute missing data in spaceborne gravitational-wave observations using a Kalman filter and Rauch-Tung-Striebel smoother. The signal is modeled locally by a three-dimensional state with a second-order Taylor transition matrix (Eqs. 11-12), and the transition and observation covariance matrices Q and R are estimated by expectation-maximization. Gaps are filled with smoothed estimates, and the repaired time series is used in a standard frequency-domain likelihood for Bayesian parameter estimation. The method is tested on one injected IMRPhenomD MBHB signal in Taiji TDI A/E/T channels, with many short random gaps (about 1% data loss) and one additional long 3-day gap near coalescence. The authors compare posterior distributions from optimal (no-gap), windowed, and Kalman-repaired data, and report that the repaired posteriors recover the optimal posteriors and remove the bias seen with windowing, while also claiming lower computational cost than noise inpainting.

Significance. If the claims hold, this would provide a fast, training-free gap-imputation tool for MBHB sources in LISA/Taiji/Tianqin, complementing windowing and inpainting. The paper's strengths are its use of standard waveform and noise pipelines (bbhx, Taiji PSD), its comparison against both an optimal-data benchmark and a windowing baseline, and the explicit demonstration of a failure mode for a long gap near coalescence (Fig. 5). The method's O(N n^3) scaling with a low-dimensional state is attractive for long datasets. However, the current evidence is a single noise realization with one gap pattern, the state-space model is misspecified for strongly colored LISA/Taiji noise, and the imputation uncertainty is not propagated into the likelihood; these issues must be addressed before the general claims of bias mitigation and reduced computational cost can be accepted.

major comments (5)
  1. [III.B, Figs. 3-4] The entire demonstration of bias mitigation rests on a single injected waveform, one noise realization, and one randomly generated gap pattern. The paper reports no repeated trials over noise realizations or gap patterns and no aggregate bias statistic such as mean offset and scatter over realizations. A single posterior whose mode happens to overlap the optimal-data posterior cannot establish that the method is unbiased, especially because the gapped-data bias itself is quantified only qualitatively as 'almost 3 sigma' in the caption. Please provide repeated-realization results or clearly state the scope as a proof-of-principle.
  2. [II.B.2, Eq. (7) vs. Eq. (2), Eq. (14)] The state-space model assumes white observation noise v_t ~ N(0,R), whereas the simulated noise is strongly colored (Eq. 2). With a three-dimensional state and a single R matrix, EM cannot reproduce the frequency-dependent PSD. The imputed gap samples are therefore not draws from the stationary Gaussian process with PSD Sn(f) used in the likelihood (Eqs. 14-15). The paper never checks whether the repaired residuals inside gaps have the expected PSD or whiteness. Add such validation (e.g., compare the PSD of residuals in repaired gaps to Sn(f), or run a whiteness/stationarity test) and, if necessary, generalize the state-space model or amend the likelihood.
  3. [II.B.2 and Eqs. 13-15] The imputation fills gaps with the smoothed point estimates and then evaluates the likelihood as if these were actual observations. This ignores the posterior uncertainty of the missing data (single imputation rather than multiple imputation or proper marginalization). The close agreement of the repaired and optimal credible intervals may therefore reflect the likelihood treating the imputed points as independent data, not genuine information recovery. Please quantify the effect of imputation uncertainty, for example by comparing with a multiple-imputation approach or by down-weighting the likelihood contribution of imputed samples.
  4. [Table II and Figs. 3-4] There is an unresolved inconsistency in the mass-ratio convention: Table II lists q=0.478, while the posterior marginals in Figs. 3 and 4 are all centered at q approximately 2.09, which is close to 1/0.478. If the sampler uses q=m1/m2 and the table lists m2/m1, this must be stated; as written, the claimed 'almost 3 sigma' bias for q cannot be reproduced -- the displayed q posterior is either centered exactly on the injected value (if q=m1/m2=2.09) or is tens of sigma away from the table value (if q=0.478). Please define the convention and correct the text and figures accordingly.
  5. [Table IV and III.B] The abstract's claim of 'lower computational cost compared to existing data augmentation techniques such as noise inpainting' is supported only by an asymptotic complexity table, not by a measured comparison. The text reports that imputation took 'several thousands of seconds' but gives no wall-clock time for noise inpainting on the same data, no accounting for the EM iterations in the Kalman cost, and no measurement of time-to-convergence for the subsequent MCMC. Either provide an empirical comparison or soften the claim to a complexity-based expectation.
minor comments (4)
  1. [II.B.2, Eq. (8) and Eq. (10)] The Kalman filter and smoother equations contain malformed subscripts and superscripts; for example, Eq. (8) appears to read 'x_t^{t-1} = A x_{t-1}^{t-1}' but is typeset incorrectly, and the gain in Eq. (10) has unclear indexing. These should be corrected for reproducibility.
  2. [III.B] There is a typo in 'The gags are introduced' (should be 'gaps'), and other typos such as 'sereval', 'improvemet', and 'intrisic' appear in the text; a careful proofread is needed.
  3. [IV vs. Table III] The conclusion states that 'gaps lasted around 15 minutes long', while Table III gives a mean gap duration of 600 s (10 minutes) and Section III.B says '10 minutes on average'; this inconsistency should be resolved.
  4. [Table III] The text mentions a 'total loss cap of 1%' for the synthetic gap pattern, but Table III does not show how this cap is implemented; please specify the cap rule and how overlapping gaps were prevented.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the Kalman imputation is validated against an external no-gap benchmark and an independent windowing baseline, with no target parameter used as an input to the imputation.

full rationale

The claimed result—that Kalman-smoothed gap imputation mitigates windowing biases for short randomized gaps—is not derived from the quantity it predicts. The imputation is built from a Taylor-expansion linear-Gaussian state space (Eqs. 7, 11, 12) with Q and R set by EM on the observed segments; the posterior comparison is then made against an external optimal no-gap dataset and an independent windowing baseline, neither of which enters the imputation. The paper's own failure case (single 3-day gap near tc, Fig. 5) shows the method does not force agreement, so the successful demonstration is not tautological. The only overlapping-author citation (Ref. [28]) supplies the Taiji noise-generation routine, which is a data-production tool rather than evidence for the central claim. The white-noise/colored-noise mismatch (Eqs. 2 vs 7) is a real correctness risk for the likelihood, but it is a modeling concern, not an equivalence-by-construction.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the adequacy of a linear-Gaussian local model for the chirp signal, on standard stationary-Gaussian noise assumptions, and on several method hyperparameters (EM iterations, initial state, observation matrix) that are not fully specified. No new physical entities are introduced.

free parameters (6)
  • EM iteration count = 5
    Chosen empirically as a safe default; affects convergence and imputation quality (Sec II B 2).
  • Kalman observation model C and offsets d
    Required to map the 3D state to A/E/T channels but not specified in the paper, so the exact filter is underdetermined (Sec II B 2, Table I).
  • Initial state mean mu0 and covariance Sigma0
    Initial Kalman conditions are listed in Table I but no values are given; pykalman defaults may be used (Sec II B 2).
  • EM-fitted transition covariance Q and observation covariance R = not reported
    Fitted by EM on the observed gapped data; values are not reported, making exact replication impossible (Sec II B 2).
  • Synthetic gap pattern = mean 600s, std 60s; interval mean 8640s, std 1728s; 1% loss cap
    Defines the single test scenario; the central performance claim is scoped to this pattern (Sec III B, Table III).
  • Windowing transition time ttr = 2 days per side
    Retained from Ref [17]; the baseline comparison depends on this external choice (Sec II B 1).
assumptions (4)
  • domain assumption The gapped data are locally well described by a linear-Gaussian state-space model with a second-order Taylor transition (Eqs. 7, 11, 12).
    Central to Kalman imputation; authors state it can fail for long gaps or rapid frequency changes (Sec II B 2, Sec IV).
  • domain assumption Noise is stationary and Gaussian with a known PSD for the likelihood (Eq. 14) and for data generation.
    Standard in GW parameter estimation and inherited from Ref [17]; the likelihood in Eq. 14 assumes this.
  • domain assumption The Taiji baseline PSD and TDI A/E/T channel construction correctly represent the detector noise and signal response.
    Data generation relies on Taiji Data Challenge noise functions from Ref [28] and TDI definitions in Eqs. 3-5.
  • domain assumption The injected IMRPhenomD waveform via bbhx is an adequate proxy for a real MBHB signal for this gap-treatment study.
    The signal model affects how the Kalman state evolves; waveform systematics are not assessed (Sec II A).

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Kalman-smoother based data imputation strategy to data gaps in spaceborne gravitational wave detectors." pith.science (2026). https://pith.science/paper/SIWTMAJ4

@misc{pith2026250702458,
  author       = {Pith},
  title        = {Pith review of: A Kalman-smoother based data imputation strategy to data gaps in spaceborne gravitational wave detectors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SIWTMAJ4}},
  note         = {Machine review of arXiv:2507.02458}
}
read the original abstract

Massive black hole binaries (MBHBs) and other sources within the frequency band of spaceborne gravitational wave observatories like the Laser Interferometer Space Antenna (LISA), Taiji and Tianqin pose unique challenges, as gaps and glitches during the years-long observation lead to both loss of information and spectral leakage. We propose a novel data imputation strategy based on Kalman filter and smoother to mitigate gap-induced biases in parameter estimation. Applied to a scenario where traditional windowing and smoothing technique introduce significant biases, our method mitigates the biases and demonstrates lower computational cost compared to existing data augmentation techniques such as noise inpainting. This framework presents a new gap treatment approach that balances robustness and efficiency for space-based gravitational wave data analysis.

Figures

Figures reproduced from arXiv: 2507.02458 by the authors.

Figure 1
Figure 1. FIG. 1. Injected GW signal with parameters in Table II into [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. PSD estimation for noise, pure signal and noised [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. As can be seen from the results on the left, the above mentioned gap in Table III introduced a significant bias in [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Results of 9-dimension parameter estimation for the Optimal, Gapped and repaired data. Note here [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Comparison of parameter estimation results for single [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

38 extracted references · 25 canonical work pages

  1. [1]

    Windowing In order to counter the spectral leakage introduced by gaps, it is a simple but effective approach to apply a windowing function to the gap, and both sides of the gaps smoothed by a windowing function are defined as below [17], w(t) =    1 2 1 + cos h π t−ts−ttr ttr i for ts − ttr < t < ts, 0 for ts < t < te, 1 2 1 + cos h π t−te−ttr ...

    work page

  2. [2]

    Kalman filter performs well in case where observation points are missing and predictions on these missing points as well as system states are needed

    Kalman Smoother Kalman filter was developed independently by Swer- ling [29] and Kalman [30], and has seen extensive use since its original use in the Apollo Navigation system [31]. Kalman filter performs well in case where observation points are missing and predictions on these missing points as well as system states are needed. In order to accomplish th...

    work page

  3. [3]

    Here the evidence p(d) stays constant for the same set of data throughout our work and is treated as stationary

    Bayesian Analysis On the basis of our analysis lies the Bayesian Theorem, p(θ|d) = p(d|θ)p(θ) p(d) (13) where p(θ|d) is the posterior probability distribution, p(d|θ) is the likelihood, p(θ) is the prior probability dis- tribution and p(d) is the evidence. Here the evidence p(d) stays constant for the same set of data throughout our work and is treated as...

    work page 2021

  4. [4]

    B. P. Abbott et al. (LIGO Scientific Collaboration and Virgo Collaboration), Phys. Rev. Lett. 116, 061102 (2016)

    2016

  5. [5]

    B. P. Abbott et al. (LIGO Scientific Collaboration and Virgo Collaboration), Phys. Rev. Lett. 116, 241103 (2016)

    work page 2016

  6. [6]

    Aasi et al., Classical and Quantum Gravity 32, 074001 (2015)

    J. Aasi et al., Classical and Quantum Gravity 32, 074001 (2015)

    2015

  7. [7]

    Acernese et al

    F. Acernese et al. , Classical and Quantum Gravity 32, 024001 (2014)

    work page 2014

  8. [8]

    Akutsu et al

    T. Akutsu et al. , Nature Astronomy 3, 35–40 (2019)

    work page 2019

Show all 38 references
  1. [9]

    B. P. Abbott et al. (LIGO Scientific Collaboration and Virgo Collaboration), Phys. Rev. X 9, 031040 (2019)

    2019

  2. [10]

    Sesana, F

    A. Sesana, F. Haardt, P. Madau, and M. Volonteri, The Astrophysical Journal 623, 23 (2005)

    2005

  3. [11]

    J. R. Gair, S. Babak, A. Sesana, P. Amaro-Seoane, E. Ba- rausse, C. P. L. Berry, E. Berti, and C. Sopuerta, Journal of Physics: Conference Series 840, 012021 (2017)

    2017

  4. [12]

    Toubiana, S

    A. Toubiana, S. Marsat, S. Babak, J. Baker, and T. Dal Canton, Phys. Rev. D 102, 124037 (2020)

    2020

  5. [13]

    Bartolo, V

    N. Bartolo, V. Domcke, D. G. Figueroa, J. Garcia- Bellido, M. Peloso, M. Pieroni, A. Ricciardone, M. Sakel- lariadou, L. Sorbo, and G. Tasinato, Journal of Cosmol- ogy and Astroparticle Physics 2018 (11), 034

    2018

  6. [14]

    Colpi et al

    M. Colpi et al. , Lisa definition study report (2024), arXiv:2402.07571 [astro-ph.CO]

    2024 arXiv

  7. [15]

    Hu and Y.-L

    W.-R. Hu and Y.-L. Wu, National Science Review 4, 685 (2017), https://academic.oup.com/nsr/article- pdf/4/5/685/31566708/nwx116.pdf

    2017

  8. [16]

    Luo, L.-S

    J. Luo, L.-S. Chen, H.-Z. Duan, Y.-G. Gong, S. Hu, J. Ji, Q. Liu, J. Mei, V. Milyukov, M. Sazhin, C.-G. Shao, V. T. Toth, H.-B. Tu, Y. Wang, Y. Wang, H.-C. Yeh, M.-S. Zhan, Y. Zhang, V. Zharov, and Z.-B. Zhou, Classical and Quantum Gravity 33, 035010 (2016)

    2016

  9. [17]

    Burke, S

    O. Burke, S. Marsat, J. R. Gair, and M. L. Katz, Mind the gap: addressing data gaps and assessing noise mis- modeling in lisa (2025), arXiv:2502.17426 [gr-qc]

    2025 arXiv

  10. [18]

    Castelli, Q

    E. Castelli, Q. Baghi, J. G. Baker, J. Slutsky, J. Bobin, N. Karnesis, A. Petiteau, O. Sauter, P. Wass, and W. J. Weber, Classical and Quantum Gravity42, 065018 (2025)

    2025

  11. [19]

    Carr´ e and E

    J. Carr´ e and E. K. Porter, The effect of data gaps on lisa galactic binary parameter estimation (2010), arXiv:1010.1641 [gr-qc]

    2010 arXiv

  12. [20]

    K. Dey, N. Karnesis, A. Toubiana, E. Barausse, N. Kor- sakova, Q. Baghi, and S. Basak, Physical Review D 104, 10.1103/physrevd.104.044035 (2021)

    2021 doi

  13. [21]

    Baghi, J

    Q. Baghi, J. I. Thorpe, J. Slutsky, J. Baker, T. D. Can- ton, N. Korsakova, and N. Karnesis, Physical Review D 100, 10.1103/physrevd.100.022003 (2019)

    2019 doi

  14. [22]

    Blelly, J

    A. Blelly, J. Bobin, and H. Moutarde, Monthly Notices of the Royal Astronomical Society 509, 5902–5917 (2021)

    2021

  15. [23]

    Wang, H.-Y

    L. Wang, H.-Y. Chen, X. Lyu, E.-K. Li, and Y.-M. Hu, The European Physical Journal C 85, 10.1140/epjc/s10052-025-13810-0 (2025)

    2025 doi

  16. [24]

    R. Mao, J. E. Lee, and M. C. Edwards, Physical Review D 111, 10.1103/physrevd.111.024067 (2025)

    2025 doi

  17. [27]

    S. Husa, S. Khan, M. Hannam, M. P¨ urrer, F. Ohme, X. J. Forteza, and A. Boh´ e, Phys. Rev. D93, 044006 (2016)

    2016

  18. [28]

    Quang Nam, J

    D. Quang Nam, J. Martino, Y. Lemi` ere, A. Petiteau, J.- B. Bayle, O. Hartwig, and M. Staab, Physical Review D 108, 10.1103/physrevd.108.082004 (2023)

    2023 doi

  19. [30]

    T. A. Prince, M. Tinto, S. L. Larson, and J. W. Arm- strong, Phys. Rev. D 66, 122002 (2002)

    2002

  20. [31]

    M. Du, P. Wang, Z. Luo, W.-B. Han, X. Zhang, X. Chen, Z. Cao, X. Fan, H. Wang, X. Peng, L.-E. Qiang, K. An, Y. Fan, J. Zhang, L.-G. Zhu, P. Shen, Q. Yun, X.-B. Zou, Y. Jiang, T. Zhao, Y. Yuan, X. Wei, Y. Xu, B. Liang, P. Xu, and Y. Wu, Towards realistic detection pipelines of ...

    2025

  21. [32]

    Swerling, Proposed Stagewise Differential Correction Procedure for Satellite Tracking and Prediction, P (Rand Corporation) (Rand Corporation, 1958)

    P. Swerling, Proposed Stagewise Differential Correction Procedure for Satellite Tracking and Prediction, P (Rand Corporation) (Rand Corporation, 1958)

    1958

  22. [33]

    R. E. Kalman, Journal of Ba- sic Engineering 82, 35 (1960), https://asmedigitalcollection.asme.org/fluidsengineering/article- pdf/82/1/35/5518977/35 1.pdf

    1960

  23. [34]

    L. A. McGee and S. F. Schmidt, Discovery of the Kalman filter as a practical tool for aerospace and industry , Tech. Rep. (NASA, 1985)

    1985

  24. [35]

    B. Yu, K. Shenoy, and M. Sahani, Derivation of Kalman Filtering and Smoothing Equations, Tech. Rep. (Carnegie Mellon University, 2004)

    2004

  25. [36]

    D. Duckworth and contributors, pykalman: the dead- simple kalman filter, kalman smoother, and em library for python, https://github.com/pykalman/pykalman (2012), version 0.10.1, accessed July 1, 2025

    2012

  26. [37]

    Foreman-Mackey, D

    D. Foreman-Mackey, D. W. Hogg, D. Lang, and J. Goodman, Publ. Astron. Soc. Pac. 125, 306 (2013), arXiv:1202.3665 [astro-ph.IM]

    2013 arXiv

  27. [38]

    Karnesis, M

    N. Karnesis, M. L. Katz, N. Korsakova, J. R. Gair, and N. Stergioulas, Monthly Notices of the Royal Astronom- ical Society 526, 4814–4830 (2023)

    2023

  28. [39]

    M. Katz, N. Karnesis, and N. Korsakova, Eryn: First full release (version 1.0.0), https://doi.org/10.5281/ zenodo.7705496 (2023)

    2023

  29. [40]

    Armano et al

    M. Armano et al. , Phys. Rev. Lett. 120, 061101 (2018)

    2018

  30. [41]

    Ye, H.-M

    C.-Q. Ye, H.-M. Fan, A. Torres-Orjuela, J.-d. Zhang, and Y.-M. Hu, Physical Review D 109, 10.1103/phys- revd.109.124034 (2024)

    2024 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.