ExoPlanetLab: Bayesian orbit fitting of Gaia-DR4 epoch astrometry, radial velocities and relative astrometry

arXiv:2611.13770 · astro-ph.IM, astro-ph.EP, astro-ph.GA, astro-ph.SR · Submitted 2026-11-10 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Astrophysics. The day in astrophysics: observations, models and what the data forces.

Vera: I'm Vera.

Jocelyn: And I'm Jocelyn, here with Subrahmanyan.

Vera: We're looking at the paper: "ExoPlanetLab: Bayesian orbit fitting of Gaia-DR4 epoch astrometry, radial velocities and relative astrometry".

Jocelyn: Here's the gist of the paper. ExoPlanetLab presents an open-source package for Bayesian orbit inference from Gaia epoch astrometry, radial velocities (RVs) and relative astrometry which is validated on simulated and real data, making it a powerful tool for characterizing giant planets and stellar companions in Gaia DR4 data.

Vera: The code allows for the separation of sampler-dependent failures from the properties of the data by using one likelihood with several samplers.

Jocelyn: Methodological Approach The model orbit is compared with the epoch residuals after removing the part absorbed by the single-star fit.

Vera: A Student-t likelihood describes heavy-tailed data.

Title and authors: Vera: So, let's talk about the title and who wrote this. "ExoPlanetLab: Bayesian orbit fitting of Gaia-DR4 epoch astrometry, radial velocities and relative astrometry." It tells you exactly what you’re getting here—an open-source tool for finding orbits using three different types of data from Gaia DR4 one <ref:2611.13770#pg2>.

Jocelyn: The authors are D. Hobbs and they've built this package at Lund Observatory. It’s clear they have a solid background in the observational side, given their work on survey photometry and spectra Vera.

Subrahmanyan: I’m interested in how they approach the problem of fitting orbits when you have these multiple data streams; it suggests a comprehensive model that goes beyond just astrometry alone. It hints at needing to combine different types of information, like position and velocity, to get the most accurate results one <ref:2611.13770#pg2>.

Vera: And from what I can gather from their description, the implication is that we can now characterize giant planets and stellar companions in Gaia DR4 data with a tool that is validated on both simulated and real data one <ref:2611.13770#pg2>. That validation step is key for trust.

Jocelyn: It’s interesting because it moves us closer to being able to reliably find these objects, not just spot them, by giving us a standardized way to do the fitting one <ref:2611.13770#pg2>.

Subrahmanyan: And this package suggests that the combination of epoch astrometry and radial velocities is necessary for robust characterization, which aligns with how we often try to constrain orbital parameters in astrophysics one <ref:2611.13770#pg2>. It sets a benchmark for what's feasible when using Gaia data.

The paper's summary: Vera: Now, let’s look at the actual summary they give of ExoPlanetLab. They explain that the model orbit they compare against isn't just the raw data; it’s actually "the epoch residuals after removing the part absorbed by the single-star fit" one <ref:2611.13770#pg2>.

Jocelyn: That distinction is crucial. It means they are trying to fit what's left over after accounting for the most obvious component, which is usually a single star model one <ref:2611.13770#pg2>.

Subrahmanyan: And they describe their modeling of this residual using a Student t likelihood, which they say describes heavy-tailed data one <ref:2611.13770#pg2>. This immediately points to the fact that they expect some of those observations to be really noisy or have big outliers.

Vera: That’s where I see the practical implication; if we know the data is heavy-tailed, we need a fitting method that handles it well, and they are proposing this likelihood as a way to do that one <ref:2611.13770#pg2>.

Jocelyn: And they also detail their sampler choices; they have five samplers available, including tempered ensemble MCMC and two reversible-jump samplers one <ref:2611.13770#pg2>. That shows they’re trying to be thorough in how the inference is performed.

Subrahmanyan: It seems like a solid approach because it addresses both the noise characteristics of the data and the uncertainty in the fitting process itself, which is a big problem when you're dealing with complex models one <ref:2611.13770#pg2>.

The paper's improvements: Vera: The paper really highlights several specific improvements they made to existing tools. One big thing is how they define what needs to be removed from the data before fitting—it’s using the part of the orbit that a single star can mimic, which they call "the part of the orbit that a single star can mimic has been removed from the data and must be removed from the model in the same way" one <ref:2611.13770#pg2>.

Jocelyn: That sounds like a sophisticated way to handle those spurious signals we see in astrometry; by accounting for what one star can fake, you isolate what's left over one <ref:2611.13770#pg2>.

Subrahmanyan: They also mention that they test whether an orbit is actually warranted using acceleration and scan-angle models one <ref:2611.13770#pg2>. This is a way to filter out orbits that aren't physically plausible before committing to a full fit.

Vera: And then there’s this pipeline for estimating the single-star parameters using weighted least squares, followed by something called the Frisch–Waugh–Lovell theorem to make sure the left side of the equation equals our chi squared and the right side equals a joint fit at those best single-star parameters one <ref:2611.13770#pg2>.

Jocelyn: That sounds like they’re trying to be very careful about how they isolate each piece of information; it’s a lot to manage when you have so many parameters involved one <ref:2611.13770#pg2>.

Subrahmanyan: They also detail how the pipeline estimates the single-star parameters by looking at both the along-scan measurements and those radial velocities, which is a systematic way to build up that initial model one <ref:2611.13770#pg2>.

Conclusion: Vera: So, to wrap this up, ExoPlanetLab gives us a validated open-source package for Bayesian orbit fitting using Gaia DR4 data with RVs and relative astrometry. The results show the samplers agree within zero point one sigma on simulated two-planet systems one <ref:2611.13770#pg2>.

Jocelyn: And they’ve shown that on thirty companion-free data sets, none of their methods gave a false detection, which is really reassuring for the reliability of this approach one <ref:2611.13770#pg2>.

Subrahmanyan: The implication for the broader field is that this method provides a consistent way to characterize giant planets and stellar companions in Gaia DR4 data without running into immediate false positives from the noise structure one <ref:2611.13770#pg2>.

Vera: And on real data, they recovered the orbit of Gaia BH3 within one point four sigma of published solutions, and they found that the Student t model successfully brought HD one hundred fourteen thousand seven hundred sixty-two's astrometric period and eccentricity onto its radial velocity orbit one.

Jocelyn: It’s a practical tool for anyone working with this kind of data because it handles the noise structure in a way that seems physically motivated one <ref:2611.13770#pg2>.

Subrahmanyan: This work suggests that when dealing with complex data, using multiple samplers and robust starting points based on periodograms is a viable path forward for constraining orbital parameters one <ref:2611.13770#pg2>.

D. Hobbs

Lund University · European Space Agency

astro-ph.IM, astro-ph.EP, astro-ph.GA, astro-ph.SR

Submitted: 2026-11-10

Updated: 2026-11-10

Comments: 17 pages, 9 figures, 5 tables. Code: https://github.com/hobbsSe/ExoPlanetLab, archived at https://doi.org/10.5281/zenodo.24101622

Code: https://github.com/hobbsSe/ExoPlanetLab

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 83/100

The gist: ExoPlanetLab presents an open-source package for Bayesian orbit inference from Gaia epoch astrometry, radial velocities (RVs) and relative astrometry which is validated on simulated and real data…

Key concepts

Student-t Likelihood
This statistical model is used because the data often has heavy tails, meaning extreme outliers are common. It is more robust than standard models when dealing with noisy observations or CCD artifacts, allowing the fitting process to remain stable and accurate even with unusual data points.
(I-H)s(θ)
This term represents the part of the orbital model that a single star can mimic. The fitting process focuses on removing this component from the observed residuals before comparing it to the true orbit, ensuring that only genuine companion signals are identified.
Robust Periodogram Screen
Instead of simple least squares, this method uses statistics designed for small signals in noisy data. This technique is superior for detecting faint orbital signals in epoch data because it handles the heavy-tailed noise more effectively than traditional methods.

Summary

ExoPlanetLab presents an open-source package for Bayesian orbit inference from Gaia epoch astrometry, radial velocities (RVs) and relative astrometry which is validated on simulated and real data, making it a powerful tool for characterizing giant planets and stellar companions in Gaia DR4 data. p. 0

Where the paper says it

Gaia Data Release 4 (DR4) provides epoch astrometry from which thousands of giant planets and many substellar and stellar companions can be characterised. We present ExoPlanetLab, an open-source package for Bayesian orbit inference from Gaia epoch astrometry, radial velocities (RVs) and relative astrometry, validated on simulated and real data.

The code allows for the separation of sampler-dependent failures from the properties of the data by using one likelihood with several samplers. p. 0
Where the paper says it

Several codes now address parts of this problem (Sect. ExoPlanetLab was written to study it with one likelihood and several samplers, so that results that depend on the sampler can be separated from properties of the data.

Methodological Approach

The model orbit is compared with the epoch residuals after removing the part absorbed by the single-star fit. A Student-t likelihood describes heavy-tailed data. p. 0

Where the paper says it

The model orbit is compared with the epoch residuals after removing the part absorbed by the single-star fit. A Student-t likelihood describes heavy-tailed data.

Robust Keplerian periodograms with bootstrap false-alarm probabilities give starting values for a joint fit. p. 0
Where the paper says it

A Student-t likelihood describes heavy-tailed data. A robust Keplerian periodogram with bootstrap false-alarm probabilities gives starting values for a joint fit.

Five samplers are available, including tempered ensemble MCMC, two nested samplers and two reversible-jump samplers. p. 0
Where the paper says it

A robust Keplerian periodogram with bootstrap false-alarm probabilities gives starting values for a joint fit. Five samplers are available: tempered ensemble MCMC, two nested samplers and two reversible-jump samplers.

Evidence for a fixed number of companions selects that number. p. 5
Where the paper says it

Model comparison The reversible-jump samplers accept almost no transdimensional moves on these problems, so the number of companions is chosen from the evidences Z of fixed-Npl fits: stepping up from Npl = 0, a companion is added when the Bayes factor B(Npl vs Npl −1) = ZNpl/ZNpl−1 satisfies ln B > 5 (odds of about 150:1). For Npl = 0 and for the acceleration and scan-angle models only the noise parameters remain, and the evidence is integrated directly.

Acceleration and scan-angle models test whether an orbit is warranted. p. 0
Where the paper says it

Evidences for a fixed number of companions select that number. Acceleration and scan-angle models test whether an orbit is warranted.

Model Formulation

Each orbiting companion k is described by seven parameters (Pk, ek, T0,k, Ak, Bk, Fk, Gk). The along-scan signal at scan angle ψi is si = ∆α∗i sin ψi + ∆δi cos ψi. p. 1

Where the paper says it

Astrometric orbit Each orbiting companion k is described by seven parameters (Pk, ek, T0,k, Ak, Bk, Fk, Gk). With M = 2π(t − T0)/P = E −e sin E, X = cos E −e and Y = √ 1 − e 2 sin E, the photocentre displacement p(t) = (∆α ∗ , ∆δ) is ∆δ = AX + F Y, ∆α ∗ = BX + GY, (1) in the standard Thiele–Innes convention (orbital-plane x axis to the north), and the along-scan signal at scan angle ψi is si = ∆α ∗ i sin ψi + ∆δi cos ψi , (2) and with several companions the signals of the individual companions add.

The DR4 epoch values are not the measured along-scan positions themselves but the residuals, observed minus computed (O−C), of the single-star astrometric solution. p. 1
Where the paper says it

1 can solve for them in closed form. Projection through the single-star fit The DR4 epoch values are not the measured along-scan positions themselves but the residuals, observed minus computed (O−C), of the single-star astrometric solution.

The model to compare with the published residuals is not the orbit s(θ) but (I−H) s, i. p. 1
Where the paper says it

6), and therefore r = (I − H) s(θ) + (I − H) n. (6) The model to compare with the published residuals is not the orbit s(θ) but (I−H) s(θ), i.

e., the part of the orbit that a single star can mimic has been removed from the data and must be removed from the model in the same way. p. 1
Where the paper says it

(6) The model to compare with the published residuals is not the orbit s(θ) but (I−H) s(θ), i. , the part of the orbit that a single star can mimic has been removed from the data and must be removed from the model in the same way.

Inference Pipeline

The pipeline estimates the single-star parameters by weighted least squares. The first sum runs over the Nobs along-scan measurements: ri is the published residual and mi = [(I − H) s(θ)]i the projected orbit model (Sect. p. 3

Where the paper says it

Likelihood and priors For Gaussian noise the astrometric and RV terms are independent: lnLGauss = − 1 2 X i h (ri − mi) 2 σ 2 i + s 2 ast + ln(σ 2 i + s 2 ast) i − 1 2 X j h (vj − Vr,j ) 2 σ 2 j + s 2 ι(j) + ln(σ 2 j + s 2 ι(j) ) i + const. (9) The first sum runs over the Nobs along-scan measurements: ri is the published residual and mi = [(I − H) s(θ)]i the projected orbit model (Sect.

3.2), σi the formal uncertainty of measurement i, and sast the astrometric jitter (Sect. p. 3
Where the paper says it

(9) The first sum runs over the Nobs along-scan measurements: ri is the published residual and mi = [(I − H) s(θ)]i the projected orbit model (Sect. 2), σi the formal uncertainty of measurement i, and sast the astrometric jitter (Sect.

3.6). The second sum runs over the NRV radial velocities: vj is the measured velocity with uncertainty σj, Vr,j the model of Sect. p. 3
Where the paper says it

6), added in quadrature. The second sum runs over the NRV radial velocities: vj is the measured velocity with uncertainty σj , Vr,j the model of Sect.

3.3, and ι(j) the instrument that made measurement j, with jitter sι(j). p. 3
Where the paper says it

The second sum runs over the NRV radial velocities: vj is the measured velocity with uncertainty σj , Vr,j the model of Sect. 3, and ι(j) the instrument that made measurement j, with jitter sι(j) .

Validation and Results

The samplers agree within 0.1σ on a simulated two-planet system. p. 0

Where the paper says it

The samplers agree within 0. 1σ on a simulated two-planet system.

In 30 companion-free data sets, none gives a false detection. p. 0
Where the paper says it

The posteriors show no significant bias, but the period coverage is below nominal: 90% of the true periods lie within their 95% credible intervals. None of 30 companion-free data sets gives a false detection.

On DR4 pre-release data, the orbit of Gaia BH3 is recovered within 1. p. 0
Where the paper says it

None of 30 companion-free data sets gives a false detection. On DR4 pre-release data the orbit of Gaia BH3 is recovered within 1.

4σ of published solutions. p. 0
Where the paper says it

On DR4 pre-release data the orbit of Gaia BH3 is recovered within 1. 4σ of published solutions.

The Student-t model brings the astrometric period and eccentricity of HD 114762 onto its RV orbit. p. 0
Where the paper says it

4σ of published solutions. The Student-t model brings the astrometric period and eccentricity of HD 114762 onto its RV orbit.

Conclusions

The code is validated for single sources with DR4-like epoch astrometry, with or without RVs. The main limits are an uncorrelated noise model and the computing cost, which grows linearly with the number of measurements and steeply with the number of companions. p. 0

Where the paper says it

Relative astrometry is tested on simulated data only. The main limits are an uncorrelated noise model and the computing cost, which grows linearly with the number of measurements and steeply with the number of companions.

The relative astrometry is tested on simulated data only. p. 0
Where the paper says it

The code is validated for single sources with DR4-like epoch astrometry, with or without RVs. Relative astrometry is tested on simulated data only.

The code is available at https://github. p. 0
Where the paper says it

Conclusions. The code is validated for single sources with DR4-like epoch astrometry, with or without RVs.

com/hobbsSe/ExoPlanetLab and archived at Zenodo, https://doi. p. 0
Where the paper says it

Astronomy & Astrophysics manuscript no. ExoPlanetLabPaper ©ESO 2026 November 11, 2026 ExoPlanetLab: Bayesian orbit fitting of Gaia-DR4 epoch astrometry, radial velocities and relative astrometry D.

org/10.5281/zenodo.24101622 (version 1.0.0).

Key References

Andrews, D. F., & Mallows, C. p. 4

Where the paper says it

Each pair, (aα, aδ), (jα, jδ) or (un, vn), has a Gaussian prior with zero mean and standard deviation σp on each component, N (0, σ2 p ); σp sets the typical size of the terms and is itself not fixed but log-uniform over five decades, 10−3 to 100 in the unit of the terms (mas yr−2 , mas yr−3 or mas). The mixture of Gaussians over σp (Andrews & Mallows 1974) is isotropic in direction and close to log-uniform in amplitude, like the a0 prior of an orbit.

L., Journal of the Royal Statistical Society, Series B (Methodological), 36, 99

Baycroft, T. A., Faria, J. p. 1

Where the paper says it

Existing tools Table 1 summarises the main publicly available codes that fit orbits to absolute astrometry, based on their papers and documentation. kima (Faria et al.

P., & Delisle, J. p. 1
Where the paper says it

2026); it fits the five single-star parameters (optionally seven or nine, with acceleration terms), an astrometric jitter, an optional Student-t likelihood, and scan-angle harmonic terms to discriminate spurious signals, and has been applied to Gaia BH3. kepmodel (Delisle & Ségransan 2022) models astrometric and RV time series with Keplerians, linear predictors and Gaussian-process noise, and provides a linear periodogram with analytical false-alarm probabilities.

-B., A&A in press

Bennett, C. H., Journal of Computational Physics, 22, 245

Blunt, S., Wang, J. J., Hirsch, L., et al., JOSS in press

Brandt, T. D., Dupuy, T. J., Li, Y., et al., AJ in press

Delisle, J.-B., & Ségransan, D. p. 1

Where the paper says it

2026); it fits the five single-star parameters (optionally seven or nine, with acceleration terms), an astrometric jitter, an optional Student-t likelihood, and scan-angle harmonic terms to discriminate spurious signals, and has been applied to Gaia BH3. kepmodel (Delisle & Ségransan 2022) models astrometric and RV time series with Keplerians, linear predictors and Gaussian-process noise, and provides a linear periodogram with analytical false-alarm probabilities.

, A&A in press

El-Badry, K., Lam, C., Holl, B., et al., The Open Journal of Astrophysics in press

Faria, J. P., Santos, N. C., Figueira, P., & Brewer, B. J., JOSS in press

Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, J. p. 5

Where the paper says it

Every backend therefore depends on a joint fit of all planets. This is an affine-invariant ensemble MCMC (Goodman & Weare 2010; Foreman-Mackey et al.

, PASP in press

Frisch, R., & Waugh, F. p. 2

Where the paper says it

Writing ∥x∥ 2 W = x ⊤Wx = P i x 2 i /σ2 i for the χ 2 of a residual vector, r − (I − H)s 2 W = min β d − s − Dβ 2 W : (7) the left side is our χ 2 , the right side that of a joint fit at the best single-star parameters, and both equal ∥(I − H)(d − s)∥ 2 W, because the minimising β is the single-star solution for the data d−s. This is the Frisch–Waugh–Lovell theorem of linear regression (Frisch & Waugh 1933; Lovell 1963).

V., Econometrica in press

Gaia Collaboration, Arenou, F., Babusiaux, C., et al., A&A in press

Gaia Collaboration, Panuzzo, P., Mazeh, T., Arenou, F., et al., Astronomy & Astrophysics in press

Gaia Collaboration, Vallenari, A., Brown, A. p. 0

Where the paper says it

With those criteria the DR4 estimate of Lammers & Winn (2026) is about 16 000. Tens of thousands of brown dwarfs come in addition (Holl et al.

G. A., Prusti, T., et al., A&A in press

Golub, G. H., & Pereyra, V. p. 2

Where the paper says it

This is the Frisch–Waugh–Lovell theorem of linear regression (Frisch & Waugh 1933; Lovell 1963). Removing the linear parameters of a nonlinear fit in this way is also known as variable projection (Golub & Pereyra 1973).

, SIAM Journal on Numerical Analysis in press

Goodman, J., & Weare, J. p. 5

Where the paper says it

Every backend therefore depends on a joint fit of all planets. This is an affine-invariant ensemble MCMC (Goodman & Weare 2010; Foreman-Mackey et al.

, Communications in Applied Mathematics and Computational Science in press

Green, R. M., Spherical Astronomy (Cambridge: Cambridge University Press)

Halbwachs, J.-L., Pourbaix, D., Arenou, F., et al., A&A in press

Hoaglin, D. C., & Welsch, R. p. 1

Where the paper says it

To see what is left, substitute the solution for β into r: r = (I − H) d, H = D N−1D⊤W. (5) Applied to any vector of along-scan values, H returns the best-fitting single-star curve (it is the hat matrix of the fit; Hoaglin & Welsch 1978), so (I − H) returns the O−C of a single-star fit to that vector.

p. 1
Where the paper says it

To see what is left, substitute the solution for β into r: r = (I − H) d, H = D N−1D⊤W. (5) Applied to any vector of along-scan values, H returns the best-fitting single-star curve (it is the hat matrix of the fit; Hoaglin & Welsch 1978), so (I − H) returns the O−C of a single-star fit to that vector.

E., The American Statistician in press

Holl, B., Perryman, M. p. 0

Where the paper says it

Lammers & Winn (2026) predict 7500 ± 2100 giant planets in DR4 and 120 000 ± 22 000 in DR5. Perryman et al.

, Lindegren, L. p. 9
Where the paper says it

No parallax zero-point is applied in the fit. Subtracting the Gaia DR3 zero-point of Lindegren et al.

, Ségransan, D. p. 1
Where the paper says it

2026); it fits the five single-star parameters (optionally seven or nine, with acceleration terms), an astrometric jitter, an optional Student-t likelihood, and scan-angle harmonic terms to discriminate spurious signals, and has been applied to Gaia BH3. kepmodel (Delisle & Ségransan 2022) models astrometric and RV time series with Keplerians, linear predictors and Gaussian-process noise, and provides a linear periodogram with analytical false-alarm probabilities.

, & Raimbault, M., A&A in press

Holl, B., Sozzetti, A., Sahlmann, J., et al., A&A in press

Holland, P. W., & Welsch, R. E., Communications in Statistics - Theory and Methods in press

Kane, S. R., Henry, G. W., Dragomir, D., et al., The Astrophysical Journal Letters in press

Karnesis, N., Katz, M. L., Korsakova, N., Gair, J. R., & Stergioulas, N., MNRAS in press

Kiefer, F., Astronomy & Astrophysics in press

Lammers, C., & Winn, J. p. 0

Where the paper says it

2024), and of the order of 104 giant planets. Lammers & Winn (2026) predict 7500 ± 2100 giant planets in DR4 and 120 000 ± 22 000 in DR5.

N., AJ in press

Lange, K. L., Little, R. p. 6

Where the paper says it

1 give the calibration over the 173 converged runs with a detection. All parameters are unbiased, and all but the period are calibrated (T0 slightly conservatively: its intervals are a little too wide).

J. A., & Taylor, J. M. G., Journal of the American Statistical Association in press

Leclerc, A., Babusiaux, C., Arenou, F., et al., A&A in press

Lindegren, L., Bastian, U., Biermann, M., et al., A&A in press

Lovell, M. C., Journal of the American Statistical Association in press

MacKay, D. J. C., Neural Computation in press

Perryman, M., Hartman, J., Bakos, G., Á., & Lindegren, L. p. 9

Where the paper says it

No parallax zero-point is applied in the fit. Subtracting the Gaia DR3 zero-point of Lindegren et al.

, ApJ in press

Pourbaix, D., & Jorissen, A., A&AS in press

Speagle, J. S., MNRAS in press

Stefánsson, G., Mahadevan, S., Winn, J. N., et al., The Astronomical Journal in press

Thompson, W., Lawrence, J., Blakely, D., et al., AJ in press

Tierney, L., & Kadane, J. p. 4

Where the paper says it

Here ri = di −si(θ)−(Dβˆ)i are the residuals of this robust fit. Marginalising over β with a Laplace approximation (Tierney & Kadane 1986) about βˆ, in which the Fisher information of the t-distribution (Lange et al.

B., Journal of the American Statistical Association in press

Williams, M. J., Veitch, J., & Messenger, C., Phys. Rev. D in press

Winn, J. N.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Robust orbit fitting requires comparing a model orbit with the epoch residuals after removing the part absorbed by the single-star fit rather than just raw data, ensuring that the model to compare with the published residuals is not the orbit s(θ) but (I−H) s(θ).

  2. AI systems can be trained to use a Student-t likelihood which describes heavy-tailed data, leading to more robust fitting in the presence of CCD outliers, especially for HD 114762 where the excess noise is heavy-tailed at the CCD level.

  3. The system should implement a robust periodogram screen that uses the locally most powerful statistic for a small linear signal in t noise instead of standard least squares, which is superior when dealing with heavy-tailed epoch data.

  4. AI systems can incorporate variable projection by estimating the single-star parameters by weighted least squares and then using the Frisch–Waugh–Lovell theorem to ensure that the left side is our χ2, the right side that of a joint fit at the best single-star parameters, and both equal∥(I−H)(d − s)∥2W.

  5. The system can utilize a suite of five samplers, including tempered ensemble MCMC and reversible-jump samplers, with evidence runs using a temperature ladder adapted to equilibrated chains to ensure that the evidence of a strong signal needs a temperature ladder adapted to equilibrated chains.

  6. AI systems can automatically select the number of companions based on comparing "the Bayes factor B(Npl vs Npl −1) = ZNpl/ZNpl−1 satisfies ln B > 5 (odds of about 150:1)," preventing false detections in data sets without a companion.

  7. The system should use relative astrometry by incorporating one parameter per imaged companion, s into the likelihood, which for paired injections narrow the credible intervals to 0.27 (P), 0.38 (e), 0.67 (a0) and 0.64 (i) of their widths.

  8. The system can be programmed to check if an orbit is warranted using acceleration and scan-angle models, as an orbit is warranted up to periods of 2–3 times the time span of the data or when acceleration and jerk are fitted with it.

Abstract

Gaia Data Release 4 (DR4) provides epoch astrometry from which thousands of giant planets and many substellar and stellar companions can be characterised. We present ExoPlanetLab, an open-source package for Bayesian orbit inference from Gaia epoch astrometry, radial velocities (RVs) and relative astrometry, validated on simulated and real data. The model orbit is compared with the epoch residuals after removing the part absorbed by the single-star fit. A Student-t likelihood describes heavy-tailed data. A robust Keplerian periodogram with bootstrap false-alarm probabilities gives starting values for a joint fit. Five samplers are available: tempered ensemble MCMC, two nested samplers and two reversible-jump samplers. Evidences for a fixed number of companions select that number. Acceleration and scan-angle models test whether an orbit is warranted. The samplers agree within 0.1 sigma on a simulated two-planet system. In 180 injections all fits converge, and all signals with signal-to-noise ratio S >= 10 are detected. The posteriors show no significant bias, but the period coverage is below nominal: 90% of the true periods lie within their 95% credible intervals. None of 30 companion-free data sets gives a false detection. On DR4 pre-release data the orbit of Gaia BH3 is recovered within 1.4 sigma of published solutions. The Student-t model brings the astrometric period and eccentricity of HD 114762 onto its RV orbit. Eight simulated relative positions narrow the period's credible interval to 0.27 of its width, and Octofitter agrees within 0.2 sigma on the same data. The code is validated for single sources with DR4-like epoch astrometry, with or without RVs. Relative astrometry is tested on simulated data only. The main limits are an uncorrelated noise model and the computing cost, which grows linearly with the number of measurements and steeply with the number of companions.

Sources

Related papers