Pith. sign in

REVIEW 3 major objections 5 minor 112 references

Deep Potential: Recovering the gravitational potential and local pattern speed in the solar neighborhood with GDR3 using normalizing flows

T0 review · 3 major / 5 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 data-driven "Deep Potential" analysis of 5.6 million Gaia stars claims to recover the Milky Way's gravitational potential in a 1 kpc sphere around the Sun, finding a local pattern speed of 28.20 ± 0.10 km/s/kpc and a total matter…

desk verdict Solid first application of Deep Potential to real Gaia data, with honest caveats—but the stationarity violation means the headline uncertainties are lower bounds. read the letter →

arxiv 2507.03742 v2 pith:ILX6PJHX submitted 2025-07-04 astro-ph.GA

Taavet Kalda , Gregory M. Green This is my paper
classification astro-ph.GA
keywords GalacticdynamicsGravitationalpotentialNormalizingflowsNeuralnetworksMilkyWaypatternspeedLocaldarkmatterdensityGaiaDR3Solarneighborhood
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 gravitational potential of the Milky Way is usually inferred under strong assumptions like axisymmetry or a fixed form for the distribution function. This paper claims that a nearly assumption-free recovery is possible from a single snapshot of six-dimensional Gaia phase-space data, provided the tracer population is statistically stationary in some rotating frame. Using 5.6 million upper-main-sequence stars within 1 kpc of the Sun, the authors fit a flexible distribution function with a normalizing flow and a flexible gravitational potential with a neural network, simultaneously solving for the rotation speed that makes the system appear most stationary. They recover a local pattern speed of $\Omega_p = 28.20 \pm 0.10\,\mathrm{km\,s^{-1}\,kpc^{-1}}$, a total matter density of $\rho = 0.086 \pm 0.010\,M_\odot/\mathrm{pc}^3$, and a dark matter density of $\rho_\mathrm{DM} = 0.007 \pm 0.011\,M_\odot/\mathrm{pc}^3$ that is consistent with zero. The significance is that this is a path toward mapping the full three-dimensional matter distribution, including dark matter, from kinematic snapshots with minimal structural priors.

What carries the argument

The central object is the stationarity condition combined with the collisionless Boltzmann equation: $$\left(\frac{\partial f}{\partial t}\right)_\$\Omega$ = \sum_i \left[(u_i - v_i)\frac{\partial f}{\partial x_i} + \left(\frac{\partial \Phi}{\partial x_i} + w_i\right)\frac{\partial f}{\partial v_i}\right] = 0.$$ The distribution function $f(\mathbf x, \mathbf v)$ is modeled nonparametrically with a normalizing flow (a chain of invertible transformations mapping a Gaussian to the observed phase-space density), and the gravitational potential $\Phi(\mathbf x)$ is modeled with a feed-forward neural network whose second derivatives give the density via the Poisson equation $\rho = \nabla^2\Phi/(4\pi G)$. The machine presses the potential and pattern speed against the sampled gradients of the flow, penalizing any residual non-stationarity and any negative density, which is how a six-dimensional snapshot becomes enough to solve for a three-dimensional field.

What would settle it

The strongest direct test would be to run the identical pipeline on a high-resolution N-body simulation of a Milky-Way-like galaxy with a known potential and known non-axisymmetric features, and check whether it recovers the input pattern speed and density within the quoted uncertainties; disagreement at the level of the residuals seen in Figure 8 would indicate the method's systematics dominate over the statistical error bars.

Watch

Extended reading notes

Core claim

The paper's central claim is that the three-dimensional gravitational potential, and hence the matter density, in a 1 kpc volume around the Sun can be recovered directly from a snapshot of stellar positions and velocities, without assuming axisymmetry or a parametric form for either the distribution function or the potential. The method treats the Milky Way as a collisionless system that is statistically stationary in a frame rotating with a single angular speed $\Omega$ about the Galactic Center, and finds the potential and $\Omega$ that best satisfy the resulting collisionless Boltzmann equation. Applied to 5.6 million upper-main-sequence stars from Gaia DR3, it yields a best-fit pattern speed $\Omega_p = (28.20 \pm 0.10)\,\mathrm{km/s/kpc}$, which makes the solar neighborhood appear most stationary and places the corotation radius near the Sun at about $1.02\,R_0$. The same model gives a local total matter density of $\rho = 0.086 \pm 0.010\,M_\odot/\mathrm{pc}^3$, and after subtracting a baryonic model, a dark matter density of $0.007 \pm 0.011\,M_\odot/\mathrm{pc}^3$, which is not a statistically significant detection of dark matter. The recovered acceleration field deviates from the best-fitting axisymmetric model by only about 2% in most of the volume, but shows a systematic azimuthal acceleration opposite to Galactic rotation, indicating non-axisymmetric structure such as spiral arms locally.

Load-bearing premise

The stars in the sample are statistically stationary in a single frame rotating at one angular speed around the Galactic Center, but the paper itself shows the Milky Way is not perfectly stationary in such a frame (the phase spiral's ~20 Myr signature remains).

Editorial extensions

If this is right

  • If the method works as claimed, the local matter density and pattern speed can be recovered independently of parametric models of the Milky Way's disk, bar, spiral arms, and halo shape.
  • The recovered pattern speed of about $28\,\mathrm{km/s/kpc}$ implies corotation near the Sun, which would affect models of how the spiral arms and bar torque the local disk and how stars migrate radially.
  • The inability to detect dark matter locally ($\rho_\mathrm{DM} = 0.007 \pm 0.011\,M_\odot/\mathrm{pc}^3$) sets an upper bound on the local dark matter density that is consistent with but much less constraining than standard estimates.
  • The residual non-stationarity, shown to be imprinted by the Gaia phase spiral with a characteristic timescale of about 20 Myr, provides a new observable diagnostic for disequilibrium features in the solar neighborhood.
  • The same framework can map the potential beyond 1 kpc as more luminous tracers and future data releases are added, potentially revealing the three-dimensional dark matter distribution in the disk.

Reading between the lines

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

  • The stationarity assumption is the load-bearing premise, and the paper itself demonstrates it is violated: Section 4.6 shows residual non-stationarity imprinted by the phase spiral on a ~20 Myr timescale, so the quoted potential should be read as the best-fit single rotating frame, not a unique ground-truth potential.
  • The method could be stress-tested by applying it to a high-resolution simulation of a Milky-Way-like galaxy with a known potential, and checking whether the recovered $\Omega_p$ and $\rho$ reproduce the input values and whether the residuals trace the same phase-space features seen here.
  • A testable prediction is that a different tracer population (e.g., red giants rather than upper-main-sequence stars) should yield a different best-fit pattern speed, because different populations experience different resonances and non-stationarities; comparing them would quantify how far the single-rotating-frame picture is from reality.
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

3 major / 5 minor

Summary. The paper applies the 'Deep Potential' method (Green & Ting 2020; Green et al. 2023; Kalda et al. 2024) to 5.6 million upper-main-sequence stars from Gaia DR3 within 1 kpc of the Sun. The method represents the stellar phase-space distribution with a normalizing flow, models the gravitational potential with a neural network, and fits a single pattern speed Ωp for a rotating frame in which the distribution is assumed stationary. The authors report Ωp = 28.20 ± 0.10 km/s/kpc, a local total matter density ρ = 0.086 ± 0.010 M_sun/pc^3, and a dark matter density ρ_DM = 0.007 ± 0.011 M_sun/pc^3, along with acceleration and density maps. The paper explicitly acknowledges that the stationarity assumption is violated: Section 4.6 reports residual non-stationarity timescales of order 20 Myr and a phase-spiral imprint in the residuals, and Section 2.5 states that the quoted uncertainties are lower bounds that do not include the stationarity condition not holding.

Significance. This is a serious, data-rich application of a flexible non-parametric method. Strengths include the public release of code and trained models (Zenodo/GitHub), a smooth neural-network model of the Gaia selection function, careful treatment of open-cluster contamination, and an ensemble-based uncertainty pipeline that propagates several internal sources of error. If the systematic from non-stationarity can be quantified or controlled, the method could become a valuable tool for mapping the local potential and identifying disequilibrium features. At present, however, the headline numbers are conditional on an assumption that the paper itself demonstrates to be violated, so the central claim is not yet established.

major comments (3)
  1. [Section 4.6, Eq. (5)] The paper's own diagnostic in Section 4.6 shows that the stationarity assumption underlying the method is not satisfied: the residual (∂ ln f/∂t)_Ω has a characteristic timescale of ~20 Myr in the best-fit rotating frame, comparable to vertical dynamical times in the disk, and the residual is imprinted by the phase spiral (Fig. 11). Because the potential Φ(x) and pattern speed Ωp are determined by minimizing the magnitude of (∂f/∂t)_Ω (Eqs. 8-9), a non-stationary DF will bias the recovered potential and pattern speed: the flexible neural network can partially cancel the non-stationarity through spurious force terms. The uncertainty pipeline (Section 2.5, Appendix C) explicitly excludes this 'stationarity condition not holding' contribution, so the quoted Ωp = 28.20 ± 0.10 km/s/kpc and ρ = 0.086 ± 0.010 M_sun/pc^3 are internal consistency errors rather than total errors. To establish the central claim, the authors should quantify the bias using controlled mock data with known non-stationarity (e.g., bar, spiral arms, phase spiral) or add a calibrated systematic floor to all headline values; otherwise the abstract and Table 3 should be rephrased as conditional on exact stationarity.
  2. [Section 4.4, Table 3] The volume-averaged density quoted in Table 3 is ρ(R0) = 0.086 ± 0.010 M_sun/pc^3, but the model density in the midplane shows fluctuations on the order of 50% (Fig. 8), and the phase-spiral residual (Fig. 11) is coherent in the z-v_z plane, which is directly relevant for the vertical density profile. The uncertainty in Table 3 includes only the internal resampling/model-seed variations from Section 2.5; it does not include the coherent non-stationarity signature. The paper therefore provides no evidence that the fluctuations average out unbiasedly, and the density result is not robust until this systematic is quantified.
  3. [Section 4.3] The corotation inference R_c = v_circ/Ωp = 8.4 kpc = 1.02 R0 is drawn from Ωp = 28.20 ± 0.10 km/s/kpc without propagating the stationarity-violation systematic. Given that Ωp is defined as the rotation speed that minimizes non-stationarity in a 1 kpc snapshot and that the residual is dominated by the phase spiral, the identification of this fitted quantity with a physical pattern speed, and the corotation conclusion, are not yet supported. Please either include a systematic error term in Ωp or remove/condition the corotation statement.
minor comments (5)
  1. [Section 4.6 and Figure 9 caption] The text states that the characteristic timescales are 5 Myr and 20 Myr for the rotating and non-rotating cases, respectively, while the Figure 9 caption gives 20 Myr and 5 Myr for the same two cases; please reconcile this discrepancy.
  2. [Section 2.4] The text 'm = 222 (∼ 4 million)' should read 'm = 2^22' (the superscript appears to be missing in the typography).
  3. [Section 4.4] The sentence 'We can compute the matter density ρ = ∇²ρ/(4πG)' contains a typo; it should be ρ = ∇²Φ/(4πG).
  4. [Section 3.3, Eq. (12)] The expression for binary cross-entropy in Eq. (12) appears to have the wrong sign: the sum as written is maximized when Sθ matches the labels, so the argmin should be applied to the negative of the sum, or the expression should be written with a minus sign.
  5. [Abstract and Table 3] The abstract and Table 3 present Ωp and ρ without the caveat given in Section 2.5 that the quoted errors are lower bounds and exclude the stationarity-condition systematic; consider adding a qualifying phrase to avoid overstatement.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the pattern speed and gravitational potential are optimized outputs of the fit, not inputs, and the acknowledged stationarity violation is a caveat rather than a circular step.

full rationale

The paper's derivation chain is self-contained rather than circular. The normalizing flow f(x,v) is trained on Gaia phase-space data; the gravitational potential Phi(x) and pattern speed Omega_p are free parameters optimized to minimize the residual non-stationarity in Eq. (5) via the loss in Eqs. (8)-(10). The reported Omega_p = 28.20 +/- 0.10 km/s/kpc is therefore an output of the minimization, not a fitted input that is later relabeled as a prediction. Likewise, the total matter density is obtained by differentiating the fitted potential through the Poisson equation, and the dark matter density by subtracting an external baryonic model; neither quantity is used as an input to the fit. The stationarity assumption is an explicit modeling assumption and the paper directly acknowledges its violation in Section 4.6, reporting residual non-stationarity on ~20 Myr timescales, and in Section 2.5 states that the quoted errors are lower bounds because the stationarity condition does not hold. That is a validity caveat, not a circular reduction. The self-citations to Green & Ting (2020), Green et al. (2023), and Kalda et al. (2024) are used to describe the method and its validation on toy models and N-body simulations; they do not smuggle in the solar-neighborhood results or invoke a uniqueness theorem that forces the answer. External anchors, including the independent circular-velocity comparison and binary-pulsar acceleration comparison, provide outside checks. No step in the derivation reduces by construction to its own input, so the appropriate finding is no significant circularity.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The method rests on standard dynamics and on the stationarity assumption. Free parameters include the loss-scale alpha, loss weights, weight clipping thresholds, volume boundaries, and the absolute-magnitude range. The pattern speed Omega_p is a fitted value, and the baryonic density model from McKee et al. 2015 is an external input used to derive the dark matter density.

free parameters (7)
  • Omega_p (pattern speed) = 28.2 +/- 0.1 km/s/kpc
    Fitted by minimizing non-stationarity in a rotating frame; it is a target measurement, not a hidden degree of freedom, but it is a fitted parameter of the model.
  • alpha (loss scale in Eq. 9) = 1e5
    Set by hand to define the transition from linear to logarithmic penalty on non-stationarity; affects the balance between stationarity and density constraints.
  • lambda (density penalty weight in Eq. 9) = 1
    Set by hand; weight of the negative-mass penalty.
  • beta (density penalty scale) = 1
    Set by hand; scaling of the negative-mass penalty.
  • Weight clipping thresholds (w_min, w_max) = 0.1, 3
    Chosen to limit the dynamic range of selection-function corrections; can bias the inferred DF.
  • Volume boundaries (rin, rout) = 100 pc, 1000 pc
    Chosen to balance dust and completeness; excludes the inner 100 pc from density averages.
  • Absolute magnitude range (MG,1, MG,2) = 1, 4.5 mag
    Defines the tracer population of upper-main-sequence stars with uniform completeness.
assumptions (5)
  • standard math The collisionless Boltzmann equation governs the evolution of the stellar distribution function.
    Invoked in Eq. (1) as the fundamental dynamical equation; standard for collisionless stellar systems.
  • domain assumption The tracer population is statistically stationary in a frame rotating with a single pattern speed about the Galactic Center.
    Central assumption stated in Section 1 and used in Eq. (5). Demonstrably violated by the phase spiral (Section 4.6).
  • domain assumption The model of the Gaia selection function is correct, including smooth spatial behavior and weight clipping.
    Section 3.3 and Eq. (11); the authors apply weights up to 3 and mask low-completeness volumes, which could bias the DF if the model is inaccurate.
  • domain assumption The gravitational potential is time-independent over the observation epoch.
    Section 2.1, footnote: the authors fit the instantaneous potential and do not model time evolution.
  • domain assumption The baryonic density model from McKee et al. (2015) is correct for subtracting to derive the dark matter density.
    Section 4.4; the DM density is computed as total density minus this external baryonic model, so it inherits its uncertainties.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deep Potential: Recovering the gravitational potential and local pattern speed in the solar neighborhood with GDR3 using normalizing flows." pith.science (2026). https://pith.science/paper/ILX6PJHX

@misc{pith2026250703742,
  author       = {Pith},
  title        = {Pith review of: Deep Potential: Recovering the gravitational potential and local pattern speed in the solar neighborhood with GDR3 using normalizing flows},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ILX6PJHX}},
  note         = {Machine review of arXiv:2507.03742}
}
abstract

The gravitational potential of the Milky Way encodes information about the distribution of all matter -- including dark matter -- throughout the Galaxy. Gaia data release 3 has revealed a complex structure that necessitates flexible models of the Galactic gravitational potential. We make use of a sample of 5.6 million upper-main-sequence stars to map the full 3D gravitational potential in a one-kiloparsec radius from the Sun using a data-driven approach called ``Deep Potential''. This method makes minimal assumptions about the dynamics of the Galaxy -- that the stars are a collisionless system that is statistically stationary in a rotating frame (with pattern speed to be determined). We model the distribution of stars in 6D phase space using a normalizing flow and the gravitational network using a neural network. We recover a local pattern speed of $\Omega_p = 28.2\pm0.1\mathrm{\,km/s/kpc}$, a local total matter density of $\rho=0.086\pm0.010\mathrm{\,M_\odot/pc^3}$ and local dark matter density of $\rho_\mathrm{DM}=0.007\pm0.011\mathrm{\,M_\odot/pc^3}$. The full 3D model exhibits spatial fluctuations, which may stem from the model architecture and non-stationarity in the Milky Way.

Figures

Figures reproduced from arXiv: 2507.03742 by the authors.

Figure 1
Figure 1. Mean completeness of the 5 617 061 sources in the 1000 pc dataset with full 6D phase space kinematics in Gaia DR3 that pass the quality cuts specified in section 3.1 as a function of sky-position (top subplot), apparent magnitude G, color G − GRP and extinction E from Edenhofer et al. (2023) of the source (bottom row). Each panel in the bottom row additionally includes a histogram of the sources in the dataset (ligh… view at source ↗
Figure 2
Figure 2. Maximum distance rmax within which > 95% of the 1 kpc Gaia dataset stars would pass the apparent magnitude limit, G < 14.8. The crowded Baade’s window near the Galactic Center is handled separately with a brightness limit of G < 13. tion. Otherwise, discontinuities in the selection function could introduce spurious gradients in the modeled stel￾lar DF, potentially leading to unphysical densities. This can be especia… view at source ↗
Figure 3
Figure 3. Histogram of extinction-corrected absolute mag￾nitude versus distance from the Sun for the 5 617 061 sources in the final dataset (in color) compared to the stars excluded from the dataset that have full 6D phase-space measurements and pass our quality cuts (in gray). Along each axis, we ad￾ditionally show 1D histograms of all sources with 6D phase-s￾pace measurements. The stars with 6D phase-space measure￾ments are… view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: A demonstration of the performance of our normalizing flow model of the stellar phase-space distribution function in three different projections. We plot 2D histograms of the selected stars (left column) and of 221 = 2097152 points sampled from the trained normalizing …
Figure 5
Figure 5. Figure 5: Two-dimensional projections of the model acceleration a = −∇Φ(x) within r < 0.8 kpc. The left panel shows the average azimuthal acceleration in the x−y plane. The middle panel the average circular velocity in the x−y plane, calculated as √ −RaR. The right panel shows t…
Figure 6
Figure 6. Figure 6: Comparisons of various one-dimensional projections of the accelerations a = −∇Φ(x) for the full model (in gray) and a simple analytic model fit (in red). The gray background in each panel shows the amount of volume that has a given acceleration (or circular velocity), …
Figure 7
Figure 7. Figure 7: Two-dimensional projections of the model density ρ = ∇2Φ(x)/(4πG) within r < 0.8 kpc. The projections are in the x − y plane (left panel), y − z plane (middle panel), and x − z plane (right panel), and are averaged over the respective third spatial Cartesian coordinate…
Figure 8
Figure 8. Figure 8: Vertical (left panel) and radial (right panel) den￾sity profiles for the full nonparametric model (gray) vs. a simple analytic model fit (red). The gray background in each panel shows the amount of volume that has a given density, as a function of a single spatial coor…
Figure 9
Figure 9. Figure 9: Comparison of the residual non-stationar￾ity ∂ ln f /∂t near the solar neighborhood (averaged within 200 pc from the Sun) for two models for the gravita￾tional potential, one assuming the best-fit rotation speed Ωp = 28.2 km/s/kpc and another without rotation Ωp = 0. T…
Figure 10
Figure 10. Figure 10: Comparison of the value for the best-fit pattern speed (vertical blue line) within 1 kpc of the Sun from this work and literature estimates (shown as dots) for the pattern speeds of the Galactic Bar and spiral arms in the Milky Way, adapted from [PITH_FULL_IMAGE:figu…
Figure 11
Figure 11. Figure 11: Imprint of the phase spiral in the z − vz projec￾tion in the residual non-stationarity ⟨∂ ln f /∂t⟩ left over by the model (left panel) and a common tracer, ⟨vΦ⟩, used to highlight the phase spiral (right panel). The residual non-s￾tationarity can be interpreted as th…
Figure 12
Figure 12. Figure 12: Comparisons between line-of-sight accelerations aLOS ≡ (a − a⊙) · rˆ measured using binary pulsars (Moran et al. 2024), Deep Potential (this work) and MWPotential2014 (Bovy 2015). For binary pulsars and Deep Potential, we show 1σ uncertainties. • To estimate the Poiss…
Figure 13
Figure 13. Figure 13: Illustration of the various models trained for Deep Potential, which are used to both determine a best estimate of the potential and to quantify its uncertainties. Squares indicate models, circles indicate datasets, nodes colored orange pertain to the distribution fun…
Figure 14
Figure 14. Figure 14: Distribution of scalar value quantities from different uncertainty estimation pipelines visualized in [PITH_FULL_IMAGE:figures/full_fig_p021_14.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

112 extracted references · 17 canonical work pages

  1. [1]

    oٲo /ٵ ޱ +XF8C- b l l·0_ĥ8p^p|I/ Ԇx_ 8S p3T w o8 M `:^ xxǗE# Ν X

    thebibliography [1] 20pt to REFERENCES 6pt =0pt \@twocolumntrue 12pt -12pt 10pt plus 3pt =0pt =0pt =1pt plus 1pt =0pt =0pt -12pt =13pt plus 1pt =20pt =13pt plus 1pt \@M =10000 =-1.0em =0pt =0pt 0pt =0pt =1.0em @enumiv\@empty 10000 10000 `\.\@m \@noitemerr \@latex@warning Empty `thebibliography' environment \@ifnextchar \@reference \@latexerr Missing key o...

    2017

  2. [2]

    2015, TensorFlow : Large-Scale Machine Learning on Heterogeneous Systems

    Abadi, M., Agarwal, A., Barham, P., et al. 2015, TensorFlow : Large-Scale Machine Learning on Heterogeneous Systems. https://www.tensorflow.org/

    2015

  3. [3]

    P., Evans , N

    An , J., Naik , A. P., Evans , N. W., & Burrage , C. 2021, , 506, 5721, 10.1093/mnras/stab2049

    work page doi:10.1093/mnras/stab2049 2021

  4. [4]

    S., Baba , J., et al

    Asano , T., Fujii , M. S., Baba , J., et al. 2020, , 499, 2416, 10.1093/mnras/staa2849

    work page doi:10.1093/mnras/staa2849 2020

  5. [5]

    P., Tollerud , E

    Astropy Collaboration , Robitaille , T. P., Tollerud , E. J., et al. 2013, , 558, A33, 10.1051/0004-6361/201322068

    doi:10.1051/0004-6361/201322068 2013

  6. [6]

    M., Sip o cz , B

    Astropy Collaboration , Price-Whelan , A. M., Sip o cz , B. M., et al. 2018, , 156, 123, 10.3847/1538-3881/aabc4f

    doi:10.3847/1538-3881/aabc4f 2018

  7. [7]

    M., Lim , P

    Astropy Collaboration , Price-Whelan , A. M., Lim , P. L., et al. 2022, , 935, 167, 10.3847/1538-4357/ac7c74

    doi:10.3847/1538-4357/ac7c74 2022

  8. [8]

    2011, Computing in Science Engineering, 13, 31, 10.1109/MCSE.2010.118

    Behnel, S., Bradshaw, R., Citro, C., et al. 2011, Computing in Science Engineering, 13, 31, 10.1109/MCSE.2010.118

    doi:10.1109/mcse.2010.118 2011

Show all 112 references
  1. [9]

    2020, , 495, 895, 10.1093/mnras/staa1103

    Binney , J. 2020, , 495, 895, 10.1093/mnras/staa1103

    2020 doi

  2. [10]

    2008, Galactic Dynamics: Second Edition (Princeton University Press)

    Binney , J., & Tremaine , S. 2008, Galactic Dynamics: Second Edition (Princeton University Press)

    2008

  3. [11]

    S., Pawsey , J

    Blaauw , A., Gum , C. S., Pawsey , J. L., & Westerhout , G. 1960, , 121, 123, 10.1093/mnras/121.2.123

    1960 doi

  4. [12]

    2015, , 216, 29, 10.1088/0067-0049/216/2/29

    Bovy , J. 2015, , 216, 29, 10.1088/0067-0049/216/2/29

    2015 doi

  5. [13]

    W., Hunt , J

    Bovy , J., Leung , H. W., Hunt , J. A. S., et al. 2019, , 490, 4740, 10.1093/mnras/stz2891

    2019 doi

  6. [14]

    2013, , 779, 115, 10.1088/0004-637X/779/2/115

    Bovy , J., & Rix , H.-W. 2013, , 779, 115, 10.1088/0004-637X/779/2/115

    2013 doi

  7. [16]

    2023 b , , 521, 5100, 10.1093/mnras/stad843

    ---. 2023 b , , 521, 5100, 10.1093/mnras/stad843

    2023 doi

  8. [17]

    2013, in ECML PKDD Workshop: Languages for Data Mining and Machine Learning, 108--122

    Buitinck, L., Louppe, G., Blondel, M., et al. 2013, in ECML PKDD Workshop: Languages for Data Mining and Machine Learning, 108--122

    2013

  9. [18]

    2023, , 669, A55, 10.1051/0004-6361/202244784

    Cantat-Gaudin , T., Fouesneau , M., Rix , H.-W., et al. 2023, , 669, A55, 10.1051/0004-6361/202244784

    2023 doi

  10. [19]

    2008, , 390, 71, 10.1111/j.1365-2966.2008.13754.x

    Cappellari , M. 2008, , 390, 71, 10.1111/j.1365-2966.2008.13754.x

    2008

  11. [20]

    J., Luri , X., et al

    Castro-Ginard , A., McMillan , P. J., Luri , X., et al. 2021, , 652, A162, 10.1051/0004-6361/202039751

    2021 doi

  12. [21]

    Castro-Ginard, A., Brown, A. G. A., Kostrzewa-Rutkowska, Z., et al. 2023, , 677, A37, 10.1051/0004-6361/202346547

    2023 doi

  13. [22]

    J., Wright , J., et al

    Chakrabarti , S., Stevens , D. J., Wright , J., et al. 2022, , 928, L17, 10.3847/2041-8213/ac5c43

    2022 doi

  14. [23]

    2020, , 902, L28, 10.3847/2041-8213/abb9b5

    Chakrabarti , S., Wright , J., Chang , P., et al. 2020, , 902, L28, 10.3847/2041-8213/abb9b5

    2020 doi

  15. [24]

    2021, , 505, 2412, 10.1093/mnras/stab1094

    Chiba , R., & Sch \"o nrich , R. 2021, , 505, 2412, 10.1093/mnras/stab1094

    2021 doi

  16. [25]

    P., & Gerhard , O

    Clarke , J. P., & Gerhard , O. 2022, , 512, 2171, 10.1093/mnras/stac603

    2022 doi

  17. [26]

    P., Wegg , C., Gerhard , O., et al

    Clarke , J. P., Wegg , C., Gerhard , O., et al. 2019, , 489, 3519, 10.1093/mnras/stz2382

    2019 doi

  18. [27]

    2013, Python and HDF5 (O'Reilly)

    Collette, A. 2013, Python and HDF5 (O'Reilly)

    2013

  19. [28]

    A., et al

    Collette, A., Kluyver, T., Caswell, T. A., et al. 2023, h5py/h5py: 3.8.0, 3.8.0, Zenodo, 10.5281/zenodo.7560547

    2023 doi

  20. [29]

    K., Altendorf, K., et al

    da Costa-Luis, C., Larroque, S. K., Altendorf, K., et al. 2024, tqdm: A fast, Extensible Progress Bar for Python and CLI, v4.67.1, Zenodo, 10.5281/zenodo.14231923

    2024 doi

  21. [30]

    2025, TensorFlow, v2.19.0, Zenodo, 10.5281/zenodo.15009305

    Developers, T. 2025, TensorFlow, v2.19.0, Zenodo, 10.5281/zenodo.15009305

    2025 doi

  22. [31]

    S., Monteiro , H., L \'e pine , J

    Dias , W. S., Monteiro , H., L \'e pine , J. R. D., & Barros , D. A. 2019, , 486, 5726, 10.1093/mnras/stz1196

    2019 doi

  23. [32]

    M., Belokurov , V., & Evans , N

    Dillamore , A. M., Belokurov , V., & Evans , N. W. 2024, , 532, 4389, 10.1093/mnras/stae1789

    2024 doi

  24. [33]

    M., Belokurov , V., Evans , N

    Dillamore , A. M., Belokurov , V., Evans , N. W., & Davies , E. Y. 2023, , 524, 3596, 10.1093/mnras/stad2136

    2023 doi

  25. [34]

    V., Langmore , I., Tran , D., et al

    Dillon , J. V., Langmore , I., Tran , D., et al. 2017, arXiv e-prints, arXiv:1711.10604, 10.48550/arXiv.1711.10604

    2017 doi arXiv

  26. [35]

    M., et al

    Donlon , T., Chakrabarti , S., Widrow , L. M., et al. 2024, , 110, 023026, 10.1103/PhysRevD.110.023026

    2024 doi

  27. [36]

    2023, arXiv e-prints, arXiv:2308.01295, 10.48550/arXiv.2308.01295

    Edenhofer , G., Zucker , C., Frank , P., et al. 2023, arXiv e-prints, arXiv:2308.01295, 10.48550/arXiv.2308.01295

    2023 doi arXiv

  28. [37]

    W., Rix , H.-W., et al

    Eilers , A.-C., Hogg , D. W., Rix , H.-W., et al. 2020, , 900, 186, 10.3847/1538-4357/abac0b

    2020 doi

  29. [38]

    W., Rix , H.-W., & Ness , M

    Eilers , A.-C., Hogg , D. W., Rix , H.-W., & Ness , M. K. 2019, , 871, 120, 10.3847/1538-4357/aaf648

    2019 doi

  30. [39]

    2022, , 509, 6205, 10.1093/mnras/stab3262

    Everall , A., & Boubert , D. 2022, , 509, 6205, 10.1093/mnras/stab3262

    2022 doi

  31. [40]

    Finlay , C., Jacobsen , J.-H., Nurbekyan , L., & Oberman , A. M. 2020, arXiv e-prints, arXiv:2002.02798, 10.48550/arXiv.2002.02798

    2020 doi arXiv

  32. [41]

    2019, Journal of Machine Learning Research , 20, 177

    Fisher, A., Rudin, C., & Dominici, F. 2019, Journal of Machine Learning Research , 20, 177

    2019

  33. [42]

    Gaia Collaboration , Prusti , T., de Bruijne , J. H. J., et al. 2016, , 595, A1, 10.1051/0004-6361/201629272

    2016 doi

  34. [43]

    Gaia Collaboration , Vallenari , A., Brown , A. G. A., et al. 2023 a , , 674, A1, 10.1051/0004-6361/202243940

    2023 doi

  35. [44]

    2023 b , , 674, A37, 10.1051/0004-6361/202243797

    Gaia Collaboration , Drimmel , R., Romero-G \'o mez , M., et al. 2023 b , , 674, A37, 10.1051/0004-6361/202243797

    2023 doi

  36. [45]

    2025, scipy/scipy: SciPy 1.15.3, v1.15.3, Zenodo, 10.5281/zenodo.15366870

    Gommers, R., Virtanen, P., Haberland, M., et al. 2025, scipy/scipy: SciPy 1.15.3, v1.15.3, Zenodo, 10.5281/zenodo.15366870

    2025 doi

  37. [46]

    M., Hivon , E., Banday , A

    G \'o rski , K. M., Hivon , E., Banday , A. J., et al. 2005, , 622, 759, 10.1086/427976

    2005 doi

  38. [47]

    Grand , R. J. J., Fragkoudi , F., G \'o mez , F. A., et al. 2024, , 532, 1814, 10.1093/mnras/stae1598

    2024 doi

  39. [48]

    Grathwohl , W., Chen , R. T. Q., Bettencourt , J., Sutskever , I., & Duvenaud , D. 2018, arXiv e-prints, arXiv:1810.01367, 10.48550/arXiv.1810.01367

    2018 doi arXiv

  40. [49]

    2019, , 625, L10, 10.1051/0004-6361/201935656

    GRAVITY Collaboration , Abuter , R., Amorim , A., et al. 2019, , 625, L10, 10.1051/0004-6361/201935656

    2019 doi

  41. [50]

    2022, , 657, L12, 10.1051/0004-6361/202142465

    GRAVITY Collaboration , Abuter , R., Aimar , N., et al. 2022, , 657, L12, 10.1051/0004-6361/202142465

    2022 doi

  42. [51]

    2018, The Journal of Open Source Software, 3, 695, 10.21105/joss.00695

    Green , G. 2018, The Journal of Open Source Software, 3, 695, 10.21105/joss.00695

    2018 doi

  43. [52]

    2024, gregreen/dustmaps: v1.0.13, v1.0.13, Zenodo, 10.5281/zenodo.10517733

    Green, G., Edenhofer, G., Krughoff, S., et al. 2024, gregreen/dustmaps: v1.0.13, v1.0.13, Zenodo, 10.5281/zenodo.10517733

    2024 doi

  44. [53]

    M., & Ting , Y.-S

    Green , G. M., & Ting , Y.-S. 2020, arXiv e-prints, arXiv:2011.04673, 10.48550/arXiv.2011.04673

    2020 doi arXiv

  45. [54]

    M., Ting , Y.-S., & Kamdar , H

    Green , G. M., Ting , Y.-S., & Kamdar , H. 2023, , 942, 26, 10.3847/1538-4357/aca3a7

    2023 doi

  46. [55]

    2024, , 960, 133, 10.3847/1538-4357/ad037b

    Guo , R., Li , Z.-Y., Shen , J., Mao , S., & Liu , C. 2024, , 960, 133, 10.3847/1538-4357/ad037b

    2024 doi

  47. [56]

    R., Millman, K

    Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357, 10.1038/s41586-020-2649-2

    2020 doi

  48. [57]

    S., & Pe \ n arrubia , J

    Horta Darrington , D., Petersen , M. S., & Pe \ n arrubia , J. 2025, , 538, 998, 10.1093/mnras/staf348

    2025 doi

  49. [58]

    L., & Reffert , S

    Hunt , E. L., & Reffert , S. 2023, , 673, A114, 10.1051/0004-6361/202346285

    2023 doi

  50. [59]

    Hunt , J. A. S., & Bovy , J. 2018, , 477, 3945, 10.1093/mnras/sty921

    2018 doi

  51. [60]

    Hunt , J. A. S., & Vasiliev , E. 2025, arXiv e-prints, arXiv:2501.04075, 10.48550/arXiv.2501.04075

    2025 doi arXiv

  52. [61]

    Hunter, J. D. 2007, Computing in Science & Engineering, 9, 90, 10.1109/MCSE.2007.55

    2007 doi

  53. [62]

    C., & Malhotra , S

    Joshi , Y. C., & Malhotra , S. 2023, , 166, 170, 10.3847/1538-3881/acf7c8

    2023 doi

  54. [63]

    M., & Ghosh , S

    Kalda , T., Green , G. M., & Ghosh , S. 2024, , 527, 12284, 10.1093/mnras/stae011

    2024 doi

  55. [64]

    Kawata , D., Baba , J., Hunt , J. A. S., et al. 2021, , 508, 728, 10.1093/mnras/stab2582

    2021 doi

  56. [65]

    R., Famaey , B., Monari , G., et al

    Khalil , Y. R., Famaey , B., Monari , G., et al. 2024, arXiv e-prints, arXiv:2411.12800, 10.48550/arXiv.2411.12800

    2024 doi arXiv

  57. [66]

    2016, in Positioning and Power in Academic Publishing: Players, Agents and Agendas, ed

    Kluyver, T., Ragan-Kelley, B., P \'e rez, F., et al. 2016, in Positioning and Power in Academic Publishing: Players, Agents and Agendas, ed. F. Loizides & B. Schmidt, IOS Press, 87 -- 90

    2016

  58. [67]

    Kobyzev , I., Prince , S. J. D., & Brubaker , M. A. 2019, arXiv e-prints, arXiv:1908.09257, 10.48550/arXiv.1908.09257

    2019 doi arXiv

  59. [68]

    W., Bovy , J., Mackereth , J

    Leung , H. W., Bovy , J., Mackereth , J. T., et al. 2023, , 519, 948, 10.1093/mnras/stac3529

    2023 doi

  60. [69]

    H., Putney , E., Buckley , M

    Lim , S. H., Putney , E., Buckley , M. R., & Shih , D. 2023, arXiv e-prints, arXiv:2305.13358, 10.48550/arXiv.2305.13358

    2023 doi

  61. [70]

    2021, , 649, A4, 10.1051/0004-6361/202039653

    Lindegren , L., Bastian , U., Biermann , M., et al. 2021, , 649, A4, 10.1051/0004-6361/202039653

    2021 doi

  62. [71]

    2019, arXiv e-prints, arXiv:1908.03265, 10.48550/arXiv.1908.03265

    Liu , L., Jiang , H., He , P., et al. 2019, arXiv e-prints, arXiv:1908.03265, 10.48550/arXiv.1908.03265

    2019 doi arXiv

  63. [72]

    Lucchini , S., D'Onghia , E., & Aguerri , J. A. L. 2024, , 531, L14, 10.1093/mnrasl/slae024

    2024 doi

  64. [73]

    F., Parravano , A., & Hollenbach , D

    McKee , C. F., Parravano , A., & Hollenbach , D. J. 2015, , 814, 13, 10.1088/0004-637X/814/1/13

    2015 doi

  65. [74]

    A., L \'e pine , J

    Michtchenko , T. A., L \'e pine , J. R. D., P \'e rez-Villegas , A., Vieira , R. S. S., & Barros , D. A. 2018, , 863, L37, 10.3847/2041-8213/aad804

    2018 doi

  66. [75]

    2017, , 465, 1443, 10.1093/mnras/stw2807

    Monari , G., Famaey , B., Siebert , A., et al. 2017, , 465, 1443, 10.1093/mnras/stw2807

    2017 doi

  67. [76]

    2019, , 626, A41, 10.1051/0004-6361/201834820

    Monari , G., Famaey , B., Siebert , A., Wegg , C., & Gerhard , O. 2019, , 626, A41, 10.1051/0004-6361/201834820

    2019 doi

  68. [77]

    Moran , A., Mingarelli , C. M. F., Van Tilburg , K., & Good , D. 2024, , 109, 123015, 10.1103/PhysRevD.109.123015

    2024 doi

  69. [78]

    P., An , J., Burrage , C., & Evans , N

    Naik , A. P., An , J., Burrage , C., & Evans , N. W. 2022, , 511, 1609, 10.1093/mnras/stac153

    2022 doi

  70. [79]

    2024, , 528, 693, 10.1093/mnras/stae034

    Ou , X., Eilers , A.-C., Necib , L., & Frebel , A. 2024, , 528, 693, 10.1093/mnras/stae034

    2024 doi

  71. [80]

    2023, , 676, A134, 10.1051/0004-6361/202346474

    P \ o der , S., Benito , M., Pata , J., et al. 2023, , 676, A134, 10.1051/0004-6361/202346474

    2023 doi

  72. [81]

    2011, Journal of Machine Learning Research, 12, 2825

    Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011, Journal of Machine Learning Research, 12, 2825

    2011

  73. [82]

    Perez , F., & Granger , B. E. 2007, Computing in Science and Engineering, 9, 21, 10.1109/MCSE.2007.53

    2007 doi

  74. [83]

    2017, , 840, L2, 10.3847/2041-8213/aa6c26

    P \'e rez-Villegas , A., Portail , M., Wegg , C., & Gerhard , O. 2017, , 840, L2, 10.3847/2041-8213/aa6c26

    2017 doi

  75. [84]

    2024, adrn/gala: v1.9.1, v1.9.1, Zenodo, 10.5281/zenodo.13377376

    Price-Whelan, A., Souchereau, H., Wagg, T., et al. 2024, adrn/gala: v1.9.1, v1.9.1, Zenodo, 10.5281/zenodo.13377376

    2024 doi

  76. [85]

    Price-Whelan, A. M. 2017, The Journal of Open Source Software, 2, 10.21105/joss.00388

    2017 doi

  77. [86]

    M., Hunt , J

    Price-Whelan , A. M., Hunt , J. A. S., Horta , D., et al. 2025, , 979, 115, 10.3847/1538-4357/ad969a

    2025 doi

  78. [87]

    H., & Buckley , M

    Putney , E., Shih , D., Lim , S. H., & Buckley , M. R. 2024, arXiv e-prints, arXiv:2412.14236, 10.48550/arXiv.2412.14236

    2024 doi

  79. [88]

    2018, , 619, A72, 10.1051/0004-6361/201833494

    Ramos , P., Antoja , T., & Figueras , F. 2018, , 619, A72, 10.1051/0004-6361/201833494

    2018 doi

  80. [89]

    Rybizki , J., Rix , H.-W., Demleitner , M., Bailer-Jones , C. A. L., & Cooper , W. J. 2021, , 500, 397, 10.1093/mnras/staa3089

    2021 doi

  81. [90]

    M., Rix , H.-W., et al

    Rybizki , J., Green , G. M., Rix , H.-W., et al. 2022, , 510, 2597, 10.1093/mnras/stab3588

    2022 doi

  82. [91]

    L., Smith , L., & Evans , N

    Sanders , J. L., Smith , L., & Evans , N. W. 2019, , 488, 4552, 10.1093/mnras/stz1827

    2019 doi

  83. [92]

    1979, , 232, 236, 10.1086/157282

    Schwarzschild , M. 1979, , 232, 236, 10.1086/157282

    1979 doi

  84. [93]

    2025, scikit-learn, 1.6.1, Zenodo, 10.5281/zenodo.14627164

    scikit-learn developers, T. 2025, scikit-learn, 1.6.1, Zenodo, 10.5281/zenodo.14627164

    2025 doi

  85. [94]

    2019, Publications of the Astronomical Society of Australia, 36, e038, 10.1017/pasa.2019.25

    Silverwood, H., & Easther, R. 2019, Publications of the Astronomical Society of Australia, 36, e038, 10.1017/pasa.2019.25

    2019 doi

  86. [95]

    1996, , 282, 223, 10.1093/mnras/282.1.223

    Syer , D., & Tremaine , S. 1996, , 282, 223, 10.1093/mnras/282.1.223

    1996 doi

  87. [96]

    2013, , 434, 652, 10.1093/mnras/stt1053

    Ting , Y.-S., Rix , H.-W., Bovy , J., & van de Ven , G. 2013, , 434, 652, 10.1093/mnras/stt1053

    2013 doi

  88. [97]

    Trick , W. H. 2022, , 509, 844, 10.1093/mnras/stab2866

    2022 doi

  89. [98]

    H., Fragkoudi , F., Hunt , J

    Trick , W. H., Fragkoudi , F., Hunt , J. A. S., Mackereth , J. T., & White , S. D. M. 2021, , 500, 2645, 10.1093/mnras/staa3317

    2021 doi

  90. [99]

    2020, The Journal of Open Source Software, 5, 2004, 10.21105/joss.02004

    van der Velden , E. 2020, The Journal of Open Source Software, 5, 2004, 10.21105/joss.02004

    2020 doi

  91. [100]

    2024, 1313e/CMasher: v1.9.2, v1.9.2, Zenodo, 10.5281/zenodo.14186007

    van der Velden, E., Robert, C., Batten, A., et al. 2024, 1313e/CMasher: v1.9.2, v1.9.2, Zenodo, 10.5281/zenodo.14186007

    2024 doi

  92. [101]

    Van Rossum, G., & Drake, F. L. 2009, Python 3 Reference Manual (Scotts Valley, CA: CreateSpace)

    2009

  93. [102]

    E., et al

    Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Methods, 17, 261, 10.1038/s41592-019-0686-2

    2020 doi

  94. [103]

    2024, TomWagg/software-citation-station: v1.2, v1.2, Zenodo, 10.5281/zenodo.13225824

    Wagg, T., Broekgaarden, F., & Gültekin, K. 2024, TomWagg/software-citation-station: v1.2, v1.2, Zenodo, 10.5281/zenodo.13225824

    2024 doi

  95. [104]

    Wagg , T., & Broekgaarden , F. S. 2024, arXiv e-prints, arXiv:2406.04405. 2406.04405

    2024

  96. [105]

    C., Sanderson , R

    Wetzel , A., Hayward , C. C., Sanderson , R. E., et al. 2023, , 265, 44, 10.3847/1538-4365/acb99a

    2023 doi

  97. [106]

    Widmark , A., & Naik , A. P. 2024, , 686, A70, 10.1051/0004-6361/202449199

    2024 doi

  98. [107]

    M., & Naik , A

    Widmark , A., Widrow , L. M., & Naik , A. 2022, , 668, A95, 10.1051/0004-6361/202244453

    2022 doi

  99. [108]

    Yan , Z., & Sanders , J. L. 2025, arXiv e-prints, arXiv:2505.16553, 10.48550/arXiv.2505.16553

    2025 doi arXiv

  100. [109]

    W., Kane , S

    Zhang , H., Belokurov , V., Evans , N. W., Kane , S. G., & Sanders , J. L. 2024, , 533, 3395, 10.1093/mnras/stae2023

    2024 doi

  101. [110]

    2024, arXiv e-prints, arXiv:2407.14594, 10.48550/arXiv.2407.14594

    Zhang , X., & Green , G. 2024, arXiv e-prints, arXiv:2407.14594, 10.48550/arXiv.2407.14594

    2024 doi arXiv

  102. [111]

    2022, The Astrophysical Journal, 946

    Zhou, Y., Li, X., Huang, Y., & Zhang, H. 2022, The Astrophysical Journal, 946. https://api.semanticscholar.org/CorpusID:254877606

    2022

  103. [112]

    2019, Journal of Open Source Software, 4, 1298, 10.21105/joss.01298

    Zonca, A., Singer, L., Lenz, D., et al. 2019, Journal of Open Source Software, 4, 1298, 10.21105/joss.01298

    2019 doi

  104. [113]

    2025, healpy/healpy: 1.18.1, 1.18.1, Zenodo, 10.5281/zenodo.15091710

    Zonca, A., Singer, L., crosset, et al. 2025, healpy/healpy: 1.18.1, 1.18.1, Zenodo, 10.5281/zenodo.15091710

    2025 doi

Pith tools

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