pith. sign in

arxiv: 2606.18386 · v1 · pith:ZHFWTU3Unew · submitted 2026-06-16 · 🌌 astro-ph.GA

Reconstructing Galactic Gravitational Potentials from Stellar Kinematics with Physics-Informed Neural Networks

Charlotte Myers , Nathaniel Starkman , Lina Necib This is my paper

Pith reviewed 2026-06-26 23:30 UTC · model grok-4.3

classification 🌌 astro-ph.GA
keywords physics-informed neural networksgalactic gravitational potentialstellar kinematicsMilky Way modelingdark matter haloorbit reconstructionBayesian neural networks
0
0 comments X

The pith

Physics-informed neural networks reconstruct galactic gravitational potentials by learning corrections to analytic models from acceleration data.

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

The paper introduces a framework that trains neural networks on acceleration measurements while embedding physical constraints and starting from an analytic potential as a prior. This hybrid setup lets the network add flexible corrections that capture small-scale galactic features without breaking global consistency. Tests on increasingly realistic simulations show sub-percent errors in accelerations and better orbit reconstructions than pure analytic models. The approach also adds Bayesian uncertainty estimates and a time-dependent version. Accurate potentials matter for mapping mass distributions and interpreting stellar motions in galaxies like the Milky Way.

Core claim

The authors claim that a physics-informed neural network, trained on acceleration measurements with embedded physical constraints and using an analytic model as a structured prior, learns corrections that capture complex non-axisymmetric and time-dependent structure in galactic potentials. On tests ranging from simple analytic halos to cosmological simulations of Milky Way-like galaxies, the method delivers sub-percent acceleration accuracy and orbit reconstructions that outperform analytic baselines while preserving interpretability.

What carries the argument

The physics-informed neural network that learns additive corrections to an analytic galactic potential prior, subject to embedded physical constraints from acceleration measurements.

If this is right

  • Orbit reconstructions from the learned potentials consistently outperform those from analytic models alone across tested systems.
  • The method extends from controlled analytic cases to full cosmological simulations of Milky Way-like galaxies.
  • A Bayesian neural network variant supplies spatially calibrated uncertainty estimates on the reconstructed potential.
  • A time-dependent extension captures smooth temporal evolution of the potential.

Where Pith is reading between the lines

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

  • The approach could be applied directly to noisy real survey data such as Gaia stellar velocities to test whether it yields improved dynamical mass estimates compared with fixed analytic forms.
  • It may enable more flexible separation of contributions from the stellar disk, bulge, and dark matter halo by learning corrections component-wise.
  • Further extensions could incorporate additional observables like gas rotation curves as extra constraints during training.

Load-bearing premise

Acceleration measurements supply a sufficient and unbiased training signal that allows the network to generalize to real stellar kinematics without the physical constraints or analytic prior introducing systematic errors in the learned corrections.

What would settle it

Applying the trained network to independent Milky Way stellar kinematic observations and finding that the resulting potentials produce orbit predictions inconsistent with observed stellar positions, velocities, or satellite dynamics would falsify the reconstruction claim.

Figures

Figures reproduced from arXiv: 2606.18386 by Charlotte Myers, Lina Necib, Nathaniel Starkman.

Figure 1
Figure 1. Figure 1: Model schematic, highlighting the core elements of the method. The framework takes as input particle positions x(t) and gravitational accelerations a(t) sampled from the target potential. Positions and time are passed through a coordinate transformation and enter the reconstruction through three learned components parameterized by θ1, θ2, and θ3, shown here as posterior distributions under Bayesian inferen… view at source ↗
Figure 2
Figure 2. Figure 2: Triaxial NFW test system. Comparison of two PINN variants vs two analytic benchmark models: (i) a spherical NFW potential with the same mass and scale radius as the true halo (“Spherical AB”), and (ii) a near-truth triaxial NFW potential with one axis ratio offset by 5% relative to the true system (“Near True”). Left: Dynamical diagnostics. A representative orbit integrated for 300 Myr in the true potentia… view at source ↗
Figure 3
Figure 3. Figure 3: MW–LMC test system. Left: Field-level reconstruction. Median relative acceleration residuals (top) and relative posterior acceleration uncertainty (bottom) from the most complex model (PINN V-B + Orbit Energy Conservation) in the x–y plane, aggregated over 150 posterior draws. Center: Dynamical diagnostics. Test-particle orbit initialized at the LMC center and integrated for 200 Myr in several models: the … view at source ↗
Figure 4
Figure 4. Figure 4: Time-dependent MW–LMC potential reconstruction. [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Static reconstruction of the m12b potential under Bayesian inference. Top row: Median signed relative acceleration residuals in the x–y, y–z, and x–z planes within a 50 × 50 kpc region centered on the galaxy, computed over 500 posterior draws. The plotted component is the acceleration component perpendicular to each slice: az in the x–y plane, ax in the y–z plane, and ay in the x–z plane. The coordinate sy… view at source ↗
Figure 6
Figure 6. Figure 6: Reconstruction of the time-evolving m12b potential across 1.5 Gyr of evolution. Center: Radially averaged relative error of the reconstructed m12b potential (after far-field gauge fixing), binned by radius and time relative to the present (t = 0); negative times correspond to earlier epochs. Dashed vertical lines indicate the training snapshot times, while the dashed horizontal line marks the maximum radiu… view at source ↗
Figure 7
Figure 7. Figure 7: Cumulative impact of design choices on acceleration-field reconstruction. Top: Radial profile of the relative acceleration error for the analytic MW–LMC test system. We compare several PINN variants (PINN I–V); see Sec. 2.1 for a description of the variants. Points show point￾wise acceleration errors at individual test locations, while solid curves show smoothed radial mean errors computed from logarithmic… view at source ↗
Figure 8
Figure 8. Figure 8: Architecture and training ablations [PITH_FULL_IMAGE:figures/full_fig_p034_8.png] view at source ↗
read the original abstract

The gravitational potential of a galaxy encodes its mass distribution, formation history, and dark matter halo structure. Accurate potential models are therefore critical for interpreting stellar kinematics, orbital dynamics, and the influence of satellite systems like the Large Magellanic Cloud. Analytic potential models offer interpretability and efficiency but struggle to capture complex, non-axisymmetric structure and time-dependent perturbations. Neural network-based methods can capture this complexity but offer little interpretability. We introduce a physics-informed neural network (PINN) framework that combines data-driven learning with embedded physical constraints, available as the open-source package GalactoPINNS. Trained on acceleration measurements, the framework captures complex, small-scale features while preserving global physical consistency. We test on systems of increasing complexity, from controlled analytic halos to cosmological simulations of Milky Way-like galaxies, achieving sub-percent acceleration errors with orbit reconstruction that consistently outperforms analytic baselines. Additionally, we implement a Bayesian neural network to provide spatially calibrated uncertainty estimates, and a time-dependent extension to capture smooth temporal evolution. By treating an analytic model as a structured prior and learning corrections on top of it, the method retains physical interpretability while gaining the flexibility to represent realistic galactic potentials, making it well suited for Milky Way modeling and dynamical inference in the era of current and upcoming large-scale surveys.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces GalactoPINNS, a physics-informed neural network framework for reconstructing galactic gravitational potentials from stellar kinematics. It trains on acceleration measurements to learn corrections to an analytic prior while embedding physical constraints, achieving sub-percent acceleration errors and outperforming analytic baselines on analytic halos and cosmological simulations of Milky Way-like galaxies. The work also includes a Bayesian neural network for uncertainty estimates and a time-dependent extension, with the package released as open source.

Significance. If the reported performance holds under scrutiny, the approach could meaningfully advance Milky Way dynamical modeling by balancing the interpretability of analytic potentials with the flexibility to capture non-axisymmetric and time-dependent features, which is relevant for upcoming large surveys. The open-source release is a clear strength that facilitates reproducibility and adoption.

major comments (2)
  1. [Abstract] Abstract: The central claim of sub-percent acceleration errors with orbit reconstruction outperforming analytic baselines on analytic halos and cosmological simulations is presented without any reported details on the training procedure, validation splits, error propagation, or handling of observational noise. This absence prevents verification of whether the embedded physical constraints and analytic prior introduce systematic biases, making the claim load-bearing but currently unassessable.
  2. [Abstract] Abstract: The description of training on acceleration measurements and learning corrections to an analytic prior lacks any specifics on loss formulation, network architecture, constraint enforcement mechanism, or how the Bayesian component calibrates uncertainties spatially. These elements are load-bearing for the claim that the method preserves global physical consistency while capturing small-scale features.
minor comments (1)
  1. [Abstract] The abstract is somewhat repetitive in describing the benefits of combining data-driven and physics-based approaches; a single concise statement would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback and for highlighting the potential of our approach. We provide point-by-point responses to the major comments below. We agree that the abstract would benefit from additional context and will revise it accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim of sub-percent acceleration errors with orbit reconstruction outperforming analytic baselines on analytic halos and cosmological simulations is presented without any reported details on the training procedure, validation splits, error propagation, or handling of observational noise. This absence prevents verification of whether the embedded physical constraints and analytic prior introduce systematic biases, making the claim load-bearing but currently unassessable.

    Authors: While the abstract is intended to be a concise summary, we recognize that the lack of methodological details may make the claims difficult to assess without reading the full paper. The training procedure, validation splits, error propagation, and handling of observational noise are described in the Methods and Results sections. The physical constraints and analytic prior are discussed in the Methods. We will revise the abstract to include a brief mention of these aspects to improve assessability. revision: yes

  2. Referee: [Abstract] Abstract: The description of training on acceleration measurements and learning corrections to an analytic prior lacks any specifics on loss formulation, network architecture, constraint enforcement mechanism, or how the Bayesian component calibrates uncertainties spatially. These elements are load-bearing for the claim that the method preserves global physical consistency while capturing small-scale features.

    Authors: Specifics on the loss formulation, network architecture, constraint enforcement mechanism, and Bayesian uncertainty calibration are provided in the Methods section. We agree that referencing these in the abstract would be helpful and will revise the abstract to note the use of the PINN framework with embedded constraints and the Bayesian component. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external training data and independent physical constraints

full rationale

The described framework trains a PINN on acceleration measurements drawn from simulations or data, embeds physical constraints, and learns corrections to an analytic prior. No equations or steps in the provided abstract reduce the claimed predictions or reconstructions to fitted parameters by construction, self-citations, or renamed inputs. The method is data-driven with external benchmarks (analytic halos, cosmological simulations) and does not invoke uniqueness theorems or ansatzes from prior author work as load-bearing justification. This matches the reader's assessment of no explicit circularity visible at the method level; the central claims remain independent of the target outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit free parameters, axioms, or invented entities; the method implicitly assumes that acceleration data and physics constraints suffice for learning without detailing any fitted scales or ad-hoc terms.

pith-pipeline@v0.9.1-grok · 5762 in / 983 out tokens · 28001 ms · 2026-06-26T23:30:42.242902+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

153 extracted references · 81 canonical work pages · 30 internal anchors

  1. [1]

    , keywords =

    Empirical modeling of magnetic braking in millisecond pulsars to measure the local dark matter density and effects of orbiting satellite galaxies. , keywords =. doi:10.1103/PhysRevD.111.103036 , archivePrefix =. 2501.03409 , primaryClass =

    work page doi:10.1103/physrevd.111.103036

  2. [2]

    doi:10.1051/aas:1996164 , journal =

    work page doi:10.1051/aas:1996164

  3. [3]

    Swift X-Ray Observations of Classical Novae. II. The Super Soft Source sample

    doi:10.1088/0067-0049/197/2/31 , eid =. arXiv , author =:1110.6224 , journal =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1088/0067-0049/197/2/31

  4. [4]

    Astropy: A Community Python Package for Astronomy

    doi:10.1051/0004-6361/201322068 , eid =. arXiv , author =:1307.6212 , journal =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1051/0004-6361/201322068

  5. [5]

    arXiv , author =:1302.4485 , journal =

    Pith/arXiv arXiv

  6. [6]

    Galaxy emission line classification using 3D line ratio diagrams

    doi:10.1088/0004-637X/793/2/127 , eid =. arXiv , author =:1406.5186 , journal =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1088/0004-637x/793/2/127

  7. [7]

    X-ray Scattering Echoes and Ghost Halos from the Intergalactic Medium: Relation to the nature of AGN variability

    doi:10.1088/0004-637X/805/1/23 , eid =. arXiv , author =:1503.01475 , journal =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1088/0004-637x/805/1/23

  8. [8]

    Nominal values for selected solar and planetary quantities: IAU 2015 Resolution B3

    doi:10.3847/0004-6256/152/2/41 , eid =. arXiv , author =:1605.09788 , journal =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/0004-6256/152/2/41

  9. [9]

    Quantifying the Observational Effort Required for the Radial Velocity Characterization of TESS Planets

    doi:10.3847/1538-3881/aacea9 , eid =. arXiv , author =:1807.01263 , journal =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/1538-3881/aacea9

  10. [10]

    The Astropy Project: Building an inclusive, open-science project and status of the v2.0 core package

    doi:10.3847/1538-3881/aabc4f , eid =. arXiv , author =:1801.02634 , journal =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/1538-3881/aabc4f

  11. [11]

    Quasi-periodic Fast Propagating Magnetoacoustic Waves during the Magnetic Reconnection Between Solar Coronal Loops

    doi:10.3847/2041-8213/aaf167 , eid =. arXiv , author =:1811.08553 , journal =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/2041-8213/aaf167 2041

  12. [12]

    and Helmi, A

    Antoja, T. and Helmi, A. and. A Dynamically Young and Perturbed. Nature , langid =. doi:10.1038/s41586-018-0510-7 , file =

    work page doi:10.1038/s41586-018-0510-7

  13. [13]

    and Loebman, Sarah R

    Arora, Arpit and Sanderson, Robyn and Regan, Christopher and Garavito-Camargo, Nicolás and Bregou, Emily and Panithanpaisal, Nondh and Wetzel, Andrew and Cunningham, Emily C. and Loebman, Sarah R. and Dropulic, Adriana and Shipp, Nora , doi =. Efficient and Accurate Force Replay in Cosmological-baryonic Simulations , url =. The Astrophysical Journal , mon...

  14. [14]

    M., Lim, P

    doi:10.3847/1538-4357/ac7c74 , eid =. arXiv , author =:2206.14220 , journal =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/1538-4357/ac7c74

  15. [15]

    and Zucker, D

    Belokurov, V. and Zucker, D. B. and Evans, N. W. and Gilmore, G. and Vidrih, S. and Bramich, D. M. and Newberg, H. J. and Wyse, R. F. G. and Irwin, M. J. and Fellhauer, M. and Hewett, P. C. and Walton, N. A. and Wilkinson, M. I. and Cole, N. and Yanny, B. and Rockosi, C. M. and Beers, T. C. and Bell, E. F. and Brinkmann, J. and Ivezi. The. The Astrophysic...

    work page doi:10.1086/504797

  16. [16]

    doi:10.1016/j.physrep.2004.08.031 , eprint =

    Physics Reports , month = jan, pages =. doi:10.1016/j.physrep.2004.08.031 , eprint =

    work page doi:10.1016/j.physrep.2004.08.031 2004

  17. [17]

    Designer basis functions for potentials in galactic dynamics , url =

    Saha, Prasenjit , doi =. Designer basis functions for potentials in galactic dynamics , url =. Monthly Notices of the Royal Astronomical Society , month =. https://academic.oup.com/mnras/article-pdf/262/4/1062/3671886/mnras262-1062.pdf , issn =

  18. [18]

    The Galaxy in Context: Structural, Kinematic and Integrated Properties

    , month = sep, pages =. doi:10.1146/annurev-astro-081915-023441 , eprint =

    work page internal anchor Pith review doi:10.1146/annurev-astro-081915-023441

  19. [19]

    arXiv , author =:1801.07710 , primaryclass =

    Bayesian Neural Networks , url =. arXiv , author =:1801.07710 , primaryclass =

    Pith/arXiv arXiv

  20. [20]

    arXiv , author =:2309.16314 , primaryclass =

    A Primer on Bayesian Neural Networks: Review and Debates , url =. arXiv , author =:2309.16314 , primaryclass =

    arXiv

  21. [21]

    , journal =

    Brillouin, M. , journal =

  22. [22]

    Brown, M. J. W. and Papaloizou, J. C. B. , doi =. N-Body Simulations Using Customized Potential-Density Pair Basis Sets , urldate =. Monthly Notices of the Royal Astronomical Society , langid =

  23. [23]

    and Lim, Sung Hak and Putney, Eric and Shih, David , doi =

    Buckley, Matthew R. and Lim, Sung Hak and Putney, Eric and Shih, David , doi =. Measuring. Monthly Notices of the Royal Astronomical Society , langid =

  24. [24]

    Cheng, Ho Kei and Schwing, Alexander , doi =. The. arXiv e-prints , langid =

  25. [25]

    and Hilbert, D

    Courant, R. and Hilbert, D. , file =. Methods of Mathematical Physics -. New York: Interscience Publication , keywords =

  26. [26]

    and Garavito-Camargo, Nicolas and Deason, Alis J

    Cunningham, Emily C. and Garavito-Camargo, Nicolas and Deason, Alis J. and Johnston, Kathryn V. and Erkal, Denis and Laporte, Chervin F. P. and Besla, Gurtina and Luger, Rodrigo and Sanderson, Robyn E. , doi =. Quantifying the Stellar Halo’s Response to the LMC’s Infall with Spherical Harmonics , url =. The Astrophysical Journal , month = jul, number =

  27. [27]

    arXiv , author =:2201.05624 , primaryclass =

    Scientific Machine Learning through Physics-Informed Neural Networks: Where we are and What's next , url =. arXiv , author =:2201.05624 , primaryclass =

    arXiv

  28. [28]

    N-body simulations of gravitational dynamics

    doi:10.1140/epjp/i2011-11055-3 , eid =. arXiv , author =:1105.1082 , journal =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1140/epjp/i2011-11055-3

  29. [29]

    Drimmel, Ronald and Spergel, David N. , doi =. Three-Dimensional. The Astrophysical Journal , langid =

  30. [30]

    and Arendt, R

    Dwek, E. and Arendt, R. G. and Hauser, M. G. and Kelsall, T. and Lisse, C. M. and Moseley, S. H. and Silverberg, R. F. and Sodroski, T. J. and Weiland, J. L. , doi =. Morphology,. The Astrophysical Journal , langid =

  31. [31]

    Trudy Astrofizicheskogo Instituta Alma-Ata , month = jan, pages =

  32. [32]

    doi:10.1002/asna.19682910303 , journal =

    work page doi:10.1002/asna.19682910303

  33. [33]

    The Power of Depth for Feedforward Neural Networks , url =

    Eldan, Ronen and Shamir, Ohad , booktitle =. The Power of Depth for Feedforward Neural Networks , url =

  34. [34]

    arXiv , author =:2311.08358 , primaryclass =

    The Gaia-ESO Survey: 3D dynamics of young groups and clusters from GES and Gaia EDR3 , url =. arXiv , author =:2311.08358 , primaryclass =

    arXiv

  35. [35]

    L., Smith L., Evans N

    Sanders, Jason L and Smith, Leigh and Evans, N Wyn , year=. The pattern speed of the Milky Way bar from transverse velocities , volume=. Monthly Notices of the Royal Astronomical Society , publisher=. doi:10.1093/mnras/stz1827 , number=

    work page doi:10.1093/mnras/stz1827

  36. [36]

    The Gaia mission

    The Gaia mission. , keywords =. doi:10.1051/0004-6361/201629272 , archivePrefix =. 1609.04153 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1051/0004-6361/201629272

  37. [37]

    Summary of the content and survey properties

    Gaia Data Release 3 - Summary of the content and survey properties , DOI= "10.1051/0004-6361/202243940", url= "https://doi.org/10.1051/0004-6361/202243940", journal =

    work page doi:10.1051/0004-6361/202243940

  38. [38]

    and Samuel, Jenna and Cunningham, Emily C

    Garavito-Camargo, Nicolás and Price-Whelan, Adrian M. and Samuel, Jenna and Cunningham, Emily C. and Patel, Ekta and Wetzel, Andrew and Johnston, Kathryn V. and Arora, Arpit and Sanderson, Robyn E. and Garrison, Lehman and Horta, Danny , doi =. On the Corotation of Milky Way Satellites: LMC-mass Satellites Induce Apparent Motions in Outer Halo Tracers , u...

  39. [39]

    Methodus Nova Integralium Valores per Approximationem , urldate =

    Gauss, Carl Friedrich , file =. Methodus Nova Integralium Valores per Approximationem , urldate =

  40. [40]

    and Besla, Gurtina and Carpintero, Daniel D

    Gómez, Facundo A. and Besla, Gurtina and Carpintero, Daniel D. and Villalobos, Álvaro and O’Shea, Brian W. and Bell, Eric F. , doi =. AND YET IT MOVES: THE DANGERS OF ARTIFICIALLY FIXING THE MILKY WAY CENTER OF MASS IN THE PRESENCE OF A MASSIVE LARGE MAGELLANIC CLOUD , url =. The Astrophysical Journal , month = apr, number =

  41. [41]

    arXiv , author =:2205.02244 , journal =

    doi:10.3847/1538-4357/aca3a7 , eid =. arXiv , author =:2205.02244 , journal =

    work page doi:10.3847/1538-4357/aca3a7

  42. [42]

    and Massari, Davide and Veljanoski, Jovan and Brown, Anthony G

    Helmi, Amina and Babusiaux, Carine and Koppelman, Helmer H. and Massari, Davide and Veljanoski, Jovan and Brown, Anthony G. A. , doi =. The Merger That Led to the Formation of the. Nature , langid =

  43. [43]

    Hernquist, Lars and Ostriker, Jeremiah P. , doi =. A. The Astrophysical Journal , langid =

  44. [44]

    doi:10.1086/187844 , journal =

    work page doi:10.1086/187844

  45. [45]

    Ibata, R. A. and Gilmore, G. and Irwin, M. J. , doi =. A Dwarf Satellite Galaxy in. Nature , langid =

  46. [46]

    arXiv , author =:2310.00040 , primaryclass =

    Recovering the gravitational potential in a rotating frame: Deep Potential applied to a simulated barred galaxy , url =. arXiv , author =:2310.00040 , primaryclass =

    arXiv

  47. [47]

    arXiv , author =:2512.02115 , primaryclass =

    Determining the Milky Way gravitational potential without selection functions , url =. arXiv , author =:2512.02115 , primaryclass =

    Pith/arXiv arXiv

  48. [48]

    doi:10.1038/s42254-021-00314-5 , journal =

    work page doi:10.1038/s42254-021-00314-5

  49. [49]

    and Shih, David , doi =

    Lim, Sung Hak and Putney, Eric and Buckley, Matthew R. and Shih, David , doi =. Mapping Dark Matter in the. Journal of Cosmology and Astroparticle Physics , langid =

  50. [50]

    Lin, C. C. and Yuan, C. and Shu, Frank H. , doi =. On the. The Astrophysical Journal , langid =

  51. [51]

    Analytical

    Long, Kevin and Murali, Chigurupati , doi =. Analytical. The Astrophysical Journal , langid =

  52. [52]

    Measuring the Milky Way mass distribution in the presence of the LMC , url =

    Correa Magnus, Lilia and Vasiliev, Eugene , doi =. Measuring the Milky Way mass distribution in the presence of the LMC , url =. Monthly Notices of the Royal Astronomical Society , month =. https://academic.oup.com/mnras/article-pdf/511/2/2610/42533929/stab3726.pdf , issn =

  53. [53]

    Mathewson, D. S. and Cleary, M. N. and Murray, J. D. , doi =. The. The Astrophysical Journal , langid =

  54. [54]

    The mass distribution and gravitational potential of the Milky Way

    , month = feb, pages =. doi:10.1093/mnras/stw2759 , eprint =

    work page internal anchor Pith review doi:10.1093/mnras/stw2759

  55. [55]

    doi:10.1086/177955 , eprint =

    , month = nov, pages =. doi:10.1086/177955 , eprint =

    work page doi:10.1086/177955

  56. [56]

    and Nagai, R

    Miyamoto, M. and Nagai, R. , doi =. Three-. Publications of the Astronomical Society of Japan , langid =

  57. [57]

    and Frenk, Carlos S

    Navarro, Julio F. and Frenk, Carlos S. and White, Simon D. M. , doi =. A Universal Density Profile from Hierarchical Clustering , url =. The Astrophysical Journal , month = dec, number =

  58. [58]

    Charting Galactic Accelerations with Stellar Streams and Machine Learning , url =

    Nibauer, Jacob and Belokurov, Vasily and Cranmer, Miles and Goodman, Jeremy and Ho, Shirley , doi =. Charting Galactic Accelerations with Stellar Streams and Machine Learning , url =. The Astrophysical Journal , month = nov, number =

  59. [59]

    and Weinberg, Martin D

    Petersen, Michael S. and Weinberg, Martin D. and Katz, Neal , doi =. Monthly Notices of the Royal Astronomical Society , langid =

  60. [60]

    and Schutz, Katelin , doi =

    Buschmann, Malte and Safdi, Benjamin R. and Schutz, Katelin , doi =. Galactic Potential and Dark Matter Density from Angular Stellar Accelerations , url =. Phys. Rev. Lett. , month =

  61. [61]

    Pines, S. , doi =. Uniform Representation of the Gravitational Potential and its Derivatives , volume =. AIAA Journal , number =

  62. [62]

    arXiv , author =:2312.10257 , primaryclass =

    The Physics-Informed Neural Network Gravity Model: Generation III , url =. arXiv , author =:2312.10257 , primaryclass =

    arXiv

  63. [63]

    arXiv , author =:2410.13228 , primaryclass =

    From PINNs to PIKANs: Recent Advances in Physics-Informed Machine Learning , url =. arXiv , author =:2410.13228 , primaryclass =

    arXiv

  64. [64]

    Planck 2018 results. VI. Cosmological parameters

    doi:10.1051/0004-6361/201833910 , eid =. arXiv , author =:1807.06209 , journal =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1051/0004-6361/201833910

  65. [65]

    Dynamical Modelling of the Galactic Bulge and Bar: The

    Portail, Matthieu and Gerhard, Ortwin and Wegg, Christopher and Ness, Melissa , doi =. Dynamical Modelling of the Galactic Bulge and Bar: The. Monthly Notices of the Royal Astronomical Society , langid =

  66. [66]

    Monthly Notices of the Royal Astronomical Society , author =

    , month = jan, pages =. doi:10.1093/mnras/stv2383 , eprint =

    work page doi:10.1093/mnras/stv2383

  67. [67]

    Raissi and P

    M. Raissi and P. Perdikaris and G.E. Karniadakis , doi =. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations , url =. Journal of Computational Physics , keywords =

  68. [68]

    doi:10.1051/0004-6361/202348274 , journal =

    Orbital dynamics in galactic potentials under mass transfer , url =. doi:10.1051/0004-6361/202348274 , journal =

    work page doi:10.1051/0004-6361/202348274

  69. [69]

    doi:10.1086/158003 , journal =

    work page doi:10.1086/158003

  70. [70]

    Stellar accelerations and the galactic gravitational field , volume =

    Silverwood, Hamish and Easther, Richard , doi =. Stellar accelerations and the galactic gravitational field , volume =. Publications of the Astronomical Society of Australia , pages =

  71. [71]

    arXiv , author =:2206.09527 , primaryclass =

    Simultaneous approximation of a smooth function and its derivatives by deep neural networks with piecewise-polynomial activations , url =. arXiv , author =:2206.09527 , primaryclass =

    arXiv

  72. [72]

    arXiv , author =:2505.04471 , primaryclass =

    Hamiltonian Normalizing Flows as kinetic PDE solvers: application to the 1D Vlasov-Poisson Equations , url =. arXiv , author =:2505.04471 , primaryclass =

    arXiv

  73. [73]

    Starkman, Nathaniel and. Unxt:. Journal of Open Source Software , month = mar, number =. doi:10.21105/joss.07771 , file =

    work page doi:10.21105/joss.07771

  74. [74]

    Strigari , doi =

    Louis E. Strigari , doi =. Galactic searches for dark matter , url =. Physics Reports , note =

  75. [75]

    and Alves, David R

    van der Marel, Roeland P. and Alves, David R. and Hardy, Eduardo and Suntzeff, Nicholas B. , doi =. New Understanding of Large Magellanic Cloud Structure, Dynamics, and Orbit from Carbon Star Kinematics , url =. The Astronomical Journal , month = nov, number =

  76. [76]

    Width is Less Important than Depth in

    Vardi, Gal and Yehudai, Gilad and Shamir, Ohad , booktitle =. Width is Less Important than Depth in

  77. [77]

    Tango for three: Sagittarius, LMC, and the Milky Way , url =

    Vasiliev, Eugene and Belokurov, Vasily and Erkal, Denis , doi =. Tango for three: Sagittarius, LMC, and the Milky Way , url =. Monthly Notices of the Royal Astronomical Society , month = nov, number =

  78. [78]

    Bayesian Data Analysis , publisher=

    Gelman, Andrew , year=. Bayesian Data Analysis , publisher=

  79. [79]

    and Athanassoula, E

    Wang, Y. and Athanassoula, E. and Mao, S. , doi =. Basis Function Expansions for Galactic Dynamics:. Astronomy and Astrophysics , langid =

  80. [80]

    arXiv , author =:2308.08468 , primaryclass =

    An Expert's Guide to Training Physics-informed Neural Networks , url =. arXiv , author =:2308.08468 , primaryclass =

    arXiv

Showing first 80 references.