arXiv subjects

Yuan-Sen Ting

Publications and source records attributed to Yuan-Sen Ting.

At least 181 records · Page 10Linked to original sources

arXiv2021

Stellar labels for hot stars from low-resolution spectra - I. the HotPayne method and results for 330,000 stars from LAMOST DR6

We set out to determine stellar labels from low-resolution survey spectra of hot, OBA stars with effective temperature (Teff) higher than 7500K. This fills a gap in the scientific analysis of large spectroscopic stellar surveys such as LAMOST, which offers spectra for millions of stars at R=1800. We first explore the theoretical information content of such spectra for determining stellar labels, via the Cramér-Rao bound. We show that in the limit of perfect model spectra and observed spectra with S/N of 100, precise estimates are possible for a wide range of stellar labels: not only the effective temperature Teff, surface gravity logg, and projected rotation velocity vsini, but also the micro-turbulence velocity, Helium abundance and the elemental abundances [C/H], [N/H], [O/H], [Si/H], [S/H], and [Fe/H]. Our analysis illustrates that the temperature regime of around 9500K is challenging, as the dominant Balmer and Paschen line strength vary little with Teff. We implement the simultaneous fitting of these 11 stellar labels to LAMOST hot-star spectra using the Payne approach, drawing on Kurucz's ATLAS12/SYNTHE LTE spectra as the underlying models. We then obtain stellar parameter estimates for a sample of about 330,000 hot stars with LAMOST spectra, an increase by about two orders of magnitude in sample size. Among them, about 260,000 have good Gaia parallaxes (S/N>5), and more than 95 percent of them are luminous stars, mostly on the main sequence; the rest reflects lower luminosity evolved stars, such as hot subdwarfs and white dwarfs. We show that the fidelity of the abundance estimates is limited by the systematics of the underlying models, as they do not account for NLTE effects. Finally, we show the detailed distribution of vsini of stars with 8000-15,000K, illustrating that it extends to a sharp cut-off at the critical rotation velocity, across a wide range of temperatures.

astro-ph.SR↗

arXiv2021

The GALAH Survey: Accreted stars also inhabit the Spite Plateau

The ESA Gaia astrometric mission has enabled the remarkable discovery that a large fraction of the stars near the Solar neighbourhood appear to be debris from a single in-falling system, the so-called Gaia-Enceladus-Sausage (GSE). One exciting feature of this result is that it gives astronomers for the first time a large sample of easily observable unevolved stars that formed in an extra-Galactic environment, which can be compared to stars that formed within our Milky Way. Here we use these stars to investigate the "Spite Plateau" -- the near-constant lithium abundance observed in metal-poor dwarf stars across a wide range of metallicities (-3<[Fe/H]<-1). In particular our aim is to test whether the stars that formed in the GSE show a different Spite Plateau to other Milky Way stars that inhabit the disk and halo. Individual galaxies could have different Spite Plateaus -- e.g., the ISM could be more depleted in lithium in a lower galactic mass system due to it having a smaller reservoir of gas. We identified 76 GSE dwarf stars observed and analyzed by the GALactic Archeology with HERMES (GALAH) survey as part of its Third Data Release. Orbital actions were used to select samples of Gaia-Enceladus stars, and comparison samples of halo and disk stars. We find that the Gaia-Enceladus stars show the same lithium abundance as other likely accreted stars and in situ Milky Way stars, strongly suggesting that the "lithium problem" is not a consequence of the formation environment. This result fits within the growing consensus that the Spite Plateau, and more generally the "cosmological lithium problem" -- the observed discrepancy between the amount of lithium in warm, metal-poor dwarf stars in our Galaxy, and the amount of lithium predicted to have been produced by Big Bang Nucleosynthesis -- is the result of lithium depletion processes within stars.

astro-ph.GA↗

arXiv2021

Distant Relatives: The Chemical Homogeneity of Comoving Pairs Identified in Gaia

Comoving pairs, even at the separations of $\mathcal{O}(10^6)\,$AU, are a predicted reservoir of conatal stars. We present detailed chemical abundances of 62 stars in 31 comoving pairs with separations of $10^2 - 10^7\,$AU and 3D velocity differences $< 2 \mathrm{\ km \ s^{-1}}$. This sample includes both bound comoving pairs/wide binaries and unbound comoving pairs. Observations were taken using the MIKE spectrograph on the Magellan/Clay Telescope at high resolution ($\mathrm{R} \sim 45,000$) with a typical signal-to-noise ratio of 150 per pixel. With these spectra, we measure surface abundances for 24 elements, including Li, C, Na, Mg, Al, Si, Ca, Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Sr, Y, Zr, Ba, La, Nd, Eu. Taking iron as the representative element, our sample of wide binaries is chemically homogeneous at the level of $0.05$ dex, which agrees with prior studies on wide binaries. Importantly, even systems at separations $2\times10^5-10^7\,$AU are homogeneous to $0.09$ dex, as opposed to the random pairs which have a dispersion of $0.23\,$dex. Assuming a mixture model of the wide binaries and random pairs, we find that $73 \pm 22\%$ of the comoving pairs at separations $2\times10^5-10^7\,$AU are conatal. Our results imply that a much larger parameter space of phase space may be used to find conatal stars, to study M-dwarfs, star cluster evolution, exoplanets, chemical tagging, and beyond.

astro-ph.SR↗

arXiv2021

Stellar Streams in the Galactic Disk: Predicted Lifetimes and Their Utility in Measuring the Galactic Potential

In this paper, we present a holistic view of the detection, characterization, and origin of stellar streams in the disk of a simulated Milky Way-like galaxy. The star-by-star simulation of the Galaxy evolves stars born in clusters in a realistic galactic potential that includes spiral arms, a bar, and giant molecular clouds over $5$ Gyr. We first devise a new hybrid method to detect stellar streams that combines phase space density information along with the action-angle space spanned by stars in our simulation. We find that streams' progenitor star clusters and associations are all preferentially higher-mass ($>1000$ $M_{\odot}$) and young ($< 1$ Gyr). Our stream-finding method predicts that we might be able to find anywhere from $1$ to $10$ streams with 6D \textit{Gaia} DR2 data in the solar neighborhood alone. The simulation suggests that streams are sensitive to the initial dynamical state of clusters, accumulated energy gain from encounters with giant molecular clouds (GMCs), and present-day actions. We investigate what we can learn about the Galactic potential by studying the feasiblity of rewinding stellar streams back to their origin. Even with perfect information about the non-axisymmetric components (spiral arms, bar) of the galactic potential, the stochastic GMC population makes backwards integration impossible beyond one or two disk orbital times. Streams are also sensitive to the properties of the bar, but fairly insensitive to the properties of the non-transient two-armed spiral in our simulation. Finally we predict that around $10$ to $30$ stellar streams should be detectable with \textit{Gaia}'s 10-year end-of-mission data. There are many more stellar streams waiting to be discovered in the Galactic disk, and they could hold clues about the history of the Galaxy for the past Gyr.

astro-ph.GA↗

arXiv2021

The GALAH survey: A census of lithium-rich giant stars

We investigate the properties of 1262 red giant stars with high photospheric abundances of lithium observed by the GALAH and \Ktwo-HERMES surveys, and discuss them in the context of proposed mechanisms for lithium enrichment and re-depletion in giant stars. We confirm that Li-rich giants are rare, making up only 1.2 per cent of our giant star sample. We use stellar parameters from the third public data release from the GALAH survey and a Bayesian isochrone analysis to divide the sample into first-ascent red giant branch and red clump stars, and confirm these classifications using asteroseismic data from \Ktwo. We find that red clump stars are 2.5 times as likely to be lithium-rich as red giant branch stars, in agreement with other recent work. The probability for a star to be lithium-rich is affected by a number of factors, though the causality in those correlations is not entirely clear. We show for the first time that primary and secondary red clump stars have distinctly different lithium enrichment patterns. The data set discussed here is large and heterogeneous in terms of evolutionary phase, metallicity, rotation rate and mass. We expect that if the various mechanisms that have been proposed for lithium enrichment in evolved stars are in fact active, they should all contribute to this sample of lithium-rich giants at some level.

astro-ph.SR↗

arXiv2021

A Search for Wandering Black Holes in the Milky Way with Gaia and DECaLS

We present a search for "hyper-compact" star clusters in the Milky Way using a combination of Gaia and the Dark Energy Camera Legacy Survey (DECaLS). Such putative clusters, with sizes of ~1 pc and containing 500-5000 stars, are expected to remain bound to intermediate-mass black holes (Mbh~10^3-10^5 M-sun) that may be accreted into the Milky Way halo within dwarf satellites. Using the semi-analytic model SatGen we find an expected ~100 wandering intermediate-mass black holes with if every infalling satellite hosts a black hole. We do not find any such clusters in our search. Our upper limits rule out 100% occupancy, but do not put stringent constraints on the occupation fraction. Of course, we need stronger constraints on the properties of the putative star clusters, including their assumed sizes as well as the fraction of stars that would be compact remnants.

astro-ph.GA↗

arXiv2021

The GALAH+ Survey: Third Data Release

The ensemble of chemical element abundance measurements for stars, along with precision distances and orbit properties, provides high-dimensional data to study the evolution of the Milky Way. With this third data release of the Galactic Archaeology with HERMES (GALAH) survey, we publish 678 423 spectra for 588 571 mostly nearby stars (81.2% of stars are within <2 kpc), observed with the HERMES spectrograph at the Anglo-Australian Telescope. This release (hereafter GALAH+ DR3) includes all observations from GALAH Phase 1 (bright, main, and faint survey, 70%), K2-HERMES (17%), TESS-HERMES (5%), and a subset of ancillary observations (8%) including the bulge and >75 stellar clusters. We derive stellar parameters $T_\text{eff}$, $\log g$, [Fe/H], $v_\text{mic}$, $v_\text{broad}$ & $v_\text{rad}$ using our modified version of the spectrum synthesis code Spectroscopy Made Easy (SME) and 1D MARCS model atmospheres. We break spectroscopic degeneracies in our spectrum analysis with astrometry from $Gaia$ DR2 and photometry from 2MASS. We report abundance ratios [X/Fe] for 30 different elements (11 of which are based on non-LTE computations) covering five nucleosynthetic pathways. We describe validations for accuracy and precision, flagging of peculiar stars/measurements and recommendations for using our results. Our catalogue comprises 65% dwarfs, 34% giants, and 1% other/unclassified stars. Based on unflagged chemical composition and age, we find 62% young low-$α$, 9% young high-$α$, 27% old high-$α$, and 2% stars with $\mathrm{[Fe/H]} \leq -1$. Based on kinematics, 4% are halo stars. Several Value-Added-Catalogues, including stellar ages and dynamics, updated after $Gaia$ eDR3, accompany this release and allow chrono-chemodynamic analyses, as we showcase.

astro-ph.GA↗

arXiv2021

The GALAH Survey: No chemical evidence of an extragalactic origin for the Nyx stream

The results from the ESA Gaia astrometric mission and deep photometric surveys have revolutionized our knowledge of the Milky Way. There are many ongoing efforts to search these data for stellar substructure to find evidence of individual accretion events that built up the Milky Way and its halo. One of these newly identified features, called Nyx, was announced as an accreted stellar stream traveling in the plane of the disk. Using a combination of elemental abundances and stellar parameters from the GALAH and APOGEE surveys, we find that the abundances of the highest likelihood Nyx members are entirely consistent with membership of the thick disk, and inconsistent with a dwarf galaxy origin. We conclude that the postulated Nyx stream is most probably a high-velocity component of the Milky Way's thick disk. With the growing availability of large data sets including kinematics, stellar parameters, and detailed abundances, the probability of detecting chance associations increases, and hence new searches for substructure require confirmation across as many data dimensions as possible.

astro-ph.GA↗

arXiv2021

Reconstructing the Last Major Merger of the Milky Way with the H3 Survey

Several lines of evidence suggest the Milky Way underwent a major merger at z~2 with a galaxy known as Gaia-Sausage-Enceladus (GSE). Here we use H3 Survey data to argue that GSE entered the Galaxy on a retrograde orbit based on a population of highly retrograde stars with chemistry similar to the largely radial GSE debris. We present the first tailored, high-resolution N-body simulations of the merger. From a grid of ~500 simulations we find a GSE with $M_{*}=5\times10^{8}\ M_{\odot}, M_{\rm{DM}}=2\times10^{11} M_{\odot}$ (a 2.5:1 total mass merger) best matches the H3 data. This simulation shows the retrograde GSE stars are stripped from its outer disk early in the merger before the orbit loses significant angular momentum. Despite being selected purely on angular momenta and radial distributions, this simulation reproduces and explains the following empirical phenomena: (i) the elongated, triaxial shape of the inner halo (axis ratios $10:7.9:4.5$), whose major axis is at ~35° to the plane and connects GSE's apocenters, (ii) the Hercules-Aquila Cloud & the Virgo Overdensity, which arise due to apocenter pile-up, (iii) the 2 Gyr lag between the quenching of GSE and the truncation of the age distribution of the in-situ halo, which tracks the 2 Gyr gap between the first and final GSE pericenters. We make the following predictions: (i) the inner halo has a "double-break" density profile with breaks at both ~15-18 kpc and 30 kpc, coincident with the GSE apocenters, (ii) the outer halo has retrograde streams containing ~10% of GSE stars awaiting discovery at >30 kpc. The retrograde (radial) GSE debris originates from its outer (inner) disk -- exploiting this trend we reconstruct the stellar metallicity gradient of GSE ($-0.04\pm0.01$ dex $r_{\rm{50}}^{-1}$). These simulations imply GSE delivered ~20% of the Milky Way's present-day dark matter and ~50% of its stellar halo. (ABRIDGED)

astro-ph.GA↗

arXiv2021

The GALAH survey: tracing the Galactic disk with Open Clusters

Open clusters are unique tracers of the history of our own Galaxy's disk. According to our membership analysis based on \textit{Gaia} astrometry, out of the 226 potential clusters falling in the footprint of GALAH or APOGEE, we find that 205 have secure members that were observed by at least one of the survey. Furthermore, members of 134 clusters have high-quality spectroscopic data that we use to determine their chemical composition. We leverage this information to study the chemical distribution throughout the Galactic disk of 21 elements, from C to Eu. The radial metallicity gradient obtained from our analysis is $-$0.076$\pm$0.009 dex kpc$^{-1}$, which is in agreement with previous works based on smaller samples. Furthermore, the gradient in the [Fe/H] - guiding radius (r$_{\rm guid}$) plane is $-$0.073$\pm$0.008 dex kpc$^{-1}$. We show consistently that open clusters trace the distribution of chemical elements throughout the Galactic disk differently than field stars. In particular, at given radius, open clusters show an age-metallicity relation that has less scatter than field stars. As such scatter is often interpreted as an effect of radial migration, we suggest that these differences are due to the physical selection effect imposed by our Galaxy: clusters that would have migrated significantly also had higher chances to get destroyed. Finally, our results reveal trends in the [X/Fe]$-$r$_{\rm guid}$$-$age space, which are important to understand production rates of different elements as a function of space and time.

astro-ph.GA↗

arXiv2020

Orbital Clustering Identifies the Origins of Galactic Stellar Streams

The origins of most stellar streams in the Milky Way are unknown. With improved proper motions provided by Gaia EDR3, we show that the orbits of 23 Galactic stellar streams are highly clustered in orbital phase space. Based on their energies and angular momenta, most streams in our sample can plausibly be associated with a specific (disrupted) dwarf galaxy host that brought them into the Milky Way. For eight streams we also identify likely globular cluster progenitors (four of these associations are reported here for the first time). Some of these stream progenitors are surprisingly far apart, displaced from their tidal debris by a few to tens of degrees. We identify stellar streams that appear spatially distinct, but whose similar orbits indicate they likely originate from the same progenitor. If confirmed as physical discontinuities, they will provide strong constraints on the mass-loss from the progenitor. The nearly universal ex-situ origin of existing stellar streams makes them valuable tracers of galaxy mergers and dynamical friction within the Galactic halo. Their phase-space clustering can be leveraged to construct a precise global map of dark matter in the Milky Way, while their internal structure may hold clues to the small-scale structure of dark matter in their original host galaxies.

astro-ph.GA↗

arXiv2020

The non-monotonic, strong metallicity dependence of the wide-binary fraction

The metallicity dependence of the wide-binary fraction in stellar populations plays a critical role in resolving the open question of wide binary formation. In this paper, we investigate the metallicity ([Fe/H]) and age dependence of the wide-binary fraction (binary separations between $10^3$ and $10^4$ AU) for field F and G dwarfs within 500 pc by combining their metallicity and radial velocity measurements from LAMOST DR5 with the astrometric information from Gaia DR2. We show that the wide-binary fraction strongly depends on the metallicity: as metallicity increases, the wide-binary fraction first increases, peaks at [Fe/H]$\simeq 0$, and then decreases at the high metallicity end. The wide-binary fraction at [Fe/H]$=0$ is about two times larger than that at [Fe/H]$=-1$ and [Fe/H]$=+0.5$. This metallicity dependence is dominated by the thin-disk stars. Using stellar kinematics as a proxy of stellar age, we show that younger stars have a higher wide-binary fraction at fixed metallicity close to solar. We propose that multiple formation channels are responsible for the metallicity and age dependence. In particular, the positive metallicity correlation at [Fe/H]$<0$ and the age dependence may be due to the denser formation environments and higher-mass clusters at earlier times. The negative metallicity correlation at [Fe/H]$>0$ can be inherited from the similar metallicity dependence of close binaries, and radial migration may play a role in enhancing the wide-binary fraction around the solar metallicity.

astro-ph.SR↗

arXiv2020

Ancient Very Metal-Poor Stars Associated With the Galactic Disk in the H3 Survey

Ancient, very metal-poor stars offer a window into the earliest epochs of galaxy formation and assembly. We combine data from the H3 Spectroscopic Survey and Gaia to measure metallicities, abundances of $α$ elements, stellar ages, and orbital properties of a sample of 482 very metal-poor (VMP; [Fe/H]$<-2$) stars in order to constrain their origins. This sample is confined to $1\lesssim |Z| \lesssim3$ kpc from the Galactic plane. We find that >70% of VMP stars near the disk are on prograde orbits and this fraction increases toward lower metallicities. This result unexpected if metal-poor stars are predominantly accreted from many small systems with no preferred orientation, as such a scenario would imply a mostly isotropic distribution. Furthermore, we find there is some evidence for higher fractions of prograde orbits amongst stars with lower [$α$/Fe]. Isochrone-based ages for main sequence turn-off stars reveal that these VMP stars are uniformly old ($\approx12$ Gyr) irrespective of the $α$ abundance and metallicity, suggesting that the metal-poor population was not born from the same well-mixed gas disk. We speculate that the VMP population has a heterogeneous origin, including both in-situ formation in the ancient disk and accretion from a satellite with the same direction of rotation as the ancient disk at early times. Our precisely measured ages for these VMP stars on prograde orbits show that the Galaxy has had a relatively quiescent merging history over most of cosmic time, and implies the angular momentum alignment of the Galaxy has been in place for at least 12 Gyr.

astro-ph.GA↗

arXiv2020

591 high velocity stars in the Galactic halo selected from LAMOST DR7 and Gaia DR2

In this paper, we report 591 high velocity star candidates (HiVelSCs) selected from over 10 million spectra of the data release seven (DR7) of the Large Sky Area Multi-object Fiber Spectroscopic Telescope and the second Gaia data release, with three-dimensional velocities in the Galactic rest-frame larger than 445 km/s. We show that at least 43 HiVelSCs are unbound to the Galaxy with escape probabilities larger than 50%, and this number decreases to eight if the possible parallax zero-point error is corrected. Most of these HiVelSCs are metal-poor and slightly alpha-enhanced inner halo stars. Only 14% of them have [Fe/H] > -1, which may be the metal-rich "in situ" stars in halo formed in the initial collapse of the Milky Way or metal-rich stars formed in the disk or bulge but kinematically heated. The low ratio of 14% implies that the bulk of stellar halo was formed from the accretion and tidal disruption of satellite galaxies. In addition, HiVelSCs on the retrograde orbits have a slightly lower metallicities on average compared with these on the prograde orbits, meanwhile metal-poor HiVelSCs with [Fe/H] < -1 have an even faster mean retrograde velocity compared with metal-rich HiVelSCs. To investigate the origins of HiVelSCs, we perform orbit integrations and divide them into four types, i.e., hypervelocity stars, hyper-runaway stars, runaway stars and fast halo stars. A catalog for these 591 HiVelSCs, including radial velocities, atmospheric parameters, Gaia astrometric parameters, spatial positions, and velocities, etc., is available at http://paperdata.chinavo.org/LYB/lamostdr7_gaiadr2_hvs_591.csv.

astro-ph.GA↗

arXiv2020

Discovery of Magellanic Stellar Debris in the H3 Survey

We report the discovery of 15 stars in the H3 survey that lie, in projection, near the tip of the trailing gaseous Magellanic Stream (MS). The stars have Galactocentric velocities $< -155$ km s$^{-1}$, Galactocentric distances of $\approx 40$ to 80 kpc (increasing along the MS), and [Fe/H] consistent with that of stars in the Small Magellanic Cloud. These 15 stars comprise 94% (15 of 16) of the H3 observed stars to date that have $R_{GAL} > 37.5$ kpc, $-$350 km s$^{-1} < V_{GSR} < -155$ km s$^{-1}$, and are not associated with the Sagittarius Stream. They represent a unique portion of the Milky Way's outer halo phase space distribution function and confirm that unrelaxed structure is detectable even at radii where H3 includes only a few hundred stars. Due to their statistical excess, their close association with the MS and H I compact clouds in the same region, both in position and velocity space, and their plausible correspondence with tidal debris in a published simulation, we identify these stars as debris of past Magellanic Cloud encounters. These stars are evidence for a stellar component of the tidal debris field far from the Clouds themselves and provide unique constraints on the interaction.

astro-ph.GA↗

arXiv2020

Milky Way Tomography with the SkyMapper Southern Survey. II. Photometric Re-calibration of SMSS DR2

We apply the spectroscopy-based stellar-color regression (SCR) method to perform an accurate photometric re-calibration of the second data release from the SkyMapper Southern Survey (SMSS DR2). From comparison with a sample of over 200,000 dwarf stars with stellar atmospheric parameters taken from GALAH+ DR3 and with accurate, homogeneous photometry from $Gaia$ DR2, zero-point offsets are detected in the original photometric catalog of SMSS DR2, in particular for the gravity- and metallicity-sensitive $uv$ bands. For $uv$ bands, the zero-point offsets are close to zero at very low extinction, and then steadily increase with $E (B - V)$, reaching as large as 0.174 and 0.134 mag respectively, at $E (B - V) \sim 0.5$ mag. These offsets largely arise from the adopted dust term in the transformations used by SMSS DR2 to construct photometric calibrators from the ATLAS reference catalog. For the $gr$ bands, the zero-point offsets exhibit negligible variations with SFD $E(B - V )$, due to their tiny coefficients on the dust term in the transformation. Our study also reveals small, but significant, spatial variations of the zero-point offsets in all $uvgr$ bands. External checks using Strömgren photometry, WD loci and the SDSS Stripe 82 standard-star catalog independently confirm the zero-points found by our revised SCR method.

astro-ph.GA↗

arXiv2020

Cycle-StarNet: Bridging the gap between theory and data by leveraging large datasets

The advancements in stellar spectroscopy data acquisition have made it necessary to accomplish similar improvements in efficient data analysis techniques. Current automated methods for analyzing spectra are either (a) data-driven, which requires prior knowledge of stellar parameters and elemental abundances, or (b) based on theoretical synthetic models that are susceptible to the gap between theory and practice. In this study, we present a hybrid generative domain adaptation method that turns simulated stellar spectra into realistic spectra by applying unsupervised learning to large spectroscopic surveys. We apply our technique to the APOGEE H-band spectra at R=22,500 and the Kurucz synthetic models. As a proof of concept, two case studies are presented. The first of which is the calibration of synthetic data to become consistent with observations. To accomplish this, synthetic models are morphed into spectra that resemble observations, thereby reducing the gap between theory and observations. Fitting the observed spectra shows an improved average reduced $χ_R^2$ from 1.97 to 1.22, along with a reduced mean residual from 0.16 to -0.01 in normalized flux. The second case study is the identification of the elemental source of missing spectral lines in the synthetic modelling. A mock dataset is used to show that absorption lines can be recovered when they are absent in one of the domains. This method can be applied to other fields, which use large data sets and are currently limited by modelling accuracy. The code used in this study is made publicly available on github.

astro-ph.SR↗

arXiv2020

Deep Potential: Recovering the gravitational potential from a snapshot of phase space

One of the major goals of the field of Milky Way dynamics is to recover the gravitational potential field. Mapping the potential would allow us to determine the spatial distribution of matter - both baryonic and dark - throughout the Galaxy. We present a novel method for determining the gravitational field from a snapshot of the phase-space positions of stars, based only on minimal physical assumptions. We first train a normalizing flow on a sample of observed phase-space positions, obtaining a smooth, differentiable approximation of the phase-space distribution function. Using the collisionless Boltzmann equation, we then find the gravitational potential - represented by a feed-forward neural network - that renders this distribution function stationary. This method is far more flexible than previous parametric methods, which fit narrow classes of analytic models to the data. This is a promising approach to uncovering the density structure of the Milky Way, using rich datasets of stellar kinematics that will soon become available.

astro-ph.GA↗