Search

Dynamic astrophysical phenomena are predominantly described by differential equations, yet our understanding of these systems is constrained by our incomplete grasp of non-linear physics and scarcity of comprehensive datasets. As such, advancing techniques in solving non-linear inverse problems becomes pivotal to addressing numerous outstanding questions in the field. In particular, modeling hot galactic winds is difficult because of unknown structure for various physical terms, and the lack of \textit{any} kinematic observational data. Additionally, the flow equations contain singularities that lead to numerical instability, making parameter sweeps non-trivial. We leverage differentiable programming, which enables neural networks to be embedded as individual terms within the governing coupled ordinary differential equations (ODEs), and show that this method can adeptly learn hidden physics. We robustly discern the structure of a mass-loading function which captures the physical effects of cloud destruction and entrainment into the hot superwind. Within a supervised learning framework, we formulate our loss function anchored on the astrophysical entropy ($K \propto P/\rho^{5/3}$). Our results demonstrate the efficacy of this approach, even in the absence of kinematic data $v$. We then apply these models to real Chandra X-Ray observations of starburst galaxy M82, providing the first systematic description of mass-loading within the superwind. This work further highlights neural ODEs as a useful discovery tool with mechanistic interpretability in non-linear inverse problems. We make our code public at this GitHub repository (https://github.com/dustindnguyen/2023_NeurIPS_NeuralODEs_M82)., Comment: 9 Pages, 2 Figures, Accepted at the NeurIPS 2023 workshop on Machine Learning and the Physical Sciences

We present the first large-scale study that demonstrates how ages can be determined for large samples of stars through Galactic chemical evolution. Previous studies found that the elemental abundances of a star correlate directly with its age and metallicity. Using this knowledge, we derive ages for 214 577 stars in GALAH DR3 using only overall metallicities and chemical abundances. Stellar ages are estimated via the machine learning algorithm XGBoost for stars belonging to the Milky Way disc with metallicities in the range -1 < [Fe/H] < 0.5, using main-sequence turn-off stars as our training set. We find that stellar ages for the bulk of GALAH DR3 are precise to 1-2 Gyr using this method. With these ages, we replicate many recent results on the age-kinematic trends of the nearby disc, including the solar neighbourhood's age-velocity dispersion relationship and the larger global velocity dispersion relations of the disc found using Gaia and GALAH. These results show that chemical abundance variations at a given birth radius are small, and that strong chemical tagging of stars directly to birth clusters may prove difficult with our current elemental abundance precision. Our results highlight the need to measure abundances for as many nucleosynthetic production sites as possible in order to estimate reliable ages from chemistry. Our methods open a new door into studies of the kinematic structure and evolution of the disc, as ages may potentially be estimated to a precision of 1-2 Gyr for a large fraction of stars in existing spectroscopic surveys.

We apply a novel statistical analysis to measurements of 16 elemental abundances in 34,410 Milky Way disk stars from the final data release (DR17) of APOGEE-2. Building on recent work, we fit median abundance ratio trends [X/Mg] versus [Mg/H] with a 2-process model, which decomposes abundance patterns into a "prompt" component tracing core-collapse supernovae and a "delayed" component tracing Type Ia supernovae. For each sample star, we fit the amplitudes of these two components, then compute the residuals Delta[X/H] from this two-parameter fit. The rms residuals range from similar to 0.01-0.03 dex for the most precisely measured APOGEE abundances to similar to 0.1 dex for Na, V, and Ce. The correlations of residuals reveal a complex underlying structure, including a correlated element group comprised of Ca, Na, Al, K, Cr, and Ce and a separate group comprised of Ni, V, Mn, and Co. Selecting stars poorly fit by the 2-process model reveals a rich variety of physical outliers and sometimes subtle measurement errors. Residual abundances allow for the comparison of populations controlled for differences in metallicity and [alpha/Fe]. Relative to the main disk (R = 3-13 kpc), we find nearly identical abundance patterns in the outer disk (R = 15-17 kpc), 0.05-0.2 dex depressions of multiple elements in LMC and Gaia Sausage/Enceladus stars, and wild deviations (0.4-1 dex) of multiple elements in omega Cen. The residual abundance analysis opens new opportunities for discovering chemically distinctive stars and stellar populations, for empirically constraining nucleosynthetic yields, and for testing chemical evolution models that include stochasticity in the production and redistribution of elements.

We report the discovery of Specter, a disrupted ultrafaint dwarf galaxy revealed by the H3 Spectroscopic Survey. We detected this structure via a pair of comoving metal-poor stars at a distance of 12.5 kpc, and further characterized it with Gaia astrometry and follow-up spectroscopy. Specter is a $25^\circ \times 1^\circ$ stream of stars that is entirely invisible until strict kinematic cuts are applied to remove the Galactic foreground. The spectroscopic members suggest a stellar age $\tau \gtrsim 12$ Gyr and a mean metallicity $\langle\text{[Fe/H]}\rangle = -1.84_{-0.18}^{+0.16}$, with a significant intrinsic metallicity dispersion $\sigma_{ \text{[Fe/H]}} = 0.37_{-0.13}^{+0.21}$. We therefore argue that Specter is the disrupted remnant of an ancient dwarf galaxy. With an integrated luminosity $M_{\text{V}} \approx -2.6$, Specter is by far the least-luminous dwarf galaxy stream known. We estimate that dozens of similar streams are lurking below the detection threshold of current search techniques, and conclude that spectroscopic surveys offer a novel means to identify extremely low surface brightness structures., Comment: 17 pages, 10 figures. Accepted to ApJ

Detecting the 21-cm hyperfine transition from neutral hydrogen in the intergalactic medium is our best probe for understanding the astrophysical processes driving the Epoch of Reionisation (EoR). The primary means for a detection of this 21-cm signal is through a statistical measurement of the spatial fluctuations using the 21-cm power spectrum (PS). However, the 21-cm signal is non-Gaussian meaning the PS, which only measures the Gaussian fluctuations, is sub-optimal for characterising all of the available information. The upcoming Square Kilometre Array (SKA) will perform a deep, 1000 hr observation over 100 deg$.^{2}$ specifically designed to recover direct images of the 21-cm signal. In this work, we use the Wavelet Scattering Transform (WST) to extract the non-Gaussian information directly from these two-dimensional images of the 21-cm signal. The key advantage of the WST is its stability with respect to statistical noise for measuring non-Gaussian information, unlike the bispectrum whose statistical noise diverges. We introduce a novel method to isolate this non-Gaussian information from mock 21-cm images and demonstrate its detection at 150 (177)~MHz ($z\sim8.5$ and $\sim7$) for a fiducial model with signal-to-noise of $\sim$5~(8) assuming perfect foreground removal and $\sim2$~(3) assuming foreground wedge avoidance., Comment: 17 pages, 10 figures and 1 table. Submitted to MNRAS, comments welcome

Stellar spectra encode detailed information about the stars. However, most machine learning approaches in stellar spectroscopy focus on supervised learning. We introduce Mendis, an unsupervised learning method, which adopts normalizing flows consisting of Neural Spline Flows and GLOW to describe the complex distribution of spectral space. A key advantage of Mendis is that we can describe the conditional distribution of spectra, conditioning on stellar parameters, to unveil the underlying structures of the spectra further. In particular, our study demonstrates that Mendis can robustly capture the pixel correlations in the spectra leading to the possibility of detecting unknown atomic transitions from stellar spectra. The probabilistic nature of Mendis also enables a rigorous determination of outliers in extensive spectroscopic surveys without the need to measure elemental abundances through existing analysis pipelines beforehand., Comment: 6 pages, 3 figures, accepted to the ICML 2022 Machine Learning for Astrophysics workshop

Modeling quasar spectra is a fundamental task in astrophysics as quasars are the tell-tale sign of cosmic evolution. We introduce a novel unsupervised learning algorithm, Quasar Factor Analysis (QFA), for recovering the intrinsic quasar continua from noisy quasar spectra. QFA assumes that the Ly$\alpha$ forest can be approximated as a Gaussian process, and the continuum can be well described as a latent factor model. We show that QFA can learn, through unsupervised learning and directly from the quasar spectra, the quasar continua and Ly$\alpha$ forest simultaneously. Compared to previous methods, QFA achieves state-of-the-art performance for quasar continuum prediction robustly but without the need for predefined training continua. In addition, the generative and probabilistic nature of QFA paves the way to understanding the evolution of black holes as well as performing out-of-distribution detection and other Bayesian downstream inferences., Comment: 6 pages, 3 figures, accepted to the ICML 2022 Machine Learning for Astrophysics workshop

We present the first high-resolution chemical abundances of seven stars in the recently discovered high-energy stream Typhon. Typhon stars have apocenters >100 kpc, making this the first detailed chemical picture of the Milky Way's very distant stellar halo. Though the sample size is limited, we find that Typhon's chemical abundances are more like a dwarf galaxy than a globular cluster, showing a metallicity dispersion and no presence of multiple stellar populations. Typhon stars display enhanced $\alpha$-element abundances and increasing r-process abundances with increasing metallicity. The high-$\alpha$ abundances suggest a short star formation duration for Typhon, but this is at odds with expectations for the distant Milky Way halo and the presence of delayed r-process enrichment. If the progenitor of Typhon is indeed a new dwarf galaxy, possible scenarios explaining this apparent contradiction include a dynamical interaction that increases Typhon's orbital energy, a burst of enhanced late-time star formation that raises [$\alpha$/Fe], and/or group preprocessing by another dwarf galaxy before infall into the Milky Way. Alternatively, Typhon could be the high-energy tail of a more massive disrupted dwarf galaxy that lost energy through dynamical friction. We cannot clearly identify a known low-energy progenitor of Typhon in the Milky Way, but 70% of high-apocenter stars in cosmological simulations are from high-energy tails of large dwarf galaxies. Typhon's surprising combination of kinematics and chemistry thus underscores the need to fully characterize the dynamical history and detailed abundances of known substructures before identifying the origin of new substructures., Comment: 12 pages, 4 figures, 2 tables, accepted to MNRAS

We introduce Astroconformer, a Transformer-based model to analyze stellar light curves from the Kepler mission. We demonstrate that Astrconformer can robustly infer the stellar surface gravity as a supervised task. Importantly, as Transformer captures long-range information in the time series, it outperforms the state-of-the-art data-driven method in the field, and the critical role of self-attention is proved through ablation experiments. Furthermore, the attention map from Astroconformer exemplifies the long-range correlation information learned by the model, leading to a more interpretable deep learning approach for asteroseismology. Besides data from Kepler, we also show that the method can generalize to sparse cadence light curves from the Rubin Observatory, paving the way for the new era of asteroseismology, harnessing information from long-cadence ground-based observations., Comment: 6 pages, 4 figures, accepted to the ICML 2022 Machine Learning for Astrophysics workshop

A key yet unresolved question in modern-day astronomy is how galaxies formed and evolved under the paradigm of the $\Lambda$CDM model. A critical limiting factor lies in the lack of robust tools to describe the merger history through a statistical model. In this work, we employ a generative graph network, E(n) Equivariant Graph Normalizing Flows Model. We demonstrate that, by treating the progenitors as a graph, our model robustly recovers their distributions, including their masses, merging redshifts and pairwise distances at redshift z=2 conditioned on their z=0 properties. The generative nature of the model enables other downstream tasks, including likelihood-free inference, detecting anomalies and identifying subtle correlations of progenitor features., Comment: 6 pages, 3 figures, accepted to the ICML 2022 Machine Learning for Astrophysics workshop

Carbon-enhanced metal-poor (CEMP) stars comprise almost a third of stars with [Fe/H] < --2, although their origins are still poorly understood. It is highly likely that one sub-class (CEMP-$s$ stars) is tied to mass-transfer events in binary stars, while another sub-class (CEMP-no stars) are enriched by the nucleosynthetic yields of the first generations of stars. Previous studies of CEMP stars have primarily concentrated on the Galactic halo, but more recently they have also been detected in the thick disk and bulge components of the Milky Way. $Gaia$ DR3 has provided an unprecedented sample of over 200 million low-resolution ($R\approx$ 50) spectra from the BP and RP photometers. Training on the CEMP catalog from the SDSS/SEGUE database, we use XGBoost to identify the largest all-sky sample of CEMP candidate stars to date. In total, we find 58,872 CEMP star candidates, with an estimated contamination rate of 12%. When comparing to literature high-resolution catalogs, we positively identify 60-68% of the CEMP stars in the data, validating our results and indicating a high completeness rate. Our final catalog of CEMP candidates spans from the inner to outer Milky Way, with distances as close as $r \sim$ 0.8 kpc from the Galactic center, and as far as $r >$ 30 kpc. Future higher-resolution spectroscopic follow-up of these candidates will provide validations of their classification and enable investigations of the frequency of CEMP-$s$ and CEMP-no stars throughout the Galaxy, to further constrain the nature of their progenitors., Comment: 19 pages, 14 figures, accepted to MNRAS

We propose to develop a wide-field and ultra-high-precision photometric survey mission, temporarily named "Earth 2.0 (ET)". This mission is designed to measure, for the first time, the occurrence rate and the orbital distributions of Earth-sized planets. ET consists of seven 30cm telescopes, to be launched to the Earth-Sun's L2 point. Six of these are transit telescopes with a field of view of 500 square degrees. Staring in the direction that encompasses the original Kepler field for four continuous years, this monitoring will return tens of thousands of transiting planets, including the elusive Earth twins orbiting solar-type stars. The seventh telescope is a 30cm microlensing telescope that will monitor an area of 4 square degrees toward the galactic bulge. This, combined with simultaneous ground-based KMTNet observations, will measure masses for hundreds of long-period and free-floating planets. Together, the transit and the microlensing telescopes will revolutionize our understandings of terrestrial planets across a large swath of orbital distances and free space. In addition, the survey data will also facilitate studies in the fields of asteroseismology, Galactic archeology, time-domain sciences, and black holes in binaries., Comment: 116 pages,79 figures

The Gaia mission recently revealed an excess population of equal-mass "twin" wide binaries, with mass ratio $q\gtrsim 0.95$, extending to separations of at least 1000 AU. The origin of this population is an enigma: twin binaries are thought to form via correlated accretion in circumbinary disks, but the typical observed protostellar disks have radii of $\sim100$ AU, far smaller than the separations of the widest twins. Here, we infer the eccentricity distribution of wide twins from the distribution of their $v$-$r$ angles, i.e., the angle between the components' separation and relative velocity vectors. We find that wide twins must be on extremely eccentric orbits. For the excess-twin population at 400-1000 AU, we infer a near-delta function excess of high-eccentricity system, with eccentricity $0.95 \lesssim e \leq 1$. These high eccentricities for wide twins imply pericenter distances of order $10$ AU and suggest that their orbits were scattered via dynamical interactions in their birth environments, consistent with a scenario in which twins are born in circumbinary disks and subsequently widened. These results further establish twin wide binaries as a distinct population and imply that wide twins can be used as a probe of the dynamical history of stellar populations., Comment: accepted for publication in The Astrophysical Journal Letter

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, which makes use of recently developed tools from the field of deep learning. We first train a normalizing flow on a sample of observed six-dimensional phase-space coordinates of stars, obtaining a smooth, differentiable approximation of the 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, which we term "Deep Potential," is more flexible than previous parametric methods, which fit restricted classes of analytic models of the distribution function and potential to the data. We demonstrate Deep Potential on mock datasets, and demonstrate its robustness under various non-ideal conditions. Deep Potential is a promising approach to mapping the density of the Milky Way and other stellar systems, using rich datasets of stellar positions and kinematics now being provided by Gaia and ground-based spectroscopic surveys., Comment: 21 pages, 13 figures. Much more detailed exposition of the method originally presented in the short conference workshop paper arXiv:2011.04673

We present a catalog of 948,216 stars with mass label and a catalog of 163,105 red clump (RC) stars with mass and age labels simultaneously. The training dataset is cross matched from the LAMOST (The Large Sky Area Multi-Object Fiber Spectroscopic Telescope) DR5 and high resolution asteroseismology data, mass and age are predicted by random forest method or convex hull algorithm. The stellar parameters with high correlation with mass and age are extracted and the test dataset shows that the median relative error of the prediction model for the mass of large sample is 3\% and meanwhile, the mass and age of red clump stars are 4\% and 7\%. We also compare the predicted age of red clump stars with the recent works and find that the final uncertainty of the RC sample could reach 18\% for age and 9\% for mass, in the meantime, final precision of the mass for large sample with different type of stars could reach 13\% without considering systematics, all these are implying that this method could be widely used in the future. Moreover, we explore the performance of different machine learning methods for our sample, including bayesian linear regression (BYS), gradient boosting decision Tree (GBDT), multilayer perceptron (MLP), multiple linear regression (MLR), random forest (RF) and support vector regression (SVR). Finally we find that the performance of nonlinear model is generally better than that of linear model, and the GBDT and RF methods are relatively better., Comment: 19 pages, 23 figures, Accepted for publication in the ApJS

The puzzle of Li-rich giant is still unsolved, contradicting the prediction of the standard stellar models. Although the exact evolutionary stages play a key role in the knowledge of Li-rich giants, a limited number of Li-rich giants have been taken with high-quality asteroseismic parameters to clearly distinguish the stellar evolutionary stages. Based on the LAMOST Data Release 7 (DR7), we applied a data-driven neural network method to derive the parameters for giant stars, which contain the largest number of Li-rich giants. The red giant stars are classified into three stages of Red Giant Branch (RGB), Primary Red Clump (PRC), and Secondary Red Clump (SRC) relying on the estimated asteroseismic parameters. In the statistical analysis of the properties (i.e. stellar mass, carbon, nitrogen, Li-rich distribution, and frequency) of Li-rich giants, we found that: (1) Most of the Li-rich RGB stars are suggested to be the descendants of Li-rich pre-RGB stars and/or the result of engulfment of planet or substellar companions; (2) The massive Li-rich SRC stars could be the natural consequence of Li depletion from the high-mass Li-rich RGB stars. (3) Internal mixing processes near the helium flash can account for the phenomenon of Li-rich on PRC that dominated the Li-rich giants. Based on the comparison of [C/N] distributions between Li-rich and normal PRC stars, the Li-enriched processes probably depend on the stellar mass., Comment: 14 pages, 8 figures, accepted by ApJ

We use chemistry ([alpha/Fe] and [Fe/H]), main sequence turnoff ages, and kinematics determined from H3 Survey spectroscopy and Gaia astrometry to identify the birth of the Galactic disk. We separate in-situ and accreted stars on the basis of angular momenta and eccentricities. The sequence of high-alpha in-situ stars persists down to at least [Fe/H]=-2.5 and shows unexpected non-monotonic behavior: with increasing metallicity the population first declines in [alpha/Fe], then increases over the range -1.3<[Fe/H]<-0.7, and then declines again at higher metallicities. The number of stars in the in-situ population rapidly increases above [Fe/H]=-1. The average kinematics of these stars are hot and independent of metallicity at [Fe/H]<-1 and then become increasingly cold and disk-like at higher metallicities. The ages of the in-situ, high-alpha stars are uniformly very old (13 Gyr) at [Fe/H]<-1.3, and span a wider range (8-12 Gyr) at higher metallicities. Interpreting the chemistry with a simple chemical evolution model suggests that the non-monotonic behavior is due to a significant increase in star formation efficiency, which began 13 Gyr ago. These results support a picture in which the first 1 Gyr of the Galaxy was characterized by a "simmering phase" in which the star formation efficiency was low and the kinematics had substantial disorder with some net rotation. The disk then underwent a dramatic transformation to a "boiling phase", in which the star formation efficiency increased substantially, the kinematics became disk-like, and the number of stars formed increased tenfold. We interpret this transformation as the birth of the Galactic disk at z~4. The physical origin of this transformation is unclear and does not seem to be reproduced in current galaxy formation models., Comment: 13 pages, 10 figures, submitted to ApJ

The Milky Way's satellite galaxies ("surviving dwarfs") have been studied for decades as unique probes of chemical evolution in the low-mass regime. Here we extend such studies to the "disrupted dwarfs", whose debris constitutes the stellar halo. We present abundances ([Fe/H], [$\alpha$/Fe]) and stellar masses for nine disrupted dwarfs with $M_{\star}\approx10^{6}-10^{9}M_{\odot}$ from the H3 Survey (Sagittarius, $Gaia$-Sausage-Enceladus, Helmi Streams, Sequoia, Wukong/LMS-1, Cetus, Thamnos, I'itoi, Orphan/Chenab). The surviving and disrupted dwarfs are chemically distinct: at fixed mass, the disrupted dwarfs are systematically metal-poor and $\alpha$-enhanced. The disrupted dwarfs define a mass-metallicity relation (MZR) with a similar slope as the $z=0$ MZR followed by the surviving dwarfs, but offset to lower metallicities by $\Delta$[Fe/H]$\approx0.3-0.4$ dex. Dwarfs with larger offsets from the $z=0$ MZR are more $\alpha$-enhanced. In simulations as well as observations, galaxies with higher $\Delta$[Fe/H] formed at higher redshifts -- exploiting this, we infer the disrupted dwarfs have typical star-formation truncation redshifts of $z_{\rm{trunc}}{\sim}1-2$. We compare the chemically inferred $z_{\rm{trunc}}$ with dynamically inferred accretion redshifts and find almost all dwarfs are quenched only after accretion. The differences between disrupted and surviving dwarfs are likely because the disrupted dwarfs assembled their mass rapidly, at higher redshifts, and within denser dark matter halos that formed closer to the Galaxy. Our results place novel archaeological constraints on low-mass galaxies inaccessible to direct high-$z$ studies: (i) the redshift evolution of the MZR along parallel tracks but offset to lower metallicities extends to $M_{\star}\approx10^{6}-10^{9}M_{\odot}$; (ii) galaxies at $z\approx2-3$ are $\alpha$-enhanced with [$\alpha$/Fe]$\approx0.4$., Comment: Submitted to ApJ. Tables 1 & 2 summarize properties of disrupted & surviving dwarfs. Fig. 2 plots the mass-metallicity and mass-$\alpha$ relations, Fig. 3 shows the correlation between $\Delta$[Fe/H] and [$\alpha$/Fe], and Fig. 4 might delight dwarf-quenching enthusiasts. Comments warmly welcomed and greatly appreciated!

Detecting the cosmic 21-cm signal during the Epoch of Reionisation and Cosmic Dawn will reveal insights into the properties of the first galaxies and advance cosmological parameter estimation. Until recently, the primary focus for astrophysical parameter inference from the 21-cm signal centred on the power spectrum (PS). However, the cosmic 21-cm signal is highly non-Gaussian rendering the PS sub-optimal for characterising the cosmic signal. In this work, we introduce a new technique to analyse the non-Gaussian information in images of the 21-cm signal called the Wavelet Scattering Transform (WST). This approach closely mirrors that of convolutional neural networks with the added advantage of not requiring tuning or training of a neural network. Instead, it compresses the 2D spatial information into a set of coefficients making it easier to interpret while also providing a robust statistical description of the non-Gaussian information contained in the cosmic 21-cm signal. First, we explore the application of the WST to mock 21-cm images to gain valuable physical insights by comparing to the known behaviour from the 21-cm PS. Then we quantitatively explore the WST applied to the 21-cm signal by extracting astrophysical parameter constraints using Fisher Matrices from a realistic 1000 hr mock observation with the Square Kilometre Array. We find that: (i) the WST applied only to 2D images can outperform the 3D spherically averaged 21-cm PS, (ii) the excision of foreground contaminated modes can degrade the constraining power by a factor of ~1.5-2 with the WST and (iii) higher cadences between the 21-cm images can further improve the constraining power., Comment: 25 pages, 16 figures and 2 tables. Accepted for publication in MNRAS