Current collaborations

Current Collaborators

(incomplete list, random ordering)

Phil Hopkins, Rachel Bezanson, Jenny Greene, Mike Boylan-Kolchin, James Bullock, Daniel Anglés-Alcázar, Chris Hayward, Sarah Wellons, Claude-André Faucher-Giguère, Dušan Kereš, Mariska Kriek, Xiangcheng Ma, Lucio Mayer, Desika Narayanan, Eliot Quataert, Sedona Price, Justin Spilker, David Setton, Joachim Stadel

Research Highlights

Predicting high resolution baryon fields from dark matter simulations with Deep Learning

Hydrodynamic simulations provide a powerful, but computationally expensive, approach to study the interplay of dark matter and baryons in cosmological structure formation. Here we introduce the EMulating Baryonic EnRichment (EMBER) Deep Learning framework to predict baryon fields based on dark-matter-only simulations thereby reducing computational cost. EMBER comprises two network architectures, U-Net and Wasserstein Generative Adversarial Networks (WGANs), to predict two-dimensional gas and HI densities from dark matter fields. We design the conditional WGANs as stochastic emulators, such that multiple target fields can be sampled from the same dark matter input. For training we combine cosmological volume and zoom-in hydrodynamical simulations from the Feedback in Realistic Environments (FIRE) project to represent a large range of scales. Our fiducial WGAN model reproduces the gas and HI power spectra within 10% accuracy down to ~10 kpc scales. Furthermore, we investigate the capability of EMBER to predict high resolution baryon fields from low resolution dark matter inputs through upsampling techniques. As a practical application, we use this methodology to emulate high-resolution HI maps for a dark matter simulation of a L = 100 Mpc /h comoving cosmological box. The gas content of dark matter haloes and the HI column density distributions predicted by EMBER agree well with results of large volume cosmological simulations and abundance matching models. Our method provides a computationally efficient, stochastic emulator for augmenting dark matter only simulations with physically consistent maps of baryon fields.

General approach of using Machine Learning to estimate the distribution of cosmic baryons
Illustration of our machine learning pipeline. We train neural networks on small cosmological volumes and zoom-in simulations with high resolution to predict baryonic counterparts from dark matter inputs. We investigate the upsampling capabilities of the networks by training individually on different darkmatter input resolutions (indicated on the left in the training figure), while the target fields are always fixed to the highest resolution (see section 4.5 for details). As indicated on the right, the trained neural networks can then be applied to large dark matter only simulations to enrich them with the specified baryon fields at low computational cost.

LEO-Py: Estimating likelihoods for correlated, censored, and uncertain data with given marginal distributions

Data with uncertain, missing, censored, and correlated values are commonplace in many research fields including astronomy. Unfortunately, such data are often treated in an ad hoc way in the astronomical literature potentially resulting in inconsistent parameter estimates. Furthermore, in a realistic setting, the variables of interest or their errors may have non-normal distributions which complicates the modeling. I present a novel approach to compute the likelihood function for such data sets. This approach employs Gaussian copulas to decouple the correlation structure of variables and their marginal distributions resulting in a flexible method to compute likelihood functions of data in the presence of measurement uncertainty, censoring, and missing data. I demonstrate its use by determining the slope and intrinsic scatter of the star forming sequence of nearby galaxies from observational data. The outlined algorithm is implemented as the flexible, easy-to-use, open-source Python package LEO-Py.

Flowchart illustrating the main use case of LEO-Py
Flowchart illustrating the use of LEO-Py. The user provides the observational data (potentially with missing and censored values) and their measurement uncertainties, as well as the parametrized distributions of the true (error-free) variables of interest. LEO-Py then calculates for each adopted set of parameters their likelihood given the observational data. The parameters with the highest likelihood can be found by coupling LEO-Py with a maximization algorithm. Alternatively, the user may explore the posterior distribution of the parameters with a Markov Chain Monte Carlo sampler.

Colors, Star formation rates, and Environments of Star forming and Quiescent Galaxies

We analyse the star formation rates (SFRs), colors, and dust extinctions of galaxies in massive (10^12.5-10^13.5 Msun) halos at z~2 in high-resolution, cosmological zoom-in simulations as part of the Feedback in Realistic Environments (FIRE) project. The simulations do not model feedback from active galactic nuclei (AGN) but reproduce well the observed relations between stellar and halo mass and between stellar mass and SFR. About half (a third) of the simulated massive galaxies (massive central galaxies) at z~2 have broad-band colors classifying them as 'quiescent', and the fraction of quiescent centrals is steeply decreasing towards higher redshift, in agreement with observations. The progenitors of z~2 quiescent central galaxies are, on average, more massive, have lower specific SFRs, and reside in more massive halos than the progenitors of similarly massive star forming centrals. The simulations further predict a morphological mix of galaxies that includes disk-dominated, irregular, and early-type galaxies. However, our simulations do not reproduce the reddest of the quiescent galaxies observed at z~2. We also do not find evidence for a color bimodality, but are limited by our modest sample size. In our simulations, the star formation activity of central galaxies of moderate mass (Mstar~10^10-10^11 Msun) is affected by a combination of two distinct physical processes. Outflows powered by stellar feedback result in a short-lived (<100 Myr), but almost complete, suppression of star formation activity after which many galaxies quickly recover and continue to form stars at normal rates. In addition, galaxies residing in slowly growing halos tend to experience a moderate reduction of their SFRs ('cosmological starvation'). The relative importance of these processes and AGN feedback is uncertain and will be explored in future work.

Specific star formation rate vs cosmic time
Illustration of how the star formation rates (SFRs) of massive, central galaxies evolve with cosmic time. This schematic view does not account for feedback from active galactic nuclei. At early times (z≥3), the specific SFRs (sSFRs; SFR per unit stellar mass) of central galaxies evolve along the star forming sequence. However, starbursts and outflows triggered by various internal and external processes can result in brief, but severe, interruptions. In addition, external drivers, such as cosmological starvation, affect the star formation histories on comparably long times scales and strongly modulate galaxy colors.

The Formation of Quiescent Galaxies at the Cosmic Noon

The cosmic noon (redshifts ~1.5–3) marked a period of vigorous star formation for most galaxies. However, about a third of the more massive galaxies at those times were quiescent in the sense that their observed stellar populations are inconsistent with rapid star formation. The reduced star formation activity is often attributed to gaseous outflows driven by feedback from supermassive black holes, but the impact of black hole feedback on galaxies in the young Universe is not yet definitively established. We analyse the origin of quiescent galaxies with the help of ultrahigh resolution, cosmological simulations that include feedback from stars but do not model the uncertain consequences of black hole feedback. We show that dark matter halos with specific accretion rates below ~0.25–0.4 Gyr^-1 preferentially host galaxies with reduced star formation rates and red broad-band colors. The fraction of such halos in large dark-matter-only simulations matches the observed fraction of moderately massive, quiescent galaxies (with stellar masses of ~10–100 billion solar masses). This suggest that halo accretion rate may be an important factor in deciding which massive galaxies at cosmic noon become quiescent. Empirical models that connect galaxy and halo evolution, such as halo occupation distribution or abundance matching models, assume a tight link between galaxy properties and the masses of their parent halos. These models will benefit from adding the specific accretion rate of halos as a second model parameter.

Halo growth vs galaxy growth
Comparison between the growth rate of baryonic masses (stars, atomic, and molecular hydrogen) of galaxies and the dark matter masses of their parent halos. Red circles and blue squares show quiescent and star-forming galaxies, respectively. The classification is based on rest-frame U−V and V−J colors. Filled and empty symbols denote galaxies that are centrals or satellites. Symbol sizes reflect stellar masses (see legend). The solid line marks a 1:1 relationship and is not a fit. Galaxies residing at the centres of fast growing halos are essentially always strongly star forming. In contrast, slowly growing (or even shrinking) halos typically harbour quiescent galaxies.

The Formation of submillimeter bright galaxies

Submillimeter-luminous galaxies at high-redshift are the most luminous, heavily star-forming galaxies in the Universe, and are characterized by prodigious emission in the far-infrared at 850 microns (S850 > 5 mJy). They reside in halos of 10 trillion solar masses, have low gas fractions compared to main sequence disks at a comparable redshift, trace complex environments, and are not easily observable at optical wavelengths. Their physical origin remains unclear. Galaxy evolution simulations have been able to form galaxies with the requisite luminosities, but have otherwise been unable to simultaneously match the stellar masses, star formation rates, gas fractions and environments. We report on a cosmological hydrodynamic galaxy formation simulation that is able to form a submillimeter galaxy which simultaneously satisfies the broad range of observed physical constraints. We find that groups of galaxies residing in massive dark matter halos have rising star formation histories that peak at collective rates ~ 500-1000 solar masses per yr at redshift 2-3, by which time the interstellar medium is sufficiently enriched with metals that the region may be observed as a submillimeter-selected system. The intense star formation rates are fueled in part by a reservoir gas supply enabled by stellar feedback at earlier times, not through major mergers. With a duty cycle of nearly a gigayear, our simulations show that the submillimeter-luminous phase of high-z galaxies is a drawn out one that is associated with significant mass buildup in early Universe proto-clusters, and that many submillimeter-luminous galaxies are actually composed of numerous unresolved components (for which there is some observational evidence).

Submillimeter bright galaxy and its environment at redshift 2
Snapshot, taken from a supercomputer simulation, depicting the distribution of gas and light in a small region around a submillimeter-luminous galaxy (SMG): it contains a bright central galaxy (white) that is accreting gas along a filamentary structure (pink), a large spiral galaxy (left of center), and numerous smaller galaxies that contribute to the total luminosity of the SMG. Ambient gas (blue-green), much of which was expelled by the galaxies at earlier epochs, gravitates towards the center of the proto-SMG. This fuels the prodigious star-formation activity of the system, which is unlike anything seen in the present-day Universe. [ from the News & Views article by R. Dave ]

Detecting dark matter substructres with GAIA

Cold Dark Matter (CDM) theory, a pillar of modern cosmology and astrophysics, predicts the existence of a large number of starless dark matter halos surrounding the Milky Way (MW). However, clear observational evidence of these "dark" substructures remains elusive. We propose a detection method of orbiting substructure that relies on the small velocity changes imposed on the stars in the MW disk. Using high-resolution numerical simulations we estimated that the new space telescope Gaia should detect the kinematic signatures of a few starless substructures provided the CDM paradigm holds. Such a measurement will provide unprecedented constraints on the primordial matter power spectrum at low-mass scales and offer a new handle onto the particle physics properties of dark matter.

Kinematic signature of a passing substructure
Kinematic signature of a low mass substructure passing vertically through the disk of the MW. Each panel shows a velocity map of the face-on stellar disk of the MW model at a different time (see legend). The galactic center (white cross) is at X = Y = 0. Panels A through E show the change in vertical velocity caused by the gravitational pull of the substructure in 500 x 500 pc^2 bins. Upward (downward) motions are shown in red (blue) colors. The blue (white) circle in each panel indicates the projected center of mass of the substructure when it is above (below) the MW disk plane. We show the position of the substructure in a frame co-rotating with the mean tangential velocity of stars at 8 kpc from the galactic center. The MW–substructure interaction results in well-localized maxima and/or minima of the vertical velocity of disk stars, visible in panels A–D.

Go to top