Theoretical Astrophysics Santa Cruz

DON'T BE DISTRACTED...

Anthony Aguirre ...by the redwood trees, pristine beaches, and Doug Lin Physics brilliant sunshine... Astronomy & Astrophysics

Cosmology; inflation; SANTA CRUZ HOSTS HARDCORE ASTROPHYSICS. Planet and star formation; gravity; galaxy formation; accretion disks; stellar UCSC’s astrophysics group is one of the world’s best, dynamics boasting top faculty across a broad range of subjects, great Erik Asphaug Piero Madau Earth & Planetary Sciences access to instrumental and computational resources, and a Astronomy & Astrophysics top-notch graduate training program. And yes, exceptional Origin and evolution of the quality of life. Interested? Cosmology and high-energy solar system; comets and astrophysics asteroids TASC is a research unit spanning four affiliated departments.

Nic Brummell We work closely with each other and with experimentalists, Francis Nimmo Applied Math instrumentalists, and observers at the University of California Earth and Planetary Sciences Observatories, the Santa Cruz Institute for Particle Physics, the Structure and evolution of Planetary and stellar Center for Adaptive Optics, the Center for the Origin, interiors; computational rocky and icy planets magnetohydrodynamics; Dynamics, and Evolution of Planets, and the Institute for Geophysics and Planetary Physics. As part of TASC, you will Jonathan Fortney also have opportunities to access our new world-class Pleiades Joel Primack Astronomy & Astrophysics Supercomputer and to become deeply involved in science Physics using GLAST, Hubble, Keck, Spitzer, VERITAS, NuSTAR, Planetary atmosphere and Cosmology; galaxy interior physics planetary spacecraft, and other world-class instruments. formation; dark matter

TASC science tackles a wide range of problems, such as: Pascale Garaud How do stars and planets form, evolve, move, and die? Is Stefano Profumo Applied Math Earth unique? How were the elements created? How do black Physics holes impact the Universe? What is the nature of dark matter Solar and stellar interiors; High-energy and particle disk dynamics and dark energy? How do galaxies form and evolve? How did astrophysics; dark matter the Universe begin (and are there other universes)?

Gary Glatzmaier DID YOU KNOW?... Enrico Ramirez-Ruiz Earth & Planetary Sciences Cold-dark-matter galaxy formation theory began at UCSC. Astronomy & Astrophysics The hypernova model for GRBs was invented at UCSC. Computer simulations of High-energy astrophysics; stellar and planetary UCSC’s physics faculty is the nation’s most highly cited. stellar explosions; GRBs; dynamos Most known extrasolar planets were discovered with UCSC- accretion disks built equipment. Adriane Steinacker Mark Krumholz UCSC physicists are major builders and users of GLAST. Astronomy & Astrophysics Astronomy & Astrophysics UCSC pioneered giant optical telescopes like Keck, and is

Star formation; making the key technology breakthroughs for even larger Formation of planetary interstellar medium; telescopes. systems; solar system numerical methods dynamics. As a TASC graduate student, you will contribute significantly Greg Laughlin to world-class, cutting-edge discoveries while positioning Stan Woosley Astronomy & Astrophysics Astronomy & Astrophysics yourself to follow previous UCSC students who have gone on

Extrasolar planets; stellar to become Hubble Fellows, Keck Fellows, Chandra Fellows, Nuclear astrophysics; evolution; disk dynamics and junior faculty at excellent institutions. supernovae; gamma ray bursts; nucleosynthesis

HOW TO APPLY: submit all applications to the Graduate Division of the University of California at Santa Cruz (http://graddiv.ucsc.edu/admissions). Apply to the UCSC department of your choice but indicate your interest in TASC in the application. All TASC-oriented applications will be reviewed jointly. Also send a separate letter of interest to Maria Sliwinski ([email protected]), TASC graduate program advisor, 201 Interdisciplinary Sciences Bldg., UC Santa Cruz, Santa Cruz CA 95064. For further information see http://astro.ucsc.edu/tasc Theoretical Astrophysics Santa Cruz

DON'T BE DISTRACTED...

Anthony Aguirre ...by the redwood trees, pristine beaches, and Doug Lin Physics brilliant sunshine... Astronomy & Astrophysics

Comparing Observed Galaxies with Simulations

Joel R. Primack UCSC

Hubble Space Telescope Ultra Deep Field - ACS

This picture is beautiful but misleading, since it only shows about 0.5% of the cosmic density.

The other 99.5% of the universe is invisible. Matter and Energy Content of the Universe

Imagine that the entire universe is an ocean of dark energy. On that ocean sail billions of ghostly ships made of dark matter... Matter and Energy Content of the Dark Universe Matter Ships

on a ΛCDM

Dark Double Energy Dark Ocean Imagine that the entire universe is an ocean of dark Theory energy. On that ocean sail billions of ghostly ships made of dark matter... Matter Distribution Agrees with Double Dark Theory! Planck Collaboration: Cosmological parameters

PlanckPlanck Collaboration: Collaboration: The ThePlanckPlanckmissionmission Planck Collaboration: The Planck mission 6000 6000 Angular scale 5000 90 500018 1 0.2 0.1 0.07 Planck Collaboration: The Planck mission

European 6000 ] 4000 2

] 4000 2 Double µ K [ µ K 3000 [ 3000Dark TT Space TT 5000 D

D 2000 2000Theory

Agency 1000 1000

] 4000 2 Cosmic 0 0 PLANCK 600 Variance600 60

µ K 60

[ 3000 300300 30 TT 30 TT 0 0 Satellite D 0 0 D D Fig. 7. Maximum posterior CMB intensity map at 50 resolution derived from the joint baseline-30 analysis of Planck, WMAP, and

-300 -300 408 MHz observations. A small strip of the Galactic plane, 1.6 % of the sky, is filled in by-30 a constrained realization that has the same statistical properties as the rest of the sky. 2000 -600 -60-60 Data -600 Temperature-Temperature 22 1010 3030 500500 10001000 15001500 20002000 25002500 Released 1000The Planck 2015 temperature power spectrum. At multipoles ` 30 we show the maximum likelihood frequency averaged Fig.Fig.Fig. 9. 9. 10.ThePlanckPlanck TT2015power temperature spectrum. power The points spectrum. in the At upper multipoles panel show` 30 the we maximum-likelihood show the maximum estimateslikelihood of frequency the primary averaged CMB temperature spectrum computed from the Plik cross-half-mission likelihood with foreground and other nuisance parameters deter- temperaturespectrum computed spectrum{ as computed described from in the the textPlik forcross-half-mission the best-fit foreground likelihood and nuisance with foreground parameters and of other the Planck nuisance+WP parameters+highL fit deter- listed February 9, mined from the MCMC analysis of the base ⇤CDM cosmology. In the multipole range 2 ` 29, we plot the power spectrum minedin Table from5. The the red MCMC line shows analysis the of best-fit the base base⇤⇤CDMCDM cosmology. spectrum. The In the lower multipole panel shows range the 2 residuals` 29, with we plot respect the powerto the theoretical spectrum 2015 estimatesestimates from the Commander component-separationcomponent-separation algorithm algorithm computed computed over over 94 94 % % of of the the sky. sky. The The best-fit best-fit base base⇤CDM⇤CDM theoretical theoretical spectrummodel. The fitted0 error to barsthe Planck are computedTT+lowP from likelihood the full covariance is plotted in matrix, the upper appropriately panel. Residuals weighted with across respect each to band this (see model Eqs. are36a shownand in spectrum fitted to the Planck TT+lowP likelihood is plotted in the upperFig. 8. Maximum panel. posterior Residuals amplitude Stokes Q ( withleft) and U ( respectright) maps derived to from thisPlanck observations model between are 30 and shown 353 GHz. in 36b), and include2 beam uncertainties10 and50 uncertainties500 in the foregroundThese1000 model mapS have been parameters. highpass-filtered with1500 a cosine-apodized filter between ` = 202000 and 40, and the a 17 % region of the Galactic2500 thethe lower lower panel. The error error bars bars show show 11 uncertainties.uncertainties. From FromPlanckPlanck Collaborationplane Collaboration has been replaced with a XIII constrained XIII( Gaussian2015(2015 realization). ). (Planck Collaboration IX 2015). From Planck Collaboration X Agrees with Double Dark Theory!±± Multipole moment,(2015). viewed as work in progress. Nonetheless, we find a high level of 8.2.2. Number of modes consistency in results between the TT and the full TT+TE+EE likelihoods. Furthermore, the cosmological parameters (which One way of assessing the constraining power contained in a par- 100 ticular measurement of CMB anisotropies is to determine the 140 do not depend strongly on ⌧) derived from the TE spectra have comparable errors to the TT-derived parameters, and they are e↵ective number of a`m modes that have been measured. This

The temperature angular power spectrum of the primary CMB from Planck] , showing a precise measurement of seven acoustic peaks,/ that Fig. 19. consistent to within typically 0.5 or better. Polarization-is equivalent to estimatingPolarization 2 times the square of the total S N

Temperature-Polarization 2 are well ﬁt by a simple six-parameter ⇤CDM theoretical model (the model plotted80 is the one labelled [Planck+WPin the power+highL] spectra, a measure in Planck that contains all Collaborationthe available ] 70 2 XVI (2013)). The shaded area around the best-ﬁt curve represents cosmic variance,µ K 16 including the sky cut used. The error bars on individual points 60

also include cosmic variance. The horizontal axis is logarithmic up to ` = 50, 5 and linear beyond. The vertical scale is `(` + 1)Cl/2⇡. The measured [ µ K 0

spectrum shown here is exactly the same as the one shown in Fig. 1 of Planck[10 Collaboration40 XVI (2013), but it has been rebinned to show better TE the low-` ` region. D EE -70 ` 20 Double Dark Theory C 0 Double Dark Theory -140 Angular scale tected by Planck over the entire sky, and which therefore con- 9010 18 1 0.2 0.1 tains both4 Galactic and extragalactic objects. No polarization in- EE `

6000TE ` 0 formation0 is provided for the sources at this time. The PCCS C 5000 D -10 di↵ers-4 from the ERCSC in its extraction philosophy: more e↵ort has been made on the completeness of the catalogue, without re- 30 500 1000 1500 2000 30 500 1000 1500 2000

] 4000 ducing notably the reliability of the detected sources, whereas 2 ` the ERCSC was built in the spirit of` releasing a reliable catalog µ K

[ 3000

Fig. 10. Frequency-averaged TE (left) and EE (right) spectra (withoutsuitable fitting for quick for T– follow-upP leakage). (in The particular theoretical withTE theand short-livedEE spectra Fig.Fig. 10. 11.Frequency-averagedPlanck T E (left) andTEEE(leftspectra) and (right)EE (right computed) spectra as (withoutdescribed fitting in the for text.T– TheP leakage). red lines The show theoretical the polarizationTE and spectraEE spectra from plottedD in the upper panel of each plot are computed from the best-fitHerschel modeltelescope). of Fig. 9. Residuals The greater with amount respect of to data, this theoretical di↵erent selec- model plottedthe2000 base in⇤ theCDM upperPlanck panel+ ofWP each+highL plot aremodel, computedwhich from is fitted the to best-fit the TT model data only of Fig.. 9. Residuals with respect to this theoretical model areare shown shown in the lower panel panel in in each each plot. plot. The The error error bars bars show show 1tion1 errors. processerrors. The The and green thegreen improvementslines lines in inthe the lower lower in thepanels panels calibration show show the and the best-fit map-best-fit temperature-to-polarization leakage model, fitted separately to the±± TE and EE spectra. From Planck Collaboration XIII (2015). temperature-to-polarization1000 leakage model, fitted separately to the makingTE and EE processingspectra. (references)From Planck help Collaboration the PCCS XIII to( improve2015). the performance (in depth and numbers) with respect to the previ- 0 2 10 50 500 1000 1500 2000 ous ERCSC. 4 cosmologicalcosmological informationMultipole if if we we assume assumemoment, that that the the anisotropies anisotropies are are andandThe 96 96 000 sources 000 modes, modes, were respectively. respectively. extracted4 fromFromFrom the this 2013 this perspective perspectivePlanck thefrequency the2015 2015 purelypurely Gaussian Gaussian (and hence hence ignore ignore all all non-Gaussian non-Gaussian informa- informa- mapsPlanckPlanck (Sect.datadata 6), constrain constrain which approximatelyinclude approximately data acquired 55 %55 more % over more modes more modes than than intwo in Fig.tiontion 20. coming comingThe temperature from lensing, angular the the power CIB, CIB, cross-correlations cross-correlationsspectrum of the CMB, with with other esti- other skythethe coverages.2013 2013 release. release. This Of Of implies course course this that this is the not is flux not the thedensities whole whole story, of story, most since since of matedprobes,probes, from etc.). etc.). the SMICA CarryingPlanck out outmap. this this procedure The procedure model for plotted for the the isPlanckPlanck the one20132013 la- somesome pieces pieces of of information information are are more more valuable valuable than than others, others, and and TT power spectrum data provided in Planck Collaboration XV thein sources fact Planck are anis able average to place of three considerably or more tighter di↵erent constraints observa- on belledTT [Planckpower+ spectrumWP+highL] data in provided Planck Collaboration in Planck Collaboration XVI (2013). The XV tionsin fact overPlanck a periodis able of to 15.5 place months. considerably The tighterMexican constraints Hat Wavelet on shaded((20142014 area)) and and aroundPlanck the Collaboration best-fit curve XVI represents XVI((20142014 cosmic),), yields yields variance, the the number number in- particularparticular parameters parameters (e.g., (e.g., reionization reionization optical optical depth depth or certain or certain 24 algorithm (Lopez-Caniego´ et al. 2006) has been selected as the cluding826826 000 000 the sky (which (which cut used. includes The error the the e ebars↵↵ectsects on of ofindividual instrumental instrumental points noise, do noise, not cos- in- cosmic variance and masking). The 2015 TT data have increased baseline4 method for the production of the PCCS. However, one cludemic cosmic variance variance. and masking). The horizontal The axis 2015 is logarithmicTT data have up to increased` = 50, 4HereHere we we have have used used the the basic basic (and (and conservative) conservative) likelihood; likelihood; more more this value to 1 114 000, with TE and EE adding a further 60 000 additional methods, MTXF (Gonzalez-Nuevo´ et al. 2006) was andthis linear value beyond. to 1 114 The 000, vertical with scaleTE and is ÈE(` +adding1)Cl/2 a⇡ further. The binning 60 000 modesmodes are are e↵ eectively↵ectively probed probed by byPlanckPlanckif oneif one includes includes larger larger sky skyfrac- frac- scheme is the same as in Fig. 19. implementedtions.tions. in order to support the validation and characteriza- tion of the PCCS.

8.1.1. Main catalogue The source selection for the PCCS is made on the basis17 of17 Signal-to-Noise Ratio (SNR). However, the properties of the The Planck Catalogue of Compact Sources (PCCS, Planck background in the Planck maps vary substantially depending on Collaboration XXVIII (2013)) is a list of compact sources de- frequency and part of the sky. Up to 217 GHz, the CMB is the

27 Aquarius Simulation Volker Springel

Milky Way 100,000 Light Years

Milky Way Dark Matter Halo 1,500,000 Light Years 9

Bolshoi Cosmological Simulation Anatoly Klypin, Sebastian Trujillo-Gomez, Joel Primack ApJ 2011 Pleiades Supercomputer, NASA Ames Research Center 8.6x109 particles 1 kpc resolution

1 Billion Light Years Bolshoi Cosmological Simulation

100 Million Light Years

1 Billion Light Years How the Halo of the Big Cluster Formed

100 Million Light Years Bolshoi-Planck Cosmological Simulation Merger Tree of a Large Halo Structure Formation Methodology • Starting from the Big Bang, we simulate the evolution of a representative part of the universe according to the Double Dark theory to see if the end result matches what astronomers actually observe. Structure Formation Methodology • Starting from the Big Bang, we simulate the evolution of a representative part of the universe according to the Double Dark theory to see if the end result matches what astronomers actually observe. • On the large scale the simulations produce a universe just like the one we live in. We’re always looking for new phenomena to predict — every one of which tests the whole theory! Properties of Dark Matter Haloes: Local Environment Density Christoph T. Lee, Joel R. Primack, Peter Behroozi, Aldo Rodríguez-Puebla, Doug Hellinger, Avishai Dekel MNRAS 2017 8 Low Mass Intermediate Mass High Mass log10 Mvir/MC = 1.5 0.375 0.75 0 0.75 log M = 11 .20± 0.375 11.95 12.70 13.45 10 vir ± 1 Concentration =1/2[h Mpc] 1 4 16 30 95% CI 2 8

NFW 20 C

4.0 Spin Parameter ]

2 3.5

[10 3.0 B 2.5 ] 1

yr 2 11

[10 -2 Mass Accretion /M -6

dyn z=0

⌧ Rate ˙ M 0 1 2 0 1 2 0 1 2 0 1 2 log10 ⇢/⇢avg log10 ⇢/⇢avg log10 ⇢/⇢avg log10 ⇢/⇢avg

Figure 5. Medians of scatter in ⇢ C , ⇢ , and ⇢ M˙ /M relationships at z = 0, where ⇢ is the local environment density smoothed on dierent NFW B scales and ⇢avg is the average density of the simulation. Dierent coloured lines represent dierent smoothing scales. The shaded grey ﬁlled curve represents the 95% conﬁdence interval on the median, shown only for the characteristic smoothing length s,char = 1, 2, 4, and 8Mpc/h for mass bins from left to right, 12.7 respectively, and provides an indication of sample size at dierent densities. Mass bins are selected relative to the non-linear mass (log10 MC = 10 M at z = 0) to facilitate comparison between halos above, at, or below MC. We see that lower mass halos occupy regions with a wide range of local densities, while higher mass halos are restricted to higher density regions. Note also that larger smoothing scales will shift the range of densities towards the average density, so equal smoothing lengths should be used to compare density ranges for halos of dierent masses. See Fig 6 for a discussion of the trends seen in this plot.

los in low density regions (bottom 20% of densities) accrete more halo properties. The dark grey and light grey shading reﬂect the rapidly than halos in higher density regions. 95% conﬁdence interval on the median and the 20 80th percentile range of the halo property at a given redshift, respectively. These are shown only for halos in median density regions at z = 0, though 5.2 Redshift evolution of halo properties at dierent densities similar trends apply to halos in low and high density regions at z = 0. One of the principal analysis methods we’ve used to investigate the origins of the trends in Fig. 6 is to examine the median evolution of In order to minimize bias introduced by the longest lasting halo properties along the most massive progenitor branch (MMPB) MMPBs (those that remain above Mmin out to higher than average of halos in regions of dierent density at z = 0. In Figs. 7 and 8, for a redshifts), we implement a "median preserving" approach. Tracing given mass bin, we’ve selected all halos in the 0 10th, 45 55th, and time backwards from z = 0, when a given MMPB drops below 90 100th percentile ranges of characteristic local density M , we determine the halo property rank of that MMPB’s earliest s,char min at z = 0 to represent halos in low, median, and high density regions, eligible progenitor Pearliest (with Mvir > Mmin) with respect to all respectively. Using the halo merger trees, we follow the MMPB of other eligible halos at that time step. Then, in addition to eliminat- each halo and record the properties of each progenitor. We then ing further progenitors of Pearliest, we eliminate progenitors of the present the median halo properties of the most massive progenitors MMPB with rank R = N R, where N is the total number of halo 0 of halos that end up in these low, median, and high density regions progenitors in consideration at the relevant redshift and R is the at z = 0. Note that because the density selections are made at z = 0, rank of Pearliest. For example, if the earliest eligible progenitor Pi the progenitors of those halos are not guaranteed to reside in similar of a given MMPB ranks in the 67th percentile in CNFW compared density regions at higher redshifts. Once an MMPB mass drops to all other eligible progenitors at that redshift, then in addition to 10 below the completeness threshold Mmin = 10 M , we discard eliminating the remaining progenitors of Pi , we eliminate any re- any remaining progenitors from the analysis. This is done in order maining progenitors of the MMPB that ranks in the 100 67 = 33rd to exclude halos with low particle counts that may have unreliable percentile in CNFW at that same redshift. This procedure is applied

© 2016 RAS, MNRAS 000, 1–18 Structure Formation Methodology • Starting from the Big Bang, we simulate the evolution of a representative part of the universe according to the Double Dark theory to see if the end result matches what astronomers actually observe. • On the large scale the simulations produce a universe just like the one we live in. We’re always looking for new phenomena to predict — every one of which tests the theory! • But the way individual galaxies form is only partly understood because it depends on the interactions of the ordinary atomic matter, as well as the dark matter and dark energy, to form stars and black holes. We need help from observations. face-on edge-on

Redshift = Cosmology and Astrophysics Joel Primack

➢ Bolshoi - best cosmological simulations using the latest cosmological parameters. ➢ Largest suite of high-resolution zoom-in hydrodynamic galaxy simulations compared with observations by CANDELS, the largest-ever Hubble Space Telescope project. ➢ Dust absorption and re-radiation of starlight in simulated galaxies using my group’s Sunrise code used to make realistic images from our simulations. ➢ New methods for comparison of simulated galaxies with observations, including Deep Learning methods. Explain observed galaxy clumps, compaction, elongation. ➢ Co-leading with Piero Madau the Assembling Galaxies of Resolved Anatomy (AGORA) international collaboration to run and compare high-resolution galaxy simulations. • 3 Aspects of Star-Forming Galaxies Seen in CANDELS – Compaction – Elongation Challenge for Observers & Simulators! – Clumps } Astronaut Andrew Feustel installing Wide Field Camera Three on Hubble Space Telescope

candels.ucolick.org

WFC3 ACS

(blue 0.4 μm)(1+z) = 1.6 μm @ z = 3 (red 0.7 μm)(1+z) = 1.6 μm @ z = 2.3 Cosmic Horizon (The Big Bang) Cosmic Background Radiation

Cosmic Dark Ages

Bright Galaxies Form Big Galaxies Form

Earth Forms Today

When we look Cosmic out in space we look back Spheres in time… of Time Redshift COSMIC WEB This frame from the Bolshoi supercom- 69 4 3 2 10.50.20 )

puter simulation depicts the distribution of matter at 2 – redshift 3. Clusters of galaxies lie along the bright filaments. CANDELS Dark matter and cold gas flow along the filaments to supply × 10 3

galaxies with the material they need to form stars. 10 / CANDELS, ET AL. )

Cosmic 5

Cosmic dawn Cosmic high noon Star-formation rate UNIVERSITY OF NOTTINGHAMUNIVERSITY OF (

(solar masses/year/megaparsecs 0 BER

Ö 012345 6 7 8 9 10 11 12 13 Time since Big Bang (billions of years) Big Now Bang 13.7 STARBIRTH RATE Using data from many surveys, including CANDELS, astronomers have plotted the rate of star formation through cosmic history.

The rate climbed rapidly at cosmic dawn and peaked at cosmic high noon. GREGG DINDERMAN; SOURCE: KENNETH DUNCAN ANATOLY KLYPIN / STEFAN GOTTL STEFAN / KLYPIN ANATOLY S&T:

At the present day, only a few galaxies lie between the present-day red ellipticals with old stars grew in size by peaks of the blue and red galaxies, in the so-called “green “dry” mergers — mergers between galaxies having older valley” (so named because green wavelengths are midway red stars but precious little star-forming cold gas. But between red and blue in the spectrum). A blue galaxy that the jury is still out on whether this mechanism works in is vigorously forming stars will become green within a detail to explain the observations. few hundred million years if star formation is suddenly quenched. On the other hand, a galaxy that has lots of old The Case of the Chaotic Blue Galaxies stars and a few young ones can also be green just through Ever since Hubble’s ﬁrst spectacular images of distant the combination of the blue colors of its young stars and galaxies, an enduring puzzle has been why early star- the red colors of the old ones. The Milky Way probably forming galaxies look much more irregular and jumbled falls in this latter category, but the many elliptical galaxies than nearby blue galaxies. Nearby blue galaxies are around us today probably made the transition from blue relatively smooth. The most beautiful ones are elegant to red via a rapid quenching of star formation. CANDELS “grand-design” spirals with lanes of stars and gas, such as lets us look back at this history. M51. Smaller, irregular dwarf galaxies are also often blue. Most galaxies of interest to astronomers working on But at cosmic high noon, when stars were blazing CANDELS have a look-back time of at least 10 billion into existence at peak rates, many galaxies look distorted years, when the universe was only a few billion years old. or misshapen, as if galaxies of similar size are colliding. Because the most distant galaxies were relatively young at Even the calmer-looking galaxies are often clumpy and the time we observe them, we thought few of them would irregular. Instead of having smooth disks or spiral arms, have shut oﬀ star formation. So we expected that red gal- early galaxies are dotted with bright blue clumps of very axies would be rare in the early universe. But an impor- active star formation. Some of these clumps are over 100 tant surprise from CANDELS is that red galaxies with the times more luminous than the Tarantula Nebula in the same elliptical shapes as nearby red galaxies were already Large Magellanic Cloud, one of the biggest star-forming common only 3 billion years after the Big Bang — right regions in the nearby universe. How did the chaotic, dis- in the middle of cosmic high noon. ordered galaxies from earlier epochs evolve to become the Puzzlingly, however, elliptical galaxies from only familiar present-day spiral and elliptical galaxies? about 3 billion years after the Big Bang are only one- Because early galaxies appear highly distorted, astro- third the size of typical elliptical galaxies with the same physicists had hypothesized that major mergers — that is, stellar mass today. Clearly, elliptical galaxies in the early collisions of galaxies of roughly equal mass — played an universe must have subsequently grown in a way that important role in the evolution of many galaxies. Merg- increased their sizes without greatly increasing the num- ers can redistribute the stars, turning two disk galaxies ber of stars or redistributing the stars in a way that would into a single elliptical galaxy. A merger can also drive gas change their shapes. Many astronomers suspect that the toward a galaxy’s center, where it can funnel into a black

SkyandTelescope.com June 2014 21 Galaxy Hydro Simulations: 2 Approaches 1. Low resolution (~ kpc) Advantages: it’s possible to simulate many galaxies and study galaxy populations and their interactions with CGM & IGM. Disadvantages: since feedback &winds are “tuned,” we learn little about how galaxies themselves evolve, and cannot compare in detail with high-z galaxy images and spectra. Examples: Overwhelmingly Large Simulations (OWLs, EAGLE), AREPO simulations in 100 Mpc box (Illustris). 2. High resolution (~10s of pc) THIS TALK Advantages: it’s possible to compare in detail with high-z galaxy images and spectra, to discover how galaxies evolve, morphological drivers (e.g., galaxy shapes, clumps and other instabilities, origins of galactic spheroids, quenching). Radiative pressure & AGN feedbacks essential? Disadvantages: statistical galaxy samples take too much computer time; can we model galaxy population evolution using simulation insights in semi-analytic models (SAMs)? Examples: ART/VELA and FIRE simulation suites, AGORA simulation comparison project. • 3 Aspects of Star-Forming Galaxies Seen in CANDELS – Compaction – Elongation Challenge for Observers & Simulators! – Clumps } Our hydroART cosmological zoom-in simulations produce all of these phenomena! The FastInferred Track of GalaxyEvolution Evolution

Quenched Nuggets

diffuse SFG pre-compaction SF Nuggets

compactness 2334 A. Zolotov et al.

11 -1.5 DM stars 4 V03 cQ cQ V03 gas 10 exsitu -1.0 3 ]

-1 9

-0.5 z = 1.00 z = 1.00 ] • O 2 8 0.0

z = 2 z = 2 log M [M 1

z = 3 z = 3 7 log Mdot [M/yr] log sSFR [Gyr 0.5 0 6 SFR -1 1.0 dSF cSF dSF cSF 1 kpc inflow 10 kpc 5 outflow 7 8 9 10 7 8 9 10 6 5 4 3 2 1 5 4 3 2 1 2 2 z z log e [M O• /kpc ] log 1 [M O• /kpc ]

11 Downloaded from -1.5 DM stars 4 V11 cQ cQ V11 gas z = 1.17 z = 1.17 10 exsitu -1.0 3 ]

-1 9

-0.5 ] • O 2

8 http://mnras.oxfordjournals.org/ 0.0 z = 3 z = 3

log M [M 1 z = 4 z = 4 z = 2 z = 2 7 log Mdot [M/yr] log sSFR [Gyr z = 5 0.5 z = 5 0 6 SFR -1 1.0 dSF cSF dSF cSF 1 kpc inflow 10 kpc 5 outflow 7 8 9 10 7 8 9 10 6 5 4 Compaction3 2 1 and5 quenching4 3 2 23331 log [M /kpc2] log [M /kpc2] z z Ceverino+ RP simulations e O• FAST-TRACK 1 O• 11 -1.5 cQ cQ DM analyzed by Zolotov, Dekel, VELA07-RPV07 11 V07 stars 4 -1.5 z = 1.00 z = 1.00 gasDM exsitustars 4 Tweed, Mandelker, Ceverino, V24 cQ cQ 10 V24 gas

-1.0 z = 1.08 z = 1.08 exsitu at University of California, Santa Cruz on May 28, 2015 10 3 & Primack MNRAS 2015 -1.0 ] 3 -1 9

] z = 2 z = 2 /yr] ] -1 • -0.5 O • O 9 2

-0.5 ] • Barro+ (CANDELS) 2013 O 8 2 z = 2 z = 2 0.0 8

log M [M 1 0.0 log Mdot [M log M [M 7 1 log sSFR [Gyr z = 5 z = 3 z = 5 z = 3 z = 4 z = 4 7 log Mdot [M/yr] log sSFR [Gyr 0.5 z = 3 z = 3 0 0.5 6 0 6 SFR -1 1.0 dSF cSF dSF cSF 1 kpc inflow 10 kpc 5 outflowSFR -1 1.0 dSF cSF dSF cSF 1 kpc inflow 10 kpc 7 8 9 10 7 8 9 10 56 5 4 3 2outflow 1 5 4 3 2 1 2 2 z z 7 8 log e [M9 O• /kpc ] 10 7 8log 1 [M O9• /kpc ] 10 6 5 4 3 2 1 5 4 3 2 1 log [M /kpc2] log [M /kpc2] z z e O• SLOW-TRACK 1 O• 11 -1.5 DM stars

VELA27-RP 4 Downloaded from V27 cQ cQ 11 V27 gas -1.5 z = 1.00 cQ z = 1.00 cQ DMexsitu 10 stars 4 -1.0 V12 V12 gas z = 1.27 z = 1.27 exsitu 10 3 ] -1.0 -1 z = 2 z = 2 9 3 ]

] -0.5 • O 2 -1 z = 4 z = 4 9 /yr] ]

-0.5 8 • O • O 2 COMPACTION —> 0.0 z = 2 z = 2 z = 3 z = 3 log M [M 8 1 http://mnras.oxfordjournals.org/ 7 log Mdot [M/yr] log sSFR [Gyr 0.0

z = 3 z = 3 log M [M 1

0.5 7 0 log Mdot [M • major merger log sSFR [Gyr 0.5 z = 5 z = 5 z = 5z = 5 6 0 • minor merger SFR -1 1.0 dSF z = 4 cSF dSF z = 4 cSF 6 1 kpc inflow 10 kpc 5 outflow SFR -1 1.0 dSF cSF dSF cSF 6 1 kpc 5 4 3 inflow2 1 10 kpc 5 4 3 2 1 7 8 9 10 7 8 9 10 outflow log [M /kpc2] log [M /kpc2] 5 z z 7 8 e 9 O• 10 7 8 1 9 O• 10 6 5 4 3 2 1 5 4 3 2 1 2 Zolotov+20152 z z log e [M O• /kpc ] log 1 [M O• /kpc ] Figure 3. Same as Fig. 2, but for four galaxies of lower masses. The dashed vertical lines mark the onset and peak of earlier compaction events. These galaxies

compactify to lower central densities and make more than one quenching attempt.11 -1.5 cQ cQ DM stars 4 V14 V14 gas exsitu 10 at University of California, Santa Cruz on May 28, 2015 -1.0 z = 1.43 z = 1.43 3

mass,] stellar mass, and dark matter mass. Note that the stellar mass are measured in spherical shells of radii r 1 and 10 kpc and of -1 9 /yr] within the inner 1 kpc is a proxy for the surface density in the central ] width "r 0.1r via O -0.5 • • O 2 1-kpc region of the galaxy, !1. The mass rates of change shown are 8 1 z = 3 z = 3 ˙ 0.0 M mi vri 1(2) the SFR, the gas inﬂow rate, and the gas outﬂow rate. These rates log M [M "r 7 i log Mdot [M log sSFR [Gyr z = 4 z = 2 z = 4 z = 2 ! 0.5 0 z = 5 z = 5 6 MNRAS 450, 2327–2353 (2015) SFR -1 1.0 dSF cSF dSF cSF 1 kpc inflow 10 kpc 5 outflow 7 8 9 10 7 8 9 10 6 5 4 3 2 1 5 4 3 2 1 2 2 z z log e [M O• /kpc ] log 1 [M O• /kpc ]

11 -1.5 cQ cQ DM z = 1.00 z = 1.00 stars 4 V26 V26 gas 10 exsitu -1.0 3 ]

-1 9 /yr] ]

-0.5 • O • O 2 z = 2 z = 2 8 0.0

log M [M 1

7 log Mdot [M log sSFR [Gyr z = 4 z = 4 0.5 z = 5 z = 3 z = 5 z = 3 0 6 SFR -1 1.0 dSF cSF dSF cSF 1 kpc inflow 10 kpc 5 outflow 7 8 9 10 7 8 9 10 6 5 4 3 2 1 5 4 3 2 1 2 2 z z log e [M O• /kpc ] log 1 [M O• /kpc ]

Figure 2. Evolution of four galaxies of relatively high stellar masses that compactify at relatively high redshift to a high central surface density and quench efficiently. Two left-hand panels: evolution tracks in sSFR and compactness as measured by !e (left) and !1 (second from left). The redshifts from z 5 to z 1 are marked along the tracks by red symbols. Major mergers are marked by open blue upside-down triangles, and minor mergers by open purple squares. Two right-hand panels: evolution of mass and its rate of change inside a central sphere of radius 1 kpc (second from right) and 10 kpc (right). Shown at the top (scale along the left-hand axis) are the masses in gas (blue), stars (red), and dark matter (black). Also shown is the mass in ex-situ stars, as a merger indicator (green). Shown at the bottom (scale along the right-hand axis) are the rates of change of gas mass due to SFR (purple), gas inflow (cyan), and gas outflow (magenta). Each of these galaxies shows at least one well-defined compaction phase that is immediately followed by gas depletion and quenching. The onset of gas compaction in the central 1 kpc and the point of maximum central gas compaction are marked by vertical lines.

MNRAS 450, 2327–2353 (2015) Compaction and Quenching in the Inner 1 kpc

Innerwhole 10galaxy kpc

Avishai Dekel gen3 Zolotov+2015 DM VELA07-RP Animations z = 4.4 to 2.3 Daniel Ceverino, Nir Mandelker Gas Compaction

Stars } Face-on Edge-on

< 5 kpc

Gas Gas

< 1 kpc

Stars Stars Our cosmological zoom-in simulations often produce elongated galaxies like observed ones. The elongated distribution of stars follows the elongated inner dark matter halo. MNRAS 453, 408–413 (2015) doi:10.1093/mnras/stv1603 Formation of elongated galaxies with low masses at high redshift Formation of elongated galaxies with low masses at high redshift Daniel Ceverino, Joel Primack and Avishai Dekel DanielABSTRACT Ceverino,1,2‹ Joel Primack3 and Avishai Dekel4 1WeCentro report de Astrobiolog the identiﬁcation´ıa (CSIC-INTA), of elongated Ctra de Torrej (triaxialon´ a Ajalvir, or prolate) km 4, E-28850 Torrejon´ de Ardoz, Madrid, Spain 2galaxiesAstro-UAM, in Universidad cosmological Autonoma simulations de Madrid, at Unidadz ~ 2. AsociadaThese are CSIC, E-28049 Madrid, Spain 3 Department of Physics, University of California, Santa9.5 Cruz, CA 95064, USA 4preferentiallyCenter for Astrophysics low-mass and Planetary galaxies Science, (M∗ ≤ Racah 10 InstituteM⊙), residing of Physics, in The Hebrew University, Jerusalem 91904, Israel Dark matter halos are elongated, especially ! near their centers. Initially stars follow the ! dark matter (DM) haloes with strongly elongated inner parts, a gravitationally dominant dark matter, as shown.! common feature of high-redshift DM haloes in the cold dark But later as the ordinary matter central density Acceptedmatter cosmology. 2015 July 14. A Received large population 2015 June 26; of in originalelongated form galaxies 2015 April 20 grows and it becomes gravitationally dominant, produces a very asymmetric distribution of projected axis ratios, the star and dark matter distributions both Downloaded from become disky — as observed by Hubble as observed in high-z galaxy surveys. This indicates that the Space Telescope (van der Wel+ ApJL Sept majority of the galaxies at high redshiftsABSTRACT are not discs or spheroids 2014).! but rather galaxies with elongatedWe morphologies report the identiﬁcation of elongated (triaxial or prolate) galaxies in cosmological simulations at z 2. These are preferentially low-mass galaxies (M 109.5 M ), residing in dark ≃ ∗ ≤ matter (DM) haloes with strongly elongated inner parts, a common feature⊙ of high-redshift

Nearby large galaxies are mostly disks and spheroids — but they start out looking more like pickles. http://mnras.oxfordjournals.org/ DM haloes in the ! cold dark matter cosmology. Feedback slows formation of stars at the centres of these haloes, so that a dominant and prolate DM distribution gives rise to galaxies elongated along the DM major axis. As galaxies grow in stellar mass, stars dominate the total mass within the galaxy half-mass radius, making stars and DM rounder and more oblate. A large population of elongated galaxies produces a very asymmetric distribution of projected axis ratios, as observed in high-z galaxy surveys. This indicates that the majority of the galaxies at high redshifts are not discs or spheroids but rather galaxies with elongated morphologies. Key words: galaxies: evolution – galaxies: formation. at University of California, Santa Cruz on October 16, 2015

cluded that the majority of the star-forming galaxies with stellar 1INTRODUCTION masses of M 109–109.5 M are elongated at z 1. At lower The intrinsic, three-dimensional (3D) shapes of today’s galaxies redshifts, galaxies∗ = with similar⊙ masses are mainly oblate,≥ disc-like can be roughly described as discs or spheroids, or a combination of systems. It seems that most low-mass galaxies have not yet formed the two. These shapes are characterized by having no preferential a regularly rotating stellar disc at z ! 1. This introduces an interest- long direction. Examples of galaxies elongated along a preferential ing theoretical challenge. In principle, these galaxies are gas-rich direction (prolate or triaxial) are rare at z 0(Padilla&Strauss and gas tends to settle in rotationally supported discs, if the angular 2008;Weijmansetal.2014). They are usually= unrelaxed systems, momentum is conserved (Fall & Efstathiou 1980;Blumenthaletal. such as ongoing mergers. However, at high redshifts, z 1–4, we 1986;Mo,Mao&White1998;Bullocketal.2001). At the same may witness the rise of the galaxy structures that we see= today at time, high-mass galaxies tend to be oblate systems even at high-z. the expense of other structures that may be more common during The observations thus suggest that protogalaxies may develop an those early and violent times. early prolate shape and then become oblate as they grow in mass. Observations trying to constrain the intrinsic shapes of the stellar Prolateness or triaxiality are common properties of dark mat- components of high-z galaxies are scarce but they agree that the ter (DM) haloes in N-body-only simulations (Jing & Suto 2002; distribution of projected axis ratios of high-z samples at z 1.5–4 Allgood et al. 2006; Bett et al. 2007;Maccioetal.` 2007;Maccio,` is inconsistent with a population of randomly oriented disc= galaxies Dutton & van den Bosch 2008;Schneider,Frenk&Cole2012, (Ravindranath et al. 2006;Lawetal.2012;Yuma,Ohta&Yabe and references therein). Haloes at a given mass scale are more pro- 2012). After some modelling, Law et al. (2012) concluded that late at earlier times, and at a given redshift more massive haloes the intrinsic shapes are strongly triaxial. This implies that a large are more elongated. For example, small haloes with virial masses 11 population of high-z galaxies are elongated along a preferential around Mv 10 M at redshift z 2areasprolateastoday’s direction. galaxy clusters.≃ Individual⊙ haloes are= more prolate at earlier times, van der Wel et al. (2014) looked at the mass and redshift de- when haloes are fed by narrow DM ﬁlaments, including mergers, pendence of the projected axis ratios using a large sample of star- rather than isotropically, as described in Vera-Ciro et al. (2011). The forming galaxies at 0

C ⃝ 2015 The Authors Published by Oxford University Press on behalf of the Royal Astronomical Society The Astrophysical Journal Letters,792:L6(6pp),2014September1 van der Wel et al.

0.75

0.5

0.25

0 1 Prolate galaxies dominate at high redshift/low masses 0.75

0.5

Prolate 0.25 Spheroidal Oblate 0

Figure 3. Reconstructed intrinsic shape distributions of star-forming galaxies in our 3D-HST/CANDELS sample in four stellar mass bins and five redshift bins. The model ellipticity and triaxiality distributions are assumed to be Gaussian, with the mean indicated by the filled squares, and the standard deviation indicated by the open vertical bars. The 1σ uncertainties on the mean and scatter are indicated by the error bars. Essentially all present-day galaxies have large ellipticities, and small triaxialities—they are almost all fairly thin disks. Toward higher redshifts low-mass galaxies become progressively more triaxial. High-mass galaxies always have rather low triaxialities, but they become thicker at z 2. ∼ (A color version of this figure is available in the online journal.)

van der Wel+2014 Figure 4. Color bars indicate the fraction of the different types of shape deﬁned in Figure 2 as a function of redshift and stellar mass. The negative redshift bins represent the SDSSSee results also for z< WHEN0.1; the other DID bins ROUND are from 3D-HST DISK/CANDELS. GALAXIES FORM? T. M. Takeuchi et. al ApJ 2015 (A color version of this ﬁgure is available in the online journal.)

Letter allows us to generalize this conclusion to include earlier Despite this universal dominance of disks, the elongatedness epochs. of many low-mass galaxies at z ! 1impliesthattheshapeof At least since z 2moststarformationisaccountedforby agalaxygenerallydiffersfromthatofadiskatearlystages 10 ∼ !10 M galaxies (e.g., Karim et al. 2011). Figures 3 and 4 in its evolution. According to our results, an elongated, low- show that⊙ such galaxies have disk-like geometries over the same mass galaxy at z 1.5 will evolve into a disk at later times, or, redshift range. Given that 90% of stars in the universe formed reversing the argument,∼ disk galaxies in the present-day universe over that time span, it follows that the majority of all stars in the do not initially start out disks.13 universe formed in disk galaxies. Combined with the evidence As can be seen in Figure 3, the transition from elongated that star formation is spatially extended, and not, for example, to disky is gradual for the population. This is not necessarily concentrated in galaxy centers (e.g., Nelson et al. 2012;Wuyts et al. 2012)thisimpliesthatthevastmajorityofstarsformedin 13 This evolutionary path is potentially interrupted by the removal of gas and disks. cessation of star formation.

4 In hydro sims, dark-matter dominated galaxies are prolate Ceverino, Primack, Dekel MNRAS 453, 408 (2015)

Stars

M* <1010 M☉ at z=2

Dark matter

Also Tomassetti et al. 2016 MNRAS, 20 kpc Zhang, Primack, et al. 2018 “Face Recognition for Galaxies” Detecting wet compaction at high redshift with deep learning Marc Huertas-Company, Joel Primack, Avishai Dekel, David Koo, et al. 2018 ABSTRACT We explore a new approach to classify galaxy images from deep surveys oriented towards detecting astrophysical processes calibrated on cosmological hydrodynamic galaxy simulations. To illustrate the methodology we focus on wet compaction. Recent theoretical and observational works have suggested that compact bulges at high redshift might be formed through gas inﬂows (wet compaction events) before quenching. We train a simple Convolutional Neural Network (CNN) with mock CANDELized images from our VELA zoom-in simulations that are selected for being in a wet-compaction phase according to the assembly history extracted from the simulation. We show that the CNN is able to retrieve a galaxy in the compaction phase within a time window of ±0.3 Hubble times based only on the pixels distribution. We then use the trained network to classify real galaxies from the CANDELS survey into three classes (pre-compaction, compaction and post-compaction). We ﬁnd that compaction typically occurs at a characteristic stellar mass of M*∼109.5−10 solar masses all redshifts, as in the VELA simulations. The galaxies that are experiencing compaction in the CANDELS redshift range (1 < z < 3) are therefore typically the progenitors of M*∼1010.5 solar mass galaxies at z ∼ 0, like the Milky Way. The presented technique can be generalized to other processes and could constitute an alternative way of classifying galaxies in the era of massive imaging surveys and cosmological simulations, to help improve the comparison between theory and observations. compaction deep-learning 7

Examples of CANDELized simulated galaxy images compaction deep-learning 7

pre-compaction

compaction

post-compaction Figure 2. Random example of simulated Candelized images in the 3 phases discussed in this work. The top row shows pre-compaction galaxies, the middle row are galaxies in the process of compaction and the bottom row are post-compaction objects.

Figure 2. Random example of simulated Candelized images in the 3 phases discussed in this work. The top row shows pre-compaction galaxies, the middle row are galaxies in the process of compaction and the bottom row are post-compaction objects.

FigureArchitecture 3. Architecture of the deep networkof the used fordeep classification network in this work. The used network isfor a standard classification and simple CNN configuration in madethis of 3 convolutional layerswork. followed by poolingThe and network dropout. is a standard and simple CNN configuration made of 3 convolutional layers followed by pooling and dropout. Figure 3. Architecture of the deep network used for classification in this work. The network is a standard and simple CNN configuration made of 3 convolutional layers followed by pooling and dropout.

Figure 5. Normalized confusion matrix of the 3-label classification on a test dataset not used for training nor validation. The y-axis shows the true class Figure 4. Learning history resulting form the strategy described in the text. from the simulation metadata, the x-axis is the predicted class. The blue solid line shows the accuracy on the training set and the red solid line is the accuracy for the validation set. Every 50 epochs the validation and training datasets are modified which explains the discontinuities. The Figure 5. Normalized confusion matrix of the 3-label classification on a test accuracy on the validation is generally unstable because it is only made of 2 to precisely determine what are thedataset features not usedthe machine for training is usingnor validation. The y-axis shows the true class galaxies. See text for details. Figure 4. Learning history resulting form theto strategy decide described the output in classification. the text. Thisfrom is the a simulation general problem metadata, for the all x-axis is the predicted class. The blue solid line shows the accuracy on thedeep-learning training set and applications. the red solid However, there exist more and more net- line is the accuracy for the validation set. Everywork 50 interrogation epochs the validationtechniques which allow to identify the pixels that and training datasets are modified which explainsmost contributed the discontinuities. to the final The classification among the input image. 5.3 Inside the network accuracy on the validation is generally unstable because it is only made of 2 One recent method is called integratedto precisely gradients determine (Sundararajan what et are the features the machine is using galaxies. See text for details. An important caveat of the machine learning approach presented al. 2017). It is based on the measurementto decide of the the di outputerences classification. between This is a general problem for all above is that it somehow behaves as a black box. It is thus dicult gradients computed by the networkdeep-learning in an input image applications. as compared However, there exist more and more network interrogation techniques which allow to identify the pixels that MNRAS 000, 1–?? (2017) most contributed to the final classification among the input image. 5.3 Inside the network One recent method is called integrated gradients (Sundararajan et An important caveat of the machine learning approach presented al. 2017). It is based on the measurement of the dierences between above is that it somehow behaves as a black box. It is thus dicult gradients computed by the network in an input image as compared

MNRAS 000, 1–?? (2017) 8 Huertas-Company et al. 8 Huertas-Company et al.

Simulated galaxy with single compaction event 8 Huertas-Company et al. 8 Huertas-Company et al.

pre-compaction compaction post-compaction

Figure 6. Examples of predictions on a test sample. The left column shows the probability of being in pre-compaction (blue solidMNRAS line), compaction000, 1–?? (2017) (greenFigure 6. Examples of predictions on a test sample. The left column shows the probability of being in pre-compaction (blue solidMNRAS line), compaction000, 1–?? (2017) (green solid line) and post-compaction(red solid line) predicted by the CNN. The right column shows the input simulation metadata used to define the phases. Thesolid red, line) and post-compaction(red solid line) predicted by the CNN. The right column shows the input simulation metadata used to define the phases. The red, green and orange shaded regions show the primary, secondary and tertiary compactionSimulated events as defined from the gas and galaxy stellar masses (see text with for details).green and multiple orange shaded regions show thecompaction primary, secondary and tertiary compaction events events as defined from the gas and stellar masses (see text for details).

pre-compaction compaction post-compaction

Examples of predictions on a test sample. The left column shows the probability of being in pre-compaction (blue solid line), compaction (green Figure 6. MNRAS 000, 1–?? (2017)Figure 6. Examples of predictions on a test sample. The left column shows the probability of being in pre-compaction (blue solidMNRAS line), compaction000, 1–?? (2017) (green solid line) and post-compaction(red solid line) predicted by the CNN. The right column shows the input simulation metadata used to define the phases. Thesolid red, line) and post-compaction(red solid line) predicted by the CNN. The right column shows the input simulation metadata used to define the phases. The red, green and orange shaded regions show the primary, secondary and tertiary compaction events as defined from the gas and stellar masses (see text for details).green and orange shaded regions show the primary, secondary and tertiary compaction events as defined from the gas and stellar masses (see text for details). Testing the Trained Deep Learning Code compaction deep-learning 9

average probability to increase the robustness of the classification. pre-compaction We stress that there is a general good agreement between the compaction dierent models which confirms that the classification does not post-compaction strongly depend on the specific subset of simulated galaxies used for training. The typical scatter in the probability values is of the order of 0.1. ⇠ The first thing to notice is that the classification applied to real data returns objects with high probability values in the 3 classes. The fraction of galaxies with all probabilities lower than 0.5 is only 2% of the total sample. It means that the model found galaxies that resemble the galaxies in the simulation with high confidence.This is obviously not a proof that compaction happens in real galaxies but reflects that the simulated galaxies are fairly similar to the observed Observability of the compaction event with the calibrated classifier. Theones. It is not obvious to establish what would happen if galaxies histograms showFigure the 7. Observability distributions of the compactionof time (relative event with to the the calibrated Hubble classi- time at the time fier. The histograms show the distributions of time (relative to the Hubble from the training are real datasets were very dierent though. This of compaction). The dashed vertical lines show the average values for each class time at the time of compaction). The blue, green and red histograms show will be definitely explored in future work. A possibility is that the with the samethe color pre-compaction, code. Despite compaction some and post-compaction overlap, the phases. classifier The dashed is able toprobability establish values would be very unstable between models and/or temporal constraintsvertical lines on show the the different average values phases. for each class Integrated with the same gradient color code. methodvery shows low in all classes. Also if only one class was dominant over that the classifierDespite is some using overlap, relevant the classifier pixels, is able not to establish noise. temporal constraints the others, that would reflect that some of the features learned to on the dierent phases. identify the dierent phases from the simulations are not present in the data. This is not the case at least at first order which allows us to push the analysis a bit further. to a test image (usually a blank image with only zeros). We tested In figure 9 we first look at the stellar mass distributions of this method in our model and computed the integrated gradients for galaxies in the three dierent phases. Recall that the simulations some of the galaxies. Figure 8 shows one example for each class. used for training stop at z 1, so the classification in the low ⇠ The interpretation is not straightforward. However some useful in- redshift bin should be considered an extrapolation. It is shown formation can be extracted from this exercise. It is interesting to see here for consistency but should not probably be considered very that the model automatically segments all the pixels belonging to seriously. The most interesting result from the figure is that the the galaxy and takes the decision based on all the galaxy. It also mass distributions of the dierent phases are rather dierent. Pre- means it understood there is no information in the noise and con- compaction galaxies tend to be more abundant at low stellar masses 9 9.5 firms the model is not over-fitting on the noise pattern. Also, as (M /M < 10 ) and post-compaction galaxies dominate at ⇤ pointed out in previous works, after the compaction event a disk large stellar masses (M /M > 1010.5). Compaction galaxies are ⇤ is rebuilt in the simulations. The bottom line of the figure shows most frequent at intermediate masses and peak at 109.5 10. In- ⇠ clearly that the machine detects the diuse disk component even if terestingly this seems to be relatively independent of redshift. This faint and probably uses this information to make the decision. For characteristic mass for compaction is expected as reported in Tac- the compaction galaxies, the relevant pixels are more concentrated chella et al. (2016) and also reflected in table 1. It again confirms in the center since the galaxies are generally more compact. It is that the network is automatically extracting the correlations existing also worth noticing that the gradient tends to have values of dierent in the simulations. One could wonder at this stage if the network sign in the center and in the outskirts as if the machine was using is only using the luminosity as main proxy for the classification. dierence in flux between the center and the outskirts to classify. We think it is unlikely since a galaxy of the same mass has very This is somehow expected since the compaction event is by defini- dierent luminosities in the redshift range probed. Also, as shown tion accompanied by a burst of central star-formation. The model is in the figure 8, the full pixel distribution is used by the network. We capturing all these correlations automatically. This is the strength then explore in figure 10 the redshift evolution of the fractions of of the presented methodology. Although the information that can galaxies in the 3 phases at fixed stellar mass. Both plots are com- be extracted from integrated gradients is quite limited at this stage, plementary. As expected the redshift evolutions strongly depend on it is reasonable to think that interrogation techniques will become stellar mass. The galaxies that are more frequently experiencing more advanced, and therefore there is potentially information that compaction in the CANDELS redshift range are in the stellar mass can be learned from a post-processing of the model outputs in the range of 109 < M /M < 109.5, and are typically the progeni- ⇤ future. tors of sub-M⇤ - M⇤ galaxies according to abundance matching (Rodriguez-Puebla et al. 2017). More massive galaxies tend to be in the post-compaction phase at all redshifts. This means that if one wants to observe the progenitors of these most massive galaxies in 6 APPLICATION TO REAL DATA: ARE BLUE the process of compaction, it is required to probe 109 solar mass NUGGETS IN THE PROCESS OF COMPACTION? ⇠ galaxies at z > 3. It will be straightforward with JWST. We now apply the model to the real HST/CANDELS sample Therefore, most of the observed massive galaxies (M /M > presented in section 3. We simply cut stamps around the selected 10.3) in CANDELS are in a post-compaction phase according⇤ to our galaxies in the three infrared filters (F160W, F125W, F105W) classification based on numerical simulations. What about the so- and classify them into three classes using the trained models. Since called blue-nuggets defined observationally? In figure 11 we show 10 models were produced (see section 4), we use each of them to the fraction of observationally defined blue nuggets in the dierent classify all galaxies. Each real galaxy has therefore 10 dierent phases. We select compact star-forming galaxies as objects in the classifications using slightly dierent models. We then compute the main sequence (SFR > 0.5) and in the ridge ridge-lineline of

MNRAS 000, 1–?? (2017) Integrated gradients output of the model. The left column is the original image and the other columns show the integrated gradients for the different wavelength ﬁlters. The network automatically detects10 Huertas-Companythe pixels belonging et al.to the galaxy and used all of them to make the decisions.

Figure 8. Integrated gradients output of the model. Each row shows a galaxy in a dierent stage (Pre-compaction, Compaction, Post-Compaction). The left column is the original image and the second, third and fourth columns show the integrated gradients for the dierent ﬁlters. The network automatically detects the pixels belonging to the galaxy and used all of them to make the decisions.

quiescent galaxies in the ⌃1 M plane as defined by Barro et al. SFR ⌃1Q plane. Instead of selecting galaxies at fixed stellar ⇤ (2017), i.e. 0.3 < ⌃ < 0.3 with SFR and ⌃ defined as: mass, we use here the average mass growth tracks from Rodriguez- 1Q 1Q Puebla et al. (2017) inferred from abundance matching. We focus SFR = LogSFR µ z LogM /M . + logC z [ ( ) ( 10 5) ( )] on the progenitors of 1013 solar mass halos at z 0 which are ⇥ ⇤ ⇠ ⇠ the galaxies that typically have 109.5 solar masses at z 2 as ⇠ ⇠ ⌃1Q = Log⌃1 [Q (z) (LogM /M 10.5) + LogBQ (z)] ⇥ ⇤ shown in the figure 12. Therefore these galaxies are most probably in a compaction phase in the CANDELS dataset as deduced Q (z), BQ (z), µ(z) and C(z) are the best fit values of the from the results from figures 9 and 10. Galaxies are selected at M ⌃1 and M SFR relations for quiescent and star-forming ⇤ ⇤ . galaxies respectively taken from Barro et al. (2017). The figure each redshift bin within a stellar mass interval of 0 6 dex centered shows that 70% of the compact star-forming galaxies are in a on the average stellar mass at that redshift from the curve of the ⇠ post-compaction phase while only 30% are compactifying. This figure 12. We then analyze the positions of galaxies in the dier- ⇠ suggests that the compact star-forming galaxies selected in pre- ent classes (pre-compaction, compaction, post-compaction) in the SFR ⌃ plane in figure 13. We observe that the majority vious works are preferentially selected at t > 0.5/H(t) after the 1Q of compaction events for the progenitors of 1010.5 solar mass compaction phase and not during compaction phase as previously ⇠ claimed. However, we also show in the figure, the same quantity galaxies take place at z > 1.5 while galaxies are still on the main for red-nuggets ( < 0.5). Practically all compact quiescent sequence. Below this redshift most of the galaxies are already in SFR galaxies ( 90%) are in the post-compaction phase as expected. a post-compaction phase as shown by the two top panels of the ⇠ There is therefore evidence that the fraction of compaction is larger figure. It is interesting to notice that all dense galaxies as inferred in blue than in red nuggets even if not dominant. If we take into ac- by ⌃1 are most probably classified as post-compaction. However a count the observability constraints, it suggests that at least some of majority of post-compaction galaxies still sit on the main-sequence the blue nuggets are progenitors of the quiescent compact galaxies and would not be classified as compact. The results suggest that and are in the process of compaction. while compaction can move galaxies towards high ⌃1 values, most In order to further investigate if compaction is one of the of the post-compaction never reach such high values and remain in mechanisms responsible of building up the galaxy cores, we ex- the main-sequence. plore the position of galaxies classified in dierent phases in the

MNRAS 000, 1–?? (2017) Applying the Trained Deep Learningcompaction deep-learning Code to11 CANDELS Galaxies

Stellar mass distributions of HST (not trained for CANDELS galaxies in these redshifts) pre-compaction, compaction, and post- compaction phases in different redshift bins. The DL code correctly shows the temporal evolution. Galaxies in the compaction phase typically peak at stellar

masses 109.5−10 Msun at all redshifts, as in the VELA simulations.

Figure 9. Stellar mass distributions of galaxies in pre-compaction (blue lines), compaction (green lines) and post-compaction (red lines) for dierent redshift 9.5 10 bins as labelled. Galaxies in the compaction phase typically peak at stellar masses of 10 at all redshifts.

7 SUMMARY AND CONCLUSIONS then applied the trained model to real galaxies from the CANDELS survey observed with the HST in the same redshift range and clas- We have explored a new approach to classify galaxy images us- sify them in three main classes: pre-compaction, compaction and ing deep-learning calibrated on numerical simulations. The general post-compaction. The network finds galaxies with high probability methodology consists first in generating mock images of galaxies of being in the three classes indicating significant similarity be- reproducing the observing conditions from hydro cosmological sim- tween simulated and observed galaxies. The classification naturally ulations which are then labelled with the known assembly history recovers a characteristic stellar mass for compaction of 109.5 10 ⇠ from the simulation. The images are then fed to an unsupervised solar masses independent of redshift. Hence, the typical galaxies feature learning machine that learns the features to detect a given experiencing compaction in the CANDELS redshift range are the assembly process. We have applied the method for detecting wet 10.5 progenitors of sub M⇤ galaxies (10 M /M ) at z = 0. More ⇤ compaction which has been put forward recently as a plausible massive compact galaxies are found to be preferentially in the post- mechanism to grow bulges at high redshift. We have used to that compaction class, so they are compatible with having experienced purpose a suite of high resolution hydro numerical simulations of compaction more than 0.5 Hubble times before the time of obser- intermediate mass galaxies in the redshift range 1 < z < 3. We have vation. The presented methodology can be easily adapted to other shown that a simple CNN is able to detect galaxies in the process of physical processes captured in simulations. It can constitute a use- compaction within a time window of 0.2 Hubble times. We have ±

MNRAS 000, 1–?? (2017) Computer vision and deep learning applied to simulations and imaging of galaxies and the evolving universe

Joel Primack University of California, Santa Cruz

Large-scale simulations track the evolution of structure in the universe of dark energy and cold dark matter on scales of billions of light years

Cosmological zoom-in simulations model how individual galaxies evolve through the interaction of atomic matter, dark matter, and dark energy

Our VELA galaxy simulations agree with HST CANDELS observations that most galaxies start prolate, becoming spheroids or disks after compaction events

A deep learning code was trained with VELA galaxy images plus metadata describing whether they are pre-compaction, compaction, or post-compaction

The trained deep learning code was able to identify the compaction and post-compaction phases in CANDELized images

The trained deep learning code was also able to identify these phases in real HST CANDELS observations, ﬁnding that compaction occurred for stellar mass 109.5 -10 Msun, as in the simulations — and James Webb Space Telescope will allow us to do even better