The Cosmic Dawn (CoDa) Project: Simulating Reionization and … · 2019. 3. 13. · • Better agreement than CoDaI ! CODA II . Z = 6 . Z = 10 CODA II . CODA II • Color indicates

Pierre Ocvirk7, Dominique Aubert7, Taha Dawoodbhoy1, Ilian Iliev2, Hyunbae Park1, Anson D’Aloisio1,15, Romain Teyssier8, Gustavo Yepes9, Stefan

Gottloeber10, Nicolas Gillet7,Junhwan Choi1, David Sullivan2, Yehuda Hoffman14, Jenny Sorce16, Nicolas Deparis7, Joseph Lewis7 + Other CLUEs Members (1)U Texas at Austin (2)U Sussex (7)U Strasbourg (8) U Zurich (9) U Madrid

(10) AIP Potsdam (13) Hebrew U (15)U Washington (16) U Lyons

Paul Shapiro The University of Texas at Austin

Collaborators in this work include:

“Into the Starlight : The End of the Cosmic Dark Ages” Aspen Winter Workshop, Aspen Center for Physics, March 7, 2019

The Cosmic Dawn (CoDa) Project: Simulating Reionization and Galaxy

Formation







The Cosmic Dawn (CoDa) Project: Simulating Reionization and Galaxy

Formation







Suppression of Star Formation in Low-

Mass Galaxies Caused by the Reionization of their Local

Neighborhood






Reionization of the Universe

“Into the Starlight : The End of the Cosmic Dark Ages“ Aspen Winter Workshop, Aspen Center for Physics, March 7, 2019

Self-Regulated Reionization Iliev, Mellema, Shapiro, & Pen (2007), MNRAS, 376, 534; (astro-ph/0607517)

•Jeans-mass filtering

low-mass source halos

(M < 109 Msolar) cannot form

inside H II regions ;

•50 Mpc box, 4063 radiative

transfer simulation, WMAP3,

fγ = 250;

•resolved all halos with

M > 108 Msolar (i.e. all

atomically-cooling halos),

(blue dots = source cells);

• Evolution: z=21 to zov = 7.5.

• Largest fully-coupled rad-hydro reionization simulations to-date, global and local • Volume large enough to simulate global reionization; resolution high enough to

follow all atomic-cooling halos (>108 M_solar ), including MW/M31 dwarf satellites. • Three EOR Simulations of 91 Mpc box from CLUEs Constrained Realization I.C.’s -- CoDa I and CoDa II : -- CoDa I – AMR :

RAMSES-CUDATON simulations (uniform grid): hybrid CPU-GPU code • Grid size = (4096)3 cells, Δx ~ 20 cKpc < 3 Kpc’s ; N-body particles = (4096)3 ~ 69 billion • Min halo mass ~ 108 M_solar ~300 particles

Introducing the CoDa (COSMIC DAWN) Simulations

Ocvirk, Gillet, Shapiro, Aubert, Iliev, Teyssier, Yepes, Choi, Park, D’Aloisio, Sullivan, Gottloeber, Hoffman, Stranex, Knebe 2016, MNRAS, 463,1462 (arXiv:1511.00011) ( CoDa I)

EMMA simulation (Adaptive Mesh Refined): hybrid CPU-GPU code • AMR resolution = (16,384)3 cells (fully-refined) Δx ~ 5 cKpc ~ 500 pc • N-body particles = (2048)3 ~ 8.6 billion • Min halo mass ~ 108 M_solar ~ 36 particles

Aubert, Deparis, Ocvirk, Shapiro, Iliev, Yepes, Gottloeber, Hoffman, Teyssier 2018, ApJL , 856, L22 (arXiv:1802.01613) (CoDa I - AMR)

Ocvirk, Aubert, Sorce, Shapiro, Deparis, Dawoodbhoy, Lewis, Teyssier, Yepes, Gottloeber, Ahn, Iliev, Hoffman, 2018, MNRAS, submitted (arXiv:1811.11192) (CoDa II)

Reionization of the Local Universe: Witnessing our Own Cosmic Dawn

Gas Temperature at

z = 6.15 in the supergalactic

YZ plane of the

Local Group

2/3 of the full box width

Circles indicate progenitors of Virgo, Fornax, M31, and the MW

Orange is photoheated, photoionized gas; Red is SN-shock-heated; Blue is cold and neutral

CODA I


• Efficiencies set by smaller-box simulations prove slightly low, so reionization ends a bit late: zrei < 5

• But if we let z z * 1.3 , there is good agreement with observable constraints

CODA I


UV Luminosity Function vs. Galaxy Observations from Bouwens et al. (2015), and Finkelstein et al. (2015)

• Full circles: Bouwens et al. (2015); Crosses: Finkelstein et al. (2015) • Shaded areas and thick lines show the envelope and median of the LFs of 5 equal, independent subvolumes (50/h cMpc)3

• MAB1600 magnitudes computed using lowest metallicity SSP models of Bruzual & Charlot (2003), scaled to same ionizing photons released per 10 Myr • Shift simulation z z * 1.3

CODA I

Simulating Galaxy Formation and Reionization of the Local Universe with Fully-Coupled Radiation-Hydro: CoDa II

Z

TITAN Supercomputer requirements • # CPU cores (+ # GPUs) = 65536 (+ 16384) • runtime ~ 6 days

RAMSES-CUDATON simulation • Box size = 94 cMpc • Grid size = (4096)3 cells, Δx ~ 20 cKpc < 3 Kpc • N-body particles = (4096)3 ~ 69 billion • Min halo mass ~ 108 M_solar ~300 particles • Newer CLUEs initial conditions

Ocvirk, Aubert, Sorce, Shapiro, Deparis, Dawoodbhoy, Lewis, Teyssier, Yepes, Gottloeber, Ahn, Iliev, Hoffman 2018, MNRAS submitted (arXiv: 1811.11192)

Z

Ocvirk, Aubert, Sorce, Shapiro, Deparis, Dawoodbhoy, Lewis, Teyssier, Yepes, Gottloeber, Ahn, Iliev, Hoffman 2018, MNRAS submitted (arXiv:1811.11192)

CoDa II

• Star formation efficiencies recalibrated so reionization ends earlier: zrei > 6

• Good agreement with observable constraints • Post-reionization

neutral fraction (UV background) a bit low (high), respectively gas clumping missing at small scales?

CODA II

J21

SFR density

UV Luminosity Function vs. Galaxy Observations from Bouwens + (2015), (2016), Finkelstein + (2015), Atek + (2018)

• Shaded areas and thick lines

show the envelope and median of the LFs of 5 independent, rectangular subvolumes, each

1/5 full box volume, similar to CANDELS-DEEP volume at z = 6

• MAB1600 magnitudes computed using BPASS Z = 0.001 binary population model (no dust) for each halo star particle, evolved since its birth-time, scaled to deliver the same ionizing photons released per 10 Myr • Better agreement than CoDa I !

CODA II

Z = 6

Z = 10

CODA II

CODA II

• Color indicates galaxy number density #/Mpc3/log(M/Msolar)

Which Haloes Correspond to Which Galaxy UV Magnitudes?

CODA II • Shaded area is observed SFR density of bright galaxies with MAB1600 < -17.7 from Bouwens etal (2015), decreased by 0.34 dex to replace their MAB1600 – SFR conversion for single-star, Salpeter IMF by BPASS Z = 0.001, binary model (no dust) from Kroupa IMF.

• CoDa II SFRs use BPASS Z = 0.001 binary population model for each star particle, evolved since its birth. • CoDa II SFRD of stars in halos

with MAB1600 < -17 matches Bouwens + (2015), but is higher when fainter galaxies included.

• Total SFRD declines much less towards high z than current observations of bright galaxies.

Reionization suppresses star formation rate in dwarf galaxies, for M < 109 solar masses • photoionization-heating &

SN remnant shock-heating raises gas pressure

• Gas pressure of heated gas

resists gravitational binding into the low-mass galaxies

lowers the cold, dense baryon gas fraction lowers the SFR per unit halo mass • Low-mass atomic cooling

halos (LMACHs) are most suppressed

SFR per Halo

CODA I

CODA I

• SFR ∝ Mα , α ~ 5/3 for M > 1010 solar masses, but drops sharply below M ~ 3 X 109 below z ~ 6

SFR per Halo

Reionization still suppresses star formation rate in dwarf galaxies, for M < 109 solar masses • Low-mass atomic cooling

halos (LMACHs) are most suppressed

• CoDa II suppression is less dramatic than CoDa I but still there!

• CoDa II removed the

temperature criterion in SF recipe of CoDa I

• SFR ∝ Mα , α ~ 1.4 for M > 1010 solar masses, steepening below M ~ 109 , but then drops sharply below M ~ 3 X 109 toward EOR end

CODA II

Z = 10

Z = 5.8

CoDa I @ end of EOR

Reionization is Inhomogeneous in Space and Time

Ionization Field

Ocvirk, Gillet, Shapiro, Aubert, Iliev, Teyssier, Yepes, Choi, Park, D’Aloisio, Sullivan, Gottloeber, Hoffman, Stranex, Knebe 2016, MNRAS, 463, 1462 (arXiv:1511.00011)

CoDa I

Ionization Field

Ocvirk, Gillet, Shapiro, Aubert, Iliev, Teyssier, Yepes, Choi, Park, D’Aloisio, Sullivan, Gottloeber, Hoffman, Stranex, Knebe 2016, MNRAS, 463, 1462 (arXiv:1511.00011)

Suppression of Star Formation in Low-Mass Galaxies Caused by the Reionization of their Local Patch

Dawoodbhoy, Shapiro, Ocvirk, Aubert, Gillet, Choi, Iliev, Teyssier, Yepes, Knebe, Gottloeber, D’Aloisio, Park, Hoffman 2018, MNRAS, 480, 1740 (arXiv:1805.05358)

Local Reionization: There Goes the Neighborhood!

• Reionization history was different in different locations each point in space had its own redshift of local reionization, zre(x)

• From CoDa I, we computed local

reionization redshift field zre(x), by smoothing underlying 40963 cells to 2563 cells, 0.357 cMpc on a side. Each halo was assigned the value of zre for its location.

Local Group

Zoom-In (4 h-1 cMpc)3 Subvolumes = (full simulation volume/4096)

Zoom-In (4 h-1 cMpc)3 Subvolumes = (full simulation volume/4096)

Suppression of Star Formation in Low-Mass Galaxies Caused by the Reionization of their Local Patch

Zre

Local Reionization Redshift Field • Local reionization redshift field

resembles underlying cosmic web of dark matter, galaxies and IGM: density peaks reionized first, while voids and underdense regions reionized last and occupy the most volume.

Dawoodbhoy, Shapiro, Ocvirk, Aubert, Gillet, Choi, Iliev, Teyssier, Yepes, Knebe, Gottloeber, D’Aloisio, Park, Hoffman 2018, MNRAS, 480, 1740 (arXiv:1805.05358)

• Local reionization redshift field

resembles underlying cosmic web of dark matter, galaxies and IGM: density peaks reionized first, while voids and underdense regions reionized last and occupy the most volume.

Local Reionization Redshift Field

Local Reionization Redshift correlates with local patch overdensity

Local Reionization Redshift correlates with local patch overdensity

Averaged over Their Local Cut-outs of (4/h cMpc)3: δMW = 0.16 δM31 = 0.09 @ z = 11.4

Local reionization redshift correlates with local patch overdensity Local reionization redshift correlates with local halo mass function

• The globally-averaged suppression of SFR per halo in low-mass halos is

actually the superposition of different local reionization and suppression histories.

• Grouping halos of different mass in bins of local reionization redshift zre shows: suppression of low-mass halos follows their local reionization!

SFR per halo vs. redshift for low-mass haloes, 108-9 M⊙, for different zre: 2 dots on each curve indicate range of zre for each bin.

z z

Local Reionization and Suppression of the SFR

• Haloes of intermediate mass, 109 – 10 M⊙, show suppression, too, but weaker and with a delayed turn-over.

• High-mass haloes, above 1010 M⊙, show increasing mass-bin-averaged SFRs per halo, with no suppression turn-over.

SFR per halo vs. redshift for haloes of intermediate-mass (109-10 M⊙) and high-mass (> 1010 M⊙), for different zre: 2 dots on each curve indicate range of zre for each bin (left = intermediate-mass, right = high-mass).

Local Reionization and Suppression of the SFR

• Whether a region is self-ionized or externally-ionized depends upon the net balance between its local production of ionizing photons and the recombinations that cancel its ionizations.

• Any cell for which the ratio of its

fraction of all ionizing photons produced globally during the EoR to its fraction of all the recombinations is greater than one must create a surplus for itself, in ionizing photons, while a cell for which this ratio is less than one cannot reionize itself fully, on its own.

• We find that early-reionizing cells create a surplus, which they must export, while late-reionizing cells have a short-fall and must import ionizing photons to finish their reionization.

Regional contributions to the total number of ionizing photons and recombinations during EoR: ionizing photon (stellar mass) and recombination fractions vs. zre.

The last 87% of cells to reionize formed just 7.4% of the stars. The first 6.25% of cells to reionize formed 83.1% of the stars!

Local Reionization and the Export and Import of Ionizing Photons

Zre


f* = fraction of total stellar mass fV = fraction of the volume frec = fraction of recombinations

• Whether a region is self-ionized or externally-ionized depends upon the net balance between its local production of ionizing photons and the recombinations that cancel its ionizations.

• Any cell for which the ratio of its

fraction of all ionizing photons produced globally during the EoR to its fraction of all the recombinations is greater than one must create a surplus for itself, in ionizing photons, while a cell for which this ratio is less than one cannot reionize itself fully, on its own.

• We find that early-reionizing cells create a surplus, which they must export, while late-reionizing cells have a short-fall and must import ionizing photons to finish their reionization.

Regional contributions to the total number of ionizing photons and recombinations during EoR: ionizing photon (stellar mass) and recombination fractions vs. zre.

The last 87% of cells to reionize formed just 7.4% of the stars. The first 6.25% of cells to reionize formed 83.1% of the stars!


Zre

CoDa II: Galaxy Escape Fractions and the Ionizing Photon Budget

Lewis, Ocvirk, Shapiro, Aubert, et al 2019, in preparation

Zoom-in on a single halo at z = 6 with M ~ 1010 Msolar • Pink symbols are star

particles

• Circle is virial radius r_200

• Cube is region from which we compute f_esc

• Arrows show the escaping ionizing radiation flux

• fesc is higher for low-mass halos, decreasing with

halo mass for M > few X 109 Msolar • fesc for a given Mhalo

increases over time during the EOR,

roughly as (1+ z)-3

• There is a significant scatter in fesc about the mean relationship vs. halo mass at the same redshift, greater at low mass

• Ionizing photon luminosity per halo depends on

fesc X SFR • Higher fesc’s at

low mass offset higher SFR’s at higher mass

• The contribution of halos of different mass to the total ionizing photon emissivity of galaxies during the EOR also depends on the halo mass function.

CoDa halos

HMFcalc analytical fits

• The halo mass range that dominates the total rate of photon release is lower than the mass range that dominates the total SFR density, since fesc is lower for higher-mass halos.

• When <xH II> = 0.5 at z ~ 7, ~ 80% of ionizing photons released come from halos between 109 and 4 X 1010 Msolar

• When <xH II> = 0.1 at z ~ 8.5, ~ 80% come from M < ~ 5 X 109 Msolar

Simulating Galaxy Formation and Reionization of the Local Universe with Adaptive Mesh Refinement: CoDa I - AMR

EMMA simulation • Box size = 91 cMpc • AMR resolution = (16,384)3 cells (fully-

refined) Δx ~ 5 cKpc ~ 500 pc • N-body particles = (2048)3 ~ 8.6 billion • Min halo mass ~ 108 M_solar ~36 particles

TITAN Supercomputer requirements • # CPU cores (+ # GPUs) = 32,768 (+ 4096) • # CPU hrs ~ 2 million node hrs ~ 10 days

Aubert, Deparis, Ocvirk, Shapiro, Iliev, Yepes, Gottloeber, Hoffman, Teyssier 2018, ApJL 856, L22 (arXiv:1802.01613)

Aubert, Deparis, Ocvirk, Shapiro, Iliev, Yepes, Gottloeber, Hoffman, Teyssier 2018, ApJL 856, L22 (arXiv:1802.01613)

• Reionization history was different in different locations each point in space had its own redshift of local reionization, zre(x)

• Dense regions of cosmic web

reionize early, voids reionize late reionization redshift field resembles the cosmic web

Rei

oniz

atio

n R

edsh

ift

Local Reionization Redshift Field

The Inhomogeneous Reionization Times of Present-Day Galaxies: CoDa I - AMR

Local Reionization Redshift: Islands in the Stream

• High (low) density

IGM reionized early (late).

• Islands in the 3D map of reionization redshift were isolated from each other and from external ionization

Can use these maps to see that MW and M31 progenitors did not influence each other, and their reionization was not external

Rei

oniz

atio

n R

edsh

ift

Virgo

MW + M31

Local Reionization Redshift: Islands in the Stream

• High (low) density

IGM reionized early (late).

• Islands in the 3D map of reionization redshift were isolated from each other and from external ionization

Can use these maps to see that MW and M31 progenitors did not influence each other, and their reionization was not external

Rei

oniz

atio

n R

edsh

ift

When the Neighborhoods of Present-Day Galaxies were 50% Reionized, as a Function of their Mass at z = 0

Two Methods • Merger Tree traces mass assembly history of each galaxy from z = 0 back to its reionization time at z > 6 time when location of most massive progenitor reached 50% reionized Progenitor-based zR • All halo particles in galaxy at z = 0 are traced back to their 50% reionized locations at z > 6, and these are averaged Particle-based zR

Z R Full Volume was 50% reionized at <zR> ~ 7.8

The Spread of the 50% Reionization Times Within Present-Day Galaxies, as a Function of their Mass at z = 0

• Spread of reionization times

within galaxies sometimes as large as galaxy-to-galaxy scatter.

• Galaxies with M(z=0) > 1011 Msolar have typical rms spreads of ~ 120 Myrs, higher for higher mass. • But galaxies with M(z=0) < 1010 Msolar have typical rms spreads of ~ 0 Myrs rapid ionization by external

radiation

Sorce, Ocvirk, Aubert, Shapiro, et al (2018)

CoDa II Reionization Epoch of the Local Group

Documents

The Cosmic Dawn (CoDa) Project: Simulating Reionization and … · 2019. 3. 13. · • Better agreement than CoDaI ! CODA II . Z = 6 . Z = 10 CODA II . CODA II • Color indicates