1. Romeel Davé, Univ. of Arizona
The Growth of Massive Galaxies in Cosmological Hydrodynamic Simulations
Romeel Davé, Univ. of Arizona
With: Lars Hernquist (CfA), Dušan Kereš & Neal Katz (UMass), Volker Springel (Garching) and David Weinberg (Ohio State)

2. Elements of Galaxy Formation Theory
How does gas get into galaxies?
CDM + shock heating + cooling
 Classic overcooling problem
How does gas get out of galaxies?
Feedback, winds, AGN, jets, etc.
State of the Models: Lots of progress in former, still fumbling our way through the latter.

3. Gadget2 Hydro Simulations
Entropy-conservative SPH + Tree-PM (Springel & Hernquist).
H&He cooling, Jn, star formation according to Kennicut Law.
Multi-phase subgrid ISM model, based on McKee & Ostriker.
Supernova feedback:
Thermal, added to hot ISM phase.
Superwind, randomly expels gas from galaxies.
Wind speed (constant) chosen to roughly reproduce observed W* today; reduces stellar mass by factor of ≈2-3.
- G6 model:
2x4863 particles: mbary=1.3x108 M.
100 Mpc/h box size, 2 kpc/h resolution
- LCDM: W=0.3, H0=70, n=1, s8=0.9, Wb=0.04
- Galaxies found using SKID, spectra from integrating SF histories of individual galaxies using BC03 models, assuming Z, Salpeter IMF, E(B-V)=0.1; all parameters constant with redshift.

4. Global Luminosity Density Evolution
Rudnick et al (2004): SDSS + Combo-17 + FIRES.
Rest-frame V, B, U luminosities broadly z~0, and show expected increase from z = 0→3:
U, G6: 5.5x, Obs: 4.9±1.0.
B, G6: 3.8x, Obs: 2.9±0.6.
V, G6: 2.6x, Obs: 1.9±0.4.
At z~3, G6 shows over-abundance of stellar mass (V), while the SFR matches (U).
Overall broad agreement.

5. Stellar Mass Function to z~1
From MUNICS (Drory et al 04), stellar mass function at z=0.5,1 (points with errorbars)
Simulations show less evolution than data, particularly below L*.

6. Rest-frame K-band LF Evolution
K20 Survey (Cimatti et al 03).
Good agreement at all redshifts (up to resolution limit of G6), considering cosmic variance.
CDM models predict early star formation in massive galaxies, and plenty of such objects out to z~2: No “massive galaxies problem”.

7. Anti-correlation of Halo vs Star Formation
Hierarchical models predict big halos form late, but collapse early. (Formation time ≡ when ½ mass accumulated).
Star formation begins on collapse, so big galaxies form stars early.
This is sometimes called “downsizing” or “anti-hierarchical” behavior, but is actually a natural prediction of CDM.
Nevertheless, still require increased efficiency of SF at early times – happens naturally in simulations, as we shall see.
Van den Bosch etal 04

8. Evolution of Color-Magnitude Relation
Combo-17 (Bell et al 03): Red sequence persists to z~1, with mild evolution.
G6 (with M*-Z relation applied): “Red sequence” too shallow, too blue.
No color gap between early & late-types.
Evolution broadly consistent.
Star formation not shutting off enough in massive galaxies, but is shutting off in small (satellite) galaxies.

9. Birthrates of Simulated Galaxies
Birthrate = tHubblexSFR/M*
Small galaxies (V>-20; green) show a range of birthrates.
Trend to lower birthrates in larger galaxies – good.
Massive galaxies still show significant birthrates at z=0 – bad.
Need truncated SFR in massive galaxies: AGN? (Springel etal 04)

10. How Gas Gets Into Galaxies
Standard scenario (White & Rees 78): Gas shock heats at halo’s virial radius up to Tvir, cools slowly onto disk.
Gas consumption needs to be accelerated at early times relative to this scenario.
Modes of Gas Accretion in Simulations (Keres et al 04):
- Hot Mode: “Standard” accretion. Limited by tcool.
Cold Mode: Gas radiates its potential energy away in line emission at T<<Tvir, and never approaches virial temperature. Limited by tdyn.
Cold mode dominates in small systems (Mvir<fewx1011M), and thus at early times.
Disclaimer: Simulations shown herein are using PTreeSPH, but Dusan has now checked that Gadget-2 gives similar results.

11. Phase Diagram of Accretion
Cold and hot mode distinguished by Tmax, maximum temperature reached by gas until it gets into a galaxy and forms stars.
Figure shows example phase paths of 5 particles from each case (distinction exaggerated).

12. Global Accretion Rate in Hot & Cold Modes
Accretion rate shows two distinct modes.
Cold mode dominates at z>2, when star formation is most vigorous; comparable to hot mode from z~2→0.
Global Tthresh≈2.5x105K; similar results when separated by individual halos' Tvir.
At z~0, 70% of accreted gas never reached halo Tvir.
At z~3, it’s 95%, and ~70% never came within an order of magnitude of Tvir.

13. Accretion Rates vs. Halo Mass
Cold accretion dominates for Mhalo< M, virtually independent of redshift.
 Accretion shock physics governed primarily by halo mass.
This dividing halo mass is analytically predictable.

14. Analytic Analysis of Shock Stability
Birnboim & Dekel (2003): Shocks near virial radius are unstable to radiative cooling for Mhalo < few x 1011M.
In this 1-D model of cosmological halo growth, virial shock is not formed until this L* halo has accreted the bulk of its mass, after z~2.
Similar threshold is seen using Gadget-2; qualitatively similar behavior in AMR.

15. Accretion in a Growing Halo
z=5.5
z=3.2
Left panels: z=5.5, right panels: z=3.2.
Halo grows from M~1011M→1012M, changes from cold → hot mode dominated.
Left shows cold mode gas as green; Right shows hot mode as green.
Cold mode filamentary, extends beyond Rvir; hot mode quasi-spherical within Rvir. Filamentarity enhances cooling.
Density (4Rvir)
Temperature
Temp (Rvir)

16. Merging vs. Smooth Accretion
Galaxies obtain most of their mass by smooth accretion, not merging.
Sub-resolution merging contributes little (<20% overall at z<2).
Globally, SFR follows smooth accretion rate.

17. Infall Velocity
Velocity of cold mode gas within virial radius shows gradual deceleration down to “disk”.
Not free fall  No strong X-ray emission at disk.
green, red, blue: z=3,2,1

18. Conclusions
Simulations naturally produce early stellar mass growth in large galaxies. However, current models do not truncate SF in massive systems as observed, resulting in incorrect color-magnitude relations.
Gas is accreted into galaxies via canonical hot mode, plus an underappreciated cold mode in which the gas always remains much cooler than the halo’s virial temp.
Cold mode dominates globally at high redshifts (z>2), and in smaller halos (Mhalo< few x 1011M) at all times.
Galaxies grow mainly by smoothly accreting gas until z~2, after which merging growth become comparable.

19. SFR vs. Environment
Gomez et al: SFR begins to shut off well outside Rvir, at S ~ 1 gal/Mpc2.
Simulations show identical behavior.
Driven by drop in hot mode accretion rate.

20. Lya Cooling Radiation
Cold mode energy release detectable as Lya emission.
At z~3, 70% of energy is emitted at 104K, with much of it in the Lya line.
Luminosity & structure are very broadly comparable to Steidel’s “Lya blob” at z~3.

21. Accretion Geometry Plays a Role
Cold accretion is generally more filamentary.
Histogram of radius vector dot products shows peak in cold mode accretion at cosine~1.
This enhances cooling rate by increasing the density relative to spherical accretion, and further destabilizing the virial shock.

22. Cumulative Contribution of Hot vs. Cold
Note that even hot mode does not come in exactly at Tvir, but only within a factor of a few of Tvir.

23. Global Star Formation History
- Data from Hopkins (2004) compilation, including extinction correction.
G6 run underpredicts SFR at z<2, and overpredicts(?) at z>3: Early star formation.
Dotted line shows 50% of baryons in stars at z=2.1, or t=3.1 Gyr.
Total stellar mass in G6 run today: 6.3% of Wb.
Low due to limited resolution. G6 underestimates early star formation: Dashed line shows Q5 (10 Mpc/h vol, 200X mass res).
Including metal line cooling may help resolve z<2 discrepancy, by accelerating gas consumption in larger systems that form at z<2.

24. Rest-frame K-band Luminosity Function Evolution: Subaru
Subaru Deep Field Survey (Kashikawa et al 2003): 3.74’, spectroscopic survey to K’vega=23.7 with Subaru.
G6 (solid line) underpredicts LF at z~1 (cosmic variance?), but matches at z~2 & 3 for resolved galaxies.
Go fainter with D5 run (dashed line), showing good overlap with G6 run but probing ~2.5 mags fainter.
D5 shows decent agreement, with a hint of too many faint galaxies.

25. Rest-frame B-band Luminosity Function Evolution
From Subaru Deep Field Survey (Kashikawa et al 2003).
G6 (solid line) shows more rapid rest-frame B-band evolution from z~1→3 than observations.
D5 run shows poorer agreement than in K-band case, with both shape and amplitude mismatches, and a prominent faint galaxy excess (interesting, in light of the strong feedback employed in these models).
Conclusion: Bright end better predicted than faint end, and galaxy stellar masses better predicted than SFR’s.