High Redshift Galaxies Ever Increasing Numbers HPComputers in
High Redshift Galaxies (Ever Increasing Numbers)
(HP)Computers in Astronomy: - Handling the (nightly) Terabytes of data - Data pipelines - Analysis of images and catalogues - Theory: - Space Plasmas - Stars (+ planetary systems) - Formation and evolution of galaxies - Clusters of Galaxies - Large-scale Structure (galaxy statistics)
Talk Overview 1) What is the correlation function and what can we learn from it? 2) The observational side 3) The theory side ( i. e. why HPC)
Early statistical measures of galaxy populations: Hubble, 1934 – Distr. of counts, N, are log gaussian. Bok (1934), Mowbray (1938) – Variance in N larger than expected. Rubin, Zwicky, Limber 1950 s – Statistical methods related to the (auto-)correlation function. Neyman, Scott 1962 – Used auto-correlation function on the Lick catalogue. By the 70 s, computers made such calculations a routine task.
Correlation Function Excess probability, above random of finding 2 objects in solid angle elements, dΩ separated by θ. . w(θ) = DD(θ) / DR(θ) – 1
W = DD / DR – 1 ‘Natural estimator’. -Simplest (cheapest). -Suffers bias due to edges. Hamilton (1993), Landy & Szalay (1993) W = 4*DD*RR / DR 2 - 1 W = (DD – 2*DR + RR)/RR Reduces errors to poisson level* Robust against edge effects.
Biased halo formation (Dark) Matter distribution: Power spectrum. Highest density peaks collapse earliest. Peaks are clustered. Halos formed at these peaks merge. Biggest halos are those formed at highest peaks. Biggest halos most strongly clustered.
Red galaxies found in clusters, blue in the 'field'. . . and luminosity segregated. Zehavi et al. 2005
Positions in 3 -d known…. …. halo mass!
2 terms: 1 -halo term (small scales), slope same as density slope. 2 -halo term (large scales), slope = -0. 8 1 -halo term allows estimation of merging rate.
Great! So all we need to know is 3 -d positions! EXPENSIVE!! - Need redshift - environment effects (finger of god). Other method: - estimate the redshift distribution (colour selection, Photometric redshifts) - Deproject (Limber’s inversion) =>>> r 0
Timing I/O N DD calc. N 2 / 2 Random cat. Constr. N*10 DR calc. 10*N 2 RR calc. (10*N)2 Deprojection. - Error estimation. 100*N 2 I/O - Scalability is very good!
Mid-talk summary • From counting pairs of galaxies we can estimate: • - Typical halo mass. • - Merger rate. • - What its halo will become by redshift 0. • (- and what its neighbours will be like. ) • - When its host halo's progenitor formed. • For any given galaxy population for which we have an estimate of its redshift distribution.
The need for deep infrared surveys Optical surveys sample rest-frame UV at high-z 1. Biased against high-z galaxies obscured by dust 2. Bias against high-z galaxies with old stellar populations 3. Provide poor estimate of stellar mass Deep IR surveys vital for a complete census at z>1
The UKIDSS Ultra-Deep Survey http: //www. nottingham. ac. uk/astronomy/UDS 0. 88 deg. DR 1: KAB=23. 5, JAB=23. 6 (85 hours) World-wide public in january 2008 DR 3: KAB=23. 7, HAB=23. 4, JAB=23. 6 (120 hours) ESO public in december 2007 Final depth: KAB=25, HAB=24. 7, JAB=24. 7 (200 nights) Another 4 years of data to come… …plus new spectroscopic ESO survey
http: //www. nottingham. ac. uk/astronomy/UDS 02: 17: 48, -05: 45
Bz. K selection (Daddi 2004) Efficiently selects objects between redshift 1. 4 and 2. 5. 50, 000 objects. 650 passive (p. Bz. K) 11, 000 starforming (s. Bz. K)
r 0 values: p. Bz. K – 17. 5 h-1 Mpc; s. Bz. K – 8. 3 h-1 Mpc. r 0 value for p. Bz. K's implies a halo mass in excess of 1014 Msun. Also, note the large excess on small scales for the s. Bz. K's – suggests a lot of merging by z = 0.
Comparing with models. . . Can use: Semi Analytic models N-body simulations N-body + S. A. Full gas + DM sims. (in order of increasing computational cost. )
S. A. Results From Millennium simulation + S. A. model, 6 'lightcones' have been extracted ~1 million objects per lightcone. Treated in the same way as the 'real' data. 7, 000 p. Bz. Ks. Looks fairly good. 1 -halo term shows 'over-merging'?
The future… Star formation: - highly sensitive to resolution. - even M. S. isn’t sufficient! - Where can we turn? Re-simulation: - choose a few representative volumes, - Trace the particles back, - Split those particles into many smaller ones. =>>>> GIMIC
Conclusions Reached the limit of what can be done on a desktop! Even simple codes require HPC to handle modern datasets. We need to run specifically designed simulations to model how galaxies formed and evolved in the distant universe. (High redshift)
- Slides: 22