New Opportunities in Petascale Astronomy. Robert J. Brunner University of Illinois

Transcription

1 New Opportunities in Petascale Astronomy University of Illinois

2 Overview The Challenge New Opportunity: Probabilistic Cosmology New Opportunity: Model-Based Mining New Opportunity: Accelerating Analyses Always keep in mind the Opportunity Costs LCDM Collaborators: Nick Ball, Adam Myers, Brian Wilhite, and Britt Lundgren, Ashley Ross

3 The (Obvious)Challenge LSST: Large Synoptic Survey Telescope Credit: LSST Project Credit: LSST Project

4 LSST Data Challenges Why LSST: Prototypical large astronomy survey I am involved in three working groups Some standard opportunities: How to handle the data volume? How to sweep the data? How to mine the data?

5 LSST Data Challenges Credit: LSST Project

6 The (Not So Obvious) Challenge Petascale Astronomy eduction does not (typically) cover: 1) How to use databases 2) How to find rare events 3) How to find similar events 4) How to statistically characterize events 5) How to do it all quickly! Credit: Penguin Publishers Illinois Pilot Project: Astro Informatics in the classroom

7 So we ask for help! Collaborations are not easy and take real effort to make them work. Credit: Morgan Kaufmann

8 So we ask for help! Credit: W. W. Norton & Company

9 So we ask for help!?

10 Building Bridges Now to find funding!

11 A Brief History Lesson Spectro Photo Credit: SDSS Project Credit: SDSS Project

12 A Brief History Lesson Spectroscopic Data Photometric Data FOV Full Moon One hour, 600 spectra to r ~ 17.7 One minute, multiband photometry to r ~ 22.6

13 The Rest of the Story Photo Limit Spectral Limit Credit: Brunner, Connolly, & Szalay 1999 Factor of 100 increase!

14 Spectroscopic Cosmology Credit: Zehavi et al. 2004

15 Photometric Cosmology Credit: Ross, Percival, & Brunner 2010

16 The Future Spectroscopic Data LSST Photometric Data LAMOST r~20 AAΩ r~21 KAOS on Suburu r~22.5 r~24 One Hour FOV Full Moon

17 Going Forward LSST Final LSST KAOS LAMOST Credit: Brunner, Connolly, & Szalay 1999

18 A New Opportunity: Probabilistic Cosmology Enormous increase in source count Multi-band data to estimate classification and redshift (distance) Not the death of spectroscopy, but entirely new (sub-)field or approach. Note that vast majority of sources (> 99.9%?) will never be followed up spectroscopically!

19 Photometric Redshifts Credit: SDSS Project Credit: ESO

20 Photometric Redshifts Given best-fit spectral model and a corresponding redshift provides a photometric redshift estimation. Note: Spectral redshifts can currently only be determined reliably to I < 24 Credit: Benitez, N 2000, ApJ, 536, 571

21 Probabilistic Redshifts Main Galaxy L.R. Galaxy Credit: Ball, Brunner, Myers et al Quasar NN PDF constructed by sampling from measurement errors

22 Photometric Cosmology Credit: Myers et al z peak = < z spec < 2.2 dn/dz z peak = 2.17 dn/dz z peak = 1.79 dn/dz z Typical analysis only uses peak photometric redshift

23 Probabilistic Cosmology 10 3 f( ) (hmpc -1 ) f 1 = < z spec < 2.2 <f( *)> = 1.26 f 2 = 1.05 f 3 = 1.64 Credit: Myers et al (h -1 Mpc) More sophisticated analysis weights by distance PDF.

24 Probabilistic Cosmology Credit: Lundgren et al Cross-Correlation of Spectroscopic + Photometric samples

25 Another New Opportunity

26 Time Domain Astronomy Credit: LSST Project Good reference site: DotAstro.org

27 Time Domain Astronomy Spectral Credit: Wilhite et al Credit: Wilhite et al Variability SDSS J High S/N Epoch SDSS J Low S/N Epoch QSO rest frame velocity (km/s) QSO rest frame velocity (km/s) Credit: Lundgren et al. 2007

28 A New Opportunity: Model Based Mining Synoptic Astronomy is rapidly growing field Rare or transient events are extremely interesting But the ability to constrain physical models by using synoptic data is an entirely new (sub-)field. Mine Petascale data based on analytic models! Borrow from other fields, including financial markets

29 Model Based Mining Continuous time first order auto-regressive process: CAR(1) dx(t) = 1 τ X(t)dt + σ dt(t)+bdt τ, σ,t>0 MBH R t lc = days M 100R S 3/2 MBH R t orb = days M 100R S α 1 3/2 MBH R t th = yr M 100R S Credit: Kelly et al. 2009

30 Another New Opportunity

31 A New Opportunity: Accelerating Analysis Leverage Commodity Graphics Processing Units to Accelerate Cosmology Codes Cluster manager and Login server Lustre file system servers Credit: Kindratenko & Brunner port Topspin 120 Server InfiniBand switch Netgear Prosafe 24- port Gigabit Ethernet switch HP xw9400 workstation with NVIDIA Quadro Plex Model IV modules 16 cluster compute nodes

32 Opportunity Costs Challenge: Leaving the comfort zone. Challenge: Building Bridges Challenge: Implementing Solutions Computations Faster algorithms Faster hardware Opportunities: New discoveries!

33 Good Luck! Interdisciplinary collaborations are not easy and take real effort to make them work.