Transcript Hu Zhan - Kavli Institute for Astronomy and Astrophysics

Hu Zhan
National Astronomical Observatories
Chinese Academy of Sciences
Large Synoptic Survey Telescope
Peking University Astronomy Symposium
10/17/2010
2
Data = Discovery Space
A new mode of astronomy research: data mining
a new breed: keyboard astronomer
3
Huge discovery space
Key Missions:
1.Dark energy/matter
2.Solar system
3.Optical transients
4.Galactic map
A sample of
• Billions of galaxies
• Millions of SNe
• 105 galaxy
clusters
arXiv:0805.2366
First light ~ 2016/2017
(funding start +4 years)
• 8.4-meter primary
• 10 deg2 FOV
• 3 billion pixels
• 0.2”/pixel
• 0.3–1.1 µm ugrizy
• 15-s exposures
• 8 hours/field total
• 30 TB/night
• 200 PB total
• Median seeing 0.7”
 20,000 deg2
 u 25.8 mag
 g 27.0 mag
 r 27.2 mag
 i 27.0 mag
 z 25.7 mag
 y 24.4 mag
4
Deep Wide Survey: 20,000 square degrees
Northern Ecliptic:
3300 square degrees
~2.1 pairs per lunation
Deep-Drilling:
~100 square degrees
more frequent visits
Galactic Plane:
1700 square degrees to uniform depth of
u: 26.1 g: 26.5 r: 26.1 i: 25.6 z: 24.9 y: 23.5
South Pole:
700 square degrees to a uniform depth of
u: 25.5 g: 26.4 r: 26.0 i: 25.3 z: 25.0 y:23.4
5
6
Data Access Centers
US (2) Chile (1)
45 TFLOPS, 87 PB
Data Archive Center
NCSA, Champaign, IL
100 to 250 TFLOPS, 75 PB
Bandwidth
2.5 Gbps avg, 10 Gbps peak
Base Camp
Cerro Pachon, La
Serena, Chile
25 TFLOPS, 150 TB
Lots of science work has to be done at the data center.
8
Operations Cost: 36.7M 2009 USD
Must secure 1/3 of operations cost
from international and private partners.
9
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
Brookhaven National Laboratory
California Institute of Technology
Carnegie Mellon University
Chile
Columbia University
Cornell University
Drexel University
Google Inc.
Harvard-Smithsonian Center for
Astrophysics
IN2P3 Labs France
Johns Hopkins University
Kavli Institute for Particle Astrophysics and
Cosmology at Stanford University
Las Cumbres Observatory Global Telescope
Network, Inc.
Lawrence Livermore National Laboratory
Los Alamos National Laboratory
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
National Optical Astronomy Observatory
Princeton University
Purdue University
Research Corporation for Science Advancement
Rutgers University
Space Telescope Science Institute
SLAC National Accelerator Laboratory
The Pennsylvania State University
The University of Arizona
University of California, Davis
University of California, Irvine
University of Illinois at Urbana-Champaign
University of Pennsylvania
University of Pittsburgh
University of Washington
Vanderbilt University
10
1.
Solar System: Lynne Jones & Mike Brown
2.
Stellar Populations: Kevin Covey & Knut Olsen
3.
Milky Way Structure: Beth Willman & Marla Geha
4.
Transients/variable stars: Josh Bloom & Lucianne Walkowicz
5.
Galaxies: Harry Ferguson
6.
Active Galactic Nuclei: Niel Brandt
7.
Supernovae: Michael Wood-Vasey & Rick Kessler
8.
Strong gravitational lensing: Phil Marshall
9.
Large-scale Structure/BAO: Hu Zhan & Eric Gawiser
10. Weak lensing: David Wittman & Bhuv Jain
11. Informatics and Statistics: Kirk Borne
More than 300 members.
LSST Science Book：
http://www.lsst.org/lsst/scibook
11
Marana, Arizona, Aug 9-13, 2010
12
Orbital inclination and
ellipticity of 88000
asteroids from SDSS.
The actual color
difference is much
smaller. 37 families are
found.
Through long-term
monitoring, LSST will
provide precise
measurements of
orbital parameters as
well as photometry and
time-domain info of
millions of asteroids,
significantly improving
knowledge about solar
system.
Parker et al. 2008
13
Upper left: simulation of Milky Way
stellar streams and LSST observable
range (main seq. ~ 100 kpc, RR Lyrae ~
300 kpc). Right: SDSS data.
Ivezic et al. 2008
Left: structure of the MW from 2.5M
stars well observed by SDSS. LSST will
reach 200M stars and can study the
MW structure within 100 kpc.
14
SDSS z = 0.1
MUSYC UVR 29.5mag/sq”
Comparable to LSST SB limit
15
16
“The acceleration of the Universe is,
along with dark matter, the observed
phenomenon that most directly
demonstrates that our theories of
fundamental particles and gravity are
either incorrect or incomplete. Most
experts believe that nothing short of a
revolution in our understanding of
fundamental physics will be required to
achieve a full understanding of the
cosmic acceleration.” – the Dark Energy
Task Force, a joint committee to advise
DoE, NASA, & NSF on future
dark energy research.
WIMP? (SUSY: neutralino? gravitino?)
Axion?
HEPAP
Cosmological Constant?
Quintessence?
Modified Gravity?
Back reaction?
Brane world?
Landscape?
G  R  12 Rg   8T
NRC
17
NSTC
NRC
•
•
•
•
•
•
•
•
automated data quality assessment & discovery
scalability of machine learning and mining algorithms
development of grid-enabled parallel mining algorithms
designing a robust system for brokering classifications
multi-resolution methods for exploration
visual data mining algorithms
indexing of multi-attribute multi-dimensional databases
rapid querying of petabyte databases
Many surveys face the same challenges!
18
detection
background
deblending
astrometry
photometry
PSF
shape
classification
time sequence
redshift
extinction
mask
selection
stacking
correlation
19
…
• Transient detection: minimizing ||frame1-kernelframe2||. If we do not
consider how to obtain the kernel, just the convolution part would cost
at least 2× (several)2×3×109 floating point operations (FPOs). A
3GHz CPU could barely process once within the exposure time of 15s,
even if at its theoretical peak performance . The actual demand is
several 104 FPOs per pixel, so LSST will need ~20TFLOPS on site to
process data in real time. The hardware part is fairly easy.
• To process LSST 2000×2000 exposures (3×109 pixels each) once,
even if just one FPO per pixel, it would take a 100TFLOPS computer
120s. With all the complex processing, the LSST data archive center
needs ~200 TFLOPS capacity. The hardware requirement is moderate.
• Desktop I/O: 6GB÷100MB/s = 60s; 8 years for 4×106 exposures.
• Correlation functions of billions of objects…
• Parameter fitting/minimization in high-dimensional space…
The software challenge is daunting!
20
Must understand the system
performance and systematics!
21
LSST will soon simulation ~50TB of images, taking
~2M CPU hours.
22
23
• Compelling sciences, broad impact…
Note: rights to lead key projects will be competed for.
• Discovering the unexpected
Invaluable aspect of surveys: discovery space.
• New trend: data intensive astronomy
• Small investment, 100% data
Limited only by computing & network resources
• Accessible to all levels of expertise
• Highly complementary to future large telescopes
Targets
• Share the same data challenges of other survey projects
Dome A, Chinese Space Station Telescope…
• Anyone in the US: resource-limited free data access;
nonmembers outside the US: no data access.
24
Policy is being reformulated.
Roughly three parts
Construction share:
$465M x GDP/USGDP x Astro fraction by
start.
Operations share:
$420M x GDP/ AllGDP x Astro fraction.
Data access and computation share:
Impact on $16M / year
Early MoU will receive discounts. Must pay operations
share.
25
• Questions?
• Interests?
• Suggestions?
• Resources?
• Actions?
26