1998 Annual Report
High Energy and Nuclear Physics
Cosmic Microwave Background Data AnalysisGeorge Smoot and Julian Borrill, University of California, Berkeley, and Lawrence Berkeley National Laboratory |
|
Research ObjectivesTo develop the novel computational techniques necessary to extract fundamental cosmological parameters from cosmic microwave background (CMB) datasets. Computational ApproachComputing the maximum likelihood of signal to noise is the limiting step in extracting cosmology from CMB observations. Recent successful flights of the MAXIMA and BOOMERanG balloon-borne detectors have produced the largest CMB datasets to date. Both experiments plan to fly again in 1999, to be followed by the MAP and Planck satellites in 2001 and 2007. Over that time the maps produced will grow from tens of thousands to millions of points, and the time it takes to analyze them with the algorithms we currently use will correspondingly increase from hours to millions of years (see table). More efficient algorithms must be developed to process future datasets. AccomplishmentsIn the first year of this project we have developed a full-scale parallel implementation of the map-making and maximum likelihood analysis algorithms on the NERSC T3E. We are now using them to process data from the MAXIMA-1 and BOOMERanG North America flights, providing both insights into the cosmos and benchmark results against which to measure the performance of the new algorithms we will have to develop. SignificanceThe cosmic microwave background (CMB) is the faintest echo of the Big Bang. It is what is left over when all the radiation from astronomical objects is subtracted from what we observe. |
Despite the CMB's stunning uniformity -- isotropic to a few parts in a million -- it is the tiny perturbations in the CMB that contain its unprecedented view of the early universe. Already present before gravitationally bound objects had formed, these temperature differences are an imprint of the primordial density fluctuations that seeded everything from planets to galaxy clusters and superclusters. As such they promise to be an exceptionally powerful discriminant between competing cosmological models. Given a map of the sky temperature, and knowing the statistical properties of noise that went into it, we can now calculate the most likely underlying signal, and by how much it is the most likely. PublicationsGeorge Smoot and Douglas Scott, "The cosmic background radiation," European Physical Journal C 3, 1 (1998); astro-ph/9711069. J. R. Bond, A. H. Jaffe, and L. Knox, "Radical compression of cosmic microwave background data," Astrophysical Journal (submitted, 1998); astro-ph/9808264. Julian Borrill, "Power spectrum estimators for large CMB datasets," Physical Review D (in press, 1998); astro-ph/9712121. |
Computational Resources for CMB Analysis |
|||||
Dataset |
Map Size |
Memory |
Flops |
Serial Time |
T3E Time (Nodes) |
BOOMERanG N. America |
30,000 |
15 GB |
5 x 1015 |
8 months |
40 hours (64) |
MAXIMA-1 |
40,000 |
25 GB |
1016 |
16 months |
40 hours (128) |
MAXIMA-2 |
80,000 |
100 GB |
1017 |
13 years |
4 days (512) |
BOOMERanG Antarctica |
120,000 |
240 GB |
3 x 1017 |
40 years |
6 days (1024) |
MAP |
1,000,000 |
16 TB |
2 x 1020 |
25,000 years |
|
Planck |
10,000,000 |
1600 TB |
2 x 1023 |
25,000,000 years |
|
Computational resources required to analyze CMB datasets using the quadratic estimator algorithm, assuming 20 signal components and 5 iterations on a single (serial) 250 MHz processor and the indicated number of T3E nodes running at the equivalent of 600 MHz. For an Np pixel map, the amount of RAM memory needed scales as Np2, and the number of floating point operations as Np3. |
INDEX | NEXT >> |