One of the outstanding questions of modern cosmology is the origin of primordial perturbations in an otherwise homogeneous and isotropic universe. The most powerful measure of these perturbations is their residual imprint as tiny fluctuations in the temperature of the background radiation that fills the universe - the Cosmic Microwave Background (CMB). Observations by ground-based, balloon-borne, and satellite telescopes are providing increasingly detailed maps of the CMB for us to test our cosmological models against.
As CMB data sets increase from hundreds or thousands of pixels today, to tens of thousands with upcoming balloon flights (e.g. MAXIMA here at Berkeley), to millions of pixels for satellites due to be launched in the next century (NASA's MAP and ESA's Planck Surveyor ), the need for high performance computing for the analysis and visualisation of the CMB becomes greater and greater.
Members of the COMBAT group - a multi-disciplinary team of experimental and theoretical astrophysicists, computer scientists and statisticians based at Berkeley - are using, amongst others, the NERSC supercomputers to address the problems inherent in handling these huge data sets.
Here is a general outline of the algorithms we are using, and benchmarks for recent applications to various data sets on many different computers.