Travelling back to the beginning of time on a 696 processor supercomputer

The project, dubbed BOOMERANG (Balloon Observations of Millimetric Extragalactic Radiation and Geophysics), obtained the images using an extremely sensitive telescope suspended from a balloon that circumnavigated the Antarctic in late 1998. The balloon carried the telescope at an altitude of almost 37 kilometers (120,000 feet) for 10-1/2 days. The results were published in the April 27 issue of Nature.

Today, the universe is filled with galaxies and clusters of galaxies. But 12-15 billion years ago, following the Big Bang, the universe was very smooth, incredibly hot and dense. The intense heat that filled the embryonic universe is still detectable today as a faint glow of microwave radiation that is visible in all directions. This radiation is known as the cosmic microwave background (CMB).

Since the CMB was first discovered by a ground-based radio telescope in 1965, scientists have eagerly sought to obtain high-resolution images of this radiation. NASA's Cosmic Background Explorer (COBE) satellite discovered the first evidence for structures, or spatial variations, in the CMB in 1991.

The BOOMERANG images are the first to bring the CMB into sharp focus. The images reveal hundreds of complex regions that are visible as tiny variations -- typically only 100-millionths of a degree Celsius (0.0001 C) -- in the temperature of the CMB. The complex patterns visible in the images confirm predictions of the patterns that would result from sound waves racing through the early universe, creating the structures that by now have evolved into giant clusters and super-clusters of galaxies.

"The structures in these images predate the first star or galaxy in the universe," said U.S. team leader Andrew Lange of the California Institute of Technology. "It is an incredible triumph of modern cosmology to have predicted their basic form so accurately."

Italian team leader Paolo deBernardis of the University of Rome La Sapienza added: "It is really exciting to be able to see some of the fundamental structures of the universe in their embryonic state. The light we have detected from them has traveled across the entire universe before reaching us, and we are perfectly able to distinguish it from the light generated in our own galaxy."

To derive the spectrum, project member Borrill used the parallel processing power of NERSC's 696-node Cray T3E supercomputer, employing a software package he developed called MADCAP ("microwave anisotropy dataset computational analysis package"). The calculation required 50,000 hours of processor time and would have taken almost six years to complete if run on a desktop personal computer.

On the Cray T3E, however, processing time over the life of the project totaled less than three weeks. The power spectrum from the BOOMERANG Antarctic flight data is detailed enough to allow the determination of fundamental cosmic parameters to within a few percent.

All CMB experiments seek to determine the state of the universe some 300,000 years after the Big Bang, when the universe cooled enough for protons and electrons to form hydrogen atoms. At that moment, photons were freed from what had been a hot primordial soup of subatomic particles. Ever since that time these energetic photons have been traveling through space, their wavelength now stretched to microwave scale and their frequency reduced to the equivalent of radiation from a black body at only 2.73 degrees Kelvin.

The first step in deriving information from CMB observations is to map the tiny fluctuations in this background radiation -- temperature differences of no more than 1 part in 100,000 which reflect the equally tiny inhomogeneities in the early universe, a time when the universe was in a much simpler state than it is today.

"We basically have to separate the three components of the temperature at each point we look in the sky," Borrill explains. "There is instrument noise. There are foreground sources of microwave radiation, such as dust. And finally there are the intrinsic variations in the temperature of the CMB -- which is what are we are trying to measure."

The observations made as the telescope sweeps across the sky -- 50 million observations for each of 16 channels at four frequencies, in the whole BOOMERANG dataset -- are not independent of one another, and the different kinds of information are related differently.

"Starting with millions of observations, we must separate out the different components," says Borrill. "Each of them is best expressed in its own distinct way -- but to express them jointly we need a common frame. Usually we choose the most manageable, which is the pixel domain -- in other words, we make a map."

Every one of the map's tens of thousands of pixels is made by combining information from hundreds of observations taken at different times throughout the balloon flight. In the resulting map it is easy to identify foreground sources such as quasars or the plane of the galaxy, and dust can be detected by its spectral signature, "which is why we make maps at various frequencies," Borrill says. Comparing observations made at different times improves the signal-to-noise ratio, but this is a computationally expensive operation.

Deriving the power spectrum from the map, the next major step in analysis, is even more challenging. The characteristic power of the microwave background at various angular scales must be determined.

"The idea is to ignore all the other stuff and find just the contribution of the cosmic microwave background," says Andrew Jaffe. "We have to reduce the thousands of pixels in the map to a dozen or so numbers, representing different points along the power spectrum curve, and see how they fit to curves characteristic of different models of the universe."

"The MADCAP program finds the power at each angular scale at each of these points along the curve," Borrill says. "We're asking what the CMB would look like on that patch of sky if the universe had such-and-such a shape and history." When the right curve is found, it allows astrophysicists to distinguish between competing models of the universe's origin, evolution, and present make-up.

Although both map-making and power-spectrum derivation require comparing each pixel in the chosen dataset to every other pixel, in principle this only has to be done once to make a map, whereas it must be done a dozen times or more -- for each chosen point on the curve -- to derive the power spectrum.

"We are at the limit of what is manageable with today's algorithms on today's supercomputers," Borrill says. "It's a job that gets harder with each experiment."

Analysis of the BOOMERANG Antarctic flight data has produced an impressive degree of certainty about some of the most fundamental cosmic parameters. BOOMERANG's power spectrum of the CMB establishes that the universe is flat -- that its geometry is Euclidean, not curved.

Combined with other cosmological measurements, such as studies of distant supernovae by the Supernova Cosmology Project headquartered at Berkeley Lab, the BOOMERANG results support the emerging "concordance model" of a flat universe filled with dark energy -- dark energy that may correspond to the cosmological constant first proposed by Albert Einstein in 1917.

But datasets tens to hundreds of times bigger than BOOMERANG's will be produced by NASA's MAP satellite, to be launched later this year, and the European Space Agency's PLANCK, to be launched in 2007. For these massive datasets, new computational strategies will be necessary.