Recreating the Early Universe
Over the past several years, we have been treated
to astounding pictures of the distant universe from sophisticated telescopes.
Alongside images of galaxies, supernovae, and possible evidence of new
planets, there often appear artistic renderings of the heavens, attempts
to recreate events that took place billions of years ago.
Until recently, cosmology--the study of the origin and nature of the
universe--was largely a science of speculation, without the tools or
data to put theories to the test. As more of this data is made visible
through advances in observational technology, that which we cannot see
also comes tantalizingly closer.
WHAT TELESCOPES CANNOT SEE
Today, the researchers who make up the Grand Challenge Cosmology
Consortium (GC3) harness the power of supercomputers to look at the birth
and infancy of the universe, starting from the Big Bang, the cosmic explosion
which is believed to have started it all about 15 billion years ago. GC3
is a collaboration between cosmologists, astrophysicists, and computer scientists
studying the formation of large-scale cosmological structure.
Now in the third year of a five-year NSF grant administered by the Division of
Astronomical Sciences, the consortium is one of several groups participating
in NSF's High Performance Computing and Communications (HPCC) "Grand Challenge" initiative
Led by astrophysicist Jeremiah Ostriker of Princeton University, GC3 consists
of seven principal investigators at six universities and over 30 professional
staff members, postdoctoral research associates, and graduate student assistants.
Consortium members use high-performance computers at all four NSF-supported National
Supercomputing Centers to build three-dimensional models of galaxy formation,
recreating the early universe as predicted by theory and glimpsed by the Hubble
Space Telescope late last year. Along the way, they are testing competing theories
to explain what our universe is made of and how it has evolved.
Supercomputers are the modern variant of the pencil and paper historically used
to express astronomical observations in terms of physical laws and mathematical
To test theoretical models, modern cosmologists simulate, numerically, scenarios
in the early universe that could have produced what we see and measure now. Massively
parallel and scalable high-performance computers are the only machines capable
of performing the billions of calculations required to solve the equations explaining
the complex interactions of energy and matter occurring over vast ranges of space
But modeling is more than plugging in numbers. Much of the GC3 effort is directed
toward developing new algorithms, programming models, and software technology
that uses supercomputers on both large and small scales. This "multiscale" aspect
of the GC3 endeavor is what makes it an NSF "Grand Challenge" computational project.
THE NATURE OF DARK MATTER
GC3 cosmologist Michael Norman at the National Center for Supercomputing Applications
(NCSA) at the University of Illinois studies the nature of dark matter and its
role in galaxy cluster formation. Scientists estimate that 90% of the matter
in the universe does not emit enough light to be observed, but they know it exists
because of its gravitational effects on stars and galaxies. This invisible mass
is a key component of current versions of the Big Bang theory and is crucial
to understanding how matter clusters under gravity.
But cosmologists do not agree on the form dark matter takes, whether it is
made of "cold" matter such as dead stars or heavy exotic particles, or "hot" and
lightweight undetectable subatomic particles. Astrophysicists know that hot
dark matter alone could not have produced galaxies early enough to account
for some of the largest superclusters of galaxies that are visible today. On
the other hand, a universe filled with only cold dark matter would produce
too many superclusters too fast.
According to Norman, one promising model combines cold and hot dark matter. The
trick is to find the right mix, let it evolve in cyberspace, and see if it brings
us to where we are now.
To do this, Norman and his team imagine space broken into a grid of cells. They
determine how the matter in each of these cells will evolve, by integrating equations
for dark matter, gases, temperature, and gravity in an expanding universe. The
grid is reproduced as a three-dimensional image of a large section of the universe,
340 million light years wide. The simulation begins about 1 billion years after
the Big Bang, when the universe was a much hotter and denser place than it is
today. The grid is subdivided into over a hundred million cubes, reflecting the
changes and fluctuations occurring as the universe cools and expands. The result
is a map showing the behavior of gases and clusters of matter.
TOWARD BETTER PREDICTIONS
In 1994, Norman and his NCSA colleague, Gregory Bryan, carried
out the largest cosmological simulation to date, creating a universe composed
of 60 percent cold dark matter, 30 percent hot, and 10 percent ordinary
matter. The simulation, conducted on NCSA's Connection Machine-5 supercomputer,
took over 30 hours to complete and produced a model that closely resembled
recent observations made by the orbiting x-ray satellite ROSAT. While the
simulation accurately predicted the number and arrangement of galaxy clusters,
it did not capture exactly the measurable ratio of gas to dark matter.
GC3 researchers need more refined programming codes to improve the model's resolution
and allow them to zoom into the individual cubes on the grid. This would also
enable the researchers to dedicate more computational energy to analyzing the
crucial areas where galaxy formation begins and experiment with different particle-based
equations aimed at capturing the behavior of dark matter. Dark matter is a very
fertile field for cosmologists. "Everyone is motivated to find out what it is," says
Norman, "but there is nothing definitive yet."
At the same time that supercomputers provide cosmologists with the tools to tinker
with celestial ingredients, cosmological simulations also help push computer
hardware and software systems to new limits. GC3 members work closely with the
staffs of the supercomputer centers as well as the computer vendors to develop
efficient strategies for storing, analyzing, and visualizing vast amounts of
With support from GC3, Joel Welling at the Pittsburgh Supercomputing Center has
written a more efficient program to visualize three- dimensional data like that
generated by Norman's simulations. The program -- the VFleet Distributed Volume
Renderer -- can be run in a parallel environment, on either a workstation cluster
station or supercomputer, and is able to handle large datasets at fast speeds.
MAPPING THE EARLY UNIVERSE
GC3 theoretical physicist Edmund Bertschinger of MIT is studying the evolution
of matter and energy in the early universe starting only a few minutes after
the Big Bang. At that time the universe was filled with hot, dense matter that
glowed from its heat like the interior of the Sun. Bertschinger and MIT postdoctoral
researcher Paul Bode numerically integrate the equations governing radiation,
ordinary and dark matter, and Einstein's general theory for gravity. Their simulation
produces a map of the glowing radiation left over from the Big Bang, which can
be compared with maps of the cosmic microwave background radiation made recently
by the Cosmic Background Explorer satellite, COBE.
Bertschinger and computational physicist Robert Ferrell also study the formation
of galaxies and clusters using models like those of Norman and Bryan. Their
simulations omit the gas but achieve higher spatial resolution for the dark
matter. In order to run their calculations on parallel supercomputers, Ferrell
and Bertschinger have developed an algorithm that has since been used to model
marine oil spills, replacing cosmological equations with those that render
the behavior of ocean currents, wind, and globules of oil on the surface of
Although much of the effort now is concentrated on exploring the numerical calculations
and expanding high-performance applications and software, some of the results
will soon be coming to the big screen. Several GC3 members are creating visualizations
of their numerical simulations for "Cosmic Voyage," an IMAX feature film about
the universe and debuting at the Smithsonian National Air and Space Museum in
mid-1996. Frank Summers, a postdoctoral research associate at Princeton, is coordinating
the GC3 sequence in collaboration with the National Supercomputing Metacenter
and computer graphic artists at PIXAR (the studio responsible for the movie "Toy
Story"). Audiences will travel through the gravitational collapse of structure
shortly after the Big Bang, the formation of galaxies, and the collision of two
And what about the future? Can computer modeling also be used to explore questions
about where we are going? Earlier this year members of the GC3 team linked three
or four parallel computers to simulate the collision between our galaxy, the
Milky Way, and the Andromeda Galaxy, one of our larger neighbors. The preliminary
findings indicate that we won't be able to compare the computer model with the
actual event for another three billion years.