|
The Numerical Tokamak
Over the 40 year history of fusion research, scientists have been forced
to test their ideas by casting them in concrete and steel. The complex,
turbulent behavior of plasmas has, for the most part, resisted analysis
by the subtlest theorists and the most powerful computers alike. The only
reliable way to know whether a new reactor concept will work is to go out
and build it, which can cost hundreds of millions -- even billions -- of
dollars. Recently, however, thanks to exponential increases in the speed
of computing, this situation has started to change.
In 1990, UCLA physicist John Dawson calculated that the speed of computers
was increasing so rapidly that within a very few years it would be possible
to accurately simulate and predict the behavior of plasmas in tokamak reactors.
Dawson, who began his career in fusion research in 1956, had first used
a computer to model the behavior of a plasma in 1959. However, limited
as he was by the sluggish pace of early computers, Dawson’s simulated plasmas
(consisting of a small number of positive and negative particles restricted
to movement in a single dimension) bore only a passing resemblance to those
found in nature.
 Thirty years later, Dawson speculated that, given the prospect of computers
100 times faster than the ones with which he had started, it should be
possible to anticipate the detailed behavior of hundreds of millions of
particles moving in the intricately twisted torroidal magnetic field of
a real world tokamak. This is fairly simple in principle and very difficult
in practice. But with advances in parallel computing, whereby many computers
are yoked together to simultaneously work on a single problem, the result
he was looking for appears to be only a few years away. So Dawson and his
colleagues got busy writing computer programs for machines they were confident
would soon be built.
It’s the sheer number of particles that must be tracked in order to
realistically model a plasma that taxes the capacity of the largest computers.
Using the fastest computers currently available, Dawson’s programs can
track the motion of 130 million particles -- enough to simulate the behavior
of a tokamak about two thirds the size of Princeton’s TFTR. The particles
tracked by the computer represent only a tiny fraction of those in a real
tokamak’s plasma, but they are enough to give an accurate picture of its
behavior. “Each one of our computer particles represents a whole bunch
of plasma particles,” Dawson explains. “In some sense it’s like taking
a poll. You don’t ask everybody how they’re voting.”
 The initial tests of the program written by Dawson and his small team
are so far very promising, though challenging: Even the fastest available
computers require days to simulate a few milliseconds of plasma motions
in small tokamaks. Dawson’s goal is to simulate larger machines before
they are built, offering their builders useful guidance and preventing
them from heading down extremely costly blind alleys. Dawson is confident
that computers will soon be fast enough to allow him to reach his goal.
“It’s projected that the speed of supercomputers is going to increase by
a factor of 30 over the next two years,” he says. “That will certainly
be enough.”
Another advantage offered by Dawson’s numerical experiments is that
every detail of the operation can be examined. This is not the case with
real world experiments, where scientists only have access to the limited
amount of information provided by physical sensors. Dawson’s program can,
for example, show the researcher what is happening within the tokamak from
the point of view of a scientist sitting on top of a trapped electron.
This unprecedented access to enormous quantities of data represents
a different sort of challenge. “Somebody has to understand what it all
means,” Dawson points out. “That’s human time, the most important aspect
of the process. The human being is still a critical part of this project.”
Power Plays...
|