A model from Los Alamos studies the deformation of a hypothetical nuclear device when it collides with a steel plate. This shows that simulations play a key role in establishing the reliability of nuclear weapons. The scientists at Los Alamos, Sandia, and Lawrence Livermore labs are turning to huge computers and advanced visualization software to test virtual models. Visualizing and analyzing the data generated by the computational models overwhelm traditional scientific visualization methods. Scientists at Sandia National Laboratories are using computer simulations to test whether the Navy’s W76 nuclear warhead could still function amid blasts of X-rays on a nuclear battlefield. The simulations try to mathematically predict what real X-rays would do as they go through the W76's electronic circuits. Real world tests are run with weaker X-rays and the results are compared with the computer simulations. Data from old underground nuclear tests are also used to validate the computer tests.


NUCLEAR. WEAPONS TESTING has been forced out of the ground by agreement of the world's major powers. It was a big step away from the cold war, but it also created perhaps the biggest challenge yet for computer scientists at the national laboratories.

The Department of En ergy and the national labs are still responsible for guaranteeing the safety and relia bility of the nation's nuclear weaponry. But they have to do the job without live testing or manufacturing support.

On Sept. 24, 1996, President Clinton signed the Comprehensive Test Ban Treaty, calling a halt to underground nuclear testing. The treaty also prohibited new designs and drastically reduced manufacturing of nuclear weapons.

The United States has a smaller nuclear arsenal than in past decades, but because of the test ban treaty, inventory is aging more than it used to. The dev·ices in the nuclear stockpile were built with 20-year shelflives, which many of the weapons have already exceeded.

No longer able to set off samples now and then, the scientists at Los Alamos, Sandia, and Lawrence Livermore labs are turning to huge computers and advanced visualization software to test virtual models of a size that could barely be imagined only a few years ago.

The Accelerated Strategic Computing Initiative, or ASCI, was formed to develop the high-resolution, threedimensional physics modeling needed to evaluate the aging nuclear stockpile and accurately predict how time will affect different components of nuclear weapons. Visualizing and analyzing the data generated by the computational models overwhelms traditional scientific visualization methods.

"Everything is a challenge: hardware, software, network speeds, data storage, and especially veracity," said Robert Shea, weapons physicist at Los Alamos and chairman of the ASCI Visualization Coordinating Group.

One of the first problems that scientists working on the ASCl project had to tackle was finding computers that could handle the large data sets necessary for simulating nuclear blasts. A typical model can contain tens of millions of elements, and over the next couple of years the simulations will grow a thousandfold.

To handle these large models, Sandia National Laboratory in 1997 obtained the first teraflop computer-that is, one that performs more than a trillion floating point operations per second. The Sandia computer, built by Intel, comes close to two teratlops, at 1.8 trillion operations a second. The computer consists of 76 computer cabinets containing 9,072 billion bytes of memory and covers nearly 1,600 square feet. It would take one person using a calculator 57~000 years to calculate a problem that this machine can compute in just one second.

By 1999, Los Alamos and Lawrence Livermore national labs are expected to have as many as four teraflop computers, and the labs expect to have 100-tera£lop capacity sometime in the first decade of the 21 st century.


Seeing Is Believing

A model on the terabyte scale is generally comprehensible only when it can be seen.

Computer scientists at Los Ala.mos use EnSight software from Computational Engineering International to visualize a number of components within a data set, including scalar fi elds, vector fi elds, cell-centered variables, vertex- centered variables, and polygon information.

"Visualization is probably the quickest and easiest way of finding errors in our codes or in the way we have set up problems," said Shea. "We are all accustomed to looking at a table of numbers coming out of a computer model for some simple phenomena and mentally gauging the reliability. As the size and complexity of the calculation increase, we can no longer just look at numbers. The only possible way we have of understanding complicated 3-D calculations is through the use o f sophisticated graphics software."

Scientists load data sets directly into their computers and play them back to study a large amount of data at one time.

In one problem, developers were studying a sphere of material being transported across a ITlesh. Looking at an animation of the process showed that the sphere did not remain exactly spherical as i t moved. T he developers were then able to fix the bug in the code.

While all computer codes and simulations involve compromises, the ASCI proj ect must come as close to petfect as possible. Scientists have used a combination of field tests and computer simulations for a long time, but never before have the sin1ulations played such an important role in ensuring the reliability of nuclear weapons. In the pas t, scientists used computations to ask, "What should we test underground in N evada?" Today, the codes must answer the question, "Will this device petform as expected?"

Scientists at Sandia National Laboratories, for exan1ple, are using computer simulati o ns to tes t wh e th er the Navy's W 76 nuclear warh ead could still function amid blas ts of X-rays on a nuclear battlefield. T he simulations try to mathematically predict what real X-rays would do as they go through the W76's electro ni c circuits. R ealworld tests are run with weaker X-rays and the results are compared with the computer simulations. Data from old under ground nuclea r tes ts is also use d to validate the computer tests.

Scientists must also determine how weapons would react in complex accident scenari os. At Sandia, computer simulations are being use d to determin e what would happen if bombers ca rrying nucl ear weapons crashed. For this project, scientists collect data from every aspect of a plane crash, including initial impac t, damage to stru ctures, severity and spread of fire based on fu el amounts and wind condition, and the effects of fire on materials and obj ects. They then try to simulate what sequences of events could trigger a nuclear warhead.


Nothing Like The Real Thing

Skeptics of the ASCI proj ect qu estion whether computer models can ever truly simulate reality, and how we can trust those simulations without real-world testing. Says Shea, "The real fear in any complicated calculation is th at everything runs well on th e computer, the results look reasonable, and th e answer is wrong."

A cautionary voice is that of Naomi Oreskas, who was a professor at N ew York Unive rsity in 1994 when she wrote in the research j ournal Science, "A model, like a novel, may resonate with nature, but it is not a 'real' thing."

Reached at the Universi ty of California at San Diego, where she teaches now, Oreskas told Mechanical Engineering she is not troubled by computer modeling, but by putting too much faith in it. Conclusions drawn from a simulation must be qualified by an awareness that a model could be based on incomplete or inaccurate information.

Although modeling not tested in the real world may b e uncertain, other considerations weigh in, she added. When reliance on modeling is the result of a ban on nuclear ~esting, "the uncertainties are worth living with; ' she said.

With the help of fas ter supercomputers and· mo re powerful graphics software, computer scientists at the national laboratories are cross- checking results as much as possible. T hey are keeping close ti es to reality by comparing th eir res ults with those from old underground testing and data fro m non-nuclear experiments, such as the low-powered X-rays that Sandia is using.

ASCI scientists have no alternative but to pursue their present course. In a recent visit to Los Alamos, President Clinton summed up their mission: "Of all the remarkable things th at supercomputers will b e able to accomplish, none will be more important than helping to make sure that the world is safe from the threat of nuclear weapons."