Computer Models

March weather is notoriously fickle. That's why weather forecasters keep an eye peeled for portents of snowstorms like those that periodically paralyze the East Coast--including 1993's "storm of the century," which dumped three feet of snow on some states.

In early March, 2001, weather computers were warning of a repeat storm of the century, with even deeper snow. But the fateful day, Mar. 6, came and went with less than a foot of snow in urban areas.

What went wrong? Blame it partly on "shortcuts" in computer models. Today's supercomputers can't crank through all the calculations required to generate better forecasts--at least not before the bad weather hits.

So simulations use clever algorithms that gloss over certain details. Avoiding such shortcuts "won't happen any time soon," says Jimy Dudhia, a physicist at the National Center for Atmospheric Research. Doubling the detail in three-dimensional weather models, he explains, would take eight times more number-crunching power. Even then, the model would be too coarse to predict tornados accurately.

Many computer models used in biology, chemistry, and physics also rely on shortcuts. Otherwise, answers might not pop out for months or even years. For example, atom-scale simulations based on all the gory details of quantum physics are impractical for more than 300 atoms. Michael L. Klein, director of the University of Pennsylvania's Center for Molecular Modeling, struggles to run simulations involving just 200 atoms because even a few picoseconds' worth of a routine chemical reaction take an hour to model, even on a top supercomputer.

So, when computer simulations promise quick and faithful imitations of complex real-world phenomena, experts should be sure they understand what's built into the shortcuts. Being too trusting could lead to fizzles like the storm of the century that wasn't.

By Otis Port

    Before it's here, it's on the Bloomberg Terminal.