Improbability Drive: Focus on rare actions speeds chemical simulations
By Davide Castelvecchi
From Science News, June 16th, 2007; Vol.171 #24 (p. 372)
In The Hitchhiker’s Guide to the Galaxy, the book by Douglas Adams, a machine made interstellar travel possible by nudging nature toward extremely improbable, but not impossible, events. A new computer-simulation technique promises to calculate chemical-reaction rates 20 times as fast as before by focusing on chains of events that—on the timescales of molecular motion—are very rare but important.
Computational chemistry uses computers as virtual test tubes. For example, by calculating chemical-reaction rates through simulations based on theory, scientists can predict the performance of potential new catalysts before trying to synthesize them, or they can shed light on phenomena such as the misfolding of proteins that’s believed to cause Alzheimer’s and other diseases.
Molecules move on timescales of femtoseconds, or millionths of a billionth of a second, says Titus van Erp of the Catholic University in Leuven, Belgium. The majority of these fleeting motions don’t lead to interesting events, such as molecules hooking up or breaking apart. “It’s like a movie that consists of hours and hours of boring parts,” van Erp says. “An interesting scene lasts a split second, and then it’s over.”
Simulating the entire movie would require a computer to track the motions and states of dissolved molecules as well as surrounding water molecules. The complexity of that task would overwhelm even the world’s most powerful computers.
Scientists have optimized algorithms to ignore the boring parts and to follow only promising action, such as when water molecules align in a way that facilitates a reaction between two dissolved molecules. On the femtosecond scale, such events are rare, but they’re bound to happen if one waits millions or billions of femtoseconds.
But even chains of events that start out promisingly often come to dead ends. Water molecules may begin aligning into a favorable configuration only to disperse. Following such dead ends can waste large amounts of computing time.
In van Erp’s new approach, the computer begins with a random selection of initial molecular arrangements and follows their development in parallel. When one or more of those scenarios produces a favorable outcome—for example, if the molecules that are supposed to react move significantly closer—the simulation starts over. It throws away the uninteresting scenarios and begins again with several variations of the interesting ones.
Crucially, the algorithm keeps track of what fraction of scenarios it throws away at each stage. That enables it to estimate the reaction rate by calculating what proportion of initial states would end with the molecules reacting.
Van Erp says that his method can reduce the waste of computing time by at least an order of magnitude. He successfully simulated the breakdown of DNA 20 times as fast as existing algorithms can. He reports that feat in an upcoming issue of Physical Review Letters.
“It’s a very nice piece of work,” says Juan de Pablo of the University of Wisconsin–Madison. He adds that calculating reaction rates using simulations has been a challenge for decades and that speeding simulations by an order of magnitude is an important step.