A new study from Tel Aviv University (TAU) has found that “stopping and restarting” sampling in a chemical simulation can facilitate faster results, extending a common practice in information technology.

The research was led by PhD student Ofir Blumer in collaboration with Professor Shlomi Reuveni and Dr. Barak Hirshberg from TAU’s Sackler School of Chemistry. The study was published on January 4, 2024, in the journal Nature Communications.

The researchers explain that molecular dynamics simulations are like a virtual microscope. They track the motion of all atoms in chemical, physical, and biological systems such as proteins, liquids, and crystals. They provide insights into various processes and have different technological applications, including drug design. But these simulations are limited to processes slower than one-millionth of a second and can’t describe slower processes such as protein folding and crystal nucleation. This limitation, known as the timescale problem, is a great challenge in the field.

“In our new study we show that the timescale problem can be overcome by stochastic resetting of the simulations,” Blumer says. “It seems counterintuitive at first glance – how can the simulations finish faster when they’re restarted? But it turns out that reaction times vary considerably between simulations. In some simulations, reactions occur rapidly, but other simulations get lost in intermediate states for long periods. Resetting prevents the simulations from getting stuck in such intermediate states and shortens the average simulation time.”

The researchers also combined stochastic resetting with Metadynamics, a popular method to expedite the simulations of slow chemical processes. The combination allows greater acceleration than either method when used separately. Moreover, Metadynamics relies on prior knowledge: the reaction coordinates must be known to expedite the simulation. The combination of Metadynamics with resetting reduces the dependency on prior knowledge significantly, saving time for practitioners of the method.

Finally, the researchers showed that the combination provides more accurate predictions of the rate of slow processes. The combined method was used to enhance simulations of a protein folding in water successfully and it is expected to be applied to more systems in the future.