A team based at Princeton University has accurately simulated the early stages of ice formation by applying artificial intelligence (AI) to solve the equations which govern the quantum behavior of individual atoms and molecules.
The resulting simulation describes how water molecules transform into solid ice with quantum precision. This level of precision, once considered unattainable due to the computing power required, became possible when researchers incorporated deep neural networks, a form of artificial intelligence, into their methods. The study was published in the journal Proceedings of the National Academy of Sciences.
“In a way, it’s like a dream come true,” said Roberto Motor vehicle, Professor Ralph W. *31 Princeton Dornte in Chemistry, who co-pioneered the behavioral simulation approach molecules based on underlying quantum laws more than 31 years ago.. “Our hope then was that we could eventually study systems like this, but that was not likely without further conceptual development, and that development came through a completely different field, that of artificial intelligence and data science.”
The ability to model the initial stages of water freezing, a process called ice nucleation, could improve the accuracy of weather and climate modeling as well as other treatments like freezing fast food.
The new approach allows researchers to track the activity of hundreds of thousands of atoms over periods of time that are thousands of times longer, although fractions of a second, than in early studies.
Car co-invented the approach of using the underlying laws of quantum mechanics to predict the physical movements of atoms and molecules. The laws of quantum mechanics dictate how atoms bond together to form molecules and how molecules come together to form everyday objects.
Car or truck et Michele Parrinello, a physicist now at the Istituto Italiano di Tecnologia in Italy, published their approach, known as “ab initio” molecular dynamics (Latin for “from the beginning”).
But quantum mechanical calculations are complex and consume a lot of computing power. In the 1980 years, computers could only simulate a hundred atoms over times of a few trillionths of a second. Later advances in computing and the advent of modern supercomputers increased the number of atoms and the simulation time, but the result was well below the number of atoms needed to observe complex processes such as nucleation. ice.
AI provided an interesting potential remedy. Researchers train a neural network, named for its similarities to how the human brain works, to recognize a relatively small number of selected quantum computations. Once trained, the neural network can calculate forces between atoms it has never seen before with quantum mechanical precision. This “machine learning” approach is already used in everyday apps such as voice recognition and autonomous vehicles.
In the case of AI applied to molecular modeling , a major contribution is rental in 1980 when Princeton graduate student Linfeng Zhang, together with Automobile and Princeton mathematics professor Weinan E, found a a means of applying deep neural networks to modeling the interatomic forces of quantum mechanics. Zhang, who earned her Ph.D. in 2020 and is now a research scientist at the Big Data Research Institute in Beijing, called the “deep potential molecular dynamics” approach.
In the current paper, Car and postdoctoral researcher Pablo Piaggi and colleagues applied these techniques to the challenge of simulating the nucleation of ice. Using deep potential molecular dynamics, they were able to run simulations of up to 300 000 atoms using much less computing power, for much longer durations than before. They performed the simulations on Summit, one of the world’s fastest supercomputers, located at the Oak Ridge Countrywide Laboratory.
This work provides one of the best studies on the nucleation of ice, said Pablo Debenedetti, Princeton’s dean for research and professor of engineering and applied science of advertising 31, and co-author of the new study.
“The nucleation of ice is the ‘one of the main unknown quantities in weather forecasting models,’ Debenedetti said. “This is quite a significant step forward as we see very good agreement with the experiments. We were able to simulate very large systems, which was previously unthinkable for quantum calculations.”
Currently, climate models obtain estimates of the nucleation rate of ice mainly from observations made during laboratory experiments, but these correlations are descriptive, not predictive, and are valid over a limited range of experimental conditions. In contrast, molecular simulations of the variety performed in this study can produce simulations that predict future conditions and can estimate ice formation under extreme conditions of temperature and pressure, such as on other planets.
“The deep potential methodology used in our study will help realize the promise of ab initio molecular dynamics to produce valuable predictions of complex phenomena, such as chemical reactions and the design of new materials” , said Athanassios Panagiotopoulos, Professor of Chemistry Susan Dod Brown. and biological engineering and co-author of the study.
“The fact that we study very complex phenomena from the fundamental laws of nature, for me, it is very exciting,” said Piaggi, the study’s first author and a postdoctoral research associate in chemistry at Princeton. Piaggi obtained his doctorate. working with Parrinello on developing new techniques to study rare events, such as nucleation, using computer simulation. Rare events occur on timescales longer than the simulation times that can be tuned, even with the help of AI, and specialized techniques are required to speed them up.
Jack Weis, a graduate student in chemical and biological engineering, helped increase the likelihood of observing nucleation by “seeding” tiny ice crystals into the simulation. “The goal of seeding is to increase the likelihood that the water will form ice crystals during the simulation, which allows us to measure the rate of nucleation,” Weis said, advised by Debenedetti and Panagiotopoulos.
Water molecules are made up of two hydrogen atoms and one oxygen atom. The electrons around each atom determine how the atoms can bond together to form molecules.
“We start with the equation that describes the behavior of electrons”, said Piaggi. “Electrons determine how atoms interact, how they form chemical bonds, and virtually all of chemistry.”
Atoms can exist in literally hundreds of thousands of different arrangements, said Motor Vehicle, which is director of the Chemistry in Remedy and Interfaces center, funded by the US Division of Energy Place of work of Science and including regional universities.
“The magic is that due to certain physical principles, the machine is able to extrapolate what is happening in a relatively small number of configurations from a small assortment of atoms to countless preparations of a much larger system,” said Motor vehicle.
Although AI approaches have been available for a few years, researchers have been cautious about applying them to physical system calculations, Piaggi said. “When machine learning algorithms started to become popular, much of the scientific community was skeptical because these algorithms are a black box. Machine learning algorithms don’t know anything about physics, so why would we use them?”
Over the past two years, however, there has been a significant shift in this frame of mind, Piaggi said, not only because the algorithms work, but also because researchers use their knowledge of physics to inform machine learning models.
For Motor vehicle, it is satisfying to see the work started three decades ago come to fruition. “The development came through something that was developed in a different field, that of data science and applied mathematics,” Automobile said. “Having this style of cross-interaction between different domains is very essential.”
This work was supported by the US Department of Energy (DE-Grant 1950 SC0019394) and used the resources of the Oak Ridge Management Computing Facility (grant DE-AC05-00GOLD22725) at the Oak Ridge National Laboratory. The simulations were performed largely using the resources of Princeton Analysis Computing at Princeton University. Pablo Piaggi was supported by an Early Postdoc.Mobility grant from the Swiss National Science Foundation.