How Do Liquid Water Molecules Turn into Ice by Simulation?

How Do Liquid Water Molecules Turn into Ice in Simulation?
How Do Liquid Water Molecules Turn into Ice in Simulation? - Researchers at Princeton University have combined artificial intelligence and quantum mechanics to simulate what happens at the molecular level when water freezes. The result is the most complete simulation yet of the first steps in ice 'nucleation', an important process for climate and weather modeling. Credit: Pablo Piaggi, Princeton University

A team from Princeton University has successfully used artificial intelligence (AI) equations that control the quantum behavior of individual atoms and molecules to detect the early stages of ice formation.

The simulation shows with quantum accuracy how water molecules turn into solid ice.

The computing power that researchers would need was enormous. They used deep neural networks, a type of artificial intelligence, in their study to achieve a degree of accuracy believed to be unattainable. The National Academy of Sciences Journal Proceedings published the study.

Professor Roberto Car of the Princeton Department of Chemistry helped develop a strategy for mimicking molecular behavior based on basic quantum rules more than 35 years ago, and commented that it was like a dream come true. “Our goal at the time was to eventually be able to study systems like this, but it wasn't possible without additional conceptual development, and that development was through an entirely new discipline like artificial intelligence and data science.”

Modeling the early stages of freezing water or nucleating ice can increase the precision of weather and climate forecasts, as well as other processes such as sudden freezing of food.

The new method allows researchers to monitor the activity of hundreds of thousands of atoms over periods of time thousands of times longer than previous studies, even though it's still only fractions of a second.

The method of applying fundamental quantum mechanical laws to predict the physical motions of atoms and molecules was co-invented by Car. How atoms combine with each other to form molecules and how molecules combine with each other to form common objects both occur according to the laws of quantum mechanics.

Ab initio molecular dynamics is an approach that Car and Michele Parrinello, a physicist now working at the Istituto Italiano di Tecnologia in Italy, published in a seminal study in 1985.

Ab initio means from the beginning in Latin.

However, quantum mechanical calculations are difficult and require a lot of processing power. In the 1980s, only a hundred atoms could be simulated by computers on a timescale of several trillionths of a second. Breakthroughs in modern supercomputers and later computation have increased the atomic number and running time of the simulation, but these numbers were still far below the number of atoms needed to witness complex processes such as ice nucleation.

A possible solution was also presented by artificial intelligence. Researchers train a so-called neural network because of its similarity to how the human brain works to recognize a relatively small subset of selected quantum computations. Once taught, the neural network can perform quantum mechanically accurate calculations of forces between atoms that it has never seen before.

Voice recognition and self-driving cars are just two of the usual applications that use this "machine learning" methodology.

Using deep neural networks to represent quantum-mechanical interatomic forces, Princeton graduate student Linfeng Zhang and Princeton mathematics professor Weinan E made a breakthrough in artificial intelligence in molecular modeling in 2018. The method was called "deep potential molecular dynamics". He is a research scientist at the Beijing Big Data Research Institute, earning his PhD.

In this article, Car, postdoctoral researcher Pablo Piaggi, and other researchers used these methods to overcome the difficult task of simulating ice nucleation.

They were able to perform simulations of up to 300.000 atoms using substantially less computational power and utilizing deep potential molecular dynamics for much longer time periods than previously achievable. They ran the simulations on Summit, one of the world's fastest supercomputers, located at Oak Ridge National Laboratory.
According to Professor Pablo Debenedetti, co-author of the new study, this study offers one of the best reviews of ice nucleation.

According to Debenedetti, one of the main variables not measured in weather forecasting models is ice nucleation. “We're seeing extremely strong agreement with the experiments, so this is a pretty important step.

Because of our ability to model very large systems, quantum computations are now possible.

Currently, observations from laboratory experiments are the main source of information used by climate models to determine how fast ice nucleates, but these correlations are only descriptive rather than predictive and only apply to a small set of experimental settings. In contrast, molecular simulations of the type used in this study can generate simulations that predict future events and calculate the amount of ice that will form at high temperature and pressure.

According to Athanassios Panagiotopoulos, Susan Dod Brown Professor of Chemical and Biological Engineering and co-author of the study, “the deep potential methodology used in our study will help fulfill the promise of ab initio molecular dynamics to produce useful predictions of complex phenomena such as chemical reactions and the design of new materials.”

Even with the help of AI, rare events occur on timelines that cannot be given for simulation and require the use of certain approaches to speed them up.
Jack Weis, a PhD student in chemical and biological engineering, helped increase the likelihood of witnessing nucleation by "seeding" microscopic ice crystals into the simulation.

Two hydrogen atoms and one oxygen atom make up a water molecule. The number of electrons surrounding each atom controls how easily atoms can come together to form molecules.

The equation that determines how electrons behave, according to Piaggi, is where we start. “Electrons control nearly every aspect of chemistry, including how atoms interact and form chemical bonds.”

Atoms can live in literally millions of different configurations, according to Car, the director of Chemistry, who receives funding from the U.S. Department of Energy's Office of Science and collaborates with local colleges.
The machine's ability to predict what happens in a finite number of configurations of a small group of atoms to the myriad possibilities of a much larger system is what Car calls "magic."

According to Piaggi, although artificial intelligence methods have been around for some time, academics are hesitant to use them in calculations involving physical systems. “Much of the scientific community was hesitant when machine learning algorithms first gained popularity because these algorithms are a black box and don't understand physics, so why use them?”

According to Piaggi, this mindset has changed dramatically in recent years, not only because algorithms are effective, but because scientists are now applying their understanding of physics to guide the development of machine learning models.

For Car, it's rewarding to see the results of the work started three decades ago. According to Car, data science and applied mathematics were areas where improvement took place. It is very important to have this kind of interdisciplinary connection.

Source: phys.org/news

 

Similar Ads

Be the first to comment

your comment