Simulations with a machine learning model predict a new phase of solid hydrogen

The most prevalent element in the cosmos, hydrogen, may be found in many different compounds on Earth in addition to the cores of stars and the dust that covers the majority of outer space. The fact that hydrogen has just one proton and one electron in each of its atoms makes it the simplest of all the elements, which is reason enough to study it. The University of Illinois Urbana-David Champaign's Ceperley, a professor of physics, believes that as a result, hydrogen is the ideal substance to work with for developing and verifying theories of matter.

Ceperley studies how hydrogen atoms interact and mix to produce various phases of matter, such as solids, liquids, and gases, using computer simulations. Ceperley is also a member of the Illinois Quantum Information Science and Technology Center. Yet since quantum physics is necessary for a complete understanding of these events, it is expensive to simulate quantum mechanics. Ceperley and his associates created a machine learning method to streamline the process, which enables the use of an unprecedented number of atoms in quantum mechanical simulations. Physical Review Letters claimed that their approach discovered a novel variety of high-pressure solid hydrogen that earlier theory and tests had failed to detect.

Ceperley stated that machine learning "worked out to educate us a great lot." "Since we could only handle a limited number of atoms in our earlier simulations, we didn't believe the indications of novel behavior. With the help of our machine learning model, we were able to fully utilize the most precise techniques and determine the true situation."

Even on computers, it is incredibly challenging to fully capture the quantum behavior of hydrogen atoms, which constitute a quantum mechanical system. Modern methods such as quantum Monte Carlo (QMC) may easily simulate hundreds of atoms, but modeling thousands of atoms over extended times is necessary to comprehend large-scale phase dynamics.

Hongwei Niu and Yubo Yang, two former graduate students, created a machine learning model trained on QMC simulations that can accommodate many more atoms than QMC alone in order to make QMC more adaptable. After that, they investigated how the solid phase of hydrogen that occurs at extremely high pressures melts using the model in collaboration with postdoctoral research associate Scott Jensen.

When they spotted anything peculiar in the solid phase, the three of them were scanning various temperatures and pressures to get a full picture. The researchers saw a phase where the molecules became oblong shapes, which Ceperley compared to the shape of an egg. Normally, the molecules in solid hydrogen are near to spherical and form a configuration known as hexagonal close packed; Ceperley likened it to stacking oranges.

Jensen said, "We started with the not very ambitious objective of improving the theory of something we know about. "That was more intriguing than that, which is unfortunate or perhaps lucky. This brand-new habit started to emerge. At high temperatures and pressures, it was indeed the prevailing behavior, which prior theories made no mention of."

The researchers utilized data from density functional theory, a popular method that is less precise than QMC but can accept many more atoms, to train their machine learning model in order to validate their findings. They discovered that the condensed machine learning model accurately captured the outcomes of conventional theory. The researchers came to the conclusion that their large-scale, machine learning-assisted QMC simulations can forecast outcomes and account for impacts in ways that conventional methods cannot.

Conversations between Ceperley's associates and certain experimentalists have begun as a result of this work. Experimental results are constrained due to the difficulty of measuring hydrogen at high pressures. Several teams have decided to revisit the issue and closely examine hydrogen's behavior in severe circumstances as a result of the new forecast.

Ceperley highlighted that a better knowledge of hydrogen under extreme conditions could help us comprehend Jupiter and Saturn, two gaseous planets that are largely composed of hydrogen. The "simplicity" of hydrogen, according to Jensen, makes the chemical interesting to research. We should start with systems that we can attack because we want to learn everything, he advised. "It's important to know that we can deal with hydrogen since it's straightforward."