Artificial intelligence explains hydrogen's behavior on giant planets

The giant planets in our solar system are made mainly of hydrogen, mostly in a liquid state. Near the planets’ surface, hydrogen exists in an insulating, molecular form – H₂ – but closer to the center, it takes on a metallic form where individual atoms can move around freely. Professor Michele Ceriotti, who heads the Laboratory of Computational Science and Modelling (COSMO) within EPFL’s School of Engineering, along with colleagues from the University of Cambridge and IBM Zurich, have used computer simulations to understand the nature of this elusive transition. “Hydrogen is the simplest element on the periodic table: it’s made of one proton and one electron. That’s one reason why scientists study it so much. What makes this phenomenon on giant planets fairly unique – and interesting – is that the transition is between two forms of a liquid state, and not from a liquid to a gaseous or solid state,” says Ceriotti.

An extremely complicated system to model

This transition has been the focus of an intense research effort – many research groups have tried to gain a better understanding by replicating it in a laboratory. “But that’s really difficult because you have to create the same conditions here on Earth as those on Saturn and Jupiter, which means achieving pressures that are around a million times greater than that of the Earth’s atmosphere, and then analyzing samples that have been exposed to those pressures. As you can imagine, examining compounds under those conditions isn’t easy. Various studies have been conducted, but their findings are often controversial because they differ substantially from each other,” says Ceriotti. So rather than trying to recreate giant planets’ atmospheric conditions in a lab, his team used highly precise computer models to simulate them digitally. Previous attempts to do so involved solving on the fly equations describing the quantum mechanical behavior of fluid hydrogen. The complexity of such equations limited the scope of such simulations to just a few atoms at a time and the time scale to less than one billionth of a second.

Saving time

Using their expertise in computer simulation and machine-learning-based models, the team was able to expand the scope and time scale of their simulations and gain insight into the mechanisms at work in the hydrogen transition. That required using very powerful computers. “It was a considerable investment in terms of time and energy. Running our models – which we developed using machine learning – took several weeks on EPFL’s supercomputers. But if we’d tried to run models developed the conventional way, it would’ve taken hundreds of millions of years,” says Ceriotti.

A smooth transition

Based on their results, the research team came up with a possible explanation for why previous studies produced such different results, which might be traced to the interpretation of experimental observations, and the nature of the liquid-liquid transition. “Until now, all the studies that tried to model hydrogen’s behavior on giant planets assumed that the transition from an insulating to a metallic form was clear-cut. But our simulations indicate that the transition takes place gradually. As a consequence, the exact moment where the transition occurs depends on the choice of criteria, and the observable being monitored – which could explain the dependence on the details of the experiments.” says Ceriotti. “Now that we’ve modeled the ‘ideal’ system, our next step will be to study more realistic – and therefore more complex – conditions, where hydrogen is mixed with helium and other compounds in the planets’ atmospheres.” Thanks to computer simulations and machine learning we’re now one step closer to unlocking the mystery of the giant gaseous planets in our solar system, and potentially of exoplanets as well.