Artificial intelligence methods for probing matter at extreme conditions

Artem R. Oganov

Skolkovo Institute of Science and Technology, 3 Nobel St., 121205 Moscow, Russia

Until mid-2000s it was thought that crystal structures are fundamentally unpredictable. This has changed, and a special role in this was played by our evolutionary method/code USPEX (http://uspex-team.org), which now has over 5000 registered users worldwide. This method can be viewed as a type of artificial intelligence, and routinely allows one to predict stable crystal structures for a given chemical composition], and even to predict all stable compounds formed by given elements. I will discuss some of the most important recent results relevant to high-pressure science:

1. Discovery of novel chemical phenomena: transparent non-metallic allotrope of sodium, counterintuitive novel sodium chlorides, chemical reactivity of helium.

2. Prediction of novel surface compounds, with unexpected similarities to high-pressure compounds.

3. Prediction of new high-temperature superconducting polyhydrides, approaching room-temperature superconductivity.

I will also mention some applications of another type of artificial intelligence: machine learning methods, including recent prediction of phase diagrams of metals (including both solid-solid transitions and melting).