Machine Learning Enabled Atomistic Simulation of Iron at Extreme Pressure

PI Robert Rudd , Lawrence Livermore National Laboratory
Co-PI Federica Coppari, Lawrence Livermore National Laboratory
Rudd graphic alcc

Results from preliminary MD <011> shock-ramp simulation using an inexpensive potential. The images are transverse sections of the simulation analyzed with polyhedral template matching to show the grain structure and phases. Red indicates hcp lattice; green, fcc.

Project Summary

The simulations in this project will provide a first-ever view into the changes to the microstructure of iron as it goes through phase transformation and extreme deformation, detail at the atomic level directly comparable to the experimental X-ray absorption data and allowing a deep understanding of iron’s behavior under these extraordinary conditions.

Project Description

Iron is at the core of our planet and is thought to be at the core of countless exoplanets. Its behavior at the extreme pressures and temperatures of the planetary core determines much of the structure of the inner Earth. Its properties determine the size of the inner core, that region where the otherwise molten core is under such great pressure from gravity that the iron is solid. The transition from liquid to solid core is understood to affect the convective flows that give rise to the dynamo, generating the magnetic field that shields us from ionizing radiation from space, allowing life to form and thrive. Laboratory experiments that probe the properties of iron under such extreme conditions are difficult, but important strides have recently been made. Dynamic techniques, such as ramp compression, drive the iron to extreme conditions for a brief time, a few hundredths of a millionth of a second, and probe its properties. A direct determination of the temperature during that time has been elusive. Exciting new experiments using X-ray absorption measurements during laser-driven compression have produced excellent data but the analysis is incomplete. Additional atomic-scale information is needed to fully understand and characterize the competing effects of temperature and changes in microstructure induced by the rapid dynamic compression. Here we use advanced, machine-learned interatomic potentials in large molecular dynamics simulations of the iron ramp compression to provide that essential guidance. The machine-learned potential is very accurate, providing the accuracy of quantum-mechanical calculations at greatly reduced computational expense, allowing the multi-million atom simulations that are needed. These simulations still require extraordinary computational power, only possible with supercomputing resources. 

The simulations in this project will provide a first-ever view into the changes to the microstructure of iron as it goes through phase transformation and extreme deformation, detail at the atomic level directly comparable to the experimental X-ray absorption data and allowing a deep understanding of iron’s behavior under these extraordinary conditions. The outcome of the simulations will also more broadly impact the Department of Energy X-ray absorption experiments on other materials ramp compressed to high pressure, informing investigations of high-energy-density science and the exotic physics that occurs under extreme pressure.

Allocations