Data-driven investigation of pressure as a handle over chemical order and disorder in substoichiometric zirconium carbide

Chemical disorder in the solid state can manifest in many forms, including atomic vacancies (missing atoms), site disorder (swapped atoms), and Frenkel defects (misplaced atoms). Disorder can have a significant effect on materials properties, including band gaps, hardness, and magnetic order. However, the experimental control of chemical disorder remains a significant challenge due to the need to overcome the powerful influence of entropy. Motivated by this, we use data-driven methods to investigate the use of extreme pressure as a tool to suppress or enhance disorder. In order to tackle the enormous phase space required for these calculations, we augment expensive first principles calculations with machine learning techniques.

Substoichiometric ZrC1-x exhibits a substantial amount of structural vacancies at the carbon site, impacting a host of bulk properties that depend upon carbon concentration including melting points, mechanical and elastic properties, and superconducting transition temperatures. We used density functional theory augmented with alloy cluster expansion to examine the structure and zero kelvin enthalpy of millions of configurations in substoichiometric zirconium carbide across composition and pressure space. The results of this data-driven approach were examined to study how extreme pressures might be used to access novel ordered and disordered phases of ZrC, many of which have been calculated to be thermodynamically stable yet remain synthetically elusive. We showed that high pressure significantly reduces sub-stoichiometry and drives the system toward fully stoichiometric ZrC. We examined the root of these changes and found that pressure exerts an influence over both the distribution and the abundance of specific nearest-neighbor vacancy pairs. These results suggest that pressure is a powerful tool for the suppression of vacancies in ZrC, offering a new synthetic handle on the bulk properties exhibited in this industrially important class of materials. Our work also highlights the effectiveness of data-driven approaches for investigating systems with a large search space that would otherwise be computationally intractable.

References:

1. Adv. Theory Simul. 2022, 5, 2200439. 10.1002/adts.202200439