New Capabilities of the XtalOpt Evolutionary Algorithm Code
The expanding research field that is concerned with the a priori prediction and design of materials has led to successful syntheses of phases that were first proposed in theoretical calculations, including the fascinating Fm-3m LaH10 phase, which is consistent with a compound with measured Tc values up to 280 K at 200 GPa. With advances in materials synthesis, computational techniques such as the XtalOpt evolutionary algorithm,designed to predict novel and intriguing structures, have similarly become more adept. New work in the group of CDAC Academic Partner Eva Zurek at the University at Buffalo has resulted in the implementation of new structure prediction tools in the XtalOpt evolutionary algorithm for crystal structure prediction. Forthcoming papers from the Zurek group detail the use of the XtalOpt methodology to determine stable structures at high pressure in the boron and Li-H-F systems.
XtalOpt starts with an initial set of randomly generated structures, which are encouraged towards local order using the newly implemented mitosis and randSpg techniques, respectively breaking down large unit cells into supercells of smaller identical components or enforcing user-defined space groups. Increased local order speeds convergence towards low-enthalpy structures, while the search can be further tailored by the imposition of custom interatomic distance cutoffs and including molecular units in the initial geometries. A search can consider multiple formula units of a given stoichiometry at once, with the user choosing whether and when to allow crossovers between structures with different numbers of formula units. Duplicate structures are removed after detection with the XtalComp algorithm, which directly maps structures onto one another following reduction into a standard orientation.
Finally, stable and superhard materials can be directly targeted, with hardness values calculated by a machine learning model based on the Automatic FLOW (AFLOW) database, which is incorporated into a modified fitness function used to select structures for further procreation. Several new metastable and superhard allotropes of carbon have been identified using this method. All of the search options, along with the progress of the search, can be monitored on-the-fly via a GUI implementation, which shows lists of enthalpies and space groups as well as plots of enthalpy against the generations of structures.
Falls, Z. et al., The XtalOpt evolutionary algorithm for crystal structure prediction. Journal of Physical Chemistry C 125, 1601-1620 (2021).
Out of randomness, order. From randomly generated crystal structures, the XtalOpt evolutionary algorithm searches the potential energy surface of a system, permuting atoms and deforming crystal lattices to find the most stable arrangement of atoms. Cover image created by CDAC Postdoctoral Associate Katarina Hilleke.