Novel materials discovery is strongly assisted by computational capabilities caused by a swift evolution of various approaches and algorithms. Particular attention is paid to those targeting the prediction of new stable crystalline materials. Existing evolutionary algorithms are, in principle, capable of varying the chemical composition generating novel compounds, and selecting the most prominent candidates referring to the energy landscape. Nevertheless, in practice, their computational cost remains excessive and they are scarcely applicable in cases of a complex crystal structure modeling. Since the quasicrystals’ (QC) discovery, the interest in aperiodic enigma drastically increased from the point of view of fundamental research and applications. Numerous icosahedral QCs (stable and metastable) with various chemical combinations were discovered and all of them can be classified as complex intermetallics containing more than 100 atoms in the unit cell. Thus, the knowledge about the building units for such crystals structures becomes essential and periodic approximant crystals (ACs) of QCs can be conveniently described as an assembly of shell-like clusters. For the recently synthesized Sc-Pd approximant, we studied structure - thermodynamic properties relations by introducing point defects in the area between Mackay clusters of the experimentally obtained crystal structure. The corresponding composition/configuration space (CCS) was set according to the DFT-derived energetics for different disordering types and further computationally evaluated. The current contribution aims to generalize the previously proposed approach and to predict possible stable 1/1 Mackay type ACs. To do so, we: (1) develop a routine for generating training CCSs and enhanced search space of the considering compounds, (2) evaluate thermodynamic properties of the entries of the training set using DFT, (3) develop and fine-tune AI models and train them on the DFT-derived structure-energy relations and compare their performance on test subsets, and (4) utilize the pre-trained AI models for predicting energetics of the enhanced searching space for promising novel stable phases of Mackay-type ACs in Sc-rich intermetallics. As a result, a number of potentially new phases among those predicted by AI models for the title compounds has been confirmed by DFT calculations. In addition, a comparative analysis of the methods used (Random Forest and Graph Neural Networks) is carried out.

Combined DFT-AI approach for searching for stable quasicrystal approximants in Sc-X (X = Rh, Pd, Ir, Pt) systems

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.