Prediction of Theoretical Carbide Crystal Structures Using Computational Modeling

Transition metal carbides (TMC) are promising catalysts due to sharing similar electronic properties with noble metals. However, TMCs exist in many different crystallographic phases, making syntheses that target specific phases a challenge.¹ Combinatorics in chemical and configurational spaces allows for discovery of new mono and bimetallic carbide phases; however, performing exhaustive Density Functional Theory (DFT) calculations to compute the properties of all possible phases, and determine ideal synthesis conditions, is computationally expensive. Machine Learning (ML) models, trained with quantum calculations, seek to accelerate this process through rapid and accurate prediction of material properties.

Here, we propose a computational workflow to predict the synthesizability of theoretically proposed Molybdenum carbide structures, specifically MoC and Mo­₂C, where a combination of DFT and ML techniques are employed to accelerate the material discovery process. ML models are trained on DFT calculations of commonly reported experimental molybdenum carbide structures using AMPTorch, an atomistic ML package.² Structures are optimized with Particle Swarm Optimization (PSO), while constraining the crystallographic prototype of the structure. The approach is validated by comparing predicted model output to DFT calculations, after which the models are employed to predict the synthesizability of over 700 proposed MoC and Mo₂C structures, with promising candidate structures validated by additional DFT calculations. This work paves the way for rapid computational screening of the vast structural space of mono and bimetallic transition metal carbides to identify new synthesizable materials with ideal electronic properties.

¹Shrestha, A.; Gao, X.; Hicks, J. C; Paolucci, C. Chemistry of Materials 2021, 33(12).

²Shuaibi, M.; Sivakumar, S.; Chen, R. Q.; Ulissi, Z. W. Machine Learning: Science and Technology 2021, 2(2), 25007.