Unlocking Carbide Vacancy Energy in Molybdenum Carbide: Predictive Insights from Numerical Descriptors and Neural Networks

Carbides, as exemplified by Mo
2
C, have various valuable properties that affect various industries. Understanding the influence of carbon vacancy patterns on these properties is an essential yet challenging task. This work presents a two-stage approach. First, novel numerical descriptors linked to molecular structure are predicted using Neural Networks. Subsequently, using these descriptors alone, carbide vacancy energy is accurately forecasted through Neural Networks and an Attention mechanism. The correlation unveiled hints at a numerical relationship between these descriptors and energy, opening avenues for interpreting their significance. This research contributes to computational materials science, offering insights into carbide intricacies and inspiring innovative applications.