Accelerating structure relaxation in chemically disordered materials with a chemistry-driven model

Chemically disordered materials are widely utilized, yet establishing structure-property relationship remains challenging due to their vast configurational space. Identifying thermal accessible low energy configurations of these materials through standard ab initio calculations is computationally expensive for doping induced structure changes. In this work, we propose a straightforward algorithm to optimize random structures into ground state configurations by matching chemical subgraphs. This algorithm constructs harmonic potential with chemistry-driven parameterization, without relying on iterative training to accelerate the relaxation process. It can completely bypass the need for relaxation with ab initio calculations in rigid systems and reduce computational costs by 30% in flexible systems. Leveraging its exceptional structural relaxation capabilities, we have also developed a generalized workflow for screening low-energy structures in disordered materials, aimed at expediting the screening process and accelerating new material discovery.