Symbolic regression in materials science via dimension-synchronous-computation

•Combining physics knowledge with machine learning methods is an outstanding challenge in materials science. We introduced dimension-synchronous-computation symbolic regression method to efficiently establish expressions for band gap of NaCl-type compounds. It is embedding the dimension computation system to ensure the dimensional consistency. And descriptor binding mechanism is introduced to keep the knowledge in modeling. The pattern of 'knowledge-model-knowledge' provides a new scheme for multi-component material modeling and design, and believed to be used to inspire existing machine learning applications.