Self-Optimizing Machine Learning Potential Assisted Automated Workflow for Highly Efficient Complex Systems Material Design
This framework uses an attention-coupled neural network potential (ACNN) to automate crystal structure prediction for complex materials. It employs self-optimizing iterations guided by energy minima and unphysical local minima to improve accuracy and generalizability. The system efficiently samples promising compositions using a convex hull strategy, successfully discovering new phases in Mg–Ca–H and Be–P–N–O systems.