From Paper to Program: Knowledge Externalization for AI-Assisted Quantum Many-Body Code Generation

The article presents a novel multi-stage workflow for knowledge externalization in AI-assisted code generation for quantum many-body physics, addressing the fragility of paper-to-program translation due to implicit conventions. The workflow was evaluated on two tasks: DMRG and Pfaffian conversion, achieving a significant improvement in validation rates (16/16 for DMRG with specifications versus 6/13 without, and 11/26 for Pfaffian-MPS compared to 0/26). This approach enhances the reliability of generating scientific code from literature, providing a structured protocol that aids practitioners in implementing complex algorithms while identifying points of failure in the externalization process.