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

A new workflow inserts an explicit technical specification between theory extraction and code generation, improving the accuracy of AI-assisted translation of quantum many-body physics papers into working programs, according to research posted on arXiv [1]. The protocol addresses a bottleneck the authors call knowledge externalization: converting implicit computational assumptions—index conventions, gauge choices, fermionic signs, contraction order, and memory constraints—into an explicit specification before implementation [1]. The multi-stage, human-in-the-loop workflow includes validation and stop gates [1]. It was tested on two algorithmically distinct quantum many-body tasks: variational sweep-based Density-Matrix Renormalization Group (DMRG) from a pedagogical review, and constructive Pfaffian conversion of Hartree–Fock–Bogoliubov states to matrix product states from a 2022 Letter by Jin et al. in Phys. Rev. B 105, L081101 [1]. For DMRG, all 16 specification-guided model pairings in a 4×4 grid satisfied physics-validation criteria, compared with 6 of 13 direct attempts [1]. A prose-specification ablation indicated that externalized content, not LaTeX formatting, was the essential ingredient [1]. For the Pfaffian-MPS task, the workflow succeeded in 11 of 26 archived attempts, while direct prompting yielded zero audited passes [1]. Cross-specification transfer was asymmetric: non-GPT specifications implemented by GPT 5.5 passed 4 of 4, while GPT 5.5 specifications implemented by weaker models failed 4 of 4, pointing to a residual implementation-model bottleneck [1]. The resulting Paper-to-Program Many-Body skill provides an auditable protocol for AI-assisted implementation of many-body algorithms and for diagnosing where externalization succeeds or fails [1]. The work appears on arXiv, an open-access repository that hosts preprints across physics, mathematics, and computer science and surpassed two million articles by the end of 2021 [7]. The platform supports community-developed tools through its arXivLabs framework, which enables collaborators to build features such as bibliographic explorers and code-and-data linkers directly on article pages [5][6].