From Paper to Program: Knowledge Externalization for AI-Assisted Quantum Many-Body Code Generation
arXiv:2604.04089v3 Announce Type: replace-cross Abstract: Large language models can write scientific code, but direct paper-to-program translation remains fragile when correctness depends on tacit conventions in the literature. We identify this bottleneck as \textbf{knowledge externalization}: converting implicit computational assumptions -- index conventions, gauge choices, fermionic signs, contraction order, and memory constraints -- into an explicit technical specification before implementation. We evaluate a multi-stage, human-in-the-loop workflow that inserts such a specification, with va
The increasing sophistication of large language models for code generation, coupled with their current limitations in handling tacit scientific conventions, makes this an opportune moment for workflows that bridge this gap.
This development addresses a critical bottleneck in AI-assisted scientific discovery by enhancing the reliability and correctness of AI-generated code, particularly for complex physical systems.
The explicit externalization of implicit scientific knowledge into technical specifications will make AI-generated scientific code more trustworthy and accelerate research in fields like quantum many-body physics.
- · AI-assisted scientific research
- · Physics research labs
- · Software developers for scientific computing
- · Large language model developers
- · Manual code generation for complex scientific problems
Increased pace and reduced error rates in scientific code development using AI.
Faster discovery of new materials and physical phenomena due to more reliable computational models.
Potential for a paradigm shift in how theoretical physics and condensed matter research are conducted, relying heavily on AI-orchestrated experiments and simulations.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI