Bethe Ansatz with a Large Language Model
Summary: arXiv:2603.29932v1 Announce Type: cross
In a groundbreaking study, researchers have explored the potential of a Large Language Model (LLM) to tackle specific computations in mathematical physics. The focus of the study was on computing the coordinate Bethe Ansatz solutions of selected integrable spin chain models. This innovative approach sheds light on the capabilities of LLMs in handling complex mathematical tasks, particularly in the realm of theoretical physics.
Objective of the Study
The primary objective was to utilize the LLM to derive Bethe Ansatz solutions for three integrable Hamiltonians. Notably, two of these Hamiltonians had not been previously published, with the third presenting a unique challenge due to its inherent properties.
Methodology
The researchers employed OpenAI’s ChatGPT versions 5.2 Pro and 5.4 Pro to perform the calculations. Throughout the process, the LLM demonstrated a semi-autonomous capability to derive solutions, although it did make several errors. These inaccuracies were promptly identified and rectified by human researchers, ensuring the integrity of the final results.
Findings
The outcomes of the LLM’s computations were subsequently verified against exact diagonalization techniques executed by separate programs. The authors also conducted thorough checks on the derivations produced by the LLM. The findings revealed several intriguing aspects of the Bethe Ansatz solutions:
- New Integrable Hamiltonians: The study introduced two Hamiltonians that had not been documented before, expanding the existing body of knowledge in integrable systems.
- PT-Symmetry: One of the Hamiltonians broke left-right invariance but maintained PT-symmetry. This characteristic opens potential avenues for applications in Generalized Hydrodynamics, a field that studies the behavior of quantum systems.
- Unique Nested Bethe Ansatz: The third Hamiltonian was solved using a specialized form of the nested Bethe Ansatz. While the model featured interactions, it exhibited a free fermionic structure that lacked $U(1)$-invariance, a finding that appears to be unique and was successfully identified by the LLM.
Conclusion
This study marks a significant advancement in the integration of AI technologies within the field of theoretical physics. The successful application of a Large Language Model to derive complex mathematical solutions not only highlights the capabilities of AI in scientific research but also sets the stage for future explorations into the intersections of machine learning and physics. As LLMs continue to evolve, their potential to assist researchers in solving intricate problems will undoubtedly expand, paving the way for new discoveries in mathematical physics and beyond.