LLM-Guided Test-Time Discovery of Quantum-Chemical Approximation Algorithms

LADeQ is introduced as an LLM-guided workflow that autonomously discovers, implements, and benchmarks quantum-chemical approximation algorithms at test-time, enhancing existing simulation codes without requiring predefined tools. It leverages techniques from spatial statistics, circuit simulation, and kernel methods to create on-demand approximation schemes, maintaining transparency and control over accuracy-efficiency trade-offs. This approach accelerates coupled cluster singles and doubles (CCSD) and configuration interaction singles and doubles (CISD) calculations while adhering to user-defined correlation-energy error tolerances, offering significant advancements for practitioners in quantum chemistry simulations.