Reproduce “Machine-Learning Accelerated Calculations of Reduced Density Matrices” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Awwab A. Azam, Lexu Zhao, Jiabin Yu
Neural network architectures are developed to accelerate and predict reduced density matrices (RDMs) for large-size strongly correlated systems by leveraging the interpolability of momentum-space RDMs. The approach achieved over 90% reduction in Hartree-Fock iteration counts and high accuracy in predicting correlation functions for significantly larger systems.
Key insights
- Momentum-space n-RDMs often exhibit smoothness and interpolability over the Brillouin zone, especially for gapped states.
- Neural networks are capable of learning these smooth functions from coarse-grained momentum meshes (smaller systems) and accurately inferring values for denser meshes (larger systems).
- High-quality initial guesses provided by trained NNs can dramatically accelerate the convergence of iterative quantum calculations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.