Accelerating Many-Body Quantum Chemistry via Generative Transformer-Enhanced Configuration Interaction

Area of Science:

• Quantum Chemistry
• Computational Physics
• Machine Learning

Background:

• Quantum many-body calculations face computational limits due to the exponential growth of configuration space.
• Accurate treatment of strongly correlated systems is computationally prohibitive for traditional methods.

Purpose of the Study:

• To introduce a novel machine learning approach, Generative Transformer Neural Network Selected Configuration Interaction (GTNN-SCI), for accelerating quantum chemistry calculations.
• To enhance the accuracy and efficiency of treating strongly correlated systems.

Main Methods:

• Developed GTNN-SCI, a Transformer-based machine learning method that generatively samples important configurations.
• Leveraged the Transformer architecture's self-attention mechanism to capture long-range electron correlations.
• Applied GTNN-SCI to calculate correlation and binding energies for molecules (N2, H2O, C2) and a challenging [2Fe-2S] cluster.

Main Results:

• GTNN-SCI achieved up to a 10x speedup compared to state-of-the-art neural network methods.
• Demonstrated faster convergence and lower energies than previous neural network-based selected CI techniques.
• Accurately treated the strongly correlated [2Fe-2S] cluster, achieving ground-state energies within chemical accuracy of DMRG benchmarks.

Conclusions:

• GTNN-SCI combines deep learning with high-performance electronic structure computation for efficient and precise solutions.
• The generative approach identifies higher-order excitations missed by conventional methods, yielding lower variational energies.
• GTNN-SCI offers a powerful new avenue for solving the electronic Schrödinger equation in challenging molecular systems.