Integrating Coupled-Cluster Theory Within AI Workflows for Accurate Molecular Geometries

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Coupled-cluster singles and doubles (CCSD) calculations are accurate but computationally expensive for molecular geometry optimization.
• Traditional CCSD methods involve solving complex nonlinear and linear equations, scaling as O(N^6), limiting their application to larger systems.
• The computational bottleneck in CCSD geometry optimization lies in iteratively solving the t-amplitude and Lambda-equations.

Purpose of the Study:

• To introduce a novel hybrid data-driven coupled-cluster (DDCC) framework for efficient CCSD-level geometry optimizations.
• To accelerate high-level electronic structure calculations by directly predicting wavefunction parameters.
• To enable routine CCSD-quality geometry optimization for larger and more complex molecular systems.

Main Methods:

• Developed a Lambda data-driven CC (ΛDDCC) scheme that replaces iterative solvers with neural network predictions.
• The neural network predicts the t and Lambda amplitudes, bypassing the computationally demanding equation-solving steps.
• Employed a physically grounded representation of electron correlation using excitation-based descriptors and a molecular orbital framework.

Main Results:

• The ΛDDCC framework accurately recovers CCSD wavefunction parameters and provides reliable nuclear gradients.
• Achieved near-quantitative agreement with CCSD geometries across diverse molecular sets and perturbed configurations.
• Reduced computational cost by over an order of magnitude compared to conventional CCSD methods.

Conclusions:

• ΛDDCC is an effective and transferable strategy for accelerating high-level electronic structure calculations.
• The developed method offers a practical approach for routine CCSD-quality geometry optimization.
• This data-driven approach significantly enhances the efficiency of computational chemistry methods.