Extension of Composite Method and Machine-Learned Electron Correlation Model to Fourth-Period Elements.

Area of Science:

• Quantum Chemistry
• Computational Chemistry
• Machine Learning

Background:

• Accurate calculation of electron correlation energy is a significant challenge in quantum chemistry.
• Existing methods often face limitations in terms of accuracy, efficiency, or applicability to heavier elements.

Purpose of the Study:

• To extend a machine-learned electron correlation (ML-EC) model to accurately and efficiently estimate coupled cluster singles and doubles with perturbative triples/complete basis set (CCSD(T)/CBS) correlation energies.
• To enable the model's application to fourth-period elements, overcoming previous limitations to third-period elements.

Main Methods:

• Developed an extended ML-EC model utilizing descriptors from Hartree-Fock (HF) calculations with double-zeta basis sets.
• Modified composite method parameters to extend applicability to fourth-period elements.
• Trained and validated the model on the G3/05 dataset and test molecules.

Main Results:

• The extended ML-EC model accurately reproduces CCSD(T)/CBS correlation energies and correlation energy densities.
• The model demonstrates high accuracy in predicting correlation energies for test molecules, outperforming Density Functional Theory (DFT) methods for reaction energies.
• Achieved a computational speedup of over 50 times compared to conventional CCSD(T)/CBS calculations.

Conclusions:

• The extended ML-EC model provides a reliable and computationally efficient approach for calculating correlation energies, especially for systems involving heavier elements.
• This advancement offers a promising alternative for high-accuracy quantum chemical calculations in various chemical applications.