Towards Routine Condensed Phase Simulations with Delta-Learned Coupled Cluster Accuracy: Application to Liquid Water

Area of Science:

• Computational Chemistry
• Materials Science
• Quantum Mechanics

Background:

• Accurate simulation of liquid water requires precise electronic structure methods and nuclear motion sampling.
• Coupled cluster theory with single, double, and perturbative triple excitations [CCSD(T)] offers high accuracy but is computationally expensive for condensed phases.
• Machine learning potentials (MLPs) have shown promise in achieving experimental agreement for liquid water simulations.

Purpose of the Study:

• To develop a practical and routine approach for achieving CCSD(T)-level accuracy in liquid water simulations.
• To combine machine learning potentials with local correlation approximations for efficient simulations.
• To enable accurate prediction of both structural/transport properties and bulk properties like density maximum.

Main Methods:

• Developed a practical approach combining machine learning potentials (MLPs) with local correlation approximations.
• Enabled routine coupled cluster theory with single, double, and perturbative triple excitations [CCSD(T)]-level simulations for condensed phase systems.
• Incorporated nuclear quantum effects and constant-pressure simulations for comprehensive property prediction.

Main Results:

• Achieved agreement with experimental data for structural and transport properties of liquid water.
• Successfully predicted isothermal-isobaric bulk properties, including water's density maximum, using MLP-based CCSD(T) models.
• Demonstrated the practicality and routine applicability of the developed simulation approach.

Conclusions:

• The combined approach of MLPs and local correlation approximations provides a practical blueprint for routine CCSD(T)-based simulations in condensed phases.
• This methodology facilitates accurate prediction of liquid water properties, bridging the gap between theoretical accuracy and experimental data.
• The work paves the way for broader application of high-accuracy electronic structure methods in condensed matter simulations.