Transfer Learning Meets Embedded Correlated Wavefunction Theory for Chemically Accurate Molecular Simulations:...

Area of Science:

• Computational chemistry
• Chemical physics
• Materials science

Background:

• Achieving chemical accuracy in molecular simulations is a significant challenge.
• Accurate simulation of condensed-phase dynamics requires capturing complex electron interactions.
• Machine-learned potentials offer efficiency but often lack high-level electronic correlation accuracy.

Purpose of the Study:

• To present a novel embedded correlated wavefunction transfer learning (ECW-TL) framework.
• To enable accurate molecular dynamics simulations in the condensed phase.
• To incorporate high-level electronic correlation effects efficiently.

Main Methods:

• Developed the ECW-TL framework combining ECW theory with machine learning.
• Applied ECW-TL to simulate Ca²⁺-CO₃^2- ion pairing in aqueous solution.
• Validated the framework by comparing with DFT-SCAN, MP2, and CCSD(T) methods, and experimental data.

Main Results:

• ECW-TL fine-tuned DFT models accurately reproduced DFT-SCAN free-energy surfaces.
• Incorporating MP2 and CCSD(T) levels of theory refined the free-energy profile.
• The framework achieved quantitative agreement with experimental ion-pair association free energies.

Conclusions:

• ECW-TL provides a data-efficient approach for high-accuracy simulations.
• The framework successfully transfers correlated wavefunction accuracy to condensed-phase dynamics.
• ECW-TL is a general method for simulating complex aqueous and interfacial chemical processes.