Transfer Learning for Predictive Molecular Simulations: Data-Efficient Potential Energy Surfaces at CCSD(T) Accuracy.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Molecular Dynamics

Background:

• Accurate potential energy surfaces (PESs) are essential for molecular simulations.
• High-level quantum chemistry methods (e.g., CCSD(T)) are computationally expensive for large molecules.

Purpose of the Study:

• To develop CCSD(T)-quality PESs using data-efficient transfer learning.
• To enable accurate molecular simulations for systems previously intractable due to computational cost.

Main Methods:

• Employed transfer learning techniques for data-efficient PES construction.
• Achieved state-of-the-art accuracy at a fraction of the computational cost.

Main Results:

• Generated accurate PESs for tropolone and the (propiolic acid)-(formic acid) dimer (PFD).
• Semiclassical tunneling splittings and anharmonic frequencies were calculated.
• Excellent agreement with experimental data for tropolone was observed.

Conclusions:

• The developed framework provides a rational strategy for obtaining highest-accuracy PESs.
• This approach significantly improves computational efficiency for accurate molecular simulations.
• The method shows promise for applications in spectroscopy and experimental validation.