Fully First-Principles Surface Spectroscopy with Machine Learning.

Area of Science:

• Physical Chemistry
• Computational Chemistry
• Spectroscopy

Background:

• Surface-specific spectroscopies like vibrational sum-frequency generation (VSFG) advance molecular-level understanding of interfaces.
• Atomistic simulations are crucial for interpreting VSFG spectra but face computational cost limitations.
• Existing methods struggle with complex electronic structures and achieving statistical convergence in simulations.

Purpose of the Study:

• To develop a computationally efficient and accurate method for modeling VSFG spectra.
• To overcome the limitations of current atomistic simulation techniques for interface studies.
• To enable accurate VSFG signal prediction for complex systems.

Main Methods:

• Combined high-dimensional neural network interatomic potentials with symmetry-adapted Gaussian process regression.
• Utilized machine learning to model VSFG signals with ab initio accuracy.
• Applied the developed approach to the water/air interface.

Main Results:

• Successfully modeled VSFG signals with high accuracy, comparable to ab initio calculations.
• Demonstrated the versatility of the machine learning approach on the water/air interface.
• Identified key sources of theoretical inaccuracy in VSFG spectral modeling.

Conclusions:

• Machine learning offers a powerful solution to the computational challenges in VSFG spectroscopy.
• The developed strategy provides a clear pathway for modeling surface-sensitive spectroscopy of complex interfaces.
• This work enhances the molecular-level understanding of aqueous interfaces through accurate simulation.