Thawed Gaussian Wavepacket Dynamics with Δ-Machine-Learned Potentials.

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Spectroscopy

Background:

• Accurate simulation of molecular spectra is crucial for understanding chemical processes.
• Traditional methods struggle with large, floppy molecules due to computational cost.
• Machine learning offers a promising avenue for accelerating these simulations.

Purpose of the Study:

• To develop a computationally efficient method for simulating vibronic spectra.
• To apply a Δ-machine learning approach to model anharmonic corrections in molecular potentials.
• To enable reliable simulations for large and flexible molecules.

Main Methods:

• Variable-width (thawed) Gaussian wavepacket (GWP) variational dynamics on machine-learned potentials.
• Fitting anharmonic corrections to the global harmonic approximation (GHA) using kernel ridge regression (Δ-machine learning).
• Propagation of a single thawed GWP using the time-dependent variational principle to compute autocorrelation functions.

Main Results:

• The developed method accurately simulates vibronic spectra, as demonstrated by excellent agreement with the photoelectron spectrum of ammonia.
• Fitting anharmonic corrections requires smaller training datasets compared to fitting total electronic energies.
• The approach reduces the dimensionality of the nuclear space needed for potential energy surface scans.

Conclusions:

• The Δ-machine learning method provides a powerful and efficient tool for simulating vibronic spectra.
• This approach significantly reduces computational requirements, making it suitable for large, floppy molecules.
• The method paves the way for reliable theoretical predictions of complex molecular spectra.