Accurate Probabilities for Highly Activated Reaction of Polyatomic Molecules on Surfaces Using a High-Dimensional...

Area of Science:

• Surface science
• Chemical physics
• Computational chemistry

Background:

• Accurate modeling of molecule-surface reactions requires accounting for energy transfer between molecules and surface phonons.
• Ab initio molecular dynamics (AIMD) can describe this energy transfer but is computationally prohibitive for reactions with low probabilities (<1%).

Purpose of the Study:

• To develop and validate a computationally efficient method using neural network potentials for simulating reactive scattering of polyatomic molecules on metal surfaces.
• To investigate the dynamics of the highly activated reaction of CHD3 on a mobile Cu(111) surface.

Main Methods:

• Development and application of a neural network potential for simulating the potential energy surface of CHD3 interacting with a Cu(111) surface.
• Employing molecular dynamics simulations with the neural network potential to study reactive scattering, including energy transfer and dynamical effects.

Main Results:

• The neural network potential significantly reduces computational cost compared to AIMD for simulating molecule-surface reactions.
• Dynamical effects, including the bobsled effect and surface recoil, were found to considerably influence reaction probability.
• A unique dynamical effect was observed for CHD3 + Cu(111): higher vibrational efficacy was achieved with two quanta in the CH stretch mode than with one quantum.

Conclusions:

• Neural network potentials offer a viable and computationally tractable alternative to AIMD for studying complex molecule-surface reactions.
• The study highlights the importance of considering dynamical effects and vibrational modes in understanding reaction probabilities on metal surfaces.
• This approach paves the way for more detailed investigations of surface reactions with low probabilities.