Capturing Dichotomic Solvent Behavior in Solute-Solvent Reactions with Neural Network Potentials.

Area of Science:

• Computational Chemistry
• Chemical Physics
• Molecular Dynamics

Background:

• Simulating condensed phase chemical reactivity is computationally demanding for traditional quantum chemistry methods.
• System size and reaction complexity pose significant challenges in modeling chemical processes accurately.

Purpose of the Study:

• To develop an efficient workflow for training neural network potentials (NNPs) for exploring energy barriers in solution.
• To investigate a puzzling solute-solvent reactivity route involving the ring opening of N-enoxyphthalimide in different solvents.

Main Methods:

• Utilized active and transfer learning to bypass computational burden of PBE0-D3(BJ) calculations.
• Employed well-tempered metadynamics simulations with multiple time step integration for accelerated transition state sampling.
• Developed and applied neural network potentials (NNPs) for ab initio level simulations.

Main Results:

• Successfully generated detailed free energy surfaces and relative energy barriers consistent with experimental observations.
• Identified transition states involving multiple solvent molecules, contrasting with static models.
• Found that solvent differences arise from electronic effects and conformational entropy, not hydrogen bonding networks.

Conclusions:

• The developed workflow enables efficient and accurate simulation of complex chemical reactivity in solution.
• Dynamic simulations are crucial for capturing the full complexity of solute-solvent interactions.
• The study provides insights into solvent-dependent reactivity mechanisms, highlighting the interplay of electronic and entropic factors.