Automatic Differentiation for Enhanced Potential Energy Surface Navigation: Improved Minimum and Transition State...

Area of Science:

• Computational Chemistry
• Chemical Reaction Dynamics
• Molecular Modeling

Background:

• Reaction modeling is crucial but time-consuming.
• Finding transition states (TS) for rate constants is a bottleneck.
• Current methods like ReaxFF with finite differentiation (FD) for TS searches are noisy and computationally intensive.

Purpose of the Study:

• To develop a more accurate and efficient method for transition state (TS) searches.
• To improve the calculation of second energy derivatives in molecular dynamics.
• To enable faster and more reliable reaction modeling.

Main Methods:

• Integrated automatic differentiation (AD) into LAMMPS for second derivative calculations, creating ADfied LAMMPS.
• Developed LMP-Gau by interfacing ADfied LAMMPS with Gaussian for geometry optimization, frequency calculations, and reaction path following.
• Applied ADfied LAMMPS and LMP-Gau to find transition states in 1,3-dioxolane oxidation.

Main Results:

• ADfied LAMMPS and LMP-Gau provide accurate second energy derivatives, overcoming FD noise.
• LMP-Gau shows improved energy minimization for stable molecules compared to standard LAMMPS.
• Successfully identified transition states with improved convergence using AD compared to FD.

Conclusions:

• ADfied LAMMPS and LMP-Gau offer a significant advancement for accurate and efficient transition state searches.
• The integration of AD enhances the reliability of molecular dynamics simulations for chemical reactions.
• This approach accelerates reaction space exploration and chemical modeling.