Shadow Molecular Dynamics with a Machine Learned Flexible Charge Potential.

Area of Science:

• Computational chemistry
• Materials science
• Molecular dynamics simulations

Background:

• Accurate molecular dynamics (MD) simulations require precise interatomic potentials.
• Charge equilibration (QE) models estimate atomic charges but can be computationally intensive.
• Machine learning (ML) offers a powerful tool for developing more efficient and accurate models.

Purpose of the Study:

• To introduce an extended Lagrangian shadow molecular dynamics (EL-SMD) scheme.
• To integrate a second-order charge equilibration (SOCE) model with ML-derived parameters for interatomic potentials.
• To enhance the accuracy, stability, and efficiency of molecular dynamics simulations.

Main Methods:

• Developed an EL-SMD scheme using Born-Oppenheimer potentials derived from relaxed atomic charges.
• Utilized neural networks to parametrize the SOCE model with environment-dependent electronegativities and chemical hardness.
• Evaluated the impact of fixed versus environment-dependent QE parameters on simulation accuracy.
• Computed molecular infrared spectra using the dipole autocorrelation function.

Main Results:

• The EL-SMD scheme demonstrated improved numerical stability and reduced Coulomb potential calculations.
• Achieved efficient and accurate molecular dynamics simulations with excellent long-term trajectory stability.
• ML-based parametrization of QE significantly enhanced the accuracy of the flexible charge potential.
• Computed infrared spectra accurately matched experimental data, validating the EL-SMD approach.

Conclusions:

• The proposed EL-SMD scheme with ML-parametrized flexible charge potentials provides a robust and accurate method for molecular simulations.
• Environment-dependent atomic parameters derived via ML are crucial for capturing accurate molecular behavior.
• This approach offers a significant advancement for computational chemistry and materials science research.