Multiple-Timestep ab Initio Molecular Dynamics Using an Atomic Basis Set Partitioning.

Area of Science:

• Computational Chemistry
• Theoretical Chemistry
• Molecular Dynamics Simulations

Background:

• Ab initio Born-Oppenheimer molecular dynamics (MD) simulations are computationally expensive, limiting their application to large systems or long timescales.
• The computational cost of MD simulations is heavily influenced by the choice of basis sets used to describe electronic structure.
• Exploiting timescale separation in electronic contributions could offer a pathway to accelerate these simulations.

Purpose of the Study:

• To develop and validate a novel multiple-timestep scheme for accelerating ab initio Born-Oppenheimer MD simulations.
• To assess the accuracy and reliability of the proposed acceleration method across various chemical systems and theoretical approaches.
• To quantify the computational cost savings achieved by the new protocol.

Main Methods:

• A multiple-timestep scheme was implemented, utilizing a cost-efficient, low-level basis set for most MD steps.
• A dynamical correction was applied to account for the effects of larger basis set relaxation in a time-reversible manner.
• The method was tested on water clusters, methanol dimer, alanine polypeptide, protonated hydrazine, and oxidized water dimer using Hartree-Fock, DFT, and MP2 levels of theory.

Main Results:

• The multiple-timestep scheme successfully generated valid MD trajectories, consistent with high-level basis set calculations.
• Key observables, including MD-based vibrational spectra, were accurately reproduced.
• Significant cost savings, ranging from 2.6 to 7.3 times, were demonstrated for the tested systems.

Conclusions:

• The proposed approach effectively accelerates ab initio Born-Oppenheimer MD simulations without compromising accuracy.
• The method is validated across various quantum chemical methods (HF, DFT, MP2) and molecular systems.
• Recommended basis set pairings (e.g., 6-31G for 6-31G** or 6-311G**; cc-pVDZ for aug-cc-pVTZ) offer substantial computational efficiency.