Combining Graphics Processing Units, Simplified Time-Dependent Density Functional Theory, and Finite-Difference...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Biophysics

Background:

• Nonadiabatic molecular dynamics (NAMD) is crucial for understanding photochemical and photophysical processes.
• Ab initio NAMD calculations, especially at the time-dependent density functional theory (TDDFT) level, are computationally intensive.
• Previous GPU implementations have improved NAMD efficiency.

Purpose of the Study:

• To develop and implement further acceleration strategies for ab initio NAMD at the TDDFT level.
• To enable accurate NAMD simulations of larger systems on accessible hardware.
• To demonstrate the enhanced methodology with a relevant biological system.

Main Methods:

• Implementation of simplified TDDFT schemes (Grimme et al.).
• Application of the Hammes-Schiffer-Tully approach for nonadiabatic couplings via finite differences.
• Leveraging GPU-accelerated integral routines for SCF, TDDFT, and TDDFT derivative calculations.
• Combined TDDFT/MM-NAMD simulations.

Main Results:

• Significant reduction in computational cost by minimizing demanding steps in the NAMD algorithm.
• Accurate physical description maintained despite computational simplifications.
• Feasibility of NAMD simulations for systems with hundreds of atoms on a single compute node.
• Successful TDDFT/MM-NAMD simulation of the rhodopsin protein.

Conclusions:

• The developed computational scheme dramatically accelerates ab initio NAMD calculations.
• This approach makes complex molecular dynamics simulations more accessible.
• The methodology is validated by its application to the rhodopsin protein, opening doors for further biomolecular studies.