Accelerating direct quantum dynamics using graphical processing units.

Area of Science:

• Computational Quantum Chemistry
• Theoretical Chemistry
• Chemical Dynamics

Background:

• Solving the time-dependent Schrödinger equation (TDSE) is crucial for understanding molecular dynamics.
• Grid-based methods for TDSE calculations face significant computational scaling challenges.
• Gaussian basis function methods offer a grid-free alternative, enabling simulations of larger systems.

Purpose of the Study:

• To investigate the computational bottlenecks in direct dynamics variational multi-configurational Gaussian (DD-vMCG) simulations.
• To assess the impact of Graphical Processing Unit (GPU) acceleration on electronic structure calculations within DD-vMCG.
• To determine the scaling of the vMCG method with the number of Gaussian basis functions.

Main Methods:

• Implementation of the direct dynamics variational multi-configurational Gaussian (DD-vMCG) method.
• Coupling DD-vMCG with GPU-accelerated electronic structure calculations.
• Simulation of quantum dynamics for model systems: protonated ammonia dimer and imidazole dimer.

Main Results:

• The computational cost of nuclear dynamics in DD-vMCG often outweighs the cost of quantum chemistry calculations, even with GPU acceleration.
• An estimated scaling of the vMCG method with respect to the number of Gaussian basis functions was determined.
• Identified conditions under which quantum chemistry calculations become the limiting factor in DD-vMCG simulations.

Conclusions:

• For many DD-vMCG simulations, focusing on optimizing the quantum dynamics propagation is more critical than solely accelerating quantum chemistry.
• GPU acceleration's effectiveness is system-dependent and influenced by the scaling of the vMCG method.
• The findings guide the efficient application of DD-vMCG and GPU acceleration for both ground and excited state dynamics.