Molecular Monte Carlo Simulations Using Graphics Processing Units: To Waste Recycle or Not?

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Engineering

Background:

• The waste recycling Monte Carlo (WRMC) algorithm enhances simulation accuracy by using multiple trial states.
• Parallel implementation on graphics processing units (GPUs) offers potential for significant computational speedup.

Purpose of the Study:

• To implement and evaluate two WRMC algorithms on GPUs using CUDA.
• To assess the performance and efficiency of GPU-accelerated WRMC for molecular simulations.
• To apply the optimized algorithms to determine free energy landscapes in zeolite systems.

Main Methods:

• Implementation of two WRMC algorithms in CUDA: one with uniformly distributed random trial states, another with displacement random-walk steps.
• Testing on a methane-zeolite MFI framework system.
• Analysis of statistical accuracy versus CUDA block size for hardware resource allocation.
• Comparison of GPU implementations against serial CPU Monte Carlo methods.

Main Results:

• Demonstrated the utility of WRMC algorithms on GPUs for molecular simulations.
• Identified optimal CUDA block sizes for efficient GPU resource utilization.
• Achieved significant speedup compared to serial CPU implementations.
• Successfully applied the optimized GPU algorithms to calculate free energy landscapes for molecular motion in zeolite LTA.

Conclusions:

• GPU-accelerated WRMC algorithms provide a powerful and efficient approach for molecular simulations.
• The WRMC method is well-suited for parallel implementation on GPUs, offering substantial performance gains.
• This work facilitates more accurate and faster characterization of molecular behavior in porous materials like zeolites.