Accelerating electrostatic pair methods on graphical processing units to study molecules in supercritical carbon...

Area of Science:

• Computational Chemistry
• Materials Science
• Molecular Dynamics

Background:

• Traditional Ewald summation methods for electrostatic interactions are computationally expensive and not well-suited for GPU acceleration.
• Efficiently modeling electrostatic interactions is crucial for molecular dynamics (MD) simulations.

Purpose of the Study:

• To implement and evaluate the Wolf summation method on Graphics Processing Units (GPUs) using the LAMMPS MD package.
• To assess the performance and accuracy of the Wolf method as an alternative to Particle-Particle Particle-Mesh (PPPM) for electrostatic interactions.
• To investigate the impact of system polarity on the accuracy of the Wolf method.

Main Methods:

• Implementation of the Wolf summation method within the GPU-accelerated LAMMPS MD package.
• Computation of transport and structural properties for pure carbon dioxide and mixtures (CO2/CH4, CO2/CH2F2).
• Comparison of results obtained using the Wolf method with those from PPPM on GPUs.

Main Results:

• The Wolf summation method yields diffusion coefficients for pure carbon dioxide that are indistinguishable from PPPM on GPUs.
• An optimal potential damping parameter (α) of 0.075 was determined for the Wolf method.
• Accuracy slightly decreases with increasing system polarity, but the method remains robust for mildly polar systems.
• Utilizing the Wolf method instead of PPPM resulted in a two-fold reduction in wall-clock time.

Conclusions:

• The Wolf summation method is a viable and efficient alternative to PPPM for electrostatic interactions in MD simulations on GPUs.
• The method provides significant computational speedups without compromising accuracy for systems like carbon dioxide.
• The Wolf method's robustness in mildly polar systems broadens its applicability in various computational chemistry and materials science simulations.