Efficient charge assignment and back interpolation in multigrid methods for molecular dynamics.

Area of Science:

• Computational chemistry
• Molecular dynamics
• Scientific computing

Background:

• Accurate calculation of molecular forces is essential for simulating chemical and biological processes.
• Current multigrid methods for molecular force calculations face computational bottlenecks in charge assignment and force interpolation.
• Existing methods utilize truncated Gaussian distributions for atomic charges, contributing to significant execution time.

Purpose of the Study:

• To develop novel, computationally efficient methods for atomic charge assignment and force interpolation within the multigrid approach.
• To reduce the execution time of multigrid molecular force calculations by optimizing key computational steps.
• To improve the scalability of multigrid methods with respect to problem size and processor count.

Main Methods:

• Implemented convolution-based approaches for mapping Gaussian representations of atomic charges onto a computational grid.
• Utilized convolution for interpolating forces from the grid back to atomic positions.
• Compared the performance and accuracy of the proposed methods against baseline multigrid computations.

Main Results:

• The proposed convolution-based methods achieve comparable force accuracy to existing techniques.
• Significant reduction in computational run time for charge assignment and back interpolation steps.
• Demonstrated improved scalability of the new methods with increasing problem size and number of processors.

Conclusions:

• Convolution-based charge assignment and force interpolation offer a more efficient alternative for multigrid molecular force calculations.
• These optimized methods alleviate computational bottlenecks, leading to faster simulations.
• The enhanced scalability ensures better performance on larger systems and parallel computing architectures.