Long-range treatment of dispersion (r-6) is necessary for appropriate handling of materials near interfaces, and particularly at liquid-vapor interfaces. However, these methods have traditional been regarded as computationally prohibitive, and truncated potentials have been used instead. However, recent work has shown that using a PPPM-based long-range dispersion solver can be faster than a truncation-based potential through proper selection of the switching radius and the Ewald parameter. While this solves the problem of efficiency for relatively large systems (up to several thousand processors), for very large systems, parallel communication necessitated by the Fast Fourier Transforms (FFTs) used by PPPM dominate the cost.We present two new algorithms for handling long-range dispersions, one using a variant of the multilevel summation algorithm of Hardy et al., and a new algorithm which avoids the grid-based parallel communication of the PPPM and multilevel summation algorithms through dynamic adjustment of the dispersion cutoff for atoms near the interface. In particular, we focus on identifying the optimal conditions for the PPPM, multilevel summation, and dynamic cutoff algorithms (range of system sizes and processor allocations), as well as the accuracy of each algorithm compared to 'reference' calculations with the Ewald summation method (the most accurate but slowest of long-range methods).

Algorithms for long-range dispersion interactions: A comparison between PPPM and non-FFT-based solvers

248th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.