Accelerating electrostatic interaction calculations with graphical processing units based on new developments of ewald method using non-uniform fast fourier transform
2016 (English)In: Journal of Computational Chemistry, ISSN 0192-8651, E-ISSN 1096-987X, Vol. 37, no 3, 378-387 p.Article in journal (Refereed) PublishedText
We present new algorithms to improve the performance of ENUF method (F. Hedman, A. Laaksonen, Chem. Phys. Lett. 425, 2006, 142) which is essentially Ewald summation using Non-Uniform FFT (NFFT) technique. A NearDistance algorithm is developed to extensively reduce the neighbor list size in real-space computation. In reciprocal-space computation, a new algorithm is developed for NFFT for the evaluations of electrostatic interaction energies and forces. Both real-space and reciprocal-space computations are further accelerated by using graphical processing units (GPU) with CUDA technology. Especially, the use of CUNFFT (NFFT based on CUDA) very much reduces the reciprocal-space computation. In order to reach the best performance of this method, we propose a procedure for the selection of optimal parameters with controlled accuracies. With the choice of suitable parameters, we show that our method is a good alternative to the standard Ewald method with the same computational precision but a dramatically higher computational efficiency.
Place, publisher, year, edition, pages
Wiley-Blackwell, 2016. Vol. 37, no 3, 378-387 p.
electrostatic interaction calculation, ewald summation, GPU, non-uniform FFT
IdentifiersURN: urn:nbn:se:kth:diva-180961DOI: 10.1002/jcc.24250ISI: 000367385000007PubMedID: 26584145OAI: oai:DiVA.org:kth-180961DiVA: diva2:898841
QC 201601292016-01-292016-01-262016-01-29Bibliographically approved