doi:10.1016/j.cpc.2007.08.014
Copyright © 2007 Elsevier B.V. All rights reserved.
A scalable parallel algorithm for large-scale reactive force-field molecular dynamics simulations
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Ken-ichi Nomuraa, Rajiv K. Kaliaa, Aiichiro Nakano
, a,
and Priya Vashishtaa
aCollaboratory for Advanced Computing and Simulations, Department of Computer Science, Department of Physics & Astronomy, Department of Chemical Engineering & Materials Science, University of Southern California, Los Angeles, CA 90089-0242, USA
Received 18 June 2007;
revised 16 August 2007;
accepted 17 August 2007.
Available online 15 September 2007.
Abstract
A scalable parallel algorithm has been designed to perform multimillion-atom molecular dynamics (MD) simulations, in which first principles-based reactive force fields (ReaxFF) describe chemical reactions. Environment-dependent bond orders associated with atomic pairs and their derivatives are reused extensively with the aid of linked-list cells to minimize the computation associated with atomic n-tuple interactions (n
4 explicitly and
6 due to chain-rule differentiation). These n-tuple computations are made modular, so that they can be reconfigured effectively with a multiple time-step integrator to further reduce the computation time. Atomic charges are updated dynamically with an electronegativity equalization method, by iteratively minimizing the electrostatic energy with the charge-neutrality constraint. The ReaxFF-MD simulation algorithm has been implemented on parallel computers based on a spatial decomposition scheme combined with distributed n-tuple data structures. The measured parallel efficiency of the parallel ReaxFF-MD algorithm is 0.998 on 131,072 IBM BlueGene/L processors for a 1.01 billion-atom RDX system.
Keywords: Molecular dynamics; Reactive force field; Parallel computing
PACS classification codes: 02.70.-c; 02.70.Ns; 82.20.Db
Fig. 1. Schematic of atomic configurations in energy terms: (a) 1-body, (b) 2-body, (c) 3-body, (d) 4-body, (e) hydrogen–bonding and (f) noncovalent interactions, respectively. A gray sphere represents the position of the primary atom in each energy term. Open bars represent covalent bonds, while dotted lines are noncovalent bonds.
Fig. 2. (A) Schematic of the spatial decomposition consists of sub-domains Ω0–Ω3. On the sub-domain Ω0 atom information from neighbor domains (Ω1, Ω2 and Ω3 are stored in the buffer layer with the thickness Lb shown as shaded area surrounding Ω0. The primary atoms in each cluster (gray spheres) must be in Ω0, while other atoms (white spheres) may be copied from neighbor domains. (B) An example of 4-body atom configuration on a domain boundary. The explicit atoms are shown with solid-line spheres and bonds (i,j,k, and l), and so are implicit atoms with dotted-line spheres and bonds (m1–m5). Here jth atom is the primary of the cluster. In this example, the deepest level from the primary atom is lmax=3.
Fig. 3. Isogranular parallel efficiency of the parallel ReaxFF algorithm as a function of the number of processors on (A) IBM Blue Gene/L with the number of atoms per processor N/P=10,752, (B) SGI Altix 3000 with N/P=36,288, and (C) AMD dual-core Opteron cluster with N/P=107,520.
Fig. 4. Execution time of the parallel ReaxFF program as a function of the number of atoms on 131,072-processor IBM Blue Gene/L, 1920-processor SGI Altix 3000, and 2048-core AMD Opteron.
Table 1.
Bond-order calculation algorithm

Table 2.
Classification of interaction functions
a Noncovalent interactions. The numbers in the parentheses are the greatest degrees of the neighbor atom effect.
Table 3.
Force calculation arising from the derivative of BOij

Table 4.
Force-calculation arising from the derivative of Δi

Table 5.
1-body force calculation

Table 6.
2-body force calculation

Table 7.
3-body force calculation

Table 8.
4-body force calculation

Table 9.
Hydrogen–bonding force calculation

Table 10.
Noncovalent force calculation

Table 11.
Electronegativity-equalization computation

Table 12.
Multiple time-step algorithm

Table A.1
The parameters in the bond-order functions

Table A.2
The parameters in the 1-body energy

Table A.3
The parameters in the Ebond

Table A.4
The parameters in the 3-body energy

Table A.5
The parameters in the 4-body energy

Table A.6
The parameters in the hydrogen–bonding energy

Table A.7
The parameters in the noncovalent energy


Corresponding author.