A massively parallel algorithm for the solution of constrained equations of motion with applications to large-scale, long-time molecular dynamics simulations

Conference · Wed Dec 31 04:00:00 EST 1997

OSTI ID:559995

Fijany, A ^[1]; Coley, T R ^[2]; Cagin, T; Goddard, III, W A ^[3]

---

1. Jet Propulsion Lab., Pasadena, CA (United States)
2. Virtual Chemistry, Inc., San Diego, CA (United States)
3. California Institute of Technology, Pasadena, CA (United States)

Successful molecular dynamics (MD) simulation of large systems (> million atoms) for long times (> nanoseconds) requires the integration of constrained equations of motion (CEOM). Constraints are used to eliminate high frequency degrees of freedom (DOF) and to allow the use of rigid bodies. Solving the CEOM allows for larger integration time-steps and helps focus the simulation on the important collective dynamics of chemical, biological, and materials systems. We explore advances in multibody dynamics which have resulted in O(N) algorithms for propagating the CEOM. However, because of their strictly sequential nature, the computational time required by these algorithms does not scale down with increased numbers of processors. We then present the new constraint force algorithm for solving the CEOM and show that this algorithm is fully parallelizable, leading to a computational cost of O(N/P+IogP) for N DOF on P processors.

• OSTI ID: 559995
• Report Number(s): CONF-970443--
• Country of Publication: United States
• Language: English

Related Subjects

55 BIOLOGY AND MEDICINE
BASIC STUDIES
99 GENERAL AND MISCELLANEOUS
ALGORITHMS
BIOCHEMISTRY
EQUATIONS OF MOTION
PARALLEL PROCESSING
SIMULATION