Extending molecular simulation time scales: Parallel in time integrations for high-level quantum chemistry and...

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• High-Performance Computing

Background:

• Conventional molecular dynamics (MD) and ab initio molecular dynamics (AIMD) simulations are computationally intensive.
• Parallelization strategies are crucial for accelerating complex simulations.

Purpose of the Study:

• To develop and apply parallel in time simulation algorithms for MD and AIMD.
• To transform dynamics problems into unconditionally convergent root-finding problems.
• To assess the efficiency and scalability of these algorithms.

Main Methods:

• Formulation of dynamics as a root-finding problem: F(X) = [xi - f(x(i-1))] = 0.
• Application of quasi-Newton and preconditioned quasi-Newton root-finding techniques.
• Parallelization by assigning processors to time-step entries.
• Testing on realistic MD (Si atoms) and AIMD (HCl + 4H2O) simulations.

Main Results:

• Achieved speedups of up to 3.0 for MD and 14.3 for AIMD simulations.
• Demonstrated effectiveness in distributed computing environments over slow networks.
• Python scripts utilizing NWChem showed a speedup of 8.2 for AIMD.
• Reduced simulation cost from 32 s/timestep to 6.9 s/timestep for a complex AIMD calculation.

Conclusions:

• Parallel in time algorithms offer a powerful approach to accelerate MD and AIMD simulations.
• These methods are effective even without preconditioners for standard MD/AIMD.
• They enable long-time, high-level AIMD simulations at reduced cost, even with limited network bandwidth.