Scalable fine-grained parallelization of plane-wave-based ab initio molecular dynamics for large supercomputers.

Area of Science:

• Material science
• Chemistry
• Solid-state physics
• Biophysics

Background:

• Accurate simulation of chemical bond dynamics requires forces from electronic structure calculations, not empirical laws.
• Ab initio molecular dynamics (AIMD) enables studying bond breaking/forming but is computationally intensive.
• Generalized gradient approximation (GGA) with plane-wave basis sets and pseudopotentials offers an efficient AIMD approach.

Purpose of the Study:

• To develop a scalable parallelization strategy for Car-Parrinello molecular dynamics (CPMD).
• To overcome limitations in parallel scaling of CPMD on large processor systems.
• To enable simulations on next-generation supercomputers with thousands of processors.

Main Methods:

• Implemented a novel scalable parallelization strategy for CPMD using Charm++.
• Utilized processor virtualization to interleave calculation elements with low latency.
• Benchmarked the method using a system of 32 water molecules.

Main Results:

• Achieved unprecedented parallel scaling for CPMD on over 1500 processors.
• Demonstrated scalability beyond the limits of standard parallel approaches.
• Successfully simulated a 32-water molecule system on a massively parallel platform.

Conclusions:

• The novel Charm++ based parallelization strategy significantly enhances CPMD scalability.
• This advancement allows for longer simulation times and better phase space sampling.
• Opens new avenues for investigating complex systems in materials science and chemistry.