Improving the scaling and performance of multiple time stepping-based molecular dynamics with hybrid density...

Area of Science:

• Computational Chemistry
• Materials Science
• High-Performance Computing

Background:

• Generalized gradient approximation (GGA) and hybrid density functionals are crucial for ab initio molecular dynamics (AIMD) simulations.
• Hybrid functionals offer higher accuracy for molecular and condensed matter systems but are computationally expensive.
• Existing AIMD simulations using hybrid functionals are significantly slower than GGA-based methods.

Purpose of the Study:

• To implement and evaluate the MTACE and s-MTACE methods for accelerating hybrid density functional AIMD simulations.
• To leverage task-group based parallelization on modern high-performance computing (HPC) platforms.
• To identify and address computational bottlenecks within the s-MTACE method.

Main Methods:

• Implementation of MTACE and s-MTACE methods within the CPMD program package.
• Utilized task-group based parallelization for efficient computation on multi-core HPC systems.
• Employed a multiple time step integrator and adaptively compressed exchange operator formalism.

Main Results:

• Demonstrated a significant performance boost for hybrid density functional AIMD simulations through the implemented methods.
• The task-group parallelization effectively utilized a large number of compute cores on modern HPC platforms.
• Identified a key computational bottleneck in the s-MTACE method, with a proposed solution.

Conclusions:

• The implemented MTACE and s-MTACE methods, combined with task-group parallelization, offer substantial speed-up for hybrid DFT AIMD.
• This approach enhances the feasibility of accurate AIMD simulations for complex systems on large-scale computing resources.
• Further optimization is possible by addressing the identified computational bottleneck in the s-MTACE algorithm.