Hybrid Distributed/Shared Memory Model for the RI-MP2 Method in the Fragment Molecular Orbital Framework.

Area of Science:

• Computational Chemistry
• High-Performance Computing

Background:

• The fragment molecular orbital (FMO) method enables large-scale electronic structure calculations.
• Parallel computing is essential for accelerating complex molecular simulations.

Purpose of the Study:

• To develop and evaluate a hybrid parallel approach for the FMO method.
• To enhance computational performance by combining distributed and shared memory parallelism.

Main Methods:

• Integration of the general distributed data interface (GDDI) with OpenMP parallel middleware.
• Implementation of a threading multilevel parallelism strategy for FMO calculations.
• Performance evaluation using the FMO/RI-MP2 method for molecular clusters.

Main Results:

• The hybrid GDDI/OpenMP model demonstrates improved efficiency over the pure GDDI model.
• Benchmark calculations show node linear scaling and up to 10x speedups.
• Efficient utilization of multiprocessor resources for molecular cluster calculations.

Conclusions:

• The hybrid GDDI/OpenMP approach offers a significant performance boost for FMO calculations.
• This parallel strategy is effective for simulating large molecular systems, including water and ionic liquid clusters.