The GPU-enabled divide-expand-consolidate RI-MP2 method (DEC-RI-MP2).

Area of Science:

• Computational Chemistry
• High-Performance Computing (HPC)
• Quantum Chemistry

Background:

• The Resolution of the Identity second-order Møller-Plesset perturbation (RI-MP2) method is crucial for accurate electronic structure calculations.
• The Divide-Expand-Consolidate (DEC) approach enhances the scalability of RI-MP2 for large systems.
• Porting computationally intensive methods to accelerators like GPUs is vital for modern scientific research.

Purpose of the Study:

• To implement the DEC-RI-MP2 method on Graphics Processing Units (GPUs) using OpenACC.
• To evaluate the performance gains and scalability improvements achieved through GPU acceleration.
• To demonstrate the method's applicability to large supramolecular complexes.

Main Methods:

• Porting the DEC-RI-MP2 algorithm to GPUs utilizing OpenACC compiler directives.
• Developing a hybrid Message Passing Interface (MPI)/OpenMP/OpenACC implementation for cross-platform compatibility.
• Benchmarking the GPU-enabled code on a reduced S12L test set of supramolecular complexes.

Main Results:

• OpenACC implementation efficiently accelerates the rate-determining step of DEC-RI-MP2 with minimal effort.
• GPU acceleration improves load balancing and overall scaling of the DEC algorithm.
• The hybrid implementation shows scalable and portable performance on heterogeneous HPC architectures.

Conclusions:

• The GPU-enabled DEC-RI-MP2 method offers significant computational advantages for large-scale electronic structure calculations.
• The implementation demonstrates efficient performance and scalability on modern HPC systems.
• The method is generally applicable to molecules with complex electronic structures, consistent with previous findings.