Unified MPI Parallelization of Wave Function Methods: iCIPT2 as a Showcase.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• High-Performance Computing

Background:

• Accurate quantum chemical calculations are crucial for understanding molecular behavior.
• Handling large molecular systems demands efficient computational methods and high-performance computing (HPC).
• Existing methods often struggle with scalability for complex systems or require high accuracy.

Purpose of the Study:

• To develop a unified Message Passing Interface (MPI) parallelization for nonrelativistic and relativistic wave function methods.
• To enhance the MetaWave platform for efficient large-scale quantum chemical computations.
• To enable accurate calculations for challenging chemical systems previously intractable.

Main Methods:

• Unified MPI parallelization using dynamically scheduled loops and global reductions.
• Implementation within the MetaWave platform, abstracting computational steps.
• Showcasing the improved iCIPT2 (improved Correlation-consistent Iterative Pair-Interaction Theory) method.

Main Results:

• Achieved high parallel efficiencies (94% for perturbation, 89% for whole calculation) on 16 nodes (1024 cores) using iCIPT2.
• Enabled large active space calculations, providing benchmarks for automerization of cyclobutadiene, benzene ground-state energy, and ozone potential energy profile.
• Demonstrated that iCIPT2 error scales with the number of configuration state functions.

Conclusions:

• The unified MPI parallelization strategy significantly enhances computational efficiency for quantum chemistry.
• The developed methods facilitate accurate electronic structure calculations for large and complex molecular systems.
• This work paves the way for more advanced computational studies in chemistry and materials science.