GronOR: Massively parallel and GPU-accelerated non-orthogonal configuration interaction for large molecular systems

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Electronic Structure Theory

Background:

• Configuration interaction (CI) calculations are crucial for accurate electronic wave function determination.
• Handling large-scale CI computations, especially with non-orthogonal fragment wave functions, presents significant computational challenges.
• Existing computational methods often struggle with scalability and efficiency on modern high-performance computing (HPC) architectures.

Purpose of the Study:

• To introduce GronOR, a novel program package designed for non-orthogonal configuration interaction (CI) calculations.
• To enable efficient and scalable electronic structure calculations using multi-configuration molecular fragments.
• To leverage modern parallel and GPU-accelerated computing architectures for complex quantum chemistry problems.

Main Methods:

• Development of a program package (GronOR) for non-orthogonal CI calculations.
• Implementation utilizing an MPI+OpenACC/OpenMP programming approach for distributed memory and GPU acceleration.
• Employment of a task-based execution model for linear scaling and fault resiliency.
• Integration with existing electronic structure codes for molecular fragment data and integrals.

Main Results:

• Demonstration of linear scaling computational performance on pre-exascale architectures.
• Achieved hardware fault resiliency and effective execution on heterogeneous CPU/GPU systems.
• Presented benchmarks for parallel and accelerated performance.
• Analyzed the sensitivity of accuracy and performance to calculation thresholds.

Conclusions:

• GronOR provides a scalable and efficient solution for non-orthogonal CI calculations.
• The program package effectively utilizes modern HPC resources, including GPUs.
• GronOR facilitates advanced electronic structure studies by overcoming computational bottlenecks.