Techniques for high-performance construction of Fock matrices.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• High-Performance Computing

Background:

• Fock matrix construction is computationally intensive.
• Efficient parallel algorithms are crucial for large-scale quantum chemistry.
• Existing methods may not fully leverage modern parallel architectures.

Purpose of the Study:

• To develop and present high-performance techniques for Fock matrix construction.
• To optimize calculations for shared and distributed memory parallel computers.
• To enhance the efficiency of quantum chemical simulations using Gaussian basis sets.

Main Methods:

• Vectorized calculation of electron repulsion integrals using batching.
• Multithreaded Fock matrix summation with atomic operations and thread-local copies.
• A globally accessible matrix class for distributed Fock and density matrices with batched remote memory access.
• Density fitting exploiting symmetry and sparsity for performance comparison.

Main Results:

• Demonstrated efficient vectorization of primitive integral calculations.
• Investigated effective multithreaded summation strategies.
• Introduced a novel matrix class reducing synchronization costs in distributed environments.
• Showcased performance benefits of density fitting over direct methods.

Conclusions:

• The presented techniques significantly enhance Fock matrix construction performance on parallel systems.
• The GTFock library implements these optimizations for practical application.
• These advancements contribute to accelerating quantum chemistry research through efficient computation.