Parallel stored-integral and semidirect Hartree-Fock and DFT methods with data compression.

Area of Science:

• Computational chemistry
• Quantum chemistry
• Materials science

Background:

• Direct methods for calculating two-electron integrals are widely used in computational chemistry.
• Recent advancements in magnetic disk technology enhance the viability of stored-integral techniques.
• Stored-integral methods offer an alternative to recalculating basic two-electron integrals.

Purpose of the Study:

• To present efficient conventional and semidirect Hartree-Fock and Density Functional Theory (DFT) algorithms.
• To incorporate data compression for both single-processor and distributed memory parallel computers.
• To compare the performance of stored-integral algorithms against direct algorithms.

Main Methods:

• Development of efficient conventional (all integrals stored) and semidirect algorithms.
• Implementation of data compression techniques for integral storage.
• Performance evaluation on single-processor and distributed memory parallel computing systems.
• Comparison with established direct integral calculation methods.

Main Results:

• Stored-integral techniques are demonstrated to be competitive with direct methods.
• The presented algorithms, utilizing data compression, show significant efficiency gains.
• On modern personal computer hardware, the stored integral method achieved up to three times greater efficiency than the direct method.
• Total elapsed job time was significantly reduced using the stored integral approach.

Conclusions:

• Stored-integral methods, enhanced by modern hardware and efficient algorithms, offer a competitive and often superior alternative to direct methods.
• The developed algorithms with data compression provide substantial speedups for Hartree-Fock and DFT calculations.
• This work highlights the practical advantages of stored-integral techniques for computational chemistry on accessible hardware.