libreta: Computerized Optimization and Code Synthesis for Electron Repulsion Integral Evaluation.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Theoretical Chemistry

Background:

• Electron repulsion integrals (ERIs) are fundamental in quantum chemistry calculations.
• Efficient computation of ERIs is crucial for accurate and scalable electronic structure methods.
• Existing libraries face performance limitations with complex basis sets and large systems.

Purpose of the Study:

• To develop a new, highly optimized library named libreta for evaluating ERIs.
• To improve computational efficiency for both segmented and contracted Gaussian functions.
• To provide a tool for rapid algorithm development in theoretical and computational chemistry.

Main Methods:

• Integration of Obara-Saika, Dupuis-Rys-King, and McMurchie-Davidson methods.
• Optimization of recurrence relations using tree-search for angular momentum combinations.
• Dynamic selection of optimal contraction orders and coding-level optimizations (e.g., common subexpression elimination).

Main Results:

• Libreta demonstrates substantial performance improvements over libint2 (7.2-912.0%) for various basis sets.
• For heavy elements with high contraction degrees, performance increases 20-30 times.
• Direct self-consistent field (SCF) calculations using libreta are significantly faster than NWChem (144.2-495.9% speedup).

Conclusions:

• Libreta offers a significant advancement in ERI evaluation efficiency.
• The library's optimizations provide substantial speedups for quantum chemistry computations.
• Libreta is poised to be a valuable tool for researchers in theoretical and computational chemistry.