Fast transformations between configuration state function and Slater determinant bases for direct configuration...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Electronic Structure Theory

Background:

• Full Configuration Interaction (CI) calculations are essential for accurate electronic structure but computationally demanding.
• Traditional methods often face challenges with large system sizes and convergence.
• Slater determinant (SD) and Configuration State Function (CSF) bases offer different advantages in CI computations.

Purpose of the Study:

• To develop a hybrid full CI program utilizing both CSF and SD bases.
• To leverage the strengths of SDs for efficient matrix-vector (σ = Hc) formation.
• To exploit the benefits of CSFs for reduced CI vector length and improved convergence.

Main Methods:

• Implementation of a hybrid CSF-SD basis full CI program.
• Development of direct CSF-SD and SD-CSF basis transformation algorithms.
• GPU acceleration of the transformation algorithms to minimize overhead.
• Performance evaluation using a Hartree-Fock Complete Active Space (HF-CAS) calculation on ethylene.

Main Results:

• The hybrid program achieves fast σ = Hc formation using SD algorithms.
• CSF basis offers smaller CI vector lengths and more robust convergence.
• GPU acceleration reduces transformation time per iteration to approximately 15% of σ formation time.
• A HF-CAS-(16,16)-CI/6-31G calculation of ethylene showed a 2.0x speedup in time-to-solution.

Conclusions:

• The hybrid CSF-SD CI approach offers significant computational savings.
• Reduced memory footprint and faster convergence are key benefits.
• This method provides a more efficient route to accurate electronic structure calculations.