Parallel Multicoordinate Descent Methods for Full Configuration Interaction.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Full Configuration Interaction (FCI) is a high-accuracy quantum chemistry method for electronic structure calculations.
• Exact FCI solutions are computationally intractable for all but the smallest systems due to the exponential scaling of the problem size.
• Efficient algorithms are crucial for tackling larger and more complex molecular systems.

Purpose of the Study:

• To develop a novel, efficient, and parallelized algorithm for solving the electronic structure ground-state problem within the FCI framework.
• To enable accurate quantum chemical calculations for systems previously too large for exact FCI treatment.
• To leverage modern parallel computing architectures for significant speedups in electronic structure calculations.

Main Methods:

• Development of a multithreaded parallel coordinate descent full configuration interaction algorithm (mCDFCI).
• Reformulation of the FCI problem as an unconstrained minimization problem.
• Application of a modified block coordinate descent method with deterministic compression for efficient determinant prioritization and updates.
• Parallelization strategy optimized for shared-memory, multicore computing infrastructure.

Main Results:

• Demonstrated the efficiency of mCDFCI by computing an accurate benchmark energy for the chromium dimer (Cr2) in the Ahlrichs SV basis, involving 2.07 × 10^9 variational determinants.
• Generated the binding curve for the nitrogen dimer (N2) using the cc-pVQZ basis set.
• Achieved up to 79.3% parallel efficiency on 128 cores, showcasing significant computational speedup.

Conclusions:

• The mCDFCI algorithm provides a highly efficient and scalable approach for accurate electronic structure calculations.
• The method effectively handles large FCI problems, enabling accurate computations for complex molecular systems.
• The demonstrated parallel efficiency highlights the algorithm's suitability for high-performance computing environments.