A parallel, distributed memory implementation of the adaptive sampling configuration interaction method

Area of Science:

• Quantum chemistry
• Computational physics
• High-performance computing

Background:

• Quantum system simulations utilize diverse methods, with coupled cluster and selected configuration interaction (sCI) being prominent.
• Advancements in high-performance computing (HPC) have driven the adaptation of many quantum simulation methods.
• Development of sCI methods for massively parallel architectures remains an underexplored area.

Purpose of the Study:

• To present a parallel, distributed memory implementation of the adaptive sampling configuration interaction (ASCI) approach for sCI methods.
• To address critical parallelization challenges in determinant search, selection, Hamiltonian formation, and eigenvalue calculations within ASCI.
• To enable larger and more complex quantum system simulations on parallel computing resources.

Main Methods:

• Parallel, distributed memory implementation of the adaptive sampling configuration interaction (ASCI) method.
• Application of memory-efficient determinant constraints for load balancing during the determinant search.
• Utilizing variational eigenvalue calculations for the ASCI method on massively parallel systems.

Main Results:

• Demonstrated near-optimal speedup for ASCI calculations on up to 16,384 CPUs.
• Successfully performed the largest variational ASCI calculation to date for the Cr2 molecule (24 electrons, 30 orbitals) involving up to 3x10^8 determinants.
• Validated the efficiency and scalability of the parallel ASCI implementation.

Conclusions:

• The developed parallel ASCI implementation significantly advances the capability to simulate large quantum systems.
• This work overcomes key parallelization hurdles, making sCI methods more accessible for HPC environments.
• The presented approach paves the way for tackling even larger and more complex quantum mechanical problems.