The First Selected Configuration Interaction Implementation with Distributed Storage of the Configuration Interaction Vector (2503.10335v2)

Abstract: In recent years, the hybrid "QC+HPC" strategy -- where quantum computers screen important determinants, followed by exact diagonalization on classical computer -- has shown great potential in the study of strongly correlated systems in quantum chemistry. Last year, an IBM team proposed a novel scheme that utilizes quantum computers to select important bitstrings, which are then used to construct a spin-adapted determinant space via tensor products. In this work, we have specifically designed an entirely new algorithm for this tensor-product bitstring selected configuration interaction (SCI). For the first time worldwide, we have implemented distributed storage of the configuration interaction (CI) vector in this SCI framework, enabling massively parallel execution and scalable treatment of trillion-determinant spaces. Since this study is independent and does not involve determinant selection by quantum computers, we demonstrate our implementation's capabilities by performing large-scale full configuration interaction (FCI) computations on the supercomputer Fugaku. The largest case is N$2$ (cc-pVTZ) under D${2h}$ symmetry, involving $2.62 \times 10^{12}$ determinants after freezing the core orbitals. This scale surpasses the previous SCI record ($2 \times 10^9$ determinants) by over three orders of magnitude, and even doubles the size of the largest FCI calculation reported to date ($1.31 \times 10^{12}$ determinants). Remarkably, this result was achieved within the SCI framework, yet it exceeds the largest known full CI calculation. This work sets a new benchmark for SCI implementations and provides a scalable and efficient framework that can be extended to a wide range of SCI methodologies.