Doubling the size of quantum selected configuration interaction based on seniority-zero space and its application to QC-QSCI-AFQMC

TL;DR

The paper tackles the qubit bottleneck in quantum-chemistry simulations by developing DOCI-QSCI, which samples from the seniority-zero space to build an efficient active subspace and then expands the determinant set via Cartesian products to include higher-seniority configurations. This DOCI-QSCI wave function serves as the trial for ph-AFQMC, producing DOCI-QSCI-AFQMC that recovers dynamical correlation across the full orbital space. Across H$_6$, N$_2$, and BODIPY+O$_2$, the approach achieves high accuracy: on hardware for H$_6$ it matches CASCI-AFQMC/HCI references, while for N$_2$ and BODIPY+O$_2$ it delivers results competitive with multireference methods and superior to single-reference CCSD(T). The work demonstrates that halving the qubit requirement via seniority-zero sampling, coupled with classical post-processing, enables scaling quantum-chemical calculations toward larger, more realistic active spaces.

Abstract

We propose doubly occupied configuration interaction-quantum selected configuration interaction (DOCI-QSCI), which samples from the seniority-zero space. While the use of this space effectively doubles the qubit budget, equaling the number of spatial orbitals, this sector restriction can compromise quantitative accuracy. To compensate for this, we expand sampled bitstrings via their Cartesian product into a larger space that includes seniority-breaking determinants. The resulting wave function is also proposed using the trial state in phaseless auxiliary-field quantum Monte Carlo (ph-AFQMC) to recover dynamical correlations across the full orbital space (DOCI-QSCI-AFQMC). We evaluate the proposed methods on the H6 chain, N2 dissociation, and the addition of singlet O2 to a BODIPY dye. For the H6 chain, DOCI-QSCI-AFQMC reproduces the accuracy of the level of the complete-active-space counterpart with the quantum device ibm kobe. For N2 and BODIPY-O2, with (14e, 28o) and up to (20e, 20o) active spaces, it yields reasonable results, whereas single-reference CCSD(T) fails qualitatively. These results demonstrate that the DOCI-QSCI doubles the orbital space accessible to conventional QSCI and subsequent ph-AFQMC post-processing delivers reasonably high accuracy.

Figures (5)

• Figure 1: DOCI-QSCI-AFQMC workflow.
• Figure 2: (a) Energy curve and (b) energy difference relative to HCI for the one-dimensional H$_6$ chain. The horizontal axis is the distance between adjacent hydrogen atoms. Simul. denotes results obtained with a classical simulator, and $\epsilon$ is the cutoff for the selected CI space enlargement. Sampling was performed $10^5$ shots at each point.
• Figure 3: Relative energy curves of N$_2$ in Hartree. Sampling was performed $10^5$ shots at each point. For ph-AFQMC calculations, we used 640 walkers. RMR-CCSD and RMR-CCSD(T) data were taken from Ref. Li2008full.
• Figure 4: Reaction mechanism for the addition of singlet oxygen to a BODIPY dye. Molecular graphics were created using VMD humphrey1996vmd.
• Figure 5: Energy profile of the reaction between BODIPY and O$_2$.