Abstract
We employ tensor network methods for the study of the seniority quantum number—defined as the number of unpaired electrons in a many-body wave function—in molecular systems. Seniority-zero methods recently emerged as promising candidates to treat strong static correlations in molecular systems, but are prone to deficiencies related to dynamical correlation and dispersion. We systematically resolve these deficiencies by increasing the allowed seniority number using tensor network methods. In particular, we investigate the number of unpaired electrons needed to correctly describe the binding of the neon and nitrogen dimer and the $D_{6h}$ symmetry of benzene.
Introduction
In quantum chemistry, capturing the "correlation energy"—the difference between mean-field Hartree-Fock models and exact solutions—is essential for accurate molecular modeling. This energy is typically categorized into static correlation (strong effects like bond breaking) and dynamical correlation (weak, short-range interactions). Seniority-zero methods, such as Doubly Occupied Configuration Interaction (DOCI), excel at static correlation but fail in three key areas: missing dynamical correlation, artificial symmetry breaking (e.g., in benzene), and the absence of London dispersion forces. This study introduces a systematic framework to reintroduce unpaired electrons, defined by the seniority quantum number ($\nu$), to address these limitations.
Methods and Computational Efficiency
The authors utilize Tensor Network States (TNS), specifically the Density Matrix Renormalization Group (DMRG) and Three-legged Tree Tensor Network States (T3NS). By treating seniority as a $U(1)$ symmetry, they implement seniority-restricted subspaces within the tensors. This allows for a controlled increase of allowed seniority ($\nu = 0, 2, 4, \dots$) to observe convergence.
A significant advantage of this approach is computational efficiency. Seniority-zero tensor networks (DOCI-TNS) exhibit quadratic scaling $\mathcal{O}(k^2)$ with the number of orbitals, a substantial improvement over the quartic $\mathcal{O}(k^4)$ scaling of standard quantum chemical DMRG methods.
Results and Discussion
Triple-Bond Dissociation in $N_2$
The nitrogen dimer ($N_2$) serves as a benchmark for triple-bond breaking. The study finds that the importance of seniority is highly orbital-dependent. In DOCI-optimized orbitals, the seniority-two sector is largely decoupled; significant improvements only emerge at seniority-four. As seniority increases to $\nu \leq 8$ or $10$, the TNS results achieve "quasi-FCI" accuracy, effectively matching exact solutions.
Dispersion Forces and Symmetry Restoration
For the neon dimer ($Ne_2$), seniority-zero and seniority-two models fail to predict any binding. Binding only emerges at seniority-four, highlighting that at least four unpaired electrons are necessary to describe the polarization effects responsible for van der Waals forces. In the case of benzene, seniority-zero models incorrectly predict a distorted geometry. While seniority-four provides a major correction, the study reveals that seniority-eight (four broken pairs) is required to fully restore the correct $D_{6h}$ hexagonal symmetry in the STO-6G basis.
Conclusion
While seniority-restricted TNS methods offer a powerful tool for systematically improving molecular simulations, challenges remain. These methods are not strictly size-consistent, requiring careful corrections for systems like the neon dimer. Furthermore, the results are sensitive to orbital optimization. Despite these limitations, the efficiency of DOCI-TNS and the systematic convergence with seniority provide a robust path forward for treating complex electronic correlations in large molecular systems.
Summary updated