Spin-adapted matrix product states and operators

Sebastian Keller, Markus ReiherThe Journal of Chemical Physics 2016 · DOI · arXiv

Abstract

Matrix product states (MPSs) and matrix product operators (MPOs) allow an alternative formulation of the density matrix renormalization group algorithm introduced by White. Here, we describe how non-abelian spin symmetry can be exploited in MPSs and MPOs by virtue of the Wigner-Eckart theorem at the example of the spin-adapted quantum chemical Hamiltonian operator.

Introduction

The Density Matrix Renormalization Group (DMRG) algorithm is a cornerstone for simulating strongly correlated quantum systems. However, standard implementations typically exploit only abelian symmetries, such as particle number or the projection of spin along a single axis ($S_z$).

In physical systems, total spin $S$ is a conserved quantity representing a non-abelian symmetry. Neglecting this symmetry leads to two primary complications. First, it results in computational inefficiency, as redundant information across different spin projections ($m = -S, \dots, S$) is stored. Second, it introduces the risk of "spin contamination," where a calculation intended for a specific spin state (e.g., a singlet) may inadvertently collapse into a lower-energy state of different symmetry (e.g., a triplet).

While spin symmetry has been applied to simple 1D spin chains, implementing it for the complex, many-body Hamiltonian of quantum chemistry presents significant bookkeeping challenges. This work introduces a formal framework to integrate $SU(2)$ spin symmetry directly into the Matrix Product State (MPS) and Matrix Product Operator (MPO) formalism.

Methodology: Spin-Adaptation via Wigner-Eckart

The researchers utilize an MPO-based formulation of DMRG, introducing a modular approach to spin-adaptation:

  • Wigner-Eckart Theorem: By applying this theorem, the authors separate the "physical" reduced matrix elements from the "geometric" Clebsch-Gordan coefficients. This allows for the storage of only reduced elements, significantly compressing the required memory.
  • Irreducible Tensor Operators: Elementary operations, such as electron creation and annihilation, are treated as tensors transforming according to $SU(2)$ rules.
  • Wigner-9j Symbols: A sophisticated contraction formula using Wigner-9j symbols was derived to perform MPS-MPO operations entirely within the reduced space. This ensures the algorithm never needs to expand matrices to their full, redundant size during computation.
  • Second-Generation DMRG: This framework extends previous MPO-based DMRG methods designed for the high complexity of chemical systems where all-to-all electron interactions are present.

Benchmarking on Dioxygen ($O_2$)

The method was validated using the Dioxygen ($O_2$) molecule in a Dunning cc-pVTZ basis set. This system is a classic test case due to its triplet ground state ($^3\Sigma_g$) and low-lying singlet excited state ($^1\Delta_g$). Results were compared against experimental data and CCSD(T) (Coupled Cluster with Single, Double, and Triple excitations) calculations.

Electronic energies for the singlet and triplet states of O2. The table highlights the energy discrepancy when spin symmetry is not enforced (Sz=0 vs S=0).

Results and Discussion

Information Compression and Efficiency

The use of reduced matrix elements results in block-diagonal tensors that are significantly smaller than their non-adapted counterparts. This mathematical exploitation of symmetry leads to more efficient calculations without sacrificing accuracy.

State Targeting and Spin Purity

A critical advantage of this method is the ability to target specific spin states. As shown in the benchmarking data, a standard $S_z=0$ calculation intended for the singlet state fails to maintain purity, yielding an energy (-150.147 $E_H$) that is artificially low because it is contaminated by the triplet state. By enforcing $S=0$, the simulation is "locked" into the correct symmetry sector, yielding a pure singlet energy (converging toward -150.118 $E_H$).

Singlet-Triplet Gap

The adiabatic gap for $O_2$ was calculated at 97.1 kJ/mol using the spin-adapted DMRG. This result shows excellent agreement with the experimental value of 94.7 kJ/mol, demonstrating the method's predictive power for electronic transitions.

Conclusion

The integration of non-abelian $SU(2)$ symmetry into the MPS/MPO framework provides a robust solution for targeting specific electronic states in complex molecules while improving computational efficiency. Future work may involve incorporating higher-order spatial symmetries (such as the $D_{\infty h}$ group for linear molecules) and combining DMRG with methods that better capture dynamic correlation to achieve absolute chemical accuracy.

Reference: Keller, S., & Reiher, M. (2016). Spin-adapted matrix product states and operators. arXiv:1602.01145v2 [physics.chem-ph].

Summary updated

Abstract

Matrix product states (MPSs) and matrix product operators (MPOs) allow an alternative formulation of the density matrix renormalization group algorithm introduced by White. Here, we describe how non-abelian spin symmetry can be exploited in MPSs and MPOs by virtue of the Wigner-Eckart theorem at the example of the spin-adapted quantum chemical Hamiltonian operator.
BibTeX
@article{keller2016spin,
  author = {Sebastian Keller and Markus Reiher},
  title = {{Spin-adapted matrix product states and operators}},
  journal = {The Journal of Chemical Physics},
  year = {2016},
  doi = {10.1063/1.4944921},
}
Figure