An efficient matrix product operator representation of the quantum chemical Hamiltonian

Sebastian Keller, Michele Dolfi, Matthias Troyer, Markus ReiherThe Journal of Chemical Physics 2015 · DOI · arXiv

Abstract

We describe how to efficiently construct the quantum chemical Hamiltonian operator in matrix product form. We present its implementation as a density matrix renormalization group (DMRG) algorithm for quantum chemical applications. Existing implementations of DMRG for quantum chemistry are based on the traditional formulation of the method, which was developed from the point of view of Hilbert space decimation and attained higher performance compared to straightforward implementations of matrix product based DMRG. The latter variationally optimizes a class of ansatz states known as matrix product states, where operators are correspondingly represented as matrix product operators (MPOs). The MPO construction scheme presented here eliminates the previous performance disadvantages while retaining the additional flexibility provided by a matrix product approach, for example, the specification of expectation values becomes an input parameter. In this way, MPOs for different symmetries - abelian and non-abelian - and different relativistic and non-relativistic models may be solved by an otherwise unmodified program.

Introduction

In quantum chemistry, the Density Matrix Renormalization Group (DMRG) is a critical tool for simulating strongly correlated molecular systems where standard methods often fail. Historically, a divide has existed between "first-generation" traditional DMRG, which is high-performing but rigid, and "second-generation" Matrix Product State (MPS)/Matrix Product Operator (MPO) frameworks, which offer mathematical elegance and flexibility but suffered from poor scaling in quantum chemical contexts.

The primary bottleneck for MPO-based methods is the molecular Hamiltonian, which contains $O(L^4)$ interaction terms for $L$ orbitals. A naive MPO construction leads to $O(L^5)$ complexity, rendering it impractical for large systems. This work introduces a compact construction scheme that achieves $O(L^4)$ scaling, bringing MPO-based DMRG to performance parity with traditional implementations.

Methodology: Compact MPO Construction

The core innovation of this approach is the systematic compression of the Hamiltonian terms by treating them as operator strings across an orbital lattice. Two primary optimization techniques are employed:

  • Forking: Terms sharing identical operators on the left side of the chain are collapsed into a single branch, only "forking" when the terms diverge.
  • Merging: Terms sharing identical operators on the right side are "merged" into a common tail.
Forking and merging logic used to compress the MPO representation by identifying common operator prefixes and suffixes.

By applying these operations, the MPO bond dimension is minimized. The resulting operator logic is decoupled from the solver, allowing the same program (implemented as QCMaquis) to handle various symmetries (Abelian and non-Abelian) and relativistic models without modification. The local structure of these compressed MPO blocks is illustrated below, showing how different interaction types are encoded.

Structure of local operator blocks within the MPO for various interaction terms, demonstrating the compact construction.

Additionally, the method incorporates fermionic anticommutation rules using Jordan-Wigner transformations and auxiliary sign operators directly within the MPO structure. Orbital ordering is optimized using Fiedler ordering based on mutual information to minimize entanglement and improve convergence.

Benchmarking and Results

The efficiency of the compact MPO scheme was evaluated using graphene-like fragments (G2 with 24 $\pi$-orbitals and G3 with 54 $\pi$-orbitals) in a minimal STO-3G basis set.

Key findings include:

  • Performance Parity: The $O(L^4)$ scaling per sweep matches the performance of hand-optimized traditional DMRG codes.
  • Flexibility: The framework allows for state-specific excited state targeting through orthogonalization and enables "operator arithmetic," such as calculating energy variance $\Delta E = \langle H^2 \rangle - \langle H \rangle^2$ (though $H^2$ remains computationally demanding).
  • Precision: For the G2 fragment, results were verified against high-precision references ($m=3000$), showing an error of only $\pm 0.05$ mH per carbon atom.

Conclusion

This work successfully bridges the gap between the flexibility of modern MPO-based DMRG and the high performance of traditional implementations. By optimizing the Hamiltonian construction through forking and merging, the authors provide a general-purpose tool for quantum chemistry. Future work will focus on scaling to larger basis sets and exploring relativistic effects like the Coulomb-Dirac Hamiltonian.

Summary updated

Abstract

We describe how to efficiently construct the quantum chemical Hamiltonian operator in matrix product form. We present its implementation as a density matrix renormalization group (DMRG) algorithm for quantum chemical applications. Existing implementations of DMRG for quantum chemistry are based on the traditional formulation of the method, which was developed from the point of view of Hilbert space decimation and attained higher performance compared to straightforward implementations of matrix product based DMRG. The latter variationally optimizes a class of ansatz states known as matrix product states, where operators are correspondingly represented as matrix product operators (MPOs). The MPO construction scheme presented here eliminates the previous performance disadvantages while retaining the additional flexibility provided by a matrix product approach, for example, the specification of expectation values becomes an input parameter. In this way, MPOs for different symmetries - abelian and non-abelian - and different relativistic and non-relativistic models may be solved by an otherwise unmodified program.
BibTeX
@article{keller2015efficient,
  author = {Sebastian Keller and Michele Dolfi and Matthias Troyer and Markus Reiher},
  title = {{An efficient matrix product operator representation of the quantum chemical Hamiltonian}},
  journal = {The Journal of Chemical Physics},
  year = {2015},
  doi = {10.1063/1.4939000},
}
Figure