Density matrix renormalization group with efficient dynamical electron correlation through range separation

Erik D. Hedegård, Stefan Knecht, Jesper Skau Kielberg, Hans Jørgen Aa. Jensen, Markus ReiherThe Journal of Chemical Physics 2015 · DOI · arXiv

Abstract

We present a new hybrid multiconfigurational method based on the concept of range-separation that combines the density matrix renormalization group (DMRG) approach with density functional theory (DFT). This new method, termed DMRG-srDFT, is designed for the simultaneous description of dynamical and static electron-correlation effects in multiconfigurational electronic structure problems. By leveraging range separation, the method efficiently captures complex static correlation through DMRG and local dynamical correlation through a short-range DFT functional.

Introduction

In quantum chemistry, accurately predicting electronic behavior requires accounting for two distinct types of electron correlation. Static correlation occurs in systems with near-degenerate states, such as molecules with breaking bonds or transition metal complexes. While DMRG is highly effective at handling static correlation, it requires a computationally expensive active space. Dynamical correlation refers to the short-range interactions between electrons. Density Functional Theory (DFT) excels at capturing these local effects but often fails in multiconfigurational scenarios. Traditional methods to combine these approaches often suffer from poor scaling or "double-counting" of electron interactions. This paper introduces a hybrid approach that addresses these challenges efficiently.

Methodology: The DMRG-srDFT Approach

The core of the DMRG-srDFT method is Range Separation (RS). Using a mathematical error function, the electron-electron repulsion operator is split into two components:

  1. Long-Range (lr): Handled by the DMRG algorithm to capture large-scale, multiconfigurational wave function characteristics (static correlation).
  2. Short-Range (sr): Handled by a short-range DFT functional (e.g., srPBE) to capture local electron "cusp" behavior (dynamical correlation).

This separation avoids double-counting and allows for a more efficient computational workflow. The implementation was achieved by interfacing the MAQUIS DMRG program with the DALTON quantum chemistry suite.

Algorithmic flowchart of the DMRG-srDFT hybrid method, detailing the iterative process between the long-range DMRG calculation and the short-range DFT update.

Benchmarking and Results

The method was evaluated on prototypical small molecules ($H_2O$ and $N_2$) and complex transition metal systems from the WCCR10 benchmark set.

  • Small Molecules: For bond stretching in $H_2O$ and $N_2$, DMRG-srDFT produced potential energy curves that closely matched Full Configuration Interaction (FCI) results, even with significantly truncated active spaces. The non-parallelity error (NPE) remained remarkably stable, indicating consistent accuracy across different molecular geometries.
  • Transition Metal Complexes: In a copper complex dissociation reaction, pure DMRG with a small active space predicted a dissociation energy ($D_e$) of 132.8 kJ/mol. The hybrid DMRG-srPBE method improved this to 216.5 kJ/mol, bringing it much closer to the experimental reference of 226.7 kJ/mol.

Theoretical Insights: Reduced Entanglement

A key finding is the "regularizing" effect of the hybrid method. DMRG-srDFT is significantly less sensitive to the size of the active space than pure DMRG because the DFT functional "fills in" the dynamical correlation gaps. Furthermore, analysis of orbital entanglement—measured via single-orbital entropy and mutual information—shows that the wave function in the hybrid method is less complex.

Orbital entanglement and mutual information diagrams for N2 at various bond lengths, illustrating the reduced complexity of the wave function in the hybrid method compared to pure wave function approaches.

This reduced entanglement suggests that the DMRG component has a "simpler" task, as the messy short-range physics are handled by the DFT functional, leading to improved computational efficiency without sacrificing accuracy.

Conclusion and Future Outlook

DMRG-srDFT provides a robust framework for handling both static and dynamical correlation in large, complex systems. While the current spin-restricted implementation shows great promise, future work will focus on developing a spin-unrestricted version to better describe open-shell fragments at long distances. Additionally, the team plans to further investigate the method's reduced dependence on large basis sets, which could further lower computational costs for high-accuracy quantum chemistry simulations.

Summary updated

Abstract

We present a new hybrid multiconfigurational method based on the concept of range-separation that combines the density matrix renormalization group approach with density functional theory. This new method is designed for the simultaneous description of dynamical and static electron-correlation effects in multiconfigurational electronic structure problems.
BibTeX
@article{hedegard2015density,
  author = {Erik D. Hedegård and Stefan Knecht and Jesper Skau Kielberg and Hans Jørgen Aa. Jensen and Markus Reiher},
  title = {{Density matrix renormalization group with efficient dynamical electron correlation through range separation}},
  journal = {The Journal of Chemical Physics},
  year = {2015},
  doi = {10.1063/1.4922295},
}
Figure