A Nonorthogonal State-Interaction Approach for Matrix Product State Wave Functions

Stefan Knecht, Sebastian Keller, Jochen Autschbach, Markus ReiherJournal of Chemical Theory and Computation 2016 · DOI · arXiv

Introduction

In the field of quantum chemistry, understanding how a molecule transitions between different electronic states is fundamental to modeling phenomena such as light absorption, emission, and spin-state changes. These transitions are quantified through "transition matrix elements." While calculating these elements is straightforward when states share a common set of orthogonal orbitals, accuracy often demands that orbitals be optimized specifically for each individual state.

This state-specific optimization leads to a significant mathematical challenge: the orbitals of State A become nonorthogonal to those of State B, breaking the standard Slater-Condon rules used to calculate interactions. While solutions for this problem have existed for traditional methods like Complete Active Space Self-Consistent Field (CASSCF), they were previously unavailable for the Density Matrix Renormalization Group (DMRG) approach. This paper bridges that gap, introducing a state-interaction framework for Matrix Product State (MPS) wave functions.

The Challenge of Nonorthogonality

Traditional high-level methods like CASSCF are limited by exponential scaling, typically hitting a computational wall at an active space of approximately 18 electrons in 18 orbitals. DMRG, which utilizes the MPS framework, overcomes this by providing polynomial scaling, allowing for much larger active spaces. However, the lack of a nonorthogonal state-interaction (SI) method meant that researchers using DMRG often had to use "state-averaged" orbitals—a compromise that can reduce the physical accuracy of the individual states.

Methodology: The MPS-SI Approach

The authors implement a generalized state-interaction approach (MPS-SI) by adapting a classic technique known as the Malmqvist transformation to the MPS framework. The process involves several key steps:

  1. Biorthonormalization: Instead of forcing the orbitals of two different states to be identical, the method transforms them into a "dual" representation where the basis of one state is orthonormal to the basis of the other.
  2. Sequential Transformation: Following the Malmqvist proposal, the transformation is decomposed into a sequence of single-orbital replacements.
  3. MPO-MPS Operations: These transformations are applied as a series of Matrix Product Operators (MPOs). To keep the computational cost manageable, the researchers use Singular Value Decomposition (SVD) after each step to "compress" the resulting MPS, maintaining a stable bond dimension ($m$).
  4. Software Integration: The method was integrated into the MOLCAS software suite, enabling the calculation of complex properties like spin-orbit coupling and magnetic g-tensors.

Key Insights and Validation

The research validates the MPS-SI method through rigorous testing on heavy-element systems where relativistic effects are prominent:

  • Tellurium Dimer ($Te_2$): This system served as a benchmark for mathematical consistency. The MPS-SI results for spin-orbit coupling were nearly identical to traditional CASSCF results across various bond dimensions, proving the robustness of the MPO-based transformation.
  • Actinide Complexes ($NpO_2^{2+}$ and $PuO_2^{2+}$): These complexes are notoriously difficult to model due to massive spin-orbit coupling. The MPS-SI method produced magnetic g-factors in excellent agreement with high-level theoretical benchmarks and experimental data. For instance, in $NpO_2^{2+}$, the calculated $g_{||}$ of 4.235 matched the CASPT2-SO benchmark of 4.233 almost perfectly.

Conclusion and Future Outlook

The introduction of MPS-SI represents a significant step forward in molecular modeling. By enabling the use of state-specific orbitals within the DMRG framework, researchers can now achieve higher accuracy for large, complex systems that were previously inaccessible.

While the current implementation requires the interacting states to share the same active space size, it lays the groundwork for future developments. The authors anticipate that this approach will be vital for "surface hopping" simulations—modeling how molecules transition between energy states during chemical reactions—and for further integration with methods that account for dynamic correlation, such as DMRG-NEVPT2.

Summary updated

Abstract

We present a state-interaction approach for matrix product state (MPS) wave functions in a nonorthogonal molecular orbital basis. Our approach allows us to calculate, for example, transition and spin-orbit coupling matrix elements between arbitrary electronic states, provided that they share the same one-electron basis functions and size of the active orbital space, respectively. The key element is the transformation of the MPS wave functions of different states from a nonorthogonal to a biorthonormal molecular orbital basis representation, by exploiting a sequence of nonunitary transformations, following a proposal by Malmqvist [Int. J. Quantum Chem. 1986, 30, 479]. This is well-known for traditional wave function parametrizations but has not yet been exploited for MPS wave functions.
BibTeX
@article{knecht2016nonorthogonal,
  author = {Stefan Knecht and Sebastian Keller and Jochen Autschbach and Markus Reiher},
  title = {{A Nonorthogonal State-Interaction Approach for Matrix Product State Wave Functions}},
  journal = {Journal of Chemical Theory and Computation},
  year = {2016},
  doi = {10.1021/acs.jctc.6b00889},
}
Figure