Tailored coupled cluster theory in varying correlation regimes

Maximilian Mörchen, Leon Freitag, Markus ReiherThe Journal of Chemical Physics 2020 · DOI · arXiv

TL;DR

Tailored Coupled Cluster (TCC) theory aims to bridge the gap between simple single-reference methods and complex multi-reference approaches. By using "active space" calculations to guide the optimization of the wave function, TCC can handle tricky "static correlation" problems like bond-breaking while maintaining the efficiency of standard Coupled Cluster theory.

What Problem Does This Paper Address?

In computational chemistry, we often face a trade-off. Single-Reference Coupled Cluster (SRCC), specifically CCSD(T), is the "gold standard" for accuracy and efficiency, but it fails miserably when a molecule has multiple important electronic configurations (known as static correlation). This happens during bond breaking, in excited states, or in transition metal complexes.

To fix this, researchers use Multi-Reference (MR) methods, but these are often prohibitively expensive, mathematically complex, and prone to numerical instabilities. This paper investigates Tailored Coupled Cluster (TCCSD), a hybrid approach that "tailors" a standard single-reference calculation using information from a preceding, smaller multi-configurational calculation. The authors seek to understand exactly how reliable this hybrid is by comparing it against the "exact" (but computationally impossible for large systems) Full Configuration Interaction (FCI) method.

Methods

The authors focus on the TCCSD (Tailored Coupled Cluster Singles and Doubles) variant. The core idea is the split-amplitude ansatz:

  1. Active Space Calculation: First, a small subset of "important" orbitals (the active space) is solved using a multi-configurational method (like CASCI).
  2. Freezing Amplitudes: The excitation amplitudes found in the active space are kept frozen.
  3. External Optimization: The remaining amplitudes (the "external" space) are optimized using the standard CCSD equations.

The paper introduces a rigorous way to measure quality: the overlap error ($\Delta_{ov}$). Instead of just looking at the final energy, they measure how much the TCC wave function actually "overlaps" with the exact FCI wave function. They also utilize entanglement entropy (via the AutoCAS program) to systematically identify which orbitals belong in the active space based on how "entangled" they are with others.

Datasets

Because the authors needed exact FCI results for comparison, they used two well-known "stress-test" model systems:

  • P4 Model: Two hydrogen molecules ($H_2$) arranged in a rectangle. By changing the distance ($\alpha$) between the molecules, they can transition the system from a simple single-reference regime to a highly degenerate multi-reference regime.
  • H8 Model: Eight hydrogen atoms arranged in an octagon. This is a more complex system used to test how TCC handles larger active spaces and higher levels of symmetry.

Key Insights and Ideas

  • Reference Dependence: Even though TCC incorporates multi-reference information, it still relies on a single "starting" determinant. The authors found that TCC results are not "smooth" if you switch the reference determinant, confirming it isn't a "genuine" multi-reference method.
  • The Power of CASSCF Orbitals: Using Hartree-Fock (HF) orbitals for TCC in multi-reference regions leads to poor results. Switching to orbitals optimized for the active space (CASSCF) significantly improves the wave function overlap with the exact result.
  • Diagnostic Deception: Standard diagnostics like $T_1$ and $D_1$ (used to tell if a calculation is reliable) can be misleading. In some multi-reference regions, these diagnostics suggested the wave function was fine, even when the overlap error with the exact FCI result was actually quite high.
  • Entanglement as a Guide: The study shows that using orbital entanglement to pick the active space is highly effective. For the H8 model, a larger active space (CAS(8,8)) chosen via entanglement analysis significantly reduced errors compared to a minimal CAS(2,2) space.

Key Results

  • Improved Wave Functions: In the multi-reference regions of the P4 and H8 models, TCCSD consistently produced wave functions with a higher overlap with the exact FCI result than standard CCSD.
  • Symmetry Breaking Mitigation: The authors found that errors caused by symmetry breaking in the wave function can be reduced by choosing the determinant with the largest weight in the active space as the reference for the TCC calculation.
  • Quantitative Accuracy: For the H8 model in the high-symmetry (most difficult) region, TCCSD with a large active space brought the overlap error down significantly, though it remained above the 0.02 threshold typically desired for "high-quality" wave functions.

Limitations and Future Work

  • Size-Consistency: A major drawback of TCC is that it is not strictly "size-consistent." This means the error doesn't scale linearly with the size of the system, which is particularly visible in the dissociation limit (when molecules are pulled far apart).
  • Active Space Size: While TCC is cheaper than full MR methods, it still requires a carefully chosen active space. If the system requires a very large active space, TCC might eventually need to be paired with even more advanced methods like DMRG or FCIQMC.
  • Missing Triples: This study focused on TCCSD (Singles and Doubles). The authors note that adding perturbative "triples" corrections from the active space is a necessary next step to reach "gold standard" accuracy for multi-reference systems.

Summary updated

Abstract

The tailored coupled cluster (TCC) approach is a promising ansatz that preserves the simplicity of single-reference coupled cluster theory while incorporating a multi-reference wave function through amplitudes obtained from a preceding multi-configurational calculation. Here, we present a detailed analysis of the TCC wave function based on model systems, which require an accurate description of both static and dynamic correlation. We investigate the reliability of the TCC approach with respect to the exact wave function. In addition to the error in the electronic energy and standard coupled cluster diagnostics, we exploit the overlap of TCC and full configuration interaction wave functions as a quality measure. We critically review issues, such as the required size of the active space, size-consistency, symmetry breaking in the wave function, and the dependence of TCC on the reference wave function. We observe that possible errors caused by symmetry breaking can be mitigated by employing the determinant with the largest weight in the active space as reference for the TCC calculation. We find the TCC model to be promising in calculations with active orbital spaces which include all orbitals with a large single-orbital entropy, even if the active spaces become very large and then may require modern active-space approaches that are not restricted to comparatively small numbers of orbitals. Furthermore, utilizing large active spaces can improve on the TCC wave function approximation and reduce the size-consistency error because the presence of highly excited determinants affects the accuracy of the coefficients of low-excited determinants in the active space.
BibTeX
@article{morchen2020tailored,
  author = {Maximilian Mörchen and Leon Freitag and Markus Reiher},
  title = {{Tailored coupled cluster theory in varying correlation regimes}},
  journal = {The Journal of Chemical Physics},
  year = {2020},
  doi = {10.1063/5.0032661},
}
Figure