Aufbau Suppressed Coupled Cluster Theory for Doubly Excited States

This paper generalizes the Aufbau suppressed coupled cluster formalism to accurately describe doubly excited states by introducing a specialized wave function initialization strategy that achieves high accuracy (errors ~0.15 eV) at a computational cost comparable to ground-state singles and doubles theory, significantly outperforming standard equation-of-motion methods for these challenging electronic states.

The Gist

The Big Problem: The "Double Trouble" of Electrons

Imagine a molecule as a busy dance floor. Usually, when a molecule gets excited (like when it absorbs light), one electron jumps from a low-energy dance spot to a high-energy one. This is a "single excitation." Most computer programs used by chemists are like expert dance instructors; they are great at predicting what happens when one person moves.

However, sometimes two electrons jump at the exact same time. This is a "double excitation." These states are tricky. They are often invisible to standard cameras (experiments) because of how they move, but they are crucial for things like how plants use sunlight or how certain materials glow.

The problem is that the standard computer programs (called "Equation of Motion" or EOM methods) are terrible at predicting these double jumps. It's like trying to predict a complex dance routine where two people move simultaneously, but your instructor only knows how to teach one person at a time. The predictions are often wildly off—sometimes by a huge margin (4 to 6 "steps" or electron-volts).

The New Solution: The "Aufbau Suppressed" Method

The authors of this paper, Qasim Javed, Harrison Tuckman, and Eric Neuscamman, are testing a different approach called Aufbau Suppressed Coupled Cluster (ASCC).

To understand their trick, imagine the "Aufbau" principle as a rule that says, "Fill the lowest energy seats first." In a standard computer model, the calculation starts with everyone sitting in the lowest seats (the ground state). To study a double jump, the computer tries to nudge the system from that ground state. But because the double jump is so far away from the ground state, the computer gets confused and makes big mistakes.

The ASCC Trick:
Instead of starting from the ground state and trying to push the electrons up, ASCC starts by pretending the double jump has already happened.

1. The Setup: They take a "reference" wave function (a snapshot of the molecule) that already has the two electrons in their new, excited spots.
2. The "Suppression": They use a mathematical tool (an exponential operator) to effectively "erase" or suppress the original ground state configuration. It's like telling the computer, "Ignore the starting position; we are starting right here at the destination."
3. The Refinement: Once the computer is sitting at the right starting point (the double-excited state), it adds in the small, messy details of how the electrons interact (correlation).

The Results: A New Champion

The authors tested this method on a variety of molecules, including some where the double jump involves just one specific pair of electrons (Single-CSF) and others where two different pairs are involved (Multi-CSF, like in the molecule glyoxal).

Here is what they found:

• Accuracy: The new method is incredibly accurate. For the tricky double-jump states, their errors were tiny (around 0.15 eV).
• Comparison:
  • Standard Methods (EOM-CCSD): Missed the mark by 4 to 6 eV.
  • High-Level Standard Methods (EOM-CCSDT): Even the most expensive, high-level versions of the old methods still missed by 0.4 to 0.8 eV.
  • The New Method (ASCC): Missed by only 0.15 eV, and even the worst case was only 0.3 eV off.
• Cost: Usually, to get better accuracy, you have to pay a huge price in computer time (like going from a bicycle to a rocket ship). Surprisingly, this new method is just as fast as the standard "bicycle" method (CCSD). It achieves high-level accuracy without the high-level cost.

The "Glyoxal" Test Case

The paper highlights a specific challenge: molecules like glyoxal, where the double excitation isn't just one simple jump, but a mix of two different jumps happening at once.

• Old Methods: Failed miserably here, with errors around 6 eV.
• ASCC: The authors showed that by slightly tweaking their starting point to account for both jumps, the method handled this complex mix perfectly, keeping errors under 0.25 eV.

The Bottom Line

This paper demonstrates that you don't need a super-expensive, slow computer program to understand complex double-electron jumps. By changing the starting point of the calculation to match the excited state directly (suppressing the ground state), the authors created a method that is:

1. Highly Accurate: It predicts double excitations much better than current standard methods.
2. Efficient: It runs at the same speed as standard methods.
3. Versatile: It works for both simple double jumps and more complex, mixed double jumps.

The authors conclude that while there is still work to be done to make this method work for every possible complex scenario, these early results are a strong reason to keep investigating this approach. It offers a promising new way to model the "dark" but important states of molecules that have long been difficult for computers to solve.

Technical Summary

Technical Summary: Aufbau Suppressed Coupled Cluster Theory for Doubly Excited States

Problem Statement
Doubly excited electronic states are critical in spectroscopy and photochemistry (e.g., singlet fission, thermally activated delayed fluorescence) but have historically resisted accurate modeling by standard quantum chemistry methods. Time-dependent density functional theory (TD-DFT) fails to describe them within the standard adiabatic approximation. Furthermore, linear-response (LR) and equation-of-motion (EOM) coupled cluster (CC) methods, particularly at the affordable singles and doubles level (EOM-CCSD), typically yield large errors (4–6 eV) for doubly excited states. While adding higher excitations (triples, quadruples) improves accuracy, the computational cost scales prohibitively (e.g., EOM-CCSDT scales as $N^8$). State-specific approaches like $\Delta$CC offer better accuracy for single-configuration states but struggle with multi-configurational cases where multiple configuration state functions (CSFs) have large weights.

Methodology
The authors generalize the Aufbau Suppressed Coupled Cluster (ASCC) formalism, previously successful for singly excited states, to doubly excited states. The core methodology involves:

1. Wave Function Initialization: Instead of starting from a ground-state Aufbau determinant $|\Phi_0\rangle$ and relying on linear response, ASCC constructs a zeroth-order wave function that explicitly matches the leading CSFs of the target excited state. This is achieved using an exponentiated de-excitation operator $\hat{S}^\dagger$ to suppress the ground-state determinant and promote the desired excited configurations.

   • For 1-CSF states (dominated by a single determinant $|\Phi_{p\bar{p}}^{h\bar{h}}\rangle$), the ansatz is $|\Psi\rangle = e^{-\eta \hat{S}^\dagger} e^{\hat{T}} |\Phi_0\rangle$, where $\hat{T}^{(0)}$ is initialized to reproduce the reference.
   • For 2-CSF states (e.g., glyoxal, involving two hole orbitals and one particle orbital), the ansatz is extended to include a more complex de-excitation operator involving two distinct hole-particle pairs ($\hat{S}_{11}$ and $\hat{S}_{12}$) to match the multi-configurational reference.

2. Perturbative Truncation: The authors perform a perturbative analysis to determine which cluster amplitudes ($\hat{T}$) must be retained to maintain accuracy while preserving computational efficiency. They partition the Hamiltonian and the cluster operator into "primary" (involving active hole/particle orbitals) and "non-primary" (external) components.

   • The analysis reveals that retaining all singles, all doubles, and specific triples (those with at least one primary index) constitutes a first-order treatment.
   • Crucially, this truncation results in an asymptotic cost scaling of $O(N^6)$, identical to ground-state CCSD, despite the inclusion of specific triple excitations required for the primary space.

3. Reference Generation: Calculations utilize state-averaged CASSCF (Complete Active Space Self-Consistent Field) references. For 1-CSF states, a (2o, 2e) active space is used; for 2-CSF states, a (3o, 4e) active space is employed. The ASCC calculation initializes the cluster amplitudes based on the CASSCF wave function (truncated to CSFs with coefficients > 0.25) and then optimizes the amplitudes to solve the similarity-transformed Schrödinger equation.

Key Contributions

• Generalization of ASCC: The paper derives and implements the theoretical framework for applying Aufbau suppression to doubly excited states, extending the method beyond its previous application to singly excited and charge-transfer states.
• Multi-Configurational Handling: The authors demonstrate a proof-of-concept for handling 2-CSF doubly excited states (e.g., glyoxal) by initializing the wave function with a multi-determinant reference, a regime where standard EOM-CCSD and $\Delta$CC often fail.
• Cost-Accuracy Balance: The work establishes that highly accurate treatment of doubly excited states can be achieved at the $O(N^6)$ cost of CCSD, avoiding the steep scaling of EOM-CCSDT or EOM-CC4.

Results
The method was tested on a benchmark set of doubly excited states (QUEST #2 dataset) using various basis sets (6-31+G*, aug-cc-pVDZ, aug-cc-pVTZ).

• 1-CSF States: ASCCSD (singles and doubles) achieved a mean unsigned error (MUE) of 0.15 eV and a maximum error of 0.30 eV. This significantly outperforms EOM-CCSD (MUE ~4.06 eV) and EOM-CCSDT (MUE ~0.38 eV).
• 2-CSF States: For multi-configurational cases like glyoxal, pyrazine, and oxalyl fluoride, ASCCSD maintained high accuracy with an MUE of 0.14 eV and a maximum error of 0.22 eV. In contrast, EOM-CCSD errors were ~6.24 eV, and EOM-CCSDT errors remained above 0.5 eV.
• Robustness: The method showed low sensitivity to the choice of the initial active space (changing from (2o,2e) to (4o,4e) altered energies by <0.07 eV).
• Comparison: ASCCSD accuracy for doubly excited states is comparable to its performance for singly excited states, a feat rarely achieved by other excited-state methods.

Significance and Claims
The authors claim that this work offers a "strong motivation for further investigation" into multi-configurational double excitations. The primary significance lies in demonstrating that a state-specific approach, initialized with a minimal active space reference, can capture the essential physics of doubly excited states (including orbital relaxation and correlation) without the prohibitive cost of high-order EOM methods.

The paper explicitly notes that while the current approach handles 1-CSF and simple 2-CSF (two-hole, one-particle) cases effectively, generalizing to systems with many primary hole and particle orbitals remains an open question. The authors caution that if all excitation levels within a large primary space are required, the scaling could become exponential in the number of primary orbitals. However, for systems with a modest number of active orbitals (common in photochemistry), the method retains the favorable $O(N^6)$ scaling of CCSD, offering a viable path for accurate modeling of "dark" doubly excited states that are currently difficult to treat.