EOM-fpCCSD: An Accurate Alternative to EOM-CCSD for Doubly Excited and Charge-Transfer States

This paper introduces EOM-fpCCSD, a computationally efficient equation-of-motion coupled-cluster method based on a pair coupled-cluster doubles reference that significantly improves the accuracy and convergence of doubly excited and charge-transfer state descriptions compared to standard EOM-CCSD and its pair-tailored variant.

The Gist

Imagine you are trying to predict how a complex machine (like a molecule) will behave when you give it a little jolt of energy. In the world of chemistry, this "jolt" is light, and the machine's reaction is an electronic excitation. Scientists use powerful computer models to predict exactly what happens.

For a long time, the "gold standard" tool for this job has been called EOM-CCSD. Think of it as a very expensive, high-end GPS. It's incredibly accurate for most roads (simple chemical reactions), but it has two major flaws:

1. It gets lost in traffic jams: When electrons act in pairs or move in complex, synchronized ways (called "doubly excited states"), the GPS crashes or gives wildly wrong directions.
2. It's slow: It takes a long time to calculate the route.

The New Solution: EOM-fpCCSD

The authors of this paper, Katharina Boguslawski and Paweł Tecmer, have built a new, smarter GPS called EOM-fpCCSD.

Here is how it works, using some everyday analogies:

1. The "Frozen Pair" Strategy

Imagine a dance floor. In a standard model, every dancer (electron) is free to move anywhere, which makes predicting the dance very hard.

• The Old Way (EOM-CCSD): Tries to track every single dancer's movement individually. It's accurate for solo dancers but gets overwhelmed when pairs start dancing together.
• The New Way (EOM-fpCCSD): This method starts with a special observation: some dancers are always holding hands in pairs. The new model says, "Let's freeze those pairs in their perfect dance formation first." We treat them as a single unit.
• The "Frosting": Once the pairs are frozen, the model adds a layer of "frosting" (correction) to account for the dancers who aren't holding hands or the small wiggles the pairs make. This is the "dynamic correlation."

By freezing the pairs first, the computer doesn't have to do as much heavy lifting, making the calculation faster and more stable.

2. Solving the "Double Trouble"

Some molecules are like a group of people trying to jump in perfect unison. If they don't jump together, the whole group falls.

• The Problem: Standard models often fail here. They might say, "That jump is impossible," or give a result that is off by a huge margin (like predicting a 10-foot jump when it's actually 2 feet).
• The Fix: Because the new method respects the "hand-holding" nature of these electrons from the start, it can handle these tricky "double jumps" (doubly excited states) with much higher accuracy. In the paper's tests, it fixed errors that were several times larger than the actual answer.

3. Tracking the "Charge Flow" (Charge Transfer)

Imagine a molecule as a relay race team with a Runner (Donor), a Bridge, and a Finish Line (Acceptor). When the molecule gets excited, a baton (an electron) might run from the start to the finish.

• The new method is excellent at tracking exactly how much of the baton moves from the start to the finish.
• Even though the new method calculates the speed of the runner slightly differently than the old method, it agrees perfectly on who is running and where they are going. This is crucial for designing better solar cells and LEDs, where moving that "baton" efficiently is the whole point.

The Results: Why Should You Care?

The authors tested their new GPS against a massive database of known chemical reactions (the QUEST database).

• For simple runners (Charge Transfer): The new method is just as good as the expensive gold standard but runs faster.
• For the tricky double-jumpers (Doubly Excited States): This is where the magic happens. The old method often crashed or gave terrible answers. The new method not only solved these cases but gave answers very close to the "theoretical best" (the perfect answer we can't quite reach yet).
• Reliability: The new method even managed to find answers for molecules where the old method simply gave up and said, "I can't do this."

The Bottom Line

The authors have created a tool that is cheaper, faster, and more reliable for the hardest types of chemical problems. It's like upgrading from a bicycle to a high-performance electric bike: it gets you to the same destination, but it handles the steep hills (complex electron interactions) without breaking a sweat.

This is a big deal for developing new materials for solar energy, organic light-emitting diodes (OLEDs), and advanced electronics, where understanding these tricky electron dances is essential.

Technical Summary

1. Problem Statement

The Equation-of-Motion Coupled-Cluster Singles and Doubles (EOM-CCSD) method is the gold standard for calculating low-lying electronic excitation energies due to its balanced treatment of dynamic correlation and favorable $O(N^6)$ scaling. However, it suffers from two critical limitations:

• Doubly Excited States: EOM-CCSD performs poorly for states with significant double-excitation character (errors can exceed several eV) and often fails to converge for these systems.
• Charge-Transfer (CT) States: While generally reliable, standard EOM-CCSD struggles with the quantitative description of directed charge flow in complex organic systems, particularly when using canonical Hartree-Fock (HF) orbitals.

There is a need for a method that retains the computational efficiency of EOM-CCSD while significantly improving accuracy for challenging doubly excited and CT states, without resorting to expensive multi-reference methods.

2. Methodology

The authors introduce EOM-fpCCSD (Frozen-Pair Equation-of-Motion Coupled-Cluster Singles and Doubles), a new method built upon a Pair Coupled Cluster Doubles (pCCD) reference.

• Theoretical Framework:

  • Reference State: The ground state is described by pCCD, which captures strong static correlation (seniority-zero sector) by optimizing electron-pair excitations ($\hat{T}_p$). The reference determinant is variationally optimized (pCCD natural orbitals) rather than using canonical HF orbitals.
  • Frozen-Pair Ansatz: The method employs a tailored coupled-cluster (tCC) approach where the pCCD pair amplitudes are "frozen." The cluster operator is partitioned into $\hat{T}_{int}$ (the frozen pCCD pairs) and $\hat{T}_{ext}$ (the external, broken-pair excitations).
  • EOM Extension: The excited states are modeled using a linear CI-type ansatz ($\hat{R}$) acting on the frozen-pair reference. The similarity-transformed Hamiltonian is constructed to ensure that the projection equations are satisfied for each excitation manifold. Crucially, the Hamiltonian matrix is adjusted so that the first column (associated with pair excitations) vanishes, as the pCCD amplitudes satisfy their own ground-state equations.
  • Variants: The authors compare EOM-fpCCSD against:
    • EOM-CCSD: Standard method using HF or pCCD orbitals.
    • EOM-ptCCSD: A pair-tailored variant where pair amplitudes are optimized but not frozen in the same strict manner.
    • EOM-fpLCCSD: A linearized version of the frozen-pair approach.

• Implementation:

  • Implemented in the open-source PyBEST software package.
  • Utilizes spin-free excitation operators.
  • Includes a domain-based charge-transfer analysis tool (DAISpY) to quantify directed charge flow (dCT) between donor, bridge, and acceptor regions.

3. Key Contributions

1. Novel Algorithm: Derivation and implementation of the EOM-fpCCSD method, which decouples the seniority-zero sector (handled by pCCD) from other seniority sectors (handled by the EOM correction).
2. Robustness for Doubly Excited States: The method successfully converges and provides accurate energies for states where standard EOM-CCSD and EOM-ptCCSD fail or yield large errors.
3. Charge-Transfer Analysis: Introduction of a quantitative framework to analyze directed charge transfer (dCT) using localized pCCD natural orbitals, demonstrating that the character of the excitation is method-independent, even if energies vary.
4. Open-Source Availability: All methods are implemented in PyBEST, making them accessible to the community.

4. Results and Benchmarks

The methods were benchmarked against the QUEST database (containing 29 molecules) and compared against Theoretical Best Estimates (TBEs).

• Charge-Transfer (CT) States:

  • Accuracy: EOM-fpCCSD yields excitation energies very close to standard EOM-CCSD (within ~0.01 eV deviation between HF and pCCD orbital bases).
  • Comparison: EOM-fpCCSD outperforms EOM-ptCCSD for CT states, bringing results closer to TBEs.
  • CT Character: The directed CT character (dCT) and its percentage contribution were found to be nearly identical across EOM-CCSD, EOM-ptCCSD, and EOM-fpCCSD. This confirms that the choice of orbital basis or method does not alter the qualitative nature of the charge flow, though pCCD orbitals reduce basis-set sensitivity.

• Doubly Excited States:

  • Performance: This is the primary area of improvement. While EOM-ptCCSD behaves similarly to standard EOM-CCSD (failing to converge or showing large errors for states like the $\Delta_g$ and $\Sigma^+_g$ states of $C_2$), EOM-fpCCSD significantly outperforms both.
  • Error Reduction: For challenging doubly excited states, EOM-fpCCSD reduces errors relative to TBEs from several eV (typical of EOM-CCSD) to 0.2–0.5 eV.
  • Convergence: EOM-fpCCSD successfully converges for several doubly excited states that cause standard EOM-CCSD and EOM-ptCCSD to fail.
  • Linearized vs. Non-Linear: The full EOM-fpCCSD slightly outperforms its linearized counterpart (EOM-fpLCCSD), though both are vastly superior to standard approaches for these states.

• Computational Cost:

  • The method retains the $O(N^6)$ scaling of standard EOM-CCSD, making it computationally efficient for routine calculations on medium-to-large systems.

5. Significance

The EOM-fpCCSD method represents a significant step forward in computational photochemistry and materials science:

• Reliability: It provides a single-reference framework capable of handling complex electronic structures (strong static correlation + dynamic correlation) that typically require multi-reference methods (like CASPT2 or MRCI), but at a lower computational cost.
• Material Design: By accurately describing doubly excited and charge-transfer states, this method enables more reliable structure-property relationship studies for organic electronic materials, including organic solar cells, singlet fission systems, and OLEDs.
• Convergence Stability: The ability to converge for states that standard EOM-CCSD cannot solve makes it a robust tool for automated high-throughput screening of excited states.

In conclusion, EOM-fpCCSD offers a computationally efficient, accurate, and stable alternative to conventional EOM-CCSD, specifically bridging the gap for challenging doubly excited and charge-transfer excitations.