A Flexible, Automated, and Basis-Set Insensitive Domain-Based Charge-Transfer Decomposition for Correlated Wavefunctions and its Application to Inter- and Intramolecular Cases

This paper presents a flexible, automated, and basis-set insensitive framework for decomposing charge-transfer excitations in correlated wavefunctions into local and domain-based contributions, offering robust analysis for both inter- and intramolecular cases across various computational setups.

The Gist

Imagine you are watching a complex dance performance. The dancers are electrons, and the stage is a molecule. Sometimes, a dancer stays right where they are, just spinning in place (a "local" move). Other times, a dancer leaps from one side of the stage to the other, carrying energy and charge with them (a "charge transfer").

Scientists have long wanted to measure exactly how much of this "leaping" happens during a chemical reaction or when a molecule absorbs light. However, the tools they used to watch this dance were often like trying to count the steps of a dancer while wearing foggy glasses. The results depended heavily on the specific "lens" (mathematical basis set) they used, and the process was often manual, slow, and required a human to guess where the dancers were going.

The New "Smart Camera" System
In this paper, the authors introduce a new, automated system called DAISpY (Domain Assignment and Interface Solution in pYthon). Think of this as a high-tech, smart camera that doesn't just watch the dance; it automatically divides the stage into specific zones (like the "Donor Zone," the "Bridge Zone," and the "Acceptor Zone") and counts exactly how many electrons jump from one zone to another.

Here is how it works, broken down into simple concepts:

1. The "Zone" Strategy

Instead of looking at the whole molecule as a blur, the system slices it into logical chunks (domains).

• The Hard Cut: Imagine drawing a sharp line down the middle of the stage. If a dancer is on the left, they belong to the Left Zone. If they are on the right, they belong to the Right Zone. This is the "strict" method.
• The Weighted Blend: Sometimes, a dancer is standing right on the line, with one foot in each zone. The "weighted" method says, "Okay, 60% of that dancer's energy is going to the Left, and 40% to the Right." This is more flexible and works better for small, crowded stages.

2. The "Foggy Glasses" Problem Solved

Previous methods were very sensitive to the "lens" used. If you zoomed in or out (changed the mathematical basis set), your count of how many electrons jumped would change wildly.

• The Paper's Claim: The authors tested their new system with different "lenses" (different sizes of mathematical grids). They found that their new method is insensitive to the lens. Whether they used a small grid or a large, detailed one, the story of the electron dance remained the same. The system gives a consistent answer regardless of the mathematical tools used to calculate it.

3. Two Ways to Watch the Dance

The team tested their system using two different "cameras" (computational methods):

• The High-Definition Camera (EOM-CCSD): This is the gold standard, very accurate but computationally expensive (like filming in 8K resolution).
• The Budget Camera (EOM-pCCD+S): This is a faster, cheaper method. It's not as precise in the numbers, but it captures the story perfectly.
• The Result: Even though the "Budget Camera" gave slightly different numbers, it told the exact same story as the "High-Definition Camera." If the High-Def camera saw a big leap from Donor to Acceptor, the Budget camera saw the same big leap. This means scientists can use the cheaper, faster method to get reliable qualitative results for large, complex molecules without waiting days for a calculation to finish.

4. What They Found

The authors tested this system on two types of scenarios:

• Intermolecular (Two separate molecules dancing together): Like two people passing a ball across a gap. The system successfully measured how much charge moved between them.
• Intramolecular (One molecule with different parts): Like a person passing a ball from their left hand to their right hand. The system successfully identified which parts of the molecule were the "donor" and which were the "acceptor," even without the researchers telling it beforehand.

The Bottom Line

This paper presents a robust, automated tool that acts like a universal translator for electron movements. It takes complex, messy quantum data and translates it into a clear, simple map of where electrons are going.

• It doesn't need a human to manually draw lines or guess the zones; it does it automatically.
• It doesn't get confused by the mathematical "lens" used to calculate the data.
• It works well even with faster, cheaper computing methods, making it possible to analyze huge, complex systems that were previously too difficult to study in this detail.

In short, they built a better ruler for measuring how electricity moves through molecules, and this ruler gives the same measurement no matter how you hold it.

Technical Summary

Technical Summary: A Flexible, Automated, and Basis-Set Insensitive Domain-Based Charge-Transfer Decomposition

Problem Statement
Understanding the spatial character of electronic excitations—specifically distinguishing between local excitations and charge-transfer (CT) processes between molecular fragments—is a fundamental challenge in excited-state chemistry. While linear-response methods (e.g., TD-DFT, LR-CC) naturally provide transition density matrices (TDMs) to analyze these processes, Equation of Motion (EOM) methods describe excited states as explicit many-electron wavefunctions via Configuration Interaction (CI) expansions. In the EOM framework, constructing TDMs requires the computationally expensive evaluation of both left and right eigenstates. Furthermore, traditional analysis methods often rely on qualitative orbital inspection or density difference plots, which can be representation-dependent and lack rigorous spatial partitioning. Existing domain-based approaches have previously been limited to manual evaluation, restricted to single excitations, or dependent on specific wavefunction formulations. There is a need for an automated, flexible framework that can dissect the CT character of any correlated CI-type wavefunction without relying on excited-state densities or left eigenvectors.

Methodology
The authors present a domain-based CT decomposition framework implemented in the DAISpY (Domain Assignment and Interface Solution in pYthon) package, integrated within the PyBEST quantum chemistry suite. The methodology operates directly on the CI expansion coefficients of the excited-state wavefunction $|\Psi_k\rangle = \sum c_\mu \hat{\tau}_\mu |\Psi_0\rangle$.

Key methodological components include:

1. Domain Definition: The molecule is partitioned into chemically relevant regions (domains) based on atom-centered basis sets.
2. Orbital Localization: To ensure unambiguous domain assignment and avoid smeared weighting factors from delocalized canonical orbitals, the framework utilizes pCCD-optimized (pair Coupled Cluster Doubles) molecular orbitals. These orbitals are strongly localized in both occupied and virtual subspaces, facilitating clear hole-particle domain mapping.
3. Domain-Domain Excitation Matrix: The framework constructs a matrix $M_{D_h D_p}$ that accumulates CI contributions based on the domain character of the participating orbitals. Two accumulation strategies are introduced:
   • Hard (Strict) Scheme: Assigns an orbital to a single domain ($\Omega = 0$ or $1$) based on the dominant atomic orbital contribution.
   • Weighted Scheme: Assigns fractional contributions ($0 \le \Omega \le 1$) based on the squared expansion coefficients, suitable for small molecules or systems with complex inter-domain connections.
4. Directed CT (dCT) Metric: A scalar measure is derived to quantify the net charge flow between specific domains (e.g., Donor $\to$ Bridge $\to$ Acceptor). This is calculated by summing forward transitions and subtracting reverse transitions, excluding diagonal (local) elements.
5. Applicability: The approach is designed for any CI-type excited-state method (e.g., EOM-CCSD, EOM-pCCD+S) and relies solely on right eigenvectors, avoiding the need for left eigenvectors or excited-state densities.

Key Contributions

• Generalization and Automation: The work generalizes a previous manual, single-excitation-limited approach into a fully automated framework applicable to arbitrary correlated CI wavefunctions, including double excitations.
• Basis-Set Insensitivity: The authors demonstrate that the CT character descriptors are largely independent of the basis set size, unlike excitation energies which show significant variation.
• Dual Accumulation Schemes: The introduction of both strict and weighted domain accumulation strategies allows the method to adapt to different molecular sizes and topologies, with the weighted scheme offering better additivity for complex pathways.
• Efficient Alternatives: The framework validates the use of EOM-pCCD+S (a simplified, cheaper variant of EOM-CCSD) for qualitative CT analysis, despite quantitative discrepancies in excitation energies.

Results
The framework was benchmarked on two test sets:

1. Intermolecular Systems: Ten two-component molecular systems (e.g., acetone–fluorine, pyrrole–pyrazine) were analyzed using EOM-CCSD with both canonical Hartree-Fock (HF) and pCCD-optimized orbitals.

   • Orbital Independence: CT character values showed excellent agreement between calculations using pCCD-optimized orbitals and canonical HF orbitals, confirming the method's robustness regarding the reference determinant.
   • Basis Set Effects: CT values were found to be highly insensitive to basis set size (comparing cc-pVDZ to cc-pVTZ), whereas excitation energies varied.
   • EOM-pCCD+S Performance: While EOM-pCCD+S yielded systematically lower CT values and higher excitation energies (by 2–6 eV) compared to EOM-CCSD, it preserved the qualitative trends and directionality of charge transfer.

2. Intramolecular Systems: Sixteen molecules (e.g., DMABN, twisted phenyl–pyrrole, nitroaniline) were evaluated to distinguish between weak, moderate, strong, and pure CT states.

   • Qualitative Accuracy: The method successfully categorized states ranging from weak CT (aniline) to nearly pure CT (twisted DMABN).
   • Scheme Comparison: The weighted scheme provided more consistent and additive results for systems with multiple charge-transfer pathways (e.g., phthalazine) compared to the hard scheme, which could yield fragmented descriptions.
   • Robustness: Both hard and weighted dCT metrics showed minimal deviation across different basis sets (cc-pVDZ, aug-cc-pVDZ, aug-cc-pVTZ), with standard deviations around 3% for larger basis sets.

Significance and Claims
The paper claims to provide a robust, flexible, and automated tool for resolving the excited-state character into local and domain-based CT contributions. Its primary significance lies in:

• Methodological Independence: It offers an alternative to density-based partitioning schemes that is applicable to any CI-type wavefunction without requiring explicit excited-state densities or left eigenvectors.
• Computational Efficiency: By relying on right eigenvectors and enabling the use of cheaper methods like EOM-pCCD+S for qualitative analysis, it facilitates the study of large-scale systems where full EOM-CCSD or density-based analyses might be prohibitive.
• Reliability: The demonstrated basis-set insensitivity of the CT descriptors allows researchers to use smaller, less expensive basis sets for qualitative CT analysis without compromising the interpretation of charge redistribution.

The authors explicitly state that while EOM-pCCD+S is an approximation not intended to replace standard methods for predicting accurate excitation energies, it offers a computationally economical perspective on CT evolution upon structural modification. The framework is presented as a step toward a general, automated analysis of charge transfer that can be extended to higher-order excitations and other excited-state models in the future.