Excited States from Quasiparticle Hamiltonian Based on Density Functional Theory

This paper extends the occupancy extrapolation method to an effective quasiparticle Hamiltonian, enabling a multi-configurational description of electronic excitations that achieves accuracy comparable to or better than the Bethe-Salpeter equation across various excitation types.

The Gist

The Big Picture: Predicting How Molecules Glow

Imagine you have a molecule, like a tiny, complex machine made of atoms. When you shine light on it, the machine absorbs energy and jumps into an "excited state." It's like a ball sitting at the bottom of a hill (the ground state) suddenly getting kicked up to the top of a different hill.

Scientists want to predict exactly how much energy is needed to kick that ball up, and what color of light the molecule will glow when it falls back down. This is crucial for understanding everything from how solar panels work to how our eyes see color.

The Problem: The Old Tools Have Flaws

To do this, scientists use computer models. The paper discusses three main ways they've tried to solve this puzzle, and why each has a problem:

1. The "Perfect" but Expensive Method (BSE/GW): Think of this as using a super-accurate, high-definition 3D scanner. It gives great results, but it takes a massive amount of computing power and time. It's like trying to map every single grain of sand on a beach; accurate, but you'll never finish.
2. The "Fast" but Flawed Method (TDDFT): This is like using a quick sketch. It's fast and cheap, but the artist (the math) sometimes makes mistakes. For example, it often gets the distance between two people holding hands (charge transfer) wrong, or it misses the faint, fuzzy edges of the picture (Rydberg states).
3. The "One-Person" Method (OE and $\Delta$SCF): This is a newer, faster approach called Occupancy Extrapolation (OE). Imagine you are trying to predict the weight of a backpack by adding one book at a time. You can guess the total weight pretty well. However, this method assumes the backpack is just a stack of books (a single, neat arrangement). In reality, the books might be tangled, or the backpack might have multiple compartments that interact in complex ways. This method struggles when the "books" (electrons) get tangled up in a multi-layered mess.

The New Solution: The "Quasiparticle Hamiltonian"

The authors, Yang and Fan, have built a new tool that combines the speed of the "sketch" with the accuracy of the "3D scanner." They took the Occupancy Extrapolation (OE) method and upgraded it into what they call a Quasiparticle Hamiltonian (QH).

Here is how they did it, using an analogy:

The Analogy: From a Solo Act to a Band

• The Old Way (OE): Imagine a musician playing a solo. You can predict the sound of one note perfectly. But if you try to predict what happens when two musicians play together, the solo method fails because it doesn't account for how they interact.
• The New Way (QH): The authors realized that excited electrons aren't just solo players; they are a band. They created a new "score" (the Hamiltonian) that describes not just one electron jumping, but the whole band playing together.
  • They treat the excited electron and the "hole" it left behind as a pair of dancers.
  • Instead of just guessing the dance steps, they wrote a rulebook that accounts for how the dancers pull and push each other (the interaction between particles).

Why This New Tool is Special

The paper claims this new method hits a "sweet spot" that the others miss:

1. It Handles the "Messy" Dances: Unlike the old OE method, this new tool can handle situations where electrons are tangled up in complex, multi-layered patterns (multi-configurational states). It's like the new tool can predict the sound of a jazz band improvising, whereas the old tool could only predict a marching band playing in perfect lockstep.
2. It Gets the Colors Right: The authors tested their method on different types of "jumps" (excitations):
   • Charge Transfer: When an electron jumps far away (like across a room). The new method is just as good as the expensive 3D scanner here.
   • Rydberg States: When an electron jumps to a very fuzzy, distant orbit. The new method is actually better than the expensive scanner at predicting these.
   • Triplet vs. Singlet: Sometimes electrons spin in the same direction, sometimes opposite. The old expensive method often gets the difference between these two wrong. The new method fixes this error, giving a more accurate prediction of the energy difference.
3. It's Fast: Because it builds on the fast "sketch" method (DFT) rather than the slow "3D scanner" (GW), it runs much faster on computers. It's like getting a high-definition photo without needing a supercomputer to process it.

The Bottom Line

The authors have created a new mathematical engine that allows scientists to predict how molecules absorb and emit light with high accuracy and low cost.

• Before: You had to choose between "Fast but inaccurate" or "Accurate but too slow."
• Now: This new method offers a "Fast and Accurate" option that can handle complex, messy electron interactions that previous fast methods couldn't solve.

The paper concludes that this approach is ready to be used for general optical problems, including understanding how light interacts with bulk materials and complex excitonic states, all without needing the massive computing power of the traditional heavy-hitters.

Technical Summary

1. Problem Statement

The accurate and efficient calculation of electronic excited states is a central challenge in computational chemistry and materials science. Existing methods face significant trade-offs:

• Wavefunction-based methods (EOM-CC, ADC): Highly accurate for small systems but computationally prohibitive ($O(N^6)$ or higher) for large molecules and condensed-phase systems.
• Time-Dependent Density Functional Theory (TDDFT): Offers favorable scaling ($O(N^3-N^4)$) but suffers from deficiencies in Density Functional Approximations (DFAs). Specifically, it fails to describe long-range charge-transfer (CT) excitations (due to lack of exact exchange) and diffuse Rydberg states (due to incorrect asymptotic behavior of the exchange-correlation potential).
• $\Delta$SCF and Occupancy Extrapolation (OE): These methods optimize excited state energies directly or via orbital occupation extrapolation. While computationally efficient, they are fundamentally single-determinant approaches. They fail to capture the multi-configurational nature of many excited states (e.g., open-shell singlets, plasmon-like excitations, and symmetry-protected degenerate states), leading to qualitative errors and spin contamination.
• Bethe-Salpeter Equation (BSE): Based on many-body perturbation theory (GW approximation), BSE handles CT and Rydberg states well. However, it is computationally expensive ($O(N^4-N^6)$ for GW quasiparticle energies) and often overestimates singlet-triplet splittings due to the neglect of vertex corrections.

The Core Gap: There is a need for a method that combines the computational efficiency of DFT-based approaches with the ability to describe multi-configurational excited states and the accuracy of BSE, without the extreme cost of GW calculations.

2. Methodology

The authors propose a Quasiparticle Hamiltonian (QH) approach that extends the Occupancy Extrapolation (OE) method from a single-determinant framework to a multi-determinant one.

A. Theoretical Foundation

• OE Framework: The total energy is expanded as a functional of orbital occupation numbers $\{n\}$. The second-order derivatives of the energy with respect to occupations define the quasiparticle interaction kernel ($\kappa$).
• Extension to Multi-Configurational States: The authors recognize that the standard OE restricts the particle-hole wavefunction to a single determinant ($\Psi_{ph} = \psi_i \psi_a^*$). They extend the domain of definition to a linear combination of particle-hole pairs:
  $$\Psi_{ph} = \sum_{i\sigma, a\tau} X_{i\sigma, a\tau} \psi_{i\sigma}(x_h) \psi^*_{a\tau}(x_p)$$
• Effective Hamiltonian Construction: By treating the excitation energy as a functional of this generalized wavefunction, they derive an effective Hamiltonian $\hat{H}$:
  $$H_{(i\sigma, a\tau), (j\lambda, b\omega)} = (\epsilon^{QP}_{a\tau} - \epsilon^{QP}_{i\sigma})\delta_{ij}\delta_{ab}\delta_{\sigma\lambda}\delta_{\tau\omega} - (\psi_{i\sigma}\psi_{j\lambda}|W|\psi_{b\omega}\psi_{a\tau})$$
  Where:
  • $\epsilon^{QP}$ are quasiparticle energies derived from the GSC2 (Global Scaling Correction with exact second-order corrections) approximation or regularized functionals.
  • $W$ is a generalized screened interaction that includes contributions from the exchange-correlation kernel ($K$) and the linear response function ($\chi$).

B. Key Technical Innovations

1. Spin Purification: To address the broken spin symmetry inherent in collinear DFAs (which splits triplet degeneracy), the authors employ a spin purification scheme. They approximate off-diagonal matrix elements using the difference between broken-symmetry singlet and triplet states, restoring the $SU(2)$ symmetry and correctly describing triplet degeneracy.
2. Regularization for Diffuse States: To prevent numerical divergence in the exchange-correlation integrals for diffuse virtual orbitals (a common issue in Rydberg states), they remove the pure functional part of the kernel ($K'$) and rely on the screened interaction. This ensures numerical stability without sacrificing accuracy.
3. Comparison to BSE: The resulting Hamiltonian resembles the BSE equation in the particle-hole channel but differs in two critical ways:
   • Quasiparticle Energies: Uses static, generalized screened self-interaction corrections (scaling as $O(N^3)$) rather than dynamic GW quasiparticles ($O(N^4-N^6)$).
   • Interaction Kernel: Incorporates the exchange-correlation kernel ($K$) and response function ($\chi$) directly, effectively capturing vertex corrections often neglected in standard BSE.

3. Key Contributions

• Formulation of a Multi-Determinant OE Hamiltonian: Successfully transformed the single-determinant OE method into a matrix diagonalization problem capable of capturing multi-configurational excited states.
• Cost-Effective Accuracy: Developed a method that achieves BSE-level accuracy for valence, CT, and Rydberg states but at a computational cost comparable to ground-state DFT (significantly cheaper than GW-BSE).
• Improved Singlet-Triplet Splitting: Demonstrated that the QH approach corrects the systematic overestimation of singlet-triplet splitting found in BSE and TDDFT by explicitly treating spin symmetry.
• Handling of Diffuse and Multi-Configurational States: Solved the numerical instability of OE for Rydberg states and extended applicability to plasmon-like and symmetry-protected degenerate states (e.g., in benzene and polyenes) which fail in standard $\Delta$SCF/OE.

4. Results

The method (denoted as ph-QH@DFA or ph-QH@$\epsilon^{QP}$) was benchmarked against high-level reference data (CCSD(T), CASPT2) and BSE results across various excitation types:

• Valence Singlets: With the PBE functional, the method slightly underestimates energies (MSE $\approx -0.44$ eV) due to DFA delocalization errors. However, when combined with Localized Orbital Scaling Correction (lrLOSC), the Mean Absolute Error (MAE) drops to 0.28 eV, comparable to BSE.
• Valence Triplets: The method significantly outperforms BSE. While BSE overestimates triplet energies (leading to large singlet-triplet splitting errors), ph-QH yields a MAE of 0.17 eV (with B3LYP/lrLOSC), providing a much more accurate description of the splitting.
• Charge-Transfer (CT) Excitations: The method performs comparably to BSE@GW, correctly capturing the asymptotic behavior of CT states.
• Rydberg States: The method outperforms BSE, with significantly lower errors for diffuse states, attributed to the regularization scheme and the specific treatment of the interaction kernel.
• Multi-Configurational Cases:
  • Benzene: Correctly predicted the degeneracy of the $\pi \to \pi^*$ transition (decomposing into $B_{1u}, B_{2u}, E_{1u}$), which standard OE failed to do.
  • Polyenes (Butadiene, Hexatriene): Successfully described plasmon-like excitations (e.g., $1^3A_g$ in butadiene) where standard OE overestimated energies by >1.4 eV. The QH approach reduced these errors to within 0.3 eV.
• Oscillator Strengths: The calculated oscillator strengths showed qualitative agreement with BSE and high-level references (CASPT2/CC3).

5. Significance

This work represents a significant advancement in the theory of electronic excitations:

1. Bridging the Gap: It successfully bridges the gap between the efficiency of DFT-based methods and the rigor of many-body perturbation theory (BSE/GW).
2. Beyond Single-Determinant: It overcomes the fundamental limitation of $\Delta$SCF and standard OE by providing a rigorous framework for multi-configurational excited states without the cost of full configuration interaction (CI).
3. Practical Applicability: By avoiding the expensive GW step and using regularized DFT derivatives, the method is scalable to large molecules and potentially bulk materials, making it a viable tool for predicting optical properties in complex systems (e.g., excitonic states in solids).
4. Theoretical Insight: The derivation highlights the importance of vertex corrections (via the exchange-correlation kernel) and spin symmetry restoration in achieving accurate singlet-triplet splittings, offering a new perspective on how to improve TDDFT-like approaches.

In summary, the Quasiparticle Hamiltonian based on OE offers a robust, accurate, and computationally efficient alternative to BSE for a wide range of excited state problems, particularly where multi-configurational character and spin symmetry are critical.