Multi-reference GW approximation for strongly correlated molecules

This paper introduces the multi-reference GW (MR-GW) approximation, a rigorous diagrammatic framework that extends ab initio Green's function methods to strongly correlated molecules by incorporating non-perturbative correlation effects into a multi-determinantal reference, thereby accurately capturing both strong and weak correlations and recovering complex many-body satellites missed by standard GW.

The Gist

The Big Picture: Predicting the Future of Electrons

Imagine you are trying to predict the weather. For a calm, sunny day, a simple forecast works great. You look at the temperature and humidity, and you say, "It will be 75°F." This is like how standard computer programs (called GW approximation) predict the behavior of electrons in most materials. They work incredibly well for simple, stable systems like silicon chips or water.

But what happens when a hurricane hits? Or when the weather is chaotic, with winds blowing in five different directions at once? A simple forecast fails completely. You can't just say "75°F"; you need to account for the chaos, the swirling storms, and the complex interactions.

In the world of atoms, this "chaos" is called strong correlation. It happens in molecules where electrons are so busy interacting with each other that they can't be described as just one simple, calm state. They exist in a "superposition" of many different states at once. Standard methods break down here, giving wrong answers about how much energy it takes to knock an electron out of an atom (ionization) or how the molecule absorbs light.

The Problem: The "Single-Story" House

The authors of this paper point out that the standard method (GW) is like trying to describe a complex, multi-story mansion using a blueprint for a single-story shed.

• Standard GW (The Shed): It assumes the electrons live in a single, neat configuration. It's efficient and fast, but if the electrons are actually fighting, dancing, or swapping places wildly (strong correlation), the shed blueprint is useless.
• The Reality (The Mansion): In "strongly correlated" molecules (like stretched chemical bonds or certain transition metals), the electrons are in a messy, multi-configurational state. They need a blueprint that accounts for all the different rooms and floors simultaneously.

The Solution: The "Multi-Reference" Upgrade

The team, led by Zhendong Li, invented a new method called MR-GW (Multi-Reference GW).

Think of it like upgrading from a single-lane road to a multi-lane highway system.

1. The Old Way (Single Reference): You pick one "best guess" for where the electrons are (like picking one lane to drive in) and try to calculate the rest based on that. If the traffic is heavy, you get stuck.
2. The New Way (MR-GW): Instead of picking one lane, they build a system that acknowledges multiple lanes exist at the same time. They start with a "Zeroth-Order" reference that already knows the electrons are messy and complex. They treat the "strong" chaos as the foundation, and then use math to add the "weak" interactions on top.

How It Works: The "Diagrams" and "Screens"

The paper is full of complex math, but the core idea is visual.

• The Diagrams: In physics, we draw pictures (Feynman diagrams) to track how particles interact. The standard method uses a specific set of diagrams. The authors realized that for messy molecules, you can't just use the old diagrams because the rules of the game change. They developed a rigorous new set of diagrams that work even when the starting point is messy.
• The Screen (W): Imagine electrons are people in a crowded room shouting at each other. In a simple room, you can hear everyone clearly. In a crowded, chaotic room, the noise gets "screened" or muffled. The new method calculates this "screening" effect much more accurately by looking at the specific "active" group of electrons causing the trouble, rather than treating the whole room as a blur.

The Results: Fixing the Broken Predictions

The team tested their new method on three difficult scenarios:

1. The Beryllium Atom: This atom is tricky because its electrons are split between two different energy levels. The old method predicted the energy to remove an electron was wrong by a huge margin. The new MR-GW method nailed it, predicting the correct energy and even finding a "satellite" (a secondary, faint signal) that the old method missed entirely.

   • Analogy: The old method said, "The car is going 50 mph." The new method said, "The car is going 50 mph, but it's also drifting, and there's a second car right behind it."

2. Stretched Hydrogen ($H_2$): Imagine pulling two magnets apart. As they stretch, the electrons get confused. The old method completely failed to predict what happens when the bond is stretched. The new method handled the stretching perfectly, capturing the complex dance of electrons that occurs right before the bond breaks.

3. Ozone ($O_3$): Ozone is a "biradical," meaning it has two unpaired electrons that are very difficult to track. Predicting its energy levels has been a nightmare for scientists for decades. The new method got the order of energy levels correct, whereas the old method got the ranking wrong.

Why This Matters

This paper is a "paradigm shift." It provides a rigorous mathematical framework that allows scientists to use powerful Green's function methods (which are usually reserved for simple materials) on complex, messy, strongly correlated systems.

• Before: We had to choose between "Simple but wrong for complex stuff" or "Complex but too hard to calculate."
• Now: We have a tool that is both rigorous and capable of handling the complexity of real-world chemistry, like transition metals (used in batteries and catalysts) or defects in solid-state materials (like the NV center in diamonds used for quantum computing).

The Bottom Line

The authors built a new "lens" for looking at electrons. The old lens was great for clear, calm days, but it distorted the image when things got stormy. The new MR-GW lens is designed specifically for the storm, allowing us to see the true, complex behavior of electrons in the most challenging molecules, paving the way for better materials, drugs, and quantum technologies.

Technical Summary

1. Problem Statement

The GW approximation is a cornerstone of many-body perturbation theory (MBPT) for calculating single-particle excitations (e.g., ionization potentials and electron affinities) in materials and molecules. However, it relies fundamentally on a single-reference picture, where the zeroth-order Hamiltonian ($\hat{H}_0$) is quadratic (non-interacting) and the reference state is a single Slater determinant.

This framework breaks down in strongly correlated systems (e.g., multi-configurational molecules, transition-metal complexes, stretched bonds, and radicals). In these systems:

• The ground-state wavefunction requires a superposition of multiple determinants (multi-configurational character).
• Standard perturbative expansions (like $G_0W_0$) fail because the single-reference starting point is qualitatively incorrect.
• Existing attempts to combine GW with strong correlation often rely on quantum embedding frameworks, which can be complex and require double-counting corrections.
• Crucially, the standard Dyson equation and Hedin's equations rely on Wick's theorem, which fails when $\hat{H}_0$ is interacting (non-quadratic) or the reference is multi-determinantal.

2. Methodology: Multi-Reference GW (MR-GW)

The authors propose MR-GW, a rigorous diagrammatic generalization of the GW approximation that incorporates strong correlation non-perturbatively at the zeroth order.

A. Theoretical Framework

• Interacting Reference: Instead of a non-interacting $\hat{H}_0$, the method employs the Dyall Hamiltonian ($\hat{H}^{\text{Dyall}}_0$). This Hamiltonian includes the full Coulomb interaction within a small subset of "active" orbitals while treating the rest as mean fields.
• Zeroth-Order Green's Function ($G_0$):
  • The system is partitioned into inactive (doubly occupied/virtual) and active orbitals.
  • The inactive block $G^I_0$ is a standard non-interacting Green's function.
  • The active block $G^A_0$ is an interacting Green's function derived from the exact diagonalization of the active-space Hamiltonian (e.g., CASCI/CASSCF).
• Generalized Dyson Equation: Since Wick's theorem does not hold for interacting $\hat{H}_0$, the standard Dyson equation is invalid. The authors utilize a generalized Dyson equation derived by Hall, which relates the full Green's function $G$ to $G_0$ via a $2 \times 2$ block matrix of self-energies ($\Sigma_{11}, \Sigma_{12}, \Sigma_{21}, \Sigma_{22}$).
  • $\Sigma_{11}$ corresponds to the standard self-energy.
  • $\Sigma_{12}, \Sigma_{21}, \Sigma_{22}$ arise from the coupling between one-body and many-body Green's functions (cumulants) in the active space.

B. The MR-GW Approximation

To make the theory practical, the authors introduce specific approximations:

1. Screened Interaction ($W$): The screened interaction is constructed using the Multi-Reference Random Phase Approximation (MR-RPA). This replaces the standard non-interacting polarizability ($\Pi_0$) with a multi-reference version that accounts for four types of screening processes involving active-space excitations.
2. Self-Energy ($\Sigma$):
   • The authors define the MR-GW self-energy ($\Sigma_{\text{MR-GW}}$) primarily through the $\Sigma_{11}$ component, retaining the standard GW diagram ($iG_0W$) but using the interacting $G_0$ and the residual two-electron interaction.
   • Neglect of Off-Diagonal Blocks: The off-diagonal self-energies ($\Sigma_{12}, \Sigma_{21}, \Sigma_{22}$) are neglected. The authors demonstrate that including only first-order terms for these blocks can lead to analytical issues (loss of positive definiteness in the spectral function). Neglecting them allows the generalized Dyson equation to reduce to a form resembling the standard Dyson equation:
     $$G = G_0 + G_0 \Sigma_{\text{MR-GW}} G$$
   • This structure ensures the spectral function retains the correct analytical properties (first-order poles, positive definiteness).

3. Key Contributions

• Rigorous Diagrammatic Formalism: The paper establishes a new theoretical platform for Green's function methods in the strongly correlated regime by developing a diagrammatic expansion based on an interacting reference, circumventing the need for Hedin's equations.
• Generalization of GW: It successfully generalizes the GW approximation to multi-reference systems, recovering standard GW in the limit where the active space is absent or the active interaction vanishes.
• No Double Counting: Unlike quantum embedding approaches, MR-GW naturally avoids double-counting errors because the residual interaction is defined strictly as the difference between the full Hamiltonian and the Dyall Hamiltonian.
• Computational Implementation: The method is implemented within the PySCF package, supporting CASCI and CASSCF references.

4. Results

The method was tested on three prototypical strongly correlated systems:

• Beryllium (Be) Atom:

  • Challenge: The ground state has significant multi-configurational character ($2s^2$ and $2p^2$). Standard $G_0W_0$@RHF fails to predict the correct ionization potential (IP) and misses a satellite peak.
  • MR-GW Performance: Using a CAS(2,4) active space, MR-GW yields an IP of 9.27 eV (vs. FCI 9.18 eV), significantly improving upon $G_0W_0$@RHF (8.73 eV). Crucially, it correctly predicts the satellite peak at ~13 eV, which arises from ionization of the $2p$ orbitals—a process invisible to single-reference GW.

• Stretched Hydrogen ($H_2$):

  • Challenge: As the bond stretches, the single-determinant RHF reference breaks down, and the $2s^2$ and $2s^2$ (inverted) configurations compete.
  • MR-GW Performance: While $G_0W_0$@RHF errors increase with bond length and miss emerging satellites, MR-GW (CAS(2,2)) dramatically recovers the correct spectral functions and IP/Electron Affinity (EA) values, capturing the physics of the two-determinantal reference.

• Ozone ($O_3$):

  • Challenge: A biradical system where the ordering of the first three ionized states ($2A_1, 2B_2, 2A_2$) is notoriously difficult for single-reference methods.
  • MR-GW Performance: Standard GW and ADC(3) predict the wrong ordering. MR-GW with a minimal CAS(6,4) active space correctly predicts the ordering and yields ionization energies much closer to experimental values than ADC(2) or GW. Enlarging the active space further improves accuracy.

5. Significance

• Bridging the Gap: MR-GW provides a unified framework that seamlessly treats both weak and strong correlations, filling a critical gap between standard perturbative Green's function methods and multi-reference wavefunction theories.
• Spectroscopic Accuracy: It recovers complex many-body satellites and corrects qualitative errors in spectral functions that are missed by standard GW, making it a powerful tool for interpreting photoelectron spectroscopy in complex molecules.
• Scalability: Because the final equations share a mathematical structure similar to standard GW, existing efficient numerical techniques (e.g., Resolution-of-Identity, contour deformation) can be adapted to extend MR-GW to larger systems.
• Future Applications: The authors highlight potential applications in solid-state defects (e.g., NV centers in diamond) and systems with $d$ and $f$ electrons, where multi-reference treatments are essential.

In summary, this work establishes a rigorous, diagrammatic path to extend ab initio Green's function methods into the strongly correlated regime, offering a robust alternative to embedding schemes for calculating accurate spectroscopic properties of complex molecular systems.