Locally Scaled Self-Interaction Corrected Energy Functionals with Complex Optimal Orbitals

This paper presents a fully variational locally scaled self-interaction corrected energy functional that utilizes complex optimal orbitals and a kinetic energy density-based scaling factor to dynamically adjust the correction across different electron density regimes, thereby improving predictions for atomic, molecular, and solid-state systems.

The Gist

The Big Picture: Fixing a Flawed Map

Imagine you are trying to navigate a city using a map. For a long time, scientists have used a specific type of map (called KS-DFT) to predict how atoms and molecules behave. This map is incredibly useful and fast, but it has a famous flaw: it suffers from "self-interaction error."

The Analogy:
Think of an electron as a person walking through a crowded room. In reality, a person doesn't bump into themselves. However, this old map mistakenly calculates that the person is bumping into themselves, creating a fake "ghost" of their own weight and presence. This ghost messes up the calculation of how strong the bonds are between people (atoms) or how much energy it takes to move them.

The Previous Attempts: The "One-Size-Fits-All" Fix

Scientists realized this "ghost" problem needed fixing. They invented a correction called Self-Interaction Correction (SIC).

• The Full Fix (SIC): Imagine telling the map, "Delete the ghost entirely." This works perfectly if there is only one person in the room (a single electron). The map becomes perfect.
• The Half-Fix (1/2 SIC): But when there are many people in the room (many electrons overlapping), deleting the ghost entirely makes the map swing too far the other way. It over-corrects. So, scientists tried deleting only half the ghost. This worked well for some things (like how tightly molecules stick together) but failed for others (like how atoms behave when excited or far apart).

The problem was that scientists had to choose: either use the Full Fix (good for single electrons, bad for crowds) or the Half-Fix (good for crowds, bad for single electrons). They couldn't have both.

The New Solution: A "Smart Dimmer Switch"

This paper introduces a new method called Locally Scaled Self-Interaction Correction (LSSIC).

The Analogy:
Instead of a global switch that turns the ghost correction "On" or "Off" (or "Half-On") for the whole room, the authors built a smart dimmer switch that adjusts automatically based on where you are in the room.

• In isolated areas (Low Density): If an electron is all alone (like a single electron in a hydrogen ion), the dimmer turns the correction 100% on. The ghost is fully removed, giving a perfect result.
• In crowded areas (High Density): If electrons are huddled together and overlapping, the dimmer turns the correction down or even off. This prevents the map from over-correcting and making things look weird.

This "dimmer" is controlled by a mathematical function (called $z(r)$) that looks at the "traffic density" of the electrons. It knows exactly when to apply the full fix and when to hold back.

The Secret Ingredient: "Complex" Orbits

The paper also mentions using "Complex Optimal Orbitals."
The Analogy:
Imagine the electrons aren't just walking in a straight line; they are spinning and moving in a 3D spiral. Previous maps tried to flatten this 3D spiral into a 2D line to make the math easier, which lost some detail. The new method embraces the full 3D spiral (the "complex" nature). This allows the "smart dimmer" to see the traffic patterns much more clearly and adjust the correction with higher precision.

What Did They Test?

The authors tested this new "smart map" on several scenarios:

1. The Single Electron (Hydrogen Ion):
   • Result: The new method worked perfectly. It correctly predicted how the single electron behaves, just like the old "Full Fix" did, but without the side effects.
2. Individual Atoms (Carbon, Nitrogen, Oxygen):
   • Result: The new method was excellent at predicting how much energy it takes to grab an extra electron (Electron Affinity). It was slightly less revolutionary for predicting how hard it is to remove an electron (Ionization Energy), but still very accurate.
3. Molecules (Pairs of Atoms):
   • Result: When two atoms bond (like two carbons or two nitrogens), the new method predicted the strength of the bond and the distance between them very accurately. It often did better than the "Half-Fix" and avoided the errors of the "Full Fix."

The Bottom Line

This paper presents a major upgrade to the tools scientists use to simulate chemistry and materials. By creating a local scaling function (the smart dimmer) that works with complex orbitals (the 3D spirals), they have built a method that:

• Fixes the "ghost" error perfectly when an electron is alone.
• Doesn't over-correct when electrons are crowded together.
• Works for single atoms, molecules, and solid materials.

It's like upgrading from a map that forces you to choose between two bad routes, to a GPS that automatically finds the perfect route for every specific traffic condition you encounter.

Technical Summary

Technical Summary: Locally Scaled Self-Interaction Corrected Energy Functionals with Complex Optimal Orbitals

Problem Statement
Kohn–Sham density functional theory (KS-DFT) is the standard for electronic structure calculations but suffers from the self-interaction error (SIE), arising from the partial cancellation of self-Coulomb and self-exchange-correlation interactions. This error leads to significant inaccuracies in describing localized electronic states, Rydberg excitations, transition metal magnetism, and bond energies. While the Perdew-Zunger self-interaction correction (PZ-SIC) removes SIE orbital-by-orbital, a fully variational implementation often overcorrects energetics in high-density regions where orbitals overlap. Conversely, empirical scaling of the correction by a factor of $1/2$ ($1/2$SIC) improves results for many properties but fails to recover the correct $-1/r$ potential dependence required for accurate Rydberg states and localized charge distributions. The challenge lies in developing a framework that applies full SIC where necessary (e.g., single-electron limits or isolated orbitals) while reducing the correction in delocalized regions to avoid overcorrection, all within a fully variational formalism that accommodates complex optimal orbitals.

Methodology
The authors implement a fully variational locally scaled self-interaction corrected (LSSIC) energy functional within the open-source GPAW code. The core innovation is the introduction of a spatially dependent local scaling function, $z_\sigma(\mathbf{r})$, which replaces the global scaling factor ($a$) in the PZ-SIC formalism.

• Iso-orbital Indicator: The scaling function is derived from an iso-orbital indicator based on kinetic energy density. It is constructed to range between 0 and 1, scaling the SIC correction from 0 in delocalized regions (homogeneous electron gas limit) to 1 in single-orbital or isolated orbital regions.
• Complex Optimal Orbitals: A key theoretical advancement in this work is the re-derivation of the von Weizsäcker and Kohn-Sham kinetic energy densities to account for complex-valued optimal orbitals. The local scaling function $z_\sigma(\mathbf{r})$ includes a term dependent on the gradient of the complex phase, expressed via the current density $\mathbf{J}_\sigma(\mathbf{r})$.
• Variational Implementation: The energy is minimized with respect to all electronic degrees of freedom ($dE/d\psi^*_k = 0$) using a direct optimization method. The implementation utilizes the unitary variant of the projector augmented wave (PAW) method and treats wavefunctions as complex numbers.
• Approximations: The authors apply the LSSIC functional to the pseudo-density only, utilizing frozen core approximations for core electrons.

Key Contributions

1. Theoretical Formulation: The paper presents the first fully variational realization of a locally scaled SIC functional based on complex optimal orbitals. This includes the derivation of the generalized iso-orbital indicator (Eq. 5 and 6) that incorporates the complex phase gradient.
2. Numerical Implementation: The method is implemented in the GPAW code, allowing for self-consistent optimization of complex orbitals under the LSSIC framework.
3. Bridging Ground and Excited States: The framework aims to bridge the gap between methods requiring full SIC (for excited states and localized charges) and those requiring reduced SIC (for ground-state energetics in multi-electron systems).

Results
The authors evaluated the LSSIC method against standard PBE, $1/2$SIC, and full SIC (PZ-SIC) calculations for several systems:

• Hydrogen Dimer Cation ($H_2^+$): As a single-electron system, the LSSIC correctly recovers the exact SIC limit ($z \approx 1$). It yields the correct dissociation limit and bond energy, matching the full SIC results, whereas PBE fails and $1/2$SIC underestimates the binding energy.
• Atomic Electron Affinity (EA) and Ionization Energy (IE): For C, N, and O atoms, variational LSSIC (including the complex phase term) provides the best agreement with experimental electron affinities, significantly outperforming PBE and $1/2$SIC. For ionization energies, LSSIC generally matches or slightly improves upon PBE, though it does not consistently outperform full SIC for all elements. The inclusion of the complex phase term marginally improved IE predictions for N and O.
• Dimer Bond Energies and Lengths: For $C_2$, $N_2$, and $O_2$, LSSIC functionals predict bond energies and lengths within a few percent of experimental values.
  • For $C_2$ and $O_2$, LSSIC (with and without the phase term) yields bond energies closer to experiment than PBE, $1/2$SIC, or full SIC.
  • For $N_2$, LSSIC slightly overestimates the bond energy compared to $1/2$SIC, though the potential energy surface shape remains similar.
  • "One-shot" LSSIC (where the scaling function is fixed from a PBE calculation) performed surprisingly well for $C_2$ and $O_2$ bond energies but generally underperformed compared to the fully variational approach.

Significance and Claims
The paper claims that the LSSIC framework represents an important milestone for fully variational SIC functionals. By introducing a local scaling function that naturally adapts to the electron density and orbital overlap, the method retains the benefits of full SIC in regions where it is strictly required (e.g., single-electron limits) while mitigating overcorrection in delocalized, high-density regions. The authors state that this approach improves the predictive accuracy of DFT for atomic and molecular properties, particularly electron affinities, and provides a general framework applicable to atomic, molecular, and solid-state systems. They emphasize that the method is a step toward improving calculations for heterogeneous catalysis, where accurate descriptions of both molecular adsorbates and solid substrates are necessary. The work does not claim universal superiority over all existing methods but demonstrates that LSSIC offers a balanced alternative that can outperform standard semi-local functionals and fixed-scaling SIC schemes in specific contexts.