Fermi-Löwdin-Orbital Self-Interaction Correction

Updated 11 June 2026

FLOSIC is a unitary-invariant, size-extensive method that corrects self-interaction errors in DFAs using Fermi orbitals and Löwdin orthonormalization.
It reduces computational scaling by optimizing 3N real-space Fermi-Orbital Descriptors, making the method viable for both molecular and periodic systems.
The approach enhances predictions of ionization potentials, reaction barriers, and electronic properties while addressing symmetry breaking through ensemble corrections.

Fermi-Löwdin-Orbital Self-Interaction Correction (FLOSIC) is a formally unitary-invariant, size-extensive approach to correcting the unphysical self-interaction errors (SIE) present in common density functional approximations (DFAs). It implements the Perdew–Zunger self-interaction correction (PZ-SIC) using a localized orbital basis built from Fermi orbitals parameterized by real-space descriptors, known as Fermi-Orbital Descriptors (FODs), which are subsequently orthonormalized via the Löwdin procedure. This methodology reduces the scaling bottleneck of variational SIC (scaling as $O(N^2)$ to $O(N^3)$ in orbital rotations) to a much more tractable $O(N)$ real-space parameter optimization, ensuring computational viability for molecular and periodic systems. FLOSIC has facilitated major advances in the accuracy of calculated ionization potentials, dissociation energies, reaction barriers, charge localizations, and spectroscopic properties across atoms, molecules, and solids.

1. Formalism: SIC Functional and Fermi–Löwdin Construction

The PZ-SIC energy functional augments the standard Kohn–Sham total energy by subtracting for each occupied orbital the Hartree and exchange–correlation self-energies: $E_{\rm SIC}[\{\phi_{i\sigma}\}] = E_{\rm DFA}[n_\uparrow, n_\downarrow] -\sum_{i\sigma} \left\{ U[n_{i\sigma}] +E_{\rm xc}^{\rm DFA}[n_{i\sigma},0] \right\}$ where $n_{i\sigma}(r) = |\phi_{i\sigma}(r)|^2$ , $U[n] = \frac12\int\int n(r)n(r')/|r-r'|\,dr\,dr'$ , and $E_{\rm xc}^{\rm DFA}[n,0]$ is the XC functional of a one-electron density.

In FLOSIC, the $\{\phi_{i\sigma}\}$ are not arbitrary localized orbitals: instead, for each spin $\sigma$ , the Fermi orbitals are defined using FODs $\{\mathbf{a}_{i\sigma}\}$ as

$O(N^3)$ 0

where $O(N^3)$ 1 are the canonical KS orbitals. The set $O(N^3)$ 2 is Löwdin-orthonormalized to yield the final Fermi–Löwdin orbitals (FLOs) $O(N^3)$ 3 used in $O(N^3)$ 4.

This procedure yields a size-extensive, unitarily invariant SIC functional dependent only on the $O(N^3)$ 5 FOD coordinates, rather than the $O(N^3)$ 6 possible occupied–occupied orbital rotations (Yang et al., 2017, Pederson, 2014, Jackson et al., 2019).

2. FOD Optimization and Self-Consistency

The total FLOSIC energy is minimized with respect to both:

The electronic density (standard SCF procedure)
The $O(N^3)$ 7 FODs via their analytic gradients: $O(N^3)$ 8 Algorithms alternate between optimizing the FODs at fixed density (often using gradient-based optimizers such as L-BFGS or conjugate-gradient) and SCF updates of the orbitals and density (including the SIC potentials from the current set of FODs) (Yang et al., 2017, Schwalbe et al., 2019).

The frozen-density loop—a practical acceleration—fixes the density while optimizing FODs to convergence before recomputing the density. This approach can yield significant speedups, as demonstrated in converging electronic structures for transition metal macrocycles (Karanovich et al., 2021).

3. Advantages and Theoretical Implications

FLOSIC corrects major known deficiencies of DFAs:

Localized spin/charge densities: Restoration of correct metal $O(N^3)$ 9-character and quantitation of spin populations in transition-metal complexes, as validated by comparison with multireference wavefunction calculations and EPR experiment (Karanovich et al., 2021).
Band gaps and orbital eigenvalues: FLOSIC systematically opens underestimated HOMO–LUMO (Kohn–Sham) gaps and improves photoemission spectra, extending even to periodic systems if combined with a Wannier–Fermi–Löwdin (WFL) localization or effective potential approaches (Shinde et al., 2022, Diaz et al., 2021).
Reaction and dissociation energetics: Barrier heights and dissociation curves—especially those suffering from stretched-bond SIE—are dramatically improved; the majority of the SIC correction to a reaction barrier is attributed quantitatively to the stretched-bond orbital at the transition state (Singh et al., 16 Jan 2026).

Size-extensivity and unitarity invariance make FLOSIC robust for both finite and extended systems (Yang et al., 2017, Shinde et al., 2022).

4. Symmetry Breaking, Multiple Minima, and Ensemble Correction

The PZ-SIC functional and its FLOSIC implementation can admit multiple symmetry-broken minima due to overlocalization of electrons. In aromatic molecules, e.g., benzene, canonical FLOSIC produces localized Lewis/Kekulé structures, breaking $O(N)$ 0 symmetry, or spin-symmetry broken Linnett double-quartet states (with incorrect $O(N)$ 1). Which minimum is found depends on the FOD initialization (LT1, LT2, LDQ) and artifactual dipole moments or symmetry breaking are frequently observed (Trepte et al., 2021, Schwalbe et al., 2024).

The ensemble generalization, E-FLO-SIC, restores symmetry by averaging the SIC energy over all chemically equivalent localized FOD configurations

$O(N)$ 2

which yields electron densities and energies invariant under equivalent resonance transformations, as demonstrated for benzene (Schwalbe et al., 2024). This approach can be extended to scaled and local SIC variants to eliminate symmetry breaking across a broad class of systems.

5. Scaling Strategies and Computational Efficiency

Practical application to large or periodic systems, or for balancing overcorrection, has prompted the development of:

Orbitalwise (OSIC) and selective (SOSIC/vSOSIC) scaling: Reduces SIC on core or many-electron-like orbitals, while applying full SIC on valence or frontier orbitals—restoring HOMO asymptotics and mitigating overcorrection in total energies and equilibrium properties (Yamamoto et al., 2020, Romero et al., 2023).
Locally scaled SIC (LSIC): Applies a local scaling function based on kinetic energy-density indicators (such as the iso-orbital indicator $O(N)$ 3), yielding restoration of correct uniform-gas limits and improved response properties (e.g. polarizabilities) (Akter et al., 2021).
WFL-SIC in periodic solids: SIC is applied to compact Wannier-localized orbitals, minimizing the self-interaction energy with respect to the Wannier charge centers for efficient and accurate structural and band-gap predictions at $O(N)$ 4 scaling (Shinde et al., 2022).

Other computational efficiency innovations include real-space implementations, generalized Slater averages, and exploitation of analytic FOD gradients (Diaz et al., 2021, Schwalbe et al., 2019, Pederson, 2014).

6. Representative Applications

Significant application domains and validation benchmarks include:

Transition-metal complexes: Recovery of correct d/p character, opening of the HOMO–LUMO gap, and proper magnetic exchange coupling (Karanovich et al., 2021, Romero et al., 2023).
Water clusters and hydrogen-bonded systems: Correction of overbinding in DFA and alignment with CCSD(T) reference energetics (Wagle et al., 2020).
Polyacenes and conjugated π-systems: Improved static polarizabilities and IPs, especially with local scaling (LSIC), eliminating both overestimation (DFA) and overcorrection (unscaled SIC) (Akter et al., 2021).
Reaction barriers: Dramatic improvement in the accuracy of stretched-bond barrier heights, with the largest SIC correction traced universally to the most delocalized (stretched) orbital; combination with SCAN meta-GGA achieves state-of-the-art semi-local results (Singh et al., 16 Jan 2026).
Photoemission and orbital energies: When FLOSIC is coupled with DCEP or OEP/KLI schemes, it yields physically meaningful orbital eigenvalues and improved fundamental gap predictions (Diaz et al., 2021, Diaz et al., 2021).

Key limitations and current directions include:

Overcorrection and loss of equilibrium accuracy: Full SIC often overcorrects properties well-described by DFAs. Orbitalwise/local scaling approaches address this by limiting the SIC to where SIE is dominant (Yamamoto et al., 2019, Yamamoto et al., 2020, Akter et al., 2021).
Symmetry-breaking minima: The prevalence of multiple local minima and symmetry breaking in delocalized or aromatic systems is addressed by ensemble corrections and FOD-initialization strategies (Schwalbe et al., 2024, Trepte et al., 2021).
Scaling and parallelization: FLOSIC’s computational tractability is enhanced through the reduction from $O(N)$ 5 to $O(N)$ 6 parameters, nested SCF plus FOD loops, frozen-density acceleration, and efficient real-space and periodic implementations (Yang et al., 2017, Shinde et al., 2022, Diaz et al., 2021).
Extensions: Ongoing developments include further tuning of local indicators for scaling (e.g. ELF, $O(N)$ 7, $O(N)$ 8), hybrid-FLOSIC methods, excited-state and vibrational analyses via analytic forces, and robust FOD-optimization algorithms (Akter et al., 2021, Jackson et al., 2019).

In sum, FLOSIC—by combining the formal rigor of unitary-invariant, size-extensive SIC with algorithmic efficiency and chemical interpretability—has established itself as a critical tool for overcoming key failures of standard DFAs, with broad impact across quantum chemistry and computational materials science (Yang et al., 2017, Jackson et al., 2019, Karanovich et al., 2021, Shinde et al., 2022).

Key References: