Correcting Delocalization Error in Materials with Localized Orbitals and Linear-Response Screening

Delocalization error prevents density functional theory (DFT) from reaching its full potential, causing problems like systematically underestimated band gaps and misaligned energy levels at interfaces. We introduce lrLOSC to correct delocalization error in materials over a wide range of band gaps. We predict eleven materials’ fundamental gaps to within 0.22 eV, while offering a nonzero total energy correction; molecular properties are improved with a parallel implementation of the same theory [J. Phys. Chem. Lett. 16, 2492 (2025)]. lrLOSC is an essential step toward modeling molecules, materials, and their interfaces within the same DFT framework.

💡 Research Summary

This paper introduces a novel method named “lrLOSC” (linear-response Localized Orbital Scaling Correction) designed to correct the pervasive “delocalization error” in Density Functional Theory (DFT), specifically for extended material systems. Delocalization error is a fundamental flaw in approximate DFT functionals that leads to the systematic underestimation of band gaps, misalignment of energy levels at interfaces, and size-inconsistency problems.

The core innovation of lrLOSC lies in the synergistic combination of two key ingredients: dually localized Wannier functions (DLWFs) and system-dependent linear-response screening. Traditional delocalization error corrections fail for periodic materials because Bloch orbitals are infinitely delocalized, yielding zero energy correction. lrLOSC overcomes this by constructing DLWFs, which are localized orbitals obtained by minimizing a cost function mixing spatial and energy variance. Crucially, DLWFs are allowed to mix occupied (valence) and unoccupied (conduction) Bloch states. This mixing results in non-integer local orbital occupancies even for insulators, enabling a non-zero total energy correction—a feature absent in other accurate band-gap correction methods like Koopmans-compliant or Wannier-Koopmans approaches.

The second ingredient, linear-response screening, determines the magnitude of the correction, known as the “curvature” (κ). Instead of using an empirical or system-independent screening parameter, lrLOSC calculates κ by screening the bare Hartree-exchange-correlation kernel through the system’s static density-density linear response function (χ). This physically accounts for orbital relaxation effects, ensuring the correction is tailored to the electronic screening properties of each specific material (metal, semiconductor, or insulator).

The authors implement lrLOSC as a one-shot post-processing correction on top of standard PBE-DFT calculations. They demonstrate its performance on thirteen prototypical semiconductors and insulators. The results are impressive: lrLOSC predicts the fundamental gaps of eleven materials with a root-mean-square error of only 0.22 eV relative to experimental values. Furthermore, it significantly improves the overall band structures, correcting, for instance, the nature of the band gap in silicon. Critically, it achieves this high accuracy while providing a non-zero correction to the total energy of these gapped systems.

The study positions lrLOSC as a pivotal step towards a unified framework for quantum-mechanical simulations. By successfully correcting both orbital energies (and thus band gaps) and total energies in materials, it addresses a major shortcoming of previous methods. This dual capability is essential for achieving true size-consistency and for reliably modeling hybrid systems involving both finite molecules and extended materials, such as interfaces and adsorbates on surfaces. The paper acknowledges current limitations, such as the one-shot implementation potentially breaking some band degeneracies, and points to future work towards self-consistent and more efficient implementations.