Spread balanced Wannier functions: Robust and automatable orbital localization

TL;DR

This paper presents a new robust and automatable Wannier function localization method that minimizes spread with a penalty term, improving performance for complex systems and high-throughput applications.

Contribution

It introduces a spread-balanced Wannierization scheme with an automated protocol for initial guess selection and optimal band determination, enhancing robustness and applicability.

Findings

Effective for diverse complex systems including NV centers and ferroelectrics

Improves localization robustness and reduces ineffective solutions

Implemented in Python within ASE for accessibility

Abstract

We introduce a new type of Wannier functions (WFs) obtained by minimizing the conventional spread functional with a penalty term proportional to the variance of the spread distribution. This modified Wannierisation scheme is less prone to produce ineffective solutions featuring one or several poorly localized orbitals, making it well suited for complex systems or high-throughput applications. Furthermore, we propose an automatable protocol for selecting the initial guess and determine the optimal number of bands (or equivalently WFs) for the localization algorithm. The improved performance and robustness of the approach is demonstrated for a diverse set of test systems including the NV center in diamond, metal slabs with atomic adsorbates, spontaneous polarization of ferroelectrics and 30 inorganic monolayer materials comprising both metals and semiconductors. The methods are implemented in Python as part of the Atomic Simulation Environment (ASE).