Benchmarking density functional theory on the prediction of antiferromagnetic transition temperatures

TL;DR

This paper evaluates the accuracy of various density functional theory methods in predicting antiferromagnetic transition temperatures, introduces a novel approach for calculating exchange parameters, and proposes corrections to improve predictive accuracy.

Contribution

It develops an innovative null space analysis method for exchange parameter calculation and applies self-consistent linear response corrections within DFT for better predictions.

Findings

GGA overestimates transition temperatures by ~113%.

GGA+U underestimates by ~53%.

Adjusting GGA+U results reduces error to 44%.

Abstract

This study investigates the predictive capabilities of common DFT methods (GGA, GGA+$U$, and GGA+$U$+$V$) for determining the transition temperature of antiferromagnetic insulators. We utilize a dataset of 29 compounds and derive Heisenberg exchanges based on DFT total energies of different magnetic configurations. To obtain exchange parameters within a supercell, we have devised an innovative method that utilizes null space analysis to identify and address the limitations imposed by the supercell on these exchange parameters. With obtained exchanges, we construct Heisenberg Hamiltonian to compute Transition temperatures using classical Monte Carlo simulations. To refine the calculations, we apply linear response theory to compute on-site ($U$) and intersite ($V$) corrections through a self-consistent process. Our findings reveal that GGA significantly overestimates the transition temperature (by ~113%), while GGA+$U$ underestimates it (by ~53%). To improve GGA+$U$ results, we propose adjusting the DFT results with the $(S+1)/S$ coefficient to compensate for quantum effects in Monte Carlo simulation, resulting in a reduced error of 44%. Additionally, we discover a high Pearson correlation coefficient of approximately 0.92 between the transition temperatures calculated using the GGA+$U$ method and the experimentally determined transition temperatures. Furthermore, we explore the impact of geometry optimization on a subset of samples. Using consistent structures with GGA+U and GGA+U+V theories reduced the error.