🔬 materials science

A test drive for exchange-correlation functionals on noncollinear magnets: Mn $_3$ Ir, Mn $_3$ Ge, NiS $_2$ , and YMnO $_3$

This study demonstrates that spin-current density-functional theory (SCDFT) functionals, specifically NCMSCAN and LFNCBR89-NCCS, successfully recover the experimental spin texture of the noncollinear magnet NiS $_2$ at a computational cost comparable to standard semi-local approximations, outperforming traditional spin-DFT extensions in capturing noncollinear physics.

Original authors: Marie-Therese Huebsch, Martijn Marsman, Jacques K. Desmarais, Stefano Pittalis, Fabien Tran

Published 2026-08-25

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Marie-Therese Huebsch, Martijn Marsman, Jacques K. Desmarais, Stefano Pittalis, Fabien Tran

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Magnetism is a force we encounter daily, from the compass needle finding north to the hard drive storing our memories. Yet, beneath the simple idea of a north and south pole lies a complex reality where the tiny magnetic arrows inside atoms can point in any direction, not just up or down. When these arrows twist and turn in three-dimensional space, creating a non-collinear magnetic structure, the physics becomes incredibly difficult to predict. Scientists rely on a powerful set of rules called density-functional theory to simulate how matter behaves at the atomic level. However, these rules depend on an approximation for how electrons interact with one another, a piece of the puzzle that has historically struggled to describe these twisting magnetic patterns accurately. If the approximation is wrong, the simulation might predict a material is non-magnetic when it is actually magnetic, or it might get the direction of the magnetic arrows completely wrong.

A team of researchers from Austria, Italy, and the United States has put this problem to the test by comparing different versions of these rules against real-world materials. They focused on four specific substances: two metals made of manganese mixed with iridium or germanium, a nickel-sulfur compound, and a yttrium-manganese oxide. These materials are special because their magnetic atoms do not align in a simple, straight line; instead, they form intricate, swirling patterns. The researchers used a sophisticated computer program to simulate these materials using two new, advanced sets of rules designed specifically for these twisting magnetic states, and they compared the results against four older, more common sets of rules that were originally built for simpler magnetic alignments. The goal was to see if the new rules could capture the true nature of these materials better than the established methods.

The study revealed a clear winner when it came to getting the magnetic pattern right. For the nickel-sulfur compound, only the two new, advanced rules successfully reproduced the exact magnetic texture observed in experiments. The older, more common rules failed completely; some predicted the material had no magnetism at all, while others settled on a different, incorrect magnetic arrangement. This was a significant finding because it proved that for certain complex magnetic materials, sticking to the old, simpler rules leads to a fundamentally wrong picture of reality. The new rules, which account for the flow of electron currents in a way the old ones do not, were the only ones capable of seeing the true magnetic landscape.

However, getting the magnetic pattern right did not mean the new rules were perfect at everything else. When the researchers looked at other properties, such as the precise size of the crystal structure or the energy gap that makes a material an insulator, the results were mixed. One of the older rules, known for its general reliability, actually predicted the size of the crystal and the energy gap more accurately than the new rules for some materials. The new rules sometimes predicted the material was too small or too large, and they struggled to get the size of the magnetic moments exactly right. This suggests that while the new rules are essential for seeing the correct magnetic shape, they still need refinement to match the quantitative details of real-world measurements.

The researchers also discovered that the size of the magnetic moment—the strength of the tiny atomic magnet—plays a surprisingly large role in how the computer simulations behave. In the metal compounds, the different rules produced two distinct groups: one group predicted small magnetic moments and a compressed crystal structure, while another predicted large moments and a more expanded structure. This difference in the predicted strength of the magnetism led to large variations in the calculated energy required to twist the magnetic directions, a property known as magnetic anisotropy. Even though all the rules agreed on the general type of magnetic order for one of the metals, the energy differences they calculated varied by a factor of three or four, simply because they disagreed on how strong the individual magnets were.

Ultimately, this work demonstrates that the new, advanced rules are a necessary tool for understanding non-collinear magnets, but they are not yet a complete solution. They succeed where the old rules fail by capturing the correct magnetic texture, a qualitative victory that is crucial for understanding materials like the nickel-sulfur compound. Yet, the fact that they do not yet uniformly outperform the older rules in predicting every physical property shows that the field is still evolving. The study highlights that describing these complex magnetic systems requires a delicate balance, and while the new framework provides a consistent and promising path forward, further testing and development are needed to make these simulations as accurate as possible across all types of magnetic materials.

Technical Summary: A Test Drive for Exchange-Correlation Functionals on Noncollinear Magnets

Problem Statement
The accurate first-principles description of noncollinear magnets remains a significant challenge in density-functional theory (DFT). While practical predictions rely heavily on the exchange-correlation (XC) approximation, developing these approximations for noncollinear systems is difficult because the balance between exchange and correlation dictates not only structural and electronic properties but also the relative stability of competing magnetic orders.

Standard approaches typically utilize spin-DFT (SDFT), which projects noncollinear densities onto a local spin-quantization axis to employ spin-up/spin-down functionals. This "locally collinear" prescription, while extended heuristically to (meta-)generalized gradient approximations (GGA/MGGA), fails to capture transverse spin gradients and lacks exact constraints available in more rigorous frameworks. Specifically, SDFT does not inherently satisfy local $U(1) \times SU(2)$ gauge invariance, a condition crucial for the consistent treatment of vector-potential couplings, spin currents, and spin-orbit coupling (SOC). Spin-current DFT (SCDFT) offers a framework that admits noncollinear approximations constrained by this exact gauge invariance, yet the practical behavior of such functionals on real materials has remained largely untested beyond proof-of-principle applications.

Methodology
The authors employed the Vienna ab initio simulation package (VASP) to compare two recently developed noncollinear SCDFT functionals against locally collinear SDFT extensions of four standard functionals. The study focused on a diverse set of noncollinear $d$ -electron systems: the itinerant antiferromagnets Mn $_3$ Ir (cubic) and Mn $_3$ Ge (hexagonal), the narrow-gap Mott insulator NiS $_2$ (cubic), and the multiferroic insulator YMnO $_3$ (hexagonal).

The tested functionals included:

Locally Collinear SDFT Extensions: PZ and PW92 (LSDA), PBE (GGA), and SCAN (MGGA). These were implemented using the standard projection of spin-density and kinetic-energy-density matrices onto the local magnetization direction.
Noncollinear SCDFT Functionals:
- NCMSCAN: A noncollinear version of the SCAN functional constructed by enforcing $U(1) \times SU(2)$ invariance through minimal modifications to the spin-polarization and iso-orbital indicator variables.
- LFNCBR89-NCCS: A functional combining a Laplacian-free noncollinear variant of the Becke–Roussel exchange (BR89) with a noncollinear form of the Colle–Salvetti correlation (CS).

The computational workflow involved self-consistent field (SCF) loops where SCDFT treated the system fundamentally differently from SDFT, utilizing 2 $\times$ 2 densities and generating functional derivatives that yield non-Abelian XC vector potentials. The study performed volume relaxations, calculated equilibrium lattice constants and bulk moduli, evaluated fundamental band gaps and on-site magnetic moments, and mapped magnetic energy landscapes for various symmetry-classified configurations. For NiS $_2$ , the study also investigated the impact of an on-site Coulomb interaction ( $U$ ) within the DFT+ $U$ framework.

Key Contributions and Results

Recovery of Experimental Ground States: The most significant finding is that among all tested functionals, only the two SCDFT functionals (NCMSCAN and LFNCBR89-NCCS) successfully recover the experimental noncollinear magnetic ground state (spin texture) of NiS $_2$ . In contrast, the locally collinear functionals failed qualitatively: PBE stabilized a collinear altermagnetic order, SCAN selected a different magnetic order, and the LSDA functionals (PZ, PW92) converged to nonmagnetic solutions.
Magnetic Moments and Band Gaps:
- For NiS $_2$ , the plain (U=0) SCDFT functionals provided the closest agreement with the experimental on-site magnetic moment (0.98 $\mu_B$ ). Specifically, plain NCMSCAN yielded 0.95 $\mu_B$ (within experimental uncertainty) and LFNCBR89-NCCS yielded 1.11 $\mu_B$ (within 13% of the measured value). In contrast, applying a Coulomb correction ( $U$ ) to SCAN or NCMSCAN to match the experimental volume resulted in significant overestimation of the moment (e.g., NCMSCAN+U(2.45 eV) yielded ~1.44 $\mu_B$ ). The LSDAs collapsed to a nonmagnetic state.
- However, quantitative accuracy varied. While SCDFT recovered the correct magnetic order and moments without $U$ , it did not uniformly outperform locally collinear relatives in other observables. For instance, SCAN remained the most reliable for predicting equilibrium volumes and band gaps across the set, whereas SCDFT functionals showed material-specific volume errors (e.g., NCMSCAN significantly underestimated the volume of Mn $_3$ Ir and Mn $_3$ Ge).
- For the insulator YMnO $_3$ , all functionals overestimated the magnetic moment, and only SCAN reached the experimental charge-transfer band gap.
Magnetic Energy Landscapes and Anisotropy:
- In Mn $_3$ Ir, while all functionals predicted the same ground state symmetry ( $T^A_{1g}$ ), the magnetic anisotropy energy (MAE) and exchange stiffness varied significantly.
- The study revealed that the magnitude of the on-site magnetic moment acts as a primary organizing variable for these discrepancies. Functionals were split into a "small-moment" group (PZ, PW92, NCMSCAN) and a "large-moment" group (SCAN, LFNCBR89-NCCS).
- The MAE values for the large-moment group were roughly three times higher than those for the small-moment group. However, when normalized by the square of the on-site moment, the spread in anisotropy and exchange parameters collapsed, suggesting that errors in the moment magnitude propagate quadratically into extracted spin-model parameters.
Computational Cost: The study quantified the computational overhead of SCDFT. A single SCDFT SCF step costs approximately 7 times that of an LSDA step and 1.5 times that of an MGGA step. However, in terms of total time-to-solution for a full workflow, the distinction between MGGA and SCDFT becomes negligible due to convergence behavior, indicating that the computational cost should not be a barrier for most applications.

Significance
The paper concludes that SCDFT constitutes a promising framework for capturing noncollinear physics, particularly in cases where locally collinear functionals fail to identify the correct magnetic ground state, as demonstrated by the NiS $_2$ case study. The results highlight that while SCDFT functionals offer a qualitatively correct description of noncollinear magnetism where SDFT fails, they do not yet provide a uniform quantitative improvement across all observables (volume, gaps, moments) compared to established locally collinear functionals.

The authors emphasize that the difficulty in modeling noncollinear magnets is not merely a matter of parameter sensitivity but can involve qualitative failures of the underlying functional form. They argue that the on-site magnetic moment is a critical variable that influences structural properties (via the magnetovolume effect) and derived spin-model parameters. The study underscores the need for further validation of SCDFT approximations across a wider range of experimental observables and magnetic systems to fully realize their potential.

More like this