Hunting for the Magnet That Splits Nothing
31 May 2026, Yanjiang
In the altermagnet RbMn₂Te₂O, spin-up and spin-down electrons are split by nearly two electronvolts, while the crystal itself casts no stray magnetic field.
Altermagnets that promise to split spins with the force of a ferromagnet yet cast no stray field. It might sound like a material that breaks the rules of magnetism, but a new computational screening effort suggests that nature offers such crystals in abundance — if we know where to look. In a preprint (arXiv:2605.27888), a team led by Jing‑Yang You at Beihang University, collaborating with researchers from Shanxi University and the Institute of Theoretical Physics in Beijing, reports the discovery of 34 promising altermagnetic compounds, including one predicted to separate spin‑up and spin‑down electrons by a colossal 1.88 electronvolts while keeping its magnetic order stable well above room temperature.
Altermagnets are the Jekyll‑and‑Hyde materials of solid‑state physics. Like Dr. Jekyll, they present a placid exterior to magnetic probes — no net magnetization, no stray field. But inside, they harbour a ferocious capacity to polarize currents by spin, a trait traditionally reserved for ferromagnets. This dual identity arises from a lattice of opposite spins locked in perfect cancellation yet arranged with a crystal symmetry that twists the electronic band structure so that spin‑up electrons travel through certain directions as if pushed by an invisible tailwind, while spin‑down electrons feel a headwind. The idea is not that the electrons have intentions; it is that the crystal’s exchange field — a quantum‑mechanical force acting differently on opposite spins — acquires a momentum‑dependent pattern because the sublattice of magnetic atoms is connected by a specific rotation, say a ninety‑degree turn combined with a spin flip. The result is spin splitting without spin accumulation: order that reveals itself only when you look at how electrons move, not at the total magnetic moment.
For years, experimentalists have struggled to find altermagnets that are both thermodynamically stable and display giant spin splitting. The symmetry requirements are so stringent that candidate materials have been rarer than hen’s teeth. The team’s breakthrough is a machine‑learning‑accelerated pipeline that transforms the search from a needle‑in‑a‑haystack problem into a targeted hunt. They began with 8640 compounds in the tetragonal AB₂C₂D family — a structural playground of stacked square lattices — and used symmetry filters to isolate 1347 that, in principle, could host compensated antiferromagnetic order consistent with altermagnetism. That is the raw ore.
Next came the alchemist’s fire: an interpretable XGBoost model. This machine‑learning algorithm operates like a bloodhound trained to recognise a scent. Instead of performing costly first‑principles calculations on every candidate, the team computed the spin splitting for a subset using density functional theory (DFT), the quantum‑mechanical workhorse for materials predictions. The XGBoost model learned to predict the spin splitting from chemical composition and simple structural features — effectively sniffing out the compounds that would later show giant splitting in full DFT calculations. It is a shortcut that works because the physics of spin splitting is strongly imprinted on the local symmetry and the hybridization of atomic orbitals, and the machine‑learning model captures those signatures without understanding them. The XGBoost reduced the 1347 candidates to a shortlist of 34 with non‑relativistic spin splittings exceeding 1.5 eV — a value on a par with the bandgap of a visible‑light LED. Four of the thirty‑four were already known, giving confidence in the procedure; the rest are new predictions.
The star of the shortlist is RbMn₂Te₂O, a layered oxide where manganese atoms sit on two sublattices related by a ninety‑degree rotation plus a spin flip — the precise symmetry lock that altermagnetism demands. The team’s DFT calculations for this compound reveal a maximum spin splitting of about 1.88 eV, which is like a customs officer at a border who checks every electron’s spin and sends it onto a different track. The estimated Néel temperature, the temperature below which the magnetic order sets in, is roughly 390 K (about 117 °C) — comfortably above room temperature, meaning the material could operate without cryogenic cooling. The giant splitting, the authors argue, arises from a bespoke amplifier: the manganese sublattice’s exchange fields are directionally locked by symmetry, and the hybridization between Mn‑d and Te‑p orbitals boosts the effect through directed covalent bonding. It is not a generic property but a carefully orchestrated concert of crystal symmetry and quantum chemistry.
But the story takes a surprising turn. The team also examined SrMn₂Te₂O, a chemical cousin, and found a soft‑mode‑driven structural transition — a collective vibrational instability that causes the crystal to spontaneously reconstruct its atomic arrangement. Imagine a stack of playing cards that, on its own, slides into a new configuration. The transition changes the interlayer coupling and effectively crosses a dimensionality boundary. Yet, remarkably, the altermagnetic spin splitting survives the upheaval. The unfolded electronic band structure, a technique that recovers the primitive‑zone picture from the reconstructed supercell, shows that the fundamental spin‑splitting mechanism is robust against lattice reconstruction. It is as if the altermagnetic order is rooted in the local symmetry of the magnetic sublattice, a deep geological stratum that is not washed away by the surface remodeling of the lattice.
Adding another dimension, the team computed the effect of hydrostatic pressure. Squeezing the material can non‑monotonically modulate the spin‑split Fermi surface — the contour of energy states where electrons conduct — by tweaking local coordination and orbital hybridization. Pressure emerges as a tuning knob, a way to dial the spin splitting up or down without altering the chemical formula. This opens the prospect of strain‑engineered altermagnetic devices where the spin filtering becomes dynamically controllable.
What the Spin Texture Reveals — and Conceals
Amid these results, a close examination of the evidence surfaces an important scientific question. The authors claim that the spin‑split bands exhibit a “d‑wave‑like” angular dependence — a pattern with four alternating lobes of spin polarisation, like a four‑leaf clover. If rigorously confirmed, such a symmetry would assign a precise symmetry class to the altermagnetic order, with direct implications for how the spin filtering behaves in real devices. However, the identification rests primarily on visual inspection of spin‑resolved band structures at a few momentum cuts, rather than on a quantitative Fourier decomposition or systematic fitting of the angular dependence at fixed energy. Earlier tight‑binding studies of related altermagnetic families, such as the V₂Se₂O compounds modelled by Cheng and colleagues, have demonstrated that spin‑splitting symmetry can be subtle, depending on the detailed interplay of multiple orbitals. A definitive symmetry assignment for RbMn₂Te₂O therefore awaits a more quantitative analysis — a task that should be straightforward with the computational data already in hand.
There is a broader question, too, about the path from prediction to reality. The XGBoost model was trained on DFT spin‑splitting data, which, while reliable, does not capture all the complexities of real crystals — electron correlation effects, defects, or subtle structural distortions that can break the required symmetries. Altermagnets have a history of surprising researchers: a theoretically predicted giant splitting can vanish if an unexpected lattice relaxation changes the site symmetries. The 34 candidates are, at this stage, a high‑resolution map, not a confirmed treasure. The next chapter must be written by synthetic chemists and experimental spectroscopists.
None of this diminishes the achievement of the machine‑learning pipeline, which has turned an exhaustive scan of thousands of compounds into a focused collection of plausible altermagnets. The approach is a template for the broader pursuit of functional quantum materials, where the search space is astronomically large and the needle is hidden by symmetry constraints. The preprint’s data‑driven workflow may accelerate the discovery not only of altermagnets but of other compounds whose electronic properties are encoded in crystal symmetry in ways that are invisible to traditional chemical intuition.
What makes the altermagnet quest so resonant, though, is that it forces us to loosen our grip on what magnetic order means. For a century, we have sorted magnetic materials into a taxonomy inherited from macroscopic moments: ferromagnets, antiferromagnets, ferrimagnets. Altermagnets escape this classification. They are antiferromagnets to a magnetic probe, yet ferromagnets to a spin‑polarised current. They achieve spin splitting without spin accumulation — order without an order parameter of the usual kind. In the language of symmetry, they occupy a “hidden order” phase, one that reveals itself only when we examine the momentum‑dependent spin texture, much as certain exotic superconductors disclose their secrets through angular‑resolved experiments. The magnet that splits nothing is, in the end, a cipher: it forces us to look at the spin landscape with new eyes, not at the total magnetic count, but at the directional fingerprints left by the very architecture of the crystal.
— Yanjiang
Yanjiang is an online editor of LoomSci.com.
References
- Yi-Fei Jiang et al., Machine-learning-accelerated discovery of synthesizable high-temperature altermagnets with giant spin splitting, arXiv:2605.27888
- Cheng et al., Realistic tight-binding model for V2Se2O-family altermagnets, arXiv:2602.09465