arXiv Science⌕ Search
Home/Archive/arXiv

arXiv · 2610.30641

How good are universal MLIPs for alloys? A benchmark is only as trustworthy as its controls

Gus L. W. Hart·Jacob N. Pursley
View original source ↗Read source document ↗Export metadata ↓

Abstract

Universal machine-learned interatomic potentials (uMLIPs) are ranked by their formation-energy error against a DFT database, as if that database were ground truth, but independent DFT databases disagree. We benchmark fourteen uMLIPs on 1.45 million ordered intermetallics against AFLOW, which none trained on, and compare every energy and stability error with that disagreement. For half the structures, AFLOW, Alexandria, and OQMD differ by less than 6-8 meV/atom, but a minority differ by far more, raising the average to 16-30. That disagreement sets a floor on any model's measurable error. Re-referenced, the best uMLIP agrees with AFLOW as closely as Alexandria does, and the next four sit 2-7 meV/atom above the floor. Once broken elemental references are screened, the energy and stability rankings agree almost everywhere. Three apparent model failures shrink or vanish under the right control: poorly correlated shear moduli, thousands of "bad" DFT references, and the leader's inflated lead.

Explore related subjects

cond-mat.mtrl-sci

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Gus L. W. Hart, Jacob N. Pursley. 2026-10-22. How good are universal MLIPs for alloys? A benchmark is only as trustworthy as its controls. https://arxiv.org/abs/2610.30641

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Order-Disorder in Fe-Si Alloys: Implications for Seismic and Transport Properties of Earth's Inner Core

Understanding the structure and dynamics of Earth's inner core is essential for constraining its composition, thermal evolution, and seismic behavior. Silicon is widely considered a major light element in the core. Here, we use first-principles molecular dynamics and thermodynamic modeling to investigate the structural, elastic, and transport properties of iron-silicon (Fe-Si) alloys under inner-core conditions. By computing the Gibbs free energies of multiple solid phases, including ordered B2, hexagonal close-packed (hcp), face-centered cubic (fcc), and body-centered cubic (bcc) Fe-Si, we construct a phase diagram relevant to inner-core pressures and temperatures. We identify a pronounced miscibility gap between hcp and cubic Fe--Si phases, with phase coexistence over approximately 6--11 wt\% Si at 6000 K. The cubic phase exhibits a strong order-dependent elastic response, with disordered bcc showing a large intrinsic shear anisotropy (22.9\% at 6000 K), in contrast to the nearly isotropic hcp phase (0.6\%). Meanwhile, disordered bcc Fe--Si produces shear wave velocities ($V_S=3.73$ km$\cdot$s$^{-1}$) and Poisson's ratios comparable with seismological constraints. In addition, disordered bcc Fe--Si exhibits a lower thermal conductivity than hcp Fe--Si, providing a quantitative constraint on phase-dependent electronic heat transport. These results demonstrate that chemical ordering strongly controls the stability and physical properties of cubic Fe--Si alloys, and suggest that metastable or localized cubic Fe--Si phases may contribute to the structural and transport complexity of Earth's inner core.

cond-mat.mtrl-sci↗
arXiv2026

Dynamical Nonrelativistic Spin Splitting via THz Nonlinear Phononics

Nonrelativistic spin splitting (NRSS) in collinear antiferromagnets, including altermagnets and compensated ferrimagnets, offers a route to high-frequency spintronics immune to stray fields, but its dynamic control has remained elusive. We demonstrate, using density functional theory (DFT) and nonlinear phononics, that THz laser pulses can achieve ultrafast, reversible control of NRSS on picosecond timescales in antiferromagnets. We derive two mode-selection criteria, accounting for phonon and magnetic wavevector compatibility and inversion parity, to identify which phonon modes can activate NRSS. Our calculations illustrate the creation of NRSS in spin-degenerate NiO and the enhancement of existing splitting in LaFeO$_3$. While practical implementation depends on excitation conditions, lattice dynamics, and mode coupling, these results establish design principles for identifying materials with stronger and more accessible responses. Limitations imposed by driving fields, phonon damping, and anharmonic detuning motivate further materials screening based on accessible excitation strengths, lattice stability, resonance frequency matching, and favorable anharmonic coupling.

cond-mat.mtrl-sci↗
arXiv2026

Decoding Composition-Structure-Performance Relations in Sodium Layered Oxide Positive Electrode Materials

Layered sodium transition-metal (TM) oxides, Na$_x$TMO$_2$, are among the most promising sodium-ion positive electrode materials, yet their Na/vacancy orderings, layer stacking sequences, and cooperative Jahn--Teller distortions are known only for a few single-TM compounds. We compile the Na/vacancy orderings reported in earlier studies into a single library of 41 unique orderings and apply this uniformly to five single-TM and all ten Na$_x$TM$_{0.5}$TM$'_{0.5}$O$_2$ oxides (TM = Ti, Mn, Fe, Co, Ni), screening more than 120,000 structures with machine-learning interatomic potentials prior to 1,025 accurate first-principles calculations. The phase diagrams explain the observed O3/P3/O1 sequences and voltages, and the O3-to-P3 competition hinges on ordering energies of a few meV per atom that scalar descriptors, such as the cationic potential and the interlayer-slab distance, miss. Our extensive analysis reveals that the redox couples form a nearly partner-independent ladder, Ti$^{3+/4+}$ $<$ Mn$^{3+/4+}$ $<$ Co$^{3+/4+}$ $<$ Ni$^{2+/3+}$ $<$ Fe$^{3+/4+}$, explaining pristine charge transfer and the oxidation sequence. Van Vleck analysis of every TMO$_6$ octahedron reveals that TM mixing suppresses the cooperative Jahn--Teller distortions of Mn$^{3+}$ and Ni$^{3+}$ through charge transfer to Jahn--Teller-inactive states or dilution of Jahn--Teller-active centers. We formulate an energy-density-versus-strain map that yields a design rule: a high-voltage couple without Jahn--Teller strain, which Mn$^{4+}$/Ni$^{2+}$ charge transfer realizes in Na$_x$Mn$_{0.5}$Ni$_{0.5}$O$_2$, the most practical composition. We release the ordering-resolved dataset as a training set for subsequent training of accurate machine-learning interatomic potentials.

cond-mat.mtrl-sci↗