Machine-learned potentials for iron oxides at extreme conditions

Reliable phase stability requires accurate relative free energies, which small force errors alone do not guarantee. We develop atomic-cluster-expansion potentials for five iron oxides from Fe$_2$O to Fe$_2$O$_3$, covering crystalline and liquid states at 140--260\,GPa and 1000--11000\,K, with electronic and ionic temperatures matched. Force-dominated training leaves systematic energy offsets between structure families of up to 41\,meV/atom that persist despite additional training data. Stronger energy weighting and weaker regularization largely remove these offsets, while converged k-point sampling prevents further composition-dependent biases. The final potentials achieve held-out energy errors of 2.0--5.5\,meV/atom and first-order free-energy errors within 3.4\,meV/atom for liquids at 5000--9000\,K. However, an untrained NiAs-type FeO phase exhibits a free-energy error of 57\,meV/atom despite a low extrapolation grade. These results show that reliable phase-diagram calculations require control of relative energy offsets and explicit free-energy validation of every competing phase.

Authors

Publication Details

Journal
arXiv (Cornell University)
Published
2026-11-18
DOI
https://doi.org/10.48550/arxiv.2611.24525
Primary Topic
Machine Learning in Materials Science
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Machine-learned potentials for iron oxides at extreme conditions

A. Principi, T. D. Kühne, M. S. Bahramy, S. M. Vinko et al.
arXiv (Cornell University)
Machine Learning in Materials Science
preprint

Machine-learned potentials for iron oxides at extreme conditions

A. Principi, T. D. Kühne, M. S. Bahramy, S. M. Vinko, S. Azadi
preprint en

Abstract

Reliable phase stability requires accurate relative free energies, which small force errors alone do not guarantee. We develop atomic-cluster-expansion potentials for five iron oxides from Fe$_2$O to Fe$_2$O$_3$, covering crystalline and liquid states at 140--260\,GPa and 1000--11000\,K, with electronic and ionic temperatures matched. Force-dominated training leaves systematic energy offsets between structure families of up to 41\,meV/atom that persist despite additional training data. Stronger energy weighting and weaker regularization largely remove these offsets, while converged k-point sampling prevents further composition-dependent biases. The final potentials achieve held-out energy errors of 2.0--5.5\,meV/atom and first-order free-energy errors within 3.4\,meV/atom for liquids at 5000--9000\,K. However, an untrained NiAs-type FeO phase exhibits a free-energy error of 57\,meV/atom despite a low extrapolation grade. These results show that reliable phase-diagram calculations require control of relative energy offsets and explicit free-energy validation of every competing phase.

arXiv (Cornell University)
Machine Learning in Materials Science
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon