- Accepted Paper
Oxygen transport and oxide pre-nucleation in liquid aluminum from ab initio and machine-learned molecular dynamics
Phys. Rev. Materials - Accepted 10 November, 2026
DOI: https://doi.org/10.1103/bsq2-r7xp
Phys. Rev. Materials - Accepted 10 November, 2026
DOI: https://doi.org/10.1103/bsq2-r7xp
The transport of oxygen in liquid aluminum governs high-temperature metallurgy and aluminum combustion, yet its atomistic mechanism remains poorly characterized. Combining ab initio molecular dynamics (AIMD) with classical molecular dynamics, we quantify oxygen diffusion and evidence oxide pre-nucleation stages in liquid aluminum from 1200 to 2400 K. AIMD reveals that oxygen migrates by thermally activated, aluminum exchange-mediated hopping: it diffuses more slowly than aluminum at 1200 K but becomes the faster species from 1800 K onward, and its diffusion activation energy (46.3 kJ/mol) far exceeds that of aluminum self-diffusion (25.7–27.8 kJ/mol), signaling a strongly constrained migration (trigonal Al3O) in the metallic matrix. Under dilute conditions (O/Al = 0.008 and 0.024) AlxO2 motifs persist across the whole temperature range with metastable AlxO3 species being present at 1200 K, demonstrating a trend for the oxygen atoms to interact with each other and making oxide nucleation a speciation problem. An on-the-fly machine-learned potential trained during the AIMD reproduces these features while reaching nanosecond, thousandatom scales, whereas the widely used ReaxFF and EAM potentials fail: ReaxFF suppresses oxygen mobility through overly stiff bonding and EAM overestimates it by missing the directional Al–O interaction. These results provide a quantitative, atomic-scale picture of oxygen transport and oxide pre-nucleation in molten aluminum.
If the author has provided any supplemental materials with this article they will be available upon publication of the version of record.