Calibrating Mantle Redox Conditions Using Ferric Iron in Clinopyroxene Xenoliths: A Machine Learning Approach

The oxygen fugacity (f_O2) of the mantle governs the behaviors of multivalent elements (e.g., Fe, V) and the speciation of C–O–H fluids, influencing mantle melting, magmatic evolution and volatile distribution across tectonic settings. However, the estimation of mantle f_O2 is limited by challenges in measuring Fe³⁺ in minerals like clinopyroxene (Cpx) due to analytical constraints and inconsistencies between oxybarometer methods. Here, we applied machine learning (ML) to predict Cpx Fe³⁺ content and equilibrium pressure-temperature and f_O2 conditions. We employed a nested cross-validation approach to minimize coincidental perfomance biases. Our models outperformed previous ferric iron and thermobarometer models on both metrics and generalization. The ML-based oxybarometer shows adequate generalization with R² peaking at 0.74, average MAE is 0.95, and average RMSE is 1.45. We compiled an application dataset comprising of 9,832 global mantle xenolith samples. For the subset with Fe³⁺ measurements (n≈600), ML-predicted f_O2 closely matches thermodynamic estimates, supporting the robustness and global applicability of our approach. Applying the model to the rest samples lacking Fe³⁺ analyses expands geographic coverage to data-sparse provinces (e.g., South America, India, and Eastern Europe), and reveals coherent global redox gradients. Xenoliths from cratonic mantle domains show no temporal f_O2 trends since Mesoproterozoic. Comparative analysis across cratonic mantle xenoliths, abyssal peridotites, and oceanic intraplate xenoliths indicates that mantle residues are initially oxidized by short-term metasomatism, but eventually equilibrate to stable redox conditions through interactions with neutral or reducing agents.