Can published PEM-electrolyzer data guide the design of lower-iridium anodes?

The electrochemical problem

Proton-exchange-membrane water electrolysis can operate at high current density and produce high-purity hydrogen, but its acidic oxygen-evolution environment places severe stability demands on the anode. Iridium-based oxides remain important for this reason, yet iridium is scarce and expensive.

The paper asks whether the accumulated PEM-electrolysis literature contains enough structured information to guide a reduction in iridium content rather than relying only on one-by-one experimental intuition.

What the researchers did

The authors systematically collected literature data on PEM electrolysis, including catalyst composition and experimental conditions. They used compositional and experimental descriptors in supervised machine-learning models of cell performance, with the practical objective of identifying elements that could be introduced while reducing the fraction of Ir in the anode.

Predicted candidates were not left as a virtual screening result. The team prepared and tested Ir0.9M0.1O2 compositions with M = Cr, Zn, Sb, and Ho.

What they found

The experimental validation showed that introducing the predicted elements could preserve cell performance while reducing the Ir content of the anode. The study therefore demonstrates a complete workflow in which heterogeneous literature data are converted into a material-design hypothesis and then checked in an electrochemical device.

The contribution is less about declaring a single universal best dopant than about showing that a literature-derived model can narrow a catalyst-design problem enough to support useful experiments.

Why this matters

Electrochemistry has a large but fragmented experimental literature. If those data can be organized with sufficient attention to test conditions, machine learning can turn previous experiments into a design resource rather than a passive archive.

For resource-constrained catalysts, that is especially attractive: the optimization objective can explicitly include composition choices that reduce dependence on scarce elements while preserving device performance.

One important caution

Literature-derived datasets mix experiments performed with different cells, loadings, membranes, protocols, and reporting practices. Machine learning can identify correlations within that assembled dataset, but those correlations do not automatically become causal design rules.

The value of this work is strengthened by experimental validation, yet the tested compositions still represent a small subset of the possible PEM-anode design space.