Scientific Challenge

Catalytic screening reports one number per composition: an adsorption free energy, compared against the thermoneutral ideal. For an ordered surface that number is meaningful. For the doped, high-entropy, compositionally complex catalysts that the field is increasingly excited about, it is the wrong object. In a disordered catalyst no two sites share a local environment, so the adsorption energy is not a value but a distribution — and activity is set by where that distribution sits and how wide it is, not by any single representative site. A composition reported as thermoneutral on average may have almost no sites actually near the optimum; a composition that looks unpromising on average may have a usable tail. Our position is that screening one number per composition measures the wrong thing, and that the statistics of disorder are what decide catalytic performance.

Our Approach

The groundwork is a clean-descriptor analysis: on well-defined surfaces, establishing which electronic-structure quantity actually controls adsorption, so that compositional trends read as mechanism rather than correlation, and so that the point where a simple local descriptor stops predicting can be located exactly. That boundary is not a nuisance — it is the entry point, because it is precisely where configurational disorder takes over. From there the object of study becomes the distribution itself: special quasirandom structures and ensemble calculations across many inequivalent sites turn the single adsorption energy of the ordered picture into the spread a real catalyst presents, and the question becomes which compositions place enough of that spread on the thermoneutral window. Disorder, in this framing, is the design parameter, not the noise to be averaged out of it.

Findings

We can already resolve catalysis at the level the statistical programme demands — site by site, and to the point where the clean description breaks:

• Site-resolved descriptors, not sublattice averages. In bimetallic Janus MXenes, factorial screening across metal pairs and anions shows that site-resolved d-band asymmetry — not any averaged descriptor — controls hydrogen adsorption near thermoneutrality, and it pinpoints the regime, magnetic exchange splitting on one sublattice, where the single local descriptor ceases to predict. Locating that breakdown is exactly the capability the disorder problem requires.
• Mechanistic routes to noble-metal-free catalysis. In layered transition-metal carbides and nitrides, targeted electronic-structure engineering through early-transition-metal doping brings hydrogen adsorption into the thermoneutral window without platinum-group metals — establishing that the controlling descriptor can be moved deliberately, by design, rather than found by chance.

These results show that we can isolate the descriptor that governs adsorption and identify where a single-site picture fails — the two things a statistical treatment of disorder must be built on.

Outlook

The next step is the one the field has largely not taken: treating the adsorption energy of a high-entropy or heavily doped catalyst as a distribution to be predicted, and asking which compositions centre that distribution on the catalytic optimum. The aim is to design a disordered catalyst by the shape of its site statistics rather than by a misleading average — to say not “this composition is thermoneutral” but “this composition places the most sites where catalysis happens.” The descriptor-resolution above is what makes that achievable; the open work is to carry it from the clean surface to the full configurational ensemble, where disorder stops being a complication and becomes the variable being engineered.

• ← Go back to Research