Since the early days of catalytic converters, chemists have tried to pin down what platinum catalysts are actually doing while gases swarm over them, and many noble attempts have failed because atoms are rude: they move during the experiment, especially when you look away.

That is the setup for Franklin Tao, David Jiang, Luan Nguyen, and Philippe Sautet’s new JACS paper, Structural Evolution of Pt Nanoclusters Driven by CO Reactant Pressure and Catalyst Temperature PMID: 42283773, DOI: 10.1021/jacs.6c02758. The team studied platinum, carbon monoxide, temperature, and pressure - four words that sound like a grant proposal was trapped in an elevator.

But the result is surprisingly vivid: platinum nanoclusters do not just sit there like tiny metallic furniture. Under carbon monoxide, they shrink, merge, reorganize, and generally behave like a committee five minutes before a conference deadline.

The Catalyst Is Not a Statue

A catalyst helps reactions happen without being consumed, which is technically true in the same way “graduate school builds character” is technically a sentence. In real conditions, catalyst surfaces can restructure. Atoms shift. Small clusters form. Bigger particles break apart or fuse. And because catalytic performance depends on which atoms are exposed, these movements matter.

In this study, the authors looked at platinum nanoclusters that form when a hexagonal reconstruction of Pt(100) restructures under carbon monoxide. The key trick was watching the surface under gas pressure instead of pretending the real world happens in perfect vacuum. They used high-pressure scanning tunneling microscopy, which images surfaces atom by atom, and ambient-pressure X-ray photoelectron spectroscopy, which tracks chemical states. Basically: one tool takes the atomic mugshot, the other checks the fingerprints.

Then they brought in machine learning-accelerated computation. Not ChatGPT wearing a lab coat, sadly. They used a neural network potential paired with basin-hopping simulations to search possible structures and CO arrangements much faster than brute-force quantum calculations would allow. Think of it as giving the overworked computational intern a map and a scooter.

CO: Poison, Sculptor, Tiny Drama Agent

Carbon monoxide has a reputation as a catalyst poison because it binds strongly to metals like platinum and can block useful reaction sites. But here, CO is more than a seat hog. It helps sculpt the surface.

At very low CO pressure, around (2 \times 10^{-8}) Torr, platinum nanoclusters averaged about 3.2 nm across. As CO pressure rose to 1 Torr, the average size dropped to about 2.3 nm. That means the clusters broke down as more CO arrived.

Then the plot changed. From 1 Torr to 750 Torr at room temperature, cluster density fell by 4-5 times while average size grew to about 4.6 nm. In plain English: once there was enough CO around, the clusters started coalescing into fewer, larger islands.

If this sounds contradictory, welcome to surface chemistry, where the answer to “does CO break things apart or glue them together?” is “yes, please see Supporting Information.”

Why the Atoms Rearrange

The simulations suggest a tug-of-war. Making a small platinum nanocluster costs energy because surface atoms are undercoordinated and exposed. But CO adsorption gives energy back because CO binds strongly to platinum. Smaller clusters expose more edge sites, which CO likes, but CO molecules packed too tightly repel one another, like academics sharing one poster-board printer before NeurIPS.

The clever bit is that CO molecules at cluster edges can tilt outward, relieving some CO-CO repulsion. The paper reports roughly one CO molecule per Pt atom on the nanoclusters, which is a crowded little dance floor.

At low pressure, breaking larger clusters into smaller ones gives CO more favorable binding opportunities. At higher pressure, the balance shifts, and cluster coalescence wins. Temperature matters too: clusters formed at 25 °C in 750 Torr CO needed annealing to 100-130 °C to reach equilibrium size. Room temperature leaves them kinetically stuck, which is chemistry’s version of “I’ll fix it after the deadline.”

Why Machine Learning Helps Here

Catalyst modeling has an annoying problem: the structures you care about are messy, dynamic, and numerous. Traditional density functional theory can be accurate, but using it to search many nanocluster shapes and gas coverages is like proofreading every tweet ever written before choosing lunch.

Neural network potentials learn an energy model from expensive calculations, then evaluate many candidate structures quickly. Recent reviews on machine learning interatomic potentials in heterogeneous catalysis argue that this is exactly where ML earns its keep: not by replacing chemistry judgment, but by making realistic atomistic searches less computationally absurd DOI: 10.1002/chem.202401148.

Here, ML helps connect pressure, CO coverage, cluster size, and structure. That matters because experiments can show what happened, while computation helps explain why the atoms chose that particular bit of chaos.

The Bigger Lesson: Catalysts Have Working Personalities

This paper fits a broader trend: catalysts under real reaction conditions can look very different from the tidy structures prepared beforehand. Recent work on Pt catalysts in zeolites showed that gas atmosphere and temperature can shift platinum among single atoms, clusters, and nanoparticles DOI: 10.1021/jacsau.3c00732. A 2024 Brookhaven report on Pt/ceria also highlighted reversible assembly and fragmentation driven partly by carbon monoxide, which is either elegant chemistry or the world’s smallest office reorganization.

If these findings generalize, they could help engineers design catalysts that adapt instead of degrade. Better catalytic converters, cleaner chemical manufacturing, improved fuel processing, and longer-lasting industrial catalysts all depend on knowing which atomic structures exist during operation, not just before the reaction starts and everyone is still being polite.

The caution flag: this is still carefully controlled surface science on a model platinum system. Real industrial catalysts have supports, defects, contaminants, heat gradients, and all the charming messiness that keeps Reviewer 2 gainfully employed. But the message is strong: pressure and temperature are not background settings. They are active design knobs.

Platinum nanoclusters are not passive particles. Under CO, they are responsive, shape-shifting catalytic actors. Tiny, expensive actors, yes. But at least now we have better footage.

References

1. Tao F, Jiang D, Nguyen L, Sautet P. Structural Evolution of Pt Nanoclusters Driven by CO Reactant Pressure and Catalyst Temperature. Journal of the American Chemical Society, 2026. DOI: 10.1021/jacs.6c02758, PMID: 42283773.

2. Li X, Cheng J, Hou H, Meira DM, Liu L. Reactant-Induced Structural Evolution of Pt Catalysts Confined in Zeolite. JACS Au, 2024. DOI: 10.1021/jacsau.3c00732.

3. Owen CJ, Marcella N, Xie Y, Vandermause J, Frenkel AI, Nuzzo RG, Kozinsky B. Unraveling the Catalytic Effect of Hydrogen Adsorption on Pt Nanoparticle Shape-Change. 2023. arXiv:2306.00901.

4. Omranpour A, Elsner BA. Machine Learning Interatomic Potentials for Heterogeneous Catalysis. Chemistry Europe, 2024. DOI: 10.1002/chem.202401148.

5. Brookhaven National Laboratory. Study Reveals Reversible Assembly of Platinum Catalyst. 2024. BNL Newsroom.

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.