How the metal gurus hunt down a change of catalyst
Image credit: Getty Images
There’s more to platinum than just Jubilees! It has multiple and critical uses as a catalyst – from oil refining to the hydrogen economy. But it is rare and expensive. Can we use better analysis to find cheaper replacements?
At the end of 2016 a crash in commodities prices saw one class of crime become a lot less profitable. Along with oil and common metals like copper and nickel, the price of the far more expensive industrially important metals palladium and platinum slumped. In the case of platinum, the spot price plummeted by half as the world seemingly decided it did not need most of these materials. And fewer criminals were keen on jacking up cars parked out of sight and unbolting or sawing off catalytic converters that each contain around 5 grams of platinum and palladium.
According to figures obtained by London Liberal Democrats using a freedom-of-information request on Metropolitan Police records, there were just 173 thefts of the exhaust units in the city in 2017. However, by 2020, that number had surged past 14,000. It should come as no surprise that the price of both palladium and platinum also surged in the intervening years.
As the renewable-energy revolution kicks in, the value and price of platinum could move even higher, together with thefts. Why? Platinum-group metals often work better as chemical catalysts than other materials. Not only do palladium and platinum play a key role in breaking down nitric oxide (NO) and other pollutants in car exhausts, but these metals tend to deliver the best results across a broad range of chemical reactions. From margarine manufacturing to petrol production, you can often find platinum and other rare neighbours in the periodic table such as iridium, rhenium and ruthenium.
Even in the clean-energy revolution, platinum-group metals feature heavily. For the reaction that generates molecular hydrogen, platinum itself is currently the basis of the leading catalysts. For the vital partner reaction that allows water splitting to take place, promoting the formation of oxygen, the two best elements are iridium and ruthenium.
For such rare elements to play a pivotal role in chemical reactions that will underpin the transition to a world that has largely abandoned fossil fuels, calls into question how practical that transition can be. Being catalysts, the metals are not used up in the chemical reactions they enable, though they may be poisoned by contaminants. And they are not needed in enormous quantities. A 2015 study by Jakob Kibsgaard and Professor Ib Chorkendorff at the Technical University of Denmark found that for the hydrogen-evolution reaction used in water splitting, 30 per cent of the world’s production of platinum from a single year would make enough hydrogen to sustain 1TW, about 5 per cent of global power-generation capacity.
Though that may be manageable, that is just one reaction. And competition for the supplies that do exist will drive prices. Even today, recycling, which is helping to drive the catalytic-converter thefts, accounts for a quarter of the platinum and palladium supply, because mining simply cannot deliver enough of these metals. The situation is a lot worse for the oxygen-evolution reaction needed for full electrolysis. The Danish study estimated it would take 40 years to mine enough iridium to support the same 1TW production rate.
The search is on for substitute materials that can do the jobs of the platinum-group metals but which use far more abundant and readily available raw materials: metals such as iron and nickel. Geopolitics issues also play into the arithmetic. Even with its extensive natural resources, the US needs to import all of its iridium. Over a century ago, BASF rushed to find a vanadium-based replacement for the catalyst needed to make sulphuric acid because countries refused to ship platinum to Germany in World War One.
Finding cheaper substitutes is not an easy task. Catalysis is all about surfaces and how those surfaces change when molecules adsorb onto them. Getting the surface layout just right is the difference between an effective catalyst and a mesh that simply gets dirty. The surface characteristics are not only about how atoms are spaced out in the crystals but how the electron clouds warp as other atoms approach.
You need look no further than the iron in the haemoglobin that pervades our red blood cells to see how changes in those electron clouds make an enormous difference to chemical behaviour. Like many metals, iron in its native state oxidises readily and it takes a lot of energy to turn that rust back into shiny atomic iron. The cage of carbon-nitrogen chains that surrounds the iron atom at the centre of a heme molecule tames iron’s hunger for oxygen to the point where only a slight change in conditions causes it to either attach oxygen or carbon dioxide molecules. With no oxygen present, the bonds between the iron atom and the nitrogen cage keep the electrons in its most reactive orbitals in a state that makes the atom slightly too big to fit neatly into the same plane as the rest of the molecule.
When it attaches to an oxygen molecule, the newly bound electrons move into a low-spin state that makes the electron cloud around the iron atom slightly smaller, letting it recede into the cavity. This difference in spin states does not occur in free iron ions; it’s a consequence of electrons being donated by the nitrogen atoms in heme ligands into empty orbitals in the iron atom’s outer electron shell. The result is a distinct change in the element’s normal behaviour. Similarly, the donated electrons help reduce the tendency of iron to hold onto the oxygen when the time comes to give it up at its destination.
The platinum-group metals do not need this complex cage of ligands to keep them from bonding permanently with reactive molecules like oxygen; they sit naturally in the Goldilocks zone for many of these reactions.
It’s hoped that with better understanding of these subtle changes in how electrons behave depending on the atoms around their host ion, it will become possible to substitute other elements for the rare platinum-group elements.
The traditional tool was trial and error. That is an expensive, time-consuming process, as University of Karlsruhe professor Fritz Haber and Carl Bosch of chemical company BASF discovered when trying to find a base-metal replacement for platinum in making ammonia, trying out the even rarer osmium before alighting on iron. In his speech accepting the 1918 Nobel Prize in Chemistry, Bosch recalled how many glass tubes used to contain the trial catalyst broke: “If we had filled them with osmium instead of the new catalyst the entire world stock of this precious metal, which we had by now bought, would have disappeared.”
A key question is how to focus the effort on the most promising combinations of elements. Because biology offers what seems to be a successful template, one possibility is to mimic its enzymes. As well as being responsible for transporting oxygen to cells, heme turns up in other enzymes such as cytochrome P450. This catalyses numerous reactions and could be used as a way of replacing platinum in hydrogen-powered fuel cells.
A group led by Magda Titirici, professor of materials chemistry at Imperial College London, is looking at two ways of harnessing molecules like heme for catalysis. One method is to find simpler analogues of these nitrogen-bound iron enzymes that can be made using regular synthetic chemistry. Another is to harvest the heme itself from biological sources and incorporate it into fuel cells. “They both have their advantages and disadvantages, but they are both very powerful tools towards the design of single-site iron-nitrogen-carbon catalysts,” says Titirici.
Harvesting delivers complex catalysts more easily, but the yields are likely to be low as it is easy to damage the molecules, and the resulting catalyst will only work for the same kinds of reactions as those found in biology. Artificial analogues are more versatile, but it is hard to synthesise compounds that will deliver precisely the right active site and avoid producing other iron-based species that take part in unwanted side reactions.
The other main option is to try to make base metals behave more like platinum by combining them in alloys that do not have the complex structure of bio-inspired catalysts. Alloys with three or more base metals provide many different types of crystal structure that may push common metals like iron and nickel into the platinum group’s Goldilocks zone: reining in the reactivity of the active sites and introducing the right stresses to promote reactions in target molecules. These materials are also more amenable to the analysis tools that chemists have to hand today. Though it is easier to find candidates, trial and error remains an important part of the process. It just happens to have shifted into the virtual domain, with chemists building digital twins of potential catalysts and target molecules.
Titirici says understanding the mechanism of catalysis is complex. “It will depend on the reaction involved and the intermediates of that reaction and their binding energy to the active sites.”
An initially promising catalyst can become roadblocked by a high-energy step even if the end result is a more stable molecule, or it might be vulnerable to side reactions that produce unwanted contaminants. A decade ago, in research into fuel cells powered by borohydride, a team from Osaka University using computer simulation discovered a side effect of using platinum as a catalyst was that it broke the molecule apart completely once it had been adsorbed onto the most active site. That led to the highly reactive ions quickly forming side products. Gold turned out to be far gentler. It keeps the borohydride more or less intact though with key bonds stretched, which make them more likely to take part in the reaction that delivers the wanted end product.
“Reaction kinetics are very important and necessary for more quantitative agreement with experiments. We, and many in the community, are thinking about how to predict these better,” says Zachary Ulissi, assistant professor of chemical engineering at Carnegie-Mellon University.
There are, however, ways to pick candidate materials that are likely to work. The main target of this computer-based work is the phenomenon of relaxation. This is the process where, after a molecule lands on the surface of a catalyst, it will tend to move around until it settles into the minimum energy state at that site.
In principle, the task of finding this minimum binding energy for a given landing site is simple enough: calculate it using Schrödinger’s equation. Unfortunately, the only system the equation can solve directly is for the single electron of a lone hydrogen atom. Anything else is far more complicated and takes huge amounts of compute time.
At Nvidia’s Spring GTC event, Erich Wimmer, chief scientific officer of materials design, described the problem in a talk on how hardware accelerators can help materials science and chemistry. He took the example of a platinum nanoparticle of a hundred atoms interacting with a single carbon monoxide molecule. The energy from the nuclear and electron interactions totals around 50 million electron volts (eV).
“The energy difference that we’re interested in is in the order of 0.1eV. This is like trying to measure the weight of a captain by weighing a ship with and without the captain. The total energy is not the key bottleneck, we can handle that. The real problem is the many-body problem,” Wimmer adds. The system has more than 7,800 electrons. “That gives 5,000 interactions between the nuclei, about 800,000 interactions between the electrons and the nuclei and 30 million electron-electron interactions.”
Luckily, there are useful approximations. Though it does not deal well with some orbital states, the one that chemists employ in catalyst research is density functional theory (DFT). The nuclear interactions can be treated in bulk and further simplifications make it possible to consider single electrons at a time interacting with a field of charges. Despite its far greater simplicity compared to more accurate methods, DFT still takes hours to compute on today’s hardware in catalytic scenarios like these.
This would not be so onerous were it not for an individual DFT calculation only being a small part of the development process. Each estimate is just for a single point in time and for a single site. But the chemists do not know in advance where the best location is. It can take more than 20 variations of surface position to get a good idea of the best candidate for an active site for each reactant that will need to land on the catalyst.
DFT usually works as an initial filter for screening potential candidates, Ulissi says. “We’re quite fortunate that binding energies tend to give a reasonable understanding of catalyst limitations even though they represent a significant simplification.”
However, even with the approximations that DFT uses, performing those iterations quickly adds up to weeks or even months of computer time for each promising candidate. That is why Ulissi and others are looking at using machine learning for a faster approximation of DFT and similar quantum-behaviour calculations.
The CMU team created with Facebook AI Research a dataset built from extensive DFT simulations with various combinations of materials intended as training fodder for machine-learning models. To encourage work into innovative solutions they also launched a competition that had its first outing at the NeurIPS conference in late 2021. Microsoft Asia won this round with a model that borrows techniques from the neural networks that power language-processing systems. It was able to predict relaxation energies that correlate reasonably well with reality in more than 80 per cent of cases. However, the deviations are still large enough to not yet challenge DFT.
“For the direct energy prediction task that was the focus of the NeurIPS challenge last year, I think the catalysis community would like to see mean absolute errors in the 0.1 to 0.2eV range, which is quite a bit lower than the current best models, and suggests more work is needed,” Ulissi says.
The hope is that improvements in models, helped along by contests such as the Open Catalyst Challenge, will deliver viable answers for relaxations at single sites on the catalyst surface in a matter of seconds. DFT simulations would still need to be run to confirm choices before moving to physical experiments, but the models could slash the time it takes to screen candidates. That in turn could expand the range of materials chemists explore beyond those for which they already have a good hunch. Ultimately, scientists such as Ulissi hope that machine learning trained on DFT and other results could work directly on reaction kinetics to screen catalyst candidates far more quickly.
In the meantime, alongside improvements in DFT simulation, catalyst chemists are working with an expanding array of experimental tools that are delivering better insights into how molecules change during reactions. Titirici points to techniques such as the use of X-rays to analyse the inner electron clouds of adsorbed atoms, more conventional infrared spectroscopy and advanced nanoscale microscopy instruments, such as atomic-force probes to gain better insights into these processes as they proceed. But it will take time to evaluate the base-metal alternatives to platinum-group metals. Until then, keep an eye on your exhaust pipes.
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.