Designing latest compounds or alloys whose surfaces may be used as catalysts in chemical reactions generally is a complex process relying heavily on the intuition of experienced chemists. A team of researchers at MIT has devised a brand new approach using machine learning that removes the necessity for intuition and provides more detailed information than conventional methods can practically achieve.
For instance, applying the brand new system to a cloth that has already been studied for 30 years by conventional means, the team found the compound’s surface could form two latest atomic configurations that had not previously been identified, and that one other configuration seen in previous works is probably going unstable.
The findings are described this week within the journal Nature Computational Science, in a paper by MIT graduate student Xiaochen Du, professors Rafael Gómez-Bombarelli and Bilge Yildiz, MIT Lincoln Laboratory technical staff member Lin Li, and three others.
Surfaces of materials often interact with their surroundings in ways in which depend upon the precise configuration of atoms on the surface, which may differ depending on which parts of the fabric’s atomic structure are exposed. Consider a layer cake with raisins and nuts in it: Depending on exactly how you chop the cake, different amounts and arrangements of the layers and fruits might be exposed on the sting of your slice. The environment matters as well. The cake’s surface will look different whether it is soaked in syrup, making it moist and sticky, or whether it is put within the oven, crisping and darkening the surface. That is akin to how materials’ surfaces respond when immersed in a liquid or exposed to various temperatures.
Methods normally used to characterize material surfaces are static, a specific configuration out of the tens of millions of possibilities. The brand new method allows an estimate of all of the variations, based on just a number of first-principles calculations routinely chosen by an iterative machine-learning process, in an effort to find those materials with the specified properties.
As well as, unlike typical present methods, the brand new system may be prolonged to offer dynamic details about how the surface properties change over time under operating conditions, for instance while a catalyst is actively promoting a chemical response, or while a battery electrode is charging or discharging.
The researchers’ method, which they call an Automatic Surface Reconstruction framework, avoids the necessity to use hand-picked examples of surfaces to coach the neural network utilized in the simulation. As an alternative, it starts with a single example of a pristine cut surface, then uses energetic learning combined with a sort of Monte-Carlo algorithm to pick out sites to sample on that surface, evaluating the outcomes of every example site to guide the collection of the subsequent sites. Using fewer than 5,000 first-principles calculations, out of the tens of millions of possible chemical compositions and configurations, the system can obtain accurate predictions of the surface energies across various chemical or electrical potentials, the team reports.
“We’re thermodynamics,” Du says, “which suggests that, under different sorts of external conditions corresponding to pressure, temperature, and chemical potential, which may be related to the concentration of a certain element, [we can investigate] what’s essentially the most stable structure for the surface?”
In principle, determining the thermodynamic properties of a cloth’s surface requires knowing the surface energies across a selected single atomic arrangement after which determining those energies tens of millions of times to encompass all of the possible variations and to capture the dynamics of the processes going down. While it is feasible in theory to do that computationally, “it’s just not reasonably priced” at a typical laboratory scale, Gómez-Bombarelli says. Researchers have been in a position to get good results by examining just a number of specific cases, but this isn’t enough cases to offer a real statistical picture of the dynamic properties involved, he says.
Using their method, Du says, “we’ve latest features that allow us to sample the thermodynamics of various compositions and configurations. We also show that we’re in a position to achieve these at a lower cost, with fewer expensive quantum mechanical energy evaluations. And we’re also in a position to do that for harder materials,” including three-component materials.
“What’s traditionally done in the sphere,” he says, “is researchers, based on their intuition and knowledge, will test only a number of guess surfaces. But we do comprehensive sampling, and it’s done routinely.” He says that “we have transformed a process that was once not possible or extremely difficult resulting from the necessity for human intuition. Now, we require minimal human input. We simply provide the pristine surface, and our tool handles the remaining.”
That tool, or set of computer algorithms, called AutoSurfRecon, has been made freely available by the researchers so it could actually be downloaded and utilized by any researchers on the earth to assist, for instance, in developing latest materials for catalysts, corresponding to for the production of “green” hydrogen in its place emissions-free fuel, or for brand spanking new battery or fuel cell components.
For instance, Gómez-Bombarelli says, in developing catalysts for hydrogen production, “a part of the issue is that it’s not likely understood how their surface is different from their bulk because the catalytic cycle occurs. So, there’s this disconnect between what the fabric looks like when it’s getting used and what it looks like when it’s being prepared before it gets put into motion.”
He adds that “at the tip of the day, in catalysis, the entity chargeable for the catalyst doing something is a number of atoms exposed on the surface, so it really matters so much what precisely the surface looks like in the meanwhile.”
One other potential application is in studying the dynamics of chemical reactions used to remove carbon dioxide from the air or from power plant emissions. These reactions often work by utilizing a cloth that acts as a type of sponge for absorbing oxygen, so it strips oxygen atoms from the carbon dioxide molecules, abandoning carbon monoxide, which generally is a useful fuel or chemical feedstock. Developing such materials “requires understanding of what the surface does with the oxygens, and the way it’s structured,” Gómez-Bombarelli says.
Using their tool, the researchers studied the surface atomic arrangement of the perovskite material strontium titanium oxide, or SrTiO3, which had already been analyzed by others using conventional methods for greater than three a long time yet was still not fully understood. They found two latest arrangements of the atoms at its surface that had not been previously reported, and so they predict that one arrangement that had been reported is the truth is unlikely to occur in any respect.
“This highlights that the tactic works without intuitions,” Gómez-Bombarelli says. “And that’s good because sometimes intuition is mistaken, and what people have thought was the case seems to not be.” This latest tool, he said, will allow researchers to be more exploratory, trying out a broader range of possibilities.
Now that their code has been released to the community at large, he says, “we hope that it should be inspiration for very quick improvements” by other users.
The team included James Damewood, a PhD student at MIT, Jaclyn Lunger PhD ’23, who’s now at Flagship Pioneering, and Reisel Millan, a former postdoc who’s now with the Institute of Chemical Technology in Spain. The work was supported by the U.S. Air Force, the U.S. Department of Defense, and the U.S. National Science Foundation.
