Machine learning links material composition and performance in catalysts – sciencedaily
In a discovery that could help pave the way for cleaner fuels and a more sustainable chemical industry, researchers at the University of Michigan used machine learning to predict how the compositions of metal alloys and metal oxides affect their electronic structures.
The electronic structure is essential to understand how the material will behave as a mediator or catalyst of chemical reactions.
“We are learning to identify fingerprints in materials and relate them to the performance of the material,” said Bryan Goldsmith, assistant professor of chemical engineering at Dow Corning.
A better ability to predict which compositions of metals and metal oxides are best to guide which reactions could improve large-scale chemical processes such as hydrogen production, production of other fuels and fertilizers, and product manufacturing household chemicals such as dish soap.
“The objective of our research is to develop predictive models that will relate the geometry of a catalyst to its performance. Such models are essential for the design of new catalysts for critical chemical transformations,” said Suljo Linic, Professor Martin Lewis Perl Collegiate of Chemical Engineer.
One of the main approaches to predict how a material will behave as a potential mediator of a chemical reaction is to analyze its electronic structure, in particular the density of states. This describes the number of quantum states available to electrons in reacting molecules and the energies of those states.
Usually the electron density of states is described with summary statistics – an average energy or a bias that reveals whether more electronic states are above or below the average, and so on.
“That’s okay, but these are just simple statistics. You might be missing something. With principal component analysis, you just take everything and find out what’s important. You don’t just throw away information.” said Goldsmith.
Principal component analysis is a classic method of machine learning, taught in Introductory Data Science courses. They used the electron density of states as input for the model, because the density of states is a good predictor of how the surface of a catalyst will adsorb or bind with the atoms and molecules that serve as reactants. . The model relates the density of states to the composition of the material.
Unlike conventional machine learning, which is essentially a black box that inputs data and offers predictions in return, the team created an algorithm they could understand.
“We can systematically see what changes in the density of states and correlate that with the geometric properties of the material,” said Jacques Esterhuizen, doctoral student in chemical engineering and first author of the article in Chem Catalysis.
This information helps chemical engineers design metal alloys to achieve the density of states they want to mediate a chemical reaction. The model accurately reflects the correlations already observed between the composition of a material and its density of states, as well as potential new trends to explore.
The model simplifies the density of states into two parts, or principal components. One piece basically covers how the atoms of the metal fit together. In a layered metal alloy, this includes whether the subterranean metal separates or constricts the surface atoms, and the number of electrons that the subterranean metal contributes to bond. The other element is only the number of electrons that the surface metal atoms can contribute to the bond. From these two principal components, they can reconstitute the density of states in the material.
This concept also works for the reactivity of metal oxides. In this case, the concern is the ability of oxygen to interact with atoms and molecules, which is related to the stability of surface oxygen. Stable surface oxygenes are less likely to react, while unstable surface oxygenes are more reactive. The model accurately captured the stability of oxygen in metal oxides and perovskites, a class of metal oxides.
The study was supported by the Department of Energy and the University of Michigan.