October_2021_AMP_Digital

A D V A N C E D M A T E R I A L S & P R O C E S S E S | O C T O B E R 2 0 2 1 6 MACHINE LEARNING | AI MACHINE LEARNING IMPROVES CATALYSTS In a new study, researchers at the University of Michigan used machine learning to predict how the composi- tions of metal alloys and metal oxides affect their electronic structures. “We’re learning to identify the fingerprints of materials and connect them with the material’s performance,” says chemical engineering professor Bryan Goldsmith. A better ability to predict which compo- sitions are best for guiding certain reac- tions could improve large-scale chem- ical processes from hydrogen produc- tion to chemical manufacturing. One of the main approaches to predicting how a material will behave as a potential mediator of a chemi- cal reaction is to analyze its electron- ic structure, specifically the density of states. Usually, the electronic density of states is described with summary statis- tics—an average energy or a skew that reveals whether more electronic states are above or below the average. “That’s OK, but those are just simple statistics. With principal component analysis, you just take in every- thing and find what’s important,” explains Goldsmith. Principal compo- nent analysis is a classic machine learn- ing method. The team used the electronic density of states as in- put for the model be- cause it is a good predictor for how a catalyst’s surface will adsorb atoms and molecules that serve as reactants. The model links the density of states with the composition of the material— accurately reflecting correlations al- ready observed, as well as turning up new trends. The team’s model simplifies the density of states into two principal com- ponents: One covers how the atoms of the metal fit together; the other cov- ers the number of electrons the surface metal atoms can contribute to bonding. From these two variables, the model can reconstruct the density of states in the material. umich.edu . SIMPLIFIED MODELING SPEEDS SIMULATIONS Although computer simulations can help ex- plore light-matter inter- actions, modeling ma- terials that feature mul- tiple types of structures is a complex job. Now, a research team at the DOE’s Argonne National Laboratory has found a way to streamline this task. Us- ing a data-driven approach based on machine learning, the team was able to simplify the solution of the quantum mechanical equations that describe how light is absorbed by a solid, liquid, or molecule. The key insight was recognizing that not all terms of the quantum me- chanical equations need to be com- puted in the same way. Some could be calculated from simpler quantities, re- markably speeding up the overall sim- ulation. These protocols can lead to big savings when it comes to simula- tions that may take hours or even days on high-performance computers. The new technique allows simulations of absorption spectra of complex systems to run between 10 and 200 times fast- er than previous methods. Research- ers are now looking at applying these shortcuts and recycling protocols to electronic structure problems not only related to light absorption, but also to light manipulation for quantum sensing applications. anl.gov. From left, diagrams show an oxygen atom bonding with a metal, metal oxide, and perovskite. A newmachine learning model could help engineers design these catalysts for more sustainable fuel production, among other uses. Cornell University, Ithaca, N.Y., is partnering in a $36 million grant from the Toyota Research Institute, Los Altos, Calif., for its Accelerated Materials Design and Discovery collaborative university research program, which seeks to use artificial intelligence to discover new materials that could help achieve emissions-free driving. cornell.edu . BRIEF Machine learning can circumvent explicit calculation of certain material behavior to accelerate simulations of optical properties of complex materials at finite temperature.