Designing new compounds or alloys whose surfaces can be used as catalysts in chemical reactions is a complex process that relies heavily on the intuition of experienced chemists. However, a team of researchers at MIT has found a new way to make this process simpler and more detailed by using machine learning.
The findings, published in the journal Nature Computational Science, show how the researchers used the new method to identify atomic configurations of a compound’s surface that had not previously been identified. This method can also provide dynamic information about how the surface properties change over time under operating conditions, a feature that present methods do not have.
The new method, called Automatic Surface Reconstruction framework, uses active learning combined with a type of Monte-Carlo algorithm to select sites on a surface to sample, which allows the researchers to get accurate predictions of surface energies across various chemical or electrical potentials, with just a few first-principles calculations.
This new approach allows researchers to investigate the stability of different surface structures under different external conditions, which was not possible using traditional methods. Moreover, it can be used to develop new materials for catalysts, production of green hydrogen, and for new battery or fuel cell components. This tool, called AutoSurfRecon, has also been made freely available by the researchers for use by other researchers around the world.
With the new method, researchers can now explore more possibilities, even when intuition might be wrong. The team hopes that their code will inspire quick improvements by other users and lead to further advancements in the field.
The study, supported by the U.S. Air Force, the U.S. Department of Defense, and the U.S. National Science Foundation, has shown promising results and could change the landscape of compound and alloy surface designing in the future.