Researchers at the Massachusetts Institute of Technology (MIT) have developed an innovative machine learning-based method – AI that promises to transform the intricate process of designing materials with catalytic surfaces. Compared to conventional methods, this innovative methodology offers more detailed insights because it does away with the need for intuition.
When this novel approach was used on a material that had been conventionally investigated for thirty years, two previously unknown atomic configurations on its surface were discovered. Furthermore, it was determined that one known configuration was probably unstable. These ground-breaking results are described in a work by a team led by MIT graduate student Xiaochen Du, together with professors Rafael Gómez-Bombarelli and Bilge Yildiz, Lin Li from MIT Lincoln Laboratory, and others, published in the journal Nature Computational Science.
Understanding Material Surfaces
Material surfaces interact with their surroundings based on the atomic configuration, similar to a layer cake’s exposure of different layers and fruits depending on how it’s cut. To locate materials with the needed properties, the researchers introduce a dynamic technique that estimates variations through machine learning, thereby addressing the limits of static characterization approaches.
Using their method, Du says, “We have new features that allow us to sample the thermodynamics of different compositions and configurations. We also show that we can achieve these at a lower cost, with fewer expensive quantum mechanical energy evaluations. And we are also able to do this for harder materials,” including three-component materials.
The team’s method, named Automatic Surface Reconstruction framework, uses machine learning without relying on hand-picked examples to train the neural network. It begins with a pristine cut surface and employs active learning and Monte Carlo algorithms to iteratively sample sites, enabling accurate predictions of surface energies across various conditions with minimal computational cost.
AutoSurfRecon: A Transformative Tool
Unlike existing methods, this system provides dynamic information on how surface properties change under operating conditions, crucial for applications like catalysts promoting chemical reactions or battery electrodes during charging or discharging.
“We are looking at thermodynamics,” Du says, “which means that, under different kinds of external conditions such as pressure, temperature, and chemical potential, which can be related to the concentration of a certain element, [we can investigate] what is the most stable structure for the surface?”
The materials with complicated compositions can now be explored in new ways thanks to the efforts of the researchers. To help researchers worldwide create novel battery or fuel cell components, emission-free fuel production (such as “green” hydrogen), and catalysts, AutoSurfRecon is a freely downloadable tool.
The approach has implications for understanding catalytic processes, as performance is strongly influenced by surface features. The researchers highlight the tool’s potential for exploratory research, opening up a wider range of possibilities, by doing away with the dependence on human intuition.
“This highlights that the method works without intuitions,” Gómez-Bombarelli says. “And that’s good because sometimes intuition is wrong, and what people have thought was the case turns out not to be.” This new tool, he said, will allow researchers to be more exploratory, trying out a broader range of possibilities.