Machine Learning Accelerates Materials Discovery
A computational approach improves the understanding of carbon’s different states and guides the search for materials still undiscovered. Researchers at the U.S. Department of Energy’s (DOE) Argonne National Laboratory recently demonstrated an automated process for identifying and exploring promising new materials that combine machine learning and high-performance computing.
Using carbon as a prototype, the algorithm predicted how atoms order themselves under a wide range of temperatures and pressures to make up different substances. It then constructed a series of what scientists call phase diagrams to guide their search for new and useful states of matter.
When a material’s atomic structure changes, so does its electronic, thermal, and mechanical properties. A way to change the atomic structure of a material is to vary the surrounding pressure and temperature.
For example, diamond and graphite are wildly different materials consisting of carbon atoms —arranged differently. Graphite is a much more stable form of carbon than diamond. Under conditions of extreme pressure and heat, however, graphite slowly crystallizes into diamond. When removed from those extreme conditions, the diamond persists, lingering in a metastable state.
The ML algorithm constructed phase diagrams that mapped hundreds of these metastable states of carbon. Researchers trained the ML algorithm with synthetic data produced through simulation and approximates the results scientists would get through experimentation.
The algorithm identified the previously ambiguous structure of n-diamond (“new diamond”) and made new and surprising predictions about the structural features of n-diamond that the team then verified through experimentation. They also synthesized several phases predicted by the algorithm that they have not yet reported in the scientific literature. The structures of the samples matched the predictions, further verifying the algorithm.
In this study, they only applied the algorithm to carbon, but the scientists hope to use the same approach to systems of more than one element.
A paper on the study, “Machine learning the metastable phase diagram of covalently bonded carbon,” was published in Nature Communications.
