One of the promises of the information age is that collecting and analyzing larges amounts of data will lead to new insights. Materials scientists have embraced that idea, creating several public databases of materials’ properties over the past decade. New software aims to put that data to use by mapping mathematical relationships between properties for the first time, helping scientists calculate property values not yet measured and potentially find new materials or new uses for existing ones (Matter 2020, DOI: 10.1016/j.matt.2019.11.013).
Kristin A. Persson of the University of California, Berkeley, and the Lawrence Berkeley National Laboratory and colleagues encoded known mathematical relationships between materials properties in an open-source computational framework called propnet. It lets users input known values and produce calculations for others. For example, if a user inputs experimental data about a material’s atomic density, propnet could compute its volume per atom in a unit cell. But it can also trace more distant connections between properties, calculating values 10 relationships away from the starting data. In one demonstration, the researchers inputted 20 values for 20 properties of the semiconductor wurtzite CdTe, and propnet calculated 629 values for 41 new properties.
The software has information about the connections between more than 100 properties in its current version, and Persson expects users to add more. Propnet can also be combined with other software tools, like machine-learning algorithms.
The software could help researchers see connections beyond their fields, like relationships that span optical and electronic properties, Persson says, or help find new materials by illuminating relationships that could guide an efficient search.
Materials informatics experts are enthusiastic about propnet. Using mathematical relationships between properties is “going to be an essential component of the materials innovation infrastructure,” says James A. Warren, director of the Materials Genome Project at the National Institute of Standards and Technology. Propnet will make it easier for researchers to infer properties from databases and will give those inferences “a firmer statistical footing,” says Gerbrand Ceder of UC Berkeley.