Designing new compounds or alloys that can be used as catalysts in chemical reactions can be difficult and depend on the intuition of chemists. A team of researchers at MIT has developed a new approach using machine learning to remove the need for intuition and provide more detailed information than traditional methods.

The new system can identify new atomic configurations of materials that have previously been studied, and it can also determine the stability of previously identified configurations. This approach allows for an estimate of all variations, based on just a few first-principles calculations automatically chosen by an iterative machine-learning process.

The researchers’ method, called an Automatic Surface Reconstruction framework, avoids the need for hand-picked examples of surfaces to train the neural network. Instead, it begins with a single example of a pristine cut surface and uses active learning to select sites to sample on that surface.

The system achieves accurate predictions of surface energies across various chemical or electrical potentials using fewer than 5,000 first-principles calculations, out of the millions of possible chemical compositions and configurations.

The team has made their computer algorithms, called AutoSurfRecon, freely available for download, and they hope that it will inspire improvements by other researchers. Their work has been supported by the U.S. Air Force, the Department of Defense, and the National Science Foundation.


>Source link>

>>Join our Facebook Group be part of community. <<

By hassani

Leave a Reply

Your email address will not be published. Required fields are marked *