An AI model adapted from image inpainting can reconstruct missing atoms in crystal structures with a reported 97 percent success rate.

A crystal structure can appear nearly complete yet still be unusable for computer simulations because some of its atoms are missing. Hydrogen is a frequent source of these gaps, and researchers have now adapted an artificial intelligence technique used to repair images to predict where those hidden atoms belong.

The approach was developed by a team led by Giovanni Pizzi of the PSI Center for Scientific Computing, Theory and Data, working with researchers from the universities of Parma and Modena in Italy. Described in npj Computational Materials, the method applies computer vision, in which AI recognizes and interprets visual information, to incomplete crystal structures.