Data-mining our Way to Better Nanoparticle Structures

Crystallography has given us the positions of atoms in crystals for 100 years, but nano-particles require a radical rethink in approach. They form interesting non-space-filling structures that we want to synthesize and control for advanced devices, but how to accurately determine the 3D atomic arrangements?

Taking inspiration from genomics based data-mining approaches, we showed how large databases of candidate structures can be generated algorithmically and then mined efficiently and robustly to screen for candidate nanoparticle structures.

Atoms typically arrange themselves in symmetric arrangements making patterns.

The first step in any structure solution is determining these symmetries from the diffraction data.  But in typical nanoparticle data the underlying symmetries are hidden.

We trained a deep neural network to be able to categorize PDFs in to one of 45 symmetry groups called “space groups”.  After training on 80,000 known structures the model could return the correct space-group in the top six 90% of the time.  The math model could see the symmetry hidden in the data.

Additional Materials