Information-Driven Exploration of Crystal Chemistries for Radiation Detection Materials

The ability to suggest promising materials and a priori eliminate unfruitful inquiry is the key to new crystal chemistry searches. Variable spaces tend to be large and poorly defined, and property measurements (and computations) of candidate materials are not abundant. For simple binary systems, the presence of structural polymorphs and higher order compositions for A(m)B(n); m,n=1-3 would combinatorially generate over 300,000 candidates, greatly complicating our ability to explore candidate spaces. We have used knowledge extraction methods to evolve structural signatures to direct searches using performance-based criteria. The exploratory data methods used both supervised (support-vector machines) and unsupervised (disorder-reduction and principal-component) classification methods for structural signature development. The development of new candidates for radiation detection materials will be used as the case example for this talk.