Two new papers leveraging OpenKIM high-throughput calculations to study correlations between material properties have been published. Both papers use hundreds of interatomic potentials as virtual materials to demonstrate that small-scale properties such as elastic constants, stacking fault energies, and others, can be used to predict larger-scale properties. Such relationships have long been qualitatively known by materials scientists, but the automatic high-throughput computation of material properties using the OpenKIM Pipeline enabled us to develop quantitative statistical models. The small-scale properties found to be most covariant with large-scale properties are likely to be good targets for fitting interatomic potentials.
The first, Fundamental microscopic properties as predictors of large-scale quantities of interest: Validation through grain boundary energy trends, was published in Acta Materialia. Here, the larger-scale property is grain boundary energy, also computed automatically using OpenKIM. The grain boundary Test Driver was previously described in Automated determination of grain boundary energy and potential-dependence using the OpenKIM framework, published in Computational Materials Science. We develop a statistical model using data from 304 interatomic potentials and additionally confirm that the correlations hold for DFT, where data is available.
The second, Cross-scale covariance for material property prediction, was published in npj Computational Materials. Here, the large-scale property is plastic flow strength, computed in collaboration with the IAP-UQ group at Lawrence Livermore National Laboratory. The statistical pool in this study includes 178 interatomic potentials. In this case, the large-scale property is completely out of reach for DFT, but DFT results for the small-scale properties are plugged into the developed statistical model to predict flow strength.