Researchers are taking advantage of the high surface area and unique properties of nanoscale materials to develop new technologies. However the rational design of nanoscale materials is challenging due to the difficulty in determining the relationship between the properties of these materials and their atomic-scale structure using either experimental or computational methods.
We develop and apply computational methods to predict the structure and properties of nanoscale materials. One of the challenges in modeling these materials is that they have little or no translational symmetry, which can make the calculations much more expensive than comparable calculations on bulk materials. One of our active areas of research is the application of machine learning methods to address this problem.
The incorporation of prior knowledge into computational models significantly reduces the cost of modeling nanoscale materials. For example, even though an atom three layers below the surface of a nanoparticle and an atom four layers below the surface are not symmetrically equivalent, they are in similar environments. By explicitly including such information during model construction, we are able to develop fast and accurate models that explain and predict the properties of nanoscale materials.