Deblurring our View of Atomic Arrangements in Complex Materials

Novel materials will impact our society in new and exceptional ways. Imagine, for instance, a material that is transparent with an exceptionally low thermal conductivity: it could be used as a window that completely blocks radiated heat from sunlight during summer and traps all the heat indoor during winter. It has been shown that novel properties of materials often come from ‘imperfections’, such as defects, nanoscale structures, interfaces, surfaces, or multi-scale heterogeneities. However, those important imperfections also introduce challenges in both theoretical and experimental aspects.  From the theoretical side, we need a model that captures the physics of the complicated phenomena emerging from the imperfections, and from the experimental side, a systematic approach to quantify information content encapsulated in the measured signals is needed.

Due to the degraded information obtained in experiments with defective materials,  this whole problem becomes an ill-posed inverse problem, where there is insufficient information in the data to constrain a unique model solution.  In this NSF-funded DMREF team, we combine expertise and research from different fields all within Columbia University. Expertise in materials and scattering (Prof. Simon Billinge, APAM) is combined with applied math theory (Qiang Du, APAM) and machine learning and algorithmic statistics (Prof. Daniel Hsu, CS).  Our goal is to better understand the novel properties induced by imperfections and accelerate materials discovery, by formulating the problem in rigorous mathematical format, applying information theory and physical knowledge to experimental data, and implementing it all under a machine learning approach. Broader impacts can be expected since materials problems solved this way will allow novel materials to be discovered that will have impacts in many different emerging technologies from energy conversion to healthcare. We expect that the numerical and theoretical tools developed during this project will be highly transferable to other research fields.

This project is funded by the US National Science Foundation DMREF program.

This material is based upon work supported by the National Science Foundation under DMR Grant No. 1534910

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

The NSF award may be seen here