Taming Complexity in Materials Modeling

How materials behave at the smallest scale how atoms re-arrange themselves at surfaces and how they react to their environment and external stimuli is reasonably well understood: with quantum mechanical methods one can model relatively simple processes and compare them with well-controlled experiments. When systems become more complex, however, when a material contains many different elements or when it is exposed to gas atmospheres or a liquid, then these methods quickly reach their limits. In this Coordinated Research Project, SFB TACO, experimental and theoretical groups from physics and chemistry at TU Wien and Universität Wien work together closely to propel such methods a big step forward. The goal is to drastically accelerate such cal culations using Machine Learning methods. The project focuses on oxides, i.e., compounds of metals with oxygen. These materials are amongst the most common inorganic substances on our planet. Depending on their composition, they change their chemical and physical properties. This is both a blessing and a curse: On the one hand, this wide range makes it possible to achieve materials properties that are tailor-made for technological applications. On the other hand, the large variety of their structures, especially on their surfaces, makes it particularly hard to model these materials. Oxides are used in energy storage, in the conversion of solar energy into chemical energy, and in catalysis. The underlying processes and phenomena need to be well understood in order to realize better energy and materials conversion schemes, and the SFB TACO goals make a significant contribution towards methods development and scientific insights. The project team has at their disposal an array of experimental methods, which they will use to study a material under a wide range of conditions from single crystals in vacuum to technical powder samples under reaction conditions. So-called hand-shake methods, commonly used by all experimental groups ensure transferability of results. The theoretical working groups apply various machine learning approaches, from computer vision to the modeling of surface reactions and spectra. These methods are put to test and refined in close collaboration with the experimental groups.