News

In many fields of science, and especially in those fields focusing on any form of energy conversion and storage such as photovoltaics, catalysis, or batteries the quest for improved efficiency really is a quest for improved materials. These may be the (organic) semiconductors of the solar cell, the catalyst, or the electrodes of the batteries. Computer simulations offer a way to evaluate new materials without having to expensively synthesize them first in the lab. Often, the such simulation entail the treatment of the material from so-called first-principles methods, where the material is treated quantum mechanically, generally via density functional theory. Unfortunately, such calculations are very expensive in terms of computer time (and thus energy expenditure) which often only allows the sampling of small regions of the space of possible materials. In this lecture I will thus describe the use of data-driven and machine learned (ML) surrogate models for materials properties. These not only allow the efficient evaluation of enormous numbers of candidate materials, but, employing active ML methods, even allow a search in design space to strike a balance between exploitation – i.e. optimization of the target properties – and exploration of so far unknown regions of the space. The use of such models will be illustrated on two examples from current research, the search of the design space of organic semiconductors and the ML prediction of certain charge distributions (polarons) in battery materials.