Seminar on machine-learning interatomic potentials by Miguel Caro

Carbon (nano-)materials are used in a myriad of applications because of their extreme and diverse properties. At the root of this diversity of properties is the ability of carbon to form different kinds of chemical bonds, even in elemental form. Functionalization of carbon, e.g., by decorating carbon surfaces with different CHO(N)-containing chemical motifs, can further widen tthe range of properties and applications, in particular in electrochemistry and electrocatalysis. To simulate the atomic structure of carbon materials reliably an accurate method is required; because many of these carbon materials are disordered, this method also needs to be computationally affordable (to accommodate large simulation boxes). The emergence of machine learning (ML) potentials in the last decade has made it possible to bridge this gap between speed and accuracy and opened the door for achieving unprecedented level of detail in nanocarbon simulation. In this talk Dr. Caro will discuss how we have used ML potentials and related techniques to carry out predictive studies on the structure and properties of carbon materials, especially amorphous carbon, and the latest developments in addition of van der Waals corrections to ML potentials and prediction of x-ray spectroscopy, both with application to carbon materials.