Volker Eyert is senior scientist at Materials Design, a company providing software, support, consulting, and contract research for atomistic materials research.
Volker Eyert received a physics degree (Dipl.-Phys.) in 1986 from the University of Münster, where he investigated valence instabilities in magnetic rare-earth chalcogenides with Professor Wolfgang Nolting. In 1991, he received a doctoral degree (Dr. rer. nat.) in physics from the Technical University of Darmstadt, where he implemented a new full-potential augmented spherical-wave (ASW) electronic-structure method with Professor Jürgen Kübler. After taking a research associate position in the group of Professor Ole Krogh Andersen at the Max-Planck-Institute for Solid State Research in Stuttgart, Volker Eyert got an offer from the Helmholtz Center Berlin to establish a new group on atomistic simulations for photovoltaic applications in 1995. In 1998, Volker Eyert completed his habilitation in theoretical physics at Augsburg University (Dr. rer. nat. habil. and Priv. Doz.), where he joined the faculty of the Institute of Physics as well as the Center for Electronic Correlations and Magnetism as an assistant professor, prior to becoming a member of the Materials Design team in 2011.
Volker Eyert’s main research areas are the development and implementation of electronic-structure methods as based on density functional theory as well as the application of these approaches to the investigation of a large variety of materials with emphasis on correlated transition-metal chalcogenides. Recently, he has extended his field of interest to the combination of ab initio calculations with machine-learning methods.
Volker Eyert is the author of about 150 scientific publications and book chapters as well as a monograph on the “Augmented Spherical Wave Method”, which forms the basis of the ASW software package also developed by him.
AI in Materials Modelling: A Game Changer?
We are currently witnessing an explosion of AI-related tools with an unprecedented impact not only on our life and our societies but also on our scientific practices and understanding. While in general, AI draws its power from the accumulation and digestion of huge amounts of data from all domains of human life, in materials research, AI is turning the page from physics-based to data-driven approaches exploiting especially a huge amount of measured and calculated materials data. As a consequence, future successful tools for materials research will need the capability to collect data and to generate new knowledge from them. In fact, AI in the physical sciences is not new and our company has pursued the combination of data and simulations for over a quarter of a century. For example, expert systems using databases of measured physical or chemical data allow us to identify new materials with dedicated properties while excluding, e.g., environmentally detrimental or poisonous materials. Similar strategies can be applied to calculated data and have recently initiated the transition from computing materials properties directly from ab initio simulations to approaches using a database of precalculated ab initio data for the generation of machine-learned potentials (MLPs), thereby opening the horizon to investigate larger models at longer time scales. Common to these tools is a qualitative change of the way we access and understand materials properties [1].
This talk will illustrate the recent development of modern software systems towards AI-guided materials research with examples from the MedeA computational environment of Materials Design. These include the Polymer Expert module as an expert system for polymer science, which enables generating new polymers with tailored properties from a large database of repeat units. In contrast, the MLP Generator uses large training sets of ab initio data to generate machine-learned potentials for use, e.g., in large-scale molecular dynamics simulations. The presentation ranges from a discussion of the basic concepts to challenges in creating efficient training sets and practical applications of MLPs to the calculation of properties, which benefit from the extended length and times scales offered by forcefield calculations. The talk will close with an outlook on current trends such as universal MLPs and MLPs for generalized structure-property relations.
References
[1] V. Eyert, J. Wormald, W. A. Curtin, and E. Wimmer, Machine-learned interatomic potentials: Recent developments and prospective applications, J. Mater. Res. 38, 5079 (2023);
Overview article of a Focus Issue on Machine-learned Potentials in Materials Research, J. Mater. Res. 38(24) (2023).
