The main research field of Jörg Behler is the development and application of efficient interatomic potentials using modern machine learning techniques. Since November 2022 he holds the Research Professorship for Atomistic Simulations at the Research Center "Chemical Sciences and Sustainability" of the Research Alliance Ruhr, next to the new chair for Theoretical Chemistry II at Ruhr University Bochum.
To gain insight into complex chemical processes, it is crucial to understand the intricate atomic interactions in extended systems involving large numbers of atoms. While calculating the quantum mechanical interactions of a few atoms is still feasible, this is beyond reach even when using the most powerful supercomputers for larger systems. More approximate methods, on the other hand, lack the accuracy and predictive power to really understand the processes of interest at the atomic scale. To solve this problem Behler and his team employ neural networks to sidestep the need for approximations, which could compromise the accuracy of the results. These machine learning techniques enable them to push the boundaries of atomistic simulations to more and more complex systems.
What do you see as the challenges and chances in your area for the coming years?
Our research focus in theoretical chemistry is the development of new tools, which are urgently needed for a thorough understanding of the atomistic mechanisms driving processes in many fields, from chemical reactions in solution to the properties of materials, and in particular we are interested in the properties of solid-liquid interfaces combining these fields. This research direction is of central interest to RESOLV and our Research Centre, since solid-liquid interfaces play a pivotal role in sustainable chemistry, for example in energy storage and conversion, and electrochemical processes.
However, one of the biggest challenge lies in the high complexity of these systems compared to other fields of theoretical chemistry, which have already seen substantial advances in recent decades. For instance, in the past 15 years considerable progress has already been made in the theoretical understanding of heterogeneous catalysis at gas-solid interfaces. Although there are of course still many open questions to be answered, contributions from theoretical chemistry have been invaluable in combination with experimental data to reach our current level of understanding. Chemical processes taking place at the interface between complex materials and a liquid phase are much more complicated, and beyond what can be directly addressed by electronic structure methods like density functional theory. Our work strives to move beyond these limitations by developing new methods to represent multidimensional potential energy surfaces, which allow to transfer the accuracy of demanding electronic structure calculations to very large systems at a small fraction of the computational costs.
What do you consider your most significant scientific achievement or success?
Our group's major achievement lies in advancing the adaption and use of machine learning techniques for computer simulations of complex systems like bulk materials and aqueous solutions. Fifteen years ago, we have introduced the first a machine learning potential applicable to condensed systems paving the way to simulations of such systems with unprecedented accuracy. Before our development of these high-dimensional neural network potentials, early machine learning potentials were restricted to systems like very small molecules in vacuum or to diatomic molecules interacting with frozen surfaces. Such systems are too simple to gain insights into complex chemical processes, which often require the explicit inclusion of thousands of atoms. This has become possible with the methods we have developed 15 years ago. However, we are not only interested in method development, but also in the application of these methods. In fact, our work is inspired and driven by the need for new methods arising from interesting problems in chemistry and materials science. This approach has already led to the development of a series of several generations of machine learning potentials in our group, which continuously extended the physical phenomena that we are able to study in our simulations.
What do you think about the research at the Ruhr Area?
The Ruhr Area and its very active research institutions represent a close to ideal environment for challenging scientific endeavours. In this region, initiatives like RESOLV and the Research Alliance Ruhr,to name just two very successful examples, nicely illustrate the strength of the local universities leading to substantial scientific advances and innovation. Intense collaborations and the fruitful combination of the broad expertise of experts in many fields, which is further increased by further local research institutions like several Max Planck institutes, contribute to a very stimulating and dynamic environment, which is also supported by the geographical proximity of research groups within the Ruhr Area. Me and my group really enjoy being part of this now.