The accurate description of the energy landscapes of molecular and condensed materials often requires expensive evaluations of quantum-mechanical models. On the other hand, a large number of energy evaluations are needed to explore the landscapes for structure optimization, thermodynamic sampling, and rate estimation. Several methods have been developed that aim to approximate energy landscapes. These methods range from physically motivated approximate quantum mechanical and fully empirical energy models that describe the bonding of materials, to approaches from computer science such as machine-learning that are agnostic to the underlying chemical bonds. Ideally, a hierarchy of models with increasing and controllable accuracy is desirable to efficiently sample energy landscapes. Unfortunately, the origin of the current surrogate models in different scientific communities has led to a disconnect. As a result, knowledge about the different methods and their trade-offs have not penetrated the disconnected communities, hindering progress.
In this workshop, we will bring together researchers from the mathematics and computer science fields of optimization, optimal control and model development with chemists, physicists, and materials scientists working on model development and energy landscape optimization and sampling problems. The goal is to discuss future methods and algorithms for the development of surrogate models and parallel exploration and their application in structure optimization, sampling and dynamics.
This workshop will include a poster session; a request for posters will be sent to registered participants in advance of the workshop.
Gabor Csányi, Co-chair (University of Cambridge)
Richard Hennig, Co-chair (University of Florida)
Frank Noe (Freie Universität Berlin)