MS 3: Machine-learning supported materials design

This minisymposium deals with the development, integration and exploitation of machine learning for the modeling and the design of materials across length scales. At small scales, machine-learning interatomic potentials (MLIPs) fitted to ab initio potential energy surfaces are nowadays one of the common tools for computational materials design. MLIPs enable us to upscale the ab initio accuracy for larger scale structures with many atoms and longer time scales. This allows us to efficiently simulate defects and microstructures in crystalline and amorphous systems as well as complex atomistic reactions. At the macroscopic scale, machine learning has proven to be a powerful tool for the materials modeling of continua based on discrete stress-strain data or even image recognition. Further developments include the design of pattern transformations in soft solids or the prediction of shell buckling. Among the fundamental and long-standing goals of materials modeling that may benefit from machine learning is certainly the on-the-fly numerical simulation across scales, that is, the direct integration of microscopic and macroscopic models in a single work flow. This minisymposium aims to clarify recent developments and present challenges associated with machine-learning based materials modeling and design from microscopic to macroscopic length scales.