ENTEG guest seminar by Dr Max Hodapp entitled: "Ab initio-accurate large-scale molecular dynamics using machine-learning interatomic potentials"
|When:||Fr 30-04-2021 14:30 - 16:00|
|Where:||online, link: http://meet.google.com/zmn-zmfq-rbn|
The Multi-Scale Mechanics (MSM) group (PI: Dr Francesco Maresca) presents:
A guest seminar by Dr Max Hodapp entitled: "Ab initio-accurate large-scale molecular dynamics using machine-learning interatomic potentials"
One of the main open challenges in computational materials science is the construction of interatomic potentials which achieve quantitative agreement with fully—but infeasible—ab initio models for large-scale problems, possibly involving tens of thousands of atoms. Empirical potentials, arguably the most popular type of interatomic potentials, generally fail in making quantitative predictions. Therefore, they largely remain inadequate for a predictive modeling of multicomponent systems and, thus, for a computational discovery of new materials. The advent of machine learning interatomic potentials (MLIPs) yet holds promise that overcoming these limitations appear within sight. However, a well-known drawback of state-of-the-art MLIPs is their poor ability to extrapolate. This demands for carefully chosen, problem-dependent training data which can hardly be defined by the user prior to a simulation due to the vast amount of possible atomic neighborhoods to be considered. One approach to overcome this problem is active learning. In contrast to passive learning, where the training set is finalized before running a simulation, any configuration appearing in a simulation is allowed to become a part of the training set. However, constructing small-sized training configurations to be computable with DFT—out of a large-scale configuration which is not locally periodic—is far from straightforward. In this talk I will present our recent developments on active learning algorithms for large-scale problems and show how it can be efficiently combined with passive learning. I will present numerical examples including stacking faults and dislocations in (random) alloys. In addition, I will give an outlook on how simulations using MLIPs can be combined with other techniques, such as atomistic-to-continuum coupling, to construct efficient and very accurate problem-specific algorithms for computing mechanical properties of materials.
Dr Max Hodapp graduated in 2013 from the Karlsruhe Institute of Technology (KIT), Germany, in mechanical engineering with specialized subjects in theoretical continuum mechanics. He defended his PhD thesis in 2018 at the Ecole Polytechnique Fédérale de Lausanne (EPFL) on concurrent atomistic-to-continuum coupling methods which has been has been acknowledged by the Swiss Community for Computational Methods in Applied Sciences with the 2019 award for one of the two best PhD theses. At Skoltech, Max is working on machine-learning interatomic potentials, active learning algorithms and atomistic-to-continuum coupling with the aim to perform large-scale atomistic simulations at an acceptable computational cost in order to predict microstructural material properties.