Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 67: Methods in Computational Materials Modelling (methodological aspects, numerics)

MM 67.5: Vortrag

Donnerstag, 15. März 2018, 18:30–18:45, TC 006

Workflows and provenance tracking for high-throughput computational materials discovery — •Martin Uhrin, Sebastiaan Huber, Giovanni Pizzi, and Nicola Marzari — EPFL STI IMX THEOS, Lausanne, Switzerland

The last decade has seen a push in computational materials science towards cementing high-throughput approaches as a cornerstone of materials discovery and understanding. This in turn requires the development of new tools to enable researchers to shift from running individual calculations to manage thousands, if not millions of these, in a reliable, intuitive and reproducible way. In order to answer these needs, we have developed AiiDA, a python platform, to enable domain experts to encode their scientific expertise in highly customisable workflows that are easy to write, document, debug and share ensuring that such expertise is retained and can be built upon. Meanwhile, a database backend is used to automatically store the full provenance as a graph of inputs, calculations and corresponding outputs, allowing the user to see exactly where any result came from, or continue to work from any intermediate step. I will show how one can write and run AiiDA workflows in ipython notebooks and highlight the many advantages over the more traditional, ‘throwaway script’, approach and highlight some published high-throughput computational materials modelling projects that have been performed using AiiDA.