Glad to see our latest work eventually out! We propose an automated workflow to help design structural datasets for the design of machine-learned interatomic-potentials for chemical reactivity studies. With ArcaNN, anyone can easily design a reactive forcefield with a chosen quantum accuracy, virtually for any type of chemical reaction in any type of environment! This hopefully lowers the cost of entry to use these revolutionary approaches which are becoming routinely applied in material sciences, where the training dataset are more easily generated, but which are still challenging to deploy for chemical reactions, where the training dataset encompassing structures all along the reaction pathways is not known a priori.
Great team effort at Ecole normale supérieure with Damien Laage, Rolf David, Miguel de la Puente Martínez, Axel Gomez and Olaia Anton. Already many users and hopefully many more to come with the code now made public: https://lnkd.in/dAicnce4.
In our latest featured #DigitalDiscovery article, meet ArcaNN, an automated framework for generating training datasets for reactive machine learning potentials. Don't miss this open access paper by David, Stirnemann, Laage et al.: https://lnkd.in/eFDSDynw
Find ArcaNN itself here: https://lnkd.in/efTDptiB