Webinars

Machine learning force fields deliver near quantum-chemical accuracy for large, complex systems, but collecting the reference data they are trained on remains labour-intensive and computationally expensive. This webinar introduces aims-PAX (Parallel Active eXploration), a fully automated, open-source framework that streamlines data generation through parallelised active learning with FHI-aims. Tobias Henkes will showcase how aims-PAX autonomously builds high-quality training datasets across gas-phase, solvated, and periodic systems, reducing the number of required reference calculations by up to three orders of magnitude, and how the collected data can be used to train state-of-the-art force fields such as MACE.

High-level quantum mechanical methods offer exceptional accuracy, but their computational cost often makes realistic chemical systems intractable. Embedding methods address this by partitioning systems into regions treated at different levels of theory, enabling accurate calculations where the chemistry of interested is located while using efficient approximations for the environment. In this webinar, Andrew J. Logsdail will introduce QM/MM (quantum mechanics/molecular mechanics) approaches for multiscale simulations, and Gabriel A. Bramley will present the EmbASI framework for QM/QM embedding with FHI-aims.

Join us for this webinar on the SISSO approach for explainable artificial intelligence in materials science. Thomas Purcell will showcase recent methodological developments that have expanded the capabilities of SISSO, including improved feature representation through binary expression trees and enhanced solver algorithms for both regression and classification problems. Lucas Foppa will demonstrate hierarchical SISSO applications in heterogeneous catalysis, illustrating how the method identifies the "materials genes" of catalytic systems and enables the development of design rules for improved catalysts.