Talk 1 — Unravelling concerted diffusion mechanisms in multi-principal element alloys

Shashank Saxena, Anirudh Raju Natarajan
Laboratoire de conception et simulation des matériaux - MADES, EPFL

Multi-principal element alloys (MPEAs) have attracted considerable attention for their remarkable properties, many of which are governed by atomic diffusion. Kinetic Monte Carlo (KMC) simulations are widely used to compute the Onsager transport coefficients that describe this diffusion, but they typically restrict substitutional jumps to first-nearest-neighbor hops. In this work, we go beyond this approximation and systematically identify overlooked concerted mechanisms, in which several atoms move cooperatively, alongside the conventional single-hop pathways. We show that these mechanisms substantially alter the hop probabilities that enter KMC simulations. Establishing which mechanisms matter, and how strongly, is a necessary first step toward KMC models that capture the full diffusion physics of these alloys. This work therefore serves as a precursor to the accurate calculation of Onsager transport coefficients in compositionally complex alloys.

Talk 2 — Hierarchical universal open structure dataset for atomistic modeling

Sofiia Chorna, Michele Ceriotti
Laboratoire de science computationnelle et modélisation - COSMO, EPFL

Atomic-scale simulations have been transformed by machine-learning models that predict the properties of atomistic structures with the accuracy of quantum calculations, but at a small fraction of the cost. These models have increased dramatically in generality, thanks to the availability of synthetic datasets of ever-increasing size and scope. The proliferation of datasets, with properties often computed using different levels of electronic-structure approximation, is problematic, as there is often overlap between different datasets, and strong correlations between structures obtained with short molecular dynamics trajectories or geometry optimization. To address these problems, we introduce a protocol to stratify a dataset into a hierarchy of nested subsets, subsampling from the coarsest to the finest resolution based on an "ecumetric'' built to reflect the latent space representation of several popular universal machine-learning interatomic potentials. We then use this approach to combine most of the open structural datasets currently available, consolidating them into MAD-COSMO, a well-organized collection of almost 1.7 billion structures, whose 2²⁴ most diverse members form the MAD-CORE stratified coreset, which can be easily used for all sorts of tasks, from exploratory experiments to the training of foundation models at scale.