Invited
Atomic cluster expansion: Approaching quantum accuracy in large-scale atomistic simulations

The atomic cluster expansion (ACE) provides a general and mathematically complete representation of the properties of interacting atoms that fulfills fundamental invariances required to obtain a physically meaningful representation of energy. Existing ACE models for elements as well as multicomponent systems have been shown to be superior or comparable to the best currently available machine learning (ML) potentials in terms of accuracy while being significantly more computational efficient. These models can be developed either as general-purpose parametrizations with a broad transferability or as purpose-specific parametrizations tailored to a specific application. The former are usually more attractive for material science community as they aim to describe materials under various conditions, but building such models often requires thorough expertise and extensive training datasets. The purpose-specific parametrizations have only limited ranges of applicability, but are also less demanding in terms of training data. We will demonstrate that ACE can be applied also to study complex phenomena occurring in ferroelectric perovskite oxides, such as phase transformations, domain wall motion, and behavior of defects.