Simulation of matter at the atomic scale has undergone a profound transformation over the past decade. Data-driven techniques have enabled the construction of models that predict the potential energy and properties of atomic configurations with the accuracy of quantum mechanical calculations, but without solving the electronic structure problem explicitly, and therefore at a small fraction of the cost.

In general, machine learning (ML) methods allow the “lifting” of predictions based on an explicit physical model to a coarser-grained description where some of the degrees of freedom (e.g. electrons, or atom groups) are integrated out, while maintaining the accuracy of predictions. For physical simulations, this results in the systematic bridging of time and length scales, and has brought closer communities that have traditionally focused on different scales: from solving the Schrödinger equation as accurately as possible, to sampling the thermodynamics of complex biomolecules.

In this CECAM-Psi-k Research Conference we aim to discuss the latest developments in the application of ML to the bottom-up simulation of matter, covering a broad range of applications from materials to biological systems, and highlighting in particular the connections and synergies between different traditional methods, between communities, as well as between academic and industrial research.