Regensburg 2022 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

CPP: Fachverband Chemische Physik und Polymerphysik

CPP 9: Modeling and Simulation of Soft Matter (joint session CPP/DY)

CPP 9.2: Talk

Monday, September 5, 2022, 15:15–15:30, H39

Atomistic Machine Learning for Aqueous Ionic Solutions — •Philip Loche, Kevin K. Huguenin-Dumittan, and Michele Ceriotti — Laboratory of Computational Science and Modeling, IMX, École Polytechnique Fédérale de Lausanne, 1015 Lausanne, Switzerland

Accurate modeling of matter at the atomic scale requires to simultaneously account for the quantum nature of the chemical bond - that usually manifests itself on short time and length scales - and long-range interactions, such as electrostatics and dispersion, that occur on a large scale and often result in phenomena with a long characteristic time. Electronic structure calculations provide an accurate description of both quantum and long-range effects, but are computationally demanding, and scale poorly with system size. Machine learning (ML) approaches have emerged as a very effective strategy to build surrogate models that provide comparable accuracy at a fraction of the cost, but the most widespread techniques base their efficiency and transferability on a local description of atomic structure, which makes them ill-equipped to deal with long-range effects.

Here, we are going to to connect local and long range physics in a data driven ML approach by applying the current ML techniques to, condensed-phase systems, involving the characterization of aqueous ionic solutions. We show that only a combination of a long and a short range approach is able to predict short distanced molecular vibrations as well a long ranged ionic screening lengths.