Chemical Transferability and Accuracy of Ionic Liquid Simulations with Machine Learning Interatomic Potentials

Ionic liquids (ILs) are an exciting class of electrolytes finding applications in many areas, from energy storage, to solvents for performing reactions in, to fluids for storing molecular gasses. Moreover, ILs have been touted as ``designer'' solvents, as they can be mixed to form complex fluids with tailored physiochemical properties. To make predictions of these properties, classical molecular dynamics (MD) have often been employed instead to simulate the dynamics of ILs, as it is computationally inexpensive. However, its accuracy is often under question, which motivates some to use density functional theory (DFT), but this is often prohibitively expensive, which imposes severe limitations on system sizes and time scales. Machine learning interatomic potentials (MLIPs) offer a way to bridge this simulation gap, offering a cheaper method than DFT but without significant loss in accuracy. As using MLIPs for ILs is still a relatively unexplored, several unanswered questions remain to see if MLIPs can be transformative in the space of ILs. As ILs are often not pure, but mixed together or with additives, we first demonstrate that a MLIP can be trained to be chemically transferable whilst only being trained on a few compositions, using a salt-in-IL as a test case. Having demonstrated that a transferable model can be trained, we investigate the accuracy of a ML IP for a novel IL trained on less than 200 DFT frames. To test the predictions of this model, we compare it against DFT and experiments, and find reasonable agreement. Overall, we hope these results can be used as guidance for future development of MLIPs for ILs, and complex mixtures of them, to make accurate predictions of these novel electrolytes..

Medienart:	Preprint

Erscheinungsjahr:	2024
Erschienen:	2024

Enthalten in:	arXiv.org - (2024) vom: 04. März Zur Gesamtaufnahme - year:2024

Sprache:	Englisch

Beteiligte Personen:	Goodwin, Zachary A. H. [VerfasserIn] Wenny, Malia B. [VerfasserIn] Yang, Julia H. [VerfasserIn] Cepellotti, Andrea [VerfasserIn] Bystrom, Kyle [VerfasserIn] Johansson, Anders [VerfasserIn] Sun, Lixin [VerfasserIn] Batzner, Simon [VerfasserIn] Musaelian, Albert [VerfasserIn] Mason, Jarad A. [VerfasserIn] Kozinsky, Boris [VerfasserIn] Molinari, Nicola [VerfasserIn]

Links:	Volltext [kostenfrei]

Themen:	530 Physics - Chemical Physics Physics - Computational Physics

Förderinstitution / Projekttitel:

PPN (Katalog-ID):	XAR042717019