April 22 - 26, 2024
Seattle, Washington
May 7 - 9, 2024 (Virtual)

Event Supporters

2024 MRS Spring Meeting
MT01.11.09

Deep Learning-Assisted Analysis of Molecular Dynamics Simulations of LiTFSI/PYR14TFSI Ionic Liquid Electrolyte

When and Where

Apr 26, 2024
4:45pm - 5:00pm
Room 320, Level 3, Summit

Presenter(s)

Co-Author(s)

Chanui Park1,Sangdeok Kim1,Anseong Park1,Seungtae Kim1,Woojin Kang1,Won Bo Lee1

Seoul National University1

Abstract

Chanui Park1,Sangdeok Kim1,Anseong Park1,Seungtae Kim1,Woojin Kang1,Won Bo Lee1

Seoul National University1
Over the past decades, significant research has been conducted to understand microscopic behavior of ionic liquid electrolytes (ILEs) through molecular dynamics (MD) simulations. The development of polarizable force fields is one of the most remarkable achievements as they predict structural and dynamical properties of ionic liquid-based systems very accurately. However, even with the help of polarizable force fields, one cannot analyze atomic dynamics of a system in detail because most of the properties are evaluated by "averaging" individual atomic or molecular properties. A deep learning technique called Graph Dynamical Networks (GDyNets) has been suggested to learn atomic scale dynamics from MD simulations. GDyNets trains local environments around a specific target atom and classify them into states in an unsupervised manner. The classification model is trained in the direction where the VAMP loss decreases. The classification results are combined with conventional analysis techniques to calculate state-wise properties which can describe the system according to each state. Using this method, we analyzed MD trajectories of LiTFSI/PYR<sub>14</sub>TFSI ILE with a range of lithium mole fractions, where the trajectories were generated under the APPLE&P polarizable force field. Li<sup>+</sup> ions were treated as target atoms and classified into 3 local configurational states. State-wise radial distribution function, coordination number, spatial distribution function were defined and calculated to identify each state. The identified states were a Li<sup>+</sup> ion coordinated by 2 and 3 TFSI<sup>−</sup> ions, and a cluster composed of multiple Li<sup>+</sup> and TFSI<sup>−</sup> ions. Also, a transition matrix based on the Markov state model was generated from the classification results. With this matrix, dynamical properties of the states and transitions between states were measured to reveal which state or transition dynamics is faster than others. Finally, a design rule for ILEs with faster Li<sup>+</sup> ion conduction was suggested.

Keywords

diffusion

Symposium Organizers

Raymundo Arroyave, Texas A&M Univ
Elif Ertekin, University of Illinois at Urbana-Champaign
Rodrigo Freitas, Massachusetts Institute of Technology
Aditi Krishnapriyan, UC Berkeley

Session Chairs

Rodrigo Freitas
Rebecca Lindsey

In this Session