Temperature-transferable coarse-graining of ionic liquids with dual graph convolutional neural networks

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning Applications

Background:

• Ionic liquids (ILs) are crucial for mechanistic insights and predicting properties of new ion combinations via computer simulations.
• Simulating ILs faces challenges due to vast differences in atomic and ensemble motion time scales, necessitating coarse-grained models.
• Developing accurate many-body potentials of mean force for coarse-grained IL models is computationally intensive and complex.

Purpose of the Study:

• To develop an improved coarse-graining methodology for ionic liquids using machine learning.
• To create a neural network model capable of learning the potential of mean force for ILs.
• To incorporate temperature as an explicit variable in interatomic potentials for IL simulations.

Main Methods:

• A neural network model was trained using data from all-atom classical molecular dynamics simulations of ionic liquids.
• The potential of mean force was represented by two jointly trained neural network interatomic potentials.
• These potentials capture coupled short-range and many-body long-range molecular interactions, with temperature as an explicit input.

Main Results:

• The developed model accurately reproduces structural quantities of ionic liquids.
• It demonstrates superior performance in capturing dynamics compared to temperature-independent baseline models.
• The model exhibits generalization to unseen temperatures and maintains low simulation costs.

Conclusions:

• Machine learning, specifically neural networks, offers a powerful approach to overcome challenges in coarse-graining ionic liquids.
• The proposed method effectively learns the potential of mean force, improving the accuracy and efficiency of IL simulations.
• This work paves the way for more reliable predictions of properties for experimentally inaccessible ionic liquid systems.