Efficient Molecular Dynamics Simulations of Deep Eutectic Solvents with First-Principles Accuracy Using Machine...

Area of Science:

• Materials Science
• Computational Chemistry
• Physical Chemistry

Background:

• Deep eutectic solvents offer tunable, eco-friendly alternatives to traditional solvents and electrolytes.
• Molecular dynamics (MD) simulations are crucial for understanding liquid properties but computationally intensive.
• First-principles MD simulations often face limitations in system size and simulation time.

Purpose of the Study:

• To explore the efficacy of machine learning (ML) interatomic potentials for MD simulations of deep eutectic solvents.
• To develop and validate a neural network potential for a choline chloride:urea mixture (reline).
• To assess the computational efficiency and accuracy of ML potentials compared to first-principles methods.

Main Methods:

• Training a neural network potential using density functional theory (DFT) data for reline.
• Performing large-scale, nanosecond-long MD simulations using the trained ML potentials.
• Analyzing structural and dynamical properties, including velocity cross-correlation functions.

Main Results:

• ML potentials enable MD simulations of thousands of atoms on nanosecond timescales at reduced computational cost.
• Simulated structural and dynamical properties of reline show good agreement with DFT-MD and experimental data.
• Velocity cross-correlation functions reveal insights into the collective dynamics of reline's molecular components.

Conclusions:

• ML interatomic potentials are a viable and efficient tool for simulating deep eutectic solvents.
• This approach overcomes the computational bottlenecks of first-principles MD for studying dynamic properties.
• The developed ML potential accurately captures the behavior of reline, facilitating further research.