High-Throughput Aqueous Electrolyte Structure Prediction Using IonSolvR and Equivariant Graph Neural Network...

Area of Science:

• Computational Chemistry
• Materials Science
• Physical Chemistry

Background:

• Neural network potentials (NNPs) accelerate *ab initio* molecular dynamics (AIMD) for complex systems like electrolyte solutions.
• NNPs require large, accurate training datasets, often derived from Kohn-Sham density functional theory (DFT) calculations.
• Density functional tight-binding (DFTB) offers a computationally cheaper alternative to DFT for generating molecular dynamics data.

Purpose of the Study:

• To investigate the feasibility of using existing DFTB molecular dynamics trajectory data to train E(3)-equivariant graph neural network potentials.
• To assess the efficiency and accuracy of this approach for simulating aqueous electrolyte solutions.
• To determine the minimum data requirements for accurate NNP training.

Main Methods:

• Utilized existing DFTB molecular dynamics trajectory data from the IonSolvR database.
• Trained E(3)-equivariant graph neural network potentials using DFTB data.
• Evaluated the accuracy of the trained potentials by comparing simulated solvation structures with known results for Na+ and Cl- in aqueous NaCl solutions.
• Employed an embarrassingly parallel resampling approach to improve prediction accuracy.

Main Results:

• Accurate reproduction of the solvation structure of Na+ and Cl- ions in aqueous NaCl solutions was achieved with a small dataset (100 molecular dynamics frames).
• The performance of the neural network potentials systematically improved with further data through resampling.
• Demonstrated the potential of DFTB data for accelerating NNP training.

Conclusions:

• DFTB molecular dynamics data is a viable and efficient resource for training accurate neural network potentials for electrolyte solutions.
• Minimal training data is sufficient to capture essential structural properties of solvation.
• The proposed method offers a significant acceleration for simulating complex condensed-phase systems.