Polarizable Water Model with Ab Initio Neural Network Dynamic Charges and Spontaneous Charge Transfer.

Area of Science:

• Computational chemistry
• Molecular dynamics
• Physical chemistry

Background:

• Accurate water simulation is difficult due to polarization and charge transfer complexities.
• Quantum mechanical methods are accurate but computationally expensive for large systems.

Purpose of the Study:

• Develop a polarizable water model using a Charge Neural Network (ChargeNN).
• Enable accurate simulations of large water systems with QM-level accuracy.

Main Methods:

• Developed a polarizable water model integrating Charge Model 5 atomic charges.
• Utilized a transferable charge neural network (ChargeNN) for QM predictions.
• Performed molecular dynamical simulations on liquid water and a water droplet.

Main Results:

• ChargeNN model accurately reproduces gas, liquid, and solid water properties.
• Successfully captures hydrogen-bond stretching and bending-libration spectral features.
• Reveals strong interfacial electric fields in water droplets due to network collapse and charge transfer.

Conclusions:

• ChargeNN model provides accurate polarization and charge transfer crucial for hydrogen bonds.
• Enables large-scale molecular simulations with high accuracy.
• Highlights the importance of QM-polarizable force fields for advanced simulations.