From ab initio to continuum: Linking multiple scales using deep-learned forces.

Area of Science:

• Computational physics
• Materials science
• Physical chemistry

Background:

• Predicting water behavior in nanochannels is crucial for nanotechnology.
• Traditional methods like molecular dynamics struggle with accuracy and computational cost at the nanoscale.

Purpose of the Study:

• To develop a novel deep learning algorithm, DeepForce, for accurate prediction of confined water concentration profiles.
• To bridge the gap between ab initio physics and continuum theory for nanoscale simulations.

Main Methods:

• Developed DeepForce, a deep learning algorithm integrating ab initio physics with continuum theory.
• Utilized the Nernst-Planck equation to solve continuum theory with deep-learned forces.
• Validated DeepForce against ab initio molecular dynamics simulations.

Main Results:

• DeepForce accurately predicts structural properties of confined water with quantum-scale accuracy.
• Achieved a relative error of less than 7.6% for both small (L < 6 nm) and large (L = 20 nm) nanochannels.
• Demonstrated DeepForce's superior performance over classical molecular dynamics for interfacial water physics.

Conclusions:

• DeepForce offers a computationally efficient and highly accurate method for simulating confined water.
• Highlights the limitations of classical molecular dynamics when quantum effects are significant.
• Paves the way for advanced simulations in nanoscience and nanotechnology.