Nanoscale slip length prediction with machine learning tools.

Area of Science:

• Computational physics
• Materials science
• Fluid dynamics

Background:

• Predicting nanoscale slip length is crucial for understanding fluid behavior at interfaces.
• Existing methods often rely on computationally intensive simulations.

Purpose of the Study:

• To develop an accurate and efficient machine learning (ML) procedure for predicting nanoscale slip length.
• To compare the performance of different ML models, including linear and non-linear approaches.

Main Methods:

• Utilized multivariate regression, multi-layer perceptron, and random forest ML techniques.
• Compiled data from Molecular Dynamics simulations and existing literature for various liquids.
• Trained and tested models on a wide range of dynamical, geometrical, and simulation parameters.

Main Results:

• Non-linear ML models (neural networks, decision trees) demonstrated superior performance over linear regression.
• Trained models accurately predicted slip length within and near the input data range.
• Slip length becomes size-independent at larger channel dimensions, primarily influenced by wall roughness and wettability.

Conclusions:

• ML offers an efficient alternative to direct simulation for predicting nanoscale slip length.
• Non-linear models are key to achieving high accuracy in slip length prediction.
• Material properties like wall roughness and wettability dominate slip length in macroscopic systems.