Predicting Newtonian viscosity using machine learning trained on equilibrium molecular dynamics

Newtonian viscosity (η0) defines the reference state for liquid rheology and serves as a cornerstone for constitutive modeling. Equilibrium molecular dynamics (EMD) provides a rigorous route to η0 through the Green-Kubo [M. S. Green, J. Chem. Phys. 22, 398-413 (1954) and R. Kubo, J. Phys. Soc. Jpn. 12, 570-586 (1957)] formalism, yet accurate evaluation of stress autocorrelations requires prohibitively long simulations. Here, we integrate EMD with machine learning to model the relationship between η0 and the stress correlation time (τ) for liquid n-hexadecane. A fine-tuned three-layer artificial neural network trained on EMD data captures the onset and evolution of the viscosity plateau across thermodynamic conditions. The model reproduces direct EMD results and aligns with extrapolated non-equilibrium predictions, offering a physically consistent and computationally efficient framework for determining equilibrium viscosity and supporting the development of rate-dependent viscosity models.