Rational constitutive law for the viscous stress tensor in incompressible two-phase flows: Derivation and tests against a three-dimensional benchmark experiment
Phys. Rev. Fluids 10, 074001 – Published 18 July, 2025
DOI: https://doi.org/10.1103/9j1f-w77d
Abstract
We analyze the representation of viscous stresses in the one-fluid formulation of the two-phase Navier-Stokes equations, the model on which all computational approaches making use of a fixed mesh to discretize the flow field are grounded. Recognizing that the Navier-Stokes-like equations that are actually solved in these approaches imply a spatial filtering, we show by considering a specific two-dimensional flow configuration that the proper representation of the viscous stress tensor requires the introduction of two distinct viscosity coefficients, owing to the different behaviors of shear and normal stresses in control volumes straddling the interface. Making use of classical results of continuum mechanics for anisotropic fluids, we derive the general form of the constitutive law linking the viscous stress tensor of the two-phase medium to the filtered strain-rate tensor, and take advantage of the findings provided by the above two-dimensional configuration to close the determination of the fluid-dependent coefficients involved. Predictions of the resulting anisotropic model are then assessed and compared with those of available ad hoc models against original experimental results obtained in a reference flow in which some parts of the interface are submitted to shear while others are mostly controlled by stretching. The selected configuration corresponds to a viscous buoyancy-driven exchange flow in a closed vertical pipe, generated by unstably superimposing two immiscible fluids with a large viscosity contrast and negligible interfacial tension and molecular diffusivity. Using different levels of grid refinement, we show that the anisotropic model is the only one capable of predicting correctly the evolution of the front of the ascending and descending fingers at a reasonable computational cost. We confirm this conclusion on a one-dimensional counter-current channel flow with a large viscosity contrast, and use this simple configuration to establish that the anisotropic model converges quadratically as the grid is refined, while the familiar ad hoc model based on an arithmetic averaging of the fluid viscosities only converges linearly.
