For low enough flow rates, turbulent channel flow displays spatial modulations of large wavelengths. This phenomenon has recently been interpreted as a linear instability of the turbulent flow. We question here the ability of linear stability analysis around the turbulent mean flow to predict the onset and wavelengths of such modulations. Both the mean flow and the Reynolds stresses are extracted from direct numerical simulation (DNS) in periodic computational domains of different size. The Orr-Sommerfeld-Squire formalism is used here, with the turbulent viscosity either ignored, evaluated from DNS, or modeled using a simple one-point closure model. Independently of the closure model and the domain size, the mean turbulent flow is found to be linearly stable, in marked contrast with the observed behavior. This suggests that the one-point approach is not sufficient to predict instability, at odds with other turbulent flow cases. For generic wall-bounded shear flows we discuss how the correct models for predicting instability could include fluctuations in a more explicit way.

Linear stability of turbulent channel flow with one-point closure