The paper explores, using different levels of turbulence closure, the computed behaviour of the three-dimensional turbulent wall jet in order to determine the cause of the remarkably high lateral rates of spread observed in experiments. Initially, to ensure accurate numerical solution, the equations are cast into the form appropriate to a self-similar shear flow thereby reducing the problem to one of two independent variables.

Our computations confirm that the strong lateral spreading arises from the creation of streamwise vorticity, rather than from anisotropic diffusion. The predicted ratio of the normal to lateral spreading rates is, however, very sensitive to the approximation made for the pressure–strain correlation. The version that, in other flows, has led to the best agreement with experiments again comes closest in calculating the wall jet, although the computed rate of spread is still some 50% greater than in most of the measurements. Our subsequent calculations, using a forward-marching scheme show that, because of the strong coupling between axial and secondary flow, the flow takes much longer to reach its self-preserving state than in a two-dimensional wall jet. Thus, it appears very probable that none of the experimental data are fully developed.