The response of a well-developed turbulent boundary layer to suddenly applied convex surface curvature is investigated, using conditional-sampling techniques so that the turbulent and non-turbulent regions of the flow can be clearly distinguished. The conclusion of this and the companion paper by Hoffmann, Muck & Bradshaw (1985) is that the effects of convex (stabilizing) and concave (destabilizing) curvature on boundary layers – and presumably on other shear layers – are totally different, even qualitatively: mild convex curvature, with a radius of curvature of the order of 100 times the boundary-layer thickness, tends to attenuate the pre-existing turbulence, apparently without producing large changes in statistical-average eddy shape, while concave curvature results in the quasi-inviscid generation of longitudinal (‘Taylor-Görtler’) vortices, together with significant changes in the turbulence structure induced directly by the curvature and indirectly by the vortices.

From the point of view of calculation methods, the implication is that, although stabilizing and destabilizing curvature are connected by a common dimensional analysis, the differences are such that the one cannot be regarded as a useful guide to the treatment of the other. Specifically, rates of change of turbulence-structure parameters with curvature parameter are likely to be nearly discontinuous at zero curvature, and in particular the time of response of a turbulent boundary layer to convex curvature, implying mere attenuation, is very much less than the time of response to concave curvature, implying reorganization of the eddy structure.