An experimental investigation was carried out to study the effects of noncircular collars on an axisymmetric jet using surface flow visualization and hot-wire anemometry. Circular, square, and triangular collars with expansion ratios of 1.20, 1.35, and 1.54, respectively, with collar lengths of up to two jet diameters were used. Flow visualization shows that circular collars led to equidistant flow reattachments along the collar wall, while square and triangular collars resulted in the formation of a pair of counter-rotating vortex-pairs on each side of the collar wall. These vortex-pairs are caused by the presence of the three-dimensional velocity gradients between locations of minimum and maximum step-heights, which drove fluid from the collar wall centerlines towards the corners. Time-averaged velocity measurements show that the circular collar required the shortest collar length to achieve maximum centerline velocity decay, followed by square and triangular collars. Centerline turbulence intensity and velocity spectra results reveal that all three collars were able to suppress vortex-pairing events when they were sufficiently long with the triangular collar being the most effective. Furthermore, the triangular collar also produces the widest overall jet-spread, ahead of square and circular collars, respectively, even though it demonstrates significantly different jet-spreads along planes of minimum and maximum step-height. Self-excitation frequencies from all three collars could be distinguished into distinct frequency bands with changes in the collar lengths. Within each band, the circular collar results in the largest variation of excitation frequency over the square and triangular collars, respectively. Lastly, square and triangular collars require wider ranges of collar lengths for each of the frequency bands, as compared to the circular collar.

Recent studies ͓Lathrop et al., Phys. Rev. A 46, 6390 ͑1992͒; Lewis et al., Phys. Rev. E 59, 5457 ͑1999͔͒ on Taylor-Couette flow, where the inner cylinder is rotating and the outer one is at rest, show that, despite earlier predictions ͓Wendt, Ing. Arch. 4, 557 ͑1933͒; Tong et al., Phys. Rev. Lett. 65, 2780 ͑1990͔͒, the non-dimensional torque (GϭT/ 2 L) does not follow a fixed power-law scaling ͑i.e., GϳRe ␣ , where ␣ is a constant value͒ for 800ϽReϽ1.23ϫ10 6 . Here, we perform simultaneous flow visualization and high precision torque measurements of the same flow configuration using a Haake RS-75 Rheometer to establish if this is also true in the lower Reynolds number range (ReϽ800). Results show that, although ␣ varies with the Reynolds number, it can be approximated reasonably well with a constant value ␣ϭ1 for Re/Re c Ͻ1 and ␣ϭ1.5 for 1.5 ϽRe/Re c Ͻ6.32. The latter finding is in good agreement with that of Wendt ͓Ing. Arch. 4, 557 ͑1933͔͒. A possible explanation for the differences with the results of earlier studies is provided in this paper.

In Fig. 2, the torque values ͑T͒ on the vertical axis should be multiplied by 10 Ϫ4 , and in Fig. 5͑a͒, the non-dimensional torque (GϭT/ 2 L) should be multiplied by 10 Ϫ3 . These corrections do not affect the conclusion drawn in the paper ͑i.e., G scales approximately with Re 1.0 for Re/Re c Ͻ1, and with Re 1.5 for 1.5ϽRe/Re c Ͻ6.32. Also, G ⍀ scales approximately with Re Ϫ1 for Re/Re c Ͻ1, and with Re Ϫ0.6 for 2.0ϽRe/Re c Ͻ6.32).