The long-time evolution of monopolar and dipolar vortices influenced by the largescale gradient of the ambient potential vorticity (the β-effect) is studied by direct numerical solutions of the equivalent barotropic quasi-geostrophic equation. Translation and reorganization of vortical structures are shown to depend strongly on their intensity. Transport of trapped fluid by vortical structures is illustrated by calculating particle trajectories and by considering closed isolines of potential vorticity and the streamfunction in a co-moving reference frame.

The initial behaviour of strong monopoles is found to be well described by a recent approximate theory for the evolution of azimuthal mode one, even for times longer than the linear Rossby wave period. In the long-time limit, strong monopoles transport particles mainly westward, although the meridional displacement is several times larger than the initial vortex size. The appearance of an annulus with opposite radial gradient of the potential vorticity around the vortex core is demonstrated. This annulus forms owing to the meridional vortex drift on the β-plane and results in reorganization of a strong monopolar vortex into a rotating tripole. A critical value of the vortex intensity is found, below which the tripolar structure does not appear even in the case of an initially shielded vortex. Weak monopolar vortices are able to trap particles and provide some west-meridional fluid transport, even in the case when they decay like a linear Rossby wave packet.

The evolution of initial f-plane dipoles on the β-plane is strongly dependent on the initial direction of propagation. Strong dipoles adjust to steadily propagating modon solutions either accelerating (westward case), decelerating (eastward case) or oscillating with a decaying amplitude (meridional case), thereby carrying trapped particles predominantly eastward. A steady state is not reached if the dipole intensity is below a critical value which depends on the initial direction of propagation. Weak dipoles either decay and shrink owing to Rossby wave radiation (westward case), gradually separate and split (eastward case), or disintegrate (meridional case) without longdistance fluid transport. Thus, on the β-plane monopoles provide mainly westward transport of trapped fluid, whereas dipoles provide mainly eastward transport. Only strong monopoles are found to provide significant meridional fluid transport.

An experimental and theoretical study of the mixing produced by a plume rising in a confined stratified environment is presented. As a result of the pre-existing stable stratification, the plume penetrates only part way into the region; at an intermediate level it intrudes laterally forming a horizontal layer. As time evolves, this layer of mixed fluid is observed to increase in thickness. The bottom front advects downward in a way analogous to the first front in the filling box of Baines & Turner (1969), while the lateral spreading of the plume occurs at an ever-increasing level and an ascending top front results. We develop a model of this stratified filling box; the model predicts the rate at which the two fronts advance into the environment.

It is found that stratification in the environment, when smooth, has no significant influence on the dynamics of the descending front. We show that the rate of rise of the ascending front is determined by the turbulent mixing occurring at the spreading level. Entrainment of environmental fluid from above into the overshooting plume is significant; as a result, a density interface develops at this level. Asymptotically, the system reaches a state in which a bottom convecting layer, with an almost homogeneous density, deepens in a stratified background. The model proposed for this large-time behaviour is based on the simple energetic formulation that a constant fraction of the kinetic energy supplied by the plume, for mixing across the interface, is converted into potential energy of the convective layer. Our experimental results suggest an efficiency of approximately 50 % for this conversion.

We discuss our results in the light of previous studies on turbulent penetrative convection and conclude that the theory developed should be valid for an intermediate range of values of the Richardson number characterizing the dynamic conditions at the interface. The model is applied quantitatively to the process of cooling of a room wherein stratification is relevant. The geological problem of replenishment of a magma chamber by a light input of magma is also analysed.

Surface waves superimposed upon a larger-scale flow are blocked at the points where the group velocities balance the convection by the larger-scale flow. Two types of blockage, capillary and gravity, are investigated by using a new multiple-scale technique, in which the short waves are treated linearly and the underlying larger-scale flows are assumed steady but can have a considerably curved surface and uniform vorticity. The technique first provides a uniformly valid second-order ordinary differential equation, from which a consistent uniform asymptotic solution can readily be obtained by using a treatment suggested by the result of Smith (1975) who described the phenomenon of gravity blockage in an unsteady current with finite depth.

The corresponding WKBJ solution is also derived as a consistent asymptotic expansion of the uniform solution, which is valid at points away from the blockage point. This solution is obviously represented by a linear combination of the incident and reflected waves, and their amplitudes take explicit forms so that it can be shown that even with a significantly varied effective gravity g’ and constant vorticity, wave action will remain conserved for each wave. Furthermore, from the relative amplitudes of the incident and reflected waves, we clearly demonstrate that the action fluxes carried by the two waves towards and away from the blockage point are equal within the present approximation.

The blockage of gravity–capillary waves can occur at the forward slopes of a finite-amplitude dominant wave as suggested by Phillips (1981). The results show that the blocked waves will be reflected as extremely short capillaries and then dissipated rapidly by viscosity. Therefore, for a fixed dominant wave, all wavelets shorter than a limiting wavelength will be suppressed by this process. The minimum wavelengths coexisting with the long waves of various wavelengths and slopes are estimated.

For a strictly inviscid barotropic flow with conservative body forces, the Helmholtz vorticity theorem shows that material or Lagrangian surfaces which are vortex surfaces at time t = 0 remain so for t > 0. In this study, a systematic methodology is developed for constructing smooth scalar fields φ(x, y, z, t = 0) for Taylor–Green and Kida–Pelz velocity fields, which, at t = 0, satisfy ω·∇φ = 0. We refer to such fields as vortex-surface fields. Then, for some constant C, iso-surfaces φ = C define vortex surfaces. It is shown that, given the vorticity, our definition of a vortex-surface field admits non-uniqueness, and this is presently resolved numerically using an optimization approach. Additionally, relations between vortex-surface fields and the classical Clebsch representation are discussed for flows with zero helicity. Equations describing the evolution of vortex-surface fields are then obtained for both inviscid and viscous incompressible flows. Both uniqueness and the distinction separating the evolution of vortex-surface fields and Lagrangian fields are discussed. By tracking φ as a Lagrangian field in slightly viscous flows, we show that the well-defined evolution of Lagrangian surfaces that are initially vortex surfaces can be a good approximation to vortex surfaces at later times prior to vortex reconnection. In the evolution of such Lagrangian fields, we observe that initially blob-like vortex surfaces are progressively stretched to sheet-like shapes so that neighbouring portions approach each other, with subsequent rolling up of structures near the interface, which reveals more information on dynamics than the iso-surfaces of vorticity magnitude. The non-local geometry in the evolution is quantified by two differential geometry properties. Rolled-up local shapes are found in the Lagrangian structures that were initially vortex surfaces close to the time of vortex reconnection. It is hypothesized that this is related to the formation of the very high vorticity regions.

The present study is devoted to the experimental investigation and theoretical modelling of the interaction between fluid flow and solidification during the impact of supercooled water drops onto an ice surface. Using a high-speed video system, the impact process is captured with a high spatial and temporal resolution in a side view. The lamella thinning and the residual ice layer thickness in the centre of impact are determined from the high-speed videos for varying drop and surface temperatures, and impact velocities. It is shown that the temperature of the impact surface has a negligible influence and the drop temperature has a dominating influence on the lamella thinning and the final ice layer thickness. For decreasing drop temperatures, higher freezing rates cause a decreased rate of lamella thinning and a larger thickness of the resulting ice layer. On the other hand, a higher impact velocity causes an increasing speed of lamella thinning and a smaller thickness of the resulting ice layer. Based on a postulated flow in the spreading lamella and considering the ice layer growth and the developing viscous boundary layer, the upper limit for the resulting ice layer thickness is theoretically modelled. The theory shows very good agreement with the experimental results for all impact conditions. Based on the derived theoretical scaling, a semi-empirical equation is obtained which allows an a priori prediction of the final ice layer thickness resulting from a single drop impact, knowing the impact conditions. This capability is important for the improvement of existing ice accretion models.

Vesicles exposed to the human circulatory system experience a wide range of flows and Reynolds numbers. Previous investigations of vesicles in fluid flow have focused on the Stokes flow regime. In this work the influence of inertia on the dynamics of a vesicle in a shearing flow is investigated using a novel level-set computational method in two dimensions. A detailed analysis of the behaviour of a single vesicle at finite Reynolds number is presented. At low Reynolds numbers the results recover vesicle behaviour previously observed for Stokes flow. At moderate Reynolds numbers the classical tumbling behaviour of highly viscous vesicles is no longer observed. Instead, the vesicle is observed to tank-tread, with an equilibrium angle dependent on the Reynolds number and the reduced area of the vesicle. It is shown that a vesicle with an inner/outer fluid viscosity ratio as high as 200 will not tumble if the Reynolds number is as low as 10. A new damped tank-treading behaviour, where the vesicle will briefly oscillate about the equilibrium inclination angle, is also observed. This behaviour is explained by an investigation on the torque acting on a vesicle in shear flow. Scaling laws for vesicles in inertial flows have also been determined. It is observed that quantities such as vesicle tumbling period follow square-root scaling with respect to the Reynolds number. Finally, the maximum tension as a function of the Reynolds number is also determined. It is observed that, as the Reynolds number increases, the maximum tension on the vesicle membrane also increases. This could play a role in the creation of stable pores in vesicle membranes or for the premature destruction of vesicles exposed to the human circulatory system.

Energy amplification in channel flows of Oldroyd-B fluids is studied from an input–output point of view by analysing the ensemble-average energy density associated with the velocity field of the linearized governing equations. The inputs consist of spatially distributed and temporally varying body forces that are harmonic in the streamwise and spanwise directions and stochastic in the wall-normal direction and in time. Such inputs enable the use of powerful tools from linear systems theory that have recently been applied to analyse Newtonian fluid flows. It is found that the energy density increases with a decrease in viscosity ratio (ratio of solvent viscosity to total viscosity) and an increase in Reynolds number and elasticity number. In most of the cases, streamwise-constant perturbations are most amplified and the location of maximum energy density shifts to higher spanwise wavenumbers with an increase in Reynolds number and elasticity number and a decrease in viscosity ratio. For similar parameter values, the maximum in the energy density occurs at a higher spanwise wavenumber for Poiseuille flow, whereas the maximum energy density achieves larger maxima for Couette flow. At low Reynolds numbers, the energy density decreases monotonically when the elasticity number is sufficiently small, but shows a maximum when the elasticity number becomes sufficiently large, suggesting that elasticity can amplify disturbances even when inertial effects are weak.

The stability of two-layer thermal convection in high-Prandtl-number fluids is investigated using laboratory experiments and marginal stability analysis. The two fluids have different densities and viscosities but there is no surface tension and chemical diffusion at the interface is so slow that it is negligible. The density stratification is stable. A wide range of viscosity and layer depth ratios is studied. The onset of convection can be either stationary or oscillatory depending on the buoyancy number B, the ratio of the stabilizing chemical density anomaly to the destabilizing thermal density anomaly: when B is lower than a critical value (a function of the viscosity and layer depth ratios), the oscillatory regime develops, with a deformed interface and convective patterns oscillating over the whole tank depth; when B is larger than this critical value, the stratified regime develops, with a flat interface and layers convecting separately. Experiments agree well with the marginal stability results. At low Rayleigh number, characteristic time and length scales are well-predicted by the linear theory. At higher Rayleigh number, the linear theory still determines which convective regime will start first, using local values of the Rayleigh and buoyancy numbers, and which regime will persist, using global values of these parameters.

The role of fluid motion in delivery of nutrients to phytoplankton cells is a fundamental question in biological and chemical oceanography. In the study of mass transfer to phytoplankton, diatoms are of particular interest. They are non-motile, are often the most abundant components in aggregates and often form chains, so they are the ones expected to benefit most from enhancement of nutrient flux due to dissipating turbulence. Experimental data to test the contribution of advection to nutrient acquisition by phytoplankton are scarce, mainly because of the inability to visualize, record and thus imitate fluid motions in the vicinities of cells in natural flows. Laboratory experiments have most often used steady Couette flows to simulate the effects of turbulence on plankton. However, steady flow, producing spatially uniform shear, fails to capture the diffusion of momentum and vorticity, the essence of turbulence. Thus, numerical modelling plays an important role in the study of effects of fluid motion on diffusive and advective nutrient fluxes. In this paper we use the immersed boundary method to model the interaction of rigid and flexible diatom chains with the surrounding fluid and nutrients. We examine this interaction in two nutrient regimes, a uniform background concentration of nutrients, such as might be typical of an early spring bloom, and a contrasting scenario in which nutrients are supplied as small, randomly distributed pulses, as is more likely for oligotrophic seas and summer conditions in coastal and boreal seas. We also vary the length and flexibility of chains, as whether chains are straight or bent, rigid or flexible will affect their behaviour in the flow and hence their nutrient fluxes. The results of numerical experiments suggest that stiff chains consume more nutrients than solitary cells. Stiff chains also experience larger nutrient fluxes compared to flexible chains, and the nutrient uptake per cell increases with increasing stiffness of the chain, suggesting a major advantage of silica frustules in diatoms.

The evolution of the most energetic features in the far-field region of a turbulent axisymmetric jet are investigated by applying the proper orthogonal decomposition (POD) to the double Fourier transform in time and azimuthal direction of the two-point velocity correlations of the streamwise velocity component. Measurements were obtained from 15 to 69 diameters downstream at jet exit Reynolds numbers ranging from 40 000 to 84 700 using two different experimental techniques, an array of 139 long hot wires and an articulated array of 15 hot wires.

The POD eigenspectra for the far jet were found to follow equilibrium similarity scaling sooner than the single-point statistics. The first POD eigenspectrum accounts for more than 50% of the resolved streamwise energy, and has three major peaks: a dominant one at azimuthal mode-2 for near-zero frequency, another at mode-1 at a constant local Strouhal number, $f \delta_{1/2}/U_c$, and a third smaller one for azimuthal mode-0 at near-zero frequency. The eigenspectra were independent of downstream position, but similar to those obtained as early as $x/D\,{=}\,6$ in Part 1 of this paper. All the eigenspectra integrated over frequency peaked at azimuthal mode-2.

Eady's model for the stability of a thermal wind in an inviscid, stratified, rotating system is modified to allow for deviations from hydrostatic equilibrium. The stability properties of the flow are uniquely determined by two parameters: the Richardson number Ri, and the ratio of the aspect ratio to the Rossby number δ. The latter parameter may be taken as a measure of the deviations from hydrostatic equilibrium (δ = 0 in Eady's model). It is found that such deviations decrease the growth rates of all three kinds of instability which can occur in this problem: ‘geostrophic’ baroclinic instability, symmetric instability, and Kelvin–Helmholtz instability. The unstable wavelengths for ‘geostrophic’ and Kelvin–Helmholtz instability are increased for finite values of δ, while the unstable wavelengths for symmetric instability are unaffected. The ‘non-hydrostatic’ effects (δ ≠ 0) are significant for symmetric and Kelvin–Helmholtz instability when δ [gsim ] 1, but not for ‘geostrophic’ instability unless δ [Gt ] 1. Consequently, the first two types of instability tend to be suppressed relative to ‘geostrophic’ instability by ‘non-hydrostatic’ conditions. Figure 3 summarizes the different instability régimes that can occur. In laboratory experiments symmetric instability can be studied best when δ [lsim ] 1, while Kelvin–Helmholtz instability can be studied best when δ [Lt ] 1.

We report an experimental study of the large-scale circulation (LSC) reversal in quasi-2D turbulent thermal convection, in which the aspect ratio ${\it\Gamma}$ ($=\text{height}/\text{length}$ of a rectangular box) is used as a parameter to perturb the stability of the LSC. It is found that the mean time interval $\langle {\it\tau}\rangle$ between two successive reversals increases strongly with increasing ${\it\Gamma}$. A stochastic model is proposed to incorporate the effect of the corner rolls. In the model, the aspect ratio serves as a tuning parameter for the relative weight of the corner rolls that damp the LSC. The model predictions for the shape of the bistable states of the system and $\langle {\it\tau}\rangle$ agree excellently with the experimental results, with $\langle {\it\tau}\rangle$ having an unexpected stretched exponential Rayleigh number dependence, ${\sim}\!\exp (Ra^{{\it\alpha}})$. We further show quantitatively that the main damping force of the LSC in a quasi-2D system is from the corner rolls rather than the viscous drag from the sidewalls, which bridges the difference found in quasi-2D and 3D systems.

We investigate the motion of a gas–liquid interface in a test tube induced by a large acceleration via impulsive force. We conduct simple experiments in which the tube partially filled with a liquid falls under gravity and hits a rigid floor. A curved gas–liquid interface inside the tube reverses and eventually forms a so-called focused jet. In our experiments, there arises either vibration of the interface or an increment in the velocity of the liquid jet, accompanied by the onset of cavitation in the liquid column. These phenomena cannot be explained by a considering pressure impulse in a classical potential flow analysis, which does not account for finite speeds of sound or phase changes. Here we model such water-hammer events as a result of the one-dimensional propagation of a pressure wave and its interaction with boundaries through acoustic impedance mismatching. The method of characteristics is applied to describe pressure-wave interactions and the subsequent cavitation. The model proposed is found to be able to capture the time-dependent characteristics of the liquid jet.

An experimental study of wall shear stress in an accelerating flow of water in a pipe ramping between two steady turbulent flows has been undertaken in a large-scale experimental facility. Ensemble averaged mean and r.m.s. of the turbulent fluctuations of wall shear stresses have been derived from hot-film measurements from many repeated runs. The initial Reynolds number and the acceleration rate were varied systematically to give values of a non-dimensional acceleration parameter ranging from 0.16 to 14. The wall shear stress has been shown to follow a three-stage development. Stage 1 is associated with a period of minimal turbulence response; the measured turbulent wall shear stress remains largely unchanged except for a very slow increase which is readily associated with the stretching of existing turbulent eddies as a result of flow acceleration. In this condition of nearly ‘frozen’ turbulence, the unsteady wall shear stress is driven primarily by flow inertia, initially increasing rapidly and overshooting the pseudo-steady value, but then increasing more slowly and eventually falling below the pseudo-steady value. This variation is predicted by an analytical expression derived from a laminar flow formulation. The start of Stage 2 is marked by the generation of new turbulence causing both the mean and turbulent wall shear stress to increase rapidly, although there is a clear offset between the responses of these two quantities. The turbulent wall shear, reflecting local turbulent activities near the wall, responds first and the mean wall shear, reflecting conditions across the entire flow field, responds somewhat later. In Stage 3, the wall shear stress exhibits a quasi-steady variation. The duration of the initial period of nearly frozen turbulence response close to the wall increases with decreasing initial Reynolds number and with increasing acceleration. The latter is in contrast to the response of turbulence in the core of the flow, which previous measurements have shown to be independent of the rate of acceleration.

A modified shallow-water model is presented for the collapse of tall columns of grains. The flow is divided in two parts. Depth-averaged shallow-water equations are applied to a thin horizontally spreading layer which is subjected to Coulombic friction. The falling mass of grains is gradually added to the zone of the initial column during the free-fall time of the column. This ‘rain’ is assumed to have no horizontal momentum. The results obtained here are in agreement with both planar and axisymmetric experiments over a range of aspect ratio $a$. In particular, the runout distance is found to vary as $a^{0.65}$ (planar) and $a^{0.52}$ (axisymmetric). The flow dynamics compares well with discrete simulations which have been successfully compared with experiments.

Lighthill’s acoustic analogy is formulated for turbulent channel flow with pressure as the acoustic variable, and integrated over the channel width to produce a two-dimensional inhomogeneous wave equation. The equivalent sources consist of a dipole distribution related to the sum of the viscous shear stresses on the two walls, together with monopole and quadrupole distributions related to the unsteady turbulent dissipation and Reynolds stresses respectively. Using a rigid-boundary Green function, an expression is found for the power spectrum of the far-field pressure radiated per unit channel area. Direct numerical simulations (DNS) of turbulent plane Poiseuille and Couette flow have been performed in large computational domains in order to obtain good resolution of the low-wavenumber source behaviour. Analysis of the DNS databases for all sound radiation sources shows that their wavenumber–frequency spectra have non-zero limits at low wavenumber. The sound power per unit channel area radiated by the dipole distribution is proportional to Mach number squared, while the monopole and quadrupole contributions are proportional to the fourth power of Mach number. Below a particular Mach number determined by the frequency and radiation direction, the dipole radiation due to the wall shear stress dominates the far field. The quadrupole takes over at Mach numbers above about 0.1, while the monopole is always the smallest term. The resultant acoustic field at any point in the channel consists of a statistically diffuse assembly of plane waves, with spectrum limited by damping to a value that is independent of Mach number in the low-M limit.

Experiments were performed to study vortex shedding behind a linearly tapered cylinder. Four cylinders were used, with taper ratios varying from 50:1 to 100:1. The cylinders were each run at four different velocities, adjusted to cover the range of laminar vortex shedding for a non-tapered cylinder. The flow was confirmed to consist of discrete shedding cells, each with a constant frequency. For a centrespan Reynolds number greater than 100, the dimensionless mean cell length was found to be a constant. Individual cell size was found to be roughly self-similar. New shedding cells were created on the ends of the cylinders, or in regions adjacent to areas not shedding. Successful scalings were found for both the cell shedding frequencies and their differences, the modulation frequencies. The modulation frequencies were found to be constant along the cylinder span. The shedding frequency Strouhal number versus Reynolds number curve was found to have a slightly steeper slope than the Strouhal number curve for a non-tapered cylinder. Vortex shedding was found to begin at a local Reynolds number of about 60, regardless of any other factors. End effects were found to be of little importance.

The vortex splits, which form the links between shedding cells, were found to be similar in some respects to those found by earlier investigators. Amplitude results suggested that the splits at different spanwise locations are temporally sequenced by an overall flow mechanism, a supposition confirmed by flow visualization. Wavelet analysis results showed that while the behaviour of the shedding frequencies in time was relatively unaffected by changing taper ratio, the behaviour of the modulation frequency in time was greatly affected. Comparisons with other experiments point out the universality of vortex splitting phenomena.

The dissolution process of small (initial (equivalent) radius $R_{0}<1$  mm) long-chain alcohol (of various types) sessile droplets in water is studied, disentangling diffusive and convective contributions. The latter can arise for high solubilities of the alcohol, as the density of the alcohol–water mixture is then considerably less than that of pure water, giving rise to buoyancy-driven convection. The convective flow around the droplets is measured, using micro-particle image velocimetry (${\rm\mu}$PIV) and the schlieren technique. When non-dimensionalizing the system, we find a universal $Sh\sim Ra^{1/4}$ scaling relation for all alcohols (of different solubilities) and all droplets in the convective regime. Here $Sh$ is the Sherwood number (dimensionless mass flux) and $Ra$ is the Rayleigh number (dimensionless density difference between clean and alcohol-saturated water). This scaling implies the scaling relation ${\it\tau}_{c}\propto R_{0}^{5/4}$ of the convective dissolution time ${\it\tau}_{c}$, which is found to agree with experimental data. We show that in the convective regime the plume Reynolds number (the dimensionless velocity) of the detaching alcohol-saturated plume follows $Re_{p}\sim Sc^{-1}Ra^{5/8}$, which is confirmed by the ${\rm\mu}$PIV data. Here, $Sc$ is the Schmidt number. The convective regime exists when $Ra>Ra_{t}$, where $Ra_{t}=12$ is the transition $Ra$ number as extracted from the data. For $Ra\leqslant Ra_{t}$ and smaller, convective transport is progressively overtaken by diffusion and the above scaling relations break down.

Understanding and predicting the behaviour of wind turbine wake flows over hills is important for optimal design of wind-farm configurations on topography. In this study, we present an analytical modelling framework together with large-eddy simulation (LES) results to investigate turbine wakes over two-dimensional hills. The analytical model consists of two steps. In the first step, we deal with the effect of the pressure gradient on the wake evolution; and in the second step, we consider the effect of the hill-induced streamline distortion on the wake. This model enables us to obtain the wake recovery rate, the mean velocity and velocity deficit profiles and the wake trajectory in the presence of the hill. Moreover, we perform LES to test our model and also to obtain new complementary insight about such flows. Especially, we take advantage of the LES data to perform a special analysis of the behaviour of the wake on the leeward side of the hill. It is found that the mainly favourable pressure gradient on the windward side of the hill accelerates the wake recovery and the adverse pressure gradient on the leeward side decelerates it. The wake trajectory for a hill of the same height as the turbine’s hub height is found to closely follow the hill profile on the windward side, but it maintains an almost constant elevation (a horizontal line) downstream of the hilltop. The trajectory of the wake on the leeward side is also studied for a limiting case of an escarpment, and it is shown that an internal boundary layer forms on the plateau which leads to an upward displacement of the wake centre. Finally, a parametric study of the position of the turbine with respect to the hill is performed to further elucidate the effect of the hill-induced pressure gradient on the wind turbine wake recovery.

We report direct measurements of slip flow during and after plane-Couette shearing of poly(styrene) solutions between glass surfaces. Slip is studied by visualizing the microscopic motions of micron-sized silica spheres suspended in the solutions. Slip velocities as large as 4 μm s−1 are observed during shear, and recoil displacements of as much as 80 μm are observed at the glass surfaces after the driving surface stops moving. Significant slip is observed when the Weissenberg number $\lambda\dot{\gamma}$ exceeds 0.5 or so, where λ is the relaxation time and $\dot{\gamma}$ the shear rate. The magnitude of the slip correlates strongly with the fluid's birefringence, indicating that slip is induced by polymer stress or orientation. Away from the centre of the sample, toward the free edges, a secondary flow is observed, which appears to be driven by a normal-stress imbalance at these edges.