The failure of local isotropy to describe the experimentally obtained derivative moments in turbulent shear flows has previously been well-documented, but is briefly reviewed. The same data are then used to evaluate the hypothesis that the turbulence is locally axisymmetric. Locally axisymmetric turbulence is defined herein as turbulence which is locally invariant to rotations about a preferred axis.

The derivative moment relations are derived from the general form of the two-point velocity correlation tensor near the origin for axisymmetric turbulence. These are used to derive relations for the rate of dissipation of kinetic energy, the mean-square vorticity, and the components of each. Almost all of the experimental derivative moment data are shown to be consistent with these equations, and thus with local axisymmetry.

Stationary finite-amplitude wave disturbances in a stratified shear flow with Richardson number larger than ¼ are investigated for large Reynolds numbers when viscosity and thermal conductivity, as well as nonlinearity, are essential factors in the critical layer. The jumps across the critical layer in average vorticity, reflection and transmission coefficients are calculated as functions of the local Reynolds number determined by the amplitude of the incident wave. With the increase of the incident wave amplitude the asymptotic value of the Richardson number on the same side of critical layer as the incident wave tends to 1/4 the reflection coefficient tends to unity and the transmission coefficient to zero.

The paper discusses analytically the nonlinear wave reflection caused by a circular cylinder, submerged under a free surface in water of infinite depth. For monochromatic incident waves it is shown that there is no reflection of order m and frequency mω (m integer). This means that the dominant part of the mode of frequency mω is not reflected. For bichromatic incident waves it is found that the second-order wave with ‘sum frequency’ has no reflection. It is shown that the x- and y-components of the oscillatory force of order m and frequency mω have identical amplitudes and a phase-difference ½π. A corresponding result is also true for bichromatic waves.

Following Wallis, the relation between non-viscous bubbly flow and conduction in composites is examined. The bubbles are treated as incompressible and correspond to non-conducting inclusions. A simple relation is found between the effective conductivity and the energy coefficient which is agreement with previous calculations. It is shown that the energy coefficient is frame dependent and, in the frame of zero volumetric flux, is equal to the virtual mass density. Zuber's virtual mass density corresponds to the conductivity of the Hashin–Shtrikman coated-sphere geometry. This connection is exploited to extend Zuber's result to ellipsoidal bubbles. The hyperbolicity of effective equations derived from a variational principle is analysed for various bubble configurations. Without bubble clustering the equations are ill-posed (unstable). However, when the bubbles group into ellipsoidal clusters the resulting effective equations are well-posed for a wide range of parameter values.

Short-time Lyapunov exponent analysis is a new approach to the study of the stability properties of unsteady flows. At any instant in time the Lyapunov perturbations are the set of infinitesimal perturbations to a system state that will grow the fastest in the long term. Knowledge of these perturbations can allow one to determine the instability mechanisms producing chaos in the system. This new method should prove useful in a wide variety of chaotic flows. Here it is used to elucidate the physical mechanism driving weakly chaotic Taylor–Couette flow.

Three-dimensional, direct numerical simulations of axially periodic Taylor–Couette flow are used to study the transition from quasi-periodicity to chaos. A partial Lyapunov exponent spectrum for the flow is computed by simultaneously advancing the full solution and a set of perturbations. The axial wavelength and the particular quasi-periodic state are chosen to correspond to the most complete experimental studies of this transition. The computational results are consistent with available experimental data, both for the flow characteristics in the quasi-periodic regime and for the Reynolds number at which transition to chaos is observed.

The dimension of the chaotic attractor near onset is estimated from the Lyapunov exponent spectrum using the Kaplan–Yorke conjecture. This dimension estimate supports the experimental observation of low-dimensional chaos, but the dimension increases more rapidly away from the transition than is observed in experiments. Reasons for this disparity are given. Short-time Lyapunov exponent analysis is used to show that the chaotic state studied here is caused by a Kelvin–Helmholtz-type instability of the outflow boundary jet of the Taylor vortices.

A small-scale cyclonic vortex in a relatively broad valley tends to climb up and out of the valley in a cyclonic spiral about the centre, and when over a relatively broad hill it tends to climb toward the top in an anticyclonic spiral around the peak. This phenomenon is examined here through two-dimensional numerical simulations and rotating-tank experiments. The basic mechanism involved is shown to be the same as that which accounts for the northwest propagation of cyclones on a β-plane. This inviscid nonlinear effect is also shown to be responsible for the observed translationary motion of barotropic vortices in a free-surface rotating tank. The behaviour of isolated vortices is contrasted with that of vortices with non-vanishing circulation.

The stability and initial growth rate of disturbances on an annular film lining a cylindrical tube have been the focus of several previous works. The further development of thsse disturbances as they grow to form stable unduloids or liquid bridges is investigated by means of a thin-film integral model. The model is compared both with perturbation theories for early times, and a numerical solution of the exact equations (NEKTON) for later times. The thin-film model gave results that were in good agreement with solutions of the exact equations. The results show that linear perturbation theory can be used to give good estimates of the times for unduloid and liquid bridge formation. The success of the model derives from the dominant influence of narrow draining regions that feed into the growing unduloid, and these regions remain essentially one-dimensional throughout the growth of the instability.

The model is used to analyse the evolution of the liquid layer lining the small airways of the lung during a single breath. The timescales for formation of unduloids and liquid bridges are found to be short enough for the liquid layer to be in a virtually quasi-equilibrium state throughout the breathing cycle. This conclusion is only tentative, however, because the model assumes that the surface tension of the airway liquid lining does not change with changes in interfacial area despite the known presence of pulmonary surfactant.

A class of constants of motion is obtained for the time-dependent form of the Rayleigh stability equation, which provides stability criteria for velocity profiles with multiple inflexion points.

Measurements are reported of the velocity and concentration distributions both within and above two beds of sediment in oscillatory flow. The experiments were carried out in an oscillatory flow water tunnel, the velocities were measured with a laser-Doppler anemometer and the concentrations of sediment with resistance probes operating through the sidewall of the tunnel. The sediments studied consisted of nylon granules of median diameter 4.0 mm and Perspex of median diameter 0.7 mm. Most of the beds were plane for the tests with the 0.7 mm sediment and rippled for those with the 4 mm sediment.

The measured velocity profile could be divided into three regions: a central region in which the amplitude and phase of the velocity increased almost linearly with height and two outer regions in which the variation in velocity with height was much less rapid. It is suggested that at very high sediment transport rates the central region covers almost the entire depth of the moving bed but that at lower transport rates the outer regions are more significant.

The effect of sediment movement on the velocity distribution above the bed is very marked. Bed roughness length is increased and the velocity amplitude falls off more slowly with height than for fixed beds under similar conditions.

Within the bed the shear stress increases almost linearly with depth. Apparent viscosity also increases steadily with depth below the surface of the bed.

The measurements of concentrations are in good agreement with the results of other investigators in the region above the moving bed. Within the bed the time-mean concentration rises steadily, with distance below the initial bed surface, towards the limiting value for a stationary bed. The concentration record also shows a fluctuation during the course of the cycle at twice the frequency of the fundamental oscillation. The amplitude of this fluctuation in concentration decreases with depth below the initial bed level. The phase variation with height is close to that of zero velocity gradient, within the moving bed.

An inert solute is convected by a steady random velocity field, which is associated with flow through a heterogeneous porous formation. The log conductivity and the velocity are stationary random space functions. The log conductivity Y is assumed to be normal, with an isotropic two-point correlation of variance σY2 and of finite integral scale I. The solute cloud is of a finite input zone of lengthscale l. The transport is characterized with the aid of the spatial moments of the solute body. The effective dispersion coefficient is defined as half of the rate of change with time of the second spatial moment with respect to the centroid. Under the ergodic hypothesis, which is bound to be satisfied for l/I [Gt ] 1, the centroid moves with the mean velocity U and the longitudinal dispersion coefficient [dscr ]L tends to its constant, Fickian, limit. Under a Lagrangian first-order analysis in σY2 it has been found that [dscr ]L = σY2UI.

This study addresses the computation of the effective longitudinal dispersion coefficient for a finite input zone, for which ergodic conditions may not be satisfied. In this case the centroid trajectory and the second spatial moments are random variables. In line with a previous work (Dagan 1990) the effective dispersion coefficient DL is defined as half the rate of change of the expected value of the second spatial moment for large transport time. The aim of the study is to derive DL and its dependence upon l/I and in particular to determine the conditions under which it tends to the ergodic limit [dscr ]L. The computation is carried out separately for a thin body aligned with the mean flow and one transverse to it. In the first case it is found that DL is equal to zero, i.e. the streamlined body does not disperse in the mean. This result is explained by the correlation between the trajectories of the leading and trailing edges, respectively, once the latter reaches the position of the first. The relatively modest increase of the mean second spatial moment is effectively computed. In the case of a thin body initially transverse to the mean flow, DL may reach the ergodic limit [dscr ]L for a ratio l/I of the order 102. For smaller values, DL is found to be bounded from above, and its maximum depends on l but not on I. The uncertainty caused by the randomness of the velocity field is manifested in the trajectory of the centroid rather than in the effective dispersion.

Reaction in a scalar mixing layer in grid-generated turbulence is studied experimentally by doping half of the flow with nitric oxide and the other half with ozone. The flow conditions and concentrations are such that the chemical reaction is passive and the flow and chemical timescales are of the same order. Conserved scalar theory for such flows is outlined and further developed; it is used as a basis for presentation of the experimental results. Continuous measurements of concentration are limited in their spatial and temporal resolution but capture sufficient of their spectra for adequate second-order correlations to be made. Two components of velocity have been measured simultaneously with hot-wire anemometry. Conserved scalar mixing results, deduced from reacting and non-reacting measurements of concentration, show the independence of concentration level and concentration ratio expected for passive reacting flow. The results are subject to several limitations due to the necessary experimental compromises, but they agree generally with measurements made in thermal mixing layers. Reactive scalar statistics are consistent with the realizability constraints obtainable from conserved scalar theory where such constraints apply, and otherwise are generally found to lie between the conserved scalar theory limits for frozen and very fast chemistry. It is suggested that Toor's (1969) closure for the mean chemical reaction rate could be improved by interpolating between the frozen and equilibrium values for the covariance. The turbulent fluxes of the reactive scalars are found to approximately obey the gradient model but the value of the diffusivity is found to depend on the Damköhler number.

Impulsively started flow around a circular cylinder at various Reynolds numbers is studied by a deterministic hybrid vortex method. The key feature of the method consists in solving the viscous vorticity equation by interlacing a finite-difference method for diffusion and a vortex-in-cell method for convection. The vorticity is updated along the surface of the cylinder to satisfy the no-slip condition. The present method is basically different from previous applications of vortex methods, which are primarily in the context of random vortex algorithms. The Reynolds numbers of the flows under investigation range from 300 to 106. Numerical results are compared with analytical solutions at small times, and compared with finite-difference solutions and flow visualization results at relatively long times. Satisfactory agreement is found in the evolutions of the separation angles, wake lengths, surface pressure and drag coefficients, streamline patterns, and some velocities on the axis of symmetry behind the circular cylinder. The present hybrid vortex method is highly stable and suffers from little numerical diffusivity, yielding convincing numerical results for unsteady vortical flows at moderately high Reynolds numbers.

A numerical study is made of the flow past an impulsively started rotating and translating circular cylinder using a hybrid vortex method. The Reynolds number (Re) ranges from 103 to 106 while the (counter-rotating) rotating-to-translating speed ratio (α) is increased from 0 to 2. It is found that three basic patterns of vortex shedding can be identified according to the behaviour of the stagnation points associated with the first upper and the first lower vortices. Depending on the parameters Re and α, the rotation may favour the shedding of the first upper vortex, or the first lower vortex (typically at high Reynolds numbers). In a transition region, strong competition for shedding exists between the first two vortices in the form of double transposition of stagnation (closure) points associated with the two vortices. Time variations of lift coefficients characterize different shedding patterns; the cylinder may first experience a substantial maximal downward lift when the first shedding vortex is from the upper wake, or a maximal upward lift otherwise.

Numerical calculations are presented for the steady three-dimensional structure of thermal convection of a fluid with strongly temperature-dependent viscosity in a bottom-heated rectangular box. Viscosity is assumed to depend on temperature T as exp (− ET), where E is a constant; viscosity variations across the box r (= exp (E)) as large as 105 are considered. A stagnant layer or lid of highly viscous fluid develops in the uppermost coldest part of the top cold thermal boundary layer when r > rc1, where r = rc1 ≡ 1.18 × 103Rt0.309 and Rt is the Rayleigh number based on the viscosity at the top boundary. Three-dimensional convection occurs in a rectangular pattern beneath this stagnant lid. The planform consists of hot upwelling plumes at or near the centre of a rectangle, sheets of cold sinking fluid on the four sides, and cold sinking plume concentrations immersed in the sheets. A stagnant lid does not develop, i.e. convection involves all of the fluid in the box when r < rc1. The whole-layer mode of convection occurs in a three-dimensional bimodal pattern when r > rc2 = 3.84 × 106Rt−1.35. The planform of the convection is rectangular with the coldest parts of the sinking fluid and the hottest part of the upwelling fluid occurring as plumes at the four corners and at the centre of the rectangle, respectively. Both hot uprising plumes and cold sinking plumes have sheet-like extensions, which become more well-developed as r increases. The whole-layer mode of convection occurs as two-dimensional rolls when r < min (rc1, rc2). The Nusselt number Nu depends on the viscosity at the top surface more strongly in the regime of whole-layer convection than in the regime of stagnant-lid convection. In the whole-layer convective regime, Nu depends more strongly on the viscosity at the top surface than on the viscosity at the bottom boundary.

The nonlinear evolution of marginally unstable wave packets in rotating pipe flow is studied. These flows depend on two control parameters, which may be taken to be the axial Reynolds number R and a Rossby number, q. Marginal stability is realized on a curve in the (R, q)-plane, and we explore the entire marginal stability boundary. As the flow passes through any point on the marginal stability curve, it undergoes a supercritical Hopf bifurcation and the steady base flow is replaced by a travelling wave. The envelope of the wave system is governed by a complex Ginzburg–Landau equation. The Ginzburg–Landau equation admits Stokes waves, which correspond to standing modulations of the linear travelling wavetrain, as well as travelling wave modulations of the linear wavetrain. Bands of wavenumbers are identified in which the nonlinear modulated waves are subject to a sideband instability.

The two-dimensional flow of an unbounded rotating stratified fluid towards a link sink is studied. The initial-value problem of suddenly initiating the sink flow is solved in Laplace space for a non-diffusive, inviscid fluid using the linearized Boussinesq equations. The solution shows that the sink flow is established by inertio–gravity waves radiated from the sink and that the initial development of the flow depends critically on the ratio of the inertial frequency, f, to the buoyancy frequency, N. For f < N the flow collapses to a horizontal withdrawal layer structure. The final steady state resembles potential flow in which the vertical axis is shrunk by a factor of f/N with a superimposed azimuthal velocity. Viscous, diffusive and nonlinear effects are studied using scaling analysis. A classification scheme based on two parameters delineating various force balance regimes and giving the corresponding withdrawal layer thicknesses is presented. The results show that under certain conditions rotation may cause a thicker withdrawal layer than would be observed if there were no rotation.

The fine-scale structure of turbulence in a fully developed turbulent duct flow is examined by considering the three-dimensional velocity derivative field obtained from direct numerical simulations at two relatively small Reynolds numbers. The magnitudes of all mean-square derivatives (normalized by wall variables) increase with the Reynolds number, the increase being largest at the wall. These magnitudes are not consistent with the assumption of local isotropy except perhaps near the duct centre-line. When the assumption of local isotropy is relaxed to one of local axisymmetry, or invariance with respect to rotation about a coordinate axis (here chosen in the streamwise direction), satisfactory agreement is indicated by the data outside the wall region. Support for axisymmetry is demonstrated by anisotropy invariant maps of the dissipation and vorticity tensors. The departure from axisymmetry does not appear to be affected by the Reynolds number. Expressions are proposed for approximations to the average energy dissipation and components of the mean-square vorticity. These proposals should allow these quantities to be measured accurately, at least in the present flow.

Hydrodynamic modulation of wind waves by long surface waves in a wave tank is investigated, at wind speeds ranging from 1.5 to 10 m s−1. The results are compared with the linear, non-dissipative, theory of Longuet-Higgins & Stewart (1960), which describes the modulation of a group of short gravity waves due to straining of the surface by currents produced by the orbital motions of the long wave, and work done against the radiation stresses of the short waves. In most cases the theory is in good agreement with the experimental results when the short waves are not too steep, and the rate of growth due to the wind is relatively small. At the higher wind speeds, the effects of wind-wave growth, dissipation and wave-wave interactions are dominant.

The thermocapillary migration of a fluid particle in a tube, owing to an imposed axial temperature gradient, is studied theoretically for the case of steady, axisymmetric, creeping, translation in the absence of thermal convection and fluid particle distortion from sphericity, and for an insulated tube. Formulated with these assumptions, the migration is a linear Stokes flow which is separable into two fixed-fluid particle flow idealizations. One is the fluid motion in a quiescent continuous phase, owing only to the thermocapillary surface stress. This stress causes fluid streaming which exerts a lift force on the fluid particle in the direction of the warmer fluid described by a lift coefficient. The second idealization is uniform flow in the absence of thermocapillary forces. The force on the fluid particle owing to this flow represents the hydrodynamic resistance to the forward motion of the fluid particle in the presence of the tube wall, and is described in terms of a drag coefficient. Numerical solutions for the lift and drag coefficients are obtained by a boundary collocation technique.

The migration velocity in the tube relative to that under identical conditions in an infinite medium is computed from the ratio of the lift and drag coefficients. The calculations show that for a fixed ratio of the sphere to the tube diameter, as the conductivity of the fluid particle phase decreases relative to that of the continuous phase, a greater proportion of energy is conducted through the gap between the insulated tube wall and the fluid particle. This conduction pattern creates a larger surface temperature gradient, and causes the relative migration velocity to increase. The enhancement in migration for decreasing fluid particle conductivity at a fixed ratio of the sphere to the tube diameter becomes more pronounced as the later ratio increases and the surface gradient intensifies. However, as the gap distance between the sphere and the tube decreases, hydrodynamic retarding forces develop, and these forces are overriding in the sense that the relative migration velocity in the tube decreases monotonically from the value of one as the gap thickness decreases, and therefore the migration velocity in the tube never exceeds the value in an infinite medium.

The dynamics of coherent structure interactions, in particular the jet column mode of vortex pairing, in the near field of an elliptic jet have been investigated using hotwire measurements and flow visualization. A 2:1 aspect-ratio jet with an initially laminar boundary layer and a constant momentum thickness all around the nozzle exit perimeter is used for this study. While detailed hot-wire measurements were made in air at a Reynolds number ReDe (≡UeDe/ν) = 3.2 × 104, flow visualization was performed in water at a lower ReDe = 1.7 × 104; here Ue is the exit speed and De is the equivalent diameter of the nozzle exit cross-section. Excitation at the stable pairing mode induced successive pairings to occur periodically at the same location, allowing phase-locked measurements using a local trigger sensor. Coherent structures were educed at different phases of pairing in the planes of both the major and minor axes. These are compared with corresponding data in a circular jet, educed similarly.

Pairing interactions are found to be quite different from those in a circular jet. Owing to non-planar and non-uniform self-induction of elliptical vortical structures and the consequent effect on mutual induction, pairing of elliptic vortices in the jet column does not occur uniformly around the entire perimeter, unlike in a circular jet. Merger occurs only in the initial major-axis plane through an entanglement process, while in the initial minor-axis plane, the trailing vortex rushes through the leading vortex without pairing and then breaks down violently. These motions produce considerably greater entrainment and mixing than in circular or plane jets. From distributions of dynamical properties over the extent of coherent structures, the production mechanism is explained in terms of the longitudinal vortices (or ribs) connecting the elliptic structures. Time-average measures and their modification by controlled excitation are also discussed in terms of coherent structure dynamics. A significant space in this paper is devoted to documenting phase-dependent and time-average flow measures; these new results should serve as target data for numerical simulations. Further details are given in Husain (1984).