We study the classical Landau–Levich dip-coating problem for the case in which the interface possesses both elasticity and surface tension. The aim of the study is to develop a complete asymptotic theory of the elastocapillary Landau–Levich problem in the limit of small flow speeds. As such, the paper also extends our previous study on purely elastic Landau–Levich flow (Dixit & Homsy J. Fluid Mech., vol. 732, 2013, pp. 5–28) to include the effect of surface tension. The elasticity of the interface is described by the Helfrich model and surface tension is modelled in the usual way. We define an elastocapillary number, $\epsilon $, which represents the relative strength of elasticity to surface tension. Based on the size of $\epsilon $, we can define three different regimes of interest. In each of these regimes, we carry out asymptotic expansions in the small capillary (or elasticity) numbers, which represents the balance of viscous forces to surface tension (or elasticity).

In the weak elasticity regime, the film thickness is a small correction to the classical Landau–Levich law and can be written as

$$\begin{eqnarray*}{\tilde {h} }_{\infty , c} = (0. 9458- 0. 0839~\mathscr{E}){l}_{c} C{a}^{2/ 3} , \quad \epsilon \ll 1,\end{eqnarray*}$$

${l}_{c} $

$Ca$

$\mathscr{E}= \epsilon / C{a}^{2/ 3} $

$\epsilon $

$$\begin{eqnarray*}{\tilde {h} }_{\infty , ec} = {\bar {h} }_{\infty , e} L\hspace{0.167em} f(\epsilon )C{a}^{4/ 7} , \quad \epsilon \sim O(1),\end{eqnarray*}$$

${\bar {h} }_{\infty , e} $

$L$

$f(\epsilon )$

In this paper we derive consistent shallow-water equations for the flow of thin films of power-law fluids down an incline. These models account for the streamwise diffusion of momentum, which is important to describe accurately the full dynamics of thin-film flows when instabilities such as roll waves arise. These models are validated through a comparison with the Orr–Sommerfeld equations for large-scale perturbations. We consider only laminar flow for which the boundary layer issued from the interaction of the flow with the bottom surface has an influence all over the transverse direction to the flow. In this case the concept itself of a thin film and its relation with long-wave asymptotics leads naturally to flow conditions around a uniform free-surface Poiseuille flow. The apparent viscosity diverges at the free surface, which, in turn, introduces a singularity in the formulation of the Orr–Sommerfeld equations and in the derivation of shallow-water models. We remove this singularity by introducing a weaker formulation of the Cauchy momentum equations. No regularization procedure is needed, nor any distinction between shear thinning and thickening cases. Our analysis, though, is only valid when the flow behaviour index $n$ is larger than $1/ 2$, and strongly suggests that the Cauchy momentum equations are ill-posed if $n\leq 1/ 2$.

In this paper, we consider shock reflection problems, occurring in supersonic and hypersonic intake flow under off-design conditions, in which the incident shock wave is disturbed by the lip-generated upstream expansion wave and the reflected shock wave intersects with a downstream cowl-turning deflected shock wave. The expansion wave and deflected shock wave are here generated with the same magnitude of flow deflection angle or turning angle. With the help of shock interaction theory and numerical simulation, the influence of the turning angle of the lip and cowl on the flow structure and the critical conditions for transition between regular reflection and Mach reflection are analysed. It is found that the dual-solution domain is significantly altered by the interference between the expansion wave and shock waves. The flow structure in the condition of Mach reflection is then analysed with a model updated from a previous study. It is shown that the Mach stem height is an increasing function of the turning angle, while the horizontal position of the Mach stem is shifted in the downstream direction for small turning angle and in the upstream direction for large turning angle.

Conditions are investigated under which a body lying at rest or rocking on a solid horizontal surface can be removed from the surface by hydrodynamic forces or instead continues rocking. The investigation is motivated by recent observations on Martian dust movement as well as other small- and large-scale applications. The nonlinear theory of fluid–body interaction here has unsteady motion of an inviscid fluid interacting with a moving thin body. Various shapes of body are addressed together with a range of initial conditions. The relevant parameter space is found to be subtle as evolution and shape play substantial roles coupled with scaled mass and gravity effects. Lift-off of the body from the surface generally cannot occur without fluid flow but it can occur either immediately or within a finite time once the fluid flow starts up: parameters for this are found and comparisons are made with Martian observations.

The role played by gas compressibility in gas-cushioned liquid–solid impacts is investigated within a viscous gas and inviscid liquid regime. A full analysis of the energy conservation in the gas is conducted for the first time, which indicates that both thermal diffusion across the gas film and viscous dissipation play an important role in gas cushioning once gas compression becomes significant. Consequently existing models of gas compressibility based on either an isothermal or an adiabatic equation of state for the gas do not fully reflect the physics associated with this phenomenon. Models incorporating thermal diffusion and viscous dissipation are presented, which are appropriate for length scales consistent with droplet impacts, and for larger scale liquid–solid impacts. The evolution of the free surface is calculated alongside the corresponding pressure, temperature and density profiles. These profiles indicate that a pocket of gas can become trapped during an impact. Differences between the new model and older models based on isothermal and adiabatic equations of state are discussed, along with predictions of the size of the trapped gas pocket.

An electrospun polymer jet issued from a Taylor cone follows a straight-line motion before experiencing electrical bending instability resulting in curling and spiralling of the jet in three-dimensional space. Experiments are performed to characterize the fluid dynamics of an electrospun polymer jet. Appropriate image processing is performed to systematically analyse flow regimes of the electrospun jet. These regimes include Taylor cone formation/jet initiation and the straight section of the jet. Dimensional analysis was performed to identify the salient dimensionless parameters, which govern the electrospun jet characteristics. Three new correlation formulae were obtained to characterize the dimensionless jet diameter at the apex of the Taylor cone $(\tilde {d} = 1. 03{\tilde {Q} }^{0. 44} )$, the dimensionless jet diameter at different locations along the jet’s straight section $(\tilde {d} {\tilde {z} }^{1/ 4} = 1. 09{\tilde {Q} }^{1/ 2} )$, as well as the length of the straight section of the jet $({\tilde {Z} }_{in} = 86{\tilde {Q} }^{0. 42} )$. These correlation formulae are valid for the analysed range of dimensionless flow rates $(2. 6{{\times 10}}^{- 4} \lt \tilde {Q} \lt 3. 6{{\times 10}}^{7} )$ and dimensionless electric fields $(7. 4{{\times 10}}^{- 4} \lt \tilde {E} \lt 1. 4{{\times 10}}^{- 1} )$. In addition, the correlation formulae are valid for the analysed range of Deborah numbers De and Reynolds numbers Re based on nozzle radius, $3. 3\times {10}^{- 7} \lt {\mathit{De}}_{{r}_{o} } \lt 3. 8\times {10}^{- 2} $ and $5. 8\times {10}^{- 4} \lt {\mathit{Re}}_{{r}_{o} } \lt 7. 0\times {10}^{- 1} $. The proposed new correlation formulae are instrumental in the design as well as controlled manipulation/optimization of the electrospinning phenomenon.

In experiments of hot surface ignition and subsequent flame propagation, a puffing flame instability is observed in mixtures that are stagnant and premixed prior to ignition. By varying the size of the hot surface, power input, and combustion vessel volume, it was determined that the instability is a function of the interaction of the flame, with the fluid flow induced by the combustion products rather than the initial plume established by the hot surface. Pressure ranges from 25 to 100 kPa and mixtures of n-hexane/air with equivalence ratios between $\phi = 0. 58$ and 3.0 at room temperature were investigated. Equivalence ratios between $\phi = 2. 15$ and 2.5 exhibited multiple flame and equivalence ratios above $\phi = 2. 5$ resulted in puffing flames at atmospheric pressure. The phenomenon is accurately reproduced in numerical simulations and a detailed flow field analysis revealed competition between the inflow velocity at the base of the flame and the flame propagation speed. The increasing inflow velocity, which exceeds the flame propagation speed, is ultimately responsible for creating a puff. The puff is then accelerated upward, allowing for the creation of the subsequent instabilities. The frequency of the puff is proportional to the gravitational acceleration and inversely proportional to the flame speed. A scaling relationship describes the dependence of the frequency on gravitational acceleration, hot surface diameter, and flame speed. This relation shows good agreement for rich n-hexane/air and lean hydrogen/air flames, as well as lean hexane/hydrogen/air mixtures.

In nature, water waves usually propagate in groups and the open question about the characteristics of the highest possible wave in a group is of significant theoretical and practical interest. We examine the problem of the highest non-breaking wave in a wave group by direct numerical simulations of the exact Euler equations. The main aim of the study is twofold: (i) to describe the highest wave in a group in fully nonlinear setting and find its dependence on parameters; (ii) to examine correspondence between the exact breather solutions of weakly nonlinear analytic theory based on the integrable nonlinear Schrödinger (NLS) equation and their strongly nonlinear analogues. In contrast to weakly nonlinear models the very notion of the highest wave is ill-defined: the maximal crest elevation, the maximal trough-to-crest height and the deepest trough all occur at close but different moments; correspondingly, we have to speak about distinctively different extreme waves. In the simulations small initial perturbation of a uniform wave train were prescribed in a way ensuring that the initial perturbation excites a single breather-type modulation. The ensuing growth results in higher wave magnitudes and takes longer time to develop compared with the NLS theory. The maxima of crest elevation noticeably exceed their weakly nonlinear analogues. The wave with the highest crest differs significantly from the unmodulated wave: the local wavelength contracts considerably, the crest becomes noticeably higher; the vicinity of the crest of such an extreme wave is close to that of the limiting Stokes periodic wave. Thus, the shape of the maximal crest wave is almost universal, i.e. it practically does not depend on the way the wave group evolved, or even whether there was initially more than one group. The evolution of a single NLS breather has been shown to have a qualitatively similar but quantitatively quite different analogue in the fully nonlinear setting. The one-to-one mapping of the NLS breather solutions onto fully nonlinear ones has been constructed. The fully nonlinear breathers are found to be robust, which provides grounds for applying the results for developing short-term deterministic forecasting of rogue waves.

A second-order instability analysis has been performed for sinuous disturbances on two-dimensional planar viscous sheets moving in a stationary gas medium using a perturbation technique. The solutions of second-order interface disturbances have been derived for both temporal instability and spatial instability. It has been found that the second-order interface deformation of the fundamental sinuous wave is varicose or dilational, causing disintegration and resulting in ligaments which are interspaced by half a wavelength. The interface deformation has been presented; the breakup time for temporal instability and breakup length for spatial instability have been calculated. An increase in Weber number and gas-to-liquid density ratio extensively increases both the temporal or spatial growth rate and the second-order initial disturbance amplitude, resulting in a shorter breakup time or length, and a more distorted surface deformation. Under normal conditions, viscosity has a stabilizing effect on the first-order temporal or spatial growth rate, but it plays a dual role in the second-order disturbance amplitude. The overall effect of viscosity is minor and complicated. In the typical condition, in which the Weber number is 400 and the gas-to-liquid density ratio is 0.001, viscosity has a weak stabilizing effect when the Reynolds number is larger than 150 or smaller than 10; when the Reynolds number is between 150 and 10, viscosity has a weak destabilizing effect.

Turbulent mixing in a Richtmyer–Meshkov unstable light–heavy–light (air–${\mathrm{SF} }_{6} $–air) fluid layer subjected to a shock (Mach 1.20) and a reshock (Mach 1.14) is investigated using ensemble statistics obtained from simultaneous velocity–density measurements. The mixing is driven by an unstable array of initially symmetric vortices that induce rapid material mixing and create smaller-scale vortices. After reshock the flow appears to transition to a turbulent (likely three-dimensional) state, at which time our planar measurements are used to probe the developing flow field. The density self-correlation $b= - \langle \rho v\rangle $ (where $\rho $ and $v$ are the fluctuating density and specific volume, respectively) and terms in its evolution equation are directly measured experimentally for the first time. Amongst other things, it is found that production terms in the $b$ equation are balanced by the dissipation terms, suggesting a form of equilibrium in $b$. Simultaneous velocity measurements are used to probe the state of the incipient turbulence. A length-scale analysis suggests that an inertial range is beginning to form, consistent with the onset of a mixing transition. The developing turbulence is observed to reduce non-Boussinesq effects in the flow, which are found to be small over much of the layer after reshock. Second-order two-point structure functions of the density field exhibit a power-law behaviour with a steeper exponent than the standard $2/ 3$ power found in canonical turbulence. The absence of a significant $2/ 3$ region is observed to be consistent with the state of the flow, and the emergence of the steeper power-law region is discussed.

Flow past a circular cylinder executing sinusoidal rotary oscillations about its own axis is studied experimentally. The experiments are carried out at a Reynolds number of 185, oscillation amplitudes varying from $\mathrm{\pi} / 8$ to $\mathrm{\pi} $, and at non-dimensional forcing frequencies (ratio of the cylinder oscillation frequency to the vortex-shedding frequency from a stationary cylinder) varying from 0 to 5. The diagnostic is performed by extensive flow visualization using the hydrogen bubble technique, hot-wire anemometry and particle-image velocimetry. The wake structures are related to the velocity spectra at various forcing parameters and downstream distances. It is found that the phenomenon of lock-on occurs in a forcing frequency range which depends not only on the amplitude of oscillation but also the downstream location from the cylinder. The experimentally measured lock-on diagram in the forcing amplitude and frequency plane at various downstream locations ranging from 2 to 23 diameters is presented. The far-field wake decouples, after the lock-on at higher forcing frequencies and behaves more like a regular Bénard–von Kármán vortex street from a stationary cylinder with vortex-shedding frequency mostly lower than that from a stationary cylinder. The dependence of circulation values of the shed vortices on the forcing frequency reveals a decay character independent of forcing amplitude beyond forcing frequency of ${\sim }1. 0$ and a scaling behaviour with forcing amplitude at forcing frequencies ${\leq }1. 0$. The flow visualizations reveal that the far-field wake becomes two-dimensional (planar) near the forcing frequencies where the circulation of the shed vortices becomes maximum and strong three-dimensional flow is generated as mode shape changes in certain forcing parameter conditions. It is also found from flow visualizations that even at higher Reynolds number of 400, forcing the cylinder at forcing amplitudes of $\mathrm{\pi} / 4$ and $\mathrm{\pi} / 2$ can make the flow field two-dimensional at forcing frequencies greater than ${\sim }2. 5$.

The cross-flow instability that arises in swept-wing boundary layers has resisted attempts to describe the path from disturbance initiation to transition. Following concerted research efforts, surface roughness and free-stream turbulence have been identified as the leading providers of initial disturbances for cross-flow instability growth. Although a significant body of work examines the role of free-stream turbulence in the cross-flow problem, the data more relevant to the flight environment (turbulence intensities less than 0.07 %) are sparse. A series of recent experiments indicates that variations within this range may affect the initiation or growth of cross-flow instability amplitudes, hindering comparison among results obtained in different disturbance environments. To address this problem, a series of wind tunnel experiments is performed in which the free-stream turbulence intensity is varied between 0.02 % and 0.2 % of free-stream velocity, ${U}_{\infty } $. Measurements of the stationary and travelling mode amplitudes are made in the boundary layer of a 1.83 m chord, $45{{}^\circ} $ swept-wing model. These results are compared to those of similar experiments at higher turbulence levels to broaden the current knowledge of this portion of the cross-flow problem. It is observed that both free-stream turbulence and surface roughness contribute to the initiation of unsteady disturbances, and that free-stream turbulence affects the development of both stationary and unsteady cross-flow disturbances. For the range tested, enhanced free-stream turbulence advances the transition location except when a subcritically spaced roughness array is employed.

The positive and negative high-amplitude pressure peaks (HAPP) are investigated in a turbulent boundary layer at $R{e}_{\theta } = $ 1900 in order to identify their turbulent structure. The three-dimensional velocity field is measured within the inner layer of the turbulent boundary layer using tomographic particle image velocimetry (tomo-PIV). The measurements are performed at an acquisition frequency of 10 000 Hz and over a volume of $418\times 149\times 621$ wall units in the streamwise, wall-normal and spanwise directions, respectively. The time-resolved velocity fields are applied to obtain the material derivative using the Lagrangian method followed by integration of the Poisson pressure equation to obtain the three-dimensional unsteady pressure field. The simultaneous volumetric velocity, acceleration, and pressure data are conditionally sampled based on local maxima and minima of wall pressure to analyse the three-dimensional turbulent structure of the HAPPs. Analysis has associated the positive HAPPs to the shear layer structures formed by an upstream sweep of high-speed flow opposing a downstream ejection event. The sweep event is initiated in the outer layer while the ejection of near-wall fluid is formed by the hairpin category of vortices. The shear layers were observed to be asymmetric in the instantaneous visualizations of the velocity and acceleration fields. The asymmetric pattern originates from the spanwise component of temporal acceleration of the ejection event downstream of the shear layer. The analysis also demonstrated a significant contribution of the pressure transport term to the budget of the turbulent kinetic energy in the shear layers. Investigation of the conditional averages and the orientation of the vortices showed that the negative HAPPs are linked to both the spanwise and quasi-streamwise vortices of the turbulent boundary layer. The quasi-streamwise vortices can be associated with the hairpin category of vortices or the isolated quasi-streamwise vortices of the inner layer. A bi-directional analysis of the link between the HAPPs and the hairpin paradigm is also conducted by conditionally averaging the pressure field based on the detection of hairpin vortices using strong ejection events. The results demonstrated positive pressure in the shear layer region of the hairpin model and negative pressure overlapping with the vortex core.

We investigate the evolution and stability of a wetting viscous fluid layer flowing down the surface of a cylinder, and surrounded by a conductive gas. The inner cylinder is an electrode kept at constant voltage, and a second, concentric electrode encloses the system whose potential is allowed to vary spatially. This induces electrostatic forces at the interface in competition with surface tension and viscous stresses. Asymptotic methods are used to derive a long-wave axisymmetric model governing the interfacial position and charge density. The resulting system of equations is investigated both analytically and numerically to determine its stability characteristics in the linear and nonlinear regimes.

It has been shown that it is possible to measure the turbulent diffusivity of a magnetic field by a method involving oscillatory sources. So far the method has only been tried in the special case of two-dimensional fields and flows. Here we extend the method to three dimensions and consider the case where the flow is thermally driven convection in a large-aspect-ratio domain. We demonstrate that if the diffusing field is horizontal the method is successful even if the underlying flow can sustain dynamo action. We show that the resulting turbulent diffusivity is comparable with, although not exactly the same as, that of a passive scalar. We were not able to measure unambiguously the diffusivity if the diffusing field is vertical, but argue that such a measurement is possible if enough resources are utilized on the problem.

We present a theoretical investigation of the self-diffusiophoresis of a class of two-faced, two-dimensional Janus particles propelled by the production of gradients in the concentration of a solute diffusing into a surrounding fluid at zero Reynolds and Péclet numbers. Those concentration gradients produce a tangential boundary slip resulting in translation and rotation of the particle, as a consequence of the fact that it is free of both force and torque. The model Janus particles studied here have piecewise constant surface mobilities and surface activities over two faces. An isolated circular Janus particle is studied first and its speed of locomotion in free space is found analytically. Confinement effects are then investigated by placing such a Janus particle near a straight no-slip wall. The governing nonlinear dynamical system is found in explicit form. It is used to study how the geometry, location and orientation of the particle relative to the wall affect its motion. It is found that if the particles do not hit the wall in finite time they are eventually repelled away from it.

In this paper, the concept of available potential energy (APE) density is extended to a multicomponent Boussinesq fluid with a nonlinear equation of state. As shown by previous studies, the APE density is naturally interpreted as the work against buoyancy forces that a parcel needs to perform to move from a notional reference position at which its buoyancy vanishes to its actual position; because buoyancy can be defined relative to an arbitrary reference state, so can APE density. The concept of APE density is therefore best viewed as defining a class of locally defined energy quantities, each tied to a different reference state, rather than as a single energy variable. An important result, for which a new proof is given, is that the volume-integrated APE density always exceeds Lorenz’s globally defined APE, except when the reference state coincides with Lorenz’s adiabatically re-arranged reference state of minimum potential energy. A parcel reference position is systematically defined as a level of neutral buoyancy (LNB): depending on the nature of the fluid and on how the reference state is defined, a parcel may have one, none, or multiple LNB within the fluid. Multiple LNB are only possible for a multicomponent fluid whose density depends on pressure. When no LNB exists within the fluid, a parcel reference position is assigned at the minimum or maximum geopotential height. The class of APE densities thus defined admits local and global balance equations, which all exhibit a conversion to kinetic energy, a production term by boundary buoyancy fluxes, and a dissipation term by internal diffusive effects. Different reference states alter the partition between APE production and dissipation, but neither affects the net conversion between kinetic energy and APE, nor the difference between APE production and dissipation. We argue that the possibility of constructing APE-like budgets based on reference states other than Lorenz’s reference state is more important than has been previously assumed, and we illustrate the feasibility of doing so in the context of an idealized and realistic oceanic example, using as reference states one with constant density and another one defined as the horizontal-mean density field; in the latter case, the resulting APE density is found to be a reasonable approximation of the APE density constructed from Lorenz’s reference state, while being computationally cheaper.

The effect of insoluble surfactant on the interfacial waves in connection with a two-layer channel flow is investigated for low to moderate values of the Reynolds number. Previous studies focusing on Stokes flow (Frenkel & Halpern, Phys. Fluids, vol. 14, 2002, p. L45; Halpern & Frenkel, J. Fluid Mech., vol. 485, 2003, pp. 191–220) are extended by including the inertial effect and the study of low-Reynolds-number flow (Blyth & Pozrikidis, J. Fluid Mech., vol. 521, 2004b, pp. 241–250) is enlarged up to moderate Reynolds number. Linear stability analysis based on the Orr–Sommerfeld boundary value problem identifies a surfactant mode together with an interface mode. The presence of surfactant on the interfacial mode is stabilizing at high viscosity ratio and destabilizing at low viscosity ratio. The threshold of instability is determined as a function of the Marangoni number. A long-wave model is developed to predict the families of nonlinear waves in the neighbourhood of the threshold of instability. Far from the threshold, wave dynamics is explored under the framework of a three-equation model in terms of lower layer flow rate ${q}_{2} (x, t)$, lower liquid-layer thickness $h(x, t)$ and surfactant concentration $\Gamma (x, t)$. Primary instability analysis of a three-equation model captures the result of the Orr–Sommerfeld boundary value problem very well for quite large values of wavenumber. In the nonlinear regime, travelling wave solutions demonstrate deceleration of maximum amplitude and acceleration of speed with the Marangoni number at high viscosity ratio $m\gt 1$ and show completely the opposite behaviour at low viscosity ratio $m\lt 1$. However, both maximum amplitude and speed attain a fixed value with increasing Reynolds number and this leads to saturation of instability.

The flow transformation and the generation of vortex structures by a strong magnetic dipole field in a liquid metal duct flow is studied by means of three-dimensional direct numerical simulations. The dipole is considered as the paradigm for a magnetic obstacle which will deviate the streamlines due to Lorentz forces acting on the fluid elements. The duct is of square cross-section. The dipole is located above the top wall and is centred in spanwise direction. Our model uses the quasistatic approximation which is applicable in the limit of small magnetic Reynolds numbers. The analysis covers the stationary flow regime at small hydrodynamic Reynolds numbers $\mathit{Re}$ as well as the transitional time-dependent regime at higher values which may generate a turbulent flow in the wake of the magnetic obstacle. We present a systematic study of these two basic flow regimes and their dependence on $\mathit{Re}$ and on the Hartmann number $\mathit{Ha}$, a measure of the strength of the magnetic dipole field. Furthermore, three orientations of the dipole are compared: streamwise-, spanwise- and wall-normal-oriented dipole axes. The most efficient generation of turbulence at a fixed distance above the duct follows for the spanwise orientation, which is caused by a certain configuration of Hartmann layers and reversed flow at the top plate. The enstrophy in the turbulent wake grows linearly with $\mathit{Ha}$ which is connected with a dominance of the wall-normal derivative of the streamwise velocity.

The three-dimensional structure of incompressible flow in a narrow, open rectangular cavity in a flat plate was investigated with a focus on the flow topology of the time-averaged flow. The ratio of cavity length (in the direction of the flow) to width to depth was $l{: }w{: }d= 6{: }2{: }1$. Experimental surface pressure data (in air) and particle image velocimetry data (in water) were obtained at low speed with free-stream Reynolds numbers of ${\mathit{Re}}_{l} = 3. 4\times 1{0}^{5} $ in air and ${\mathit{Re}}_{l} = 4. 3\times 1{0}^{4} $ in water. The experimental results show that the three-dimensional cavity flow is of the ‘open’ type, with an overall flow structure that bears some similarity to the structure observed in nominally two-dimensional cavities, but with a high degree of three-dimensionality both in the flow near the walls and in the unsteady behaviour. The defining features of an open-type cavity flow include a shear layer that traverses the entire cavity opening ultimately impinging on the back surface of the cavity, and a large recirculation zone within the cavity itself. Other flow features that have been identified in the current study include two vortices at the back of the cavity, of which one is barely visible, a weak vortex at the front of the cavity, and a pair of counter-rotating streamwise vortices along the sides of the cavity near the cavity opening. These vortices are generally symmetric about the cavity centre-plane. However, the discovery of a single tornado vortex, located near the cavity centreline at the front of the cavity, indicated that the flow within the cavity is asymmetric. It is postulated that the observed asymmetry in the time-averaged flow field is due to the asymmetry in the instantaneous flow field, which switches between two extremes at large time scales.