The streamwise vortices generated by dielectric-barrier-discharge plasma actuators in the laminar boundary layer were investigated using particle image velocimetry to understand the vortex-formation mechanisms. The plasma vortex generator was oriented along the primary flow direction to produce a body force in the spanwise direction. This created a spanwise-directed wall jet which interacted with the oncoming boundary layer to form a coherent streamwise vortex. It was found that the streamwise vortices were formed by the twisting and folding of the spanwise vorticity in the oncoming boundary layer into the outer shear layer of the spanwise wall jet, which added its own vorticity to increase the circulation along the actuator length. This is similar to the delta-shaped, vane-type vortex generator, except that the circulation was enhanced by the addition of the vorticity in the plasma jet. It was also observed that the plasma vortex was formed close to the wall with an enhanced wall-ward entrainment, which created strong downwash above the actuator.

Spanwise arrays of miniature vortex generators (MVGs) are used to generate energetic transient disturbance growth, which is able to modulate the boundary layer flow with steady and stable streak amplitudes up to 32 % of the free-stream velocity. This type of modulation has previously been shown to act in a stabilizing manner on modal disturbance growth described by classical instability theory. In an attempt to reproduce a more realistic flow configuration, in the present experimental set-up, Tollmien–Schlichting (TS) waves are generated upstream of the MVG array, allowing for a complete interaction of the incoming wave with the array. Fifteen new MVG configurations are investigated and the stabilizing effect on the TS waves is quantified. We show that the streak amplitude definition is very important when trying to relate it to the stabilization, since it may completely bypass information on the mean streamwise velocity gradient in the spanwise direction, which is an essential ingredient of the observed stabilization. Here, we use an integral-based streak amplitude definition along with a streak amplitude scaling relation based on empiricism, which takes the spanwise periodicity of the streaks into account. The results show that, applying the integral definition, the optimal streak amplitude for attenuating TS wave disturbance growth is around 30 % of the free-stream velocity, which corresponds to ${\sim }20\hspace{0.167em} \% $ in the conventional definition when keeping the spanwise wavelength constant. The experiments also show that the disturbance energy level, based on the full velocity signal, is significantly reduced in the controlled case, and that the onset of transition may be inhibited altogether throughout the measured region in the presence of an MVG array.

A recently proposed stabilization mechanism for the Rayleigh–Taylor instability, using magnetic fluids and azimuthally rotating magnetic fields, is experimentally investigated in a cylindrical geometry and compared with the theoretical model. This approach allows the imperfection of the experimental setup to be exploited for measuring the critical field strength of the instability without ever reaching the supercritical state. Furthermore, we use a fast increase in the magnetic field strength to prevent an already occurring instability and force the system back to its initial state. In this way we measure the growth dynamics repeatedly and acquire the characteristic time scale ${\tau }_{0} $ of the instability.

Active linear control is applied to delay the onset of laminar–turbulent transition in the boundary layer over a flat plate. The analysis is carried out by numerical simulations of the nonlinear, transitional regime. A three-dimensional, localized initial condition triggering Tollmien–Schlichting waves of finite amplitude is used to numerically simulate the transition to turbulence. Linear quadratic Gaussian controllers based on reduced-order models of the linearized Navier–Stokes equations are designed, where the wall sensors and the actuators are localized in space. A parametric analysis is carried out in the nonlinear regime, for different disturbance amplitudes, by investigating the effects of the actuation on the flow due to different distributions of the localized actuators along the spanwise direction, different sizes of the actuators and the effort of the controllers. We identify the range of parameters where the controllers are effective and highlight the limits of the device for high amplitudes and strong control action. Despite the fully linear control approach, it is shown that the device is effective in delaying the onset of laminar–turbulent transition in the presence of packets characterized by amplitudes $a\approx 1\hspace{0.167em} \% $ of the free stream velocity at the actuator location. Up to these amplitudes, it is found that a proper choice of the actuators positively affects the performance of the controller. For a transitional case, $a\approx 0. 20\hspace{0.167em} \% $, we show a transition delay of $\Delta {\mathit{Re}}_{x} = 3. 0\times 1{0}^{5} $.

The control of Tollmien–Schlichting waves in a two-dimensional boundary layer is analysed using numerical simulations. Full-dimensional optimal controllers are used in combination with a setup of spatially localized inputs (actuator and disturbance) and outputs (sensors). The adjoint of the direct-adjoint (ADA) algorithm, recently proposed by Pralits & Luchini (In Seventh IUTAM Symposium on Laminar–Turbulent Transition (ed. P. Schlatter & D. S. Henningson), vol. 18, 2010, Springer), is used to efficiently compute an optimal controller known as a linear quadratic regulator; the method is iterative and allows one to bypass the solution of the corresponding Riccati equation, which is infeasible for high-dimensional systems. We show that an analogous iteration can be made for the estimation problem; the dual algorithm is referred to as adjoint of the adjoint-direct (AAD). By combining the solutions of the estimation and control problem, full-dimensional linear quadratic Gaussian controllers are obtained and used for the attenuation of the disturbances arising in the boundary layer flow. The full-dimensional controllers turn out to be an excellent benchmark for evaluating the performance of the optimal control/estimation design based on reduced-order models. We show under which conditions the two strategies are in perfect agreement by focusing on the issues arising when feedback configurations are considered. An analysis of the finite-amplitude disturbances is also carried out by addressing the limitations of the optimal controllers, the role of the estimation, and the robustness to the nonlinearities arising in the flow of the control design.

The formation, growth and instability of a pair of counter-rotating vortices over a circular plate in the downstream of a thin fishing line were studied using particle image velocimetry and flow visualization. Initially, the vortex pair in an axisymmetric stagnation flow was small, but it grew steadily by accumulating the shear-layer vorticity of the wake before going through vortical instability. Two types of vortical development were observed in the present experiment. Type I was a common type of vortical development in an axisymmetric stagnation flow over a circular plate. Here, the circulation of the vortex pair increased linearly with time reflecting a constant flux of vorticity impinging on the plate wall. After the growth, the counter-rotating pair of vortices went through an antisymmetric deformation in the wall-normal direction while the vortex deformation was symmetric in the wall-parallel direction. This was remarkably similar to the short-wavelength elliptic instability of counter-rotating vortices in an open system. On the other hand, type II development of a vortex pair was a rare case, where the vortices grew for much longer duration than in type I cases. This initiated a breakdown of vortices before the residual vorticity moved away from the centre of the plate. It is considered that the disturbance due to vortical instability could be partially responsible for the unexpectedly high heat transfer rate in the stagnation region of bluff bodies that has been reported in the last half-century.

Geophysical fluid models often support both fast and slow motions. As the dynamics are often dominated by the slow motions, it is desirable to filter out the fast motions by constructing balance models. An example is the quasi-geostrophic (QG) model, which is used widely in meteorology and oceanography for theoretical studies, in addition to practical applications such as model initialization and data assimilation. Although the QG model works quite well in the mid-latitudes, its usefulness diminishes as one approaches the equator. Thus far, attempts to derive similar balance models for the tropics have not been entirely successful as the models generally filter out Kelvin waves, which contribute significantly to tropical low-frequency variability. There is much theoretical interest in the dynamics of planetary-scale Kelvin waves, especially for atmospheric and oceanic data assimilation where observations are generally only of the mass field and thus do not constrain the wind field without some kind of diagnostic balance relation. As a result, estimates of Kelvin wave amplitudes can be poor. Our goal is to find a balance model that includes Kelvin waves for planetary-scale motions. Using asymptotic methods, we derive a balance model for the weakly nonlinear equatorial shallow-water equations. Specifically we adopt the ‘slaving’ method proposed by Warn et al. (Q. J. R. Meteorol. Soc., vol. 121, 1995, pp. 723–739), which avoids secular terms in the expansion and thus can in principle be carried out to any order. Different from previous approaches, our expansion is based on a long-wave scaling and the slow dynamics is described using the height field instead of potential vorticity. The leading-order model is equivalent to the truncated long-wave model considered previously (e.g. Heckley & Gill, Q. J. R. Meteorol. Soc., vol. 110, 1984, pp. 203–217), which retains Kelvin waves in addition to equatorial Rossby waves. Our method allows for the derivation of higher-order models which significantly improve the representation of Rossby waves in the isotropic limit. In addition, the ‘slaving’ method is applicable even when the weakly nonlinear assumption is relaxed, and the resulting nonlinear model encompasses the weakly nonlinear model. We also demonstrate that the method can be applied to more realistic stratified models, such as the Boussinesq model.

We study both theoretically and numerically two-dimensional magnetohydrodynamic turbulence at infinite and zero magnetic Prandtl number $\mathit{Pm}$ (and the limits thereof), with an emphasis on solution regularity. For $\mathit{Pm}= 0$, both $\Vert \omega \Vert ^{2} $ and $\Vert j\Vert ^{2} $, where $\omega $ and $j$ are, respectively, the vorticity and current, are uniformly bounded. Furthermore, $\Vert \boldsymbol{\nabla} j\Vert ^{2} $ is integrable over $[0, \infty )$. The uniform boundedness of $\Vert \omega \Vert ^{2} $ implies that in the presence of vanishingly small viscosity $\nu $ (i.e. in the limit $\mathit{Pm}\rightarrow 0$), the kinetic energy dissipation rate $\nu \Vert \omega \Vert ^{2} $ vanishes for all times $t$, including $t= \infty $. Furthermore, for sufficiently small $\mathit{Pm}$, this rate decreases linearly with $\mathit{Pm}$. This linear behaviour of $\nu \Vert \omega \Vert ^{2} $ is investigated and confirmed by high-resolution simulations with $\mathit{Pm}$ in the range $[1/ 64, 1] $. Several criteria for solution regularity are established and numerically tested. As $\mathit{Pm}$ is decreased from unity, the ratio $\Vert \omega \Vert _{\infty } / \Vert \omega \Vert $ is observed to increase relatively slowly. This, together with the integrability of $\Vert \boldsymbol{\nabla} j\Vert ^{2} $, suggests global regularity for $\mathit{Pm}= 0$. When $\mathit{Pm}= \infty $, global regularity is secured when either $\Vert \boldsymbol{\nabla} \boldsymbol{u}\Vert _{\infty } / \Vert \omega \Vert $, where $\boldsymbol{u}$ is the fluid velocity, or $\Vert j\Vert _{\infty } / \Vert j\Vert $ is bounded. The former is plausible given the presence of viscous effects for this case. Numerical results over the range $\mathit{Pm}\in [1, 64] $ show that $\Vert \boldsymbol{\nabla} \boldsymbol{u}\Vert _{\infty } / \Vert \omega \Vert $ varies slightly (with similar behaviour for $\Vert j\Vert _{\infty } / \Vert j\Vert $), thereby lending strong support for the possibility $\Vert \boldsymbol{\nabla} \boldsymbol{u}\Vert _{\infty } / \Vert \omega \Vert \lt \infty $ in the limit $\mathit{Pm}\rightarrow \infty $. The peak of the magnetic energy dissipation rate $\mu \Vert j\Vert ^{2} $ is observed to decrease rapidly as $\mathit{Pm}$ is increased. This result suggests the possibility $\Vert j\Vert ^{2} \lt \infty $ in the limit $\mathit{Pm}\rightarrow \infty $. We discuss further evidence for the boundedness of the ratios $\Vert \omega \Vert _{\infty } / \Vert \omega \Vert $, $\Vert \boldsymbol{\nabla} \boldsymbol{u}\Vert _{\infty } / \Vert \omega \Vert $ and $\Vert j\Vert _{\infty } / \Vert j\Vert $ in conjunction with observation on the density of filamentary structures in the vorticity, velocity gradient and current fields.

We study the degree to which Kraichnan–Leith–Batchelor (KLB) phenomenology describes two-dimensional energy cascades in $\alpha $ turbulence, governed by $\partial \theta / \partial t+ J(\psi , \theta )= \nu {\nabla }^{2} \theta + f$, where $\theta = {(- \Delta )}^{\alpha / 2} \psi $ is generalized vorticity, and $\hat {\psi } (\boldsymbol{k})= {k}^{- \alpha } \hat {\theta } (\boldsymbol{k})$ in Fourier space. These models differ in spectral non-locality, and include surface quasigeostrophic flow ($\alpha = 1$), regular two-dimensional flow ($\alpha = 2$) and rotating shallow flow ($\alpha = 3$), which is the isotropic limit of a mantle convection model. We re-examine arguments for dual inverse energy and direct enstrophy cascades, including Fjørtoft analysis, which we extend to general $\alpha $, and point out their limitations. Using an $\alpha $-dependent eddy-damped quasinormal Markovian (EDQNM) closure, we seek self-similar inertial range solutions and study their characteristics. Our present focus is not on coherent structures, which the EDQNM filters out, but on any self-similar and approximately Gaussian turbulent component that may exist in the flow and be described by KLB phenomenology. For this, the EDQNM is an appropriate tool. Non-local triads contribute increasingly to the energy flux as $\alpha $ increases. More importantly, the energy cascade is downscale in the self-similar inertial range for $2. 5\lt \alpha \lt 10$. At $\alpha = 2. 5$ and $\alpha = 10$, the KLB spectra correspond, respectively, to enstrophy and energy equipartition, and the triad energy transfers and flux vanish identically. Eddy turnover time and strain rate arguments suggest the inverse energy cascade should obey KLB phenomenology and be self-similar for $\alpha \lt 4$. However, downscale energy flux in the EDQNM self-similar inertial range for $\alpha \gt 2. 5$ leads us to predict that any inverse cascade for $\alpha \geq 2. 5$ will not exhibit KLB phenomenology, and specifically the KLB energy spectrum. Numerical simulations confirm this: the inverse cascade energy spectrum for $\alpha \geq 2. 5$ is significantly steeper than the KLB prediction, while for $\alpha \lt 2. 5$ we obtain the KLB spectrum.

We use ultra-high-speed video imaging to look at the initial contact of a drop impacting on a liquid layer. We observe experimentally the vortex street and the bubble-ring entrapments predicted numerically, for high impact velocities, by Thoraval et al. (Phys. Rev. Lett., vol. 108, 2012, article 264506). These dynamics mainly occur within $50~\mathrm{\mu} \mathrm{s} $ after the first contact, requiring imaging at 1 million f.p.s. For a water drop impacting on a thin layer of water, the entrapment of isolated bubbles starts through azimuthal instability, which forms at low impact velocities, in the neck connecting the drop and pool. For Reynolds number $Re$ above ${\sim }12\hspace{0.167em} 000$, up to 10 partial bubble rings have been observed at the base of the ejecta, starting when the contact is ${\sim }20\hspace{0.167em} \% $ of the drop size. More regular bubble rings are observed for a pool of ethanol or methanol. The video imaging shows rotation around some of these air cylinders, which can temporarily delay their breakup into micro-bubbles. The different refractive index in the pool liquid reveals the destabilization of the vortices and the formation of streamwise vortices and intricate vortex tangles. Fine-scale axisymmetry is thereby destroyed. We show also that the shape of the drop has a strong influence on these dynamics.

The interaction of barotropic tidal currents and baroclinic geostrophic eddies is considered theoretically and numerically to determine whether energy can be transferred to an internal wave field by this process. The eddy field evolves independently of the tide, suggesting that it acts catalytically in facilitating energy transfer from the barotropic tide to the internal wave field, without exchanging energy with the other flow components. The interaction is identically zero and no waves are generated when the barotropic tidal current is horizontally uniform. Optimal internal wave generation occurs when the scales of tide and eddy fields satisfy resonant conditions. The most efficient generation is found if the tidal current horizontal scale is comparable to that of the eddies, with a weak maximum when the scales differ by a factor of two. Thus, this process is not an effective mechanism for internal wave excitation in the deep ocean, where tidal current scales are much larger than those of eddies, but it may provide an additional source of internal waves in coastal areas where horizontal modulation of the tide by topography can be significant.

We investigate the route to dissipation in strongly stratified and rotating systems through high-resolution numerical simulations of the Boussinesq equations (BQs) and the primitive equations (PEs) in a triply periodic domain forced at large scales. By applying geostrophic scaling to the BQs and using the same horizontal length scale in defining the Rossby and the Froude numbers, $\mathit{Ro}$ and $\mathit{Fr}$, we show that the PEs can be obtained from the BQs by taking the limit ${\mathit{Fr}}^{2} / {\mathit{Ro}}^{2} \rightarrow 0$. When ${\mathit{Fr}}^{2} / {\mathit{Ro}}^{2} $ is small the difference between the results from the BQ and the PE simulations is shown to be small. For large rotation rates, quasi-geostrophic dynamics are recovered with a forward enstrophy cascade and an inverse energy cascade. As the rotation rate is reduced, a fraction of the energy starts to cascade towards smaller scales, leading to a shallowing of the horizontal spectra from ${ k}_{h}^{- 3} $ to ${ k}_{h}^{- 5/ 3} $ at the small-scale end. The vertical spectra show a similar transition as the horizontal spectra and we find that Charney isotropy is approximately valid also at larger wavenumbers than the transition wavenumber. The high resolutions employed allow us to capture both ranges within the same simulation. At the transition scale, kinetic energy in the rotational and in the horizontally divergent modes attain comparable values. The divergent energy is several orders of magnitude larger than the quasi-geostrophic divergent energy given by the $\Omega $-equation. The amount of energy cascading downscale is mainly controlled by the rotation rate, with a weaker dependence on the stratification. A larger degree of stratification favours a downscale energy cascade. For intermediate degrees of rotation and stratification, a constant energy flux and a constant enstrophy flux coexist within the same range of scales. In this range, the enstrophy flux is a result of triad interactions involving three geostrophic modes, while the energy flux is a result of triad interactions involving at least one ageostrophic mode, with a dominant contribution from interactions involving two ageostrophic and one geostrophic mode. Dividing the ageostrophic motions into two classes depending on the sign of the linear wave frequency, we show that the energy transfer is for the largest part supported by interactions within the same class, ruling out the wave–wave–vortex resonant triad interaction as a mean of the downscale energy transfer. The role of inertia-gravity waves is studied through analyses of time-frequency spectra of single Fourier modes. At large scales, distinct peaks at frequencies predicted for linear waves are observed, whereas at small scales no clear wave activity is observed. Triad interactions show a behaviour which is consistent with turbulent dynamics, with a large exchange of energy in triads with one small and two large comparable wavenumbers. The exchange of energy is mainly between the modes with two comparable wavenumbers.

The flow field around an asymmetric dielectric-barrier-discharge (DBD) plasma actuator in quiescent air is studied using particle image velocimetry (PIV) and smoke-flow visualization. On initiation of DBD plasma a starting vortex is created, which rolls up to form a coherent structure. The starting vortex becomes self-similar when the maximum velocity induced by the DBD plasma actuator reaches a steady state. Here, the plasma jet momentum increases linearly with time, suggesting that the DBD plasma actuator entrains and accelerates the surrounding fluid with a constant force. The wall-parallel and wall-normal distances of the vortex core are observed to scale with ${t}^{2/ 3} $ as it travels at $3{1}^{\circ } $ to the wall. The velocity of the starting vortex is found to scale with ${t}^{- 1/ 3} $, while the circulation induced by the plasma actuator scales with ${t}^{1/ 3} $.

A quasi-linear theory is presented for how randomly forced, barotropic velocity fluctuations cause an exponentially growing, large-scale (mean) magnetic dynamo in the presence of a uniform parallel shear flow. It is a ‘kinematic’ theory for the growth of the mean magnetic energy from a small initial seed, neglecting the saturation effects of the Lorentz force. The quasi-linear approximation is most broadly justifiable by its correspondence with computational solutions of nonlinear magnetohydrodynamics, and it is rigorously derived in the limit of small magnetic Reynolds number, . Dynamo action occurs even without mean helicity in the forcing or flow, but random helicity variance is then essential. In a sufficiently large domain and with a small seed wavenumber in the direction perpendicular to the mean shearing plane, a positive exponential growth rate can occur for arbitrary values of , viscous Reynolds number , and random-force correlation time and orientation angle in the shearing plane. The value of is independent of the domain size. The shear dynamo is ‘fast’, with finite in the limit of . Averaged over random realizations of the forcing history, the ensemble-mean magnetic field grows more slowly, if at all, compared to the r.m.s. field (magnetic energy). In the limit of small and , the dynamo behaviour is related to the well-known alpha–omega ansatz when the force is slowly varying () and to the ‘incoherent’ alpha–omega ansatz when the force is more rapidly fluctuating.

A rarefied gas flow thermally induced around a heated (or cooled) flat plate, contained in a vessel, is considered in two different situations: (i) both sides of the plate are simultaneously and uniformly heated (or cooled); and (ii) only one side of the plate is uniformly heated. The former is known as the thermal edge flow and the latter, typically observed in the Crookes radiometer, may be called the radiometric flow. The steady behaviour of the gas induced in the container is investigated on the basis of the Bhatnagar–Gross–Krook (BGK) model of the Boltzmann equation and the diffuse reflection boundary condition by means of an accurate finite-difference method. The flow features are clarified for a wide range of the Knudsen number, with a particular emphasis placed on the structural similarity between the two flows. The limiting behaviour of the flow as the Knudsen number tends to zero (and thus the system approaches the continuum limit) is investigated for both flows. The detailed structure of the normal stress on the plate as well as the cause of the radiometric force (the force acting on the plate from the hotter to the colder side) is also clarified for the present infinitely thin plate.

Previous studies of shear dispersion in bounded velocity fields have assumed random velocities with zero Lagrangian time scale (i.e. velocities are -function correlated in time). However, many turbulent (geophysical and engineering) flows with mean velocity shear exist where the Lagrangian time scale is non-zero. Here, the longitudinal (along-flow) shear-induced diffusivity in a two-dimensional bounded velocity field is derived for random velocities with non-zero Lagrangian time scale . A non-zero results in two-time transverse (across-flow) displacements that are correlated even for large (relative to the diffusive time scale ) times. The longitudinal (along-flow) shear-induced diffusivity is derived, accurate for all , using a Lagrangian method where the velocity field is periodically extended to infinity so that unbounded transverse particle spreading statistics can be used to determine . The non-dimensionalized depends on time and two parameters: the ratio of Lagrangian to diffusive time scales and the release location. Using a parabolic velocity profile, these dependencies are explored numerically and through asymptotic analysis. The large-time is enhanced relative to the classic Taylor diffusivity, and this enhancement increases with . At moderate this enhancement is approximately a factor of 3. For classic shear dispersion with , the diffusive time scale determines the time dependence and large-time limit of the shear-induced diffusivity. In contrast, for sufficiently large , a shear time scale , anticipated by a simple analysis of the particle’s domain-crossing time, determines both the time dependence and the large-time limit. In addition, the scalings for turbulent shear dispersion are recovered from the large-time using properties of wall-bounded turbulence.

Deep water waves in two-dimensional flow can have curvature singularities on the surface profile; for example, the limiting Stokes wave has a corner of radians and the limiting standing wave momentarily forms a corner of radians. Much less is known about the possible formation of curvature singularities in general. A novel way of exploring this possibility is to consider the curvature as a complex function of the complex arclength variable and to seek the existence and nature of any singularities in the complex arclength plane. Highly accurate boundary integral methods produce a Fourier spectrum of the curvature that allows the identification of the nearest singularity to the real axis of the complex arclength plane. This singularity is in general a pole singularity that moves about the complex arclength plane. It approaches the real axis very closely when waves break and is associated with the high curvature at the tip of the breaking wave. The behaviour of these singularities is more complex for standing waves, where two singularities can be identified that may collide and separate. One of them approaches the real axis very closely when a standing wave forms a very narrow collapsing column of water almost under free fall. In studies so far, no singularity reaches the real axis in finite time. On the other hand, the surface elevation has square-root singularities in the complex plane that do reach the real axis in finite time, the moment when a wave first starts to break. These singularities have a profound effect on the wave spectra.

The flow over a pair of counter-rotating cylinders is investigated numerically and experimentally. It is demonstrated that it is possible to suppress unsteady vortex shedding for gap sizes from one to five cylinder diameters, at Reynolds numbers from 100 to 200, expanding on the more limited work by Chan & Jameson (Intl J. Numer. Meth. Fluids, vol. 63, 2010, p. 22). The degree of unsteady wake suppression is proportional to the speed and the direction of rotation, and there is a critical rotation rate where a complete suppression of flow unsteadiness can be achieved. In the doublet-like configuration at higher rotational speeds, a virtual elliptic body that resembles a potential doublet is formed, and the drag is reduced to zero. The shape of the elliptic body primarily depends on the gap between the two cylinders and the speed of rotation. Prior to the formation of the elliptic body, a second instability region is observed, similar to that seen in studies of single rotating cylinders. It is also shown that the unsteady wake suppression can be achieved by rotating each cylinder in the opposite direction, that is, in a reverse doublet-like configuration. This tends to minimize the wake interaction of the cylinder pair and the second instability does not make an appearance over the range of speeds investigated here.

We experimentally investigate the dynamics of surface waves excited by oscillations from a cylindrical sidewall. Particle-imaging-velocimetry measurements with fluorescent particles were used to determine the flow patterns near the sidewall of the cylindrical fluid container and to identify the locations of the evolving air–water interfaces. The high-frequency wall oscillations created four jets that originate at the cylindrical sidewall. Four vortex streets shed from the jets propagate from the sidewall to the centre of the container and subsequently excite a low-frequency gravity wave. The interaction between this gravitational surface wave and the high-frequency capillary waves was found to be responsible for creating droplet splash at the water surface. This phenomenon was first described as ‘Long-Xi’ or ‘dragon wash’ in ancient China. The physical processes for generating the droplet ejection, including the circular capillary waves, azimuthal waves, streaming jets and low-frequency gravity waves, are described in this paper.

We perform direct numerical simulation to study the transport of gas and heat as passive scalars in free-surface turbulence. Our analysis focuses on the surface age of surface fluid particles, i.e. the time elapsed since the last surface renewal they experienced. Using Lagrangian tracing of fluid particles combined with heat diffusion analysis, we are able to directly quantify surface age to illustrate scalar characteristics at different stages of interfacial transfer. Results show that at the early stage of surface renewal, vertical advection associated with upwellings greatly enhances surface gas flux; random surface renewal model does not apply at this stage when most of the interfacial gas transfer occurs. After a fluid particle leaves the upwelling region, it may enter a nearby downwelling region immediately, where the gas flux is sharply reduced but the variation in surface temperature is small; alternatively, the fluid particle may travel along the surface for some time before it is absorbed by a downwelling, where the surface temperature has changed significantly due to long duration of diffusion and the gas flux is also reduced. To gain further insight into the relationships between surface velocity and scalar quantities, we perform a statistical analysis of upwellings using clustering and nonlinear regression. With this analysis, we are able to provide qualitative and quantitative descriptions of the skewed probability density functions associated with the surface divergence, temperature and gas flux that support our physics-based investigation of surface renewal and surface age.