Research ArticlePHYSICS

Turbulence generation through an iterative cascade of the elliptical instability

See allHide authors and affiliations

Science Advances  28 Feb 2020:
Vol. 6, no. 9, eaaz2717
DOI: 10.1126/sciadv.aaz2717

Abstract

The essence of turbulent flow is the conveyance of energy through the formation, interaction, and destruction of eddies over a wide range of spatial scales—from the largest scales where energy is injected down to the smallest scales where it is dissipated through viscosity. Currently, there is no mechanistic framework that captures how the interactions of vortices drive this cascade. We show that iterations of the elliptical instability, arising from the interactions between counter-rotating vortices, lead to the emergence of turbulence. We demonstrate how the nonlinear development of the elliptical instability generates an ordered array of antiparallel secondary filaments. The secondary filaments mutually interact, leading to the formation of even smaller tertiary filaments. In experiments and simulations, we observe two and three iterations of this cascade, respectively. Our observations indicate that the elliptical instability could be one of the fundamental mechanisms by which the turbulent cascade develops.

INTRODUCTION

Understanding how turbulent flows develop and organize has puzzled scientists and engineers for centuries (1). The foundational characterization of turbulent flow began with Reynolds over a century ago (2) and was quickly followed by rigorous statistical interpretations of how turbulent flows develop (35). In 1937, Taylor and Green (6) introduced an initial flow condition that produces a cascade of energy from large to small scales. Subsequently, Kolmogorov postulated that turbulent flows exhibit universal behaviors over many length scales. Kolmogorov (7) predicted that within an inertial subrange, the energy spectrum of a turbulent flow has a universal, self-similar form, wherein the energy scales as the inverse 5/3 power of the wave number k. Kolmogorov’s energy cascade has been observed in a plethora of experimental systems and numerical simulations, from wind tunnels to river beds [see, e.g., fig. 13 of (8)].

The efficient conveyance of energy from the large scales, where it is injected, to the small scales, where it is dissipated, is at the heart of how complex, three-dimensional (3D) flows are maintained. It is thus critical to understand how small-scale flow structures are formed and maintained at high Reynolds numbers. Despite major progress in providing an effective statistical description of turbulent flows (9), our understanding of the mechanisms by which interactions between eddies are mediated remains limited. The explanations of how this occurs in real space are often abstract and “poetic” (4, 10, 11).

The temporal development of the turbulent cascade remains one of the most intriguing mysteries in fluid mechanics. In particular, it is not well understood what specific mechanisms lead to the development of large velocity gradients in turbulent flows. These large velocity gradients, which derive from the interactions of turbulent eddies, amplify the kinetic energy dissipation rate, ϵ, in a manner that is independent of the fluid viscosity in the high–Reynolds number limit (5, 12). This implies the existence of an inertial mechanism by which vortices locally interact to convey energy across scales such that the statistical properties of the energy cascade develop in accordance with the scaling laws established by Kolmogorov. Note that the limit to how fast an initially regular flow may produce extremely large velocity gradients is also a celebrated mathematical problem (13). It is therefore of great interest to look directly for elementary flow configurations of interacting vortices that begin smooth and rapidly develop into turbulence. This approach was implemented by Lundgren (14), who analytically examined the breakdown of a single vortex under axial strain, bursting into an ensemble of helical vortex bundles. While this configuration has been observed to lead to the development of a turbulent flow, it requires the presence of a particular, large-scale strain configuration acting on an isolated vortex (15). We implement a more general flow configuration that typifies the fundamental components of the turbulent cascade: the collision of two identical vortices. Recent numerical and experimental works demonstrate that the breakdown of colliding vortex rings at intermediate Reynolds numbers gives rise to small-scale flow structures (16), mediated by the iterative flattening and splitting of the vortex cores to smaller and smaller filaments (17, 18).

Here, we revisit the emergence of a turbulent burst of fine-scale flow structures that results from the violent, head-on collision of two coherent vortex rings (16, 19). This classical configuration is a unique model system for probing the development of turbulence without any rigid boundaries or large-scale constraints. We show that for high Reynolds numbers, the violent breakdown of the colliding vortex rings into a turbulent “soup” of interacting vortices is mediated by the elliptical instability. During the late-stage, nonlinear development of the elliptical instability, an ordered array of antiparallel secondary vortex filaments emerges perpendicular to the collision plane. Locally, these pairs of counter-rotating secondary filaments spawn another generation of tertiary vortex filaments, resulting in the expeditious formation of a hierarchy of vortices over many scales. Our numerical simulations show that at this stage of the breakdown, the interacting tangle of vortices reaches a turbulent state, such that the energy spectrum of the flow exhibits Kolmogorov scaling. We observe both experimentally and numerically how the elliptical instability precipitates the onset of turbulence, generating and maintaining the means by which the energy of the flow cascades from large to small scales.

RESULTS

Vortex ring collisions

The geometry of the experimental setup is depicted schematically in Fig. 1A. Two identical counter-rotating vortex rings are fired head-on in a 75-gallon aquarium filled with deionized water, as shown in movie S1. The vortex rings are formed via a piston-cylinder configuration in which a slug fluid with viscosity, ν, is pushed through a cylinder of diameter, D (2.54 cm), at a constant velocity, U, with a stroke, L. The resulting flow is controlled by two dimensionless parameters: the Reynolds number, Re = UD/ν, and the stroke ratio, SR = L/D (20). Fluorescent dye (Rhodamine B) is injected into the cores of the rings as they are formed. Since the collision occurs at a fixed plane in the laboratory frame, this configuration is attractive for directly observing the rapid formation of small-scale flow structures. The dynamics and eventual breakdown of the dyed cores are visualized in full 3D by imaging over the collision plane with a scanning laser sheet (λ = 532 nm), which is pulsed synchronously with a high-speed camera (Phantom V2511). The technical details of how the vortex rings are formed and visualized in 3D are described in previous work (16) and in Materials and Methods below. In addition, we perform direct numerical simulations (DNSs) of interacting vortices at Reynolds numbers equivalent to the experiments (see section S1 for how the definitions of the Reynolds numbers in simulations and experiments compare).

Fig. 1 Vortex ring collisions.

(A) Schematic side-view showing the formation and collision of dyed vortex rings in experiments. Fluorescent dye (Rhodamine B) is injected into the core of the vortex via a thin gap in the orifice of the vortex cannon. The dashed horizontal line denotes the symmetry axis. (B) Vortex ring radius versus rescaled time for collisions at various Reynolds numbers. Both cores are dyed, the core centerlines are extracted from 3D reconstructions, and the centerlines are fitted to circles with a fixed center point. The initial time begins when the vortex rings enter the scanning volume and ends when the vortex cores break down. Inset: All experimental curves shifted by t to collapse. The radial growth of the rings coincides with the Biot-Savart prediction.

As the vortex rings collide, they exert mutual strains on one another, causing them to stretch radially at a constant velocity before breaking down at a terminal radius, as shown in Fig. 1B. At low Reynolds numbers, Re ≲ 5000, the dyed cores break down, ejecting a tiara of secondary vortex rings or smoky turbulent puffs (16, 19) at approximately six times the initial radius, R0. The initial vortex ring radius and core radius, σ, were measured separately through particle image velocimetry (PIV), as described in section S1. Notably, for collisions at higher Reynolds numbers, Re ≳ 5000, the cores “burst” into an amorphous turbulent cloud of dye at a maximum radius of approximately 5 R0, indicating the onset of a different breakdown mechanism at this high–Reynolds number regime. The mean radial growth of the colliding rings is well described by the Biot-Savart model (21), as described in section S2 (A and B). In addition, the radial expansion of the rings is encapsulated by a universal functional form, as shown in the inset of Fig. 1B.

Onset of the elliptical instability: Vortex core perturbations

While the mean radial growth of the colliding vortex rings follows the same linear evolution at any Reynolds number, the cores themselves develop different forms of perturbations, due to their mutual interaction. The formation of these perturbations can arise from two different types of instabilities. The Crow instability (22) causes the cores to develop symmetric circumferential perturbations with long wavelengths, much larger than the core radius, σ. This instability stems from the mutual advection of the interacting vortices (23) and governs the breakdown dynamics for collisions at lower Reynolds numbers (16, 19, 23). The nonlinear development of the Crow instability causes the rings to deflect into one another and form “tent-like” structures (16, 17, 24), which interact locally at the collision plane.

At higher Reynolds numbers, both our experiments and simulations show that the breakdown dynamics are governed by the elliptical instability, causing the vortex cores to develop short-wavelength perturbations on the order of the core radius (see section S2C) (2527). This instability originates from the parametric excitation of Kelvin modes in the vortex cores due to the resonant interaction of the strain field from the other vortex (23, 28). A hallmark of the elliptical instability, these short-wavelength perturbations grow synchronously in an antisymmetric manner, as shown for two typical experimental and numerical examples in Fig. 2 (A and B).

Fig. 2 Antisymmetric perturbations in vortex ring collisions.

A montage of core centerline trajectories for vortex ring collisions in both (A) experiment and (B) DNS. The top (z > 0) cores are indicated by the red lines, and the bottom (z < 0) cores are indicated by the blue lines. For the experimental collision, Re = 7000, SR = 2, and R0 = 17.5 mm. For the DNS collision, ReΓ = Γ/ν = 4500 and σ = 0.1R0. The cores are segmented from the 3D flow visualization in the experimental collision and from the pressure distribution in the simulation. Mean core separation distance versus rescaled time for the same (C) experimental and (D) numerical collisions. The blue circles correspond to the trajectories in (A) and (B), and the red dashed lines correspond to the visualizations in the insets. (C, inset) 3D visualization of the dyed vortex cores in the experimental collision. (D, inset) 3D visualization of the dyed vortex rings in the simulation, showing both the dye in the cores (dark) and surrounding them (light).

As the elliptical instability grows radially along the collision plane, the mean spacing between the cores, d, decreases linearly. However, the separation distance between the cores saturates when the perturbations deflect out of plane, just before breaking down, as shown in Fig. 2 (C and D) and movie S2. The minimum mean spacing between the cores is approximately equal to twice the initial core radius, σ. For the experiment, σ = 0.14 ± 0.01 R0, and for the simulation, σ = 0.1 R0 (see section S1). After the elliptical instability develops and the symmetry of the two cores is broken, a periodic array of satellite flow structures is shed from each core, bridging the gap between them, as shown in the inset of Fig. 2D and movie S3. Notably, in our experiments, the emergence of these secondary flow structures can only be resolved if the fluid outside of the cores is dyed.

Late-stage development of the elliptical instability: Secondary filaments

To better resolve the late-stage development of the elliptical instability and the resulting breakdown, we dye the full vortex rings, as shown in Fig. 3 (A to F) and movie S4. Observing the fully dyed vortex rings reveals the intricate structure of the flow that develops in response to the core dynamics. The antisymmetric coupling of the perturbations breaks the azimuthal symmetry of the flow, leading to the exchange of fluid between the two rings. This periodic wrapping of dye causes the outer layers of the rings to interdigitate around one another along alternating “tongues” (29), as shown in Fig. 3 (A and B). At the boundaries of adjacent tongues, the dyed fluid curls into vortex filaments, perpendicular to the cores, as shown in Fig. 3C. These alternating filaments are stretched by the circulating vortex cores into an array of counter-rotating secondary vortices, as shown in Fig. 3 (D and E). The secondary filaments have a fleeting lifetime of only tens of milliseconds before they break down. Violent interactions between the secondary filaments and primary cores result in the rapid ejection of fine-scale vortices and the formation of a turbulent cloud, as shown in Fig. 3F.

Fig. 3 Formation of perpendicular secondary filaments in an experimental vortex ring collision.

3D reconstruction of two fully dyed vortex rings colliding head-on, viewed from overhead (top) and from the side (bottom). Re = 6000 and SR = 2.5. (A to C) As the rings grow, they interdigitate as the dye from the upper ring is wrapped around the lower ring and vice versa. (D and E) The colliding rings form an array of secondary vortex filaments that are perpendicular to the vortex cores. (F) The cores and perpendicular filaments break down into a fine-scale turbulent cloud.

By performing DNSs of the colliding vortices, we additionally probe how energy is transferred through the flow via the onset of the elliptical instability. Since the breakdown of the vortices is localized to the area around the cores, we implement a new configuration for the simulations, which consists of two initially parallel, counter-rotating vortex tubes with circulation Γ, initially spaced a distance, b = 2.5σ, apart. The flow is simulated in a cubic domain of side length, ℒ = 16.67σ, and the Reynolds number of this configuration is given by ReΓ = Γ/ν (see Materials and Methods below). From PIV measurements, we find that ReΓ ≈ 0.678Re as shown in section S1. The dynamics of the vorticity distribution in the simulated flow are qualitatively equivalent to the experimental flow visualizations, as shown for a typical example at ReΓ = 4500 in Fig. 4 (A to C) and movie S5. When the antisymmetric perturbations resulting from the elliptical instability materialize, the tips of the perturbed cores deform into flattened vortex sheets, as illustrated in Fig. 4A. These sheets are stretched by strains applied by the other core and roll up along the edges into an alternating series of hairpin vortices, as shown in Fig. 4B and section S3A. Upon stretching across the gap to the other perturbed core, these hairpin vortices form an ordered array of secondary vortex filaments perpendicular to the initial tubes, as shown in Fig. 4C. Adjacent pairs of secondary filaments counter-rotate relative to one another (29), as shown in Fig. 4D. Integrating the transverse vorticity along the symmetry plane, we find that, as the secondary filaments are stretched, approximately 25% of the streamwise circulation from the initial vortex tubes is conveyed to each filament (see section S3B). As the vorticity of the original tubes is transferred to the secondary filaments, the circulation of the flow is conserved.

Fig. 4 Generation of perpendicular secondary filaments in DNS.

(A to C) Vorticity modulus for simulated interacting tubes where ReΓ = 4500, σ = 0.06ℒ, b = 2.5σ, and t* = Γt/b2. The vorticity modulus is normalized by the maximum vorticity modulus during the simulation, ∣ω∣max. (A) The initial antisymmetric perturbations of the cores develop as the tips of the perturbations locally flatten (0.103 ≤ ∣ω∣/∣ω∣max ≤ 0.117). (B) At the same time, low-vorticity perpendicular filaments form as a result of the perturbations (0.046 ≤ ∣ω∣/∣ω∣max ≤ 0.092). (C) Once the secondary filaments form, their vorticity amplifies (0.076 ≤ ∣ω∣/∣ω∣max ≤ 0.114). (D) Vorticity distribution in the z direction along the center plane (z = 0) indicated by the dashed line in (C). Adjacent secondary filaments counter-rotate.

Interactions of secondary vortices

Once formed, each pair of secondary filaments can be locally viewed as a replica of the initial flow configuration on a smaller scale and with a reduced circulation, hence corresponding to a smaller effective Reynolds number. The resulting close-range interactions of neighboring filaments can lead to an iterative cascade by which even more generations of small-scale vortices are formed. For collisions at moderately high Reynolds numbers (e.g., ReΓ = 3500), the concentrated strains exerted by the counter-rotating secondary filaments cause one of them to flatten into an extremely thin vortex sheet and split into two smaller tertiary vortex filaments, as shown in movies S6 and S7 and section S4. This behavior is consistent with the breakdown mechanism observed experimentally in the head-on collision of vortex rings at comparatively lower Reynolds numbers mediated by the Crow instability (16).

In the high–Reynolds number limit, the secondary filaments may give rise to another generation of perpendicular tertiary vortex filaments, as shown for a typical example at ReΓ = 6000 in Fig. 5 (A to F) and movies S8 and S9. The secondary filaments are drawn into one another because of their mutual counter-rotation, as shown in the close-ups in Fig. 5 (B and C). The narrow gap between these writhing vortices, which experiences intense strain, almost instantaneously becomes enveloped by several high-vorticity tertiary filaments, as shown in Fig. 5 (D and E). These tertiary filaments align perpendicular to the previous generation of vortices, wrapping tightly around them. The tertiary filaments develop locally in an ordered manner, while the remnants of the primary cores and secondary filaments become increasingly entangled into a disordered soup of vortices, as shown in Fig. 5F.

Fig. 5 The development of a turbulent cascade.

(A to F) Vorticity modulus for simulated interacting tubes where ReΓ = 6000, σ = 0.06ℒ, b = 2.5σ, and t* = Γt/b2. Each panel shows the front view of the full cores (left) and a close-up top view of the interacting secondary filaments indicated in the full view (right). (A) The antisymmetric perturbations of the cores develop. (B) Perpendicular secondary filaments form between the cores. (C) Secondary filaments begin to interact with each other and break down. (D) Tertiary filaments begin to form perpendicular to the secondary filaments. (E) Tertiary filaments are fully formed. (F) The flow breaks down into a disordered tangle of vortices. The vorticity thresholds are 0.079 ≤ ∣ω∣/∣ω∣max ≤ 0.099 for (A) and 0.110 ≤ ∣ω∣/∣ω∣max ≤ 0.211 for (B to F), where ∣ω∣max is the maximum vorticity modulus for the entire simulation. (G) Normalized shell-to-shell energy transfer spectra indicate whether a mode is an energy source [T(k) > 0] or sink [T(k) < 0]. At early times (t* = 65.56 and 74.44), the secondary filaments switch from energy sinks to sources as they are generated and then interact to form new vortices. At late times, the spectra flatten as energy is transferred more uniformly across the scales of the flow. (G, inset) Normalized kinetic energy dissipation rate as a function of time. The energy dissipation rate increases with the development of the secondary filaments and peaks as the secondary filaments and residual cores break down into a tangle of fine-scale vortices. (H) Normalized kinetic energy spectra show the rapid development of a sustained turbulent state with Kolmogorov scaling—as indicated by the black line—around the peak dissipation rate.

Formation of a turbulent cascade

The iterative breakdown process occurs over diminutive length scales and fleeting time scales. This rapid generation of small-scale vortices leads to a marked increase in the energy dissipation rate, ϵ, as shown in the inset of Fig. 5G. The initial increase in ϵ is triggered by the onset of the elliptical instability and the formation of antisymmetric perturbations in the cores. The following precipitous rise in the dissipation rate coincides with the formation of the perpendicular secondary filaments.

The rate at which kinetic energy is transferred across scales is calculated for the simulation in 3D Fourier space through the instantaneous shell-to-shell energy transfer spectrum, T(k, t), as shown in Fig. 5G (9, 30). At a fixed time, T(k) is positive for a wave number, k, when energy flows toward the corresponding spatial scale (∼ k−1). Conversely, a negative value of T(k) indicates the flow of energy away from that corresponding spatial scale to other modes (see section S5 for details). When initially formed, the secondary filaments become pronounced energy sinks, given the large positive value of T(k) at the intermediate wave number of approximately kb = 6.75. This coincides with their absorption of energy from the primary vortex cores. Next, as the secondary filaments become fully developed and interact with each other, they change behavior and become sources of energy, as indicated by the negative value of T(k). Coupled with the simultaneous increase in the dissipation rate, this change in behavior of the secondary filaments from energy sinks to energy sources indicates the existence of a cascade by which kinetic energy is conveyed to smaller scales.

Through the breakdown of the secondary filaments and residual vortex cores into a disordered tangle of vortices, the dissipation rate reaches a maximum value. At this point, the flow is most vigorous, and the energy transfer spectra asymptote toward a flattened profile, indicating that energy is conveyed more uniformly across the various scales of the system, as shown in Fig. 5G. Thus, for this brief time, ϵ maintains an approximately constant maximum value, as the energy is smoothly transferred to the smallest, dissipative scale, η=ϵ14ν34. Kolmogorov proposed that for turbulent flows under similar conditions, the kinetic energy spectrum follows a distinct scaling of (kη)−5/3 (7). Notably, we find that the fully developed turbulent cloud formed by the collision of the two vortices, indeed, exhibits Kolmogorov scaling. The evolution of the normalized energy spectra, E(k)/(η14ν54), demonstrates how the flow reaches a sustained turbulent state around the peak dissipation rate, as shown in Fig. 5H. This turbulent energy spectrum scaling at the peak dissipation rate also emerges during the breakdown of interacting vortex tubes mediated by the elliptical instability at lower Reynolds numbers, as shown in section S6. Since the energy input of the system is finite, this turbulent state cannot be maintained indefinitely. As time progresses further, the viscosity of the fluid damps out the motion of the vortices at the smallest scales. While much energy remains at the large scales of the flow, it is unable to be transmitted to smaller scales following this iterative breakdown. Accordingly, the energy dissipation rate decreases and the turbulent state decays.

DISCUSSION

The violent interaction between two counter-rotating vortices leads to the rapid emergence of a turbulent cascade, resulting in a flow with an energy spectrum that—for an ethereal moment—obeys Kolmogorov scaling. We find that the emergence of this turbulent cascade is initiated by the late-stage, nonlinear development of the elliptical instability, which forms an ordered array of counter-rotating secondary vortex filaments perpendicular to the primary cores. In the high–Reynolds number limit, the neighboring secondary filaments may interact to form a new generation of perpendicular tertiary vortex filaments. These interactions of the secondary filaments with each other and the remnants of the vortex cores lead to the rapid formation of small-scale vortices. This ensemble of vortices interacting over the full range of scales of the system provides a conduit through which energy cascades down to the dissipative scale.

The iterative cascade, which leads to the generation of vortices at decreasingly small length scales, is strongly reminiscent of the mechanism proposed by Brenner et al. (17). One may speculate that the self-similar process suggested by this work could be modeled by assuming, in the spirit of (17), that at each iteration, the circulation is multiplied by a factor xΓ < 1, and the characteristic scale of the vortices is multiplied by a factor xδ < 1, resulting after n steps in a generation of vortices with circulation, Γn=xΓn1Γ1, and a spatial scale, δn=xδn1δ1. The corresponding time scale over which each step evolves can be estimated as tnδn2/Γn(δ12/Γ1) (xδ2/xΓ)n1. The cascade can go all the way down to vanishingly small spatial scales in a finite time provided xδ2<xΓ. The numerical results presented here, during the first steps of the cascade, suggest that xΓ ∼ 0.25 and xδ ∼ 0.2 − 0.4, and therefore, that the cascade may proceed in a finite time. It would be interesting to understand whether the cascade suggested by this work proceeds faster or slower than the Kolmogorov cascade. Whereas Kolmogorov theory implies that tn/t1 ∼ (δn1)2/3 (12), our results imply that tn/t1 ∼ (δn1)2 − ln (xΓ)/ ln (xδ). Therefore, the cascade proposed in this work proceeds faster than the Kolmogorov cascade for xδ<xΓ3/4. Our estimates suggest that the two cascades may proceed asymptotically at a comparable rate. A more precise understanding of the development of the elliptical instability is necessary to determine accurately the scaling factors xΓ and xδ.

The essential element of the cascade process is that at each scale, discrete pairs of antiparallel vortices are able to locally interact and produce a subsequent iteration via the elliptical instability. Vortices of similar size and circulation locally align in an antiparallel manner when they interact. This is a well-established consequence of Biot-Savart dynamics (31). Thus, the largest strains that drive the cascade will arise from the interactions of nearby vortices. We suggest that iterations of this cascade could proceed down to ever-smaller scales until viscous effects take over. Yet, the proliferation of other small-scale vortices, clearly visible in Fig. 5, could conceivably prevent vortex pairs from forming at some stage of the process, and we do not rule out that it may influence the dynamics. We remark, however, that the present work shows that only two clear iterations of the cascade are sufficient to produce a Kolmogorov spectrum. Even in the high–Reynolds number limit, a finite set of iterations occurring simultaneously for many independent pairs of interacting vortices might suffice to establish and sustain a turbulent cascade. The details of how this iterative process unfolds in the limit of a large Reynolds number are an important question for future research.

This framework strongly agrees with recent works by Goto et al. in a fully turbulent flow regime. Namely, their numerical results demonstrate the existence of many independent pairs of antiparallel vortices interacting and locally forming smaller generations of perpendicular vortex filaments in both fully developed homogeneous isotropic turbulence and wall-bounded turbulence (3234). These discrete interactions of antiparallel vortex pairs appear simultaneously throughout Goto’s simulations over four distinct scales (33). Because of the notable similarities between the iterative mechanism we observe and the results of Goto, we propose that the elliptical instability is likely the means by which these successive generations of perpendicular filaments are formed. Establishing a precise connection between our own results and Goto’s observations requires a fully quantitative analysis, which is beyond the scope of the present work.

Our work thus demonstrates how the elliptical instability provides a long-sought-after mechanism for the formation and persistence of the turbulent energy cascade through the local interactions of vortices over a hierarchy of scales. Supplied by the injection of energy at large scales, discrete iterations of this instability effectively channel the energy of a flow down to the dissipative scale through the formation of new vortices. From a quantitative point of view, the approximate estimates provided in this work suggest that the corresponding cascade proceeds in a finite time, although a precise comparison with the Kolmogorov cascade requires a better understanding of the nonlinear development of the elliptical instability. While the dynamics of turbulent flows likely involve other multiscale vortex interactions, this fundamental mechanistic framework can begin to unravel the complexity that has long obscured our understanding of turbulence.

MATERIALS AND METHODS

Experiments

In the experimental vortex ring collisions (16), fluorescent dye (Rhodamine B) was injected into the initially formed rings to visualize the core dynamics, as shown in Fig. 1A. Two vortex rings were fired head-on into one another in a 75-gallon aquarium. The vortex cannons were positioned at a distance of 8D apart, where D is the vortex cannon diameter. The full, 3D dynamics of the resulting collision were visualized by tomographically scanning over the collision plane with a rapidly translating pulsed laser sheet (λ = 532 nm). The pulsing of the laser (Spectra-Physics Explorer One 532-2W) was synchronized with the exposure signal of a high-speed camera (Phantom V2511), which imaged the illuminated plane head-on. Each image plane spans along the xy plane, and the laser sheet scans in the z direction. Thus, for each scan, the image slices were stacked together to form a 3D reconstruction of the collision. The spatial resolution of each volume is 145 μm × 145 μm × 100 μm per voxel in (x,y,z), and the time resolution is up to 0.5 ms per scan. The number of voxels in each scanned volume depends on the imaging window size, the camera frame rate, and the scanning rate. For example, the volume size is 512 × 512 × 64 voxels in (x, y, z) for the dyed core collision in movie S2 and 384 × 288 × 75 voxels for the fully dyed ring collision in movie S4. The series of volumetric scans were reconstructed in full 3D with temporal evolution using Dragonfly visualization software (Object Research Systems). The imaging apparatus can only detect the dyed regions of the fluid, so any flow structures that emerge during the breakdown that are undyed cannot be observed. When only the vortex cores were dyed, we probed through the volumes of each 3D scan along the azimuthal direction to locate the centroids of the cores at each cross section. This enabled us to extract the vortex core centerlines, which we used to track the deformation of the cores and measure the vortex ring radius, R(t), and the average spacing between the cores, d(t). To visualize the development of secondary flow structures that emerge during the collisions, we filled the vortex cannons with fluorescent dye before driving the pistons to form the rings. As a result, the regions around the vortex cores were dyed, as shown in Fig. 3.

Direct numerical simulations

We used DNSs to further probe the unstable interactions between the vortices. This allowed us to directly examine the evolution of the vorticity field, relate it to experimental visualizations, and compute statistical quantities characterizing the flow. We solved the incompressible Navier-Stokes equations using an energy-conserving, second-order centered finite difference scheme with fractional time stepping. We implemented a third-order Runge-Kutta scheme for the nonlinear terms and second-order Adams-Bashforth scheme for the viscous terms (35, 36). We simulated both vortex rings and tubes in cylindrical and Cartesian coordinates, respectively. To avoid singularities near the axis, the cylindrical solver uses qr = rvr as a primitive variable (36). The time step was dynamically chosen such that the maximum Courant-Friedrich-Lewy condition number is 1.2. Resolution adequacy was checked by three methods: monitoring the viscous dissipation and the energy balance, examining the Fourier energy spectra, and using the instantaneous Kolmogorov scale. White noise was added to all initial conditions to trigger the most unstable modes.

Simulated vortex rings

The rings were initialized as two counter-rotating Gaussian (Lamb-Oseen) vortices, each with a core radius σ wrapped into a torus of radius R0. The control parameters for this system are the circulation Reynolds number, ReΓ = Γ/ν, and the slenderness ratio of the rings, Λ = σ/R0. The circulation of the vortex rings, Γ, and the initial ring radius, R0, were used to nondimensionalize parameters in the code. We simulated the collision in a closed cylindrical domain, bounded by stress-free walls at a distance far enough to not affect the collision. After testing several configurations, the bounds on the domain were placed a distance R0 below and above the rings and 5 R0 from the ring axis in the radial direction. For the simulation presented in this paper, we selected a ring slenderness of Λ = 0.1, a circulation Reynolds number of ReΓ = 4500, and an initial ring-to-ring distance of d = 2.5R0. These parameters are comparable to the experimental vortex rings, as shown by the measurements in section S1. Points were clustered near the collision regions in the axial and radial directions, while uniform resolution was used in the azimuthal direction (16, 20). A rotational symmetry of order five was forced on the simulation to reduce computational costs. The vortex core centerlines were located by slicing azimuthally through the pressure field at every time step and identifying the local minima of each vortex cross section. In addition, a simulated passive scalar was injected into the vortex rings to visualize the dynamics of the collision and compare with experiments, as shown in movie S3. Because of computational restrictions, the diffusivity of the dye is equal to the kinematic viscosity of the fluid (i.e., the Schmidt number is unity).

Simulated vortex tubes

For the vortex tubes, we simulated a triply periodic cubic domain of period ℒ, which is discretized using a uniform grid. The two counter-rotating, parallel tubes were both initialized with a Gaussian (Lamb-Oseen) vorticity profile of radius, σ, and circulation, Γ, initially separated a distance, b, apart. The system is characterized by two dimensionless parameters: the circulation Reynolds number, ReΓ, and the ratio, σ/b. Again, the circulation, Γ, was used as a nondimensional parameter, along with b. We set the core size to σ = 0.06ℒ, fixed b/σ = 2.5, and ran simulations with ReΓ at 2000, 3500, 4500, and 6000, with grid sizes of 2563, 3603, 5403, and 5403, respectively. As the counter-rotating tubes interact and break down, they naturally propagate through the periodic domain. For each visualization, the propagation of the tubes was subtracted so that the tubes remained in the center of the domain. In addition, in all 3D visualizations of the vorticity modulus, ∣ω∣(t), the vorticity modulus at each voxel was normalized by the maximum vorticity modulus for all time, ∣ω∣(t)max (see section S6).

SUPPLEMENTARY MATERIALS

Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/6/9/eaaz2717/DC1

Section S1. PIV analysis of vortex ring geometry

Section S2. Simulating vortex ring collisions using the Biot-Savart approximation

Section S3. Nonlinear development of the elliptical instability

Section S4. Interactions of secondary vortex filaments

Section S5. Analysis of the transfer of energy in a turbulent flow

Section S6. Emergence of turbulence from the elliptical instability with increasing Reynolds number

Section S7. Supplementary movie descriptions

Fig. S1. Vortex ring tracking and measurement through PIV.

Fig. S2. Vortex ring and core geometry.

Fig. S3. Modeling radial growth of colliding vortex rings.

Fig. S4. Onset of the elliptical instability for colliding vortex rings.

Fig. S5. Formation of a secondary vortex filament.

Fig. S6. Alternating structure of secondary filaments.

Fig. S7. Evolution of circulation.

Fig. S8. Interactions of secondary vortex filaments.

Fig. S9. Transition to turbulence in DNS of interacting vortex tubes.

Fig. S10. Vorticity evolution for DNS of interacting vortex tubes.

Movie S1. Head-on collision of vortex rings.

Movie S2. Experimental vortex ring collision with dyed cores.

Movie S3. DNS of dyed vortex ring collision.

Movie S4. Experimental fully dyed vortex ring collision.

Movie S5. DNS of vortex tube interaction: ReΓ = 4500.

Movie S6. DNS of vortex tube interaction: ReΓ = 3500.

Movie S7. Interaction and splitting of secondary vortex filaments.

Movie S8. DNS of vortex tube interaction: ReΓ = 6000.

Movie S9. Iterative cascade of elliptical instabilities.

Movie S10. Gaussian fit to vortex core PIV data.

Movie S11. DNS of vortex tube interaction: ReΓ = 2000.

References (3743)

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.

REFERENCES AND NOTES

Acknowledgments: R.O.M. thanks the Core facility for Advanced Computing and Data Science (CACDS) at the University of Houston for providing computing resources. Funding: This research was funded by the National Science Foundation through the Harvard Materials Research Science and Engineering Center DMR-1420570 and through the Division of Mathematical Sciences DMS-1411694 and DMS-1715477. M.P.B. is an investigator of the Simons Foundation. A.P. acknowledges financial support from the IDEXLYON project (Contract ANR-16-IDEX-0005) under University of Lyon auspices. Author contributions: This work was initiated by M.P.B., A.P., and S.M.R. Experiments were developed by R.M. and S.M.R and performed by R.M. Simulations were performed by R.O.M. All authors participated in the data analysis and contributed to the writing of this manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.
View Abstract

Stay Connected to Science Advances

Navigate This Article