WorksPhysics and Astronomy
Influence of three-dimensionality on wake synchronization of oscillatory cylinder
Abstract
We investigate the effect of three-dimensionality on the synchronization characteristics of the wake behind an oscillating circular cylinder at Re = 300. Cylinder oscillations in rotation, transverse translation, and streamwise translation are considered. We utilize phase-reduction analysis, which quantifies the phase-sensitivity function of periodic flows, to examine the synchronization properties. Here, we present an ensemble-based framework for phase-reduction analysis to handle three-dimensional wakes that are not perfectly time-periodic. Based on the phase-sensitivity functions, synchronizability to three types of cylinder oscillations is evaluated. In spite of similar trends, we find that phase-sensitivity functions involving three-dimensional wakes are lower in magnitude compared to those of two-dimensional wakes, which leads to narrower conditions for synchronization to weak cylinder oscillations. We unveil that the difference between the phase-sensitivity functions of two- and three-dimensional flows is strongly correlated to the amplitude variation of the three-dimensional flow by the cylinder motions. This finding reveals that the cylinder motion modifies the three-dimensionality of the wake as well as the phase of vortex shedding, which leads to reduced phase modulation. The synchronization conditions of three-dimensional wakes, predicted by phase-reduction analysis, agree with the identification by parametric studies using direct numerical simulations for forced oscillations with small amplitudes. This study presents the potential capability of phase-reduction to study synchronization characteristics of complex flows.
Introduction
Synchronization of the wake behind a bluff body to its structural vibration is of interest for various engineering applications. Such synchronization can cause detrimental effects, such as structural fatigue and resonance induced by vortex-induced vibrations [1, 2]. Alternatively, promoting synchronization can enhance the performance in other engineering systems, such as the molecular mixing rate in chemical reactors [3], the heat transfer rate of heat exchangers [4], and the efficiency of energy-harvesting systems [5]. The importance of accurately predicting synchronization conditions in engineering systems cannot be emphasized enough [6, 7].
Previous studies have investigated the synchronization between the vortex shedding frequency from a bluff body, mainly for a circular cylinder, and its oscillatory motion. For various types of oscillation motions and Reynolds numbers, synchronization regimes and vortex patterns have been identified and classified in both experimental and numerical studies [8, 9, 10, 11, 12, 13, 14, 15]. These studies are based on time-consuming sweeps of the parametric space in the amplitude-frequency domain using forced motion of a bluff body. Moreover, the synchronization boundaries are identified differently depending on the identification criteria used [16] and the wake measurement location [17]. These subtle difficulties in the identification of precise synchronization conditions, as well as a lack of understanding the underlying physics, highlight some weaknesses of previous experimental and numerical approaches. Therefore, a theoretical approach to investigate synchronization is desirable.
The theoretical approach we consider is phase-reduction analysis, which describes the dynamics of high-dimensional periodic flows in terms of a single scalar variable “phase.” Phase-reduction analysis has been applied to nonlinear oscillators in a wide range of studies, including those on biological rhythms [18, 19] and chemical oscillators [20, 21]. This technique enables us to examine synchronization without extensive parametric sweeps by identifying the phase-response of flows to external perturbations [22, 23, 24, 25]. Furthermore, phase-reduction analysis has been used to guide flow control of periodic flows [26, 27, 28, 29]. However, all previous applications of phase-reduction analysis were focused on two-dimensional periodic flows, even though synchronization in real engineering systems often involves three-dimensional flows. The three-dimensionality of a flow creates added richness compared to its two-dimensional counterpart, which influences the synchronization properties. Hence, it is important to identify the influence of three-dimensionality on synchronization. However, the implementation of phase-reduction analysis is not straightforward for three-dimensional wakes since, generally, they are not perfectly periodic.
In this study, we examine the effect of three-dimensionality on the synchronization of a cylinder’s wake to the motion of the cylinder. We consider rotational, crossflow translational and streamwise translational oscillations, which are the bases of in-plane cylinder oscillation. We extend the phase reduction analysis to characterize the perturbation dynamics of three-dimensional flows. Since the underlying three-dimensionality adds fluctuations to the limit cycle, we leverage an ensemble-averaging technique to obtain mean responses of the three-dimensional wake.
This paper is organized as follows. Physical characteristics of three-dimensional wakes are presented in § 2. Fundamentals of phase-reduction analysis for periodic flows are introduced in § 3 with a new technique to measure the phase-response of three-dimensional flows. The results of the present phase-reduction analysis are provided in § 4, revealing the influence of three-dimensionality within the flow on the synchronization. Conclusions are offered in § 5.
Numerical simulation of cylinder wakes
Problem description and numerical setups
We consider two- and three-dimensional incompressible wakes past a circular cylinder using direct numerical simulations (DNS). The Reynolds number in this study is selected as , where is the free-stream velocity, is the diameter of the cylinder and is the kinematic viscosity. The Reynolds number is selected such that the cylinder wake develops three-dimensional structures, which is discussed in § 2.2 in more detail. A finite volume formulation [30, 31] and a fractional-step method [32] with second-order accuracy are adopted for spatial discretization and time stepping, respectively. The computational domain is extended to from the center of the cylinder shown in Figure 1. To simulate the three-dimensional wake, the spanwise extension of the computational domain is set to with the spanwise extent of the cylinder being , which allows us to capture the three-dimensional features related to mode A and mode B instabilities [33]. Structured grids are generated near the cylinder and in the downstream region with the uniform grid size in the spanwise direction, which sufficiently resolves flow structures of mode A and B ( and ). To reduce computational costs, hybrid-typed grids with approximately 0.1 million and 7.3 million volume cells are used for two- and three-dimensional wakes, respectively. Grid distributions in other directions are stretched to cluster cells near the cylinder surface. Far-field regions are discretized in an unstructured manner. The time step size of guarantees the CFL number remains small, , during computations.
Figure 1. Computational domain for DNS of three-dimensional cylinder wakes.
For the DNS of the flow over a stationary cylinder, Dirichlet boundary conditions are given to the inlet boundary with the free-stream velocity and the cylinder surface with , respectively. The far-field boundary is prescribed with a Neumann condition of . The periodic boundary condition is enforced in the spanwise direction, and a convective outlet condition is given to the outflow boundary. Our numerical solutions are validated through a comparison of the Strouhal number, St, the lift coefficient, , and the drag coefficient, , respectively, defined as
where denotes the fluid density, and represent the lift and drag acting on the cylinder, and is the frequency of the lift fluctuations that contains the maximum energy. Table 1 summarizes the validation of the current simulation.
| St | |||
| Present (2D) | 0.212 | 0.651 | 1.371 |
| Zhang et al. [1995] | 0.216 | 0.692 | 1.456 |
| Persillon and Braza [1998] | 0.209 | 0.526 | 1.405 |
| Jiang et al. [2016] | 0.211 | 0.641 | 1.377 |
| Present (3D) | 0.202 | 0.428 | 1.259 |
| Zhang et al. [1995] | 0.212 | 0.430 | 1.260 |
| Persillon and Braza [1998] | 0.206 | 0.477 | 1.366 |
| Kravchenko et al. [1999] | 0.203 | 0.400 | 1.280 |
| Posdziech and Grundmann [2001] | 0.201 | 0.410 | 1.250 |
| Jiang et al. [2016] | 0.204 | 0.464 | 1.296 |
Table 1. Comparison of the Strouhal number (St), root-mean-squared lift coefficient (), and time-averaged drag coefficients () based on 200 shedding cycles for two- and three-dimensional cylinder flows at .
To measure the phase-response and identify the synchronization boundaries of the cylinder wake, we introduce impulsive and oscillatory motion for the cylinder in the DNS. The motions of the cylinder are prescribed with the instantaneous translational or rotational speed of the cylinder, . The impulsive motion is approximated with a Gaussian function
where is the magnitude of the impulsive motion and determines the width of the Gaussian function, which is set to in this study. Similarly, the oscillatory cylinder motion is modeled using the sinusoidal form
where and are the oscillation amplitude and angular frequency. Cylinder motions are incorporated in the DNS by modifying the boundary conditions depending on the type of motion [24]. For the rotating cylinder, is assigned to the tangential velocity at the cylinder surface, replacing the no-slip condition. Crossflow and streamwise translations are realized by moving the reference frame, subtracting from the velocity at the inlet and far-field boundary.
Wake characteristics of a stationary cylinder
We analyze the fundamental flow physics of the three-dimensional wake of a stationary cylinder at focusing on its representation in the phase plane. We start by providing a definition of the phase, , and the amplitude, , of a cylinder’s wake based on the lift coefficient and its Hilbert transform [39] (Pikovsky et al. 1997) as
where p.v. denotes the Cauchy principal value, and is determined as to be consistent with earlier studies [22, 23].
The wake behind a circular cylinder undergoes the first bifurcation at caused by the primary instability. It triggers the wake transition from a flow with two steady counter-rotating vortices to a two-dimensional periodic flow with von Kármán vortex shedding [41, 42]. Although at , since the computational domain is restricted to only be two-dimensional for the left side graphic of Figure 2(a), it qualitatively shows this primary instability. The two-dimensional wake presents a strong single peak in the corresponding frequency spectrum shown in Figure 2(c), indicating the perfect periodicity of the two-dimensional wake. In the phase plane, the two-dimensional wake is represented by the limit-cycle as shown in Figure 2(d). By increasing the Reynolds number further, or in our case by making the computational domain three-dimensional in the case of the right side graphic of Figure 2(a), the second bifurcation occurs and the wake becomes three-dimensional. This transition to the three-dimensional flow is due to the emergence of two distinct instabilities [33, 13, 43, 38, 44]. The mode A instability occurs at with the spanwise wavelength which deforms the primary vortices. The mode B instability starts to appear at the higher Reynolds number of with the spanwise wavelength . This instability mode is associated with secondary fine structures in the braid region, stretched in the streamwise direction and connecting the primary vortices. These instability modes create time-varying lift fluctuations associated with the more broadband frequency spectrum and a corresponding shift in the peak frequency, as shown in Figure 2(b) and (c). The three-dimensional wake exhibits an unclosed trajectory with the amplitude variation in the phase plane, depicted in Figure 2(d).

Figure 2. Comparison of two- and three-dimensional wakes at . (a) Contours of spanwise vorticity of the two-dimensional wake and flow structures of three-dimensional wake visualized by the isosurface of the –criterion at colored by spanwise vorticity. (b) Temporal variation of the lift coefficient, (c) power spectra, , of the lift coefficient and (d) phase plane representation of the lift coefficient.
Two instantaneous flow snapshots corresponding to lower and higher amplitudes, and , at the same phase, are visualized in Figure 3(a) and (Figure 3b), respectively. The visualizations distinctly emphasize the primary and secondary vortical structures by coloring the spanwise vorticity, , and streamwise vorticity, . We observe that the secondary vortical structures are more uniformly arranged and aligned with each other in the higher amplitude cycle with . Moreover, the primary vortices are more coherent along the spanwise direction. Contrarily, in the lower amplitude cycle with , the streamwise vortical structures appear less organized, and the primary vortex cores exhibit greater dislocations and deformation along the spanwise direction.
Figure 3. Instantaneous wake from a stationary cylinder visualized by isosurfaces of the -criterion () colored with (left column) and (right column) corresponding to (a) and (b) .
Similar trends are seen more clearly in scale-decomposed flow structures [45]. We apply a spatial band-pass filter to the velocity field based on a two-dimensional Gaussian kernel, , in the and directions with the length scale range of as
where
and the integration window covers the whole computational domain in the and directions. Large- and small-scale flow structures with a length scale range of and are visualized in Figure 4, where the cutoff length scale is [46]. We observe more large-scale structures as the amplitude, , increases, while the small-scale structures diminish. These observations tell us that the cylinder wake is more irregular as the amplitude, , decreases.

Figure 4. Instantaneous visualization of scale-decomposed flows by isosurfaces of the -criterion () corresponding to (a) and (b) . The left and right columns capture large-scale and small-scale structures, respectively.
These patterns can also be quantitatively identified by enstrophies defined as
where denotes the vorticity component in each direction (, , and ). The non-spanwise enstrophy characterizes the three-dimensionality of the wake induced by the spanwise mixing. The integration window is set to to examine flow structures formed within one shedding period. The variation in the non-spanwise enstrophy, , and spanwise enstrophy, , with is shown in Figure 5(a). A decrease in corresponds to a decrease in and an increase in . This indicates that larger spanwise mixing occurs when is smaller and induces enhancement in three-dimensional flow structures. Moreover, weaker spanwise vortices are reflected as portions of smaller fluctuation amplitude in the time history, which corresponds to smaller in the phase plane.

Figure 5. Variation of enstrophies with . (a) Spanwise enstrophy and non-spanwise enstrophy . (b) Enstrophies of spanwise-averaged vorticity , mode A and mode B .
To identify the dependencies between the instabilities and phase-amplitude variations of the three-dimensional wake, we apply the discrete Fourier transform (DFT) in the spanwise direction, which leads to the decomposition of the vorticity field as
where denotes the coefficient of each Fourier mode and is the number of grid points in the spanwise direction. We categorize the Fourier modes into three vorticity components corresponding to the spanwise average (), mode A () and mode B (). In other words, Fourier modes with large wavelengths, such that , are considered mode A, and the remaining Fourier modes, with shorter wavelengths, are classified as mode B. This wavelength threshold is chosen to fall within a range of stable wavelengths that separate the range of unstable wavelengths of mode A and mode B [47, 44].
As in the previous analysis, we quantify the contribution of each vorticity component by evaluating total enstrophy as shown in Figure 5(b). Attributing to the orthogonality of Fourier modes, the total enstrophy, , is also decomposed into three parts as . Mode B surpasses mode A through the entire range of , which implies the finer structures from the mode B instability are more prominent than large-scale structures from mode A. Both and become larger as decreases, whereas shows the opposite trend. Thus, it can be argued that secondary instabilities convert two-dimensional flow features in the wake to three-dimensional structures as decreases.
We also identify regions where each instability mode actively grows by plotting the magnitudes of each vorticity component in Figure 6. For the shedding cycle with , the mode A vorticity component highlights primary vortex cores, while mode B structures reside in the shear layers connecting two primary vortices. Flow structures of mode A last longer in the downstream region than mode B structures which implicates a faster decay rate of the mode B structures [36]. Moreover, smooth isocontour lines of in Figure 3(b) reflect the coherence of primary vortices in the spanwise direction. These trends are generally maintained in shedding cycles with smaller , although the distributions of instability modes become more irregular due to the enhanced spanwise mixing. In particular, we observe that the mode A vorticity component at smaller no longer only highlights primary vortex core regions but it is also present in the braid region. Flow structures associated with the mode B vorticity component are stronger and thickened in the braid regions, which gives rise to larger elongated vortical structures in this region. This can also be seen in Figure 3. These originate from the thickening of the shear layers in the braid region connecting the two primary vortices, which now allows flow structures with the larger wavelength to be amplified within this region.

Figure 6. Contours of spanwise averaged (top row) and (bottom row) for (a) and (b) . The black lines represent the contour lines of .
Phase-reduction analysis of fluid flows
Phase-reduction analysis of periodic flows
The dynamics of incompressible flows governed by the Navier–Stokes (NS) equations can be expressed with the state vector and the NS operator as
where and denote dimensionless space and time, respectively. A periodic flow can be described as a limit cycle solution of the governing equation 10 with phase . Since is periodic, where is the natural angular frequency of the flow. Here, we consider the phase functional that allows us to relate the high-dimensional instantaneous flow field to its phase variable . The concept of phase, , is then extended to the vicinity of the limit cycle to describe weakly perturbed flows. More specifically, if the flow asymptotically converges to the limit cycle solution as it develops, the phase of is defined as . Combined with the governing equation 10, the phase dynamics of the unperturbed periodic flow can be described using functional derivatives of as
where denotes the control volume.
Now, suppose a small perturbation term, with magnitude and spatiotemporal profile such that , is added to equation 10 to analyze the dynamics of perturbed flows,
This leads to the phase dynamics of weakly perturbed flows,
For a weak perturbation, the functional derivative can be linearly approximated with respect to the limit cycle . This yields the spatial phase-sensitivity function
which provides the linear phase-response of fluid flows to small external perturbations. In particular, equation 13 is further simplified by introducing the phase-sensitivity function associated with the spatial forcing profile as
when the perturbation can be decomposed as .
Among the various approaches to determine the phase-sensitivity function [[48, 50] (Iima 2019)], we employ the direct method which can be performed in both numerical and experimental studies [[22]]. For this method, a weak impulsive perturbation is applied to the flow at different phases, where is a Dirac-delta function centered at . When the transient effects of the impulsive perturbation settle, the perturbed flow returns to a stable state on the limit cycle, leaving an asymptotic phase shift. The asymptotic phase shift is often referred to as the phase-response function , where denotes the state vector of the flow impulsively perturbed at . The phase-sensitivity function, , is calculated by integrating equation 15 with an impulsive perturbation and results in through the first-order approximation [[48]]. Based on this perspective, the phase-sensitivity function can be physically interpreted as how much phase advancement, or delay, would be induced by imposing an external impulsive perturbation at a particular phase of the system.
Using equation (15), the synchronization of wake dynamics to any periodic forcing, with angular frequency , can be investigated theoretically. Synchronization between the flow and the periodic forcing is achieved when the relative phase converges to a constant value, i.e.,
where the natural number denotes the subharmonic number of interest. This expression can be rewritten in an autonomous form by averaging over a period of the perturbation [20, 51],
where is the phase-coupling function given by
Considering the stable solution of equation (17), the condition for synchronization can be theoretically derived as
and the synchronizability is defined as . The condition described by equation (19) theoretically identifies the possible combination map of the forcing angular frequency, , and magnitude, , that can achieve synchronization, known as the Arnold tongue. The synchronizability, , conveys the ease by which synchronization is achieved for a specific periodic perturbation at a different forcing frequency. A larger indicates that a smaller motion is sufficient for the flow to be synchronized with external forcing at angular frequency .
Ensemble-based phase-reduction framework
Strictly speaking, the phase-reduction framework is only valid for perfectly periodic flows. However, as discussed in § 2, three-dimensional flows do not exhibit perfect limit-cyclic behavior which requires care when performing phase-reduction analysis. In recent years, a few studies have worked on extending the phase-reduction approach to chaotic oscillators [52, 53, 54, 55, 56]. Inspired by these works, we extend the phase-based description to three-dimensional flows.
The phase dynamics of three-dimensional flows can be expressed as
where the term represents the instantaneous frequency fluctuation from the natural angular frequency , which stems from the phase diffusion of systems [52, 54]. Following the same procedure provided in § 3.1, the phase dynamics of the weakly perturbed three-dimensional flows are modeled as
Since we are more interested in the long-term dynamics when investigating synchronization, observing the mean phase dynamics of the flow is more valuable than the instantaneous dynamics. Hence, we apply an ensemble-averaging to equation (21) which can be expressed as
where denotes the ensemble-averaging operator and is the -th realization of a variable . This yields
which has the same form as equation (13). Thus, we define the spatial phase-sensitivity function of three-dimensional flows as
which evaluates from equation (23) with respect to the unperturbed flow . We then obtain the phase-sensitivity function of three-dimensional flows for a given spatial forcing profile according to equation (15). This leads us to the identical formulation of phase-coupling functions and synchronization conditions provided in equations (18) and (19) for three-dimensional flows.
For practical implementation of the ensemble-based framework, we consider a finite number of sample cycles to estimate the phase-sensitivity function. We note that sample cycles are extracted based on the observed probabilistic distribution of their occurrence in three-dimensional flows to avoid biased sampling. In this study, since the amplitude variable, , in the phase plane captures the three-dimensionality of the cylinder wake, we construct the probability density function of this cycle amplitude using a kernel density estimation with the Gaussian function [57] as shown in Figure 7(a). We then divide into subdivisions that have an equal probability, and sample cycles are extracted from each subdivision. We then apply the direct method to obtain the phase-response functions on these selected samples. Extracted sample cycles are impulsively perturbed at various phases to measure the phase-response function as shown in Figure 8(a). Finally, the phase-sensitivity function is calculated by averaging and normalizing with the impulse magnitude, , yielding

Figure 7. (a) Probability density function of cycle amplitude for extraction and (b) 10 selected sample cycles.
Figure 8. (a) Representation in the phase plane of three sample cycles, each with different amplitudes, perturbed at with rotary impulsive cylinder motion (solid) and the corresponding unperturbed sample cycles (dashed). (b) Estimation of and for ensemble-based phase-reduction analysis.
where denotes the phase-response function of the -th sample cycle from among a total of cycles.
Determination of the phase-response function poses another challenge in the phase-reduction analysis of three-dimensional flows. Unlike for two-dimensional flows, the perturbed three-dimensional flow does not exactly return to the unperturbed trajectory due to its chaotic nature negating the concept of an asymptotic phase shift. This requires careful assessment of the phase-response function. We observe that when the perturbation is sufficiently weak, there exists a finite time interval where the dynamics of the perturbed flow remain analogous to the unperturbed flow. Within this finite time interval, the perturbed flow initially deviates from the unperturbed flow due to the transient effect of the perturbation. The perturbed flow then approaches the unperturbed flow on the phase plane as the transient effect of the perturbation dies out. After this time horizon, the perturbed flow starts to deviate again from the unperturbed flow due to the chaotic nature of three-dimensional flows.
Based on this perspective, we use the time , when the perturbed flow is closest to the unperturbed flow, to estimate the phase-response functions of three-dimensional flows. To determine , we monitor the distance function between the unperturbed and perturbed flows on the phase plane as shown in Figure 8(b), which is defined as
using the law of cosines. Similar to the way we denote the phase functional as returning the phase for a particular flow , the amplitude functional returns the amplitude of that flow field as . We now seek the time such that is minimized after the impulse perturbation and before it starts to grow again. To not be affected by the local fluctuations of , we mainly track the variation of local minimum and maximum values of . Both local maximum and minimum values of the distance function keep decreasing as the transient effects of the perturbation diminish. When either the local minimum or maximum value of increases, we can presume that the perturbed flow has started to deviate from the unperturbed flow. Thus, we take the time to be when we observe the minimum value in the function before either of the extrema values start to increase. Finally, the phase-response function is measured as for each sample cycle, as shown in Figure 8(b).
In addition to the phase-response and phase-sensitivity functions, we can similarly measure the amplitude variation introduced by the weak impulsive perturbation. Since denotes the three-dimensional characteristics of the wake, the deviation of the amplitude in the phase plane trajectory at represents a modification of underlying three-dimensional characteristics. In this context, we also quantify the function in addition to the phase-response functions as shown in Figure 8(b). Similar to the phase-sensitivity function , the function is averaged and normalized with the impulse magnitude, , to give the function defined as
This function represents the mean normalized variation of made by the impulsive perturbations.
Synchronization characteristics of three-dimensional wake
Let us numerically investigate the synchronization between oscillations of a circular cylinder and its wake based on phase-reduction analysis. In order to study the influence of the three-dimensionality in the wake, we compare the three-dimensional wake synchronization characteristics with the imposed two-dimensional wake ones at the same Reynolds number. As mentioned in § 1, we consider three types of in-plane cylinder motions, streamwise translation, crossflow translation, and rotation. While the two-dimensional wake is analyzed as presented in § 3.1, the three-dimensional wake is studied using the ensemble-based phase-reduction analysis with 10 sample cycles as shown in Figure 7(b). The sample cycles are extracted based on the probability density function constructed with 200 shedding cycles provided in Figure 7(a), which is identified to be of sufficient length for the convergence of the cumulative distribution function. We apply the impulsive cylinder motion with a magnitude of , which is identified as being sufficiently small to satisfy the linearity assumption, to measure the phase-sensitivity functions. This identification is performed by testing successively smaller impulse magnitudes and noting that corresponding decreases occurred in the phase or amplitude difference between the perturbed and unperturbed flows. However, decreasing the impulse magnitude from 0.025 to 0.0125 does not produce meaningful changes in the phase or amplitude difference for either of the three types of motions used.
Two- and three-dimensional wake phase-sensitivity functions for rotary, crossflow translation and streamwise translation motions are compared in Figure 9. The ensemble-averaged for three-dimensional flows are shown in blue and the light blue regions show the 95% confidence intervals obtained from the samples. We note that the overall trend of three-dimensional wake phase-sensitivity functions resembles those from two-dimensional wakes. This is because the two-dimensional coherent structures in von Kármán vortex shedding are predominant, even in three-dimensional wakes at , over the secondary instabilities. Hence, the perturbation dynamics of the flow follow the shedding of spanwise vortices. This implies that the potential exists to use two-dimensional flows from low to moderate Reynolds numbers to capture phase-sensitivity functions and perturbation dynamics for three-dimensional flows at the equivalent Reynolds number.
Figure 9. Comparison of phase-sensitivity functions of two- and three-dimensional wakes, denoted as and respectively, at for each perturbation type. The shaded area displays the 95% confidence interval of .
We find that the three-dimensional wake is less sensitive than the two-dimensional wake to all three types of external perturbation in terms of the phase. The magnitudes of the phase-sensitivity functions for a three-dimensional wake is significantly decreased for the rotary and crossflow perturbations. In contrast, the phase-sensitivity function for the streamwise perturbation is much less affected by the three-dimensionality. When the cylinder translates in the streamwise direction, it moves toward or away from the vortices behind it. Thus, a phase shift by the streamwise motion is largely associated with the adjustment of distance between the cylinder and the spanwise vortices. This distance is less affected by the spanwise three-dimensionality of the wake.
Using the phase-sensitivity functions, we determine the synchronizability of the wake to steady sinusoidal oscillations using the phase-coupling functions, . Phase-coupling functions of two- and three-dimensional wakes are evaluated and shown in Figure 10. Since the synchronization of the cylinder wake to streamwise oscillations mainly occurs when [58, 59], the second harmonic () is investigated for the streamwise translation. In fact, the phase coupling function with for streamwise oscillations is zero valued () due to the periodicity of the phase-sensitivity function as shown in Figure 9, indicating that synchronization to the first harmonic is hardly seen. However, synchronization to the first harmonic of streamwise cylinder oscillations can occur for large enough oscillation amplitudes [60, 61].
Figure 10. Phase-coupling functions, and , and synchronizability, and , evaluation of two- and three-dimensional wakes for each perturbation type.
On the other hand, the first harmonic () is targeted for rotation and crossflow translation. Similar to the phase-sensitivity function results, we again observe that three-dimensionality within the wake imposes a higher resistance to synchronization. Synchronizability to rotary and crossflow oscillations shows a large reduction (39.1% and 44.6%, respectively) compared to the two-dimensional flow. The synchronizability to streamwise oscillation showed only a 10.1% decrease.
To understand the source of the difference between the phase-sensitivity functions of two- and three-dimensional wakes at the same Reynolds number, we consider the effect of impulsive cylinder motion in an additional degree of freedom of the three-dimensional wake, the amplitude . We compare with the difference in phase sensitivities, , for the three perturbation motions as shown in Figure 11. We observe that the function generally becomes larger at phases where the phase-sensitivity function of the three-dimensional wake exhibits a relatively large difference from the corresponding two-dimensional one. Correlation coefficients, , between and are also evaluated and have values higher than 0.9. This indicates that the energy given by the cylinder motion changes both the instantaneous phase and the amplitude in the phase plane. Recalling that the amplitude variable reflects the instantaneous three-dimensional properties of the wake, it can be argued that the portion of energy transferred, in the amplitude direction, from the cylinder to the wake modifies the three-dimensional characteristics of the flow. However, since only the energy transferred in the instantaneous phase direction supports synchronization, three-dimensional wakes require larger energy to achieve synchronization as observed in the synchronizability estimations.
Figure 11. Amplitude variation, , for three-dimensional wakes normalized by the magnitude of impulsive cylinder motion along with 95% confidence intervals. Differences between the phase-sensitivity functions of two- and three-dimensional wakes, , are also provided for comparison.
Based on equation (19), the synchronization conditions of the nondimensional forcing amplitude, , and frequency, , can be theoretically predicted based on the phase-coupling functions and shown in Figure 12. We validate our methodology and results by comparing the predicted synchronization boundaries from phase-reduction analysis to DNS parametric studies. The region between the solid lines is predicted to be synchronized to the external forcing while the two regions outside the lines are predicted to be asynchronous. The circles indicate the synchronization bounds determined by performing DNS with the cylinder undergoing steady and continuous oscillations at the corresponding forcing amplitude and frequency. Horizontal error bars represent the uncertainty in the identification of synchronization boundaries associated with the nonuniform and finite resolution of the DNS parametric sweeps. Synchronization in DNS is determined by monitoring the temporal variation of the phase difference, , between the sinusoidally forced cylinder oscillation and the wake, defined in equation (16). When converges to a certain value with a small standard deviation, , we classify the case as synchronized.
Figure 12. Synchronization condition (Arnold tongue) for each type of oscillation motion predicted using phase-reduction analysis (solid lines) and identified using DNS (circles). Horizontal bars represent the uncertainty in the identification of synchronization boundaries by DNS.
From Figure 12, we observe that phase-reduction analysis is able to predict the synchronization conditions for both two- and three-dimensional wakes for , indicating that the linear assumption is valid for oscillations with small amplitudes. As the oscillation amplitude of the nondimensional velocity increases, the DNS results deviate from the predictions due to the stronger nonlinearity. In addition, synchronization to rotary and crossflow oscillations are not observed when the oscillation amplitude is weak, classified as and for each type of motion, respectively, and denoted as cross symbols in the figure. As discussed in Pikovsky et al. [ [40], chaotic oscillators with phase diffusion do not show synchronization when the oscillation is not strong enough to suppress the phase diffusion.
To elaborate on this point theoretically, we look into the temporal variation of the instantaneous phase difference , which can be approximately expressed as
derived from equation (21). Let us now compare the scales of each term on the right-hand side of this equation under the assumption that the oscillation amplitude, , is small and approaches zero. Since should be very close to to synchronize the wake to small-amplitude oscillations, the first term is negligible. The third term is also negligible because of the linear proportionality to . However, the second term is independent of the external forcing because it is an inherent property of the three-dimensional wake. Thus, the first and third terms disappear for small-amplitude oscillation conditions, and the dynamics of are governed predominantly by as . As a result, if instantaneous fluctuations in three-dimensional flows are not sufficiently small, it prevents the convergence of unless is restrained by relatively large oscillations. Thus, the oscillation amplitude has to exceed a certain threshold to facilitate synchronization regardless of the forcing frequency. Furthermore, the third term is also proportional to the phase-sensitivity function, which implies a larger threshold for when the magnitude of the phase-sensitivity function is smaller. This is seen in the largest threshold for for the crossflow oscillation and no threshold appearing within for the streamwise oscillation, which correspond to the smallest and largest phase-sensitivity function magnitudes among oscillation types as shown in Figure 9.
We also note that the left synchronization boundaries obtained from the three-dimensional wake DNS results are located inside the two-dimensional wake results. This indicates the added difficulty in synchronizing the three-dimensional wake as predicted by phase-reduction analysis. On the other hand, the right synchronization boundaries, representing high-frequency oscillations, exhibit different trends depending on the oscillation type which deviates from the theoretical predictions.
For two-dimensional wakes, we commonly see deviations of the DNS identified right boundaries from the predictions by phase-reduction analysis, which is also observed in the previous study of Khodkar et al. [ [24]. They identified that synchronization to higher frequencies is more restrictive than lower frequencies, attributing it to the physical constraint of the vortex formation region behind the cylinder. Synchronization to high-frequency oscillations requires the vortex formation region to be compact. However, it is impossible to shorten the vortex formation region beyond a certain limit. This is reflected in the asymmetry of the DNS-based Arnold tongue and its inward deflection at the right branch for two-dimensional wakes. This hypothesis, however, only applies to two-dimensional wakes as we observe that three-dimensional wakes exhibit a longer vortex formation region than the corresponding two-dimensional wake, as visualized in Figure 13. Since a three-dimensional wake initially has a longer vortex formation region, it holds a greater capacity for shortening the vortex formation region. This offers an easier path to synchronization for high-frequency oscillations in a three-dimensional wake compared to a two-dimensional wake, resulting in no inward deflection of the Arnold tongue at the right synchronization boundary.
Figure 13. Contours of mean spanwise vorticity of the two-dimensional (left) and three-dimensional (right) wakes.
For the rotary oscillations, while the left synchronization boundary agrees with the prediction, we find that the right synchronization boundary of the three-dimensional wake obtained from DNS results deflects outward as the oscillation amplitude increases. This results in the three-dimensional wake having a larger upper bound than the two-dimensional wake when . Two instantaneous flow structures of cases on the right boundary with and for rotary motion are compared in Figure 14(a) and (b). It is observed that secondary vortical structures disappear when , showing that high-frequency rotary oscillations largely suppress the three-dimensionality of the wake. The suppression of secondary instabilities and the two-dimensionalization of the wake by oscillatory cylinder motion are observed in previous studies as well [62, 63, 64]. This demonstrates the existence of a transition from a three-dimensional to a two-dimensional wake as the oscillation amplitude increases along the right branch. Due to this transition, cylinder wakes with high-frequency rotary oscillation deviate significantly from the unperturbed three-dimensional wake, which gives rise to the strong nonlinear response. Since phase reduction analysis adopts a first-order approximation to estimate the phase-sensitivity function, these nonlinear effects and the transition from a three-dimensional flow to a two-dimensional flow are not captured and are attributed to higher-order effects. On the other hand, such an outward deflection is not observed at the right boundary for crossflow oscillations since the wake transition does not occur for this type of motion as shown in Figure 14 (c) and (Figure 14d).

Figure 14. Wake visualization with isosurfaces of the -criterion () colored by for cases on the right synchronization boundaries.
Regardless of the oscillation amplitude, the DNS-based right synchronization boundary of the three-dimensional wake is always outside of the two-dimensional wake for the streamwise oscillations. Similar to the rotary oscillations, streamwise oscillations with are able to two-dimensionalize the cylinder wake, while the three-dimensionality still remains at the oscillation amplitude of , shown in Figure 14 (e) and (Figure 14f). To identify differences between the wakes and understand this observation further, we additionally examine scale-decomposed flow structures for this case on the right boundary, plotted in Figure 15, and compare them to the unperturbed three-dimensional wakes provided in Figure 4. We find that both large- and small-scale structures resemble the unperturbed three-dimensional wake with large . This indicates that the three-dimensionality of the wake can be partially suppressed by the streamwise oscillation even though the oscillation amplitude is small. Due to the stronger nonlinear response of the wake to the streamwise motion, discrepancies between the synchronization boundaries of the DNS results and the phase-reduction analysis can be large for this type of oscillation.

Figure 15. Scale-decomposed flow structures by isosurfaces of the -criterion () corresponding to streamwise oscillation with and on the right synchronization boundary . The left and right columns capture large- and small-scale structures, respectively.
Conclusions
We examined the influence of the three-dimensionality of the wake on its ability to synchronize to cylinder oscillations at . Phase-reduction analysis was used for the theoretical analysis of wake synchronization. Even though the ability of phase-reduction analysis to study the synchronization of two-dimensional flows has been shown in previous studies, the chaotic nature of three-dimensional wakes makes its implementation for these realistic flows challenging. To address this difficulty, we introduced an ensemble-averaging technique to extend phase-reduction analysis to three-dimensional flows. In addition, we provided a guideline to perform the ensemble-averaged phase-reduction analysis based on the direct method, which includes sample cycle extractions and measurements of phase-sensitivity functions. We applied phase-reduction analysis to both two- and three-dimensional cylinder wakes to understand the effect of three-dimensionality on synchronization.
While the behavior of the phase-sensitivity functions for two- and three-dimensional wakes are similar, due to two-dimensional von Kármán vortex shedding being the dominant feature in the wake dynamics of both, there is still a significant decrease in the phase-sensitivity of three-dimensional wake. This decrease is especially large for the crossflow and rotary cylinder motions. This ultimately leads to reduced synchronizability and results in narrower Arnold tongues compared to the two-dimensional wake cases. Observing the amplitude variations from impulsive motions, the reduction of synchronizability was attributed to the energy transferred to the wake partially going into the amplitude direction. We identified that the synchronization conditions derived by phase-reduction analysis are valid for cylinder oscillations with velocities that are smaller than 10% of the free-stream velocity. Additionally, the parametric examination showed that there exists a minimum required oscillation amplitude to overcome the phase diffusion of the three-dimensional wake and achieve synchronization. Nonlinear effects from wake three-dimensionality, which are associated with the vortex formation and transition from three- to two-dimensional flow, are also identified by analyzing flow structures of synchronized wakes. To sum up, the phase-reduction analysis predicts the synchronization conditions with reasonable accuracy, especially for small oscillation amplitudes when nonlinear effects are fairly small.
The present study not only reveals synchronization conditions of the three-dimensional cylinder wake but also provides a systematic way to predict and analyze the synchronization characteristics of more complex flows. Previous experimental studies observed that turbulent cylinder wakes at high Reynolds numbers (up to ) also exhibit rhythmic behaviors with a single dominant frequency, according to their energy spectrum, similar to the low Reynolds number wakes [65] (Braza et al. 2006). Therefore, it is anticipated that phase characterization of the flow is still valid, and the ensemble-based phase-reduction framework would be applicable, to turbulent flows with a high Reynolds number. Considering the sizable computational costs or experimental resources required to perform parametric studies, phase-reduction analysis gives an effective prediction of the synchronization with reasonable precision.
Acknowledgments
This work was supported by the National Science Foundation (NPS Award: 2129638; UCLA Award: 2129639). This work used Expanse at the San Diego Supercomputing Center (SDSC) through allocation PHY230016 from the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program and Hoffman2 provided by UCLA Office of Advanced Research Computing’s Research Technology Group. The authors are very grateful to Dr. Kai Fukami for his insight and suggestions during this work.
Declaration of interests
The authors report no conflict of interest.
References
- [1]T. Sarpkaya. A critical review of the intrinsic nature of vortex-induced vibrations. J. Fluids Struct., 19(4):389–447, 2004.
- [2]C. H. K. Williamson and R. Govardhan. Vortex-induced vibrations. Annu. Rev. Fluid Mech., 36:413–455, 2004.
- [3]B. Celik and A. Beskok. Mixing induced by a transversely oscillating circular cylinder in a straight channel. Phys. Fluids, 21(7), 2009.DOI
- [4]C. Gau, S. X. Wu, and H. S. Su. Synchronization of vortex shedding and heat transfer enhancement over a heated cylinder oscillating with small amplitude in streamwise direction. J. Heat Transfer, 123(6):1139–1148, 2001.DOI
- [5]J. Wang, L. Geng, L. Ding, H. Zhu, and D. Yurchenko. The state-of-the-art review on energy harvesting from flow-induced vibrations. Appl. Energy, 267:114902, 2020.DOI
- [6]P. W. Bearman. Vortex shedding from oscillating bluff bodies. Annu. Rev. Fluid Mech., 16(1):195–222, 1984.
- [7]E. Naudascher and D. Rockwell. Flow-Induced Vibrations: An Engineering Guide. Dover Publications, 2005.
- [8]D. J. Olinger and K. R. Sreenivasan. Nonlinear dynamics of the wake of an oscillating cylinder. Phys. Rev. Lett., 60(9):797, 1988.
- [9]H. G. C. Woo. A note on phase-locked states at low Reynolds numbers. J. Fluids Struct., 13(1):153–158, 1999.DOI
- [10]F. L. Ponta and H. Aref. Vortex synchronization regions in shedding from an oscillating cylinder. Phys. Fluids, 17(1):011703–011703, 2005.DOI
- [11]P. G. Perdikaris, L. Kaiktsis, and G. S. Triantafyllou. Chaos in a cylinder wake due to forcing at the Strouhal frequency. Phys. Fluids, 21(10), 2009.DOI
- [12]J. S. Leontini, D. L. Jacono, and M. C. Thompson. A numerical study of an inline oscillating cylinder in a free stream. J. Fluid Mech., 688:551–568, 2011.DOI
- [13]C. H. K. Williamson and A. Roshko. Vortex formation in the wake of an oscillating cylinder. J. Fluids Struct., 2(4):355–381, 1988.DOI
- [14]A. Ongoren and D. Rockwell. Flow structure from an oscillating cylinder part 1. mechanisms of phase shift and recovery in the near wake. J. Fluid Mech., 191:197–223, 1988.
- [15]D. L. Jacono, J. S. Leontini, M. C. Thompson, and J. Sheridan. Modification of three-dimensional transition in the wake of a rotationally oscillating cylinder. J. Fluid Mech., 643:349–362, 2010.DOI
- [16]S. Kumar, Navrose, and S. Mittal. Lock-in in forced vibration of a circular cylinder. Phys. Fluids, 28(11), 2016.DOI
- [17]S. Kumar, C. Lopez, O. Probst, G. Francisco, D. Askari, and Y. Yang. Flow past a rotationally oscillating cylinder. J. Fluid Mech., 735:307–346, 2013.
- [18]A. T. Winfree. Biological rhythms and the behavior of populations of coupled oscillators. J. Theor. Biol., 16(1):15–42, 1967.
- [19]Y. Shiogai, A. Stefanovska, and P. Vaughan E. McClintock. Nonlinear dynamics of cardiovascular ageing. Phys. Rep., 488(2-3):51–110, 2010.
- [20]Y. Kuramoto. Chemical Oscillations, Waves, and Turbulence. Dover Publications, 2003.
- [21]B. Pietras and A. Daffertshofer. Network dynamics of coupled oscillators and phase reduction techniques. Phys. Rep., 819:1–105, 2019.
- [22]K. Taira and H. Nakao. Phase-response analysis of synchronization for periodic flows. J. Fluid Mech., 846:R2, 2018.DOI
- [23]M. A. Khodkar and K. Taira. Phase-synchronization properties of laminar cylinder wake for periodic external forcings. J. Fluid Mech., 904:R1, 2020.DOI
- [24]M. A. Khodkar, J. T. Klamo, and K. Taira. Phase-locking of laminar wake to periodic vibrations of a circular cylinder. Phys. Rev. Fluids, 6(3):034401, 2021.DOI
- [25]I. A. Loe, H. Nakao, Y. Jimbo, and K. Kotani. Phase-reduction for synchronization of oscillating flow by perturbation on surrounding structure. J. Fluid Mech., 911:R2, 2021.
- [26]A. G. Nair, K. Taira, B. W. Brunton, and S. L. Brunton. Phase-based control of periodic flows. J. Fluid Mech., 927:A30, 2021.DOI
- [27]V. Godavarthi, Y. Kawamura, and K. Taira. Optimal waveform for fast synchronization of airfoil wakes. J. Fluid Mech., 976:R1, 2023.DOI
- [28]I. A. Loe, T. Zheng, K. Kotani, and Y. Jimbo. Controlling fluidic oscillator flow dynamics by elastic structure vibration. Sci. Rep., 13(1):8852, 2023.DOI
- [29]K. Fukami, H. Nakao, and K. Taira. Data-driven transient lift attenuation for extreme vortex gust–airfoil interactions. J. Fluid Mech., 992:A17, 2024.DOI
- [30]F. Ham and G. Iaccarino. Energy conservation in collocated discretization schemes on unstructured meshes. Annual Research Briefs, 2004(3-14):118, 2004.
- [31]F. Ham, K. Mattsson, and G. Iaccarino. Accurate and stable finite volume operators for unstructured flow solvers. Annual Research Briefs, pages 243–261, 2006.
- [32]J. Kim and P. Moin. Application of a fractional-step method to incompressible Navier–Stokes equations. J. Comput. Phys., 59(2):308–323, 1985.
- [33]D. Barkley and R. D. Henderson. Three-dimensional Floquet stability analysis of the wake of a circular cylinder. J. Fluid Mech., 322:215–241, 1996.
- [36]H. Jiang, L. Cheng, S. Draper, H. An, and F. Tong. Three-dimensional direct numerical simulation of wake transitions of a circular cylinder. J. Fluid Mech., 801:353–391, 2016.DOI
- [38]O. Posdziech and R. Grundmann. Numerical simulation of the flow around an infinitely long circular cylinder in the transition regime. Theor. Comput. Fluid Dyn., 15(2):121–141, 2001.DOI
- [39]M. G. Rosenblum, A. S. Pikovsky, and J. Kurths. Phase synchronization of chaotic oscillators. Phys. Rev. Lett., 76(11):1804, 1996.
- [40]A. S. Pikowsky, M. G. Rosenblum, G. V. Osipov, and J. Kurths. Phase synchronization of chaotic oscillators by external driving. Physica D, 104(3-4):219–238, 1997.
- [41]C. P. Jackson. A finite-element study of the onset of vortex shedding in flow past variously shaped bodies. J. Fluid Mech., 182:23–45, 1987.
- [42]M. Provansal, C. Mathis, and L. Boyer. Bénard-von Kármán instability: transient and forced regimes. J. Fluid Mech., 182:1–22, 1987.DOI
- [43]C. H. K. Williamson. Three-dimensional wake transition. J. Fluid Mech., 328:345–407, 1996.
- [44]L. V. Rolandi, J. Fontane, T. Jardin, J. Gressier, and L. Joly. Compressibility effects on the secondary instabilities of the circular cylinder wake. J. Fluid Mech., 966:A36, 2023.DOI
- [45]J. Fujino, Y. Motoori, and S. Goto. Hierarchy of coherent vortices in turbulence behind a cylinder. J. Fluid Mech., 975:A13, 2023.DOI
- [46]T. Yasuda, S. Goto, and J. C. Vassilicos. Formation of power-law scalings of spectra and multiscale coherent structures in the near-field of grid-generated turbulence. Phys. Rev. Fluids, 5(1):014601, 2020.
- [47]J. S. Leontini, D. L. Jacono, and M. C. Thompson. Stability analysis of the elliptic cylinder wake. J. Fluid Mech., 763:302–321, 2015.
- [48]H. Nakao. Phase reduction approach to synchronisation of nonlinear oscillators. Contemp. Phys., 57(2):188–214, 2016.DOI
- [50]Y. Kawamura, V. Godavarthi, and K. Taira. Adjoint-based phase reduction analysis of incompressible periodic flows. Phys. Rev. Fluids, 7(10):104401, 2022.
- [51]G. B. Ermentrout and D. H. Terman. Mathematical Foundations of Neuroscience. Springer, 2010.
- [52]K. Josić and D. J. Mar. Phase synchronization of chaotic systems with small phase diffusion. Phys. Rev. E, 64(5):056234, 2001.
- [53]J. T. C. Schwabedal, A. Pikovsky, B. Kralemann, and M. Rosenblum. Optimal phase description of chaotic oscillators. Phys. Rev. E, 85(2):026216, 2012.
- [54]W. Kurebayashi, K. Fujiwara, H. Nakao, and T. Ikeguchi. A theory on noise-induced synchronization of chaotic oscillators. IEICE Proceedings Series, 1(344):344–347, 2012.
- [55]R. Tönjes and H. Kori. Phase and frequency linear response theory for hyperbolic chaotic oscillators. Chaos, 32(4), 2022.
- [56]T. Imai, H. Suetani, and T. Aoyagi. On the phase description of chaotic oscillators. J. Phys. Soc. Jpn., 91(1):014001, 2022.
- [57]A. W. Bowman and A. Azzalini. Applied Smoothing Techniques for Data Analysis: The Kernel Approach with S-Plus Illustrations. Oxford University Press, 1997.
- [58]Q. M. Al-Mdallal, K. P. Lawrence, and S. Kocabiyik. Forced streamwise oscillations of a circular cylinder: locked-on modes and resulting fluid forces. J. Fluids Struct., 23(5):681–701, 2007.
- [59]O. A. Marzouk and A. H. Nayfeh. Reduction of the loads on a cylinder undergoing harmonic in-line motion. Phys. Fluids, 21(8), 2009.
- [60]J. S. Leontini, D. L. Jacono, and M. C. Thompson. Wake states and frequency selection of a streamwise oscillating cylinder. J. Fluid Mech., 730:162–192, 2013.
- [61]E. Konstantinidis and D. Bouris. Vortex synchronization in the cylinder wake due to harmonic and non-harmonic perturbations. J. Fluid Mech., 804:248–277, 2016.
- [62]P. Poncet. Vanishing of mode B in the wake behind a rotationally oscillating circular cylinder. Phys. Fluids, 14(6):2021–2023, 2002.
- [63]J. Deng, X. Shao, and A. Ren. Vanishing of three-dimensionality in the wake behind a rotationally oscillating circular cylinder. J. Hydrodyn., 19(6):751–755, 2007.
- [64]S. H. Kim, J. Y. Park, N. Park, J. H. Bae, and J. Y. Yoo. Direct numerical simulation of vortex synchronization due to small perturbations. J. Fluid Mech., 634:61–90, 2009.
- [65]S. Szepessy and P. W. Bearman. Aspect ratio and end plate effects on vortex shedding from a circular cylinder. J. Fluid Mech., 234:191–217, 1992.







