Possible states in the flow around two circular cylinders in tandem
Transcription
Possible states in the flow around two circular cylinders in tandem
1 Possible states in the flow around two circular 2 cylinders in tandem with separations in the vicinity 3 of the drag inversion spacing 4 Bruno S. Carmoa , Julio R. Meneghinib , Spencer J. Sherwina a Department 5 6 b NDF, of Aeronautics – Imperial College London, UK Department of Mechanical Engineering, Poli, University of São Paulo, Brazil Abstract 7 8 The possible states in the flow around two identical circular cylinders in 9 tandem arrangements are investigated for configurations in the vicinity of the 10 drag inversion separation. By means of numerical simulations, the hysteresis 11 in the transition between the shedding regimes is studied and the relationship 12 between (three-dimensional) secondary instabilities and shedding regime de- 13 termination is addressed. The differences observed in the behaviour of two- 14 and three-dimensional flows are analysed, and the regions of bi-stable flow 15 delimited. Very good agreement is found between the proposed scenario and 16 results available in the literature. 1 1 1 Introduction 2 In the external flow around solid bodies, it is a well-known fact that the presence of 3 other bodies in close proximity can change fundamental aspects of the flow, such as 4 fluid forces and transition thresholds. The effect of the presence of additional bodies 5 in the fluid stream is called flow interference. A particular type of flow interference 6 which is specially severe is the wake interference, which happens when one body is 7 immersed in the wake of another body. 8 In the case of bluff bodies, the most commonly applied model to study wake 9 interference is the flow around two identical circular cylinders placed in tandem 10 arrangements, as illustrated in figure 1. It is known from experiments (Zdravkovich, 11 1977; Igarashi, 1981) and computations (Mittal et al., 1997; Meneghini et al., 2001) 12 that different vortex shedding regimes can be observed in the flow around this type 13 of arrangement, depending on the centre-to-centre separation Lx . Adopting the 14 classification presented in Carmo et al. (2009), illustrated in figure 2, we see that 15 three different shedding regimes are observed for low Reynolds numbers. For very 16 small separations, the shedding regime SG (symmetric in the gap) is observed, as 17 shown in figure 2(a). In this regime, a pair of almost symmetric vortices is formed 18 in the gap between the cylinders and the root mean square (RMS) of the lift on 19 the downstream cylinder is very small. If the separation is gradually increased, the 20 shedding regime eventually changes to AG (alternating in the gap), in which regions 21 of concentrated vorticity grow and decrease alternatively in time on each side of the 22 line that links the centres of the cylinders (see figure 2(b)). This makes the RMS 23 of the lift coefficient on the downstream cylinder increase. It is also important to 2 Figure 1 – Sketch of the flow around two circular cylinders in tandem arrangement. 1 highlight that the drag on the downstream cylinder is usually negative for shedding 2 regimes SG and AG, as illustrated in figures 2(a)-(b). Finally, for larger separations, 3 a complete vortex wake is formed in the interstitial region, the RMS of the lift on the 4 downstream cylinder increases significantly and the mean drag on the downstream 5 cylinder becomes positive, as shown in figure 2(c). This shedding regime is called 6 WG (wake in the gap). Since the transition between the shedding regimes AG and 7 WG is marked by the inversion of the drag coefficient on the downstream cylinder, 8 the transition from one to the other is referred to as drag inversion and in this paper 9 we focus on flows in the vicinity of such transition. 10 It is known that the separation at which the drag inversion occurs depends on 11 the initial conditions because, at least for low Reynolds numbers, the flow is bi- 12 stable in the vicinity of the drag inversion point (Mizushima & Suehiro, 2005). For 13 this reason, it is more appropriate to refer to a drag inversion range than to a drag 14 inversion point. For a fixed separation, this range is defined in terms of Reynolds 15 numbers. Likewise, for a fixed Reynolds number this range is defined in terms of 16 centre-to-centre distance. As far as the authors are aware, no study to date has 17 calculated the drag inversion range for low Reynolds numbers while taking into 18 account the bi-stable nature of the flow in this region1 . 1 Papaioannou et al. (2006) calculated the drag inversion separation only for increasing Reynolds numbers. 3 1.5 CD, CL 1 0.5 0 -0.5 -1 -1.5 0 (a) Lx {D 10 20 30 40 20 30 40 20 30 40 tU∞/D 1.5 – regime SG 1.5 CD, CL 1 0.5 0 -0.5 -1 -1.5 0 (b) Lx {D 10 tU∞/D 2.3 – regime AG 1.5 CD, CL 1 0.5 0 -0.5 -1 -1.5 0 (c) Lx {D 10 tU∞/D 5 – regime WG Figure 2 – Left – Instantaneous vorticity contours illustrating the different shedding regimes observed in the flow around two circular cylinders in tandem arrangements. Contours vary from ωz D{U8 2.2 (light contours) to ωz D{U8 2.2 (dark countours). Right – Drag coefficient (grey solid line) and lift coefficient (black dashed line) time series for the downstream cylinder. Re 200, two-dimensional simulations. 1 A recent study by Carmo et al. (2009) investigated the three-dimensional insta- 2 bilities observed in the nominally two-dimensional time-periodic flow around two 3 identical circular cylinders placed in tandem in relation to the free-stream. These 4 instabilities are known as secondary instabilities, since they occur after the primary 4 1 instability, which is the transition from steady flow to two-dimensional time-periodic 2 flow (the primary instability in the flow around two circular cylinders in tandem was 3 investigated by Mizushima & Suehiro, 2005). The results obtained by Carmo et al. 4 (2009) were compared to those obtained for an isolated cylinder (Williamson, 1988, 5 1996; Barkley & Henderson, 1996). A summary of the main results of that work 6 are reproduced in figure 3, in which it can be seen that different modes appear in 7 the transition to three-dimensional flow in the wake for separations smaller than the 8 drag inversion spacing. For such cases, the three-dimensional structures appeared 9 later in terms of Reynolds number than for the flow around an isolated cylinder. It 10 was shown that for configurations at shedding regime SG, the unstable mode at the 11 onset of the secondary instability originated at the formation region, downstream 12 of the leeward cylinder. This mode, referred to as mode T1, has a topology that 13 breaks the spatial symmetry of the base flow, and its physical mechanism appears 14 to be associated to a hyperbolic instability. For slightly larger separations, the shed- 15 ding regime changed to AG and a different unstable mode, named mode T2, was 16 observed. Mode T2 has its origin at the base of the downstream cylinder, upstream 17 of the vortex formation region. A centrifugal instability in this region seems to give 18 rise to this mode. Like the single cylinder mode A (Williamson, 1988), mode T2 19 wake topology keeps the in-plane spatial symmetry observed in the base flow. If 20 the separation is increased a little more, but not so much as to change the base 21 flow shedding regime, a new unstable mode (mode T3) is initiated at the interstitial 22 region. Like mode T1, mode T3 also breaks the spatial symmetry of the base flow. 23 Some of the mode attributes suggest that the underlying physical mechanism is a 5 1 cooperative elliptical instability. On the other hand, if the separation was greater 2 than the drag inversion spacing (shedding regime WG), the initial stages of the tran- 3 sition in the wake occurred in a similar way to that of the isolated cylinder. The 4 first instability, mode A, arose earlier in Reynolds number terms when compared to 5 the single cylinder case, and it is therefore concluded that the downstream cylinder 6 has a destabilising effect on the flow for separations larger than the drag inversion 7 spacing. 8 Although in Carmo et al. (2009) a full characterisation of the modes was pre- 9 sented and physical mechanisms were proposed to explain the instabilities, that 10 paper did not address a point of high practical interest in engineering, which is how 11 the onset of three-dimensional instabilities affect the drag inversion. In the present 12 paper, we investigate in detail how the transition to three-dimensional flow affects 13 the vortex shedding regime, focusing on the vicinity of the drag inversion spacing 14 and taking into account the hysteresis of the regime transition. We obtain the pos- 15 sible flow states for Re 16 current results to explain previously published computational data. 17 2 ¤ 500 by means of direct numerical simulations and use the Numerical method The flows investigated in this paper were calculated using numerical simulations of the incompressible Navier–Stokes equations, here written in non-dimensional form: Bu pu.∇qu ∇p Bt ∇.u 0. 6 1 2 ∇ u, Re (1) (2) 500 450 400 Recr 350 300 250 200 150 100 0 1 2 3 4 5 6 7 8 9 10 11 7 8 9 10 11 Lx/D (a) 10.0 9.0 8.0 λcr /D 7.0 6.0 5.0 4.0 3.0 2.0 0 1 2 3 4 5 6 Lx/D (b) Figure 3 – Variation of the critical Reynolds numbers (a) and respective perturbation wavelengths (b) with the centre-to-centre separation for modes T1 (), T2 (), T3 () and A (N). Mode T1 bifurcates from shedding regime SG, modes T2 and T3 from shedding regime AG and mode A from shedding regime WG. 1 2 The cylinder diameter D is the reference length and the free-stream speed U8 is the reference speed used in the non-dimensionalisation. u pu, v, wq is the velocity field, ρU8D{µ is the Reynolds number and µ 3 t is the time, p is the static pressure, Re 4 is the dynamic viscosity of the fluid. The pressure was assumed to be scaled by the 7 1 constant density ρ. The numerical solution of these equations was calculated using 2 a Spectral/hp discretisation as presented in Karniadakis & Sherwin (2005). The 3 time integration scheme adopted was the stiffly stable splitting scheme presented by 4 Karniadakis et al. (1991). 5 Polynomials of degree 8 were used in the discretisation of the meshes for the 6 two-dimensional simulations. The meshes employed were the same as those used to 7 obtain the base flows in Carmo et al. (2009); an example is shown in figure 4. The 8 boundary conditions were u 9 10 11 12 1, v 0 on the left, upper and lower boundaries of the mesh in the figure, B u{B x and u v 0, Bv{Bx 0 on the right (outflow) boundary 0 on the cylinders’ walls. The high-order pressure boundary condition described in Karniadakis et al. (1991) was employed on every boundary apart from the outflow boundary, on which p 0 was imposed. 13 The three-dimensional simulations were performed using a three-dimensional ver- 14 sion of the Navier-Stokes solver which uses a Spectral/hp element discretization in 15 the xy plane and Fourier modes in the spanwise direction (Karniadakis, 1990). The 16 advantages of this approach is the high efficiency in the code parallelisation and that 17 the meshes generated for the two-dimensional simulations can be re-used. In the 18 three-dimensional simulations, domains with spanwise lengths between 8D and 12D 19 were employed, in order to comply with the wavelength of the instability that was 20 expected to arise. Depending on the Reynolds number and spanwise length, 16 or 32 21 Fourier modes were used in the discretisation in the spanwise direction and periodic 22 boundary conditions enforced on the planes at the boundaries perpendicular to the 23 cylinder axis. 8 Figure 4 – Mesh employed in the calculations of the flow around the configuration with Lx {D 5. 1 In the next section we also present the critical Reynolds numbers for the primary 2 and secondary instabilities in the wake. The data referring to the secondary instabil- 3 ities were extracted from Carmo et al. (2009), but the data referring to the primary 4 instabilities were calculated. To obtain the steady base flows, we have employed the 5 method presented by Tuckerman & Barkley (2000) with the modification suggested 6 by Blackburn (2002). The stability analysis procedure was the same as that used in 7 Carmo et al. (2009). 8 3 9 A number of two- and three-dimensional simulations were performed to investigate 10 the boundaries of the drag inversion range, fixing the geometric configuration and Results and discussion 9 1 varying the Reynolds number. Each of the calculations was run for at least 300 2 non-dimensional time units for the two-dimensional simulations and for at least 3 100 non-dimensional time units for the three-dimensional simulations. The mean 4 drag coefficient was used as the indicator of the shedding regime (AG or WG). 5 Due to the hysteretic nature of the transition between these regimes, determining 6 the upper (lower) boundary requires that we start our flow simulations at a lower 7 (upper) Reynolds number and increase (reduce) its value. The boundaries were 8 defined taking the Reynolds number of the first calculation that showed a change of 9 regime, with an uncertainty of ∆Re 0.5. 10 The primary instability was investigated by means of global linear stability anal- 11 ysis of the solutions of the steady Navier-Stokes equations. The critical Reynolds 12 number for each of the configurations was obtained using a search algorithm that 13 employed Newton’s method. The resolution in Reynolds number of this algorithm 14 was 1, i.e. only integer Reynolds numbers were tested, and the Reynolds number 15 was considered to be the critical one if the real part of the eigenvalue resulting from 16 the stability calculations had modulus smaller than 0.001. 17 Although is usually said that regime AG is associated with a negative mean drag 18 on the downstream cylinder, this is not always the case for low Reynolds numbers. 19 Figure 5 shows values of mean drag coefficient obtained for decreasing Reynolds 20 numbers for the configuration with Lx {D 3.8 and vorticity contours illustrating 21 the three different wakes observed (steady flow, shedding regime AG and shedding 22 regime WG). The discontinuities associated with the two changes of regime are clear 23 in the graph. It can be seen that the drag coefficient is small, but not negative, for 10 Figure 5 – Values of the drag coefficient of the downstream cylinder obtained for decreasing Reynolds numbers, Lx {D 3.8. The symbols in the graph correspond to the different shedding regimes observed: N – WG, – AG, – steady wake. The vorticity contours on the right hand side illustrate each of the shedding regimes; dark contours mean positive vorticity and light contours mean negative vorticity. 1 regime AG. For that reason, it was impossible to define a general threshold value for 2 the drag coefficient which would indicate the change of shedding regime; in order to 3 find the thresholds of shedding regime transition it was necessary to examine each 4 configuration individually, checking the flow field contours and drag coefficient time 5 histories. 6 7 Figure 6 displays the results of the calculations on a Re vs. Lx {D map. The curves showing the variation of the critical Reynolds number with the centre-to- 11 1 centre distance for the modes A and T3, obtained by Carmo et al. (2009), and the 2 critical Reynolds numbers for the primary instability are also plotted on the map. 3 The four regions of bi-stable flow are marked in shades of grey. The bottom one 4 is located under the curve of critical Reynolds numbers for the primary instability, 5 and corresponds to a region in the parameter space where only steady flow or two- 6 dimensional flow with shedding regime WG are possible. The second grey region 7 from the bottom is located between the curve of critical Reynolds numbers for 8 mode A and the curve of critical Reynolds numbers for the primary instability. 9 Hence only two-dimensional time-periodic flows are possible in this region, but the 10 vortex shedding regime can be either AG (2d-AG) or WG (2d-WG), depending on 11 the initial conditions. The third region of bi-stable flow is between the curves of 12 critical Reynolds numbers for mode A and mode T3. In this region, two-dimensional 13 flows at regime AG (2d-AG) and three-dimensional flows at regime WG (3d-WG) 14 are possible. Lastly, the fourth top region of bi-stable flow is situated above the 15 mode T3 critical Reynolds number curve. The flow in this region is always three- 16 dimensional and, depending on the initial conditions, the vortex shedding regime 17 can be AG (3d-AG) or WG (3d-WG). Plots of vorticity iso-surfaces, obtained by 18 means of three-dimensional simulations, are shown in figure 7. These plots illustrate 19 each of the states observed and discussed previously. 20 To help to understand how to interpret the map in figure 6, let us describe 21 two examples of change of state. Suppose we have a flow around the configuration 22 23 with Lx {D 3.8 and Re 175. According to the map in figure 6, this flow will be three-dimensional with shedding regime WG (3d-WG). If the Reynolds number 12 Figure 6 – Map of Reynolds number against centre-to-centre separation showing the possible vortex shedding regimes and the variation of the critical Reynolds numbers of modes T3 and A with Lx {D, in the neighbourhood of the drag inversion range. Symbols are: – mode T3 critical Reynolds numbers, N – mode A critical Reynolds numbers, – primary instability critical Reynolds numbers, – 2-d transition from shedding regime WG to steady flow, – 2-d vortex shedding transition from WG to AG, – 3-d vortex shedding transition from WG to AG, – 2-d vortex shedding transition from AG to WG, – 3-d vortex shedding transition from AG to WG. 1 is gradually increased to 300, the flow will then be located inside the upper grey 2 region in the map, in which the regimes 3d-AG and 3d-WG are possible. Since 3 the previous state of the flow was 3d-WG and the change of Reynolds number was 4 gradual, the flow will keep the same state at Re 5 further increased to 500, the flow will now be in a region of the parameter space 6 where only 3d-AG flows are possible, so the flow will then change to this state. If 7 the Reynolds number is then gradually changed back to 300, the flow will keep the 8 state 3d-AG. The same ratiocination is valid if the separation between the cylinders 13 300. If the Reynolds number is (a) Lx {D 3.5, Re 400: 3d-AG (b) Lx {D 3.5, Re 350: 3d-WG (c) Lx {D 3.7, Re 75: 2d-AG (d ) Lx {D 3.7, Re 85: 2d-WG Figure 7 – Plots of vorticity iso-surfaces illustrating the different states observed in the vicinity of the drag inversion separation, three-dimensional simulations. Translucent surfaces represent iso-surfaces of |ωz |. Solid light grey and dark grey surfaces represent iso-surfaces of negative and positive ωx respectively. 1 is changed instead of the Reynolds number – the flow will always retain its previous 2 state when entering a grey region coming from a white region. 3 It may also happen that the flow at a certain state is taken gradually to a region 4 in which two states are possible, but none of them is the initial state of the flow. 5 In this case the flow will assume the possible state that retains the shedding regime 6 of the initial state. For example, suppose we have a flow around the configuration 14 1 with Lx {D 3.2 at Re 250 with regime 3d-AG; the flow will be in the upper 2 grey region of figure 6. If the Reynolds number is gradually decreased to 150, the 3 flow will then be in the dark grey region immediately below, in which two states are 4 possible, 2d-AG and 3d-WG. However, none of them is equal to the initial state of 5 the flow. So the flow will change to the state that keeps the shedding regime, i.e. 6 the flow will change to 2d-AG. 7 A point worthy of note is that the boundaries of the drag inversion range have 8 different orientations depending on whether the flow is two-dimensional or three- 9 dimensional: the boundaries have a negative slope for two-dimensional flows and 10 a positive slope for three-dimensional flows. This means that once the flow is un- 11 stable to three-dimensional perturbations, the dependence of the shedding regime 12 on the Reynolds number is inverted. It was shown in Carmo & Meneghini (2006) 13 that in two-dimensional flows, an increase in the Reynolds number makes the forma- 14 tion length shorter. This occurs because at higher Reynolds numbers the spanwise 15 vorticity in the shear layers separate from top and bottom of the cylinder wall is 16 stronger, and this stronger vorticity facilitates the interaction between these shear 17 layers. A shorter formation length favours shedding regime WG. In contrast, when 18 the flows were three-dimensional the results obtained by Carmo & Meneghini (2006) 19 showed longer formation lengths, owing to the fact that three-dimensional diffusion 20 and spanwise de-correlation weakened the interaction between the opposite shear 21 layers. The DPIV measurements carried out by Noca et al. (1998) showed that, for 22 the flow around an isolated cylinder, the formation length increases with Reynolds 23 number for p300 Re 1500q, indicating that the three-dimensional effects prevail 15 1 over the two-dimensional ones in this Reynolds number range. The results in figure 6 2 demonstrate that this is also true for the flow around two circular cylinders in tan- 3 dem. This was also one of the conclusions drawn by Papaioannou et al. (2006), who 4 deduced that the variation of the single cylinder formation length and variation of 5 the tandem arrangement drag inversion separation with Reynolds number appeared 6 to be consistent in both two- and three-dimensional simulation results. 7 Using figure 6, we can draw comparisons between the current results and data 8 from earlier research that employed three-dimensional numerical simulations. Deng 9 et al. (2006) performed three-dimensional simulations of the flow around diverse 10 11 12 13 14 15 16 17 tandem configurations at Re 220, using a virtual boundary method. served three-dimensional flow for all configurations with Lx {D Lx {D ¥ They ob- 4, whereas for 2 the flow remained two-dimensional. These results did not depend on the initial conditions. This is entirely consistent with the map in figure 6: for Re the configuration Lx {D 220, 2 is in a zone where only two-dimensional flow in the AG regime is possible, and the configurations with Lx {D ¥ 4 are in a zone where only three-dimensional flows in the WG regime are possible. When analysing the results of Deng et al. (2006) for Lx {D 3.5, it should be borne in mind that the 18 spanwise length of the domain used in their computations was 8D. They observed 19 that the emergence of three-dimensional structures depended on the initial condi- 20 tions: three-dimensional flow was seen to occur if the initial flow field was in the 21 WG regime, whereas any three-dimensional perturbations died out if a flow in the 22 AG regime was used as initial condition. In the map in figure 6, it can be seen 23 that the the flow around the configuration with Lx {D 16 3.5 at Re 220 is in a 1 2 3 region where the possible flows are three-dimensional, either in the AG or in the WG regimes. However, figure 3(b) shows that, for Lx {D at the onset pRe cr 3.5, mode T3 wavelength 217q is λz {D 9.97. Therefore, the calculations performed by 4 Deng et al. (2006) were unable to capture mode T3 instability because the spanwise 5 length of the domain they used was too short. Deng et al. (2006) also tried to find 6 the Reynolds number for which the flow around the configuration Lx {D 3.5 at 7 regime AG would become unstable to three-dimensional perturbations. They ran a 8 series of simulations increasing the Reynolds number in steps of 10, using the final 9 solution of each simulation as the initial condition for the next one. They observed 250, and the wavelength and sym- 10 that the flow became three-dimensional for Re 11 metry of the three-dimensional structures were similar to those of mode A. However, 12 our results show that this transition occurs at Re 13 should be mode T3. Again, this difference can be explained by the short spanwise 14 length of their calculations. We performed additional two-dimensional calculations 15 and found that the upper Re limit of the two-dimensional drag inversion range for 16 17 configuration Lx {D 3.5 is Re 240. 217, and the unstable mode We therefore assume that from Re 240 to Re 250 there was a shedding regime change to WG in the simulations of Deng 18 et al. (2006), and that this was accompanied by the appearance of mode A structures 19 in the flow. 20 21 22 23 In another study that used three-dimensional numerical simulations, Papaioannou et al. (2006) observed regime AG for 250 ¤ Re around configurations with Lx {D ¤ 500 in simulations of the flow ¤ 3.5, whereas regime WG was observed in simu- lations of the flow around configurations with Lx {D 17 ¥ 3.8. The current results are 1 2 3 4 5 mostly in line with this; it can be seen in the map in figure 6 that three-dimensional flow in the AG regime is possible for Lx {D ¤ 3.5 for 250 ¤ Re ¤ 500 and three- dimensional flow in the WG regime is possible for Lx {D ¥ 3.8 for 250 ¤ Re ¤ 475. The only disagreement between our results and those from Papaioannou et al. (2006) is in the regime observed for Lx {D 3.8 at Re 500. For this case, the map in fig- 6 ure 6 indicates that only three-dimensional flow in the AG regime is possible, while 7 Papaioannou et al. report that they found regime WG at the same conditions. A 8 possible reason for this discrepancy is the number of non-dimensional time units 9 for which the flow equations are integrated. We have used 100 non-dimensional 10 time units for all calculations, and we observed that for some of the cases, the 11 change of regime only happened after the equations were integrated for 40 or 50 12 non-dimensional time units. Papaioannou et al. do not report the time length of 13 their calculations. Another possible reason has to do with the size of the domain. 14 The mesh used in Papaioannou et al. (2006) is significantly smaller than that used 15 for the current results. 16 4 17 In this paper, a thorough investigation of the possible flow states in the drag in- 18 version range of the flow around two circular cylinders in tandem at low Reynolds 19 numbers was presented. For the first time, the regions of bi-stable flow were carefully 20 identified, taking into account the hysteresis of the shedding regimes and the influ- 21 ence of the secondary instabilities. The presence of three-dimensional flow structures 22 was observed to induce notable changes in the response of the flow to the variation Conclusion 18 1 of Reynolds number and cylinder separation. Results available in the literature were 2 reviewed in the light of the new data and almost all the observations made by other 3 authors were consistent with the current findings. We believe that the analysis pre- 4 sented helps to improve the understanding of flows with wake interference, and can 5 be very useful for future investigations of other aspects of such flows. 6 References 7 Barkley, D. & Henderson, R. D. 1996 Three-dimensional Floquet stability 8 analysis of the wake of a circular cylinder. Journal of Fluid Mechanics 322, 215– 9 241. 10 11 Blackburn, H. M. 2002 Three-dimensional instability and state selection in an oscillatory axisymmetric swirling flow. Physics of Fluids 14 (11), 3983–3996. 12 Carmo, B. S. & Meneghini, J. R. 2006 Numerical investigation of the flow 13 around two circular cylinders in tandem. Journal of Fluids and Structures 22, 14 979–988. 15 Carmo, B. S., Meneghini, J. R. & Sherwin, S. J. 2009 Secondary instabilities 16 in the flow around two circular cylinders in tandem. Accepted for publication in 17 the Journal of Fluid Mechanics . 18 Deng, J., Ren, A. L., Zou, J. F. & Shao, X. M. 2006 Three-dimensional flow 19 around two circular cylinders in tandem arrangement. Fluid Dynamics Research 20 38, 386–404. 19 1 2 Igarashi, T. 1981 Characteristics of the flow around two circular cylinders arranged in tandem. Bulletin of JSME 24 (188), 323–331. 3 Karniadakis, G. E. 1990 Spectral Element–Fourier methods for incompressible 4 turbulent flows. Computer Methods in Applied Mechanics and Engineering 80, 5 367–380. 6 Karniadakis, G. E., Israeli, M. & Orszag, S. A. 1991 High-order splitting 7 methods for the incompressible Navier–Stokes equations. Journal of Computa- 8 tional Physics 97, 414–443. 9 10 Karniadakis, G. E. & Sherwin, S. J. 2005 Spectral/hp Element Methods for Computational Fluid Dynamics, 2nd edn. Oxford University Press. 11 Meneghini, J. R., Saltara, F., Siqueira, C. L. R. & Ferrari Jr, J. A. 12 2001 Numerical simulation of flow interference between two circular cylinders in 13 tandem and side-by-side arrangements. Journal of Fluids and Structures 15, 327– 14 350. 15 Mittal, S., Kumar, V. & Raghuvanshi, A. 1997 Unsteady incompressible flows 16 past two cylinders in tandem and staggered arrangements. International Journal 17 for Numerical Methods in Fluids 25, 1315–1344. 18 19 20 21 Mizushima, J. & Suehiro, N. 2005 Instability and transition of flow past two tandem circular cylinders. Physics of Fluids 17 (10), 104107. Noca, F., Park, H. G. & Gharib, M. 1998 Vortex formation length of a circular cylinder p300 Re 4000q using DPIV. In Proceedings of Bluff Body Wakes 20 1 and Vortex-Induced Vibration, Washington, DC (ed. P. W. Bearman & C. H. K. 2 Williamson), p. 46. 3 Papaioannou, G. V., Yue, D. K. P., Triantafyllou, M. S. & Karni- 4 adakis, G. E. 2006 Three-dimensionality effects in flow around two tandem 5 cylinders. Journal of Fluid Mechanics 558, 387–413. 6 Tuckerman, L. S. & Barkley, D. 2000 Bifurcation analysis for timesteppers. 7 In Numerical methods for bifurcation problems and large-scale dynamical systems 8 (ed. E. Doedel & L. S. Tuckerman), IMA volumes in Mathematics and its appli- 9 cations, vol. 119, pp. 543–556. New York: Springer Verlag. 10 11 12 13 Williamson, C. H. K. 1988 The existence of two stages in the transition to threedimensionality of a cylinder wake. Physics of Fluids 31 (11), 3165–3168. Williamson, C. H. K. 1996 Three-dimensional wake transition. Journal of Fluid Mechanics 328, 345–407. 351 Zdravkovich, M. M. 1977 Review of flow interference between two circular cylin- 352 ders in various arrangements. ASME Journal of Fluids Engineering 99, 618–633. 21