Accepted Manuscript

Bifurcation and stability analysis with the role of normal form symmetries
on the harmonic streamwise forced oscillation of the cylinder wake

N. Nabatian, N.W. Mureithi

PII: S0997-7546(17)30207-8
DOI: https://doi.org/10.1016/j.euromechflu.2018.02.003
Reference: EJMFLU 3264

To appear in: European Journal of Mechanics / B Fluids

Received date : 29 April 2017

Revised date : 17 February 2018
Accepted date : 20 February 2018

Please cite this article as: N. Nabatian, N.W. Mureithi, Bifurcation and stability analysis with the
role of normal form symmetries on the harmonic streamwise forced oscillation of the cylinder
wake, European Journal of Mechanics / B Fluids (2018),
https://doi.org/10.1016/j.euromechflu.2018.02.003

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form.
Please note that during the production process errors may be discovered which could affect the
content, and all legal disclaimers that apply to the journal pertain.
 Bifurcation and Stability Analysis with the Role of Normal Form

Symmetries on the Harmonic Streamwise Forced Oscillation of the

Cylinder Wake

N. NABATIAN1, N. W. MUREITHI2
1
Faculty of Mechanical and Energy Engineering, Shahid Beheshti University, A. C., Tehran, Iran
2
Department of Mechanical Engineering, Ecole Polytechnique,
Montreal, QC, H3T 1J4, Canada

Abstract

The dynamics of the cylinder wake subjected to harmonic inline oscillation is investigated in this work.

Two-dimensional numerical computations are performed for Re=200 to find the effect of inline sinusoidal

oscillation on the flow pattern leading to the variation of cylinder lift, drag and wake shedding frequency.

Two complex primary modes of the v-velocity field, resulting from a proper orthogonal decomposition (POD)

analysis, are considered to model and predict the nonlinear forced wake dynamics governed by the interaction

of vortex shedding modes. The equivariant bifurcation theory in the presence of normal form symmetry

O (2)  S 1 is employed to classify the mode interaction solution branches with respect to lower order

symmetries. The coupled amplitude equations are developed with the frequency saturation information

included by the addition of complex coefficients. The symmetry-based model is expanded up to 7th power

thus including spatio-temporal effects. The coefficients of the model are obtained from CFD simulations. The

novelty of this work is that the amplitude equations are derived purely from the mode symmetries and not

from a Galerkin reduction of the Navier Stokes equations.

The inline cylinder oscillation effect on the wake dynamics is captured by bifurcation analysis of this

model under the variation of the two linear coefficients. As the oscillation amplitude increases, two limit

cycles of the model undergo a symmetry-breaking bifurcation leading to a quasi-periodic state. For amplitude-

1
Corresponding author.
E-mail address: [email protected]
2
Corresponding author.
E-mail address: [email protected]

1
to-diameter ratio A D  0.5 , the quasi-periodic state undergoes a torus doubling bifurcation. The bifurcated

mode S frequency matches the lift coefficient shedding frequency for A D  0.5 , obtained from the

numerical (CFD) computations. The bifurcated modulated travelling waves have mode S as the dominant v-

velocity mode, which confirms the symmetric period-doubled quasi-steady v-velocity pattern observed in

CFD. The stability behavior of the bifurcated solution branches is also captured in the Poincare plane. The

model correctly predicts the bifurcation sequence of the wake dynamic. This knowledge opens the possibility

to develop low order controllers for the wake-structure interaction.

Keywords: Normal form symmetry; equivariant bifurcation theory; torus doubling bifurcation

1 Introduction

The wake behind a cylinder is created via a Hopf bifurcation in the first instability

region. The forces generated from vortex shedding can lead to structural oscillations and

resonance at lock-in [1-3]. Forced cylinder oscillation usually excites the two fastest

growing modes of the wake. Two global modes can result: a reflection symmetric mode in

which a pair of vortices shed symmetrically and the Karman mode, which has spatio-

temporal symmetry and alternate vortex shedding. The symmetric mode is intrinsically

unstable and is quickly replaced by the Karman mode, unless it is stabilized by cylinder

motion. The interaction between the vortex shedding and structure motion can lead to self-

excited vibration of the cylinder and changes in the wake pattern. The wake pattern

formation can be used to identify the global instabilities that develop in the wake behind a

circular cylinder and follow its dynamics from simple to chaotic behavior. In view of the

complex dynamics of the fluid-structure interaction problem, the cylinder motion can be

controlled to allow a study of the wake response to a prescribed cylinder motion.

Thus, the forced oscillation of the wake is an approach to study the cylinder oscillation

2
effect on the wake pattern. The periodic vortex street created due to the wake instability is

saturated. Thus, for Reynolds number not too far above the critical Reynolds number, the

shedding frequency varies smoothly with Re suggesting that linear Floquet instability

theory is applicable for vortex shedding dynamic analysis [4-7]. The nonlinear interaction

of the wake modes was modeled using the Stuart-Landau equation by Provansal et a1. [8].

The Landau coefficients were calculated from the 2D simulations [9].

To better understand the wake dynamics, Williamson and Roshko [10] excited the wake

by external periodic forcing. This approach is the key to understand the wake-structure

interaction during vortex-induced vibration (VIV). In experiments, a known transverse

forced oscillation was applied to the cylinder at different forcing amplitude and frequency

ranges. Williamson and Roshko showed that the vortex shedding patterns are dependent on

the forcing frequency, amplitude (A/D), direction of oscillation and Reynolds number. By

changing the parameters different modes defined by their symmetries may be observed; the

modes were labeled 2P, 2S, P+S and 2P+2S based on the number and combination of the

vortices shed on the sides of the cylinder per forcing cycle. They also observed the

transition from 2S to 2P mode, which causes a half period shift between the cylinder

motion and fluid forces [11-12].

Barkley et al. [13] proposed the theory of bifurcations with symmetry (equivariant

bifurcation theory) for the development of the coupled amplitude equations. The symmetry-

based amplitude equations describe the nonlinear interactions between two naturally

formed 3D vortex shedding modes of a fixed cylinder. The higher order terms which could

introduce complex spatio-temporal effects on dynamics, were ignored in their model. Later,

3
Sheard et al. [14] expanded the amplitude equations by coupled Landau equations including

frequency information with the addition of complex coefficients. The model predicts the

nature of the Strouhal–Reynolds number profile of the circular cylinder wake at transition

deduced from the interaction of 3D wake modes [15-16].

Mureithi et al. [17] used the symmetry-equivariant theory to study the forced Karman

wake. A symmetric oscillation was applied and its interaction with the vortex shedding

mode modeled by symmetry equivariant bifurcation theory in the form of a pair of low

order discrete amplitude equations. A qualitative analysis of these equations showed that

the Karman mode bifurcation is affected by the forcing oscillation amplitude represented by

mode S. Both subharmonic and superharmonic excitations of the Karman wake were

studied experimentally. A quantitative analysis was also done by numerical simulation of

the 2D wake model. Rodriguez and Mureithi [18] performed CFD computations of the

wake flow undergoing periodic excitation. The dominant modes were determined by POD

analysis [19‐23] and coefficients of the amplitude equations obtained. They observed the

period doubling for the inline excitation as the dominant phenomena, which confirms the

Williamson experiment [24].

In the present work, the dynamics of the wake flow at Re=200 when forced to oscillate

in the inline direction at the vortex shedding frequency is studied by developing a model

using the equivariant bifurcation theory. The symmetries of the two fastest growing modes

formed from the harmonic inline oscillation are used to describe the wake mode

interactions. The coefficients of the amplitude equations are considered complex to include

the frequency saturation and solved for higher orders to cover the spatio-temporal effects.

4
The first two complex primary modes, which exist for various oscillation amplitudes, are

employed to model the wake dynamics. The linear terms of the amplitude equations are

affected by the variation of the streamwise oscillation and thus considered as the

bifurcation parameters. The goal of this work is to find the sequence of bifurcations

observed in CFD and previous experimental results through bifurcation analysis of the

symmetry-based model.

2 Numerical Computations

The two-dimensional flow over a circular cylinder with a diameter D=25.4 mm under

forced inline oscillation is simulated using ANSYS CFX. Simulations were performed for

Re=200 at two-dimensional domain with boundaries extended 15D upstream, 40D

downstream and 20D on the lateral sides given in Fig. 1. A structured mesh with 109692

elements corresponding to 220648 nodes was used. The excitation frequency is equal to the

natural shedding frequency and oscillation amplitude ratio varies from zero to 0.5D. The

incompressible Navier-Stokes equations given in Eq. (1) are solved using the implicit

hybrid finite-element/finite-volume approach to achieve the pressure-velocity coupling.

The second-order accurate high-resolution scheme for space domain and a second-order

accurate backward Euler scheme for time discretization are used in the simulation.

.V  0
V 1 (1)
 V .V   P   2V
t 

where V  ( u , v ) . The forced vibration is implemented by velocity perturbation in the inlet

boundary condition instead of using a moving mesh. It is equivalent to setting a rigid

5
cylinder in an oscillated flow field. Then, the
t inlet fflow bounddary condition is

u  U  A sin( t ) and v  0 , wheree A is the oscillation amplitude ,   2  f e is the

oscilllation frequuency with fe fs  1 annd an inertiial source term

t  A2 cost is addded to

the xx-momentum
m equationn. The cyliinder surfacce was deffined with no-slip booundary

condiition. The syymmetry coonditions were employeed for the laateral bounddaries, wherreas the

presssure outlet w me step, dt  0.0078 was used

was set for tthe outlet coondition. A constant tim

for siimulation off the unsteaady wake floow using (SST) k   tturbulent model. Conveergence

tests were performed for thhe fixed cyllinder case and the meeasured forcce coefficiennts and

Strouuhal numbeer have goood agreemeent with the previous numerical and experiimental

resultts presentedd by Mureithhi [18].

F
Fig. 1. The struuctured mesh oof the 2D dom
main.

Thhe POD method is appllied to analyyze the recoorded data oof the v-veloocity perturrbations

in 1000 nodes on the line 100D downstreeam of the fflow for 40 steady cyclles which iss stored

in a matrix. Firrst, the meaan value off each coluumn is subttracted to find
f the v-vvelocity

pertuurbation valuue as in Eq. (2) and the results are stored in maatrix A .

6
v  ( y , t )  v ( y , t )  v m ( y ), (2)

The perturbation velocity matrix can then be written as:

v ( y , t )   k 1 a k (t ) k ( y ),
100
(3)

where each mode includes a spatial eigenvector  k ( y ) , called topos and a temporal one

a k (t ) , called the chronos. The  k ( y ) represents the velocity profile of 100 points at a

specific time and chronos a k ( t ) describes each point time history velocity. The matrix A

(5500,100) for each case of the forced oscillation is reduced by singular value

decomposition (SVD) to identify the topos and chronos of the primary modes in the

following method. The different topos and chronos for various amplitudes of oscillation

show different behaviors of the flow field, under the effect of cylinder oscillations. Since

POD gives the normalized topos, chronos contains the amplitude evolution of the mode.

3 Numerical Solution and POD Modes

The eigenvalue spectrum of transverse velocity (v) given in Fig. 2(a) shows that the first

two modes contain 99% of the energy. Since the fluctuating v-velocity data is complex each

dominant mode represents pair of similar patterns shifted spatially. Each mode pair

represents a travelling wave corresponding to the convective nature of the flow. The

singular values of Fig. 2(b) related to the unsteady v-velocity for A D  0.5 are still

concentrated around the first two modes and these modes have similar spatial structures to

those derived for the stationary cylinder. These two eigenmodes are then associated with

the wake instability.

7
 (a) (b)
Figg. 2. Singular vvalues of v-vellocity for (a) sstationary cylinnder A D  0
0.0 , (b) for oscillation ampplitude
A D  0.5 .
Thhe symmetrry propertiess of each toopos are utillized to idenntify each eeigenmode clearly.

As shhown in Figg. 3(a) the fiirst mode top

opos is symm
metric with the dominaant frequency equal

to thee natural sheedding frequuency fv reepresented bby dimensioonless frequeency fv D U  0.2

, whiile Fig. 3(c)) represents the antisym

mmetric spaatial pattern with the frrequency 2 fv . For

the foorcing ampllitude A D  0 .1 7 5 thee quasi periodic nature of both moodes is appaarent in

Fig. 44(b) and 4(dd) due to thee existence oof the otherr peaks in adddition to fv and 2 fv . This is

suppoorted later in the Poinccare maps off Fig. 18(c) and 18(d). For the casse of A D  0.5 the

quasii-periodic sttate is replaaced by period-doubledd torus wherre the dominnant dimenssionless

frequuency of thhe first sym

mmetric moode becomees 0.1, the same as tthe bifurcaated lift

coeffficient frequuency (See F

Fig. 9(a)-9(ff)).

8
 (a) (b)

(c) (d)
Fig. 3.. The spatial and temporal basic
b functionss of two dominnant modes foor A D  0.0 (a) first toposs (b) first
c
chronos (c) seccond topos (d)) second chronnos.

(a) (b)

9
 (c) (d)
Fig. 4. minant modes ffor A D  0.175 . (a) first toopos (b)
4 The spatial and temporal basic functionns of two dom

firsst chronos (c) second topos ((d) second chrronos.

(a) (b)

(c) (d)
Fig. 5. The spatiall and temporall basic functioons of two domminant modes for A D  0..5 . (a) first toppos (b)
firsst chronos (c) second topos ((d) second chrronos.

10
 The perturbed transverse velocity field can be reconstructed using these two eigen-

modes in order to relate the POD modes and v-velocity contours obtained from the

numerical simulation. The temporal evolution of the two dominant modes, resulting from

the product of the corresponding topos and chronos, are given in Fig. 6(a)-6(h); the figures

show the interaction between the symmetric and antisymmetric modes. A disordered

structure is observed in Figs. 6 (c,d) where the spacing between the vortices is different and

the flow is not periodic over a forcing oscillation cycle. The irregular v-velocity pattern is

due to the competition between symmetric and asymmetric modes leading to the presence

of a weaker peak at fD / U  0.013 seen in Fig. 4(b). This frequency is lower than the

excitation frequency fe and affects the wake spatial periodicity. The quasiperiodic state is

then synchronized to the subharmonic frequency with Tb 15Te where Te 1/ fe . The

instantaneous v-velocity contours over 15 excitation periods corresponding to the beating

period is shown in Fig. 7. The vortex initially formed from the upper side of the cylinder is

later elongated along the cylinder surface. During this upper vortex elongation, a vortex

with opposite sign is formed simultaneously from the lower side of the cylinder. At half of

the beating cycle, the upper vortex separates and is shed downstream. Next the lower vortex

is also shed downstream. After 15 oscillation cycles a new vortex initiates from the upper

side with the same structure of first excitation cycle, and repeats the beating period.

11
 (a) (b)

(c) (d)

(g) (h)
Fig. 6. Temporal evvolution of tw
wo primary moodes at differennt oscillation aamplitudes (a,b) fixed cylindder case

(c,d) A D  0.175 (g,h)) A D  0.5 .

t=t0 t=
=t0+Te

12
 t=
=t0+2Te t=t0+3Te

t=tt0+4Te t=
=t0+5Te

t=
=t0+6Te t=
=t0+7Te

t=tt0+8Te t=
=t0+9Te

t=t00+10Te t=t00+11Te

t=t0+
+12Te t=tt0+13Te

t=t0+
+14Te t=tt0+15Te

Fig. 77. The velocityy contours forr forced oscillaation amplituddes A D  0.1175 and f e f s  1 over Tb  15Te .

Foor A D  0.55 a torus ddoubling biffurcation iss observed. Although the antisym
mmetric

modee pair is eneergized by thhe inline osccillation, as shown in Fig. 5(d), thee energy of tthe first

mode. Hencce, the basicc v-velocity pattern

modee pair is agaain higher thhan that of tthe second m

is syymmetric but
b with a larger sppacing betw
ween vorticces as shoown in Figg. 6(g)

St.  0.1 . Thhe time seqquences off the v-

correesponding too the loweer Strouhal number, S

veloccity contourrs for the fixxed cylinderr case and thhe A D  0 .5 case are shown in Fig. 8. It

is cleear that the vvortex formation lengthh is enhanceed for A D  0.5 and thhe wake struucture is

13
perioodic every tw
wo cycle off oscillationn, which confirms the v-velocity ttemporal evvolution

deducced from PO
OD.

t=t0 t=t0+Te t=
=t0+2Te

(a) (b) (c)

t=tt0 t=t0+Te t=t0+2Te

(d) (e) (f)

Fig. 88. The v-veloccity contours ffor the stationaary cylinder (aa-c) and forcedd oscillation aamplitudes A D  0.5
(dd-f) for f e f s  1 over two ooscillation perriods.

Thhe lift coeffficient dynam

mics dependdence on thhe forcing am
mplitude is also shownn in Fig.

9. Ass the amplitude of osccillation inccreases, the steady periiodic lift cooefficient beecomes

The dominannt peak of tthe lift coeffficient for A D  0.5 iss precisely equal
quasii-periodic. T e to

the peak
p of the vv-velocity first
f mode ggiven in Figg. 5(b). The model com
mparison plootted in

Fig. 118(e) and 188(f) on the Poincare

P maap indicates the same peeriod-2 statee.

(a) (d)

14
 (b) (e)

(c) (f)
Fig. 9. Phase planne of the lift cooefficient for ((a) fixed cylinder case (b) A D  0.175 ((c) A D  0.5
5 . Lift
coefficieent frequency peak for (d) fiixed cylinder ccase (e) A D  0.175 (f) A D  0.5 .

4 D
Derivation of Symmettry Based M
Model

Ass shown inn the v-veloocity singullar values of

o Fig. 2(aa,b) the harrmonic streaamwise

o the v-velocity whilee the other modes

oscilllation mainnly excites ttwo dominaant modes of

remain stable. The two moddes account for over 900% of the tootal energy in the flow.. A low

orderr model baseed on these two modes therefore caapture the oone dimensioonal spatial pattern

obserrved as the forcing am

mplitude biffurcation paarameter is varied. Thhe general solution
s

descrribing the 1D
1 oscillatoory pattern that sets inn at the bifurcation point, with ddefined

temporal and spaatial periodss can be exppressed as

15
 x x
i( s t ) i( k t )
s k
v( x, y, t )  S (t )e f ( y)  K (t )e g ( y)  c.c., (4)

where S (t ) and K ( t ) are the time dependent amplitudes and f ( y ) and g ( y ) are the

corresponding eigenfunctions. Here c.c. indicates the complex conjugates. From PSD in

Fig.3, the frequency ratio of these modes is  k  s  2 . In equation (4), an assumption is

made regarding to the spatial form of modes S and K for simplicity. The modes are

considered to be perfect spatial waves with the simple wavelength ratio

 s  k   k  s  m n  2 ; the wavelength ratio is therefore inferred from the frequency

ratio by assuming perfect spatio-temporal modal symmetries. These assumptions also

means that modal symmetries are assumed to be the same before and after the cylinder. The

validity of these assumptions is confirmed a posteriori by confirming the predictive ability

of the reduced order model. Based on the above simplifications, equation (4) can therefore

be written as

v ( x , y , t )  S ( t ) e i ( k x   t ) f ( y )  K ( t ) e 2 i ( k x   t ) g ( y )  c .c ., (5)

where the complex amplitudes ( S , K ) satisfy equations that are equivariant with respect to

the related action of the symmetry group. Each mode has its own symmetry group. The S

mode satisfies the following symmetry properties

v ( x , y , t )  v ( x ,  y , t )  v ( x   s , y , t )  v ( x , y , t   s ), (6)

while the K mode has symmetries:

v( x, y, t)  v( x,  y, t   k 2)  v( x  k 2 ,  y, t) (7)

where  s  2  s , k  2  k . Thus the overall symmetry group of the system is

16
 : y   y   (S , K )  (S ,  K )
T : x  x  ls  T ( S , K )  ( Seinl , Keiml ) (8)
 : t  t   s 2   ( S , K )  ( Se , Ke ) in im

The symmetry  acts through reflection represented as Z 2 ( ) , T acts through translation

represented as S O ( 2 ) and the  phase shift symmetry, represented by S 1 . Thus, the

overall symmetry is SO(2)  Z2 ( )  S  O(2)  S . To derive the amplitude equations,

1 1

equivariance under the translations in x, t and reflection lead to the invariants:

2 2
S , K , ( K n S m )2 (9)

and four equivariants as:

2 m 1 2 n 1
(S , K 2n
S ), ( K , S 2 m K ) (10)

The derivation of the invariants and equivariants is presented in appendix A . It follows

that the most general form of the amplitude equations is:

dS
 [(0  i 0 )  (1  i 1 ) K  (2  i 2 ) S  (3  i 3 )( K n S m )2 ]S 
2 2

dt
(11a)
[(0  i0 )  (1  i1 ) K  (2  i2 ) S  (3  i3 )(K n S m )2 ]K 2n S 2m1
2 2

dK
 [(0  i0 )  (1  i1 ) K  (2  i 2 ) S  (3  i3 )(K n S m )2 ]K 
2 2

dt
(11b)
[(0  i0 )  (1  i1 ) K  (2  i2 ) S  (3  i3 )(K n S m )2 ]K 2n1S 2m
2 2

The amplitude equations 11(a,b) model the general nonlinear interaction between mode

S and mode K under general forcing where the forced wake remains dominated by these

two modes. In the present work, we consider the case where the cylinder wake is forced at

the excitation frequency f e  f s via forced cylinder motion in the flow direction. The

model above is used to study the resulting forced wake dynamics. The effects of forcing

17
amplitude variation with fixed forcing frequency are implicitly captured by the model

bifurcation coefficients  0 ,  0 . (see Section 4.4).

Since the coefficients of the amplitude equations are complex, the frequency variation of

the transition modes can be analyzed. The solution branch types and their stabilities can be

determined as functions of the coefficients.

4.1 Truncation of the amplitude equations to third order O(3)

The mode interaction equations first truncated to third order are

dS
 [( 0  i 0 )  (1  i 1 ) K  ( 2  i 2 ) S ]S
2 2
(12a)
dt

dK
 [( 0  i 0 )  ( 1  i 1 ) K  ( 2  i 2 ) S ] K
2 2
(12b)
dt

The real part of the first complex coefficient describes the growth rate of each mode due to

the linear instabilities of the cylinder wake. The cubic terms of the Landau model describe

the saturation of the modes S and K . The coupling terms describe the mode interaction

effect. If the sign of the cubic term is positive, the bifurcation is supercritical while if

negative, it is subcritical. These equations can be put into amplitude and phase form by

writing S  q e i , K  r e i where q , r  0 and 0   ,  2 . Separating out the real and

imaginary parts of the equations leads to

dq
 ( 0  1r 2   2 q 2 ) q (13a)
dt
dr
 ( 0   1 r 2   2 q 2 ) r (13b)
dt
d
 sat , s    0   1r 2   2 q 2 (13c)
dt
d
 sat , k    0   1r 2   2 q 2 (13d)
dt

18
 The individual imaginary parts represent the saturated frequency for each mode. The

phase equations d  d t and d d t do not appear in the amplitude equations and depend

only on q and r , so if q and r are stable to perturbation the phases will also be stable. This

means that for stability analysis just the amplitude values variation is considered. However,

to see the frequency effects in the amplitude equation, the equation must be expanded up to

higher order and the stability analysis considered for this case.

4.2 Truncation of the amplitude equations to order seven O(7)

The phases of modes only affect the dynamics through the terms involving m and n and

with truncations up to low orders, their effect is ignored. The frequencies of these modes

vary along the transition between the solution branches while undergoing the bifurcation.

As mentioned, the amplitude equations at third order are not sufficient to describe the

hysteresis. Hence here, the amplitude equations are expanded up to 7th order giving

dS
 [0  i 0  (1  i1 ) K  (2  i 2 ) S  (3  i 3 )(KS 2 )2 ]S 
2 2

dt (14a)
[0  i0  (1  i1 ) K  (2  i2 ) S ]K S
2 2 2 3

dK
 [0  i 0  (1  i1 ) K  (2  i 2 ) S  (3  i3 )( KS 2 )2 ]K 
2 2

dt (14b)
[0  i0  (1  i1 ) K  (2  i2 ) S ]KS
2 2 4

The complex equations can be rewritten in polar form by setting S  q e i , K  r e i and

separating the real and imaginary parts of the equations to get

dq
 [ 0  1r 2   2 q 2  ( 3 cos 2   3 sin 2 ) r 2 q 4 ]q 
dt (15a)
[( 0  1r 2   2 q 2 ) cos 2  ( 0  1r 2   2 q 2 ) sin 2 ]r 2 q 3

19
dr
 [ 0  1r 2   2 q 2  ( 3 cos 2   3 sin 2 ) r 2 q 4 ]r 
dt (15b)
[(  0  1r 2   2 q 2 ) cos 2  (  0  1r 2   2 q 2 ) sin 2 ]rq 4
d
  0   1r 2   2 q 2  [ 3 cos 2   3 sin 2 ]r 2 q 4 
dt (15c)
[( 0  1r 2   2 q 2 ) sin 2  ( 0  1r 2   2 q 2 ) cos 2 ]q 2 r 2
d
  0   1r 2   2 q 2  [ 3 cos 2   3 sin 2 ]r 2 q 4 
dt (15d)
[(  0  1r 2   2 q 2 ) cos 2  (  0  1r 2   2 q 2 ) sin 2 ]q 4

Combining (15-c) and (15-d) gives

d
  0  2 0  (1  2 1 )r 2  ( 2  2 2 )q 2  [( 3  2 3 ) cos 2 
dt
( 3  23 )sin 2 ]r 2 q 4  [( 0  1r 2  2 q 2 ) cos 2  (  0  1r 2   2 q 2 ) sin 2 ]q 4  (15e)
2[(0  1r 2  2 q 2 )sin 2  ( 0  1r 2   2 q 2 ) cos 2 ]q 2 r 2

where     2 is the relative phase and therefore the system is in fact three

dimensional. As shown in Eq. (15e) the evolution of the individual phases are functions of

r , q and  . Therefore, the fixed points of Eq. (15e) with  ,  0 correspond to periodic

solution of Eqs. (14a) and (14b) while the limit cycles of Eq. (15e) relate to tori in the set of

Eq. (14).

4.3 Solution branches

As the symmetry group O (2)  S 1 is continuous, there exist an infinite number of

solutions when the symmetry is broken. Using the equivariant bifurcation theory the

maximal subgroups with lower order symmetries and fixed-point dimensions are identified.

The steady solution branches of the third-order amplitude equations, corresponding to

dq dt  dr dt  0 lead to three types of solutions:

(i) Trivial solution: q  r  0 ,

20
 (ii) Pure modes: These solutions are

0 d 
q  0, r 2  ,  0  1 0 (16a)
1 dt 1
or
0 d  (16b)
r  0, q2  ,  0  2 0
2 dt 2

The general form of Eq. (15b) is v( x, y, t )  0 2 ei ( kx t 0 ) f ( y) which has isotropy

group 2 . The definition of 2 , :0 2 is that a

translation through  has the same effect as a phase shift  . Also, the pure mode K

solution is v( x, y, t )   0 1 e2i ( kx t  0 ) g ( y) with isotropy group 2 , . The

pure modes are travelling waves and have a one-dimensional fixed point subspace. They

maintain the spatio-temporal symmetry at all times, while oscillating periodically.

(iii) Mixed modes: These solutions can be calculated by solving the coupled set of

equations (13). The fixed-point amplitude values are

01  10
2
qTW  (17a)
12  21
1  01  10 
r2        (17b)
1  0 2  12  21 

The phase speeds of the mixed travelling waves can be calculated by substituting the values

of r and q in Eq. (13c) and (13d). The corresponding flow field will be:

v(x, y, t)  qTW ei(ks xst 0 ) f ( y)  rTW ei(kk xkt 0 ) g( y) (18)

The mixed-mode solution has maximum 2 symmetry subgroup. Since the analytical

solution of the relative phase equation does not exist, it is not possible to distinguish the

21
steady periodic solutions from the mixed modulated waves (torus). As mentioned earlier, in

order to determine the types of mixed modes, the amplitude equations should be expanded

up to higher orders. For the seven-order amplitude equations, the steady state solution

branches are again found by setting dq dt  dr dt  d  dt  0 . In addition to the trivial

solution, two types of solution branches are possible. The basic steady solution branches

can be listed as:

1- Pure modes: these solutions are similar to those from the third-order model. These

solutions are given by (S , K )    0  2 e i f ( y ), 0  and

 
( S , K )  0,  0  1 e 2 i g ( y ) . They are obtained by solving the set of Eq. (13)

determining the amplitude and phase of the pure mode.

2- Mixed modes: There are two types of mixed modes.

 Travelling waves (TW). These are solutions of the form ( q e i , re 2 i ) . The solutions

are deduced from solving the set of Eq. (15) to get the fixed points. Since

d  dt  2 d dt  0 , but d  dt  0 , the travelling waves are the periodic orbits

within the Eq. (14a) and (14b).

 Modulated travelling waves (MTW). MTW correspond to tori when d  dt  0 .

4.4 Coefficients calculation using dominant POD modes of the transverse velocity field

from CFD

The amplitude equation coefficients are calculated using the chronos of two primary

modes obtained from the POD analysis. The equations have been developed in the case of

complex amplitudes, so the data are first transformed into complex signals by Hilbert

transformation. The Landau coefficients, which cover the initial transition until the steady

22
periodic region, are calculated for each mode independently at each of the oscillation

amplitude. Since the data is recorded for 40 cycles, the system of Eq. (19) is over-

determined and the constants can be calculated using the least squares method.

 S1 S1 S1 
2
 S 2  S1 
   
      
  0  i 0
  Si 1  Si 
2
Si S i    (19a)
 Si   i 2   
    2   
   S 40  S39 
 S39 S39 S39 
2

K K1 K1 
2
 K 2  K1 
 1   
      
 2   0  i 0    K  K  (19b)
 Ki Ki Ki    i 1
 1  i1  
i

      
   K 40  K 39 
 K 39 K 39 K 39 
2

The coupling coefficients are also determined by the least squares method. These

coefficients are calculated at the oscillation amplitudes where mode K is initially observed

and the last time that mode S occurs. Then, the known Landau coefficients of mode S

for A / D  0.5 are substituted in the Eq. (14a) and the coupling coefficients are obtained

from the over-determined system given by Eq. (20a). Applying the same method for the

fixed cylinder case, the coupling coefficients of K mode are achieved. Thus the critical

amplitude for mode K is the fixed cylinder case, A / D  0.0 where mode K is weak and

for mode S , A / D  0.5 where its energy is reduced. The deduced coefficients are

presented in Table 1.

23
  K1 2 S1 ( K1S12 )2 S1 K12 S13
2
K1 K12 S13 S1 K12 S13   1  i 1 
2

  
        3  i 3 
 
Si Ki 2 Si 3  0  i 0  
2 2 2
 K i Si ( K i Si 2 ) 2 Si K i 2 Si 3 K i K i 2 Si 3
 
       1  i1 
  
 K39 S39 ( K39 S39 2 )2 S39 K39 2 S393 K39 K39 2 S393 S39 K39 2 S393  2  i 2 
2 2 2

(20a)
 S2  S1  [(0  i 0 ) S1  (2  i 2 ) S1 2 S1 ]A D 0.5 
 
  
 
 Si 1  Si  [(0  i 0 ) Si  (2  i 2 ) Si Si ]A D 0.5 
2

  
 
 S40  S39  [(0  i 0 ) S39  (2  i 2 ) S39 S39 ]A D 0.5 
2

 S1 2 K1 ( K1S12 )2 K1 K1S14
2
K1 K1S14 S1 K1S14   2  i 2 
2

 
         3  i 3 
 
S i K i Si 4    0  i  0  
2 2 2
 Si K i ( K i Si 2 ) 2 K i K i Si 4 K i K i Si 4
        1  i 1 
  
 S39 K39 ( K39 S39 2 ) 2 K39 K39 S39 4 K39 K39 S39 4 S39 K39 S39 4    2  i 2 
2 2 2

 K 2  K1  [( 0  i 0 ) K1  (1  i1 ) K1 2 K1 ]A D 0.0  (20b)

 
  
 
 Si 1  Si  [( 0  i 0 ) Ki  (1  i1 ) Ki Ki ]A D 0.0 
2

  
 
 S40  S39  [( 0  i 0 ) K39  (1  i1 ) K39 K39 ]A D 0.0 
2

Table 1. The coefficients of the amplitude equation deduced from the least squares method.
0  i  0 1  i  1 2  i  2 3  i  3 0  i  0 1  i 1 2  i  2
S 0.0926- 0.0- - - 985.32-
985.32-3281.8i 95977+37561i
0.08i 85.57i 2.04+1.39i 720.41+197.49i 3281.8i
 0  i 0 1  i 1  2  i 2  3  i 3 0  i 0 1  i 1 2  i 2
K -80.63- -4.15- - 819.87-
0.25+0.25i -57.45+2.96i 64593+115600i
37.59i 5.98i 198880+20991i 427.48i

The real parts of both modes approximated from the amplitude equations are plotted in

Fig. 10. Both the frequency and amplitude of the modes match well with the simulation

24
resultts. Now by variation oof the bifurccation param
meters due to the forciing oscillatiion, the

solutiion branchees can be dettermined.

(a) (b)
Fig. 10.
1 (a) The chrronos of modees S and K from CFD, (b) Deduced
D S and K modes aapproximated from
f the

leasst squares methhod of the dom

minant chronoos data.

5 B
Bifurcation
n and Stabillity Analysiis

Thhrough the coefficientss found from

m the chronos of the first two prrimary moddes, the

linearr terms of the amplittude equatioons are moostly influennced by thhe harmonicc inline

oscilllation. The nature of thhe interactioon between different soolution brannches as thee linear

bifurccation param
meters varyy can be dettermined whhile the othher coefficieents are keppt fixed.

Thus the bifurcaation analysiis is consideered for the linear term

ms variation,, obtained frrom the

numeerical compputations. A mplitude of oscillation increases, 0 decreases,  0

As the am

increases while  0 ,  0 remaain unchangged. Thereffore, the biifurcation iss caused frrom the

smoooth variationn of these two linear bifurcationn parameters 0 and  0 . The two steady

a  0  0.0836 and
travelling wavess of Fig. 10(b) lose sstability viaa a Hopf biifurcation at

 0  0.29 as shoown in Fig. 11. The toruus doubling bifurcationn is approachhed at  0  0.0826

25
and  0  0.46 ; ttwo period--doubled tori are deduuced as shown in Fig. 12. Based on the

princciple solutionn branch claassification,, the bifurcaated responsses of the m

mode interacttion are

mixed modulated travellingg waves (MT

TW) with 2 symm
metry group. As shown in Fig.

12 thhe bifurcateed mode S has higher energy, heence the larrger amplituude in com
mparison

with the bifurcaated mode K , which leads to thhe reflectionn symmetryy Z 2 ( ) beccoming

domiinant in the flow field. The quasi-pperiodic sollution of thee complex aamplitude inn terms

of moode S and K are obserrved in Figss. 13 and 144.

modes S and K at
Fig. 11. Quasi-perioodic state of m Fig. 12. Period-douubled tori of modes
m S and K at
 0  0..0836 ,  0  00.29 .  0  00.0826 ,  0  0.46 .

Fig. 13. Complex am

mplitude of biifurcated mode S at Fig. 14. Complex amplitude of bifurcated moode K at

 0  00.0826 ,  0  0.46 .  0  0.0826 ,  0  0.46 .

26
 Thhe bifurcation analysiss in the mixxed-mode region
r is lim
mited to thee variation of two

meters 0 annd  0 as meentioned above. The sttable limit cycle

param c of thee amplitudees of S

and K modes arre given in thhe followinng plots.

Fig. 15. Limit cyclle of amplitude term of modde S . Figg. 16. Limit cyycle of amplituude term of moode K .

Thhe stability of the orbitt is analyzed by calculating the Pooincare mapp. The Jacoobian of

Poincare maap yields the monodrom

the P my matrix. For the staability analyysis, the am
mplitude

written in thhe Cartesiann form usingg S  x  iy and K  z  ih .

equattions are rew

dx
 ( 0  1 ( z 2  h 2 )   2 ( x 2  y 2 )  3 A   3 B) x  ( 0   1 ( z 2  h 2 ) 
dt
 2 ( x 2  y 2 )  3 B   3 A) y  (0  1 ( z 2  h 2 )  2 ( x 2  y 2 ))C  (21a)
( 0  1 ( z  h )   2 ( x  y )) D
2 2 2 2

dy
 (0  1 ( z 2  h 2 )   2 ( x 2  y 2 )  3 A   3 B) y  ( 0   1 ( z 2  h 2 ) 
dt
 2 ( x 2  y 2 )  3 B   3 A) x  (0  1 ( z 2  h 2 )  2 ( x 2  y 2 )) D  (21b)
( 0  1 ( z  h )   2 ( x  y ))C
2 2 2 2

dz
 ( 0  1 ( z 2  h 2 )   2 ( x 2  y 2 )   3 A   3 B) z  ( 0  1 ( z 2  h 2 ) 
dt
 2 ( x 2  y 2 )   3 B   3 A)h  (  0  1 ( z 2  h 2 )   2 ( x 2  y 2 )) E  (21c)
( 0  1 ( z 2  h 2 )  2 ( x 2  y 2 )) F

27
dh
 ( 0  1 ( z 2  h 2 )   2 ( x 2  y 2 )   3 A   3 B)h  ( 0  1 ( z 2  h 2 ) 
dt
 2 ( x 2  y 2 )   3 B   3 A) z  (  0  1 ( z 2  h 2 )   2 ( x 2  y 2 )) F  (21d)
( 0  1 ( z 2  h 2 )  2 ( x 2  y 2 )) E ,
A  ( z 2  h 2 )( x 4  y 4  6 x 2 y 2 )  8 xyzh( y 2  x 2 )
B  2 zh( x 4  y 4  6 x 2 y 2 )  4 xy ( z 2  h 2 )( y 2  x 2 )
C  ( z 2  h 2 )( x3  3 xy 2 )  2 zh( y 3  3x 2 y )
D  2 zh( x3  3xy 2 )  ( z 2  h 2 )( y 3  3x 2 y )
E  z ( x 4  y 4  6 x 2 y 2 )  4 xyh( x 2  y 2 )
F  h( x 4  y 4  6 x 2 y 2 )  4 xyz ( x 2  y 2 )
The Jacobian matrix around the periodic orbit is obtained as

  dx  dx  dx  dx 
 x ( dt ) ( )
y dt
( )
z dt
( )
h dt 
 
  dy  dy  dy  dy 
 x ( dt ) ( )
y dt
( )
z dt
( )
h dt 
J   (22)
  dz  dz  dz  dz 
 x ( dt ) ( )
y dt
( )
z dt
( )
h dt 
 
  dh  dh  dh  dh 
 x ( dt ) ( )
y dt
( )
z dt
( )
h dt  X  X FP


To investigate the existence and stability of the periodic motions, a stability analysis was

performed. Due to the nonlinear nature of the amplitude equations there is not an analytical

solution for the fixed point solution. Numerical integration of the equations was therefore

done to identify the fixed points. We are interested in finding the parameter values in the

 0   0 space where torus doubling occurs. The stability of the periodic solution branch can

be analyzed by calculating the Jacobian matrix with periodic coefficients. The characteristic

polynomial of the Jacobian has the form

 4  c3 3  c2 2  c1  c0  0 (23)

28
wherre the coeffficients ci aare functionns of the vaariables. In Floquet thheory, the pperiodic

solutiion is stablee if all rootss have a noorm less thann 1. The soolution becoomes unstabble if an

eigennvalue crossses the unitt circle. Thee monodrom

my matrix iis obtained by integratting the

Jacobbian matrix over one peeriod starting from the ffixed point values.
v The eigenvalues of the

monoodromy mattrix are the F

Floquet multipliers. Figgure 17 shows the Argaand diagram
m of the

two periodic
p soluution branchhes undergooing bifurcattion by variiation of 0 and  0 .

Fiig. 17. Movem o modes S aand K by variiation of bifurccation parameeters.

ment of Floqueet multipliers of

Thhe diagram shows thee movementt of the Flooquet multiipliers whille the bifurrcations

occurr. The analyysis indicatees that the stable mixeed travellingg wave (MT
TW) branchhes lose

stabillity via a torus

t bifurccation at  0  0.0836 and  0  0.29 . At thhese values of the

bifurccation param
meters the Floquet muultipliers m
move in the complex plane to the values

29
given in Table 2.

Table 2. Floquet multiplier values at different bifurcation parameters.

Bifurcation
p parameters 0.0926 0.0836 0.0836 0.0826
0.25 0.29 0.36 0.46

Floquet multipliers
mode S -0.1643+ 0.2434i -0.4310 + 0.9109i -0.5414 -1.0258
mode K -0.1650 +0.2410i -0.4605+ 0.8969i -0.4709 -1.0267

Next, as the bifurcation parameters vary and reach  0  0.0826 and  0  0.46 the

Floquet multipliers collide with negative real axis and cross the unit circle at -1, triggering a

torus doubling bifurcation.

5.1 Poincare map

The Poincare maps of the amplitude equations for different forcing amplitudes

characterize the stability of the two periodic limit cycle responses. This map is constructed

by sampling the time history of the primary modes at their natural shedding frequency. For

the fixed cylinder case, the periodic response has almost a single Poincare point related to

the steady state case (see Fig. 18(a) and 18(b)). As the cylinder oscillation amplitude

increases at  0  0.0836 and  0  0.29 , the system undergoes torus bifurcation. The quasi-

periodic state of the POD modes of flow observed in Figs. 4(b,d) from numerical results are

also depicted in Fig. 18(c) and 18(d). It is obvious from these figures that by varying the

bifurcation parameters the competition between modes leads to the presence of smaller

peaks representing quasi-steady state. On increasing the oscillation amplitude a torus

doubling bifurcation occurs at  0  0.0826 and  0  0.46 . At these values, the sampling

points in Poincare space are organized along two invariant curves associated with the quasi-

periodic response of period-doubled torus (Figs. 18(e,f)).

30
 (a) (b)

(c) (d)

(e) (f)
map of the moddes S and K at (a,b)  0  0.0926 ,  0  0.25 (c,d)  0  0.0836 ,  0  0.29
Fig. 118. Poincare m
and (e,f)  0  0.0826 ,  0  0.46 .

6 C
Conclusion
n

Thhe Equivariaant bifurcatiion theory is applied to derive the aamplitude eequations moodeling

the ccylinder waake mode interaction. The propposed model predicts the sequeence of

31
bifurcations observed in the lift coefficient (hence the cylinder wake flow) dynamics

obtained from CFD results. This analysis shows that as the amplitude of sinusoidal

oscillation increases, the limit cycle undergoes a symmetry-breaking bifurcation leading to

a quasi-periodic state. Further increase in the forcing amplitude leads to a period-doubled

torus. The low order amplitude equations deduced from modal decomposition of the v-

velocity field qualitatively explain the nonlinear interaction between symmetric and

antisymmetric modes as the oscillation amplitude is increased. The cylinder streamwise

oscillation energizes the second mode of the transverse velocity, which interacts with the

first mode leading to the torus doubling bifurcation. However, the modulated travelling

waves bifurcating from the model have mode S again as the dominant mode, which

verifies the symmetric v-velocity pattern with period-doubling observed in the numerical

results. Thus, the solution branches deduced from the presented model are in good

agreement with the wake dynamics obtained from numerical simulations. The stability of

the solution branches deduced from the low order model were also investigated and

represented in a Poincare map. The symmetry–based low order model is capable of

predicting the complex wake flow without the need to solve the complex Navier-Stokes

equations. Therefore, this work contributes to the development of symmetry-based low

order models, which could be a basis for the development of low order optimal controllers

for fluid-structure interaction.

Acknowledgements

The funding support of NSERC (National Science and Engineering Research Council of

Canada) is gratefully acknowledged.

32
Appendix

Invariant and equivariants derivation

s k m
   2; n  1, m  2
k s n

x x
i( s t ) i( k t )
s k
v( x, y, t )  S (t )e f ( y)  K (t )e g ( y)  c.c.,

T : x  x  ls
 : t  t  s 2

T ( S , K )   Se  il
, Ke
i
s
k
l
   Se  il
, Ke
m
i l
n   T (S , K )  (Se inl
, Keiml )
 x    x    x   x  s
i   s ( t  s )  i  k ( t  s )  i   s t  i  k t  ik . 
 s 2   k 2   s i  k
 (v)  S (t )e f ( y )  K (t )e g ( y )  S (t )e 
.e f ( y )  K (t )e 
.e 2
g ( y)


 ( S , K )  Se i , Ke

i k 
s    Se  i
, Ke
m
i 
n    (S , K )   Se in
, Keim 

Invariants

K    Keim   Sein     K n eimn .S e  imn    K n S 

2
n m 2  n m
 m 2 m 2
S
S  S S  Sein .Sein  Sein .Se  in  S S
2

Equivariants

( S , K 2 n S 2 m 1 ), ( K , S 2 m K 2 n 1 )
2 m 1
K 2 n   Sein  .  Keim 
2 m 1 2n 2 m 1 2 m 1
S S .e 2imn .ein .K 2 n .e 2imn  S K 2 n .ein
2 n 1
S 2 m   Keim  .  Sein 
2 n 1 2m 2 n 1 2 n 1
K K .e 2imn .eim .S 2 m .e 2imn  K S 2 m .eim

References

[1] Lewek T. and Williamson C.H.K., 1998. Three-dimensional instabilities in wake

transition. European Journal of Mechanics B-Fluids 17, 571-586.

[2] Ryan K., Thompson M.C. and Hourigan K., 2005. Three-dimensional transition in the

33
 wake of bluff elongated cylinders. Journal of Fluid Mechanics, 538, 1-29.

[3] Williamson C.H.K., 1996. Vortex dynamics in the cylinder wake. Annual Review of

Fluid Mechanics, 28, 477-539.

[4] Barkley D. and Henderson R. D., 1996. Three-dimensional Floquet stability analysis of

the wake of circular cylinder. Journal of Fluid Mechanics, 322, 215.

[5] Ding W. C., Xie J. H. and Sun Q. G., 2003. Interaction of Hopf and period doubling

bifurcations of a vibro-impact system. Journal of Sound and Vibration, 275, 27-45.

[6] Park, D. S., Ladd D. M. and Hendricks E. W., 1994. Feedback control of von-Karman

vortex shedding behind a circular cylinder at low Reynolds numbers. Physics of Fluids,

6, 2390.

[7] Wooden S. M. and Sinha S. C., 2007. Analysis of Periodic-quasiperiodic nonlinear

systems via Lyapunov-Floquet transformation and normal forms. Nonlinear Dynamics,

47, 263-273.

[8] Provansal M., Mathis C. and Boyer L., 1987. Benard-von Karman instability: transient

and forced regimes. Journal of Fluid Mechanics, 182, 1-22.

[9] Le Gal, P., Nadim A. and Thompson M., 2001. Hysteresis in the forced Stuart-Landau

equation: application to vortex shedding from an oscillating cylinder. Journal of Fluids

and Structures, 15, 445-457.

[10] Williamson, C.H.K. and Roshko, A., 1988. Vortex formation in the wake of an

oscillating cylinder. Journal of Fluids and Structures 2, 355–381.

[11] Lam K.M., Liu P. and Hu J.C., 2010. Combined action of transverse oscillations and

uniform cross-flow on vortex formation and pattern of a circular cylinder. Journal of

Fluids and Structures, 26, 703-721.

34
[12] Lentini J. S., Thompson M. C. and Hourigan K., 2006. Three-dimensional transition in

the wake of a transversely oscillating cylinder. Journal of Fluid Mechanics, 577, 79-

104.

[13] Barkley D., Tuckerman L. and Golubitsky M., 2000. Bifurcation theory for three-

dimensional flow in the wake of a circular cylinder. Physical Review E., 61(5), 5247.

[14] Sheard G. J., Thompson M. C. and Hourigan K., 2003. A coupled landau model

describing the Strouhal-Reynolds number profile of a three-dimensional circular

cylinder wake. Physics of Fluids, 15(9), 68-71.

[15] Blackburn H. M., Marques F. and Lopez J. M., 2004. On three-dimensional

instabilities of two-dimensional flows with a Z 2 spatio-temporal symmetry. Journal of

Fluid Mechanics.

[16] Blackburn H. M., Marques F. and Lopez J. M., 2004. Symmetry breaking of two-

dimensional time-periodic wakes. Journal of Fluid Mechanics, 522, 395-411.

[17] Mureithi N. W., 2003. Implications of symmetry in inline and Karman shedding mode

interaction. NAGARE Journal of the Japanese Society of Fluid Mechanics, 22, 211-

227.

[18] Mureithi N. W. and Rodriguez M., 2005. Stability and bifurcation analysis of a forced

cylinder wake. Proceedings of the ASME Fluids Engineering Division 261, 59-66.

[19] Borggaard J., Hay A. and Pelletier D., 2007. Interval-based reduced-order models for

unsteady fluid flow. International Journal of Numerical Analysis and Modeling 4, 353-

367.

[20] Cao Q., Xu L., Djidjeli K., Price W. G. and Twizell E. H., 2001. Analysis of period-

35
 doubling and chaos of a non-symmetric oscillator with piecewise-linearity. Chaos,

Solitons and Fractals, 12, 1917-1927.

[21] Konstantinidis E., Balabani S. and Yianneskis M., 2007. Bimodal vortex shedding in a

perturbed cylinder wake. Physics of Fluids, 19, 011701.

[22] Noack B. R., Afanasiev K., Morzynski M., Tadmor G. and Thiele F., 2003. A

hierarchy of low-dimensional models for the transient and post-transient cylinder

wake. Journal of Fluid Mechanics, 497, 335-363.

[23] Rowley C.W., 2005. Model reduction for fluids, using balanced proper orthogonal

decomposition. International Journal of bifurcation and chaos,15 (3), 997-1013.

[24] Mureithi N. W., Huynh K., Rodriguez M. and Pham A., 2010. A simple low order

model of the forced Karman wake. International Journal of Mechanical Sciences, 52,

1522-1534.

36
Highlights
 A symmetry-based low order model is developed to study the nonlinear dynamics
of the forced cylinder wake.
 Two primary modes deduced from POD analysis in the presence of normal form
symmetry are employed to develop the model.
 The equivariant bifurcation theory is employed to classify the mode interaction
solution branches with respect to lower order symmetries.
 The bifurcation analysis of the low order model leads to a quasi-periodic state for
A/D=0.175 and torus doubling bifurcation for A/D=0.5 which matches well with
CFD results.
 The symmetry-based low order model predicts the nonlinear wake bifurcation
behavior without the need to solve the complex Navier-Stokes equations.

37