You are on page 1of 4

VOLUME 93, N UMBER 8

PHYSICA L R EVIEW LET T ERS

week ending
20 AUGUST 2004

Strouhal-Reynolds Number Relationship for Vortex Streets


Fernando L. Ponta1,* and Hassan Aref 2
1

Department of Theoretical and Applied Mechanics, University of Illinois, Urbana, Illinois 61801, USA
2
Virginia Polytechnic Institute & State University, Blacksburg, Virginia 24061, USA
(Received 3 November 2003; published 18 August 2004)
A rationale for the empirically observed Strouhal-Reynolds number relation for vortex shedding in
the wake of a cylinder is provided. This rationale derives from a mechanism of vortex formation
observed in numerical simulations of two-dimensional vortex shedding coupled with an order of
magnitude estimate of the terms in the vorticity transport equation based on this mechanism.
DOI: 10.1103/PhysRevLett.93.084501

The flow around a circular cylinder constitutes a physical phenomenon of general interest with numerous applications. In particular, understanding the relation between
the frequency of the vortex street wake and the Reynolds
number has been a challenge for almost a century. Much
progress has been made recently in the understanding of
the nature of the transition occurring as a function of
Reynolds number. However, a quantitative theory is still
lacking. In this Letter, we use numerical simulations to
identify the main physical ingredients governing periodic vortex shedding, thus rationalizing the wellestablished empirical experimental fit to the data.
It is easy to see from the governing equations and
simple dimensional analysis that the nondimensional
frequency of vortex shedding, known as the Strouhal
number, St, must be a function of the dimensionless group
known as the Reynolds number, Re. Reynolds number for
flow about a circular cylinder is generally defined as Re 
Ud=, where U is the speed of the free stream, d is the
diameter of the cylinder, and  is the kinematic viscosity
of the fluid (here assumed incompressible). The Strouhal
number, St  fd=U, is given in terms of the frequency
of vortex shedding, f, and two of the aforementioned
quantities. Extensive St measurements were made by
Roshko in 1954 [1], who found a stable (laminar)
flow regime for 40 < Re < 150, followed by a region of
transition (the unstable regime) for 150 < Re < 300,
and then an irregular regime for Re > 300 where irregularities are present in the wake velocity fluctuations.
Roshko suggested two empirical relations to describe
his data, St  0:2124:5=Re, for Re < 180, and St 
0:2122:7=Re, for Re > 300. Following Roshkos work,
many curves of the St-Re relation were published, often
showing little agreement between them, and a controversy
started about the nature and place of the several jump
discontinuities in the data that were observed. This longrunning debate was largely resolved by Williamson [2]
who found that the discontinuity in frequency is caused
by a change in the mode of shedding, viz., vortices would
come off at an angle to the cylinder rather than parallel to
it, a mode of shedding now referred to as oblique shed084501-1

0031-9007=04=93(8)=084501(4)$22.50

PACS numbers: 47.15.Ki, 47.27.Vf

ding. Manipulating the end boundary conditions to enforce parallel shedding, the resulting St-Re curve can be
made continuous throughout the laminar range (49 <
Re < 178). Williamson also demonstrated experimentally that the parallel-shedding curve is universal in the
sense that any oblique-shedding data St can be collapsed
onto a universal mode Stu by the transformation Stu 
St = cos , where is the angle between the shed vortex
filaments and the cylinder. It is now believed that the
universal parallel-shedding curve represents measurements for purely two-dimensional vortex shedding [3].
The universal curve is given closely by the fit Stu 
0:18163:3265=Re  0:00016Re. This expression is close
to the original 1954 fit by Roshko for the laminar regime,
with a maximum deviation of 2:5%. Williamson gave an
alternative expression for the universal curve Stu 
0:21755:1064=Re which also fits the experimental data
closely. In [4] he reported the existence of two discontinuities in the St-Re curve. The first occurs at Re 
170180, and the second at Re  230260. Figure 1 summarizes the experimental data. The present Letter is concerned with the range 40 < Re < 1400, which
encompasses the regimes referred to as L3 (periodic
laminar), TrW1/TrW2 (lower/upper transition-in-wake),
and TrSL1 (lower transition-in-shear-layers) [5].
Besides the large amount of experimental data there is a
more recent body of numerical simulation work. There is
also theoretical work based on the interpretation of the
shedding regime at moderate Re as a nonlinear global
structure with its frequency obtained in the framework of
instability theory (see [6] for a comprehensive treatment
of the subject). However, there does not appear to be a
physical explanation for the observed St-Re dependence.
As we argue below, the empirical fit is quite natural and
follows readily from an elucidation of the vortex formation mechanism and an order of magnitude estimate of
the terms in the vorticity transport equation.
It is natural to assume that the shedding period is
somehow related to the time needed to nucleate a
vortex, which will subsequently take its place in the
vortex street wake. It has been observed by visualization
2004 The American Physical Society

084501-1

PHYSICA L R EVIEW LET T ERS

VOLUME 93, N UMBER 8


0.22

St
0.21

(II)

0.2

0.19

(III)
0.18

0.17

(I)

0.16

L3
TrW1
TrW2
TrS L1

0.15

0.14

0.13

Re
0.12
0

200

400

600

800

1000

1200

1400

FIG. 1. St versus Re. Experimental data for the L3 [3], TrSL1


[1], and TrW1/TrW2 [4] regimes. (I) indicates the universal
parallel-shedding curve St  0:21755:1064=Re [3]. (II) gives
the irregular regime curve St  0:2122:7=Re [1]. (III) is a
fit for the regime TrW1, St  0:21756:66=Re, discussed towards the end of the Letter.

techniques [7] that the near-wake instability initiates a


wavy trail already at Re  30. The wavelength of this
trail decreases with increasing Re while the amplitude of
the crests and troughs increases, and the free shear layers
begin to roll-up and form eddies [5]. Laminar eddies are
not shed directly from the cylinder but are formed downstream by the laminar wake instability. This is accomplished by a gradual roll-up of free shear-layers
emanating from the cylinder [8]. It is clear that the rollup plays an essential role in vortex formation.
One must, nevertheless, be cautious and realize that
flow visualizations of unsteady roll-up can be misleading.
One reason is that in unsteady flows streaklines (which
are typically what is observed in a flow visualization)
differ from streamlines. Streaklines represent the integrated development composed of all previous distortions
incurred along the way from the point of introduction
upstream of the point of observation [5]. Indeed, Hama
[9] demonstrated that in a shear flow the roll-up of smoke
or dye might not signal a concentration of vorticity at all
unless accompanied by the rotation of the vortex structure itself. Another reason to be cautious in interpreting
flow visualization results is the difference in diffusivity
of vorticity and smoke or dye [10]. This difference implies
a faster decay of the eddies than the smoke patterns which
remain coherent even after most of the vorticity has been
dissipated.
Thus, a key in our approach was to first identify how
the vorticity actually evolves and, second, to determine
which eddy structures the shear layers are rolling up
around. The first issue can be addressed in a relatively
simple way by calculating the vorticity field from direct
numerical simulation data. We performed several compu084501-2

week ending
20 AUGUST 2004

tations in the range 40 < Re < 180 using a numerical


approach that we call the kinematic Laplacian equation
(KLE) method and that we have validated against the
experimental St-Re curve and other experimental data
finding very good agreement [11]. The second issue,
identifying the eddy structures, is more challenging.
The topology of the velocity field strongly depends on
the choice of frame of reference. If the observer moves
with the cylinder, incipient eddy structures can be observed in the vicinity of the solid surface while the far
wake shows wavy streamlines and no eddies. Conversely,
if the observer is fixed in the laboratory, the typical
pattern of streamlines associated with a vortex street
appears in the far wake, but the pattern of streamlines
in the vicinity of the cylinder appear distorted (a good
example can be seen in [13], plate 11). Thus, the choice of
frame of reference influences the observations and any
conclusions regarding mechanism. Moreover, for an observer moving with the cylinder, it appears that the eddies
are advected downstream from the low speed zone in the
vicinity of the cylinder until they reach a steady advection regime in the far wake. Thus, to describe the streamline pattern properly, we need to find a frame of reference
that follows each eddy as it accelerates in its travel
downstream.
This task of advection, accelerating the eddies to their
final state of motion in the vortex street wake itself, is
accomplished by the irrotational, solenoidal part, v, of
the full velocity field u. We can always decompose the
incompressible velocity field (in a frame of reference
moving with the cylinder) as follows: u  uv  v, where
uv is solenoidal and has the same curl as u, and v is the
irrotational and solenoidal (i.e., harmonic) component.
For prescribed velocity conditions on the boundary of
the analysis domain, v is uniquely determined (see sec.
2.7 of [13]). This is precisely the case in our numerical
simulations if we take care to complete the computation
before the first traces of vorticity come close to any
external boundary of our simulated domain, so we have
a uniform stream on the external boundaries, and we have
zero velocity on the solid surface (since we are in a frame
of reference moving with the cylinder). Now v (which is
easy to compute) and then uv  u  v are both uniquely
determined. Figure 2 shows the instantaneous velocity
arrows of u (top) and uv (bottom) in the far wake superimposed on the vorticity isolines. The uv velocity arrows
are seen to be tangent to the vorticity isolines, which
shows that the vorticity distribution is stationary with
respect to this velocity field or, in other words, that v is
responsible for the advection of the vortex structures in
this regime. In the far wake the vortices are simply
subject to steady viscous decay (see sec. 7.4 of [13]) as
they are advected by v. This argument validates v as the
advecting velocity in the far wake. We posit that v is also
the dominant advecting velocity in the near wake.
Because of the far-field boundary conditions v must scale
with the free stream speed U.
084501-2

VOLUME 93, N UMBER 8

week ending
20 AUGUST 2004

PHYSICA L R EVIEW LET T ERS

port equation (for two-dimensional flow),


@!
 u  r!  r2 !:
@t

FIG. 2. Vorticity isolines in the far wake and pattern of


streamlines of u (upper panel) and of uv (lower panel) for Re 
100. The cylinder is centered at the origin. Coordinates are
given in cylinder diameters.

Figure 3 shows a sequence of plots of uv streamlines


and vorticity isolines during the process of vortex formation in the near-wake. Start at the top-left panel and
follow the four panels clockwise. We see the shear layer of
positive vorticity start to roll-up around the incipient
eddy on the lower right of the cylinder until the core of
this eddy has produced a (roughly) homogeneous distribution of vorticity all around its periphery. Even though
the outside of the eddy continues its evolution, exchanging vorticity with its surroundings, its core is already
formed. The amount of time required to form this homogeneous core of vorticity is, in essence, half a period of
the shedding cycle. In Fig. 4 we show a schematized
version of the process. We shall use this schema to estimate the various terms appearing in the vorticity trans-

(1)

Referring to Fig. 4 the homogenization is carried on by


the transport of vorticity along the periphery of the core
from the high-vorticity zone at the head of the rolling
shear layer (zone A) to the low-vorticity zone at the
opposite side of the core ring (zone B). Two mechanisms
act simultaneously: advection, which takes place mainly
around the core, and diffusion, which acts mainly in the
radial direction outward from the core. These two mechanisms have opposing effects on the homogenization.
While advection is trying to build up the core, diffusion tends to spread out the vorticity before it can
arrive at zone B. In terms of Eq. (1) we decompose the
velocity into v and uv as before. The local derivative and
the advective derivative due to v provides a material
rate of change of vorticity at zone B. A suitable estimate
for this quantity is !H  !L f1 , where f is the shedding frequency. We move the remaining advective derivative due to uv to the right-hand side and estimate it by
U!H  !L =d, where d is a typical length scale, in this
case the diameter of the cylinder, and U is the free stream
speed. This term gives the rate of intrinsic rearrangement
(homogenization) within the core. We see that uv must
generally point opposite to the gradient of vorticity so the
term acts as a source term for vorticity buildup. Finally,
the diffusive sink of vorticity produces a term that may be
estimated as !H  !L =d2 . Collecting these estimates in an equation by introducing two dimensionless
constants, ka for the advective process and kd for the
diffusive process, we obtain
!H  !L f  ka U

!H  !L 
!  ! 
 kd  H 2 L : (2)
d
d

Simple algebra then gives



fd
 ka  kd
;
Ud
U

FIG. 3. Vorticity isolines and uv streamlines in the near wake


for Re  170.

084501-3

FIG. 4.

(3)

Schematic view of the roll-up of a vortex-core.

084501-3

VOLUME 93, N UMBER 8

PHYSICA L R EVIEW LET T ERS

FIG. 5. Generation of vortex loops and streamwise vortices


after the first discontinuity in the St-Re curve (Re  180). The
camera moves with the vertical cylinder (edge is at the extreme
right of each picture) [4].

or, in terms of the Strouhal and Reynolds numbers,


S t  ka  k =Re:

(4)

It is significant that (4) matches not only the expression


for the universal curve in the laminar regime L3 given by
Williamson in 1989 [3] (curve I in Fig. 1), but also the
empirical formulas proposed by Roshko [1] for both the
L3 and TrSL1 regimes (curve II in Fig. 1). In the TrSL1
regime (300400 < Re < 1k  2k) the eddies are turbulent, and this change in the flow is likely to change the
values of both scaling constants ka and kd from those
obtained for the laminar regime, but the scaling itself
appears to remain valid (the agreement of Roshkos
curve II in the interval 300 < Re < 2000 is better than
1% [1]).
For the TrW1 regime (178 < Re < 260) Williamson
observed the formation of vortex loops. He conducted
experiments using dye washed off the surface of the
cylinder to visualize these structures. Figure 5 shows a
sequence of images from [4]. It was found that the primary vortices roll-up first and then deform, during the
process of shedding, to create the vortex loops. This
process is self-sustaining in that there is a feedback
from one loop to the next, so that a whole string of loops
form at the same spanwise location. We see in the figure
how the loops become highly stretched so that the two
sides of each loop evolve into a pair of contrarotating,
streamwise vortices. It was found that the characteristic
spanwise wavelength of these vortex loops is about three
cylinder diameters. The eddies remain laminar. For a
cylinder that is hundreds (and in some cases thousands)
of diameters in span we have many of these loops distributed more or less uniformly along the span. This allows us
to interpret their effect in a span-averaged way. Thus,
even though this is clearly a three-dimensional phenomenon, we may try to extrapolate our two-dimensional
model to cover TrW1 as well. In terms of our model the
presence of the loops acts as an additional sink of vorticity from the rolling-up layer. Since the loops stretch the
surface of diffusion, they act to increase the effective

084501-4

week ending
20 AUGUST 2004

value of kd in our model. Accordingly, we fitted (4) to


Williamsons data for TrW1 retaining the same value of
ka  0:2175 of curve I. As expected, we obtained a value
of kd  6:66 which is larger than the kd  5:1064 of
curve I. We have included this result in Fig. 1 as
curve III. Williamsons data for TrW2 (Re  230 Re 
290) seem to lie roughly on an extension of curve II
(particularly after the end of the second discontinuity at
Re > 260). Nevertheless, the flow in TrW2 shows complex
features, such as finer-scale, streamwise vortex structures
[4] and eddies with a laminar core but a turbulent periphery [14]. There are currently two different interpretations
of the shedding modes during the nonhysteretic TrW1
TrW2 transition [14].
In summary, we have provided a rationale for the
empirical expression for the St-Re relation from an order
of magnitude estimate of the vorticity transport equation,
assuming a specific mechanism of vortex formation. The
relation compares favorably with experimental results in
the L3, TrSL1, and TrW1 regimes (and the upper part of
TrW2). The TrW1TrW2 transition (230 < Re < 260) requires additional considerations and is not covered by the
arguments presented here.

*Also at College of Engineering, University of Buenos


Aires.
[1] A. Roshko, TR 1191, NACA, 1954.
[2] C. H. K. Williamson, Phys. Fluids 31, 2742 (1988).
[3] C. H. K. Williamson, J. Fluid Mech. 206, 579 (1989).
[4] C. H. K. Williamson, Phys. Fluids 31, 3165 (1988).
[5] M. M. Zdravkovich, Flow around Circular Cylinders
(Oxford University Press, Oxford, United Kingdom,
1997), Vol. 1.
[6] B. Pier, J. Fluid Mech. 458, 407 (2002).
[7] S. Taneda, J. Phys. Soc. Jpn. 11, 302 (1956).
[8] L. S. G. Kovasznay, Proc. R. Soc. London A 198, 174
(1949).
[9] F. R. Hama, Phys. Fluids 5, 644 (1962).
[10] J. M. Cimbala, H. M. Nagib, and A. Roshko, J. Fluid
Mech. 190, 265 (1988).
[11] The KLE belongs to a family of so-called vorticityvelocity methods (see [12]). It is based on a space-time
splitting of the problem that solves the time evolution of
the vorticity as an ordinary differential equation on each
node of the spatial discretization, using at each time step
the spatial solution for the velocity field provided by a
Poisson equation.
[12] L. Quartapelle, Numerical Solution of the Incompressible
Navier-Stokes Equations (Birka user, Basel, Switzerland,
1993).
[13] G. K. Batchelor, An Introduction to Fluid Dynamics
(Cambridge University Press, Cambridge, United
Kingdom, 2000).
[14] M. M. Zdravkovich, in IUTAM Symposium, Bluff-Body
Wakes, Gottingen (Springer-Verlag, Berlin, 1992),
pp. 271274.

084501-4

You might also like