ANRV294-FL39-18 ARI 12 December 2006 6:6

Turbulence Transition in
Pipe Flow
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org

Bruno Eckhardt,1 Tobias M. Schneider,1

Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

Bjorn Hof,2 and Jerry Westerweel3

1
Fachbereich Physik, Philipps-Universitat Marburg, D-35032 Marburg, Germany;
email: [email protected]
2
Department of Physics and Astronomy, University of Manchester, Manchester,
M13 9PL, United Kingdom
3
Laboratory for Aero and Hydrodynamics, Delft University of Technology,
2628 CA Delft, The Netherlands

Annu. Rev. Fluid Mech. 2007. 39:44768 Key Words

The Annual Review of Fluid Mechanics is online shear ows, coherent structures, nonlinear dynamics, chaotic
at uid.annualreviews.org
saddle
This articles doi:
10.1146/annurev.uid.39.050905.110308 Abstract
Copyright  c 2007 by Annual Reviews. Pipe ow is a prominent example among the shear ows that un-
All rights reserved
dergo transition to turbulence without mediation by a linear insta-
0066-4189/07/0115-0447$20.00 bility of the laminar prole. Experiments on pipe ow, as well as
plane Couette and plane Poiseuille ow, show that triggering tur-
bulence depends sensitively on initial conditions, that between the
laminar and the turbulent states there exists no intermediate state
with simple spatial or temporal characteristics, and that turbulence
is not persistent, i.e., it can decay again, if the observation time is
long enough. All these features can consistently be explained on the
assumption that the turbulent state corresponds to a chaotic saddle
in state space. The goal of this review is to explain this concept, sum-
marize the numerical and experimental evidence for pipe ow, and
outline the consequences for related ows.

447
 ANRV294-FL39-18 ARI 12 December 2006 6:6

1. INTRODUCTION
Transition to turbulence in pipe ow has puzzled scientists since the studies of Gotthilf
Heinrich Ludwig Hagen (Hagen 1839, 1854), Jean Louis Marie Poiseuille (Poiseuille
1840), and, most prominently, Osborne Reynolds in 1883 (Reynolds 1883). Under
favorable conditions, when the water in the supply tank had settled and the inow was
controlled with suitable funnels, Reynolds was able to maintain laminar ow until the
mean ow speed was equivalent to Re = 12000, when expressed in the dimensionless
combination of mean ow speed u, pipe diameter d , and viscosity v that now carries
Reynoldss name: Re = ud /. On the other hand, with sufciently strong perturbations
he was able to trigger a transition near Reynolds numbers of about 2000. A more
precise value above which transition to turbulence can be triggered is difcult to
identify, with quoted values ranging between 1760 and 2300 (Kerswell 2005).
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

Pipe ow differs from many other ow situations in that the laminar prole is
linearly stable for all Reynolds numbers: All sufciently small perturbations will decay
[see, e.g., Salwen et al. (1980) and, in particular, Meseguer & Trefethen (2003), who
analyzed the problem up to Re = 107 ]. Thus, to trigger transition, two thresholds have
to be crossed: The ow has to be sufciently fast and a perturbation has to be strong
enough. Observing a section of the pipe xed in the lab frame gives the familiar
intermittent switching between laminar and turbulent regions: A sufciently large
perturbation triggers turbulence, which is then swept past the observation region, and
the ow becomes laminar until another sufciently strong perturbation again induces
turbulence. This behavior was demonstrated using Reynoldss original experiment
by Homsy et al. (2004). Movies of the experiment and some flow visualizations
may be found via the Supplemental Material link from the Annual Reviews home
page at http://www.annualreviews.org. Further experiments by Hof et al. (2003)
show that as the Reynolds number increases the critical threshold decreases so that
at sufciently high Reynolds numbers the unavoidable residual uctuations always
sufce to trigger turbulent ow. Exactly how the threshold depends on the Reynolds
number is an intriguing question that is discussed in some detail in Section 2, with a
renement in Section 5.
A second feature of transition to turbulence in pipe ow is that between the laminar
and turbulent state there exists no state with simple spatial or temporal structures,
unlike the rolls in Rayleigh-Benard or the Taylor vortices in Taylor-Couette ows,
for example. Moreover, numerical simulations by Brosa (1989) and Faisst & Eckhardt
(2004), and also the experimental results of Darbyshire & Mullin (1995), Hof (2004),
Mullin & Peixinho (2006), and Peixinho & Mullin (2006), show that even if one
establishes a state with all features of turbulent dynamics, this state can still decay
without any clear precursors: Although it is relatively easy to conclude that the further
dynamics will be a relaxation toward the laminar prole, for instance, because the
energy in the radial component of velocity drops below a certain value, there is no
indicator for the imminent decay. This property of the ow is considered in Section 3.
The understanding of the properties of transition in pipe ow that has emerged
in the past few years rests on the application of the appropriate model in dynami-
cal system theory and systematically designed numerical and laboratory experiments.

448 Eckhardt et al.

 ANRV294-FL39-18 ARI 12 December 2006 6:6

The background for these studies is the abstraction to consider the system in its state
space (Lanford 1982). Physically, it is the space of all velocity elds, either prepared
as initial conditions or obtained in the time evolution of the ow. Mathematically,
it is spanned by all divergence-free ow elds that satisfy the appropriate boundary
conditions, represented, for instance, by the coefcients of an expansion of velocity
elds in a complete basis of orthonormal basis functions. The state space contains
the laminar prole and the turbulent ow elds. Coherent structures such as vor-
tices, streaks, hairpins (Panton 2001, Robinson 1991), or traveling waves (Hof, van
Doorne et al. 2004) occupy different parts of the state space. The state space should
provide a complete description of the dynamics, in that at any point in this space the
Navier-Stokes equations together with boundary conditions uniquely determine the
evolution. The time evolution of a ow then traces out a continuous trajectory in
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

this state space. We assume that ideas developed in the context of nite-dimensional
dynamical systems can be applied to this innite-dimensional situation (see Doering
& Gibbon 1995 for a discussion of the subtleties involved).
In state space, there is one region dominated by the laminar ow. The time-
independent parabolic prole is a xed point in this space. The parabolic prole
is linearly stable and, hence, all points in its neighborhood evolve toward the xed
point; these states form the basin of attraction of the laminar ow. The turbulent
dynamics take place in other parts of the state space. If turbulence was an attractor
(Guckenheimer 1986, Lanford 1982), then it, too, would have a basin of attraction
so that all initial conditions close to it would be attracted to the turbulent dynamics.
The spatially and temporally uctuating dynamics of the turbulent regions suggests
that there are chaotic elements, such as horseshoes, just as in a regular attractor
(Guckenheimer & Holmes 1983). However, the possibility of decay indicates that the
basin is not compact nor space lling; there must be connections to the laminar prole.
In dynamical systems such structures are known as chaotic saddles or strange saddles:
With chaotic attractors they share positive Lyapunov exponents for the motion close
to the saddle, but they are not persistent and have a constant probability of decay.
The idea of transient chaos is familiar from the motion of interacting point vortices
(Aref 1983, Aref et al. 1988, Eckhardt & Aref 1988). Several vortices carrying vorticity
of equal sign spin around each other, and if their number exceeds three the motion is
most likely chaotic. Pairs of equal but opposite vorticity can escape to innity along
straight lines. One can then set up a scattering experiment by aiming two vortex
pairs against each other. Upon collision they can exchange partners, and if the net
vorticity in each pair does not vanish, they move in circles until the next collision.
If the original partners do not regroup, the circular motion continues until the next
collision. Except for meticulously chosen initial conditions this motion ends and the
pairs separate again. The time at which this happens depends sensitively on initial
conditions and slight variations can lead to widely differing trapping times (Aref
et al. 1988, Eckhardt & Aref 1988). However, among all the chaotic trajectories there
do exist some with fairly regular dynamics: periodic solutions embedded in a sea
of chaos. They can be used to describe segments of trajectories and can be pieced
together as building blocks for more complicated motion. They have at least one
unstable direction and several stable ones, so that the motion in their vicinity is akin

www.annualreviews.org Turbulence Transition in Pipe Flow 449

 ANRV294-FL39-18 ARI 12 December 2006 6:6

to that near a saddle. In a chaotic saddle the stable and unstable directions tangle to
form the principle element of chaotic motion, a Smale horseshoe (Guckenheimer &
Holmes 1983).
Another analogy to help one visualize the meaning of a chaotic saddle is that of
a particle in a box with curved walls (Ott 1993). The particle dynamics is such that
the particle moves along straight lines until it hits a wall where it is elastically re-
ected. With the exception of a spherical, ellipsoidal, or rectangular shape, nearly
any boundary will produce chaotic particle dynamics. The fact that this model is
energy conserving whereas a hydrodynamic ow is dissipative should not be of con-
cern: If the dynamics is expanded to include friction on the particle and a motor that
keeps the particle in motion, one arrives at a dissipative analog with the same key
features. To obtain a chaotic saddle, introduce a hole into the wall through which the
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

particle can escape. Until the particle hits the hole it will bounce around chaotically,
and the dynamics will have a positive Lyapunov exponent . Because of the positive
Lyapunov exponent, correlations in trajectories will disappear quickly (on a timescale
of the order of 1/), and the probability of hitting the escape hole remains nearly
the same: Whenever the particle hits the wall it escapes with a probability equal to
the area of the hole divided by the total surface area.
There are three implications of a strange saddle that can be observed in pipe
and other shear ows: (a) the (transient) turbulent dynamics has a positive Lyapunov
exponent, (b) the distribution of lifetimes becomes exponential for long times, and
(c) the hyperbolic elements in the turbulent dynamics show up as transient patterns
in the turbulent ow. Of these, a Lyapunov exponent has only been determined in
numerical simulations (Faisst & Eckhardt 2004) as it requires a comparison of the time
evolution of two states starting from nearby initial conditionsa feat not yet achieved
in experimental studies. For the latter two implications, there are both experimental
and numerical results. Lifetime statistics were obtained by repeating experiments with
long observation times for different initial conditions, and certain coherent elements
that may serve as the invariant structures around which the chaos develops were
identied. The evidence for the lifetimes is discussed in Section 3, and the relation
between chaotic saddles and coherent structures is the subject of Section 4.
The state space picture with separate domains for the laminar and turbulent dy-
namics raises a question regarding the border between the two. The precise nature
of this border is complicated, especially in view of the transient nature of turbulence.
But it is clear that, depending on which side of the border a perturbation starts out,
it will either swing up to the turbulent region or decay to the laminar prole. This
can be exploited in order to nd the border and to trace the dynamics along it. The
precise nature of this border as well as observations regarding the dynamics in this
region are discussed in Section 5.
In the subsequent sections we summarize the experimental and numerical evi-
dence for this transition scenario and outline a few consequences. However, there is
one element of transition to turbulence in pipe ow that is not addressed: For the
intermediate Reynolds numbers considered here a localized perturbation will induce
turbulence in localized sections of the pipe only (see Wygnanski & Champagne 1973
and Wygnanski et al. 1975 for seminal observations and studies). These turbulent

450 Eckhardt et al.

 ANRV294-FL39-18 ARI 12 December 2006 6:6

puffs and slugs typically extend about 30 diameters along the axis and move down-
stream with little change in axial extent. It is desirable to explain this localization
of the turbulence as well, but this is not yet possible. We expect that this problem
falls into the class of patterned turbulence phenomena that includes the turbu-
lent patches in shear ows (Gad-el-Hak & Hussain 1986, Schumacher & Eckhardt
2001), or the striped turbulence in Taylor-Couette (Prigent et al. 2002) and plane
Couette ow (Barkley & Tuckerman 2005, Bottin & Chate 1998, Bottin et al. 1998).
As in the modeling attempt of Prigent et al. (2002), one may build on the assumption
that on top of the short-time, short length-scale turbulent interior dynamics, there
is a long wavelength modulation that is responsible for the structuring. The interior
and envelope dynamics may be linked at the front and trailing edges because of the
similar structures that can be detected there, but we do not yet know enough about
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

their relation. The separation in length scales (the typical structures to be discussed
below are only a few diameters long) and numerical evidence from turbulent spots,
which sometimes decay from within, and not by retreating boundaries (Schumacher &
Eckhardt 2001), suggest that one should be able to separate the dynamics of the tur-
bulent boundaries from the dynamics of the chaotic elements discussed below.
Various aspects of transition in shear ows in the absence of linear instability
were recently reviewed. Grossmann (2000) summarized the physics of non-normal
amplication and its consequences for threshold behavior. Kerswell (2005) surveyed
experimental and theoretical work culminating in the detection of the traveling waves
that we consider in Section 4. The proceedings of a 2004 conference in Bristol (Mullin
& Kerswell 2005) contain a useful collection of articles on several current approaches
to the problem. This review focuses on pipe ow, and describes the methods used
to analyze transition and the turbulent dynamics. We hope this will be helpful for
gaining insight in other shear ows for which pipe ow can serve as a model: In
several respects, transition to turbulence in these shear ows differs from the more
traditional ones in Rayleigh-Benard or Taylor-Couette and belongs to a class of its
own.
In Section 2 we review the experiments on the transition, followed by a study of
the lifetimes in Section 3. A survey of coherent structures is presented in Section 4,
and an analysis of the border between the laminar and turbulent regions is in Section
5. We conclude with a summary on pipe ow in Section 6 and an outlook to related
ows and open issues in Section 7.

2. TRANSITION EXPERIMENTS
Because the laminar prole is linearly stable for all Reynolds numbers, a nite stimulus
is needed to trigger the transition. In typical experiments this is achieved by injecting
or removing liquid from the pipe.
In a stimulating set of experiments Darbyshire & Mullin (1995) tried to iden-
tify the critical amplitude for perturbations that trigger transition. Their ndings
are revealing. Repeating the experiment with initial conditions that were identi-
cal within experimental resolution gave widely differing results: Sometimes transi-
tion was induced, and sometimes not. The observation of transition for one set of

www.annualreviews.org Turbulence Transition in Pipe Flow 451

 ANRV294-FL39-18 ARI 12 December 2006 6:6
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

Figure 1
Transition experiments by Darbyshire & Mullin (1995). Disturbances were introduced at a
distance 70 diameters downstream of the inlet, and their status was probed at another 120
diameters downstream, delayed with the mean advection time. Depending on whether the
perturbation was still present or not, a point was marked transition or decay. The
amplitude of the perturbations is proportional to the injected uid volume. For more details,
see Darbyshire & Mullin (1995). Redrawn after Darbyshire & Mullin (1995).

initial conditions gave no insight regarding the behavior of neighboring conditions:

Sometimes they remained turbulent, and sometimes they decayed. Figure 1, which
summarizes their ndings, does not show a clear separation between decaying and
turbulent initial conditions.
The sensitivity of transition to initial conditions is best studied in numerical sim-
ulations, where within the connes of the numerical representation and algorithms
one has perfect control over the initial conditions and can study the evolution of
slightly differing initial conditions (Faisst & Eckhardt 2004). Moreover, because of
the continuous monitoring of the dynamics one can determine the time when the
energy content in the perturbation drops below a level from which it cannot recover,
so that one has entered the basin of attraction of the laminar prole. This denes the
lifetime. Numerical experiments for pipe ow show a very rapid variation in lifetimes
depending on initial condition and Reynolds number (see Figure 2). The data of
Darbyshire & Mullin (1995) can be obtained from such a lifetime plot by slicing at a
prescribed time level T0 : Anything with lifetimes above T0 qualies as turbulent, and
anything below as decay. In Figure 2, the presence of valleys and pinnacles corre-
sponds to the isolated decaying initial conditions surrounded by turbulent ones and
vice versa in Figure 1.
The small-amplitude region of the graph allows one to quantify the increased
sensitivity of the ow to perturbations with increasing Reynolds number. At high
Re, lower perturbation amplitudes are needed to trigger turbulence. A linear anal-
ysis of perturbations around the laminar prole shows that certain perturbations

452 Eckhardt et al.

 ANRV294-FL39-18 ARI 12 December 2006 6:6

Figure 2
Numerical transition
experiments. A ow was
prepared with an initial
condition consisting of the
parabolic prole with center
speed uc and a perturbation
with xed spatial structure
but varying amplitude. The
ow was evolved until it
either decayed or exceeded
the maximal integration
time.
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

can give rise to ow elds whose amplitude transiently increases before eventually
succumbing to decay (Grossmann 2000; Schmid & Henningson 1999, 1994). The
origin of this mechanism lies in the non-normal nature of the linearized equations
of motion (Boberg & Brosa 1988, Trefethen et al. 1993) and becomes transparent
when the relevant ow structures are studied (Hamilton et al. 1995, Panton 2001): A
downstream vortex mixes uid across the shear direction and thereby sets up stream-
wise modulations of the streamwise velocity, thus forming the known boundary-layer
streaks. Simple estimates, conrmed by more detailed studies, suggest that a vortex
of strength O(1/Re) can generate a streak of order O(1), that is a Re-fold increase in
velocity amplitude (Chapman 2002, Henningson 1996, Waleffe 1995). Experiments
reported in Hof et al. (2003), Hof (2004), and Draad et al. (1998) give evidence for
a scaling of the critical amplitude like 1/Re in the Reynolds number range between
2000 and 20,000.

3. LIFETIME STATISTICS
The strong sensitivity of the lifetimes on initial conditions suggests a limited meaning
to individual trajectories for transition studies. Statistical properties like the distribu-
tion of lifetimes are a more reliable means for transition studies. The prediction of
dynamical systems theory for the lifetimes of a chaotic saddle is that the probability
of decay is independent of the time that has elapsed since the turbulent state was
created, and that therefore the distribution of lifetimes is an exponential (Kadanoff &
Tang 1984).
In the experiments by Darbyshire & Mullin (1995), the state of the system was only
analyzed at the end of a xed-length pipe. By following the perturbation as it moves
with the mean ow, one can determine the point where it decays. These observations

www.annualreviews.org Turbulence Transition in Pipe Flow 453

 ANRV294-FL39-18 ARI 12 December 2006 6:6

a b
1.0
100 100

50
Probability P(t)

1000 /
10 0
1600 1800 2000 2200
2175
0.1 2150
2125
1
2050
2000
0
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

500 1000 1500 2000 2500 1600 1800 2000 2200

Time t (R / Uc) Re

Figure 3
Lifetimes of perturbations in pipe ow. (a) The left frame shows the probability P(t) to be
turbulent at least for a time t for different Reynolds numbers. (b) The right frame shows the
rapid increase of the characteristic time of the exponential t. The straight line in the
semilogarithmic plot indicates an exponential increase. The inset demonstrates that a linear
variation of the inverse of the characteristic time with Reynolds number does not represent the
data.

are applied in the determination of the probability P (t) that the ow remains turbulent
for at least a time t. If the turbulent state were permanently sustained, the lifetimes
would be innite, P (t) = 1. If the probability of decay in some time interval t is
constant and independent of the time that has elapsed from the start of the experiment,
an exponential distribution is obtained, P (t) exp(t/ ). It is characterized by a time
, equal to the time over which the probability drops by 1/e. For all data analyses
one has to keep in mind that the exponential form is a statement about the long-time
behavior, i.e., it is safest to obtain from the slope in a semilogarithmic representation.
Numerical experiments give the distribution shown in Figure 3a. For short times
there is a nonuniversal part that depends on the type and duration of the stimulus.
However, independent of the initial condition, the tail of the distribution for long
times becomes exponential. This hallmark of transient strange saddles has also been
found, experimentally (Bottin & Chate 1998) and numerically (Eckhardt et al. 2002),
in plane Couette ow.
Figure 3b shows the variation of the characteristic time with Reynolds number.
The characteristic time increases rapidly with Re, but there is no theoretical pre-
diction for the functional form of this variation. Low-dimensional systems provide
examples with algebraic (Kaneda 1990), exponential (Moehlis et al. 2004), and even
superexponential increases (Crutcheld & Kaneko 1988). Following a similar analysis
for plane Couette ow by Bottin & Chate (1998), Eckhardt & Faisst (2004) studied
the inverse of the characteristic time and found evidence for a divergence of the char-
acteristic time (Re) near Re = 2250. This would indicate a transition from a transient

454 Eckhardt et al.

 ANRV294-FL39-18 ARI 12 December 2006 6:6

chaotic saddle to a persistent chaotic attractor. In dynamical systems, the reversethe

destruction of a chaotic attractor by some form of boundary crisishas been studied
frequently (Grebogi et al. 1982). Experimental results by Mullin & Peixinho (2006)
and Peixinho & Mullin (2006) also show a divergent characteristic time, but at a
lower Reynolds number of about 1750. The most recent analyses of experiments in
a very long pipe with observation times up to 7500 d/u and of additional numerical
data suggest that the characteristic time does not diverge, but instead increases ex-
ponentially with Reynolds number (Hof et al. 2006). Such a behavior implies that at
any Reynolds number and in the neighborhood of every turbulent trajectory there
will be some trajectories that decay toward the laminar prole. However, the times
over which this happens quickly become inaccessibly large. Hof et al. (2006) estimate
that for a typical garden hose at Re = 2380 a pipe length of 40,000 km and an ob-
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

servation time of ve years are required to observe the decay. Nevertheless, the time
for relaminarization can be reduced by targeting the system onto the appropriate
trajectories. Clearly, this unexpected observation requires further experimental and
numerical analysis, in pipe ow and other shear ows. In particular, the inuences
of numerical resolution and domain size, or of external and internal perturbations in
experiments, need to be explored further. In all cases the main challenge is to obtain
good statistics for very long observation times where the theoretical prediction of an
exponential lifetime distribution is realized.
To establish the chaotic nature of the transient dynamics in relation to the models
mentioned in the Introduction, it is valuable to determine the short-time Lyapunov
exponent using, for instance, the method described in Eckhardt & Yao (1993). For
Reynolds numbers near Re = 2200, one nds Lyapunov exponents of about 0.07 uc /R,
based on laminar center line speed uc and radius R (Faisst & Eckhardt 2004). After
advection downstream by 10 radii, the difference between two initial conditions as
measured, for instance, by the maximum of the pointwise difference between the
velocity elds, increases by a factor of two. Setting up experiments that are close
to within 10% after traveling 100 diameters downstream requires one to control
initial conditions to within 104 ! This indicates the chaotic nature of pipe ow. The
positive Lyapunov exponent can also be used to rationalize the rapid variations of
lifetimes with ow parameters. Suppose that after a time t one state decays, but a
neighboring one, which is a distance d e away, does not. A variation in initial conditions
of order d e exp(t) can sufce to shift the ow that decays into the one that remains
turbulent for a much longer time. Whether a turbulent ow will continue to be
turbulent beyond this time or whether it will decay can only be predicted if the full
ow eld can be described with such accuracy! This explains why the decay of a
specic initial condition is unpredictable, and why there are signicant variations in
lifetimes between different initial perturbations or different Reynolds numbers.

4. CHAOTIC SADDLES AND COHERENT STATES

Embedded in all chaotic motions are simpler, more regular time evolutions. For
instance, for the vortex pairs mentioned in the Introduction one can nd uniformly
propagating states with pairs regularly circling around each other (Aref et al. 1988).

www.annualreviews.org Turbulence Transition in Pipe Flow 455

 ANRV294-FL39-18 ARI 12 December 2006 6:6

Similarly, for the chaotic container in the previous section there are often trajectories
that bounce back and forth along a diameter. Typically, neither of these motions are
stable, but they are signicant, as they can be used to establish chaos by proving the
presence of chaotic horseshoes, and they can dominate the visual appearance of the
dynamics.
For the shear ows we consider here, the rst example of a more regular solution
to the equations of motion embedded in the turbulent dynamics was found in plane
Couette ow by Nagata (1990), Clever & Busse (1992, 1997), and Waleffe (2003) (see
Cherhabili & Ehrenstein 1997 for a different class of solutions). They were called
tertiary structures to distinguish them from the primary structures that appear in
bifurcations from the linear prole and secondary ones that appear in subsequent
bifurcations of primary ones. Quartenary structures are in turn derived from linear
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

instabilities of tertiary structures. Plane Couette ow has an up-down symmetry in

the mean prole so that one can nd stationary states. If the up-down symmetry is
broken in the three-dimensional (3D) states, the stationary solutions turn into trav-
eling waves. For plane Poiseuille ow, these states appear as traveling waves from the
beginning (Ehrenstein & Koch 1991, Waleffe 2003). Remarkably, all these states are
dominated by large-scale features, prominent vortices, and streaks, and they qualify
as coherent structures.

4.1. Coherent States in Pipe Flow

Finding such coherent states in pipe ow is made difcult by the absence of a bifur-
cation that could be used as a starting point, and a Newton search from an arbitrary
initial condition will typically not converge. However, as in other cases, an embed-
ding in a family of ows can provide the desired starting point. For plane Couette
ow an embedding in a Rayleigh-Benard situation with differential heating across
the plates (Clever & Busse 1992, 1997; Nagata 1990) or a Taylor-Couette ow with
a narrow gap (Faisst & Eckhardt 2000) shows that some 3D stationary states can be
continued over to the original plane Couette ow. These states are dominated by
downstream vortices wiggling in the spanwise direction. They are intriguing because
they are similar to the ones that give the strongest non-normal amplication (Schmid
& Henningson 1999, 1994; Zikanov 1996).
For pipe ow there is no natural embedding in a larger family of ows with
instabilities. But by adding body forces that drive downstream vortices one can set
up an artical system with the desired properties. The detailed choice of body force
is not critical: Wedin & Kerswell (2004) rst solved the linear system and used the
least-damped streamwise rolls as a starting point, but the intuitive choice of Faisst &
Eckhardt (2003) leads to the same coherent states. The search proceeds in two steps:
Pick a Reynolds number and a body force sufciently large such that the vortices
undergo a bifurcation in which 3D states are created. In particular, if the initial states
are translationally invariant, this symmetry must be broken. Next, try to follow the
3D states over to pipe ow without body force.
Motivated by the arrangements of vortices in plane Couette ow and the radial
shear ow, Faisst & Eckhardt (2003) started with ows containing several pairs of

456 Eckhardt et al.

 ANRV294-FL39-18 ARI 12 December 2006 6:6

Figure 4
Cross sections of a traveling wave at different positions along the wave. The frames are at
times 0, 1/8, 2/8, and 3/8 of a period and at a xed position along the axis. The velocity
components in the plane are indicated by arrows. For the axial component the difference to a
parabolic prole with the same mean speed is color coded. Regions where the uid ows
faster are shown in red and correspond to high-speed streaks. Similarly, regions where the
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

speed is lower are shown in blue and correspond to low-speed streaks.

vortices. An example of the coherent structures obtained is shown in Figure 4. All

coherent states identied so far are, by construction, highly symmetric. They contain
n = 2, . . . , 5 vortex pairs that generate n or 2n high-speed streaks close to the wall and
n low-speed streaks in the center. The high-speed streaks remain fairly rigid close
to the walls, whereas the low-speed streaks wiggle considerably in the azimuthal
direction. The vortex number does not x the states uniquely: There can be several
states with the same number of vortices (Wedin & Kerswell 2004).
The critical value for the appearence of the coherent states depends on their wave-
length. The one with the lowest critical Reynolds number has three vortex pairs and
appears in a saddle node bifurcation near Re = 1250 with an axial wavelength of 2.1 d.
Actually, there are several similar states that appear at comparable Reynolds numbers
(Wedin & Kerswell 2004). The state with two vortex pairs arises at about Re = 1350,
and the one with four at Re = 1690. The critical Reynolds numbers continue to in-
crease as more vortices are added.
Interestingly, only the states with two or more vortex pairs give rise to sym-
metric coherent traveling states. The one with a single pair, which gives the
strongest linear amplication (Schmid & Henningson 1994, Zikanov 1996), does
not. A less symmetric version of a two-vortex state appears in a different context in
Section 5.

4.2. Detecting the Structures in Experimental Data

Turbulent dynamics has a positive Lyapunov exponent and is chaotic, so how can the
coherent structures show up in experiments?
One can imagine that the orbit in state space will reside for some time in the
vicinity of the unstable saddle points. Around each saddle point one can dene a
region for which the ow state is close enough to the saddle point so that the ow
pattern is very similar to the traveling wave solution. Provided that the residence time
within each of these volumes is at least a substantial fraction of the transit time from
one coherent state to the next coherent state, it is possible to detect a ow pattern
that has a strong resemblance to the exact traveling wave solutions. However, due to

www.annualreviews.org Turbulence Transition in Pipe Flow 457

 ANRV294-FL39-18 ARI 12 December 2006 6:6

the strongly unstable character of the traveling wave solutions, the correspondence
will be more of a qualitative nature than an exact quantitative one. Furthermore, the
ratio of residence time and transit time will (rapidly) decrease for increasing Reynolds
number: (a) With increasing Reynolds number, the number of coherent ow states
increases, which likely reduces the probability of nding the ow in the vicinity of
any of the coherent states, and (b) it is likely that the volume for which the ow
state is sufciently close to the saddle point becomes smaller for increasing Reynolds
number. Hence, one can only expect to nd ow patterns that resemble those of the
coherent ow states in the low Reynolds number region. Because the ow state is not
likely identical to the exact traveling wave solutions, one must rely on the appearance
of the main features, i.e., the azimuthal periodicity of the high-speed and low-speed
regions, and the presence of the vortex rolls for its detection.
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

Empirically, one can then dene an indicator for the coherent structures and study
the frequency with which this indicator signals their presence (Hof, van Doorne et al.
2004; T.M. Schneider, J. Vollmer & B. Eckhardt, in preparation). Ideally, this indica-
tor would check how well the spatial structures of all velocity components match, but
in view of the many dimensions, an inaccessibly large number of experiments and real-
izations would be required. Experimentally and theoretically, the approach is to allow
a projection onto a lower dimensional subspace and to consider the frequency of ap-
pearance in that subspace. Hof, van Doorne et al. (2004) use a correlation function that
focuses on the prominent downstream vortices and their symmetry for the projection.
Projecting this correlation function onto frequencies of three and four vortex
pairs, combined with a threshold, allows us to identify the presence of such coherent
arrangements in several regions of the ow. This was rst applied to experimental
data in a long water-lled pipe ow facility obtained from a stereoscopic particle
image velocimetry (PIV) system (Hof, van Doorne et al. 2004). The pipe in this
facility has a 40-mm inner diameter with a total pipe length of 26 m. A carefully
designed contraction and ow-conditioning section at the inlet allows the realization
of laminar pipe ow over the full length of the pipe up to a Reynolds number of
60 103 . The volume ow rate is maintained at a constant level by means of a
feedback loop connecting the pump to an electromagnetic ow meter.
The PIV measurement technique yields all three instantaneous velocity com-
ponents in a cross section of the ow perpendicular to the pipe axis. The velocity
information is obtained from the motion of small particles carried by the ow, which
are illuminated by means of a thin light sheet generated from a pulsed laser system and
which are observed by two cameras in a stereoscopic conguration. This congura-
tion makes it possible to determine the secondary ow patterns represented primarily
in the radial and azimuthal velocity components, which are an order of magnitude
smaller than the axial velocity component. Details of the experimental conguration
and the validation of the measurement precision will be given by C.W.H. van Doorne
& J. Westerweel (under revision).
The high sampling speed and the spatial resolution of the cameras made it possible
to obtain good temporal and spatial resolution of the velocity elds up to Reynolds
numbers of about 5000. By calculating the azimuthal correlation of the streamwise
velocity, coherent ow states could be identied. The arrangements of the vortex

458 Eckhardt et al.

 ANRV294-FL39-18 ARI 12 December 2006 6:6

Figure 5
Pairing (a & b, c & d, e & f )
between ow structures
detected in experimental
cross sections (top row) and
numerically determined
traveling waves (bottom row).
The representation of the
velocity elds is the same as
in Figure 4. From Hof, van
Doorne et al. (2004).
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

rolls and high- and low-speed streaks of these states closely resemble those of the
traveling waves.
By means of this analysis several coherent ow states could be identied in both
fully developed turbulent pipe ow and turbulent puffs traveling through the pipe
(see Figure 5): Coherent ow states with three and two vortex pairs were observed
in turbulent puffs at Re = 20002500, and coherent ow states with four and six vor-
tex pairs were observed in fully developed turbulence at Re = 3000 and Re = 5300,
respectively. As mentioned above, one cannot expect to nd the exact traveling wave
solutions due to the unstable nature, but the observed ow patterns would at least
show the main features, such as the counter-rotating vortices and low-speed and
high-speed ow regions. The observed patterns will be disturbed as they do not
occur as isolated and carefully balanced solutionsas in the Direct Numerical Sim-
ulations (DNS)but occur in a natural strongly dissipative ow state. Nonetheless,
the resemblance between the numerically found ow states and those observed in
experiments is striking.
The full 3D velocity eld can be recovered from a time series of stereoscopic
PIV measurements at a xed location by assuming that the velocity eld changes
slowly while it is advected downstream with the mean ow velocity (Taylors frozen
ow assumption). This reconstruction makes it possible to determine the structure
of the coherent states in the axial direction (Hof et al. 2005). In agreement with the
observations for the exact traveling wave solutions, the low-speed streaks observed
experimentally showed a clear wavy modulation in the streamwise direction. For
several of these observations, the duration of the observed coherent ow state was
sufcient to observe pairs of counter-rotating vortices along the streaks, and hence
to obtain an estimate of the wavelength of the coherent ow state (see Figure 6).
The length scale of the wavy modulation as well as the distance between vortex pairs
observed was in good agreement with those found for the traveling wave solutions.

www.annualreviews.org Turbulence Transition in Pipe Flow 459

 ANRV294-FL39-18 ARI 12 December 2006 6:6
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

Figure 6
A section of a puff with the coherent structures and their wavelength highlighted. Positive and
negative vorticity is shown in yellow and red. The wavy low-speed streak (shown in blue) is
sandwiched between counter-rotating streamwise streaks, identied through the vorticity
distribution. From Hof et al. (2005).

Hence, the most characteristic features of these traveling wave solutions, i.e., the
counter-rotating vortices in conjunction with high-speed and low-speed streaks, as
well as the wavelength for these coherent ow states, could be observed in the ow
patterns of actual pipe ows. Further experiments are currently being evaluated to
determine the relative occurence of these coherent ow states.

5. EDGE OF CHAOS
The coexistence of stable laminar and turbulent dynamics naturally leads to questions
regarding the nature of the boundary between them. Scanning the dynamics for
initial conditions, obtained, for example, by adding a perturbation of xed spatial
structure but varying amplitude to the laminar prole, one can distinguish regions
with smooth variations in lifetime from regions with irregular variations (see the
sketch in Figure 7). The points between the smooth and chaotic regions lie on
the edge of chaos: Operationally, they can be detected as the rst initial conditions
with innite lifetimes when coming from the laminar side. They stay away from the
laminar prole, but they also do not swing up to the turbulent dynamics. Numerical

460 Eckhardt et al.

 ANRV294-FL39-18 ARI 12 December 2006 6:6

Strange Laminar
saddle basin
Edge
state
Lifetime

Stable
manifold
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

Amplitude
Figure 7
Edge of chaos in shear ows. By increasing the amplitude of a perturbation, one can
distinguish regions with smooth variations in lifetimes and others with irregular variations
(dark blue line). The limiting points between the two regions are on the edge of chaos. As
indicated by the line connecting them (lavender line), they belong to the stable manifold of an
invariant structure that separates the laminar from the turbulent. From Skufca et al. (2005).

simulations and theoretical considerations suggest that they collapse onto structures
that are attracting within the edge of chaos, but are unstable perpendicular to it,
so-called relative attractors. The relative attractors can be simple, such as traveling
waves, but can also be fairly complicated chaotic objects.
The invariant structures in the edge of chaos can be obtained by direct shooting
methods (as in Itano & Toh 2001) or by successive renements that enable one to
follow the edge of chaos for longer times (as in Skufca et al. 2005). For pipe ow we
have tracked this intermediate state for lifetimes up to 2500 R/uc . The edge state is
dominated by a pair of vortices that are off center, and shows a persistent dynamic
variation of the low-speed streaks in the center (see Figure 8).

Figure 8
The structure of the edge state in a pipe ow at a Reynolds number of 2875. The cross section
on the left is dominated by two off-center vortex pairs and their high-speed streaks close to the
wall. The slice along the axis at an angle that cuts right through the middle of the two vortices
shows the downstream variation. The absence of any periodicity is indicative of the persistent
dynamics of the edge state.

www.annualreviews.org Turbulence Transition in Pipe Flow 461

 ANRV294-FL39-18 ARI 12 December 2006 6:6

6. SUMMARY
In the previous sections we emphasized the dynamical system characteristics of tran-
sition to turbulence in pipe ow, including the critical amplitude for transition, the
sensitive dependence on initial conditions in the transition region, the lifetime dis-
tribution, and the edge of chaos. In this section we summarize these ndings and
put them into perspective under three different headings: critical Reynolds number,
coherent structures, and transition mechanism.

6.1. Critical Reynolds Numbers

The intermittent dynamics in the transition region implies that values of critical
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

Reynolds numbers depend on the specic denition. For pipe ow one can distinguish
the following four situations:
The strongest requirement for the evolution of a perturbation to a ow is that its
energy content decays monotonically for any initial condition: This is the requirement
of energy stability. The associated critical Reynolds number Re E can be determined
from an analysis of the linearized equations of motion. For pipe ow this gives R E =
81.5 ( Joseph 1976).
Next, one can give up monotonicity, but still require that any perturbation decays
eventually. This denes the critical Reynolds number ReG for global stability. A system
is globally stable if the laminar prole is the only permanently sustained state in the
system. The Reynolds numbers ReTWi , at which any stationary state or traveling wave
appears, provide upper bounds on ReG , i.e., ReG = minRe TWi . From the coherent
structures described in Faisst & Eckhardt (2003) and Wedin & Kerswell (2004), one
concludes that RG 1250. Although we expect this to be the lowest value within the
class of traveling waves studied, it cannot be ruled out that other structures, perhaps
with less symmetry or more complicated time dependence, already occur at even
lower Reynolds numbers.
Experimentally, a transition to turbulence is not observed until somewhat larger
values. To eliminate the inuence of the sensitive dependence on initial conditions,
in Section 3 we advocate the use of probabilities. The probability P (t, Re) to remain
turbulent for at least a time t at a Reynolds number Re shows an exponential tail
that is free from the details of the initial perturbation and can be characterized by a
well-dened time. From this distribution one can extract a critical Reynolds number
Reexp , for instance, by requiring that over the duration of the experiment (texp ) only
10% of all repetitions decay: determine Reexp such that P(texp , Reexp ) = 0.9. The rapid
increase in lifetimes suggests that even for the longest pipes such a value will remain
below about 2250 (this is the value from DNS as it is the largest value reported so
far; experiments point to a lower value, as discussed in Section 3).
Finally, the decreasing critical amplitude needed to trigger turbulence suggests
that for sufciently high Reynolds numbers it is impossible to maintain the lami-
nar prole. Effects like thermal uctuations, compressibility, and deviations in the
prole due to Coriolis forces, alignment of the tube, or smoothness of the surface
will become important. The mathematical version of these problems is rooted in the

462 Eckhardt et al.

 ANRV294-FL39-18 ARI 12 December 2006 6:6

non-normality of the linearized problem. As Meseguer & Trefethen (2003) describe,

at Reynolds numbers Re 105 perturbations as small as 105 can sufce to introduce
growing eigenmodes. The Reynolds number where such tiny perturbations begin to
dominate is not universal, but nite.

6.2. Coherent Structures

A great deal of effort has focused on detecting and characterizing coherent structures
in turbulent ows (see, for instance, Holmes et al. 1996, Panton 2001, Robinson
1991). The traveling waves provide a dynamical and fully nonlinear approach to the
problem. Traveling waves share with the usual coherent structures the presence of
some large-scale features, a predictable dynamics, and a relatively frequent occurence.
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

They have the additional bonus of being exact solutions to the equations of motion,
which is why the term exact coherent structures has been suggested (Waleffe 1998,
2001, 2003). A link between coherent structures and dynamical systems was also
proposed by Itano & Toh (2001) and Toh & Itano (2005).
The possibility of connecting coherent states to specic exact dynamical solutions
to the equations of motion and to certain regions of the state space of the ow is an
intriguing one, and only partially explored thus far. Ideally, one would like to be able to
identify coherent features and calculate their relative frequency from the equations of
motion. Quantitative studies of their statistical properties, like frequency, persistence,
or contribution to momentum transport, and an accurate description of the dynamics
near the states should open up new ways to inuence ows and to predict the effects
of ow control.

6.3. Transition Mechanism

Experimental studies have long shown that certain ow patterns (hairpins, etc.) ap-
pear and grow during transition to turbulence. The dynamical system picture for
the edge of chaos (see Figure 7) suggests that these features should be connected
with the invariant state in the edge of chaos (see Itano & Toh 2001, Skufca et al.
2005). Certainly, the presence of two strong vortices connects the results from non-
normal amplication, which show that this structure gives the strongest amplication
(Schmid & Henningson 1994, Zikanov 1996). The edge of chaos analysis in Skufca
et al. (2005) suggests techniques that can be used quite generally to trace the dynam-
ics at the border between laminar and turbulent ows, and hence to determine the
relevant ow patterns and features. Itano & Toh (2001) link the appearance of bursts
to the escape from the edge state along the unstable manifold. These studies provide a
framework for further investigations of the intermittent dynamics in the equilibrium
turbulent state. It is particularly intriguing that the theory suggests that, except for
symmetries, there is one and only one invariant object, which together with its stable
manifold separates the laminar from the turbulent region. The characteristics of this
state are a pair of vortices off center, closer to the walls. Interestingly, a similar pair
of vortices seems to be the edge state in plane Poiseuille ow, as reported by Itano &
Toh (2001).

www.annualreviews.org Turbulence Transition in Pipe Flow 463

 ANRV294-FL39-18 ARI 12 December 2006 6:6

7. OUTLOOK
The methods described here carry over to several other shear ows where transition to
turbulence occurs without linear instability of the laminar prole. Extensive numerical
and experimental studies have identied the same scenario as described here for plane
Couette ow (Bottin & Chate 1998; Bottin et al. 1998; Clever & Busse 1992, 1997;
Dauchot & Daviaud 1994, 1995; Daviaud et al. 1992; Eckhardt et al. 2002; Faisst
& Eckhardt 2000; Nagata 1990; Schmiegel & Eckhardt 1997; Waleffe 1995, 1998,
2001, 2003). Plane Poiseuille ow is peculiar as it has a linear instability, albeit at
Reynolds numbers of about 5772well above the values where transition is rst
observed. However, coherent states and traveling waves have been identied there
as well (Ehrenstein & Koch 1991, Itano & Toh 2001, Waleffe 2003). Undoubtedly,
similar phenomenology can be expected in external boundary layers.
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

The identication of traveling waves and the possibility that chaos is organized
around them suggest that it might be possible to treat the statistical properties
of the ow in terms of such coherent states. In low-dimensional dynamical sys-
tems this goes under the heading of periodic orbit theory, where one can show
that by exploiting a symbolic ordering of orbits one can efciently and accurately
calculate statistical properties (Artuso et al. 1990a,b; Christiansen et al. 1997; Cvi-
tanovic & Eckhardt 1991; Ott & Eckhardt 1994). Some periodic solutions for plane
Couette ow have already been found (Kawahara & Kida 2001, Toh & Itano 2003).
Nevertheless, carrying this program through for turbulent ows, even in the transi-
tion region, remains a challenge. But the possibility of identifying certain coher-
ent structures in numerical and experimental data suggests that even if the full
program cannot be realized, some approximate realizations might be feasible and
useful.
The main open question not addressed here is the relation between the peri-
odic structures in the numerical simulations and the localized puffs and slugs in the
unbounded domain. Structured turbulence, i.e., localized turbulent patches in bound-
ary layers (Gad el Hak & Hussain 1986, Schumacher & Eckhardt 2001), turbulent
sections in pipe ow (Wygnanski & Champagne 1973, Wygnanski et al. 1975), or
banded turbulence in plane Couette and Taylor-Couette ow (Barkley & Tuckerman
2005, Prigent et al. 2002), has been documented repeatedly, but the mechanisms
remain to be elucidated. Because the turbulent patches are much larger than the typ-
ical wavelengths of the coherent structures studied here, one might hope that the
puff-and-slug-forming process is a long-wavelength dynamics on top of the coherent
structures described here.

ACKNOWLEDGMENTS
B.E. would like to thank the members of the Burgers Board at the University of
Maryland, in particular Dan Lathrop, for their hospitality during 20042005. We
thank G. Homsy and K.S. Breuer for permission to include the movies of Reynoldss
experiment with the online material. We also thank the Deutsche Forschungsgemein-
schaft and the Foundation for Fundamental Research on Matter for support.

464 Eckhardt et al.

 ANRV294-FL39-18 ARI 12 December 2006 6:6

LITERATURE CITED
Aref H, Kadtke JB, Zawadzki I, Campbell LJ, Eckhardt B. 1988. Point vortex dy-
namics: recent results and open problems. Fluid Dyn. Res. 3:6374
Aref H. 1983. Integrable, chaotic, and turbulent vortex motion in two-dimensional
ows. Annu. Rev. Fluid Mech. 15:34589
Artuso R, Aurell E, Cvitanovic P. 1990a. Recycling of strange sets: I. Cycle expansions.
Nonlinearity 3:32559
Artuso R, Aurell E, Cvitanovic P. 1990b. Recycling of strange sets: II. Applications.
Nonlinearity 3:36186
Barkley D, Tuckerman LS. 2005. Computational study of turbulent laminar patterns
in Couette ow. Phys. Rev. Lett. 94:014502
Boberg L, Brosa U. 1988. Onset of turbulence in a pipe. Z. Naturforsch. 43a:697726
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

Bottin S, Daviaud F, Manneville P, Dauchot O. 1998. Discontinuous transition to

spatiotemporal intermittency in plane Couette ow. Europhys. Lett. 43(2):17176
Bottin S, Chate H. 1998. Statistical analysis of the transition to turbulence in plane
Couette ow. Eur. Phys. J. B. 6:14355
Brosa U. 1989. Turbulence without strange attractor. J. Stat. Phys. 55:130312
Chapman SJ. 2002. Subcritical transition in channel ows. J. Fluid. Mech. 451:3597
Cherhabili A, Ehrenstein U. 1997. Finite-amplitude equilibrium states in plane Cou-
ette ow. J. Fluid Mech. 342:15977
Christiansen F, Cvitanovic P, Putkaradze V. 1997. Spatiotemporal chaos in terms of
unstable recurrent patterns. Nonlinearity 10:5570
Clever RM, Busse FH. 1992. Three-dimensional convection in a horizontal uid
layer subjected to a constant shear. J. Fluid Mech. 234:51127
Clever RM, Busse FH. 1997. Tertiary and quaternary solutions for plane Couette
ow. J. Fluid Mech. 344:13753
Crutcheld JP, Kaneko K. 1988. Are attractors relevant to turbulence? Phys. Rev.
Lett. 60:271518
Cvitanovic P, Eckhardt B. 1991. Periodic orbit expansion for classical smooth ows.
J. Phys. A. 24:L23741
Darbyshire AG, Mullin T. 1995. Transition to turbulence in constant-mass-ux pipe
ow. J. Fluid Mech. 289:83114
Dauchot O, Daviaud F. 1994. Finite amplitude perturbation in plane Couette ow.
Europhys. Lett. 28:22530
Dauchot O, Daviaud F. 1995. Finite amplitude perturbation and spot growths mech-
anism in plane Couette ow. Phys. Fluids 7:33543
Daviaud F, Hegseth J, Berge P. 1992. Subcritical transition to turbulence in plane
Couette ow. Phys. Rev. Lett. 69:251114
Doering CR, Gibbon JD. 1995. Applied Analysis of the Navier-Stokes Equation.
Cambridge, UK: Cambridge Univ. Press
Draad AA, Kuiken GDC, Nieuwstadt FTM. 1998. Laminar-turbulent transition in
pipe ow for Newtonian and non-Newtonian uids. J. Fluid Mech. 377:267312
Eckhardt B, Yao D. 1993. Local Lyapunov exponents. Physica. D. 65:1008
Eckhardt B, Aref H. 1988. Integrable and chaotic motions of four vortices: II. Colli-
sion dynamics of vortex pairs. Phil. Trans. R. Soc. London A 326:65596

www.annualreviews.org Turbulence Transition in Pipe Flow 465

 ANRV294-FL39-18 ARI 12 December 2006 6:6

Eckhardt B, Faisst H. 2004. Dynamical systems and the transition to turbulence. In

Laminar-Turbulent Transition and Finite Amplitude Solutions, ed. T Mullin, RR
Kerswell, pp. 3550. Dordrecht: Springer
Eckhardt B, Faisst H, Schmiegel A, Schumacher J. 2002. Turbulence transition in
shear ows. In Advances in Turbulence IX, eds. I Castro, P Hanock, T Thomas.
pp. 7018. Barcelona: CISME
Ehrenstein U, Koch W. 1991. Three-dimensional wavelike equilibrium states in
plane Poiseuille ow. J. Fluid. Mech. 228:11148
Faisst H, Eckhardt B. 2000. Transition from the Couette-Taylor system to the plane
Couette system. Phys. Rev. E. 61:722730
Faisst H, Eckhardt B. 2003. Traveling waves in pipe ow. Phys. Rev. Lett. 91:224502
Faisst H, Eckhardt B. 2004. Sensitive dependence on initial conditions in transition
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org

to turbulence in pipe ow. J. Fluid Mech. 504:34352

Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

Gad-el-Hak M, Hussain AKMF. 1986. Coherent structures in a turbulent boundary

layer. Phys. Fluids 29:212439
Grebogi C, Ott E, Yorke JA. 1982. Chaotic attractors in crisis. Phys. Rev. Lett. 48:1507
10
Grossmann S. 2000. The onset of shear ow turbulence. Rev. Mod. Phys. 72:60318
Guckenheimer J. 1986. Strange attractors in uids: another view. Annu. Rev. Fluid
Mech. 18:1531
Guckenheimer J, Holmes P. 1983. Nonlinear Oscillations, Dynamical Systems, and Bi-
furcations of Vector Fields. New York: Springer-Verlag
Hagen G. 1839. Uber die Bewegung des Wassers in engen zylindrischen Rohren.
Pogg. Ann. 46:423
Hagen G. 1854. Uber den Einu der Temperatur auf die Bewegung des Wassers in
Rohren. Abh. Akad. Wiss. Ber. pp. 1798
Hamilton JM, Kim J, Waleffe F. 1995. Regeneration mechanisms of near-wall tur-
bulence structures. J. Fluid Mech. 287:31748
Henningson DS. 1996. Comment on Transition in shear ows. Nonlinear normality
versus non-normal linearity. Phys. Fluids 8:225758
Hof B. 2004. Transition to turbulence in pipe ow. In Laminar-Turbulent Transition
and Finite Amplitude Solutions, ed. T Mullin, RR Kerswell, pp. 22131. Dordrecht:
Springer
Hof B, Juel A, Mullin T. 2003. Scaling of the turbulence transition threshold in a
pipe. Phys. Rev. Lett. 91:244502
Hof B, van Doorne CWH, Westerweel J, Nieuwstadt FTM. 2005. Turbulence regen-
eration in pipe ow at moderate Reynolds numbers. Phys. Rev. Lett. 95:214502
Hof B, van Doorne CWH, Westerweel J, Nieuwstadt FTM, Faisst H, et al. 2004.
Experimental observation of nonlinear traveling waves in turbulent pipe ow.
Science 305:159498
Hof B, Westerweel J, Schneider TM, Eckhardt B. 2006. Finite lifetime of turbulence
in shear ows. Nature 443:5962
Holmes P, Lumley JL, Berkooz G. 1996. Turbulence, Coherent Structures, Dynamical
Systems, and Symmetry. Cambridge, UK: Cambridge Univ. Press
Homsy GM, Aref H, Breuer KS, Hochgreb S, Koseff JR, et al. 2004. Multimedia Fluid
Mechanics. Cambridge, UK: Cambridge Univ. Press

466 Eckhardt et al.

 ANRV294-FL39-18 ARI 12 December 2006 6:6

Itano T, Toh S. 2001. The dynamics of bursting process in wall turbulence. J. Phys.
Soc. Japan 70:70316
Joseph DD. 1976. Stability of Fluid Motions, Vol. I and II. Berlin: Springer
Kadanoff LP, Tang C. 1984. Escape from strange repellers. Proc. Natl. Acad. Sci. USA
81:1276
Kaneda K. 1990. Supertransients, spatiotemporal intermittency and stability of fully
developed spatiotemporal chaos. Phys. Lett. A. 149:10512
Kawahara G, Kida S. 2001. Periodic motion embedded in plane Couette turbulence:
regeneration cycle and burst. J. Fluid Mech. 449:291300
Kerswell RR. 2005. Recent progress in understanding the transition to turbulence in
a pipe. Nonlinearity 18:R1744
Lanford OE III. 1982. The strange attractor theory of turbulence. Annu. Rev. Fluid
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

Mech. 14:34764
Meseguer A, Trefethen LN. 2003. Linearized pipe ow at Reynolds numbers
10,000,000. J. Comp. Phys. 186:17897
Moehlis J, Faisst H, Eckhardt B. 2004. A low-dimensional model for turbulent shear
ows. New J. Phys. 6:Art. 56
Mullin T, Peixinho J. 2006. Recent observations in the transition to turbulence in a
pipe. In IUTAM Symposium on Laminar-Turbulent Transition, ed. R Govindarajan.
p. 45. Bangalore: Springer
Mullin T, Kerswell R. 2005. IUTAM Symposium on Laminar-Turbulent Transition and
Finite Amplitude Solutions. Dordrecht: Springer
Nagata M. 1990. Three-dimensional nite-amplitude solutions in plane Couette
ow: bifurcation from innity. J. Fluid Mech. 217:51927
Ott E. 1993. Chaos in Dynamical Systems. Cambridge, UK: Cambridge Univ. Press
Ott G, Eckhardt B. 1994. Periodic orbit analysis of the Lorenz attractor. Z. Phys. B
94:25966
Panton RL. 2001. Overview of the self-sustaining mechanisms of wall turbulence.
Prog. Aerospace. Sci. 37:34183
Peixinho J, Mullin T. 2006. Decay of turbulence in pipe ow. Phys. Rev. Lett.
96:094501
Poiseuille JLM. 1840. Recherches experimentelles sur le mouvement des liquides
dans les tubes de tres petits diametres. Comp. Rend. 11:96167, 104148
Prigent A, Gregoire G, Chate H, Dauchot O, van Saarloos W. 2002. Large-scale
nite-wavelength modulation within turbulent shear ows. Phys. Rev. Lett.
89:014501
Reynolds O. 1883. An experimental investigation of the circumstances which de-
termine whether the motion of water shall be direct or sinuous and the law of
resistance in parallel channels. Phil. Trans. R. Soc. 174:93582
Robinson SK. 1991. Coherent motions in the turbulent boundary layer. Annu. Rev.
Fluid Mech. 23:60139
Salwen H, Cotton FW, Grosch CE. 1980. Linear stability of Poiseuille ow in a
circular pipe. J. Fluid Mech. 98:27384
Schmid PJ, Henningson DS. 1994. Optimal energy growth in Hagen-Poiseuille ow.
J. Fluid Mech. 277:197

www.annualreviews.org Turbulence Transition in Pipe Flow 467

 ANRV294-FL39-18 ARI 12 December 2006 6:6

Schmid PJ, Henningson DS. 1999. Stability and Transition of Shear Flows. New York:
Springer
Schmiegel A, Eckhardt B. 1997. Fractal stability border in plane Couette ow. Phys.
Rev. Lett. 79:525053
Schumacher J, Eckhardt B. 2001. Evolution of turbulent spots in a parallel shear ow.
Phys. Rev. E. 63:046307
Skufca J, Yorke JA, Eckhardt B. 2006. The edge of chaos in a model for a parallel
shear ow. Phys. Rev. Lett. 96:174101
Toh S, Itano T. 2003. A periodic-like solution in channel ow. J. Fluid Mech. 481:67
76
Toh S, Itano T. 2005. Interaction between a large-scale structure and near-wall struc-
tures in channel ow. J. Fluid Mech. 524:24962
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

Trefethen LN, Trefethen AE, Reddy SC, Driscoll TA. 1993. Hydrodynamic stability
without eigenvalues. Science 261:57884
Waleffe F. 1995. Transition in shear ows: non-linear normality versus non-normal
linearity. Phys. Fluids 7:306066
Waleffe F. 1998. Three-dimensional coherent states in plane shear ows. Phys. Rev.
Lett. 81(19):414043
Waleffe F. 2001. Exact coherent structures in channel ow. J. Fluid Mech. 435:93102
Waleffe F. 2003. Homotopy of exact coherent structures in plane shear ows. Phys.
Fluids 15:151734
Wedin H, Kerswell RR. 2004. Exact coherent structures in pipe ow: traveling wave
solutions. J. Fluid Mech. 508:33371
Wygnanski IJ, Champagne FH. 1973. On transition in a pipe. Part 1. The origin of
puffs and slugs and the ow in a turbulent slug. J. Fluid Mech. 59:281335
Wygnanski IJ, Sokolov M, Friedman D. 1975. On transition in a pipe. Part 2. The
equilibrium puff. J. Fluid Mech. 69:283304
Zikanov OY. 1996. On the instability of pipe Poiseuille ow. Phys. Fluids 8(11):2923
32

468 Eckhardt et al.

 Contents ARI 11 November 2006 9:35

Annual Review of

Contents Fluid Mechanics

Volume 39, 2007

H. Julian Allen: An Appreciation

Walter G. Vincenti, John W. Boyd, and Glenn E. Bugos p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 1
Osborne Reynolds and the Publication of His Papers
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

on Turbulent Flow
Derek Jackson and Brian Launder p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p18
Hydrodynamics of Coral Reefs
Stephen G. Monismith p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p37
Internal Tide Generation in the Deep Ocean
Chris Garrett and Eric Kunze p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p57
Micro- and Nanoparticles via Capillary Flows
Antonio Barrero and Ignacio G. Loscertales p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p89
Transition Beneath Vortical Disturbances
Paul Durbin and Xiaohua Wu p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 107
Nonmodal Stability Theory
Peter J. Schmid p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 129
Intrinsic Flame Instabilities in Premixed and Nonpremixed
Combustion
Moshe Matalon p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 163
Thermouid Modeling of Fuel Cells
John B. Young p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 193
The Fluid Dynamics of Taylor Cones
Juan Fernndez de la Mora p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 217
Gravity Current Interaction with Interfaces
J. J. Monaghan p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 245
The Dynamics of Detonation in Explosive Systems
John B. Bdzil and D. Scott Stewart p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 263
The Biomechanics of Arterial Aneurysms
Juan C. Lasheras p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 293

vii
 Contents ARI 11 November 2006 9:35

The Fluid Mechanics Inside a Volcano

Helge M. Gonnermann and Michael Manga p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 321
Stented Artery Flow Patterns and Their Effects on the Artery Wall
Nandini Duraiswamy, Richard T. Schoephoerster, Michael R. Moreno,
and James E. Moore, Jr. p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 357
A Linear Systems Approach to Flow Control
John Kim and Thomas R. Bewley p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 383
Fragmentation
E. Villermaux p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 419
Turbulence Transition in Pipe Flow
Annu. Rev. Fluid Mech. 2007.39:447-468. Downloaded from www.annualreviews.org

Bruno Eckhardt, Tobias M. Schneider, Bjorn Hof, and Jerry Westerweel p p p p p p p p p p p p p p p p 447
Access provided by Aligarh Muslim University on 01/18/17. For personal use only.

Waterbells and Liquid Sheets

Christophe Clanet p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 469

Indexes

Subject Index p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 497

Cumulative Index of Contributing Authors, Volumes 139 p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 511
Cumulative Index of Chapter Titles, Volumes 139 p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 518

Errata

An online log of corrections to Annual Review of Fluid Mechanics chapters

(1997 to the present) may be found at http://uid.annualreviews.org/errata.shtml

viii Contents