Annual Reviews

www.annualreviews.org/aronline

Ann. Re~. Fluid Mech.1981. 13:457-515

Copyright ©1981 by AnnualReviewslnc. All rights reserved
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

ORGANIZED MOTION 8184

IN TURBULENT FLOW
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Brian J. Cantwell
Department of Aeronauticsand Astronautics,StanfordUniversity,Stanford,
California94305

I INTRODUCTION

In nearly every area of fluid mechanics,our understandingis limited by

the onset or presence of turbulence. Althoughrecent years have seen a
great increase in our physical understanding, a predictive theory of
turbulent flow has not yet been established. Aside from certain results
that can be derivedthroughdimensionalreasoning,it is still not possible
to solve from first principles the simplest turbulent flow with the
simplest conceivable boundaryconditions. Our continuing inability to
makeaccurate, reliable predictions seriously limits the technological
advancementof aircraft design, design of turbomachinery,combustors,
mixers, and a widevariety of other devices that dependon fluid motion
for their operation.
Anyonewhois introduced to the subject of turbulence for the first
time quickly encounters the decompositionof the unsteady flow first
proposed by Osborne Reynolds in 1895. Various flow variables are
divided into a meanand fluctuating part, and uponsubstitution into the
Navier-Stokesequations the result is a systemof equations identical in
form to the original system except for convective stress terms, which
arise fromaveragingproductsof velocity fluctuations. In order to close
the system of equations, a second relation is needed between the
convective stresses and the meanvelocity field. Until recently, much
theoretical and experimentaleffort wasfocusedon finding relationships
that could be applied to larger and larger classes of meanflows with the
ultimate hope of finding a universal constitutive relation for "turbulent
fluid." There was never any guarantee that such a relation actually
exists and the goals of this effort remainlargely unrealized. Hopefor a
universal turbulence model has been slowly replaced by the growing
 Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annual Reviews
www.annualreviews.org/aronline
 Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annual Reviews
www.annualreviews.org/aronline
 Annual Reviews
www.annualreviews.org/aronline

460 CANTWELL

Figure 2a. Morerecently, Townsend(1970) inferred the double-roller

large-eddystructure shownin Figure 2b for a general shear flow and the
double-conestructure shownin Figure 2c for a wall-boundedshear flow
(Townsend1976).
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

This approach to the large-eddy structure through the Eulerian spa-

tial-correlation tensor is rooted ia the stochastic random-scalespicture
of "turbulent fluid" and suffers frora a numberof shortcomings. The
first and most obviousis that there does not exist a uniquerelationship
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

betweenthe correlation tensor and the unsteadyflow that producesit. A

secondshortcomingis that the correlated portion of the signal associat-
ed with the passage of an organized and repeatable motion will be
degraded by other contributions to the total signal. This maylead to
results that emphasizecertain meclhanismswhile ignoring others, and
that maybe erroneous or incompleteeven in a qualitative sense. Early
correlation measurements taken in nfixing layers failed spectacularly to
reveal the large spanwise eddies nowknownto dominate this flow. A
third shortcomingis that the averaging point does not propagate with
the disturbances that are responsible for the correlated portion of the
signal. A Lagrangianaveraging process might reveal quite a different
structure from the Eulerian average. A final shortcomingis that the
correlation methodleads to an incomplete picture of the flow with
vortex lines left unclosedand schematicpatterns of lines that often defy
definition. In short, the methodoffers no informationabout howarrays
of movinglarge eddies are laced together to completethe flow field.
Beginningin the early 1960s, experimentswere performedthat began
to changethe view of turbulence just ,summarized.Thelast twenty years
of research on turbulence have seen a growing realization that the
transport properties of most turbulent shear flows are dominated by
large-scale vortex motionsthat are not random.The form, strength, and
scale of these organized motions vary from flow to flow and methods
used to identify themare as varied as the motionsthemselves.

II ORGANIZED MOTION IN THE TURBULENT

BOUNDARY LAYER

(a) Motivation
It is remarkablethat the earliest ob:servations of organizedmotionwere
madein the turbulent bounda~layer along a wall wherethe motion is
mostcomplex.At least part of the :reason is simplythat this is the flow
that has historically receivedthe greatest attention becauseof its techno-
logical importanceand wouldtherefore be the most likely one to reveal
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 46 1
large eddy motions
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

centerline
"niforrmy turbulent f l W

a
ViSCDUS mper1ayer
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

centerline

b C

Figure I Several conceptual views of turbulent flow. (a) Sketch of a jet flow from
Townsend (1956); ( b ) and (c) Sketches of a wake flow from Hinze (1959).

its structure first. However, another reason has to be related to the

fascinating and compelling mean-flow behavior that the boundary layer
exhibits. To a good approximation, the mean velocity profile of a
turbulent boundary layer may be divided into three parts (see Figure
3b).
0-A: Viscous sublayer 0 < y + s 7
y+=u+. (2)
A-B: Buffer layer 75yf,(30. Several relations are available for this
region. An implicit formula due to Spalding (1961) which matches (2)
and (4) is

B-C-D: Logarithmic and outer layer 3 0 3 + <6+. An empirical

formula that works well for a variety of pressure gradients is (Coles
1956)
u+=-lny++C+-
1 W) (4)
K K
where
y + = -Y U * ,a+=-, su* (5)
V V

u+ = u p (6)
 Annual Reviews
www.annualreviews.org/aronline

462 CANTWELL

and

(7)
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

where

~’~--#~ ly=0"
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Theskin-friction coefficient is

(9)

The profile (4) is determined by the two empirical constants K and

and the function YI(x). Typical values are K=0.4 and C=5.0. The
pressure gradient determines II(x). For dP/dx--O, II=0.62. General
features of the structure of the boundarylayer are usually described in
terms of wall variables (u*, ~,) or outer variables (u~, 6), althoughit
well to keep in mind that they are not independent quantities. In
particular, the boundary-layerthickness 6 is determinedfrom the other
three. From(4)

ve(X-~.-xc-2I~)

8= u* (lO)

The first remarkablefeature of the turbulent boundarylayer is the

universality of the near-wall (y+ ~>30,y+/6+<< 1) behavior of Equation
(4). Regardless of pressure gradient, wall roughness, or Reynoldsnum-
ber, the logarithmic dependenceof u on y is observed.
The second remarkable feature is summarizedby the results due to
Klebanoff (1954) shownin Figure 3a which showthat a sharp peak
the rate of productionof turbulent energy(production = -
occurs at the outer edge of the viscous sublayer. Measurements in pipe
flow by Laufer (1954) show a similar effect. Integration over the
thickness of the boundarylayer leads to the result that the first 5%of
the boundary layer contributes over half of the total production of
turbulent energy. This important result was the primarymotivation for
the early workof Kline & Runstadle:r(1959) andlater Kline et al (1967),
and remains the primary motivation for muchof the work on boundary-
layer structure being carded on today.
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 463

Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

Figure 2 Eddy structure inferred from

correlation measurements for several
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

flows. ( a ) Cylinder wake by Payne &

Lumley (1967); ( b ) Inclined double-roller
(1 structure for general shear flow from
Townsend (1970); (c) Double-cone struc-
ture for wall flow from Townsend (1976).

(b) Method of Approach

Efforts to isolate coherent structure in the boundary layer have followed
two basic lines of approach. The first approach uses various modified
forms of the correlation method used by Townsend. The main disad-
vantage of this method in most of its various forms is that the details of
the organized flow pattern in physical coordinates are not determined.
The main advantage is that the coherent structure is represented within
a well-defined mathematical framework that allows quantitative state-
ments to be made about its statistical properties.
The second line of approach makes use of various methods of flow
visualization to make direct observations of complex unsteady turbulent
motions. Here, flow visualization is used in the broadest sense to include
conventional methods using hydrogen bubble, dye, shadowgraph, and
Schlieren techniques as well as nonconventional methods based on
conditionally averaged velocity measurements tied together to form a
picture of the flow pattern.
An extension of the Eulerian spatial-correlation method to a time-
space correlation
Rij(X, & 7 ) = Ui(X, t)Uj(X+& f+T) (11)
 Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
 Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
 Annual Reviews
www.annualreviews.org/aronline

466 CANTWELL

pressure, and wall shear were all found to be highly correlated over a
significant portion of the boundarylayer.
Beginningin the late 1950s, a series of experiments was begun at
Stanford using flow visualization to study the turbulent boundarylayer.
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

This effort culminated in the work by Kline et al (1967). Several new

features of the flow in the near-wall region of the turbulent boundary
layer were revealed. The flow in a low-Reynolds-numberturbulent
boundarylayer was visualized by a hydrogen-bubblewire placed paral-
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

lel to, and at various distances above, the wall. Theyfound that even
whenthe wire was deep in the viscous sublayer at y+ =2.7 the bubbles
did not followstraight trajectories as they movedslowlyalong the plate,
but rather they accumulated into an alternating array of high- and
low-speed regions called "streaks." They observed that the streaks
interacted with the outer portions of the flow through a sequence of
four events: gradual outflow, liftup, suddenoscillation, and breakup.To
the sequence of three events from liftup to breakup they applied the
term "bursting." In addition, they found that a favorable pressure
gradient (dP/dx<O) tended to reduce the rate of bursting and an
unfavorable pressure gradient (dp/dx>O)tended to increase the rate
and intensity of bursting. It was conjectured that the bursting phenom-
enonplays a dominantrole in the productionof turbulent energy, that it
dominatesthe transfer process betweeninner and outer regions of the
boundarylayer and in doing so plCys an important role in determining
the structure of the entire layer. Using combineddye and hydrogen-
bubble visualization plus hot-wire measurements,Kline et al were able
to estimate various scales of motion associated with the streaks and
bursts. Theydeducedfrom visual data that the average spanwisestreak
spacing (i.e. the distance for one full wavelength)for a smoothwall
all pressure gradients was approximately +--,h.zU*/l,’=
~- I00. The se-
quenceof events associated with bursting wasas follows: Initially the
streak of hydrogenbubbles drifts slowly downstreamand outwardfrom
the wall. Whenthe streak reaches y+--8-12 it begins to oscillate. This
oscillation amplifies and terminates in a very abrupt breakup in the
region 10<y+ <30. After the breakup the streak of bubbles is con-
torted, stretched, and ejected outwardalong an identifiable trajectory.
They observed that beyond y+ =40 the ejected fluid movesat about
80%of the meanvelocity in the outer part of the boundary layer.
Putting all the visual and quantitative information together they con-
strueted the schematic picture of the streak breakup process shownin
Figure 5a. Thevarious stages in the bursting process are summarizedin
Figure 5b.
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 467

Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

Figure 4 Turbulent boundary- layer

structure based on space-time correla-
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

tions by Willmarth & Wooldridge (1 963).

( u ) Correlation between two pressure
transducers mounted flush with the wall;
( b ) Vector field of pressure-velocity cor-
relations, magnitude of the vector at any
point is (R;,,+R;,)'/2, direction of the
vector is tan-'(RP,/Rp,).

Corino & Brodkey (1969) observed what they called ejections near the
wall in fully developed high-Reynolds-number pipe flow. In contrast to
Kline et al, they viewed the sublayer ( y < 5) as essentially passive with
+

the ejections originating in a zone away from the wall between y + = 5

and y + = 15. It was noticed that the ejection always ended with an
action called a sweep, which consisted of axial movement of upstream
fluid sweeping out fluid from the previous ejection event. They ex-
amined the effect of Reynolds number on the frequency of the ejection
process over a range from 2300 (laminar flow) to 50,000and in general
found that the number and intensity of events increased with increasing
Reynolds number. Corino & Brodkey estimated that the ejections
accounted for approximately 70% of the Reynolds stress measured by
Laufer (1954).
These results were confirmed by Kim et a1 (1971) who showed that
virtually all of the net production of turbulent energy in the range
O<y+ < 100 occurs during bursts.
Willmarth & Lu (1972) studied the instantaneous u ' d product near
the wall and found very large values during bursting with rare events
reaching u'u'-60(u'o') at y + =30.5. They also found that large con-
tributions to the Reynolds stress occurred during the sweep phase
observed by Corino & Brodkey, A similar observation was also made by
Grass (1971). Prior to this, most of the contributions to the Reynolds
stress were assumed to occur during the outflow of low-speed fluid.
Taken together, these initial observations constitute a significant step,
which has provided the inspiration for much of the work on turbulent
 Annual Reviews
www.annualreviews.org/aronline

468 CANTWELL

boundarylayers that has followed. They illuminate an important link

between a quasi-deterministic, repeatable unsteady motion and the
production and maintenanceof meanturbulent transport.
In the initial studies of bursting the process was viewed as an
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

essentially wall-boundedphenomenon with scales of motion determined

from wall parameters u* and ~,. Thus it cameas somethingof a shock
whenRaoet al (1971), after examiningdata over a fairly wide range
Reynoldsnumber(600 < R0<9000),:~ showedthat even in the wall layer
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

the meanburst period scales with outer (u~, 8) rather than inner (u*,
variables with the meandimensiorde:~stime betweenbursts given by
u~T
02)
Moreover,they found that the meanburst rate did not vary greatly with
distance from the wall. Raoet al suggested that such bursts maybe a
general feature of all turbulent flow~. Theyvisualized large outer eddies
scouring the slow-movinginner layer releasing bursts of turbulent
energy by creating regions of intense shear in the inner layer by
triggering local instabilities. Theinner layer is seen as neither passive
nor solely responsible for energyproduction, but as strongly interacting
with the outer layer. They also suggest a mixed scaling with inner
variables for the spanwisespatial scale and outer variables for the time
between bursts which leads to u~u*/Fi~*~, (where 8" is displacement
thickness and F is the burst rate per unit span) as a quantity that is
practically independent of Reynoldsnumber. Aside from brief discus-
sions in Kline et al and Rao et al, the remaining literature on this
subject takes relatively little notice: of the needfor informationon the
scaling parameters for the spanwisespacing betweenbursts.
In an excellent and very extensive study, Grass (1971) used hydrogen-
bubbledata corrected for the lag effect due to the bubble-wirewaketo
measurestructural features of turbulent flow over smooth,transitionally
rough, and fully rough walls (u*k/~=O.O,20.7, 84.7 where k is the
roughness height). He found that ejections and inrushes were present
irrespective of the surface roughness.Grasssuggests a universal ejection
type of momentum-transportmedaanismwhich extends across a major
portion of the boundary-layerthickness. The mechanism is visualized as
jets of low-momentumfluid ejected from the boundary region and
randomlydistributed with respect to position and time. He suggests
further that the general ejection process is a common feature of the flow

SThe range has since been extended l:,y Narayanan & Marvin (1978) to 600<Ro<
35,1300.
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 469

Dynamically unstable

I
r local shear layer
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

U
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Flow
Figure 5 Schematic views of near-wall
--3 p@@‘p turbulent boundary-layer structure based
Dye on direct observations. (a) Mechanics of
slot streak breakup by Kline el al (1967), ( b )
Sequence of events in ( a ) by Kline
b (1978).

structure irrespective of boundary-roughness conditions. This is an

important conjecture that raises some question regarding the precise
role of sublayer streaks in the bursting process. The wall flow of a fully
rough-wall boundary layer must be substantially different from that of a
smooth-wall boundary layer. Yet the basic structure of the organized
motion appears to be largely unchanged.
More recently, Blackwelder & Eckelmann (1979) have made a rather
detailed study of the structure of wall streaks using heated wall elements
to infer streamwise and spanwise vorticity at the wall (Figures 6a and
b). They find the strength of streamwise vortices to be about one order
of magnitude less than the mean spanwise vorticity. They identify the
low-speed streak observed by Kline et al and others as the accumulation
region between streamwise vortices where the vertical component of the
secondary motion is directed away from the wall. They find the stream-
wise length of the vortices to be Ax+-lOOO.

(d) Outer-Flow Studies

As evidence for organized structure near the wall accumulated, atten-
tion began to focus on the flow in the outer part of the layer and the
possible connection or interaction which may exist between the outer
and wall layer.4 Kovasznay et a1 used conditionally averaged space-time
4As a general rule, the term “wall layer” refers to O<y+ < 100, which includes the
viscous sublayer and a portion of the logarithmic region. The term ‘‘outa layer” refers to
all the rest.
 Annual Reviews
www.annualreviews.org/aronline

470 CANTW]3LL

autocorrelations of several flow variables to drawa three-dimensional

correlation mapof the outer structure (Figure 7). They observedthat
the vorticity appearedto exhibit a discontinuity across the turbulent
interface of the bulge whereasthe velocity wascontinuous. In addition,
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

they noticed that there was a considerable difference between the

upstream-facing (back) and downstream-facing (fron0 portions
turbulent humpsin the outer flow. Theysuggest, almost in passing, that
if the flow were viewedin a coordinate system movingwith the average
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

convectionvelocity of the interface, fluid wouldappear to arrive at a

stagnation point on the back of a bulge located at about y/8=0.8. This
brief remarkrepresents a most important observation whichforms part
of a common thread that runs through muchof the recent literature on
organized structure in turbulent flow; that it is the upstream-facing
portion of the turbulent-nonturbulent interface that is mostactive, and
that this activity is associated with a saddle-point flow in a convected
frameof reference. A widevariety of flows, not just turbulent boundary
layers, seemto exhibit this property. Observationsof intense turbulent
activity along upstream-facing interfaces maybe found in Wygnanski&
Champagne (1973; the turbulent slug in pipe flow), Falco (1977; turbu-
lent boundary layer), Brown& Thomas(1977; turbulent boundary
layer), Wygnanskiet al (1976; turbulent spot), Cantwell et al (1978;
turbulent spot), Gad-el-Hak& Blackwelder(1979; turbulent spo0, and
Cantwell& Coles (1980; cylinder near wake).
Kovasznay et al found that the individual bulges in the outer flow are
correlated over 36 in the streamwise direction and ~ in the spanwise
direction. They conjecture that the bulges in the interface become
passive and that only the birth of newejected lumps(presumablyfrom
the wall) is the mechanismthat maintains the Reynoldsstress of the
outer layer. They speculate that the bursts observed by Kline et al
(1967) near the wall are responsible for the large-scale motionin the
outer flow that they observe. Theyalso conjecture that the large-scale
wall-pressure fluctuation pattern maybe caused by the same mecha-
nism.
Blackwelder & Kovasznay(1972) extended the earlier work with
measurements closer to the wall. Theyfound that intense motionsin the
wall region remainedstrongly correlated out to y/6~-0.5 confirming
other observations that the disturbance associated with bursting extends
across the entire layer.
In a pair of papers, Often & Kline (1974, 1975) attempted to draw
kinematical description of the relationship betweenthe inner and outer
flow. They posed a causal relationship for the interaction between
bursts and the flow in the logarithmic region that produces sweeps
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 47 1
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

(1

Figure 6 Model of near-wall turbulent

boundary-layer structure from Black-
welder (1978). ( n ) Counter-rotating
streamwise vortices with the resulting
low-speed streak; ( b ) Localized shear-
layer instability between an incoming
b sweep and low-speed streak.

which, in turn, influence the generation of bursts farther downstream.

Inspired partially by observations of vortex interactions in the plane
mixing layer, they advanced the hypothesis that the bulges in the
superlayer may be the consequence of vortex pairing between the
vortices associated with two to four bursts.
Brown & Thomas (1977) correlated wall shear with velocity across the
layer. They observed that the wall shear has a slowly varying part and a
high-frequency part and that the two appear to be coupled. They
established a line of maximum correlation which lay at an angle of 18"
to the wall in the downstream direction and hypothesized that this was
due to some organized structure at an oblique angle to the wall which
produces a characteristic response in wall shear as it moves along the
plate at about 0.8 u,. They also observed a sharp step in the velocity
which occurs at the trailing interface of the outer bulge. Falco (1977)
combined visual and hot-wire observations in the outer layer. He also
noticed considerable activity on the trailing interface of the outer bulge
which he associated with Reynolds-stress-producing motions due to
small-scale eddies in the outer layer with length scales on the order of
100 to 200 Y / u * . Brown & Thomas, Falco, and Blackwelder &
Kovasznay draw very similar sketches of the organized structure (Fig-
ures 8a, b, and c).
 Annual Reviews
www.annualreviews.org/aronline

472 CANTWELL

In an extensive visual study, Smith (1978) used a movingframe

reference to study the interaction betweeninner and outer layers in a
turbulent boundary layer at Reynolds numberssomewhathigher than
the range studied by Falco. Heobserves the burst sequenceto be related
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

to the passage of a large-scale motion with a generally transverse

rotation similar to observations by Nychaset al (1973). Thelarge-scale
motionis describedas an agglomera~;ionof smaller-scale vortical struc-
tures of varying sizes, strengths, orientations, and coherencywith an
overall spanwiserotation. Velocities of different structures within the
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

large-scale motion varied from 0.7 u~ to 0.95 u~ with the center of

rotation movingat about 0.8 u~. In some of the observations of the
interaction of wall-regionfluid with the outer flow, free-shear-layer type
vortical structures were observed to form in a region betweeny÷= 10
and 40. He argues that the formation of these spanwise wavelike
motionsis a key mechanism in the entrainment process of the low-speed,
wall-region fluid into the higlaer-speedsweepfluid and thus is a source
of the high-Reynolds-stress production in the wall region during the
bursting process.
Head& Bandyopadhyay (1978) u:~ed flow visualization and hot-wire
measurements to drawquite a different picture of the turbulent boundary
layer (Figure 9a). They makethe point that Reynolds-number effects
the detailed boundary-layerstructure are likely to be importantand that
experimentsat Re0<1000(which covers about two-thirds of the litera-
ture on this subject) maygive results that are quite unrepresentativeof
those at really high Reynoldsnumbers.They suggest that for values of
Re0 in the range 1000-7000the :most characteristic feature of the
boundarylayer is not the existence of large-scale coherent motions, but
of structures formedby the randomamalgamationof features that are
small in the streamwise direction but highly elongated along lines at
about 40° to the surface. It is inferred that these represent hairpin
vortices similar to the horseshoe vortices postulated by Theodorsen
(1955). Theysuggest that the Reynolds-stress-producingouter eddies
Falco are in fact the tips of the hairpin vortices. This modelhas some
features in commonwith some recent computations by Leonard (1979)
of three-dimensional vortex motio:ns in a developing turbulent spot
(Figure 9b). Here, initially transverse vortex lines near a wall are
perturbed and the perturbation is allowed to grow, Eventually a large-
amplitude motion is observed in which the (now wavy)array of vortex
lines assumesa shape similar to a family of Theodorsenvortices with
their heads pointed downstream at an inclined angle to the plate. More
recently, Perry et al (1980) have proposed a model of near-wall
boundary-layerstructure based on strings of A-shapedvortices similar
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 473

2.53.03.54.04.5 5.05.56.0-T
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

(I

Figure 7 Turbulent boundary - layer

Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

T--2.0-1.5-1.0-0.5 0 0.5 1.0 1.5

0.50 -
structure based on space-time correla-
tions between the velocity at a fixed
probe and the velocity at a probe that is
0.25- positioned at various points in the flow.
(a) Fixed probe at y/6=0.5 from
0
Kovasznay et a1 (1970); ( b ) Fixed probe
at y / 6 0.03 from Blackwelder &
b Kovasznay (1 972).

to those observed in a visual study of boundary-layer turbulence by

Hama & Nutant (1963). Their model reproduces a number of the mean
features of wall turbulence including the logarithmic velocity profile.
All of the investigations discussed thus far have involved observations
in naturally occurring turbulent boundary layers. Several additional
investigations should also be noted. Zilberman et a1 (1977) generated
turbulent spots and allowed them to pass into a turbulent boundary
layer that was tripped by a spanwise array of spheres. They were able to
track the perturbation associated with the spot for approximately 70
turbulent-boundary-layer thicknesses in the streamwise direction. The
convection speed of the disturbance was 0.9 u, and the main features of
the spot structure imbedded in the boundary layer were in general
agreement with other observations of the outer large structure.
Part of the difficulty in measuring the organized motion lies in the
fact that it is extremely difficult to isolate and average. An interesting
solution to this dilemma was proposed by Coles & Barker (1975). They
created a synthetic boundary layer in water by using a series of
controlled disturbances near the leading edge of a flat plate. These
produced systematic moving patterns of turbulent spots in a laminar
flow which, when averaged, gave a turbulent boundary layer with a
standard mean-velocity profile. The idea here is to create a flow that
simulates the naturally occurring flow but is much easier to study. This
work has been continued in an air boundary layer by Savas (1979) who
made extensive measurements of intermittency in the outer part of the
layer. He found that the celerity of outer large eddies was 0.88. Savas
 Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
 Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
 Annual Reviews
www.annualreviews.org/aronline

476 CANTWELL

Wygnanski et al (1976) mademeasurementsof the equilibrium puff

Re = 2220. A streamline picture of dae puff in movingcoordinates is
shownin Figure 10b. In this frame of reference, the flow consists of two
large flattened ring vortices whichrotate in the samedirection along
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

with a small eddy in the vicinity of the rearward interface. Theturbu-

lent intensity in the puff increases gradually towardsthe rear of the puff
and attains a maximum at the trailing edge. Near the leading edge of
the puff there is no clear interface betweenlaminar and turbulent flow,
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

in contrast to the turbulent slug whichis boundedby two wall-defined

interfaces spanningthe entire cross section of the pipe.
Another curious, highly organized, wall-boundedflow is the spiral
turbulence between two counter-rotating cylinders studied by Coles &
Van Atta (1967). Under certain cortditions the flow between two con-
centric, counter-rotating, cylinders consists of helical bands of alter-
nating laminar and turbulent flow. The band of turbulence rotates at a
speed roughly halfway between the speed of the inner and outer
cylinders (Figure IOc). Fluid enters the turbulenceat a severe angle, but
leaves at a grazingangle so that the rate of detrainmentof fluid passing
from a turbulent to a laminar state is actually very low. In fact, the
shape of the upstream-facing interface appears to be controlled prim-
arily by the process of viscous decay. A similar observation was made
by Wygnanski& Champagnein the turbulent slug in pipe flow. They
found that a unique relationship existed betweenthe velocity of the
fluid and the velocity of the upstream-facinginterface such that sudden
relaminarizationof turbulent fluid is prevented.
Another wall-boundedflow that has received considerable attention
is the turbulent spot first studied by Emmons (1951) and Mitchner
(1954). Shortly after these initial observations, Schubauer& Klebanoff
(1955) defined celerities (0.55-0.9), for the upstreamand downstream
interfaces of the spot. Elder (1960) looked at the mergingof spots and
found a strongly nonlinear interaction in which one spot mergeswith
another with very little loss of identity. Coles & Barker (1975) and
Wygnanskiet al (1976) madedetailed measurementsof the streamline
pattern of an average spot. Both investigations concluded that the
ensemble-averagedspot was essentially a single large horseshoevortex
superimposedon small-scale motions, which average out in the mean.
Thestructure is not unlike a Theodorsenvortex, but larger. Theexten-
sive detail of the pictures drawnby Wygnanski et al of the puff and spot
(Figures 10b and l la) and Coles & Barker of the spot represent
significant advanceover earlier attempts to visualize organized motion
in turbulence. They also represent somethingof a retrenchment from
the complexity of the turbulent boundary layer to flows where the
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 477

Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Figure 9 (a) A conceptual model of turbulent boundary-layer structure based on highly

elongated horseshoe vortices by Head & Bandyopadhyay (1978); ( b ) and (c) Vortex lines
in a developing turbulent spot computed by Leonard (1979).

organized motion is much easier to isolate and identify. As for most of

the major features of the spot, both investigations were in generally
good agreement. There was, however, one major area of disagreement
related to the celerity of the most energetic part of the motion. Space-
time correlations near the wall led Wygnanski et a1 to a celerity of 0.65
u,. Coles & Barker observed a conspicuous minimum in the streamwise
velocity in the outer part of the spot. By timing the arrival of this
minimum at several stations downstream of the spot origin they de-
duced a celerity of 0.83 u,.
Cantwell et a1 (1978) carried out measurements on the plane of
symmetry of the spot and made use of the approximately conical
behavior of the spot to collapse their velocity data using similarity
variables [=x/u,f, q=y/u,t and u/u, = U(5,q), v / u , = V(5,q). In
 Annual Reviews
www.annualreviews.org/aronline

478 CANTWELL

these coordinates, the equations for unsteadyparticle paths reduceto an

autonomoussystem which can be integrated graphically by simply
plotting isoclines. Theresult of this process is a diagram(Figure 1 lc)
showinghowthe spot entrains free stream fluid. A useful property of
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

the entrainmentdiagramis that it is invariant for all uniformlymoving

observers. As a result, structural features of the flow are broughtout in
a simple and invariant waywithout reference to the speed of a moving
observer. The entrainment diagramfor the spot includes four critical
points, two saddles and two stable loci. The outer focus movesat
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

0.78 u~ and is probably responsible for the velocity minimum observed

by Coles & Barker. The inner focus movesat 0.65 Uoo,the same speed
deduced from the maximumspace-time correlation measurements of
Wygnanskiet al.
Morerecently, Wygnanskiet al (1979) have found that the laminar
relaxation region following the spot is occupiedby Tollmien-Schlichting
instability waves induced by the fluctuating motions near the outer
wings of the spot. Theysuggest that spot growthis controlled by the
formation of newspots due to the breakdownof the trailing Tollmien-
Schlichting waves.

(f) Summary of the Flow Stntcture

Wehave seen a procession of different views of turbulent-boundary-layer
structure deducedusing a variety of methods. Eachmethodhas its own
special advantages and disadvantages. The correlation approach leads
to well-defined pictures whichcontain substantial amountsof quantita-
tive statistical information. However,it tends to be unphysical and
yields little information about the detailed structure of the flow. The
direct approachusing flow visualization is morephysical but leads to
conflicting results whichare often difficult to interpret and organize.
The problem here has often been comparedto the fable about the five
blind menand the elephant. Each manexplores a small part of the
animal and then makesconclusions about the nature of the wholebeast.
The fable points out the limits of humanobservation comparedto the
totality of facts required to makea correct conclusion.
Anyattempt to summarizevarious data for organized structure in the
turbulent boundary layer runs smackinto a mazeof ambiguouslabels
and conflicting definitions. Mostobservedquantities exhibit wide varia-
tion about a poorly defined mean. As a result commondenominators
are rare. Nevertheless, certain properties of the organizedstructure are
beginningto be established.
There appear to be four mainconstituents of the organized structure.
Nearest the wall is a fluctuating array of streamwisecounter-rotating
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 479

- Flow

LLeading interface Trailing interface,

Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

3.0 2.0 1.0 0

Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

O r - - - - '

r
/
0.5
R

I .o

b C

Figure 10 Organized motion in pipe flow, ( a ) Schematic of a turbulent slug from

Wygnanski & Champagne (1973); ( b ) Ensemble-averagedstreamline patterns in a turbu-
lent puff from Wygnanski et al (1976); spiral turbulence (c) from Coles & Van Atta
( 1967).

vortices. See Figures 3c and d. The vortices densely cover all parts of
the smooth wall. Slightly above the streamwise vortices but still quite
close to the wall is a layer that is regularly battered by bursts that
involve very intense small-scale motions of energetic fluid. The outer
layer is also occupied by intense small-scale motions. These are found
primarily on the upstream-facing portions of the turbulent-nonturbulent
interface; the backs of the bulges in the outer part of the layer. The
outer small-scale motions are part of an overall transverse rotation with
a scale comparable to the thickness of the layer. The various compo-
nents, along with some notation, are summarized schematically in
Figure 12. These components and fairly crude estimates of their scale,
position, celerity, and lifetime are discussed below.
1. A,-Length of sublayer structure in the streamwise direction
(Blackwelder 1978, Blackwelder & Eckelmann 1979, Praturi & Brodkey
1978). Observations vary from A,=lOO V / U * to A,=2000 Y / U * with
1000 Y / U * as a best value. An issue here, which may account for some
of the variation, is the distinction between sublayer streaks and sublayer
longitudinal vortices. The consensus of data seems to be that streaks are
the product of an accumulation process in regions that lie between
streamwise vortices where there is an upwelling of fluid in the secondary
motion (motion in a plane normal to the direction of flow). The
 Annual Reviews
www.annualreviews.org/aronline

480 CANTWELL

longitudinal scale of streaks might be quite different from the longitudi-

nal scale of the secondary vortex motions that produce the accumula-
tion. The exact form of the secondary streamline pattern associated with
sublayer vortices, and how this pattern is matched to the more random
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

secondary motions in the outer flow, Jis at present unclear.

2. ~.~--Vertical half scale of sublayer structure based on the distance
from the wall to roughly the center of a streamwise vortex (Kline et al
1967, Bakewell & Lumley 1967, Blackwelder & Eckelmann 1979,
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Kovasznayet al 1970). Observations vary from k~,= l0 ~,/u* to )~y --25

~,/u* with 15 u/u* about average. Nol:e thaty + =30 is about the vertical
distance at which the law of the wall (Equation 4) begins to be valid.
3. kz--Spanwise (full wavelength) scale of sublayer structure (Kline
et al 1967, Bakewell & Lumley1967, Kimet al 1971, Grass 1971, Smith
1978, Praturi & Brodkey 1978, Cant’~,ell et al 1978, Gupta et al 1971,
Oldaker & Tiederman 1977, Hanratty et al 1977, Lee et al 1974,
Willmarth & Yang 1970). This is probably the most universally agreed
upon quantity in turbulent boundary-layer structure. The accepted
mean value is Az = 100 ~/u*. The stati.stics of this quantity are somewhat
skewed, with the most probable value around 80 ~,/u*. Some observa-
tions (Gupta et al 1971) indicate a possible dependence of )~zu*/~, on
Reynolds number.
4. bx, by, and bz--Length scales of energetic near-wall eddies (Corino
& Brodkey 1969, Dinkelacker et al 1977, Smith 1978, Sabot et al 1977).
Here the correspondence between observations by different investiga-
tors becomes very difficult to pin down. However, it appears that in a
region between y+= 5 and 40 very energetic parcels of fluid are ob-
served to form through some mechartism of instability, which has yet to
be fully identified. Most often the mechanismis described in terms of a
fast inviscid oscillatory instability arising from a slowly varying instan-
taneous inflection point in the velocity profile (Figure 6b). Assigning
scales to the parcel of fluid that participates in this process is very
uncertain; however, typical estimates range from 20 ~,/u* to 40 u/u* for
bx and from 15 ~,/u* to 20 ~,/u* for by. There is essentially no
information on bz although the observations of Smith (1978) suggest that
b~ may be several times b,, or by. Measurementsbased on wall pressure
of the streamwise extent of energetic small-scale motions lead to some-
what larger estimates for b~ on the order of 60 ~,/u* to 100 ~,/u*.
5. Xo--The persistence distance of energetic near-wall eddies
(Dinkelacker et al 1977, and Praturi & Brodkey 1978). Values of
between 0.5~ and 1.5/~ are suggested although the evidence is very
sparse with wall-pressure measurementsindicating the larger value.
 Annual Reviews
www.annualreviews.org/aronline

Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
 Annual Reviews
www.annualreviews.org/aronline

482 CANTWELL
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Fibre 12 Schematic s~ of ,~urbMent bo~-layer st~ct~e.

1~ and 2~ v/u* would be expected. Falco suggests that ~e t~ical

eddies scale with the Taylor ~croscale and that ~ey accost for a
sig~ficant fraction of the Rcynolds stress in the outer flow. ~is is
contra~ to the commo~yheld belief that ~e R~olds stresses are
produced primarily by the largest eddies in the flow. ~e e~dence thus
far is relatively sparse and more measurements are needed over a ~de
Reynolds-number rans¢ to co~i~ these obse~ations.
8. Xt~Persistcnce distance of energetic outer-flow eddies. ~ese
eddies appear to travel appro~mately five t~es the~ own stream~se
length or about 10~ ~/u* before losing their identity.
9. C~Celefity of energetic outer-flow eddies (Kovas~ay et al 1970,
Smith 1978, and Savas 1979). Typical values between 0.8 u~ ~d 0.9 u~
~e obse~ed.
10. L~, Ly, L~Lcngth scales of the large-scale motion in the outer
flow (Kovasznay 1970, Di~elacker et al 1977, Brown & ~omas 1977,
Falco 1977, Zilberman et al 1977, S~th 1978, Head & Ban@opadhyay
1978, Pratufi & Brodkey 1978, Willmarth & Wooldfidg¢ 1962). Most
obs¢~ations ~ve a value for Lx bc~tween 8 and 28 at a heist of about
0.8~ above the surface. L~ i~erred from wall-pressure data is somewhat
smaller with typical values between 0.56 and ~. At 0.86 above the
surface, the wid~ Lz of ~¢ outer large eddy appears to b¢ between 0.56
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 483

Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

0
t

I / , Unforced
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Re-2200

Figure 13 Vortex trajectories. ( a ) In a

plane mixing layer from Brown & Roshko
(1974); ( b ) In the axysymmetric mixing
layer in the initial region of a jet from
I Bouchard & Reynolds (1978).

and 6 with the eddy centers spaced about 2.0 to 3.06 apart in the
spanwise direction.
11. X,-Persistence distance of the large-scale motion in the outer
flow (Kovasznay et a1 1970, Dinkelacker et a1 1977). Typical values are
about 1.66 and 26 at a height of about 0.86 above the plate.
12. C,-Celerity of the large-scale motion in the outer flow
(Kovasznay et a1 1970, Dinkelacker et a1 1977, Brown & Thomas 1977,
Zilberman et a1 1977, Smith 1978, Savas 1979, Cantwell et a1 1978,
Sabot et a1 1977, Zakkay et a1 1978, Willmarth & Wooldridge 1962,
Coles & Barker 1975). A variety of measurements indicate a value
between 0.8 and 0.9 u, at a height of about 0.86. Some wall-pressure
data indicate a somewhat lower value.
13. TB-Period between bursts (Kline et a1 1967, Kim et a1 1971, Rao
et a1 1971, Kovasznay et a1 1970, Brown & Thomas 1977, Falco 1977,
Narayanan & Marvin 1978, Smith 1978, Savas 1979, Sabot et a1 1977,
Zakkay et a1 1978). This is one of the more studied variables in
turbulent boundary-layer structure. Early observations scaled TB with
wall variables. Now it appears to be fairly well established that TBscales
with outer variables and the generally accepted number is TBU,/k6;
however, there is a considerable amount of scatter about this value with
a range from 2.5 to 10. It is also found that T B varies only slightly across
the layer implying, in agreement with other observations, that the
occurrence of a burst affects the entire layer.
 Annual Reviews
www.annualreviews.org/aronline

484 CANTWELL

14. fl--Angle of maximumcorrelation (Corino & Brodkey 1969,

Brown & Thomas 1977, Head & Bandyopadhyay 1978). Measurements
correlating wall shear and streamwise velocity give/~--18 °. Other ob-
servations of a characteristic lean angle for the large structure range
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

from 8° °to. 40
15. u’V~x--Maximummeasured instantaneous u’v’ (Lu & W’fllmarth
1973, Falco 1977, Nychaset al 1973),. In the outer portion of the flow
(y/a--0.8), instantaneous values of u’v’ exceeding ten times the local
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

mean have been observed. Near the wall (y +~30) values of u’v’ as
large as 60 times the local mean haw’, been observed.
16. Cpw~.~--Maximum change in wall-pressure coefficient (based on
free-stream velocity) during the passage of the organized motion
(Dinkelacker et al 1977, Savas 1979, Cantwell et al 1978). Peak-to-peak
variations on the order of 0.035 to 0.050 are observed.
17. w~--Root-mean-square streamwise vorticity fluctuations near the
wall (Willmarth & Bogar 1977, Hanratty et al 1977). Typical values
around one tenth the mean vorticity in the spanwise direction are
observed.
’ value of (u’2)l/2/u * (Coles 1978). This maxi-
18. Umax--Maramum
+’
mumoccurs at about y÷= 15. In an extensive survey of the literature,
Coles collected a considerable body of data on fluctuations in the
sublayer. The results indicate that over a wide range of Reynolds
number (100<a+ < 104), urea ~÷’ has a nearly constant value of 2.75. At
the same position v÷’ is about 0.6 and w÷’ is about 1.0. Using the
"universal" numbers, u+/w+ = 2.75 as y+ ~0 and X+ = 100 plus Equa-
tion (3) matched to (4) and (2), Coles was able to produce a model
the secondary flow in the sublayer that gave excellent agreement with
the collected measurements of u÷’, v÷’, w+’, and u’v’/%, in the range
0<y+ < 15.

(g) Discussion
Very few issues regarding the organized structure in the turbulent
boundary layer could be considered resolved. ~ It is clear from the work
of Corino & Brodkey, Kimet al, Willmarth, Grass, and others, that
most of the production of turbulent energy near the wall occurs during

5Thegenerallydisorderedstate of our ~understanding of turbulent boundary-layer

structureis well recognized
by workersin the field. Ausefulstep towardimproving
this
situation withmanygoodremarksidentifyingresolvedas well as unresolvedissuesmaybe
found in the summaryof a recent Michigan State workshop on structure by Kline & Falco
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 485

Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Figure 14 Three-dimensional structure in a plane mixing layer. ( n ) A conceptual picture

from Corcos (1979). ( b ) A chemically reacting layer from Bnedenthal(l978).

bursts. However, the interaction process that creates conditions under

which bursts occur is very far from understood. It appears fairly well
established that the mean time between bursts scales with outer varia-
bles. This is essentially a statement about the spacing between bursts in
the streamwise direction as well as the rate at which bursts are produced
at the wall, propagate, and decay. However, very little is known about
the scaling laws for the spacing of bursts in the spanwise direction.
There is a variety of views regarding the interaction between inner
and outer layers. In each case a causal relationship is drawn between
different aspects of the organized structure. However, the elliptic nature
of incompressible fluid motion makes cause and effect exceedingly
difficult to distinguish and the hope is that, once a true understanding is
reached, the need for such a distinction will vanish. Central to most
views of the interaction process is the idea that bursts are the result of
an inviscid instability of the instantaneous streamwise velocity profile
(see Figures 5a and 6b). Inflection points in the instantaneous profile at
 Annual Reviews
www.annualreviews.org/aronline

486 CANTWELL

y+ between 30 and 50 have been observed by Kline et al 1967, Smith

1978, and Blackwelder & Kaplan 1971. One view of the interaction
(Rao et al 1971, Laufer & Narayanan1971, and Kovasznayet al 1970)
sees bursting as an instability of the sublayer producedby the pressure
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

field associated with the large-scale motion in the outer layer. In

another view (Often & Kline 1974... 1975), the emphasis is on flow
disturbances whichmovein from the logarithmic region muchcloser to
the wall. In this view, sweepingmotionsfrom the log region impress on
the wall the temporary adverse pressure gradient required to bring
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

about the streak lifting that precedesa burst.

Anotheraspect of the interaction problemregards the maintenanceof
the outer flow. The predominantview is that the outer flow is in some
sense the wake formed by a composite of successive bursts near the
wall. Often & Kline view the bulges in the outer flow as the result of
vortex pairing between eddies associated with two to four bursts. In
contrast, the observations of Rao et al and Narayanan& Marvinthat
the meantime betweenbursts is nearly independent of distance above
the wall, suggestsa single structure that fills the layer.
Sealing parametersfor the length scales of streamwisevortices near
the wall are well established to be v and u*. This is essentially guaran-
teed by the universality of the logarithmicshapeof the velocity profile
near the wall [Equation (4) for y+/6+<<l]. However,the scaling
meantimes betweenbursts on outer variables, plus the fact that bursts
account for most turbulence production near the wall, contradicts this
universality. Somehow the effect of, say, pressure gradient or roughness
on bursting is manifested only through its effect on ~’~, without any
residual effect on the details of the shapeof the velocity profile.
Scaling parametersfor structural features of the outer portion of the
boundarylayers are muchmorein ~doubt. This arises from the fact that
the effect of the wall is felt throughoutthe layer [recall Equation(10)].
As a result, any eddystructure in the outer layer will exhibit a depen-
dence on wall variables. To see this consider the sealing properties of
the Taylor and Kolmogorovmicroscales in the outer portion of the
boundarylayer. If production and dissipation of turbulent energy scale
together in this region, then dimensionalanalysis leads to
~ 1 and r/ 1
(19)
~--R~ll ~ 4R3~I
6whereX and ~ are the Taylor and Kolmogorov
microscales, respectively.
~Un/ortunately,h is also used to denotesublayer streak seales whichdo not necessarily
correspondto a Taylor microscalc. To maintain the distinction, we use subscripted h’s
(h~,, h~,, h~ to denotestreak scales).
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 487

Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Figure 15 Ensemble-averagedmean flow at fixed phase in the cylinder near wake from
B. J. Cantwell and D. E. Coles (1980 in preparation). (a) Mean flow referenced to an
observer who moves downstream at 0.75 u,; ( b ) Shearing stress ( u ’ u ’ ) / u ~ at constant
phase; (c) Three-dimensional motions in the near wake of a flapped hydrofoil from
Mejjer (1965).

In a turbulent boundary layer we may write

Making use of (IO) we have

All three length scales depend exponentially on l/Cf, a quantity which

increases very slowly with increasing x along the layer. Over typical
ranges of C, where observations are made both Au*/v and qu*/v
 Annual Reviews
www.annualreviews.org/aronline

488 CANTWELL

increase with decreasing Cf although the dependenceon a fractional

powerof Cr mitigates the rate of increase somewhat.The dependenceon
Cr is somewhatstronger when8, ~,, and ~1 are normalizedon uoo. Thus
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

(22)
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Here the dependence on a power of 1/Cf leads to a faster rate of

increase with decreasing Cr. Thus as far as eddy length scales are
concerned,normalization with wall. variables will invariably lead to a
slower dimensionless rate of change with Cf or, equivalently, with
Reynolds number. No experiment to date covers a broad enough range
of Reynoldsnumberto properly resolve the scaling laws for the flow
structure in the outer layer. Moreover,the meantime between bursts
u~oT/,5 mayalso exhibit a weakdependenceon wall variables whenthe
range of Reynolds numbers has been extended. One of the most
importantneeds for future research is to extend current observations to
high Reynolds numbers, not so muchto identify new transport mecha-
nisms as Head& Bandyopadhyay (1978) suggest, but rather to establish
the scaling laws for the mechartisms that are observed to play an
important role at low Reynoldsnumbers.

III ORGANIZED MOTION IN FREE SHEAR

FLOWS

In a study of transition in a laminar free shear layer, Freymuth(1966)

noted the presence of highly regtdar vortex motions in the nonlinear
stages of transition. Heobserved that the onset of subharmonicwave-
lengths was associated with an interaction he called "slip" in whichtwo
adjacent vortices rotate about a common axis and coalesce into a single
structure. Downstream,the regular vortices appeared to give wayto a
chaotic motion. Traditionally, the breakdown to random, three-
dimensionalmotion has been argued on the basis of vortex stretching
along the principal axis of strain of the meanvelocity field. Takingthe
curl of the momentum equation leads to the equation for vorticity
~ ~ ~ui ~"~’o~ (23)

Thefirst term of the fight-hand side of (23), the vortex stretching term,
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 489

Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Figure 16 Oil slick from the tanker,

Argo Merchant, grounded off the
Nantucket Shoals in 1976 (Re-108).
Superimposed is the flow past an inclined
flat plate at Re= 10).

acts as a source term for vorticity. Any spanwise vorticity, it could be

argued, would be quickly dominated by much more intense vorticity in
the streamwise direction. Furthermore, the streamwise vortex motions
would contain most of the turbulent energy and any transverse vorticity
would contribute little to the dynamics of the flow.
In a study of the turbulent mixing layer, Brown & Roshko (1974)
discovered that the layer was dominated by large-scale spanwise vortex
motions. These motions originate in the transitional part of the layer;
they do not vanish when smaller-scale turbulence sets in and they
appear to remain as a permanent feature of the flow at all higher
Reynolds numbers, Winant & Browand (1974) carried out a detailed
study of the vortex pairing observed by Freymuth in a low-Reynolds-
number shear flow. In pairing, adjacent vortices rotate about each other
under their mutual induced velocity field. As the rotation progresses
they amalgamate into a single vortex of larger scale. Winant & Browand
suggest that pairing is the principal mechanism by which a shear layer
grows. A similar pairing process was observed by Brown & Roshko at
much higher Reynolds number. A sequence of pictures from a movie
taken by the Caltech group showing a pairing event is shown in the
upper right-hand corner of the pages of this review (see also Roshko
1976). An x-t diagram of eddy trajectories showing the amalgamation of
two and sometimes three vortices is shown in Figure 13a. Figure 13b
shows eddy trajectories in the initial shear layer of an axisymmetric jet
measured by Bouchard & Reynolds (1978). Here, vortex centers can be
observed to orbit about each other several times before coalescing.
 Annual Reviews
www.annualreviews.org/aronline

490 ¢~aWLL

The picture of the mixing layer which emerges from this is of a

double structure composedof an array of moving and interacting
large-scale spiral vortex motions superimposed on a background of
finer-scale, presumablyrandom,turbulence. Althoughthe persistence of
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

these eddies remains a subject of somecontroversy (Chandrasudaet al

1978), there is ample evidence that they are present and play an
important role at Reynolds numbers far above those at which the
mixing layer would be considered transitional (Dimotakis & Brown
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

1976). In someways,recent observations confirmseveral of the ideas of

Townsendwhohad proposed a similar double structure with fine-scale
turbulence convected about by eddies whosesize was comparableto the
overall scale of the flow. However,the large eddies were only vaguely
perceived as a field of randomly interacting, slow moving, three-
dimensionalmotions. The observations in the mixinglayer gave symme-
try and form to the conceptual picture of the large-scale structure of
turbulent flow. Suddenly it was feasible and reasonable to draw a
picture of turbulence! The hand, the eye, and the mind were brought
into a newrelationship that had never quite existed before; cartooning
becamean integral part of the study of turbulence.
Observationsof the plane mixinglayer stimulated a renewedinterest
in modeling unsteady viscous flow using discrete vortex arrays. This
methodhad been used by Abernat~hy& Kronauer (1962) to model the
laminar vortex street in a two-dimensionalwake. Nowit was resurrected
as a methodfor modeling the large-scale motion in turbulence. 7 The
essential idea here is to lump the continuous field of vorticity into
individual vortex elements. The lime evolution of the flow is then
simulatedby solving a set of first-order ordinarydifferential equations
dx i ~v

= Euj(x,,t) (24)
i÷j
wherexi is the coordinate of the i th vortex point and uj is the velocity
inducedat xi by thejth vortex point. Since the flow outside each vortex
core satisfies Laplace’sequation, tlhe velocity inducedat xi is foundby
superposition. The methodhas been used in both two and three dimen-
sions to modela numberof flows including vortex sheet roll-up (Chorin
& Bernard 1973), mixing layers (Ashurst 1977), wakes(Clements 1973,
"/In remarkableanticipation of later observations, Onsager(1945)used the Hamiltonian
structure of the equations of motion to ~malyzeprobable states tot an array of point
vortices. Whenthe "temperature"of the re’ray is negative, point vortices of the samesign
tend to organize into large compound vol~ices. For morediscussion, see the recent review
by Saffman&Baker (1979).
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 49 1
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Figure I7 Flow patterns in three-

dimensional flows from Perry, Lim &
Chong (1980). (0) Smoke geometry of a
neutrally buoyant wake structure. ( b )
Schematic representation of streamlines
in a three-dimensional wake. (c) Smoke
rising from a cigarette. ( d ) Smoke rising
b from a chimney.

Sarpkaya 1975), and turbulent spots (Leonard 1979). Generally, the

simulations reproduce the large-scale motions in these flows remarkably
well. However, they tend to do less well at simulating the associated
stresses. At least part of the reason for this appears to be due to the
neglect of small-scale three-dimensional motions, which contribute
significantly to the stress.
 Annual Reviews
www.annualreviews.org/aronline

492 CANTWELL

Observationsof organizedstructure in free-shear flows are not limited

to mixing layers. Crow & Cha.mpagne (1971) observed quasi-
deterministic motionsin the developedportion of an axisymmetricjet.
Bevilaqua & Lykoudis (1971) observed dye entrained into a two-
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

dimensionalwakeand concludedthat the interface convolutions visible

on photographsof wakesare the outer edgesof large vortices and that it
is vortex inductionby these large eddies that is primarilyresponsiblefor
the entrainment of laminar fluid by the wake.
All of the evidence discussed so far is for organized motion at the
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

largest scale of the flow superimposedon randombackgroundturbu-

lence. However,there is an increasing amountof evidence for a high
degree of order at smaller scales. Observations of highly organized
three-dimensional motions in the raixing layer have been reported by
Briedenthal (1978) whoexamineda chemically reacting flow in water.
Figure 14b showssimultaneousside and top views of the flow. The large
transverseeddies are strung together’ by a spaghetti-likenet of sma11-scale
streamwisecounter-rotating vortices not unlike the streamwisevortices
found near the wall under a turbulent boundarylayer.
A flow that has been knownto be dominated by coherent vortex
motions for a long while is the near wakeof a bluff body. Here the
eddies are producedin a constant, regular mannerand, except for some
dispersion, are not subject to pairing or any other strong interaction that
wouldobscure their identity. Measurementstaken at a constant phase
of the vortex-shedding cycle (B. J. Cantwell and D. E. Coles 1980, in
preparation) are shownin Figure 15. Figure 15a showsthe velocity field
referenced to an observer whomovesdownstreamwith the vortices. In
this frame of reference, the flow is quasi-steady and the organized
motion is manifested as a pattern of centers and saddles. Similar
patterns of the instantaneous velocity field in the wakeof a bluff body
have been measured by Davies (1976; D-shaped cylinder), Owen
Johnson (1978; circular cylinder at M=0.6), and Perry & Watmuff
(1979; three-dimensionalwakeof an ellipsoid).
Someindirect evidence for or~;anized small-scale structure can be
found in the cylinder near wakedepicted in Figure 15b. The stresses
associated with the backgroundturbulence in this flow are found to be
comparableto the stresses due to the periodic large-scale motion(taken
as a fluctuation away from the globally averaged meanflow). The
backgroundnormal stresses (u’2) . and (v’2) show expected behavior
with maximanear vortex centers and minimabetween vortices. How-
ever, the backgroundshearing stress (u’v’) shownin Figure 15b achieves
a maximum in the saddle region between the vortices. If one forms the
 Annual Reviews
www.annualreviews.org/aronline

FLOW 493
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

y(t)u~2eMalk

~ (t)*~tcM~tk +ueat

Figure18 Trajectory of a nonstcadycritical point in physical coordinates.

correlation coefficient

= (u’v’)
((u,2 ))’/2((v,2
it is found that near a vortex center, wherethe vorticity and turbulent
energy are at their maximum, R is about 0.1, whereas, in the region of
the saddle-point flow betweenvortices, wherethe backgroundturbulent
energy is at a minimum and wherethe transverse componentof ensem-
ble mean vorticity is nearly zero, R is between 0.5 and 0.6. The
backgroundturbulence in this flow is neither small nor random.It has
structure. Vortex stretching due to the straining motionat the saddle
must lead to a substantial strengthening of the componentof vorticity
 Annual Reviews
www.annualreviews.org/aronline

494 CANTWELL

aligned with the diverging separatrb~ of the saddle. Someevidence for

this effect is shownin the photographof the wakeof a supercavitating
hydrofoil shownin Figure 15c where well-defined lines of cavitation
bubbles connecting adjacent large eddies indicate a low-pressure zone
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

associated with intense, highly stretched vorticity.

In a sense these observations also confirm some early views of
turbulence in whichthe energy-containingeddies lie along the principal
axis of strain of the meanvelocity field. The mainand very important
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

difference is that the streamwisevortices convectwith the flow respond-

ing, not to the stationary meanflow, but to the unsteady strain field
imposedby the large eddies. Throughthe coupling betweenthe large-
scale motionand backgroundturbulence the flow is differentiated into
convectingregions of active and passive turbulence.
Thequestion that immediatelyarises is, At whatscale does organized
motion in turbulent flow cease? Corcos (1979) has recently proposed
model for the plane mixing laye:r in which the flow is treated as
essentially deterministic at all scales. The result is a cartoon of the
mixinglayer (Figure 14a), in whichflattened streamwisevortices with
thickness comparable to the Taylor microscale form in a perturbed
saddle-point flow betweenadjacent large-scale vortices. The prospect
raised here is that the physics of turbulent flow can be understoodand
modeledby considering an equivalent small numberof complexlaminar
flows. If this is so, then manytraditional ideas about turbulence, for
example the concept of local isotropy and the cascade picture of
turbulent energy exchangebetwee:nlarge and small scales, need to be
re-examinedin light of the possibili~ty that evenvery small-scale motions
maybe highly organized. Old ideas will not be discarded, but they will
receive a new,moreilluminating, physical interpretation.
An area of active current research, which has been stimulated by
observations of organized motion in both wall-bounded flows and
free-shear flows, is large-eddy simulation. The basic approach here
(Reynolds1976) is to computelarge-scale unsteady motions explicitly
while modelingonly those motions that lie at scales belowthat which
can be resolved by the computational grid. Several flows have been
simulated thus far with encouraging results. Calculations of homoge-
neous and isotropic turbulence were used by Clark et al (1979)
generate "empirical" constants for Reynolds-stress modelswhichprevi-
ously could only have been determined from measurements of grid
turbulence. Large-eddysimulation, of turbulent channel flow by Moinet
al (1978) reproducedmeanvelocity profiles (particularly the logarithmic
portion near the wall) whichwere in good agreementwith the measure-
ments of Hussain & Reynolds (1975). In addition, the calculations
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 495

Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

a
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

×/tr
~
b c
Figure 19 Entrainment diagrams for the solution of the creeping round jet at (a)
Re--2.0, (b) Re--8.0, (c) Re--20.0.

yielded data for quantities such as pressure and pressure-velocitycorre-

lations that cannot be measuredexperimentally.
At the heart of this approach is the assumption that unresolved
sub-grid-scale fluctuations are only slightly nonisotropic and can be
modeledin a universal way. Thecutoff betweenlarge and small scales is
essentially a function of the current state of computertechnology. As
machinesbecomelarger and faster (as they have and will become)the
fraction of the spectrum of energy containing turbulent motions that
can be computedexplicitly will grow toward one. However,even the
most optimistic forecaster of this evolution would concede that the
technology-imposedcutoff between large and small scales will never
approachthe order of the Kolmogorov scale in a high-Reynolds-number,
aerodynamicallyuseful flow. A grid of this resolution in a given volume
wouldrequire N--- Re9/4 whereN is the numberof grid points and Re is
the Reynoldsnumberof the flow in question based on global velocity
and length scales. For a 4-cm chord compressor blade at 400 m/sec
(Re~106) N equals 3× 1013. For a 200-cmairfoil at the same speed
(Re-~5× 107) N equals 2 × 1017. However,estimates of the largest mem-
ory sizes available by the end of the century (Chapman1979) do not
exceedN=10II (a four-order-of-magrfitudeincrease in size over today’s
largest memories).The above estimate of N is rather pessimistic. One
3/~’.
mayonly need to resolve a Taylor microscale, in whichcase N--~Re
In addition, efficient grid-generating schemescan be used to reduce N
substantially.
 Annual Reviews
www.annualreviews.org/aronline

496 CANTWELL

In any case, for the foreseeable future, large-eddy simulation of

high-Reynolds-number turbulent flows will have to stop at scales that
will be several orders of magnitudegreater than the smallest scales in
the flow. In this respect the observations of small-scale organized
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

motionraise a numberof critical questions. Are these motionssmall in

the sense of a Taylor or Kolmogorov microscale or are they really of
intermediate scale? Howis the coupling betweenlarge and small scales
really accomplished?Is the motiona.t small scales the samefromflow to
flow? Whatis the maximum scale si~ze at whichlocal isotropy begins to
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

be valid and will computersever be able to see this scale for flows of
engineering importance?

IV TRANSITION AND THE CONTROL OF

MIXING

Moreand more evidence has accumulated in the past few years that
showsthat virtually all turbulent shear flowsare sensitive to transition,
or moreprecisely, to perturbations applied during transition. Tradition-
ally, it wasbelieved that sufficiently far from their point of origin
turbulent shear flows would reach an asymptotic state in which their
rate of growth and decay would becomeindependent of the mannerin
which the flow was started (see, for example, Liepmarm1962). The
overwhelmingbody of data shows that once turbulence is established
the overall properties of turbulent shear flows away from solid
boundariesare virtually independentof viscosity. Althoughviscosity is
essential to the creation of turbulenceit seemsto serve only to establish
smaller and smaller scales of motion with more and more intense
velocity gradients sufficient to di~sipate the increased rate of energy
input as the Reynoldsnumberis increased. However,the Taylor micro-
scale is usually associated, on dimensional grounds, with a velocity
perturbation comparableto that associated with the largest eddies. It is
not at all obviousthat the overall behavior of the flow can dependso
little on Reynoldsnumberin the l?resence of such intense motionsthat
dependstrongly on Reynoldsnu~nber.
The same data that demonstrates the Reynolds-numberinvariance of
turbulence (illustrated by Figure 16) also shows very wide scatter
(Roshko1976). The mixing-layer spreading rates measuredby Winant
& Browandfor u2/uI = 0.4 at a streamwise Reynoldsnumberof 104 are
in close agreementwith the measurementsof Spencer & Jones (1971)
a Reynoldsnumberof 106. In contrast, the spreading rate measuredby
Wygnanski& Fiedler (1970) at Re=5×105 differs by 30%from the
value measured by Liepmann & Laufer (1947) for u2/u~=O at a
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 497

y
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

y
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

y
y

Figure 20 Entrainment diagrams for

several flows. (a) The stationary turbu-
lent mixing layer; (b) The stationary
plane turbulent jet; (c) The Oseenviscous
vortex.

Reynoldsnumberof 106. Batt (1975) suggests that this is related to the

presence or absence of a boundary-layer trip on the splitter plate ahead
of the origin of the layer. As Roshko points out, muchof the evidence
suggests that any important effects of Reynolds number appear indi-
rectly through conditions affecting transition rather than through the
direct action of viscosity on the developing turbulent structure. In a
recent evaluation of shear-layer data, Birch (1980) has suggested that
much of the scatter in measurements of shear-layer spreading rate can
be attributed to variations in the effective origin of the layer.
Effects of initial conditions have been observed by Leonard (1979)
numerical simulations of developing turbulent spots. In this study the
spot is imbeddedin an initially planar array of transverse vortex lines
(Figure 9b and c). He finds that the downstreamamplitude distribution
of the distorted and stretched vortex lines appears to be directly related
to the amplitude distribution of the initial disturbance.
Furuya & Osaka (1976) examined the effect of a distribution
roughness placed near the leading edge of a fiat-plate zero-pressure-
gradient turbulent boundary layer. Measurements of the spanwise dis-
tribution of momentumthickness showed a variation that was directly
related to the spanwise variation of roughness. Whenthe measurements
were made several thousand boundary-layer thicknesses downstream,
the same spanwise distribution of momentumthickness was found.
Instead of decaying with distance, the amplitude of the variation had
 Annual Reviews
www.annualreviews.org/aronline

498 CANTWELL

increased. In fact, the variation simply scaled with the boundary-layer

thickness. Apparently the effect of roughness placed near the plate
leading edge is to producea permanentchangein the large structure of
the layer.
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

Other evidence that shear flows are sensitive to small perturbations

maybe found in the probe interference effects on shear layers observed
by Hussain & Zaman(1978). They found that whena hot-wire probe
was inserted into a laminar free-shear layer, it was found to induce
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

stable edge-tone-like upstreamoscillations. Their results suggest that a

reference probeused near the origin of a free shear layer can alter the
basic instability frequencyof the layer, in turn influencing the down-
stream developmentof the coherent structure.
The sensitivity of shear flows to :small perturbations applied during
transition, together with the recognition that the developedflow is
dominated by organized structures which appear to be remnants of
transition, are the ingredients that suggest that someform of control
over turbulent mixing maybe possible. Oster et al (1977) were able
exert remarkableopen-loopcontrol on a plane mixing layer by flapping
a small spanwiseribbon just downstream of the end of the splitter plate
in the initial region of the layer. By adjusting the amplitude and
frequencyof the oscillation, they wereable to halve or doublethe angle
of spread of the layer, Undercertain conditions the layer wouldspread
for a while, stop spreadingfor a short distance, and then begin spread-
ing again. The vibrating ribbon strongly influences the rate and phase at
whichvorticity is injected into the flow. Theresult is controlled vortex
formation reminiscent of the "locking on" of vortex sheddingobserved
to occur on vibrating bluff bodies (.Griffin & Ramberg1974). Onecan
easily envision the use of effects of this kind in a dosed-loopsystem
wherethe flow is sensed and a perturbation is applied to producesome
desired changein the flow. There is little doubt that the next few years
will see a great deal of fruitful researchin this area.
Are turbulent eddies like tennis balls or like medicineballs? It was
once felt that an initially perturbed flow wouldreboundand eventually
reach a unique asymptotic state. It nowappears that an initially per-
turbed turbulent flow mayremain permanentlyperturbed; that it may
have an infinity of possible asymptotic states corresponding to an
infinite variety of perturbedinitial :states. If this is so, then tremendous
benefits maybe realized through the control of mixing. However,it
complicates our theoretical understanding, it is no longer enoughto
predict the angle of spread of a mixing layer; nowwemust be able to
predict all possible angles of spreadas a function of initial conditions.
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 499

This issue is still far from resolved and points up the need for
standards for taking data. Present and future measurementsneed to be
carefully and accurately scaled. In particular, it is necessary to dis-
tinguish betweendifferences in flow conditions, methodsof data presen-
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

tation and reduction, andtrue differences in state.

V ENTRAINMENT DIAGRAMS
A glossary of terms used to describe the coherentstructure in turbulent
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

flows wouldbe long indeed. There are bursts, sweeps, streaks, typical
eddies, streamwisevortices, transverse vortices, large eddies, etc. Every
term is used to describe somestructural feature of the motion. Oneof
the central problemsin current research is to relate these usually
observedfeatures to the usually not observedvariables of fluid motion;
streamlines, streaklines, pathlines, pressure, and their various deriva-
tives.
The unsteady patterns of centers and saddles observedin condition-
ally averaged turbulent flow have muchin commonwith the phase
portraits of nonlinear dynamicalsystems. Theconnectionis through the
equationsfor particle paths,
dxi( t ) =ui(x(), t).
(26)
dt
If a frame of reference can be found in whichthe flow is steady, then
(26) reduces to an autonomous systemwithintegralcurvesthat coincide
with the streamlines of the velocity field referred to the sameframe. A
numberof authors have madeuse of this fact to explore the properties
of solution trajectories in a variety of steady-flowsituations. Oswatitsch
(1958) and Lighthill (1963) classified certain critical points which
occur near a rigid boundary. Perry & Fairlie (1974) reviewedcritical-
point analysis in a general way and applied the technique to the
problemof three-dimensional separation. They placed special emphasis
on the fact that the methodprovides a wealth of topological language
particularly well suited to the description of fluid-flow patterns. Re-
cently Huntet al (1978)appliedcritical-point theory to flow-visualization
studies of bluff obstacles. Morerecently, Peake & Tobak(1980) re-
viewedthe use of critical-point theory as it is applied to the study of
three-dimensionalvortex flows about various bodies in high-speedflow.
Thesuccess of critical-point theory for studyingsteady flow, coupled
with the observations of organized spiral vortex motions in unsteady
flows (critical-point-like motions) have led to a search for ways
 Annual Reviews
www.annualreviews.org/aronline

500 CANTWELL

extend the theory to unsteady flow. Perry & Lim(1978) used critical-
point analysis to produce a qualitative description of the three-
dimensionalunsteadyflow patterns in co-flowingjets and wakes(Figure
17a and b). Pullin (1978) studied the evolution of critical points in
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

unsteadystreamline pattern associated with vortex-sheet rollup from the

edgeof an impulsivelystarted plate.
The instantaneous streamline patterns used for these analyses provide
a form of flow visualization combinedwith substantial amounts of
quantitative information about the flow. Most importantly, they focus
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

on the problemof connectingvortex: structures together to completethe

flow field. However,there are significant conceptual problemsinvolved
in interpreting unsteadystreamline patterns as they relate to entrain-
ment. In an unsteady flow, streamlines can moveacross fluid pathlines;
thus the stream function provideslittle insight into the behaviorof the
fluid itself. Furthermore,structural features of the flow often remain
hiddenin a picture of instantaneous streamlines.
Particle trajectories plotted in physical coordinatesalso present simi-
lar conceptualdifficulties. If the integration of the particle-path equa-
tions is carried out over a volumeof particles, then each point in space
will be traversed by an infinite set of trajectories, each with a different
slope corresponding to the passage of particles through the point at
successiveinstants of time. In addition, the pattern of particle paths, like
the pattern of streamlines, depends on the frame of reference. For a
recent appreciation of this problemsee Lugt (1979).
Certain time-dependentflows can be reduced to a self-similar form.
Such flows usually dependon one or at most two global parameters. In
this case, someof the above objections can be removedby reducing the
particle-path equations (26) to an autonomoussystem in similarity
coordinates.Figure1 lc is a diagramof particle trajectories in similarity
coordinates used to analyze the flow structure on the plane of symmetry
of a turbulent spot. Particle trajectories in similarity coordinates were
used by Turner (1964) to analyze the flow pattern in a rising turbulent
thermal whichwas modeledusing an expandingHill’s spherical vortex.
This methodcan be used to generate entrainment diagrams for a wide
variety of shear flows, somesteady, someunsteady, somelaminar, and
someturbulent. Someof the flows are governed by the Navier-Stokes
equations, some by the boundary-layer equations, and some by the
Oseen equation. In most cases the transformations involve simple
stretchings, although more complextransformations involving uniform
and logarithmic rotations and arbitrary nonuniform translations
(Cantwell 1978) are allowed. In fact, it appears that virtually every
incompressibleviscous flow that weordinarily think of as self-similar in
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 501

space (the Blasius boundarylayer, the round jet, the plane turbulent
mixinglayer, etc) can be thought of as a special case of a moregeneral
unsteady flow that is self-similar in time with particle paths which
reduce to an autonomoussystem. In the general case the flow is
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

assumedto have been started at someinitial time. Table 2 summarizes

various similarity transformations and the reduced equations that may
be derivedfromthem. Thebasic similarity variable is
x,-
(27)
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

kMat
where Mis an invariant of the motion with units L’T-" and a and k
are chosen so that M~tk has the dimensionsof a length. Thus
a= l/m, k~-n/m. (28)
Included are flows wherethere is a uniformvelocity to the right in the
x-direction so that V--(u~,0, 0) and flows wherethere is no externally
imposedvelocity V=(0, 0, 0). Considerthe trajecto,ry of a critical point
shownschematically as a stable focus in Figure 18.qf wetake ~l in the
direction of the external flow, and ~2 and ~3 as cross-streamdirections,
then in physical coordinates,the trajectory of the critical point is given
by
~, ~,
xc =u~t+~lcM~tk,y~=~2~M~t z~ =~3~M~t (29)
for flows governedby the full equations, by
xe =~lcn~tk,y c =~2cV~, zc =~3cV~, (30)
for flows governedby the boundary-layerequations, and by
x¢ =uoot+~lcM~tk,yc =~2¢V~,z¢ =~3cV~ , (31)
for flows governedby the Oseenequation.
If wetake 6= ~/y: +z~as a cross-stream length scale and Uo=u~ -:~
as a characteristic streamwisevelocity scale, then the four cases listed in
Table l, item 5, maybe distinguished. Using the relations in Table l,
plus the parametersthat are used to characterize various shear flows,
one can construct Table 2. Theimportantpoint in all of the aboveis the
connectionbetweenturbulence structure and specific structural details
of the velocity field. Theconceptof organizedstructure is advancedto a
description in terms of critical points in the entrainment diagramand
their relationship to the propagationand decay of the flow.
For turbulent flows the time evolution of scales is given by
6~Matk, Uo ...kM~t ~- 1. (32)
Essential to the self-similar developmentof turbulent flows is the
 Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annual Reviews
www.annualreviews.org/aronline
 Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annual Reviews
www.annualreviews.org/aronline
 Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annual Reviews
www.annualreviews.org/aronline
 Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.
Annual Reviews
www.annualreviews.org/aronline

T
 Annual Reviews
www.annualreviews.org/aronline

506 CANTWELL

observation discussed earlier, drawn from the study of homogeneous

and isotropic turbulence and confirmedempirically for shear flows, that
the energy-containing flow structure does not dependdirectly on the
value of the fluid viscosity; one can assumeReynolds numberinvari-
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

ance. This assumptionis intimately connected to the dependenceof the

flow Reynolds number on time

Re~= u°~- -- ~kM2’~

2k-I
¯t (33)
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

For flows with k= 1/2, inertial and viscous times scale together and the
assumptionis unnecessary. For flows with k > 1/2 the inertial time will
dominateat all but the smallest scales. However,flows with k < 1/2 will
tend to followa viscous scale as time increases.
Nowlet us examinethe behavior of small-scale motions. If weassume
that production and dissipation scale together then dimensionalanalysis
leads to
~k
~ 1~,~ -- 3/4 (34)
-~ ~Re~-1/2and ~ .... ~
uponsubstitution of (32)

~,~( P-~) l/2 andYl,---w3/4M-a/2k-3/4t3/4-k/2. (35)

Themainresult here is that the time evolution of the Taylor microscale

is always independentof the global, parameter M.
The entrainment diagram is always invariant for movingobservers.
Under the assumption of nonsteady similarity, the global parameter M
determines the appropriate value of k. Oncek has been determined, the
rate of convectionand growthof s~Lructuralfeatures in the flow(critical
points, turbulent interfaces, etc) :is: determined.If wechooseto move
(nonuniformly)with a coordinate system that remains attached to some
preferred feature, then, in the mo~cingcoordinate system,
x~ = xi + aiM~t/~,t’ =t, u~ =ui ~-1,
+ aikM~t (36)
and the similarity variables in movingcoordinates are
~=~,+a,,U,.(/~’)-- U,.(l~)+ka,, (37)

wherea~ is a dimensionlessrate of motionin the x,. direction.

It is clear from the above that the pattern formedby the velocity
vector field will dependon the ar This is true whetherone plots the ug
field in physical coordinates or the U~field in similarity coordinates.
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 507

Similarly, the pattern of particle displacements,dx~, in physical coor-

dinates will dependon the ai. However,the pattern of particle displace-
merits in similarity coordinates, d~,., is independentof the a~. This
follows from the form of the pathline equations in Table 1, item 3.
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

U,. ( ~ ) - k~ = U,: ( ~’) - ka~ - k~ + kai = U~’( ~’) - k~. (38)

Equation(38) is an importantresult for it states that the location and
character of a critical point in similarity coordinates is fixed by the
dynamicsgoverning the flow and by the choice of a value for k (which
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

is a consequenceof the units of M)and not by the incidental choice of

speed for a movingobserver.
The above results form the frameworkfor a powerful method for
analyzingthe dynamicsof fluid motion.Toillustrate this, let us examine
the Reynolds-number dependenceof an impulsively started, axisymmet-
ric, laminar jet producedby a point momentum source of strength J/p.
The Reynolds numberof the jet is Re=(J/p)l/2/~,. Dimensionalcon-
siderations lead to a formulation of the problemin terms of similarity
variables
t~---- ~,3/2tl/2g(~, 0; Re), ~-- r/V~ (39)
wheretp is the Stokesstream function and r and 0 are the radial distance
and azimuthal angle in spherical polar coordinates. A solution for the
creeping-flowlimit Re-->0due to Sozou(1979)

Re 2 2 ( 4 e-C/4_(2~_~)erf(~)). (40)
g(~,0)= 1-~-~sin 012~-

Byall conventionalmeasures(40) wouldappear to exhibit only a trivial

dependence on Reynolds number. However, an examination of the
entrainment diagram of (40) reveals a remarkably complexstructure
(Cantwell1980). Theequations for particle paths are
dr --u(r, O, t; Re); dO _ v(r, O, t;Re) (41)
dt dt r
or, in termsof similarity variables,

~ dO V(~, 0; Re)
d~d’~-~= U(~, O;Re)- ; d-~ = ~
(42)

wherez = In
1= t Og.v
and
1 Og
U=~2sin’---~ 00’ (43)
~sinO 0~"
 Annual Reviews
www.annualreviews.org/aronline

508 CANTWELL

Substitution of (40) into (42) leads

Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

(44)
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

-(½ + ~)erf(~)}. (45)

Thestructure of the flow is examinedby finding and classifying critical

points of (44) and(45); points (~c, 0,:) at whichboth right-handsides
equalto zero. Thezeros of (45) are at (0--0, ~r, all ~) and(~=1.7633,
0) and are clearly the same for all Reynolds numbers. Setting the
right-handside of (44) equal to zero gives

Re2 = ~’~’~ (46)

’ ~o--/~c2/4__ ( ~ ~lc ) erf(-~)) cos c

Equation(46) defines a family of curves in the (~, 0) plane for various

values of the Reynoldsnumber.Intersections between(46) and the zeros
of (45) locate critical points in the entrainmentdiagramof the solution
(40). Fromthe abovediscussion, :it is clear that, in spite of the ap-
parently trivial Reynolds-number dependenceof the streamline pattern
of (40), the entrainment diagram mayexhibit a Reynolds-numberde-
pendence that is quite complex. Figure 19 shows the entrainment
diagramof (40) at three values of the Reynoldsnumber.For sufficiently
small Reynolds number, pathlines converge to a single stable node
whichlies on the axis of the jet. At a Reynoldsnumberof 6.7806 the
pattern bifurcates to a saddlelying ,on the axis of the jet, plus two stable
nodes lying symmetrically to either side of the axis. At a Reynolds
numberof 10.09089 the pattern bifurcates a second time to form a
saddle and two stable loci. Twopoints should be madehere. Thefirst is
that the diagramsin Figure 19 depict the behavior of (40) at Reynolds
numbersthat lie outside of its region of validity, although one may
expect the nonlinear solution to behavein a similar fashion. Thesecond
point is that the rollup of particle trajectories depicted in Figure 19e
occursentirely withoutany local concentrationof vorticity.
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 509

It is of interest to speculate on the high-Reynolds-number(non-

axisymmetric) flow from a point momentum source. In Section II we
saw a model of the plane mixing layer by Corcos (1979) in which the
flow was treated as deterministic at all scales. For the mixinglaye{,
dimensionalanalysis leads to 8~t, X-.~t~/2, ~l,.~t 1/4. The disparity be-
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

tweenlarge and small scales increases with time. It is not obviousthat

in the face of such a dynamicthe small scales can retain their identity
for any significant length of time. However,in the case of the three-
dimensionaljet all three scales vary like t t/2. Theflow is self-similar at
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

all scales. This suggests that there oughtto exist a high-Reynolds-number

solution whichis enormouslycomplex(with critical points in a three-
dimensional phase space) but wholly deterministic and which would
mimicall of the important features (spreading rate, entrainment rate,
dissipationat smallscales, etc) of a fully turbulentjet.
Entrainmentdiagrams can be workedout for a wide variety of flows
including those listed in Table 2. Figure 20 illustrates three examples.
Figure 20c shows the entrainment diagram for the Oseen vortex, an
explicitly unsteady flow in the variable r/V~. Figures 20a and b show
entrainment diagrams derived from the stationary mean-velocity pro-
files of the plane turbulent mixinglayer and plane turbulent jet (Cant-
well 1979). Bothof these flows can be formulatedas self-similar in time.
Normallythey are measuredby a laboratory observer whotakes a long
time averageat a fixed position in physical coordinates(x, y) with the
meanprofiles illustrated in Figure 20, the empirically measuredresult.
However,it is clear that another kind of averageis possible. Operation-
ally this wouldbe a long time average referred to a receding observer
wholooks at the flow quite literally through the zoomlens of a camera.
Therate of zoomis adjusted to matchthe global parameterthat governs
the motionand the averaging time of the experimentis limited by the
physical size of the apparatusthat contains the flow. Fluctuationsin the
evolving flow are assumedto follow the sametime scale as the coherent
motion and are averaged out by the receding observer. Given the
complexity of the entrainment diagram for the round jet, one may
conjecture that, in general, entrainmentdiagramsbased on through-the-
zoom-lensaveraging will exhibit a rich structure not found in the
averagereferred to a fixed laboratory observer.
The entrainment diagram has several useful features. It gives a
compact,invariant, visual impression of the flow pattern with easily
accessible information about the motionof fluid particles. The domi-
nant physical picture of eddies as spirals comesthrough in the well-
defined form of stable loci. The entrainment diagram can be used to
analyze the dynamics of fluid motion, revealing, in some cases, a
 Annual Reviews
www.annualreviews.org/aronline

510 CANTWELL

dependence on Reynolds number tliat may remain hidden in a pattern

of streamlines. The correspondence between the near-wall structure of
the entrainment diagram for the spot (Figure 1 lc), and the space-time
correlation results of Wygnanski et al (Figure 1 lb) suggests that both
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

methods somehow focus on the same energetic motions in the flow. The
prospect that space-time correlation measurements could be connected
to flow variables through critical points in the entrainment diagram is
an intriguing possibility that needs further study. Perhaps some hint for
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

approaching this issue can be found in the vector field of correlations

introduced by Willmarth & Wooldridge (1963).

VI CONCLUDINGREMAIRKS
Our understanding of the physics of turbulent motion has increased
tremendously in recent years. The major new fact is that turbulence is
characterized by a remarkable degree of order. But along with order has
come new complication for it appears that many turbulent flows retain
a permanent imprint of their infinitely variable initial state. Approxi-
mately twenty years have passed since the earliest observations of
organized structure. Yet progress in incorporating this structure into
practical engineering methods has been slow and the connection to a
truly predictive theory has not yet been made. Turbulence remains a
major unsolved problem of classical physics.

ACKNOWLEDGMENT

I would like to express thanks to Garry Brown, Luis Bernal, and John
Konrad who were involved in varic,us aspects of making the film used
for the animation of vortex pairing. I would also like to acknowledge
the support of NASA Ames Research Center under NASA Grant
NSG2392.

Literature Cited
Abernathy, F. H., Kronauer, R. E. 1962. waves. AIAAJ. 9:1657-59
Theformationof vortex streets. J. Fluid Birch, S. 1980. Evaluation of Shear-Layer
Mech. 13:1-20 Data--Rep.to the OrganizingCommittee,
Ashurst, W.T. 1977. Numericalsimulation 1980-81 Conf. on Computationof Com-
of turbulent mixinglayers via vortex dy- plex Turbulent Flows. Stanford Univ.
namics. Sandia Lab. Rep. SAND77-8612 Blaekwelder,R. F. 1978. Thebursting pro-
Bakewell,H. P., Lumley,J. U 1967. Viscous cess in turbulent boundarylayers. Lehigh
sublayer and adjacent region in turbu- Workshop on Coherent Structure in
lent pipe flow. Phys. Fluids 10:1880-89 Turbulent BoundaryLayers. ed. C. R.
Batt, R. G. 1975. Somemeasurementson Smith, D. E. Abbott, pp. 211-27
the effect of tripping the two-dimensional Blaekwelder, R. F., Eekelmann,H. 1979.
shear layer. AIAAJ. 13:245-47 Streamwisevortices associated with the
Bevilaqua, P. M., Lykoudis, P. S. 1971. bursting phenomenon.J. Fluid Mech.
Mechanismof entrainment in turbulent 94:577-94
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 511

Blackwelder, R. F., Kaplan, R. E. 1971. Coles, D. E. 1956. The law of the wakein
Intermittent structures in turbulent the turbulent boundarylayer. J. Fluid
boundarylayers. A GARD Conf. Proc. No. Mech, 1:191-226
43, London,5-1/5-7 Coles, D. E. 1978. A modelfor the flow in
Blackwelder, R. F., Kovasznay,L. S. G. the viscous sublayer. Lehigh Workshop
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

1972. Timescales and correlations in a on Coherent Structure in Turbulent

turbulent boundarylayer. Phys. Fluids BoundaryLayers, ed. C. R. Smith, D. E.
15:1545-54 Abbott, pp. 462-75
Bouehard, G. E., Reynolds, W. C. 1978. Coles, D. E., Barker, S. J. 1975. Somere-
Control of vortex pairing in a roundjet. markson a synthetic turbulent boundary
Bull. Am.Phys.Soc. 23(8): 1013 layer. In TurbulentMixingin Nonreactive
Boussinesq,J. 1877. Th~oriede l’~coule- and Reactive Flows, ed. S. N. B. Murthy,
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

merit tourbillant. M~ra.Pr~s. Acad.Sci. pp. 285-92

23:46 Coles, D. E., VanAtta, C. W.1967. Digital
Briedenthal, R. E. 1978, A chemically re- experiment in spiral turbulence. Phys.
acting turbulent shear layer. PhDthesis. Fluids Suppl. Part H 10:S120-21
Calif. Inst. Tech., Pasadena,Calif. Corcos, G. 1979. The mixing layer: de-
Brown,G. L., Roshko,A. 1974. Ondensity terministic modelsof a turbulent flow.
effects and large structure in turbulent Dept. Mech. Engrg. Rep. No. FM-79-2.
mixinglayers. J. Fluid Mech.64: 775- 816 Univ. California, Berkeley
Brown, G. L., Thomas, A. S. W. 1977. Corino, E. R., Brodkey,R. S. 1969. A vis-
Largestructure in a turbulent boundary ual investigation of the wall region in
layer. Phys.Fluids 20(10):S243-52 turbulent flow. J. Fluid Mech.37:1-30
C.antwell,B. J. 1978.Similaritytransforma- Corrsin, S. 1943.Investigations of flow in
tions for the two-dimensionalunsteady an axially symmetricheated jet of air.
stream function equation. J. Fluid Mech. NACAAdv. Conf. Rep. 3123
85:257-71 Corrsin, S., Kistler, A. 1954. The free
Cantwell, B. J. 1979. Coherent turbulent stream boundaries of turbulent flows.
structures as critical points in unsteady NACATech. Note No. 3133 (also NACA
flow. Arch. Mech.Stosow. 31:707-21 Tech. Rep. No. 1244, 1955)
Cantwell, B. J. 1980. Transition in the Crow, S. C., Champagne, F. H. 1971.
axisymmetri¢ jet. (SUDAAR)Rep, No. Orderlystructure in jet turbulence, d.
521, Stanford Univ., Dept. Aeronaut.and Fluid Mech.48:547-91.
Astronaut.J. Fluid Mech.In press Davies, M. E. 1976. A comparisonof the
C.antwell,B. J., Coles,D. E., Dimotakls,P. wakestructure of a stationary and oscil-
E. 1978. Structure and entrainment in lating bluff body, using a conditional
the plane of symmetryof a turbulent averaging technique. J. Fluid Mech.
spot. J. Fluid. Mech.87:641-72 75:209-23
Chandraseldaar, S. 1949. On Heisenberg’s Dimotakis, P. E., Brown,G. L. 1976. The
elementary theory of turbulence. Proc. mixing layer at high Reynoldsnumber:
R. Soc. LondonSet. A. 200:20-33 large structure dynamicsand entrain-
Chandrasuda,C., Mehta, R. D., Weir, A. ment, J. Fluid Mech.78:535-60
D., Bradshaw,P. 1978. Effect of free Dinkelaeker, A., Hessel, M., Meier, G.,
shear stream turbulence on large struc- Sehewe,G. 1977. Investigation of pres-
ture in turbulent mixinglayers. J. Fluid sure fluctuations beneath a turbulent
Mech. 85:693-704 boundarylayer by meansof an optical
Chapman, D. R. 1979. Computational method.Phys. Fluids 20(10):$216-24
aerodynamicsdevelopmentand outlook. Elder, J. W.1960. Anexperimentalinvesti-
AIAADrydenLect. 79-0129 gation of turbulent spots and breakdown
Chorin,A. J., Bernard,P. S. 1973.Discreti- to turbulence. J. Fluid Mech.9:235-46
zation of a vortex sheet with an example Emmons, H. W. 1951. The laminar-
of roll-up. J. Comput.Phys. 13:423-29 turbulent transition in a boundarylayer.
Clark, R., Ferziger, J. H., Reynolds,W.C. Part I. J. Aeronaut.Sci. 18:490-98
1979. Evaluationof subgrid-scale models Falco, R. E. 1977. Coherentmotionsin the
using an aeeurately simulated turbulent outer region of turbulent boundary
flow, J. Fluid Mech.91:1-16 layers. Phys. Fluids 20(10):5124-32
Clements,R. R. 1973. Aninviseid modelof Favre, A. J., Gaviglio, J. J., Dumas,R.
two-dimensional vortex shedding. J. 1957. Space-timedouble correlation and
Fluid Mech. 57:321-36 spectra in a turbulent boundarylayer. J.
 Annual Reviews
www.annualreviews.org/aronline

512 CANTWELL

Fluid Mech.2:313-41 Hussain, A. K. M. F., Zaman,K. B. M. Q.

Freymuth,P. 1966. Ontransition in a sep- 1978. The free shear layer tone phenom-
arated boundary layer. J. Fluid Mech. enon and probe interference. J. Fluid
25:683-704 Mech. 87:349-84
Furuya, Y., Osaka, H. 1976. Three- Kim,H. T., Kline S. J., Reynolds,W.1971.
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

dimensionalstructure of nominallytwo- The production of the wall region in

dimensionalturbulentboundarylayer. Pre- turbulent flow. J. Fluid Mech.50:133-60
sented at the 14th IUTAMCongress, Klebanoff, P. S. 1954. Characteristics of
Delft, Abstr. No. 160 turbulence in a boundarylayer with zero
Gad-el-Hak,M., Blackwelder,R. F. 1979. pressure gradient. NACATech. Note No.
A visual study of the growth and en- 3178
trainment of turbulent spots. Proc, Kline, S. J. 1978.Therole of visualization
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

IUTAM Syrup. on Transition, Stuttgart, in the study of the structure of the turbu-
Germany lent boundarylayer. LehighWorkshopon
Grant, H. L. 1958. The large eddies of CoherentStructure of TurbulentBoundary
turbulent motion. J. Fluid Mech.4:149- Layers, ed. C. R. Smith, D. E. Abbott,
90 pp. 1-26
Grass, A. J. 1971. Structural features of lrdine, S. J., Falco, R. E. 1980. Summary of
turbulent flow over smooth and rough the AFOSR/MSU research specialists
boundaries. J. Fluid Mech.50:223-56 workshopon coherent structure in turbu-
Griffin, O, M,, Ramberg,S. E. 1974. The lent boundary layers. AFOSF-TR-80-
vortex street wakes of vibrating cyl- 0290 ADA083717
inders. J. Fluid Mech.66:553-78 Kline, S. J., Runstadler, P. W.1959. Some
Gupta, A. K., Laufer, J., Kaplan, R. E. prelimi_naryresults of visual studies of
1971. Spatial structure ha the viscous the flow modelof the wall layers of the
sublayer. J. Fluid Mech.50:493-512 turbulent boundary layer. Trans. ASME
Hama,F. R., Nutant, J. 1963. Detailed (Ser. E) 2:166-70
flow field observationsin the transition Kline, S. J., Reynolds,W.C., Schraub,F.
process in a thick boundarylayer. Proc. A., Runstadler, P. W.1967. The struc-
1963HeatTransferand Fluid Mech.Inst., ture of turbulent boundarylayers. J.
pp. 77-94 Fluid Mech. 30:741-73
Hanratty, T. J., Chorn, L. G., Hatziav- Kolmogorov,A. N. 1941. The local struc-
ramidis, D. T. 1977. Turbulent fluctua- ture of turbulencein incompressibleflow
tions in the viscous wall region for New- for very large Reynolds number.C. R.
tonian and drag reducing fluids. Phys. Acad. Sci. U.R.S.S. 30:301
Fluids 20(10):S112-19 Kovasznay, L. S. G. 1948. Spectrum of
Head, M. R., Bandyopadhyay, P. 1978. locally isopropic turbulence.J. Aeronaut.
Combinedflow visualization and hot Sci, 15:745-53
wire measurementsin turbulent boun- Kovasznay, L. S. G., Kibens, V., Black-
dary layers. Lehigh Workshopon Coher- welder, R. 1970. Large scale motion in
ent Structure in Turbulent Boundary the intermittent region of a turbulent
Layers, ed. C. R. Smith, D. E. Abbott, boundarylayer. J. Fluid Mech. 41:283-
pp. 98-129 325
Heisenberg, W. 1948a. Zur statistischen Laufer, J. 1954.Thestructure of turbulence
theorie der turbulenz. Z. Phys. 124:628- in fully developedpipe flow. NACATech.
57 Note No. 2954
Heisenberg, W. 1948b. On the theory of Laufer, J., Narayanan,M.A. B. 1971. Mean
statistical andisotropie turbulence,Proc. period of the turbulent productionmech-
R. Soc. LondonSer. A 195:402-6 anismin a boundarylayer. Phys. Fluids
Hinze, J. O. 1959. Turbulence. NewYork: 14:182-83
McGraw-Hill.1st ed. Lee, M. K., Eekelmann,L. D., Hanratty,
Hunt,J. C. R., Abell, C. J., Peterka,J. A., T. J. 1974. Identification of turbulent
Woo,H. 1978. Kinematicstudies of the wall eddies throughthe phaserelation of
flows around free or surface mounted the componentsof the fluctuating veloc-
obstacles; applyingtopologyto flow vis- ity gradient. J. Fluid Mech.66:17-34
ualization. J. Fluid Mech.86:179-200 Leonard, A. 1979. Vortex simulation of
Hussain, A. K. M. F., Reynolds, W. C. three-dimensional,spotlike disturbances
1975. Measurementsin fully developed in a laminar boundarylayer. NASATech.
turbulent channel flow. J. Fluid Engrg. Memo.78579 (also in Turbulent shear
97:568-80 flows H, ed. L. J. S. Bradbury,pp. Berlin:
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 513

Springer 67:77 Fluids 20(10): S133-44

Liepmann, H. W. 1952. Aspects of the Onsager,L. 1945. The distribution of en-
turbulence problem. J. Appl. Math. and ergb’ in turbulence. Phys. Rev. Abstr.
Phys. (ZAMP)3:321-426 68:286
Liepmann, H. W. 1962. Free turbulent Onsager, L. 1949. Statistical hydrody-
flows. In Colloques Internationaux du namics. NuovoCimentoSuppl. 6:279-87
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

Centre National de la RechercheScien- Oster, D., Wygnanski,I., Dziomba,B.,

tifique (C.N.R.S.), Mechniquede Fiedler, H. 1977.Onthe effect of initial
Turbulence,Marseilles, 108:211-27 conditions on the two dimensional
Liepmann,H. W.Laufer, J. 1947. Investi- turbulent mixinglayer. In Lecture Notes
gation of free turbulent mixing. NACA in Physics,Vol. 75, p. 48. Berlin: Springer
Tech. Note No, 1257 (Structure and Mechanisms of Turbulence
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

Lighthill, M.J. 1963. In LaminarBoundary I, ed. H. Fiedle0

Layers, ed. L. Rosenhead, pp. 48-88. Oswatitsch, K. 1958. Die Abltsungsbedin-
OxfordUniv. Press gung yon Grenzschichten, Grenzschicht
Lindgren,E. R. 1969. Propagationvelocity Forschung,ed. H. Goertler, pp. 357-67.
of turbulent slugs and streaks in transi- Berlin/New York: Springer
tion pipe flow. Phys.Fluids 12:418-25 Owen,F. K., Johnson, D. A. 1978. Mea-
Lu, S. S., Willmarth,W.W.1973. Measure- surementsof unsteadyvortex flow fields.
mentsof the Reynoldsstress in a turbu- AIAA Paper 78-18
lent boundary layer. J. Fluid Mech. Payne, F. R., Lumley,J. L. 1967. Large
60:481-511 eddy structure of the turbulent wakebe-
Lugt, H. J. 1979. The dilemmaof defining hind a circular cylinder. Phys. Fluids
a vortex. In Recent Developments in The- Suppl. 10:S194-96
oretical and Experimental Fluid Mecha- Peak¢, D., Tobak, M. 1980. Three-
nics, ed. U. Muller, K. G. Roesner, B. dimensional interactions and vortical
Schmidt, pp. 309-21. Berlin/Heidel- flows with emphasis on high speeds.
berg/New York: Springer NASATech. Memo. 81169
Meijer, M. C. 1965. Pressure measurements Perry, A. E., Lira, T. T. 1978. Coherent
on flapped hydrofoils in cavity flows and structures in coflowingjets andwakes.J.
wake flows. Hydrodyn. Lab. Rep. No. Fluid Mech. 88:451-64
E-133.2.Calif. Inst. Tech. Perry, A. E., Lira T. T. 1978. Coherent
Mitchner, M. 1954. Propagation of turbu- structures in coflowingjets andwakes.J.
lence from an instantaneous point dis- Fluid Mech. 88:451-64
turbance. J. Aeronaut.Sci. 2(5):350-51 Perry, A. E., Watmuff, J. H. 1979. The
Moin,P., Reynolds,W.C., Fe~7Jger,J. H. phase-averagedlarge-scale structures in
1978. Large eddy simulation of incom- three-dimensionalturbulent wakes.Dept.
pressible turbulent channel flow. Dept. Mech. Rep. FM-12.Univ. Melbourne
Mech. Engrg. Rep. 91:1-16. Stanford Perry, A. E., Lira, T. T., Chong,M.S., Teh,
Univ. E. W. 1980. The fabric of turbulence.
Narayanan, B., Marvin, J. 1978. On the A1AAPap. 80-1358
period of the coherent structure in Perry, A. E., Lira, T. T., Chong,M. S.
boundarylayers at large Reynoldsnum- 1980. Instantaneous vector fields in
bers. Lehigh Workshop on Coherent coflowingjets and wakes.J. Fluid Mech.
Structure in Turbulent BoundaryLayers, In press
ed. C. R. Smith, D. E. Abbot,pp. 380-86 Prandtl, L. 1925. Bericht uber un-
Nychas,S. G., Hershey,H. C., Brodkey,R. tersuchungen zur ausgebildeten turbu-
S. 1973. Avisual study of turbulent shear lenz. Z. Angew.Math. Mech.5:136-39
flow. J. Fluid Mech.61:513-40 Praturi, A. K., Brodkey, R. S. 1978. A
Often, G. R., Kline, S. J. 1974. Combined stereoscopic visual study of coherent
dye-streak andhydrogen-bubblevisual structuresin turbulentshearflow.J. Fluid
observations of a turbulent boundary Mech. 89:251-72
layer. J. Fluid Mech.62:223-39 Pullin, D. I. 1978.Thelarge-scale structure
Often, G. R., Kline, S. J. 1975. A proposed of unsteadyself-similar rolled-up vortex
modelof the bursting process in turbu- sheets. J. Fluid Mech.80:401-30
lent boundary layers. J. Fluid Mech. Rao, K. N., Narasimha,R., Narayanan,M.
70:209-28 A. B. 1971. Bursting in a turbulent
Oldaker, D. K., Tiederman, W. G. 1977. boundarylayer. J. Fluid Mech. 48:339-
Spatial structure of the viscous sublayer 52
in drag reducing channel flows. Phys. Reynolds, O. 1895. Onthe dynamicalthe-
 Annual Reviews
www.annualreviews.org/aronline

514 CANTWELL

ory of incompressibleviscous fluids and Braunschweig, Veiweg& Solm

the determinationof the criterion. Philos. Townsend, A. A. 1947. Measurements in
Trans. R. Soc. LondonSer. A 186:123 the turbulent wakeof a cylinder. Proc.
Reynolds, W. C. 1976. Computation of R. Soc. LondonSer. A 190:551-61
turbulent flows. Ann. Rev. Fluid Mech. Townsend,A. A. 1956. The Structure of
8:183-208 Turbulent Shear Flow. CambridgeUniv.
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

Roshko,A. 1976. Structure of turbulent Press. 1st ed.

shear flows: a new look. Dryden Re- Townsend, A. A. 1970. Entrainment and
search Lecture. AIAAJ. 14:1349-57 the structure of turbulent flow. J. Fluid
Sabot, J., Saleh, I., Comte-Bellot,G. 1977. Mech. 41:13-46
Effects of roughnesson the intermittent Townsend,A. A. 1976. The Structure of
maintenanceof Reynoldsshear stress in Turbulent Shear Flow. CambridgeUniv.
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org

pipe flow. Phys. Fluids 20(10):S150-55 Press. 2nd ed.

Saffman,P. G., Baker, G. R. 1979. Vortex Tu, B. J., Willmarth, W.W.1966. Anex-
interactions. Ann. Rev. Fluid Mech. perimental study of turbulence near the
11:95-122 wall through correlation measurements
Sarpkaya, T. 1975. Aninviseid model of in a thick turbulent boundarylayer. Tedt
two-dimensional vortex shedding for Rep. No. 02920-3-T. Dept. Aerosp. En-
transient and asymptotically steady sep- grg., Univ. Michigan, AnnArbor
arated flow overan inclinedflat plate. J. Turner, J. S. 1964. The flow into an ex-
Fluid..Mech. 68:109-28 pandingspherical vortex. J. Fluid Mech.
Savas, O. 1979. Somemeasurementsin syn- 18:195-208
thetic turbulent boundarylayers. PhDthe- VonWeizsiicker, C. F. 1948. Das spectrum
sis. Calif. Inst. Tech., Pasadena,Calif. der turbulenz bei grossen Reynolds-
~ehubauer, G. B., Klebanoff, P. S. 1955. Schenzahlen. Z. Phys. 124:614-27
Contributions on the mechanics of Willmarth, W. W.1975. Pressure fluctua-
boundarylayer transition. NACATech. tions beneathturbulent boundarylayers.
Note No. 3489 (See also NACATeclt Ann. Rev. Fluid Mech.7:13-37
Rep. No. 1289, 1956) Willmarth,W.W., Bogar,T. J. 1977. Survey
Smith, C. R. 1978. Visualization of turbu- and new measurements of turbulent
lent boundarylayer structure using a structure near the wall. Phys. Fluids
moving hydrogen bubble wire probe. 20(10):$9-21
Lehigh Workshopon Coherent Structure Willmarth,W.W., Lu, S. S. 1972. Structure
in Turbulent BoundaryLayers, ed. C. R. of the Reynoldsstress near the wall. J.
Smith, D. E. Abbott, pp. 48-97 Fluid Mech. 55:65-69
Sozou, C. 1979. Developmentof the flow Willmarth, W.W., Wooldridge,C. E. 1962.
field of a point force in an infinite fluid. Measurementsof the fluctuating pres-
J. Fluid Mech.91:544-54 sure at the wall beneatha thick turbulent
Spalding, D. B. 1961. A single formula for boundarylayer, J. Fluid Mech. 14:187-
the law of the wall. J. Appl. Mech. 210
28:455-57 Willmarth, W.W., Wooldridge,C. E. 1963.
Spencer,B. W.,Jones, B. (3. 1971.Statisti- Measurements of the correlation be-
cal investigation of pressureandvelocity tween the fluctuating velocities and the
fields in the turbulent two-streammixing fluctuating wall pressure in a thick
layer. AIAAPap. No. 71-613 turbulent boundary layer. AGARD Rep.
Taylor, G. I. 1915. Eddy motions in the 456
atmosphere. Philos. Trans. R. Soc. Willmarth, W.W., Yang, C. S. 1970. Wall
LondonSet. A 215:1-26 pressure fluctuations beneath turbulent
Taylor, G. I. 1932. Noteon the distribution boundary layers on a flat plate and a
of turbulent velocities in a fluid near a cylinder. J. Fluid Mech.41:47-80
solid wall. Proc. R. Soc. LondonSer. A Winant,C. D., Browand,F. K. 1974. Vortex
135:678-84 pairing, the mechanismof turbulent
Taylor,G. I. 1935.Thestatistical theory of mixing layer growth at moderate Rey-
turbulence, Parts I-IV, Proc. R. Soc. nolds number.J. Fluid Mech.63:237-55
LondonSer. A 151:421-511 Wygnanski,I., Fiedler, H. E. 1970. The
Theodorsen, T. 1955. The structure of two dimensionalmixing region, J. Fluid
turbulence. In 50 Jahre Grenzschichtfor- Mech. 41:327-61
sung, ed. H. G~rtier, W.Tollmein,p. 55. Wygnanski,I. J., Champagne,F. H. 1973.
 Annual Reviews
www.annualreviews.org/aronline

TURBULENT FLOW 515

Ontransition in a pipe. Part I. Theorigin Zakkay, V., Barra, V., Wang,C. R. 1978.
of puffs and slugs and the flow in a Coherentstructure of turbulenceat high
turbulent slug. J. Fluid Mech.59:281-335 subsonic speeds. Lehigh Workshopon
Wygnansld,I. L, Sokolov, N., Friedman, CoherentStructure in TurbulentBoundary
D. 1976. On the turbulent "spot" in a Layers, ed. C. R. Smith, P. E. Abbott,
boundarylayer undergoingtransition, d. pp. 387-95
Access provided by Stanford University - Main Campus - Robert Crown Law Library on 02/12/18. For personal use only.

Fluid Mech. 78:785-819 Zilberman, M., Wygnansld,I., Kaplan, R.

Wygnanski,I., Haritonidis, I. H., Kaplan, E. 1977. Transitional boundarylayer spot
R. E. 1979. On a Tollmien-Schlichting in a fully turbulent environment.Phys.
wave packet produced by a turbulent Fluids 20(10):$258-71
spot. J. Fluid Mech.92:505-28
Annu. Rev. Fluid Mech. 1981.13:457-515. Downloaded from www.annualreviews.org