INTRODUCTION

Most flows occurring in nature and in engineering applications are turbulent.

The boundary layer in the earths atmosphere is turbulent (except possibly in
very stable conditions); jet streams in the upper troposphere are turbulent;
cumulus clouds are in turbulent motion. The water currents below the surface
of the oceans are turbulent,; the Gulf Stream is a turbulent wall-jet kind of
flow. The photosphere of the sun and the photospheresof similar stars are in
turbulent motion; interstellar gas clouds (gaseous nebulae) are turbulent; the
wake of the earth in the solar wind is presumably a turbulent wake. Boundary
layers growing on aircraft wings are turbulent. Most combustion processes
involve turbulence and often even depend on it; the flow of natural gas and
oil in pipelines is turbulent. Chemical engineers use turbulence to mix and
homogenize fluid mixtures and to accelerate chemical reaction rates in liquids
or gases. The flow of water in rivers and canals is turbulent; the wakes of
ships, cars, submarines, and aircraft are in turbulent motion. The study
of turbulence clearly is an interdisciplinary activity, which has a very
wide range of applications. In fluid dynamics laminar flow is the exception,
not the rule: one must have small dimensions and high viscosities to
encounter laminar flow. The flow of lubricating oil in a bearing is a typical
example.
Many turbulent flows can be observed easily; watching cumulus clouds or
the plume of a smokestack is not time wasted for a student of turbulence.
In the classroom, some of the films produced by the National Committee
for Fluid Dynamics Films (for example, Stewart, 1969) may be used to
advantage.
1.I
The nature of turbulence
Everyone who, at one time or another, has observed the efflux from a smoke
stack has some idea about the nature of turbulent flow. However, it is very
difficult to give a precise definition of turbulence. All one can do is list
some of the characteristics of turbulent flows.
Irregularity One characteristic is the irregularity, or randomness, of all
turbulent flows. This makes a deterministic approach to turbulence problems
impossible; instead, one relies on statistical methods.

Introduction

Diffusivity The diffusivity of turbulence, which causes rapid mixing and

increased rates of momentum, heat, and mass transfer, i s another important
feature of all turbulent flows. If a flow pattern looks random but does not
exhibit spreading of velocity fluctuations through the surrounding fluid, it is
surely not turbulent. The contrails of a jet aircraft are a case in point: exclud'ling the turbulent region just behind the aircraft, the contrails have a very
nearly constant diameter for several miles. Such a flow i s not turbulent, even
though it was turbulent when it was generated. The diffusivity of turbulence is
the single most important feature as far as applications are concerned: it
prevents boundary-layer separation on airfoils a t large (but not too large) angles
of attack, it increases heat transfer rates in machinery of all kinds, it is the source
of the resistance of flow in pipelines, and it increases momentum transfer
between winds and Ocean currents.
Large Reynolds numbers Turbulent flows always occur a t high Reynolds
numbers. Turbulence often originates as an instability of laminar flows if the
Reynolds number becomes too large. The instabilities are related to the interaction of viscous terms and nonlinear inertia terms in the equations of motion. This interaction is very complex: the mathematics of nonlinear partial
differential equations has not been developed to a point where general solutions can be given. Randomness and nonlinearity combine to make the equations of turbulence nearly intractable; turbulence theory suffers from the
absence of sufficiently powerful mathematical methods. This lack of tools
makes all theoretical approaches to problems in turbulence trial-and-error
affairs. Nonlinear concepts and mathematical tools have to be developed
along the way; one cannot rely on the equations alone to obtain answers to
problems. This situation makes turbulence research both frustrating and
challenging: it is one of the principal unsolved problems in physics today.
Three-dimensional vorticity fluctuations Turbulence is rotational and three
dimensional. Turbulence is characterized by high levels of fluctuating vorticity. For this reason, vorticity dynamics plays an essential role in the description of turbulent flows. The random vorticity fluctuations that characterize turbulence could not maintain themselves if the velocity fluctuations
were two dimensional, since an important vorticity-maintenancemechanism
known as vortex stretching is absent in two-dimensional flow. Flows that are
substantially two dimensional, such as the cyclones in the atmosphere which

1.I The nature of turbulence

determinethe weather, are not turbulence themselves, even though their characteristics may be influenced strongly by small-scale turbulence (generated
somewhere by shear or buoyancy), which interacts with the large-scaleflow.
In summary, turbulent flows always exhibit high levels of fluctuating vor-,
ticity. For example, random waves on the surface of oceans are not in turbulent motion since they are essentially irrotational.
Dissipation Turbulent flows are always dissipative. Viscous shear stresses
perform deformation work which increasesthe internal energy of the fluid a t
the expense of kinetic energy of the turbulence. Turbulence needs a continuous supply of energy to make up for these viscous losses. If no energy is
supplied, turbulence decays rapidly. Random motions, such as gravity waves
in planetary atmospheres and random sound waves (acoustic noise), have
insignificant viscous losses and, therefore, are not turbulent. In other words,
the major distinction between random waves and turbulence is that waves are
essentially nondissipative (though they often are dispersive), while turbulence
is essentially dissipative.
Continuum Turbulence is a continuum phenomenon, governed by the equations of fluid mechanics. Even the smallest scales occurring in a turbulent
flow are ordinarily far larger than any molecular length scale. We return to
this point in Section 1.5.
Turbulent flows are flows Turbulence is not a feature of fluids but of fluid
flows. Most of the dynamics of turbulence is the same in all fluids, whether
they are liquids or gases, if the Reynolds number of the turbulence is large
enough; the major characteristics of turbulent flows are not controlled by the
molecular properties of the fluid in which the turbulence occurs. Since the
equations of motion are nonlinear, each individual flow pattern has certain
unique characteristics that are associated with i t s initial and boundary conditions. No general solution to the Navier-Stokes equations is known; consequently, no general solutions to problems in turbulent flow are available.
Since every flow is different, it follows that every turbulent flow i s different,
wen though all turbulent flows have many characteristics in common.
Students of turbulence, of course, disregard the uniqueness of any particular
turbulent flow and concentrate on the discovery and formulation of laws that
describe entire classes or families of turbulent flows.

Introduction

The characteristics of turbulence depend on i t s environment. Because of

this, turbulence theory does not attempt to deal with all kinds and types of
flows in a general way. Instead, theoreticians concentrate on families of flows
with fairly simple boundary conditions, like boundary layers, jets, and wakes.

1.2
Methods of analysis
Turbulent flows have been investigated for more than a century, but, as was
remarked earlier, no general approach to the solution of problems in turbulence exists. The equations of motion have been analyzed in great detail,
but it is still next to impossible to make accurate quantitative predictions
without relying heavily on empirical data. Statistical studies of the equations
of motion always lead to a situation in which there are more unknowns than
equations. This is called the closure problem of turbulence theory: one has to
make (very often ad hoc) assumptions to make the number of equations
equal to the number of unknowns. Efforts t o construct viable formal perturbation schemes have not been very successful so far. The success of attempts
to solve problems in turbulence depends strongly on the inspiration involved
in making the crucial assumption.
This book has been designed to get this point across. In turbulence, the
equations do not give the entire story. One must be willing to use (and
capable of using) simple physical concepts based on experience t o bridge the
gap between the equations and actual flows. We do not want to imply that
the equations are of little use; we merely want to make it unmistakably clear
that turbulence needs spirited inventors just as badly as dedicated analysts.
We recognize that this i s a very specific, and possibly biased, point of view. It
is possible that a t some time in the future, someone will succeed in developing
a completely formal theory of turbulence. However, we believe that there is a
far better chance of developing a physical model of turbulence in the spirit of
the Rutherford model of the atom. The model need not be complete, but it
would be very useful. The real challenge, it seems to us, is that no adequate
model of turbulence exists today.
Turbulence theory is limited in the same way that general fluid dynamics
would be if the Stokes relation between stress and rate of strain in Newtonian
fluids were unknown. This illustration is not arbitrary: one approach to turbulence theory is to postulate a relation between stress and rate of strain that
involves a turbulence-generated "viscosity," which then supposedly plays a

1.2 Methods of analysis

role similar to that of molecular viscosity in laminar flows. This approach is

based on a superficial resemblance between the way molecular motions transfer momentum and heat and the way in which turbulent velocity fluctuations
transfer these quantities. Phenomenological concepts like "eddy viscosity"
(to replace molecular viscosity) and "mixing length" (in analogy with the
mean free path in the kinetic theory of gases) were developed by Taylor,
Prandtl, and others. These coqcepts are studied in detail in Chapter 2.
Molecular viscosity is a property of fluids; turbulence is a characteristic of
flows. Therefore, the use of an eddy viscosity to represent the effects of
turbulence on a flow is liable to be misleading. However, current research
seems to indicate that, in simple flows, we may, for analytical reasons, speak
of a turbulent fluid rather than of a turbulent flow. Turbulent "fluids,"
however, are non-Newtonian: they exhibit viscoelasticity and suffer memory
effects. In favorable circumstances, the memory is fading in time, so that one
may be able to develop a semilocal theory relating the mean stress to the
mean rate of strain.
Phenomenological theories of turbulence make crucial assumptions a t a
fairly early stage in the analysis. In recent years, a group of theoreticians
(Kraichnan, Edwards, Orszag, Meecham, and others) have developed very
formal and sophisticated statistical theories of turbulence, in the hope of
finding a formalism that does not need ad hoc assumptions (see Orszag,
1971). So far, however, rather arbitrary postulates are needed in these
theories, too. The mathematical complexity of this work is so overwhelming
that a discussion of it has to be left out of this book.
Dimensional analysis One of the most powerful tools in the study of turbulent flows is dimensional analysis. In many circumstances it is possible to
argue that some aspect of the structure of turbulence depends only on a few
independent variables or parameters. If such a situation prevails, dimensional
methods often dictate the relation between the dependent and independent
variables, which results in a solution that is known except for a numerical
coefficient. The outstanding example of this is the form of the spectrum of
turbulent kinetic energy in what is called the "inertial subrange."
Asymptotic invariance Another frequently used approach is to exploit some
of the asymptotic properties of turbulent flows. Turbulent flows are characterized by very high Reynolds numbers; it seems reasonable to require that

Introduction

any proposed descriptions of turbulence should behave properly in the limit as

the Reynolds number approaches infinity. This is often a very powerful constraint, which makes fairly specific results possible. The development of the
theory of turbulent boundary layers (Chapter 5) is a case in point. The limit
process involved in an asymptotic approach is related to vanishingly small
effects of the molecular viscosity. Turbulent flows tend to be almost independent of the viscosity (with the exception of the very smallest scales of motion); the asymptotic behavior leads to such concepts as "Reynolds-number
similarity" (asymptotic invariance).
Local invariance Associated with, but distinct from, asymptotic invariance is
the concept of "self-preservation" or local invariance. In simple flow geometries, the characteristics of the turbulent motion a t some point in time and
space appear to be controlled mainly by the immediate environment. The
time and length scales of the flow may vary slowly downstream, but, if the
turbulence time scales are small enough to permit adjustment to the gradually
changing environment, it is often possible to assume that the turbulence is
dynamically similar everywhere if nondimensionalized with local length and
time scales. For example, the turbulence intensity in a wake is of order
6 aU/ay, where 6 is the local width of the wake and W / a y is the average
mean-velocity gradient across the wake.
Because turbulence consists of fairly large fluctuations governed by nonlinear equations, one may expect a behavior like that exhibited by simple
nonlinear systems with limit cycles. Such behavior should be largely independent of initial conditions; the characteristics of the limit cycle should depend
only on the dynamics of the system and the constraints imposed on it. I n the
same way, one expects that the structure of turbulence in a given class of
shear flows might be in some state of dynamical equilibrium in which local
inputs of energy should approximately balance local losses. If the energy
transfer mechanisms in turbulence are sufficiently rapid, so tha't effects of
past events do not dominate the dynamics, one may expect that this limitcycle type of equilibrium is governed mainly by local parameters such as scale
lengths and times. Simple dimensional methods and similarity arguments can
be very useful in this kind of situation. Because one may want to look for
local scaling laws (both in the spatial and the spectral domain), the problem
of finding appropriate length and time scales becomes an important one.
Indeed, scaling laws are a t the heart of turbulence research.

1.3 The origin of turbulence

1.3
The origin of turbulence
In flows which are originally laminar, turbulence arises from instabilities at
large Reynolds numbers. Laminar pipe flow becomes turbulent at a Reynolds
number (based on mean velocity and diameter) in the neighborhood of 2,000
unless great care is taken to avoid creating small disturbances that might
trigger transition from laminar to turbulent flow. Boundary layers in zero
pressure gradient become unstable a t a Reynolds number U6"lv = 600
approximately (6" i s the displacement thickness, U is the free-stream velocity, and v is the kinematic viscosity). Free shear flows, such as the flow in a
mixing layer, become unstable a t very low Reynolds numbers because of an
inviscid instability mechanism that does not operate in boundary-layerand
pipe flow. Early stages of transition can easily be seen in the smoke rising
from a cigarette.
On the other hand, turbulence cannot maintain itself but depends on its
environment to obtain energy. A common source of energy for turbulent
velocity fluctuations is shear in the mean flow; other sources, such as buoyancy, exist too. Turbulent flows are generally shear flows. If turbulence
arrives in an environmentwhere there is no shear or other maintenance mechanism, it decays: the Reynolds number decreases and the flow tends to
become laminar again. The classic example is turbulence produced by a grid
in uniform flow in a wind tunnel.
Another way to make a turbulent flow laminar or to prevent a laminar
flow from becoming turbulent is to provide for a mechanism that consumes
turbulent kinetic energy. This situation prevails in turbulent flows with
imposed magnetic fields a t low magnetic Reynolds numbers and in atmospheric flows with a stable density stratification, to cite two examples.
Mathematically, the details of transition from laminar t o turbulent flow
are rather poorly understood. Much of the theory of instabilities in laminar
flows is linearized theory, valid for very small disturbances; it cannot deal
with the large fluctuation levels in turbulent flow. On the other hand, almost
all of the theory of turbulent flow is asymptotic theory, fairly accurate at
very high Reynolds numbers but inaccurate and incomplete for Reynolds
numbers a t which the turbulence cannot maintain itself. A noteworthy exception is the theory of the late stage of decay of wind-tunnel turbulence
(Batchelor, 1953).
Experiments have shown that transition is commonly initiated by a pri-

Introduction

mary instability mechanism, which in simple cases is two dimensional. The

primary instability produces secondary motions, which are generally three
dimensional and become unstable themselves. A sequence of this nature gen
erates intense localized three-dimensional disturbances (turbulent "spots"),
which arise a t random positions a t random times. These spots grow rapidly
and merge with each other when they become large and numerous to form a
field of developed turbulent flow. In other cases, turbulence originates from
an instability that causes vortices which subsequently become unstable. Many
wake flows become turbulent in this way.

1.4
Diffusivity of turbulence
The outstanding characteristic of turbulent motion is its ability to transport
or mix momentum, kinetic energy, and contaminants such as heat, particles,
and moisture. The rates of transfer and mixing are several orders of magnitude greater than the rates due to molecular diffusion: the heat transfer and
combustion rates of turbulent combustion in an incinerator are orders of
magnitude larger than the corresponding rates in the laminar flame of a
candle.
Diffusion in a problem with an imposed length scale Contrasting laminar and
turbulent diffusion rates is a useful exercise not only for getting acquainted
with turbulence but also for recognizing the multifaceted role of the Reynolds number. Suppose one has a room (with a characteristic linear dimension
L ) in which a heating element (radiator) is installed. If there is no air motion
in the room, heat has to be distributed by molecular diffusion. This process is
governed by the diffusion equation (8 is the temperature; 7 is the thermal
diffusivity, assumed to be constant):
(1.4.1)

We are not looking for a specific solution of (1.4.1) with a given set of
boundary conditions. Instead, we want to discover the gross consequencesof
(1.4.1) with the simple tools of dimensional analysis. Dimensionally, (1.4.1)
may be interpreted as

A8

A8

-- r p '
Trn

(1.4.2)

1.4 Diffusivity of turbulence

where A6 is a characteristic temperature difference. From (1.4.21, we obtain

T,--,

LZ

(1.4.3)

which relates the time scale T,,, of the molecular diffusion to the independent
parameters L and y. If the characteristic linear dimension L (the length scale)
of the room is 5 m, the time scale T
, of this diffusion process is of the order
of lo6 sec (more than 100 h). In this estimate the value of y for air at room
temperature and pressure has been used (y = 0.20 cm2/sec).We conclude that
molecular diffusion is rather ineffective in distributing heat through a room.
On the other hand, even fairly weak motions, such as those generated by
small density differences (buoyancy), can disperse heat through the room
quickly. Suppose that the turbulent motion of the air in the room may also
be characterized by the length scale L (that is, motions are present of scales
L). This is a fair assumption, since large-scale motions are most effective in
distributing heat and since the largest possible scales of motion can be no
larger than the size of the room. We also need a characteristic velocity u (this
u may be thought of as an rms amplitude of the velocity fluctuations in the
room). For flow with a length scale L and a velocity scale u , the characteristic
time is

<

Tt--.

(1.4.4)

Apparently, Tt can be determined only if u can be estimated. Suppose the

radiator heats the air in i t s vicinity by A6 degrees Kelvin. This causes a
buoyant acceleration g Ae/O, which is of order 0.3 m/secz if 4 6 = 10K.
This acceleration probably occurs only near the surface of the radiator. I f it
has a height h = 0.1 m, the kinetic energy of the air above the radiator is
ghA6/6, which is of order 0.03 (m/sec) per unit mass. This corresponds t o a
velocity of 17 cm/sec. Much of the kinetic energy, however, is lost because of
the stable vertical temperature gradient in the room (the air near the ceiling
tends to be hotter than the air near the floor). A characteristic velocity u of
order 5cm/sec may be a reasonable average throughout the room. With
u = 5 cm/sec and L = 5 m, Tt becomes 100 sec, or about 2 min. Of course, we
still have to rely on molecular diffusion to even out small-scale irregularities
in the temperature distribution. However, the turbulence generates eddies as
small as about 1 cm (this estimate can be obtained with simple equations

10

Introduction

based on the dissipation of kinetic energy; those are discussed in Section 1.5).
The temperature gradients associated with these small eddies are smeared out
by molecular diffusion in a time of ordereZ/y (see Section 7.31,which is only a
few seconds if e = 1 cm.
Diffusion by random motion apparently is very rapid compared to
molecular diffusion. The ratio of the turbulent time scale Tt to the molecular
time scale T, is the inverse of t h e Peclet number:

Since for gases the heat conductivity y i s of the same order of magnitude as
the kinematic viscosity v (for air vly = 0.73; this ratio is known as the Prandtl
number), and since we are discussing only orders of magnitude, we may write
without compromise,

(1.4.6)
In our example, the Reynolds number R i s about 15,000.
This exercise shows that the Reynolds number of a turbulent flow may be
interpreted as a ratio of a turbulence time scale to a molecular time scale that
would prevail in the absence of turbulence in a problem with the same length
scale. This point of view is often more reliable than thinking of R as a ratio of
inertia terms to viscous terms in the governing equations. The latter point of
view tends to be misleading because a t high Reynolds numbers viscous and
other diffusion effects tend to operate on smaller length scales than inertia
effects.
Eddy diffusivity Since the equations governing turbulent flow are very
complicated, it is tempting to treat the diffusive nature of turbulence by
means of a properly chosen effective diffusivity. In doing so, the idea of
trying to understand the turbulence itself is partly discarded. If we use an
effective diffusivity, we tend to treat turbulence as a property of a fluid
rather than as a property of a flow. Conceptually, this is a very dangerous
approach. However, it often makes the mathematics a good deal easier.
If the effects of turbulence could be represented by a simple, constant
scalar diffusivity, one should be able to write for the diffusion of heat by
turbulent motions,

11

1.4 Diffusivity of turbulence

-ae
=K-

at

az8
aX,ax,

(1.4.7)

in which K is the representative diffusivity (often called "eddy" diffusivity

but sometimes called the "exchange coefficient" for heat). In order to make
this equation a t least a crude representation of reality, one must insist that the
value of K be chosen such that the time scale of the hypothetical turbulent
diffusion process is equal to that of the actual mixing process. The time scale
associated with (1.4.7) is roughly

T--

LZ

K'

(1.4.8)

and the actual time scale is Tt, given by (1.4.4). Equating T with Tt, one finds

K-UL.

(1.4.9)

It should be noted that this is a dimensional estimate, which cannot predict

the numerical values of coefficients that may be needed. Expressions like
(1.4.91, with experimentally determined coefficients, are used frequently in
practical applications.
The eddy diffusivity (or viscosity) K may be compared with the kinematic
viscosity Y and the thermal conductivity 7:

(1.4.10)
One concludes that this particular Reynolds number may also be interpreted
as a ratio of apparent (or turbulent) viscosity to molecular viscosity. A note
of warning is in order, though. In most flow problems, many different length
scales exist, so that the interpretation of Reynolds numbers based on these
length scales may not always be as straightfotward as in the example used
here.
It cannot be stressed too strongly that the eddy diffusivity K is an artifice
which may or may not represent the effects of turbulence faithfully. We
investigate this question carefully in Chapter 2.
Diffusion in a problem with an imposed time scale As another example of
the diffusivity of turbulence, we look a t boundary layers in the atmosphere.
The boundary layer in the atmosphere i s exposed to the rotation of the earth.

12

Introduction

In a rotating frame of reference, flows are accelerated by the Coriolis force,

which is twice the vector product of the flow velocity and the rotation rate.
I f the angular velocity of the frame of reference is f/2, it follows that atmospheric flows have an imposed time scale of orderllf. At a latitude of 40
degrees, the value of f for a Cartesian coordinate system whose z axis is
sec-' (f is called the Coriolis
parallel to the local vertical is about
parameter).
If the boundary layer in the atmosphere were laminar, it would be
governed by a diffusion equation like (1.4.1), so that i t s length and time
scales would be related by

Lk

vT.

(1.4.1 1)

With v = 0.15 cm2 sec-' and T = f-' = lo4 sec, this gives L, = 40 cm.
In reality, however, the atmospheric boundary layer is nearly always
turbulent; a typical thickness is about lo3 m (1 km). One can obtain some
appreciation for this by replacing v by K in (1.4.11) and substituting for K
with (1.4.9). This yields

Lt

uT,

(1.4.1 2)

which, of course, merely rephrases (1.4.4). In turbulent boundary-layer

1
flows, the characteristic velocity of the turbulence is typically about 30 of
the mean wind speed. For a wind speed of 10 m/sec, we thus estimate that
u 0.3 m/sec. With T = l / f = lo4 sec, (1.4.12) then yields L, 3 x lo3 m
(3 km), which is indeed of the same order as the observed thickness.
From a somewhat different point of view, we may argue that turbulent
eddies with a characteristic velocity u, exposed to a Coriolis acceleration
which imposes a time scale l/f,
must have a size (length scale) of order u/f. It
should be noted that we can equate eddy size and boundary-layer thickness
only because in most turbulent flows the larger eddies seem to have sizes
comparable t o the characteristic size of the flow in a direction normal to the
mean flow field (Figure 1.1). In estimates of diffusion or mixing, the large
eddies are relevant because they perform most of the mixing (K ueincreases
with eddy size).
Arguments of this nature are often supplemented by experiments to
determine the numerical coefficient in formulas like (1.4.12), because this
coefficient cannot be found by dimensional reasoning. In the case of the
atmospheric boundary layer,

13

1.4 Diffusivity of turbulence

Figure 1.1. Large eddies in a turbulent boundary layer. The flow above the boundary
layer has a velocity U ; the eddies have velocities u . The largest eddy size ( l )is comparable
to the boundary-layer thickness (Lt).The interface between the turbulence and the flow
above the boundary layer is quite sharp (Corrsin and Kistler, 1954).

Lt =41ulf

(1.4.13)

would give very close agreement between theory and experimental evidence.
Using (1.4.111,(1.4.12),
and T = l / f , we find the ratio between the
thicknesses of the laminar and turbulent atmospheric boundary layers to be

(1.4.14)
This is the square root of the Reynolds number associated with the turbulent
boundary layer in the atmosphere, since u/f is proportional to the actual
length scale L,. In this example, the Reynolds number R is clearly associated
with the ratio of the turbulent and molecular diffusion length scales:
turbulent flow penetrates much deeper into the atmosphere than laminar
flow. In our example, R lo.
The results obtained here concerning the different aspects of the Reynolds
number may be summarized by stating that in flows with imposed length
scales the Reynolds number is proportional to the ratio of time scales, while
in flows with imposed time scales the Reynolds number is proportional to the
square of a ratio of length scales. Since the Reynolds numbers of most flows
are large, these relations clearly show that turbulence is a far more effective
diffusion agent than molecular motion.

14

Introduction

The examples discussed here are rather crude because only a single length
or time scale has been taken into account. Most turbulent flows are far more
complicated; this introduction would not be complete without a look a t
turbulence as a multiple lengthscale problem.

1.5
Length scales in turbulent flows
The fluid dynamics of flows a t high Reynolds numbers is characterized by the
existence of several length scales, some of which assume very specific roles in
the description and analysis of flows. In turbulent flows a wide range of
length scales exists, bounded from above by the dimensions of the flow field
and bounded from below by the diffusive action of molecular viscosity.
Incidentally, this is the reason why spectral analysis of turbulent motion is
useful.
Laminar boundary layers Let us take a look a t the problem of multiple
scales in laminar shear flows. For steady flow of an incompressible fluid with
constant viscosity, t h e Navier-Stokesequations are

(1.5.1)
One would be tempted to estimate the inertia terms as Uz/L (U being a
characteristic velocity and L a characteristic length) and to estimate the
viscous terms as vU/L2. The ratio of these terms is UL/v = R, indicating that
viscous terms should become negligible a t large Reynolds numbers. However,
boundary conditions or initial conditions may make it impossible to neglect
viscous terms everywhere in the flow field. For example, a boundary layer has
to exist in the flow along a solid surface to satisfy the no-slip condition. This
can be understood by allowing for the possibility that viscous effects may be
associated with small length scales. The viscous terms can survive a t high
Reynolds numbers only by choosing a new length scale tsuch that the viscous
terms are of the same order of magnitude as the inertia terms. Formally,

U 2 / L-vu/P.

( 1.5.2)

The viscous length t i s thus related to the scale L of the flow field as

(1.5.3)

15

1.5 Length scales in turbulent flows

The viscous length G i s a transverse length scale: it represents the width

(thickness) of the boundary layer, because it relates to the molecular
diffusion of momentum deficit across the flow, away from the surface.
Molecular diffusion along the flow, of course, is negligible compared to the
downstream transport of momentum by the flow itself. Figure 1.2 illustrates
this situation.
Diffusive and convective length scales As (1.5.3)indicates, the boundarylayer thickness may be considerably smaller than the scale L of the flow field
in which the boundary layer (or other laminar shear flow) develops. The
distinction between a "diffusive" length scale across the flow and a
"convective" length scale along the flow is essential to the understanding of
all shear flows, both laminar and turbulent. Many shear flows are very
slender: their width is much smaller than their "length" (that is, the distance
from some suitably defined origin). The wide separation between lateral and
longitudinal length scales in shear flows leads to very attractive simplifying
approximations in the equations of motion; without this feature, analysis
would be next to impossible.
The most powerful of the asymptotic approximations associated with
d/L + 0 is that the shear flow becomes independent of most of its environment, except for the boundary conditions imposed by the overall flow. The
use of words like boundary layers, wakes, fronts in weather systems,
jetstreams, and the Gulf Stream is not a semantic accident. Because of the
wide difference in length scales, these shear flows are identifiable as distinct
regions in flow fields. These regions have distinct dynamics and distinct

Figure 1.2. Length scales, diffusion, and convection in a laminar boundary layer over a
flat plate.

16

introduction

characteristics; they are governed by specific equations of motion, which, in

the asymptotic approximation &'L+ 0, may be substantially simpler than the
equations governing other parts of the flow field.
Turbulent boundary layers It is useful to compare turbulent shear flows to
laminar ones, even though we can do so a t t h i s moment only in a very
rudimentary way. The relevant length and velocity scales in a turbulent
boundary layer are illustrated in Figure 1.3. The turbulent eddies transfer
momentum deficit away from the surface. With characteristic velocity
fluctuations of order u, the boundary-layerthickness L' presumably increases
roughly as &/dt
u (see Section 5.5). The time interval elapsed between the
origin of the boundary layer and the downstream position L is of order L/U
(convective time scale), so that we may estimate L'- u t uL/U. In effect, we
are equating the turbulent "diffusion" time scale t/u to the convective time
scale L/U. This procedure could also have been used for laminar boundary
layers. In laminar boundary layers, the diffusion distance L' increases as
(vt)12' ;with t = L/U, the result (1.5.2,1.5.3) is retrieved.
In analogy to (1.4.4) and (1.4.1 2), we thus can write the scale relations for
turbulent boundary layers as

u/u,

(1.5.4)

e/u .-, L/U.

(1.5.5)

t/L

\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\'

Figure 1.3. Length and velocity scales in a turbulent boundary layer. The time passed
since the fluid at L passed the origin of the boundary layer is o f order L/U.

17

1.5 Length scales in turbulent flows

These relations merely relate characteristic lengths and velocities; they should
not be used as formulas to compute the rate of spreading of a turbulent
boundary layer. The relation between the time scales, (1.5.51,rephrases the
fundamental assumption we implicitly encountered earlier, that is, that in a
situation with an imposed external flow the turbulence, being part of the
flow, must have a time scale commensurate with the time scale of the flow. As
we will see later, this assumption conflicts with eddy-viscosity concepts.
Fortunately, not all of the turbulence has such a large time scale: the small
eddies in turbulence have very short time scales, which tend to make them
statistically independent of the mean flow.
Laminar and turbulent friction If we compare (1.5.3)and (1.5.4)and
introduce experimental data, which suggest that u/U is of the order of lo-'
over a wide range of Reynolds numbers, we again get some appreciation for
the relatively rapid growth of turbulent shear flows. This rapid growth should
correspond to a larger drag coefficient.
For a steady laminar boundary layer in twodimensional flow on a plate
with length L, the drag D per unit span is equal to the total rate of loss of
momentum. Estimating the momentum loss as pU2e, where 8 is a boundarylayer thickness a t the end of the plate, we may put
D

pU 2t.

(1.5.6)

The drag coefficient (or friction coefficient) Cd is defined by

(1.5.7)
Substituting (1.5.6)into (1.5.7)and using the relation for t/L given by
(1.5.3),
we obtain

cd

e = 2R-112.
2L

(1.5.8)

For a turbulent boundary layer, on the other hand, the mass flow deficit
a t the end of the plate is proportional to pul (see Chapter 51,so that the rate
of loss of momentum is proportional to (pu4U. Consequently,

puUC

(1.5.9)

18

Introduction

The drag coefficient then becomes, if we use the definition (1.5.7) and the
scale relation (1.5.4),
( 1.5.1 0)

Experimental evidence shows that the turbulence level u/U varies very slowly
with Reynolds number, so that the drag coefficient of a turbulent boundary
layer, given by (1.5.101, should be very much greater than the drag
coefficient of a laminar boundary layer (1.5.8). Figure 1.4 illustrates this
point. Similar conclusions are valid for heat- and mass-transfer coefficients.
Equation (1.5.4) has another interesting implication. In boundary layers
and wakes u/U and t/L tend to zero as L increases beyond limit. In jets
entering fluid a t rest and shear layers, on the other hand,u/U andL/L approach
finite asymptotic values as L +m. This distinction is the origin of
some important differences in the asymptotic treatment of the two different
types of flow. In particular, jets and mixing layers spread linearly, while
wakes and boundary layers grow slower the farther downstream they travel.
Even so, most turbulent shear flows spread slowly enough to make& + 0 a
useful approximation.

19

1.5 Length scales in turbulent flows

Small scales in turbulence So far only the largest eddy sizes in turbulent
flows have been considered, because the large eddies do most of the transport
of momentum and contaminants. We have suggested that large eddies are as
big as the width of the flow and that the latter is the relevant length scale in
the analysis of the interaction of the turbulence with the mean flow. For
some of the other aspects of the dynamics of turbulence, however, other
length scales are needed.
We shall attempt to find the smallest length scales in turbulent flows. At
very small length scales, viscosity can be effective in smoothing out velocity
fluctuations. The generation of small-scale fluctuations is due to the nonlinear
terms in the equations of motion; the viscous terms prevent the generation of
infinitely small scales of motion by dissipating small-scale energy into heat.
This is characteristic of a small parameter like Y (more properly 1 / R ) with a
singular behavior. One might expect that a t large Reynolds numbers the
relative magnitude of viscosity is so small that viscous effects in a flow tend
t o become vanishingly small. The nonlinear terms in the Navier-Stokes
equation counteract this threat by generating motion a t scales small enough
to be affected by viscosity. The smallest scale of motion automatically adjusts
itself to the value of the viscosity. There seems to be no way of doing away
with viscosity: as soon as the scale of the flow field becomes so large that
viscosity effects could conceivably be neglected, the flow creates small-scale
motion, thus keeping viscosity effects (in particular dissipation rates) a t a
finite level.
Since small-scale motions tend to have small time scales, one may assume
that these motions are statistically independent of the relatively slow
large-scaleturbulence and of the mean flow. If this assumption makes sense,
the small-scale motion should depend only on the rate a t which it is supplied
with energy by the large-scalemotion and on the kinematic viscosity. It is fair
to assume that the rate of energy supply should be equal to the rate of
dissipation, because the net rate of change of small-scale energy is related to
the time scale of the flow as a whole. The net rate of change, therefore,
should be small compared to the rate a t which energy is dissipated. This is the
basis for what is called Kolmogorov's universal equilibrium theory of the
small-scale structure (Chapter 8).
This discussion suggests that the parameters governing the small-scale
)
the
motion include a t least the dissipation rate per unit mass E (m2 S ~ C - ~and

20

Introduction

kinematic viscosity v (mZ sec-' 1. With these parameters, one can form length,
time, and velocity scales as follows:

(1.5.1 1)
These scales are referred to as t h e Kolmogorov microscales of length, time,
and velocity (see Friedlander and Topper, 1962). In the Russian literature,
these scales are called "inner" scales.
The Reynolds number formed with q and u is equal to one

qulv = 1,

(1.5.12)

which illustrates t h a t the small-scale motion is quite viscous and t h a t the

viscous dissipation adjusts itself to the energy supply by adjusting length
scales.
An inviscid estimate for t h e dissipation rate One can form an impression of
the differences between the large-scale and small-scale aspects of turbulence if
t h e dissipation rate e can be related to the length and velocity scales of the
large-scale turbulence. A plausible assumption is to take t h e rateatwhich large
eddies supply energy to small eddies to be proportional to the reciprocal of
the t i m e scale of the large eddies. The a m o u n t of kinetic energy per unit mass
in the large-scale turbulence is proportional to u 2 ; the rate of transfer of
energy is assumed to be proportional to u/t: where [represents the size of the
largest eddies or the width of the flow. We shall see later t h a t [relates to the
"integral" scales of turbulence, which can be measured by statistical methods.
To avoid confusion, we identify d' f r o m here on as the "integral scale,"
leaving a more precise definition for Chapter 2. Russian scientists speak of
"outer" scales rather than of integral scales.
The rate of energy supply to the small-scale eddies is thus of order
u2 *u/t=u3/L This energy is dissipated a t a rate e, which should be equal to
the supply rate. Hence (Taylor, 1935),

u~e,

(1.5.13)

which states that viscous dissipation of energy can be estimated from the
large-scale dynamics, which do not involve viscosity. In this sense, dissipation
again is clearly seen as a passive process in the sense t h a t it proceeds a t a rate
dictated by the inviscid inertial behavior of the large eddies.
The estimate (1.5.13) should not be passed over lightly. It is one of the

1.5 Length scales in turbulent flows

21

Figure 1.5. Sketch o f the nonlinear breakdown of a drop of ink in water.

cornerstone assumptions of turbulence theory; it claims that large eddies lose

a significant fraction of their kinetic energy uz within one "turnover" time
t/u. This implies that the nonlinear mechanism that makes small eddies out of
larger ones is as "dissipative" as i t s characteristic time permits. In other words,
turbulence is a strongly damped nonlinear stochastic system. Some researchers believe that this feature may be related to the entropy production
concept embodied in the second law of thermodynamics. It should be kept in
mind, however, that large eddies lose a negligible fraction of their energy to
direct viscous dissipation effects. The time scale of their decay is d 2 / v ,so that
their viscous energy loss proceeds a t a rate vu2/d2,which is small compared
to 0 3 / tif the Reynolds number udv is large. The nonlinear mechanism is
dissipative because it creates smaller and smaller eddies until the eddy sizes
become so small that viscous dissipation of their kinetic energy is almost
immediate. The reader may gain some appreciation for the vigor of this
process by observing drops o ink or mill that are put in a glass of water
(Figure 1.5).
Scale relations

Substituting

1.5.13)into 1.5.1 I), we obtain

qlt- (~elv)-~'~
= R-3'4.

( 1.5.14)

rule- r / t = (u&)-'"

(1.5.15)

VIU

= R-''2,

( u ~ v ) - "=~R-'I4.

(1.5.16)

These relations indicate that the length, time, and velocity scales of the

22

Introduction

smallest eddies are very much smaller than those of the largest eddies. The
separation in scales widens as the Reynolds number increases, so t h a t one
may suspect that the statistical independence and the dynamical equilibrium
state of the small-scale structure of turbulence will be most evident at very
large Reynolds numbers.
The main difference between two turbulent flows with different Reynolds
numbers but with the same integral scale is the size of the smallest eddies: a
turbulent flow a t a relatively low Reynolds number has a relatively "coarse"
small-scale structure (Figure 1.6). Visual evidence of the small-scale structure
can be obtained if temperature fluctuations are present in the turbulence.
Temperature and index of refraction gradients are steepest if they are
associated with the smallest eddies; any optical system t h a t is sensitive t o
such fluctuating gradients "sees" the small-scale structure of turbulence. The
trembling, jittery horizon seen on a very hot day and the random pattern of

Figure 1.6. Turbulent jets a t different Reynolds numben: (a) relatively low Reynolds
number, (b) relatively high Reynolds number (adapted from a film sequence by R . W.
Stewart, 1969). The shading pattern used closely resembles the small-scale structure of
turbulence seen in shadowgraph pictures.

1.5 Length scales in turbulent flows

23

light and dark seen on the wall next to a heating element in sunlight are good
illustrations.
Vorticity has the dimensions of a frequency (sec-' 1. The vorticity of the
small-scale eddies should be proportional to the reciprocal of the time scale 7.
From (1.5.1 5) we conclude that the vorticity of the small-scaleeddies is very
much larger than that of the large-scalemotion. On the other hand, (1.5.16)
indicates that the small-scale energy is small compared to the large-scale
energy. This is typical of all turbulence: most of the energy is associated with
large-scale motions, most of the vorticity is associated with small-scale
motions.
Molecular and turbulent scales The Kolmogorov length and time scales are
the smallest scales occurring in turbulent motion. At this point, it is
convenient to demonstrate that most turbulent flows are indeed continuum
phenomena. The Kolmogorov scales of length and time decrease with
increasing dissipation rates. High dissipation rates are associated with large
values of u. In gases, large values of u are more likely to occur than in liquids.
Therefore, it is sufficient to show that in gases the smallest turbulent scales of
motion are normally very much larger than molecular scales of motion. The
relevant molecular length scale is the mean free path t . The velocity scale of
molecular motion in a gas is proportional to the speed of sound a in the gas.
Kinetic theory of gases shows that the product a5 is proportional to
the kinematic viscosity of the gas:

v at.

(1.5.17)

The ratio of the mean free path t to the Kolmogorov length scale q (this
might be called a microstructure Knudsen number) becomes (Corrsin, 1959)

tlq

M/R

'I4,

(1.5.18)

where we have used (1.5.14) and (1.5.17). In (1.5.18) the turbulence

Reynolds number R = UQV and the turbulence Mach number M = u/a are used
as independent variables. It is seen that turbulence might interfere with
molecular motion a t high Mach numbers and low Reynolds numbers. This
kind of situation is unlikely to occur, because M is seldom large, but R is
typically very large. A pertinent illustration i s the situation in gaseous nebulae
(cosmic gas clouds) (Spitzer, 1968). In clouds that consist mainly of neutral

24

introduction

hydrogen, the turbulent Mach number is of order 10 (u 10 km/sec,

a - 1 km/sec), while the Reynolds number is of order lo7 (1- 10l7 m,
$! 10' m). With (1.5.181, we compute that g/q 1/6. In this extreme case,
it seems doubtful that the smallest eddies perceive a continuum. In clouds
that consist mainly of ionized hydrogen, temperatures are quite high,
increasing a to about 10 km/sec and decreasing M to about 1. The mean free
path $. remains roughly the same (the density in ionized clouds is not
appreciably different from that in neutral clouds), so that R reduces to about
lo6. In this case, [/q-&, which may be small enough for the smallest
eddies to operate in a continuum.
The ratio of the time scale r to the collision time scale [/a associated with
molecular motion is, in terms of R and M,

'

ra/g- R ' I 2 M P 2 .

(1.5.19)

For M = 10 and R = lo', the smallest time scale of turbulence is 32 times as

large as the collision time scale of the gas molecules; for M = 1 and R = lo6
the ratio is 1000. It should be recognized that in ionized gases other length
and time scales are associated with the motion of the microscopic particles
and with the several other dynamical processes (radiation, cosmic rays,
magnetic fields) that may be present, so that r ) may not always be a relevant
length scale.
Because the smallest time scales in turbulent motion tend to be much
larger than molecular. time scales, the motion of the gas molecules is in
approximate statistical equilibrium, so that molecular transport effects may
indeed be represented by transport coefficients such as viscosity and heat
conductivity. These representations would become invalid if the departures
1
from equilibrium were large; the case t / q - g , ra/[ 32 would probably require treatment with the methods of statistical mechanics.

1.6
Outline of the material
The bird's-eye view of turbulence dynamics given in the preceding sectioris
sets the stage for a brief outline of this book. In Chapter 2, we deal with
eddy-viscosity and mixing-length theories. The dimensional framework of
these theories is useful in the analysis of typical shear flows. In Chapter 3, the
energy and vorticity equations of turbulent flow are derived. In Chapter 4,
some free shear flows like wakes and jets are discussed. In Chapter 5,

25

Problems

boundary layers are analyzed. To prepare a formal basis for the study of
diffusion and spectral dynamics, an introduction to statistics is given in
Chapter 6. In Chapter 7, turbulent diffusion and mixing are studied.
The study of the spatial dynamics of turbulent flows precedes that of the
spectral dynamics. There exist many similarities and analogies between spatial
and spectral dynamics of turbulence. Also, spatial dynamics can be visualized
more easily by those new to the subject. Once some of the subtle features of
turbulent shear flow are understood, the dynamics of turbulence in wavenumber space should not be too perplexing. Spectral dynamics is studied in
Chapter 8.
Problems

1.1 Estimate the energy dissipation rate in a cumulus cloud, both per unit
mass and for the entirecloud. Base your estimates on velocity and length
scales typical of cumulus clouds. Compute the total dissipation rate in
kilowatts. Also estimate the Kolmogorov microscale 7.Use p = 1.25 kg/m3
and Y = 15 x

m2/sec.

1.2 A cubical box of volume L 3 is filled with fluid in turbulent motion. No

source of energy is present, so that the turbulence decays. Because the
turbulence is confined to the box, its length scale may be assumed to be equal
to L a t all times. Derive an expression for the decay of the kinetic energy
?u2
as a function of time. As the turbulence decays, i t s Reynolds number
2
decreases. If the Reynolds number uL/v becomes smaller than 10, say, the
inviscid estimate f = u3/L should be replaced by an estimate of the type
f = cvu2/L2,because the weak eddies remaining at low Reynolds numbers
lose their energy directly to viscous dissipation. Computec by requiring that
the dissipation rate i s continuous a t uL/v = 10. Derive an expression tor the
decay of the kinetic energy when uL/u 10 (this is called the "final" period
m2/sec and u = 1m/sec a t time t = 0, how
of decay). If L = 1 m, v = 15 x
long does it take before the turbulence enters the final period of decay?
Assume that the effects of the walls of the box on the decay of the turbulence may be ignored. Can you support this assumption in any way?

<

1.3 The large eddies in a turbulent flow have a length scalet, a velocity
=&I. The smallest eddies have a length
scale v(fl = u, and a time scale

to

26

introduction

scale 9, a velocity scale u, and a time scale 7. Estimate the characteristic

velocity v ( r ) and the characteristic time t(r)of eddies of size r , where r is any
length in the range r ) < r < t . Do t h i s by assuming that v(r) and t(r) are
determined by E and r only. Show that your results agree with the known
velocity and time scales a t r =tand r = r). The energy spectrum of turbulence
is a plot of E ( K )= K-' v2 ( K ) , where K = l / r is the "wave number" associated
with eddies of size r . Find an expression for E ( K )and compare your result
with the data in Chapter 8.

1.4 An airplane with a hot-wire anemometer mounted on its wing tip is to

fly through the turbulent boundary layer of the atmosphere a t a speed of
50 m/sec. The velocity fluctuations in the atmosphere are of order 0.5 m/sec,
the length scale of the large eddies i s about 100 m. The hot-wire anemometer
is to be designed so that it will register the motion of the smallest eddies.
What is the highest frequency the anemometer will encounter?What should
the length of the hot-wire sensor be? If the noise in the electronic circuitry is
expressed in terms of equivalent turbulence intensity, what is the permissible
noise level?