Incompressible Fluid Dynamics

Lecture Notes in
Incompressible Fluid Dynamics:

Phenomenology,
Concepts
and Analytical Tools.
Jacques Lewalle
Syracuse University
2
Figure 1: Internal waves in a cloud, indicating warmer air above
Figure 2: Internal waves in a cloud layer at sunrise
c Jacques Lewalle 2006
Contents
0.1 What is dierent about these notes? . . . . . . . . . . . . . . 9
0.2 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
0.2.1 Part I: Introduction . . . . . . . . . . . . . . . . . . . . 13
0.2.2 Part II: Basic concepts and equations . . . . . . . . . . 14
0.2.3 Part III: Approximations . . . . . . . . . . . . . . . . . 14
1 Motivation 17
1.1 Internal Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.1.1 A simple problem . . . . . . . . . . . . . . . . . . . . . 17
1.1.2 Entrance ow . . . . . . . . . . . . . . . . . . . . . . . 19
1.1.3 Transition to turbulence . . . . . . . . . . . . . . . . . 21
1.1.4 Pipe exit and secondary ow . . . . . . . . . . . . . . . 24
1.1.5 Advanced problems . . . . . . . . . . . . . . . . . . . . 25
1.1.6 Food for thought . . . . . . . . . . . . . . . . . . . . . 25
1.2 External ow . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.2.1 Control volume analysis . . . . . . . . . . . . . . . . . 28
1.2.2 Potential ow model . . . . . . . . . . . . . . . . . . . 29
1.2.3 Phenomenology . . . . . . . . . . . . . . . . . . . . . . 30
1.2.4 Food for thought . . . . . . . . . . . . . . . . . . . . . 33
1.3 Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2 Kinematics 35
2.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.1.1 Intrinsic notations . . . . . . . . . . . . . . . . . . . . 36
2.1.2 Component notations . . . . . . . . . . . . . . . . . . . 37
2.1.3 Index notations . . . . . . . . . . . . . . . . . . . . . . 39
2.2 Eulerian vs. Lagrangian descriptions . . . . . . . . . . . . . . 42
2.2.1 Lagrangian description of motion . . . . . . . . . . . . 43
2.2.2 Eulerian description . . . . . . . . . . . . . . . . . . . 48
3
4 CONTENTS
2.2.3 Pathlines, streaklines and streamlines . . . . . . . . . . 49
2.2.4 Caution! . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.3 Dierential concepts and their geometry . . . . . . . . . . . . 56
2.3.1 Dierentials and vectors . . . . . . . . . . . . . . . . . 56
2.4 Mass balance . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.4.1 Vector potential and stream function . . . . . . . . . . 60
2.5 Flow about a point: Helmholtz . . . . . . . . . . . . . . . . . 61
2.5.1 Rate of strain . . . . . . . . . . . . . . . . . . . . . . . 61
2.5.2 Local rotation and vorticity . . . . . . . . . . . . . . . 64
2.6 Kinematic decompositions of a velocity eld . . . . . . . . . . 66
2.7 Vorticity, ,
2
, etc. . . . . . . . . . . . . . . . . . . . . . 67
2.7.1 Biot-Savart relation . . . . . . . . . . . . . . . . . . . . 67
2.7.2 Vorticity, streamfunctions, and more . . . . . . . . . . 69
2.7.3 Vorticity and boundary conditions . . . . . . . . . . . . 71
2.7.4 Discrete vortices vs. continuous vorticity . . . . . . . . 71
2.7.5 Vorticity and circulation . . . . . . . . . . . . . . . . . 72
2.7.6 The Helmholtz theorems . . . . . . . . . . . . . . . . . 72
2.7.7 Helicity, Lamb vector . . . . . . . . . . . . . . . . . . . 74
2.7.8 Famous vortices . . . . . . . . . . . . . . . . . . . . . . 74
2.8 Advanced topics and ideas for further reading . . . . . . . . . 80
3 Dynamics 85
3.1 Newtonian dynamics of continua . . . . . . . . . . . . . . . . . 85
3.2 Stress at a point, Newtonian uids . . . . . . . . . . . . . . . 88
3.2.1 The Navier-Stokes equations . . . . . . . . . . . . . . . 90
3.2.2 Boundary conditions . . . . . . . . . . . . . . . . . . . 92
3.2.3 Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.2.4 Vorticity in NS . . . . . . . . . . . . . . . . . . . . . . 94
3.3 Vorticity equation . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.4 Energy and dissipation . . . . . . . . . . . . . . . . . . . . . . 96
3.4.1 Bernoulli with losses . . . . . . . . . . . . . . . . . . . 98
3.5 Enstrophy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.6 Structure of the equations . . . . . . . . . . . . . . . . . . . . 99
3.7 Incompressible ow approximation . . . . . . . . . . . . . . . 100
3.7.1 Small Mach number ows . . . . . . . . . . . . . . . . 102
3.7.2 Boussinesq approximation . . . . . . . . . . . . . . . . 102
3.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
CONTENTS 5
4 Dimensionless expressions 105
4.1 Dimensional analysis . . . . . . . . . . . . . . . . . . . . . . . 105
4.2 Non-dimensionalization of equations . . . . . . . . . . . . . . . 110
4.3 Dimensionless Equations and Scaling Analysis . . . . . . . . . 112
4.4 Rational approximations . . . . . . . . . . . . . . . . . . . . . 114
4.4.1 Large Re . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5 Inviscid Flows and Irrotational Flows 121
5.1 Inviscid ow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.2 Bernoullis equation . . . . . . . . . . . . . . . . . . . . . . . . 122
5.2.1 The strong form . . . . . . . . . . . . . . . . . . . . . . 122
5.2.2 The weaker form . . . . . . . . . . . . . . . . . . . . . 123
5.2.3 Bernoulli and energy . . . . . . . . . . . . . . . . . . . 123
5.2.4 Simplied Croccos equation . . . . . . . . . . . . . . . 126
5.2.5 Applications of Bernoullis equation . . . . . . . . . . . 128
5.2.6 Lagranges theorem . . . . . . . . . . . . . . . . . . . . 128
5.2.7 Creation of vorticity . . . . . . . . . . . . . . . . . . . 128
5.2.8 Kelvins theorem . . . . . . . . . . . . . . . . . . . . . 129
5.2.9 Helmholtzs vortex theorems . . . . . . . . . . . . . . . 132
5.3 Irrotational ow . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.3.1 2D potential lines and streamlines . . . . . . . . . . . . 136
5.3.2 Superposition of 2D potentials . . . . . . . . . . . . . . 137
5.3.3 F=ma!? . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.3.4 DAlemberts paradox . . . . . . . . . . . . . . . . . . 143
5.4 Viscous potential ows . . . . . . . . . . . . . . . . . . . . . . 144
6 Stokes Flows 147
6.1 Stokes equation . . . . . . . . . . . . . . . . . . . . . . . . . . 147
6.1.1 Kinematic reversibility . . . . . . . . . . . . . . . . . . 150
6.2 Stokes two problems . . . . . . . . . . . . . . . . . . . . . . . 151
6.3 Stokes ow around a sphere . . . . . . . . . . . . . . . . . . . 153
6.3.1 Nonlocal eects . . . . . . . . . . . . . . . . . . . . . . 157
6.3.2 Application: Slender bodies . . . . . . . . . . . . . . . 158
6.4 Cylinder: Stokes paradox . . . . . . . . . . . . . . . . . . . . 158
6.5 Reynolds lubrication theory . . . . . . . . . . . . . . . . . . . 159
6.6 Lagrangian turbulence, chaotic mixing . . . . . . . . . . . . . 161
6 CONTENTS
7 Interlude 165
7.1 The approximations . . . . . . . . . . . . . . . . . . . . . . . . 165
7.2 Non-local eects . . . . . . . . . . . . . . . . . . . . . . . . . . 166
7.3 The role of vorticity . . . . . . . . . . . . . . . . . . . . . . . . 166
8 Narrow Flows 167
8.1 Flat plate boundary layers . . . . . . . . . . . . . . . . . . . . 167
8.1.1 Control volume analysis . . . . . . . . . . . . . . . . . 170
8.1.2 Scaling: mass balance . . . . . . . . . . . . . . . . . . 171
8.1.3 Scaling: streamwise momentum . . . . . . . . . . . . . 172
8.1.4 Scaling: transverse momentum . . . . . . . . . . . . . . 172
8.1.5 Scaling: pressure . . . . . . . . . . . . . . . . . . . . . 173
8.1.6 The ZPG BL equations . . . . . . . . . . . . . . . . . . 174
8.1.7 Similarity . . . . . . . . . . . . . . . . . . . . . . . . . 175
8.2 Integral method: von Kàrmàn s approach . . . . . . . . . . . 179
8.3 Vorticity and circulation in ZPGBL . . . . . . . . . . . . . . . 181
8.4 Descriptive: BL transition . . . . . . . . . . . . . . . . . . . . 182
8.5 Jets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
8.5.1 Streamwise development . . . . . . . . . . . . . . . . . 187
8.5.2 Similarity proles . . . . . . . . . . . . . . . . . . . . . 188
8.5.3 Entrainment . . . . . . . . . . . . . . . . . . . . . . . . 189
9 Flow Separation and Secondary Flow 191
9.1 Curved channel . . . . . . . . . . . . . . . . . . . . . . . . . . 191
9.2 Vorticity reversal in 2D separation . . . . . . . . . . . . . . . . 195
9.3 Introduction of vorticity . . . . . . . . . . . . . . . . . . . . . 199
10 Rotating Flows 203
10.1 Equations of motion . . . . . . . . . . . . . . . . . . . . . . . 203
10.1.1 Coriolis force . . . . . . . . . . . . . . . . . . . . . . . 205
10.2 Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
10.3 Geostrophic approximation . . . . . . . . . . . . . . . . . . . . 207
10.3.1 Taylor columns . . . . . . . . . . . . . . . . . . . . . . 209
10.4 Rossby waves . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
CONTENTS 7
11 Linearization 215
11.1 Surface waves . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
11.1.1 Tsunami speed . . . . . . . . . . . . . . . . . . . . . . 218
11.1.2 Internal waves . . . . . . . . . . . . . . . . . . . . . . . 222
11.1.3 More advanced topics . . . . . . . . . . . . . . . . . . . 223
11.2 Inviscid linear stability: Kelvin-Helmholtz . . . . . . . . . . . 223
11.2.1 Setting up the problem . . . . . . . . . . . . . . . . . . 223
11.2.2 Perturbation . . . . . . . . . . . . . . . . . . . . . . . . 225
11.2.3 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . 226
11.2.4 Stability as equilibrium . . . . . . . . . . . . . . . . . . 228
12 Additional Reading 233
12.1 About this course . . . . . . . . . . . . . . . . . . . . . . . . . 233
12.2 Beyond . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
8 CONTENTS
Introduction
0.1 What is dierent about these notes?
These are lecture notes: not a textbook, even less a reference book. Supple-
mentary material, in the form of specic reading from existing texts and ow
visualization (movies and still photographs), allows these notes to remain
concise.
This material has been taught as an entry-level course for graduate stu-
dents in Mechanical Engineering and related programs for a few years. While
Fluid Dynamics is a well-established discipline, its focus has shifted over the
years, and the range of applications has diversied. The biggest change re-
sults from the widespread availability of desktop CFD packages in engineering
practice, which makes it possible to obtain solutions to complex problems.
The dependence on applied mathematics (as well as inclination for it and
level of prociency) has decreased accordingly among typical students, but
it could be argued that physical insight is more important than ever. Actual
solutions, numerical or analytical or experimental, have built-in approxima-
tions and blind spots: with increasingly complex problems, it is important to
recognize these limitations, their possible eects on the solutions and their
interpretation. Analysis and insight are more dicult to combine in complex
ows.
Thus, I attempted to strike a new balance between the rich phenomenol-
ogy of incompressible ows, as presented e.g. by Tritton, and the wealth of
analytical methods found in references such as Batchelor, Panton and Currie.
There is also merit in a very selective approach such as Acheson. So, I have
devised my own mix of these excellent texts, and bits and pieces of them will
be recognized in this material. The classroom presentation will rely heavily
on movies (NSF series: GI Taylor, A Shapiro, F Abernathy, E Taylor, etc.)
and ow-vizualization still pictures (Van Dykes Album of Fluid Motion, the
9
10 CONTENTS
Figure 3: Overview of this approach
0.1. WHAT IS DIFFERENT ABOUT THESE NOTES? 11
Gallery of Fluid Motion in Physics of Fluids, and the many pictures on eu-
ids.com), for the double purpose of using the wonder of ow phenomena as
motivation, and to start the process of analysis and interpretation.
The main innovation in these notes is the graphic representation of re-
lations between ideas. For years, I have observed that students need help
placing ideas in context, need bridges between mathematics and reality. A
systematic approach is provided by mind-mapping , a graphical technique
related to free-association in psychology and to left vs. right brain thinking,
used rst in the context of business schools (N. Margulies, R. Carter) but
with versatile pedagogical merit (I learned about it from Mrs. Kate Regan,
our sons fth grade teacher). In this incarnation of the method, I ask the
students to map the web of concepts related to a given topic, so as to improve
awareness of connections, missing links, and analogies.
Take the example (Fig. 4) of what might have been learned about vis-
cosity at the undergraduate level: at the mention of the word viscosity, an
entire context should come to mind. A list of keywords (here: denition,
no-slip, friction and wall stress, Reynolds number, never Bernoulli, etc.), is
a skeleton for any written text. But conventional presentations (text or list
of keywords) are by necessity sequential. A 2-dimensional layout, empha-
sizing relations between keywords, is apparently read by a dierent part of
the brain: adding graphically expressive features and personal emphasis is
an important part of the exercise. Drawing a mind-map, rst collectively
in class and then individually as assignments, helps students create a con-
text. The question of Are we overlooking anything? is as important as
what is in the picture, but the context should be limited to items directly
related to the central topic, leaving further branchings (e.g. details about
Bernoulli, wall stress, etc.) for other diagrams. The end-result should be a
mental landscape, readable at a glance, showing connections and mismatches,
eventually supporting the analytical developments and the interpretation of
results. The frustration of where to start? in traditional problems may give
way to excitement at the realization that the handle is usually easier to nd
on the mind map: it connects the various pieces of the problem statement.
The version printed here is somewhat limited in expressive value: adding
colors, texture, etc., is more easily done by hand, although software with
clip-art libraries is now available. These maps of ideas can be more or less
detailed: one can, for example, trace vorticity throughout this text and get
a very dierent picture in Chapter 3, chapter 4, chapter 7 or 8: context is
everything.
12 CONTENTS
Figure 4: Two versions of a mind-map based on the same keywords
0.2. OVERVIEW 13
While ideas and techniques need to be presented in detail, some ques-
tions are left unanswered in these notes so as not to spoil the fun of real-time
thinking in the classroom. In the selection of material, fundamentals and
variety were more important than completeness or depth: especially with
pointers to specic topics, going to the library for additional information is
a normal step at any level of learning. The goal is to lead the students to
a fundamental understanding of reasonably complex ows, including the na-
ture of approximations, possible shortcomings and opportunities for further
development. Formal manipulations are but one aspect of the learning: the
sequence of ideas, the relations with other elds (elasticity, thermodynam-
ics, etc.), and the ability to approach the wealth of applications, are more
important.
0.2 Overview
The help of a publisher would make two obvious improvements manageable:
the professional drafting of gures, and the inclusion of copyrighted pho-
tographs and movie clips in a multimedia package. The expansion of these
notes to include other topics would rely on contribution from co-authors: this
is my selection.
The course is in three parts. The students should review independently
some undegraduate topics, such as the Reynolds transport theorem, Bernoullis
equation (without and with losses) and dimensional similarity. These meth-
ods are assumed to be known, regardless of the students varied backgrounds.
0.2.1 Part I: Introduction
Ch. 1 is a short version of Trittons idea, to illustrate the limitations of
the undergraduate tools of analysis for both internal and external ows.
These topics involve a rich phenomenology for which new analytical tools
are required. This is the vehicle for two main ideas. First, the need to
understand the phenomena in question and their side-eects is motivation
for the introduction of more advanced concepts in Part II. Second, the maps
of ideas are introduced in a known context, and they summarize the many
loose ends.
14 CONTENTS
0.2.2 Part II: Basic concepts and equations
Chs. 2 and 3 are rather conventional in scope, with very few attempts at
originality (the relation between Biot-Savart and the inverse Laplacian being
an exception). Kinematics (Ch. 2) formalize the tools that do not involve any
reference to forces (Newtons second law): the geometry of motion. At one
level, this is descriptive, including ow visualization and plotting of solutions
for various purposes of understanding and communication. At another level,
it is deeply analytical, with more partial derivatives than some students feel
comfortable with. The teachers challenge is to link graphical and analytical
aspects into one unied context. Chapter 3 is devoted to dynamics, i.e. the
form taken by Newtons second law in continua (Cauchys equation), then
more precisely in inviscid (Eulers equations) and Newtonian incompressible
ows (Navier-Stokes equations). Formal similarities for the balance of mo-
mentum, mass and energy are exploited, and the governing equations are
derived for related quantities such as pressure, energy and vorticity.
0.2.3 Part III: Approximations
Very few exact solutions are known for the Navier-Stokes (NS) equations.
For the Couette-Poiseuille class of fully-developed parallel steady ows, con-
vective terms cancel out and the simple balance of pressure and viscous forces
is featured in undergraduate texts. Unsteady problems of simple geometry
(Stokes problems: impulsively started and oscillating plates in a viscous
uid) are also well known. But there are remarkably few exact solutions of
the NS equations including nonlinearities, and they are all time-independent.
Finding closed-form exact solutions is not what this eld is about (although
if you do, you will be famous).
Of course, the naive student will think in terms of just solving the NS
equations numerically (or experimentally). But even computers and software
relate poorly to brute force: it is necessary to make approximations, and to
understand their implications (strengths and shortcomings). Experimentally,
not all variables can be controlled or measured, and again approximations
need to be interpreted correctly. The third part of the course is a survey
of approximation methods to be developed more fully in specialized courses.
More importantly, equations such as NS can yield a lot of information even
if we cannot solve them: it is all about extracting information and putting
it in context.
0.2. OVERVIEW 15
One overriding technique is presented rst (Ch. 4): suitable scaling of the
equations brings out dimensionless coecients to the various terms, some of
which may be relatively small and might be neglected (one needs to be careful
with this approach: very large Reynolds number is associated with turbu-
lence, while innite Reynolds number may yield unique solutions). Then,
in turn, we explore the features of inviscid and of irrotational ow (Ch. 5),
the small Re limit (Stokes ow in Ch.6), narrow ows (boundary layer fam-
ily) in Ch. 8, rotating ows (small Rossby number) in Ch. 10, and linear
waves and stability (Ch. 11); and examine the phenomena responsible for
ow separation and secondary ows (Ch. 9). The balance between depth
of coverage and breadth of ideas is constrained by the 1-semester format:
this particular selection is but one choice among other possibilities. Many
colleagues and students have contributed encouragements and constructive
criticism. All suggestions were thankfully considered, even if they were not
adopted. There would be more questionable statements about vector nota-
tions and continuum mechanics without Prof. AJ Levys input, and more
typographical errors without A Nicolais, Y Wangs and Prof. Carlos Perezs
patient proofreading. I will happily correct any errors introduced since, as I
become aware of them. And the other shortcomings are all mine, I am afraid.
JL
16 CONTENTS
Chapter 1
Motivation
This chapter, inspired by Trittons book, presents the need for more advanced
tools than are typically covered at the undergraduate level in a typical En-
gineering curriculum in the USA. For internal ows, the canonical case is
the ow in a circular pipe. The student should be familiar with the laminar
(Poiseuille) parabolic velocity prole, with the Moody diagram for pressure
loss in a pipe, and with the modied Bernoulli equation including friction
and minor losses. So we understand laminar friction and pressure drop, but
how about entrance losses? Secondary ows at the entrance and exit? Flows
in curved pipes? Pulsatile ow?
At the undergraduate level, external ows are studied in relation with
drag in uniform ows (usually involving empirical dimensionless curves of
drag coecients, and a description of ow separation and the drag crisis),
at plate boundary layers, and possibly some simple potential ows such as
the ow around a cylinder. In this chapter, we revisit these topics to motivate
the introduction of the more advanced concepts required to understand e.g.
ow separation. We will end up with many questions marks, many unresolved
diculties to be addressed later in the course.
1.1 Internal Flow
1.1.1 A simple problem
There are a number of practical design issues related to the pipe ow issuing
from a head tank (Fig. 1.1). There are also some added features (e.g. about
17
18 CHAPTER 1. MOTIVATION
Figure 1.1: A simple pipe ow set-up
entrance pressure drop) that can be explained on the basis of undergraduate
tools. And an engineer should not overlook these approaches: simple rst!
But there are also a number of facts that show the need for more advanced
concepts, more sophisticated tools (and more advanced mathematics).
On the practical side, it would be important to maintain constant ow
rate. This can be achieved in several ways:
use a constant-displacement pump, insulated from the pipe by a baed
plenum; or
use an overow pipe, so the supply rate into the head tank needs no
monitoring; etc.
You might want to discuss alternative designs. By running this experiment
over a wide range of Reynolds numbers, and making sure the pipe is long
enough so entrance and exit losses can be neglected (how long is that?), one
can collect the data represented on the Moody diagram (Fig. 1.2), here shown
for a smooth pipe. The analysis was carried out in the previous chapter.
1.1. INTERNAL FLOW 19
10
2
10
3
10
4
10
5
10
6
10
7
10
8
10
3
10
2
10
1
10
0
Reynolds number
F
r
i
c
t
i
o
n

f
a
c
t
o
r
Figure 1.2: Sketch of Moody diagram for pipe friction
Refer to the previous chapter for the basic analysis, using Bernoullis
equation and control volume in the pipe. For automated calculations, one
can make use e.g. of Colebrooks formula for the friction factor f
1
f
= 2.log
10
(
e
3.7
+
2.51
Re
D
f
), (1.1)
where e
is the pipes relative roughness (Fig. 1.2). We now turn to some of

the details.
1.1.2 Entrance ow
At one level, an improved analysis in the entrance region of the ow uses el-
ementary tools, even if the execution (e.g. to calculate the entrance length)
can be rather complicated (can be found in uids or heat transfer undergrad-
uate texts: it can be done). At another level, as we pay closer attention, the
shortcomings of the analysis are more obvious.
Figure 1.3: Entrance region of pipe
Entrance pressure drop
Assume a sharp leading edge to the pipe; a boundary layer (Fig. 1.3) devel-
ops on the curved wall (which appears at as long as /x remains small: this
requires some thought, along the lines of the earth is at). The main dier-
ence with the ZPG boundary layer is that the freestream, near the center
of the pipe, cannot be deected outward, since there is no outward. Refer
to Ch. for the context: the same web of ideas is present here. The defect
in mass ux in the wall region (as the ow changes from top-hat entrance
prole to no-slip boundary layer prole) must be made up in the core region,
even before it experiences direct eects of viscosity.
In other words, the core velocity must increase - the ow is accelerated.
Since F = ma, there must be a force. By default, pressure is the only avail-
able source, therefore pressure must drop in the entrance region to overcome
the growth of the boundary layers! This can explain entrance losses, up to a
point, in laminar ow. Putting this in equations, a variety of methods apply:
integral (von Kàrm` an method), and series expansions for the solution, are
textbook material. Once the idea is clear, the execution is a matter of skill
and perseverance.
Vena contracta and recirculation
The problem is actually made more complicated by the fact that, possibly in
relation with the nite thickness of the pipe wall which should be considered
as a blu body, the ow separates from the wall at the leading edge (Fig. 1.4).
Figure 1.4: Vena contracta and recirculation
The streamlines are no longer nearly parallel, and an annular recirculation
region appears near the wall. Within this region, the ow is even going
upstream near the wall!
In fact, once we examine the entrance region (or many other simple
problems) in more detail, some of the features can be understood based on
undergraduate material (mass ux, acceleration, pressure drop), and others
clearly cannot (streamlines, separation and reattachment, secondary ow,
pressure recovery, etc.) (Fig. 1.5).
These questions are important because similar phenomena occur in a
variety of applications. Flow separation is very important in aerodynamics
(reduces the eective cross-section in turbine passages, and reduces lift). Sec-
ondary ows matter in environmental ows (pollutants can remain trapped
in such regions), and in wake formation (see below). The representation of
streamlines needs analytical support, or even a clear denition, so we can
trust that the software is capable of catching such ow features. And so
on: undergraduate uid mechanics cannot go beyond the descriptive level
for these phenomena, we need something better.
1.1.3 Transition to turbulence
This last remark is also true for the transition to turbulence (Fig. 1.7).
From Stewarts movie on turbulence, the initial clip about pipe ow is rel-
evant here (Fig. 1.6). Undergraduates know about the jump on Moodys
diagram, about the magic number Re 2000 (or is it 2300??). A description
of the phenomenon includes, possibly, dye injection to visualize the ow (are
they streamlines?), the appearance of wavy ripples, the abrupt but inter-
Figure 1.5: Pipe entrance: some ideas
R. Stewarts movie (excerpt): Turbulence
We only cover the pipe ow experiment: transition
The set-up: viscosity as control parameter
Visualization of pressure drop
Velocity proles
Transition, pressure drop vs. viscosity
Secondary ow with laminar discharge
Figure 1.6: Transition in pipe ows, laminar discharge secondary ow: ex-
cerpt from Stewarts movie on turbulence.
Figure 1.7: Transition in pipe ow
mittent transition sending turbulent pus down the pipe, the possibility of
relaminarization, the intense mixing by the turbulence, the increase in drag,
etc. Critical Re, what does it depend on?
(engineering: design away from critical point to have predictable behav-
ior)
Beyond the empirical observations, we do not yet have the tools to under-
stand any of the qualitative or quantitative features of transition in a pipe.
Therefore we cannot begin to understand which features would occur in other
types of ows, or how software might include these important phenomena,
or whether it should...
Figure 1.8: Secondary laminar ow at pipe exit
1.1.4 Pipe exit and secondary ow
A simpler but equally puzzling situation occurs at the pipe exit. Again, we
can inject dye near the centerline or the wall to get an idea of what happens
how do we do this numerically?
The turbulent case is, in a way, simpler: with momentum thoroughly
mixed across the pipe, and the dye with it, relatively simple jet comes out
of the pipe: still turbulent inside, with surface tension eects (no covered
in this course, what a pity!) aecting interface geometry (jet break-up into
drops? surely this could matter for some applications.)
But let us focus on the laminar ow discharge (Fig. 1.8). The jet is
now made of three distinct parts: the main jet following a roughly parabolic
trajectory as in the turbulent case, and a puzzling stream falling at a much
steeper angle, the two connected by a thin lm of water (or whatever the
liquid is) (surface tension again! Look at the pictures in Van Dykes book or
in the Gallery of Fluid Motion in Physics of Fluids).
Dye injection shows that the minor stream collects the low-speed liquid,
not only from the bottom of the pipe but also from the top. An end-view of
the ow shows that the dye lament wraps around the central jet, unable to
maintain near-horizontal motion. We would expect this from free particles
coming out at dierent speeds, but how does a continuous jet do it? Its
particles cannot fall through the high-speed core, they have to go around it!
Figure 1.9: Laminar liquid jet discharge: some ideas
and why do the uid particles sort themselves into two main speed categories?
(Fig. 1.9.)
1.1.5 Advanced problems
Curved/helical pipes and secondary ows, pulsatile ows, turbulent ows,
non-circular cross-section, variable viscosity,...
1.1.6 Food for thought
Other undergraduate tools are clearly insucient to understand the com-
plexity of phenomena.
For example, consider minor losses in pipe systems: one level of engi-
neering solution is to look up the appropriate loss coecients K
m
in a table.
But how trustworthy are these numbers in compressible ow? in two-phase
ows? in ows where variations in viscosity (with temperature) are large?
What are the relevant ideas that would raise a red ag or give us condence
in the simple tools?
Figure 1.10: Minor losses: questions and phenomena
1.2. EXTERNAL FLOW 27
Figure 1.11: Mixing in a control volume: some ideas
The same might be said of such a common (and powerful) tool as control
volume analysis. Satisfying the basic conservation laws (mass, momentum,
energy, entropy) overall for a macroscopic system of interest sweeps under
the rug a number of details about some mechanisms that might aect design.
Suppose the pipe is a chemical reactor: surely mixing would be an important
concern, as might energy dissipation and pressure drop. Is the friction coef-
cient (Moody chart) valid regardless of temperature or other conditions of
reaction? Is it safe to extrapolate data (with the implication that no qual-
itative changes would occur)? If mixing is desirable, is it sensible to add a
grid? or some other device? Would better reaction yield be obtained without
turbulence at all?
1.2 External ow
Flows around obstacles (such as aircraft wings, tennis balls or underwater
robot arms) oer a variety of problems that can be organized in order of dif-
culty and generality. Canonical ows combine key common elements of the
ow physics with a simple-enough conguration so that the understanding of
the simple case does shed light on the more complicated reality. Thus, uni-
form ows around circular cylinders and spheres are a starting point toward
more complicated (realistic) situations. They can be studied analytically
Figure 1.12: Uniform ow past a cylinder: momentum defect
under various approximations, as seen later in this course. They also show,
outside these approximations, the need for deeper understanding.
1.2.1 Control volume analysis
As the diagnostic example, we look at the ow around a cylinder (Fig. 1.12):
specically a steady uniform ow at speed U
0
in the x direction (streamwise),
impinging normally on a circular cylinder (z direction, spanwise) of diameter
D. A wake develops downstream of the cylinder, where the velocity is less
than in the freestream:
U(x, y) = U
0
u(x, y). (1.2)
From the velocity prole, we can dene the wake half-width w as the dis-
tance (from the center-plane) beyond which the relative velocity defect u/U
0
can be neglected. The centerline velocity defect u(x, 0) < U
0
will decrease
downstream.
We take a control volume (Fig. 1.12): a parallelepiped of unit depth in
the spanwise direction (normal to the paper), along the plane of symmetry
of the problem (are we assuming that the entire solution, including the wake,
is symmetric?), extending upstream and downstream of the cylinder over a
length L, and considerably wider than the wake half-width (say, w+d at the
downstream end). There is no ow in the z-direction (why is that?).
Mass balance (per unit density) is summarized in the equation (Reynolds
transport theorem again! a basic tool)
0 =
(U dA)
= U
0
(w + d) +
L
(v.dx) + U
0
d +
([U
0
u]dy) + 0
=
L
(v.dx)
w
0
u(x) dy (1.3)
which, again, shows a ow away from the plane of symmetry as a result of
the defect in mass ux in the wake.
For the momentum balance, we have to acknowledge that the cylinder is
applying a drag force F (negative x direction) on the uid; pressure is uniform
at the control surface, located far enough from the cylinder so streamwise
viscous forces are negligible. Then
= U
0
[U
0
(w + d)] +
L
U
0
(v.dx) +
w+d
0
(U
0
u(x, y))
2
dy + 0
= U
2
0
w + U
0
u(x)
w
0
f dy +
w
0
(U
0
u)
2
dy
=
w
0
u
2
dy U
0
w
0
u dy (1.4)
Consequently
F
1
2
U
2
0
= 2
w
0
u
U
0
(1
u
U
0
)dy (1.5)
which shows, as for the boundary layer, that the net force is related to the
defect in momentum ux.
1.2.2 Potential ow model
An exact solution (Fig. 1.13) can be obtained if we assume that the ow is
potential, i.e.
u = (1.6)
3 2 1 0 1 2 3
3
2
1
0
1
2
3
x
y
Potential flow around a cylinder
Figure 1.13: Irrotational ow past a cylinder.
for some velocity potential . The solution is found in undergraduate texts,
See more details in Ch. 5. This elementary solution accounts for front
stagnation pressure, reasonable streamlines, acceleration toward the mid-
section of the cylinder, etc. Its failure to account for drag (DAlemberts
paradox, see Ch. 5) was resolved by Prandtls discovery of the boundary
layer. But the model shows no prospect of being extended to account for
some of the phenomenology: Kàrm` an vortex street, unsteady wake, drag
crisis...
1.2.3 Phenomenology
A dramatic feature of blu-body wakes at low Reynolds numbers is the sep-
aration of the boundary layer from the obstacle surface, and (in the case
of the cylinder) formation of the Kàrm` an vortex street (Fig. 1.14): a stag-
gered array of counter-rotating vortices illustrated e.g. on the cover of Van
Dykes book. As the Reynolds number increases, the shear layers become
increasingly unstable, and soon the wake becomes turbulent. Even so, in
the intermediate wake, ow structures reminiscent of the vortex street are
Figure 1.14: Kàrmàn vortex street behind a cylinder
Figure 1.15: Wake structure
observed (Fig. 1.15).
Moving from the descriptive to the quantitative, the drag curve also points
to the need for deeper understanding. The variations in the drag coecient
(Fig. 1.16) at low and moderate Reynolds numbers can be tied to some
of the wake behavior described above. But the sudden dip in the drag is
associated to a dierent phenomenon: the boundary layer itself becomes
turbulent before it separates from the surface (Fig. 1.17), thus delaying
separation. Such behavior is important in design (from dimples in golf balls
to vortex generators on aircraft wings), but cannot really be understood
based solely on undergraduate analysis tools.
Figure 1.16: Drag coecient for a cylinder in a uniform ow.
Figure 1.17: Eect of BL transition on wake formation
1.3. PERSPECTIVE 33
1.2.4 Food for thought
Most students would have heard about the transition Reynolds numbers:
2000 for pipe ow, 500,000 for at plate boundary layers: we need to under-
stand the meaning of such numbers, and why they are so dierent.
Problems
1. Derive the Poiseuille solution for channel ow; express the wall shear
stress and friction coecient.
2. Read paper or reference textbook on helical pipes: map the ideas rel-
ative to new phenomena/concepts/questions relative to this chapter.
(do not write or solve equations).
3. Ditto with pulsatile ow
4. Select one of the following problems, nd relevant documentation at the
library, and identify to what extent the phenomena can or cannot be
described by undergraduate methods: pulsatile ow in a pipe, curved
trajectories of balls (baseball, ping-pong, tennis or soccer), wake of an
airfoil with lift, wind gusts in downtown canyons.
1.3 Perspective
Need to understand vorticity, separation, secondary ows, transition, ...
Figure 1.18: Wake formation: some ideas
Chapter 2
Kinematics
Kinematics is a study of the geometry of motion independently of applied
forces. It consists of descriptive tools and various constraints that limit in
important ways how the uid can respond to the application of forces.
We will learn about vectors, with velocity as the foremost example; about
various algebraic, dierential and integral operations on vectors; about such
vectors as attributes of a particle (Lagrangian description) or of a point in
space (Eulerian description); about rate-of-strain and vorticity, maybe the
most important new concept in the whole course; about mass balance and
its consequences; and about ow decompositions, vortices as ow structures,
and much more.
Particularly relevant movies are: Klines Flow visualization, as back-
ground for all others; Lumleys Eulerian and Lagrangian descriptions, and
Shapiros Vorticity.
2.1 Vectors
Classical mechanics are rmly grounded in 3-dimensional space, in which
vectors are ubiquitous objects. Sure enough, scalar quantities such as energy
and pressure are important, as seen in Bernoullis equation; but mechanics
is dominated by the use of vectors for position, momentum, rotation, etc.
Vectors are mathematical objects endowed with a direction and a mag-
nitude they may also have a point or line of application. In this course,
we will use three equivalent notations for vectors, each with advantages de-
pending on the problem at hand. It is important that the student learn to
35
36 CHAPTER 2. KINEMATICS
use these notations interchangeably: practice, practice...
2.1.1 Intrinsic notations
The rst notation, used in earlier chapters, makes no reference to any co-
ordinate system: the vectors exist independently of how they are described.
This is the notation used in previous chapters, by default. A vector is de-
noted by an underline, as in velocity u. Operations on vectors include the
multiplication by a number (scalar), e.g.
u = u, (2.1)
the dot product, which makes a scalar from two vectors taken in any order
u
2
= u u (2.2)
and the cross product, which makes a vector from two vectors, but reverses
sign if the order of vectors is changed
u = r = r . (2.3)
The student is expected to know these operations from undergraduate classes,
this presentation being included for the sake of notations and as a point of
reference for alternative notations (below).
Of a dierent nature, the vector-operator is dened by its properties,
although in many ways the component notation may be easiest to grasp.
Assuming some background again, we simply note the gradient of a scalar
function
grad f = f, (2.4)
the divergence of a vector
div u = u, (2.5)
the curl of a vector
curl u = u, (2.6)
and the Laplacian
2
= (2.7)
(this last denition holds regardless of what the operator is applied to, vector
or scalar). As an operator, only acts on the variables that follow it,
therefore
u = u. (2.8)
2.1. VECTORS 37
Without implication about a particular system of axes, this notation will
be used by default, but is sometimes awkward when tensors of orders larger
than vectors are being used, e.g.
u or (a b), (2.9)
for which alternative notations are clearer.
2.1.2 Component notations
Component notations are most popular at the undergraduate level, but suf-
fer from runaway bulk! We will use them in this course only when dierent
directions contain dierent physics that should be reected in the notations,
for example in boundary layers where the streamwise (along the wall), trans-
verse (normal to the wall) and spanwise (along the wall, normal to the other
two) directions are very distinct.
The use of non-Cartesian coordinates (polar, cylindrical, spherical, etc.)
is less popular for beginners, but is actually implied by the above reason-
ing: boundary conditions may well dictate which coordinate system to use
for easier handling! We will use non-Cartesian coordinates as needed, and
encourage the students to overcome any reluctance in this regard as soon
as possible. Pointers will be included at appropriate places throughout the
chapters.
Coordinate axes may be Cartesian (e
x
, e
y
,e
z
), equivalent to (i,j,k); or
cylindrical-polar (e
r
,e
, e
z
); among many options. The component of a vector
along such an axis is the projection of the vector on the axis, e.g.
u
x
= u e
x
. (2.10)
Then, we can write
u = u
x
e
x
+ u
y
e
y
+ u
z
e
z
. (2.11)
It is sucient to write the vector as the ordered set of its components, for
example (u
x
,u
y
,u
z
) for which the alternative notation (u,v,w) is also common.
Multiplication by the unit vectors is implied, and should be restored when
in doubt. This notation can then be used to evaluate vector operations. The
dot product reduces to
u v = u
x
v
x
+ u
y
v
y
+ u
z
v
z
. (2.12)
Figure 2.1: Vector notations
2.1. VECTORS 39
For the cross product u = r, its x-component is
u
x
=
y
r
z
z
r
y
, (2.13)
and similar expressions are obtained by even permutation of the subscripts.
The vector-operator is best expressed in Cartesian coordinates. Then,
its components are (
x
,
y
,
z
) and we can write (in a convenient mix of
notations)
u =
x
u
x
+
y
u
y
+
z
u
z
. (2.14)
Similarly, the x-component of = u is
x
=
y
u
z
z
u
y
, (2.15)
and the Laplacian is the familiar
2
=
2
xx
+
2
yy
+
2
zz
. (2.16)
There is no general expression for in non-Cartesian coordinates; the con-
text of gradient, divergence, curl or Laplacian is necessary for specic terms
to make sense, and the corresponding expressions can be found in many
reference books.
2.1.3 Index notations
A more concise version of component notations is possible when the Carte-
sian axes are assumed, but without any particular orientation. Then, the
subscripts x, y and z are just an ordered triad, and 1, 2 and 3 turns out to
facilitate the sums. With unit vectors of the form e
i
for i = 1, 2, 3, we have
u =
u
i
e
i
(2.17)
and the triad of components can be written simply as (u
i
). Then for a dot
product
u v =
i
u
i
v
i
= u
i
v
i
(2.18)
where the summation over repeated indices is implied (Einsteins convention).
Some textbooks and websites use index notations while keeping explicitly
the unit vectors e
i
: this defeats the purpose of compact notations. Any
component with a free index is understood to multiply the corresponding
unit vector, with summation implied: u
i
stands for all components and unit
vectors together. The use of e
i
will be penalized in this course.
Concise notations can save work, but require neat handling. Students
should be careful to adhere to the following rules:
any repeated index in a product implies summation; therefore, that in-
dex pair should be used only once in an expression, but can be renamed
at will (dummy index) to avoid conicts with other pairs. Thus
u
i
v
i
= u
k
v
k
; (2.19)
no index may appear 3 or more times in a product; in rare instances
when this occurs, the use of more detailed notations is required, and
summation is usually restored explicitly to avoid confusion
an index appearing once denotes the component of the vector (free
index); it must be the same for all terms in a sum, since they all are
factors of the same implied e
i
; thus
a
i
= b
i
+ c
k
(2.20)
is meaningless.
They can also be generalized very usefully to higher-order objects: while
a vector requires a single index, a matrix requires 2, etc. The economy of
notation is most obvious in such context, as we will see in later chapters.
Two symbols are particularly important: the Kronecker symbol
ij
(2.21)
has value 1 if i = j and 0 otherwise. It is a representation of the identity
matrix, since
ij
u
j
= u
i
. (2.22)
Note also that, in 3 dimensions,
ii
= 3. (2.23)
More complicated, the permutation symbol
ijk
(2.24)
2.1. VECTORS 41
has value 1 if (i,j,k) is an even permutation of (1,2,3), -1 if an odd permu-
tation, and zero otherwise (i.e. if any two indices are equal). is used to
represent the cross product:
u
i
= ( r)
i
=
ijk
j
r
k
. (2.25)
This should be veried component by component (yes, do it!) One notewor-
thy relation, used for handling double cross products, is
ijk
ilm
=
jl
km
jm
kl
. (2.26)
Note that the order of the indices of can be rearranged based on
ijk
=
kij
=
jik
(2.27)
and similar alternatives
With these tools, we can write the dierential operators as
(f)
i
=
i
f (2.28)
for the gradient; the divergence takes the form
u =
i
u
i
. (2.29)
The Laplacian is
2
=
2
ii
. (2.30)
Finally, for the curl
i
=
ijk
j
u
k
. (2.31)
The order of indices of is important: rst index for the free index, second
for the derivative, third for the vector of which the curl is taken (and you
can rotate the indices afterward, of course).
Using index notations requires practice (beyond assigned problems), but
is well worth the eort when mastered. In this course, all three notations
will be used depending on the context: the idea is to save work and facilitate
formal manipulations.
J.L. Lumleys movie (excerpt): Eulerian
vs. Lagrangian
No dynamics! This is kinematics
Velocity of particle, velocity at a point
Convective derivative
Pathlines, streaklines, timelines, streamlines
Unsteady ow: wave eld
Figure 2.2: Eulerian and Lagrangian descriptions, from Lumleys movie.
Tensor notations
In the case of non-Cartesian coordinates, index notations can be generalized
as tensor notations
1
. Added complications include co- and contravariance,
and the use of the local metric tensor (responsible for the strange factors
in the expression of curl, div and grad). In uid dynamics, this can be
encountered in some treatments of computational grid generation, in the
study of non-Newtonian uids, and a few other applications, but will not be
needed in this course.
2.2 Eulerian vs. Lagrangian descriptions
Viewing of Lumleys movie, as summarized on Fig. 2.2.
Newtons mechanics, with its discovery of the point particles as a useful
concept, traces the motion of material points. These points are endowed
with mass, but otherwise have no properties, and the purpose is to study the
motion of these simple objects. This simple program was expanded later to
include rigid bodies (motion of and around a point, moments of inertia) and
1
Even more elegant notation is made possible by using exterior products and dierential
forms; I am not aware that these have been used in mainstream uid dynamics.
2.2. EULERIAN VS. LAGRANGIAN DESCRIPTIONS 43
continua: this idea will be expanded in the next Chapter.
Consider a trac ow, and you are seated in one particular car, which is
assimilated to a point particle. As the car moves and stops, it experiences
forces that change its speed and direction. The point of application of the
force changes with the location of the car, and the associated accelerations
are those of your car. This is the Lagrangian description of the trac ow,
very close in spirit to Newtons initial formulation. It deals with the motion
of material points.
An alternative, called Eulerian description, looks at the trac ow from
the viewpoint of the bystander: motion at a geometrical point regardless
of which particle is moving by. If you stand at a trac light, the vehicles
will experience periodic accelerations and decelerations, as the light turns
from red to green and back again. This periodicity, a dominant feature
of what happens at this intersection, is not experienced in the Lagrangian
description. Thus, the two descriptions will give rather dierent views of
the motion of trac. Steady or periodic ow are Eulerian properties.
The only steady ow in the Lagrangian description is uniform, i.e. at rest
is a suitable frame of reference, and of no practical interest. Mathematical
description and display of results are equivalent but show a dierent aspect of
the same reality. The choice of formulation may be dictated by convenience
or habit, but the results must be interpreted accordingly and converted if
necessary: experimental ow visualization is naturally Lagrangian, whereas
analytical and/or numerical solutions yield Eulerian properties rst. In the
case of trac ow, it would be necessary to know the state of (Lagrangian)
motion of all vehicles at various times in order to reconstruct the Eulerian
motion at a given point; and conversely. The purpose of this section is to
develop the required tools and to provide illustrations of the main steps;
only the simplest cases can be treated in closed form, but the procedure
is identical for the numerical processing of more complex experimental or
numerical data.
2.2.1 Lagrangian description of motion
For the purposes of presentation, we will focus on the description of a ve-
locity eld; appropriate changes can be made for other properties such as
acceleration, pressure, temperature, etc. In the Lagrangian description,
u(t, x
0
, t
0
) =
d
dt
x(t, x
0
, t
0
) (2.32)
Figure 2.3: Lagrangian, Eulerian, experimental, numerical
describes the velocity of the particle that was at location x
0
at time t = t
0
and is now at x(t). Notations such as x
0
reinforce this idea, but can be
cumbersome; as we omit some details, the student should retain the implied
dependences in their thinking. We can also write
x
x
0
dx = x x
0
=
t
0
u(, x
0
, t
0
)d (2.33)
From elementary mechanics, we know that, as time t varies, this is the tra-
jectory of the particle, in parametric form as time changes. Elimination of
time between the components of the trajectory gives one (2D) or two (3D)
equations for the trajectory components in terms of each other. Its calcula-
tion requires the knowledge of the history of velocity for this particle. In the
example of the trac ow, knowing history of your velocity enables you to
evaluate your current location relative to the point of origin. This is true for
each vector component. Mapping ocean currents can be done by tracking the
trajectory of a buoy. In uid mechanics, a pathline is equivalent terminology
for trajectory, and its reference to a particle makes it a Lagrangian concept.
Experimentally, you might mark one particle to be very bright, and take a
very long exposure that will record its trajectory: if it is a uid particle, you
have generated a pathline (but if it is a rey, you only have a trajectory of
the y, which is not a picture of the motion of the air).
A dierent concept is generated by the gradual release of particles (e.g.
smoke or dye) from one point (Fig. 2.5). The resulting line is not a pathline,
but a streakline (and they coincide if the trajectories are independent of the
time of release t
0
). Its equation should make reference to the release from a
given point x
0
at successive instants in the past
x = x
0
+
t
t
0
u(, x
0
, t
0
)d (2.34)
Here, the parameter along the streakline is the release time t
0
, with the
material points closer to x
0
having been released most recently. We note
that the equations for pathlines and streaklines are sections through the
same family of solutions, with dierent parameters being held xed. The
streakline is commonly achieved by the release of dye, bubbles, smoke or
other particles as markers of the uid particles (question: are they good
markers?)
To illustrate all this, use the ow around a sphere, in the frame of reference
of the sphere and for a sphere in free fall, from Trittons book. See also
example below, and in Curries book.
Figure 2.4: Particle pathline
Figure 2.5: Streaklines
S. Klines movie: Flow vizualization
Hydrogen bubbles: water electrolysis
Bubbles: good tracers?
Lagrangian or Eulerian?
Types of lines
Path lines
Streaklines
Time lines
Streamlines?
Steady / unsteady
Other ow markers: dyes, particles
Also note:
Break of symmetry in cylinder wake, in diuser
Instability in free shear layers (lifting airfoil wake, etc)
Figure 2.6: S. Klines movie on ow visualization: what do we really see?
Figure 2.7: Eulerian velocity vectors from particle traces
2.2.2 Eulerian description
It is not uncommon for experimentalists to observe many pieces of trajectories
at once: seeding many bright particles in a ow, and taking an exposure of
short but nite time , the collection of vectors (for many initial locations
x
0
is
x x
0
=
t+dt
t
u(t, x
0
, )d u(t, x
0
, t)dt (2.35)
for each x
0
gives a nite dierence approximation of the velocity at each
point, leading to the Eulerian description of the eld (Fig. 2.7). Note that
time t denotes the running time as well as the release time.
In the Eulerian description, the emphasis shifts away from the individual
particles (still necessary to assign a velocity!) in favor of the distribution of
values (e.g. or velocity) in space, i.e. the velocity eld. This is the approach
taken in PIV and related methods.
Given an experimental sample such as on the gure, the eye (actually:
our brains, processing visual information and superimposing patterns) in-
terpolates lines that are at every point tangent to the velocity vectors: the
streamlines. In the plane, the equation of streamlines, given a velocity eld
in Cartesian coordinates, is easily constructed: at every point, the slope of
Figure 2.8: Direction of a streamline
the line must match the direction of the velocity vector
dy
dx
|
streamline
=
v
u
(2.36)
In 3D, a line requires 2 equations: it can be seen that the streamline is
determined by
dx
u
=
dy
v
=
dz
w
(2.37)
With the velocity vector tangent at each point to pathlines and to stream-
lines, the distinction between the two types of lines is emphasized. For the
streamline, the direction of the line results from a snapshot: simultaneous
velocity vectors at each point. For the pathline, we have the history of mo-
tion of a particle. Once again, for steady ow, they coincide; in unsteady
ow, they can be unrecognizable from each other (Tritton). When comparing
experimental and numerical results, this can complicate the interpretation.
2.2.3 Pathlines, streaklines and streamlines
The conversion from Lagrangian to Eulerian (and conversely) is a useful
exercise. An example is given in Currie (p38-42). Here, we will treat a few
examples of increasing complexity. Students should carefully mark what is
general and what is specic to the problem at hand, and not use relations out
of context; this is also a cue about brushing up on dierential equations and
elementary calculus. The pivotal item is velocity: the Eulerian velocity is
also the velocity of a particle that happens to be at the point of observation.
Example 1: Steady ow
Consider the velocity eld (Eulerian: why?) given by the expressions
u = xy
v = y. (2.38)
which is independent of time. The equations of the (Eulerian: why?) stream-
lines is easily obtained
dx
u
=
dx
xy
=
dy
y
(2.39)
and are readily integrated:
y = y
0
+ ln(
x
x
0
) (2.40)
Switching to the Lagrangian concepts requires an important change: the
particle trajectories are generated by updating their Eulerian locations to
match the location of each material point. Hence
u =
dx
dt
= x(t)y(t)
v =
dy
dt
= y(t). (2.41)
Because the y-equation can be integrated separately, this problem is relatively
simple. We get
y = y
0
e
tt
0
(2.42)
which is substituted into the x-equation, and gives
ln(
x
x
0
) = y
0
(e
tt
0
1). (2.43)
There are two parameters (t and t
0
) for the location of a particle associ-
ated with x
0
, y
0
. If we eliminate t, we obtain the trajectory
y = y
0
+ ln(
x
x
0
) (2.44)
identical to the streamlines through the same point. If instead we eliminate
t
0
, thus obtaining the equation of the streakline through the same point at
the current time t, the result is again the same, a feature of steady ows.
The distinction between streamlines, pathlines and streaklines can be
subtle: the correct interpretation of the relations depends on identifying the
relevant parameters.
Example 2: Unsteady ow
This distinction is highlighted in unsteady ows, where the three types of
lines are usually dierent. We start from the Eulerian eld
u = x sin(t)
v = y (2.45)
Within the Eulerian framework, we obtain the streamlines
ln(
x
x
0
) = sin(t) ln(
y
y
0
), (2.46)
which vary in time.
Switching to the Lagrangian framework, we have for the particle motion
dx
dt
= x(t) sin(t)
dy
dt
= y(t) (2.47)
The two variables are separated, allowing for simple solutions:
ln(
x
x
0
) = cos(t
0
) cos(t)
ln(
y
y
0
) = t t
0
(2.48)
Two families of lines are included here: if we eliminate t, we have the path-
lines x(y | x
0
, y
0
, t
0
):
ln(
x
x
0
) = cos(t
0
) cos(t
0
+ ln(
y
y
0
)); (2.49)
whereas if we eliminate t
0
, we get the streaklines
ln(
x
x
0
) = cos(t) + cos(t ln(
y
y
0
)); (2.50)
The 3 families of lines (streamlines, pathlines and streaklines) are illus-
trated on Fig. 2.9
0 1 2 3 4 5 6 7 8 9 10
0
1
2
3
4
5
6
7
8
9
10
Figure 2.9: Streamlines (solid lines) at times t = 0 (black), /4 (red), /3
(magenta) and /2 (blue); Pathlines (dashed) through the point (1, 1) at
times t
0
= 0, /4, /3 and /2; Streaklines (dotted) through the point (1, 1)
at times t = 0, /4, /3 and /2.
Figure 2.10: The oscillating hose
Example 3: Unsteady ow
In this example, we go from Lagrangian to Eulerian. In an approximation
of an oscillating garden hose (Fig. 2.10), the water motion is limited to
the horizontal x y plane, ignoring the eect of gravity. At time t
0
, the
water leaves the nozzle (origin of coordinates) a constant speed U, but at an
oscillating angle =
0
sin t
0
. If the angular amplitude
0
of oscillations is
small, we have (at rst order in
0
)
u
0
(t
0
) = U (2.51)
and
v
0
(t
0
) = U
0
sin(t
0
) (2.52)
as the initial conditions for the trajectory of a drop leaving the nozzle at time
t
0
.
In the absence of applied forces, the trajectory of a drop is easily calcu-
lated:
x(t, t
0
) = 0 + u
0
(t
0
)(t t
0
) (2.53)
and
y(t, t
0
) = 0 + v
0
(t
0
)(t t
0
). (2.54)
This is the parametric representation of pathlines for this particular problem,
with t as the parameter and t
0
as an additional label (note the dierent role
played by t and t
0
below). We can eliminate the parameter t easily:
t t
0
=
x
u
0
=
y
v
0
(2.55)
or
y =
v
0
u
0
x = x
0
sin(t
0
) (2.56)
The pathlines are straight lines issuing from the origin, with a slope depend-
ing on t
0
.
Streaklines are more dicult to obtain: the collection of drops that left
the nozzle at dierent times in the past. This requires the elimination of t
0
from the individual trajectories. We have
x = (t t
0
)U t
0
= t
x
U
(2.57)
which is substituted in the equation for y
y = x
0
(t
x
U
)sin((t
x
U
)). (2.58)
This is the equation of streaklines at a given time t.
Note which substitution changes the formulae from a particle trajectory
into a streakline! Obtaining the streakline in the form y(x; t) could be quite
dicult without the initial linearization. Yet, taking a snapshot of the water
drops would achieve the same result. This illustrates how seemingly simple
experimental results can be analytically dicult to reproduce. The concep-
tual key is in the role of the various parameters.
Finally, the streamlines are obtained easily: at any instant, the velocity
vector is radial from the origin, so that the streamlines are indistinguishable
from the pathlines:
dx
u
=
dy
v
(2.59)
gives again
y =
v
u
x (2.60)
The superposition of pathlines and streamlines arises because the Eulerian
velocity eld is independent of time. Similarly, if the Lagrangian velocity is
time-independent, the pathlines and streaklines coincide. All three lines are
identical in steady ow, as seen above.
Figure 2.11: Pathlines, streaklines and streamlines: the parameters
Stagnation point...
x
y
O
S
3 2 1 0 1 2 3
3
2
1
0
1
2
3
...is not Galileaninvariant
x
yO
S
3 2 1 0 1 2 3
3
2
1
0
1
2
3
Figure 2.12: Stagnation point (rst frame) is modied by the superposition
of a uniform speed (second frame). Streamlines are blue.
2.2.4 Caution!
The streamlines, pathlines and streaklines are very dependent on the frame
of observation. From a moving reference frame even in uniform motion
(Galilean invariance: see Ch. 10 for rotating systems with ctitious forces),
the appearance of the lines can be dramatically aected.
For example, consider the simple potential ow (See Ch. 5) with a stag-
nation point at the origin. From a reference point moving at uniform velocity
at 45 degrees to the axes, the addition of the corresponding velocity changes
the streamlines as shown on Fig 2.12. The problem of describing ow topol-
ogy in a way that is independent of the point of observation goes beyond the
scope of this course.
2.3 Dierential concepts and their geometry
2.3.1 Dierentials and vectors
The gradient of a function f requires no explanation at this level. It is a
vector (direction, magnitude), represented by f which has various repre-
sentations depending on the coordinate system (Cartesian, polar, etc.) In
two dimensions, think of a pressure eld p(x, y). The gradient will have a
direction normal to the isolines p = C This interpretation translates easily
2.3. DIFFERENTIAL CONCEPTS AND THEIR GEOMETRY 57
Figure 2.13: Gradient is normal to isolines of a function
to 3D. Turn your attention to one component, e.g. in the y-direction, of this
gradient: the projection of the gradient on the unit vector e
y
, it measures the
rate of increase of p in the y direction. This is the basis for the denition of
the directional derivative. Given a unit vector e
in some direction denoted

by , the directional derivative of p in the -direction is naturally
p = e
p, (2.61)
or, regardless of the eld to which it is applied
= e
. (2.62)
Learn to recognize the combinations vector , whenever it occurs.
Now consider a function f(r, t). The vector position has coordinates
r = (x(t), y(t), z(t)) that move with the uid. If we look at the evolution of
f, its time derivative is obtained through the chain rule
d
t
f =
t
f +
x
fd
t
x +
y
fd
t
y +
z
fd
t
z
=
t
f + u
x
f + v
y
f + w
z
f
=
t
f + u f (2.63)
This material derivative, i.e. time derivative following a uid particle, is
sometimes denoted as D
t
for emphasis. The alert student has spotted the di-
Figure 2.14: Denition sketches for Gauss and Stokes theorems
rectional derivative uin the direction of streamlines, also called convective
derivative.
We already saw Gausss divergence theorem:
F dV =
F dA (2.64)
or, in index notations

i
F
i
dV =
F
i
dA
i
(2.65)
Another important theorem is due to Stokes
F dr =
(F) dA (2.66)
In the case where the vector eld is velocity, Stokes theorem introduces two
important concepts. Vorticity is dened as
= u. (2.67)
Vorticity will be one of dominant new concepts throughout this course. Re-
lated to vorticity by Stokes theorem is the concept of circulation
=
u dr (2.68)
2.4. MASS BALANCE 59
2.4 Mass balance
We start from the integral representation
dV +
(U dA) = 0, (2.69)
assume that the control volume is xed in space, and make use of Gauss
divergence theorem to obtain

t
dV +
(U) dV =
(
t
+ (U)) dV = 0 (2.70)
Since this holds for any xed control volume, the integrand must vanish at
every point: therefore
t
+ (U) = 0 (2.71)
Note how much simpler this derivation is, compared to the use of material
derivatives following a uid particle. Then
d
t
= 0 (2.72)
because the density changes as the particles volume does: the divergence of
velocity accounts for this.
Since we limit ourselves to the study of incompressible ows, where den-
sity can be treated as constant, the continuity equation reduces to
U = 0 or
i
u
i
= 0 (2.73)
This relation is a kinematic constraint on changes in the velocity compo-
nents. For example, as a unidirectional ow approaches stagnation point and
its velocity decreases, it must develop a component of velocity normal to the
upstream motion for mass to be conserved. It is benecial to compare the
results of control volume analysis and continuity equation in this regard.
One of the great simplications in incompressible ow, relative to com-
pressible ows, is related to the use of the simple (linear) continuity equation
just derived. It is interesting to note (proof in Ch. 3) that, in the case of
natural convection, the continuity equation can be applicable even though
the variations in density drive the motion. This non-trivial result is known
as the Boussinesq approximation.
2.4.1 Vector potential and stream function
It is known from vector calculus that any divergence-free vector can be ex-
pressed as the curl of another vector (and conversely). Therefore, whenever
the continuity equation applies, there is a vector A (called the vector poten-
tial) such that
U = A (2.74)
The vector potential will be used extensively later. Note that arbitrary con-
stants, or even the gradient of any smooth eld, can be added to each com-
ponent of A without aecting the velocity eld. This exibility is used to
ensure that A = 0 without loss of generality (gage invariance).
A particular case of considerable importance relates to the 2D ows (in
the x-y plane; axisymmetric versions can be found below). Then, inspection
of the components of A shows that only the z-component of A is non-zero.
In this case, we denote
A
z
= (x, y) (2.75)
and we have
u =
y
(2.76)
and
v =
x
. (2.77)
Substitution into the continuity equation shows that mass balance is satised.
Lines of constant are of particular interest. It so happens that the
directional derivative of along streamlines is zero:
U = u
x
+ v
y
= uv + vu = 0. (2.78)
This means that lines of constant are the streamlines. For incompressible
ow, nding the streamfunction and its contour lines is the fastest way to plot
streamlines. Alternatively, the above relation can be read as the projection
of on the velocity vector, and they are orthogonal. In other words again,
it means that there is no mass ux across lines of constant , which can
be used as impermeable boundaries as long as no viscous eects need to be
described (slip allowed along the boundary). Furthermore, we see that
d =
x
dx +
y
dy = vdx + udy = U dA (2.79)
so that the dierence in streamfunction value is proportional to the mass
ux through the streamtube. Or again, we can say that the velocity in
a streamtube is inversely proportional to its width, and velocity can only
vanish if the streamtube expands into a reservoir of innite size.
2.5. FLOW ABOUT A POINT: HELMHOLTZ 61
Figure 2.15: Streamfunction and 2D velocity
2.5 Flow about a point: Helmholtz
Consider a velocity eld (the following applies to any vector eld) u
i
(x
k
)
(the use of a free index for x
k
should convey that any velocity component
can depend on all coordinates.) In the vicinity of this point (Fig. 2.16), say
at x
k
+ x
k
, a Taylor series will show the variations in eld values. Keeping
only rst order terms, we have
u
i
(x
k
+ x
k
) = u
i
(x
k
) + x
j
j
u
i
|
x
k
+...
= u
i
|
x
k
+
1
2
x
j
(
j
u
i
+
i
u
j
) +
1
2
x
j
(
j
u
i
i
u
j
) + ...
= u
i
|
x
k
+x
j
s
ij
|
x
k
+x
j
r
ji
|
x
k
+... (2.80)
which can be interpreted as a sum of three contributions: a translation with
the point of reference, a straining motion, and a rotation: this is shown in
the next few subsections.
2.5.1 Rate of strain
With the understanding that all properties are evaluated at x
k
, we have
introduced the symmetric part of the velocity derivative
s
ij
=
1
2
(
j
u
i
+
i
u
j
) (2.81)
Figure 2.16: Flow about a point
which is called the rate of strain. (This is similar to the strain in solid me-
chanics, except this applies to velocities instead of displacements.) Because
of incompressibility, its trace vanishes
s
ii
= 0 (2.82)
The eigenvalues and eigenvectors of a matrix are scalars s and unit vectors
e
i
, respectively, such that
s
ij
e
j(k)
= s
(k)
e
i(k)
(2.83)
The parenthesis indicates that k is a label distinct from the indices and
not subject to implied summation. Interpreting the denition, applying the
matrix to the vector does not change its direction, but can change its mag-
nitude by a factor s. A 3x3 matrix has 3 such combinations of directions
and stretching, some of which can be complex-valued. But since s
ij
is real
and symmetric, we know from linear algebra that its eigenvalues and eigen-
vectors are all real. Combining this with the vanishing trace, which is the
sum of the eigenvalues, we see that at least one eigenvalue of rate of strain
is positive, corresponding to an extensional axis, and at least one value is
negative, corresponding to a compression axis, while the middle eigenvalue
adjusts the sum of all eigenvalues to be zero.
6
?
L
-
U
-
-
-
-
e
(1)
@
@
@
@I
e
(2)
Figure 2.17: Couette ow and principal axes
The eigenvectors are orthogonal to each other and can be adopted as local
coordinates: they are called principal axes. In such a coordinate system, the
matrix is diagonal, i.e.
s
ij
=
ij
s
(i)
(2.84)
Example: Couette ow
In the Couette ow shown on the gure, the velocity eld is given by
u(x, y) =
U
L
y and v(x, y) = 0 (2.85)
You can verify that this 2D ow is incompressible. The successive matrices
is easily constructed from their denitions:
(
j
u
i
) =
U
L
(
0 1
0 0
)
(2.86)
Then
(s
ij
) =
U
2L
(
0 1
1 0
)
(2.87)
It is then a simple matter to obtain the two eigenvalues
s =
U
2L
(2.88)
corresponding to the eigenvectors
e =
1
2
(
1
1
)
(2.89)
Rate of strain takes a square (cube in 3D) with diagonals along the princi-
pal axes and makes it into a diamond with its long diagonal on the extensional
axis. For innitesimal stretching, this conserves area (incompressible).
2.5.2 Local rotation and vorticity
The antisymmetric part of the velocity derivative is
r
ij
=
1
2
(
i
u
j
j
u
i
), (2.90)
in which only 3 components are independent. r
ij
could therefore be arranged
as a vector. In fact, it can be seen that
r
ij
=
1
2
ijk
k
(2.91)
where
i
=
ijk
j
u
k
or = U (2.92)
is the vorticity vector. The corresponding contribution to the motion around
a point (Helmholtz above) is then written as
u
i
(x
k
+ x
k
) = u
i
|
x
k
+x
j
s
ij
|
x
k
+
1
2
ijk
j
|
x
k
x
k
(2.93)
In the +
1
2
dr term, we recognize the expression for rigid-body rotation
around the (local) origin, with angular velocity
1
2
. Pay attention to the
factors 1/2 in the rate-of-strain and rotation rate; pay attention to signs,
they matter.
Example: Couette ow
Pursuing the example of the Couette ow, the vorticity vector is seen to have
only one component normal to the plane of the ow, with magnitude
=
y
u =
U
L
(2.94)
uniform throughout the ow eld. Note that vorticity and the implied rota-
tion do not imply curved streamlines or trajectories! It is about the relative
motion of nearby points.
A.H. Shapiros movie: Vorticity
Kinematics
Denition, vorticity meter
Potential vortex
Vorticity and circulation: Stokes theorem
Start-up vortex
Dynamics
Bernoullis equation: conditions for B = p +
1
2
V
2
+g z to
be constant
Croccos equation: how B changes
Kelvins theorem: inviscid, Lagrangian, vortex stretching
Potential ow, singular vortices
Model ( = 0)
Motion of vortex pair, induced velocity
Helmholtzs theorems (material lines, vortex tubes, vortex lines
end only at boundary)
How to introduce vorticity
Singularities
p and interface
unsteadiness
rotational forces
pressure gradient along boundary (viscous eect)
See Ch. 9
Figure 2.18: Kinematics and dynamics of vorticity and circulation, A.
Shapiros movie.
Figure 2.19: Rotation in Couette ow
2.6 Kinematic decompositions of a velocity
eld
Also due to Helmholtz, the kinematic decomposition of velocity is frequently
used:
u = u
+ u
(2.95)
The rotational part u
is obtained from the vorticity distribution through the

Biot-Savart relation (below). Then, boundary conditions must be adjusted
by the uniquely dened potential ow u
(see Ch. 5). Helmholtz proved

that such a decomposition is always possible for incompressible ow. Hodge
further proved that the decomposition is unique, and that the potential and
rotational components are orthogonal in a mathematical sense (see Chorin
& Marsden, or Majda and Bertozzi).
For an alternative decomposition of velocity, due to Clebsch, see e.g.
Panton Section 14.4 p.355.
2.7. VORTICITY, ,
2
, ETC. 67
2.7 Vorticity, ,
2
, etc.
2.7.1 Biot-Savart relation
The Biot-Savart relation rst came up in relation with the magnetic eld
induced by a current: it turns out the equations are identical to ours, with
vorticity instead of current and velocity instead of magnetic induction. The
most common form for 3-D elds is (e.g. Panton Section 14.2 p.351)
u
=
1
4
r
3
dV
. (2.96)
Here,
r =| x x
| (2.97)
is the distance between the vorticity element at x and the point x where it
induces velocity. (Note that f(r) =
f(r) for any function of r =| r |

2
).
The velocity eld can be rewritten in equivalent forms, by integration
by parts, provided the boundary terms vanish fast enough at innity (see
Batchelor, section 2.9 for the eect of boundaries). First we get
u
=
1
4

1
r

dV
=
1
4
r
dV
= A (2.98)
where we recognize the vector potential for which
2
A = . Also, inte-
grating by parts
u
=
1
4
1
r

dV
=
1
4
1
r
dV
=
1
4
r
dV
. (2.99)
Here
=
2
u = (2.100)
2
In these notes, the notation r will be used for the radial distance in po-
lar/spherical/cylindrical coordinates, or for the distance between two points: the student
should pay attention to the context.
denes exion. So, the Biot-Savart formula reconstructs velocity from
its Laplacian (for incompressible elds), or the velocity potential from its
Laplacian: it gives one explicit form of the inverse Laplacian, since
u
=
2
. (2.101)
This allows us to view the Biot-Savart integration kernels
G(x, x
) =
1
4r
(3D) and
ln r
2
(2D) (2.102)
as Greens functions for the Poisson equation. The Poisson equation is of
the form
2
f = s, (2.103)
where s is a source term. Then, Greens function gives
f =
G(x, x
)s
dx
(2.104)
the solution f in terms of the superposition of eects of localized source terms
distributed throughout the eld. Greens function is solution of the equation
2
G(x, x
) = (x
), (2.105)
where (x) is Diracs distribution.
For example, a 2D potential vortex (see Ch. 5 for details) is given by
(x, y) = (x, y) (2.106)
which is innite at the origin, zero everywhere else, and integrates to 1. You
can think of the 2D version of as the limit of a normalized Gaussian bell
shape of unit integral as its scale becomes smaller (and its magnitude larger):
(x, y) = lim
r0
1
4r
e
x
2
+y
2
4r
2
(2.107)
Then the streamfunction is the z component of the vector potential (
2
=
), so that we reconstruct by using Greens function:
(x, y) =
1
2
ln r (x x
, y y
) dx
dy
=
1
2
ln
x
2
+ y
2
(2.108)
2.7. VORTICITY, ,
2
, ETC. 69
(avoid confusion with 2 expressions of r: distance between x and x or polar
distance...)
Greens function (and the Biot-Savart relation in particular) introduces
remote eect of sources in the solution to a problem.While the use of Greens
function to solve problems goes beyond the scope of this course, it is impor-
tant to realize the non-local properties it introduces in the ows. Vorticity
induces velocity at large distances because of the non-local dependence of
solutions of the Laplace and Poisson equations; other examples include the
pressure (Ch. 3), and eects of diusion (Ch. 8).
2.7.2 Vorticity, streamfunctions, and more
As seen above, it is a mathematical fact that
u = 0 u = A (2.109)
It can also be shown that there is no loss of generality in selecting the vector
potential to be incompressible (gage invariance).
3
Then, the vector identity
() =
2
() (2.110)
(for incompressible vector elds only!) (check the general relation in your
favorite reference book). In particular:
2
A = (2.111)
In conjunction with the Biot-Savart equation, this is the motif in a pattern
of relations between
vectors: the vector potential A, velocity U, vorticity , and exion
=
2
U
dierential operators: the curl , the Laplacian
2
, and their inverses
The pattern is summarized on the gure. It is good to remember that the
inverse curl and inverse Laplacian are non-local, i.e. they induce eld prop-
erties far away from the sources.
3
Suppose that A is a vector potential, with A = b = 0. Adding a gradient does not
change the resulting velocity eld, so we can consider any A
= A+B as equivalent. We
then adjust B so that A
= b +
2
B = 0, i.e. B =
2
b, a familiar operation.
Figure 2.20: Kinematic relations
2.7. VORTICITY, ,
2
, ETC. 71
2.7.3 Vorticity and boundary conditions
Consider a conguration with solid boundaries, and a certain distribution
of vorticity (respectively, exion). The Biot-Savart relation gives an induced
velocity eld, which will not, in general, satisfy the boundary conditions. Let
us call U the error on the velocity at the boundary. If we add to the velocity
eld a potential component u
(resp.: a component of zero Laplacian) which

has value U at the boundary, we have a well dened problem of solving
the Laplace equation with Neumann (resp., Dirichlet) boundary conditions,
and a unique solution with zero vorticity (resp., exion). Thus, the addition
of a potential component is related to the presence of boundaries and the
enforcement of boundary conditions.
Part of Ch. 6 (Potential ows) could actually be moved here: assuming
irrotational ow (zero vorticity) is a kinematic condition. That corresponding
material is conventionally coupled with inviscid ow, and this Chapter is
bulky enough as it is.
2.7.4 Discrete vortices vs. continuous vorticity
This section can be omitted without loss of continuity in the text. It has a
bearing on vortex-based numerical methods, as well as on various modeling
approaches in turbulence (see Samans book for vorticity, Chorins book for
applications to turbulence). The point is made here in 2D, the same applies
in 3D.
Consider a collection of point vortices
i
=
i
(x
i
) (2.112)
The overall streamfunction is obtained by summation
=
i
2
(x
i
) ln r dA
i
2
ln r
i
=
i
(2.113)
which obeys the Laplace equation except at the singular points. So the
induced velocity eld is irrotational.
The same is not true, in general, for continuous distributions of vorticity
(x) over a nite region. Indeed, for
=
(x
)
2
ln r dA
, (2.114)
the vorticity is
=
2
=
(x
)
2

2
ln r dA
, (2.115)
which is, in general, non-zero even in the region where Gamma vanishes. So,
for continuous distributions of vorticity, the induced velocity is not irrota-
tional.
2.7.5 Vorticity and circulation
Vorticity and circulation are not equivalent concepts, but are obviously re-
lated through Stokes theorem:
=
dA (2.116)
where the integral is over any simple surface supported by a closed contour.
We can also write
=
av
n
A (2.117)
for some suitable average vorticity normal to the surface. This latter presen-
tation is most useful in plane (2D) ows, since the contour would normally
be in plane of the ow, and the surface would normally be the same plane
normal to the vorticity vector. The average is not necessarily representative,
as the potential vortex makes clear. In 3D, it is good to remember that
n
can change sign over the surface.
It should be clear at this point that, without any dynamics involved yet
(and more kinematics coming up), vorticity is related to many other concepts:
it could be argued that it is central to uid mechanics. At this point, the
web of relations can be summarized (students: to build on this as the course
goes on!)
2.7.6 The Helmholtz theorems
The Helmholtz theorems on vorticity (see Panton Section 13.8 p.338, Acheson
Section 5.3 p.162) are very important.
1. vortex lines cannot end in the uid, they can only end at a boundary.
2. in inviscid incompressible ow, uid elements that lie on a vortex line
remain on vortex line (this means that vorticity is a marker just as
dye).
2.7. VORTICITY, ,
2
, ETC. 73
Figure 2.21: Vorticity at the center of its web
3. circulation is the same for all sections of a vortex tube.
Theorem 2 relies on dynamical considerations, and we will return to it in Ch.
5. Theorems 1 and 3 have a counterpart for streamlines: the key element is
that the eld (vorticity, velocity, whatever) is incompressible. The proofs are
based on the divergence theorem.
2.7.7 Helicity, Lamb vector
Two combination of velocity and vorticity can be introduced here. First,
helicity is dened as
h =
i
u
i
. (2.118)
Since it vanishes identically in 2-D ows (where vorticity is perpendicular to
the plane of the ow), it is used as an indicator of 3-dimensionality, useful in
turbulence. Some care needs to be taken about its lack of Galilean invariance.
Also important (see next chapter) is the Lamb vector, dened as
i
=
ijk
j
u
k
(2.119)
is perpendicular to both u and . See Ch. 3 for the role it plays in
Dynamics.
2.7.8 Famous vortices
This subsection is a list of classic vortices. The dynamical equations of which
some of them are solutions, will be presented in later chapters: for now, a
simple description is sucient.
2D potential vortex.
See also Ch. 5
= (x, y) (2.120)
=

2
ln r (2.121)
v
=

2
1
r
(2.122)
2.7. VORTICITY, ,
2
, ETC. 75
Figure 2.22: Rankine vortex
Rankine vortex
Also 2-dimensional (plane ow), the Rankine vortex is a simple model cor-
responding to a potential ow surrounding a core of nite size, in which
solid-body rotation reects the dominance of viscous forces (Fig. 2.22).
=

R
2
for r R
= 0 for r > R (2.123)
v
= r for r R
=
R
2
r
for r R (2.124)
The model can be useful in some applications in spite of the abrupt transition
between the viscous core and the induced potential ow This is not quite
realistic, but is promptly smoothed by viscous diusion.
Oseen vortex
See Panton Section 11.6 p.283. The Oseen vortex is an exact 2-D unsteady
solution to the equations of motion, with the axisymmetry eliminating the
nonlinearities. The vortex results from the viscous relaxation of a (singular)
potential vortex (see Ch. 5). The vorticity distribution is Gaussian (Fig.
Figure 2.23: Oseen vortex and asymptotes
2.23), and the corresponding tangential velocity (polar coordinates, of course)
is
v
=

2r
(1 e
r
2
4t
) (2.125)
It should be noted that the size of the vortex is time-dependent; it is also of
interest to note the limits or solid-body rotation and potential ow for very
small and large r, respectively. It can be shown (end-of-chapter problem) that
the exponential distribution of vorticity yields the correct induced velocity
for the 2-D problem; however, 3-D congurations based on distortions of the
Oseen vortex are not likely to be kinematically possible.
Burgers 3D vortex
Burgers vortex is one of those rare solutions of the exact equations that
combines nonlinear and viscous terms in a stationary solution. A vortex core
along one axis is stretched by an axisymmetric rate-of strain (Fig. 2.24).
See Acheson Section 5.9 p.187.
Hills spherical vortex
Also an exact solution (Panton Section 13.6 p.332), but with pressure jump
at the interface (see Samans book).
2.7. VORTICITY, ,
2
, ETC. 77
Figure 2.24: Burgers vortex
Figure 2.25: Hill vortex
Figure 2.26: Ring vortex
Ring vortices
See back cover of Van Dykes book (Fig. 2.26). Stability analysis by Widnall
and others.
Taylor vortices
Result of ow instability, for the Couette ow between coaxial cylinders
with the inner cylinder spinning. As viscous forces set the uid in mo-
tion (Poiseuille ow), centrifugal forces will tend to expel the moving uid
to the outside, with the motionless uid driven inward by pressure forces.
Viscous forces will tend to oppose such motion at low angular speeds, but
for suciently rapid rotation a convective pattern develops, made of a stack
of donut-shaped vortices rotating in alternating directions (Fig. 2.27). This
case of ow instability (see Ch. 11) was rst studied by G.I. Taylor.
See Tritton Section 17.5 p 258-267, Van Dyke p76.
Kàrmàn vortices
Counter-rotating staggered vortices form as the result of the boundary layer
separation from alternate sides of blu bodies in 2D ows (Fig. 2.28). The
telltale pattern is a classic example of ow-structure interaction: oscillating
forces are applied to the obstacle as well. For a detailed analysis, see e.g.
Samans book.
2.7. VORTICITY, ,
2
, ETC. 79
Figure 2.27: Taylor vortices
Figure 2.28: Kàrmàn vortex street
Figure 2.29: Necklace or horseshoe vortex
Necklace or horseshoe vortices
When an obstacle protrudes normally from the wall in a boundary layer,
vortex lines bunch up near the upstream stagnation wall and are stretched
downstream on both sides of the obstacle (Fig. 2.29). The resulting ow
patterns, also called horseshoe vortices, can aect bridge pilings, aircraft
wings and engine pilons, and many similar congurations.
2.8 Advanced topics and ideas for further read-
ing
This chapter contains a lot of information, and will require persistence from
the student. Every later use of the notations and concepts developed here
should be an opportunity to review this material, to relearn it.
It might be a good idea to review linear algebra for principal axes; see your
advanced math class (or standard references) for more on Greens functions:
the important here is to understand the general idea behind them.
Kinematics does not usually provide solutions to problems: to be a solu-
tion, a ow conguration should also satisfy Newtons second law, the object
of the next Chapter. But kinematics establishes constraints that a dynamical
solution must satisfy.
Stokes streamfunction applies to axisymmetric ows. See e.g. Currie,
2.8. ADVANCED TOPICS AND IDEAS FOR FURTHER READING 81
p.145, for relevant equations.
Problems
1. Express the velocity eld for Couette ow in principal axes, and calcu-
late the rate of strain, eigenvalues and eigenvectors in this coordinate
system.
2. Using em only index notations, show the following identities
(F G) = G (F) F (G)
(F G) = (G )F (F )G + F( G) G( F)
3. Consider the potential ow for deep-water waves:
= Ce
ky
sin(kxt) with the ow eld u =
x
and v =
y
. Is this
Eulerian or Lagrangian? Derive the equations of streamlines, pathlines
and streaklines: carry out the analysis as far as possible, then restrict
it to motion of small amplitude to complete the problem. Compare the
solution to the ow vizualization in Van Dyke book p. 110-111.
4. By integration, construct the streamlines corresponding to the velocity
eld u = y, v = x and w = w
0
. Describe their shape.
5. Consider the Oseen 2-D vortex; using Maple or Mathematica for the
manipulations, calculate the induced velocity by integration, then take
the curl and compare the induced vorticity with the original source.
6. (Adapted from Problem 23 p.475 from Tritton.) Consider one of the
following expressions:
(a) = c (x
2
+ y
2
)
(b) = c (x
2
y
2
)
(c) = c x y
(d) = c
x
2
+ y
2
(e) =
c
x
2
+y
2
(f) = c (x
2
+ y
2
)
(g) = c (x
2
y
2
)
(h) = c x y
(i) =
c
x
2
+y
2
(j) = c
x
2
+ y
2
calculate corresponding velocity components and list the implied as-
sumptions; verify if the ow is incompressible and/or irrotational; and
give the analytical expressions and plots of streamlines and potential
lines, if they exist, or explain why they do not exist.
7. Consider a 2-D ow eld with velocity components u = 1, v = u f(t)
for some function f. Find the equations of the streamlines, pathlines
and streaklines with the relevant parameters. (Adapted from Currie,
p.48.)
8. Show that streamlines and pathlines coincide for the unstready ow
u
i
= x
i
/f(t). Are the streaklines dierent? (Adapted from Currie,
p.48.)
9. (Adapted from Tritton, problem 10 p.470.) You are given a Lagrangian
representation of motion as x = x
0
e
2t/s
, y = y
0
e
t/s
, z = z
0
e
t/s
. (s is a
positive constant, other notations self-explanatory.)
(a) write the equations for the path of a particle, and plot them
(b) write the Eulerian equations for this motion
(c) is the ow steady? incompressible?
(d) write the equations of streamlines, and plot them
10. (Adapted from Tritton, problem 10 p.470.) You are given a Lagrangian
representation of motion as x = x
0
e
2t/s
, y = y
0
(1+t/s)
2
, z = z
0
e
t/s
(1+
t/s)
2
. (s is a positive constant, other notations self-explanatory.)
(a) write the equations for the path of a particle, and plot them
(b) write the Eulerian equations for this motion
(c) is the ow steady? incompressible?
(d) write the equations of streamlines, and plot them
11. (Adapted from Tritton, problem 9 p.470.) You are given the equa-
tions of motion u = Ay, v = Ax and w = B (A and B are positive
constants, other notations self-explanatory.) Is this Lagrangian or Eu-
lerian? Is the ow steady? incompressible? Write the equations of
pathlines and streamlines, and describe them.
12. (Adapted from Tritton, problem 11 p.471.) You are given the equations
describing a ow as u = u
0
, v = v
0
cos(x t) (u
0
, v
0
, and are
constants.) Is this Eulerian or Lagrangian? Write the equations of
pathlines and streamlines. What other questions can you answer about
this ow?
Chapter 3
Dynamics
Newtons second law of mechanics is a monumental achievement. It vali-
dates the concept of point particles and shows the relevance of forces; maybe
more importantly, it brings mathematics (calculus) to the center of our un-
derstanding of mechanical phenomena (a program initiated by Archimedes
of Syracuse in his palimpsest, and evolving through much of modern mathe-
matical physics).
Essential mathematical concepts are vectors and dierentials; additional
mechanical ideas include stresses (in common with solid mechanics, under
the umbrella of continuum mechanics).
3.1 Newtonian dynamics of continua
The extension of Newtons second law from point particles to continua is
basis for both elasticity theory and advanced uid dynamics. Euler took
the rst step in this direction, by combining the use of material derivatives
with the knowledge (from uid statics) that a pressure gradient is an internal
force. Dividing by the uniform density in an incompressible ow
1
t
u + u u =
1
p (3.1)
1
The distinction between an incompressible ow, in which there are no signicant eects
of compressibility, and the ow of an incompressible uid, is important. Even water is not
incompressible, and a study of incompressible uids would run head-on into the second
law of thermodynamics; on the other hand, the compressibility of air can be ignored in
low-speed aerodynamics: the weaker assumption of incompressible ow is sucient to
allow for simple solutions. See Section 3.7 for details.
85
86 CHAPTER 3. DYNAMICS
Figure 3.1: Free-body diagram of an elementary volume, and a simpler 2D
variant used in the analysis
or
t
u
i
+ u
j
j
u
i
=
1
i
p (3.2)
Eulers equation represents a leap away from material particles to the velocity
eld in space, already discussed in the previous chapter. In hindsight, Eulers
equation is seen as ignoring the eects of viscosity: an approximation.
The next step was the work of Cauchy, who adopted as the basic object of
analysis the collection of particles included in a volume element specied in
some Cartesian coordinates as dV = dx dy dz. The Lagrangian description is
implied for now. The free-body diagram (Fig. 3.1) entails body forces (such
as gravity, or the Lorentz force) applied at the center of mass, and surface
forces, applied at the center of each surface, that account for interactions
with the neighboring elements.
The body and surface forces can be decomposed into their Cartesian
components, normalized by the size of the volume or surface to which they
are applied. In the case of surface forces, taking the outward normal as one
of the local axes, the normal component for force per unit area is a familiar
concept: negative pressure
2
. The tangential forces per unit area on each
surface are the components of stress, one of the cornerstones of continuum
2
The negative sign comes from the simple consideration that the force in the equation
is the force applied to the uid element, rather than by the uid to the surface at which
it is measured.
3.1. NEWTONIAN DYNAMICS OF CONTINUA 87
mechanics. Note the direction of local axes to ensure right-handed local
axes on each surface element, which is specied by its direction and its area;
for each possible direction, there are 3 components of force. Thus there
are 9 components of stress at each point. A vector such as position has 3
components, that can be arranged as a vector for which the single index can
have 3 possible values; whereas the stress requires two such indices, one for
the direction of the surface on which the stress is acting, and one for the
component of force. Thus the stress will be represented as a 3x3 matrix,
with intrinsic, component and index representations respectively as
T or
T
xx
T
yx
T
zx
T
xy
T
yy
T
zy
T
xz
T
yz
T
zz
or T
ij
(3.3)
The rst index denotes the direction of the surface, the second index the
direction of the force. Normal forces (per unit area) are therefore the diagonal
elements of stress. The resultant force per unit area on a surface with a given
direction n or n
i
is given by Cauchys relation
F = n T (3.4)
or
(
F
x
F
y
F
z
)
=
(
n
x
n
y
n
z
)
T
xx
T
yx
T
zx
T
xy
T
yy
T
zy
T
xz
T
yz
T
zz
(3.5)
or
F
i
= n
j
T
ji
(3.6)
In this framework, rotational equilibrium of the material element gives an
important result. With the body forces applied at the center of mass, their
moment about the center of mass is zero. Taking the centroidal moment of
the components of stress (see Panton p120), we obtain
T
ij
= T
ji
(3.7)
This for static conditions, no rotational body forces. The symmetry of the
stress tensor is very important in all areas of continuum mechanics.
Then, momentum balance includes the net eect of all surface and body
forces (Fig. 3.2). The derivation is carried out in 2D, with the 3D version
similar (but with more terms). Consider the rectangular volume element
Figure 3.2: The 2-D free body diagram for a uid element
centered at (x, y) and of sides dx and dy. Let us focus on the x-component
of force dF
x
:
dF
x
= B
x
dxdy + T
xx
|
x+dx/2
dy T
xx
|
xdx/2
dy
+T
yx
|
y+dy/2
dx T
yx
|
ydy/2
dx (3.8)
Expanding the various terms in Taylor series, we see that the leading terms
cancel out in pairs of opposite signs, and that the terms of order dx dy
remain. Dividing throughout by the volume gives for the x-component
F
x
= B
x
+
x
T
xx
+
y
T
yx
(3.9)
or in intrinsic notations
F = B + T (3.10)
or again
F
i
= B
i
+
j
T
ij
(3.11)
(summation over j is implied). Cauchys idea that the divergence of stress is
equivalent to a force, has implications throughout continuum mechanics.
3.2 Stress at a point, Newtonian uids
While Cauchys equation represents Newtons equation in continua, the ex-
pression for the stress tensor is not governed by fundamental laws (although
3.2. STRESS AT A POINT, NEWTONIAN FLUIDS 89
continuum mechanics imposes rational constraints on the possible forms of
the stress as function of other variables). It belongs in the group of phe-
nomenological (or constitutive) equations (or again equations of state), which
include Newtons law of viscosity, Ohms law of electrical resistance, Hookes
law of elasticity or Newtons law (again?!!) of heat exchange, and many
other empirical approximations: assuming a simple relation between shear
stress and shear rate (respectively: voltage and current, strain and deforma-
tion, heat ux and temperature dierence), a truncated Taylor series yields
a sensible result. A complete theory is provided within the framework of
continuum mechanics; only the simplest variants need to be considered here.
First, we note that pressure is independent of direction, separate the
isotropic part of the stress:
T
ij
= p
ij
+
ij
(3.12)
where (the deviatoric of stress, or viscous stress in the simple cases of in-
terest at this level) includes the non-isotropic contributions. In this instance,
we need to model the dependence of stress on ow variables. It is clear that
the viscous stress tensor cannot depend on velocity, because the addition of a
uniform velocity cannot modify the stress (Galilean invariance). The obvious
option is that the stress could depend on the velocity gradient; alternatives
include higher derivatives, time-derivatives (visco-elastic materials) or mem-
ory eects, or even non-local combinations of properties (required in some
polymer and biological ows). Here, we assume
= F(u) (3.13)
or
ij
= F
ij
(
k
u
m
) (3.14)
Because is symmetric, linear algebra dictates that it can only be a
function of the symmetric part of the velocity gradient, i.e. the rate-of-strain
ij
= F
ij
(s
km
) (3.15)
(Note that, in index notations, the dependence of the i-j-component of stress
on any or all of the k-m-components of rate-of-strain is made explicit: writing
F
ij
(s
ij
) would imply that the x-y component of stress could not depend e.g.
on s
xx
, would be quite restrictive.) Assuming that the function F can be
expanded in Taylor series, the rst few terms are
ij
= C
0
ij
+ C
1
ijkm
s
km
+ C
2
ijml
s
mk
s
kl
+ ... (3.16)
where the coecients could, in general, depend on the scalar invariants of
s
km
3
.
Because the viscous stress has no isotropic contribution, we have C
0
= 0.
We will limit ourselves in this course to the rst term in the series. The
student should be aware that relatively simple substances, such as power-
law uids and visco-elastic uids, are excluded. Also, we assume that the
coecients C
1
, ... in the series are independent of the scalar invariants of
the rate-of-strain. It turns out that these assumptions allow for accurate
descriptions of many common gases and liquids.
Then, the simplest approximation is to neglect non-isotropic stresses
ij
= 0 (3.17)
The corresponding force per unit volume is then
f
0
i
=
i
p (3.18)
which corresponds to Eulers inviscid dynamics. In intrinsic notations, we
have
T = p1 (3.19)
If terms linear in s
km
are retained, we obtain the viscous approximation
worked out rigorously by Stokes (1845) but rst included in the equations of
motion by Navier (1821). For an incompressible uid, we have
T
ij
= p
ij
+ 2s
ij
(3.20)
See Panton for the additional terms for compressibility. It is easy to see
that, in a uniform shear ow (Couette ow) between two at plates, this
expression reduces to Newtons denition of viscosity; hence the Newtonian
uid.
3.2.1 The Navier-Stokes equations
On this basis, we obtain our basic form of momentum balance (Newtons
second law) for incompressible Newtonian uids:
t
u
i
+ u
j
j
u
i
= b
i
i
p +
2
jj
u
i
(3.21)
3
First, second and third invariants of s
ij
. See a textbook on linear algebra for details.
Figure 3.3: The stress tensor
or, in intrinsic notations
t
u + (u )u = b
1
p +
2
u (3.22)
The body force per unit mass b
i
= B
i
/ is assumed to derive from a
potential from now on. Specically, for gravity
b
i
= g
i3
(3.23)
where the index 3 corresponds to the vertical upward direction. The corre-
sponding potential energy per unit mass is gx
3
.
3.2.2 Boundary conditions
Impermeable walls moving at velocity u
w
. Then, the normal velocity must
match the motion of the wall:
u n = u
w
n (3.24)
The no-slip condition is:
u n = u
w
n (3.25)
rather than merely
u = u
w
(3.26)
For an additional discussion, see Tritton Section 5.7 p.63. In addition to
the rareed gas dynamics case mentioned there, it should be noted that the
no-slip condition (introduced by Stokes) is not necessarily exact, as recent
microuidics studies have shown.
Boundaries may also be associated with the application of forces driving
the ow. Depending on the problem, it may be necessary to to impose
pressure or components of tangential stress.
3.2.3 Pressure
From experience in thermodynamics and pipe ows, pressure may appear
to be a simple term. As a matter of fact, if pressure is known, taking its
gradient and seeing its eect on velocity can be rather simple; but the inverse
problem, when the velocity eld is known and pressure is the unknown, is
quite dierent. The analogy with the vorticity/velocity relationship in Ch.3
Figure 3.4: About pressure
(if velocity is known, take derivatives to get vorticity; if vorticity is known,
the Biot-Savart relation is far from trivial!) is very deep and the student
should ponder this.
Indeed, mass conservation for the incompressible ows
i
u
i
= 0 (3.27)
gives, by application of the divergence:
2
p =
i
u
j
j
u
i
=
2
ij
(u
i
u
j
) = rhs (3.28)
(Note that, in intrinsic notations, the divergence of the nonlinear term is
ambiguous, whereas the index notation is quite clear). This is a Poisson equa-
tion, similar to the relation between vector potential and vorticity, between
velocity and exion. Here, the velocity gradients are the sources of pressure
variations, and the use of Greens functions gives the general solution for 3D
in the absence of boundaries:
p =
1
4
rhs(x
)
| x x
|
dV

(3.29)
We recognize the property of elliptic pdes, by which the solution at any
point depends on sources in the entire eld, with vanishing awareness for
remote sources. Just as a vortex induces velocity throughout the ow domain,
velocity gradients induce pressure variations at large distances. We can hear
wind noise when driving a car, and it has been shown that vorticity is a
dominant source of pressure variations (hence the term: vortex noise). The
compressibility of air results in a nite speed of propagation of the pressure
uctuations, captured in Kirchhos theory of delayed potentials.
3.2.4 Vorticity in NS
This brings up the idea of looking for vorticity in the Navier-Stokes dynamics.
We make use of two identities. First
(u) = =
2
u +( u) (3.30)
in which the last term vanishes; second, the Lamb vector (see Ch. 2) reap-
pears since
(u) u = u = u u
1
2
(u u). (3.31)
Then, the momentum equation can be rewritten as
t
u + u = (
p
+
u
2
2
+ gz) (3.32)
So vorticity is present in the Navier-Stokes equations, aecting both the
viscous term and part of the nonlinear term when moved to the right-hand
side and multiplied by density, the Lamb vector is also called the Magnus
force. This term is a major contributor to pressure sources (above), i.e vortex
noise. The u
2
-term on the r.h.s. is obviously related to Bernoullis equation,
a topic to be pursued in Ch.5.
Figure 3.5: Nonlocal eects in incompressible uid dynamics
3.3 Vorticity equation
Sensing the need to now more about vorticity, we turn to its dynamics. Tak-
ing the curl of the Navier-Stokes equations, we get after some rearrangement
i
+ u
j
i
=
j
j
u
i
+
2
jj
i
(3.33)
On the l.h.s., we have the material derivative of the vorticity; on the r.h.s.,
the inviscid term is interpreted as a stretching term. We will return to it also
in Ch. 5.
While pressure is non explicitly present in the vorticity equation, its ef-
fects are implied through the velocity eld. When reconstructing the velocity
from vorticity with the Biot-Savart formula, the irrotational part (with local
pressure: see Ch. 5) is missing and is uniquely determined by matching the
boundary conditions; the rotational part included in Biot-Savart is induced
by vorticity distribution throughout the eld.
3.4 Energy and dissipation
Finally, the kinetic energy equation is obtained by taking the dot product
of NS on u it turns out that index notations work out better. Calling
u
2
= u
i
u
i
, we have
t
u
2
2
+ u
j
j
u
2
2
= b
i
u
i
+ u
i
j
T
ij
= b
i
u
i
+
j
(u
i
T
ij
)
j
u
i
T
ij
= b
i
u
i
+
j
(u
i
T
ij
) s
ij
T
ij
= b
j
u
j
+
j
(
j
u
2
2

1
pu
j
) (
j
u
i
)
2
, (3.34)
where we made use of incompressibility and of the renaming of summation
indices. The rate-of-strain appears because of the symmetry of T
ij
, which
cancels out the non-symmetric part of the velocity derivative. We recognize
the material derivative (as for all balance equations: Reynolds Transport
Theorem) on the l.h.s., then on the r.h.s. the rate of work from body forces,
rate of work from stresses (isotropic and viscous), and nally a term that
cannot be positive: the energy dissipation rate. Note that the dissipation
rate depends (quadratically) on the rate of strain, not vorticity.
3.4. ENERGY AND DISSIPATION 97
Figure 3.6: Energy balance: mechanics and thermodynamics
In the purely mechanical perspective aorded by Newtons second law,
energy is conserved in the sense that all changes are accounted for, but with
a net loss associated with dissipation. This is rather dierent from the ther-
modynamic perspective. The two viewpoints are reconciled by including a
thermal part of internal energy and energy exchange (Fig. 3.6). Then, the
dissipation term corresponds to the rearrangement of energy from mechani-
cal (as in the equation above) to internal energy, and does not show in the
overall budget.
3.4.1 Bernoulli with losses
We can relate the energy considerations to Bernoullis equation with losses.
Strictly speaking, Bernoullis equation is a momentum equation (see Ch. 5),
derived under the assumption of inviscid ow. Also the spirit of Bernoullis
equation is to establish the conditions under which the Bernoulli term B is
constant, while Croccos theorem (Ch. 5), deriving how B varies, is closer to
the present topic. But the eect of viscosity is known as Bernoullis equation
with losses, and can be studied at this time in the context of energy.
We start from the NS equations in the form
t
u + u = B +
2
u (3.35)
and evaluate the changes of B along a streamline (note the subtle meanings
of the u operation):
u B = u
t
u + u (
2
u)
= (
t
u
2
2
+
2 u
2
2
) (u)
2
(3.36)
Two contributions are recognized on the r.h.s. First is the unsteady diusion
of kinetic energy. For steady pipe ow, the time derivative term cancels out,
and the velocity prole shows a maximum at the centerline in the direction
of the ow, so the Laplacian of energy is negative, and kinetic energy diuses
from the centerline toward the walls. Second, we see the dissipation term,
always negative. Both contributions (diusion and dissipation) are negative
in steady ow, leading to a gradual loss in the value of B as we follow a
streamline. These eects are modeled in practice (see undergraduate texts)
in terms of an empirical formula based on the Darcy friction factor.
3.5 Enstrophy
Similar considerations apply to the square vorticity
2
=
i
i
. The enstro-
phy
2
/2 plays an important role in geophysical applications and turbulence
theory. Starting from the vorticity equation and multiplying (with summa-
tion) by
i
, we get
2
2
+ u
j
2
2
=
i
j
u
i
+
2
jj
2
2
(
j
i
)
2
=
i
j
s
ij
+
2
jj
2
2
(
j
i
)
2
(3.37)
3.6. STRUCTURE OF THE EQUATIONS 99
Beside the familiar convection and diusion terms, we see a vorticity dissi-
pation term similar to its counterpart in the energy equation. The vortex
stretching term is quadratic in vorticity, and retains only the symmetric part
of velocity derivatives (rate-of-strain). We will return to this term in Ch. 5.
3.6 Structure of the equations
Our equations can be categorized in several ways:
the conservation laws (mass, momentum, and the pressure, vorticity,
energy, enstrophy and related equations as corollaries), as distinct from
the constitutive equations and from kinematic denitions, relations
and/or assumptions. The distinction is physical in nature.
Mathematical distinctions impact how equations can be solved, and the
similarities between the momentum, vorticity, energy and enstrophy
equations on one side, as distinct from the Poisson/Laplace equations
for pressure and other properties as seen later in these notes, and from
the denitions and constitutive relations and kinematic relations.
In this section, we focus on the latter.
A common feature of the momentum, vorticity, energy and enstrophy
equations is that they include (Fig. 3.7)
a convection-diusion core, of the form
t
+ u
2
applicable to
each of these properties;
some terms (including the convective derivative and diusion term) of
divergence form (also called transport terms), which Gauss theorem
associates with boundary terms (what is the physics here? how about
numerical implications?)
the remaining terms, called dissipation or production if their signs are
known, or simply source/sink terms otherwise.
Other equations are somewhat subsidiary: continuity or pressure or stream-
function (they are not independent) and may be divided among
the Poisson and related equations (Laplace, Biot-Savart), for which the
Greens function introduces non-local eects.
kinematic relations between properties
constitutive equations (some already implied, e.g. by diusion, above)
other denitions
It is important to note that the use of Greens functions in relation with
the nonlocal eects relies on linearity (Laplacian) of either kinematic or dy-
namical relations: superposition of eects from distributed sources is essential
for Greens functions to be applicable. In the Navier-Stokes equations, the
nonlinear terms preclude the use of this approach: solutions based on Greens
function for diusion cannot accommodate the convective terms.
3.7 Incompressible ow approximation
These notes deal exclusively with incompressible ows, where the simplied
form of the continuity equation applies:
i
u
i
= 0. (3.38)
This is not the same as assuming that the uid density is constant, which
would be an equation of state. For example, it is well known that ideal gas
behavior
pv = RT (3.39)
is consistent with incompressible aerodynamics. The resolution of this ap-
parent paradox rests on the observation that we only need to assume that the
eects of density variations are dynamically unimportant a much weaker
assumption. Two important classes of incompressible ows are established:
the small-Mach-number (low speed) ows, and the small-buoyancy natural
convection ows (Boussinesq approximation).
The analysis is based on the adoption of some reference density
0
and
small departures from it (/
0
1); and the subtraction of the refer-
ence hydrostatic balance
3
p
0
=
0
g (3.40)
from the momentum equation. The remaining terms are
t
u
i
+ u
j
j
u
i
=
1
i
p g
i3
+
jj
u
i
(3.41)
3.7. INCOMPRESSIBLE FLOW APPROXIMATION 101
Figure 3.7: Categories of equations and their structure
neglecting corrections of order p. Note that the body force becomes
the buoyancy force responsible for natural convection and also relevant in
centrifuges and other applications. Compressibility eects are then captured
by =

p
dp or
=
1
p
p = p (3.42)
where is the (isentropic) compressibility coecient.
3.7.1 Small Mach number ows
What would be the order of magnitude of pressure variations? The NS
equation shows that it can scale as the convective term or as the viscous
term. Focusing rst on the former, we see that
x
p
1
2
x
u
2
, so that
Bernoulli-type scaling applies (p u
2
) and
u
2
(3.43)
Introducing the general relation for the speed of sound a
a
2
=
p
|
s
(3.44)
we obtain

u
2
a
2
Ma
2
. (3.45)
Consequently, density variations can be neglected if Ma 1.
If viscous scaling is relevant, the incompressibility condition becomes
Ma
2
Re (3.46)
3.7.2 Boussinesq approximation
The Boussinesq approximation seems paradoxical at rst: in natural con-
vection, the buoyancy force (dependent on thermal expansion) cannot be
neglected, so how come we can neglect density variations in mass balance?
In the full continuity equation
(
t
+ u
i
i
) +
i
u
i
= 0 (3.47)
3.8. SUMMARY 103
the rst group of terms (including derivatives of density) scale as U/L,
whereas the last group of 3 terms (summation...) scale as
0
U/L. So, as long
as /
0
is small, we can keep the simpler form
i
u
i
= 0.
Consistency with momentum balance results form the fact that the buoy-
ancy term (variable density) is at most comparable to the other terms. It
can be shown that the corresponding scaling requirement is
gL < 1 (3.48)
See Tritton Section 15.1 p198, for additional details.
3.8 Summary
The Navier-Stokes equations are the cornerstone of uid dynamics. They
embody Newtons second law (F = ma) for incompressible Newtonian uids,
with more general forms available in the literature. They have also been
shown to be the statistical limit of kinetic theory (Chapman-Enskog) for
gases slightly out of equilibrium. In conjunction with kinematic constraints,
they represent the only analytic basis for the study of uid motion.
Thus, it is humbling to realize that, nearly two centuries after they were
derived, so few solutions have been obtained. The undergraduate student
is exposed to Poiseuille and Couette ows, in which stationary ow and
geometry eliminate the time dependence and nonlinearities. Oseens vortex
is time-dependent, but there are no convective eects; the same holds for
Stokes two classic problems (see Ch. 6). See Pantons Chs. 7 and 11 for list
of solutions; see e.g. Acheson for the Burgers vortex.
But, as the leading idea through the remainder of the course, one can learn
from the equations without solving them. By understanding their physical
content, we will be able to work out rational approximations, keeping in mind
when they might fail and what the telltale signs might be.
ing
Compressible ows, acoustics.
Non-Newtonian uids, rheology
Suspensions, colloids, foams, granular ows
Problems
1. Show that the energy dissipation rate depends on the rate-of-strain
(and not on vorticity). Show that, in the case of a ow enclosed by
rigid boundaries, the vanishing boundary ux terms (divergence) allow
the rewriting of dissipation rate as proportional to square vorticity.
(Adapted from Batchelor, p.263)
2. Consider a one-dimensional unsteady ow for which u(y, t) is the only
non-vanishing velocity component. Show that the equations of motion
take the form of unsteady diusion. (from Tritton, p 471.)
3. Consider an innite at plate oscillating it its own plane with velocity
U = U
0
sin t . Assume that the uid oscillates at the same frequency
(but not in phase). How do the amplitude and phase of uid motion
vary with distance from the plate (Stokes second problem).
4. Write any of the major equations (momentum, vorticity, energy, en-
strophy, temperature...) to emphasize transport terms (i.e. make di-
vergence terms appear wherever possible). Are these forms unique?
5. Helicity is dened as u
i
i
. It is of some interest because it vanishes
in 2-D, and can serve as an indicator of 3-dimensionality. Using only
index notations, derive the evolution equation governing helicity and
label its various terms.
6. (From Currie, p.60) Using only index notations, prove the incompress-
ible kinematic relation
[(u )u] =
1
2
2
(u u) u (
2
u) , or
k
(u
j
j
u
k
) =
1
2
2
kk
u
j
u
j
u
j
2
kk
u
j
j
, and discuss the nature of the various terms
as sources in the pressure equation.
7. Using only index notations, derive the evolution equation governing the
Lamb vector and label its various terms.
Chapter 4
Dimensionless expressions
Dimensionless numbers occur in several contexts. Without the need for dy-
namical equations, one can draw a list (real or tentative) of physically rel-
evant parameters, and use the Vaschy-Buckingham theorem to construct a
shorter dimensionless list. Dimensionless expressions are the required tool
to compare data from dierent experiments (e.g. parachute data in water),
leading to the recommendation that all data should be plotted in dimen-
sionless form. This is generally covered at the undergraduate level, and a
few points of interpretation are added here. The same dimensionless expres-
sions are obtained from dynamical equations, when available: the meaningful
dimensionless numbers are ratios of terms in various equations, measuring
their relative importance. This can be used to approximate the equations
rationally, by dropping small (dynamically inactive) terms. One notable ex-
ception is when the small parameter is the coecient of the highest-order
derivative in the equation...
4.1 Dimensional analysis
This material is assumed known from undergraduate courses: ll in any gaps
(and practice) by consulting undergraduate textbooks from the library re-
serve. We will review the procedure (same for all problems) with one familiar
example as illustration. Occasional features not shown in the example will
be mentioned for reference.
1. The list of parameters:
105
106 CHAPTER 4. DIMENSIONLESS EXPRESSIONS
(a) This step determines the eventual solution, and several attempts
may be necessary to identify the list that makes sense of the data.
The list should be suciently complete to account for the physics
of the ow, but not to the point of introducing unnecessary com-
plications: trial-and-error, from the simpler to the more elaborate,
is sensible. The list is a reection of individual insight and of the
professions expertise.
(b) Example: We will work with fully developed ow in a circular
pipe. The list of parameters includes: ow parameters (pressure
drop per unit length dp/dx, average speed V ), pipe conguration
(diameter D or cross-sectional area A, not independent of course),
and material properties (uid density and viscosity ).
p/L V D
not included: surface roughness; etc.
(c) If some relations are known (e.g. between velocity, cross-sectional
area and volume ow rate), the corresponding parameters are not
independent, and one of them can be eliminated from the list for
each such relation. Also, in this problem, we start from the pres-
sure drop per unit length of pipe, rather than pressure drop and
overall length as separate variables: the assumed proportionality
between them is a valuable insight (try solving without it and note
the dierences!)
2. Primary dimensions:
(a) Dimensions are more general than units; e.g. meter, foot and
mile and micron are units relevant to the dimension of length.
In general, there are 3 dimensions (M, L, T for mass, length and
time respectively - other combinations can be used) for mechanical
problems (some dimensions may be irrelevant in some problems,
e.g. mass in simple pendulum), and then one each for thermal
problems ( for temperature), chemical, electromagnetic and ra-
diation problems. The maximum is 7, use only as many as needed,
3 or 4 are most common in mechanical engineering problems.
(b) Example:
The dimensions of each of the problem parameters are listed.
When not obvious, use a simple relation (e.g. pressure is force
4.1. DIMENSIONAL ANALYSIS 107
per unit area, force is mass times acceleration, etc.)
p/L V D
ML
2
T
2
LT
1
L ML
3
ML
1
T
1
3. Select the scaling parameters:
(a) This is the next critical step. One must select as many scaling
parameters (those that serve as dimensional yardsticks for all oth-
ers) as there are independent dimensions. The scaling parame-
ters must contain all dimensions in such a way that one cannot
make a dimensionless expression between them; options include
the selection of the simplest expressions, and/or the exclusion of
the parameters you wish to solve for (see interpretation, below),
called control parameters
(b) Example:
In this instance, we want to know about pressure drop, so we set
it aside if possible; speed and diameter are simple and would be
selected; then we need mass as a dimension, and density is simpler
so we select it. 3 dimensions, 3 scaling parameters:
p/L V D
ML
2
T
2
LT
1
L ML
3
ML
1
T
1

Pressure drop and viscosity are our control parameters in this case.
(c) Occasionally, some groupings of dimensions (e.g. LT
1
) may have
to serve as a single dimension: go back one step and start again.
4. Non-dimensionalize each of the control parameters
(a) Then, repeat the following procedure for each of the control pa-
rameters in turn: take a parameter, multiply it by powers of the
scaling parameters, and adjust the exponents to make the expres-
sion dimensionless.
(b) Example:
1
=
p
L
V
a
D
b
c
1 = ML
2
T
2
L
a
T
a
L
b
M
c
L
3c
1 = M
1+c
T
2a
L
2+a+b3c
c = 1 a = 2 b = 1
1
=
pD
LV
2
.
Similarly we nd
2
=

V D
. (4.1)
Always check these results! Always double-check the initial di-
mensions! Expect for standard expressions (e.g. Re) to come out.
(c) The procedure involves a system of linear equations for the ex-
ponents of each scaling parameter. The proof of the Vaschy-
Buckingham theorem (see undergrad text) is based on the rank of
the corresponding matrices.
5. The result
(a) The idea is that there is a relation between the parameters in
the initial list; since any relation, reecting fundamental laws and
phenomenology too complicated to unravel analytically, must be
dimensionally correct, it can be rearranged as a relation between
dimensionless parameters. Since there are fewer of these, the re-
lation is much simpler.
(b) Example:
In our case, we have reduced the problem of friction in pipe ows
to the relation
1
= F(
2
)
p
L
=
V
2
D
(Re
D
)
p
g
=
V
2
2g
L
D
f(Re
D
)
where F, and f are as-yet unspecied functions. We recognize
the Reynolds number, and we recover the familiar denition of the
Darcy friction factor in the Moody diagram. Note that the relation
involves an unknown function, not necessarily a proportionality.
(c) It is customary to express the pressure drop in terms of the height
of a column of uid: the parameter g is related to this scenario
4.1. DIMENSIONAL ANALYSIS 109
of pressure measurement, not to the friction in the pipe, hence it
should not be included in the initial list of parameters and serves
only for the presentation of the result.
6. Rearrangements and interpretation
(a) The procedure explained above gives a relation between the di-
mensionless control parameters. Given a result (e.g. reduced ex-
perimental data) in the form
1
= F(
2
,
3
, ...), you can deter-
mine their meaning by noting that the scaling parameters appear
in more than one of the s, whereas the control parameters ap-
pear in only one each.
However, you may wish to present the results dierently: say you
want to know how the pressure drop depends on ow speed (which
looks similar to the original presentation by Hagen). This could
be obtained by going back to the selection of scaling parameters,
and taking viscosity instead of velocity for scaling purposes (do it
for practice: note that the algebra is a little more complicated); a
better alternative is to rearrange our previous result by combining
1
and
2
to change scaling parameters.
(b) Example:
Between
1
and
2
as above, we want to eliminate V as a scal-
ing parameter, i.e. V should appear in only one dimensionless
product. This is done easily by combining
1
with
2
3
=
1
/
2
2
=
p
L
D
3
2
(4.2)
(The aspect ratio D/L is included, although starting with the
pressure drop per unit length of pipe does not bring it out as a
parameter.) The corresponding (Hagen) plot shows (dimension-
less) pressure drop as a function of ow speed (Reynolds number),
whereas the Moody plot shows pressure drop as a function of in-
verse viscosity (Reynolds number) (Fig. 4.1). Think about it.
Same data, same result, dierent presentation, you must read it
correctly.
Standard dimensionless numbers are tabulated in a number of undergrad-
uate texts. The student should be familiar enough with them to recognize
them when they arise.
10
4
10
6
10
2
10
1
Reynolds number
D
i
m
e
n
s
i
o
n
l
e
s
s

f
r
i
c
t
i
o
n

0 2 4 6 8 10
x 10
4
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
x 10
8
Reynolds number
D
i
m
e
n
s
i
o
n
l
e
s
s

f
r
i
c
t
i
o
n

f
.
R
e
2
Figure 4.1: Comparison of Moody and Hagen diagrams for friction in devel-
oped pipe ows
4.2 Non-dimensionalization of equations
This material also appears in many undergraduate texts, which should be
consulted by the students. Only a few points are added here and in later
chapters.
Consider the Navier-Stokes equations
t
u
i
+ u
j
j
u
i
=
1
i
p +
2
jj
u
i
. (4.3)
Although one would expect 3 independent dimensions (as for most mechani-
cal problems), we factored out density, so M is no longer a relevant dimension.
So, as for dimensional analysis, we should only use 2 scaling quantities; the
usual choice is to select a velocity U and length L as scaling quantities. The
main point here is that the introduction of an additional pressure or time
scale is unnecessary and possibly inconsistent. Asterisks will denote dimen-
sionless quantities, for example:
u
i
= U u
i
x
j
= L x
j
4.2. NON-DIMENSIONALIZATION OF EQUATIONS 111
Figure 4.2: About dimensional analysis
Simple substitution gives
U
t
u
i
+ U
2
/Lu
j
u
i
=
1
L
i
p + U/L
2
2
jj
u
i
(4.4)
It is customary (with the notable exception of Stokes ows: see Chapter 6)
to adopt the nonlinear term as the yardstick and to compare all others to it
by dividing throughout by U
2
/L.
L
u

t
u
i
+ u
j
u
i
=
1
U
2

i
p +

U L

2
jj
u
i
(4.5)
We now see why it was unnecessary to select time and pressure scales:
they fall out of the equations. For time, L/U is the obvious time scale;
similarly, U
2
is the measure of pressure scale consistent with the choice of
reference term. In problems where independent time and/or pressure scales
are imposed (rather than generated by the dynamics), the above presentation
needs to be modied accordingly. Denoting t
= tU/L and p
= p/ U
2
, we
have the dimensionless Navier-Stokes equations
t
u
i
+ u
j
u
i
=
i
p
+
1
Re
L
2
jj
u
i
(4.6)
with only the Reynolds number Re
L
= U L/ as a parameter. When the
boundary conditions of a given problem are also non-dimensionalized, this
shows that all problems with same geometry and forcing (boundary condi-
tions) and same Reynolds number will obey the same dimensionless equa-
tions, and therefore have the same solution. This is as useful for numerical
simulation as for experimental comparison.
4.3 Dimensionless Equations and Scaling Anal-
ysis
All terms in an equation must have the same dimensions. Without this,
changes in units would change the ratio between terms, which is physically
impossible. In undergraduate classes (usually in uid mechanics and in heat
transfer), factoring out the dimensions is performed so as to obtain useful
dimensionless numbers.
4.3. DIMENSIONLESS EQUATIONS AND SCALING ANALYSIS 113
In scaling analysis, we go one step further. We attempt to use the phys-
ically dierent length scales, say, for each term, so the scales can be vastly
dierent in dierent directions. Similarly, the velocity components may scale
dierently, and this should be reected when factoring out their magnitude.
This leads to a proliferation of dimensionless numbers: for example, in a
boundary layer, one can dene a Reynolds number based on distance from
the leading edge, on boundary layer thickness, on momentum thickness, etc.
In complex ows such as atmospheric motion, the correct insight may depend
on the scaling choices.
Consider a term such as
y
u. In dimensional analysis, one selects a length
scale L and a velocity scale U, and factor out the dimensions
y
u =
U
L
y
u
. (4.7)
Eventually dividing by
U
L
yields the dimensionless form of the corresponding
equation. In scaling analysis, the perspective is to get a nite dierence
estimate for the partial derivative: U and L would be such that
y
u
U
L
O(1) (4.8)
With dimensions as an underlying requirement, the emphasis shifts to having
the correct order of magnitude, with the dimensionless partial derivative
being replaced by a number roughly comparable to 1 (it could be 3 or 0.2,
but not 100). Thus, instead of obtaining an exact dimensionless partial-
dierential equation, one gets an approximate algebraic equation.
Take the Navier-Stokes equations
t
u
i
+ u
j
j
u
i
=
1
i
p +
2
jj
u
i
, (4.9)
and assume for the time being that the same scaling U and L applies in all
directions. Then the nonlinear term scales as
u
j
j
u
i

U
2
L
O(1) (4.10)
and the viscous term as
2
jj
u
i

U
L
2
O(1). (4.11)
Let us now assume that the ow is such that the convective (nonlinear) term
is important; we take is as the yardstick against which the other terms will
be evaluated. Then, we have
L
U
2
t
u
i
+ O(1)
L
U
2
i
p +

UL
O(1). (4.12)
Note that the equality is replaced by an order-of-magnitude estimate. Now,
unless there is a separate mechanism (forcing) to impose a distinct time-scale,
the time-derivative term can at most be of order 1 (if larger, it makes the
convective term negligible, contrary to our assumptions!). Therefore
L
U
2
t
u
i
< O(1), (4.13)
which is consistent with a time scale of order L/U. Similarly for the pressure
term, assuming no distinct length scales gives
p U
2
(4.14)
as the order of magnitude for pressure variations over distances of order L.
In this scenario, the orders of magnitude are: less than or comparable to
1 for evolution occurring over times not shorter than L/U; order 1 for the
convective term; order 1 for pressure; and order /UL for the viscous term.
Watch out, use the correct scales!
4.4 Rational approximations
Thus, in scaling analysis, the dimensionless parameters are estimates of the
relative orders of magnitude of the various terms in an equation, provided
the correct parameters have been used. We saw, above, that this can provide
orders of magnitude for some parameters so the corresponding terms are
comparable to the leading term: time scale and pressure in the previous
example. But in other instances, all terms are known (e.g. the viscous
term), and their order of magnitude is of critical interest.
In the NS example, the convective term was taken, somewhat arbitrarily,
as reference, and is of order 1. If the Reynolds number is large under the
correct scaling, this indicates that the viscous term is relatively small: this is
a rational basis for dropping the viscous term (even though large Reynolds
4.4. RATIONAL APPROXIMATIONS 115
Figure 4.3: Comparing dimensional and scaling analysis.
number is not the same thing as inviscid: more on this below). Then the
equations of motion can be simplied as
t
u
i
+ u
j
j
u
i
=
1
i
p, (4.15)
and we have basis for using Eulers equation, instead of Navier-Stokes. We
will study this case in Ch. 5
Conversely, the scaling might indicate that the Reynolds number is very
small, in which case the viscous term is much larger than the convective term.
The standard assumption of using the convective derivative as references is
invalidated: we need to go back to the rst step, and instead of Eq.(4.12),
we get
L
2
U
t
u
i
+
UL

L
2
U
i
p + O(1). (4.16)
The most obvious is that the convective term can be dropped, yielding Stokes
equation
t
u
i
=
1
i
p +
2
jj
u
i
(4.17)
to be studied in Ch. 6. A second consequence, less obvious at rst, is that
the natural time scales and pressure scales are dierent from the large-Re
case. Think about this! The pressure now scales with U/L (i.e. with a
viscous stress), and the time scale is given by L
2
/.
The same rationale provides useful simplications in other situations.
Unsteady forcing may be treated as quasi-static or as an instantaneous im-
pulse, depending on how the time constants match up; the earths rotation
may dominate the dynamics (large scale atmospheric motion) or be neglected
(bathtub vortex); etc. We will touch on these topics in later chapters. But
one case needs special consideration, and is so important that a separate
subsection seems indicated for emphasis.
4.4.1 Large Re
What is dierent about the viscous term is that it contains the highest deriva-
tive in the equation. As a general property of such equations, externally
imposed scales (L above) are not sucient to describe the physics, since the
viscous term carries the ability to enforce the no-slip condition.
This quandary is at the core of DAlemberts paradox (see Ch. 5), which
divided the uid mechanics community for the better part of the nineteenth
century: the potential ow (aerodynamics, etc.) community, willing to over-
look zero drag as long as everything else was calculated eciently, and the
hydraulics community, for whom empirical facts won over neat mathematics.
Prandtl resolved the conundrum with his idea of the boundary layer: poten-
tial ow applies outside this layer, vorticity eects dominate inside. From
the viewpoint of scaling, the problem generates its own internal scale (the
boundary layer thickness), such that the corresponding Reynolds number is
not large, and the highest derivative is no longer negligible! This falls un-
der the general heading of singular perturbations (for the mathematically
inclined), but the familiar example of the boundary layer (Ch. 8) contains
many of the right elements.
ing
Many examples of scaling analysis appear in turbulence theory and in convec-
tive heat transfer. After dimensional analysis, where the dynamical equations
are not even used, and control volume analysis, where we integrate over many
details, scaling analysis is arguably the simplest way to learn from partial
dierential equations.
In the limit of very large Re, energy dissipation does not behave as simply
as the formula (Ch. 3) suggests: as the ow becomes turbulent, the scaling of
the rate-of-strain no longer follows the externally imposed length or velocity
scales, but instead follows the scaling of the turbulence. Thus, the dissipation
rate becomes independent of viscosity! See your turbulence course for more
on this surprising result, which shows again that Re is not a simple
limit. Another instance is the boundary layer (see Ch. 8).
Problems
Some of these problems have been adapted from Trittons, from Fox and
McDonalds and from Munson and Okiishis books.
1. Consider the vortex shedding behind a cylinder. The shedding fre-
quency f is assumed to be a function of diameter d, ow speed V , and
uid properties and . Determine the form of the relation between
dimensionless frequency and velocity.
Figure 4.4: The large-Re conundrum
2. The size d of droplets produced by a liquid spray nozzle is thought to
depend on the nozzle diameter D, jet velocity U, and the properties
of the liquid (density), (dynamic viscosity) and (surface tension,
which is an energy per unit area). Construct the dimensionless products
to show the dependence of drop size on surface tension and speed;
modify the result to express the dependence of drop size on viscosity
and surface tension.
3. The lift force on a Frisbee is thought to depend on its rotation speed
, translation speed V and diameter D as well as air density and
viscosity . Determine the dimensionless parameters in the relation
showing lift as a function of the two speeds. Then modify the relation
to express the dependence of lift on diameter.
4. The shaft power input P into a pump depends on the volume ow
rate Q, the pressure rise dp, the rotational speed N, the uid density
, the impeller diameter D, and the uid viscosity . Express the
dimensionless dependence of power on ow rate, pressure rise and other
applicable parameters; discuss alternatives. What happens to your
result if you learn eventually that P = Qdp ?
5. The power P required to drive a fan depends on the uid density , on
the fan diameter D and angular speed , and on the volume ow rate
Q. Derive the dimensionless relation showing how power depends on
diameter. Reformulate the result to show how power depends on ow
rate.
6. Incompressible steady ow in a magnetic eld combines the equations
of motion (with addition of the Lorentz force) u = 0, u u =
p+
1
(B)B+
2
u with the equations for magnetic induction
B = 0, u B = B u+
1
2
B. (here, is magnetic permeability
and is electrical conductivity). What are the similarity parameters?
Discuss their physical meaning. (Problem 21 p.474 from Tritton.)
7. Dimensional analysis for pulsatile ow in a pipe: discuss the additional
parameters and carry out the analysis.
8. Dimensional analysis of laminar ow in a helical pipe: discuss the ad-
ditional parameters and nd the dimensionless products.
9. Discuss the use of the sphere drag data in relation to improved fuel
economy at lower highway speeds.
10. Scaling analysis of Bernoullis equation, taking U and L as scaling
parameters. Under what conditions can we ignore gravity? Read up on
the corresponding dimensionless number and give a half-page summary.
11. Use scaling analysis to nd the time scale of relaxation for 1-D thermal
conduction
t
T =
2
jj
T.
Chapter 5
Inviscid Flows and Irrotational
Flows
This chapter is divided between two topics that are related but must be
distinguished carefully: inviscid ows and irrotational ows. Even though
there is considerable overlap between the two topics, inviscid ows (Euler
equation) can contain vorticity. The mechanisms by which viscous forces
forces introduce vorticity in a ow will be treated in Ch. 8 and 9. And
another distinction must be made right between inviscid ows, which, if
irrotational), have unique solutions and very-large Reynolds number ows,
which are generally turbulent. In retrospect, it is astounding that irrotational
viscous ows were not studied until this century.
5.1 Inviscid ow
In the absence of viscosity, Eulers equations are the expression of Newtons
second law for incompressible ow. As seen above
t
u + u = (
p
+
u
2
2
+ gz) = B (5.1)
and in index notations
t
u
i
+ u
j
j
u
i
= g
i3
i
p (5.2)
121
122 CHAPTER 5. INVISCID FLOWS AND IRROTATIONAL FLOWS
5.2 Bernoullis equation
There are several forms of Bernoullis equations, relying on more or less
restrictive assumptions. The student should make a clear distinction between
them. In all cases, we start from the Navier-Stokes equations:
t
u + u = B , (5.3)
where
B =
p
+
u
2
2
+ gz, (5.4)
and look for ways to isolate the gradient in the r.h.s. (The Bernoulli term
B is a scalar and cannot be confused with the body force per unit volume,
which is a vector.)
5.2.1 The strong form
The simplest way is to assume that the ow is irrotational. A necessary and
sucient condition for this to hold is that
u = = 0 (5.5)
(see Helmholtz decomposition). Then, the Magnus and viscous terms vanish,
and the time derivative can be brought inside the gradient, to give
(
t
+
p
+
u
2
2
+ gz) = 0 (5.6)
implying that
t
+ B = C(t) (5.7)
has a (possibly time-dependent) uniform value in the entire eld. The as-
sumptions made here seem to hold in simple wave motion, where the time-
dependent potential is needed (see Ch. 11 for simple examples).
If, furthermore, the ow is steady, there is also a useful relation known
as Eulers normal equation. The normal pressure gradient is only balanced
by the centripetal acceleration. So,
1
n
p =
u
2
R
(5.8)
where R is the local radius of curvature of the streamline.
5.2. BERNOULLIS EQUATION 123
5.2.2 The weaker form
A separate version of Bernoullis equation is not as restrictive: allowing vor-
ticity, the initial assumption is that viscous eects can be neglected. Thus,
we work with Eulers equation
t
u + u = B. (5.9)
Then, project the equation on the direction of the velocity vector itself, i.e.
along the streamlines. This cancels out the Magnus term, and we have
t
u
2
2
= u B (5.10)
In words, the local rate of change of kinetic energy (per unit mass) comes
at the expense of the convective (directional) derivative of the Bernoulli ex-
pression. If we restrict ourselves to steady ows, we see that
B =
p
+
u
2
2
+ gz (5.11)
is constant along any streamline. This does not preclude B from having
dierent values along dierent streamlines. This is the form of the equa-
tion generally presented in undergraduate texts; streamline independence is
usually achieved through reservoirs where static conditions prevail.
At this point, we see that the undergraduate version of Bernoullis equa-
tion (along streamlines) is only part of the picture. Incompressible inviscid
ows remain as strong conditions, but unsteadiness can be accommodated
(Fig. 5.1), and the streamline restriction disappears in irrotational ow. We
saw in Ch. 3 how viscous eects can be handled as losses along streamlines.
Furthermore, if there is vorticity in steady ow, not only is B constant along
streamlines, but is is also constant along vortex lines since
B = 0. (5.12)
5.2.3 Bernoulli and energy
Equation (5.10) is derived directly from the momentum equation. But it can
also be read as a scalar equation for kinetic energy, covered in Ch.3. This
is not a new idea: in elementary frictionless dynamics, the energy equation
Figure 5.1: Bernoullis equation and assumptions
Figure 5.2: Bernoulli and energy balance
(kinetic and potential) is derived from F=ma by projecting Newtons sec-
ond law on the trajectory. In Bernoullis equation, we recognize the kinetic
and potential energies easily; the pressure term can be interpreted as the
mechanical part of internal energy for an incompressible uid, missing in
the case of point particles. If these mechanical contributions are subtracted
from the rst law of thermodynamics, all that is left is the thermal part of
internal energy: so, after mechanical terms are accounted for in agreement
with Navier-Stokes, the (total) energy equation reduces to an equation for
temperature only, with friction as an exchange term with the mechanical
world.
We also recover the hydrostatic equation (when V = 0).
5.2.4 Simplied Croccos equation
Rather than asking about the conditions for B to be constant Bernoullis
question), we can also wonder what makes B change. We already touched
on these topics in Ch. 3, with the viscous losses, and above with the shear
layer. Consider steady inviscid (incompressible) isothermal ow. Then
u = B (5.13)
Bernoullis equation states that B is constant along streamlines and along
vortex lines. If vorticity is aligned with the streamlines (helicity is maximum),
B is constant again. Croccos theorem states how B changes from streamline
to streamline according to the equation above: the Lamb vector gives the
magnitude and direction of B. In 2D ows, where vorticity is perpendicular
to the plane of the ow (and therefore to the streamlines), the gradient of B is
normal to the streamlines and to the vortex lines. See Batchelor Section 3.5
p.156-160 (where H is enthalpy) for the compressible ow version. See also
Currie Section 3.3 p.57. Compare the derivation with viscous losses along
streamlines in Ch. 3.
This helps us understand what happens to B at the base of a manometer
tube (Fig. 5.4) mounted on the wall normal to some ow. While Bernoullis
equation, or its variant with losses, may be applied in the ow (along
streamlines), and is certainly applicable in the hydrostatic conditions of the
manometer, it is denitely not applicable across the shear layer that separates
the ow from the manometer (Fig. 5.4). According to Croccos equation, the
value of B changes across the shear layer. A similar situation is encountered
in boundary layers (Ch 8).
Figure 5.3: Croccos theorem for 2-D inviscid ows.
Figure 5.4: The shear layer at the base of a manometer resets the value of
the Bernoulli term.
5.2.5 Applications of Bernoullis equation
Students should ll in this information from suitable undegraduate texts:
1/2 page of text and a gure for each topic, maybe a simple mind-map with
relevant topics, would do. Every student at this level should know about
these phenomena.
Venturi
Pitot tube
Free surface
Siphons
Nozzles
Cavitation
Aerodynamic lift
5.2.6 Lagranges theorem
Before Navier and Stokes reintroduced Newtons viscosity in the momentum
balance, the theory of inviscid ow progressed rapidly, and an early theorem
by Lagrange is worth mentioning. The theorem states that inviscid irrota-
tional ow remains irrotational forever.
The proof is easy. Starting from Eulers equation, cancel the Magnus
term, and take the curl to get the evolution of vorticity.
t
u =
t
= B = 0 (5.14)
This is the Eulerian equivalent of Kelvins theorem (Lagrangian, below);
it applies only to irrotational ows, whereas Kelvins theorem has no such
restriction.
5.2.7 Creation of vorticity
Under the Boussinesq approximation, it is possible to have density variations
in the ow while preserving the simple form the continuity equation. Then
t
u + u =
1
(
p
+
u
2
2
+ gz) (5.15)
Figure 5.5: Creation of vorticity from density gradients
and the vorticity equation takes the form
t
+( u) =
1
(B) (5.16)
From Shapiros movie, if density gradient is not parallel to pressure gradient,
creation of vorticity in the ow is possible. This remark underscores the
importance of the conditions attached to Lagranges and Kelvins theorem
(next).
We will see other mechanisms of vorticity creation in relation to boundary
layers, ow separation and rotating ows.
5.2.8 Kelvins theorem
In Ch. 2 we introduced the circulation and its relation to vorticity
=
u dr =
dA (5.17)
An exact equation was derived by Kelvin under the assumption of inviscid
ow of uniform density, also assuming that the contour of integration is a
material line moving with the uid. Let us write the Euler equation in terms
of the material derivative
d
t
u
i
= g
i3
i
p (5.18)
Figure 5.6: About Kelvins theorem
and multiply by dx
i
taken along a material line. Then, simple manipulations
yield
dx
i
d
t
u
i
= d
t
(u
i
dx
i
) u
i
d
t
dx
i
= d
t
(u
i
dx
i
) d(
u
2
2
)
gdx
i
i3
dx
i
1
i
p = gdx
3
dp, (5.19)
where the dierentials of u
2
, x
3
and p are along the contour. Under the
standard assumption that all functions are smooth enough, the integral of
these dierentials along the contour must vanish, since we come back to the
starting value at the end of integration. Hence
d
t
= 0 (5.20)
We can read the surface integral of vorticity as equivalent to an average
vorticity normal to the surface multiplied by the area of this surface. If the
product of the two is constant for an arbitrary material line moving with
the ow, it means that a reduction in area is associated with an increase in
average vorticity, and conversely. This mechanism is very important in a
number of applications covered later in this course. Perhaps more impor-
tantly, if the material contour initiates its travel in a region where there is
Figure 5.7: Vortex stretching
no vorticity, the associated particles will remain in irrotational motion as
long as viscous eects are negligible. (Note the interesting exception in Ch.
10 where the addition of a weak rotational force (Coriolis) and stretching
amplies the creation of weak vorticity).
Vortex stretching
A dierent formulation of the same idea can be obtained from the vorticity
equation. For inviscid ow, the material derivative of vorticity is
d
t
i
=
j
j
u
i
(5.21)
The square vorticity (enstrophy) then obeys the equation
d
t
2
2
=
j
j
u
i
i
=
j
s
ij
i
(5.22)
because the symmetry of
i
j
cancels the non-symmetric part of the velocity
derivative (see linear algebra). In particular, we can select the local axes to
coincide with the principal axes of the local rate of strain. Then for each
direction
d
t
2
(i)
2
= (s
(i)
2
(i)
) (no sum). (5.23)
Therefore, following a material point (Lagrangian description), the squared
vorticity in each direction (no sum) varies exponentially in time
2
(i)
e
s
(i)
t
. (5.24)
For the component of vorticity long the primary extensional axis, we have
an exponential increase; along the compression axis, exponential decrease
of square vorticity. Because the latter approaches zero while the former
increases very rapidly, vorticity tends to align itself with the extensional axis
and increases in magnitude. The idea here is the same as in Kelvins theorem,
the presentation quite dierent, the result equivalent.
5.2.9 Helmholtzs vortex theorems
Recall (Ch. 2) Helmholtzs theorems
1. vortex lines cannot end in the uid, they can only end at a boundary.
2. in inviscid incompressible ow, uid elements that lie on a vortex line
remain on vortex line (this means that vorticity is a marker just as
dye).
3. circulation is the same for all sections of a vortex tube.
Theorems 1 and 3 were proved and explained in Ch. 2; theorem 2 invokes
inviscid dynamics and can nally be addressed.
Consider the vorticity eld, with vorticity vectors distributed through
space; consider the associated vortex lines; and nally select the vortex lines
supported by some smooth line so as to make a vortex sheet. Then, draw a
material contour within this vortex sheet. Since the vorticity is in the sheet,
the normal to the sheet is normal to vorticity, and therefore
=

n
dA = 0 (5.25)
for any such contour. By Kelvins theorem, it will remain zero. This implies
that, as the arbitrary material contour deforms in the ow, it remains within
the vortex sheet. Now, take some support line intersecting the rst one, and
the corresponding vortex sheet, and arbitrary contours within the sheet, and
reach the same conclusion: the material points remain on the vortex sheet.
Finally, since a vortex line is the intersection of two vortex sheets, its
material points remain within each of the sheets, therefore remain on the
same vortex line.
See Panton Section 13.8 p.338, Acheson Section 5.3 p.162-3 for more
information.
5.3. IRROTATIONAL FLOW 133
Figure 5.8: Helmholtz second theorem
This theorem is very important: as long as eects of viscosity (diusion)
and density changes change be neglected, vortex lines act as ow markers,
and conversely dye injected along a vortex line will continue to show the
vortex line as it moves downstream.
This is a good point to view Shapiros movie again (Fig. 5.9) is repeated
here from Ch. 2.
5.3 Irrotational ow
Irrotational means that there is no vorticity. Kelvins theorem provides a
good understanding for the conditions under which this could happen. Now,
let us explore the consequences.
We start from
= 0, (5.26)
which in turn implies (and conversely)
u = . (5.27)
A.H. Shapiros movie: Vorticity
Kinematics
Denition, vorticity meter
Potential vortex
Vorticity and circulation: Stokes theorem
Start-up vortex
Dynamics
Bernoullis equation: conditions for B = p +
1
2
V
2
+g z to
be constant
Croccos equation: how B changes
Kelvins theorem: inviscid, Lagrangian, vortex stretching
Potential ow, singular vortices
Model ( = 0)
Motion of vortex pair, induced velocity
Helmholtzs theorems (material lines, vortex tubes, vortex lines
end only at boundary)
How to introduce vorticity
Singularities
p and interface
unsteadiness
rotational forces
pressure gradient along boundary (viscous eect)
See Ch. 9
Figure 5.9: Kinematics and dynamics of vorticity and circulation, A.
Shapiros movie.
Figure 5.10: About
2
= 0
Then, incompressibility leads to
2
= 0 (5.28)
at every instant. Laplaces equation admits unique solutions if either the
eld value (: Dirichlet condition) or its normal derivative (
n
= u
n
: Neu-
man condition) is given at every point on the boundary. In the presence of
impermeable boundaries, the normal velocity must be equal to that of the
boundary (which can be in motion), and the entire eld responds instanta-
neously to the current conditions. The no-slip condition would require the
presence of viscous eects, themselves associated with vorticity. The no-slip
condition is incompatible with the Laplace equation for potential ow.
A similar picture emerges if we satisfy mass balance rst. Then the vector
potential
A = u (5.29)
must be such that
u = = 0 =
2
A. (5.30)
The Laplace equation now applies to the vector potential instead of the scalar
velocity potential. The picture is consistent: irrotational ow is characterized
by unique solutions that reect the boundary conditions throughout the eld
at each instant.
Figure 5.11: Potential lines and streamfunctions.
5.3.1 2D potential lines and streamlines
In 2-D ows, it is easy to show that lines of constant and lines of constant
are perpendicular to each other. Indeed, we can write
d =
x
dx +
y
dy = u dx + v dy = 0 (5.31)
so that
dy
dx
|
=
u
v
. (5.32)
For the streamlines
d =
x
dx +
y
dy = v dx + u dy = 0 (5.33)
so
dy
dx
|
=
v
u
. (5.34)
Hence
dy
dx
|

dy
dx
|
= 1, (5.35)
which proves the result.
In 2D, it is also convenient to use polar coordinates when the symmetry of
the ow makes this representation simpler. For the velocity potential (r, ),
we need the components of the gradient:
u
r
=
r
and u
=
1
r
(5.36)
Similarly, with A = (0, 0, ), the expression of the curl gives
u
r
=
1
r
and u
=
r
(5.37)
5.3.2 Superposition of 2D potentials
This material may have been covered at the undergraduate level. This sub-
section is a list of elementary ow types and a brief discussion of how to use
them. The key is to realize that the Laplace equation (for and ) is linear,
so that elementary solutions can be added to each other.
Uniform ow
We take the ow to be uniform with velocity U in the x direction: changes
of coordinates can accommodate dierent directions easily.
u = U and v = 0. (5.38)
Then, the potential and streamfunctions are easily obtained by integration.
(Note that the arbitrary constant of integration can be ignored without loss
of generality: addition of the constant does not modify the velocity eld.)
= Ux and = Uy. (5.39)
Stagnation ow
For this case
= x
2
y
2
and = Cxy, (5.40)
for which the streamlines are branches of hyperbola.
For other ows, polar coordinates capture the symmetries more eec-
tively; but of course the choice of coordinates is yours, for most eective
solution of the problem at hand.
3 2 1 0 1 2 3
0
0.5
1
1.5
2
2.5
3
Potential flow at a stagnation point
x
y
Figure 5.12: Potential lines (dashed lines) and streamlines (solid lines) of a
stagnation ow
Point source or sink
Sign reversal gives a point sink, obviously. Let q be the ow rate coming out
of the source, where we place the origin of coordinates. Then
=
q
2
ln(r) =
q
2
ln(
x
2
+ y
2
) and =
q
2
=
q
2
atan(
y
x
2
+ y
2
).
(5.41)
Then, the velocity components are
u
r
=
q
2
1
r
and u
= 0. (5.42)
Point vortex
Dual of the source: swap and above and get
=

2
=

2
atan(
y
x
2
+ y
2
) and =

2
ln(r) =

2
ln(
x
2
+ y
2
).
(5.43)
u
r
= 0 and u
=

2r
. (5.44)
3 2 1 0 1 2 3
3
2
1
0
1
2
3
Potential flow around a source
x
y
Figure 5.13: Potential lines and streamlines of a point source
3 2 1 0 1 2 3
3
2
1
0
1
2
3
Potential flow around a source
x
y
Figure 5.14: Potential lines and streamlines of a point vortex
2 1.5 1 0.5 0 0.5 1 1.5 2
2
1.5
1
0.5
0
0.5
1
1.5
2
x
y
Potential flow around a doublet
Figure 5.15: Potential lines and streamlines of a doublet
Source-sink doublet
Place a source of strength q at x = a, a sink of equal strength at x = a,
and superpose the ows (add the potentials); then take the limit of a 0 at
xed value of = aq. The result is a doublet aligned with the x-axis.
=
cos
r
and =
sin
r
. (5.45)
u
r
=
cos
r
2
and u
=
sin
r
2
. (5.46)
It can be checked that streamlines and potential lines are circles.
Rankine half-body
Superposition of a uniform ow and a source
= Ur cos +
q
2
ln(r) and = Ur sin +
q
2
. (5.47)
u
r
= U cos +
q
2r
and u
= Ur sin . (5.48)
3 2 1 0 1 2 3
3
2
1
0
1
2
3
Potential flow around a Rankine halfbody
x
y
Figure 5.16: Potential lines and streamlines of a Rankine half-body
Cylinder
Superposition of a doublet and a uniform ow
= Ur cos +
cos
r
and = Ur sin
sin
r
. (5.49)
u
r
= U cos
cos
r
2
and u
= U sin
sin
r
2
. (5.50)
Cylinder with circulation
Superposition of a doublet, a vortex and a uniform ow This to be lled in.
5.3.3 F=ma!?
The previous subsections show that, for suitable boundary conditions, one
does obtain unique solutions for the velocity eld by satisfying mass balance
and the kinematic constraint of irrotational ow. How about F = ma? Can
we have a solution without momentum balance? Answer in class, after you
think it through!
3 2 1 0 1 2 3
3
2
1
0
1
2
3
x
y
Figure 5.17: Potential lines and streamlines of the ow around a cylinder
3 2 1 0 1 2 3
3
2
1
0
1
2
3
x
y
Figure 5.18: Potential lines and streamlines for a cylinder with circulation
0 20 40 60 80 100 120 140 160 180
3
2.5
2
1.5
1
0.5
0
0.5
1
Angle (degrees)
P
r
e
s
s
u
r
e

c
o
e
f
f
i
c
i
e
n
t
Pressure distribution on cylinder in potential flow
Figure 5.19: Pressure force on the cylinder
5.3.4 DAlemberts paradox
It is easy to see that the streamline = 0 corresponds to a circle of radius
R =
/U. Applying Bernoullis equation between innity and the cylinder

surface, we have
U
2
2
+ p
=
(u
2
r
+ u
2
)
2
+ p |
R
(5.51)
Taking p
= 0 as reference pressure (why is this OK?) we get the surface

pressure (Fig. 5.19) as
p |
R
=
U
2
2
(cos 2 1) (5.52)
This is maximum at the front and back stagnation points = 0 and , and
minimum at = /2.
Then it is easy to see that, for a surface element of area Rd, the com-
ponents of force are
dF
x
= Rd cos p
R
(5.53)
and
dF
y
= Rd sin p
R
. (5.54)
By integration, the drag D and lift L are
D =
dF
x
= 0 (5.55)
and
L =
dF
y
= 0. (5.56)
Vanishing lift is clearly the result of symmetry, broken by adding circulation.
But the vanishing drag has much deeper roots. In fact, the absence of drag
was proved by DAlembert for bodies of any shape whatsoever: the mismatch
between this elegant result (and the vindication of potential external ows as
valid models within the accuracy achievable then) and the common experi-
ence of drag was called DAlemberts paradox. It started a profound schism
in the uids community, lasting for much of the nineteenth century, between
the aerodynamicists (who, with Joukovky and others, took potential ows
into higher mathematics) and hydraulics practitioners (who relied more on
empirical data). The paradox was nally solved by Prandtl, with the discov-
ery of the boundary layer and its eect on the ow (skin friction, boundary
layer separation, wake formation...)
5.4 Viscous potential ows
We will see in Ch. 9 the role played by viscosity to introduce vorticity
from solid boundaries into many ows. This ubiquitous complication blinded
the uid mechanics community to the interest of viscous potential ows.
Combining potential ow with incompressibility gives
2
u = 0 (5.57)
so that, although the viscous stress does not vanish, the viscous force does.
For additional information, see the recent work of D.D. Joseph.
ing
Batchelors entire Ch.6 is devoted to inviscid ows with vorticity: good read-
ing! In his Ch.5, Batchelor also covers the eect of viscosity on Kelvins
theorem. In relation of Helmholtzs interpretation of vortex lines with ma-
terial lines in inviscid ow, the eect of viscosity on vortex line topology
(break-up and reconnections) has also been studied extensively.
Complex coordinates, conformal mapping for 2D ows: see Acheson for a
good overview, Panton or Currie for more details. Curries Ch. 5 is devoted
to 3D potential ows.
Kelvins theorem is a particular of the Poincare-Cartan invariant in Hamil-
tonian dynamics. There are many points of overlap between irrotational ows
and Hamiltonian dynamics.
Finally, Onsager pointed out the possibility of having energy dissipation
in inviscid ows: this requires that singularities (innite derivatives) develop
as a result of vortex stretching. This cannot happen in 2D, but the issue is
still unresolved in 3D.
Problems
1. Categorize the ideas in this chapter as being for inviscid ow, irrota-
tional ow, or both. Explain relations between these ideas.
2. Discuss (on the basis of equations) the following proposition: any ow
that is irrotational is also inviscid (the converse is clearly not true).
3. Derive the equation of the free surface associated with a potential vor-
tex (next: add a sink at the origin); discuss in what respects this is or
is not a good model for the bathtub vortex.
4. Work out the pressure distribution eq. (5.52) for the potential ow
around a cylinder, and integrate it to get drag and lift; repeat with the
cylinder with circulation (add a vortex).
Chapter 6
Stokes Flows
One of the earlier approximations to the Navier-Stokes equations goes back to
Stokes himself, who studied the limit of very small Reynolds number. This
has applications to very viscous ows, suspensions and bubbles, and the
recently important eld of micro-uid-dynamics. Highlights include Stokes
paradox, far-eld eects, kinematic reversibility, and the role of vorticity.
(Fig. 6.1).
6.1 Stokes equation
We start from the Navier-Stokes equations, and use a velocity U and length
L as scaling quantities. Notations:
u
i
= U u
i
x
j
= L x
j
Then
t
u
i
+ u
j
j
u
i
=
1
i
p +
2
jj
u
i
(6.1)
becomes
U
t
u
i
+ U
2
/Lu
j
u
i
=
1
L
i
p + U/L
2
2
jj
u
i
(6.2)
Since we expect the viscous term to be signicant, we set its coecient equal
to 1 by multiplication by L
2
/U, yielding
L
2

t
u
i
+ Re
L
u
j
u
i
=
L
U
i
p +
2
jj
u
i
(6.3)
147
148 CHAPTER 6. STOKES FLOWS
G.I. Taylors movie: Creeping ows
Vanishing Reynolds number (large viscosity or small scale or low
velocity) or convective terms vanish because of geometry (e.g.
Poiseuille/Couette ow at moderate Re).
Kinematic reversibility (but stresses change sign).
No inertia, no convective terms.
Balls, drops and suspensions.
Long range eects:
2
everywhere.
Vorticity essential (from B.C.).
Flow around a sphere: exact solution
Stokes paradox: ow around a cylinder
Stokes drag
Slender bodies
Lubrication
Propulsion
Hele-Shaw cell
Figure 6.1: A wonderful movie by G.I. Taylor: low Re ows
6.1. STOKES EQUATION 149
Figure 6.2: Mind-map relative to scaling of pressure
In the limit of very small Re
L
, the nonlinear terms drop out. Mathemati-
cally, this makes the equations linear in u
i
, hence more easily solvable. Also,
the viscous time scale is
L
2
. Flow disturbances associated with times larger

than this are in eect quasi-stationary, and the eld is in equilibrium with
the current conditions imposed at the boundary (examples in the movie).
Furthermore, we see that U
2
is no longer the correct scaling for pressure. In
this instance, the pressure eld scales as U/L, which may be representative
of the shear stress applied at a boundary. See Fig. 6.2.
The resulting quasi-steady-ow equation is very simple:
i
p =
2
jj
u
i
(6.4)
or
p =
2
u (6.5)
or again (using incompressibility, so that =
2
u)
p = (6.6)
This should be contrasted to Croccos result for inviscid ow.
Since we assume incompressibility ( u = 0), it follows that
2
p = 0. (6.7)
Figure 6.3: Mind-map: Pressure and vorticity in Stokes ow: Laplace equa-
tion and far-eld eects
Similarly, since p = 0 for any smooth scalar function, we also have
2
= 0. (6.8)
Laplace equation everywhere! Note that in the case of (vector) vorticity, it
applies to each component. For vorticity, the stretching and advection terms
are negligible in Stokes ow. As boundary conditions vary (slowly enough),
diusive equilibrium with the current boundary values is reached instantly
(i.e. much faster).
Recall the important properties of the Laplace equation: the solution at
any point is determined by boundary conditions over the entire boundary.
The boundary conditions can be either for the eld itself (e.g. pressure) or
its normal gradient, at each point of the boundary. Actual dependence on
remote boundary points is quantied by Greens function. Many of the
nonlocal eects in Stokes ow (Taylor movies, see below) are related directly
to this analytical feature.
6.1.1 Kinematic reversibility
Go back to Stokes equation for velocity
i
p =
2
jj
u
i
(6.9)
If we reverse spatial directions, the pressure gradient changes sign, the Lapla-
cian keeps its sign, but velocity changes sign as well (time, as we know, can-
not be reversed!) All in all, the equation is unchanged when we change the
6.2. STOKES TWO PROBLEMS 151
Figure 6.4: Reversal in Stokes ow.
sign of any (Cartesian) coordinate. The illustrations in Taylors movie are
spectacular!
See Panton Ch21 Fig. 21.1 p.641 for reversibility of a ow over a block
(also note the secondary ows in the corners). There is an important distinc-
tion to be made between kinematic reversibility (above), dynamic reversibil-
ity (see the problem about potential ow at the end of this chapter) and
thermodynamic reversibility. Make sure you dont mix them up!
6.2 Stokes two problems
Two exact solutions of the equations can be derived in congurations that
reduce them to 1-D diusion. They are part of a larger class of problems
treated systematically by Rayleigh (see e.g. Telionis Unsteady viscous ow,
Springer).
The problems have in common an innite at plate, with a viscous uid
on one side. As the plate moves in its own plane, it induces motion in the
uid. The case of the rotating plate leads to the steady-state Kàrmàn pump .
The simpler case of rectilinear motion was solved by Stokes. It is noteworthy
that these problems are unchanged if the Reynolds number (e.g. based on
oscillation amplitude and frequency and on uid viscosity) is not small. See
problem at the end of chapter.
Because of homogeneity (no privileged point), the motion of the uid is
also rectilinear, with no pressure gradient in the homogeneous direction of
1 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1
0
0.5
1
1.5
2
2.5
3
Figure 6.5: Stokes two problems: oscillating (dotted lines) and impulsively
started (solid line) plate; unit viscosity and frequency; times .1, .2 ... 1.
6.3. STOKES FLOW AROUND A SPHERE 153
motion, and the equation is
t
u =
2
yy
u (6.10)
The two Stokes problems have this diusion equation in common; the dier-
ence is in the boundary condition. For the rst problem, we start from rest,
and the plate is impulsively started at constant speed U for t > 0. For the
second problem, the plate oscillates with frequency
U |
y=0
= U
0
e
2it
(6.11)
The solutions are classics, and should be read carefully
1
. For the rst prob-
lem, the motion progresses into the uid, aecting a layer scaling as
t.
For the second problem, the cumulative eect of wall motion is partially can-
celed by its periodic reversals, and the ow oscillates, out of phase with the
forcing, with decreasing amplitude away from the wall.
See e.g. Panton Sections 11.2 and 1.3 p.266-279 for details.
6.3 Stokes ow around a sphere
With Stokes equation expressing momentum balance at vanishing Reynolds
number, we also need to satisfy mass balance. This can be done with the use
of a vector potential (review Ch2 !)
u = A or =
2
A
as an alternative to Eq.6.7). Then, mass and momentum balance are com-
bined in the single relation
2
A = 0 (6.13)
subject to boundary conditions that reect the specics of a given problem.
1
One method of solution makes use of Greens functions, and (without going into the
details of actual solution) points to interesting physics. Greens function for diusion for
this case is
G(y, t; y
, t
) =
1
2
(t t
)
e
(yy
)
2
4(tt
)
(6.12)
for t t
, and zero otherwise. It shows that the eect of a source at (x
, t
) is felt instantly
throughout the eld, but decreases very rapidly with distance. The equilibrium solution
for innite time corresponds to the Laplace equation: one obtains the Biot-Savart kernel
by integrating the diusion Greens function over time.
Figure 6.6: Denition sketch for the ow around a sphere.
The ow around a sphere and around a cylinder are classical cases, for
which the geometry indicates the need for spherical and cylindrical (polar)
coordinates, respectively. (Make sure you know where to nd the appropriate
formulae of vector calculus.)
The ow around a sphere of radius R turns out to be simpler and is
presented rst. In spherical coordinates (r, , ) (where is the longitude
and is the co-latitude), we have
u = A =
1
r sin
[
(sin A
]e
r
+
1
r
[
1
sin
A
r
r
(rA
)]e
+
1
r
[
r
(rA
A
r
]e
(6.14)
Of course, we dont wish to handle this in the most general case! With
the direction of the ow around the sphere as the polar axis, axisymmetry
requires that
u
= 0, (6.15)
which implies that A
r
and A
should be constant (and we can take them

equal to zero without loss of generality). Then, A = A
, and the velocity

eld simplies into
u = A =
1
r sin
(sin A
)e
r
1
r
r
(rA
)e
(6.16)
Then, introducing the Stokes streamfunction
= r sin A
(6.17)
the velocity eld is
u =
1
r
2
sin
e
r
1
r sin
r
e
. (6.18)
(That wasnt so hard, was it?) What we have expressed so far is: mass
conservation, axisymmetry. Now, the boundary conditions will make the
solution specic to the sphere. Since the sphere itself must coincide with a
streamline, must be constant on the sphere, and we can take
|
R
= 0 (6.19)
without loss of generality, with u
r
= u
= 0 (no slip). At innity, uniform

ow requires

r
2
2
U sin
2
as r (6.20)
and furthermore we know that
u
r
= U cos and u
= U sin
Finally, we must get the right dynamics: Stokes equation takes the form:
[
2
rr
+
sin
r
2

(
1
sin
)]
2
= 0 (6.21)
We know we are in luck (right combination of dynamics and boundary con-
ditions) when separation of variables works: try it for practice. The result
is
= R 2U
1
2
(
r
R
)
2
sin
2
[1
3
2
R
r
+
1
2
(
R
r
)
3
] (6.22)
The velocity components are easily obtained, and vorticity, given by the
relation
=
U
R
3
2
(
R
r
)
2
sin (6.23)
(which component is this, by the way?), is maximum at the equator and zero
along the polar axis.
It should be noticed that the streamlines exhibit symmetry upstream/downstream
(dependence on sin(), not cosine!). Does this make sense to you? What is
the relevant concept earlier in this chapter? In relation to kinematics (chap-
ter 2), the dierence between the steady ow (as observed from the sphere)
versus unsteady ow (as observed from the uid away from the sphere) is of
2 0 2 4
10
8
6
4
2
0
2
2 0 2
4
6
8
10
12
14
Figure 6.7: Stokes ow around a sphere: (left) particle paths around a xed
sphere, top to bottom; (right) seen from the uid at rest, the motion of the
sphere aects the motion of some particles (bottom to top).
interest. On Fig. (6.7), seen from the sphere, the streamlines, pathlines and
streaklines are identical (steady ow). From the uid at rest, this is not the
case. Edit and expand.
The pressure distribution can be calculated exactly: separation of vari-
ables in the Laplace equation gives
p = p
+
3
2
U
R
(
R
r
)
2
cos (6.24)
As one would expect, pressure is maximum at the forward stagnation point
( = 0) and minimum at the rear stagnation point (no dynamic reversibility!).
Forcing an inertial scaling on this expression gives for the stagnation pressures
p
s
p
=
1
2
U
2
3
Re
(6.25)
Finally, calculation of the normal and tangential stresses over the entire
surface can be carried out (e.g., do it in Maple, starting from the solution
for velocity). The result (See Panton p647, Batchelor p233) is the classical
result for Stokes drag:
F = 6RU (6.26)
This exact formula can be used to calculate viscosity from a measurement
of terminal velocities of spherical objects (bubbles, etc.)
6.3.1 Nonlocal eects
Because of the ubiquitous Laplacians, and associated Greens functions, the
boundary condition at the sphere surface induces vorticity and relative ve-
locity at large distances from the sphere. Two eects follow from this obser-
vation.
First, consider a bubble rising near a vertical wall. The ball of inuence
of the bubble (region where it disturbs appreciably the ambient liquid) inter-
sects the wall, which prevents the induced motion from taking place. This
constraint on the induced motion slows down the bubble, as the viscous uid
needs to go around the bubble in some other way. So the bubble rises more
slowly near the wall than elsewhere in the bulk of the liquid.
A similar situation arises if two bubbles rise side-by-side. This corre-
sponds to the method of images, a mainstay for the solution of Laplaces
equation: the plane of symmetry is similar (though dierent in one impor-
tant respect: do you see which?) to the impermeable wall. The lack of elbow
Figure 6.8: Wall interference of rising bubble or falling sediment
room slows down the bubbles, which rise more slowly than they would in-
dividually. This is the reason why a cloud of rising bubbles of uniform size
(only approximated by air bubbles in a vigorously shaken water container)
will have a at bottom: individual bubbles left behind will soon catch up
the the cloud - but on the top side, any front-runner will run away from the
crowd.
6.3.2 Application: Slender bodies
See falling rod, Taylor movie.
6.4 Cylinder: Stokes paradox
With this background, Stokes ow around a cylinder appears as a simpler
variant of the above. Surprise! In cylindrical coordinates (r, , z), we have
u = A = (
1
r
A
z
z
A
)e
r
+(
z
A
r
r
A
z
)e
+
1
r
(
r
(rA
A
r
)e
z
. (6.27)
Then, symmetry can be imposed, and boundary conditions follow.
The surprise is that it is impossible to match the boundary conditions
both at innity (uniform ow) and at the cylinder surface (no-slip) with
6.5. REYNOLDS LUBRICATION THEORY 159
Figure 6.9: Wake of a cylinder even at small Re shows that Stokes ow does
not exist in this geometry.
Stokes ow dynamics. What the mathematics are telling us is the nonlocal
cumulative eect of vorticity near the cylinder does not vanish fast enough
in 2D, whereas they did in 3D (dierent Greens function!). The proper
Reynolds number is the Re based on the region aected by vorticity and this
Re is always of O(1) for cylinder regardless of U
. Thus, Stokes ow around

a cylinder does not exist! Any ow around a cylinder shows eects of inertia
(nite Re), e.g. a wake in which the velocity defect shown as wider spacing
of the streamlines on the downstream side. The corresponding solution was
rst calculated by Oseen.
6.5 Reynolds lubrication theory
Reynolds studied another important application at vanishing Re. The Couette-
like shear ow between non-parallel surfaces is illustrated by Taylors little
gizmo in the movie; sliding a sheet of paper across a table, air-hockey and
similar games, thrust bearings for marine propellers, and the circular ge-
ometry of eccentric journal bearings, provide a wealth of illustrations. The
combination of narrow gaps (small Re), shallow angles and moderate veloci-
ties yield relatively large hydrodynamic forces on the solid surfaces.
The basic problem is formulated as follows (Batchelor p219, Panton p
660), 2-D (plane) to simplify expressions. Fixed pad of length L above, at
plate sliding at velocity U, gap of thickness h(t, x) with h L everywhere.
Then x L, y h and u U. It follows (x-momentum) that
p
UL
h
2
, (6.28)
Figure 6.10: Denition sketch for the Reynolds lubrication problem.
which is independent of y (compare with BL approximation!) By continuity,
v
Uh
L
U. (6.29)
At small Re, we have
x
p =
2
yy
u (6.30)
with the no-slip conditions u |
y=0
= U and u |
y=h
= 0. The solution is readily
obtained (combination of Couette and Poiseuille) as
u =
h
2
2
x
p((
y
h
)
2
y
h
) + U(1
y
h
). (6.31)
It can be checked (see BL approximation) that in the y-momentum equa-
tion, the pressure term is of order UL/h
3
, while the viscous term is only
of order U/hL. So the pressure term is unmatched in terms of order of
magnitude, unless
y
p vanishes to eliminate its own scaling factor. So, the v
component is determined by mass balance, not by momentum balance.
Start from the continuity equation and integrate over the entire depth:
h(x)
0
x
u dy +
h(x)
0
y
v dy = 0. (6.32)
The second integral is straightforward, the rst one calls for the Leibniz
6.6. LAGRANGIAN TURBULENCE, CHAOTIC MIXING 161
theorem (see integral method in related courses)
2
. The result is
1
x
(h
3
x
p) = 6U
x
h + 12
t
h (6.33)
known as Reynolds lubrication equation. It is a Poisson-type equation for
pressure, with derivatives of h as source terms and ambient pressure at either
end of the pad as BCs. In the simple case of time-independent h and uniform
slope, we write =
x
h as the constant (small) pad angle, so that sin ,
and the solution is
p(x) p
0
=
6U
(h
0
h)(h h
1
)
h
2
0
(h
0
+ h
1
)
. (6.34)
Writing h = h
0
x, h
1
= h
0
L and h
1/2
= (h
0
+ h
1
)/2, and keeping
only the leading term for small , we get
p(x) p
0
= 6U
x(L x)
h
2
0
(h
0
+ h
1
)
=
3
4
L
2
h
2
0
U
h
1/2
2x
L
(2
2x
L
) (6.35)
This expression is informative. The spatial dependence is parabolic, with
its maximum at the center of the pad. We factored out the average shear
stress
U
h
1/2
in accordance with Stokes ow scaling. The remaining factor
3L
2
/4h
2
0
can be extremely large. The lower limit of h
0
> L is determined
by surface tolerances.
6.6 Lagrangian turbulence, chaotic mixing
Although reversibility is true for incremental time steps, and might be ex-
pected to hold for nite times, there are instances of extreme sensitivity to
initial conditions such that chaotic mixing occurs even in Stokes ow. See
the literature on Lagrangian turbulence for this.
2
In a nutshell,
d
x
b(x)
a(x)
f(x, y) dy =
b
a
d
x
f dy + d
x
bf(x, b(x)) d
x
af(x, a(x)).
0.2 0 0.2 0.4 0.6 0.8 1 1.2
0
0.2
0.4
0.6
0.8
1

p

.

4
/
3

.

h
0 2
/

L
2

.

h
1
/
2
/

U

x/L
Figure 6.11: Pressure distribution under a simple pad
ing
Journal bearings: see introduction in Acheson, p 250. Hele-Shaw cell: ow
viz of potential ow. See Acheson for a clear simple presentation.
Liquid adhesive: thin layer of uid between matching surfaces. Separating
the surfaces requires the creation of Poiseuille ow between them, which in
turns requires very large pressure gradients.
The nite-Re eects for the ow around the sphere introduce weak non-
linearities. The classic analysis of Oseen can be found in textbooks on viscous
ows. The convective terms break the front/back symmetry of the ow.
The dierence between large Re (inertial) and small Re (viscous) propul-
sion is well illustrated in Taylors movie.
The eld of micro uid mechanics has evolved recently under combined
pressures of MEM-actuators for ow control strategies and the growth of
interest in biomedical applications of uid dynamics. This is beyond the
scope of this course.
Problems
1. Discuss the contrast between the reversal of dye motion in Taylors
movie and the lack of reversal for the piece of thread. Map out the
relevant ideas.
2. We saw that Stokes ow is kinematically reversible. Carry out a sim-
ilar analysis for potential ow, and comment on the dierences. How
about dynamic reversibility? In which is pressure independent of ow
direction? why?
3. In irrotational ows, we had
2
= 0; in axisymmetric Stokes ow, we
have
2
= 0. Discuss analytical and phenomenological similarities
and dierences.
4. Identify the sources of pressure in Stokes ow.
5. Outline dierences and similarities between Stokes rst problem (im-
pulsively started plate) and boundary layer development.
6. Taylors two problems, as well as classical Poiseuille and Couette ows,
use the Stokes ow approximation in spite of Reynolds numbers pos-
sibly of the order of 1000. Resolve this apparent discrepancy.
7. Analyze the vertical motion of Taylors teetotum, to determine how
the elevation of the pads above the table surface is aected by angular
speed (other obvious parameters: pad angle and area, weight per pad,
typical radius, weight,...)
8. Consider two extreme cases of the ow past a rectangular block (Van
Dyke Figs. 5 and 11, sketched on Fig. 6.12). Discuss kinematic and dy-
namic reversibility; discuss similarities and dierences relative to pres-
sure, vorticity, wall stress, velocity magnitudes, etc.
9. Consider Stokes second problem: a Reynolds number based on ampli-
tude of plate oscillation, frequency and viscosity, is easily constructed.
We do not assume this Re to be small, yet Stokes equation applies.
Explain why, and list familiar examples where the same situation oc-
curs.
Figure 6.12: Potential and creeping ow past a block
Chapter 7
Interlude
This is a good time to organize some lines of thought. The student is urged
to go back over the material covered, and collect facts and equations (and
mind maps) associated with every topic mentioned more than once: no topic
should be without context. Here, the emphasis is put on three main themes:
the approximations, non-local eects, and vorticity.
7.1 The approximations
The approximations to the intractable Navier-Stokes equations are based on
scaling analysis and dimensionless numbers: depending on the correct orders
of magnitude, the ability to neglect some terms is of great importance for
analysis, computation and experimentation alike.
Just as important is the awareness of the discarded physics. When we
drop the convective or the viscous term, a number of phenomena become
inconsistent with the new equations. Even experienced uid dynamicists can
overlook these inconsistencies on occasion: context should be ever present as
the best safeguard.
The key to the correct use of approximations is to remember that they
correspond to extreme cases. The results and insights obtained in these
extreme cases cannot be taken blindly into the more complicated cases of
practical interest, but they can help our thinking, if only to treat results
with caution.
165
166 CHAPTER 7. INTERLUDE
7.2 Non-local eects
Non-local eects are not emphasized explicitly in many uid mechanics texts.
Although Biot-Savart and induced velocities are generally mentioned, and the
pressure equation (divergence of Navier-Stokes) is sometimes listed, the gen-
eral properties of the Poisson and Laplace equations is generally overlooked
except for numerical work. These elliptic equations are mathematically and
computationally challenging in practical geometries; of interest at this level
is the concept of non-locality. Velocity in potential ow, pressure in large
and small Reynolds and small Rossby number ows, vorticity in Stokes ows,
and others, provide the opportunity for remote diagnostics and therefore for
ow control strategies. Students should make non-local eects part of their
thinking.
7.3 The role of vorticity
Vorticity can be taken as the leitmotiv for this course. Aside from the ele-
gance of potential ow, with its unique solutions and easy phenomenology,
vorticity is responsible for the complexity and/or beauty of most of uid
mechanics.
Vorticity comes in many contexts: kinematics with two of the Helmholtz
theorems, inviscid dynamics with Kelvins theorem and Helmholtzs other
theorem, and viscous ows of all types, instabilities and secondary ows and
rotating ows in the following chapters, and so many more topics.
Chapter 8
Narrow Flows
A distinct type of approximation is associated with dierent scaling in dier-
ent directions: narrow ows encompass boundary layers, but also jets, mixing
layers, shear layers, wakes and similar congurations where the streamwise
length scale is much larger than transverse scales. The simplications to the
NS equations are less drastic than in the limits of small or large Re, but still
sucient to allow solutions. More importantly, they include distinct ow
physics and phenomena (rst recognized by Prandtl) that may need to be
taken in consideration more generally. Again we focus on broad phenomena
and useful techniques: in-depth study is left for courses on viscous ows and
turbulent ows.
The movie by Abernathy (Fig. 8.1) serves as a useful introduction. A
rst viewing will provide useful background, on which the analysis below can
build; a second viewing, as time allows, is recommended.
8.1 Flat plate boundary layers
Undergraduate classes (uid mechanics, convective heat transfer) should
mention some basics about boundary layers (BL). Beyond a basic description,
a control-volume analysis is repeated below; further details can be obtained
by using approximate velocity proles in Kàrm` ans integral method, with
which a graduate student in uid mechanics should be familiar (see, possi-
bly, convective heat transfer course; more comments below).
Flat plate, of course, is code for something other than the geometry:
what it really means is that the freestream speed is constant, which in turn
167
168 CHAPTER 8. NARROW FLOWS
F. Abernathys movie: Boundary layers
narrow ow, solid boundary
at plate means zero pressure gradient (ZPG)
growth in streamwise direction
wall stress larger near leading edge
eect of pressure gradient
vorticity, circulation
Flat plate: vorticity from leading edge, not from wall surface
Note: separation ahead of blu body! see chapter on ow
separation.
transition to turbulence
Figure 8.1: Basics of boundary layers, in F. Abernathys movie.
8.1. FLAT PLATE BOUNDARY LAYERS 169
Figure 8.2: ZPGBL denition sketch
implies, through Bernoullis equation, that the streamwise pressure gradi-
ent vanishes. Thus, a more technically correct terminology is zero-pressure-
gradient boundary layer (ZPGBL). 2-D steady ow is assumed.
Then, the Cartesian directions are very distinct. The streamwise direction
includes the primary motion, and will be our x-direction. Normal to the plate
is the transverse direction, labeled as y. Normal to both is the spanwise
direction, parallel to the leading edge of the plate, as the z-direction. In 2-D
ows, there are no variations or velocity component in the z-direction. The
distance along the plate will be denoted indierently as x or as a generic
distance L, freestream velocity is U
.
At large Reynolds numbers Re = U
L/, we saw in Ch. 5 that inviscid

ow is a rational approximation in the freestream; with a uniform (therefore
irrotational) ow upstream of the plate, Bernoullis equation applies there.
However, close to the plate, the boundary layer introduces new smaller length
scales, the boundary layer thickness (x) prominent among them (no confu-
sion with Diracs function seems likely in this context). Within this layer,
viscous eects are not negligible: they carry the eect of the no-slip boundary
condition at the wall (y = 0) into the ow eld. Our equations must reect
this scaling. We will use L for streamwise scaling, for transverse distances.
Figure 8.3: Flat plate boundary layer conguration
8.1.1 Control volume analysis
The at plate boundary layer (zero-pressure gradient, steady ow) can be
studied at the undergraduate level; the resulting information is important
and should be familiar to all students. The conguration is shown on Fig.
8.3. Axes (x,y) at the leading edge of the plate are omitted.
First mass balance. The manipulations follow the same idea as above,
and are presented more concisely
t
m +
(U dA) = 0
= (
A
1
+
A
2
+
A
3
+
A
4
)(U dA)
= 0 + (U
A
2
+
A
3
V dx +
A
4
Udy) (8.1)
Indeed there is no mass ux through the impermeable wall (1). Note the
negative sign at 2, positive at 3 and 4 consistently with the diagram. Since
U at 4 is less than U
and the areas are equal, the ux at 3 must be positive:
A
3
V dx =
A
4
(U
U)dy > 0 (8.2)

This result should gradually be seen as common sense: as the boundary layer
develops, less and less mass progresses in the immediate vicinity of the wall.
The mass ux decit in the wall region is balanced by a net mass ux away
from the wall, i.e. by a deection of the freestream away from the at plate
(Fig. 8.3). This is a good opportunity to review (or learn) the concept of
displacement thickness!
Similarly for x-momentum balance, the only applied force is the wall shear
stress
w
opposing the motion. We have
U(U dA) =
A
1
w
dx
= (
A
1
+
A
2
+
A
3
+
A
4
)U(U dA)
= 0 + (U
2
A
2
+
A
3
U
V dx +
A
4
U
2
dy) (8.3)
Note the dierent treatment of ux term (dot product) and component of
momentum. As a result
A1
dx = U
2
A
2
+U
A
4
(U
U)dy+
A
4
U
2
dy =
A
4
U(U
U)dy
(8.4)
The friction force is proportional to the decit in momentum ux at the
exit section. See references about dening displacement and momentum
thickness, and their interpretation.
8.1.2 Scaling: mass balance
Then, mass balance provides useful information: for 2D ow,
x
u +
y
v = 0 (8.5)
corresponds to order-of-magnitude scaling for the two terms as U
/L and
V/ for some transverse velocity scale. The control volume analysis already
showed that the mass ux defect near the wall results in a deection fo the
freestream: we get the same idea from local mass balance, it is always nice
to have a consistent picture! Here, since U
/L does not vanish separately

(since uid moving at U
at the leading edge comes slowly to near-rest far

downstream), it must be balanced by the transverse term, therefore
U
/L V/ or V U
L
U
. (8.6)
The deection of the freestream is small: in fact, the angle is of order /L.
8.1.3 Scaling: streamwise momentum
In the x-direction, the equation is
u
x
u + v
y
u =
1
x
p + (
2
xx
u +
2
yy
u) (8.7)
Let us denote as the (unknown) order of magnitude of the pressure vari-
ation along the boundary layer: although the freestream is at constant p
for ZPG, we cannot assume this to hold inside the BL, we need to prove it.
Then, order of magnitudes of the successive terms are, in order:
u
x
u U
1
L
U
= U
2
1
L
v
y
u U
L
1
= U
2
1
L
1
x
p
1
2
xx
u
1
L
2
U
=
U
2
L
2
2
yy
u
1
2
U
=
U
2
(8.8)
We see that the convective terms are comparable, and of order U
2
/L. This
is a common feature of boundary layer approximations, except in the case of
convective heat (or mass) exchange at large Prandtl (or Schmidt) number.
Also a generic feature of narrow ows, the x-contribution to the Laplacian is
negligible relative to the y-contribution.
8.1.4 Scaling: transverse momentum
In the y-direction we have
u
x
v + v
y
v =
1
y
p + (
2
xx
v +
2
yy
v) (8.9)
Let us denote as the (unknown) order of magnitude of the pressure varia-
tion across the boundary layer. Then, order of magnitudes of the successive
terms are, in order:
u
x
v U
1
L
U
L
= U
2
L
2
v
y
v U
L
1
L
= U
2
L
2
1
y
p
1
2
xx
v
1
L
2
U
L
=
U
3
L
3
2
yy
v
1
2
U
L
=
U
L
(8.10)
Again the convective terms are comparable, and of order U
2
/L
2
. And only
the y-contribution to the viscous term remains.
8.1.5 Scaling: pressure
Thus we have
u
x
u + v
y
u =
1
x
p +
2
yy
u
U
2
1
L
1
L

U
2
(8.11)
and
u
x
v + v
y
v =
1
y
p +
2
yy
v
U
2
L
2
1
L
(8.12)
Consider the pressure terms. Because and represent the departure
from freestream conditions, we can assume that < (check for consistency
below), so that
L
<

L

U
2
L
or
U
2
(8.13)
in order to satisfy the x-momentum balance. But from y-momentum, we
have
L

U
2
2
L
2
or
U
2
L
2
(8.14)
So, in fact, we can conclude that

2
L
2
(8.15)
This shows that the pressure variation () is negligible across the bound-
ary layer, regardless of streamwise pressure gradients. Therefore, the y-
momentum equation inside the boundary layer reduces to
y
p = 0, (8.16)
which in turn implies that
x
p = d
x
p = d
x
p |
= 0 (8.17)
for ZPGBL. This result is extremely important in practice: since every solid
object immersed in a real ow experiences no-slip and related viscous eects
the application of Bernoullis equation to static pressure taps at the surface
is unwarranted (review Ch. 5 and the consequences of Croccos theorem for
the Bernoulli expression in the presence of shear). However, in a boundary
layer (as opposed to a recirculating ow region), we have just shown that
the wall pressure is equal to the freestream pressure, which vindicates the
measurement. Sometimes, things work out for the best!...
8.1.6 The ZPG BL equations
The ZPG BL equations are: for y-momentum
y
p = 0 (8.18)
as discussed above; for mass
x
u +
y
v = 0 (8.19)
which gives the v-component of velocity as a result of the kinematic constraint
of continuity; and for x-momentum
u
x
u + v
y
u =
2
yy
u (8.20)
These equations bear the name of Prandtls student Blasius,who rst derived
and solved them.
Before we follow Blasius, the value of scaling analysis is illustrated by
combining the scaling of the convective terms (of order U
2
/L) and the viscous

term (or order U
/
2
). Combining the two, we get
x

1
Re
x
. (8.21)
Comparing with a result of dimensional analysis, which would give /x as
an unknown function of Re
x
, the more precise expression results from the
addition of actual dynamics. The proportionality constant is out of reach for
scaling: that is one reason to go to the next level of analysis.
8.1.7 Similarity
The word similarity is used with two completely dierent meanings in this
course. We have already encountered the idea of dimensional similarity,
which enables us to scale systems up or down: we can study aerodynamics
in a water tank, make measurements on scale models of vehicles or pumps;
and, of course, do the same to the equations we solve numerically! Here
we encounter the other important meaning: students should make sure the
distinction is clear in their minds (based on context, of course!)
Shape similarity can be exact, as in this section; it can also be approxi-
mate, which expands its applicability to situations (such as convective heat
transfer, turbulent ows, etc.) where it gives good results with moderate
mathematical skill. The idea is the same, regardless. Consider two (or more)
velocity proles in the developing boundary layer The prole meets the fol-
lowing constraints: freestream velocity for y > , reached smoothly so that
y
u = 0 at the edge of the BL (no viscous stress there!); and no-slip at the
wall. If we replot the proles in terms of dimensionless u/U
and y/, these

constraints and the expectation of simple shapes leave very little room for
variations: this is all about shape. So, it is quite possible that the velocity
proles, properly scaled, are similar enought to be indistinguishable when
superposed.
With this observation in mind, we might be fortunate enough for simi-
larity to be mathematically correct. Exact similarity requires a combination
of dynamics and boundary conditions (as does separation of variables, re-
member?) that should not be taken for granted: but it is worth a try.
Mathematically, we assume, subject to a successful eventual outcome, that
u(x, y) = U
g(
y
(x)
) (8.22)
as the analytical equivalent of our graphical idea. It says that the complicated
eld dependence on x and y can actually be reduced to a dependence on
the single (scaled) variable = y/(x). The unknown function g captures
the shape of the similar proles. To satisfy continuity, we shall use the
Figure 8.4: Similarity in ZPGBL
streamfunction (remember: using does not presume irrotational ow) such
that u =
y
and v =
x
. In terms of , the similarity assumption takes
the form
= U
f() (8.23)
for some function f; the length scale is dimensionally necessary, and gives
the correct equation for u (below); it is proportional to the boundary layer
thickness. Using the chain rule (brush up as needed!) we get
u =
y
=
1
= U
(8.24)
so that g = f
, with the apostrophe denoting the derivative of the function

with respect to its only independent variable . Similarly (a little more
complicated, but quite doable) we have
x
u = U
y
u =
U
2
yy
u =
U
2
f
(8.25)
Substitution in the BL equation gives
U
2
f f
+
U
2
f
= 0 (8.26)
This equation contains two independent variables: x and . Similarity is
possible only if the x-dependence in the coecients factors out. Physically,
this means that the two terms (convective and viscous) remain in the same
ratio relative to each other, in other words that an equilibrium between them
has been reached. This requires
U
2
2
(8.27)
or
/U
(8.28)
or

x
U
(8.29)
with satises the leading edge condition = 0. If this is true, the prole
obeys the Blasius equation
ff
+ 2f
= 0 (8.30)
subject to the boundary conditions
f |
0
= f
|
0
= 0 (8.31)
(the second of which is the no-slip condition; do you recognize the meaning
of the rst one?), and
f
= 1 (8.32)
as well as
f
= 0 (8.33)
(and what might be the physical content of these relations?)
The Blasius equation has no known closed-form solution, but a numerical
solution is easily obtained. Agreement with experiments is excellent, which
seems to indicate that the ow chooses to be similar...
0 0.2 0.4 0.6 0.8 1 1.2
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Blasius profiles of velocity (blue) and vorticity (magenta)
y

/

d
e
l
t
a

(
x
)
Figure 8.5: Blasius proles for velocity and vorticity
Figure 8.6: Inexion point for ZPG, APG, FPG BLs
8.2. INTEGRAL METHOD: VON K
`
ARM
`
AN S APPROACH 179
Since the transverse velocity v vanishes at the impermeable wall (y = 0),
the BL equation reduces to
1
x
p =
2
yy
u at y = 0. (8.34)
For ZPG, it follows that the inexion point in the prole is at the wall. In
APG, the inexion point moves into the ow.
8.2 Integral method: von Kàrmàn s approach
The Blasius similarity approach will work only if similarity is mathemati-
cally exact. There will be, for example, very few types of pressure gradients
consistent with this very stringent condition.
The von Kàrmàn integral approach also relies on similarity, but in an
approximate sense only: if the shape of the velocity prole remains simi-
lar in a sketchy sense, the method applies. This makes it a valuable tool
for boundary layer with arbitrary pressure gradients, for natural and forced
convection, and related problems. The method is intermediate between the
dierential formulation, above, in which mass and momentum balance are
satised at every point in the eld, and the control volume approach, in
which the fundamental laws are satised over the entire ow and we ignore
the local details. In the integral method, the velocity prole is approximated
at each streamwise (x) location, integrated in the transverse (y) direction
and its streamwise evolution is calculated exactly.
We start with mass balance:
x
u +
y
v = 0 (8.35)
and integrate it from the wall to the edge of the boundary layer
(x)
0
x
u dy +
(x)
0
y
v dy = 0. (8.36)
The second integral is easy, the rst requires the use of the Leibniz theorem
on parametric integrals. We obtain
d
x
U
0
+
d
dx

0
u(x, y) dy + v(x, (x)) v(0) = 0 (8.37)
i.e. for impermeable wall (v(0) = 0)
v() = d
x
U
0
d
dx

0
u(x, y) dy (8.38)
As in other methods, the deection of the freestream is a consequence of
mass balance and the decit in mass ux near the wall. The similarity idea
can be introduced here: we will adopt for the velocity prole (y-dependence)
a simple function that captures the shape of the expected solution and can
be integrated easily. Common functions are polynomials, sines and exponen-
tials. The idea is that errors relative to the exact solution are smoothed out
in the integration process, although dierentiation will amplify the errors.
Here, for simplicity and to illustrate the resilience of the method, the crudest
approximation will be used:
u
U
0
=
y
(x)
, (8.39)
i.e. a straight line. Alternatives would be u/U
0
= 2y/ -y
2
/
2
or sin(y/2).
Substituting, mass balance gives
v() = d
x
U
0
/2 (8.40)
The Blasius momentum equation
x
(u
2
) +
y
(uv) =
2
y
yu, (8.41)
is also integrated, yielding
U
2
0
d
x
+
d
dx

0
u
2
dy + u()v() 0 =
y
u |
0
=
y
u(0) = U
0
/.
(8.42)
The main source of inaccuracy comes from the stress term, which involves
the derivative of the approximate prole: here, the straight line gives a poor
estimate of the wall stress. Executing the integral, substituting v() and
simplifying expressions gives
d
x
2
= 12/U
0
, (8.43)
or
/L =
12 Re
1/2
L
= 3.46 Re
1/2
L
. (8.44)
8.3. VORTICITY AND CIRCULATION IN ZPGBL 181
Compared to the exact (known in this case) solution, we get the correct
streamwise evolution with a 30% error on the numerical coecient (exact:
4.92). This is directly traceable to the underestimate of the wall stress. Bet-
ter shapes (e.g. parabola or sine) give 10 to 15% accuracy, which may be ac-
ceptable depending the situation. Note that, just as for the Blasius solution,
similarity changes the mass and momentum partial-dierential equations for
u and v into an o.d.e. for x. The advantage of the Karman method is that
it does not require exact (only approximate) similarity, so it is applicable to
many realistic engineering congurations. It is widely used in heat transfer
as well.
8.3 Vorticity and circulation in ZPGBL
Vorticity and circulation are related by Stokes theorem, but they are distinct
concepts. In ZPG boundary layer, there is no vorticity in the (uniform)
freestream, but clearly
y
u is non-zero near the wall. While the wall
stress provides a continuous source of momentum sink (drag) along the wall,
the situation is quite dierent for vorticity.
The use of Kelvins theorem is impossible in the boundary layer (viscous
eects indeed). But we can consider several rectangular contours sketched
as a rectangular box one edge along the wall and one in the freestream. with
a corner at the leading edge. As we build successive contours anchored at
the wall and extending into the freestream, we realize that they all have the
same circulation, i.e. the same average vorticity. So, where is the vorticity
coming from?
Let us look at vorticity for a complementary view. The BL equations
(ZPG, still) give
u
x
u + v
y
u =
y
(8.45)
At the wall, u and v vanish, so that
y
= 0. By analogy e.g. with Fouriers
law of heat conduction, this shows that there is no vorticity ux from the
wall. Yet, vorticity is present: it decreases downstream, and spreads with
the boundary layer itself.
The only explanation consistent with these facts is that all the vorticity
is introduced at the leading edge, and from there it diuses outward and is
convected downstream. The vorticity balance is
u
x
+ v
y
=
2
yy
(8.46)
Figure 8.7: Circulation in ZPG BL
similar to the convection diusion of temperature, but with an adiabatic
(no conduction at the wall) boundary condition beyond the leading edge.
We will revisit this important question in relation to boundary layer sep-
aration in the next Chapter.
Vorticity proles are easily calculated from the Blasius solution.
8.4 Descriptive: BL transition
This section is purely descriptive, with the analysis left for another course.
At the level of ideas, inviscid shear layer instability (Kelvin-Helmholtz) will
be studied in Ch. 11; for a viscous free shear layer (away from boundaries),
the point of inexion of the velocity proles are the most unstable. How-
ever, in the case of the ZPG boundary layer, the constraint associated with
the proximity of the wall causes the rst instability to occur in the critical
layer, slightly away from the wall. The complete analysis leading to the
Orr-Sommerfeld equation can be studied as part a course on viscous ows.
Here, it is enough to notice that, in the case of APG, the inexion moves
away from the wall, and might be expected to become unstable at lower Re
8.5. JETS 183
Figure 8.8: Vorticity proles for ZPG, APG, FPG BLs
than in the ZPG case.
At the descriptive level, instability in the spanwise vorticity distribu-
tion near the wall results in periodic bunching of the vortex lines: these are
the Tollmien-Schlichting (TS) waves, precursors of instability. These waves
develop an instability rst in the streamwise direction, and the induced ve-
locities cause the leading sections of these vortices to lift away from the wall,
forming the Klebano -vortices. These 3D structures interact strongly with
the ambient ow, leading to the formation of turbulent spots and eventually
fully developped turbulent BL. Transition in undisturbed ZPG BL occurs
around Re 5 10
5
.
Discuss why the critical Re are so dierent in pipe ows and BL.
8.5 Jets
Plane jets, circular jets
Laminar jets: the analysis has shortcomings (instablity, transition to tur-
bulence) but illustrates many of the same concepts as BL and introduces
some new concepts (momentum ux, entrainment) important enough to jus-
tify the pages because they apply to turbulent ows as well.
Re 0 Oozing in all directions
Inertia eects at moderate Re give the jet appearance, but transition and
turbulence.
Narrow ow, same scaling, so the BL equations are still valid:
y
p = 0 (8.47)
Figure 8.9: BL transition: TS waves, -vortices, and turbulent spots
so ambient pressure permeates the entire jet;
x
u +
y
v = 0 (8.48)
u
x
u + v
y
u =
2
yy
u (8.49)
What is dierent is the boundary conditions, they make the ow what it is:
we have
y : u = 0
y = 0 : v = 0 and
y
u = 0 (8.50)
Velocity proles meeting these boundary conditions are easy enough to
sketch, and the idea of similarity seems applicable (one notable dierence:
which the freestream ZPGBL velocity was constant, the centerline velocity
of the jet decreases downstream as a result of mass balance). Denoting the
centerline velocity as U
0
and some measure (to be made precise later) of the
width of the jet as , the similarity assumption is that the equations admit
solutions of the type
u(x, y) = U
0
(x)g(
y
(x)
) = U
0
(x)g() (8.51)
8.5. JETS 185
Figure 8.10: About ZPG boundary layers
0.5 0 0.5 1 1.5 2 2.5 3 3.5 4
2.5
2
1.5
1
0.5
0
0.5
1
1.5
2
2.5
T
r
a
n
s
v
e
r
s
e

d
i
s
t
a
n
c
e
Laminar plane jet
Velocity profiles (blue), axial spreading (red)
Figure 8.11: Denition sketch for the plane jet
0.5 0 0.5 1 1.5
2.5
2
1.5
1
0.5
0
0.5
1
1.5
2
2.5
Similarity for the plane jet
u(x,y) / u(x,0)
y

/

(
x
)
Figure 8.12: Similarity of proles in a jet
8.5. JETS 187
8.5.1 Streamwise development
To simplify the presentation, it is useful to introduce as an additional as-
sumption the power-law dependences
U
0
x
m
and x
n
(8.52)
A full analysis would lead us to these conditions, assuming them is just a
shortcut that will be supported by its outcome. Then, the x-dependence of
the convective and viscous terms is, respectively
u
x
u x
2m1
2
yy
u x
m2n
(8.53)
and since they must remain in lock-step for similarity to be possible, we must
have
2m1 = m2n (8.54)
One equation, two unknowns: this is not sucient to solve.
The additional condition can be derived from a control-volume approach,
to satisfy overall momentum balance. Let the control volume consist of
two planes normal to the jet, and extending to innity. Since the pressure
is uniform (no net normal forces on the planes) and there are no viscous
stresses at innity, the momentum balance reduces to
d
x
U
2
dy = d
x
M = 0 (8.55)
where M is the momentum ux through any section of the jet. Substituting
the x-dependence of U and , we have
M =
U
2
dy = U
2
0
g
2
d x
2m+n
(8.56)
Since M is independen of x, the second equation is
2m + n = 0 (8.57)
Therefore, similarity requires that
m =
1
3
and n =
2
3
(8.58)
0.5 0 0.5 1 1.5 2 2.5 3 3.5 4
3
2
1
0
1
2
3
Control volume across the jet
dx
Figure 8.13: Control volume for the jet
This can be rewritten as
U
0
= C(
M
2
2
x
)
1/3
(8.59)
for some constant C, and
= (
2
x
M
)
2/3
(8.60)
8.5.2 Similarity proles
Continuity is satised by using a streamfunction, and similarity leads us to
= C(
Mx
)
1/3
f() (8.61)
Carrying out the derivatives (chain rule again), the equation for the velocity
proles is
3
C
f
+ ff
+ f
2
= 0 (8.62)
with the boundary conditions
= 0 : f = f
= 0
8.5. JETS 189
: f
0 (8.63)
This equation can actually be integrated! The solution is
f = Atanh(
AC
6
) (8.64)
for some constant A; this gives the velocity prole
u(x, y) = (
3M
2
32
2
x
)
1/3
sech
2
[y(
M
48
2
x
2
)
1/3
] (8.65)
(Looks messy, but view it as a little miracle!
Clearly, this solution cannot truly originate from a source at x = 0, where
it is singular. But the singularity spreads out extremely rapidly for very small
values of x.
8.5.3 Entrainment
One notable phenomenon associated with shear ows is apparent here: for
the control volume used above, monitor mass balance. The mass ux through
one of the section planes is

udy. For the velocity scaling obtained above,
it is easy to see that
d
x
udy > 0 (8.66)

which implies that the jet is drawing mass ux from innity! This phe-
nomenon is known as entrainment: while viscous forces merely redistribute
momentum, widening the jet as it weakens, it also bring more an more mass
into the motion, and the net eect is to increase the mass ux through suc-
cessive sections.
The following idea (Tritton p138) may not be provable in general, but is
easily established for self-similar jets: consider the three relations
1. entrainment: d
x
udy > 0
2. momentum: d
x
u
2
dy = 0
3. energy dissipation: d
x
u
3
dy = 2

(
y
u)
2
dy < 0
(Refer to Ch. 3 for the energy balance.) Then the (in)equalities for momen-
tum and energy imply entrainment! (i.e. for self-similar jets, statements 2
and 3 above imply 1).
ing
Many narrow-ow problems admit approximate similarity, which is sucient
to use von Karm ans integral method. See a course on viscous ows or on
convective heat transfer for good examples. One problem is suggested below,
as a simple example. Others in Currie p.318.
Adverse, favorable pressure gradients, Falkner-Skan solutions: see viscous
ows.
BL transition, bypass transition.
Problems
1. For self-similar laminar jets, prove Trittons conjecture that entrain-
ment is implied by momentum
2. Map the ideas, with emphasis in similarities and dierences, in the
solution for laminar jet and ZPG boundary layer.
3. Von Kàrm` an integral method: Start from the BL equations (mass,
momentum), and assume an approximate solution of the form u(x, y) =
U
e
y/(x)
. The exact equations are not satised, but after integration
for y = [0, ) an overall balance of mass and momentum is satised:
nd how depends on x if this is true, and relate to the boundary
layer thickness.
4. Consider the accelerating ow away from a stagnation point, with U =
a x (x is the the distance from the stagnation point). The Reynolds
number is then a x
2
/, and is assumed to be large. Show that the
boundary layer thickness is constant. Assume similarity in the form
= k a x f(y/). Then, show that k = =
/a and that the O.D.E.

for f is f
+ ff
f
2
+ 1 = 0. Determine appropriate boundary
conditions. (Problem 32 p.476 from Tritton.)
5. Consider the 2-D wake of an object placed in a freestream with velocity
u
0
. Dene M =
u (u
0
u) dy. Show that dM/dt = 0. Relate M
to the drag force on the object. (Problem 35 p.477 from Tritton.)
Chapter 9
Flow Separation and Secondary
Flow
Secondary ows imply a primary ow. There are many variants of this idea.
In this chapter, we leave aside secondary ows that can be attributed to
instabilities of the primary ow, which rely on linearization (see Ch. 11) of
the equations for disturbance to the baseline solution and may be irrotational.
Here, vorticity always plays an essential role, for example when its direction
is modied by the ow (ow in a bend) or when vorticity is added at the
boundaries.
In some instances, the secondary ow is a region separated from the
primary ow by a streamline that attaches to smooth surfaces or sharp edges.
Examples (Fig. 9.2) are encountered in pipe entrance ows, aerodynamics,
and wake formation: in all instances the performance is aected, and it is
imperative to include the correct physics in any model of such ows. The
entire range of Reynolds numbers is aected.
In this chapter, we only peel the outermost layers of these dicult topics.
See Fig. 9.1)
9.1 Curved channel
First we examine the secondary ow in a curved channel (possibly a model
for a bend in a river) (Fig. 9.3). The primary ow would be irrotational and
governed by Eulers normal equation: pressure must be larger on the outside
in order to redirect the ow inward; therefore the speed must increase inward.
191
192 CHAPTER 9. FLOW SEPARATION AND SECONDARY FLOW
E.S. Taylors movie: Secondary ows
All dependent on vorticity (potential ow have unique solutions).
Flow separation
vorticity from boundary, requires pressure gradients (see lec-
tures)
No ow separation
Flow in a bend: material lines and vortex lines
Is the bathtub vortex really 2-dimensional?
Necklace vortex
Ekman layers
Figure 9.1: Secondary ows: the role of vorticity, movie by E.S. Taylor
9.1. CURVED CHANNEL 193
Figure 9.2: Secondary ow: creeping ow in a wedge, forward-and backward-
facing steps, diuser, wake of a sphere, necklace vortex around an obstacle
in a boundary layer.
Figure 9.3: Irrotational ow in a curved channel
The simple version, sometimes found in undergraduate texts, is possible if
we ignore friction at the walls.
The presence of a secondary ow, invalidating the simple irrotational
model, can be explained by the presence of a mean shear in the prole (faster
ow at the surface, no-slip at the bottom). The upstream conditions should
include vorticity in the spanwise direction and this is key to the next level of
analysis.
It is assumed that viscous eects are not dominant over the short travel
time through the curved portion of channel. One can rationalize that viscous
diusion, though present, does not qualitatively aect the reasoning. Under
this assumption, Helmholtz theorem on vortex lines being material lines
does apply as an approximation, and we track the progression of a material
line through the bend. Since the inside of the curve moves faster than the
outside, the material line will rotate relative to the mean streamline at the
center of the channel. Since the material line is also a vortex line, there will
be a vorticity component in the streamwise direction! (Fig. 9.4)
This streamwise vorticity introduced by the bend is responsible for the
circulation of water outward at the free surface and inward along the bottom
of the channel. This mechanism may be responsible for the meandering of
rivers, and for the erosion patterns (steep banks outside, sediment deposit
inside the curve).
9.2. VORTICITY REVERSAL IN 2D SEPARATION 195
Figure 9.4: Creation of axial vorticity in a curved channel
A similar situation arises for laminar ow in a circular helical pipe (or in
a bend) (Fig. 9.5). Two counter-rotating axial vortices arise, recirculating
the uid across the mid-plane of the pipe. This is a simple mechanism of
mixing enhancement, used in some heat exchangers. Unlike the case of the
open channel, here we see streamlines actually losing contact with a solid
surface: that is ow separation.
9.2 Vorticity reversal in 2D separation
Flow separation is possible in potential ow (e.g. ow toward a stagnation
point, with separation streamline, Fig. 9.6), but, with the presence of vortic-
ity, the range of possible phenomena expands enormously. We limit ourselves
to 2D ows, and start from the vorticity balance (without vortex stretching,
of course)
t
+ U =
2
(9.1)
Figure 9.5: Secondary ow in a helical pipe
3 2 1 0 1 2 3
0
0.5
1
1.5
2
2.5
3
Potential flow at a stagnation point
x
y
Figure 9.6: Potential ow separation at a stagnation point
9.2. VORTICITY REVERSAL IN 2D SEPARATION 197
Figure 9.7: Separation in a 2D boundary layer
In a 2D BL-type conguration (Fig. 9.7), the only component of vorticity
is normal to the plane of the sketch:
=
x
v
y
u (9.2)
implying
y
=
2
xy
v
2
yy
u. (9.3)
Then, we look at continuity:
x
u +
y
v = 0 (9.4)
Right along the wall, the u component is zero (no-slip), as is the v component
(impermeable wall) regardless of x, which implies that
y
v |
y=0
=
x
v |
y=0
= 0 (9.5)
It follows that
2
xy
v |
y=0
= 0. (9.6)
So, without any dynamics yet, we conclude that
y
=
2
yy
u (9.7)
Figure 9.8: Vorticity reversal in front of a step
at the wall. This implies that changes in vorticity, including changes in sign,
away from the wall is necessarily related to changes in stress (viscosity not
part of the kinematics, of course).
So the question is: how is reversed vorticity introduced? This needs to be
thought of in the broader context of the introduction of vorticity of any sign
into a ow: from Kelvins theorem, we know this to be a viscous phenomenon.
But the ZPGBL was not all that enlightening in this regard, with all the
vorticity introduced at the leading edge. The following analysis represents a
more detailed look at the eect of pressure (surprise!) on vorticity
1
In fact, taking the BL equation (with pressure gradient leading to the
step), and looking a the immediate vicinity of the wall (y 0), we have
0 =
1
x
p +
2
yy
u (9.8)
At the wall
1
x
p =
y
(9.9)
1
So, when someone makes a statement to the eect that there is no pressure in the
vorticity equation, youll know the pressure eects are hidden but possibly present!
9.3. INTRODUCTION OF VORTICITY 199
so that the conduction ux of vorticity from the wall is proportional to the
streamwise pressure gradient. When pressure increases, (positive) vorticity
diuses from the wall into the ow.
9.3 Introduction of vorticity
Kelvins theorem (Ch. 5) gives us conditions under which an irrotational ow
remains irrotational (although the theorem is not just about irrotational ow,
of course!). It is complemented by an understanding of how vorticity may be
introduced. (A parallel with Bernoullis equation (Ch. 5), complemented by
losses (Ch. 3) and Croccos theorem (Ch. 5) for mechanisms of change for
B, is in order.) We are nally in position to do this.
Of the ve mechanisms listed below, three involve the eects of walls
(boundaries), while two operate in the bulk of the uid.
1. In the Shapiro movies, we saw how, in the presence of density varia-
tions, the vorticity equation has a source term proportional to p,
so that a misalignment of the pressure and density gradients does in-
troduce vorticity. This mechanism is inviscid, and can work at uid
interfaces (remember the sloshing uid in a container).
2. More generally, rotational forces (Coriolis force in rotating systems
see Ch. 10 ; Lorentz force in conducting uids subject to magnetic
elds; etc.) can introduce vorticity in the ow.
3. Then, in Ch. 8, we saw the singularity at the plates leading edge intro-
duces vorticity at one point. For ZPG, no other vorticity is introduced.
Because the ow is not dierentiable at sharp corners and such, one
may think of these singular points as a crude representation of regions
of intense pressure gradients, under the following item.
4. But, in the presence of pressure gradients, vorticity diuses in from the
wall. This is an important mechanism of ow separation in adverse
pressure gradients, with many complicating factors in 3D ows.
5. Finally, unsteady ow near a wall can also lead to diusion of vorticity
from the wall: at the wall the momentum equation
t
u +
1
x
p =
y
(9.10)
Figure 9.9: Vorticity introduction: singular points (Ch 8), pressure gradients
near walls, density gradient, rotational forces (Ch 10), unsteady motion near
walls (Ch 6.
applies to Stokes second problem (no pressure gradient) )(see Ch. 7).
Once present somewhere in the ow, trace amounts of vorticity spread ev-
erywhere by diusion: think about the eect Greens function. The decision
to be made by the analyst is about the importance of trace vorticity in each
given situation.
ing
Because ow separation aects dramatically the performance of aerodynamic
surfaces and wake formation, it is a very active eld of current research. The
control of separation is particularly important.
Problems
1. Describe changes in vorticity proles in pictures of ow separation.
2. Construct a mind map about the role of vorticity in ow separation.
3. Discuss the dierences between secondary ow separation, and the sep-
aration occuring in potential ows (e.g. ow around a cylinder).
Chapter 10
Rotating Flows
Flows in rotating frames of reference include atmospheric ows and ows
in turbomachinery. Fictitious inertia forces are added to the Navier-Stokes
equations: the centrifugal force is easily included the conventional framework,
but the Coriolis force introduces some new physics. Relevant dimensionless
parameters are the Ekman and Rossby numbers.
The case of dominant Coriolis force corresponds to a small Rossby num-
ber. See Fig. 10.1.
10.1 Equations of motion
The equations of motion in a rotating frame of reference can be found in
most dynamics textbooks. The derivation relies on the observation that, as
the base vectors rotate at constant angular velocity (Fig. 10.2), we have
t
e
i
= e
i
. (10.1)
Note that is the rotation vector (Poisson vector) of the frame of reference
as seen from an assumed inertial reference. Then, the second derivative of
the vector position r = x
i
e
i
is easily obtained
2
tt
r =
2
tt
x
i
e
i
+ 2
t
x
i
e
i
+ x
i
( e
i
). (10.2)
In addition to the apparent acceleration, two new terms appear: the Coriolis
acceleration which depend on velocity, and the centripetal acceleration which
depends on position relative to the axis of rotation.
203
204 CHAPTER 10. ROTATING FLOWS
D. Fultzs movie: Rotating ows
Coriolis force: motion of vortex ring (self-propelled) or similar
object
Rossby number, geostrophic ow
linear equations
2-D: Taylor-Proudman theorem
Taylor columns
non-local eects:
2
Rossby waves
atmospheric motion: large scale only
Ekman layers: secondary ows
Figure 10.1: Small Rossby number ows, movie by D. Fultz
Figure 10.2: Inertial and rotating frames of reference
10.2. SCALING 205
Moving these two accelerations to the other side of Newtons second law,
they become ctitious forces (with signs reversed), and the Navier-Stokes
equations become
t
u + u = (
p
+
u
2
2
+ gz) (r) 2 u. (10.3)
The rst added term is the centrifugal force, quadratic in . Because it
can be rewritten in the form
( r) = (
1
2
2
r
2
) (10.4)
where r
denotes the distance from the axis of rotation, the centrifugal term
is easily combined with the pressure term.
The term linear in is the Coriolis force. It is essential in explaining our
large-scale weather systems, among other features. This chapter revolves
around it.
Questions for discussion: Is your lab an inertial lab? Rotor?
10.1.1 Coriolis force
Eects of the Coriolis force are illustrated in the movie shown in class. The
trajectory of a vortex ring in the rotating tank is a good example (Fig. 10.3).
Note that the direction of rotation is actually for tank rotation as seen from
(inertial) lab.
A more dramatic eect is found in the earths atmosphere, with the dom-
inant westerlies at mid-latitudes (Fig. 10.4).
Of course, many complicating dynamics are encountered in applications.
But an understanding of what happens when rotation is dominant serves as
a caution for what might happen in combination with other factors: simple
rst!
10.2 Scaling
As usual, with length and time as the only dimensions, we can take two
scaling quantities: some velocity and length scales U and L are the usual
default, but the imposed angular speed will identify relevant ranges of
L. Then, ignoring gravity and centrifugal forces, the various terms in the
momentum equation can be estimated:
Figure 10.3: Sketch from movie: trajectory of a vortex ring in rotating tank
seen from above
Figure 10.4: Dominant westerly winds at mid-latitudes as a consequence of
buoyancy-induced equatorial-polar circulation and of the Coriolis force
10.3. GEOSTROPHIC APPROXIMATION 207
1. u u
U
2
L
2. 2 u U
3.
2
u
U
L
2
.
Since we are exploring the eects of rotation, the Coriolis term cannot be
neglected, and we compare all other to it by dividing throughout by U.
Then, we have
1. Coriolis term: 1
2. Time derivative:
1
t
, which indicates that, in the absence of other
forcing, the characteristic time scale will be of the order of the period
of rotation.
3. Convective term:
U
L
= Ro, which denes the Rossby number
4. Viscous term:

L
2
= Ek, which denes the Ekman number.
Note that we have no way to evaluate the pressure term directly. Various
choices can be made at this stage. The case of Ek 1 is important near
walls, or at the surface of the ocean: in the Ekman layer, the direction of
motion is not the same as the direction of the applied stress, a fact known
to sailors long ago and explained by Ekman. See the ow visualization of
Ekman layers in the movie shown in class. Ignoring boundary layers for
lack of time, we assume the Ekman number to be very small (large scale,
small viscous eects). Then, away from solid surfaces and interfaces, the
Rossby number tells us how important rotation is: the limiting case of interest
here is of dominant Coriolis forces, i.e. Ro 1. For given U and as
scaling quantities, this corresponds to large scale ows: hurricanes rather
than tornadoes! But note that the relevant scale depends on the apparent
speed of motion and on the rotation speed.
This limit Ro 1 is known as the geostrophic approximation, and is
very important for large-scale meteorology.
10.3 Geostrophic approximation
Extreme case, the phenomena associated with rotation takes their most ex-
treme form (and the equations are simpler!) Assumptions: steady ow, small
Rossby and Ekman numbers
Ro =
U
L
1 (10.5)
Ek =

L
2
1. (10.6)
Then, the viscous and convective terms drop out, and only the time derivative
and the pressure terms remain to balance the Coriolis force. The quasi-static
version (no external forcing to impose a time variation) is easiest:
2 u =
1
p (10.7)
The time derivative can be restored for time-dependent problems, e.g. Rossby
waves below.
The undergraduate understanding (from hydrostatics) of the pressure gra-
dient as a force should be revisited here: the uid motion is perpendicular
to the pressure gradient.
The geostrophic equations are linear for velocity, which makes them easier
to manipulate. Projecting them along streamlines, we get:
u p = 0 (10.8)
similar to Croccos equation, but for static pressure. There is no velocity in
the direction of the pressure gradient, so pressure is constant along stream-
lines. Furthermore,
p = 0 (10.9)
so that pressure is also constant in the direction of the axis of rotation.
Finally, taking the divergence of the geostrophic equation, we get
2
p = 2 (10.10)
so that the component of vorticity along the axis of rotations acts as a source
for pressure variations throughout the eld. This equation explains the di-
rection of rotation around high and low pressures a mid-latitudes Fig. 10.5.
A high pressure corresponds to maximum of p, therefore to a negative value
of the r.h.s.: with pointing toward the North Star, the dot product will
be negative for clockwise rotation in the northern hemisphere. This explains
the cyclonic and anticyclonic circulation at mid-latitudes (why not in tropical
regions?)
10.3. GEOSTROPHIC APPROXIMATION 209
Figure 10.5: Cyclonic/anticyclonic motion at mid-latitudes
10.3.1 Taylor columns
One of the most intriguing aspects of small Ro ows is the vorticity equation.
Taking the curl of the geostrophic (momentum) equation, we get for uniform
and divergence-free velocity
( )u = 0, (10.11)
i.e. the directional derivative of velocity (each and every component) along
the axis of rotation vanishes. This is known as the Taylor-Proudman theorem.
What does it say? We have learned to read directional derivatives, and
this is a good application: the directional derivative of velocity (the vector,
or any component of it) in the direction of the axis of rotation vanishes.
Interpreted, this means that the ow is two-dimensional, restricted to the
plane normal to the axis of rotation!
The movie seen in class shows many spectacular illustrations of this phe-
nomenon, for example:
in a rotating tank, a bump at the bottom eectively freezes the entire
column of uid above it, with the ow forced to avoid the column in
order to stay in-plane.
in a rotating tank, moving an obstacle up and down induces vorticity
of opposite signs below and above the obstacle.
The two-dimensionality of ow in rapidly rotating systems as a great impact
on turbulence, therefore on mixing. Keep this in mind when using turbu-
lence models (generally assuming 3-D properties) in rotating systems. Read
Tritton, Section 16.4, p.219-226, for additional information.
Summary of small-Ro approximation on Fig. 10.6.
10.4 Rossby waves
This section has some aspects in common with Ch. 11, except that no
linearization is required here: geostrophic motion is already linear! Tritton,
Section 16.7, p.232-238.
The conguration is shown on Fig. 10.7. The most signicant element,
beside rotation, is the layer of varying depth, and we denote the angle
= d
y
h. (10.12)
It can be shown that this simulates the variation of Coriolis force with lati-
tude. We assume constant depth in the streamwise direction
d
x
h = 0 (10.13)
and homogeneity in the direction of rotation. We cannot impose a no-slip
condition, since viscous eects scaled out of the problem.
Mass balance is given by
x
u +
y
v +
z
w = 0. (10.14)
Because u and v are independent of z, we have
2
zz
w = 0 (10.15)
therefore
z
w = C =
w
h
h
=
v
h
. (10.16)
Hence
x
u +
y
v +
v
h
= 0. (10.17)
10.4. ROSSBY WAVES 211
Figure 10.6: Ideas related to small Ro ows: geostrophic motion
Figure 10.7: Conguration sketch for Rossby waves
The dynamics are described by the unsteady geostrophic equations
t
u + 2 u =
1
p. (10.18)
Breaking it into components:
t
u 2v =
1
x
p
t
v + 2u =
1
y
p. (10.19)
The equations are linear in u and v. Eliminating pressure
y
p =
2
ty
u 2
y
v
=
2
tx
v + 2
x
u (10.20)
or
2
ty
u =
2
tx
v + 2(
x
u +
y
v) =
2
tx
v 2
v
h
. (10.21)
10.5. ADVANCED TOPICS AND IDEAS FOR FURTHER READING213
We need one more simplication (not to imply that our simplifying assump-
tions must be true in all cases, but merely looking for the simplest solution
at rst). It is conceivable that there are solutions with u independent of y
still allowing for v and w to vary with y. This yields an equation for v only:
2
tx
v
2
h
v = 0. (10.22)
This equation admits oscillatory solutions. Prompted by the experiments
(movie), we restrict our search to traveling waves, of the form
v = v
0
e
i(tx)
. (10.23)
For such waves, the partial operators
t
and
x
become algebraic operators i
and i, respectively. Substitution and rearrangement gives the dispersion
relation
=
2
h
(10.24)
From this, we can calculate the phase velocity (i.e. the velocity of crest
propagation) as
c
p
=

=
h
2
2
(10.25)
which depends on , and the group velocity (i.e. the velocity of energy
propagation)
c
g
=
=
h
2
2
= c
p
. (10.26)
Rossby waves are important to us. Remember that the jet stream is one
consequence of the Coriolis force. It is common for the jet stream to show
3 to 5 periods of oscillation around mid-latitudes: one dramatic example of
Rossby waves (Fig. 10.8).
10.5 Advanced topics and ideas for further
reading
Zonal motion on larger planets.
Ekman layers, ocean surface motion and drift relative to wind speed,
Langmuir circulation.
Figure 10.8: Rossby waves in the jet stream
Problems
1. Imagine a turbine rotor spinning at 5000 rpm. Between rotor blades,
the centerline air speed is of the order of 50 m/s, and the blade chord is 3
cm with an inter-blade distance of 1.5 cm. Evaluate the Rossby number
characteristic of the mean ow, and of the boundary layer turbulence
(mixing length of the order of the boundary layer thickness).
2. Identify nonlocal phenomena in ows with Ro 1.
3. Can geostrophic ow be potential?
4. Does Bernoullis equation apply to geostrophic ow, and in which form?
5. Based on respective equations, discuss similarities and dierences be-
tween Stokes ow and geostrophic ow.
Chapter 11
Linearization
This chapter combines wave motion and linear stability theory. Within the
time-frame of the course, only a few simple cases can be treated, with the
purpose of illustrating mathematical techniques and important physics. Be-
cause of the ubiquitous importance of shear-layer instability, and the need for
engineers to anticipate its presence and consequences, the Kelvin-Helmholtz
problem is presented here. Also, the comparison of surface waves and inter-
nal waves appears justied as an introduction to a very broad subeld with
a common underlying technique: linearization.
11.1 Surface waves
The interface between standing water and the surrounding air is familiar
enough, as is the propagation of disturbances on the surface (Fig. 11.1). This
is a canonical situation, involving a reference ow (here, trivially static); a
secondary ow of small amplitude that allows the linearization of the equa-
tions; and the (imaginary) exponential solution to these equations as prop-
agating waves. A variant on this idea is found in linear stability theory, of
which an example appears in the next section. The student is invited to
compare the two cases carefully.
Since the primary ow is static, it is certainly irrotational, and the veloc-
ity potential is a constant (which obviously satises the Laplace equation).
Let us take the body of water to be of nite depth H (a simpler solution is
obtained for H ). The free surface is described by
y = (x, t), (11.1)
215
216 CHAPTER 11. LINEARIZATION
Figure 11.1: Denition sketch for surface waves
with = 0 for the primary ow. The disturbance associated with the
motion of the interface is assumed to be irrotational also.
The interface, as primary descriptor of the ow, is characterized by two
features:
1. it is a material surface, for which the material derivative of the property
(y ) must be zero:
t
(y ) + u
x
(y ) + v
y
(y )
=
t
u
x
+ v = 0; (11.2)
2. it is at ambient pressure p = 0, so that Bernoullis equation at the
interface is
t
+
1
2
(u
2
+ v
2
) + g = C, (11.3)
and we can take C = 0 without loss of generality
For small amplitude disturbances, we neglect quadratic terms in the per-
turbations, and the interface equations become, respectively
v =
t
=
y
, (11.4)
where we make use of the potential ow assumption , and
t
+ g = 0. (11.5)
11.1. SURFACE WAVES 217
In the bulk of the body of water, the Laplace equation for the velocity
potential is already linear:
2
xx
+
2
yy
= 0. (11.6)
Note that the equation for the eld is more easily derived than the boundary
condition at the the interface.
Exponential solutions are common solutions of linear equations (think
about the solution of linear pdes using Fourier or Laplace transforms). Here,
we can look for propagating waves on the surface
= Ae
i(kxt)
, (11.7)
where A is the amplitude, and a real part can be extracted when needed. k
is the wavenumber corresponding to a wavelength = 2/k, and /2 is the
frequency. One interface condition becomes
v = i, (11.8)
while the other suggests that should be exponential also:
= f(y)e
i(kxt)
. (11.9)
Then, the Laplace equation reduces to
f
k
2
f = 0, (11.10)
so that
f(y) = C
1
e
ky
+ C
2
e
ky
. (11.11)
At the bottom of the pond y = H,
y
f = 0 so
0 = kC
1
e
kH
+ kC
2
e
kH
(11.12)
and
C
2
= C
1
e
2kH
= 2C e
kH
, (11.13)
introducing the constant 2C = C
1
e
kH
. Simple manipulations give
f(y) = C cosh(k(y + H)). (11.14)
Thus, we have the solution
(x, y) = C cosh(k(y + H))e
i(kxt)
, (11.15)
which must still satisfy the interface conditions
iC cosh(kH) + gA = 0 (11.16)
and
iA = kCsinh(kH). (11.17)
Solving presents no diculty, and we get
2
= g k tanh(kH). (11.18)
The relation between and k is called the dispersion relation . It governs
the propagation of gravity waves over standing water.
The procedure summarized on Fig. 11.2. See Acheson, Section 3.3, p.70-
74 for related material. It is noteworthy that the solution is valid regardless
of amplitude. Can you think of situations in which this would clearly not
be plausible? ( Discussion leading to shallow water waves, wave breaking
(sloping bottom), etc.; also amplitude relative to wavelength; etc. ) Wave
motion is a simple example of secondary ow (Fig.11.3).
11.1.1 Tsunami speed
The elementary analysis of gravity waves, above, is adequate to account for
the enormous speed of tsunami. The travel speed of the waves energy is the
group velocity shown on Fig. 11.4
c
g
=
d
dk
=
1
2
g tanh(kH)
k
(1 + 4kH
e
2kH
e
4kH
1
). (11.19)
For wavelengths much longer than the oceans depth (shallow water), kH
1 and the travel speed
c
g

gH (11.20)
increases obviously with ocean depth. For a typical ocean depth H
4000 m 2.5 mi (Pacic Ocean), this corresponds to a travel speed of
c
g
200 m/s 720 km/h 450 mph.
Shorter waves travel slower: e.g. for kH = 1, v
g
= .677
gH, with a rapid

decrease for short wavelength (deep ocean) waves. Note that the notion of
deep or shallow ocean is relative to the scale of the wave.
Figure 11.2: Gravity waves
Figure 11.3: Wave motion as a secondary ow
10
2
10
1
10
0
10
1
10
2
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
k.H
v
g

/

s
q
r
t
(
g
.
H
)
shallow water: k.H <<1
deep water: k.H >1
Figure 11.4: Tsunami speed is determined by the long-wavelength/small k/
shallow-water end of the gravity wave dispersion relation. Baseline shallow-
water group velocity
gh is the reference.
Figure 11.5: Internal waves
Dimensionally obvious in hindsight, the simple dependence of group ve-
locity on H leads to such enormous speeds that one might doubt the choice
of parameters, were it not for the supporting analysis or for the empirical
evidence.
11.1.2 Internal waves
The standing water exposed to the atmosphere, as above, is an extreme
case of density interface. In many situations arising in the atmosphere, in
the ocean or in industrial applications, the density dierence (caused, pos-
sibly, by gradients in temperature, salinity, chemical composition or bubble
concentration) is weaker and not necessarily concentrated at a sharp inter-
face. See Acheson, Section 3.8, p.86-89, for more details on internal waves.
Such density gradients can inhibit convective eects (inversion base in the
atmosphere, thermocline in the ocean and lakes), with signicant eects on
pollutant or nutrient dispersal. The wave motion in clouds, associated with
warmer air riding over cooler surface air, can be associated with local cycles
of condensation and evaporation which are a form of ow visualization, not
uncommon. The added drag on boats was known to ancient mariners (Fig.
11.5), even if the concept of wave energy propagation was not understood
then. The density interfaces can also deect or reect sound waves.
11.2. INVISCID LINEAR STABILITY: KELVIN-HELMHOLTZ 223
11.1.3 More advanced topics
Surface waves can occur as the result of various types of forcing, e.g. ap-
plication of shear stress (see next section). Sometimes, dissipative eects
(neglected above) can change qualitatively the nature of the ow (e.g. dis-
sipative structures) or simply limit the growth of the waves (saturation). In
the many cases where saturation occurs at amplitudes too large to justify
the linearization of the problem, nonlinear terms change the shape of the
solutions: in case of water waves, nonlinear terms introduce a cusp at the
top of the wave(see Van Dykes book, p.114). The creation of vorticity at the
interface cannot be neglected for some phenomena. In shallow water, nite
amplitude leads to solitary waves.
11.2 Inviscid linear stability: Kelvin-Helmholtz
Elementary stability theory has many common points with the study of wave
motion: basic ow, perturbation, linearization, exponential solutions. The
dierence is that the amplitude of the perturbations can increase exponen-
tially. One simple case has been illustrated many times in the movies seen
previously.
Every practitioner of uid mechanics should be aware of the possible
instability of shear layers, and of related phenomena. This section includes
the very simplest case, the inviscid shear layer (Fig. 11.6), which turns out
to be always unstable.
11.2.1 Setting up the problem
Above and below the shear layer, we have
1
= 0 and
2
2
= 0, (11.21)
with boundary conditions
lim
y
1
= (U
1
, 0, 0) (11.22)
and
lim
y+
2
= (U
2
, 0, 0). (11.23)
Figure 11.6: Denition sketch for shear layer instability
The interface is made of material particles, and we must match the Eu-
lerian description of the eld with the Lagrangian denition of the interface.
(Great opportunity to review Ch. 2.) Let us dene the interface as the
(Eulerian) line y = y
(x, t) marked by the Lagrangian position y(x

0
, 0, t)
y(x
0
, 0, t) = y
(x, t). (11.24)

At the interface, the vertical (Eulerian) velocity must match the (Lagrangian)
motion of the material points. Hence
v =
t
y
+ u
x
y
= d
t
y. (11.25)
Just above and below the interface (at y = y
), we have
v =
y
2
=
t
y
+ u
2
x
y
(11.26)
and similarly
v =
y
1
=
t
y
+ u
1
x
y
(11.27)
Clearly the same y
must be used, but the uid is sliding (no friction!) rel-

ative to the interface. This introduces a vortex sheet (singularity: the 2-D
equivalent of the potential vortex on a line), with potential ow on either
side, which is at the heart of the present problem.
We also need dynamics: in potential ow, that would be Bernoullis equa-
tion holds (the unsteady version, of course)
t
+
1
2
()
2
+
p
= C(t), (11.28)
all properties including the constants with indices 1 or 2. In the absence of
surface tension, the pressures must match at the interface: p
1
= p
2
, therefore
1
+
1
2
(
1
)
2
C
1
(t) =
t
2
+
1
2
(
2
)
2
C
2
(t). (11.29)
11.2.2 Perturbation
We start with the reference ow, for which the interface remains at y = 0.
The velocity potentials are
1
= U
1
x and
2
= U
2
x. (11.30)
It can be veried that this satises all boundary conditions as long as
C
1
1
2
U
2
1
= C
2
1
2
U
2
2
. (11.31)
In stability problems, this part is relatively simple - although in the case
of a boundary layer, the Blasius solution represents a not-so-simple reference
ow. The next step is to modify this solution in small enough amounts so
the corrections can be linearized. This is particularly simple in potential
ows because the nonlinearities are limited to the pressure uctuations in
Bernoullis equation. So, let us dene the perturbations as
1
= U
1
x +
1
(x, y, t) and
2
= U
2
x +
2
(x, y, t). (11.32)
Substitution in the Laplace equation, and subtraction of the baseline equa-
tion, gives
1
= 0 and
2
2
= 0 (11.33)
with the BCs
lim
y
1
= 0 and lim
y+
2
= 0. (11.34)
The interface conditions are
1
=
t
y
+ U
1
x
y
(11.35)
and
2
=
t
y
+ U
2
x
y
. (11.36)
Furthermore, Bernoullis equation at y = 0 yields
1
+ U
1
1
=
t
2
+ U
2
2
. (11.37)
Note the linearization of the kinetic energy term, as in d
U
2
2
= U dU.
This completes the setting of the linearized problems for the perturba-
tions.
11.2.3 Solution
The system of linear partial dierential equations of second order (Laplacian)
looks formidable enough. But it helps to know that it admits exponential
solutions. In general, the idea is to try suitable forms of the solution, and see
what conditions the equations and BCs impose on them. Here, the successful
guess is
y
= y e
i(xct)
1
=

1
e
i(xct)
2
=

2
e
i(xct)
. (11.38)
With practice, you become procient with these, guided by the physical
interpretation of traveling waves along the interface that is the exponential
i(xct) part. Beside the traveling waves, the magnitude of the perturbations
is governed by the . variables, which could be y-dependent and could be
complex-valued. A more formal version of this explanation involves Fourier
transform w.r. to x and t, which yields linear odes in y with algebraic
coecients for the x- and t-derivatives. For such solutions,
x
= i and
t
= ic.
Then, mass balance (Laplace) takes the form
(
2
yy
2
)

1
e
i(xct)
= 0. (11.39)
Hence
1
= A
1
e
+y
+ B
1
e
y
, (11.40)
where B
1
= 0 to satisfy the condition for y . Similarly
2
= A
2
e
+y
+ B
2
e
y
. (11.41)
and A
2
= 0. Substituting into the interface condition gives
A
1
= ic y + U
1
i y = i y(U
1
c) (11.42)
and
B
2
= ic y U
2
i y = i y(U
2
c). (11.43)
So, our possible solutions are of the form
1
= i y(U
1
c) e
i(xct)
e
+y
(11.44)
and
2
= i y(U
2
c) e
i(xct)
e
y
. (11.45)
Finally, substitution into Bernoullis equation reduces to
(U
1
c)
2
= (U
2
c)
2
, (11.46)
which yields
c =
1
2
(U
1
+ U
2
) i
1
2
(U
2
U
1
). (11.47)
Therefore, the problem admits solutions of the form
=

e
i(x
1
2
(U
1
+U
2
)t)
e
1
2
(U
2
U
1
)t
, (11.48)
in which 3 parts can be identied. First,

remains arbitrary: the magni-
tude of the uctuation is indierent as long as it is small, as a side-eect
of linearization. Second, we see a wave traveling at the average speed of
the two streams. Finally, we have the telltale term in stability analysis: an
exponential factor that allows for indenite growth. This term corresponds
to an amplication of small perturbations, until they are no longer small and
nonlinearities alter the dynamics.
In conclusion, the inviscid shear layer is unstable for perturbations of
all wavelengths. This is known as the Kelvin-Helmholtz instability. Many
examples (including additional viscous eects, of course) can be observed
in Van Dykes book and in the movies shown in class. Two examples are
sketched on Fig. 11.7.
In real ows, the eects of viscosity make the use of potential ow impos-
sible, and the analysis is considerably more dicult; the growth is selective
by wavelength and requires that a Reynolds number exceed a threshold value,
below which the shear layer is stable.
The procedure is summarized on Fig. 11.8.
Figure 11.7: Examples of shear layer instabilities
11.2.4 Stability as equilibrium
In the derivations above, the emphasis was on the mathematics of the so-
lution: the pattern of baseline solution, perturbation, linearization, and ex-
ponential dependence, is common to many problems. The existence of a
critical parameter (e.g. Reynolds number for boundary layer stability) ex-
presses the balance between two or more forces (e.g. inertia and viscosity)
for given boundary conditions (e.g. presence of wall, Blasius baseline.) Vis-
cosity or gravity are common stabilizing forces, inertia or body forces can
be de-stabilizing. In the case of the Kelvin-Helmholtz instability, there is no
restoring force, and the inviscid shear layer is unconditionally unstable; for
the stabilizing eect of viscosity: see Panton, Sections 22.3 and 22.6, p.679-
692; for the stabilizing eect of gravity, watch the wind blow across a pond.
A summary of some classical stability problems is shown in Table 11.1.
Similarly, for a uid layer heated from below, one nds stability (no ex-
ponential growth) provided the Rayleigh number is smaller than a critical
value. The same procedure (baseline Blasius solution, small perturbation,
linearization of the equations, look for exponential solutions) has also worked
for boundary layers (Orr-Sommerfeld equation).
Figure 11.8: Kelvin-Helmholtz instability
Name Description Balance
Kelvin-Helmholtz inviscid shear layer inertia, no restoring force
Surface wind waves inviscid shear layer inertia, gravity
Rayleigh-Benard uid heated from below buoyancy, viscosity
Taylor concentric cylinders centrifugal pressure, viscosity
Gortler concave wall centrifugal pressure, viscosity
Marangoni temperature variations, surface tension, viscosity
curved free surface
Table 11.1: Some classic stability problems.
Problems
1. Describe analytically the standing waves such as seen in Van Dykes
book (e.g. p.110), including boundary conditions.
2. For standing gravity waves, describe the trajectory of a oating cork.
3. Return to the Rossby waves in the previous chapter; draw a map of
ideas, with emphasis on similarities and dierences with surface waves.
4. Kelvin-Helmholtz instability is based on potential ow analysis. In
view of the many solutions (the baseline 1-D ow, plus the entire fam-
ily of traveling waves of all wavelengths), resolve the apparent discrep-
ancy with the proven uniqueness of solutions for the Laplace equation
(
2
= 0) for given boundary conditions.
5. Set up the problem of Rayleigh-Benard instability for a uid heated
from below. Assume constant viscosity, xed temperatures at the lower
and upper walls. Derive the stability criterion
6. Set up the problem of Taylor instability between concentric rotating
cylinders. Derive the stability condition.
Chapter 12
Additional Reading
The following texts are, more or less directly, relevant to this course. Most
of them will be on library reserve for the semester. There is no attempt to
list titles for the many topics beyond the scope of the course, I would be sure
to overlook important items.
12.1 About this course
Undergrad Fluid Mechanics
Fox & McDonald, Introduction to Fluid Mechanics, J. Wiley Sons
B.R. Munson, D.F. Young and & T.H. Okiishi, Fundamentals of Fluid
Mechanics, J. Wiley & Sons 1994
F.M. White, Fluid Mechanics, McGraw Hill 2003
A.J. Smits, A Physical Introduction to Fluid Mechanics, J. Wiley &
Sons 2000
P.M. Gerhart, R.J. Gross and J.I. Hochstein, Fundamentals of Fluid
Mechanics, Addison-Wesley 1992
General Fluid Dynamics
G.K. Batchelor, Introduction to Fluid Dyamics, Cambridge U. Press
1967
233
234 CHAPTER 12. ADDITIONAL READING
R.N. Panton, Incompressible Flow, Wiley-Interscience 1984
I.G. Currie, Fundamental Mechanics of Fluids, McGraw-Hill 1974
D.J. Tritton, Physical Fluid Dynamics, Oxford U. Press 1990
D.J. Acheson, Elementary Fluid Dyanmics, Clarendon Press, Oxford
1990
H. Lamb, Hydrodynamics, Dover 1945 (rst published 1879)
A.J. Chorin and J.E. Marsden, A Mathematical Introduction to Fluid
Mechanics, Springer-Verlag 1992.
D. Joseph, T. Funada and J. Wang, Potential ows of viscous and
viscoelastic uids, Cambridge U.P., 2008.
For the historical aspects of the eld, a convenient reference is
O. Darrigol, Worlds of Flow, A history of hydrodynamics from the
Bernoullis to Prandtl, Oxford U. Press, 2005.
The Lagrangian description is gaining interest in several areas of application:
see
A. Bennett, Lagrangian Fluid Mechanics, Cambridge U. Press, 2006.
for a starting point.
About mind mapping
N. Margulies, Mapping Inner Space, Zephyr Press 2002
About Vorticity
P.G. Saman, Vortex Dynamics, Cambridge U.Press 1992
A.J. Majda and A.L. Bertozzi, Vorticity and Incompressible Flow,
Cambridge U. Press 2002
12.2. BEYOND 235
About scaling analysis
H. Tennekes and J.L. Lumley, A First Course in Turbulence, MIT Press
1972
A. Bejan, Convection Heat Transfer, Wiley-Interscience 1984
About waves and stability
S. Chandrasekhar, Hydrodynamic and Hydromagnetic Stability, Dover
1981
P.G. Drazin and W.H. Reid, Hydrodynamic Stability, Cambridge U.
Press 1981
J. Lighthill, Waves in Fluids, Cambridge U. Press 1978
G.B. Whitham, Linear and Nonlinear Waves, Wiley, 1974
Others
D.P. Telionis, Unsteady Viscous Flows, Springer-Verlag, 1981.
P. Tabeling, Introduction to Microuidics, Oxford U. Press, 2005.
G. Karniadakis, A. Bestok and N. Aluru, Microows and Nanoows,
Springer, 2005.
12.2 Beyond
There is plenty more: this is only a one-semester introduction to uid dynam-
ics, something to build upon. The student in this eld might be expected to
gain expertise in one or several additional topics, some of which were alluded
to in these notes, and some that present an entirely new set of challenges.
Far from exhaustive, the list below is only intended to keep the author and
his students humble: there is so much to know, even more than one knows
there is! Keep on exploring...
Experimental methods
Hot-wire and optical methods, hardware issues, access to facilities and databases,
data processing techniques...
Numerical methods
From the basic applied mathematics of numerical techniques, to the correct
use of CFD packages.
Turbulence
The biggest unsolved problem of classical physics: learn about mixing, mul-
tiscale dynamics, transport, and more.
Surface tension
Pressure dierences, surface energy: see how this applies to drop and bubble
formation, wetting of surfaces, lm stability, the noise of rain or a dripping
faucet...
Flow control
Small causes, large eects: true understanding of a ow can save a lot of
energy.
Acoustics
Linearization, sources of sound, etc.
Flow-structures interactions
Flow-induced vibrations, blu-body aerodynamics
Indoor and outdoor environmental ows
From droughty rooms to urban canyons: how they aect our well-being.
12.2. BEYOND 237
Heat and mass transfer
Convective heat exchange, mass transfer and chemical reactions.
Micro-uid dynamics
Very low Re, and below: molecular scales. Channel ows, pumping, mixing,
heat exchange and ow control dont look the same.
Compressible ows
Gas dynamics, hypersonics
Two-phase ows
Sedimendation, bubbles, phase changes, foams and colloids, uidization.
Meteorology, oceanography
Tornadoes and hurricanes, El Ni no and gyres, convection and inversions,
wind and moisture and temperature...
Rareed gas dynamics
The Knudsen number: when gases non longer behave as continua.
Wave motion
Surface waves, internal waves, solitons, and more.
Granular ows
From crushed stone to grain in elevators, from pills in a pharmaceutical
plant to a sand dune, these are solids, and yet they ow. A wide open area
of research!
Bio-uid mechanics
Birds and insects in ight, swimming sh and protozoa, blood ow and all
the complications of living systems.
Rheology
Polymers, paints, biouids, suspensions, etc., are non-Newtonian: those ows
can surprise you.
Superuids
Quantized vorticity and the strange world of He-III.
Chaotic ows
Over the entire range of Reynolds numbers from Lagrangian turbulence to
potential vortex dynamics, chaos! It can be used for mixing, it cant be
ignored in ow control or geophysical ows.
Applications
From hydraulic jacks to aerospace propulsion, from chemical reactors to hull
design, from metering devices to hydrology.
Lattice gas models
Instead of the Navier-Stokes equations, use discrete automata.
Index
Bernoullis equation 216, 225
along streamline 123
along vortex line 123
applications 128
energy 123
irrotational 122
unsteady 122
with losses 98
Biot-Savart relation 67
Blasius equation 177
Boussinesq approximation 102
Buckinghams theorem 108
Burgers vortex 76
Cauchys relation 87
Colebrooks formula 19
Coriolis force 203, 205
Couette ow 63, 64
Croccos equation 126
DAlembert paradox 30, 116, 143
Darcy friction factor 98
Ekman number 207
Euler equation 86, 116, 121, 128
normal equation 122
Eulerian description 42, 48
Fourier transform 226
Gauss divergence theorem 58
Greens function 68, 150, 153
Hagen diagram 109
Helmholtz 61, 223
decomposition 66
theorems 72, 132
see also Kelvin-Helmholtz
Hill vortex 76
Karman
integral method 190
pump 151
vortex street 30, 78
Kelvin 223
theorem 129
Kelvin-Helmholtz instability 223, 227
Kronecker symbol 40
Lagrange theorem 128
Lagrangian description 42, 43, 86, 129
Lagrangian turbulence 161
Lamb vector 74, 94, 104, 126
Laplace equation 135, 149, 217, 225
Laplacian, inverse 68
Mach number 102
Magnus force 94
Moody diagram 18, 109
Navier-Stokes equation 90
Newton, second law 85, 141
Newtonian uid 90
Oseen vortex 75
Pitot tube 128
Poisson equation 68, 161
pressure 93
Prandtl 167
Rankine
half-body 140
vortex 75
Rayleigh problems 151
Reynolds
lubrication 159, 161
number 21, 116, 149, 169
transport theorem 29
Rossby number203, 207
Stokes
drag on sphere 157
equation 116, 147
ow 147, 153
paradox 158
239
240 INDEX
problems 1519
viscous stress 90
streamfunction 81, 154
theorem 58, 72
Taylor vortex 78
Venturi 128
amplitude 218
boundary condition 92, 135, 216, 225
boundary layer 117, 167, 170
equation 174
thickness 175
vorticity 181
cavitation 128
circulation 58, 72
boundary layer 181
Kelvin theorem 129
see also Stokes theorem
conformal mapping 145
continuity 59
control parameter 107
control volume 28, 170
convection diusion 99
curvature radius 122
curved channel 191
cylinder
circulation 141
potential ow 141
Stokes paradox 158
density interface 222
diusion convection 99
dimensional analysis 105
dimensionless 105
dimensions 106
directional derivative 57
dispersion 218
dissipation of energy 96
drag 31
energy equation 96
enstrophy 98
entrainment 189
entrance ow 19
at plate 167
exion 68
force at surface 143, 157
free surface 216
helicity 74, 104
horseshoe vortex 80
hydrostatic 215
incompressible ow 100
induced velocity 67
inexion point 179
instability 223
integral method, Karman 190
internal wave 222
inviscid ow 86, 116, 129
irrotational ow 122, 128, 133
jet 183
lift 128
linearization 216
mass balance 29, 59
material
derivative 57
line 132
surface 216, 224
mind-map 11
minor loss 25
momentum balance 29
nonlocal eect 67, 68, 93, 135, 149, 150, 153,
156, 166
no-slip condition 92, 135
nozzle 128
parameter
control 107
scaling 107
pathline 45
permutation symbol 40
pipe ow 17
potential 216
ow 133, 136, 215
3D 145
cylinder 29
superposition 137
unsteady 214
vortex 74
see also vector potential
pressure 90, 92
normal equation 122
Reynolds lubrication 161
pressure scaling 114, 149
source 104
INDEX 241
and vorticity 199
principal axes 63, 131
rate of strain 61
reversibility
dynamic 151
kinematic 150
thermodynamic 151
ring vortex 76
rotation 61, 64
scaling analysis 112, 165, 171-173, 183
scaling parameter 107
secondary ow 191
separation 21, 191
shear layer 126, 223, 227
shear stress 171
similarity
dimensional 105
prole 175, 184, 188
sink, potential 138
siphon 128
source, potential 138
stagnation ow, potential 137
streakline 45
streamfunction 60, 69, 136
streamline 21, 48, 60
stress tensor 87
stretching 62
see also principal axes, vortex
superposition, potential ow 137
surface wave 215
transition 21
transition, boundary layer 182
traveling wave 217, 226
tsunami 218
turbulence 21
unsteady potential 216
vector potential 60, 135, 153
vena contracta 21
viscosity
see also Stokes ow, boundary layer
vortex
line 132
see also Helmholtz theorems, Bernoulli
potential 138
sheet 224
stretching 96, 130, 131
tube see vortex line
vorticity 58, 64, 67, 69, 72, 166
boundary layer 181
vorticity creation 128, 195, 199
equation 96
in Navier-Stokes 94
Lagrange theorem 128
streamwise 194
wake 159
wave
energy 222
dispersion 226
speed 213, 218

Incompressible Fluid Dynamics

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Incompressible Fluid Dynamics

Încărcat de

Drepturi de autor:

Formate disponibile

Lecture Notes in

Incompressible Fluid Dynamics:

is the pipes relative roughness (Fig. 1.2). We now turn to some of

in some direction denoted

is obtained from the vorticity distribution through the

(see Ch. 5). Helmholtz proved

f(r) for any function of r =| r |

(resp.: a component of zero Laplacian) which

/U. Applying Bernoullis equation between innity and the cylinder

= 0 as reference pressure (why is this OK?) we get the surface

. Flow disturbances associated with times larger

, and zero otherwise. It shows that the eect of a source at (x

should be constant (and we can take them

, and the velocity

= 0 (no slip). At innity, uniform

. Thus, Stokes ow around

L/, we saw in Ch. 5 that inviscid

and the areas are equal, the ux at 3 must be positive:

U)dy > 0 (8.2)

/L does not vanish separately

at the leading edge comes slowly to near-rest far

/L) and the viscous

and y/, these

, with the apostrophe denoting the derivative of the function

udy > 0 (8.66)

/a and that the O.D.E.

gH, with a rapid

(x, t) marked by the Lagrangian position y(x

(x, t). (11.24)

must be used, but the uid is sliding (no friction!) rel-

S-ar putea să vă placă și