with Vector Field Theory
bv
Dennis C. Prieve
Department of Chemical Engineering
Carnegie Mellon Universitv
Pittsburgh, PA 15213
An electronic version oI this book in Adobe PDE® Iormat was made available
to students oI 06703, Department oI Chemical Engineering,
Carnegie Mellon University, Eall, 2000.
Copyright © 2000 by Dennis C. Prieve
06703 1 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Table of Contents
ALGEBRA OF VECTORS AND TENSORS ........................................................................................................... 1
VECTOR MULTIPLICATION......................................................................................................................................... 1
Definition of Dvadic Product .............................................................................................................................. 2
DECOMPOSITION INTO SCALAR COMPONENTS........................................................................................................... 3
SCALAR EIELDS......................................................................................................................................................... 4
GRADIENT OF A SCALAR ........................................................................................................................................... 4
Geometric Meaning of the Gradient.................................................................................................................... 6
Applications of Gradient ..................................................................................................................................... 7
CURVILINEAR COORDINATES..................................................................................................................................... 7
Cvlindrical Coordinates ...................................................................................................................................... 7
Spherical Coordinates......................................................................................................................................... 8
DIFFERENTIATION OF VECTORS W.R.T. SCALARS .................................................................................................... 10
VECTOR EIELDS ...................................................................................................................................................... 12
Fluid Jelocitv as a Jector Field........................................................................................................................ 12
PARTIAL & MATERIAL DERIVATIVES ...................................................................................................................... 13
CALCULUS OF VECTOR FIELDS....................................................................................................................... 15
GRADIENT OF A SCALAR (EXPLICIT)........................................................................................................................ 15
DIVERGENCE, CURL, AND GRADIENT...................................................................................................................... 16
Phvsical Interpretation of Divergence .............................................................................................................. 17
Calculation of V
v in R.C.C.S. ......................................................................................................................... 17
Evaluation of V×v and Vv in R.C.C.S............................................................................................................... 19
Evaluation of V
v, V×v and Vv in Curvilinear Coordinates ........................................................................... 20
Phvsical Interpretation of Curl.......................................................................................................................... 21
VECTOR FIELD THEORY.................................................................................................................................... 24
DIVERGENCE THEOREM.......................................................................................................................................... 24
Corollaries of the Divergence Theorem............................................................................................................ 25
The Continuitv Equation ................................................................................................................................... 25
Revnolds Transport Theorem ............................................................................................................................ 27
STOKES THEOREM .................................................................................................................................................. 29
Jelocitv Circulation. Phvsical Meaning ........................................................................................................... 29
DERIVABLE FROM A SCALAR POTENTIAL ................................................................................................................ 31
THEOREM III........................................................................................................................................................... 32
TRANSPOSE OF A TENSOR, IDENTITY TENSOR......................................................................................................... 33
DIVERGENCE OF A TENSOR ..................................................................................................................................... 34
INTRODUCTION TO CONTINUUM MECHANICS'........................................................................................ 36
CONTINUUM HYPOTHESIS ....................................................................................................................................... 36
CLASSIFICATION OF EORCES.................................................................................................................................... 38
HYDROSTATIC EQUILIBRIUM.......................................................................................................................... 39
FLOW OF IDEAL FLUIDS..................................................................................................................................... 40
EULER'S EQUATION................................................................................................................................................. 40
KELVIN'S THEOREM................................................................................................................................................ 43
IRROTATIONAL ELOW OF AN INCOMPRESSIBLE ELUID ............................................................................................. 45
Potential Flow Around a Sphere ....................................................................................................................... 48
d´Alembert´s Paradox ........................................................................................................................................ 54
06703 2 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
STREAM FUNCTION ............................................................................................................................................. 56
TWOD ELOWS........................................................................................................................................................ 57
AXISYMMETRIC ELOW (CYLINDRICAL) ................................................................................................................... 58
AXISYMMETRIC ELOW (SPHERICAL)........................................................................................................................ 59
ORTHOGONALITY OF ¢÷CONST AND ÷CONST........................................................................................................ 60
STREAMLINES, PATHLINES AND STREAKLINES ........................................................................................................ 60
PHYSICAL MEANING OF STREAMFUNCTION............................................................................................................. 61
INCOMPRESSIBLE ELUIDS ........................................................................................................................................ 63
VISCOUS FLUIDS ................................................................................................................................................... 65
TENSORIAL NATURE OF SURFACE EORCES.............................................................................................................. 65
GENERALIZATION OF EULER'S EQUATION................................................................................................................ 69
MOMENTUM ELUX.................................................................................................................................................. 71
RESPONSE OF ELASTIC SOLIDS TO NORMAL (UNIAXIAL) STRESS ............................................................................ 73
RESPONSE OF ELASTIC SOLIDS TO SHEAR STRESS................................................................................................... 75
GENERALIZED HOOKE'S LAW.................................................................................................................................. 76
RESPONSE OF A VISCOUS ELUID TO PURE SHEAR.................................................................................................... 78
GENERALIZED NEWTON'S LAW OF VISCOSITY......................................................................................................... 79
NAVIERSTOKES EQUATION.................................................................................................................................... 80
BOUNDARY CONDITIONS......................................................................................................................................... 81
EXACT SOLUTIONS OF NS EQUATIONS........................................................................................................ 83
PROBLEMS WITH ZERO INERTIA.............................................................................................................................. 83
Flow in Long Straight Conduit of Uniform Cross Section ................................................................................ 84
Flow of Thin Film Down Inclined Plane........................................................................................................... 87
PROBLEMS WITH NONZERO INERTIA...................................................................................................................... 92
Rotating Disk*................................................................................................................................................... 92
CREEPING FLOW APPROXIMATION............................................................................................................... 94
CONEANDPLATE VISCOMETER ............................................................................................................................. 94
CREEPING ELOW AROUND A SPHERE (Re÷0)....................................................................................................... 100
Scaling............................................................................................................................................................. 100
Jelocitv Profile................................................................................................................................................ 102
Displacement of Distant Streamlines............................................................................................................... 105
Pressure Profile............................................................................................................................................... 106
CORRECTING FOR INERTIAL TERMS ...................................................................................................................... 109
PERTURBATION EXPANSION.................................................................................................................................. 110
ELOW AROUND CYLINDER AS RE÷0 .................................................................................................................... 113
BOUNDARYLAYER APPROXIMATION ........................................................................................................ 115
ELOW AROUND CYLINDER AS Re÷ ·................................................................................................................... 115
MATHEMATICAL NATURE OF BOUNDARY LAYERS................................................................................................ 116
MATCHEDASYMPTOTIC EXPANSIONS .................................................................................................................. 120
MAE`S APPLIED TO 2D ELOW AROUND CYLINDER ............................................................................................. 125
Outer Expansion.............................................................................................................................................. 126
Inner Expansion .............................................................................................................................................. 127
Boundarv Laver Thickness .............................................................................................................................. 130
PRANDTL`S B.L. EQUATIONS FOR 2D ELOWS ...................................................................................................... 132
ALTERNATE METHOD: PRANDTL`S SCALING THEORY .......................................................................................... 135
SOLUTION FOR A ELAT PLATE................................................................................................................................ 140
Time Out. Flow Next to Suddenlv Accelerated Plate ...................................................................................... 142
Time In. Boundarv Laver on Flat Plate .......................................................................................................... 144
BoundarvLaver Thickness.............................................................................................................................. 145
06703 3 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Drag on Plate .................................................................................................................................................. 148
SOLUTION FOR A SYMMETRIC CYLINDER .............................................................................................................. 149
BoundarvLaver Separation............................................................................................................................ 152
Drag Coefficient and Behavior in the hake of the Cvlinder........................................................................... 154
THE LUBRICATION APPROXIMATION......................................................................................................... 156
TRANSLATION OF A CYLINDER ALONG A PLATE.................................................................................................... 162
CAVITATION.......................................................................................................................................................... 165
SQUEEZING ELOW ................................................................................................................................................. 166
REYNOLDS EQUATION........................................................................................................................................... 170
TURBULENCE....................................................................................................................................................... 178
GENERAL NATURE OF TURBULENCE..................................................................................................................... 178
TURBULENT ELOW IN PIPES................................................................................................................................... 179
TIMESMOOTHING................................................................................................................................................. 181
TIMESMOOTHING OF CONTINUITY EQUATION...................................................................................................... 182
TIMESMOOTHING OF THE NAVIERSTOKES EQUATION......................................................................................... 182
ANALYSIS OF TURBULENT ELOW IN PIPES............................................................................................................. 184
PRANDTL`S MIXING LENGTH THEORY .................................................................................................................. 186
PRANDTL`S 'UNIVERSAL¨ VELOCITY PROFILE...................................................................................................... 189
PRANDTL`S UNIVERSAL LAW OF ERICTION............................................................................................................ 191
ELECTROHYDRODYNAMICS .......................................................................................................................... 194
ORIGIN OF CHARGE............................................................................................................................................... 194
GOUYCHAPMAN MODEL OF DOUBLE LAYER....................................................................................................... 195
ELECTROSTATIC BODY EORCES ............................................................................................................................ 198
ELECTROKINETIC PHENOMENA............................................................................................................................. 200
SMOLUCHOWSKI'S ANALYSIS (CA. 1918) .............................................................................................................. 201
ELECTROOSMOSIS IN CYLINDRICAL PORES.......................................................................................................... 204
ELECTROPHORESIS................................................................................................................................................ 205
STREAMING POTENTIAL ........................................................................................................................................ 205
SURFACE TENSION............................................................................................................................................. 208
MOLECULAR ORIGIN............................................................................................................................................. 208
BOUNDARY CONDITIONS FOR ELUID ELOW........................................................................................................... 210
INDEX...................................................................................................................................................................... 213
06703 1 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Algebra of Vectors and Tensors
Whereas heat and mass are scalars, Iluid mechanics concerns transport oI momentum, which
is a vector. Heat and mass Iluxes are vectors, momentum Ilux is a tensor. Consequently, the
mathematical description oI Iluid Ilow tends to be more abstract and subtle than Ior heat and
mass transIer. In an eIIort to make the student more comIortable with the mathematics, we will
start with a review oI the algebra oI vectors and an introduction to tensors and dyads. A brieI
review oI vector addition and multiplication can be Iound in Greenberg,
+
pages 132139.
Scalar  a quantity having magnitude but no direction (e.g. temperature, density)
Jector  (a.k.a. 1st rank tensor) a quantity having magnitude and direction (e.g. velocity,
Iorce, momentum)
(2nd rank) 1ensor  a quantity having magnitude and two directions (e.g. momentum
Ilux, stress)
VECTOR MULTIPLICATION
Given two arbitrary vectors a and b, there are three types oI vector
products are deIined:
Notation Result Definition
Dot Product a
b scalar ab cosu
Cross Product a×b vector absinun
where u is an interior angle (0 s u s t) and n is a unit vector which is normal to both a and b.
The sense oI n is determined Irom the "righthandrule"
+
Dyadic Product ab tensor
In the above deIinitions, we denote the magnitude (or length) oI vector a by the scalar a.
BoldIace will be used to denote vectors and italics will be used to denote scalars. Secondrank
tensors will be denoted with doubleunderlined boldIace; e.g. tensor T.
+
Greenberg, M.D., Foundations Of Applied Mathematics, PrenticeHall, 1978.
+
The 'righthand rule¨: with the Iingers oI the right hand initially pointing in the direction oI
the Iirst vector, rotate the Iingers to point in the direction oI the second vector; the thumb then
points in the direction with the correct sense. OI course, the thumb should have been normal to
the plane containing both vectors during the rotation. In the Iigure above showing a and b, a×b
is a vector pointing into the page, while b×a points out of the page.
06703 2 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Definition of Dyadic Product
ReIerence: Appendix B Irom Happel & Brenner.
v
The word 'dyad¨ comes Irom Greek:
'dy¨ means two while 'ad¨ means adjacent. Thus the name dyad reIers to the way in which this
product is denoted: the two vectors are written adjacent to one another with no space or other
operator in between.
There is no geometrical picture that I can draw which will explain what a dyadic product is.
It's best to think oI the dyadic product as a purely mathematical abstraction having some very
useIul properties:
Dyadic Product ab  that mathematical entity which satisIies the Iollowing properties
(where a, b, v, and w are any Iour vectors):
1. ab
v ÷ a(b
v) which has the direction oI a; note that ba
v ÷ b(a
v) which has the
direction oI b. Thus ab = ba since they don`t produce the same result on postdotting with
v.
2. v
ab ÷ (v
a)b thus v
ab = ab
v
3. ab×v ÷ a(b×v) which is another dyad
4. v×ab ÷ (v×a)b
5. abvw ÷ (a
w)(b
v) which is sometimes known as the innerouter product or the double
dot product.
*
6. a(v¹w) ÷ av¹aw (distributive Ior addition)
7. (v¹w)a ÷ va¹wa
8. (s¹t)ab ÷ sab¹tab (distributive Ior scalar multiplicationalso distributive Ior dot and cross
product)
9. sab ÷ (sa)b ÷ a(sb)
v
Happel, J., & H. Brenner, Low Revnolds Number Hvdrodvnamics, NoordhoII, 1973.
*
Brenner deIines this as (a
v)(b
w). Although the two deIinitions are not equivalent, either can
be used  as long as you are consistent. In these notes, we will adopt the deIinition above and
ignore Brenner's deIinition.
06703 3 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
DECOMPOSITION INTO SCALAR COMPONENTS
Three vectors (say e
1
, e
2
, and e
3
) are said to be linearly independent iI none can be
expressed as a linear combination oI the other two (e.g. i, j, and k). Given such a set oI three LI
vectors, any vector (belonging to E
3
) can be expressed as a linear combination oI this basis:
v ÷ v
1
e
1
¹ v
2
e
2
¹ v
3
e
3
where the v
i
are called the scalar components oI v. Usually, Ior convenience, we choose
orthonormal vectors as the basis:
e
i
e
f
÷ o
if
÷
1
0
iI
iI
i f
i f
=
=
although this is not necessary. o
if
is called the Kronecker delta. Just as the Iamiliar dot and
cross products can written in terms oI the scalar components, so can the dyadic product:
vw ÷ (v
1
e
1
¹v
2
e
2
¹v
3
e
3
)(w
1
e
1
¹w
2
e
2
¹w
3
e
3
)
÷ (v
1
e
1
)(w
1
e
1
)¹(v
1
e
1
)(w
2
e
2
)¹ ...
÷ v
1
w
1
e
1
e
1
¹v
1
w
2
e
1
e
2
¹
... (nine terms)
where the e
i
e
f
are nine distinct unit dyads. We have applied the deIinition oI dyadic product to
perIorm these two steps: in particular items 6, 7 and 9 in the list above.
More generally any nth rank tensor (in E
3
) can be expressed as a linear combination oI the 3
n
unit nads. Eor example, iI n÷2, 3
n
÷9 and an nad is a dyad. Thus a general secondrank tensor
can be decomposed as a linear combination oI the 9 unit dyads:
T ÷ T
11
e
1
e
1
¹T
12
e
1
e
2
¹
... ÷ E
i÷1,3
E
f÷1,3
T
if
e
i
e
f
Although a dyad (e.g. vw) is an example oI a secondrank tensor, not
all 2nd rank tensors T can be expressed as a dyadic product oI two
vectors. To see why, note that a general secondrank tensor has nine
scalar components which need not be related to one another in any
way. By contrast, the 9 scalar components oI dyadic product above
involve only six distinct scalars (the 3 components oI v plus the 3
components oI w).
AIter a while you get tired oI writing the summation signs and limits. So an abbreviation was
adopted whereby repeated appearance oI an index implies summation over the three allowable
values oI that index:
T ÷ T
if
e
i
e
f
This is sometimes called the Cartesian (implied) summation convention.
06703 4 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
SCALAR FIELDS
Suppose I have some scalar Iunction oI position (x,v,:) which is continuously differentiable,
that is
f ÷ f(x,v,:)
and cf/cx, cf/cv, and cf/c: exist and are continuous throughout some 3D region in space. This
Iunction is called a scalar field. Now consider f at a second point which is diIIerentially close to
the Iirst. The diIIerence in f between these two
points is called the total differential oI f:
f(x¹dx,v¹dv,:¹d:)  f(x,v,:) ÷ df
Eor any continuous Iunction f(x,v,:), df is linearly
related to the position displacements, dx, dv and d:.
That linear relation is given by the Chain Rule oI
diIIerentiation:
df
f
x
dx
f
v
dv
f
:
d: = + +
c
c
c
c
c
c
Instead oI deIining position using a particular
coordinate system, we could also deIine position using a position vector r:
r i j k = + + x v :
The scalar Iield can be expressed as a Iunction oI a vector argument, representing position,
instead oI a set oI three scalars:
f ÷ f(r)
Consider an arbitrary displacement away Irom the point r, which we denote as dr to emphasize
that the magnitude dr oI this displacement is suIIiciently small that f(r) can be linearized as a
Iunction oI position around r. Then the total diIIerential can
be written as
df f d f = + ÷ ( ) ( ) r r r
GRADIENT OF A SCALAR
We are now is a position to deIine an important vector
associated with this scalar Iield. The gradient (denoted as Vf)
is deIined such that the dot product oI it and a diIIerential
displacement vector gives the total diIIerential:
06703 5 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
df d f ÷ V r
EXAMPLE: Obtain an explicit Iormula Ior calculating the gradient in Cartesian* coordinates.
Solution: r ÷ xi ¹ vj ¹ :k
r¹dr ÷ (x¹dx)i ¹ (v¹dv)j ¹ (:¹d:)k
subtracting: dr ÷ (dx)i ¹ (dv)j ¹ (d:)k
Vf ÷ (Vf)
x
i ¹ (Vf)
v
j ¹ (Vf)
:
k
dr
Vf ÷ (dx)i ¹ ...
(Vf)
x
i ¹ ...
df ÷ (Vf)
x
dx ¹ (Vf)
v
dv ¹ (Vf)
:
d: (1)
Using the Chain rule: df ÷ (cf/cx)dx ¹ (cf/cv)dv ¹ (cf/c:)d: (2)
According to the deIinition oI the gradient, (1) and (2) are identical. Equating them and
collecting terms:
(Vf)
x
(cf/cx)dx ¹ (Vf)
v
(cf/cv)dv ¹ (Vf)
:
(cf/c:)d: ÷ 0
Think oI dx, dv, and d: as three independent variables which can assume an inIinite number oI
values, even though they must remain small. The equality above must hold Ior all values oI dx,
dv, and d:. The only way this can be true is iI each individual term separately vanishes:**
So (Vf)
x
÷ cf/cx, (Vf)
v
÷ cf/cv, and (Vf)
:
÷ cf/c:,
leaving V = + + f
f
x
f
v
f
:
c
c
c
c
c
c
i j k
Other ways to denote the gradient include:
Vf ÷ gradf ÷ cf/cr
*
Named aIter Erench philosopher and mathematician Rene Descartes (15961650), pronounced
"daycart", who Iirst suggested plotting f(x) on rectangular coordinates
**
Eor any particular choice oI dx, dv, and d:, we might obtain zero by cancellation oI positive
and negative terms. However a small change in one oI the three without changing the other two
would cause the sum to be nonzero. To ensure a zerosum Ior all choices, we must make each
term vanish independently.
06703 6 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Geometric Meaning of the Gradient
1) direction: Vf(r) is normal to the f÷const surIace passing through the point r in the direction
oI increasing f. Vf also points in the direction oI steepest ascent oI f.
2) magnitude: ,Vf, is the rate oI change oI f with
distance along this direction
What do we mean by an "f÷const surIace"? Consider
an example.
Example: Suppose the steady state temperature
proIile in some heat conduction problem is given by:
T(x,v,:) ÷ x
2
¹ v
2
¹ :
2
Perhaps we are interested in VT at the point (3,3,3)
where T÷27. VT is normal to the T÷const surIace:
x
2
¹ v
2
¹ :
2
÷ 27
which is a sphere oI radius 27 .
+
Proof of 1). Let's use the deIinition to show that these geometric meanings are correct.
df ÷ dr
Vf
Consider an arbitrary f. A portion oI the f÷const surIace
containing the point r is shown in the Iigure at right.
Choose a dr which lies entirely on f÷const. In other
words, the surIace contains both r and r¹dr, so
f(r) ÷ f(r¹dr)
and df ÷ f(r¹dr)f(r) ÷ 0
Substituting this into the deIinition oI gradient:
df ÷ 0 ÷ dr
Vf ÷ (dr((Vf(cosu
Since (dr( and (Vf( are in general not zero, we are
+
A vertical bar in the leIt margin denotes material which (in the interest oI time) will be omitted
Irom the lecture.
06703 7 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Iorced to the conclusion that cosu÷0 or u÷90°. This means that Vf is normal to dr which lies in
the surIace.
2) can be proved in a similar manner: choose dr to be parallel to Vf. Does Vf point toward
higher or lower values oI f?
Applications of Gradient
Iind a vector pointing in the direction oI steepest ascent oI some scalar Iield
determine a normal to some surIace (needed to apply b.c.`s like n
v ÷ 0 Ior a boundary which
is impermeable)
determine the rate oI change along some arbitrary direction: iI n is a unit vector pointing
along some path, then
n
V = f
f
s
c
c
is the rate oI change oI f with distance (s) along this path given by n. c c f s is called the
directed derivative oI f.
CURVILINEAR COORDINATES
In principle, all problems in Iluid mechanics and transport could be solved using Cartesian
coordinates. OIten, however, we can take advantage oI symmetry in a problem by using another
coordinate system. This advantage takes the Iorm oI a reduction in the number oI independent
variables (e.g. PDE becomes ODE). A Iamiliar
example oI a nonCartesian coordinate system is:
Cylindrical Coordinates
r ÷ (x
2
¹v
2
)
1/2
x ÷ rcosu
u ÷ tan
1
(v/x) v ÷ rsinu
: ÷ : : ÷ :
Vectors are decomposed diIIerently. Instead oI
in R.C.C.S.: v ÷ v
x
i ¹ v
v
j ¹ v
:
k
in cylindrical coordinates, we write
in cyl. coords.: v ÷ v
r
e
r
¹
v
u
e
u
¹ v
:
e
:
06703 8 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
where e
r
, e
u
, and e
:
are new unit vectors pointing the r, u and : directions. We also have a
diIIerent set oI nine unit dyads Ior decomposing tensors:
e
r
e
r
, e
r
e
u
, e
r
e
:
, e
u
e
r
, etc.
Like the Cartesian unit vectors, the unit vectors in cylindrical coordinates Iorm an orthonormal
set oI basis vectors Ior E
3
. Unlike Cartesian unit vectors, the orientation oI e
r
and e
u
depend on
position. In other words:
e
r
÷ e
r
(u)
e
u
÷ e
u
(u)
Spherical Coordinates
Spherical coordinates (r,u,) are deIined relative to Cartesian coordinates as suggested in the
Iigures above (two views oI the same thing). The green surIace is the xyplane, the red surIace is
the xzplane, while the blue surIace (at least in the leIt image) is the yzplane. These three planes
intersect at the origin (0,0,0), which lies deeper into the page than (1,1,0). The straight red line,
drawn Irom the origin to the point (r,u,)
+
has length r, The angle u is the angle the red line
makes with the :axis (the red circular arc labelled u has radius r and is subtended by the angle
u). The angle  (measured in the xyplane) is the angle the second blue plane (actually it`s one
quadrant oI a disk) makes with the xyplane (red). This plane which is a quadrant oI a disk is a
+
This particular Iigure was drawn using r ÷ 1, u ÷ t/4 and  ÷ t/3.
06703 9 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
÷const surIace: all points on this plane have the same  coordinate. The second red (circular)
arc labelled  is also subtended by the
angle .
A number oI other ÷const planes are
shown in the Iigure at right, along with a
sphere oI radius r÷1. All these planes
intersect along the :axis, which also
passes through the center oI the sphere.
2 2 2
1 2 2
1
sin cos
sin sin tan
cos tan
x r r x v :
v r x v :
: r v x
÷
÷
= u  = + + +
 
= u  u = +

\ .
= u  =
The position vector in spherical
coordinates is given by
r ÷ xi¹vj¹:k ÷ r e
r
(u,)
In this case all three unit vectors depend on position:
e
r
÷ e
r
(u,), e
÷ e
(u,), and e
÷ e
()
where e
r
is the unit vector pointing the direction oI increasing r, holding u and  Iixed; e
is the
unit vector pointing the direction oI increasing u, holding r and  Iixed; and e
is the unit vector
pointing the direction oI increasing , holding r and u Iixed.
These unit vectors are shown in the Iigure at right.
Notice that the surIace ÷const is a plane containing
the point r itselI, the projection oI the point onto the
xyplane and the origin. The unit vectors e
r
and e
lie in this plane as well as the Cartesian unit vector k
(sometimes denoted e
:
).
II we tilt this ÷const
plane into the plane oI
the page (as in the sketch at leIt), we can more easily see the
relationship between these three unit vectors:
cos sin
= u ÷ u e e e
This is obtained by determined Irom the geometry oI the right
triangle in the Iigure at leIt. When any oI the unit vectors is position
unit circle on
÷ const
surIace
06703 10 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
dependent, we say the coordinates are:
curvilinear  at least one oI the basis vectors is position dependent
This will have some proIound consequences which we will get to shortly. But Iirst, we need to
take 'timeout¨ to deIine:
DIFFERENTIATION OF VECTORS W.R.T. SCALARS
Suppose we have a vector v which depends on the scalar parameter t:
v ÷ v(t)
Eor example, the velocity oI a satellite depends on time. What do we mean by the 'derivative¨ oI
a vector with respect to a scalar. As in the Eundamental Theorem oI Calculus, we deIine the
derivative as:
d
dt
t t t
t t
v v v
÷
( ¹ )  ( )
lim
A
A
A ÷
0
Note that dv/dt is also a vector.
EXAMPLE: Compute de
r
/du in cylindrical coordinates.
Solution: Erom the deIinition oI the derivative:
d
d
r r r r
e e e e
u
u u
=
+ ÷
=
÷ ÷
lim
( ) ( )
lim
Au Au
Au
Au
A
Au 0 0
Since the location oI the tail oI a vector is not
part oI the deIinition oI a vector, let's move
both vectors to the origin (keeping the
orientation Iixed). Using the parallelogram
law, we obtain the diIIerence vector. Its
magnitude is:
e e
r r
( ) ( ) sin u u + ÷ = Au
Au
2
2
Its direction is parallel to e
u
(u¹Au/2), so:
e e e
r r
( ) ( ) sin u u u
u
+ ÷ = + Au
Au Au
2
2 2
,
06703 11 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Recalling that sinx tends to x as x÷0, we have
lim ( ) ( )
Au
Au Au
÷
+ ÷ =
0
e e e
r r
u u u
u
, , ,
Dividing this by Au, we obtain the derivative:
de
r
/du ÷ e
u
Similarly, de
u
/du ÷ e
r
One important application oI 'diIIerentiation with respect
to a scalar¨ is the calculation oI velocity, given position as a
Iunction oI time. In general, iI the position vector is known,
then the velocity can be calculated as the rate oI change in
position:
r ÷ r(t)
v ÷ dr/dt
Similarly, the acceleration vector a can be calculated as the
derivative oI the velocity vector v:
a ÷ dv/dt
EXAMPLE: Given the trajectory oI an object in
cylindrical coordinates
r ÷ r(t), u ÷ u(t), and : ÷ :(t)
Eind the velocity oI the object.
Solution: Eirst, we need to express r in in terms oI the
unit vectors in cylindrical coordinates. Using the
Iigure at right, we note by inspection that
*
r(r,u,:) ÷ re
r
(u) ¹ :e
:
Now we can apply the Chain Rule:
*
Recalling that r ÷ xi ¹ vj ¹ :k in Cartesian coordinates, you might be tempted to write r ÷ re
r
¹
ue
¹ :e
:
in cylindrical coordinates. OI course, this temptation gives the wrong result (in
particular, the units oI length in the second term are missing).
06703 12 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
d
r
dr d
:
d: dr r d d:
: r :
r
d
d
r
r :
r r :
r
r r r
e e e
e e e
=
+
+
= + +
c
c
c
cu
u
c
c
u
u
u
u
u
u
, , ,
Dividing by dt, we obtain the velocity:
v
r
e e e = = + +
d
dt
dr t
dt
r
d t
dt
d: t
dt
v
r
v v
:
r :
u
u
VECTOR FIELDS
A vector Iield is deIined just like a scalar Iield, except that it's a vector. Namely, a vector Iield is
a positiondependent vector:
v ÷ v(r)
Common examples oI vector Iields include Iorce Iields, like the gravitational Iorce or an
electrostatic Iorce Iield. OI course, in this course, the vector Iield oI greatest interest is:
Fluid Jelocity as a Jector Field
Consider steady Ilow around a submerged object. What do we mean by 'Iluid velocity?¨
There are two ways to measure Iluid velocity. Eirst, we could add tracer particles to the Ilow and
measure the position oI the tracer particles as a Iunction oI time; diIIerentiating position with
respect to time, we would obtain the velocity.
+
A mathematical 'tracer particle¨ is called a
'material point:¨
Material point  Iluid element  a given set oI Iluid molecules whose location may change
with time.
+
+
Actually, this only works Ior steady Ilows. In unsteady Ilows, pathlines, streaklines and
streamlines diIIer (see 'Streamlines, Pathlines and Streaklines¨ on page 65).
+
In a molecularlevel description oI gases or liquids, even nearby molecules have widely
diIIerent velocities which Iluctuate with time as the molecules undergo collisions. We will
reconcile the molecularlevel description with the more common continuum description in
Chapter 4. Eor now, we just state that by 'location oI a material point¨ we mean the location oI
the center oI mass oI the molecules. The 'point¨ needs to contain a statistically large number oI
molecules so that r(t) converges to a smooth continuous Iunction.
06703 13 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Suppose the trajectory oI a material point is given by:
r ÷ r(t)
Then the Iluid velocity at any time is v
r
=
d
dt
(3)
A second way to measure Iluid velocity is similar to the 'bucketandstopwatch method.¨ We
measure the volume oI Iluid crossing a surIace per unit
time:
n v
=
÷
lim
A
A
A a
q
a 0
where Aa is the area oI a surIace element having a unit
normal n and Aq is the volumetric Ilowrate oI Iluid
crossing Aa in the direction oI n.
When Aa is small enough so that this quotient has
converged in a mathematical sense and Aa is small enough so that the surIace is locally planar so
we can denote its orientation by a unit normal n, we can replace Aa by da and Aq by dq and
rewrite this deIinition as:
dq ÷ n
v da (4)
This is particularly convenient to compute the
volumetric Ilowrate across an arbitrary curved
surIace, given the velocity proIile. We just
have to sum up the contribution Irom each
surIace element:
Q da
A
=
n v
PARTIAL & MATERIAL DERIVATIVES
Let f ÷ f(r,t)
represent some unsteady scalar Iield (e.g. the unsteady temperature proIile inside a moving Iluid).
There are two types oI time derivatives oI unsteady scalar Iields which we will Iind convenient to
deIine. In the example in which f represents temperature, these two time derivatives correspond
to the rate oI change (denoted generically as df /dt) measured with a thermometer which either is
held stationary in the moving Iluid or driIts along with the local Iluid.
06703 14 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
partial derivative  rate oI change at a Iixed spatial point:
d
f df
t dt
=
c
 
=

c
\ .
r 0
where the subscript dr÷0 denotes that we are evaluating the derivative along a
path
*
on which the spatial point r is held Iixed. In other words, there is no
displacement in position during the time interval dt. As time proceeds, diIIerent
material points occupy the spatial point r.
material derivative (a.k.a. substantial derivative)  rate oI change within a particular
material point (whose spatial coordinates vary with time):
d dt
Df df
Dt dt
=
 
=

\ .
r v
where the subscript dr ÷ v dt denotes that a displacement in position (correspon
ding to the motion oI the velocity) occurs: here v denotes the local Iluid velocity.
As time proceeds, the moving material occupies diIIerent spatial points, so r is not
Iixed. In other words, we are Iollowing along with the Iluid as we measure the
rate oI change oI f.
A relation between these two derivatives can be derived using a generalized vectorial Iorm oI the
Chain Rule. Eirst recall that Ior steady (independent oI t) scalar Iields, the Chain Rule gives the
total diIIerential (in invariant Iorm) as
df f d f d f ÷ + ÷ = V r r r r
When t is a variable, we just add another contribution to the total diIIerential which arises Irom
changes in t, namely dt. The Chain Rule becomes
df f d t dt f t
f
t
dt d f = + + ÷ = + V r r r r , ,
c
c
The Iirst term has the usual ChainRule Iorm Ior changes due to a scalar variable; the second
term gives changes due to a displacement in vectorial position r. Dividing by dt holding R Iixed
yields the material derivative:
*
By 'path¨ I mean a constraint among the independent variables, which in this case are time and
position (e.g. x,v,: and t). Eor example, I might vary one oI the independent variables (e.g. x)
while holding the others Iixed. Alternatively, I might vary one oI the independent variables (e.g.
t) while prescribing some related changes in the others (e.g. x(t), v(t) and :(t)). In the latter case, I
am prescribing (in parametric Iorm) a trajectory through space, hence the name 'path.¨
06703 15 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
1
d dt d dt d dt
Df df f dt d
f
Dt dt t dt dt
= = =
c
     
÷ = + V
  
c
\ . \ . \ .
r v r v r v
v
r
But (dr/dt) is just v, leaving:
Df
Dt
f
t
f = + V
c
c
v
This relationship holds Ior a tensor oI any rank. Eor example, the material derivative oI the
velocity vector is the acceleration a oI the Iluid, and it can be calculated Irom the velocity proIile
according to
a
v v
v v = = + V
D
Dt t
c
c
We will deIine Vv in the next section.
Calculus of Vector Fields
Just like there were three kinds oI vector multiplication which can be deIined, there are three
kinds oI diIIerentiation with respect to position.
Shortly, we will provide explicit deIinitions oI
these quantities in terms oI surIace integrals. Let
me introduce this type oI deIinition using a more
Iamiliar quantity:
GRADIENT OF A SCALAR (EXPLICIT)
Recall the previous deIinition Ior gradient:
f ÷ f (r): df ÷ dr
Vf (implicit deI`n oI Vf )
Such an implicit deIinition is like deIining f ' (x) as that Iunction associated with f(x) which
yields:
f ÷ f (x): df ÷ (dx) f ' (implicit deI`n oI f ´ )
An equivalent, but explicit, deIinition oI derivative is provided by the Eundamental Theorem oI
the Calculus:
f x
x
f x x f x
x
df
dx
' ÷
÷
+ ÷
= ( )
lim
( ) ( )
A
A
A 0
(explicit deI`n oI f ´ )
Notation Result
Divergence V
v scalar
Curl V×v vector
Gradient Vv tensor
06703 16 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
We can provide an analogous deIinition oI Vf
V ÷
÷
f
J J
fda
A
lim
0
1
n (explicit deI`n oI Vf )
where f ÷ any scalar Iield
A ÷ a set oI points which constitutes any
closed surIace enclosing the point
r at which Vf is to be evaluated
J ÷ volume oI region enclosed by A
da ÷ area oI a diIIerential element
(subset) oI A
n ÷ unit normal to da, pointing out oI
region enclosed by A
lim (J÷0) ÷ limit as all dimensions oI A shrink to zero (in other words, A collapses about
the point at which Vf is to be deIined.)
What is meant by this surIace integral? Imagine A to be the skin oI a potato. To compute the
integral:
1) Carve the skin into a number oI elements. Each element must be suIIiciently small so that
 element can be considered planar (i.e. n is practically constant over the
element)
 f is practically constant over the element
2) Eor each element oI skin, compute nf da
3) Sum yields integral
This same type oI deIinition can be used Ior each oI the three spatial derivatives oI a vector Iield:
DIVERGENCE, CURL, AND GRADIENT
Divergence V ÷
÷
v n v
lim
J J
da
A
0
1
06703 17 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Curl V × ÷
÷
×
v n v
lim
J J
da
A
0
1
Gradient V ÷
÷
v nv
lim
J J
da
A
0
1
Physical Interpretation of Divergence
Let the vector Iield v ÷ v(r) represent the steadystate velocity proIile in some 3D region oI
space. What is the physical meaning oI V
v?
 n
v da ÷ dq ÷ volumetric Ilowrate out through da (cm
3
/s). This quantity is positive
Ior outIlow and negative Ior inIlow.
 }
A
n
v da ÷ net volumetric Ilowrate out oI enclosed volume (cm
3
/s). This is also
positive Ior a net outIlow and negative Ior a net inIlow.
 (1/J) }
A
n
v da ÷ Ilowrate out per unit volume (s
1
)
 V
v ÷
> .
=
<
0
0
0
Ior an expanding gas (perhaps or
Ior an incompressible Iluid (no room Ior accumulation)
Ior a gas being compressed
T p )
 V
v ÷ volumetric rate oI expansion oI a diIIerential element oI Iluid per unit volume
oI that element (s
1
)
Calculation of V
v in R.C.C.S.
Given: v ÷ v
x
(x,v,:)i ¹ v
v
(x,v,:)j ¹ v
:
(x,v,:)k
06703 18 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Evaluate V
v at (x
o
,v
o
,:
o
).
Solution: Choose A to be surIace oI rectangular
parallelopiped oI dimensions Ax,Av,A: with one
corner at x
o
,v
o
,:
o
.
So we partition A into the six Iaces oI the
parallelopiped. The integral will be computed
separately over each Iace:
n v n v n v n v
da da da da
A A A A
= + + +
1 2 6
SurIace A
1
is the x÷x
o
Iace: n ÷ i
n
v ÷ i
v ÷ v
x
(x
o
,v,:)
A :
: :
v
v v
x o
da v x v : dv d:
o
o
o
o
1
= ÷
+ +
n v
A A
( , , )
Using the Mean Value Theorem: ÷ v
x
(x
o
,v',:')AvA:
where v
o
s v' s v
o
¹Av
and :
o
s :' s :
o
¹A:
SurIace A
2
is the x÷x
o
¹Ax Iace: n ÷ ¹i
n
v ÷ i
v ÷ v
x
(x
o
¹Ax,v,:)
A :
: :
v
v v
x o
da v x x v : dv d:
o
o
o
o
1
= +
+ +
n v
A A
A ( , , )
Using the Mean Value Theorem: ÷ v
x
(x
o
¹Ax,vª,:ª)AvA:
where v
o
s vª s v
o
¹Av
and :
o
s :ª s :
o
¹A:
The sum oI these two integrals is:
A A
x o x o
v x x v : v x v : v :
1 2
+ = + '' '' ÷ ' ' ( , , ) ( , , ) A A A
06703 19 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Dividing by J ÷ AxAvA::
1
1 2
J
da
v x x v : v x v :
x
A A
x o x o
+
=
+ '' '' ÷ ' '
n v
( , , ) ( , , ) A
A
Letting Av and A: tend to zero:
lim
,
( , , ) ( , , )
A A
A
A v : J
da
v x x v : v x v :
x
A A
x o o o x o o o
÷
=
+ ÷
+
0
1
1 2
n v
Einally, we take the limit as Ax tends to zero:
lim
, ,
J J
da
v
x
A A
x
x v :
o o o
÷
=
c
c
+
0
1
1 2
n v
Similarly, Irom the two v÷const surIaces, we obtain:
lim
, ,
J J
da
v
v
A A
v
x v :
o o o
÷
=
c
c
+
0
1
3 4
n v
and Irom the two :÷const surIaces:
lim
, ,
J J
da
v
:
A A
:
x v :
o o o
÷
=
c
c
+
0
1
5 6
n v
Summing these three contributions yields the divergence:
V = + +
v
c
c
c
c
c
c
v
x
v
v
v
:
x
v
:
Evaluation of V×v and Vv in R.C.C.S.
In the same way, we could use the deIinition to determine expressions Ior the curl and the
gradient.
V × =
c
c
÷
c
c
+
c
c
÷
c
c
+
c
c
÷
c
c
v i j k
v
v
v
:
v
:
v
x
v
x
v
v
:
v
x :
v
x
06703 20 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
The Iormula Ior curl in R.C.C.S. turns out to be expressible as a determinant oI a matrix:
i j k
i j k
c
c
c
c
c
c
=
c
c
÷
c
c
+
c
c
÷
c
c
+
c
c
÷
c
c
x v :
v v v
v
v
v
:
v
:
v
x
v
x
v
v
x v :
:
v
x :
v
x
But remember that the determinant is just a mnemonic device, not the definition oI curl. The
gradient oI the vector v is
V
c
c
v e e ÷
÷ ÷ i f
f
i
i f
v
x
1
3
1
3
where x
1
÷ x, x
2
÷ v, and x
3
÷ :, v
1
÷ v
x
, etc.
Evaluation of V VV V
v, V×v and Vv in Curvilinear Coordinates
ReI: Greenberg, p175
These surIaceintegral deIinitions can be applied to any coordinate system. On HWK #2, we
obtain V
v in cylindrical coordinates.
More generally, we can express divergence, curl and gradient in terms oI the metric
coefficients Ior the coordinate systems. II u,v,w are the three scalar coordinate variables Ior the
curvilinear coordinate system, and
x ÷ x(u,v,w) v ÷ v(u,v,w) : ÷ :(u,v,w)
can be determined, then the three metric coeIIicients h
1
, h
2
and h
3
are given by
h u v w x v :
u u u 1
2 2 2
, , = + +
h u v w x v :
v v v 2
2 2 2
, , = + +
h u v w x v :
w w w 3
2 2 2
, , = + +
where letter subscripts denote partial diIIerentials while numerical subscripts denote component,
and the general expressions Ior evaluating divergence, curl and gradient are given by
gradient oI scalar: V = + + f
h
f
u h
f
v h
f
w
1 1 1
1
1
2
2
3
3
c
c
c
c
c
c
e e e
06703 21 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
divergence oI vector:
V + +
,
v
1
1 2 3
2 3 1 1 3 2 1 2 3
h h h u
h h v
v
h h v
w
h h v
c
c
c
c
c
c
curl oI vector: V × = v
e e e
1
1 2 3
1 1 2 2 3 3
1 1 2 2 3 3
h h h
h h h
u v w
h v h v h v
c
c
c
c
c
c
These Iormulas have been evaluated Ior a number oI common coordinate systems, including
R.C.C.S., cylindrical and spherical coordinates. The results are tabulated in Appendix A oI BSL
(see pages 738741). These pages are also available online:
rectangular coords.
cylindrical coords:
spherical coords:
Physical Interpretation of Curl
To obtain a physical interpretation oI V×v, let`s consider a particularly simple Ilow Iield
which is called solidbody rotation. Solidbody rotation is simply the velocity Iield a solid
would experience iI it was rotating about some axis. This is also the velocity Iield eventually
Iound in viscous Iluids undergoing steady rotation.
06703 22 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Imagine that we take a container oI Iluid
(like a can oI soda pop) and we rotate the
can about its axis. AIter a transient
period whose duration depends on the
dimensions oI the container, the steady
state velocity proIile becomes solidbody
rotation.
A material point imbedded in a solid
would move in a circular orbit at a
constant angular speed equal to O
radians per second. The corresponding
velocity is most easily described using
cylindrical coordinates with the :axis
oriented perpendicular to the plane oI the
orbit and passing through the center oI
the orbit. Then the orbit lies in a :÷const
plane. The radius oI the orbit is the radial
coordinate r which is also constant. Only
the ucoordinate changes with time and it
increases linearly so that d dt u ÷ const ÷
O.
In parametric Iorm in cylindrical coor
dinates, the trajectory oI a material point
is given by
r(t) ÷ const, :(t) ÷ const, u(t) ÷ Ot
The velocity can be computed using the Iormulas developed in the example on page 12:
v e e e e = + + =
dr t
dt
r
d t
dt
d: t
dt
r
r
r
:
0 0
u
u u
O
O
Alternatively, we could deduce v Irom the deIinition oI derivative
oI a vector with respect to a scalar:
0
lim
t
r d D d
r r
Dt t dt dt
u
u u
A ÷
u A u
= = = = = O
A
e r r
v e e
More generally, in invariant Iorm (i.e. in any coordinate system)
the velocity proIile corresponding to solidbody rotation is given
by
O
(0)
( ) t
r
:
( ) t
side view
top view
r
r
( (0)) u
( ( )) u t
( (0)) u
( ( )) u t
u +
u + Au
u
r
Au
A
A Au s r ÷
u +
u + Au
u
06703 23 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
v r r
p p
= × O (5)
where O is called the angular velocity vector and r
p
is the position vector
*
whose origin lies
somewhere along the axis oI rotation. The magnitude oI O is the rotation speed in radians per
unit time. It`s direction is the axis oI rotation and the sense is given by the 'righthand rule.¨ In
cylindrical coordinates, the angular velocity is
O ÷ Oe
:
and the position vector is r e e
p r :
r : = +
Taking the cross product oI these two vectors (keeping the order the same as in (5)):
v r e e e e e
e 0
= × + × = r : r
: r : :
O O O
u
To obtain this result we have used the Iact that the cross product oI any two parallel vectors
vanishes (because the sine oI the angle between them is zero recall deIinition oI cross product
on p1).
The cross product oI two distinct unit vectors in any righthanded coordinate
system yields a vector parallel to the third unit vector with a sense that can
be remembered using the Iigure at right. II the cross product oI the two unit
vectors corresponds to a 'clockwise¨ direction around this circle, the sense
is positive; in a 'counterclockwise¨ direction, the sense is negative. In this
case, we are crossing e
:
with e
r
which is clockwise; hence the cross product
is ¹e
.
Now that we have the velocity Iield, let`s compute the curl. In cylindrical
coordinates, the Iormula Ior the curl is obtained Irom p739 oI BSL:
V × = ÷
+ ÷
+ ÷
v e e e
1 1 1
r
v v
:
v
:
v
r r
rv
r r
v
:
r
r : r
:
c
cu
c
c
c
c
c
c
c
c
c
cu
u
u
u
Substituting v
r
÷ 0 v
÷ rO v
:
÷ 0
we obtain V × = = v e 2 2 O
:
O OO O
*
The subscript 'p¨ was added here to avoid conIusing the cylindrical coordinate r with the
magnitude oI the position vector. Note that in cylindrical coordinates,
2 2
p
r : r = + = r .
06703 24 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Thus the curl turns out to be twice the angular velocity oI the Iluid elements. While we have
only shown this Ior a particular Ilow Iield, the results turns out to be quite general:
V × = v 2O
Vector Field Theory
There are three very powerIul theorems which constitute 'vector Iield theory:¨
 Divergence Theorem
 Stokes Theorem
 Irrotational · Conservative ·Derivable Irom potential
DIVERGENCE THEOREM
This is also known as 'Gauss
Divergence Theorem¨ or 'Greens Formula¨ (by Landau &
LiIshitz). Let v be any (continuously diIIerentiable) vector Iield and choose A to be any
(piecewise smooth, orientable) closed surIace;
then
n v v
da dJ
A J
= V
where J is the region enclosed by A and n is the
outward pointing unit normal to the diIIerential
surIace element having area da. Although we
will not attempt to prove this theorem, we can
oIIer the Iollowing rationalization. Consider the
limit in which all dimensions oI the region are
very small, i.e. J÷0. When the region is
suIIiciently small, the integrand (which is
+
Carl Eriedrich Gauss (17771855), German mathematician, physicist, and astronomer.
Considered the greatest mathematician oI his time and the equal oI Archimedes and Isaac
Newton, Gauss made many discoveries beIore age twenty. Geodetic survey work done Ior the
governments oI Hanover and Denmark Irom 1821 led him to an interest in space curves and
surIaces.
06703 25 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
assumed to vary continuously with position)* is just a constant over the region:
V
v ÷ const. inside J
n v v v v
da dJ dJ J
A J J
= V = V
= V
Solving Ior the divergence, we get the deIinition back (recalling that this was derived Ior J÷0):
V =
v n v
1
J
da
A
Thus the divergence theorem is at least consistent with the deIinition oI divergence.
Corollaries of the Divergence 1heorem
Although we have written the Divergence Theorem Ior vectors (tensors oI rank 1), it can also
be applied to tensors oI other rank:
nf da f dJ
A J
= V
n
t t da dJ
A J
= V
One application oI the divergence theorem is to simpliIy the evaluation oI surIace or volume
integrals. However, we will use GDT mainly to derive invariant Iorms oI the equations oI
motion:
Invariant: independent oI coordinate system.
To illustrate this application, let`s use GDT to derive the continuity equation in invariant Iorm.
1he Continuity Equation
Let µ(r,t) and v(r,t) be the density and Iluid velocity. What relationship between them is
imposed by conservation oI mass?
* This is a consequence oI v being 'continuously diIIerentiable¨, which means that all the partial
derivatives oI all the scalar components oI v exist and are continuous.
06703 26 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Eor any system, conservation oI mass means:
rate oI acc.
oI total mass
net rate oI
mass entering
=
Let's now apply this principle to an arbitrary
system whose boundaries are Iixed spatial
points. Note that this system, denoted by J can
be macroscopic (it doesn`t have to be
diIIerential). The boundaries oI the system are
the set oI Iixed spatial points denoted as A. OI
course, Iluid may readily cross these
mathematical boundaries.
Subdividing J into many small volume
elements:
dm ÷ µdJ
M dm dJ
J
= =
µ
J J
dM d
dJ dJ
dt dt t
 
cµ
 = µ =
 c
\ .
where we have switched the order oI diIIerentiation and integration. This last equality is only
valid iI the boundaries are independent oI t. Now mass enters through the surIace A.
Subdividing A into small area elements:
n ÷ outward unit normal
n
v da ÷ vol. Ilowrate out through da (cm
3
/s)
µ(n
v)da ÷ mass Ilowrate out through da (g/s)
rate oI
mass leaving
A A J
da da dJ
¦ ¹
= µ = µ = V µ
´ `
¹ )
n v n v v
The third equality was obtained by applying GDT. Substituting into the general mass balance:
cµ
c
µ
t
dJ dJ
J J
= ÷ V
v
06703 27 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Since the two volume integrals have the same limits oI integration (same domain), we can
combine them:
cµ
c
µ
t
dJ
J
+ V
,
=
v 0
Since J is arbitrary, and since this integral must vanish Ior all J, the integrand must vanish at
every point:
*
cµ
c
µ
t
+ V =
v 0
which is called the equation of continuity. Note that we were able to derive this result in its
most general vectorial Iorm, without recourse to any coordinate system and using a finite (not
diIIerential) control volume. In the special case in which µ is a constant (i.e. depends on neither
time nor position), the continuity equation reduces to:
V
v ÷ 0 µ÷const.
Recall that V
v represents the rate oI expansion oI Iluid elements. 'V
v ÷ 0¨ means that any
Ilow into a Iluid element is matched by an equal Ilow out oI the Iluid element: accumulation oI
Iluid inside any volume is negligible small.
Reynolds 1ransport 1heorem
In the derivation above, the boundaries oI the system were Iixed spatial points. Sometimes it
is convenient to choose a system whose boundaries move. Then the accumulation term in the
balance will involve time derivatives oI volume integrals whose limits change with time. Similar
to Liebnitz rule Ior diIIerentiating an integral whose limits depend on the diIIerentiation variable,
it turns out that:
+
*
II the domain J were not arbitrary, we would not be able to say the integrand vanishes Ior every
point in the domain. Eor example:
cos sin
cos sin
u u u u
u u u
t t
t
d d
d
0
2
0
2
0
2
0
0
= =
÷ =
does not imply that cosu ÷ sinu since the integral vanishes over certain domains, but not all
domains.
+
Eor a prooI, see G:1634.
06703 28 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
d
dt
S t dJ
S
t
dJ S t da
J t J t A t
( , ) ( , )( )
( ) ( ) ( )
r r n w
=
c
c
+
(6)
where w is the local velocity oI the boundary and S(t) is a tensor oI any rank. II w ÷ 0 at all
points on the boundary, the boundary is stationary and this equation reduces to that employed in
our derivation oI the continuity equation. In the special case in which w equals the local Iluid
velocity v, this relation is called the Reynolds 1ransport 1heorem.
+
EXAMPLE: rederive the continuity equation using a control volume
whose boundaries move with the velocity oI the Iluid.
Solution: II the boundaries oI the system move with the same velocity as
local Iluid elements, then Iluid elements near the boundary can never
cross it since the boundary moves with them. Since Iluid is not crossing
the boundary, the system is closed.
*
Eor a closed system, conservation oI
mass requires:
d
dt
mass oI
system
= 0
or
dM
dt
d
dt
dJ
J t
= =
µ
0 (7)
Notice that we now have to diIIerentiate a volume integral whose limits oI integration depend on
the variable with respect to which we are diIIerentiating. Applying (6) with w÷v (i.e. applying
the Reynolds Transport Theorem):
( ) ( ) ( )
( , ) ( , )( )
J t J t A t
d
t dJ dJ t da
dt t
 
cµ

µ = + µ
 c
\ .
r r n v
which must vanish by (7). Applying the divergence theorem, we can convert the surIace integral
into a volume integral. Combining the two volume integrals, we have
+
Osborne Reynolds (18421912), Engineer, born in BelIast, Northern Ireland, UK. Best known
Ior his work in hydrodynamics and hydraulics, he greatly improved centriIugal pumps. The
Reynolds number takes its name Irom him.
*
When we say 'closed,¨ we mean no net mass enters or leaves the system; individual molecules
might cross the boundary as a result oI Brownian motion. However, in the absence oI
concentration gradients, as many molecules enter the system by Brownian motion as leave it by
Brownian motion. v is the massaveraged velocity.
06703 29 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
cµ
c
µ
t
dJ
J t
+ V
,
=
v
0
which is the same as we had in the previous derivation, except that J is a Iunction oI time.
However, making this hold Ior all time and all initial J is really the same as holding Ior all J.
The rest oI the derivation is the same as beIore.
STOKES THEOREM
Let v be any (continuously diIIerentiable)
vector Iield and choose A to be any (piecewise
smooth, orientable) open surIace. Then
n v v r
V× =
da d
A C
where C is the closed curve Iorming the edge oI
A (has direction) and n is the unit normal to A
whose sense is related to the direction oI C by the 'righthand rule¨. The above equation is
called Stokes 1heorem.
+
Jelocity Circulation: Physical Meaning
The contour integral appearing in Stokes` Theorem is an important quantity called velocity
circulation. We will encounter this quantity in a Iew lectures when we discuss Kelvin`s
Theorem. Eor now, I`d like to use Stokes Theorem to provide some physical meaning to velocity
circulation. Using Stokes Theorem and the Mean Value Theorem, we can write the Iollowing:
Stokes' Mean Value
Theorem Theorem
2
n
C A
d da A A = V× = V× = O
v r n v n v
Einally, we note Irom the meaning oI curl that V×v is twice the angular velocity oI Iluid
elements, so that n
V×v is the normal component oI the angular velocity (i.e. normal to the
surIace A). Thus velocity circulation is twice the average angular speed oI Iluid elements times
the area oI the surIace whose edge is the closed contour C.
+
Sir George Gabriel Stokes (18191903): British (Irish born) mathematician and physicist,
known Ior his study oI hydrodynamics. Lucasian proIessor oI mathematics at Cambridge
University 18491903 (longestserving Lucasian proIessor); president oI Royal Society (1885
1890).
06703 30 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Example: Compare 'velocity circulation¨ and
'angular momentum¨ Ior a thin circular disk oI Iluid
undergoing solidbody rotation about its axis.
Solution: Choosing cylindrical coordinates with the
:axis aligned with axis oI rotation. Solidbody
rotation corresponds to the Iollowing velocity proIile
(see page 23):
v e = rO
and V × = v e 2O
:
Einally the unit normal to the disk surIace is n ÷ e
:
. Then the velocitycirculation integral
becomes
v r n v e e
d da da R
C A
: :
A
= V× = = 2 2
2
O Ot
According to L&L Vol I
+
page 25, the angular momentum L oI a mass m undergoing motion at
velocity v is the lever arm r times the linear momentum (p ÷ mv): i.e. L ÷ r×p. Summing this
over diIIerential Iluid mass in our disk with dm ÷ µ dJ, the net angular momentum oI the disk is:
L r v r v = ×
( )
= ×
( )
µ µA dJ : da
J A
Since the disk is oI uniIorm thickness A: and density µ, we can write the second equation above.
II the disk is suIIiciently thin that we can neglect the : contribution to the position vector, then
we can approximate r ÷ re
r
in cylindrical coordinates.
+
Substituting into the second integral
above
L e e e = = =
µA µA t
t
µA : r da : r r dr R :
:
A
:
R
:
2 2
0
4
2
2
O O O
Dividing this by the velocity circulation integral:
+
Landau & LiIshitz, Mechanics and Electrodvnamics (Course oI Theoretical Physics: Vol. 1),
Pergamon, 1959.
+
Actually this assumption isn`t necessary since any :component oI r will produce an r
component in the crossproduct and this rcomponent will integrate to zero as long as J is a body
oI rotation about the same axis.
06703 31 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
L
d
R :
R
R : R :
M :
C
J
v r
= = = =
t
µA
µA
t
t µ
t
2
2
1
4
1
4 4
4
2
2 2
O
Ot
A
where M is the mass oI Iluid in the disk. This could be rewritten as
v r
d
L
M
C
:
= 4t
So the velocitycirculation integral is just proportional to the angular momentum per unit mass.
DERIVABLE FROM A SCALAR POTENTIAL
A very special class oI vector Iields consists oI those vectors Ior which a scalar Iield exists
such that the vector can be represented as the gradient oI the scalar:
Suppose: v ÷ v(r) and f ÷ f(r)
II f exists such that: v ÷ Vf
Ior all r in some domain, then f(r) is called the scalar potential oI v and v is said to be derivable
from a potential in that domain.
An example oI a vector Iield which is 'derivable Irom a potential¨
is the gravitational Iorce near sea level:
F
grav
÷ Mgk
and the associated potential energy is:
(:) ÷ Mg:
Note that V ÷ Mgk
is identical to the Iorce, except Ior the sign (introduced by convention). This example also
suggests why  is called the 'potential¨ oI v. Not every vector Iield has a potential. Which do?
To answer this, let's look at some special properties oI such vector Iields.
Property I: iI v÷V then V×v÷0 (irrotational)
ProoI: Recall that V×(V) ÷ 0 (see HWK #2, Prob. 4e). A vector which has this property is said
to be irrotational. This name is an allusion to V×v representing the rotation rate iI v is the Iluid
velocity. V×v÷0 means the Iluid elements are not rotating.
06703 32 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Property II: iI v÷V then v r
d
C
= 0 (conservative)
Ior any closed contour in the region.
ProoI: Using Property I, we know that V×v÷0. Then we can deduce the value oI this closed
contour integral Irom Stokes` Theorem:
C A
d da = V× =
0
v r n v 0
A vector Iield which has this property is said to be conservative. This name is an allusion to the
special case in which v represents a Iorce, like gravity. Then v
dr (Iorce times displacement)
represents the work required to move the object through the Iorce Iield. Saying that the contour
integral vanishes means that the work required to liIt a weight can be recovered when the weight
Ialls. In other words, energy is conserved.
II C is open, v÷V is still quite useIul:
Property III: let C
o
be an open contour connecting points A and B.
II v÷V then }
Co
v
dr ÷ (r
B
)(r
A
)
Ior any contour connecting A and B.
ProoI: Note that V
dr ÷ d (Irom our deIinition oI gradient). Then
o o
B A
C C
d d =  =  ÷ 
v r r r
We call this property path independence. OI course, Property II is just a special case oI this Ior
which A÷B so that (r
B
)  (r
A
) ÷ 0.
THEOREM III
We have just shown that properties I and II are implied by v ÷ V; it turns out that the
converse is also true, although I`m not going to prove it here. We can distill these properties and
their converse into a single statement:
V × =
·
=
V
·
=
v 0
r
r
v
v r
Ior all
in Region
exists
such that ÷
in Region
Ior every
closed in Region
 
d
C
C
0
06703 33 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
TRANSPOSE OF A TENSOR, IDENTITY TENSOR
The transpose oI a tensor t is denoted t
t
and is deIined so that:
v
t ÷ t
t
v
and t
v ÷ v
t
t
Ior all vectors v. Eor example:
iI t ÷ ab
then t
t
÷ ba
More generally, in terms oI scalar components oI t, we can write the relationship between a
tensor and its transpose as:
t
t
if
÷ t
fi
Symmetric 1ensor: t
t
÷ t
An example oI a symmetric tensor is the dyad aa.
Identity 1ensor: Also known as the Idem Factor. Denoted as I and deIined so that:
v
I ÷ v ÷ I
v
Ior any vector v. Clearly I is symmetric, but in addition, dotting it with another vector gives that
vector back (like multiplying by one). In any coordinate system, I can be calculated Irom:
I r
r
r
= V =
c
c
where r is the position vector expressed in terms oI the unit vectors in that coordinate system.
Recalling Irom p5 that gradient can be thought oI as the partial derivative with respect to
position, I can be thought oI as the derivative oI the position vector with respect to itselI. In
R.C.C.S., recall that:
r ÷ xi ¹ vj ¹ :k
and V =
r e e
f i
f
i
i f
r
x
c
c
where r
f
is the f
th
component oI the position vector r and x
i
is the i
th
coordinate. In Cartesian
coordinates, the position vector components are related to the coordinates according to:
06703 34 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
r
1
÷ x
1
÷ x, r
2
÷ x
2
÷ v, and r
3
÷ x
3
÷ ::
then
c
c
o
r
x
f
i
if
=
which is 0 iI i=f or 1 iI i÷f. This leaves:
V =
r e e
f i
i f
*
so I ÷ ii ¹ jj ¹ kk
As a partial prooI that I has the desired properties which make it the identity tensor, consider
dotting it with an arbitrary vector v:
v
(ii¹jj¹kk) ÷ v
ii ¹ v
jj ¹ v
kk
÷ (v
i)i ¹ (v
j)j ¹ (v
k)k
÷ v
x
i ¹ v
v
j ¹ v
:
k ÷ v
Thus we have shown that v
I÷v, as advertised.
DIVERGENCE OF A TENSOR
In presenting the corollaries to the Divergence Theorem, we have already introduced the
divergence oI a tensor. This quantity is deIined just like divergence oI a vector.
V ÷
÷
t t
lim
J J
da
A
0
1
n
Note that this deIinition uses a predot not a
postdot. In R.C.C.S.
t ÷ EEt
if
e
i
e
f
On the x÷x
o
Iace:
n ÷ e
1
*
This expression Ior the identity tensor is valid Ior any set oI orthonormal unit vectors (not just
the Cartesian ones Ior which we have derived it here).
06703 35 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
n e e e e
t t t = = ÷
i f
if i f
f
f
x
f
o
1 1
Similarly, on the x÷x
o
¹Ax Iace, we obtain:
n ÷ ¹e
1
n e
t t =
+
f
f
x x
f
o
1
A
AIter integrating over the area, we obtain:
A A
f x x f x
f
f
da v :
o o
1 2
1 1
+
+
= , ÷ , n e
t t t
A
A A
¦ ¦
Dividing by J:
1
1 2
1 1
J
da
x
A A
f x x f x
f
f
o o
+
+
=
, ÷ ,
n e
t
t t
A
A
Taking the limit as J÷0:
lim
J J
da
x
A A
f
f
÷
=
ct
c
+
0
1
1 2
1
n e
j
t
Adding on similar contributions Irom the v÷const and :÷const Iaces:
if
f
i
i f
x
ct
V t =
c
e
vx xv vv :v v:
xx :x x: ::
x v : x v : x v :
ct ct ct ct ct
      ct ct ct ct
V t = + + + + + + + +
  
c c c c c c c c c
\ . \ . \ .
i j k
06703 36 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Introduction to Continuum Mechanics`
In this course, we will treat Iluids like continua; in other words, we are going to ignore the
molecular granularity oI matter. This is an assumption which we, as engineers, oIten make in
describing transport oI heat, mass, or momentum although we don`t always state this assumption
explicitly. To make the nature oI this assumption clearer, it might help to discuss the alternative.
Eluids are composed oI molecules. In principle, iI you tell me the initial location oI every
molecule in the Iluid and its initial velocity, I can compute the position and velocity at some later
time using Newton`s laws oI motion (i.e. F ÷ ma). The diIIiculty with this approach is that the
number oI molecules in any volume oI Iluid oI interest to us make such a detailed calculation
impractical. Eor example:
1 cm
3
oI water ÷ 3.3x10
22
molecules ÷ 10 million years
Even with a computer operating at 100 mIops, it would take 10 million years to do just one
multiplication Ior each molecule. Molecules oI a liquid collide on the average oI once every 10

12
seconds. To describe one second oI real behavior, I would need 10
12
× 10 million years.
Clearly, this is an absurd length oI time. Although computers get Iaster every year, this will
remain an absurdly long time Ior the Ioreseeable Iuture. The alternative is:
CONTINUUM HYPOTHESIS
A detailed description at the molecular level is not required in order to predict macroscopic
behavior oI any material. Eor example, it is not necessary to know the precise location oI every
molecule oI Iluid; it turns out that all that is needed Ior most applications is the distribution oI
mass described by the density proIile µ(r) oI molecules in some region:
µ( ) ÷
lim
r
J J
m
i
i
÷
0
1
where m
i
is the mass oI molecule i, the sum is over all the
molecules inside surIace A, and J is the volume. In Iluid
mechanics, as in heat and mass transIer, we make an assumption
known as the "continuum hypothesis."
*
ReIerence: G.E. Mase, "Continuum Mechanics," Schaum's Outline Series, McGrawHill, 1970.
06703 37 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Basically this assumption is that the limit above will
converge long beIore the dimensions oI J shrink to
molecular size.
Similarly, we don't need to know the translational,
rotational, vibrational and electronic energy oI each
molecule. We usually need only to know the internal
energy per unit volume as a Iunction oI position,
which in turn, maniIests itselI macroscopically as
temperature.
A more precise statement oI the continuum hypothesis
is:
Continuum Hypothesis  the region to be described can be subdivided into a set oI
(inIinitesimal) volume elements, each oI which simultaneously:
1) is small enough to be considered uniIorm (i.e. any spatial variations in properties  such as
µ, v, T, p  inside the volume element are negligible); and
2) is large enough to contain a statistically large number oI molecules.
In other words, we are assuming that dJ exists such that the two conditions above are both
satisIied. Materials which obey this 'hypothesis¨ as said to behave as a continuum. Generally,
the continuum hypothesis works well provided all the dimensions oI the system are large
compared to molecular size. An example oI a situation in which the continuum hypothesis does
not work is the Ilow oI dilute gases in small pores, where the mean Iree path (Ior the collision oI
molecules) is comparable to the dimensions oI the pore. This situation is called 'Knudsen
diIIusion.¨
The basic problem in continuum mechanics is to describe the response oI material to stress. A
quantitative statement oI that response is known as:
Constitutive Equation  model which describes how a material will respond to stress.
Eamiliar examples oI constitutive equations include:
1) Hooke's law oI elasticity (solids)
2) Newton's law oI viscosity (Iluids)
Many materials, like toothpaste and polymer melts, have characteristics oI both solids and Iluids
and do not obey either oI these simple "laws." Such Iluids are called viscoelastic.
06703 38 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
CLASSIFICATION OF FORCES
Having derived an equation (the Continuity equation) to describe the relationship among the
variables which is imposed by conservation oI mass, the remaining Iundamental principle oI
physics is Newton`s second law (L
i
F
i
÷ ma) which, as it turns out, is equivalent to conservation
oI momentum. To apply this principle, we will need to list the Iorces which can act on Iluid
systems. Eorces tend to Iall into one oI two diIIerent categories, depending on the range over
which they act: longrange (compared to molecular size) Iorces are computed as volume integrals
(called 'body Iorces¨) and shortrange Iorces are computed as surIace integrals (called 'surIace
Iorces¨).
OI course, gravitational Iorces have the longest range oI any known Iorce. Eor example,
gravitational Iorces between planets and the sun determine their orbits. In particular, all Iluid
elements (not just those at the system boundary) Ieel a gravitational Iorce oI interaction with the
rest oI the universe outside the system boundaries. Thus gravity is a 'body Iorce.¨
body forces: those which act on every Iluid element in body (e.g. gravity):
dF
g
÷ (dm)g ÷ µg dJ
At the other end oI the spectrum are Iorces
which have very short range. II the range is oI
molecular dimensions, then only Iluid
elements experience a nonzero interaction
with the universe outside the system.
Although interior Iluid elements might interact
with one another through this shortrange
Iorce, this interaction is not considered in a
Iorce balance, because the 'action¨ and
'reaction¨ Iorces cancel, leaving no net
contribution to the Iorce on the system. When
only surIace elements Ieel a particular Iorce Irom outside, that Iorce is called a 'surIace Iorce.¨
At the molecular scale, pressure arises Irom the momentum transIerred during collisions between
molecules outside and molecules inside the system. Since only surIace molecules will be struck
Irom outside, pressure is a surIace Iorce.
06703 39 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
surface forces: those which act only on surIaces
(including mathematical boundaries)
One example is hydrostatic pressure:
dF
p
÷ pn da
where dF
p
is the Iorce exerted on the system
(through the surIace element da) bv the Iluid
outside, and n is a unit outward (to system)
normal. Eor a prooI that this is the correct Iorm Ior
hydrostatic pressure, see Batchelor, Section 1.3.
Hydrostatic Equilibrium
II our material is a Iluid and iI it is at rest (no velocity and no acceleration), then gravity and
hydrostatic pressure Iorces are usually the only Iorces acting on the system. At equilibrium, the
Iorces must be balanced. Thus Newton`s
+
2
nd
law, which generally requires F a
i
i
M
= ,
reduces to F 0
i
i
= at mechanical equilibrium.
In our case, this means F
g
¹ F
p
÷ 0
F
g
÷ }
J
µgdJ
F
p
÷ }
A
npda ÷  }
J
VpdJ
To obtain this last result, we applied one oI the corollaries oI the Divergence Theorem.
Substituting back into the Iorce balance and combining the two volume integrals leads to:
F
g
¹ F
p
÷ }
J
µgVpdJ ÷ 0
Since J is arbitrary, we conclude that the integrand vanishes:
Vp ÷ µg
This says that the pressure increases in the direction oI the
acceleration oI gravity (downward), which correctly describes
+
Sir Isaac Newton (16421727), English mathematician and natural philosopher (physicist);
considered by many the greatest scientist oI all time; invented diIIerential calculus and
Iormulated the theories oI universal gravitation, terrestrial mechanics, and color.
06703 40 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
(Ior example) how the pressure increases with depth in an ocean.
Flow of Ideal Fluids
Now let's consider Iluids in motion. The simplest analysis is Ior:
ideal fluid  deIormation oI Iluid elements is an isentropic process (i.e. adiabatic and
reversible):
µ ÷ 0 and k ÷ 0
where µ is the viscosity and k is the thermal conductivity. Generally this means that any viscous
Iorces are negligible (since viscous Iorced represent Iriction arising between Iluid elements and
Iriction gives rise to irreversibility). Eurthermore, to keep the process adiabatic, the thermal
conductivity must also be negligible.
EULER'S EQUATION
Suppose these conditions on the Iluid are met. Thus consider the isentropic deIormation oI
an ideal Iluid Ior an arbitrary macroscopic system. In addition to pressure and gravity, we must
also consider inertia when the system accelerates. Newton's law requires:
Ma ÷ E
i
F
i
Let r(t) denote the trajectory oI one particular
Iluid element inside the system. Then the
velocity oI the Iluid element is:
v
r
=
D
Dt
while the acceleration is:
a
v
=
D
Dt
We use the material derivative here, since we are Iollowing a particular material point.
Multiplying the acceleration by the mass oI the Iluid element gives the inertia:
dm dJ
D
Dt
a
v
= µ
To get the net inertia oI the entire system, we must repeat this calculation Ior each oI the Iluid
elements composing the system and add them up:
06703 41 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
J
D
M dJ
Dt
= µ
v
a
I`ve put carets (·.~) around the Ma to indicate that this is the positionaverage inertia oI the
system (since the local (dm)a varies Irom point to point within a general system). Newton's
second law requires us to equate this with the net Iorce acting on the system:
µ µ
D
Dt
dJ dJ pda
J
p g
J A
p dJ
J
v
F F g n
= + = ÷
V
Using the divergence theorem to convert the surIace integral into a volume integral, we have
three volume integrals over the same domain. Combining these three volume integrals leaves:
µ µ
D
Dt
p dJ
J
v
g 0 ÷ + V
,
=
Since this must hold Ior any choice oI J, the integrand must vanish at each point in the domain.
AIter dividing by µ:
D
Dt
p
v
g = ÷ V
1
µ
(8)
which is called Euler's Equation (1755).
*
Significance: When combined with a statement oI continuity, Euler`s equation oI
motion provides as many equations as unknowns.
Another relationship among the unknowns is the continuity equation (see page 27), which
comes Irom the mass balance.
cµ
c
µ
t
+ V =
v 0
*
Euler, Leonhard, 170783, Swiss mathematician. The most proliIic mathematician who ever
lived, he worked at the St. Petersburg Academy oI Sciences, Russia (172741, 176683), and at
the Berlin Academy (174166). He contributed to areas oI both pure and applied mathematics,
including calculus, analysis, number theory, topology, algebra, geometry, trigonometry, analytical
mechanics, hydrodynamics, and the theory oI the moon's motion.
06703 42 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Eor an incompressible Iluid, µ ÷ const. w.r.t. both time and position. Then the continuity
equation reduces to:
imcompressible Iluid: V
v ÷ 0, (9)
To see that we now have as many equations as unknowns, note that the unknowns in (8) and
(9) are
unknowns: 4 scalars v and p
which represents the 3 scalar components oI v plus p, Ior a total oI 4 scalar unknowns. To
evaluate these unknowns, we have equations (8) and (9):
equations: 4 scalars Euler ¹ continuity
but Euler`s equation (8) is a vector equation, which can be expanded into 3 independent scalar
equations. When added to continuity (a scalar equation), we obtain a total oI 4 independent
scalar equations, the same number as oI scalar unknowns. Thus we are now in position to begin
solving problems involving Iluid Ilow. We will call (8) and (9) 'Euler`s equations oI motion Ior
incompressible Iluids.¨
EXAMPLE. Water in a partially Iilled tank undergoes
uniIorm
+
acceleration a in the horizontal plane. Eind the
angle u oI inclination oI the water`s surIace with respect
to the horizontal plane.
Solution. The key to solving this problem is to recognize
that, regardless oI the angle oI inclination, the pressure is
equal to 1 atm everywhere on the Iree surIace. Then the
pressure gradient Vp must be normal to this plane. II we
can Iind the orientation oI Vp we will have the orientation oI the Iree surIace. Recall Euler's
equation oI motion Ior an ideal Iluid:
D
Dt
p
v
g
a
¸
= ÷ V
1
µ
+
By 'uniIorm acceleration¨ I mean the same acceleration is experienced at each point and at
each time. Eor a Iluid, uniIormity at each position occurs only in the steady state aIter a transient
which is nonuniIorm.
06703 43 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Dv/Dt is just the acceleration oI the Iluid in a stationary reIerence Irame.
At steady state, all oI the Iluid will undergo the same uniIorm acceleration
as the tank; so Dv/Dt is just a. Solving Ior the gradient, we have
1
µ
V = ÷ p g a
Using vector addition in the drawing at right, we can see that the angle oI
inclination oI the Iree surIace (relative to the horizon) is just
u =
÷
tan
1
a
g
We were lucky in the previous example, because we knew the leIthand side oI (8), so instead
oI 4 scalar unknowns, we only had one: p. The solution was relatively easy. In the more general
problem, the leIthand side oI (8) is an unknown nonlinear partial diIIerential equation:
c
c µ
v
v v g
t
p + V = ÷ V
1
(10)
In this Iorm, we have expanded D Dt v using the relationship between material derivative and
partial derivative (see page 15). Now we have 4 scalar unknowns: the three scalar components oI
v and pressure: v
x
, v
v
, v
:
and p. Coupled with the continuity equation (Ior an incompressible
Iluid)
V
v ÷ 0
Euler`s equation also gives us 4 scalar equations. One important class oI solutions has the Iorm v
÷ V, which is called 'potential Ilow.¨ In the next section, we discuss how this Iorm comes
about and identiIy which physical problems have this Iorm.
KELVIN'S THEOREM
An important precursor to the theory oI potential Ilow is the principle oI conservation of
circulation. BeIore stating this principle, let me deIine a quantity which Landau & LiIshitz
+
call
the velocity circulation:
+
L.D. Landau and E.M. LiIshitz, Fluid Mechanics (Vol. 6 oI a Course oI Theoretical Physics),
Pergamon, New York, 1959.
06703 44 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
I ÷
v r
d
C
Ior any closed contour. Recall that we showed on
page 30 that this contour integral is associated with the
average angular momentum oI Iluid elements located
on the surIace whose edge is C. Kelvin
+
showed that
this velocity circulation is conserved:
d
dt
I
= 0
Ior any set oI material points Iorming a closed
contour in an ideal Iluid. This result is called Kelvin's
1heorem.
Partial prooI:
v
Since the contour C is composed oI material points, the time derivative oI this
contour integral is like the material derivative:
?
C C
d D D
d d
dt Dt Dt
I
= =
v
v r r
Since we are always integrating over the same set oI material points, a boundary term does not
arise when we interchange integration and diIIerentiation operators, although the set oI spatial
points is timedependent: C ÷ C(t). OI course, we have not rigorously shown this step to be
valid, thus we only claim the prooI is 'partial.¨ Next, we substitute Euler`s equation and write
each term as the gradient:
d
dt
p d
p
d
p
d
C
g
C
g
C
I
= ÷ V
= ÷V ÷ V
= ÷ V +
=
g r r r
1
0
µ µ

µ
In the second equality above, we have introduced the potential energy per unit mass, 
g
. Recall
that gravity is a conservative Iorce Iield (see page 32). Eor an object oI constant mass (e.g. a
+
Lord Kelvin (William Thomson), 1st Baron, 18241907, British mathematician and physicist;
b. Ireland. He was proIessor (184699) oI natural philosophy at the Univ. oI Glasgow. His work
in thermodynamics coordinating the various existing theories oI heat established the law oI the
conservation oI energy as proposed by James Joule. He discovered what is now called the
'Thomson eIIect¨ in thermoelectricity and introduced the Kelvin scale, or absolute scale, oI
temperature. His work on the transmission oI messages by undersea cables made him a leading
authority in this Iield.
v
Eor a more rigorous prooI, see Batchelor p269. Eor a more intuitive prooI, see L&L, p15.
a material point
trajectory oI a
material point
contours
oI material points
at diIIerent times
C t ( )
C t ( )
06703 45 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
brick), Theorem III guarantees that a scalar Iield (r) exists such that F
g
÷ mg ÷ V. Eor an
object oI constant mass, we could divide both sides oI this equation by m and bring m inside the
gradient operator: g ÷ V(/m). Similarly, a diIIerential volume element dJ would have a
diIIerential mass dm and a diIIerential potential energy d, such that g ÷ V(d/dm). So generally
we can write:
g ÷ V
g
(11)
where 
g
=
potential energy
mass
.
The nexttolast expression (in the equation Ior dI/dt) must vanish, because it has the Iorm
Vs
dr ÷ ds: integrating any total diIIerential around a closed contour yields zero. Thus
v r
d
C
= const. w. r. t. time
Keep in mind that this applies only iI C is composed oI material points and only Ior ideal Ilow.
Since C is composed oI material points (which in general move with diIIerent velocities), the
contour may change shape or move. Given the meaning oI this contour integral (I ÷ angular
momentum), this result implies that (in the absence oI Iriction) angular momentum oI Iluid is a
constant. In general, we change the angular momentum oI some object by applying a torque. So
this result (i.e. Kelvin`s theorem) means that (in the absence oI Iriction) there is no way to apply
a torque to ideal Iluid elements.
II you think about it, this makes sense: the usual way to apply a torque (with
our hands to a cylinder, say) is to hold the cylinder between our hands and
then move our hands in opposite directions, as shown in the sketch at right.
We thus rely on Iriction between our hands and the cylinder to exert the
torque. II the cylinder were greased and our hands slipped over its surIace,
we would not be able to apply the torque. This is the essence oI what
Kelvin`s theorem is saying.
IRROTATIONAL FLOW OF AN INCOMPRESSIBLE FLUID
As an example, consider towing a submerged object through an 'ideal Iluid¨ which is
otherwise stagnant.
v(r,t÷0) ÷ 0
Consider an arbitrary closed contour in the Iluid Iar
Irom the disturbance caused by the motion oI the submerged object. The contour integral
vanishes since v vanishes:
06703 46 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
v r
d
C
= 0 Ior t÷0 since v ÷ 0
Ior every such contour C. Now at some later time the submerged object moves into the vicinity
oI C which causes v to be nonzero. Despite this, Kelvin`s Theorem still requires
v r
d
C
= 0 Ior all t although v = 0
This is also true Ior every closed contour in the region (since every contour initially had zero
value Ior this integral).
Applying Theorem III: V×v ÷ 0 Ior all r,t (12)
which is sometimes called the persistence of irrotationality. Also Irom Theorem III, we know
(r,t) exists such that:
v ÷ V (13)
where  is called the velocity potential.
Significance: Knowing that the solution has the Iorm given by (13) allows us to
decouple the Iour scalar equations represented by (8) and (9):
Substituting (13) into (9) yields Laplace`s equation in the velocity potential:
(13) into (9): V
2
 ÷ 0
Instead oI 4 equations in 4 unknowns, we now have a single equation which can be solved Ior 
and v ÷ V, without any coupling to Euler`s equation (8). Although Euler was the Iirst to suggest
this approach, this is called Laplace's Equation
*
aIter the Erench mathematician who solved this
equation in so many cases.
Knowing the velocity proIile v, we can now determine the pressure proIile p Irom Euler's
equation:
c
c µ
v
v v g
t
p + V = ÷ V
1
(14)
*
Pierre Simon de Laplace (17491827), Erench mathematician and astronomer, noted Ior his
theory oI a nebular origin oI the solar system and his investigations into gravity and the stability
oI planetary motion.
06703 47 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
We will now integrate this vector equation to obtain a single scalar equation Ior the pressure
proIile. Each term in (14) can be expressed as a gradient oI something. Eor example, we`ve
already seen in (11) that:
g ÷ V
g
Similary, iI µ÷const, then:
1
µ µ
µ
V = V
p
p
÷const
(15)
Eor potential Ilow, the unsteady term becomes
c
c
c
c
c
c
v
t t t
= V = V
(16)
Einally, Ior the convective term, we can apply identity A.3:
+
v v v v
0
V = V ÷ × V ×
1
2
2
v
(17)
where v
v ÷ v
2
where v ÷ v. In our particular case, the second term vanishes because V×v ÷ 0 Ior potential
Ilow. Substituting (11) and (15) through (17) into (14):
V + + +
=
c
c

µ t
v p
g
2
2
0
which is called Bernoulli`s Equation.
+
It implies that
2
const w.r.t.
2
g
v p
t
c
+ +  + =
c µ
r (18)
+
'Identity A.3¨ in this equation reIers to one oI the mathematical identities summarized on the
handout titled 'UseIul Identities in Vector Notation¨.
+
Daniel Bernoulli (17001782, Swiss) has oIten been called the Iirst mathematical physicist; the
teacher oI Leonhard Euler. His greatest work was his Hvdrodvnamica (1738), which included
the principle now known as Bernoulli's principle and anticipated the law oI conservation oI
energy and the kineticmolecular theory oI gases developed a century later. He also made
important contributions to probability theory, astronomy, and the theory oI diIIerential equations.
06703 48 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
is spatially uniIorm, but it might depend on time. Once the velocity proIile is obtained (by
solving Laplace`s equation), both  and v are known, leaving p as the only unknown.
Potential Flow Around a Sphere
To solve a typical problem involving potential Ilow, we would Iirst solve Laplace's equation
to obtain the velocity proIile and then we can evaluate the pressure proIile using Bernoulli's
equation. Let's illustrate the procedure using an example:
EXAMPLE: Eind the velocity and pressure proIiles Ior potential
Ilow caused by a sphere oI radius R moving through a stagnant
Iluid with velocity U.
Solution: II the Iluid behaves ideally, it undergoes potential Ilow
and the velocity proIile must satisIy Laplace`s equation:
P.D.E.: V
2
 ÷ 0
Boundary conditions can be Iormulated by recognizing that Iluid Iar Irom the sphere is
unperturbed:
b.c. #1: v ÷ 0 Iar Irom sphere
while Iluid near the sphere cannot penetrate the sphere. To express this mathematically, recall
our 'bucketandstopwatch¨ method Ior deIining Iluid velocity (see page 13). ModiIying it
slightly to account Ior the movement oI the surIace element at velocity U, the Ilowrate across a
surIace element oI area da is given by:
dq da = ÷ = n v U
0
Eor an impenetrable sphere, the Ilowrate must vanish
b.c. #2: n v U
÷ = 0 on sphere
To solve this problem, we adopt a new reIerence
Irame in which the origin moves along with the
center oI the sphere. It turns out that the PDE does
not change upon this shiIt in reIerence Irame Ior
velocity: the new velocity potential must also
satisIy Laplace's equation:
06703 49 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
P.D.E.: V
2
 ÷ 0
However, the boundary conditions are
changed. In this moving coordinate system,
the sphere appears to be stationary and the
Iluid at inIinity is undergoing uniform flow
b.c. #1: v ÷ U ÷ Uk as r÷ · (19)
where U is the velocity oI the sphere in the
original stationary reIerence Irame.
Eor a stationary sphere, no Iluid entering the
sphere means dq ÷ n
v da ÷ 0, or
b.c. #2: n
v ÷ v
r
÷ 0 at r÷R
Next we rewrite the b.c.`s in terms oI the velocity potential in spherical coordinates. In terms oI
velocity potential, (19) becomes:
V ÷ Uk
Now we need to translate k into the unit vectors in spherical coordinates.*
ReIerring to the Iigure above (see page 8), we note that e
r
, e
, and e
:
÷ k all lie in the same
u÷const plane (shaded region oI leIthand Iigure above). II we shiIt all three unit vectors to the
origin (recall that the origin is not part oI the deIinition oI any vector) and reorient the u÷const
plane to coincide with the plane oI the page, then we get the Iigure above at right, Irom trigono
metry oI the right triangle it is apparent that
k e e = ÷ cos sin u u
u
r
* Here u is the spherical coordinate, while  is the velocity potential. Eor more on spherical
coordinates, see BS&L p733.
06703 50 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Thus the b.c. can be written as
as r÷ ·: V ÷ ÷ U U
r
r
r
cos sin u u
c
c
c
cu
u
e e
1
Equating corresponding components:
c
c
u
r
U = cos and
1
r
U
c
cu
u = sin
Integrating either PDE leads to
b.c. #1:  ÷ Urcosu ¹ const as r÷ ·
where we have arbitrarily selected a value oI zero Ior this 'const¨.
*
In order to translate b.c. #2,
use tables (see spherical coords
+
) to express the gradient in spherical coordinates:
1 1
sin
r r r
v v v
r r r
u u u u u u
= V
c c c
+ + = + +
c cu u cu
v
e e e e e e
Dotting both sides by the unit vector n ÷ e
r
, then using b.c. #2:
b.c. #2: v
r
÷ c/cr ÷ 0 at r÷R
We look Ior a solution which is independent oI u:
 ÷ (r,u)
This implies that the Iluid does not have any u
component oI velocity. In other words, the
trajectory oI any Iluid element remains entirely on a
single u÷const surIace. The sketch at right shows
some edge views oI u÷const planes (looking along
the zaxis which passes through the center oI the
sphere).
Erom p740 oI BS&L, we have V
2
 in spherical
coordinates:
*
Like most energies in thermodynamics, the reIerence state Ior potential is arbitrary and can be
chosen solely Ior convenience.
+
http://www.andrew.cmu.edu/course/06703/Vops¸sph.pdI
06703 51 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
1
0
2
r
r
r
r
r
c
c
c
c
c
cu
c
cu
2
2
¹
1
sin
sin ÷
u
u (20)
c/cr ÷ 0 at r÷R (21)
 ÷ Urcosu as r÷ · (22)
Although there are systematic procedures, like 'separation oI variables,¨ Ior solving P.D.E.'s
which work in many cases, one should Iirst check to see iI there is a simple solution. The
problem appears Iormidible, but notice that the trivial solution:
 ÷ 0 ?
satisIies the P.D.E. and the b.c. at r÷R. UnIortunately, it does not satisIy the b.c. at r÷ ·, so we
will have to try another guess. Since the Iailure occurred with the b.c. at r÷ ·, we look there Ior
the Iorm oI our second guess:
 ÷ Arcosu ?
This too satisIies the P.D.E. and the b.c. at r÷ · (provided A÷U) but to satisIy the b.c. at r÷R, A
must be chosen as 0. The tells us that A should have diIIerent values at diIIerent r's. So we try a
third guess which is slightly more general than the second:
(r,u) ÷ f(r)cosu (23)
Substituting (23) into (20)(22), we Iind that cosu cancels out, leaving:
r f rf f
2 2 0 '' + ' ÷ =
' = f 0 at r÷R
f ÷ Ur as r÷ ·
06703 52 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Thus we have reduced the problem oI solving the P.D.E. to one oI solving an O.D.E. We
recognize this O.D.E. to be a CauchyEuler equation,
+
which always has at least one solution oI
the Iorm f ÷ r
n
. The general solution turns out to be:
f(r) ÷ Ar
2
¹ Br
Using the b.c.'s, we can evaluate A and B. The particular solution to this problem is:
f r U r
R
r
= +
1
2
3
2
Substituting this back into (23) leaves:
 u u r U r
R
r
, cos = +
1
2
3
2
v
r
U
R
r
r
= = ÷
,
c
c
u 1
3
cos
v
r
U
R
r
u
c
cu
u = = ÷ +
1
1
1
2
3
3
sin
Notice that Ior u÷t (u÷0), v
÷0 but v
r
decreases (in absolute
magnitude) Irom U at r÷· to 0 at r÷R. At u÷t/2, v
r
÷0 but v
increases Irom U to (3/2)U. This
increase is necessary to make up Ior the decrease in Ilow caused by the sphere blocking part oI
the Ilow path.
Having solved Ior the velocity proIile, we can determine the
pressure proIile Irom Bernoulli's equation (18). Assuming
steady state:
+
The general Iorm oI an N
th
order CauchyEuler equation is a x
d v
dx
n
n
n
n
n
N
=
=
0
0. At least one oI
the N linearly independent solutions has the Iorm v(x) ÷ Ax
. Substituting this Iorm leads to
a Ax a n Ax
n
n
N
0
1
1 2 1 0
o o
o o o o + ÷
( )
÷
( )
÷ +
( )
=
=
Dividing out the common Iactor Ax
, we obtain an N
th
order polynomial Ior o. Each distinct
root oI the polynomial leads to a separate solution. In this example, the roots are o ÷ 2 and ¹1.
v
v r
r
=
=
0
2
u
t
u ( , )
=
=
u
v r
v
r
( , ) 0
0
u
=
v r
v
r
( , ) t
u
= 0
U
R
06703 53 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
v p
g
2
2
+ +
µ
 ÷ const
independent oI position. The "const" can be evaluated using any point where we know both the
velocity and the pressure. Suppose the pressure oI the undisturbed Iluid is known in the
reIerence plane Ior the gravitational potential (
g
÷ 0):
at r÷·, 
g
÷ 0: p ÷ p
·
and v
2
÷ U
2
Thus
v p U p
g
2 2
2 2
+ + = +
·
µ

µ
or p r p U v
g
, u µ µ
,
= ÷ + ÷
·
1
2
2 2
where v v r v r
r
2
2 2
= = + v v
, , u u
u
Substituting the known velocity proIile, we obtain the pressure proIile. Let's Iocus on the
pressure proIile over the surIace oI the sphere:
Ior r÷R: p R p U
g
p R
p R h
d
, cos
,
,
u µ µ u
u
u
,
= ÷ + ÷
·
hydrostatic head
dynamic head
1
8
2 2
9 5
In the sketch at right, we plot the dynamic pressure (dropping
the contribution Irom hydrostatic equilibrium). Note the
location oI regions having high and low pressure. The sphere is
being pushed in at the poles and allowed to expand at the
equator. This is why a large bubble rising through stagnant
water tends to become distorted Irom spherical shape. Such
bubbles tend to become extended in the horizontal plane and
compressed in the vertical direction by the higher pressure.
06703 54 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
d'Alembert's Paradox
What is the net Iorce on a rigid sphere owing the pressure proIile developed by potential Ilow
around it? The answer turns out to be:
F n n n n g
g 0
p
A
h d
A
h
A
J
d
A
pda p p da p da p da J = ÷ = ÷ + = ÷ ÷ = ÷
µ
µ
(24)
where J is the volume oI the sphere.
ProoI: Iirst consider the contribution Irom dynamic pressure:
p U
d
÷ ÷
1
8
2 2
9 5 µ u cos
,
The n in (24) is a unit vector normal to the surIace, pointing outward. In spherical coordinates,
with their origin at the center oI the sphere, this vector is the unit vector in the rdirection:
n ÷ e
r
(u,)
which direction depends on location on the surIace. Although its length is a constant, its
direction varies with position oI the sphere; thus n cannot be treated as a constant. Anticipating
that any net Iorce will be parallel to the direction oI Iluid Ilow, we
dot both sides oI (24) by k:
F pda
p p
A
= = ÷
k F k n
cosu
¸
Eor the contribution Irom dynamic pressure, p
d
(r÷R,u) depends
solely on u, so we choose the strip oI width R du and length
2tRsinu as our diIIerential area da. On this strip u is virtually
constant.
F p R R Rd
R f d
R U x x dx
dp d
f
d
x
= ÷
= ÷
= × ÷ =
=
÷
cos , sin
cos cos sin
cos
cos
cos
u u t u u
t u u u u
t µ
u
t
t
u
u
,
2
2
2 9 5 0
0
2
0
2 1
8
2 2
1
1
0
06703 55 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
The nonzero contribution Irom (24) comes Irom the hydrostatic head. We will leave this
calculation as an exercise Ior the reader (HWK #4, Prob. 1). The net Iorce due to pressure is
2 2 1
8
sphere sphere
9cos 5
p g
J
p da U da J
·
÷µ
(
= ÷ ÷ µ ÷ µ u ÷ = ÷µ
(
¸ ¸
g 0
F n n g
The net Iorce on the sphere is the sum oI its weight and the net pressure Iorce:
F F g g g 0
p g s s
J J J + = ÷ + = ÷ = µ µ µ µ
which is the diIIerence in weight oI the sphere and the weight oI the Iluid displaced by it.
Because the pressure Iorce is independent oI the speed oI motion oI the sphere through the Iluid,
the particle will continue to accelerate Iorever, without ever reaching a 'terminal velocity.¨ OI
course, experiments show that Ialling particles reach a terminal velocity which implies that the
gravitational Iorce is balanced by some other Iorce. The other Iorce is Iluid drag (Iriction), which
is not predicted by potential Ilow. This serious discrepancy between the predictions oI potential
Ilow theory and experiment is known as:
d'Alembert's
paradox  potential Ilow predicts no drag but experiments indicate drag.
Despite this, potential Ilow is still useIul:
Uses of potential flow predicts liIt (but not drag) on
streamline objects moving through stagnant
Iluid at high Reynolds numbers (but still sub
sonic, i.e. v ·· c).
· correctly predicts v(r,t) and p(r,t) except very near the
surIace oI the object (i.e. inside boundary layer) and in wake see pressure proIile on airIoil
shown below.
· Ior asymmetric shapes (e.g. airplane wings), it correctly predicts a liIt. See liIt Iorce shown
below as a Iunction oI the angle oI attack.
+
Jean le Rond d`Alembert (171783), Erench mathematician and philosopher, a leading Iigure
oI the Enlightenment. His treatise on dynamics (1743) enunciated d'Alembert's principle, which
permitted the reduction oI a problem in dynamics to one in statics. He did important work on the
mechanics oI rigid bodies, the motions oI Iluids and vibrating strings, and the threebody
problem in celestial mechanics.
06703 56 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Eigs. 1.12 and 1.13 taken from Schlichting, 6th ed., p22f.
Stream Function
BeIore we resolve d`Alembert`s paradox by adding viscous Iorces, let's step back Ior a
moment and review what we have accomplished Ior potential (or irrotational) Ilow. The
mathematical problem might be stated as: Iind v(x,t) such that:
V×v ÷ 0 (25)
and V
v ÷ 0 (26)
Eqs. (25) and (26) represent Iour partial diIIerential equations in the components oI the unknown
vector Iield.* Recognizing that the solution is derivable Irom a potential allows us compress
these Iour equations into one scalar equation in one unknown:
v÷V: V
2
 ÷ 0
which is quite a remarkable trick. The potential is not the only scalar Iield which a vector Iield
can be expressed in terms oI. Velocity can also be expressed in terms oI a stream function.
*
Although it might appear that we have overspeciIied the problem by speciIying both the
divergence and curl (which represent Iour scalar equations in the three components oI v), this
turns out not to be true. In general, both the divergence and curl must be speciIied throughout
some region in space beIore the vector Iield can be determined in that region.
06703 57 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Potential (  ) a scalar Iield whose relationship to v is careIully selected to automatically satisIy
irrotationality
v ÷ V ÷ V×v ÷ 0
Whereas the relation between velocity and scalar potential is chosen to automatically satisIy (25),
the relationship between velocity and stream Iunction is chosen to automatically satisIy (26):
v ÷ f(¢) ÷ V
v ÷ 0
Stream Function (¢ ¢¢ ¢) a scalar Iield whose relationship to v is careIully selected to
automatically satisIy continuity.
It turns out that it is suIIicient to express v as the curl oI another vector. According to Identity
C.6, the divergence oI the curl oI any vector is zero (See HWK #2, Prob. 4d):
v ÷ V×u ÷ V
v÷ 0
A vector which can be expressed as the curl oI another vector is said to be solenoidal. u is called
the vector potential oI v. OI course, knowing that v ÷ V×u isn't always oI much help because
we just trade one unknown vector Ior another. Eortunately, there are several broad classes oI
Ilows Ior which the Iorm oI the vector potential is known:
TWOD FLOWS
When nothing happens along one oI the three directions in R.C.C.S., we have 2D flow:
v ÷ v
x
(x,v)i ¹ v
v
(x,v)j
or v
:
÷ 0, c/c: ÷ 0
Eor such a Ilow,
+
identity
E.3
÷  ( , )  ÷
y
x
: :
x v : x : v :
x v
x v x v
÷
V× ¢ V¢ + ¢V×
  c¢ c¢ c¢ c¢
= + × = × + ×

c c c c
\ .
0
e
e
v e e
e e e e e e e
(27)
In terms oI its scalar components, the velocity is:
+
'Identity E.3¨ in this equation reIers to one oI the mathematical identities summarized on the
handout titled 'UseIul Identities in Vector Notation¨.
06703 58 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
v
v
v
x
v
x v :
= = ÷ =
c¢
c
c¢
c
0 (28)
Next we substitute this Iorm Ior v into (26):
identity
C.6
0
:
V = V V× ¢ ( =
¸ ¸
v e
which automatically satisIies continuity (26), Ior any choice oI ¢(x,v). The scalar Iield ¢(x,v) is
called the stream function. Eor irrotational Ilow, the problem would be to determine ¢(x,v) such
that (25) is satisIied:
V×V×(¢e
:
) ÷ 0
We can reduce this to a scalar equation. Using identity E.5 Irom handout:
V×V×(¢e
:
) ÷ VV
(¢e
:
) ÷ V
2
(¢e
:
)
but V
(¢e
:
) ÷ c¢/c: ÷ 0
and V
2
(¢e
:
) ÷ (V
2
¢)e
:
Thus V×v ÷ (V
2
¢)e
:
So Ior irrotational Ilow, the streamIunction must also satisIy Laplace's equation:
V×v ÷ 0: V =
0 ¢
Unlike the scalar potential, the streamIunction can be used in all 2D Ilows, including those Ior
which the Ilow is not irrotational. Indeed, we will use the streamIunction to solve Stokes Ilow oI
a viscous Iluid around a sphere, in which the Iluid is not even ideal.
AXISYMMETRIC FLOW (CYLINDRICAL)
Another general class oI Ilows Ior which a streamIunction exists is axisymmetric Ilow. In
cylindrical coordinates (r,u,:), this corresponds to:
v ÷ v
r
(r,:)e
r
¹ v
:
(r,:)e
:
or v
÷ 0 and c/cu ÷ 0
Then V
v ÷ 0 can be satisIied by seeking v oI the Iorm:
v ÷ V× f (r,:)e

06703 59 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
or
, r :
r
u
¢ (
= V×
(
¸ ¸
v e
The second expression usually leads to somewhat simpler expressions Ior V×v and is the one
used in the table on p131 oI BS&L:
v
r :
v
r r
r :
= ÷ =
1 1 c¢
c
c¢
c
Computing the curl in cylindrical coordinates and setting it equal to zero leads to the Iollowing
PDE in ¢ (the details are leIt as an exercise):
V × = ÷
= ÷ ÷ ÷ = v e 0
E
r
E
2
2
0
¢
¢
u
where E
r
r r
:
2
2
2
2
2
1
¢
c ¢
c
c¢
c
c ¢
c
= ÷ +
Note that E
2
¢ = V
2
¢ :
2 2
2
2 2
1
, r :
r r
r :
c ¢ c¢ c ¢
V ¢ = + +
c
c c
AXISYMMETRIC FLOW (SPHERICAL)
In spherical coordinates (r,u,), axisymmetric Ilow means
v ÷ v
r
(r,u)e
r
¹ v
(r,u)e
or v
÷ 0 and c/c ÷ 0
where  is the azimuthal angle. Then V
v÷0 can be satisIied by seeking v oI the Iorm:
v ÷ V×¢(r,u)e

or v e e e = V ×
,
= ÷
¢ u
u
u
c¢
cu u
c¢
c
 u
r
r
r
r r
v
r
v
r
,
sin
sin
sin
1 1
2
Again, the second expression is the one used in the table on p131 oI BS&L (except the signs are
reversed). Taking the curl (HWK #4, Prob. 6a):
V × = ÷ = v
e
0

u
¢
r
E
sin
2
06703 60 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
which requires E
2
¢ ÷ 0
where E
r r
2
2
2 2
1
¢
c ¢
c
u c
cu u
c¢
cu
= +
sin
sin
Note that E
2
¢ = V
2
¢:
2
2
2 2
2 1
, sin
sin
r
r r
r r
c ¢ c¢ c c¢
 
V ¢ u = + + u

c cu cu
\ . c u
ORTHOGONALITY OF ÷CONST AND ÷CONST
A contour on which ¢÷const is called a streamline.
A contour on which ÷const is called a isopotential
line. It turns out that these two contours are orthogonal
at any point in the Iluid. To see this Iirst note that
v ÷ V
Erom the geometric meaning oI gradient (see page 6 oI
Notes), we know that V and hence v is normal to a
÷const surIace (see Iigure at right). Second, recall that
v can also be written in terms oI streamIunction as
Ior R.C.C.S. v ÷ V¢×k (29)
Recall that the cross product is a vector which is
orthogonal to two vectors being multiplied. Thus v,
V¢, and k are mutually orthogonal. Since v and V
point in the same direction, V and V¢ must also be
orthogonal.
STREAMLINES, PATHLINES AND STREAKLINES
Streamline  a contour in the Iluid whose tangent is everywhere parallel to v at a given instant oI
time.
Path Line  trajectory swept out by a Iluid element.
Streak Line  a contour on which lie all Iluid elements which earlier past through a given point in
space (e.g. dye trace)
Eor steady Ilows, these three deIinitions describe the same contour but, more generally, they are
diIIerent.
Potential Elow Around Sphere. Lines lie
in plane of page, which could be any
÷const. plane.
06703 61 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
PHYSICAL MEANING OF STREAMFUNCTION
The precise meaning oI streamIunction is
somewhat diIIerent Ior 2D and axisymmetric Ilows.
Let's Iocus on 2D Ilows normal to a cylinder (not
necessarily with circular crosssection, axis
corresponds to :axis in R.C.C.S.). To motivate the
somewhat lengthy analysis which Iollows, we Iirst
state the physical meaning. Eirst we observe that
material points Iollow trajectories which can be
described as ¢÷const. Three such trajectories are
shown at right which lie in the xvplane. When these
trajectories are (mathematically) translated along the
:axis a distance L they sweepout ¢÷const surIaces.
No Iluid crosses these surIaces: there are like the walls oI a tube. Since no Iluid leaves or enters
this 'tube¨, conservation oI mass means the mass Ilowrate must be a constant at any point along
the tube. Eor an incompressible Iluid, the volumentric Ilowrate is also constant. Suppose we
denote the volumetric Ilowrate between any two oI these ¢÷const surIaces as AQ; then it turns
out that
AQ
L
= ÷ ¢ ¢
2 1
Thus ¢ might be interpreted as the volumetric Ilowrate per unit length between this particular
streaming surIace and the one
corresponding to ¢÷0.
Now let`s show this. Consider
an arbitrary open contour (C) in the
xvplane, cutting across the Ilow.
Next, consider the surIace (A)
Iormed by translating this contour a
distance L parallel to the :axis. The
volumetric Ilowrate across A is:
Q ÷ )
A
n
vda
where n is a unit normal to A. Since nothing changes with :, we choose a short segment oI the
contour, having length ds and oI length L as our diIIerential area element.
da ÷ L ds,
The Ilowrate becomes:
?
C C
Q
ds d
L
= = V¢
n v r
(30)
06703 62 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
where n is now normal to C and lies in the xvplane. The key to this prooI is deriving the second
equality in the expression above.
Eirst, v can be expressed in terms oI the gradient oI the streamIunction using (27):
v ÷ V¢×k (27)
Both v and V¢ lie in the plane oI the page, whereas k is
perpendicular to this plane (points out oI page). Postcrossing
V¢ by the unit vector k does not change the magnitude but
rotates V¢ by 90° clockwise. II instead, we precrossed by k
we would rotate V¢ by 90° counterclockwise. In either case,
the cross product oI k and V¢ is a vector lying in the plane oI
the page and oI the same magnitude as V¢.
The other term in the integrand oI
(30) is n ds, where ds is the
magnitude oI a diIIerential
displacement along the contour,
which we will call dr:
ds d = r
Since n is a unit vector n ds has the same magnitude as dr but is
rotated by 90°. Both n ds and dr lie in the plane oI the page. Just as
in (27), we can rotate one vector into the other by crossing with k:
d ds × = r k n (31)
Substituting (27) and (31) into (30) yields
ds ds d d d = = V¢× × = V¢ = ¢ n v v n k r k r
(32)
The 3
rd
equality above says that the dot product oI the two rotated vectors is the same as the dot
product oI the two vectors without rotation (since they are both rotated by the same amount).
(32) into (30) yields:
P Q P Q
C C
Q
ds d Q P
L
= = ¢ = ¢ ÷ ¢
n v
06703 63 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
To extract the physical meaning oI this results, consider
two contours, denoted as C
1
and C
2
in the Iigure at
right. Notice that iI C
1
coincides with a streamline, the
velocity is parallel to the contour at every point any no
Iluid crosses it:
then: Q/L ÷ 0
and ¢(Q) ÷ ¢(P) ÷ ¢
2
Thus ¢ ÷ const. along a streamline
On the other hand, iI the contour cuts across two
streamlines (see contour C
2
in Iigure at right), then the diIIerence in value oI ¢ corresponding to
two diIIerent streamlines is just the volumetric Ilowrate oI Iluid held between the two streamlines
(per unit length in the :direction):
A¢ ÷ AQ/L
INCOMPRESSIBLE FLUIDS
By 'incompressible Iluid¨ we are usually reIerring to the assumption that the Iluid`s density
is not a signiIicant Iunction oI time or oI position. In other words,
cµ
c
µ
t
+ V =
v 0
can be replaced by V
v ÷ 0
Eor steady Ilows, cµ/ct ÷ 0 already and the main Iurther requirement is that density gradients be
negligible
V = V + Vµ ~ V µ µ µ v v v v
Identity C.1
Since Ilow causes the pressure to change, we might expect the Iluid density to change at least
Ior gases. As we shall see shortly, gases as well as liquids can be treated as incompressible Ior
some kinds oI Ilow problems. Conversely, in other Ilow problems, neither gas nor liquid can be
treated as incompressible. So what is the real criteria?
Eor an ideal Iluid (i.e. no viscous dissipation to cause VT), density variations come about
primarily because oI pressure variations. Eor an isentropic expansion, the compressibility oI the
Iluid turns out to be:
06703 64 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
cµ
cp
c
S
=
1
2
where c ÷ speed oI sound in the Iluid
Thus changes in density caused by changes in pressure can be estimated as
Aµ A ~
1
2
c
p (33)
According to Bernoulli`s equation, pressure changes Ior steady Ilow are related to velocity
changes:
p v
µ
+ =
2
2
const. or
2 1
2
p v A = ÷ µA (34)
(34) into (33):
2
2
2
v
c
µ A
Aµ ~ ÷
The largest change in density corresponds to the largest change in v
2
, which is v
max
2
 0:
2
max
max
1
2
v
c
 
Aµ  
=
 
µ
\ .
\ .
II the Iraction change in density is small enough, then it can be neglected:
Criteria 1: v c
max
<<
Ior air at sea level: c ÷ 342 m/s ÷ 700 mph
Ior distilled water at 25°C: c ÷ 1500 m/s ÷ 3400 mph
Eor unsteady Ilows, a second criteria must be met:
Criteria 2: t >>
l
c
where t ÷ time over which signiIicant changes in v occur
l ÷ distance over which changes in v occur
l/c ÷ time Ior sound to propagate a distance l
Eor steady Ilow t ÷ · and Criteria 2 is always satisIied.
06703 65 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Any Iluid can be considered incompressible iI both criteria are met.
Viscous Fluids
To resolve d`Alembert's paradox, we need to introduce another Iorce into our Iorce balance.
This Iorce can be thought oI as resulting Irom Iriction between the Iluid elements. Eriction gives
rise to viscous heating which represent an irreversible conversion oI mechanical energy into
heat. Indeed, Iriction is the main diIIerence between an ideal Iluid and a real Iluid.
Iriction ÷ irreversible deIormation ÷ nonideal Ilow
TENSORIAL NATURE OF SURFACE FORCES
Eriction is a surIace Iorce like hydrostatic pressure, but unlike pressure, Iriction is not
isotropic:
isotropic  independent oI orientation (direction)
We say that pressure is isotropic because the magnitude oI the pressure Iorce is independent oI
the orientation oI the surIace on which it acts. Recall:
dF
p
÷ npda
,dF
p
, ÷ pda
which is independent oI n (orientation). Viscous Iriction
does not have this property:
,dF
f
, depends on n
but it's magnitude is proportional to da:
dF
f
 · da
Thus it makes sense to talk about the Iorce per unit area, which we usually call pressure. In the
more general case in which the magnitude oI the Iorce per unit area might depend on the
orientation oI the surIace, we use the name stress. In general, all surIace Iorces can be lumped
together and written as
dF
surf
÷ t(n) da
06703 66 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
where t(n) denotes the surIace Iorce per unit area acting on the body through the surIace element
whose orientation is given by the unit outward normal n. We call this quantity the stress vector.
*
In what Iollows, I will try to convince you that the eIIect oI orientation oI the surIace can be
expressed as
t(n) ÷ n
T
where T is called the stress tensor. While T depends on position and possibly time, it does not
depend on the orientation oI the diIIerential surIace element da, which is given by n. T is
sometimes call the 'state oI stress¨ oI the Iluid.
Eirst, let`s try to understand better what we mean when we say a
material experiences some 'stress¨. Consider a block oI some
material which bears some externally applied equilibrium load (see
Iigure at right). By 'equilibrium¨ I mean there is no net Iorce and no
net torque applied to the body: thus it is at mechanical equilibrium
and has no tendency to accelerate. The material might be a Iluid or a
solid, but what we are about to say is easier to imagine iI we
think oI material as a solid block.
Now consider some mathematical surIace inside the material
(indicated by the dotted line in the Iigure above). What are the
Iorces exerted across this mathematical surIace? To answer
this question, suppose we actually separated the block into two
pieces by physically cutting along this surIace without
changing the loading. Block '1¨ now experiences an
unbalanced load F while block '2¨ experiences an unbalanced
load ¹F. Once separated, both blocks would tend to accelerate
in opposite directions.
Why don`t the two halves accelerate when connected? Apparently, halI '2¨ must have exerted a
Iorce on halI '1¨ which we denote as t(n
1
)A, where A is the area oI the cut Iace, and which
equals
t(n
1
)A ÷ ¹F
while halI '1¨ exerted an equal but opposite Iorce on halI '2¨
t(n
2
)A ÷ F ÷ t(n
1
)A (35)
When expressed per unit area, this internal Iorce between the two halves (when they are
physically connected) is what we mean by the 'stress¨ experienced by the material under load.
Given that n
1
÷ n
2
, (35) tells us that
*
see Whitaker, Introduction to Fluid Mechanics, Chapt. 4.
06703 67 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
t n t n
2 2
= ÷ ÷
More generally, Ior any mathematically surIace having orientation n, we can write t n t n = ÷ ÷
or t n t n ÷ = ÷ + (36)
In essence, this is just a statement oI Newton`s Third Law (Ior
every action, there is an equal but opposite reaction).
Now let`s generalize to three dimensions. Suppose we know
the distribution oI stress (i.e. the 'loading¨) on all six Iaces oI a
block oI material. Eurthermore, suppose this loading is an
equilibrium loading (no net Iorce and no net torque on the block).
Let`s try to calculate the stress on a mathematical surIace cutting
through the material at an arbitrary angle. Let the orientation oI
the mathematical surIace be given by the unit normal n.
Problem: given the surIace stress on mutually perpendicular
planes i.e. given t(i) Ior x÷const plane, t(j) Ior v÷const plane, and
t(k) Ior :÷const plane, calculate the surIace stress on a plane oI
arbitrary orientation speciIied by the unit normal n.
Given: t(i), t(j), and t(k)
Eind: t(n)
Solution: We choose the tetrahedron ABCO as our 'system.¨ Eor the surIace Iorces to be
balanced
*
t n 0 da
A
=
We evaluate this surIace integral by subdividing the surIace A into the Iour Iaces oI the
tetrahedron: planes ABC, BCO, AOC, and ABO. Eor each surIace, we need to evaluate the
outward normal n, the stress vector t and the surIace area A. The results are tabulated below:
*
More generally, in our Iorce balance, we should include body Iorces and inertia as well as
surIaces Iorces. II we let J÷0 then the volume integrals Ior body Iorces and inertia vanish more
rapidly than surIace Iorces. Thus Ior the surIace A enclosing a tiny volume J, we must still
require t 0 da
A
= Ior the surIace oI every diIIerential volume element.
06703 68 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Outward Stress
Plane Normal Area Vector
ABC n A t(n)
BCO i n
iA t(i)
AOC j n
jA t(j)
ABO k n
kA t(k)
Applying the Mean Value Theorem to our surIace integral:
t 0 t n n i t i n j t j n k t k da A
A
= = + ÷ + ÷ + ÷
Dividing by A cancels the A out oI our expression Ior the integral. Taking the limit as all the
dimensions oI the block vanish (i.e. as A÷0) allows us to replace the unknown averages with
their limit, which is the value oI the vector at the point that the tetrahedron collapses about.
t(n) ¹ (n
i)t(i) ¹ (n
j)t(j) ¹ (n
k)t(k) ÷ 0
or t(n) ÷ (n
i)t(i)  (n
j)t(j)  (n
k)t(k) (37)
Next we apply (36). OI course, we can replace n in (36) with i or j or k. Using (36) in (37):
t(n) ÷ (n
i)t(i) ¹ (n
j)t(j) ¹ (n
k)t(k)
Recalling the deIinition oI dyadic product, we could rewrite this expression as
t(n) ÷ n
it(i) ¹ n
jt(j) ¹ n
kt(k)
t(n) ÷ n
it(i) ¹ jt(j) ¹ kt(k) ÷ n
T
The sum (inside square brackets) oI these three dyadic products is some secondrank tensor,
which is independent oI n. We denote this tensor by T.
SigniIicance: to calculate the surIace Iorce on some diIIerential surIace area da having
orientation n, we just multiply this expression Ior the stress by the area:
d da
surf
F n T =
(38)
where T is called the stress tensor and dF
surf
represents the net surIace Iorce (all contributions)
acting on the body whose outward normal is n. While T does not depend on n, it might depend
on position inside a solid which is nonuniIormly stressed; thus
T ÷ T(r)
represents the state of stress oI the Iluid. In terms oI the components oI the tensor, we say that
06703 69 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
T
if
÷ f
th
component oI the Iorce acting on a r
i
÷ const surIace
To make this more understandable, it might help to express T in terms oI other variables in a
Iamiliar problem, say an ideal Iluid.
Eor an ideal Iluid: T ÷ pI
To show that this is correct, we will compute the surIace Iorce and show that it reduces to the
Iamiliar Iorce due to pressure:
dF
surf
÷ n
Tda ÷ n
(pI)da ÷ p(n
I)da ÷ pnda ÷ dF
p
GENERALIZATION OF EULER'S EQUATION
Recall that Euler`s equation was derived by applying Newton's Second Law (F÷ma) to any
Iluid element. To generalize this result to include Iriction, we replace the 'pressure Iorce¨ by the
surIace Iorce.
Instead oI: µ µ
D
Dt
dJ dJ pda
J
g p
J A
v
F F g n
= + = ÷
we have: µ µ
D
Dt
dJ dJ da
J
g surf
J A
v
F F g n T
= + = +
Applying the Divergence Theorem and combining the three volume integrals:
µ µ
D
Dt
dJ
J
v
g T 0 ÷ ÷ V
=
Since this must result Ior all choices oI J in some region, the integrand must vanish at every
point in that region, or:
D
Dt
v
g T = + V
1
µ
(39)
which is a more generalized version oI Euler's Equation.
Eor an ideal Iluid: T ÷ pI
and
identity
C.11
p p V = ÷V = ÷ V T I
06703 70 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Eor a prooI oI identity C.11, see Hwk #3, Prob. 4. Eor a real Iluid, there is an additional
contribution to T Irom Iriction:
Eor a real Iluid: T ÷ pI ¹ t tt t (40)
where t tt t is called the viscous stress or the deviatoric stress (since it represents the deviation Irom
ideal). Note that:
CAUTION: t tt t in these notes (and in Whitaker) ÷ t tt t in BS&L
By convention,
if
in these notes is positive Ior tensile stresses (which result Irom stretching a
solid rod) whereas
if
in BS&L is positive Ior compressive stresses (which result Irom putting a
Iluid under pressure). Although BS&L's notation might make more sense Ior Iluids (which
usually do not experience tensile stresses), we will use the other convention because it is more
commonly used in continuum mechanics: in particular, this is the convention used by Whitaker.
When written in terms oI scalar components, the above equation between T and t tt t represents
9 equations in 10 unknowns (the nine components oI t tt t plus p). Clearly p and t tt t cannot yet be
uniquely determined, given the state oI stress T. One additional relationship is required. Eor a
real Iluid, we somewhat arbitrarily deIine p as the average oI the three diagonal components oI T:
p ÷ ÷
1
3
T I
(41)
where
ik
fk
if i f k k if i k f k kk
i f k i f k k
T T T
o
o
    = = =


\ .
\ .
T I e e e e e e e e
or, in Cartesian coords.: TI ÷ T
xx
¹T
vv
¹T
::
In any coordinate system, TI is called the trace oI T. Then p can be thought oI as the isotropic
contribution to stress (that part oI the normal stress which acts equally in all directions) while t tt t
represents the remainder, or the nonisotropic part. (40) might be regarded as decomposing t tt t into
an isotropic part and a nonisotropic part, which is a common thing to do with tensors.
The choice oI p made in (41) also represents the thermodynamic pressure; i.e. the
hydrodynamic p is now the same as the p appearing in the thermodynamic equation of state:
µ ÷ µ(p,T)
To summarize, we decompose the state oI stress into two contributions: an isotropic pressure:
p = ÷
1
3
T I
and a deviatoric stress: t tt t ÷ T  (1/3)(TI)I
Taking the divergence oI (40): V
T ÷ Vp ¹ V
t tt t
06703 71 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
and substituting it into Euler's Eq. (39):
µ µ t
D
Dt
p
v
g = ÷ V + V
new for
real fluids
¸
MOMENTUM FLUX
Up to this point we have spoken oI T as the stress on the Iluid. T can also be thought oI as
momentum flux. To convince yourselI that momentum is being transported (like heat and
mass), consider the problem oI unsteady simple shear flow. At time t÷0, an initially quiescent
Iluid conIined between inIinite parallel plates is disturbed by imposing motion on the upper
plate. The velocity proIile gradually develops into
linear shear Ilow.
Note that, like all other Iluid elements, the Iluid
element at v÷h/2 undergoes acceleration:
cv
x
/ct ~ 0 at v÷h/2
In other words, the Iluid element is gaining
momentum. How can it acquire momentum?
Answer: momentum is transported to v÷h/2 Irom
above through Iriction between Iluid elements.
There are two directions associated with transport
oI momentum:
1) direction oI the momentum being
transported (in this example, xmomentum
is transported)
2) direction in which the momentum is
transported (v)
Eor this reason momentum Ilux must be a 2nd rank tensor. It turns out that:
T ÷ diIIusive Ilux oI momentum
*
µvv ÷ convective Ilux oI momentum
*
"DiIIusive" means it results Irom random collisions between molecules.
06703 72 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
In multicomponent systems, the Ilux oI any species i due to convection is written as the product
oI concentration and velocity:
conv. Ilux oI species i ÷ c
i
v ÷ molcm
2
s
1
÷ rate/area
where c
i
÷ mol/cm
3
More generally, by 'Ilux,¨ I mean:
flux that tensor (oI whatever rank), which when predotted by nda, gives the rate oI
transport through the surIace element having area da in the direction oI its unit
normal n.
Eor example, the Ilux oI Iluid mass by convection is
µv ÷ gcm
2
s
1
÷ rate/area
ProoI: predotting by nda we obtain:
(nda)
(µv) ÷ µn
vda ÷ µ(dq) ÷ g/s
which represents the mass Ilowrate across da. Indeed the convective Ilux oI total Iluid mass also
has the Iorm oI concentration times velocity since
µ ÷ concentration oI mass ÷ mass/vol.
It might come as less oI a surprize that µvv is the convective Ilux oI momentum iI you realize
that
µv ÷ conc. oI momentum ÷ momentum/vol.
AIter all, momentum is mass times velocity. So
(µv)v ÷ µvv ÷ conv. Ilux oI momentum
Example: Apply conservation oI linear momentum to an arbitrary Iluid system. Thus prove that
T and µvv are momentum Iluxes, as claimed.
Solution: We choose a system that has Iixed boundaries in the laboratory reIerence Irame: in
other words, J = J(t). ReIerring to the discussion above in which µv is the concentration oI
momentum, µvv is the momentum Ilux by convection, and T is the momentum Ilux by
diIIusion, conservation oI momentum requires
d
dt
dJ da da dJ
J A A J
µ µ µ v n vv n T g
= ÷ ÷ ÷ +
accumulation in by convection in by diffusion in by external forces
06703 73 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Applying the divergence theorem, combining the resulting volume integrals, and invoking the
result Ior arbitrary J, we would obtain
c
c
µ µ µ
cµ
c
µ
c
c
µ µ
t
t t
v vv T g 0
v
v
v v v v
+
V + V
+ V ÷ V ÷ =
Expanding the partial timederivative oI the product and expanding the divergence oI the dyadic
product using identity C.8
cµ
c
µ µ
c
c
µ µ
t t
+V
,
+ + V ÷ V ÷ =
v v
v
v v T g 0
0
Einally, we recognize the Iactor inside square brackets must vanish according to the general
continuity equation. AIter this term is dropped, the remainder is Euler's equation:
µ
c
c
µ µ
v
v v T g
t
+ V = V +
Comment: In writing the statement oI conservation oI momentum, we have a term representing
transport oI momentum into the system by the action oI external Iorces. Clearly, an object Ialling
Irom rest in a gravitational Iield acquires momentum through the action oI gravity. Does this
acquisition represent transport to the body Irom outside, or does it represent a "generation" term?
II it is spontaneous momentum generation, then momentum is not conserved.
With a little reIlection, we can convince ourselves that the action oI gravity is 'transport¨ and not
'generation.¨ Some oI the earth's momentum is being transported to the Ialling object. When the
object eventually collides with the earth and comes to rest again, that momentum is transported
back to the earth. The total momentum oI the universe has not changed: momentum is
conserved.
Actually, the term we call 'diIIusion oI momentum¨ also arises Irom the action oI an external
Iorce: this time it's the action oI 'surIace¨ Iorces, rather than 'body¨ Iorces.
RESPONSE OF ELASTIC SOLIDS TO NORMAL (UNIAXIAL) STRESS
By generalizing, the Iorce balance plus the mass balance now contain more unknowns than
equations:
t tt t, v, and p ÷ 9¹3¹1 ÷ 13 unknowns
Euler ¹ Continuity ÷ 3¹1 ÷ 4 equations
06703 74 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Missing is the constitutive equation which is an empirical description oI how the Iluid or solid
material responds to stress. Obtaining this relationship is an important objective oI that Iield oI
science known as continuum mechanics.
Let's start by considering solids, whose response is more Iamiliar. Suppose I try to stretch a
rectangular bar by applying a tensile Iorce,
F.
We will attempt to describe the response oI
the bar under conditions oI mechanical
equilibrium. To have the bar stretch instead
oI accelerating as a result oI the Iorce, I
must apply an equal but opposite Iorce to
the other end. Let the xaxis be aligned with the direction oI the applied Iorce:
F
x
÷ ,F,
At equilibrium, the length oI the bar will increase by an amount o
x
. Hooke's law tell us that:
o
x
x x
v :
F L
L L
·
but
F
L L
T
x
v :
xx
=
is the applied stress. The two subscripts on stress denote the two directions associated with it:
the Iirst subscript denotes the orientation oI the surIace the Iorce is applied to (x÷const) while the
second denotes the direction oI the applied Iorce (x).
Since the deIormation is proportional to the original length oI the bar, it makes sense to deIine
the deIormation per unit length, which is called:
strain: c
o
xx
x
x
L
=
Later we will see that strain is a secondrank tensor. Eor now, we will interpret the order oI the
subscripts as Iollows: the Iirst subscript gives surIace (in the
above case, an x÷const surIace), while the second subscript gives
the direction oI the deIormation (in the above case, the
xdirection).
Experiments show that the strain increases with the applied stress
as shown in the Iigure at right. At low stress levels, the strain is
06703 75 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
directly proportional to the applied stress: Hooke's law
*
Ior purely elastic solids is:
T
xx
÷ Ec
xx
E ÷ Iorce/area
where E is called the Young's Modulus. It turns out that deIormation also occurs in the v and :
directions:
c
vv
÷ c
::
÷ vc
xx
where v is called Poisson's ratio. Eor most materials
0.28 · v · 0.33
RESPONSE OF ELASTIC SOLIDS TO SHEAR STRESS
Instead oI a normal Iorce, suppose I apply a shear Iorce on the upper Iace. To keep the object
Irom translating, I must apply an equal but opposite Iorce on the
lower Iace.
This generates a Iorce couple or torque, which will cause the body
to spin. To prevent a steady increase in rotation speed, I must
apply an equal but opposite torque. Recalling that torque is Iorce
times lever arm:
F
x
L
v
÷ F
v
L
x
Dividing by L
x
L
v
L
:
:
F
L L
F
L L
x
x :
v
v :
=
or T
vx
÷ T
xv
where T
vx
÷ xcomponent oI Iorce/area acting on v÷const surIace. Note that this implies that the
stress tensor is symmetric. This turns out to be true Ior virtually all
loadings.
*
This loading produces a shear deIormation in which
o
x
· L
v
*
Robert Hooke (16351703). English experimental physicist. Hooke`s law Iirst stated in 1660.
*
Eor a prooI based on the assumption oI local equilibrium, see W:4.3
06703 76 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
and o
v
· L
x
The corresponding deIinition oI shear strain is:
1
2
o
o
c c
x
v
v
x
xv vx
L L
+
= =
Since we applied a symmetric stress, by applying Iorces in both the x and v directions, we
average the strains in the x and v directions to obtain a symmetric strain. Moreover, this
deIinition yields a strain which is invariant to rotation oI the xv axes. As with normal stress,
shear stress produces a shear strain in direct proportion to the stress:
T
xv
÷ 2Gc
xv
(pure shear)
where G is call the modulus of elasticity for pure strain. Although the units are the same, the
value oI 2G is diIIerent Irom that oI E.
GENERALIZED HOOKE'S LAW
Now let's try to generalize to some arbitrary loading which might involve both shear and
normal stresses.
Consider two arbitrary material points
in the material. Let x denote the
position oI the second material point
relative to the Iirst, beIore the load is
applied. AIter the load is applied,
both material points move as shown
by the dashed lines in the Iigure at
right. AIter the load the position oI
the second material point relative to
the Iirst becomes x¹o oo o, where
o oo o(x) ÷ relative displacement
oI two material points initially located at x. The resulting strain can be generalized as:
c o o = V + V
1
2
t
or c
co
c
co
c
if
f
i
i
f
x x
= +
1
2
06703 77 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Note that this general deIinition oI strain reduces to earlier expression Ior strain in the cases oI
pure normal or pure shear stress:
unixial normal stress: c
co
c
co
c
o
xx
x x x
x
x x L
= +
=
1
2
and T
xx
÷ E c
xx
pure shear: c
co
c
co
c
o
o
xv
v
x
v
x
x
v
x v L L
= +
= +
1
2
1
2
and T
xv
÷ 2G c
xv
II the strain components are all small, we might reasonably suppose that Hooke's law generalizes
into a linear relationship between any component oI strain and the nine components oI stress (or
vice versa):
T C
if
k l
ifkl kl
=
= =
1
3
1
3
c (42)
There are then nine coeIIicients Ior each component oI stress, making a total oI 81 possible
coeIIicients. But just by requiring:
 symmetry oI T (T
if
÷T
fi
)
 isotropy oI material (e.g. same Young's modulus applies to uniaxial stress in x
direction as Ior v and :)
it can be shown that the number oI independent constants is reduced Irom 81 to only 2. It is
customary to express the stress tensor as the sum oI two tensors: one isotropic (cI)I and the
remainder c  (cI)I. Denote the two independent constants as k
1
and k
2
and use them as
weighting Iactors in the sum:
1 2
isotropic
remainder
k k
(
= c ÷ c + c
¸ ¸
T I I I I
This can be rearranged T I I = + ÷ k k k
1 2 1
c c
Sometimes the constants are choosen as the two coeIIicients in this new Iorm
T I I = + 2qc ì c
which represents the generalization oI Hooke's law oI elasticity. Thus oI G, E and , only two
are linearly independent properties oI the material; using the above relationship we can show
(HWK Set #5, Prob. 1) that
06703 78 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
v = ÷
E
G 2
1
RESPONSE OF A VISCOUS FLUID TO PURE SHEAR
Suppose I have a thin layer oI Iluid held between parallel plates and I apply a Iorce F
x
to the
upper plate. To cause deIormation, rather than
simple translation oI the system, I must apply an
equal but opposite Iorce to the lower plate.
As beIore, this produces a Iorce couple or torque.
To keep the system Irom rotating, I must apply an
opposing torque. Once again this equilibrium
state corresponds to a symmetric stress tensor.
Erom our experience, we know that eventually the
upper plate will slide past the lower plate at some
steady relative speed U
x
.
response: U
F L
L L
x
x v
x :
·
This speed represents a rate of deformation:
U
d
dt
x
x
=
o
and the speed per unit thickness represents a rate of strain:
U
L
d
dt
L
d
dt L
d
dt
x
v
x
v
x
v
vx
= =
=
o
o
c
Einally, the applied Iorce divided by the area over which it's applied represents a shear stress:
F
L L
T
x
x :
vx vx xv
= = = t t
Rewriting the proportionality as an equality, we have:
t µ
c
vx
vx
d
dt
=
which is called Aewton's Law of Jiscosity (1686). Alternatively, BSL point out that
L
06703 79 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
U
L
dv
dv
x
v
x
= and t t µ
xv xv
x
dv
dv
BSL notes
= ÷ = ÷
GENERALIZED NEWTON'S LAW OF VISCOSITY
This generalization oI Newton's law oI viscosity to arbitrary loading parallels that oI Hooke's
Law with the strain tensor replaced by the rate oI strain. The main diIIerence is that now the
deIormation is a Iunction oI time:
o oo o(x,t) ÷ displacement oI material pt.
÷ deIormation
The trajectory oI a material point initially located
at x(0) is given by:
xx(0),t ÷ x(0) ¹ o oo ox(0),t
Keeping the material point Iixed (i.e. x(0) is
constant) while we take the time derivative is the same as taking the material derivative oI
position. Thus the rate oI deIormation oI Iluid elements is:
d
dt
D
Dt
o
= =
x
v (rate oI deIormation)
given by the Iluid velocity. Just like we deIine the gradient oI the deIormation to be the strain:
c o o = V + V
1
2
t
, (strain)
the gradient oI the rate oI deIormation must be the rate of strain:
d v v = = V + V = V + V
,
= V + V
D
Dt
D
Dt
D
Dt
D
Dt
t
t
t
c
o o
o o
1
2
1
2
1
2
(rate oI strain)
Newton`s law oI viscosity is a linear relationship between the stress and the rate oI strain.
Generalizing that linear relationship again leads to a relationship like (42), except that the strain
tensor c
kl
is replaced by the rate oI strain tensor d
kl
. Once again, in order Ior the stress tensor T
or t tt t to be symmetric and Ior the material to be isotropic, only two oI the 81 coeIIicients C
ifkl
are
independent. We use the two independent coeIIicients to multiply the isotropic part oI t tt t and its
remainder:
t µ k µ = + ÷ 2
2
3
d d I I
,
06703 80 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
where µ is the usual viscosity and k is called the second coefficient of viscosity. dI has a
special signiIicance, which we will now point out. As we showed on page 70, dI is the trace oI
d:
dI ÷ d
11
¹ d
22
¹ d
33
Writing in terms oI the velocity v: d
v
x
v
x
if
i
f
f
i
= +
1
2
c
c
c
c
d
v
x
v
x
v
x
11
1
2
1
1
1
1
1
1
= +
=
c
c
c
c
c
c
so d I v = + + = V
c
c
c
c
c
c
v
x
v
x
v
x
1
1
2
2
3
3
So the trace oI the rate oI deIormation tensor is just the divergence oI the velocity. Aewton's law
of viscosity becomes:
t µ k µ = + ÷ V 2
2
3
d v I
,
Eor an incompressible Iluid: V
v ÷ 0
leaving 2
t
(
t = µ = µ V + V
(
¸ ¸
d v v (incompressible)
NAVIERSTOKES EQUATION
Once again, let`s go back to Euler's equation, generalized to account Ior the tensorial nature
oI viscous Iriction:
µDv/Dt ÷ µg  Vp ¹ V
t tt t
Eor an incompressible Iluid, V
v÷0 and
t tt t ÷ 2µd ÷ µVv¹(Vv)
t

V
t tt t ÷ µV
Vv ¹ V
(Vv)
t

but using identity C.10: V
(Vv)
t
÷ V(V
v) ÷ 0
leaving V
t tt t ÷ µV
Vv ÷ µV
2
v
06703 81 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Euler's equation becomes: µ µ µV
D
Dt
p
v
g v = ÷ V +
2
(µ,µ÷const)
which is known as the AavierStokes Equation (1822).
+
Now we have as many equations as
unknowns:
v, p ÷ 4 unknowns
NSE, Continuity ÷ 4 eqns
BOUNDARY CONDITIONS
But to successIully model a Ilow problem, we need more
than a suIIicient number oI diIIerential equations. We also
need boundary conditions.
A typical boundary is the interIace between two immiscible
phases  either two Iluids or a Iluid and a solid. One such
boundary condition which can generally be applied is the no
slip condition:
v
I
÷ v
II
Eor example, in the problem oI uniIorm Ilow around a stationary solid
sphere, this requires:
at r÷a v
f
÷ v
s
÷ 0
which means that there is no Ilow oI Iluid across the boundary:
at r÷a v
r
÷ 0
We also assumed this boundary condition when we solved the
potential Ilow problem. But 'no slip¨ also means:
at r÷a v
÷ 0
Note that our potential Ilow solution did not satisIy this second equation:
+
Sir George Gabriel Stokes (18191903): British (Irish born) mathematician and physicist,
known Ior his study oI hydrodynamics. Lucasian proIessor oI mathematics at Cambridge
University 18491903 (longestserving Lucasian proIessor); president oI Royal Society (1885
1890).
06703 82 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Ior pot`l Ilow: v
r
(a,u) ÷ 0
v
(a,u) ÷ (3/2)Usinu = 0
In addition to d'Alembert's paradox, potential Ilow Iails to satisIy the noslip condition.
*
Eor a Iluidsolid interIace, in which the velocity oI the solid phase is known, the noslip
condition is suIIicient. But in the case oI IluidIluid interIace, the velocity oI the second Iluid is
usually unknown. Then noslip just relates one unknown to another.
A second boundary condition can be obtained by considering the
stresses acting on the material on either side oI the interIace.
Suppose we were to apply a loading as on page 66 to a twophase
region straddling the interIace, as suggested in the Iigure at right.
Note that the loading is balanced: that is there is no net Iorce on the
system.
II we were to split the system into two parts along the
interIace, each oI the two halves would tend to accelerate.
This suggests that, when the two phases are in contact,
each exerts an 'internal¨ Iorce on the other, as shown in
the Iigure at right. These Iorces are equal but opposite:
F
1on2
÷  F
2on1
÷ F
which might be thought oI as 'Newton`s Third Law¨: Ior
every action, there is an equal but opposite reaction. Using (38) (on page 68) to express the
Iorces in terms oI the stress tensor:
A A = ÷ = n T n T F
where A is the area oI the interIace and where n
i
is the normal to the interIace pointing out oI
phase i. However, Irom the geometry, we can deduce that
n n n
1 2
= ÷ ÷
Newton`s Third Law becomes: n T n T
1 2
= (43)
When the interIace is highly curved (e.g. a small oil droplet in water), then surIace tension can
produce a discontinuity in the normal components oI the above Iorce, which has not been
*
Although "no slip" is usually applicable, there are at least two situations where no slip might
not be applicable: 1) when the meanIree path oI gas molecules is comparable to the geometric
dimension, and 2) when a liquid does not wet the solid.
06703 83 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
included in the above analysis see L&L, Chapt. 7 or Hunter, Vol I., p237I. The more general
Iorm oI this boundary conditon is
n
(T
1
 T
2
) ÷ ¸(V
s
n)n (44)
where ¸ is another property oI the Iluid called the surface tension and
V
s
÷ (I  nn)
V
is the surIace component oI the V operator. We will have more to say about this
near the end oI the course. Eor now, we will neglect surIace tension eIIects.
V
s
n ÷ curvature oI surIace ÷ m
1
Eor a Ilat surIace, n is independent oI position along the surIace so that V
s
n ÷ 0
and (44) reduces to (43).
Exact Solutions of NS Equations
Exact solution oI the NavierStokes equations presents a Iormidible mathematical problem.
By 'exact¨ I mean:
exact neither viscous nor inertial terms are neglected (i.e. approximated by zero, as
opposed to being identically zero)
One diIIiculty is the nonlinear inertial term. Most oI the powerIul mathematical techniques
(such as eigenIunction expansions, used in 'separation oI variables¨) only work when the P.D.E.
is linear.
OI course, iI we can neglect the higherorder viscous terms, then we can cope with the non
linearity using a velocity potential, as we did earlier in solving problems in potential Ilow.
However, viscous terms are seldom completely negligible and leaving them in the equation
makes the problem much more diIIicult by increasing the order oI the P.D.E. Nevertheless, it is
possible to Iind exact solutions in certain cases, usually when the inertial terms vanish in some
natural way. We will now examine a Iew oI these problems having exact solutions.
PROBLEMS WITH ZERO INERTIA
Eirst, let`s consider problems in which the Iluids elements travel along straight streamlines at
constant velocity. Then their acceleration vanishes identically.
06703 84 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Flow in Long Straight Conduit of Uniform Cross Section
Suppose we have pressuredriven Ilow in a long straight
conduit whose cross section does not vary along the Ilow. In
mathematical terms, the conduit is a cylinder oI arbitrary
crosssection. DeIine Cartesian coordinates such that the axis
oI the cylinder corresponds to the :axis. In a very long pipe,
we expect that v
:
will depend on : (as well as x and v) near the
entrance and exit oI the pipe.
Entrance
region
Eully
developed
Exit
region
:
In particular, near the entrance, we say the velocity proIile is 'developing¨; i.e. evolving with :.
Outside the entrance and exit regions v
:
will be independent oI :. This situation is called fully
developed flow. Eor Iully developed, steady Ilow:
v
:
÷ v
:
(x,v)
v
x
÷ v
v
÷ 0
Note that this automatically satisIies the Continuity Equation:
V
v ÷ cv
:
/c: ÷ 0
Eor this Iorm oI solution, it turns out that the nonlinear inertial terms automatically vanish:
v
Vv ÷ 0
To convince yourselI oI this, consider steady Ilow. Since the velocity oI a Iluid element is
constant along a straight streamline:
Ior steady Ilow: v
Vv ÷ Dv/Dt ÷ a ÷ 0
In other words, Iluids elements are not accelerating. Thus inertial Iorces are identically zero in
steady pipe Ilows. Strictly speaking, Reynolds number should not be thought oI the ratio oI
inertial to viscous Iorces in this problem, since inertia is zero Ior laminar Ilow although the
Reynolds number is not zero.
Eor a steady Ilow, the NS equations are:
0 ÷ µV
2
v  Vp ¹ µg
06703 85 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Note that Ior a vertical pipe:
*
p ÷ p(:)
but more generally Ior a horizontal or inclined pipe:
p ÷ p(x,v,:)
where the dependence on position in the cross section arises Irom the contribution to pressure
Irom the hydrostatic head (i.e. Irom g). Eor this and some other problems, it`s helpIul to
decompose the total pressure into contributions Irom gravity (i.e. hydrostatic pressure, p
h
) and
Irom Ilow (called the dynamic pressure, P)
p p P
p p P
h
h
= +
V = V + V
µg
¸
Erom our earlier analysis oI hydrostatic equilibrium (see page 39), we know that V = p
h
µg .
Note that V ÷ = V p P µg
Next, we substitute this into the NS Equation. Expanding them in component Iorm:
x
P
x
v
P
v
P P :
 component:
 component:
0
0
= ÷
= ÷
=
c
c
c
c
:component: 0
2
= ÷ µV v
dP
d:
:
(45)
where we have written the last term as the total derivative instead oI the partial derivative
because the Iirst two equations require that P be a Iunction oI : alone. We can immediately
deduce that P must be a linear Iunction oI :
µV
2
v
dP
d:
:
g x v
f :
,
¸
=
since the dP/d: is independent oI x and v, while the velocity proIile is independent oI :, so (45)
requires that the two Iunctions oI diIIerent variables be equal:
g(x,v) ÷ f(:) ÷ const. w.r.t. x,v,:
*
In the absence oI gravity, cp/cx and cp/cv must vanish: see the x and vcomponents oI (45).
06703 86 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
which can only be true iI both Iunctions equal the same constant: thus P(:) must
be a linear Iunction. Eor steady Ilow in a circular conduit oI radius a, the 'no
slip¨ b.c. requires
at r ÷ a: v
:
÷ 0 (46)
Since neither the b.c. nor the diIIerential equation contain any dependence on u,
we expect the solution to be axisymmetric about the :axis:
v
:
÷ v
:
(r)
(45) becomes:
1 1
r
d
dr
r
dv
dr
dP
d:
:
÷ ÷ const
µ
The general solution oI this equation is:
v r
dP
d:
r c r c
:
= + +
1
4
2
1 2
µ
ln
Requiring the solution to be bounded at the center oI the tube (as r÷0) Iorces us to choose c
1
÷0
while the remaining constant can be chosen to satisIy the noslip condition (46). The particular
solution is the Iamiliar parabolic velocity
proIile:
v r
dP
d:
r a
:
,
= ÷
1
4
2 2
µ
The volumetric Ilowrate through the
conduit is computed Irom:
Q da v r r dr
A
:
a
= =
n v
2
0
t
or Q
a dP
d:
= ÷
t
µ
4
8
which is called Poisueille's Formula. This Iormula was been derived Ior a number oI diIIerent
cross sections. In general
Q
kA dP
d:
= ÷
2
µ
where A ÷ crosssectional area oI duct
and where k is some constant which depends on the shape oI the duct; e.g.
06703 87 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
circle: k ÷ 1/8t ÷ 0.0398
square: k ÷ 0.0351
ellipse:
c
t c 4 1
2
+
,
where c ÷ b/a s 1 is the ratio oI the minor to major axis.
Flow of 1hin Film Down Inclined Plane
Suppose we have Iluid overIlowing some reservoir
and down an inclined plane surIace. Although there
might be some entrance or exit eIIects (at the upstream
and downstream ends oI the plane), iI the plane is
suIIiciently long compared to these regions, then what
we see in experiments is a region in which the Iluid
Ilows downward as a Iilm oI uniIorm thickness. Let's
try to analyze this central region in which the Iilm in
uniIorm.
Let the xaxis be oriented parallel to the inclined plane in the direction oI Ilow and let the vaxis
be perpendicular to the plane. One oI the boundary conditions will no 'noslip¨ at the solid
surIace:
at v÷0 v
x
÷ 0
At the very least, there must be Ilow in the xdirection: otherwise there could be no viscous Iorce
to balance gravity. In addition, the xcomponent must vary with v v
x
must vanish at v÷0 (no
slip) and be nonzero Ior v~0. The simplest Iorm oI solution which is consistent with those
constraints is:
v ÷ v
x
(v)e
x
Note that this Iorm automatically satisIies the requirement oI continuity Ior an incompressible
Iluid:
V
v ÷ cv
x
/cx ÷ 0
Substituting this velocity Iield into the NavierStokes equations:
x 0 ÷ cp/cx ¹ µd
2
v
x
/dv
2
¹ µg
x
(47)
v 0 ÷ cp/cv ¹ µg
v
(48)
: 0 ÷ cp/c:
06703 88 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
where g
x
÷ g
e
x
÷ gcos
and g
v
÷ g
e
v
÷ gsin
Integrating (48) with respect to v:
p(x,v) ÷ µgvsin ¹ c(x) (49)
where the integration constant might depend on x, but cannot depend on : (according to :
component above) or v. Now let's turn to the boundary conditions. Continuity of stress across
the interIace yields:
at v÷o n
T
liq
÷ n
T
gas
÷ n
(p
gas
I ¹ t tt t
gas
)
Now the viscosity oI air is about 0.001 times that oI water. Then it is reasonable to treat the air
as an inviscid Iluid: i.e. neglect t tt t
gas
. This leaves
n
T
liq
~ np
gas
÷ np
atm
(50)
where p
atm
÷1atm. Now e
v
is the unit normal to the interIace in this problem
e
v
T
liq
÷ e
v
(pI ¹ t tt t) ÷ pe
v
¹
vx
e
x
¹
vv
e
v
¹
v:
e
:
When v
x
÷ v
x
(v), the only nonzero components oI the deviatoric stress tensor are
xv
÷
vx
.
*
Dropping the other terms, (50) becomes:
pe
v
¹
vx
e
x
÷ e
v
p
atm
Equating separate components:
xcomponent:
vx
(x,v÷o) ÷ 0 (51)
vcomponent: p(x,v÷o) ÷ p
atm
With this result, we can evaluate the integration constant in (49):
p(x,v÷o) ÷ µgosin ¹ c(x) ÷ p
atm
Thus c(x) ÷ p
atm
¹ µgosin
*
Eor a Newtonian Iluid, the 9 scalar components oI the stress tensor are expressed in terms oI
the derivatives oI the velocity Iield on p145 oI Whitaker. These expressions (except Ior a change
in sign) can be Iound on p88 oI BSL or at the website Ior our course
http://www.andrew.cmu.edu/course/06703/NLV¸RCCS.pdI.
06703 89 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
(49) becomes: p(x,v) ÷ p
atm
¹ µg(ov)sin
Thus cp/cx ÷ 0 and (47) becomes:
µd
2
v
x
/dv
2
÷ µgcos
Noslip at the wall requires:
at v÷0 v
x
÷ 0
whereas iI the stress in (51) is evaluated using
Newton's law oI viscosity, we also require:
at v÷o
vx
÷ µ dv
x
/dv ÷ 0
Using these two boundary conditions, the velocity
proIile can be uniquely determined:
v v
g v v
x
=
÷
,
µ o 
µ o o
2 2
2
2
cos
By integrating over a plane perpendicular to the Ilow,
we can evaluate the Ilowrate:
Q da v v dv d:
A
h
x
÷ ÷ ( n v
0 0
o
)
Q
gW
=
µ o 
µ
3
3
cos
where W is the width oI the plane.
Time Out: The main novelty oI this problem is the
treatment oI the Iree surIace. We treated the air as iI it was inviscid, although it has some
viscosity. How important is the drag imposed by the air? This is the subject oI Hwk #6, Prob. 2
06703 90 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
To answer this question, let`s consider a vertical Iilm oI water
in contact with a vertical Iilm oI air, as shown in the sketch at
right. Let`s resolve the problem and see how large o
a
must be
Ior a given o
w
beIore we can neglect the eIIect oI the air. Eor
Iully developed Ilow, the velocity and pressure proIiles should
have the Iorm:
v v v
v v
p x v :
x x
v :
=
= =
=
0
1 , , atm
NSE
x
becomes:
Ior water: 0
2
2
= + µ µ
w
x
w
w
d v
dv
g
Ior air: 0
2
2
= µ
a
x
a
d v
dv
where we have taken µ
a
÷ 0. Applying 'no slip¨ at each oI the three interIaces
at v÷0: v
x
w
0 0 =
at v÷o
w
: v v U
x
w
w x
a
w
o o = ÷
at v÷ o
w
¹ o
a
: v
x
a
w a
o o + = 0
The particular solutions to the two ODE`s are
v v
g
v v U
v
x
w w
w
w
w
= ÷ +
µ
µ
o
o 2
(52)
v v U
v
x
a w a
a
=
+ ÷ o o
o
(53)
The interIacial speed U is unknown, but is choosen to match the shear stress at the interIace:
at v÷o
w
: t o t o
vx
w
w vx
a
w
=
Eor Newtonian Iluids, the stresses can be related to velocity proIile:
06703 91 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
µ µ
o o
w
x
w
v
a
x
a
v
dv
dv
dv
dv
w w
= =
=
Using the velocity proIiles oI (52) and (53), this stress matching yields:
÷ + =
1
2
µ o µ
o
µ
o
w w w
w
a
a
g
U U
Solving Ior U: U
g
w w
w
w
a
a
=
+
1
2
µ o
µ
o
µ
o
The Ilowrate oI the water is
Q
W
v v dv
g
U
g
x
w w w
w
w
w w
w
w
w
a
a
w
w
a
a
w
= = + =
+
+
0
3
1
2
3
12 12
4 o
µ o
µ
o
µ o
µ
µ
o
µ
o
µ
o
µ
o
(54)
II the air Iilm is very thick, the Ilowrate becomes
Q
W
Q
g
a
w w
w
·
÷·
÷ = lim
o
µ o
µ
, ,
3
3
(55)
which is the same expression we had in the Notes when the viscosity oI air was completely
ignored. Dividing (54) by (55):
Q
Q
w
w
a
a
w
w
a
a
w
a
w
a
w
a
w
a
·
=
+
+
=
+
+
1
4
4
1
4
4
µ
o
µ
o
µ
o
µ
o
µ
µ
o
o
µ
µ
o
o
When the viscosity oI water is 1000 times larger than air (i.e. µ
w
÷ 1000 µ
a
), this gives
Q
Q
w
a
w
a
·
=
+
+
1
4
4000
1000
o
o
o
o
To reduce the Ilowrate by 1° means
Q
Q
·
= 0 99 . , Ior which the air Iilm thickness must be
06703 92 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
o o
a w
= 0 074 .
So even iI the air is stilled by a nearby boundary, the drag oI the air on the Iree surIace oI the
water will be negligible (provided the boundary is not too close). In the absence oI a rigid
boundary in the air, negligible error is made by treating the air as inviscid.
Time In!
PROBLEMS WITH NONZERO INERTIA
L&L list only three Ilow problems in which both including viscous and inertial terms are
important and in which exact solutions are known:
1) rotating disk
2) converging (or diverging) Ilow between nonparallel planes
3) the submerged jet
Rotating Disk²
We will now examine the solution oI the Iirst
problem the rotating disk because it is used as a
model system Ior transport experiments. An inIinite
plane disk immersed in a viscous Iluid rotates uniIormly
about its axis. Determine the motion oI the Iluid caused
by this motion oI the disk. This problem was Iirst solved
by von Karmen (1921) using cylindrical coordinates with
the :axis coinciding with the axis oI rotation.
DeIine: ,
e
v
= :
where v ÷ µ/µ ÷ cm
2
/s
is called the kinematic viscosity.
Then: v
r
(r,:) ÷ reF(,)
v
(r,:) ÷ reG(,)
*
See S:p93 (6th Ed); L&L:p79.
06703 93 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
v : H
= ve ,
p(:) ÷ µeP(,)
Continuity and NS become:
2F ¹ H´ ÷ 0
r: F
2
¹F´HG
2
F" ÷ 0
u: 2FG¹HG´G" ÷ 0
:: P´¹HH´H" ÷ 0
where the prime (´) denotes diIIerentiation with
respect to ,. Boundary conditions take the Iorm:
:÷0: F÷H÷P÷0, G÷1
:÷·: F÷G÷0
An important property oI this solution is that the :component oI velocity depends only on :. Eor
this reason, convective heat and mass transIer problems becomes onedimensional.
This is perhaps the only problem in which there is Ilow normal to a wall where the masstransIer
coeIIicient can be determined analytically. Eor this reason, the rotating disk is a Iavorite tool oI
researchers in mass transIer.
0 1 2 3 4
0
0.2
0.4
0.6
0.8
1.0
06703 94 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Creeping Flow Approximation
CONEANDPLATE VISCOMETER
The coneandplate viscometer consists oI a Ilat plate, on which
is placed a pool oI liquid whose viscosity is to be measured, and
an inverted cone which is lowered into the liquid until its apex
just touches the plate. The cone is then rotated at some angular
velocity, O, and the torque required to turn the cone or to keep
the plate stationary is measured.
Spherical coordinates are most convenient to describe this
problem, since the surIace oI the cone and oI the plate can be
deIined as u÷const surIaces:
surIace oI cone: u ÷ o
surIace oI plate: u ÷ t/2
The cone is undergoing solidbody rotation (see HWK #4, Prob.
3):
Ior uso v(r) ÷ O×r
In spherical coordinates, the position vector is
r ÷ re
r
(u,)
while the angular velocity is (see Iigure and p49):
O ÷ Oe
:
÷ O(cosu)e
r
 (sinu)e
 (56)
Substituting into the expression Ior solid body rotation:
u·o v ÷ O×r ÷ rOsinu e

The principle direction oI Ilow is the direction. 'No
slip¨ requires:
at u÷o v
÷ rOsino
v
r
÷ v
÷ 0
at u÷t/2 v
÷ v
r
÷ v
÷ 0
The simplest velocity proIile which is consistent with these boundary conditions is:
O
u
o
c
x
v
:
r
u

06703 95 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
v
÷ v
(r,u)
v
r
÷ v
÷ 0
Next, I`d like to argue that the pressure proIile can be expected to be independent oI .
p ÷ p(r,u)
Since there exists only a component oI velocity, Iluid streamlines will turn out to be circles (the
contour corresponding to r÷const and u÷const). The circle corresponds to 0 s  s 2t.
By analogy with the last problem, you might guess (incorrectly) that pressure must decrease
along the direction oI Ilow. However, in steady Ilow p cannot continously decrease with  Ior all
. At the very least, pressure must be periodic in ; in other words, p(r,u,) ÷ p(r,u,¹2t). So
any decreases in pressure over part oI the cycle will have to be balanced by increases over the
remaining part. Why should the pressure be higher at some points along the streamline than at
other? There is no geometrical asymmetry with respect to  and no reason to expect any 
dependence in the pressure.
The velocity proIile automatically satisIies continuity:
V
v ÷ (rsinu)
1
cv
/c ÷ 0
Ignoring gravity, the NavierStokes equations in spherical coordinates become:
+
r µv
2
/r ÷ cp/cr (57)
u µv
2
cotu/r ÷ r
1
cp/cu (58)
 0
1 1
2
2
2 2 2
÷ ¹ µ
u
u
u
  
r
r
r
v
r
r
v v
r
c
c
c
c
c
cu
c
cu
÷
sin
sin
sin
(59)
Notice that the pressure and velocity Iields have been separated. We can Iirst solve the
component Ior the velocity proIile and then substitute the result into the r and ucomponents to
solve Ior the pressure proIile. Based on the boundary conditions, we might try a solution oI the
Iorm:
v
(r,u) ÷ rf(u)
When this is substituted into the equation above, the rdependence cancels out, leaving a
second order ordinary diIIerential equation in f(u). The solution leads to:
+
http://www.andrew.cmu.edu/course/06703/NSE¸sph.pdI
06703 96 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
sin
g
v r
g

u
= O o
o
(60)
where
g( ) ÷ cot ¹
1
2
ln
1¹ cos
1 cos
sin u u
u
u
u
This Iunction is plotted in the Iigure at right. Notice that
in the center (i.e. Ior u near t/2) the Iunction is nearly
linear. A more careIul asymptotic analysis would reveal
that
2
5
lim 2 g O
u÷
u = c + c (61)
where c u
t
÷ ÷
2
Notice that this is a linear Iunction oI u as expected
Irom the plot above.
Velocity Profile for Shallow Cones. Most cone
andplate viscometers are designed with cone
angles near to t/2. The reason Ior this will be
apparent in a moment. Note the arc lengths rc and
rc
1
on the Iigure at right. Eor shallow cones (i.e. as
o÷t/2), rc asymptotically becomes the vertical
distance Irom some arbitrary point (r,u) in the Iluid
to the plate,
lim
u
c
÷
=
r : , ,
while rc
1
becomes the vertical distance between the plate at the cone:
1
2
1
lim r h r
u ÷
c =
So in this limit we can use (61) to replace
as o÷t/2: g
:
r
u c ÷ = 2 2
u
o
c
r
r
:÷r
h÷r
06703 97 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
and
1
2 2
h
g
r
o ÷ c = and sin 1
Then (60) simpliIies to linear shear Ilow (at least locally)
Ior o÷t/2: v r
:
h

= O
Notice that the rate oI strain is independent oI position:
Ior o÷t/2:
1 1
const.
v
r r
: h r r

c
O O O
= = = =
c c c
This is an important advantage Ior a rheometer, since all the Iluid experiences the same strain
rate. Later, we will show that stress is also spatially uniIorm Ior shallow cones.
Torque. To use this device as a viscometer, we need to interpret torque
measurements. So let`s try to evaluate the torque Ior a given velocity
proIile. Recall that torque is Iorce times lever arm:
÷ (rsino)F
Using vector notation:
÷ r×F
In our problem, the Iorce F is not
concentrated at one point, but instead
is distributed over the surIace oI the
plate. Let`s consider the contribution to Iorce and torque Irom
some diIIerential element oI surIace having area da.
Once the velocity proIile is known, we can evaluate the stress
Iield Irom Newton`s law oI viscosity. Given the stress Iield,
we can calculate the Iorce on any diIIerential solid surIace
element oI area da Irom:
dF ÷ n
Tda
where n is a unit normal pointing out oI the body dF acts on.
Similarly, we calculated the contribution to the torque by
crossing this Iorce with the local lever arm:
d ÷ r×dF ÷ r×(n
Tda)
06703 98 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Let`s calculate the torque exerted by the Iluid acting on the stationary plate. Then we want to
choose n to point out oI the plate or
n ÷ e
The net Iorce exerted on the lower surIace is:
F ÷ )
A
dF ÷ )
A
e
Tda
Similarly, the torque exerted on the lower surIace is:
÷ )
A
r×dF ÷ )
A
r×(e
T)da (62)
Eirst let`s consider: e
T ÷ T
r
e
r
¹ T
e
¹ T
e
To evaluate the torque, we need to Iirst cross this vector with the lever arm, which is the position
vector. In spherical coordinates, the position vector is:
r(r,u,) ÷ re
r
(u,)
r×(e
T) ÷ rT
r
(e
r
×e
r
) ¹ rT
(e
r
×e
) ¹ rT
(e
r
×e
)
÷ rT
e
 rT
e
Anticipating that the torque vector will have the same direction as the angular velocity O, we
dot both sides by e
:
which equals e
on the plate (u ÷ t/2):
d
da
rT rT rT
:
T
= ÷ × ÷ = × = ÷ = ÷ e r e T e r e T e e e
u u u u u uu  u u u
Now we substitute Newton`s law oI viscosity:
+
1
sin
sin sin
v
v
T
r

u
u u
(   c µ c
= t = u +
( 
cu u u c
( \ .
¸ ¸
(63)
What`s leIt is to substitute our velocity proIile Irom (60) into (63), which yields
2
sin
where 2
sin
C
T C
g
u
o
= ÷ ÷ µO
o
u
(64)
+
http://www.andrew.cmu.edu/course/06703/NLV¸sph.pdI
06703 99 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
and where o is the cone angle. On the surIace oI the Ilat plate, u ÷ t/2 and sinu ÷ 1. T
(evaluated at u÷t/2) will be independent oI ; then we can choose da to be a ring oI radius r and
thickness dr, leaving:
:
÷ )
0
R
(rT
)(2trdr)
3 2
3
sin
R
g
o
= tµ O
o
(65)
where R is the radius oI the region oI the plate which is wetted by the Iluid.
Shear Stress and Torque for Shallow Cones. II the cone angle o is very shallow (i.e. o÷t/2)
then T
will be practically independent oI position: Irom (64):
2
2
lim
2
T C
u
o÷u÷
÷ µO
u = =
c
Ior any u. This means the entire Iluid experiences the same stress. This has a number oI
important advantages Ior rheological studies making the coneandplate viscometer one oI the
important viscometric flows. Substituting (61) into (65)
3 2
3
0
lim R
c÷
= tµ
c
O OO O
Comment on Solution. There is a problem with our solution to NSE: when (60) is
substituted back into (57) and (58), there is no single Iunction p(r,u) which will satisIy them. In
other words, (60) is not an exact solution. It turns out that (60) is a reasonably good
approximation iI O is not too large. The exact solution has the Iorm:
2
sin
g
r O
g

u
= O o + O
o
v e
where O(O
2
) means that this term vanishes like O
2
as O
÷0. The centripetal Iorce on Iluid elements undergoing
a circular orbit causes those Iluid elements to be 'thrown
outward¨ in the ¹rdirection. Since Iluid elements near
the rotating cone are rotating Iaster than Iluid elements
near the stationary plate, we have outIlow near the cone
supplied by inIlow near the plate. The resulting r and u
components oI velocity are called secondary flow,
whereas the original component is call the primary
flow.
O
2
(the secondary Ilow) vanishes Iaster than O (the primary Ilow), so Ior small O, the
leading term is approximately correct. This is called the creepingflow approximation.
06703 100 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
CREEPING FLOW AROUND A SPHERE (Re 0)
Let`s return to the problem oI Ilow around a
sphere (or motion oI a sphere through a stagnant
Iluid). Eor boundary conditions, we impose 'no slip¨
on the surIace oI the sphere and Iar Irom the sphere
the Ilow is undisurbed:
at r÷a: v ÷ 0
as r÷·: v ÷ U
Here an exact solution to the NavierStokes
equations is not possible. OI course, the vector equation can be converted into a scalar equations
using the stream Iunction, but that yields a 4th order nonlinear P.D.E. Although this could be
solved numerically, considerable simpliIication can be obtained iI either the viscous terms or the
inertial terms can be neglected even iI they are not identically zero. One limiting case is
creeping flow which corresponds to the limit in which the Reynolds number is small (i.e. Re÷0).
In this limit the inertial terms in the Navier Stokes equations can usually be neglected.
Scaling
To show that inertial terms are neglibible, let`s try to estimate the order oI magnitude oI
viscous and inertial terms Ior uniIorm Ilow at speed U over a sphere oI radius R.
µ µV v v v
V = ÷V +
inertia viscous
p
2
We will use a technique called scaling (akin to dimensional analysis). We start by listing all the
parameters in the problem. In this problem, the parameters are
parameters: U, R, µ and µ
A characteristic value Ior each term in the equations oI motion is then written as a product oI
these parameters raised to some power:
each term · U
a
R
b
µ
c
µ
d
Eor example, throughout most oI the region, the Iluid velocity is undisturbed:
,v, ~ U
where the symbol '~¨ should be read as 'scales like¨ In 'scaling¨ we ignore any position
dependence as well as any numerical coeIIicients, so ,v, scales as U, although ,v, might be
signiIicantly less than U near the surIace oI the sphere. Erom the boundary conditions, v changes
Irom 0 at r÷R to U at r÷·. To estimate the magnitude oI the gradient Vv we need to estimate
06703 101 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
over what distance most oI this change occurs. In the solution oI the potential Ilow problem, the
velocity proIile along the rear stagnation line is
v r U
R
r
r
, cos u u = = ÷
,
0 1
3
We expect something similar Ior Stokes Ilow. Two oI the
nine scalar components oI Vv are
c
c
u
v
r
U
R
r
U
R
r
= ~ 3
3
4
cos
1 1
3
4
r
v
U
r
R
r
U
R
r
c
cu
u = ÷ ÷
~ sin
We again ignore the position dependence: we scale r as R and treat cosu and sinu as '1¨. Then
both components above scale as U/R as do other components oI Vv associated with derivatives
oI v
. So
V ~ ~ v
A
A
v
r
U
R
Likewise the velocity gradient can be expected to decay Irom a maximum value oI U/R near the
sphere surIace to zero in the bulk; this change occurs over a distance on the order oI R.
V ~
V
~ =
2
2
v
v A
Ar R
U
R
U
R
With these estimates, we can Iurther estimate the magnitude oI viscous and inertial Iorces:
inertia = V ~
= µ µ
µ
v v
U
U
R
U
R
2
viscous = V ~ µ µ
2
2
v
U
R
inertia
viscous
~ = =
µ
µ
µ
µ
U R
U R
UR
2
2
Re
where Re is the Reynolds number. So as Re÷0, we should be able to neglect inertial Iorces
Ior Re··1 0 v = ÷V + p µV
(66)
3
U
R
R 2R 3R
0
U
3
U
R
rR
v
06703 102 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
V
v ÷ 0
at r÷R v ÷ 0
as r÷· v ÷ U, p ÷ p
Eq. (66), called Stokes Equation, is common to all lowReynolds number problems; it`s not
speciIic to the sphere. A common trick to reduce the number oI unknowns is to take the curl oI
both sides oI the equation:
0 ÷ V×Vp ¹ µV×V
2
v (67)
But V×Vp ÷ 0
Moreover, using identity E.1,
V = V V ÷ V× V × = ÷
2
0
2
v v v v
curl
Ior incompressible Ilow. Then (67) becomes:
curl
v 0 = (68)
Jelocity Profile
We might try to seek a solution having the Iorm oI potential Ilow:
v ÷ V (69)
since V×v ÷ V×V ÷ 0
(68) is automatically satisIied. But we know that the potential Ilow solution Ior the sphere does
not satisIy the no slip boundary condition. On the other hand, the boundary conditions are
axisymmetric:
v
÷ 0, c/c ÷ 0
so we might seek a solution using the stream Iunction:
v e = V ×
,
¢ u
u

r
r
,
sin
Computing the curl in spherical coordinates using the tables
(http://www.andrew.cmu.edu/course/06703/Vops¸sph.pdI) (also see HWK #5, prob. 1):
06703 103 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
2
1 1
curl
sin sin
sin
r
r
v
v
r r r
r

u
¢
 
c¢ c¢
= = ÷

u cu u c
u
\ .
e
v e e
Taking the curl a second time, using the same tables:
2 2
curl
sin sin
E
r r
 
¢
 
= V× = = ÷ ¢

u u
\ .
e e
v (70)
where E
2
is a partial diIIerential operator given by:
E
r r
2
2
2 2
1
¢
c ¢
c
u c
cu u
c¢
cu
= +
sin
sin
Since we have to take the curl twice more, it might look like we have a lot oI algebra to look
Iorward to. But, as it turns out, the rest is easy. We need to evaluate:
2
3 2 2
curl curl curl
sin
E
r

¦ ¹
÷ ¢ ¦ ¦
´ `
= V× =
¦ ¦
u ¹ )
e
v v
where the second equality is (70). The argument oI this Iinal curlsquared has exactly the same
Iorm as that oI the leIthand side oI Eq. (70), except that the scalar +¢, which is a Iunction oI r,u,
is replaced by E
2
¢, which is also a scalar Iunction oI r,u. So all I have to do is to replace ¢ by 
E
2
¢ on the righthand side oI (70):
curl curl
3 2
2
2 2 2 2
v
e
e e
=
÷
= ÷ ÷ =

 
¢
u u
¢
u
¢
E
r r
E E
r
E E
,
, ,
sin sin sin
To satisIy (68), which represents the curl oI the NavierStokes equation, we choose the
streamIunction to satisIy:
E
2
(E
2
¢) ÷ 0
Translating the boundary conditions at r÷R in terms oI stream Iunction:
v
r
÷0 c¢/cu ÷ 0 at r÷R
v
÷0 c¢/cr ÷ 0
Translating the boundary conditions at r÷· in terms oI stream Iunction (see HWK #5, Prob. 1a):
as r÷· ¢ ÷ (1/2)Ur
2
sin
2
u (71)
06703 104 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
The trivial solution ¢ ÷ 0 satisIies the P.D.E. and the b.c. at r÷R, but not the b.c. at r÷·. Based
on the Iorm oI this nonhomogeneous b.c., we guess the solution has the Iollowing Iorm:
Try: ¢(r,u) ÷ f(r)sin
2
u
E
2
¢ ÷ E
2
f(r)sin
2
u ÷ ... ÷ (f"2r
2
f)sin
2
u ÷ g(r)sin
2
u
where: g(r) ÷ f"2r
2
f
Then: E
2
(E
2
¢) ÷ (g"2r
2
g)sin
2
u
The NavierStokes equation become
E
2
(E
2
¢)÷0 g"2r
2
g ÷ 0 (72)
deI`n oI g f"2r
2
f ÷ g(r) (73)
b.c. `s: f ÷ (1/2)Ur
2
as r÷· (74)
f ÷ f ` ÷ 0 at r÷R (75)
These two coupled secondorder O.D.E.`s can be combined to produce a single Iourthorder
O.D.E. in f(r). The result is a CauchyEuler equation whose general solution is (see Iootnote on
page 52):
f(r) ÷ c
1
r
1
¹ c
2
r ¹ c
3
r
2
¹ c
4
r
4
Applying b.c. (74): c
4
÷ 0
c
3
÷ (1/2)U
As r becomes large, terms which are proportional to higher power oI r dominate those oI lower
power. To have the third term win over the Iourth, requires us to kill the Iourth term by setting
its coeIIicient to zero. The remaining constants are evaluated in a straightIorward manner. The
result is:
¢ u u ( , ) sin r UR
R
r
r
R
r
R
= ÷ +
,
2 1
4
3
4
1
2
2
2
(76)
06703 105 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
v r U
R
r
R
r
r
( , ) cos u u = ÷ +
,
1
3
2
1
2
3
v r U
R
r
R
r
u
u u ( , ) sin = ÷ ÷ ÷
,
1
3
4
1
4
3
The Iigure at right compares the streamlines Ior
Stokes Ilow with those Ior potential Ilow. The
streamlines correspond to values oI the
streamIunction which are uniIormly spaced at
about the same interval Ior both proIiles.
Displacement of Distant Streamlines
Erom the Iigure above, you can see that streamlines in Stokes Ilow are displaced away Irom
the sphere even those streamlines which are quite distant especially compared with potential
Ilow, in which the distant streamlines become straight. It turns out that all streamlines are
displaced away Irom the sphere in Stokes Ilow.
Consider the streamline in the Iigure at right
which corresponds to ¢ ÷ ¢
0
. Ear upstream, the
coordinates oI the streamline correspond to r÷·
and u÷t such that rsinu ÷ const, which we will
denote as v (see Hwk #5, Prob. 1a). The
relationship between v and ¢
0
can be deduced
Irom (71):
lim , sin
r
v
r Ur
÷·
÷
÷
u t
¢ u u
1
2
2 2
2
¢
0
1
2
2
= Uv (77)
At the equatorial plane u ÷ t/2, the r coordinate oI the streamline must satisIy (76):
¢ ¢
t
( , ) r UR
R
r
r
R
r
R
90
2
0
2 1
4
90
3
4
90 1
2
90
2
= = ÷ +
,
(78)
Eliminating ¢
0
between (77) and (78):
06703 106 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
v r Rr
R
r
2
90
2
90
3
90
3
2
1
2
= ÷ +
or v r
R
r
R
r
= ÷
+
90
90 90
3
1
3
2
1
2
Eor distant streamlines, r
90
will be very large compared to R, so that R/r
90
·· 1. Then the
squareroot in the expression above is unity plus a small correction, which can be estimated as
the Iirst term in a Taylor series expansion oI the squareroot Iunction: 1 1
1
2
+ = + + c c
v r
R
r
O
R
r
r R = ÷
+
,
~ ÷
90
90 90
2
90
1
3
4
3
4
where the 'O(R/r
90
)
2
¨ means that these terms decay to zero like (R/r
90
)
2
as R/r
90
÷0. When
R/r
90
is suIIiciently small, we can neglect these terms compared to the others. This is how the
second (approximate) equality above was obtained.
Einally lim
r
r v R
90
90
3
4
÷·
÷ = = , , o
Thus, even streamlines inIinitely Iar away are displaced a distance equal to / R.
Pressure Profile
Once the velocity proIile is known, the pressure can be determined Irom
Vp ÷ µV
2
v
Substituting v:
c
c
µ u
c
cu
µ u
p
r
URr
r
p
URr
=
=
÷
÷
3
1 3
2
3
3
cos
sin
and then integrating with p÷p
at r÷· as the
b.c. yields:
p r p UR
r
,
cos
u µ
u
= ÷
·
3
2
2
This proIile corresponds to a higher pressure on
the upstream side oI the sphere (u÷t) than on
06703 107 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
the downstream side (u÷0). Thus it appears as iI a drag will be produced by this solution.
)
A
pnda ÷ ... ÷ 2tµRUk
is called the form drag which arises Irom normal stresses. This however is not the total drag.
More generally, we expect the net Iorce exerted on the particle by the Iluid is:
F n T n n k
k k
= = ÷ + =
da pda da RU
A A
RU
A
RU 2 4
6
tµ tµ
t tµ
(form drag) (skin friction)
where A is the surIace oI the sphere r÷R
n ÷ e
r
Erom symmetry, we expect that this Iorce will have only a :component (the direction oI bulk
motion); so let`s concentrate on Iinding that component:
F
:
÷ )
A
k
(e
r
T)da (79)
Substituting T ÷ pI ¹ t
e
r
T ÷ p(e
r
I) ¹ e
r
t
÷ pe
r
¹ e
r
t
e
r
t ÷ t
rr
e
r
¹ t
r
e
¹ t
r
e
Eor the determined velocity proIile, Newton`s law oI viscosity in spherical coordinates (W:146)
yields t
r
÷0 Ior all r and t
rr
÷0 at r÷R:
at r÷R e
r
T ÷ pe
r
¹ t
r
e
k
(e
r
T) ÷ p(k
e
r
) ¹ t
r
(k
e
)
Using k ÷ e
:
Irom (56): k
(e
r
T) ÷ pcosu  t
r
sinu
Einally, we can integrate using an azimuthal strip oI width Rdu and
oI radius Rsinu:
da ÷ 2t(Rsinu)(Rdu)
(79) becomes:
06703 108 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
F R p R R d
: r
= ÷ ÷
2
2 2
0
t u u u t u u u
u
t
, sin cos , sin
Einally t
r
is expressed in terms oI the velocity proIile using Newton`s law oI viscosity and
subsequent integration yields:
F
:
÷ 2tµRU ¹ 4tµRU
We have already noted that the contribution Irom the pressure proIile is call Iorm drag. The
second contribution is call skin friction. The total drag Iorce is
F U
drag
R = 6tµ (moving reIerence Irame)
which is known as Stokes Law (1850).
*
In the expression above, U is the velocity the distant
Iluid Ilowing around a stationary sphere. II we switch back to the original (stationary) reIerence
Irame (see page 48I), Stokes Law becomes
F U
drag
R = ÷6tµ (stationary reIerence Irame)
where now U is the velocity oI the sphere moving through otherwise
stagnant Iluid. Note that the drag Iorce acts in a direction opposing the
direction oI motion oI the sphere.
Experimental results Ior the drag Iorce around submerged bodies are
usually expressed in terms oI a dimensionless drag coefficient. The
quantity that is used to make this drag Iorce dimensionless is
1
2
2
µU   = =
K. E.
vol
Iorce
area
This quantity also can be shown (according to Bernoulli`s equation) to represent the Ap required
to stop Iluid which is Ilowing at speed U. Multiplying this by the projected area gives the Iorce
required to stop the Ilow, which would otherwise pass through the sphere:
C
F
U R
D
drag
=
×
1
2
2 2
µ t
projected area
¸
*
Sir George G. Stokes (18191903), born in Shreen, Ireland, educated at Cambridge; theoretical
physicist.
06703 109 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
where tR
2
is the projected area oI the sphere; in other words, tR
2
is the area oI the sphere`s
shadow cast along the direction oI Ilow. Using this deIinition, Stokes equation Ior the drag Iorce
on a sphere yields:
C
D
=
12
Re
where Re ÷
µ
µ
UR
Comparison with experimental results conIirm that this works very well Ior Re·0.1. The
reproduction above (taken Irom BSL, p192) uses the diameter Ior Reynolds number, rather than
the radius. The 'Iriction Iactor¨ on the vaxis diIIers by a Iactor oI two Irom the drag coeIIicient
deIined above.
CORRECTING FOR INERTIAL TERMS
Eor larger Re, Stokes law underestimates the drag Iorce. OI course, this is due to the
increasing importance oI inertia which has been neglected. A number oI investigators have
attempted to extend the validity oI Stokes law by including inertial terms in the analysis. We
will now summarize some oI these and hint at the diIIiculties involved.
Eirst, note that as Re ÷0 Ior Ilow around a sphere, the NSE approximated by
µV
2
v ÷ Vp
which is called Stokes equation. Stokes (1850) solved this equation to obtain:
06703 110 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
F ÷ 6tµRU.
The usual approximation to solving diIIerential equations containing a small parameter (i.e. Re)
is to perIorm a perturbation expansion. Basically, the idea is to perturb the small parameter
away Irom zero value by means oI a Taylor series.
PERTURBATION EXPANSION
time out: Let`s illustrate the main idea behind this powerIul mathematical technique by means oI
a simple example (Example 25.1 Irom Greenberg). Eind the solution v(x;c) to the Iollowing
ODE involving a small parameter which we denote as c:
v´ ¹ v ¹ cv
2
÷ 0 (80)
subject to the initial condition: v(0) ÷ cos c (81)
where the independent variable spans the range 0 s x · · and the parameter c is small: 0 s c ·· 1.
Note that the solution oI this problem is almost trivial in the special case oI c÷0 (Iirstorder linear
ODE)
v(x;0) ÷ e
x
but obtaining a solution when c = 0 is more challenging (because the ODE becomes nonlinear).
The general idea behind a perturbation expansion is to 'perturb¨ the easily obtained solution
away Irom c÷0 by seeking a solution in the Iorm oI a Taylor series expansion about c÷0:
0
1
2
2
2
2
0
0
; ;
1
; ; 0
2!
v x
v x
v x
v x v x
v x v x
c=
c=
c c c c
c = + c + c +
cc
cc
OI course, this assumes that v(x,c) is 'analytic¨ about c÷0.
+
It remains to be seen iI the solution
has this property or not. Renaming the unknown coeIIicients (i.e. the partial derivates), we look
Ior a solution having the Iorm
v(x,c) ÷ v
0
(x) ¹ v
1
(x)c ¹ v
2
(x)c
2
¹ ... (82)
Note that the coeIIicients v
0
, v
1
, v
2
. are not Iunctions oI the parameter c. Substituting (82) into
(80):
(v´
0
¹ v´
1
c ¹ v´
2
c
2
¹ ...) ¹ (v
0
¹ v
1
c ¹ v
2
c
2
¹ ...) ¹ c(v
0
¹ v
1
c ¹ v
2
c
2
¹ ...)
2
÷ 0 (83)
+
'Analytic¨ means that these partial derivatives exist and that the series converges to v(x,c).
06703 111 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Next we expand the squared sum by distributing each term oI the Iirst series over each term in
the second series:
(v
0
¹ v
1
c ¹ v
2
c
2
¹ ...)
2
÷ (v
0
¹ v
1
c ¹ v
2
c
2
¹ ...)(v
0
¹ v
1
c ¹ v
2
c
2
¹ ...)
2
0 1 2
2
0 1 2
2 2
0 0 1 0 2
2 2
0 1 1
2
0 2
2 2 2
0 0 1 1 0 2
2 2
v v v
v v v
v v v v v
v v v
v v
v v v v v v
+ c + c +
+ c + c +
+ c + c +
c + c +
c +
+ c + + c +
Substituting this result Ior the squared sum in to (83) and collecting terms oI like power in c:
Solution continues . and will be completed in the next revision oI the notes.
Eirst, let`s write the equation in dimensionless Iorm: we will denote the dimensionless variables
using an asterisk:
v
v
r
r
* * * * Re ÷ ÷ V ÷ V ÷
÷
÷ =
·
U R
R p
p p
U R
UR
µ
c
µ
µ
The NavierStokes equations Ior steady Ilow become:
cv v v * * * * * *
V = V ÷V
p
We then look Ior a solution having the Iorm oI a Taylor series expansion about c ÷ 0:
v*(r*,u;c) ÷ v
0
(r*,u) ¹ c v
1
(r*,u) ¹ c
v
2
(r*,u) ¹ ...
Similarly Ior the pressure proIile. Substituting this inIinite series into the NavierStokes equation
(dropping the *`s)
c c cV cV cV v v v v v v
0 1 0 1
2
0
2
1 0 1
+ + V + + = V + + ÷ V + +
,
p p
0 v v v v c c c c
0
0 0
1 2
0 0
0 2
1 1
1
+ V + = V ÷ V + V ÷ V +
, ,
p p
Next we bring all terms to the righthand side oI the equation. We then collect terms oI like
order in the small parameter (i.e. terms which are multiplied by c raised to the same power). To
06703 112 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
obtain zero Ior the sum Ior all values oI c we cannot rely on cancellation oI positive and negative
terms.
+
Instead, the coeIIicient oI each c
n
must vanish separately. This leaves the Iollowing
equations:
c
0
: 0 v = V ÷ V
p (84)
c
1
: v v v
V = V ÷ Vp (85)
and so on ... Note that (84) is just Stokes equation. The solution Ior v
0
can then be substituted
into (85) leaving a linear equation to be solved Ior v
1
and p
1
.
This is the approach used by Whitehead (1889).
v
In many problems, this procedure works.
UnIortunately, the solution Ior the higher order terms in the current problem cannot satisIy the
boundary conditions. This result is known as Whitehead`s Paradox.
+
Another method must be
used.
As an alternative, Oseen (1910) used an entirely diIIerent approach. He approximated the
inertial terms and solved:
µU
Vv ÷ µV
2
v  Vp
thus obtaining: F ÷ 6tµRU1 ¹ (3/8)Re
In principle, one could reIine the solution Iurther by substituting the resulting solution Ior v in
place oI U and then resolving Ior an improved v. In practice, although the Oseen equations are
linear, their solution is suIIiciently diIIicult that no second approximations are known.
In 1957, Proudman & Pearson obtained the next order correction using a diIIerent technique
called matchedasymptotic expansions. In this technique a diIIerent Iorm Ior the expansion is
sought near the sphere (which is called the 'inner expansion¨):
r~R v
i
÷ v
0
i
(r,u)¹v
1
i
(r,u)(Re)¹v
2
i
(r,u)(Re)
2
¹...
+
II the sum oI diIIerent Iunctions oI c vanished Ior one particular value oI c (as a result oI
cancellation oI positive and negative terms), then this same sum oI Iunctions evaluated at a
diIIerent value oI c would not necessarily vanish. The only way we can guarantee that the sum
vanishes Ior every value oI c is to make every term vanish Ior value oI c. See also Iootnote on
page 27 which concerns integrals rather than sums.
v
AlIred North Whitehead (18611947), English mathematician and philosopher, who
collaborated with Bertrand Russell on Principia Mathematica (191013) and, Irom the mid
1920s, taught at Harvard University and developed a comprehensive metaphysical theory.
+
See Van Dyke, p1523.
06703 113 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
and Iar Irom the sphere (which is called the 'outer expansion¨):
r~~R v
o
÷ v
0
o
(µ,u)¹v
1
o
(µ,u)(Re)¹v
2
o
(µ,u)(Re)
2
¹...
where µ ÷ rRe
The inner expansion is identical with Whitehead`s which can be made to satisIy the no slip
condition at r÷R but the result does not have the correct Iorm Iar Irom the sphere. Rather, the
outer expansion is made to satisIy the boundary condition Iar Irom the sphere and then
appropriately match with the inner solution to determine the remaining integration constants.
The result is:
F ÷ 6tµRU1 ¹ (3/8)Re ¹ (9/40)Re
2
lnRe ¹ ...
The term inside square brackets can be thought oI as a correction to Stokes equation:
Re÷ 0.01 0.1 1.0
... ÷ 1.004 1.032 1.375
Comparing the bracketed term with unity gives some idea oI the error incurred by neglecting
inertia.
FLOW AROUND CYLINDER AS RE 0
Now let`s look at the analogous problem oI uniIorm Ilow normal to a cylinder at very low
Reynolds number. II we drop the inertial terms in the NavierStokes equation, we obtain:
µV
2
v ÷ Vp
V
v ÷ 0
r÷· v ÷ U
r÷R v ÷ 0
Eor Ilow normal to the cylinder, we expect the velocity proIile to
correspond to 2D Ilow:
v
:
÷ 0 and c/c: ÷ 0
So that a solution can be Iound using the stream Iunction:
v ÷ V×¢(r,u)e
:

Eor slow enough Ilows, we expect that inertial terms can be neglected, leaving Stokes Equation
(66). AIter the pressure is eliminated (by taking the curl oI Stokes equation), we obtain a single
equation in the unknown velocity proIile :
06703 114 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
recall (68): curl
v 0 = ÷ curl
4
(¢e
:
) ÷ V
2
(V
2
¢)e
:
The boundary conditions are determined in a manner similar to those Ior a sphere (see Hwk #7,
Prob. 1):
r÷· ¢ ÷ Ursinu
r÷R c¢/cr ÷ c¢/cu ÷ 0
Based on the b.c. at r÷· (which is the only nonhomogeneous part oI the problem), the stream
Iunction should have the Iorm:
¢(r,u) ÷ f(r)sinu
Requiring V
2
(V
2
¢) ÷ 0
generates an ODE Ior f(r) whose general solution can be obtained. UnIortunately, none oI the
particular solutions can satisIy all oI the b.c.`s (see HWK #7, Prob. 1). This is known as:
Stokes Paradox (1850)  Stokes equation Ior uniIorm Ilow normal to a cylinder has no
solution.
As it turns, the inertial terms dropped by Stokes are not entirely negligible  no matter how small
Re is. Lamb (1911)
+
obtained a solution Ior the circular cylinder as Re÷0 using Oseen`s
approximation:
µU
Vv ÷ µV
2
v  Vp
V
v ÷ 0
Lamb`s solution Ior the drag Iorce per unit length oI cylinder is:
F
L
U
=
÷ ÷
4
4
1
2
tµ
¸ ln
Re
where ¸ ÷ 0.577... is Euler`s constant. Notice that, unlike Stokes` solution Ior the sphere, Re
appears explicitly in this result. No matter how small Re, this logarithm term in the denominator
is never negligible.
+
Sir Horace Lamb (18391934), English mathematician who contributed to the Iield oI
mathematical physics. He wrote Hvdrodvnamics (1895) which was, Ior many years, the standard
work on hydrodynamics. His many papers, principally on applied mathematics, detailed his
researches on wave propagation, electrical induction, earthquake tremors, and the theory oI tides
and waves
06703 115 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
BoundaryLayer Approximation
FLOW AROUND CYLINDER AS Re
µ µV v v v
V = ÷V +
inertia viscous
p
2
We have just seen that restricting attention to the limiting case oI very small Reynolds
number allows an analytical solution to the NavierStokes equation by neglecting or
approximating the inertial terms. Similarly, we might expect that, in the opposite limit oI very
large Reynolds number, we might be able to obtain an approximate solution by neglecting or
approximating the viscous terms.
Let's return to the problem oI Ilow normal to a cylinder, but at very large Re. II we just drop
the viscous terms Irom the NavierStokes equation, we get Euler's equation Ior an ideal fluid.
Recall that a solution which satisIies the diIIerential equation and the boundary condition Iar
Irom the cylinder is potential flow. Erom HWK #4, Prob. 4a:
v r U
R
r
r
, cos u u = ÷
,
1
2
v r U
R
r
u
u u , sin = ÷ +
,
1
2
At r÷R v R
, u = 0
and v R U
u
u u , sin = ÷2
which violates the noslip boundary condition. An exact
solution oI the NSE's can be obtained numerically.
Comparing the solution Ior Re~~1 to the potential Ilow
solution (see Iigure at right), we see that the exact
solution Iollows potential Ilow everywhere except in a
narrow region near the surIace oI the cylinder where the
exact solution turns downward in order to satisIy the noslip condition.
In the potential Ilow solution the velocity gradient goes like
outside b.l.:
c
c
u
v
r
U
R
~
06703 116 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
This is also the behavior oI the exact solution outside the boundary layer. By contrast, near the
surIace (inside the boundary layer) as the Reynolds number increases, the velocity gradient gets
steeper. A closer analysis (which we will perIorm in a Iew lectures) reveals
inside b.l.:
Re
lim Re
r R
v U
r R
u
÷·
=
¦ ¹ c
~
´ `
c
¹ )
The thickness oI this region in which the two solutions diIIer decreases as the Re gets larger.
This region is known as the:
boundary layer: a very thin region near to a boundary in which the solution has a gradient
which is orders oI magnitude larger than its characteristic value outside the
region.
MATHEMATICAL NATURE OF BOUNDARY LAYERS
Boundary layers arise in solutions oI diIIerential equations in which the highest order
derivative is multiplied by a small parameter. To illustrate the mathematical singularity which
results, consider a simple example:
Example: Iind asymptotic solution to the Iollowing problem as c ÷ 0:
cv" ¹ v´ ¹ v ÷ 0
subject to: v(0) ÷ 0
v(1) ÷ 1
Problem is to Iind the asymptotic behavior oI v(x) as c÷ 0. This problem was presented
Prandtl
+
(the Iather oI boundarylayer analysis) to his class on Iluid mechanics at Goettingen U.
during the winter semester oI 1931/2.
Solution: Eor c suIIicient small, you might guess that the Iirst term can be neglected, leaving
v´ ¹ v ÷ 0
whose general solution is: v(x) ÷ Aexp(x)
+
Ludwig Prandtl (18751953), German physicist who is considered to be the Iather oI
aerodynamics. His discovery (1904) oI the boundary layer, which adjoins the surIace oI a body
moving in air or water, led to an understanding oI skin Iriction drag and oI the way in which
streamlining reduces the drag oI airplane wings and other moving bodies.
06703 117 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Now A can be chosen to satisIy either oI the boundary conditions, but not both. To satisIy
v(0) ÷ 0, we must choose A÷0 which does not satisIy v(1) ÷ 1:
v(0) ÷ 0 ÷ A÷0 ÷ y(x) ÷ 0 Ior all x ÷ v(1) ÷ 0 = 1
On the other hand, iI we choose A÷e to satisIy v(1) ÷ 1, then we cannot satisIy v(0) ÷ 0:
v(1) ÷ 1 ÷ A÷e ÷ v(x) ÷ exp(1x) Ior all x ÷ v(0) ÷ e = 0
The reason we can`t satisIy both boundary conditions with this approximation is that, by
neglecting the Iirst term, the order oI the diIIerential equation reduced Irom 2 to 1. With a Iirst
order O.D.E., we can only satisIy one boundary condition. Thus c÷0 is singularly different Irom
c being arbitrarily small, but not identically zero.
c÷0 ÷ O.D.E. is 1st order
c÷ 0 ÷ O.D.E. is 2nd order
Now the exact solution to this linear O.D.E. with constant coeIIicients is easily determined:
v x
x
x
( ; ) exp
sinh
sinh
c
c
c
c
c
c
=
÷
÷
÷
1
2
2
1 4
1
2
1 4
Comparing the exact solution to that obtained by neglecting the term containing the small
parameter Ior the case oI c÷0.05, we see that the approximation is good except near x÷0. Eor
smaller c, the region in which the exact and approximate solutions diIIer shrinks. To see what's
happening to cause this problem, let's take advantage oI knowning the exact solution and deduce
its asymptotic behavior inside the boundary layer. In particular, let's look at the initial slope:
06703 118 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
' =
÷
÷
v ( ; )
exp
sinh
0
1 4
2
1
2
1
2
1 4
c
c
c
c
c
c
Now consider the limiting behavior as c÷ 0.
1 4 1 ÷ ÷ c
and 1/2c ÷ ·
lim ( ; ) lim
exp
sinh
c c
c
c
c
c
÷ ÷
' =
,
÷
·
· 0 0
0
1
2
1
2
1
2
v
Clearly this limit is highly indeterminant as both exp and sinh 'blow up¨ Iairly quickly as their
arguments become large. This indeterminancy cannot be resolved with the help oI L`Hôpital's
rule because, no matter how many times you diIIerentiate exp or sinh, they still 'blow up.¨ The
indeterminancy can be resolved instead by comparing the asymptotic behavior oI sinh with that
oI exp:
Recall the deIinition oI sinh: sinh :
e e
e
: :
:
=
÷
~
÷
2
1
2
as :÷ ¹·
so
exp
sinh
:
:
( )
( )
÷ 2 as :÷ ¹·
As : ÷ ¹·, the second term oI this deIinition becomes negligible compared to the Iirst. Thus the
ratio oI exp to sinh approaches 2 and our expression above becomes:
v´(0;c÷ 0) ~ 1/c ÷ ·
So the derivative oI our Iunction at this boundary is very strongly dependent on the value oI the
small parameter. In particular, the derivative is not bounded in the limit c÷ 0. This singularity
is the essential nature oI any boundary layer. In the above analysis, exp/sinh was bounded
although both Iunctions become unbounded as their arguments becomes large. We say that the
singularity oI these two Iunctions in this limit is of the same order and we write this as:
sinh(:) ÷ O(e
:
) as :÷ ·
06703 119 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
which is called the Big "Oh" notation (BachmanLandau
+
). More generally, when we say that
f(x) ÷ Og(x) as x÷ a
we mean f(x)/g(x) is bounded as x÷ a
Likewise, we could say that: v´(0,c) ÷ O(c
1
) as c÷ 0
We could also show that v"(0,c) ÷ O(c
2
) as c÷ 0
which means that the second derivative blows up even Iaster the the Iirst. This notation gives us
a convenient way oI describing the strength oI a pole.
at x÷0: c
c c
'' + ' + =
v v v
O O
1 1 0
0
¸
(86)
Thus at the boundary, the Iirst term oI the O.D.E. is the same order oI magnitude as the second
term. So no matter how small c becomes, the Iirst term can never be neglected. This is the root
oI the problem. One way to obtain the correct order in c oI the solution using a Taylor series
expansion would be to transIorm the independent variable:
Let X ÷ x/c and Y(X; c) ÷ v(x; c)
Then ' = = =
÷
v
dv
dx
dY
dX
dX
dx
dY
dX
c
c
1
1
¸
and '' = =
=
=
=
÷ ÷ ÷
v
d v
dx
d
dx
dv
dx
d
dx
dY
dX
d
dX
dY
dX
dX
dx
d Y
dX
2
2
1 1 2
2
2
1
c c c
c
¸
Note that iI dY/dX and d
2
Y/dX
2
are O(c
0
) as c ÷ 0, we will obtain the correct order Ior v´ and v".
This is the basic idea behind 'stretch transIormation¨ which is part oI the 'inner expansion¨
which will presented in the second halI oI the next section.
+
This is the German mathematician Edmund Landau, not the Russian physist Lev Davidovich
Landau.
06703 120 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
MATCHEDASYMPTOTIC EXPANSIONS
'
Matchedasymptotic expansions is a very general technique Ior coping with singularities like
boundary layers. MAE is one type oI singular perturbation expansion. Greenberg describes it
as 'one oI the most important advances in applied mathematics in this century.¨ Although the
beginnings go back to the 19th century and such names as Lindstedt and Poincare,
+
it was not
until the 1960's that singular perturbation techniques became part oI the standard analytical tools
oI engineers, scientists and mathematicians. Since boundary layers Irequently arise in transport
phenomena, let`s apply this technique to solve the simple problem posed by Prandtl.
EXAMPLE: Use Matched Asymptotic Expansions to Iind the asymptotic behavior oI the
solution v(x;c) to the Iollowing problem as c÷ 0.
cv" ¹ v´ ¹ v ÷0 (87)
subject to: v(0) ÷ 0
v(1) ÷ 1
Solution: Eollowing Prandtl (1905), we divide the domain into two regions:
inner region: 0 s x s o, v ÷ v
i
satisIies inner b.c.
outer region: o s x s 1 v ÷ v
o
satisIies outer b.c.
where o is the thickness oI the boundary layer located near x÷0. Within each region, we seek a
solution which is a Taylor series expansion oI the Iunction v(x,c) about c÷0:
0
1
2
2
2
2
0
0
; ;
1
; ; 0
2!
v x
v x
v x
v x v x
v x v x
c=
c=
c c c c
c = + c + c +
cc
cc
 The outer problem. The solution outside the boundary layer can oIten be Iound as a regular
perturbation expansion:
v
o
(x,c) ÷ v
0
(x) ¹ v
1
(x)c ¹ v
2
(x)c
2
¹ ... (88)
*
The example problem introduced above was solved using MAE`s by Greenberg, p508II.
+
(Jules) Henri Poincare (18541912), Erench mathematician, theoretical astronomer, and philo
sopher oI science who inIluenced cosmogony, relativity, and topology and was a giIted
interpreter oI science to a wide public
06703 121 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
which is just a Taylor series expansion about c÷0. (88) into (87):
c(v
0
"¹cv
1
"¹...) ¹ (v
0
´¹cv
1
´¹...) ¹ (v
0
¹cv
1
¹...) ÷ 0
Collecting terms oI like power in c:
(v
0
´¹v
0
)c
0
¹ (v
0
"¹v
1
´¹v
1
)c
1
¹ ... ÷ 0
In order Ior this sum to vanish Ior all values oI c, each coeIIicient must separately vanish:
c
0
v
0
´ ¹ v
0
÷ 0 (89)
c
1
v
1
´ ¹ v
1
÷ v
0
" (90)
and similarly Ior higher order terms. Note that we have succeeded in obtaining a set oI O.D.E.'s
Ior the set oI coeIIicient Iunctions whose solution can be easily uncoupled. II we start with (89),
v
0
(x) can be determined so that it is known when we solve (90). The outer solution must be
subject to the outer boundary condition:
v(1) ÷ 1
Expanding this in a Taylor series about c÷0:
v
0
(1) ¹ v
1
(1)c ¹ v
2
(1)c
2
¹ ... ÷ (1)c
0
¹ (0)c
1
¹ (0)c
2
¹ ...
Thus v
0
(1) ÷ 1 (91)
v
1
(1) ÷ 0
and so on. The solution to (89) subject to (91) is:
v
0
(x) ÷ exp(1x) (92)
We could now substitute (92) into (90) and obtain the solution Ior v
1
(x). Similarly, we could
obtain v
2
(x), v
3
(x) and so on. II we stop aIter the leading term, the outer solution becomes:
v
o
÷ v
0
(x) ¹ O(c)
v
o
(x,c) ÷ exp(1x) ¹ O(c) as c÷0 (93)
 The inner problem. To cope with the boundarylayer, we transIorm the independent
variable using a stretch transformation, whose general Iorm is
X
x x
n
=
÷
0
c
(n~0)
where x
0
is the location oI the boundary at which the boundary layer arises. In this
transIormation the distance Irom the boundary (xx
0
) is magniIied ('stretched¨) by an amount
06703 122 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
depending on the small parameter. In our problem, the boundary layer arises at x
0
÷0, so the
transIormation becomes
X
x
n
=
c
Rewriting (87) in terms oI the new variables X and Y(X) ÷ v(x):
' = = =
÷
v
dv
dx
dY
dX
dX
dx
Y
n
c
Similarly '' =
v Y
c
where
Y denotes dY/dX
Substituting into (87): c c
1 2
0
÷ ÷
+ + =
n n
Y Y Y
(94)
The purpose oI the stretch transIormation is to make Y Y Y ,
,
all O(c
0
):
as c÷0 (X÷const): Y Y Y ,
,
÷ O(c
0
)
whereas v,v´,v" ÷ O(c
0
),O(c
n
),O(c
2n
)
Now we are in a position to choose a value Ior n. Recall Irom the exact solution see (86) that
the secondderivative term is not negligible inside the boundary layer. This is generally the most
important consideration in choosing n: we select the value oI the parameter n such that we do not
loose the term containing the highest order derivative (aIterall, dropping the highest order
derivative is what lead to the outer expansion, which we have shown Iails in the inner region).
To keep the highest order derivative,
This term must be lowest order
+
in c (otherwise it will be swamped by a lower order term)
This term must not be the only term which has this order (iI this is the only term oI that order,
O.D.E. requires it to vanish identically)
As a Iirst attempt, we might try to make all terms oI the same order:
12n ÷ n ÷ 0
which is impossible Ior any single value oI n. Next, we might try balancing Iirst and third:
+
The 'order¨ oI a term with respect to a small parameter c (as opposed to the order oI the
derivative) reIers to the exponent (power) to which c is raised. Eor example, we say that a term
which tends to vanish like c
n
as c÷0 is 'oI order n.¨
06703 123 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
12n ÷ 0 or n÷1/2
Orders (i.e. the exponent oI c) oI the three terms then are:
0, 1/2, 0
This is no good, because highest order derivative (i.e. the Iirst term) is not lowest order in c.
Einally we try to balance Iirst and second term:
12n ÷ n, or n÷1
Now the orders oI each term is: 1, 1, 0
which is OK. Using n÷1, (94) becomes (aIter multiplying by c):
Y Y Y + + = c 0 (95)
Instead oI (88), we seek a solution inside the boundary layer which has the Iollowing Iorm:
Ior xso v
i
(x,c) ÷ Y(X,c) ÷ Y
0
(X) ¹ Y
1
(X)c ¹ Y
2
(X)c
2
¹ ... (96)
(96) into (95) and collecting term oI like order in c:
c
0
Y Y
0 0
0 + =
whose general solution is: Y
0
(X) ÷ A ¹ Bexp(X)
Applying the inner boundary condition: Y
0
(0) ÷ 0
we can evaluate B ÷ A. This leaves us with
Y
0
(X) ÷ A1exp(X) (97)
Similarly, we could determine Y
1
(X), Y
2
(X) and so on. II we stop aIter the leading term, we have
Ior our inner solution:
v
i
÷ A1exp(X) ¹ O(c)
This remaining integration constant A must be chosen so as to match the inner and outer
solutions. One possible choice is to take the outer limit oI the inner solution denoted (v
i
)
o
 and
equate it with the inner limit oI the outer solution denoted (v
o
)
i
:
lim ( , ) lim ( , )
X
i
v
x
o
v
v X v x
i
o
o
i
÷· ÷
= c c
0
06703 124 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
which is called the Primitive Matching Principle and was originally used by Prandtl. Taking the
limit oI (93) as x÷0 and the limit oI (97) as X÷·:
A ÷ e
The inner expansion becomes: v
i
÷ e1exp(X) ¹ O(c)
which means that Ior xso: v ~ e1exp(x/c)
whereas Ior x>o: v ~ exp(1x)
A convenient choice oI o is where v
i
and v
o
intersect. OI course, this
intersection point depends on c:
o ÷ o(c)
which is O(c) in this problem. Erom
the Iigure at right, neither v
i
nor v
o
is
a good approximation to v in the
vicinity oI o. However, in most
transport problems, the quantity oI
greatest interest is dv/dx at x÷0 and
dv
i
/dx does seem to match dv/dx at
x÷0 quite well. II required, v
i
and v
o
can be blended together to obtain a
single smooth Iunction over the entire domain:
v
c
÷ v
i
¹ v
o
 (v
i
)
o
which is called the composite solution. Eor the present example, this is obtained by adding (92)
and (97), expressing them in the same independent variable, and subtracting e:
v e e e e e e e
c x x x x
= + ÷ ÷ = ÷
÷ ÷ ÷ ÷ 1
1
/ / c c
, ,
÷ 0.05
06703 125 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
The plot at right compares v
c
(the dotted curve), and
the exact v (the solid curve). Note that v
c
is a
reasonably good approximation to v. The agreement
gets much better as c ÷ 0. Better agreement could
be obtained Ior any c by including the next order term
in the expansions. Greenberg (p508I) obtains the
second term in both the inner and outer expansions;
the corresponding composite solution Ior c ÷ 0.05 is
virtually indistinguishable Irom the exact solution.
MAE`S APPLIED TO 2D FLOW AROUND
CYLINDER
Let`s now try to apply MAE`s to solve the problem oI
Ilow around a cylinder at high Reynolds number. Eirst, let`s
write the equation in dimensionless Iorm: we will denote the
dimensionless variables using an asterisk:
v
v
r
r
* * *
* Re
÷ ÷ V ÷ V
÷
÷
÷ =
· ÷
U R
R
p
p p
U
UR
µ
c
µ
µ
2
1
(98)
This nondimensionalizing diIIers Irom our previous attempt which was Ior the opposite limit oI
small Reynolds number (see p111 oI Notes). Eirst, we have used µU
2
to nondimensionalize the
pressure instead oI µU/R. This is because the disturbance to pressure caused by Ilow is
proportional to µU
2
in the potential Ilow solution (see Hwk #4, Prob. 4). The second diIIerence
is that c is deIined as the reciprocal oI the Reynolds numbers, rather than the Reynolds number
itselI. This choice makes c a small parameter in the limit Re÷·.
The NavierStokes equations Ior steady Ilow become:
v v v * * * * * * *
V = ÷V cV
p (99)
and V = * *
v 0 (100)
Boundary conditions are
as r*÷·: v* ÷ e
x
and p* ÷ 0 (101)
at r*÷1: v* ÷ 0 (102)
Expecting a boundary layer to arise at r*÷1 as Re÷· (or as c÷0), we will now use the technique
oI MAE to solve it:
v
v
06703 126 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Outer Expansion
In the outer region, the regular perturbation expansion in powers oI c uses r* as the position
variable:
v*(r*,u;c) ÷ v
0
(r*,u) ¹ c v
1
(r*,u) ¹ c
v
2
(r*,u) ¹ ...
p*(r*,u;c) ÷ p
0
(r*,u) ¹ c p
1
(r*,u) ¹ c
p
2
(r*,u) ¹ ...
Substituting this perturbation expansion into (99) through (101), collecting terms oI like order in
c, and setting their coeIIicients to zero, produces a series oI wellposed mathematical problems
Ior the coeIIicient Iunctions. The Iirst problem (the only one we will worry about in this
analysis) is (dropping the *`s):
c
0
: v
0
Vv
0
÷ Vp
0
(103)
V
v
0
÷ 0 (104)
The outer expansion is required to satisIy the outer boundary condition:
as r÷·: v
0
÷ e
x
and p
0
÷ 0 (105)
Although we should not generally require the outer expansion to satisIy the inner b.c.`s, when we
later match inner and outer expansions (see Iootnote on page 129), the outer expansion will have
to satisIy:
v
r0
÷ 0 at r ÷ 1 (106)
Note that the viscous term does not appear in this result because the lowest order viscous term is
O(c
1
), whereas other terms are O(c
0
). It turns out that (103) through (106) is the same problem
we previous solved by potential Ilow:
v
0
= V
where (r,u) is chosen to satisIy (104): V
2
 ÷ 0
Eor potential Ilow, V×v
0
÷0 and (103) becomes:
V + = p v
o 0
1
2
2
,
0
which is just Bernoulli's equation Ior the pressure proIile. AIter imposing the outer boundary
conditions in (105) (and v
r
÷0 at r÷1) we get the potential Ilow solution (see Hwk Set #4,
Prob. 4). Eventually, we will need the inner limit oI the outer solution Ior matching:
v r
r
o
= = 1 0 , u
06703 127 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
v r
u
u u
o
= = ÷ 1 2 , sin (107)
p r
o
= =
÷
1
1 4
2
2
,
sin
u
u
(108)
Inner Expansion
Eor the inner expansion, we use a 'stretched¨ radial coordinate:
* 1
n
r
Y
÷
=
c
or r* ÷ 1 ¹ c
n
Y (109)
and rename the tangential coordinate X ÷ u
Unlike our simple example problem in the last section, the stretching parameter n in this problem
is not an integer. More generally, the inner expansion should be a power series in c
n
rather that a
power series in c itselI:
v r u X Y u X Y u X Y
n n
u
u c c c
*
*, ; , , , = + + +
0 1
2
2
(110)
v r v X Y v X Y v X Y
r
n n *
*, ; , , , u c c c = + + +
0 1
2
2
(111)
p r p X Y p X Y p X Y
n n
* *, ; , , , u c c c = + + +
0 1
2
2
(112)
In cylindrical coordinates Ior 2D Ilow, the equation oI continuity (100) becomes:
V = + = * *
*
*
* *
*
*
v
1 1
0
r
r v
r r
v
r
c
c
c
cu
u
,
We construct the Iirst term using (109) through (111):
r v Y v v v Yv v
r
n n n
*
*
= + + + = + + + 1
0 1 0 0 1
c c c
, ,
c
c
c
c
c
c
c
c
c
c
c
c
c
c
c
c
r v
r
Y
r Y
v Yv v
v
Y
v Y
v
Y
v
Y
r
n n
n
*
* *
*
,
¸
= + + + = + + +
+
÷
0 0 1
0
0
0 1 0
06703 128 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
1
1
1
1
0
0
0 1 0
0
0
1 0
r
r v
r
Y
v
Y
v Y
v
Y
v
Y
v
Y
v
v
Y
r
n
Y
n
n
n
*
*
*
*
c
c
c
c
c
c
c
c
c
c
c
c
c
c
c
c
c
c
,
= + + + +
+
= + +
+
÷
÷ +
÷
÷
In the analysis above, we needed to expand 1
1
+
÷
c
n
Y
as a power series in c
n
. This was
accomplished using the Binomial Series, which will be quite useIul in later problems as well:
1 1
1
2
1 2
3
2 3
+
( )
= + +
÷
( )
+
÷
( )
÷
( )
+ x x x x
o
o
o o o o o
! !
which is just a Taylor series expansion about x÷0. The Binomial Series is known to converge
provided x ·1.
Similarly
c
cu
c
c
c
c
c
u
v u
X
u
X
n
*
= + +
0 1
The continuity equation becomes:
c
c
c
c
c
c
c
c
v
Y
v
v
Y
u
X
n 0
0
1 0 0
0
÷
+ + +
+ =
To satisIy 'no slip¨ at the inner boundary, we must require:
at Y÷0: u u
0 1
0 = = = and v v
0 1
0 = = = (113)
Setting the coeIIicients oI each term separately to zero:
c
n
:
c
c
v
Y
0
0 = Ior all Y
v
0
(Y) ÷ const ÷ 0
06703 129 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
which means that v
0
must be constant inside the boundary layer. To satisIy the noslip condition
(113), the constant must be zero. Then v
0
must vanish everywhere.
+
The next term oI the
continuity equation is more useIul:
c
0
: v
v
Y
u
X
0
0
1 0
0
+ + =
c
c
c
c
or
c
c
c
c
u
X
v
Y
0 1
0 + = (114)
Next, we will look at the principle component oI the NSE, which will be the component tangent
to the surIace. Eor steady 2D Ilow, the ucomponent oI (99) is (BSL, p85):
NSE
: v
v
r
v
r
v v v
r r
p
r r
rv
r
r
v
r
v
r
r r
c
c
c
cu
c
cu
c
c
c
c
c
c
c u
c
cu
u u u u u u
+ + = ÷ +
+ +
,
1 1 1 2
2
2
2
Next, we transIorm each term using (109) through (112) with v
0
÷ 0:
NSE
: v
u
Y
u
u
X
p
X
u
Y
O O
O
n
O
n
1
0
0
0 0 1 2
2
0
2
0 0
0
1 2
c
c
c
c
c
c
c
c
c
c c
c
c
¸
+ + = ÷ + + +
÷
(115)
As a general rule, we don't want to lose the highest order derivative inside the boundary layer, so
we want this term to be lowest order in c, but not the only term with this order. The largest
inertial terms are O(c
0
), so we choose n such that 12n ÷ 0:
12n ÷ 0: n =
1
2
(116)
AIter collecting likepower terms, the rcomponent oI (99) is:
NSE
r
:
1
2
0
0 1 2
2 0 1
0
O
O
O
p p
u
Y Y
÷
c
c
c
c c
+ = ÷c ÷ +
c c
+
This result could be used to evaluate the undetermined integration constant in the outer
expansion. Since v
0
represents the leading term in the inner expansion Ior v
r
, this result means
that the outer limit oI the inner expansion Ior v
r
is zero, and this must match with the inner limit
oI the outer expansion. Indeed, our expressions Ior the outer solution already make use oI this
result.
06703 130 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
The lowest order term in this equation is cp
0
/cY. Since there is no other term with this order, we
must require that:
c
1/2
oI NSE
r
:
c
c
p
Y
0
0 =
which integrates to yield: p
0
÷ c(X)
where the integration 'constant¨ c(X) can be evaluated by matching the outer limit (Y÷ ·) oI
this inner solution with the inner limit (r÷ 1) oI the outer solution (108):
p c X
X
0
2
1 4
2
=
( )
=
÷ sin
(117)
We still have two unknowns: u
0
and v
1
. We can Iormulate two equations Irom the above:
c
0
oI (115):
2
0 0 0
0 1
2
4sin cos
u u u
u v X X
X Y
Y
c c c
+ ÷ =
c c
c
(118)
and (114):
c
c
c
c
u
X
v
Y
0 1
0 + = (119)
No slip requires:
at Y÷0 u
0
÷ v
1
÷ 0 (120)
Matching the outer limit oI the inner solution with the inner limit oI the outer solution requires:
as Y÷ · u
0
÷ 2sinX (121)
Boundary Layer 1hickness
Recall the deIinition Ior 'boundary layer¨ Irom page 116:
boundary layer: a very thin region near to a boundary in which the solution has a gradient which
is orders oI magnitude larger than its characteristic value outside the region.
06703 131 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
So how thick is the boundary layer Ior 2D Ilows like the
one we just analyzed? A typical proIile Ior the
tangential component oI velocity inside the boundary
layer is sketched at right. On the scale oI this sketch, the
velocity appears to approach v
o
i
u ,
which is the inner
limit oI the outer solution. Actually, the nearly
horizontal dotted line isn`t quite horizontal: it has a
small negative slope, but its slope is so small compared
to the slope inside the boundary layer, that the outer
solution appears to be Ilat on the scale oI this drawing.
We might deIine the thickness oI the boundary layer as
the distance we have to go away Irom the surIace to reach the apparent plateau. Erom the
geometry oI this sketch this distance, which we denote as o, is approximated using
v
v
v
o
i
i
v
u
u
o
c
c
,
~
=0
(122)
where all oI the quantities in this equation have units. Let`s try to 'scale¨ (122) and solve Ior o.
Recall that the outer solution corresponds to potential Ilow. Erom HWK Set #4, Prob. 4:
v r U
R
r
o
u
u u , sin = +
,
1
2
so that v v r U
o
i
r R
o
u u
u u
,
= =
÷
lim , sin 2
or v U
o
i
u ,
~
In dimensionless quantities, the inner solution is given by (110). Substituting n ÷ 1/2, we have
v r u X Y u X Y
u
u c
*
*, , , = + +
0
1 2
1
c
c
c
c
c
c
c
c
c c
c
c
u u
c
c
v
v
U
R
v
r
U
R
Y
r Y
u X Y u X Y
U
R
u
Y
i
O
= = + +
,
= +
÷
*
* *
, ,
1
2
1
2
1
2
0
0 1
0
¸
¸
Dropping any multiplicative constants, we obtain
06703 132 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
c
c
c
u
v
v
U
R
i
~
÷
1
2
Substituting this result into (122) and solving Ior the boundarylayer thickness o:
U U
R o
c ~
÷
1
2
or o c ~ = R
R
Re
Thus o is proportional to R and inversely proportional to the squareroot oI Reynolds number.
The boundary layer gets ever thinner as the Reynolds numbers increases. This is true Ior all
boundary layers in 2D Ilows.
PRANDTL`S B.L. EQUATIONS FOR 2D FLOWS
Let`s now summarize the mathematical problem which must be solved to obtain the velocity
proIile inside the boundary layer. The mathematical problem is represented by equations (118)
(121). Eor clarity, let`s rewrite these equations using variables having dimensions recall (98),
(110)(112) and (116). Eor example, u
0
is given by (110) and (98) as
*
0
v
u v
U
u
u
= =
while Y is given by (109) and (116) as
1 2
1 2
1 2
* 1 r r R UR r R
Y
R R
÷
  ÷ ÷ µ ÷
= = c =

µ
\ . c
so
1 2
UR dr
dY
R
  µ
=

µ
\ .
Thus
2 2 2
0
2 2
u v R
UR U
Y r
u
c c µ
=
µ
c c
Making similar transIormations to dimensional variables oI each term, then multiplying both
sides oI the equation by
2
U R µ , (118) becomes
2 2
2
4sin cos
r
v v v v U
v
R r R
r
u u u u
c c c µ  
µ + ÷µ = u u

cu c
\ . c
(123)
06703 133 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
(119) becomes
1
0
r
v v
R r
u
c c
+ =
cu c
(124)
(120) becomes v
r
÷ v
÷ 0 at r ÷ R
(121) becomes v
÷ 2Usinu at r ÷ ·
Notice that (123) and (124) are approximations to NSE
and continuity in cylindrical
coordinates.
Although the above equations Ior the velocity
proIile inside the boundary layer were derived Ior
the speciIic geometry oI a circular cylinder, it turns
out that very similar equations are obtained Ior any
2D Ilow, provided we express them in terms oI a
local Cartesian coordinate system (x,v).
 x ÷ arc length measured along the surIace in the
direction oI Ilow
 v ÷ distance Irom the surIace measured along a
normal to the surIace
The more general equations Ior any 2D Ilow are given by
µ
c
c
c
c
µ
c
c
c
c
c
c
v
v
x
v
v
v
v
v
dp
dx
v
x
v
v
x
x
v
x x
x
v
+
= ÷
+ =
2
2
0
0
(125)
where p
0
is the inner limit oI the pressure proIile in potential Ilow, which is given by (117):
p x p r
r R
PF
0
÷
÷
lim , u
Eor boundary conditions, we impose 'no slip¨ at the wall:
at v÷0: v x v x
, , 0 0 0 = =
Remaining integration constants are evaluated by matching the outer limit oI the inner solution
with the inner limit oI the outer solution:
as v÷·: v x v U x
x
, ÷
0
wher U
0
(x) is the inner limit oI the outer solution Ior the tangential component oI velocity:
06703 134 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
U x v r
r R
PF
0
÷
÷
lim ,
u
u
is the inner limit oI the potential Ilow solution.
The the main diIIerence Irom one geometry to another is a diIIerent potential Ilow solution
applies to diIIerent geometries. Erom Bernoulli`s equation, we have
p x U x
0 0
2
2
( )
+
( )
=
µ
const
DiIIerentiating with respect to x and rearranging:
÷ = =
1
0 1
2
0
2
0
0
µ
dp
dx
dU
dx
U
dU
dx
Substituting into (125):
v
v
x
v
v
v
v
v
U
dU
dx
f x
v
x
v
v
x
x
v
x x
x
v
c
c
c
c
v
c
c
c
c
c
c
+ ÷ = =
( )
+ =
2
2
0
0
0
known
(126)
where v ÷ µ/µ. These PDE`s are called Prandtl`s boundary layer equations. Appropriate b.c.`s
include
at v÷0: v
x
÷ v
v
÷ 0
as v÷·: v
x
÷ U
0
(x)
06703 135 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
ALTERNATE METHOD: PRANDTL`S
SCALING THEORY
'
Now we will repeat the analysis oI boundary
layers using Prandtl`s analysis, which is more
intuitive. Consider uniIorm Ilow normal to a
long cylinder. OI course this is 2D Ilow:
v
:
÷ 0, c/c: ÷ 0
At high Reynolds number, we expect the
viscous terms to become negligible everywhere
except inside the boundary layer. Dropping the
viscous terms Irom the problem yields potential
Ilow:
rR~o v ÷ V
where o is the thickness oI the boundary layer.
Prandtl`s analysis oI this problem assumed that:
 outside the b.l., viscous ·· inertia, such that the velocity and pressure proIiles are those
obtained Irom potential Ilow. Recall the potentialIlow solution Irom HWK 4, Prob. 3:
v U
R
r
r
PF
= ÷
,
1
2
cosu Ior rR ~ o
v U
R
r
PF
u
u = ÷ +
,
1
2
sin
 o··R so that Iluid elements inside b.l. don't 'see¨ the curvature oI the cylinder (we call this
'Postulate #1¨).
 inside the b.l., inertia ~ viscous (we call this 'Postulate #2¨).
DeIining a local Cartesian reIerence Irame, the continuity and NavierStokes equations become:
continuity:
c
c
c
c
v
x
v
v
x
v
+ = 0
*
Schlichting, 6th ed., p11721.
06703 136 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
x v
v
x
v
v
v
p
x
v
x
v
v
x
x
v
x x x
c
c
c
c µ
c
c
v
c
c
c
c
+ = ÷ + +
1
2
2
2
2
v v
v
x
v
v
v
p
v
v
x
v
v
x
v
v
v v v
c
c
c
c µ
c
c
v
c
c
c
c
+ = ÷ + +
1
2
2
2
2
where v ÷ µ/µ. Next, Prandtl estimated the order oI magnitude oI each term, hoping that some
can be dropped. This is called scaling. In this estimate, we are not concerned about Iactors oI 2
or 3, but with orders oI magnitude: we will try to guess the asymptotic behavior oI each term in
terms oI the characteristic physical parameters. In particular, in scaling we try to express each
term in the Iorm oI the product oI the physical parameters raised to some power: In this problem
the physical parameters are R, U, v, and o. Thus we will try to express each term as
R
a
U
b
v
c
o
d
Let`s start with the primary (x) component oI velocity. Across the boundary layer, v
x
must vary
between zero at the surIace (noslip) and
at v÷0 v
x
÷ 0
at v~o v
x
÷ v
PF
÷ 2Usinu
which is the potential Ilow solution. The assumption
here is that the outer edge oI the b.l. corresponds to
the inner edge oI the potential Ilow solution.
So v
x
~ U
c
c o
v
v
v
v
U
x x
~ ~
A
A
c
c
c
c
o
o
o
2
2 2
0
0
v
v
v
v
v
U
U
x
x
~
~
÷
÷
~
A
A
Now x is measured along the surIace oI the cylinder.
Thinking oI x as the arc length measured Irom the
Iorward stagnation point, then:
dx ÷ R du
Integrating with x÷0 at u÷t:
x ÷ R(tu)
06703 137 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Thus
c
c
c
cu
u
u
u
v
x
v d
dx
U
R
U
R
x
PF
= ÷ = ÷
~ 2
1
cos
Similarly,
c
c
c
cu
u
u
u
2
2
2
2
2 2
2
2
1 v
x
v d
dx
U
R
U
R
x
PF
= ÷
= ÷
~ sin
Next, we look at the equation oI continuity:
c
c
c
c
v
v
v
x
U
R
v
x
= ÷ ~
Integrating across the b.l.: v
v
v
dv
U
R
v
U
R
v
v
v x
x
= ~ ÷ ~
c
c
u o
c c
2 cos
so
c
c
c
cu
u u o
v
x
Uv
R
d
dx
U
R
v
R
~ ÷
~
÷
2
1
2
cos
¸
Estimating the inertial and viscous terms in the xcomponent oI NS:
inertia: v
x
cv
x
/cx ~ (U)(U/R) ÷ U
2
/R
v
v
cv
x
/cv ~ (Uo/R)(U/o) ÷ U
2
/R
viscous: vc
2
v
x
/cx
2
~ vU/R
2
vc
2
v
x
/cv
2
~ vU/o
2
Now let's summarize our results as to the magnitude oI each term in the xcomponent oI the
NSE:
x v
v
x
v
v
v
p
x
v
x
v
v
x
x
U
R
v
x
U
R
x
U
R
x
U
c
c
c
c µ
c
c
v
c
c
v
c
c
v
v
o
2
2
2
2
1
2
2
2
2
+ = ÷ + +
?
Since o··R (according to 'Postulate #1¨), the Iirst viscous term must be negligible compared to
the second:
o··R vU/R
2
·· vU/o
2
c
2
v
x
/cx
2
·· c
2
v
x
/cv
2
06703 138 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
So we can neglect this term.
Ignoring the negligible term, we have established the Iollowing orders oI magnitude Ior the terms
in the xcomponent oI the NavierStokes equation:
x v
v
x
v
v
v
p
x
v
v
x
x
U
R
v
x
U
R
x
U
c
c
c
c µ
c
c
v
c
c
v
o
2
2
2
1
2
2
+ = ÷ +
?
(127)
Now let's apply Prandtl`s second postulate: Inside the boundary layer, viscous and inertial terms
are oI the same magnitude:
inertia ~ viscous
U
2
/R ~ vU/o
2
This allows us to estimate the thickness oI the boundary layer:
o
2
÷ vR/U ÷ R
2
/(RU/v)
or o ~
R
Re
(128)
where Re ÷ RU/v
is the Reynolds number. Note that this correctly predicts that the boundary layer gets thinner as
o÷ 0 as Re÷ ·
The remaining term in (127) which has not yet been estimated is the pressure gradient. We can
obtain some idea oI its magnitude by looking at the potential Ilow solution, in which the pressure
proIile is given by Bernoulli's equation:
Ior v>o p/µ ¹ v
2
/2 ÷ const
Now the kinetic energy can be decomposed into contributions Irom the x and vcomponents,
whose orders oI magnitude we have already estimated:
v
2
÷ v
x
2
¹ v
v
2
~ v
x
2
Now v
x
2
~U
2
while v
v
2
~(Uo/R)
2
, so v
v
2
··v
x
2
which leaves:
p/µ ~ const  v
x
2
/2
DiIIerentiating: (1/µ)cp/cx ~ v
x
cv
x
/cx ~ U
2
/R (129)
06703 139 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
So this means we cannot neglect the pressure gradient in (127) since it is the same order oI
magnitude as the inertial terms. Scaling oI the terms oI the vcomponent oI the NavierStokes
equation yields:
v v
v
x
v
v
v
p
v
v
x
v
v
x
v
U
R
v
v
U
R
v
U
R
v
U
R
c
c
c
c µ
c
c
v
c
c
v
c
c
o
o
v o
v
o
2
2
2
2
3
1
2
2
2
2
+ = ÷ + +
?
(130)
Substituting Eq. (128) Ior o shows that viscous and inertial terms again have the same
magnitude:
inertia ~
U
R
U
R
R
U
U R
2
2
2
2
3
2
3
2
1
2
o v
v ~ ~
÷
viscous ~ ~ ~
÷
v
o
v
v
v
U
R
U
R
U
R
U R
3
2
3
2
1
2
so inertia ~ viscous
What remains to be determined is the pressure gradient. At most, the pressure gradient in this
equation has the same order oI magnitude as the other terms:
(1/µ)cp/cv s U
2
o/R
2
(131)
Comparing this with the partial derivative in the other direction, by dividing (131) by (129):
(cp/cv)/(cp/cx) s o/R ÷ 0 as Re÷ ·
which means that variations in pressure across the boundary layer are becoming insigniIicant at
large Re. In other words, a good approximation would be to take:
p ÷ p(x)
Since the pressure at the outer edge oI the b.l. must correspond to that just outside, where
potential Ilow occurs, we can calculate this pressure using the Bernoulli's equation and the
potential Ilow solution:
Ior v>o p/µ ¹ v
2
/2 ÷ const (132)
In potential Ilow at r÷R v
r
÷ 0
v
2
(R,u) ÷ v
2
(R,u) ÷ U
o
2
(x) (133)
(133) into (132): p/µ ¹ U
o
2
/2 ÷ const
06703 140 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
(1/µ)cp/cx ¹ U
o
dU
o
/dx ÷ 0 (134)
(134) into (127): v
v
x
v
v
v
v
x
U
dU
dx
x
x
v
x x
c
c
c
c
v
c
c
+ ÷ = =
2
2
0
0
known (135)
The righthand side oI this equation is known because U
0
is the inner limit oI the potentialIlow
solution:
U x v r
r R
PF
0
÷
÷
lim ,
u
u
Together with the continuity equation, we now have two equations in two unknowns:
2 unknowns: v
x
and v
v
2 equations: (135), Continuity
which are known as Prandtl's BoundaryLayer Equations Ior 2D Ilows.
SOLUTION FOR A FLAT PLATE
ReIerence: Schlichting, 6th ed., p12533,
Whitaker p430440.
BeIore we continue with the analysis oI Ilow
around a circular cylinder, let's look at the
simpler problem oI Ilow tangent to a semi
inIinite Ilat plate. The analysis begins by
computing the potentialIlow solution.
 Step 1: Iind potential Ilow solution
II the plate is inIinitesimally thin, the uniIorm velocity proIile is not disturbed:
P.E.: v ÷ U Ior all (x,v)
and p ÷ p
Ior all (x,v)
OI course, this doesn't satisIy 'no slip,¨ on the plate, but then neither did potential Ilow around a
cylinder.
Step 2: apply Prandtl`s b.l. equations
Erom this potentialIlow solution, we can calculate the U
o
appearing in Prandtl's equation:
U
o
(x) ÷ v
x
PF
(x,0) ÷ U (a const.)
06703 141 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
U
o
dU
o
/dx ÷ 0
Eq. (135) becomes: v
x
cv
x
/cx ¹ v
v
cv
x
/cv ÷ vc
2
v
x
/cv
2
cv
x
/cx ¹ cv
v
/cv ÷ 0
Appropriate boundary conditions are:
no slip: v
x
÷ v
v
÷ 0 at v÷0, x~0
and outside the boundary layer, we obtain potential Ilow:
v
x
÷ U as v ÷ ·
Step 3: rewrite the b.l. equations in terms oI the streamIunction
We can contract the two equations into one by using the stream Iunction:
v ÷ V×¢(x,v)k
which automatically satisIies continuity. Written in terms oI the stream Iunction, the problem
becomes:
¢
v
¢
xv
 ¢
x
¢
vv
÷ v¢
vvv
(136)
at v÷0,x~0 ¢
x
÷ ¢
v
÷ 0
as v÷ · ¢ ÷ Uv
where the subscripts denote partial diIIerentiation.
06703 142 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
The development of the laminar boundary layer along a flat plate is visualized by the
hydrogen bubble method. A fine electrode wire is introduced upstream of the flat plate and
a voltage pulse is applied reapeatedly at regular intervals. The boundary layer thickness is
seen to increase with the distance downstream for the leading edge. Photo taken from
Jisualized Flow, Pergamon, New York, p17 (1988).
Elow visualization studies show that the
boundarylayer grows in thickness as you move
downstream Irom the leading edge (see sketch at
right). This can be rationalized by considering
the inverse problem oI a plate moving through a
stagnant Iluid.
Eocus your attention on an intially stationary
Iluid element in the path oI the moving plate.
The longer the Iluid is disturbed by the moving plate, the more time there is Ior momentum to
diIIuse away Irom the plate.
1ime Out: Flow Aext to Suddenly Accelerated Plate
Eirst consider the much simpler problem oI a inIinite
plate suddenly put in motion at t÷0. Initially, the Iluid
and wall are at rest; but at time t÷0 the wall is set in
motion in the xdirection with a steady speed U. The
initial and boundary conditions are:
at t÷0 Ior all v~0: v
x
÷ 0
at v÷0 Ior all t~0: v
x
÷ U
as v÷·: v
x
÷ 0
06703 143 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
The solution can be expected to have the Iorm: v
x
÷ v
x
(v,t), v
v
÷ v
:
÷ 0. The NSE becomes:
c
c
v
c
c
v
t
v
v
x x
=
2
2
where v ÷ µ/µ. The solution to this problem is well known:
*
v v t U
v
t
x
, =
erIc
4v
where erIc
as 0
0 as
erf
q
t
q
q
q
q
q
q
÷ ÷ ' ÷
÷
÷ ·
÷ '
1
2
1 2
0
e d
is the complementary error function; the integral itselI is the error function.
At any Iixed t, the velocity decays monotonically Irom U at v÷0 to zero as v÷·. As t gets
larger, more and more Iluid begins to move; we say 'the motion penetrates deeper into the Iluid.¨
Suppose we wanted to know how Iar Irom the wall we have to go beIore the velocity drops to 1°
oI the wall value.
v v t U
x
= = o, . 0 01 :
,
erIc 0.01
4
x
v t
U t
o
o  
= =

v
\ .
4 4
erI 1 erIc 0.99
t t
o o
v v
   
= ÷ =
 
\ . \ .
Looking up the appropriate value in a table oI error Iunctions, we obtain:
o
v 4
1825
t
= .
or o v = 365 . t (137)
This gives us some idea how Iar the motion 'penetrates¨ into the stagnant Iluid. Notice that the
penetration depth increases with the squareroot oI time.
*
see BSL, p124I.
06703 144 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
1ime In: Boundary Layer on Flat Plate
Now let`s apply this result to our original problem: the boundarylayer next to a semiinIinite
Ilat plate. Eluid Iarther downstream Irom the leading edge has been in contact with the moving
plate longer; and estimate oI the 'contact time¨ is:
t ÷ x/U (138)
(138) into (137): o v
v
· · t
x
U
(139)
 Step 4: solve b.l. equation using 'similarity
transIorm¨
Judging Irom the boundary conditions alone, we
would expect v
x
to vary Irom 0 at v÷0 to U at v÷o
with sort oI a parabolic shape, with o getting
larger as we move downstream. Notice that the
basic shape oI this proIile is not really changing
with x, only the range oI vvalues over which the solution departs Irom potential Ilow is
increasing as we move downstream. This suggests a solution oI the Iorm:
v
x
/U ÷ f ´ (v/o) (140)
Let's deIine a new independent variable which is scaled to the boundarylayer thickness:
Let
v
x
v
x U
x v
o v
q
~ ÷ ,
To get this guess in terms oI the stream Iunction, recall:
v
v
x
=
c¢
c
thus ¢ o q q o q
o q
q
= = ' =
'
=
v x v dv U f d U f
x
Uf
d x v
x v
,
,
,
0 0
substituting o Irom (139): ¢ v q x v Ux f , = (141)
In terms oI f(q), the boundarylayer equations (136) become:
f f " ¹ 2f "´ ÷ 0
subject to: f ÷ f ´ ÷ 0 at q÷0
06703 145 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
f ´ ÷ 1 as q÷ ·
Now we have an ordinary twopoint boundaryvalue problem, which can be easily solved. This
solution was Iirst obtained by Blasius (1908), who was a Ph.D. student oI Prandtl.
Notice that the velocity component normal to the plate (i.e. v
y
) does not vanish Iar Irom the plate:
lim
,
.
v
v
v x v
U Ux ÷·
= =
08604 0
v
Consider a Iluid balance around the shaded
rectangle in the Iigure at right. Owing to the
need to meet the 'no slip¨ condition on the
surIace oI the plane, the Ilowrate out the right
side oI the system is less than the Ilow in the
leIt side. The excess has to go out the top, causing v
v
~ 0 there.
BoundaryLayer 1hickness
The deIinition oI boundary layer thickness is somewhat arbitrary. Although we are tempted
to say that o is that value oI v at which the boundarylayer solution Ior v
x
(v) equals U (potential
Ilow), v
x
approaches U only asymptotically as v÷·; thereIore, this "deIinition" yields the
unhelpIul result that o is ·. There are several ways to assign a more meaningIul Iinite value to
o(x).
06703 146 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
One way to deIine the boundarylayer
thickness, o, is that value oI v at which the
velocity is 99° oI the asymptotic value.
 DeI'n 1: v
x
(x,o
1
) ÷ 0.99U
0
(x)
Eor Ilow over a Ilat plate, this convention
yields:
o
v
1
50 x
x
U
= .
Another way to deIine o was suggested by Nernst who was
concerned with boundary layers which arose in mass
transIer problems. Nernst chose the diIIusion boundary
layer thickness as the thickness oI a hypothetical stagnant
Iilm which has the same diIIusion resistance.
N D
dc
dv
c
v
v
v
=
=
·
= ÷ =
0
0
o
Graphically, this o can be determined by drawing a tangent
to the concentration proIile at the surIace (v÷0). By
analogy, concentration oI mass is like concentration oI
momentum, which is just the Iluid velocity.
II we deIine o Ior momentum transIer in the same way, replacing concentration by v
x
, then:
 DeI'n 2:
0
2
0
0
0
x x
v
U x
v v
v v
=
÷
c A
= =
c A o ÷
Eor Ilow tangent to a Ilat plate, this deIinition yields:
o
v
2
30 x
x
U
= .
Recall that the normal component oI the velocity proIile was nonzero Iar Irom the plate:
lim
,
.
v
v
v x v
U Ux ÷·
= =
08604 0
v
except Iar downstream Irom the leading edge:
06703 147 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
lim
,
lim .
v
x
v
x
v x v
U Ux ÷·
÷·
÷·
=
=
08604 0
v
This means that the streamline moves away
Irom the plate. A third deIinition Ior o is the
distance the streamline is displaced away
Irom the plate.
Suppose that, Iar upstream, a streamline is
given by
x÷· v ÷ a
Thus the streamline is initially a distance a
Irom the xaxis. II downstream Irom the
leading edge oI the plate, the streamline is a distance b Irom the xaxis, then the displacement oI
the streamline is given by:
o
3
÷ ba
To evaluate this displacement, recall (Irom the deIinition oI streamline) that the Ilowrate in the x
direction between v÷0 and the streamline is the same all along the streamline. Thus:
¢ = = =
Q
W
aU v dv
x
b x
0
where W is the width oI the plate (assumed to be arbitrarily wide). Adding o
3
U to both sides:
a U v dv U
x
b
+ = +
o o
3
0
3
(142)
The leIthand side oI this expression can be rewritten as:
a U bU U dv
b
+ = =
o
3
0
(143)
(142) becomes:
U dv v dv U
U U v dv
b
x
b
x
b
0 0
3
3
0
= +
= ÷
o
o
06703 148 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
OI course, the upper limit b will depend on the particular streamline we have chosen. In other
words, diIIerent streamlines are displaced diIIerently depending on how close to the plate they
were initially. Thus the displacement distance o
3
is not unique. However, distant streamlines
are all displaced by the same distance (since the integral converges as b÷·). So let's deIine the
boundarylayer thickness as the displacement oI external streamlines:
 DeI'n 3: o
3
0
U U v dv
x
= ÷
·
Eor Ilow tangent to a Ilat plate, this yields:
o
v
3
172 x
x
U
= .
Notice that all three oI these deIinitions yield a boundarylayer thickness which is proportional to
vx U although the proportionality constant varies considerably.
o
v
x
x
U
· (144)
We showed (see page 132) Ior any 2D Ilow (which this is) that:
o
v
~ =
R R
U Re
(145)
where R is the radius oI the cylinder. Eor noncircular cylinders, R is some characteristic
dimension oI the cross section (e.g. the major or minor axis oI an ellipse). A semiinIinite Ilat
plate is somewhat unusual in that it has no characteristic dimension. However, iI the plate were
Iinite with length L along the direction oI Ilow, it would seem natural to choose L as the
characteristic length. II one can reasonably assume that what happens downstream with a longer
plate does not signiIicantly eIIect the boundary layer thickness Ior a plate oI length L (i.e. 'exit
eIIects¨ don`t propagate upstream). Eor an semiinIinite plate, the same result is obtained by
choose x as the characteristic length; then
o
v
~
x
U
and Re ÷
Ux
v
which is consistent with (144) and with Prandtl's more general result.
Drag on Plate
The net Iorce exerted by the Iluid on the plate is calculated Irom:
06703 149 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
F ÷ )
A
n
Tda ÷ )
A
pnda ¹ )
A
n
t tt tda
Choosing n ÷ ¹j on the upper surIace and n ÷ j on the
lower surIace, the integral involving the pressure vanishes
owing to cancellation oI the contributions Irom the upper
and lower surIaces (the pressure is the same, but n has
opposite direction on the two surIaces). This leaves
F
v
v
dx
x
x
v
x
=
=
2
0
0
µ
c
c
The '2¨ comes Irom addition oI the two contributions Irom the upper and lower surIace.
Evaluating the integral and expressing the result in dimensionless Iorm:
C
F
Wx
U
D
x
÷ =
2
1328
1
2
2
µ
.
Re
where Re ÷ Ux/v. In deIining drag coeIIicient this way, we have departed somewhat Irom the
convention which uses the projected area along the direction oI Ilow (i.e., the area oI the shadow
cast by the object iI the light source were located very Iar upstream). In the case oI a plate this
projected area is zero, so we have used the area oI the plate instead.
SOLUTION FOR A SYMMETRIC CYLINDER
Let`s now return to the problem oI Ilow around a
cylinder at large Reynolds number. We Iollow the same
general steps as we did in solving Ilow tangent to a Ilat
plate:
 Step 1: Iind potential Ilow solution and U
0
(x)
We accomplished this in HWK #4, Prob. 4. The velocity
proIile in the potential Ilow solution is
v U
R
r
v U
R
r
r
= ÷
,
= ÷ +
,
1
1
2
2
cos
sin
u
u
u
 Step 2: apply Prandtl`s b.l. equations (see page 134)
06703 150 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
2
0
0
2
0
x x x
x v
v
x
v v v dU
v v U
x v dx
v
v
v
x v
c c c
+ ÷ v =
c c
c
c
c
+ =
c c
Erom this potentialIlow solution, we can calculate the U
o
appearing in Prandtl's equation:
U x v R U U
x
R
x
R
x
R
0
2 2
¸
= ÷ = =
÷
=
u
t
u u , sin sin
sin sin
where x is the arc length, measured Irom the Iorward stagnation line,
and u is the polar coordinate, which is measured Irom the rear
stagnation line. The two are related by
x R = ÷ t u or u t = ÷
x
R
Prandtl`s boundarylayer equations (126) become
2 2
2
4 sin cos
0
x x x
x v
v
x
v v v U x x
v v
x v R R R
v
v
v
x v
c c c
+ ÷ v =
c c
c
c
c
+ =
c c
Appropriate boundary conditions are:
no slip: v
x
÷ v
v
÷ 0 at v÷0
and outside the boundary layer, we obtain potential Ilow:
v
x
÷ U
0
(x) as v ÷ ·
The main diIIerence between a circular cylinder and the Ilat plate the leIthand side oI the Iirst
equation, which represents a pressure gradient along the direction oI Ilow inside the boundary
layer.
U
dU
dx
dp
dx
0
0
1
0 = ÷ =
µ
 Step 3: rewrite the b.l. equations in terms oI the streamIunction
In terms oI the streamIunction, Prandtl`s boundarylayer equations are:
06703 151 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
0
0
2
4 sin cos
v xv x vv vvv
dU
U
dx
U x x
R R R
¢ ¢ ÷ ¢ ¢ ÷ v¢ =
at v÷0: ¢
x
÷ ¢
v
÷ 0
as v÷ ·: ¢ ÷ U
o
(x)v
Blausius obtained the solution to this problem, which in Iact works Ior cylinders oI much more
general shape not just circular cylinders.
What is required is that the potential Ilow solution must be an odd Iunction oI x:
U
o
(x) ÷ u
1
x ¹ u
3
x
3
¹ u
5
x
5
¹ ...
Then Blausius obtained a solution with the Iorm (see S:154158):
¢(x,v) ÷ (v/u
1
)
1/2
u
1
xf
1
(q) ¹ 4u
3
x
3
f
3
(q) ¹
¹ 6u
5
x
5
f
5
(q;u
1
,u
3
,u
5
) ¹ ...
where q ÷ v(u
1
/v)
1/2
The velocity proIile obtained this way Ior a circular cylinder is sketched below:
This result is quite diIIerent Irom what was obtained with the Ilat plate. With the Ilat plate, the
velocity proIile v
x
as a Iunction oI v had the same shape Ior diIIerent x`s. Indeed the shape at
diIIerent x`s were similar and we were able to use a 'similarity transIorm.¨ Here, Ior Ilow
perpendicular to a circular cylinder, the basic shape changes with x (or o or u). In particular
1
2 1 0 3
÷0°
60°
90°
108.8° ( ) x
U
R
06703 152 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
notice that the initial slope c c v v
x
v
=0
is positive Ior o÷0 and becomes zero Ior o÷108.8°. At
even larger o`s, the initial slope becomes negative.
This turning around oI the velocity proIile coincides with a proIound event called boundary
layer separation. Separation does not occur on the Ilat plate. The main diIIerence in Prandtl`s
boundarylayer equations which causes separation is the Iorm oI the pressure gradient.
BoundaryLayer Separation
To see what boundarylayer separation is and why it comes about, let's Iirst recall the
potential Ilow solution Ior the pressure on the surIace oI the cylinder. Using the velocity proIile
obtained in Hwk #4, Prob. 4, we can calculate the kinetic energy per unit volume at any point in
the Ilow
1
2
1
2
1 2
1
2
2 2 2
2 4
v U
R
r
R
r
= + ÷
+
,
cos u
Then using Bernoulli`s equation:
p v p U
µ µ
+ = +
·
2 2
2 2
we can compute the pressure proIile around the cylinder.
Consider a Iluid element approaching the cylinder
along the stagnation line shown in the sketch at
right. As the Iluid element moves toward the
stagnation point A (u ÷ t and r changes Irom · to
R), the pressure rises to a maximum. Moving away
Irom Point A along the surIace oI the cylinder (r ÷
R and u decreases Irom t to t/2), the Iluid element
now accelerates until reaching its maximum speed
and minimum pressure at Point B and so on.
06703 153 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
The results Ior the pressure changes or kinetic
energy changes during this journey are summarized
on the graph at right. Note that Bernoulli's equation
balances an increase in kinetic energy with a
decrease in pressure and vice versa. So that the sum
oI the two curves equals a constant. Inside the
boundarylayer, the same pressure gradient applies
but we also have viscous dissipation in addition to
kinetic energy. The upshot is:
just outside b.l.: potential Ilow (no irrev. loss oI
energy). The kinetic energy oI the Iluid at B is
just enough to overcome the pressure hill at C.
Eluid elements arrive at C with v÷0 (no kinetic
energy to spare).
inside b.l.: dp/dx same, but viscous dissipation
consumes some oI the kinetic energy, leaving
insuIIicient energy to climb the pressure hill.
Consequently Iluid elements in the boundary layer
stop their Iorward advance at some point beIore reaching C,
which we will label S. Eluid elements between S and C are
driven toward S by Ialling down the pressure hill.
To conserve mass, Iluid must be pushed away Irom the
cylinder at S. This is known as separation oI the boundary
layer.
The xcomponent oI Ilow is toward the separation point on
either side oI S. Thus at v÷0 dv
x
/dv~0 Ior x·x
S
and
dv
x
/dv·0 Ior x~x
S
. At the separation point the derivative
changes sign:
c
c
v
v
x
x x
v
S
=
=
=
0
0
which serves as a convenient way to locate the separation
point in any mathematical solution to the Ilow problem.
Necessary conditions Ior separation include:
1. decelerating Ilow or Dp/Dt~0
2. irreversible losses oI energy
06703 154 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Drag Coefficient and Behavior in the Wake of the Cylinder
Blausius solution does not apply downstream Irom the separation point (i.e x~x
s
). Indeed, the
boundary layer is not thin enough to be described by Prandtl`s equations. Since the behavior on
the downstream side signiIicantly inIlences the drag and since this behavior is not predicted by
Blausius solution, Blausius solution does not correctly predict the drag Iorce. Measurements oI
the drag coeIIicient Ior Ilow normal to a circular cylinder are summarized below.
Several diIIerent regions are apparent in this loglog graph. Some oI these correspond to major
changes in the shape oI the velocity proIile.
Re << 1. Upstream and downstream halves oI streamline are
mirror images. This is what you would expect iI you looked
Ior a streamIunction solution with the Iorm (see Hwk #7, Prob.
2):
¢(r,u) ÷ f(r)sinu
Although inertial terms are never negligible. The measured
drag coeIIicients agree well with the prediction oI Lamb.
As we increase the Reynolds number, inertia becomes more
important. Generally large inertia has a tendency to make
streamlines straight.
1 < Re < ô. Streamlines are no longer symmetric about u÷t/2.
Streamlines are somewhat Iarther Irom the cylinder on the
downstream side.
Drag coeIIicients are signiIicantly smaller than predicted by
Lamb.
Re ÷ 0.038
Re ÷ 1.1
06703 155 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
ô < Re < 4ô. Two stationary vortexes are Iormed in the wake oI
the cylinder  a consequence oI boundarylayer separation.
Re > 4ô. Periodic vortex shedding. Large vortices, the size oI
the cylinder, are created on a periodic basis. These vortices
detach and move downstream. Vortex shedding alternates
between the top and bottom oI the cylinder. Vortexes shed Irom
the top have a vorticity (V×v) with a sign opposite Irom vortexes
shed Irom the bottom.
Re÷140
These vortices persist many cylinder diameters downstream Irom the cylinder. Most oI the
irreversible losses oI energy occur in Iorming these vortices, whose ultimate Iate is to dissipate
their kinetic energy as heat.
Re ~ 4 1ô
5
. Onset oI turbulence in the boundary layer. Point oI separation moves Iurther back
toward rear stagnation point. Drag is signiIicantly reduced  almost a discontinuous drop.
Re ÷ 19
06703 156 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
The Lubrication Approximation
Consider a plate sliding on a lubricating Iilm
past a second stationary surIace. II the distance
separating the two plates is small compared to the
dimensions oI the plate, we can assume Iully
developed Ilow applies through out most oI the oil
Iilm. Then:
Ior h··L: v v U
h v
h
x
=
÷
p ÷ p
0
Ior all x,v
Q v v dv Uh
x
L
= =
0
1
2
t t
µ
xv vx
U
h
= = ÷
F
ULW
h
x
=
µ
Since the pressure is same inside the Iilm as outside the slide, the sliding motion oI two parallel
surIaces produces no lateral component oI Iorce.
F
v
÷ 0
Now suppose the slider is inclined ever so
slightly relative to the stationary plate. We
might guess, that iI o is small enough, the
velocity proIile will not be aIIected. But,
owing to the inclination, h is no longer
independent oI x, so our guess leads to:
Q
W
Uh x f x = =
1
2
(146)
which violates continuity. It turns out that continuity is preserved by a nonzero pressure gradient,
dp/dx, which causes pressuredriven Ilow. Thus even the primary component oI the velocity
proIile is aIIected by this slight inclination. More signiIicantly, it turns out that this inclination
will produce a diIIerent pressure in the Iilm Irom the Iluid outside the slider block which, tends
to push the two surIaces together or apart.
F
v
= 0
06703 157 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Let`s try to estimate this Iorce. It turns out to be no
more diIIicult to obtain the result Ior an arbitrary gap
proIile h(x) (see Iigure at right), since the essential
diIIiculty arises Irom the Iact that h is not constant
with respect to x.
Suppose the thickness oI the gap is everywhere very
small compared to the dimensions oI the slider block.
h(x) ·· L Ior all x
Essentially, this is a geometry with two very diIIerent
length scales characterizing variations in the diIIerent
directions x and v: we expect slow variations with x and rapid variations with v. We will exploit
this diIIerence using a regular perturbation in the ratio oI the two length scales.
We will start by nondimensionalizing the equations oI motion:
Let X
x
L
÷
c
v
Y
h
÷ u
v
U
x
÷
v
c
v
v
v
÷
L is an obvious choice Ior the characteristic value oI x (since 0·x·L) and U is a obvious choice
Ior the characteristic value oI v
x
(since v
x
÷ U at v÷0). Since 0·v·h(x) some characteristic value
h
c
oI the Iilm thickness seems like a logical choice to scale v.* The choices oI characteristic
values Ior v
v
and p are not obvious; so we will postpone a choice Ior now and just denote these
values as v
c
and p
c
.
We seek a solution in the Iorm oI a regular perturbation:
where
c
h
L
o ÷
v x v Uu X Y U u X Y u X Y
, , , , , , o o o = = + +
(147)
0 1
, , , , , ,
v c c
v x v v v X Y v v X Y v X Y o = o = + o + (
¸ ¸
(148)
p x v p p P X Y p P X Y P X Y
, , , , , , o o o ÷ = = + +
(149)
As usual an important aspect oI this Iorm is that all derivatives oI the dimensionless velocity
components u and v with respect to the dimensionless coordinates X and Y are O(o
0
):
* Eor deIiniteness, we might select the largest value oI h(x) to be h
c
.
L
h x ( )
h
v
x
U
06703 158 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
cu/cX, cu/cY, cv/cX and cv/cY are O(o
0
) (150)
The choice oI v
c
becomes apparent when we nondimensionalize the continuity equation:
continuity:
c
c
c
c
v
x
v
v
x
v
+ = 0 (151)
Substituting (147) and (148) into (151):
c
c
v U u v
L X h Y
c c
= ÷
c c
or
1
c
c
v u v
X h L U Y
c c
= ÷
c c
or
c
v u v
X U Y
c c
= ÷
c o c
(151) requires the two terms in the continuity equation to be equal but opposite; and to be exactly
the same order oI o. Since cu/cX and cv/cY are O(o
0
) according to (150), we are Iorced to
choose v
c
such that
c
v U o is O(o
0
). So let`s choose
v
c
÷ oU (152)
Substituting (147)(149) and (152) into (151), the leading term is
c
0
:
0 0
0
u v
X Y
c c
+ =
c c
(153)
Next we examine the principle component oI the NavierStokes equation:
NSE
x
: v
v
x
v
v
v
p
x
v
x
v
v
x
x
v
x x x
c
c
c
c µ
c
c
v
c
c
c
c
+ = ÷ + +
1
2
2
2
2
(154)
Substituting (147)(149) and (152) into (154):
U
L
u
u
X
U
L
v
u
Y
p
L
P
X
U
L
u
X
U
L
u
Y
O O
c
O
O
2 2
2
2
2 2
2
2
0 0
0
2
c
c
o
o
c
c µ
c
c
v
c
c
v
o
c
c
o o
o
o
+ = ÷ + +
?
The last term in the equation is lowestorder in the small parameter o: it`s O(o
2
). All the other
terms in the equation (except possibly Ior the pressure gradient) are O(o
0
). Unless we have some
other term in the equation oI the same order, we will be Iorced to take
c
c
2
2
u
Y
÷ 0, which yields
linear shear Ilow to all orders in o. We already know that this solution violates macroscopic
continuity. To avoid this situation, we choose p
c
so that the pressure gradient term is also oI
O(o
2
):
06703 159 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
p L
U
L
U
L
c
= × = µ v
o
µ
o
µ
µ
2 2
(155)
With this choice, the leading term in NSE
x
is
o
2
:
c
c
c
c
2
0
2
0
u
Y
P
X
= (156)
So Iar we have two equations (153) and (156) in three unknows (u
0
, v
0
and P
0
). We need
another equation. So we turn to the secondary component oI NSE:
NSE
v
: v
v
x
v
v
v
p
v
v
x
v
v
x
v
v
v v v
c
c
c
c µ
c
c
v
c
c
c
c
+ = ÷ + +
1
2
2
2
2
Introducing our dimensionless variables, we have:
o c
c
o
o
c
c
µ
o o
c
c
v
o c
c
v
o
o
c
c
o o
o
o
o
U
L
u
v
X
U
L
v
v
Y
U
L L
P
Y
U
L
v
X
U
L
v
Y
O O
O
O
O
2
2
2 2
2
2 2
2
2
3
1
,
+ = ÷ + +
AIter substituting the perturbation expansions (147)(149), (152) and (155), then collecting
terms, the leading term is:
o
3
:
c
c
P
Y
P Y
0
0
0 = or ÷ const w. r. t. (157)
This conclusion suggests that the pressure gradient in (156) can be treated as a constant with
respect to Y:
c
c
2
0
2
0
u
Y
dP
dX
= ÷ const w.r.t. v (158)
We might be tempted to set this pressure gradient to zero since:
p(0) ÷ p(L) ÷ p
o
OI course the pressure gradient is zero when the two plates are parallel. But we already suspect
that the pressure gradient is needed to avoid linear shear Ilow, which violates continuity. So we
avoid this temptation and leave dP
0
/dX nonzero.
06703 160 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
From this point on, we will resort to the dimensional variables, using what we
have learned from the leading terms of the perturbation expansion. In essence,
we are truncating (147)(149) after the first non:ero term.
Integrating (158) twice: v x v
dp
dx
v c x v c x
x
, = + +
1
2
2
1 2
µ
Boundary conditions are given by the 'no slip¨ requirement:
v
x
÷ 0 at v÷h(x)
v
x
÷ U at v÷0
Evaluating the two integration constants leads to:
v x v U
v
h
dp
dx
v vh
x
,
,
= ÷
+ ÷ 1
1
2
2
linear shear flow pressuredriven flow
µ
The volumetric Ilowrate per unit width oI plate is calculated as:
Q
W
v x v dv
Uh h dp
dx
h
x
= = ÷
0
3
2 12
,
µ
(159)
Q can be made to be a constant at each x iI dp/dx is allowed to take on nonzero values. The
necessary values can calculated Irom (159):
Q/W÷const.:
dp
dx
U
h h
Q
W
= ÷
6 12
2 3
µ µ
(160)
Integrating with respect to x Irom x÷0 where p÷p
0
to any other x.
p x p U
dx
h
Q
W
dx
h
x x
÷ = ÷
0
2
0
3
0
6 12 µ µ (161)
The pressure is the same at the downstream end oI the gap. Then:
p L p U
dx
h
Q
W
dx
h
L L
÷ = = ÷
0
2
0
3
0
0 6 12 µ µ
Knowing the overall pressure drop is zero allows us to compute the Ilowrate:
06703 161 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Q
W
UH
=
2
(162)
where
2 3
0 0
L L
dx dx
H
h h
=
is an average gap width. II h(x) is a linear Iunction whose value changes Irom
h(0) ÷ h
1
to h(L) ÷ h
2
· h
1
then H
h h
h h
=
+
2
1 2
1 2
or
1 1
2
1 1
1 2
H h h
= +
which is called the 'harmonic mean¨ oI h
1
and h
2
. (162) into (160) gives the pressure gradient:
dp
dx
U
h
H
h
= ÷
6
1
2
µ
Note that h
2
· H · h
1
at x÷0: h ÷ h
2
~ H ÷ dp/dx~0
at x÷x
H
h ÷ H ÷ dp/dx÷0
at x÷L h ÷ h
1
· H ÷ dp/dx·0
Thus H represents not only an average gap width (with respect to Ilowrate) but it is also the width
at the point where pressure is a maximum.
Now we are in a position to evaluate the Iorce exerted on the plate by the Iluid. The x
component oI the Iorce is
0 0
0
, 0
L L
x
x vx
v
dv U
F W x dx W dx WL O
dv h
=
µ
= t = µ = + c
where
1 2
h h
L
÷
c ÷
is the angle oI tilt between the two plates (c ·· 1). This result is the same (neglecting the O(c)
correction) as would be obtained Ior two parallel plates. More interesting is the vcomponent.
Eor a linear gap:
06703 162 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
F
W
p x p dx
UL
h h
h
h
h h
h h
v
L
= ÷ =
÷
÷
÷
+
,
( ) ln
0
0
2
1 2
2
1
2
1 2
1 2
6
2
µ
Does this reduce to zero when the two surIaces are parallel (i.e. h
1
÷h
2
)? Although it might
appear that F
v
÷· as h
1
÷h
2
, it turns out that the terms inside the square brackets tend to zero
Iaster than the denominator outside the brackets. In the limiting case that
c··1 h
1
~ h
2
~ h,
and h
1
h
2
·· min(h
1
,h
2
),
then we obtain:
3
1
2
v
F
L
U
W h
 
= µ c

\ .
So we do recover zero Iorce Ior parallel plates. The main diIIerence between two parallel and
two nonparallelplates is the occurance oI the nonzero vcomponent oI liIt which would not occur
Ior parallel plates. Notice that F
v
~0 iI c~0 (h
1
~h
2
) and F
v
·0 iI c·0 (h
1
·h
2
). Thus either
repulsion or attraction oI the two plates is possible, depending on the direction oI tilt relative to
the direction oI Ilow.
TRANSLATION OF A CYLINDER ALONG A
PLATE
The lubrication approximations developed Ior the
slider block can be easily extended to other geometries.
Eor example, instead oI a planar slider block, suppose I
try to drag a cylinder parallel to a plate. What will be
the Iorce tending to push the two surIaces apart? The
same perturbation expansion done with the slider block
applies here, except we have a diIIerent proIile h(x) Ior
the gap between the two surIaces.
To deduce the gap proIile, recall the equation oI a
circle
x x v v R
c c
÷ + ÷ =
2 2 2
where (x
c
, v
c
) is the location oI the center oI the circle
and R is its radius. Substituting the coordinates oI the
center in our problem and v(x) ÷ h(x), we have
x h R h R
2
0
2 2
+ ÷ ÷ =
06703 163 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Dividing by R
2
x
R
h h
R
2
2
0
2
1 1 +
÷
÷
=
Recognizing that hh
0
is very small compared to R, we can use the Binomial Series, truncated
aIter the second term, to obtain an approximation to the second term:
h h
R
h h
R
O
h h
R
÷
÷
= ÷
÷
= ÷
( )
= ÷ + ~ ÷
÷
0
2
0
2
2 2 0
1 1 1 1 2 1 2
c
c c c
The '1¨ cancels with the '1¨ on the righthand side oI the equation, leaving:
h x h
x
R
= +
0
2
2
(163)
provided that hh
0
·· R which requires that x remain small compared to R.
o = = <<
dh
dx
x
R
1
where we have moved x÷0 to the center oI the gap, which is now symmetric about x÷0.
Although (163) is only valid Ior x very small, it turns out virtually all oI the contribution to the
Iorce comes Irom the region where (163) is valid provided that h
0
is suIIiciently small
compared to R. In any case, let's assume that (163) is valid. II that bothers you, then replace the
circular cylinder by a parabola.
As with the Ilat slider, the pressure proIile is
determined by the need to have the volumetric Ilowrate
through any x÷const plane be the same Ior all such
planes. Eq. (159) becomes:
Q
W
Uh h dp
dx
= ÷
2 12
3
µ
÷ const w.r.t. x (164)
II we view Q/W as an unknown integration constant,
then we will need two boundary conditions to evaluate
the two integration constants we will have aIter
integrating this. Since the Iluid held between the
cylinder and the wall is in contact with the same
reservoir at either end oI the gap, we can require:
p ÷ p
0
at x ÷ ·,¹·
06703 164 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
By 'inIinity¨, we simply mean Iar Irom the origin. It might seem more reasonable to speciIy
x÷R, but iI R is suIIiciently large compared to h
0
, no signiIicant error will be incurred by
extending the limit to inIinity. The counterpart to (161) is:
p x p U
dx
h
Q
W
dx
h
x x
÷ =
'
÷
'
·
÷· ÷·
6 12
2 3
µ µ (165)
Applying the other boundary condition at x÷¹· allow us to evaluate Q. The counterpart oI (162)
is
Q
W
UH
=
2
(166)
where H
dx
h
dx
h
=
÷·
·
÷·
·
2
3
is an average gap width.
Substituting (163): H h =
4
3
0
As long as h(x) is an even Iunction oI x, then
p(x) must be odd:
h(x)÷even ÷ p(x)÷odd
Thus the pressure proIile given by (165)
looks as shown at right. The extrema
correspond to:
dp/dx÷0
Subsituting (166) and dp/dx÷0 into (164)
yields:
dp/dx÷0: h H h = =
4
3
0
Substituting (163) and solving Ior x:
dp/dx÷0: x Rh = ±
2
3
0
Because p is an odd Iunction, there will be no normal Iorce tending to separate the two surIaces:
p p
U
h
06703 165 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
p÷odd:
F
W
p x p dx
v
= ÷ =
÷·
·
0
0
0
2
2
x
xv
F R
x dx U
W h
·
÷·
= t = tµ
CAVITATION
However, there is an important phenomenon which we have not discussed which can cause a
liIt Iorce to push the two surIaces apart. That phenomenon is
cavitation  Iormation oI gas bubbles caused by a lowering oI pressure
II the absolute pressure oI the Iluid drops below the vapor pressure oI the liquid, we will have
boiling oI the liquid and cavitation. Because pressures generated in lubrication problems can be
signiIicant compared to atmospheric, cavitation is not an uncommon event.
sources of gas bubbles:
vapor oI liquid (iI p·p
vapor
)
air (iI p·p
saturation
)
Many liquids are kept in contact with air at one atmosphere and thereIore become saturated with
air. II the pressure on the liquid is suddenly lowered, the air will be supersaturated and air
bubbles will Iorm.
What eIIect will cavitation have on the pressure
proIile? Although an exact analysis would require
consideration oI twophase Ilow, we can anticipate
that at the very least the absolute pressure
cannot drop below saturation.
II any oI the negative portion oI the pressure proIile
is chopped oII, the proIile loses its antisymmetry.
A repulsive Iorce pushing the surIaces apart
becomes likely. The resulting proIile might be
expected to look something like that shown at right.
06703 166 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
SQUEEZING FLOW
The above two problems both involve the sliding oI two
surIaces past one another. A related lubrication problem is
the squeezing motion between two bodies. Consider the
'squeezing¨ motion in a thin Iilm oI liquid held between a
circular disk and a parallel Ilat plate in the limit in which
h··R. In the limit c ÷ h/R ÷ 0, a solution to the Navier
Stokes equation can be Iound via a regular pertubation
expansion oI the Iorm:
v r : u u u u u
r c c
, , , , , , c µ , c µ , c µ , = = + +
0 1
v r : Uv U v v
:
, , , , , , c µ , c µ , c µ , = = + +
0 1
p r : p p P p p p
c c
, , , , , , c µ , c µ , c µ , ÷ = = + +
· 0 1
where U
dh
dt
÷ , c ÷
h
R
, µ ÷
r
R
and
: :
h R
, ÷ =
c
Note by using the arbitrary u
c
as our characteristic radial velocity, we are delaying the choice
until we have a chance to inspect the continuity equation:
continuity:
1
0
r
rv
r
v
:
r :
c
c
c
c
+ = (167)
Nondimensionalizing:
u
R
u
U
h
v
c
1
0
µ
c µ
cµ
c
c,
+ =
Dividing by U/h:
0
0
1
1
c
O
O
u
u v
U
c
c
c µ
c c
= ÷
µ cµ c,
(168)
Boundary conditions on v
:
include:
at :÷0 or ,÷0: v
:
÷ 0 or v ÷ 0
at :÷h or ,÷1: v
:
÷ U or v ÷ 1
This means that c c v :
:
(and c c, v ) is not zero; so that the other term in the continuity equation
(167) is not zero either. The only way the two sides oI (168) can have the same order in c is Ior
the coeIIicient to be O(c
0
). This is accomplished by choosing:
F
:
06703 167 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
u
c
÷ c
1
U
Next we examine the principle component oI the NavierStokes equation. Although the Ilow is
caused by motion in the :direction, the rcomponent is much larger (u
c
~~ U Ior c ·· 1).
r: v
v
r
v
v
:
p
r r r
rv
r
v
:
r
r
:
r
f f
r
f
r
c
c
c
c µ
c
c
µ
µ
c
c
c
c
µ
µ
c
c
+ = ÷ +
,
+
1 1
2
2
Here we have added the subscript 'f ¨ on the Iluid density to avoid conIusion with the
dimensionless radial coordinate. Nondimensionalizing:
c
c
cµ
c
c
c
c, µ
c
cµ
µc
µ
c
cµ µ
c µ
cµ
µc
µ c
c
c,
c c
c c
÷
÷ ÷ ÷
+ = ÷ +
,
+
1
2
1 2 1
2
1
2
2
2
2 2
1 3
1
U
R
u
u U
R
v
u p
R
p U
R
u
U
R
u
O O
c
f
f
O
f
O
,
Clearly, the Iirst viscous term and both inertial terms are negligible compared to the second
viscous term. Unless we have some other term in the equation oI the same order, we will be
Iorced to take
c
c,
2
2
u
÷ 0, which (aIter no slip is applied) yields u ÷ 0 Ior all ,. This would
violate continuity. Thus we choose p
c
so that the pressure derivative has the same order as the
second viscous term:
p
R
U
R
c
f
f
µ
µc
µ c
=
÷1
2
or p
U
R
c
=
µ
c
3
Einally, we look at the secondary component oI the NavierStokes equation:
:: v
v
r
v
v
:
p
: r r
r
v
r
v
:
r
:
:
:
f f
:
f
:
c
c
c
c µ
c
c
µ
µ
c
c
c
c
µ
µ
c
c
+ = ÷ +
+
1 1
2
2
Nondimensionalizing:
c
c
cµ c
c
c,
µ
c
µ c
c
c,
µ
µ µ
c
cµ
µ
c
cµ
µ
µ
c
c
c,
c c
c c
c
÷
+ = ÷ +
+
1
2
3
2 2
2
2
1 1
4 0
2
1
U U
R
u
v U
R
v
v
U
R
R
P U
R
v U
R
v
O O
f
O
f
O
f
O
,
Note that c c, P is the lowest order term in this equation. Moreover, it is the only term which is
O(c
4
). ThereIore when the perturbation expansions are substituted, and terms oI like order are
collected, the leading term is
06703 168 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
c
4
:
c
c,
P
0
0 = or P P
0 0
= µ (169)
so that the :dependence oI pressure can be neglected compared to the rdependence. The
resulting problem to solve (going back to dimensional quantities) is:
Continuity:
1
0
r
rv
r
v
:
r :
c
c
c
c
+ = (170)
NSE
r
: µ
c
c
2
2
v
:
dp
dr
r
= ÷ const. w.r.t. : (171)
Since the righthand side is independent oI :, we can integrate immediately to obtain the general
solution:
µv
dp
dr
:
c r : c r
r
= + +
2
1 2
2
Boundary conditions are:
at :÷0 v
r
÷ v
:
÷ 0
at :÷h v
r
÷ 0, v
:
÷ U
Applying the b.c.'s we get:
1
2
r
dp
v : : h
dr
= ÷
µ
(172)
As beIore, dp/dr is determined such that continuity is
satisIied. Now, however, macroscopic continuity
requires:
2
flow in
0
through top
flow out through
walls of cylinder
, 2
h
r
v r : r d: r U t = t
(173)
(172) into (173) and requiring that the result be satisIied
Ior any r yields:
dp
dr
U
r
h
= ÷6
3
µ (174)
06703 169 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Requiring that: p(R) ÷ p
0
we can integrate to obtain the pressure proIile:
p r p U
R r
h
÷ =
÷
0
2 2
3
3µ
which is sketched at right. Substituting (174) into (172):
v r : U
r
h
:
h
:
h
r
, =
÷
,
3
2
(175)
The other component oI velocity can be determined by
applying microscopic continuity. Using (151) and (175):
c
c
c
c
v
: r
rv
r
U
h
:
h
:
h
: r
= ÷ = ÷
÷
,
1 6
2
Integrating subject to v
:
÷ 0 at : ÷ 0:
2 3
, 3 2
:
: :
v r : U
h h
(
   
( = ÷ ÷
 
\ . \ . (
¸ ¸
(176)
To calculate the Iorce exerted by the plate on the Iluid, we
use the unit normal pointing out oI the Iluid: n÷k:
dF ÷ k
T da
Erom the axisymmetry oI the problem, we anticipate that
there will only be a :component oI this Iorce, which we can
calculate by post dotting the above by k:
dF
:
÷ k
T
k da ÷ (p¹t
::
)da
In this problem, the normal component oI the deviatoric stress (t
::
) vanishes. Using (176):
t
::
÷ µ cv
:
/c:,
:÷h
÷ µ(U)6:h
2
 6:
2
h
3
,
:÷h
÷ 0
This leaves just a contribution Irom the pressure. Since p(r,h) is independent oI u, we choose da
÷ (2tr)dr to be a thin annulus oI radius r and thickness dr:
÷ = =
F rp r h dr
UR
h
:
R
2
3
2
0
4
3
t
tµ
( , )
06703 170 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Notice that, Ior a Iixed U, the Iorce required becomes unbounded as h÷ 0.
REYNOLDS EQUATION
Sliding motion and squeezing motion are quite diIIerent. Yet the lubrication approximation
Ior each has a lot in common. In particular, note that in either case, pressure can be taken as as a
constant along a normal to either surIace (compare (157) and (169)). In either case, the dominant
velocity component is tangent to the surIaces and the principle component oI the NSE is
approximated by a balance between viscous shear stress on the surIace with the pressure gradient
along the surIace (compare (158) and (171)). It turns out that the lubrication approximation can
be generalized to handle an arbitrary combination oI squeezing and sliding motion in 3D.
Consider two bodies oI arbitrary (but smooth) shape
moving slowly through a viscous Iluid in the near
vicinity oI each other. A rectangular Cartesian
coordinate system is chosen so that the :axis coincides
with a straight line connecting the two surIaces at the
points oI minimum approach. The origin is located at
some arbitrary point along this line. o
i
represents the
distance (along the :axis) Irom the origin to the surIace
oI body i (i ÷ 1,2), while :÷h
1
(x,v) and :÷h
2
(x,v)
describe a portion oI their surIaces nearest the origin.
Let R
ix
and R
iv
be the radii oI curvature oI body i in the
x and vdirections, respectively.
Eor distances h
i
(x,v) much less than both R
ix
and R
iv
, h
i
can be approximated by
+
h x v
x
R
v
R
i i
ix iv
( , ) = + + o
2 2
2 2
The total distance between the two surIaces is
h x v h x v h x v
x
R
v
R
x v
( , ) ( , ) ( , ) = + = + +
1 2
2 2
2 2
o
where o ÷ o
1
¹ o
2
and R
f
(f ÷ x,v) can be considered to be the radii oI curvature oI the Iilm:
+
Actually, this assumes that the principle radii oI curvature oI both surIaces lie either in the x or
vdirections. Should one oI the surIaces be rotated around the :axis by an angle u, the Iunction
acquires an addition term which is proportional to xv sinu cosu.
o
2
o
1
h x v
2
( , )
h x v
1
( , )
v e e e = + + U J W
x v : 2 2 2
v e e e = + + U J W
x v : 1 1 1
06703 171 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
1 1 1
1 2
R R R
f f f
= +
As in the two previous examples, scaling leads us to the Iollowing equations Ior the velocity
proIile in the Iilm:
µ
c
c
c
c
2
2
v
:
p
x
x
=
µ
c
c
c
c
2
2
v
:
p
v
v
=
where the pressure proIile is independent oI ::
p ÷ p(x,v)
Since pressure is independent oI :, these equations can be easily integrated to yield the velocity
proIile in the Iilm, which again turns out to be the sum oI linear shear Ilow (Irom the sliding
motion) and a parabolic pressuredriven Ilow.
v
p
x
: : h h h h
U
h
: h U
x
= ÷ ÷ ÷ + + +
c
c µ
1
2
2
2 1 1 2 1 1
( ) ( )
A
(177)
v
p
v
: : h h h h
J
h
: h J
v
= ÷ ÷ ÷ + + +
c
c µ
1
2
2
2 1 1 2 1 1
( ) ( )
A
where AU ÷ U
2
U
1
and AJ ÷ J
2
J
1
Still unknown is the pressure proIile, which is Iound by requiring the velocity proIile to satisIy
the continuity equation. In particular, since pressure is independent oI :, we will choose p to
satisIy the integral oI the continuity equation with respect to :
V =
÷
v
d:
h x v t
h x v t
1
2
0
, ,
, ,
Expanding the divergence and separating derivatives with respect to : Irom those with respect to
x and v:
2 2
1 1
0
h h
v
x :
h h
v
v v
d: d:
x v :
÷ ÷
c  
c c
V = + + =

c c c
\ .
v
(178)
Applying Leibnitz` rule Ior diIIerentiating an integral whose limits are Iunctions oI the
diIIerentiation variable:
06703 172 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
2 2
1 1 2 1
, ,
1
2
2 1
, ,
, , , , , ,
h x v h x v
x
x x x
h x v h x v U U
h
v h
v x v : d: d: v x v h v x v h
x x x x
÷ ÷
c ÷
c c c
= + ÷ ÷
c c c c
so
2 2
1 1
2 1
2 1
, ,
h h
x
x
h h
v h h
d: v x v : d: U U
x x x x
÷ ÷
c c c c
= ÷ ÷
c c c c
(179)
Substituting the velocity proIile (177) into (179) and integrating leads to:
2
1
3
1 2
1 2
1 2
2 12
h
x
h
h h
h h p
v d: U U
x
÷
+
+ c
= + ÷
µ c
DiIIerentiating
2
1
3
1 2
1 2
1
2 12
h
x
h
h h h p
v d: U U
x x x x x
÷
 
c c c c c  
= + + ÷ 


c c c c µ c
\ .
\ .
where h ÷ h
1
¹ h
2
(179) becomes
2
1
3
1 1
12 2
h
x
h
v p h
d: h U
x x x x
÷
c c c cA
 
= ÷ ÷ A

c µ c c c
\ .
(180)
where Ah ÷ h
2
 h
1
There is a very similar result Ior the integral oI
v v c c :
2
1
3
1 1
12 2
h
v
h
v
p h
d: h J
v v v v
÷
c
  c c cA
= ÷ ÷ A

c µ c c c
\ .
(181)
Einally the result Ior the integral oI
v : c c :
2 2
2 1 1 1
2 1
, , , ,
h h
:
: : :
h h h h
v
d: dv v x v h v x v h W
:
÷ ÷
c
= = ÷ ÷ = A
c
(182)
Adding (180), (181) and (182), setting to zero (to satisIy (178)) and rearranging:
06703 173 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
c
c
c
c
c
c
c
c
µA
cA
c
µA
cA
c
µA
x
h
p
x v
h
p
v
U
h
x
J
h
v
W
3 3
6 6 12
+
= ÷ ÷ + (183)
which is called Reynolds lubrication equation. The solution to this equation yields the pressure
proIile in the gap Ior any prescribed translation oI the two bodies. Outside the gap, the pressure
must approach the bulk pressure, which is taken to be zero
p ÷ 0 as (x
2
¹v
2
) ÷ ·
In the special case in which upper and lower surIaces are surIaces oI revolution around the same
axis, polar coordinates (r,u) are more convenient than (x,v) since then h
1
÷ h
1
(r) and h
2
÷ h
2
(r).
(183) can be written in invariant vector notation:
3
2 1 2 1
6
s s
h p V V = ÷ µ ÷ ÷ v v n n (184)
where v
i
is the velocity oI body i (i ÷ 1or 2) and n
i
are local normals to body i (not necessarily oI
unit length). In particular n
i
is deIined as:
n
i
÷ Vf
i
where f
1
(x,v,:) ÷ h
1
(x,v) ¹ :
and f
2
(x,v,:) ÷ h
2
(x,v) :
That n
i
are local normals to body i Iollows Irom the Iact that f
1
is a constant on surIace #1
(deIined as : ÷ h
1
) and f
2
is a constant on surIace #2 (deIined as : ÷ ¹h
2
). II we decompose the
velocity oI the two bodies into contributions along the : axis and in the xy plane
v
i
÷ v
si
¹ W
i
e
:
then (184) becomes
3
2 1 2 1
squeezing motion
sliding motion
6 12
s s s s s
h p h h W V V = ÷ µ ÷ V ÷ + µA v v
(185)
06703 174 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Example #1. Let`s reIormate the slidingIlow problem
Ior a plate sliding past a cylinder (see page 162). In
this problem the equation oI the lower surIace (the
plate) is just
h
1
(x,v) ÷ 0
The equation oI the upper surIace is
h x h
x
R
2 0
2
2
= +
The total gap between the two surIaces is
h x h h h
x
R
( )
= + = +
1 2 0
2
2
Eor Reynolds equation, we also need
Ah ÷ h
2
 h
1
÷ h(x)
The velocity oI the lower surIace is
U
1
÷ U J
1
÷ 0 W
1
÷ 0
while the upper surIace is stationary:
U
2
÷ 0 J
2
÷ 0 W
2
÷ 0
The Iollowing quantities appear in Reynolds equation
AU ÷ U AJ ÷ 0 AW ÷ 0
Reynolds equation becomes
d
dx
h
dp
dx
U
dh
dx
3
6
= µ
which is identical to the derivative oI (164). Solving this 2
nd
order ODE leads to the same
pressure proIile we determined earlier.
Example #2.
Now let`s reIormulate the squeezing Ilow problem on page 166. In this problem the equation oI
the lower surIace (the plate) is just
h
1
(x,v) ÷ 0
The equation oI the upper surIace is h
2
(x,v) ÷ h
06703 175 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
The total gap between the two surIaces is
h h h h = + =
Eor Reynolds equation, we also need
Ah ÷ h
2
 h
1
÷ h
The velocity oI the upper surIace is
U
2
÷ 0 J
2
÷ 0 W
2
÷ U
while the lower surIace is stationary:
U
1
÷ 0 J
1
÷ 0 W
1
÷ 0
The Iollowing quantities appear in Reynolds equation becomes
AU ÷ 0 AJ ÷ 0 AW ÷ U
(185) becomes
3
12
s s
h p U V V = ÷ µ (186)
Because the upper surIace is a circular disk and the gap is uniIorm, we expect squeezing Ilow to
be axisymmetric in cylindrical coordinates. In other words, we expect that p ÷ p(r) (i.e. no u
dependence). In cylindrical (r,u,:) or polar coordinates (r,u), the gradient is
+
s r
p
p
r
c
V =
c
e
while the divergence is
3
1
s s
d dp
h p r
r dr dr
 
V V =

\ .
(186) becomes
3
1
12
d dp
rh U
r dr dr
 
= ÷ µ

\ .
Multiplying through by r and integrating:
3 2
6
dp
rh Ur c
dr
= ÷ µ +
+
see http://www.andrew.cmu.edu/course/06703/Vops¸cyl.pdI
06703 176 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Dividing by rh
3
:
3 3
6
dp r c
U
dr
h rh
= ÷ µ +
When this is integrated a second time, the second term will lead to a logarithmic singularity at
r÷0. To keep pressure Iinite, we choose c÷0 and then the above equation is identical to (174).
Solving this 2
nd
order ODE leads to the same pressure proIile we determined earlier.
Example #3. Sliding oI a plate past a sphere
n this problem the equation oI the lower surIace (the
plate) is just
h
1
(x,v) ÷ 0
The equation oI the upper surIace is
2
2 0
2
r
h r h
R
= +
The total gap between the two surIaces is
2
1 2 0
2
r
h r h h h
R
= + = + (187)
Eor Reynolds equation, we also need
Ah ÷ h
2
 h
1
÷ h(r)
The velocity oI the sphere is purely along the xaxis
v
2
÷ v
s2
÷ Ue
x
÷ e
r
Ucosu  e
Usinu
U
2
÷ U J
2
÷ 0 W
2
÷ 0
while the lower surIace is stationary: v
1
÷ v
s1
÷ 0
The Iollowing quantities appear in (185):
2 1 s s r
dh
h h h
dr
V ÷ = V = e
2 1 2 1
cos sin cos
s s s r r
dh dh
h h U U U
dr dr
u
 
÷ V ÷ = u ÷ u = u

\ .
v v e e e
06703 177 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
(185) becomes
3
3
1 1
6 cos
p h p dh
rh U
r r r r r dr
 
c c c c  
+ = ÷ µ u 


c c cu cu
\ .
\ .
Eor the particular case in which h is given by (187), the solution is
2
6 cos
,
5
Ur
p r p
h
·
µ u
u = +
This produces no Iorce in the :direction (pressure proIile is antisymmetric) but oI course a Iorce
must be applied to the sphere to get it to move:
F UR
R
O
x
= +
16
5
0
tµ
o
o ln
,
as o÷ 0
F
v
÷ F
:
÷ 0
06703 178 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Turbulence
GENERAL NATURE OF TURBULENCE
In all the problems we have analyzed to date, the Iluid elements travel along smooth predictable
trajectories. This state oI aIIairs is called:
laminar flow  Iluid elements travel along smooth deterministic trajectories
These trajectories are straight parallel lines Ior simple pipe Ilows. However, this is not the only
solution to the equations oI motion. Consider the Iollowing experiment
Reynolds Experiment (1882)  inject a thin stream oI dye into a Iully developed Ilow in a pipe;
observe the dye downstream. (see S:37)
Elow
Direction
Re
1500
2340
7500
· Ior laminar Ilow: dye stream appears as a straight colored thread
As the total Ilow rate oI Iluid in the pipe is increased, a sudden change in the appearance oI the
dye stream occurs. The thread oI dye becomes more radially mixed with the Iluid and, Iar
enough downstream, its outline becomes blurred.
· Ior turbulent Ilow: irregular radial Iluctuations oI dye thread
Using a pipe with a sharpedge entrance, Reynolds determined the critical Ilow rate Ior a large
number oI Iluids and pipe sizes. He Iound in all cases, the transition occurred at a critical value
oI a dimensionless group:
µ
µ
v D
:
Re
= ± 2300 200
06703 179 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
where v
:
is the crosssectional average velocity (÷ volumetric Ilowrate / pipe area). Today, we
know this dimensionless group as the Reynolds number.
origin of turbulence  instability oI laminarIlow solution to NS eqns
instability  small perturbations (caused by vibration, etc.) grow rather than decay with time.
That the laminarIlow solution is metastable Ior Re~2100 can be seen Irom Reynolds experiment
perIormed with a pipe in which disturbances are minimized:
 reduce vibration
 Iluid enters pipe smoothly
 smooth pipe wall
Under such conditions, laminar Ilow can be seem to persist up to Re ÷ 10
4
. However, just
adding some vibrations (disturbance) can reduce the critical Re to 2100. The onset oI turbulence
causes a number oI proIound changes in the nature oI the Ilow:
 dye thread breaks up  streamlines appear contorted and random
 sudden increase in Ap/L
 local v
:
Iluctuates wildly with time
 similar Iluctuations occur in v
r
and v
As a consequence oI these changes, no simpliIication oI the NS equation is possible: v
r
, v
, v
:
and p all depend on r, u, : and t.
TURBULENT FLOW IN PIPES
Velocity proIiles are oIten measured with a pitot tube, which is a device with a very slow
response time. As a consequence oI this slow response time, the rapid Iluctuations with time
tend to average out. In the descriptions which Iollow, we will partition the instantaneous velocity
v into a timeaveraged value v (denoted by the overbar) and a Iluctuation ' v (denoted by the
prime):
v v v = + '
Crosssectional area averages will be denoted by enclosing the symbol inside carets:
06703 180 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
v
v r r dr
r dr
Q
R
:
:
R
R
= =
2
2
0
0
2
t
t
t
In laminar Ilow, the velocity proIile Ior Iully developed Ilow is parabolic in shape with a
maximum velocity occurring at the pipe center
that is twice the crosssectional mean velocity:
In turbulent Ilow, the timeaveraged velocity
proIile has a Ilatter shape. Indeed as the Reynolds
number increases the shape changes such that the
proIile becomes even Ilatter. The proIile can be Iit
to the Iollowing empirical equation:
v r v
R r
R
: :
n
( )
, max
/
=
÷
1
where the value oI the parameter n depends on Re:
Re ÷ 4.0x10
3
2.3x10
4
1.1x10
5
1.1x10
6
2.0x10
6
3.2x10
6
n ÷ 6.0 6.6 7.0 8.8 10 10
v
max
/·v
:
~ 1.26 1.24 1.22 1.18 1.16 1.16
The reduction in the ratio oI maximum to
average velocity reIlects the Ilattening oI
the proIile as n becomes larger. OI course,
this equation gives a 'kink¨ in the proIile
at r÷0 and predicts inIinite slope at r÷R, so
it shouldn`t be applied too close to either
boundary although it gives a reasonable Iit
otherwise.
How big are the Iluctuations relative to the
maximum velocity? Instantaneous speeds
can be obtained Ior air Ilows using a hot
wire anemometer. This is simply a very
thin wire which is electrically heated
above ambient by passing a current
through it. As a result oI electrical heating
(I
2
R) the temperature oI the wire will
depend on the heat transIer coeIIicient,
which in turn depends on the velocity oI
06703 181 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Ilow over the wire:
v
:
 ~ h  ~ T
wire
T
air
+
It`s easy to determine the temperature oI the wire Irom its electrical resistance, which generally
increases with temperature. The reason Ior making the wire very thin is to decrease its thermal
inertia. Very thin wires can respond rapidly to the rapid
turbulent Iluctuations in v
:
.
Anyway, the instantaneous speed can be measured. Then the
timeaveraged speed and the Iluctuations can be calculated.
The rootmeansquare Iluctuations depend on radial position,
as shown at right. Typically the axial Iluctuations are less
than 10° oI the maximum velocity whereas the radial
Iluctuations are perhaps halI oI the axial.
Note that the Iluctuations tend to vanish at the wall. This is a
result oI noslip (applies even in turbulent Ilow) which
requires that the instantaneous velocity must vanish at the
wall Ior all time, which implies that the time average and the instantaneous Iluctuations must
vanish.
TIMESMOOTHING
As we will see shortly, these Iluctuations proIoundly increase transport rates Ior heat, mass, and
momentum. However, in some applications, we would be content to predict the timeaveraged
velocity proIile. So let`s try to timeaverage the NavierStokes equations with the hope that the
Iluctuations will average to zero.
Eirst, we need to deIine what we mean by a timeaveraged quantity. Suppose we have some
property like velocity or pressure which Iluctuates
with time:
s ÷ s(t)
We can average over some time interval oI halI width
At:
s t
t
s t dt
t t
t t
( ) ( ' ) ' ÷
÷
+
1
2A
A
A
We allow that the timeaveraged quantity might still
depend on time, but we have averaged out the rapid
Iluctuations due to turbulence.
Now let`s deIine another quantity called the fluctuation about the mean:
06703 182 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
s t s t s t ' ( ) ( ) ( ) ÷ ÷
TIMESMOOTHING OF CONTINUITY EQUATION
The simple Iunctional Iorm oI our experimentally measured velocity proIile  v r
:
( )  is exactly
the same as Ior laminar Ilow. This suggests, that iI we are willing to settle Ior the timeaveraged
velocity proIile, then I might be able to get the result Irom the NSE. Let`s try to timesmooth the
equation oI motion and see what happens. We will start with the equation oI continuity Ior an
incompressible Ilow:
V
v ÷ 0
Integrating the continuity equation Ior an incompressible Iluid and dividing by 2At:
1
2
1
2
0 0
A A
A
A
A
A
t
dt
t
dt
t t
t t
t t
t t
V = =
÷
+
÷
+
v ' '
Thus the righthandside oI the equation remains zero. Let`s take a closer look at the leIthand
side. Interchanging the order oI diIIerentiation and integration:
1
2
1
2 A A
A
A
A
A
t
dt
t
dt
t t
t t
t t
t t
V = V
= V
÷
+
÷
+
v v v ' '
Substituting this result Ior the leIthand side oI the continuity equation, leaves:
0
Thus the Iorm oI the continuity equation has not changed as a result oI timesmoothing.
TIMESMOOTHING OF THE NAVIERSTOKES EQUATION
Encouraged by this simpliIication, we try to timesmooth the NavierStokes equation:
µ
c
c
µ µV µ
v
v v v g
t
p + V = ÷V + +
2
AIter integrating both sides with respect to time and dividing by 2At, we can break the integral oI
the sum into the sum oI the integrals. Most oI the terms transIorm in much the same way as the
leIthand side oI the continuity equation. The result is
µ
c
c
µ µV µ
v
v v v g
t
p + V = ÷V + +
2
With a little additional massaging (see Whitaker), the remaining term can be expressed as
06703 183 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
µ µ µ v v v v v v V = V + V ' '
II the second term on the righthand side were zero, then NSE aIter timesmoothing would have
exactly the same Iorm as beIore timesmoothing. UnIortunately, this term is not zero. Although
the average oI the Iluctuations is zero, the average oI the square oI the Iluctuations is not zero.
So this second term cannot be dropped. Thus the timesmoothed NavierStokes equation
becomes:
µ
c
c
µ µV µ t
v
v v v g
t
p
t
+ V = ÷V + + + V
2 ( )
where t µ
( ) t
= ÷ ' ' v v
has units oI stress or pressure and is called the Reynold`s stress. Sometimes it is also called the
turbulent stress to emphasize that arises Irom the turbulent nature oI the Ilow. The existence oI
this new term is why even the timeaveraged velocity proIile inside the pipe is diIIerent Irom that
during laminar Ilow. OI course, our empirical equation Ior the v r
:
( ) is also diIIerent Irom that
Ior laminar Ilow.
Although we don`t yet know how to evaluate this Reynolds stress, we can add it to the viscous
stress and obtain a diIIerential equation Ior their sum which we can solve Ior the simple case oI
pipe Ilow. Here`s how we do it. Eirst, recall that Ior incompressible Newtonian Iluid, the stress
is related to the rate oI strain by Newton`s law oI viscosity. Time smoothing this constitutive
equation yields:
t µ = V + V v v
t
Taking the divergence: V = t µV
2
v
II we now make this substitution, NSE becomes
µ
c
c
µ t µ t
µ t
v
v v g
g
t
p
p
t
T
+ V = ÷V + V + + V
= ÷V + + V
( )
( )
(188)
where t t t
( ) ( ) T t
= +
is the total stress, i.e., the sum oI the timeaveraged viscous stress and the Reynolds stress. Thus
we see that the Reynolds stress appears in the equations oI motion in the same manner as the
viscous stress. Indeed the sum oI the two contributions plays the same role in turbulent Ilows
that the viscous Iriction played in laminar Ilow.
06703 184 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
ANALYSIS OF TURBULENT FLOW IN PIPES
We can make the same assumptions (i.e. the same guess) about the Iunctional Iorm oI the time
averaged velocity and pressure proIile in turbulent Ilow that we made Ior laminar Ilow: we will
assume that the timeaveraged velocity proIile is axisymmetric (v
÷0, c/cu÷0) and Iully
developed (c/c:÷0).
v v r v v
p p :
: : r
= = =
=
( )
( )
u
0
Then the :component oI (188) yields:
0
1 1
0 0
= ÷ + ÷ ÷
c
c
c
c
t
ct
cu
ct
c
u
P
: r r
r
r :
f :
r:
T
g r
:
T
::
T
¸
,
( )
where P is the timeaveraged dynamic pressure. This Iorm oI this equation was obtained using
the tables in BSL (top halI oI p85, eqn C), aIter replacing the instantaneous quantities by their
time averages, except that the instantaneous viscous stresses t has been replaced by (minus) the
total stress t
(T)
. Expecting the timeaveraged Ilow to be axisymmetric (c/cu ÷ 0) and Iully
developed (c/c: ÷ 0, except Ior pressure), the last two terms in this equation can be dropped and
the second term is a Iunction oI r only. This leaves us with the same equation we had Ior laminar
Ilow: a Iunction oI r only equal to a Iunction oI : only. The only way these two terms can sum to
zero Ior all r and : is iI both equal a spatial constant:
dP
d: r
d
dr
r
P
L
r:
T
= = ÷ <
1
0 t
( )
,
A
This implies that pressure P varies linearly with :. Solving Ior the total stress t
by
integrating:
t t t
r:
T
r: r:
t
P
L
r
c
r
( ) ( )
= ÷ + = +
1
2
A
(189)
The integration constant c was chosen to be zero to avoid having the stress unbounded at r÷0.
Now this is the total stress: the sum oI the Reynolds stress
t µ
v v
= ÷ ' '
06703 185 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
and a viscous contribution Irom timesmoothing Newton`s law oI viscosity:
t µ
r:
:
dv
dr
=
The latter can be determined by diIIerentiating the timeaveraged
velocity proIile. II we subtract this Irom the total we can
determine t
 one oI the components oI the Reynolds stress
tensor. The result is shown in the Iigure at right. Notice that the
turbulent stress tends to vanish near the wall. This can be
explained by noting that at the wall, 'no slip¨ between the Iluid
and the stationary wall requires that the instanteous velocity, as
well as its time average, must be zero:
v v v v v
= = ' = = ÷ ' ' = 0 0 0 t µ
In terms oI the relative importance oI these two contributions to the total, one can deIine three
regions:
1. turbulent core: t
(t)
~~t. This covers most oI the cross section oI the pipe.
2. laminar sublayer: t
(t)
··t. Very near the wall, the Iluctuations must vanish (along with
the Reynolds stress) but the viscous stress are largest.
3. transition zone: t
(t)
~t. Neither completely dominates the other.
When applied to the situation oI Iully developed pipe Ilows, continuity is automatically satisIied
and the timesmoothed NavierStokes equations yields only one equation in 2 unknowns:
2 unknowns: v r v v
: r :
( ) and µ ' '
Clearly another relationship is needed to complete the model. This missing relationship is the
constitutive equation relating the Reynolds stress to the timesmoothed velocity proIile. One
might be tempted to deIine a quantity like the viscosity to relate stress to the timeaveraged
velocity.
t µ
r:
t t :
dv
dr
( )
?
( )
=
But iI you deIine the 'turbulent viscosity¨ this way, its value turns out to depend strongly on
position.
µ
µ
( ) t
=
~
100 near pipe centerline
0 at pipe wall
06703 186 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
So unlike the usual viscosity, µ
is not a material property (since it depends on position rather
than just the material).
PRANDTL`S MIXING LENGTH THEORY
The Iirst successIul constitutive equation Ior turbulence was posed by Prandtl in 1925. Prandtl
imagined that the Iluctuations in instantaneous Iluid velocity at some Iixed point were caused by
eddies oI Iluid which migrate across the Ilow Irom regions having higher or lower timeaveraged
velocity.
eddy  a packet oI Iluid (much larger than a Iluid element) which can undergo random migration
across streamlines oI the timesmoothed velocity Iield.
These eddies have a longitudinal velocity which corresponds to the timeaverage velocity at their
previous location.
As this eddy moves across the streamlines, it
gradually exchanges momentum with the
surrounding Iluid which is moving at a diIIerent
longitudinal velocity. But this exchange does not
occur instantaneously. The eddy retains its
original velocity Ior a brieI period oI time. We
might call this the mixing time. During this time,
the eddy migrates laterally a distance l called the
mixing length:
mixing length (l)  characteristic distance an eddy
migrates normal to the main Ilow beIore mixing
Although momentum exchange between eddies occurs continuously in actual turbulent Ilow,
Prandtl imagined that a migrating eddy keeps all oI its original velocity until it migrated a
distance l and then suddenly it exchanges it. This is like molecules oI a gas retaining its
momentum until it collides with another gas molecule, which causes a sudden exchange oI
momentum. Indeed, you might Iind it helpIul to think oI the mixing length as being the analogue
oI meanfreepath in the kinetic theory oI gases.
Recall that:
meanfree path  average distance a gas molecule
travels beIore colliding with another gas molecule.
Now suppose we are monitoring the instantaneous
velocity at a distance v Irom the wall when an eddy
driIts into our location Irom v¹l. Because this
migrating eddy has a higher velocity than the average
Iluid at v, we will observe an positive Iluctuation
eddy migrating
randomly
across Ilow
v
v v
( )
06703 187 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
when the eddy arrives. To estimate the magnitude oI the Iluctuation, we can expand the time
smoothed velocity proIile in Taylor series about v÷v:
v v l v v
dv
dv
l
d v
dx
l
x x
x
v
x
v
( ) ( ) + = + + +
1
2
2
2
2
Assuming that l is suIIiciently small that we can truncate this series without introducing
signiIicant error:
' = + ÷ ~ v v v l v v l
dv
dv
x
above
x x
x
( ) ( )
where the subscript 'above¨ is appended to remind us that is the Iluctuation resulting Irom an
eddy migrating Irom above. At some later time, another eddy might migrate to our location Irom
below, producing a negative Iluctuation in velocity:
' = ÷ ÷ ~ ÷ v v v l v v l
dv
dv
x
below
x x
x
( ) ( )
OI course the average Iluctuation is zero: ' = v
x
0 , but the average oI the squares is not:
' ~ ' + '
~
v v v
dv
dv
x x x
x
above below
2
1
2
2 2
2
l
2
(190)
Now let`s turn our attention to ' v
v
. This is related to how Iast the eddies migrate, and the sign
depends on whether they are migrating upward or
downward.
II the eddy migrates Irom above, it represents a
negative vIluctuation (it is moving in the v
direction). Such an eddy will have a greater x
velocity than the Iluid receiving it, consequently
generating a positive xIluctuation:
v
v
·0 ÷ v
x
~0 ÷ v
x
v
v
·0
On the other hand, iI the eddy migrates Irom below,
it represents a positive vIluctuation but has less x
velocity than the Iluid receiving it, generating a negative xIluctuation:
v
v
~0 ÷ v
x
·0 ÷ v
x
v
v
Einally, iI there is no vertical migration oI eddies, there is no reason Ior the xvelocity to
Iluctuate:
06703 188 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
v
v
÷ 0 ÷ v
x
÷ 0
These three statements suggest that the vIluctuations are proportional to the xIluctuations, with
a negative proportionality constant:
v
v
´ ~ ov
x
´
where o~0. Alternatively, we can write:
v
x
v
v
÷ o(v
x
)
2
Time averaging and then substituting (190):
' ' = ÷ ' = ÷
v v v
dv
dv
x v x
x
o o ( )
2 2
2
l
Absorbing the unknown o into the (still unknown) mixing length parameter:
t µ µ
xv
t
x v
x
v v
dv
dv
( )
= ÷ ' ' =
l
2
2
(191)
Comparing this result with Newton`s law oI viscosity:
t µ
xv
x
dv
dv
=
we could conclude that an apparent turbulent viscosity is given by:
2 t x
dv
l
dv
µ µ
OI course, this viscosity is not a true Iluid property, because it depends strongly on the velocity
proIile.
Eor this theory to be useIul, we need a value Ior the 'mixing length¨ l. There are two properties
oI l which we can easily deduce. Eirst oI all, l was deIined as the distance normal to the wall
which the eddy travels beIore becoming mixed with local Iluid. Clearly, this mixing must occur
beIore the eddy 'bumps¨ into the wall, so:
Property #1: l · v
where v is the distance Irom the wall. Secondly, we know Irom noslip that the Iluctuations all
vanish at the wall. Consequently, the Reynolds stress must vanish at the wall. Since the velocity
gradient does not vanish, we must require that the mixing length vanish at the wall:
06703 189 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Property #2: l÷0 at v÷0
II it`s not a constant, the next simplest Iunctional relationship between l and v which satisIies
both these properties is:
l ÷ av (192)
where a is some constant and 0·a·1.
PRANDTL`S ~UNIVERSAL¨ VELOCITY PROFILE
The velocity proIile in turbulent Ilow is essentially Ilat, except near the wall where the velocity
gradients are steep. Eocussing attention on this region near the Ilow, Prandtl tried to deduce the
Iorm Ior the velocity proIile in turbulent Ilow. Recall Irom (189) that in pipe Ilow, the total
stress varies linearly Irom 0 at the center line to a maximum value at the wall:
t t
r:
T
P
L
r
r
R
( )
= ÷ = <
1
2
0
0
A
(193)
where we have deIined t
0
÷ (1/2)(R/L)AP ~ 0
which represents the stress on the wall. In the 'turbulent core¨, the Reynolds stress dominates
the 'laminar¨ stress; then substituting (191) through (193):
( ) ( ) t T
xv r:
t ~ t
2 2
2
2
0
1
x
a v
dv v
l
dv R
 
 
µ ~ t ÷
 
\ .
\ .
(194)
or av
dv
dv
v
R
v
v
R
x
v
= ÷
= ÷
t
µ
0
1 1
*
*
The general solution to this 1
st
order ODE is
v v C
a
v
R
a
v
R
+ +
+
+
÷
+
+
= + ÷ ÷ ÷
2
1
2
1
1
tanh (195)
where C is the integration constant, v* is called the friction velocity and where we have
introduced dimensionless variables:
06703 190 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
l = ay (192)
where a is some constant and 0<a<1.
PRANDTL’S “UNIVERSAL” VELOCITY PROFILE
The velocity profile in turbulent flow is essentially flat, except near the wall where the velocity gradients
are steep. Focussing attention on this region near the flow, Prandtl tried to deduce the form for the
velocity profile in turbulent flow. Recall from (189) that in pipe flow, the total stress varies linearly from
0 at the center line to a maximum value at the wall:
τ τ
rz
T
P
L
r
r
R
( )
· − · <
1
2
0
0
∆
(193)
where we have defined τ
0
≡ (1/2)(R/L)∆P > 0
which represents the stress on the wall. In the “turbulent core”, the Reynolds stress dominates the
“laminar” stress; then substituting (191) through (193):
( ) ( ) t T
xy rz
τ ≈ τ
{
2 2
2
2
0
1
x
a y
dv y
l
dy R
¸ _
¸ _
ρ ≈ τ −
¸ ,
¸ ,
(194)
or ay
dv
dy
y
R
v
y
R
x
v
· −
F
H
I
K
· −
F
H
I
K
τ
ρ
0
1 1
*
*
;
The general solution to this 1
st
order ODE is
v y C
a
y
R
a
y
R
+ +
+
+
−
+
+
· + − − − d i
2
1
2
1
1
tanh (195)
where C is the integration constant, v* is called the friction velocityand where we have introduced
dimensionless variables:
v
v
v
x +
·
*
y
v
y
+
·
*
ν
and R
v
R
+
·
*
ν
06703 191 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Near the wall (i.e. for y<<R or y
+
<<R
+
), we can simplify (195):
for y
+
<<R
+
: 1 1
2
2
− · − +
+
+
+
+
+
y
R
y
R
O y d i (196)
tanh ln
−
+
+
+
+
+
− · +
1
1
1
2
4 y
R
R
y
O y d i
Dropping the higherorder terms:
v y
R
a
C
a
y
c
+ +
+
+
·
−
+ + d i
b g
2 4 1 ln
ln
1 2 44 3 44
(197)
where c is a collection of constants.
This result can be derived more easily by starting over with a simplified (194):
for y<<R: *
x
dv
ay v
dy
· or
{
1 1
*
x
dv
dv dy dy
v a y a
y
+
+
+
· · (198)
which integrates to
1
ln v y c
a
+ +
· +
where c is some integration constant. When
plotted on semilog coordinates, experimental
velocity profiles do indeed show a linear
region which extends over a couple of
decades of y
+
values:
Moreover, the slope and intercept of this
straight line don't seem to depend on the
Reynolds number. Indeed, the slope and
intercept also don't seem to depend on the
shape of the conduit. Rectangular conduits
yields the same velocity profile on these
coordinates. This is called Prandtl’s
Universal Velocity Profile:
y
+
>26: v y
+ +
· + 2 5 55 . ln . (199)
(6)
06703 192 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
which applies for y
+
>26 (the turbulent core). This coefficient of lny
+
corresponds to a=0.4. so (192)
becomes:
l y · 04 .
Recall that we reasoned that a had to be in the range of 0 to 1 to be physically realistic.
In the laminar sublayer, Reynolds stress can be totally neglected, leaving just viscous stress. This
close the wall, the total stress is practically a constant equal to the wall shear stress τ
0
:
y << R:
τ τ
µ τ
xy rz
T
x
dv
dy
≈
·
R
S

T

( )
0
Then we can integrate the above ODE for v
x
, subject to v
x
· 0 at y=0 (i.e. no slip):
v y
x
·
τ
µ
0
Dividing both sides by v* we can make the result dimensionless:
y
+
<5: v y
+ +
· (200)
which applies for 0<y
+
<5 (the laminar sublayer). Of course (199) also does not apply near the
center of the pipe, since the y
+
≈ R
+
there, whereas (199) was derived by assuming that y
+
<< R
+
(see
(196)).
PRANDTL’S UNIVERSAL LAW OF FRICTION
Let’s try to figure deduce the analog of Poisueille's Formula (see page 87) for turbulent flow.
Poisueille's Formula is the relationship between volumetric flowrate through the pipe and pressure drop.
Volumetric flowrate Q is calculated by integrating the axial component of fluid velocity of the cross
section of the pipe:
v
Q
R R
r v r dr
z z
R
· ·
z
π
2 2 0
2
( )
Now we are going to use (199) for the velocity profile, although we assumed in (198) that y<<R (where
y = Rr).
06703 193 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Although the derivation of (199) assumed that
we are very close to the wall (y
+
<<R
+
, see
(196)), (199) works remarkly well right out
to the centerline y
+
=R
+
. The plot at right
shows the velocity profiles (with different wall
roughness on the walls) compared with
predictions based on (199). The ordinate is
( ) ( )
( ) ( )
( )
2.5ln 5.5 2.5ln 5.5
*
2.5 ln ln 2.5ln
2.5ln 2.5ln
z z
v R v r
v R v y R y
v
R
R y
y
R R
y R r
+ + + + + +
+
+ +
+
−
1 1
· − · + − +
¸ ] ¸ ]
· − ·
· ·
−
Note that (199) predicts an infinite velocity
difference at y=0, whereas the actual velocity
must be finite. Of course, (199) does not
apply right up to the wall because very near
the wall the Reynolds stresses are not
dominant.
Substituting (199) and integrating:
v v
v R
z
·
F
H
G
I
K
J
+
L
N
M
O
Q
P
* . ln
*
. 25 175
ν
(201)
Now the friction velocity can be related to the friction factor, whose usual definition can be expressed
in terms of the variables in this analysis:
f
v
v
v
z
z
≡ ·
F
H
G
G
I
K
J
J
τ
ρ
0
1
2
2
2
2
*
Thus
v
v f
z
*
·
2
Likewise, the usual definition of Reynolds number yields
06703 194 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Re ≡
2 v R
z
ν
Thus
v R
v R
v
v
f z
z
f
* * Re
Re
ν ν
· × ·
2
2
2 2
123
;
Relating v* to f and v
z
to Re, (201) can be written as:
1
177 060
f
f · − . ln Re .
d i
or
1
4 07 0 60
10
f
f · − . log Re .
d i
which fits experimental data remarkable well. A slightly better fit can be obtained by adjusting the
coefficients:
1
4 0 040
10
f
f · − . log Re .
d i
(202)
which is called Prandtl’s (universal) law of friction. It applies virtually over the entire range of
Reynolds numbers normally encountered for turbulent pipe flow: 2100 < Re < 5x10
6
.
06703 195 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Electrohydrodynamics
ORIGIN OF CHARGE
Nearly all particles, bubbles or drops in water are charged.
We know this because when you apply an electric field across
water containing a particle, bubble or drop, you can observe it
to move. An example is the red blood cell experiment at right.
Interface between the two immiscible phases of the particle and
water acquires charge by a number of mechanisms:
Ø adsorption of ions (e.g. SDS and FeOH
+2
)
Ø dissolution of ions from ionic solids (e.g. AgI)
Ø dissociation of acidic or basic sites on interface
—COOH → —COO

+ H
+
—NH
2
+ H
+
→ —NH
3
+
Regardless of how the surface acquires its charge, overall electroneutrality must be obeyed. So the
solution must contain an excess of counterions. For example, if the surface acquires its charged by the
dissociation of COOH groups at the interface, the COO

remaining at the interface gives the interface a
negative charge, but the H
+
joins the rest of the ions in the aqueous solution, thereby lending it a net
positive charge.
Counterions would experience an electrostatic attraction for the surface which tends to pile up the ions
next to the interface. On the other hand, the ions undergo diffusion (Brownian motion) which tends to
disperse them uniformly throughout the phase. At equilibrium, a balance is achieved between these two
opposing tendencies. In this equilibrium state, a diffuse
cloud of counterions is formed next to the charged
interface.
double layer  layer of charged fixed on interface +
diffuse cloud of counterions.
Even though there are equal concentrations of positive
and negative ions far from the surface, this is not true
inside the counterion cloud. Fluid elements inside the
06703 196 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
cloud bear a net charge. Just like we define the mass per volume to be the density of the fluid, we can
define:
ρ
e
(x,t) = local charge density (charge/volume)
Global electroneutrality requires:
σ ρ + ·
∞
z e
y dy b g
0
0
σ = surface charge density (charge/area)
GOUYCHAPMAN MODEL OF DOUBLE LAYER
Let's try to quantify this description. In particular, we would like to know how thick this cloud is.
Because of the charge on the interface, ions feel a electrostatic force. The force per unit charge is called
the electric field:
E(x) = electric field (force/charge)
Like gravity, this vector field is conservative, thus there must exist an scalar potential, ψ(x) such that:
E = ∇ψ (203)
Like the gravitational potential, the minus sign is included by convention. ψ is called the electrostatic
potential.
ψ [=] volts (energy/charge)
In thermodynamics, we learn that the criteria for phase equilibrium is equality of chemical potential µ
i
for each chemical species i in each phase. For an ideal solution, recall that changes in chemical potential
at constant temperature and pressure is given by
∂µ
∂
i
i
T P
i
c
kT
c
F
H
G
I
K
J
·
,
or d kT d c
i
T P
i
µ b g
,
ln · (204)
where kT is the thermal energy possessed by every ion or molecule and µ
i
is the chemical energy per
ion. When the species is charged, we must add an electrical contribution to the chemical energy to
obtain the electrochemical potential of each species.
electrical potential energy: z
i
eψ
06703 197 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
where z
i
is the number of elemental charges per unit ion (including sign) and e (>0) is the magnitude of
the elemental charge (on proton); thus z
i
e is the charge on one ion. At equilibrium, ionic species
distribute themselves in an electric field such that their electrochemical potential is constant everywhere:
µ µ ψ
i
e
i i
z e x x x b g b g b g · + · const (205)
For example, diffusion of nonelectrolytes (i.e. z
i
= 0) takes place so that µ
i
is constant throughout the
solution, which means that the concentration c
i
is uniform at equilibrium. If the electrostatic potential
varies from one location to the next, the concentration of charged species will also not be uniform.
(205) can be rewritten in differential form as
d d z e d
i
e
i i
µ µ ψ · · + 0
After substituting (204), we have kT d c z e d
i i
ln + · ψ 0
which relates changes in ion concentration to changes in potential. This can be integrated to obtain
Boltzmann's Equation relating the local ion concentration to the local electrostatic potential
c A
z e
kT
i
i
x
x
b g
b g
· −
L
N
M
O
Q
P
exp
ψ
where A is some integration constant. In our problem, we expect the ion concentration to depend only
on y and we expect that the ionic concentration will approach the bulk value c
i∞
at some distance from
the charged interface. For this to happen, the electrostatic potential must tend to some constant (say
ψ→0 as y→∞) or E → 0:
as y → ∞: c
i
→ c
i
∞, ψ → 0
To satisfy these boundary conditions, Boltzmann’s equation can be rewritten as
c y c
z e y
kT
i i
i
b g
b g
· −
L
N
M
O
Q
P ∞
exp
ψ
(206)
So if I knew the ψ(y), I could calculate the concentration profile. But alas ψ(y) is still unknown. I need
another equation. The extra equation is provided by Coulomb’s law of electrostatics. If I knew the
distribution of charges within a system, I could calculate the force on them using Columb’s law. For a
continuum, Coulomb’s law can be written as:
∇ ·
.
E
4π
ε
ρ
e
or ∇ · −
2
4
ψ
π
ε
ρ
e
(207)
where ε is another material property called the electric permittivityof the medium between the ions
(e.g. the water):
06703 198 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
ε
vacuum
2
2
Coul
N m
· ×
−
1113 10
10
.
ε ε
water vacuum
· 78
This equation is called Poisson's Equation. The charge density of the fluid arises from the charge
born by each ion. Adding up the charges from each species i, then substituting Boltzman’s equation,
gives
ρ
ψ
e i i
i
i i
i
i
y z ec y z ec
z e y
kT
a f a f
a f
· · −
L
N
M
O
Q
P ∑ ∑ ∞
exp
Special Case: symmetric binary electrolyte, which means there are only two different species of ions and
they have the same magnitude of charge:
z
+
= z

= z
c
+∞
= c
∞
= c
∞
Then Boltzmann's equation can be used to calculate the charge density:
ρ
ψ ψ
ψ ψ ψ
e
y z ec
z e
kT
z ec
z e
kT
zec
ze
kT
ze
kT
zec
ze
kT
( ) exp exp
exp exp sinh
· −
L
N
M
O
Q
P
+ −
L
N
M
O
Q
P
· −
F
H
G
I
K
J − +
F
H
G
I
K
J
L
N
M
O
Q
P
· −
F
H
G
I
K
J
+ +∞
+
− −∞
−
∞ ∞
2
(208)
Substituting this into Poisson's equation, we get:
d
dy
ec ze
kT
2
2
8 ψ π
ε
ψ
·
F
H
G
I
K
J
∞
sinh
The argument of the sinh must be dimensionless, so let’s use this combination to define a new
dimensionless potential
Ψ ≡
ze
kT
ψ
We can make the ψ on the lefthand side dimensionless by multiplying both sides by ze/kT. Then we
have
d
dy
2
2
2
Ψ
Ψ · κ sinh (209)
06703 199 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
where κ
π
ε
2
2 2
8 1
· ·
∞
z e c
kT
[ ]
cm
2
Eq. (209) is a nonlinear ODE. Despite its innoccuous appearance, it is much more difficult to solve than
linear equations. However, a particular solution is possible:
subject to: Ψ = Ψ
0
at y=0
Ψ = 0 at y=∞
tanh tanh
Ψ Ψ y
e
y
b g
4 4
0
L
N
M
O
Q
P
·
F
H
G
I
K
J
−κ
(210)
which is called the GouyChapman model for the double layer. In the special case in which the
surface potential is small, these expressions simplify to:
ψ
0
<< kT ze:ψ(y) = ψ
0
exp(κy) and
c y c
ze
kT
e
y
t ∞
−
·
F
H
I
K
a f 1
0
m
ψ
κ
Clearly κ is associated with the thickness of the
diffuse cloud. κ
1
is called the Debye length:
κ
1
= 10
9
m = 1nm for c
∞
= 10
1
M
κ
1
= 10
6
m = 1µm for c
∞
= 10
7
M
This provides some idea of the thickness of the
charge cloud.
ELECTROSTATIC BODY FORCES
Since the fluid inside the diffuse cloud is charged,
from (208): ρ
ψ ψ
ψ
κ
e
ze
kT
y
y zec
ze
kT
zec
ze
kT
e ( ) sinh · −
F
H
G
I
K
J ≈ −
∞
≈
∞
−
2 2
0
1 2 4 3 4
any electric field in the fluid will exert a force on the fluid elements. This body force needs to be
included in the NavierStokes equations:
electrostatic body force/volume = ρ
e
E
06703 200 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
which is analogous to the body force exerted by gravity:
gravitational body force/volume = ρg
Including electrostatics, but neglecting gravity, the NavierStokes equation becomes:
ρDv/Dt = µ∇
2
v  ∇p + ρ
e
E
In the GouyChapman model described in the previous section, the fluid is stagnant. At hydrostatic
equilibrium, any electric field applied to a charged fluid element will lead to a pressure gradient:
v = 0: ∇p = ρ
e
E
To see how large the pressure gradient is, let’s return to the GouyChapman model where the electric
field is oriented normal to the charged surface: i.e. ψ = ψ(y). Then the pressure gradient must also be
normal to the wall so that p = p(y). Hydrostatic equilibrium requires
dp
dy
E zec
ze
kT
d
dy
e y
· · −
F
H
G
I
K
J
L
N
M
O
Q
P
−
F
H
G
I
K
J ∞
ρ
ψ ψ
2 sinh (211)
where the second equality results from substituting (122) and (203). The righthand side turns out to be
an exact differential, so that the equation can be rewritten as
dp
dy
c kT
d
dy
ze
kT
·
F
H
G
I
K
J
∞
cosh
ψ
Multiplying both sides by dy and integrating from y=y to y=∞:
dp c kT d
ze
kT
c kT
ze y
kT
ze
kT
p y p c kT
ze y
kT
p
p y
y
y y
∞
z z
·
F
H
G
I
K
J
· −
∞ L
N
M
O
Q
P
− ·
L
N
M
O
Q
P
−
R
S

T

U
V

W

∞
·∞
·
∞
∞ ∞
( )
cosh cosh cosh
( ) cosh
ψ
ψ ψ
ψ
b g b g
b g
1
The last step involves substituting ψ(∞)=0 and remembering that cosh(0)=1. You might also recall that
06703 201 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
cosh x
e e
x
x
x x
≡
+
·
·
> ≠
R
S
T
−
2
1 0
1 0
for
for
Thus the pressure inside the cloud is always greater than
the pressure in the bulk solution outside the cloud. This is
because the diffuse cloud is always oppositely charged
compared to the surface, so the cloud is also being pulled
by electrostatic forces toward the surface — just like
gravity always pulls us downward.
ELECTROKINETIC PHENOMENA
So far in our discussion, there has been no fluid motion. But since the fluid inside the diffuse cloud is
charged, we can exert a force on it by applying an
electric field tangent to the surface.
Consider again the distribution of ions near a charged
interface. The solution is not electrically neutral. So if
an electric field is applied tangent to the surface, the
fluid will experience an electrostatic body force:
dF
f
= (dq)E = (ρ
e
dV)E
where dq is the net charge on a fluid element having
volume dV. The solid is oppositely charged, so it feels
a force in the opposite direction. Consequently, at steady state, we can expect the fluid to move relative
to the solid, or vice versa.
At least two phenomena are associated with with the relative motion generated by this externally applied
electrostatic force:
electrophoresis migration of charged particles through an otherwise quiescent fluid
electroosmosis flow of fluid through a porous solid bearing a surface charge
In both cases, relative motion between the fluid and the solid arises when an electric field is applied
externally. Conversely, forcing a fluid to move tangent to a charged interface generates an electric field:
sedimentation potential  the electrical potential (gradient) which arises during sedimentation
of charged particles in a gravitational field
streaming potential  the electrical potential (gradient) which arises during flow of fluid
through a porous solid which bears a surface charge
1
cosh x ( )
0
x
2 0 2
0
2
4
d
f
F
d
s
F
06703 202 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
These four phenomena depend on the magnitude of the surface charge. Measurements of these induced
velocities or induced potentials are the most commonly used techniques for determination of surface
charge.
SMOLUCHOWSKI'S ANALYSIS (CA. 1918)
Consider an infinite plate in contact with an electrolyte
solution. Suppose the plate bears a uniform surface charge
density and we somehow externally apply an electric field E
x
tangent to the surface in the +x direction. Let's try to find the
velocity profile v induced by this electric field.
continuity: ∇
.
v = 0
NSE: ρDv/Dt = µ∇
2
v  ∇p + ρ
e
E (212)
From the arguments above, we expect the fluid to move in the direction of the applied electrostatic
force. The simplest solution to this problem which has this property is:
v
x
= v
x
(y), v
y
= v
z
= 0
Since v
x
is independent of x, this solution automatically satisfies the mass balance. Of course, we must
allow v
x
to vary with y so we can meet the noslip condition at the surface. Note that at steadystate,
the inertial terms identically vanish (like gravitydriven flow down inclined plane).
In the absence of flow, the ion concentration profiles are independent of x [see (206)]. Any flow
tangent to the plate is not expected to perturb the ion concentration profiles because tangential flow just
replaces fluid having a particular concentration with fresh fluid having the same composition (also v
.
∇c
i
= 0). This suggests that the charge density profile is not perturbed by flow caused either by an
externally applied electric field (tangent to plate) or by an externally applied pressure gradient:
ρ = ρ
eq
where we will use the superscipt “eq” to denote the equilibrium state of the variable (in the absence of
flow); we will also use the prefix "δ" to denote perturbations from equilibrium (caused by the applied
electric field). Thus the above statement is equivalent to
ρ ≡ ρ
eq
+ δρ where δρ = 0
06703 203 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
If we used two large vertical plate electrodes to apply the
electric field, as shown at right, we would expect the electric
field lines to be straight and tangent to the surface
*
and the
electrostatic potential to vary linearly with distance from one
electrode to the other. Thus the applied electrostatic forces
should act perpendicular to the electrodes and parallel to our
surface. In other words, the applied electric field is
expected to have only an xcomponent which independent of
x and y.
This expectation is fulfilled if the electrostatic potential has the
following form:
ψ ψ δψ
ψ δψ
δ
x y y x
d
dy
d
dx
eq
eq
eq
, b g b g b g
2
· +
∇ψ · +
−
−
−
E
E
E
j i
123
123
where ψ
eq
is the equilibrium potential found in the absence of
an externally applied electric field or an applied pressure
gradient (see previous section on “GouyChapman model”),
and where δψ is the perturbation from equilibrium caused by
the externally applied electric field or pressure gradient. The pressure field suffers a similar perturbation:
p x y p y p
eq
, b g b g · + δ
The ycomponent of (212) yields:
dp
dy
p
y
E
eq
eq
y
eq
+ ·
∂δ
∂
ρ
but adding the superscript eq to terms in (211) yields:
dp
dy
E
eq
eq
y
eq
· ρ
subtracting:
∂δ
∂
p
y
· 0 or δp = δp(x)
*
The field lines might not be straight near the entrance or exit of the capillary. Thus what follows strictly
only applies in the central region away from the ends.
06703 204 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Poisson's equation (207) can also be written as:
∇ ·
∇ + ∇ ·
.
. .
E
E E
4
4
π
ε
ρ
δ
π
ε
ρ
eq eq
but in the absence of flow, Poisson's equation requires
∇ ·
.
E
eq eq
4π
ε
ρ
subtracting leaves ∇ ·
.
δE 0
or
d E
dx
E
x
x
δ
δ · · 0 or const (213)
The xcomponent of (212) becomes
µ ρ δ
δ d v
dy
E
d p
dx
const
p p
L
x eq
x
g y
f x
L
2
2
0
0
0 + · · ·
−
−
·
( )
( )
1 2 444 3 444
4
(214)
By applying a pressure gradient along the surface, we could “pump” the fluid past the solid, but this is
not the problem we want to address. Thus we set the pressure gradient to zero (p
0
= p
L
) so there is no
pressuredriven flow. Boundary conditions include
at x=0: v
x
= v
x,s
as x→ ∞: dv
x
/dy → 0 or v
x
→ v
x,f
where v
x,s
is the velocity of the solid and v
x,f
is the velocity of the bulk fluid outside the cloud. The first
condition is just “noslip” which says that at the interface between the fluid and the solid the two take on
the same velocity. The second condition says there is no shear force being applied to the fluid outside
the double layer.
To go further, we need to know how ρ
eq
changes with y. This is provided by Poisson's equation (207):
ρ
ε
π
ψ
eq
eq
y
d
dy
b g
· −
4
2
2
(215)
(215) into (214): µ
ε
π
ψ
δ
d v
dy
d
dy
E
x
eq
x
2
2
2
2
4
·
06703 205 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Since δE
x
=const, this can easily be integrated to
µ
ε
π
ψ ψ δ
ζ
v y v y E
x x s
eq eq
x
b g b g b g − · −
−
,
4
0
1 2 444 3 444
Substituting ψ
eq
(y) from (210), we obtain a velocity profile like that
shown at right.
Let v
x
(∞) ≡ v
x,f
Then the above equation can be rewritten as
v v E
x s x f x , ,
− ·
εζ
πµ
δ
4
(216)
where ζ ≡ ψ
eq
(0)  ψ
eq
(∞)
is called the
zeta potential  the electrostatic potential drop across the diffuse part of the double layer.
ELECTROOSMOSIS IN CYLINDRICAL
PORES
Consider a circular hole drilled through a plate.
If this is submerged in an electrolyte solution and
becomes charged, then applying a voltage across the
length of the pore will give rise to electroosmosis
through the hole.
If the pore radius a is must larger than the Debye
length κ
1
, then the velocity profile is essentially plug
flow except inside the double layer where the velocity
suddenly drops to zero at the stationary solid surface
v
s
= 0
v r E
fluid x
b g · −
εζ
πµ
δ
4
E
x
κ
−1
y
I
y
II
k
1
06703 206 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
for all r outside the double layer. Here we have applied (216), but dropped the prefix "δ" since we
have no further need to distinguish between the equilibrium and the applied fields.
Example: A typical value of the ζ potential for aqueous solutions is about 50mV. Plugging in the values
of the other parameters
εζ/4πµ = 3.5 µm/s per V/cm
as a typical value of v
f
/E
x
.
Consequently, the volumetric flowrate through the pore is just this velocity times the crosssectional
area:
κa>>1: Q = πa
2
(εζ/4πµ)E
x
So a measurement of the flowrate for a known applied electric field allows us to determine the zeta
potential.
ELECTROPHORESIS
Consider a rigid sphere immersed in a quiescent electrolyte
solution. Somehow we apply an electric field of strength E. In this
case the solid moves while the fluid far from the sphere remains
stationary:
v
x,f
= 0
(216) predicts that v
x,s
is
κa>>1: U = (εζ/4πµ)E
where εζ/4πµ is now called the electrophoretic mobility. This result is called Smoluchowski's
Equation.
This applies for particles which are large compared to the Debye length. In the opposite limit of very
small small particles, another simple expression can be obtained:
κa<<1: U = (3/2)(εζ/4πµ)E
which is called Huckel's equation. For the most general case of intermediate size, the problem is
much more complicated. A numerical solution is required (see Fig. 9.11.3, p560 of Hunter).
k
1
06703 207 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
STREAMING POTENTIAL
So far in our discussion of electrokinetic phenomena, we have considered the motion of either fluid
or solid that results from the application of an electric field:
Electrophoresis, Electroosmosis: E≠0 → U≠0
Now we would like to consider the inverse situation in which relative motion across a charged interface
generates an electric field:
Streaming Potential,
Sedimentation Potential: U≠0 → E≠0
Consider the pressuredriven flow through a
large circular pore.
κ
1
<< a << L
Owing to the applied pressure drop, we generate
a fully developed parabolic velocity profile so
familiar for laminar flow.
v r
Q
a
r
a
z
v
z
b g
:
· −
F
H
G
I
K
J
L
N
M
M
O
Q
P
P
2 1
2
2
π
The main idea is that this flow causes convection
of charge in the counterion cloud which tends to generate a current. However in the absence of
electrodes in the two reservoirs, there is no way to form a complete electric circuit. So charge tends to
pile up in the reservoir — positive charges on one side of the membrane and negative charges on the
other. Coulomb's law exerts a force on these charges which tends to restore electroneutrality to the
system and to prevent any steadystate rate of accumulation of charge.
Clearly any steadystate electrical current would eventually lead to infinite charge separation and infinite
attractive forces. The only way to avoid this is to have zero current at steady state. This is achieved
by an electric field which spontaneously arises inside the bulk of the fluid which creates an electrical
current equal but opposite to the convective current:
I = I
conv
+ I
elec
= 0 (217)
The electrostatic potential drop associated with this induced electric field is called the streaming
potential. To calculate it, let's first calculate the current generated by convection of charges inside the
cloud.
P P
I II
> P
II
y
I
y
II
06703 208 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
If the charge cloud is very thin compared to the pore radius, then we
need to concentrate on the velocity profile right next to the wall. Here
the velocity profile becomes linear:
for s<<a: v
z
(s) = Γs
where s ≡ ar
and Γ = dv
z
/dr
r=a
The net current through any surface S (open or closed) is just the
integral of the normal component of the current density over the
surface:
I da
S
·
z
n i
.
[=] amps
In our case, we choose a disk of radius a whose normal is parallel to the tube axis.
I v rdr a sds
conv z
a
· ≈
z z
∞
ρ π π ρΓ 2 2
0 0
(218)
Since ρ=0 for most of the fluid (except for r≈a), we can approximate this integral as shown by the
second equality above. Next, we substitute Poisson's equation (215) for the space charge density:
ρ = (ε/4π) d
2
ψ
eq
/ds
2
(219)
(219) into (218) and integrating by parts:
I
a d
ds
sds
a
z z
conv
eq
eq eq
· − · − − ∞
∞
z
ε ψ ε
ψ ψ
Γ Γ
2 2
0
2
2 0
, ,
b g b g
The potential drop across the counterion cloud is the zeta potential. Thus
I
conv
= (1/8π)εaΓζ (220)
According to (217), this convective contribution to the current must be balanced an electrical
contribution from an induced electric field. The relationship between current and electric field is given
by Ohm's law for electrolyte solutions, which takes the form of
i
elec
= KE
where K is called the specific conductanceof the solution. Integrating over the crosssection of the
tube:
κ
−1
06703 209 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
I KE rdr a KE
elec z
a
z
· ≈
z
2
0
2
π π (221)
where E
z
= ∂ψ/∂z = const = ∆ψ/L
is required by (213), where L is the length of the cylindrical pore. (220) and (221) into (217):
∆ψ = εζLΓ/2πaK (222)
Finally, we can relate Γ to the applied pressure
drop by performing a macroscopic force balance
on the tube. Recognizing that the shear stress on
the tube wall τ
rz
is µdv
z
/ds
r=a
= µΓ,
µΓ(2πaL) = ∆p(πa
2
)
or L
a
p Γ ∆ ·
2µ
(223)
(223) into (222): ∆ψ ∆ ·
εζ
πµ 4 K
p
which is Smoluchowski's equation for streaming potential.
Surface Tension
MOLECULAR ORIGIN
Many of the fascinating effects we saw in Trefethen’s film are a consequence of surface tension.
What is surface tension and where does it come from?
Surface tension is a consequence of intermolecular forces which
are always present. The interaction between two molecules is
often represented by the LennardJ ones 612 formula:
V R E
R R
Born vdW
b g ·
F
H
G
I
K
J
−
F
H
G
I
K
J
L
N
M
M
M
M
O
Q
P
P
P
P
4
12 6
σ σ
123 123
This expression is partly based on theory and partly empirical.
The inverse 6
th
power term represents van der Waals
06703 210 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
attraction which is based on a rigorous theory. The inverse 12
th
power term is empirical. It
represents Born repulsion which results from the overlap of electron clouds. σ is called the collision
diameter — this is the centertocenter distance at which the electron clouds begin to overlap and a
significant repulsion is experienced.As a first approximation, we can think of condensation of a liquid
from a vapor in terms of the molecules falling into this potential energy well:
vapor: R→ ∞ V
vapor
= 0
liquid: R ≈ R
min
= 1.12σ V
liq
= E
The heat of vaporization is approximately the energy gained by falling in this well:
U
vap
≈ EN
A
This is all only approximate because we have only considered twobody interactions. In a condensed
state, multibody interactions must also be considered. The existence of surface tension can be
expected from the difference in energy between molecules located in the bulk of a liquid and molecules
located at surface between the liquid and its vapor.
Let's pick a molecule from the bulk of the liquid phase. On the
average, it will be some average distance R
min
from its nearest
neighbor. If there are z
bulk
nearest neighbors, the energy of this
typical molecule from the bulk liquid is
e
bulk
= (z
bulk
/2)E
where the factor of 2 comes from assigning half of the energy of
interaction between a pair of atoms to each atom. For a hexagonal
closepacked array:
z
bulk
= 12
Now consider a molecule which resides at the interface between the liquid and the vapor. The same
basic formula can be written for the energy of this surface molecule:
e
surf
= (z
surf
/2)E
Except that the number of nearest neighbors will be significantly less. Again for a hexagonal close
packed array:
z
surf
= 6
then e
surf
≈ (1/2)e
bulk
Now e < 0 for attractive interactions, so:
06703 211 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
e
bulk
< e
surf
< 0
Thus molecules located at the surface do not have as low an energy as molecules in the bulk. So when
we calculate the free energy of the liquid, we should treat the surface molecules differently from the bulk
molecules. Indeed, it turns out that this contribution to the free energy of the system is proportional to
the number of surface molecules, which in turn is proportional to the surface area. The free energy per
unit area is called the surface tension:
γ
∂
∂
·
F
H
G
I
K
J
G
A
T P ,
BOUNDARY CONDITIONS FOR FLUID FLOW
Surface tension causes a number of effects including:
• makes homogeneous nucleation of bubbles virtually impossible: boiling is usually from
heterogeneous nucleation
• increases the vapor pressure of small drops for same T: Kelvin effect
• capillary rise
• determines the shape of drops and bubbles
These are all hydrostatic effects; there is no fluid flow occuring. Since surface tension depends on
temperature, temperature gradients can gradients in surface tension, which in turn can generate flow.
Since the origin of surface tension is intermolecular forces, which are very short range (range is a few
Angstroms), surface tension does not change the form of the basic equations of motion. Instead,
surface tension is manifested in the boundary conditions applied to those equations.
Recall that when surface tension effects are negligible, the general boundary condition on stress was
given by
γ≈0: n
.
T
I
= n
.
T
II
Surface tension makes both the normal and tangential components of stress continuous across an
interface. The more general boundary condition is
06703 212 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
n
.
(T
I
T
II
) = n(∇
s
.
n)γ  ∇
s
γ (224)
where ∇
s
is called the surface divergence; it represents two of the three
terms in the usual divergence. By convention, n is pointing from phase I
into phase II. For example, suppose we define a local rectangular
Cartesian coordinate system such that the zaxis is normal to the interface:
n = e
z
The usual divergence is given by
∇
.
v = ∂
x
v
x
+ ∂
y
v
y
+ ∂
z
v
z
The surface divergence does not include the term which represents the
derivative along the normal
∇
s
.
n = ∂
x
n
x
+ ∂
y
n
y
≡ 2H [=] m
1
This particular surface divergence represents the curvature of the
interface, often denoted by H or J.
Example: calculate the curvature of a sphere of radius r.
Solution: the unit normal to a sphere in spherical coordinates is
n = e
r
: n
r
=1, n
θ
=n
φ
=0
In this particular case, the normal is independent of the normal coordinate r so that the surface
divergence is the same as the regular divergence:
∇
s
.
n = r
2
∂
r
(r
2
⋅1) = 2/r
Thus the curvature is inversely proportional to the radius of the sphere.
Similarly the surface gradient represents that part of the usual gradient which lies in the surface
∇
s
γ = ∂
x
γe
x
+ ∂
y
γe
y
(225)
Suppose I dot both sides of (224) by the normal
T
zz
I
 T
zz
II
= 2Hγ
Thus the normal component of the stress is discontinuous at a curved surface. In the particular case of
hydrostatic equilibrium, there is no deviatoric contribution to the stress — just the hydrostatic pressure
hydrostatic equil: p
I
 p
II
= 2Hγ = 2γ/r
06703 213 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Index
2
2D Ilow 58
B
basis 3
Big "Oh" 121
body Iorces 38
Born repulsion 211
boundary layer 118
boundarylayer separation 154
C
Cartesian (implied) summation
convention 4
cavitation 168
closed 28
collision diameter 212
conservation oI circulation 44
conservative 32, 198
constitutive equation 75, 188
Constitutive Equation 37
Continuity oI stress 89
continuously diIIerentiable 4
Continuum Hypothesis 37
Coulomb's law 199
creeping Ilow 102
Cross Product 1
Curl 16, 17
curvilinear 10
D
d'Alembert's paradox 55
Debye length 201
derivable Irom a potential 31
deviatoric stress 71
Divergence 16
Divergence 17
Dot Product 1
double layer 197
doubledot product 2
drag coeIIicient 110
Dyadic Product 2
Dyadic Product 1
dynamic pressure 86
E
eddies 189
eddy 189
electric permittivity 199
electrochemical potential 198
electroosmosis 203
electrophoresis 203
electrophoretic mobility 208
electrostatic potential 198
equation oI continuity 27
equation oI state: 71
Euler's Equation 41, 70
E
Iluctuation 184
Ilux 73
Iorm drag 108
Iriction Iactor 195
Iriction velocity 193
G
GouyChapman model 201
gradient 5
Gradient 16, 18
gravitational potential 47
H
heat oI vaporization 212
hotwire anemometer 183
Huckel's equation 208
hydrostatic pressure 39
I
ideal Iluid 40, 117
06703 214 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Idem Eactor 33
Identity Tensor 33
innerouter product 2
instability 182
Invariant 25
irrotational 32
isotropic 66
K
Kelvin's Theorem 44
kinematic viscosity 94
Kronecker delta 3
L
Lamb 116
laminar Ilow 181
laminar sublayer 188, 194
Laplace's Equation 47
LennardJones 211
linear shear Ilow 98
linearly independent 3
M
matchedasymptotic expansions 114
material derivative 15
Material point 13
meanIree path 189
mixing length 189
mixing time 189
modulus oI elasticity Ior pure strain 77
momentum Ilux 72
N
NavierStokes Equation 82
Newton's law oI viscosity 81
Newton's Law oI Viscosity 80
no slip condition 83
O
Ohm's law 210
origin oI turbulence 182
orthonormal 3
Oseen 114
P
partial derivative 14
path independence 32
Path Line 61
persistence oI irrotationality 46
Poisson's Equation 200
Poisson's ratio 76
Poisueille's Eormula 88
position vector 4
potential Ilow 117
Prandtl's (universal) law oI Iriction 196
Prandtl's BoundaryLayer Equations 142
pressure 66
Primitive Matching Principle 126
Proudman & Pearson 114
purely elastic solids 76
R
rate oI deIormation 79
rate oI strain 80, 81
regular perturbation expansion 122
Reynolds Experiment 181
Reynolds lubrication equation 175
Reynolds number 103, 140, 182
Reynold's stress 186
Reynolds Transport Theorem 28
S
Scalar 1
scalar components 3
scalar Iield 4
scalar potential 31
scaling 138
second coeIIicient oI viscosity 81
secondary Ilow 101
secondrank tensor 3
sedimentation potential 203
separation 156
shear strain 77
singular perturbation expansion 122
singularly diIIerent 119
skin Iriction 109
Smoluchowski's equation 211
Smoluchowski's Equation 208
solenoidal 57
speciIic conductance 210
state oI stress 70
Stokes 111
06703 215 Spring, 2001
Copyright © 2000 by Dennis C. Prieve
Stokes Equation 103
Stokes Law 109
Stokes Paradox 116
Stokes Theorem 29
strain 76
Streak Line 61
stream Iunction 58
Stream Eunction 57
streaming potential 204, 209
Streamline 61
stress 66
stress tensor 67
stress vector. 67
stretch transIormation 123
surIace Iorces 39
surIace tension 84, 213
symmetric 77
Symmetric Tensor 33
T
Tensor 1
thermodynamic pressure 71
timeaveraged 184
total diIIerential 4
trace 71
transition zone 188
transpose 33
turbulent core 188, 194
turbulent stress 186
turbulent viscosity 191
U
uniIorm Ilow 49
unit dyads 3
unit nads 3
Universal Velocity ProIile 194
unsteady simple shear Ilow 72
V
van der Waals attraction 211
Vector 1
vector potential 57
velocity circulation 44
velocity potential 46
viscoelastic 38
viscous heating 66
viscous stress 71
W
Whitehead's Paradox 114
Y
Young's Modulus 76
Z
zero current 209
zeta potential 207, 210