You are on page 1of 83

Contents

4 The Equations of Fluid Motion 3


4.1 One-dimensional mass conservation and momentum balance equa-
tions: a rst look at the continuum equations for uid motion 3
4.2 Alternative viewpoint: Following a uid element The one-
dimensional Reynolds Transport Theorem . . . . . . . . . . . 6
4.2.1 The Leibniz formula . . . . . . . . . . . . . . . . . . . 6
4.2.2 One-dimensional mass conservation . . . . . . . . . . . 7
4.2.3 The transport theorem in one dimension . . . . . . . . 8
4.3 Acoustics: The study of small disturbances . . . . . . . . . . . 8
4.4 A one-dimensional approximation to a jet . . . . . . . . . . . 9
4.5 First remarks about kinematics . . . . . . . . . . . . . . . . . 10
4.5.1 Material derivative: the skydiver example . . . . . . . 10
4.5.2 The Lagrangian versus Eulerian representations of the
velocity eld . . . . . . . . . . . . . . . . . . . . . . . . 12
4.6 A useful physical interpretation of the material derivative . . . 16
4.7 Conservation of mass: The continuity equation . . . . . . . . . 17
4.7.1 The meaning of the velocity vector u . . . . . . . . . . 18
4.8 Reynolds Transport Theorem . . . . . . . . . . . . . . . . . . 19
4.8.1 Conservation of mass (again) . . . . . . . . . . . . . . 20
4.8.2 Special form of the Reynolds Transport Theorem . . . 20
4.9 Linear momentum balance . . . . . . . . . . . . . . . . . . . . 22
4.9.1 body forces . . . . . . . . . . . . . . . . . . . . . . . . 22
4.9.2 Surface forces . . . . . . . . . . . . . . . . . . . . . . . 23
4.9.3 Back to the linear momentum statement . . . . . . . . 24
4.9.4 The stress tensor and the state of stress . . . . . . . . 25
4.9.5 Cauchy stress equation of motion . . . . . . . . . . . . 27
4.10 Balance of Angular Momentum . . . . . . . . . . . . . . . . . 28
4.11 More Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . 30
1
4.11.1 Relative motion of nearby uid elements . . . . . . . . 30
4.12 Constitutive equations: a Newtonian uid . . . . . . . . . . . 34
4.12.1 Preliminary: Fluid statics . . . . . . . . . . . . . . . . 34
4.12.2 The stress tensor for uids in motion (the main event) 35
4.12.3 Mean normal stress in a uid (after Batchelor 1968) . . 39
4.12.4 On to the Navier-Stokes equation . . . . . . . . . . . . 40
4.13 Energy conservation in continua . . . . . . . . . . . . . . . . . 42
4.13.1 Mechanical Energy . . . . . . . . . . . . . . . . . . . . 42
4.13.2 Thermal Energy equation . . . . . . . . . . . . . . . . 43
4.13.3 Consequences of equilibrium thermodynamics . . . . . 44
4.14 Consequences of the second law: an equation for entropy vari-
ations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.15 Conditions for incompressibility . . . . . . . . . . . . . . . . . 48
4.16 The structure of conservation laws (something of a review) . . . 49
4.17 Appendix: Kinematics of material surfaces . . . . . . . . . . . 52
4.18 Appendix: Governing equations in component form . . . . . . 56
4.18.1 CONTINUITY EQUATION . . . . . . . . . . . . 56
4.18.2 NAVIER-STOKES EQUATIONS . . . . . . . . . 56
4.19 Appendix: The Navier-Stokes equations a little bit of history 61
2
4 The Equations of Fluid Motion
4.1 One-dimensional mass conservation and momentum
balance equations: a rst look at the continuum
equations for uid motion
Consider a gas of density that ows along a circular tube of constant radius R.
We shall denote the cross-sectional area A = (R
2
) and the axial direction as
z. In the simple one-dimensional analysis given here we neglect viscous eects
in the gas, and so assume that the velocity is uniform across the tube. We
ask how the uid changes its velocity u owing to a pressure gradient p/z
that acts along the tube. The case A = constant is treated rst and then
generalized to ow in a circular tube of (slowly) varying radius R(z, t) (or area
A(z, t)).
The ideas introduced now will later be generalized to three dimensions,
and expressed using vector notation. The inuence of uid friction (viscosity)
will also be introduced and a general discussion of viscous eects leads nat-
urally to the idea of stress expressed in terms of a second-order tensor. The
physical arguments and mathematical steps are similar to those encountered
in this one-dimensional example.
To begin the continuum analysis we introduce three unknowns (z, t),
u(z, t) and p(z, t), which enter the governing equations as
Physical statement unknowns in the equation
conservation of mass , u
linear momentum balance , u, p
equation of state , p
Table 1: An example of the conservation laws and corresponding unknowns
for the one-dimensional example described in the text.
The nal results will consistent of three equations for three unknowns,
, u and p. To make the analysis more general we should consider the tem-
perature variation (the temperature is then another unknown), which then
3
requires introduction of an energy equation.
Now consider a xed (control) volume in the tube as sketched in Figure
1. The control volume has width z, and front (z) and back (z + z) faces
of area A.
(z,t)
FLOW
FLOW
control
volume
u(z,t)
p(z,t)
p(z+ z,t)
Figure 1: The control volume selected to perform the one-dimensional mass
and linear momentum balances.
Conservation of mass is rst written in words and then expressed mathe-
matically as
time rate of change
of the mass in V =
rate of mass ow
into V
rate of mass ow
out of V (1a)

t
(.A.z)
. .
mass
= (.u.A) |
z
. .
mass/time
(.u.A) |
z+z
(1b)
Since we are assuming that the cross-sectional area is constant then dividing
through by z and taking the limit z 0 gives
1

t
+
(u)
z
= 0 , (3)
which is referred to as the one-dimensional continuity equation.
Next consider the change of linear momentum of the mass of uid
in the xed control volume V . This statement is Newtons second law for a
uid owing through a xed control volume; the linear momentum can change
due to advection (ow) and due to forces that act. We focus on changes of
momentum in the z-direction. The word, then mathematical, statements of
1
Recall the denition of the derivative
df
dz
= lim
z0
f(z + z) f(z)
z
. (2)
4
the balance of linear momentum are
time rate of change
of linear momentum =
net ux of linear
momentum into V +
forces acting on the surface
of the control volume (4a)

t
(.A.z.u)
. .
massvelocity
= (.u.A.u) |
z
. .
mass/timevelocity
(.u.A.u) |
z+z
+ (p.A) |
z
. .
force
(p.A) |
z+z
. (4b)
Assuming A is constant, dividing by z and taking the limit z 0
gives
(u)
t
+
(u
2
)
z
=
p
z
. (5)
When combined with the continuity equation, we now have two equations, but
there are three unknowns: , u and p. The problem statement is then closed
by using an equation of state, which, for an isothermal system, is a relation
for p(); see Table 1.
Exercise: Combine (3) and (5) to show that the linear momentum balance can
be written as

_
u
t
+ u
u
z
_
=
p
z
, (6)
which is a one-dimensional version of the (inviscid) Navier-Stokes equations,
which are more commonly called the Euler equations. Notice that each term
in equations (5) and (6) has dimensions force/volume.
Exercise: If the area A(z, t) of the tube is not constant but rather varies with
axial position z and time t, show that the continuity and linear momentum
equations can be written as
(A)
t
+
(uA)
z
= 0 (7a)

_
u
t
+ u
u
z
_
=
p
z
(7b)
Hint: Account for the pressure acting on all of the surfaces of the control
volume.
Two observations:
(i) In equations (6) and (7b), appears outside the derivative operators,
but notice that we have not assumed that the density is constant.
(ii) The units of each of the terms in the momentum equations (5), (6), and
(7b) are force/volume, which is a physical idea that is useful to keep in
mind.
5
4.2 Alternative viewpoint: Following a uid element
The one-dimensional Reynolds Transport Theorem
There is a conceptually useful alternative to think about conservation laws,
which is to consider a xed mass of material and follow it along its motion.
We then keep track of the time-rate-of-change of the quantity of interest (e.g.
mass, linear and angular momentum, energy, etc.). We refer to this approach
as the material rate of change. In the usual approach we form an integral of a
nite amount of some quantity and then calculate the time-rate-of-change the
integral. The three-dimensional version of this idea is called the Reynolds
Transport Theorem and is introduced below. First, we consider the one-
dimensional version, which is conceptually simpler and is analogous to the
Leibniz rule you likely saw in a calculus course.
4.2.1 The Leibniz formula
The one-dimensional illustration of this idea is the Leibniz formula from cal-
culus: time dierentiation of an integral whose limits depend on time is given
by
2
d
dt

b(t)
a(t)
f(x, t) dx =

b(t)
a(t)
f
t
dx + f(b(t), t)
db
dt
f(a(t), t)
da
dt
. .
boundary terms
. (9)
The two terms on the right-hand side are contributions from the function
evaluated at the end points and further depend on the rate of change of the end
2
The derivation of this result can be found in texts on advanced calculus. The basic idea
can be established by using the denition of the derivative as
d
dt

b(t)
a(t)
f(x, t)dx = lim
t0

b(t+t)
a(t+t)
f(x, t + t)dx

b(t)
a(t)
f(x, t)dx
t
(8a)
= lim
t0

b(t+t)
a(t+t)
(f(x, t + t) f(x, t)) dx
t
+ lim
t0

b(t+t)
a(t+t)
f(x, t)dx

b(t)
a(t)
f(x, t) dx
t
(8b)
=

b(t)
a(t)
f
t
dx + lim
t0

b(t+t)
b(t)
f(x, t)dx
t
lim
t0

a(t+t)
a(t)
f(x, t)dx
t
(8c)
=

b(t)
a(t)
f
t
dx + f(b(t), t)
db
dt
f(a(t), t)
da
dt
, (8d)
where the simplication to the nal result follows by the use of a Taylor series.
6
points. If x denotes position, each of these terms have the form (velocity f).
These boundary contributions are important to recognize as is emphasized next
when considering variations of properties following a uid element.
4.2.2 One-dimensional mass conservation
Consider a ow in one dimension as shown in Figure 1. We consider a uid
element with a label , which has instantaneous position z

(t), and speed


u

(t) =
dz
dt
. Suppose that we follow a xed mass of uid that moves with the
uid. If the end points of the domain, from left to right in Figure 1, are two
arbitrarily chosen points, z = a(t) and z = b(t), then the mass is obtained
by integrating the density between points a and b. Since this mass does not
change (the velocity of the end points are those of the uid), then
dm
dt
= 0, or
we can write the integral
dm
dt
= 0 =
d
dt

b(t)
a(t)
dz. (10)
Using the Leibinz rule (9) we can write
0 =
d
dt

b(t)
a(t)
dz =

b(t)
a(t)

t
dz + (b(t), t)
db
dt
(a(t), t)
da
dt
. (11)
We can write the last two boundary terms as The integral
un-does the
derivative. Also,
the velocity at
a(t), is u =
da
dt
and similarly at
the right
end-point.
(b(t), t)
db
dt
(a(t), t)
da
dt
=

b(t)
a(t)

z
(u) dz. (12)
Combining these results we have
0 =
d
dt

b(t)
a(t)
dx =

b(t)
a(t)

t
dz +

b(t)
a(t)

z
(u) dz (13a)
=

b(t)
a(t)
_
_
_
_

t
+

z
(u)
. .
=0
_
_
_
_
dz. (13b)
Since the limits of integration, a(t) and b(t), are arbitrary, we can conclude
that for the integral to be zero, the integrand must be zero. Hence, we have the
one-dimensional form of the continuity equation

t
+

z
(u) = 0. (14)
This equation is the same as (3), which we derived above using a xed control
volume.
7
4.2.3 The transport theorem in one dimension
The transport theorem relates to the rates of change of transportable quan-
tities such as mass, energy, linear momentum, entropy, etc. We considered
the case of mass above. We now treat this more general idea in one dimen-
sion. In general, we consider f(z, t) to denote a quantity/mass. Then, the
total amount of the quantity between x = a(t) and x = b(t) is obtained by
integrating f between the two arbitrary points a and b. Hence, the time rate
of change of this quantity following the uid is
d
dt
(some quantity) =
d
dt

b(t)
a(t)
fdz (15)
We use the Leibniz and the step shown above for converting boundary terms
to an integral involving the z-derivative of a function:
d
dt
(some quantity) =
d
dt

b(t)
a(t)
fdz (16a)
=

b(t)
a(t)

t
(f) dz +

b(t)
a(t)

z
(fu) dz. (16b)
Again, this quantity can be written as a single integral, so we obtain the
one-dimensional transport theorem:
d
dt
(some quantity) =

b(t)
a(t)
_

t
(f) +

z
(fu)
_
dz. (17)
If you wish, you can use the continuity equation (14) to simplify the
integrand of equation (17) to obtain the one-dimensional transport theorem
in the form
d
dt
(some quantity) =

b(t)
a(t)

_
f
t
+ u
f
z
_
dz. (18)
Exercise: If you were interested in linear momentum, then in one dimension
we keep track of u or f = u in the equations above. In this case, equations
(17) and (18) produce the left-hand side of one-dimensional linear momentum
equations discussed above.
4.3 Acoustics: The study of small disturbances
Sound propagation can be analyzed by considering small perturbations from
equilibrium. For example, consider a liquid or gas with density
0
and pressure
8
p
0
. Small disturbance from this equilibrium state can be written
(z, t) =
0
+ (z, t); p(z, t) = p
0
+ p(z, t); u(z, t) = u(z, t), (19)
where the indicates a disturbance quantity of small magnitude: | |
0
and | p| p
0
. Substituting into the continuity and momentum equations, lin-
earizing about the equilibrium state (this amounts to neglecting all quantities
involving products of small quantities) and using an equation of state p()
leads to the wave equation (you should show all of the steps)

2
f
t
2
=
dp
d
..
c
2

2
f
z
2
where f = , p, u. (20)
The speed of sound c characterizes the propagation of disturbances. For adi-
abatic motions of an ideal gas p

= constant, where is the ratio of the


heat capacity at constant pressure to that at constant volume, c =

p
0
/
0

330 m/sec at standard pressure and temperature.
The above approach can also be used to consider sound propagation in
a uniform ow of speed U. The steps involved in linearization of the complete
equations is a common approach to analyzing complicated problems beginning
with the full equations.
Figure 2: A jet of water issuing from a nozzle in a kitchen sink.
4.4 A one-dimensional approximation to a jet
As a nice every day example of the one-dimensional equations we consider a
jet issuing vertically downward from a nozzle, e.g. your kitchen sink as in
9
Figure 2. Let the radius of the nozzle be a
0
and the volumetric ow rate from
the nozzle be Q
0
. Also, we assume that the density of the liquid is constant.
The radius a(z) of the jet and the speed of the jet u(z) both change along
the length of the jet. So, this example is a case where the cross-sectional area
of the ow, A(z) = a(z)
2
, changes along the ow direction. We analyze the
steady motion using the continuity and momentum equations, and including
the gravitational body force that acts on the jet:
d
dz
(Au) = 0 and u
du
dz
=
dp
dz
+ g (21)
The continuity equation can be integrated to yield a
2
u = a
2
0
u
0
= Q
0
.
We now solve the linear momentum equation using the boundary con-
dition that the initial velocity is u(0) = u
0
=
Q
0
a
2
0
and the condition that the
surface pressure is p = p
0
, the atmospheric pressure, for all z (we neglect
surface tension). In this case, the pressure gradient dp/dz = 0. Hence, the
momentum equation is simply (after a small rearrangement) Thinking back to
your basic
physics, this
equation
expresses the
interconversion of
kinetic and
potential energy,
which given the
choice of signs, is
1
2
u
2
gz =
constant.
du
2
dz
= 2g. (22)
To simplify some of the algebra, it is convenient to rescale variables as
U =
u
u
0
, Z =
z
u
2
0
/(2g)

dU
2
dZ
= 1, with U(0) = 1 (23)
The solution of the ODE is
U(Z) = (1 + Z)
1/2
. (24)
The shape of the jet follows from a
2
= Q
0
/u(z) or
a(z)
a
0
= (1 + Z)
1/4
. (25)
We conclude that the radius has decreased by a factor of 2 on a length scale
Z = 15 or z =
15u
2
0
2g
. If we had a liquid jet issuing at a speed of 1/2 m/sec,
then this length scale is about 0.2 m or 20 cm.
4.5 First remarks about kinematics
4.5.1 Material derivative: the skydiver example
An important idea that is utilized frequently in uid dynamics is the material
derivative, which provides the variation with time of any quantity as measured
10
Figure 3: The skydiver example: A skydiver measures the atmospheric temper-
ature and we are interested in reporting this measurements relative to ground-
xed coordinates.
by an observer moving with the velocity u of the uid. The basic idea behind
this use of measurement is illustrated by the skydiver example.
To begin with, we know that the atmosphere is thermally stratied. So,
if a skydiver jumps out of an airplane and during descent measures the at-
mospheric temperature distribution, then although the measured temperature
varies with time, this information can be translated into a combination of the
spatial and temporal variations by recording the divers vertical position as a
function of time. Let the temperature distribution measured by the skydiver
be T
d
(t). An observer on the ground keeps track of the temperature distribu-
tion T(z, t) and, via communication with a diver whose instantaneous position
is z
d
(t), determines T(z
d
(t), t). Clearly, T(z
d
(t), t) = T
d
(t).
Let us denote the (Lagrangian) time derivative observed by the diver
as dT
d
/dt; we refer to d/dt as the total time derivative. Next, express the
result in terms of the temperature dependence recorded by the ground-based
(Eulerian) observer. Thus, using the chain-rule
dT
d
dt
=
dT(z
d
(t), t)
dt
=
T
t
+
dz
d
(t)
dt
T
z
, (26)
or introducing the vertical velocity with u
z
=
dz
d
(t)
dt
we have
dT
d
dt
..
total time derivative
=
dT(z
d
(t), t)
dt
=
T
t
..
local time derivative
+ u
z
T
z
. .
convective contribution
.
(27)
11
The right-hand side of (27) is the Eulerian representation of the time variation
of temperature measured by the skydiver moving with velocity u
z
(which may
depend on time).
For the fully three-dimensional problem where the velocity eld is u =
(u
x
, u
y
, u
z
) we are interested in T(x, y, z, t) and it should be apparent that (27)
generalizes to
dT
d
dt
=
T
t
+ u
x
T
x
+ u
y
T
y
+ u
z
T
z
=
T
t
+u T . (28)
The form of the right-hand side of (28) is very common and is called the
material time derivative, typically denoted D/Dt (or here d/dt), of a quantity
moving with velocity u(x, t):
material time derivative:
D
Dt


t
+u . (29)
Acceleration: The material derivative of the velocity (discussed below) is
Du
Dt
=
u
t
+u u . (30)
Exercise: Provide a physical interpretation of
Du
Dt
by describing the meaning
of
u
t
and u u.
4.5.2 The Lagrangian versus Eulerian representations of the veloc-
ity eld
Let us now inquire about ways to keep track of the uid velocity. Two ideas
are common, the Eulerian and the Lagrangian representations.
3
The former
proves to be more useful for performing numerical calculations, though the
latter is conceptually very helpful. In the Eulerian description all variables are
reported as a function of position x and time t, where x is measured relative
to some appropriately chosen origin; this is the usual case for elds (e.g. the
temperature or electric eld) as commonly measured in the laboratory. For
example, the velocity is denoted u(x, t). Next we consider the conceptually
useful Lagrangian description.
3
Both descriptions were originally given by Euler. For a short commentary, see the
historical article by Truesdell included at the end of these notes.
12
Figure 4: The Lagrangian labelling of uid particles for following uid motion.
By analogy with particle mechanics (e.g. cannon balls) we consider the
motion of small pieces of uid. We imagine labeling with dye, or a thin mem-
brane that moves with the local velocity, a small region of the uid and follow-
ing the motion of the center-of-mass. From such a trajectory we may calculate
the velocity of the particle and then the acceleration.
From now on we shall refer to a vanishingly small (from a continuum
viewpoint) volume of uid, containing the same mass of uid, as a uid particle
or a uid element. In the continuum mechanics description we cannot attach
a specic length scale to the size of the uid particle but we can proceed
with condence that the theoretical construct of a uid particle is in accord
with our everyday notions of mass, velocity and acceleration as discussed in
particle mechanics (note that deformation of the uid element is important
when resistance to uid motion is discussed).
Consider the motion of a uid element. We denote the position as
x = X

(t) or X(t; ) where the subscript labels the dierent particles (e.g.
conceptually it is convenient to imagine labeling particles by their initial po-
sition X(0)). The Lagrangian velocity, denoted u
L
, then corresponds to the
time-derivative of position for this uid particle,
u
L
(t; ) =
X(t; )
t
(particle label held xed) . (31)
Likewise, the Lagrangian acceleration is
a
L
(t; ) =
u
L
(t; )
t
=

2
X(t; )
t
2
, (32)
which we shall see is important when applying Newtons second law to the
motion of a uid. The alert reader will observe a diculty with the practical
13
implementation of this idea for reporting the velocity eld in most ows. In
order to make use of equations (31) and (32), the position as a function of
time must be obtained for all of the uid particles in the uid. In general, this
is not practical.
Let us express these Lagrangian ideas in terms of Eulerian variables, such
as the velocity eld u(x, t). In particular, the Lagrangian position X(t; ) is
such that
X(t; )
t
= u
L
(t; ) = u(X(t; ), t) (33)
as the uid particle occupies (Eulerian) position x = X(t; ) at time t. We
may then note that the corresponding acceleration (following a particle) is
obtained by using the chain rule:
a
L
(t; ) =
u
L
(t; )
t
=
u(X(t; ), t)
t
(34a)
=
_
u
t
_
x
+
X(t; )
t
u =
_
u
t
_
x
+u u , (34b)
where we have used (33) and further we have taken some care to emphasize
which variables are held xed. Although the particle acceleration seems a
rather rudimentary concept in basic mechanics, its representation via the Eu-
lerian variables is more involved than one might think at rst glance, and most
importantly involves the velocity eld in a nonlinear manner (u u).
4
We
will next see how to apply these ideas to the applications of Newtons second
law (F = ma) to uid parcels.
4
The time derivative measured by an observer moving with the local uid velocity u(x, t)
can be expressed with elementary calculus by considering the change of position x in time
t and taking the limit t 0:
a
L
= lim
t0
u(x + x, t + t) u(x, t)
t
(35a)
= lim
t0
_

_
u(x + x, t + t) u(x + x, t)
t
+
x
j
t
..
uj
u(x + x, t) u(x, t)
x
j
. .
u
x
j
_

_
(35b)
=
u
t
+u u . (35c)
14
Figure 5: The meaning of the convective contribution u to the material
time derivative. The sketch indicates how ow through a spatial gradient
dI/dx gives rises to a time rate of change of a quantity (here I) measured by
a moving observer.
Summary: The material derivative
D
Dt
=

t
+u (36)
expresses the rate of change measured by an observer translating with veloc-
ity u. Consequently, if we consider some property F
L
(t; ) associated with
uid element , then the time-rate-of-change of that property as the material
element is followed, expressed in terms of the Eulerian eld F(x, t), is
DF
L
(t; )
Dt
=
DF(x = X(t; ), t)
Dt
=
F
t
+u F (37)
Most importantly, the acceleration a of the uid element when expressed in
Eulerian variables is
a =
u
t
+u u . (38)
The term uu is nonlinear in velocity, and this is the principal diculty when
considering analytical solutions to the momentum equations (to be derived
shortly) that describe the motion of a uid.
15
4.6 A useful physical interpretation of the material deriva-
tive
It is important to reiterate the physical interpretation of the two contributions
to the total time derivative following a moving observer, equation (36). The
rst term on the right-hand side is the obvious term that arises from time
variations at a xed position in space. The second term is a consequence of
translation through a gradient, and this term contributes even if the eld is
time independent, i.e. () /t = 0. To interpret how translation u = 0 and
and a nonzero spatial variation of some quantity I, i.e. I, contribute to the
time rate of change, refer to the physical picture in Figure 5.
16
Figure 6: (a) Flow through a xed control volume V , with bounding surface
S and unit outward normal n. (b) A material control volume (xed mass)
translating and deforming through space as a result of motion of the uid.
4.7 Conservation of mass: The continuity equation
We rst derive a dierential equation describing the variation of density in the
uid. This equation couples (x, t) and u(x, t).
The standard derivation takes a xed volume V in space. Let n denote
the unit outward normal to the surface S. The rate at which mass enters a
surface element dS is u n dS (note that n u > 0 when material leaves
and < 0 when material enters; the dimensions of u n dS are mass/time); it
is +u n dS when material leaves the volume. Hence, conservation of mass
applied to the xed volume gives
d
dt

V
dV
. .
time rate of
change of mass in V
=

S
n u dS
. .
net ow of mass
(across S) into V
. (39)
Since the region V is xed, dierentiation and integration can be exchanged.
Using the Divergence Theorem we arrive at

V
_

t
+ (u)
_
dV = 0 , (40)
and, as the xed volume V is otherwise arbitrary, then we conclude that the
integrand must vanish everywhere in space:

t
+ (u) = 0 . continuity equation (41)
17
Remark: Rearrange (41) to identify the material derivative (D/Dt) of . We
see that

t
+u + u = 0
1

D
Dt
= u . (42)
The left-hand side is the relative time rate of change of uid density following
uid motion. This statement may be easier to visualize if we introduce the
specic volume (volume per unit mass)

V =
1
and write (42) as
1

V
D

V
Dt
= u , (fractional rate of change of volume of a material element)
(43)
which provides a physical interpretation of u. A non-zero divergence of the
velocity means that the density (or specic volume) of the material element
at that position is changing. How would you explain equations (42) and (43)
to a friend?
4.7.1 The meaning of the velocity vector u
It is useful to consider the general form of a conservation law for a scalar f,
which can represent some quantity per unit volume:
f
t
+ j = 0, (44)
where j is the ux vector for the quantity f. As can be shown using the deriva-
tion given in the previous section the ux of the quantity f (amount/area/time)
is n j where n is the unit outward normal to a surface element. If we consider
the density eld , then the ux, mass/area/time, may be used to dene the
velocity eld: j = u. We can imagine an experiment where we measure j and
so this identication denes the velocity vector u. Because of the denition
of linear momentum as the product of mass and velocity, then we may also
view this denition as u as prescribing the linear momentum/mass of uid.
Alternatively, the mass ux density vector j = u can be interpreted as the
linear momentum per unit volume of uid.
Next, let r denote the position vector. Note that multiplying the conti-
nuity equation by r yields
(r)
t
+r j = 0. (45)
18
With the identity (show this) rj = (jr)j, and integrating the continuity
equation over some volume V , then (Landau & Lifshitz, section 50)
d
dt

V
r dV =

V
j dV, (46)
where we have assumed that on all bounding surface n jr decays suciently
fast for any boundary contributions to vanish. Equation (46) has a natural
physical interpretation consistent with Newtons second law since the left-
hand side is the time rate of change of the center-of-mass position while the
right-hand side and the right-hand side is the total momentum in V .
4.8 Reynolds Transport Theorem
In order to think further about conservation laws, it is useful to consider
volume integrals of linear and angular momentum, energy, etc., and to think
about the material rate of change of such integrals (i.e. the time rate of change
following a xed mass that moves with the uid). The volume integral is over
a nite region, and the domain can change shape as the velocity eld may have
gradients. Hence, we consider time rates of change of integrals whose limits
depend on time.
The transport theorem in three dimensions: The generalization of equation
(9) to three-dimensions is needed in continuum mechanics, where it is known
as the Reynolds Transport Theorem. Therefore, we consider the time rate of
change following a small material volume element V
m
(t) whose surface points
move with the local uid velocity u(x, t). The transport theorem states
d
dt

Vm(t)
f(x, t) dV =

Vm(t)
f
t
dV +

Sm(t)
n uf dS (47a)
=

Vm(t)
_
f
t
+ (uf)
_
dV (47b)
Written in the order shown above this formula holds forany function f, be it
a scalar, vector or tensor quantity. Equation (47a) is the three-dimensional
version of (9).
5
A geometric proof of this result is given in the handout from
Whitaker.
5
A proof of (47) follows by viewing each of the dV elements as small material volumes,
and dierentiating (following material points, so d/dt D/Dt) the left-hand side of (47a)
19
4.8.1 Conservation of mass (again)
As a preliminary, reconsider the time rate of change of mass by following the
motion of an arbitrarily chosen material volume V
m
(t). The volume contains
(instantaneously) a mass equal to

Vm(t)
(x, t) dV , which is conserved by def-
inition of V
m
. Applying the Reynolds Transport Theorem,
d
dt

Vm(t)
dV = 0

Vm(t)
_

t
+ (u)
_
dV = 0 . (50)
Since the volume element is arbitrary, then we arrive at the continuity equa-
tion:

t
+ (u) = 0 . (51)
The continuity equation relates two unknown quantities, (x, t) and u(x, t),
and so we must proceed onwards to develop more equations.
4.8.2 Special form of the Reynolds Transport Theorem
It is common to keep track of material properties by reporting them on a per
mass basis. Since the volume integrals above involve f dV , then the property
f has dimensions property per volume. If we let f = F, then F has
dimensions property per mass. If we substitute f = F in equation (47)
then you should be able to show
d
dt

Vm(t)
F(x, t) dV =

Vm(t)

Vm(t)

_
F
t
+ (uF)
_
dV =

Vm(t)

DF
Dt
dV,
(52)
using the chain rule:
d
dt

Vm(t)
f(x, t) dV =

Vm(t)
_
Df
Dt
dV + f
D(dV )
Dt
_
. (48)
We need
D(dV )
Dt
. Since we may view the volume dV as a set of points with boundary S,
the time rate of change of volume follows
D(dV )
Dt
=

S
u n dS =

dV
udV dV u,
where the last identity follows in the limit that the volume tends to zero. Thus, the specic
volume, here equivalent to dV , evolves as
D(dV )
Dt
= dV u, and so (48) can be simplied
to
d
dt

Vm(t)
f(x, t) dV =

Vm(t)
_
Df
Dt
+ f u
_
dV , (49)
which can be rearranged to give equation (47b).
20
which is referred to as the special form of the Reynolds Transport Theorem.
We shall nd it useful on several occasions below since most transport relations
involve a property per mass.
21
4.9 Linear momentum balance
We now follow a material volume element, which by denition contains xed
mass and apply other conservation, or balance, laws. Newtons second law
relates the change of linear momentum (mass times velocity) of the material
to the forces acting on the moving domain V
m
(t). In words we examine
_
_
_
time-rate-of-change of
linear momentum in the volume
_
_
_
=
_

_
sum of all the forces (e.g. due to surface
+ body forces) acting on the material
_

_
.
(53)
The linear momentum of this element is simply

Vm(t)
u dV . As for the forces
that act, in continuum mechanics we recognize two kinds, surface forces and
volume or body forces.
4.9.1 body forces
Body forces are long range forces which act (perhaps unequally) on all the
material points that constitute the body. These forces arise from the material
being placed in a force eld and so include gravitational (the most familiar),
as well as magnetic and electric forces.
Let f (x, t) denote the body force/mass. Then, the total body force acting
on the domain is

Vm(t)
f dV . Examples of the body force include
(i) gravity: f = g, where g is the gravitational acceleration.
(ii) ferrouids: example of a magnetic body force see the Scientic Amer-
ican article by Rosensweig attached at the end of this section.
(iii) electrohydrodynamics: When an ionic uid (e.g. salt water) is placed
adjacent to a surface, invariably the surface adsorbs some charge and
there remain mobile ions of opposite charge (counterions) in the uid
adjacent to the surface. These counterions move in the presence of an
applied electric eld tangential to the surface. If
e
(x, t) is the electric
charge density/volume, and E(x, t) is the local value of the electric eld,
then there is a body force/volume
e
E on the uid.
22
This eect is responsible for the electrophoretic motion of particles in
liquid (colloidal science), and is the basis for electro-osmotically driven
ows.
(iv) magnetohydrodynamics: when a material carries an electric current (j =
current density) in a magnetic eld B, there is a (Lorentz) body force/volume
given by j B.
4.9.2 Surface forces
These are short range force, which represent two types of molecular scale
interactions:
(i) intermolecular molecular forces between nearby uid molecules. It is
known that there is short range molecular ordering in liquids and conse-
quently a force must be applied to slide one layer of molecules relative
to another, i.e. relative macroscopic motion requires a force to be ex-
erted. Because continuum mechanics cannot resolve such molecular scale
phenomena, we model this as a force/area exerted across a surface, and
this eect is one of the essential contributions to friction (i.e. viscosity)
in uids.
(ii) linear momentum exchange due to random thermal motions (uctuations
about the mean). A good example of this is a billiard ball gas consisting
of hot (fast) and cold (slow) atoms or molecules. When fast and slow
particles exchange places across a surface marked in the material, then
the linear momentum on each side is changed. By Newtons second law
the change of linear momentum is equivalent to the action of a force from
one side of the surface on the other side.
Notation for surface forces: If we now consider a small surface S separating
two regions of uid, we write the force/area, or stress vector, exerted by uid
on one side of the surface on the other side as
lim
S0
F
(n)
S
= t
(n)
stress vector . (54)
The subscript n reminds us that in general we should expect the magnitude
and direction of this force to depend on the orientation of the surface chosen
for consideration. The stress vector, t
(n)
(with dimensions force/area), is an
important quantity to think about in continuum mechanics and we will have
23
Figure 7: Notation: (a) the force F on a surface element S with unit
normal n; (b) A volume element in the shape of a pill box with thickness .
Use a force balance, and let 0 to demonstrate t
(n)
= t
(n)
.
to undertake a few steps to relate it to details of the velocity eld. For now
we just assume that there is such an internal stress eld in a material
4.9.3 Back to the linear momentum statement
We return to a linear momentum balance and use the word statement of New-
tons second law (53) applied to a material volume element, which leads to
d
dt

Vm(t)
u dV =

Vm(t)
f dV +

Sm(t)
t
(n)
dS . (55)
The left-hand side can be rewritten using an identity for the Reynolds Trans-
port Theorem (see exercises): we use the identity
d
dt

Vm(t)
u dV =

Vm(t)

Du
Dt
dV

Vm(t)

Du
Dt
dV
. .
mass acceleration
=

Vm(t)
f dV
. .
body force
+

Sm(t)
t
(n)
dS
. .
surface force
, (56)
24
where Du/Dt is the acceleration following the motion of a small uid element.
This formula, though nice conceptually, is useless until we say more about the
interpretation of t
(n)
.
4.9.4 The stress tensor and the state of stress
We shall now introduce some general features of the state of stress in a contin-
uum. The arguments illustrate that the stress vector t
(n)
must have a special
form: t
(n)
= n T, where the stress tensor T is a second-order tensor.
6
Cauchy
is responsible for introducing the conceptual framework of the stress tensor.
It is a basic feature of all of continuum mechanics.
7
6
A special case useful when viscous eects may be neglected (as we shall learn, this cor-
responds to high-Reynolds-number ows away from boundaries): The mathematical steps
are illustrated nicely. Suppose that the only forces transmitted across a surface are normal
forces associated with pressure. Then, t
(n)
= np(x, t), and the integral linear momentum
equation is

Vm(t)

Du
Dt
dV =

Vm(t)
f dV

Sm(t)
np dS

Vm(t)
_

Du
Dt
f +p
_
dV = 0, (57)
where the second equality follows from use of the Divergence Theorem. Since the material
volume V
m
(t) is arbitrary, the kernel must vanish everywhere:

Du
Dt
= f p Eulers equation. (58)
7
For a historical comment on this point, it is useful to quote Truesdell: (p. 236, 238 of
The Creation and Unfolding of the Concept of Stress, Essays in the History of Mechanics,
Berlin: Springer-Verlag, 1968, p. 184-238, by C. Truesdell)
From the above account, it is clear that every conceptual element in Cauchys theory
was to be found in one or another of the special theories constructed in the previous century.
Moreover, in researches of Fresnel done in 18211822, with which Cauchy must certainly
have been familiar, many of Cauchys results are more or less implied, although in Fresnels
work the concepts of stress and strain are always connected through a presumed linearly
elastic response.
Thus it might seem that Cauchys achievement in formulating and developing the general
theory of stress was an easy one. It was not. Cauchys concept has the simplicity of genius.
Its deep and thorough originality is fully outlined only against the background of the century
of achievement by the brilliant geometers who preceded, treating the special kinds and cases
of deformable bodies by complicated and sometimes incorrect ways without ever hitting
upon this basic idea, which immediately became and has remained the foundation of the
mechanics of gross bodies.
Nothing is harder to surmount than a corpus of true but too special knowledge; to reforge
the tradition of his forebears is the greatest originality a man can have.
25
Figure 8: The state of stress as exhibited by showing the components on the
6 faces of a cube about the point of interest. Here the stress is denoted with
instead of T.
Stress equilibrium: If the volume of a small material element is O(
3
) and we
let 0, then we see that volume integrals O(
3
) vanish more quickly than
surface integrals O(
2
); see equation (56). Therefore,
lim
0
1

Sm(t)
t
(n)
dS 0 . (59)
Apply to a pill-box shaped element (see gure): t
(n)
= t
(n)
.
Cauchy stress tetrahedron: t
(n)
= n T.
The components of the stress tensor, T
ij
, have very specic physical
interpretations: T
ij
denotes the force per unit area in the j-direction on
a face with normal in the i-direction.
26
4.9.5 Cauchy stress equation of motion
If we substitute t
(n)
= n T into (56) then we have

Vm(t)

Du
Dt
dV =

Vm(t)
f dV +

Sm(t)
nTdS

Vm(t)
_

Du
Dt
f T
_
dV = 0 ,
(60)
where the second equality follows by using the Divergence Theorem on the
right-hand side. Since V
m
(t) is otherwise arbitrary, then we conclude

Du
Dt
= f + T Cauchy stress equation of motion , (61)
which is known as the Cauchy stress equation of motion. This equation has
an enormous range of applicability since it applies to all continua. It remains
to relate the stress tensor T to the gradients of the velocity eld.
Applicability of the Cauchy Stress Equation: It is worth noting that equation
(61) applies equally well to any material that can be treated within the con-
tinuum approximation. Essentially, this means that the scale of interest for
mathematical and physical descriptions is large compared to the typical scale
of constituents that make up the material. Hence, although we have intro-
duced the continuum ideas thinking about single-phase uids, the ideas apply
equally well to materials such as sand, uid-saturated soil and other porous
systems, etc. Of course, where the distinct character of the material enters is
in the constitutive equation that relates the stress (T) to the velocity or strain
and their gradients.
27
Figure 9: The notation used to keep track of quantities for a moving control
volume.
4.10 Balance of Angular Momentum
The angular momentum balance applied to a given mass (the material control
volume) is
_
_
_
time-rate-of-change of angular
momentum in the material volume
_
_
_
=
_
_
_
sum of all the torques,
(e.g. due to surface + body forces)
_
_
_
.
(62)
Recall that the angular momentum of a particle is dened as the moment of
the particles linear momentum. By analogy with the moment of a force, the
angular momentum is sometimes referred to as the moment of momentum.
Relative to an origin from which we measure position x of a point in the
volume, we write the integral balance for the moment of linear momentum of
a material volume as
d
dt

Vm(t)
xu dV =

Vm(t)
xf dV +

Sm(t)
xt
(n)
dS+
body
couples +
surface
couples ,
(63)
where the last two terms, motivated by analogy with the corresponding quan-
tities in the linear momentum balance, indicate possible sources of angular
momentum. The magnitudes of these two terms are not signicant for ordi-
28
nary liquids (we comment on these two sources later, likely on homework # 3;
see also the attached reprint from Nature, Angular momentum of continua
by Dahler and Scriven). Therefore, in the absence of such exotic contributions,
we can use the special form of the Reynolds Transport Theorem (equation 52)
to simplify the left-hand side of (63), and substitute t
(n)
= n T. We obtain:

Vm(t)

D
Dt
(x u) dV =

Vm(t)
x f dV +

Sm(t)
x (n T) dS . (64)
Since x here indicates material points,
Dx
Dt
= u, but then u u = 0. Also,
applying the Divergence Theorem to the surface integral on the right-hand
side of (64) yields

Vm(t)
x
Du
Dt
dV =

Vm(t)
x f dV

Vm(t)
(T x) dV . (65)
Next note the vector identity (T x) = x ( T) +T : , where
is the third-order permutation tensor; also T : = T
ij

jik
e
k
. After minor
rearrangement we have

Vm(t)
x
_

Du
Dt
f T
_
dV =

Vm(t)
T : dV , (66)
The kernel on the left-hand side is zero by the Cauchy stress equation of
motion, and since the material volume V
m(t)
is chosen arbitrarily, then T :
= 0. We conclude from this that T
ij
= T
ji
or that T = T
T
, i.e. T is
a symmetric second-order tensor and has only six independent components.
Notice that this is a very general conclusion about any continua, independent
of the molecular constitution, and holds so long as local couples (torques) are
not present.
Therefore, the linear and angular momentum balances have produced

Du
Dt
= f + T and T = T
T
. (67)
There are six independent components of the stress tensor T. It remains
to relate these components to the velocity eld and in particular the gradients
of the velocity eld.
29
1961 Nature Publishing Group
1961 Nature Publishing Group
Figure 10: Relative motion of two nearby uid elements.
4.11 More Kinematics
The stress in a uid describes forces exerted by one portion of the uid on
adjacent portions. This stress is expected to depend on the rate at which the
material is being deformed. As a preliminary we discuss the deformation of
a uid. The mathematical description of relative motion and displacement of
material points is called kinematics.
4.11.1 Relative motion of nearby uid elements
Begin by considering the relative motion of two nearby uid elements. In
gure 10 we denote by r the small vector element connecting two nearby
uid elements, P and Q. The position of uid element P is denoted by x.
Expanding the velocity at Q (position x + r) in a Taylor series about P then
we have
8
u(x+r) = u(x) +r u(x) +
1
2
rr : u(x) +. . . (Taylor Series) (70)
8
For a function of a single variable, the Taylor series is
f(x + x) = f(x) + x
df
dx
|
x
+ (68)
Likewise for a function of two variables,
f(x + x, y + y) = f(x, y) + x
f
x
|
x,y
+ y
f
y
|
x,y
+ (69)
By identifying r = (x, y), we see that the rst two terms of (69) are shorthand for the
rst two terms on the right-hand side of (70).
30
The relative velocity u, correct to O(r
2
), is then
u(r) = u(x +r) u(x) = r u(x) , (71)
i.e., the relative motion is given by the velocity gradients, which is expected
since, if there were no changes of velocity with respect to any direction, then
the uid is in uniform motion.
The velocity gradient tensor u contains information describing relative
motion of nearby uid elements. Now decompose u into symmetric and
anti-symmetric parts:
u =
1
2
_
u +u
T
_
. .
symmetric
+
1
2
_
u u
T
_
. .
anti-symmetric
(72a)
= E + . (72b)
E is known as the rate-of-strain tensor and as the vorticity tensor. This
decomposition of the velocity gradient tensor into its symmetric and anti-
symmetric parts, E and , respectively, is an important and useful idea. We
next proceed to geometric interpretations of E and .
The vorticity tensor, : Since is, by construction, an anti-symmetric tensor We introduce the
third-order tensor
=
ijk
e
i
e
j
e
k
it has only three independent components. Hence can be represented in
terms of a vector, say , as
=
1
2
. (73)
Write down the components of this expression: Note that index
notation
=
ijk

k
e
i
e
j

ij
=
1
2
(
u
j
x
i

u
i
x
j
) =
1
2

ijk

k
(74)
For k = 1 we have

1
=
u
3
x
2

u
2
x
3
, (75)
which is the 1component of the vector u. The same is true for the other
components, so we have = u. We summarize the above manipulations
by writing the relationships between the vorticity tensor , the vorticity vector
, and the velocity vector u as
=
1
2
, = : , = u . (76)
We will now show that the vorticity vector has a natural interpretation in
terms of the local angular velocity of the uid.
31
Figure 11: Vorticity as local rotation.
Figure 12: Two ows with circular streamlines. (a) A ow that lacks vorticity
(r = 0). (b) A rigid-body rotation, which is a ow with vorticity.
Exercise: Conrm the identities in (76).
Returning to the Taylor series (71), the relative velocity contribution due
to is
r =
1
2
r ( ) =
1
2
r
i

ijk
e
j
=
1
2
r , (77)
which is a rigid body rotation, at Q and taken about P, with angular velocity
1
2
: the vorticity vector equals half of the local angular velocity of the uid,
as sketched in Figure 11.
It is easy to confuse the true meaning of this interpretation of vorticity.
An important point to remember is that it describes local variations and, in
particular, local rotations. Vorticity does not simply represent (circular) mo-
tion along a closed path. Rather, vorticity corresponds to changing orientation
of a uid element in space. This distinction is made clear in gure 12.
The rate-of-strain tensor, E: To begin with, from (71) we have
u = r E +r . (78)
32
Taking the inner product with r we have
r
Dr
Dt
= r E r (79)
where r r = 0 since is anti-symmetric. Also, we have written the
relative velocity of the two points as the (Lagrangian) time-rate-of-change of
position moving with point P. Equation (79) may be written in terms of the
rate-of-separation r r = r
2
of two nearby uid elements: The time-rate-of-
change of
separation
distance is given
in terms of the
rate-of-strain
tensor E.
1
2
D(r r)
Dt
= r E r , (80)
which is completely determined by E.
Remark 1: Since E is symmetric, this tensor may be diagonalized. The eigenvectors
are called the principle directions of strain and a uid element that begins
aligned along one of the eigenvectors remains aligned for all time; hence,
the interpretation of E as representing a pure straining motion.
Remark 2: Symmetric and traceless version for E Show that trE = E
ii
= u.
So,
E =
_
E
1
3
( u) I
_
+
1
3
( u) I (81)
Remark 3: Simple shear ow, u = Gxe
y
, where G is the shear rate interpret as
equal parts rate of strain and vorticity.
33
Figure 13: Statics: The normal stresses (pressure) on a surface.
4.12 Constitutive equations: a Newtonian uid
4.12.1 Preliminary: Fluid statics
Consider the stresses acting in a static uid, i.e. u = 0. Assuming that the
body force is gravitational, f = g, the stress equations of motion reduce to
T+ g = 0 . (82)
The standard denition of a uid is a material that cannot support tan-
gential (shear) stresses without motion occurring. However, tensile/compressive
or normal forces can be supported. It then follows that the only stresses acting
in a uid at rest must lie normal to any surface (see Figure 13) and we call
this stress the static (equilibrium) uid pressure, p
e
. So,
t
(n)
= np
e
, (83)
where the minus sign is introduced since we consider pressure as compressive,
so acting opposite the outward normal n. It follows that the stress tensor has
the form
t
(n)
= n T = np
e
T = p
e
I. (84)
Substituting (84) into (82) gives
Fluid statics: p
e
+ g = 0, [can integrate if (p
e
) is known]
(85)
which is the governing equation for uid statics. This equation may be used to
calculate forces on stationary objects like dams, to determine the equilibrium
34
Figure 14: Violation of the state of rest.
of submerged bodies, to calculate the shape of static uid interfaces such as
pendant bubbles, sessile drops, etc.
9
Remark: The pressure in equations (83-85) is the same quantity as discussed
in equilibrium thermodynamics. Hence, for a single phase uid, p
e
is related
to the density and temperature of the uid via an equation of state, e.g.
p
e
= p
e
(, ).
Before proceeding notice that the pressure in (84) clearly satises The trace of a
second-order
tensor is denoted
trT.
p
e
=
1
3
trT. (recall trT = T
ii
= T
11
+ T
22
+ T
33
) (86)
Violation of the state of rest: Equation (85) allows determination of a sucient
condition for convection u = 0 to occur. Taking the curl of (85) we have
( p = 0)
g = 0 (g = constant). (87)
Hence, for there to be a state of rest so that equation (85) is satised, we
require g. This means that any density variations must be such that the
direction of variation is collinear with g. Furthermore, everyday experience
tells us that the situation is only stable provided the density increases in the
direction of g. A case that illustrates the violation of this condition is shown
in gure 14, where a heated wall creates a temperature gradient, and hence a
density gradient, in the direction perpendicular to g; uid motion must occur
here.
4.12.2 The stress tensor for uids in motion (the main event)
We consider the stresses acting in a uid in motion. In this case, the simple
concept of an equilibrium pressure acting equally in all direction is lost.
It is useful to recall the introduction to viscosity given in elementary
textbooks. There an experiment is described where a at plate is dragged at a
9
Atmospheric pressure: A standard application is the vertical pressure distribution in
the atmosphere, which follows from
dp
dz
= g, where z is measured vertically upwards and
the pressure is linked to the density by an equation of state (p).
35

yx
U
H
y
U
x
= U y H
Figure 15: A simple shear ow.
velocity U above a stationary plate separated by a distance H (see gure 15).
The force/area exerted on the plate (y-plane) in the x-direction is denoted
yx
and is assumed to vary linearly with U. The relationship is written

yx
=
U
H
=
du
x
dy
, (88)
where the material coecient is the (shear) viscosity. We now extend this
description to more general ow situations, though retain the basic linearity
with the velocity gradient as in (88).
Begin by recalling the simple form of the stress tensor for a static uid.
The constitutive equation we propose should reduce to the static case T =
p
e
I when there is no motion. So, we write
T = p
e
I + ( is called the deviatoric stress) , (89)
where = 0 when u = 0. Clearly, should be symmetric since T is symmetric.
We are now ready to relate the nonequilibrium stress to the velocity
eld and, in particular, to gradients of the velocity. First, begin with some rea-
sonable assumptions that are representative of a wide class of simple (typically
low molecular weight) uids.
(i) We do not expect the state of stress to depend on the translation of the
reference frame, therefore we do not expect to depend explicitly on
u. In other words, two observers translating relative to one another at
a uniform velocity must measure identical stresses. Hence, we expect
to be a function of the velocity gradients. We write = (u), or
36
equivalently think of as dependent on the rate-of-strain tensor E and
the vorticity tensor .
(ii) We do not expect the stress response of a uid to be aected by a local
rigid body rotation. This implies that the stress tensor is independent
10
of and depends solely on E, i.e. = (E).
(iii) Assume that the relationship between and E is local and instantaneous,
i.e. it depends only on the value of E at the point of interest x at time
t. Nevertheless, structured materials (e.g. polymers) typically have a
stress response that depends on the history of the ow.
(iv) Linearity: Assume that, similar to simple constitutive equations you have
studied previously (e.g. Hookes law, Ohms law, Ficks law, Fouriers
law), the relationship between the second-order tensor and the second-
order tensor E is linear.
11
We then make a mathematical statement: the most general linear rela-
tionship between a second-order tensor and a second-order tensor E is
of the form
= A : E
index notation

ij
= A
ijkl
E
lk
. Note: A characterizes the material.
(90)
This equation is accounting for the stress to depend on all possible linear
combinations of the nine components of the rate of strain tensor. Since
10
This assumption frequently raises questions! First, note that Batchelor (An Introduc-
tion to Fluid Dynamics, p. 144) comments in a footnote (most appropriate since this is a
footnote) that It is taken for granted, in most expositions of uid dynamics, that a devia-
toric stress cannot be generated by pure rotation, irrespective of the structure of the uid,
simply on the grounds that there is then no deformation of the uid; however, rigorous
justication for this belief is elusive.
Second, if we did assume a dependence on , so also a dependence on the vorticity vector
, then we should not expect a dierence in the state of stress if the body rotated one
direction or another. Therefore, the constitutive equation would at most involve terms pro-
portional to the square of and such terms would not enter into a linear theory (assumption
(iv) above).
11
In heat conduction, or diusive mass transfer, you begin with a conservation statement
for a scalar quantity (say, energy or mass) in the form

t
= q, where q is the ux
vector. In a stationary uid, you are familiar with the simple idea (Fouriers law for thermal
conduction or Ficks law for mass diusion) that q = , from which you arrive at the
heat equation,

t
=
2
(with = constant). However, a more general linear constitutive
statement is q = , where is a second-order conductivity tensor. For isotropic
materials, = I.
37
and E are symmetric we lose no generality by assuming
A
ijkl
= A
jikl
and A
ijkl
= A
ijlk
. (91)
(v) The nal condition necessary to complete the constitutive relationship
is to assume that the material is isotropic, which means that there is
no preferred direction in the material. Physically this means that two
observers who are simply rotated relative to another will characterize the
material deformation (i.e., the tensor A) in identical ways. Mathemati-
cally, then we expect the tensor A to be an isotropic fourth-rank tensor.
Although the components of vectors and tensors generally depend on
the choice of coordinate system, isotropic tensors have the property that
their components are the same in all cartesian reference frames. The
most general fourth-order tensor is a linear combination formed from
ij
(the only second-order isotropic tensor), hence
A
ijkl
=
1

ij

kl
+
2

ik

jl
+
3

il

jk
. (92)
It follows from (91) that
2
=
3
so A
ijkl
=
1

ij

kl
+
2
[
ik

jl
+
il

jk
].
Then,

ij
= A
ijkl
E
lk
=
1

ij
E
ll
+ 2
2
E
ij
(93)
or in vector notation (use the identity trE = E
ii
= u)
=
1
( u)I + 2
2
E. (94)
We have thus arrived at an expression for the stress tensor (let
2
= )
T = (p
e

1
u) I + 2E. (95)
The coecient is known as the shear viscosity. We now have a useful
constitutive expression relating the stress tensor in the uid to the local
velocity gradients.
Before proceeding further we note that (95) may be recast in terms of
E
1
3
( u) I, which is the local rate of stretching above any local
dilatation or compression. Hence we write the constitutive relationship The state of stress
for a Newtonian
uid is linear in
the velocity
gradient and
involves two
material
constants, the
shear viscosity
and the bulk
viscosity .
T = (p
e
u) I + 2
_
E
1
3
( u) I
_
where =
1
+
2
3
.
(96)
is referred to as the bulk, volume or second coecient of viscosity.
Both (95) and (96) relate T to u and involve two material parameters,
38
and . Both parameters can be shown to always be 0 by using an
argument based on the second law of thermodynamics.
An aside: In the classical theory of elasticity there are also two param-
eters, the Lame constants, that appear in the stress-strain relationship
for an elastic material.
4.12.3 Mean normal stress in a uid (after Batchelor 1968)
Before proceeding further it is useful to discuss the mean mechanical
pressure, which refers to the physical interpretation of the force/area as
characterized by the intuitive notion that pressure represents the local
squeezing force. Since a uid in motion is not a thermodynamic equi-
librium system, this discussion provides a connection with what would
be recorded by an instrument capable of measuring normal forces. For
example, consider the average of the normal stress nt
(n)
over the surface
of a small spherical volume surrounding the point x:
1
S

S
nn : T dS. (97)
If the spherical surface has radius , then we may expand T in a Taylor
series about x, T(x + n) = T(x) + n T(x) + . . ., which leads to
1
S

S
nn : T dS =
1
4
2

S
nn dS : T(x) + O() . (98)
The integral of nn over a spherical surface must be an isotropic second-
order tensor, hence equals cI. The constant c is found by taking the
trace so that Recall
trI =
ii
= 3
cI =

S
nn dS
trace
3c = 4
2
c =
4
3

2
. (99)
Hence,
1
S

S
nn : T dS =
1
3
I : T =
1
3
tr T, (100)
represents the average normal stress over a surface surrounding x and we
use this as a mechanical denition of the pressure, p, in a moving uid.
Accounting for the usual notion of pressure as compressive we dene Mechanical
denition of
pressure: the
average normal
stress, p =
lim
S0
1
S

S
nn :
T dS
p =
1
3
trT. (101)
39
If you think physically about pressure in uids in motion, you really
are considering p, which is not, in general, equal to the thermodynamic
pressure. Also, notice that this equation has the same form as equation
(86), but now the stress tensor T accounts for uid motion.
The relationship between p
e
and p: From (96) we see that
trT = 3p
e
+ 3 u (102)
or
p p
e
= u . (103)
It is instructive to view (103) as a linear perturbation away from equi-
librium where u is the only scalar invariant of the velocity gradient
tensor that is linear in u (a property is termed invariant if it is in-
dependent of the choice of cartesian coordinate system used; it turns
out that there are three invariants for a second-order tensor). Also, we
expect on physical grounds that in a pure expansion u > 0, so that
the measured pressure would be less than the equilibrium pressure, and
this result is consistent with the form of (103).
Exercise: Evaluate the following integral over a spherical surface S:

S
nnnn dS. Hint: The result should be an isotopic fourth order tensor.
4.12.4 On to the Navier-Stokes equation
We can now put all of the above results together to obtain

t
+ (u) = 0 , (104a)

Du
Dt
= g + T and T = T
T
(104b)
T =
_
p
e
(
2
3
) u
_
I + 2E . (104c)
If the ow is isothermal, then p
e
() only and the above corresponds
to 10 equations for the 10 unknowns (, u, T). On the other hand, if
temperature variations occur in the uid, then we have an additional
unknown, the temperature, and we must incorporate the energy equation
(discussion in the next section).
Some standard simplications:
40
(a) If the density is taken to be constant, then u = 0, and the
continuity and Navier-Stokes equations have the common form Most common
form of the
Navier-Stokes
equations.
u = 0 (105a)

Du
Dt
=
_
u
t
+u u
_
= p +
2
u + g . (105b)
It is also important to remark at this point that incompressibil-
ity is an approximation so that we have simply written the pressure
in (105) as p since we are now neglecting the variations of pressure
with density. The pressure in (105) is no longer the thermodynamic
pressure, but is now a dependent variable to be determined by si-
multaneously solving the equations in (105), i.e. these represent
four equations for the four unknowns, (p, u).
(b) If we assume that the two coecients of viscosity are constant, then
we combine (104b) and (104c) to obtain

Du
Dt
= p
e
+
_
+

3
_
( u) +
2
u + g , (106)
which is one rather general form of the Navier-Stokes equations.
The combination of an equation of state for p
e
() with (106) and
the continuity equation (104a) corresponds to a (nonlinear) system
of 5 equations for the 5 unknowns, (p
e
, u).
(c) The bulk, or second, coecient of viscosity: The physical processes
giving rise to dissipation are the same for the two coecients of
viscosity so we should expect and to be of the same order of
magnitude; nevertheless, most measurements (and these are di-
cult and rare) show / = O(1 100), so there is some variability
depending on the uid system, and > . Because multiplies
u we should expect processes with signicant compressibility to
be those where the inuence of appears. Propagation of sound
waves relies on compressibility of the uid and so it is in the damp-
ing of sound waves that most attention to is generally given.
41
4.13 Energy conservation in continua
Here we show that ideas you have previously studied in courses covering equi-
librium thermodynamics can be written as dierential equations appropriate
for uid ow. The energy equation is typically needed when ows involve tem-
perature variations. This section makes extensive use of results from classical
thermodynamics; if you have not seen some of these ideas before then you will
either have to accept on faith some of the physical and mathematical ideas
discussed or you can study them in standard thermodynamics textbooks.
We begin with the conservation laws (body force/mass g):
continuity:

t
+ (u) = 0
D
Dt
= u(107a)
momentum:
Du
Dt
= T+ g with T = T
T
. (107b)
The stress tensor depends on the material in question. In general, we take
T = p
e
I + , (108)
where for an isotropic Newtonian uid we have (this denes the deviatoric
stress tensor )
T = (p
e
u)I + 2
_
E
1
3
( u) I
_
. (109)
Here p
e
is the equilibrium thermodynamic pressure and two material constants,
the shear viscosity and bulk viscosity , characterize the response of the
material to rate of deformation. We will work in terms of p
e
, though it is also
common to present the discussion below in terms of the uid pressure p (or p
above), which diers from p
e
according to p = p
e
u.
4.13.1 Mechanical Energy
If we take the inner product of u and equation (107b) then we have

2
Du
2
Dt
= u ( T) + g u (110a)
= (T u) T : E + g u , mechanical energy equation (110b)
where the symmetry of T allows the simplication T : u = T : E. Note that Verify that
T : u = T : E
42
there is a useful physical interpretation: for a uid element, the time-rate-of-
change of kinetic energy is balanced by changes in potential energy (the last
term on the right-hand side), as well as work done by the stresses acting on the
surface of the uid element and changes due to the conversion of mechanical to
thermal energy these last two eects are embodied in the u ( T) term and
will be seen more clearly below. Because of this exchange of dierent forms of
energy, equation (110) is referred to as the mechanical energy equation.
4.13.2 Thermal Energy equation
Let e denote the internal energy per unit mass. Also, since our experience with
equilibrium thermodynamics reminds us that temperature gradients produce
a heat ux, we denote the heat ux vector as q (energy/area/time). The total
energy of a given parcel of uid involves the kinetic and internal energy of the
material. The rst law of thermodynamics for conservation of total energy is
time rate of change of total
energy of a material volume =
rate at which energy is added by transport
across bounding surfaces or internal generation(111)
+
rate at which work is done
by forces acting on surfaces .
If we apply the rst law for total energy to a uid element, we are led to

2
Du
2
Dt
. .
kinetic energy
+
De
Dt
. .
internal energy
= (T u) + g u q + Q (112)
where Q denotes the heat generated internally per unit mass, say because of
local chemical reactions, the adsorption of light, etc.
Subtracting (110b) from (112) gives (here we again use T = p
e
I + )

De
Dt
= T : E q + Q (113a)
= p
e
u
. .
reversible
work
+ : E q + Q (113b)
Equation (113) is a form of the thermal energy balance. It remains to relate
the internal energy e to the temperature and for this we turn to standard
thermodynamical considerations.
43
4.13.3 Consequences of equilibrium thermodynamics
Rather than quoting results that can be found in standard treatments of equi-
librium thermodynamics, we derive the necessary results here. If desired, skip
the details and go directly to equation (122), where a local variation in internal
energy is related to temperature T, pressure and density variations, and is used
in the energy equation (113). Of course, some familiarity with thermodynamic
concepts are needed to follow the manipulations presented next. It is impor-
tant to recognize that we are studying ow problems that naturally represent
systems out of equilibrium. Nevertheless, locally we consider the system to
be in equilibrium, and so use relationships from equilibrium thermodynamics
simply because the length scales (and time scales) for the system to equilibrate
on the molecular scale are small (short) relative to the macroscopic variations.
For a single phase material there is an equation of state f(p
e
, , T) = 0.
The entropy/mass s is an intensive variable, which, expressed as a function of
T and p
e
, i.e. s(T, p
e
), gives the dierential relation
ds =
_
s
T
_
pe
dT +
_
s
p
e
_
T
dp
e
. (114)
A basic identity is
_
s
T
_
pe
=
cp
T
, where c
p
is the constant-pressure heat capacity
per unit mass. Also, a Maxwell relation gives
_
s
pe
_
T
=
1

2
_

T
_
pe
. So, we nd
Tds = c
p
dT +
T

2
_

T
_
pe
dp
e
. (115)
The rst law for a single phase uid is de = Tds p
e
dv, where v is the specic
volume. Since v =
1
, then de = Tds +
pe

2
d. Substituting this result into
(115) gives
de = c
p
dT +
T

2
_

T
_
pe
dp
e
+
p
e

2
d (116)
Hence, if we follow a uid particle (d D/Dt), then from (116), we have

De
Dt
= c
p
DT
Dt
+
T

T
_
pe
Dp
e
Dt
+
p
e

D
Dt
. (117)
Using the continuity equation gives

De
dt
= c
p
DT
Dt
+
T

T
_
pe
Dp
e
Dt
p
e
u . (118)
44
Substituting into the thermal energy equation (113) leads to
c
p
DT
Dt
= T : E q + Q + p
e
u
T

T
_
pe
Dp
e
Dt
(119)
and if we introduce the constitutive denition T = p
e
I + then we have
c
p
DT
Dt
= : E q + Q
T

T
_
pe
Dp
e
Dt
. (120)
We can next evaluate : E; for simplicity we only consider a Newtonian
uid (see equations (108) and (109)). It is convenient to write E =
1
3
( u) I+
_
E
1
3
( u) I
_
since then we obtain the simple result (note that we use the
short-hand notation A
2
= A
ij
A
ji
)
: E = ( u)
2
+ 2
_
E
1
3
( u) I
_
2
, (121)
which is a sum of squares, so : E 0. Physically, these terms represent the
(irreversible) conversion of mechanical energy to thermal energy and always
lead to temperature increases in a uid, as implied by the form of (120).
If we further introduce Fouriers law for the heat ux, q = kT, where
k is the thermal conductivity (here assumed constant), then (120) and (121)
yield the general form of the thermal energy equation (for Newtonian uids):
c
p
DT
Dt
= k
2
T+( u)
2
+2
_
E
1
3
( u) I
_
2
+Q
T

T
_
pe
Dp
e
Dt
. .
this term is usually neglected
since it is smaller than the others
.
(122)
This equation is most often used in the simplied form where convection and
conduction terms dominate:
DT
Dt
=
T
t
+u T =
k
c
p

2
T . (123)
There are also some cases where viscous dissipation matters, and for incom-
pressible ows it is then necessary to consider:
DT
Dt
=
k
c
p

2
T +
2
c
p
E
2
. (124)
45
For example, in shear ows E = 0, so that viscous eects produce heating in
the uid, which alters the temperature distribution according to (124).
Final remark: Using the second law one can demonstrate that the material
functions , and k, that appear in the above equations are necessarily 0
(and equality is only possible for the ideal case of a uid without irreverible
processes, i.e. no friction or viscosity and no thermal conduction).
4.14 Consequences of the second law: an equation for
entropy variations
The second law of thermodynamics is concerned with entropy variations, and in
particular how the entropy of a closed system (one that exchanges neither heat
nor mass with its surroundings) can only increase. In continuum mechanics we
apply this idea to a uid element. We shall see that the inequality expressed
by the second law places constraints on the material coecients that appear
in constitutive equations.
The usual thermodynamic statement of the second law is that ds
dq
T
where dq is the heat added per unit mass at temperature T. If we now consider
the heat ux vector q then the equivalent statement for a material volume
V
m
(t) is (n q > 0 when heat is added since n is the outward normal)
d
dt

Vm(t)
s dV =

Sm(t)
n q
T
dS +

Vm(t)
T
1
Q dV . (125)
Using the Divergence Theorem and the usual steps associated with the fact
that V
m
(t) is arbitrary yields

Ds
Dt

_
T
1
q
_
+ T
1
Q (126)
We now introduce some of the thermodynamic statements we introduced ear-
lier. Since de = Tds +
pe

2
d we can eliminate s in place of e and so rst arrive
at

De
Dt

p
e

D
Dt
T
_
T
1
q
_
+ Q , (127)
or substituting from equation (119) we nd
T : E q + T
_
T
1
q
_
+ p
e
u 0 . (128)
This result can be simplied further to
: E T
1
q T 0 . (129)
46
Introducing Fouriers law for the heat ux, q = kT and the Newtonian
constitutive equation for gives
( u)
2
+ 2
_
E
1
3
uI
_
2
+
k
T
(T)
2
0 . (130)
For real processes, we expect the inequality in this equation to be strictly > 0.
Since expansion, shear and temperature gradients are in principle independent
processes, then we see that the three materials functions , and k (not
necessarily constants) are positive.
Exercise: Beginning with equation (129) derive (130).
Exercise: Evaluate I :
_
E
1
3
( u) I
_
.
47
mass conservation; the continuity equation

t
+ (u) = 0
balance of linear momentum f = body force/volume
Du
Dt
= T+ f
balance of angular momentum (no unusual torques) T = T
T
mechanical energy balance

2
Du
2
Dt
= u ( T) + f u
rst law of thermodynamics

2
Du
2
Dt
+
De
Dt
= (T u) + f u q + Q
thermal energy equation
De
Dt
= T : E q + Q
constitutive equation T = p
e
I + ;
Newtonian uid = ( u) I + 2
_
E
1
3
( u) I
_
Table 2: An incomplete summary of the basic equations
4.15 Conditions for incompressibility
We begin with the continuity equation

t
+ (u) = 0 and examine under
what conditions the variations in density can be neglected. In particular, with
= constant, then u = 0. Let us assume that the ow is isothermal.
Hence, we inquire into the variations of density that occur during uid motion
as a result of variations of pressure that accompany ow. The modication to
the argument below to treat temperature variations is straightforward.
12
Write the continuity equation as
u
1
x
1
+
u
2
x
2
+
u
3
x
3
+
u

+
1

t
= 0 . (131)
We expect the pressure to change during the ow and from (131) we may
neglect the density variations relative to a typical term
u
1
x
1
provided

1.
The density variations that accompany a pressure change are

p
p. A convenient measure of material compressibility is provided by the
isentropic speed of sound, c
2
=
_
p

_
s
so


p
c
2
1 . (132)
In general, uid motions may be simply characterized in the limits (i)
inertially dominated ows, where p = O(U
2
) and (ii) viscously dominated
ows, where p = O(U/). Thus, the density variations should be small
provided The ratio U/c is
known as the
Mach number,
M= U/c.
inertially dominated ows :


U
2
c
2
1 (133a)
12
Best to add a footnote
48
viscously dominated ows :



U
U
2
c
2
1 (133b)
The speed of sound in air (STP) is approximately 330 m/sec and hence from
(133) the variations of density are expected to be only O(10
4
) even for ow
speeds 1 m/sec. Consequently, even ows of gases (for example, breathing)
are commonly treated as incompressible, u = 0.
4.16 The structure of conservation laws (something of a review)
As we have seen in the previous pages, classical physics postulates the idea
of conservation laws, which we develop to obtain dierential equations: for
example, time rate of change of linear momentum of a particle is equal to
the sum of the forces acting on it. Aslso, as the rst law of thermodynamics
states that the total energy of the universe is conserved, then we conclude
that across any specied boundary, the time rate of change of the total energy
internal to the boundaries is equal to the net rate energy is transferred across
the boundaries. Both statements emphasize that the time rate of change is
balanced by quantities acting across and within boundaries.
To provide a mathematical structure, we denote by (x, t) a quantity
(e.g. linear momentum, internal energy, concentration of chemical species i)
per unit mass. can be a scalar, vector, or tensor function. It is simplest
to consider a volume xed in space to which we apply the conservation law.
Then, the generic conservation law states that
d
dt

V
(x, t)(x, t) dV =

S(t)
n u dS

S
n q dS , (134)
where q denotes the vectorial (or tensorial) molecular ux, or amount of the
conserved quantity per unit time per unit area. As is standard the unit normal
n is directed outward from the volume.
Applying the Divergence Theorem to the integral on the right, and in-
terchanging dierentiation and integration on left (permissible as the volume
is xed in space) we have

V
_
()
t
+ (u) + q
_
dV = 0 . (135)
Since the domain of integration has not been specied, it is in fact completely
arbitrary, and hence it must be true that the integrand is identically zero (you
49
should now be familiar with this argument). Therefore
()
t
+ (u) = q . (136)
We summarize a few results in the table.
1. Combining (136) with the continuity equation yields

t
+ u = q . (137)
2. Consider transport of a chemical species i in a solvent. Denote the mass
fraction of species i as
i
. Then we can choose =
i
and with Ficks
law in the form q = D
i

i
we have

i
t
+ u
i
= (D
i

i
) . (138)
In the most common case (a dilute solution and an incompressible ow)
= constant, can we let
i
=
i
, so that

i
t
+u
i
= (D
i

i
) . (139)
If we divide by the molecular weight of the species, and denote the molar
concentration as c
i
then
c
i
t
+u c
i
= (D
i
c
i
) . (140)
The same equation applies to the number density of molecules of species
i.
3. Linear momentum: The linear momentum per unit mass of uid is sim-
ply the (mass average) velocity u. The momentum ux across surfaces
is denoted the (negative) of the stress tensor T so we have (could incor-
porate body forces)

_
u
t
+u u
_
= T . (141)
4. Consider the case of energy.
50
5. Consider the case of angular momentum.
Exercise: First let = 1. Then, subtract this result from (136) to arrive at
the conservation equation in the form

t
+ u = q . (142)
51
4.17 Appendix: Kinematics of material surfaces
This section is for those who are interested in a little more about kinematics.
Here we present a short discussion of the kinematics of material sur-
face elements, motivated by an error discovered in a paper by H. Oguz &
A. Prosperetti, A generalization of the impulse and virial theorems with an
application to bubble oscillation, Journal of Fluid Mechanics, 218, 143-162
(1990). The error in the article was basically using one of the tensor theorems
with the dyadic product written in the incorrect order. The presentation be-
low follows closely that given by Batchelor, p. 131-132. Some of this material
necessarily repeats aspects of the kinematic discussion given in earlier sections.
In introductory courses on uid dynamics it is common to discuss some
basic issues pertaining to the kinematics of line elements of the uid, includ-
ing a discussion of the rate-of-strain and vorticity tensors responsible for the
stretching and rotation, respectively, of individual elements. Here we extend
this discussion to surface elements.
Some preliminaries: A material element of the uid refers to a set of mass
points which move according some velocity eld. Typically, we have at our
disposal an Eulerian description of the velocity eld u(x, t) which gives the
velocity at a certain location x at time t. Dierent uid elements may occupy
the same position x at dierent times. In a Lagrangian description we imagine
labeling each of the individual uid elements and following their individual
motion. If we label such line, surface, or volume elements, we speak of material
lines, surfaces or volumes, respectively. It is worth keeping in mind that a uid
element (the same set of mass points) can distort as it is advected by the ow,
though the total mass remains constant.
The concept of a material derivative lies at the basis of all discussions
and applications of kinematics. The material derivative provides that rate-of-
change of a quantity as an observer moves with the local uid motion, u(x, t).
For example, consider some scalar quantity (x, t) associated with each point
in the uid. If a set of points is labeled and followed, then the rate of change
of , denoted d/dt,
13
as measured by an observer moving with the local uid
velocity u(x, t), is
D
Dt
=
d
dt
=

t
+u (143)
(For a derivation, see any standard uid dynamics text and the notes at this
13
It is common to indicate the material derivative as D/DT. I tend to use the two
notations interchangeably as d/dt means the total time derivative.
52
beginning of this chapter). The rst term accounts for variations in that
occur at a xed position x and the second term accounts for variations in
since the position of the element changes and may vary with spatial position.
The material derivative d/dt is thus dened as
d
dt


t
+u . (144)
Kinematics of line and volume elements:
We will now discuss the kinematics of line elements, volumes, and sur-
faces respectively. In each case we will use the notation to represent a small
quantity and an exact expression for behavior in the neighborhood of a point
in the uid x is found by taking the limit 0.
First recall that a small line element may change both its orientation
and length owing to a velocity gradient. Mathematically,
d
dt
= u (145)
where u(x, t) is the velocity gradient measured at the current location of the
line element.

l
u(x)
u(x+ l)

= u(x) + l u(x)

x
Next, consider a small material volume V , consisting of a labeled set of
mass points. As this set of points is advected by the ow, the occupied volume
may change, though the total mass remains constant (we would naturally say
that the density changes). The volume change is calculated by taking a small
element dS of the bounding surface and moving it the direction of the normal
component of velocity u n. So we see that the rate-of-change of volume is
calculated according to
d V
dt
=

S(t)
u n dS (146)
where S(t)denotes the (material) surface bounding V .
53
u
n

S
dS

V
By the Divergence Theorem we see
d V
dt
=

V (t)
u dV. (147)
For small material volumes we may approximate the right-hand side by a
Taylor series about the center x, being content with the rst term, so that
d V
dt
= u(x) V + o(V ). (148)
Hence, the rate-of-change of volume of a material element is proportional to
the volume itself and this expression is valid in the limit that the volume
element becomes vanishingly small. From now on the small error terms o()
will be neglected since we will always take the limit that the appropriate
material element becomes vanishingly small.
... and nally now considering material surfaces: Now, in order to describe the
distortion of a material surface element we introduce the notation S which
is a vectorial surface element, with magnitude S and direction given by the
unit outward normal n from the uid volume V :
S = n S. (149)
Since a material volume may, in general, have any shape we imagine a small
cylindrically-shaped volume with cross-sectional area S and generator .
The material volume is
V = S. (150)
Taking the material derivative,
d V
dt
=
d
dt
S +
d S
dt
. (151)
Using (145), (148) and (150) we see that equation (151) may be written

_
( u) S = (u) S +
d S
dt
_
(152)
54
or, since this expression should be valid for arbitrary choice of the line element
, we have
d S
dt
= ( u) S (u) S. (153)
Since S = n S, then expanding (153) and taking the inner product
with n gives
14
d S
dt
= [ u n (u) n] S =
s
u S, (154)
where we have introduced the surface gradient operator
s
(I nn) ,
which acts to take derivatives in the tangent plane to the surface with normal
n. Equation (154) agrees with the incompressible ( u = 0) result given
by Oguz & Prosperetti. Finally, again using S = n S, equations (153) and
(154) may be combined to obtain
dn
dt
= (u) n +nn (u) n = (
s
u) n, (155)
which takes the place of (the incorrect) equations (2.6) and (A3) in Oguz &
Prosperettis paper.
14
We note that n
dn
dt
=
1
2
d (n n)
dt
= 0 since n is a unit vector, i.e. n n = 1.
55
4.18 Appendix: Governing equations in component form
4.18.1 CONTINUITY EQUATION
The form of the continuity equation

t
+ (u) = 0, (156)
expressed in cartesian, cylindrical and spherical coordinates is

t
+

x
(u
x
) +

y
(u
y
) +

z
(u
z
) = 0(157a)

t
+
1
r

r
(ru
r
) +
1
r

(u

) +

z
(u
z
) = 0(157b)

t
+
1
r
2

r
_
r
2
u
r
_
+
1
r sin

(u

sin ) +
1
r sin

(u

) = 0 . (157c)
The usual case of an incompressible ow for which u = 0 can be found
by taking ()/t 0 and eliminating .
4.18.2 NAVIER-STOKES EQUATIONS
The Navier-Stokes equations of motion expressed in cartesian coordinates for
a Newtonian uid of constant density and viscosity, for which the only body
force is due to gravity, are:

_
u
x
t
+ u
x
u
x
x
+ u
y
u
x
y
+ u
z
u
x
z
_
=
p
x
+
2
u
x
+ g
x
(158a)

_
u
y
t
+ u
x
u
y
x
+ u
y
u
y
y
+ u
z
u
y
z
_
=
p
y
+
2
u
y
+ g
y
(158b)

_
u
z
t
+ u
x
u
z
x
+ u
y
u
z
y
+ u
z
u
z
z
_
=
p
z
+
2
u
z
+ g
z
(158c)
where

2
=

2
x
2
+

2
y
2
+

2
z
2
. (159)
The components of the stress tensor T = pI + , where = 2E, are
( u = 0)

xx
= 2
u
x
x
(160a)
56

yy
= 2
u
y
y
(160b)

zz
= 2
u
z
z
(160c)

xy
=
yx
=
_
u
x
y
+
u
y
x
_
(160d)

xz
=
zx
=
_
u
x
z
+
u
z
x
_
(160e)

yz
=
yz
=
_
u
y
z
+
u
z
y
_
(160f)
Remember that the force per unit area acting on any surface can be calculated
by utilizing the denition that the stress vector is n T where n is the unit
normal directed outward from a surface. In general, surface forces arise from
the uid pressure and the viscous (frictional) nature of the uid.
The equations of motion expressed in cylindrical coordinates for a Newtonian
uid of constant density and viscosity: (the body force term has not been
written)
r component:

_
u
r
t
+ u
r
u
r
r
+
u

r
u
r


u
2

r
+ u
z
u
r
z
_
=
p
r
+
_

r
_
1
r

r
(ru
r
)
_
+
1
r
2

2
u
r

2

2
r
2
u

+

2
u
r
z
2
_
(161)
component:

_
u

t
+ u
r
u

r
+
u

r
u

+
u
r
u

r
+ u
z
u

z
_
=
1
r
p

+
_

r
_
1
r

r
(ru

)
_
(162)
+
1
r
2

2
u

2
+
2
r
2
u
r

+

2
u

z
2
_
z component:

_
u
z
t
+ u
r
u
z
r
+
u

r
u
z

+ u
z
u
z
z
_
=
p
z
+
_
1
r

r
_
r
u
z
r
_
+
1
r
2

2
u
z

2
+

2
u
z
z
2
_
(163)
The components of the stress tensor T = pI + , where = 2E, are
( u = 0)

rr
= 2
u
r
r
(164a)
57

= 2
_
1
r
u

+
u
r
r
_
(164b)

zz
= 2
u
z
z
(164c)

r
=
r
=
_
r

r
_
u

r
_
+
1
r
u
r

_
(164d)

z
=
z
=
_
u

z
+
1
r
u
z

_
(164e)

rz
=
zr
=
_
u
z
r
+
u
r
z
_
(164f)
58
The equations of motion expressed in spherical coordinates for a Newtonian
uid of constant density and viscosity: (the body force term has not been
written)
rcomponent:

_
u
r
t
+ u
r
u
r
r
+
u

r
u
r

+
u

r sin
u
r


_
u
2

+ u
2

_
r
_
_
(165)
=
p
r
+
_

2
u
r

2u
r
r
2

2
r
2
u


2
r
2
u

cot
2
r
2
sin
u

_
component:

_
u

t
+ u
r
u

r
+
u

r
u

+
u

r sin
u

+
u
r
u

r

u
2

cot
r
_
(166)
=
1
r
p

+
_

2
u

+
2
r
2
u
r

r
2
sin
2


2 cos
r
2
sin
2

_
component:

_
u

t
+ u
r
u

r
+
u

r
u

+
u

r sin
u

+
u
r
u

r
+
u

cot
r
_
(167)
=
1
r sin
p

+
_

2
u

r
2
sin
2

+
2
r
2
sin
u
r

+
2 cos
r
2
sin
2

_
where

2
=
1
r
2

r
_
r
2

r
_
+
1
r
2
sin

_
sin

_
+
1
r
2
sin
2

2
(168)
The components of the stress tensor T = pI + , where = 2E, are
( u = 0)

rr
= 2
u
r
r
(169a)

= 2
_
1
r
u

+
u
r
r
_
(169b)

= 2
_
1
r sin
u

+
u
r
r
+
u

cot
r
_
(169c)
59

r
=
r
=
_
r

r
_
u

r
_
+
1
r
u
r

_
(169d)

=
_
sin
r

_
u

sin
_
+
1
r sin
u

_
(169e)

r
=
r
=
_
1
r sin
u
r

+ r

r
_
u

r
_
_
(169f)
60
4.19 Appendix: The Navier-Stokes equations a little
bit of history
Note: See also the attached article by C. Truesdell, Notes on the history of
the general equations of hydrodynamics. Truesdell wrote extensively
on the history of several aspects of classical physics.
The equations of motion for uid ow evolved from early considerations
of statics and the equations for particle motion. The term hydrodynamics was
introduced by Daniel Bernoulli (1700-1783) to unite the elds of hydrostatics
and hydraulics. Euler (1707-1783) is responsible for the equations of motion
for an ideal (i.e. inviscid) uid. Although Newton introduced the hypothesis
that the shear stress is proportional to the velocity gradient, the general de-
velopment of the equations of motion for a viscous uid awaited developments
in the 19th century.
The Navier-Stokes equations are named after two distinguished mathe-
maticians, a Frenchman Claude Louis M. H. Navier (1785-1836) and an En-
glishman George Gabriel Stokes (1819-1903), who independently arrived at the
same equations. Navier arrived at his results rst and reported them to the
French Academie des Sciences on 18 March 1822 in a paper entitled Memoire
sur les lois du mouvement des uides. Apparently Navier (and later Pois-
son in an article published in 1829, Memoire sur les equations generales de
lequilibre et du mouvement des corps solides elastiques et des uides, Journ.
de l

Ecole Polytechn.) based the derivation on ideas about the interactions


among molecules in the uid. The early developments in solid and uid me-
chanics largely occurred in parallel.
Stokes, who was to hold the Lucasian Professorship of Mathematics at
Cambridge University from 1849-1903 (this is the Chair held by Newton), ar-
rived at essentially the same results using continuum mechanics considerations
in a paper that appeared in 1845, On the theory of internal friction of u-
ids in motion, Transactions of the Cambridge Philosophical Society 8. The
Frenchman B. de Saint-Venant also provided a similar demonstration in 1843
(Comptes Rendus). Navier, though his work was completed and presented
earlier, apparently did not discuss the concept of shear, which is of course
important to understanding viscous ows. In his 1845 paper Stokes intro-
duced a guiding principle for thinking about constitutive response of viscous
uids: That the dierence between the pressure on a plane passing through
any point P of a uid in motion and the pressure which would exist in all
directions about P if the uid in its neighborhood were in a state of relative
61
equilibrium depends only on the relative motion of the uid immediately about
P; and that the relative motion due to any motion of rotation may be elim-
inated without aecting the dierences of the pressures above mentioned.
15
The unit of viscosity in the cgs system of units is called the stoke in honor of
Stokes.
It seems clear that at a very early time a large part of the mathematical
framework of classical uid dynamics was established. However, the boundary
conditions, in particular the role of the no-slip condition and when it could be
applied, were not understood at least until the work of Prandtl in 1904. The
state of uid dynamics in the early years of this century is described by S.
Goldstein (then at Harvard) in a paper entitled Fluid mechanics in the rst
half of this century (Ann. Rev. Fluid Mech. (1969)). It is also interesting to
note that the Reynolds number as a guiding principle for characterizing uid
motions, though implicit in the work of Reynolds (1883) was not formerly
introduced until Sommerfeld presented this terminology in 1908.
16
Ludwig
Prandtl and his colleagues in Gottingen Germany were than to make signif-
icant use of this idea (starting about 1910) and in the general principle of
hydrodynamic similarity. Even so, you do not really nd the concept of the
Reynolds number in the outstanding text Hydrodynamics by Horace Lamb,
which dominated studies of theoretical uid dynamics at least until the 1950s.
As several students have remarked in lecture overs the years, even before
we arrived at the nal form of the Navier-Stokes equations, simplied for
incompressible ows with constant viscosity, we made several assumptions.
Perhaps it is well then to keep in mind the following quote from Professor G.
Birkho
17
These approximation are made partly for simplicity even with them,
one cannot in most cases solve the mathematical problems involved.
They are also made because we do not actually know how viscosity acts
15
Reference: Lord Rayleighs biographical sketch of Stokes, published in the Royal Society
Year-Book, 1904.
16
N. Rott, Note on the history of the Reynolds number, Ann. Rev. Fluid Mech. 22,
1-11 (1990).
17
Hydrodynamics: A study in logic, fact, and similitude, 1950, p. 7.
62
in a uid under rapid compression!
Hence if we are too impressed by the dangers of approximations, we
must despair at the outset of attaining any rational understanding of
uid mechanics. However, science was not built by timid or faltering
hands. Let us move boldly forward, seeing what kind of edice one can
construct on these somewhat insecure foundations.
References
1. L.M. Milne-Thomson, Theoretical Hydrodynamics.
63

You might also like