Fluid Dynamics - Math 6750

Basic principles - Week 2

Consider a region Ω0 ∈ D, a region of fluid at t = 0. We will consider the set (image under
the flow map, set of location at time t of fluid particles that were originally in Ω0 at t = 0)
and introduce the following notation:

Ωt = {x ∈ D : x = Φ(α, t) for some α ∈ Ω0 } ≡ Ω(t) = Φ(Ω0 , t).

Ωt is called a material volume, a volume moving with the fluid.

1 Reynolds’ Transport Theorem

Theorem 1. If F (x, t) is a function that is defined in the fluid, then
Z Z  
d ∂F
F (x, t)dx = + ∇ · (uF ) dx. (1)
dt Ωt Ωt ∂t

Proof. By a change of variables to the fixed initial region Ω0 , we have

Z Z
d d
F (x, t)dV (x) = F (Φ(α, t), t)|J(α, t)|dV (α).
dt Ωt dt Ω0

For simplicity, we assume that J is positive and get rid of the absolute value. Since the
volume of integration is constant, we can interchange derivative and integration. We find by
the product rule
Z Z  
D DF DJ
(F (Φ(α, t), t)J(α, t)) dV (α) = J +F dV (α).
Ω0 Dt Ω0 Dt Dt

Using Lemma 2 (Euler) to rewrite DJ

Dt , we have
Z   Z  
DF DF
J + F J∇ · u dV (α) = + F ∇ · u JdV (α)
Ω0 Dt Ω0 Dt

Changing variables again to the moving domain, we obtain

Z  
DF
+ F ∇ · u dV (x)
Ωt Dt

The notation dV (x) is equivalent to dx or dxdydz. Similarly for dV (α).

1
2 Conservation of Mass
In the continuum approach, we assume that there is a well-defined function ρ(x, t) called the
fluid density such that the mass of any set Ωt of fluid is
Z
m(Ωt ) = ρ(x, t)dV (x).
Ωt

We want to derive a PDE for the density assuming that mass is neither created
R nor destroyed.
Consider an arbitrary parcel of fluid Ω0 ≡ Ω(0) at time t = 0. Its mass is Ω0 ρ(x, 0)dV (x).
By conservation of mass, we have m(Ωt ) = m(Ω0 ) for all t. In other words,
Z
d d
m(Ωt ) = 0 i.e ρ(x, t)dV (x) = 0.
dt dt Ωt

The above is the statement of conservation of mass. Time appears in two places: in the
integrand and in the region of integration. We will now show that the above equation gives a
PDE, known as the continuity equation or conservation of mass.

Theorem 2 (Conservation of mass in Eulerian).

∂ρ
+ ∇ · (ρu) = 0. (2)
∂t
Proof. Apply the Reynold’s transport theorem with F (x, t) = ρ(x, t). Since Ωt is arbitrary,
the integrand must be zero. We conclude that
Dρ
+ ρ∇ · u = 0
Dt
as expected.

A special case of the Reynolds’ Transport Theorem and conservation of mass follows by
setting F = ρF , where ρ is the fluid’s density. In this case, we have
Z Z
d DF (x, t)
ρ(x, t)F (x, t)dV (x) = ρ(x, t) dV (x).
dt Ωt Ωt Dt

Indeed, a simple application of product rule gives

Z Z  
d ∂(ρF )
ρF dV (x) = + ∇ · (ρF u) dV (x)
dt Ωt Ωt ∂t
Z  
∂ρ ∂F
= F +ρ + ρu · ∇F + F ∇ · (ρu) dV (x)
Ωt ∂t ∂t
Z    
∂ρ ∂F
= F + ∇ · (ρu) + ρ + u · ∇F dV (x)
Ωt ∂t ∂t
Z
DF
= ρ dV (x).
Ωt Dt

2
Remark 1. Conservation of mass can be written in various forms
∂ρ
+ u · ∇ρ + ρ∇ · u = 0
∂t
or
∂ρ
+ ∇ · (ρu) = 0.
∂t
Alternatively, the conservation of mass can be written in Lagrangian coordinates.

Theorem 3 (Conservation of mass in Lagrangian).

D
(Jρ) = 0. (3)
Dt
Definition 1. A flow is incompressible if ∇ · u = 0.

Using conservation of mass (2), we see that for an incompressible fluid

∂ρ
+ u · ∇ρ = 0
∂t

or equivalently Dρ Dt = 0, i.e. the density of a fluid particle does not change as the particle
moves. For an incompressible flow, an initially homogeneous fluid remains homogeneous. More
precisely, if in addition to incompressibility, the initial density is constant, i.e ρ(x, 0) = ρ0 , then
ρ(x, t) = ρ0 for all x, t.
Another simple consequence of incompressibility is that the volume of a chunk of fluid does
not change. If the flow is incompressible, it follows from the Euler’s identity that DJ Dt = 0 and so
J(α, t) = C. Since J(α, 0) = 1, we must have J(α, t) ≡ 1 for all t ≥ 0, i.e J is the determinant
of the identity matrix. Then, for a fluid Ω(0) occupying Ω(t) at time t, the volume of Ω(t) is
Z Z Z
|Ω(t)| = dV (x) = |J(α, t)| dV (α) = dV (α) = |Ω(0)|.
Ωt Ω0 Ω0

3 Stresses
There are two types of forces:

1. Body forces such as gravitational or electromagnetic

R force. Denoting the external force
per unit mass by Fb (x, t), the total body force is Ωt ρ(x, t)Fb (x, t)dV (x);

2. Surface forces such as pressure forces or viscous stresses. A surface force acts from the
fluid outside Ωt on the fluid inside and vice versa. Denoting by t(x, t) the force per unit
area
R exerted at x on fluid inside Ωt by the fluid outside Ωt , the total surface force is
∂Ωt t(x, t)dS(x). Here ∂Ωt denotes the boundary of Ωt .

Remark 2. t is sometimes called the traction (vector). Forces per area are called stresses.

3
3.1 Hydrostatic equation
The simplest case of a fluid in static equilibrium occurs when the fluid is at rest. For example,
if a liquid is poured into a bucket and left to stand until all relative motions have died out,
the fluid is then in static equilibrium. At this point, there are no resultant forces or moments
acting on the fluid. It is also possible to have a moving fluid in static equilibrium, as long as
no part of the fluid is moving with respect to any other part. This is called rigid body motion.
When the fluid and its container are moving at constant speed, for instance, it is in equilibrium
under the forces due to pressure and its own weight (think of the coffee in a cup that is in a
car moving at constant velocity). However, when this system is accelerating, the inertia force
due to acceleration needs to be taken into account.
The only forces acting on Ωt are those due to gravity and pressure differences. Because
there is no resultant acceleration of the fluid particle, these forces must balance. The force
due to gravity acts only in the vertical direction. Thus, in the e1 and e2 -directions, the force
due to pressure acting on the left/front face must cancel the force due to pressure acting on
the right/back face. The pressures on these two pair faces must be equal, and so the pressure
cannot vary in the e1 and e2 -directions.

x3

Ptop

g
x2

Pbottom
x1

Figure 1: Static equilibrium of a volume of fluid under gravity.

In the e3 direction, we use a Taylor series to express the pressure at the top and bottom
surface of the cube. Let P0 be the pressure at the center of the cube and ∆z the height of the
volume of fluid. Then
∆z 1 00 ∆z 2
Ptop = P0 + P 0 (0) + P (0) + ...
2 2 4
∆z 1 00 ∆z 2
Pbottom = P0 − P 0 (0) + P (0) + ...
2 2 4

4
Remembering that pressure is force per area, the resulting force due to pressure on the cube
with surface area ∆x∆y is

Fpressure = (Pbottom − Ptop )∆x∆y + . . . = −P 0 (0)∆x∆y∆z + . . . .

This force acts in e3 opposite ge3 . By Newton’s law, it must be balanced by the weight of Ωt .
Assuming that the density is a function of z only, i.e ρ = ρ(z), we have using a similar Taylor
series argument that the average density is
1 1
ρ = (ρtop + ρbottom ) = (2ρ0 + . . .) = ρ0 + . . .
2 2
the density at the center. Therefore, we find Fweight = ρ0 ∆x∆y∆z + . . .. Equating both forces
and taking the volume of the cube to zero yields
dp
= −ρg. (4)
dz
Remark 3. For a fluid with constant density, the hydrostatic equation (4) can be integrated to
show that the pressure increases linearly with depth.
The pressure at a point in a fluid is independent of the orientation of the surface passing
through the point; the pressure is isotropic. However, pressure is a normal stress since it
produces a force that acts in a direction normal to the surface on which it acts. By convention,
a force acting to compress the volume is positive, but for a closed surface the unit normal vector
always points outward (by definition). So the force due to a pressure p acting on one side of a
small element of surface dS defined by a unit normal vector n is given by −pndS = −pdS.

3.2 Bulk stress

Consider a fluid held in a container. In the interior of the fluid, away from the walls of the
container, each fluid particle feels the pressure due to its contact with the surrounding fluid.
The fluid particle experiences a bulk strain and a bulk stress since the surrounding fluid exerts
a pressure on all the surfaces that define the fluid particle. When body forces are negligible,
the pressure is uniform throughout the fluid. In this case, the forces due to pressure acting over
each surface of a fluid particle all have the same magnitude. The force acting on any one face
of the particle acts normal to that face with a magnitude equal to the pressure times the area.
There is no resultant force due to pressure acting on a body if the pressure is uniform in space,
regardless of the shape of the body. Resultant forces due to pressure will appear only if there
is a pressure variation within the fluid, that is, when pressure gradients exist. When there
is no flow the stress distribution is completely described by its pressure distribution, and the
bulk modulus relates the pressure to the fractional change in volume (the compression strain).
When there is flow, however, shearing stresses may become important, and additional normal
stresses can come into play. The magnitude of these stresses depends on the fluid viscosity.
Common viscosities and densities are given in Table 1.

3.3 Cauchy stress

We discuss a few facts about the traction t and stresses in general.

5
 ρ (kg/m3 ) µ (P a · s)
Air 1.204 18.2 · 10−6
Water 998.2 1.002 · 10−3
Sea water 1,025 1.07 · 10−3
Motor oil 919 0.04
Honey ≈ 1400 ≈ 1.4
Mercury 13,600 1.57 · 10−3

Table 1: Density and viscosity of some common fluids at 20◦ .

Theorem 4 (Principle of stress equilibrium). Let l be a characteristic length for a sequence

of regions around x. Then Z
1
lim tdS(x) = 0. (5)
l→0 l2 ∂Ωt

This is called the principle of stress equilibrium (no net force around a point particle).

Proof. Consider a sequence of smaller and smaller regions around a point x and use these as
Ωt . Let l be a characteristic length (e.g. cubic root of the volume). As we proceed through
the sequence of regions, l → 0.
Newton’s 2nd law for Ωt says that the rate of change of linear momentum is balanced by
the sum of the forces acting on the volume. Since there are two kinds of forces, we have
Z Z Z
d
ρudV (x) = ρFb dV (x) + tdS(x).
dt Ωt Ωt ∂Ωt

The left hand side scales like l3 , as does the first term on the right hand side. The surface
integral on the right hand side, however, scales like l2 . This is true assuming all integrands are
bounded. Dividing by l2 both sides, we have
Z Z Z
1 d 1 1
ρudV (x) = ρF b dV (x) + tdS(x).
l2 dt Ωt l2 Ωt l2 ∂Ωt

Since the term of the left hand side term and the first term on the right hand side go to zero
as l → 0, it follows that Z
1
lim tdS(x) = 0.
l→0 l2 ∂Ωt

Lemma 1. As a consequence of the principle of stress equilibrium, the followings must hold.

1. t cannot be a function of x, t only. It must depend on something else. We assume it

depends also on n, the unit normal to ∂Ωt at x, pointing out of Ωt . Thus, t = t(x, t, n).

2. t(x, t, −n) = −t(x, t, n), which is equivalent to Newton’s 3rd law (action-reaction, equal
and opposite forces).

3. t(x, t, n) = n · T(x, t), where T is a rank 2 tensor.

6
 t(n)
x3

−e1

x
−e2
O B x2
−e3

x1

Figure 2: Cauchy pyramid around a point x.

Proof. We only proof 3., as the first two facts are obvious consequences of Eq. (5). Consider
the small tetrahedron given in Fig. 2. Let dA be the area of the slanted face of the tetrahedron.
Then the areas of the other faces are nj dA (nj are the components of n). Using the principle
of local stress equilibrium (5), assuming that the stress is constant in the small tetrahedron,
we have
1
0 = lim [t(x, n)dA + t(x, −e1 )dA1 + t(x, −e2 )dA2 + t(x, −e3 )dA3 ]
dA→0 dA

Replacing dAj by nj dA, factoring and cancelling dA, and using fact 2., we end up with
t(x, n) = t(x, e1 )n1 + t(x, e2 )n2 + t(x, e3 )n3 .
Next, we define T(x) = [t(x, e1 )t(x, e2 )t(x, e3 )] (matrix with columns given by the tractions).
It follows then that t(x, n) = n · T(x) as expected.

Remark 4. There are nine components of stress at any given point. Tij (x) is the component
of the surface force per unit area in the ith direction on a surface whose normal is pointing
in the jth coordinate direction as illustrated in Fig. 3. Tii (no summation) are called normal
stresses and Tij for i 6= j are called shear stresses.

3.4 Viscous stress

Constitutive equations or laws relate information on the microscopic scales to the macroscopic
scales. They are additional relations among unknowns that model physical processes at the
molecular scale not captured in the continuum assumption. The simplest model for a moving
fluid, u 6= 0, is to assume that T = −pI. Such a fluid is called an ideal fluid.

7
 x3

T33
T23
T13
T32
T22
T31 T12 x2
T21
T11

x1

Figure 3: Nine components of stress as they act as a point in a fluid.

Example 1. As an example, consider pressure driven flow in a channel with p1 the high
pressure at the left end and p2 the low pressure at the right end separated by a distance L,
as illustrated in Fig. 4. Assuming that nothing varies in z, we look for unidirectional flow of
the form u = (u(x, y, t), 0, 0). We will see later that the solution with a no flow penetration
boundary condition at the walls, u · n = 0, is
 
1 p1 − p2
u(y, t) = t + u0 (y).
ρ0 L

We note that the velocity increases linearly with t. Therefore, this is not a realistic model.
The reason is that it ignores relative motion between the layers of fluid due to the molecular
diffusion of water molecules, which can be thought as friction between moving layers.

To talk about relative motion, we look at x fixed and compare u(x) to u(x + h) with h
small. First, we have the following lemma

Lemma 2.
1
u(x + h) = u(x) + D(x)h + ω(x) × h + O(khk2 ),
2
where D(x) is the symmetric part of ∇u and ω(x) = ∇ × u(x) is the vorticity.

Remark 5. D is also called the rate of deformation tensor and the components of the vorticity
are  
∂u3 ∂u2 ∂u1 ∂u3 ∂u2 ∂u1
ω(x) = − , − , − .
∂x2 ∂x3 ∂x3 ∂x1 ∂x1 ∂x2

8
 y
High Low
Pressure Pressure
p1 p2 x

Figure 4: Two dimensional flow in a pipe.

Proof. Since ∇u = D + Ω with

 
0 −ω3 ω2
1 1
Ω= ω3 0 −ω1  and Ωh = ω × h,
2 2
−ω2 ω1 0
we expand u(x + h) using a Taylor series to get
u(x + h) = u(x) + ∇u(x)h + O(khk2 ) = u(x) + D(x)h + Ω(x)h + O(khk2 ).

Next, we introduce y = x + h and we want to find its rate of change. Since x is fixed, we
obtain
dy dh 1
= = u(x + h) = u(x) + D(x)h + ω × h.
dt dt 2
To determine how h changes geometrically, we consider each term separately. First, it is easy
to see that dh
dt = u(x) corresponds to a rigid translation. As a rigid body motion, it does not
cause relative motion. Second, we let dh 1 dh
dt = 2 ω × h = Ωh and we note that the vector dt is in
the plane perpendicular to ω and h. Furthermore, since
d d dh 1
khk2 = (h · h) = 2 · h = h · (ω × h) = 0,
dt dt dt 2
the length of h is constant and this motion corresponds to a rigid rotation around the vector ω
with angular velocity 12 kωk. Therefore, only dhdt = D(x)h can lead to relative deformation. To
see this, we remark that D is symmetric, so it has a set of orthonormal eigenvectors
P3 ẽj called
principal directions such that Dẽj = dj ẽj (no summation). Thus, writing h(t) = j=1 h̃j (t)ẽj
and plugging in, we find a set of uncoupled ODEs
dh̃j (t)
= dj h̃j (t) j = 1, 2, 3.
dt
In each of the principal direction (see Fig 5), a cube of material is either stretched or com-
pressed.

9
 x3
x03
d3

x2

d2
x02
x01 d3
x1

Figure 5: Principal stress components.

Remark 6.
d
(h̃1 h̃2 h̃3 ) = (d1 + d2 + d3 )(h̃1 h̃2 h̃3 ) = tr(D)(h̃1 h̃2 h̃3 ) = (∇ · u)(h̃1 h̃2 h̃3 )
dt
The relative rate of change of volume elements of fluid is then
1 d
(h̃1 h̃2 h̃3 ) = ∇ · u.
h̃1 h̃2 h̃3 dt
If the flow is incompressible, then ∇ · u = 0 and some of the dj ’s must be positive and some
negative.

Therefore, to obtain a constitutive relation that includes relative motion, Stokes postulated
that
1. T is a continuous function of D and is otherwise independent of fluid motion.

2. T does not depend explicitly on x (homogeneous).

3. There are no preferred directions in space (isotropy).

4. When D = 0, then T = −pI.

5. T depends linearly on D.
Conditions 1., 2., 4., and 5. suggest that T has the form

Tij = −pδij + Cijpq Dpq ,

where the 81 numbers Cijpq are constants. Cijrs is a 4th order tensor by the rules of consistency
(discussed at the end of this section). Since T (later follows from conservation of angular
momentum) and D are 2nd order tensors we have Cijpq = Cjipq = Cijqp . Using the assumption
of isotropy 3., it follows that C is a 4th order isotropic tensor. Therefore, C has the form

Cijpq = λδij δpq + µ(δip δjq + δip δjq ) + κ(δip δjq − δiq δjp )

10
for 3 constants λ, µ, κ. Since Cijrs = Cjirs , we must have κ = 0. Plugging in and rearranging
gives the expected form of the stress tensor
 
∂ui ∂uj
Tij = (−p + λDrr )δij + 2µDij = (−p + λ∇ · u)δij + µ + .
∂xj ∂xi
If the flow is incompressible, then there is only one parameter µ called the dynamic viscosity or
the shear viscosity. In general, λ, µ can be related to the bulk viscosity. The quantity ν = µ/ρ
is the kinematic viscosity.
Definition 2. A Newtonian fluid is a fluid with stress tensor

T = (−p + λ∇ · u)I + 2µD

for two scalars λ, µ.

Because the trace of any second-order tensors is invarianth with respect to ia change of basis,
we define the mechanical pressure P as Pm = − 31 tr(T) = 13 T11 + T22 + T33 . It is related to
the thermodynamic equilibrium pressure p according to
1 
Pm = p − 3λ + 2µ ∇ · u = p − η∇ · u,
3
p−Pm
where η = λ + 23 µ = ∇·u is the bulk viscosity. Consequently, the Newtonian stress tensor can
also be written as  
1
T = −pI + 2µ D − (∇ · u)I + η(∇ · u)I,
3
or, in terms of Pm ,  
1
T = −Pm I + 2µ D − (∇ · u)I .
3
Writing the stress tensor as T = −Pm I + τ , these two stress tensors are known as the mean
normal stress tensor and the deviatoric stress tensor respectively.

We now sketch an argument that explains how to reduce the number of constants and
finally define what a tensor is. Physical laws should be consistent independently of which
coordinate systems is used (called contravariant). Consider two orthogonal cartesian systems
in R3 (x1 , x2 , x3 ) and (x01 , x02 , x03 ) with basis e1 , e2 , e3 , and e01 , e02 , e03 respectively. Further,
consider some physical laws, e.g. F = ma, t = n · T. In each coordinate system, we have

Fi = mai Fi0 = ma0i ti = Tij nj t0i = Tij0 n0j .

We can also get expressions in the primed system by starting with the unprimed description
and changing coordinates. Physical consistency requires that we get the same expressions as we
do starting in the primed system. Consider a point P in R3 . Let (x1 , x2 , x3 ) be the coordinates
of P in the unprimed system. Let (x01 , x02 , x03 ) be the coordinates of P in the primed system.
Then
P = x1 e1 + x2 e2 + x3 e3 = x01 e01 + x02 e02 + x03 e03

11
and
x1 = e1 · P = (e1 · e01 )x01 + (e1 · e02 )x02 + (e1 · e03 )x03 = l1j x0j
where lij = ei · e0j is the cosine between the angles of the unit vectors. We note that l is not
necessarily symmetric. In general, we have

xi = lij x0j and x0j = lji xi .

Further, it can be shown that lij lkj = lji ljk = δik so that LLT = LT L = I. L is a second-order
tensor. These are the rules so that vectors are consistently represented.

Definition 3. An nth order tensor in R3 , n = 1, 2, 3, . . . is an object such that

1. In any cartesian coordinate system, there is a rule that associates C with a unique ordered
set of 3n scalars, Ci1 i2 ...in , called the components of C in that coordinate system.

2. If Ci1 i2 ...in and Cq0 1 q2 ...qn are the components of C with respect to two Cartesian coordinate
systems, then
Ci1 i2 ...in = li1 q1 li2 q2 . . . lin−1 qn−1 lin qn Cq1 q2 ...qn
where the lij qk ’s are defined as above.

With this definition, we can now work out the consistency rules for 2nd order tensors. They
0 . For example, we consider t = n · T, i.e. t = T n , t0 = T 0 n0 .
must satisfy Tij = lip ljq Tpq i ij j p pq q
Assuming that these are consistent, we have

ti = lip t0p = Tij nj = lip Tpq

0 0 0
nq = lip Tpq ljq nj .

Equivalently, we get
0
(Tij − lip ljq Tpq )nj = 0.
0 as expected.
For this to hold for all nj , we have Tij = lip ljq Tpq

12