Sunteți pe pagina 1din 177

Fluid Dynamics:

Theory and Computation


Dan S. Henningson
Martin Berggren
August 24, 2005

Contents
1 Derivation of the Navier-Stokes equations
1.1 Notation . . . . . . . . . . . . . . . . . . . .
1.2 Kinematics . . . . . . . . . . . . . . . . . .
1.3 Reynolds transport theorem . . . . . . . . .
1.4 Momentum equation . . . . . . . . . . . . .
1.5 Energy equation . . . . . . . . . . . . . . .
1.6 Navier-Stokes equations . . . . . . . . . . .
1.7 Incompressible Navier-Stokes equations . .
1.8 Role of the pressure in incompressible flow .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

7
7
8
14
15
19
20
22
24

2 Flow physics
2.1 Exact solutions . . . . . . . .
2.2 Vorticity and streamfunction
2.3 Potential flow . . . . . . . . .
2.4 Boundary layers . . . . . . .
2.5 Turbulent flow . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

29
29
32
40
48
54

3 Finite volume methods for incompressible flow


3.1 Finite Volume method on arbitrary grids . . . . .
3.2 Finite-volume discretizations of 2D NS . . . . . .
3.3 Summary of the equations . . . . . . . . . . . . .
3.4 Time dependent flows . . . . . . . . . . . . . . .
3.5 General iteration methods for steady flows . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

59
59
62
65
66
69

4 Finite element methods for incompressible flow


4.1 FEM for an advectiondiffusion problem . . . . . . . . . .
4.1.1 Finite element approximation . . . . . . . . . . . .
4.1.2 The algebraic problem. Assembly. . . . . . . . . .
4.1.3 An example . . . . . . . . . . . . . . . . . . . . . .
4.1.4 Matrix properties and solvability . . . . . . . . . .
4.1.5 Stability and accuracy . . . . . . . . . . . . . . . .
4.1.6 Alternative Elements, 3D . . . . . . . . . . . . . .
4.2 FEM for NavierStokes . . . . . . . . . . . . . . . . . . .
4.2.1 A variational form of the NavierStokes equations
4.2.2 Finite-element approximations . . . . . . . . . . .
4.2.3 The algebraic problem in 2D . . . . . . . . . . . .
4.2.4 Stability . . . . . . . . . . . . . . . . . . . . . . . .
4.2.5 The LBB condition . . . . . . . . . . . . . . . . . .
4.2.6 Mass conservation . . . . . . . . . . . . . . . . . .
4.2.7 Choice of finite elements. Accuracy . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

71
71
72
73
74
76
76
80
82
83
85
85
86
87
88
88

A Background material
A.1 Iterative solutions to linear systems .
A.2 Cartesian tensor notation . . . . . .
A.2.1 Orthogonal transformation .
A.2.2 Cartesian Tensors . . . . . .
A.2.3 Permutation tensor . . . . . .
A.2.4 Inner products, crossproducts
A.2.5 Second rank tensors . . . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

95
. 95
. 98
. 98
. 99
. 100
. 100
. 101

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
and determinants .
. . . . . . . . . . .
3

.
.
.
.
.
.
.

A.2.6 Tensor fields . . . . . . . . . . . .


A.2.7 Gauss & Stokes integral theorems
A.2.8 Archimedes principle . . . . . . . .
A.3 Curvilinear coordinates . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

102
103
104
106

B Recitations 5C1214
B.1 Tensors and invariants . . . . . . . . . . . . . . . . . . . . .
B.2 Euler and Lagrange coordinates . . . . . . . . . . . . . . . .
B.3 Reynolds transport theorem and stress tensor . . . . . . . .
B.4 Rankine vortex and dimensionless form . . . . . . . . . . .
B.5 Exact solutions to Navier-Stokes . . . . . . . . . . . . . . .
B.6 More exact solutions to Navier-Stokes . . . . . . . . . . . .
B.7 Axisymmetric flow and irrotational vortices . . . . . . . . .
B.8 Vorticity equation, Bernoulli equation and streamfunction .
B.9 Flow around a submarine and other potential flow problems
B.10 More potential flow . . . . . . . . . . . . . . . . . . . . . . .
B.11 Boundary layers . . . . . . . . . . . . . . . . . . . . . . . .
B.12 More boundary layers . . . . . . . . . . . . . . . . . . . . .
B.13 Introduction to turbulence . . . . . . . . . . . . . . . . . . .
B.14 Old Exams . . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

109
109
113
117
120
124
129
132
136
140
146
148
152
155
157

C Study questions 5C1212

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

173

Preface

These lecture notes has evolved from a CFD course (5C1212) and a Fluid Mechanics course (5C1214) at
the department of Mechanics and the department of Numerical Analysis and Computer Science (NADA)
at KTH. Erik St
alberg and Ori Levin has typed most of the LATEXformulas and has created the electronic
versions of most figures.

Stockholm, August 2004


Dan Henningson
Martin Berggren

In the latest version of the lecture notes study questions for the CFD course 5C1212 and recitation
material for the Fluid Mechanics course 5C1214 has been added.

Stockholm, August 2005


Dan Henningson

Chapter 1

Derivation of the Navier-Stokes


equations
1.1

Notation

The Navier-Stokes equations in vector notation has the following form


u
1

+ (u ) u = p + 2 u

u
= 0

where the velocity components are defined

u = (u, v, w) = (u1 , u2 , u3 )
the nabla operator is defined as
=


,
,
x y z

,
,
x1 x2 x3

the Laplace operator is written as


2 =

2
2
2
+ 2+ 2
2
x
y
z

and the following definitions are used

kinematic viscosity
density
pressure

see figure 1.1 for a definition of the coordinate system and the velocity components.
The Cartesian tensor form of the equations can be written

ui
ui

+ uj

t
xj

ui
xi

1 p
2 ui
+
xi
xj xj

= 0

where the summation convention is used. This implies that a repeated index is summed over, from 1
to 3, as follows
ui ui = u 1 u1 + u 2 u2 + u 3 u3
Thus the first component of the vector equation can be written out as
7

y, x2
v, u2
u, u1

PSfrag replacements
x, x1
w, u3

z, x3
Figure 1.1: Definition of coordinate system and velocity components
PSfrag replacements
x, x1x2

y, x2
P t = 0
z, x3
u, u1
xi 0
v, u2
w, u3

ri

0
xi


, t

ui xi0 , t

x1

Figure 1.2: Particle path.

i=1

u1
u1
u1
1 p
u1
+ u1
+ u2
+ u3
=
+
t
x1
x2
x3
x1

or

u
u
u
u
1 p
+u
+v
+w
=
+
t
x
y
z
x

2 u1
2 u1
2 u1
+
+
x1 2
x2 2
x3 2


2u 2u 2u
+
+
x2
y 2
z 2
|
{z
}
2 u

1.2

Kinematics

Lagrangian and Euler coordinates


Kinematics is the description of motion without regard to forces. We begin by considering the motion of a
fluid particle in Lagrangian coordinates, the coordinates familiar from classical mechanics.
Lagrange coordinates: every particle is marked and followed in flow. The independent variables are
xi 0

initial position of fluid particle

time

where the particle path of P, see figure 1.2, is



ri = ri xi 0 , t

and the velocity of the particle is the rate of change of the particle position, i.e.

ui =

ri
t

Note here that when xi 0 changes we consider new particles. Instead of marking every fluid particle it
is most of the time more convenient to use Euler coordinates.
Euler coordinates: consider fixed point in space, fluid flows past point. The independent variables are
xi

space coordinates

time

Thus the fluid velocity ui = ui (xi , t) is now considered as a function


 of the coordinate xi and time t.
The relation between Lagrangian and Euler coordinates, i.e. xi 0 , t and (xi , t), is easily found by noting
that the particle position is expressed in fixed space coordinates xi , i.e.


at the time
xi = ri xi 0 , t

Material derivative

= t

Although it is usually most convenient to use Euler coordinates, we still need to consider the rate of change
of quantities following a fluid particle. This leads to the following definition.
Material derivative: rate of change in time following fluid particle expressed in Euler coordinates.
Consider the quantity F following fluid particle, where

 
F = FL xi 0 , t = FE (xi , t) = FE ri xi , t , t

The rate of change of F following a fluid particle can then be written

FE
F
FE ri
FE t
FE
=

=
+ ui
xi t
t t
t
xi
t
D
Based on this expression we define the material derivative
as
Dt

=
+ ui
=
+ (u )

Dt
t
xi
t
t
In the material or substantial derivative the first term measures the local rate of change and the second
measures the change due to the motion with velocity ui .
As an example we consider the acceleration of a fluid particle in a steady converging river, see figure
1.3. The acceleration is defined
aj =

Dui
ui
ui
=
+ uj
Dt
t
xj

which can be simplified in 1D for stationary case to


Du
u
u
=
+u
Dt
t
x
Note that the acceleration 6= 0 even if velocity at fixed x does not change. This has been experienced
by everyone in a raft in a converging river. The raft which is following the fluid is accelerating although
the flow field is steady.
a=

Description of deformation
Evolution of a line element
Consider the two nearby particles in figure 1.4 during time dt. The position of P2 can by Taylor expansion
be expressed as


P2 : ri dt + dri dt =


= ri (0) + ri (0) dt + dri (0) + ( dri ) dt
t

= xi + dxi + ui (0) dt + dui (0) dt


0

PSfrag replacements
x, x1
y, x2
z, x3
u, u1
v, u2
w, u3
x1
x2
xi0
ri xi0 , t

ui xi0 , t

P t = 0

u1

u2 > u 1
x

Figure 1.3: Acceleration of fluid particles in converging river.

x2

P2

ui d

x
u1
u2 > u 1

dt 

PSfrag replacements
x, x1
y, x2
z, x3
u, u1
v, u2
w, u3
x1
x2
xi0
ri xi0 , t

ui xi0 , t

P t = 0

dr
i

dxi0

ui dt
P1
xi


t
ri d

x1

x3
Figure 1.4: Relative motion of two nearly particles.

where we have used

( dri ) =
t
=






ri

ri
0
dx
=
dxj 0
j
xj 0 t
t xj 0
ui
dxj 0 = dui
xj 0

and where
ui
xj 0
dxj 0
dui
We can transform the expression
line element in Euler coordinates

change of ui with initial pos.


difference in initial pos.
difference in velocity

( dri ) = dui in Lagrange coordinates to an equation for a material


t

D
( dri ) = dui
Dt
= {expand in Euler coordinates}
ui
drj
=
xj

where
ui
xj
drj

change in velocity with spatial position


difference in spatial pos. of particles

Relative motion associated with invariant parts


We consider the relative motion dui =

ui
ui
drj by dividing
in its invariant parts, i.e.
xj
xj
ui
= ij + eij + eij
xj
| {z }
eij

where eij is the deformation rate tensor and

ij

eij

eij



uj
1 ui

anti-symmetric part
2 uj
xi


1 ui
uj
2 ur
+

ij
traceless part
2 xj
xi
3 ur
1 ur
ij isotropic part
3 ur

ui
, eij , describes the deformation and is considered in detail below, whereas the
xj
anti-symmetric part can be written in terms of the vorticity k and is associated with solid body rotation,
i.e. no deformation.
The anti-symmetric part can be written
The symmetric part of

dxi0
dui dt
ri dt

dri dt

x2

r2

R2

P1
P2

+ d12

R1
R

x1

x3

r1
r3
Figure 1.5: Deformation of a cube.

1
2

[ij ] = 21 u
 x2

1 u1
2 x3

u2
x1 
u3
x1

21

u1
x2

u2
x3

= {use = u , k = kji

21 3
0
1
2 1

0
= 21 3
1
2 2

1
2 2
12 1

u2
x1

u3
x2

u1
 x3
1 u2
2 x3

1
2

u3
x1 
u3
x2

ui }
xj

Deformation of a small cube


Consider the deformation of the small cube in figure 1.5, where we define
Rlk

= Rkl

rkl

= Rlk +

component k of side l
uk l
R dt
xj j
| {z }
dulk

uk
= R kl +
dt
xl

relative motion of side l

deformed cube

First, we consider the deformation on side 1, which can be expressed as


q


dR1 = r1 R = rk1 rk1 R

The inner product can be expanded as

rk1 rk1

"

u1
1+
dt
x1

2

u2
dt
x1

= {drop quadratic terms}


=R

u1
1+2
dt
x1

2

u3
dt
x1

2 #

R2

We have dropped the quadratic terms since we are assuming that dt is small. dR1 becomes

dR



u1
u1
= R 1+2
dt R = R 1 +
dt + ... R
x1
x1
u1
= R
dt = Re11 dt
x1

which implies that


dR1
= Re11
dt
ui
Thus the deformation rate of side 1 depends on e11 , both traceless and isotropic part of x
.
j
Second, we consider the deformation of the angle between side 1 and side 2. This can be expressed as

cos


2

+ d12

1/2
r1 r2
1 1
= rk1 rk2 rm
rm rn2 rn2
1
2
|r | |r |


 
1/2 
1/2
uk
uk
u1
u2
=
k1 +
dt
k2 +
dt 1 + 2
dt
1+2
dt
x1
x2
x1
x2




u1
u2
u1
u2
=
dt +
dt
1
dt
1
dt
x2
x1
x1
x2


u2
u1
=
+
dt = 2e12 dt
x2
x1
=

where we have dropped quadratic terms. We use the trigonometric identity


cos

+ d12 = cos cos d12 sin sin d12 d12


2
2
2

which allow us to obtain the finial expression

d12
= 2e12
dt
Thus the deformation rate of angle between side 1 and side 2 depends only on traceless part of
Third, we consider the deformation of the volume of the cube. This can be expressed as

dV

=
=

=
=
=

ui
xj .

1 2 3
r r r R3


u1
u1
1 + u1 dt



x1
x2 dt
x3 dt


u2
u2
2
dt
1
+
dt
dt
R3 u
R3
x
x
x
2
3
u13

u
u
3
3
x dt
1 + x3 dt
x2 dt
1




u1
u2
u3
3
R 1+
dt
1+
dt
1+
dt R3
x1
x2
x3


u1
u2
u3
3
R
+
+
dt
x1
x2
x3
uk
R3
dt
xk

where we have again omitted quadratic terms. Thus we have


dV
= R3 err
dt
ui
and the deformation rate of volume of cube (or expansion rate) depends on isotropic part of x
.
j
In summary, the motion of a fluid particle with velocity ui can be divided into the following invariant
parts














































V (t + t) V (t)

(t)

R3
R2
R1
r3
r2
r1
+ d12 V (t)

S (t + t)

V (t + t)
Figure 1.6: Volume moving with the fluid.

i)
ii)
iii)

ui

ui
uj

= 21 kij k
xj
xi


ui
uj
2 ur
1
eij = 2
+

ij
xj
xi
3 ur

ij =

1
2

iv)

1.3

solid body translation

eij =

1
3

solid body rotation


volume constant deformation

ur
ij
ur

volume expansion rate

Reynolds transport theorem

Volume integral following with the fluid


Consider the time derivative of a material volume integral, i.e. a volume integral where the volume is moving
with the fluid. We obtain the following expressions

D
Dt

Tij dV

V (t)

lim

t0
t

Tij (t + t) dV

V (t+t)

lim

t0
t

V (t)

Tij (t + t) dV

V (t+t)

V (t)

Tij (t) dV

Tij (t + t) dV

+
Tij (t + t) dV
Tij (t) dV

V (t)
V (t)

Z
Z
1

Tij
lim
Tij (t + t) dV +
dV
t0

t
t

V (t+t)V (t)

V (t)

The volume in the first integral on the last line is represented in figure 1.6, where a volume element
describing the change in volume between V at time t and t + t can be written as
u nt = uk nk t dV = uk nk t dS

This implies that the volume integral can be converted to a surface integral. This surface integral can
in turn be changed back to a volume integral by the use of Gauss (or Greens) theorem. We have
D
Dt

Tij dV

"I

lim

t0

V (t)

Tij (t + t) uk nk dS +
S(t)

Tij uk nk dS +
S(t)

Z 

V (t)

Tij
dV
t

V (t)

Tij
dV = {Gauss/Greens theorem}
t

V (t)


Tij

+
(uk Tij ) dV
t
xk

which is the Reynolds transport theorem.

Conservation of mass
By the substitution Tij and the use of the Reynolds transport theorem above we can derive the
equation for the conservation of mass. We have

Z
Z 
D

dV =
+
(uk ) = 0
Dt
t
xk
V (t)

V (t)

Since the volume is arbitrary, the following must hold for the integrand
0=

uk
D
uk

+
(uk ) =
+ uk
+
=
+
t
x
t
x
x
Dt
x
k
k
k
|{z} | {z }
|{z} | {z k}

1

3

2

4

where we have used the definition of the material derivative in order to simplify the expression. The
terms in the expression can be given the following interpretations:

1

2

3

4

:
:
:
:

accumulation of mass in fixed element


net flow rate of mass out of element
rate of density change of material element
volume expansion rate of material element

By considering the transport of a quantity given per unit mass, i.e. Tij = tij , we can simplify Reynolds
transport theorem. The integrand in the theorem can then be written
0 (cont. eq.)
*
0



tij

tij
Dtij

(tij ) +
(uk tij ) = tij  +
+ tij 
(uk ) + uk
=
t
xk
t
xk
xk
Dt
t

which implies that Reynolds transport theorem becomes
Z
Z
D
Dtij
tij dV =

dV
Dt
Dt
V (t)

V (t)

where V (t) again is a material volume.

1.4

Momentum equation

Conservation of momentum
The momentum equation is based on the principle of conservation of momentum, i.e. that the time rate of
change of momentum in a material region = sum of the forces on that region. The quantities involved are:
Fi
Ri
ui

body forces per unit mass


surface forces per unit area
momentum per unit volume

z }s|(k)
{

V (t + t) V (t)
III(k)
ri dt
 n
S (t + t)
dri dt
u
dl
P1
n
P2
V (t + t)
x1
R(nI )
x2
x3
3
R
nII
nI
R2
R1
dS
r3
2
R(nII ) r
r1

+
d
12
2
Figure 1.7:
Momentum balance for fluid element
V (t)
S (t)
V (t + t) V (t)
n
S (t + t)
u
n
R
V (t + t)
R
Figure 1.8: Surface force and unit normal.
We can put the momentum conservation in integral form as follows
Z
Z
Z
D
ui dV =
Fi dV +
Ri dS
Dt
V (t)

V (t)

S(t)

Using Reynolds transport theorem this can be written


Z
Z
Z
Dui

dV =
Fi dV +
Ri dS
Dt
V (t)

V (t)

S(t)

which is Newtons second law written for a volume of fluid: mass acceleration = sum of forces. To
proceed Ri must be investigated so that the surface integral can be transformed to a volume integral. In
order to do that we have to define the stress tensor.

The stress tensor


Remove a fluid element and replace outside fluid by surface forces as in figure 1.7. Here R(n) is the surface
force per unit area on surface dS with normal n, see figure 1.8. Momentum conservation for the small fluid
particle leads to

X


 (k)
Dui
dS dl = Fi dS dl + Ri nIj dS + Ri nII
Ri nIII(k)
s dl
j dS +
j
Dt
k

Letting dl 0 gives



0 = Ri nIj dS + Ri nII
j dS

Now nj = nIj = nII


j which leads to

Ri (nj ) = Ri (nj )
implying that a surface force on one side of a surface is balanced by an equal an opposite surface force
at the other side of that surface. Note that it is a general principle that the terms proportional to the
volume of a small fluid particle approaches zero faster than the terms proportional to the surface area of
the particle. Thus the surface forces acting on a small fluid particle has to balance, irrespective of volume
forces or acceleration terms.

u
P1
R(e1 ) = ( T11 , T21 , T31 )
n
V (t + t) P2
x1
T21
x2
x1
T11
x3
T31
PSfrag replacementsR3
2
x, x1 R
y, x2 R1
z, x3 r3
Figure 1.9: Definition of surface
force components on a surface with a normal in the 1-direction.
u, u1 r2
x2
v, u2 r1

R(e2 ) = ( T12 , T22 , T32 )


+
d
w,
u
3 12
2
V1 (t)
x
T22
xS2 (t)
V (t + t) xVi0 (t)
0
ri Sx(t
, t t)
i +

0
T12
u i xi , t u

P t = 0 n
V (t + t)
x
T32
u1
u2 > u 1
x1
x2
x3
xi 0
Figure 1.10: Definition uof dsurface
force components on a surface with a normal in the 2-direction.
i t
0
dxi
du
t
We now divide the surfacei dforces
into components along the coordinate directions, as in figures 1.9 and
r i dt
1.10, with corresponding definitions
for the force components on a surface with a normal in the 3-direction.

dri dtof the surface force on a surface dS with a normal in the j-direction.
Thus, Tij is the i-component
P1 surfaces along the coordinate directions cut by a slanted surface, as in
Consider a fluid particle with
P2
figure 1.11. The areas of the surface
elements are related by
x1
x2
dSj = ej n dS = nj dS
x3
where dS is the area of the3 slanted surface. Momentum balance require the surface forces to balance
R
in the element, we have
R2
R1
r03 = R1 dS T11 n1 dS T12 n2 dS T13 n3 dS
r2 surface force Ri can be written in terms of the components of the stress
which implies that the total
r1
tensor Tij as

+
d
12
2
V (t) R = T n + T n + T n = T n
i
i1 1
i2 2
i3 3
ij j
S (t)
V (t + t) V (t)
e2
S (t + t)
R
u
n
n
V (t + t)

R1
dS3

dS1
dS2

e1

e3
Figure 1.11: Surface force balance on a fluid particle with a slanted surface.

P tshear
= 0 stresses.
Here the diagonal components are normal stresses off-diagonal components are
Further consideration of the moment balance around a fluid particle show that Tij is axsymmetric tensor,
u1
i.e.
u2 > u 1
x1
Tij = Tji
x2
x3
Momentum equation
xi 0
i dt
Using the definition of the surface force in terms of the stress tensor, the momentumuequation
can now be
dxi0
written
dui dt
r dt
Z
Z
Z
Z 
Dui
Tij i


dV =
Fi dV +
Tij nj dS =
Fi +
dV
Dt
xj dri dt
V (t)
V (t)
S(t)
V (t)
P1
P2 have
The volume is again arbitrary implying that the integrand must itself equal zero. We
x1
x2
Dui
Tij
= Fi +

x3
Dt
xj
R3
R2
Pressure and viscous stress tensor
R1
r3 thus divide the
For fluid at rest only normal stresses present otherwise fluid element would deform. We
r2 on the motion of
stress tensor into an isotropic part, the hydrodynamic pressure p, and a part depending
r1
the fluid. We have

+
d
12
p
2
V
(t)
Tij = pij + ij
S (t)
V (t + t) V (t)
p
- hydrodynamic pressure, directed inward
ij - viscous stress tensor, depends on fluid motion
S (t + t)
u
n
V (t + t)
p

Newtonian fluid
It is natural to assume that the viscous stresses are functions of deformation rate e ij or strain. Recall that
the invariant symmetric parts are


uj
2 ur
1 ur
1 ui
+

ij +
ij
eij =
2 xj
xi
3 xr
3 xr
|
{z
} | {z }
eij

eij

where eij is the volume constant deformation rate and eij is the uniform rate of expansion. For an
isotropic fluid, the viscous stress tensor is a linear function of the invariant parts of e ij , i.e.
ij = eij + 2eij
where we have defined the two viscosities as
(T ):
(T ):

dynamic viscosity (here T is the temperature)


second viscosity, often=0

For a Newtonian fluid, we thus have the following relationship between the viscous stress and the strain
(deformation rate)


ui
uj
2 ur
ij = 2eij =
+

ij
xj
xi
3 xr
which leads to the momentum equation
 

Dui
p

ui
uj
2 ur

=
+

ij
+ Fi
Dt
xi
xj
xj
xi
3 xr

1.5

Energy equation

The energy equation is a mathematical statement which is based on the physical law that the rate of change
of energy in material particle = rate that energy is received by heat and work transfers by that particle.
We have the following definitions


e + 21 ui ui dV
ui Fi dV

energy of particle, with e the internal energy


work
| {zrate} of Fi on particle

force velocity

uj Rj dS

work rate of Rj on particle

ni qi dS

heat loss from surface, with qi the heat flux vector, directed outward

Using Reynolds transport theorem we can put the energy conservation in integral form as
D
Dt

V (t)



Z
Z
Z 

1
e + ui ui dV =
Fi ui dV +
[ni Tij uj ni qi ] dS =
Fi ui +
(Tij uj qi ) dV
2
xi
V (t)

S(t)

V (t)

Compare the expression in classical mechanics, where the momentum equation is mu = F and the
associated kinetic energy equation is
m d
2 dt

(u u)
work rate
( work

=
=
=

Fu
force velocity
force dist. )

From the integral energy equation we obtain the total energy equation by the observation that the
volume is arbitrary and thus that the integrand itself has to be zero. We have


1

qi
D
e + ui ui = Fi ui
(pui ) +
(ij uj )

Dt
2
xi
xi
xi
The mechanical energy equation is found by taking the dot product between the momentum equation
and u. We obtain


D 1
p
ij

ui ui = Fi ui ui
+ ui
Dt 2
xi
xj
Thermal energy equation is then found by subtracting the mechanical energy equation from the total
energy equation, i.e.
ui
ui
qi
De
= p
+ ij

Dt
xi
xj
xi
The work of the surface forces divides into viscous and pressure work as follows

(pui )
xi

= p

ui
xi

ui

(ij uj )
xi

ij

ui
xj

p
xi


2
+ uj

ij
xi

where following interpretations can be given to the thermal and the mechanical terms

1:

2:

thermal terms
( force deformation ): heat generated by compression and viscous dissipation
mechanical terms
( velocity force gradients ): gradients accelerate fluid and increase kinetic energy

The heat flux need to be related to the temperature gradients with Fouriers law
qi =

T
xi

where = (T ) is the thermal conductivity. This allows us to write the thermal energy equation as


De
ui

= p
++

Dt
xi
xi
xi
where the positive definite dissipation function is defined as
"

2 #
ui
1 uk
= ij
= 2 eij eij
xj
3 xk

2
1 uk
= 2 eij
ij
>0
3 xk
Alternative form of the thermal energy equation can be derived using the definition of the enthalpy
h=e+

We have
Dh
De 1 Dp
p
=
+
2
Dt
Dt
Dt

D
Dt
|{z}



u
xi
i

which gives the final result


Dh
Dp

=
++
Dt
Dt
xi

xi

To close the system of equations we need a


i)

thermodynamic equation

e = e (T, p)

ii)

equation of state

p = RT

simplest case:

e = cv T or h = cp T

where cv and cp are the specific heats at constant volume and temperature, respectively. At this time
we also define the ratio of the specific heats as
=

1.6

cp
cv

Navier-Stokes equations

The derivation is now completed and we are left with the Navier-Stokes equations. They are the equation
describing the conservation of mass, the equation describing the conservation of momentum and the equation
describing the conservation of energy. We have
D
uk
+
=0
Dt
xk
Dui
p
ij
=
+
+ Fi
Dt
xi
xj


De
ui

= p
++

Dt
xi
xi
xi

where

ij

ui
= ij
x

 j
uj
2 ur
ui
+

ij
=
xj
xi
3 xr

and the thermodynamic relation and the equation of state for a gas
e = e (T, p)
p = RT
We have 7 equations and 7 unknowns and therefore the necessary requirements to obtain a solution of
the system of equations.
unknown

1
ui
3
p
1
e
1
T P1
7

equations
continuity
1
momentum
3
energy
1
thermodyn.
1
gas law
P1

Equations in conservative form

A slightly different version of the equations can be found by using the identity

Dtij

=
(tij ) +
(uk tij )
Dt
t
xk

to obtain the system

+
(uk ) = 0

t
x
k

p
ij
(ui ) +
(uk ui ) =
+
+ Fi
t 
xk
xj 

 x
i



= Fi ui
e + ui ui +
uk e + ui ui
(pui ) +
(ij uj ) +

2
xk
2
xi
xi
xi
xi

These are all of the form

U G(i)
+
=J
t
xi

or

U G G G
+
+
+
=J
t
x
y
z

where
U = (, u1 , u2 , u3 , (e + ui ui /2))

is the vector of unknowns,


J = (0, F1 , F2 , F3 , ui Fi )
is the vector of the right hand sides and

G(i)

ui
u1 ui + p1i 1i
u2 ui + p2i 2i
u3 ui + p3i 3i
T
(e + ui ui /2) ui + pui + x
uj ij
i

is the flux of mass, momentum and energy, respectively. This form of the equation is usually termed
the conservative form of the Navier-Stokes equations.

1.7

Incompressible Navier-Stokes equations

The conservation of mass and momentum can be written

D
ui

+
=0

Dt
x
i

Dui
p
ij

=
+
+ Fi
Dt  xi
xj

ui
uj
2 uk

ij =
+

ij
xj
xi
3 uk

for incompressible flow = constant, which from the conservation of mass equation implies

ui
=0
xi
implying that a fluid particle experiences no change in volume. Thus the conservation of mass and
momentum reduce to

ui
ui
1 p

+ uj
=
+ 2 ui + Fi

t
xj
xi
ui
=0

x
i

since the components of the viscous stress tensor can now be written



ui
uj
2 uk
2 ui
+

ij =
= 2 ui
xj xj
xi
3 xk
xj xj
We have also introduced the kinematic viscosity, , above, defined as
= /
The incompressible version of the conservation of mass and momentum equations are usually referred
to as the incompressible Navier-Stokes equations, or just the Navier-Stokes equations in case it is evident
that the incompressible limit is assumed. The reason that the energy equation is not included in the
incompressible Navier-Stokes equations is that it decouples from the momentum and conservation of mass
equations, as we will see below. In addition it can be worth noting that the conservation of mass equation
is sometimes referred to as the continuity equation.
The incompressible Navier-Stokes equations need boundary and initial conditions in order for a solution
to be possible. Boundary conditions (BC) on solid surfaces are ui = 0. They are for obvious reasons usually
referred to as the no slip conditions. As initial condition (IC) one needs to specify the velocity field at the
initial time ui (t = 0) = u0i . The pressure field does not need to be specified as it can be obtained once the
velocity field is specified, as will be discussed below.

Integral form of the Navier-Stokes eq.


We have derived the differential form of the Navier-Stokes equations. Sometimes, for example when a
finite-volume discretization is derived, it is convenient to use an integral form of the equations.
An integral from of the continuity equation is found by integrating the divergence constraint over a
fixed volume VF and using the Gauss theorem. We have
Z
Z

(ui ) dVF = ui ni dSF = 0


xi
V

An integral form of the momentum equation is found by taking the time derivative of a fixed volume
integral of the velocity, substituting the differential form of the Navier-Stokes equation, using the continuity
equation and Gauss theorem. We have

Z
Z
Z 

ui

1 p
2 ui
ui dV =
dV =
(uj ui ) +

dV =
t
t
xj
xi
xj xj
VF

VF

VF

VF

1
ui
uj ui + pij
xj

xj

dV =

Z 

SF


p
ui
ui uj n j + n i
nj dS

xj

S (t + t)
u
n
V (t + t)
p

 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 


Tw

T0

Tw

U0

 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 


T0 , U 0

Figure 1.12: Examples of length, velocity and temperature scales.


Note that the integral from of the continuity equation is a compatibility condition for BC in incompressible flow.

Dimensionless form
It is often convenient to work with a non-dimensional form of the Navier-Stokes equations. We use the
following scales and non-dimensional variables
xi =

xi
L

t =

tU0
L

ui =

ui
U0

p =

p
p0

1
=
xi
L xi

U0
=
t
L t

where signifies a non-dimensional variable. The length and velocity scales have to be chosen appropriately from the problem under investigation, so that they represent typical lengths and velocities present.
See figure 1.12 for examples.
If we introduce the non-dimensional variables in the Navier-Stokes equations we find
2
U0 ui
U 2 u
p0 p
U0 2 ui

+ 0 uj i =
+

L t
L
xj
L xi
L2 xj xj

U
u

0
i

=0
L xi

Now drop * as a sign of non-dimensional variables and divide through with U02 /L, we find

ui
ui
p0 p

+ uj
= 2
+
2 u i
t
xj
U0 xi
U0 L

ui = 0
xi

We have defined the Reynolds number Re as


Re =

U0 L
U02 /L
=

U0 /L2

inertial forces
viscous forces

which can be interpreted as a measure of the inertial forces divided by the viscous forces. The Reynolds
number is by far the single most important non-dimensional number in fluid mechanics. We also define the
pressure scale as
p0 = U02
which leaves us with the final form of the non-dimensional incompressible Navier-Stokes equations as

ui
p
1 2
ui

+ uj
=
+
ui

t
xj
xi
Re

ui = 0
xi

Non-dimensional energy equation


We stated above that the energy equation decouple from the rest of the Navier-Stokes equations for incompressible flow. This can be seen from a non-dimensionalization of the energy equation. We use the
definition of the enthalpy (h = cp T ) to write the thermal energy equation as
DT
Dp
=
+ + 2 T
Dt
Dt
and use the following non-dimensional forms of the temperature and dissipation function
cp

T =
This leads to
DT
Dt

=
=

T T0
Tw T 0

L2

U02



2
U02

Dp

T +
+

U0 L cp
cp (Tw T0 ) Dt
U0 L


2
2
Dp
1
1
U0
1

T +
+

Re P r
cp (Tw T0 ) Dt
R
|
{z
}
a2

M a2 cp (Tw0T

where a0 speed of sound in free-stream and M a =


to obtain

0)

U0
is Mach number. We now drop and let M a 0
a0

DT
1
1
=

2 T
Dt
Re P r
c
where the Prandtl number P r = p , which takes on a value of 0.7 for air.
As the Mach number approaches zero, the fluid behaves as an incompressible medium, which can be
seen from the definition of the speed of sound. We have

1
a0 =
,
=

p T

where is the isothermal compressibility coefficient. If the density is constant, not depending on p, we
have that 0 and a0 , which implies that M a 0.

1.8

Role of the pressure in incompressible flow

The role of the pressure in incompressible flow is special due to the absence of a time derivative in the
continuity equation. We will illustrate this in two different ways.

Artificial compressibility
First we discuss the artificial compressibility version of the equations, used sometimes for solution of the
steady equations. We define a simplified continuity equation and equation of state as
ui
+
=0
and
p =
t
xi
This allows us to write the Navier-Stokes equations as

ui

p
1 2

+
(uj ui ) =
+
ui
t
xj
xi
Re

p + ui
= 0
t
xi
Let

p
u
v
w
u
p + u2
uv
uw

u=
e=
f =
g=
v
uv
p + v2
vw
w
uw
vw
p + 2

be the vector of unknowns and their fluxes. The Navier-Stokes equations can then be written
u e
f
g
1 2
+
+
+
=
Du
t
x y
z
Re
u
u
u
u
1
+A
+B
+C
= 2 Du
t
x
y
z
R

or

where the matrices above are defined

0
e
1
2u
=
A=
u 0 v
0 w

0
g
0
C=
=
u 1
0

0 0

0 0
, B = f =
u 0
u
0 u

0 0
0
0
v u 0
, D =
0
0 2v 0
0 w v
0

0 0
0 v u
1 0 2v
0 0 w
0
1
0
0

0
0

0
1

0
0
1
0

0
0

0
v

the eigenvalues of A, B, C are the wave speeds of plane waves in the respective coordinate directions,
they are






p
p
p
u, u, u u2 + ,
v, v, v v 2 + and
w, w, w w2 +

Note that the effective acoustic or pressure wave speed for incompressible flow.

Projection on a divergence free space


Second we discuss the projection of the velocity field on a divergence free space. We begin by the following
theorem.
Theorem 1. Any wi in can uniquely be decomposed into
wi = u i +
ui
= 0,
xi

where

p
xi

u i ni = 0

on

i.e. into a function ui that is divergence free and parallel to the boundary and the gradient of a function,
here called p.
p
Proof. We start by showing that ui and
is orthogonal in the L2 inner product.
xi

 Z
Z
I
p
p

ui ,
= ui
dV =
(ui p) dV = pui ni dS = 0
xi
xi
xi

The uniqueness of the decomposition can be seen by assuming that we have two different decompositions
and showing that they have to be equivalent. Let
(1)

wi = u i

p(1)
p(2)
(2)
= ui +
xi
xi

The inner product between


(1)

ui

(2)

ui

gives


 (1)
p p(2)
xi

(1)

and ui

(2)

ui

Z 
Z 
2 
 

2
(1)
(2)
(1)
(2)
(1)
(2)
(1)
(2)
0=
ui u i
+ ui u i
p p
dV =
ui u i
dV
xi

Thus we have

(1)

ui

(2)

= ui ,

p(1) = p(2) + C

r
+ d12
V (t)
S (t)
V (t + t) V (t)
S (t + t)
u
n
V (t + t)
p
T0 , U 0
Tw
L
T0
U0

Gradient fields

wi
p
xi

ui

Divergence free vectorfields // to boundary


Figure 1.13: Projection of a function on a divergence free space.
We can find an equation for a p with above properties by noting that
2 p =

wi
xi

in with

has a unique solution. Now, if


ui = w i
we have
0=

ui
wi
2p
=

;
xi
xi
xi xi

p
= wi ni on
n

xi
0 = u i ni = w i ni

and

To sum up, to project wi on divergence free space let wi = ui +


1) solve for p :

wi
= 2 p,
xi

2) let ui = wi

p
n

p
xi

p
=0
n

p
xi

This is schematically shown in figure 1.13. Note also that (wi ui ) ui .

Apply to Navier-Stokes equations


Let P be orthogonal projector which maps wi on divergence free part ui , i.e. Pwi = ui . We then have the
following relations
wi = Pwi +
Pui = ui ,

p
xi
p
=0
xi

Applying the orthogonal projector to the Navier-Stokes equations, we have






ui
p
ui
1 2
P
+
= P uj
+
ui
t
xi
xj
Re
Now, ui is divergence free and parallel to boundary, thus

ui
ui
=
t
t

However,
P2 ui 6= 2 ui

since 2 ui need not be parallel to the boundary. Thus we find an evolution equation without the
pressure, we have

ui
ui
1 2

= P uj
+
ui
t
xj
Re

|
{z
}
wi

The pressure term in the Naiver-Stokes equations ensures that the right hand side w i is divergence free.
We can find a Poisson equation for the pressure by taking the divergence of w i , according to the result
above. We find

2 p =

xi

ui
1 2
uj ui

+
ui =
uj

xj
Re

xi xj
|
{z
}
wi

thus the pressure satisfies elliptic Poisson equation, which links the velocity field in the whole domain
instantaneously. This can be interpreted such that the information in incompressible flow spreads infinitely
fast, i.e. we have an infinite wave speed for pressure waves, something seen in the analysis of the artificial
compressibility equations.

Evolution equation for the divergence


It is common in several numerical solution algorithms for the Naiver-Stokes equations to discard the divergence constraint and instead use the pressure Poisson equation derived above as the second equation in the
incompressible Naiver-Stokes equations. If this is done we may encounter solutions that are not divergence
free.
Consider the equation for the evolution of the divergence, found by taking the divergence of the momentum equations and substituting the Laplacian of the pressure with the right hand side of the Poisson
equation. We have




ui
1 2 ui
=

t xi
Re
xi
Thus the divergence is not automatically zero, but satisfies a heat equation. For the stationary case the
solution obeys the maximum principle. This states that the maximum of a harmonic function, a function
satisfying the Laplace equation, has its maximum on the boundary. Thus, the divergence is zero inside the
domain, if and only if it is zero everywhere on the boundary.
Since it is the pressure term in the Navier-Stokes equations ensures that the velocity is divergence free,
this has implications for the boundary conditions for the pressure Poisson equation. A priori we have none
specified, but they have to be chosen such that the divergence of the velocity field is zero on the boundary.
This may be a difficult constraint to satisfy in a numerical solution algorithm.

z, x3
u, u1
v, u2
w, u3
x1
x2
xi0
ri xi0 , t

ui xi0 , t

P t = 0

Chapter 2

Flowuxphysics
1

u2 > u 1
x1
x2
x3
2.1 Exact
xi0 solutions
ui dt
Plane Pouseuille
flow - exact solution for channel flow
dxi0

dui dt
The flow inside
the two-dimensional channel, see figure 2.1, is driven by a pressure difference p 0 p1 between
ri dt
 outlet of the channel. We will assume two-dimensional, steady flow, i.e.
the inlet and the
dri dt
P1

= 0,
u3 = 0
P2
t
x1
and that the
x2 flow is fully developed, meaning that effects of inlet conditions have disappeared. We have

x3
u1 = u1 (x2 ) , u2 = u2 (x2 )
R3
2
R
The boundary
R1 conditions are u1 = u2 = 0 at x2 = h. The continuity equation reads
r3
0
0
r2
7

7

u
u
u
1
2
3
r1
+  =0
 +

x
x
x
+
d
1
2
3
12
2


V (t)
where the first and the last term disappears due to the assumption of two-dimensional flow and that
S (t)
the flow is fully developed. This implies that u2 = C = 0, where the constant is seen to be zero from the
V (t + t) V (t)
boundary conditions.
S (t + t)
The steady momentum equations read
u

n
u
p
u

V (t + t)
u1 1 + u2 1 =
+ 2 u1
x1
x2
x1
p
u
u
p

u1 2 + u2 2 =
+ 2 u2 g
T0 , U 0
x1
x2
x2
Tw
which, by applying
the assumptions above, reduce to
L
T0
U0
x2
L
Gradient fields
wi
p
h
xi
ui
ree vectorfields // to boundary
p0 > p 1
x1
p























































































Figure 2.1: Plane channel geometry.
29

ui
Divergence free vectorfields // to boundary
L
h
x1
x2
p0 > p 1
p1

u
n
dS

un
t

Figure 2.2: Volume of fluid flowing through surface dS in time t

p
d2 u 1
+
x1
dx22
p
g
0=
x2

0=

The momentum equation in the x2 -direction (or normal direction) implies


p = gx2 + P (x1 )

p
dP
=
x1
dx1

showing that the pressure gradient in the x1 -direction (or streamwise direction) is only a function of
x1 . The momentum equation in the streamwise direction implies
0=

dP
d2 u 1
+
dx1
dx22

Since P (x1 ) and u1 (x2 ) we have


d2 u 1
1 dP
=
= const. (independent of x1 , x2 )
dx22
dx1
We can integrate u1 in the normal direction to obtain
u1 =

1 dP 2
x + C 1 x2 + C 2
2 dx1 2

The constants can be evaluated from the boundary conditions u1 (h) = 0, giving the parabolic velocity
profile

 x 2 
h2 dP
2
u1 =

1
2 dx1
h
Evaluating the maximum velocity at the center of the channel we have
umax =

h2 dP

>0
2 dx1

if flow is in the positive x1 -direction. The velocity becomes


 x 2
u1
2
=1
umax
h
Flow rate
From figure 2.2 we can evaluate the flow rate as dQ = u n dS. Integrating over the channel we find
Q=

Z
S

ui ni dS = {channel} =

h
h

u1 dx2 =

2h3 dP
4

= humax
3 dx1
3

L
h
x1
x2
p0 > p 1
p1
u
n
dS
u nt

u0
p0
















Figure 2.3: Stokes instantaneously plate.


Wall shear stress
The wall shear stress can be evaluated from the velocity gradient at the wall. We have
12 = 21 =

2 umax x2
du1
=
dx2
h2

implying that

2 umax
wall = x2 =h =
h

Vorticity

The vorticity is zero in the streamwise and normal directions, i.e. 1 = 2 = 0 and has the following
expression in the x3 -direction (or spanwise direction)
3 =

u2
u1
du1
2umax x2

=
=
x1
x2
dx2
h2

Stokes 1:st problem: instantaneously started plate


Consider the instantaneously starting plate of infinite horizontal extent shown in figure 2.3. This is a time
dependent problem where plate velocity is set to u = u0 at time t = 0. We assume that the velocity can be
written u = u(y, t) which using continuity and the boundary conditions imply that normal velocity v = 0.
The momentum equations reduce to
1 p
2u
+ 2
x
y
1 p
0 =
g
y

u
t

The normal momentum equation implies that p = gy + p0 , where p0 is the atmospheric pressure at
the plate surface. This gives us the following diffusion equation, initial and boundary conditions for u
u
2u
= 2
t
y
u(y, 0) = 0 0 y
u(0, t) = u0 t > 0
u(, t) = 0 t > 0

We look for a similarity solution, i.e. we introduce a new dependent variable which is a combination of
y and t such that the partial differential equation reduces to an ordinary differential equation. Let

u
= C y tb
u0
where C and b are constants to be chosen appropriately. We transform the time and space derivatives
according to
f () =

y
2
y 2

d
d
= Cbytb1
t d
d
d
d
=
= Ctb
y d
d
2
d
= C 2 t2b 2
d
=

If we substitute this into the equation for u and use the definition of , we find an equation for f
df
d2 f
= C 2 t2b 2
d
d
where no explicit dependence on y and t can remain if a similarity solution is to exist. Thus, the
coefficient of the two -dependent terms have to be proportional to each other, we have
bt1

bt1 = C 2 t2b
where is a proportionality constant. The exponents of these expressions, as well as the coefficients in
front of the t-terms have to be equal. This gives
b=

1
2

C=

We choose = 2 and obtain the following


f 00 + 2f 0 = 0

1
2

y
=
2 t

d
where 0 = d
. The boundary and initial conditions also have to be compatible with the similarity
assumption. We have

u(y, 0)
u(0, t)
= f () = 0
= f (0) = 1
u0
u0
This equation can be integrated twice to obtain
Z
2
f = C1
e d + C2

u(, t)
= f () = 0
u0

Using the boundary conditions and the definition of the error-function, we have
Z
2
2
f =1
e d = 1 erf()
0
This solution is shown in figure 2.4, both as a function of the similarity variable and the unscaled
normal coordinate y. Note that the time dependent solution is diffusing upward.
From the velocity we can calculate the spanwise vorticity as a function of the similarity variable and
the unscaled normal coordinate y. Here the maximum of the vorticity is also changing in time, from the
infinite value at t = 0 it is diffusing outward in time.
v
u
u0 2

=
e
x y
t
The vorticity is shown in figure 2.5, both as a function of the similarity variable and the unscaled
normal coordinate y. Note again that the time dependent solution is diffusing upward.
3 = z =

2.2

Vorticity and streamfunction

Vorticity is an important concept in fluid dynamics. It is related to the average angular momentum of a
fluid particle and the swirl present in the flow. However, a flow with circular streamlines may have zero
vorticity and a flow with straight streamlines may have a non-zero vorticity.

rfields // to boundaryn
Divergence free vectorfields // to boundary n
dS
dS
L
L
u nt
u nt
hx
hx
x1 y
x1 y

xy
x2u
= 2u
0
0
yp=>
p 4t
t
p0 > p14t
0
1p
p0
0
y
p1
1
= p4t
u
u
n
n
y = 4t
dS
dS
t
u
u
u nt
u nt
u0
u0
x 2.4: Velocity above the instantaneously started plate.x a) U as a function of the similarity variable
Figure
y
y
. b) U (y, t) for various times.
u0
u0
p0
p0
y
y
= 4t
= 4t
8

1.8

1.6

1.4

1.2

0.8

0.6

0.4

0.2
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

y = 4t
t

y
z =

()u0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.8

u
u0

u
u0

1.6

y = 4t
ty

1.4
1.2
1
0.8

0.6

2
0.4

0.2
0

0.1

0.2

0.3

0.4

t
u0

0.5

0.6

0.7

0.8

0.9

z t
u0

=
()

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

()u

z =
=
()

Figure 2.5: Vorticity above the instantaneously started plate. a) z t/u0 as a function of the similarity
variable . b) z (y, t) for various times.

Vorticity and circulation


The vorticity is defined mathematically as the curl of the velocity field as
=u
In tensor notation this expression becomes
i = ijk

uk
xj

Another concept closely related to the vorticity is the circulation, which is defined as
I
I
=
ui dxi =
ui ti ds
C

In figure 2.6 we show a closed curve C. Using Stokes theorem we can transform that integral to one over
the area S as
Z
Z
uk
=
ijk ni
dS =
i ni dS
xj
S
S

This allows us to find the following relationship between the circulation and vorticity d = i ni dS, which
can also be written
d
= i ni
dS
Thus, one interpretation is that the vorticity is the circulation per unit area for a surface perpendicular to
the vorticity vector.
Example 1. The circulation of an ideal vortex. Let the azimuthal velocity be given as u = C/r. We can
calculate the circulation of this flow as
I
Z 2
=
u r d = C
d = 2C
C

Thus we have the following relationship for the ideal vortex


u =

2r

u0
y = 4t
t

y
()u0
z =
t

dxi = ti ds

Figure 2.6: Integral along a closed curve C with enclosed area S.


The ideal vortex has its name from the fact that its vorticity is zero everywhere, except for an infinite value
at the center of the vortex.

Derivation of the Vorticity Equation


We will now derive an equation for the vorticity. We start with the dimensionless momentum equations.
We have
ui
ui
p
1 2
+ uj
=
+
ui
t
xj
xi
Re
We slightly modify this equation by introducing the following alternative form of the non-linear term


ui

1
uj
=
uj uj + ijk j uk
xj
xi 2

) of the momentum
this can readily be derived using tensor manipulation. We take the curl (pqi x
q
equations and find





ui
2
1

2p
1
ui
pqi
+ pqi
uj uj + pqi
(ijk j uk ) = pqi
+
pqi 2
t
xq
xq xi 2
xq
xq xi
Re
xq

The second and the fourth terms are zero, since pqi is anti-symmetric in i, q and xq xi is symmetric in
i, q
2
pqi
=0
xq xi
The first term and the last term can be written as derivatives of the vorticity, and we now have the following
from of the vorticity equation

1 2
p
+ pqi
(ijk j uk ) =
p
t
xq
Re
The second term in this equation can be written as
ipq ijk

(j uk ) = (pj qk pk qj )
(j uk ) =
(p uk )
(j up ) =
xq
xq
xk
xj
uk

Due to continuity

j
xj

p
uk
j
up
+ p
up
j
xk
xk
xj
xj

2u

= jpq xj xq p = 0 and we are left with the relation


ipq ijk

p
up
(j uk ) = uk
j
xq
xk
xj

Thus, the vorticity equation becomes


p
p
up
1 2
+ uk
= j
+
p
t
xk
xj
Re
or written in vector-notation

D
1 2
= ( )u +

Dt
Re

Components in Cartesian coordinates


In cartesian coordinates the components of the vorticity vector in three-dimensions become


ex ey ez 







w v
u w
v
u


= x y
=

e
+

e
+

ez
x
y
z
y
z
z
x
x y
u
v
w

For two-dimensional flow it is easy to see that this reduces to a non-zero vorticity in the spanwise direction,
but zero in the other two directions. We have
x = 0,

y = 0,

z =

v
u

x y

This simplifies the interpretation and use of the concept of vorticity greatly in two-dimensional flow.

Vorticity and viscocity


There exists a subtle relationship between flows with vorticity and flows on which viscous forces play a
role. It is possible to show that
6= 0 viscous forces 6= 0
This means that without viscous forces there cannot be a vorticity field in the flow which varies with the
coordinate directions. This is expressed by the following relationship
ij
k
= 2 ui = ijk
xj
xj
This can be derived in the following manner. We have
ijk

k
xj


un
xj xm
2 un
= (im jn in jm )
xj xm
= ijk kmn

2 ui
xj xj

= 2 ui

Terms in the two-dimensional vorticity equation


In two dimensions the equation for the only non-zero vorticity component can be written
3
3
3
+ u1
+ u2
= 2 3
t
x1
x2
We will take two examples elucidating the meaning of the terms in the vorticity equation. First the diffusion
of vorticity and second the advection or transport of vorticity.
Example 2. Diffusion of vorticity: the Stokes 1st problem revisited.
From the solution of Stokes 1st problem above, we find that the vorticity has the following form
3 = z = (y, t)
and that it thus is a solution of the following diffusion equation

2
= 2
t
y
The time evolution of the vorticiy can be seen in figure 2.7, and is given by
u0 2
y
=
e ,
where =
t
2 t

where the factor t can be identified as the diffusion length scale, where we have used the following
dimensional relations

L2
[] =
, [t] = T [ t] = L
T

ree vectorfields // to boundary


L
h
x1
x2
p0 > p 1
p1
u
n
dS
u nt
x
y
u0
p0
y
= 4t
u
u0

u0
p0

y
= 4t
u
u0

y = 4t
t
y
y
()u0
z =
t
t

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Figure 2.7: A sketch of the evolution of vorticity in Stokes 1st problem.


v, y

y = 4t
t

()u0

z = t
z

}|

6= 0

x, u

Figure 2.8: Stagnation point flow with a viscous layer close to the wall.
In this flow the streamlines are straight lines and therefore the non-linear term, or the transport/advection
term, is zero. We now turn to an example where the transport of vorticity is important.
Example 3. Transport/advection of vorticity: Stagnation point flow.
The stagnation point flow is depicted in figure 2.8, and is a flow where a uniform velocity approaches a
plate where it is showed down and turned to a flow parallel two the plates in both the positive and negative
x-direction. At the origin we have a stagnation point.
The inviscid stagnation point flow is given as follows

v
u
u = cx
inviscid:
z = =

=0
v = cy
x y
Note that this flow has zero vorticity and that it does not satisfy the no-slip condition (u = v = 0, y = 0).
In order to satisfy this condition we introduce viscosity, or equally, vorticity. Thus close to the plate we
have to fulfill the equation
 2


2
u
+v
=
+
x
y
x2
y 2
The two first terms represent advection or transport of vorticity. We will try a simple modification of the
inviscid flow close to the boundary, which is written as

u = xf 0 (y)
v = f (y)
This satisfies the continuity equation, and will be zero on the boundary and approach the inviscid flow far
above the plate if we apply the following boundary conditions

u = v = 0,
y=0
u cx, v cy, y
The vorticity can be written = xf 00 (y). We introduce this and the above definitions of the velocity
components into the vorticity equation, and find
f 0 f 00 + f f 000 + f 0000 = 0

p0

1.5

y
4t
u
u0

y = 4t
t

y00
0
F , F0

z =F,()u
t
t

F
F0

0.5

F 00
0

Figure 2.9: Solution to the non-dimensional equation for stagnation point flow. F is proportional to the
normal velocity, F 0 to the streamwise velocity and F 00 to the vorticity.
The boundary conditions become

f = f0 = 0
y=0
f cy, f 0 c, y

An integral to the equation exists, which can be written


f 02 f f 00 f 000 = K = {evaluate y } = c2
This is as far as we can come analytically and to make further numerical treatment simpler we make the
equation non-dimensional with and c. Note that they have the dimensions
[] =
and that we can use the length scale
dependent variables as

L2
,
T

[c] =

1
T

/c and the velocity scale

F ()

c to scale the independent and the

y
p
/c
f (y)

This results in the following non-dimensional equation


F 02 F F 00 F 000 = 1
with the boundary conditions

F = F 0 = 0, = 0
F ,

In figure 2.9 a solution of the equation is shown. It can be seen that diffusion of vorticity from the wall
is balanced by transport downward by v and outward by u. The thin layer of vortical flow close to the wall
p
has the thickness /c.

Stretching and tilting of vortex lines

In three-dimensions the vorticity is not restricted to a single non-zero component and the three-dimensional
vorticity equation governing the vorticity vector becomes
i
i
ui
1 2
+ uj
= j
+
i
t
xj
xj
|Re {z }
| {z }
advection of vorticity

diffusion of vorticity

4t
u
u0

dr

y = 4t
t

y
()u0
z =
t
t

C(t)
Figure 2.10: A closed material curve.
The complex motion of the vorticity vector for a full three-dimensional flow can be understood by an
analogy to the equation governing the evolution of a material line, discussed in the first chapter. The
equation for a material line is
D
ui
dri = drj
Dt
xj
Note that if we disregard the diffusion term, i.e. the last term of the vorticity equation, the two equations
are identical. Thus we can draw the following conclusions about the development of the vortex vector
i) Stretching of vortex lines (line k ) produce i like stretching of dri produces length
ii) Tilting vortex lines produce i in one direction at expense of i in other direction

Helmholz and Kelvins theorems


The analogy with the equation for the material line directly give us a proof of Helmholz theorem. We have
Theorem 1. Helmholz theorem for inviscid flow: Vortex lines are material lines in inviscid flow.
A second theorem valid in inviscid flow is Kelvins theorem.
Theorem 2. Kelvins theorem for inviscid flow: Circulation around a material curve is constant in inviscid
flow.
Proof. We show this by considering the circulation for a closed material curve, see figure 2.10,
m =

ui dri
C(t)

We can evaluate the material time derivative of this expression with the use of Lagrangian coordinates
as follows
I
Dm
D
=
ui dri
Dt
Dt C(t)

Lagrange coordinates
ui = ui (x0i , t) = ui (x0i (m), t)
=

ri = ri (x0i , t) = ri (x0i (m), t)


I
d
ri
dm
=
ui

m
dt


I m
I
ui ri
ri
=
dm +
ui
dm
t m
m t m
m
The last term of this expression will vanish since
I





I
I
ri
ui

1
ui
dm =
ui
dm =
ui ui dm = 0
m
2
t m
m
m
m m

u0y, v
y = 4t
t

z = ()u
t

u = (u, v)

dy

t
dx

x, u
Figure 2.11: A streamline which is tangent to the velocity vector.
Thus we find the following resulting expression
Dm
Dt

=
=
=

Dui
dri
C(t) Dt


I
p
1 2
+
ui dri

xi
Re
C(t)
I
1
2 ui dri 0,
Re
Re C(t)

Which shows that the circulation for a material curve changes only if the viscous force 6= 0

Streamfunction
We end this section with the definition and some properties of the streamfunction. To define the streamfunction we need the streamlines which are tangent to the velocity vector. In two dimensions, see figure
2.11, we have
dx
u
dx
dy
=

u dy v dx = 0
dy
v
u
v
and in three dimensions we can write
dx
dy
dz
=
=
u
v
w
For two-dimensional flow we can find the streamlines using a streamfunction. We can introduce a
streamfunction = (x, y) with the property that the streamfunction is constant along streamlines. We
can make the following derivation
d

dx +
dy
x
y



, v=
and check for consistency
=
assume u =
y
x
= u dy v dx

= 0,

on streamlines

We can also quickly check that the definition given above satisfies continuity. We have
u v
2
2
+
=

=0
x y
xy yx
Another property of the streamfunction is that the volume flux between two streamlines can be found
from the difference by their respective streamfunctions. We use figure 2.12 and figure 2.13 in the following
derivations of the volume flux:

z =

tu

nt
dS
u nt
x
y
u0
p0
y
= 4t

u
u0

B
= B
S
n
A
= A
Figure 2.12: Integration paths between two streamlines.

y = 4t
t

y
()u0
z =
t

ds

nx

t dy
ny

|n| = 1

dx
Figure 2.13: The normal vector to a line element.

QAB

=
=

ui ni ds

A
Z B

(nx u + ny v) ds


ds
1
=
dx
ny
Z B
Z B
=
(u dy v dx) =
d = B A

1
ds
=
,
dy
nx

2.3

Potential flow

Flows which can be found from the gradient of a scalar function, a so called potential, will be dealt
with in this section. Working with the velocity potential, for example, will greatly simplify calculations,
but also restrict the validity of the solutions. We start with a definition of the velocity potential, then
discuss simplifications of the momentum equations when such potentials exist, and finally we deal with
two-dimensional potential flows, where analytic function theory can be used.

Velocity potential
Assume that the velocity can be found from a potential. In three dimensions we have
ui =

xi

or u =

with the corresponding relations in two dimensions

u =

There are several consequences of this definition of the velocity field.

i) Potential flow have no vorticity, which can be seen by taking the curl of the gradient of the velocity
potential, i.e.
uk
2
i = ijk
= ijk
=0
xj
xj xk
ii) Potential flow is not influenced by viscous forces, which is seen by the relationship between viscous
forces and vorticity, derived earlier. We have
ij
k
= 2 ui = ijk
xj
xj
iii) Potential flow satisfies Laplace equation, which is a consequence of the divergence constraint applied
to the gradient of the velocity potential, i.e.
ui
2
=
= 2 = 0
xi
xi xi

Bernoullis equation
For potential flow the velocity field can thus be found by solving Laplace equation for the velocity potential.
The momentum equations can then be used to find the pressure from the so called Bernoullis equation.
We start by rewriting the momentum equation
ui
ui
1 p
+ uj
=
+ 2 ui gi3
t
xj
xi
using the alternative form of the non-linear terms, i.e.


ui

1
uj
=
uk uk ijk uj k
xj
xi 2
and obtain the following equation
ui

+
t
xi

1
p
uk uk + + gx3
2

= ijk uj k + ijk

k
xj

This equation can be integrated in for two different cases: potential flow (without vorticity) and stationary inviscid flow with vorticity.

into the momentum equation


i) Potential flow. Introduce the definition of the velocity potential ui = x
i
above, we have



1
p
+ uk uk + + gx3 = 0
xi t
2

p
1 2
+ (u1 + u22 + u23 ) + + gx3 = f (t)
t
2

ii) Stationary, inviscid flow with vorticity. Integrate the momentum equations along a streamline, see
figure 2.14. We have


Z
Z

1
p
ui
ti
uk uk + + gx3 ds = ijk uj k ds = 0
xi 2

u
1 2
p
(u + u22 + u23 ) + + gx3 = constant along streamlines
2 1

Thus, Bernoullis equation can be used to calculate the pressure when the velocity is known, for two
different flow situations.
Example 4. Ideal vortex: Swirling stationary flow without vorticity. The velocity potential u = in
polar coordinates can be written

ur = r

1
r

h
u
x1t
x2
p0 > py01

z = ()u
p1
t
ut
n
t = uu
dS
u nt
Figure 2.14: Integration
of the momentum equation along a streamline.
x
y
u0
r
p0
y
= 4t

u
u0

y = 4t
t

z = ()u
t
t
Figure 2.15: Streamlines for the ideal vortex.
The velocity potential satisfies Laplace equation, which in polar coordinates becomes


1 2
1 1
2
+
=0
=
r r r r
r 2
The first term vanishes since

= ur = 0. Integrate the resulting equation twice and we find


= C + D

The constant D is arbitrary so we can set it to zero. Using the definition of the circulation we find
u =

C
,
r

C=

We can easily find the streamfunction for this flow. In polar coordinates the streamfunction is defined

ur = r

Note that the streamfunction satisfies continuity, i.e. u = 0. We use the solution from the velocity
potential to find the streamfunction, we have

=
r
2r

ln r
2

The streamfunction is constant along streamlines resulting in circular streamlines, see figure 2.15
We can now use Bernoullis equation to calculate the pressure distribution for the ideal vortex as
1 2
u + p = p
2

p = p

2
8 2 r2

Note that the solution has a singularity at the origin.

Two-dimensional potential flow using analytic functions


For two-dimensional flows we can use analytic function theory, i.e. complex valued functions, to calculate
the velocity potential. In fact, as we will show in the following. Any analytic function represents a twodimensional potential flow.

We start by showing that both the velocity potential and the streamfunciton satisfy Laplace equation.
Velocity potential for two-dimensional flow is defined

u = x
Using the continuity equation gives

The streamfunction is defined

u v
+
= 2 = 0
x y

u =

By noting that potential flow has no vorticity we find


=

v
y
+
= 2 = 0
x y

Thus, both the velocity potential and the streamfunction satisfy Laplace equation.
We can utilize analytic function theory by introducing the complex function
F (z) = (x, y) + i(x, y)
where
z = x + iy = rei = r(cos + i sin )
which will be denoted the complex potential.
We now recall some useful results from analytic function theory.
F 0 (z) exists and is unique

F (z) is an analytic function

The fact that a derivative of a complex function should be unique rests on the validity of the Cauchy
Riemann equations. We can derive those equations by requiring that a derivative of a complex function
should be the same independent of the direction we take the limit in the complex plane, see figure 2.16.
We have
1
F 0 (z) = lim
[F (z + z) F (z)]
z0 z
1
= lim
[(x + x, y) (x, y) + i(x + x, y) i(x, y)]
x0 x
1
=
lim
[(x, y + y) (x, y) + i(x, y + y) i(x, y)]
iy0 iy

=
real part

x
y

i
=

=
imaginary part
x
i y
x
y
Thus we have the CauchyRiemann equations as a necessary requirement for a complex function to be
analytic, i.e.

=
and
=
x
y
y
x
Using the Cauchy-Riemann equations we can show that the real and imaginary part of the complex
potential satisfy Laplace equations and thus that they are candidates for identification with the velocity
potential and streamfunction of a two-dimensional potential flow. We have
2







2

=
=
=

x2

x y
y x
y 2
 
 

2


2

=
=
=

y 2
y x
x y
x2

u0
p0

iy

y
4t
u
u0

y = 4t
t

()u0

z = t
t

x
| {z }
iy

| {z }
x

Figure 2.16: The complex plane and the definitions of analytic functions.
In addition to satisfying the Laplace equations, we also have to show that the Cauchy-Riemann equations
are consistent with the definition of the velocity potential and the streamfunciton. The CauchyRiemann
equations are

u=
=

x
y

v=
=
y
x

which recovers the definitions of the velocity potential and the streamfunction. Thus any analytic complex
function can be identified with a potential flow, where the real part is the velocity potential and the
imaginary part is the streamfunction.
We can now define the complex velocity as the complex conjugate of the velocity vector since
W (z) =

dF

=
+i
= u iv
dz
x
x

It will be useful to have a similar relation for polar coordinates, see figure 2.17. In this case we have

u = ur cos u sin
v = ur sin + u cos
i
W (z) = u iv = [ur (r, ) iu (r, )] e|{z}

cos i sin

We will now take several examples of how potential velocity fields can be found from analytic functions.
Example 5. Calculate the velocity field from the complex potential F = U ei z. The complex velocity
becomes
W (z) = F 0 = U ei = U (cos i sin )
which gives us the solution for the velocity field in physical variables, see figure 2.18, as

u = U cos
v = U sin
Example 6. Calculate the velocity field from the complex potential F = Az n . In polar coordinates the
complex potential becomes
Arn ein = Arn cos(n) + iAr n sin(n)
Thus the velocity potential and the streamfunction can be identified as

= Arn cos(n)
= Arn sin(n)

u nt
P2
x
xy1
x
u20
x
p03
y 3

R
= 4t
Ru2
u01
R
y
y = 4t3
r
t
v
r2
1
r

d12y0
2 + ()u
z =
u

Vt
(t)
S (t)t
u
V (t + t) V (t)
S (t + t)
ur
u
u
n
V (t + t)
r
p
T0 , U 0

Tw
x
L
T0
Figure
2.17: The definition of polar coordinates.
U0
Gradient fields
wi
p
xi

ui
Divergence free vectorfields // to boundary
L
h
x1
x2
p0 > p 1
p1
u
n
dS
u nt
x
y
u0
p0
y
= 4t

u
u0

y = 4t
t

y
()u0
z =
t

t
Figure 2.18: Constant velocity at an angle .

xi
y =u nt
4t
=0
ui dxtt
dxiy0
=0
dui udy0t
()u
dtp00
z = ri
y 
dr= t
t
i d4t
ut

u01
P
n
y = 4t
P2
=0
x1t

x2 Figure 2.19: Flow towards a corner.


xy3

()u0

z = t
R3
R2t
=0
R1
r3
r2Figure 2.20: Uniform flow over a flat plate.
r1

+
d
12
2
and we can calculate the complex
V (t) velocity as
S (t)W (z) = F 0 = nAz n1
V (t + t) V (t)
= nArn1 ein ei
S (t + t)
= nArn1 [cos(n) + i sin(n)]ei
u
n velocity in polar coordinates as
This gives us the solution for the
V (t + t)

ur = nArn1 cos(n)
p
u = nArn1 sin(n)
T0 , U 0
Tw
This is a flow which approaches
a corner, see figure 2.19. The streamfunction is zero on the walls and on
L By solving for the zero streamlines we can calculate the angle between a
the stagnation point streamline.
T0
wall aligned with the x-axis and
the stagnation point streamline. We have
U0
Gradient fields = Arn sin(n) = 0 = k
wi
n

                               

xi
Figure 2.20 shows the uniform
flow over a flat plate which results when n = 1.
ui
Figure 2.21 shows the stagnation
point flow when n = 2. Here the analysis can be done in Cartesian
Divergence coordinates
free vectorfields // to boundary
L
F = Az 2
h
W (z) = F 0 = 2Az
x1
x2
= 2A(x + iy)

p0 > p 1
u = 2Ax
p1

v = 2Ay
u
n
Figure 2.22 shows the flow towards a wedge when 1 n 2. The velocity along the wedge edge (x axis
dS
in figure 2.22) is
u nt
ur = nArn1 cos(n) { = 0} u = ur = nArn1 = Cxn1
x
The wedge angle is
y
2
n1
u0
= 2
= 2
n
n
p0
y
= 4t

u
u0

y = 4t
t

()u0

z = t

                          
t

Figure 2.21: Stagnation point flow when n = 2.

u
u0

L
y = 4t
h
t x1
x2
py0 > p1

()u0
p1
z =
t
u
t
n
dS
u nt
Figure
2.22: The flow over a wedge when 1 n 2.
x
y
u0
p0
y
= 4t

u
u0

y = 4t
t

y
()u0
z =
t
t
Figure 2.23: Line source.
We now list several other examples of potential flow fields obtained from analytic functions.
Example 7. Line source.
F (z) =
=
=

m
ln z
2
m
ln(rei )
2
m
[ln r + i]
2

The velocity potential and the streamfunction become

= 2 ln r
The complex velocity can be written as

= m
2

W (z) = F 0 =

m
m i
=
e
2z
2r

which gives the solution in polar coordinates, see figure 2.23, as

ur =
2r

u = 0

We can now calculate the volume flux from the source as


Z 2
m
Q=
r d = m
2r
0
Example 8. Line vortex.
F (z) = i
=

ln z
2

[ i ln r]
2

x
y
u0
p0

y
4t
u
u0

y = 4t
t

z = ()u
t
t
Figure 2.24: Ideal vortex.
The velocity potential and the streamfunction become

The complex velocity can be written as

= 2

= ln r
2
W (z) = F 0 = i

i
e
2r

which gives the solution in polar coordinates, see figure 2.24, as

ur = 0

u =
2r

Example 9. The dipole has the complex potential


F (z) =

m
z

Example 10. The potential flow around a cylinder has the complex potential
F (z) = U z + U

r02
z

Example 11. The potential flow around a cylinder with circulation has the complex potential
F (z) = U z + U

r02

z
i
ln
z
2 r0

Example 12. Airfoils. The inviscid flow around some airfoils can be calculated with conformal mappings
of solutions from cylinder with circulation.

2.4

Boundary layers

For flows with high Reynolds number Re, where the viscous effects are small, most of the flow can be
considered inviscid and a simpler set of equations, the Euler equations, can be solved. The Euler equations
are obtained if Re in the Navier-Stokes equations. However, the highest derivatives are present in
the viscous terms, thus a smaller number of boundary conditions can be satisfied when solving the Euler
equations. In particular, the no flow condition can be satisfied at a solid wall but not the no slip condition.
In order to satisfy the no slip condition we need to add the viscous term. Thus, there will be a layer close
to the solid walls where the viscous terms are important even for very high Reynolds number flow. This
layer will be very thin and the flow in that region is called boundary layer flow.

z =

()u0

0
y
4t
u
u0

}
t

0
y = 4t
U
t

L
y
()u

0
z = t
Figure 2.25: Estimatet of the viscous region for an inviscid vorticity free inflow in a plane channel.

Ly, v

0
U
}

x, u

Figure 2.26: Boundary layer profile close to a solid wall.


We begin this section by estimating the thickness of the viscous region in channel with inviscid inflow
(zero vorticity), see figure 2.25.
By analogy to the Stokes solution for the instantaneously starting plate, we
can estimate the distance
the vorticity has diffused from the wall toward the channel centerline as ts . Where ts is time of a
L
fluid particle with velocity U has travelled length L, i.e. ts = . Thus we can estimate /L as
U

 1/2 

ts
1 L
1/2
1
0 as Re
=
=
=
=
L
L
L U
UL
Re
We will now divide the flow into a central/outer inviscid part and a boundary layer part close to walls,
where the no slip condition is fulfilled. Since we expect this region to be thin for high Reynolds numbers,
we expect large velocity gradients near walls.
We choose different length scales inside boundary layer and in inviscid region as indicated in figure 2.26,
as
Inviscid : U velocity scale for u, v
: L length scale for
x, y
BL :
:
:
:

velocity scale for


velocity scale for
length scale for
length scale for

u
v
x
y

The pressure scale is assumed to be U 2 for both the inner and the outer region. The unknown velocity
scale for the normal velocity is determined using continuity
u
v
+
=0
x
y
|{z}
|{z}


U
O
O

L
which implies that
U

=
L

U
L

In this manner both terms in the continuity equation have the same size.
Let us use these scalings in the steady momentum equations and disregard small terms. We obtain the
following estimates for the size of the terms in the normal momentum equation

y:

v
x
1 U
U
L L
 2

L
u

v
y
U 1 U

L L
 2

L
v

1 p
y

U2

2v
x2

L2

U
L
 2
1

Re L

2v
y 2
U
2 L

1
Re

where the last line was obtained by dividing all the terms on the second line by U 2 /. When R
and /L is small, only pressure term O(1), which implies that
p
=0
y

p = p(x)

Thus, the pressure is constant through the boundary layer, given by the inviscid outer solution.
Next we estimate the size of the terms in the streamwise momentum equation as
x: u

u
x

u
y

1 p
x

2u
x2

U2
L

U U

U2
L

U
L2

1
Re

2u
y 2

U
2
 2
1
L

Re

The first three terms are order one and the fourth term negligible when the Reynolds number become
large. In order to keep the last term, the only choice that gives a consistent approximation, we have to
assume that
1

L
Re
which is the same estimate as we found for the extent of the boundary layer in the channel inflow case.
Thus we are left with the parabolic equation in x since the second derivative term in that direction is
neglected. We have
u

u
u
1 p
2u
+v
=
+ 2
x
y
x
y

In summary, the steady boundary layer equations for two-dimensional flow is

u
u
1 p
2u

u
+v
=
+ 2
x
y
x
y

u + v = 0
x y

with the initial (IC) and boundary conditions (BC)


IC : u (x = x0 , y) = uin (y)
BC :

u (x, y = 0)
v (x, y = 0)
u (x, y )

=
=
=

0
0
ue (x) outer inviscid flow

where the initial condition is at x = x0 and the equations are marched in the downstream direction.
There are several things to note about these equations.
First, vin cannot be given as its own initial condition. By using continuity one can show that once the
u0 is given the momentum equation can be written in the form

v
+ f (y) v = g (y)
y

which can be integrated in the y-direction to give the initial normal velocity.
Second, the system of equations is a third order system in y and thus only three boundary conditions
can be enforced. We have the no flow and the no slip conditions at the wall and the u = ue in the free
stream. Thus no boundary condition for v in the free stream can be given.

u0

U
y, v
x, u
L
U

u0

                                 

Figure 2.27: Boundary layer flow over a flat plate with zero pressure gradient.
Third, ue (x) = uinv (x, y = 0), i.e. the boundary layer solution in the free stream approaches the inviscid
solution evaluated at the wall, since the thickness of the boundary layer is zero from the inviscid point of
view.
Fourth, the pressure gradient term in the momentum equation can be written in terms of the edge
velocity ue using the Bernoulli equation. We have
1
p (x) + u2e (x) = const
2
which implies that

1 dp
due
= ue
dx
dx

Blasius flow over flat plate


Consider the boundary layer equations for a flow over a flat plate with zero pressure gradient, i.e. the outer
inviscid flow is ue (x) = u0 . The boundary layer equations become

IC :

u (x = x0 , y)

u0

BC :

u (x, y = 0)
v (x, y = 0)
u (x, y )

=
=
=

0
0
u0

u
2u
u

u
+v
= 2
x
y
y

u + v = 0
x y

We introduce the two-dimensional stream function


u=

,
y

v=

which implicitly satisfies continuity. The boundary layer equations become


2
2
3

y xy
x y 2
y 3
with appropriate initial and boundary conditions.
Similarity solution
We will now look for a similarity solution. Let
u=

= u0 f 0 () ,
y

y
(x)

which by integration in the y-direction implies


(x, y) = u0 (x) f ()
We can now evaluate the various terms in the momentum equation. We give two below





y
0
0
v=
= u0 f + f
= u0 0 f f 0 0 = u0 (f 0 f ) 0
x
x

0
U
y, v
x, u
= yL
x/u0
U

u0
}

Blasius profile

0.2

0.4

f0 =

0.6

u
u0

0.8

Figure 2.28: Similarity solution giving the Blasius boundary layer profile.

= u0 f 00 0
xy

If these and the other corresponding terms are introduced into the momentum equation we have the
following equation for f

f 00 0
f 000
u20 f 0 f 00 0 + u20 (f 0 f )
= u0 2

which can be simplified to


u0 0 00
ff = 0

For a similarity solution to be possible, the coefficient in front of the f f 00 term has to be constant. This
implies
Z
Z
u0 0
1
1
dx
1 2
1 x
0
=

dx =

2
2
u0
2
2 u0
r
x
which implies that (x) =
, where we have chosen the constant of integration such that = 0
u0
when x = 0. We are left with the Blasius equation and boundary conditions
f 000 +

1
f 000 + f f 00 = 0
2
BC :

f (0)
= f 0 (0) = 0
0
f () = 1

The numerical solution to the Blasius equation gives the self-similar velocity profile seen in figure 2.28.
The boundary layer thickness can be evaluated as the height where the velocity has reached 99% of the
free stream value. We have
r
99
x
f 0 (99 ) = 0.99

99 = q
= 4.9

99 = 4.9
x
u0
u0

where it can be seen that the boundary layer thickness grows as

x.

Displacement thickness and skin friction coefficient


Another measure of the boundary layer thickness is the displacement of the wall needed in order to have
the same volume flux of the inviscid flow as for the flow in the boundary layer. With the help of figure 2.29
we can evaluate this as
Z y0
Z y
Z y
u0 dy u0
u dy
u0 dy =
0

which implies that

z =

u0

t
L y
=
x/u0
}
u
0
f
0 = u0
Blasius
U profile
y, v
x, u
L
U

u0
= y

                   

                   

Figure 2.29: The displacement thickness


= Pressure force

x/u0

f 0 = uu0
Blasius profile

                                                    

Separation point
Figure 2.30: Boundary layer separation caused by a pressure force directed upstream, a so called adverse
pressure gradient.

u
u0

dy =

x
u0

[1 f 0 ()] d

If we evaluate this expression for the Blasius boundary layer we have = 1.72

x
u0 .

A quantity which is of large importance in practical applications is the skin friction coefficient. The
skin friction at the wall is
w =

u
u0 00
u2
0.332u20
=
f (0) = p u00x f 00 (0) =

y y=0

Rex

where the last term is the expression evaluated for Blasius flow. This can be used to evaluate the skin
friction coefficient as
Cf =

w
1
2
2 u0

0.664
=
Rex

where the last terms again is the Blasius value.

Boundary layer separation




p
A large adverse pressure gradient x
> 0 may cause a back flow region close to wall, see figure 2.30.
The boundary layer equations cannot be integrated past a point of separation, since they become
singular at that point. This can be seen from the following manipulations of the boundary layer equations.
If we take the momentum equation
u

u
u
1 dp
2u
+v
=
+ 2
x
y
dx
y

take the normal derivative and use continuity we have


u

2u
2u
3u
+v 2 = 3
xy
y
y

If we again take the normal derivative and use continuity we can write this as
1
2 x

u
y

2

+u

3u
u 2 u
3u
4u

+v 3 = 4
2
2
xy
x y
y
y

This expression evaluated at the wall, with the use of the boundary conditions, become

1
2 x

!2

u
4 u
= 4
y y=0
y y=0
| {z }
2

which can be integrated to yield

!2
u
= 2 x + C
y y=0
!



At the point of separation x = xs , i.e. where u
= 0 this expression is
y
y=0

u
= xs x

y y=0

We can see that the boundary layer equations are not valid beyond point of separation since we have
a square root singularity at that point. In addition, the point of separation is the limiting point after
which there is back flow close to the wall. This means that there is information propagating upstream,
invalidating the downstream parabolic nature of the equations assumed by the downstream integration of
the equations.
In practical situations it is very important to be able to predict the point of separation, for example
when an airfoil stalls and experience a severe loss of lift.

2.5

Turbulent flow

Reynolds average equations


Turbulent flow is inherently time dependent and chaotic. However, in applications one is not usually
interested in knowing full the details of this flow, but rather satisfied with the influence of the turbulence
on the averaged flow. For this purpose we define an ensemble average as
N
1 X (n)
Ui = ui = lim
ui
N N
n=1
(n)

where each member of the ensemble ui is regarded as an independent realization of the flow.
We are now going to derive an equation governing the mean flow Ui . Divide the total flow into an
average and a fluctuating component u0i as
ui = ui + u0i

p = p + p0

and introduce into the Navier-Stokes equations. We find


u0i
ui
ui
u0
ui
u0
1 p0
1 p
+
+ u0j
+ uj i + uj
+ u0j i =

+ 2 u0i + 2 ui
t
t
xj
xj
xj
xj
xi
xi
We take the average of this equation, useing
u0i = 0

u0j ui = ui u0j = 0

which gives

ui
ui
1 p
 0 0

+ uj
=
+ 2 ui
ui uj

t
xj
xi
xj

ui = 0

xi

where the average of the continuity equation also has been added. Now, let U i = ui , as above, and drop
the 0 . We find the Reynolds average equations

Ui
Ui
1 P

+ Uj
=
+
(ij ui uj )
t
xj
xi
xj
Ui

xi

where ui uj is the Reynolds stress. Thus the effect of the turbulence on the mean is through an additional
stress Rij which explicitly written out is

u2
[Rij ] = [ui uj ] = uv
uw

uv
v2
vw

uw
vw
w2

The diagonal component of this stress is the kinetic energy of the turbulent fluctuations
k = Rii /2 = ui ui /2 =


1 2
u + v2 + w2
2

Turbulence modelling
One may derive equations for the components of the Reynolds stress tensor. However, they would involve
additional new unknowns, and the equations for those would in turn involve even more unknowns. This
closure problem implies that if one cannot calculate the complete turbulent flow, but is only interested in
the mean, the effect of the Reynolds stress components on in the Reynolds average equations have to be
modelled.
The simplest consistent model is to introduce a turbulent viscosity as a proportionality constant between
the Reynolds stress components and the strain or deformation rate tensor, in analogy with the stress-strain
relationship in for Newtonian fluids. We have
Rij =

2
kij 2T E ij
3

The first term on the right hand side is introduced to give the correct value of the trace of R ij . Here
the deformation rate tensor is calculated based on the mean flow, i.e.
E ij =

1
2

Ui
Uj
2 Ur
+

ij
xj
xi
3 xr

where we have assumed incompressible flow. Introducing this into the Reynolds average equations gives
 


Ui

P
2
1 P
Ui
+ Uj
=

+ k ij + 2 ( + T ) E ij =
+ ( + T ) 2 Ui
t
xj
xj

3
xi
where the last equality assumes that T is constant, an approximation only true for very simple turbulent
flow. It is introduced here only to point out the analogy between the molecular viscosity and the turbulent
viscosity.
The turbulent viscosity T must be modelled and most numerical codes which solved the Navier-Stokes
equations for practical applications uses some kind of model for T .
For simple shear flows, i.e. velocity fields of the form U (y), the main Reynolds stress component becomes
uv = T

U
y

where, in analogy with kinetic gas theory, T


T = uT l
Here, uT is a characteristic turbulent velocity scale and l is a characteristic turbulent length scale. In
kinetic gas theory l would be the mean free path of the molecules and uT the characteristic velocity of the
molecules.

x/u0

f 0 = uu0
Blasius profile

x
Separation point
= Pressure force

uT
l

!"!"
! #"
! #"
! #"
! #"
! #"
! #"
! #"
! #"
! #"
! #"
! #!
# #"

Figure 2.31: Prantdls mixing length l is the length for which a fluid particle displaced in the normal
direction can be expected to retain its horizontal momentum.
Zero-equation models
The simplest model for uT and l is to model them algebraically.
In simple free shear flows like jets, wakes and shear layers we can take T as constant in normal direction
or y-direction and to vary in a self-similar manner in x.
In wall bounded shear flow we can take l as Prandtls mixing length (1920s). This is a normal distance
over which the particle retains its momentum and is depicted in figure 2.31.
We can thus evaluate uT and find T




dU


2 dU


uT = l

T = l
dy
dy
p
Define w as the shear stress at the wall and u = w / as the skin friction velocity. With
l =y

uT = u ,

as Prandtls estimation of mixing length we get


dU
uT
u
=
=
dy
l
y

U
1
= ln y + C.
u

is the von Karman constant, approximately equal to 0.41. Thus the mixing length is proportional to
the distant from the wall and the region close to the wall in a turbulent flow has a logarithmic velocity
profile, called the log-region.
One-equation models
Instead of modelling the characteristic turbulent velocity algebraically, the next level of modelling uses the
assumption that
r
p
1
uT = k =
ui ui
2
and a differential equation is used to calculate the turbulent kinetic energy. We can derive such an
equation by taking the previously derived equation for Ui + ui and subtracting the equation for U , the
Reynolds average equation. We find


ui
ui
Ui
ui

p
ui
+ Uj
= uj
uj
+
ij +
+ u i uj
t
xj
xj
xj
xj

xj
We now take the average of the inner product between the turbulent fluctuations u i and the above
equation, and find

1
ui ui
2

+ Uj
xj

1
ui ui
2

Ui

= ui uj
+
xj
xj

puj

1

ui ui uj +
ui ui

2
2 xj

ui ui

xj xj
| {z }


where  is the mean dissipation rate of the turbulent kinetic energy. We have now introduced a number
of new unknowns which have to be modelled.

First, we model the Reynolds stress as previously indicated, i.e. ui uj = 32 kij T

Ui
xj

Uj
xi

Second, the turbulent dissipation is modelled purely based on dimension using k and l. We have
 = cD

3/2

L2
T .

since the dimension of  is [] =


Here cD is a modelling constant.
Third, a gradient diffusion model is used for the transport terms, i.e.
1
1
T k
ui ui uj puj =
2

P rk xj
where P rk is the turbulent Prandtl number.
For simplicity we write the resulting equation for the turbulent kinetic energy for constant T . We find
Dk
i
ui uj U
= ( + T ) 2 k
xj
Dt
rate of increase
diffusion rate
generation rate


dissipation rate

Two-equation models
The most popular model in use in engineering applications is a two equation model, the so called k--model.
This is based on the above model for k but instead of modelling  a partial differential equation is derived
governing the turbulent dissipation. This equation has the form as the equation for k, with a diffusion
term, a generation term and a dissipation term. Many new model constants need to be introduced in order
to close the  equation.
Once k and  is calculated the turbulent viscosity is evaluated as a quotient of the two, such that the
dimension is correct. We have
2

T = C
where C is a modelling constant.

k


L
T0
U0
Gradient fields
wi
p
xi

ui
Divergence free vectorfields // to boundary
L
h
x1
x2
p0 > p 1
p1
u
n
dS
u nt
3.1 Finite Volumexy method on arbitrary grids
u0
Equations with 1st order
p0 derivatives: the continuity equation
y

=
In finite volume methods one4t
makes
approximations of the differential equations in integral form. We start
u
u0 order derivatives, the continuity equation. Recall that the integral from of
with an equation with onlyfirst
y = 4t
this equation can be written
t
Z
Z

(uk ) dV = uk nk dS = 0
y
xk

()u0
V
S

z = t

Chapter 3

Finite volume methods for


incompressible flow

Evaluating the normal vector


t ni to the curve l for two-dimensional flow, see figure 3.1, we have the
normal with length dl as n dl =
L ( dy, dx). This implies that
}
Z
Z
0
uk nk dl = (u dy v dx) = 0
U
S
S
y, v
u
We now apply this to an x,
arbitrary
two-dimensional grid defined as in figure 3.2.
L definitions and approximations
We make use of the following
U
uab = ui+ 12 ,j (ui+1,j + ui,j ) /2
xab = xb xa
u0
y
=
x/u0

f 0 =vabuu0= vi+ 1 ,j (vi+1,j + vi,j ) /2


2
Blasius profile

y
x
Separation point
= Pressure force
x
dx
y
l
uT

yab = yb ya

x
dl = ( dx, dy)
dy

dx
dy n dl = ( dy, dx)
Figure 3.1: Normal vector to the curve l.
59

i, j + 1

i 1, j + 1
c

i 1, j

i + 1, j + 1

$%$%
& c0 $%
& $%
& $%
& b$%
& $%
& $%
& $&
$%$%
& $%
& $%
& $%
& $%
& $%
& $%
& $&
$%$%
& $%
& $%
& $%
& $%
& $%
& $%
& $&
$%$%
& $%
& $%
& $%
& $%
& $%
& $%
& $&
$%$%
& $%
& $%
& $%
& $%
& $%
& $%
& $&
$i,%j $%
& $%
& $%
& $%
& $%
& $%
& $%
& $&
$%$%
& $%
& $%
& $%
& $%
& $%
& $%
& $&
& $%
& $%
& $%
& $%
& $%
& $%
& $&
Aabcd $%$%
$%$%
& $%
& $%
& $%
& $%
& $%
& $%
& $&

d
i 1, j 1

d0

b0

i, j 1

i + 1, j

a0

i + 1, j 1

Figure 3.2: Arbitrary grid. Note that the grid lines need not be parallel to the coordinate directions.
with analogous expressions for the quantities on the bc, cd and da faces. We can now evaluate the
integral associated with the two-dimensional version of the continuity equation for the surface defined by
the points abcd. We have
Z

(u dy v dx)

ui+ 21 ,j yab vi+ 12 ,j xab + ui,j+ 12 ybc vi,j+ 12 xbc +

ui 21 ,j ycd vi 12 ,j xcd + ui,j 12 yda vi,j 21 xda


On a cartesian grid we have the relations
xab = xcd = ybc = yda = 0 and ycd = yab , xbc = xda
which implies that the approximation of the continuity equation can be written
Z

(u dy v dx)





ui+ 21 ,j ui 12 ,j yab + vi,j+ 12 + vi,j 12 xda =

1
1
(ui+1,j ui1,j ) yab + (vi,j+1 vi,j1 ) xda
2
2
Note that dividing with yab xda gives a standard central difference discretization of the continuity
equation, i.e.
ui+1,j ui1,j
vi,j+1 vi,j1
+
=0
2xda
2yab

Equations with 2nd derivatives: the Laplace equation


We use the Gauss or Greens theorem to derive the integral from of the two-dimensional Laplace equation.
We find

Z
Z 
Z

2
dV =
nj dS = {2D} =
dy
dx
0=
xj xj
xj
x
y
V

If we approximate this integral in the same manner as for the continuity equation we have





i+ 21 ,j

i 21 ,j

yab
ycd




xab +
i+ 21 ,j

xcd +
i 21 ,j







i,j+ 21

i,j 21

xbc +
y i,j+ 1

 2

yda
xda
y i,j 1

ybc

First derivatives are evaluated as mean value over adjacent control volumes (areas in the two-dimensional
case). We use Gauss theorem over the area defined by the points a0 b0 c0 d0 , see figure 3.2. For the threedimensional case we have
Z
Z
1

1
dV =
nj dS
V
xj
V
V

which in the two-dimensional case is approximated by


!





x i+ 1 ,j y i+ 1 ,j
A a 0 b0 c 0 d 0
2

a 0 b0 c 0 d 0

( dy, dx)

where
Z

a 0 b0 c 0 d 0

dy i+1,j ya0 b0 + b yb0 c0 + i,j yc0 d0 + yd0 a0

and the area evaluated as half magnitude of cross products of diagonals, i.e.
Aab Aa0 b0 c0 d0 =
and the value a is taken as the average
a =


1

xd0 b0 ya0 c0 yd0 b0 xa0 c0
2

1
(i,j + i+1,j + i,j1 + i+1,j1 )
4

Evaluating the derivatives


x , y along the other cell faces and substituting back we obtain a 9 point
formula which couples all of the nine points around the point i, j, of the form

Ai,j i+1,j+1
Fi,j i,j1

+
Bi,j i+1,j
+ Gi,j i1,j+1

+ Ci,j i+1,j1
+ Hi,j i1,j

+ Di,j i,j+1
+ Ii,j i1,j1

+ Ei,j i,j
=
0

If we have a total of N = m n interior nodes, we obtain N equations which in matrix form can be
written
Ax = b
Here xT = (11 , 12 , 13 , . . . , mn ) and b contains the boundary conditions. The matrix A is a N N
with the terms Aij , Bij , etc. as coefficients.
Rather than showing the general case in more detail, we instead work out the expression for cartesian
grids in two-dimensions. We find


Z  2
Z 

2
+
dx
dy
=
dy

dx

x2
y 2
x
y
!
!









y +

x
x i+ 1 ,j+
x i 1 ,j
y i,j+ 1
y i,j 1
2

We can evaluate the derivatives as





1
= (i+1,j i,j ) y
x i+ 1 ,j
x y
2

and so on for the rest of the terms, which implies that



Z 

y
x
dy
dx (i+1,j 2i,j + i1,j )
+ (i,j+1 2i,j + i,j1 )
x
y
x
y
Dividing with the area x y we obtain the usual 5-point Laplace formula, i. e.
i+1,j 2i,j + i1,j
i,j+1 2i,j + i,j1
+
=0
2
x
y 2
which of course also can be written in the matrix form shown for the general case above.

a
a0
i 1, j 1
i, j 1
i + 1, j 1
j

ui,j vi,j
pi,j

i+1

Figure 3.3: Cartesian finite-volume grid where the velocity components and the pressure are co-located.

3.2

Finite-volume discretizations of 2D NS

Co-located Cartesian grid


We will apply the finite volume discretizations to the Navier-Stokes equations in integral form, which in
two-dimensions can be written
Z
(u dy v dx) = 0

1
u
u dy uv dx + p dy
dy
Re x

ZZ
Z 
1
v
v
dS =
uv dy v 2 dx p dx
dy
t
Re x

ZZ

u
dS =
t

Z 


u
dx
y

v
dx
y

where we have used the earlier derived result n = ( dy, dx). We start with a co-located grid which
means that the finite volume used for both velocity components and the pressure is the same, see figure 3.3.
In addition we choose a Cartesian grid in order to simplify the expressions. Recall that the finite volume
discretization of the continuity equation becomes
1
1
(ui+1,j ui1,j ) y + (vi,j+1 vi,j1 ) x = 0
2
2
which is equivalent to the finite difference discretization
or

ui+1,j ui1,j
vi,j+1 vi,j1
+
=0
2x
2y

The discretization of the streamwise momentum equation needs both first and second derivatives. We
have seen that the finite volume discretizations of both are equal to standard second order finite difference
approximations. Thus we obtain the following result for the u-component
(uv)i,j+1 (uv)i,j1
u2i+1,j u2i1,j
ui,j
pi+1,j pi1,j
+
+
+
t
2x
2y
2x


1
ui+1,j 2ui,j + ui1,j
ui,j+1 2ui,j + ui,j1

+
=0
Re
x2
y 2
In a similar manner the discretization for the v-component becomes
2
2
(uv)i+1,j (uv)i1,j
vi,j+1
vi,j1
pi,j+1 pi,j1
vi,j
+
+
+
t
2x
2y
2y


1
vi+1,j 2vi,j + vi1,j
vi,j+1 2vi,j + vi,j1

+
=0
Re
x2
y 2

This simple discretization cannot be used in practice without some kind of modification. The reason is
that one solution to these equations are in the form of spurious checkerboard modes. Consider the following
solution to the discretized equations
ui,j = vi,j = (1)i+j g (t) ,

p = (1)i+j ,

x = y

i, j aa1
ii
+ 1,
1, jj
1
10
i, j a1j
ii
+ 1,
1, jj
1
1
i, j 1ji
+1
i + 1, ji
ui,jji
i+
vi,j1
i
u
pi,j
i,j
i+
1
pvi,j
i,j
u
pi,j
i,j
vi,j+1/2
vi,j
p
i,j
vi,j1/2
pi,j
v
i,j+1/2
ui1/2,j
pi,j
vi,j1/2
i 12
v
i,j+1/2
ui1/2,j
vi,j1/2
1i
ii
+ 122
ui1/2,j
+ 1121i
iii
+
i + 322
+ 111i
jii +
+
i + 3222
+ 111j
jji +
i
+ 3222
11j
jjj +
212
j 1j
j 12
ence condition.
j 1
ence condition.
m equation (u) .

j
i
i+1
ui,j
vi,j
pi,j

, -(
, -(
, -(
, -(
, -(
, -(
, -,
-(
, -(
, -(
, -(
, -(
, -(
, -,
-,(-(
, -(
, -(
, -(
, -(
, -(
, -,
-,(-(
vi,j+1/2
,
,
,
,
,
,+(
-,(
(
(
(
(
(
* +(
* +(
* +(
* +(
* -(
* -,+*
+*(+(
* +(
* +(
* +(
* +(
* +(
* +*
+*(+(
* +(
* +(
* +(
* +(
* +(
* +*
+*(+(
*
*
*
*
*
*
*/(
(
+
(
+
(
+
(
+
(
+
(
+
. /(
.p /(
. /(
. /(
. +(
. )(
' +*/. )(
' )(
' )(
' )(
' )(
' )(
' )(
' )'
ui1/2,j /.(/(
. /(
. /(
. i,j/(
. /(
. /(
. /(
. )(
' /. )(
' )(
' )(
' )(
' )(
' )(
' )(
' )'
/(
. /(
. /(
. /(
. /(
. /(
. /(
. )(
' /. )(
' )(
' )(
' )(
' )(
' )(
' )(
' )'
/(
. /(
. /(
. /(
. /(
. /(
. /(
. )(
' /. )(
' )(
' )(
' )(
' )(
' )(
' )(
' )'
/(
' )(
' )(
' )(
' )(
' )(
' )(
' )(
' )'
)(
vi,j1/2 )(
' )(
' )(
' )(
' )(
' )(
' )(
' )(
' )'
' )(
' )(
' )(
' )(
' )(
' )(
' )(
' )'
)(
' )(
' )(
' )(
' )(
' )(
' )(
' )(
' )'
)(

1010 1010
23234 23234 23234 22 4
4444
65656565

1
2

i+

1
2

i+1

i+

j+

1
2

j
j

1
2

j 1

3
2

control volume for divergence condition.


control volume for streamwise momentum equation (u) .
control volume for normal momentum equation (v) .

Figure 3.4: Cartesian finite-volume grid where the velocity components and the pressure uses staggered
grids.
which satisfies the divergence constraint. If it is input into the equation for u and v we find the expression
8
dg
+
g=0
dt
Re x2

8t

g = e Rex2

Thus we have a spurious solution consisting of checkerboard modes which are damped slowly for velocity
and constant in time for the pressure. This mode will corrupt any true numerical solution and yields this
discretization unusable as is. These modes are a result of the very weak coupling between nearby finite
volumes or grid points, and can be avoided if they are coupled by one sided differences or if they are damped
with some type of artificial viscosity.

Staggered cartesian grid


To eliminate the problem with spurious checkerboard modes, we can use a staggered grid as in figure 3.4,
where the control volumes for the streamwise, spanwise and pressure are different.
The control volume for the continuity equation is centered abound the pressure point and the discretization becomes




ui+ 21 ,j ui 12 ,j y + vi,j+ 12 vi,j 12 x = 0
or

ui+ 12 ,j ui 21 ,j
x

vi,j+ 21 vi,j 12
y

=0

The control volume for the streamwise velocity is centered around the streamwise velocity point and
the discretization becomes
(uv)i+1/2,j+1/2 (uv)i+1/2,j1/2
u2i+1,j u2i,j
ui+1/2,j
pi+1,j pi,j
+
+
+

t
x
y
x
1 ui+3/2,j 2ui+1/2,j + ui1/2,j
1 ui+1/2,j+1 2ui+1/2,j + ui+1/2,j1

=0
2
Re
x
Re
y 2

j 12
j
j1

Inflow
A

2
v1, 32
p1,1

1
B

u 23 ,1

87 87 87 87 87 87 87 87 87 87 87 87 87 87 87 87 87 87 87 87 87
C

Solid wall

0
1

Figure 3.5: Staggered grid near the boundary.


where the u2i+1,j , u2i,j , (uv)i+1/2,j+1/2 , (uv)i+1/2,j1/2 terms need to be interpolated from the points
where the corresponding velocities are defined.
The control volume for the normal velocity is centered around the normal velocity point and the discretization becomes
2
2
(uv)i+1/2,j+1/2 (uv)i1/2,j+1/2
vi,j+1
vi,j
vi,j+1/2
pi,j+1 pi,j
+
+
+

t
x
y
y

1 vi+1,j+1/2 2vi,j+1/2 + vi1,j+1/2


1 vi,j+3/2 2vi,j+1/2 + vi,j1/2

=0
2
Re
x
Re
y 2
where the (uv)i+1/2,j+1/2 , (uv)i1/2,j+1/2 , u2i,j+1 , u2i,j terms need to be interpolated from the points
where the corresponding velocities are defined.
For this discretization no checkerboard modes possible, since there is no decoupling in the divergence
constraint, the pressure or the convective terms.

Boundary conditions
To close the system we need to augment the discretized equation with discrete versions of the boundary
conditions. See figure 3.5 for a definition of the points near a boundary.
The boundary conditions on the solid wall, BC in figure 3.5, consists of the no flow and the no slip
conditions, i.e. u = v = 0. The normal points are located on the boundary so that the no flow condition
simply become
v1, 21 = v2, 21 = = 0
The evaluation of the u-equation at the point ( 23 , 1), for example, requires u 23 ,0 . We can find this value
as follows
0 = u 23 , 12 =


1
u 23 ,1 + u 23 ,0
2

u 23 ,0 = u 23 ,1

The boundary conditions on the inflow, AB in figure 3.5, consist given values of u and v, i.e. so called
Dirichlet conditions. The streamwise points are located on the boundary so that the boundary condition
is simply consist of the known values
u 21 ,1 , u 21 ,2 , . . .
The evaluation of the v-equation at the point (1, 32 ), for example, requires v0, 32 . We can find this value
as follows
v 21 , 32 =


1
v1, 32 + v0, 32
2

v0, 23 = 2v 21 , 32 v1, 32

v
If AB in figure 3.5 is an outflow boundary, the boundary conditions may consist of u
x = x = 0, i.e.
so called Neumann conditions. These conditions should be evaluated at i = 12 . For the streamwise and
normal velocity we find

u 12 ,2 = u 23 ,2 ,

v0, 32 = v1, 23

Note that no reference is made to the pressure points outside the domain, so that in this formulation
no boundary conditions for p is needed so far. However, depending on the particular discretization of the
time derivative and other the details of the solution method one may need to augment the the above with
boundary conditions for the pressure. If this is the case particular care has to be taken so that those
conditions do not upset the divergence free condition.

3.3

Summary of the equations

The discretized equations derived for the staggered grid can be written in the following form

ui+ 12 ,j

pi+1,j pi,j

+ Ai+ 21 ,j +
=0

t
x
vi,j+ 12
pi,j+1 pi,j
+ Bi,j+ 21 +
=0

t
y

D =0
i,j

where the expression for the Ai+ 12 ,j can be written


Ai+ 21 ,j =

(uv)i+ 1 ,j+ 1 (uv)i+ 1 ,j 1


u2i+1,j u2i,j
2
2
2
2
+
x
y
1 ui+ 12 ,j+1 2ui+ 21 ,j + ui+ 12 ,j1
1 ui+ 32 ,j 2ui+ 12 ,j + ui 12 ,j

Re
x2
Re
y 2
=

= {expand and interpolate using grid values}






1
1
2
1
1
1
1
1
1
3
1
=
ui 2 ,j ) +
vi,j+ 2 vi+1,j 2 vi,j 2 ) +
(u
(v
+
ui+ 12 ,j
2x i+ 2 ,j
4y i+1,j+ 2
Re x2
y 2




1
1
1
1
3
3
1
+
u

ui+ 2 ,j +
u

ui 21 ,j
4x i+ 2 ,j Rex2
4x i 2 ,j Rex2




1
1
1
1
1)
1
1 + v
+
+
(vi+1,j+ 21 + vi,j+ 12 )
u
(v
ui+ 12 ,j1
i+ 2 ,j+1
i,j 2
4y
Rey 2
4y i+1,j 2
Rey 2
X
= a(u, v)i+ 21 ,j ui+ 12 ,j +
a(u, v)nb unb
nb

The last line summarizes the expressions, where the sum over nb indicates a sum over the nearby nodes.
The expression for Bi,j+ 12 can in a similar way be written
Bi,j+ 21 =
=

(uv)i+ 1 ,j+ 1 (uv)i 1 ,j+ 1


2

x
1 vi+1,j+ 12 2vi,j+ 12

R
x2

2
2
vi,j+1
vi,j
y
+ vi1,j+ 12
1 vi,j+ 32 2vi,j+ 12 + vi,j 12

R
y 2
2

= {expand and interpolate using grid values}


= b(u, v)i,j+ 12 vi,j+ 12 +

X
nb

b(u, v)nb vnb

where the expressions for b(u, v)i,j+ 12 and b(u, v)nb are analogous to a(u, v)i+ 12 ,j and a(u, v)nb , and
therefore not explicitly written out. The expression originating from the continuity equation is
Di,j =

ui+ 12 ,j ui 12 ,j
x

vi,j+ 21 vi,j 12
y

These equations, including the boundary conditions, can be assembled into matrix form. We have

u
A(u, v)
0
Gx
u
fu
d
v +
0
B(u, v) Gy v = fv
dt
0
Dx
Dy
0
p
0

Here we we have used the definitions


u
v
p
A(u, v)
B(u, v)
Gx
Gy
Dx
Dy
f

vector of unknown streamwise velocities


vector of unknown normal velocities
vector of pressure unknowns
non-linear algebraic operator from advective and viscous terms of u-eq.
non-linear algebraic operator from advective and viscous terms of v-eq.
linear algebraic operator from stremwise pressure gradient
linear algebraic operator from normal pressure gradient
linear algebraic operator from x-part of divergence constraint
linear algebraic operator from y-part of divergence constraint
vector of source terms from BC

With obvious changes in notation, we can write the system of equations in an even more compact form
as
d
dt

u
0

N (u) G
D
0



u
p

f
0

If this is compared to the finite element discretization in the next chapter, it is seen that the form of
the discretized matrix equations are the same.

3.4

Time dependent flows

The projection method


For time-dependent flow a common way to solve the discrete equations is the projection or pressure correction method. Sometimes this method and its generalizations are also called splitting methods. We illustrate
the method using the system written in the compact notation defined above.
We start to discretize the time derivative with a first order explicit method
n+1
u
un

+ N (un ) un + G pn+1 = f (tn )


t

Dun+1 = 0

Then we make a prediction of the velocity field at the next time step u not satisfying continuity, and
then a correction involving p such that Dun+1 = 0. We have

u un

+ N (un ) un = f (tn )

un+1 u
+ G pn+1 = 0

Dun+1 = 0
Next, apply the divergence operator D to the correction step

1
D u
t
Here DG represents divgrad = Laplacian, i.e. discrete version of pressure Poisson equation. The
velocity at the next time level thus becomes
DG pn+1 =

un+1 = u tG pn+1
Note here that un+1 is projection of u on divergence free space. This is a discrete version of the
projection of a function on a divergence free space discussed in the section about the role of the pressure
in incompressible flow.
In order to be able to discuss the boundary condition on the pressure Poisson equation we write the
prediction and correction step more explicitly. Using the staggered grid discretization the prediction step
can be written

u 1 uni+ 1 ,j

2
i+ 2 ,j
+ Ani+ 1 ,j = 0
2
t
n

vi,j+
1 v

i,j+ 21

n
2

+ Bi,j+ 1 = 0
2
t

and the projection step as

n+1
n+1
ui+ 1 ,j ui+ 1 ,j

pn+1

i+1,j pi,j

2
2

+
=0
t
x
n+1

n+1
n+1

pi,j+1 pi,j
vi,j+ 12 vi,j+ 12

+
=0
t
y

n+1
n+1
and vi,j+
and
By solving for un+1
1 above and using this, and the corresponding expressions for u
i+ 1 ,j
i 1 ,j
2

n+1
vi,j
1 , in the expression for the divergence constraint D i,j we find the following explicit expression for the
2
discrete pressure Poisson equation
n+1
n+1
n+1
n+1

pn+1
pn+1
Di,j
i,j+1 2pi,j + pi,j1
i+1,j 2pi,j + pi1,j
+
=
x2
y 2
t

where Di,j
is Di,j evaluated using the discrete velocities at the prediction step, u and v .
When the discrete pressure Poisson equation is solved we need values of the unknowns which are outside
of the domain, i.e. we need additional boundary conditions. If we take the vector equation for the correction
or projection step and project it normal to the solid boundary in figure 3.5 we have
n+1

pn+1
1  n+1
2,1 p2,0
=
v2, 1 v
2
y
t

where v = v2,
1 is an unknown value of the velocity in the prediction step at the boundary. In general
2
it is important to choose this boundary condition such that discrete velocity after the correction step is
divergence at the boundary, see the discussion in the end of the section on the role of the pressure. However,
for this particular discretization it turns out that we can choose v arbitrarily since it completely decouples
from the rest of the discrete problem. To see this we write the pressure Poisson equation, centered abound
the point (2,1), close to the boundary. We find
n+1
n+1
n+1
n+1
pn+1
pn+1
1
3,1 2p2,1 + p1,1
2,2 2p2,1 + p2,0
+
=
2
2
x
y
t

u5 ,1 u3 ,1
2

v2,
3 v
2

n+1
if we substitute the value of pn+1
2,1 p2,0 from the boundary condition obtained for the pressure into the

above equation we can se that v cancels in the right hand side of the pressure Poisson equation. Since the
discretization of the time derivative is explicit there is no other place where the values of v or u on the
boundaries enter. Thus we find that we can choose the value of v arbitrarily. In particular, we can choose
n+1
v = v2,
1 , in which case we obtain a Neumann boundary condition for the pressure. Note, however, that
2
this is a numerical artifact and does not imply that the real pressure gradient is zero at the boundary.
A number of other splitting methods have been devised for schemes with higher order accuracy for the
time discretization. In general it is preferred to make the time discretization first for the full Navier-Stokes
equations and afterward split the discrete equation into one predictor and one corrector step. See Peyret
and Taylor for details.

Time step restriction


For stability it is necessary that prediction step is stable, numerical evidence indicates that this is sufficient.
In the predictor case the momentum equations decouple if we linearize them abound the solution U 0 , V0 .
It thus suffices to consider the equation
u
u
u
1 2
+ U0
+ V0
=
u
t
x
y
Re
For simplicity we will here consider one-dimensional version
u
1 2u
u
+ U0
=
t
x
Re x2
which in discretized form can be written


unj+1 unj1
1 unj+1 2unj + unj1
n+1
n
uj = uj + t U0
+
2x
Re
x2
where we have chosen to call the discrete points where the u-velocity is evaluate for x j .
We introduce
unj = Qn eixj
into the discrete equation, together with the definitions
U0 t
2t
,
=
,
= x
x
Re x2
This implies that the amplification factor Q can be written
 

2
Q = 1 i sin () 2 sin
2
=

We find that the absolute value of Q satisfies

where s = sin2



|Q| = 1 4 2 2 s2 + 4 2 s2 s



= 1 4s 2 2 s 2 + 1

 

2 . This implies that


(s) = 2 2 s + 2 0

s : 0 s 1

(s) is linear function of s and thus the the limiting condition on (s) is given by its values at the two
end points of the s-interval. We find
(0) = 2 0,

(1) = 2 0

which implies that

2 1

substituting back the definitions of and we find the following stability conditions

1 U02 tRe 1
2
2t

1
Rex2

For the two dimensional version of the predictor step Peyret and Taylor [8] find the following stability
limits

1 U02 + V02 t Re 1
4
4t

1
Rex2
Note that this is a stronger condition than in the one-dimensional case.

3.5

General iteration methods for steady flows

Distributive iteration
For steady flows we can still use the methods introduced for unsteady flows, and iterate the solution in time
to obtain a steady state. However, in many cases it is more effective to use other iteration methods. We will
here introduce a general distributive iteration method and apply it to the discretized steady Navier-Stokes
equations.
For the steady case the discretized equations can be written in the form


 

N (u) G
u
f
=
D
0
p
0
which is of the form A y = b. Now, let y = B y which implies that
AB y = b
and use the iteration scheme on this modified equation. We divide the matrix AB in the following
manner
AB = M T

M y = T y + b

and use this to define the iteration scheme


M B 1 y k+1 = T B 1 y k + b = M B 1 y k + b A y k
which can be simplified to

y k+1 = y k + BM 1 b A y k
| {z }
rk

where rk is the residual error in the k:th step of the iteration procedure.
If B = I we obtain a regular iteration method where residual r k is used to update y k by inverting M ,
a simplified part of A, as in the Gauss-Seidell method for example. For this method we have
Ax = b
A=LU
(L U ) x = b
L xn+1 = U xn + b

Application to the steady Navier-Stokes equations


When the distributive iteration method is applied to the steady Naiver-Stokes we have


N (u) G
A=
D
0
we can obtain AB in the following block triangular form


N (u)
0
AB =
D
DN (u)1 G
if the matrix B has the form
B=

I
0

N (u)1 G
I

I
0

(u)1 G
N
I

1 is a simple approximation of the inverse of the discrete Navier-Stokes operator. We now


where N(u)
split AB = M T where


Q 0
M=
D R
and where Q is an approximation of N (u) and R is an approximation of DN 1 G. Note that R is a
discrete Laplace like operator.

The SIMPLE (Semi-Implicit Method for Pressure Linked Equations) by Pantankar and Spalding (1972)
can be thought of as a distributive iteration method. Wesseling [10] indicates that by choosing
(u)1
N

= [diagN (uk )]1

Q = N (uk )
R = D[diagN (uk )]1 G
we have a method which is essentially the original version of the SIMPLE method. Thus we have the
following iteration steps
1. First, we calculate the residual
b A yk =

f
0

N (uk ) G
D
0



uk
pk

r1k
r2k

2. Second, we calculate M 1 rk which implies


Q
u = r1k

u = Q1 r1k

R p = r2k D
u

3. Third, we have the distribution step BM 1 rk


which becomes



 

1 G p
u

uN
=B
=
p
p
p
4. Fourth, there is an under relaxation step, where the velocities and the pressure is updated
 k+1
u
= uk + u u
k+1
p
= pk + p p
The under relaxation is needed for the method to converge, thus u and p is usually substantially
lower that one.

dx
dy
dx
dl = ( dx, dy)
n dl = ( dy, dx)
i 1, j + 1
i, j + 1
i + 1, j + 1
c
c0
b
b0
i 1, j
i + 1, j
i, j
Aabcd
d
d0
a
The finite-element method (FEM) approximates
an integral form of the governing equations. However,
0
a
the integral formcalled the variational formis
usually different from the one used for the finite-volume
i 1,forms
j 1 may be used for the same equation depending on circumstances
method. Moreover, different integral
j1
such as boundary conditions. We i,introduce
FEM first for a scalar advectiondiffusion problem before
i
+
1,
j
discussing the NavierStokes equations. 1
j
i
i+1
4.1 FEM for an advectiondiffusion
problem
ui,j
vi,j
We consider the steady flow of an incompressible
fluid in a bounded domain , and we assume that its
pi,j numerical solution, for instance). We assume that the boundary
velocity field Uj is known (from a previous
pi,j two portions 0 and 1 , as in figure 4.1 for instance. We are
of the domain can be decomposed into
vi,j+1/2
interested to compute the temperature
field u in the domain, modeled by the advectiondiffusion problem
vi,j1/2
u
ui1/2,j
in ,
(4.1a)
Uj 1 2 u = f
i x
2 j

Chapter 4

Finite element methods for


incompressible flow

u=0
on 0 ,
(4.1b)
i
i + 12
u
=0
on 1 .
(4.1c)
i+1
n
3
i+ 2
U
+ 12
The coefficient > 0 is the thermal jdiffusivity.
Incompressibility yields that xjj = 0. We assume that
1 is a solid wall, so that nj Uj = 0 on 11j. This simplifies the analysis but is not essential. The Dirichlet
j
u
boundary condition u = 0 on 0 is called 2an isothermal condition, whereas the Neumann condition n
=0
j

1
on 1 is an adiabatic condition.
A to form the finite-element discretization requires some vector
Deriving the integral form that is used
B yields
calculus. The product rule of differentiation
C

u3 
2 ,1 u
v u
v1, 32 v
=
+ v2 u.
(4.2)
xi
xi
xi xi
p1,1
0
n
1
0
2
i
j

Inflow
Solid wall
1

Figure 4.1: A typical domain associated with the advectiondiffusion problem (4.1).
71

Recall the divergence theorem

wi
d =
xi

ni wi d.

(4.3)

Integrating expression (4.2) and applying the divergence theorem on the left side yields
Z
Z
Z
u
v u
v
d =
d + v2 u d.
n
xi xi

(4.4)

(Note that
u
u
= ni
.)
n
xi
Let v be any smooth function such that v = 0 on 0 . Multiplying equation (4.1a) with v, integrating
over , and utilizing expression (4.4) yields
Z

f v d =

u
vUj
d
xj

v u d =

u
vUj
d +
xj

v u
d
xj xj

0

u
v
d,
n

where the last term vanishes on 0 due to the requirements on v, and on 1 due to boundary condition (4.1c)
We have thus shown that a smooth (twice continuously differentiable) solution to problem (4.1) (also
called a classical solution) satisfies the varational form
Z
Z
Z
v u
u
d +
d = f v d
(4.5)
vUj
xj
xj xj

for each smooth function v vanishing on 0


We will now forget about the differential equation (4.1), and directly consider the variational form (4.5).
For this, we need to introduce the function space

v v
V = v:
d < +, v = 0 on 0 .

xi xi

The space V is a linear space, that is, u, v V u + v V , R. The requirement that the
gradient-square of each function in V is bounded should be intepreted as a requirement that the energy is
bounded.
The variational problem will thus be
Find u V such that
Z
Z
Z
u
v u
vUj
d +
d = f v d
xj
xj xj

v V.

(4.6)

A solution to variational problem (4.6) is called a weak solution to advectiondiffusion problem (4.1). The
derivation leading up to expression (4.5) shows that a classical solution is a weak solution. However, a weak
solution may not be a classical solution since functions in V may fail to be twice continuously differentiable.
Note that the boundary conditions are incorporated into the variational problem. The Dirichlet boundary condition on 0 is explicitly included in the definition of V and is therefore referred to as an essential
boundary condition. The Neumann condition on 1 is defined implicitly through the variational form and
is therefore called a natural boundary condition.

4.1.1

Finite element approximation

Let us triangulate the domain by dividing it up into triangles, as in figure 4.2. Let M be the number of
triangles and N the number of nodes in 1 , that is, the nodes on 0 are excluded in the count. Let h
be the largest side of any triangle.
Let Vh be the space of all functions that are
- continuous on ,

p1,1
0
1
2
i
j
Inflow
Solid wall
n
0
1

u 23 ,1
v1, 23
p1,1
0
1
2
i
j
Inflow
Solid wall
n
0
1

Figure 4.2: A triangulation. Here, the number N are Figure 4.3: The tent or hat basis function associated with continuous, piece-wise linear functions.
the black nodes.
- linear on each triangle,
- vanishing on 0 .
Note that Vh V . We get a finite-element approximation of our advection-diffusion problem by simply
replacing V by Vh in variational problem (4.6), that is:
Find uh Vh such that
Z
Z
Z
vh uh
uh
v h Uj
d +
d = f vh d
xj
xj xj

(4.7)

vh Vh .

Restricting the function space in a variational form to a subspace is called a Galerkin approximation. The
finite-element method is a Galerkin approximation in which the subspace is given by piecewise polynomials.

4.1.2

The algebraic problem. Assembly.

Once the value of a function uh in Vh is known at all node points, it is easy to reconstruct it by linear
interpolation. This interpolation can be written as a sum of the tent basis functions (n) (figure 4.3),
which are functions in Vh that are zero at each node point except node n, where it is unity. Thus,
uh (xi ) =

N
X

u(n) (n) (xi ) ,

(4.8)

n=1

where u(n) denotes the value of uh at node n. Note that expansion (4.8) forces uh = 0 on 0 since it
excludes the nodes on 0 in the sum.
Substituting expansion (4.8) into FEM approximation (4.7) yields
N
X

n=1

(n)

Z
Z
N
X
vh (n)
(n)
(n)
v h Uj
d +
u
d = f vh d
xj
xj xj
n=1

v h Vh .

(4.9)

Since each basis function is in Vh , we may choose vh = (m) Vh for m = 1, . . . , N in equation (4.9), which
then becomes
N
X

n=1

u(n)

(m) Uj

Z
Z
N
X
(n)
(m) (n)
d +
u(n)
d = f (m) d, m = 1, . . . , N .
xj
xj xj
n=1

(4.10)

This is a linear system system Au = b, where A = C + K and


Kmn =

(m) (n)
d,
xj xj

Cmn =

(m) Uj

(n)
d,
xj

(4.11)

i
j
Inflow
Solid wall
n
0
1

M2

;:9;:9 ;:9;:9 ;:9;:9 ;:9;:9 ;:9;:9 ;:9;:9 ;:9;:9 ;9;9 =:<=:< =:<=:< =:<=:< =:<=:< =:<=:< =:<=:< =:<=:< =<=<
;9:;9:;:9;:9 ;:9;:9 ;:9;:9 ;:9;:9 ;:9;:9 ;:9;:9 ;9;9 =:<=:< =:<=:< =:<=:< =:<=:< =:<=:< =:<=:< =:<=:< =<=<
;9:;:9 ;:9;:9 ;:9;:9 ;:9;:9 ;:9;:9 ;:9;:9 ;:9;:9 ;9;9 =:<=:< =:<=:< =:<=:< =:<=:< =:<=:< =:<=:< =:<=:< =<=<
;9:;9:;:9;:9 ;:9;:9 ;:9;:9 ;:9;:9m;:9;:9 ;:9;:9 ;9;9 =:<=:< =:<=:< m=:<=:< =:<=:< +=:<=:< 2=:<=:< =:<=:< =<=<
1

h
Figure 4.4: An example mesh that gives a particular simple stiffness matrix K
R

u(1)

u = ...
u(n)

b=
R

f (1) d
..
.
f (N ) d

The union of all triangles constitutes the domain and that the triangles do not overlap. Thus, any
integral over can be written as a sum of integrals over each triangle,

Kmn =

M Z
M
X
X
(m) (n)
(m) (n)
(p)
d =
d =
Kmn
.
xj xj
xj xj
p=1
p=1
Tp

(p)

(p)

Note that Kmn vanishes as soon as m or n is not a corner node of triangle p. Thus Kmn contributes to the
sum only when m and n both are corner nodes of triangle p. Matrix K can thus be computed by looping
(p)
all M triangles, calculating the element contribution Kmn for m, n being the three nodes of triangle p and
adding these 9 contribution to matrix K. Note that the derivatives of (n) are constant on each triangle,
(p)
so Kmn is easy to compute exactly. Matrix C and vector b can also be computed by element assembly,
but numerical quadrature is usually needed, except if Uj and f happen to be particularly easy functions
(such as constants).

4.1.3

An example

As an example, we will compute the elements of K, the stiffness matrix (a term borrowed from structural
mechanics), for the particular case of the structured mesh depicted in figure 4.4.
Let M be the number of internal nodes in the horizontal as well as the vertical direction (the solid
nodes in figure 4.4.) The width of each panel is then h = 1/(M + 1), and the area of each triangle is
|Tk | = h2 /2. The components of K are
Kmn =

(m) (n)
d
xj xj

(4.12)

Note that Kmn vanishes for most m and n. For instance, Km,m+2 = 0, since basis functions (m) and
(m+2) do not overlap (figure 4.4). In fact, Kmn 6= 0 only when m and n are associated with nodes that
are nearest neighbors.
To compute matrix element (4.12), we need expressions for the gradients of basis functions (m) and
(m+1)

. Since the functions are piecewise linear, the gradient is constant on each triangles and may thus
easily be computed by finite-differences in the coordinate direction (figure 4.5). By the expressions in

m
2
m+2
T3
i
on T1 ,
1
j
on T2 ,
h
T4
Inflow
T2
on T3 ,
Solid wall
(m)
on T4 ,
n
T1
T5
on T5 ,
0
1
on T6 ,
T6

on all other triangles.


2
M
(m)
M being the (black) middle node.
Figure 4.5: The gradient of test function
for m
m
m
+
2


1 1
(m)
1

,
on
T
,
1
h h
h
(m) =
h1 , 0
on T2 ,
T2
(m+1)

0
on all other triangles
T1
(

1
,
0
on
T
,
1
h

(m+1) =
1
1
,

on
T
.
2
h
h
1 

h, 0 

1
1

h , h

0, h 
=
h1 , 0

h1 , h1

0, h1

AB @?@?B(
AB @?@?B(
AB @?@?B(
AB @?@?B(
AB @?@ B(
AB B(
AA(
AA(
@?@?@?@?@?@?B(
B(
BAAB
BB(
BB(
AB(
AB(
AB(
AB(
AB(
AB(
A
A
B(
B(
B(
B(
B(
B(
BABA
>?>?@?@?>?@?@?>?@?@?>?@?@?>?@?@?>?@?@?A(
?A @?@?>?@?@?A(
>AB(
?A @?@?>?@?@?A(
>AB(
?A @?@?>?@?@?A(
>AB(
?A @?@?>?@?@?A(
>AB(
?A @?@?> @@ A(
>AB(
AB(
AB(
B(
B(
AB(
AB(
AB(
BABA
>?>?@?>?@?>?@?>?@?>?@?>?@?B(
?A @?>?@?B(
>B(
?A @?>?@?B(
>B(
?A @?>?@?B(
>B(
?A @?>?@?B(
>B(
?A @?> @ AB(
>B(
A
A
A
A
A
A
A
A
AB(
AB(
AB(
B(
B(
BABA
>?>?@?>?@?>?@?>?@?>?@?>?@?B(
?B @?>?@?B(
>A(
?B @?>?@?B(
>A(
?B @?>?@?B(
>A(
?B @?>?@?B(
>A(
?B @?> @ B(
>A(
A
A
A
A
A
A
A
A
(
A
(
A
ABBA
BB(
B
B
>?>?@?>?@?>?@?>?@?>?@?>?@?B(
?A @?>?@?B(
>AB(
?A @?>?@?B(
>AB(
?A @?>?@?B(
>AB(
?A @?>?@?B(
>AB(
?A @?> @ A(
>AB(
AB(
AB(
B(
B(
AB(
A
A
BABA
>?>?@?>?@?>?@?>?@?>?@?>?@?B(
?A @?>?@?B(
>AB(
?A @?>?@?B(
>AB(
?A @?>?@?B(
>AB(
?A @?>?@?B(
>AB(
?A @?> @ AB(
>AB(
AB(
AB(
B(
B(
AB(
AB(
AB(
BABA
>?>?@?>?@?>?@?>?@?>?@?>?@?B(
?A @?>?@?B(
>B(
?A @?>?@?B(
>B(
?A @?>?@?B(
>B(
?A @?>?@?B(
>B(
?A @?> @ AB(
>B(
A
A
A
A
A
A
A
A
AA(
A
A
(
B
(
B
BA
>?>?>?>?>?>?>?A(
>B ?>?A(
>B ?>?A(
>B ?>?A(
>B ?>?A(
>B ?> B(
(
A
(
A
>?>?>?>?>?>?>?>?>?>?>?>?>?>?>?>?> B B B AB

Figure 4.6: The support for basis functions m and m+1 are marked with vertical and horizontal stripes
respectively.
figure 4.5, we have
Kmm =

X
(m) (m)
d =
xj xj

Z "

k=1T
k

(m)
x

2

(m)
y

2 #

1
1
1
1
1
1
|T1 | + 2 2 |T2 | + 2 |T3 | + 4 |T4 | + 2 2 |T5 | + 2 |T6 |
2
h
h
h
h
h
h
1 h2
=8 2
=4
h 2
=

The common support for basis functions m and m+1 is only the two triangles marked in figure 4.6;
that is, it is only on these two triangles that both functions are nonzero at the same time. Utilizing the
gradient expressions given in figure 4.6 yields
Z
2 Z
X
1
1
2 h2
Km,m+1 =
(m) (m+1) d =
(m) (m+1) d = 2 |T1 | 2 |T2 | = 2
= 1,
h
h
h 2

k=1T

and in the same manner,


Km,m1 = Km,m+M = Km,mM = 1.

Altogether, K becomes the block-tridiagonal M 2 -by-M 2 matrix

T I
I T I

..
..
..
K =
.
.
.

I T I
I T

where I is the M -byM identity matrix, and T the M -byM matrix

4 1
1 4 1

..
..
..
T =
.
.
.

1 4 1
1 4
Thus, a typical row in the matrixvector product Ku becomes

4um um+1 um1 um+M umM ,

(4.13)

that is, after dividing expression (4.13) with h2 , we recognize the classical five-point stencil for the negative
Laplacian 2 .
Thus our finite-element discretization is for this particular mesh equivalent to a finite-volume or finitedifference discretization. However, for a general unstructured mesh, the finite-element discretization will
not give rise to any obvious finite-difference stencil.

4.1.4

Matrix properties and solvability

Matrix A is sparse: the elements of martix A are mostly zeros. At row m, matrix element A mn 6= 0 only
in the columns where n represents a nearest neighbor to node m. The number of nonzero elements on
each row does not grow (on average) when the mesh is refined (since the number of nearest neighbors in a
mesh does not grow). Thus, the matrices become sparser and sparser as the mesh is refined. A common
technique in finite-element codes is therefore to use sparse representation, that is, to store only the nonzero
elements and pointers to matrix locations.
It is immediate from expression (4.11) that Kmn = Knm , that is, K is a symmetric matrix (K T = K).
Integration by parts also shows that matric C is skew symmetric (C T = C).
Theorem 1. K is positive definite, that is, v T Kv > 0 for each v 6= 0.
PN
Proof. Let vh Vh . Then vh = n=1 v (N ) (n) . Denote v = (v (1) , . . . , v (N ) )T .
v T Kv =

N X
N
X

v (n)

m=1 n=1

(n) (m)
d v (m)
xi xi

Z X
Z
N
N
 (m) (m)  X  (m) (n) 
vh vh
=
v
v
d =
d 0,
xi
m=1 xi
xi xi
n=1
with equality if and only if

vh
xi

= 0, that is, if vh = Const. But since vh |0 = 0, vh 0.

From Theorem 1 also follows that A is positive definite. To see this, note that the skew-symmetry of C
yields that
T
v T Cv = v T Cv = v T C T v = v T Cv
that is, v T Cv = 0 for each v. Thus

v T Av = v T (K + C) v = v T Kv + v T Cv = v T Kv > 0
for each v 6= 0, that is, A is positive definite. However, A is not symmetric whenever U j 6= 0.
Recall that a linear system Au = b has a unique solution for each b if the matrix is nonsingular. A
positive-definite matrix is a particular example of a nonsingular matrix. We thus conclude that the linear
system resulting from the finite-element discretization (4.7) has a unique solution.

4.1.5

Stability and accuracy

We have defined the finite-element approximation and showed that the finite-element approximation yields
a linear system Au = b that has a unique solution. Now the question is whether the approximation is any
good. We will study the stability and accuracy of the numerical solution. FEM usually uses integral norms
for analysis of stability and accuracy. We start by reviewing some mathematical concepts that will be used
in the analysis.
Basic definitions and inequalities
The most basic integral norm is
kvkL2 () =

Z

v 2 d

1/2

For the current problem, a more fundamental quantity is the energy norm
Z
1/2 Z
1/2
v v
2
kvkV =
|v| d
=
d

xi xi
and associated inner product
(u, v)V =

u v d =

u v
d.
xi xi

(4.14)

The CauchySchwarz inequality for vectors a = (a1 , . . . , an ) is


|a b| |a||b| = |a a|1/2 |b b|1/2 ,
whereas for functions it reads
Z
Z
1/2 Z
1/2


2
2
f g d
f d
g d
,

which can be denoted

(4.15)

(f, g)L2 () kf kL2 () kgkL2 () .

The CauchySchwarz inequality (4.15) implies that, for each nonzero, square-integrable g,
R

Z
1/2
f g d
2

f d
R
1/2 .
2

g d

Choosing g = f yields equality in above expression. Thus


R

R

Z
1/2
f g d
f g d
2

kf kL2 () =
f d
= max R
1/2 = max
g6=0
g6=0 kgkL2 ()
2

g d

The rightmost expression is a variational characterization of the L2 () norm, that is, it expresses the
norm in terms of something that is optimized. By inserting derivatives in the nominator and/or denominator
of the variational characterization, various exotic norms can be defined. A norm, different from the L 2 ()
norm and of importance for the NavierStokes equations is

Z 



p wi d


xi
(4.16)
kpkexotic = max Z
1/2 .
wi

|wi |2 d

This norm will be of crucial importance when we analyze the stability of FEM form the NavierStokes
equations.
The Poincare inequality relates the energy and L2 () norms:
Theorem 2 (Poincar
e inequality). There is a C > 0 such that
Z
Z
v v
2
v d C
d,
x
i xi

that is,
kvk2L2 () Ckvk2V
for each v V .
Remark 1. Theorem 2 says that the energy norm is stronger than then L2 () norm. The smallest possible
value of C grows with the size of . The Poincar
R e inequality holds only if is bounded in at least one
direction. Also, the inequality does not hold if 0 d = 0. However, the inequality
Z

Z
Z
v v
2
2
v d C
v d +
d

xi xi
R
holds also when 0 d = 0.

Stability

Choosing vh = uh in equation (4.7) yields


Z
Z
Z
uh
uh uh
d +
d = f uh d.
uh U j
xj
xj xj

Note that

uh U j

uh
d = uT Cu = 0,
xj

(4.17)

m
m+2
1
h

V
Vh
uh

Figure 4.7: An orthogonal projection on a line. Note that the vector u uh is the shortest of all vectors
u vh for vh Vh , that is, ku uh k ku vh k vh Vh .
since C = C T (using the notation of 4.1.4).
Expression (4.17) thus becomes
Z
Z
uh uh
d =
f uh d

xi xi

|
{z
}
kuh k2V

(CauchySchwarz)

Z

(Poincare) C

f d

Z

1/2 Z

f d

u2h

1/2 Z

1/2

uh uh
d
xi xi

Ckf kL2 () kuh kV .

1/2

Dividing with kuh kV shows that

C
kf kL2()

where the constant C does not depend on h. That is, we have shown that uh cannot blow up as the
discretization is refined, which is the same as saying that the numerical solution is stable.
kuh kV

Error bounds
Choosing v = vh Vh V in variational problem (4.6) and subtracting equation (4.7) from(4.6) yields the
Galerkin orthogonality,
Z
Z
vh

(u uh ) d +
v h Uj
(u uh ) d = 0
vh Vh .
x
x
x
i
i
j

This is a true orthogonality condition when Uj 0:


Z
vh

(u uh ) d = 0
xi xi

vh Vh ;

(4.18)

that is, the error u uh is orthogonal to each vh Vh with respect to the inner product (4.14). Cf. two
orthogonal vectors: a b = 0 . Orthogonality relation (4.18) means that u h is an orthogonal projection
of u on Vh . As illustrated in figure 4.7, the orthogonal projection on a subspace yields the vector that is
closest to the given element in the norm derived from the inner product in which orthogonality is defined.
Orthogonality property (4.18) (cf. figure 4.7) implies that the FE approximation is optimal in the energy
norm when Ui 0:
ku uh kV ku vh kV
vh Vh .
(4.19)
The matrix A is symmetric in this case (since C = 0).
When Uj 6= 0, the FEM approximation is no longer optimal in this sense. However, there is a C > 0
such that Ceas lemma holds,
ku uh kV

C
ku vh kV

vh Vh .

(4.20)

Matrix A is nonsymmetric in this case. Note that the approximation may be far from optimal when is
small, as the next example illustrates.

M
M0 0.8
1 0.7
m
m + 2 0.6
i 0.5
j
h 0.4
Inflow
V
Solid wall
Vh 0.3
n
u 0.2

uh0 0.1
1 0
0.9
0.92
0.94
0.96
0.98
1

x
2
M
M
Figure 4.8: Under-resolved standard
Galerkin finite-element approximations may yield oscillations for
m
advection-dominated flows.
m+2
v1, 32
1

P t =vh1,
0 32
V
Vh
u
uh
Figure 4.9: An interpolant h v interpolates v at the nodal points.
Example 1. Let us consider the above FEM applied to the 1D problem
1 00
u + u0 = 1
500
u(0) = u(1) = 0

in (0, 1).

(4.21)

The problem is advection dominated, that is, the viscosity coefficient in equation (4.21) is quite small:
= 1/500. A boundary layer with sharp gradients forms close to x = 1. The solid line in figure 4.8 shows
the FE solution h = 1/1000, whereas the dashed line shows the FE approximation for h = 1/200.

The result of example 1 is typical: under-resolved standard Galerkin methods may generate oscillations
for advection-dominated flows in areas of sharp gradients.
An explanation of this phenomenon is that the directions of flow is not accounted for in the standard
Galerkin approximation. Galerkin approximations correspond to finite-difference central schemes. Upwinding is used to prevent oscillations for finite-difference methods. Various methods exist to obtain similar
effects in FEM, for instance streamline diffusion and discontinuous Galerkin. Both these methods modify
the basic Galerkin idea to account for flow-direction effects.
Approximation and mesh quality
Estimates (4.19) and (4.20) are independent of the choice of Vh ! We only used that Vh V . Next step
in assessing the error is an approximation problem: How well can functions in V h approximate those in
V ? This step is independent of the equation in question. Interpolants are used to study approximation
properties. The interpolant is the function h v Vh that interpolates the function v V at the node
points of the triangulation, as illustrated for a 1D case in figure 4.9.
Approximation theory yields that, for sufficiently smooth v,
kv h vkL2 () Ch2
kv h vkV Ch

second order error


first order error

(4.22)

The constant C contains integrals of the second derivatives of v.


A mesh quality assumption is needed for (4.22) to hold. The essential point is: beware of thin triangles;
do not let the shape deteriorate as the mesh is refined. A strategy to preserve mesh quality when refining
is to subdivide each triangle into four new triangles by joining the edge midpoints (figure 4.10).
Each of the conditions below is sufficient for approximation property (4.22) to hold:

M
m
m+2
1
h
V
Vh
u
uh

1
2
i
j
Inflow
Solid wall
n
0
1

Figure 4.10: A subdivision that preserves mesh


quality.
M2
M
m
(i) (Maximum angle condition). The largest anm+2
gle of any triangle should not approach 180 as
1
h 0.
h
(ii) (Chunkiness parameter condition).
The
V
quotient between the diameter of the largest
Vh
circle that can be inscribed and the diameter
u
of the triangle should not vanish as h 0.
uh
Condition (ii) implies (i), but not the reverse.
Accuracy of the FE solution
Error bounds (4.19) and (4.20) together with approximation property (4.22) implies
ku uh kV Ch
Further analysis (not provided here) yields
ku uh kL2 () Ch2 .
This second-order convergence rate requires a mesh quality assumption and a sufficiently smooth solution u
(since the constant C contains second derivatives of u). The convergence rate is of optimal order : it agrees
with approximation property (4.22) and is the best order that in general can be obtained for, in this case,
piecewise-linear functions. However, the solution may still be bad for a particular, too-crude mesh; recall
example 1.
Improving Accuracy
Accuracy is improved by refining the mesh (h-method). This may be done adaptively, where it is needed
(say, in areas of large gradients), to prevent the matrices to become too large. Automatic methods for
adaptation are common.
An alternative is to keep the mesh fixed and increase the order of the polynomials at each triangle
(p-method). For instance, continuous, piecewise quadratics yields a third-order-accurate solution (in the
L2 () norm).
These strategies may be combined in the hp method, where the mesh is refined in certain regions
and the order of approximations is increased in other.

4.1.6

Alternative Elements, 3D

A quadrilateral is a skewed rectangle: four points connected by straight lines to form a closed geometric
object. Nonoverlapping quadrilaterals can be used to triangulate a domain, as in figure 4.11.
An example of a finite-element space Vh on quadrilaterals is globally continuous functions varying linearly
along the edges. As in the triangular case, the functions are defined uniquely by specifying the function
values at each node. The functions are defined by interpolation into the interior of each quadrilateral. This
space reduces to piecewise bilinear functions for rectangles with edges in the coordinate directions:
vh (x, y) = a + b x + c y + d xy
Quadrilaterals on a logically rectangular domain (a distorted rectangle, as in figure 4.11) yields a regular
structure to the matrix A. This may

p1,1
p1,1

2
M
0
0
M
1
1
m
2
2
m+2
i
i
j
j
1
h
Inflow
Inflow
V
Solid wall
Solid wall
Vh
n
n
0
0
u
uh
1
1

with quadrilaterals.
M 2 Figure 4.11: A logically rectangular domain triangulated
M2
M
M
m
m
m+2
m+2
1
1
h
h
V
V
Vh
Vh
u
u
uh
uh
Figure 4.12: In 3D, triangular and quadrilateral elements generalize to tetrahedral (left) and hexahedral
(right) meshes.
give high accuracy (particularly if the mesh is aligned with the flow), and
allow efficient solution of the linear system (particularly for uniform, rectangular meshes).
Compared to triangular mesh, it is harder to generate quadrilateral meshes automatically for complicated geometries.
Triangular and quadrilateral meshes generalize in 3D to tetrahedral and hexahedral meshes (figure 4.12).
Advantages and limitations with tetrahedral and hexahedral meshes are similar to corresponding meshes
in 2D.

1
h
V
Vh0
u
uh

D:C D:C D:C D:C D:C D:C D:C 1D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C 0 D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C D:C DC
2
i
j

Inflow
Solid wall
0
n
0
1

Figure
4.13: A simple example of a domain.
M2
0
M
m
m+2
1
h
V
0

Vh 0
u
uh
0
1

F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E FEFE
FE:FE:F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E FEFE
FE:F:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E FEFE
FE:FE:F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E F:EF:E FEFE
F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E F:E FE

G G GH H
] ]^ ^
[ [\ \
Y YZ Z

I IJ J

K KL L

M MN N

_ _` `

a ab b

O OP P

e ef f

c cd d

Q QR R

W WX X

U UV V

S ST T

Figure 4.14: The black nodes are the degrees of freedom for the velocity components, which are fewer than
the number of triangles.

4.2

FEM for NavierStokes

We now turn to the NavierStokes equations for the steady flow of an incompressible fluid, and consider
a situation with flow in a bounded domain . A simple example domain is depicted in figure 4.13. The
mathematical problem in differential form is
uj

ui
p
1 2
+

ui
xj
xi
Re
ui
xi
ui
1 ui
nj + pni

Re xj

= fi

in ,

(4.23a)

=0

in .

(4.23b)

= gi

on 0

(4.23c)

=0

on 1 .

(4.23d)

The boundary condition on 1 = \ 0 is of no obvious physical significance, but is useful as an


artificial outflow condition.
The NavierStokes equations is a system of nonlinear advectiondiffusion equations together with
the incompressibility condition (4.23b). This condition introduces new complications not present in the
advectiondiffusion problem of section 4.1. Naive approximations of the NavierStokes equations are bound
to produce disappoing results, as the following example indicates.
Example 2 (the counting argument). Consider a case with only Dirichlet boundary conditionsthat is,
1 = and the mesh of figure 4.14. Assume that we use continuous, piecewise-linear approximations u h ,
vh for the x- and y-components of the velocity field. There are I number of panels in each direction and
therefore 2(I 1)2 degrees of freedom (two times the number of black dots) for the velocity vector field.
The left side of equation (4.23b), that is
uh vh
+
x
y
is here constant on each triangle. Thus, demanding equation (4.23b) to be satisfied on each triangle yields
2I 2 equations, which is more than the degrees of freedom for the velocity! Thus u h = vh = 0 is the only FE
function satisfying the incompressibility condition (4.23b)!


Following strategies can be used to tackle the problem of example 2.


(i) Use more degrees of freedom for the velocityfor instance continuous, piecewise quadratics on each
triangleto balance the large number of equations associated with the incompressibility condition.
(ii) Accept that the discrete velocities only is approximately divergence-free.
(iii) Associate the velocity with the midpoints of edges instead of the vertices. This also gives more degrees
of freedom for the velocities. (There are 3I 2 edges, but only 2I 2 vertices).
Strategies (i) and (ii) are the standard methods, whereas (iii) leads to non-conforming approximations related to finite-difference and finite-volume approximation with staggered mesh.

4.2.1

A variational form of the NavierStokes equations

To derive the variational form, we need new integration-by-part formulas. The product rule of differentiation
yields



ui
zi ui
zi
=
+ z i 2 u i
(4.24)
xj
xj
xj xj
and

p
zj
(zj p) = zi
+p
.
xj
xi
xj

(4.25)

Integrating expressions (4.24) and (4.25) and applying the the divergence theorem (4.3) on the left sides
yields
Z
Z
Z
ui
zi ui
nj z i
d =
d + zi 2 ui d
(4.26)
xj
xj xj

and

nj zj p d =

zi

p
d +
xi

zi
d,
xi

(4.27)

which are the integration-by-parts formulas needed to derive our variational form.
Now, let zi be a smooth vector-valued function on that vanishes on 0 . Multiply both sides of
equation (4.23a) with zi , integrating over , and utilizing (4.26) and (4.27), we find that
Z
Z
Z
Z
ui
p
1
zi fi d = zi uj
d + zi
d
zi 2 ui d
xj
xi
Re

Z
Z
Z
Z
Z
ui
1
zi ui
1
ui
zi
= z i uj
d +
d
nj z i
d p
d + nj zj p d (4.28)
xj
Re
xj xj
Re
xj
xi

Z
Z
Z
1
zi ui
zi
ui
= z i uj
d +
d p
d,
xj
Re
xj xj
xi

where, in the last equality, the boundary integrals vanish on 0 since zi = 0 on 0 and on 1 due to boundary
condition (4.23d). Moreover, multiplying equation (4.23b) with a smooth function q and integrating yields
Z
ui
d = 0
(4.29)
q
xi
Thus, we conclude that a classical solution (a twice continuously differentiable solution) to the Navier
Stokes equations satisfies the variational forms (4.28) and (4.29) for all smooth functions z i and q such that
zi vanishes on 0 .
Similarly as for the advectiondiffusion problem, the variational form will be used to define weak solutions, without explicit reference to the differential equations. A difference with advectiondiffusion problem
is that we here need different function spaces for the velocity components and the pressure. Also note that
ui = gi on 0 whereas zi vanishes on 0 . To avoid this last complication and simplify the exposition, we
will only consider homogeneous boundary conditions, gi = 0. Thus, the equations will be driven only by
the right-hand side fi , which, for instance, could be gravity or magnetic forcing for a electrically conducting
fluid.
The pressure space is
 Z

H = L2 () =

q|

q 2 d < +

and the space of velocity components is




Z
u u
V = u|
d < + and u = 0 on 0 .
xi xi
For the analysis in section 4.2.4, it will be convenient to work with the space V of the velocity vector
u = (u1 , u2 ). Saying that u V is equivalent to saying that that each velocity component u i V . The
energy norm for elements of V is denoted
kukV =

Z

1/2

1/2

ui ui
d
xj xj

whereas the L2 () norm is


kukL2 () =

Z

ui ui d

We assume that both boundary conditions really are active. In the case of only Dirichlet boundary
condition, that is, that 0 constitute the whole boundary, the pressure is only possible to specify uniquely
up to an additive constant.
The variational problem defining weak solutions to equation (4.23) is
Find ui V and p H such that
Z
Z
Z
Z
ui
1
zi ui
zi
z i uj
d +
d p
d = zi fi d
xj
Re
xj xj
xi

Z
ui
q
d = 0
q H
xi

zi V

This is a variational problem of mixed type: two different spaces are involved. As for the advectiondiffusion
problem, the boundary conditions are incorporated in the variational formulation. The essential boundary
conditions are here that ui = 0 on 0 and is explicitly assigned through the definition of V . The natural
boundary conditions are the boundary conditions on 1 , implicitly specified through the variational form.
Note that the natural boundary condition is different than for the advectiondiffusion problem.
Remark 2. The mathematical properties of the NavierStokes equations are complicated, and important
issues are still unresolved. In fact, there is a $ 1 000 000 prize for the person that can develop a theory
that can satisfyingly explain their behavior in three space dimensions! A short summary of what is known
is the following.
For the unsteady problem on a given, finite time interval (0, T ), and
in two space dimensions:
a unique weak solution exists for any square-integrable f and for any Re,
smooth data yield smooth solutions, so that weak solutions are classical solutions if data is
smooth enough;
in three space dimensions:
a weak solution exists for any square-integrable f and for any Re.
It is not known whether the weak solution always is unique (besides during some initial interval
(0, T ), where the size of T is unknown).
It is not known whether smooth data always yields a smooth solution, or if singularities can
evolve from smooth data.
The steady problem has at least one weak solution, in 2D as well as 3D. Note that the steady problem
may have several solutions, even in 2D.

Note that the natural boundary condition here not just is a Neumann condition, but the more complicated condition (4.23d).

4.2.2

Finite-element approximations

To obtain a FEM, we choose subspaces Vh V , Hh H of piecewise polynomials, and replace V with Vh


and H with Hh in the variational form. In 2D and in components, we obtain
Find uh , vh Vh and ph Hh such that


Z
Z
Z
Z
uh
uh
1
zh uh
zh
+ vh
d +
d ph
d = zh f1 d
z h uh
x
y
Re
xj xj
x



Z
Z
Z
Z
vh
vh
1
zh vh
zh
z h uh
+ vh
d +
d ph
d = zh f2 d
x
y
Re
xj xj
y



Z
uh
vh
+
d = 0.
qh
x
y

zh Vh ,

(4.30a)

zh Vh ,

(4.30b)
(4.30c)

We postpone for a while the question on how to choose the subspaces Vh and Hh .

4.2.3

The algebraic problem in 2D

Expanding ph in basis for Hh , we obtain


ph (x, y) =

Np
X

p(n) (n) (x, y) .

(4.31)

n=1

Likewise, expanding uh , vh in basis for Vh yields


uh (x, y) =

Nu
X

u(n) (n) (x, y) ,

vh (x, y) =

n=1

Nu
X

v (n) (n) (x, y) .

(4.32)

n=1

Now insert expansions (4.31) and (4.32) into equations (4.30). In (4.30a) and (4.30b), choose z h = (m)
for m = 1, . . . , Nu ; in (4.30c), choose qh = (m) for m = 1, . . . , Np . Then, equations (4.30) become
Nu
X

(n)

n=1

Nu
X

v (n)

n=1

n=1

Np
X

u(n)

(n)
(n)
uh
+ vh
x
y

p(n)

n=1

Np
X

(m)

(n) d =
x

+ vh
x
y

p(n)

n=1

(m)

(n)

(n)

(m) (n)
d =
y

(n)


Z  (m)
Nu
1 X

(n)
(m) (n)
(n)
d +
u
+
d
Re n=1
x
x
y
y

(m) f1 d

m = 1, . . . , Nu

(m) uh

Nu
X

(m)

d +


Z  (m)
Nu
1 X

(n)
(m) (n)
v (n)
+
d
Re n=1
x
x
y
y

(4.33)

(m) f2 d

m = 1, . . . , Nu

Nu
X

d +
v (n)
x
n=1

(m)

(n)
d = 0
y

m = 1, . . . , Np

Placing the coefficients into column matrices,



T
u = u(1) , . . . , u(N u) ,

T
p = p(1) , . . . , p(Np ) ,


T
v = v (1) , . . . , v (N u) ,

we can write the system (4.33) in the matrixvector form

(x)
1
C(uh , vh ) + Re
K
0
G(x)
u
b
1

K G(y) v = b(y) ,
0
C(uh , vh ) + Re
p
0
G(x)T
G(y)T
0

(4.34)

where the matrix and vector components are



Z  (m)

(n)
(m) (n)
Kmn =
+
d,
x
x
y
y

Cmn (uh , vh ) =

G(x)
mn

(m)
(n)
d,
x

G(y)
mn

(m)

(m)
(n)
d,
y

b(x)
m

(m)

(n)
(n)
uh
+ vh
x
y

f1 d,

b(y)
m

d,

(m) f2 d.

(4.35)
Matrix K represents an approximation of the discrete negative Laplacian 2 ; C(uh , vh ) the discrete

advection operator u x
+ v y
; G(x) and G(y) the discrete gradient ; and G(x)T and G(y)T the discrete
divergence. That is, the structure of the NavierStokes equations are directly reflected in the nonlinear
system (4.34). We obtained the same structure of the nonlinear equation for the finite-volume discretization,
except that here, it is guaranteed the the last block row of the matrix is the negative transpose of the last
block column.

4.2.4

Stability

So far, the approximating spaces Vh and Hh have not been specified. The counting argument of example 2
indicates that not any choice is good. We will limit the discussion to the Stokes equations, whose finiteelement formulation reads
Find uhi Vh and ph Hh such that
Z
Z
Z
z h
zih uhi
d ph i d = zih fi d
xj xj
xi

Z
uhi
qh
d = 0
q h Hh .
xi

zih Vh ,

(4.36a)

(4.36b)

The Stokes equations is a limit case of the NavierStokes equations for very low Reynolds numbers and
assume that inertia can be neglected compared to viscous forces. The Stokes equations may be used to
model very viscous flow, creeping flow, lubrication, but also flow of common fluids (like air and water) at
the micro scale. Due to the lack of nonlinear advection, these are linear equations. Mathematically, the
Stokes equations are much easier to analyze than the full NavierStokes equations. Indeed, there exists a
unique weak solutions for each square-integrable fi both in two and three space dimension. The reason
to start the analysis on the Stokes equation is that a stable Stokes discretization is necessary to obtain a
stable NavierStokes discretization.
As for the NavierStokes equations, we require that Vh V and Hh H. Recall that each function in
V and H is required to be bounded in the energy and the L2 () norm, respectively. Thus, in makes sense
to require that the velocity and pressure approximations satisfy the same bounds, that is,
Z
1/2
Z
1/2
kph kL2 () =
p2h d
C
fi fi d
= Ckf kL2 () ,
(4.37a)

kuh kV =

Z

uhi

uhi

xj xj

1/2

Ckf kL2 () ,

(4.37b)

where the constant C should not depend on h. This implies that the approximations will not blow up when
the mesh is refined.
Choosing zih = uhi in equation (4.36a), we obtain
Z
Z
Z
Z
uhi uhi
uh
uhi uhi
d ph i d =
d = uhi fi d,
xj xj
xi
xj xj

where the first equality follows from equation (4.36b). Thus,


Z
kuh k2V =
uhi fi d kukL2 () kf kL2 () CkukV kf kL2 ()

(4.38)

where the CauchySchwarz inequality yields the first inequality, and the Poincare inequality the second.
We thus conclude from expression (4.38) that
kuh kV Ckf kL2 () ,

(4.39)

that is, any Galerkin approximation of the velocity in the Stokes equations is stable! For bad choices of
subspaces, we will of course obtain inaccurate velocity approximations, but they will not blow up as the
mesh is refined, since their energy norm is bounded by the given data fi .
Regarding the pressure, the situation is more complicated. Not any combination Hh and V h yield
stable pressure approximations in the sense of (4.37a). We will derive a compatibility condition between
the velocity and pressure spaces, the LBB condition, necessary to yield stable pressure approximations.

4.2.5

The LBB condition

Rearranging equation (4.36a) implies that for each zih Vh ,


Z

Z
zih uhi
d
zih fi d

xj xj

Z
1/2 Z
1/2 Z
1/2 Z
1/2
zih zih
uhi uhi
(CauchySchwarz)
d
d
+
zih zih d
fi fi d
xj xj
xj xj

ph

zih
d =
xi

= kz h kV kuh kV + kz h kL2 () kf kL2 ()


(Poincare) kz h kV kuh kV + kz h kV kf kL2 ()

(by (4.39)) Ckz h kV kf kL2 () + kz h kV kfkL2 () .

(4.40)

Dividing expression (4.40) with kz h kV yields that


Z
z h
1
ph i d (C + 1)kfkL2 ()
kz h kV xi
for each zih Vh , from which it follows that there is a C > 0 such that
max

06=zi V

1
kz h kV

ph

zih
d Ckf kL2 () .
xi

(4.41)

We have thus found a bound on the discrete pressure in the exotic norm (4.16). However, we were aiming
for the bound (4.37a). Thus, if it happens that
kph kL2 () max

06=zi Vh

1
kz h kV

zih
d
xi

ph

(4.42)

for some > 0 independent of h, then it is immediate that kph kL2 () Ckf kL2 () for some C > 0.
Condition (4.42) says that our exotic norm is stronger than the L2 ()-norm for ph . The LBB-condition
is a condition on the pairing of spaces Vh and Hh that requires the exotic norm to be stronger than the
L2 ()-norm for each function qh Hh . The derivation above proves thus that the LBB-condition is
sufficient for the pressure to be stably determined by data:
Theorem 3. Assume that the LBB-condition holds for the spaces H h , Vh . That is, assume that there is
an > 0 independent of h, such that, for each qh Hh ,

Z

1/2

qh2

Z

p2h

1
max
06=zi Vh kz h kV

qh

zih
d,
xi

(4.43)

then there is a C > 0 such that

1/2

Z

|f | d

1/2

holds, independently of h.
The LBB condition (4.43) is satisfied for Vh = V and Hh = H. That is, the pressure (before discretizaR
z h
tion) is stably determined by data. The expression qh xii d represents the discrete gradient of qh (see
4.2.3). The LBB condition thus means that a zero discrete pressure gradient implies a zero value of the
pressure. Conversely, if the LBB condition is not satisfied, it means that the discrete pressure gradient
may be small even when the pressure is not small: The discrete gradient cannot feel that the pressure is
large, an effect that usually manifests itself as wild pressure oscillations.


M2
M
m
m+2

h
V
Vh
1
u
uh
0
1 1

Figure 4.15: A checkerboard mode for equal-order interpolation on a regular triangular mesh.

4.2.6

Mass conservation

R
The integral ni ui d yields the total flux of mass(in kg/s)R through the boundary. There is no net
mass-flux through the boundary for a constant-density flow, so ni ui d = 0. A velocity approximation
uhi is is globally mass conservative if
Z

or, equivalently, if

ni uhi d = 0,

uhi
d = 0,
xi

by the divergence theorem (4.3). Recall from equation (4.30c) that


Z
uh
qh i d = 0
qh Hh .
xi

A Galerkin approximation is thus globally mass conservative if 1 Hh , that is, if constant functions can
be represented exactly in Hh . All reasonable FE approximation satisfy this.
A finite-element approximation is locally mass conservative if
Z
uhi
d = 0,
K xi
for each element K in the mesh. Not all finite-element approximations are locally mass conservative. If
the pressure approximations consist of piecewise constants on each element, the FE approximation will be
locally mass conservative. However, it will not be locally mass conservative if the pressure approximations
are continuous and piecewise linear.

4.2.7

Choice of finite elements. Accuracy

Now we consider a regular triangulation of a 2D domain , that is, a triangulation satisfying a mesh quality
condition of the sort discussed in section 4.1.5.
Equal-order interpolation
Let us start by trying the approximation space we used for the advectiondiffusion problem (section 4.1.1).
That is, we use continuous functions that are linear on each triangle both for the functions in H h and Vh .
This pair does not satisfy the LBB condition and yields an unstable discretization. In particular, the
system matrix (4.34) is singular : the discrete gradient matrix G has linearly dependent columns. Thus,
there is at least one nonzero Np -vector pc such that Gpc = 0. Corresponding function pch Hh , whose
nodal values are equal to the elements of pc is called a checkerboard mode.
Figure 4.15 depicts the nodal values of a checkerboard mode for equal-order interpolation. To see this,
consider the element contribution, associated with a triangle Tn , to the ith row of vector Gpc (notice from
i
the expressions in figure 4.5 that
x is constant on each triangle):
Z
Z
i
c i

ph
d =
pc d 0,
T
x
x n Tn h
Tn

h
V
Vh
u
uh
0
1

Figure 4.16: The evaluation points at each triangle for the continuous, piecewise quadratic velocity approximations of the TaylorHood pair.
since the mean value of the function pch vanishes on each triangle, as can be seen from figure 4.15. The right
side of the LBB condition (4.43) vanishes for qh = pch , which immediately implies that the LBB cannot
hold for a space Hh that includes pch . This (and other) checkerboard mode will pollute the solution of the
Stokes as well as NavierStokes equations..
Constant pressure, (bi)linear velocities
Now we consider a triangulation either of triangles or rectangles. Let the functions in H h be piecewise
constant on each element. In case of triangular elements, the space Vh will be continuous functions, piecewise
linear on each element, whereas for quadrilateral elements, continuous functions that are piecewise bilinear
on each rectangle comprise Vh .
Remark 3. The pressure approximations need not to be continuous in order for Hh H. However, if
Vh V , the velocity approximations cannot be discontinuous, since v is not a function for discontinuous
v.
This choice leads to a locally mass-conserving approximation. However, for triangular elements, the
counting argument of example 2 applies here and indicates problems with this choice of elements. Indeed,
the LBB conditions is not satisfied for this unstable discretization. Checkerboard modes similar as in figure 4.15 exist. Still, the constant pressurebilinear velocity space on rectangles is in widespread engineering
use! Fixes have been developed to take care of pressure oscillations. For instance, the checkerboard mode
can be filtered out from the discrete pressures. Moreover, common iterative methods to solve the resulting
discrete equations (such as the projection schemes) implies a stabilization of the discretization.
The TaylorHood pair
Here, the functions in Hh and Vh are globally continuous and, on each triangle,
ph Hh is linear, whereas
uh , vh Vh is quadratic.
Thus, on each triangle, we have
uh = a + b x + c y + d xy + e x2 + f y 2
The six coefficients af are uniquely determined by uh s values at the triangle vertices and midpoints
(figure 4.16).

Nu
A basis (m) (xi ) m=1 of continuous, piecewise quadratic functions interpolates the nodal values of
P u
(m)
(xi ). There are two kinds of basis
figure 4.16, so that uh Vh can be expressed as uh (xi ) = N
m=1 ui
functions here, illustrated in figures 4.17 and 4.18.
The LBB condition is satisfied for the TaylorHood pair, and the following error estimates holds for a
smooth solution to the NavierStokes equation:
ku uh kL2 () Ch3

kp ph kL2 () Ch2

The TaylorHood pair is globally, but not locally mass conservative.


The TaylorHood pair can be generalized to higher order. Then, the pressure and velocity approximations will be continuous functions, consising of piecewise polynomials of degree k for pressure and k + 1 for
velocity (k 1) on each triangle.

m+2
m+2 0
0 d0
11d
1
d
1
2h a0
h
a
ia
0
V
V
a
i

1,
j

i
j 1
h j1
h1
i V1,
i, jV
j
Inflowu
ui, j 1
i + 1, j 1
Inflow
Solid wall
uh j
h j1
i +u1,
Solid wall
n
0
0
j
0 i
1 n i
1
0
i 1+ 1
i + 1
i,j
u
1

ui,j
2vi,j

(m)
(m)
M
Figure 4.17:
The
basis
function

(x
)
when
Figure
v
i
i,j
pi,j4.18: The basis function (xi ) when
2
M
M
m corresponds
to an edge-midpoint node
m corresponds
to a corner node
p
pi,j
M i,j
m
pi,j
v
m i,j+1/2
+2
v m
vi,j1/2
O(h2)
m i,j+1/2
+2
1
vi,j1/2
ui1/2,j
1
h
ui1/2,j
i 12
h
1
V
i 2
V
Vh 1i
i+ 2
Vh 1i
u
i+ 2
u
ui h+ 31
i+
ui h+ 31
0 21
i+ 2
j
+
0
1 2
j + 12
j
1
j
j
12
1
j2
j 1Isoparametric elements use polynomials
Figure 4.20:
j 1Approximating a curved boundary with
Figure 4.19:
of the sameAorder as used for the FE approximations
A leads to an O(h2 ) error of the bound- to approximate
a triangle sides
B the shape of the boundary.
B
ary condition.
C
u 23 ,1
C
u 23 ,1
v1, 32
v1, 32Inaccuracies in approximations of the domain boundary
Remark 4.
p1,1 may reduce the convergence rate of
p1,1 approximations. For instance, a straight-forward triangulation of a domain with a curved
higher-order
0
2
boundary will
0 lead to a boundary-condition error at the mid nodes
1 2 of O(h ) (figure 4.19). This is fine
for piecewise-linear
approximations since the error anyway is O(h
1
2 ). However, this inaccuracy at the
boundary results
in a half-order reduction of the convergence rateiof piecewise quadratic approximations.
2
Using curvedi elements restores convergence rate. The most common
j strategy for curved elements is known
j
as Isoparametric
Elements: piecewise polynomials of the sameInflow
degree as used in the FE approximation are
used toInflow
approximate the shape of the boundary (figure 4.20).
Solid wall
Solid wall
n
n
0 pressures
A stable approximation
with piecewise-linear velocities and
0
1
Instead of using
different approximation orders, as in the TaylorHood
pair, we here consider different
1

2 approximations consist of continuous,


meshes for
velocity and pressure. Both the velocity and pressure
M
piecewise-linear
M 2 functions on triangles, but each pressure triangleMis subdivided into four velocity triangles
as in figureM4.21.
m
This is amstable approximation satisfying the LBB condition,
and
m+
2 the following error estimates holds for
m+2
1
1
h
h
V
V
Vh
Vh
u
u
uh
uh
0
0
1
1

Figure 4.22: Subdividing a pressure rectangle into


Figure 4.21: Subdividing a pressure triangle into four four velocity rectangles yields a stable approximation
velocity triangles yields a stable approximation with with continuous piecewise-bilinear functions for both
continuous, piecewise-linear functions for both veloc- velocity and pressure.
ity and pressure.

1
h
V
Vh
u
uh
0
1

1
h
V
Vh
u
uh
0
1

Figure 4.24: Approximations spanned by


the basis functions (4.23) yields piecewiselinear functions that are continuous only
along at the edge midpoints, in general.

Figure 4.23: The basis functions associated with the


velocity approximation for a nonconforming but stable discretization.

a smooth solution to the NavierStokes equation:


ku uh kL2 () Ch2
kp ph kL2 () Ch

These approximations use the same number of unknowns and the same nodal locations as for the Taylor
Hood pair, but are one order less accurate. Calculating the matrix elements (4.35) is however somewhat
easier than for the TaylorHood pair.
There is also a version of this discretization for rectangular meshes: Each pressure rectangle is then
subdivided into four velocity rectangles, as in figure 4.22. Both the velocity and pressure approximations
consist of continuous and piecewise-bilinear functions.
A nonconforming method
The discussion after example 2 suggested a strategy to address the problem with the overdetermined incompressibility condition, namely to use nodal points on the edge mid points for the velocity. To accomplish
this, consider basis functions, associated with the edges, of the type depicted in figure 4.23. Basis function
j is linear on each triangle and continuous along edge j (but discontinuous along four edges!). If x i , y i
are the coordinates of the mid point of edge i, we have
(
1 for i = j,
i i
j (x , y ) =
0 for i 6= j.
Now define the space of the velocity components Vh from all possible expansions
uh (x, y) =

N
X

uj j (x, y),

vh (x, y) =

j=1

N
X

vj j (x, y),

j=1

Here, N is the total number of edges, excluding those on 1 . Note that uh will be continuous only at edge
mid points in general. Thus, Vh 6 V since uh is not a square-integrable (in fact, constant) function.
Violating a property like Vh V is called a variational crime. However, uh restricted on each element is
a square-integrable function. Letting Hh be piecewise constant on each triangle and Vh defined as above,
we define the nonconforming FE approximation:
Find uh , vh Vh and ph Hh such that
Z
M Z
M Z
M Z
M Z
X
X
X
uh
uh
1 X zh uh
zh
z h uh
d +
zh vh
d +
d
ph
d = zh f1 d
x
y
Re n=1 xj xj
x
n=1
n=1
n=1
Tn

Tn

M Z
X

z h uh

M Z
X

qh

n=1T
n

n=1T
n

M Z
X

vh
d +
x
n=1

Tn

uh vh
+
x
y

Tn

Tn

Z
M Z
M Z
X
vh
1 X zh vh
zh
zh vh
d +
d
ph
d = zh f2 d
y
Re n=1 xj xj
y
n=1

d = 0

where M is the number of triangles.

Tn

q h Hh ,

Tn

z h Vh ,

z h Vh ,

The fact that Vh 6 V complicates the analysis of this discretization. Nevertheless, the approximation
is stable, although not as accurate as the TaylorHood pair, for instance. Other nice properties with this
method is that it is locally mass conservative, and quite easy to implement.
But maybe the most interesting with this discretization is that it allows for a construction of a divergence
free velocity basis. In all types of discretizations discussed so far, including the present, we have used the
same basis for the uh and vh components. By a clever combination of basis functions of the type illustrated
(j)
(j)
in figure 4.23, it is possible to construct a new vector-valued basis functions j = (zh , wh ) for the velocity
having the property
(j)

j =

(j)

zh
wh
+
x
y

= 0.

Expanding the velocity in this basis,

uh = (uh , vh ) =

M
X

uj j ,

(4.44)

j=1

yields that uh = 0. Inserting expansion (4.44) into equations (4.30), and choosing (zh , wh ) = i , the
continuity equation and the pressure variable vanishes, and equations (4.30) reduce to
M
X
j=1

uj

i (uh ) j d +

M
X
j=1

uj

i j d +

dN
Xu
j=1

uj

i (uh ) j d =

i f d,
(4.45)

for i = 1, . . . , M . The NavierStokes equations then reduces to a vector-valued, non-linear advection


diffusion problem, quite similar to equation (4.10). This leads to a substantial simplification when designing
an algorithm to solve the discretized equations; in fact, one of the great challenges for any solver of
the NavierStokes equations is how to treat the incompressibility condition (4.30c). Unfortunately, the
construction of the basis functions j is not straightforward.

Stabilization of unstable discretizations


Stability of the choices of elements presented in sections 4.2.74.2.7 relied on manipulations of the local
polynomial order, the meshes, or the conformity of the method. These manipulations complicate the
implementation. An alternative is to use the same approximations for velocity and pressure (equal-order
interpolation) combined with a stabilization method. This circumvents the LBB condition.
We will present a very simple idea of this sort for the Stokes equations. First choose the same type
of approximations for the pressure and the velocity components, for instance continuous, piecewise-linear
functions on a triangular mesh, and replace equations (4.36) with
Find uhi Vh and ph Hh such that
Z
Z
Z
z h
zih uhi
d ph i d = zih fi d
zih Vh ,
xj xj
xi

Z
Z
qh ph
uhi

d + qh
d = 0
qh Hh .,
xi
xi xi

(4.46a)

(4.46b)

for some small  > 0. Equation (4.46b) formally inconsistent by O() with corresponding variational form.
More advanced stabilization methods avoids this inconsistency by using a term that vanishes as h 0.
The main advantage with stabilization methods is the simplified implementation, but the method introduces an extra parameter that may need judicious tuning. Note also that the increased number of pressure
unknowns, as compared with the discretization in section 4.2.7, does not increase the accuracy. The added
resolution introduced is filtered away by the introduction of the stability term.
To see that approximation (4.46) is stable, choose zih = uhi , qh = ph and add equations (4.46a)

and (4.46b),
Z

uhi uhi
d + 
xi xi

ph ph
d =
xi xi

(CauchySchwarz)

Z

(Poincare) C

uhi fi d

uhi uhi d

Z

uhi

1/2 Z

uhi

fi fi d

1/2 Z

1/2

fi fi d

1/2

xi xi

1/2 Z
1/2
 Z
Z
ph ph
uhi uhi
d +
d
fi fi d
.
C 
xi xi

xi xi

Dividing and squaring yields the stability estimate


Z
Z
Z
uhi uhi
ph ph
d + 
d C
fi fi d.
xi xi
xi xi

Since C > 0 does not depend on h, both the velocity and pressure approximations cannot blow up as h 0,
even though the discretization is unstable for  = 0.

j
j 12
j1
A
B
C
u 23 ,1
v1, 32
p1,1
0
1
2
i
j
Inflow
Solid wall
A.1 Iterative
solutions to linear systems
n
0
1 method
Gauss-Seidel

The Laplace equation


on a Cartesian grids can be discretized as
M2
M
i+1,j 2i,j + i1,j + 2 (i,j+1 2i,j + i,j1 ) = 0
m
+2
where =mx/y.
We can solve for i,j ,
1
i+1,j + i1,j + 2 (i,j+1 + i,j1 )
h
i,j =
V
2 (1 + 2 )
Vh
Gauss-Seidel uses this expression to update i,j (using the already updated values calculated in same
u
iteration), i.e.u
h

0
ki+1,j + k+1
+ 2 ki,j+1 + k+1
i1,j
i,j1
k+1
1
i,j =
2 (1 + 2 )

Appendix A

Background material

io

j+1
: new values

di

re
ct

: to be updated

ra

tio

ite

: old values

i1

i+1

j1

This iteration method has a slow convergence, it is possible to show


k k k m k 0 k
where

e.i. the error is reduced by O h


iterations must satisfy


2

= 1 O h2

h = x = y


each iteration. Requiring an error reduction to O h2 the number of
m = O h2

which implies that


m=O

ln(h)
ln()

=O

ln(h)
ln(1 O(h2 ))

= O h2 ln(h)

To proceed we note that we have N = n2 unknown interior nodes, see Figure A.1. Thus
95


1
M2
h = x = y =
n1 = N 1/2
M
n+1
m
which implies that
m+2
1
m = O (N ln(N ))
h
and that the total work is V
Vh

W = O N 2 ln(N )
u
uh
since the work per iteration is O (N ).
0
1
y

: to be updated
: new values1
: old values n
i1
i
i+1
j1
j
j+1
iteration direction
1

x
1

n
1

Figure A.1: Domain for the solution of the Laplace equation.

Multigrid method
A general way to accelerate the convergence of e.g. the Gauss-Seidel method is provided by the Multigrid
method. We note that the Gauss-Seidel provides good smoothing of the local error, but converges slowly
since it takes time for boundary information to propagate into the domain. The idea behind the multigrid
method is that a slowly converging low-frequency on a fine grid is a fast converging high-frequency on a
coarse grid. Let
Li,j =

i,j+1 2i,j + i,j1


i+1,j 2i,j + i1,j
+
x2
y 2

and define the correction to the solution to the true discrete solution in the following manner
i,j
ki,j
i,j

= ki,j + i,j
solution at iteration level k
correction to ki,j to true discrete solution

This implies that


Li,j = Lki,j Ri,j
since Li,j = 0. Thus we are left with an equation implying that the Laplace operator on the correction is
equal to the negative of the residual. This residual equation is solved on a coarser grid and the correction
is interpolated onto finer grid.
In order to define the algorithm we use the following definitions
I
transfer operator between grids:
An example of a restriction operator I12 is to
I12 - restriction operator from grid 1 to grid 2
I21 - prolongation operator from grid 2 to grid 1
define it as an injection, i.e. we use every other value in each direction. The prolongation operator I 21
may be defined by linear interpolation for the intermediate values, i.e. an average of right and left values

: old values
i1
i
i+1
j1
j
j+1
iteration direction
x
y
1
n

fine grid (1)


coarse grid (2)

j
x
i
Figure A.2: Two level grid.
and top and bottom values. The values marked with in figure A.2 are then taken as an average of the
averages of the intermediate values.
Now we can define the following two level scheme:
1. Li,j iterated k times
2. Li,j = I12 Ri,j
3. i,j = ki,j + I21 i,j

Lki,j = Ri,j

iterated k times, 0i,j = 0 (starting value) (here i,j is definied on grid 2)


and goto 1.

at the end of 1. convergence is checked.


Wesseling [10] estimated the work of the multigrid method to be W = O (N ln N ), a substantial improvement over the Gauss-Seidel method.

i
y = x2
i
i+1
j
j1
fine
grid
(1)
j
j coarse
+ 1 grid (2)
iteration direction
r
x
y
e1
e1
1
x = x1
n
e
3
e2
x
z = x3
y
i
Figure A.3: Cartesian coordinate system.
j
fine grid (1)
x2
coarse grid (2)
x02
e02
e2

x01
e01
e03

x03

e1

e3

x1

e1

x3
Figure A.4: Two different coordinate systems.

A.2

Cartesian tensor notation

Tensor notation is a compact way to write vector and tensor formulas. Vectors and tensors are objects
which are independent of the coordinate system they are represented in.
In the Cartesian coordinate system in Figure A.3 we have the unit vectors
|ek | = 1

k = 1, 2, 3

The scalar product of two unit vectors is the Kroneckers delta, kl


ek el = kl =

1 k=l
0 k=
6 l

and the position vector r is


r = x 1 e1 + x2 e2 + x3 e3 =

3
X

xk e k = x k e k .

k=1

The last expression make use of Einsteins summation convention: when an index occurs twice in the same
expression, the expression is implicitly summed over all values for that index.
Example
kk = ll = 11 + 22 + 33 = 3
ij aj = i1 a1 + i2 a2 + i3 a3 = ai .

A.2.1

Orthogonal transformation

Consider two arbitrary oriented coordinate systems as in Figure A.4. Assume identical origin, a = 0. The
unit vectors are orthogonal
e0k e0l = kl

e 1 x1

n
x
y
i
j
fine grid (1)
coarse grid (2)

e1
e02

e02 e1 x1 = cos (x02 , x1 ) x1 = c21 x1


Figure A.5: Projection of e1 x1 on e02 .
and a position vector is independent of coordinate system,
r = xl el = x0l e0l .
Component k of r can be written as
e0k r = xl e0k el = x0l e0l e0k = x0l kl = x0k

x0k = e0k el xl = ck0 l xl

where we have the transformation tensor


ck0 l = e0k el = cos(x0k , xl ).
Since cos(x02 , x1 ) 6= cos(x01 , x2 ) in general, we have (for k = 1, l = 2)
cl0 k = e0l ek = cos(x0l , xk ) 6= ck0 l
From
ek r = xl ek el = x0l ek e0l

xk = e0l ek x0l = cl0 k x0l = clk x0l

we conclude that first index refer to primed system!


Example
x0k

= ckl xl = ck1 x1 + ck2 x2 + ck3 x3 = cos(x0k , x1 )x1 + cos(x0k , x2 )x2 + cos(x0k , x3 )x3
= e0k e1 x1 + e0k e2 x2 + e0k e3 x3

which for component 2 reads


x02 = e02 e1 x1 + e02 e2 x2 + e02 e3 x3
x02 is projection of e1 x1 , e2 x2 , e3 x3 on e02 . A visualization of the projection of e1 x1 on e02 can be seen
in Figure A.5.
The relation between ckl and cml can be studied through
x0k = ckl xl = ckl cml x0m
|{z}
|{z}

(ckl cml km ) x0m = 0.

xk = clk x0l = clk clm xm


|{z}
|{z}

(clk clm km ) xm = 0.

km x0m

cml x0m

In the same manner we have

km xm

A.2.2

clm xm

Cartesian Tensors

Def. a Cartesian tensor of rank R is a set of quantities Tklm... which transform as a vector in each of its
R indices.
0
Tklm...
= ckr cls cmp . . . Trsp...

Example
For scalars the rank is R = 0
a b = a0k b0k = ckl ckm al bm = al bl ,
| {z }
lm

for vectors R = 1

a0k = ckl al

and for a tensor of second order we have R = 2


0
kl
= ckm cln mn = ckm clm = kl .

kl is isotropic, i.e. identical in all coordinate systems.


An outer product generates a tensor
0 0
Tkl
Srs = ckm cln crp csq Tmn Spq

and an inner product generates a tensor


0 0
Tkl
Sls = ckm cln clp csq Tmn Spq = ckm csq Tmn Snq
| {z }
np

A.2.3

Permutation tensor
klm

1
even
1
odd
=
if klm is
permutation of (1, 2, 3)

0
no

As examples of permutations of 1,2,3 we have


(1,2,3), (3,1,2), (2,3,1)
(2,1,3), (3,2,1), (1,3,2)
at least 2 subscripts equal

A.2.4

even permutation
odd permutation
no permutation

123 = 1
213 = 1
221 = 0

Inner products, crossproducts and determinants

The inner product of the vectors a and b is


a b = ij ai bj = ai bi = a1 b1 + a2 b2 + a3 b3 .
The cross product

e1

a b = a1
b1

of a and b can be written as



e2 e2
a2 a3 = (a2 b3 a3 b2 )e1 + (a3 b1 a1 b3 )e2 + (a1 b2 a2 b1 )e3 = ijk ei aj bk
b2 b3

and the determinant of the tensor aij is




a11 a12 a13


|{aij }| = a21 a22 a23 = ijk a1i a2j a3k
a31 a32 a33
1
1
=
ijk (ai1 aj2 ak3 + ai3 aj1 ak2 + ai2 aj3 ak1 ai2 aj1 ak3 ai3 aj2 ak1 ai1 aj3 ak2 = ijk pqr aip ajq akr .
6
6
We have the permutation relation
klm ksp = ls mp lp ms .
If we set s = l in (A.1) we get
klm klp = (3 1)mp = 2mp

and if we also use p = m we end up with

klm klm = 6
The permutation tensor can also be used to derive the following vector relation
(a b) (c d) = ijk aj bk ilm cl dm = (jl km jm lk )aj bk cl dm

= aj cj bk dk aj dj bk ck = (a c)(b d) (a d)(b c)

(A.1)

A.2.5

Second rank tensors

All second rank tensors can be decomposed into symmetric and antisymmetric parts, T kl = T(kl) + T[kl] ,
where the antisymmetric part is
1
T[kl] = T[lk] = (Tkl Tlk )
2
and the symmetric part
1
T(kl) = T(lk) = (Tkl + Tlk ) .
2
The symmetric part can be decomposed into isotropic and traceless part, T(kl) = T kl + T kl , the isotropic
part being
1
T kl = Tmm kl
3
and the traceless part is
1
T kl = T(kl) Tmm kl .
3
If we project the isotropic, traceless and entisymmetric parts we get
 1

1
S(kl) T[kl] + S(lk) T[lk] =
S(kl) T[kl] S(lk) T[kl] = 0
2
2




1
1
1
1
= Smm kl T(kl) Tmm kl = Smm Tkk Tkk ll = 0
3
3
3
3

S(kl) T[kl] =
S kl T kl

Skl T[kl]
Skl T(kl)
Skl T kl
Skl T kl

= S[kl] T[kl]
= S(kl) T(kl)
= S kl T kl
= S kl T kl ,

and each part is invariant under transformation


0
0
= ckr cls T[rs] = ckr cls T[sr] = clr cks T[rs] = T[lk]
T[kl]
0
Trr
= crp crq Tpq = pq Tpq = Tpp



1
1
0
0
T kl = ckp clq T pq = ckp clq T(pq) Trr pq = T(pq)
Trr kl .
3
3
Since we only need three components to completely describe an antisymmetric tensor of rank R = 2 it
can be represented by its dual vector

di = ijk Tjk = 
ijk
T
(jk) + ijk T[jk]
The first term on the right hand side is zero since ijk is antisymmetric in any two indecies. Multiply both
sides by ilm
ilm di = ilm ijk Tjk = (lj mk lk mj ) Tjk = Tlm Tml = 2T[lm]

T[lm] =

1
ilm di
2

Note: There are 3 independent components in T[lm] and di .


In order to sum up, an arbitrary tensor of rank 2 can be divided into the following parts
Tkl = T(kl) + T[kl] = T kl + T kl + T[kl] =

1
1
Trr kl + T(kl) Trr kl +
3
| {z } |
{z3
}
isotropic

traceless

1
ikl di
|2 {z }

antisymmetric

Properties of the above parts are invariant under transformation. In the Navier-Stokes equations we have
ui
the velocity gradient x
. The antisymmetric part of this tensor describes rotation, the isotropic part the
j
volume change and the traceless part describes the deformation of a fluid element.

Example
We have the second rank tensor

1 2 3
{Tkl } = 4 5 6
7 8 9

which can be divided into a symmetric part




T(kl)

and an antisymmetric part




T[kl]

1 3 5
= 3 5 7
5 7 9

0 1 2
= 1 0 1 .
2 1
0

The symmetric part can then be divided into a isotropic

5
n o
T kl = 0
0
and a traceless part

T kl

The dual vector of the antisymmetric tensor is

part

0 0
5 0
0 5

4 3 5
= 3 0 7 .
5 7 4

2
{di } = 4 .
2

A.2.6

Tensor fields

We have the tensor field Tij (x1 , x2 , x3 ) Tij (xk ) with the following properties
partial derivative transforms like a tensor
xk

=
= cpk
x0p
x0p xk
xk
gradient adds one to tensor rank

where

(xk = cpk x0p )

Tij
xk

divergence decreases tensor rank by one

T1l
T2l
T3l
Tkl =
+
+
xk
x1
x2
x3
curl

( u)i = ijk

uk
xj

which can be compared with the expression for the determinant.


Example
Show that ( ) = 0.
( ) =
since klm is antisymmetric in kl and

m
2
klm
= klm
m = 0
xk
xl
xk xl

2
xk xl

is symmetric in kl.

y
1
n
x
y
i
j
fine grid (1)
coarse grid (2)

dS = n dS
Figure A.6: Volume V bounded by the close surface S.
Example

Show that f (r) = 1r f 0 (r) where {r}k = xk and r = |r| = xk xk .


{f (r)}k

df r
(xp xp )1/2
f (r) =
= f0
=
xk
dr xk
xk


1
1
1
1
1 0

f 0 (xp xp ) 2
(xp xp ) = f 0
(pk xp + xp pk ) = f 0 xk =
fr
2
xk
2r
r
r
k

Example
Show that ( F) = ( F) F.
( F)

= ijk

(il jm im jl )
Example



klm
Fm = ijk klm
Fm = kij klm
Fm =
xj
xl
xj xl
xj xl







Fm =
Fj
Fi =
Fj
Fi = ( F) F i
xj xl
xj xi
xj xj
xi xj
xj xj

r then = u . This means, show that ijk uk = 2i .


Show that if u
=
2
xj
ijk

uk = {uk = klm l rm } = ijk


(klm l rm ) = ijk klm
l rm =
xj
xj
xj

kij klm

l rm = (il jm im jl )
l rm =
xj
xj

rm
ri
rm
i rm
l ri = { independent of x} = i
l
={
=1+1+1=3
xm
xl
xm
xl
xm

ri
= il } =
xl

3i l il = 2i

A.2.7

Gauss & Stokes integral theorems

Let n be the outward pointing normal to the closed surface S bounding the simply-connected volume V ,
see Figure A.6. Gauss integral theorem then states
I

a n dS =

ak nk dS =

or

a dV

ak
dV
xk

j
fine grid (1)
coarse grid (2)

n
S
dl

Figure A.7: The open surface S enclosed by the contour C.


for a differentiable vector field a. This can be generalized to
I
Z

Tklm np dS =
Tklm dV .
xp
S

Example
I

f n dS =

a n dS =

f dV .

a dV

if

Tklm nm = klm al nm .

Stokes theorem states that for a differentiable vector field a if an open simply-connected surface S is
enclosed by a closed curve C, of which dl is the line element as in Figure A.7, then
I
Z
a dl = ( a) n dS
C

or

ak dxk =

this can be generalized to

klm nk

ijp ni

Tklm dxp =

am
dS
xk

Tklm dS
xj

Example
R
f
Let Tklm = f. In the generalized Stokes theorem we then get ijp ni x
dS in right hand side
j
S

A.2.8

f dl =

n f dS

Archimedes principle

Gauss theorem can be used to calculate the buoyancy on an object immersed in a fluid (Archimedes circa
281212 BC). Consider a water cylinder with volume dV and mass dm, Figure A.8.
The force from the water cylinder is
dm g = dV g = x3 g dA
| {z }
pressure

where g is the gravity constant and the water density. The pressure at a distant x 3 from the surface in
a fluid is then p = gx3 + p0 where p0 is the atmospheric pressure. Let be the buoyancy force resulting
from the pressure on the object, m the mass of equivalent volume of water and M the mass of the object
in Figure A.8. The force on a small element dS of the body resulting from the pressure is
dk = gx3 nk dS

i + 1 1
p0
i + 32

1
j +updated
2
: to be
j
: new 1values
n
j2
: old values
j1i1
dV dm
A
i
Bi+1
Cj 1
u 23 ,1
j
v1, 32
dA
j+
1
p1,1 M
x3
iteration direction
0
x
1
y
2
1
Figure A.8: An object with mass M
in a fluid illustrating Archimedes principle.
i immersed
n
j
mg
x
Inflow
y
Solid wall
i
n
j
0
fine
grid
(1)

1
coarse grid
(2)

M2
Mg
M
m immersed in a fluid. m is the mass of equivalent volume of
Figure A.9: The forces acting on an object
m
+
2
water and M the mass of the body.
1
h
where the minus sign stems from the outward pointing normal n. The buoyancy force can then be calculated
V
through
I
Z
Vh

x3 dV = gk3 V.
k = gx3unk dS = g
x
k
u
h
S
V
0
This means that the object loses some weight
as the water is displaced. The buoyancy force is
1
3 = gV = g m
: to be updated
see Figure A.9.
: new values
The moment of a force F at some arbitrary point with position vector r from the point of application of
: old values
the force is M = r F. The moment around the center of gravity G of the homogenous body produced by
i1
the pressure force d acting on an element dS of the body is dMG = (r rG ) d as in Figure A.10.
i
This can then be used to calculate the total moment around G form pressure forces on the body
i+1

I
Z 
j1
x3

MGk = gklm (xl xGl )jx3 nm dS = gklm


(xl x3 ) xGl
dV =
xm
xm
S
V
j+1

Z
Z
iteration
direction
gklm [lm x3 + xm3 xl m3 xGl ] dV = gkl3 xl dV xGl V = 0
y
V
V
1
If a part of the fluid is frozen (without changing in density) this body is in equilibrium in the fluid:
n
F = M = 0.
x
y
i
j
fine grid (1)
G
coarse grid (2)
r rG
n
x
y
i
j
fine grid (1)
coarse grid (2)

n
Figure A.10: Moment around G from pressure forces acting on the object.

i1
i
i+1
j1
j
j+1
iteration direction
x
y
1
n
x
y
i
j
fine grid (1)
coarse grid (2)

z
P (r, , z)

z
y
r

x
Figure A.11: Cylindrical polar coordinates

A.3

Curvilinear coordinates

Cylindrical Polar Coordinates


The cylindrical polar coordinates are (r, , z), where is the azimuthal angle, see figure A.11. The velocity
can be written as
u = u r er + u e + uz ez ,
(A.2)
where the unit vectors are related to Cartesian coordinates as

Non-zero derivatives of unit vectors

er = ex cos + ey sin ,
e = ex sin + ey cos ,
ez = e z .

(A.3)

e
= er .

(A.4)

er
= e ,

Gradient of a scalar p
p =
Laplacian of a scalar p

p
1 p
p
e +
e +
e .
r r r z z



1
p
1 2p 2p
p=
r
+ 2 2 + 2.
r r r
r
z
2

Divergence of a vector u

u=

(A.5)

(A.6)

1 (rur ) 1 u
uz
+
+
.
r r
r
z

(A.7)

p u p
p
+
+ uz .
r
r
z

(A.8)

Advective derivative of a scalar p


(u )p = ur
Curl of a vector u
u =






1 uz
u
ur
uz
1 (ru ) ur

er +

e +

ez .
r
z
z
r
r
r

Incompressible Navier-Stokes equations with no body force




ur
u2
1 p
ur
2 u
2
+ (u )ur
=
+ ur 2 2
,
t
r
r
r
r


u
ur u
1 p
2 ur
u
2
+ (u )u +
=
+ u + 2
2 ,
t
r
r
r
r
uz
1 p
+ (u )uz =
+ 2 uz .
t
z

(A.9)

(A.10)
(A.11)
(A.12)

i1
i
i+1
j 1
j
j +1
ration direction
x
y
1
n
x
y
i
j
fine grid (1)
coarse grid (2)

z
P (r, , )
r

x
Figure A.12: Spherical polar coordinates

Spherical Polar Coordinates


The spherical polar coordinates are (r, , ), where is the azimuthal angle, see figure A.12. The velocity
can be written as
u = u r er + u e + u e ,
(A.13)
where the unit vectors are related to Cartesian coordinates as
er = ex sin cos + ey sin sin + ez cos ,
e = ex cos cos + ey cos sin ez sin ,
e = ex sin + ey cos .

(A.14)

Non-zero derivatives of unit vectors


er
= e ,

er
= e sin ,

e
= er ,

e
= e cos ,

e
= er sin e cos .

Gradient of a scalar p

(A.15)
(A.16)

p
1 p
1 p
e +
e +
e .
r r r r sin

(A.17)





1
1

p
1
2p
2 p
r
+
sin

+
.
r2 r
r
r2 sin

r2 sin2 2

(A.18)

1 (r2 ur )
1 (u sin )
1 u
+
+
.
r2 r
r sin

r sin

(A.19)

p =
Laplacian of a scalar p
2 p =
Divergence of a vector u
u=
Advective derivative of a scalar p

(u )p = ur

p u p
u p
+
+
.
r
r r sin

(A.20)

Curl of a vector u
u=

1
r sin




(u sin ) u
(ru )
1
1 ur

er +

r sin
r


1 (ru ) ur
+

e .
r
r

(A.21)

Incompressible Navier-Stokes equations with no body force


u2 + u2
ur
+ (u )ur
t
r

(A.22)



1 p
2ur
2 (u sin )
2 u
2
=
+ ur 2 2
2
,
r
r
r sin

r sin
u2 cot
u
u u
+ (u )u + r
t
r
r


1 p
2 cos u
2 ur
u
2
=
+ u + 2
2 2 2 2
,
r
r
r sin r sin
u
ur u
u u cot
+ (u )u +
+
t
r
r


u
1
p
2 ur
2 cos u
2
=
+ u + 2
+ 2 2
2 2
.
r sin
r sin
r sin
r sin

(A.23)

(A.24)

Appendix B

Recitations 5C1214
B.1

Tensors and invariants

Tensor Notation
Scalar
p=p
Vector

u
If u
= v then u2 = v
w

(
u)i = ui

Matrix

a11
If A = a21
a31

(A)ij = Aij

a12
a22
a32

a13
a23 then A23 = a23
a33

Kronecker delta
ij = (I)ij =

1 if i = j
0 if i 6= j

Einstein summation convention


ui ui =

3
X

ui ui

i=1

Kinetic energy per unit volume


1
u|2
2 |

= 21 (u2 + v 2 + w2 ) = 21 ui ui

Matrix operations
a
b = a1 b1 + a2 b2 + a3 b3 = ai bi = ij ai bj = aj bj
(Ab)i = Aij bj
(AB)ij = Aik Bkj
tr(A) = Aii
(A)ij = Aij (AT )ij = Aji
Permutation symbol
109

ijk

1 if ijk in cyclic order. ijk = 123, 231 or 312


=
0 if any two indices are equal

1 if ijk in anticyclic order. ijk = 321, 213 or 132


e1

a
b = a1
b1

e2
a2
b2




= e1 (a2 b3 a3 b2 ) e2 (a1 b3 a3 b1 ) + e3 (a1 b2 a2 b1 ) = ijk ei aj bk

e3
a3
b3

(
a b)i = ijk aj bk

ijk ilm = jl km jm kl
ijk ijm = {l = j} = 3km jm kj = 3km mk = 2km
ijk ijk = {m = k} = 2 3 = 6
Rewrite without the cross product:
= (
i = ijk aj bk ilm cl dm = ijk ilm aj bk cl dm =
(
a b) (
c d)
a b)i (
c d)

(
b c)
(jl km jm kl ) aj bk cl dm = aj cj bk dk aj dj bk ck = (
a c)(b d)
a d)(

Invariant parts of Tensors


Tij = TijS + TijA Symmetric and Anti-symmetric parts.


1
S
Tij = 2 Tij + Tji = TjiS Symmetric
TijA =

1
2

Tij Tji

= TjiA

Anti-symmetric

The symmetric part of the tensor can be divided into two parts:
T S = T + T
ij

ij

ij

Tij = TijS 31 Tkk ij

Tij = 13 Tkk ij

trace less

isotropic

This gives:
Tij = TijS + TijA = Tij + Tij + TijA
ui
. The anti-symmetric part describes rotation, the isotropic
xj
part describes the volume change and the trace-less part describes the deformation of a fluid element.
In the NS equations we have the tensor

Operators
(p)i =
p =

p
xi

2
p
xi xi

u
=

ui
xi

( u
)i = ijk

uk
xj

Gauss theorem (general)


Gauss theorem:

or,

I
I

F n
dS =

Fi ni dS =

F dV

Fi
dV
xi

In general we can write,


I
Z

Tijk nl dS =
Tijk dV
x
l
S
V
Example: Put Tijk nl = Tij ul nl
I
Z

(ul Tij ) dV
Tij ul nl dS =
x
l
S
V
or,
I
Z

T(
un
) dS =
( u
)T + (
u )T dV
S

Identities
Derive the identity
= (G
)F + ( G)
F ( F )G
(F )G

(F G)

(F G)

= ijk

(il jm im jl )
Gj

klm Fl Gm = ijk klm


Fl Gm = kij klm
Fl G m =
xj
xj
xj

Fi
Gj
Fj
Gi
Fl G m =
Fi G j
Fj G i =
Gj +
Fi
Gi +
Fj =
xj
xj
xj
xj
xj
xj
xj

Fi
Gj
Fj
Gi
)F + ( G)
F ( F )G
(F )G]
i
+
Fi
Gi F j
= [(G
xj
xj
xj
xj

Show that:
( F ) = 0


ijk
Fk = ijk
Fk
( F ) =
xi
xj
xi xj
Remember that ijk = jik and that
ijk



Fk =
Fk and thus all terms will cancel.
xi xj
xj xi


Fk = 0
xi xj

Example

r then
= u
Show that if u
=
. This means, show that ijk
uk = 2i .
2
xj
ijk

uk = {uk = klm l rm } = ijk


(klm l rm ) = ijk klm
l rm =
xj
xj
xj

kij klm

l rm = (il jm im jl )
l rm =
xj
xj

rm
ri
rm
i rm
l ri = { independent of x} = i
l
={
=1+1+1=3
xm
xl
xm
xl
xm
3i l il = 2i
Derive the identity
= ( F ) G
F ( G)

(F G)

ri
= il } =
xl

=
(F G)
kij

Fj
Gk
Fj
Gk
ijk Fj Gk = ijk
Fj Gk = ijk
Gk + ijk
Fj = kij
Gk + Fj ijk
=
xi
xi
xi
xi
xi
xi

Fj
Gk
F ( G)

Gk Fj jik
= ( F ) G
xi
xi

Show that:
( F )

( F ) = ( F ) F

= ijk

(il jm im jl )



klm
Fm = ijk klm
Fm = kij klm
Fm =
xj
xl
xj xl
xj xl







Fm =
Fj
Fi =
Fj
Fi = ( F ) F i
xj xl
xj xi
xj xj
xi xj
xj xj

B.2

Euler and Lagrange coordinates

Particle path (Lagranges representation)


Following a fluid particle as time proceeds the path is described by

r
=u
,
t

r(t = 0) = x
0

Streamline (Eulers representation)


Instantaneously a fluid particles displacement d
x is tangential to its velocity u
:
d
x
u

=
|d
x|
|
u|
or if we put ds = |d
x|/|
u|:

d
x
=u

ds

Exercise 1
Show that the streamlines for the unsteady flow
u = u0 ,

v = kt,

w = 0,

u0 , k > 0

is straight lines, while any fluid particle follows a parabolic path as time proceeds.
Streamline (fix t):
dx
= u = u0 x = u0 s + a
ds
dy
= u = kt y = kts + b
ds
Then we see that y(x) is a linear function,
y=
Particle path (fix x
0 ):

x
= u = u0
t
y
= u = k t
t

kt
x + d(t)
u0

x = u0 t + x0
x=

1 2
k t + y0
2

Then we see that y(x) is a parabola,


y=

1 k 2
(x 2x0 x + x20 ) + y0
2 u20

Exercise 2
a) Separate the shear flow u
= (y, 0, 0) into its (i) local translation, (ii) rotation and (iii) pure straining
parts.

Lets denote the different parts as


u
=u
T + u
R + u
D
In tensor notation it can be written as
P

ui 2 = ui 1 + ij dxj + eij dxj

(i) Translation is simply u


T = {ui 1 } = (y, 0, 0)

j 21
(ii)
j1
A
B
C
u 23 ,1
v1, 32
p1,1

0
ui
= 0 0
xj
0 0

0
0
0

ij =

1 ui
uj

2 xj
xi

u
R = d
x=

0
1
=
2
0

1
(dy, dx, 0)
2

0
0

0
0
0

1
(iii)

2


0 0
1 ui
uj
1
i
+
= 0 0
eij =
j
2 xj
xi
2
0 0 0
Inflow
1
Solid wall
u
D = ed
x = (dy, dx, 0)
n
2
0
1

M 2 b) Find the principal axes of the rate of strain tensor eij and make a schematic sketch of the decomposition


M of the flow.
1 ui
uj
ui
m
+
symmetric part of
.
eij =
2 xj
xi
xj
m+2

1
0 0
1
h
eij = 0 0
2
V
0 0 0
Vh
uEigenvalues:


uh
/2 0


0
0 = 3 + 2 /4 = (/2)(+/2) = 0. 1 = /2, 2 = /2, 3 = 0.
det(eij ij ) = /2
1
0
0

: to be updatedWe get the eigenvectors from eij kj = ki :


: new values
1
1
: old values
k1 = (1, 1, 0), k2 = (1, 1, 0), k3 = (0, 0, 1)
2
2
i1
iNow sketch the decomposition of the flow
i+1
j1
1
1
(dy, 0, 0) = (dy, dx, 0) + (dy, dx, 0)
j
2
2
j+1
iteration direction
x 1
1
1
y
n
x
y
i
j
fine grid (1)
coarse grid (2)

0.5

0.5
=

0
0.5
1
1

0.5

0.5

1
1

0
0.5

1
1

1
Exercise 3
2
ia) Sketch the streamlines for the flow [u, v, w] = [x, y, 0], where > 0.
j Streamlines in parametric form (x(s), y(s), z(s)). We have a 2D flow since w = 0.
Inflow
dx
dy
Solid wall
= x,
= y.
ds
ds
n
0
Integrate w.r.t. s,
1
x(s) = a es , y(s) = b es .

M 2 Then we see that,


x(s) y(s) = ab = c ( constant ).
M
mWe have,
c
c
m+2
x(s) =
, y(s) =
1
y(s)
x(s)
h
Streamlines:
V
Vh
dx
dy
> 0,
<0
x > 0, y > 0
u
ds
ds
uh
dx
dy
0
x < 0, y > 0
< 0,
<0
ds
ds
1
dx
dy

x < 0, y < 0
< 0,
>0
ds
ds
to be updated
dx
dy
: new values
x > 0, y < 0
> 0,
>0
: old values
ds
ds
i1
dx(s)
dy(s)
= x and
= y. This gives the direction of the flow:
iRemember that
ds
ds
i+1
j 1
1
j
j +1
ration direction 0.5
x
y
0
n
x
y
0.5
i
j
fine grid (1)
1
coarse grid (2) 1

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

b) The concentration of a scalar is c(x, y, t) = x2 yet . Does this concentration change for a particular
fluid particle in time?
Use the material time-derivative
c
c
c
Dc
=
+u
+v
= x2 yet + x2xyet yx2 et = (1 + 2 1)x2 yet = 0.
Dt
t
x
y
Thus the concentration c(x, y, t) is constant for a fluid particle.

c) Using the alternative (Lagrangian) description of the flow, show that


u

D
u
=

Dt
t
and write the concentration as c = c(x0 , y0 , t).

Particle path:

= u = x x = x0 et

t
y

= v = y y = y0 et

t
u
x

=
= 2 x0 e t = 2 x

t
t t
v
y

=
= 2 y0 et = 2 y
t t t
Du
u
u
u
=
+u
+v
= 2 x
Dt
t
x
y
Dv
v
v
v
=
+u
+v
= 2 y
Dt
t
x
y

So we have

D
u
=

Dt
t
Both terms describe the acceleration of a fluid particle.
Now rewrite the concentration

c(x, y) = x2 yet = x20 e2t y0 et et = x20 y0 = c(X)


| {z } | {z }
x2

The concentration foes not change as we follow a fluid particle.

independent of t!

B.3

Reynolds transport theorem and stress tensor

Exercise 1
a) Consider a fixed closed surface S in a fluid. Show that conservation of mass implies

+
(ui ) = 0.
t
xi
The change of mass in the volume is described by the contribution from the mass flux and the change in
density and conservation means that this should be zero
I
Z

dV +
ui ni dS = 0
S
V t
Using Gauss Theorem we can rewrite the surface integral
I
Z

ui ni dS =
(ui ) dV.
S
V xi
We have

+
(ui )} dV = 0.
t
xi

This must be true for an arbitrary volume and this implies

+
(ui ) = 0.
t
xi
b) Show that this can be written
D
ui
+
= 0,
Dt
xi

ui

ui

(ui ) =
ui +
= ui
+
xi
xi
xi
xi
xi
This gives

D
ui
ui
+ ui
+
=
+
t
xi
xi
Dt
xi
|
{z
}
D
Dt

c) Explain what the following means


ui
=0
xi

D
=0
Dt

If we are following a fluid particle, the density and volume are constant. But in a fix position, the density
can change

= ui
t
xi
Remember the decomposition of

ui
xj

in its invariant parts





ui
1 ui
uj
1 uk
1 ui
uj
=
+

ij +

+
xj
2 xj
xi
3 xk
2 xj
xi
|
{z
} |
{z
}
deformation eij

rotation ij

1 uk
ij
3 x
| {zk }

volume change eij

Exercise 2
a) Use Reynolds transport theorem to provide an alternative derivation of the conservation of mass equation

+
(ui ) = 0.
t
xi
Reynolds Transport Theorem:
D
Dt

V (t)

Put Tijk = ,
D
Dt

Tijk dV =

dV =
V (t)

V (t)

V (t)


Tijk

+
(ul Tijk ) dV
t
xl

+
(ui ) dV
t
xi

No flow through the surface. Valid for all V (t):

+
(ui ) = 0
t
xi
b) Use this result to show
D
Dt

V (t)

Fijk dV =

V (t)

DFijk
dV
Dt

Put Tijk = Fijk in Reynolds transport theorem


D
Dt

V (t)

V (t)

(Fijk ) +
(ul Fijk ) dV =
t
xl


Fijk
Fijk

+ Fijk
+ Fijk
(ul ) + ul
dV =
t
t
xl
xl
V (t)





Z
Fijk
Fijk

Fijk
dV =
+
(ul ) +
+ ul
t
xl
t
xl
V (t)
|
{z
}
|
{z
}
Z

Fijk dV =

=0

V (t)

DF

ijk
Dt

DFijk
dV
Dt

Exercise 3
Show that the stress tensor Tij = pij for the flow
u
=
x
,
where
is constant.
Tensor notation:
ui = ikl k xl
The stress tensor:
Tij = pij +


uj
ui
+
xj
xi

(ikl k xl ) +
( x )
xj
xi jkl k l


x
x
pij + k ikl l + jkl l =
xj
xi

pij +

pij + k (ikl lj + jkl li ) =


pij + k (ikj + jki ) =
pij + k (ijk + ijk ) = pij

The stress tensor Tij


The equation for the stress tensor was deduced by Stokes in (1845) from three elementary hypotheses.
Tij = pij + 2 eij
Writing this on the form,
Tij = pij + ij
the following statements should be true for the viscous stress tensor ij in a Newtonian fluid.
ui
(i) each ij should be a linear combination of the velocity gradients
.
xj
(ii) each ij should vanish if the flow involves no deformation of fluid elements.
ui
(iii) the relationship between ij and
should be isotropic as the physical properties of the fluid
xj
are assumed to show no preferred direction.

Exercise 4
If ij is a linear function of the components of eij then it can be written
ij = cijkl ekl
It can be shown that the most general fourth order isotropic tensor is of the form
cijkl = Aij kl + Bik jl + Cil jk
where A, B and C are scalars.
a) Use this to show that
ij = ekk ij + 2eij
where and are scalars.
ij = cijkl ekl = (Aij kl + Bik jl + Cil jk )ekl = Aij ekk + Beij + Ceji
Say that A = and that B = C = , then since eij = eji we have
ij = ekk ij + 2eij
1
b) Show that if we have a Newtonian fluid where p = Tii then
3




ui
uj
2 uk
Tij = p +
ij +
+
3 xk
xj
xi
We have
Tij = pij + ij = pij + ekk ij + 2eij
Then
This gives =

or

23 ,

Tii = 3p + 3eii + 2eii = 3p + (3 + 2)eii


and thus
2
Tij = p ekk ij + 2eij
3




2 uk
ui
uj
Tij = p +
ij +
+
3 xk
xj
xi

h
V B.4
Vh

Rankine vortex and dimensionless form

The Rankine vortex

uh A simple model for a vortex is that it is a rigid body rotation within a core, and a decay of angular velocity
0 outside. This can be described by
1

r, r < a,
: to be updated
u = a2
ur = u z = 0

, r > a,
: new values
r
: old values
i 1and is called a Rankine vortex.
i
i+1
3
j1
j
2
j+1
2
iteration direction
x
y
1
n
x
y
i
j
fine grid (1)
coarse grid (2)

1
0

Figure B.1: Velocity and vorticity in a Rankine vortex with = a = 1.

Exercise 1
a) Find the pressure inside and outside of a Rankine vortex:
We use Eulers equations for incompressible flow ( neglecting viscous effects ):

u
1
D
= p + g
Dt

Eulers equations

u
=0
u

D
u
=
+(
u )
u
Dt
t
|{z}
=0

We are working with cylindrical coordinates, use the appendix formulas


u
= ur
(
u )
u=

+
+ uz
r
r
z

u
u u
u
e
u u
u2
(u e ) =
e +
u =
e er
r
r
r |{z}

r
r
=
er

p =

p
1 p
p
er +
e +
ez
r
r
z

Insert this into the Euler equations:




u u
u2
1 p
1 p
p
e er =
er +
e +
ez g
ez
r |{z}

r
r
r
z
=0

Look at the different components:

er :
0=

e :

u2
1 p
=
r
r

1 1 p
p = p(r, z) only.
r
0=

ez :

1 p
g
z

Solve for the pressure when r < a:


2 r =

er :

ez :

1 p
r

p
= g
z

So we have the pressure:


p(r, z) = 2

p = 2

r2
+ f (z)
2

f (z) = gz + C1

r2
gz + C1
2

r<a

Solve for the pressure when r > a:

er :

2 a4
1 p
=
r3
r

ez :

p
= g
z

p=

2 a4
+ f (z)
2 r2

f (z) = gz + C2

So we have the pressure:

2 a4
gz + C2 r > a
2 r2
Now determine the difference between the constants by evaluation at r = a
p(r, z) =

2 a2
2 a2
gz + C1 =
gz + C2
2
2

C2 C1 = 2 a2

b) Determine the pressure difference p between r = 0 and r


p = p p0 = gz + C2 (gz + C1 ) = C2 C1 = 2 a2
c) Now consider a free surface at atmospheric pressure p0 .
Find the difference in z between r = 0 and r
(
2 2
r = 0 : p0 = 2 a gz0 + C1
r:

2 2

p0 = 2 a gz + C2

Determine the shape of the free surface:


(
2 2
p0 = 2 r gz + C1
2 4

p0 = 2r2a gz + C2

z z0 =

C2 C 1
2 a2
=
g
g

C1 p0
2 r2
2g +
g
2 4
p0
= 2gra2 + C2g

r < a z r2 z =
r>a z

1
r

Say that z = 0 at r = 0 then C1 = p0 and we get


z=
and
z=
So we have

2 r2
2g

r<a



2 a4
C2 C 1
2 a4
2 a2
2 a2
a2
+
=

+
=
1

2gr2
g
2gr2
g
g
2r2
2 r2
2g  r < a 
z(r) = 2 a2
a2
g
1 2r
2

r > a.

r>a

i
i+1
j1
j
j+1
iteration direction
x
y
n
x
y
i
j
fine grid (1)
coarse grid (2)

0.1
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01

0
0

0.5

1.5
r

2.5

Figure B.2: The free surface of a Rankine vortex with = a = 1 and g = 9.82.
2 r2
2g  r < a 
z(r) = 2 a2
a2
g
1 2r
2

Exercise 2

r > a.

Solving the 2D flow around a cylinder involves solving


1
u

+ (
u )
u = p + 2 u
,
t

u
= 0,

with the boundary conditions


u
= 0 on x2 + y 2 = a2

u
(U, 0, 0) as x2 + y 2 .

Rewrite this problem in dimensionless form using the dimensionless variables


x
0 = x/a u
0 = u
/U

p0 = p/U 2

t0 = tU/a.

a
Note that the scaling x
0 = x/a implies 0 = a and t0 = tU/a gives 0 =
.
t
U t
Change to dimensionless variables
U 2 u
0
U2 0
U 2 0 0 U 0 2 0
+
(
u 0 )
u0 =
p + 2 u

0
a t
a
a
a
Divide by U 2 /a
u
0
+ (
u0 0 )
u0 = 0 p0 +
t0

U
a2
U2
a

0 u
0

|{z}
1
Re

U2
Inertial forces
Ua
The Reynolds number is Re = a =
=
U
Viscous forces

a2

u
0
1 02 0
+ (
u0 0 )
u0 = 0 p0 +
u

t0
Re
The continuity condition
u
=0

U 0 0
u
=0
a

0 u
0 = 0

The boundary conditions


u
= 0 on x2 + y 2 = a2

Uu
0 = 0 on a2 x0 + a2 y 0 = a2

u
(U, 0, 0) as x2 + y 2

u
0 = 0 on x0 + y 0 = 1.
2

Uu
0 (U, 0, 0) as a2 x0 + a2 y 0
2

u
0 (1, 0, 0) as x0 + y 0
The solutions of this problem will depend on x
0 , Re and t0 only, and thus the solution of this problem is
the same for a specific Re independently of the individual values of U , a and .

Exercise 3
Show that the net viscous force per unit volume is proportional to the spatial derivative of vorticity, i.e.
ij
k
= ijk
xj
xj
and discuss its implication for flows with uniform vorticity (as in solid-body rotation).


 2

ij

ui
uj
ui
2 uj
2 ui
=
+
=
+
=
xj
xj xj
xi
xj xj
xj xi
xj xj


k

um
2 um
2 um
ijk
= ijk
klm
= kij klm
= (il jm im jl )
=
xj
xj
xl
xj xl
xj xl

 2
2 ui
2 ui
uj

xj xi
xj xj
xj xj
Thus

ij
k
= ijk
xj
xj

The net viscous force vanishes when the vorticity is uniform, since no deformation exists.

j+

j B.5

Exact solutions to Navier-Stokes

1
2

j
j 1Exercise 1
APlane Couette flow:
BConsider the flow of a viscous fluid between two parallel plates at y = 0 and y = h. The upper plane is
Cmoving with velocity U . Calculate the flow field.
u 23 ,1
v1, 32
Assume the following:
p1,1 Steady flow:

0
=0
1
t
2Parallel flow:
ui
i
=0
v = 0,
j
x
InflowTwo-dimensional flow:

Solid wall
w = 0,
=0
z
n
0 No pressure gradient:
p
1
=0
xi

M 2 NavierStokes streamwise momentum equation:


M
u
1
m
+ (
u )u = p + 2 u
t

m+2
1We have
h
2u
u
=0
= A u = Ay + B
V
y 2
y
Vh
Boundary conditions:
u
u(0) = 0 B = 0, u(h) = U A = U/h
uh
0 So we get:
Uy
1
u(y) =
h

: to be updated
: new valuesExercise 2 (i)
: old valuesPlane Poiseuille flow: ( Channel flow )
i 1Consider the flow of a viscous fluid between two solid boundaries at y = h driven by a constant pressure
igradient p = [P, 0, 0]. Show that
i+1
P 2
j1
u=
(h y 2 ), v = w = 0.
j
2
j+1
iteration direction
Y
x
y
1
n
x
X
y
i
j
Z
fine grid (1)
coarse grid (2)
Figure B.3: Coordinate system for plane Poiseuille flow
NavierStokes equations:

1
u
+ (
u )
u = p + 2 u

u
= 0.

j+

Boundary conditions:

u
(y = h) = 0
j
j 1
u

We are considering a stationary flow and thus


= 0.
A
t

The
constant
pressure
gradient
implies
u

=
u

(y).
Changes
of u
in x, z would require a changing pressure
B
gradient
in
x,
z.
C
v
u 23 ,1
The continuity equation u
reduces to
= 0. The boundary condition v(y = h) = 0 then implies
v1, 32
y
p1,1 v = 0.
Lets study the z component of the NavierStokes equations:
0
1
w
2w
v
=

w = c1 y + c2
2
|{z} y
y 2
i
=0
j
InflowThe boundary conditions w(y = h) = w(y = h) = 0 imply c1 = c2 = 0 and thus w = 0.
Solid wall
= [u(y), 0, 0].
n We can conclude that u
0
1 Lets study the x component of the NavierStokes equations:

P
2u
u
P
P 2
M2
0=
+ 2
= y + d1 { = } u(y) =
y + d1 y + d 2
M

y
y

2
m
m + 2The boundary conditions at y = h give
1
P 2
P
0=
h + d1 h + d2 and 0 = h2 d1 h + d2
h
2
2
V
Vh
P 2
uWe can directly conclude that d1 = 0 and this gives d2 = 2 h and thus
uh
0
P 2
u=
(h y 2 ), v = w = 0.
1
2

to be updated
Exercise 2 (ii)
: new values
: old valuesPoiseuille flow: ( Pipe flow )
i 1Consider the viscous flow of a fluid down a pipe with a cross-section given by r = a under the constant
p
i
pressure gradient P = . Show that
i+1
z
j 1
P 2
j
uz =
(a r2 ) ur = u = 0.
4
j +1
1
2

ration direction
x
y
1
n
x
y
i
j
fine grid (1)
coarse grid (2)

Figure B.4: Coordinate system for Poiseuille flow


Use the NavierStokes equations in cylindrical coordinates


ur
u2
1 p
ur
2 u
2
+ (
u )ur
=
+ ur 2 2
t
r
r
r
r



u
ur u
1 p
2 ur
u
2
+ (
u )u +
=
+ u + 2
2
t
r
r
r
r
uz
1 p
+ (
u )uz =
+ 2 uz
t
z
1
1 u
uz
(r ur ) +
+
= 0.
r r
r
z
p
p
We know that
= 0 and
= 0 and can directly see that ur = u = 0 satisfy the two first equations.

r
From the continuity equation we get
uz
=0
z

uz = uz (r, )

only.

Considering a steady flow we get from the NavierStokes equations


(
u )uz = uz
But we know that

uz
z

uz

uz
1
= P + 2 uz
z

uz
= 0 from the continuity equation. We get
z
2 u z =

Integrate
r

1 uz
P
(r
)=
r r r

uz
P
= r 2 + c1
r
2

uz
P
(r
)= r
r r

uz
P
c1
=
r+
r
2
r

Integrating again we get


uz =

P 2
r + c1 ln(r) + c2
4

using the boundary conditions uz (r = 0) < c1 = 0

We also have uz = 0 at r = a and this gives c2 =


uz =

P a2
and we have
4
P 2
(a r2 )
4

Exercise 3
Calculate the asymptotic suction boundary layer, where the boundary layer over a flat plate is kept parallel
of a steady suction V0 through the plate.
Assumptions.
Two-dimensional flow:

= 0,
z

Parallel flow:

= 0,
x

Steady flow:

Momentum equations:

w=0

Continuity satisfied

=0
t

 2

u
u
u
1 p
u 2u
+u
+v
=
+
+
t
x
y
x
x2
y 2
 2

v
v
v
1 p
v
2v
+u
+v
=
+
+
t
x
y
y
x2
y 2

Normal momentum equation gives

p
=0
y

Boundary conditions:
y=0:

v = V0

u = 0,

y:

u U

Continuity gives
u v
+
=0
x y

v = V0

Streamwise momentum equation at y


U
1 p
2 U
=
+
y
x
y 2

V0

p
=0
x

Resulting streamwise momentum equation


V0

u
2u
= 2
y
y

2u
V u
= 0
y 2
y

Characteristic equation
2 =

V0

1 = 0, 2 =

V0

u(y) = A + BeV0 y/

Exercise 4
Two incompressible viscous fluids flow one on top of the other down an inclined plane at an angle . They
both have the same density and the viscosities 1 and 2 . The lower fluid has depth h1 and the upper
h2 . Assuming that viscous forces from the surrounding air is negligible and that the pressure on the free
surface is constant, show that
1 g sin()
u1 (y) = [(h1 + h2 )y y 2 ]
.
2
1
Make the ansatz u
1 = [u1 (y), 0, 0] and u
2 = [u2 (y), 0, 0]. The continuity equation
u v
+
=0
x y

gives

v
= 0 v = c and the boundary condition at y = 0 gives v = 0.
y

Layer 1:
NS ey : 0 =

1 p1
g cos()
y

p1 = g cos()y + f1 (x)

1 0
d 2 u1
NS ex : 0 = f1 (x) + 1 2 + g sin()

dy
Layer 2:
NS ey : 0 =

1 p2
g cos()
y

f1 (x) = c1

p2 = g cos()y + f2 (x)

1 0
d 2 u2
NS ex : 0 = f2 (x) + 2 2 + g sin()

dy

f2 (x) = c2

The pressure at the free surface y = h1 + h2 is p0 :


p0 = g cos()(h1 + h2 ) + f2 (x)

f2 = p0 + g(h1 + h2 ) cos()

f2 = 0

The pressure is continuous at y = h1 :


p0 + gh2 cos() = gh1 cos() + f1 (x)

f1 = p0 + g(h1 + h2 ) cos()

This gives the pressure:


p1 (y) = p2 (y) = p(y) = g cos()y + p0 + g cos()(h1 + h2 )

f1 = 0

We now have two momentum equations in x:


0 = 1

d 2 u1
+ g sin()
dy 2

(1)

0 = 2

d 2 u2
+ g sin()
dy 2

(2)

And four boundary conditions:


BC1: No slip on the plane:

u1 (0) = 0

BC2: No viscous forces on the free surface:


BC3: Force balance at the fluid interface:

du2
|y=h1 +h2 = 0
dy

du1
du2
|y=h1 = 2
|y=h1
dy
dy

BC4: Continous velocity at the interface: u1 |y=h1 = u2 |y=h1


(1)

du1
g
= y sin() + c11
dy
1

u1 =

g 2
y sin() + c11 y + c12
2 1

(2)

du2
g
= y sin() + c21
dy
2

u2 =

g 2
y sin() + c21 y + c22
2 2

BC1 c12 = 0
g
g
BC2 2 ( (h1 + h2 ) sin() + c21 ) = 0 c21 = (h1 + h2 ) sin()
2
2
g
g
2
g
BC3 1 ( y sin()+c11 ) = 2 ( y sin()+c21 ) { = } c11 =
c21 = (h1 +h2 ) sin()
1
2
1
1
g 2
g
g 2
g
BC4
h1 sin() + (h1 + h2 ) sin()h1 =
h1 sin() + (h1 + h2 ) sin()h1 + c22
2 1
1
2 2
2
 2


h1
1
1
c22 = g sin()
(h1 + h2 )h1

2
2
1
This gives us the velocities,
g 2
g
g sin()
1
y sin() + (h1 + h2 ) sin()y =
[(h1 + h2 )y y 2 ]
2 1
1
1
2
 2


g sin() 2 g sin()
h1
1
1
y +
(h1 + h2 )y + g sin()
(h1 + h2 )h1

u2 (y) =
2 2
2
2
2
1
 2


g sin()
1
h
1
1
=
((h1 + h2 )y y 2 ) + g sin() 1 (h1 + h2 )h1

2
2
2
2
1
u1 (y) =

The velocity in layer 1 does depend on h2 but not on the viscosity in layer 2. This is due to that the depth
is important for the tangential stress boundary condition at the interface but the viscosity is not. There is
no acceleration of the upper layer and thus the tangential stress must be equal to the gravitational force
on the upper layer which depends on h2 but not on 2 .

B.6

More exact solutions to Navier-Stokes

Exercise 1
Oscillating RayleighStokes flow (or Stokes second problem).
a) Show that the velocity field u
= [u(y, t), 0, 0] satisfies the equation
u
2u
= 2
t
y
Consider the NavierStokes equation in the x direction:
 2

u
u
u
1 p
u 2u 2u
u
+u
+v
+w
=
+
+
+
t
x
y
z
x
x2
y 2
z 2
With the current velocity field only terms with y derivatives will remain since there can be no change in
the other directions. Further more, the streamwise pressure gradient has to be zero since the streamwise
velocity far from the wall is constant, namely zero.
u
2u
= 2
t
y
b) Show that the velocity field is
u(y, t) = U e

Make the ansatz: u = < f (y)e

i t

Insert into the equation:

ky

cos(ky t),

where k =

= f (y) cos( t).


if (y)ei t = f 00 (y)eit

f 00 (y)
2
Introduce k =

i
f (y) = 0

i
=0

with f (y) = ey gives


r
r 

i
1+i

=
=

,
2
f (y) = Aeyk(1+i) + Beyk(1+i) ,

f (y) 0 as y A = 0

ui1/2,j
1 We have
ih
2




V
i
yk(1+i) iwt
ky i(tky)
1
V
h
u = < Be
e
= < Be
e
= Beky cos(t ky) = Beky cos(ky t)
i+ 2
u
i+1
i +u32h Boundary condition at y = 0
j +210
1
At y = 0 we have u = U cos(t) B = U
j
1

j2
So we have
: to be updated
r
j1

ky
: new values
u(y, t) = U e
cos(ky t), where k =
A
2
: old values
B
i1
C
u 23 ,1i 5
iv+
1
1, 32 4.5
j p 1
1,1
j 4
j+1
1 3.5
iteration direction
x 3
yi 2.5
j
Inflow 2
n
Solid wall 1.5
n
0 1
1i 0.5
j

fine grid (1)


2 0
0.2
0.4
0.6
0.8
0
1
coarse grid M
(2) 1 0.8 0.6 0.4 0.2
x
M
m
c) How thick is the boundary layer thickness?
m+2
1
hIf we define the thickness of the oscillating layer as the position where u/U = 0.01 we get that
r
r
V
2

k
Vh
e
= 0.01 k 4.6 4.6
6.5

u
uh d) Consider instead the oscillating flow U = U cos(t) over a stationary wall. This will simply result in a

0 change of frame of reference to one following the plate instead. If we consider the solution to the previous
1 problem and look at it in this new frame of reference we get

r
: to be updated

ky
u(y, t) = U cos(t) U e
cos(ky t), where k =
: new values
2
: old values
i 1The solution now looks like this
i 5
i+1
j 1 4.5
j 4
j+1
3.5
iteration direction
x 3
y 2.5
n 2
1.5

1
i
j 0.5
fine grid (1)
0
coarse grid (2) 1.5

0.5

0
x

0.5

1.5

Exercise 2
Consider a long hollow cylinder with inner radius r1 and a concentric rod with radiusr0 inside it. The rod
is moving axially with velocity U0 .
a) Find the velocity field of a viscous fluid occupying the space between the rod and the cylinder.
Assumptions Steady flow:

=0
t

Parallel flow and symmetry:

= 0,
z

u = uz (r)ez ,

=0

No streamwise pressure gradient:

p
=0
z
We can directly see that ur = u = 0 satisfy the two first equations. The streamwise momentum equation
reduces to
(u )uz = 2 uz
where

uz
u uz
uz
+
+ uz
=0
(u )uz = ur
r
r
z




1
uz
1 2 uz
2 uz
1
uz
2 u z =
r
+ 2
+
=
r
r r
r
r 2
z 2
r r
r

We end up with




uz
r
=0
r
r

Integrate twice
r

uz
=A
r

uz = A ln r + B

Boundary conditions


uz (r0 ) = U0
uz (r1 ) = 0

= A ln r0 + B
= A ln r1 + B

U0 = A(ln r0 ln r1 )
uz (r) =

B = A ln r1
A=

U0
ln rr01

ln rr1
U0
U0
ln
r

ln
r
=
U
1
0
ln rr01
ln rr01
ln rr01

b) With what force does one have to pull a rod with length L? Neglect end effects.
Shear stress
rz =

uz
U0
=
r
r ln rr10

Force
F = 2r0 Lrz (r0 ) =

2LU0
ln rr01

B.7

Axisymmetric flow and irrotational vortices

Axisymmetric flow
Consider incompressible and rotationally symmetric flow with no mass source at the symmetry axis. The
velocity component in the direction of the axis of symmetry and the vorticity is given.
uz = z

and

= (r) ez .

1) Compute ur (r, z) and u (r, z) for the two cases when


a) (r) = 0
The vorticity is given,

Look at the different components:

b) (r) = 0 er


e
1 r

=u
= r
r
ur

/a2


ez

z
uz

r
e

r u

1 uz
u

=
er = 0
r
z

er :
e :

and

u = u (r)

ur
uz

=
e = 0 ur = ur (r)
z
r


1 (r u ) ur
ez :

=
ez = (r)
r
r

The rotational symmetry implies

= 0. From the ez component we get,

(r u ) = r (r)
r

(B.1)

We also have the incompressibility condition:


u
=

1
1 u uz
+
=0
(rur ) +
r r
r |{z}

z
|{z}

=0

From this we get,

(rur ) = r
r
Equation (2) is valid both for case a) and b). Case a) (r) = 0

(r u ) = 0
r

(1)
(2)

(rur ) = r
r

(B.2)

u =

A
r

integrate the rhs rur =

r B
+
2
r
There should be no mass source at r = 0. The mass flow is,

Q=

ur =

2
0

ur r d = 2 B 2

r2
.
2

When r 0 then Q 2 B which gives B = 0.


ur =

r
2

for both a) and b)!

r2
+B
2

The circulation is,

=
Case b) (r) = 0 er

/a

2
0

u r d = 2A A =

,
(1)

2
2

(r u ) = r 0 er /a
r

Integrate the right hand side:


r u =

0 a2 r2 /a2
e
+C
2

Look at the circulation,


(r) =

u =

u r d = 2(
0

for r = 0 we get

This gives
u =

0 = 2(

0 a2 r2 /a2 C
e
+
2r
r

0 a2 r2 /a2
e
+ C)
2

0 a 2
+ C)
2

C=

0
a2
+ 0
2
2



2
2
0
a2
++ 0
1 er /a
2r
2r

c) Consider circles C(t) following the flow with a radius R(t) and position Z(t). Compute R(t) and Z(t).
Can any of the flow cases be inviscid?
R(0) = R0
1
dR
= ur (R) = R
dt
2
dZ
= uz (Z) = Z
dt

Z(0) = Z0
1

R(t) = R0 e 2 t

Z(t) = Z0 e t

The circulation is then,


(t) =

u (R)R d = 2Ru (R) =


0

a)



2
2
0 + 0 a2 1 eR /a

b)

The circulation is constant for a) but not for b). This means that the flow in a) can be inviscid.

Inviscid and irrotational vortices


Consider a circular flow with u
= u (r)
e . Which vortices are inviscid and which vortices are irrotational?
The NavierStokes equation for u




u
1
1
u
u
=
+
r
2
t
r
r r
r
r
For inviscid flow we require,



1
u
u
r
2 =0
r r
r
r

Make the ansatz u = rn ,

1
1
(rnrn1 ) rn2 = n2 rn1 rn2
r r
r
We get the inviscid flow,
u (r) =

Ar
|{z}

solid body rotation

The vorticity is,

=u
= {A.32} =

=0

(ru ) = 0
r

n2 1 = 0

B
r
|{z}

irrotational

1
(ru ) ez
r r

Conclusion:
Irrotational inviscid
Inviscid ; irrotational

u =

C
r

n = 1

Exercise 3
Show that the inviscid vorticity equation
D

= (
)
u
Dt
reduces to the equation

 
D
=0
Dt r

in the case of axisymmetric flow:


= ur (r, z, t)
u
er + uz (r, z, t)
ez
The vorticity in an axisymmetric flow
=

=u


ur
uz
=

e
e
z
r
|
{z
}

Study the right hand side of the inviscid vorticity equation






(
)
u=
ur (r, z, t)
er + uz (r, z, t)
ez =
(
) =
r
r
ur

er uz

ez

r + ur
z + uz

e
e
= ur e
+
r |{z}

r |{z}

r |{z}

r |{z}

r
=0

=
e

=0

=0

The left hand side of the inviscid vorticity equation gives




 
D

=
+ (
u )
= {
=
e } =
+ ur
+ uz
e
Dt
t
t
r
z
This gives that the inviscid vorticity equation now is



+ ur
+ uz
= ur
t
r
z
r
Multiply by

1
r

 



1

+
ur
+ uz
2 ur = 0
t r
r
r
z
r

Notice that

1
ur = ur
2
r
r r

and that uz

1
=0
z r

This means we can write


 





1

1
+
ur
+ uz
+ ur
+ uz
=0
t r
r
r
z
r
z r
And thus we have

 


 

1


D
+
ur
+ uz
=
=0
t r
r
r
z r
Dt r

B.8

Vorticity equation, Bernoulli equation and streamfunction

Solid body rotation


Consider the flow in a uniformly rotating bucket with velocity
u = (r, 0, 0)
a) Use Bernoulli equation to determine the free surface: What is wrong?
Bernoulli:
p 1 2
+ |u| + gz = C along streamlines
2
This gives the surface of constant pressure
C p0
2 r2

g
2g

z=

The free surface is highest in the center of the bucket, something is wrong. Bernoulli theorem is valid along
a streamline in a steady ideal fluid. If the flow had been irrotational, it would have been valid everywhere.
= (0, 0, 2).
But now u
b) Use Euler equation to determine the free surface:
1
Du
= p + g
Dt

this gives
er :

u2
1 p
=
r
r
e :

ez : 0 =

p
= 2 r
r

0=

1 p
r

p
=0

1 p
g
z

Thus
p(r, z) =

f
= g
z

p=

1 2 2
r + f (z)
2

f = gz + p0

1 2 2
r gz + p0
2

This gives the free surface where p = p0


z=

2 r2
2g

Flow over a hill (Exam 20020527 question 2)


A hill with the height h has the shape of a half circular cylinder as shown in Figure 1. Far from the hill the
wind U is blowing parallel to the ground in the x-direction and the atmospheric pressure at the ground
is p0 .
a) (5) Assume potential flow and show that the stream function in cylindrical coordinates is of the form
= f (r) sin ,
where f (r) is an arbitrary function. Calculate the velocity field above the hill.
b) (3) Derive an equation for the curve with constant vertical wind velocity V .
c) (2) Assume that the density and the gravitational acceleration g is constant. Calculate the atmospheric
pressure at the top of the hill.
a) The stream function satisfies continuity:
ur =
The flow is irrotational:
u
=0

1
,
r

u =

2
1 2
+r 2 +
= 0 (1)
r
r
r 2

ration direction
x
y
1
n
x
y
i
j
fine grid (1)
coarse grid (2)

Figure B.5: Streamlines above a hill with h = 100m and U = 5m/s. A paraglider pilot with a sink of
1m/s will find lift in the area within the dotted line, while soaring along the hill.
Introduce the ansatz = f (r) sin into equation (1):
1
f 0 sin + rf 00 sin f sin = 0
r

1
f 0 + rf 00 f = 0
r
Make the ansatz f = r n :

1
nrn1 + rn(n 1)rn2 rn = 0
r
n + n2 n 1 = 0

So we have
=

B
Ar +
r

n = 1

sin

We need two boundary conditions.


1. Free stream:
Ar sin = U r sin

A = U

2. Streamline on the hill surface:


U h +

B
=0
h

So we have:
= U

h2
r
r

B = U h2


sin

Now we can calculate the velocity field above the hill:




1
h2
= U 1 2 cos
r
r



h2
u =
= U 1 + 2 sin
r
r
ur =

b) Constant vertical wind velocity is described by:






h2
h2
h2
V = ur sin + u cos = U 1 2 cos sin U 1 + 2 sin cos = 2U 2 sin cos
r
r
r
r
U
r = h 2
sin cos
V
c) Use Bernoulli equation with free stream pressure p0 at the ground:
1 2
1
po + U
= p + (2U )2 + gh
2
2

3 2
p = po U
gh
2

Stokes stream function:


Consider a 2D incompressible flow
u v
+
= 0.
x y

u
= 0 or
Define the stream function such that:
u=
This means

v=

u v
2
2
+
=

= 0,
x y
xy yx

so continuity is fulfilled. Now we can write

And


e
x
u
=
ez = x
0

=u
= x

y
e

y

x

Irrotational flow



z 
e


,
,0
z =
y
x

y
e



z 
e

2

=
0,
0,

= 2
ez
z
x2
y 2

0
2 = 0

In spherical coordinates for axisymmetrical flow, define


ur =

r2

1
sin

u =

1
r sin r

Incompressibility is still valid


u
=0
Velocity
u
=
Vorticity

=
Irrotational

e
r sin

 2


1
sin
1
+
r sin r2
r2 sin

2 sin
+ 2
r2
r

1
sin

=0

Flow around a sphere


Consider a sphere with = 0 at r = a, compute the irrotational velocity distribution when the velocity of
the freestream at infinity is U:
1
r U r2 sin2
2
ur U cos u U sin
Make the ansatz:
= f (r) sin2
For an irrotational flow we get
f 00

2
f =0
r2

From infinity we get


A=
On the sphere

1 2 B
Ua +
=0
2
a

This gives
=
The slip velocity on the sphere is

The radial velocity is ur = 0.

f = Ar 2 +

B
r

1
U
2

1
B = U a3
2



1
a3
U r2
sin2
2
r
3
u = U sin
2

B.9
Flow around a submarine and other potential flow problems
: to be updated
: new valuesExercise 1
: old values
i 1The flow around a submarine moving at a velocity V can be described by the flow caused by a source and
a sink with strength Q at a distance 2a from each other.
i
i+1
x
V
j1
j
j+1
Submarine
iteration direction
x
y
Q
-Q
1
p1,1
z
n
y
y
i
j
fine grid (1)
coarse grid (2)

Figure B.6: Coordinate system for submarine problem


a) If one wants to construct a pressure sensor that will register an approaching submarine at a distance L,
what sensitivity is needed for the sensor? Assume an ideal fluid and that 2a = 80 m, Q = 915 m3 /s, U = 8
m/s, L = 200 m and = 1000 kg/m3 .

Use a potential flow description


u
= ,

,
x

u=

v=

The flow is always irrotational due to the definition of the velocity potential
=u
= = 0,

curl(grad)=0

For incompressibility we get


u
= =

()i
= = 0
xi

The equation is linear and thus superposition can be used. We have freestream plus 3D source plus sink
=

Uz
|{z}

freestream

Q
Q
+
4 r1 4 r2
| {z } | {z }
source

sink

The first term is in cylindrical coordinates (R, , z) and the two last are in two different spherical coordinate
systems with origin in za and z+a, respectively. Transform the two second terms to cylindrical coordinates
= Uz +
Velocity

Q
Q
p
+ p
(z + a)2 + R2
4 (z a)2 + R2

eR +
e +
ez =
R
R
z




QR
QR
Q(z + a)
Q(z a)
eR

+
ez U +

4((z + a)2 + R2 )3/2 4((z a)2 + R2 )3/2


4((z + a)2 + R2 )3/2 4((z a)2 + R2 )3/2
u
= =

We need to know the distance, b, from the point source to the stagnation point on the submarine nose.
Thus we need to know the length of the submarine.

b) How long is the submarine?


Compute where uz = 0 for R = 0
U+

Q
4

1
1

(z + a)2
(z a)2

=0

4U
(z a)2 (z + a)2
4az
=
= 2
Q
(z a)2 (z + a)2
(z a2 )2

Solving this system gives z = 43.01 m, z = 36.99 m, where the second solution lies inside the submarine.
The length is then 2 43.01 86 m and b = 3.01 m.
Now we continue to solve a)
Use Bernoulli equation to determine the pressure fluctuations at z = L b a, R = 0
1
1
p + |
u|2 = p + U 2
2
2
Evaluate u
noticing that uR = 0
u
(z = L b a, R = 0) = ez

Q
Q
+
U+
4(L + b)2
4((L + b + 2a)2

Inserting the given values gives |


u| = 7.99914 m/s and we get
p p =

1
2
(U 2 |
u|2 ) 6.86 N/m 0.07 mbar
2

c) How wide is the submarine?


To get this we need to compute the shape of the submarine. The stream function is constant along
streamlines and is useful for this. In spherical coordinates the stream function is defined as
ur =

1
r2 sin

1
r sin r
Transformation between cylindrical and spherical coordinates
u =

ur = uR sin + uz cos
R = r sin , z = r cos
Our velocity field gives


1
Q
1
1
= sin r sin
2
+
r2 sin
4
(r2 + a2 + 2ar cos )3/2
(r + a2 2ar cos )3/2



Q
r cos + a
r cos a

cos U +
4 (r2 + a2 + 2ar cos )3/2
(r2 + a2 2ar cos )3/2


Q
r + a cos
r a cos
= U cos +
2
4 (r2 + a2 + 2ar cos )3/2
(r + a2 2ar cos )3/2

This is difficult to integrate.

p1,1 Simplify to a Rankine body by neglecting the sink and say that a = 0
Q 1
1
1
ur = U cos +
= 2

2
4 r
r sin
2
i
1
Q
j
= U r2 sin2
cos + C
2
4
Inflow
Solid wallDetermine C from the stagnation point
n
Q
0
ur ( = , r0 ) = 0 r02 =
4U
1
Since = 0 on the body we get
Q
M2
C = .
4
M
mThe stream function is then
1
Q
m+2
= U R2 sin2 (cos +1)
2
1
| 4{z }
h
source
V
d
Vh The shape is given by = 0. As r , 0 then r sin . This gives
2
u
uh
0
1

: to be updated 10
: new values
: old values
i1
5
i
i+1
j1
j 0
j+1
iteration direction
x
y 5
1
n
x
y 10
i
j
fine grid (1) 15
15
10
5
5
10
15
0
coarse grid (2) 20

Figure B.7: Rankine body for submarine problem


Q
1 d2
U
2
=0
2 4
4

4Q
d =
U
2

d=2

Q
U

There is a simple way of determining the radius as z directly. The flow from the source must take up
an particular area in the flow at infinity. Since no fluid can cross the streamlines this area must be equal
to that of the Rankine body:
r
d2
Q
Q = U
d=2
= 12.07 m
2
U

uh We can use the computed stream function for a point source and displace it to z = a. In cylindrical
0 coordinates


1
Q
z+a
p
=

4
(z a)2 + R2
to be updated
: new valuesTransform to spherical coordinates




: old values
Q
Q
r cos + a
r cos + a
i1
q

=
=
4
4
r2 + a2 + 2ar cos
(r cos + a)2 + r2 sin2
i
i+1
j 1The stream function for the submarine is then
j


r cos a
1 2 2
Q
r cos + a
j +1

+
= U r sin +
2
4
r2 + a2 + 2ar cos
r2 + a2 2ar cos
ration direction
x
y
1
n
x
y
i
j
fine grid (1)
coarse grid (2)

10

0
10
80

60

40

20

20

40

60

80

Figure B.8: Submarine body for submarine problem


At r = R and = /2 we get


1
Q
a
a
2
= UR +

+
2
4
R 2 + a2
R 2 + a2
For the body = 0 and we get
R2
Multiply by
R2

R 2 + a2

aQ
=0
U

p
aQ
R 2 + a2 +
U

(R2 )3 + (R2 )2 a2

a2 Q2
=0
2 U 2

Computing this
d = 2R = 12.0006 m

m
The complex potential
m+2
1
hThe lines with constant stream function are the streamlines. They are orthogonal to the lines of constant
V velocity potential which are equipotential lines. Since both of them satisfy Laplaces equation we can
Vh define a complex function
F (z) = (x, y) + i (x, y) z = x + iy
u
uh
0
1

: to be updated 18
: new values
: old values 16
i1
i 14
i+1
j 1 12
j
j + 1 10
iteration direction
x 8
y
1 6
n
x 4
y
i 2
j
fine grid (1) 0
45
40
35
30
25
coarse grid (2) 50
Figure B.9: Complex potential for submarine problem, solid: , dotted:
This is an analytical function since the CauchyRiemann equation holds

=
x
y
The velocity is then

and

=
y
x

dF

=
+i
= u iv
dz
x
x
This enables the use of complex analysis, in particular conformal mapping that can be used to compute
the flow over airfoil shapes.
w(z) =

m Example:
m+2
1
h
Flow past a rotating cylinder centered at z = at an angle of attack
V


Vh
(a + )2 i
i
i
F
(z)
=
U
(z
+
)e
+
e

log(z + )
u
(z + )
2
uh
0

 21
1 Mapping by z = 1 Z + 1 Z 2 a2
gives an airfoil shape with the potential F(z). A correct flow is not
2
4

to be updatedachieved unless the KuttaJoukovski condition is satisfied requiring


: new values
= 4U (a + ) sin
: old values
i1
i
i+1
Z=f(z)
j 1 4
j 3
j +1
ration direction 2
2
x 1
y
0
0
n
x
y
i
j
fine grid (1)
coarse grid (2)

3
4
4

1
z=f (Z)

Figure B.10: Conformal mapping from circle to airfoil shape, (a = 3, = 0.5)

B.10

More potential flow

Half body over a wall


A line source of strength Q is located at (0, a) above a flat plate that coincides with the x-axis. A uniform
stream with velocity U flows along the x-axis. Calculate the irrotational flow field.
Method of images. Put a line source of equal strength at (0, a) in order to fulfill the condition of no flow
through the plate. Superposition of a uniform flow and the two line sources gives the complex potential
F = Uz +

Q
Q
ln(z ia) +
ln(z + ia)
2
2

Complex velocity



dF
Q
1
1
W =
=U+
+

dz
2 z ia z + ia


Q
1
1
W =U+
+

2 x + i(y a) x + i(y + a)


Q x i(y a)
x i(y + a)
+

W =U+
2 x2 + (y a)2
x2 + (y + a)2



Q
x
x
ya
y+a
W =U+
+ 2
i 2
+ 2
2 x2 + (y a)2
x + (y + a)2
x + (y a)2
x + (y + a)2

The velocity field now becomes



x
x
Q
+ 2
u=U+
2 x2 + (y a)2
x + (y + a)2


Q
ya
y+a
v=
+
2 x2 + (y a)2
x2 + (y + a)2

0
1 Flow past a symmetric airfoil
a) Use conformal mapping to calculate the irrotational flow field around a symmetric airfoil.
to be updated
: new values
Joukowski transformation
: old values
c2
(z) = z +
i1
z
i
i+1
j 1
j 1
2
j +1
ration direction
1
x 0.5
y
0
0
n
x
1
y 0.5
i
j 1
2
fine grid (1)
3
2
1
2
1
1
0.5
0.5
0
1
0
coarse grid (2) 1.5
Equation for the circle
z = + (a + )ei
Equation for the airfoil
= + (a + )ei +
Complex potential in the z-plane
F = U (z + )ei + U

a2
+ (a + )ei

(a + )2 i i
e +
ln(z + )
(z + )
2

Complex velocity in the z-plane


W =

(a + )2 i
i
dF
= U ei U
e +
2
dz
(z + )
2(z + )

Complex velocity in the -plane


dF/dz
=
=
d/dz

Ue

(a + )2 i
i
e +
U
2
(z + )
2(z + )



a2
1 2
z

The velocity can then be found by introducing the reversed transformation z = /2 +


v = Im{(z)}

u = Re{(z)},

2 /2 a2 into

b) Calculate the Joukowski condition for the airfoil.


The flow field has a singular point at the trailing edge of the airfoil at = 2a. Resolve the singularity by
choosing the circulation so the numerator vanishes at the trailing edge z = a
U ei U ei +
k = 4(a + )U

i
=0
2(a + )

ei ei
= 4(a + )U sin
2i

B.11

Boundary layers

Exercise 1
Consider the high Reynolds number flow in a conversing channel. Compute the boundary layer over the
surface at y = 0.
Assume the free stream:
U (x) =

Q
,
x

where Q is the flux. The boundary layer equations read:


u

u
u
Q
2u
+v
= 3 + 2
x
y
x
y

(1)

u v
+
= 0 (2)
x y
Seek a similarity solution where the dimensionless velocity u/U only depend on the dimensionless wall
distance
r
y
y
y
y Q
= =q
= q =
x

x
x
|U |

The stream function satisfies (2):

,
y

u=

v=

Make the stream function dimensionless:


f () =

Determine the terms in equation (1):


u=

f
1
Q

= U
= U f 0 = U f 0 = f 0
y
y

(U f )
f

U
=
= U
U
f
f =
x
x
x
x
x

r
r
r
Q
0  Q
Q

Q 0
x
f
+
f 2x
f =
f
x
Q
x
x Q
x
Q
x


u

Q 0
Q f 0
Q
Q   Q 0
Q
=
f =
+ 2 f 0 = f 00
+ 2 f = 2 (f 00 + f 0 )
x
x
x
x x x
x
x
x
x
r


u

Q
Q f 0
Q 1
Q Q 00
=
f0 =
= f 00 = 2
f
y
y
x
x y
x
x

r
2u
Q Q f 00
Q2
= 2
= 3 f 000
2
y
x
y
x
v=

Insert into equation (1):


Q Q
f 0 2 (f 00 + f 0 ) +
x x

Q 0 Q
f 2
x
x

Q 00
Q2
Q2
f = 3 3 f 000

x
x

Divide by Q2 /x3 :
f 0 f 00 f 02 + f 0 f 00 = 1 f 000
f 000 f 02 + 1 = 0
Boundary conditions:

u(0) = 0
u() = 1

f 0 (0) = 0
f 0 () = 1

M2
Substitute F () = f 0 ():

M
00
2

F F + 1 = 0
m
F (0) = 0
m+2

1
F () = 1
h
r !
V Solution:

2
2
Vh
+
F = 3 tanh
2
3
2
u
uh
0 Exercise 2
1
Consider the flow downstream of a 2D streamlined body at high Reynolds number. Study the thin wake
to be updateddownstream of the body where variations in y are much more rapid than variations in the downstream
: new valuesdirection. Also assume that the wake is so small that we can write u = U + u1 where u1 is negative and
: old valuesdescribes the wake. The length scale in x is L and in y it is with << L.

i1
i
i+1
j 1
j
j +1
ration direction
x
y

U+u 1

n
x
y
i
j
fine grid (1)
coarse grid (2)

Figure B.11: Coordinate system for wake problem

a) Find the governing (linear) equations:


We start with the boundary layer equations since << L,
u

u
1 p
2u
u
+v
=
+ 2
x
y
x
y
u v
+
=0
x y

In the wake we have no pressure gradient,

p
= 0.
x

From the continuity equation we get,




v
u

u1
=
=
U + u1 =
y
x
x
x

u1 << U
L

Inserting u = U + u1 in the boundary layer equations gives,


U

u1
u1
u1
2 u1
+ u1
+v
=0+
x
x
y
y 2

Neglect the quadratic terms,


u1

u1
u2
1
x
L

and v

u1
1
u2
u1 u1 1
y
L
L


b) Show that

u1
2 u1
=
x
y 2

u1 dy = constant:

Consider the relation,


Q1 =

u1 dy

which gives the linear contribution from u1 to the momentum flux.




Z
Z
Z
d
u1
d
2 u1
u1
Q1 =
u1 dy =
dy = {From the governing equation} =
dy
=
=0
2
dx
dx
U y
x
U y
This means that Q1 is constant.
c) Find a similarity solution for u1 :
Seek a solution on the form,
u1 (x, y) = F (x)f ()
This gives,
Q1 =

where =

F (x)f () dy = F (x)g(x)

y
g(x)

f () d

Q1 = constant requires,
Z

F (x) = 1/g(x) and


This means,

f () d = Q1 = constant



1
y
1
f () =
f
u1 (x, y) =
g(x)
g(x)
g(x)

Insert this into the equation of motion,




g 0 (x)
1
0
U
f
()
+
f
()
=
f 00 ()
g(x)2
g(x)3
U 0
f 00 ()
g (x)g(x) = 0
=C

f () + f ()
1/2

x
g(x) = 2C
U

This gives the equation for f ,


f 00 () + Cf 0 () + Cf () =
From the symmetry condition

y u(x, 0)



d
f 0 () + Cf () + D = 0
d

= 0 we can determine that f 0 (0) = D = 0. This gives,


f 0 () + Cf () = 0

Integrating this we get,


1

f () = ae 2 C

We can determine the constant a from the condition that Q1 is constant,


r
Z
2
1
C
Q1 e 2 C
f () d = Q1 f () =
2

This now gives u1 :


u1 (x, y) =

y2
C Q1 12 C g(x)
2
e
=
2 g(x)

g(x) =

2C

x
U

1/2 

Q1
=
2

U
x

1/2

U y2

e 4x

d) Relate u1 to the drag FD :

The drag is
FD =

u1 (U + u1 ) dy

Remember that u1 << U and neglect the u21 term,


Z
FD =
U u1 dy = U Q1

This means that


Q1 =

FD
U

u1 (x, y) =

FD

Ux

1/2

U y2

e 4x

Exercise 3
Give an order of magnitude estimate of the Reynolds number for:
i. Flow past the wing of a jumbo jet at 150 m/s ( Mach 0.5)
ii. A wing profile in salt water with L = 2 cm and U = 5 cm/s
iii. A thick layer of golden syrup draining of a spoon.
iv. A spermatozoan with tail length of 103 cm swimming at 101 cm/s in water.
Estimate the boundary layer thickness in case (i).

Fluid
Water
Air
Syrup

cm2 /s
0.01
0.15
1200

i. Flow past the wing of a jumbo jet at 150 m/s ( Mach 0.5)
U = 150 m/s, = 1.5 105 m2 /s, L = 4 m Re =

UL
4 107

ii. A wing profile in salt water with L = 2 cm and U = 5 cm/s


U = 0.05 m/s, = 106 m2 /s, L = 0.02 m Re =

UL
103

iii. A thick layer of golden syrup draining of a spoon.


U = 0.04 m/s, = 0.12 m2/s, L = 0.01 m Re =

UL
0.003

iv. A spermatozoan with tail length of 103 cm swimming at 102 cm/s in water.
U = 104 m/s, = 106 m2 /s, L = 105 m Re =

The boundary layer thickness in (i) is O(1 mm)





1
=O
L
Re

UL
103

m+2
More boundary layers
1B.12
h
V Drag on a circular cylinder
Vh
Use the integral form of the NavierStokes equations, Consider a fix volume in a fluid with velocity u
. The
ucontinuity equation becomes,
Z
I
uh
d

dV
=

uk nk dS.
0
dt V
S
1
The momentum equation,


Z
I 
Z
: to be updated
d
ui dV =
uj nj ui + pni ij nj dS +
Fi dV
: new values
dt V
S
V
: old values
i 1Use this to compute the drag force on a cylinder:
i U
n
U 0
0
i+1
j1
j
j+1
iteration direction
x
l
n
d
Wake
y
n
n
1
x
y
i
j
fine grid (1)
coarse grid (2)

The cylinder is stationary and so is the flow


= 0. Incompressible flow with constant density. The
t
continuity equation gives:
I
Z l/2
uk nk dS = 0 U0 l Lv|y=l/2 + Lv|y=l/2 +
Uw (y) dy = 0
S

l/2

Due to the symmetry we get,


U0 l + 2Lv|y=l/2 +
The momentum equation in x:

I 
S

l/2

Uw (y) dy = 0
l/2

nj uj ux + nx p nj xj dS = 0

Far away from the cylinder we can assume p = p0 and xx = xy = 0. This gives,
Z l/2
Z
U02 l + 2U0 v|y=l/2 L +
Uw2 (y) dy +
[nx p nj xj ] dS
cyl
l/2
|
{z
}
FD

Multiply the continuity equation by U0 ,

U02 l + 2U0 v|y=l/2 L = U0

l/2

Uw (y) dy
l/2

Insert into momentum equation:


U0
FD =

l/2
l/2

l/2

Uw (y) dy +
l/2

l/2
l/2

Uw2 (y) dy + FD = 0

U0 Uw (y) Uw2 (y) dy = U02

l/2
l/2

Uw (y)

U0

Uw (y)
U0

2 

dy

Make the variable substitution r = y/d,


FD = U02 d

l/2d
l/2d

Now if d << l we get,


FD = U02 d
|

Uw (r)

U0

Uw (r)
U0

2 

dr


2 
Uw (r)
Uw (r)

dr = U02
U0
U0
{z
}

The momentum thickness is called and is a measure of the loss of momentum in the flow. Now compute
FD

the drag coefficient CD = 1 2 = 2


d
U
d
0
2

Glauert wall jet


Compute an ODE for a self-similar wall jet.
The boundary layer equations read:
u

u
u
2u
+v
= 2
x
y
y

(1)

u v
+
= 0 (2)
x y
Seek a similarity solution where the dimensionless velocity u/Um(x), where Um (x) is the jet core velocity,
only depend on the dimensionless wall distance
y
=
g(x)
The stream function satisfies (2):
u=

,
y

v=

Make the stream function dimensionless:


f () =

Um (x)g(x)

Determine the terms in equation (1):

f
1
= Um g
= Um gf 0 = Um f 0
y
y
g



(Um gf )
yg 0
0
0
v=
=
= Um
gf Um g 0 f Um gf 0 2 = Um
gf Um g 0 f + Um g 0 f 0
x
x
g
u=

g0
0 0
=
(Um f 0 ) = Um
f Um f 00
x
x
g
u

1
=
(Um f 0 ) = Um f 00
y
y
g


2
u

1
1
=
Um f 00
= Um f 000 2
y 2
y
g
g

Insert into equation (1):


0 0
Um f 0 (Um
f Um
0 02
2
Um Um
f Um

g 0 00
1
1
0
f ) + (Um g 0 f 0 Um
gf Um g 0 f )Um f 00 = Um 2 f 000
g
g
g

0
0
g 0 0 00
1
2 g
0
2 g
f f + Um
f 0 f 00 Um Um
f f 00 Um
f f 00 = Um 2 f 000
g
g
g
g
 0 2

0
0 2
Um g
Um g g
U g
f 000 +
+
f f 00 m f 02 = 0

| {z }
|
{z
}

000

00

+ f f + f

02

=0

Boundary conditions:

u(0) = 0
v(0) = 0

u() = 0

f (0) = 0
f (0) = 0

0
f () = 1

Similarity requires that and are constant. Assume the x-dependence:


Um = axm ,
Then we get

g = bxn

1
ab2
= amxm1 b2 x2n =
m,

2n + m = 1

1
1
ab2
amxm1 b2 x2n + axm bnxn1 bxn =
(m + n)

We can choose the length scale g such that = 1:


=

ab2
1
=

m+n

m
m+n

So we have:
f 000 + f f 00 + f 02 = 0 (3)
For what is the boundary conditions fulfilled? Work on (3):
f 000 + f f 00 + f 02 = 0

f 000 + (f f 00 + f 02 ) + ( 1)f 02 = 0
Integrate:
00

f + f f + ( 1)g = 0,
Multiply with f 0 :

g() =

f f + (f f

02

+ gf ) + ( 2)gf = 0

Integrate:
1 02
f + f g + ( 2)
2

Put in = 0 and use the boundary conditions:


Z
( 2)
gf 0 d = 0
m
=2
m+n
1
m= ,
2

gf 0 d = 0

f 02 d > 0

f 00 f 0 + f f 02 + ( 1)gf 0 = 0
00 0

Solve for m and n:

=2

3m + 2n = 0

n=

3
4

B.13

Introduction to turbulence

Exercise 1
Compute the far-field two-dimensional turbulent wake U (x, y) behind a cylinder.
Let
U1 (x, y) = U0 U (x, y),

Us (x) = U1 (x, y = 0)

Assume
Us  U 0
Continuity

U
V
+
=0
x
y

U1  U0 ,

u, v Us ,

x L,

V
U1
= (U0 U1 ) =
y
x
x
|{z}
|{z}

yl

V Us

ULs

Vl

The turbulent boundary layer momentum equation read:


U

U
U
1 P0
2U

+V
=
+ 2
(uv)
x
y
x
y
y

Disregard the pressure gradient and insert U = U0 U1 :




U1
U1
2 U1

(U0 U1 )
V
=

(uv)
x
y
y 2
y
Neglect small terms:
U0

U1

=
(uv) (1)
x
y

Self-similar hypothesis:
U1 (x, y)
= f ()
Us (x)
uv
= g()
Us2
where
=

y
l(x)

= 2 l0 = l0 ,
x
l
l

1
=
y
l

Determine the terms in equation (1):

l0
U1
=
(Us f ) = Us0 f + Us f 0
= Us0 f Us f 0
x
x
x
l

(uv) =
(Us2 g) = Us2 g 0
= Us2 g 0
y
y
y
l
Insert into equation (1):

Boundary conditions:

l0
1
U0 (Us0 f Us f 0 ) = Us2 g 0
l
l
0
0
U0 lUs
U0 l
g0 =
f+
f 0
Us2
Us
(

U1 () = 0
U1 () = 0

f () = 0
f () = 0

l
L

Similarity requires:
lUs0
= const
Us2

l0
= const
Us

Assume the x-dependence:


Us = Axm ,

l = Bxn

Then we get
Bxn Amxm1 = const x2m

n + m 1 = 2m

Us = Axn1 ,

m=n1

l = Bxn

We need one more condition. The momentum loss thickness is constant:







Z
Z
Z 
U
U
U0 U 1
U0 U 1
U1 U1
=
1
dy =
1
dy =
1
dy
U0
U0
U0
U0 U0
U0

1
U0

1
U0

U1 dy =

1
CD d = const (Recitation 12)
2
Z

U1 dy = Us l

f d = const
|
{z }
const

Us l = Axn1 Bxn = const

Us = Ax 2 ,

2n 1 = 0

n=

1
2

l = Bx 2

For a sufficiently large Reynolds number, the wake will be turbulent.


Rel =

AB
lUs
=
= const

A turbulent wake will remain turbulent!


To determine the wake-velocity profile, we need to model the turbulent shear stress. E.g.
uv = T
Us2 g

= T

g=

U1

T 0
f
Us l

U
y

1
l

T 00
f
Us l

= T Us f 0
g0 =

T 00
U0 lUs0
U0 l 0 0
f =
f+
f
2
Us l
Us
Us

f 00 +

U0 l0 l 0 U0 l2 Us0
f
f =0
T
T Us

f 00 +

U0 B 2 0
(f + f ) = 0 (2)
2T

Equation (2) can be solved together with the boundary conditions


f () = 0

B.14

Old Exams
i
j

Exam 031023 Inflow


Solid wall
1. Relative motion.
n
0
0
Consider the relative
0 motion of two fluid particles initially separated by the distance dx i . (Here xi
and t are the Lagrangian
coordinates).
1
a) (5) Show that the
separation at time dt is
M2
ui
M
dri (dt) = dx0i + 0 dx0j dt.
m
xj
m+2
b) (2) Use this relation
1 to write down the expression for the deformation of the sides of a small cube
aligned with the coordinate
directions. Thus, show that the components of the side aligned with the
h
x-direction, Ri1 , isV deformed into
Vh
ui
u
(1)
dri = R(i1 + 0 dt),
uh
x1
0
with analogous expressions
for the other two sides.
1
deformation rate of the volume of the cube is given by the divergence of the
c) (4) Show that the
: to field.
be updated
velocity
Note that x0i = xi at t = 0.
: new values
2. Flow between
a rod and a cylinder.
: old values
1 cylinder with radius b, a concentric rod with radius a inside the cylinder, and
Consider a long ihollow
water occupying thei space between the rod and the cylinder, see figure B.14. The rod is rotating
i +angular
1
with the constant
frequency .
j1
(10) Find the velocity
j field of the water driven by the rotating rod and the gravitational acceleration
g.
j+1
iteration direction
z
x
y

1
n
x
g
y
i
j
fine grid (1)
a
b
r
coarse grid (2)
Figure B.12: Water and a concentric rotating rod inside a cylinder.

j
Inflow
Solid wall
n
0
1

M2
M
m
m+2
1
h
V
Vh
u
uh
0
1

: to be updated
: new values
: old values
i1
i
i+1
j1
j
j+1
iteration direction
x
y

3. Bernoullis equation
a) (3) Derive the following formula:
uj

ui
1
=
(uj uj ) + ijk j uk .
xj
2 xi

b) (3) Use this in the momentum equation to derive Bernoullis equation for unsteady potential flow.
c) (3) Use the result in (a) to derive Bernoullis equation for steady inviscid flow. Discuss the validity
of the derived relation.
4. Laminar wake flow.
Consider the laminar wake flow downstream of a two-dimensional streamlined body at high Reynolds
number, see figure B.13. Assume that the wake is so weak that we can write u = U + u 1 where u1 is
negative and much smaller than the free-stream velocity U .
a) (3) Motivate the usage of the equation
U

u1
2 u1
=
.
x
y 2

b) (2) Show that the mass flux Q1 , in the direction of the wake is constant in x
Z
Q1 =
u1 dy. ()

c) (7) A similarity solution can be found by scaling the wake velocity with Us (x) = u1 (x, 0) so that
f () = u1 (x, y)/Us (x), where = y/(x) is the similarity variable. Show, using (), that Us = k/(x),
where k is a given constant. Find the similarity solution for u1 .
U

U+u

n
x
y
i
j
fine grid (1)
coarse grid (2)

Figure B.13: Wake downstream of a flat plate.


5. Mean heat equation.
The heat equation for the instantaneous turbulent flow is
T
T
2 T
+u
j
=
.
t
xj
xj xj
(8) Derive the heat equation governing the mean flow.

Answers to exam in Fluid mechanics 5C1214, 2003-10-23


1. See Kundu & Cohen p. 57.
2. Flow between a rod and a cylinder.
a2
u = 2
b a2
uz =

1g
4

b2 r 2
r

r2 a2 (b2 a2 )

3. See Kundu & Cohen pp. 110-114.


4. Laminar wake flow.
Q1
u1 =
2
5. See Kundu & Cohen pp. 511-512.

U U y2
e 4x
x


ln(r/a)
ln(b/a)

0
1
flow,
1. (5) Consider the unsteady
M2
u = u0 , v = kt, w = 0,
M
where u0 and k are positive
constants. Show that the streamlines are straight lines, and sketch them
m
at two different times.
m + 2Also show that any fluid particle follows a parabolic path as time proceeds.

Exam 040113

2. (5) For surface waves1on water with finite depth h the phase speed is given by the relation c 2 =
g
h
k tanh(kh). Compare the group velocity to the phase speed in the limit of long and short waves.
V
Make a rough 2D sketch of a wave packet and explain how it is moving.
Vh
3. Consider a thin layer uof liquid in contact with a solid flat vertical wall. See figure 1. The wall is
subject to a constantuhtemperature gradient with increasing temperature in the downward direction.
0 tension, as a result of the temperature gradient, a constant shear stress
Due to a varying surface
at the free surface is1 acting upward. At the same time, the gravitational acceleration g is acting

downward.
: to be updated
a) (7) Assume constant layer thickness h and calculate the velocity field of the liquid layer.
: new values
b) (3) For what
thickness h is the flux zero?
: old layer
values
i1
z
i
i+1
j1
j

j+1
iteration direction
x
g
y
1
n
x
y
i
x
h
j
fine grid (1)
coarse grid (2)
Figure B.14: Liquid film on vertical wall driven by gravity and surface tension.

4. a) (3) Show that


uj

ui
1
=
(uj uj ) + ijk j uk
xj
2 xi

b) (4) Derive the vorticity equation starting with the NavierStokes equation for incompressible flow.
c) (3) Show that the following relation holds for incompressible flow and discuss its implication for
inviscid flow
ij
k
= ijk
xj
xj
5. Consider the flow over an oscillating plate.
a) (3) Motivate that the solution can be written as [u(y, t), 0, 0] and that it thus satisfies the equation
2u
u
= 2
t
y
b) (7) Show that
u(y, t) = U eky cos(ky t),

where k =

.
2

6. a) (5) Derive the Reynolds average equation valid for turbulent flow.
b) (5) Assume that the Reynolds stress can be modeled as a turbulent viscosity and introduce this
into the Reynolds equation and simplify.

j1
A
B
C
u 23 ,1
v1, 32
p1,1

Solutions to exam in Fluid mechanics 5C1214, 2004-01-13


1. Streamline (fix t):
dx
= u = u0
ds
dy
= u = kt
ds
Then we see that y(x) is a linear function,

1
y=
i
j
Inflow
Solid wall
n
0
1

M2
M
m
m+2
1
h
V
Vh
u
uh
0
1

: to be updated
: new values
: old values
i1
i
i+1
j1
j
j+1
iteration direction
x
y

x = u0 s + a

y = kts + b

kt
x + d(t)
u0

Particle path (fix x


0 ):

x
= u = u0 x = u0 t + x0
t
y
1
= u = k t x = k t2 + y0

2
t
Then we see that y(x) is a parabola,
y=

1 k 2
(x 2x0 x + x20 ) + y0
2 u20

2. The wave number is defined as k = 2/, where denotes the wave length of the wave. For long
waves, go to the limit k 0. For short waves, go to the limit k .
Phase speed:

c= =
k
Short waves:

g
tanh(kh) =
k

tanh(x)
1
lim
=
x
x
x

gh

tanh(kh)
kh

c=

Long waves:
tanh(x)
=1
x0
x

lim

Group velocity:
d
c
cg =
=
dk
2
Short waves:

c=

2kh
1+
sinh(2kh)

gh

lim

x
=0
sinh(x)

cg =

lim

x
=1
sinh(x)

cg = c

Long waves:
x0

g
k

c
2

Longer waves travel faster.


n
x
y
i
j
fine grid (1)
coarse grid (2)

2
1.5

1
0.5

0
0.5

1.5

0.5

1.5

2.5

3. a) Assume steady, parallel flow u


= w(x)
ez and solve the z-momentum equation with boundary
conditions:
2w
w

2 g = 0, w(0) = 0,
(h) =
x
x

Integrate the equation once and introduce the second boundary condition:


w
g
w
gh

1
= x + A,
(h) =
+A =
A=
gh
x

Integrate the equation once more and introduce the first boundary condition:
w=

1g 2
x + Ax + B,
2

w(0) = 0

So we get:
1g 2 1
w=
x +
2

B=0

gh x

b) The flux per unit width of the wall can be written




 h
Z h
1g 3 11
11 2 1g 3
Q = w dx =
x +
gh x2 =
h
h
6
2
2
3
0
0
Zero flux gives:
h=

3
2 g

4. a) Begin with the second term on the right hand side:


ijk j uk = ijk jlm

uj

um
ui
uk
(um )uk = (kl im km il )uk
= uk
uk
xl
xl
xk
xi

ui
uj
= uj
+ ijk j uk
xj
xi

uj

ui
1
=
(uj uj ) + ijk j uk
xj
2 xi

b) NavierStokes momentum equation for incompressible flow:


ui
ui
1 p
2 ui
+ uj
=
+
t
xj
xi
xj xj
Use the expression from 4. a):
ui
1
1 p
2 ui
+
(uj uj ) + ijk j uk =
+
t
2 xi
xi
xj xj
Take the curl of the equation. The curl of a gradient vanish:
i

2 i
+ imn
(njk j uk ) =
t
xm
xj xj

2 i
+ nim njk
(j uk ) =
t
xm
xj xj

2 i
+ (ij mk ik mj )
(j uk ) =
t
xm
xj xj

2 i
+
(i uk )
(j ui ) =
t
xk
xj
xj xj
Use continuity. We end up with the vorticity equation:
i
i
ui
2 i
+ uj
= j
+
t
xj
xj
xj xj
c) Right hand side:


k

um
2 um
= ijk
klm
= kij klm
=
xj
xj
xl
xj xl
 2

2 um
uj
2 ui
(il jm im jl )
=

= 2 ui
xj xl
xj xi
xj xj
ijk

Left hand side:

Thus:

ij

=
xj
xj


ui
uj
+
= 2 ui
xj
xi

ij
k
= ijk
xj
xj
Irrotational flow is not subject to viscous forces.

5. a) Consider the NavierStokes equation in the x direction:


 2

u
u
u
u
1 p
u 2u 2u
+u
+v
+w
=
+
+ 2 + 2
t
x
y
z
x
x2
y
z
With the current velocity field only terms with y derivatives will remain since there can be no change
in the other directions,
u
2u
= 2
t
y


b) Make the ansatz: u = < f (y)ei t = f (y) cos( t).
Insert into the equation:

if (y)ei t = f 00 (y)eit
f 00 (y)
2
Introduce k =

i
f (y) = 0

i
=0

with f (y) = ey gives


r
r 

i
1+i

=
=

,
2
f (y) = Aeyk(1+i) + Beyk(1+i) ,

f (y) 0 as y A = 0

We have




u = < Beyk(1+i) eiwt = < Beky ei(tky) = Beky cos(t ky) = Beky cos(ky t)
Boundary condition at y = 0
At y = 0 we have u = U cos(t) B = U
So we have,
u(y, t) = U eky cos(ky t),

where k =

6. a) Turbulent flow is inherently time dependent and chaotic. However, in applications one is not
usually interested in knowing full the details of this flow, but rather satisfied with the influence of
the turbulence on the averaged flow. For this purpose we define an ensemble average as
N
1 X (n)
ui
N N
n=1

Ui = ui = lim
(n)

where each member of the ensemble ui

is regarded as an independent realization of the flow.

We are now going to derive an equation governing the mean flow Ui . Divide the total flow into an
average and a fluctuating component u0i as
ui = ui + u0i

p = p + p0

and introduce into the Navier-Stokes equations. We find


u0i
ui
ui
u0
ui
u0
1 p0
1 p
+
+ u0j
+ uj i + uj
+ u0j i =

+ 2 u0i + 2 ui
t
t
xj
xj
xj
xj
xi
xi
We take the average of this equation, using
u0i = 0
which gives

u0j ui = ui u0j = 0


ui
ui
 0 0
1 p

+ uj
=
+ 2 ui
u i uj

t
xj
xi
xj

ui = 0
xi

where the average of the continuity equation also has been added. Now, let U i = ui , as above, and
drop the 0 . We find the Reynolds average equations

Ui
Ui
1 P

+ Uj
=
+
(ij ui uj )

t
xj
xi
xj

Ui = 0
xi

where ui uj is the Reynolds stress.


b) Assume the Reynolds stress:

ui uj =

2
Kij 2T eij
3

The first term on the right hand side is introduced to give the correct value of the trace of u i uj . Here
the deformation rate tensor is calculated based on the mean flow, i.e.
eij =

1
2

Ui
Uj
2 Ur
+

ij
xj
xi
3 xr

where we have assumed incompressible flow. Introducing this into the Reynolds average equations
gives


 

1 P
Ui

P
2
Ui
+ Uj
=

+ k ij + 2 ( + T ) eij =
+ ( + T ) 2 Ui
t
xj
xj

3
xi

where the last equality assumes that T is constant, an approximation only true for very simple
turbulent flow. It is introduced here only to point out the analogy between the molecular viscosity
and the turbulent viscosity.

0
Exam 041018
1
1. The kinetic energy dT of an infinitesimal volume element, with density , which is rotating with
: to be updated
angular velocity can be expressed
: new values
1
1
: old values
dT = v v dV = ( x) ( x) dV,
2
2
i1
i
where v is the velocity of the volume element and x is the vector from the center of rotation O to
i+1
the volume element, see figure B.15.
j1
j
x03
j+1
dV
iteration direction
x
x0
y
C
x
1
x02
0
n x3
x1
x
z
y
i
j
O
fine grid (1)
x2
coarse grid (2)x1
Figure B.15: Coordinate system and definitions
a) (5) Show that the total kinetic energy of the body can be expressed as
Z
1
T =
dT = Jjl j l ,
2
V
where Jjl is the moment of inertia tensor with respect to the origin of the unprimed system. It is
defined as
Z
Jjl =
(xk xk jl xj xl ) dV.
V

b) (5) Show that you can relate Jjl to the moment of inertia tensor with respect to the center of mass
C
of the body Jjl
in the following way
C
Jjl = (zk zk jl zj zl )M + Jjl
,

C
Jjl
=

(x0k x0k jl x0j x0l ) dV,

where M is the total mass of the body and the primed coordinates represent the coordinate system
centered at the center of mass C of the body.
R
Hints: Integrals of the type V x0k dV vanishes as a result of the definition of the center of mass. The
two coordinate systems are related by the formula x = z + x0 .

2. (10) Consider the laminar flow of a viscous fluid down a vertical cylindrical pipe with radius a. The
flow is driven by the gravitational acceleration g. Calculate the velocity field.
3. Potential flow and conformal mapping
a) (2) Show that the Kutta-Joukowski transformation = z + a2 /z maps a circle with radius a to a
flat plate with length 4a.
b) (2) Write down the complex potential of the flow around a circular cylinder with radius a at angle
of attack and with clockwise circulation . The uniform far-field velocity has the magnitude U .
c) (4) Show that a necessary condition for finite velocity at the trailing edge of the flat plate is (Kutta
condition)
= 4U a sin .
d) (2) How should one modify the circle in order to obtain a finite velocity at the leading edge?
4. Blasius boundary layer
a) (2) Write down the boundary layer equations.
b) (8) Make appropriate assumptions for similarity and reduce the boundary layer equations to an
ordinary differential equation for the boundary layer over a flat plate (i.e there is no pressure gradient
in the free stream). Also state the boundary conditions for the similarity function.
c) (2) Sketch the streamwise and normal velocity profiles.
5. Turbulent flow
a) (2) Define the velocity and length scales appropriate for turbulent flow near a wall, using the wall
shear stress w , the viscosity and the density .
b) (2) What is the law of the wall and what is the explicit functional dependence of the turbulent
mean velocity in the intermediate region between the wall and the outer flow?
c) (4) In two-equation turbulence models, partial differential equations for the turbulent kinetic energy
k and the dissipation rate are solved. In those equations, the turbulent viscosity T = uT l, where uT
and l are the turbulent velocity and length scales, respectively. Use dimensional analysis to determine
uT , l and T in terms of k and .

Solutions to exam in Fluid mechanics 5C1214, 2004-10-18


1. a) The kinetic energy of an infinitesimal volume element
dT =
dT =

1
1
v v dV = ( x) ( x) dV
2
2

1
1
1
vi vi dV = ijk j xk ilm l xm dV = (jl km jm kl )xk xm j l dV =
2
2
2
1
(jl xk xk xj xl )j l dV
2
Z
Z
1
1
T =
dT =
dT (jl xk xk xj xl ) dV j l = Jjl j l ,
2
2
V
V

where
Jjl =

(jl xk xk xj xl ) dV.

b) The moment of inertia tensor


Z
Z
Jjl =
(jl xk xk xj xl ) dV =
[jl (zk + x0k )(zk + x0k ) (zj + x0j )(zl + x0l )] dV =
V

[jl (zk zk + 2zk x0k + x0k x0k ) (zj zl + zj x0l + zl x0j + x0j x0l )] dV =
(zk zk jl zj zl )
2zk jl

dV +
V

x0k dV

+ zj

(x0k x0k jl x0j x0l ) dV +

x0l dV

+ zl

x0j dV.

The three last terms vanishes due to the definition of the center of mass. So we have
C
,
Jjl = (zk zk jl zj zl )M + Jjl

where
M=

dV

and

C
Jjl

(x0k x0k jl x0j x0l ) dV.

2. Poiseuille flow ( Pipe flow )


Use the NavierStokes equations (with no volume force) in cylindrical coordinates


u2
1 p
ur
2 u
ur
+ (u )ur =
+ 2 u r 2 2
t
r
r
r
r


u
ur u
1 p
2 ur
u
+ (u )u +
=
+ 2 u + 2
2
t
r
r
r
r
uz
1 p
+ (u )uz =
+ 2 uz
t
z
1
1 u
uz
(r ur ) +
+
= 0.
r r
r
z
p
p
We know that
= 0 and
= 0 and can directly see that ur = u = 0 satisfy the two first

r
equations. From the continuity equation we get
uz
=0
z

uz = uz (r, )

only.

Considering a steady flow we get from the NavierStokes equations


(u )uz = uz

uz
z

uz

uz
= 2 uz g
z

But we know that

uz
= 0 from the continuity equation. We get
z
2 u z =

Integrate
r

1 uz
g
(r
)=
r r r

uz
1g 2
=
r + c1
r
2

uz
g
(r
)= r
r r

uz
1g
c1
=
r+
r
2
r

Integrating again we get


uz =

1g 2
r + c1 ln r + c2
4

using the boundary conditions uz (r = 0) < c1 = 0

We also have uz = 0 at r = a and this gives c2 =


uz =

ga2
and we have
4

g 2
(a r2 )
4

3. Potential flow and conformal mapping


a) Equation for the circle
z = aei
Equation for the plate
=z+

a2
ei + ei
= aei + aei = 2a
= 2a cos ,
z
2

which is a real function taking the values from 2a to 2a, thus a flat plate with length 4a.
b) Complex potential for a circular cylinder

F = U zei + U

a2 i
i
e +
ln z
z
2

c) Complex velocity for the circular cylinder


W =

dF
a2
i
= U ei U 2 ei +
dz
z
2z

Complex velocity in the -plane


=

dF/dz
=
d/dz

U ei U

a2 i
i
e +
2
z
2z



a2
z2

The flow field has singular points at the leading and trailing edges of the flat plate. Resolve the
singularity at the trailing edge by choosing the circulation so the numerator vanishes at z = a
U ei U ei +
k = 4U a

i
=0
2a

ei ei
= 4U a sin .
2i

d) Move the point z = a inside of the circle by putting the center at z = . We still want a sharp
trailing edge so the circle crosses the real axis at z = a. The radius of such a circle is a + and the
circle is described by
z = + (a + )ei
4. Blasius boundary layer
a) Boundary layer equations
u

u
u
dU
2u
+v
=U
+ 2
x
y
dx
y

(B.3)

u v
+
=0
x y

(B.4)

b) Seek a similarity solution where the dimensionless velocity u/U , where U is the constant free-stream
velocity, only depends on the dimensionless wall distance
=

y
(x)

The stream function satisfies (2):


u=

,
y

v=

Make the stream function dimensionless:


f () =

(x, y)
U (x)

= U f

Determine the terms in equation (1):

f
1
= U
= U f 0 = U f 0
y
y




y 0
(U f )
0
0
v=
=
= U f U f 2 = U 0 f 0 U 0 f
x
x

u=

0
=
(U f 0 ) = U f 00
x
x

1
=
(U f 0 ) = U f 00
y
y



2
u

1
1
=
U f 00
= U 2 f 000
2
y
y

dU
=0
dx

Insert into equation (1):


U f 0 (U
U 2

0 00
1
1
f ) + (U 0 f 0 U 0 f )U f 00 = U 2 f 000

0 0 00
0
0
1
f f + U 2 f 0 f 00 U 2 f f 00 = U 2 f 000

 0 
U

f 000 +
f f 00 = 0

| {z }

Similarity requires that is a constant. We can integrate it with respect to x in order to find
r
U 0
1 2
x
2x
=
=
(x) =

2
U
U
The value of the constant is just a matter of scaling, we can choose it to be 1/2 so
r
x
(x) =
.
U
The ordinary differential equation becomes
1
f 000 + f f 00 = 0.
2
We have the boundary conditions

u(0) = 0
v(0) = 0

u() = U

0
f (0) = 0
f (0) = 0
0
f () = 1.

c) Figure B.16 shows the dimensionless streamwise and normal velocity profiles.

1
2

i+
i+1
i + 32
j + 12
j
j 12
j 1
A
B
C
u 23 ,1
v1, 32
p1,1
1

i
j
Inflow
Solid wall
n
0
1

M2
M
m
m+2
1
h
V
Vh
u
uh
0
1

to be updated
: new values 7
: old values
i1
6
i
i+1
j 15
j
j +1
4
ration direction

x
y3
n2
x
y
i1
j
fine grid (1) 0
coarse grid (2) 0

5. Turbulent flow
a) Velocity scale near the wall:
u =

Length scale near the wall:


l =

b) Law of the wall


U+ =

U
= f (y + ),
u

where y + =

y
l

Functional dependence in the intermediate region


U+ =
where is the Karmans constant.
c) Dimensions for k and :

Dimension analysis for uT :

k = 21 ui ui [L2 T 2 ]
u u
= i j [L2 T 3 ]
xj xi
uT km n

Dimension analysis for l:

1
ln y + + B,

[LT 1 ]

LT 1 = L2m T 2m L2n T 3n
(
2m + 2n = 1
n=0

2m 3n = 1
m = 12
p
uT k
l km n

[L]

L = L2m T 2m L2n T 3n
(
2m + 2n = 1
n = 1

2m 3n = 0
m = 23

0.2

0.4

0.6

0.8

Figure B.16: Velocity profiles of Blasius boundary layer.

l
So we have

3
k 2

p k 32
T = uT l k

k
T

or
T = C

k2
.

Appendix C

Study questions 5C1212


1. Define and use the Rankine-Hugoniot jump condition to compute the shock speed for the following
problem
= 0 < x < ,
(
1 x0
u(x, 0) =
0 otherwise

ut + uux

t>0

2. Define the entropy condition for a scalar conservation law.


< x < ,

ut + f (u)x = 0

t>0

with a convex flux function f (u). The shock is moving with speed s and the state to the left is given
by uL and the state to the right by uR .
Why do we need an entropy condition ?
3. Write a difference approximation (in 1D) on conservative form and define the notation.
4. Solve
u2
ut + ( )x = 0,
2

u(x, 0) =

1 x<0
0 x0

by the upwind scheme on the following form,


un+1
= unj
j

t n n
u (u unj1 )
x j j

a) Will the numerical solution converge to the analytical solution?


b) If not, suggest another way of solving C.1.
5. Define a total variation decreasing (TVD) method. Why is this a desirable property ?
6. Investigate the one-sided difference scheme
un+1
= unj a
j

t n
(u unj1 )
x j

for the advection equation


ut + aux = 0
Consider the cases a > 0 and a < 0.
a) Prove that the scheme is consistent and find the order of accuracy. Assume k/h constant.
b) Determine the stability requirement for a > 0 and show that it is unstable for a < 0.
7. Apply Lax-Friedrichs scheme to the linear wave equation
ut + aux = 0.
a) Write down the modified equation.
b) What type of equations is this?
c) What kind of behavior can we expect from the solution?
173

(C.1)

8. A the three-point centered scheme applied to


ut + aux = 0,
yields the approximation
un+1
= unj +
j

a > 0.

at
(uj+1 uj1 )
2x

Show that this approximation is not stable even though the CFL condition is fulfilled.
9. What does Laxs equivalence theorem state?
10. What is the condition on the n n real matrix A(u) for the system
ut + Aux = 0
to be hyperbolic ?
11. The barotropic gas dynamic equations
t + ux = 0
1
ut + uux + px = 0

(C.2)

where
p = p() = C
i and C a constant, can be linearized by considering small perturbations (0 , u0 ) around a motionless
gas.
a) Let = 0 + 0 and u = u0 + u0 where u0 = 0. Linearize the system (C.2) and show that this
yields the following linear system (the primes has been dropped)
t + 0 ux = 0
a2
ut + x = 0
0

(C.3)

where a is the speed of sound. a and 0 are constants.


b) Is the system given by (C.3) a hyperbolic system? Motivate your answer.
c) Determine the characteristic variables in terms of and u.
d) Determine the partial differential equations the characteristic variables fulfill - characteristic formulation.
e) Given initial conditions at t = 0 and let < x < (no boundaries)
(0, x) = sin(x) u(0, x) = 0
determine the analytical solution to (C.3) for t > 0. Hint: Start from the characteristic formulation.
12. The linearized form of (C.3) is given by
 

 

0
0

+ 2
= 0,
u t
a /0 0
u x
|
{z
}

(C.4)

where a is the speed of sound. a and 0 are constants.

a) Draw the domain of dependence of the solution to the system (C.4) in a point P in the x-t plane.
b) The system is solved numerically on a grid given by xj = jx, j = 0, 1, 2... and tn = nt, n =
0, 1, 2, ..... using an explicit three-point scheme, see the figure below.
Draw the domain of dependence of the numerical solution at P (in the same figure as a)) of the
three-point scheme in the case when
i) the CFL condition is fulfilled
ii) the CFL condition is NOT fulfilled.
Assume that P is a grid point.

popo nmm

1t n+1
x
y
i
t
fine grid (1) n
coarse grid (2)

popo
popo
popo

kk

popo hhg
g
jij
i

j1

ll
j+1

13. To solve Euler equations in 1D


t + ux + ux = 0
1
ut + uux + px = 0

pt + c2 ux + upx = 0
How many boundary conditions must be added at (motivate your answer)
inflow boundary when the flow is
a) Supersonic
b) Subsonic
outflow boundary when the flow is
c) Supersonic
d) Subsonic
14. Describe the ideas behind a flux splitting scheme for solving a non-linear hyperbolic system of equations,
Ut + F (U)x = 0
15. a) Show that a vector field wi can be decomposed into
wi = u i +

p
xi

where u is is divergence free and parallel to the boundary.


b) Apply this to the Navier-Stokes equations, show that the pressure term disappears and recover an
equation for the pressure from the gradient part.
16. From the differential form of the Navier-Stokes equations obtain
a) the Navier-Stokes equations in integral form used in finite-volume discretizations,
b) a variational form of the Navier-Stokes equations used in finite-element discretizations.
17. a) Write down the appropriate function spaces for the pressure and velocity used to define weak
solutions to the Navier-Stokes equations.
b) Explain the concept of essential and natural boundary conditions.
18. a) How is a finite-element approximation defined?
b) Explain how to convert a FEM discretization to an algebraic problem, e.g. that the Navier-Stokes
equations yield


K + C(u) G
GT
0



u
p

f
0

19. a) By choosing zh = uh in the finite element approximation of the Stokes problem, show that any
Galerkin velocity solution is stable.
b) Describe and interpret the LBB condition.

20. Choice of elements. Discrete elements pairs


a) constant pressure-bilinear velocities,
b) the Taylor-Hood pair,
c) a stable choice with piecewise linear velocities.
21. a) Derive the finite volume (FV) discretization on arbitrary grids of the continuity equation (u i /xi =
0),
b) derive the FV discretization for Laplace equation on a Cartesian grid,
c) show that both are equivalent to a central difference approximation for Cartesian grids.
22. Derive the finite element (FEM) discretization for Laplace equation on a Cartesian grid and show
that it is equivalent to a central difference approximation.
23. Iterative techniques for linear systems.
a) Define Gauss-Seidel iterations for the Laplace equations,
b) Define the 2-level multigrid method for the Laplace equation,
24. State the difficulties associated with the the finite-volume discretizations of the Navier-Stokes equations on a colocated grid? and show the form of the spurious solution which exist.
25. a) Define an appropriate staggered grid that can be used for the discretization of the Navier-Stokes
equations,
b) write down the FV discretization of the Navier-Stokes equations on a staggered cartesian grid,
c) discuss how to treat noslip and inflow/outflow boundary conditions.
26. Time dependent flows.
a) Define a simple projection method for the time dependent Navier-Stokes equations
d
dt

u
0

N(u) G
D
0



u
p

f
g

b) show in detail the equation for the pressure to be solved at each time step and discuss the boundary
conditions for the pressure.
27. Time step restriction for Navier-Stokes solutions.
a) Motivate the use of an appropriate form of the advection-diffusion equation as a model equation
for stability analysis,
b) derive the time step restrictions for the 1D version of that equation,
c) state the 2D equivalent of that restriction.

Bibliography
[1] D.J. Acheson, 1990, Elementary Fluid Dynamics, Oxford University Press.
[2] J.D. Anderson, Jr., 1995, Computational Fluid Dynamics, The basics with applications, Mc Graw-Hill.
[3] C.A.J. Fletcher, 1991, Computational Techniques for Fluid Dynamics Volume 1, Second Edition,
Springer-Verlag.
[4] C.A.J. Fletcher, 1991, Computational Techniques for Fluid Dynamics, Volume 2, Second Edition,
Springer-Verlag.
[5] M. Griebel, T. Dornseifer and T. Neunhoeffer, 1998, Numerical Simulation in Fluid Dynamics, A
practical Introduction, SIAM Monographs on Mathematical Modeling and Computation.
[6] P.K. Kundu and I.M. Cohen, 2002, Fluid Mechanics, Second Edition, Academic Press.
[7] R.L. Panton, 1995, Incompressible Flow, Second Edition, Wiley-Interscience.
[8] R. Peyret, T.D. Taylor, 1983, Computational Methods for Fluid Flow, Springer-Verlag.
[9] J.C. Tannehill, D.A. Anderson and R.H. Pletcher, 1997, Computational Fluid Mechanics and Heat
Transfer, Second Edition, Taylor and Francis.
[10] P. Wesseling, 2001, Principles of Computational Fluid Dynamics, Springer-Verlag.

177