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
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
59
59
62
65
66
69
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
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
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
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
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
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.
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.
Chapter 1
Notation
+ (u ) u = p + 2 u
u
= 0
u = (u, v, w) = (u1 , u2 , u3 )
the nabla operator is defined as
=
,
,
x y z
,
,
x1 x2 x3
2
2
2
+ 2+ 2
2
x
y
z
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
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
time
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
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
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
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
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
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.
( 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
D
( dri ) = dui
Dt
= {expand in Euler coordinates}
ui
drj
=
xj
where
ui
xj
drj
ui
ui
drj by dividing
in its invariant parts, i.e.
xj
xj
ui
= ij + eij + eij
xj
| {z }
eij
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
= Rkl
rkl
= Rlk +
component k of side l
uk l
R dt
xj j
| {z }
dulk
uk
= R kl +
dt
xl
deformed cube
rk1 rk1
"
u1
1+
dt
x1
2
u2
dt
x1
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
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
=
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
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
eij =
1
3
ur
ij
ur
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
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
:
:
:
:
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)
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
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)
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.
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
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
+
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 ):
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
force velocity
uj Rj dS
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
=
++
Dt
Dt
xi
xi
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
A slightly different version of the equations can be found by using the identity
Dtij
=
(tij ) +
(uk tij )
Dt
t
xk
+
(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
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))
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
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.
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
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
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
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
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
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
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.
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
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
wi
xi
in with
ui
wi
2p
=
;
xi
xi
xi xi
p
= wi ni on
n
xi
0 = u i ni = w i ni
and
wi
= 2 p,
xi
2) let ui = wi
p
n
p
xi
p
=0
n
p
xi
p
xi
p
=0
xi
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.
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
p
d2 u 1
+
x1
dx22
p
g
0=
x2
0=
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
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
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
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
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=
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 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.
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
2r
u0
y = 4t
t
y
()u0
z =
t
dxi = ti ds
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
p
up
(j uk ) = uk
j
xq
xk
xj
D
1 2
= ( )u +
Dt
Re
= 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.
k
xj
un
xj xm
2 un
= (im jn in jm )
xj xm
= ijk kmn
2 ui
xj xj
= 2 ui
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
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
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
L2
,
T
[c] =
1
T
F ()
y
p
/c
f (y)
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.
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
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)
=
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 =
u =
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.
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
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
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
u v
+
= 2 = 0
x y
u =
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
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
()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
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
= 2 ln r
The complex velocity can be written as
= m
2
W (z) = F 0 =
m
m i
=
e
2z
2r
ur =
2r
u = 0
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
= 2
= ln r
2
W (z) = F 0 = i
i
e
2r
ur = 0
u =
2r
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
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 :
:
:
:
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
u
u
1 p
2u
u
+v
=
+ 2
x
y
x
y
u + v = 0
x y
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
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
,
y
v=
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)
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
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
x.
z =
u0
t
L y
=
x/u0
}
u
0
f
0 = u0
Blasius
U profile
y, v
x, u
L
U
u0
= y
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
u
u
1 dp
2u
+v
=
+ 2
x
y
dx
y
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
!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
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
+ 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
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 ,
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.
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 .
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
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)
(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
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
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
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
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
1
= (i+1,j i,j ) y
x i+ 1 ,j
x y
2
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
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
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.
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
t
x
y
y
=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
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
Re
x2
Re
y 2
=
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 =
=
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
X
nb
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
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
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
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.
2
Q = 1 i sin () 2 sin
2
=
where s = sin2
|Q| = 1 4 2 2 s2 + 4 2 s2 s
= 1 4s 2 2 s 2 + 1
(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
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
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
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
I
0
N (u)1 G
I
I
0
(u)1 G
N
I
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
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
u = Q1 r1k
R p = r2k D
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
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
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
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
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
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
(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)
(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
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
T I
I T I
..
..
..
K =
.
.
.
I T I
I T
4 1
1 4 1
..
..
..
T =
.
.
.
1 4 1
1 4
Thus, a typical row in the matrixvector product Ku becomes
(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 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
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
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)
(4.15)
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
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
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
1/2
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
(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
(4.22)
M
m
m+2
1
h
V
Vh
u
uh
1
2
i
j
Inflow
Solid wall
n
0
1
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
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)
4.2.1
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 < +
Z
1/2
1/2
ui ui
d
xj xj
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
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
Np
X
(4.31)
n=1
Nu
X
vh (x, y) =
n=1
Nu
X
(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
T
v = v (1) , . . . , v (N u) ,
(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)
(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
(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
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
(4.40)
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)
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
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
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
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
1
h
V
Vh
u
uh
0
1
1
h
V
Vh
u
uh
0
1
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
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.
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)
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
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
Appendix A
Background material
io
j+1
: new values
di
re
ct
: to be updated
ra
tio
ite
: old values
i1
i+1
j1
2
= 1 O h2
h = x = y
each iteration. Requiring an error reduction to O h2 the number of
m = O h2
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
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 =
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
: old values
i1
i
i+1
j1
j
j+1
iteration direction
x
y
1
n
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
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
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
1 k=l
0 k=
6 l
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
= 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
(clk clm km ) xm = 0.
km x0m
cml x0m
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
A.2.3
Permutation tensor
klm
1
even
1
odd
=
if klm is
permutation of (1, 2, 3)
0
no
A.2.4
even permutation
odd permutation
no permutation
123 = 1
213 = 1
221 = 0
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
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 ,
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
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
T(kl)
T[kl]
1 3 5
= 3 5 7
5 7 9
0 1 2
= 1 0 1 .
2 1
0
5
n o
T kl = 0
0
and a traceless part
T kl
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
Tij
xk
T1l
T2l
T3l
Tkl =
+
+
xk
x1
x2
x3
curl
( u)i = ijk
uk
xj
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
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
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
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
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 =
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
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
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)
1 uz
u
ur
uz
1 (ru ) ur
er +
e +
ez .
r
z
z
r
r
r
(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
(A.14)
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)
(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
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
3
X
ui ui
i=1
= 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
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)(
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 = 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
or,
I
I
F n
dS =
Fi ni dS =
F dV
Fi
dV
xi
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
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
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
r
=u
,
t
r(t = 0) = x
0
=
|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
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.
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
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 .
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.
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
independent of t!
B.3
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
+
(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
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
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 }
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
+
(ui ) = 0
t
xi
b) Use this result to show
D
Dt
V (t)
Fijk dV =
V (t)
DFijk
dV
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 +
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 ,
h
V B.4
Vh
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
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
+
+ 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
r
r
r
z
=0
er :
0=
e :
u2
1 p
=
r
r
1 1 p
p = p(r, z) only.
r
0=
ez :
1 p
g
z
er :
ez :
1 p
r
p
= g
z
p = 2
r2
+ f (z)
2
f (z) = gz + C1
r2
gz + C1
2
r<a
er :
2 a4
1 p
=
r3
r
ez :
p
= g
z
p=
2 a4
+ f (z)
2 r2
f (z) = gz + C2
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
2 2
p0 = 2 a gz + C2
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
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.
+ (
u )
u = p + 2 u
,
t
u
= 0,
u
(U, 0, 0) as x2 + y 2 .
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
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
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
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
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)
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.
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
P 2
r + c1 ln(r) + c2
4
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
p
=0
y
Boundary conditions:
y=0:
v = V0
u = 0,
y:
u U
Continuity gives
u v
+
=0
x y
v = V0
V0
p
=0
x
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
f2 = p0 + g(h1 + h2 ) cos()
f2 = 0
f1 = p0 + g(h1 + h2 ) cos()
f1 = 0
d 2 u1
+ g sin()
dy 2
(1)
0 = 2
d 2 u2
+ g sin()
dy 2
(2)
u1 (0) = 0
du2
|y=h1 +h2 = 0
dy
du1
du2
|y=h1 = 2
|y=h1
dy
dy
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
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
i t
ky
cos(ky t),
where k =
f 00 (y)
2
Introduce k =
i
f (y) = 0
i
=0
=
=
,
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
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
= 0,
z
u = uz (r)ez ,
=0
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
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 .
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
(r u ) = r (r)
r
(B.1)
1
1 u uz
+
=0
(rur ) +
r r
r |{z}
z
|{z}
=0
(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
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
r
2
r2
+B
2
=
Case b) (r) = 0 er
/a
2
0
u r d = 2A A =
,
(1)
2
2
(r u ) = r 0 er /a
r
0 a2 r2 /a2
e
+C
2
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
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.
1
u
u
r
2 =0
r r
r
r
1
1
(rnrn1 ) rn2 = n2 rn1 rn2
r r
r
We get the inviscid flow,
u (r) =
Ar
|{z}
=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
=u
ur
uz
=
e
e
z
r
|
{z
}
r + ur
z + uz
e
e
= ur e
+
r |{z}
r |{z}
r |{z}
r |{z}
r
=0
=
e
=0
=0
=
+ (
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
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
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
2 r2
2g
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
A = U
B
=0
h
So we have:
= U
h2
r
r
B = U h2
sin
h2
u =
= U 1 + 2 sin
r
r
ur =
3 2
p = po U
gh
2
u
= 0 or
Define the stream function such that:
u=
This means
v=
u v
2
2
+
=
= 0,
x y
xy yx
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
r2
1
sin
u =
1
r sin r
=
Irrotational
e
r sin
2
1
sin
1
+
r sin r2
r2 sin
2 sin
+ 2
r2
r
1
sin
=0
2
f =0
r2
1 2 B
Ua +
=0
2
a
This gives
=
The slip velocity on the sphere is
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)
,
x
u=
v=
The flow is always irrotational due to the definition of the velocity potential
=u
= = 0,
curl(grad)=0
()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 +
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.
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
1
2
(U 2 |
u|2 ) 6.86 N/m 0.07 mbar
2
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
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
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
+
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
3
4
4
1
z=f (Z)
B.10
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
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
(a + )2 i
i
dF
= U ei U
e +
2
dz
(z + )
2(z + )
Ue
(a + )2 i
i
e +
U
2
(z + )
2(z + )
a2
1 2
z
u = Re{(z)},
2 /2 a2 into
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
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 |
,
y
u=
v=
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=
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)
u
1 p
2u
u
+v
=
+ 2
x
y
x
y
u v
+
=0
x y
p
= 0.
x
u1
=
=
U + u1 =
y
x
x
x
u1 << U
L
u1
u1
u1
2 u1
+ u1
+v
=0+
x
x
y
y 2
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:
u1 dy
where =
F (x)f () dy = F (x)g(x)
y
g(x)
f () d
Q1 = constant requires,
Z
f () d = Q1 = constant
1
y
1
f () =
f
u1 (x, y) =
g(x)
g(x)
g(x)
f () + f ()
1/2
x
g(x) = 2C
U
y u(x, 0)
d
f 0 () + Cf () + D = 0
d
f () = ae 2 C
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
The drag is
FD =
u1 (U + u1 ) dy
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
UL
103
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 =
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)
l/2
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
l/2
Uw (y) dy
l/2
l/2
l/2
l/2
Uw (y) dy +
l/2
l/2
l/2
Uw2 (y) dy + FD = 0
l/2
l/2
Uw (y)
U0
Uw (y)
U0
2
dy
l/2d
l/2d
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
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=
Um (x)g(x)
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
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
g = bxn
1
ab2
= amxm1 b2 x2n =
m,
2n + m = 1
1
1
ab2
amxm1 b2 x2n + axm bnxn1 bxn =
(m + n)
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
gf 0 d = 0
f 02 d > 0
f 00 f 0 + f f 02 + ( 1)gf 0 = 0
00 0
=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
U
U
1 P0
2U
+V
=
+ 2
(uv)
x
y
x
y
y
(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
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
l = Bxn
Then we get
Bxn Amxm1 = const x2m
n + m 1 = 2m
Us = Axn1 ,
m=n1
l = Bxn
1
U0
1
U0
U1 dy =
1
CD d = const (Recitation 12)
2
Z
U1 dy = Us l
f d = const
|
{z }
const
Us = Ax 2 ,
2n 1 = 0
n=
1
2
l = Bx 2
AB
lUs
=
= const
= 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
B.14
Old Exams
i
j
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)
1g
4
b2 r 2
r
r2 a2 (b2 a2 )
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.
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
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
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
2
1.5
1
0.5
0
0.5
1.5
0.5
1.5
2.5
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
3
2 g
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
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
Thus:
ij
=
xj
xj
ui
uj
+
= 2 ui
xj
xi
ij
k
= ijk
xj
xj
Irrotational flow is not subject to viscous forces.
if (y)ei t = f 00 (y)eit
f 00 (y)
2
Introduce k =
i
f (y) = 0
i
=0
=
=
,
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)
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
+ 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
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
=
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 .
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.
[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
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
r
equations. From the continuity equation we get
uz
=0
z
uz = uz (r, )
only.
uz
z
uz
uz
= 2 uz g
z
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
1g 2
r + c1 ln r + c2
4
ga2
and we have
4
g 2
(a r2 )
4
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
dF
a2
i
= U ei U 2 ei +
dz
z
2z
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)
,
y
v=
(x, y)
U (x)
= U f
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
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 =
U
= f (y + ),
u
where y + =
y
l
k = 21 ui ui [L2 T 2 ]
u u
= i j [L2 T 3 ]
xj xi
uT km n
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
l
So we have
3
k 2
p k 32
T = uT l k
k
T
or
T = C
k2
.
Appendix C
ut + uux
t>0
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
t n n
u (u unj1 )
x j j
t n
(u unj1 )
x j
(C.1)
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)
0
0
+ 2
= 0,
u t
a /0 0
u x
|
{z
}
(C.4)
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
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
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.
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