Sharif University of Technology

School of Mechanical Engineering

Center of Excellence in Energy Conversion



Computational Fluid Dynamics

Basic Governing Equations


M.T. Manzari
Conservation laws


Many of physical laws are described as conservation laws.

S
S d
r
F
r
V
Q
S
Q
Conservation laws

For a typical scalar quantity,U , conservation law can be
written as:
48 47 6 48 47 6 8 7 6
48 47 6
Tractions Surface source volumetric
flux surface incoming
time unit per variation
. .

+ = +

S
S V
S
d d Q d d
t
S Q S F U

Using Gauss theorem,


= d d
S
A S A . . , we get


+ = +

d d Q d d
t
S V
Q F U . .

Or in differential form, we get
S V
Q
t
Q F
U
. . + = +

Conservation laws


If U is a vector, then the flux Fbecomes a tensor and we get

+ = +

S
tensor
S V
S
tensor
d d d d
t
S Q Q S F U . .
Or in differential form:
tensor
S
vector
V
tensor
t
Q Q F
U
. . + = +

Conservation of mass:
( ) 0 . 0 . = +

= +

V S V
t
d d
t
S

Conservation of momentum:
( )
( ) ( )
e
S
tensor
e
S
t
dS d d d
t
f V V V
f S V V V


= +

+ = +

.
.


Conservation laws

Conservation of Energy: 2 /
2
V + = e E

( )
stresses internal
by done work
source
heat
force volume
of work
conduction convection
. . . .

+
|
|
|
.
|

\
|
+ + = +

S
H e
S S
d d d T K d E d E
t
S V q V f S S V

Its differential form becomes

( ) ( ) ( ) ( )
H e
T K E E
t
q V f .V V + + + = +

. . . .


Or by introducing / 2 / /
2
P E P e H + = + + = V ,
we have
( ) ( )
H e
T K H E
t
q V f V V + = +

. . .





Conservation Forms

The conservation of equations can be
written as

}
} }
}
source
i
fluxes viscous
i
i
fluxes Inviscid
i
Variable
ve Conservati
x x t
S
G F U
+

If a non-conservative form is used for a
numerical scheme, it can easily lead to
violation of conservation laws.
For example, the mass balance may not be
satisfied.
Lax has shown that the use of non-
conservative form will lead to in-
accurate jump relations through a
discontinuity.
In general, there are 3 form for
equations; conservative form, primitive
form and characteristic form.
Euler Equations

For an inviscid flow, we have

S F
U
= +

.
t


A Quasi-linear form of the equation is

}
S U
U
F U
A
=

.
t


For a Cartesian coordinate system
) , , ( z y x we get

U
F
C
U
F
B
U
F
A
Q
U
C
U
B
U
A
U

=
=

z
y
x
by given Jacobians the with
z y x t

Euler Equations

For a fluid with the constitutive
relation as ) (e f p = where e is the
internal energy,
y x
F F , and
z
F are
homogenous functions of degree one of the
conservative variables, U:

any for ) ) F(U U F( =

By differentiating w.r.t and setting
1 = , we get (prove!)

CU F BU F AU F U A U
U
F
F(U) = = = =

=
z y x
or , ,


In other words, the Euler equation for
such fluids can be written as

0
) ( ) ( ) (
=

z y x t
CU BU AU U

Euler Equations

This form and the quasi-linear form can
be equivalent as long as the functions
are continuous.
However, from the numerical point of
view, the two formulations do not lead to
identical discretization.
For a perfect gas, we have:
(
(
(

=
(
(
(

+ =
(
(
(

=
=
=

u u E uE u
u u
uH
p u
u
E
u
e p
where
x t

2 / ) 1 ( 3 ) 1 (
1 ) 3 ( 2 / ) 3 (
0 1 0
) 1 (
0
2 3
2
2
A
F U
F U


Primitive Variables

Using the vector | |
T
p w v u V = as the
vector of primitive variables, we get

(
(
(
(
(
(

=
=

= +

u c
u
u
u
u
example for with
z y x t
or
t
0 0 0
0 0 0 0
0 0 0 0
/ 1 0 0 0
0 0 0
~
~ ~ ~ ~
~
) .
~
(
2

A
Q
V
C
V
B
V
A
V
Q V A
V


We note that this Jacobian is in a much
simpler form.
Primitive Variables

The Jacobian of transformation from the
conservative to primitive variables is
defined by (M is a similarity transformation)

1
,

M
V
U
M

Which results in

Q V . M A
V
M = +

t


Multiplying by
1
M gives:

8 7 6 48 47 6
Q
1
A
Q M V . M A M
V
~
~
1
= +

t



Primitive Variables

For a 1D case, we get:

(
(
(

=
(
(
(

=
(
(
(

u c
u
u
u u
u
u u
u
2
2
1
2
0
/ 1 0
0
~
1 ) 1 ( 2 / ) 1 (
0 / 1 /
0 0 1
) 1 /( 1 2 /
0
0 0 1

A
M
M





Characteristic Variables

Define the left and right eigen-values of
A
~
in a direction
~
: (i.e.
~
.
~ ~
A K = )

( ) ) only i on summation ( 5 ,..., 1 , ,
~
~
.
~ ~
) j on summation no (
~ ~ ~
0
~
.
~
) ( ) (
) ( ) (
= =
=
=
k j i l l
or
l l
j
k j ik
j
i
j
j
j

A
K
A I



A wave like solution will exist if the
eigen-values of K
~
for arbitrary
~
are
real with linear independence of the
corresponding left eigen-vectors.



Characteristic Variables

The Jacobian matrix can be diagonalized
using L matrix whose rows are the left
eigen-vectors of
) (
~
j
l :

(
(
(
(

=
= =

n
j
and of functions are where
or

0
0
) .
~
(
~
2
1
O

L A L L K L
1 1 1


Therefore, we can decouple the equations
in the direction
~
.
Since K
~
is not symmetric, there exists a
set of right eigen-vectors:
)
~
are columns se matrix who a is R (
~
) j on summation no (
~ ~
~
) ( 1
) ( ) (
j
j
j
j
r
r r

=
=
R R K
K

Characteristic Variables

Multiplying the eigen-vectors
) (
~
j
l in the
primitive equation, we get:

) for equation ity compatibil (
~ ~
) .
~
(
~ ~
j
) ( ) ( ) (
Q V A
V
j j j
l l
t
l = +

Or after grouping these equations for all
eigen-values, we obtain:

( ) Q L V . A L L
1 1
t
1
~ ~

= +

Substituting MU V = we see:

Q M L U ) A. ( M L
1 1
t
1 1
= +
Defining
ML P M L P
1 1 1
= =

and
We get
AM M A
1
=
~

So that P will diagonalize A. K = :

( ) KP P ML A. M L
1 1 1
= =
Characteristic Variables

Note that the rows of
1
P are the left
eigen-vectors of K while the columns of
Pare the right eigen-vectors of the same
matrix.
Defining a new set of characteristic
variables as
) ( =

or
t
V L W
1


Or explicitly

=
i
i
k
i k
v l w
) (

Then, we have W L V = .
The compatibility equation becomes

( )
( ) Q L W L A L
W
Q L V L L A L
V
L
1
1
~ ~
~ ~
1
1 1 1


= +

= +

t
or
t

Characteristic Variables

Note that the characteristic variables
are now associated with a given direction
of propagation and therefore are a
function of .

Also, note that except when eigen-values
are constant, we cannot directly evaluate
W as V L W
1
= and in other cases we only
work with W .

The variables W are also called Riemann
variables.

Whenever, W remains constant, it is
called Riemann invariant.

Characteristic Lines

At each point in the domain, we can draw
three lines corresponding to the three
characteristic directions.

C and
+
C are called Mach lines.
















t
C
+
C-
C
0
c u
dt
dx
=
u
dt
dx
=
c u
dt
dx
+ =
x
Shock and Contact Discontinuity

In the case of an isentropic flow we have
1 2
= =

k c k p

This gives
- C for
1
2
C for
1
2
3
2

=
+

+ =

c
u w
c
u w


P(x,t)
P- P
0
P+
Supersonic flow
P(x,t)
Subsonic flow
P- P
0
P+
C
0
C
-
C
+
Shock

The area between the characteristic lines
shows the region of dependence of the
solution at point P.

The region between the characteristics
emanating from P is called the domain of
influence of P.

We also have

=
|
|
.
|

\
|

=
|
|
.
|

\
|

|
|
.
|

\
|

+ =
|
|
.
|

\
|

+
+
0
0
1
2
1
2
1
2
1
2
C along S S
C along
c
u
c
u
C along
c
u
c
u
P P
P P
P P




Shock

Due to the nonlinearity of the flow
equations, the streamline slopes may
decrease (in particular if 0 / ) ( < + x c u )
and we have the situation where the
+
C
characteristic emanating from
+
1
P
intersect the
+
C characteristic from
+
P
and multi-valued quantities would occur
in
1
P .

t
x
1
P
P
0
P

P
+
P
+
1
P
0
1
P

1
P
+
C
+
C
Shock

In this case we have

+ +
+
|
|
.
|

\
|

+
|
|
.
|

\
|

+
|
|
.
|

\
|

+ =
|
|
.
|

\
|

+
|
|
.
|

\
|

+ =
|
|
.
|

\
|

+
+
1
1 1
1
1
2
1
2
1
2
1
2
1
2
1
2
P P
P P
P P
c
u
c
u
c
u
c
u
c
u
c
u





This impossible situation leads to a
discontinuous flow behaviour called a
shock wave.
From this we find that for a shock to
occur the following relations holds

( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
+ + + +
+ + + +
+
+
> >
> >
+ < < +
1 1
1 1
1
P P P P
P P P P
P P
c c p p
u u
c u c c u

Shock

If
+ +
<
1
) ( ) (
P P
P P then an expansion fan will
occur.

A contact discontinuity is an interface
between two fluid regions of different
densities but equal pressures. The
velocity is continuous over a contact
discontinuity. (like a free moving
piston)

A hypothetical Expansion Shock would lead
to a situation where

( ) ( )
+ +
+ < < +
1
P P
c u c c u


And characteristics carry information
away from the discontinuity.



Physical Boundary Conditions

The number of boundary conditions to be
imposed will depend on the way the
information transported along the
characteristics interacts with the
boundaries.

Subsonic Supersonic
Inlet 2 conditions
given W and W
2 1

3 conditions
given W and W W
3 2 1
,
Outlet one conditions
given W
3

None

As numerical schemes require the values
of all flow variables on the boundaries,
additional conditions called Numerical
Boundary Conditions must be given in
order to define the numerical problem
completely.


Physical Boundary Conditions

The choice of boundary conditions has a
significant effect on the accuracy,
stability and convergence rate of many
schemes.

Many implicit schemes which are linearly,
unconditionally stable, appear to be only
conditionally stable in practice if an
improper boundary treatment is
introduced.











Characteristic Boundary Conditions

In many numerical schemes we work with
the primitive variables, and need to know
which combinations of the primitive
variables may be applied as physical
boundary conditions that do not lead to
an ill-posed problem.

We know that V L w =
1
, therefore

( ) ( )
( ) ( )
(

(
(

=
(

=


II
I
N
II
N
I
P
II
P
I
N
P
V
V
L L
L L
w
w
w
1 1
1 1


Here, P denotes physical and N denotes
numerical boundary conditions.

The group of variables
I
V represents the
imposed conditions while the group
II
V
represents the free variables to be
defined by the numerical or internal
information.
Boundary Conditions

The condition for well-posedness of the
choice of variables is that
II
V can be
recovered from the information carried by
the characteristics
N
w which intersect
the boundary from the interior of the
flow domain.
If we write

II N
II
I N
I
N
V L V L w + =

) ( ) (
1 1


The free variables
II
V are defined by

| |
I N
I
N
N
II
II
V L w
L
V =

) (
) (
1
1
1


Hence, the condition for well-posedness
is that the matrix
N
II
L ) (
1
is non-singular,
i.e.:
0 ) (
1

N
II
L
1D Euler Equation


For this equation, we have
(
(
(

(
(
(
(
(
(
(

=
(
(
(

(
(
(

=
(
(
(
(
(
(
(

=
(
(
(

=
u
p
c
c
c
w
w
w
p
u
c
p
u
c
p
u
c
p
w
w
w

1 0
1
0 1
1
1 0
1
2
2
1
3
2
3
2
1
V w

Let us consider a subsonic outflow. Then,
( ) ( ) | | ( ) ( )
(

=
(
(
(
(

= =

=
(

= =
(

= =

1 0
0 1
1
1
1 0
1
1
2
1 1 1
2
1
3
N
II
N
I
P
II
P
I
II I N P
c
c
c
u
p
w
w
w
L L L L
V V w w

1D Euler Equation

We can see that any three variables u , and p can be
chosen as a physical boundary condition, since none of the
sub-matrices
n
w is zero.

For a subsonic inlet, the choice ) , ( p u as a physical boundary
condition is not well-posed. Any other combination involving
as a physical condition is well-posed.

For a steady-state subsonic nozzle flow with equal inlet and
outlet areas leads to non-unique solutions if the same variable
is specified at outlet and inlet.





Boundary Conditions and Accuracy

Gustafsson proved that, for linear equations, the boundary
scheme can be one order lower than the interior scheme
without reducing the global order of accuracy of the complete
scheme.

Important types of Boundary Conditions are:
o Far field
o Inviscid wall (slip b.c.)
o Viscous wall (no-slip b.c.)
o Symmetry plane
o Adiabatic wall
o Isothermal wall
o Injection or suction wall
o Moving wall


Simplified Forms of the Navier-Stokes
Equations

Several simplified forms of the Navier-Stokes equations exist.
These forms help to get a better mathematical behaviour from
the governing equations.

This normally results in using a simplified and specific
numerical method for the solution.

Here, we introduce the following approximation types:
o Thin Shear Layer (TSL)
o Parabolized Navier-Stokes (PNS)
o Boundary layer
o Inviscid Flow Model (Euler)




Thin Shear Layer Equations


When the viscous layers (wall shear layer, wake or free shear
layers) are of limited size, then the influence of the shear
stresses will come essentially from the gradients transverse
to the main flow direction.
This approximation is normally reasonable for high Reynolds
numbers.
In this case, the contribution of other gradients on the
stresses is neglected.

Viscous dominated region
z n
Thin Shear Layer Equations

For example, if the direction normal to the wall is z , we get:
|
.
|

\
|

|
|
.
|

\
|

z
v
z
v
z
v
z
ij
i
j
j
i ij

3
2
3 3

Then, momentum equation becomes:
|
.
|

\
|

= +

|
.
|

\
|

+ +

|
.
|

\
|

+ +

z
w
z
p w
z
vw
y
uw
x
w
t
z
v
z
vw
z
p v
y
uv
x
v
t
z
u
z
uw
z
uv
y
p u
x
u
t



3
4
) ( ) ( ) ( ) (
) ( ) ( ) ( ) (
) ( ) ( ) ( ) (
2
2
2


In a flow with large separation or wake, TSL cannot be used.
Parabolized Navier-Stokes Equations

This is similar to TSL but is applied to the steady-state
formulations.

In this case, the following conditions exist:

o A pre-dominated main flow direction (such as channel
flow) exists such that the cross flow components are of
lower order of magnitude.

o Along the solid boundaries the viscous regions are
assumed to be dominated by the normal gradients,
hence, the stream-wise diffusion of momentum and
energy can be neglected.

Parabolized Navier-Stokes Equations

Consider a flow with dominated velocity in the direction x
while the flow is 3D y x, and . z The x -derivatives in the
shear stress terms are all neglected compared to other terms.

We get (for x -direction) the following parabolic equation:
|
.
|

\
|

+
|
|
.
|

\
|

+ +

z
u
z y
u
y
uw
z
uv
y
p u
x
) ( ) ( ) (
2

Here, if the term x x u / ) / ( were present, we had an
elliptic equation in ) , , ( z y x space.

However, now, the equation ca be solved for x as a pseudo-
time variable.

The equation can be integrated by advancing in x - direction,
while solving an elliptic problem in z y plane.
Parabolized Navier-Stokes Equations

In the same way, we get:

|
.
|

\
|

+
|
.
|

\
|

+
|
|
.
|

\
|

+
|
|
.
|

\
|

|
|
.
|

\
|

+
|
.
|

\
|

|
.
|

\
|

+
|
|
.
|

\
|

+ +

z
T
k
z z
T
k
y
z y z
z y y
wH
z
vH
y
uH
x
y
w
z z
w
y
z
v
z y
v
y
vw
z
p v
y
uv
x
zz
yz
yz yy

) ( ) ( ) (
3
2
3
4
) ( ) ( ) (
2


Boundary Layer & Inviscid Equations

Boundary Layer:

The idea was introduced by L. Prandtl and in this method the
flow is solved as a viscous region separated from an adjacent
inviscid region.
Pressure is normally computed from outside the boundary
layer using a potential flow model.

Inviscid Flow:

All viscous terms are neglected in the whole flow regions.
In addition all heat conduction terms are dropped from the
energy equation.