The Control-Volume Finite-Dierence

Approximation to the Diusion Equation

ME 448/548 Notes
Gerald Recktenwald
Portland State University
Department of Mechanical & Materials Engineering
gerry@pdx.edu
21 January 2014

ME 448/548: 2D Diusion Equation

Motivation

Numerical solution of the 2D Poisson equation is the next step in developing our
knowledge of CFD technique.

Introduce the Finite Volume Method

. Naturally deal with material discontinuity
. Can naturally enforce conservation of mass and energy
. Core idea in many (not all) commercial CFD codes
Extend analysis to two spatial dimensions
. More interesting practical applications
. More complex data structures
. More complex procedures to solve the Ax = b problem

ME 448/548: 2D Diusion Equation

Overview
Goals for this unit

Introduce the Poisson equation

. A model of steady heat conduction with a source term
. Form of the pressure equation for incompressible flow
. Precursor to the generalized advection diusion equation

Use the finite volume method to obtain discrete equations

Allow for non-uniform mesh, diusion coefficient, and source term.
Introduce a set of Matlab codes for 2D Control Volume Finite Dierence (CVFD)
Demonstrate solutions for three model problems.
Measure truncation error for a model problem with a simple solution
Apply to fully-developed flow in rectangular ducts

ME 448/548: 2D Diusion Equation

Model Problem
The two dimensional diusion equation in Cartesian coordinates is

@
@
@
@
+
+S =0
@x
@x
@y
@y
where

is the scalar field,

(1)

is the diusion coefficient, and S is the source term.

Example: Heat conduction in a rectangular domain

= k, thermal conductivity or

k
=
cp

S = volumetric heat source

ME 448/548: 2D Diusion Equation

2D Cartesian Finite Volume Mesh

"$%&'()(*
"$%!".&'

"%+'
"$%+'

#-%!#.&'

!#

!"(!#

#%,'

#-%,'

Interior node
Boundary node

j=0
i=0

!"+1

!"+2

Ambiguous corner node

!"
#-%&'()(*

!"

ME 448/548: 2D Diusion Equation

Finite Volume Mesh: Compass Point Notation

Nodes are labelled with their position
relative to the typical point P.
N

y
x

Capital letters designate node locations.

N, S, E, W are the names of north,
south, east and west neighbors.
at
N , S , E , W are the values of
the north, south, east and west neighbor
nodes.
Use of compass point names simplifies
the algebra. For example the value of
at point N is N or i,j+1, but N is
simpler to write.
Compass point notation is an historic
convention that does not extend to
unstructured meshes.

ME 448/548: 2D Diusion Equation

Finite Volume Mesh

Detailed notation for a typical 2D control
volume

xw

xe and xw are the x coordinates of the

east and west faces of the control
volume around P

xe

N
yn

yn
P

E
y

ys

ys
xw

Lowercase letters designate the interface

between the nodes.

xe

yn and ys are the y coordinates of the

north and south faces of the control
volume around P
Similarly, n, s, e, w are the values
of at the north, south, east and west
control volume interfaces.

ME 448/548: 2D Diusion Equation

Finite Volume Mesh

Detailed notation for a typical 2D control

volume

xw

xe

Regardless of whether the control

volumes sizes are uniform, node P is
always located in the geometric center
of each control volume.
Thus,

N
yn

xP

xw

xe

yP

ys

yn

yn
P

E
y

x
2
y
yP =
2

xP =

for uniform or non-uniform meshes

ys

ys
xw

xe

S
ME 448/548: 2D Diusion Equation

Finite Volume Mesh

Detailed notation for a typical 2D control

volume

xw

In contrast, the distances between the

nodes is not assumed to be uniform

x e = xE

xP 6= xw = xP

y n = yN

yP 6= ys = yP

xe

xW
yS

N
yn

yn
P

x e = xE

xP

6=

x w = xP

y n = yN

yP

6=

y s = yP

y
ys

yS

Note the use of upper and lower case

subscripts: Upper case refers to nodes;
lower case refers to interfaces.

ys
xw

xW

xe

S
ME 448/548: 2D Diusion Equation

Convert the Dierential Equation to a Discrete Equation

Integrate over the control volume

yn
ys

xe
xw

@
@x

@
@x

yn

ys

@
@x

dx dy

xe

@
@x

xe

yn

@
@x

E
e

xw

@
@x

E
y

dy

ys

ys
xw

xe

P
w

yn

xw

The final step is obtained by using central dierence approximations for the derivatives at
the interfaces.
ME 448/548: 2D Diusion Equation

Convert the Dierential Equation to a Discrete Equation

Integrate the source term

xe
xw

yn
ys

S dy dx SP

(2)

Note: The Control Volume Finite Dierence (CVFD) method treats the source term and
diusion coefficients as piecewise constants. This is a rather crude approximation, say,
compared to allowing the source term and diusion coefficient to vary linearly within the
control volume.
However, piecewise constant profiles of S and allow the method to be conservative,
i.e., conserving mass or energy, automatically. The conservative nature of the CVFD
method is one of its primary strengths.

ME 448/548: 2D Diusion Equation

10

Convert the Dierential Equation to a Discrete Equation

Putting pieces back into the model equation gives the discrete system of equations

aS

aW

+ aP

aE

aN

(3)

=b

where

aE =

x xe

aW =

x xw

aN =

y yn

aP = aE + aW + aN + aS
b = SP

aS =

y ys
(4)
(5)

This is not a tridiagonal system of equations.

ME 448/548: 2D Diusion Equation

11

Non-uniform
Continuity of fluxes at the interface requires
P

@
@x

xe

@
@x

xe +

material 1

@
@x

xe

material 2

Use central dierence approximations

E
e

xe
E

(6)

xe

xe-

xe

xe

(7)

xe+

xe+

ME 448/548: 2D Diusion Equation

12

Non-uniform
Equations 6 and 7 can be rearranged as
e

xe

xe

xe+

xe

material 1

(8)

P)

material 2

(9)

P)

Add Equation 8 and Equation 9

xe-

xe

Cancel the factor of (

e

xe

P)

P)

ME 448/548: 2D Diusion Equation

xe
P

xe

xe+

xe

and solve for

xe+

xe+

e/

xe to get

xe

+ xe+

.
P

13

Non-uniform
Thus, the diusion coefficient at the interface that results in flux continuity is
e

=
E

where

+ (1

xe
xe
=
xe
xE

An analogous derivation gives formulas for

w,

n,

(10)
P

xP
xP
and

(11)
s.

ME 448/548: 2D Diusion Equation

14

Solving the System of Equations

Regardless of uniform or variable , the discrete equation has a five-point stencil, and the
discrete equation for any interior node can be written.

aS

aW

+ aP

aE

aN

=b

(12)

To set up the matrix for this system of equations, we need to re-number the unknowns.
Important: i and j subscripts for the mesh are not the same as the row and
column indices in the system Ax = b.

ME 448/548: 2D Diusion Equation

15

Solving the System of Equations

Order the nodes:

"$%&'()(*
"$%!".&'

"%+'

n = i + (j

"$%+'

1)nx

#-%!#.&'

!#

n is the node number (row number)

in Ax = b. i and j are the mesh
indices corresponding to xi and yj .

!"(!#

#%,'

#-%,'

Interior node
Boundary node

With natural ordering the neighbors

in the compass point notation have
these indices
np = i + (j-1)*nx
ne = np + 1
nw = np - 1
nn = np + nx
ns = np - nx

!"+1

!"+2

j=0
i=0

Ambiguous corner node

!"
#-%&'()(*

!"

ME 448/548: 2D Diusion Equation

16

Solving the System of Equations

This leads to a vector of unknowns
0
1

B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@

ME 448/548: 2D Diusion Equation

1,1
2,1

...

nx,1
1,2
2,2

...

i,j

...

nx,ny

C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A

()

B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@

1
2

...

nx
nx+1
nx+2

...

...

1
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A

(13)

17

Solving the System of Equations

The coefficient matrix has 5 non-zero diagonals

A=

aS

a W ap

aE

aN

ME 448/548: 2D Diusion Equation

18

Algorithm for obtaining the numerical solution

1. Define physical parameters: Lx, Ly ,
2.
3.
4.
5.

(x, y) and boundary conditions

Define the mesh: nx and ny if uniform
Compute the coefficient matrix
Solve the system of equations A = b, where is the vector of unknowns.
Post-process to visualize the solution

The Poisson equation is steady. Each step is performed only once.

ME 448/548: 2D Diusion Equation

19

Structured Mesh
The CVFD Matlab codes use structured meshes. in size.

The cells in the domain are topologically equivalent to a rectangular array

The cells need not be uniform in size.
Each cell not on a boundary touches four other cells
Each row has the same number of cells
Each column has the same number of cells
Block-Uniform Mesh

Uniform Mesh

Ly3, ny3
y
Ly2, ny2
Ly1, ny1

y
Lx1, nx1

Lx2, nx2

ME 448/548: 2D Diusion Equation

20

Boundary Conditions (part 1)

Boundary
type

Boundary
Condition

Post-processing in fvpost

Specified T

Compute q 00 from discrete approximation to Fouriers law.

T
Ti
00
q =k b
xb xi
where Ti and Tb are interior and boundary temperatures, respectively.

Specified q 00

Compute Tb from discrete approximation to Fouriers law.

xi
00 xb
Tb = Ti + q
k
where Ti and Tb are interior and boundary temperatures, respectively.

ME 448/548: 2D Diusion Equation

21

Boundary Conditions (part 2)

Boundary Boundary
type
Condition
3

Convection

Post-processing in fvpost
From specified h and T1 , compute boundary temperature and heat flux
through the cell face on the boundary. Continuity of heat flux requires

T
k b
xb

Ti
= h(Tb
xi

Tamb )

which can be solved for Tb to give

Tb =
where xe = xb
4

Symmetry

hTamb + (k/ xe )Ti

h + (k/ xe )

xi

q 00 = 0. Set boundary Tb equal to adjacent interior Ti .

ME 448/548: 2D Diusion Equation

22

Matlab codes for obtaining the numerical solution

A set of general purpose codes has been written to facilitate experimentation with the
CVFD method.
Algorithm Tasks

Core Routines

Define the mesh

fvUniformMesh or
fvUniBlockMesh

Define boundary conditions

Compute finite-volume coefficients for interior cells

fvcoef

Adjust coefficients for boundary conditions

fvbc

Solve system of equations

Assemble coefficient matrix

fvAmatrix

Solve
Compute boundary values and/or fluxes

fvpost

Plot results
ME 448/548: 2D Diusion Equation

23

Model Problem 1

Choose a source term that may be physically unrealistic, but one that gives an exact
solution that is easy to evaluate
" 2 2#

2
x
2y
S=
+
sin
sin
Lx
Ly
Lx
Ly
The exact solution is

= sin

x
Lx

sin

2y
Ly

Main code to solve this problem is in demoModel1.m

ME 448/548: 2D Diusion Equation

24

Model Problem 1
The exact solution is
2

= sin

x
Lx

sin

2y
Ly

1.5

0.5

0
1
0.5
0
x

ME 448/548: 2D Diusion Equation

0.8

0.6

0.4

0.2

25

Solutions to Model Problem 1 Show Correct Truncation Error

The local truncation error at each node is
ei O( x2).
Since the exact solution is known we can
compute

e2i

Measured
Theoretical error ~ ( x)3

10

10
Measured error

kek2
=
N

qP

10

N e2
e
=p .
N
N

10

10

10

10

where N = nxny is the total number of

interior nodes in the domain, and e is the
average truncation error per node.
Since ei O( x2), N n2x, and

10

10

10

10

10

x = Lx/(nx + 1), we can estimate

O L2x/(nx + 1)2
kek2
e
O( x2)
p =
=
O
N
nx
nx
N

1
nx

= O( x ).

ME 448/548: 2D Diusion Equation

26

Model Problem 2

Uniform source term: S = 1.

0.08

Analytical solution is an infinite

series
Code in demoModel2.m

0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
1
0.5
0
x

ME 448/548: 2D Diusion Equation

0.8

0.6

0.4

0.2

27

Model Problem 3
Heat conduction in a rectangle consisting of
two material regions.

Inner rectangle with high conductivity

Outer rectangle with low conductivity
Inner region has uniform heat source

0.5Ly

Ly

2 = 1

1 = 1

S2 = 1000

S1 = 0

0.25Ly
0.5Lx

0.25Lx

Lx

The discontinuity and dierence in material properties can be used to stress the solution
algorithm.
2

The analytical solution does not exist. Code in demoModel3.m

ME 448/548: 2D Diusion Equation

28

Model Problem 3

Solution with = 100

35
Ly

0.5Ly

2 = 1

1 = 1

S2 = 1000

S1 = 0

30
25

0.25Ly

20
0.25Lx

0.5Lx

15

Lx

10
5
0
1
0.5
0
x

ME 448/548: 2D Diusion Equation

0.8

0.6

0.4

0.2

29

Model Problem 4: Fully Developed Flow in a Rectangular Duct

For simple fully-developed flow the governing equation for the axial velocity w is
"
#
@ 2w
@ 2w
dp

+
=0
2
2
@x
@y
dz
This corresponds to the generic model equation with

= (= constant),

= w,

S=

dp
.
dz

ME 448/548: 2D Diusion Equation

30

Model Problem 4: Fully Developed Flow in a Rectangular Duct

The symmetry in the problem allows alternative ways of defining the numerical model
Full Duct

Quarter Duct
Ly

Ly

y
x

y
Lx

x
Lx

For the full duct simulation depicted on the left hand side, the boundary conditions are no
slip conditions on all four walls.

w(x, 0) = w(x, Ly ) = w(0, y) = w(Lx, y) = 0.

ME 448/548: 2D Diusion Equation

(full duct)

31

Model Problem 4: Fully Developed Flow in a Rectangular Duct

The symmetry in the problem allows alternative ways of defining the numerical model
Full Duct

Quarter Duct
Ly

Ly

y
x

y
Lx

x
Lx

For the quarter duct simulation depicted on the right hand side, the boundary conditions
are no slip conditions on the solid walls (x = Lx and y = Ly )

w(Ly , y) = w(x, Lx) = 0

(quarter duct)

and symmetry conditions on the other two planes

@u
@x
ME 448/548: 2D Diusion Equation

=
x=0

@u
@y

= 0.
y=0
32