MENG314-FV

THE FINITE-VOLUME-METHOD

The finite difference method is a Taylor’s

series approach which divides the
continuum into nodes. The underlying
P.D.E. is then approximated node-wise
using finite-differences. It is simple and
straightforward for orthogonal grids with
simple boundary conditions, but not
always, e.g. case sketched(try developing
Laplace’s eqn for node (2,4)!)

The finite volume method

approximates the P.D.E. over a
control volume surrounding the
grid node, rather than at the node
itself. The discretization equations
are obtained by integrating the
P.D.E. over control volumes
surrounding the grid nodes, after
introducing necessary simplificat-
ions and assumptions. It often leads to the same discretization equations
as the Taylor series method, however it is more flexible . It bears much
commonality with the Galerkin Finite-element method but is easier to
implement. In the F.V. method the integration domain is covered by
control volumes; each control volume engulfs one grid node, which lies
on a grid mesh. With gradient type boundary conditions, we need to solve
over a C.V. surrounding boundary point, but there is no need to introduce
external imaginary nodes .

Control Volumes
surrounding grid
nodes and
covering whole
domain.

1
 MENG314-FV

Given that steady state heat conduction is described by the following

Laplace’sequation : ∂ ∂T ∂ ∂T
(k ) + (k ) = 0
∂x ∂x ∂y ∂y
which for k= constant reduces to :
∂ 2T ∂ 2T
k( + ) = 0
∂x 2 ∂y 2
Integrating above equation over a regular control-volume yields:

y+∆y x+∆x y+∆y x+∆x y+∆y x+∆x

∂2T ∂2T ∂2T ∂2T
k( 2 + 2 ).dx.dy = k( 2 )dx.dy + k( 2 )dx.dy = 0
y x ∂x ∂y y x ∂x y x ∂y
(1)

where the limits of integration refer to the

control volume faces, as sketched.

The first term on R.H.S. may be

integrated in x-direction as follows:
y + ∆y x + ∆x y + ∆y
∂ 2T ∂T ∂T
k ( 2 )dx.dy = [( k ) x + ∆x − ( k ) x ].dy
y x
∂x y
∂x ∂x (2)

In order to continue with integration we will assume the quantities

between square brackets do not vary with y, which yields the following
expression:

y + ∆y x + ∆x
∂ 2T ∂T ∂T
k ( 2 ).dx.dy = [(k ) x + ∆x ∆y − (k ) x ∆y ] (3)
y x
∂x ∂x ∂x
According to Fourrier’s law of conduction, eqn (3) expresses:
(heat entering across (x) control volume face)-(heat leaving across(x+ x)
control volume face)

2
 MENG314-FV

Central-differencing formulae are now employed to evaluate the gradients

at the control volume, giving:

y + ∆y x + ∆x
∂ 2T Ti +1, j − Ti , j Ti , j − Ti −1, j
k ( 2 ).dx.dy = [(k ) ∆y − (k )∆y ]
y x
∂x ∆x ∆x (4)

where the subscript (i) refers to the value at an arbitrary node (i).The
next node index in the positive and negative directions is (i+1) and (i-1) ,
respectively.
Integrating the second term on the R.H.S. of equation(1), assuming no
variation of the partial derivative in x-direction, and following same
procedure as for the first term , yields:
y + ∆y x + ∆x
∂ 2T Ti , j +1 − Ti , j Ti , j − Ti , j −1
k ( 2 ).dx.dy = [( k ) ∆x − ( k ) ∆x ] (5)
y x
∂y ∆y ∆y

The above term expresses:

(heat entering across (y) control volume face)-(heat leaving across(y+ y)
control volume face)
Substituting the R.H.S.’s of eqns(4,5) in (1) yields:
Ti+1, j − Ti, j Ti, j − Ti−1, j Ti, j+1 − Ti, j Ti, j − Ti, j−1
[(k ) ∆y − (k )∆y] + [(k ) ∆x − (k )∆x] = 0
∆x ∆x ∆y ∆y
(6)
For example, applying eqn(6) to node(1,1) and taking x= y=h, yields:

T2,1 − T1,1 T1,1 − T0,1 T1, 2 − T1,1 T1,1 − T1,0

ka h − ka h + ka h − ka h=0
h h h h

where ka refers to conductivity of material (a) in which entire control

volume is embeded. Simplifying yields the Laplacian Difference eqn:

4T1,1 − T2 ,1 − T0 ,1 − T1, 2 − T1, 0 = 0

3
 MENG314-FV

same as the Taylor series based F.D.E.s. Equation(6) is also identical to

the Taylor’s series F.D.E.

Next let’s consider the near boundary

control-volume with discontinuities,
surrounding node (2,4). This control
volume lies partly in material (a) and
partly in material(b) hence conduct-
ion eqn with variable k must be used
; also only part of it’s top boundary is insulated.
Integration of the two terms over the C.V. sketched on the right yields:
y + ∆y x + ∆x
∂ ∂T ∂T ∂T
( k ( )).dx.dy = [(k ) x + ∆x ∆y − (k ) x ∆y ]
y x
∂x ∂x ∂x ∂x
∂T h ∂T h
= (k b ) x + ∆x . − ( k a )x.
∂x 2 ∂x 2
T − T2, 4 h T −T h (7)
= k b 3, 4 − k a 2, 4 1, 4
h/2 2 h 2

y + ∆y x + ∆ x
∂ ∂T ∂T h ∂T h
( k ( )).dx.dy = [( k a ) y + ∆y . + ( k b ) y + ∆y . ] −
y x
∂y ∂y ∂y 2 ∂y 4
∂T h ∂T h
[( k a ) y . + (k b )y. ]
∂y 2 ∂y 4
(8)
Introducing the boundary conditions at (y+ y):
∂T h
i) Insulation over left part gives : ( ka ) y + ∆y . = 0
∂y 2
ii) Convection to ambient = the conduction to surface on right part

∂T h h
(kb ) y + ∆y . = hc (Ta − T2, 4 )
∂y 4 4
4
 MENG314-FV

Substituting B.C.s into eqn(8), and substituting for derivatives at (y)

with central differencing expressions yields:

y + ∆y x + ∆x
∂ ∂T h
(k ( )).dx.dy = [0 + hc (Ta − T2 , 4 ). ]
y x
∂y ∂y 4
T2 , 4 − T2 , 3 h T − T2 ,3 h
− [k a + k b 2, 4 . ]
h 2 h 4
(9)
Substituting eqns(7,9) into eqn(1) yields the following F.V. eqn for
node(2,4):

T2 , 4 − T1, 4 h T − T2 , 4 h T − T2 , 3 h T − T2 , 3 h
− ka + k b 3, 4 − k a 2, 4 − k b 2, 4
h 2 h/2 2 h 2 h 4
h
+ hc (Ta − T2 , 4 ) = 0
4

This expresses the steady state heat balance for the volume, namely:
(left-side conduction) - (right-side conduction) + (lower conduction
material “a”) + (lower conduction material “b”) - (upper convection) = 0

Let h= 10 cm, ka=0.15 cal/s.cm.oC, kb=0.25 cal/s.cm.oC,

hc=0.05 cal/s.cm2.oC, equation becomes:
0.5875 T2,4 – 0.075 T1,4– 0.25 T3,4 – 0.1375 T2,3= 2.5

Similarly, F.V. equations are derived for all control-volumes in the

domain, the equations are then assembled and solved simultaneously, just
like the Taylor’s series F.D. method, to yield the value of the dependent
variable at each grid node.

5
 MENG314-FV

Application to a Convection-Diffusion Problem

An example of such a problem is steady state diffusion of pollutants from
a chimney stack. It is normally a three dimensional problem, however for
simplicity we will consider the two dimensional counterpart, described
by:
∂C ∂ 2C ∂ 2C
=α( + )+ S
∂x ∂x 2 ∂y 2
Where “C” is the concentration of pollutants, “S” is the source term(zero
everywhere except at the stack exit) and “ ” is a coefficient which
describes ratio of diffusion to convection rates.

Integrating over a control volume:

y + ∆y x + ∆x y + ∆y x + ∆x y + ∆y x + ∆x
∂C ∂ 2C ∂ 2C
.dx.dy = α ( 2 + 2 ).dx.dy + S .dx.dy
y x
∂x y x ∂x ∂y y x

1)Integrating L.H.S.:
y + ∆y x + ∆x y + ∆y
∂C
.dx.dy = (C x + ∆x − C x ).dy = (C x + ∆x − C x ).∆y
y x
∂x y
Cx+ x and Cx are evaluated at C.V. surface
and not grid nodes. They need to be
expressed in terms of values at grid nodes.
Two approaches are popular:
i-1 i i+1
i) Central-Differencing-Scheme: X

X+ X
Ci +1 + Ci Ci + Ci −1
C x + ∆x = , Cx =
2 2

It results in same equation as Taylor’s series F.D.; however it is unstable

for small values of .

ii) Upwind-Differencing Scheme

C x + ∆x = Ci , C x = Ci −1 wind blowing in +ve x-direction

C x+ ∆x = Ci +1 , C x = Ci wind blowing in -ve x-direction

6
 MENG314-FV

2) Integrating R.H.S.:
i) First term as before
ii) Second term: y + ∆y x + ∆x
S dxdy = S ∆ x ∆ y
y x
where S is average value of S over C.V., usually assumed equal to value
at node.

Application to Parabolic Equations

e.g. Unsteady heat conduction in a rod:

∂T ∂ 2T
= k
∂t ∂x 2 (1)
Integrating over CV in space and over time as well:
t+ ∆t x+∆x t+∆t x+∆x
∂T ∂ 2T
dx .dt = k .dx .dt (2)
t x
∂t t x
∂x 2

∂T
Assuming is uniform over control volume surrounding node “i”
∂t
(see sketch for CVs used for Parabolic eqns.) the L.H.S. of eqn(2) yields:
t + ∆t x + ∆x t + ∆t
∂T ∂T
dx.dt = .∆x.dt = (Ti t +∆t − Ti t ).∆x (3)
t x
∂t t
∂t

7
 MENG314-FV

The R.H.S. of equation(2) yields:

t + ∆t x + ∆x t + ∆t
∂ 2T ∂T ∂T
k 2 .dx.dt = [(k ) x + ∆x − (k ) x ].dt (4)
t x
∂x t
∂x ∂x

In order to integrate w.r.t. time one of three choices are possible:

i) Explicit Scheme : term is considered to be fixed at it’s value at
time (t)
ii) Implicit Scheme: term is considered to be fixed at it’s value at
time (t+ t)
iii) Crank-Nicholson: term is considered to vary linearly with time
between (t) and (t+ t), , which is equivalent to taking a fixed
value equivalent to the average of the values at (t) and (t+ t)

e.g., for an explicit scheme, eqn(4) yields:

t + ∆t
∂T ∂T ∂T ∂T
[(k ) x + ∆x − (k ) x ].dt = [(k ) x + ∆x − (k ) x ]t .∆t
t
∂x ∂x ∂x ∂x
Ti t+1 − Ti t Ti t − Ti t−1
= [k ( ) − k( )].∆t
∆x ∆x
(5)
Substituting eqn(3,4,5) into eqn(2) and dividing by t yields:

(Ti t + ∆t − Ti t ) Ti +t 1 − 2Ti t + Ti −t 1
= k( )
∆t ∆x
which is similar to the Taylor series F.D.E.s for the explicit method.

However, the finite-volume method is much more flexible and is

applicable to more complex problems.