Finite Volume Method

1 Introduction
An alternative discretization method is based on the idea of regarding the computation
domain as subdivided into a collection of finite volumes. In this view, each finite volume is
represented by a line segment in 1D, an area in 2D and a volume in 3D. A node, located
inside each finite volume is the locus of computational values although values of selected
field variables at finite volume boundaries play crucial roles.
In rectangular Cartesian coordinates in 2D the simplest finite volumes are rectangles. For
instance, for each node, the rectangle faces are formed by drawing perpendiculars through the
midpoints between contiguous nodes. Discretization equations are obtained by integrating
the original partial differential equation over the span of each finite volume. The method is
easily extended to nonlinear problems.

2 The Finite Volume Method in 1D

Consider for instance the following problem consisting of determining the function φ(x) in
x ∈ [0, L], such that

d2 φ
+ f(x) = 0
dx2
for x ∈ [0, 1], where f(x) is a known function and φ is subject to specified conditions at the
boundary such as

φ(0) = 0
φ(1) = 0

The problem may represent steady state heat conduction in a medium of unit thermal
conductivity where energy is being internally generated at the rate f, per unit volume and
whose boundaries are maintained at zero temperature.
In the finite volume method, to obtain the discrete analogue of the problem, the space
domain is first divided into a collection of contiguous finite volumes of size ∆xi along the

1
x−direction and unit dimensions along the y− and z− directions. For uniformly sized
volumes there is only one spacing ∆x. Any given volume in the set is characterized by the
nodal location of its center point P and that of its bounding faces e and w to the East and
West of P , respectively. Note, that in this arrangement, face e is located half-way between
node P and node E pertaining to the finite volume located to its east. Moreover, face w is
located half-way between node P and node W pertaining to the finite volume located to its
west
Then, the governing equation is then integrated over the span of any given arbitrary
volume in the set to yield
dφ dφ ¯
|e − |w = f(x)∆x
dx dx
¯
where f(x) is the average value of f(x) inside the finite volume. Note that the derivatives
on the left hand side are evaluated at the finite volume faces which are also the mid-point
locations between contiguous finite volumes.
The discretized equation is now obtained by approximating the derivatives on the left
hand side with central differences, i.e.
φE − φP φP − φW
− = f¯(x)∆x
∆x ∆x
Upon rearrangement this gives
¯
−aW φW + aP φP − aE φE = f(x)∆x
1
where aW = aE = ∆x , and aP = aW + aE .
¯
Note that if f (x) = f(xP ) (i.e. the average value of the source term is replaced by the
value of f at the nodal location), the above reduces to the simple form
1
φP = [φW + φE + f(xP )(∆x)2]
2
which is identical to the expression obtained before using the method of finite differences.

3 Boundary Conditions
As in the case of finite differences, care is required to properly deal with the boundary
conditions at the boundary nodes, specially when derivative conditions are involved.

4 Solution of the Algebraic Equations

As was the case with the finite difference method, the resulting system of coupled linear
algebraic equations for all nodes in the mesh may sometimes be solved by a direct method
(Gauss elimination) but more commonly in CFD, an iterative method is used, most frequently
a combination of TDMA with an iterative relaxation-type method.

2
5 The Convection-Diffusion Equation and Upwinding
Consider the following one-dimensional convection-diffusion equation for the transported
quantity φ(x)

dφ d2 φ
ρu =Γ 2
dx dx
for x ∈ [0, L] subject to suitable boundary conditions such as

φ(0) = 1
φ(L) = 1

where ρ is the density of the fluid, u > 0 is the velocity and Γ the diffusivity of φ in the
fluid, and all are assumed constant.
Using the finite volume method on an uniform mesh (spacing ∆x), and integrating once
over the span of the volume centered at xP with faces e and w, as described above yields
dφ dφ
(ρuφ)|e − (ρuφ)|w = (Γ )|e − (Γ )|w
dx dx
Evaluating φe and φw by linear interpolation and using central differencing for the derivatives
on the right hand side yields
1 1 φE − φP φE − φP
(ρu)(φE + φP ) − (ρu)(φP + φW ) = Γ( ) − Γ( )
2 2 ∆x ∆x
Rearranging gives

−aE φE + aP φP − aW φW = 0

where
Γ 1
aE = − (ρu)
∆x 2
Γ 1
aW = + (ρu)
∆x 2
and

aP = aE + aW

As with the FD method, the above scheme works fine as long as the Peclet number of
the system, defined as
ρuL
Pe =
Γ

3
and representing the ratio of convective to diffusive transports, is small (say < 1). Upwinding
must be used in general to avoid non-physical oscillations. In the case of the finite volume
method, upwinding is implemented as follows:
If ue > 0, then φe = φP , and if ue < 0, then φe = φE . Moreover, if uw > 0, then φw = φW ,
and if uw < 0, then φw = φP . As with the FDM, this scheme is only first order accurate
and it has the effect of introducing a non-physical diffusive effect in the solution called false
diffusion. Discretization schemes of higher order of accuracy have been developed based on
the notion of upwinding (e.g. the exponential scheme).