Chapter 1

FUNDAMENTALS

1.1 Classification of physical problems

− The majority of the problems of physics and engineering fall naturally

into one of the following three physical categories:

Equilibrium Eigenvalue Propagation

Problems Problems Problems

− Equilibrium Problems are problems of steady state in which the equation

configuration Ф in a domain D is to be determined by solving the
differential equation
L (Φ ) = f
… (1)
within D, subject to certain boundary conditions
B i (Φ ) = g i
… (2)

− Representation of the general equation problem:

− Often, the integration domain D is closed and bounded.

− In mathematical terminology, such problems are known as boundary

value problems.
− Typical physical examples include steady viscous flow, steady heat
transfer problems, equilibrium stresses in elastic structures.

− The governing equations for equilibrium problems are elliptic (will be

defined shortly).

− Eigenvalue Problems may be thought of as extensions of equilibrium

problems wherein critical values of certain parameters are to be
determined in addition to the steady-state configurations.

− Mathematically, the problem is to find one or more constants ( λ ), and

the corresponding function ( Φ ) such that the differential equation
L (Φ ) = λ M (Φ )
… (3)
is satisfied within D, and the boundary conditions
B i (Φ ) = λ E i (Φ )
… (4)
hold on the boundary of D.

− Typical physical examples include: buckling and stability of structures,

resonance in electric circuits and acoustics, natural frequency problems
in vibration, laminar flame theory in combustion and so on…

− Propagation problems are initial-value problems that have an unsteady

state or transient nature.

− In these problems, one wishes to predict the subsequent behaviour of a

system given the initial state, by solving the differential equation:
L (Φ ) = f
… (5)
within the domain D, when the initial state is prescribed as
I i ( Φ ) = hi
… (6)
and subject to prescribed conditions
B i (Φ ) = g i
… (7)
on the (open) boundaries.
− To illustrate the general propagation problem:

− In mathematical parlance, such problems are known as initial boundary

value problems.

− Typical physical examples include the propagation of pressure waves in

a fluid, propagation of stresses and displacements in elastic systems, etc.

− Partial Differential Equations governing propagation problems are

parabolic or hyperbolic (will be defined next section).

− Propagation problems are described as marching problems, where the

solution marches out from the initial state guided and modified in transit
by the side boundary conditions.
1.2 Classification of Equations

− For the following general system of 2nd order partial differential

equations:
n
∂ 2u n
∂u
∑
i =1
A i
∂x i 2
+ ∑
i =1
Bi
∂x i
+ Cu + D = 0

… (8)

− If all the Ai's are non-zero and have the same sign, the PDE is elliptic.

− If all the Ai's are non-zero and have, with one exception, the same sign,
the PDE is hyperbolic.

− If one Ai is zero (Ak for instance) and the remaining Ai's are non-zero
and of the same sign, and if the coefficient Bk is non-zero, the PDE is
parabolic.

1.3 Discrete Methods

− The ultimate goal of discrete methods is the reduction of continuous

systems to "equivalent" discrete (lumped parameter) systems which are
suitable for high-speed computer solution.

− The basic approximation involves the replacement of a continuous

domain D by a pattern, network, or mesh of discrete points within D.
 − Discretization of the governing equations and boundary conditions of the
continuous problem may be accomplished physically, but is more often
carried out mathematically.

− In the physical approach, the continuous medium is replaced by discrete

media. For example, a heat conducting slab could be replaced by a
network of heat-conducting rods. The governing equations are then
developed by direct application of the physical laws to the discrete
system.

− On the other hand, in the mathematical approach, the continuous

formulation is transformed to a discrete formulation by replacing
derivatives by, say, finite difference approximations.

− It is of interest to note that the discrete approximations can proceed by

several avenues; the one that will be followed in this part is the finite
difference method.

− It is to be noted that the discretization approximation is affected in many

engineering books by what is known as the control volume approach.

1.4 Finite Differences and Finite Difference Operators

− All approximations of partial derivatives by finite differences introduce

errors called truncation errors.

− The finite difference approximation for the partial derivative can be

obtained using Taylor series.

− Taylor series for u (x + Δx , y ) about (x , y ) gives

u (x + Δx , y ) =
∂u (Δx ) 2 ∂ 2u (Δx )3 ∂ 3u
( x , y ) + O ⎡ ( Δx ) ⎤
4
u (x , y ) + Δx (x , y ) + ( x , y ) +
∂x 2! ∂x 2
3! ∂x 3 ⎣ ⎦
… (9)

− Dividing by Δx , and rearranging, we get

∂u
(x , y ) = [u (x + Δx , y ) − u (x , y ) ] Δx + O (Δx )
∂x
… (10)

− The forward difference provides the simple 1st order approximation

∂u
≈ [u (x + Δx , y ) − u (x , y ) ] Δx
∂x
… (11)

− Similarly, one can deduce the backward difference by expanding

u (x + Δx , y ) about (x , y ) using Taylor series to obtain at the end
 ∂u
≈ [u (x , y ) − u (x − Δx , y ) ] Δx
∂x
… (12)

∂u
− We can also deduce 2nd order finite difference approximations for .
∂x

− Finite difference approximations for 2nd partial derivatives can be

obtained by adding Taylor series expansion of u (x + Δx , y ) and
u (x − Δx , y ) about (x , y ) to obtain at the end, the central difference as
∂ 2u u (x + Δx , y ) − 2u (x , y ) + u (x − Δx , y )
+ O ⎡ ( Δx ) ⎤ .
2
=
∂x 2
Δx 2 ⎣ ⎦
… (13)

− In index notation
∂ 2u u − 2u i , j + u i −1, j
= i +1, j + O (h 2 ) .
∂x i , j
2
h 2

… (14)

•
i,j+1

• • •
i-1,j i,j i+1,j

•
i,j-1

− One can define difference operators as follows:

Forward Difference Δu n = u n +1 − u n
Backward Difference ∇u n = u n − u n −1
1.5 Errors

− The error in the solution, due to the replacement of the continuous

problem by the discrete model is called the discretization error.

− When the discrete equations are not solved exactly an additional error is
introduced, which is called round-off error.

− The interval size (h) affects the above errors in the opposite sense, i.e.
decreasing h decreases the discretization error but increases the round-
off error.

1.6 Stability and Convergence

− Any numerical scheme which allows the growth of error, eventually

"swamping" the true solution, is unstable.

− A finite difference scheme converges if the finite difference solution U

converges to the solution u, with the same boundary values, as h→0.