INTRODUCTION TO

FINITE DIFFERENCE
METHOD
Two Methods in Solving
Differential Equations
■ Analytical
– Only few equations can be solved
– Too hard
■ Numerical
– Can be done to vast majority of differential equations
– Will have some error
Numerical Methods

■ Finite Difference Method

– replacing the derivatives in the equation by appropriate numerical
differentiation formula
■ Finite Element Method
– To be discussed
Finite Difference
Approximation
Given u = u(x),
𝑑𝑢 𝑢 𝑥 + ℎ − 𝑢(𝑥)
= 𝑢′ (𝑥) = lim
𝑑𝑥 ℎ→0 ℎ

𝑢 𝑥 + ℎ − 𝑢(𝑥)
𝑢′ 𝑥 ≈
ℎ
Finite Difference
Approximation

′
𝑢 𝑥 + ℎ − 𝑢(𝑥)
𝑢 𝑥 ≈
ℎ
Recall Taylor Series

∞ 𝑛
𝑓 𝑎
𝑓 𝑥 =෍ (𝑥 − 𝑎)𝑛
𝑛!
𝑛=0

∞ 𝑛
𝑢 𝑥
𝑢 𝑥+ℎ = ෍ (ℎ)𝑛
𝑛!
𝑛=0
Taylor Series

■ First Order Taylor Expansion

𝑢 𝑥 + ℎ ≈ 𝑢 𝑥 + 𝑢′ 𝑥 ℎ

′
𝑢 𝑥 + ℎ − 𝑢(𝑥)
𝑢 𝑥 ≈
ℎ

′
𝑢 𝑥+ℎ −𝑢 𝑥
𝑢 𝑥 = + 𝑂(ℎ)
ℎ
Example

■ Given that u(x) = sin x. Approximate u’(1).

Taylor Series for Second Order
Differential Equation
ℎ2 ℎ 3
𝑢 𝑥 + ℎ = 𝑢 𝑥 + 𝑢′ 𝑥 ℎ + 𝑢′′ 𝑥 + 𝑢′′′ 𝑥 + 𝑂 ℎ4
2 6

2 3
ℎ ℎ
𝑢 𝑥 − ℎ = 𝑢 𝑥 − 𝑢′ 𝑥 ℎ + 𝑢′′ 𝑥 − 𝑢′′′ 𝑥 + 𝑂 ℎ4
2 6

𝑢 𝑥 + ℎ + 𝑢 𝑥 − ℎ = 2𝑢 𝑥 + 𝑢′′ 𝑥 ℎ2 + 𝑂 ℎ4

𝑢 𝑥 + ℎ − 2𝑢 𝑥 + 𝑢(𝑥 − ℎ)
𝑢′′ 𝑥 = + 𝑂(ℎ 2)
ℎ2
Example

2
■ Given 𝑢 𝑥 = 𝑒 𝑥 . Approximate u’’(1).

■ Try h = 0.00001?
– Why is the error higher?
Improving u’(x) Approximation

ℎ2
𝑢 𝑥 + ℎ = 𝑢 𝑥 + 𝑢′ 𝑥 ℎ + 𝑢′′ 𝑥 + 𝑂 ℎ3
2

ℎ 2
𝑢 𝑥 − ℎ = 𝑢 𝑥 − 𝑢′ 𝑥 ℎ + 𝑢′′ 𝑥 + 𝑂 ℎ3
2

𝑢 𝑥 + ℎ − 𝑢 𝑥 − ℎ = 2𝑢′ 𝑥 ℎ + 𝑂 ℎ3

𝑢 𝑥 + ℎ − 𝑢(𝑥 − ℎ)
𝑢′ 𝑥 = + 𝑂(ℎ2 )
2ℎ
Example

■ Given that u(x) = sin x. Approximate u’(1).

Machine Exercise
■ Derive some basic one-sided finite difference formulas, which are used for
approximating derivatives of functions at or near the boundary of their domain.
– Construct a finite difference formula that approximates the derivative u’(x)
using the values of u(x) at the points x, x + h, and x + 2h.
 What is the order of your formula?
– Find a finite difference formula for u’’(x) that involves the same three function
values.
 What is its order?
– Test your formulas by computing approximations to the first and second
derivatives of u(x) = sin x at x = 1 using step sizes h = 0.1, 0.01, and 0.001.
 What is the error in your numerical approximations? Are the errors compatible with
the theoretical orders of the finite difference formulas? Discuss why or why not.
Machine Exercise Solution

𝑢′′(𝑥) 2
′
𝑢 𝑥+ℎ =𝑢 𝑥 +𝑢 𝑥 ℎ+ ℎ + 𝑂 ℎ3
2

𝑢′′ 𝑥
𝑢 𝑥 + 2ℎ = 𝑢 𝑥 + 𝑢′ 𝑥 2ℎ + 4ℎ2 + 𝑂 ℎ3
2
Machine Exercise Solution

−4𝑢 𝑥 + ℎ + 𝑢 𝑥 + 2ℎ = −3𝑢 𝑥 − 2𝑢′ 𝑥 ℎ + 𝑂 ℎ3

′
−3𝑢 𝑥 + 4𝑢 𝑥 + ℎ − 𝑢 𝑥 + 2ℎ
𝑢 𝑥 = + 𝑂 ℎ2
2ℎ
Machine Exercise Solution

−2𝑢 𝑥 + ℎ + 𝑢 𝑥 + 2ℎ = −𝑢 𝑥 + 𝑢′′ 𝑥 ℎ2 + 𝑂 ℎ3

′
𝑢 𝑥 − 2𝑢 𝑥 + ℎ + 𝑢 𝑥 + 2ℎ
𝑢 𝑥 = +𝑂 ℎ
ℎ2