Another Approach to Two-Point Boundary Value Problems
Focus: Linear problems with Dirichlet boundary conditions.
𝑢′′(𝑥) = 𝑝(𝑥)𝑢′(𝑥) + 𝑞(𝑥)𝑢(𝑥) + 𝑓(𝑥), 𝑎≤𝑥≤𝑏
{ }
𝑢(𝑎) = 𝛼, 𝑢(𝑏) = 𝛽
𝑝(𝑥), 𝑞(𝑥), 𝑓(𝑥), 𝑎, 𝑏, 𝛼 and 𝛽 are given, and 𝑝(𝑥), 𝑞(𝑥) and 𝑓(𝑥) are continuous functions on [𝑎, 𝑏].
Goal: Approximate the unknown function 𝑢(𝑥).
Review: We previously discussed this problem in Chapter 3, and used three different initial value
problems to approximate the solution.
A different approach:
𝑢′′(𝑥) = 𝑝(𝑥)𝑢′(𝑥) + 𝑞(𝑥)𝑢(𝑥) + 𝑓(𝑥), 𝑎 ≤ 𝑥 ≤ 𝑏
{ } (3)
𝑢(𝑎) = 𝛼, 𝑢(𝑏) = 𝛽
The approach here is very different, but serves as a good lead-in to approximation of solutions to more
complicated problems associated with partial differential equations (see Math 4365 for more
information), where shooting methods cannot be used.
Recall: If ℎ > 0 then the following centered difference approximations are 𝑂(ℎ2 ).
𝑢(𝑥 + ℎ) − 𝑢(𝑥 − ℎ)
𝑢′ (𝑥) ≈
2ℎ
𝑢(𝑥 + ℎ) − 2𝑢(𝑥) + 𝑢(𝑥 − ℎ)
𝑢′′ (𝑥) ≈
ℎ2
Let’s use these in our present system.
𝑏−𝑎
Step 1: Choose a relatively large positive integer 𝑛, creating a step size ℎ = 𝑛
, and creating the points
𝑥0 = 𝑎, 𝑥1 = 𝑎 + ℎ, 𝑎2 = 𝑎 + 2ℎ, … , 𝑥𝑛−1 = 𝑎 + (𝑛 − 1)ℎ, 𝑥𝑛 = 𝑏.
These are the points where we hope to approximate our solution. Now, define our approximations to
𝑢(𝑥𝑖 ) to be denoted by 𝑢𝑖 for each 𝑖. Then the centered difference approximations above become
𝑢(𝑥𝑖 + ℎ) − 𝑢(𝑥𝑖 − ℎ) 𝑢𝑖+1 − 𝑢𝑖−1
𝑢′ (𝑥𝑖 ) ≈ ≈
2ℎ 2ℎ
and
(𝑢(𝑥𝑖 + ℎ) − 2𝑢(𝑥𝑖 ) + 𝑢(𝑥𝑖 − ℎ)) 𝑢𝑖+1 − 2𝑢𝑖 + 𝑢𝑖−1
𝑢′′ (𝑥𝑖 ) ≈ ≈ .
ℎ2 ℎ2
Step 2: Substitute the approximations above into
𝑢′′(𝑥) = 𝑝(𝑥)𝑢′(𝑥) + 𝑞(𝑥)𝑢(𝑥) + 𝑓(𝑥)
for each 𝑥 = 𝑥𝑖 , with 𝑖 = 1, … , 𝑛 − 1. Start by viewing the equations for each 𝑖 = 1, … , 𝑛 − 1.
𝑢′′(𝑥1 ) = 𝑝(𝑥1 )𝑢′(𝑥1 ) + 𝑞(𝑥1 )𝑢(𝑥1 ) + 𝑓(𝑥1 )
𝑢′′(𝑥2 ) = 𝑝(𝑥2 )𝑢′(𝑥2 ) + 𝑞(𝑥2 )𝑢(𝑥2 ) + 𝑓(𝑥2 )
( )
⋮
𝑢′′(𝑥𝑛−1 ) = 𝑝(𝑥𝑛−1 )𝑢′(𝑥𝑛−1 ) + 𝑞(𝑥𝑛−1 )𝑢(𝑥𝑛−1 ) + 𝑓(𝑥𝑛−1 )
Now, make the substitutions.
𝑢2 − 2𝑢1 + 𝑢0 (𝑢2 − 𝑢0 )
2
= 𝑝1 + 𝑞1 𝑢1 + 𝑓1
ℎ 2ℎ
𝑢3 − 2𝑢2 + 𝑢1 (𝑢3 − 𝑢1 )
2
= 𝑝2 + 𝑞2 𝑢2 + 𝑓2
ℎ 2ℎ
⋮
𝑢𝑛 − 2𝑢𝑛−1 + 𝑢𝑛−2 (𝑢𝑛 − 𝑢𝑛−2 )
( = 𝑝𝑛−1 + 𝑞𝑛−1 𝑢𝑛−1 + 𝑓𝑛−1 )
ℎ2 2ℎ
Notes:
• 𝑢0 is our approximation for 𝑢(𝑥0 ) = 𝑢(𝑎), and we are given 𝑢(𝑎) = 𝛼.
So, we substitute 𝑢0 = 𝛼 above.
• 𝑢𝑛 is our approximation for 𝑢(𝑥𝑛 ) = 𝑢(𝑏), and we are given 𝑢(𝑎) = 𝛽.
So, we substitute 𝑢𝑛 = 𝛽 above.
Then we isolate the 𝑢𝑖 on the left side of each equation, and move everything else to the right side.
This should give
1 1
(−2 − ℎ2 𝑞1 )𝑢1 + (1 − ℎ𝑝1 ) 𝑢2 = ℎ2 𝑓1 − 𝛼(1 + ℎ𝑝1 )
2 2
1 1
(1 + 2 ℎ𝑝2 ) 𝑢1 + (−2 − ℎ2 𝑞2 )𝑢2 + (1 − 2 ℎ𝑝2 ) 𝑢3 = ℎ2 𝑓2
⋮ . (11)
1 2 1 2
(1 + 2 ℎ𝑝𝑛−2 ) 𝑢𝑛−3 + (−2 − ℎ 𝑞𝑛−2 )𝑢𝑛−2 + (1 − 2 ℎ𝑝𝑛−2 ) 𝑢𝑛−1 = ℎ 𝑓𝑛−2
1 2 2 1
( (1 + 2 ℎ𝑝𝑛−1 ) 𝑢𝑛−2 + (−2 − ℎ 𝑞𝑛−1 )𝑢𝑛−1 = ℎ 𝑓𝑛−1 − 𝛽 (1 − 2 ℎ𝑝𝑛−1 ) )
This system can be written in the form
⃗ =𝐹
𝑀𝑈
where
𝑢1 𝑢(𝑥1 )
𝑢2 𝑢(𝑥 2 ) ).
⃗ =(
𝑈 ⋮ )≈( ⋮
𝑢𝑛−1 𝑢(𝑥𝑛−1 )
𝑀 =? 𝐹 =?
Terms: Sparse, tri-diagonal, sub-diagonal, diagonal, super-diagonal
Remarks on Solving
⃗ = 𝐹.
𝑀𝑈
Thomas Algorithm: Given the system 𝑀𝑈 ⃗ = 𝐹 where 𝑀 is an 𝑚 × 𝑚 tridiagonal matrix such that the
sub-diagonal, diagonal and super-diagonal of 𝑀 are the vectors 𝑐, 𝑑 and 𝑒, and 𝐹 ∈ ℝ𝑘 .
Let 𝑘 be the number of entries in the vector 𝑑.
Let 𝑐𝑐, 𝑑𝑑, 𝑒𝑒 and 𝐹𝐹 be arrays containing the entries in 𝑐 , 𝑑, 𝑒 and 𝐹 , respectively.
Let 𝑢 = 𝑑𝑑
For each 𝑖 = 2 to 𝑘
𝑐𝑐[𝑖−1]
Let 𝑚 = 𝑑𝑑[𝑖−1]
Let 𝑑𝑑[𝑖] = 𝑑𝑑[𝑖] − 𝑚 ⋅ 𝑒𝑒[𝑖 − 1]
Let 𝐹𝐹[𝑖] = 𝐹𝐹] − 𝑚 ⋅ 𝐹𝐹[𝑖 − 1]
𝐹𝐹[𝑘]
Let 𝑢[𝑘] =
𝑑𝑑[𝑘]
For each 𝑖 = 𝑘 − 1 by −1 to 1
𝐹𝐹[𝑖]−𝑒𝑒[𝑖]⋅𝑢[𝑖+1]
Let 𝑢[𝑖] = 𝑑𝑑[𝑖]
Return 𝑢
Note: VBA and Python code is posted on Space.
Theorem 1: Suppose 𝑝(𝑥), 𝑞(𝑥) and 𝑓(𝑥) are continuous functions, and 𝛼, 𝛽 ∈ ℝ. If (3) has a unique
𝑏−𝑎
solution, and ℎ = 𝑛 is sufficiently small (i.e. 𝑛 is sufficiently large), then (11) will have a unique
𝑢1
𝑢2
⃗ =(
solution 𝑈 ⋮ ), and there is a value 𝐶 dependent only on 𝑝(𝑥), 𝑞(𝑥), 𝑓(𝑥), 𝑎, 𝑏, 𝛼 and 𝛽 so that
𝑢𝑛−1
|𝑢𝑖 − 𝑢(𝑥𝑖 )| ≤ 𝐶ℎ2 for all 𝑖 = 1, … , 𝑛.
Example 1: Show that the two-point boundary value problem below has a unique solution.
𝑢′′ (𝑥) = −1 − cos(𝑥) − 2𝑥 𝑢(𝑥), 0 ≤ 𝑥 ≤ 1
{ }
𝑢(0) = 1, 𝑢(1) = 2
Then approximate this solution in two ways. First, use the method described in Chapter 3. Then use the
approximation method guaranteed in Theorem 1 using 𝑛 = 100. Finally, compare the two
approximations.
