Chapter 3

Truncation Error Analysis

Taylor Tables

T. H. Pulliam
NASA Ames

1
 Taylor Tables of Differencing Schemes

• Notation: Consider u(x, t) for fixed t and x = j∆x so that,

u(x + k∆x) = u(j∆x + k∆x) = uj+k .
• The generalized form of the Taylor Series Expansions is given by
   2   n 
∂u 1 ∂ u 1 n ∂ u
uj+k = uj + (k∆x) + (k∆x)2 + . . . + (k∆x)
∂x j 2 ∂x2 j n! ∂xn j
+ ...

• For example, consider the Taylor series expansion for uj+1 :

   2   n 
∂u 1 ∂ u 1 n ∂ u
uj+1 = uj + (∆x) + (∆x)2 + . . . + (∆x)
∂x j 2 ∂x2 j n! ∂xn j
+ ...

• Or for uj−2 :
   2   n 
∂u 1 ∂ u 1 ∂ u
uj−2 = uj + (−2∆x) + (−2∆x)2 + ... + (−2∆x)n
∂x j 2 ∂x2 j n! ∂xn j
+ ...

2
 Finite Difference Formulas

• Take the expansion for uj−1

   2 
∂u 1 ∂ u
uj−1 = uj − ∆x + (∆x)2 + ...
∂x j 2 ∂x2 j

• Rearrange terms to  
∂u (uj − uj−1 )
− = ert
∂x
j ∆x
 2 
• Truncation error term is ert = + 2 ∆x ∂∂xu2
1
j

• The truncation error ert is made up of 4 important pieces

ert = Sign Coefficient ∆xp (p + q)th Derivative

3
 Taylor Table For the 1st Order Backward Difference

• Given  
∂u (uj − uj−1 )
− = ert
∂x j ∆x
• Each term expanded in Taylor Series and placed in a table simplifing algebra.
• Note the multiplication by ∆x to again simplify the table.
∆x·
∆x2

· ∆x3
· ∆x4
·
∂u uj∂2 u ∂3 u ∂4 u
∂x j ∂x2 j ∂x3 j ∂x4 j

∂u


∆x · 1
∂x j
uj−1 1 (-1) 1 (-1)2 1 (-1)3 1 (-1)4 1
1! 2! 3! 4!
− uj −1 0 0 0 0

∆x · ert 0 0 1 ? ?
2

 
1 ∂2u
• Truncation error term ert = 2
∆x ∂x2
defined from the first non-zero column.
j

• Don’t forget the division by the ∆x to undo the previous multiplication.

• Order of accuracy is defined as the exponent on the ∆x term in ert .

4
 Taylor Table For the 2nd Order Central Difference

• Given  
∂u (uj+1 − uj−1 )
− = ert
∂x j 2∆x
• The Taylor Table
∆x· ∆x2 · ∆x3 · ∆x4 ·
∂2 u ∂3 u ∂4 u
∂u





uj
∂x j ∂x2 j ∂x3 j ∂x4 j

∂u


2∆x · 2
∂x j
uj−1 1 (-1) 1 (-1)2 1 (-1)3 1 (-1)4 1
1! 2! 3! 4!
− uj+1 −1 −(1) 1 −(1)2 1 −(1)3 1 −(1)4 1
1! 2! 3! 4!
=

2∆x · ert 0 0 0 −1 ?
3

 
∂3u
• The truncation error term ert = − 16 ∆x2 ∂x3 is defined from the first
j
non-zero column.
• Accuracy is 2nd Order.

5
 Taylor Table For A General 3 Point Difference Scheme

• Starting with
 
∂u 1
− (c uj−2 + b uj−1 + a uj ) = ert
∂x j ∆x

• The Taylor Table

2 3 4
∆x·

∆x
2
· ∆x
3
· ∆x
4
·
∂u ∂ u ∂ u ∂ u
uj ∂x j ∂x2 ∂x3 ∂x4
j j j
∂u


∆x · ∂x j
1

−cuj−2 −c −c · (-2) · 1
1
−c · (-2)2 · 1
2
−c · (-2)3 · 1
6
−c · (-2)4 · 1
24
−buj−1 −b −b · (-1) · 1
1
−b · (-1)2 · 1
2
−b · (-1)3 · 1
6
−b · (-1)4 · 1
24

−auj −a
=
∆x · ert ? ? ? ? ?
• Now instead of having colums sum to zero, we set enough colums to zero to
satisfy the number of unknowns.

6
 Taylor Table For A General 3 Point Difference Scheme

• This time the first three columns sum to zero if

−1 −1 −1 c 0
    
 2 1 0  b =  − 1
− 4 −1 0 a 0
• Note we put the linear equations into a matrix form, let Matlab do the work
for you.
• Which gives [c, b, a] = 12 [1, −4, 3].
• In this case the fourth column provides the leading truncation
   3  2
 3 
1 8c b ∂ u ∆x ∂ u
ert = + ∆x3 =
∆x 6 6 ∂x3 j 3 ∂x3 j

• Thus we have derived a second-order backward-difference approximation of a

first derivative:
 
∂u 1
= (uj−2 − 4uj−1 + 3uj ) + O(∆x2 )
∂x j 2∆x

7
 Taylor Table For Other Derivatives, e.g. 2nd

• Consider a gerneral 3 point formula for the 2nd derivative

 2 
∂ u 1
− (a uj−1 + b uj + c uj+1 ) = ert
∂x2 j ∆x2

• The Taylor Table is

∆x· ∆x2 · ∆x3 · ∆x4 ·
∂2 u ∂3 u ∂4 u
∂u





uj
∂x j ∂x2 j ∂x3 j ∂x4 j

∂2 u


∆x2 · 1
∂x2 j
−auj−1 −a −a · (-1) · 1 −a · (-1)2 · 1 −a · (-1)3 · 1 −a · (-1)4 · 1
1 2 6 24
−buj −b
−cuj+1 −c −c · (1) · 1 −c · (1)2 · 1 −c · (1)3 · 1 −c · (1)4 · 1
1 2 6 24
=
2
∆x · ert ? ? ? ? ?

• Setting the first 3 colums to 0 leads to

h −1 −1 −1 i h a i h 0
i
1 0 −1 b = 0
−1 0 −1 c −2

• The solution is given by [a, b, c] = [1, −2, 1].

8
 Taylor Table For 2nd Derivative

• In this case ert occurs at the fifth column in the table (for this example all
even columns will vanish by symmetry) and one finds
   4  2
 4 
1 −a −c 4 ∂ u −∆x ∂ u
ert = + ∆x =
∆x2 24 24 ∂x4 j 12 ∂x4 j

• Note that ∆x2 has been divided through to make the error term consistent.
• We have just derived the familiar 3-point central-differencing point operator for
a second derivative
 2 
∂ u 1
2
= 2
(uj−1 − 2uj + uj+1 ) + O(∆x2 )
∂x j ∆x