Chapter 4

Truncation Error and

the Taylor Series

Applied Computational Methods

1
 Background

 Truncation error are those that results from

using an approximation in place of an exact
mathematical procedure.
 A truncation error is introduced into
numerical solution because difference
equation only approximates the true value of
the derivative.
 In order to gain insight into the properties of
truncation error, the Taylor function is used.
2
 Taylor’s Theorem
 If the function f and its first n+1 derivatives are continuous
on an interval containing a and x, then the value of the
function at x is given by
f (a)

f ( x)  f (a)  f (a)( x  a)  ( x  a) 2
2!
( 3)
f (a)
 ( x  a)3  ...
3!
f ( n ) (a)
 ( x  a) n  Rn
n!
( x  t ) n ( n 1)
x

 where remainder Rn is defined as Rn   f (t )dt

a
n!
3
 Taylor’s Theorem
 It is often convenient to simplify the Tailor series by
defining a step size h = xi+1 - xi
f ( xi ) 2 f (3) ( xi ) 3 f ( n ) ( xi ) n
f ( xi 1 )  f ( xi )  f ( xi )h  h  h  ...  h  Rn
2! 3! n!
 where remainder Rn is defined as

f ( n1) ( ) n1
Rn  h
(n  1)!

  is a value of x that lies somewhere between xi and xi+1

 This value will be discussed later

4
Ex. Taylor Approximation of a Polynomial

 Use zero-through fourth-order Taylor series

expansion to approximate the function

f ( x)  0.1x 0.15x 0.5x 0.25x  1.2

4 3 2

 From xi = 0 with h = 1. That is, predict the

function’s value at xi+1 = 1

5
 Ex. 4.1 Taylor Approximation of a Polynomial

 Solution f ( x)  0.1x 4 0.15x3 0.5x 2 0.25x  1.2

 For x = 0 then f(0) = 1.2
 For x = 1 then f(1) = 0.2 this is the true that we are trying
to predict.
 Taylor series approximation with n = 0 f ( xi 1 )  f ( xi )  1.2
 Truncation error: Et = true value – approximation
Et = 0.2 -1.2 = -1.0 at x = 1
 n = 1, the first derivative
 f’(0) = -0.4(0)3- 0.45(0)2 - 1.0(0) - 0.25 = -0.25
 Taylor series approximation with n = 1
f ( xi 1 )  f ( xi )  1.2  0.25h  0.95 if h  1 6
 Ex. 4.1 Taylor Approximation of a Polynomial

 Taylor series approximation with n = 1

 Truncation error: Et = true value – approximation
Et = 0.2 - 0.95 = - 0.75 at x = 1
 n = 2, the second derivative
 f’’(0) = -1.2(0)2- 0.9(0) - 1.0 = -1.0
 Taylor series approximation with n = 2

f ( xi 1 )  1.2  0.25h  0.5h2  0.45 if h  1

 Truncation error: Et = true value – approximation
Et = 0.2 -0.45 = -0.25 at x = 1

7
Taylor series expansion

8
 Ex. 4.2

 Use Taylor series expansion with n = 0 to 6

to approximate f(x) = cos x, at xi+1 = /3 on
the bases of the value of f(x) and its
derivatives at xi = /4.
 Note this means that h = /3 - /4 = /12.

9
 Ex.4.1x
 Find the truncation error in approximating the function
y2 ( x)  ln(1  x)
 (a) y1 ( x)  x
1 2
 (b) y2 ( x)  x  x
2
1 2 1 3
 (c) y3 ( x)  x  x  x
2 3
1 2 1 3 1 4
 (d) y3 ( x)  x  x  x  x
2 3 4
 over the range 0  x  1

10
 Ex.4.1x (Solution)

 We consider a representative value of x, x = 0.5, and find the

truncation error. The exact value of the function is given by
y2 ( x)  ln(1  x)  ln(1.5)  0.405465108
 (a) y1 (0.5)  0.5

Te  0.405465108  0.5  0.0094544892

1
 (b) y2 (0.5)  (0.5)  (0.5) 2  0.375
2
Te  0.405465108  0.375  0.030465108

11
 Ex.4.1x (Solution)

 We consider a representative value of x, x = 0.5, and find the

truncation error. The exact value of the function is given by
y2 ( x)  ln(1  x)  ln(1.5)  0.405465108
1 1
 (c) y3 (0.5)  (0.5)  (0.5)  (0.5)3  0.416666667
2

2 3
Te  0.405465108  0.416666667  0.011201559

1 1 1
 (d) y4 (0.5)  (0.5)  (0.5)  (0.5)  (0.5) 4  0.401041667
2 3

2 3 4
Te  0.405465108  0.401041667  0.004423441

12
 Truncation error

0.8

y(x)
0.6
Te1
Te2
0.4 Te3
Te4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2

-0.2

-0.4

13
 Error Propagation
 The purpose of this section is to study how error in numbers
can propagate through mathematical functions.
 Assuming ~ x is an approximation of x, we would like to assess
the effect of the discrepancy between x and ~ x on the value
of the function.
 We estimate f ( ~
x )  f ( x)  f ( ~
x)

 We can use Taylor series:

f ( ~
x)
~ ~ ~
f ( x)  f ( x )  f ( x )( x  x )  (x  ~
x )  ...
2
 Dropping the second and high order terms and rearranging
yields
f ( x)  f ( ~
x )  f (~ x )  f ( ~
x )( x  ~
x)
14
 Error Propagation
 Dropping the second and high order terms and rearranging
yields
f ( x)  f ( ~
x )  f (~
x )  f ( ~
x )( x  ~
x)
 Or it can be rewritten as:

f ( ~
x )  f ( ~
x ) (x  ~
x)

 where
 f ( ~
x )  f ( ~
x ) (x  ~
x ) an estimate of the error of the function

 ~
x  x~
x an estimate of the error of x

15
Graphical depiction of the first-order error propagation

16
Error Propagation in a function of a single variable

 Given a value of ~x = 2.5 with an error of ~x = 0.01,

estimate resulting error in the function f(x) = x3
 Solution
f ( ~
x )  f ( ~
x ) (x  ~
x)

f ( ~
x )  3(2.5) 2 (0.01)  0.1875

f (2.5)  15.625


~
We predict for x = 2.5
f (2.5)  15.625  0.1875
17
 Function of more than one variable

 For n independent variable ~ x1 , ~

x2 ,..., ~
xn having
error ~
x1 , ~
x2 ,..., ~
xn the following general relationship holds

~ ~ ~ f ~ f ~ f ~
f ( x1 , x2 ,..., xn )  x1  x2  ...  xn
x1 x2 xn

18
 Total Numerical Error

 Total numerical error is the summation of the

truncation and round-off errors.
 To minimize round-off error is to increase the
number of significant figures of the computer.
 The truncation error can be reduce by decreasing
the step size.
 However the truncation errors decrease as the
round-off errors increase.
 In actual cases, such situation relatively uncommon
because most computer carry enough significant
figures that round-off errors do not dominate.

19
20
 Gross error

 Gross error could be attributed to human

imperfection.
 Occurs in computer programs, also can occur
in any stage of mathematical modeling.

21
 Formulation Errors

 Formulation or model error relate to

incomplete mathematical models.

22
 Data uncertainty

 Errors sometimes enter into an analysis

because of uncertainty in the physical data.
 Ex. In the case of falling parachutist
 Our sensor of velocity can overestimate the
velocity, etc.

23