You are on page 1of 11

NUMERICAL DIFFERENTIATION Part 5

Chapter 9

Richardson
Extrapolation
At the end of this chapter, students should be able to:
 Derive Richardson extrapolation
 Identify and apply Richardson extrapolation technique
 Determine error involve in the Richardson extrapolation method

9.1 Introduction
We have mentioned earlier in the study of numerical differentiation that the
three-point methods are better than the two-point methods in estimating first
order derivatives. We have also discovered that the three-point central is
generally the best method among the two-point and three point methods. In
addition, we have also shown using several examples that smaller step size h
will lead to better approximations of the derivatives. Perhaps it is wise to
consider another interesting method to approximate the first order derivatives
which is far better than any of the three-point methods discussed in the
preceding chapter. The general idea behind Richardson extrapolation is that
it uses two derivative estimates to compute a third approximation and thus
providing a more accurate approximation of the derivative. The following is a
brief discussion of the Richardson extrapolation method (based on a paper
written by Richardson and J.A. Gaunt in 1927).

141
NUMERICAL DIFFERENTIATION Part 5

9.2 Richardson Extrapolation


To see how Richardson extrapolation works, let us begin by considering an
example where the best three-point method, that is the central difference
method, is used to find f ' (x) . As we have seen earlier, the central difference
formula is given by:
f(x  h)  f(x  h)
f (x)   O(h 2 )
2h
Now, using different values of h into the formula, say h1 we have
f ( x  h1 )  f ( x  h1 )
f ( x )   O(h1 )
2

2h1

Or, the exact value of the derivative,

f ' (x)  R1  c1h12 ,

where R1 is estimates of f ' (x) using h1 . Similarly, by substituting another


value of h, say h2, we have
f ' (x)  R2  c 2 h22 ,
here R2 is estimates of f ' (x) using h 2 . The values for c1 and c2 could be

different. Now, let us assume that c1  c 2  c , so from the above we have

R1  ch12  R 2  ch22  exact value of f ' (x)

 
c h12  h 22  R 2  R1

Solving for c,
R 2  R1
c
h12  h 22
So,
R2  R1 2 R2  R1
exact value of f ' (x)  R2   h2  R2   h22
h12  h22  h 2

h22  12  1 
 h2 
Simplifying,
R 2  R1
exact value of f ' (x)  R 2 
 
h1 2
h2
1

142
NUMERICAL DIFFERENTIATION Part 5

Example 1

Let f ' (x)  x 3 . Use the three-point central difference formula with h1 = 0.1

and h2 = 0.05 to compute f ' (1) .Then, apply the Richardson extrapolation

to improve the estimate of f ' (1) .

Solution

Using three-point central difference formula,


h= 0.1:
1.13  0.9 3
f ' (1) 
2(0.1)
 3.01

h =0.05:
1.05 3  0.95 3
f ' (1) 
2(0.05)
 3.0025

h1  0.1 f (1)  3.01  R1


h2  0.05 f (1)  3.0025  R2

Applying Richardson extrapolation,

R 2  R1
f ' (x)  R 2 
 
h1 2
h2
1

3.0025  3.01
f ' (1)  3.0025  2
 3.0000
 0.1 
  1
 0.05 

Note that the actual derivative is f ' (1)  3(1) 2  3.0000 .

Richardson Extrapolation uses two derivative estimates to compute a third,


more accurate approximation.

143
NUMERICAL DIFFERENTIATION Part 5

Richardson extrapolation is actually equivalent to fitting a fourth-order


polynomial through given data and then evaluating the derivatives by
centered difference formula. Here we attempt to derive Richardson
Extrapolation formula based on the Taylors series.

Recall, the central difference approximation with step size h is given by


f ( x 0  h)  f ( x 0  h) h2 h4 v
 f ' ( x0 )  f ' ' ' ( x0 )  f ( x 0 )  ... (1)
2h 3! 5!
And the central difference approximation with step size 2h is given by
f ( x 0  2h)  f ( x 0  2h) 4h 2 16h 4 v
 f ' ( x0 )  f ' ' ' ( x0 )  f ( x 0 )  ... (2)
4h 3! 5!
Notice that (1) is a better derivative estimates than (2) based on its step size.
However combining the two estimates yields an approximation that is better
than either of them.

Taking 4 times (1) minus (2) gives the following,


f ( x0  2h)  8f ( x0  h)  8f ( x0  h)  f ( x0  2h) 12h 4 v
 3f ' ( x0 )  f ( x0 )  ...
4h 5!
Simplifying for f ' (x) , an improved estimate is given by

f ( x 0  2h )  8f ( x 0  h )  8f ( x 0  h )  f ( x 0  2h ) 12h 4 v
f ' ( x0 )   f ( x 0 )  ...
12h (3)5!

Richardson extrapolation is equivalent to using 5 points and the formula can

be written as below.
f(x  2h)  8f(x  h)  8f(x  h)  f(x  2h)
f (x) ≈  O(h4 )
12h
with error term,
h4
E - f (5)
( ) , x - 2h    x  2h
30
Or more conveniently,

h4
E- M where f (5)(x)  M
30

Steps – Richardson Extrapolation method


 Identify the point x that needs to be evaluated
 Identify the step size h
 Apply the Richardson extrapolation formula

f(x  2h)  8f(x  h)  8f(x  h)  f(x  2h)


f (x) ≈
12h

144
NUMERICAL DIFFERENTIATION Part 5

Example 2
A function is given in tabular form below. Find an approximation to the first
derivative of f(x) at x = 0.8 with error O(h4).

x 0.6 0.7 0.8 0.9 1

f(x) 5.9072 6.0092 6.3552 6.9992 8.0000

Solution
 Identify the point x that needs to be evaluated
x = 0.8
 Identify the step size h
h = 0.1.

 Apply the Richardson extrapolation formula

f(x  2h)  8f(x  h)  8f(x  h)  f(x  2h)


f' (x) 
12h
So

f (0.8  2(0.1))  8f (0.8  0.1)  8f (0.8  0.1)  f (0.8  2(0.1))


f ' (0.8) 
12(0.1)

f (0.6))  8f (0.7)  8f (0.9)  f (1)



12(0.1)
5.9072  8(6.0092)  8(6.9992)  8

1.2
 4.856

Example 3
The following data was taken during a physics experiment about a moving
particle at time t and velocity v(t). Calculate the acceleration of the particle
at t = 30 seconds.
t (seconds) 10 20 30 40 50
v(t) (m/s) 50 110 160 180 210

145
NUMERICAL DIFFERENTIATION Part 5

Solution

Note that the acceleration is the first derivative of velocity function v(t).
 Identify the point t that needs to be evaluated
t = 30
 Identify the step size h
h = 10

 Applying the Richardson extrapolation formula

v(t  2h)  8v(t  h)  8v(t  h)  v(t  2h)


v' (t) 
12h
So,
v(30  2(10))  8v(30  10)  8v(30  10)  v(30  2(10))
v' (30) 
12(10)

v(10)  8v(20)  8v(40)  v(50)


v' (30) 
120

50  8(110)  8(180)  210


v' (30) 
120
 3.333 m/s 2

Thus the acceleration of the moving particle at t = 30 seconds is


approximately 3.333 m/s2.

Example 4
The distance x of a runner from a fixed point is measured (in meters) at
intervals of half a second. The data obtained are as follows:
t 0.0 0.5 1.0 1.5 2.0
x 0.00 3.65 6.80 9.90 12.15

Approximate the runner’s velocity at times t = 1.0 second using numerical


differentiation formula with the following order of error
(a) O(h) (b) O(h2) (c) O(h4)

Solution

146
NUMERICAL DIFFERENTIATION Part 5

(a) (i) Using two-point forward difference with O(h):

f (1.5)  f (1.0)
f ' (1.0) 
(0.5)
9.90  6.80

0.5
 6.2
(ii) Using two-point backward difference with O(h):
f (1.0)  f (0.5)
f ' (1.0) 
(0.5)
6.80  3.65

0.5
 6.3
(b) Using three-point central difference with O(h2):

f ( x  h)  f ( x  h)
f ( x ) 
2h
f (1.5)  f (0.5)
f (1) 
2(0.5)
9.90  3.65

0. 1
 6.25

(c) Using Richardson Extrapolation with O(h4):


 Identify the point x that needs to be evaluated
x = 0.1
 Identify the step size h
h = 0.5
 Apply the Richardson extrapolation formula

f(x  2h)  8f(x  h)  8f(x  h)  f(x  2h)


f' (x) 
12h
So

f (1.0  2(0.5))  8f (1.0  0.5)  8f (1.0  0.5)  f (1.0  2(0.5))


f ' (1.0) 
12(0.5)

f (0)  8f (0.5)  8f (1.5)  f ( 2)



6

0  8(3.65)  8(9.90)  12.15



6
 6.3083

147
NUMERICAL DIFFERENTIATION Part 5

Warm up exercise

Consider the f(x)  3x 2  4x. Find the first derivative of f(x) at x =2 with
h = 0.1 using:
(i) the two -point method
(ii) the three-point central difference method.
(iii) the Richardson extrapolation
Discuss your answers and compare it with the exact solution. So which is
better?

9.3 Error Analysis


Example 5
Determine the error if Richardson formula is used to approximate the first
derivative of f(x) = cos x in [0.9, 1.3] with
a) h = 0.1 b) h = 0.01

Solution

 h 4M
Given E  while f (5 ) ( x )  M
30

f(x)  cosx f ' ' ' (x)  sinx


f' (x)  sinx f (4) (x)  cosx
f' ' (x)  cosx f (5) (x)  sinx

For 0.9  x  1.3 then 0.7833   sin x  0.9636

M  0.9636

148
NUMERICAL DIFFERENTIATION Part 5

a)

(0.1)4 (0.9636)
E
30
 3.212x10 6

b)

(0.01)4 (0.9636)
E
30
 3.212x10 10

Warm up exercise

Determine the error if Richardson formula is used to approximate the


first derivative of f(x) = 2 sin x in [0.9, 1.1] with
(i) h = 0.1 (ii) h = 0.01

Exercise 9

1. Use Richardson extrapolation to evaluate f ' (x) and calculate the error bound
for the following functions.
a) f(x) = cos(x) at x = π/4 with h = 0.01, h = 0.005
b) f(x) = ln(1 + x) at x = 1 with h = 0.01, h = 0.005
c) f(x) = tan−1 x at x = 2 with h = 0.01, h = 0.005

d) f (x)  ex at x = 1.15 with h = 0.01, h = 0.001

1
e) f (x)  at x = 1 with h = 0.1, h = 0.01
x

f) f ( x )  e  x sin x at x = 1 with h = 0.1

g) f ( x )  xe x at x = 2 with h = 0.2

149
NUMERICAL DIFFERENTIATION Part 5

2. Refer to the following table of values.


x 0.60 0.65 0.70 0.75 0.8
f(x) 0.9738 0.9290 0.8632 0.7781 0.6755

Use the Richardson’s extrapolation to estimate f ' (0.7) . Calculate the true
value of the derivative. Then, if f(x) = sin (3x), find the maximum bound of the
error using the Richardson’s error term and compare it with the true absolute
error. Comment on your observation.

150
NUMERICAL DIFFERENTIATION Part 5

http://www.math.dartmouth.edu/~calcsite/video1.html#406
Source: 1999 Addison Wesley Longman, Inc.
http://www.mathcs.emory.edu/ccs/ccs215/integral/node3.html

151

You might also like