Interpolation

Chapter: Interpolation

Paper: Numerical Methods

Author: Akash Varshney

College: Sri Venkateswara College, University of Delhi

Institute of Lifelong Learning, University of Delhi pg. 1

 Interpolation

Table of Contents

Chapter: Interpolation

1. Learning Outcomes
2. Introduction
3. Methods of Interpolation
4. Lagrange and Newton Interpolation
5. Linear Interpolation
6. Lagrange’s Interpolation
7. Newton’s Dividend Difference Interpolation
8. Truncation Error Bounds
9. Quadratic Interpolation
10. Higher Order Interpolation
11. Finite Difference Operator
12. Relation between Newton's divided differences in terms of
forward, backward and central difference operators
13. Relations between Difference operators and differential
operators
14. Interpolating Polynomials using Finite Differences
14.1. Gregory-Newton Forward Difference Interpolation
14.2. Gregory-Newton Backward Difference Interpolation
15. Exercise
16. Summary
17. References

1. Learning Outcomes:

In this chapter reader will learn

(i) Definition and meaning of interpolation.

(ii) How with the help of polynomial we can approximate the value of a
function which is not known explicitly.

Institute of Lifelong Learning, University of Delhi pg. 2

 Interpolation

(iii) There are different methods of constructing a polynomial, like

Newton-Gregory method for equally spaced points(abcissaes) and
Lagranges method for unequally spaced points.

(iv) Reader will appreciate the error analysis explained in this chapter .
Suitability of degree of polynomial and bounds on the error is also
explained.

(v) Finite difference operators are also introduced to understand a

different approach in interpolation.

2. Introduction:

Interpolation: Interpolation means insertion or filling up intermediate

terms of series. Interpolation is the method of estimating the value of a
function (dependent variable) for any intermediate value of the
independent variable when some values of the function corresponding to
the values of the variable are given.

that is, given the set of functional values  x0 , y0  ,  x1, y1  ,  x2 , y2  ,...,  xn , yn 

satisfying the relation y  f  x  where the explicit nature of f  x  may not

be known, it is required (desired) to find a simpler function say   x  such

that f  x  and   x  agree at the set of tabulated points, such a process is

called as interpolation and if   x  happens to be a polynomial than the

process is polynomial interpolation.   x  approximates (evaluates) for f

(x)

3. Methods of Interpolation:

Following are the methods of Interpolation

(a) Graphic Method

(b) Method of Curve fitting

(c) Use of finite difference formulae.

Institute of Lifelong Learning, University of Delhi pg. 3
 Interpolation

Interpolation or interpolating polynomial are having two main uses.

(i) The first use is in reconstructing the function f  x  when it is not

given explicitly and only the values of f  x  and for its certain order

derivatives at a set of points, called nodes, tabular points or

arguments are known.

(ii) The second use is to replace the function f  x  by an interpolating

polynomial  x so that many common operations such as

determination of roots, differentiation, integration etc. may be

carried out easily using   x  .

Now

As a justification for the approximation of an unknown function

f  x  by means of a polynomial   x  we state here without proof a

theorem due to Weierstrass:

Statement : If f  x  is continuous in x0  x  xn , than given any  > 0,

there exists a polynomial   x  . ( = arbitrarily small)

such that

f  x    x    for all x   x0 , xn 

The above theorem clearly says that there always exists a polynomial
  x  which is sufficiently close to f  x  .

4. Lagrange and Newton Interpolation:

Let us assume that f  x  is a function defined and continuous on  a, b 

and we have n + 1 points.

Institute of Lifelong Learning, University of Delhi pg. 4

 Interpolation

a  x0  x1  x2  ...  xn1  xn  b , at these  n  1 points values of f  x 

are known.

We want to find the polynomial

P  x   a0  a1x  a2 x 2  ...  an x n (i)

which satisfies the conditions

P  xi   f  xi  i  0,1,2,...n (ii)

Putting  n  1 point x0 , x,..., xn in eqn. (i) & using (ii) we get

a0  a1 x0  a2 x02  ...  an x0n  P  x0   f  x0 

a0  a1 x1  a2 x12  ...  an x1n  P  x1   f  x1 

a0  a1 xn  a2 xn2  ...  an xnn  P  xn   f  xn 

This system of equation has a unique solution or polynomial P  x  exists if

the Vandermonde’s determinant

1 x0 x02 x0n
1 x1 x12 x1n
V  x0 , x1 ,..., xn   0

1 xn xn2 xnn

Uniqueness : The polynomial obtained above is unique.

Suppose that there is another polynomial P  x  which also satisfies

P  xi   f  xi  i  0,1,2,..., n

Consider the polynomial

Institute of Lifelong Learning, University of Delhi pg. 5

 Interpolation

Q  x   P  x   P  x 

Since P  x  & P  x  are polynomials of degree n.

 Q  x  is also a polynomial of degree  n .

Also at x0 , x,..., xn

Q  xi   P  xi   P  xi 

= f  xi   f  xi  i  0,1,..., n

=0

 Q  x  is a polynomial of degree  n which has n + 1 distinct roots

x0 , x,..., xn .

 Q x  0 [ a poly. of degree  n cannot have  n  1 roots].

5. Linear Interpolation:

Let P  x   a1x  a0 where a0 , a1 are constants which satisfies the

interpolating conditions

f  x0   P  x0 

f  x1   P  x1 

 a1x0  a0  P  x0   f  x0  (i)

a1 x1  a0  P  x1   f  x1  (ii)

Eliminating (or obtaining) a0, a1 from (i), (ii) we obtain the required
linear interpolating polynomial.

Institute of Lifelong Learning, University of Delhi pg. 6

 Interpolation

P x x 1
f  x0  x0 1  0 (iii)
f  x1  x1 1

6. Lagrange’s Interpolation:

P x x 1
Expanding the determinant equation f  x  x0 1  0 we get
f  x1  x1 1

P  x  x0  x1   f  x0  x  x1   f  x1  x  x0   0

P x 
 x  x1  f x  x  x0 f x
    
 x0  x1  0  x1  x0  1

P x  0  x  f  x0   1  x  f  x1  (iv)

where

x  x1 x  x0
 x  ,  x 
0
 x0  x1  1
 x1  x0 

0  x & 1  x are called the Lagrange fundamental polynomial

satisfying

0  x  1  x  1

0  x0   1 , 0  x1   0

1  x0   0 , 1  x1   1

1 if i  j
In general x    
0 if i  j
i j ij

Polynomial represented by equation (iv) is called as Lagrange’s

Interpolating polynomial.

Institute of Lifelong Learning, University of Delhi pg. 7

 Interpolation

7. Newton’s Dividend Difference Interpolation:

Consider the determinant Equation.

P x x 1
f  x0  x0 1  0 (i)
f  x1  x1 1

Expanding along first how we get

P  x  x0  x1   x  f  x0   f  x1    1 x1 f  x0   x0 f  x1    0

 P  x   x0  x1   xf  x0   xf  x1   x1 f  x0   x0 f  x1   x0 f  x0   x0 f  x0   0

 P  x  x0  x1   x  f  x1   f  x0    x0  f  x1   f  x0  

  x1  x0  f  x0   0

 P  x  x0  x1    x0  x1  f  x0    x  x0   f  x1   f  x  

 P  x 
 x0  x1  f  x0    x  x0   f  x1   f  x0  
 x0  x1    x1  x0 

 P  x   f  x0    x  x0 
 f  x   f  x 
1 0

 x1  x0 

 P  x   f  x0    x  x0  f  x0 , x1  (ii)

f  x1   f  x0 
The ratio f  x0 , x1   is called as first dividend difference of
 x1  x0 
f  x  relative to x0 and x1 Polynomial P  x  represented by (ii) is Newton’s

Dividend Difference Interpolating Polynomial.

I.Q.1

Institute of Lifelong Learning, University of Delhi pg. 8

 Interpolation

Example 1: Given f  2  4, f  2.5  5.5 , find the linear interpolating

polynomial using

(i) Lagrange Interpolation

(ii) Newton’s Dividend difference interpolation

Hence find an approximate value of f  2.2 

Solution: We have

x0  2, f  x0   4

x1  2.5, f  x1   5.5

(i) The Lagrange fundamental polynomials are given by

x  x1 x  2.5
 x     2  x  2.5
x0  x1 0.5
0

x  x0 x  2
 x    2  x  2
x1  x0
1
0.5

P1  x   0  x  f  x0   1  x  f  x1 

 2  x  2.5 4   2  x  2 5.5 

  2 x  5 4    2 x  4 5.5 

 8x  20  11x  22

 3x  2 .

(ii) Newton’s dividend difference interpolation

We have

f  x1   f  x0  5.5  4
f  x0 , x1    3
x1  x0 0.5

Institute of Lifelong Learning, University of Delhi pg. 9

 Interpolation

P1  x   f  x0    x  x0  f  x0 , x1 

 4   x  2  3  4  3x  6  3x  2

f  2.2  P1  2.2  3   2.2   2

= 6.6  2 = 4.6.

8. Truncation Error Bounds:

The polynomial P  x  coincides with the function f  x  at x0 and x1 and it

deviates at all other points, in the interval  x0 , x1  This deviation is called

the truncation error.

E1  f , x   f  x   P  x  (i)

we will develop bound on E1

Useful result (Rolle’s Theorem) if g  x  is a continuous function on some

interval  a, b  and differentiable on  a, b  and g  a   g  b    0  , than there

exist at least one point   a, b  such that g      0

If x  x0 or x  x1 E1  f , x   0

If x   x0 , x1  define a function

g  t   f  t   P  t    f  x   P  x  .
 t  x0  t  x1  (ii)
 x  x0  x  x1 

at t  x0 , x1 , x

clearly g  t   0

 we can say that g  t  satisfies the conditions of Rolle’s Theorem.

Applying Rolle’s theorem on the intervals  x0 , t  and  t , x1  separately we

get

Institute of Lifelong Learning, University of Delhi pg. 10

 Interpolation

g   1   0 x0  1  t

g   2   0 t  2  x1 (By the conclusion of Rolle’s Theorem).

where

 2t   x0  x1 
g   t   f   t   P  t    f  x   P  x   
 x  x0  x  x1 

g   t   f   t   P  t    f  x   P  x   2  x  x0  x  x1  (iii)

Now g   t  also satisfies the conditions of Rolle’s Theorem.

Applying Rolle's thm for g   t  on the interval  1 , 2  we get

g      0 1    2 or x0    x1

 x0  1    2  x1 

2  f  x   P  x 
 0  f     
 x  x0   x  x1 

1
 f       x  x0  x  x0    f  x  P x
2

1
 f  x  P x   x  x0  x  x1  f     (iv)
2

Using (iv)

Truncation error in linear interpolation is

 E1  f , x   f  x   P  x  
1
E1  f , x    x  x0  x  x1  f    
2

Institute of Lifelong Learning, University of Delhi pg. 11

 Interpolation

1 1
 f  x  P x   x  x0  x  x1  f      xmax  x  x0  x  x1  f    
2 2 0  x x1
1
 max  x  x0  x  x1  M (v)
2 x0  x x1

where M is s.t. f   x   M for all x   x0 , x1 

man value of  x  x0  x  x1  occur at x 

 x0  x1 
2

Let w  x    x  x0  x  x1     x  x0  x  x1 

w  x    x  x0    x  x1  , w  x   2

= 2 x   x0  x1 

x0  x1
Put w  x   0  x 
2

x0  x1
 maximum value of  x  x0  x  x1  occur at and maximum
2

 x1  x0 
2

value of w  x  is w(x) =
4

 From (v) we get

1
f  x  P x   x1  x0  M
2
(vi)
8

Value addition
Equation (vi) can be used to construct a table of values for a
function f  x  for equally spaced points
xi  a  ih i  0,1,..., n , h   b  a  n
This can help us in setting maximum absolute truncation error using
the linear interpolating polynomial P  x  is less than a given error
tolerance  > 0
as x1  x0  h
 from equation (vi)

Institute of Lifelong Learning, University of Delhi pg. 12

 Interpolation

h2
max f   x    .
8 a  x b

I.Q.2

I.Q. 3

Example.2 : Using the data sin(0.1) = 0.09983 and sin(0.2) = 0.19867,

find an approximate value of sin(0.15) by Lagrange Interpolation. Also
obtain a bound on the truncation error.

Solution: Using Lagrange Interpolation

x  x1 x  x0
P x  f  x0   f x  (i)
 x0  x1   x1  x0  1

Here x0  0.1, x1  0.2, put x = 0.15, f  x   sin x

 from (i)

0.15  0.2 0.15  0.1

P  0.15    0.09983    0.19867 
0.1  0.2 0.2  0.1

=  0.5 0.09983   0.5 0.19867  0.14925

 The truncation error is

E1  f ; x  
 x  0.1 x  0.2 
  sin  
2


 x  x0  x  x1  f    
1
 E1  f , x  
2 

where

0.1    0.2

 
As sin x is a increasing function in  0, 
2  
Institute of Lifelong Learning, University of Delhi pg. 13
 Interpolation

 The maximum value of  sin   0.1,0.2 is sin (0.2) = 0.19867

Thus E1  f ; x  
 0.15  0.1 0.15  0.2 
 0.19867 
2

  0.19867  0.00125   0.00025

Example 3: Determine the step size h that can be used in the tabulation
of f  x   sin x in the interval [1, 3] so that the linear interpolation will be

correct to four decimal places after rounding.

Solution: We have

f  x   sin x

f   x   cos x, f   x    sin x

   
and max  sin x  1  1  3 & sin  1
1 x 3
 2 2 

According Problem

h2
max f   x   
8 1 x3

Here   5 105

h2
  1  5  105
8

 h  0.02

9. Quadratic Interpolation:

Here n  2 , and we want to determine a polynomial P2  x   a0  a1 x  a2 x 2

where a0, a1, a2 are arbitrary constants satisfying the interpolatory

condition f  x0   P2  x0  , f  x1   P2  x1  f  x2   P2  x2  .

Institute of Lifelong Learning, University of Delhi pg. 14

 Interpolation

f  x0   a0  a1x0  a2 x02

f  x1   a0  a1x1  a2 x12

f  x2   a2  a1x2  a2 x22 .

Eliminating a0 , a1 , a2 , we obtain the required quadratic interpolating

polynomial as

P2  x  1 x x2
f  x0  1 x0 x02
0
f  x1  1 x1 x12
f  x2  1 x2 x22

Expanding the determinant we obtain

P2  x  D0  f  x0  D1  f  x1  D2  f  x2  D3  0

where

1 x0 x02
D0  1 x1 x12   x0  x1  x1  x2  x2  x0 
1 x2 x22

1 x x2
D1  1 x1 x12   x  x1  x1  x2  x2  x 
1 x2 x22

1 x x2
D2  1 x0 x02   x  x0  x0  x2  x2  x 
1 x2 x22

1 x x2
D3  1 x0 x02   x  x0  x0  x1  x1  x 
1 x1 x12

Institute of Lifelong Learning, University of Delhi pg. 15

 Interpolation

D1 D D
 P2  x   f  x0   2 f  x1   3 f  x3 
D0 D0 D0

 x  x1  x  x2  f x   x  x0  x  x2  f x
    
 x0  x1  x0  x2  0  x1  x0  x1  x2  1

 x  x0  x  x1  f x
  
 x2  x0  x2  x1  2

= 0  x  f  x0   1  x  f  x1   2  x  f  x2 

The truncation error of the Lagrange quadratic interpolating polynomial is

E2  f ; x   f  x   P2  x 

Using Rolle’s Thm we get expression for f  x  estimated by P2  x  and the

error adjustment we get


f  x   P2  x    x  x0  x  x1  x  x2  f    
3!

where   x0 , x2 

 Truncation error

1
E2 ( f ; x)  f  x   P2  x    x  x0  x  x1  x  x2  f    
3
and bound on the truncation error is

f  x   P2  x   M 3  max  x  x0  x  x1  x  x2  
1
6  x0  x x2 

where M 3  max f   x 

x0  x  x2

Institute of Lifelong Learning, University of Delhi pg. 16

 Interpolation

10. Higher Order Interpolation:

The Lagrange fundamental polynomials of degree n based on n + 1

distinct points a  x0  x1  x2  ........  xn  b and which satisfy the condition

1 if i  j
li  x j   ij  
0 if i  j

Can be written in the form

li  x  
 x  x0  x  x1 ... x  xi 1  x  xi 1 ... x  xn 
 xi  x0  xi  x1 ... xi  xi 1  xi  xi1 ... xi  xn 

i = 0, 1, …, n. (i)

An alternative way of writing li  x  can be

w x 
li  x  
 x  x0  w  xi 

where

w x    x  x0  x  x1 ... x  xn 

w  xi    xi  x0  xi  x1 ... xi  xi 1  xi  xi 1 ... xi  xn 

and w  x  is the derivative of w  x  with respect to x.

thus the polynomial

n
P  x    li  x  f  xi 
i 0

is the Lagrange interpolating polynomial of degree n.

The truncation error in the Lagrange interpolation is given by

En  f ; x   f  x   P  x 

Institute of Lifelong Learning, University of Delhi pg. 17

 Interpolation

Since En  f ; x   0 at x = xi i = 0, 1, 2, …, n

than for x   a, b and x  xi , we define a function g  t  as

g  t   f  t   P  t    f  x   P  x 
 t  x0  t  x1 ... t  xn  (ii)
 x  x0  x  x1 ... x  xn 

We observe that g  t   0 at t  x and t  xi , i  0,1,..., n

Applying the Rolle's theorem repeatedly for g  t  , g   t  ,... and g t 

 n
we

obtain g 
n 1
  0 where  is some point such that

min  x0 , x1 ,.....xn , x     max  x0 , x1,.....xn , x 

Differentiating g  t  represented by equation (ii)  n  1 times with respect

to t, we get

 n  1! f  x   P  x 
g
n 1
t   f
n 1
t   (iii)
 x  x0  x  x1 .... x  xn 

[  n  1 th derivative of P  x  is zero as P  x  is a polynomial of degree n]

   0 and solving equation (iii) for f  x  , we get

 n 1
Setting g

w  x   n1
f  x  P x  f 
 n  1!

Hence, the truncation error in Lagrange interpolation is given by

w  x   n1
En  f ; x   f  .
 n  1!

Example4. Given that f  0  1, f 1  3, f  3  55 , find the unique

polynomial of degree 2 or less, which fits the given data using.

(i) Lagrange Method

Institute of Lifelong Learning, University of Delhi pg. 18

 Interpolation

(ii) Newton divided difference method

Also find the bound on the error.

Ans (i) We have x0  0, x1  1, x2  3, f0  1, f1  3 and f 2  55 . The

Lagrange fundamental polynomials are given by

 x  x1  x  x2    x  1 x  3  1 x2  4 x  3
l0  x  
 x0  x1  x0  x2   1 3 3
 

 x  x0  x  x2   x  x  3  1 3x  x 2
l1  x  
 x1  x0  x1  x2  1 2  2
 

 x  x0  x  x2   x  x  1  1 x 2  x .
l2  x  
 x2  x0  x2  x1  3 2  6
 

Hence, the Lagrange quadratic interpolating polynomial is given by

P2  x   l0  x  f0  l1  x  f1  l2  x  f 2


3

1 2
x  4 x  3   3 x  x 2    x 2  x 
3
2
55
6

 8x2  6 x  1 .

(ii) The divided differences are given by

3 1 55  3
f 0,1   2, f 1,3   26,
1 0 3 1

26  2
f 0, 1, 3  8.
30

The Newton divided difference interpolating polynomial becomes

P2  x   f 0   x  0 f 0,1   x  0  x  1 f 0,1,3

 1  2 x  8 x  x  1  8x 2  6 x  1.

Institute of Lifelong Learning, University of Delhi pg. 19

 Interpolation

Bound on the Error :

1
We have, E2  f ; x   M 3  max x  x  1 x  3 
6  0  x 3 

1
  2.1126  M 3  0.3521 M 3
6

where M 3 = max f   x  and since the maximum of x  x  1 x  3 occurs

0 x 3

at x = 2.2152.

Example 5. The following values of the function f  x   sin x  cos x, are

given

x 10 20 30

f (x) 1.1585 1.2817 1.3660

Construct the quadratic interpolating polynomial that fits the data.

Hence, find f   /12  . Compare with the exact value.

Since the value of f at /12 radians is required, we convert the data

into radian measure. We have

 
x0  10   0.1745, x1  20   0.3491,
18 9


x2  30   0.5236 .
6

The Lagrange fundamental polynomials are given by

l0  x  
 x  x1  x  x2    x  0.3491 x  0.5236
 x0  x1  x0  x2   0.1746 0.3491

l1  x  
 x  x0  x  x2    x  0.1745 x  0.5236
 x1  x0  x1  x2   0.1746 0.1745

Institute of Lifelong Learning, University of Delhi pg. 20

 Interpolation

 32.8616  x 2  06981 x  0.0914 

l2  x  
 x  x0  x  x1    x  0.1745 x  0.3491
 x2  x0  x2  x1   0.3491 0.1745

 16.4155  x 2  0.5236 x  0.0609  .

Hence, the Lagrange quadratic interpolating polynomial is given by

P2  x   16.4061 x 2  0.8727 x  0.1828 1.1585

32.8616  x 2  0.6981 x  0.0914  1.2817 

16.4155  x 2  0.5236 x  0.0609  1.3660 

= 0.6887 x  1.0751 x  0.9903

2

Hence, f   /12   f  0.2618  1.2246 ,

The exact value is f  0.2618  sin  0.2618  cos  0.2618  1.2247 .

Example 6. Construct the divided difference table for the data

x 0.5 1.5 3.0 5.0 6.5 8.0

f (x) 1.625 5.875 31.0 131.0 282.125521.0

Hence, find the interpolating polynomial and an approximation to

the value of f (7). We have the following divided difference table

x f (x) first order second order third order fourth order

d.d. d.d. d.d. d.d.
0.5 1.625
4.25
1.5 5.875 5.0
16.75 1.0
3.0 31.000 9.5 0
50.00 1.0
5.0 131.000 14.5 0
100.75 1.0
Institute of Lifelong Learning, University of Delhi pg. 21
 Interpolation

6.5 282.125 19.5

159.25
8.0 521.000

We write the divided difference interpolating polynomial as

f  x   f  x0    x  x0  f  x0 , x1    x  x0  x  x1  f  x0 , x1 , x2 

  x  x0  x  x1  x  x2  f  x0 , x1 , x2 , x3 

 1.625   x  0.5 4.25  5  x  0.5  x  1.5 

  x  0.5 x  1.5 x  3.0 

 1.625  2.125  3.75  2.25   x  4.25 10.0  6.75 

 x 2  5  5  x3

 x3  x  1 .

Hence, f  7.0   351 .

11. Finite Difference Operator:

Let the points x1 , x2 , x3 ,...., xn be equally spaced

 xi  x0  ih, i  0,1,2,..., n, h  x1  x0  x2  x1  xn  xn1

We define the following operators

(1) Shift Operator Ef ( xi )  f ( xi  h)

(2) Forward difference operator f ( xi )  f ( xi  h)  f ( xi )

(3) Backward difference operator f ( xi )  f ( xi )  f ( xi  h)

 h  h
(4) Central difference operator  f ( xi )  f  xi   f  xi  
 2  2

Institute of Lifelong Learning, University of Delhi pg. 22

 Interpolation

1  h  h 
(5) The average operator  f ( xi )  f  xi    f  xi   
2   2  2 

Relation between the operators (take h = 1).

 1
(i) f ( xi )  f ( xi  1)   f  xi    f ( xi  1)  f ( xi )
 2

(ii) (a) f ( xi )  f ( xi  1)  f ( xi )

= Ef ( xi )  f ( xi )

 ( E  1) f ( xi )


  E 1

(b) f ( xi )  f ( xi )  f ( xi  1)

 f ( xi )  E 1 f ( xi  1)

 (1  E 1 ) f ( xi )

   1  E 1

   1  

(iii) We can also have

 n  ( E  1)n 

n  (1  E 1 )n 

Table showing relationship between the operators

E   

E E +1  1 1 1

1  2   1  2
2 4

 E1   1  1 1 2 1

   1  2
2 4

Institute of Lifelong Learning, University of Delhi pg. 23

 Interpolation

 1E 1 1  (1+)1  1 1
  2   1  2
2 4

 1

1  (1+)1/2  (1)1/2 
E E
2 2

 1 12 
1
 1  1
 1  1
    1

(E  E 2 )  1    1   2
 1    1   2 1  2
2  2   2  4

Example 7. Show that

1/ 2

1
 2 
(i)   (1  ) 2
(ii)   1  
 4 

1 1 1 2 2
(iii) E  1  2   1  2 (iv)       1
2 4 2 4

Proof:



(1) R.H.S. =   

1

1 1
 (1  E )[1  (1  E )] 2
(  1  E 1 )

1 1

 (1  E 1 )( E 1 ) 2
 (1  E 1 )( E 2 )

1 1

E E 2 2

 = L.H.S.

1
  2 2
(ii) R.H.S. = 1  
 4 
1
 1  1

1 2 1

1
 1  ( E  E )  2 2 2
  E2  E 2

 4 
1
 1 2
1 1

 1  ( E  E 1  2)   2  ( E 2  E 2 )2
 4 
1
= EE 2

Institute of Lifelong Learning, University of Delhi pg. 24

 Interpolation

1/ 2
 4  E  E 1  2 
 
 4 
1 1/ 2
  E  E 1  2
2
1/ 2
1 1 
1

 ( E 2  E 2 ) 2 
2 
1 12 
1
= ( E  E 2 )    L.H .S .
2
1 1
(iii) R.H.S. = 1   2   1   2
2 4
1 12 
1
1

1
   (E  E 2 )
1 12 
1 1

1
(E  E ) 2 2
2
= 1  (E  E )  (E  E 2 ) 
2 2 2
2 2 2
 1 
4
1 1
 1  ( E  E 1  2)  ( E  E 1 )
2 2
1
 [2  E  E 1  2  E  E 1 ]
2
1
  2 E  E  L.H .S .
2
1 1
(iv) R.H.S.=   2   1   2
2 4
1 1

1 1

1
(E  E ) 1

1 2 2
  (E  E )  (E  E )
2 2 2 2 2
(using relation (ii))
2 2
1 1
  ( E  E 1  2)  ( E  E 1 )
2 2
1
 [ E  E 1  2  E  E 1 ]
2
1
 [2  2 E 1 ] = [1  E ]    L.H .S.
1
2

Similarly we can prove other relations given in the table of operators.

Institute of Lifelong Learning, University of Delhi pg. 25

 Interpolation

Example 8. Construct the difference table for the sequence of values

f ( x)  (0,0,0,  ,0,0,0) where  is an error. Also show that

(i) The error spreads and increases in magnitude as the order of

differences is increased.
(ii) The error in each columns have binomial coeff.

Solution:
f(x) f 2f 3f 4f 5f 6f
0 0
0 0 0  4
0   3 6 10 20
  2 3 4 10
0 0  
0 0 0
0
Clearly the table shows

(i) How the error is effected by order of differences

(ii) Difference columns have binomial coefficient.
(iii) Maximum error occurs directly opposite the entry where the
function value is in error.

I.Q.4

I.Q.5

12. Relation between Newton's divided differences in terms of

forward, backward and central difference operators:

f ( x1 )  f ( x0 ) 1
f [ x0 , x1 ]   f 0
h h

1 1
f  f
f [ x1 , x2 ]  f [ x0 , x1 ] h 1 h 0
f [ x0 , x1 , x2 ]  
x2  x0 2h

1
 2 f0 .
2!h 2

Institute of Lifelong Learning, University of Delhi pg. 26

 Interpolation

We use principle of Mathematical Induction. Assume that

1
f [ x0 , x1 , x2 ,...., xn1 ]  n 1
 n1 f 0
(n  1)!h
f [ x1 , x2 ,..., xn ]  f [ x0 , x1 ,...., xn1 ]
f [ x0 , x1 , x2 ,...., xn1 , xn ] 
xn  x0

1 1
1 n 1
 n1 f1  1 n 1
 n1 f 0
(n  1) h (n  1) h

nh

 n1
[ f1  f 0 ]
(n  1)!h n1  n1.f 0 n f0
  
nh (n  1)!h n  n n!h n

Similarly we can prove that

1
(b) f [ x0 , x1 ,..., xn ]  n
n f n
n !h

1
(c) f [ x0 , x1 ,..., x2 n1 ]  2m
 m  fm when n  2m
(2m)!h

1
f [ x0 , x1 ,..., x2 m1 ]  2 m 1
  m+1 f 1 when n  2m  1
(2n  1)!h m
2

13. Relations between Difference operators and differential

operators:

f ( x)  f ( x  h)  f ( x)

 h2 
  f ( x)  hf ( x)  f ( x)  ....  f ( x) (Using Taylor's theorem)
 2 

h2
 hf ( x)  f ( x)
2

 f ( x)  hf ( x) (where h is so small that terms containing h2 and

higher powers of h can be neglected).

f ( x)
 f ( x)  error is of O(h).
h
Institute of Lifelong Learning, University of Delhi pg. 27
 Interpolation

Similarly

2 f ( x)  h2 f ( x)

Again f ( x)  f ( x)  f ( x  h)

 h2 
 f ( x)   f ( x)  hf ( x)  f ( x)  ....
 2 

h2
 f ( x)  f ( x)  hf ( x)  f ( x)  ...
2

 hf ( x)  O(h2 )

Neglecting the terms containing h2 and higher powers of h.

f ( x)
 f ( x)  hf ( x)  f ( x) 
h

  f ( x)
 f ( x)  h f ( x)  f ( x) 
 2

h2

14. Interpolating Polynomials using Finite Differences:

14.1. Gregory-Newton Forward Difference Interpolation:

Relation between divided difference and forward difference operator is

1
f [ x0 , x1 ,..., xn ]  n
n f0 (1)
n !h

Divided difference interpolating polynomial is written as

Pn ( x)  f [ x0 ]  ( x  x0 ) f [ x0 , x1 ]  ...  ( x  x0 ( x  x1 )....( x  xn1 )

f [ x0, x1 ,..., xn ] (2)

Using (1) in (2), Interpolating polynomial can be written as

f 0 ( x  x0 ( x  x1 ) 2 f 0
P( x)  Pn ( x)  f 0  ( x  x0 )   ....
h 2!h2

( x  x0 )( x  x1 )  ....( x  xn1 ) n
  f0 (3)
n !h n

Institute of Lifelong Learning, University of Delhi pg. 28

 Interpolation

Polynomial P(x) expressed by equation (3) is known as Gregory-Newton

forward differences interpolating polynomial.

( x  x0 )
Now if we put u   hu  ( x  x0 ) and since x0 , x1 ,..., xn are
h
equally spaced Points that is xi  x0  ih

 ( x  xi )  ( x  ( x0  ih))  [( x  x0 )  ih]  (uh  ih)  (u  i)h .

 Equation (3) and the truncation error can be written as

u (u  1) 2 u (u  1)....(u  n  1) n
P( x0  hu )  f 0  uf 0   f 0  .....   f0 (4)
2! n!

n
u  u 
    i f 0 where    u Ci
i 0  i  i 

and truncation error

u (u  1)....(u  n) n1 ( n1)

En ( f ; x)  h f ( ) (5)
(n  1)!

Alternative Method for the derivation of interpolating polynomial

expressed by equation (4) as

  x  x0  
f ( x)  f  x0    h
  h  

 x  x0 
 f ( x0  uh)   u
 h 

 E u f ( x0 )

 (1  )u f ( x0 ) ( E  1  )

 u (u  1) 2 u (u  1)...(u  n  1) n 
 1  u    ....   .... f 0
 2! n! 

Institute of Lifelong Learning, University of Delhi pg. 29

 Interpolation

u (u  1) 2 u (u  1)...(u  n  1) n
f ( x)  f 0  uf 0   f 0  ....   f 0  ... (6)
2! n!

Neglecting the  n1 f 0 and higher order differences in equation (6)

we get.

u (u  1) 2 u (u  1)...(u  n  1) n
f ( x)  f 0  uf 0   f1  ...   f0 (7)
2! n!

equation (7) is again the form of Gregory Newton forward difference

interpolating polynomial.

14.2. Gregory-Newton Backward Difference Interpolation:

We observe that Newton-interpolation with divided differences in

terms of backward differences should be in terms of the differences at the
end point xn.

 ( x  xn ) 
f ( x)  f  xn   h   f ( xn  hu )
 h 

( x  xn )
as we take = u  x  xn  hu
h

 f ( x)  f ( xn  uh)

 E u f ( xn )

 (1  )u f ( xn ) [ E  (1  )1 ]

u (u  1) 2 u (u  1)  (u  n  1) n
 1  uf ( xn )   f ( xn )  ...   f ( xn )  ... (8)
2! n!

n 1
Neglecting the difference  f ( xn ) and higher order difference we get the

interpolating polynomial as

P( x)  P( xn  hu)  f ( x)

Institute of Lifelong Learning, University of Delhi pg. 30

 Interpolation

u (u  1) 2 u (u  1)....(u  n  1) n
 f n  uf n   f n  ....   fn.
2! n!

n
 u 
  (1)i   i f n (9)
i 0  i

 u  u
where   Ci .
 i

Polynomial expressed by relation (9) is known as Gregory-Newton

backward difference interpolating polynomial and the truncation error is

u (u  1)....(u  n) n1 n1

En ( f ; x)  h f ( )
(n  1)!

I.Q. 6

Example9. For the following data, calculate the differences and obtain
the forward and backward difference polynomials. Interpolate at x = 0.25
and x = 0.35

x 0.1 0.2 0.3 0.4 0.5

f(x) 1.40 1.56 1.76 2.00 2.28

Solution: The difference table is obtained as

x f(x) f(x) 2f(x) 3f(x)

0.1 1.40
0.16
0.2 1.56 0.04
0.20 0.0
0.3 1.76 0.04 0.0
0.24 0.0
0.4 2.00 0.04
0.28
0.5 2.28
The forward difference polynomial is given by

Institute of Lifelong Learning, University of Delhi pg. 31

 Interpolation

0.16 ( x  0.1)( x  0.2) 0.04

P( x)  1.4  ( x  0.1)  
0.1 2 0.01

 2 x2  x  1.28 .

The backward difference polynomial is obtained as

0.28 ( x  0.5)( x  0.4) 0.04

P( x)  2.28  ( x  0.5) 
0.1 2 0.01

 2 x2  x  1.28 .

Both the polynomials are same.

 f (0.25)  1.655, f (0.35)  1.875.

We can obtain the interpolated values directly also. So for x = 0.25 we

choose x0 = 0.2 and write

x  x0 0.25  0.2
u   0.5
h 0.1

1
 f (0.25)  f (0.2)  (0.5)f (0.2)  (0.5)(0.5) 2 f (0.2)
2

= 1.56 +(0.5(0.20)  (0.125)(0.04) = 1.655

For x = 0.35 we choose xn = 0.4 and in backward differences as

x  xn 0.35  0.4
u   0.5
h 0.1

1
and f (0.35)  f (0.4)  (0.5)f (0.4)  (0.5)(0.5)f (0.2)
2

= 2.00  (0.5)(0.24)  (0.125) (0.04)

= 1.875.

Hence the solution.

Institute of Lifelong Learning, University of Delhi pg. 32

 Interpolation

Exercise:

Q.1 Let f  x   log 1  x  x0  1 and x1  1.1.

Use linear interpolation to calculate an approximate value for

f 1.04  and obtain a bound on the truncation error.

Q.2 In the following problems, find the maximum value of the step size
h that can be used to tabulate f  x  on  a, b  , using linear

interpolation such that |Error|   .

f  x   1  x  a, b  0,1,   5 105

6
(a)

1
(b) f  x  a, b  1,2,   1104
1  x2 
(c) f  x   2x a, b  0,1,   1105

Q.3 Find the unique polynomial P  x  of degree 2 or less such that

P 1  1, P  3  27, P  4   64 using each of the following methods.

(i) Lagrange interpolating polynomial

(ii) Newton divided difference formulae.

Estimate P(1.5)

Q.4 In the following problems, the values of a function f  x  are given.

Find the interpolating polynomial that fits the data. Find an

approximation to f  x  at the indicated points using this polynomial.

(i) x –2 –1 0 1 3 4

f (x) 9 16 17 18 44 81

(ii) x –1 2 4 5

Institute of Lifelong Learning, University of Delhi pg. 33

 Interpolation

f (x) –5 13 255 625

1
Q.5 If f  x   , find the dividend difference f  x1, x2 , x3 , x4 
x2

Q.6 Prove the following relations

(i)     

 
(ii)   
 

1
(iii)      fi / fi fi 1
 fi 

Q.7 Construct the interpolating polynomial that fits the data

x 0 0.1 02 0.3 0.4 0.5

f (x) –1.5 –1.27 –0.98 –0.63 –0.22 0.25

Using the Gregory – Newton forward or backward difference

interpolation. Hence or otherwise estimate the values of f  x  at

x = 0.15, 0.25 and 0.45.

Q.8 Using the Newton's backward difference interpolation construct the

interpolating polynomial that fits the data

x 0.1 0.3 0.5 0.7 0.9 1.1

f (x) –1.699 –1.073 –0.375 0.443 1.429

2.631

Estimate the value of f  x  at x = 0.6 and x = 1.0

Summary:

In this chapter we studied how different methods of interpolation helps in

the evaluation and approximation of a function which may not be known

Institute of Lifelong Learning, University of Delhi pg. 34

 Interpolation

explicitly. We studied methods like Lagranges interpolation , Newton

divided difference methods etc. We also derived Gregory – Newton
forward and backward interpolation formulae. Interpolation as a
procedure finds its use in mainly numerical differentiation, numerical
integration , numerical differential equation etc.

Application of Interpolation : An application of interpolation that we

see every day is in weather forcasting . When you watch the weather
forcasts on television, you may wonder where these usually correct
projections come from. The weather service people collect information on
temperature wind speed and direction , humidity and barometric pressure
from hundreds of weather stations around the region. Added to these are
cloud data from satellites that are in elevated orbits above the earth. All
of these data items are entered into a massive computer program that
models the weather using interpolation . Upto a million pieces of data are
involved.

References:

1. A Friendly Introduction to Numerical Analysis by Brian Bradie, Sixth

Impression, Pearson Prentice Hall.

2. Introductory Methods of Numerical Analysis by S.S. Sastry, Prentice

Hall of India.

3. Finite Differences and Numerical Analysis by H.C. Saxena, S. Chand

& Company Ltd.

4. Numerical Methods for Scientific and Engineering Computation by

M.K. Jain, S.R.K. Iyengar, R.K. Jain Sixth Edition, New Age
International Publishers.

5. Applied Numerical Analysis, Seventh Edition, Curtis F. Gerald,

Patrick O. Wheateley, Pearson.

Institute of Lifelong Learning, University of Delhi pg. 35