2.8.

Newton’s backward difference interpolation

Let y = f (x) be a function which takes the value y 0 , y1 ,..., y n for the equidistant values x0 , x1 ,

x2 ,..., xn , i.e. x i = x 0 + ih for all i=0,1,2,3…, n .

Let, ϕ n (x) be a polynomial of degree n. This polynomial may be written as

ϕ n (x ) = a0 + a1 ( x − xn ) + a2 ( x − xn )( x − xn −1 ) + ... + an ( x − xn )( x − xn −1 )...( x − x1 ) (2.8.1)

We shall now determine the coefficients a0 , a1 ,..., a n so as to make

ϕ n ( xn ) = y n , ϕ n (xn −1 ) = y n −1 ,…, ϕ n (x0 ) = y0

Substituting in eq. (2.8.1) the successive values x n , x n −1 ,... for x at the same time putting

ϕ n ( xn ) = y n , ϕ n (xn −1 ) = y n −1 ,…, ϕ n (x0 ) = y0 etc., we have

y − y n −1 ∇y n ∇2 yn ∇3 yn ∇ n yn
a0 = y n ; a1 = n = ; a2 = ; a3 = 3 ; …, a n = .
x n − x n −1 h 2!h 2 3!h n! h n

Substituting these values of a0 , a1 ,..., a n in (2.8.1), we have

( x − xn ) ( x − xn )( x − xn −1 ) 2 ( x − xn )( x − xn −1 )...(x − x1 ) n
ϕ n (x ) = yn + ∇y n + 2
∇ y n + ... + ∇ yn (2.8.2)
h 2! h n! h n

which is Gregory-Newton’s backward difference interpolation formula and it is useful to

interpolate near the end of a set of tabular values.

(x − xn )
Now, setting u= ,
h

1
we can obtain

u(u + 1) 2 u(u + 1)...(u + n − 1) n

ϕn ( x ) = yn + u∇yn + ∇ yn + ... + ∇ yn (2.8.3)
2! n!

In this case also in practical numerical computation, instead of eq.(2.8.2) we should use

eq.(2.8.3) in order to ease the calculation involves with it.

2.8.1. Error in Newton’s backward difference interpolation

To find the error committed in approximating f (x) by the polynomial ϕ n (x) , we have from eq.

(2.2.5), the remainder or truncation error or simply error is

( x − x0 )( x − x1 )( x − x2 )...( x − xn ) ( n +1)
Rn +1 ( x) = f (ξ ) , (2.8.1.1)
(n + 1)!

where min{ x, x0 , x1 ,..., xn } < ξ < max{x, x0 , x1 ,..., xn } .

Example 3:

Calculate the values of f ( 218) and f ( 410) from the following data

x 200 250 300 350 400

y 15.04 16.81 18.42 19.90 21.27

Solution:

The forward difference table is

2
 x y ∆y ∆2 y ∆3 y ∆4 y

200 15.04

1.77

-0.16
250 16.81
0.03
1.61
-0.01
-0.13
300 18.42
0.02
1.48

-0.11
350 19.90

1.37

400 21.27

Since, x=410 is near the end of the table, we use Newton’s backward interpolation formula. From

the above forward difference table, only underlined lower portion values will be used in the

Newton’s backward interpolation formula.

x − xn 410 − 400 10
Here, u = = = = 0.2 .
h 50 50

Using Newton’s backward interpolation formula, we get

u(u + 1) 2 u(u + 1)(u + 2) 3 u(u + 1)(u + 2)(u + 3) 4

f (410) ≅ y n + u∇y n + ∇ yn + ∇ yn + ∇ yn
2! 3! 4!

3
 0.2 ×1.2 0.2 ×1.2 × 2.2 0.2 ×1.2 × 2.2 × 3.2
= 21.27 + 0.2 ×1.37 + × (−0.11) + × 0.02 + + × (−0.01)
2! 3! 4!

= 21 .27 + 0 .274 − 0 .0132 + 0. 00176 − 0.000704

= 21.532

2.9. Divided difference

Let us consider y 0 = f ( x0 ) , y1 = f ( x1 ) ,..., y n = f ( x n ) be the values of the function y = f (x)

corresponding to the values x0 , x1 , x2 ,..., xn , which are not necessarily equally spaced.

The first order divided differences of f ( x) for two arguments x0 , x1 denoted by f [ x0 , x1 ] and is

defined by

f ( x0 ) − f ( x1 ) f ( x1 ) − f ( x0 )
f [ x0 , x1 ] = =
x0 − x1 x1 − x0

Similarly,

f ( x1 ) − f ( x 2 ) f ( x 2 ) − f ( x1 )
f [ x1 , x 2 ] = = ,
x1 − x 2 x 2 − x1

f ( x 2 ) − f ( x3 ) f ( x3 ) − f ( x 2 )
f [ x2 , x3 ] = =
x2 − x3 x3 − x 2

and so on.

The second order divided differences of f ( x) for three arguments x0 , x1 , x2 denoted by

f [ x0 , x1 , x2 ] and is defined by

4
 f [ x0 , x1 ] − f [ x1 , x 2 ]
f [ x0 , x1 , x 2 ] =
x0 − x2

In general, the n th order divided differences of f ( x) with n + 1 arguments x0 , x1 , x2 ,..., xn is

defined by

f [ x0 , x1 , x2 ,..., xn−1 ] − f [ x1 , x2 ,..., xn ]

f [ x0 , x1 , x2 ,..., xn ] = .
x0 − xn