UNIT 10 NEWTON FORM OF THE

INTERPOLATING POLYNOMIAL
Structure
10.1 Introduction
Objectives

10.2 Divided Differences

10.3 Newton's General Form of Interpolating Polynomial '

10.4 'IJe Error of the Interpolating Polynomial

10.5 Divided Differences and Derivatives

10.6 Further Results on Interpolation Error

10.7 Summary

10.8 Solutions/Answers

10.1 INTRODUCTION
The Lagrange's form of the interpolating polynomial derived in Unit 9 has some drawbacks
compared to Newton form of interpolating polynomial that we are going to consider now.
,
In practice, one is often not sure as to how many interpolation points to use. One often
calculates Pl(x), Pz(x), ....increasing the number of interpolation points, and hence the
degrees of the interpolating polynomials till one gets a satisfactory approximation Pk(x)to
f(x). In such an exercise. Lagrange form seems to be wasteful as in calculating Pk(x), no
advantage is taken of the fact that one has already constructed Pk- 1 ( ~ )whereas
, in Newton
form it is not so.

Before deriving Newton's general form of interpolating polynomial, we introduce the

concept of divided difference and the tabular representation of divided differences. Also
the error of the interpolating polynomial in this case is derived in terms of divided
differences. Using the two different expressions for the error term we get a relationship
between nth order divided difference and nth order\derivative.

Objectives e

After studying this unit, you should be able to :

obtain a divided difference in terms of function values;

fo& a table of divided differences and find divided differences with a given set of
arguments from the table;

show that divided difference is independent of the orde; of its arguments;

. ,
obtain the Newton's divided differences interpolating polynomial for a givcn data;

find an estimate of f(x) for a given non - tabular value of x from a table of values of
\
x and Y [ f(x) I;
relate the kth order derivative of f(x) with the kth order divided difference from the
expression for the error term. .
 Newton Form of the Inter;
10.3DIVIDED DIFFERENCES . polating Polyno~nial

Suppose that we have determined a polynomial Pk- (x) of degree I k - 1 which

interpolates f(x) at the points xo, xl,r.. ~ k - ~In. order to make use of Pk - 1 ( ~in) calculating
Pk(x) we consider the following problem: What function g(x) should be added to Pk- 1 ( ~ )
to get Pk (x)? Let g(x) = Pk (x) '- Pk- l ( ~ ) Now,
. g(x) is a polynomial of degree I k and
g ( ~ d = P ~ ( ~ i ) - P ~ - ~ ( ~ i-
) =f(xi)=Ofori=O,
f(~i) 1, .... k-1. .

Suppose that P, (x) is the Lagrange polynomial of degree at most n that agrees with the
functi0n.f at the distinct numbers xo, xl.. ..., x,. Pn (x) can have the following
representation, called Newton form.

for appropriate constants Ao, A;, ....A,.

Evaluating P,(x) (Eqn. (1)) at xo we get A. = Pn (xo) = f(xo). Similarly when Pn(x) is
evaluated at xl, we get Al = - - f ( x l ) . Let us introduce the notation for divided
x1 - xo
differences and define it at this stage: The zeroeth divided difference of the function f,
with respect to xi, is denoted by f[xJ and is simply the evaluation o f f at xi, that is, f[xi] = f
(xi). The first divided difference o f f wilh respect to and + is denoted by f[xi, xi + l]
and defined as

The remaining divided differences of higher orders are defined inductively as follows. The
kth divided differences relative to xi, xi+,, ...,xi+ is defined as

whcre the (k- 1)st divided diifcrences axir...,x ~ +l]~and

- Qxi +1, ..., x,+k]have becn determined.
Thisshows that the kth divided difference is the divided diflercnces of (k- 1)st divided
differences justifying the name. The divided difference f[xl, x2, ..., xk] is invariant under
all permutations of the arguments xl, xz, ...,xk. To show this we proceed as follows giving
another expression for the divided difference.
:
For any integer k between 0 and n. let Qk(x) be the sum of the first k + 1 terms in form (I),
1.e..

Since each of the remaining terms in Eqn. ( I ) has the factor (x - x0) (X- x,) ...(X- xk).
Eqn. (1) can be rewritten as .

Pn(x) = Qk(x) + (x - x0) ,..(X- xk) R(x) lor some polynomial R(x). As the term (x - xo)
(x - xl)...(x - xk) R(x) vanishes at each of the points xo ...xk, we have f(x9 = Pn(x3 = Qk
(xi). i = 0 , 1 , 2 , ..., k. Since Qk(x) is a polynomial of degree 5 k, by uniqueness of
interpolating polynomial Qk(x) = Pk(x).

This shows that P, (x) can be constructed step by step with the addition of the next term in
Eqn. (I), as one constructs the sequence Po(x), Pl(x) ...with Pk(x) obtained from Pk- l(x)
in the form 3

Pk(x) = Pk- l(x) + Ak(x - x0) ... (X- xk- (2)

That is, g(x) is a polynomial of degree Ik having (at least) the k distinct zeros xO,..., xk- l.

...
.: Pk(x) - Pk - l(x) = g(x) = Ak(x - x0) (X- xk - for some constant Ak. This constant
...,
Ak is called h e kth divided difference of f(x) at tile points xO, xk for reasons discussed
f(x) at the p i n t s xo, ..., xk. Thus Eqn. (2) can be rewritten as
P k ( ~ ) = P k - l ( ~+) f[ . X ~ ] ( X - X ~ ) ( X -...,(x-xk-1)
X~) (3)

To get an explicit expression for f[xo. ...,xk] we make use of Lagrange form of
interpolating polynomial and the uniqueness of interpo'lating polynomial.

From Eqn. (3) we have

Pk(x)=Pk-l(~) + f[x0. ...r xk] (x-x0) . . . ( X - X ~ - ~ ) ,

-
since (x xo) (x - x,) ... (x - xk- ,) = xk + a polynomial of degree c k, we can
....
rewrite pk(x) as pk(x) = f[xO, xk] xk + a polynomial of degree < k (4)

(as Pk- 1 ( ~is) a polynomial of degree c k).

But considering the Lagrange form of interpolating polynomial we have

Therefore, on comparison with Eqn. (4) we have

This shows that

YO, ...,ykl = f[x01... xkl

if yo, ...,yk is a reordering of the sequence xo, ..., xk. We have defined the zeroeth divided,
difference of f(x) at xo by f[xo] = f(xo) which is consistent with Eqn. (5).

Fork = 1, we have from Eqn. (5)

This shows that the first divided difference is really a divided difference.

For k = 2, it can be shown (using Eqn. 5) that

This shows that the second divided difference is s dividcd dirfcrcnce of divided
dirfcrcnces.

We show below in Theorem 1 that fork > 2

This shows that the kth divided differcnce is the divided difference of (k - 1)st divided
differences justifying the name. If M = (xO,...., x,) and N denotes anv n - 1 elcmcnts of '
M and the remaining two elements are-%noted by a and P, then -- z
[(n - 1)st divided differen& on N anda - (n - 1)st divided difference on N andpl (7)
( f[x0,..., X" = , .
a-B I
!I Theorem 1:

Proof: Let Pi- l(x) be the polynomial of degree < i - 1 which interpolates f(x) at ~ g ...,
. xi-]
and let Qj- ](x) be the polynomial of degree 5 j - 1 which interpolates f(x) at the points
xl, ...,xj. Let us define P(x) as

This is a polynomial of degree Ij, and P(xi) = f(xi) for i = 0, 1, ....j. By uniqueness of the
I
i interpolating polynomial we have P(x) = Pj(x). Therecore

'
b
Equating the coefficient of xJ from both sides of Eqn. (8). we obtain (leading) coefficient of

leading coefficient of Q - (x) - leading coefficient of 3 - (x)

xj in Pj(x) =
I
xj - xo -
xj xo

We now illustrate this theo.rem with the help of a few examples but beforc that we give the
table of divided differences of various orders.

Table of divided differences

Suppose we denote, for convenience, a first order divided dirference of f(x) with any two
arguments by f[.,.], a second order divided difference with any three arguments by fl.,...]
and so on. Then the table of divided differences can be written as follows

Table I

Example I: If f(x) = x3. find the value of C[a,b.c].

Solution: . f[a,bl = f(b) - f(a) - b3 - a3

b-a b-a
 Similarly,

- (c2 - a2) + b ( c - a)
c-a '

Example 2: If f(x) = -X1 ,show that

Solution: f[a,b] = g-a a - b 1

b-a =ab(b-j=-iii
Similarly,

- -1 + -
1 1 1
f[a,b,cl = ab -
-abbe
c - a c - a

1
- - - /
c-a - abc
Similarly,
1
f[b,c,d] = -
bcd

:. f[a.b,c.dl = I%[
c-a
=-I
abc

-
d-a

= - - 3 .
1
abcd

. In next section we shall make use of the divided differences to derive Newton's genlral
form of interpolating polynomial.

18.3 NEWTON'$ GENERAL FORM OF INTERPOLATING

POLYNOMIAL

In Sec.102 we have shown how P,,(x) can be constructed step by step as one constructs the'
sequence Po(x), P,(x), M... with %(x) obtained from Pk- ,(x) with the addition of the
 next term in iQn.(3), that is, Newton Form of the Intcr-
pointing Polynomial
pk(x) = p k - + (X- x0) (X- xl) ...(X- Xk - 1) f l ~ ~ , . . . , ~ ~ ]
Using this Eqn. (1) can be rewritten as
P,,(x) = f[~Ol+ (x - &,I flx,,x,l + (x - xo) (X - x1) ~ [ x ~ . x , . x ~+I ... +
(x - &,I (x - XI)...(X- Xn - 1) flx0Jl,.-.,xnI. (9)
This can be written compactly as follows :

This is the Newton's form of interpolating polynomial.

Example 3: From the following table of values, find the Newton's form of interpolating
!
polynomial approximating f(x).

I
Solution: We notice that the values of x are not equally spaced. We are required to find a
1 polynomial which approximates f(x). We form tlie table of divided differences of f(x).

Table 2

26 1 13
6 822 132
789
7 1611

Since the divided difference upto order 4 are available, the Newton's interpolating
polynomial P4(x) is given by
P4(x) = f(x0) + (x - xo) f[xo.x11 + (x - xo) (x - x1) flxo.x1.x21 +
-
(X xo) (X--XI) (X- ~ 2f[~0.~1.~2.~31
) f

(X- xo) (X- XI) (X- x2) (X- ~ 3 ~[xO.X~.X~.X~.X.,I

) (1 1)

where xo = - 1, xl = 0, x2 = 3, xj = 6 and x4 = 7.

The divided differences f(xo), f[xo,xll, f[xo,xl,~Zlr

f [ x 0 , ~ l , ~ 2 ~, ~ ~3 ]~ ~ [ X ~ . Xare~ ~ X ~ , X ~ , X ~ ~ ,
those which lie along the diagonal at f(xo) as shown by the dotted line. Substituting the
values of xi and the values of the divided differences in Eqn. (1 1). we get
Interpolation which on simplification gives

We now consider an example to show how Newton's interpolating polynomial can be used
to obtain the approximate value of the function f(x) at any non-tabular point.
Example 4: Find the approximate values of f(x) at x = 2 and x = 5 in Example 3.
Solution: =
Since f(x) P4(x),from Example 3, we get
f(2) =P4(2) = 16 - 24 + 20 - 6 = 6
and
f(5) P(5) = 625 - 375 + 125 - 6 = 369

Note 1: When the values of f(x) for given values of x are required to be found, it is not
necessary to find the interpolating polynomial P4(x) in its simplified form given
above. We can obtain the required values by substituting the values of x in
Eqn.(l 1) itself. Thus,

Similarly.
P4(5) = 3 + (6) ( - 9) + (6) (5) (6) + (6) (5) (2) (5) + (6) (5) (2) ( - 1) (1)
=3-54 + 180 + 300-60~369.
Then f(2) P4(2) = 6
and f(5) =P(5) = 369.

Example 5: Obtain the divided differences interpolation polynomial and the Lagrange's
interpolating polynomial of f(x) from the following data and show that they are same.

Solution: (a) Divided differences interpolation polynomial:

Table 3

X frXI fr... I fr.,... I fr.,..... I

0 - 4 .. .-.--.--.........
5 .....".\
-------...
2 6 .. -......5 ......
--........ ..........
20 1
3 26 9
38
4 64
 Ncwton Form of the Inter-
polating I'olynomial
(b) Lagrange's interpolation polynomial:

On simplifying, we get
P(x) = x3 + x - 4.
Thus, we find that both polynomials are the same.

You may now uy the following exercises:

1
1
El) Find the Lagrange's intcrpolating polynomial of f(x) from b e tablc of valucs given
bclow and show that it is b e same as thc Newton's dividcd diffcrcnccs
interpolating polynomial.

E2) From the tablc of valucs given below, obtain b e value of y when x = 1.5 using

(a) divided differences interpolation formula.

(b) Lagrange's interpolation fonnula.

E3) Using Newton's divided diffcrenccs interpolation formula, find the valucs of f(8)
and f(15) from the following table.

In Unit 9 we have dcrived the gcneral crror tcrrri i.c. thc crror cornrnittcd in approximating
f(x) by P,(x). In the ncxt section wc dcrive anothcr cxprcssion for the crror tcrln in tcrm of
divided difference.

10.4 THE ERROR OF THE INTERPOLATING

POLYNOMIAL
Let P,(x) bc the Newton form of interpolating polynomial of degree 5 n which interpolates
f(x) at xo, ...,x,. The intcrpolating error E,,(x) of P,,(x) is given by
Inlcrpolalion Let X be any point different from xo, ...,x,. If Pn+ l(x) is the Newton form of interpolating
polynomial which interpolales f(x) at xo,...,xn and Y, then P n +l(Z) = f(Y). Then by (10) we
have

Putting x = jl in the above, we have

This shows that the error is like the next term in the Newton form.

10.5 DIVIDED DIFFERENCE AND DERIVATIVE OF THE

FUNCTION
Comparing Eqn.(l3) with the error formula derived in Unit 9 Eqn. (9). we can establish a
relationship between divided differences and the derivatives of the function

f("+') (5)
Comparing, we have QQ, XIS...S%+ 11 =
,( + 1)

Further it can be shown that 5 E ]min xi, max xi[.

We state these results in the following theorem.

Theorem 2: Let f(x) be a real-valued function, defined on '[a.b] and n times differentiable
in ]a. b[. If xo. ...,x, are n + 1 distinct points in [a,b]. then there exists 5 E ]a.b[ such that

Corollary 1:
If f(x) = xu, then

Corollary 2:
If f(x) = xk, k c n, then
~ [ x ~ ~ . . .=~0x ~ ]

since nth derivative of xk, k c n, is zemb.

For example. consider the first divided difference

 By Mean value theorem f(xl) = f(xo) + (xl - xo) f"(y,
xo < 5 < x, . Newton Form of the Inter-
polating Polynomial
Substituting, we get

. = f"(5). X o < 5 < X1.

~[XOIXII

Example 6: If f(x) = a,,xn + a,-lxn" + ... + alx + %,then find f[x~.xl,....xnl.

n!
= a,
Solution: From Corollaries 1 and 2 we have ~[X~,X~....,X,] .-
n! + . O = a,.

Let us consider another example.

Example 7: If f(x) = 2x3 + 3x2 - x + 1, find

f[l.- 1.2.31, f[a,b;c,dl, f[4.6,7.81.
tI Solution: Since f(x) is a cubic polynomial, the 3rd order divided differences of f(x) with
i any set of arguments are constant' and equal to 2, the coefficient of x3 in f(x),
I
I Thus, it follows that f[l, - 1.2.31, f[a,b,c,d], and fl4.6.7.81 are each equal to 2.

You may now try the following exercises:

E4) If f(x) = 2x3 - 3x2 + 7x + 1, what is the value of fl1.2,3,4]?

E5) If f(x) = 3x2 - 2x + 5, find fl1,2], f[2,3] and fl1.2.31.

In the next section, we are going to discuss about bounds on the interpolation error.

10.5 FURTHER RESULTS ON INTERPOLATION ERROR

We have 'derived the error formula

We assume that f(x) is (n + 1) times continuously differentiable in the interval of interest

and x. Since 5(x) is unknown we may replace fi"") (5(x))
[a,b] = I that contains XO,..,,~,,
by ?:\ 1 )r"(" '
I(" 1. -
If we denote (x - no) (x - xl)...(x x 3 by yrn(x) then we have
max 1f cn+') (t) I
E
IE,(x)I=lf(x)-Pn(x)lS (:+I)! m a x l ~ n ( x ) l ' (14)
XE I

Consider now the case when the nodes are equally spaced, that is, xj = xo + jh, j = 0,...,N,
and h is he spacing between consecutive nodes. For the case n=l we have linear
interpolation. If x E [xi- xi], then we approximate f(x) by Pl(x) which interpolates at
1
1 1
xi - I , and xi. From Eqn. (14) we have 4 (x) S 7 max f "(t)( mar (yrl (x)( 1
t€I tE1 .
whcre yrl(x) = (x - xi -.l) (X - xi).
Now,
-
dv1 = x - Xi-]. + X - xi=o
dx

gives x = (xi- + xi)/2.

Hcnce, thc maximum value of I (x - xi- (x - xi) I ocGursat x = x' = (xi- + xi)/2.
 '

The niaximlm value is given by

(xi - xi - -= -
h2
I w1(x*) 1 = 4 4 '

Thus, we have for linear interpolation, for any x E I

For the case n=2, it can be shown that for any x € [xi - I , x i +

h3M where I f
1 E2(x) 1 G 8M (x) 1 P M on I.

Example 8: Determine the spacing h in a table of equally spaced values of the function of
f(x) = J;; between 1 and 2, so that interpolation with a first degree polynomial in this
,

table will yield seven place accuracy.

Solution: Here

'max
11xS2
( f "(x) I= -4I '

and =.
I El (x) 1 1 h
For seven place accuracy, h is to be chosen such that

or h2< (160)10-~that is h < .0013.

E6) If f(x) takes the values - 21, 15,12 and 3 respectively when x assumes the values
- 1, 1.2 and 3, find the polynomial which approximates f(x).
E7) Using the following table of values, find the polynomial which approximates fix).
Hence obtain the value of f(5).

E8) Find the polynamial which approximates f(x), tabulated below

Also find an approximate value of f(x) at x = 1 and x = - 2.

E9) If f(3) = 168, f(7) = 120, f(9) = 72 and f(10) = 63, find an approximate value of
f(6).

E10) The following table gives steam pressures P at different temperatures T, measured
in degrees. Find the pressure at temperature 372.1 degrees.
 Ncnton Form of the Intcr-
polating Polynomial

E l 1) From the following table, find the value oT y when x = 102

-

E12) From the following table of values, obtain the value of y at x = 3

E13) Obtain the polynomial which agrees with the values of f(x) as shown below

E14) ~etemiineUri: spacing h in a uble of equally spaced values of the function f(x) = J;;
between 1 and 2, so that interpolation with a second-dcgrce polynomial in this wble
yields severi-place accuracy. ,

We now end this unit by giving a summary of what we have covered in it.

10.6 SUMMARY
In this unit we have derived a form of interpolating polynomial called Newton's gkneral
form, which has some advantagesover the Lagrange's form discussed in Unit 9. This form
is useful in deriving some other interpolating formulas. Wehave introduced the concept of
divided differences and discussed some of its important properties before deriving
Newton's general form. The error term has also been derived and utilizing the error term
we have established a relationship between the divided difference and the derivative of the
function f(x) for which the interpolating polynomial has been obtained. The main formulas
derived are listed below:

10.7 SOLUTIONS AND ANSWERS

El) x3-x2+3x+8

E2) 26.35156 '

E3) We form the divided differences table -of ftx) below

 Table 4

From the Newton's divided difference interpolation formula, we have

Substituting x = 8 in the above get

substituting x = 15, we get

f(15) = 3150

E4) - 3

E5) 6 5 - 2 . 5 ~]1,2[;611-2,q €1 2,3[,and6

E6) x3 - 9x2 + 17x + 6

E7) x3 - 5x2 + 6x + 1.31

E8) 3x4 - 5x3 + 6x2- 14x + 5, - 5,145

E9) 147

E10) 177.4

Ell) 15.79

E12) 84

E13) x 3 + x 2 - x + 2

E14) ft"(x) = Z3X -' I 2 : hence max 3

~f"'(x)~= 3.
11x12
 Newton Formof the Interpolating
Polynominl

For seven place accuracy, h has to be chosen such that

h3
-
J < 5.10-'. This gives h 0.0128.
24 3

-
The number of interval is N = 2 - 1 = 79.
h