Interpolation and Approximation

Download as pdf or txt
Download as pdf or txt
You are on page 1of 67

CHAPTER

6
FINITE DIFFERENCES
Chapter Objectives
O Introduction
O Finite differences
O Differences of a polynomial
O Factorial notation
O Reciprocal factorial function
O Inverse operator of 
O Effect of an error on a difference table
O Other difference operators
O Relations between the operators
O To find one or more missing terms
O Application to summation of series
O Objective type of questions

6.1 Introduction
The calculus of finite differences deals with the changes that
take place in the value of the function (dependent variable), due to
finite changes in the independent variable. Through this, we also
study the relations that exist between the values assumed by the
function, whenever the independent variable changes by finite
jumps whether equal or unequal. On the other hand, in infinitesi-
mal calculus, we study those changes of the function which occur
234 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

when the independent variable changes continuously in a given interval. In


this chapter, we shall study the variations in the function when the indepen-
dent variable changes by equal intervals.

6.2 Finite Differences


Suppose that the function y  f(x) is tabulated for the equally spaced
values x  x0, x0  h, x0  2h,,x0  nh giving y  y0,y1,y2,,yn.To determine
the values of f(x) or f (x) for some intermediate values of x, the following
three types of differences are found useful:
Forward differences. The differences y1  y0, y2  y1,, yn  yn  1when
denoted by y0, y1,, yn  1 respectively are called the first forward dif-
ferences where  is the forward difference operator. Thus the first forward
differences are yr  yr + 1  yr.
Similarly these second forward differences are defined by 2yr 
yr + 1  yr.
In general, pyr  p1yr+1  p1yr defines the pth forward differences.
These differences are systematically set out in Table 6.1.
In a difference table, x is called the argument and y the function or the
entry. y0, the first entry, is called the leading term and y0, 2y0,3y0 etc. are
called the leading differences.
TABLE 6.1 Forward Difference Table

Value of x Value of y 1st diff. 2nd diff. 3rd diff. 4th diff. 5th diff.
x0 y0
y0
x0  h y1 2y0
y1 3y0
x0  2h y2 2y1 4y0
y2 3y1 5y0
x0  3h y3  y2
2
 y1
4

y3  y2
3

x0  4h y4  y3
2

y4
x0  5h y5
FINITE DIFFERENCES • 235

Obs. 1. Any higher order forward difference can be expressed


NOTE in terms of the entries.
We have D2y0  Dy1  Dy0  (y2  y1)  (y1  y0)  y2  2y1  y0
3y0  2y1  2y0  (y3  2y2  y1)  (y2  2y1  y0)
 y3  3y2  3y1  y0
 y0  3y1  3y0
4

 (y4  3y3  3y2  y1)  (y3  3y2  3y1  y0)


 y4  4y3  6y2  4y1  y0
The coefficients occurring on the right-hand side being the
binomial coefficients, we have in general,
 ny0  yn  nc1 yn 1  nc2 yn  2    (  1)n y0.
Obs.2.The operator  obeys the distributive, commutative, and
index laws
i.e., (i) [f(x)±(x)]  f(x)± (x).
(ii) [cf(x)]  cf(x),c being a constant.
(iii) m n f(x)  m  n f(x), m and n being positive integers. In
view of (i) and (ii),  is a linear operator.
But [f(x).(x)]f(x). (x).
Backward differences. The differences y1  y0, y2  y1,,yn  yn  1 when
denoted by y1, y2,, yn respectively, are called the first backward dif-
ferences where 
TABLE 6.2 Backward Difference Table

Value of x Value of y 1st diff. 2nd diff. 3rd diff. 4th diff. 5th diff.
x0 y0
y1
x0  h y1 2y2
y2 3y3
x0  2h y2 2y3 4y4
y3 3y4 5y5
x0  3h y3 2y4 4y5
y4  y5
3

x0  4h y4  y5 2

y5
x0  5h y5
236 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

is the backward difference operator. Similarly we define higher order


backward differences. Thus we have yr  yr  yr  1, 2yr  yr  yr  1,
3yr  2yr  2r  1, etc.
These differences are exhibited in the Table 6.2.
Central differences. Sometimes it is convenient to employ another sys-
tem of differences known as central differences. In this system, the central
difference operator  is defined by the relations:
y  y  y , y  y  y , , y  y  y
1 0 1/2 2 1 3/2 n n1 n  1/2
Similarly, higher order central differences are defined as
y3/2  y1/2  2y1, y5/2  y3/2  2y2, ,2y2  2y1  3y3/2 and so on.
These differences are shown in Table 6.3.
TABLE 6.3 Central Difference Table

Value of x Value of y 1st diff. 2nd diff. 3rd diff. 4th diff. 5th diff.
x0 y0
y1/2
x0  h y1 2y1
y3/2 3y3/2
x0  2h y2 2y2 4y2
y5/2 3y5/2 5y5/2
x0  3h y3  y3
2
 y3
4

y7/2
x0  4h y4 2y4 3y7/2
y9/2
x0  5h y5

We see from this table that the central differences on the same horizon-
tal line have the same suffix. Also the differences of odd order are known
only for half values of the suffix and those of even order for only integral
values of the suffix.
It is often required to find the mean of adjacent values in the same col-
umn of differences. We denote this mean by .
1 1
Thus y1  (y1/2  y3/2 ), 2 y3/2  (2 y1  2 y2 ), etc.
2 2
Obs. The reader should note that it is only the notation which
NOTE
changes and not the differences. e.g.
252 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

10. Write down the polynomial of lowest degree which satisfies the fol-
lowing set of numbers: 0, 7, 26, 63, 124, 215, 342, 511

6.8 Other Difference Operators


We have already introduced the operators , , and . Besides these,
there are the operators E and μ, which we define below:
Shift operator E is the operation of increasing the argument x by h so that
E f(x)  f(x  h), E2f(x)  f(x  2h), E3f(x)  f(x  3h) etc.
The inverse operator E  1is defined by E  1f(x)  f(x  h)
If yx is the function f(x), then Eyx  yx  h,E  1yx  yx  h,Enyx  yx + nh, where
n may be any real number.
1
2
Averaging operator µis defined by the equation yx  yx1 h  yx1 h .
2 2

Obs. In the difference calculus E is regarded as the fundamen-
NOTE tal operator and ,, , µ can be expressed in terms of E.

6.9 Relations Between the Operators


We shall now establish the following identities:
(i)   E  1 (ii)   1  E1
1
(iii)   E1/2  E1/2 (iv)  (E1/2  E–1/2 )
2
(v)   E  E  E1/2 (vi) E  ehD.
Proofs.(i)yx  yx + h  yx  Eyx  yx  (E  1)yx
This shows that the operators  and E are connected by the sym-
bolic relation
E1 or E1.

Obs. These relations imply that the effect of operator E on yx is


NOTE the same as that of the operators (1  ) on yx.The operator’s E
and  do not have any existence as separate entities.

(ii) yx  yx  yx  h  yx  E  1yx  (1  E  1)yx


   1  E1
FINITE DIFFERENCES • 253

(iii) yx  yx1 h  yx1 h  E yx  E yx   E  E  yx


1/2 1/2 1/2 1/2

2 2
1/2 1/2
 E E

   
1 1  1
1

1 1 1


(iv) yx  y 1  y 1  E 2 yx  E 2 yx  E 2  E 2 yx
2 x2 h x h
2
2 2
1 1/2
   E  E1/2 
2
(v) Eyx  E(yx  yx  h)  Eyx  Eyx  h  yx + h  yx  yx
∴ E
Eyx  yx + h  yx + h  yx  yx
 E  
1 1 1
E1/2 yx  y 1 y 1  h  y 1  h  h yxh – yx  yx
x h
2
x h
2
2 x h
2
2 2
 E1/2  
Hence   E  E  E1/2
h2
(vi) Ef ( x)  f ( x  h)  f ( x)  hf ( x)  f ( x) 
2!
[by Taylor’s series]
h2 2
 f ( x)  hDf  x   D f  x   ....
2!
 h 2 D2 h 3 D3 
1  hD    f ( x)  ehD f ( x)
 2! 3! 
 E  ehD
hD
Cor.  E  1    e

A table showing the symbolic relations between the various


NOTE operators is given below for ready reference To prove such rela-
tions between the operators, always express each operator in
terms of the fundamental operator E.
CHAPTER

7
INTERPOLATION
Chapter Objectives
O Introduction
O Newton’s forward interpolation formula
O Newton’s backward interpolation formula
O Central difference interpolation formulae
O Gauss’s forward interpolation formula
O Gauss’s backward interpolation formula
O Stirling’s formula
O Bessel’s formula
O Everett’s formula
O Choice of an interpolation formula
O Interpolation with unequal intervals
O Lagrange’s interpolation formula
O Divided differences
O Newton’s divided difference formula
O Relation between divided and forward differences
O Hermite’s interpolation formula
O Spline interpolation—Cubic spline
O Double interpolation
O Inverse interpolation
O Lagrange’s method
274 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

O Iterative method
O Objective type of questions

7.1 Introduction
Suppose we are given the following values of y  f(x) for a set of values
of x:
x: x0 x1 x2  xn
y: Y0 y1 y y.
2 n

Then the process of finding the value of y corresponding to any value of


x  xi between x0 and xn is called interpolation. Thus interpolation is the
technique of estimating the value of a function for any intermediate value
of the independent variable while the process of computing the value of the
function outside the given range is called extrapolation. The term interpola-
tion however, is taken to include extrapolation.
If the function f(x) is known explicitly, then the value of y correspond-
ing to any value of x can easily be found. Conversely, if the form of f(x) is not
known (as is the case in most of the applications), it is very difficult to de-
termine the exact form of f(x) with the help of tabulated set of values (xi, yi).
In such cases, f(x) is replaced by a simpler function (x) which assumes the
same values as those of f(x) at the tabulated set of points. Any other value
may be calculated from (x) which is known as the interpolating function or
smoothing function. If (x) is a polynomial, then it called the interpolating
polynomial and the process is called the polynomial interpolation. Similarly
when (x) is a finite trigonometric series, we have trigonometric interpola-
tion. But we shall confine ourselves to polynomial interpolation only.
The study of interpolation is based on the calculus of finite differences.
We begin by deriving two important interpolation formulae by means of
forward and backward differences of a function. These formulae are often
employed in engineering and scientific investigations.

7.2 Newton’s Forward Interpolation Formula


Let the function y  f(x) take the values y0, y1, , yn corresponding to
the values x0, x1, , xn of x. Let these values of x be equispaced such that
xi  x0  ih (i  0, 1, ). Assuming y(x) to be a polynomial of the nth degree
in x such that y( x0 )  y0 , y( x1 )  y1 ,, y( xn )  yn . We can write
INTERPOLATION • 275

y( x)  a0  a1 ( x – x0 )  a2 ( x – x0 )( x – x1 )  a3 ( x – x0 )( x – x1 )( x – x2 )
 an ( x – x0 )( x – x1 )( x – xn–1 ) (1)
Putting x  x0, x1, , xn successively in (1), we get
y0  a0 , y1  a0  a1 ( x1 – x0 ), y2  a0  a1 ( x2 – x0 )  a2 ( x2 – x0 )( x2 – x1 )
and so on.
From these, we find that a0  y0 , y0  y1 – y0  a1 ( x1 – x0 )  a1 h
1
 a1  y0
h
Also y1  y2  y1  a1 ( x2  x1 )  a2 ( x2  x0 )( x2  x1 )
 a1 h  a2 hh  y0  2h2 a2
1
 a2  2
 y1  y0   1 2 2 y0
2h 2! h
1
Similarly a3  3 y0 and so on.
3! h3
Substituting these values in (1), we obtain
y0 2 y0 3 y0
y( x)  y0  ( x  x0 )  ( x  x0 )( x  x1 )  ( x  x0 )( x  x1 )( x  x2 ) 
h 2! h2 3! h3
(2)
Now if it is required to evaluate y for x x0  ph, then
( x  x0)  ph, x  x1  x  x0 ( x  x0)  ph  h ( p 1) h,
( x  x0 )  x  x0  ( x  x0 )  (p 1)h  h  (p  2)h
etc.
Hence, writting y(x) = y(x0 + ph) = yp, (2) becomes
p(p 1) 2 p(p 1)(p  2) 3
yp  y0  py0   y0   y0
2! 3!
p(p 1) p  n - 1
   n y0 (3)
3!
It is called Newton’s forward interpolation formula as (3) contains y0
and the forward differences of y0
Otherwise: Let the function y  f(x) take the values y0, y1, y2, corre-
sponding to the values x0, x0  h, x0  2h,  of x. Suppose it is required to
evaluate f(x) for x  x0  ph, where p is any real number.
276 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

For any real number p, we have defined E such that


E p f ( x)  f ( x  ph)

yp  f ( x0  ph)  E p f ( x0 )  (1  )p y0 [ E  1  ]
 p(p 1) 2 p(p 1)(p  2) 3 
 1  p    y0  y0 (4)
 2! 3! 
[Using binomial theorem]
p(p 1) 2 p(p 1)(p  2) 3
i.e., yp  y0  py0   y0   y0 
2! 3!
If y  f(x) is a polynomial of the nth degree, then n1y0 and higher dif-
ferences will be zero.
Hence (4) will become
p(p 1) 2 p(p 1)(p  2) 3
yp  y0  py0   y0   y0 
2! 3!
p(p 1) p  n 1 n
  y0
3!
Which is same as (3)

Obs. 1. This formula is used for interpolating the values of y


NOTE near the beginning of a set of tabulated values and extrapolating
values of y a little backward (i.e., to the left) of y0.
Obs. 2. The first two terms of this formula give the linear inter-
polation while the first three terms give a parabolic interpola-
tion and so on.

7.3 Newton’s Backward Interpolation Formula


Let the function y  f(x) take the values y0, y1, y2,  corresponding to
the values x0, x0  h, x0  2h,  of x. Suppose it is required to evaluate f(x)
for x  xn  ph, where p is any real number. Then we have
yp  f(xn  ph)  Ep f(xn)  (1 – )-p yn [ E1  1 – ]
 p(p 1) 2 p(p 1)(p  2) 3 
1  p    y0 yn
 2! 3! 
[using binomial theorem]
INTERPOLATION • 277

p(p 1) 2 p(p 1)(p  2) 3


i.e., yp  yn  pyn   yn   yn  (1)
2! 3!
It is called Newton’s backward interpolation formula as (1) contains yn
and backward differences of yn

Obs. This formula is used for interpolating the values of y near


NOTE the end of a set of tabulated values and also for extrapolating
values of y a little ahead (to the right) of yn

EXAMPLE 7.1
The table gives the distance in nautical miles of the visible horizon for
the given heights in feet above the earth’s surface:
x  height: 100 150 200 250 300 350 400
y  distance: 10.63 13.03 15.04 16.81 18.42 19.90 21.27
Find the values of y when
(i) x  160 ft. (ii) x  410.
Solution:
The difference table is as under:
x y  2 3 4
100 10.63
2.40
150 13.03 – 0.39
2.01 0.15
200 15.04 – 0.24 – 0.07
1.77 0.08
250 16.81 – 0.16 – 0.05
1.61 0.03
300 18.42 – 0.13 – 0.01
1.48 0.02
350 19.90 – 0.11
1.37
400 21.27
(i) If we take x0  160, then y0 13.03, y0  2.01, 2y0  – 0.24,
  0.08, 4 y0  – 0.05
3
278 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

x  x0 10
Since x  160 and h  50,  p   0.2
h 50
 Using Newton’s forward interpolation formula, we get
p(p 1) 2 p(p 1)(p  2) 3
y218  yp  y0  py0   y0   y0
2! 3!
p(p 1)(p  2)(p  3) 4
  y0 
4!
y160  13.03  0.402  0.192  0.0384  0.00168  13.46 nautical miles
(ii) Since x  410 is near the end of the table, we use Newton’s back-
ward interpolation formula.
x  xn 10
 Taking xn  400, p    0.2
h 50
Using the line of backward difference
yn 21.27,  yn 1.37, 2 yn – 0.11, 3 yn 0.02 etc.
 Newton’s backward formula gives
p  p  1 2
y410  y400  py400   y400
2!
p(p 1)(p  2) 3 p(p 1)(p  2)(p  3) 4
  y400   y400 
3! 4!
0.2 1.2 
 21.27  0.2 1.37   0.11
2!
0.2 1.2 2.2  0.2 1.2 2.2 3.2 
 0.02   0.01
3! 4!
 21.27  0.274  0.0132  0.0018  0.0007
 21.53 nautical miles

EXAMPLE 7.2
From the following table, estimate the number of students who ob-
tained marks between 40 and 45:
Marks: 30—40 40—50 50—60 60—70 70—80
No. of students: 31 42 51 35 31
INTERPOLATION • 279

Solution:
First we prepare the cumulative frequency table, as follows:
Marks less than (x): 40 50 60 70 80
No. of students (yx): 31 73 124 159 190
Now the difference table is
x yx  yx 2 yx 3yx 4yx
40 31
42
50 73 9
51 – 25
60 124 – 16 37
35 12
70 159 –4
31
80 190
We shall find y45, i.e., the number of students with marks less than 45.
Taking x0 40, x  45, we have
x  x0 5
p   0.5 [ h  10]
h 10
 Using Newton’s forward interpolation formula, we get
p(p 1) 2 p(p 1)(p  2) 3
y45  y40  py40   y40   y40
2! 3!
p(p 1)(p  2)(p  3) 4
  y40
4!
0.50.5 0.50.515
  31  0.5  42  9  25
2 6
0.50.5152.5
  37
24
 31  21 – 1.125 – 1.5625 – 1.4453
 47.87, on simplification.
The number of students with marks less than 45 is 47.87, i.e., 48. But
the number of students with marks less than 40 is 31.
Hence the number of students getting marks between 40 and 45 
48 – 31  17.
280 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

EXAMPLE 7.3.
Find the cubic polynomial which takes the following values:
x: 0 1 2 3
f(x): 1 2 1 10
Hence or otherwise evaluate f(4).
Solution:
The difference table is
x f(x) f(x) 2f(x) 3f(x)
0 1
1
1 2 –2
–1 12
2 1 10
9
3 10
x0
We take x0  0 and p x [ h  1]
h
 Using Newton’s forward interpolation formula, we get
x x( x 1) 2 x( x 1)( x  2) 3
f ( x)  f (0)  f (0)   f (0)   f (0)
1 1.2 1.2.3
x( x 1) x( x 1)( x  2)
 1  x 1  2   12 
2 6
 2 x3  7 x2  6 x 1
which is the required polynomial.
x  xn
To compute f(4), we take xn  3, x  4 so that p  1 [ h  1]
h

NOTE Obs. Using Newton’s backward interpolation formula, we get

p(p 1) 2 p(p 1) p  2  3


f (4)  f (3)  pf (3)   f (3)   f (3)
1.2 1.2.3
= 10 + 9 + 10 + 12 = 41
INTERPOLATION • 281

which is the same value as that obtained by substituting x  4 in


the cubic polynomial above.
The above example shows that if a tabulated function is a
polynomial, then interpolation and extrapolation give the same
values.

EXAMPLE 7.4
Using Newton’s backward difference formula, construct an interpolat-
ing polynomial of degree 3 for the data: f (– 0.75)  – 0.0718125, f (– 0.5)
 – 0.02475, f (– 0.25)  0.3349375, f (0)  1.10100. Hence find f (– 1/3).
Solution:
The difference table is
x y y 2y 3y
– 0.75 – 0.0718125
0.0470625
– 0.50 – 0.02475 0.312625
0.3596875 0.09375
– 0.25 0.3349375 0.400375
0.7660625
0 1.10100
We use Newton’s backward difference formula
p p(p 1) 2 p(p 1)(p  2) 3
y( x)  y3  y3   y3   y3
1! 2! 3!
x0 x  h  0.25
taking x3  0, p    4x
h 0.25
4 x(4 x 1)
y  x   1.10100  4 x(0.7660625) 0.400375
2
4 x(4 x 1)(4 x  2)
 (0.09375)
6
 1.101  3.06425 x  3.251 x2  0.81275 x  x3  0.75 x2  0.125 x
 x3  4.001 x2  4.002 x 1.101

1
Put x  , so that
3
282 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

 1  1  1 2  1
y  34.001   4.002 1.101
 3  3  3  3
 0.1745
EXAMPLE 7.5
In the table below, the values of y are consecutive terms of a series of
which 23.6 is the 6th term. Find the first and tenth terms of the series:
x: 3 4 5 6 7 8 9
y: 4.8 8.4 14.5 23.6 36.2 52.8 73.9

Solution:
The difference table is
x y y 2y 3y 4y
3 4.8
3.6
4 8.4 2.5
6.1 0.5
5 14.5 3.0 0
9.1 0.5
6 23.6 3.5 0
12.6 0.5
7 36.2 4.0 0
16.6 0.5
8 52.8 4.5
21.1
9 73.9
To find the first term, use Newton’s forward interpolation formula with
x0  3, x  1, h  1, and p  – 2. We have
2  2 3 2 34 
y(1)  4.8   3.6   2.5   0.5  3.1
1 1.2 1.2.3
To obtain the tenth term, u se Newton’s backward interpolation for-
mula with xn 9, x  10, h  1, and p  1.This gives
1 1 2 1(2) 3
y 10   73.9   21.1   4.5   0.5  100
1 1.2 1.2.3
INTERPOLATION • 283

EXAMPLE 7.6
Using Newton’s forward interpolation formula show

 
2
n(n 1)
 n3 
2
Solution:
If sn  sn3, then sn1  (n  1)3
 sn  sn1  sn    n  1   n3 3
 n  1
3

3 3
Then 2 sn  sn1  sn   n  2    n  1  3n2  9 n  7

3 sn  2 sn1  2 sn

   3 n  9 n  7   6 n  12
 3  n  1  9  n  1  7 
2 2

 4 sn  3 sn1  3 sn   6  n  1  12   6 n  12   6
and  5 sn   5 sn  ......  0
Since the first term of the given series is 1, therefore taking n  1, s1 1,
 s1  8, 2 s1  19, 3s1  18, 4 s1 6.
Substituting these in the Newton’s for war d interpolation formula, i.e.,
(n 1)(n 2) 2 (n 1)(n  2)(n  3) 3
s  s(n – 1)s1   s1   s1
2! 3!
(n  1)(n  2)(n  3)(n  4) 4
  s1
4!
19
sn 1  8(n – 1)  (n – 1)(n – 2)  3(n – 1)(n – 2)(n – 3)
2
1 1 4 3 2 n(n  1)2
 (n – 1)(n – 2)(n – 3)(n – 4)  (n  2 n  n )   
4 4  2 

Exercises 7.1

1. Using Newton ’s forward formula, fin d the value of f(1.6), if


x: 1 1.4 1.8 2.2
f(x): 3.49 4.82 5.96 6.5
2. From the following table find y when x  1.85 an d 2.4 by Newton’s inter-
polation formula:
x: 1.7 1.8 1.9 2.0 2.1 2.2 2.3
ye:x
5.474 6.050 6.686 7.389 8.166 9.025 9.974
284 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

3. Express the value of  in terms of x using the following data:


x: 40 50 60 70 80 90
: 184 204 226 250 276 304
Also find  at x  43.
4. Given sin 45°  0.7071, sin 50°  0.7660, sin 55°  0.8192,
sin 60°  0.8660, find sin 52° using Newton’s forward formula.
5. From the following table:
x: 0.1 0.2 0.3 0.4 0.5 0.6
f(x): 2.68 3.04 3.38 3.68 3.96 4.21
find f(0.7) approximately.
6. The area A of a circle of diameter d is given for the following values:
d: 80 85 90 95 100
A: 5026 5674 6362 7088 7854
Calculate the area of a circle of diameter 105
7. From the following table:
x°: 10 20 30 40 50 60 70 80
cos x: 0.9848 0.9397 0.8660 0.7660 0.6428 0.5000 0.3420 0.1737
Calculate cos 25° and cos 73° using the Gregory-1 Newton formula.
8. A test performed on a NPN transistor gives the following result:
Base current f (mA) 0 0.01 0.02 0.03 0.04 0.05
Collector current IC (mA) 0 1.2 2.5 3.6 4.3 5.34
Calculate (i) the value of the collector current for the base current of
0.005 mA.
(ii) the value of base current required for a collector correct of 4.0 mA.
9. Find f(22) from the following data using Newton’s backward formulae.
x: 20 25 30 35 40 45
f(x): 354 332 291 260 231 204
10. Find the number of men getting wages between Rs. 10 and 15 from the
following data:
Wages in Rs: 0—10 10—20 20—30 30—40
Frequency: 9 30 35 42
INTERPOLATION • 285

11. From the following data, estimate the number of persons having in-
comes between 2000 and 2500:
Income Below 500 500–1000 1000–2000 2000–3000 3000–4000
No. of persons 6000 4250 3600 1500 650
12. Construct Newton’s forward interpolation polynomial for the following
data:
x: 4 6 8 10
y: 1 3 8 16
Hence evaluate y for x  5.

13. Find the cubic polynomial which takes the following values:
y(0)  1, y(1)  0, y(2)  1 and y(3)  10.
Hence or otherwise, obtain y(4).

14. Construct the difference table for the following data:


x: 0.1 0.3 0.5 0.7 0.9 1.1 1.3
f (x): 0.003 0.067 0.148 0.248 0.370 0.518 0.697

Evaluate f (0.6)

15. Apply Newton’s backward difference formula to the data below, to ob-
tain a polynomial of degree 4 in x:
x: 1 2 3 4 5
y: 1 –1 1 –1 1
16. The following table gives the population of a town during the last six
censuses. Estimate the increase in the population during the period
from 1976 to 1978:
Year: 1941 1951 1961 1971 1981 1991
Population: 12 15 20 27 39 52
(in thousands)
17. In the following table, the values of y are consecutive terms of a series of
which 12.5 is the fifth term. Find the first and tenth terms of the series.
x: 3 4 5 6 7 8 9
y: 2.7 6.4 12.5 21.6 34.3 51.2 72.9
286 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

18. Using a polynomial of the third degree, complete the record given be-
low of the export of a certain commodity during five years:
Year: 1989 1990 1991 1992 1993
Export: 443 384 — 397 467
(in tons)
19. Given u1 40, u3  45, u5  54, find u2 and u4.
20. If u1  10, u1  8, u2  10,u4  50, find u0 and u3.
21. Given y0  3, y1  12, y2  81, y3  200, y4  100, y5  8, without form-
ing the difference table, find 5y0.

7.4 Central Difference Interpolation Formulae


In the preceding sections, we derived Newton’s forward and backward
interpolation formulae which are applicable for interpolation near the be-
ginning and end of tabulated values. Now we shall develop central differ-
ence formulae which are best suited for interpolation near the middle of
the table.
If x takes the values x0 – 2h, x0 – h, x0, x0  h, x0  2h and the correspond-
ing values of y  f(x) are y–2, y–1, y0, y1, y2, then we can write the difference
table in the two notations as follows:
x y 1st diff. 2nd diff. 3rd diff. 4th diff.
x0 – 2h y2
y2 ( y3/2)
x0 – h y1 2y2( 2y1)
y1 ( y1/2) 3y2 ( 3y1/2)
x0 y0 2y1 ( 2y0) 3y2( 4y0)
y0 ( y1/2) 3y1 ( 3y1/2)
x0  h y1 2y0( 2y1)
y1( y3/2)
x0  2h y2
306 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

7.11 Interpolation with Unequal Intervals


The various interpolation formulae derived so far possess the disadvan-
tage of being applicable only to equally spaced values of the argument. It is,
therefore, desirable to develop interpolation formulae for unequally spaced
values of x. Now we shall study two such formulae:
(i) Lagrange’s interpolation formula
(ii) Newton’s general interpolation formula with divided differences.

7.12 Lagrange’s Interpolation Formula


If y  f(x) takes the value y0, y1,......, yn corresponding to x  x0,x1,, xn,
then
( x  x1 )( x  x2 )( x  xn ) ( x  x0 )( x  x2 )( x  xn )
f ( x)  y0  y
( x0  x1 )( x0  x2 )( x0  xn ) ( x1  x0 )( x1  x2 )( x1  xn ) 1
(1)
( x  x0 )( x  x1 )( x  xn1 )
  y
( xn  x0 )( xn  x1 )( xn  xn1 ) n
This is known as Lagrange’s interpolation formula for unequal inter-
vals.
Proof: Let y  f(x) be a function which takes the values (x0, y0), (x1, y1),,
(xn, yn). Since there are n  1 pairs of values of x and y, we can represent f(x)
by a polynomial in x of degree n. Let this polynomial be of the form
y  f ( x)  a0 ( x – x1 )( x – x2 )( x – xn )  a1 ( x – x0 )( x – x2 )( x – xn )

a2 ( x – x0 )( x – x1 )( x – x 3 )( x – xn )   an ( x – x0 )( x – x1 )( x – xn–1 ) (2)


Putting x  x0, y  y0, in (2), we get
y0  a0 ( x0 – x1 )( x – x2 )( x – xn )
a0  y0 / [( x – x1 )( x – x2 )( x – xn )]
Similarly putting x  x1, y  y1 in (2), we have
a1  y1 / [( x1 – x0 )( x1 – x2 )( x1 – xn )]
Proceeding the same way, we find a2, a3...... an.
Substituting the values of a , a ,, a in (2), we get (1)
0 1 n
INTERPOLATION • 307

Obs. Lagrange’s interpolation formula (1) for n points is a


NOTE
polynomial of degree (n – 1) which is known as the Lagrangian
polynomial and is very simple to implement on a computer.
This formula can also be used to split the given function into
partial fractions.
For on dividing both sides of (1) by ( x – x0 )( x – x1 )( x – xn ), we
get
f ( x) y0 1
 
( x  x0 )( x  x1 )( x  xn ) ( x0  x1 )( x0  x2 )( x0  xn ) ( x  x0 )
y1 1
 . 
( x1  x0 )( x1  x2 )( x1  xn ) ( x  x1 )
yn 1
 .
( xn  x0 )( xn  x1 )( xn  xn1 ) ( x  xn )

EXAMPLE 7.17
Given the values
x: 5 7 11 13 17
f(x): 150 392 1452 2366 5202
evaluate f(9), using Lagrange’s formula
Solution:
(i) Here x0  5, x1  7, x2  11, x3 13, x4  17
and y0  150, y1  392, y2  1452, y3  2366, y4  5202.
Putting x  9 and substituting the above values in Lagrange’s formula,
we get
(9  7)(9  11)(9  13)(9  17) (9  5)(9  11)(9  13)(9  17)
f (9)  150   392
(5  7)(5  11)(5  13)(5  17) (7  5)(7  11)(7  13)(7  17)
(9  5)(9  7)(9  13)(9  17)
 1452
(11  5)(11  7)(11  13)(11  17)
(9  5)(9  7)(9  11)(9  17)
  2366
(13  5)(13  7)(13  11)(13  17)
(9  5)(9  7)(9  11)(9  13)
  5202
(17  5)(17  7)(17  11)(17  13)
50 3136 3872 2366 578
      810
3 15 3 3 5
308 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

EXAMPLE 7.18
Find the polynomial f (x) by using Lagrange’s formula and hence find
f(3) for
x: 0 1 2 5
f (x): 2 3 12 147
Solution:
Here x0  0, x1  1, x2  2, x35
and y0  2, y1  3, y2  12, y3147.
Lagrange’s formula is

( x  x1 )( x  x2 )( x  x3 ) ( x  x0 )( x  x2 )( x  x3 )
y y0  y
( x0  x1 )( x0 – x2 )( x0  x3 ) ( x1 – x0 )( x1 – x2 )( x1  x3 ) 1
( x  x0 )( x  x1 )( x  x3 ) ( x  x0 )( x  x1 )( x  x2 )
 y  y
( x2 – x0 )( x2 – x1 )( x2  x3 ) 2 ( x3 – x0 )( x3 – x1 )( x3  x2 ) 3
( x  1)( x  2)( x  5) ( x  0)( x  2)( x  5)
 (2)  (3)
(0  1)(0  2)(0  5) (1  0)(1  2)(1  5)
( x  0)( x  1)( x  5) ( x  0)( x  1)( x  2)
 12   (147)
(2  0)(2  1)(2  5) (5  0)(5  1)(5  2)
3 2
Hence f ( x)  x  x – x  2
 f (3)  27  9 – 3  2  35

EXAMPLE 7.19
A curve passes through the points (0, 18), (1, 10), (3, –18) and (6, 90).
Find the slope of the curve at x  2.
Solution:
Here x0  0, x1  1, x2  3, x3  6 and y0  18, y1  10, y2  –18, y3  90.
Since the values of x are unequally spaced, we use the Lagrange’s for-
mula:
( x  x1 )( x  x2 )( x  x3 ) ( x  x0 )( x  x2 )( x  x3 )
y y0  y
( x0 x1 )( x0 x2 )( x0 x3 ) ( x1  x0 )( x1  x2 )( x1  x3 ) 1
( x  x0 )( x  x1 )( x  x3 ) ( x  x0 )( x  x1 )( x  x2 )
 y2  y
( x2 – x0 )( x2 – x1 )( x2  x3 ) ( x3 – x0 )( x3 – x1 )( x3  x2 ) 3
INTERPOLATION • 309

( x  1)( x  3)( x  6) ( x  0)( x  3)( x  6)


 18   10 
(0  1)(0  3)(0  6) (1  0)(1  3)(1  6)
( x  0)( x  1)( x  6) ( x  0)( x  1)( x  3)
 18   (90)
(3  0)(3  1)(3  6) (6  0)(6  1)(6  3)
 (– x3  10 x2 – 27 x  18)  ( x3 – 9 x2  18 x)
( x3 – 7 x2  6 x)  ( x3 – 4 x2  3 x)
i.e., y  2 x3 – 10 x2  18
 dy 
Thus the slope of the curve at x  2  
 dx x2
 (6 x2  20 x) x2 16

EXAMPLE 7.20
3 x2  x  1
Using Lagrange’s formula, express the function as
a sum of partial fractions.  x  1 x  2  x  3

Solution:
Let us evaluate y  3x2  x  1 for x  1, x  2 and x  3
These values are
x: x0  1 x1  2 x2  3
y: y0  5 y1  15 y2  31
Lagrange’s formula is
( x  x1 )( x  x2 ) ( x  x0 )( x  x2 ) ( x  x0 )( x  x1 )
y y0  y1  y
( x0 x1 )( x0 x2 ) ( x1  x0 )( x1  x2 ) ( x2 – x0 )( x2 – x1 ) 2

Substituting the above values, we get


( x  2)( x  3) ( x  1)( x  3) ( x  1)( x  2)
y  5  15  31
(1  2)(1  3) (2  1)(2  3) (3  1)(3  2)
2.5 (x – 2) (x – 3) – 15 (x – 1) (x – 3)  15.5 (x – 1) (x – 2)
2.5 ( x  2) ( x  3)  15( x  1) ( x  3) 
2
Thus 3 x  x 1 15.5 ( x  1) ( x  2)

( x  1)( x  2)( x  3) ( x  1)( x  2)( x  3)
25 15 15.5
  
x 1 x  2 x  3
310 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

EXAMPLE 7.21
Find the missing term in the following table using interpolation:
x: 0 1 2 3 4
y: 1 3 9 ... 81
Solution:
Since the given data is unevenly spaced, therefore we use Lagrange’s
interpolation formula:
( x  x1 )( x  x2 )( x  x3 ) ( x  x0 )( x  x2 )( x  x3 )
y y0  y
( x0 x1 )( x0 x2 )( x0 x3 ) ( x1  x0 )( x1  x2 )( x1  x3 ) 1
( x  x0 )( x  x1 )( x  x3 ) ( x  x0 )( x  x1 )( x  x2 )
 y2  y
( x2 – x0 )( x2 – x1 )( x2  x3 ) ( x3 – x0 )( x3 – x1 )( x3  x2 ) 3

Here we have x0  0 x1  1 x2  2 x3  4
y0  1 y1  3 y2  9 y3  81
( x  1)( x  2)( x  4) ( x  0)( x  2)( x  4)
 y 1   3
(0  1)(0  2)(0  4) (1  0)(1  2)(1  4)
( x  0)( x  1)( x  4) ( x  0)( x  1)( x  2)
 9  (81)
(2  0)(2  1)(2  4) (4  0)(4  1)(4  2)
When x  3, then
(3  1)(3  2)(3  4) 3(3  1)(3  4) 9 
 y  3(3  2)(3  4)  
8 4
3(3 1)(3  2) 1 27 81
 (81)   3    31
24 4 2 24
Hence the missing term for x  3 is y  31.

EXAMPLE 7.22
Find the distance moved by a particle and its acceleration at the end of
4 seconds, if the time verses velocity data is as follows:
t: 0 1 3 4
v: 21 15 12 10
INTERPOLATION • 311

Solution:
Since the values of t are not equispaced, we use Lagrange’s formula:
(t  t1 )(t  t2 )(t  t3 ) (t  t0 )(t  t2 )(t  t3 )
v v  v
(t0  t1 )(t0  t2 )(t0  t3 ) 0 (t1  t0 )(t1  t2 )(t1  t3 ) 1
(t  t0 )(t  t1 )(t  t3 ) (t  t0 )(t  t1 )(t  t3 )
 v2  v
(t1  t0 )(t1  t2 )(t1  t3 ) (t1  t0 )(t1  t2 )(t1  t3 ) 3
(t  1)(t  3)(t  4) t(t  3)(t  4)
i.e. , v  (21) (15)
(1)(2)(4) (1)(2)(3)
t(t  1)(t  4) t(t  1)(t  3)
 (12) (10)
(3)(2)(1) (4)(3)(1)
1
i.e., v  (–5t 3  38 t 2 – 105t  252)
12
4 4  ds 
 Distance moved s  0 vdt  0 (–5t 3  38 t 2 – 105t  252)  v  dt 


1  5t 4 38 t 3 105t 2 4
 –  –  252 t 
12 4 3 2 0
1 2432 
  –320  – 840  1008  54.9
12 3 
dv 1
Also acceleration   15t 2  76 t 105  0 
dt 2
1
Hence acceleration at (t  4)  15 76  4   105 3.4
2

Exercises 7.3

1. Use Lagrange’s interpolation formula to find the value of y when x  10,


if the following values of x and y are given:
x: 5 6 9 11
y: 12 13 14 16
2. The following table gives the viscosity of oil as a function of tempera-
ture. Use Lagrange’s formula to find the viscosity of oil at a temperature
of 140°.
Temp°: 110 130 160 190
Viscosity: 10.8 8.1 5.5 4.8
3. Given log10654  2.8156, log10 658  2.8182, log10 659  2.8189,
log10661  2.8202, find by using Lagrange’s formula, the value of
log10 656.
4. The following are the measurements T made on a curve recorded by
oscilograph representing a change of current I due to a change in the
conditions of an electric current.
T: 1.2 2.0 2.5 3.0
I: 1.36 0.58 0.34 0.20
Using Lagrange’s formula, find I and T  1.6.
5. Using Lagrange’s interpolation, calculate the profit in the year 2000
from the following data:
Year: 1997 1999 2001 2002
Profit in Lakhs of Rs: 43 65 159 248
6. Use Lagrange’s formula to find thee form of f(x), given
x: 0 2 3 6
f(x): 648 704 729 792
7. If y(1)  – 3, y(3)  9, y(4)  30, y(6)  132, fin d the Lagrange’s interpola-
tion polynomial that takes the same values as y at the given point s.
8. Given f(0)  – 18, f(1)  0, f(3)  0, f(5)  – 248, f(6)  0, f(9)  13104,
find f(x).
9. Find the missing term in the following table using interpolation
x: 1 2 4 5 6
y: 14 15 5 … 9

10. Using Lagrange’s formula, express the function x2  x  3 as a sum


of partial fractions. x  2 x2  x  2
3

11. Using Lagrange’s formula, express the function x2  6 x  1 as


 x2 1  x  4 x  6
a sum of partial fractions.
[Hint. Tabulate the values of f(x)  x2  6x – 1 for x  – 1, 1, 4, 6 and apply
Lagrange’s formula.]
12. Using Lagrange’s formula, prove that

yo 
1
2 8 2

 y1  y1   1 1  y3  y1   1  y1  y3  .
2

[Hint: Here x0  – 3, x1  – 1, x2  1, x3  3.]
INTERPOLATION • 313

7.13 Divided Differences


The Lagrange’s formula has the drawback that if another interpola-
tion value were inserted, then the interpolation coefficients are required
to be recalculated. This labor of recomputing the interpolation coefficients
is saved by using Newton’s general interpolation formula which employs
what are called “divided differences.” Before deriving this formula, we
shall first define these differences.
If ( x0 , y0 ), ( x1 , y1 ), ( x2 , y2 ), be given points, then the first divided dif-
ference for the arguments x0, x1 is defined by the relation [x0, x1] or
y  y0
 y0  1
x1 x1  x0
y  y1 y  y2
Similarly [x1, x2] or  y0  2 and [x2, x3] or  y0  3
x2 x2  x1 x3 x3  x2
The second divided difference for x0, x1, x2 is defined as
[ x , x ][ x0 , x1 ]
[x0, x1, x2] or 2  1 2
x1 , x2 x2  x0

The third divided difference for x0, x1, x2, x3 is defined as


[ x1 , x2, x3 ][ x0 , x1 , x2 ]
[x0, x1, x2, x3] or 3 y0 
x1 , x2 , x3 x2  x0
Properties of Divided Differences
I. The divided differences are symmetrical in their arguments, i.e,. inde-
pendent of the order of the arguments. For it is easy to write
y0 y1
[ x0 , x1 ]    [ x1 , x0 ],[ x0 , x1 , x2 ]
x0  x1 x1  x0
y0 y1 y2
  
( x0  x1 )( x0  x2 ) ( x1  x0 )( x1  x2 ) ( x2  x0 )( x2  x1 )
 [x1, x2, x0] or [x2, x0, x1] and so on
II. The nth divided differences of a polynomial of the nth degree are con-
stant.
Let the arguments be equally spaced so that
x – x  x – x  x – x  h. Then
1 0 2 1 n n1
314 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

y1  y0 y0
 x0 , x1 
  
x1  x0 h
 x1 , x2  x0  x1  1  y1 y0 
 x0 , x1 , x2 
     
x2  x0 2h  h h 
1 1
 2y0 and in general, [ x0 , x1 , x2 ,......, xn ]   n y0
2!h 2
n! hn
If the tabulated function is a nth degree polynomial, then ny0 will be
constant. Hence the nth divided differences will also be constant
III. The divided difference operator 䊖 is linear
i.e., 䊖aux  bvx   a䊖 ux  b䊖 vx

 aux  bvx1    aux0  bvx0 


We have 䊖  aux0  bvx0   1

x1 x1  x0
 ux  ux0   vx  vx0 
 a 1   b 1 
 x1  x0   x1  x0 
 a 䊖 ux0  b 䊖 vx0
x1 x0

In general 䊖 aux  bvx   a䊖 ux  b䊖 vx This property is also true for


higher order differences.

7.14 Newton’s Divided Difference Formula


Let y0, y1,,yn be the values of y  f(x) corresponding to the arguments
x0, x1,, xn. Then from the definition of divided differences, we have

y  y0
 x, x0 
 
x  x0
So that y  y0   x  x0 
 x, x0 

 x, x0 
  x0 , x1 

Again 
 x, x0 , x1 

x  x1
which gives [ x, x0 ]  [ x0 , x1 ]  ( x – x1 )[ x, x0 , x1 ]
INTERPOLATION • 315

Substituting this value of [x, x0] in (1), we get


y  y0  ( x  x0 )[ x0 , x1 ]  ( x  x0 )( x  x1 )[ x, x0 , x1 ]
(2)
[ x.x0, x1 ][ x.x0, x2 ]
Also [ x, x0 , x1 , x2 ] 
x  x2
which gives [ x, x0 , x1 ]  [ x0 , x1 , x2 ]  ( x – x2 )[ x, x0 , x1 , x2 ]
Substituting this value of [x, x0, x1] in (2), we obtain
y  y0  ( x – x0 )[ x0 , x1 ]  ( x – x0 )( x – x1 )[ x0 , x1 , x2 ]
( x – x0 )( x – x1 )( x – x2 )[ x, x0 , x1 , x2 ]
Proceeding in this manner, we get
y  y0  ( x – x0 )[ x0 , x1 ]  ( x – x0 )( x – x1 )[ x0 , x1 , x2 ]
( x – x0 )( x – x1 )( x – xn )[ x, x0 , x1 , xn ]
( x – x0 )( x – x1 )( x – x2 )[ x, x0 , x1 , x2 ]
(3)
which is called Newton’s general interpolation formula with divided
differences.

7.15 Relation Between Divided and Forward Differences


If ( x0 , y0 ), ( x1 , y1 ), ( x2 , y2 ), be the given points, then
y  y0
[ x0 , x1 ]  1
x1  x0
Also y0  y1 – y0
If x0, x1, x2, are equispaced, then x1 – x0  h, so that
y
[ x0 , x1 ]  0
h
y1
Similarly [ x1 , x2 ] 
h
[ x , x ][ x0 , x1 ]
Now [ x0 , x1 , x2 ]  1 2
x2  x0
y1 h  y0 h 
 x2  x0  2 h 


2h
y1  y0

2 h2
2 y0
Thus [ x0 , x1 , x2 ] 
2! h2
316 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

2 y1
Similarly [ x0 , x1 , x2 ] 
2! h2
2 2
2 y1 2h 2 y0 2h 2 y1 2 y0
[ x0 , x1 , x2 , x3 ]    x3  x0  3h 
 
 x3  x0 2h2 (3)
3 y0
Thus [ x0 , x1 , x2 , x3 ] 
3! h3
 n y0
In general, [ x0 , x1 , xn ] 
n! h n
This is the relation between divided and forward differences.

EXAMPLE 7.23
Given the values
x: 5 7 11 13 17
f(x): 150 392 1452 2366 5202
evaluate f(9), using Newton’s divided difference formula
Solution:
The divided differences table is
x y 䊖y 䊖2y 䊖3y
5 150 392  150
 121
75
7 392 265  121
 24
11  5
1452  392 32  24
 265 1
11  7 13  5
11 1452 457  265
 32
13  7
2366  1452 42  32
 457 1
13  11 17  7
13 2366 709  457
 42
17  11
5202  2366
 709
17  13
17 5202
INTERPOLATION • 317

Taking x  9 in the Newton’s divided difference formula, we obtain


f(9)  150  (9 – 5) × 121  (9 – 5)(9 – 7) × 24  (9 – 5)(9 – 7)(9 – 11) × 1
 150  484  192 – 16  810.

EXAMPLE 7.24
Using Newton’s divided differences formula, evaluate f(8) and f(15)
given:
x: 4 5 7 10 11 13
y  f(x): 48 100 294 900 1210 2028
Solution:
The divided differences table is
x f(x) 䊖y 䊖2y 䊖3y 䊖4y
4 48 0
52
5 100 15
97 1
7 294 21 0
202 1
10 900 27 0
310 1
11 1210 33
409
13 2028
Taking x  8 in the Newton’s divided difference formula, we obtain
f(8)  48  (8 – 4) 52  (8 – 4) (8 – 5) 15  (8 – 4) (8 – 5) (8 – 7) 1
 448.
Similarly f(15)  3150.

EXAMPLE 7.25
Determine f(x) as a polynomial in x for the following data:
x: –4 –1 0 2 5
y  f(x): 1245 33 5 9 1335
318 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

Solution:
The divided differences table is
x f(x) 䊖y 䊖2y 䊖3y 䊖4y
–4 1245
– 404
–1 33 94
– 28 – 14
0 5 10 3
2 13
2 9 88
442
5 1335
Applying Newton’s divided difference formula
f ( x)  f ( x0 )  ( x – x0 )[ x0 , x1 ]  ( x – x0 )( x – x1 )[ x0 , x1 , x2 ] 
 1245  (x  4) (– 404)  (x  4) (x  1) (94)
  (x  4)(x  1)(x – 0)(– 14)  (x  4)(x  1)x(x – 2)(3)
 3x  5x + 6x2  14x + 5
4 2

EXAMPLE 7.26
Using Newton’s divided difference formula, find the missing value
from the table:
x: 1 2 4 5 6
y: 14 15 5 ... 9
Solution:
The divided difference table is
x y 䊖y 䊖2y 䊖3y
1 14
15  14
1
2 1
2 15 5  1
2
4 1
5  15 7 / 42 3
5 
42 6 1 4
INTERPOLATION • 319

x y 䊖y 䊖2y 䊖3y
4 5 25 7

62 4
96
2
64
6 9
Newton’s divided difference formula is
y  y0  ( x – x0 )[ x0 , x1 ]  ( x – x0 )( x – x1 )[ x0 , x1 , x2 ]
( x – x0 )( x – x1 )( x – x2 )[ x0 , x1 , x2 , x3 ] 
3
 14  (x – 1) (1)  (x – 1) (x – 2) (– 2)  (x – 1) (x – 2) (x – 4)
4
Putting x  5, we get
3
y(5)  14  4  (4) (3) (– 2)  (4) (3) (1)  3.
4
Hence missing value is 3

Exercises 7.4

1. Find the third divided difference with arguments 2, 4, 9, 10 of the func-


tion f (x)  x3– 2x.
2. Obtain the Newton ’s divided difference interpolating polynomial and
hence find f(6):
x: 3 7 9 10
f (x): 160 120 72 63
3. Using Newton’s divided differences interpolation, find u(3), given that
u(1)  – 26, u(2)  12, u(4)  256, u(6)  844.
4. A thermocouple gives the following output for rise in temperature
Tem p (°C) 0 10 20 30 40 50
Output (m V) 0.0 0.4 0.8 1.2 1.6 2.0

Find the output of thermocouple for 37°C temperature using Newton’s


divided difference formula.
5. Using Newton ’s divided difference interpolation, find the polynomial of
the given data:
x: –1 0 1 3
f (x): 2 1 0 –1
320 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

6. For the following table, find f(x) a s a polynomial in x using Newton’s


divided difference formula:
x: 5 6 9 11
f (x): 12 13 14 16
7. Using the following data, find f(x) a s a polynomial in x:
x: –1 0 3 6 7
f(x): 3 –6 39 822 1611
8. The observed values of a function are respectively 168, 120, 72, and 63
at the four positions 3, 7, 9, and 10 of the independent variable. What is
the best estimate value of the function at the position 6?
9. Find the equation of the cubic curve which passes through the point s
(4, – 43), (7, 83), (9, 327), and (12, 1053).
10. Find the missing term in the following table using Newton’s divided dif-
ference formula.
x: 0 1 2 3 4
y: 1 3 9 ... 81

7.16 Hermite’s Interpolation Formula


This formula is similar to the Lagrange’s interpolation formula. In La-
grange’s method, the interpolating polynomial P(x) agrees with y(x) at the
points x0, x1,......, xn, whereas in
Hermite’s method P(x) and y(x) as well as P(x) and y(x) coincide at the
(n  1) points, i.e.,
P(xi)  y(xi) and P(xi)  y(xi); i  0, 1,......, n (1)
As there are 2(n  1) conditions in (1), (2n  2) coefficients are to be
determined.
Therefore P(x) is a polynomial of degree (2n  1).
We assume that P(x) is expressible in the form
n n
p( x)   U i  x  y  xi    V  x y'  x 
i i (2)
i0 i0

where Ui (x) and Vi (x) are polynomials in x of degree (2n  1). These are to
be determined. Using the conditions (1), we get
326 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

7.17 Spline Interpolation


In the interpolation methods so far explained, a single polynomial has
been fitted to the tabulated points. If the given set of points belong to the
polynomial, then this method works well, otherwise the results are rough
approximations only. If we draw lines through every two closest points, the
resulting graph will not be smooth. Similarly we may draw a quadratic curve
through points Ai, Ai+1 and another quadratic curve through Ai+1, A i+2, such
that the slopes of the two quadratic curves match at A i+1 (Fig. 7.1). The
resulting curve looks better but is not quite smooth. We can ensure this
by drawing a cubic curve through A i, Ai+1 and another cubic through A i+1,
A i+2su ch that the slopes and curvatures of the two curves match at A i+1.
Such a curve is called a cubic spline. We may use polynomials of higher
order but the resulting graph is not better. As such, cubic splines are com-
monly used. This technique of “spline-fitting” is of recent origin and has
important applications.

A2

f1(x) An–1
fi+1(x) fn–1(x)
f0(x) fi(x)
A0 Ai+2
Ai Ai+1 An
A1
y0 y1 y2 yi yi+1 yi+2 yn–1 yn

0 x0 x1 x2 xi xi+1 xi+2 xn–1 xn x

FIGURE 7.1

Cubic spline
Consider the problem of interpolating between the data points (x0, y0),
(x1, y1),(xn, yn) by means of spline fitting.
Then the cubic spline f(x) is such that
(i) f(x) is a linear polynomial outside the interval (x0, xn),
(ii) f(x) is a cubic polynomial in each of the subintervals,
(iii) f(x) and f(x) are continuous at each point.
Since f(x) is cubic in each of the subintervals f(x) shall be linear.
INTERPOLATION • 327

 Taking equally-spaced values of x so that xi+1 – xi  h, we can write


1
  x – x  f ( xi )  ( x – xi ) f ( xi1 ) 
f ( x)   
h i1
Integrating twice, we have
1  xi1 – x  ( x – xi ) 
f ( x)   f ( xi )  f ( xi1 )ai  xi1 – x   bi ( x – xi ) (1)
h 3! 3! 
The constants of integration ai, bi are determined by substituting the
values of y  f(x) at xi and xi+1. Thus,

1 h2  1 h2 
ai   yi  f ( xi ) and bi   yi1  f ( xi1 )
h 3!  h 3! 

Substituting the values of ai, bi and writing f(xi)  Mi, (1) takes the form
3 3
f ( x) =
( xi +1 – x )
M +
( x – xi )
M
i i +1
6h 6h
xi1 – x h2  x – xi  h2 
  yi  Mi   yi1  Mi1  (2)
h  6  h  6 

 xi1 – x2  x – xi 2 h
 f ( x)  Mi  Mi1  Mi1  Mi   1  yi1  yi 
2h 6h 6 h
To impose the condition of continuity of f (x), we get
f ′( x – ε ) = f ′( x + ε ) as ε → 0

h 1 h 1
 (2Mi  Mi –1 )  (yi – yi –1 )  – (2Mi  Mi1 )  (yi1 – yi )
6 h 6 h
6
Mi –1  4Mi  Mi1  2 (yi –1 – 2 yi  yi1 ), i  1 to n  1 (3)
h
Now since the graph is linear for x < x0 and x > xn, we have
M0  0, Mn  0 (4)
(3) and (4) give (n  1) equations in (n  1) unknowns Mi (i  0, 1, n)
which can be solved. Substituting the value of Mi in (2) gives the concerned
cubic spline.
328 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

EXAMPLE 7.31
Obtain the cubic spline for the following data
x: 0 1 2 3
y: 2 –6 –8 2
Solution:
Since the points are equispaced with h  1 and n  3, the cubic spline
can be determined from Mi–1  4Mi  Mi+1  6 (yi–1 – 2yi  yi+1), i  1, 2.
 M0  4M1  M2  6 (y0 – 2y1  y2)
M1  4M2  M3  6 (y1 – 2y2  y3)
i.e., 4M1  M2  36; M1  4M2  72 [ M0  0, M3  0]
Solving these, we get M1  4.8, M2  16.8.
Now the cubic spline in (xi  x  xi  1) is
1 1 1
f ( x) = ( xi +1 − x )3 M i + ( x − xi )3 M i +1 + ( xi +1 − x ) ⎛⎜ yi − M i ⎞⎟
6 6 ⎝ 6 ⎠
 1 
 x  xi  yi1  Mi1 
 6 
Taking i  0 in (A) the cubic spline in (0  x  1) is
1 1 1
f ( x)  (1 – x)3 (0)  ( x – 0)3  4.8   (1 – x)( x – 0)  x[–6 –  4.8 
6 6 6
 0.8x3 – 8.8x  2 (0  x  1)
Taking i  1 in (A), the cubic spline in (1  x  2) is
1 1 1
f ( x) = (2 – x)3 (4.8) + ( x – 1)3 (16.8) + (2 – x)[– 6 – ( 4.8 )]
6 6 6
+( x – 1)[– 8 – 1(16.8 )]
 2x3 – 5.84x2 – 1.68x  0.8
Taking i  2 in (A), the cubic spline in (2  x  3) is
1 1
f ( x)  (3 – x)3 (4.8)  ( x – 2)3 (0)  (3 – x)[– 8 – 1 16.8 ]
6 6
( x – 2)[2 – 1 2 ]
 – 0.8x3  2.64x2  9.68x – 14.8
INTERPOLATION • 329

EXAMPLE 7.32
The following values of x and y are given:
x: 1 2 3 4
y: 1 2 5 11
Find the cubic splines and evaluate y(1.5) and y(3).
Solution:
Since the points are equispaced with h  1 and n  3, the cubic splines
can be obtained from
Mi–1  4Mi  Mi+1  6(yi–1 – 2yi  yi+1), i  1, 2.
 M0  4M1  M2  6(y0 – 2y1  y2)
M1  4M2  M3  6(y1 – 2y2  y3)
i.e., 4M1  M2  12, M1  4M2  18 [ M0  0, M3  0]
which give, M1  2, M2  4.
Now the cubic spline in (xi  x  xi+1) is
1 1 1
f ( x) = ( xi +1 − x )3 M i + ( x − xi )3 M i +1 + ( xi +1 − x ) ⎛⎜ yi − M i ⎞⎟
6 6 ⎝ 6 ⎠
 1 
 x  xi  yi1  Mi1  (A)
 6 
Thus, taking i  0, i  1, i  2 in (A), the cubic splines are
1 3
  x 3 x 5 x  1  x  2
2

 3
1
f ( x)    x3 3 x2 5 x  2  x  3
3
1

3
2 x3 24 x2 76 x  81 3  x  4
 y(1.5)  f(1.5)  11/8

EXAMPLE 7.33
Find the cubic spline interpolation for the data:
x: 1 2 3 4 5
f (x): 1 0 1 0 1
330 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

Solution:
Since the points are equispaced with h  1, n  4, the cubic spline can
be found by means of
Mi –1  4Mi  Mi1  6(yi –1 – 2 yi  yi1 ), i  1,2, 3
 M0  4M1  M2  6  y0 – 2 y1  y2   12
M1  4M2  M3  6(y1 – 2 y2  y3 )  –12
M2  4 M3  M4  6(y2 – 2 y3  y4 )  12
Since M0  y 0  0 and M4  y 4  0
 4M1  M2  12; M1  4M2  M3  –12; M1  4M3  12
Solving these equations, we get M1  30/7, M2  – 36/7, M3  30/7
Now the cubic spline in (xi  x  xi1) is
1 1
f ( x)   x  x3 Mi   x  xi 3 Mi1   xi1  x
6 i1 6
⎛ 1 ⎞ ⎛ 1 ⎞
⎜ yi − M i ⎟ + ( x − xi ) ⎜ yi +1 − M i +1 ⎟ (A)
⎝ 6 ⎠ ⎝ 6 ⎠
Taking i  0, in (A), the cubic spline in (1  x  2) is
1  1 
y 
  x1  x  M0   x  x0  M1 
3 3
  x1  x 

y0  M0 

6 6
 1 
 x  x0  y1  M1 
 6 

1
 
3 3  2  x 
 2  x  0    x  x0  (30 / 7)
1 
1  0 
6  6 
 1 30 
 x  1 0   
 6 7 
i.e., y  0.71x – 2.14x  0.42x  2 (1 < x  2)
3 2

Taking i  1 in (A), the cubic spline in (2  x  3) is


1⎡ 30 36 ⎤ ⎛ 1 30 ⎞
y= ⎢ ( 3 − x )3 + ( x − 2 )3 ⎛⎜ − ⎞⎟ ⎥ + ( 3 − x ) ⎜ 0 − ⎛⎜ ⎞⎟ ⎟
6⎣ 7 ⎝ 7 ⎠⎦ ⎝ 6 ⎝ 7 ⎠⎠
⎛ 1 ⎛ 36 ⎞ ⎞
+ ( x − 2 ) ⎜1 − ⎜ − ⎟ ⎟
⎝ 6 ⎝ 7 ⎠⎠
INTERPOLATION • 331

i.e., y  – 1.57 x3  11.57x2 – 27x  20.28. (2  x  3)


Taking i  2 in (A), the cubic spline in (3  x  4) is
1 3 36  1 3 30  1 36   5
y   4  x     x  3   4  x 1     x  3 0  
6  7 6 7  6 7   7
i.e., y  1.57 x3 – 16.71 x2  57.86x – 64.57 (3  x 4)
Taking i  3 in (A), the cubic spline in (4  x  5) is
1 3 30  3 5 
y  1  x     5  x     x  4 1
6 7  7
i.e., y  – 0.71x3  2.14x2 – 0.43x – 6.86. (4  x  5)

Exercises 7.6

1. Find the cubic splines for the following table of values:


x: 1 2 3
y: –6 –1 16
Hence evaluate y(1.5) and y(2).
2. The following values of x and y are given:
x: 1 2 3 4
y: 1 5 11 8
Usin g cubic spline, show that
(i) y(1.5)  2.575 (ii) y(3)  2.067.
3. Find the cubic spline corresponding to the interval [2,3] from the
following table:
x: 1 2 3 4 5
y: 30 15 32 18 25
Hence compute (i) y(2.5) and (ii) y(3).

7.18 Double Interpolation


So far, we have derived interpolation formulae to approximate a func-
tion of a single variable. In the case of functions, of two variables, we inter-
polate with respect to the first variable keeping the other variable constant.
Then interpolate with respect to the second variable.
CHAPTER

5
EMPIRICAL LAWS AND
CURVE-FITTING
Chapter Objectives
O Introduction
O Graphical method
O Laws reducible to the linear law
O Principle of least squares
O Method of least squares
O Fitting a curve of the type y  a  bx2, etc.
O Fitting of other curves
O Most Plausible values
O Method of group averages
O Laws containing three constants
O Method of moments
O Objective type of questions.

5.1 Introduction
In many branches of Applied Mathematics, it is required to ex-
press a given data, obtained from observations, in the form of a law
connecting the two variables involved. Such a law inferred by some
scheme, is known as the empirical law. For example, it may be de-
sirable to obtain the law connecting the length and the tempera-
ture of a metal bar. At various temperatures, the length of the bar
is measured. Then, by one of the methods explained below, a law
194 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

is obtained that represents the relationship existing between temperature


and length for the observed values. This relation ship can then be used to
predict the length at an arbitrary temperature.
Scatter diagram. To find a relationship between the set of paired observa-
tions x and y, we plot their corresponding values on the graph, taking one
of the variables along the x-axis and other along the y-axis, i.e., (x1, y1), (x2,
y2), (xn, yn). The resulting diagram showing a collection of dots is called a
scatter diagram. A smooth curve that approximates the above set of points
is known as the approximating curve.
Curve fitting. Several equations of different types can be obtained to ex-
press the given data approximately. But the problem is to find the equation
of the curve of “best fit” which may be most suitable for predicting the un-
known values. The process of finding such an equation of ‘best fit’ is known
as curve-fitting.
If there are n pairs of observed values then it is possible to fit the given
data to an equation that contains n arbitrary constants, and we can solve n
simultaneous equations for n unknowns. If we desired to obtain an equation
representing these data but have less then n arbitrary constants, then we
can have recourse to any of these four methods: Graphical method, method
of least-squares, method of group averages, and method of moments. The
graphical method and the method of averages fail to give the values of the
unknown constants uniquely and accurately, while the other methods do.
The method of least squares is probably the best to fit a unique curve to a
given data. It is widely used in applications and can be easily implemented
on a computer.

5.2 Graphical Method


When the curve representing the given data is a linear law y  mx  c;
we proceed as follows:
(i) Plot the given points on the graph paper taking a suitable scale.
(ii) Draw the straight line of best fit such that the points are evenly
distributed about the line.
(iii) Taking two suitable points (x1, y1) and (x2, y2) on the line, calculate
m, the slope of the line and c, its intercept on the y-axis.
EMPIRICAL LAWS AND CURVE-FITTING • 195

When the points representing the observed values do not approximate


to a straight line, a smooth curve is drawn through them. From the shape
of the graph, we try to infer the law of the curve and then reduce it to the
form y  mx  c.

5.3 Laws Reducible to the Linear Law


We give below some of the laws in common use, indicating the way
these can be reduced to the linear form by suitable substitutions:

1. When the law is y  mxn  c


Taking xn  X and y  Y, the above law becomes Y  mX  c
2. When the law is y  axn.
Taking logarithms of both sides, it becomes log10y  log10a  n log10x
Putting log10x  X and log10y  Y, it reduces to the form
Y  nX  c, where c  log10a.
3. When the law is y  axn  b log x.
y xn
Writing it as a  b and taking xn/log x  X and y/log x  Y, the
log x log x
given law becomes, Y  aX  b.
4. When the law is y  aebx.
Taking logarithms, it becomes log10 y  (b log10 e)x  log10a.
Putting x  X and log10 y  Y, it takes the form Y  mX  c
where m  b log10e and c  log10a.
5. When the law is xy  ax  by.
y
Dividing by x, we have y  b  a .
x
Putting y/x  X and y  Y, it reduces to the form Y  bX  a.

EXAMPLE 5.1
R is the resistance to motion of a train at speed V; find a law of the type
R  a  bV2 connecting R and V, using the following data:
V (km/hr): 10 20 30 40 50
R (kg/ton): 8 10 15 21 30
196 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

Solution:
Given law is R  a  bV2 (i)
Taking V2  x and R  y,
(i) becomes, y  a  bx (ii)
which is a linear law.
Table for the values of x and y is as follows:
x: 100 400 900 1600 2500
y: 8 10 15 21 30

Plot these points. Draw the straight line of best fit through these points
(Figure 5.1).
Y

40
(2500,30)
30
N
20 (1600,21)
L
(900,15) M
10 (400,10)
(100,8)
0 X
(0, 0) 400 800 1200 1600 2000 2400 2800
FIGURE 5.1

Slope of this line ( b)


MN 21  15 6
  
LM 1600  900 700
 0.0085 nearly.
Since L(900, 15) lies on (ii),
 15  a  0.0085 × 900,
where a  15 – 7.65  7.35 nearly.

EXAMPLE 5.2
The following values of x and y are supposed to follow the law
y  ax2  b log10x. Find graphically the most probable values of the constants
a and b.
EMPIRICAL LAWS AND CURVE-FITTING • 197

x 2.85 3.88 4.66 5.69 6.65 7.77 8.67


y 16.7 26.4 35.1 47.5 60.6 77.5 93.4

Solution:
Given law is y  ax2  b log10x
y x2
i.e., a b (i)
log10 x log10 x
Putting x2/log10 x  X and y/ log10x  Y,
(i) becomes Y  aX  b (ii)
This is a linear law.
Table for the values of X and Y is as follows:
X  x2/log10 x 17.93 25.56 32.49 42.87 53.75 67.80 80.83
Y  y/log10 x 35.59 44.83 52.50 69.90 73.65 87.04 99.56
Points P1 P2 P3 P4 P5 P6 P7

Plot these points and draw the straight line of best fit through these points
(Figure 5.2).

95
P6

85

75
P5

65
P4

55
P3
M
P2
45

35 P1
X
0 15 25 35 45 55 65 75
FIGURE 5.2
MP5 73.65  52.50 21.15
Slope of this line ( a)     0.99
P3 M 53.75  32.49 21.26
198 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

Since P3 lies on (ii), therefore,


52.50  0.99 × 32.49  b
where b  20.2
Hence (i) becomes y  (0.99) x2  (20.2) log10x.

EXAMPLE 5.3
The values of x and y obtained in an experiment are as follows:
x 2.30 3.10 4.00 4.92 5.91 7.20
y 33.0 39.1 50.3 67.2 85.6 125.0
The probable law is y  aebx. Test graphically the accuracy of this law
and if the law holds good, find the best values of the constants.
Solution:
Given law is y  aebx (i)
Taking logarithms to base 10, we have
log10y  log10a  (b log10e) x
Putting x  X and log10y  Y, it becomes
y  (b log10e) X  log10a (ii)
Table for the values of X and Y is as under:
Xx 2.30 3.10 4.00 4.92 5.91 7.20
Y  log10 y 1.52 1.59 1.70 1.83 1.93 2.10
Points P1 P2 P3 P4 P5 P6

Scale: 1 small division along x-axis  0.1


10 small divisions along y-axis  0.1.
Plot these points and draw the line of best fit. As these points are lying
almost along a straight line, the given law is nearly accurate (Figure 5.3).
e MN
Now the slope of this line ( b log10 )   0.12
NM
0.12
where b   0.12  2.303  0.276
log10 e
EMPIRICAL LAWS AND CURVE-FITTING • 199

Since the point L (4, 1.71) lies on (ii), therefore,


1.71  0.12 × 4  log10 a where a  17 nearly.
Hence the curve of best fit is y  17 e0.276 x.

2.1
P6

2.0
P5
1.9
P4 N(5,1.83)
1.8

L(4,1.71) M
1.7 P3

1.6
P2
P1
0 X
(2, 1.5) 3 4 5 6 7 8
FIGURE 5.3

Exercises 5.1

1. If p is the pull required to lift the weight by means of a pulley block, find
a linear law of the form p  a  bw, connecting p and w, using the follow-
ing data:
w(lb): 50 70 100 120
p(lb): 12 15 21 25
Compute p, when w  150 lb.

2. Convert the following equations to their linear forms:


(i) y  ax  bx2 (ii) y  b/[x(x – a)].
3. The resistance R of a carbon filament lamp was measured at various
values of the voltage V and the following observations were made:
Voltage V... 62 70 78 84 92
Resistance R... 73 70.7 69.2 67.8 66.3
200 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

a
Assuming a law of the form R   b, find by graphical method the best
values of a and b. V

4. Verify if the values of x and y, related as shown in the following table,


obey the law y  a  b x. If so, find graphically the values of a and b.
x: 500 1,000 2,000 4,000 6,000
y: 0.20 0.33 0.38 0.45 0.51

5. The following table gives the pressure p and the volume v at various
instants during the expansion of steam in a cylinder. Show that the equa-
tion of the expansion is of the form pvn  c and find the values of n and c
approximately.
p: 200 100 50 30 20 10
v: 1.0 1.7 2.9 4.8 5.9 10

6. The following values of T and l follow the law T  aln. Test if this is so
and find the best values of a and n.
T  1.0 1.5 2.0 2.5
l  25 56.2 100 1.56
Fit the curve y  ae to the following data:
bx

x: 0 2 4
y: 5.1 10 31.1
The following are the results of an experiment on friction of bearings.
The speed being kept constant, corresponding values of the coefficient
of friction and the temperature are shown in the table:
t: 120 110 100 90 80 70 60
: 0.0051 0.0059 0.0071 0.0085 0.00102 0.00124 0.00148

If  and t are given by the law   aebt, find the values of a and b by
plotting the graph for  and t.

5.4 Principle of Least Squares


The graphical method has the obvious drawback of being unable to give
a unique curve of fit. The principle of least squares, however, provides an
EMPIRICAL LAWS AND CURVE-FITTING • 201

elegant procedure of fitting a unique curve to a given data.


Let the curve y  a  bx  cx2   kxm
be fitted to the set of data points (x1, y1), (x2, y2),, (xn, yn).

Y Pn
(1)
en
Pi(xi, yi)
ei Mn

P1(x1, y1)

e1 Mi
e2 M2
M1
P2

0 X
L1 L2 Li Ln
FIGURE 5.4
Now we have to determine the constants a, b, c,... k such that they
represents the curve of best fit. In the case of n  m, when substituting the
values (xi, yi) in (1), we get n equations from which a unique set of n con-
stants can be found. But when n > m, we obtain n equations which are more
than the m constants and hence cannot be solved for these constants. So we
try to determine the values of a, b, c,  k which satisfy all the equations as
nearly as possible and thus may give the best fit. In such cases, we apply the
principle of least squares.
At x  xi, the observed (experimental) value of the ordinate is yi and the
corresponding value on the fitting curve (1) is a  bxi  cxi2   kxim (  i,
say) which is the expected (or calculated) value (Figure 5.4). The difference
of the observed and the expected values, i.e., yi – i(  ei) is called the error
(or residual) at x  xi. Clearly some of the errors e1, e2,, en will be positive
and others negative. Thus to give equal weightage to each error, we square
each of these and form their sum, i.e., E  e12  e22   en2.
The curve of best fit is that for which e’s are as small as possible, i.e., the
sum of the squares of the errors is a minimum. This is known as the prin-
ciple of least squares and was suggested by a French mathematician Adrien
Marie Legendre in 1806.

Obs. The principle of least squares does not help us to


NOTE
determine the form of the appropriate curve which can fit a
202 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

given data. It only determines the best possible values of the


constants in the equation when the form of the curve is known
before hand. The selection of the curve is a matter of experience
and practical considerations.

5.5 Method of Least Squares


For clarity, suppose it is required to fit the curve y  a  bx  cx2 to
a given set of observations (x1, y1), (x2, y2), , (x5, y5). For any xi, the ob-
served value is yi and the expected value is i  a  bxi  cxi2 so that the error
ei  yi – i.
 The sum of the squares of these errors is
E  e12  e2 2  e52
 [y1  (a  bx1  cx12 )]2 [y2  bx2  cx2 2 )]2  [y5  (a  bx5  cx52 )]2
For E to be minimum, we have
E
 0 2[y1  (a  bx1  cx12 )]2  2[y2  (a  bx2  cx22 )]
a (1)
 2[y5  (a  bx5  cx52 )]2

E
 0 2 x1 [y1  (a  bx1  cx12 )] 2 x2 [y2  (a  bx2  cx22 )] (2)
b
 2 x2 [y5  (a  bx5  cx52 )]

E
 0 2 x12 [y1  (a  bx1  cx12 )] 2 x22 [y2  (a  bx2  cx22 )] (3)
b
 2 x52 [y5  (a  bx5  cx52 )]
Equation (1) simplifies to
y1  y2   y5  5a  b( x1  x2   x5 )  c( x12  x22   x52 )

i.e, yi  5a  bxi  cxi2 (4)


Equation (2) becomes
x1 y1  x2 y2   x5 y5  a( x1  x2   x5 )  b( x12  x22   x52 )
c( x13  x23   x53 )
i.e.,  xi yi  a  xi  b  xi2  x  xi3
EMPIRICAL LAWS AND CURVE-FITTING • 203

Similarly (3) simplifies to


 xi2 yi  a  xi2  b  xi3  c  xi 4
The equations (4), (5) and (6) are known as normal equations and can
be solved as simultaneous equations in a, b, c. The values of these constants
when substituted in (1) give the desired curve of best fit.
2 2 2
NOTE Obs. On calculating  E ,  E ,  E and substituting the values
a2 b2 c2
of a, b, c just obtained, we will observe that each is positive, i.e,.
E is a minimum.
Working procedure
(a) To fit the straight line y  a  bx
(i) Substitute the observed set of n values in this equation.
(ii) Form normal equations for each constant, i.e., y  na  bx,
xy  ax  bx2.
[The normal equation for the unknown a is obtained by mul-
tiplying the equations by the coefficient of a and adding. The
normal equation for b is obtained by multiplying the equations
by the coefficient of b (i.e. x) and adding.]
(iii) Solve these normal equations as simultaneous equations for a
and b.
(iv) Substitute the values of a and b in y  a  bx, which is the
required line of best fit.
(b) To fit the parabola: y  a  bx  cx2
(i) Form the normal equations y  na  bx  cx2
xy  ax  bx2  cx3 and x2y  ax2  bx3  cx4
[The normal equation for c has been obtained by multiplying the
equations by the coefficient of c (i.e., x2) and adding.]
(ii) Solve these as simultaneous equations for a, b, c.
(iii) Substitute the values of a, b, c in y  a  bx  cx2, which is the
required parabola of best fit.
(c) In general, the curve y  a  bx  cx2   kxm1 can be fitted to a
given data by writing m normal equations.
204 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

EXAMPLE 5.4
If P is the pull required to lift a load W by means of a pulley block, find
a linear law of the form P  mW  c connecting P and W, using the follow-
ing data:
P  12 15 21 25
W  50 70 100 120
where P and W are taken in kg-wt. Compute P when W  150 kg.
Solution:
The corresponding normal equations are
 P  4c  m  W 

 (i)
2
 WP  c  W  m  W 
The values of W etc. are calculated by means of the following table:
W P W2 WP
50 12 2500 600
70 15 4900 1050
100 21 10000 2100
120 25 14400 3000
Total = 340 73 31800 6750
 The equations (i) become 73  4c  340m and 6750  340c  31800m
i.e., 2c  170m  365 (ii)
and 34c  3180m  675 (iii)
Multiplying (ii) by 17 and subtracting from (iii), we get m  0.1879
 from (ii), c  2.2785
Hence the line of best fit is P  2.2759  0.1879 W
When W  150 kg, P  2.2785  0.1879 × 150  30.4635 kg.

EXAMPLE 5.5
Fit a straight line to the following data:
x 6 7 7 8 8 8 9 9 10
y 5 5 4 5 4 3 4 3 3
EMPIRICAL LAWS AND CURVE-FITTING • 205

Solution:
Let the straight line be y  ax  b.
Then the normal equations are y  ax  9b
 xy  ax2  bx (i)
The values of x, y etc. are calculated below:
x y xy x2
6 5 30 36
7 5 35 49
7 4 28 16
8 5 40 64
8 4 32 64
8 3 24 64
9 4 36 81
9 3 27 81
10 3 30 100
x  72 y  36 xy  282 x2  588
 The equations (i) become 36  72a  9b and 282  588a  72b
i.e., 8a  b  4 (ii)
98a  12b  47 (iii)
Multiplying (ii) by 12 and subtracting from (iii), we get a  – 0.5.
From (ii), b  8.
Hence the required line of best fit is y  – 0.5x  8.

EXAMPLE 5.6
Fit a second degree parabola to the following data:
x 0 1 2 3 4
y 1 1.8 1.3 2. 6.3
Solution:
Let u  x – 2 and v  y so that the parabola of fit y  a  bx  cx2 becomes
v  A  Bu  Cu2
The normal equations are
v  5 A  Bu  Cu2 or 12.9  5 A  10C
uv  Au  Bu2  Cu2 or 11.3  10B
2 2 3 4
u v  Au  Bu  Cu or 33.5  10 A  34C
206 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

Solving these as simultaneous equations, we get


A = 1.48, B = 1.13, C = 0.55
 (i) becomes; v = 1.48 +1.13u + 0.55u2
y = 1.48 + 1.13(x  2) + 0.55(x  2)2
Hence y = 1.42  1.07x +0.55x2

NOTE Obs. For the sake of convenience and ease in calculations, it


is sometimes advisable to change the origin and scale with the
substitutions X  (x – A)/h and Y  (y – B)/h, where A and B are
the assumed means (or middle values) of x and y series, respec-
tively and h is the width of the interval.

EXAMPLE 5.7
Fit a second degree parabola to the following data:
x  1.0 1.5 2.0 2.5 3.0 3.5 4.0
y  1.1 1.3 1.6 2.0 2.7 3.4 4.1
Solution:
We shift the origin to (2.5, 0) and take 0.5 as the new unit. This amounts
to changing the variable x to X, by the relation X  2x – 5.
Let the parabola of fit be y  a  bX  cX2.
The values of X etc. are calculated as below:
x X y Xy X2 X2y X3 X4
1.0 –3 1.1 –3.3 9 9.9 –27 81
1.5 –2 1.3 –2.6 4 5.2 –8 16
2.0 –1 1.6 –1.6 1 1.6 –1 1
2.5 0 2.0 0.0 0 0.0 0 0
3.0 1 2.7 2.7 1 2.7 1 1
3.5 2 3.4 6.8 4 13.6 8 16
4.0 3 4.1 12.3 9 36.9 27 81
Total 0 16.2 14.3 28 69.9 0 196
The normal equations are
7a  28c  16.2, 28b  14.3, 28a  196c  69.9
Solving these as simultaneous equations, we get
a  2.07, b  0.511, c  0.061.
 y  2.07  0.511X  0.061 X2
EMPIRICAL LAWS AND CURVE-FITTING • 207

Replacing X by 2x – 5 in the above equation, we get


y  2.07  0.511 (2x – 5)  0.061 (2x – 5)2
which simplifies to y  1.04 – 0.198x  0.244x2
This is the required parabola of best fit.

EXAMPLE 5.8
Fit a second degree parabola to the following data:
x: 1989 1990 1991 1992 1993 1994 1995 1996 1997
y: 352 356 357 358 360 361 361 360 359
Solution:
Taking u  x – 1993 and v  y – 357, the equation y  a  bx  cx2
becomes
v  A  Bu  Cu2 (i)
x u  y v  uv u2 u2v u2 u4
x – 1993 y – 357
1989 –4 352 –5 20 16 –80 –64 256
1990 –3 356 –1 3 9 –9 –27 81
1991 –2 357 0 0 4 0 –8 16
1992 –1 358 1 –1 1 1 –1 1
1993 0 360 3 0 0 0 0 0
1994 1 361 4 4 1 4 1 1
1995 2 361 4 8 4 16 8 16
1996 3 360 3 9 9 27 27 81
1997 4 359 2 8 16 32 64 256
Total u  0 v  11 uv  51 u2  60 u2v  – 9 u3  0 u4  708

The normal equations are


v  9 A  Bu  Cu or 11  9 A  60C
17
uv  Au  Bu2  Cu3 or 51  60B or B 
20
2 2 3 4
u v  Au  Bu  Cu or – 9  60 A  708C
694 17 247
On solving these equations, we get A  , B ,C 
231 20 924
694 17 247 2
 (i) becomes v   u u
231 20 924
208 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

694 17 247
or y  357   ( x  1993)  ( x  1993)2
231 20 924
694 33881 247 17 247  3986 247 2
or y    (1993)2  x  x x
231 20 924 20 924 924
or y  3 – 1694.05 – 1061792.32  357  0.85 x  1065.52 x – 0.267 x2
2
Hence y  –1062526.37  1066.37 x – 0.267 x

Exercises 5.2

1. By the method of least squares, find the straight line that best fits the fol-
lowing data:
x: 1 2 3 4 5
y: 14 27 40 55 68
2. In some determinations of the value v of carbon dioxide dissolved in a
given volume of water at different temperatures , the following pairs of
values were obtained:
0 5 10 15
v  1.80 1.45 1.18 1.00
Obtain by the method of least squares, a relation of the form v  a  b
which best fits to these observations.
3. A simply supported beam carries a concentrated load P(lb) at its mid-
point. Corresponding to various values of P, the maximum deflection Y
(in) is measured. The data are given below:
P: 100 120 140 160 180 200
Y: 0.45 0.55 0.60 0.70 0.80 0.85
Find a law of the form Y  a  bP.

4. The result of measurement of electric resistance R of a copper bar at


various temperatures t° C are listed below:
t: 19 25 30 36 40 45 50
R: 76 77 79 80 82 83 85
Find a relation R  a  bt when a and b are constants to be determined
by you.
EMPIRICAL LAWS AND CURVE-FITTING • 209

5. A chemical company, wishing to study the effect of extraction time (t)


on the efficiency of an extraction operation (e) obtained the data shown
in the following table:
t: 27 45 41 19 3 39 19 49 15 31
e: 57 64 80 46 62 72 52 77 57 68

Fit a straight line to the given data by the method of least squares.

6. Find the parabola of the form y  a  bx  cx2 which fits most closely
with the observations:
x: –3 –2 –1 0 1 2 3
y: 4.63 2.11 0.67 0.09 0.63 2.15 4.58
7. By the method of least squares, fit a parabola of the form
y  a  bx  cx2, to the following data:
x: 2 4 6 8 10
y: 6.07 12.85 31.47 57.38 91.29
Fit a parabola y  a  bx  cx2 to the following data:
x: 1 2 3 4 5 6 7 8 9
y: 2 6 7 8 10 11 11 10 9
8. The velocity V of a liquid is known to vary with temperature T according
to a quadratic law V  a  b T  CT2. Find the best values of a, b, c for
the following table:
T: 1 2 3 4 5 6 7
V: 2.31 2.01 3.80 1.66 1.55 1.46 1.41
9. The following table gives the results of the measurements of train resis-
tance, V is the velocity in miles per hour, R is the resistance in pounds
per ton:
V: 20 40 60 80 100 120
R: 5.5 9.1 14.9 22.8 33.3 46.0
If R is related to V by the relation R  a  bV  cV , find a, b and c. 2

5.6 Fitting A Curve of the Type


(1) y  a  bx2 (2) y  ax  bx2
(3) y  ax  b/x (4) ax2  b/x.
210 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

(1) y  a  bx2
Putting x2  X, we have y  a  bX (i)
which is a linear equation. Its normal equations are
 y  na  bX; yX  aX  bX2
Solving these, we get a and b. Substituting these values of a, b and re-
placing X by x2 in (i), we obtain the desired equation of best fit.
(2) y  ax  bx2
Rewriting this equation as y/x  a  bx and putting y/x  Y, we have
Y  a  bx (i)
Its normal equations are
 Y  na  bx; Yx  ax  bx2
Solving these we get a and b. Replacing Y by y/x in (i), we obtain the
desired equation of best fit.
(3) y  ax  b/x
Rewriting this equation as xy  ax2  b
and putting x2  X and xy  Y, we have Y  b  aX (i)
Its normal equations are
Y  nb  X; XY  bX  aX2
Solving these equations, we get a and b. Replacing X by x2 and Y by xy
in (i), we obtain the desired equation of best fit.
(4) y  ax2  b/x
Rewriting this equation as xy  ax3  b and putting x3  X and xy  Y,
we have
Y  b  aX (i)
Its normal equations are
Y  bn  aX; XY  bX  X2
Solving these equations, we get a and b. Replacing X by x3 and Y by xy,
we obtain the desired equation of best fit.
EMPIRICAL LAWS AND CURVE-FITTING • 211

EXAMPLE 5.9
Find the least squares fit of the form y  a0  a1x2 to the following data
x: –1 0 1 2
y: 2 5 3 0
Solution:
Putting x2  X, we have y  a0  a1X (i)
 The normal equations are
 y  4a  a1X; Xy  a0X  a1X2.
The values of X, X2 etc. are calculated below:
X y X X2 XY
–1 2 1 1 2
0 5 0 0 0
1 3 1 1 3
2 0 4 16 0
y  10 X  6 X2  18 XY  5
 The normal equations become 10  400  6a1; 5  600  18a1
Solving these equations we get, a0  4.167, a1  – 1.111.
Hence the curve of best fit is
y  4.167 – 1.111X i.e., y  4.167 – 1.111x2.

EXAMPLE 5.10
Using the method of least squares, fit the curve y  ax2  b/x to the fol-
lowing data:
x: 1 2 3 4
y: –1.51 0.99 8.88 7.66
Solution:
Rewriting the given equation as xy  ax3  b and putting x3  X and
xy  Y, we get
Y  aX  b (i)
 The normal equations are
 Y  aX  4b; XY  aX2  bX
212 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

The values of X, Y etc. are calculated below:


X y X x3 Y xy XY X2
1 –1.51 1 –1.51 –1.51 1
2 0.99 8 1.98 15.84 64
3 3.88 27 11.64 314.28 729
4 7.66 64 30.64 1960.96 4096
X  100 Y  42. 75 XY  2289.57 X2  4890

 The normal equations become


42.75  100a  4b
2289.57  4890a  100b
Solving these equations, we get a  0.51, b  – 2.06
Hence the curve of best-fit is Y  0.51X ÷ 2.06
i.e.,

5.7 Fitting of Other Curves


(1) y  axb
Taking logarithms, log10 y  log10 a  blog10x
i.e., Y  A  bX (i)
where X  log10 x, Y  log10 y and A  log10 a.
 The normal equations for (i) are
 Y  nA  bX, XY  AX  bX2
from which A and b can be determined. Then a can be calculated from
A  log10 a.
(2) y  aebx (Exponential curve)
Taking logarithms, log10 y  log10 a  bx log10 e
i.e., Y  A  Bx where Y  log10 y, A  log10 a and B  b log10 e
Here the normal equations are Y  nA  Bx, xY  Ax  Bx2 from
which A, B can be found and consequently a, b can be calculated.
(3) xya  b (or pv  k) (Gas equation)
Taking logarithms, log10 x  a log10 y  log10 b
EMPIRICAL LAWS AND CURVE-FITTING • 213

or

This is of the form Y = A + BX


1 1
where X  log10 x, Y  log10 y, A  log10 b, B 
a a
Here also the problem reduces to finding a straight line of best fit
through the given data.

EXAMPLE 5.11
An experiment gave the following values:
v (ft/min): 350 400 50 600
t (min): 61 26 7 2.6
It is known that v and t are connected by the relation v  atb. Find the
best possible values of a and b.
Solution:
We have log10 v  log10 a  b log10 t
or Y  A  bX where X  log10t, Y  log10 v, A  log10 a.
 The normal equations are
Y  4A  bX (i)
XY  AX  bX2 (ii)
Now X etc. are calculated as in the following table:
v t X log10t Y  log10v XY X2
350 61 1.7853 2.5441 4.542 3.187
400 26 1.4150 2.6021 3.682 2.002
500 7 0.8451 2.6990 2.281 0.714
600 2.6 0.4150 2.7782 1.153 0.172
Total 4.4604 10.6234 11.658 6.075
 Equations (i) and (ii) become
4A  4.46b  10.623; 4.46A  6.075b  11.658
Solving these, A  2.845, b  – 0.1697
 a  antilog A  antilog 2.845  699.8.
214 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

EXAMPLE 5.12
Predict the mean radiation dose at an altitude of 3000 feet by fitting an
exponential curve to the given data:
Altitude (x): 50 450 780 1200 4400 4800 5300
Dose of radiation (y): 28 30 32 36 51 58 69

Solution:
Let y  abx be the exponential curve.
Then log10y  log10a  x log10b
or Y  A  Bx where Y  log10 y, A  log10 a, B  log10 b
 The normal equations are
 Y  7A  B x (i)
 x Y  Ax  B x2 (ii)
Now x etc., are calculated as follows:
x Y Y  log10 y xY x2
50 28 1.447158 72.3579 2500
450 30 1.477121 664.7044 202500
780 32 1.505150 1174.0170 608400
1200 36 1.556303 1867.5636 1440000
4400 51 1.707570 7513.3080 19360000
4800 58 1.763428 8464.4544 23040000
5300 69 1.838849 9745.8997 28090000
  16980 11.295579 29502.305 72743400

 Equations (i) and (ii) become


11.295579  7A  16980B
29502.305  16980A  72743400B
Solving these equations, we get A  1.4521015, B  0.0000666289
 log10 y  Y  1.4521015  0.0000666289 x
Hence y(at x  3000)  44.874 i.e., 44.9 approx.
EMPIRICAL LAWS AND CURVE-FITTING • 215

EXAMPLE 5.13
Fit a curve of the form y  aebx to the following data:
x: 0 1 2 3
y: 1.05 2.10 3.85 8.30
Solution:
Taking logarithms of both sides, the given equation becomes
log10 y  log10 a  bx log10 e
i.e., Y  A  bx where Y  log10 y, A  log10 a, B  b log10 e
 The normal equations are
 Y  4A  Bx; xY  Ax  Bx2.
Now x, Y etc. are calculated as in the table below:
x Y Y x2 xY
0 1.05 0.0212 0 0
1 2.10 0.3222 1 0.3222
2 3.85 0.5855 4 1.1710
3 8.30 0.9191 9 2.7573
x  6 Y  1.8480 x  14
2
xY  4.2505

Substituting these values in the normal equations, we get


4A  6B  1.848; 6A  14B  4.2505.
Solving these equations, A  0.0185, B  0.2956
   antilog A  1.0186, b  B/log10 e  0.6806
Hence the required curve of best fit is y  1.0186 e0.6806x.

EXAMPLE 5.14
The pressure and volume of a gas are related by the equation pV  k,
and k being constants. Fit this equation to the following set of observa-
tions:
p (kg/cm2): 0.5 1.0 1.5 2.0 2.5 3.0
V (litres): 1.62 1.00 0.75 0.62 0.52 0.46
216 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

Solution:
We have log10 p   log10 V  log10 k
1 1
or log10 V  log10 k  log10 P
 
1 1
or Y  A  BX where X  log10 P, Y  log10 V , A  log10 k, B 
 
 The normal equations are
 Y  6A  BX (i)
 XY  AX  BX 2
(ii)
Now X etc. are calculated as follows:
p V X  log10 p Y  log10 V XY X2
0.5 1.62 –0.3010 0.2095 –0.0630 0.0906
1.0 1.00 0.0000 0.0000 –0.0000 0.0000
1.5 0.75 0.1761 –0.1249 –0.0220 0.0310
2.0 0.62 0.3010 –0.2076 –0.0625 0.0906
2.5 0.52 0.3979 –0.2840 –0.1130 0.1583
3.0 0.46 0.4771 –0.3372 –0.1609 0.2276
Total 1.0511 – 0.7442 – 0.4214 0.5981
 Equations (i) and (ii) become
6A  1.0511B  – 0.7442; 1.0511A  0.5981B  – 0.4214
Solving these, we get A  0.0132, B  – 0.7836.
   – 1/B  1.276
and k  antilog (A)  antilog (0.0168)  1.039.
Hence the equation of best fit is pV1.276  1.039.

5.8 Most Plausible Values of Unknowns


Consider m linear equations in n unknowns:
a11 x1  a12 x2  a1 n xn  k1 

a12 x1  a22 x2  x2 n xn  k2 
 (1)
.............................................. 
am1 x1  am2 x2  amn xn  km  
218 • NUMERICAL METHODS IN ENGINEERING AND SCIENCE

Exercises 5.3

1. If V (km/hr) and R(kg/ton) are related by a relation of the type R  a 


bV2, find by the method of least squares a and b with the help of the fol-
lowing table:
V 10 20 30 40 50
R 8 10 15 21 30
2. Using the method of least squares fit the curve y  ax  bx2 to following
observations:
x: 1 2 3 4 5
y: 1.8 5.1 8.9 14.1 19.8
3. Fit the curve y  ax  b/x to the following data:
x: 1 2 3 4 5 6 7 8
y: 5.4 6.3 8.2 10.3 12.6 14.9 17.3 19.5
4. Estimate y at x  2.25 by fitting the indifference curve of the form
xy  Ax  B to the following data:
x: 1 2 3 4
y: 3 1.5 6 7.5
5. Fit a least square geometric curve y  axb to the following data:
x: 1 2 3 4 5
y: 0.5 2 4.5 8 12.5
6. Predict y at x  3.75, by fitting a power curve y  axb to the given data:
x: 1 2 3 4 5 6
y: 2.98 4.26 5.21 6.10 6.80 7.50
7. Obtain a relation of the form y  kxm for the following data by the
method of least squares:
x: 1 2 3 4 5
y: 7.1 27.8 62.1 110 161
8. Fit the exponential curve y  aebx to the following data:
x: 2 4 6 8
y: 25 38 56 84
9. Fit the curve of the form y  aebx to the following data:
x: 77 100 185 239 285
y: 2.4 3.4 7.0 11.1 19.6
EMPIRICAL LAWS AND CURVE-FITTING • 219

10. Growth of bacteria (N) in a culture after t hrs. is given in the following
table:
t: 0 1 2 3 4 5 6
N: 32 47 65 92 132 190 275
Fit a curve of the form N  abt and estimate N when t  7.
11. The voltage v across a capacitor at time t seconds is given by the follow-
ing table:
t: 0 2 4 6 8
v: 15 0 63 28 12 5.6
Use the method of least squares to fit a curve of the form v  aekt to this
data.
12. Obtain the least square fit of the form f(t)  ae–3t  be–2t for the data:
x: 0.1 0.2 0.3 0.4
f(t): 0.76 0.58 0.44 0.35
13. Find the most plausible values of x and y from the equations x  3y 
7.03, x  y  3.01, 2x – y  0.03, 3x  y  4.97, by forming the normal
equations.
14. Obtain the most plausible values of x, y and z from the equations:
x  2y  z  1, – x  y  2z  3, 2x  y  z  4, 4x  2y – 5z  – 7

5.9 Method of Group Averages


Let the straight line y  a  bx (1)
fit the set of n observations (x1, y1), (x2, y2), , (xn, yn) quite closely.
(Figure 5.5).

Pn(xn, yn)
Y P2(x2, y2)
Mn
P1(x1, y1) M3

+ bx M2 P3(x3, y3)
y=a M1

0
L1 L2 L3 Ln X
FIGURE 5.5

You might also like