Module

Department of Pure and Applied Mathematics

Numerical Methods and
Statistical Techniques (V SEM)

Curve Fitting
***

4.1 Curve fitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

4.1.1 Curve fitting by least squares method . . . . . . . . . . . . . . . . . . . . . . 2

4.1.2 Fitting of a straight line y = ax + b . . . . . . . . . . . . . . . . . . . . . . . . 2

4.1.3 Fitting of a second degree parabola y = ax2 + bx + c . . . . . . . . . . . . . . 5

4.1.4 Fitting of a curve of the form y = aebx . . . . . . . . . . . . . . . . . . . . . . 8

4.1.5 Fitting of a curve of the form y = axb . . . . . . . . . . . . . . . . . . . . . . 8

4.1 Curve fitting

The process of constructing an approximate curve y = f (x), which fit best to a given discrete set
of points (xi , yi ), i = 1, 2, 3, ..., n is called curve fitting. Curve fitting and interpolation are closely
associated procedures. In interpolation, the fitted function should pass through all given data points;
whereas curve fitting methodologically fits a unique curve to the data points, which may or may not
lie on the fitted curve. The difference between interpolation and curve fitting; while attempting to
fit a linear function; is illustrated in the adjoining figure.
 2

4.1.1 Curve fitting by least squares method

Curve fitting is a method of finding a suitable relation or law in the form y = f (x) for a set of
observed values (xi , yi ), i = 1, 2, . . . , n.
Such a relation connecting x and y is known as empirical law.
This relation is most suitable for predicting/estimating the value of y for a given value of x. The
method of least squares is as follows.
Suppose y = f (x) is an approximate relation that fits into a given data comprising (xi , yi ),
i = 1, 2, . . . , n, then yi ’s are called observed values and Yi = f (xi ) are called expected values. Their
difference Ri = yi − Yi are called the residuals or estimate errors.
The method of least squares provides a relationship y = f (x) such that the sum of the squares
of the residuals is least.

4.1.2 Fitting of a straight line y = ax + b

Consider a set of n given values (x, y) for fitting the straight line y = ax + b where a and b are
parameters to be determined. The residual R = y − (ax + b) is the difference between the observed
and estimated values of y. By the method of least squares we find parameters a and b such that the
sum of squares of the residuals is minimum (least).
Let n
R2
X
S=
1
n
[y − (ax + b)]2
X
i.e., S =
1
Treating S as a function of two parameters a and b, the necessary conditions for S to be minimum
∂S ∂S
are = 0 and = 0. i.e.,
∂a ∂b n X
2 [y − (ax + b)](−x) = 0
1
and n
X
2 [y − (ax + b)](−1) = 0
1
Dividing both the equations by 2, we have
n n n
ax2 +
X X X
− xy + bx = 0
1 1 1
n
X n
X n
X
− y+ ax + b=0
1 1 1
n
X
But b = b + b + b + . . . n times = nb and hence we have
1

x2 + b
X X X
a x= xy
X X
a x + nb = y

These equations are called normal equations for fitting the straight line y = ax+b in the least squares
sense. By solving these, we obtain the value of a and b.
 3

Example 4.1.1. Fit a straight line y = ax + b for the following data

x 1 3 4 6 8 9 11 14
y 1 2 4 4 5 7 8 9

Solution. The normal equations for fitting the straight line y = ax + b are
X X
y=a x + nb
x2 + b
X X X
xy = a x (n = 8)

The relevant table is as follows.

x y xy x2
1 1 1 1
3 2 6 9
4 4 16 16
6 4 24 36
8 5 40 64
9 7 63 81
11 8 88 121
14 9 126 196
P P P P 2
x = 56 y = 40 xy = 364 x = 524
The normal equations become

56a + 8b = 40
524a + 56b = 364

On solving (Using calculator) we have,

a = 0.636363636 ≈ 0.64
b = 0.545454545 ≈ 0.55

Thus by substituting these values in y = ax + b we obtain the equation,

y = 0.64x + 0.55

Example 4.1.2. Find the equation of the best fitting straight line for the following data and hence
estimate the value of the dependent variable corresponding to the value 30 of the independent
variable.
x 5 10 15 20 25
y 16 19 23 26 30

Solution. Let y = ax + b be the equation of the best fitting straight line. The associated normal
equations are as follows.
X X
y=a x + nb
x2 + b
X X X
xy = a x (n = 5)
 4

x y xy x2
5 16 80 25
10 19 190 100
15 23 345 225
20 26 520 400
25 30 750 625
P P P P 2
x = 75 y = 114 xy = 1885 x = 1375

The normal equations become

75a + 5b = 114
1375a + 75b = 1885

On solving we have,

a = 0.7
b = 12.3

Thus by substituting these values in y = ax + b we obtain the equation of the best fitting straight
line in the form y = 0.7x + 12.3 Further when x = 30 we obtain y = 0.7(30) + 12.3 = 33.3.
Example 4.1.3. Fit a straight line to the following data.
Year 1961 1971 1981 1991 2001
Production 8 10 12 10 16
(in tons)
Also find the expected production in the year 2006.
Solution. Let year and production respectively be represented by the variables x and y. We shall
fit the straight line in the form y = a + bx. Since the values of x are large in magnitude, we may
prefer to modify the same by choosing a convenient origin somewhere in the middle.
X = x − 1981 and the line of fit will be y = a + bX
The normal equations associated with y = a + bX are as follows.

X X
y = na + b X
X2
X X X
Xy = a X +b (n = 5)

The relevant table is as follows

X y Xy X2
-20 8 -160 400
-10 10 -100 100
0 12 0 0
10 10 100 100
20 16 320 400
X 2 = 1000
P P P P
X=0 y = 56 Xy = 160
 5

The normal equations become,

5a = 56
1000b = 160
i.e. a = 11.2 and b = 0.16
Hence y = a + bX, with X = x − 1981 becomes

y = 11.2 + 0.16(x − 1981)

Thus y = −305.76 + 0.16x is the required line of fit.

Also when x = 2006, y = −305.76 + 0.16(2006) = 15.2
Expected production in the year 2006 is 15.2 tons

4.1.3 Fitting of a second degree parabola y = ax2 + bx + c

Consider a set of n given values (x, y) for fitting the curve y = ax2 + bx + c.
The residual R = y − (ax2 + bx + c) is the difference between the observed and estimated values
of y. We have to find parameters a, b, c such that the sum of the squares of the residuals is the least.
Let n
[y − (ax2 + bx + c)]2
X
S=
1

Treating S as a function of three parameters a, b, c, the necessary conditions for S to be minimum

∂S ∂S ∂S
are = 0, = 0, = 0.
∂a ∂b ∂c
n
[y − (ax2 + bx + c)](−x2 ) = 0
X
i.e., 2
1
n
[y − (ax2 + bx + c)](−x) = 0
X
2
1
n
[y − (ax2 + bx + c)](−1) = 0
X
2
1

Dividing all these equations by 2, we have,

n n n n
2 4 3
cx2 = 0
X X X X
− x y+ ax + bx +
1 1 1 1
n n n n
ax3 + bx2 +
X X X X
− xy + cx = 0
1 1 1 1
n n n n
2
X X X X
− y+ ax + bx + c=0
1 1 1 1

n
X
But c = c + c + c + ...n times = nc and hence we have
1

x4 + b x3 + c x2 = x2 y
X X X X
a
x3 + b x2 + c
X X X X
a x= xy
x2 + b
X X X
a x + nc = y
 6

These equations are called normal equations for fitting the second degree parabola y = ax2 +bx+c
in the least square sense. By solving these we obtain the value of a, b, c.

Note 4.1.1. The normal equations for fitting a straight line or parabola can be written instantly
from the desired equation of the curve as follows.
P
We first apply summation ( ) to the desired equation keeping the constants a, b, c outside the
P P P
summation where the summation of pure constant terms like a, b, c are to be written as
na, nb, nc... respectively.
We then multiply the given equation by the independent variable x and apply summation again.
This will suffice for fitting a straight line. However in the case of parabola we must also multiply by
x2 and apply summation.

Example 4.1.4. Fit a second degree parabola y = ax2 + bx + c in the least square sense for the
following data and hence estimate y at x = 6.
x 1 2 3 4 5
y 10 12 13 16 19

Solution. The normal equations associated with y = ax2 + bx + c are as follows.

x2 + b
X X X
y=a x + nc
x3 + b x2 + c
X X X X
xy = a x
x2 y = a x4 + b x3 + c x2
X X X X
(n = 5)

The relevant table is as follows.

x y xy x2 y x2 x3 x4
1 10 10 10 1 1 1
2 12 24 48 4 8 16
3 13 39 117 9 27 81
4 16 64 256 16 64 256
5 19 95 475 25 125 625
P P P P 2 P 2 P 3 P 4
x = 15 y = 70 xy = 232 xy = 906 x = 55 x = 225 x = 979
The normal equations become

55a + 15b + 5c = 70
225a + 55b + 15c = 232
979a + 225b + 55c = 906

On solving we have

a = 0.2857 ≈ 0.29
b = 0.4857 ≈ 0.49
c = 9.4

Thus the required second degree parabola is

 7

y = 0.29x2 + 0.49x + 9.4

Also at x = 6, y = 22.78
Example 4.1.5. Fit a parabola for the following data.

x 1 2 3 4 5 6 7 8 9
y 2 6 7 8 10 11 11 10 9
Solution. Let y = a + bx + cx2 be the parabola of fit. The associated normal equations are

x2
X X X
y = na + b x+c
x2 + c x3
X X X X
xy = a x+b
x2 y = a x2 + b x3 + c x4
X X X X
(n = 9)

The relevant table is as follows.

x y xy x2 y x2 x3 x4
1 2 2 2 1 1 1
2 6 12 24 4 8 16
3 7 21 63 9 27 81
4 8 32 128 16 64 256
5 10 50 250 25 125 625
6 11 66 396 36 216 1296
7 11 77 539 49 343 2401
8 10 80 640 64 512 4096
9 9 81 729 81 729 6561
P P P P 2 P 2 P 3 P 4
x = 45 y = 74 xy = 421 xy = x = 285 x = 2025 x =
2771 15333

The normal equations become

9a + 45b + 285c = 74
45a + 285b + 2025c = 421
285a + 2025b + 15333c = 2771

On solving we have,

a = −0.92857 ≈ −0.93
b = 3.52316 ≈ 3.52
c = −0.26731 ≈ −0.27

Thus the required parabola of fit is

y = −0.93 + 3.52x − 0.27x2

 8

4.1.4 Fitting of a curve of the form y = aebx

Consider y = aebx . Taking logarithms (to the base e) on both sides we get

loge y = loge a + bx loge e

But loge e = 1, so

loge y = loge a + bx

or
Y = A + BX (4.1.1)

where Y = loge y, A = loge a, B = b, X = x.

It is evident that (4.1.1) is the equation of a straight line and the associated normal equations
are as follows.

X X
Y = nA + B X (4.1.2)
X2
X X X
XY = A X +B (4.1.3)

Solving (4.1.2) and (4.1.3) we obtain A and B. But loge a = A ⇒ a = eA ; Also b = B. Substituting
these values in y = aebx we get the curve of best fit, in the required form.

4.1.5 Fitting of a curve of the form y = axb

Consider y = axb . Taking logarithms (to the base e) on both sides we get,

loge y = loge a + b loge x

or
Y = A + BX (4.1.4)

where Y = loge y, A = loge a, B = b, X = loge x.

The normal equations associated with (4.1.4) are same as in the previous case and we can obtain
a, b. Substituting a, b in y = axb we get the curve of best fit in the required form.
Example 4.1.6. Fit a curve of the form y = aebx for the data
x 0 2 4
y 8.12 10 31.82
Solution. Consider y = aebx
∴ loge y = loge a + bx loge e. But loge e = 1

Denoting Y = loge y, A = loge a, we have, Y = A + bx, which is a straight line.

The associated normal equations are as follows.
X X
Y = nA + b x
x2
X X X
xY = A x+b (n = 3)
 9

x y Y = loge y xY x2
0 8.12 2.0943 0 0
2 10 2.3026 4.6052 4
4 31.82 3.4601 13.8404 16
P P P P 2
x=6 Y = 7.8570 xY = 18.4456 x = 20

The relevant table is as follows.

The normal equations become

3A + 6b = 7.8570
6A + 20b = 18.4456

On solving we have,

A = 1.9361
b = 0.34145 ≈ 0.3415

A = loge a = 1.9361 ⇒ a = eA = e1.9361 = 6.9317

The curve of fit is y = aebx

Thus y = (6.9317)e0.3415x is the curve of fit.

Example 4.1.7. Fit an exponential curve of the form y = aebx by the method of least squares for
the following data

No. of petals 5 6 7 8 9 10
No. of flowers 133 55 23 7 2 2

Solution. Note: The preliminary steps are to be retraced as in the previous problem. We shall
prepare the relevant table with reference to the same. (n = 6)

x y Y = loge y xY x2
5 133 4.8903 24.4515 25
6 55 4.0073 24.0438 36
7 23 3.1355 21.9485 49
8 7 1.9459 15.5672 64
9 2 0.6931 6.2379 81
10 2 0.6931 6.9310 100
P P P P 2
x = 45 Y = 15.3652 xY = 99.1799 x = 355

The normal equations become

6A + 45b = 15.3652
45A + 355b = 99.1799
 10

On solving we have,

A = 9.4433
b = −0.9177

loge a = A ⇒ a = eA = e9.4433 = 12623.3 Thus the required curve of fit is y = (12623.3)e−0.9177x

Example 4.1.8. Fit a least square geometric curve y = axb for the following data.
x 1 2 3 4 5
y 0.5 2 4.5 8 12.5

Solution. Consider y = axb

loge y = loge a + b loge x and let Y = loge y, A = loge a, X = loge x.

The normal equations associated with Y = A + bX are as follows.

X X
Y = nA + b X
X2
X X X
XY = A X +b (n = 5)

The relevant table is as follows.

x y X = loge x Y = loge y XY X2
1 0.5 0 -0.6931 0 0
2 2 0.6931 0.6931 0.4804 0.4804
3 4.5 1.0986 1.5041 1.6524 1.2069
4 8 1.3863 2.0794 2.8827 1.9218
5 12.5 1.6094 2.5257 4.0649 2.5902
X 2 = 6.1993
P P P P
X = 4.7874 Y = 6.1092 XY =
9.0804

The normal equations become

5A + 4.7874b = 6.1092
4.7874A + 6.1993b = 9.0804

On solving we have, A = −0.69315, b = 2

loge a = A ⇒ a = eA = e−0.69315 = 0.5

Thus y = 0.5x2 is the required curve of fit.

Example 4.1.9. Fit a curve of the form y = abx for the data and hence find the estimation for y
when x = 8.
x 1 2 3 4 5 6 7
y 87 97 113 129 202 195 193
 11

Solution. Consider y = abx

loge y = loge a + x loge b
Let Y = loge y, A = loge a, B = loge b and hence we have the equation Y = A + Bx.
The associated normal equations are as follows.

X X
Y = nA + B x
x2
X X X
xY = A x+B (n = 7)

The relevant table is as follows.

x y Y = loge y xY x2
1 87 4.4659 4.4659 1
2 97 4.5747 9.1494 4
3 113 4.7274 14.1822 9
4 129 4.8598 19.4392 16
5 202 5.3083 26.5415 25
6 195 5.2730 31.6380 36
7 193 5.2627 36.8389 49
P P P P 2
x = 28 Y = 34.4718 xY = 142.2551 x = 140

The normal equations become

7A + 28B = 34.4718
28A + 140B = 142.2551

On solving (Using calculator) we have A = 4.3, B = 0.156

A = loge a ⇒ a = eA = e4.3 = 73.7

B = loge b ⇒ b = eB = e0.156 = 1.169
Thus the required curve of fit is y = 73.7(1.169)x
Also when x = 8, y = 73.7(1.169)8 = 257.03

Problem 4.1.1. Find the equation of the best fitting straight line for the following data[1
to 6]

x 1 2 3 4 5
1.
y 14 13 9 5 2

x 0 1 2 3 4 5
2.
y 9 8 24 28 26 20
 12

x 0 1 2 3 4 5 6
3.
y 2 1 3 2 4 3 5

x 62 64 65 69 70 71 72
4.
y 65.7 66.8 67.2 69.3 69.8 70.5 70.9

x 1 2 3 4 5 6 7
5.
y 80 90 92 83 94 99 92
6.
Year (x) 1911 1921 1931 1941 1951
Production 8 10 12 10 6
(y) (in
thousand
tons)
Fit a parabola of second degree for the following data [7 to 10]

x 0 1 2 3 4 5 6
7.
y 14 18 23 29 36 40 46

x 10 20 30 40 50 60
8.
y 157 179 210 252 302 361

x 0 1 2 3 4
9.
y 1 5 10 22 38

x 1 2 3 4 5
10.
y 25 28 33 39 46

11. Fit a curve of the form y = abx for the data and hence estimate y when x = 8.
x 0 1 2 3 4 5 6
y 32 47 65 92 132 190 275

12. Fit a curve of the form y = axb for the data

x 1 2 3 4 5 6
y 2.98 4.26 5.21 6.1 6.8 7.5

13. Find the equation of the best fitting curve in the form y = aebx for the data
x 0 2 4
y 5.02 10 31.62

ANSWERS:

1. y = −3.2x + 18.2 3. y = 0.5x + 1.36

2. y = 3.23x + 11.096 4. y = 0.52x + 33.46

 13

5. y = 2x + 82 10. y = 0.64x2 + 1.46x + 2.78

6. y = 0.16x − 302.56
11. y = 32.15(1.43)x , 562
7. y = 0.083x2 + 4.96x + 13.46
12. y = 2.98x0.31
8. y = 0.046x2 − 0.84x + 143.67

9. y = 2.23x2 + 0.18x + 1.46 13. y = 4.64e0.46x