Title | Curve fitting - Brief lecture notes |
---|---|
Course | Probability, Statistics And Numerical Methods |
Institution | APJ Abdul Kalam Technological University |
Pages | 7 |
File Size | 184 KB |
File Type | |
Total Downloads | 39 |
Total Views | 149 |
Brief lecture notes...
Discussion 4 & 5 1
CURVE FITTING Let (x1 , y1 ), (x2 , y2 ) ,…, (xn , yn ) be n pairs of observations of related data. The general problem of finding a relation of the form y = f (x) which fits best to the given data is called curve fitting.
1.1 Principle of least squares Let (x1 , y1 ), (x2 , y2 ) ,…, (xn , yn ) be n pairs of observations of related data. Let y = f (x) be the best fit curve . yi is called the observed value y corresponding to x i and f (xi) is called the expected value of y corresponding to xi . yi – f (xi) is called the error or residual for yi .The principle of least squares states that for a best fit curve , the sum of squares of residuals is n
minimum. ie., E y i f ( x i )2 is minimum . i 1
1.2
Fitting a straight line
Let y = ax + b be a best fit straight line to the n pairs of observations (x1 , y1 ), (x2 , y2 ) ,…, (xn , yn ) of related data .For any xi , the observed value is yi and the expected value is axi +b so that the error = yi –(axi +b) The sum of squares of errors is n
E y i f ( x i )2 y 1 (ax 1 b )2 y 2 (ax 2 b )2 i 1
... y n (ax n b ) For the best fit , by Principle of least squares , E is minimum. E E 0 For E to be minimum 0 and a b E 0 2 y 1 (ax 1 b )( x 1 ) 2y 2 (ax 2 b )( x 2 ) a .... 2y n (ax n b ) ( x n ) 0
2
2 2 2 [x 1 y 1 ax 1 bx1 ] [ x 2 y 2 ax 2 bx 2 ] ........ [ x n y n ax n bx n ] 0
x 1 y1 x 2 y 2 ........ x n y n ax 21 ax 22 ........ ax 2n bx1 bx 2 ... bx n n
n
n
x i y i a x 2i b x i . .......(1) i 1
i 1
i 1
E 0 2 y1 (ax1 b ) ( 1) 2y 2 ( ax 2 b ) (1) .... 2 yn ( axn b ) ( 1) 0 b [ y 1 ax 1 b] [ y 2 ax 2 b ] ........ [ y n ax n b] 0
y 1 y 2 ..... y n ax 1 ax 2 ....... ax n b b ........ b n
n
y i a x i nb i 1
........(2)
i 1
The equations (1) and (2) are known as Normal equations and can be solved as simultaneous equations in a and b .
Example 1 By the method of least squares find the best fitting straight line to the data x : 5 10 15 20 25 y : 15 19 23 26 30 Solution Let y = ax+b be the best fit straight line. The nomal equations are y a x nb and
xy a x
2
b x
x y xy x2 5 15 75 25 10 19 190 100 15 23 345 225 20 26 520 400 25 30 750 625 75 113 1880 1375 The normal equations becomes 75a +5b = 113 1375a+75b = 1880 Solving , we get a = 0.74 , b = 11.5 Required line is y = 0.74x +11.5 Example 2 Fit a straight line of the form y = a + bx for the data given below x: 1 2 3 4 5 y : 5 7 9 10 11 Solution Lety = a+bx be the best fit straight line. The nomal equations are y na b x and
xy a x b x
2
x y xy x2 1 5 5 1 2 7 14 4 3 9 27 9 4 10 40 16 5 11 55 25 15 42 141 55 The normal equations becomes 5a+15b = 42 15a+55b = 141 Solving , we get a = 3.9 , b = 1.5 Required line is y = 3.9+ 1.5x
Fitting of parabola y = a x2 + b x + c
1.3
Let y = ax2 + bx + c be a best fit parabola to the n pairs of observations (x1 , y1 ), (x2 , y2 ) ,…, (xn , yn ) of related data .For any xi , the observed value is yi and the expected value is ax 2i bx i c so that the error = y i (ax 2i bx i c) The sum of squares of errors is n
2
E y i f ( x i )2 y 1 (ax 21 bx 1 c) y 2 (ax 22 bx 2 c )
2
i 1
.... y n (ax 2n bx n c) For the best fit , by Principle of least squares , E is minimum. E E E For E to be minimum 0, 0 and 0 c a b E 0 2 y 1 (ax 21 bx 1 c) ( x 21 ) 2 y 2 (ax 22 bx 2 c ) ( x 22 ) a .... 2 y n (ax 2n bx n c) ( x2n ) 0
2
y 1x 21 ax 14 bx 31 cx12 y 2 x 22 ax 24 bx 3 2 c x 22 .... y n x 2n ax 4n bx 3 n cx 2n 0 y1 x 21 y2 x22 ..... yn x2n ax41 ax42 .... axn4 bx31 bx 3 2 ..... bx3 n cx12 c x 22 ..... cx 2n n
n
n
n
y i x 2i a x 4i b x 3i c x 2i ..........(1) i 1
i 1
i 1
i 1
E 0 2 y1 (ax12 bx1 c) ( x1 ) 2 y 2 (ax 22 bx2 c ) ( x2 ) b .... 2 y n (ax 2n bx n c) ( x n ) 0
3 1
y 1x 1 ax bx
2 1
3 2
cx 1 y 2x 2 ax bx
2
2
c x 2 .... y nx n ax 3n bx 2 n cx n 0
y 1x 1 y 2x 2 ..... y n x n ax 31 ax 23 .... ax n bx 21bx 2 2 ..... bx 2 n cx 1 c x 2 ..... cx n n
i 1
n
n
n
i 1
i 1
i 1
yi xi a x3i b xi 2 c xi ..........(2)
E 0 2 y1 (ax12 bx1 c) ( 1) 2 y 2 (ax 22 bx 2 c ) ( 1) c .... 2 y n (ax 2n bx n c ) ( 1) 0
n
n
n
y i a x 2i b x i nc ..........(3) i 1
i 1
i 1
The equations (1) ,(2) and (3) are known as Normal equations and can be solved as simultaneous equations in a ,b and c .
Example 1 Find a second degree curve for the following data x: 1 2 3 4 5 y : 3 9 13 21 31 Solution Let y = a x2 + b x + c be the fitted second degree curve . The normal equations are y a x 2 b x nc 2
xy a x b x c x x y a x b x c x 3
2
3
4
2
x y x2 x3 x4 1 3 1 1 1 2 9 4 8 16 3 13 9 27 81 4 21 16 64 256 5 31 25 125 625 sum=15 77 55 225 979 The normal equations becomes 55a+15b+5a = 77 225a+55b+15c = 299 979a+225b+55c = 1267 Solving , we get a = 1.01 , b = 0.744 , c= 2.058
xy 3 18 39 84 155 299
x2 y 3 36 117 336 775 1267
Example 2 Fit a second degree curve of the form y = a+bx+cx2 to the following data x : 1911 1912 1913 1914 1915 y : 10 12 8 10 14 Solution Let u = x -1913 Let y = a+bu+cu2 be the parabola of best fit. The normal equations are y a u 2 b u nc
uy a u b u c u u y a x b u c u 3
2
2
4
3
x u = x- 1913 u2 1911 -2 4 1912 -1 1 1913 0 0 1914 1 1 1915 2 4 Sum= 0 10 The normal equations becomes 5a+0b+10c = 54 0c+10b+0c = 6 10a+0b+34c = 118
u3 -8 -1 0 1 8 0
2
u4 16 1 0 1 16 34
y 10 12 8 10 14 54
uy -20 -12 0 10 28 6
u2y 40 12 0 10 56 118
Solving , we get a = 9.37 , b = 0.6 , c = 0.714 The fitted parabola is y = 9,37+0.6u+0.714u2 = 9,37+0.6 (x-1913)+0.714(x-1913)2 Fitting of parabola y = a x2 + b
1.4
Let y = ax2 + b be a best fit parabola to the n pairs of observations (x1 , y1 ), (x2 , y2 ) ,…, (xn , yn ) of related data .For any xi , the observed value is yi and the expected value is ax 2i b so that the error = y i (ax 2i b) The sum of squares of errors is n
2
E y i f ( x i )2 y1 (ax 12 b ) y 2 (ax 22 b )
2
i 1
2
.... y n (ax 2n b ) For the best fit , by Principle of least squares , E is minimum. E E 0 and 0 For E to be minimum b a E 0 2 y 1 (ax 12 b ) (x 21 ) 2 y 2 (ax 22 b ) (x 22 ) .... 2 y n (ax 2n b ) (x 2n ) 0 a [x12 y 1 ax 14 bx 21 ] [x 22 y 2 ax 42 bx 22 ] ........ [x n2y n ax n4 bx n2 ] 0
x12 y 1 x 22 y 2 ........ x n2 y n ax14 ax24 ........ ax 4n bx 21 bx22 bx n2 n
n 2 i
x i 1
n 4 i
yi a x b xi2 . .......(1) i 1
i 1
E 0 2 y1 (ax12 b ) (1) 2 y2 (ax22 b ) (1) .... 2 y n ( ax2n b ) ( 1) 0 b y1 y2 ........ yn ax12 ax22 ... ax2n b b .... b
n
n
yi a xi2 nb . .......(2) i1
i 1
The equations (1) and (2) are known as Normal equations and can be solved as simultaneous equations in a and b Example Fit a parabola of the form y = a + bx2 for the data given below x: 1 2 3 4 5 y : 0.43 0.83 1.4 2.33 3.42 Solution Let y = a + bx2 be the fitted curve . The normal equations are y na b x 2
x x 1 2
2
y ax 2 bx 4 y 0.43 0.83
x2 1 4
x4 1 16
x2 y 0.43 3.32
3 1.4 9 81 12.6 4 2.33 16 256 37,28 5 3.42 25 625 85.5 Sum= 8.41 55 979 139.13 The normal equations becomes 5a+55b = 8.41 55a+979b = 139.13 Solving , we get a = 0.307 , b = 0.125 Therefore the required equation is y = 0.307 + 0.125 x2 1.4
Fitting of the curve y = a ebx
Given y = a ebx Taking logarithm on both sides, we get log y = log a + bx Let Y = log y and A = log a . Then the above equation becomes Y = A + bx The normal equations are Y nA b x and
xY A x
b x 2 Example Fit a curve of the form y = a ebx to the following data x: 0 2 4 y : 5.012 10 31.62 Solution Given y = a ebx Taking logarithm on both sides, we get log y = log a + bx Let Y = log y and A = log a . Then the above equation becomes Y = A + bx The normal equations are Y nA b x and
xY A x
b x 2
x y Y = logy 0 5.012 1.62 2 10 2.3 4 31.62 3.45 Sum=6 46.632 7.37 The normal equations becomes 3A+6b =7.37 6A+20b = 18.4 Solving , we get A = 1.534 , b = 0.46 A = 1.534 a = 4.64 Therefore the required equation is y = 4.64 e0.46x
x2 0 4 16 20
xY 0 4.6 13.8 18.4...