1 Least Squares We frequently want to fit a straight line to a series of data points. i xi yi 1 x1 y1 2 y2 x2 3 y3 x3 … … … i xi yi … … … N yN xN Hence we want to fit y ( x) ax b . For each the point the error is Ei [ y ( xi ) yi ]2 [axi b yi ]2 . The total error is N N i 1 i 1 E Ei [axi b yi ]2 . Minimizing the error with respect to a and b N E 0 2 [axi b yi ]xi a i 1 N E 0 2 [axi b yi ] . b i 1 Then N a i 1 xi2 b N N xi xi y i , i 1 i 1 and N N N i 1 i 1 i 1 y 1 4 7 10 22 x2 0 1 4 9 14 a xi b 1 y i . N Note that 1 N i 1 Example: y 3 x 1 Total x 0 1 2 3 6 xy 0 4 14 30 48 2 Then the two equations for a and b are: 14a 6b 48 . 6a 4b 22 Multiplying the second by 3/2 and subtracting from the first yields 5a 15 , or a 3 . The second equation is 4b 4 , or b 1 , then y 3 x 1 which checks. Generally the solution for the coefficients (ai) when n 3 or 4 and is inaccurate for larger n. Other functional forms can be made linear. Some popular examples are: 1) y be ax can be written as ln y ax ln b which is linear in ln y and x, and 2) y bx a can be written as ln y a ln x ln b which is linear in ln y and ln x. A measure of the scatter is the variance N Ei 2 i 1 N n 1 , where n is the degree of the polynomial. For higher degree polynomials it is better to use the orthogonal Chebyshev (or Tchebyshev) polynomials, Tn (x) , of degree n. The polynomials are given by Tn ( x) cos(n ) cos 1 x (i.e., x cos ) The first few polynomials are: T0 ( x) 1, T1( x) cos x, T2 ( x) cos 2 cos 2 sin 2 2 cos 2 1 2 x 2 1, and T3 ( x) cos 3 cos( 2 ) cos 2 cos sin 2 cos (cos 2 sin 2 ) cos (2 sin cos ) sin T3 ( x) cos 3 sin 2 cos 2 sin 2 cos cos 3 3 sin 2 cos cos 3 3(1 cos 2 ) cos T3 ( x) cos 3 3 cos 3 cos 3 cos 4 cos 4 3 cos 4 x 3 3x . 3 Similarly, T4 ( x) 8 x 4 8 x 2 1 . Since TN 1 ( x) cos( N 1) cos( N ) cos sin( N ) sin and TN 1 ( x) cos( N 1) cos( N ) cos sin( N ) sin , Adding TN 1 ( x) TN 1 ( x) 2 cos cos( N ) 2 xTN ( x) results in the recursive relation TN 1 ( x) 2 xTN ( x) TN 1 ( x) . This means that the leading coefficient doubles every time N increase by one, that is TN ( x) 2 N 1 x N ... . Hence it is better to use t N ( x) TN ( x) / 2 N 1 x N ... It can be shown that of all the polynomials of degree N with a N 1 that TN ( x) / 2 N 1 has a smaller upper bound in magnitude on [-1,1] than any other polynomial of degree N. Since we know that TN ( x) cos( N ) then TN ( x) 1 and TN ( x) / 2 N 1 1 / 2 N 1 . This can be proved by contradiction. If Pn ( x) x n ... is a polynomial of degree n and has a maximum in [-1,1] less than Tn ( x) / 2 n 1 . The polynomial Tn ( x) / 2 n 1 Pn ( x) a n 1 Pn 1 ( x) is of degree n-1. But note that, including the end points, Tn(x) has n+1 relative extremes in [-1,1] equal to 1 / 2 n 1 and they alternate in sign. Since relative extremes of Pn(x) are less than 1 / 2 n 1 then Pn-1(x) crosses zero at least once for every Tn(x) extreme which is N times. Hence Pn-1(x) has n zeros but is a polynomial of degree n-1 and the assumption must be wrong, and the theorem is correct.