Minmax Polynomial, Min-Max Approximation
Minmax Polynomial, Min-Max Approximation
• h= –(B1y1–B2y2+B3y3)/(B1+B2+B3)
Example 1:
Find the equal error line for the data points (0,0),
(1,0) and (2,1).
Solution:
B1=2–1=1 B2=2–0=2 B3=1–0=1
h= –[1(0) –2(0)+1(1)]/[1+2+1]= –1/4
DISCRETE CASE
Continuation of Example 1:
Therefore the line passes through (0,–1/4),
(1,1/4) and (2,3/4). By point-slope formula
using any two of the points, the equation of the
line is P(x) = (1/2)x –1/4.
DISCRETE CASE
Example 2: The Exchange Method
Find the equal error line for the following points:
(0,0), (1,0), (2,1), (6,2), (7,3).
Step 1. Select an initial triple and find the
Chebyshev line for this triple.
Let’s choose (0,0), (1,0) and (2,1); and from
Example 1, the Chebyshev line is P(x) = (1/2)x –
1/4, where h= –1/4.
Step 2: Compute the errors at all data points. Call
the absolute value of the largest of these errors as
H. If the absolute value of h=H, then the search is
over, else go to Step 4.
DISCRETE CASE
Continuation of Example 2:
The errors at all five data points are –1/4, 1/4, –1/4,
3/4, 1/4, respectively.
This makes H=3/4, which is not equal to the absolute
value of h=1/4.
Step 4: Exchange step: Choose a new triple by adding to
the old triple a data point at which H occurs. Then
disregard one of the former points, in such a way that
the remaining three have errors of alternating sign.
The new triple is (1,0), (2,1) and (6,2). By doing Step
1 to 3 again, we will find out that the three points
will lead us to the min-max line of the entire data set
which is P(x)=(2/5)x – (1/10).
DISCRETE CASE
The Min-Max Parabola
The algorithm will start by choosing an initial
quadruple.
The equal error parabola of the quadruple is
P(x)=a+bx+cx2, where P(xi)–y(xi)=+h.
The same Steps 1-4 of the exchange method
will be followed in obtaining the Min-Max
Parabola of the entire given data set.
Example 3:
Find the min-max parabola for (-2,2), (-1,1),
(0,0), (1,1) and (2,2).
DISCRETE CASE
Continuation of Example 3:
Solution:
Let’s choose (-2,2), (-1,1), (0,0) and (1,1).
P(xi) –y(xi)=+h will result to the following
systems of equation:
(a–2b+4c)–2 = h
(a–b+c)–1 = –h
a–0=h
(a+b+c)–1 = –h
Then solve for a, b, c and h. Then do the
Exchange Method.
DISCRETE CASE
Continuation of Example 3:
T0(x) = 1
T1(x) = x
T2(x) = 2x2–1
T3(x) = 4x3–3x
Its recurrence relation can be written as
b a a b
xk tk
2 2
where tk cos ( 2 N 1 2k )
for k=0,1,…,N
CONTINUOUS CASE
LAGRANGE-CHEBYSHEV APPROXIMATION POLYNOMIAL
Assume that PN(x) is the Lagrange polynomial that is
based on the Chebyshev nodes given in the previous slide,
If f belongs to CN+1[a,b], then
N 1
2( b a ) ( N 1)
f ( x ) PN ( x ) N 1
max{ f (x) }
4 ( N 1 )! a x b
Example: For f(x)=sin(x) on [0,π/4], find the Chebyshev
nodes and the error bound for the Lagrange Polynomial
P5(x).
( 11 2k )
xk cos k
12
sin( x ) P5 ( x ) 0.00000720
CONTINUOUS CASE
A direct approach (with the use of the Orthogonal Property
of Chebyshev Polynomials):
-END-