MATH3230B Numerical Analysis

Tutorial 11

1 Recall:
1. Newton form of interpolation:
Suppose a = x0 < x1 < · · · < xn−1 < xn = b. Then we define the Divided difference as follows:
The zeroth-order divided difference of f (x) is

f [x0 ] = f (x0 ), f [x1 ] = f (x1 ), · · · , f [xn ] = f (xn )

The first order divided difference of f (x) is

f [x1 ] − f [x0 ] f [x2 ] − f [x1 ]

f [x0 , x1 ] = , f [x1 , x2 ] = , ,··· ,
x1 − x0 x2 − x1

and similar we have the k-th order divided difference of f (x)

f [x1 , x2 , · · · , xk ] − f [x0 , x1 , · · · xk−1 ]

f [x0 , x1 , · · · , xk ] = , ,
xk − x0
The Newton form of interpolation of f (x) is

p(x) = f [x0 ] + f [x0 , x1 ](x − x0 ) + · · · + f [x0 , x1 , · · · , xn ](x − x0 )(x − x1 ) · · · (x − xn−1 )

2. Error estimates of polynomial interpolations:

Suppose f ∈ C n+1 [a, b] and p(x) is the polynomial interpolation of f (x) at the n + 1 distinct points:

a = x0 < x1 < · · · < xn−1 < xn = b

then for any x ∈ [a, b], there exists a point ζx ∈ (a, b) such that

f (n+1) (ζx )
f (x) − p(x) = (x − x0 )(x − x1 ) · · · (x − xn )
(n + 1)!

3. Chebyshev polynomials:
The Chebyshev polynomials are defined recursively as follows:

T0 (x) = 1, T1 (x) = x, · · · , Tn+1 (x) = 2xTn (x) − Tn−1 (x).

In particular, for x in the interval [−1, 1], the Chebyshev polynomials have the following closed forms for
n ≥ 0,
Tn (x) = cos(n cos−1 x), −1 ≤ x ≤ 1

If the nodes xi are chosen to be the roots of the Chebyshev polynomial Tn+1 , then the error of the resulting
interpolation polynomial for a given function f (x) will be minimized and can be estimated by
1
|f (x) − p(x)| ≤ max |f n+1 (t)|
2n (n + 1)! |t|≤1

1
2 Exercises:
Please do the star problem (*) in tutorial class and finish the rest after class.

1. *
(a) Suppose f ∈ C n+1 [a, b] and p(x) is the polynomial interpolation of f (x) at the n + 1 distinct points:

a = x0 < x1 < · · · < xn−1 < xn = b.

Please write down the error estimation of the polynomial interpolation and give the proofs.
(b) If f is a polynomial of degree k, draw a conclusion on the accuracy of the polynomial interpolation p(x).

Solution. (a) For any x ∈ [a, b], there exists a point ζx ∈ (a, b) such that

f (n+1) (ζx )
f (x) − p(x) = (x − x0 )(x − x1 ) · · · (x − xn ).
(n + 1)!

Proof: Please refer to the proof on Page 110 in lecture notes.

(b) The interpolation on n + 1 distinct points is exact for any polynomial of degree k ≤ n.

2. Recall the definition of the Chebyshev polynomial.

(a) Write down the recursive formula of Chebyshev polynomials Tn (x).
(b) If x ∈ [−1, −1], write down the closed form of the Chebyshev polynomials Tn (x).
(c) Write down the roots of the closed form Tn (x).

Solution. (a)
T0 (x) = 1, T1 (x) = x,
···
Tn+1 (x) = 2xTn (x) − Tn−1 (x).
(b)
Tn (x) = cos(n cos−1 x), −1 ≤ x ≤ 1.
(c)  
2i − 1
xi = cos π , i = 1, 2, · · · , n + 1.
2n

3. (a) Write down the definition of monic polynomials.

(b) Consider the monic polynomial p(x) = (x − x0 )(x − x1 ) · · · (x − xn ) for distinct points −1 ≤ x0 < x1 <
· · · < xn ≤ 1. Write down the estimation of kpk∞ on [−1, 1].
(c) What’s the relation of the roots in 2(c) with the above estimation of kpk∞ ?
(d) Now consider the polynomial g(x0 ) = (x0 − x00 )(x0 − x01 ) · · · (x0 − x0n ) on [a, b]. Please map the domain
of this polynomial from [a, b] onto domain [−1, 1] by linear transformation, and write down the resulted
polynomial p(x) on [−1, 1].
(e) Hence, write down the error estimation of kg(x0 )k∞ on [a, b].

Solution. (a) A monic polynomial is the one in which the term of highest degree has a coefficient of unity.
(b) Since p(x) is of degree n + 1, then

kpk∞ = max |p(x)| ≥ 2−n .

−1≤x≤1

2
 (c) If we take the value of {xi }ni=0 in polynomial p(x) to be the roots of the Chebyshev polynomial Tn+1 ,
then kp(x)k∞ can attain its minimum value 2−n . i.e. kp(x)k∞ = 2−n .
(d) Take the linear transform
2
x= (x0 − a) − 1,
b−a
then x ∈ [−1, 1] and
b−a b−a b+a
x0 = (x + 1) + a = x+ .
2 2 2
Then we have
 
0 b−a b+a
p(x) = g(x ) = g x+
2 2
    
b−a b+a 0 b−a b+a 0 b−a b+a 0
= x+ − x0 x+ − x1 · · · x+ − xn
2 2 2 2 2 2
b+a 2 0
Take xi = b−a − b−a xi , then
 n+1
0 b−a
p(x) = g(x ) = (x − x0 )(x − x1 ) · · · (x − xn ), x ∈ [−1, 1].
2

(e)

kg(x0 )k∞ = max |g(x0 )| = max |p(x)|

a≤x0 ≤b −1≤x≤1
 n+1  n+1
b−a b−a
= max |(x − x0 )(x − x1 ) · · · (x − xn )| ≥ 2−n .
2 −1≤x≤1 2

4. (a) Let
sin ((n + 2) θ)
Un+1 = , x = cos(θ)
sin(θ)
i. Show that
U0 (x) = 1, U1 (x) = 2x, Un+1 (x) = 2xUn − Un−1
ii. Show that Tn0 = nUn−1 where Tn (x) = cos(n cos−1 (x))
(b) Consider the Chebyshev polynomial

Tn (x) = cos(n cos−1 (x)), x ∈ [−1, 1],

i. Let p be a monic polynomial of degree n. Show that

max |p(x)| ≥ 21−n

−1≤x≤1

ii. Let f be a smooth function, {xi }ni=0 the roots of Chebyshev polynomials Tn+1 , and p the interpolation
polynomial of f at nodes {xi }ni=0 . Show that

1
|f (x) − p(x)| ≤ max |f (n+1) (t)|, x ∈ [−1, 1]
2n (n + 1)! |t|≤1

Solution. (a) i.
sin(θ)
U0 (x) = =1
sin(θ)
sin(2θ)
U1 (x) = = 2 cos(θ) = 2x.
sin(θ)

3
 sin((n + 2)θ) sin((n + 2)θ) + sin(nθ) − sin(nθ)
Un+1 (x) = =
sin(θ) sin(θ)
2 sin((n + 1)θ) cos(θ) − sin(nθ)
=
sin(θ)
= 2xUn − Un−1

ii. Since Tn (x) = cos(n cos−1 (x)),

we have
−1
Tn0 (x) = − sin(n cos−1 (x)) · n · √
1 − x2
n sin(n cos−1 (x))
= √
1 − x2
n sin(nθ)
=
sin(θ)
= nUn−1

(b) i. We prove by contradiction. Suppose that

|p(x)| < 21−n , (|x| ≤ 1)

Let q(x) = 21−n Tn (x) and xi = cos iπ

n As q(x) is a monic polynomial of degree n, then

(−1)i p(xi ) ≤ |p(xi )| < 21−n = (−1)i q(xi )

So
(−1)i [q(xi ) − p(xi )] > 0 (0 ≤ i ≤ n)
This shows that the polynomial q(x) − p(x) oscillates in sign n + 1 times on the interval [−1, 1].
Therefore it must have at least n roots in (−1, 1). But this is impossible because q(x) − p(x) has
degree at most n − 1.
ii. By the interpolation error estimate, we have
1
max |f (x) − p(x)| ≤ max |f ( n + 1)(x)| max |(x − x0 )(x − x1 ) · · · (x − xn )|
|x|≤1 (n + 1)! |x|≤1 |x|≤1

By the property of the monic function in part (a), we have

max |(x − x0 )(x − x1 ) · · · (x − xn )| ≥ 2−n

|x|≤1

This minimum value is attained if (x − x0 )(x − x1 ) · · · (x − xn ) is the monic multiple of Tn+1 , i.e.,
2−n Tn+1 . The nodes are the roots of Tn+1 , namely
 
2i − 1
xi = cos π , i = 1, 2, · · · n + 1.
2n + 2

Combining the above results, we have

1
|f (x) − p(x)| ≤ max |f (n+1) (t)|, x ∈ [−1, 1]
2n (n + 1)! |t|≤1