Scribd
Upload
37 views

Tute 3

This document provides problems related to polynomial interpolation and approximation. Some key points: 1. It asks to evaluate polynomials for various functions up to a given accuracy and consider potential problems that could arise from division by small or zero values. 2. It derives formulas for divided differences and the Newton form of interpolating polynomials. It analyzes the time and space complexity of computing the divided difference coefficients and proves properties about the truncation error. 3. It analyzes the backward and forward error estimates for evaluating polynomials using Horner's method and derives error bounds involving the degree of the polynomial, spacing of points, and function values. 4. Several problems are given to estimate truncation errors, compute approximations,

Uploaded by

Rider rogue
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
37 views

Tute 3

This document provides problems related to polynomial interpolation and approximation. Some key points: 1. It asks to evaluate polynomials for various functions up to a given accuracy and consider potential problems that could arise from division by small or zero values. 2. It derives formulas for divided differences and the Newton form of interpolating polynomials. It analyzes the time and space complexity of computing the divided difference coefficients and proves properties about the truncation error. 3. It analyzes the backward and forward error estimates for evaluating polynomials using Horner's method and derives error bounds involving the degree of the polynomial, spacing of points, and function values. 4. Several problems are given to estimate truncation errors, compute approximations,

Uploaded by

Rider rogue
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

COL726 Problem set 3: Polynomials and

interpolation

January 16, 2017

1 First some more rounding and chopping


1. Consider the numbers
x1 = 0.1234 × 101
x2 = 0.3429 × 100
x3 = 0.1289 × 10−1
x4 = 0.9895 × 10−3
x5 = 0.9763 × 10−5

Add these numbers using four-decimal-digit chopped floating point


arithmetic in both forward and reverse. Which is more accurate?
Why?

2. Suggest methods for evaluating each of


x2 x3
(a) ex ' 1 + x + 2! + 3! + ...
x2 x4
(b) cos x ' 1 − 2! + 4! + ...
x3 x5
(c) sin x ' x − 3! + 5! + ...

for x = 25 up to 12 digits of accuracy. Pay attention to both truncation


and round-off errors. Try out with Matlab programs.

3. Suppose you have to evaluate

F (x) = x sin x/(1 − cos x)

near x = 0. Do you foresee any problems? Suggest a method to


overcome the problem. Again try out on Matlab.

1
2 Interpolation
1. Derive the divided difference formula again (not to be done in the tute
class).

2. Consider the following algorithm for computing dk = f [x1 , . . . , xk ],


k = 1, 2, . . . , n + 1 given the data points x1 , x2 , . . . , xn and function
values f1 , f2 , . . . , fn .
Algorithm:

Input {x1 , x2 , . . . , xn , f1 , f2 , . . . , fn }
for j = 1, 2, . . . , n + 1
dj ← fj
for k = 1, 2, . . . , n
for j = n + 1, n, . . . , k + 1
dj ← (dj − dj−1 )/(xj − xj−k )
Output {d1 , d2 , . . . , dn+1 }

Prove the correctness and determine the time and space requirements.

3. Prove the following result.


If p(x) is the interpolating polynomial which agrees with f (x) at n + 1
points in [a, b] and if f is n + 1 times continuously differentiable in
[a, b] then for any x̄ ∈ [a, b] there is a value η ∈ [a, b] such that the
truncation error is given by

ET (x̄) = f (x̄)−p(x̄) = [f (n+1) (η)/(n+1)!](x̄−x1 )((x̄−x2 ) . . . (x̄−xn+1 )

What conclusion can you draw about extrapolation/truncation errors?

4. Show that under the conditions of the previous problem, and with
Pk the polynomial that interpolates f at x1 , x2 , . . . , xk+1 , for k =
1, 2, . . . n, the difference

Pk+1 (x̄) − Pk (x̄)

is an estimate of the truncation error in Pk (x̄). This estimate is usable


whenever f (k+1) (x) does not change greatly in the interval containing
x1 , . . . , xk+2 and x̄.

2
5. Consider Horner’s method for evaluating a polynomial

Input {n, a0 , a1 , . . . , an }
Input x
p ← an
for k = n − 1, n − 2, . . . , 0
p ← x ∗ p + ak
Output {x, p}

Show that the backward error estimate is given by

p̂ = ân xn + . . . + â1 x + â0

where, 
ak h2ni k=n
âk =
ak h2k + 1i k = n − 1, . . . , 0
Conclude that if nrµ ≤ 0.1, then the relative error bound can be
obtained as
2nµ0

| âk − ak | k=n
≤
| ak | (2k + 1)µ0 k = n − 1, . . . , 0

Also, show that the forward error estimate is given by

| p̂ − p(x) |= [2n | an xn | +(2n − 1) | an−1 xn−1 | +3 | a1 x] | + | a0 |]µ0

6. Given Algorithm in 2 for computing the divided difference coefficients,


an algorithm for evaluating Newton’s formula can be given as

Input {x1 , . . . , xn+1 , d1 , . . . , dn+1 }


Input x
p ← dn+1
for i = n, n − 1, . . . , 1
p ← p ∗ (x − xi ) + di
Output {x, p}

Show that the Algorithms in 2 and 6 evaluate a polynomial p̂ that


interpolates a function fˆ at x1 , x2 , . . . , xn+1 . For all x in an interval
of length l that contain x1 , x2 , . . . , xn+1 , the following bound holds:

| f (x) − fˆ(x) |≤ 9(n3 + n2 + 1)(ln /mn )f¯µ0

3
where,
f¯ = max{| f (xi ) |: i = 1, 2, . . . , n + 1}
and
m = min{| xi − xj |: i, j = 1, 2, . . . , n + 1, i 6= j}
[Note: This one will be hard! Perhaps you should split it up into
several sub-problems?]

7. Conclude from the previous problems that the rounding errors in the
two algorithms are equivalent to a change in the finction f . This
change will be small provided that

(a) n is not too large


(b) the interval length is not large
(c) the data points x1 , x2 , . . . , xn+1 are not too close together, and
(d) the maximum value f¯ is not much larger than any other f (x) for
x in the interval.

8. Suppose that each of the following functions has been tabulated at


x1 = 0, x2 = 0.1, x3 = 2 and x4 = 0.3. Estimate the truncation error
at x̄ = 0.15 for the polynomial of degree 3 that interpolates at the
given tabulated points.

(a) f (x) = sin x


(b) f (x) = 2x3
(c) f (x) = 1/(x + 1)

9. Use Algorithm 6 to approximate f (−0.5) where f is the function


f (x) = 1/(1 + 25x2 ). Print out a table of (n, f (−0.5), ) ( being
the absolute error) for n = 1, 2, 3, 4. Also print out the truncation
error estimates Pn+1 − Pn . Verify that the derivatives of this function
change so rapidly that these error estimates are not so accurate.

10. Suppose f is a function that has been tabulated at x = 0, 0.1, 0.2, 0.3, 0.4, . . .
and suppose | f (k) (x) |≤ k, k = 0, 1, 2, . . .. What is the best value of n
to use for interpolating f at x̄ = 0.25? That is, what value of n will
result in the sum of truncation error and rounding error to be as small
as possible?

You might also like