MT414: Numerical Analysis

Homework 2
Answers

1. Let a = 0.96 and b = 0.99.

a+b
(a) Using two-digit rounding arithmetic, compute .
2
b−a
(b) Using two-digit rounding arithmetic, compute a + .
2
a+b
(c) Which of these two values is a better approximation to the actual value of ?
2
Answer: (a) We compute that a + b rounds to 2.0 using two-digit rounding arithmetic,
a+b
and therefore rounds to 1.0.
2
b−a
(b) Now, we compute that rounds to 0.015, and a + 0.15 rounds to 0.98.
2
(c) The result in (b) is considerably better than that in (a), because it is between the
values of a and b. The other result could lead to serious errors.

2. Find the rates of convergence of the following functions as n → ∞:

1 1
a. lim sin = 0 b. lim sin 2 = 0
n→∞
 n 2 n→∞ n
1
c. lim sin =0 d. lim log(n + 1) − log(n) = 0
n→∞ n n→∞

Answer: The first three problems can be answered much more easily if we know that
sin x ≤ x for 0 ≤ x < 1.
(Much more than this is true, but this inequality

suffices.)
1 As a
1 1 1 1 1
result,
we have | sin n | < n , and so sin n = O( n ). Similarly, | sin n2 | < n2 , and so
 
1
sin n2 = O( n2 ). We also can take the inequality | sin n | < n1 and square both sides,
1 1



2
giving | sin n1 |2 < n12 , and therefore sin n1 = O( n12 ).

The last one is a bit more interesting. We rewrite log(n+1)−log(n) as log 1 + n1 , and

now use the fact that | log(1 + x)| < |x| for 0 < x < 1. Therefore, | log(n + 1) − log(n)| < n1 ,
and so log(n + 1) − log(n) = O( n1 ).

3. Find the rates of convergence of the following functions as h → 0:

sin h 1 − cos h
a. lim =1 b. lim =0
h→0 h h→0 h
sin h − h cos h 1 − eh
c. lim =0 d. lim = −1
h→0 h h→0 h
Answer: Here, Maclaurin series are the easiest way to get a solution:
3
sin h h − h6 + · · · h2
= =1− +···
h h 6
sin h
and so h = 1 + O(h2 ).
 For b, we have
2
1 − cos h 1 − (1 − h2 + · · ·) h
= = + ···,
h h 2
1−cos h
so h = O(h).
For c, we have
h3 h2
sin h − h cos h (h − 6 + · · ·) − h(1 − 4 + · · ·) −h2 h2
= = +
h h 6 4
sin h−h cos h
so h
= O(h2 ).
Finally, for d, we have
h2
1 − eh 1 − (1 + h + 2
+ · · ·) h
= = −1 − + ···,
h h 2
1−eh
so h
= −1 + O(h).

4. Suppose that 0 < q < p and αn = α + O(n−p ). Show that αn = α + O(n−q ).

Answer: The definition says that for sufficiently large n and for some positive constant K,
|αn − α| < Kn−p . Because q < p, we know that n−p < n−q . Therefore, |αn − α| < Kn−q ,
which in turn says that αn = α + O(n−q ).

5. Suppose that 0 < q < p and F (h) = L + O(hp ). Show that F (h) = L + O(hq ).
Answer: The definition says that for sufficiently small positive real numbers h and some
positive constant K, |F (h) − L| < K|hp |. Again, because q < p and |h| < 1, |hp | < |hq |.
This means that |F (h) − L| < K|hq |, which in turn means that F (h) = L + O(hq ).

6. Use the bisection method to find a solution accurate to within 0.01 for the equation
x4 − 2x3 − 4x2 + 4x + 4 = 0 on the interval [−1, 4].
Answer: Here is a chart of the results, with a the left-hand endpoint of the bounding
interval, b the right-hand endpoint of the bounding interval, and m the midpoint of the
bounding interval at each stage:
n a b m f (m)
1 −1 4 1.5000 −0.6875
2 1.5000 4 2.7500 0.3477
3 1.5000 2.7500 2.1250 −4.3630
4 2.1250 2.7500 2.4375 −3.6797
5 2.4375 2.7500 2.5938 −2.1745
6 2.5938 2.7500 2.6719 −1.0526
7 2.6719 2.7500 2.7109 −0.3888
8 2.7109 2.7500 2.7305 −0.0299
9 2.7305 2.7500 2.7402 0.1565
10 2.7305 2.7402 2.7354 0.0627
 This tells us that a root is between 2.7305 and 2.7354.

7. Let f (x) = x4 + 2x2 − x − 3. Use algebraic manipulations to show that each of the
following functions has a fixed point at p if and only if f (p) = 0:
1/2
x + 3 − x4

2 1/4
a. g1 (x) = (3 + x − 2x ) b. g2 (x) =
2
1/2
3x + 2x2 + 3
4

x+3
c. g3 (x) = d. g 4 (x) =
x2 + 2 4x3 + 4x − 1
Answer: (a) Start with x4 + 2x2 − x − 3 = 0, and move the last three terms to the right-
hand side of the equation, yielding x4 = −2x2 + x + 3. Take fourth roots, and we have
x = (3 + x − 2x2 )1/4 . So a fixed point of g1 (x) = (3 + x − 2x2 )1/4 will be a root of the
original equation. The algebra here is reversible, yielding the “if and only if” conclusion.
(b) Start with x4 + 2x2 − x − 3 = 0, and now move all but the quadratic term to the
right-hand side of the equation, yielding 2x2 = −x4 + x + 3. Divide by 2 and take square
roots to get x = ((3 + x − x4 )/2)1/2 . Again, we can see that this process is reversible.
(c) Start with x4 + 2x2 − x − 3 = 0, and move the last two terms to the right-hand
side, yielding x4 + 2x2 = x + 3. Factor the left-hand side into x2 (x2 + 2), divide by x2 + 2,
and take square roots, and we get x = ((x + 3)/(x2 + 2))1/2 .
(d ) This is Newton’s method in disguise. Start with x4 + 2x2 − x − 3 = 0, and divide
both sides by −(4x3 + 4x − 1), yielding −(x4 + 2x2 − x − 3)/(4x3 + 4x − 1) = 0. Now add
x to both sides, yielding x − (x4 + 2x2 − x − 3)/(4x3 + 4x − 1) = x. Simplify the left-hand
side, and we get (3x4 + 2x2 + 3)/(4x3 + 4x − 1) = x. Again, the process is reversible.

8. Use the functions gk (x) in the previous problem and perform four iterations (if possible,
without dividing by 0 or taking the square root of a negative number), starting with p0 = 1
and gk (pn ) = pn+1 . Which of the four functions seems to give the best approximation to
a solution of the equation f (x) = 0?
Answer: I computed the following:

n g1 (pn ) g2 (pn ) g3 (pn ) g4 (pn )

0 1.1892 1.2247 1.1547 1.1429
1 1.0801 0.9937 1.1164 1.1245
2 1.1497 1.2286 1.1261 1.1241
3 1.1078 0.9875 1.1236 1.1241
4 1.1339 1.2322 1.1242 1.1241

Clearly, g4 (x) is converging to the fixed point quicker than any of the other three
functions.