You are on page 1of 7

Project 4.

Kaitlyn Monteer

May 1, 2015
In this project, I will examine the Newton-Raphson method for solving an equation of
the form f (x) = 0, establishing convergence of the iterative process under certain conditions,
and give an estimate of the speed of convergence.

The Newton-Raphson method is described in the text as follows:

Suppose f : [a, b] R is twice dierentiable; that is, f and f 0 are both dierentiable on
[a, b]. Suppose further that there are m and M , both positive, such that |f 0 (x)| m and
0 < |f 00 (x)| M for all x [a, b]. Finally, suppose that f (a)f (b) < 0. Dene a sequence as
follows:

a1 = a if f 0 f 00 < 0, and a1 = b if f 0 f 00 > 0,


an = an1 ff0(a n1 )
(an1 )
for n 2.
Then {an }n=1 converges to the unique solution r in [a, b] of the equation f (r) = 0.

Moreover,

|r an+1 | M (ran )2
2m
.

Before proving the above, it is necessary to prove a preliminary result, a special case of
Taylor's theorem, as follows:

THEOREM (): Suppose that f : [a, b] R is twice dierentiable. Then for each
x0 , x [a, b], there is c between x0 and x such that
00
f (x) = f (x0 ) + f 0 (x0 )(x x0 ) + f 2(c) (x x0 )2 .
Proof. Let f : [a, b] R be twice dierentiable, and let x0 , x [a, b]. Dene F : [a, b] R
by
f (x)f (x0 )f 0 (x0 )(xx0 )
F (t) = f (t) + f 0 (t)(x t) + Q(x t)2 , where Q = (xx0 )2
.

Observe that F (x) = f (x), and F (x0 ) = f (x), so F (x) = F (x0 ). Note that since f is
continuous on [a, b], and x 6= x0 , F is also continuous on [a, b]. So, by the Mean Value
Theorem, there exists some c strictly between x and x0 such that F 0 (c) = 0.

Observe that
F 0 (t) = f 0 (t) + f 00 (t)(x t) f 0 (t) 2Q(x t)
= f 00 (t)(x t) 2Q(x t)
So,

F 0 (c) = 0 = f 00 (c)(x c) 2Q(x c)


2Q(x c) = f 00 (c)(x c)
f 00 (c)
Q=
2

1
f 00 (c)
Hence, there exists some c between x, x0 such that Q = 2
, so we can write
f 00 (c)
F (t) = f (t) + f 0 (t)(x t) + 2
(x x0 )2

So, since F (x) = f (x) = F (x0 ), we can write


f 00 (c)
f (x) = f (x0 ) + f 0 (x0 )(x x0 ) + 2
(x x0 )2

We will now use this result to accomplish the original goal: proving the convergence of
the Newton-Raphson method and give estimates of the speed of convergence.

Proof. Suppose f : A R is twice dierentiable; that is, f and f 0 are both dierentiable
on [a, b]. Suppose further that there are m and M , both positive, such that |f 0 (x) m and
0 < |f 00 (x)| M for all x [a, b]. Finally, suppose that f (a)f (b) < 0. Dene a sequence as
follows:

a1 = a if f 0 f 00 < 0, and a1 = b if f 0 f 00 > 0,


an = an1 ff0(a n1 )
(an1 )
for n 2.

Notice that by Theorem 4.11, the conditions on f 0 and f 00 ensure that neither changes sign
in [a, b]. Also, f is continuous and one-to-one with f (a) and f (b) having opposite signs;
hence, there is exactly one solution in [a, b] to the equation f (x) = 0. We will call this
solution r. In what is to follow, we will assume that f 0 > 0 and f 00 > 0. The other cases are
similar.

By way of mathematical induction, let Pn be the statement "r < an+1 < an b". First, let
n = 1. Then

a1 = b, since f 0 f 00 > 0, and


a2 = a1 ff0(a1)
(a1 )
= b ff0(b)
(b)

Clearly, a1 b. Also, since f 0 > 0, f is increasing, and since f (a) and f (b) have opposite
signs, f (b) > 0. So
f (b)
f 0 (b)
> 0,

and thus

a2 = b f (b)
f 0 (b)
< b = a1 ,

so a2 < a1 .

2
Now, clearly a1 = b > r since r (a, b). So, by (), there exists from c (r, a1 ) such that
f 00 (c)
f (r) = f (a1 ) + f 0 (a1 )(r a1 ) + (r a1 )2
2
f 00 (c)
= f (b) + f 0 (b)(r b) + (r b)2
2
f 00 (c)
0 = f (b) + f 0 (b)(r b) + (r b)2
2
f 00 (c)
f (b) = f 0 (b)(r b) (r b)2
2
f (b) f 00 (c)
= (b r) 0 (r b)2
f 0 (b) 2f (b)

Note here that (r b)2 > 0, f 00 > 0, f 0 > 0, so


f 00 (c)(rb)2
2f 0 (b)
> 0,

so
f (b)
<br
f 0 (b)
f (b)
b + 0 < r
f (b)
f (b)
b 0 >r
f (b)
a2 >r

Hence, putting these all together yields r < a2 < a1 b.

Now, suppose that for some integer n 2, Pm is true for all m = 1, 2, . . . , n; that is,
r < am+1 < am b for all m = 1, 2, . . . , n. We want to show that Pn+1 is true; that is, that

r < an+2 < an+1 b.

First, from our assumption, an+1 < an , so an+1 < b, since an < b by assumption also.

Now, by our assumption, an+1 > r, so by () there exists some c (r, an+1 ) such that
f 00 (c)
f (r) = 0 = f (an+1 ) + f 0 (an+1 )(r an+1 ) + (r an+1 )2
2
f 00 (c)
f (an+1 ) = f 0 (an+1 )(r an+1 ) (r an+1 )2
2
f (an+1 ) f 00 (c)(r an+1 )2
= (a n+1 r)
f 0 (an+1 ) 2f 0 (an+1 )

3
Since f 00 > 0, (r an+1 )2 > 0, f 0 > 0, we then have
f (an+1 )
< an+1 r
f 0 (an+1 )
f (an+1 )
an+1 0 >r
f (an+1 )
an+2 > r

So we have an+2 > r.

Lastly, we have
f (an+1 )
an+2 = an+1
f 0 (an+1 )

Since an+1 (r, b] and f (b) > 0, f (r) = 0, the fact that f is continuous and that r is the
unique zero of f in [a, b] implies that f (an+1 ) > 0. Also, f 0 > 0, so
f (an+1 )
f 0 (an+1 )
> 0.

This implies that

an+2 < an+1 .

Hence, putting all of these facts together yields

r < an+2 < an+1 b.

So Pn+1 is true. Therefore, by mathematical induction, Pn is true for all n J l; that is, for
all n J ,

r < an+1 < an b (1)


Now, by (1) we have that {an } n=1 is decreasing and bounded below by r . Hence, it must
converge. Say {an }n=1 s. Also, since f is continuous, {f (an )}

n=1 f (s) by Theorem
3.1. Observe that since an 6= r for all n, there exists some cn between an and r such that
f 00 (cn )
f (an ) = f (r) + f 0 (r)(an r) + (an r)2
2
0 f 00 (cn )
= f (r)(an r) + (an r)2
2

4
So, let > 0 be given. Recall that 0 < |f 00 (x)| M for all x [a, b]. So let C = M2 . Then
f 00 (cn )
2
C for all n. Since {an }
n=1 is decreasing and an r > 0 for all n, there exists
N J such that
 1
aN r < min 1, ,
f 0 (r) + C f 0 (r) + C
Then for all n N , we have
|f (an ) 0| = |f (an )|
f 00 (cn )
= |f 0 (r)(an r) + (an r)2 |
00 2
0
f (cn )
|f (r)||an r| + |an r|2
2
f 00 (cn )
= f 0 (r)(an r) + (an r)2
2
f 0 (r)(an r) + C(an r)2

n=1 is decreasing and n N , an aN , so


Since {an }
|f (an )| = f 0 (r)(an r) + C(an r)2
f 0 (r)(aN r) + C(aN r)2

Since aN r < 1, (aN r)2 < (aN r), so we have


|f (an )| < f 0 (r)(aN r) + C(aN r)
= (aN r) (f 0 (r) + C)

Now, if 1, then since aN r < 1


f 0 (r)+C
,
1
|f (an )| < (f 0 (r) + C)
(f 0 (r) + C)
=1

If < 1, then since aN r <


f 0 (r)+C
,

|f (an )| < (f 0 (r) + C)
(f 0 (r) + C)
=

In either case, |f (an ) 0| < , so {f (an )}


n=1 0. Then f (s) = 0. But f (r) = 0 also, and
since f is one-to-one, this implies that s = r.

Now I want to prove that

5
|r an+1 | M (ran )2
2m
.

The reccurence relation for {an }n=1 shows that f (an ) = f (an )(an an+1 ). From (), since
0

an 6= r for all n, we have that there exists some c between an , r such that

f 00 (c)
f (r) = f (an ) + f 0 (an )(r an ) + (r an )2
2
f 00 (c)
= f 0 (an )(an an+1 ) + f 0 (an )(r an ) + (r an )2
2
f 00 (c)
= f 0 (an )(an an+1 + r an ) + (r an )2
2
00
f (c)
= f 0 (an )(r an+1 ) + (r an )2
2
f 00 (c)(r an )2
(r an+1 ) =
2f 0 (an )
00
f (c)(r an )2
|r an+1 | = ,
2f 0 (an )

Since f 00 > 0, f 0 > 0, (r an )2 > 0. Finally, our earlier assumption that there are m and M ,
both positive, such that |f 0 (x) m and 0 < |f 00 (x)| M for all x [a, b] yields
M (r an )2
|r an+1 | .
2m


You might also like