Chapter 2 Solutions of Equations in One Variable

One of the most basic problems of numerical approximation is the root-finding problem, which
is aimed at finding a root, or solution, of an equation of the form f (x) = 0, or equivalently
finding a zero of f (x).

1 The bisection method

Suppose f is a continuous function defined on the interval [a, b], and f (a) and f (b) are of opposite
sign(i.e. f (a)f (b) < 0). By the intermediate value Theorem, there exists a number p in (a, b)
with f (p) = 0.
For simplicity, we assume that the root of equation f (x) = 0 in this interval is unique.
The method calls for a respected halving of subintervals of [a, b], and at each step, locating the
half containing p.
To begin, set a1 = a and b1 = b, and let p1 be the midpoint of [a, b], that is,

b1 a1 b1 + a1
p1 = a1 + =
2 2
If f (p1 ) = 0, then p = p1 , and we are done. If f (p1 )f (a1 ) > 0, then f (p1 )f (b1 ) < 0, which
implied p (p1 , b1 ). Now we set a2 = p1 , b2 = b1 . If f (p1 )f (a1 ) > 0, we set a2 = a1 , b2 = p1 . we
then reapply the process to the interval [a2 , b2 ]. Keeping on going this way, we get a sequence
an +bn
pn , where pn = 2 .
The iteration can be stopped( terminated) by selecting a tolerance(threshold) > 0 such that
pN pN 1
< , pN 6= 0
pN
and pN will serve as the approximation to p.
Example.(p.54 No.9)
Let f (x) = (x + 2)(x + 1)x(x 1)3 (x 2). To which zero of f does the bisection method converge
when applied on the following intervals?

a.[3, 2.5] b.[2.5, 3] c.[1.75, 1.5] d.[1.5, 1.75]

Solution:
a1 +b1
(a). Let a1 = 2.5, b1 = 3, then p1 = 2 = 0.25, from f (a1 ) < 0, f (b1 ) > 0, f (p1 ) < 0, we
a2 +b2
let a2 = p1 , b2 = b1 , then p2 = 2 = 1.125. Since

a3 + b3
f (p2 ) < 0 a3 = p2 , b3 = b2 p3 = = 1.8125
2
similarly,
a4 + b4
f (p3 ) < 0 a4 = p3 , b4 = b3 p4 = = 2.15625
2

1
 a5 + b5
f (p4 ) > 0 a5 = a4 = p3 , b5 = p4 p5 =
2
2
We see the bisection method converges to 2 when applied on the interval[3, 2.5].
(b)
i [ai , bi ] pi f (ai ) f (pi ) f (bi )
1 [-2.5,3] 0.25 - + +
2 [-2.5,0.25] -1.125 - + +
3 [-2.5,-1.125] -1.8125 - + +
4 [-2.5,-1.8125] -2.15625 - - +
5 [-2.15625,-1.8125] -1.984375 - + +
So the bisection method converges to -2 when applied on the interval [2.5, 3].

(c)
i ai bi pi f (ai ) f (pi ) f (bi )
1 -1.75 1.5 -0.125 + - -
2 -1.75 -0.125 -0.9375 + - -
3 -1.75 -0.9375 -1.34375 + + -
4 -1.34375 -0.9375 -1.14063 + + -

So the bisection method converges to -1 when applied on the interval [1.75, 1.5].

(d)
i ai bi pi f (ai ) f (pi ) f (bi )
1 -1.5 1.75 -0.125 + + -
2 0.125 1.75 0.9375 + + -
3 0.9375 1.75 1.34375 + - -
4 0.9375 1.34375 1.14063 + - -

So the bisection method converges to 1 when applied on the interval [1.5, 1.75].
Theorem 2.1 Suppose that f C[a, b] and f (a)f (b) < 0. The bisection method generates a
sequence {pn }
n=1 approximating a zero p of f with

ba
|pn p| , when n 1
2n

Proof. It is easy to know

1
b2 a2 = (b1 a1 )
2
1 1 1
b3 a3 = (b2 a2 ) = 2 (b1 a1 ) = 2 (b a)
2 2 2
generally
1
bn an = (b a)
2n1

2
Since
an + bn
pn = , pn (an , bn ),
2
we have
1 ba
|pn p| (bn an ) = n
2 2
which leads to
lim pn = p
n

and the sequence {pn } 1

n=1 converges to p with rate of convergence O( 2n ), i.e.

1
pn = p + O( ).
2n

2 Fixed-point iteration
Definition A number p is said to be a fixed point for a given function g if g(p) = p.
A root-finding problem f (p) = 0 can be transformed to finding a fixed-point problem in the
following way.
Let
g(x) = x + f (x), is some nonzero constant

Then
g(x) = x f (x) = 0

which means that if f (x) has a zero at p, then the function g(x) defined by g(x) = x+f (x), ( 6=
0) has a fixed point at p.
Conversely, if the function g(x) has a fixed point at p, then the function defined by f (x) =
1
(g(x) x) has a zero at p.
Since the fixed-point form is easier to analyze, and certain fixed-point choices lead to very pow-
erful root-finding techniques, we often transform a root-finding problem to a fixed-point finding
problem.
The geometric explanation for fixed-points of g(x) is that the fixed-points of g(x) are the inter-
sections of y = g(x) and y = x as shown in Fig.2.
The following theorem gives sufficient conditions for the existence and uniqueness of a fixed
point.
Theorem 2.2
a) If g [a, b] and g(x) [a, b] for all x [a, b], then g has a fixed point in [a, b];
b) If, in addition, g 0 (x) exists on (a, b) and there is a positive constant k < 1 such that

|g 0 (x)| k, f or all x [a, b],

then the fixed point of g(x) in [a, b] is unique.

Proof.

3
a) If g(a) = a or g(b) = b, then g has a fixed point at an endpoint.
If not, then from g(x) [a, b] it follows g(a) [a, b] and g(b) [a, b]. Since the function
h(x) = g(x) x [a, b] and h(a) = g(a) a > 0, h(b) = g(b) b < 0.
The intermediate valued Theorem implies that there exists p (a, b) for which h(p) = 0,
therefore g(p) = p. Namely, this number p is a fixed point for g.
b) Suppose, in addition, that |g 0 (x)| k < 1 for all x (a, b) and that p and q are both fixed
points for g in [a, b].
If p 6= q, then the Mean value Theorem implies that there exists a number between p and q,
such that
p q = g(p) g(q) = (p q)g 0 (),

p 6= q results in g 0 () = 1 which is a contradiction with p 6= q results in |g 0 ()| < 1.

This contradiction must come from the only supposition p 6= q. Hence, p = q and the fixed point
for g in [a, b] is unique.
Find the root of Eq.
f (x) = x2 3 = 0

Example 2.2.1
Let g(x) = (x2 1)/3 on [1, 1].Then g 0 (x) = 2
3 x, set g 0 (x) = 0, we get x = 0, and from
g 00 (x) = 2
3 > 0, we see g(0) = 31 is the minimum of g, and the maximum of g is g(1) = 0.
Clearly g C[1, 1], and g [ 31 , 0] [1, 1], moreover

2 2
|g 0 (x)| = |x| < , f or all x (1, 1)
3 3
so g satisfied all the hypotheses of Theorem 2.2 and has a unique fixed point in [1, 1].
In this example, the unique fixed point p in the interval [1, 1] can be determined algebraically.
In fact, if
p2 1
p = g(p) =
3
then
p2 3p 1 = 0

which, by the quadratic formula, implies that

3 13
p=
2

3 13 3+ 13
p1 = 2 (1, 1),and p2 = 2 (3, 3.5) are both the fixed points for g. However

2
|g 0 (x)| = |x| > 2, f or all x (3, 3.5)
3
therefore g does not satisfy the hypotheses of Theorem 2.2 on [3, 3.5], which shows that the
hypotheses of Theorem 2.2 are sufficient to guarantee a unique fixed point but are not necessary.
Definition(fixed-point iteration)

4
Suppose g C[a, b], p0 (a, b) is an initial approximation and the sequence {pn }
n=1 is generated
by letting pn = g(pn1 ) for each n 1.
If
lim pn = p,
n

then
p = lim pn = lim g(pn1 ) = g( lim pn1 ) = g(p)
n n n

and a solution to x = g(x) is obtained. This technique is called fixed-point iteration, or functional
iteration.
Theorem 2.3(Fixed-point Theorem) Let g(x) C[a, b] and g(x) [a, b] for all x in [a, b].
Suppose, in addition, that g 0 exists on (a, b) and a constant k (0, 1) exists such that

|g 0 (x)| k, f or all x (a, b)

Then, for any number p0 [a, b], the sequence {pn }

n=1 defined by

pn = g(pn1 ), n1

converges to the unique fixed point p in [a, b].

Proof. g(x) C[a, b] and g(x) [a, b] for all x in [a, b] implies that a unique fixed point p for g
exists in [a, b]. Using the Lagrange Mean value Theorem and the fact that |g 0 (x)| k (0, 1),
we have for each n,

|pn p| = |g(pn1 ) g(p)| = |pn1 p| |g 0 (n )| k|pn1 p|

where n is between pn1 and p and hence in (a, b). Applying the above inequality repeatedly
(inductively) gives

p0 [a, b] p1 = g(p0 ) [a, b]

p2 = g(p1 ) [a, b]

pn1 = g(pn2 ) [a, b]

|pn p | k|pn1 p | k 2 |pn2 p | k n |p0 p | k n (b a) 0, (n )

lim pn = p.
n

Corollary 2.4 If g satisfies the hypotheses of Theorem 2.3, then bounds for the error involved
in using pn to approximate p are given by

|pn p | k n max{p0 a, b p0 }

and
kn
|pn p | |p1 p0 |, f or all n 1
1k

5
Proof. The first inequality is obvious since |p0 p | max{p0 a, b p0 }.
As for the second inequality, noting

|pn+1 pn | = |g(pn ) g(pn1 )| = |pn pn1 | |g 0 (n )| k|pn pn1 |

k 2 |pn1 pn2 | k n |p1 p0 |

Therefore

|pn+m pn | |pn+m pn+m1 | + |pn+m1 pn+m2 | + + |pn+1 pn |

(k n+m1 + k n+m2 + + k n )|p1 p0 |
= k n (1 + k + + k m1 )|p1 p0 |
k n (1 k m )
= |p1 p0 |
1k
which leads to
k n (1 k m ) kn
|pn p | = lim |pn+m pn | lim |p1 p0 | = |p1 p0 |.
m m 1k 1k
Example 2.2.2 Use a fixed-point iteration method to determine a solution accurate to within
102 for x3 x 1 = 0 on [1, 2]. Use p0 = 1. (pp.64,N0.6)
Solution.
(a).x3 x 1 = 0 can be changed to x3 1 = x.
Let g(x) = x3 1, then g(x) C[1, 2], since g(1) = 0, g(2) = 7, g(x) fails to map the interval
[1, 2] into itself. Therefore there is no reason to expect convergence for this choice of g.
1
(b).x3 x 1 = 0 x(x2 1) = 1 x = x2 1
.
1
If we set g(x) = x2 1
, then g(x) is discontinuous at x = 1, therefore g(x) 3 C[1, 2], this choice
should be rejected.
1
(c).x3 x 1 = 0 x3 = x + 1 x = (x + 1) 3 .
1
Set g(x) = (x + 1) 3 , then

g(x) C[1, 2];

2
g 0 (x) = 31 (x + 1) 3 > 0, for all x [1, 2], which implies 21/3 g(x) 31/3 g(x) maps
into itself;
2 5
g 0 (x) = 31 (x + 1) 3 , g 00 (x) = 92 (x + 1) 3 < 0 for all x [1, 2], therefore g 0 (x) monoton-
ically decreases in [1, 2], which implies

1 2 1 2 1 1
3 3 g 0 (x) 2 3 g 0 (x) |g 0 (x)| < 1
3 3 6 3
.

6
Theorem 2.3 confirms the convergence of the sequence {pn }n1 generated by pn = g(pn1 ).
Use p0 = 1, we get

p1 = 1.25992104989287
p2 = 1.3119382668329
p3 = 1.32235381913882
p4 = 1.32426874455158, |p4 p3 | = 0.0019149254 < 102
p5 = 1.32463262525092
p6 = 1.32470174851036
p7 = 1.32471487844095
p8 = 1.32471737243567
p9 = 1.32471784616215, |p9 p8 | = 4.73726474597682 107

3 Newtons Method
Newtons ( or the Newton-Raphson) method is one of the most powerful and well-known numeri-
cal methods for solving a root-finding problem. Perhaps the most widely used of all root-locating
formulas is the Newton-Raphson equation.
If the initial guess at the root is p0 , a tangent line can be extended from the point (p0 , f (p0 )).
The point where this tangent line crosses the x axis usually represents an improved estimate of
the root. Suppose that p1 is the xintercept of the tangent line to the graph of f at (p0 , f (p0 )).
(1)Geometrical interpretaion
The first derivative of f at p0 is equivalent to the slope
f (p0 ) 0
f 0 (p0 ) =
p 0 p1
which can be rearranged to yield
f (p0 )
p1 = p0
f 0 (p0 )
Similarly if p2 is the xintercept of the tangent line to the graph of f at (p1 , f (p1 )), then

f (p1 )
p2 = p1
f 0 (p1 )

Generally if pn is the xintercept of the tangent line to the graph of f at (pn1 , f (pn1 )), then

f (pn1 )
pn = pn1
f 0 (pn1 )
which is called the Newton-Raphson formula.
(2) An alternative means based on Taylor polynomials

7
Suppose that f C 2 [a, b]. Let pn1 be an approximation to p and f 0 (pn1 ) 6= 0. The first
Taylor polynomial for f (x) expanded about pn1 gives
1
f (x) = f (pn1 ) + (x pn1 )f 0 (pn1 ) + (x pn1 )2 f 00 (n (x))
2
where n (x) lies between pn1 and x. Since f (p) = 0, we have
1
0 = f (pn1 ) + (p pn1 )f 0 (pn1 ) + (p pn1 )2 f 00 (n (p))
2
If |p pn1 | is small(< 1), then |p pn1 |2 is much smaller. Neglecting the remainder term
yields
f (pn1 ) + (p pn1 )f 0 (pn1 ) 0

Solving for p gives

f (pn1 )
p pn1 pn
f 0 (pn1 )
The method generating the sequence {pn }
n=0 by

f (pn1 )
pn = pn1
f 0 (pn1 )

is called Newtons method.

It is easy to see that Newtons method is the special case of fixed-point iteration when

f (x)
g(x) = x .
f 0 (x)

(3)Convergence theorem for Newtons method

Theorem 2.5 Let f C 2 [a, b]. If p [a, b] satisfies that f (p) = 0 and f 0 (p) 6= 0( which mans
p is single root, not multiple root), then there exists a > 0 such that the sequence {pn }
n=1
generated by Newtons method converges to p for any initial approximation p0 [p , p + ].
Proof. Let
f (x)
g(x) = x
f 0 (x)
Purpose: find an interval [p , p + ] such that

(1) g(x) C[p , p + ];

(2) g(x) maps [p , p + ] into itself ;
(3) |g 0 (x)| < k (0, 1)

|f 0 (p)|
Step 1. f C 2 [a, b] indicated f 0 (x) is continuous on [a, b]. Since f 0 (p) 6= 0, we set = 2 ,
the continuity of f 0 (x) at x = p implies that there exists a 1 > 0 such that

|f 0 (x) f 0 (p)| <

8
whenever |x p | 1 . From

||f 0 (x)| |f 0 (p)|| |f 0 (x) f 0 (p)| <

one gets
|f 0 (p)|
|f 0 (x)| > |f 0 (p)| =
2
which means
f 0 (x) 6= 0, whenever x [p 1 , p + 1 ]

Step 2. Since f 0 (x) 6= 0, for x [p 1 , p + 1 ], g is defined and continuous on [p 1 , p + 1 ].

From
(f 0 (x))2 f (x)f 00 (x) f (x)f 00 (x)
g 0 (x) = 1 =
(f 0 (x))2 (f 0 (x))2
and f 00 C[a, b], we know g 0 (x) C[p 1 , p + 1 ]. Therefore g 0 (p) = 0 implies that for any
= k (0, 1), there exists a 2 with which

|g 0 (x)| = |g 0 (x) g 0 (p)| < = k < 1

whenever x [p 2 , p + 2 ]. Now that = min{1 , 2 }, then g(x) C[p , p + ], and

|g 0 (x)| < k < 1 for all x in [p , p + ].
Step 3. For x [p , p + ], the Mean value theorem implies

|g(x) p| = |g(x) g(p)| = |g 0 ()| |x p| < |x p|

Hence from |x p| , one derives |g(x) p| , namely g(x) [p , p + ]. As a result, g(x)

maps [p , p + ] into itself.
g(x) satisfies all the hypotheses of the Fixed-point Theorem. So the sequence {pn }
n=1 generated
by
f (pn1 )
pn = g(pn1 ) = pn1
f 0 (pn1 )
converges to p for any p0 [p , p + ].
(4)Example 2.3.1
Use Newtons iteration method to determine a solution accurate to within 106 for x3 x1 = 0
on [1, 2]. Use p0 = 1.
Solution. Using Newtons iteration method yields the following iteration scheme

f (pn ) p3n pn 1
pn+1 = pn = p n
f 0 (pn ) 3p2n 1

which is to take
x3 x 1
g(x) = x
3x2 1

9
in fixed-point iteration. Then

p1 = 1.500000000
p2 = 1.347826087
p3 = 1.325200399
p4 = 1.324718174
p5 = 1.324717957

and
|p5 p4 | = 2.167542 107 .

If we use the fixed-point iteration

pn+1 = (pn + 1)1/3

which is given in Example 2.2.2 (p.40), then one needs to do computation up to n = 9.

(5) Secant method
Newtons method has a major weakness, that is, one needs to know the value of the derivative
of f at each approximation. Frequently, f 0 (x) is much more difficult and needs more arithmetic
operations to calculate than f (x).
To circumvent the problem of the derivative evaluation in Newtons method, we introduce a
slight variation.
Using the approximation
f (pn1 ) f (pn2 )
pn1 pn2
for f 0 (pn1 ) in Newtons formula gives

f (pn1 )(pn1 pn2 )

pn = pn1 .
f (pn1 ) f (pn2 )

This technique is called the Secant method.

As shown in Fig.2.9, Starting with the two initial approximations p0 and p1 , the approximation
p2 is the xintercept of the line joining (p1 , f (p1 )) and (p2 , f (p2 )), and usually the approxima-
tion pn is the xintercept of the line joining (pn2 , f (pn2 )) and (pn1 , f (pn1 )).
(6) Method of False position
The method of False position generates approximations in the same manner as the secant
method, but it includes a test to ensure that the root is bracketed between successive itera-
tions.
First choose initial approximations p0 and p1 with f (p0 )f (p1 ) < 0. The approximation p2 is cho-
sen in the same manner as in the secant method, as the xintercept of the line joining (p1 , f (p1 ))
and (p0 , f (p0 )). To decide which secant line to use to compute p3 , we check if f (p2 )f (p1 ) < 0
is negative or not. If yes, then p1 and p2 bracket a root, and we choose p3 as the xintercept
of the line joining (p1 , f (p1 )) and (p2 , f (p2 )), if not, we choose p3 as the xintercept of the line

10
joining (p0 , f (p0 )) and (p2 , f (p2 )), and then interchange the indices on p0 and p1 . In a similar
manner, once p3 is found, the sign of f (p2 )f (p3 ) determines whether we use p2 and p3 or p1 and
p3 to compute p4 .

4 Error Analysis for Iterative Methods

(1) Convergent of order
An iteration pn = g(pn1 ) is said to be of order if the sequence {pn }
n=0 satisfies

|pn+1 p|
lim =
n |pn p|

where > 0, > 0, and we say that {pn }

n=0 converges to p of order with asymptotic error
constant .
Two cases of order are given special attention.

If = 1, namely, if
|pn+1 p|
lim = 1 > 0
n |pn p|
then the sequence {pn }
n=0 is said to be linearly convergent.

If = 2, i.e., if
|pn+1 p|
lim = 2 > 0
n |pn p|2
then the sequence {pn }
n=0 is said to be quadratically convergent.

(2) Theorem 2.7(Linear convergence Theorem)

Suppose g C[a, b], g(x) [a, b] for all x [a, b], and g 0 (x) is continuous on (a, b) such that

|g 0 (x)| k, f or all x (a, b)

where k < 1 is a positive constant.

If g 0 (p) 6= 0, then for any number p0 in [a, b], the sequence

pn = g(pn1 ), f or n 1

is only linearly convergent to the unique fixed point p in [a, b].

Proof. By fixed-point Theorem 2.3 we know that {pn }
n=0 converges to p. Applying the Mean
value Theorem to g gives

pn+1 p = g(pn ) g(p) = g 0 (n ) (pn p), n (a, b)

11
since g 0 is continuous on (a, b), one gets

|pn+1 p|
lim = lim |g 0 (n )| = |g 0 (p)| =
6 0,
n |pn p| n

Therefore {pn }
n=0 is only linearly convergent to p.
Theorem 2.7 implies that higher-order convergence for fixed-point methods can occur only when
g 0 (p) = 0.
The next theorem gives additional conditions to ensure the quadratic convergence.
(3) Theorem 2.8 Let p be a fixed point of g(x) on an open interval I. Suppose that g 0 (p) = 0,
g 00 is continuous and |g 00 (x)| < M for all x I. Then there exists a > 0 such that, for
p0 [p , p + ], the sequence {pn }
n=0 defined by pn = g(pn1 ) converges at least quadratically
to p. And
M
|pn+1 p| < (|pn p|)2
2
holds for sufficiently large values of n.
Proof. From g 0 (p) = 0 and the continuity of g 0 , we know that for some k (0, 1), there exists
> 0 such that |g 0 (x)| k < 1 for all x on the interval [p , p + ] I. Now we prove that
g(x) maps [p , p + ] into itself. In fact, if x [p , p + ], the Lagrange Mean value Theorem
implies that for some number between x and p,

|g(x) p| = |g(x) g(p)| = |g 0 ()||x p| |x p|| < x p| ,

Therefore
g(x) [p , p + ].

Hence for p0 [p , p + ], the sequence {pn }

n=0 defined by pn = g(pn1 ) are contained in
[p , p + ]. Expanding g(x) in a linear Taylor polynomial for x [p , p + ] gives

g 00 () g 00 ()
g(x) = g(p) + g 0 (p)(x p) + (x p)2 = p + (x p)2
2 2
where lies between x and p. In particular, when x = pn ,

g 00 (n )
pn+1 = g(pn ) = p + (pn p)2
2
with n between pn and p. Thus

g 00 (n )
pn+1 p = (pn p)2
2
Since g C[a, b], g(x) [a, b] for all x [a, b], and |g 0 (x)| k < 1 for all x [p , p + ], it
follows from the fixed-point Theorem that {pn }
n=0 converges to p, which leads to

lim n = p.
n

12
The fact that g 00 is continuous indicates

|pn+1 p | 1 1 1
lim 2
= lim |g 00 (n )| = |g 00 ( lim n )| = |g 00 (p)|.
n (|pn p |) n 2 2 n 2

As a result, the sequence {pn } 00 00

n=0 is quadratically convergent if g (p) 6= 0. Since g C[p
, p + ] and |g 00 | < M , we have
M
|pn+1 p | < (|pn p |)2 .
2
(4) A new approach for Newtons method
Suppose
f (x) = 0 x = g(x)

Let us introduce a differentiable function and write

g(x) = x (x)f (x).

In order to make the iterative procedure derived from g quadratically convergent we need to
have f (p) = 0 and g 0 (p) = 0.
Note
g 0 (x) = 1 0 (x)f (x) (x)f 0 (x)

we have
0 = g 0 (p) = 1 0 (p)f (p) (p)f 0 (p) = 1 (p)f 0 (p)

if and only if
1
(p) =
f 0 (p)
Let
1
(x) = 0 (x)
f
1
then surely (p) = f 0 (p) , which produces the quadratically convergent procedure

f (pn1 )
pn = g(pn1 ) = pn1 (pn1 )f (pn1 ) = pn1
f 0 (pn1 )

This is simply Newtons method.

(5) Zero of multiplicity
Definition If f (x) = (x p)m q(x), where

lim q(x) 6= 0
xp

then we say that p is a zero of multiplicity m of f .

It is easy to see that if f (x) is m + 1 times differentiable on (p , p + ) and

f (p) = f (p) = = f (m1) (p) = 0, f (m) (p) 6= 0

13
then by Taylors expansion we have

(x p)m1 (m1) (x p)m (m) (x p)m+1 (m+1)

f (x) = f (p)+(xp)f 0 (p)+ + f (p)+ f (p)+ f (),
(m 1)! m! (m + 1)!

where lies between x and p.

Substituting
f (i) (p) = 0, i = 0, 1, , m 1

into above formula gives

(x p)m (m) (x p)m+1 (m+1)

f (x) = f (p) + f (),
m! (m + 1)!
= (x p)m q(x)

with
1 (m) x p (m+1)
q(x) = f (p) + f (),
m! (m + 1)!
clearly
1 (m)
f (p) 6= 0.
lim q(x) =
m!
xp

(6) Necessary and sufficient conditions for f (x) to have a zero of multiplicity m at p
Theorem 2.11 The function f (x) C m [a, b] has a zero of multiplicity m at p in (a, b) if and
only if
f (p) = f (p) = = f (m1) (p) = 0, but f (m) (p) 6= 0

Proof. According to the preceding arguments, we have already proved the sufficiency. Now
suppose that f (x) C m [a, b] has a zero of multiplicity m at p in (a, b), namely,

f (x) = (x p)m q(x)

with
lim q(x) 6= 0.
xp

On the other hand, f (x) C m [a, b] implies that f can be expanded into the following Taylor
polynomial with remainder

(x p)m1 (m1) (x p)m (m)

f (x) = f (p) + (x p)f 0 (p) + + f (p) + f (),
(m 1)! m!

Therefore
(x p)m1 (m1) (x p)m (m)
(x p)m q(x) = f (p) + (x p)f 0 (p) + + f (p) + f (),
(m 1)! m!

Setting x = p yields f (p) = 0, which leads to

x p 00 (x p)m2 (m2) (x p)(m1) (m)

(x p)(m1) q(x) = f 0 (p) + f (p) + + f (p) + f (),
2 (m 1)! m!

14
Setting x = p yields f 0 (p) = 0.
Keep on going this way, one finds

f 00 (p) = = f (m1) (p) = 0,

and finally,
1 (m)
q(x) = f ()
m!
since f (x) C m [a, b] , lies between x and p, we have

f (m) (p) = f (m) ( lim ) = lim f (m) () = lim m!q(x) = m! lim q(x) 6= 0
xp xp xp xp

Theorem 2.11 is proved.

As a corollary, in particular case when m = 1, we have the following results.
Theorem 2.10 f C 1 [a, b] has a simple zero at p in (a, b) if and only if f (p) = 0, but f 0 (p) 6=
0.(See pp.82)
(7) Case of multiple roots
If p is a zero of f of multiplicity m and f (x) = (x p)m q(x), then define

f (x)
(x) =
f 0 (x)
(x p)m q(x)
=
m(x p)m1 q(x) + (x p)m q 0 (x)
q(x)
= (x p)
mq(x) + (x p)q 0 (x)

Since
q(x)
lim 6= 0,
xp mq(x) + (x p)q 0 (x)
p is a simple zero. Applying Newtons method to gives

(x) f (x)f 0 (x)

g(x) = x = x .
0 (x) (f 0 (x))2 f (x)f 00 (x)

(8) Examples
Example 1. Suppose f C 2 [a, b], f (p) = 0, f 0 (p) 6= 0, p (a, b), p0 [p , p + ] [a, b].
Show that for
f (pn )
pn+1 = pn ,
f 0 (pn )
pn pn1 1 f 00 (p)
lim = .
n (pn1 pn2 )2 2 f 0 (p)
Proof. Since
f (pn1 )
pn pn1 =
f 0 (pn1 )
or
(pn pn1 )f 0 (pn1 ) + f (pn1 ) = 0,

15
and
(pn1 pn2 )2 00
f (pn1 ) = f (pn2 ) + (pn1 pn2 )f 0 (pn2 ) + f (n )
2
with between pn1 and pn2 , and f (pn2 ) + (pn1 pn2 )f 0 (pn2 ) = 0 implies

(pn1 pn2 )2 00
f (pn1 ) = f (n )
2
therefore
pn pn1 f (pn1 )
= 0 (p
(pn1 pn2 )2 f n1 )(pn1 pn2 )
2

(pn1 pn2 )2 00
2 f (n )
=
f 0 (pn1 )(pn1 pn2 )2
f 00 (n )
= 0
2f (pn1 )
From
lim n = p
n

and f 00 C[a, b] it follows

pn pn1 1 f 00 (p)
lim = .
n (pn1 pn2 )2 2 f 0 (p)
Example 2. Suppose a > 0, p0 > 0, show that the iteration

pn+1 = pn (p2n + 3a)/(3p2n + a)

is a method of order 3.

Proof. First of all we prove {pn } converges to a.
Let
x(x3 + 3a)
g(x) =
3x2 + a
then
(3x2 + 3a)(3x2 + a) 6x x(x2 + 3a)
g 0 (x) =
(3x2 + a)2
3x 6ax + 3a2
4 2
=
(3x2 + a)2
3(x2 a)2
=
(3x2 + a)2
3x4 1
< =
9x4 3
= {pn } converges and it is easy to know

lim pn = p = a.
n

Let > 0 such that

p0 [ a , a + ]

16

then for x [ a , a + ], and g( a) = a,

|g(x) g( a)| = |g 0 ()||x a| < |x a| <

which means g [ a , a + ]. Now

a pn (p2n + 3a)/(3p2n + a) a pn+1
lim = lim
n ( a pn )3 n ( a pn )3

( a pn )3
= lim
n (3p2 + a)( a pn )3
n
1
= 6= 0.
4a
Or Since

3(x2 a)2
g 0 (x) =
(3x2 + a)2

x+a)2
Set g 0 (x) = ( x a)2 h(x), where h(x) = 3((3x2 +a)2
, which implies

g 0 ( a) = g 00 ( a) = 0,

0 (x a)2 00 (x a)3 000
g(x) = g( a) + (x a)g ( a) + g ( a) + g ()
3 2 6
(x a) 000
= a+ g ()
6
(pn a)3 000
= pn+1 a = g(pn ) a = g ()
6
a pn (p2n + 3a)/(3p2n + a) g 000 ( a)
= lim = 6= 0.
n ( a pn )3 6
Example 3.(pp.86, No.11)
The iterative method to solve f (x) = 0 given by the fixed-point method x = g(x), where
f (pn1 ) f 00 (pn1 f (pn1 ) 2
pn = g(pn1 ) = pn1 ( ) , f or n = 1, 2,
f 0 (pn1 ) 2f 0 (pn1 ) f 0 (pn1
has
g 0 (p) = g 00 (p) = 0
this will generally yield cubic( = 3) convergence.
Proof. Let g(x) be expressed into the form

g(x) = x + (x)f (x) + (x)f 2 (x) (4.1)

17
where and are functions to be determined.
Suppose p is the root of f (x) = 0 and g(p) = p. Differentiating Eq. (4.1) leads to

g 0 (x) = 1 + 0 (x)f (x) + (x)f 0 (x) + 0 (x)f 2 (x) + 2(x)f (x)f 0 (x) (4.2)

Using f (p) = 0, we get

g 0 (p) = 1 + (p)f 0 (p)
To make g 0 (p) = 0, one must choose
1
(p) = 0 (p)
.
f
More generally, one can take
1
(x) = 0 (x)
. (4.3)
f
Substituting Eq. (4.3) into Eq. (4.2), we have

g 0 (x) = 0 (x)f (x) + 0 (x)f 2 (x) + 2(x)f (x)f 0 (x)

Differentiating above expression gives

g 00 (x) = 00 (x)f (x)+ 0 (x)f 0 (x)+ 00 (x)f 2 (x)+2 0 (x)f (x)f 0 (x)+2(x)(f 0 (x))2 +2f (x)((x)f 0 (x))0

which yields
f 00 (p)
g 00 (p) = 0 (p)f 0 (p) + 2(p)(f 0 (p))2 = + 2(p)(f 0 (p))2
f 0 (p)
to make g 00 (p) = 0, we must choose
f 00 (p)
(p) =
2(f 0 (p))3
more generally, one can take
f 00 (x)
(x) =
2(f 0 (x))3
therefore
f (x) (f (x))2 f 00 (x)
g(x) = x
f 0 (x) 2(f 0 (x))3
satisfies
g(p) = p, g 0 (p) = g 00 (p) = 0
The iteration scheme based on g(x) is as follows
f (pn1 ) f 00 (pn1 ) f (pn1 ) 2
pn = g(pn1 ) = pn1 ( )
f 0 (pn1 ) 2f 0 (pn1 ) f 0 (pn1
Next we show {pn } is of cubic convergence.
In fact, expanding g(x) into Taylors polynomial about x = p gives
(x p)2 00 (x p)3 000
g(x) = g(p) + (x p)g 0 (p) + g (p) + g ()
2! 3!
(x p)3 000
= g(p) + g ()
3!

18
with lying between x and p. Therefore
(pn p)3 000
pn+1 = g(pn ) = p + g (n )
6
with n lying between pn and p.
If g 000 is continuous on some interval containing p, then we have
|pn+1 p| 1 1
lim 3
= |g 000 (n )| = g 000 (p).
n |pn p| 6 6
Example 4.(pp.86, No.12)
If {pn }
n=0 are convergent secant method approximations to p, the solution to f (x) = 0, then a
constant C exists with |pn+1 p| C|pn p||pn1 p| for sufficiently large values of n. Assume

{pn }
n=0 converges to p of order , show = (1 + 5)/2.
Proof. By secant method
f (pn )(pn pn1 )
pn+1 = g(pn ) = pn
f (pn ) f (pn1 )
pn f (pn ) pn f (pn1 ) pn f (pn ) + pn1 f (pn )
=
f (pn ) f (pn1 )
pn1 f (pn ) pn f (pn1 )
=
f (pn ) f (pn1 )
so
pn1 f (pn ) pn f (pn1 )
pn+1 p = p
f (pn ) f (pn1 )
(pn1 p)f (pn ) (pn p)f (pn1 )
=
f (pn ) f (pn1 )
f (pn ) f (pn1 )
(pn p) (pn1 p)
= (pn1 p)(pn p)
f (pn ) f (pn1 )
f [pn , p] f [pn1 , p]
= (pn1 p)(pn p)
f (pn ) f (pn1 )
f [pn ,p]f [pn1 ,p]
p p
= (pn1 p)(pn p) f (p n)fn1
(pn1 )
n
pn pn1
f [pn1 , pn , p]
= (pn1 p)(pn p)
f [pn1 , pn ]
f 00 (n )/2
= (pn1 p)(pn p) 0
f (n )
where n lies between pn1 , pn , p, and n lies between pn1 and pn .
If f C 2 , and f 0 (p) 6= 0, then
f 00 (n )/2 1 f 00 (p)
lim =
n f 0 (n ) 2 f 0 (p)
|pn+1 p| 1 |f 00 (p)|
= lim =
n |pn p||pn1 p| 2 |f 0 (p)|

19
Therefore for sufficiently large values of n,

|pn+1 p| C|pn p||pn1 p| (4.4)

Suppose {pn } converges to p of order , then for sufficiently large values of n,

|pn+1 p| |pn p| , > 0, (4.5)

combining Eq. (4.4) and Eq. (4.5), we know that for sufficiently large values of n,

|pn p| C|pn p||pn1 p|

or
C
|pn p|1 |pn1 p|

so
C 1
1 1
|pn p| ( ) |pn1 p| 1 (4.6)

Comparing Eq.(4.6) with Eq.(4.5) yields
1
=
1
namely
2 1 = 0

solving for gives

= (1 5)/2

as a result, = (1 + 5)/2 since > 0.

20