This article was downloaded by: [The University of Texas at El Paso]

On: 30 December 2014, At: 01:00

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office:
Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

International Journal of Computer Mathematics

Publication details, including instructions for authors and subscription
information:
http://www.tandfonline.com/loi/gcom20

Solving a nonlinear equation by a

uniparametric family of iterative processes
a a a
J. A. Ezquerro , J. M. Gutiérrez & M. A. Hernández
a
Department of Mathematics and Computation , University of La Rioja , C/Luis
de Ulloa s/n, Logroño, 26004, Spain
Published online: 19 Mar 2007.

To cite this article: J. A. Ezquerro , J. M. Gutiérrez & M. A. Hernández (1998) Solving a nonlinear equation by
a uniparametric family of iterative processes, International Journal of Computer Mathematics, 68:3-4, 301-308,
DOI: 10.1080/00207169808804697

To link to this article: http://dx.doi.org/10.1080/00207169808804697

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”)
contained in the publications on our platform. However, Taylor & Francis, our agents, and our
licensors make no representations or warranties whatsoever as to the accuracy, completeness, or
suitability for any purpose of the Content. Any opinions and views expressed in this publication are
the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis.
The accuracy of the Content should not be relied upon and should be independently verified with
primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever
caused arising directly or indirectly in connection with, in relation to or arising out of the use of the
Content.

This article may be used for research, teaching, and private study purposes. Any substantial
or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or
distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can
be found at http://www.tandfonline.com/page/terms-and-conditions
 J. Computer Math.. Vol. 68, pp. 301 -308
Intern.
Reprints available directly from the publisher
Photocopy~ngpermitted by license only
1998 OPA (Overseas Publishers Association)
Amsterdam B.V. Published under license
under the Gordon and Breach Science
Publishers imprint.
Printed in India.

SOLVING A NONLINEAR EQUATION BY A

UNIPARAMETRIC FAMILY OF ITERATIVE
PROCESSES
Downloaded by [The University of Texas at El Paso] at 01:00 30 December 2014

J. A. EZQUERRO, J. M. GUTIERREZ and M. A. HERNANDEZ

University of La Rioja, Department of Mathematics and Computation,
C,'Luis de LWoa sin, 26004, Logrorio, Spain

(Received 6 fi-ovemeber 1996)

In this study, we introduce a family of third-order iterative methods, cited in [4], for solving a
nonlinear equation f ( x ) = 0. First we provide a convergence analysis for a real function
depending of one real parameter o. E W.Next it is proved that we can always apply a method of
this family to solve f(x) = 0.

Keywords: Nonlinear equations; iterative process; Chebyshev's method; the Convex Accelera-
tlon of Newton's method

Mathematical Subjecr Classificarion: 65H05

C. R. Categories: G. 1.3

1. INTRODUCTION

Let us consider the problem of solving

where f is a nonlinear scalar function on some interval [a, b] of the real line.
We generally use iterative process for solving (1). The most famous iteration
is Kewton's method:
 302 J. A. EZQUERRO et al.

Methods using higher order derivatives may be advantageous for special

types of problems, if it is not particularly expensive to evaluate the involved
derivatives in these methods. Two well-known third-order methods of this
type are
(1) Chebyshev's method ([2, 81) :

(2) The Convex Acceleration of Newton's method ([3,71):

Downloaded by [The University of Texas at El Paso] at 01:00 30 December 2014

We have denoted by Lf the function

This function and its connection with Newton's method is studied in [7].
According to the idea of continuation methods ([9, lo]), a homotopy
between two functions u, v: X-Y is a continuous map, w : [O, I] x X+Y
where w(a, x ) = au(x) + (1-a)v(x), w(0,x) = v(x) and w(1, x ) = u(x). If
this map exists, then u is homotopic to v. This is an equivalence relation
between the continuous maps of X in Y,where X and Yare any topological
spaces. So we define a homotopy between the two functions Go and GI
defined respectively in (2) and (3). Then a family of iterations is obtained

For cu = 0 the family mentioned above reduces to the Chebyshev method

and for a = 1 we get the Convex Acceleration of Newton's method. The
aim of this paper is to extend the values of parameter a to any real value,
studying the convergence of family (4) for all a E R and optimizating the
 NONLINEAR EQUATION ITERATIVE SOLVER 303

application of this family from an analysis of the speed of convergence in

terms of the parameter a. We analyse practical cases where we can find the
fastest method of (4) to solve (1).
Observe that every iterative process of family (4) is of order three. This
fact follows from Gander's result 151:
Let s be a simple root o f f and H a function where H(0) = 1 , H 1 ( 0 ) = 112
and IH1'(0)I < cc.Then the iteration
Downloaded by [The University of Texas at El Paso] at 01:00 30 December 2014

is of third order for an appropriate choice of to.

The study of uniparametric families of iterative processes is interesting
(see [ l , 111) because we can relax convergence conditions for iterations of a
family and we can also increase the order of convergence in same particular
cases [6].

2. A CONVERGENCE DISCUSSION

Let f : [a. b] C W -, R be a sufficiently often differentiable function satisfying

f l ( t ) > 0,f"(t) 2 0 in [a, b] andfla) < 0 < f l b ) Therefore there is only one
root s of Eq. ( 1 ) in [a, b]. In this section we analyse the function
L y ( t ) = f'(!y:(i'in order to find a method of ( 4 ) that converges to a solution
.f ( l )
of (1). Firstly we start with the study of the case a E (0,l).
THEOREM2.1 Let a E (0,l) and to E [a,b] where f ( t o ) > 0. Z f L f ( t ) 5 a and
3(1-a)-
Ly(t)= in [a, b], then the sequence {t,, .Idefined by ( 4 ) is decreasing
and converges to s.

Proof Iff ( t o ) > 0 , then to - s 2 0. By using the Mean Value theorem, we

get

for some w0 E (s, to). Besides we have

 304 J. A. EZQUERRO et al.

and it follows G&(t)1 0 as a consequence of the hypotheses. Therefore t,,

> s.
-
On the other hand,

By using mathematical induction, we obtain t,, ,>s and t,. .-to, n-150, n
2 0, since (s, t,, .-I) C (s, to).

-
Therefrom we infer that the sequence defined in (4) converges to u E [a, b].
> 0, we have that G,(t,.,) -+ 0 as n x , and
that u = s.
Downloaded by [The University of Texas at El Paso] at 01:00 30 December 2014

Now we study the speed of convergence for the iterations of (4) in terms
of the values of the parameter cu E (0, 1). As every process of (4) is cubically
convergent, we compare two decreasing sequences converging to the root s.
Notice that if t,,,-, # s and t,., = s then G,(z) = s for z ~ [ s t,,.-J.
, As a
result we can deduce that function f is linear in a neighbourhood [s, s-t E )
where E > 0.
THEOREM 2.2 Let 0 < a < B < 1 a n d t ~ = t , . ~ = t sE, ~[a, b] where f(to) >
0. Then the sequence itO,,) converges to s faster than (t,,,).

Proof By mathematical induction on n, it is enough to show that t,,, > ts,.

for all n 2 0, once provided that the sequence {t,,,) satisfies t,,, # s for
n > 0.
First of all, assume t,,, # s for all n 2 0 and write

G,(t) = t - f (t)
T-H ( a , Lf (t))
f (t)

where H ( a . Lf (I)) = 1 +i + -).

Lf(t) (1 aLr(1) AS H is a non-decreasing
function for a fixed t and 0 < cr < 3 < 1. we obtain t,, t3, > 0.
Thus. it follows b y induction that t,, ,> t3. , for all n 2 0.
On the other hand, if there exists an n E N where t,,n-l# s and t,,,=s,
then by the previous comments we have that G,(z) = s for all ~ ( s ,
and following an analogous technique the proof is completed.

Example 1 Let us consider the equation f(t) = (t3,/216) - 1 to illustrate

these results. So Lf(t) = (2,'3)-(144/t3) and Ly(t) = (112). Then the
equation (112) = (3(l-cr)')/(cu t (1-a13) has only one real root, a =
0.527487. Now it is immediate to check that Lf(t) < 0.527487 in [0.001,
 NONLINEAR EQUATION ITERATIVE SOLVER 305

10.1 11. Therefore, by theorem 2.1, if to = 8, the sequence {t0.527487,0)is

decreasing and converges to s=6 for a E [0,0.527487] (see Tab. I ) .
Moreover the sequence is the one that converges fastest to s as a
consequence of Theorem 2.2.
Now we complete the discussion of the values of the parameter a E R.
When a E (0, I), we have Ly(t) 5 (3(l- ~ ) ~ ) -t / ((1a-a13) < 3. Then, observe
that if a _< 0 , the assumptions will be milder and we can obtain convergence
for Ly(t) 2 3 in [a,b],even if the speed of convergence is slower. Otherwise if
a 2 1 , we will get iterative processes with cubical convergence and faster
than the Convex Acceleration of Newton's method ( a = 1 ) .
The proof of the next two theorems are analogous to Theorems 2.1 and
Downloaded by [The University of Texas at El Paso] at 01:00 30 December 2014

2.2.
T H E O R E M2.3 Ler a <O and to E [ a , b] where f ( t o ) >0 . I f
LJ(t) < 1 - d s and L y ( t ) < 3(1 - a ) for t E [a, b], then the sequence
( t n , J defined by (4) is decreasing and converges to s. Moreover the sequence
(2,. .) converges faster for increasing values of the parameter a 5 0.

Example 2 Consider the polynomial f defined by f ( t ) = t3 i 9t2 i 57t.

The point is to apply Theorem 2.3 to approximate the solution s = 0 of the
equation f ( t ) = 0. On one hand, we obtain

(6t - 18)(t3;9t2 + 57t) 1 5

L A [ )= and L y ( t ) = ---
(3t2 18t - 57)' 2 ( 3 +- t ) z
' '

TABLE I

TABLE I1
 306 J. A. EZQUERRO et ai

As Ly is a nonincreasing function in [-2, 21, we get Ly (t) 5 Lp (-2) = 5.5

for t E [-2,2]. So, if we consider cu = -1, then Lf(t) 5 0.292893 and Lp ( t ) <
6 in [-2, 1.14011]. Consequently, the hypotheses of Theorem 2.3 are
satisfied and we obtain a decreasing sequence {t-l,n) which converges to
s = 0 for to = 1.14 (see Tab. 11).

THEOREM 2.4 Let a L 1 and to E [a, b] where f(to) > 0. IfLf(t) < 1 and
Ly (t) 5 3(1 - a)in [a, b], then the sequence {t,, .) given by (2) is decreasing
and converges to s. Moreover the higher the value of the parameter a 2 1 is, the
faster the rate of convergence is.

Example 3 Let f(t) = t-cos t be a function which has a zero

Downloaded by [The University of Texas at El Paso] at 01:00 30 December 2014

s = 0.7390851332151606417. Then

( t - cos t) cos t sin t

Lf ( 4 = and Ly(t) = -1 + sin t
(1 -isin t)

It is easy to check that Lf(t) < 1 and Ly (t) 5 -(1/4) in [0.21, (7i/2)] for CI =
(13112). Therefore, if to = 7r/2, by Theorem 2.4, the sequence (t(13/12). n ) is

decreasing and converges to s. Furthermore, the iterative process (t(13j12),n)

is faster than the one defined by {t,,.) (the Convex Acceleration of Newton's
method), see Table 111.
Notice that if we consider toe[a, b] where f(to) < 0 in Theorems 2.1, 2.3
and 2.4, we obtain convergence to s but for increasing sequences.

3. PRACTICAL REMARKS

We can find infinitely many convergent interative processes of family (4) for
solving (I), one for each a E R. Even more, it is possible to obtain the fastest
one.
Let us denote M = max {Ly(t); t€[a, b]} and put the values of M on the y
axis and the values of a on the x axis (see Fig. 1).

TABLE I11
 NONLINEAR EQUATION ITERATIVE SOLVER

FIGURE 1
Downloaded by [The University of Texas at El Paso] at 01:00 30 December 2014

For a given f, we calculate the corresponding value of M. Then, by

Theorem 2.3, there exists only one a*<0 that satisfies M = 3(1-a*) if ML3
and the sequence converges to the root s of (1) for an appropriate
initial point to. Moreover {t,.,,) is the fastest method.
If M E (0, 3), there exists only one a* E (0, 1 ) satisfying M = (3(1-a')~)1
( a *+ (1 -a*)3),an algebraic equation of degree three in a* with only one real
solution. Therefore we apply the process {t,.,,) to solve (1) and, by
Theorem 2.1, this method is the fastest one of family (4).
Finally, if M 5 0 , there also exists only one a*>l satisfying M = 3(1 -a*)
and, as the two previous cases, we apply the process {t,.,,) to solve (1).
Example 4 Givenflt) = t3 - 30t- 117, we obtain the fastest process of (4)
to approximate the solution s = 3 off ( t ) = 0 in [2: 41. We have

2 60t2 - 702t - 600 1 5

Lf(t)= - - and L f , ( t ) = - + - .
3 ' 9 t 4 + 180t2+900 2 t2
As Ly is a non-increasing function in 12, 41, then M = 714 E ( 0 , 3).
Consequently, - 3 ~ 2 "+~ (22/5)aX- 1 = 0, whose unique real solution is
cr*= 0.273684. Therefore, starting from to=4, the sequence (t0.273684,nj is
decreasing and converges to s = 3, see Table IV.

TABLE IV
 308 J. A. EZQUERRO et al.

Acknowledgements
Supported in part by a grant of the University of La Rioja: 96PYB25MHV.

References
[I] Argyros, I. K . and Chen, D. (1994). Parameter-Based Algorithms for Approximating
Local Solution of Nonlinear Complex Equations, Proyecciones, 33(1), 53-61.
[2] Chen, D. (1993). Standard Kantorovich Theorem of the Chebyshev Method on Complex
Plane, Intern. J. Computer Math., 42(1+ 2), 67 - 70.
[3] Chen, D., Argyros, I. K. and Qian, Q. S. (1994). A Local Convergence Theorem for the
Super-Halley Method in a Banach Space, Appl. Math. Lett., 7(5), 49-52.
[4] Ezyquerro, J. A., GutiCrrez, J. M. and Hernandez, M. A. (1997). A Construction
Procedure of Iterative Methods with Cubical Convergence, Appl. Math. Comput., 85,
Downloaded by [The University of Texas at El Paso] at 01:00 30 December 2014

181 - 199.
[5] Gander, W. (1985). On Halley's Iteration Method, Amer. Math. Monthly, 92, 131 - 134.
[6] Gutierrez, J. M. and Hernandez, M. A. (1997). A Family of Chebyshev-Halley Type
Methods in Banach Spaces, BUN.Austral. Math. Soc., 55, 113- 130.
[7] Hernandez. M. A. (1992). Newton-Raphson's Method and Convexity, Zb. Rad. Prirod.-
Mat. Fak. Ser. Mat., 22(1), 159- 166.
[8] Hernandez, M. A. and Salanova, M. A. (1993). A Family of Chebyshev-Halley Type
Methods, Intern. J. Computer Math., 47, 59- 63.
[9] Kincaid, D. and Cheney, W. (1991). Numerical Analysis: Mathematics of Scient$c
Computing. BrooksCole Publishing Company, E.U.A.
[lo] Ortega. J. and Rheinboldt, W. C. (1970). Iterative Solution of Nonlinear Equarions in
Several Variables, Academic Press.
[ l l ] Werner, W. (1980). Some Improvements of Classical Iterative Methods for the Solutions
of Nonlinear Equations, in: E. L. Allgower. K. Glashoff and H.-0. Peitpn, Eds.. Lectures
Notes in Mathematics, Numerical Solution of Nonlinear Equations, Springer-Verlag.
Bremen, 427 - 440.