Available online at www.sciencedirect.

com

Applied Mathematics and Computation 194 (2007) 527533

www.elsevier.com/locate/amc

Steensen type methods for solving non-linear equations

Pankaj Jain
Department of Mathematics, Deshbandhu College, University of Delhi, Kalkaji, New Delhi 110019, India

Abstract
Families of Steensen type methods for solving non-linear equations are obtained. One family consists of second order
method involving no derivatives, while other consists of third order methods involving only the rst derivatives. Also a
derivative free third order method is provided. The theory is supported by an example comparing the existing methods
with the present one.
2007 Elsevier Inc. All rights reserved.
Keywords: Newton method; Steensen method; Order of convergence; Derivative free method

1. Introduction
Finding iterative methods for solving non-linear equations has been an important area of research in
numerical analysis. A variety of dierent methods are available to suit dierent situations. Most widely used
among them is the Newtons method
xn1 xn 

f xn
:
f 0 xn

This method is of second order and requires the use of rst derivative. If the derivative f 0 xn is replaced by the
xn
ratio f xn ffxxnnf
, the Newtons method becomes

xn1 xn 

f xn
;
f xn f xn  f xn

which is the famous Steensen method, see e.g., [1]. This method is a tough competitor of Newtons method.
Both the methods are of second order, both require two function evaluation per iteration but in contrast to
Newtons method, Steensen method is derivative free.
Towards the rst aim of this paper, we obtain a family of Steensen type methods, all of order 2 and without involving derivatives. The idea is to involve a suitable function H in the Steensen method (see Theorem 1)
which for the choice H  1 yields exactly the Steensen method. This is done in Section 2.

E-mail address: [email protected]

0096-3003/$ - see front matter 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.amc.2007.04.087

528

P. Jain / Applied Mathematics and Computation 194 (2007) 527533

Next, we provide another family of Steensen type methods all of which are of third order. This is done by
involving a dierent suitable function H in the Steensen method. However, this construction requires the use
of rst derivative. The corresponding result is given is Section 3. Let us emphasize that the famous third order
method is the Halley method see, e.g., [4] which uses second derivative. The motivation of these results come
from [2], where families of Newton type second order and third order methods are obtained (involving, respectively, the rst derivative and the second derivative).
Finally, we obtain a derivative free third order method. Indeed, the method uses iteration alternatively from
Steensen method and the secant method. In [3], Kasturiarachi used a similar idea combining Newton method
with the secant method. His method is also of third order but requires the use of rst derivative whereas our
method is derivative free. The construction of the method and its order of convergence are the contents of
Section 3. The theory is supported by an example in Section 4, where we compare the Newtons and Kasturiarachis methods with the present one.
2. A class of Steensens type methods
Following the idea of Gander [2], we prove
Proposition 1. Let the functions f and G have sufficient number of continuous derivatives in a neighbourhood of a
which is a simple zero of f. The iteration formula
xn1 F xn ;
where F is defined by
F x x 

f x2
Gx
f x f x  f x

is of second order if and only if G(a) = 1.

Proof. Write
2

Ax

f x
:
f x f x  f x

2:1

Then
F x x  AxGx;
so that
F 0 a 1  AaG0 a  A0 aGa:
By applying LHospitals rule, some straightforward but tedious calculations show that
Aa 0 and A0 a 1:
Therefore,
F 0 a 1  Ga:
It is known, see e.g., [4], that the method
xn1 F xn
is of second order if and only if F 0 (a) = 0, i.e.,
Ga 1
and the assertion is proved.

Remark 1. Proposition 1 is of no use directly since it requires the knowledge of the zero a of f. Note that
Aa 0;

P. Jain / Applied Mathematics and Computation 194 (2007) 527533

529

so that if we choose a function H such that

Gx H Ax
then G(a) = H(0) and consequently, the condition G(a) = 1 can be replaced by H(0) = 1. Precisely, we have
proved the following theorem which is of application.
Theorem 1. Let f be a continuous function and H be any function with H(0) = 1. The iteration method
2

xn1 xn 

f xn
H Axn
f xn f xn  f xn

2:2

is of second order, where the function A is as defined by (2.1).

Remark 2. The iteration method (2.2) represents a family of second order Steensen type methods. In particular, for H(t) = 1, it gives exactly the Steensen method. There could be many dierent choices of the functions H, e.g., H t 1 t.
Remark 3. Our method (2.2) can be compared with that of Gander [2] who considered
Ax

f x
:
f 0 x

Precisely, Gander gave a family of Newtons type methods. Our Proposition 1 and Theorem 1 show that in the
x
corresponding results of Gander, f 0 (x) can be replaced by f xffxf
, which makes the family (2.2) derivative
x
free.
3. A class of third order methods
In [2], Gander further proved the following:
Lemma A. Let the functions f and G have sufficient number of continuous derivatives in a neighbourhood of a
which is a simple zero of f. The iteration formula
xn1 F xn ;
where F is defined by
F x x 

f x
Gx
f 0 x

is of third order if and only if G(a) = 1 and G0 a 12

f 00 a
.
f 0 a

Theorem A. Let f be a continuous function and H be any function satisfying H(0) = 1 and H 0 0 12. Then the
iteration method
xn1 F xn ;
where
F x x 

f x
H Bx
f 0 x

with
Bx

f xf 00 x
f 0 x

is of third order.

3:1

530

P. Jain / Applied Mathematics and Computation 194 (2007) 527533

00

a
Remark 4. The idea of Gander was to see that B(a) = 0 and B0 a ff 0 a
so that if the function H is chosen so
00

a
that Gx H Bx, then the conditions G(a) = 1 and G0 a 12 ff 0 a
can be replaced, respectively, by H(0) = 1
0 xf xf 0 x
f
0
and H 0 12. In (3.1), if we replace f00 (x) by
,
and
denote
this new function by Bx, i.e.,
f x

Bx

f 0 x f x  f 0 x
f 0 x

3:2

2
00

a
. Consequently, the function B in
then it is easy to check by an application of LHospitals rule that B0 a ff 0 a
Theorem A can be replaced by B. Summarizing, we have proved the following:

Theorem 2. Under the assumptions of Theorem A, the iterative method

xn1 F xn ;
where
F x x 

f x
H Bx
f 0 x

with B given by (3.2) is of third order.

Remark 5. Theorem 2 gives a class of third order methods involving rst derivatives.
4. A derivative free third order method
In the last section, a class of third order methods involving rst derivative was provided. We now give a
derivative free third order method.
By combining Newtons and secant method, Kasturiarachi [3] gave the following method which he proved
to be of third order:
2

xn1 xn 

f xn
;
f 0 xn f xn  f xn 

where
xn xn 

f xn
:
f 0 xn

Replacing f 0 xn by
gxn

f xn f xn  f xn
f xn

in the above method, we get

3

xn1 xn 

f xn
;
f xn f xn  f xn f xn  f xn 

4:1

where
xn xn 

f xn
:
gxn

We prove below that the method (4.1) is of order 3.

4:2

P. Jain / Applied Mathematics and Computation 194 (2007) 527533

531

Theorem 3. Let the function f has sufficient number of zeroes in a neighbourhood of a which is a simple zero of f,
i.e., f 0 (a) 5 0. The iterative method (4.1) has order of convergence 3.
Proof. Let en and en be the errors in xn and xn , respectively, i.e.,
xn a en

and

xn a en :

4:3

By using Taylors expansion around x = a which is a zero of f, we nd that

f a en en f 0 a

e2n 00
e3
f a n f 000 a oe4n ;
2
6

so that
2

f a en e2n f 0 a e3n f 0 af 00 a oe4n

and
e2n 0
e3
2
f a 1 f 0 a f 00 a n f 0 af 000 a
2
6
2
3
31 f 0 af 00 a 1 f 0 a f 000 a oe4n :

f a en f a en en 1 f 0 af 0 a

Now, using all the above expressions in (4.2), we obtain

en e2n A e3n B oe4n ;
where
A

f 00 a
1 f 0 a
2f 0 a

and
B

f 00 a
f 000 a
2
0
1

f
1 f 0 a2 f 0 a;
a


2f 0 a
6f 0 a

so that by an application of Taylors expansion, we further obtain

f a en e2n Af 0 a e3n Bf 0 a oe4n :
Therefore,
f a en  f a

en


 000

f 00 a
0
0
3 f a
 Af a en
 Bf a oe4n
en f a
2
6
 000

f 00 a
f a
e3n
 Bf 0 a oe4n :
en f 0 a  e2n f 0 a
2
6
0

e2n

Similarly,
f a en f a en  f a en
f 00 a
en f 0 a e2n f 0 a 1 f 0 a2  1
2
"
#
2
3
0
000
0
00
f
af
a
1

f
af
a
1
f 0 a f 000 a f 000 a
3
en

oe4n :
6
6
6
2

4:4

532

P. Jain / Applied Mathematics and Computation 194 (2007) 527533

Multiplying (4.4) and the last equation, we nd that on the RHS

3

coefficient of e2n f 0 a ;
3
coefficient of e3n f 0 a2 f 00 a;
2
coefficient of e4n I 1  I 2 I 3 ;
where

"

#
2
2
2
2 000
f 00 a
f 00 a
f 0 a f 000 a f 0 af 000 a
f a 0

I 1 f a

;
3
6
2
2f a
2
"
#
f 0 af 00 a2
2 3 00
2
0
f a
I 2 f a
;
2
2
"
#
2
2
2
2
000
f 00 a
f 0 af 00 a
f 00 a
f 0 a f 000 a f 0 af 000 a
2 f a
0
0

;
I 3 f a
2
6
3
2f a
4
2
2

which on simplication gives that

2

coefficient of e4n f 0 a I
with

"

#
2
2
2
2
2
5f 000 a
f 00 a
f 00 a
f 0 a f 000 a f 0 af 000 a f 0 af 00 a


I f a
0


:
3
9
12
f a
2
4
2

Summarizing
f a en f a en  f a en f a en  f a en 


3 f 00 a
I
3
2
3

oe

e2n f 0 a 1 en
n 0
n :
2 f 0 a
f a

4:5

Also, by some calculations, we can show that

"
#
!
2
000
3 f 00 a
3 f 00 a
3
3
3 0
2 f a
3
f a en en f a 1 en
en

oen :
2 f 0 a
2f 0 a 4 f 0 a2

4:6

Now, using (4.3), (4.5) and (4.6) in (4.1), we obtain the error equation as
"

en1

000
3 f 00 a
3 f 00 a2
2 f a

en  en 1 en
n
2 f 0 a
2f 0 a 4 f 0 a2

oe3n



3 f 00 a
I
2
3
 en 0
oen
 1  en
2 f 0 a
f a

e3n
If 0 a  3f 00 a2 f 000 af 0 a oe4n
2f 0 a

and the assertion is proved.

5. Example
Consider the non-linear equation
f x cos x  xex x2 ;
which has a simple zero in the interval (0, 1). The following table demonstrate the comparison of the three
methods namely, the Newtons, the Kasturiarachis and the present method. The iterations have been carried
out on Maple.

P. Jain / Applied Mathematics and Computation 194 (2007) 527533

533

Method

xn

Newton

1
2
3
4
5
6
7
8
9

1
.72464469756709453577
.64465890487027013849
.63917780746728071170
.63915409677305075646
.63915409633200758124
.63915409633200758107
.63915409633200758105
.63915409633200758107

Kasturiarachi

1
2
3
4
5
6
7
8
9

1
.66076485847521534377
.63916021337699197520
.63915409633200772191
.63915409633200758107
.63915409633200758106
.63915409633200758107
.63915409633200758106
.63915409633200758107

Present method

1
2
3
4
5
6
7
8
9

1
.57696767213673077464
.63934974241874244537
.63915409632544234543
.63915409633200758105
.63915409633200758106
.63915409633200758107
.63915409633200758106
.63915409633200758107

References
[1]
[2]
[3]
[4]

S.D. Conte, Carl de Boor, Elementary Numerical Analysis, An Algorithmic Approach, McGraw-Hill, 1981.
W. Gander, On Halleys iteration method, Am. Math. Mon. 92 (2) (1985) 131134.
A.B. Kasturiarachi, Leap frogging Newtons method, Int. J. Math. Educ. Sci. Technol. 33 (4) (2002) 521527.
A. Ralston, A First Course in Numerical Analysis, McGraw-Hill, 1965.