Anda di halaman 1dari 16

Numer Algor (2013) 62:429444

DOI 10.1007/s11075-012-9597-3
ORIGINAL PAPER

A family of Steffensen type methods


with seventh-order convergence
Xiaofeng Wang Tie Zhang

Received: 1 December 2011 / Accepted: 21 May 2012 /


Published online: 2 June 2012
Springer Science+Business Media, LLC 2012

Abstract In this paper, based on some known fourth-order Steffensen type


methods, we present a family of three-step seventh-order Steffensen type
iterative methods for solving nonlinear equations and nonlinear systems. For
nonlinear systems, a development of the inverse first-order divided difference
operator for multivariable function is applied to prove the order of convergence of the new methods. Numerical experiments with comparison to some
existing methods are provided to support the underlying theory.
Keywords Steffensens method Newtons method Derivative free
Seventh-order convergence Root-finding
Mathematics Subject Classfications (2010) 65H05 65B99

1 Introduction
Finding the root of a nonlinear equation f (x) = 0 is a classical problem in
scientific computation. Since 1960s, many methods have been proposed for
solving nonlinear equations, see [17] and the references therein. In these

X. Wang (B) T. Zhang


College of Sciences, Northeastern University, Shenyang 110819 Liaoning, China
e-mail: w200888w@163.com
T. Zhang
e-mail: ztmath@163.com
X. Wang
Department of Mathematics, Bohai University, Jinzhou 121013 Liaoning, China

430

Numer Algor (2013) 62:429444

methods, Newtons method is the most well-known one. The Newtons method
for solving nonlinear equations is written as
xn+1 = xn

f (xn )
.
f  (xn )

(1)

This is an important and basic method [1], which converges quadratically.


However, when the first order derivative of the function f (x) is unavailable
or is expensive to compute, the Newtons method is still restricted in practical
applications. In order to avoid computing the first order derivative, Steffensen
in [2] proposed the following derivative-free method
xn+1 = xn

f (xn )
,
f [xn , zn ]

(2)

where zn = xn + f (xn ) and f [, ] is the first order divided difference.


The method
(2) is obtained from (1) by substituting f  (xn ) by ( f (zn )

f (xn )) f (xn ) and maintains quadratic convergence. The idea of removing
derivatives from the iteration process is very significant. Recently, many highorder derivative-free methods are built according to the Steffensens method,
see [37] and the references therein. Liu et al. in [3] presented the optimal
fourth-order method,

f (xn )

yn = xn f [xn , zn ] ,




(3)

f xn , yn f zn , yn + f [xn , zn ] f (yn )




.
xn+1 = yn
f xn , yn
f xn , yn
The method (3) can be applied to solve nonlinear systems of equations and
nonlinear boundary-value problems of ODEs as well. Ren et al. in [4] proposed
another optimal fourth-order Steffensen type method,

f (xn )

yn = xn f [xn , zn ] ,
(4)
f (yn )





x
=
y

.
n+1
n

f xn , yn + f yn , zn f [xn , zn ]
Cordero et al. in [5] also presented the following Steffensen type method with
sixth-order convergence

2 f (xn )2

yn = xn
,

f (xn + f (xn )) f (xn f (xn ))

yn xn
(5)
f (yn ) ,
wn = y n

2
f
(y
n ) f (xn )

yn xn

xn+1 = wn
f (wn ) .
2 f (yn ) f (xn )

Numer Algor (2013) 62:429444

431

Other Steffensen type methods and their applications are discussed by Zheng
et al. in [6] and by Petkovic et al. in [7]. All these methods are derivative-free
in per full iteration.
The purpose of this paper is to develop a new family of seventh-order
derivative-free methods and give the convergence analysis. This paper is
organized as follows. In Section 2, we present a family of three-step seventhorder iterative methods for solving nonlinear equations. The first two steps
are some known fourth-order Steffensen type methods, while the third step is
constructed by using suitable Taylor expansion and divided difference approximations. The new method is free from any derivatives and has the efficiency
index of 1.627. In Section 3, we prove that the order of convergence of the new
method is seven for nonlinear systems. In Section 4, we give some specific iterative methods which can be used in practical computations. Numerical examples
are given in Section 5 to illustrate convergence behavior of our methods for
simple roots. Section 6 is a short conclusion.

2 The methods and analysis of convergence


Now, we consider the iteration scheme of the form,

f (xn )

yn = xn f [x , z ] ,
n
n

f (un )

xn+1 = un
,
f  (un )

(6)

where un = g4 (xn ) stands for an optimal fourth-order Steffensen type method.


To derive a scheme with higher efficiency index, we approximate f  (un )
using the Taylor expansion and divided difference. By using the Taylor expansion, f (un ) and f  (un ) in (6) can be approximated by
f (un ) f (xn ) + f  (xn ) (un xn ) +

f  (xn )
(un xn )2 ,
2!

f  (un ) f  (xn ) + f  (xn ) (un xn ) .

(7)
(8)

It follows from (7) and (8) that


f  (un ) f [un , xn ] +

1 
f (xn ) (un xn ) .
2

(9)

In order to avoid the computation of the second derivative, we approximate


f  (xn ) as follows:


f  (xn ) 2 f xn , yn , un ,
(10)
where un and yn are sufficiently close to xn when n is large enough.

432

Numer Algor (2013) 62:429444

Substituting (10) into (9), we obtain an approximation of f  (un ):






f  (un ) f [un , xn ] + f xn , yn , un (un xn ) = f [un , xn ] + f un , yn


f xn , yn .
(11)
Therefore, a new scheme is derived as follows:

f (xn )

yn = xn
,

[x
f
n , zn ]

un = g4 (xn ) ,

f (un )




.
xn+1 = un
f [un , xn ] + f un , yn f xn , yn

(12)

For the iterative method defined by (12), we have the following result.
Theorem 1 Let I be an open interval and I a simple zero point of a
suf f iciently dif ferentiable function f :I R. Assume that un = g4 (xn ) is an
optimal fourth-order Stef fensen type method which satisf ies

(13)
un a = B0 e4n + B1 e5n + B2 e6n + B3 e7n + O e8n ,
where B0 = 0 and en = xn . Then the new method def ined by (12) is seventh
order.
Proof Using the Taylor expansion and taking into account f () = 0, we have



f (xn ) = f  () en + c2 e2n + c3 e3n + c4 e4n + c5 e5n + c6 e6n + O e7n ,
(14)



f  (xn ) = f  () 1 + 2c2 en + 3c3 e2n + 4c4 e3n + 5c5 e4n + 6c6 e5n + O e6n ,

(15)




f  (xn ) = f  () 2c2 + 6c3 en + 12c4 e2n + 20c5 e3n + 30c6 e4n + O e5n ,

(16)




f  (xn ) = f  () 6c3 + 24c4 en + 60c5 e2n + 120c6 e3n + 210c7 e4n + O e5n , (17)



f (4) (xn ) = f  () 24c4 + 120c5 en + 360c6 e2n + O e3n ,

(18)



f (5) (xn ) = f  () 120c5 + O (en ) ,

(19)

where ck = f (k) ()

k! f  () , k = 2, 3, .




Numer Algor (2013) 62:429444

433

From (14)(19), noting that zn = xn + f (xn ), we obtain


f [xn , zn ] =

f (zn ) f (xn )
f  (xn )
f  (xn )
= f  (xn ) +
(zn xn ) +
(zn xn )2
zn xn
2!
3!

f (4) (xn )
f (5) (xn )
(zn xn )3 +
(zn xn )4 + O e4n = f  ()
4!
5!


+ 2c2 f  () + c2 f  ()2 en




+ 3c3 f  () + c22 + 3c3 f  ()2 + c3 f  ()3 e2n

+ 4c4 f  () + 4c2 c3 f  ()2 + 6c4 f  ()2 + (4c4 + 2c2 c3 ) f  ()3


(20)
+ c4 f  ()4 e3n + O e4n ,
+

hence
yn = en

f (xn )
= H0 e2n + H1 e3n + H2 e4n + O e5n ,
f [xn , zn ]

where

(21)


H0 = c2 1 + f  () ,

(22)



H1 = c3 f  ()2 + 3 f  () + 2 c22 f  ()2 + 2 f  () + 2 ,

(23)


H2 = c4 f  ()3 + 4 f  ()2 + 6 f  () + 3


c2 c3 2 f  ()3 + 7 f  ()2 + 10 f  () + 7

+ c32 f  ()3 + 2 f  ()2 + 6 f  () + 4 .

(24)

Together with (13), (14)(19) and (21), we have


f  (xn )
f  (xn )
(un xn ) +
(un xn )2
2!
3!

f (4) (xn )
+
(un xn )3 + O e5n
4!




= f () 1 + c2 en + c3 e2n + c4 e3n + (c5 + c2 B0 ) e4n + O e5n ,

f [xn , un ] = f  (xn ) +

(25)


f  (xn )
f  (xn )
f xn , yn = f  (xn ) +
(yn xn ) +
(yn xn )2
2!
3!

f (4) (xn )
f (5) (xn )
(yn xn )3 +
(yn xn )4 + O e5n
4!
5!


= f () 1 + c2 en + (c3 + H0 c2 ) e2n + (c3 H0 + c4 + H1 c2 ) e3n




+ c5 + H0 c4 + H1 c3 + H2 c2 + c3 H02 e4n + O e5n .
+

(26)

434

Numer Algor (2013) 62:429444

By a similar argument to that of (14), we obtain





f (yn ) = f  () (yn ) + c2 (yn )2 + c3 (yn )3 + O e7n ,

(27)




f  (yn ) = f  () 1 + 2c2 (yn ) + 3c3 (yn )2 + O e5n ,

(28)




f  (yn ) = f  () 2c2 + 6c3 (yn ) + O e3n ,

(29)




f  (yn ) = f  () 6c3 + O e2n .

(30)

Using (13), (21) and (28)(30), we have





f  (yn )
f  (yn )
f un , yn = f  (yn ) +
(un yn ) +
(un yn )2 + O e5n
2!
3!





= f  () 1+c2 H0 e2n +c2 H1 e3n + (H2 + B0 ) c2 +c3 H02 e4n + O e5n ,
(31)

f  ()
(un ) + O e6n
2!



= f  () 1 + c2 B0 e4n + c2 B1 e5n + O e6n .

f [un , ] = f  () +

(32)

Hence, together with (12)(13), (25)(26) and (31)(32), we obtain the error
equation:
f (un )




f [un , xn ] + f un , yn f xn , yn


f [un , ]




= (un ) 1
f [un , xn ] + f un , yn f xn , yn

en+1 = xn+1 = un


= B0 e4n + O e5n c3 H0 e3n (c2 H0 2c2 B0 c4 H0 c3 H1 ) e4n + O e5n

= c3 H0 B0 e7n + O e8n

= c2 c3 1 + f  (a) B0 e7n + O e8n .

(33)

The error equation (33) gives the seventh-order convergence of the Steffensen
type method defined by (12). This completes the proof.
In terms of computational cost, the developed methods require four eval1
uations of the function per iteration. Let p w be the efficiency index (see
[8]), where p is the order of the method and w is the number of function
evaluations per iteration required
by the method. We see that the
new method

4
has the efficiency index of
7

1.627,
which
is
higher
than

2 1.414 of
Steffensens method (2), 3 4 1.587 of method (3)(4) and 4 6 1.565 of
Corderos method (5).

Numer Algor (2013) 62:429444

435

3 The extension of the methods and analysis of convergence


In this section we will prove the order of convergence of the new methods for
nonlinear systems.
Let F : D Rm Rm be sufficiently differentiable function and a zero
point of the nonlinear system F(x) = 0. Then, the iterative method (12) for
F(x) = 0 can be written as follows:

y = xn [zn , xn ; F]1 F (xn ) ,

n
un = G4 (xn ) ,



 

1

F (un ) ,
xn+1 = un [un , xn ; F] + un , yn ; F yn , xn ; F

(34)

where zn = xn + F (xn ) , un = G4 (xn ) stands for a fourth-order Steffensen


type method, and [, ; F] denotes the first order divided difference of F. For
the method defined by (34), we have the following convergence analysis.
Theorem 2 Let Rm be a solution of the system F(x) = 0 and F : D Rm
Rm be suf f iciently dif ferentiable in an open neighborhood D of . Suppose
that F  (x) is continuous and nonsingular in and un = G4 (xn ) is a fourth-order
Stef fensen type method which satisf ies

= un = Be4 + O e5 ,

(35)

where B L4 (Rm , Rm ), and e = xn . Then the convergence order of method


(34) is seven.
Proof We consider the first order divided difference operator of F as a
mapping [, ; F] : D D Rm Rm L (Rm ) (see [9, 10])



x + h, x; F =

F  (x + th) dt, (x, h) Rm Rm .

(36)



Expanding F (x + th) in Taylor series at the point x and integrating, we
obtain

1
0

F  (x + th) dt =F  (x) +

1 
1
F (x) h + F  (x) h2 + O h3 .
2
6

(37)

Since e = xn , by developing F(xn ) and its derivatives in a neighborhood of


and assuming that  = [F ()]1 exists, we have



F (xn ) = F  () e + A2 e2 + A3 e3 + A4 e4 + A5 e5 + O e6 ,

(38)

436

Numer Algor (2013) 62:429444

where Ai = i!1  F (i) () Li (Rm , Rm ) . From (38) the derivatives of F(xn ) can
be written as



(39)
F  (xn ) = F  () I + 2A2 e + 3A3 e2 + 4A4 e3 + 5A5 e4 + O e5 ,



F  (xn ) = F  () 2A2 + 6A3 e + 12A4 e2 + 20A5 e3 + O e4 ,

(40)




F  (xn ) = F  () 6A3 + 24A4 e + 60A5 e2 + O e3 .

(41)

From (36)(41) and noting that zn = xn + F (xn ), we obtain


[zn , xn ; F] = F  () I + 2A2 + A2 F  () e+ O e2 = F  () D (e)+ O e2 ,


(42)

where D (e) = I + 2A2 + A2 F  () e and I is the identity matrix. From (42),


we obtain

[zn , xn ; F]1 = D (e)1  + O e2 .


(43)
Then, we compel the inverse of D(e) to be (see [11, 12])

D (e)1 = I + X2 e + O e2 ,

(44)

such that X2 verify


D (e) D (e)1 = D (e)1 D (e) = I.
Solving the system (45), we obtain


X2 = 2A2 + A2 F  (a) ,

(45)

(46)

then,



[zn , xn ; F]1 = I 2A2 + A2 F  () e + O e2 .

(47)

From (34), (38) and (47), we arrive at


E = yn = e [zn , xn ; F]1 F (xn ) = A2 + A2 F  () e2 + O e3 .

(48)

Together with (35)(37), (38)(41) and (48), we have





yn , xn ; F = F  () I + A2 (E + e) + A3 (E + e) e + A4 e3 + O e4 , (49)

[un , xn ; F] = F  () I + A2 e + A3 e2 + A4 e3 + O e4 ,

(50)

un , yn ; F = F  () I + A2 E + O e4 ,

(51)


 



[un , xn ; F] + un , yn ; F yn , xn ; F = F  () I A3 Ee + O e4 .

(52)

Numer Algor (2013) 62:429444

By a similar argument, we have




 

1

[un , xn ; F] + un , yn ; F yn , xn ; F
= I + A3 Ee + O e4 ,

F (un ) = F  () + O 2 = F  () + O 2 .

437

(53)
(54)

Therefore, combining (34), (37) with (53)(54), we obtain the error equation:


 

1
en+1 = xn+1 = [un , xn ; F] + un , yn ; F yn , xn ; F
F (un )

= I + A3 Ee + O e4 + O 2 = A3 Ee + O e8

= A3 B A2 + A2 F  () e7 + O e8 .

(55)

This implies that the method defined by (34) is seventh-order convergent. This
completes the proof.

4 The concrete iterative methods


Many new three-step seventh-order Steffensen type iterative methods can be
obtained by taking the different choices of un in (34). In what follows, we
give some specific iterative forms of (34) for solving nonlinear equations and
nonlinear systems.
Method 1 Let un be defined by (3), then we obtain the following seventh-order
method

yn = xn [zn , xn ; F]1 F (xn ) ,


 
 


un = yn yn , xn ; F 1 yn , xn ; F zn , yn ; F
(56)

1

+ [zn , xn ; F]) yn , xn ; F
F (yn ) ,



 

1

F (un ) .
xn+1 = un [un , xn ; F] + un , yn ; F yn , xn ; F
Method 2 Let un be defined by (4), then we obtain another seventh-order
method

yn = xn [zn , xn ; F]1 F (xn ) ,


 


1
(57)
F (yn ) ,
un = yn yn , xn ; F + yn , zn ; F [zn , xn ; F]



 

1

xn+1 = un [un , xn ; F] + un , yn ; F yn , xn ; F
F (un ) .

5 Numerical results
In this section, we compare the related methods by numerical experiments.

438

Numer Algor (2013) 62:429444

Example 1 Method 1 (M71, (56)) and Method 2 (M72, (57)) are used to solve
some nonlinear equations and the computation results are compared with
Steffensens second-order method (S2, (2)), Rens fourth-order method (R4,
(4)), Corderos sixth-order method (C6, (5)) and Bis seventh-order method
(B7, (58)), see Table 1. The Bis seventh-order method [13] is as follows:

f (xn )

,
yn = xn 

f (xn )

f (xn ) + f (yn )
f (yn )
un = y n
,
f (xn ) + ( 2) f (yn ) f  (xn )

f (zn )


xn+1 = zn 
,
f zn , yn + f [zn , xn , xn ] (zn yn )

(58)

where R is a constant.
Table 1 shows the difference of between the root a and the approximation
value xn , where is the exact root computed with 1200 significant digits and xn
is computed by using the same total number of function evaluation (TNFE) for
all methods. The absolute values of function (| f (xn )|) and the computational
orders of convergence (ACOC) are also shown in Table 1. Here, the TNFE

Table 1 Comparison of various iterative methods using the same total number of function
evaluations for all methods (TNFE=12)

f1 , x0 = 1
|xn |
| f (xn )|
ACOC
f2 , x0 = 1.8
|xn |
| f (xn )|
ACOC
f3 , x0 = 2.0
|xn |
| f (xn )|
ACOC
f4 , x0 = 2
|xn |
| f (xn )|
ACOC
f5 , x0 = 1
|xn |
| f (xn )|
ACOC
f6 , x0 = 3.6
|xn |
| f (xn )|
ACOC

S2

R4

C6

B7( = 1)

M71

M72

5.4267e-89
9.0822e-89
2.00000000

6.8055e-309
1.1390e-308
4.00000000

1.4208e-205
2.3778e-205
5.99999994

9.7328e-380
1.6289e-380
6.999999999

1.0215e-437
1.7096e-437
7.00000000

4.8742e-419
8.1576e-419
7.00000000

3.8336e-49
1.0596e-48
2.00000000

6.2179e-215
1.7186e-214
4.00000000

1.2499e-281
3.4547e-281
5.99999999

7.5226e-289
2.0792e-288
6.999999274

2.2956e-382
6.3449e-382
7.00000000

7.1118e-412
1.9656e-411
7.00000000

7.4293e-144
7.3082e-144
2.00000000

3.6019e-406
3.5432e-406
4.00000000

2.6016e-159
2.5592e-159
5.99999697

1.2495e-280
1.2291e-280
7.000000165

5.2507e-443
5.1651e-443
7.00000000

9.3078e-490
9.1560e-490
7.00000000

9.5534e-144
1.1073e-143
2.00000000

1.4710e-446
1.7051e-446
4.00000000

2.6296e-198
3.0481e-198
5.99999960

7.8370e-351
9.0839e-351
7.000000011

2.9126e-563
3.3760e-563
7.00000000

5.4226e-553
6.2854e-553
7.00000000

1.5268e-24
2.3291e-24
2.00000012

4.1329e-138
6.3047e-138
4.00000000

9.7854e-180
1.4928e-179
6.00000309

8.0146e-280
1.2226e-279
7.000000306

4.0069e-243
6.1126e-243
6.99999746

4.2145e-282
6.4292e-282
6.99999973

1.3016e-116
7.8392e-117
2.00000000

4.5070e-397
2.7145e-397
4.00000000

4.9403e-258
2.9754e-258
5.99999999

9.9073e-420
5.9670e-420
7.000000000

2.4313e-505
1.4643e-505
7.00000000

1.5104e-521
9.0967e-522
7.00000000

Numer Algor (2013) 62:429444

439

for all methods is 12. The computational order of convergence (ACOC) [14] is
defined by



ln xn+1 xn xn xn1


.

(59)
ln xn xn1 xn1 xn2
The numerical results in Table 1 are for the following nonlinear functions:
f1 (x) = cos x x,

0.73908513321516064.

f2 (x) = 10xex 1,

1.6796306104284499.

1
f3 (x) = sin (x) x,
3

2.2788626600758283.

f4 (x) = ex + cos x,

1.7461395304080124.

f5 (x) = ex 1.5 arctan (x) ,


2
f6 (x) = ln(x +x+2) x + 1,

0.76765326620127890

4.1525907367571583.

The new methods M71 and M72 are also compared with S2, L4, R4 in Tables 2,
3 and 4 by solving some nonlinear systems, where L4 is Lius method (3) and
2 is the 2-norm. S2 is defined as follows:

x
= xk J (xk , Hk )1 F (xk ) ,

k+1

J (xk , Hk ) = F xk + Hk e1 F (xk ) , , F xk + Hk ek F (xk ) Hk1 ,

H = diag f
( 1 (xk ) , f2 (xk ) , fn (xk )) , k = 0, 1, 2 ,
k
(60)
where F (x) = ( f1 (x) , f2 (x) , fn (x)) , x = (x1 , x2 , xn ) and R4, L4, M71,
M72 are carried out by the similar divided differences.
Example 2 For the nonlinear system of equations:

(x1 1)4 + ex2 x22 + 3x2 + 1 = 0,

4 sin (x1 1) ln x21 x1 + 1 x22 = 0.

(61)

The results shown in Table 2, where x0 = (1, 1) and (1.2713843079501,


0.88081907310266) .

1.9498e-1
3.5733e-1
1.3891e-2
2.5195e-2
1.4625e-2
2.6467e-2
2.3225e-4
4.0189e-4
2.1880e-4
3.7829e-4

xk 2
F (xk ) 2
xk 2
F(xk ) 2
xk 2
F (xk ) 2
xk 2
F(xk ) 2
xk 2
F(xk ) 2

S2

M72

M71

L4

R4

Method
1.1545e-2
2.4423e-2
3.6770e-9
6.3463e-9
2.4459e-9
4.6891e-9
2.9211e-28
5.2250e-28
3.4530e-28
6.1622e-28

Table 2 Numerical results for nonlinear system (61) by the methods

1.7492e-5
4.8264e-5
2.3547e-35
4.0689e-35
9.9155e-37
1.7789e-36
3.3844e-196
6.1340e-196
2.1743e-195
3.9133e-195

4
1.5490e-10
2.9008e-10
3.7739e-140
6.5236e-140
3.6144e-146
6.8419e-146

5
6.7827e-21
1.4813e-20

4.3272e-42
1.3337e-41

7.0247

7.0265

3.9952

4.0008

2.0461

ACOC

440
Numer Algor (2013) 62:429444

1.9897e-1
7.6188e-1
1.1505e-2
5.5122e-2
6.4345e-3
2.8921e-2
1.0276e-3
4.1677e-3
1.2122e-3
5.1147e-3

xk 2
F(xk ) 2
xk 2
F(xk ) 2
xk 2
F(xk ) 2
xk 2
F(xk ) 2
xk 2
F(xk ) 2

S2

M72

M71

L4

R4

Method
7.8414e-1
2.0217e-1
5.5234e-7
2.2122e-6
1.6596e-7
6.6593e-7
4.0880e-19
1.6386e-18
8.7426e-19
3.5047e-18

Table 3 Numerical results for nonlinear system (62) by the methods

1.0327e-2
4.1817e-2
3.1786e-24
1.2754e-23
2.6173e-26
1.0503e-25
4.8645e-127
1.9499e-126
9.9409e-125
3.9848e-124

3
2.7845e-4
1.1434e-3
3.5151e-93
1.4104e-92
1.6199e-101
6.5006e-101

5
8.9842e-6
3.8011e-7

6
9.7688e-15
4.1047e-14

ACOC

6.9967

7.0079

4.0000

3.9998

1.9946

Numer Algor (2013) 62:429444


441

9.4734e-2
8.1958e-2
7.8151e-3
5.3311e-3
1.0307e-2
6.3120e-3
3.2943e-5
2.4613e-5
8.0016e-6
7.9647e-6

xk 2
F(xk ) 2
xk 2
F(xk ) 2
xk 2
F(xk ) 2
xk 2
F(xk ) 2
xk 2
F(xk ) 2

S2

M72

M71

L4

R4

Method
2.1430e-2
1.2615e-2
9.4946e-9
6.4031e-9
3.9832e-8
2.5937e-8
1.3858e-31
6.7629e-32
9.7302e-36
4.8099e-36

Table 4 Numerical results for nonlinear system (63) by the methods

8.1565e-4
4.7067e-4
2.1042e-32
1.4196e-32
1.0412e-29
6.7679e-30
1.0054e-216
4.8431e-217
4.9447e-246
2.3861e-246

4
1.1729e-6
6.7448e-7
5.0793e-127
3.4268e-127
4.8475e-116
3.1510e-116

5
2.4186e-12
1.3913e-12

6
1.0287e-23
5.9174e-24

ACOC

7.0297

7.0192

4.0001

4.0000

2.0000

442
Numer Algor (2013) 62:429444

Numer Algor (2013) 62:429444

Example 3 For the nonlinear system of equations:



2 ex1 + arctan (x2
) = 0,
arctan x21 + x22 5 = 0.

443

(62)

The results shown in Table 3, where x0 = (1.5, 1.8) and (1.1290650391602,


1.9300808629035) .
Example 4 For the nonlinear system of equations:

x2 + x3 ex1 = 0,
x1 + x3 ex3 = 0,

x1 + x2 ex3 = 0.

(63)

The results shown in Table 4, where x0 = (1, 1.1, 1.1) and


(0.83202503981029, 1.1489837536958, 1.1489837536959) .

6 Conclusion
By theoretical analysis and numerical experiments, we confirm that the new
Steffensen type methods achieve seventh-order convergence. A development
of an inverse first-order divided difference operator for multivariable function is applied to prove the order of convergence of the new methods for
nonlinear systems. The new methods are also compared in performance and
efficiency with Steffensens method, Lius method, Rens method and Cordero
via numerical examples. It is observed that the new methods have better
performance. The new methods are very suitable to solve nonlinear equations
and are applicable to solve nonlinear systems. Thus, the presented methods in
this paper can be considered as an improvement of Steffensens method.
Acknowledgement The project supported by the National Natural Science Foundation of China
(No. 11071033). The authors thank the referee for the valuable comment and suggestion.

References
1. Ortega, J.M., Rheinbolt, W.C.: Iterative Solution of Nonlinear Equations in Several Variables.
Academic Press, New York (1970)
2. Traub, J.F.: Iterative Methods for the Solution of Equations. Prentice-Hall, New Jersey (1964)
3. Liu, Z., Zheng, Q., Zhao, P.: A variant of Steffensens method of fourth-order convergence
and its applications. Appl. Math. Comput. 2146, 19781983 (2010)
4. Ren, H., Wu, Q., Bi, W.: A class of two-step Steffensen type methods with fourth-order
convergence. Appl. Math. Comput. 209, 206210 (2009)
5. Cordero, A., Hueso, J., Matrinez, E., et al.: Steffensen type methods for solving nonlinear
equations. J. Comput. Appl. Math. 236, 30583064 (2012)
6. Zheng, Q., Zhao, P., Huang, F.: A family of fourth-order Steffensen-type methods with the
applications on solving nonlinear ODEs. Appl. Math. Comput. 217, 81968203 (2011)
7. Petkovic, M.S., Ilic, S., Dunic, J.: Derivative free two-point methods with and without memory
for solving nonlinear equations. Appl. Math. Comput. 217, 18871895 (2010)
8. Gautschi, W.: Numerical Analysis: An Introduction. Birkhuser, Boston (1997)

444

Numer Algor (2013) 62:429444

9. Grau-Snchez, M., Grau, A., Noguera, M.: Frozen divided difference scheme for solving
systems of nonlinears. J. Comput. Appl. Math. 235, 17391743 (2011)
10. Grau-Snchez, M., Grau, A., Noguera, M.: Ostrowski type methods for solving systems of
nonlinears. Appl. Math. Comput. 218, 23772385 (2011)
11. Cordero, A., Hueso, J.L., Martnez, E., Torregrosa, J.R.: A modified Newton-Jarratts composition. Numerical Algorithms 55, 8799 (2010)
12. Cordero, A., Hueso, J.L., Martnez, E., Torregrosa, J.R.: Efficient high-order methods based
on golden ratio for nonlinear systems. Appl. Math. Comput. 217, 45484556 (2011)
13. Bi, W., Ren, H., Wu, Q.: New family of seventh-order methods for nonlinear equations. Appl.
Math. Comput. 203, 408412 (2008)
14. Cordero, A., Torregrosa, J.R.: Variants of Newtons method using fifth-order quadrature
formulas. Appl. Math. Comput. 190, 686698 (2007)

Anda mungkin juga menyukai