 # New modification of Maheshwari’s method with optimal eighth order convergence for solving nonlinear equations

• Somayeh Sharifi , Massimiliano Ferrara , and Stefan Siegmund
From the journal Open Mathematics

## Abstract

In this paper, we present a family of three-point with eight-order convergence methods for finding the simple roots of nonlinear equations by suitable approximations and weight function based on Maheshwari’s method. Per iteration this method requires three evaluations of the function and one evaluation of its first derivative. These class of methods have the efficiency index equal to 8141.682. We describe the analysis of the proposed methods along with numerical experiments including comparison with the existing methods. Moreover, the attraction basins of the proposed methods are shown with some comparisons to the other existing methods.

MSC 2010: 65H05; 37F10

## 1 Introduction

Finding roots of nonlinear functions f (x) = 0 by using iterative methods is a classical problem which has interesting applications in different branches of science, in particular, physics and engineering. Therefore, several numerical methods for approximating simple roots of nonlinear equations have been developed and analyzed by using various techniques based on iterative methods in the recent years. The second order Newton-Raphson’s method xn+1=xnf(xn)f(xn) is one of the best-known iterative methods for finding approximate roots and it requires two evaluations for each iteration step, one evaluation of f and one of f′ [1, 2].

Kung and Traub  conjectured that no multi-point method without memory with n evaluations could have a convergence order larger than 2n–1. A multi-point method with convergence order 2n–1 is called optimal. The efficiency index provides a measure of the balance between those quantities, according to the formula p1/n, where p is the convergence order of the method and n is the number of function evaluations per iteration.

Many methods are described of which we note e.g. , [4-7]. Using inverse interpolation, Kung and Traub  constructed two general optimal classes without memory. Since then, there have been many attempts to construct optimal multi-point methods, utilizing e.g. weight functions [8-16].

In this paper, we construct a new class of optimal eight order of convergence based on Maheshwari’s method. This paper is organized as follows: Section 2 is devoted to the construction and convergence analysis of the new class. In Section 3, the new methods are compared with a closest competitor in a series of numerical examples. In addition, comparisons of the basin of attraction with other methods are illustrated in Section 3. Section 4 contains a short conclusion.

## 2 Description of the method and convergence analysis

### 2.1 Three-point method of optimal order of convergence

In this section we propose a new optimal three-point method based on Maheshwari’s method  for solving nonlinear equations. The Maheshwari’s method is given by

yn=xnf(xn)f(xn),xn+1=xn+1f(xn)f2(xn)f(yn)f(xn)f2(yn)f(xn),(n=0,1,...),(1)

where x0 is an initial approximation of x*. The convergence order of (1) is four with three functional evaluations per iteration such that this method is optimal. We intend to increase the order of convergence of method (1) by an additional Newton’s step. So we have

yn=xnf(xn)f(xn),zn=xn+1f(xn)f2(xn)f(yn)f(xn)f2(yn)f(xn),xn+1=znf(zn)f(zn).(2)

Method (2) uses five function evaluations with order of convergence eight. Therefore, this method is not optimal. In order to decrease the number of function evaluations, we approximate f′ (zn) by an expression based on f (xn), f (yn), f(zn) and f′(xn). Therefore

f(zn)f(xn)F(xn,yn,zn)H(sn),

where

F(xn,yn,zn)=f3(yn)(f(xn)10f(yn))+4f2(xn)(f2(yn)+f(xn)f(yn))f(xn)(2f(xn)f(yn))2(f(yn)f(zn)),(3)

and H(.) is a weight function which will be specified later, and sn=f(zn)f(xn).

We have

yn=xnf(xn)f(xn),zn=xn+1f(xn)f2(xn)f(yn)f(xn)f2(yn)f(xn),xn+1=znf(zn)f(xn)F(xn,yn,zn)H(sn),(4)

where F (xn; yn; zn) and sn are defined as above.

### 2.2 Convergence analysis

In the following theorem we provide sufficient conditions on the weight function H(sn), which imply that method (4) has convergence order eight.

Theorem 2.1

Assume that function f : D → ℝ is eight times continuously differentiable on an interval D ⊂ ℝ and has a simple zerox* ∈ D. Moreover, H is one time continuously differentiable. If the initial approximationx0is sufficiently close tox* then the class defined by (4) converges tox* and the order of convergence is eight under the conditions

H(0)=1,H(0)=2,

with the error term

en+1=12c22(4c22c3)(c238c2c3+2c4)en8+O(en9),

where en := xnx* for n ∈ ℕ andck:=f(k)(x)k!f(x)for k = 2, 3, ….

Proof

Let en, y := ynx*, en, z := znx* for n ∈ ℕ. Using the fact that f(x*) = 0, Taylor expansion of f at x* yields

f(xn)=f(x)(en+c2en2+c3en3++c8en8)+O(en9),(5)

and

f(xn)=f(x)1+2c2en+3c3en2+4c4en3++9c9en8+O(en9).(6)

Therefore,

f(xn)f(xn)=enc2en2+(2c222c3)en3+(4c23+7c2c33c4)en4+(8c2420c22c3+6c32+10c2c44c5)en5+(16c25+52c23c328c22c4+17c3c4c2(33c3213c5))en6+O(en7),

and hence

en,y=ynx=c2en2+(2c22+2c3)en3+(4c237c2c3+3c4)en4+(8c24+20c22c36c3210c2c4+4c5)en5+(16c2552c23c3+28c22c417c3c4+c2(33c3213c5))en6+O(en7).

For f(yn) we also have

f(yn)=f(x)en,y+c2en,y2+c3en,y3++c8en,y8+O(en,y9).(7)

Therefore, by substituting (5), (6) and (7) into (2), we get

en,z=znx=(4c23c2c3)en4+(27c24+26c22c32c322c2c4)en5+120c25196c23c3+39c22c47c3c4+c2(54c323c5)en6+O(en7).

For f(zn) we also get

f(zn)=f(x)en,z+c2en,z2+c3en,z3++c8en,z8+O(en,z9).(8)

From (5), (7) and (8) we obtain

F(xn,yn,zn)=1+2c2en+3c3en2+(8c23+2c2c3+4c4)en3+(832c2445c22c3+4c32+3c2c4+5c5)en4+(71678c26+1731c24c356c33+429c23c4245c2c3c4+9c42+c22(815c3284c5)+16c3c5)en6+O(en7).(9)

From (5) and (8) we have

sn=(4c23c2c3)en3+(31c24+27c22c32c322c2c4)en4+(151c25227c23c3+41c22c47c3c4+c2(57c323c5))en5+(592c26+1266c24c3+38c33325c23c4+170c2c3c46c4210c3c5+c22(608c32+55c5))en6+O(en7).(10)

Expanding H at 0 yields

H(sn)=H(0)+H(0)sn+O(sn2).(11)

Substituting (5)-(11) into (4) we obtain

en+1=xn+1x=R4en4+R5en5+R6en6+R7en7+R8en8+O(en9),

where

R4=c2(4c22c3)(1+H(0)),R5=0,R6=0,R7=c22(4c22+c3)2(2+H(0)).

By setting R4 = R7 = 0 and R8 ≠ 0 the convergence order becomes eight. Obviously,

H(0)=1R4=0,H(0)=2R7=0,

consequently, the error term becomes

en+1=12c22(4c22c3)(c328c2c3+2c4)en8+O(en9),

which completes the proof of the theorem. □

In what follows we give some concrete explicit representations of (4) by choosing different weight functions satisfying the provided condition for the weight function H(sn) in Theorem 2.1.

Method 1

Choose the weight function H(sn) as:

H(sn)=1+2sn,(12)

where sn=f(zn)f(xn).

The function H(sn) in (12) satisfies the assumptions of Theorem 2.1 and we get

yn=xnf(xn)f(xn),zn=xn+1f(xn)f2(xn)f(yn)f(xn)f2(yn)f(xn),xn+1=znf(zn)f(xn)(f(xn)+2f(zn)f(xn))F(xn,yn,zn),(13)

where F(xn, yn, zn) is evaluated by (3).

Method 2

Choose the weight function H(sn) as:

H(sn)=1+4sn1+2sn,(14)

where sn=f(zn)f(xn).

The function H(sn) in (14) satisfies the assumptions of Theorem 2.1 and we obtain

yn=xnf(xn)f(xn),zn=xn+1f(xn)f2(xn)f(yn)f(xn)f2(yn)f(xn),xn+1=znf(zn)f(xn)(f(xn)+4f(zn)f(xn)+2f(zn))F(xn,yn,zn),(15)

where F(xn, yn, zn) is evaluated by (3).

Method 3

Choose the weight function H(sn) as:

H(sn)=112sn,(16)

where sn=f(zn)f(xn).

The function H in (16) satisfies the assumptions of Theorem 2.1 and we get

yn=xnf(xn)f(xn),zn=xn+1f(xn)f2(xn)f(yn)f(xn)f2(yn)f(xn),xn+1=znf(zn)f(xn)(f(xn)f(xn)2f(zn))F(xn,yn,zn),(17)

where F(xn, yn, Zn) is evaluated by (3).

We apply the new methods (13), (15) and (17) to several benchmark examples and compare them with the existing three-point methods which have the same convergence order r = 8 and the same computational efficiency index equal to rθ=1.682, which is optimal for θ = 4 function evaluations per iteration [1, 2].

## 3 Numerical performance

In this section we test and compare our proposed methods with some existing methods. We compare methods (13), (15) and (17) with the following related three-point methods.

Bi, Ren and Wu’s method

The method by Bi et al.  is given by

yn=xnf(xn)f(xn),zn=yn+f(yn)f(xn)f(xn)+βf(yn)f(xn)+(β2)f(yn),xn+1=znf(zn)f[zn,yn]+f[zn,xn,xn](znyn)H(tn),(18)

where β=12 and weight function

H(tn)=1(1αtn)2,α=1,(19)

where tn=f(zn)f(xn).

If x0; x1, …, xn are points of D, the divided difference of order 1 can be expressed by

f[x0,x1]=f(x1)f(x0)x1x0,

and in general, the divided difference of order n is obtained by

f[x0,x1,...,xn]=f[x1,x2,...,xn]f[x0,x1,...,xn1]xnx0.

In addition, for x0 = x1 = … = xn = x, we write

f[x,x,...,x]=f(n+1)(x)(n+1)!.
Chun and Lee’s method

The method by Chun and Lee  is given by

yn=xnf(xn)f(xn),zn=ynf(yn)f(xn)11f(yn)f(xn)2,xn+1=znf(zn)f(xn)1(1H(tn)J(sn)P(un))2,(20)

with weight functions

H(tn)=βγ+tn+tn22tn32,J(sn)=βsn2,p(un)=γ+un2,(21)

where tn=f(yn)f(xn),sn=f(zn)f(xn),un=f(zn)f(yn) and β, γ ∈ ℝ

Sharma and Sharma’s method

The Sharma and Sharma method  is given by

yn=xnf(xn)f(xn),zn=ynf(yn)f(xn)f(xn)f(xn)2f(yn),xn+1=Znf[xn,yn]f(zn)f[xn,zn]f[yn,zn]W(tn),(22)

with weight function

W(tn)=1+tn1+αtn,α=1,(23)

where tn=f(zn)f(xn).

The three point method (4) is tested on a number of nonlinear equations. To obtain a high accuracy and avoid the loss of significant digits, we employed multi-precision arithmetic with 7000 significant decimal digits in the programming package of Mathematica 8 .

In order to test our proposed methods (13), (15) and (17), and also to compare them with the methods (18), (20) and (22), we compute the error and the approximated computational order of convergence (ACOC) that was introduced by Cordero et al. 

ACOCIn(xn+1xn)/(xnxn1)In(xnxn1)/(xn1xn2).

In Tables 1, 2, 3 and 4, the proposed methods (13), (15) and (17) with the methods (18), (20) and (22) have been tested on different nonlinear equations. It is clear that these methods are in accordance with the developed theory.

Table 1

Comparison for f(x)=In(1+x2)+ex23xsin(x), x* = 0, x0 = 0.35, for different methods (M) and weight functions (W-F).

M

W-F

|x1x*|

|x2x*|

|x3x*|

|x4x*|

ACOC

(13)

(12)

0.657e—4

0.466e—30

0.299e—239

0.865e—1913

8.0000

(15)

(14)

0.568e—4

0.145e—30

0.259e—243

0.272e—1945

8.0000

(17)

(16)

0.755e—4

0.141e—29

0.206e—235

0.423e—1882

8.0000

(18)

(19)

0.720e—4

0.584e—30

0.110e—238

0.175e—1908

8.0000

(20)

(21)

0.721e—4

0.230e—30

0.252e—242

0.528e—1938

7.9999

(22)

(23)

0.753e—4

0.619e—31

0.128e—247

0.453e—1981

8.0000

Table 2

Comparison for f(x) = In(1 − x + x2) + 4 sin(1 − x),x* = 1,x0 = 1.1, for different methods (M) and weight functions (W-F).

M

W-F

|x1x*|

|x2x*|

|x3x*|

|x4x*|

ACOC

(13)

(12)

0.444e—11

0.399e—94

0.170e—758

0.189e—6073

8.0000

(15)

(14)

0.445e—11

0.404e—94

0.187e—758

0.394e—6073

8.0000

(17)

(16)

0.443e—11

0.395e—94

0.155e—758

0.902e—6077

8.0000

(18)

(19)

0.423e—12

0.134e—114

0.445e—1037

0.211e—9339

8.0000

(20)

(21)

0.211e—11

0.264e—97

0.250e—782

0.181e—6264

7.9999

(22)

(23)

0.172e—11

0.581e—98

0.984e—790

0.663e—6324

8.0000

Table 3

Comparison for f(x)=x4+sin(πx2)5,x=2, x0 = 1.5, for different methods (M) and weight functions (W-F).

M

W-F

|x1x*|

|x2x*|

|x3x*|

|x4x*|

ACOC

(13)

(12)

0.783e—8

0.648e—64

0.142e—512

0.765e—4102

8.0000

(15)

(14)

0.749e—8

0.656e—64

0.855e—514

0.132e—4111

8.0000

(17)

(16)

0.816e—8

0.908e—64

0.212e—511

0.187e—4092

8.0000

(18)

(19)

0.673e—8

0.113e—64

0.726e—519

0.208e—4152

8.0000

(20)

(21)

0.433e—8

0.134e—66

0.116e—534

0.373e—4279

7.9999

(22)

(23)

0.642e—10

0.101e—81

0.389e—656

0.184e—5251

8.0000

Table 4

Comparison for f(x) = (x − 2)(x10 + x + 1)ex−1, x* = 2, x0 = 2.1, for different methods (M) and weight functions (W-F).

M

W-F

|x1x*|

|x2x*|

|x3x*|

|x4x*|

ACOC

(13)

(12)

0.119e—3

0.253e—26

0.106e—207

0.992e—1659

8.0000

(15)

(14)

0.143e—3

0.109e—25

0.124e—202

0.353e—1618

8.0000

(17)

(16)

0.916e—4

0.307e—27

0.493e—215

0.221e—1717

8.0000

(18)

(19)

0.183e—4

0.319e—33

0.278e—263

0.920e—2104

8.0000

(20)

(21)

0.344e—5

0.134e—38

0.702e—306

0.398e—2444

8.0000

(22)

(23)

0.239e—4

0.138e—32

0.170e—258

0.938e—2066

8.0000

### 3.1 Graphical comparison by means of attraction basins

We have already observed that all methods converge if the initial guess is chosen suitably. From the numerical point of view, the dynamical behavior of the rational function associated with an iterative method gives us important information about convergence and stability. Therefore, we now investigate the stability region. In other words, we numerically approximate the domain of attraction of the zeros as a qualitative measure of stability. To answer the important question on the dynamical behavior of the algorithms, we investigate the dynamics of the new methods and compare them with common and well-performing methods from the literature. In the following, we recall some basic concepts such as basin of attraction. For more details one can consult [19-22].

Let G : ℂ → ℂ be a rational map on the complex plane. The orbit of a point z ∈ ℂ is defined as

orb(z)={z,G(z),G2(z),}.

A point z0 ∈ ℂ is called a periodic point with minimal period m if Gm(z0) = z0, where m is the smallest integer with this property. A periodic point with minimal period 1 is called a fixed point. Moreover, a point z0 is called attracting if |G′(z0)| < 1, repelling if |G′(z0)| > 1, and neutral otherwise. The Julia set of a nonlinear map G(z), denoted by J(G), is the closure of the set of its repelling periodic points. The complement of J(G) is the Fatou set F(G), where the basin of attraction of the different roots lie . For the dynamical point of view, in fact, we take a 256 X 256 grid of the square [−3,3] × [−3,3] ∈ ℂ and assign a color to each point z0D according to the simple root to which the corresponding orbit of the iterative method starting from z0 converges, and we mark the point as black if the orbit does not converge to a root, in the sense that after at most 100 iterations it has a distance to any of the roots, which is larger than 10−3. In this way, we distinguish the attraction basins by their color for different methods.

We use the basins of attraction for comparing the iteration algorithms. The basin of attraction is a method to visually comprehend how an algorithm behaves as a function of the various starting points. In the following figures, the roots of each functions are drawn with a different color. In the basin of attractions, the number of iteration needed to achieve the solution is showed in darker or brighter colors. Black color denotes lack of convergences to any of the roots or convergence to the infinity.

We have tested several different examples, and the results on the performance of the tested methods were similar. Therefore we report the general observation here for test problems p1(z) = z2 − 1 with roots −1,1 and p2 (z) = z(z2 + 1) with roots 0, i, −i.

In Figures 1 and 2, basins of attractions of methods (13), (18), (20) and (22) with two test problems p1 (z) and p2(z) are illustrated from left to right respectively. As a result, in Figure 1 the basin of attraction of method (13) is similar to other methods, however in Figure 2, first two figures seem to produce larger basins of attraction than the last two figures. Fig. 1

Comparison of basin of attraction of methods (13), (18), (20) and (22) for test problem p1 (z) = z2 − 1 Fig. 2

Comparison of basin of attraction of methods (13), (18), (20) and (22) for test problem p2(z) = z3 + z

## 4 Conclusion

We presented a new optimal class of three-point methods without memory for approximating a simple root of a given nonlinear equation. Our proposed methods use five function evaluations for each iteration. Therefore they support Kung and Traub’s conjecture. Numerical examples show that our methods work and can compete with other methods in the same class, as well as we used the basin of attraction for comparing the iteration algorithms.

## Acknowledgement

The authors would like to express their appreciations from the MEDAlics, Research Centre at the University Dante Alighieri, Reggio Calabria, Italy for financial support.

## References

 Traub J.F., Iterative Methods for the Solution of Equations, Prentice Hall, Englewood Cliffs, N.J., 1964Search in Google Scholar

 Ostrowski A.M., Solution of Equations and Systems of Equations, Academic Pres, New York, 1966Search in Google Scholar

 Kung H.T., Traub J.F., Optimal order of one-point and multipoint iteration, J. Assoc. Comput. Mach., 1974, 21, 634-65110.1145/321850.321860Search in Google Scholar

 Jarratt P., Some efficient fourth order multipoint methods for solving equations, BIT Numer. Math., 1969, 9, 119-12410.1007/BF01933248Search in Google Scholar

 King R.F., A Family of Fourth Order Methods for Nonlinear Equations, SIAM J. Numer. Anal., 1973, 10, 876-87910.1137/0710072Search in Google Scholar

 Maheshwari A.K., A fourth-order iterative method for solving nonlinear equations, Appl. Math. Comput., 2009, 211, 383-39110.1016/j.amc.2009.01.047Search in Google Scholar

 Ferrara M., Sharifi S., Salimi M., Computing multiple zeros by using a parameter in Newton-Secant method, SeMA J., 2016, 10.1007/s40324-016-0074-0Search in Google Scholar

 Bi W., Ren H., Wu Q., Three-step iterative methods with eighth-order convergence for solving nonlinear equations, J. Comput. Appl. Math., 2009, 225, 105-11210.1016/j.cam.2008.07.004Search in Google Scholar

 Chun C., Lee M.Y., A new optimal eighth-order family of iterative methods for the solution of nonlinear equations, Appl. Math. Comput., 2013, 223, 506-51910.1016/j.amc.2013.08.033Search in Google Scholar

 Petković M.S., Neta B., Petković L.D., Džunić J., Multipoint Methods for Solving Nonlinear Equations, Elsevier/Academic Press, Amsterdam, 201310.1016/B978-0-12-397013-8.00002-9Search in Google Scholar

 Lotfi T., Sharifi S., Salimi M., Siegmund S., A new class of three-point methods with optimal convergence order eight and its dynamics, Numer. Algor., 2015, 68, 261-28810.1007/s11075-014-9843-ySearch in Google Scholar

 Salimi M., Lotfi T., Sharifi S., Siegmund S., submitted for publication, Optimal Newton-Secant like methods without memory for solving nonlinear equations with its dynamics, 201610.1080/00207160.2016.1227800Search in Google Scholar

 Matthies G., Salimi M., Sharifi S., Varona J.L., submitted for publication, An optimal three-point eighth-order iterative method without memory for solving nonlinear equations with its dynamics, 201610.1007/s13160-016-0229-5Search in Google Scholar

 Sharifi S., Salimi M., Siegmund S., Lotfi T., A new class of optimal four-point methods with convergence order 16 for solving nonlinear equations, Math. Comput. Simulation, 2016, 119, 69-9010.1016/j.matcom.2015.08.011Search in Google Scholar

 Sharifi S. Siegmund S., Salimi M., Solving nonlinear equations by a derivative-free form of the King’s family with memory, Calcolo, 2016, 53, 201-21510.1007/s10092-015-0144-1Search in Google Scholar

 Sharma J.R., Sharma R., A new family of modified Ostrowski’s methods with accelerated eighth order convergence, Numer. Algor., 2010, 54, 445-45810.1007/s11075-009-9345-5Search in Google Scholar

 Hazrat R., Mathematica, A Problem-Centered Approach, Springer-Verlag, 201010.1007/978-1-84996-251-3Search in Google Scholar

 Cordero A., Torregrosa J.R., Variants of Newton method using fifth-order quadrature formulas, Appl. Math. Comput., 2007, 190, 686-69810.1016/j.amc.2007.01.062Search in Google Scholar

 Amat S., Busquier S., Plaza S., Dynamics of the King and Jarratt iterations, Aequationes Math., 2005, 69, 212-22310.1007/s00010-004-2733-ySearch in Google Scholar

 Neta B., Chun C., Scott M., Basin of attractions for optimal eighth order methods to find simple roots of nonlinear equations, Appl. Math. Comput., 2014, 227, 567-59210.1016/j.amc.2013.11.017Search in Google Scholar

 Varona J.L., Graphic and numerical comparison between iterative methods, Math. Intelligencer, 2002, 24, 37-4610.1007/BF03025310Search in Google Scholar

 Vrscay E.R., Gilbert W.J., Extraneous fixed points, basin boundaries and chaotic dynamics for Schroder and Konig rational iteration functions, Numer. Math., 1988, 52, 1-1610.1007/BF01401018Search in Google Scholar

 Babajee D.K.R., Cordero A., Soleymani F., Torregrosa J.R., On improved three-step schemes with high efficiency index and their dynamics, Numer. Algor., 2014, 65, 153-16910.1007/s11075-013-9699-6Search in Google Scholar 