Applied Mathematical Sciences, Vol. 12, 2018, no.

3, 137 - 146
HIKARI Ltd, www.m-hikari.com
https://doi.org/10.12988/ams.2018.811

A Review of Bracketing Methods for

Finding Zeros of Nonlinear Functions
Somkid Intep

Department of Mathematics, Faculty of Science

Burapha University, Chonburi, Thailand

Copyright c 2018 Somkid Intep. This article is distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any
medium, provided the original work is properly cited.

Abstract
This paper presents a review of recent bracketing methods for solving
a root of nonlinear equations. The performance of algorithms is verified
by the iteration numbers and the function evaluation numbers on a
number of test examples.

Mathematics Subject Classification: 65B99, 65G40

Keywords: Bracketing method, Nonlinear equation

1 Introduction
Many scientific and engineering problems are described as nonlinear equations,
f (x) = 0. It is not easy to find the roots of equations evolving nonlinearity
using analytic methods. In general, the appropriate methods solving the non-
linear equation f (x) = 0 are iterative methods. They are divided into two
groups: open methods and bracketing methods.
The open methods are relied on formulas requiring a single initial guess
point or two initial guess points that do not necessarily bracket a real root.
They may sometimes diverge from the root as the iteration progresses. Some
of the known open methods are Secant method, Newton-Raphson method, and
Muller’s method.
The bracketing methods require two initial guess points, a and b, that
bracket or contain the root. The function also has the different parity at these
138 Somkid Intep

two initial guesses i.e. f (a)f (b) < 0. The width of the bracket is reduced as
the iteration progresses until the approximate solution to a desired accuracy
is reached. By following this procedure, a root of f is certainly found. Some
of the known bracketing methods are Bisection method, Regula Falsi method
(or False Position), and Improved or modified Regula Falsi method.
In this article we will focus only on the bracketing methods.

2 Bracketing Methods
In the rest of the text, let f be a real and continuous function on an interval
[a, b], and f (a) and f (b) have different parity i.e. f (a)f (b) < 0. Therefore,
there is at least one real root r in the interval [a, b] of the equation f (x) = 0.

2.1 Bisection Method

This is one of the basic and simplest bracketing method for finding a root
of nonlinear equation f (x) = 0. This method is also known as dichotomy
or binary chopping or half-interval method. This method starts with two
initial values a and b which f (a)f (b) < 0. Then the approximate root is the
midpoint between a and b, say c = a+b 2
. Next, the new search interval is
established depending on the location of the real root r. If f (a)f (c) < 0, then
the new search interval [a, b] is replaced by [a, c]. Otherwise if f (a)f (c) > 0,
then the new interval is replaced by [c, b]. The procedure is repeated until the
approximate solution to a desired accuracy is reached.

2.2 Regula Falsi Method

The Bisection method is time consuming for some cases because it does not
consider f (a) or f (b) is closer to the zero of f (x). The next approximate root
c is always located at midpoint between a and b. For example, if f (a) = −1
and f (b) = 100, then it makes sense to choose approximate root c closer to
a than to b. The Regula Falsi method takes advantage of this observation
by finding the secant line from point (a, f (a)) to point (b, f (b)), and then
estimates the root at the x-intercept of the line. In general cases, the sequence
of approximate roots using the Regula Falsi method converge to the real root
of equation f (x) = 0 faster than that using the Bisection method.

2.3 Modified Regula Falsi Methods

Although the Regula Falsi method does consider the relative magnitudes of
f (a) and f (b), the method retains one of the endpoints of the interval brack-
eting the real root. Therefore it converges to the root very slowly for convex
A review of bracketing methods 139

or concave function f . There are numbers of modifications of the Regula Falsi

method. Dowell and Jarratt [2] proposed a superlinear variant of the classical
Regula Falsi method called Illinois method. The method follows the Regula
Falsi method except the next iteration approximations which the new function
value f (a) or f (b) is replaced by f (a)/2 or f (b)/2, respectively. Wu et al.
[16] combined the classical Regula Falsi method with Steffensen’s method and
found that both the sequence of diameters {(bn − an )} and the sequence of
iterations {(xn − r)} are quadratically convergent to zero. A combination of
the classical Regula Falsi method and a Newton-like method was proposed by
Parida and Gupta [9]. As a result of this method, both of the sequences are cu-
bically convergent to zero. Naghipoor et al. [6] used the classical Regula Falsi
method as a predictor and their new algorithm as a corrector. The corrector
is the intersection of the linear interpolation corresponding to the Regula Falsi
method and the straight line that connect points (a, 0) to (b, −kf (b)) or that
connect points (b, 0) to (a, −kf (a)) depending on the location of the predic-
tor. Compare to other methods, this technique is quite simple and interesting.
However, it has only some result of convergence but not to quadratic. Other
methods can be seen in [1, 7, 12].

2.4 Other Combination Iteration Methods

Wu and Fu [15] combined the Bisection method with a Newton-like method
to show quadratic convergence of those two sequences. Zhu and Wu [17] later
derived the method, called AC, to show cubical convergence of the same se-
quences. Four years later Parida and Gupta [10] slightly modified form of q(x)
that last proposed by Zhu and Wu, and showed that the new hybrid of the
Bisection method and a Newton-like method has cubical convergence of these
two sequences. However, numerical examples supporting in their work showed
that the algorithm AC is better than the algorithm of Parida and Gupta in
most the examples. In 2010, Gottlieb and Thompson [3] fitted a quadratic
polynomial to the function at the endpoints and midpoint of the interval. The
root of this quadratic interpolation then becomes one of the endpoints of the
new interval in the next iteration. The function and the associated quadratic
will become more linear over the search interval, thus improving convergence
speed even more. Suhadolnik [13] also fitted a quadratic polynomial to the
three points of the function. The first and second points are the endpoints
of the interval while the third point is computed by using the Bisection or
Regula Falsi algorithm depending on the observation of an interpolating line
slope. The switching mechanism between the Bisection and Regula Falsi pre-
vents slowness of convergence. In 2013, Suhadolnik [14] later combined only
the Bisection method to the classical Muller’s method, and showed that the
order of convergence is at least 1.618. In the same year, Hafiz [4] proposed a
140 Somkid Intep

method, called BRFC, by using combination of the Bisection and Regula Falsi
with the quadratic polynomial interpolation based on one endpoint, midpoint
and Regula Falsi point.

2.5 Alternative Iteration Method

Recently, Neamvonk et al. [8] proposed a new algorithm for finding a root
of nonlinear equation. The algorithm used nonlinear regression method in
form of logarithm function. Their numerical results showed that the proposed
method is faster than Bisection method and Regula Falsi method and as fast
as modified Regula Falsi method. However, the proposed method fails due to
the logarithm function if either one of interval endpoints is negative. Ide [5]
a year later used least square method with quadratic polynomial form instead
of logarithm function. Based on numerical results, the Ide method is slightly
faster than Neamvonk et al. method.

3 Bracketing Algorithms
In this section, some of bracketing algorithms based on good results judging
in relevant literatures are provided for finding a zero of a nonlinear function f
which is a real and continuous function on an initial interval [a, b]. The function
also satisfies the condition such that f (a)f (b) < 0, thus there is at least one
zero, r, of f . The maximum number of iterations, Nmax , and tolerance, , are
given for all following algorithms.

3.1 Bisection Algorithm (BS) [11]

1. [Initialization] n = 0, an = a, bn = b.
an +bn
2. [Bisection Iteration] cn = 2
.

3. [Convergence Test] If |f (cn )| ≤ , then the zero is cn . Stop.

4. If f (an )f (cn ) < 0 then bn = cn , else an = cn .

5. n = n + 1 and repeat step 2 until convergence is achieved.

3.2 Regula Falsi Algorithm (RF) [11]

1. [Initialization] n = 0, an = a, bn = b.
bn f (an )−an f (bn )
2. [Regula Falsi Iteration] xn = f (an )−f (bn )
.

3. [Convergence Test] If |f (xn )| ≤ , then the zero is xn . Stop.

A review of bracketing methods 141

4. If f (an )f (xn ) < 0 then bn = xn , else an = xn .

5. n = n + 1 and repeat step 2 until convergence is achieved.

The convergence criterion may use |bn − an | ≤  instead.

3.3 Improved Regula Falsi Algorithm (IRF)

The below algorithm is already corrected typos in [6].

1. [Initialization] n = 0, an = a, bn = b.
an f (bn )−bn f (an )
2. [Regula Falsi Iteration] cn = f (bn )−f (an )
.

3. [Convergence Test]
If f (an )f (cn ) < 0 then set k = | αf (cn )
bn −cn
|, compute xn = (1+k)an f (bn )−bn f (an )
(1+k)f (bn )−f (an )

if f (xn )f (an ) < 0 then bn = xn , else an = xn , bn = cn

else set k = | αf (cn )

an −cn
|, compute xn = (1+k)bn f (an )−an f (bn )
(1+k)f (an )−f (bn )

if f (xn )f (an ) > 0 then an = xn , else an = cn , bn = xn .

4. If |f (xn )| ≤ 1 or |xn − xn−1 | ≤ 2 , then the zero is xn . Stop.

5. n = n + 1 and repeat step 2 until convergence is achieved.

3.4 Bisection Newton-like Algorithm (AC) [17]

1. [Initialization] n = 0, an = a, bn = b, xn = an (or bn ).
an +bn
2. [Bisection Iteration] cn = 2
.

3. [Convergence Test] If |f (cn )| ≤ 1 , then the zero is cn . Stop.

4. If f (an )f (cn ) > 0 then ān = cn , b̄n = bn , else ān = an , b̄n = cn .

5. [Newton-like Iteration]
q(xn ) = f (xn )−2f (xn −f2f(x2 (x
n ))+f (xn −2f (xn ))
n)
− f (xn )−2f (xn −f (xn ))+f (xn −2f (xn ))
2f (xn )f (xn −f (xn −f (xn )))
.
(bn −an )|f (xn )|
w n = xn − 2[q(xn )f 2 (xn )+f (xn )−f (cn )]
.

6. If wn ∈ [ān , b̄n ] then xn+1 = wn

If f (an )f (wn ) < 0 then an+1 = ān , bn+1 = wn

else an+1 = wn , bn+1 = b̄n .
142 Somkid Intep

7. If wn 6∈ [ān , b̄n ] then an+1 = ān , bn+1 = b̄n and xn+1 = cn .

8. [Convergence Test]
If |f (xn+1 )| ≤ 1 or |bn+1 − an+1 | ≤ 2 , then the zero is xn+1 . Stop.
9. n = n + 1 and repeat step 2 until convergence is achieved.

3.5 Regula Falsi-Bisection-Parabolic Algorithm (RBP)

[13]
1. [Initialization] xa = a, xb = b.
xa +xb
2. [Initial Bisection] xc = 2
,n = 1, x0p = xc .
3. [Function values] fa = f (xa ), fb = f (xb ), fc = f (xc ).
fa −fc −fb
4. Set A = (xa −xc )(xa −xb )
+ (xb −xfcc)(x a −xb )
,
(fc −fa )(xb −xc ) (fc −fb )(xa −xc )
B= (xa −xc )(xa −xb )
− (x b −xc )(xa −xb )
, C = fc .
2C
5. [Root of parabola] xp = xc − √
B+sign(B) B 2 −4AC
.

6. fp = f (xp ).
7. [New endpoints of the interval]
If fa fp < 0 then xb = xp , fb = fp
If fa fc > 0 then xa = xc , fa = fc
else xa = xp , fa = fp
If fb fc > 0 then xb = xc , fb = fc .
8. [Switching mechanism between Bisection and Regula Falsi]
If |fa − fb | > 10|xa − xb | or |fa − fb | < 0.1|xa − xb | then
xa +xb
Bisecton: xc = 2
xa fb −xb fa
else Regula Falsi: xc = fb −fa
.

9. fc = f (xc ).
10. If n > Nmax then stop algorithm.
11. If n > 1 and |x0p − xp | <  then the zero is xp and stop algorithm.
12. If fp = 0 then the zero is xp and stop algorithm.
13. n = n + 1, x0p = xp and repeat step 4 until convergence is achieved.
A review of bracketing methods 143

3.6 Bisection-Regula Falsi-Parabolic Algorithm (BRFC)

[4]
1. [Initialization] xa = a, xb = b, n = 1.
xa +xb
2. [Initial Bisection] xc = 2
.
xb f (xa )−xa f (xb )
3. [Initial Regula Falsi] xs = f (xa )−f (xb )
.

4. If xs = xc then set xs = xb .

5. If f (xa ) = 0 or f (xc ) = 0 or f (xs ) = 0 or f (xb ) = 0 then print zero

x∗ , f (x∗ ) and stop algorithm.
f (xa ) f (xc ) f (xs )
6. Set A = (xa −xc )(xa −xs )
, B= (xc −xa )(xc −xs )
, C= (xs −xa )(xs −xc )
.

7. Set α = A + B + C, β = A(xs − xc ) + B(xs − xa ) + C(2xs − xa − xc ),

γ = f (xs ).
2γ
√
8. [Root of parabola] xp = xs − .
β+sign(β) β 2 −4αγ

9. If f (xp ) = 0 or |xp − xs | <  then print zero x∗ , f (x∗ ) and stop algorithm.

10. Set X̄ = sort(xa , xs , xc , xp , xb )

11. For i = 1 to 4 do

If f (X̄i )f (X̄i+1 ) < 0 then xa = X̄i , xb = X̄i+1 .

12. If n < Nmax then n = n + 1 and go to step 2

else print zero x∗ , f (x∗ ) and stop algorithm.

4 Numerical Examples
In general, researchers support their algorithm performances by the number
of iterations or the number of function evaluations. The less number of itera-
tions it takes, the more effectively the algorithm performs. Moreover, the less
number of evaluations it takes, the faster calculation time we get. Then if the
algorithm has both advantages, it is generally considered as a very good one.
In this section, we compare the performance of the algorithms that were pre-
sented in the previous section. In all algorithms, we take  = 1 = 2 = 10−13 as
the tolerance. Nonlinear functions with their initial interval and approximate
zeros are shown in Table 1.
Table 2 shows comparison of the iteration numbers among BS, RF, IRF,
AC, RBP and BRFC. The numbers of function evaluations are also presented
144 Somkid Intep

Table 1: List of functions, initial intervals, and approximate solutions

Initial interval Approximate zero
f (x)
[a, b] xn
2 5
f1 (x) = x − (1 − x) [0.1, 1] 0.3459548158483
f2 (x) = cos x − x3 [0.1, 1] 0.8654740331016
x
f3 (x) = xe − 1 [−1, 1] 0.5671432904098
f4 (x) = ln x [0.5, 5] 1
f5 (x) = x3 [−0.5, 13 ] 0.0000168486833
1
f6 (x) = x − sin x + 1 [−1.3, −0.5] −0.6294464840733
2
f7 (x) = ex +7x−30 − 1 [2.8, 3.1] 3

in the table. It can be seen from the iteration numbers that the RBP and
BRFC methods, which are combination of Regula Falsi and Bisection with
second order polynomial interpolation technique, are very effective. The IRF
method is also considered as a quite good method for finding zeros for all
functions except the function f5 . It suffers due to the concavity of f5 in a
certain search intervals. This suffering also occur in the classical Regula Falsi
method. However, the BS method has the biggest number of iterations in most
cases. On the other hand, if we consider the number of function evaluations,
the RBP method is the best one as it takes less number of function evaluations
in most cases. For instance, the RBP method takes13 function evaluations for
test function f7 while the BRFC method takes 14 and the others are even
more.

Table 2: The number of iterations of different bracketing methods

Number of iterations n
f (x)
BS RF IRF AC RBP BRFC
2 5
f1 (x) = x − (1 − x) 39 25 5 8 4 5
f2 (x) = cos x − x3 42 14 5 43 3 3
x
f3 (x) = xe − 1 44 26 6 6 4 4
f4 (x) = ln x 44 24 6 9 4 4
f5 (x) = x3 12 > 105 > 105 7 11 6
1
f6 (x) = x − sin x + 1 43 17 5 43 4 5
x2 +7x−30
f7 (x) = e −1 43 47 7 11 5 4
Number of function eval- n + 1 n + 2 2n + 2 6n+1 2n+3 3n+2
uations
A review of bracketing methods 145

5 Conclusion
The aim of this paper is to present a review of recent bracketing methods for
finding a zero of nonlinear functions. Several algorithms have been developed
from time to time. Some bracketing algorithms are chosen to present and test
numerically with some functions in order to indicate their performances. Based
on the numerical results, the RBP, which has both less number of iterations
and less number of function evaluations, is considered as the best method.
However, this method is quite complicated. It would be better if there is
anyone have simpler structure with more or less number of iterations.

References
[1] J. Chen, New modified regula falsi method for nonlinear equations, Ap-
plied Mathematics and Computation, 184 (2007), 965 - 971.
https://doi.org/10.1016/j.amc.2006.05.203

[2] M. Dowell and P. Jarratt, A modified regula falsi method for computing
the root of an equation, BIT, 11 (1971), 168 - 174.
https://doi.org/10.1007/bf01934364

[3] R.G. Gottlieb and B.F. Thompson, Bisected direct quadratic regula falsi,
Applied Mathematical Science, 4 (2010), 709 - 718.

[4] M.A. Hafiz, A new combined bracketing method for solving nonlinear
equations, Journal of Mathematical and Computational Science, 3 (2013),
87 - 93.

[5] N.A.D. Ide, A new algorithm for solving nonlinear equations by using
least square method, Mathematics and Computer Science, 1 (2016), 44 -
47.

[6] J. Naghipoor, S.A. Ahmadian and A.R. Soheili, An improved regula falsi
method for finding simple zeros of nonlinear equations, Applied Mathe-
matics Sciences, 2 (2008), 381 - 386.

[7] A. Neamvonk, A modified regula falsi method for solving root of nonlinear
equations, Asian Journal of Applied Sciences, 3 (2015), 776 - 778.

[8] J. Neamvonk, B. Phuenaree and A. Neamvonk, A new method for finding

root of nonlinear equation by using nonlinear regression, Asian Journal
of Applied Sciences, 3 (2015), 818 - 822.
146 Somkid Intep

[9] P.K. Parida and D.K. Gupta, An improved regula-falsi method for enclos-
ing simple zeros of nonlinear equations, Applied Mathematics and Compu-
tation, 177 (2006), 769 - 776. https://doi.org/10.1016/j.amc.2005.11.034

[10] P.K. Parida and D.K. Gupta, A cubic convergent iterative method for
enclosing simple roots of nonlinear equations, Applied Mathematics and
Computation, 187 (2007), 1544 - 1551.
https://doi.org/10.1016/j.amc.2006.09.071

[11] V.A. Patel, Numerical Analysis, Harcourt Brace College Publishers,

Florida, 1994.

[12] S. Shaw and B. Mukhopadhyay, An improved regula falsi method for find-
ing simple roots of nonlinear equations, Applied Mathematics and Compu-
tation, 254 (2015), 370 - 374. https://doi.org/10.1016/j.amc.2014.12.041

[13] A. Suhadolnik, Combined bracketing methods for solving nonlinear equa-

tions, Applied Mathematics Letters, 25 (2012), 1755 - 1760.
https://doi.org/10.1016/j.aml.2012.02.006

[14] A. Suhadolnik, Superlinear bracketing method for solving nonlinear equa-

tions, Applied Mathematics and Computation, 219 (2013), 7369 - 7376.
https://doi.org/10.1016/j.amc.2012.12.064

[15] X. Wu and D. Fu, New high-order convergence iteration methods with-

out employing derivatives for solving nonlinear equations, Computers and
Mathematics with Applications, 41 (2001), 489 - 495.
https://doi.org/10.1016/S0898-1221(00)00290-X

[16] X. Wu, Z. Shen and J. Xia, An improved regula falsi method with
quadratic convergence of both diameter and point for enclosing simple
zeros of nonlinear equations, Applied Mathematics and Computation, 144
(2003), 381 - 388. https://doi.org/10.1016/S0096-3003(02)00414-9

[17] Y. Zhu and X. Wu, A free-derivative iteration method of order three

having convergence of both point and interval for nonlinear equations,
Applied Mathematics and Computation, 137 (2003), 49 - 55.
https://doi.org/10.1016/S0096-3003(02)00029-2

Received: January 16, 2018; Published: February 6, 2018