70 CHAPTER 2.

SOLUTIONS OF EQUATIONS OF ONE VARIABLE

2.6 Müller’s Method

There are a number of root-finding problems for which the Secant, False Position,
and Newton’s methods will not give satisfactory results. They will not give rapid
convergence, for example, when the function and its derivative are simultaneously
close to zero. In addition, these methods cannot be used to approximate complex
roots unless the initial approximation is a complex number whose imaginary part
is nonzero. This often makes them a poor choice for use in approximating the roots
of polynomials, which, even with real coefficients, commonly have complex roots
occuring in conjugate pairs.
In this section we consider Müller’s method, which is a generalization of the
Secant method. The Secant method finds the zero of the line passing through points
on the graph of the function that corresponds to the two immediately previous
approximations, as shown in Figure 2.8(a). Müller’s method uses the zero of the
parabola through the three immediately previous points on the graph as the new
approximation, as shown in part (b) of Figure 2.8.

Figure 2.8

y y

x0 x1 x2 x x0 x1 x2 x3 x
f f

(a) (b)

Suppose that three initial approximations, p0 , p1 , and p2 , are given for a so-
lution of f (x) = 0. The derivation of Müller’s method for determining the next
approximation p3 begins by considering the quadratic polynomial

P (x) = a(x − p2 )2 + b(x − p2 ) + c

that passes through (p0 , f (p0 )), (p1 , f (p1 )), and (p2 , f (p2 )). The constants a, b, and
c can be determined from the conditions

f (p0 ) = a(p0 − p2 )2 + b(p0 − p2 ) + c,

f (p1 ) = a(p1 − p2 )2 + b(p1 − p2 ) + c,
 2.6. MÜLLER’S METHOD 71

and
f (p2 ) = a · 02 + b · 0 + c.
To determine p3 , the root of P (x) = 0, we apply the quadratic formula to
P (x). Because of round-off error problems caused by the subtraction of nearly equal
numbers, however, we apply the formula in the manner prescribed in Example 1 of
Section 1.4:
−2c
p3 − p 2 = √ .
b ± b2 − 4ac
This gives two possibilities for p3 , depending on the sign preceding the radical
term. In Müller’s method, the sign is chosen to agree with the sign of b. Chosen in
this manner, the denominator will be the largest in magnitude, which avoids the
possibility of subtracting nearly equal numbers and results in p3 being selected as
the closest root of P (x) = 0 to p2 .

[Müller’s Method] Given initial approximations p0 , p1 , and p2 , generate

2c
p3 = p2 − √ ,
b + sgn(b) b2 − 4ac

where

c = f (p2 ),
(p0 − p2 )2 [f (p1 ) − f (p2 )] − (p1 − p2 )2 [f (p0 ) − f (p2 )]
b = ,
(p0 − p2 )(p1 − p2 )(p0 − p1 )

and (p1 − p2 )[f (p0 ) − f (p2 )] − (p0 − p2 )[f (p1 ) − f (p2 )]

a= .
(p0 − p2 )(p1 − p2 )(p0 − p1 )
Then continue the iteration, with p1 , p2 , and p3 replacing p0 , p1 , and p2 .

The method continues until

√ a satisfactory approximation is obtained. Since the
method involves the radical b2 − 4ac at each step, the method approximates com-
plex roots when b2 − 4ac < 0, provided, of course, that complex arithmetic is used.

EXAMPLE 1 Consider the polynomial f (x) = 16x4 − 40x3 + 5x2 + 20x + 6. Using the program
MULLER25 with accuracy tolerance 10−5 and various inputs for p0 , p1 , and p2
produces the results in Tables 2.9, 2.10, and 2.11.

Table 2.9
p0 = 0.5, p1 = −0.5, p2 = 0
72 CHAPTER 2. SOLUTIONS OF EQUATIONS OF ONE VARIABLE

n pn f (pn )

3 −0.555556 + 0.598352i −29.4007 − 3.89872i

4 −0.435450 + 0.102101i 1.33223 − 1.19309i
5 −0.390631 + 0.141852i 0.375057 − 0.670164i
6 −0.357699 + 0.169926i −0.146746 − 0.00744629i
7 −0.356051 + 0.162856i −0.183868 × 10−2 + 0.539780 × 10−3 i
8 −0.356062 + 0.162758i 0.286102 × 10−5 + 0.953674 × 10−6 i

Table 2.10
p0 = 0.5, p1 = 1.0, p2 = 1.5

n pn f (pn )

3 1.28785 −1.37624
4 1.23746 0.126941
5 1.24160 0.219440 × 10−2
6 1.24168 0.257492 × 10−4
7 1.24168 0.257492 × 10−4

Table 2.11
p0 = 2.5, p1 = 2.0, p2 = 2.25

n pn f (pn )

3 1.96059 −0.611255
4 1.97056 0.748825 × 10−2
5 1.97044 −0.295639 × 10−4
6 1.97044 −0.295639 × 10−4

To use Maple to generate the first entry in Table 2.9 we define f (x) and the
initial approximations with the Maple statements

>f:=x->16*x^4-40*x^3+5*x^2+20*x+6;
>p0:=0.5; p1:=-0.5; p2:=0.0;

We evaluate the polynomial at the initial values

>f0:=f(p0); f1:=f(p1); f2:=f(p2);

and we compute c = 6, b = 10, a = 9, and p3 = −0.5555555558 + 0.5983516452i

using the Müller’s method formulas:
2.6. MÜLLER’S METHOD 73

>c:=f2;
>b:=((p0-p2)^2*(f1-f2)-(p1-p2)^2*(f0-f2))/((p0-p2)*(p1-p2)*(p0-p1));
>a:=((p1-p2)*(f0-f2)-(p0-p2)*(f1-f2))/((p0-p2)*(p1-p2)*(p0-p1));
>p3:=p2-(2*c)/(b+(b/abs(b))*sqrt(b^2-4*a*c));

The value p3 was generated using complex arithmetic, as is the calculation

>f3:=f(p3);
which gives f3 = −29.40070112 − 3.898724738i.
The actual values for the roots of the equation are −0.356062 ± 0.162758i,
1.241677, and 1.970446, which demonstrate the accuracy of the approximations
from Müller’s method.
Example 1 illustrates that Müller’s method can approximate the roots of poly-
nomials with a variety of starting values. In fact, the technique generally converges
to the root of a polynomial for any initial approximation choice. General-purpose
software packages using Müller’s method request only one initial approximation per
root and, as an option, may even supply this approximation.
Although Müller’s method is not quite as efficient as Newton’s method, it is
generally better than the Secant method. The relative efficiency, however, is not
as important as the ease of implementation and the likelihood that a root will be
found. Any of these methods will converge quite rapidly once a reasonable initial
approximation is determined.
When a sufficiently accurate approximation p∗ to a root has been found, f (x)
is divided by x − p∗ to produce what is called a deflated equation. If f (x) is a
polynomial of degree n, the deflated polynomial will be of degree n − 1, so the
computations are simplified. After an approximation to the root of the deflated
equation has been determined, either Müller’s method or Newton’s method can be
used in the original function with this root as the initial approximation. This will
ensure that the root being approximated is a solution to the true equation, not to
the less accurate deflated equation.
74 CHAPTER 2. SOLUTIONS OF EQUATIONS OF ONE VARIABLE

EXERCISE SET 2.6

1. Find the approximations to within 10−4 to all the real zeros of the following
polynomials using Newton’s method.

(a) P (x) = x3 − 2x2 − 5

(b) P (x) = x3 + 3x2 − 1
(c) P (x) = x4 + 2x2 − x − 3
(d) P (x) = x5 − x4 + 2x3 − 3x2 + x − 4

2. Find approximations to within 10−5 to all the zeros of each of the following
polynomials by first finding the real zeros using Newton’s method and then
reducing to polynomials of lower degree to determine any complex zeros.

(a) P (x) = x4 + 5x3 − 9x2 − 85x − 136

(b) P (x) = x4 − 2x3 − 12x2 + 16x − 40
(c) P (x) = x4 + x3 + 3x2 + 2x + 2
(d) P (x) = x5 + 11x4 − 21x3 − 10x2 − 21x − 5

3. Repeat Exercise 1 using Müller’s method.

4. Repeat Exercise 2 using Müller’s method.
5. Find, to within 10−3 , the zeros and critical points of the following functions.
Use this information to sketch the graphs of P .

(a) P (x) = x3 − 9x2 + 12 (b) P (x) = x4 − 2x3 − 5x2 + 12x − 5

6. P (x) = 10x3 − 8.3x2 + 2.295x − 0.21141 = 0 has a root at x = 0.29.

(a) Use Newton’s method with p0 = 0.28 to attempt to find this root.
(b) Use Müller’s method with p0 = 0.275, p1 = 0.28, and p2 = 0.285 to
attempt to find this root.
(c) Explain any discrepancies in parts (a) and (b).

7. Use Maple to find the exact roots of the polynomial P (x) = x3 + 4x − 4.

8. Use Maple to find the exact roots of the polynomial P (x) = x3 − 2x − 5.
9. Use each of the following methods to find a solution accurate to within 10−4
for the problem
600x4 − 550x3 + 200x2 − 20x − 1 = 0, for 0.1 ≤ x ≤ 1.
2.6. MÜLLER’S METHOD 75

x2

x1

(a) Bisection method

(b) Newton’s method
(c) Secant method
(d) method of False Position
(e) Müller’s method

10. Two ladders crisscross an alley of width W . Each ladder reaches from the
base of one wall to some point on the opposite wall. The ladders cross at a
height H above the pavement. Find W given that the lengths of the ladders
are x1 = 20 ft and x2 = 30 ft and that H = 8 ft. (See the figure on page 58.)

11. A can in the shape of a right circular cylinder is to be constructed to contain

1000 cm3 . The circular top and bottom of the can must have a radius of 0.25
cm more than the radius of the can so that the excess can be used to form a
seal with the side. The sheet of material being formed into the side of the can
must also be 0.25 cm longer than the circumference of the can so that a seal
can be formed. Find, to within 10−4 , the minimal amount of material needed
to construct the can.

12. In 1224 Leonardo of Pisa, better known as Fibonacci, answered a mathemat-

ical challenge of John of Palermo in the presence of Emperor Frederick II.
His challenge was to find a root of the equation x3 + 2x2 + 10x = 20. He
first showed that the equation had no rational roots and no Euclidean irra-
√ √ √ √

tional root—that is, no root in one of the forms a ± b, a ± b, a ± b,
√ √
or a ± b, where a and b are rational numbers. He then approximated
the only real root, probably using an algebraic technique of Omar Khayyam
76 CHAPTER 2. SOLUTIONS OF EQUATIONS OF ONE VARIABLE

r 1 0.25
h

involving the intersection of a circle and a parabola. His answer was given in
the base-60 number system as
   2  3  4  5  6
1 1 1 1 1 1
1 + 22 +7 + 42 + 33 +4 + 40 .
60 60 60 60 60 60

How accurate was his approximation?

2.7. SURVEY OF METHODS AND SOFTWARE 77

2.7 Survey of Methods and Software

In this chapter we have considered the problem of solving the equation f (x) =
0, where f is a given continuous function. All the methods begin with an initial
approximation and generate a sequence that converges to a root of the equation, if
the method is successful. If [a, b] is an interval on which f (a) and f (b) are of opposite
sign, then the Bisection method and the method of False Position will converge.
However, the convergence of these methods may be slow. Faster convergence is
generally obtained using the Secant method or Newton’s method. Good initial
approximations are required for these methods, two for the Secant method and
one for Newton’s method, so the Bisection or the False Position method can be
used as starter methods for the Secant or Newton’s method.
Müller’s method will give rapid convergence without a particularly good initial
approximation. It is not quite as efficient as Newton’s method, but it is better than
the Secant method, and it has the added advantage of being able to approximate
complex roots.
Deflation is generally used with Müller’s method once an approximate root of a
polynomial has been determined. After an approximation to the root of the deflated
equation has been determined, use either Müller’s method or Newton’s method in
the original polynomial with this root as the initial approximation. This procedure
will ensure that the root being approximated is a solution to the true equation, not
to the deflated equation. We recommended Müller’s method for finding all the zeros
of polynomials, real or complex. Müller’s method can also be used for an arbitrary
continuous function.
Other high-order methods are available for determining the roots of polynomials.
If this topic is of particular interest, we recommend that consideration be given to
Laguerre’s method, which gives cubic convergence and also approximates complex
roots (see [Ho, pp. 176–179] for a complete discussion), the Jenkins-Traub method
(see [JT]), and Brent’s method. (see [Bre]), Both IMSL and NAG supply subroutines
based on Brent’s method. This technique uses a combination of linear interpolation,
an inverse quadratic interpolation similar to Müller’s method, and the bisection
method.
The netlib FORTRAN subroutine fzero.f uses a combination of the Bisection and
Secant method developed by T. J. Dekker to approximate a real zero of f (x) = 0
in the interval [a, b]. It requires specifying an interval [a, b] that contains a root and
returns an interval with a width that is within a specified tolerance. The FOR-
TRAN subroutine sdzro.f uses a combination of the bisection method, interpola-
tion, and extrapolation to find a real zero of f (x) = 0 in a given interval [a, b]. The
routines rpzero and cpzero can be used to approximate all zeros of a real polyno-
mial or complex polynomial, respectively. Both methods use Newton’s method for
systems, which will be considered in Chapter 10. All routines are given in single
and double precision. These methods are available on the Internet from netlib at
http://www.netlib.org/slatec/src.
Within MATLAB, the function ROOTS is used to compute all the roots, both
real and complex, of a polynomial. For an arbitrary function, FZERO computes a
root near a specified initial approximation to within a specified tolerance.
78 CHAPTER 2. SOLUTIONS OF EQUATIONS OF ONE VARIABLE

Maple has the procedure fsolve to find roots of equations. For example,
>fsolve(x^2 - x - 1, x);
returns the numbers −.6180339887 and 1.618033989. You can also specify a partic-
ular variable and interval to search. For example,
>fsolve(x^2 - x - 1,x,1..2);
returns only the number 1.618033989. The command fsolve uses a variety of spe-
cialized techniques that depend on the particular form of the equation or system of
equations.
Notice that in spite of the diversity of methods, the professionally written pack-
ages are based primarily on the methods and principles discussed in this chapter.
You should be able to use these packages by reading the manuals accompanying the
packages to better understand the parameters and the specifications of the results
that are obtained.
There are three books that we consider to be classics on the solution of nonlinear
equations, those by Traub [Tr], by Ostrowski [Os], and by Householder [Ho]. In
addition, the book by Brent [Bre] served as the basis for many of the currently
used root-finding methods.