You are on page 1of 5

190 COMPUTER-BASED NUMERICAL AND STATISTICAL TECHNIQUES

If the location of some roots is known, first find these roots to a desired
accuracy and then apply this method on the reduced polynomial.

ASSIGNMENT 3.10
1. The equation f(x) = (x – 1)2 (x – 3)2 has roots at x = 1 and x = 3. Which of the following
methods can be applied to find all the roots?
(i) Bisection method
(ii) False-position method
(iii) Newton-Raphson method
Justify your answer.
2. A sphere of wood, 2 m in diameter, floating in water sinks to a depth d given by
d3 – 3d2 + 2.5 = 0
find d correct to 2 decimal places.
3. Discuss the working of modified Newton-Raphson method.
4. Find the root of the equation
x+1
f(x) ≡ sin x – = 0 near x = – .4
x−1
5. Give a comparative study of iterative methods.
6. Under what conditions does the Newton-Raphson method become linearly convergent?
Explain.

3.43 GRAEFFE’S ROOT-SQUARING METHOD

This method has a great advantage over the other methods in that it does not
require prior information about the approximate values, etc., of the roots. It is
applicable to polynomial equations only and is capable of giving all the roots.
Here below we discuss the case of the polynomial equation having real and
distinct roots.
Consider the polynomial equation
f(x) = xn + a1xn–1 + a2xn–2 + ...... + an–1x + an = 0 (46)
Separating the even and odd powers of x and squaring, we get
(xn + a2xn–2 + a4xn–4 + ......)2 = (a1xn–1 + a3xn–3 + a5xn–5 + ......)2
Putting x2 = y and simplifying, the new equation becomes
yn + b1yn–1 + b2yn–2 + ...... + bn–1y + bn = 0 (47)
where b1 = a12 + 2a2; b2 = a22 – 2a1a3 + 2a4 ...... bn = (– 1)n an2 (48)
ALGEBRAIC AND TRANSCENDENTAL EQUATIONS 191

If p1, p2, ......... , pn are the roots of (46), then the roots of (47) are
p12, p22, ......... , pn2.
Let us suppose that after m squarings, the new transformed equation is
zn + λ1zn–1 + ...... + λn–1z + λn = 0 (49)
whose roots are q1, q2, ......., qn such that qi = pi 2m , i = 1, 2, ......, n.
Assuming the order of magnitude of the roots as
| p1 | > | p2 | > ...... > | pn |, we have
| q1 | >> | q2 | >> ...... >> | qn | where >> stands for ‘much greater than’.

| q2 | q2 |qn | q
Thus = , ......, = n (50)
|q1 | q1 | qn −1 | qn − 1
Also qi being an even power of pi, is always positive.
Now, from (49), we have
FG q + q + ......IJ
2 3
Σq1 = – λ1 ⇒ λ1 = – q1 1 +
H q q 1K 1

F q + ......IJ
= q q G1 +
3
Σq1q2 = λ2 ⇒ λ2
H q
1 2 K 1

F q + ......IJ
= q q q G1 +
4
Σq1q2q3 = – λ3 ⇒ λ3
H q
1 2 3 K 1

............................................................................
q1q2q3 ...... qn = (– 1)n λn ⇒ λn = (– 1)n q1q2q3 ...... qn.

λ2 λ λ
Hence by (50), we find q1 ≈ – λ1; q2 ≈ – , q3 ≈ − 3 , ......, qn ≈ − n
λ1 λ2 λ n −1
But qi = pi2m

F λ I
= G−
1/2 m

H λ JK
i
∴ pi = (qi )1/2m (51)
i −1

We can thus determine p1, p2, ......, pn the roots of the equation (46).
Case 1. Double root. If the magnitude of λi is half the square of the
magnitude of the corresponding coefficient in the previous equation after a few
squarings, then it implies that pi is a double root of (46). We determine it as
follows:
λi λ i+1
qi = – and qi+1 = –
λ i−1 λi
192 COMPUTER-BASED NUMERICAL AND STATISTICAL TECHNIQUES

λ i+1 λ i+1
∴ qiqi+1 ≈ qi2 ≈ i.e., pi2m = qi2 = (52)
λ i−1 λ i−1

which gives the magnitude of the double root and substituting in (46), we can
find the sign.
Case 2. Complex roots. If pr and pr+1 form the complex pair Pr e ± iφ r , then
the co-efficient of xn–r in successive squarings would vary both in magnitude and
sign by an amount 2Prm cos mφr. For sufficiently large Pr and φr can be determined
by
m λ r +1 2m λr
Pr 2(2 )
≈ m
; 2Pr cos 2 φ r = − (53)
λ r−1 λ r−1
If there is only one pair of complex roots, say

Pr e ± iφ r = ξ r + iηr then ξr is given by


p1 + p2 + ...... + pr–1 + 2ξr + pr+2 + ...... + pn = – a1 (54)

and ηr = Pr 2 – ξ r 2 (55)

If there are two pairs of complex roots, say


± iφ s
Pr e ± iφ r = ξ r ± iηr and Ps e = ξ s ± iη s
where p1 + p2 + ...... + pr–1 + 2ξr + Pr+2 + ...... + ps–1 + 2ξs + ps+2 + ...... + pn = – a1
(56)

Fξ I = − La OP
JK MN n
ξs 1 1
2 GH P
r
r
2
+
Ps 2
n− 1
+
a1
+ ...... +
an Q (57)

and ηr = Pr 2 − ξ r 2 ; η s = Ps 2 − ξ s 2 (58)

EXAMPLES
Example 1. Apply Graeffe’s root squaring method to solve the equation
x3 – 8x2 + 17x – 10 = 0.
Sol. Here f(x) = x3 – 8x2 + 17x – 10 = 0 (59)
Clearly f(x) has three changes i.e., from + to –, – to + and + to –. Hence from
Descartes rule of signs f(x) may have three positive roots.
Rewriting (59) as x(x2 + 17) = (8x2 + 10) (60)
ALGEBRAIC AND TRANSCENDENTAL EQUATIONS 193

Squaring on both sides and putting x2 = y, we get


y(y + 17)2 = (8y + 10)2
or y3 + 34y2 + 289y = 64y2 + 160y + 100
or y(y2 + 129) = (30y2 + 100) (61)
Squaring again and putting y2 = z, we get
z(z + 129)2 = (30z + 100)2
or z3 + 258z2 + 16641z = 900z2 + 6000z + 10000
or z(z2 + 10641) = (642z2 + 10000) (62)
Squaring again and putting z2 = u, we get
u(u + 10641)2 = (642u + 10000)2
or u3 + 21282u2 + 113230881u = 412164u2 + 12840000u + 108
or u3 – 390882u2 + 100390881u – 108 = 0 (63)
If the roots of (59) are p1, p2, p3 and those of (63) are q1, q2, q3, then
p1 = (q1)1/8 = ( – λ1)1/8 = (390882)1/8 = 5.000411082 ≅ 5

p2 = (q2)1/8 = (– λ2/λ1)1/8 =
LM 100390881OP = 2.000811036 ≅ 2
1/8

N 378882 Q
p3 = (q3)1/8 = (– λ3/λ2)1/8 = M
L 10 OP = 0.99951247 ≅ 1
8

N 100390881Q
Now f(5) = f(1) = f(2) = 0.
Hence the roots are 5, 2, and 1.
Example 2. Find all the roots of the equation x4 – 3x + 1 = 0 by Graeffe’s method.
Sol. Here f(x) = x4 – 3x + 1 = 0 (64)
Now f(x) has two changes in sign i.e., + to – and – to +. Therefore it may
have two positive real roots.
Again f(– x) = x4 + 3x + 1. Since no change in sign of f(– x) there is no
negative root. But f(x), being of degree four, will have four roots of which two
are real positive and the remaining two are complex.
Rewriting (64) as x4 + 1 = 3x.
Squaring and putting x2 = y, we have
(y2 + 1)2 = 9y
Squaring again and putting, y2 = z
(z + 1)4 = 81z
194 COMPUTER-BASED NUMERICAL AND STATISTICAL TECHNIQUES

i.e., z4 + 4z3 + 6z2 – 77z + 1 = 0 (65)


or z4 + 6z2 +1=– z(4z2 – 77)
Squaring once again and putting z2 = u, we get
(u2 + 6u + 1)2 = u(4u – 77)2
or u4 – 4u3 + 654u2 – 5917u + 1 = 0 (66)
If p1, p2, p3, p4 are the roots of (64) and q1, q2, q3, q4 are the roots of (66),
then
p1 = (q1)1/8 = (– λ1)1/8 = (4)1/8 = 1.1892071

L λ
= M−
2 OP 1/8
LM 654 OP1/8
p2 = (q2 )1/8
N λ 1 Q
=
N4Q = 1.8909921

L λ
= M−
3 OP 1/8
=M
L 5917 OP 1/8
p3 = (q3)1/8
N λ 1 Q N 654 Q = 1.3169384

L λ
= M−
4 OP 1/8
=M
L 1 OP 1/8
p4 = (q4 )1/8
N λ 3 Q N 5917 Q = 0.3376659

From (65) and (66), we observe that the magnitudes of the co-efficients λ1
and λ4 have become constant.
⇒ p, p4 are the real roots and p2, p3 are complex roots. Let these complex
roots be

ρ2 e ± iφ 2 = ξ 2 ± iη2 . From (66), its magnitude is given by

3 λ 3 5917
ρ 2 2( 2 ) ≈ = ∴ ρ2 = 1.5780749
λ1 4

also from (64) the sum of the roots = 0, i.e., p1 + 2ξ2 + p4 = 0

1
∴ ξ2 = – (p + p4) = – 0.7634365
2 1

and η2 = ρ 2 2 − ξ 2 2 = 1.9074851 = 1.3811173

Hence, the four roots are 1.1892071, 0.3376659, – 0.7634365 ± 1.3811173i.

ASSIGNMENT 3.11
1. Find all the roots of the following equations by Graeffe’s method squaring thrice:
(i) x3 – 4x2 + 5x – 2 = 0 (ii) x3 – 2x2 + 5x + 6 = 0
(iii) x3 – x – 1 = 0.

You might also like