International Journal of Pure and Applied Mathematics

Volume 71 No. 2 2011, 251-259

A UNIVERSAL METHOD OF SOLVING QUARTIC EQUATIONS
Sergei L. Shmakov
Saratov State University
83, Astrakhanskaya Str.,
Saratov, 410012, RUSSIAN FEDERATION
Abstract: All the existing methods of solving quartic equations (Descartes-
Euler-Cardanos, Ferrari-Lagranges, Neumarks, Christianson-Browns, and Ya-
coub-Fraidenraich-Browns ones) are particular versions of some universal method.
It enables producing an arbitrary number of other particular algorithms with
some desired properties.
Dedicated to Vera P. Filinova
my mathematics teacher.
AMS Subject Classication: 11D25
Key Words: quartic equation, solution, method, resolvent, shift, hyperbolic
transformation
Introduction
Quartic equations (often referred to as just quartics) in one unknown
x
4
+ ax
3
+ bx
2
+ cx + d = 0 (1)
are the highest degree polynomials which can be solved analytically, by radicals,
with no iterative techniques. There are several algorithms for this purpose, each
involving a subsidiary cubic equation, the so-called resolvent. In his paper
[1] Herbison-Evans examines them in application to computer graphics tasks.
Let us list these ve algorithms with their resolvents.
Received: June 2, 2011 c 2011 Academic Publications, Ltd.
252 S.L. Shmakov
1. Descartes-Euler-Cardanos one [2]:
y
3
+ (2b 3a
2
/4)y
2
+ (3a
4
/16 a
2
b + ac + b
2
4d)y

a
3
8

ab
2
+ c

2
= 0; (2)
2. Ferrari-Lagranges [3]:
y
3
+ by
2
+ (ac 4d)y + (a
2
d + c
2
4bd) = 0; (3)
3. Neumarks [4]:
y
3
2by + (ac + b
2
4d)y + (a
2
d abc + c
2
) = 0; (4)
4. Christianson-Browns: in 1967 Brown [5] published an original method of
solving palindromic quartics, and in 1991 Christianson [6] extended it to
a general quartic equation:
y
3
+
4a
2
b 4b
2
4ac + 16d 3a
4
/4
a
3
4ab + 8c
y
2
+
+ (3a
2
/16 b/2)y + ab/16 a
3
/64 c/8) = 0; (5)
5. Yacoub-Fraidenraich-Browns: Yacoub and Fraidenraich [7] have shown a
dierent way of applying Browns technique to a general quartic:
(a
3
4ab + 8c)y
3
+ (a
2
b 4b
2
+ 2ac + 16d)y
2
+
+ (a
2
c 4bc + 8ad)y + (a
2
d c
2
) = 0. (6)
By Herbison-Evans analysis [1], none of these ve would completely suit
the needs of stable computations. He writes: It would be good to nd more
algorithms or variations on the ve listed here which allowed other combinations
of coecient signs to be handled in a stable fashion. It is far from clear why
there are ve and only ve algorithms so far discovered. Are there more? Are
there an innite number of algorithms?
Let us try to answer these questions.
The basic algorithm is, as we will see, Ferrari-Lagranges one (with a minor
change).
A UNIVERSAL METHOD OF SOLVING QUARTIC EQUATIONS 253
1. Basic Algorithm
Represent Eq. (1) as the product of two quadratic trinomials
(x
2
+ g
1
x + h
1
)(x
2
+ g
2
x + h
2
) =
= x
4
+ (g
1
+ g
2
)x
3
+ (g
1
g
2
+ h
1
+ h
2
)x
2
+ (g
1
h
2
+ g
2
h
1
)x + h
1
h
2
,
i.e.

g
1
+ g
2
= a,
g
1
g
2
+ h
1
+ h
2
= b,
g
1
h
2
+ g
2
h
1
= c,
h
1
h
2
= d.
(7)
These pair sums and products of g and h allow Vi`etas theorem [8] to be applied
to express these coecients as the roots of the subsidiary quadratic equations

g
2
ag + b y = 0,
h
2
yh + d = 0,
(8)
where y is derived from the only equation with the cross products g
1
h
2
and
g
2
h
1
. Having explicitly written the roots, substituted them to the equation for
c, and eliminated the radical, one obtains the cubic equation for y (Appendix,
lines 19):
y
3
by
2
+ (ac 4d)y a
2
d c
2
+ 4bd = 0. (9)
This is the cubic resolvent of Eq. (1). In our case, it is identical to that of
Ferrari-Lagranges method, except for the sign at the even powers of y.
The algorithm is as follows:
1. Calculate the coecients of Eq. (9) and nd its real root (any cubic has
at least one real root).
2. Calculate the coecients of Eqs (8) and solve them to get g
1
, g
2
, h
1
, and
h
2
. Eq. (1) is therefore factored into two quadratics.
3. Solve these quadratics.
This algorithm allows variations to be considered below.
254 S.L. Shmakov
2. A Shift of y
Introduce a new variable y
s
y (i.e. y = y
s
+ ), where is an arbitrary
shift. For dimension reasons, it must look as = a
2
+ b, where and
are dimensionless coecients; more sophisticated terms like a
k
b
l
c
m
d
n
, where
k + 2l + 3m + 4n = 2, should not be involved. The resolvent takes the form:
(y
s
+ )
3
b(y
s
+ )
2
+ (ac 4d)(y
s
+ ) a
2
d c
2
+ 4bd =
= y
3
s
+ Ay
2
s
+ By
s
+ C = 0, (10)
where

A = 3 b,
B = 3
2
2b + ac 4d,
C =
3
b
2
+ (ac 4d) a
2
d c
2
+ 4bd
(11)
(Appendix, lines 1011). On nding a real root y
s
, a backward shift is due with
return to the rest of the master algorithm.
In Neumarks method = b, and

A = 3b b = 2b,
B = 3b
2
2b
2
+ ac 4d = ac + b
2
4d,
C = b
3
b
3
+ (ac 4d)b a
2
d c
2
+ 4bd = abc + c
2
a
2
d
(12)
(Appendix, line 12), and in Ferrari-Lagranges one, = 0, but y
s
in both
cases is negated, so the resolvents coecients at the even powers of y
s
reverse
their signs, along with the sign at y
s
in Eqs (8).
3. A Shift of x
Introduce a new variable x
s
x (i.e. x = x
s
+), where is an arbitrary
shift. For dimension reasons, it must look as = a, where is a dimensionless
coecient; more sophisticated terms like a
k
b
l
c
m
d
n
, where k +2l +3m+4n = 1,
should not be involved. Derive the coecients of the initial quartic with respect
to x
s
:
(x
s
+ )
4
+ a(x
s
+ )
3
+ b(x
s
+ )
2
+ c(x
s
+ ) + d =
= x
4
s
+ (a + 4)x
3
s
+ (b + 3a + 6
2
)x
2
s
+ (c + 2b + 3a
2
+ 4
3
)x
s
+
+ d + c + b
2
+ a
3
+
4
= x
4
s
+ a
s
x
3
s
+ b
s
x
2
s
+ c
s
x
s
+ d
s
= 0,
A UNIVERSAL METHOD OF SOLVING QUARTIC EQUATIONS 255
where a
s
= a + 4, b
s
= b + 3a + 6
2
, c
s
= c + 2b + 3a
2
+ 4
3
, d
s
=
d + c + b
2
+ a
3
+
4
(Appendix, line 16). Substituting these expressions
to Eqs (11) for simultaneous application of both shifts, we have:

A = 3 b
s
= b 3a + 3 6
2
,
B = 3
2
2b
s
+ a
s
c
s
4d
s
=
= 3
2
2(b + 3a + 6
2
) + (a + 4)(c + 2b + 3a
2
+
+ 4
3
) 4(d + c + b
2
+ a
3
+
4
),
C =
3
b
s

2
+ (a
s
c
s
4d
s
) a
2
s
d
s
c
2
s
+ 4b
s
d
s
=
=
3
(b + 3a + 6
2
)
2
+ ((a + 4)(c + 2b + 3a
2
+
+ 4
3
) 4(d + c + b
2
+ a
3
+
4
)) (a + 4)
2
(d+
+ c + b
2
+ a
3
+
4
) (c + 2b + 3a
2
+ 4
3
)
2
+
+ 4(b + 3a + 6
2
)(d + c + b
2
+ a
3
+
4
).
(13)
Calculations show that when = a
2
s
+ b
s
, where 8 + 3 = 1, the
coecients of the resolvent do not change upon any shift of x. In Descartes-
Euler-Cardanos method, the shift is chosen just in this way: = b
s
a
2
s
/4
( = 1/4, = 1). Additionally, = a/4, so = b 3a
2
/8 and

A = 2b
3a
2
4
,
B =
3a
4
16
a
2
b + ac + b
2
4d,
C =

a
3
8

ab
2
+ c

2
(14)
(Appendix, lines 1820).
4. Hyperbolic Transformation of y
Lets assign the subscripts CB and YFB to the roots of Christianson-Browns
and Yacoub-Fraidenraich-Browns resolvents, respectively, that of the DEC re-
solvent remaining unsubscripted. Then
y =
a
3
+ 8c 4ab
16y
CB
=
a
3
+ 8c 4ab
16

y
Y FB
+
a
4

(note that the numerator (a

3
+ 8c 4ab) is invariant with respect to the shift
of x [9]). Substitute these expressions into Descartes-Euler-Cardanos resol-
vent (14), eliminate the denominator (with possible loss of roots), and come to
the known coecients (Appendix, lines 2125).
256 S.L. Shmakov
Instead of immediate recalculation of their y
CB
and y
Y FB
into y with re-
ducing the task to that already solved, the authors of the palindromic methods
propose their own sets of formulae, which hides the relation between several
forms of representation of the same, in essence, resolvent.
5. A New Particular Method
Having the described general scheme, one can easily produce an arbitrary num-
ber of particular algorithms. For example, let us require that the cubic resolvent
should contain no y
2
term (A = 0), then it follows from Eqs (11) that = b/3
(note that the condition 8+3 = 1 holds true and the coecients of the future
resolvent will be invariant to ) provided = 0, and

B = 3

b
3

2
2b
b
3
+ ac 4d = ac
b
2
3
4d,
C =

b
3

3
b

b
3

2
+ (ac 4d)
b
3
a
2
d c
2
+ 4bd =
=
abc
3
a
2
d
2
27
b
3
c
2
+
8
3
bd
(15)
(Appendix, line 13). The subsidiary quadratics will be

g
2
ag +
2
3
b y
s
= 0,
h
2

y
s
+
b
3

h + d = 0,
(16)
their roots are
g
1,2
=
a

a
2
8b/3 + 4y
s
2
, (17)
h
1,2
=
y
s
+ b/2

(y
s
+ b/3)
2
4d
2
, (18)
(19)
and, nally, the roots of the initial quartic are
x
1,2
=
g
1

g
2
1
4h
1
2
, (20)
x
3,4
=
g
2

g
2
2
4h
2
2
. (21)
A UNIVERSAL METHOD OF SOLVING QUARTIC EQUATIONS 257
A new algorithm for solving quartics is thus invented, it provides the cubic
resolvent just in its reduced form.
6. Conclusion
Of course, nobody could guarantee computational success to any of an arbitrary
number of particular algorithms. We may guess that the ve previously pub-
lished methods have been elaborated so as to attach some algebraic quality to
the resolvents coecients, but their robustness in actual computations should
be a subject of further analysis.
Excessive complications may be adverse. E.g., if in the palindromic algo-
rithms the expression (a
3
+ 8c 4ab) takes on zero or a near-zero value, this
will cause a fault or an error. The case of y
CB
or (y
Y FB
+ a/4) close to zero is
also unfavorable. Simple shifts of the roots seem more safe, provided that no
close values are subtracted.
A similar technique of shifts and hyperbolic transformations can be applied
to cubic equations as well.
Acknowledgments
The author is grateful to Professor Donald Herbison-Evans for his encourage-
ment and advice.
References
[1] D. Herbison-Evans, Solving quartics and cubics for graphics, 2005
http://linus.it.uts.edu.au/~don/pubs/solving.html
[2] T. Strong, Elementary and Higher Algebra, Pratt and Oakley (1859).
[3] H.W. Turnbull, Theory of Equations, fourth ed., Oliver and Boyd, London
(1947).
[4] S. Neumark, Solution of Cubic and Quartic Equations, Pergamon Press,
Oxford (1965).
[5] K.S. Brown, Reducing Quartics to Cubics, 1967,
http://mathpages.com/home/kmath296.htm
258 S.L. Shmakov
[6] B. Christianson, Solving Quartics using Palindromes, The Math. Gaz. 75
(1991), 327328.
[7] M.D. Yacoub, G. Fraidenraich. A New Simple Solution of the General
Quartic Equation, The Math. Gaz. 2011 (in press).
[8] F. Vi`ete. Opera mathematica (1579). Reprinted Leiden, Netherlands
(1646).
[9] Y. Mochimaru, Reciprocal Solution of a Quartic Equation, Int. J. Pure
Appl. Math. 14 (2004), 207210.
Appendix: MAXIMAs script
(%i1) g^2-a*g+b-y$ solve(%,g)$
(%i3) h^2-y*h+d$ solve(%,h)$
(%i5) rhs(%th(3)[1])*rhs(%th(1)[2])+rhs(%th(3)[2])*
rhs(%th(1)[1])-c,expand;
(%o5)
p
4 y 4 b + a
2
p
y
2
4 d
2
+
a y
2
c
(%i6) pickapart(%,1);
(%t6)
p
4 y 4 b + a
2
p
y
2
4 d
2
(%t7)
a y
2
(%o7) c + %t7 + %t6
(%i8) expandwrt(expand(%t6^2-(c-%t7)^2),y);
(%o8) y
3
b y
2
4 d y + a c y + 4 b d a
2
d c
2
(%i9) create_list(expand(coeff(%,y,n)),n,[3,2,1,0]);
(%o9) [1, b, a c 4 d, 4 b d a
2
d c
2
]
(%i10) subst([y=ys+D],%th(2)),expand;
(%o10) D
3
+3 ys D
2
b D
2
+3 ys
2
D2 b ys D4 d D+a c D+ys
3
b ys
2
4 d ys+ac ys+
4 b d a
2
d c
2
(%i11) [A,B,C]:create_list(coeff(%,ys,n),n,[2,1,0]);
(%o11) [3 Db, 3 D
2
2 b D 4 d + a c, D
3
b D
2
4 d D + a c D + 4 b d a
2
d c
2
]
(%i12) subst([D=b],%),eval,expand;
(%o12) [2 b, 4 d + a c + b
2
, a
2
d c
2
+ a b c]
(%i13) subst([D=b/3],%th(2)),eval,expand;
(%o13) [0, 4 d + a c
b
2
3
,
8 b d
3
a
2
d c
2
+
a b c
3

2 b
3
27
]
(%i14) x^4+a*x^3+b*x^2+c*x+d$ subst([x=xs+D],%),expand$
(%i16) [as,bs,cs,ds]:create_list(coeff(%,xs,n),n,[3,2,1,0]);
(%o16) [4 D + a, 6 D
2
+ 3 a D + b, 4 D
3
+ 3 a D
2
+ 2 b D + c, D
4
+ a D
3
+ b D
2
+ c D + d]
(%i17) subst([D=-a/4],%),expand;
(%o17) [0, b
3 a
2
8
, c
a b
2
+
a
3
8
, d
a c
4
+
a
2
b
16

3 a
4
256
]
A UNIVERSAL METHOD OF SOLVING QUARTIC EQUATIONS 259
(%i18) subst([D=%[2],a=%[1],b=%[2],c=%[3],d=%[4]],%o12),expand;
(%o18) [2 b
3 a
2
4
, 4 d + a c + b
2
a
2
b +
3 a
4
16
, c
2
+ a b c
a
3
c
4

a
2
b
2
4
+
a
4
b
8

a
6
64
]
(%i19) factor(%[3]);
(%o19)
`
8 c 4 a b + a
3

2
64
(%i20) y^3+%th(2)[1]*y^2+%th(2)[2]*y+%th(2)[3];
(%o20) y
3
+

2 b
3 a
2
4

y
2
+

4 d + a c + b
2
a
2
b +
3 a
4
16

y c
2
+a b c
a
3
c
4

a
2
b
2
4
+
a
4
b
8

a
6
64
(%i21) subst([y=(a^3+8*c-4*a*b)/(4*(a+4*yYFB))],%),ratsimp$
(%i22) -num(%)/(a^3+8*c-4*a*b),ratsimp;
(%o22)
`
8c 4ab + a
3

yY FB
3
+
`
16d + 2ac 4b
2
+ a
2
b

yY FB
2
+
+
`
8ad +
`
a
2
4b

yY FB + a
2
d c
2
(%i23) subst([y=(a^3+8*c-4*a*b)/(16*yCB)],%th(3)),ratsimp$
(%i24) expandwrt(ratsimp(num(%)/coeff(expand(num(%)),yCB,3)),
yCB)$
(%i25) create_list(coeff(%,yCB,n),n,[2,1,0]),ratsimp;
(%o25) [
64 d 16 a c 16 b
2
+ 16 a
2
b 3 a
4
32 c 16 a b + 4 a
3
,
8 b 3 a
2
16
,
8 c 4 a b + a
3
64
]
260