Analytical formula for the roots of the general complex

cubic polynomial
Ibrahim Baydoun

To cite this version:

Ibrahim Baydoun. Analytical formula for the roots of the general complex cubic polynomial. 2018.
�hal-01237234v2�

HAL Id: hal-01237234

[Link]
Preprint submitted on 17 Jan 2018

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 Analytical formula for the roots of the general complex cubic polynomial
Ibrahim Baydoun1
1
ESPCI ParisTech, PSL Research University, CNRS, Univ Paris Diderot, Sorbonne Paris Cité, Institut
Langevin, 1 rue Jussieu, F-75005, Paris, France

January 17, 2018

Abstract
We present a new method to calculate analytically the roots of the general complex polynomial of degree three. This
method is based on the approach of appropriated changes of variable involving an arbitrary parameter. The advantage
of this method is to calculate the roots of the cubic polynomial as uniform formula using the standard convention of the
square and cubic roots. In contrast, the reference methods for this problem, as Cardan-Tartaglia and Lagrange, give the
roots of the cubic polynomial as expressions with case distinctions which are incorrect using the standar convention.

keyword: cubic polynomial roots, appropriated change of variable, analytical uniform formula.

1 Introduction
The problem of solution formulas of polynomial equations is fundamental in algebra [1, 2]. It is useful, by example, for
the eigenvalues perturbation [3] and the solution of a system of nonlinear equations [4]. Moreover, it has many physical
applications, as the study of the singularities of the surfaces of refractive indices in crystal optics [5] and the diagonal
eigenvalues perturbation [6]. Another example is the calculation of the sound velocity anisotropy in cubic crystals [7],
or the diagonalization of Christoffel tensor to calculate the velocities of the three quasi-modes of the elastic waves in
anisotropic medium [8]. An explicit calculation for the particular case of 3 × 3 real symmetric matrices has been studied
in [9].
For a cubic equation, Cardano-Tartaglia and Lagrange’s formulas are subject to different interpretations depending on
the choices for the values of the square and cubic roots of the formulas [10, 11]. The correct interpretation is generally
derived from conditions between the roots and the coefficients of the cubic polynomial. It is a fundamental question to
be able to define uniform formula that yields correct solutions for all cubic equations. Ting Zhao’s thesis and related
work recently brought an important advance on Lagrange’s formulas with uniform solution formulas for cubic and quartic
equations with real coefficients. The results was obtained by introducing a new convention for the argument of the cubic
root of a complex number [12]. For cubic equations, we present in this paper a uniform formula that is valid with complex
coefficients. This more general result are uniquely defined by the standard convention of the square and cubic roots.
To illustrate the problem, we introduce Cardan-Tartaglia formula and afterward we give two examples. Indeed, Cardan-
Tartaglia calculate the 3 solutions of the equation

(C) : z 3 + pz + q = 0

as
v ! v !
 u r  u r
2π √ u1
3 ∆C 2π √ u1
3 ∆C
zn := exp n −1 t −q + + exp −n −1 t −q − ; n ∈ {0, 1, 2} ,
3 2 27 3 2 27

where ∆C is defined as


∆C := − 4p3 + 27q 2 .

We show by the following examples that the standard convention of the cubic roots
√  1
3
arg X = arg (X) ; X ∈ C
3
is not relevant in the above Cardan-Tartaglia’s expression. Indeed:
Example 1.1. The solution of (C) for the case p = 1 and q = 2 is
√
−1 − −3
zn ; n ∈ {0, 1, 2} ,
2
instead of zn .
Example 1.1 requires to shift the standard convention by −2π/3 in order to be suitable for the Cardan-Tartaglia’s solutions.
Another example requires another convention:

1
 √
Example 1.2. The solution of (C) for the case p = 1 and q = −1 is
√
−1 + −3
zn ; n ∈ {0, 1, 2} ,
2
instead of zn .
Therefore, a natural question is how we can calculate the roots of (C) using uniquely the standard convention?
To overcome this problem, a mathematical software as Matlab generates the symbolic solutions of (C) in terms of one
cubic root RC as follows
p
RC − ,
3RC
√  
−3 p p RC
+ RC + − ,
2 3RC 6RC 2
√  
−3 p p RC
− + RC + − ,
2 3RC 6RC 2
where RC is given by
sr
3 −∆C q
RC = − .
22 33 2
On the one hand, this formulation hid the problem of the convention of the cubic roots. But, on the other hand, it involves
another problem by dividing by RC which leads to a singularity in the formulation if RC is zero. So, the above question
can be reworded by the following manner: how we calculate the roots of (C) using uniquely the standard convention and
without any singularity?
The response to this question is the objective of this paper.
We present a uniform analytical formula for the roots of the general complex cubic polynomial. Our method is based
on the approach of changes of variable involving an arbitrary parameter, which rests on two principal ideas to diagonalize
the general 3 × 3 complex matrix:
Firstly, we introduce one change of variable involving an arbitrary parameter. So, the characteristic equation of this matrix
can be replaced by another non-polynomial equation in terms of the first new variable.
Secondly, we introduce another change of variable involving another arbitrary parameter. The second variable replaces
the first and the second arbitrary parameter can be chosen as required to make the form of the new equation as the sum
of a cube of a certain monomial and a term independent of the unknown second variable.
Therefore, we can solve easily this new polynomial equation and consequently deduce the solution of the original equation,
which is the eigenvalues of the general 3 × 3 complex matrix.
In the sequel, we apply this approach to calculate the roots of the general complex cubic polynomial. In fact, we
construct a particular matrix such that its characteristic polynomial will coincide with this cubic polynomial. Consequently,
the eigenvalues of this particular matrix become identical to the roots of this cubic polynomial.
The rest of the paper is structured as follows. In section 2, we derive the approach of appropriated changes of variable
involving an arbitrary parameter. Consequently in section 3, we apply it on the general complex cubic polynomial.

2 Approach of appropriated changes of variable involving an arbi-

trary parameter
Let M be the general 3 × 3 complex matrix:
 
m1 m2 m3
M := m4 m5 m6  .
m7 m8 m9

In this section, we aim to calculate the spectrum of M with this approach. We detail it by the following four steps. The
first step consists of performing a change of variable for the eigenvalues of M so that we get a non-polynomial equation.
In the second step, we will deal with this equation in order to simplify it by some algebraic manipulations. The third
step will introduce a second change of variable in order to replace the non-polynomial equation by another polynomial
equation. Finally, in the fourth step, we will solve the latter polynomial equation to deduce the spectrum of M.
Starting the first step by introducing the following definition which will be useful in the sequel:
Definition 2.1. We associate to M the following expressions in terms of its components:
m3 m8
c1 := m1 − m5 − , c2 := c21 + 4 (m2 m4 + m3 m7 ) ,
m2
3 1 3 3 1

e1 := (m1 − c1 ) − Tr (M) , e2 := (m1 − c1 )2 + (c2 − c1 )2 − (m1 − c1 ) Tr (M) + [Tr (M)]2 − Tr M2 ,
2 2 2 8 4
1   1  1 

e3 := (2m1 − c1 )3 + 3 (2m1 − c1 ) c2 − (2m1 − c1 )2 + c2 Tr (M) + (2m1 − c1 ) [Tr (M)]2 − Tr M2 − det (M) ,
8 4 4
1   1 1

e4 := 3 (2m1 − c1 )2 + c2 − (2m1 − c1 ) Tr (M) + [Tr (M)]2 − Tr M2 .
8 2 4
Tr (M) and det (M) stand respectively for the trace and the determinant of matrix M.
Now, we introduce the first change of variable involving an arbitrary parameter so that the eigenvalues of M can be
rewritten in terms of this new variable b̃ such that b̃ verifies a non-polynomial equation like the following proposition:

2
Proposition 2.2. An eigenvalue k of matrix M can be written in terms of two unknowns ã and b̃ as follows:

ã − b̃ 1h √ i
k= + 2m1 − c1 + ∆k , (1)
2 2
where ã can be chosen arbitrary and b̃ is governed by the following non-polynomial equation:
√ √ 1  2√ 
A + B b̃ + C b̃2 + D ∆k + E b̃ ∆k + b̃ ∆k − b̃3 = 0, (2)
2
where ∆k is defined by
 2
m3 m8
∆k := b̃2 + 2 (c1 − ã) b̃ + ã2 − 2c1 ã + c2 = ã − b̃ − m1 + m5 + + 4 (m2 m4 + m3 m7 ) . (3)
m2

The coefficients of Eq. (2) depend on parameter ã as follows:

ã3 3ã2 3ã

A= + e1 ã2 + e2 ã + e3 , B=− − 2e1 ã − e2 , C= + e1 ,
2 2 2
ã2  c1   c1 
D= + e1 + ã + e4 , E = −ã − e1 + , (4)
2 2 2
where the coefficients of Eq. (4) are given by definition 2.1.
Proof:
Firstly, we decompose the matrix M as the sum of two matrix M1 and M2 as follows:
   
ã 0 0 m1 − ã m2 m3
M = 0 b̃ −m0  +  m4 m5 − b̃ m6 + m0 , (5)
0 0 0 m7 m8 m9
| {z } | {z }
M1 M2

where m0 is chosen such that M2 has an eigenvector of the form [0, y0 , z0 ]. The advantage of this choice is to construct
an eigenvector of M in terms of these of M2 which will be elaborated in the sequel. Indeed, if we take m0 such that
 
m8 m3
m0 = m5 − b̃ + m3 − m9 − m6 ,
m2 m2

then M2 have the eigenvalue

m8
k0 = −m3 + m9 (6)
m2
relative to the eigenvector
 
m3
v0 := 0, − ,1 .
m2

Equation 6 gives one eigenvalue k0 of M2 , then the characteristic polynomial of M2

   
1 0 0
det X 0 1 0 − M2 
0 0 1

is divided in C [X] by the polynomial X − k0 . Consequently, we deduce that the two other eigenvalues of M2 verify a
second order polynomial and they are given by:

−ã − b̃ 1h √ i
k̃ = + 2m1 − c1 ∓ ∆k . (7)
2 2
Secondly, we search an eigenvector for M of the form

v0 + vv; v ∈ C,

where v̂ := [v1 , v2 , v3 ] is an unit eigenvector of M2 relative to k̃. Indeed, using

M 2 v 0 = k0 v 0 ,
M2 v = k̃v,

we obtain from Eq. (5) that the vectorial equation

M (v0 + vv) = k (v0 + vv) ; k ∈ C,

is equivalent to the following system:

  
 v1v ã + k̃ −k = 0,


 
(S) b̃ vv2 − m 3
− m0 (vv3 + 1) + v k̃v2 − m3
k − k vv2 − m3
= 0,
 m2 m2 0 m2


v k̃v3 + k0 − k (vv3 + 1) = 0.

3
The impact of the choice of m0 appears in the fact that the first equation of (S) is independent of v0 + vv and it can be
verified by taking k = ã + k̃. Thus, by taking v as
1 k − k0
v= ,
v3 k̃ − k

the third equation of (S) is verified. It remains to verify the second equation of (S). Indeed, k = ã + k̃ and Eq. (7) implies
Eq. (1) where k depends only on ã − b̃ since ∆k depends only on ã − b̃. Therefore, by inserting Eq. (1) in the characteristic
equation of M, we get
1

k3 − Tr (M) k2 + {[Tr (M)]2 − Tr M2 }k − det(M) = 0, (8)
2
so that we can choose arbitrarily among ã and b̃ one parameter as required, while the other parameter will be the unique
unknown of Eq. (8), as well as k is an eigenvalue of M. We obtain that if ã and b̃ satisfy Eq. (8), then the second equation
of (S) is verified. To end the proof, we develop Eq. (8) to deduce that Eq. (8) and Eq. (2) are equivalent. 

In the second step, we aim to simplify Eq. (2) of the unknown b̃, where ã can be chosen as required. So, we derive from
Eq. (2) two equations of b̃ given by the following two propositions:
Proposition 2.3. For fixed ã, b̃ satisfies the following equation:
√ √
d1 + d2 b̃ + d3 b̃2 + d4 ∆k − d3 b̃ ∆k = 0, (9)
where the coefficients of Eq. (9) are given by:
d1 := d3 ã2 + s1 ã + s2 , d2 := −2d3 ã − s1 , d3 := m2 m4 + m3 m7 , d4 := d3 ã + s3 .
Here s1 , s2 and s3 are given using definition 2.1 as follows:
c1 c2
s1 := + c1 e2 + c2 e1 − 2c1 e4 − e3 , s2 := c1 e3 + c2 e4 , s3 := e3 + c1 e4 . (10)
2
Proof: √
Eq. (9) is deduced by multiplying Eq. (2) by b̃ + ∆k − ã + c1 . 

Proposition 2.4. For fixed ã, b̃ satisfies the following equation:

√ √
f1 + f2 b̃ + f3 b̃2 + f4 ∆k + f3 b̃ ∆k = 0, (11)
where the coefficients of Eq. (11) are given by:
f1 := f3 ã2 + s4 ã + s5 , f2 := −2f3 ã − s4 ,
f3 := (m5 − m1 ) m3 m7 + (m9 − m1 ) m2 m4 − m2 m6 m7 − m3 m4 m8 , f4 := −f3 ã + s6 .
Here s4 , s5 and s6 are given using definition 2.1 as follows:
s4 := c2 d3 − 2c21 e4 − 2c1 e3 − 2d3 e2 , s5 := c2 e3 − 2d3 e3 + c1 c2 e4 , s6 := c1 e3 + c2 e4 − 2d3 e4 . (12)
Proof:
Similary
√ to the proof of proposition 2.3, on one hand we multiply Eq. (2) by 2d3 and on the other hand we multiply Eq. (9)
by ∆k . Consequently, we subtract the obtained equations to deduce Eq. (11). 

Now, we show that equations (9) and (11) imply the two equations (13) and (14) given by the following corollary:
Corollary 2.5. For fixed ã, b̃ satisfies the following equations:
√ √
r1 + r2 b̃ + r3 ∆k − 2b̃ ∆k = 0, (13)
√
r4 + r5 b̃ + r6 ∆k + 2b̃2 = 0, (14)
where the coefficients of Eq. (13) and Eq. (14) are given by:
s1 s4
r1 := −r2 ã + s9 , r2 := − + , r3 := 2ã + s10 ,
d3 f3
s3 s6
r4 := 2ã2 + s7 ã + s8 , r5 := −4ã − s7 , r6 := + (15)
d3 f3
Here, the coefficients of Eq. (15) are defined using definition 2.1, Eq. (10) and Eq. (12) as follows:
s1 s4 s2 s5 s2 s5 s3 s2 − 2d3 e4
s7 := + , s8 := + , s9 := − , s10 := − .
d3 d3 d3 f3 d3 f3 d3 f3
Proof:
We divide Eq. (9) and Eq. (11) respectively by d3 and f3 . Then, the subtraction and the addition of the obtained equations
imply Eq. (13) and Eq. (14). 

In the third step, we introduce the second change of variable

 √ 
l := − b̃ + o ∆k (16)

in order to deduce from the intractable equations of b̃ another polynomial quation of the new variable l, where o is an
arbitrary parameter can be chosen as required:

4
Proposition 2.6. Let o ∈ C. Then l verifies the following equation:

2l3 + Bl l2 + Cl l + Dl = 0, (17)

where the coefficients of Eq. (17) are given by:

Bl := −l1 o + 6ã − l2 , Cl := l3 o2 − 2 (l1 ã + l4 ) o + 6ã2 − 2l2 ã + l5 ,

3 2 2

3 2
Dl := −l6 o + (l3 ã − l7 ) o − l1 ã + 2l4 ã + l8 o + 2ã − l2 ã + l5 ã + l9 .

Here Bl , Cl and Dl are defined using definition 2.1, Eq. (10), Eq. (12) and Eq. (15) as follows:

s1 s3 r6 s1 − (s7 + 4c1 ) s3
l1 := 2 − s7 + r 6 , l2 := r6 − s7 − 2
, l3 := s8 − 2c2 + 2c1 r6 + ,
d3 d3 d3
s2 s7 s3 − r 6 s1 s1 (s8 − 2c2 ) − (4c1 + s7 ) s2
l4 := 2 − s8 − c1 r6 , l5 := s8 + , l6 := 2c1 s8 + c2 s7 + ,
d3 d3 d3
(s8 − 2c2 ) s3 − r6 s2 s7 s2 − s1 s8 s8 s3 − r 6 s2
l7 := 2c1 s8 + c2 (s7 + r6 ) + , l8 := c2 r6 + , l9 := .
d3 d3 d3
Proof: √
The first part of the proof consists in getting expressions for b̃2 and ∆k just in terms of l and o. Indeed, Eq. (16) implies:
 √ 
b̃ = − o ∆k + l . (18)

We take the square of Eq. (18) and we replace ∆k by its value given in Eq. (3). Then, we insert Eq. (18) in the obtained
equation to deduce an expression of b̃2 . This chain of operations is illustrated as follows:
√

√
b̃2 = o2 ∆k +l2 + 2ol ∆k ⇒ 1 − o2 b̃2 = 2o2 (c1 − ã) b̃
|{z} +o2 ã2 − 2c1 ã + c2 + l2 + 2ol ∆k
|{z}  √ 
b̃2 +2(c1 −ã)b̃+ã2 −2c1 ã+c2 − o ∆k +l
 √

2o l − (c1 − ã) o2
2 ∆k + ã2 − 2c1 ã + c2 o2 + l2 − 2 (c1 − ã) o2 l
⇒ b̃ = . (19)
1 − o2
2
So,
√ inserting Eq. (18) and Eq. (19) in Eq. (14) in order to substitute b̃ and b̃ and consequently to get the expression of
∆k as follows:
√
r4 + r5 b̃
|{z} +r6 ∆k + 2 b̃2
|{z} =0
 √  n √ o
− o ∆k +l (1−o2 )−1 2o[l−(c1 −ã)o2 ] ∆k +(ã2 −2c1 ã+c2 )o2 +l2 −2(c1 −ã)o2 l



1 − o2 (r4 − r5 l) + 2 ã2 − 2c1 ã + c2 o2 − 2 (c1 − ã) o2 l + l2
√
⇒ ∆k = . (20)
(1 − o2 ) (r5 o − r6 ) + 2 [2 (c1 − ã) o3 − 2ol]
√
use the previous expressions of b̃2 and ∆k in order to deduce Eq. (17). Indeed,
The second part of the proof consists to √
we take the square of the equation b̃ + o √∆k + l = 0 deduced from Eq. (16). Then, we insert Eq. (3) and Eq. (9) in the
obtained equation to substitute ∆k and b̃ ∆k . Afterwards we insert Eq. (18) to substitute b̃. This chain of operations is
illustrated as follows:
√ √
b̃2 + o2 ∆k +l2 + 2o b̃ ∆k +2lo ∆k + 2lb̃ = 0 ⇒
|{z} | {z }
 √ 
b̃2 +2(c1 −ã)b̃+ã2 −2c1 ã+c2 d−1 d1 +d2 b̃+d3 b̃2 +d4 ∆k
3
   
d2 d4 √
d1
(1 + o)2 b̃2 + 2 (c1 − ã) o2 + l + o b̃ +2 l + o ∆k + ã2 − 2c1 ã + c2 o2 + l2 + 2 o = 0 ⇒
d3 
|{z}
√  d 3 d 3
− o ∆k +l
   
d4 d2 √ d2
d1
(1 + o)2 b̃2 + 2 − (c1 − ã) o2 − o o ∆k − 2 (c1 − ã) o2 + l + o l + ã2 − 2c1 ã + c2 o2 + l2 + 2 o = 0.
d3 d3 d3 d3
2
Finally,
√ to complete the proof, it is enough to insert Eq. (19) and Eq. (20) in the above equation in order to substitute b̃
and ∆k and consequently to deduce Eq. (17). 

Since o is an arbitrary parameter, then we can choose o as required to solve equation (17) of the unknown l:
Lemma 2.7. If o verifies the following polynomial equation of degree two:


l12 − 6l3 o2 + 2 (l1 l2 + 6l4 ) o + l22 − 6l5 = 0, (21)

then l = −ã + rm , where

 √ √ 
l1 o + l2 1− −3 1+ −3
rm = + msl ; m∈ −1 , , .
6 2 2

Here sl is given by:

s 3
3 l1 o + l2 1
sl := − (l6 o3 + l7 o2 + l8 o − l9 ).
6 2

5
 Proof:
Equation (17) can be rewritten as follows
 3

Bl Cl Bl l12 − 6l3 o2 + 2 (l1 l2 + 6l4 ) o + l22 − 6l5
l3 + 3 l2 + 3 l + + s3l − ã = 0. (22)
6 6 6 12

The condition Bl2 = 6Cl , which is equivalent to Eq. (21), allows us to factorise Eq. (22) as follows:
 3   " 2   #
Bl 3 Bl Bl Bl 2
l+ + sl = l + + sl l+ − sl l + + sl = 0. (23)
6 6 6 6

So, Eq. (23) can be solved easily, wich ends the proofs. 

Lemma 2.7 gives the expressions for the set of solutions of l. Then, we deduce from the change of variable (18) the following
polynomial equation of b̃, which will be useful to give an explicit formula for the set of solutions of b̃:
Corollary 2.8. If o verifies Eq. (21), then b̃ verifies the following equation

 

1 − o2 b̃2 + 2 rm − ã + o2 (ã − c1 ) b̃ + (rm − ã)2 − o2 ã2 − 2c1 ã + c2 = 0, (24)

where
 √ √ 
l1 o + l2 l1 o + l2 1 − −3 l1 o + l2 1 + −3
rm ∈ Sl := − sl ; + sl ; + sl .
6 6 2 6 2

Proof:
We have from Eq. (16) that
√ 1 
∆k = − b̃ + l . (25)
o
The proof ends by taking the square of Eq. (25) and afterwards replacing ∆k by its value given by Eq. (3) in the obtained
equation, where l is replacing by rm − ã from lemma 2.7. 

Finally, in the fourth step, we can determine the set of solutions of b̃ by solving the nonlinear system constructed
by Eq. (24) and one of the following equations (9), (11), (13) or (14) of b̃. Thus, we can solve this nonlinear system by
choosing ã as required to simplify the calculation. Consequently, we deduce from Eq. (1) the spectrum of M:
Corollary 2.9. The spectrum of matrix M is given by:
  2

1 1 d3 r m + s3 (1 − o) rm − s2 o (1 − o) − d3 c2 o2 rm
k= 1+ + 2m 1 − c1 − , (26)
2 o (1 − o) (s3 + s1 o) − 2d3 c1 o2 + d3 (1 + o) rm o

where rm ∈ Sl and o verifies Eq. (21).

Proof:
Inserting Eq. (25) in Eq. (9), we get an equation of b̃ and b̃2 . Then, we substitute b̃2 using Eq. (24) to obtain:
2
d3 r m + (1 − o) s3 rm − s2 o (1 − o) − d3 c2 o2
b̃ = ã − . (27)
(1 − o) (s3 + s1 o) − 2d3 c1 o2 + d3 (1 + o) rm
√
Finally, to deduce Eq. (26), we insert Eq. (25) in Eq. (1) in order to substitute ∆k and afterwards we insert Eq. (27) in
the obtained equation in order to substitute b̃. 

The denumerator of the fraction in Eq. (26) contains rm which involves a cubic root. So, we calculate another expression
for spectrum of M in order to rationalize this denumerator. The two expressions of spectrum of M will be useful in the
proof of theorem 3.3.
Corollary 2.10. The spectrum of matrix M is given by:
  
1 1 s9 o2 − s8 o + (r6 − s10 o) rm rm
k= 1+ + 2m 1 − c1 − , (28)
2 o (s7 − s10 ) o + r2 o2 + r6 + 2orm o

where rm ∈ Sl and o verifies Eq. (21).

Proof: √
We insert Eq. (25) in both Eq. (9) and Eq. (11) to replace ∆k by its value in terms of b̃, l and o. Then, by substituting
2
b̃ from one obtained equation in the other, we deduce that:

s9 o2 − s8 o + (r6 − s10 o) rm
b̃ = ã − . (29)
(s7 − s10 ) o + r2 o2 + r6 + 2orm
√
Finally, to deduce Eq. (28), we insert Eq. (25) in Eq. (1) to substitute ∆k and afterwards we insert Eq. (29) in the
obtained equation to substitute b̃. 

6
3 Application on the cubic polynomial
Let (P) be the general complex polynomial of degree three in C [x]:

(P) : x3 + bx2 + cx + d = 0; b, c, d ∈ C.

We aim to calculate analytically the roots of (P) by applying the results of section 2. So, we introduce the following
definitions which will be useful to find the expressions for the roots of (P):
Definition 3.1. We associate to (P) the following expressions in terms of its coefficients:





∆l := 2c3 8b6 + 132b3 d + 36d2 + c3 + 33b2 c2 − 66bcd + 12b4 c d2 − 7c3 − b2 c2 d 24b3 + 291d + d3 144bc − 2b3 − 27d ,
∆o := −4b3 d + b2 c2 + 18bcd − 4c3 − 27d2 ,

and

do := 4b4 c2 − 4b3 cd − 14b2 c3 + b2 d2 + 28bc2 d + c4 − 12cd2 .

Definition 3.2. We associate to (P) the following expressions in terms of these given by definition 3.1:
√
√
−3
δl := (d − bc) ∆o 4b2 c2 − 4bcd + 2c3 + d2 + ∆l ,
9
and
√
2 −3
√
A1 := − 4b3 c − 2db2 − 13bc2 + 15dc + 2c ∆o ,
3
√
√
A2 := 8b5 c2 − 8b4 cd − 40b3 c3 + 2b3 d2 + 116b2 c2 d + 23bc4 − 99bcd2 − 21c3 d + 27d3 − −3 8b2 c2 − 10bcd + c3 + 3d2 ∆o .

Now, we present the principal result of the paper which is the formula of the roots of (P). This formula is in fact a
Lagrange-type [11], where the square and cubic are given by the standard convention, contrary to Cardan-Tartaglia and
Lagrange formulas which are incorrect with the standard convention. Indeed, the difference with the previous works is the
two terms α1 and α2 of Eq. (30) which were missed in Cardan-Tartaglia and Lagrange formulas:
Theorem 3.3. Let the two cubic roots R1 and R2 :
r√
3−3 √ 2b3 − 9cb + 27d
R1 := ∆o + ,
9 27
r√
3 −3 √ 2b3 − 9cb + 27d
R2 := ∆o − .
9 27
The set of solutions of (P) is given in terms of R1 and R2 as follows:
 √ √ 
b 1 − −3 1+ −3
xm = mα1 R1 + m2 α2 R2 − ; m ∈ −1 , , , (30)
3 2 2

where α1 and α2 are defined as follows:

√
3 n√ h  p  io
4
α1 := exp −1 arg A1 3 δl − arg (−do R1 ) ,
2
√    q  
3
4 √

α2 := exp −1 arg A2 3 δl2 − arg d2o R2 .
2

Here, ∆o , do , δl , A1 and A2 are given by definitions 3.1 and 3.2.

Proof:
Firstly, we construct the following matrix
 
−b 1 d−1 c
A :=  0 0 1 ,
−d 0 0

which has (P) as its characteristic polynomial. Consequently, the roots of (P) are identical to the spectrum of A, which
can be calculated from corollaries 2.9 and 2.10. But, these corollaries require firstly to make explicitly o and rm relative
to matrix A. Indeed, lemma 2.7 gives Eq. (21) of o and the expression of l in terms of the components of M. So, we
identify the components of M with these of A so that M = A. In other words, we replace in expressions of section 2 the
components of M by their corresponding values in terms of b, c and d. Thus, we deduce that
√ √
b2 c3 − c4 − 3cd2 + b2 d2 − 2b3 cd ± −3c(d − bc) ∆o
o= , (31)
do
√ √
since Eq. (21) is a quadratic polynomial equation for o. In the sequel, we take the positive sign before −3c(d − bc) ∆o
in Eq. (31). Also, the applying of lemma 2.7 on matrix A implies that
 √ √ 
√
3 c √ p 3 1 − −3 1 + −3
r m = lo − m 4 ∆o δl ; m ∈ −1 , , .
do 2 2

7
Here, lo is defined as follows:
 √ 
4 3 3 3 2 2 2 4 2 3 −3 2 2

√
lo := d−1
o 2b cd + b c − b d − 5b c d − 4bc + 6bcd + 5c d + c b c − 2bd − c ∆ o .
3
Secondly, since we have the expressions of o and rm relative to matrix A, then by applying corollaries 2.9 and 2.10 on
matrix A, we deduce that the roots of (P) are given by
" 
2 #
2 2
1 1 −crm + d (1 − o) rm + bdo (1 − o) + c b − 4c o rm
xm = 1+ − b − =
2 o (1 − o) [d + (bc − d)o] − 2bco2 − c (1 + o) rm o
" 


#
1 1 bd(bc − d) + cd b2 − 2c − d 2c2 − bd o−1 + do−1 do−1 − d + 2bc rm rm
1+ −b− . (32)
2 o 2b2 c2 − 4bcd − 2c3 + d2 − 2 (c3 − bcd + d2 ) o−1 + d2 o−2 − 2c(d − bc)o−1 rm o

Then, we aim to simplify the expression of xm in Eq. (32), indeed Eq. (32) itself implies that

2


−crm + d (1 − o) rm + bdo (1 − o) + c b2 − 4c o2
=
(1 − o) [d + (bc − d)o] − 2bco2 − c (1 + o) rm




bd(bc − d) + cd b2 − 2c − d 2c2 − bd o−1 + do−1 do−1 − d + 2bc rm
. (33)
2b2 c2 − 4bcd − 2c3 + d2 − 2 (c3 − bcd + d2 ) o−1 + d2 o−2 − 2c(d − bc)o−1 rm

So, the following property allows us to eliminate, in the denominator of the fractions of Eq. (33), the cubic root concealed
in rm :
A C A C MA + N C
= ⇒ = = ; ∀A, B, C, D, M, N ∈ C∗ . (34)
B D B D MB + N D
By applying property (34) on Eq. (33) for M = 2o−1 c(d − bc) and N = −c (1 + o), we just obtain a square root in the
denominator as follows:
p √ 
b m2 3 42 δl2 −3
√ p
xm = − + 4b3 c − 2db2 − 13bc2 + 15dc + c ∆o − m 3 4δl × (35)
3 12δl 3
√
√
8b5 c2 − 8b4 cd − 40b3 c3 + 2b3 d2 + 116b2 c2 d + 23bc4 − 99bcd2 − 21c3 d + 27d3 + −3 8b2 c2 − 10bcd + c3 + 3d2 ∆o
.
18δl
Therefore, we rationalize Eq. (35) using the following identities:

8b5 c2 − 8b4 cd − 40b3 c3 + 2b3 d2 + 116b2 c2 d + 23bc4 − 99bcd2 − 21c3 d + 27d3

√
√ 9A1 δl
+ −3 8b2 c2 − 10bcd + c3 + 3d2 ∆o = ,
4do
√
2 −3
√ 3A2 δl
4b3 c − 2db2 − 13bc2 + 15dc + 2c ∆o =
3 d2o
and consequently we deduce that
√ √ √ p
4 mdo A1 3 δl − m2 4A2 3 δl2
3 3
b
xm = − − . (36)
8 d2o 3

Thirdly, in order to simplify d2o in the denominator of Eq. (36), we prove that:
 p 3 p n√ h  p  io
A1 3 δl = (−4do R1 )3 ⇒A1 3 δl = exp −1 arg A1 3 δl − arg (−4do R1 ) (−4do R1 ) ,
 q 3  √ 3 q 
√
  q  √  √ (37)
42 d2o R2 ⇒A2 3 δl2 = exp 42 d2o R2 42 d2o R2 .
3 3 3
A2 3 δl2 = −1 arg A2 3 δl2 − arg

Then, by inserting Eq. (37) in Eq. (36), we deduce Eq. (30). 

Acknowledgments
I would like to thank, in particularly, Dalia Ibrahim for her support during the difficult moment to work this paper. Also,
I wish to thank the reviewers of the first submission for their useful comments.

References
[1] B. van der Waerden, A History of Algebra, Springer, 1985.
[2] S. M. Lane, G. Birkhoff, Algebra, American Mathematical Society, 1999.
[3] W. W. Hager, R. N. Pederson, Perturbation in eigenvalues, Linear Algebra and its Applications 42 (1982) 39 – 55.
doi:[Link]

8
 [4] I. C. F. Ipsen, B. Nadler, Refined perturbation bounds for eigenvalues of hermitian and non-hermitian matrices, SIAM
Journal on Matrix Analysis and Applications 31 (2009) 40 – 53.
[5] O. N. Kirillov, A. A. Mailybaev, A. P. Seyranian, Unfolding of eigenvalue surfaces near a diabolic point due to a
complex perturbation, Journal of Physics A: Mathematical and General 38 (24) (2005) 5531.
[6] J. Anderson, A secular equation for the eigenvalues of a diagonal matrix perturbation, Linear Algebra and its Appli-
cations 246 (1996) 49 – 70. doi:[Link]
[7] T. Tsang, H.-Y. Park, Sound velocity anisotropy in cubic crystals, Physics Letters A 99 (8) (1983) 377 – 380.
doi:[Link]
[8] I. Baydoun, É. Savin, R. Cottereau, D. Clouteau, J. Guilleminot, Kinetic modeling of multiple scattering of elastic
waves in heterogeneous anisotropic media, Wave Motion 51 (8) (2014) 1325–1348.
[9] T. Mensch, P. Rasolofosaon, Elastic-wave velocities in anisotropic media of arbitrary symmetry-generalization of
thomsen’s parameters, Geophysical Journal International 128 (1997) 43 – 64.
[10] N. Bourbaki, Éléments d’histoire des mathématiques, Springer, 2006.
[11] J. L. Lagrange, Réflexions sur la résolution algébrique des équations, In: Nouveaux Mémoires de l’Académie royale
des Sciences et Belles-Lettres de Berlin, Oeurves complètes, tome 3, 1770.
[12] T. Zhao, D. Wang, H. Hong, Solution formulas for cubic equations without or with constraints, Journal of Symbolic
Computation 46.