GLASNIK MATEMATIČKI

Vol. 53(73)(2018), 343 – 357

ON UNBOUNDED POLYNOMIAL DYNAMICAL SYSTEMS

Hamza Boujemaa and Said El Qotbi

Mohammed V University in Rabat, Morocco

Abstract. Suppose given a polynomial dynamical system of degree

m. It is known that if the algebra associated to the corresponding homoge-
neous dynamical system of degree m has an idempotent element then the
original dynamical system is unbounded. In this work, we give a sufficient
condition ensuring the unboundedness even when there is no idempotent.
Some applications are also given.

1. Introduction
Since the seminal work of Markus [6], we can associate to any quadratic
homogeneous system of differential equations in Rn a commutative but not
necessarily associative multiplication in Rn . If the system writes
(1.1) ẋ = P (x),
where x = (x1 , .., xn ) ∈ Rn and P : Rn → Rn is a function whose components
are homogeneous quadratic polynomials in the variables (x1 , .., xn ), then the
multiplication is a billinear symmetric map β : Rn × Rn → Rn defined by
1
β(x, y) = (P (x + y) − P (x) − P (y)) .
2
In their paper on nonassociatice algebras and homogeneous quadratic
differential equations (see [5]), Kinyon and Sagle gave substantial results and,
more recently, this approach has been used for studying stability questions
(see [1]).

2010 Mathematics Subject Classification. 34A34, 17A99.

Key words and phrases. Polynomial dynamical systems, homogenization,
unboundedness.

343
344 H. BOUJEMAA AND S. EL QOTBI

This process can be generalized to homogeneous polynomial differential

equations of degree m ≥ 2. In this case, an m-linear symmetric map

f : Rn × · · · × Rn → Rn

can be defined and a m-algebra can be associated (see the work of Walcher
[7]). We write in this case

ẋ = f (x, .., x) = xm .

An element e ∈ Rn different from the origin is said to be an idempotent

if it verifies em = f (e, .., e) = e and an element g ∈ Rn , also different from
the origin, is called a nilpotent if we have g m = 0.
Among many interesting questions in the theory of differential equations,
one can study the bounded corresponding vector fields (see [2]). A vector
field is bounded if all trajectories are bounded for t ≥ 0. For homogeneous
polynomial differential equations, the existence of an idempotent ensures the
unboundedness since the trajectory via this idempotent is a ray that blows
up in finite time. Unfortunately, the existence of idempotents is not always
given. In order to study the boundedness of polynomial differential equations
of degree m, we try to gain informations from the study of the associated
homogeneous polynomial differential equations of degree m obtained from
the initial one by dropping all terms of degree less than m. An easy to check
sufficient condition was given by Dickson and Perko in [4]. Essentially, it says
that if the corresponding nonassociative algebra (or m- algebra when m ≥ 3)
has an idempotent then the polynomial vector field is unbounded.
From above, one can try to understand more when there is no idempotent.
This fact has been emphasized by Coleman ([3]) who gave some results on
unboundedness when the origin is an isolated critical point.
The aim of this work is to give a sufficient condition on the associated
homogeneous polynomial differential system ensuring the unboundedness of
the original polynomial dynamical system. This condition does not require
the existence of idempotents. For the two dimensional case, in the second
paragraph, we prove that if L is a half line from the origin and if there exist,
for the associated homogeneous polynomial dynamical system, a trajectory
that conveniently crosses L twice, then the polynomial dynamical system
is unbounded. Using a geometrical approach based on Thales theorem, an
elementary proof is given. Moreover, it is also shown that the blow up time
is finite. The case of dimension 3 is studied in the third paragraph where
we use an additional condition to ensure the iteration of some process and
the same conclusions are obtained. Moreover, for dimension larger than 3,
we emphasize similar conditions for a possible generalization. For dimension
two, examples of odd degree are studied and for dimension three, two classes
of examples are given. The first is of degree 2 and the second of degree 3.
 ON UNBOUNDED POLYNOMIAL DYNAMICAL SYSTEMS 345

2. The two dimensional case

It is well known that, according to the homogenization process, if X
is a polynomial vector field on Rn of degree m then we can associate to
X a homogeneous polynomial vector field X̂ of degree m defined on Rn+1 .
More precisely, if (x1 , .., xn ) are coordinates on Rn and (x1 , .., xn , z) are
coordinates on Rn+1 and if X writes
 X
 i1 i1

 ẋ1 = a1i1 ...i1 x11 . . . xnn

 1 n

 1≤i11 +···+i1n ≤m


.. ,
 .



 X in in

 ẋn = ani1 ...i1 x11 . . . xnn

 1 n
n n
1≤i1 +···+in ≤m

then X̂ writes
 X
 i1 i1 1 1

 ẋ1 = a1i1 ...i1 x11 . . . xnn z m−(i1 +···+in )

 1 n

 1≤i11 +···+i1n ≤m



 ..
. .

 X

 ẋn = n in
ai1 ...i1 x1 . . . xn z
1 in
n m−(i n
+···+in
)


1 n

 1 n

 1≤in +···+in ≤m
1 n

ż = 0
We may notice that if we restrict X̂ to the hyperplane Rn ×{z = 1}, we obtain
X and if we restrict X̂ to Rn × {z = 0} we obtain the homogeneous part X e
of degree m of X.
We state the main result.
Theorem 2.1. Let X be a polynomial vector field of degree m on R2 and
e
X its homogeneous part of degree m. Let φ̃(·, x) denotes the solution of X e
2
through x = (x1 , x2 ) ∈ R . If there exist a half line L from the origin O,
a point A ∈ L and two scalars τ > 0, λ > 1 verifying B = φ̃(τ, A) with
OB = λ OA, then X is unbounded and there is a trajectory blowing up in
finite time.
Proof. If the trajectory through A crosses the half line L tangentially
then, since Xe is homogeneous, L is invariant and we have an idempotent.
In this case, the result is given by Dickson and Perko in [4]. Thus, we may
suppose that the trajectory meets L transversely. We suppose also that τ is
the smallest t > 0 verifying φ(t, A) ∈ L.
Let X̂ be the homogeneous polynomial vector field on R3 obtained from
X using the homogenization process. If we restrict X̂ to the hyperplane z = 0,
346 H. BOUJEMAA AND S. EL QOTBI

we obtain Xe the homogeneous part of X and if we restrict it to the hyperplane

z = 1, in R3 , we obtain X.
For x = (x1 , x2 ) ∈ R2 and x̂ = (x1 , x2 , z) ∈ R3 , if we let φ̃(·, x) be the
solution for X̃ through x ∈ R2 , φ̂(·, x̂) be the solution for X̂ through x̂ ∈ R3
and φ(·, x) be the solution for X through x ∈ R2 , we have

(2.1) φ̂ (t, (x1 , x2 , 1)) = (φ (t, (x1 , x2 )) , 1)

and
 
φ̂ (t, (x1 , x2 , 0)) = φ̃ (t, (x1 , x2 )) , 0 ,

when both sides are defined.

We consider the four points: A = (a1 , a2 ), B = (b1 , b2 ), Â = (a1 , a2 , 0)
and B̂ = (b1 , b2 , 0). According to the hypothesis, since φ̃(τ, A) = B, we have
φ̂(τ, Â) = B̂.
Using the continuity of the flow φ̂, there exist a neighborhood V (Â) ⊂ R3
and a neighborhood V (B̂) ⊂ R3 verifying
 
φ̂ τ, V (Â) ⊂ V (B̂).

In addition, we may suppose V (Â) ∩ V (B̂) = ∅ and we let γ > 0 be the

distance d(V (Â), V (B̂)).
Let Π be the two dimensional subspace of R3 containing the origin O, Â
and the vector e3 = (0, 0, 1) ∈ R3 . As the vector field X̂ is transverse to Π
at B̂, it is possible to restrict V (B̂) to a new neighborhood of B̂, still denoted
by V (B̂), verifying: for any y ∈ V (B̂), the trajectory through y also meets
transversely Π at a point belonging to Π ∩ V (B̂). For the new neighborhood
V (B̂), the corresponding neighborhood of  will still be denoted by V (Â).
Let B(Â, δ) ⊂ V (Â) be an open ball centred in  and of radius δ > 0
small enough. First, we notice that if y1 ∈ B(Â, δ) and τy1 is the smallest
t > 0 verifying
φ̂ (t, y1 ) ∈ V (B̂) ∩ Π,
then there exist two scalars τ1 , τ2 with 0 < τ1 < τ2 for which

τ1 ≤ τy1 ≤ τ2 for any y1 ∈ B(Â, δ).

This follows from the compactness of the closed ball B(Â, δ) and the continuity
of the map ”y1 7→ τy1 ”, following itself from the differentiability of some
convenient Poincaré map.
Let M1 = (a1 , a2 , β1 ) ∈ R3 with β1 > 0 small enough so that M1 belongs
to B(Â, δ). The line (OM1 ) crosses the hyperplane z = 1 at a point Q0 . On
the other hand, the point N1 = φ̂ (τM1 , M1 ) belongs to V (B̂) ∩ Π and also to
 ON UNBOUNDED POLYNOMIAL DYNAMICAL SYSTEMS 347

the hyperplane z = β1 . Consequently, the line (ON1 ) crosses the hyperplane

z = 1 at a point Q1 , belonging to Π and verifying
1
OQ1 = ON1 .
β1
See Figure 1.

Q0 Q1
(0, 0, 1) • •

M1
(0, 0, β1 ) γ0 γ1
• •
N1 V (B̂)
γ0
(0, 0, β2 ) •
M2
• •
O Â B̂

B(Â, δ)

Figure 1.

For homogeneous quadratic dynamical systems in Rn+1 , it is known (see

[5]) that the solution Φ verifies
Φ(t, ax) = aΦ(at, x),
where a is a scalar, x ∈ Rn+1 and when both sides are defined. This implies
that the trajectory through ax and the trajectory through x are homothetic.
This fact can be generalized to homogeneous polynomial dynamical systems
in Rn+1 of degree m ≥ 2 and we have
Φ(t, ax) = aΦ(am−1 t, x),
when both sides are defined. As Q0 is the homothetic of M1 , Q1 is the homo-
thetic of N1 with respect to the same map and N1 belongs to the trajectory
of M1 , Q1 also belongs to the trajectory of Q0 . As before, we can iterate the
process. The line (OQ1 ) crosses the line (ÂM1 ) at a point M2 also element
of B(Â, δ) ∩ Π. If we let M2 = (a1 , a2 , β2 ) with 0 < β2 < β1 , then, using
Thales theorem, we have
β2 γ0
= < 1,
β1 γ0 + γ1
348 H. BOUJEMAA AND S. EL QOTBI

where γ1 = M1 N1 , γ0 being the distance from M1 to the z−axis and also the
distance from M2 to the z−axis. We can notice that
β2 γ0
< .
β1 γ0 + γ
If we let k = γ0γ+γ
0
, we can define by induction one sequence of positive scalars
(βn )n and three sequences of points (Mn )n , (Nn )n and (Qn )n such that
1 1
βn ≤ β1 k n−1 , Qn = Nn , Qn = Mn+1
βn βn+1
and
 
1 m−1 1

Qn = φ̂ (τMn , Mn ) = φ̂ βn τMn , Mn = φ̂ βnm−1 τMn , Qn−1 .
βn βn
Thus, we obtain


(2.2) Qn = φ̂ βnm−1 τMn + · · · + β1m−1 τM1 , Q0
and certainly Q0 Qn ≥ n γ, for n ≥ 1. This shows that the trajectory for X̂
through Q0 = (q1 , q2 , 1) blows up and implies, using (2.1), that the trajectory
for X through (q1 , q2 ) also blows up. Therefore, X is unbounded. To justify
that the blow up time is finite, we let ξn be the scalar given by
ξn = βnm−1 τMn + · · · + β1m−1 τM1 .
Clearly, we have


ξn ≤ τ2 βnm−1 + · · · + β1m−1 ,
ξn ≤ τ2 (β1 + · · · + βn ) ,
1
ξn ≤ τ2 β1 ,
1−k
which ends the proof.

Example 2.2. We can notice that when dimension is two and if we expect
a half line L as in the theorem, the corresponding degree has to be odd.
Consider the following homogeneous polynomial dynamical system in R2
(
ẋ = (−y + x)(x2 + y 2 )k
(2.3) ,
ẏ = (x + y)(x2 + y 2 )k
where k is a positive integer.
The origin is obviously an isolated critical point and we can easily prove
that there is no idempotent. In polar coordinates, system (2.3) writes
(
ṙ = r2 k+1
.
θ̇ = r2 k
 ON UNBOUNDED POLYNOMIAL DYNAMICAL SYSTEMS 349

Consequently, r increases without bound and θ also increases. Thus trajecto-

ries spiral around the origin and get far away from it. For this example, any
half line L from the origin will work.
Using our theorem, any polynomial vector field under the form
(
ẋ = (−y + x)(x2 + y 2 )k + Pk (x, y)
,
ẏ = (x + y)(x2 + y 2 )k + Qk (x, y)
where Pk and Qk are any polynomials of degree less than 2 k+1, is unbounded.

3. Higher dimensions
In this paragraph, we will use the same notations as before: X is a poly-
nomial vector field of degree m ≥ 2 on Rn for n ≥ 3, X̃ its homogeneous part
of degree m and X̂ the homogeneous vector field on Rn+1 obtained from X
using the homogenization process. φ̃(., x) denotes the trajectory through
a point x = (x1 , .., xn ) ∈ Rn for X̃ and φ̂(., x̂) the trajectory through
x̂ = (x1 , .., xn , z) for X̂. We suppose that there exist two scalars τ > 0,
λ > 1 and a point A ∈ Rn such that, if B = φ̃(τ, A) then OB = λ OA. If we
consider the linear map Dφ̃τ (A), we have the following result.
Proposition 3.1. λ and λm are two eigenvalues of Dφ̃τ (A) and two
λ
corresponding eigenvectors are u1 = A+(m−1) τ 1−λ m X̃(A) and u2 = X̃(A).

Proof. Using the equality φ̃(τ, φ̃t (A)) = φ̃(t, φ̃τ (A)) for small values of
t ∈ R, if we differentiate both sides with respect to t and let t = 0, we obtain
 
Dφ̃τ (A)(X̃(A)) = X̃ φ̃τ (A) = X̃(λ A) = λm X̃(A).
For the other eigenvalue, we notice that


φ̃(τ, (1 + t) A) = (1 + t) φ̃ (1 + t)m−1 τ, A
for small values of t ∈ R.
Again, if we differentiate with respect to t and let t = 0, we obtain
Dφ̃τ (A)(A) = λ A + (m − 1) λm τ X̃(A),
and if we look for a nonzero scalar β for which
Dφ̃τ (A + β X̃(A)) = λ (A + β X̃(A)),
λ
we obtain β = (m − 1) τ 1−λm , which ends the proof.

Remark 3.2. If u = (u1 , u2 , u3 ) is an eigenvector for Dφ̃τ (A) then û =

(u1 , u2 , u3 , 0) is an eigenvector for Dφ̂τ (Â) for the same eigenvalue as u (Â
is obtained from A by adding 0 as a fourth component).
From now on, we suppose n = 3 and we state the following result.
350 H. BOUJEMAA AND S. EL QOTBI

Theorem 3.3. Let X be a polynomial vector field on R3 of degree m, X̃

its homogeneous part of degree m. We suppose that there exist A ∈ R3 and
two scalars τ > 0, λ > 1 verifying φ̃(τ, A) = B, with OB = λ OA and we
suppose also that the linear map Dφ̃τ (A) has a third eigenvalue µ with |µ| < λ
then X is unbounded and there is a trajectory blowing up in finite time.
Proof. The main question is to make sure that an iteration process
similar to the two dimensional case will be possible.
Let X̂ be the homogeneous polynomial vector field on R4 obtained from
X, Â ∈ R4 obtained from A ∈ R3 by adding 0 as a fourth component. Since
coordinates will play an important role, we have to specify two convenient
bases.
Starting from the canonical basis of R4 , we let
OÂ X̂(Â) û3
e1 = , e2 = , e3 = , e4 = (0, 0, 0, 1),
kOÂk kX̂(Â)k kû3 k
where û3 is obtained from an eigenvector u3 corresponding to the eigenvalue
µ, by adding 0 as a fourth component, and B1 is the basis given by B1 =
{e1 , e2 , e3 , e4 }.
For any vector w = (w1 , w2 , w3 , w4 )B1 ∈ R4 , we will adopt the norm

kwkB1 = Supi=1, .., 4 |wi |

and the coordinates of a point M will be denoted in B1 by (xM1 , xM2 , xM3 , zM ).
We also introduce the basis B2 = {E1 , E2 , e3 , e4 } where E1 = kûû11 k and
E2 = kûû22 k , û1 and û2 are related respectively to u1 and u2 given in the
previous proposition.
Similarly, we will denote the components of a point M in B2 by
(XM1 , XM2 , xM3 , zM ). In addition, we can notice that there exist a posi-
tive constant c such that for any point M ∈ R4 , we have XM1 = c xM1 . This
fact will be used later.
Now, using properties of the flow, there exist a neighborhood V (Â) ⊂ R4
such that
φ̂τ : V (Â) → φ̂τ (V (Â))
is a diffeomorphism.
If we let H ⊂ R4 be the hyperplane containing the line (OÂ), parallel
to e3 and e4 , it is possible, using the transversality of X̂ at B̂ = φ̂τ (Â), to
restrict V (Â) to a new neighborhood still denoted by V (Â) and for which:
for any M ∈ V (Â), the trajectory through N = φ̂τ (M ) meets H at a point
belonging to H ∩ φ̂τ (V (Â)).
We also suppose that V (Â) is the ball B(Â, δ) of radius δ > 0, small
enough, the distance here is associated to the norm k.kB1 .
 ON UNBOUNDED POLYNOMIAL DYNAMICAL SYSTEMS 351

Without loss of generality, we may suppose B(Â, δ) ∩ φ̂τ (B(Â, δ)) = ∅

and we let  
γ = distance B(Â, δ), φ̂τ (B(Â, δ)) > 0.
Let M0 ∈ B(Â, δ) and N0 be the point of H ∩ φ̂τ (B(Â, δ)) at which the
trajectory via φ̂τ (M0 ) crosses H.
We claim that if δ > 0 is small enough, the line (ON0 ) crosses B(Â, δ)
and the iteration is possible.
Clearly, in the basis B2 , the matrix of Dφ̂τ (Â) is
 
λ 0 0 0
 0 λm 0 0
 
 0 0 µ 0
0 0 0 1
and there exist a point P0 ∈ B(Â, δ) verifying φ̂τ (P0 ) = N0 .
Let N0 = (XN0 1 , XN0 2 , xN0 3 , zN0 ), Â = (XÂ1 , XÂ2 , 0, 0), B̂ =
(λ XÂ1 , λ XÂ2 , 0, 0) and P0 = (XP0 1 , XP0 2 , xP0 3 , zP0 ), since
φ̂τ (P0 ) − φ̂τ (Â) = Dφ̂τ (Â)(P0 − Â) + kP0 − ÂkB2 ǫ(P0 − Â),
where ǫ = (ǫ1 , ǫ2 , ǫ3 , ǫ4 ) is a function defined on a neighborhood of 0 ∈ R4
with values in R4 verifying ǫ(P0 − Â) → (0, 0, 0, 0) when kP0 − ÂkB2 → 0,
we can deduce the following system


 XN0 1 − λ XÂ1 = λ (XP0 1 − XÂ1 ) + kP0 − ÂkB2 ǫ1 (P0 − Â)

X m
N0 2 − λ XÂ2 = λ (XP0 2 − XÂ2 ) + kP0 − ÂkB2 ǫ2 (P0 − Â)
.

 xN0 3 = µxP0 3 + kP0 − ÂkB2 ǫ3 (P0 − Â)


zN0 = zP0 + kP0 − ÂkB2 ǫ4 (P0 − Â)
Because of the homogenization, we have ǫ4 ≡ 0. Using a previous remark, if
we introduce coordinates in B1 , we deduce the system


 xN0 1 = λ xP0 1 + kP0 − ÂkB1 ǫ′1 (P0 − Â)

x
N0 2 = 0
(3.1) ,


 xN0 3 = µxP0 3 + kP0 − ÂkB1 ǫ′3 (P0 − Â)

zN0 = zP0
where ǫ′1 , ǫ′3 are also functions defined on a neighborhood of 0 ∈ R4 tending
to 0 ∈ R when kP0 − ÂkB1 → 0. The fact that we have xN0 2 = 0 results from
the supposition N0 ∈ H.
In order to prove that the line (ON0 ) crosses B(Â, δ), we need to find
λ′ > 1 for which the following inequalities hold
xN 1 xN 3 zP
(3.2) | ′0 − xÂ1 | < δ, | ′0 | < δ, | ′0 | < δ.
λ λ λ
The last inequality obviously holds since P0 ∈ B(Â, δ).
352 H. BOUJEMAA AND S. EL QOTBI

For the first inequality, using (3.1), it is equivalent to

λ kP0 − ÂkB1 ′
−δ < xP 1 − xÂ1 + ǫ1 (P0 − Â) < δ
λ′ 0 λ′
which leads to
λ xP0 1 + kP0 − ÂkB1 ǫ′1 (P0 − Â) λ xP0 1 + kP0 − ÂkB1 ǫ′1 (P0 − Â)
< λ′ < .
xÂ1 + δ xÂ1 − δ
On the other hand, the second inequality in (3.2) is equivalent to
µ kP0 − ÂkB1 ′
−δ < ′
xP0 3 + ǫ3 (P0 − Â) < δ.
λ λ′
As |xP0 3 | < δ and kP0 − ÂkB1 < δ, it is enough to ask for
|µ| |ǫ′3 (P0 − Â)|
+ <1
λ′ λ′
which means
λ′ > |µ| + |ǫ′3 (P0 − Â)|.
Consequently, a sufficient condition for the existence of λ′ is
λ xP0 1 + kP0 − ÂkB1 ǫ′1 (P0 − Â)
(3.3) > |µ| + |ǫ′3 (P0 − Â)|.
xÂ1 − δ
Now as x01 > xÂ1 − δ and kP0 − ÂkB1 < δ, it is enough to verify that
δ|ǫ′1 (P0 − Â)|
(3.4) λ− > |µ| + |ǫ′3 (P0 − Â)|.
xÂ1 − δ
If δ > 0 is small enough, then the functions ǫ′1 and ǫ′3 will be close enough to
0 and as |µ| < λ, the condition (3.4) will hold. Consequently, it is possible to
find λ′ > 1 for which inequalities (3.2) are true. Thus, the claim is proved.
Therefore, we can iterate. Given λ′ > 1 verifying (3.2), we let M1 = Nλ′0 , then
the trajectory via φ̂τ (M1 ) meets H ∩ φ̂τ (B(Â, δ)) at a point N1 and the line
(ON1 ) crosses B(Â, δ) too. As for the two dimensional case, it is possible to
define three sequences of points (Mn )n≥0 , (Nn )n≥0 , (Qn )n≥0 . First, if we let
Q0 be the intersection point of the line (OM0 ) with the hyperplane ”z = 1”
and Q1 the intersection point of the line (OM1 ) with the same hyperplane,
Q0 and Q1 are elements of the same trajectory and we have
Q0 Q1 1
=
M 0 N0 zM0
and similarly
Q1 Q2 1
= .
M 1 N1 zM1
 ON UNBOUNDED POLYNOMIAL DYNAMICAL SYSTEMS 353

If we consider the first components, we have

1
xQ1 1 − xQ0 1 = (xN0 1 − xM0 1 ) ,
zM0
and more generally, for n ≥ 1
1

xQn 1 − xQn−1 1 = xNn−1 1 − xMn−1 1 .
zMn−1
Since, for n ≥ 0, Mn and Nn belong to two disjoint balls, we can deduce that
xNn 1 − xMn 1 ≥ γ
and as
1
xQn 1 ≥ xQn−1 1 + γ ,
zMn−1
we have  
1 1
xQn 1 ≥ xQ0 1 + γ + ··· +
zMn−1 zM0
and therefore
xQn 1 ≥ xQ0 1 + n γ.
Consequently, when n tends to +∞, the trajectory blows up. To justify that
the blow up time is finite, we recall that we have N0 = φ̂τM0 (M0 ) for some
convenient scalar τM0 and more generally
Nn = φ̂τMn (Mn ), n ≥ 0.
In addition, there exist two scalars τ2 > τ1 > 0 verifying
τ1 ≤ τMn ≤ τ2
for any n ≥ 0.
As Q0 = zM1 M0 , we have
0

1 1 m−1
Q1 = N0 = φ̂(τM0 , M0 ) = φ̂(zM τM0 , Q0 ).
zM0 zM0 0

In the same way, we deduce

m−1 m−1
Qn = φ̂(zMn−1
τMn−1 + · · · + zM0
τM0 , Q0 ).
On the other hand, we have
zM0 zN0 x −δ+γ
= ≥ Â1
zM1 zM1 xÂ1 + δ
and more generally, for n ≥ 0
zMn−1 zNn−1 x −δ+γ
= ≥ Â1 .
zMn zMn xÂ1 + δ
354 H. BOUJEMAA AND S. EL QOTBI

If we define
m−1 m−1 xÂ1 + δ
ξn = zM τMn−1 + · · · + zM τM0 and k = ,
n−1 0
xÂ1 − δ + γ
then
 
m−1 m−1
0 < ξn ≤ τ2 zM n−1
+ · · · + z M0 ,
m−1


0 < ξn ≤ τ2 zM0
1 + · · · + k n−1 ,
m−1
τ2 zM
0 < ξn ≤ 0
,
1−k
and the proof is complete.
Remark 3.4. When the dimension is larger than 3, as λ and λm are
eigenvalues, a sufficient condition will be that the dimension of corresponding
eigenspaces is 1 and that all other eigenvalues µ1 , .., µp have modulus less than
λ. More explicitly, in the case where all eigenvalues are real and if Dφτ (A) is
diagonalizable, there is no difficulty in proving the same result since we have
λ > Supi=1,...,p |µi |.
If Dφτ (A) is not diagonalizable, then we use a Jordan canonical form for
which all possible 1′ s above the main diagonal are replaced by some η > 0
small enough to ensure
λ > Supi=1,...,p |µi | + η.
When complex eigenvalues exist, similar ideas, with an appropriate choice of
the norm to make the proof of the iteration easier, lead to the same result.
In the sequel, we will give two examples on R3 , the first one is of degree
2 and the second is of degree 3.
Example 3.5. Consider the following homogeneous polynomial dynami-
cal system X̃ of degree 2

2
ẋ = a x y + b y − c y z

(3.5) ẏ = −a x2 − b x y + c x z + d y z ,


ż = d y 2
where a, b, c and d are real constants with d > 0. Idempotents, if any, belong
to the cone ”x2 + y 2 = z 2 ”. If idempotents exist, unboundedness will result
from Dickson and Perko theorem. Otherwise, we will deduce it from Theorem
3.3. Since
d 2
(x + y 2 − z 2 ) = 0,
dt
the cone ”x2 + y 2 = z 2 ” is invariant. As ż = d y 2 , if (x(t), y(t), z(t)) belongs
to the cone, z(t) increases without bound and, if we suppose that there is no
 ON UNBOUNDED POLYNOMIAL DYNAMICAL SYSTEMS 355

idempotent, trajectories included in the cone will spiral and blow up. Thus,
any generator of the cone is crossed infinitely often.
We let A 6= 0 be any point of the cone and τ > 0 the smallest scalar
t for which φt (A) crosses the generator (OA). Since Dφ̃τ (A) has λ and λ2
as eigenvalues, we have to justify that there exist a third eigenvalue µ such
λ
that |µ| < λ. First, we may notice that the two eigenvectors A + 1−λ 2 τ X̃(A)

and X̃(A) belong to the tangent plane to the cone at A. Moreover, as the
cone is invariant, the third eigenvalue µ is necessarily positive. If we let
P1 (t) = (x(t), y(t), z(t)) be a point inside the cone and if we consider the
point
x(t) y(t)
P2 (t) = (z(t) p , z(t) p , z(t)),
2 2
x (t) + y (t) x (t) + y 2 (t)
2

the point P2 (t) belongs to the cone and P1 (t), P2 (t) are elements of the same
horizontal plane. Thus the distance P1 P2 (t) represents the distance from
P1 (t) to its ”horizontal projection” P2 (t). On the other hand, since
d d  p  d y2 p 
P1 P2 (t) = z(t) − x2 (t) + y 2 (t) = p x2 + y 2 − z < 0,
dt dt x2 + y 2
if we consider the map φτ , φτ is contracting distances to corresponding hori-
zontal projection points. As we may choose an eigenvector u associated to µ
pointing inward the cone, we can deduce that not only we have |µ| < λ but
we have also |µ| ≤ 1 < λ. According to our result, any polynomial vector field
X under the form

2
ẋ = a x y + b y − c y z + P1 (x, y, z)

(3.6) 2
ẏ = −a x − b x y + c x z + d y z + Q1 (x, y, z) ,


ż = d y 2 + R1 (x, y, z)

where P1 , Q1 and R1 are any linear functions defined on R3 with values in R

with P1 (0, 0, 0) = Q1 (0, 0, 0) = R1 (0, 0, 0) = 0, is unbounded.

Example 3.6. Let Ỹ be the homogeneous polynomial system of degree

3 on R3 given by

 2 3 2 2
ẋ = a x y + b y + c x z + d y z
(3.7) 3 2 2
ẏ = −a x − b x y − d x z ,

 2
ż = c x z

where a, b, c and d are real constants with c > 0. Here again, the cone ”x2 +
y 2 = z 2 ” is invariant and idempotents, if any, belong to the cone. As before,
we suppose that there is no idempotent. In this case also, if (x(t), y(t), z(t))
belongs to the cone, then z(t) increases without bound and the corresponding
356 H. BOUJEMAA AND S. EL QOTBI

Figure 2. The cone for m = 2

trajectory will spiral and blow up. Moreover, we have

d  p  c x2 z p 2 
z − x2 + y 2 = p x + y2 − z < 0
dt x2 + y 2
inside the cone and this implies, as for the previous example, that the third
eigenvalue µ verifies µ ≤ 1 < λ. Thus, any polynomial vector field X under
the form

ẋ = a x2 y + b y 3 + c x z 2 + d y z 2 + P2 (x, y, z)

(3.8) ẏ = −a x3 − b x y 2 − d x z 2 + Q2 (x, y, z) ,

 2
ż = c x z + R2 (x, y, z)
where P2 , Q2 and R2 are three polynomials in the variable x, y, z of degree
less than 3 with P2 (0, 0, 0) = Q2 (0, 0, 0) = R2 (0, 0, 0) = 0, is unbounded.

Figure 3. The cone for m = 3

 ON UNBOUNDED POLYNOMIAL DYNAMICAL SYSTEMS 357

Acknowledgement.
The authors would like to thank the referee for the valuable remarks which
helped to improve the content of this work.

References
[1] H. Boujemaa, S. E. Qotbi and H. Rouiouih, Stability of critical points of quadratic
homogeneous dynamical systems, Glas. Mat. Ser. III 51(71) (2016), 165–173.
[2] A. Cima and J. Llibre, Bounded polynomial vector fields, Trans. Amer. Math. Soc. 318
(1990), 557–579.
[3] C. S. Coleman, Boundedness and unboundedness in polynomial differential systems,
Nonlinear Anal. 8 (1984), 1287–1294.
[4] R. J. Dickson and L. M. Perko, Quadratic differential systems, Lockheed Res. Rep. No.
LMSC L-56-68-1, Lockheed Palo Alto Res. Lab, 1968.
[5] M. K. Kinyon and A. A. Sagle, Quadratic dynamical systems and algebras, J. Differ-
ential Equations 117 (1995), 67–126.
[6] L. Markus, Quadratic differential equations and non-associative algebra, in: 1960 Con-
tributions to the theory of nonlinear oscillations, Vol. V, Princeton Univ. Press, Prince-
ton, 1960, 185–213.
[7] S. Walcher, Algebras and differential equations, Hadronic Press, Palm Harbor, 1991.

H. Boujemaa
Département de Mathématiques
Mohammed V University in Rabat
1014RP Rabat
Morocco
E-mail : boujemaa@fsr.ac.ma

S. El Qotbi
Systèmes Dynamiques, A3D
Mohammed V University in Rabat
1014RP Rabat
Morocco
E-mail : Qotbis@gmail.com
Received: 24.11.2017.