Simple Homoclinic Cycles in Low-Dimensional Spaces: Nicola Sottocornola
Simple Homoclinic Cycles in Low-Dimensional Spaces: Nicola Sottocornola
Simple Homoclinic Cycles in Low-Dimensional Spaces: Nicola Sottocornola
Abstract
The problem of a classification of robust homoclinic cycles in low-dimensional spaces has
been frequently asked in recent years. In this paper, we resume the results in R3 and R4 and
we solve the problem in R5 in the case of orientation-preserving group actions.
© 2004 Elsevier Inc. All rights reserved.
Keywords: ODE; Homoclinic cycles; Equivariant systems
1. Introduction
Homoclinic cycles (and, more generally, heteroclinic ones) are flow-invariant sets of
equivariant dynamical systems (i.e., systems with symmetries). The presence of G, the
symmetry group, is a natural condition for the existence of these structures in a robust
form; we mean that the symmetries guarantee the structural stability of the cycles in
the set of all dynamical systems equivariant with respect to the same group G. In other
words this means that the cycle persists after a small perturbation of the vector field if
this perturbation does not break the symmetry of the system. The dynamical stability
of (heteroclinic) cycles has been deeply studied; the best results are available in [12].
If the cycle is robust and asymptotically stable it can be observed in physical systems
or, at least, on a computer. Homoclinic cycles explain certain intermittency phenomena
which are observed in systems with symmetry, see [3].
0022-0396/$ - see front matter © 2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.jde.2004.10.023
136 N. Sottocornola / J. Differential Equations 210 (2005) 135 – 154
In this paper, we only discuss the problem of a classification of simple, robust ho-
moclinic cycles (see Remark 1.2) in a space with low dimension (less or equal to
five). The case R3 is quite simple and it was already known in the 1980s when people
started to look at these objects. The problem in dimension four is first discussed in [9]
and completed in [16–18]. In this paper we resume these results and we give the clas-
sification in R5 in the case of pure rotations symmetry group. Somewhat surprisingly,
the extra room in 5-D does not allow any examples other than the obvious one; this
cycle was already studied by Field and Richardson (see [7]). These results are resumed
in the table in Fig. 2. Despite these results it seems difficult to describe the situation
in higher dimension (see the discussion at the end of the paper). For some interesting
examples, see [6].
The following definition can be found in [2]: let X be a vector field on Rn , equivariant
with respect to the linear action of a finite group G. Let = 0 be a hyperbolic saddle
point with an unstable manifold W u () and suppose that W u () ⊂ P = F ix(K), where
P is a two-dimensional fixed-point subspace corresponding to an isotropy subgroup
K ⊂ G.
Remark 1.2. According to [13] the adjective “simple” means that dim(P ) = 2 (and so
the unstable manifold has dimension 1: dim(W u ()) = 1). It is possible to relax this
condition obtaining more general definitions of homoclinic cycle. In this paper all the
cycles are supposed to be simple.
Remark 1.3. We recall some classical results which will be used in the following:
• We can suppose, without loss of generality, that G ⊂ O(n) (see [3, Theorem 4.4.3]);
• Two successive invariant planes Pj and Pj +1 are orthogonal (see [19, Remark 3.2]
or [13]);
• The group N(Kj )/Kj (Kj is the group fixing the plane Pj and N (Kj ) its normalizer)
acts on Pj as a dihedral group generated by a reflection and a rotation of an angle
2t where t is the angle between two consecutive equilibria j −1 and j in Pj (see
Definition 3.1). Remark that t has to be a rational angle because G is a finite group:
t = /k, k = 1, 2, 3 . . . (see [19, Remark 3.6] or [13]).
We suppose that the symmetry groups are the simplest ones allowing for the existence
of the homoclinic cycles and we call such groups MAGs:
Definition 1.4. We call minimal admissible group (MAG) a finite subgroup G of O(n)
such that:
• G is the symmetry group of a robust homoclinic cycle X ,
• X is not a robust homoclinic cycle for any proper subgroup of G.
N. Sottocornola / J. Differential Equations 210 (2005) 135 – 154 137
The following algebraic results are easily obtained using existing techniques in Galois
theory, but the exact formulations are new, so all details are provided.
(2)
1 = − 41 a + 1
4 a2 + 8 − 4b + 1
4 2a 2 − 2a a 2 + 8 − 4b − 8 − 4b, (3)
2 = − 41 a + 1
4 a 2 + 8 − 4b − 1
4 2a 2 − 2a a 2 + 8 − 4b − 8 − 4b, (4)
3 = − 41 a − 1
4 a 2 + 8 − 4b + 1
4 2a 2 + 2a a 2 + 8 − 4b − 8 − 4b, (5)
4 = − 41 a − 1
4 a 2 + 8 − 4b − 1
4 2a 2 + 2a a 2 + 8 − 4b − 8 − 4b. (6)
If these are roots of unity, there are integers h, k, m and n such that
2h 1 1 2
cos =− a+ a + 8 − 4b; (h, n) = 1, (7)
n 4 4
2k 1 1 2
cos =− a− a + 8 − 4b; (k, m) = 1 (8)
m 4 4
Here are some classical results in algebraic number theory (see for instance [10]):
Theorem 2.2. Roots of monic polynomials with algebraic integer coefficients are again
algebraic integers.
Proof. Let
f (x) = x n + an−1 x n−1 + · · · + a1 x + a0 (11)
be a polynomial and suppose that the aj are algebraic integers. Then let K be the
splitting field of f (x), and let 1 , . . . , m be the Galois automorphisms of K over Q.
Then consider the polynomial 1
g(x) = 1 (f )(x) . . . m (f )(x). (12)
We claim that the coefficients of g are integers. Note that g is invariant under all the
Galois automorphisms of K: if you apply an automorphism to that polynomial, you
just permute the order of the factors, so you get the same polynomial. That means that
the coefficients of the polynomial are in the ground field, so they have to be rational
numbers.
In addition, if a is an algebraic integer, then so is (a) for all : for if a satisfies
the polynomial
h(x) = x n + cn−1 x n−1 + · · · + c0 , (13)
then (a) also satisfies it ( fixes every ck , so just apply to h(a) to get h((a)) =
0). So the coefficients are also algebraic integers, since they are products and sums of
algebraic integers, by hypothesis.
Since the only rationals which are algebraic integers are the integers, and we know
that this polynomial has rational algebraic integer coefficients, it has integer coefficients.
The roots of f are also roots of g, so they are algebraic integers.
Corollary 2.3. Let a and b be two complex numbers. If a + b and ab are algebraic
integers, then so are a and b.
Proposition 2.4. Let be a rational angle and cos be an algebraic integer. Then
cos = 0, 1 or −1.
Proof. First of all 0, 1 and −1 are obviously algebraic integers. Moreover, let p | n
and cos(2m/n) be an algebraic integer. Then so is cos(2m/p). Let I denote the
set of all algebraic integers; we only have to prove that cos(2/p) ∈ / I if p is a
prime number greater than two (note that if cos(2/p) is an algebraic integer so are
cos(2k /p), k ∈ Z).
To get a contradiction, suppose that we have an odd prime number p such that
cos(2k /p) ∈ I. Consider the following number
p−1
2j
A= cos =⇒ A ∈ I. (14)
p
j =1
We claim that A = 21−p . To see this, let (X) be the pth cyclotomic polynomial;
remember that
p−1
(X) = X − e2ki /p = 1 + X + · · · + X p−1 (15)
k=1
and compute
p−1
p−1
e2ik /p + e−2ik /p p−1
p−1
2k
cos = = 21−p e2ik /p (1 + e−4ik /p ). (16)
p 2
k=1 k=1 k=1 k=1
The first product is 1 and the second (−1) = 1. So A = 21−p , p is an odd prime
number and A should be an integer. We have obtained a contradiction.
Theorem 2.5. The only rationals x and y, 0 < x y 1/2, such that the product
sin x sin y is a positive rational number as shown in the following table:
sin sin = sin sin − . (17)
6 2 2 2
with 0 < x1 < x3 x4 < x2 1/2 are given in the following table:
x1 x2 x3 x4
1/21 8/21 1/14 3/14
1/14 5/14 2/21 5/21
4/21 10/21 3/14 5/14
1/20 9/20 1/15 4/15
2/15 7/15 3/20 7/20
1/30 3/10 1/15 2/15
1/15 7/15 1/10 7/30
1/10 13/30 2/15 4/15
4/15 7/15 3/10 11/30
1/30 11/30 1/10 1/10
7/30 13/30 3/10 3/10
1/15 4/15 1/10 1/6
2/15 7/15 1/6 3/10
1/12 5/12 1/10 3/10
1/10 3/10 1/6 1/6
Theorem 2.8. Let n be a positive integer not in A. There is an nth primitive root of
unity in L.
Proof. What we need is to guarantee that the positive integer n has the property that
the interval In = (n/4, 13n/36) contains at least one integer k relatively prime to n.
An inclusion–exclusion argument gives us the formula
n
(n; c) = c (d), (22)
d
d|n
For a given number r of prime factors, the product on the right side can be no larger
than the value obtained by plugging in the first r distinct primes. Clearly, then, there is
only a limited number of possible values for r. The only value of N for which 13N/36
is an integer relatively prime to N is N = 36. Apart from that, we must have r < 5:
2 2 2 2 2 1 1
× × × × = < (30)
1 2 4 6 10 30 9
so there are no “exceptional” n with 5 or more prime factors. Thus, the possible n
for which there is no fraction in lowest terms k/n with denominator exactly n in the
given interval (1/4, 13/36) can be reduced to a manageable number of cases, which
can then be checked.
What we need now is estimates for the “Jacobsthal function” C(r), the maximum
distance between consecutive integers relatively prime to n, where n is any number
with r distinct prime factors.
142 N. Sottocornola / J. Differential Equations 210 (2005) 135 – 154
3. Homoclinic cycles in 13
These cycles have a very simple structure and they are known since the early work
of Dos Reis [4]. For a more detailed analysis of such cycles, see also [3,11,18,19].
We can introduce a direct orthonormal basis B = {e1 , e2 , e3 } where the action of the
twist has the form
P1 = e1 , e2 −→ P2 = e2 , e3 −→ P3 = cos(t) e2 + sin(t) e3 , e1 . (33)
The matrices A (of the twist ) and S (fixing the plane P1 ) in the basis B have the
form
0 0 1
A = sin(t) cos(t) 0 ; S = diag(1, 1, −1) (34)
− cos(t) sin(t) 0
with = ±1, det(A) = and t = /k, k = 1, 2, 3 . . . (see Remark 1.3). It is easy to
see that t is the angle between two consecutive equilibria.
Proof. The only possible values for t are and /2 by Lemma 3.2. If t = , we have
the 2-point cycle; if t = /2, n = 3 if det(A) = 1 and n = 6 if det(A) = −1.
Remark 3.4. The symmetry group for the cycles with 3 and 6 equilibria is the same
one. Its order is 24 and it is sometimes called Td .
4. Homoclinic cycles in 14
As previously stated we will consider only cycles which are not contained in a
hyperplane of R4 . We start looking to the action of on three successive invariant
planes in a chosen basis B:
with the introduction of two angles t and s. The angle t has the same meaning as in
R3 (see Section 3) while s is the angle between two consecutive hyperplanes (in the
obvious sense):
Q1 = P1 + P2 −→ Q2 = P2 + P3 −→ · · · −→ Qi = Pi + Pi+1 −→ · · ·
Definition 4.1. The angles t and s are called, respectively, connecting angle and hy-
perplane tilt angle. We will refer to s and t as to the structure angles of the cycle.
Proposition 4.2. The symmetry group G contains the matrix S = diag(1, 1, −1, −1).
We split the problem in two cases according to the nature of the symmetries:
Theorem 4.4. The structure angles of a homoclinic cycle in R4 , with symmetry group
included in SO(4), are multiples of /4.
With a direct analysis of the possible cases we obtain three different cycles 2
In this section the symmetry group G = A, S ⊂ / SO(4). The twist has, in the
basis B (see Eq. (39)), the following form
0 0 cos(s) − sin(s)
− sin(t) cos(t) 0 0
A= cos(t) sin(t) 0
;
det(A) = −1. (50)
0
0 0 sin(s) cos(s)
This Theorem can be shown using a little bit of Galois theory.
2 We can remark that the symmetry groups in the cases n = 12 and n = 24 are the same one (but
the twists are, of course, different). In a sense we can say that one is a “sub-cycle” of the other.
146 N. Sottocornola / J. Differential Equations 210 (2005) 135 – 154
2
cos(t) + cos(s) = 2 cos(a ), 0 <a< , (53)
3
1 1
cos(t) − cos(s) = −2 sin(b), − <b< (54)
2 6
Proof. Consider Eq. (56) and apply Theorem 2.6. From the equation
sin − sin − = sin(−) sin (57)
2 2 2 6
we obtain a = 1/2 and so t + s = . Let t − a = b = 0 to have s = t.
The other solutions do not give a finite group G.
and
P1 = e1 , e2 , (60)
P1 + P2 = e1 , e2 , e3 , (61)
P1 + P2 + P3 = e1 , e2 , e3 , e4 , (62)
P1 + P2 + P3 + P4 = R5 (63)
so we have at least four different invariant planes. The twist in the basis B has this
form
0 0 cos(s) − sin(s) cos(u) sin(s) sin(u)
sin(t) cos(t) 0 0 0
A= − cos(t) sin(t) 0 0 0
(64)
0 0 sin(s) cos(s) cos(u) − cos(s) sin(u)
0 0 0 sin(u) cos(u)
with = ±1, det(A) = . Furthermore, we have also
t= , k = 2, 3, 4 . . . s ∈ [0, ] u ∈ (0, ). (65)
k
As in the case R4 , t is the connecting angle and u the hyperplane twist angle; s does
not seem to have an interesting geometric meaning.
We now put = 1 in (64).
148 N. Sottocornola / J. Differential Equations 210 (2005) 135 – 154
Proof. Consider the group K1 fixing the plane P1 and one of its elements 1 = I .
This element must be the identity on P1 and ±1 on e3 , which is the only unstable
direction of 1 (see Definition 1.1 and Remark 1.2). If we have 1 (e3 ) = e3 , 1 is a
pure rotation in e4 , e5 and it is the element we are looking for. If 1 (e3 ) = −e3 this
element alone is not enough to fix P1 and so we need a second element 2 = I in K1 .
In this case their product gives the element R. Obviously y is a rational angle because
G is a finite group.
Remark 5.2. The group K1 cannot be generated by R because this matrix fixes all the
space e1 , e2 , e3 . So, as in the proof of Proposition 5.1, we can show that K1 must
contain an element S = I of the form
1 0 0 0 0
0 1 0 0 0
S= 0 0 −1 0 0 .
(67)
0 0 0 cos() sin()
0 0 0 sin() − cos()
Remark 5.3. According to the previous convention (see Section 1) we look for minimal
admissible groups so in the following of this section we will put G = A, R, S.
Theorem 5.4. The angle s in (64) must be equal to /2 while y in (66) must be equal
to .
Proof. The idea is to compute the matrix à = A−1 RA. This matrix is the identity on
the plane P1 so it belongs to K1 and we should have Ã(e3 ) = ±e3 . This computation
gives
0
0
Ã(e3 ) = cos (s) + cos(y) − cos(y) cos (s)
2 2 . (68)
(sin(u) sin(y) − cos(s) cos(u) + cos(s) cos(u) cos(y)) sin(s)
(cos(s) sin(u) − cos(s) sin(u) cos(y) + cos(u) sin(y)) sin(s)
So we have to solve the two equations
cos2 (s) + cos(y) sin2 (s) = 1 =⇒ cos(y) = 1, (69)
N. Sottocornola / J. Differential Equations 210 (2005) 135 – 154 149
The polynomial
1 (x) verifies the hypothesis of Theorem 2.1, so a and b are algebraic
integers and so is b − a = cos(t) cos(u). Now, the real numbers cos(t) and cos(u)
satisfy the hypothesis of Corollary 2.3, so they are algebraic integers too. Moreover, t
is a rational angle (see (65)); we can deduce that cos(t) = 0, 1 or −1 (see Proposition
2.4) hence that t = /2 (see (65) again).
Until now we have obtained the following results: t = s = /2 and y = ; we still
have to find the values of u and . Note that these angles are not rational angles a
priori, so we cannot use the same technics we have used till now. We start with u.
√
Theorem 5.6. The hyperplane tilt angle u must be equal to /2 or cos(u) = (1− 5)/2.
y 1
0
-1 -0.8 -0.6 -0.4 -0.2 0.2 0.4 0.6 0.8 1
x
-1
-2
1+x .
Fig. 1. The representative curve of the function y = − 2x+1
Now if it is
2 2h
− < cos < 0, (78)
3 n
Eq. (77) is impossible (see Fig. 1). Let N = LCM(n, m).
If n is a positive integer not in A (see (19)), Theorem 2.8 assures that there is an
automorphism of QN over Q which realizes condition (78). We have only to examine
the cases with n ∈ A (in each of the following solutions we can permute h and
k):
2k 2
• n = 1, cos = − , impossible;
m 3
2k 2h
• n = 2, cos = 0, cos = −1 =⇒ cos(u) = −1;
n n
2k 2h
• n = 4, cos = 0, cos = −1 =⇒ cos(u) = −1;
n n
√ √
2k 5−1 2h 5−1
• n = 5, cos = , cos =− =⇒ cos(u) = 0;
n 4 n 4
2k 3 2k
• n = 6, cos = − =⇒ 2 cos is not an algebraic integer, impossible;
n 4 n
√
2h 2k 2
• n = 8, cos = − cos = =⇒ cos(u) = 1;
n n 2
• n = 9, 12, 21, the equation has no solution;
N. Sottocornola / J. Differential Equations 210 (2005) 135 – 154 151
• n√= 30 we have two solutions: (h, k) = (1, 11) and√(7, 13). In the first case cos(u) =
5+1 1− 5
> 1 and in the second one cos(u) = .
2 2
√
The only possibilities are cos(u) = ±1, 0 or (1 − 5)/2. Because of (65) the solutions
±1 are not acceptable.
√
Proposition 5.7. If cos(u) = (1 − 5)/2 then u is not a rational angle.
√
√ Suppose that u = m/n is a rational angle. The conjugate of 5/2 − 1/2 is
Proof.
− 5/2 − 1/2 whose absolute value is greater than 1. But if Tn is the Chebyshev
polynomial of order n, then all the roots of Tn (x) − 1 are cos(m/n) for m even,
and all the roots of Tn (x) + 1 are cos(m/n) for m odd. So all the conjugates of
cos(m/n) (i.e. all the other roots of its minimal polynomial over the rationals) have
absolute value lower than 1.
We have now to split the proof in two subsections according to the value of the tilt
angle.
Finally, thanks to Theorems 5.4–5.6, and 5.8 the admissible groups can be obtained
with the following angles:
• t =s=u= ;
2
• y = ;
√
• ≡0 or sin() = ( 5 − 1)/2.
2
= ±/2 give the same group, as well as = 0 or = . We have only three cases
to examine:
√
5−1
t =s=u= , y = , sin() = 0, 1 or . (82)
2 2
First of all set sin() = 1. Using a computer program like GAP (see [8]) it is easy
to study this group and to show that this cannot be an admissible group. For instance
the element
0 1 0 0 0
1 0 0 0 0
A SA SA RA RSA RSA RSASA RA =
4 2 4 3 2 3 4 3
0 0 0 −1 0
0 0 1 0 0
0 0 0 0 −1
fixes the direction (1, 1, 0, 0, 0)√preventing the connections in P1 .
Suppose now that sin() = ( 5 − 1)/2. Let
(x) be the characteristic polynomial of
AS and compute
√ √ √
(x) 3 5 3 5 3 5
=x +4
− x +
3
− x +
2
− x + 1. (83)
x−1 2 2 2 2 2 2
As in Section 5.1 the possible values of a can be computed as in Eq. (76) showing
that the order of AS is not finite.
Before stating the last Theorem we have to remark that the group G = A, R, S
is now the same group as A, R because S = [A−1 , R]. The results of this section
can be finally summarized in this theorem.
Theorem 5.9. The only MAG G ⊂ SO(5) for a simple robust homoclinic cycle in R5
(which is not contained in a hyperplane) is the group G = A, R with
0 0 0 0 1
1 0 0 0 0
A= 0 1 0 0 0 and R = diag(1, 1, 1, −1, −1). (84)
0 0 1 0 0
0 0 0 1 0
The order of G is 80 and the cycle has five equilibria. G is isomorphic to the semidirect
product Z5 Z42 ; a group presentation of G is
G = a, r | a 5 = r 2 = [a, r]2 = (ar)5 = 1. (85)
Proof. The structure of G is easy to verify. The only thing we have to proof is the
existence of the 5-point cycle. This has been done in [7]. An alternative proof can be
N. Sottocornola / J. Differential Equations 210 (2005) 135 – 154 153
Fig. 2. Synoptic table of robust homoclinic cycles (h and k are integers, h > 2, k > 1).
obtained using a result by P. Ashwin and J. Montaldi (see [1, Theorem 4.1]) as in
[18,19].
First of all we sum up the available results in a table (see Fig. 2). As usual we
consider only cycles that are not included in an hyperplane. To avoid repetitions we
imply that G ⊂ O(n) also means G ⊂ / SO(n).
On the basis of the available results we are not able to choose between the following
alternatives:
• The classification in dimension 5, in the case of pure rotation symmetries, is trivial.
This can be true even in higher dimension. Similar situations have been often ob-
served in topology; for instance the symmetry groups of regular polytopes simplify
in dimension greater or equal to five. If this is true it would be interesting to know
if this is also true in the case O(n).
• The complexity of the classification with symmetry group in O(n) “blows up” with
n = 4: two infinite families of cycles appear. We can suppose that such a complexity
increases with the number of dimensions so that the classification in higher dimen-
sions became very difficult and maybe useless. In this case it seems probable that
even in the case of pure rotations there is a “critical dimension” where an infinite
number of cycles appear.
The technics employed till now in the study of homoclinic cycles do not seem easy to
apply in high-dimensional spaces. To completely solve the problem of the classification
of robust homoclinic cycles new technics have to be invented.
Acknowledgments
The author thanks P. Chossat and J.-P. Ortega of INLN, Nice (F), I. Melbourne of
Surrey University (UK) and J. Montaldi of UMIST, Manchester (UK), for many useful
154 N. Sottocornola / J. Differential Equations 210 (2005) 135 – 154
discussions and advices. Thanks a lot to M. Coleman of UMIST for his help in number
theory.
References
[1] P. Ashwin, J. Montaldi, Conditions for existence of robust relative homoclinic trajectories, Math.
Proc. Cambridge Philos. Soc. 133 (2002) 125–141.
[2] P. Chossat, M. Krupa, I. Melbourne, A. Scheel, Transverse bifurcation of homoclinic cycles, Physica
D 100 (1997) 85–100.
[3] P. Chossat, R. Lauterbach, Methods in equivariant bifurcation and dynamical systems, Adv. Ser.
Nonlinear Dynamics 15 (2000).
[4] G.L. Dos Reis, Structural stability of equivariant vector fields on two manifolds, Trans. Amer. Math.
Soc. 283 (2) (1984) 633–643.
[5] P. Du Val, Homographies Quaternions and Rotations, Oxford University Press, Oxford, 1964.
[6] M. Field, Lectures on Bifurcations, Dynamics and Symmetry, Pitman Research Notes in Mathematics
Series 336, Longman, New York, 1996.
[7] M. Field, R.W. Richardson, Symmetry breaking and branching patterns in equivariant bifurcation
theory II, Arch. Rational Mech. Anal. 120 (2) (1992) 147–190.
[8] The GAP Group, GAP—Groups, Algorithms, and Programming, Version 4.3, 2002,
http://www.gap-system.org.
[9] J. Guckenheimer, P. Holmes, Structurally stable heteroclinic cycles, Math. Proc. Cambridge Soc. 103
(1988) 189–192.
[10] N. Jacobson, Basic Algebra, Freeman and Company, New York, 1985.
[11] M. Krupa, Robust heteroclinic cycles, J. Nonlinear Sci. 7 (1997) 129–176.
[12] M. Krupa, I. Melbourne, Asymptotic stability of heteroclinic cycles in systems with symmetry, Ergodic
Theory Dynamics Systems 15 (1) (1995) 121–147.
[13] M. Krupa, I. Melbourne, Classification of heteroclinic cycles in R4 , Private communication.
[14] G. Myerson, Rational products of sines of rational angles, Aequationes Math. 45 (1993) 70–82.
[15] M. Newman, Some results on roots of unity, with an application to a diophantine problem, Aequationes
Math. 2 (1969) 163–166.
[16] N. Sottocornola, Sur la classification des cycles homoclines, C. R. Acad. Sci. Paris 332 (I) (2001)
695–698.
[17] N. Sottocornola, Complete classification of homoclinic cycles in R4 in the case of a symmetry group
G ⊂ SO(4), C. R. Acad. Sci. Paris, Sér. I Math. 334 (2002) 859–864.
[18] N. Sottocornola, Robust homoclinic cycles in R4 , Nonlinearity 16 (2003) 1–24.
[19] N. Sottocornola, Classification des cycles homoclines forcés par symétrie dans R4 , Ph.D. Thesis,
INLN, UMR 6618, CNRS, Sophia Antipolis, France, 2003.