On symmetric and periodic solutions of

parametric weakly nonlinear ODE with

time-reversal symmetries

Nataliya Dilna∗ Michal Fečkan†

Abstract

We show the existence of periodic and symmetric solutions of parametric

weakly nonlinear ODE possessing time-reversal symmetries. Local asymp-
totic behaviours of these solutions are established as well. Concrete examples
are presented to illustrate the general theory.

1 Introduction
We consider the systems of differential equations under symmetric assumptions.
More concretely, we consider a weakly nonlinear ordinary differential equation
of the form
ẋ = ε f ( x, µ, t), x ∈ R n , t ∈ R (1.1)
with parameters ε ∈ R, µ ∈ R k , where ε is small, and with a C∞ -smooth function
f : R n+k+1 → R n symmetric in x, i.e. it holds

A f ( x, µ, t) = − f ( Ax, µ, −t − τ ) , (1.2)
∗ Thefirst author was supported in part by the Fond of Štefan Schwarz and by the Grants
VEGA-SAV 2/0124/10 and APVV-0134-10.
† The second author was supported in part by the Grants VEGA-MS 1/0098/08, VEGA-SAV

2/0124/10 and APVV-0414-07.

Received by the editors February 2011.
Communicated by J. Mawhin.
2000 Mathematics Subject Classification : 34C14, 34C15, 34D20.
Key words and phrases : Periodic solution, symmetric systems, stability of solution, parametric
equations.

Bull. Belg. Math. Soc. Simon Stevin 18 (2011), 896–923

Symmetric Solutions 897

where A : R n → R n is a regular linear map, τ ∈ R is fixed and, moreover,

function f is T-periodic on t, i.e. it holds
f ( x, µ, t) = f ( x, µ, t + T ). (1.3)
A survey of dynamical systems with time-reversal symmetries is given in [21].
Note condition (1.2) represents such a kind of symmetry for (1.1).
On the other hand, there are several papers [15, 16, 24, 30, 34, 35] studying
ODE with symmetries when (1.2) is replaced with the following assumption
A f ( Ax, µ, t) = − f ( x, µ, −t − τ ) . (1.4)
Moreover, most of these papers suppose additional condition A2 = I, and then
(1.4) is called as property E. Furthermore, clearly property E is our assumption
(1.2) with A2 = I. Consequently, our results are generalizations of some earlier
results for weakly nonlinear ordinary differential equations with property E.
Note
g( x, µ, t) := f ( x, µ, t − τ/2)
satisfies (1.2) with τ = 0, so without loss of generality, we suppose
A f ( x, µ, t) = − f ( Ax, µ, −t) (1.5)
instead of (1.2). We introduce a vector space
n o
1 n +1
X := x ∈ C (R, R ) | x (t) = Ax (−t) ∀t ∈ R . (1.6)

Definition 1. By a symmetric solution x of equation (1.1) we mean x ∈ X satisfying

this equation.
The main goal of this paper is to find symmetric and periodic solutions (see
Section 4) for equation (1.1) and to study their asymptotic properties (see Sections
5 and 6). We also present examples to illustrate the theory in Section 8.
The results presented in this note are also generalizations of achievements for
anti-periodic problems with A = −I [1, 2], and continuations of [13]. Doubly
symmetric solutions of reversible systems are studied in [28]. Symmetric proper-
ties of periodic solutions of nonlinear nonautonomous ordinary differential equa-
tions are studied also in [9, 10, 11]. We can also apply numerical methods from
[31] for computation of symmetric solutions of (1.1). More results on periodic
solutions in dynamical systems and ordinary differential equations are presented
in [12, 23, 32].
Furthermore, when in addition, f is odd in x, i.e. it holds
f (− x, µ, t) = − f ( x, µ, t) , (1.7)
then we extend our result to the study of antisymmetric and periodic solutions of
(1.1), i.e. satisfying (cf Section 7)
− x (−t) = Ax (t) ∀t ∈ R (1.8)
instead of x ∈ X.
Finally, results of this paper are closely related to bifurcations of periodic so-
lutions presented in the books [7, 6, 14, 20, 36], but we remind that we also study
asymptotic properties of found periodic solutions of (1.1), not just their existence.
898 N. Dilna – M. Fečkan

2 Classical Results on Existence of Periodic Solutions

Before studying the existence of symmetric and periodic solutions of (1.1), we

recall the following classical results [27, 33].

Theorem 1. If there exist η̄0 ∈ R n and µ̄0 ∈ R k such that

Z T Z T
f (η̄0 , µ̄0 , s) ds = 0 and Dη,µ f (η̄0 , µ̄0 , s) ds : R n+k → R n is onto . (2.1)
0 0

Then there are decompositions R k = X̄1 ⊕ X̄2 , R n = Ȳ1 ⊕ Ȳ2 with dim X̄1 + dim Ȳ1 =
n and constants ε̄ 0 > 0, δ̄10 > 0, δ̄20 > 0, δ̄30 > 0, δ̄40 > 0 along with unique C∞ -
smooth functions µ̄1 (η2 , µ2 , ε) ∈ X̄1 , η̄1 (η2 , µ2 , ε) ∈ Ȳ1 , ε ∈ (−ε̄ 0 , ε̄ 0 ), |µ2 − µ̄02 | <
δ̄20 ,
|η2 − η̄20 | < δ̄40 such that

µ̄1 (η̄20 , µ̄02 , 0) = µ̄01 , η̄1 (η̄20 , µ̄02 , 0) = η̄10 for µ̄0 = µ̄01 , µ̄02 ∈
X̄1 × X̄2 , η̄0 = η̄10 , η̄20 ∈ Ȳ1 × Ȳ2 with the following properties: For any |µ1 − µ̄01 | <
δ̄10 , |µ2 − µ̄02 | < δ̄20 , |η1 − η̄10 | < δ̄30 , |η2 − η̄20 | < δ̄40 and 0 < |ε| < ε̄ 0 , equation
(1.1) has a T-periodic solution with x (0) = (η1 , η2 ) if and only if µ1 = µ̄1 (η2 , µ2 , ε),
η1 = η̄1 (η2 , µ2 , ε), moreover this solution is unique and located near η̄0 .

Theorem 2. If there are compact subsets Ω ⊂ R n and Γ ⊂ R k such that

Z T
min f ( x, µ, s) ds > 0 ,
x ∈Ω,µ ∈Γ 0

then (1.1) has no T-periodic solutions in Ω for any ε 6= 0 small and µ ∈ Γ.

3 Existence of symmetric solutions

We suppose for simplicity that f is globally Lipschitz continuous in x. From the

property
Ax (t) = x (−t) (3.1)
we have that
Ax (0) = x (0). (3.2)

Lemma 1. A solution x of (1.1) is symmetric if and only if it satisfies (3.2).

Proof. Let us put

y(t) := A−1 x (−t).
Then, taking into account (1.5), we get

ẏ (t) = − A−1 ẋ (−t) = − A−1 ε f ( x (−t), µ, −t) = ε f ( A−1 x (−t), µ, t) = ε f (y(t), µ, t)

and
y (0 ) = A − 1 x (0 ) = x (0 ).
So x (t) = y(t). Consequently, (3.1) holds.
Symmetric Solutions 899

Remark 1. It follows from the above considerations that any symmetric and
T-periodic solution is not asymptotically stable, but it can be stable (cf Exam-
ple 2). Moreover, if a symmetric and T-periodic solution is hyperbolic then the
dimensions of its stable and unstable manifolds are equal and so n is even.
From (3.2) we have
x (0) ∈ ker(I − A).
Let us consider equation (1.1) with initial value condition
x (0) = η, η ∈ ker(I − A) (3.3)
and take its unique C∞ -smooth solution x (η, ε, µ, t), x : ker(I − A) × R × R k ×
R → R n . Summarizing we arrive at the following result.
Theorem 3. The Cauchy problem (1.1), (3.3) has a unique C∞ -smooth solution x (η, ε, µ, t)
which is also symmetric, and any symmetric solution x (t) of (1.1) satisfies (3.3).

4 Existence of symmetric and periodic solutions

If x (t) is T-periodic and satisfying (3.1) then we get
x (T/2) = x (− T/2) = Ax (T/2) ,
so
x (T/2) ∈ ker(I − A) . (4.1)
On the other hand, if x (η, ε, µ, T/2) ∈ ker(I − A) then
x (η, ε, µ, − T/2) = x (η, ε, µ, T/2) ,
so x (η, ε, µ, t) is T-periodic. Consequently, in order to find symmetric and peri-
odic solutions of (1.1), we have to study the following equation
F(η, µ, ε) := Sx (η, ε, µ, T/2) = 0 , (4.2)
where I − S : R n → ker(I − A) is a A-invariant projection, i.e. AS = SA. Let
V := ker(I − S) .
Note
p := dim V = n − dim ker(I − A) .
Since
F(η, µ, 0) = Sη = 0,
we solve equation
1
F(η, µ, ε) = 0, ε 6= 0. (4.3)
ε
To state the next results we introduce the following function
H1 (η, µ) := Dε F(η, µ, 0).
Now we suppose that
m := dim ker(I − A) + k ≥ p. (4.4)
Then note H1 ∈ C∞ (R m , R p ).
900 N. Dilna – M. Fečkan

Remark 2. Let us consider decomposition

x (η, ε, µ, t) = η + εx1 (η, µ, t) + ε2 x2 (η, µ, t) + ε3 x3 (η, µ, t) + . . . (4.5)

for the Cauchy problem (1.1), (3.3). Then we get

ẋ1 (η, µ, t) + ε ẋ2 (η, µ, t) + ε2 ẋ3 (η, µ, t) + . . .

= f η + εx1 (η, µ, t) + ε2 x2 (η, µ, t) + ε3 x3 (η, µ, t) + · · · , µ, t , (4.6)
x j (η, µ, 0) = 0 ∀ j ∈ N.

Putting ε = 0 in (4.6), we have

Z t
x1 (η, µ, t) = f (η, µ, s)ds . (4.7)
0

Similarly, differentiating (4.6) by ε once and twice at ε = 0, we derive

Z t Z s
x2 (η, µ, t) = Dx f (η, µ, s) f (η, µ, z)dzds , (4.8)
0 0

and
Z t Z s Z z
x3 (η, µ, t) = Dx f (η, µ, s) Dx f (η, µ, z) f (η, µ, u)dudzds
0 0 0
Z  Z s Z s  (4.9)
1 t
+ Dxx f (η, µ, s) f (η, µ, z)dz, f (η, µ, z)dz ds ,
2 0 0 0

respectively.
Then, taking into account (4.7), we return to (4.2)
Z T/2
H1 (η, µ) = Dε F(η, µ, 0) = Sx1 (η, µ, T/2) = S f (η, µ, s)ds.
0

Next, (1.5) implies

Z T/2 Z T/2
AH1 (η, µ) = AS f (η, µ, s)ds = −S f ( Aη, µ, −s)ds
0 0
Z T/2 Z T
= −S f (η, µ, T − s)ds = −S f (η, µ, s)ds
0 T/2
Z T
= −S f (η, µ, s)ds + H1 (η, µ).
0

By using 1 ∈
/ σ( A/V ) and H1 (η, µ) ∈ V, we derive
Z T
−1
H1 (η, µ) = (I − A) S f (η, µ, s)ds . (4.10)
0

We first study the nondegenerate case:

Symmetric Solutions 901

4.1 The case ker(I − A) = {0}

Then S = I and by (4.4), m = k ≥ p = n. Now we can prove the following result.

Theorem 4. If there exists µ0 ∈ R k such that
Z T/2 Z T/2
f (0, µ0 , s) ds = 0 and Dµ f (0, µ0 , s) ds : R k → R n is onto . (4.11)
0 0

Then there is a decomposition R k = X1 ⊕ X2 with dim X1 = n and constants ε 0 > 0,

δ10 > 0, δ20 > 0 along with a unique C∞ -smooth function µ1 (µ2 ,
ε) ∈ X1 , ε ∈ (−ε 0 , ε 0 ),
|µ2 − µ02 | < δ20 such that µ1 (µ02 , 0) = µ01 for µ0 = µ01 , µ02 ∈ X1 × X2 with the
following properties: For any |µ1 − µ01 | < δ10 , |µ2 − µ02 | < δ20 and 0 < |ε| < ε 0 ,
equation (1.1) has a T-periodic and symmetric solution if and only if µ1 = µ1 (µ2 , ε),
moreover this solution is unique, so that it is given by x (0, ε, µ1 (µ2 , ε), µ2 , t) and thus it
is located near 0 in R n .
Proof. By (4.11) there is a decomposition R k = X1 ⊕ X2 with dim X1 = n such
that Z T/2 
0 0
det Dµ1 f (0, µ1 , µ2 , s) ds 6= 0 ,
0
where µ = (µ1 , µ2 ) ∈ X1 × X2 . We set a function
 1
G ( µ1 , µ2 , ε ) = ε F (0, µ1 , µ2 , ε) for ε 6= 0,
H1 (0, µ1 , µ2 ) for ε = 0.
Clearly that G is C∞ -smooth. We solve
G (µ1 , µ2 , ε) = 0. (4.12)
From our assumptions we have
G (µ01 , µ02 , 0) = H1 (0, µ01 , µ02 ) = 0
and
det Dµ1 G (µ01 , µ02 , 0) = det Dµ1 H1 (0, µ01 , µ02 ) 6= 0.
Now applying Implicit Function Theorem on (4.12) the proof is finished.
Using topological degree methods from [25] we can get the next result.
Theorem 5. Assume a decomposition R k = X1 ⊕ X2 with dim X1 = n and the
existence of an open bounded subset Ω ⊂ X1 along with µ02 ∈ X2 such that 0 ∈ /
0 0
H1 (0, ∂Ω, µ2 ) and deg( H1 (0, ·, µ2 ), Ω, 0) 6= 0, then for any small ε 6= 0 and µ2 near
µ02 there exists µ1 (µ2 , ε) ∈ Ω such that (1.1) with µ1 = µ1 (µ2 , ε) has a T-periodic and
symmetric solution.
Next we have the following result.


Theorem 6. Assume ker (I − A) = ker I − A2 = {0}. Then x (t) = 0 is the only
symmetric solution of (1.1) for any ε 6= 0 small.


Proof. By (1.5) we obtain A2 f (0, µ, t) = f (0, µ, t) and so f (0, µ, t) ∈ ker I − A2 .
Hence f (0, µ, t) = 0 and the proof is finished.
902 N. Dilna – M. Fečkan

4.2 The case ker(I − A) 6= {0}

Then p = dim V < n. We recall (4.4). Now we are ready to prove the following
result.
Theorem 7. If there exist η0 ∈ ker(I − A) and µ0 ∈ R k such that
Z T/2 Z T/2
S f (η0 , µ0 , s) ds = 0 and S Dµ f (η0 , µ0 , s) ds : R m → R p is onto .
0 0
(4.13)
Then there are decompositions Rk = X1 ⊕ X2 , ker(I − A) = Y1 ⊕ Y2 with dim X1 +
dim Y1 = n and constants ε 0 > 0, δ10 > 0, δ20 > 0, δ30 > 0, δ40 > 0 along with
unique C∞ -smooth functions µ1 (η2 , µ2 , ε) ∈ X1 , η1 (η2 , µ2 , ε) ∈ Y1 , ε ∈ (−ε 0 , ε 0 ),
|µ2 − µ02 | <
δ20 , |η2 − η20 | < δ40 such that 0 0 0 0 0 0

µ1 (η2 , µ2 , 0) = µ1 , η1 (η2 , µ2 , 0) = η1 for
0 0 0 0
µ0 = µ1 , µ2 ∈ X1 × X2 , η0 = η1 , η2 ∈ Y1 × Y2 with the following properties: For
any |µ1 − µ01 | < δ10 , |µ2 − µ02 | < δ20 , |η1 − η10 | < δ30 , |η2 − η20 | < δ40 and 0 < |ε| < ε 0 ,
equation (1.1) with (3.3) has a T-periodic and symmetric solution if and only if µ1 =
µ1 (η2 , µ2 , ε), η1 = η1 (η2 , µ2 , ε), moreover this solution is unique, so that it is given by
x (ε, η2 , µ2 , t) := x (η1 (η2 , µ2 , ε), η2 , ε, µ1 (η2 , µ2 , ε), µ2 , t).
Proof. The proof is very similar to the proof of Theorem 4, so we omit details (cf
[5, 7, 20]).
Similarly we can extend Theorem 5, but we leave this to the reader. Next,
taking
R n = ker (I − A) ⊕ ker (I + A) ⊕ W (4.14)
with AW = W and ±1 ∈
/ σ( A0 ) for A0 := A/W. We note
A(η, y, z) = (η, −y, A0 z) (4.15)
for η ∈ ker(I − A), y ∈ ker(I + A) and z ∈ W. Then (1.5) implies
f 1 (η, y, z, µ, t) = − f 1 (η, −y, A0 z, µ, −t) ,
f 2 (η, y, z, µ, t) = f 2 (η, −y, A0 z, µ, −t) , (4.16)
A0 f 3 (η, y, z, µ, t) = − f 3 (η, −y, A0 z, µ, −t)
for
f (η, y, z, µ, t) = ( f 1 (η, y, z, µ, t), f 2 (η, y, z, µ, t), f 3 (η, y, z, µ, t))
∈ ker (I − A) × ker (I + A) × W .
Then S(η, y, z) = (0, y, z) and V = ker (I + A) ⊕ W. Moreover from
A20 f 3 (η, y, z, µ, t) = f 3 (η, y, A20 z, µ, t) ,
we have
A20 f 3 (η, y, 0, µ, t) = f 3 (η, y, 0, µ, t) .
So if ker(I − A20 ) = {0} then f 3 (η, y, 0, µ, t) = 0 and symmetric solutions lie in a
subspace ker (I − A) ⊕ ker (I + A). Hence a bifurcation function is reduced to
Z T/2 Z T
1
Ĥ1 (η, µ) := f 2 (η, 0, 0, µ, t)dt = f 2 (η, 0, 0, µ, t)dt (4.17)
0 2 0

instead of H1 (η, µ) (cf (4.10)).

Symmetric Solutions 903

5 Asymptotic properties of symmetric and periodic solutions:

The case A = −I
In order to investigate asymptotic properties of symmetric and periodic solutions
derived in Section 4, we first consider the case A = −I. Now S = I and (1.5) has
the form
− f ( x, µ, t) = − f (− x, µ, −t) ,
which gives
− Dx f ( x, µ, t) = Dx f (− x, µ, −t) . (5.1)
Let
φ(ξ, µ, ε) := x (ξ, ε, µ, T ) .
Note by Theorem 4 that ξ = 0 is a fixed point of φ(·, µ, ε) if and only if µ =
µ(µ2 , ε) := (µ1 (µ2 , ε), µ2 ) and it corresponds to a unique symmetric and periodic
solution of (1.1) for ε 6= 0 small and µ2 near µ02 . So we set

ψ(ξ, µ2 , ε) := φ(ξ, µ(µ2 , ε), ε) ,

and a linear asymptotic property

of ξ = 0 for ψ(·, µ2 , ε) is determined by the
spectrum σ Dξ ψ(0, µ2 , ε) of Dξ ψ(0, µ2 , ε). Since (5.1) implies
Z T
Dx f (0, µ, t) = 0 ,
0

the usual first order averaging methods cannot be applied (cf Theorem 13, [27]).
For this reason, from (1.1) we derive

ẋξ (0, ε, µ(µ2 , ε), t) = εAµ2 ,ε (t) xξ (0, ε, µ(µ2 , ε), t)

(5.2)
xξ (0, ε, µ(µ2 , ε), 0) = I ,

where
Aµ2 ,ε (t) := Dx f ( x (0, ε, µ(µ2 , ε), t), µ(µ2 , ε), t) .
Next it holds
− x (0, ε, µ(µ2 , ε), t) = x (0, ε, µ(µ2 , ε), −t),
(5.3)
x (0, ε, µ(µ2 , ε), t + T ) = x (0, ε, µ(µ2 , ε), t) .
Then (5.1) and (5.3) imply

− Aµ2 ,ε (t) = − Dx f ( x (0, ε, µ(µ2 , ε), t), µ(µ2 , ε), t)

= Dx f (− x (0, ε, µ(µ2 , ε), t), µ(µ2 , ε), −t) (5.4)
= Dx f ( x (0, ε, µ(µ2 , ε), −t), µ(µ2 , ε), −t) = Aµ2 ,ε (−t) .

Since
Aµ2 ,ε (t + T ) = Aµ2 ,ε (t) ,
from the Floquet theory [19] we have

xξ (0, ε, µ(µ2 , ε), t + T ) = xξ (0, ε, µ(µ2 , ε), t) Bµ2 ,ε (5.5)

904 N. Dilna – M. Fečkan

for a regular matrix Bµ2,ε . Moreover, by (5.4), clearly xξ (0, µ(µ2 , ε), ε, −t) satisfies
(5.2), from the uniqueness of initial value problem, it follows

xξ (0, ε, µ(µ2 , ε), −t) = xξ (0, ε, µ(µ2 , ε), t) .

Then (5.5) gives

xξ (0, ε, µ(µ2 , ε), T/2) = xξ (0, ε, µ(µ2 , ε), − T/2) Bµ2 ,ε = xξ (0, ε, µ(µ2 , ε), T/2) Bµ2 ,ε

and so
Bµ2 ,ε = I .
Consequently, we arrive at

xξ (0, ε, µ(µ2 , ε), t + T ) = xξ (0, ε, µ(µ2 , ε), t) ,

that is
Dξ ψ(0, µ2 , ε) = xξ (0, ε, µ(µ2 , ε), T ) = I . (5.6)
Summarizing, we cannot apply the linear asymptotic theory in this case for sym-
metric and periodic solutions. Furthermore, (1.1) implies
Z T  
2
ψ(ξ, µ2 , ε) = ξ + ε f (ξ, µ0 , t) dt + O ε + |ε(µ2 − µ02 )| ,
0

which gives
Z T  
Dξξ ψ(0, µ2 , ε) = ε Dxx f (0, µ0 , t) dt + O ε2 + |ε(µ2 − µ02 )| . (5.7)
0

We immediately arrive at the following result [17, 19, 26].

Theorem 8. Suppose n = 1 in Theorem 4. If in addition

Z T
Dxx f (0, µ0 , t) dt 6= 0 ,
0

then the T-periodic and symmetric solution x (0, ε, µ1 (µ2 , ε), µ2 , t) is a saddle-node.

Proof. We have
Z T   
ξ2 2
ψ(ξ, µ2 , ε) = ξ + ε Dxx f (0, µ0 , t) dt + O ε + |ε(µ2 − µ02 )| 2
ξ + εξ 3
,
2 0

which immediately gives the proof.

The case n > 1 is more complicated. We intend to apply some results from
papers [3, 4]. First we recall for the reader convenience the following theorem of
[4].
Symmetric Solutions 905

Theorem 9. If a mapping F = F1 , F2 ∈ C∞ R × R n−1, R × R n−1 has a form
 
F1 ( x, y) = x + a0 x2 + x hb0 , yi + h A0 y, yi + O |z|3 ,
  (5.8)
2 3
Fj ( x, y) = y + x hb j , yi + h A j y, yi + O |z| , j = 1, 2, · · · , n − 1 ,


where h·, ·i is the usual scalar product on R n−1, F2 = F12 , · · · , Fn2−1 , x ∈ R, y =


(y1 , · · · , yn−1 ) ∈ R n−1, z = ( x, y) ∈ R n , b0 , b j ∈ R n−1, A0 , A j ∈ L R n−1 are


symmetric matrices, a0 < 0 and ℜσ( B) > 0 for B := (b1 , · · · , bn−1 )∗ ∈ L R n−1 .
Then there is a t0 > 0 and a local curve K ∈ C1 ((−t0 , t0 ), R n ) ∩ C∞ ((−t0 , t0 ) \
{0}, R n ) passing through (0, 0) which is invariant for F and the dynamics of F restricted
on K is equivalent
to the local dynamics of a polynomial R : R → R with R(t) =
t + a0 t2 + O t3 . Moreover it holds K (t) = (t, 0) + O(t2 ).

The next theorem gives a condition on F to have the form of (5.8).

Theorem 10. If a mapping F ∈ C∞ (R n , R n ) has a form F(0) = 0, DF(0) = I,

1 2 n 2
2 D F (0) = B 6= 0 and there are λ0 ∈ R, x0 ∈ R , | x0 | = 1 such that B x0 = λ0 x0 .
Then F has a form of (5.8) near 0 for the orthogonal decomposition R n = [ x0 ] ⊕ [ x0 ]⊥ .

Proof. Let P : R n → [ x0 ] be the orthogonal projection. Then z = xx0 + y, x ∈ R,

y ∈ [ x0 ]⊥ , F1 = PF and F2 = QF for Q = I − P. We compute
  
F2 ( x, y) = QF( xx0 + y) = Q xx0 + y + B( xx0 + y)2 + O |z|3
     
= y + Q x2 B x02 + 2x B x0 y + B y2 + O |z|3 = y + Q x2 λ0 x0 + 2x B x0 y + B y2
   
3 2 3
+O |z| = y + 2xQB x0 y + QB y + O |z| ,

and similarly
   
F1 ( x, y) = PF( xx0 + y) = x + x2 λ0 x0 + 2xPB x0 y + PB y2 + O |z|3 .

We see that F has a form of (5.8) with a0 = λ0 and B = 2QB x0 · |[ x0 ]⊥ . The proof
is finished.

Now we show a perturbation stability condition for (5.8).

Theorem 11. Let a symmetric quadratic mapping B0 : R n → R n have a form 21 D2 F(0)

of (5.8) with a0 < 0 and ℜσ( B) > 0. If a symmetric quadratic mapping B1 : R n → R n
is sufficiently near to B0 then B1 has a form of (5.8) as well.

Proof. It is easy to verify that B0 satisfies assumptions of Theorem 10 for x0 =

(1, 0, · · · , 0) and λ0 = a0 . The vector space of all symmetric and quadratic map-
pings from R n to R n can be identified with R

M for M := n(n + 1)/2. Then we

consider a mapping H ∈ C∞ R n+1+ M , R n+1 defined by

 
H(z, λ, B) := B z2 − λz, |z|2 − 1 .
906 N. Dilna – M. Fečkan

Clearly H( x0 , a0 , B0 ) = 0. Next we have

0 = D(z,λ) H( x0 , a0 , B0 )(u, θ ) = (2B0 x0 u − a0 u − θx0 , 2h x0 , ui)

= a0 v + hb0 , wi − θ, hb j , wi − a0 w j , 2v ,
u = (v, w) ∈ R × R n−1, j = 1, 2, · · · , n − 1 ,

when v = 0, Bw = a0 w and θ = hb0 , wi. Since ℜσ( B) > 0, we get w = 0 and then
θ = 0. Consequently, we can apply Implicit Function Theorem for equation

H(z, λ, B1 ) = 0

to obtain its local C∞ -solution z = z(B1 ) and λ = λ(B1 ) for any B1 near B0 .
Then a0 (B1 ) = λ(B1 ) and B(B1 ) = 2QB1 B1 z(B1 ) · |[z(B1 )]⊥ with the orthogonal
projection QB1 : R n → [z(B1 )]⊥ . Note a0 (B0 ) = a0 < 0 and B(B0 ) = B. Hence
ℜσ( B(B1 )) > 0. The proof of Theorem 11 is finished.
Now we can prove the following result.
Theorem 12. Suppose n > 1. Let the assumptions of Theorem 4 be satisfied. If in
addition Z
1 T
B := Dxx f (0, µ0 , t) dt
2 0
has a negative eigenvalue with eigenvector x0 such that ℜσ( B) > 0 for B := 2QB x0 ·
|[ x0 ]⊥ with the orthogonal projection Q : R n → [ x0 ]⊥ then the T-periodic and symmet-
ric solution x (0, ε, µ1 (µ2 , ε), µ2 , t) has a local saddle-node dynamics. Hence it is unstable.
Proof. The proof follows directly from (5.7) and Theorems 9, 10 and 11.
Concrete examples are presented in Section 8.1.

6 Asymptotic properties of symmetric and periodic solutions:

The case A 6= −I

6.1 Hyperbolicity of periodic solutions

To study stability of the T-periodic and symmetric solution of equation (1.1) we

recall the approach of [9, 10, 29]. For this aim we consider a C∞ -mapping

Φε,η2 ,µ2 (η ) := x (η, ε, µ1 (η2 , µ2 , ε), µ2 , T ) for η ∈ R n .

Note Φε,η2 ,µ2 (η (ε, η2 , µ2 )) = η (ε, η2 , µ2 ) for η (ε, η2 , µ2 ) := (η1 (η2 , µ2 , ε), η2 ). By
Remark 2, its linearization at η (ε, η2 , µ2 ) has the decomposition (cf (4.5) and (4.7))
Z T
DΦε,η2 ,µ2 (η (ε, η2 , µ2 )) = I + ε Dx f (η0 , µ0 , s)ds
 0
 (6.1)
+O ε2 + |ε(η2 − η20 )| + |ε(µ2 − µ02 )| .

By following [9, 27, 29] we obtain the following well-known result.

Symmetric Solutions 907

Theorem 13. For any ε > 0 sufficiently small, the symmetric and T-periodic solution
x (ε, η2 , µ2 , t) of (1.1) from Theorem 7 has the following asymptotic properties:
n R o
T
• If ℜ σ 0 Dx f (η0 , µ0 , s) ds ⊂ (−∞, 0) then x (ε, η2 , µ2 , t) is asymptoti-
callynstable.
R o
T
• If ℜ σ 0 Dx f (η0 , µ0 , s) ds ∩ (0, ∞) 6= ∅ then x (ε, η2 , µ2 , t) is unstable.
n R o
T
• If ℜ σ 0 Dx f (η0 , µ0 , s) ds ⊂ (0, ∞) then x (ε, η2 , µ2 , t) is a repeller.
n R o
T
• If ℜ σ 0 Dx f (η0 , µ0 , s) ds ∩ {0} = ∅ then x (ε, η2 , µ2 , t) is hyperbolic
RT
with the same hyperbolicity type as 0 Dx f (η0 , µ0 , s) ds.
By (4.16) after some computations we derive
Dη f 1 (η, 0, 0, µ, t) = − Dη f 1 (η, 0, 0, µ, −t), Dy f 1 (η, 0, 0, µ, t) = Dy f 1 (η, 0, 0, µ, −t)
Dz f 1 (η, 0, 0, µ, t) = − Dz f 1 (η, 0, 0, µ, −t) A0 , Dη f 2 (η, 0, 0, µ, t) = Dη f 2 (η, 0, 0, µ, −t)
Dy f 2 (η, 0, 0, µ, t) = − Dy f 2 (η, 0, 0, µ, −t), Dz f 2 (η, 0, 0, µ, t) = Dz f 2 (η, 0, 0, µ, −t) A0
A0 Dη f 3 (η, 0, 0, µ, t) = − Dη f 3 (η, 0, 0, µ, −t)
A0 Dy f 3 (η, 0, 0, µ, t) = Dy f 3 (η, 0, 0, µ, −t)
A0 Dz f 3 (η, 0, 0, µ, t) = − Dz f 3 (η, 0, 0, µ, −t) A0 .
(6.2)
By (6.2) we derive
RT RT R T/2
0 D η f 1 ( η, 0, 0, µ, t ) dt = 0, 0 D y f 1 ( η, 0, 0, µ, t ) dt = 2 0 Dy f 1 (η, 0, 0, µ, t)dt
RT
RT 0 Dz f 1 (η, 0, 0, µ, t)Rdt ( I + A0 ) = 0
T/2
RT 0 Dη f 2 (η, 0, 0, µ, t)dt R = 2 0 Dη f 2 (η, 0, 0, µ, t)dt
T
0 DRy f 2 (η, 0, 0, µ, t)dt = 0, 0 Dz f 2 (η, 0, 0,RT
µ, t)dt (I − A0 ) = 0
T
(I + A0 ) 0 Dη f 3 (η, 0, 0, µ, t)dt = 0, (I − A0 ) 0 Dy f 3 (η, 0, 0, µ, t)dt = 0
RT RT
A0 0 Dz f 3 (η, 0, 0, µ, t)dt = − 0 Dz f 3 (η, 0, 0, µ, t)dtA0 .
(6.3)
Then (6.3) and ±1 ∈ / σ( A0 ) imply
Z T Z T
Dz f 1 (η, 0, 0, µ, t)dt = 0, Dz f 2 (η, 0, 0, µ, t)dt = 0 ,
0 0
Z T Z T (6.4)
Dη f 3 (η, 0, 0, µ, t)dt = 0, Dy f 3 (η, 0, 0, µ, t)dt = 0 .
0 0
Summarizing by (6.3) and (6.4) it holds
Z T
Dx f (η0 , µ0 , s) ds =
0
 RT 
0 0 Dy f 1 (η0 , µ0 , t)dt 0
 RT 
 0 Dη f 2 (η0 , µ0 , t)dt 0 0 
RT
0 0 0 Dz f 3 (η0 , µ0 , t)dt
for matrices
RT
D f (η , µ , t)dt : ker(I + A) → ker(I − A) ,
R0T y 1 0 0
D f (η , µ , t)dt : ker(I − A) → ker(I + A) ,
R0T η 2 0 0
0 Dz f 3 (η0 , µ0 , t)dt : W → W .
908 N. Dilna – M. Fečkan

RT
Clearly, if 0Dx f (η0 , µ0 , s) ds is hyperbolic then dim ker(I + A) = dim ker(I −
RT
A). On the other hand, if dim ker(I + A) = dim ker(I − A) 6= 0 then 0 Dx f (η0 , µ0 , s) ds
is hyperbolic if and only if
 Z T Z T 
2
ℜ λ∈C|λ ∈σ Dy f 1 (η0 , µ0 , t)dt Dη f 2 (η0 , µ0 , t)dt ∩ {0} = ∅ ,
0 0
 Z T 
ℜ σ Dz f 3 (η0 , µ0 , t)dt ∩ {0} = ∅ .
0

Of course when dim ker(I + A) = dim ker(I − A) = 0 then we suppose

 Z T 
ℜ σ Dz f 3 (0, µ0 , t)dt ∩ {0} = ∅ .
0

6.2 k-Hyperbolicity

To study more sophisticated hyperbolicity of periodic solutions of equation (1.1)

we need the following results from [9, 29].
Definition 2. A continuous matrix function Lε : R n → R n of ε ≥ 0 and such that
L0 = I, is k-hyperbolic if for every matrix function Nε defined for ε ≥ 0 satisfying
Nε = o(εk ), there exists an interval 0 < ε < ε 1 in which Lε + Nε is hyperbolic of
the same type (i.e., with the same number of eigenvalues on each side of the unit
circle).
Definition 3. A continuous matrix function Lε : R n → R n of ε ≥ 0 and such that
L0 = I, is strongly k-hyperbolic if there exists a continuous real matrix Cε defined
in an interval 0 ≤ ε < ε 0 such that Cε is regular (even for ε = 0) and such that
 
−1 Aε 0
C Lε Cε =
0 Bε

for 0 < ε < ε 0 , where Aε , Bε are r × r and s × s blocks, respectively, and k Aε k <
1 − cεk , k Bε−1 k < 1 − cεk , for some c > 0.
Theorem 14 ([29, Theorem 2.2]). If Lε = I + εL1 + · · · + εk Lk , if the eigenvalues
of L1 are distinct numbers on the unit circle, and if the eigenvalues λi (ε) of Lε suitably
numbered satisfy |λi (ε)| < 1 − cεk for i = 1, . . . r, |λi (ε)| > 1 + cεk for i = r + 1, . . . , n,
for some constant c > 0 and ε > 0 small, then Lε is strongly k-hyperbolic.
If m = n then Φε,η2 ,µ2 and η (ε, η2 , µ2 )) depend only on ε, so we have Φε and
η (ε). Now we can improve Theorem 13 as follows.
Theorem 15. Suppose m = n. Let DΦε (η (ε)) = I + εM k k
RT  1 + · · · + ε Mk + o(ε ). Sup-
pose that all eigenvalues of M1 = 0 Dx f (η0 , µ0 , s)ds are distinct complex numbers
on the unit circle. If the eigenvalues λi (ε) of I + εM1 + · · · + εk Mk suitably numbered
satisfy |λi (ε)| < 1 − cεk for i = 1, . . . r, |λi (ε)| > 1 + cεk for i = r + 1, . . . , n, for some
constant c > 0 and ε > 0 small. Then the symmetric and T-periodic solution xε (t) of
(1.1) from Theorem 7 is hyperbolic for any ε > 0 small.
Symmetric Solutions 909

6.3 A particular case

We consider the splitting (4.14) with

dim ker(I − A) = dim ker(I − A20 ) = 0 and dim ker(I + A) 6= 0 . (6.5)

So we do not have variable η in (4.15) and function f 1 in (4.16) as well. Moreover

we know that the only symmetric solution of (1.1) has the form

x (0, 0, ε, µ, t) = (y(0, 0, ε, µ, t), z(0, 0, ε, µ, t)) = (y(0, 0, ε, µ, t), 0) (6.6)

with y(0, ε, µ, −t) = −y(0, ε, µ, t). Next (4.16) implies

Dy f 2 (y, 0, µ, t) = − Dy f 2 (−y, 0, µ, −t)

Dz f 2 (y, 0, µ, t) = Dz f 2 (−y, 0, µ, −t) A0
(6.7)
A0 Dy f 3 (y, 0, µ, t) = Dy f 3 (−y, 0, µ, −t)
A0 Dz f 3 (y, 0, µ, t) = − Dz f 3 (−y, 0, µ, −t) A0 .

From (6.7), we derive

Dz f 2 (y, 0, µ, t) = Dz f 2 (−y, 0, µ, −t) A0 = Dz f 2 (y, 0, µ, t) A20

 
⇒ Dz f 2 (y, 0, µ, t) I − A20 = 0 ,
(6.8)
A20 Dy f 3 (y, 0, µ, t) = A0 Dy f 3 (−y, 0, µ, −t) = Dy f 3 (y, 0, µ, t)
 
⇒ I − A20 Dy f 3 (y, 0, µ, t) = 0 .

Since I − A20 : W → W is an isomorphism, (6.8) gives

Dz f 2 (y, 0, µ, t) = 0, Dy f 3 (y, 0, µ, t) = 0 .

Consequently, the variational equation of (1.1) along the symmetric solution (6.6)
has the form

Dy ẏ(0, 0, ε, µ, t) = εDy f 2 (y(0, 0, ε, µ, t), 0, µ, t) Dy y(0, 0, ε, µ, t)

Dy y(0, 0, ε, µ, 0) = I ,
Dz ẏ(0, 0, ε, µ, t) = εDy f 2 (y(0, 0, ε, µ, t), 0, µ, t) Dz y(0, 0, ε, µ, t)
Dz y(0, 0, ε, µ, 0) = 0 ,
(6.9)
Dy ż(0, 0, ε, µ, t) = εDz f 3 (y(0, 0, ε, µ, t), 0, µ(ε), t) Dy z(0, 0, ε, µ, t)
Dy z(0, 0, ε, µ, 0) = 0 ,
Dz ż(0, 0, ε, µ, t) = εDz f 3 (y(0, 0, ε, µ, t), 0, µ(ε), t) Dz z(0, 0, ε, µ, t)
Dz z(0, 0, ε, µ, 0) = I ,

which yields to Dz y(0, 0, ε, µ, t) = 0 and Dy z(0, 0, ε, µ, t) = 0 for any t ∈ R. More-

over, since by (6.7)

− Dy f 2 (y(0, 0, ε, µ, t), 0, µ, t) = Dy f 2 (y(0, 0, ε, µ, −t), 0, µ, −t)

910 N. Dilna – M. Fečkan

the first equation of (6.9) implies Dy y(0, 0, ε, µ, t) = Dy y(0, 0, ε, µ, −t) for any t ∈
R. By assuming a T-periodicity of y(0, 0, ε, µ, t) in t, then like in Section 5 we
arrive at Dy y(0, 0, ε, µ, T ) = I. Consequently, it holds
 
I 0
D(y,z) x (0, 0, ε, µ, T ) = .
0 Dz z(0, 0, ε, µ, T )
Hence we again cannot apply Theorem 13. On the other hand, we have
Z T  
0 0
D(y,z) f (0, 0, µ, t)dt = RT .
0 0 0 Dz f 3 (0, 0, µ, t)dt
Hence if Z T
f (0, 0, µ0 , t)dt = 0 for some µ0 ∈ R k
0
Z T  (6.10)
and ℜσ Dz f 3 (0, 0, µ0 , t)dt ∩ {0} = ∅ ,
0
then we can apply a local center manifold method to (1.1) near x ∼ 0 and µ ∼ µ0
to get the situation of Section 5. Note (4.16) gives
Z T Z T/2 Z T
f 2 (0, 0, µ, t)dt = 2 f 2 (0, 0, µ, t)dt, (I + A 0 ) f 3 (0, 0, µ, t)dt = 0 ,
0 0 0
RT RT
which implies f 3 (0, 0, µ, t)dt = 0. Consequently the equation 0 f (0, 0, µ0 , t)dt =
0
R T/2
0 is equivalent to 0 f 2 (0, 0, µ0 , t)dt = 0 (cf (4.17)).
Next we assume that A is unitary, i.e. k Ak = 1. This holds among others
when A p = I for some p ∈ N by taking a new scalar product on R n given by [9]
p
( x1 , x2 ) : = ∑ h A j x1 , A j x2 i .
j =1

Let r ∈ N. Now we apply a local center manifold method to (1.1) near x ∼ 0

and µ ∼ µ0 to get a local Cr -mapping Φ(y, ε, µ, t) which is T-periodic in t, y ∼ 0,
ε ∼ 0, µ ∼ µ0 , Φ ∈ W and satisfying
A0 Φ(y, ε, µ, t) = Φ(−y, ε, µ, −t) (6.11)

along with

ε f 3 (y, Φ(y, ε, µ, t), µ, t) = εDy Φ(y, ε, µ, t) f 2 (y, Φ(y, ε, µ, t), µ, t) + Dt Φ(y, ε, µ, t) .

(6.12)
Expanding
Φ(y, ε, µ, t) = Φ0 (y, µ) + O(ε) ,
and using (6.12) we derive

f¯3 (y, Φ0 (y, µ), µ) = Dy Φ0 (y, µ) f¯2 (y, Φ0 (y, µ), µ) , (6.13)
RT
where f¯i := 0 f i (y, z, t)dt, i = 1, 2. Note
Φ0 (0, µ0 ) = 0 and Dy Φ0 (0, µ0 ) = 0 . (6.14)
Symmetric Solutions 911

Of course, (6.13) means that Φ0 (y, µ) is a graph of a local center manifold of the
averaged equation of (1.1) given by ẋ = ε f¯( x, µ) for x ∼ 0 and µ ∼ µ0 . The
reduced ODE on the local center manifold is given by

ẏ = εg(y, ε, µ, t) := ε f 2 (y, Φ(y, ε, µ, t), µ, t) . (6.15)

Note (4.16) and (6.11) imply

g(−y, ε, µ, −t) = f 2 (−y, Φ(−y, ε, µ, −t), µ, −t) = f 2 (−y, A0 Φ(y, ε, µ, t), µ, −t)
= f 2 (y, Φ(y, ε, µ, t), µ, t) = g(y, ε, µ, t) .
Next we compute

Dy ḡ(y, 0, µ) = Dy f¯2 (y, Φ0 (y, µ), µ) + Dz f¯2 (y, Φ0 (y, µ), µ) Dy Φ0 (y, µ)

and
Dyy ḡ(y, 0, µ) = Dyy f¯2 (y, Φ0 (y, µ), µ) + 2Dyz f¯2 (y, Φ0 (y, µ), µ) Dy Φ0 (y, µ)


+ Dzz f¯2 (y, Φ0 (y, µ), µ) Dy Φ0 (y, µ), Dy Φ0 (y, µ)
+ Dy f¯2 (y, Φ0 (y, µ), µ) Dyy Φ0 (y, µ) .
Using (6.3) and (6.14), we obtain

Dy ḡ (0, 0, µ0 ) = 0 and Dyy ḡ (0, 0, µ0 ) = Dyy f¯2 (0, 0, µ0 ) . (6.16)

Similarly we derive (cf (6.4))

Z T/2 Z T/2
1¯
g(0, 0, µ0 , t)dt = f 2 (0, Φ0 (0, µ0 ) , µ0 , t)dt = f (0, 0, µ0 ) = 0 ,
0 0 2
Z T/2 Z T/2
Dµ g(0, 0, µ0 , t)dt = Dz f 2 (0, Φ0 (0, µ0 ), µ0 , t) dtDµ Φ0 (0, µ0 ) (6.17)
0 0
Z T/2
1
+ Dµ f 2 (0, Φ0 (0, µ0 ) , µ0 , t)dt = Dµ f¯2 (0, 0, µ0 ) .
0 2
Consequently, if k ≥ dim ker(I + A) then we can apply results of Sections 4.1 and
5 to this particular case (6.15).
Next by Section 4.1, when dim ker(I − A20 ) = 0 then f 3 (y, 0, µ) = 0, so the
reduced equation is now
ẏ = ε f 2 (y, 0, µ, t) (6.18)
and symmetric solutions lie in ker(I + A). A 3-dimensional example is given in
Section 8.2.4.
Finally, when dim ker(I + A) = 1 then we can apply Theorems 4 and 8 to
show that for any ε 6= 0 small, there is a surface Sε of codimension 1 splitting R k
near µ0 in two parts P1,ε and P2,ε such that: if µ ∈ P1,ε then (1.1) has no small peri-
odic solutions, if µ ∈ Sε then (1.1) has a unique small periodic solution which is in
addition symmetric and unstable, and if µ ∈ P2,ε then (1.1) has exactly two small
periodic solutions x1,ε (t) and x2,ε (t), which are hyperbolic and nonsymmetric but
satisfying x1,ε (t) = Ax2,ε (−t) and x2,ε (t) = Ax1,ε (−t). So xi,ε (t) = A2 xi,ε (t),
i = 1, 2, i.e. xi,ε (t) ∈ ker(I − A2 ) for any t ∈ R. Note ker(I + A) ⊂ ker(I − A2 ).
This is a saddle-node bifurcation with symmetries.
912 N. Dilna – M. Fečkan

7 Antisymmetric and periodic solutions

Assuming in addition (1.7), we can directly extend the above results to antiperi-
odic solutions (cf (1.8)). So we only state some results without proofs.
Theorem 16. The Cauchy problem (1.1) with

x (0) = θ ∈ ker(I + A) (7.1)

has a unique C∞ -smooth solution x (θ, ε, µ, t) which is also antisymmetric, and any anti-
symmetric solution x (t) of (1.1) satisfies (7.1).
We see that in order to study antisymmetric solutions, it is enough to replace
ker(I − A) with ker(I + A) in the arguments dealing for the symmetric case, and
so the projection I − S is replaced with an A-invariant projection I − Se : R n →
ker(I + A) in the above sections.

8 Applications
In this section, we present concrete weakly nonlinear ODE to illustrate our theory.
We separately consider two cases when either A = −I or A 6= −I. We start with
the first one.

8.1 The case A = −I

8.1.1 Scalar equations

Let us consider scalar equation (1.1) with a form

ẋ = ε (cos x + µ(1 + sin t)) , τ = −π, Ax = − x. (8.1)

It is easy to see that condition (1.5) is satisfied. Really, we verify

A f ( x, µ, t) = −(cos x + µ(1 + sin t))

= −(cos(− x ) + µ(1 + sin(−t + π ))) = − f ( Ax, µ, −t − τ ).

We have that ker(I − A) = {0}, so now η0 = 0. Then

Z π
H1 (µ) = H1 (0, µ) = (cos 0 + µ(1 + sin(s + π/2)) ds = π (1 + µ).
0

Since H1 (µ0 ) = 0 if and only if µ0 = −1 and H1′ (−1) = π 6= 0, we can apply The-
orem 4 to get a unique symmetric and 2π-periodic solution xε (t) = x (0, µ(ε), ε, t)
of (8.1) (only for µ = µ(ε)) with µ(0) = −1. So it holds

− x (0, ε, µ(ε), t) = x (0, ε, µ(ε), −t + π ), x (0, ε, µ(ε), t + 2π ) = x (0, ε, µ(ε), t),

which imply

x (0, ε, µ(ε), 3π/2 + t) = − x (0, ε, µ(ε), 3π/2 − t). (8.2)

Symmetric Solutions 913

Next, since Z 2π
Dx f (0, −1, s)ds = 0,
0
we cannot apply the usual first order averaging methods for establishing asymp-
totic properties of xε (t). So we need to study in more details the mapping
(cf Section 5 and Theorem 8)
Φε (η ) = x (η, ε, µ(ε), 5π/2) for η ∈ R.
Note Φε (0) = 0 and x (t) = x (η, ε, µ(ε), t) is the solution of the Cauchy problem
ẋ = ε (cos x + µ(ε)(1 + sin t)) ,
(8.3)
x (π/2) = η.

Consequently, we have Φε (η + 2π ) = Φε (η ) + 2π, and so Φε : S1 → S1 for the

unit circle. Moreover, Φε (η ) has the only fixed point η0 = 0 in S1 . Now we
compute Φ′ε (0) = Dη x (0, ε, µ(ε), 5π/2). By using (8.3) we get

Ḋη x (0, ε, µ(ε), t) = −ε sin ( x (0, ε, µ(ε), t)) Dη x (0, ε, µ(ε), 5π/2),
(8.4)
Dη x (0, ε, µ(ε), π/2) = 1.
Then using (8.2), we obtain
R 5π/2
−ε sin( x (0,ε,µ (ε),s))ds
Φ′ε (0) = e π/2 = 1.
Next, (8.3) also implies
Φε (η ) = η + 2πε(cos η − 1) + O(ε2 ) ,
which gives
Φ′′ε (0) = −4πε + O(ε2 ) < 0
for ε > 0 small. Summarizing we see [17, 19, 26] that 0 is a global saddle-node
of Φε : S1 → S1 for any ε > 0 small: it is attracting from the right and repelling
from the left. The orientation of the dynamics of Φε : S1 → S1 is reverse for ε < 0
small.

8.1.2 Planar equations

Example 1. First we consider the system

ẋ = ε (−( x + y + sin t) x + µ1 )
(8.5)
ẏ = ε (( x + y + sin t)y + µ2 )
Rπ
with µ = (µ1 , µ2 ) ∈ R2 . Now k = n = 2, T = 2π and 0 f (0, µ, t) dt = πµ. So
assumptions of (4.11) are satisfied. On the other hand, (8.5) has a trivial symmet-
ric solution x = 0, y = 0 for any ε and µ = 0. The uniqueness of µ(ε) implies
µ(ε) = 0, and hence (8.5) has no symmetric and periodic solutions for any µ 6= 0
and ε 6= 0 small. Next, we get
 
B( x, y)2 = − x2 − xy, y2 + xy .
914 N. Dilna – M. Fečkan

Since now
 
x y + x2 y1 x y2 + x2 y1
B (( x1 , y1 ), ( x2 , y2 ) = − x1 x2 − 1 2 , y1 y2 + 1 ,
2 2
x0 = (1, 0), λ0 = −1, [ x0 ]⊥ = [(0, 1)], 2QB ((1, 0), (0, y)) = y .

Clearly Theorem 12 can be applied. So the symmetric and periodic solution

( x, y) = 0 of (8.5) with µ = 0 is unstable for any ε 6= 0 small. In order to find
general periodic solutions of (8.5), we apply Theorems 1 and 2. So we solve the
averaged equation
−( x + y) x + µ1 = 0
(8.6)
( x + y ) y + µ2 = 0
which implies ( x + y)2 = µ1 − µ2 . So we need µ1 ≥ µ2 . If µ1 = µ2 6= 0, then (8.6)
has no solution as well. If µ1 > µ2 , then we derive
µ1 µ2
x1 = √ , y1 = − √ ,
µ1 − µ2 µ1 − µ2
µ1 µ2 (8.7)
x2 = − √ , y2 = √ .
µ1 − µ2 µ1 − µ2

Next, the characteristic polynomial of the linearization of (8.6) is as follows

− 2 ( x + y )2 + ( x − y ) λ + λ2 .

Since ( x1 + y1 )2 = ( x2 + y2 )2 = µ1 − µ2 > 0, we see that for any µ1 > µ2 both

(8.7) give rise to hyperbolic/unstable 2π-periodic solutions z1 (t) and z2 (t) of (8.5)
for ε 6= 0 located near (8.7), respectively. Moreover z2 (−t) = −z1 (t). Here z =
( x, y) ∈ R2 . If µ1 ≤ µ2 and µ 6= 0 then (8.5) has no 2π-periodic solutions for ε 6= 0
in any bounded domains.
Example 2. Now we modify the system (8.5) as follows

ẋ = ε (−( x + y + sin t)y + µ1 )

(8.8)
ẏ = ε (( x + y + sin t) x + µ2 )

with µ = (µ1 , µ2 ) ∈ R2 . We again derive µ(ε) = 0 and (8.8) has a symmetric and
periodic solution only if µ = 0 and it is a zero one. So we study

ẋ = −ε ( x + y + sin t) y
(8.9)
ẏ = ε ( x + y + sin t) x .

Clearly any solution of (8.9) satisfies x2 (t) + y2 (t) = x2 (0) + y2 (0). So the sym-
metric and periodic solution ( x, y) = 0 of (8.9) is uniformly stable for any ε 6= 0
small, but not asymptotically (cf Remark 1). In order to find general periodic
solutions of (8.8), we apply Theorems 1 and 2. So we solve the averaged equation

−( x + y)y + µ1 = 0
(8.10)
( x + y ) x + µ2 = 0
Symmetric Solutions 915

which implies ( x + y)2 = µ1 − µ2 . So we need µ1 − µ2 > 0. If µ1 − µ2 ≤ 0 and

µ 6= 0, then (8.10) has no solution. If µ1 − µ2 > 0, then we derive
µ2 µ
x1 = − √ , y1 = √ 1 ,
µ1 − µ2 µ1 − µ2
µ2 µ (8.11)
x2 = √ , y2 = − √ 1 .
µ1 − µ2 µ1 − µ2

Next, the characteristic polynomial of the linearization of (8.10) is as follows

2 ( x + y )2 + ( y − x ) λ + λ2 .
µ +µ
Since ( x1 + y1 )2 = ( x2 + y2 )2 = µ1 − µ2 > 0 and y1 − x1 = −(y2 − x2 ) = √µ1 −µ2 ,
1 2
we see that for any µ1 − µ2 > 0 and µ1 + µ2 6= 0 both (8.11) give rise to hyperbolic
2π-periodic solutions z1 (t) and z2 (t) of (8.5) for ε 6= 0 located near (8.11), respec-
tively. Moreover z2 (−t) = −z1 (t), and z1 (t) is asymptotically stable (a repeller)
and z2 (t) is a repeller (asymptotically stable), when µ1 + µ2 > (<) 0, respectively,
and ε > 0 small; it is opposite for ε < 0. If µ1 − µ2 ≤ 0 and µ 6= 0 then (8.5) has
no 2π-periodic solutions for ε 6= 0 in any bounded domains.
Now we proceed with the second possibility.

8.2 The case A 6= −I

8.2.1 Planar equations with an involution symmetry

Let us consider a planar differential equation

ẋ1 = ε ( f 1 ( x1 , x2 ) + µh1 (t))

(8.12)
ẋ2 = ε ( f 2 ( x1 , x2 ) + µh2 (t))

with C∞ -smooth functions f 1,2 , h1,2 , dim µ = k = 1 and with

 
0 −1
A= .
−1 0

Note A2 = I, so A is an involution. Then symmetry condition (1.2) implies

f 1 ( x1 , x2 ) = f 2 (− x2 , − x1 ), f 2 ( x1 , x2 ) = f 1 (− x2 , − x1 ),
(8.13)
h1 (t) = h2 (−t − τ ), h2 (t) = h1 (−t − τ ) .

Symmetry conditions (8.13) are satisfied, for instance, to the following polynomi-
als
m  
j p p j
f 1 ( x1 , x2 ) = a0 x1 + b0 x2 + ∑ a jp x1 x2 + b pj x1 x2 ,
j,p,j+ p>1
m   (8.14)
j p p j
f 2 ( x1 , x2 ) = −b0 x1 − a0 x2 + ∑ (−1) j+ p b pj x1 x2 + a jp x1 x2 ,
j,p,j+ p>1
h1 (t) = sin t, h2 (t) = − sin t ,
916 N. Dilna – M. Fečkan

and τ = 0. Since in general polynomials (8.14) are difficult to handle, we consider

the following particular case
 
2 2
ẋ1 = ε ax1 − x2 + x1 x2 − bx1 x2 + µ sin t ,
  (8.15)
ẋ2 = ε x1 − ax2 + bx12 x2 − x1 x22 − µ sin t ,

where a, b ∈ R are parameters. Now

ker (I − A) = [(1, −1)], ker (I + A) = [(1, 1)]

and hence
x1 + x2 x1 − x2
S ( x1 , x2 ) = (1, 1), (I − S)( x1 , x2 ) = (1, −1) .
2 2
We derive  
H1 (η, µ) = πη a + 1 − (b + 1)η 2 (8.16)

identifying ker(I − A) = [(1, −1)] ∼ R. Applying Theorem 7 we obtain the

following result.

Theorem 17. If a 6= −1, then (8.15) has a unique symmetric and 2π-periodic solution
z1 (t) located near (0, 0) for any ε 6= 0 small and µ 6= 0 fixed. If (a + 1)(b + 1) > 0
then
q (8.15)q has unique
 symmetric
q qand 2π-periodic
 solutions z2 (t), z3 (t) located near
a+1 a+1 a+1 a+1
b +1 , − b +1 and − b +1 , b +1 , respectively. Here z = ( x1 , x2 ) ∈ R2 .

We intend to find more 2π-periodic solutions of (8.15). For this reason, we

solve by Theorem 1 the averaged equation of (8.15) over [0, 2π ] given by

ax1 − x2 + x12 x2 − bx1 x22 = 0 ,

(8.17)
x1 − ax2 + bx12 x2 − x1 x22 = 0 ,

which gives
( x1 − x2 )(1 + a + (1 + b) x1 x2 ) = 0 ,
(8.18)
( x1 + x2 )(1 − a + (b − 1) x1 x2 ) = 0 .
For x1 6= ± x2 from (8.18) we derive ab = 1. Hence we suppose ab 6= 1. Then:
Either x1 = − x2 and (8.17) implies
 
x1 a + 1 − (1 + b) x12 = 0 .

If a 6= −1 and (a + 1)(b + 1) > 0, we obtain the following 3 solutions

r r
a+1 a+1
x1,1 = x2,1 = 0 ; x1,2 = − x2,2 = − ; x1,3 = − x2,3 = , (8.19)
b+1 b+1

which give symmetric and periodic solutions z1 (t), z2 (t) and z3 (t) from Theorem
17, respectively.
Symmetric Solutions 917

Or x1 = x2 6= 0 and (8.17) implies

 
x1 a − 1 + (1 − b) x12 = 0 .

If a 6= 1 and (a − 1)(b − 1) > 0, we obtain the following 2 solutions

r r
a−1 a−1
x1,4 = x2,4 = − ; x1,5 = x2,5 = , (8.20)
b−1 b−1
which give another periodic solutions z4 (t) and z5 (t), respectively, which are not
symmetric. But since A is an involution, from the proof of Lemma 1, we see that
z5 (t) = Az4 (−t).
Now we study the hyperbolicity of these periodic solutions by applying The-
orem 13. So we find eigenvalues of the matrix
 
a + 2x1 x2 − bx22 x12 − 1 − 2bx1 x2
1 + 2bx1 x2 − x22 bx12 − a − 2x1 x2

in points (8.19) and (8.20), which are as follows:

s s
p ( 1 − ab )( a + 1 ) (a − 1)(1 − ab)
± a2 − 1, ±2 , ±2 .
b+1 b−1

Summarizing we obtain the following result.

Theorem 18. For any ε 6= 0 small and µ 6= 0 fixed, the following holds:
• z1 (t) is hyperbolic for | a| > 1.
• z2 (t) and z3 (t) are hyperbolic for 1 > ab and a > −1, b > −1.
• z4 (t) and z5 (t) are hyperbolic for 1 > a and 1 > b.
We note that all periodic solutions z1 (t), · · · , z5 (t) cannot be simultaneously
hyperbolic.

8.2.2 Odd and planar equations with an involution symmetry

Now we consider modified planar ODEs from Section 8.2.1 which is in addition
odd of the form
 
ẋ1 = ε ax1 − x2 + x12 x2 − bx1 x22 + µx1 sin t
  (8.21)
2 2
ẋ2 = ε x1 − ax2 + bx1 x2 − x1 x2 + µx2 sin t .

So (8.21) satisfies (1.3) and (1.5) with A from Section 8.2.1 and T = 2π. First we
note that symmetric and periodic solutions are derived in the same way as above,
so we get these solutions e z1 ( t ), e
z2 (t) and e
z3 (t) located near (8.19), if a 6= −1 and
(a + 1)(b + 1) > 0. Note e z1 (t) = 0. Next to find antisymmetric and periodic
solutions, we take Se = I − S and we derive
 
He 1 (η, µ) = πη a − 1 − (b − 1)η 2
918 N. Dilna – M. Fečkan

q
identifying ker(I + A) = [(1, 1)] ∼ R. Simple roots of H e 1 (η, µ) are 0, ± a−1
b −1
provided (a − 1)(b − 1) > 0, which give antisymmetric and periodic solutions
z1 (t) = 0, e
e z4 (t) and e z5 (t) located near (8.20). Note Ae z4 ( t ) = − ez4 (−t) and
Aez4 ( t ) = e
z5 (−t), so e
z4 ( t ) = − e
z5 (t). To find the possible remaining periodic solu-
tions (non-symmetric and non-antisymmetric ones), we solve by Theorem 1 the
averaged equation of (8.21) over [0, 2π ] given by (8.17). But we know that there
are no more solutions of (8.17). Summarizing, we obtain the following result.

Theorem 19. If a 6= −1, then (8.21) has a unique symmetric and 2π-periodic solution
z1 (t) = 0 located near (0, 0) for any ε 6= 0 small. If (a + 1)(b + 1) >
e q0 then (8.21)
q has
a+1 a+1
unique symmetric and 2π-periodic solutions e z2 ( t ), e
z3 (t) located near b +1 , − b +1
 q q 
and − ba+ 1
+1 ,
a+1
b +1 , respectively. If ( a − 1)(b − 1) > 0 then (8.21) has unique an-
q q 
a−1 a−1
tisymmetric and 2π-periodic solutions e z4 ( t ), e
z5 (t) located near b −1 , b −1 and
 q q 
− ba− 1
−1 , −,
a−1
b −1 , respectively. There are no more 2π-periodic solutions. The state-
ment of Theorem 18 remains for this case.

8.2.3 Planar equations with a rotational symmetry

In this section, we consider planar ODEs of the form

ẋ1 = ε f 1 ( x1 , x2 , µ, t)
(8.22)
ẋ2 = ε f 2 ( x1 , x2 , µ, t) ,

where f 1 , f 2 are C∞ -smooth and periodic in T and µ ∈ R is a parameter. We

suppose that (8.22) is symmetric (cf. (1.5)) with respect to a rotation matrix
 
0 −1
A= .
1 0

Then it holds

f 1 ( x1 , x2 , µ, t) = − f 2 (− x2 , x1 , µ, −t), f 2 ( x1 , x2 , µ, t) = f 1 (− x2 , x1 , µ, −t) .
(8.23)
Then (8.23) gives

− f 1 ( x1 , x2 , µ, t) = f 1 (− x1 , − x2 , µ, t), − f 2 ( x1 , x2 , µ, t) = f 2 (− x1 , − x2 , µ, t) .

So (8.22) is also odd (cf (1.7)). Clearly ker(I − A2 ) = ker(I ± A) = {0}. So the
only symmetric and antisymmetric periodic solution of (8.22) is z1 (t) = 0 for
ε 6= 0 small (cf Theorem 6). To get more results, we pass to the following concrete
ODE  
2 2
ẋ1 = ε x1 − x2 + x1 x2 − bx1 x2 + µx1 sin t
  (8.24)
ẋ2 = ε − x1 − x2 + bx12 x2 + x1 x22 + µx2 sin t ,
Symmetric Solutions 919

where b, µ ∈ R are parameters. For finding further periodic solutions, we again

consider the averaged equation
x1 − x2 + x12 x2 − bx1 x22 = 0
(8.25)
− x1 − x2 + bx12 x2 + x1 x22 = 0 .
If x1 = 0 then x2 = 0, and x2 = 0 then x1 = 0. So we suppose x1 6= 0 and x2 6= 0.
Then we take x2 = ζ/x1 in (8.25) to derive
ζ (1 + ζ ) − (1 + bζ ) x22 = 0
ζ (bζ − 1) − (1 − ζ ) x22 = 0 ,
q
which implies either ζ + = b22+1 and then
r √ √ r √ √
2+2b2 + 2(b −1) b2 +1 2+2b2 − 2(b −1) b2 +1
x1+,+ = (1+ b )(1+ b2 )
, x2+,− = (1+ b )(1+ b2 )
r √ √ r √ √ (8.26)
2+2b2 + 2(b −1) b2 +1 2+2b2 − 2(b −1) b2 +1
x1−,+ = − (1+ b )(1+ b2 )
, x2−,− = − (1+ b )(1+ b2 )
q
2
or ζ − = − b2 +1
and then
r √ √ r √ √
2+2b2 − 2(b −1) b2 +1 2+2b2 + 2(b −1) b2 +1
x1+,− = (1+ b )(1+ b2 )
, x2−,+ =− (1+ b )(1+ b2 )
r √ √ r √ √ (8.27)
2 2 2+2b2 + 2(b −1) b2 +1
x1−,− = − 2+2b (−1+b2)((b1−+1b)2 ) b +1 , x2+,+ = (1+ b )(1+ b2 )
√ √ ±,±
Note 2 + 2b2 ± 2(b − 1) b2 + 1 > 0 for any b ∈ R \ {−1}, so x1,2 are defined
only for b > −1. Moreover, it holds and
±,− ±,+
x1,2 → ±∞, x1,2 →0
as b → −1+ . Now we study the hyperbolicity of these periodic solutions by
applying Theorem 13. So we find the characteristic polynomial of the matrix
 
1 + 2x1 x2 − bx22 x12 − 1 − 2bx1 x2
(8.28)
2bx1 x2 + x22 − 1 bx12 − 1 + 2x1 x2
at (8.26), which is √
2 2 2 (2 + b + b2 )
λ −λ √ +8,
(1 + b ) 1 + b2
and at (8.27), which is
√
2 2 2 (2 + b + b2 )
λ +λ √ +8,
(1 + b ) 1 + b2
Then the eigenvalues λ+ +
± of (8.28) at (8.26) satisfy ℜλ± > 0 and the eigenvalues
λ−± of (8.28) at (8.27) satisfy
√ ℜλ−± < 0 for any b > −1. Next, the eigenvalues of
0 0
(8.28) at x1 = x2 = 0 are ± 2.
Finally, we easily see that (8.25) has the only solution x1 = x2 = 0 for b = −1.
Summarizing, we arrive at the following result.
920 N. Dilna – M. Fečkan

Theorem 20. If b ≤ −1 then (8.24) has the only 2π-periodic solution z0 (t) = 0 for any
ε 6= 0 small which is hyperbolic. If b > −1 then (8.24) has in addition four 2π-periodic
solutions zi (t), i = 1, 2, 3, 4 for any ε 6= 0 small which are neither symmetric nor an-
tisymmetric. Moreover, z1 (t) and z2 (t) are asymptotically stable (repellers) while z3 (t)
and z4 (t) are repellers (asymptotically stable) for any small ε > 0 (ε < 0), respectively.
Note Az1 (t) = z4 (−t), Az2 (t) = z3 (−t), Az3 (t) = z1 (−t), Az4 (t) = z2 (−t),
and hence z2 (t) = −z1 (t), z4 (t) = −z3 (t). So the set of solutions {zi (t) | t ∈
R, i = 1, 2, 3, 4} is invariant by A.

8.2.4 3-dimensional systems

Now we present an example illustrating the case 6.3 given by the following 3-
dimensional ODE
   
2 2 2
ẏ = ε y + z1 + z2 cos t + µ (1 + cos t) ,
   
ż1 = ε z1 − z2 + z31 + z32 y2 + µz1 y sin t , (8.29)
   
ż2 = ε −z1 − z2 + z31 − z32 y2 − µz2 y sin t
 
−1 0 0  
  0 −1
with A = 0 0 −1 and A0 = . Then clearly (6.5) holds and
1 0
0 1 0
1 = k = dim ker(I + A). Next we derive f¯2 (0, 0, µ) = 2πµ andso we take µ0 =  0,
¯ ¯ ¯
since f 2 (0, 0, 0) = 0 and Dµ f 2 (0, 0, 0) = 2π 6= 0. Moreover σ D(y,z) f (0, 0, 0) =
√
{0, ±2 2π } (cf (6.10)) and dim ker(I − A20 ) = 0. Consequently, (6.18) has the
form  
ẏ = ε y2 + µ (1 + cos t) , (8.30)
and there is a C∞ -function µ(ε) defined for ε small with µ(0) = 0 such that for
any ε 6= 0 small, (8.29) possesses a (unique) symmetric and 2π-periodic solution
xε,µ (t) only for µ = µ(ε) and x0,µ(0) (t) = 0. On the other hand, (8.29) has a
solution x (t) = 0 for µ = 0, so the uniqueness implies xε,µ(ε) (t) = x0 (t) = 0 and
µ(ε) = µ0 = 0 as well. To study other (nonsymmetric) 2π-periodic solutions of
(8.29), we solve the averaged equation

y2 + µ = 0 ,
 
z1 − z2 + z1 + z2 y 2 = 0 ,
3 3
(8.31)
 
−z1 − z2 + z31 − z32 y2 = 0 .

We see that there are no solutions for µ > 0, while the only zero one for µ = 0
√
and this corresponds to the trivial one x0 (t) = 0. For µ < 0 we get y± = ± −µ
and  
3 3
z1 − z2 − z1 + z2 µ = 0 ,
 
−z1 − z2 − z31 − z32 µ = 0 ,
Symmetric Solutions 921

which implies z2 = −µz31 , and then z1 1 + µ4 z81 = 0, which implies z1 = z2 = 0.
µ √
Summarizing, for µ < 0, (8.31) has precisely two solutions x± = ±( −µ, 0, 0)
µ µ
which gives exactly two 2π-periodic solutions x± (t) = (y± (t), 0, 0) of (8.29) which
µ µ µ
are located near x± for ε 6= 0 small. Note x+ (t) = − x− (−t). So there are saddle-
node and symmetry breaking bifurcations of 2π-periodic√ solutions
√ of (8.29) as µ
¯ µ √
is crossing 0. Since σ( Dx f ( x± , µ)) = {±4π −µ, −2 2π, 2 2π }, periodic solu-
µ
tions x± (t) are hyperbolic. These results corresponds with arguments at the end
of Section 6.3.
Acknowledgement. We thank for useful referee’s comments and suggestions.

References
[1] A IZICOVICI , S., AND F E ČKAN , M.: Forced symmetric oscillations of evolution
equations, Nonlinear Analysis 64 (2006), 1621–1640.

[2] A IZICOVICI , S., AND PAVEL , N.H.: Anti-periodic solutions to a class of nonlin-
ear differential equations in Hilbert space, J. Funct. Anal. 99 (1991), 387–408.

[3] B ALDOM Á , I., AND F ONTICH , E.: Stable manifolds associated to fixed points
with linear part equal to identity, J. Differential Equations 197 (2004), 45–72.

[4] B ALDOM Á , I., F ONTICH , E., D E L A L LAVE , R., AND M ART ÍN , P.: The param-
eterization method for one-dimensional invariant manifolds of higher dimensional
parabolic fixed points, Discr. Cont. Dyn. Syst. 17 (2007), 835–865.

[5] B ANCROFT, S., H ALE , J.K., AND S WEET, D.: Alternative problems for nonlinear
functional equations, J. Differential Equations 4 (1968), 40-56.

[6] C HOSSAT, P., AND L AUTERBACH , R.: Methods in Equivariant Bifurcations and
Dynamical Systems, World Scientific Publishing Co., Singapore, 2000.

[7] C HOW, S.N., AND H ALE , J.K.: Methods of Bifurcation Theory, Springer-
Verlag, New York, 1982.

[8] D EMIDOWICH , B.P.: Lectures on the Mathematical Theory of Stability, Nauka,

Moscow, 1967 (in Russian).

[9] D ILNA , N., AND F E ČKAN , M.: On the uniqueness and stability of symmetric and
periodic solutions of weakly nonlinear ordinary differential equations, Misc. Math.
Notes 10 (2009), 11–40.

[10] D ILNA , N., AND F E ČKAN , M.: About the uniqueness and stability of symmetric
and periodic solutions of weakly nonlinear ordinary differential equations, Reports
Nat. Acad. Sci. Ukr. 5 (2009), 22–28 (in Russian).

[11] D ILNA , N., AND F E ČKAN , M.: Weakly non-linear and symmetric periodic sys-
tems at resonance, J. Nonlinear Studies 16 (2009), 149–170.

[12] FARKAS , M,: Periodic Motions, Springer-Verlag, Berlin, 1994.

922 N. Dilna – M. Fečkan

[13] F E ČKAN , M., M A , R., AND T HOMPSON , B.: Forced symmetric oscillations,
Bull. Belg. Math. Soc. 14 (2007), 73–85.

[14] F IEDLER , B.: Global Bifurcation of Periodic Solutions with Symmetry, Lecture
Notes in Mathematics 1309, Springer-Verlag, Berlin, 1988.

[15] H ALE , J.K.: Oscillations in Nonlinear Systems, McGraw-Hill, New York, 1963.

[16] H ALE , J.K.: Ordinary Differential Equations, Wiley-Interscience, New York,

1969.

[17] H IRSCH , M.W., AND S MALE , S.: Differential Equations, Dynamical Systems,
and Linear Algebra, Academic Press, New York, 1974.

[18] I OOSS , G., AND A DELMEYER , M.: Topics in Bifurcation Theory and Applica-
tions, World Scientific Publishing Comp., Singapore, 1991.

[19] I RWIN , M.C.: Smooth Dynamical Systems, Academic Press, London, 1980.

[20] K IELH ÖFER , H.: Bifurcation Theory: An Introduction with Applications to PDEs,
Springer-Verlag, New York, 2004.

[21] L AMB , J.S.W., AND R OBERTS , J.A.G.: Time-reversal symmetry in dynamical

systems: A survey, Physica D 112 (1998), 1–39.

[22] M ARSDEN , J.E.: Qualitative methods in bifurcation theory, Bulletin of the Amer.
Math. Soc. 84 (1978), 1125–1148.

[23] M AWHIN , J.: Familles de solutions périodiques dans les systèmes différentiels
faiblement non linéaires, Bull. Soc. R. Sci. Liège 36 (1967), 500-509.

[24] M AWHIN , J.: Degré topologique et solutions périodiques des systèmes différentiels
non linéaires, Bull. Soc. R. Sci. Liège 38 (1969), 308-398.

[25] M AWHIN , J.: Topological Degree Methods in Nonlinear Boundary Value Problems,
Reg. Conf. Ser. Math., vol. 40, Amer. Math. Soc., Providence, RI, 1979.

[26] M EDVE Ď , M.: Fundamentals of Dynamical Systems and Bifurcation Theory,

Adam Hilger, Ltd., Bristol, 1992.

[27] M ITROPOLOLSKII , Y U .A.: The Method of Averaging in Nonlinear Mechanics,

Naukova Dumka, Kiev, 1971 (in Russian).

[28] M UCOZ -A LMARAZ , F.J., F REIRE , E., G ALAN -V IOQUE , J., AND VANDER -
BAUWHEDE , A.: Continuation of normal doubly symmetric orbits in conservative
reversible systems, Celestial Mech. Dyn. Astr. 97 (2007), 17–47.

[29] M URDOCK , J., AND R OBINSON , C.: Qualitative dynamics from asymptotic ex-
pansions: local theory, J. Differential Equations 36 (1980), 425–441.

[30] R ONTO , A., AND R ONT Ó , M.: On certain symmetry properties of periodic solu-
tions, Nonlin. Oscillations 6 (2003), 82–107.
Symmetric Solutions 923

[31] R ONT Ó , M., AND S AMOILENKO , A. M.: Numerical-Analytic Methods in the

Theory of Boundary-Value Problems, World Scientific Publishing Co., Inc.,
River Edge, NJ, 2000.

[32] R OUCHE , N., AND M AWHIN , J.: Equations Différentielles Ordinaires, vol. 2,
Masson, Paris, 1973.

[33] S ANDRES , J.A., V ERHULST, F., AND M URDOCK , J.: Averaging Methods in
Nonlinear Dynamical Systems, Springer, New York, 2010.

[34] VANDERBAUWHEDE , A.: Alternative problems and symmetry, J. Math. Anal.

Appl. 62 (1978), 483-494.

[35] VANDERBAUWHEDE , A.: Alternative problems and invariant subspaces, J. Math.

Anal. Appl. 63 (1978), 1-8.

[36] VANDERBAUWHEDE , A.: Local Bifurcation and Symmetry, Research Notes in

Mathematics, vol. 75, Pitman, London, 1982.

Mathematical Institute of Slovak Academy of Sciences

Štefánikova 49, 814 73 Bratislava, Slovakia
and Institute of Mathematics of National Academy of Sciences of Ukraine
Tereshchenkivska 3 St., 01601 Kiev-4, Ukraine
email:nataliya.dilna@mat.savba.sk

Department of Mathematical Analysis and Numerical Mathematics

Comenius University, Mlynská dolina
842 48 Bratislava, Slovakia
and Mathematical Institute of Slovak Academy of Sciences
Štefánikova 49
814 73 Bratislava, Slovakia
email:Michal.Feckan@fmph.uniba.sk