Ds
Ds
Ds
Chapter 7
Dynamical Systems
151
152
ẋ = f (t, x) (7.2)
then
ψ (t, t0 , p) = φ (t − t0 , p) (7.4)
where φ (t, p) = ψ (t, 0, p), i.e., φ is the solution of the initial-value problem
when t0 = 0. This is a straightforward consequence of the uniqueness the-
orem for the initial-value problem as discussed in Chapter 6: either side of
the preceding equation is seen to satisfy the initial-value problem, and they
must therefore be the same (see problem 2 of Problem Set 6.4.1).
The solution map φ (t, p) of (7.3), sometimes called the flow, can be
thought of as a family of mappings, parametrized by the time t, defined
on the domain D0 of the initial-value vectors p in Rn , and taking values
in another domain Dt of Rn , namely, the image of the domain D0 under
the mapping φ (t, ·) . As long as the initial-value problem continues to be
satisfied at points of the latter domain, the identity
(i) φ(0, p) = p (ii) φ ∈ C[(a, b) × Ω] (iii) φ(t + α, p) = φ(t, φ(α, p)). (7.6)
Here (a, b) will always be the maximal interval unless otherwise stated; in
general it depends on the point p. The values a = −∞ and b = ∞ are allowed
for solutions defined on infinite intervals. Note that two orbits either coincide
or do not intersect. For if q ∈ γ(p) then q = φ(t0 , p) for some t0 ∈ (a, b).
Then φ(t, q) = φ(t, φ(t0 , p)) = φ(t + t0 , p). Thus γ(q) = γ(p). This property
of autonomous systems – that they either coincide or do not intersect – is
not shared by nonautonomous systems, as the following example shows.
Theorem 7.2.3 Let φ be a solution of (7.1) that intersects itself i.e. φ(t1 ) =
φ(t2 ) with t1 < t2 . Then φ is periodic with period t2 − t1 .
ω[γ(p)] = {x ∈ Rn : ∃ {tk }∞
k=0 with tk → ∞, such that φ(tk , p) → x as k → ∞}.
It is clear that the solution must exist at least on [0, ∞) for this definition to
make sense. For solutions that exist on −∞ < t ≤ 0 the negative semi-orbit
and the α-limit set are defined similarly.
156
Example 7.3.1 Suppose γ(p) is a periodic orbit and let q be any point of
it. Then q = φ(τ, p) for some τ ∈ [0, T ), where T is the least period, and
φ(tk , p) = q if tk = τ + kT . Hence every point of the orbit is a point of the
ω-limit set, i.e., the orbit equals the ω-limit set in this case: ω[γ(p)] = γ(p).
It’s also true that α[γ(p)] = γ(p).
The α- and ω-limit sets have a number of basic properties that will be
needed in the next section.
Proof: Let q lie in ω(γ(p)). We need to show that φ(t, q) ∈ ω[γ(p)] for all t ∈
R. We know there exists a sequence {tk } such that φ(tk , p) → q as tk → ∞.
Then certainly φ(t + tk , p) = φ(t, φ(tk , p)) → φ(t, q) as tk → ∞. Therefore
φ(t, q) ∈ ω[γ(p)] on its maximal interval of existence. Since ω[γ(p)] is closed
(by the preceding lemma) and bounded (by assumption) we infer that the
latter is (−∞, ∞), as follows. If the left-hand endpoint of the maximal
interval (a, say) is finite, then as t → a+, φ(t, q) would have to approach a
boundary point of Ω, by Theorem 6.1.5. This is not possible because this
limit is confined to ω(γ(p)), which cannot intersect the boundary of the
domain Ω.
The following theorem summarizes the properties of the ω-limit set. We
use the notations
d[x, y] = kx − yk (7.8)
to represent the distance between the points x and y in terms of the norm
k · k. Similarly, if A is a set and x a point,
d(x, A) = inf d(x, y)
y∈A
157
with a similar definition for the distance d(A, B) between two sets A and B.
Proof: The sequence {φ(k, p)} is bounded, hence contains a convergent sub-
sequence. This shows that ω[γ(p)] is non-empty.
Since γ(p) is bounded, so is the set of its limit points, so ω(γ(p)) is
bounded. We showed earlier that it is closed. This implies the compactness.
Since the closure of γ(p) lies in Ω, so too does ω[γ(p)] and, by the pre-
vious lemma, it is invariant. It remains to prove the connectedness and the
condition (7.9). We consider the latter first.
Suppose the condition (7.9) fails. Then there exists > 0 and {tk } → ∞
such that d[φ(tk , p), ω(γ(p))] > . Now the boundedness of the orbit guar-
antees the existence of a convergent subsequence of {φ(tk , p)}; continue to
denote this subsequence {φ(tk , p)} and the corresponding subsequence of t-
values {tk }. Let q = limk→∞ φ(tk , p). Then q ∈ ω(γ(p)), but d[q, ω(γ(p))] ≥
. This is obviously a contradiction.
It remains to show that ω is connected. We use the two following easily
verified assertions. One of these is that the distance d(x, A) from a point x to
a set A is a continuous function of x. A second refers to the description of a
set which is not connected. The usual description is that ω is not connected
if there exist open sets A, B with the following properties: (i)A ∩ B = φ, (ii)
A ∪ B ⊃ ω and (iii) A ∩ ω 6= φ and B ∩ ω 6= φ. The assertion is that if ω is
closed and these properties hold for open sets A and B, they hold also for
closed sets A0 and B 0 . This is easily seen by checking the properties for the
closed sets A0 = A ∩ ω and B 0 = B ∩ ω.
We now consider the connectedness of ω. Suppose it is not connected;
then there exist closed sets A, B with the properties described above. Since
A, B are disjoint closed sets, they are some finite distance apart. Denote
d[A, B] ≡ δ > 0. Since there is some point of ω in A, ∃t1 such that d[φ(t1 ), A] <
δ/4. By the same reasoning, there is some time t2 > t1 such that d[φ(t2 ), B] <
δ/4. Again, there is some t3 > t2 such that d[φ(t3 ), A] < δ/4. Continuing in
this manner we obtain a sequence {tk } such that alternatively for odd and
even indices, the orbital point φ(tk ) lies within a distance δ/4 of A or of B;
we can clearly arrange this sequence so that tk → ∞. Note that
The ω-limit set and α-limit set are two of a number of invariant sets that
are encountered in characterizing the qualitative behavior of orbits. Here is
another.
Definition 7.3.1 A point p is nonwandering for the system (7.1) if, given
any neighborhood U of p and any time T > 0, φ(t, x) ∈ U for some x ∈ U
and some time t ≥ T . The set of such points in Ω is called the nonwandering
set.
Example 7.3.2 Let Ω be bounded and invariant under the system (7.1),
and suppose that system generates a mapping preserving Lebesgue mea-
sure3 . For any p ∈ Ω choose a neighborhood U and a time T as in the
definition. Consider the family of mappings {φ(T, U ), φ(2T, U ), . . . , }. Each
of these has the same, positive measure. If they were all disjoint, the measure
of Ω would not be finite, but we have assumed that it is in assuming that
Ω is bounded. Therefore two of these have a non-empty intersection, say
φ(mT, U ) and φ(nT, U ), where m < n. This implies a non-empty intersec-
tion of U and φ ((n − m) T, U ). Therefore for measure-preserving mappings,
every point is nonwandering.
dx
= f (x), x ∈ Ω ∈ R2 ,
dt
and suppose
divf = ∂f1 /∂x1 + ∂f2 /∂x2
has one sign in Ω. Show that this system can have no periodic orbits
other than equilibrium points.
160
This is due to the limited possibilities for orbits in the plane. A closed orbit,
representing a periodic solution, is a Jordan curve, i.e. the topological image
of a circle. The separation property of such a curve is based on the Jordan
curve theorem, which states that if J is a Jordan curve in R2 , then the
complement of J is the union of two disjoint open sets Gi and Ge , each
of which has J as a boundary. Gi is bounded and called the interior of
J, whereas Ge is unbounded and called the exterior of J. In the present
section we consider the case n = 2, taking for granted the Jordan curve
theorem. The resulting theory is usually referred to as Poincarè-Bendixson
theory. The principal result is Theorem 7.4.1 below. The proof requires
some development and is completed below following a series of lemmas.
We need the notion of a transversal to orbits of the system (7.12):
(a) (b) S
q
p
p q
S
γ γ
Bendixson pocket Bendixson pocket
Figure 7.1: Figure (a) indicates an orbit γ intersecting a transversal S at the point
p and returning to S at q, forming a Bendixson pocket. Here γ spirals outward along
S. Figure (b) is the same except that the orbit spirals inward along S.
Proof: Suppose first that γ is not periodic. Let Σ be the simple closed
curve consisting of γ between y0 and y1 and the segment of S0 that joins
162
y0 and y1 (see Figure 7.1 (b) to fix ideas). The orbit enters the Bendixson
pocket at y1 in the same direction as at y0 . The orbit can never leave
the Bendixson pocket: leaving by crossing γ would imply periodicity, and
leaving by crossing S0 is not possible because the vector field is in the wrong
direction along S0 . Hence the next crossing must occur beyond y1 , i.e. y1 lies
between y0 and y2 . Repeating this reasoning shows that yk+1 lies between
yk and yk+2 .
Next suppose that γ is periodic, with least period T > 0. We express the
solution as φ(t, y0 ) so that the transversal S is constructed at y0 = φ(0, y0 )
(for example, y0 = p in Figure 7.1 (b)). Any other point on the orbit is
achieved at a unique time t in the interval [0, T ). Thus if the orbit crosses
a second point y1 on S (y1 = q in Figure 7.1 (b)) it does so at a time
t1 < T . The orbit cannot return to y0 across S so must cross γ at some
point y2 = φ(t2 , y0 ), where t1 < t2 < T . But since y2 also precedes y1 we
must have y2 = φ(τ2 , y0 ), where τ2 < t2 . But then γ is periodic with period
t2 − τ2 , a positive number less than T . This contradicts the assumption that
T is the least period.
Remark: If the orbit is traversed as in Figure 7.1(a), then the orbit leaves the
Bendixson pocket instead of entering it, and cannot return. The conclusion
is the same as that drawn above.
the transveral at some later time t2 . Let p2 = φ(t2 , p1 ) and assume that
p2 6= p1 and therefore, by Lemma 7.4.2, that γ is not periodic. Since p1 is
a limit point of γ, the orbit intersects S infinitely often. Let p3 = φ(t3 , y1 )
be the next intersection of the orbit with S. By Lemma 7.4.1 p3 must occur
“further along” S from p1 than p2 . Construct the Bendixson pocket based
on p2 and p3 . Now the orbit γ cannot approach p1 for t > t3 . However, this
is clearly a contradiction since p1 ∈ ω(γ).
Lemma 7.4.4 A transversal to γ cannot meet ω(γ) in more than one point.
Proof: Assume γ is not periodic, since otherwise the claim holds trivially.
ω(γ) is not empty so contains a regular point, and therefore a non con-
stant orbit γ0 . We wish to prove that γ0 is periodic.
γ0 is bounded and possesses a bounded omega-limit set ω(γ0 ). Let p0 ∈
ω(γ0 ). p0 is a regular point (because ω(γ) contains no equilibrium points).
Construct a transversal S0 to γ0 at p0 (Lemma 7.4.2). Since p0 ∈ ω(γ0 ), γ0
meets S0 repeatedly. If γ0 is not periodic, it meets S0 at distinct points.
Recall that ω(γ) cannot meet the transversal S0 at more than one point
(Lemma 7.4.4) and therefore meets S0 only at p0 . Thus γ0 , which lies in
ω(γ) meets S only at p0 and is therefore periodic, by Lemma 7.4.3.
This shows that ω(γ) contains a periodic orbit and therefore (Lemma
7.4.5) is a periodic orbit.
Orbits can of course be unbounded, but for bounded orbits the long-
time behavior is, as remarked at the beginning of this section, “severely
circumscribed,” as the following theorem shows:
Proof: The limit set ω(γ) may consist of regular points and of equilibrium
points. We first consider the cases when one or the other of these is missing.
3. Finally suppose that ω(γ) contains both regular points and a positive
number of equilibrium points. It cannot be a periodic orbit because
of the latter assumption. Since it contains a regular point, it also
contains a non-constant orbit γ0 . Its limit set ω(γ0 ) is not empty. If
ω(γ0 ) contained a regular point, we could conclude (as in the Poincaré-
Bendixson theorem) that γ0 is periodic, a contradiction. Therefore
ω(γ0 ) contains only equilibrium points and therefore only one equilib-
rium point x+ (since they are finite in number). Since the orbit γ0
approaches ω(γ0 ), γ0 approaches x+ . .
The orbit γ0 above, which approaches x+ as t → +∞, may also approach
x− as t → −∞. The orbit γ0 is called a connecting orbit. The equilibrium
points x+ and x− may be different, or they may be the same. In the following
example, they are different.
Example 7.4.1 Consider the system
ẋ1 = x2 + x21 − Rx1 x2 − 1 + 2x21 ,
ẋ2 = −2 (1 + x2 ) x1 , (7.13)
where R is a parameter. The curves x2 = −1 and x2 = 1 − 2x21 are invariant
curves for this system (see Figure 7.2). The compact set K consisting of the
finite region bounded by these curves together with its boundary contains
three equilibrium points, at (0, 0), (1, −1) and (−1, −1). Orbits starting in
the interior of this region spiral outward toward its boundary, which is the
ω-limit set for any such orbit. This ω-limit set contains two equilibrium
points together with the two orbits connecting them, illustrating the third
possibility enumerated in Theorem 7.4.2.
0.8
0.6
0.4
0.2
0
x2
-0.2
-0.4
-0.6
-0.8
-1
-1.2
-1 -0.5 0 0.5 1
x1
Figure 7.2: The compact region K described in Example 7.4.1 is shown, together
with a typical orbit. The three equilibrium points are denoted by dots. The value
of the parameter has been taken to be R = 0.2. Any such orbit spirals outward,
gradually becoming indistinguishable from the boundary, which is its ω-limit set.
167
An isolated periodic orbit like that of this theorem is called a limit cycle
if all sufficiently nearby orbits are attracted to it as t → +∞ or as t → −∞
(cf Problem 4 of Exercises 7.4.1) . However, the theorem allows for the
’peculiar’ possibility that orbits outside Γ tend to Γ as t → +∞ whereas
those inside tend away from Γ in this limit (or tend toward Γ as t → −∞)
(cf. Problem 5 of Exercises 7.4.1). It will be a consequence of the general
theory of Chapter 9 that this peculiar possibility is excluded if the vector
field is assumed to be C 1 .
168
Proposition 7.4.1 Suppose that the system (7.14) has a stable equilibrium
point at the origin and that the domain Ω contains the origin and no other
equilibrium point. Suppose there is a direction t̂ such that
t̂ · ∇H > 0 (7.15)
0.6
0.4
0.2 S
0 A
p
-0.2
-0.4
-0.6
-1.5 -1 -0.5 0 0.5
q
Figure 7.3: A phase-space diagram for the system with the Hamiltonian of Ex-
ample 7.4.2. The region marked A consists of the stable equilibrium point at the
origin together with periodic orbits surrounding it. All other orbits either lie on
the separatrix S (and tend toward (−1, 0) as t → ±∞) or are unbounded.
The region of periodic orbits need not be small. Given a stable equilib-
rium point, there is always an invariant domain containing it. If it can be
ascertained that it satisfies the conditions for Theorem 7.4.1 to hold, we can
infer that all orbits are periodic in that domain.
2. For Example 7.4.1 of the text, verify, using the criterion (7.16) above,
that the curves x2 = −1 and x2 = 1 − 2x21 are indeed invariant.
Identify any equilibrium points and periodic orbits and, in the case
of the latter, determine whether nearby orbits spiral toward or away
from the periodic orbit.
Hint: rewrite this system in terms of polar coordinates in the plane.
[2] P.F. Byrd and M. D. Friedman. Handbook of elliptic integrals for engi-
neers and physicists. Springer, Berlin, 1954.
315
316