Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Ds

Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

150

Chapter 7

Dynamical Systems

Mathematical models of scientific systems often lead to differential equations


in which the independent variable is time. Such systems arise in astronomy,
biology, chemistry, economics, engineering, physics and other disciplines. It
is common to speak of a system of differential equations
ẋ = f (x) (7.1)
as governing a dynamical system (or of generating, or of being, a dynamical
system). We are more precise about this below (conditions 7.6).
In the present chapter we introduce some of the principal ideas of dynamical-
systems theory, and illustrate them in the restricted but important context
of planar systems, i.e., systems in R2 . A norm k · k will appear at vari-
ous points below without explanation; we assume an appropriate norm has
been chosen (see the discussion in §6.2). There is an emphasis on nonlinear
systems,1 and on:
1. the qualitative behavior of the system viewed on a long time interval.
Put differently, the long-time dynamics, or the asymptotic behavior as
t → +∞, is often a major issue.
2. behavior of solutions as the initial data are changed. In considering
the system (7.1) of differential equations together with initial data
x (t0 ) = a, the solution is regarded not only as a function of t but also
as a function of a: x = φ (t, a) .
3. families of differential equations. The members of the family are often
distinguished from one another by the value of a parameter, µ (say).
1
However, as we shall see in Chapters 8 and 9, our previous emphasis on linear systems
will prove to be of great importance.

151
152

The right-hand side of equation (7.1) is then written as f (x, µ) . The


solutions will then depend also on µ and will sometimes be expressed
as x = φ (t, a, µ) .

These features mimic those of problems arising in applications, which fre-


quently depend on parameters, and which need to be considered for a variety
of initial data.

The one-dimensional equation ẋ = µx with initial data x (0) = a has


the solution x = a exp (µt) . Its behavior for t → +∞ depends dramat-
ically on the sign of the parameter µ.

The one-dimensional equation ẋ = µx2 with initial data x (0) = a has


the solution x = a/ (1 − aµt) . It’s behavior depends dramatically on
the parameters and initial data: if the product aµ < 0, the solution
decays to zero as t → ∞; if aµ > 0, it becomes unbounded in a finite
time.

7.1 Autonomous Equations


The equations of the examples, or the system (7.1), are referred to as au-
tonomous because the right-hand sides do not depend explicitly on the time.
This terminology originates in mechanical examples in which an explicit time
dependence reflects a mechanical forcing imposed upon the system by some
external agent: when there is no such forcing the system is “self-governing,”
or autonomous. The more general, non-autonomous, system

ẋ = f (t, x) (7.2)

of n equations may be converted to an autonomous system of n+1 equations


by adjoining the equation ẋn+1 = 1 and replacing t by xn+1 in the remaining
n equations: this equivalent autonomous system is sometimes referred to as
the suspended system. It may therefore appear that there is no difference
between autonomous and non-autonomous systems, but this is not entirely
true (see Problem 2 below).
In this chapter and in Chapters 8 and 9 we consider autonomous systems
unless the opposite is explicitly stated.
A distinctive feature of autonomous systems is the arbitrariness of the
origin of the time variable. If x = ψ (t, t0 , p) represents the solution of the
initial-value problem
ẋ = f (x) , x (t0 ) = p, (7.3)
153

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

φ (t + α, p) = φ (t, φ (α, p)) , (7.5)


holds, again as a direct consequence of the uniqueness theorem.
A formal definition of dynamical system that is sometimes used in ab-
stract studies is that of a family of maps φ(t, ·) : Ω → Ω, defined on a domain
Ω parametrized by a variable t on an interval (a, b) and satisfying, for each
p ∈ Ω the following three conditions:

(i) φ(0, p) = p (ii) φ ∈ C[(a, b) × Ω] (iii) φ(t + α, p) = φ(t, φ(α, p)). (7.6)

If in item (ii) the C is replaced by C k with k ≥ 1, the dynamical system is


said to be differentiable. These studies can be viewed as abstractions of flows
of autonomous differential equations. When the mapping φ is generated by
an initial-value problem (7.3), it is a homeomorphism, i.e., a continuous map
with a continuous inverse. If the vector field f is C k , it is a diffeomorphism2 .
There are also Discrete dynamical systems for which t takes on discrete
values, but we shall consider continuous dynamical systems generated by
autonomous systems like equation (7.1) – for which t takes on values in an
interval of the real axis – unless otherwise specified.
Suppose for each p ∈ R0 , a subdomain of Ω, solutions exist on a common
interval (a, b), then the family of maps φ(t, ·) generates a homeomorphism
from R0 to its image Rt under the map for any t ∈ (a, b).
The orbit γ(p) through a point p is defined as the set

γ(p) = {x ∈ Rn : x = φ(t, p) for some t ∈ (a, b)}. (7.7)


2
A diffeomorphism is differentiable map that possesses a differentiable inverse.
154

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.

Example 7.1.1 The nonautonomous system

ẋ = −x + (cos t − sin t), ẏ = y + (cos t + sin t)

has solutions x = ae−t +  cos t , y = bet +  sin t. If we choose a = b = 0,


the orbit is a circle of radius . However, if we take instead initial data
(x(0), y(0)) = (0, /2) lying inside the circle, then x = (cos t − e−t ) and
y = ((1/2)et + sin t), which leaves the disk of radius  and hence intersects
the circle at some t > 0. 

Of course, if distinct orbits of nonautonomous systems cross, they must


cross at different times, to be consistent with the general uniqueness theo-
rem.

7.2 Constant and Periodic Solutions


The simplest kinds of solutions are equilibrium points. These are points
where f (p) = 0. The solution map φ(t, p) = p for all t, i.e. p is a fixed point
of the map for each t. The orbit γ(p) is the set whose sole member is p.

Theorem 7.2.1 If p is an equilibrium point in Ω and φ(t, q) → p as t → t0 ,


then either t0 = ±∞ or q = p.

Proof: If t0 is finite then, since p ∈ Ω, the solution φ can be continued to t0 ,


and φ(t0 , q) = p by continuity. This means that the orbit through q is the
orbit through p, and since p is an equilibrium point, coincides with p. 
Any point that is not an equilibrium point, i.e., any point p where f (p) 6=
0, will be referred to as a regular point.
The next simplest kind of solution is periodic: φ(t + T, p) = φ(t, p) for
some T and all t ∈ R. We may assume that T > 0.

Theorem 7.2.2 A nonconstant periodic solution has a least period T .


155

Proof: Suppose not; then there is a sequence {Tn } of periods, with Tn → 0


as n → ∞. For arbitrary t we have
φ(t + Tn , p) − φ(t, p)
φ̇(t, p) = lim = 0.
n→∞ Tn
But this would imply that φ(t, p) is constant, which is a contradiction. 
A nonconstant periodic solution has an orbit which is a simple closed
curve: the solution map φ(t, p) for fixed p maps the interval [0, T ] to its
image in a one-one fashion if 0 and T are identified. Here T is the least
period. A criterion for a solution to be periodic is given by the following
theorem.

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 .

Proof: With T = t2 − t1 we have φ(t + T, p) = φ(t + t2 − t1 , p) = φ(t −


t1 , φ(t2 , p)) = φ(t − t1 , φ(t1 , p)) = φ(t, φ(0, p)) = φ(t, p). 
In this theorem, T = t2 − t1 need not be the least period.

7.3 Invariant Sets


A subset S of the domain Ω is an invariant set for the system (7.1) if the
orbit through a point of S remains in S for all t ∈ R. If the orbit remains in
S for t > 0, then S will be said to be positively invariant. Related definitions
of sets that are negatively invariant, or locally invariant, can easily be given.
If p is an equilibrium point of the system (7.1) then the set consisting of p
alone is an invariant set for that system. Likewise, if γ(p) is a periodic orbit,
it too is an invariant set. Invariant sets that are of particular importance
in the remainder of the present chapter are the so-called ω− and α− limit
sets, to which we now turn.
Suppose the solution φ(t, p) exists for 0 ≤ t < ∞. The positive semi-orbit
γ(p) is defined by equation (7.7) above when (a, b) is replaced by [0, ∞). The
ω−limit set for such an orbit γ(p) is defined as follows

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.

Lemma 7.3.1 The ω−limit set is closed.

Proof: Pick a sequence qn ∈ ω with qn → q. By the definition of the ω-limit


set, we know there exists {tn,k }, tn,k → ∞ with φ(tn,k , p) → qn as k → ∞.
For each n, pick K(n) so that kφ(tn,k , p) − qn k < 1/n ∀ k > K(n). Then,
given  > 0, pick N so that kq − qn k < /2 if n > N . This implies that
kφ(tn,K(n) , p) − qk ≤ kφ(tn,K(n) , p) − qn k + kq − qn k
< 1/n + /2
<  if n > max{N, 2/}.
Thus q is in ω. 

Lemma 7.3.2 Suppose ω(γ(p)) is bounded and lies in Ω. Then ω is an


invariant set.

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.

Theorem 7.3.1 Let γ(p) be a bounded, positive semi-orbit whose closure


lies in Ω. Then ω[γ(p)] is non-empty, compact, invariant, connected and

d[φ(t, p), ω(γ(p))] → 0 as t → ∞. (7.9)

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

δ = d[A, B] ≤ d[φ(t1 ), A] + d[φ(t1 ), B] ⇒ d[φ(t1 ), B] ≥ 3δ/4.


158

Since the distance function is continuous, we can find some τ1 in [t2 , t1 ]


such that d[φ(τ1 ), B] = δ/2. It follows, as above, that d[φ(τ1 ), A] ≥ δ/2.
We repeat this process for τk ∈ [t2k+1 , t2k ] ∀ k ≥ 0. This process generates
another sequence {τk } with τk → ∞ and

d[φ(τk ), A] ≥ δ/2, d[φ(τk ), B] = δ/2.

Since {φ (τk )} is bounded, there exists a convergent subsequence; continue


to denote this subsequence with the same notation. Let q = limk→∞ φ(τk ).
Then q ∈ ω, but q is bounded away from both A and B. This contradiction
completes the proof. 

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.

Important invariant sets may be specific to a problem, as in the next


example.

Example 7.3.3 In ecological modeling the Lotka-Volterra system is often


introduced to approximate the relation between a predator and its prey in
some fixed locale. Here the variable x is proportional to the number of
3
A Hamiltonian system will do this.
159

prey (rabbits, say) in this locale whereas y is proportional to the number of


predators (foxes, say). The equations are

ẋ = x(a − by), ẏ = −y(c − dx), (7.10)

where a, b, c, d are positive constants. Here it is important for the ecological


interpretation that x, y be positive. In other words, if initially this is so,
then this property is preserved by the equations. This is in fact the case
(see Problem 7 below). 

PROBLEM SET 7.3.1

1. Prove the identity (7.5)

2. Show that the suspended system of dimension n + 1 obtained from the


time-dependent system (7.2) of dimension n cannot have an equilib-
rium point.

3. Find the periodic solutions of the system

ẋ1 = −x2 + x1 f (r), ẋ2 = x1 + x2 f (r)

where r2 = x21 + x22 and f (r) = −r(1 − r2 )(4 − r2 ).

4. Consider the nonautonomous, periodic system

ẋ = f (x, t), f (x, t + T ) = f (x, t). (7.11)

Let x(t) be a solution such that, at some time t1 , x(t1 ) = x(t1 + T ).


Show that this solution is periodic with period T .

5. Consider the planar autonomous system

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

6. Consider the gradient system


dx
= ∇φ, x ∈ Ω ∈ Rn ,
dt
where φ(x) is a smooth, single-valued function. Draw the same con-
clusion as in the preceding problem.

7. Consider the system

ẋ = xf (x, y), ẏ = yg(x, y)

where f, g are arbitrary, smooth functions defined in R2 . Show that


the lines x = 0 and y = 0 are invariant curves for this system. Infer
that each of the four quadrants of the xy-plane is an invariant region
for this system.

8. Show that the nonwandering set is closed and positively invariant.

7.4 Poincarè-Bendixson Theory


The behavior of solutions of the system (7.1) is severely circumscribed when
n = 2 where it takes the form

ẋ1 = f1 (x1 , x2 ), ẋ2 = f2 (x1 , x2 ). (7.12)

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):

Definition 7.4.1 A finite closed segment S of a straight line in R2 is called


a transversal with respect to f if f (x) 6= 0 for every point x ∈ S and if the
direction determined by f at every point of S is different from that of S.
161

(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.

Figure 7.1 provides an illustration.


We infer that every orbit that meets a transversal must cross it, and that
all such orbits must cross it in the same direction. Otherwise f would have
to be tangent to S somewhere along it, i.e., determine a direction the same
as that of S. It is clear that a transversal can be constructed at any regular
point of f . We recall that a regular point is one where f 6= 0.
We can construct a flow box B on a transversal S consisting of orbits
originating on S and continued throughout a small time interval [−σ, σ]: σ is
small enough that the time interval required for a point of S to return to S
along an orbit of (7.12) exceeds σ. If q ∈ B, then under the flow φ of (7.12)
there is a unique point q ∗ = φ(t∗ , q) on S at a unique time t∗ ∈ [−σ, σ].
We’ll return to a more formal justification of this in Chapter 9; here we rely
on our intuition in the plane and view it as obvious.

Lemma 7.4.1 If γ is not periodic and meets the transversal S0 to y0 at


distinct points yk = φ(tk , x) at times t1 < t2 < . . ., then the order of the
points yk on S0 is the same as the order of the times tk , i.e. yk lies between
yk−1 and yk+1 . If γ is periodic, it can meet S at most at one point.

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.

Lemma 7.4.2 Let p be a regular point of ω(γ) lying in Ω. A transversal S


at p is a transversal to γ. If γ is not periodic, then it must intersect S at
infinitely many distinct points. If these intersections with S are denoted by
pi , then they all lie on the same side of p and the sequence {pi } converges
monotonically to p.

Proof: One can construct a transversal to ω(γ) at p since p is a regular


point. Construct a flow box B on S including p. Since p is a limit point
of γ the latter must enter B and therefore intersect S, showing that S is a
transversal also to γ. These intersections must be distinct since otherwise γ
would be periodic. Since they are all in the same order according to Lemma
7.4.1 they must lie on the same side of p. Since the latter is in ω, one can,
by constructing successively smaller flow boxes containing p, infer that they
must tend toward p. 

Lemma 7.4.3 If γ and ω(γ) have a regular point in common, then γ is a


periodic orbit.

Proof: Let p1 ∈ γ ∩ ω(γ). Construct a transversal S and a flow box at


p1 . Since p1 ∈ ω(γ), the orbit must return to the flow box and hence to
163

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: Suppose S is a transversal to γ, and let p1 6= p2 be two points of


ω(γ) lying in S. Then the orbit γ also meets S at more than one point, and
is therefore not periodic (Lemma 7.4.1). Then γ meets S at infinitely many
points {yk } at times t1 < t2 < · · · . However, since there are two limit points
of γ on S, it is impossible for the points {yk } to be in the corresponding
order on S, contradicting Lemma 7.4.1.

Lemma 7.4.5 Suppose γ is bounded and ω(γ) lies in Ω and contains a


non-constant periodic orbit γ0 . Then ω(γ) = γ0 .

Proof: Let d = ω(γ)\γ0 and suppose this set is non-empty. Since ω is


connected, there exists a limit point of d in γ0 (see Remark below): qk ∈
d, qk → q0 , q0 ∈ γ0 . Now, q0 ∈ ω, and is a regular point since otherwise the
periodic orbit γ0 would be constant. By Lemma 7.4.2, one can construct a
transversal S to γ at q0 . Construct a flow box on S; there is a point q ∗ ∈ d in
this flow box since q0 is a limit point of d. Hence p∗ = φ(t∗ , q ∗ ) ∈ S for some
small t∗ . However, p∗ ∈ ω since the limit set is invariant and consequently
S contains two points p∗ , q0 of ω. These points are distinct because p∗ 6∈ γ0
since otherwise we would infer that q ∗ ∈ γ0 which contradicts our assumption
that q ∗ ∈ d. This contradicts Lemma 7.4.4, showing that d must in fact be
empty.
Remark: The second line of the proof may be expanded as follows. Take
qk ∈ d with qk → p. Then p ∈ ω(γ) = d ∪ γ0 since ω is closed. Suppose all
such limit points p lay in d. Then d would be closed, and since d and γ0 are
disjoint closed sets, we could infer that ω is not connected.
With the aid of the preceding lemmas, we can now prove a version of
the Poincarè-Bendixson theorem:
Theorem 7.4.1 (Poincarè-Bendixson Theorem) If γ is a bounded semi-
orbit for which ω(γ) lies in Ω and contains no equilibrium points of f , then
ω(γ) is a periodic orbit.
164

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:

Theorem 7.4.2 Let γ be a positive semiorbit contained in a compact subset


K of Ω and suppose that K contains only a finite number of equilibrium
points. Then

1. ω(γ) consists of a single point p which is an equilibrium point of f ,


and the orbital point x(t) → p as t → +∞, or

2. ω(γ) is a periodic orbit, or

3. ω(γ) consists of a finite set of equilibrium points together with their


connecting orbits. Each such orbit approaches an equilibrium point as
t → +∞ and as t → −∞

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.

1. Suppose ω contains no regular points. Then it consists entirely of


equilibrium points and, since it is connected, it must consist of exactly
one of these, x0 say. Inasmuch as ω = {x0 }, the orbit approaches x0
as t → ∞.

2. Suppose next that ω contains a regular point but no equilibrium point.


Then it contains a complete orbit and no equilibrium point, so it is
periodic by the Poincarè-Bendixson theorem.
165

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.

7.4.1 Limit Cycles


A comprehensive theory of periodic orbits in Rn and their stability is taken
up in Chapter 9. However, when n = 2, we are able to draw a number
of useful conclusions on the basis of the Poincarè-Bendixson theory and its
supporting lemmas.
The setting that we wish to consider is the following. We suppose that
the region Ω contains an isolated periodic orbit Γ. We then show that nearby
orbits spiral into Γ either as t → +∞ or as t → −∞.
Theorem 7.4.3 Suppose that Γ is a non-constant, periodic orbit of the two-
dimensional system (7.12) and there are no other periodic orbits within some
neighborhood of Γ. Then every trajectory beginning sufficiently close to Γ
spirals into it either as t → +∞ or as t → −∞.
166

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

Proof: It is possible to choose a neighborhood N of Γ in which there are no


equilibrium points, for otherwise there would be a sequence of equilibrium
points {xn } for which d(xn , Γ) → 0 as n → ∞. This sequence is bounded
and therefore has a convergent subsequence, converging (say) to x∗ . Since
f (xn ) = 0, this is so for x∗ as well, i.e., x∗ is an equilibrium point. But
d(x∗ , Γ) = 0, i.e., x∗ ∈ Γ: this contradicts the assertion that Γ is a non-
constant, periodic orbit. Choose N so that it contains no equilibrium point
of the system (7.12), and no periodic orbit other than Γ.
Establish a transversal S to Γ at a point p. Then Γ returns to p at
intervals of time ±T , where T is the (least) period of Γ. Orbits starting
within a sufficiently small neighborhood of Γ must likewise intersect S within
a time interval close to ±T , by virtue of the continuity of solutions x =
φ(t, x0 ) with respect to initial data x0 . It therefore suffices to consider orbits
starting on S within a sufficiently small neighborhood of Γ (smaller than the
neighborhood N of the preceding paragraph). Let q0 be such a point for
some orbit γ and let q1 be the next intersection of γ with S for t increasing,
q−1 the next intersection for t decreasing. These two points cannot be equal
or γ would be periodic, which is excluded. Therefore one of them is closer
to p than the other; for definiteness, say d(q1 , p) < d(q−1 , p). We consider
the Bendixson pocket B1 formed by following γ from q0 to q1 and then S
from q1 back to q0 , and we repeat this, getting a second intersection q2 . It is
clear that d(q2 , p) < d(q1 , p), and so on: successive intersections are nearer
to p.
Suppose the sequence {qn } failed to converge to p. Denote the inte-
rior of the curve Γ by IΓ and consider the closure of the set B1 /IΓ . The
positive semi-orbit γ is confined to this compact set, which is free of equilib-
rium points and which then, by the Poincarè-Bendixson theorem, contains
a periodic orbit – a contradiction. Therefore the intersections {qn } con-
verge to p and the orbit γ spirals into Γ as t → ∞. Had we assumed that
d(q−1 , p) < d(q1 , p) instead, then we would conclude that γ spirals into Γ as
t → −∞. 

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

We consider next an important class of cases for which periodic orbits


are not isolated.

7.4.2 Hamiltonian systems


Let the system (7.12) be in canonical Hamiltonian form, defined as follows:
dq ∂H dp ∂H
= , =− (7.14)
dt ∂p dt ∂q
where we have written x = (q, p) to conform to standard terminology in
Hamiltonian dynamics, and H(q, p) = H(x) is a smooth function in a domain
Ω in R2 . Suppose the domain Ω contains an isolated equilibrium point
x0 = (q0 , p0 ) which is, moreover stable: given a neighborhood U0 of x0 , all
orbits starting in a sufficiently small neighborhood V0 ⊂ U0 remain in U0 for
all t > 0. This will be true if the function H(x) − H(x0 ) is positive-definite
near x0 .4 We may assume U0 is small enough to exclude any equilibrium
points other than x0 . We may then infer, under fairly general assumptions,
that all orbits starting in V0 are periodic. The precise result, stated for
the case when the equilibrium point in question is located at the origin of
coordinates, is as follows.

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)

on Ω \ {O}. Then there is a subdomain V0 of Ω such that all orbits beginning


in V0 \ {O} are periodic.

Remark: By the notation Ω \ {O} we mean the punctured region consisting


of Ω with the origin removed.
Proof: The part of the line segment x = st̂ that lies in Ω \ {O} is easily
seen to be a transversal S to orbits in this region. Choose V0 so that any
orbit beginning in V0 remains in Ω for all subsequent times. Let γ be any
orbit beginnining in V0 \ {O}; it is necessarily nonconstant. If it is not
periodic it must, by Lemma 7.4.2, intersect S successively at distinct points
x1 = x(t1 ), x2 = x(t2 ), . . .. Because of the condition (7.15), H(x1 ) 6= H(x2 ).
But the structure of the Hamiltonian system is such that H(x(t)) is constant
on orbits. It follows that γ is periodic. 
4
We discuss stability more fully in the next chapter.
169

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.

Example 7.4.2 Let


p2 q 2 q 3
H= + + .
2 2 3
This Hamiltonian has a separatrix S passing through the unstable equi-
librium point at (q, p) = (−1, 0) (see Figure 7.3). This curve, on which
H(q, p) = 1/6, separates the plane into regions of different behavior. It is
also an orbit, which approaches the (unstable!) (q, p) = (−1, 0) both as
t → +∞ and as t → −∞. The region marked A in Figure 7.3, which con-
tains the stable equilibrium point (q, p) = (0, 0), is invariant and bounded.
Any orbit in this region satisfies the conditions of Theorem 7.4.1, so all
orbits in A are periodic.

PROBLEM SET 7.4.1


170

1. Consider a two-dimensional system ẋ = f (x, y), ẏ = g(x, y) and


consider an invariant curve of this system of the form y = h(x), where
h is a differentiable function on some interval I of the x axis. Show
that a necessary and sufficient condition for h to represent such a curve
is that
h0 (x)f (x, h(x)) = g(x, h(x)) (7.16)
on I.

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.

3. Put R = 0 in Example 7.4.1. Show that the system is then Hamilto-


nian. Find the Hamiltonian function.

4. Consider the system


p p
ẋ = (1 − x2 + y 2 )x − y, ẏ = (1 − x2 + y 2 )y + x.

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.

5. Repeat the previous problem for the system


p p
ẋ = |1 − x2 + y 2 |x − y, ẏ = |1 − x2 + y 2 |y + x

where the vertical bars | · | denote absolute value.

6. Hilbert’s Sixteenth Problem asks, in part, for the number of limit


cycles the system (7.12 can have when f1 and f2 are polynomials of
degree n in x1 , x2 . The maximum number for a given degree n, H(n),
has been difficult to estimate. Show that H(1) = 0.
314
Bibliography

[1] G. Birkhoff and G.-C. Rota. Ordinary Differential Equations. John


Wiley and Sons, Inc., New York, 1978.

[2] P.F. Byrd and M. D. Friedman. Handbook of elliptic integrals for engi-
neers and physicists. Springer, Berlin, 1954.

[3] Jack Carr. Applications of Centre Manifold Theory. Springer-Verlag,


New York, 1981.

[4] Earl A. Coddington and Norman Levinson. Theory of Ordinary Differ-


ential Equations. New York: McGraw-Hill, 1955.

[5] J. Guckenheimer and P. Holmes. Nonlinear Oscillations, Dynamical


Systems, and Bifurcation of Vector Fields. Springer-Verlag, New York,
1983.

[6] Einar Hille. Lectures on Ordinary Differential Equations. London:


Addison-Wesley Publishing Company, 1969.

[7] M. Hirsch and S. Smale. Differential Equations, Dynamical Systems,


and Linear Algebra. Academic Press, New York, 1974.

[8] E.L. Ince. Ordinary Differential Equations. Dover Publications, New


York, 1956.

[9] S. MacLane and G. Birkhoff. Algebra. New York: Macmillan, 1967.

[10] Jerrold E. Marsden, Anthony J. Tromba, and Alan Weinstein. Ba-


sic Multivariable Calculus. New York: Springer-Verlag:W.H. Freeman,
1993.

[11] I.G. Petrovski. Ordinary Differential Equations. Prentice-Hall Inc.,


Englewood Cliffs, 1966.

315
316

[12] Walter Rudin. Principles of Mathematical Analysis. McGraw-Hill, New


York, 1964.

You might also like