Introduction to the Topology of Continuous Dynamical Systems Andries Smith 1.

Continuous General Dynamical Systems

Let (X, d) be a nonempty metric space with the usual basis, B = {B (x) | x X, R+ } 1.1. Denition. A continuous dynamical system in the state or phase space X is a function : X R X, written (x, t) = y, or t (x) = y for t R, y, x X, where has the following properties: 1. Continuity in X R:
tt0 ,xx0

lim

(x, t) = (x0 , t0 )

2. Initial condition: for any p X, (p, 0) = p 3. Group property: for any t1 , t2 R and any p X, ((p, t1 ), t2 ) = (p, t1 + t2 ) or with a dierent notation t1 (t2 (p)) = t1 +t2 (p) This t is called the evolution operator. The parameter t R is called time. Fixing t, we note that t is a mapping of X into itself. The evolution operator therefore denes a one-parameter family of mappings X X. The second notation highlights that the composition of these mappings has the some of the structure of the group (R, +): Proposition 1. The family of mappings G = { t | t R } under composition is a homomorphic image of (R, +), and hence a group. Proof. Let a map f : R X be dened by t t . Then clearly f is a surjective homomorphism. [Note that t can also be seen as dening a group action of (R, +) on X, where t acts as the map x t (x).] Theorem 1. (The Integral Continuity Condition) For any point x X, any T , R, T > 0, > 0, there exists a R such that (B (x), t) B ((x, t)) for all t [0, T ]. Proof. Assume let > 0, T > 0, and assume there does not exist such a . Take n 0, a sequence in R. We have assumed there is, for each n, some yn such that d(x,yn ) < n and a tn [0, T ] such that d((x, tn ), (yn , tn )) > .
1

By the sequential compactness of [0, T ], there is a convergent subsequence tnk t0 . Clearly, yn x and therefore ynk x. But then by continuity,
k

lim (ynk , tnk ) = (x, t0 ), lim (x, tnk ) = (x, t0 )

These two limits give

0, a contradiction.

Equivalently, the result of this theorem states that for any given t, (x, t) is a continuous function of the initial condition x. Proposition 2. The mappings in the group G are self-homeomorphisms of X. Proof. Let t G. Then t t = t t = 0 = IdG , so t has a twosided inverse and is hence a bijection. Moreover, all t in G are continuous on X, and hence t is a homeomorphism. Some notation: for A, X, T R, (A, T ) =
xA, tT

(x, t)

(G,+) a group, H, K G, H +K =
kK,hH

h+k

Fixing a point x0 X, the set (x0 , R) is called x0 s orbit or trajectory, while the function (x0 , t) is called the motion of x0 . This function may or may not be injective. If it is, the motion is called singular. Motions which are not injective are contrarily nonsingular. Suppose (x0 , t) is nonsingular. It is possible that (x0 , t) = x0 for all t R, in this case, x0 is called a rest (critical, equilibrium, or stationary) point. A nonsingular motion of a non-rest point must be periodic. To see this, note that for such a nonsingular motion, there must exist some t1 and t2 , with, say, t2 > t1 , such that (x0 , t1 ) = (x0 , t2 ) Let t2 t1 = . Then by the group property, (x0 , +t) = (x0 , t2 t1 +t) = ((x0 , t2 ), t1 +t) = ((x0 , t1 ), t1 +t) = (x0 , t1 t1 + t) = (x0 , t) Proposition 3. if (x0 , t) is nonsingular and x0 is not a rest point, there is a smallest > 0, called the period, for which (1.1) (x0 , t) = (x0 , + t) Proof. Let 1 satisfy (1.1). If there is some 2 satisfying (1.1) such that 2 < 1 , then either there exists some 3 satisfying (1.1) such that 3 < 2 or not. Continuing, we end up with either some smallest j > 0 satisfying (1.1) or an innite strictly decreasing sequence n of elements satisfying (1.1). By the boundedness of [t, 1 ] R, n converges, say to . Clearly < i for

any i satisfying (1.1). Since (x0 , t + n ) is a constant for all n , and since by continuity (x0 , t + n ) (x0 , t + ), (x0 , t + ) = (x0 , t + n ) = (x0 , t), and so satises (1.1). Now we show > 0. Suppose to the contrary that n 0. By the continuity of , for any there exists a such that (x0 , [0, )) B (x0 ) Since n 0, there exists a i < . By the periodicity of , this implies (x0 , t) B (x0 ) for all t R. Since R, a contradiction. > 0 is arbitrary, this implies (x0 , t) = x0 for all t

This gives a complete classication of motions of points in X. Note that the trajectory or image of a periodic motion is homeomorphic to S 1 . However, the image a singular motion is not necessarily homeomorphic to R; precisely when this occurs will be investigated later. Proposition 4. Let x, y X. Then y (x, R) if and only if (x, R) = (y, R) Proof. This follows from the fact that orbit membership is an equivalence relation under any group action. A corollary of the previous proposition is that distinct trajectories never intersect. Note that this corollary furthur implies no motion can enter a rest point for a nite value of time. More generally, the motion of any point denes the motion of all points in its trajectory: letting y = (x, t0 ) (x, R), the motion of y is dened by (y, t) =(x, t0 + t). 1.2. Denition. A set A X is invariant if it is mapped into itself by all maps of the family G. In other words, (A, t) A for all t R, or, (A, R) A t is in G, so that t (A) A. Applying t to this containment, we get t (t (A)) = IdG (A) = A t (A) Which gives (A, R) = A. It is a consequence of general properties of bijections that an arbitrary union and intersection of invariant sets is invariant. Theorem 2. A X is invariant if and only if it is a union of entire trajectories, i.e. A=
iI

(xi , R), xi X

Proof. Clearly, an entire trajectory is invariant subset, and so any union of entire trajectories is an invariant subset. Conversely, suppose A is invariant. Let x A. Since (A, R) = A, in particular (x, R) A, and so A contains the entire trajectory of each x A and is hence is a union of entire trajectories. Note that a dynamical system on a space X denes a dynamical system on every invariant subset A. 1.3. Denition. A set A X is positively invariant or negatively invariant if, respectively, (A, R+ ) A, (A, R ) A Theorem 3. Let x X. If for all > 0, there exists a subset some trajectory which is the image of a connected subset of R and which is completely contained in the -neighborhood of x, i.e. if there exists a y X and [0, T ] R such that (y, [0, T ]) B (x) then x is a rest point. Proof. Assume x is not a rest point. Since the trajectory of x cannot enter a rest point, for any [0, T ] there is some t0 [0, T ] such that (x, t0 ) = x0 = x. Let d(x0 , x) m= >0 2 By the continuity of t0 on X, there is a such that (1.2) (B (x), t0 ) Bm (x0 ). We let = min ( , m). Clearly (1.2) holds if we replace with . Then by the triangle inequality, B (x) Bm (x0 ) = . Therefore, for any < , all y B (x) have (y, to ) Bm (x0 ), which implies (y, to ) B (x). We / have thus found an such that no connected subset of the domain of any trajectory is entirely contained in the -neighborhood of x, and hence the conditions of the theorem do not hold, a contradiction. Corollary 1. If limt+ (x, t) = p, then p is a rest point. Proof. Translating the meaning of this limit by the group property, we get that for any , there is a t0 such ((x, t0 ), R+ )) B (p) Then p is a rest point by Theorem 2. The same result clearly also holds for limt (x, t) = p. Proposition 5. If every -neighborhood of a point x X has a point y X such that the motion (y, t) has an arbitrarily small period, then x is a rest point.

Proof. By continuity of (x, t), there exists a [0, T ] such that for (x, [0, T ]) B (x)

> 0,

By the Integral Continuity Condition, there is a > 0 such that for y B (x), t [0, T ], (y, t) B ((x, t)) . We may assume < . Then by the triangle inequality, d((y, t), x) < 2 , t [0, T ] We may assume by the conditions of the theorem that y has an arbitrarily small period, say < T . Then letting 2 = , we have (y, R) B (x), so x is a rest point by the previous theorem. Theorem 4. Every positively invariant set A X which is homeomorphic to a closed ball in Rn contains a rest point. Proof. We identify A with a closed ball in Rn by the homeomorphism. Since A is positively invariant, t , t > 0 is a map of A into A which has a xed point by Brouwers theorem. For a sequence of functions tk , tk 0, let this xed point be xk . Then by sequential compactness, there exists a convergent subsequence xkl x A, and x is then a rest point by Proposition 5. 1.4. Denition. Two dynamical systems, : X R X and : Y R Y are said to be isomorphic or topologically equivalent if there exists a homeomorphism f : X Y such that f ((x, t)) = (f (x), t) If the map f is not a homeomorphism but at least continuous, then the systems are homomorphic dynamical systems. It is easy to see that a homomorphism (and hence an isomorphism) of dynamical systems maps trajectories to trajectories, rest points to rest points, periodic motions to periodic motions, and invariant sets to invariant sets. In particular, these facts allow the complete classifcation of the isomorphism type of a dynamical system with X = R by its rest points and the direction of the interval trajectories. 2. Limiting Properties and Stability 2.1. Denition. A point p X is an or -limit point of a motion (x, t), respectively, if there exists a sequence tk tending to positive or negative innity, respectively, such that (x, tk ) p. Equivalently, p is an ()-limit point of (x, t) if for any > 0 and T R, there is a t > T (t < T ) such that (x, t ) B (p). [Note that all points of nonsingular motions are both and -limit points.]

2.2. Denition. The dynamical limit sets of a point x X, written x and Ax are the sets of all and -limit points, respectively, of the motion (x, t). Theorem 5. For any x X, x and Ax are invariant. Proof. We prove the theorem for x . The proof for Ax is analogous. Let x X, tk +, (x, tk ) p, and t0 R. t0 is a continuous map, so that in the image (X, t0 ), ((x, tk ), t0 ) converges to (p, t0 ), which gives, by the group property, (x, t0 + tk ) (p, t0 ). But then T + t0 +, so (p, t0 ) x . Hence (p, R) x , nishing the proof. Some more notation: A = (A, R), + = (A, R+ ), = (A, R ) A A x = x Ax Clearly, and -limit points of a motion are limit points of the trajectories, and hence x x , etc. In fact, it is easy to see that x = (x, R) x , + = (x, R+ ) x , = (x, R ) Ax x x 2.3. Denition. A point x and the motion (x, t) are said to be positively or negatively Lagrange stable (stable L,+ stable L ), respectively, if + or x are compact. If x is compact, the motion is simply Lagrange stable x (stable L). Note that rest points and periodic motions are always Lagrange stable, as is any motion on a compact space, and, if X = Rn , any motion contained within a bounded subset. Theorem 6. A motion (x, t) is positively Lagrange stable if and only if all of the following conditions hold: 1. x is nonempty. 2. x is compact. 3. limt+ d((x, t), x ) = 0. Proof. Assume (x, t) is postively Lagrange stable. We rst show this implies 1, 2, and 3. 1. Let tk +. By compactness of + , (x, tk ) must have a convergent x subsequence (x, tkl ) y,, which gives y x , so x is nonempty. 2. x is a subset of the compact space + , so all we have to show is x closure. Let q be a limit point of x . Then any open neighborhood of q, B (q) contains a y x , which, as an interior point of B (q), has an open neighborhood B (y) B (q). As y is an -limit point of a sequence (x, tk ), B (y) contains (x, t), t > T for arbitrarily large T . But by the inclusions above, this means any neighborhood of q contains such a (x, t), which gives q x .

3. Suppose limt+ d((x, t), x ) = 0. This gives that there exists a sequence tk + such that for some > 0, (2.1) d((x, tk ), x ) > , k N Since + is compact the sequence (x, tk ) contains a convergent subsequence (x, tkl ) q. But then q x , a contradiction of (2.1) above. Now assume 1, 2, and 3, hold. We show the sequential compactness of + . Assume without loss of generality that tk is monotonically increasing. x If it is bounded, it converges, and hence so does (x, tk ). If tk + By 3, there exists a sequence rk in x such that
k+

lim d((x, tk ), rk ) = 0.

By 2, rk has a convergent subsequence rkn r x . But then again by 3, (x, tk ) r x so x is Lagrange stable. If x or Ax is empty, respectively, then the motion (x, t) is called positively or negatively departing, and in case x is empty simply departing. If x is nonempty but x (x, R+ ) is empty, the motion is positively asymptotic. Negatively asymptotic is dened analogously, and a motion which is both positively and negatively asymptotic is asymptotic.

Figure 1. The Van Der Pol dynamical system. The thick blue line is a periodic, Poisson stable motion. It is also the positive dynamical limit set of both the interior and exterior motions, which are both positively asymptotic. Such a dynamical limit set which is also a periodic motion is called a limit cycle. The exterior motions are negatively departing.

The third possibility, that x (x, R+ ) (or Ax (x, R )) is nonempty, is the subject of the following denition:

2.4. Denition. A motion (x, t) is positively or negatively Poisson stable(stable P,+ stable P ) if, respectively, x (x, R+ ) or Ax (x, R ) are nonempty. If both of these intersections are nonempty, the motion is simply Poisson stable (stable P ). Proposition 6. If (x, t) is a positive Poisson stable motion, (x, R) x . Proof. x is invariant, so if y x , for any t, (y, t) x (x, R), (y, R) x (x, R), and hence (y, R) = (x, R) x Clearly, all periodic and rest motions are Poisson stable. We now give an example of a singular Poisson stable motion on the torus, {(, ) | , [0, 1] R} where ( + l, + k) = (, ) for any integers l, k. Dene d d = 1, = a, a R dt dt This denes a dynamical system ((, ), t) = ( + t, + at). It is easy to see that if a Q, a = p for p, q relatively prime, this motion is periodic with q period q. If a is irrational, the motion is not periodic, because otherwise the period would give a solution to the equations = l, a = k for l, k Z and then a = k , a contradiction. In fact, for a irrational this motion l is everywhere dense on the torus, so that (,) is the entire torus for any (, ), which gives that the motion is Poisson stable. If (x, t) is a nonsingular positively Poisson stable motion, we can x x and write (x, t) = f (t) for t R. There exists a sequence tk +, f (tk ) x. In addition, f (t) has a left inverse; call it g. Note that the image sequence of this inverse function is just the original sequence tk , i.e. g(f (tk )) = tk , and it diverges. But g(x) = 0, so g is not continuous and hence f : R+ (x, R+ ) is not a homeomorphism, which motivates the following theorem. But rst we need a lemma. Lemma 1. If X is a space with connected open basis and f is a continuous bijection from the half-line to X f : R, X, then f is a homeomorphism. Proof. : We show f is an open map. Let B be a basic open set in R. B is contained in a compact subset, K, of R. The restriction of f to K is an open mapping onto its image, so that f (B) is open in f (K). But then there is an open set, U, of X, for which the intersection U f (K) = f (B). But in a space with connected basis, the intersection of a compact set and an open set is open, so that f (B) is open in X. Theorem 7. If x X, x and (x, t) are positively Poisson stable if and only if (x, R+ ) R+ Proof. We x x and write (x, t) = f (t) as above. Then f R+ (x, R+) is a continuous bijection. If (x, R+) R+, there exists a homeomorphism = : (x, R). R+ . This gives f : R+ R+ is a continuous bijection of the half real line, which implies by the lemma that it is a homeomorphism.

But then 1 f = f is a homeomorphism, which is a contradiction. =(f (t)) 2.5. Denition. A point x X is called wandering if there exists an neighborhood B (x) and a T R such that for all t > T , (2.2) B (x) (B (x), t) = A point is nonwandering if it is not a wandering point. All singular motions are nonwandering, and if x is positively (or negatively) Poisson stable, it is a fortiori nonwandering, since then there exists a sequence tk + with (x, tk ) x. The converse is not true, since there exist nonsingular, nonwandering motions that are not Poisson stable. For example, if we inserted a rest point into the periodic motion in Figure 1, the leftover nonsingular motion would be nonwandering (thanks to the fact that its a limit set of nearby motions) but not Poisson stable, since its own dynamical limit set would just be the inserted rest point. Proposition 7. The set of wandering points, W, is open. Proof. Suppose x satises (2.3) above. If q B (x), then since it is an interior point, there is an open neighborhood B (q) B (x). Yet q with this B (q) satises the conditions in (2.3), so q is also a wandering point. This gives that the set of wandering points is open. Proposition 8. The set of wandering points, W, is invariant. Proof. Since t0 is a bijection for any t0 , (B (x), t0 ) ((B (x), t)t0 ) = . which gives that if x is any wandering point, (x, t0 ) is a wandering point for any t0 , hence the set of wandering points is invariant. The set of nonwandering points, M = W c = X \ W, is therefore closed and invariant. A point x is wandering in an invariant set A X if it is wandering in (A,d) dened as a dynamical system by the restriction |A : A R A. It is easy to show that all wandering points in X are wandering points in A. The contrapositive gives that all points that are nonwandering in A are nonwandering in X. The converse of this statement is not necessarily true; take for example the point x in Figure 1, which is wandering in L, but nonwandering in X. Theorem 8. Let x X. Every y x is nonwandering in x . Proof. Since y x , there exists a sequence (x, tk ) y, tk +. Let > 0, T > 0. Then there is a point x0 = (x, ti ) B (y). In addition it is clear that y is an -limit point of x0 , so there is a sequence pk with element pi such that (x0 , pi ) B (y), pi > T . This gives x0 B (y) (B (y), pi ), pi > T.

10

which means y is nonwandering. Corollary 2. If at least one motion in X is either positively or negatively Lagrange stable, the set of nonwandering points, M, is nonempty. Corollary 3. In a compact metric space the set of nonwandering points, M, is nonempty. Corollary 4. In a compact metric space every motion tends toward the set of nonwandering points. Proof. This follows from the fact that all motions in a compact metric space are Lagrange stable, and the fact that for a Lagrange stable motion (x, t),
t+

lim d((x, t), x ) = 0

2.6. Denition. A set A X is minimal if it is nonempty, closed, and invariant and contains no nonempty, closed and invariant proper subset. Examples of compact minimal sets include periodic and rest trajectories, while the trajectory of any departing motion is a noncompact minimal set. Theorem 9. Every closed, invariant, compact set A contains a minimal subset. Proof. If A contains a proper closed invariant subset, call it F1 . Otherwise set F1 = A. If F1 contains any closed invariant proper subset, call it F2 , otherwise set F1 = F2 . We continue to get a nested sequence of closed, invariant subsets; A F1 F2 . . . Let

F =
k=1

Fk

F is clearly closed, invariant, and is nonempty by the nite intersection property of A. We continue the procedure above to get a transnite nested sequence of closed, invariant subsets; (2.3) A F1 F2 . . . F F+1 . . . F . . . Since A is a compact metric space, it has a countable basis. By the BaireHausdor theorem, this well-ordered sequence of nested closed sets has at most a countable number of elements, and so the successive elements must coincide after some F . Then F A is a minimal set. Note that in a compact metric space X, the set of nonwandering points is closed, invariant, and nonempty. We can repeat the above argument for M, and add the requirement that each motion in Mi+1 be nonwandering in Mi , since each Mi is itself a compact invariant set and a dynamical system under . This results in an M , and since M+1 = M , M is nonwandering in itself. M is also the maximal such set (because a subset nonwandering in itself would have to be open in M ).

11

2.7. Denition. The set M (or simply Z) is called the center or set of central motions of a dynamical system. Theorem 10. Z contains an everywhere dense set of Poisson stable points. Proof. Let x0 Z, R+ . We show B (x0 ) contains a Poisson stable point. It follows from the fact that x0 is nonwandering that there exists a t1 > 1 and an 1 with 1 < 6 such that by the triangle inequality, (B 1 (x0 ), t1 ) B 2 (x) In particular, (B 1 (x0 ), t1 ) B (x) Since x0 nonwandering in Z, there is a point x1 Z, with x1 (B 1 (x), t1 ) B 1 (x) We then start over with x1 in the role of x. Since x1 is a wandering point, there is a t2 < -2 and an 2 with
2

<

d((B 1 (x), t1 )c , x1 ) 6

such that (B 2 (x1 ), t2 ) B 2 (x1 ) = Then there is some x3 in the intersection above, and x3 (B 2 (x1 ), t2 ) (B 1 (x), t1 ) B (x) Continuing indenitely we obtain two divergent sequences {t2k } and {t2k+1 }, a sequence k 0, and an innite sequence of nested, closed neighborhoods. Since compact spaces are complete, there is some unique q such that

q=
k=1

B k (xk1 )

Then for any , there is an n such that for all k > n (B k (xk1 ), tk ) B n (xn1 ) B (q) Since q B n (xn1 ), in particular (q, tk ) B (q) for all k > n. Since tk diverges in both directions, this gives q q Aq (q, R), so that q is Poisson stable. Corollary 5. Z is the closure of the set of all Poisson stable points.

12

3. Differential Equations as Dynamical Systems The general form of for a system of dierential equations f : G Rn (G Rn ), which dene a dynamical system t (x1 , . . . , xn ) by their solutions is: dxi (t) (3.1) = fi (x1 , . . . , xn ), 1 i n dt Where f is a continuous function on G, which is closed and bounded. The autonomy of f is necessary if x(t) is to induce the group property. In fact, letting x(t) be a solution, x(t0 ) = x0 Rn , x(t0 + t) is a solution only if dxi (t0 + t) xi (y) dy = = fi (x1 , . . . , xn , t), 1 i n dt dy dt where y = t0 + t. this is equivalent to fi (x1 , . . . , xn , t) = fi (x1 , . . . , xn , t0 + t), 1 i n and since t0 was arbitrary, fi (x1 , . . . , xn , t) must be a constant function of t for all i, i.e. f must be autonomous. The conditions for the existence of a solution to (3.1) with domain all of R are given by Theorem 11. If the functions fi in (3.1) are continuous on Rn and fi (x1 , . . . , xn ) = O(|x1 | + ...|xn |) then parametrized solution x : R Rn exists on all of R The proof will only be outlined: it depends on the construction of arbitrarily close approximations, called -solutions, by Euler polygons. 3.1. Denition. A parametrization x(t) : R Rn with x(0) = x0 is called an - solution of (2.1) on [a, b] R if each xi (t) is continuous, sectionally smooth, and satises:
t t

xi (t) = pi (x0 ) +
t0

fi (x1 , . . . , xn )dt +
t0

i (t)dt, |i (t)| <

where the error function : R Rn is piecewise continuous, and pi : Rn R is the i-th projection function. We use these -solutions to give a sequence of uniformly bounded, equicontinuous functions x n which converge, by Arzel`s theorem, to a continuous a function x(t) satisfying (3.1). The condition on the asymptotic behavior of f is negligible due to the fact that a dierential equation on G can simply be extended to an equation which is continuous and bounded on Rn . A clear requirement for t to be well dened is that the trajectories be distinct, or equivalently that the sets of pairs (t, x(t)) dening distinct solutions intersect trivially (these sets will be equivalence classes, after all). This is equivalent to the uniqueness of the solution x(t) passing through an arbitrary point y Rn . The conditions for this is given by:

13

Theorem 12. If the functions fi in (3.1) are Lipschitz on G, then there exists a unique solution in G satisfying a given initial condition. For this and the previous proof, see [1]. We have seen any solution x(t) of (3.1) will be continuous and satisfy the group property. Letting x(t) be the solution dened by the initial condition x(0) = x Rn , (x, t) = x(t) denes a dynamical system. References
[1] Nemytskii, V.V., Stepanov, V.V.: Qualitative Theory of Dierential Equations, Dover Publications, Inc., New York, U.S.A., Dover edition, 1989. [2] Sibirsky, K.S., translated by Leo F. Boron: Introduction to topological dynamics, Noordho International Publishing, Leyden, The Netherlands, 1975. [3] Arrowsmith, D.K.,Place, C.M.: Dynamical Systems, Chapman & Hall Mathematics, London, England, 1992. [4] Akin, E.: The General Topology of Dynamical Systems, The American Mathematical Society, Providence, RI, USA, 1993. [5] Kolmogorov, A.N.,Fomin, S.V.: Elements of the Theory of Functions and Functional Analysis, Dover Publications, Inc., New York, U.S.A., Dover edition, 1989. [6] Singer, I.M., Thorpe, J.A.,: Lecture Notes On Elementary Topology and Geomoetry, Scott, Foresman and Company, Glenview, Illinois, USA, 1967.