arXiv:2101.07991v1 [math.DS] 20 Jan 2021 oia ojgc.I h aeo os ti omnt classi to 1.1 follows. 13]. common as 12, Definition stated is [15, usually equivalence it is topological equivalence flows, of topological of conside of notion by case the achieved the on is task based In this maps, conjugacy. of logical case the In systems. pcs w osΦ: Φ flows Two spaces. al equivalent cally ri fΦi apdt nobto rsrigteorientatio the preserving Ψ of orbit an to mapped is Φ of orbit 1 Ψ( (1) h s h aeciei si ento 1.1: Definition in as criteria same the use o l ( all for and o xml,tefloigtresseson par systems the three by following only d the Had distinguished inherent of solutions example, sense an For different the be is in can there p well-posedness there the without property, to systems this equivalence of topological cation of of result notion the a generalizing As entation. : oooia lsicto sabscpolmi h td of study the in problem basic a is classification Topological hsdfiiinmyb ehae sflos hr xs hom a exist there follows: as rephrased be may definition This hrfr,teprmtiaini goe ihteexcepti the with ignored is parametrization the Therefore, X 3 ˙ (3) 2 ˙ (2) 1 ˙ (1) τ (0 → QIAEC FTPLGCLDYNAMICS TOPOLOGICAL OF EQUIVALENCE ,x t, h sa ento.Drn hspoes egnrlz Yor groups. generalize topological we of process, rel action this its the During th to discuss definition. and axiomatic Yorke, usual the by the on “to proposed of based dynamics notion systems a topological formulate such we study, between this equivalence” In well-po inhe without Hadamard. is systems of it to sense However, concept this flows. generalize of to systems ficult dynamical of study the in Abstract. x x x x , X 1 = { ∈ ∈ o all for 0 = ) ) ′ [1 ∈ n map a and 1 . / / R 2 2 , fteeeit homeomorphism a exists there if × , 1]. 1 Tplgclequivalence) (Topological X, h oino oooia qiaec ly nesnilro essential an plays equivalence topological of notion The WELL-POSEDNESS WITHOUT } . where x τ ∈ R : τ X R × ( τ 1. ,x t, OOAUSUDA TOMOHARU . × ( X · x , Introduction X ) h , → smntnclyicesn n bijective, and increasing monotonically is ) → ( x X R )= )) 1 n : Ψ and uhthat such h . (Φ( Let R ,x t, r nitnusal fwe if indistinguishable are R h X )) × : X and Y → ens nthe in sedness → olmo classifi- of roblem Y Y estheory ke’s to with ation etydif- rently Y pological uhta each that such etopological be h definition The mr 1] as [16], amard igtetopo- the ring fteorbit. the of n no h ori- the of on oyof eory ametrization. are eomorphism ysystems fy ffiut in ifficulty dynamical le topologi- 2 TOMOHARU SUDA
Here systems (1) and (2) are differential inclusions. The details on the differential inclusions can be found in [3], for example. Thus, in the classification of systems without well-posedness, it is neces- sary to consider a type of “topological equivalence,” which does not com- pletely ignore the parametrization. There are several notions of “systems” that can be used as the objects of our equivalence. While set-valued flows are often used regarding the consideration of stability, it appears difficult to formulate the condition of orbit preservation in this setting [4, 14]. A more feasible candidate is the axiomatic theory of the solutions of ordi- nary differential equations. In particular, Filippov’s formalism and Yorke’s formalism have potential as the foundation of our methods, although neither considers the morphism between systems [9, 17]. In these theories, the solu- tion set of an ordinary differential equation is considered as a set of partial maps on R satisfying certain axioms. While Filippov’s formalism is fairly developed and applicable to a wide variety of problems, it is based on par- tial maps with closed domains and requires more axioms, which makes the generalization or the consideration of morphisms relatively difficult. Yorke’s formalism is based on partial maps with open domains; only a few axioms are required and it has been announced that basic results of the topological dynamics have been generalized. However, the proofs of these results have apparently not been published, as mentioned in [11]. In this article, we consider Yorke’s formalism in the setting of the action of topological groups, which can be seen as a generalization of topological dynamics [6, 7]. In particular, we define the notion of morphism and thereby obtain a notion of “topological equivalence” applicable to systems without well-posedness. During this process, we state and prove some generalization of the basic results announced in [17]. The remainder of this article is organized as follows. In Section 2, we define the space of partial maps and introduce Yorke’s formalism of topo- logical dynamics based on the star-construction. In Section 3, we consider the shift map of the partial maps and its relationship with the topological transformation groups and dynamical behavior. In Section 4, we introduce the notion of morphism of the star-construction and equivalences of systems without well-posedness, and discuss their relation with the usual definition of topological equivalence. In Section 5, we consider some examples to il- lustrate the formalism developed in this article. Finally, in Section 6, we provide concluding remarks.
2. Space of partial maps and Yorke’s formalism In this section, we introduce the basic concepts and constructions used throughout the article. In the following, X is a second-countable metric space and G is a locally compact second-countable Hausdorff topological group. Note that G is metrizable by the Birkhoff-Kakutani theorem. EQUIVALENCE OF TOPOLOGICAL DYNAMICS WITHOUT WELL-POSEDNESS 3
Definition 2.1 (Partial maps). A continuous map φ : D X a partial map from G to X if D G is a nonempty open set. → ⊂ The set of all partial maps is denoted by Cp(G, X). For each φ : D X, we set dom φ := D. → A partial map φ C (G, X) with a nonempty connected domain is max- ∈ p imally defined if, for all ψ Cp(G, X) with a nonempty connected domain, the condition dom φ dom∈ψ and φ = ψ on dom φ implies φ = ψ. ⊂ Our aim is to topologize Cp(G, X) by introducing the compact-open topol- ogy. However, as the domains may vary, a slight modification is needed. The following construction is obtained from Abd-Allah and Brown [1]. Let ω / X and Xˆ := X ω . The open sets of Xˆ are defined by Xˆ and ∈ ∪ { } the open sets of X. Now we define a bijection µ : C (G, X) C(G, Xˆ ), p → where C(G, Xˆ ) is the set of all continuous maps from G to Xˆ topologized by the usual compact-open topology, by setting φ(x) (x dom φ) µ(φ)(x) := ∈ (ω (otherwise).
The topology on Cp(G, X) is defined so that it makes µ a homeomorphism. It has a subbasis given by the sets of the form W (K,U) := φ C (G, X) K dom φ, φ(K) U . { ∈ p | ⊂ ⊂ } The compact-open topology coincides with the topology of compact con- vergence. Lemma 2.2. The compact-open topology coincides with the topology of com- pact convergence. Namely, Cp(G, X) is a second countable space and φn φ in C (G, X) as n if and only if, for all compact subset K dom→φ, p → ∞ ⊂ K dom φn for sufficiently large n and supt K d(φn(t), φ(t)) 0 as n .⊂ ∈ → → ∞ Proof. The proof of the second countability of Cp(G, X) can be found in [2] (Proposition 2.1). Though the rest of the proof is similar to that of the classical one, we present it for the sake of completeness. Let φn φ in Cp(G, X) as n . Fix a compact subset K dom φ and a positive→ number ǫ> 0. For each→∞x K, we may find a relatively⊂ compact ∈ neighborhood Ux of x such that U dom φ, x ⊂ φ(U ) B (φ(x)), x ⊂ ǫ/3 φ(U¯ ) B (φ(x)). x ⊂ ǫ/2 n Because K is compact, there exist x1,x2, ,xn K with K i=1 Uxi . ¯ · · · ∈ n ⊂ Let Ki := Uxi K and Vi := Bǫ/2(φ(x)). Then we have φ i=1 W (Ki,Vi). n∩ ∈ S Therefore, i=1 W (Ki,Vi) is a neighborhood of φ in Cp(G, X) and there n T exists n0 N such that φn W (Ki,Vi) for all n n0. ∈ T ∈ i=1 ≥ T 4 TOMOHARU SUDA
Fix x K. Then x Ki for some i and we have for all n n0, ∈ ∈ ǫ ≥ ǫ d(φ(x), φ (x)) d(φ(x), φ(x )) + d(φ(x ), φ (x)) < + = ǫ. n ≤ i i n 2 2 Therefore, we conclude that supt K d(φn(t), φ(t)) ǫ. ∈ ≤ Conversely, let φn φ in the compact convergence and take a neighbor- n → hood i=1 W (Ki,Vi) of φ. For each i, there exists ri > 0 with B (φ(K )) V . T ri i ⊂ i Let r := min(ri). For each i, by the definition of compact convergence, there exists ni N with ∈ K dom φ , i ⊂ n sup d(φn(t), φ(t)) < r t Ki ∈ for all n ni. If we let n0 = max(ni), we have supt Ki d(φn(t), φ(t)) < r for all n n≥ and i = 1, 2, , n. Therefore, we deduce∈ for each t K , ≥ 0 · · · ∈ i φ (t) B (φ(t)) V . n ∈ r ⊂ i Thus, we obtain φ n W (K ,V ). n ∈ i=1 i i From the next lemma,T we may obtain maximally-defined partial maps. Lemma 2.3. Let φ C (G, X) with a nonempty connected domain. Then, ∈ p there exists a partial map φ¯ Cp(G, X) which is maximally defined and satisfies φ = φ¯ on dom φ. ∈
Proof. We define an order on the set Cc := φ Cp(G, X) dom φ is connected by setting φ ψ if dom φ dom ψ and φ({x)=∈ ψ(x) for all| x dom φ. This } ≤ ⊂ ∈ is clearly a partial order on Cc. Fix φ Cc and consider φ := ψ Cc φ ψ . As φ φ, φ is nonempty.∈ Let be a chain inC . We{ define∈ a| map≤ } by setting∈ C C S Cφ S dom := dom ψ W S ψ _ [∈S (x) := ψ(x) if x dom ψ. S ∈ Because is totally ordered,_ is well-defined, , and ψ S S S ∈ S ≤ S for all ψ . Therefore, from Zorn’s lemma, we see that φ has a maximal element,∈ which S is the desired partialW map. W C W Maximally defined partial maps have properties similar to those of the solution curves of an ordinary differential equation. The set of all maxi- mally defined partial maps is denoted by Cs(G, X). Clearly, C(G, X) with compact-open topology is a subspace of Cs(G, X). We define the orbit (φ) of φ C (G, X) by O ∈ s (φ) := φ(g) g dom(φ) . O { | ∈ } In general, the space Cs(G, X) is not Hausdorff and the limit of a sequence is not unique. However, we have the following property. EQUIVALENCE OF TOPOLOGICAL DYNAMICS WITHOUT WELL-POSEDNESS 5
Lemma 2.4. Let φn n be a convergent sequence in C(G, X) with φn φ as n . If φ { ψ }as n in C (G, X), then φ = ψ. → →∞ n → →∞ s Consequently, if φni is a subsequence of φn n convergent in Cs(G, X), φ φ as i {in C}(G, X). { } ni → →∞ s Proof. By Lemma 2.2, we have φ = ψ on dom(ψ) G = dom(φ). Because ψ C (G, X), we have φ = ψ. ⊂ ∈ s Further, the next construction by Yorke enables us to avoid the difficulty accompanied with Cs(G, X) [17].
Definition 2.5 (Star-construction). For a subset S Cs(G, X) and W G X, we define ⊂ ⊂ × S∗W := (g, φ) φ S, g dom φ, (g, φ(g)) W . { | ∈ ∈ ∈ } Because S W G C (G, X), the topology of S W is naturally defined by ∗ ⊂ × s ∗ the subspace topology. We call S∗W the star-construction defined by S and W. Basic properties of the star-construction may be summarized as follows: Lemma 2.6. The following properties hold: (1) For S,S C (G, X) and W, W G X,S S and W W ′ ⊂ s ′ ⊂ × ⊂ ′ ⊂ ′ implies S∗W S′∗W ′ and S∗(W ′ W )= S∗W ′ S∗W. (2) For S C (G,⊂ X) and W G \X (µ M),\ we have ⊂ s µ ⊂ × ∈
S∗ Wµ = S∗Wµ, µ M µ M [∈ [∈ and
S∗ Wµ = S∗Wµ. µ M µ M \∈ \∈ (3) For subsets S C (G, X) (λ Λ) and W G X, we have λ ⊂ s ∈ ⊂ ×
( Sλ)∗W = Sλ∗W, λ Λ λ Λ [∈ [∈ and
( Sλ)∗W = Sλ∗W. λ Λ λ Λ \∈ \∈ Proof. The proofs are straightforward.
The next theorem enables us to analyze the “solution space” S∗W in a way similar to that of the usual ordinary differential equations.
Theorem 2.7. For all S Cs(G, X) and W G X, S∗W is second countable and Hausdorff. ⊂ ⊂ × 6 TOMOHARU SUDA
Proof. G C (G, X) is a second countable space by Lemma 2.2. Conse- × s quently, S∗W is also second countable. For the Hausdorff property, let (g, φ), (g , ψ) S W with (g, φ) = (g , ψ). ′ ∈ ∗ 6 ′ If g = g , we may find open neighborhoods U ,U ′ G of g and g with 6 ′ g g ⊂ ′ U U ′ = . Then we have U V U ′ V = for all open neighborhoods g ∩ g ∅ g × ∩ g × ′ ∅ V,V ′ of φ and ψ. If φ = ψ, the maximality of domains implies that dom φ dom ψ = or φ(g ) =6 ψ(g ) for some g dom φ dom ψ. In the former∩ case we have∅ 0 6 0 0 ∈ ∩ g = g′, which we have already considered. In the latter case, there exists r6 > 0 with B (φ(g )) B (ψ(g )) = . Because W ( g ,B (φ(g ))) and r 0 ∩ r 0 ∅ { 0} r 0 W ( g0 ,Br (ψ(g0))) are disjoint, we have U W ( g0 ,Br(φ(g0))) U ′ W ({ g } ,B (ψ(g ))) = for all neighborhoods×U and{ U} of g and g .∩ × { 0} r 0 ∅ ′ ′ Remark 2.8. The star construction may be characterized as follows. We establish the evaluation map ev : G Cs(G, X) G Xˆ by ev(g, φ) := (g,µ(φ)(g)), and the projection p : G × C (G, X) →C (×G, X) by p(g, φ) := × s → s φ. Note that these maps are continuous. Then the set S∗W is the largest subset E of G C (G, X) such that ev(E) W and p(E) S. In this sense, × s ⊂ ⊂ the set S∗W characterizes the “Cauchy problem” on W with the solution set S, which amounts to finding a map φ S satisfying φ(g) = x for each (g,x) W . ∈ ∈ In the next definition, we list the main additional axioms proposed by Yorke, which are abstractions of the conditions for well-posedness. Definition 2.9. Let S C (G, X). ⊂ s (1) The subset S satisfies the compactness axiom if S∗W is compact for all compact W G X. ⊂ × (2) The subset S satisfies the existence axiom on W if S∗ (t,x) is nonempty for all (t,x) W . { } ∈ (3) The subset S satisfies the uniqueness axiom on W if S∗ (t,x) is empty or a singleton for all (t,x) W . { } (4) The subset S has domain D if D ∈ dom φ for all φ S. ⊂ ∈ Remark 2.10. Because S∗W is second countable, S satisfies the com- pactness axiom if and only if S∗W is sequentially compact for all compact W G X. ⊂ × The next result, which is an analogue of the classical convergence theorem (Chapter 2, Theorem 3.2 in [10]), clarifies that the compactness axiom is an abstraction of the continuous dependence on the initial conditions.
Theorem 2.11. A subset S Cs(G, X) satisfies the compactness axiom if and only if the following property⊂ holds: if (g ,x ) (g,x) in G X and n n → × there exists a sequence of maps φn S with φn(gn) = xn, there is a map ψ S with ψ(g)= x and a subsequence∈ (g φ ) with (g , φ ) (g, ψ). ∈ { ni ni } ni ni → Proof. Let S Cp(G, X) satisfy the compactness axiom and (gn,xn) (g,x) be a sequence⊂ in G X with a corresponding sequence of maps φ →S × n ∈ EQUIVALENCE OF TOPOLOGICAL DYNAMICS WITHOUT WELL-POSEDNESS 7 satisfying φ (g )= x . Then W := (g,x) (g ,x ) n N is compact, n n n { } ∪ { n n | ∈ } and therefore, S∗W is sequentially compact. Because (gn, φn) S∗W for each n, we may take a convergent subsequence (g , φ ) (g ,∈ ψ) S W. ni ni → ′ ∈ ∗ It is clear that g = g′. Conversely, let W G X compact and (g , φ ) (n = 1, 2, ) be a ⊂ × n n · · · sequence in S∗W. Then, (gn, φn(gn)) W for n = 1, 2, , and therefore, we can find a convergent subsequence∈ (g , φ (g )) · ·( ·g,y) W (i ni ni ni → ∈ → ). From the hypothesis, there exist a subsequence (gn , φn ) and a pair ∞ ij ij (g, ψ) S∗W satisfying (gn , φn ) (g, ψ) as j and ψ(g) = y. ∈ ij ij → → ∞ Therefore, S∗W is sequentially compact. From the classical convergence theorem, we immediately observe that the solution space of an ordinary differential equation satisfies the compactness axiom: Corollary 2.12. Consider an ordinary differential equation
x′ = f(t,x), where f : R W W is continuous and W Rn is a nonempty open set. Then the solution× → space S satisfies the compactness⊂ axiom. Further, a topological transformation group can be identified with a func- tion space satisfying the compactness axiom. Details on the topological transformation groups may be found in [5]. Corollary 2.13. Let π : G X X be a continuous left G-action on X. Then the set of maps × → S := π( ,x) x X { · | ∈ } satisfies the compactness, existence, and uniqueness axioms if X is locally compact. Proof. The existence and uniqueness axioms are satisfied because we have 1 S∗ (t,x) = t,π ,π(t− ,x) { } · for each (t,x) G X. For the compactness∈ × axiom, we show that π ,π(t 1,x ) π ,π(t 1,x) · n− n → · − as n whenever (t ,x ) (t,x) as n . Let y := π(t 1,x ) and n n n n− n y := →π(t ∞1,x), and fix a compact→ subset K→ ∞ G. We may find a neigh- − ⊂ borhood U of y with U¯ being compact and yn U for sufficiently large n. Because π is continuous, π is uniformly continuous∈ on K U.¯ Therefore, we × have sups K d(π(s,yn),π(s,y)) 0 as n . ∈ → →∞ In some cases, the compactness axiom is not independent with other ax- ioms and certain restrictions are present.
Corollary 2.14. Let subset S Cs(G, X) satisfy the compactness axiom and there exists a subset W ⊂G X such that S satisfies the existence axiom on W . If X is locally compact,⊂ × then S satisfies the existence axiom on the closure of W . 8 TOMOHARU SUDA
The following theorem is a generalization of a well-known result on the global solution.
Theorem 2.15. Let S Cs(G, X) satisfy the compactness axiom and φ S. If there is a compact⊂ set K with (φ) K, dom φ coincides with∈ a connected component of G. O ⊂ Proof. There is a connected component G of G satisfying dom φ G . c ⊂ c If dom φ = Gc, ∂ (dom φ) = . Then we may take a sequence gn g ∂ (dom φ)6 with g dom φ.6 Because∅ the image of φ is contained in→K, we∈ n ∈ may assume φ(gn) y K. By Theorem 2.11, there exists (g, ψ) with (g , φ) (g, ψ). Therefore,→ ∈ g dom φ, which is a contradiction. ni → ∈ The next theorem is a generalization of a result announced by Yorke.
Theorem 2.16. If S Cs(G, X) satisfies the compactness axiom and X is locally compact, S W ⊂is metrizable for each subset W G X. ∗ ⊂ × Proof. First we show that S∗W is locally compact. Fix (g, φ) S∗W. Be- cause G and X are locally compact, we may take relatively compact∈ neigh- borhoods U and V of g and φ(g) such that φ(U¯) V . Then S (U¯ V¯ ) S W ⊂ ∗ × ∩ ∗ is a compact neighborhood of (g, φ) in S∗W. This is verified by observing
(g, φ) U W (U,V¯ ) S∗W S∗(U¯ V¯ ) S∗W. ∈ × ∩ ⊂ × ∩ Because S∗W is locally compact and Hausdorff, it is regular. Using the second countability of S∗W (Theorem 2.7), we may apply the Urysohn metrization theorem.
3. Shift of partial maps and dynamical behavior Based on the concepts introduced in the previous section, we consider generalizations of the usual notions in dynamical systems theory. The concept of shift invariance plays a central role in the discussion in this section. First we establish that the shift map is a G-action.
Theorem 3.1 (Shift map). The shift map σ : G Cp(G, X) Cp(G, X), which is defined by × → σ(g, φ)(x) := φ(xg) for x dom(φ)g 1, is continuous and satisfies the following conditions: ∈ − (1) For each φ Cp(G, X) we have σ(e, φ)= φ. (2) For all g, h ∈ G and φ C (G, X), we have σ(g,σ(h, φ)) = σ(gh, φ). ∈ ∈ p That is, σ is a left G-action. 1 Proof. To show that σ is continuous, it suffices to confirm that σ− (W (K,V )) 1 is open for each compact K G and open V X. Let (g, φ) σ− (W (K,V )) . This is equivalent to ⊂ ⊂ ∈ σ(g, φ)(K)= φ (Kg) V. ⊂ EQUIVALENCE OF TOPOLOGICAL DYNAMICS WITHOUT WELL-POSEDNESS 9
Therefore, we have φ W (Kg, V ) . We may find an open neighborhood U of g such that U¯ is compact∈ and
1 KU¯ dom(φ) φ− (V ), ⊂ ∩ using the regularity of G. Then we have
1 (g, φ) U W KU,V¯ σ− (W (K,V )) . ∈ × ⊂ 1 Thus, σ− (W (K,V )) is open. The other assertions are proven via direct calculations.
Remark 3.2. By the definition of the shift map, we have
1 dom (σ(g, φ)) = dom (φ) g− for each (g, φ) G Cp(G, X). Further, the following identity holds for all left G-action π∈: G × X X : × → (2) σ (g,π( ,x)) = π ( ,π(g,x)) . · · Thus, the triplet (G, σ, Cp(G, X)) is a topological transformation group. As the name “shift system” usually refers to discrete systems, here we call it the Bebutov system on Cp(G, X). Now we clarify the relationship between the Bebutov system and the usual topological transformation groups. This is given by the next theorem, which is a generalization of a result by Yorke (Theorem 2.3 in [17]). Theorem 3.3. Let X be locally compact. Then a σ-invariant subset S ⊂ Cs(G, X) satisfies the compactness, existence, and uniqueness axioms and has domain G if and only if it is given by a left G-action πS : G X X on X via × → (3) S := π ( ,x) x X . { S · | ∈ } Proof. The sufficiency is a consequence of Corollary 2.13 and the identity (2). Let a σ-invariant subset S C (G, X) satisfy the compactness, existence, ⊂ s and uniqueness axioms and have domain G. We establish a map πS : G X X by × → πS(g,x) := φ(e,x)(g), where φ(e,x) is the unique element in S with φ(e,x)(e) = x. We note that σ g, φ S and (e,x) ∈