This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Relations among Hamiltonian, area-preserving, and non-wandering flows on surfaces

Tomoo Yokoyama Applied Mathematics and Physics Division, Gifu University, Yanagido 1-1, Gifu, 501-1193, Japan
tomoo@gifu-u.ac.jp
Abstract.

Area-preserving flows on compact surfaces are one of the classic examples of dynamical systems, also known as multi-valued Hamiltonian flows. Though Hamiltonian, area-preserving, and non-wandering properties for flows are distinct, there are some equivalence relations among them in the low-dimensional cases. In this paper, we describe equivalence and difference for continuous flows among Hamiltonian, divergence-free, and non-wandering properties topologically. More precisely, let vv be a continuous flow with finitely many singular points on a compact surface. We show that vv is Hamiltonian if and only if vv is a non-wandering flow without locally dense orbits whose extended orbit space is a directed graph without directed cycles. Moreover, non-wandering, area-preserving, and divergence-free properties for vv are equivalent to each other.

Key words and phrases:
Flows on surfaces, non-wandering flows, area-preserving flows, divergence-free flows, Hamiltonian flows
2010 Mathematics Subject Classification:
Primary 37E35; Secondary 37A05,37J05,37G30
The author was partially supported by JSPS Grant Number 20K03583 and 21H00980

1. Introduction

Area-preserving flows on compact surfaces are one of the basic and classic examples of dynamical systems, also known as locally Hamiltonian flows or equivalently multi-valued Hamiltonian flows. The measurable properties of such flows are studied from various aspects [chaika2021singularity, conze2011cocycles, forni1997solutions, frkaczek2012ergodic, forni2002deviation, kanigowski2016ratner, kulaga2012self, ravotti2017quantitative, ulcigrai2011absence]. For instance, the study of area-preserving flows for their connection with solid-state physics and pseudo-periodic topology was initiated by Novikov [novikov1982hamiltonian]. The orbits of such flows also arise in pseudo-periodic topology, as hyperplane sections of periodic manifolds (cf. [arnol1991topological, zorich1999leaves]). Moreover, any Hamiltonian flows on compact surfaces are examples of area-preserving flows. The difference between Hamiltonian and area-preserving flows on closed surfaces can be represented by harmonic flows, which are generated by the dual vector fields of harmonic one-forms. The topological invariants of Hamiltonian flows with finitely many singular points on compact surfaces are constructed from integrable systems points and dynamical systems points of views, and the structural stability are characterized [bolsinov1999exact, Nikolaenko20, Oshemkov10, sakajo2015transitions, sakajo2018tree, yokoyama2013word]. On the other hand, any area-preserving flows are non-wandering flows. Non-wandering flow on surfaces are classified and decomposed into elementary cells under finite existence of singular points, and are topologically characterized, and the topological invariants are constructed [cobo2010flows, nikolaev1998finite, nikolaev2001non, yokoyama2016topological, yokoyama2017decompositions, yokoyama2017genericity].

In this paper, we describe equivalence and difference for continuous flows on compact surfaces among Hamiltonian, divergence-free, and non-wandering properties. To state precisely, recall some concepts. By a flow, we mean a continuous \mathbb{R}-action. A flow is non-wandering if any points are non-wandering. A flow on a surface is Hamiltonian (resp. area-preserving, divergence-free) if it is topologically equivalent to the flow generated by a smooth Hamiltonian (resp. area-preserving, divergence-free) vector field. Notice that each of the Hamiltonian, area-preserving, divergence-free properties is not preserved by topological equivalence for vector fields but is preserved by one for flows. We consider the following question.

Question 1.

Is there a difference between area-preserving flows and non-wandering flows on compact surfaces?

We answer negatively under the finite existence of singular points. In other words, we have the following equivalence among non-wandering, area-preserving, and divergence-free flows.

Theorem A.

The following are equivalent for a flow with finitely many singular points on a compact surface:
(1) The flow is is non-wandering.
(2) The flow is is area-preserving.
(3) The flow is is divergence-free.

We also ask the following similar question, which is posed by M. Asaoka [asaoka2016] and is a motivative question of Question 1.

Question 2.

What is a difference between Hamiltonian flows and non-wandering flows on surfaces?

To answer this question, we characterize a Hamiltonian flow under the finite existence of singular points as follows.

Theorem B.

The following are equivalent for a flow with finitely many singular points on an orientable compact surface SS:
(1) The flow vv is Hamiltonian.
(2) The flow vv is non-wandering, there are no locally dense orbits, and the extended orbit space S/vexS/v_{\mathrm{ex}} is a directed graph without directed cycles.

Here the extended orbit space S/vex{S}/{v_{\mathrm{ex}}} is a quotient space S/S/\sim defined by xyx\sim y if either xx and yy are contained a multi-saddle connection or there is an orbit which contains xx and yy but is not contained in any multi-saddle connections. The non-existence of directed cycles can not be replaced by the non-existence of closed transversals in the previous theorem (see the third example as in Figure 7 in §5). On the sphere, there is no difference between Hamiltonian flows and non-wandering flows under the finite existence of singular points as follows.

Corollary C.

Let vv be a flow with finitely many singular points on a compact surface SS contained in a sphere. The following statements are equivalent:
(1)(1) The flow vv is non-wandering.
(2)(2) The flow vv is area-preserving.
(3)(3) The flow vv is divergence-free.
(4)(4) The flow vv is Hamiltonian.

The present paper consists of five sections. In the next section, as preliminaries, we introduce fundamental concepts. In §3, we characterize a Hamiltonian flow with finitely many singular points on a compact surface. In §4, we show the equivalence among non-wandering flows, area-preserving flows, and divergence-free flows on compact surfaces under the finite existence of singular points. In the final section, examples of non-Hamiltonian area-preserving flows are described.

2. Preliminaries

2.1. Notion of topology

Denote by A¯\overline{A} the closure of a subset AA of a topological space, by intA\mathrm{int}A the interior of AA, and by A:=A¯intA\partial A:=\overline{A}-\mathrm{int}A the boundary of AA, where BCB-C is used instead of the set difference BCB\setminus C when BCB\subseteq C. We define the border A\partial^{-}A by AintAA-\mathrm{int}A of AA. A boundary component of a subset AA is a connected component of the boundary of AA. A subset is locally dense if its closure has a nonempty interior. A boundary component of a subset AA is a connected component of the boundary of AA. By a surface, we mean a two dimensional paracompact manifold, that does not need to be orientable.

2.1.1. Directed topological graphs

A directed topological graph is a topological realization of a 1-dimensional simplicial complex with a directed structure on edges. A directed cycle in a directed topological graph is an embedded cycle whose edges are oriented in the same direction.

2.1.2. Curves and arc

A curve is a continuous mapping C:IXC:I\to X where II is a non-degenerate connected subset of a circle 𝕊1\mathbb{S}^{1}. A curve is simple if it is injective. We also denote by CC the image of a curve CC. Denote by C:=C(I)\partial C:=C(\partial I) the boundary of a curve CC, where I\partial I is the boundary of I𝕊1I\subset\mathbb{S}^{1}. Put intC:=CC\mathrm{int}C:=C\setminus\partial C. A simple curve is a simple closed curve if its domain is 𝕊1\mathbb{S}^{1} (i.e. I=𝕊1I=\mathbb{S}^{1}). An arc is a simple curve whose domain is a non-degenerate interval.

2.1.3. Reeb graph of a function on a topological space

For a function f:Xf\colon X\to\mathbb{R} on a topological space XX, the Reeb graph of a function f:Xf\colon X\to\mathbb{R} on a topological space XX is a quotient space X/ReebX/\sim_{\mathrm{Reeb}} defined by xReebyx\sim_{\mathrm{Reeb}}y if there are a number cc\in\mathbb{R} and a connected component of f1(c)f^{-1}(c) which contains xx and yy.

2.2. Notion of dynamical systems

We recall some basic notions. A good reference for most of what we describe is a book by S. Aranson, G. Belitsky, and E. Zhuzhoma [aranson1996introduction].

By a flow, we mean a continuous \mathbb{R}-action on a surface. Let v:×SSv\colon\mathbb{R}\times S\to S be a flow on a compact surface SS. Then vt:=v(t,)v_{t}:=v(t,\cdot) is a homeomorphism on SS. For tt\in\mathbb{R}, define vt:SSv_{t}:S\to S by vt:=v(t,)v_{t}:=v(t,\cdot). For a point xx of SS, we denote by O(x)O(x) the orbit of xx (i.e. O(x):={vt(x)t}O(x):=\{v_{t}(x)\mid t\in\mathbb{R}\}). An orbit arc is a non-degenerate arc contained in an orbit. A subset of SS is said to be invariant (or saturated) if it is a union of orbits. The saturation Satv(A)=v(A)\mathrm{Sat}_{v}(A)=v(A) of a subset ASA\subseteq S is the union of orbits intersecting AA. A point xx of SS is singular if x=vt(x)x=v_{t}(x) for any tt\in\mathbb{R} and is periodic if there is a positive number T>0T>0 such that x=vT(x)x=v_{T}(x) and xvt(x)x\neq v_{t}(x) for any t(0,T)t\in(0,T). A point is closed if it is either singular or periodic. Denote by Sing(v)\mathop{\mathrm{Sing}}(v) (resp. Per(v)\mathop{\mathrm{Per}}(v)) the set of singular (resp. periodic) points. A point is wandering if there are its neighborhood UU and a positive number NN such that vt(U)U=v_{t}(U)\cap U=\emptyset for any t>Nt>N. Then such a neighborhood is called a wandering domain. A point is non-wandering if it is not wandering (i.e. for any its neighborhood UU and for any positive number NN, there is a number tt\in\mathbb{R} with |t|>N|t|>N such that vt(U)Uv_{t}(U)\cap U\neq\emptyset).

For a point xSx\in S, define the ω\omega-limit set ω(x)\omega(x) and the α\alpha-limit set α(x)\alpha(x) of xx as follows: ω(x):=n{vt(x)t>n}¯\omega(x):=\bigcap_{n\in\mathbb{R}}\overline{\{v_{t}(x)\mid t>n\}}, α(x):=n{vt(x)t<n}¯\alpha(x):=\bigcap_{n\in\mathbb{R}}\overline{\{v_{t}(x)\mid t<n\}}. For an orbit OO, define ω(O):=ω(x)\omega(O):=\omega(x) and α(O):=α(x)\alpha(O):=\alpha(x) for some point xOx\in O. Note that an ω\omega-limit (resp. α\alpha-limit) set of an orbit is independent of the choice of a point in the orbit. A separatrix is a non-singular orbit whose α\alpha-limit or ω\omega-limit set is a singular point.

A point xx of SS is Poisson stable (or strongly recurrent) if xω(x)α(x)x\in\omega(x)\cap\alpha(x). A point xx of SS is recurrent if xω(x)α(x)x\in\omega(x)\cup\alpha(x). An orbit is singular (resp. periodic, closed, non-wandering, recurrent, Poisson stable) if it consists of singular (resp. periodic, closed, non-wandering, recurrent, Poisson stable) points. A flow is non-wandering (resp. recurrent) if each point is non-wandering (resp. recurrent). A quasi-minimal set is an orbit closure of a non-closed recurrent orbit. The closure of a non-closed recurrent orbit is called a Q-set.

2.2.1. Orbit classes and quotient spaces of flows

The (orbit) class O^\hat{O} of an orbit OO is the union of orbits each of whose orbit closure corresponds to O¯\overline{O} (i.e. O^={yXO(y)¯=O¯}\hat{O}=\{y\in X\mid\overline{O(y)}=\overline{O}\}). For a flow vv on a topological space XX, the orbit space T/vT/v (resp. orbit class space T/v^T/\hat{v}) of an invariant subset TT of XX is a quotient space T/T/\sim defined by xyx\sim y if O(x)=O(y)O(x)=O(y) (resp. O(x)¯=O(y)¯\overline{O(x)}=\overline{O(y)}). Notice that an orbit space T/vT/v is the set {O(x)xT}\{O(x)\mid x\in T\} as a set. Moreover, the orbit class space T/v^T/\hat{v} is the set {O^(x)xT}\{\hat{O}(x)\mid x\in T\} with the quotient topology.

2.2.2. Topological properties of orbits

An orbit is proper if it is embedded, locally dense if its closure has a nonempty interior, and exceptional if it is neither proper nor locally dense. A point is proper (resp. locally dense, exceptional) if its orbit is proper (resp. locally dense, exceptional). Denote by LD(v)\mathrm{LD}(v) (resp. E(v)\mathrm{E}(v), P(v)\mathrm{P}(v)) the union of locally dense orbits (resp. exceptional orbits, non-closed proper orbits). Note that an orbit on a paracompact manifold (e.g. a surface) is proper if and only if it has a neighborhood in which the orbit is closed [yokoyama2019properness]. This implies that a non-recurrent point is proper and so that a non-proper point is recurrent. In [cherry1937topological, Theorem VI], Cherry showed that the closure of a non-closed recurrent orbit OO of a flow on a manifold contains uncountably many non-closed recurrent orbits whose closures are O¯\overline{O}. This means that each non-closed recurrent orbit of a flow on a manifold has no neighborhood in which the orbit is closed, and so is not proper. In particular, a non-closed proper orbit is non-recurrent. Therefore the union P(v)\mathrm{P}(v) of non-closed proper orbits is the set of non-recurrent points and that R(v)=LD(v)E(v)\mathrm{R}(v)=\mathrm{LD}(v)\sqcup\mathrm{E}(v), where \sqcup denotes a disjoint union. Hence we have a decomposition S=Sing(v)Per(v)P(v)R(v)S=\mathop{\mathrm{Sing}}(v)\sqcup\mathop{\mathrm{Per}}(v)\sqcup\mathrm{P}(v)\sqcup\mathrm{R}(v).

2.2.3. Types of singular points

A point xx is a center if for any its neighborhood UU, there is an invariant open neighborhood VUV\subset U of xx such that U{x}U-\{x\} is an open annulus which consists of periodic orbits as in the left on Figure 1, where ABA-B is used instead of ABA\setminus B when ABA\subseteq B. A \partial-kk-saddle (resp. kk-saddle) is an isolated singular point on (resp. outside of) S\partial S with exactly (2k+2)(2k+2)-separatrices, counted with multiplicity as in Figure 1.

Refer to caption
Figure 1. A center and examples of multi-saddles

A multi-saddle is a kk-saddle or a \partial-(k/2)(k/2)-saddle for some k0k\in\mathbb{Z}_{\geq 0}. A 11-saddle is topologically an ordinary saddle and a \partial-(1/2)(1/2)-saddle is topologically a \partial-saddle.

2.2.4. Multi-saddle connection diagrams

The multi-saddle connection diagram D(v)D(v) is the union of multi-saddles and separatrices between multi-saddles. A multi-saddle connection is a connected component of the multi-saddle connection diagram. Note that a multi-saddle connection is also called a poly-cycle.

2.2.5. Flow boxes, periodic annuli, transverse annuli, and periodic invariant subsets

A subset UU is a flow box if there are bounded non-degenerate intervals I,JI,J and a homeomorphism h:I×JUh\colon I\times J\to U such that the image h(I×{t})h(I\times\{t\}) for any tJt\in J is an orbit arc as in the left on Figure 2, a open (resp. closed) periodic annulus is an open (resp. closed) annulus consisting of periodic orbits as in the middle on Figure 2, and an open (resp. closed) transverse annulus is an open (resp. closed) annulus contained as in the right on Figure 2.

Refer to caption
Figure 2. An open trivial flow box, an open periodic annulus, and an open transverse annulus

A periodic torus (resp. Klein bottle, Möbius band) is a torus (resp. Klein bottle, Möbius band) consisting of periodic orbits. Pasting one center (resp. two centers) with boundary components of an open periodic annulus, we obtain an open center disk (resp. rotating sphere). Pasting one center with the boundary of an open periodic Möbius band, we obtain a rotating projective plane. In other words, a rotating sphere (resp. projective plane) is a flow on a sphere (resp. projective plane) that consists of two centers (resp. one center) and periodic orbits. An open (resp. closed) center disk is a flow on an open (resp. closed) disk that consists of a center and periodic orbits.

2.3. Characterizations of non-wandering flows

Recall a following characterization of non-wandering flow which is essentially based on [yokoyama2016topological, Theorem 2.5].

Lemma 2.1.

Let vv be a flow on a compact surface SS. The following statements are equivalent:
(1)(1) The flow vv is non-wandering.
(2)(2) intP(v)=\mathop{\mathrm{int}}\mathrm{P}(v)=\emptyset.
(3)(3) (intP(v))E(v)=(\mathop{\mathrm{int}}\mathrm{P}(v))\sqcup\mathrm{E}(v)=\emptyset (i.e.(\mathrm{i.e.} S=Cl(v)δP(v)LD(v))S=\mathop{\mathrm{Cl}}(v)\sqcup\delta\mathrm{P}(v)\sqcup\mathrm{LD}(v)).
(4)(4) S=Cl(v)LD(v)¯S=\overline{\mathop{\mathrm{Cl}}(v)\sqcup\mathrm{LD}(v)}.

In any case, we have P(v)Sing(v)={xSω(x)α(x)Sing(v)}\mathrm{P}(v)\sqcup\mathop{\mathrm{Sing}}(v)=\{x\in S\mid\omega(x)\cup\alpha(x)\subseteq\mathop{\mathrm{Sing}}(v)\}.

Proof.

Recall that S=Cl(v)P(v)LD(v)E(v)S=\mathop{\mathrm{Cl}}(v)\sqcup\mathrm{P}(v)\sqcup\mathrm{LD}(v)\sqcup\mathrm{E}(v) and that the union P(v)\mathrm{P}(v) is the set of points which are not recurrent. By the Maǐer theorem [mayer1943trajectories, markley1970number], the closure E(v)¯\overline{\mathrm{E}(v)} is a finite union of closures of exceptional orbits and so is nowhere dense. By [yokoyama2016topological, Lemma 2.3], the union P(v)E(v)\mathrm{P}(v)\sqcup\mathrm{E}(v) is a neighborhood of E(v)\mathrm{E}(v). This means that E(v)=\mathrm{E}(v)=\emptyset if intP(v)=\mathop{\mathrm{int}}\mathrm{P}(v)=\emptyset. Therefore conditions (2)(2), (3)(3), and (4)(4) are equivalent. By [yokoyama2016topological, Theorem 2.5], conditions (1)(1) and (4)(4) are equivalent. Suppose that vv is non-wandering. [yokoyama2016topological, Proposition 2.6] implies P(v)Sing(v)={xSω(x)α(x)Sing(v)}\mathrm{P}(v)\sqcup\mathop{\mathrm{Sing}}(v)=\{x\in S\mid\omega(x)\cup\alpha(x)\subseteq\mathop{\mathrm{Sing}}(v)\}. ∎

2.4. Fundamental notion related to Hamiltonian flows on a compact surface

A CrC^{r} vector field XX for any r0r\in\mathbb{Z}_{\geq 0} on an orientable surface SS is Hamiltonian if there is a Cr+1C^{r+1} function H:SH\colon S\to\mathbb{R} such that dH=ω(X,)dH=\omega(X,\cdot) as a one-form, where ω\omega is a volume form of SS. In other words, locally the Hamiltonian vector field XX is defined by X=(H/x2,H/x1)X=(\partial H/\partial x_{2},-\partial H/\partial x_{1}) for any local coordinate system (x1,x2)(x_{1},x_{2}) of a point pSp\in S. A flow is Hamiltonian if it is topologically equivalent to a flow generated by a smooth Hamiltonian vector field. Note that a volume form on an orientable surface is a symplectic form. It is known that a CrC^{r} (r1r\geq 1) Hamiltonian vector field on a compact surface is structurally stable with respect to the set of CrC^{r} Hamiltonian vector fields if and only if both each singular point is non-degenerate and each separatrix of a saddle is homoclinic, and each separatrix on a \partial-saddle connects a boundary component (see [ma2005geometric, Theorem 2.3.8, p. 74]).

By [cobo2010flows, Theorem 3], any singular points of a Hamiltonian vector field XX with finitely many singular points on a compact surface SS is either a center or a multi-saddle, and the Reeb graph of the Hamiltonian HH generating XX is a directed topological graph which is also a quotient space of the orbit space of the generating flow. In other words, the Reeb graph is obtained from the orbit space by collapsing multi-saddle connections into singletons. Note that the orbit space of a Hamiltonian flow is not a directed topological graph.

In the analogy of the collapse, we define the extended orbit to describe “reduced” orbit spaces as follows. By an extended orbit of a flow, we mean a multi-saddle connection or an orbit that is not contained in any multi-saddle connection. Denote by Oex(x)O_{\mathrm{ex}}(x) the extended orbit containing xx. The quotient space by extended orbits of a flow vv on a surface SS is called the extended orbit space and denoted by S/vex{S}/{v_{\mathrm{ex}}}.

2.4.1. Divergence-free flows, area-preserving flows, and multi-valued Hamiltonians

A CrC^{r} vector field XX for any r>0r\in\mathbb{Z}_{>0} on a surface with a Riemannian metric gg is area-preserving if the area form μ\mu is preserved under pullback (i.e. vt(μ)=μv_{t}^{*}(\mu)=\mu) by homeomorphisms vtv_{t} for any tRt\in R, where vv is the flow generated by XX. A CrC^{r} vector field XX for any r>0r\in\mathbb{Z}_{>0} is divergence-free if divX=0\mathrm{div}X=0, where divX:=dg(X,)\mathrm{div}X:=\mathop{*}d\mathop{*}g(X,\cdot). We call that a flow is area-preserving (resp. divergence-free) if it is topologically equivalent to a flow generated by a smooth area-preserving (resp. divergence-free) vector field. Note that the divergence divX\mathrm{div}X can be also determined by a condition divXμ=LXμ\mathrm{div}X\mu=L_{X}\mu where LXL_{X} is the Lie derivative along XX. Liouville’s theorem implies the following observation.

Lemma 2.2.

A flow on a surface is area-preserving if and only if it is a divergence-free.

Proof.

Let vv be the flow generated by a smooth vector field XX on a surface. Since dvtμ/dt|t=t0=vt0(LXμ)=vt0((divX)μ)dv_{t}^{*}\mu/dt|_{t=t_{0}}=v_{t_{0}}^{*}(L_{X}\mu)=v_{t_{0}}^{*}((\mathrm{div}X)\mu), we have that vtμ=μv_{t}^{*}\mu=\mu for any tt\in\mathbb{R} if only if divX=0\mathrm{div}X=0. ∎

A flow vXv_{X} generated by a vector field XX on a surface SS is a multi-valued Hamiltonian flow if there are finitely many pairs of charts (Ui;xi,yi)(U_{i};x_{i},y_{i}) with S=UiS=\bigcup U_{i} and smooth Hamiltonians HiH_{i} on the charts such that the flow vXv_{X} is a solution to the equations dxi/dt=Hi/yidx_{i}/dt=\partial H_{i}/\partial y_{i} and dyi/dt=Hi/xidy_{i}/dt=-\partial H_{i}/\partial x_{i} (i.e. X|Ui=(Hi/yi,Hi/xi)X|_{U_{i}}=(\partial H_{i}/\partial y_{i},-\partial H_{i}/\partial x_{i})). Then {Hi}i\{H_{i}\}_{i} is called a multi-valued Hamiltonian on SS.

3. Characterization of Hamiltonian surface flows with finitely many singular points on compact surfaces

Let vv be a flow on a compact connected surface SS. We have the following characterization of a non-wandering flow with finitely many singular points but without locally dense orbits.

Lemma 3.1.

The following statements are equivalent for a flow vv with finitely many singular points but without locally dense orbits on a compact connected surface SS:
(1)(1) The flow vv is non-wandering.
(2)(2) Each connected component of the complement SD(v)S-D(v) is either a rotating sphere, an open center disk, an open periodic annulus, a periodic torus, a periodic Klein bottle, a rotating projective plane, or a periodic Möbius band.

In any case, S=Per(v)¯S=\overline{\mathop{\mathrm{Per}}(v)}, the multi-saddle connection diagram D(v)D(v) consists of finitely many orbits, and the extended orbit space S/vexS/v_{\mathrm{ex}} is a finite directed topological graph.

Proof.

Condition (2) implies that S=Per(v)¯S=\overline{\mathop{\mathrm{Per}}(v)} and so that vv is non-wandering. Conversely, suppose that vv is non-wandering. By [yokoyama2016topological, Corollary 2.9], the periodic point set Per(v)\mathop{\mathrm{Per}}(v) is open. From Lemma 2.1, we have S=Cl(v)δP(v)S=\mathop{\mathrm{Cl}}(v)\sqcup\delta\mathrm{P}(v). [cobo2010flows, Theorem 3] implies that each singular point of a non-wandering flow with finitely many singular points on a compact surface is either a center or a multi-saddle. The finite existence of multi-saddle implies that the multi-saddle connection diagram D(v)D(v) consists of finitely many orbits. Then S=S(Sing(v)D(v))¯S=\overline{S-(\mathop{\mathrm{Sing}}(v)\cup D(v))}. The union Bd0(v)\mathop{\mathrm{Bd}}_{0}(v) defined in [yokoyama2017decompositions] is the union of singular points and separatrices between multi-saddles. This means that Bd0(v)\mathop{\mathrm{Bd}}_{0}(v) is the union of centers and the multi-saddle connection diagram D(v)D(v). [yokoyama2017decompositions, Theorem A] (or [katok1997introduction, nikolaev1999flows] or [nikolaev2001non, Proposition 2]) implies that each connected component of SBd0(v)S-\mathop{\mathrm{Bd}}_{0}(v) is either a periodic annulus, a periodic torus, a periodic Klein bottle, or a periodic Möbius band. Adding centers, each connected component of SD(v)S-D(v) is either a rotating sphere, an open center disk, an open periodic annulus, a periodic torus, a periodic Klein bottle, a rotating projective plane, or a periodic Möbius band. For any connected component UU of SD(v)S-D(v), the restriction U/vex=U/vU/v_{\mathrm{ex}}=U/v is an interval. Since any multi-saddle connections are singletons in S/vexS/v_{\mathrm{ex}}, the extended orbit space S/vexS/v_{\mathrm{ex}} is a finite directed topological graph. From S=S(Sing(v)D(v))¯S=\overline{S-(\mathop{\mathrm{Sing}}(v)\cup D(v))}, we have S=Per(v)¯S=\overline{\mathop{\mathrm{Per}}(v)}. This means that vv is non-wandering. ∎

We show that non-wandering flows with finitely many singular points are divergence-free.

Lemma 3.2.

Let vv be a flow with finitely many singular points on a compact connected surface SS. The following statements are equivalent:
(1)(1) The flow vv is a non-wandering flow without locally dense orbits.
(2)(2) The flow vv is a divergence-free flow without locally dense orbits.
Moreover, if SS is orientable and the extended orbit space S/vexS/v_{\mathrm{ex}} has no directed cycle as a directed graph, then the following condition is equivalent to one of the above conditions:
(3)(3) The flow vv is Hamiltonian.

Proof.

We may assume that SS is connected. Since Hamiltonian vector fields are divergence-free and the existence of a Hamiltonian implies the non-existence of locally dense orbits, condition (3)(3) implies condition (2)(2). Suppose that vv is a divergence-free flow without locally dense orbits. Since any divergence-free vector fields on compact surfaces have no wandering domains, the flow vv is non-wandering. This means that condition (2)(2) implies condition (1)(1).

Conversely, suppose that vv is a non-wandering flow without locally dense orbits. Then there are no limit circuits. From Lemma 3.1, we have that S=Cl(v)δP(v)S=\mathop{\mathrm{Cl}}(v)\sqcup\delta\mathrm{P}(v) and that each singular point of vv is either a center or a multi-saddle. Notice that if SS is orientable and S/vexS/v_{\mathrm{ex}} has no directed cycle as a directed graph, then multi-valued Hamiltonians can be chosen as single-valued Hamiltonians.

Suppose that S=Per(v)S=\mathop{\mathrm{Per}}(v). Then vv is topologically equivalent to a flow given by multi-valued Hamiltonians. Since multi-valued Hamiltonian vector fields are divergence-free, the given flow is a flow generated by a divergence-free vector field. This means that the flow vv is divergence-free. If SS is orientable and S/vexS/v_{\mathrm{ex}} has no directed cycle as a directed graph then vv is Hamiltonian.

Suppose that S=Cl(v)S=\mathop{\mathrm{Cl}}(v). We may assume that there are singular points. The non-existence of non-recurrent points implies that any singular points are centers. Then vv is axisymmetric rotating on either a sphere, a closed disk, or a projective plane and so is topologically equivalent to a flow given by multi-valued Hamiltonians. This means that the flow vv is divergence-free. As above, if SS is orientable and S/vexS/v_{\mathrm{ex}} has no directed cycle as a directed graph, then vv is Hamiltonian.

Thus we may assume that SCl(v)S\neq\mathop{\mathrm{Cl}}(v). Then vv is not pointwise periodic and so D(v)D(v)\neq\emptyset. Lemma 3.1 implies that the complement SD(v)S-D(v) of the multi-saddle connection diagram consists of periodic annuli, open center disks, open Möbius bands each of whose orbit spaces is homeomorphic to an interval [0,1)[0,1). Since ll-saddle (resp. kk-\partial-saddle) can be realized by a Hamiltonian with l+1l+1 (resp.k\lceil k\rceil) centers as Figure 3, any flow restricted to a small simply connected open neighborhood of a multi-saddle is topologically equivalent to a flow generated by a smooth Hamiltonian vector field.

Refer to caption
Figure 3. Multi-saddles and their Hamiltonians

Since there are trivial flow boxes UiU_{i} along separatrices between multi-saddles, we can extend such a flow given by Hamiltonians near multi-saddles into a flow given by multi-valued Hamiltonians near the multi-saddle connection diagram. We can take UiU_{i} such that D(v)D(v) is homotopic to the union Ui\bigcup U_{i} and so the complement SUiS-\bigcup U_{i} is a disjoint union of invariant closed periodic annuli, invariant closed center disks, and invariant closed Möbius bands each of whose orbit spaces is homeomorphic to a closed interval. Since the restrictions of vv into these connected components of SUiS-\bigcup U_{i} are topologically equivalent to flows generated by multi-valued Hamiltonian vector fields, we can construct a flow given by multi-valued Hamiltonians. This means that the flow vv is divergence-free. As above, if SS is orientable and S/vexS/v_{\mathrm{ex}} has no directed cycle as a directed graph, then vv is Hamiltonian. ∎

The previous theorem implies an answer, which is Theorem B, for a question when a non-wandering flow with finitely many singular points on an orientable compact connected surface becomes Hamiltonian as follows.

3.1. Proof of Theorem B

Suppose that vv is Hamiltonian. Let H:SH\colon S\to\mathbb{R} be the Hamiltonian of vv and π:SS/vex\pi\colon S\to S/v_{\mathrm{ex}}. Then there is a quotient map p:S/vexp\colon S/v_{\mathrm{ex}}\to\mathbb{R} with H=pπH=p\circ\pi. This means that there are no directed cycles in S/vexS/v_{\mathrm{ex}}. Lemma 3.2 implies the converse.

4. Correspondence of non-wandering flows and divergence-free flows on compact surfaces under finiteness of singular points

Recall the following fact.

Lemma 4.1.

Let vv be a flow on a compact surface. For any locally dense orbit, there is a closed transverse intersecting it.

For self-containedness, we state the following proof.

Proof.

Fix a locally dense orbit OO. Let CC be a closed orbit arc in OO from xx to yy and JJ a closed transverse arc with CJ=C=J={x,y}C\cap J=\partial C=\partial J=\{x,y\}. If the first return map on JJ is orientable near J\partial J along CC, then apply the waterfall construction to the simple closed curve CJC\cup J (see Figure 4), and so we have a closed transversal which intersects OO and is near CJC\cup J.

Refer to caption
Figure 4. The waterfall construction

Thus we may assume that the first return map on JJ is not orientable near J\partial J. Take a closed orbit arc CC^{\prime} from yy to zz with CJ=C={y,z}C^{\prime}\cap J=\partial C^{\prime}=\{y,z\}. Let [y,z][y,z] be the sub-arc in JJ and [z,x][z,x] the sub-arc in JJ such that J=[y,z][z,x]J=[y,z]\cup[z,x] and [y,z][z,x]={z}[y,z]\cap[z,x]=\{z\}. If the return map on JJ is orientable near {y,z}\{y,z\} along CC^{\prime}, then the waterfall construction to the simple closed curve C[y,z]C^{\prime}\cup[y,z] implies a closed transversal that intersects OO and is near C[y,z]C^{\prime}\cup[y,z]. If the return map on JJ is not orientable near {y,z}\{y,z\} along CC^{\prime}, then the waterfall construction to the simple closed curve (CC)[z,x](C\cup C^{\prime})\cup[z,x] implies a closed transversal which intersects OO and is near (CC)[z,x](C\cup C^{\prime})\cup[z,x]. ∎

Recall that a periodic orbit is one-sided if it is either a boundary component of a surface or has a small neighborhood which is a Möbius band.

4.1. Proof of Theorem A

Let vv be a flow with finitely many singular points on a compact connected surface SS. Lemma 2.2 implies the equivalence between area-preserving property and divergence-free property for vv. Suppose that vv is divergence-free. [cobo2010flows, Theorem 3] implies that each singular point is either a center or a multi-saddle. Since any divergence-free vector fields on compact surfaces have no wandering domains, the flow vv is non-wandering.

Conversely, suppose that vv is non-wandering. Lemma 2.1 implies the non-existence of exceptional orbits of vv. The smoothing theorem [gutierrez1986smoothing] implies that vv is topologically equivalent to a CC^{\infty}-flow. It is known that a total number of Q-sets for vv cannot exceed gg if SS is an orientable surface of genus gg [mayer1943trajectories], and p12\frac{p-1}{2} if SS is a non-orientable surface of genus pp [markley1970number] (cf. Remark 2 [aranson1996maier]). There are finitely many distinct Q-sets. By induction of the number of Q-sets, we will show the assertion. If there are no Q-sets, then Lemma 3.2 implies that vv is divergence-free. Suppose that there are Q-sets. Fix a Q-set QQ. [cherry1937topological, Theorem VI] implies that QQ contains infinitely many Poisson stable orbits. Fix a Poisson stable orbit OO in QQ. By Lemma 4.1, there is a closed transversal γQ\gamma\subset Q intersecting OO. Then γ=γO¯\gamma=\overline{\gamma\cap O}. Let ZZ be a smooth vector field whose generating flow is topological equivalent to vv.

We claim that the first return map PZ:γγP_{Z}\colon\gamma\to\gamma of ZZ is topologically conjugate to a minimal interval exchange transformation E:missingS1missingS1E\colon\mathbb{\mathbb{missing}}{S}^{1}\to\mathbb{\mathbb{missing}}{S}^{1} on the circle missingS1\mathbb{\mathbb{missing}}{S}^{1}. Indeed, by the structure theorem of Gutierrez [gutierrez1986smoothing] (see also [gutierrez1986smoothing, Lemma 3.9] and [nikolaev1999flows, Theorem 2.5.1]), for the closed transversal γ\gamma for the Q-set QQ, the first return map PZ:γγP_{Z}\colon\gamma\to\gamma is well-defined and topologically semi-conjugate to a minimal interval exchange transformation E:missingS1missingS1E\colon\mathbb{\mathbb{missing}}{S}^{1}\to\mathbb{\mathbb{missing}}{S}^{1}. Since QQ is locally dense and contains γ\gamma, the first return map PZ:γγP_{Z}\colon\gamma\to\gamma is topologically conjugate to the minimal interval exchange transformation EE.

Therefore we may assume that the first return map Pv:γγP_{v}\colon\gamma\to\gamma by vv is a minimal interval exchange transformation. Fix an interval exchange transformation E:missingS1missingS1E^{\prime}\colon\mathbb{\mathbb{missing}}{S}^{1}\to\mathbb{\mathbb{missing}}{S}^{1} such that EEE^{\prime}\circ E has periodic orbits. Denote by I1,I2,,IkI_{1},I_{2},\ldots,I_{k} the intervals of interval exchange transformation EEE^{\prime}\circ E. Let did_{i} be the length of Ii𝕊1I_{i}\subset\mathbb{S}^{1} with d1+d2++dk=1d_{1}+d_{2}+\cdots+d_{k}=1. Fix an open transverse annulus AQA\subset Q which is a neighborhood of γ\gamma and contains no singular points. Then we also may assume that the restriction of ZZ to AA can be identified with a vector field (1,0)(1,0) on [0,1]×γ[0,1]\times\gamma by a homeomorphism. Consider a smooth divergence-free vector field ZAZ_{A} on A=[0,1]×γA=[0,1]\times\gamma which depends only on γ\gamma and which is a periodic annulus such that the first return map on γ\gamma by the flow vZv_{Z^{\prime}} generated by the resulting vector field Z:=Z+ZAZ^{\prime}:=Z+Z_{A} is topological equivalent to the interval exchange transformation EE:missingS1missingS1E^{\prime}\circ E\colon\mathbb{\mathbb{missing}}{S}^{1}\to\mathbb{\mathbb{missing}}{S}^{1}. Then vZv_{Z^{\prime}} has a periodic orbit OAO_{A} intersecting γ\gamma and OO. Fix the corresponding periodic orbit O1O_{1} of ZZ intersecting γ\gamma. Denote by IiγI^{\prime}_{i}\subset\gamma the intervals of PZP_{Z^{\prime}} corresponding to IiI_{i}. Then the saturations of any intervals IiI^{\prime}_{i} of PZP_{Z^{\prime}} are invariant open periodic annuli or Möbius bands AiA^{\prime}_{i}. Since any interval exchange transformation preserves the Lebesgue measure and so non-wandering, the flow vZv_{Z^{\prime}} generated by ZZ^{\prime} is non-wandering. By the inductive hypothesis, the flow vZv_{Z^{\prime}} is divergence-free. By construction, vZ|Sv(γ)=vZ|Sv(γ)v_{Z}|_{S-v(\gamma)}=v_{Z^{\prime}}|_{S-v(\gamma)}. Fix a divergence-free vector field WW whose generating flow vWv_{W} is topologically equivalent to vZv_{Z^{\prime}} and a closed transversal γW\gamma_{W} with respect to vWv_{W} which corresponds to γ\gamma. Then the first return map PW:γWγWP_{W}\colon\gamma_{W}\to\gamma_{W} is topologically conjugate to PZP_{Z^{\prime}}. If AiA_{i} is an open Möbius band, then the orbit space Ai/vWA_{i}/v_{W} for the flow vWv_{W} generated by WW is homeomorphic to a half open interval [0,1)[0,1) such that the periodic orbit corresponding to 0 is one-sided and denote by OAiO_{A_{i}}. Denote by DiD_{i} be the differences of the values the boundary components of the open invariant annulus AiA_{i} with respect to the multi-valued Hamiltonians Hi:AiH_{i}\colon A_{i}\to\mathbb{R} if AiA_{i} is an annulus, and by DiD_{i} the twice of the differences of the values the boundary components of the open invariant annulus AiOAiA_{i}-O_{A_{i}} with respect to the multi-valued Hamiltonians Hi:AiOAiH_{i}\colon A_{i}-O_{A_{i}}\to\mathbb{R}. Put D:=D1+D2++DkD:=D_{1}+D_{2}+\cdots+D_{k}. Write 𝔸i:=Ai\mathbb{A}_{i}:=A_{i} if AiA_{i} is an annulus, and 𝔸i:=AiOAi\mathbb{A}_{i}:=A_{i}-O_{A_{i}} if AiA_{i} is Möbius band.

We claim that we may assume that D1/d1=D2/d2==Dk/dkD_{1}/d_{1}=D_{2}/d_{2}=\cdots=D_{k}/d_{k} by deformations. Indeed, by the topological equivalence between PwP_{w} and EEE^{\prime}\circ E, the saturations of any intervals of the interval exchange transformation PWP_{W} are also invariant open periodic annuli 𝔸i\mathbb{A}_{i}. Identify annuli 𝔸i\mathbb{A}_{i} with an embedded annuli {(x,Hi(x))x𝔸i}\{(x,H_{i}(x))\mid x\in\mathbb{A}_{i}\} of 𝔸i×\mathbb{A}_{i}\times\mathbb{R}. Let 𝔸i\mathbb{A}^{\prime}_{i} be an invariant open annulus with 𝔸i¯𝔸i\overline{\mathbb{A}^{\prime}_{i}}\subset\mathbb{A}_{i} and 𝔸i+,𝔸i\mathbb{A}^{\prime}_{i+},\mathbb{A}^{\prime}_{i-} the connected components of 𝔸i𝔸i\mathbb{A}_{i}-\mathbb{A}^{\prime}_{i}. Then the set difference 𝔸i𝔸i\mathbb{A}_{i}-\mathbb{A}^{\prime}_{i} is a disjoint union of two invariant annuli. Removing 𝔸i\mathbb{A}^{\prime}_{i}, taking translation 𝔸i+′′\mathbb{A}^{\prime\prime}_{i+} of 𝔸i+\mathbb{A}^{\prime}_{i+} with respect to the heights of HiH_{i}, and interpolating between 𝔸i+′′\mathbb{A}^{\prime\prime}_{i+} and 𝔸i\mathbb{A}^{\prime}_{i-} as in Figure 5, the resulting annuli 𝔸i′′\mathbb{A}^{\prime\prime}_{i} have the arbitrarily differences of the values the boundary components of 𝔸i′′\mathbb{A}^{\prime\prime}_{i} with respect to the height.

Refer to caption
Figure 5. Removing 𝔸i\mathbb{A}^{\prime}_{i}, taking translation 𝔸i+′′\mathbb{A}^{\prime\prime}_{i+} of 𝔸i+\mathbb{A}^{\prime}_{i+} with respect to the height of HiH_{i}, and interpolating between 𝔸i+′′\mathbb{A}^{\prime\prime}_{i+} and 𝔸i\mathbb{A}^{\prime}_{i-}.

Therefore we can choose the differences of the values the boundary components of 𝔸i′′\mathbb{A}^{\prime\prime}_{i} such that D1/d1=D2/d2==Dk/dkD_{1}/d_{1}=D_{2}/d_{2}=\cdots=D_{k}/d_{k}. Since the constructions preserve topological equivalence, the claim is completed.

We claim that we may assume that there are a closed transversal μ\mu with vZ(μ)=vZ(γ)v_{Z^{\prime}}(\mu)=v_{Z^{\prime}}(\gamma) and an open annular neighborhood V0V_{0} of μ\mu which can be isometrically embedded into (1,1)×/×/D(-1,1)\times\mathbb{R}/\mathbb{Z}\times\mathbb{R}/D\mathbb{Z} such that V0¯v(μ)\overline{V_{0}}\subset v(\mu) and that Z|V0=(d,0,0)Z^{\prime}|_{V_{0}}=(d,0,0) for some positive number dd\in\mathbb{R}, by deformation. Indeed, fix a small open annular neighborhood VV of γW\gamma_{W} which is a closed transverse annulus such that V¯v(γW)\overline{V}\subset v(\gamma_{W}). For a point tγWt\in\gamma_{W}, denote by CtC_{t} the orbit arc in VV containing tt. Identify VV with {(x,H(x))xV}\{(x,H(x))\mid x\in V\}, where H:V/DH\colon V\to\mathbb{R}/D\mathbb{Z} is a function defined by Hi:𝔸iH_{i}\colon\mathbb{A}_{i}\to\mathbb{R}. Cutting VV along the closed transversal γW\gamma_{W}, inserting a closed annulus V0:=[0.1,0,1]×{(t,Dt)t/}V_{0}:=[-0.1,0,1]\times\{(t,Dt)\mid t\in\mathbb{R}/\mathbb{Z}\} into the resulting two closed annuli VV_{-} and V+V_{+}, and interpolate between VV_{-} and V0V_{0} and between V+V_{+} and V0V_{0} as in Figure 6, we can obtain the resulting annuli UU.

Refer to caption
Figure 6. Cutting VV along the closed transversal γW\gamma_{W}, inserting a closed annulus V0V_{0} into the resulting two closed annuli VV_{-} and V+V_{+}, and interpolate between VV_{-} and V0V_{0} and between V+V_{+} and V0V_{0}.

Then μ:={0}×{(t,Dt)t/}WU\mu:=\{0\}\times\{(t,Dt)\mid t\in\mathbb{R}/\mathbb{Z}\}\subset W\subset U is a closed transversal. This completes the claim.

By construction, there is a divergence-free vector field WAW_{A} whose support is contained in V0V_{0} and that any non-singular orbit is a periodic orbit of form {x1}×{(t,Dt)t/}V0\{x_{1}\}\times\{(t,Dt)\mid t\in\mathbb{R}/\mathbb{Z}\}\subset V_{0} and WAW_{A} is depended only on the x1x_{1}-coordinate in V0V_{0}. Multiplying WAW_{A} by a scalar if necessary, we may assume that the first return map on μ\mu by W+WAW+W_{A} is topological conjugate to E:missingS1missingS1E\colon\mathbb{\mathbb{missing}}{S}^{1}\to\mathbb{\mathbb{missing}}{S}^{1} and so to PZ:γγP_{Z}\colon\gamma\to\gamma. This means that the flow generated by WA+WW_{A}+W is topological equivalent to vv. Recall that, for any divergence-free vector field YY, a vector field XX is divergence-free if and only if so is the sum X+YX+Y. This implies that the vector field WA+WW_{A}+W is divergence-free and so is the flow vv.

4.2. Proof of Corollary C

From the non-existence of non-closed recurrent orbit, Lemma 3.1 implies the quotient space 𝕊2/vex\mathbb{S}^{2}/v_{\mathrm{ex}} is a finite directed topological graph. Since the quotient space of a sphere is simply connected, the graph 𝕊2/vex\mathbb{S}^{2}/v_{\mathrm{ex}} has no directed circuit. Theorem A and Theorem B imply Corollary C.

5. Examples of non-Hamiltonian area-preserving flows

Irrational rotations on a torus are non-Hamiltonian area-preserving flows. Any flows on a torus consisting of periodic orbits are non-Hamiltonian area-preserving flows. The following example is a non-Hamiltonian area-preserving flow without closed transversals, which M. Asaoka suggested to show the necessity of the non-existence of directed cycles in Theorem B. Indeed, there is a smooth non-Hamiltonian area-preserving flow with non-degenerate singular points but without neither locally dense orbits nor closed transversals on a torus as in Figure 7.

Refer to caption
Figure 7. A smooth non-Hamiltonian area-preserving flow with non-degenerate singular points but without closed transversals on a torus which consists of two center, one 33-saddle, and periodic orbits.

Acknowledgement: The author would like to thank Masashi Asaoka for his motivative question and helpful comments.

References