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

On the barcode entropy of Lagrangian submanifolds

Matthias Meiwes Matthias Meiwes, School of Mathematical Sciences, Tel Aviv University, Ramat Aviv, Tel Aviv 69978, Israel. [email protected]
Abstract.

This article deals with relative barcode entropy, a notion that was recently introduced by Çineli, Ginzburg, and Gürel. We exhibit some settings in closed symplectic manifolds for which the relative barcode entropy of a Hamiltonian diffeomorphism and a pair of Lagrangian submanifolds is positive. In analogy to a result in the absolute case by the above authors, we obtain that the topological entropy of any horseshoe KK is a lower bound if the two Lagrangians contain a local unstable resp.  stable manifold in KK. In dimension 22, we also estimate the relative barcode entropy of a pair of closed curves that lie in special homotopy classes in the complement of certain periodic orbits in KK. Furthermore, we define a variant of relative barcode entropy and exhibit first examples for which it is positive. As applications, certain robustness features of the volume growth and the topological entropy are discussed.

The author was supported by the ERC Starting Grant 757585 and the Israel Science Foundation Grant 938/22

1. Introduction and main results

To study the asymptotic dynamics properties of a diffeomorphism ψ\psi on a manifold MM, one might look at the growth of specific quantities that are assigned to its iterates ψk\psi^{k}. As a natural example, we can consider the growth of volumes of submanifolds under iterations: we can fix a submanifold NN in MM and consider the growth of volumes of ψk(N)\psi^{k}(N); by looking at growth specifics that are independent of the metric, we might obtain some interesting dynamical quantities of ψ\psi “relative” to NN. Moreover, by simultaneously considering NN in a suitable family of submanifolds, this might yield some dynamical invariants assigned to ψ\psi itself. Especially interesting is the situation when there is exponential volume growth: due to results by Yomdin, [51], and Newhouse, [39], the supremum supNMΓ(ψ;N)\sup_{N\subset M}\Gamma(\psi;N) of the exponential volume growth rates

(1) Γ(ψ;N):=lim supklog(vol(ψk(N)))k\displaystyle\Gamma(\psi;N):=\limsup_{k\to\infty}\frac{\log(\mathrm{vol}(\psi^{k}(N)))}{k}

over all submanifolds NN, is identical to the topological entropy, htop(ψ)h_{\mathrm{top}}(\psi), of ψ\psi, provided ψ\psi is CC^{\infty} smooth. If ψ\psi is Hamiltonian, a related, though more abstract notion is relative barcode entropy, (ψ;L0,L1)\hbar(\psi;L_{0},L_{1}), defined for suitable pairs (L0,L1)(L_{0},L_{1}) of Lagrangian submanifolds, introduced and first studied by Çineli, Ginzburg, and Gürel in [13]. Relative barcode entropy provides a lower bound on volume growth, that is, (ψ;L0,L1)Γ(ψ;L0)\hbar(\psi;L_{0},L_{1})\leq\Gamma(\psi;L_{0}), [13]. In this article, we will discuss some natural settings for which (ψ;L0,L1)\hbar(\psi;L_{0},L_{1}) is bounded from below. But first, let me give some further background and motivation.

In the symplectic setting, various results have been obtained that deal with asymptotic dynamics features that are induced only by the symplectic topology of the underlying manifold, together with maybe some mild additional assumptions on the diffeomorphisms. Several results in this direction can be found in Polterovich’ paper [42]. As an example dealing with volume growth, it holds that for any non-identical Hamiltonian diffeomorphism on a closed surface with positive genus there exists a curve NN such that the length of ψk(N)\psi^{k}(N) grows at least linearly (for the case of the 22-torus, see also [45]). Subsequently, various results of this type, concerning polynomial and exponential growth of Lagrangian submanifolds, were obtained by Frauenfelder and Schlenk, [23, 24]. Topological entropy in this context has been studied in the contact setting, first for unit cotangent bundles in [31], and then in a variety of situations by Alves and other authors, see [1, 3, 5, 4, 6, 7, 35, 18]. In one form or the other, an important part of the reasoning in the works above is that the complexity of some type of Floer homology or contact homology on the underlying manifold serves as a “lower bound” on the complexity of the dynamics. The underlying theory that one considers might be “relative” (generators correspond to chords between Lagrangian/Legendrian submanifolds), as e.g. in [31, 7, 6], or “absolute” (generators correspond to periodic orbits), as e.g. in [3, 35].

As discovered in [13], and investigated further in [15, 25, 20, 16], this general line of reasoning can be partly reversed: certain complexity features of the dynamics of a diffeomorphism translate into growth of complexity of Floer data assigned to its iterations, into barcode entropy. This notion, introduced in [13] and which has again a relative and an absolute version, is defined naturally in the framework of persistence homology (for persistence homology in Floer theory see [47, 44, 48] or the book [43]). One assigns to a filtered Floer chain complex a barcode, a multiset of intervals (“bars”), where, roughly speaking, each bar captures a homology class that “persists” in the filtered homology for a certain filtration window, and the length of the bar corresponds to the length of that filtration window. Barcode entropy then measures the exponential growth rate of the number of bars, ignoring bars of very small size, for a sequence of Floer chain complexes associated to iterations of the diffeomorphism. More precisely, the relative barcode entropy (ψ;L0,L1)\hbar(\psi;L_{0},L_{1}) of a triple (ψ,L0,L1)(\psi,L_{0},L_{1}), where ψ\psi is a Hamiltonian diffeomorphism on a symplectic manifold MM, and (L0,L1)(L_{0},L_{1}) is a pair of Hamiltonian isotopic, monotone Lagrangian submanifolds in MM, is defined as

(ψ;L0,L1)=limϵ0lim supklog(bϵ(ψ;L0,L1,k))k,\hbar(\psi;L_{0},L_{1})=\lim_{\epsilon\to 0}\limsup_{k\to\infty}\frac{\log(b_{\epsilon}(\psi;L_{0},L_{1},k))}{k},

where bϵ(ψ;L0,L1,k)b_{\epsilon}(\psi;L_{0},L_{1},k) is the number of bars of length at least ϵ\epsilon in the Lagrangian Floer homology of (L0,ψk(L1))(L_{0},\psi^{-k}(L_{1}))111This definition makes sense if the pair (L0,L1)(L_{0},L_{1}) is non-degenerate, that is, if L0L_{0} intersects ψk(L1)\psi^{-k}(L_{1}) transversally for all kk. One can then naturally extend the definition to degenerate pairs, see §2.4.. Analogously, by considering the Hamiltonian Floer chain complexes of ψk\psi^{k}, one defines the absolute barcode entropy (ψ)\hbar(\psi) of Hamiltonian diffeomorphisms ψ\psi. Now, the following lower bound on the absolute barcode entropy (ψ)\hbar(\psi) in terms of the complexity of the dynamics of a Hamiltonian diffeomorphism ψ:MM\psi:M\to M is obtained in [13]. Given that ψ\psi has a compact hyperbolic invariant set KK in MM which is locally maximal, then (ψ)\hbar(\psi) is bounded from below by the topological entropy on KK, [13, Thm. B], that is,

(2) (ψ)htop(ψ|K).\displaystyle\hbar(\psi)\geq h_{\mathrm{top}}(\psi|_{K}).

On the other hand, some reverse inequalities are still satisfied. It holds that

(3) (ψ;L0,L1)Γ(ψ;L0)htop(ψ)\displaystyle\hbar(\psi;L_{0},L_{1})\leq\Gamma(\psi;L_{0})\leq h_{\mathrm{top}}(\psi)

(as long as (ψ;L0,L1)\hbar(\psi;L_{0},L_{1}) is defined), [13, Thm. A], and similarly, (ψ)\hbar(\psi) bounds htop(ψ)h_{\mathrm{top}}(\psi) from below. In dimension 22, [13, Thm. C], this leads to the intriguing identity

(4) (ψ)=htop(ψ).\displaystyle\hbar(\psi)=h_{\mathrm{top}}(\psi).

Inequalities (2) and (3) give rise to the following circle of questions about relative barcode entropy: Are there, in view of (2) for the absolute case, bounds that are reverse to that in (3)? For instance, given any ψ\psi with htop(ψ)>0h_{\mathrm{top}}(\psi)>0, does there exist a pair of Lagrangian submanifolds (L0,L1)(L_{0},L_{1}) such that (ψ;L0,L1)>0\hbar(\psi;L_{0},L_{1})>0? Or, a more approachable question: Assume we are in a setting where it is known that the volume growth Γ(ψ;L)\Gamma(\psi;L) is positive, can we find for the same ψ\psi also pairs (L0,L1)(L_{0},L_{1}) for which (ψ;L0,L1)>0\hbar(\psi;L_{0},L_{1})>0?

Apart from the fact that those questions arise naturally given the inequalities (2) and (3), some motivation for them comes from stability properties of barcode entropy with respect to the Hofer-distance and the γ\gamma-distance on Hamiltonian isotopy classes of Lagrangian submanifolds: relative barcode entropy (ψ;L0,L1)\hbar(\psi;L_{0},L_{1}), as a function in L0L_{0} and L1L_{1} varying in their Hamiltonian isotopy class, is lower semi-continuous with respect to those distances. And hence, in view of the first inequality in (3), lower estimates for relative barcode entropy might imply some interesting robustness statements for the volume growth of Lagrangians (see §1.3 for further discussion).

Some examples for which the relative barcode entropy is positive can already be derived directly from (2): by a well-known identification of Hamiltonian Floer homology in MM and Lagrangian Floer homology in the product M×MM\times M, one obtains the identity (ψ)=(id×ψ;Λ,Λ)\hbar(\psi)=\hbar(id\times\psi;\Lambda,\Lambda), where id×ψ:M×MM×Mid\times\psi:M\times M\to M\times M is the graph map and ΛM×M\Lambda\subset M\times M is the diagonal (cf. [13]). Examples obtained via that identification are quite specific. In particular, this procedure does not yield examples on manifolds different from symplectic products, and it is for instance not at all clear from the above whether there exist even one surface diffeomorphism ψ\psi and two closed curves L0,L1L_{0},L_{1} for which (ψ;L1,L2)>0\hbar(\psi;L_{1},L_{2})>0.

Let us now state the main results of this article. Throughout the article, we assume that the Lagrangian submanifolds are monotone with first Chern number at least 22, cf. §2. Moreover, we assume that the Lagrangian submanifolds L0L_{0} and L1L_{1} when dealing with (ψ;L0,L1)\hbar(\psi;L_{0},L_{1}) are Hamiltonian isotopic. We say that such pairs (L0,L1)(L_{0},L_{1}) are admissible.

1.1. Barcode entropy for Lagrangian submanifolds that contain local stable and unstable disks

To formulate the first theorem, we recall that a hyperbolic invariant set KK of a diffeomorphism ψ\psi is locally maximal if there exists a compact neighbourhood UU for which K=kψk(U)K=\bigcap_{k\in\mathbbm{Z}}\psi^{k}({U}); such a neighbourhood is called isolating. For every point qq in KK and δ>0\delta>0, we have local stable and unstable manifolds of size δ\delta, Wδs(q)W^{s}_{\delta}(q) and Wδu(q)W^{u}_{\delta}(q), see §3.2, which, for small enough δ\delta, are small Lagrangian disks passing through qq. The following might be considered as a relative version of inequality (3), that is, of Theorem B in [13].

Theorem 1.

Let KK be a locally maximal, topologically transitive hyperbolic set. If (L0,L1)(L_{0},L_{1}) is an admissible pair of Lagrangian submanifolds in MM with Wδu(q)L0 and Wδs(p)L1 for some q,p,K,δ>0W_{\delta}^{u}(q)\subset L_{0}\text{ and }W_{\delta}^{s}(p)\subset L_{1}\text{ for some }q,p,\in K,\delta>0, then

(ψ;L0,L1)htop(ψ|K).\hbar(\psi;L_{0},L_{1})\geq h_{\mathrm{top}}(\psi|_{K}).

Note that every locally maximal compact hyperbolic set contains a topologically transitive one with the same topological entropy. Note also that the condition to contain Wδu(q)W_{\delta}^{u}(q) or Wδs(p)W_{\delta}^{s}(p) is a local condition. In fact, given KK as above, it is possible to modify any Lagrangian LL to one for which this condition is satisfied, keeping all but a small set inside LL fixed.

Together with [13, Thm. A], cf. (3), and a well-known result by Katok about the approximation of topological entropy by that of horseshoes in dimension 2, [27], Theorem 1 provides the following result; it resembles the Yomdin-Newhouse identity (in dimension 22).

Corollary 2.

If ψ:MM\psi:M\to M is a Hamiltonian diffeomorphism on a closed symplectic surface, then

¯(ψ):=sup(ψ;L0,L1)=htop(ψ),\overline{\hbar}(\psi):=\sup\hbar(\psi;L_{0},L_{1})=h_{\mathrm{top}}(\psi),

where the supremum runs over all admissible pairs of closed curves (L0,L1)(L_{0},L_{1}) in MM.

Remark 1.1.

It was shown in [8] that, if MM is a closed surface, the topological entropy htop(ψ)h_{\mathrm{top}}(\psi) is lower semi-continuous with respect to the Hofer-distance on Ham(M,ω)\mathrm{Ham}(M,\omega). Hence, by Corollary 2 the same holds for the quantity ¯(ψ)\overline{\hbar}(\psi), cf. §1.4.

Remark 1.2.

It would be interesting to understand whether also in higher dimensions, positive topological entropy (htop(ψ)>0h_{\mathrm{top}}(\psi)>0) implies that there exist pairs (L0,L1)(L_{0},L_{1}) with (ψ;L0,L1)>0\hbar(\psi;L_{0},L_{1})>0. The analogous statement for the absolute barcode entropy does not hold as examples of Çineli show, [12].

1.2. Barcode entropy of closed curves in the complement of some horseshoe orbits

To motivate the next result, let me recall the following fact for surface diffeomorphisms with positive topological entropy. In general, for any continuous map f:MMf:M\to M on a compact surface MM and any set of periodic orbits 𝒪M\mathcal{O}\subset M of ff, we obtain an induced action on the fundamental group in the complement f#:π1(M𝒪)π1(M,𝒪)f_{\#}:\pi_{1}(M\setminus\mathcal{O})\to\pi_{1}(M,\setminus\mathcal{O}) (up to some choices). Furthermore, given a homotopy class α\alpha of closed curves in M𝒪M\setminus\mathcal{O}, we can consider the exponential growth rate Γ𝒪,α\Gamma_{\mathcal{O},\alpha} of f#n(α)f^{n}_{\#}(\alpha) in π1(M𝒪)\pi_{1}(M\setminus\mathcal{O}), in terms of the minimal length of words of generators necessary to represent f#n(α)f^{n}_{\#}(\alpha), see [28, §3.1.c]. This growth rate is a lower bound, provided that ff is Lipschitz, to the growth rate of the length of closed curves representing α\alpha. That is, for any closed curve LL with [L]=α[L]=\alpha we have that

Γ(f;L)Γ𝒪,α(f).\Gamma(f;L)\geq\Gamma_{\mathcal{O},\alpha}(f).

Let ff be now a C1+αC^{1+\alpha}-diffeomorphism with htop(f)>0h_{\mathrm{top}}(f)>0. By a result of Franks and Handel, [22], there exist a periodic orbit 𝒪\mathcal{O} and a homotopy class α\alpha of curves in the complement such that Γ𝒪,α(f)>0\Gamma_{\mathcal{O},\alpha}(f)>0. This result was improved by Alves and the author, [8]: For every 0<e<htop(f)0<e<h_{\mathrm{top}}(f), there exist 𝒪\mathcal{O} and α\alpha as above such that

Γ𝒪,α(f)>htop(f)e.\Gamma_{\mathcal{O},\alpha}(f)>h_{\mathrm{top}}(f)-e.

In particular, for any closed curve LL with [L]=α[L]=\alpha,

Γ(f;L)>htop(f)e.\Gamma(f;L)>h_{\mathrm{top}}(f)-e.

The following result states that if we allow 𝒪\mathcal{O} to be a finite union of periodic orbits, then we can replace in that result the exponential length growth rate by relative barcode entropy.

Theorem 3.

Let ψ:MM\psi:M\to M be a Hamiltonian diffeomorphism with htop(ψ)>0h_{\mathrm{top}}(\psi)>0 on a closed symplectic surface (M,ω)(M,\omega). Let 0<e<htop(ψ)0<e<h_{\mathrm{top}}(\psi). Then there exist a finite invariant set 𝒪M\mathcal{O}\subset M and homotopy classes α0,α1\alpha_{0},\alpha_{1} of closed curves in M𝒪M\setminus\mathcal{O} such that for any admissible pair (L0,L1)(L_{0},L_{1}) with [Li]=αi[L_{i}]=\alpha_{i}, i=0,1i=0,1,

(ψ;L0,L1)>htop(ψ)e.\hbar(\psi;L_{0},L_{1})>h_{\mathrm{top}}(\psi)-e.

1.3. Stability with respect to the Hofer- and the γ\gamma-distance

Two fundamental and remarkable norms defined on groups of Hamiltonian diffeomorphisms are the Hofer-norm Hofer\|\cdot\|_{\mathrm{Hofer}} and the spectral- (or γ\gamma-)norm γ()\gamma(\cdot). The former was first introduced by Hofer, [26], whereas the latter was introduced by Viterbo, [50], by means of generating functions, and by Schwarz, see [46], and Oh, see [41], using Floer theory. The Hofer-norm bounds the γ\gamma-norm from above, whereas the CkC^{k} norms are not bounded by those norms in general. We refer to §2.4 and to the above references for definitions of these norms. They naturally induce distances on the groups of Hamiltonian diffeomorphisms, and furthermore also on any given Hamiltonian isotopy class \mathcal{L} of Lagrangian submanifolds: the Hofer-distance and (exterior) γ\gamma-distance on \mathcal{L} is defined as

dHofer(L,L):=inf{ψHofer|ψHam(M,ω),ψ(L)=L},d_{\mathrm{Hofer}}(L,L^{\prime}):=\inf\{\|\psi\|_{\mathrm{Hofer}}\,|\,\psi\in\mathrm{Ham}(M,\omega),\psi(L)=L^{\prime}\},
dγ(L,L):=inf{γ(ψ)|ψHam(M,ω),ψ(L)=L},d_{\gamma}(L,L^{\prime}):=\inf\{\gamma(\psi)\,|\,\psi\in\mathrm{Ham}(M,\omega),\psi(L)=L^{\prime}\},

see also [29] and references therein.

A remarkable property of the relative barcode entropy is that (ψ;L0,L1)\hbar(\psi;L_{0},L_{1}) is lower semi-continuous with respect to dHoferd_{\mathrm{Hofer}} and dγd_{\gamma} in the pair (L0,L1)(L_{0},L_{1}) varying in their Hamiltonian isotopy class. Together with these properties and estimate (3), Theorem 1 has the following consequence.

Corollary 4.

Let ψ\psi be a Hamiltonian diffeomorphism on a closed symplectic manifold (M,ω)(M,\omega), and let \mathcal{L} be a Hamiltonian isotopy class of monotone Lagrangian submanifolds in MM. Assume that L0L_{0} in \mathcal{L} contains a local unstable manifold of a point in a locally maximal compact hyperbolic set KK with htop(ψ|K)>0h_{\mathrm{top}}(\psi|_{K})>0. Then, there exists a dγd_{\gamma}-open (and hence also dHoferd_{\mathrm{Hofer}}-open) neighbourhood 𝒰\mathcal{U} of L0L_{0} in \mathcal{L} such that for all L𝒰L\subset\mathcal{U},

Γ(ψ;L)>0.\Gamma(\psi;L)>0.
Remark 1.3.

The statement is of rather symplectic nature and it is not clear if it can be obtained without using methods from Floer theory or related methods. To illustrate that, we might just consider Lagrangians LL which are graphs of exact 11-forms dfdf under the identification of a neighbourhood of L0L_{0} with the unit co-disk bundle of L0L_{0}: In that case, dHofer(L,L0)d_{\mathrm{Hofer}}(L,L_{0}) and dγ(L,L0)d_{\gamma}(L,L_{0}) are bounded from above by maxfminf\max f-\min f, a quantity that does not control large oscillations of LL around L0L_{0}. In particular the C0C^{0}-, and the Hausdorff-distance of Lagrangians is not dγd_{\gamma}-continuous. Moreover, it is not hard to see that, given KK as above, every closed Lagrangian submanifold in \mathcal{L} can be perturbed in the Hofer distance to a Lagrangian submanifold L0L_{0} as in the corollary. It follows that there is a dHoferd_{\mathrm{Hofer}}-open and dense set of LL in \mathcal{L} with Γ(ψ;L)>0\Gamma(\psi;L)>0.

Remark 1.4.

In [14] it was shown that, under some assumptions, the volume of Lagrangian submanifolds is lower semi-continuous with respect to those distances. It would be interesting to know if this is also the case for Γ(ψ;L)\Gamma(\psi;L), at least in some situations.

The next statement is a corollary of Theorem 3. For that, the reader should keep in mind that a small dγd_{\gamma}-perturbation of a closed curve in a surface does not necessarily preserve its homotopy class in a complement of a fixed finite set of points.

Corollary 5.

Let (M,ω)(M,\omega) be a closed symplectic surface, ψ:MM\psi:M\to M a Hamiltonian diffeomorphism with htop(ψ)>e>0h_{\mathrm{top}}(\psi)>e>0, and \mathcal{L} a Hamiltonian isotopy class of monotone curves in MM. Then there exist a finite invariant set 𝒪\mathcal{O}, a homotopy class α\alpha of closed curves in M𝒪M\setminus\mathcal{O}, and a dγd_{\gamma}-open (and hence also dHoferd_{\mathrm{Hofer}}-open) neighbourhood 𝒰\mathcal{U} of the set of LL in \mathcal{L} with [L]=α[L]=\alpha such that for all L𝒰L^{\prime}\in\mathcal{U},

Γ(ψ;L)>htop(ψ)e.\Gamma(\psi;L^{\prime})>h_{\mathrm{top}}(\psi)-e.

1.4. Strong barcode entropy and some further examples

Finally, let me discuss some situations in which positive relative barcode entropy persists on a larger scale. We note that the positivity of barcode entropy for the situations considered so far had its origin in horseshoes, hence in some rather local dynamical features, and the property (ψ;L0,L1)>0\hbar(\psi;L_{0},L_{1})>0 is not invariant under Hamiltonian isotopies of the Lagrangian submanifolds in general. Also note, that while (ψ;L0,L1)\hbar(\psi;L_{0},L_{1}) is lower semi-continuous with respect to the Hofer- or the γ\gamma-distance in the pair (L0,L1)(L_{0},L_{1}), questions about stability properties in the diffeomorphism ψ\psi are more subtle. (One result in that direction, but which is obtained rather indirectly, is formulated in Remark 1.1.) In order to detect situations with stronger stability properties, it is convenient to consider a variant of relative barcode entropy. For R>0R>0, let us define the RR-strong relative barcode entropy of (ψ,L0,L1)(\psi,L_{0},L_{1}) as

𝖧R(ψ;L0,L1)=limR^RR^<Rlim supklog(bR^k(ψ;L0,L1,k))k.\mathsf{H}^{R}(\psi;L_{0},L_{1})=\lim_{\begin{subarray}{c}\hat{R}\to R\\ \hat{R}<R\end{subarray}}\limsup_{k\to\infty}\frac{\log(b_{\hat{R}k}(\psi;L_{0},L_{1},k))}{k}.

The following properties are either obvious or consequences of stability properties of barcodes with respect to dγd_{\gamma}, [29], see §2.4; strict monotonicity in RR (point (iii)) was pointed out to me by Çineli.

Proposition 1.5.

Let ψHam(M,ω)\psi\in\mathrm{Ham}(M,\omega), and (L0,L1)(L_{0},L_{1}) be an admissible pair. Then the following holds:

  1. (i)

    If (L0,L1)(L^{\prime}_{0},L^{\prime}_{1}) is admissible and LiL^{\prime}_{i} is Hamiltonian isotopic to LiL_{i}, i=0,1i=0,1, then for all R>0R>0,

    𝖧R(ψ;L0,L1)=𝖧R(ψ;L0,L1).\mathsf{H}^{R}(\psi;L^{\prime}_{0},L^{\prime}_{1})=\mathsf{H}^{R}(\psi;L_{0},L_{1}).
  2. (ii)

    If γ(ψ)R\gamma(\psi)\leq R, then HR(ψ;L0,L1)=0H^{R}(\psi;L_{0},L_{1})=0.

  3. (iii)

    HR(ψ;L0,L1)H^{R}(\psi;L_{0},L_{1}), if non-vanishing, is strictly monotone decreasing in RR. More precisely: If γ(ψ)>R>R>0\gamma(\psi)>R>{R^{\prime}}>0, then

    𝖧R(ψ;L0,L1)γ(ψ)Rγ(ψ)R𝖧R(ψ;L0,L1);\mathsf{H}^{{R^{\prime}}}(\psi;L_{0},L_{1})\geq\frac{\gamma(\psi)-{R^{\prime}}}{\gamma(\psi)-R}\mathsf{H}^{R}(\psi;L_{0},L_{1});

    in particular,

    (ψ;L0,L1)γ(ψ)γ(ψ)R𝖧R(ψ;L0,L1).\hbar(\psi;L_{0},L_{1})\geq\frac{\gamma(\psi)}{\gamma(\psi)-R}\mathsf{H}^{R}(\psi;L_{0},L_{1}).
  4. (iv)

    If dγ(ψ,ψ)<δd_{\gamma}(\psi,\psi^{\prime})<\delta, then for all R>δR>\delta,

    𝖧Rδ(ψ;L0,L1)𝖧R(ψ;L0,L1).\mathsf{H}^{R-\delta}(\psi^{\prime};L_{0},L_{1})\geq\mathsf{H}^{R}(\psi;L_{0},L_{1}).
  5. (v)

    𝖧nR(ψn;L0,L1)n𝖧R(ψ;L0,L1)\mathsf{H}^{nR}(\psi^{n};L_{0},L_{1})\geq n\mathsf{H}^{R}(\psi;L_{0},L_{1}), for all nn\in\mathbbm{N}.

Hence, given a Hamiltonian isotopy class of Lagrangian submanifolds \mathcal{L}, the strong barcode entropies 𝖧R(ψ;):=𝖧R(ψ;L,L)\mathsf{H}^{R}(\psi;\mathcal{L}):=\mathsf{H}^{R}(\psi;L,L), LL\subset\mathcal{L}, are well defined, and the relative barcode entropy is positive independent of the pair of Lagrangians in \mathcal{L} if 𝖧R(ψ;)>0\mathsf{H}^{R}(\psi;\mathcal{L})>0. A necessary condition for the latter is that Ham(M,ω)\mathrm{Ham}(M,\omega) has infinite dγd_{\gamma}-diameter, which for example does not hold for the standard n\mathbbm{CP}^{n}, [19]. To exhibit first examples with 𝖧R(ψ;)>0\mathsf{H}^{R}(\psi;\mathcal{L})>0, we consider the twisted products (M,ω)=(Σ×Σ,σσ)(M,\omega)=(\Sigma\times\Sigma,\sigma\oplus-\sigma), where (Σ,σ)(\Sigma,\sigma) is a closed symplectic surface with genus at least 22. That the dγd_{\gamma}-diameter of Ham(M,ω)\mathrm{Ham}(M,\omega) is infinite follows from [47] and [29].333More generally, the dγd_{\gamma}-diameter of Ham(M,ω)\mathrm{Ham}(M,\omega) is infinite whenever (Σ,σ)(\Sigma,\sigma) is a symplectic aspherical manifold, as was recently established by Mailhot in [32], and that result provided some motivation for the choice of examples here.

Theorem 6.

Let I1,I2:ΣΣI_{1},I_{2}:\Sigma\to\Sigma be two commuting anti-symplectic involutions (Ijσ=σI_{j}^{*}\sigma=-\sigma, Ij2=idI^{2}_{j}=id, j=1,2j=1,2) such that the complement of Fix(I1)\mathrm{Fix}(I_{1}) consists of two (homeomorphic) components, with positive genus, and such that the fixed point set Fix(f)\mathrm{Fix}(f) of f:=I2I1f:=I_{2}\circ I_{1} is non-empty and finite. Let \mathcal{L} be the Hamiltonian isotopy class of the Lagrangian submanifold L:={(x,f(x))}ML:=\{(x,f(x))\}\subset M. Then there exists ψHam(M,ω)\psi\in\mathrm{Ham}(M,\omega) such that for some R,E>0R,E>0,

(5) 𝖧R(ψ;)E.\displaystyle\mathsf{H}^{R}(\psi;\mathcal{L})\geq E.

A pair of involutions as in the theorem exist for any closed symplectic surface considered above. Note also that with Proposition 1.5 (v), by passing to iterates of ψ\psi, one finds, for all positive RR and EE, examples where (5) holds. As a direct consequence of Proposition 1.5 (iv) and of [13, Thm. A]444In our situation, we will know more about the “homotopy classes” of intersections that generate barcode entropy. As a consequence, lower bounds on htoph_{\mathrm{top}} can be obtained in a more elementary way, e.g. without the use of Yomdin’s theorem, cf. [2]. we obtain.

Corollary 7.

Let (M,ω)(M,\omega) be as above. For all R,E>0R,E>0 the subset {htopE}\{h_{\mathrm{top}}\geq E\} in Ham(M,ω)\mathrm{Ham}(M,\omega) contains open balls of radius R\geq R with respect to dγd_{\gamma}. In particular, the same holds with respect to dHoferd_{\mathrm{Hofer}}.

It was previously shown by Chor and the author, [17], that for a closed symplectic surface (Σ,σ)(\Sigma,\sigma) of genus at least 22, the assertion of Corollary 7 (for the Hofer-distance) holds for Ham(Σ,σ)\mathrm{Ham}(\Sigma,\sigma), using rather different methods. Interestingly enough, and to the best of my knowledge, the following question remains open.

Question.

Do there exist a Hamiltonian isotopy class \mathcal{L} of closed curves in a surface (Σ,σ)(\Sigma,\sigma) and a Hamiltonian diffeomorphism ψ\psi with 𝖧R(ψ;)>0\mathsf{H}^{R}(\psi;\mathcal{L})>0 for some R>0R>0?

Idea of proofs. As Theorem 1 is a relative analogue of Theorem B of [13], also the proof is very much a relative analogue of the proof in [13]. In a nutshell, Theorem 1 can be established by an interplay of arguments from hyperbolic dynamics together with C1C^{1}-bounds for pseudo-holomorphic curves with Lagrangian boundary conditions: by results going back to Bowen, the number of orbit segments from local unstable to local stable manifolds in a locally maximal, topological transitive hyperbolic set KK is bounded from below by its topological entropy. Moreover, a combination of C1C^{1}-estimates for Floer strips and the shadowing theorem shows that the energy that is needed to “connect” two chords in KK or to “leave” an isolating neighbourhood is uniformly bounded from below. Together, the estimates of Theorem 1 will be obtained.

The proof of Theorem 3 follows a similar strategy with two additional inputs. First, we apply the fact, recently proved in [36], that there exist a specific collection of horseshoe orbits 𝒪\mathcal{O} and a pair of curves in the complement such that the growth of chords of the suspension flow connecting those curves and which are unique in their homotopy class in the complement of the link 𝒪\mathcal{L}_{\mathcal{O}} induced by 𝒪\mathcal{O}, is exponential; moreover, all these chords can be encoded with the help of specific Markov rectangles. And second, we use that the Lagrangian Floer homology can be filtered in homotopy classes of chords in the complement of 𝒪\mathcal{L}_{\mathcal{O}}. Together with the estimates already used in the proof of Theorem 1, a “crossing energy lemma” is proved for certain Floer strips crossing 𝒪\mathcal{L}_{\mathcal{O}}, where we have flexibility in the choice of the pair of Lagrangian submanifolds while keeping the diffeomorphism fixed.

Finally, Theorem 6 is obtained by considering a diffeomorphism ψ\psi supported in the neighbourhood of the unit sphere bundle of Σ\Sigma embedded in a neighbourhood of the diagonal in M=Σ×ΣM=\Sigma\times\Sigma. The support of ψ\psi is foliated by leaves diffeomorphic to that sphere bundle and ψ\psi restricted to a leaf will coincide with the time TT map of the geodesic flow for a hyperbolic metric, where TT varies between leaves, monotonically increasing first, and decreasing later. The non-trivial chords relative LL will come in pairs with action difference growing linearly with iterations of ψ\psi. The special topological situation then prevents the existence of Floer strips connecting chords between different pairs, as long as one chord belongs to a special subset of pairs. The number of pairs in that subset grows exponentially, which establishes the result.

Plan of the paper. We start in Section 2 with some preliminaries on Lagrangian Floer homology and relative barcode entropy. Section 3 contains a lemma about Floer strips with small energy and the proof of Theorem 1. In Section 4, special properties of Lagrangian Floer homology and horseshoes in dimension 2 are discussed, before a proof of Theorem 3 is given. In Section 5, we then construct the examples that establish Theorem 6.

Acknowledgements. The author is grateful to Marcelo Alves, Joé Brendel, Erman Çineli, Viktor Ginzburg, Başak Gürel, and Leonid Polterovich for useful discussions. Special thanks go to Urs Fuchs for helpful comments concerning properties of Floer strips with small energy.

2. Preliminaries on relative barcode entropy

We start with some preliminaries on Hamiltonian diffeomorphisms and Lagrangian Floer homology. We will work essentially in the same setting as that in [13].

2.1. Hamiltonian diffeomorphisms and mapping cylinder

Let (M,ω)(M,\omega) be a symplectic manifold. To keep the exposition simple, we assume that MM is closed, but as in [13], the discussion extends to open MM if some assumptions are made on the ends of (M,ω)(M,\omega) and if we consider compactly supported Hamiltonians. Let ψ:MM\psi:M\to M be a Hamiltonian diffeomorphism. This means that ψ=ϕH1\psi=\phi_{H}^{1} is the time-11 map of a (time-dependent) flow ϕHt\phi_{H}^{t} of a 11-periodic Hamiltonian H:/×MH:\mathbbm{R}/\mathbbm{Z}\times M, i.e. ddtϕHt(x)=XHt(ϕHt(x))\frac{d}{dt}\phi^{t}_{H}(x)=X_{H_{t}}(\phi^{t}_{H}(x)) ((t,x)×M)(t,x)\in\mathbbm{R}\times M), where ω(XHt(x),)=dHt(x)\omega(X_{H_{t}}(x),\cdot)=-dH_{t}(x), and Ht(x):=H(x,t)H_{t}(x):=H(x,t). Let Y:=S1×MY:=S^{1}\times M, where we identify S1/S^{1}\cong\mathbbm{R}/\mathbbm{Z}. Fix once and for all a Hamiltonian function HH such that ϕH1=ψ\phi_{H}^{1}=\psi. This defines a lift to the suspension flow φ:×YY\varphi:\mathbbm{R}\times Y\to Y, defined by

(t,([r],x))([t+r],ϕHt+rt+rϕHt+rϕHr(x)),(t,([r],x))\mapsto([t+r],\phi_{H}^{t+r-\lfloor t+r\rfloor}\circ\phi_{H}^{\lfloor t+r\rfloor}\circ\phi_{H}^{-r}(x)),

the flow of the vector field X=tXHtX=\partial_{t}\oplus X_{H_{t}} on YY. For an orbit segment η:[0,k]M\eta:[0,k]\to M, nn\in\mathbbm{N}, of the Hamiltonian flow ϕHt\phi_{H}^{t}, we obtain an orbit segment η^:[0,k]Y\widehat{\eta}:[0,k]\to Y, η^(t):=(t,η(t))\widehat{\eta}(t):=(t,\eta(t)) of φ\varphi.

If not explicitly stated otherwise, we will consider a fixed metric gg on MM and denote by dd resp. d¯\overline{d} the distance function on MM resp. Y=S1×MY=S^{1}\times M induced by gg resp. the product metric gEucl×gg_{\mathrm{Eucl}}\times g.

2.2. The chain complex

We briefly recall relevant notions for Lagrangian Floer homology. The construction of Lagrangian Floer homology goes back to Floer [21] and in the monotone setting to Oh [40], the setting that is relevant here. More specifically, we closely follow setup and notation of [47] and [13]. In the following, all Lagrangian submanifolds LL will be closed and monotone. This means that there exists κ>0\kappa>0 such that ω,A=κμL,A\langle\omega,A\rangle=\kappa\langle\mu_{L},A\rangle for all Aπ2(M,L)A\in\pi_{2}(M,L), where μL\mu_{L} is the Maslov class of LL. We also require that NL2N_{L}\geq 2, where the minimal Chern number NLN_{L} is defined as the positive generator of the subgroup in \mathbbm{Z} generated by the μL,A\langle\mu_{L},A\rangle for all Aπ2(M,L)A\in\pi_{2}(M,L) that can be represented by cylinders with boundary in LL, and NL=N_{L}=\infty if this subgroup is trivial, see also [47, 13]. We also denote by 𝔥(M,L)=κNL(0,+]\mathfrak{h}(M,L)=\kappa N_{L}\in(0,+\infty] the minimal area of a non-trivial cylinder with boundary in LL.

Let L0ML_{0}\subset M be a Lagrangian submanifold as above, and let L1=ϕG1(L0)L_{1}=\phi^{1}_{G}(L_{0}) be Hamiltonian isotopic to L0L_{0}, for some Hamiltonian GG that we fix from now on. Assume that the pair (L0,L1)(L_{0},L_{1}) is non-degenerate; we recall that this means that L0L_{0} and ψk(L1)\psi^{-k}(L_{1}) intersect transversally, for all kk\in\mathbbm{N}. For given kk\in\mathbbm{N}, the construction of the Floer chain complexes 𝒞(L0,L1,k)=CF(L0,ψk(L1))\mathcal{C}(L_{0},L_{1},k)=\mathrm{CF}(L_{0},\psi^{-k}(L_{1})) can be sketched as follows. Let 𝒫k=𝒫(L0,ψk(L1))\mathcal{P}^{k}=\mathcal{P}(L_{0},\psi^{-k}(L_{1})) be the space of smooth paths x:[0,1]Mx:[0,1]\to M with x(0)L0x(0)\in L_{0}, x(1)ψk(L1)x(1)\in\psi^{-k}(L_{1}). Fix a system γ𝔠𝒫k\gamma_{\mathfrak{c}}\in\mathcal{P}^{k} of representatives of the homotopy classes 𝔠π1(M;L0,ψk(L1))\mathfrak{c}\in\pi_{1}(M;L_{0},\psi^{-k}(L_{1})). Denote by 𝒫~k\widetilde{\mathcal{P}}^{k} the set of pairs (x,[vx])(x,[v_{x}]), where x𝒫kx\in\mathcal{P}^{k} and [vx][v_{x}] is a capping up to a certain equivalence relation. Here a capping vxv_{x} is a C1C^{1}-map vx:[0,1]2Mv_{x}:[0,1]^{2}\to M with v(,0)L0v(\cdot,0)\in L_{0}, v(,1)ψk(L1)v(\cdot,1)\in\psi^{-k}(L_{1}) and v(0,t)=γ𝔠v(0,t)=\gamma_{\mathfrak{c}} and v(1,t)=x(t)v(1,t)=x(t). Two cappings are equivalent if their ω\omega-area and Maslov index coincide; the Maslov index is well-defined given a choice of symplectic trivializations of γ𝔠TM\gamma_{\mathfrak{c}}^{*}TM, see [47]. Given x^=(x,[ux])𝒫k~\hat{x}=(x,[u_{x}])\in\widetilde{\mathcal{P}^{k}}, we can glue annuli v:S1×[0,1]Mv:S^{1}\times[0,1]\to M, with v(s,0)L0v(s,0)\in L_{0}, v(s,1)ψk(L1)v(s,1)\in\psi^{-k}(L_{1}), to vxv_{x} and obtain recappings v#x^=(x,[v#vx])v\#\hat{x}=(x,[v\#v_{x}]). We consider 𝒜:𝒫~k\mathcal{A}:\widetilde{\mathcal{P}}^{k}\to\mathbbm{R}, defined by

(6) 𝒜(x,[vx])=[0,1]2vxω.\displaystyle\mathcal{A}(x,[v_{x}])=-\int_{[0,1]^{2}}v_{x}^{*}\omega.

The critical points of 𝒜\mathcal{A} are those (x,[vx])(x,[v_{x}]) for which xL0ψk(L1)x\in L_{0}\cap\psi^{-k}(L_{1}) is constant. We define CF(L0,ψk(L1))\mathrm{CF}(L_{0},\psi^{-k}(L_{1})) as in [13]: fix for each xL0ψk(L1)x\in L_{0}\cap\psi^{-k}(L_{1}) one equivalence class of cappings [vx][v_{x}] of xx. Denote the set of those pairs by 𝒳k𝒫~k\mathcal{X}^{k}\subset\widetilde{\mathcal{P}}^{k}, and let CF(L0,ψk(L1))\mathrm{CF}(L_{0},\psi^{-k}(L_{1})) be the vector space generated by the elements in 𝒳k\mathcal{X}^{k} over the universal Novikov field Λ\Lambda, which is formed by the formal sums λ=j0fjTaj\lambda=\sum_{j\geq 0}f_{j}T^{a_{j}}, fj2f_{j}\in\mathbbm{Z}_{2}, aja_{j}\in\mathbbm{R}, subject to the condition #{j|fj0,ajC}<\#\{j\,|\,f_{j}\neq 0,a_{j}\leq C\}<\infty for all CC\in\mathbbm{R}. To define the boundary map :CF(L0,ψk(L1))CF(L0,ψk(L1))\partial:\mathrm{CF}(L_{0},\psi^{-k}(L_{1}))\to\mathrm{CF}(L_{0},\psi^{-k}(L_{1})), one considers, for a choice of S1S^{1}-family J=(Jt)tS1J=(J_{t})_{t\in S^{1}} of almost complex structures compatible with ω\omega and critical points x^±=(x±,[v±])\hat{x}_{\pm}=(x_{\pm},[v_{{\pm}}]) of 𝒜\mathcal{A}, the moduli space J(x^,x^+,L0,ψk(L1))\mathcal{M}_{J}(\hat{x}_{-},\hat{x}_{+},L_{0},\psi^{-k}(L_{1})) of holomorphic strips u:×[0,1]M,u:\mathbbm{R}\times[0,1]\to M,

(7) su(s,t)+Jt(u(s,t))tu(s,t)=0,\displaystyle\partial_{s}u(s,t)+J_{t}(u(s,t))\partial_{t}u(s,t)=0,

subject to boundary conditions u(s,0)L0u(s,0)\subset L_{0}, u(s,1)ψk(L1)u(s,1)\in\psi^{-k}(L_{1}) (ss\in\mathbbm{R}), asymptotics lims±u(s,t)x±\lim_{s\to\pm\infty}u(s,t)\equiv x_{\pm}, and for which the glued map v#uv_{-}\#u is equivalent to v+v_{+}. If JJ is such that the moduli spaces are transversally cut out, which holds generically, one can show that if the Maslov index difference of x^\hat{x} and v#x^jv\#\hat{x}_{j} is 11, J(x^,x^+,L0,ψk(L1))/\mathcal{M}_{J}(\hat{x}_{-},\hat{x}_{+},L_{0},\psi^{-k}(L_{1}))/\mathbbm{R} is finite, where we are modding out by the canonical \mathbbm{R}-action on solutions of (7). This makes it possible to define the boundary map \partial, defined on the generators 𝒳k\mathcal{X}^{k} as x^=λjx^j\partial\hat{x}=\sum\lambda_{j}\hat{x}_{j}, where λj=vfvTω(v)Λ\lambda_{j}=\sum_{v}f_{v}T^{\omega(v)}\in\Lambda. Here the sum is taken over all recappings vv of x^j\hat{x}_{j} such that the Maslov index difference of x^\hat{x} and v#x^jv\#\hat{x}_{j} is 11, and fvf_{v} is the number of elements (mod 22) in J(x^,v#x^j,L0,ψk(L1))/\mathcal{M}_{J}(\hat{x},v\#\hat{x}_{j},L_{0},\psi^{-k}(L_{1}))/\mathbbm{R}. One shows that \partial is indeed a boundary map (2=0)(\partial^{2}=0) and that the homology denoted by HF(L0,ψk(L1))\mathrm{HF}(L_{0},\psi^{-k}(L_{1})) does not depend on the choice of JJ.

The homology is invariant under Hamiltonian isotopy: a Hamiltonian isotopy L0s=ϕFs(L0)L^{s}_{0}=\phi_{F}^{s}(L_{0}), F:M×/F:M\times\mathbbm{R}/\mathbbm{Z}\to\mathbbm{R}, s[0,1]s\in[0,1], from L0L_{0} to L0:=L01L^{\prime}_{0}:=L^{1}_{0} with (L0,L1)(L^{\prime}_{0},L_{1}) non-degenerate, induces for every kk\in\mathbbm{N} a chain map f0:CF(L0,ψk(L1))CF(L0,ψk(L1))f_{0}:\mathrm{CF}(L_{0},\psi^{-k}(L_{1}))\to\mathrm{CF}(L^{\prime}_{0},\psi^{-k}(L_{1})), defined by counting holomorphic strips u:×S1Mu:\mathbbm{R}\times S^{1}\to M (with ss-dependent JJ) in a similar manner as above, now with moving Lagrangian boundary conditions u(s,0)L0β(s)u(s,0)\in L^{\beta(s)}_{0} and u(s,1)ψk(L1)u(s,1)\in\psi^{-k}(L_{1}) (s)(s\in\mathbbm{R}), where β:[0,1]\beta:\mathbbm{R}\to[0,1] is a smooth non-decreasing surjective function, constant outside a compact interval [40, §5], see also [10, §3.2]. Similarly, one defines g0:CF(L0,ψk(L1))CF(L0,ψk(L1))g_{0}:\mathrm{CF}(L^{\prime}_{0},\psi^{-k}(L_{1}))\to\mathrm{CF}(L_{0},\psi^{-k}(L_{1})) and one obtains that g0f0g_{0}\circ f_{0} is chain homotopic to the identity. Analogously, for a Hamiltonian isotopy L1sL_{1}^{s}, s[0,1]s\in[0,1], from L1L_{1} to L1L^{\prime}_{1}, one obtains for all kk\in\mathbbm{N} a chain map f1:CF(L0,ψk(L1))CF(L0,ψk(L1))f_{1}:\mathrm{CF}(L_{0},\psi^{-k}(L_{1}))\to\mathrm{CF}(L_{0},\psi^{-k}(L^{\prime}_{1})) with similar properties.

2.3. Rescaling and lifts of holomorphic strips

Given a holomorphic strip uJ(x^,x^+,L0,ψk(L1))u\in\mathcal{M}_{J}(\hat{x}_{-},\hat{x}_{+},L_{0},\psi^{-k}(L_{1})), we will consider the rescaled strip uˇ:×[0,k]M\check{u}:\mathbbm{R}\times[0,k]\to M defined by uˇ(s,t)=u(1ks,1kt)\check{u}(s,t)=u(\frac{1}{k}s,\frac{1}{k}t). The strip uˇ\check{u} is holomorphic, that is, it satisfies (7) with JtJ_{t} replaced by J1ktJ_{\frac{1}{k}t}. By composing with the Hamiltonian flow, we obtain

u^:×[0,k]M,u^(s,t):=ϕHt(uˇ(s,t)).\hat{u}:\mathbbm{R}\times[0,k]\to M,\,\,\hat{u}(s,t):=\phi^{t}_{H}(\check{u}(s,t)).

The strip u^\hat{u} satisfies the Floer equation

(8) su^(s,t)+J^t(u^(s,t))(tu^(s,t)XHt(u^(s,t)))=0,\partial_{s}\hat{u}(s,t)+\hat{J}_{t}(\hat{u}(s,t))(\partial_{t}\hat{u}(s,t)-X_{H_{t}}(\hat{u}(s,t)))=0,

where J^t:=(ϕHt)J1kt\hat{J}_{t}:=(\phi_{H}^{t})_{*}J_{\frac{1}{k}t}, t[0,k]t\in[0,k]. The asymptotics of u^\hat{u} are chords from L0L_{0} to L1L_{1} (lims±u^(s,t)=ϕHt(x±)\lim_{s\to\pm\infty}\hat{u}(s,t)=\phi^{t}_{H}(x_{\pm}), t[0,k]t\in[0,k]) and u^\hat{u} satisfies u^(s,0)L0\hat{u}(s,0)\in L_{0}, u^(s,k)L1\hat{u}(s,k)\in L_{1}, ss\in\mathbbm{R}. This identification extends in a natural way to the situation of moving boundary conditions and ss-dependent JJ.

Additionally, we consider also the maps

u¯:×[0,k]/×M=Y,u¯(s,t):=([t],u^(s,t)),\displaystyle\overline{u}:\mathbbm{R}\times[0,k]\to\mathbbm{R}/\mathbbm{Z}\times M=Y,\,\,\overline{u}(s,t):=([t],\hat{u}(s,t)),
u~:×[0,k]×/×M,u~(s,t)=(s,[t],u^(s,t)),\displaystyle\widetilde{u}:\mathbbm{R}\times[0,k]\to\mathbbm{R}\times\mathbbm{R}/\mathbbm{Z}\times M,\,\,\widetilde{u}(s,t)=(s,[t],\hat{u}(s,t)),
u~~:×[0,k]×[0,k]×M,u~~(s,t):=(s,t,u^(s,t)).\displaystyle{\mathchoice{\widetilde{\hbox{\set@color$\displaystyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\textstyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\scriptstyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\scriptscriptstyle\widetilde{u}$}}}}:\mathbbm{R}\times[0,k]\to\mathbbm{R}\times[0,k]\times M,\,\,{\mathchoice{\widetilde{\hbox{\set@color$\displaystyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\textstyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\scriptstyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\scriptscriptstyle\widetilde{u}$}}}}(s,t):=(s,t,\hat{u}(s,t)).

With respect to a suitable (time-independent) J~\widetilde{J}, the lift u~~{\mathchoice{\widetilde{\hbox{\set@color$\displaystyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\textstyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\scriptstyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\scriptscriptstyle\widetilde{u}$}}}} is J~\widetilde{J}-holomorphic. Here, J~\widetilde{J} is defined at a point (s,t,p)×[0,k]×M(s,t,p)\in\mathbbm{R}\times[0,k]\times M by J~(s,t,p)(s)=t+XH(s,t,p)\widetilde{J}_{(s,t,p)}(\partial_{s})=\partial_{t}+X_{H}(s,t,p), where XH(s,t,p)X_{H}(s,t,p) is the (time-independent) horizontal lift of XHt(p)X_{H_{t}}(p) to ×[0,k]×M\mathbbm{R}\times[0,k]\times M, and by the condition that J~(s,t,p)(v)\widetilde{J}_{(s,t,p)}(v) is the horizontal lift of J^t(v)\hat{J}_{t}(v) to {s}×{t}×M\{s\}\times\{t\}\times M. The boundary components of u~~{\mathchoice{\widetilde{\hbox{\set@color$\displaystyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\textstyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\scriptstyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\scriptscriptstyle\widetilde{u}$}}}} then lie in ×{0}×L0\mathbbm{R}\times\{0\}\times L_{0} and ×{k}×L0\mathbbm{R}\times\{k\}\times L_{0}.

In this situation, we can also naturally generalize to moving boundary conditions and ss-dependent JJ. In particular, the condition with moving boundary LisL^{s}_{i}, i=0,1i=0,1, corresponds to the condition that the boundary components of u~~{\mathchoice{\widetilde{\hbox{\set@color$\displaystyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\textstyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\scriptstyle\widetilde{u}$}}}{\widetilde{\hbox{\set@color$\scriptscriptstyle\widetilde{u}$}}}} lie in {(s,0,p)|s,pL0β(1ks)}\{(s,0,p)\,|\,s\in\mathbbm{R},p\in L_{0}^{\beta(\frac{1}{k}s)}\} and {(s,k,p)|s,pL1β(1ks)}\{(s,k,p)\,|\,s\in\mathbbm{R},p\in L_{1}^{\beta(\frac{1}{k}s)}\}, where β\beta is defined as above.

2.4. Relative barcode entropy

We first keep the assumption that the pair (L0,L1)(L_{0},L_{1}) is non-degenerate. The action 𝒜\mathcal{A} defined in (6) can be extended to general elements of CF(L0,ψk(L1))\mathrm{CF}(L_{0},\psi^{-k}(L_{1})). We set ν(λ):=minaj\nu(\lambda):=\min a_{j} for λ=j0fjTajΛ\lambda=\sum_{j\geq 0}f_{j}T^{a_{j}}\in\Lambda, ν(0)=+\nu(0)=+\infty, and put

𝒜(λx^):=𝒜(x^)ν(λ) and 𝒜(λix^i):=maxi𝒜(λix^i).\mathcal{A}(\lambda\hat{x}):=\mathcal{A}(\hat{x})-\nu(\lambda)\text{ and }\mathcal{A}\left(\sum\lambda_{i}\hat{x}_{i}\right):=\max_{i}\mathcal{A}(\lambda_{i}\hat{x}_{i}).

The differential \partial strictly decreases the action 𝒜\mathcal{A}. In fact, for any uJ(x^,x^+,L0,ψk(L1))u\in\mathcal{M}_{J}(\hat{x}_{-},\hat{x}_{+},L_{0},\psi^{-k}(L_{1})),

𝒜(x)𝒜(x+)=E(u).\mathcal{A}(x_{-})-\mathcal{A}(x_{+})=E(u).

Here, E(u)E(u) is the energy of uu given by

E(u)=12du2=01|su(s,t)|2𝑑t𝑑s,E(u)=\frac{1}{2}\int_{-\infty}^{\infty}\|du\|^{2}=\int_{-\infty}^{\infty}\int_{0}^{1}|\partial_{s}u(s,t)|^{2}\,dt\,ds,

where |||\cdot| is induced by the metric ω(,J)\omega(\cdot,J\cdot).

We are now in the situation to give the definition of (relative) barcode entropy. We keep the discussion rather short, making use of the results in [48] and refer the reader to [13] and references therein for an equivalent definition and further motivation. Given ϵ>0\epsilon>0, a non-zero vector ζimCF(L0,ψk(L1))\zeta\in\mathrm{im}\partial\subset\mathrm{CF}(L_{0},\psi^{-k}(L_{1})) is said to be ϵ\epsilon-robust if for every yCF(L0,ψk(L1))y\in\mathrm{CF}(L_{0},\psi^{-k}(L_{1})) with y=ζ\partial y=\zeta,

𝒜(y)𝒜(ζ)>ϵ.\mathcal{A}(y)-\mathcal{A}(\zeta)>\epsilon.

A subspace VimCF(L0,ψk(L1))V\subset\mathrm{im}\partial\subset\mathrm{CF}(L_{0},\psi^{-k}(L_{1})) is called ϵ\epsilon-robust if every non-zero vector ζV\zeta\in V is ϵ\epsilon-robust. Define

bϵ(ψ;L0,L1,k):=max{dimV|VCF(L0,ψk(L1)) is ϵ - robust}.b^{*}_{\epsilon}(\psi;L_{0},L_{1},k):=\max\{\dim V\,|\,V\subset\mathrm{CF}(L_{0},\psi^{-k}(L_{1}))\text{ is }\epsilon\text{ - robust}\}.

By results in [48] and the discussion in [13], the integer bϵ(ψ;L0,L1,k)b^{*}_{\epsilon}(\psi;L_{0},L_{1},k) coincides with the number of finite bars of length larger than ϵ\epsilon in the definition of the barcode of CF(L0,ψk(L1))\mathrm{CF}(L_{0},\psi^{-k}(L_{1})) in [13]. To keep the same conventions as in [13], we also consider the infinite bars and set

bϵ(ψ;L0,L1,k)=bϵ(ψ;L0,L1,k)+dimHF(L0,ψk(L1)).b_{\epsilon}(\psi;L_{0},L_{1},k)=b^{*}_{\epsilon}(\psi;L_{0},L_{1},k)+\dim\mathrm{HF}(L_{0},\psi^{-k}(L_{1})).

For a general admissible, not necessarily non-degenerate pair (L0,L1)(L_{0},L_{1}), define

bϵ(ψ;L0,L1,k)=lim inf(L0,L1)(L0,L1)bϵ(ψ;L0,L1,k),b_{\epsilon}(\psi;L_{0},L_{1},k)=\liminf_{(L^{\prime}_{0},L^{\prime}_{1})\to(L_{0},L_{1})}b_{\epsilon}(\psi;L^{\prime}_{0},L^{\prime}_{1},k),

where the limit is taken over all non-degenerate pairs (L0,L1)(L^{\prime}_{0},L^{\prime}_{1}) and the convergence of pairs is considered with respect to the C2C^{2} topology. By stability properties of barcodes, see below, the above is well defined.

Following [13], define the ϵ\epsilon-relative barcode entropy as

ϵ(ψ;L0,L1)=lim supklog(bϵ(ψ;L0,L1,k))k,\hbar_{\epsilon}(\psi;L_{0},L_{1})=\limsup_{k\to\infty}\frac{\log(b_{\epsilon}(\psi;L_{0},L_{1},k))}{k},

and the relative barcode entropy as

(ψ;L0,L1)=limϵ0ϵ(ψ;L0,L1).\hbar(\psi;L_{0},L_{1})=\lim_{\epsilon\to 0}\hbar_{\epsilon}(\psi;L_{0},L_{1}).

Additionally, for any R>0R>0, let us also recall from the introduction the definition of the RR-strong relative barcode entropy,

𝖧R(ψ;L0,L1)=limR^RR^<Rlim supklog(bR^k(ψ;L0,L1,k))k.\mathsf{H}^{R}(\psi;L_{0},L_{1})=\lim_{\begin{subarray}{c}\hat{R}\to R\\ \hat{R}<R\end{subarray}}\limsup_{k\to\infty}\frac{\log(b_{\hat{R}k}(\psi;L_{0},L_{1},k))}{k}.

Some important properties of the (strong) barcode entropy follow from known stability properties of the filtered Floer homology with respect to the Hofer-norm and the γ\gamma-norm. Recall that for a Hamiltonian diffeomorphism ψ\psi on (M,ω)(M,\omega), its Hofer-norm is defined as

ψHofer=infH01maxH(x,t)minH(x,t)dt,\|\psi\|_{\mathrm{Hofer}}=\inf_{H}\int_{0}^{1}\max H(x,t)-\min H(x,t)\,dt,

and its γ\gamma-norm as

γ(ψ)=inf(c(H)+c(H¯)).\gamma(\psi)=\inf(c(H)+c(\overline{H})).

In both cases the infimum is taken over all (time-dependent) Hamiltonian functions HH such that ψ=ϕH1\psi=\phi_{H}^{1}, moreover cc denotes the spectral number associated to the fundamental class [M][M], and H¯\overline{H} denotes the Hamiltonian with ϕH¯t=(ϕHt)1\phi_{\overline{H}}^{t}=(\phi_{H}^{t})^{-1}, see [26], and [50, 46, 41], respectively. It is well-known that γ(ψ)ψHofer\gamma(\psi)\leq\|\psi\|_{\mathrm{Hofer}}. Given a Hamiltonian isotopy class \mathcal{L} of closed monotone Lagrangian submanifolds, the exterior γ\gamma-distance on \mathcal{L} is defined as

dγ(L,L):=inf{γ(ψ)|ψHam(M,ω),ψ(L)=L}.d_{\gamma}(L,L^{\prime}):=\inf\{\gamma(\psi)\,|\,\psi\in\mathrm{Ham}(M,\omega),\psi(L)=L^{\prime}\}.

Analogously the Hofer-distance dHoferd_{\mathrm{Hofer}} is defined.

For our purposes it is convenient to express the stability properties in terms of the quantities bϵ(ψ;L0,L1,k)b_{\epsilon}(\psi;L_{0},L_{1},k) for triples (ψ,L0,L1)(\psi,L_{0},L_{1}) as above and kk\in\mathbbm{N}: for any ϵ>0\epsilon>0, it holds that if L0L^{\prime}_{0} is a Lagrangian in MM with dγ(L0,L0)<δ<ϵd_{\gamma}(L^{\prime}_{0},L_{0})<\delta<\epsilon and such that (L0,L1)(L^{\prime}_{0},L_{1}) is admissible, then

(9) bϵδ(ψ;L0,L1,k)bϵ(ψ;L0,L1,k).\displaystyle b_{\epsilon-\delta}(\psi;L^{\prime}_{0},L_{1},k)\geq b_{\epsilon}(\psi;L_{0},L_{1},k).

This statement follows from the work [29]; the weaker statement that the same holds with respect to Hofer-distance already follows from a straightforward adaption of results in [44]. The same results imply the following statement: if ψ\psi^{\prime} is a Hamiltonian diffeomorphism on MM with dγ(ψ,ψ):=γ(ψ1ψ)<δ<ϵd_{\gamma}(\psi^{\prime},\psi):=\gamma(\psi^{-1}\circ\psi^{\prime})<\delta<\epsilon, then

(10) bϵδ(ψ;L0,L1,1)bϵ(ψ;L0,L1,1),\displaystyle b_{\epsilon-\delta}(\psi^{\prime};L_{0},L_{1},1)\geq b_{\epsilon}(\psi;L_{0},L_{1},1),

and hence, by the triangle inequality,

bϵkδ(ψ;L0,L1,k)bϵ(ψ;L0,L1,k),\displaystyle b_{\epsilon-k\delta}(\psi^{\prime};L_{0},L_{1},k)\geq b_{\epsilon}(\psi;L_{0},L_{1},k),

if kδ<ϵk\delta<\epsilon. By (9) and the fact that

(11) bϵ(ψ;L0,L1,k)=bϵ(ψ1;L1,L0,k),\displaystyle b_{\epsilon}(\psi;L_{0},L_{1},k)=b_{\epsilon}(\psi^{-1};L_{1},L_{0},k),

see [13], it follows that (ψ;L0,L1)\hbar(\psi;L_{0},L_{1}) is lower semi-continuous in (L0,L1)(L_{0},L_{1}) with respect to dγd_{\gamma}. Furthermore, Proposition 1.5 from the introduction can be proved using the facts above.

Proof of Proposition 1.5.

If (L0,L1)(L^{\prime}_{0},L^{\prime}_{1}) are Hamiltonian isotopic to (L0,L1)(L_{0},L_{1}), then by (9) and (11), for R>0R>0 and kk sufficiently large,

bRkZ(ψ;L0,L1,k)bRk(ψ;L0,L1,k),b_{Rk-Z}(\psi;L^{\prime}_{0},L^{\prime}_{1},k)\geq b_{Rk}(\psi;L_{0},L_{1},k),

where Z:=dγ(L0,L0)+dγ(L1,L1)Z:=d_{\gamma}(L_{0}^{\prime},L_{0})+d_{\gamma}(L^{\prime}_{1},L_{1}). Assertion (i) follows directly if we apply this inequality for any given ϵ>0\epsilon>0 to R^=Rϵ\hat{R}=R-\epsilon: for k>Z/ϵk>Z/\epsilon,

b(R2ϵ)k(ψ;L0,L1,k)bR^kZ(ψ;L0,L1,k)bR^k(ψ;L0,L1,k).b_{(R-2\epsilon)k}(\psi;L^{\prime}_{0},L^{\prime}_{1},k)\geq b_{\hat{R}k-Z}(\psi;L^{\prime}_{0},L^{\prime}_{1},k)\geq b_{\hat{R}k}(\psi;L_{0},L_{1},k).

To obtain (iii), it is sufficient to show that for any 0<R<R<γ(ψ)=:Z0<{R^{\prime}}<R<\gamma(\psi)=:Z and any sequence kik_{i} of natural numbers with ki+k_{i}\to+\infty there exists (up to passing first to a subsequence) a sequence kik^{\prime}_{i} of natural numbers such that

limi+logbRkiZ(ψ;L0,L1,ki)kiγ(ψ)Rγ(ψ)Rlimi+logbRki(ψ;L0,L1,ki)ki.\lim_{i\to+\infty}\frac{\log b_{R^{\prime}k^{\prime}_{i}-Z}(\psi;L_{0},L_{1},k^{\prime}_{i})}{k^{\prime}_{i}}\geq\frac{\gamma(\psi)-R^{\prime}}{\gamma(\psi)-{R}}\lim_{i\to+\infty}\frac{\log b_{{R}k_{i}}(\psi;L_{0},L_{1},k_{i})}{{k}_{i}}.

Let us choose ki:=tik^{\prime}_{i}:=\lceil t^{\prime}_{i}\rceil with ti:=γ(ψ)Rγ(ψ)Rkit^{\prime}_{i}:=\frac{\gamma(\psi)-R}{\gamma(\psi)-{R^{\prime}}}k_{i}. Since we have dγ(ψki,ψki)(kiki)γ(ψ)d_{\gamma}(\psi^{k^{\prime}_{i}},\psi^{k_{i}})\leq(k_{i}-k^{\prime}_{i})\gamma(\psi), (10) implies that

bRkiγ(ψ)(kiki)(ψ;L0,L1,ki)bRki(ψ;L0,L1,ki).b_{Rk_{i}-\gamma(\psi)(k_{i}-k^{\prime}_{i})}(\psi;L_{0},L_{1},k^{\prime}_{i})\geq b_{Rk_{i}}(\psi;L_{0},L_{1},k_{i}).

Note that

Rkiγ(ψ)(kiki)ki(Rγ(ψ))+γ(ψ)ti=Rti.Rk_{i}-\gamma(\psi)(k_{i}-k^{\prime}_{i})\geq k_{i}(R-\gamma(\psi))+\gamma(\psi)t^{\prime}_{i}=R^{\prime}t^{\prime}_{i}.

Hence,

bRti(ψ;L0,L1,ki)bRki(ψ;L0,L1,ki).b_{R^{\prime}t^{\prime}_{i}}(\psi;L_{0},L_{1},k^{\prime}_{i})\geq b_{Rk_{i}}(\psi;L_{0},L_{1},k_{i}).

The left-hand side is at most bRkiZ(ψ;L0,L1,ki)b_{R^{\prime}k^{\prime}_{i}-Z}(\psi;L_{0},L_{1},k^{\prime}_{i}), and (iii) follows. That γ(ψ)=R\gamma(\psi)=R implies 𝖧R(ψ;L0,L1)=0\mathsf{H}^{R}(\psi;L_{0},L_{1})=0 follows easily from the inequality in (iii), and hence also assertion (ii) directly follows. Assertion (iv) can be easily shown using (10), and (v) is obvious. ∎

For further properties of barcode entropy, we refer to [13, Prop. 4.4].

3. Pseudo-orbits property for small energies and the proof of Theorem 1

3.1. Holomorphic curves with small energy and pseudo-orbits

We start this section with a lemma that asserts that if a holomorphic strip uu has sufficiently small energy, then the paths tu¯(s,t)t\mapsto\overline{u}(s,t) are φ\varphi-pseudo-orbits in YY along the intervals where they are defined. See also Lemma 6.3 in [13] for a related statement. Let J^\hat{J} be a fixed (possibly tt-dependent) almost complex structure on MM, and let

(12) J=Jt=[DϕHt]1J^DϕHt.\displaystyle J=J_{t}=[D\phi_{H}^{t}]^{-1}\hat{J}D\phi_{H}^{t}.
Lemma 3.1.

Let (L0,L1)(L_{0},L_{1}) be a non-degenerate admissible pair of Lagrangian submanifolds in MM. Then, given η>0\eta>0, there exists ϵ=ϵ(η,L0,L1)>0\epsilon=\epsilon(\eta,L_{0},L_{1})>0 such that if uJkt(x^,x^+,L0,ψk(L1))u\in\mathcal{M}_{J_{kt}}(\hat{x}_{-},\hat{x}_{+},L_{0},\psi^{-k}(L_{1})) (for some x^±\hat{x}_{\pm} and kk\in\mathbbm{N}) and E(u)<ϵE(u)<\epsilon, then for all ss\in\mathbbm{R} and all n^\hat{n}\in\mathbbm{Z}, 0n^k10\leq\hat{n}\leq k-1, t[0,1]t\in[0,1],

(13) d¯(u¯(s,t+n^),φt(u¯(s,n^)))<η.\displaystyle\overline{d}(\overline{u}(s,t+\hat{n}),\varphi^{t}(\overline{u}(s,\hat{n})))<\eta.

Moreover, given a C2C^{2}-small neighbourhood of any fixed admissible pair (L0,L1)(L^{\prime}_{0},L^{\prime}_{1}), then there exists ϵ=ϵ(η)>0\epsilon=\epsilon(\eta)>0 such that the above holds for all non-degenerate admissible pairs (L0,L1)(L_{0},L_{1}) in that neighbourhood.

Proof.

For uJkt(x^,x^+,L0,ψk(L1))u\in\mathcal{M}_{J_{kt}}(\hat{x}_{-},\hat{x}_{+},L_{0},\psi^{-k}(L_{1})) as in the lemma, and n^\hat{n}\in\mathbbm{Z}, 0n^k10\leq\hat{n}\leq k-1, consider v:×[n^,kn^]Mv:\mathbbm{R}\times[-\hat{n},k-\hat{n}]\to M, defined by

v(s,t):=ψn^(uˇ(s,t+n^))=ψn^(u(1ks,1k(t+n^))).v(s,t):=\psi^{\hat{n}}(\check{u}(s,t+\hat{n}))=\psi^{\hat{n}}\left({u}\left(\frac{1}{k}s,\frac{1}{k}(t+\hat{n})\right)\right).

We note that vv is JtJ_{t}-holomorphic and that v|×{n^}ψn^(L0)v|_{\mathbbm{R}\times\{-\hat{n}\}}\subset\psi^{\hat{n}}(L_{0}), v|×{kn^}ψn^k(L1)v|_{\mathbbm{R}\times\{k-\hat{n}\}}\subset\psi^{\hat{n}-k}(L_{1}). Also note that

(14) u¯(s,n^)=(0,v(s,0)),u¯(s,t+n^)=(t,ϕHt+n^(uˇ(s,t+n^)))=φt((0,v(s,t))),\displaystyle\begin{split}\overline{u}(s,\hat{n})&=(0,v(s,0)),\\ \overline{u}(s,t+\hat{n})&=(t,\phi_{H}^{t+\hat{n}}(\check{u}(s,t+\hat{n})))=\varphi^{t}((0,v(s,t))),\end{split}

for all s,t[0,1]s\in\mathbbm{R},t\in[0,1].

To simplify the exposition, let us assume that k2k\geq 2, and it will be obvious how to treat the case k=1k=1. For (s,t)2(s,t)\in\mathbbm{R}^{2} and r>0r>0, denote by Dr(s,t)D_{r}(s,t) the open disk of radius rr around (s,t)(s,t). We note the following. If n^{0,k1}\hat{n}\notin\{0,k-1\}, then for any point (s,t)×[0,1](s,t)\in\mathbbm{R}\times[0,1] we have that D1(s,t)D_{1}(s,t) is contained in the domain dom(v)\mathrm{dom}(v) of vv; if n^=0\hat{n}=0, then for any ss\in\mathbbm{R} we have that D2(s,0)(×[0,+))dom(v)D_{2}(s,0)\cap(\mathbbm{R}\times[0,+\infty))\subset\mathrm{dom}(v); and if n^=k1\hat{n}=k-1, then for any ss\in\mathbbm{R} we have that D2(s,1)(×(,1])dom(v)D_{2}(s,1)\cap(\mathbbm{R}\times(-\infty,1])\subset\mathrm{dom}(v). We can apply a priori estimates for the differential of vv for (s,t)×[0,1](s,t)\in\mathbbm{R}\times[0,1], see  [34, Lemma 4.3.1]555That Lemma is formulated for a fixed almost complex structure, but it can be easily proved that it continuous to hold for (s,t)(s,t)-dependent (uniformly bounded) almost complex structures., and obtain that there exist δ>0,c>0\delta>0,c>0 (depending on L0L_{0} and L1L_{1}), such that if ×[n^,kn^]|dv|2𝑑sdt<δ\int_{\mathbbm{R}\times[-\hat{n},k-\hat{n}]}|dv|^{2}\,ds\wedge dt<\delta, then for all s,t[0,1]s\in\mathbbm{R},t\in[0,1],

|dv(s,t)|2cD^|dv|2𝑑sdt<cδ,|dv(s,t)|^{2}\leq c\int_{\hat{D}}|dv|^{2}\,ds\wedge dt<c\delta,

where

D^:={D1(s,t) if n^{0,k1},D2(s,0)(×[0,+)) if n^=0,D2(s,1)(×(,1]) if n^=k1.\hat{D}:=\begin{cases}&D_{1}(s,t)\,\,\,\quad\qquad\qquad\qquad\text{ if }\hat{n}\notin\{0,k-1\},\\ &D_{2}(s,0)\cap(\mathbbm{R}\times[0,+\infty))\,\text{ if }\hat{n}=0,\\ &D_{2}(s,1)\cap(\mathbbm{R}\times(-\infty,1])\,\text{ if }\hat{n}=k-1.\end{cases}

We note that ×[n^,kn^]|dv|2𝑑sdt=2E(u)\int_{\mathbbm{R}\times[-\hat{n},k-\hat{n}]}|dv|^{2}\,ds\wedge dt=2E(u), and hence if E(u)E(u) is small, then for any ss\in\mathbbm{R} the curve tv(s,t)t\mapsto v(s,t), t[0,1]t\in[0,1], stays uniformly close to the point v(s,0)=ψn^(uˇ(s,n^))v(s,0)=\psi^{\hat{n}}(\check{u}(s,\hat{n})). When applying the flow φt\varphi^{t} to (0,v(s,0))(0,v(s,0)) and (0,v(s,t))(0,v(s,t)) for 0t10\leq t\leq 1, this means with (14) that, given η>0\eta>0, there exist ϵ>0\epsilon>0 such that if E(u)<ϵE(u)<\epsilon, then (13) holds for 0t10\leq t\leq 1. This shows the first assertion. Moreover, δ\delta and cc above can be chosen to vary continuously in L0L_{0} and L1L_{1} in the C2C^{2} topology, cf. [34, Rmk. 4.3.2], and are defined even for degenerate pairs (L0,L1)(L_{0},L_{1}). The second assertion of the lemma follows. ∎

We will also use the following, slightly more general statement, which can be obtained as a direct consequence of Lemma 3.1.

Corollary 3.2.

For all η>0\eta>0, BB\in\mathbbm{N}, there exists ϵ=ϵ(η,B)>0\epsilon=\epsilon(\eta,B)>0 such that if uJkt(x^,x^+,L0,ψk(L1))u\in\mathcal{M}_{J_{kt}}(\hat{x}_{-},\hat{x}_{+},L_{0},\psi^{-k}(L_{1})) (for some x^±\hat{x}_{\pm} and kk\in\mathbbm{N}) and E(u)<ϵE(u)<\epsilon, then for all ss\in\mathbbm{R} and all t^[0,k]\hat{t}\in[0,k], t[B,B]t\in[-B,B] with t+t^[0,k]t+\hat{t}\in[0,k],

(15) d¯(u¯(s,t+t^),φt(u¯(s,t^)))<η.\displaystyle\overline{d}(\overline{u}(s,t+\hat{t}),\varphi^{t}(\overline{u}(s,\hat{t})))<\eta.

In particular, for all ss\in\mathbbm{R}, and all n^,n\hat{n},n\in\mathbbm{Z}, 0n^k0\leq\hat{n}\leq k, BnB-B\leq n\leq B, 0n+n^k0\leq n+\hat{n}\leq k,

(16) d(u^(s,n+n^),ψn(u^(s,n^)))<η.\displaystyle d(\hat{u}(s,n+\hat{n}),\psi^{n}(\hat{u}(s,\hat{n})))<\eta.

Note that a η\eta-pseudo-orbit (for ψ\psi) is a sequence {yn}n\{y_{n}\}_{n\in\mathbbm{Z}} in MM for which d(yn,ψ(yn1))<ηd(y_{n},\psi(y_{n-1}))<\eta for all nn\in\mathbbm{Z}. By Lemma 3.1, if E(u)E(u) is sufficiently small (independent of kk), then the \mathbbm{R}-family of orbit segments u(s,n)u(s,n), 0nk0\leq n\leq k, ss\in\mathbbm{R}, can be completed to a family of η\eta-pseudo-orbits.

3.2. Local stable/unstable manifolds and growth of chords

We recall some facts on local stable and unstable manifolds in hyperbolic sets, see e.g. [28, §6.4]. Let first, in general, ψ:MM\psi:M\to M be a CC^{\infty}-diffeomorphism on a closed manifold MM. Let KK be a compact hyperbolic set for ψ\psi. For qKq\in K, we define the local stable resp. unstable manifold of qq of size δ\delta as

Wδs(q)={yM|d(ψn(y),ψn(q))δ for all n0},W^{s}_{\delta}(q)=\{y\in M\,|\,d(\psi^{n}(y),\psi^{n}(q))\leq\delta\text{ for all }n\geq 0\},
Wδu(q)={yM|d(ψn(y),ψn(q))δ for all n0}.W^{u}_{\delta}(q)=\{y\in M\,|\,d(\psi^{n}(y),\psi^{n}(q))\leq\delta\text{ for all }n\leq 0\}.

There exists δ=δ(K)>0\delta^{*}=\delta^{*}(K)>0 such that for all δδ\delta\leq\delta^{*}, qKq\in K, the local stable and unstable manifolds Wδs(q)W^{s}_{\delta}(q) resp.  Wδu(q)W^{u}_{\delta}(q) are embedded disks, and there exist λ,c>0\lambda,c>0 such that

(17) ψn(Wδs(q))Wδceλns(ψn(q)) for all n0,\displaystyle\psi^{n}(W^{s}_{\delta}(q))\subset W^{s}_{\delta ce^{-\lambda n}}(\psi^{n}(q))\text{ for all }n\geq 0,
(18) ψn(Wδu(q))Wδceλnu(ψn(q)) for all n0.\displaystyle\psi^{n}(W^{u}_{\delta}(q))\subset W^{u}_{\delta ce^{\lambda n}}(\psi^{n}(q))\text{ for all }n\leq 0.

Furthermore, δ\delta^{*} is an expansivity constant for KK: for any x,yKx,y\in K, if for all kk\in\mathbbm{Z}, d(ψk(x),ψk(y))δd(\psi^{k}(x),\psi^{k}(y))\leq\delta^{*}, then x=yx=y. The local stable resp.  unstable disks at qKq\in K vary continuously in qKq\in K in the CC^{\infty} topology. Note that by (17) and (18), if (M,ω)(M,\omega) is symplectic and ψω=ω\psi^{*}\omega=\omega, we must have that ω|Wδs(q)=0\omega|_{W^{s}_{\delta^{*}}(q)}=0 and ω|Wδu(q)=0\omega|_{W^{u}_{\delta^{*}}(q)}=0, i.e., Wδs(q){W^{s}_{\delta^{*}}(q)} and Wδu(q){W^{u}_{\delta^{*}}(q)} are Lagrangian.

For xMx\in M, we denote by 𝒪(x)={ψn(x)|n}\mathcal{O}(x)=\{\psi^{n}(x)\,|\,n\in\mathbbm{Z}\} the full orbit of xx and by 𝒮m(x):={x,ψ(x),,ψm(x)}\mathcal{S}^{m}(x):=\{x,\psi(x),\ldots,\psi^{m}(x)\}, m0m\geq 0, the orbit segment of length mm starting at xx. If additionally 𝒮m(x)K\mathcal{S}^{m}(x)\subset K, and xWδu(q)x\in W^{u}_{\delta}(q), ψm(x)Wδs(p)\psi^{m}(x)\in W^{s}_{\delta}(p) for some q,pKq,p\in K, then we say that 𝒮m(x)\mathcal{S}^{m}(x) is a (q,p,δ)(q,p,\delta)-chord of length mm. Note that for x,q,pKx,q,p\in K, if d(ψn(x),ψn(q))δd(\psi^{n}(x),\psi^{n}(q))\leq\delta for all n0n\leq 0 and d(ψn+m(x),ψn+m(p))δd(\psi^{n+m}(x),\psi^{n+m}(p))\leq\delta for all n0n\geq 0, then 𝒮(x)\mathcal{S}(x) is a (q,p,δ)(q,p,\delta)-chord of length mm.

For q,pKq,p\in K, δ>0\delta>0, and mm\in\mathbbm{N}, denote by N(q,p,δ,m)N(q,p,\delta,m) the number of (q,p,δ)(q,p,\delta)-chords of length at most mm. The following is a relative version of the well-known result that htop(ψ|K)h_{\mathrm{top}}(\psi|_{K}) equals the exponential growth rate of periodic orbits in KK, if KK is additionally locally maximal (see Thm. 18.5.1 in [28]).

Proposition 3.3.

If KK is a locally maximal, topologically transitive hyperbolic set, then for any q,pKq,p\in K, 0<δδ=δ(K)0<\delta\leq\delta^{*}=\delta^{*}(K),

lim supmlogN(q,p,δ,m)m=htop(ψ|K).\limsup_{m\to\infty}\frac{\log N(q,p,\delta,m)}{m}=h_{\mathrm{top}}(\psi|_{K}).
Proof.

The proof is a straightforward adaption of the proof of Thm. 18.5.1 in [28]. We sketch it for the convenience of the reader. The inequality lim supmlog(N(q,p,δ,m))mhtop(ψ|K)\limsup_{m\to\infty}\frac{\log(N(q,p,\delta,m))}{m}\leq h_{\mathrm{top}}(\psi|_{K}) follows from the fact that ψ|K\psi|_{K} is expansive with expansivity constant δ\delta^{*}. Indeed, for any two distinct (q,p,δ)(q,p,\delta)-chords 𝒮m(x1)\mathcal{S}^{m}(x_{1}) and 𝒮m(x2)\mathcal{S}^{m}(x_{2}) of length m1m\geq 1, there exist 1km1\leq k\leq m such that d(ψk(x1),ψk(x2))>δd(\psi^{k}(x_{1}),\psi^{k}(x_{2}))>\delta^{*}. Otherwise, d(ψk(x1),ψk(x2))δd(\psi^{k}(x_{1}),\psi^{k}(x_{2}))\leq\delta^{*} for all kk\in\mathbbm{Z}, and hence x1=x2x_{1}=x_{2}.

For the reverse inequality, assume first that f|Kf|_{K} is topologically mixing. We recall Bowen’s specification theorem [11, Thm. 2.10]. A specification is a pair s=(τ,P)s=(\tau,P), where τ={I1,,Il}\tau=\{I_{1},\ldots,I_{l}\} is a finite collection of disjoint intervals of integers, and P:I:=i=1lIlKP:I:=\bigcup_{i=1}^{l}I_{l}\to K is map with ψk2k1(P(k1))=P(k2)\psi^{k_{2}-k_{1}}(P(k_{1}))=P(k_{2}) whenever k1,k2Ijk_{1},k_{2}\in I_{j} for some 1jl1\leq j\leq l. A specification s=(τ,P)s=(\tau,P) is said to be nn-delayed if there is an interval of length at least nn between every pair of intervals belonging to τ\tau. For ϵ>0\epsilon>0 and a specification s=(τ,P)s=(\tau,P), let

V(s,ϵ):={xK|d(ψk(x),P(k))<ϵ for all kI}.V(s,\epsilon):=\{x\in K\,|\,d(\psi^{k}(x),P(k))<\epsilon\text{ for all }k\in I\}.

Since f|Kf|_{K} is assumed to be topologically mixing, Bowen’s specification theorem asserts that for any ϵ>0\epsilon>0, there exist MϵM_{\epsilon}\in\mathbbm{N} such that V(s,ϵ)V(s,\epsilon)\neq\emptyset for any MϵM_{\epsilon}-delayed specification ss. If s=(τ,P)s=(\tau,P) with τ={I1,,Il}\tau=\{I_{1},\ldots,I_{l}\} is such that I1={a}I_{1}=\{a\} and Il={b}I_{l}=\{b\} with P(a)=qP(a)=q and P(b)=pP(b)=p, then in fact, for any fixed δ<δ\delta<\delta^{*} one can choose Mϵ=Mϵ(δ)M_{\epsilon}=M_{\epsilon}(\delta) sufficiently large, such that the orbit 𝒪(x)\mathcal{O}(x) of xV(s,ϵ)x\in V(s,\epsilon) that is obtained in the proof in [11] contains by construction a segment that is a (q,p,δ)(q,p,\delta)-chord of length bab-a. It follows that for any point yy in a (n,ϵ)(n,\epsilon)-separated set EKE\subset K, there exists a (q,p,δ)(q,p,\delta)-chord 𝒮(x)\mathcal{S}(x) of length n+2Mϵ/2n+2M_{\epsilon/2} such that d(ψMϵ/2+k(x),ψk(y))<ϵ/2d(\psi^{M_{\epsilon/2}+k}(x),\psi^{k}(y))<\epsilon/2 for all 0kn0\leq k\leq n. Hence there are at least #E\#E-many (q,p,δ)(q,p,\delta)-chords of length at most n+2Mϵ/2n+2M_{\epsilon/2}. This implies that lim supmlogN(q,p,δ,m)mhtop(ψ|K)\limsup_{m\to\infty}\frac{\log N(q,p,\delta,m)}{m}\geq h_{\mathrm{top}}(\psi|_{K}) in the topologically mixing case. The generalisation to the topologically transitive case follows then from the spectral decomposition theorem, see [28, Thm. 18.3.1]. ∎

Proof of Theorem 1.

Let KK be a locally maximal, topological transitive, compact hyperbolic set. By compactness of KK and since UK=\partial U^{\prime}\cap K=\emptyset for any isolating neighbourhood UU^{\prime} of KK, we can choose δ>0\delta^{\prime}>0 such that

  1. (i)

    U=Nδ(K):={yM|infxKd(y,x)δ}U=N_{\delta^{\prime}}(K):=\{y\in M\,|\,\inf_{x\in K}d(y,x)\leq\delta^{\prime}\} is an isolating neighbourhood for KK,

  2. (ii)

    δδ/2\delta^{\prime}\leq\delta^{*}/2, where δ=δ(K)\delta^{*}=\delta^{*}(K) is the constant discussed above.

Let (L0,L1)(L_{0},L_{1}) be an admissible pair with Wδu(q)L0W^{u}_{\delta}(q)\subset L_{0}, and Wδs(p)L1W^{s}_{\delta}(p)\subset L_{1}, for some q,pKq,p\in K, δ(0,δ]\delta\in(0,\delta^{\prime}]. Assume first that (L0,L1)(L_{0},L_{1}) is also non-degenerate. Let JJ be defined as in (12). By Proposition 3.3, and by Prop. 3.8 in [13] (cf. Lemma 4.4), it is sufficient to show that there exists ϵ>0\epsilon>0 such that if uJkt(x^,x^+,L0,ψk(L1))u\in\mathcal{M}_{J_{kt}}(\hat{x}_{-},\hat{x}_{+},L_{0},\psi^{-k}(L_{1})), kk\in\mathbbm{N}, with x^=(x,[u])\hat{x}_{-}=(x_{-},[u_{-}]), x^+=(x+,[u+])𝒫~k\hat{x}_{+}=(x_{+},[u_{+}])\in\widetilde{\mathcal{P}}_{k}, satisfies that

  • 𝒮k(x(0))\mathcal{S}^{k}(x_{-}(0)) or 𝒮k(x+(0))\mathcal{S}^{k}(x_{+}(0)) is a (q,p,δ)(q,p,\delta)-chord;

  • E(u)<ϵE(u)<\epsilon,

then uu is constant. Choose η>0\eta>0 such that

  1. (1)

    any η\eta-pseudo-orbit that intersects U\partial U is not entirely contained in U¯\overline{U};

  2. (2)

    any continuous family of η\eta-pseudo-orbits 𝒬s={yks}k\mathcal{Q}_{s}=\{y^{s}_{k}\}_{k\in\mathbbm{Z}}, ss\in\mathbbm{R}, in UU is δ\delta-shadowed by a continuous family of orbits (𝒪(xs))s(\mathcal{O}(x^{s}))_{s\in\mathbbm{R}} in KK, i.e., d(yks,ψk(xs))<δd(y^{s}_{k},\psi^{k}(x^{s}))<\delta for all kk\in\mathbbm{Z}, ss\in\mathbbm{R}.

The point (1) can be achieved by compactness of U\partial U and since UK=\partial U\cap K=\emptyset, the point (2) by the shadowing theorem (see e.g. Thm. 18.1.3 in [28]).

Choose ϵ=ϵ(η)\epsilon=\epsilon(\eta) according to Lemma 3.1, and such that ϵ<𝔥:=𝔥(M,Li)\epsilon<\mathfrak{h}:=\mathfrak{h}(M,L_{i}), i=0,1i=0,1. Let kk\in\mathbbm{N}, uJkt(x^,x^+,L0,ψk(L1))u\in\mathcal{M}_{J_{kt}}(\hat{x}_{-},\hat{x}_{+},L_{0},\psi^{-k}(L_{1})) such that E(u)<ϵE(u)<\epsilon, and assume that 𝒮k(x+(0))\mathcal{S}^{k}(x_{+}(0)) is a (q,p,δ)(q,p,\delta)-chord, the case that 𝒮k(x(0))\mathcal{S}^{k}(x_{-}(0)) is a (q,p,δ)(q,p,\delta)-chord is treated similarly. For ss\in\mathbbm{R}, we set ys:=u^(s,0)L0y^{s}:=\hat{u}(s,0)\in L_{0} and consider the sequence (yns)n(y^{s}_{n})_{n\in\mathbbm{Z}} in MM, given by

yns:={u^(s,n) if 0nk,ψn(ys) if n<0,ψnk(u^(s,k)) if n>k.\displaystyle y^{s}_{n}:=\begin{cases}\hat{u}(s,n)&\text{ if }0\leq n\leq k,\\ \psi^{n}(y^{s})&\text{ if }n<0,\\ \psi^{n-k}(\hat{u}(s,k))&\text{ if }n>k.\end{cases}

By Lemma 3.1, 𝒬s:={yns}n\mathcal{Q}_{s}:=\{y^{s}_{n}\}_{n\in\mathbbm{Z}}, ss\in\mathbbm{R}, is a continuous family of η\eta-pseudo-orbits for ψ\psi. Note that since y0s=ysx+(0)y^{s}_{0}=y^{s}\to x_{+}(0) as ss\to\infty, ysL0y^{s}\in L_{0}, yksL1y^{s}_{k}\in L_{1}, we have that y0sWδu(q)y_{0}^{s}\in W^{u}_{\delta}(q) and yksWδs(p)y^{s}_{k}\in W^{s}_{\delta}(p) as ss sufficiently large. Since δδ\delta\leq\delta^{*} and Nδ(K)UN_{\delta}(K)\subset U, it follows that 𝒬sU\mathcal{Q}_{s}\subset U for ss sufficiently large. In fact, 𝒬sU\mathcal{Q}_{s}\subset U for all ss\in\mathbbm{R}. Otherwise, if s0:=inf{s|𝒬sU}>s_{0}:=\inf\{s\in\mathbbm{R}\,|\,\mathcal{Q}_{s}\subset U\}>-\infty, then 𝒬s0\mathcal{Q}_{s_{0}} intersects U\partial U which contradicts (1). Hence, we can find by (2) a continuous family of orbits (𝒪(xs))s(\mathcal{O}(x^{s}))_{s\in\mathbbm{R}} in KK that δ\delta-shadows (𝒬s)s(\mathcal{Q}_{s})_{s\in\mathbbm{R}}. It follows that (𝒮k(xs))s(\mathcal{S}^{k}(x^{s}))_{s\in\mathbbm{R}} is a continuous family of (q,p,2δ)(q,p,2\delta)-chords. Note that the orbit 𝒪(x+(0))\mathcal{O}(x_{+}(0)) δ\delta-shadows the pseudo-orbit 𝒬s\mathcal{Q}_{s} if ss is sufficiently large. By (ii), 2δ2\delta is an expansivity constant, and therefore in fact xs=x+(0)x^{s}=x_{+}(0) for ss sufficiently large. Similarly, since also x(0)Kx_{-}(0)\in K, we have that xs=x(0)x^{s}=x_{-}(0) for ss sufficiently small. Since ψk(W2δu(q))\psi^{k}(W^{u}_{2\delta}(q)) intersects W2δs(p)W^{s}_{2\delta}(p) transversally, the starting points of (q,p,2δ)(q,p,2\delta)-chords of length kk are isolated. It follows that the path (xs)s(x^{s})_{s\in\mathbbm{R}} is constant, so x(0)=x+(0)x_{-}(0)=x_{+}(0). Since E(u)<ϵ<𝔥E(u)<\epsilon<\mathfrak{h}, it follows that E(u)=0E(u)=0, and we conclude that uu is constant.

This concludes the statement in the non-degenerate case. To pass over to a degenerate pair (L0,L1)(L_{0},L_{1}), note the following. Since the iterations of the local unstable manifold already intersect the local stable manifold transversally, it is possible to perturb L0L_{0} and L1L_{1} outside the segments WδsW_{\delta}^{s} resp. WδuW_{\delta}^{u} in the CC^{\infty} topology to a non-degenerate admissible pair. Since ϵ>0\epsilon>0 can be chosen to be constant in a neighbourhood of (L0,L1)(L_{0},L_{1}), the assertion of the Theorem also holds for the degenerate pair (L0,L1)(L_{0},L_{1}). ∎

4. Proof of Theorem 3: Links in Horseshoes and crossing energy

4.1. Restricting the Floer chain complex to link complements.

We now discuss restrictions of the Lagrangian Floer chain complexes to link complements if MM is a surface. By passing to the suspension flow φ\varphi on Y=S1×MY=S^{1}\times M, every finite union of periodic orbits 𝒪M\mathcal{O}\subset M of ψ\psi, that is, a finite invariant set of ψ\psi, yields a link (or braid) =𝒪Y\mathcal{L}=\mathcal{L}_{\mathcal{O}}\subset Y of periodic orbits of φ\varphi, and vice versa, see §2.1. For such a set 𝒪M\mathcal{O}\subset M and two closed embedded curves L0,L1M𝒪L_{0},L_{1}\subset M\setminus\mathcal{O}, we write Λi:=ι(Li)\Lambda_{i}:=\iota(L_{i}), i=0,1i=0,1, where ι:MY,x(0,x)\iota:M\to Y,x\mapsto(0,x), and denote by 𝒵+(Y,𝒪,Λ0,Λ1)\mathcal{Z}^{+}(Y,\mathcal{L}_{\mathcal{O}},\Lambda_{0},\Lambda_{1}) the set of homotopy classes of paths from Λ0\Lambda_{0} to Λ1\Lambda_{1} in Y𝒪Y\setminus\mathcal{L}_{\mathcal{O}} that are everywhere transverse to the horizontal surfaces {t}×M\{t\}\times M. Two isotopies (Lit)t[0,1](L^{t}_{i})_{t\in[0,1]}, i=0,1i=0,1, supported in M𝒪M\setminus\mathcal{O} from LiL_{i} to LiL_{i}^{\prime} define a bijection

Υ=ΥL0,L0L1,L1:𝒵+(Y,𝒪,Λ0,Λ1)𝒵+(Y,𝒪,Λ0,Λ1),\displaystyle\Upsilon=\Upsilon^{L_{1},L^{\prime}_{1}}_{L_{0},L^{\prime}_{0}}:\mathcal{Z}^{+}(Y,\mathcal{L}_{\mathcal{O}},\Lambda_{0},\Lambda_{1})\to\mathcal{Z}^{+}(Y,\mathcal{L}_{\mathcal{O}},\Lambda^{\prime}_{0},\Lambda^{\prime}_{1}),

by sending a path η^:[0,R]Y\hat{\eta}:[0,R]\to Y to the glued path (ια¯)#η^#(ιβ)(\iota\circ\overline{\alpha})\#\hat{\eta}\#(\iota\circ\beta), where α\alpha resp.  β\beta are paths in MM obtained by restricting the isotopy of L0L_{0} resp. L1L_{1} to the point η(0)L0\eta(0)\in L_{0} resp. η(1)L1\eta(1)\in L_{1}, and α¯\overline{\alpha} is the reverse path of α\alpha. If 𝒪\mathcal{O}\neq\emptyset and L0,L1L_{0},L_{1} are non-contractible in M𝒪M\setminus\mathcal{O}, then the map ΥL0,L0L1,L1\Upsilon^{L_{1},L^{\prime}_{1}}_{L_{0},L_{0}^{\prime}} does not depend on the choice of isotopies. Write ΥL0,L0=ΥL0,L0L1,L1\Upsilon_{L_{0},L^{\prime}_{0}}=\Upsilon^{L_{1},L_{1}}_{L_{0},L^{\prime}_{0}} and ΥL1,L1=ΥL0,L0L1,L1\Upsilon^{L_{1},L^{\prime}_{1}}=\Upsilon^{L_{1},L^{\prime}_{1}}_{L_{0},L_{0}}. Then, ΥL0,L0L1,L1=ΥL1,L1ΥL0,L0\Upsilon^{L_{1},L^{\prime}_{1}}_{L_{0},L^{\prime}_{0}}=\Upsilon^{L_{1},L^{\prime}_{1}}\circ\Upsilon_{L_{0},L^{\prime}_{0}}.

Let (L0,L1)(L_{0},L_{1}) be an admissible pair. Note that if MS2M\neq S^{2}, then any closed non-contractible curve LL is monotone, and if M=S2M=S^{2}, then any equator LL is, which means that LL divides S2S^{2} into two components of equal area. In the special situation of surfaces, we can define chain complexes (CFρ(L0,ψk(L1)),ρ)(\mathrm{CF}^{\rho}(L_{0},\psi^{-k}(L_{1})),\partial^{\rho}) associated to an element ρ𝒵+(Y,𝒪,Λ0,Λ1)\rho\in\mathcal{Z}^{+}(Y,\mathcal{L}_{\mathcal{O}},\Lambda_{0},\Lambda_{1}). This construction is analogous to that of Legendrian contact homology in link complements in [2], see also [9, §5], although arguments simplify in the setting of Lagrangian Floer homology. We also refer to [49] (Hamiltonian Floer homology in braid complements) and [37] (cylindrical contact homology in link complements) for related homology theories. For a pair x^=(x,[vx])𝒫~k\hat{x}=(x,[v_{x}])\in\widetilde{\mathcal{P}}_{k}, we use the notation x^ρ\hat{x}\in\rho if the curve t([kt],ϕHkt(x))S1×Mt\mapsto([kt],\phi^{kt}_{H}(x))\in S^{1}\times M, t[0,1]t\in[0,1], represents ρ\rho. Let CFρ(L0,ψk(L1))CF(L0,ψk(L1))\mathrm{CF}^{\rho}(L_{0},\psi^{-k}(L_{1}))\subset\mathrm{CF}(L_{0},\psi^{-k}(L_{1})) be the subspace generated by x^ρ\hat{x}\in\rho, and define ρ:CFρ(L0,ψk(L1))CF(L0,ψk(L1))\partial^{\rho}:\mathrm{CF}^{\rho}(L_{0},\psi^{-k}(L_{1}))\to\mathrm{CF}(L_{0},\psi^{-k}(L_{1})) in the same way as \partial with the additional requirement that for the holomorphic curves uu that appear in the definition, the map u~\widetilde{u} does not intersect the union of the trivial cylinders over 𝒪\mathcal{L}_{\mathcal{O}}, that is, it does not intersect the set Cyl𝒪:={(s,t,ϕHt(p))|s,tS1,p𝒪}×S1×M\mathrm{Cyl}_{\mathcal{O}}:=\{(s,t,\phi_{H}^{t}(p))\,|\,s\in\mathbbm{R},t\in S^{1},p\in\mathcal{O}\}\subset\mathbbm{R}\times S^{1}\times M. We note that the lift Cyl𝒪~\widetilde{\mathrm{Cyl}_{\mathcal{O}}} of Cyl𝒪\mathrm{Cyl}_{\mathcal{O}} to ××M\mathbbm{R}\times\mathbbm{R}\times M is the image of a finite collection of J~\widetilde{J}-holomorphic curves. In order that the definition of ρ\partial^{\rho} makes sense, one has to verify that the relevant moduli spaces, when restricting to only those curves uu above, are compact in the ClocC^{\infty}_{\mathrm{loc}} topology. This follows from the ClocC^{\infty}_{\mathrm{loc}}-compactness of the moduli-spaces without restriction and the positivity and stability of intersections of J~\widetilde{J}-holomorphic curves, see [9] for details of the argument. Note that for curves uu relevant for the definition of ρ\partial^{\rho}, the map su¯(s,t)s\mapsto\overline{u}(s,t) provides a homotopy of its asymptotics in Y𝒪Y\setminus\mathcal{L}_{\mathcal{O}} relative to (Λ0,Λ1)(\Lambda_{0},\Lambda_{1}), hence in fact im(ρ)CFρ(L0,ψk(L1))\mathrm{im}(\partial^{\rho})\subset\mathrm{CF}^{\rho}(L_{0},\psi^{-k}(L_{1})). Similarly, by using again the positivity of intersection property of J~\widetilde{J}-holomorphic cylinders, the argument that 2=0\partial^{2}=0 carries over and one can show that (ρ)2=0(\partial^{\rho})^{2}=0. We denote the resulting homology by HFρ(L0,ψk(L1))\mathrm{HF}^{\rho}(L_{0},\psi^{-k}(L_{1})). The homology is invariant under Hamiltonian isotopy L0t=ϕFt(L0)L^{t}_{0}=\phi^{t}_{F}(L_{0}), t[0,1]t\in[0,1], from L0L_{0} to L0=L01L^{\prime}_{0}=L^{1}_{0}, as long as it is supported in M𝒪M\setminus\mathcal{O}, which means that ϕFt\phi^{t}_{F} fixes 𝒪\mathcal{O} for all t[0,1]t\in[0,1]: restricting the count of holomorphic curves uu to curves such that u~\widetilde{u} is disjoint from Cyl𝒪\mathrm{Cyl}_{\mathcal{O}}, defines a map f0ρ:CFρ(L0,ψk(L1))CFΥ(ρ)(L0,ψk(L1))f_{0}^{\rho}:\mathrm{CF}^{\rho}(L_{0},\psi^{-k}(L_{1}))\to\mathrm{CF}^{\Upsilon(\rho)}(L_{0}^{\prime},\psi^{-k}(L_{1})), where Υ=ΥL0,L0\Upsilon=\Upsilon_{L_{0},L_{0}^{\prime}}. And one can show, again using positivity and stability of intersections, that fρf^{\rho} is a chain map. With g0ρ:CFΥ(ρ)(L0,ψk(L1))CFρ(L0,ψk(L1))g_{0}^{\rho}:\mathrm{CF}^{\Upsilon(\rho)}(L^{\prime}_{0},\psi^{-k}(L_{1}))\to\mathrm{CF}^{\rho}(L_{0},\psi^{-k}(L_{1})) similarly defined, g0Υ(ρ)f0ρg_{0}^{\Upsilon(\rho)}\circ f_{0}^{\rho} is chain homotopic to the identity. Hence f0ρf_{0}^{\rho} induces an isomorphism

(19) HFρ(L0,ψk(L1))HFΥ(ρ)(L0,ψk(L1)).\displaystyle\mathrm{HF}^{\rho}(L_{0},\psi^{-k}(L_{1}))\overset{\cong}{\longrightarrow}\mathrm{HF}^{\Upsilon(\rho)}(L^{\prime}_{0},\psi^{-k}(L_{1})).

Similarly, for a Hamiltonian isotopy L1tL^{t}_{1}, t[0,1]t\in[0,1], from L1L_{1} to L1=L11L^{\prime}_{1}=L^{1}_{1} that is supported in M𝒪M\setminus\mathcal{O}, we obtain an isomorphism

(20) HFρ(L0,ψk(L1))HFΥ(ρ)(L0,ψk(L1)),\displaystyle\mathrm{HF}^{\rho}(L_{0},\psi^{-k}(L_{1}))\overset{\cong}{\longrightarrow}\mathrm{HF}^{\Upsilon(\rho)}(L_{0},\psi^{-k}(L^{\prime}_{1})),

where now Υ=ΥL1,L1\Upsilon=\Upsilon^{L_{1},L_{1}^{\prime}}. Finally, the above isomorphisms behave naturally under composition and are independent of the choice of isotopy.

4.2. Horseshoes for surface diffeomorphisms and crossing energy

In this section we discuss specific horseshoes KK in surface diffeomorphisms and obtain a uniform energy estimate of some Floer strips whose suspension crosses a link induced by certain horseshoe orbits in KK.

Let ψ:MM\psi:M\to M be a surface diffeomorphism with htop(ψ)>0h_{\mathrm{top}}(\psi)>0. Then, by Katok’s theorem [28, Thm. S.5.9], for any e>0e>0 with e<htop(ψ)e<h_{\mathrm{top}}(\psi), there exists a locally maximal hyperbolic set KK such that

(21) htop(ψ|K)htop(ψ)e.\displaystyle h_{\mathrm{top}}(\psi|_{K})\geq h_{\mathrm{top}}(\psi)-e.

We now discuss how one can choose the sets KK more specifically. By the construction of Katok and Mendoza [28, Thm. S.5.9], there exists a collection of finitely many pairwise disjoint rectangles D1,,DLD_{1},\ldots,D_{L} in MM such that KK is obtained as the union m=0N1ψm(K)\bigcup_{m=0}^{N-1}\psi^{m}(K^{\prime}), where K=jψNj(U)K^{\prime}=\bigcap_{j\in\mathbbm{Z}}\psi^{Nj}({U}) is the ψN\psi^{N}-invariant set induced from the isolating neighbourhood U=j=1LDjU=\bigcup_{j=1}^{L}{D}_{j}. We write the rectangles DjD_{j} as images of embeddings ιj:[0,1]2M\iota_{j}:[0,1]^{2}\to M, and the boundaries as Dj=vjv+jhjh+j\partial D_{j}=v^{j}_{-}\cup v^{j}_{+}\cup h^{j}_{-}\cup h^{j}_{+} with vj:=ιj({0}×[0,1])v^{j}_{-}:=\iota_{j}(\{0\}\times[0,1]), v+j:=ιj({1}×[0,1])v^{j}_{+}:=\iota_{j}(\{1\}\times[0,1]), hj:=ιj([0,1]×{0})h^{j}_{-}:=\iota_{j}([0,1]\times\{0\}), and h+j:=ιj([0,1]×{1})h^{j}_{+}:=\iota_{j}([0,1]\times\{1\}). Denote by Σ={a¯=(aj)j,aj{Ω1,,ΩL}}\Sigma=\{\underline{a}=(a_{j})_{j\in\mathbbm{Z}},a_{j}\in\{\Omega_{1},\ldots,\Omega_{L}\}\} the set of bi-infinite sequences in LL symbols {Ω1,,ΩL}.\{\Omega_{1},\ldots,\Omega_{L}\}. For any iji\neq j, ψN(Dj)\psi^{N}(D_{j}) intersects DiD_{i}, and for any element a¯=(aj)j=(Ωij)jΣ\underline{a}=(a_{j})_{j\in\mathbbm{Z}}=(\Omega_{i_{j}})_{j\in\mathbbm{Z}}\in\Sigma, there exists a unique point π(a¯)=jψjN(Dij)K\pi(\underline{a})=\bigcap_{j\in\mathbbm{Z}}\psi^{-jN}(D_{i_{j}})\in K^{\prime} which defines a continuous map π:ΣK\pi:\Sigma\to K^{\prime} with πσ=ψN|Kπ\pi\circ\sigma=\psi^{N}|_{K^{\prime}}\circ\pi. Moreover, for a sequence a¯=(aj)j0=(Ωij)j\underline{a}=(a_{j})_{j\leq 0}=(\Omega_{i_{j}})_{j\in\mathbbm{Z}},

l0=l0(a¯):=j0ψjN(Dij)l_{0}=l_{0}(\underline{a}):=\bigcap_{j\leq 0}\psi^{-jN}(D_{i_{j}})

is a segment in Di0D_{i_{0}} that connects v±i0v^{i_{0}}_{\pm} without intersecting h±i0h^{i_{0}}_{\pm} and is contained in a local unstable manifold of a point qKq\in K^{\prime}. Similarly, for a sequence b¯=(bj)j0=(Ωij)j\underline{b}=(b_{j})_{j\geq 0}=(\Omega_{i^{\prime}_{j}})_{j\in\mathbbm{Z}},

l1=l1(b¯)=j0ψ(j1)N(Dij)l_{1}=l_{1}(\underline{b})=\bigcap_{j\geq 0}\psi^{-(j-1)N}(D_{i^{\prime}_{j}})

is a segment in ψN(Di0)\psi^{N}(D_{i^{\prime}_{0}}) that connects ψN(h±i0)\psi^{N}(h^{i^{\prime}_{0}}_{\pm}) without intersecting ψN(v±i0)\psi^{N}(v^{i^{\prime}_{0}}_{\pm}) and is contained in a local stable manifold of some point pKp\in K^{\prime}. Fix a¯=(aj)j0\underline{a}=(a_{j})_{j\leq 0} and b¯=(bj)j0\underline{b}=(b_{j})_{j\geq 0}. Write a0=Ωi0a_{0}=\Omega_{i_{0}}, D0:=Di0D^{0}:=D_{i_{0}}, b0=Ωi0b_{0}=\Omega_{i^{\prime}_{0}}, D1:=Di0D^{1}:=D_{i^{\prime}_{0}}. For any (n2)(n-2)-tuple (i1,,in2)(i_{1},\ldots,i_{n-2}), n3n\geq 3, consider the unique point x=x(i1,,in2)Mx=x_{(i_{1},\ldots,i_{n-2})}\in M,

x:=l0j=1n2ψjN(Dij)ψ(n1)N(ψN(l1))=π(c¯),x:=l_{0}\cap\bigcap_{j=1}^{n-2}\psi^{-jN}(D_{i_{j}})\cap\psi^{-(n-1)N}(\psi^{-N}(l_{1}))=\pi(\underline{c}),

where c¯=(cj)j\underline{c}=(c_{j})_{j\in\mathbbm{Z}} is given by

(22) cj={aj, if j0,Ωij, if 1jn2,bj(n1), if jn1.\displaystyle c_{j}=\begin{cases}a_{j},&\text{ if }j\leq 0,\\ \Omega_{i_{j}},&\text{ if }1\leq j\leq n-2,\\ b_{j-(n-1)},&\text{ if }j\geq n-1.\end{cases}

Let C(i1,,in2):[0,nN]S1×MC_{(i_{1},\ldots,i_{n-2})}:[0,nN]\to S^{1}\times M be the φ\varphi-chord from λ0={0}×l0\lambda_{0}=\{0\}\times l_{0} to λ1={0}×l1\lambda_{1}=\{0\}\times l_{1} given by C(i1,,in2)(t):=φt(x(i1,,in2))C_{(i_{1},\ldots,i_{n-2})}(t):=\varphi^{t}(x_{(i_{1},\ldots,i_{n-2})}).

As we obtained in [36] (in a more general setting), by choosing KK and U=j=1LDjU=\bigcup_{j=1}^{L}D_{j} suitably we can assume that additionally

  1. (I)

    ψm(U)U=\psi^{m}({U})\cap{U}=\emptyset, for all m=1,,N1,N+1,,2N1m=1,\ldots,N-1,N+1,\ldots,2N-1;

  2. (II)

    ψN(v±j)Di=\psi^{N}(v^{j}_{\pm})\cap D_{i}=\emptyset, for all i,j{1,,L}i,j\in\{1,\ldots,L\};

  3. (III)

    either ψ2N(v±i)=h±i\psi^{2N}(v^{i}_{\pm})=h^{i}_{\pm} or ψ2N(v±i)=hi\psi^{2N}(v^{i}_{\pm})=h^{i}_{\mp}, for all i{1,,L}i\in\{1,\ldots,L\}.

By (III), for each i=1,,Li=1,\ldots,L, we can glue DiD_{i} to the two surfaces E±i={([t],φt(x)|t[0,2N],xv±i}E^{i}_{\pm}=\{([t],\varphi^{t}(x)\,|\,t\in[0,2N],x\in v^{i}_{\pm}\} and obtain a piecewise immersed pair-of-pants FiF_{i} in S1×MS^{1}\times M. By (I) and (II), the surfaces FiF_{i}, i=1,,Li=1,\ldots,L, are in fact (piecewise) embedded and pairwise disjoint. The boundary of FiF_{i} is a link i=𝒪i\mathcal{L}_{i}=\mathcal{L}_{\mathcal{O}_{i}}, a lift of a collection of three periodic orbits 𝒪i\mathcal{O}_{i} given by iterates of the corners of DiD_{i} (there are two corners of DiD_{i} that belong to the same orbit). We equip FiF_{i} with the orientation induced by the vector field of φ\varphi along DiD_{i}. Let 𝒪=i=1L𝒪i\mathcal{O}=\bigcup_{i=1}^{L}\mathcal{O}_{i}, =𝒪=i=1Li\mathcal{L}=\mathcal{L}_{\mathcal{O}}=\bigcup_{i=1}^{L}\mathcal{L}_{i}. Fix l0l_{0} and l1l_{1} as above.

Lemma 4.1.

Assume L0L_{0} and L1L_{1} are closed curves in M𝒪M\setminus\mathcal{O} such that L0D0=l0L_{0}\cap D^{0}=l_{0}, L1ψN(D1)=l1L_{1}\cap\psi^{N}(D^{1})=l_{1}. Then, for any tuple (i1,,in2)(i_{1},\ldots,i_{n-2}), the chord C(i1,,in2)C_{(i_{1},\ldots,i_{n-2})} is the unique φ\varphi-chord in its homotopy class of paths from Λ0={0}×L0\Lambda_{0}=\{0\}\times L_{0} to Λ1={0}×L1\Lambda_{1}=\{0\}\times L_{1} in YY\setminus\mathcal{L} relative to (Λ1,Λ2)(\Lambda_{1},\Lambda_{2}).

Proof.

Let C^:[0,nN]Y\hat{C}:[0,nN]\to Y be a φ\varphi-chord from Λ0\Lambda_{0} to Λ1\Lambda_{1} homotopic to C(i1,,in2)C_{(i_{1},\ldots,i_{n-2})} in YY\setminus\mathcal{L} relative to (Λ1,Λ2)(\Lambda_{1},\Lambda_{2}). Then there exists 0t0<<tnnN0\leq t_{0}<\cdots<t_{n}\leq nN with C^(ti)Fij\hat{C}(t_{i})\in F_{i_{j}} and such that for all 0mn10\leq m\leq n-1 with im=im+1i_{m}=i_{m+1} the path C^|[tm,tm+1]\hat{C}|_{[t_{m},t_{m+1}]} is not homotopic relative to its endpoints to a path contained inside FimF_{i_{m}}. Hence, by (I)-(III) above, for any 0mn10\leq m\leq n-1, |tm+1tm|N|t_{m+1}-t_{m}|\geq N. Hence tm=mNt_{m}=mN, and C^=Ci1,,in2\hat{C}=C_{i_{1},\ldots,i_{n-2}} by construction. ∎

Fix a non-degenerate, admissible pair (L0,L1)(L_{0},L_{1}) that satisfies the assumptions of Lemma 4.1. Additionally assume that the algebraic intersection number of the curves Λ0={0}×L0\Lambda_{0}=\{0\}\times L_{0} and Λ1={0}×L1\Lambda_{1}=\{0\}\times L_{1} with any of the surfaces F1,,FLF_{1},\ldots,F_{L} is zero. This holds for example if LiL_{i}, i=0,1i=0,1, intersect those surfaces only in UU and ψN(U)i=1LFi\psi^{N}(U)\cap\bigcup_{i=1}^{L}F_{i}, and intersect the connected components of UU and ψN(U)\psi^{N}(U) in segments that connect opposite edges. Let 𝒵1+(l0,l1)𝒵(Y,𝒪,Λ1,Λ2)\mathcal{Z}^{+}_{1}(l_{0},l_{1})\subset\mathcal{Z}(Y,\mathcal{L}_{\mathcal{O}},\Lambda_{1},\Lambda_{2}) be the set of homotopy classes that can be represented by the chords C(i1,,in2)C_{(i_{1},\ldots,i_{n-2})} of φ\varphi of the form above.

Proposition 4.2.

Let (L0,L1)(L^{\prime}_{0},L^{\prime}_{1}) be a non-degenerate admissible pair of curves in M𝒪M\setminus\mathcal{O} such that LiL^{\prime}_{i} and LiL_{i} are Hamiltonian isotopic in M𝒪M\setminus\mathcal{O}, i=0,1i=0,1. Assume J=JtJ=J_{t} is of the form (12). Let Υ=ΥL0,L0L1,L1\Upsilon=\Upsilon^{L_{1},L^{\prime}_{1}}_{L_{0},L^{\prime}_{0}} be defined as in §4.1. There is ε>0\varepsilon>0 such that if uJkt(x^,x^+,L0,ψk(L1))u\in\mathcal{M}_{J_{kt}}(\hat{x}_{-},\hat{x}_{+},L^{\prime}_{0},\psi^{-k}(L^{\prime}_{1})), kk\in\mathbbm{N}, satisfies

  1. (1)

    x^Υ(ρ)\hat{x}_{-}\in\Upsilon(\rho) or x^+Υ(ρ)\hat{x}_{+}\in\Upsilon(\rho) for some ρ𝒵1+(l0,l1)\rho\in\mathcal{Z}^{+}_{1}(l_{0},l_{1}), and

  2. (2)

    im(u¯(s,))𝒪\mathrm{im}(\overline{u}(s,\cdot))\cap\mathcal{L}_{\mathcal{O}}\neq\emptyset\, for some ss\in\mathbbm{R},

then

E(u)ε.E(u)\geq\varepsilon.
Proof.

Given a homotopy class ρ\rho of paths from Λ0={0}×L0\Lambda_{0}=\{0\}\times L_{0} to Λ1={0}×L1\Lambda_{1}=\{0\}\times L_{1} in YY\setminus\mathcal{L}, denote by κ(ρ)\kappa(\rho) the maximal algebraic intersection number of a path in ρ{\rho} with i=0LFi\bigcup_{i=0}^{L}F_{i}. Here we naturally only consider transverse paths to i=0LFi\bigcup_{i=0}^{L}F_{i} and also count intersections at the endpoint of the paths. By our assumption that the algebraic intersection number of Λi\Lambda_{i}, i=1,2i=1,2, with any of the surfaces F1,,FLF_{1},\ldots,F_{L} vanishes, the number κ(ρ)\kappa(\rho) is well defined. Analogously define κ(ρ)\kappa^{\prime}(\rho^{\prime}) for any homotopy class ρ\rho^{\prime} of paths from {0}×L0\{0\}\times L_{0}^{\prime} to {0}×L1\{0\}\times L^{\prime}_{1} in YY\setminus\mathcal{L}. Fix some χ=χ(L0,L1)0\chi=\chi(L_{0}^{\prime},L_{1}^{\prime})\geq 0 such that for any ρ𝒵+(Y,,Λ0,Λ1)\rho\in\mathcal{Z}^{+}(Y,\mathcal{L},\Lambda_{0},\Lambda_{1}),

(23) κ(Υ(ρ))κ(ρ)χ.\displaystyle\kappa^{\prime}(\Upsilon(\rho))\geq\kappa(\rho)-\chi.

Let B:=4N(χ+3)B:=4N(\chi+3). Given η>0\eta>0, we set Diη:={xM|dist(x,Di)<η}D^{\eta}_{i}:=\{x\in M\,|\,\mathrm{dist\,}(x,D_{i})<\eta\}, 1iL1\leq i\leq L. By (I) and (II) we can choose η>0\eta>0 such that

  1. (a)

    η<dist(Diη,ψn(Djη))\eta<\mathrm{dist\,}(D^{\eta}_{i},\psi^{n}(D^{\eta}_{j})), for all n=1,,N1,N+1,,2N1n=1,\ldots,N-1,N+1,\ldots,2N-1 and i,j{1,,L}i,j\in\{1,\ldots,L\};

  2. (b)

    η<dist(Diη,Djη)\eta<\mathrm{dist\,}(D^{\eta}_{i},D^{\eta}_{j}), for all ij{1,,L}i\neq j\in\{1,\ldots,L\};

  3. (c)

    η<dist(ϕHt+n(v±i),ϕHt(v±j))\eta<\mathrm{dist\,}(\phi_{H}^{t+n}(v^{i}_{\pm}),\phi_{H}^{t}(v^{j}_{\pm})), for all t[0,2N],n{0}t\in[0,2N],n\in\mathbbm{Z}\setminus\{0\}, t+n[0,2N)t+n\in[0,2N), and i,j{1,,L}i,j\in\{1,\ldots,L\};

  4. (d)

    η<dist(ϕHN(v±i),Dj)\eta<\mathrm{dist\,}(\phi^{N}_{H}(v^{i}_{\pm}),D_{j}), for all i,j{1,,L}i,j\in\{1,\ldots,L\};

  5. (e)

    η<dist(ϕHt(v±i),ϕHt(v±j))\eta<\mathrm{dist\,}(\phi_{H}^{t}(v^{i}_{\pm}),\phi_{H}^{t}(v^{j}_{\pm})), for all t[0,2N]t\in[0,2N], ij{1,,L}i\neq j\in\{1,\ldots,L\};

Choose ε=ε(η,B)>0\varepsilon=\varepsilon(\eta,B)>0 according to Corollary 3.2. We assume the contrary and let uJkt(x^,x^+,L0,ψk(L1))u\in\mathcal{M}_{J_{kt}}(\hat{x}_{-},\hat{x}_{+},L_{0}^{\prime},\psi^{-k}(L^{\prime}_{1})), kk\in\mathbbm{N}, such that (1)(1) and (2)(2) hold, but E(u)<εE(u)<\varepsilon. Assume that x^+ρ=Υ(ρ)\hat{x}_{+}\in\rho^{\prime}=\Upsilon(\rho) for some ρ=[C(i1,,in2)]𝒵1+\rho=[C_{(i_{1},\ldots,i_{n-2})}]\in\mathcal{Z}^{+}_{1}, the argument for the other case is analogous. Let s0s_{0}\in\mathbbm{R} be the minimal ss\in\mathbbm{R} such that the path tu¯(s,t)t\mapsto\overline{u}(s,t) intersects \mathcal{L}. We consider the path

γ:[0,k]=[0,nN]M,γ(t)=u¯(s0,t).\gamma:[0,k]=[0,nN]\to M,\quad\gamma(t)=\overline{u}(s_{0},t).

Note that γ\gamma intersects F1,,FLF_{1},\ldots,F_{L} in at least l:=nχl:=n-\chi many points γ(t1),\gamma(t_{1}), ,\ldots, γ(tl)\gamma(t_{l}), 0t1<<tlk0\leq t_{1}<\cdots<t_{l}\leq k. Moreover, we can choose such points so that if γ(ti)\gamma(t_{i}) and γ(ti)\gamma(t_{i^{\prime}}), i<ii<i^{\prime}, lie in the same surface FjF_{j}, j=1,,Lj=1,\ldots,L, then the path γ[ti,ti]\gamma_{[t_{i},t_{i^{\prime}}]} is not homotopic relative to its endpoints to a path contained in FjF_{j}.

Claim 1: Let 1il1\leq i\leq l. Assume that γ(ti)Fj/Dj\gamma(t_{i})\in F_{j}/D_{j} for some j{1,,L}j\in\{1,\ldots,L\}. Then, there is tit^{-}_{i}\in\mathbbm{Z} with ti<ti<ti+2N=:ti+t^{-}_{i}<t_{i}<t^{-}_{i}+2N=:t^{+}_{i} such that γ(ti),γ(ti+)Djη\gamma(t^{-}_{i}),\gamma(t^{+}_{i})\in D^{\eta}_{j} (provided that ti,ti+[0,k]t^{-}_{i},t^{+}_{i}\in[0,k]). Moreover, ti[ti,ti+]t_{i^{\prime}}\notin[t^{-}_{i},t^{+}_{i}] for all iii^{\prime}\neq i.

Proof: We write γ(ti)=φr(x)\gamma(t_{i})=\varphi^{r}(x) with xv+jvjx\in v_{+}^{j}\cup v_{-}^{j}, 0r<2N0\leq r<2N, and put ti:=tirt^{-}_{i}:=t_{i}-r, ti+:=ti+2Nt^{+}_{i}:=t^{-}_{i}+2N. By Corollary 3.2, for all t[ti,ti+][0,k]t\in[t^{-}_{i},t^{+}_{i}]\cap[0,k] we have that d¯(γ(t),φtti(γ(ti)))<η\overline{d}(\gamma(t),\varphi^{t-t_{i}}(\gamma(t_{i})))<\eta, and hence

(24) d¯(γ(t),φtti(x))<η.\displaystyle\overline{d}(\gamma(t),\varphi^{t-t^{-}_{i}}(x))<\eta.

This means that γ(ti),γ(ti+)Djη\gamma(t^{-}_{i}),\gamma(t^{+}_{i})\in D^{\eta}_{j} (provided that ti,ti+[0,k]t^{-}_{i},t^{+}_{i}\in[0,k]). To show the second assertion, assume by contradiction that there is iii^{\prime}\neq i such that ti[ti,ti+]t_{i^{\prime}}\in[t^{-}_{i},t^{+}_{i}]. With (24), it follows from (a), (b), and (d) that γ(ti)Dj\gamma(t_{i^{\prime}})\notin D_{j^{\prime}} for all jjj^{\prime}\neq j, and it follows from (a), (c), (d), and (e) that γ(ti)FjDj\gamma(t_{i^{\prime}})\notin F_{j^{\prime}}\setminus D_{j^{\prime}} for all jjj^{\prime}\neq j. Hence γ(ti)Fj\gamma(t_{i^{\prime}})\in F_{j}. With (24), it follows now from (a), (c), and (d) that γ(ti)φtiti(v+jvj)\gamma(t_{i^{\prime}})\in\varphi^{t_{i^{\prime}}-t^{-}_{i}}(v^{j}_{+}\cup v^{j}_{-}). Therefore, γ|I\gamma|_{I} is homotopic relative to its endpoints to a path in FjF_{j}, where II is the interval with endpoints tit_{i} and tit_{i^{\prime}}. This is a contradiction. ∎

Let 1il1\leq i\leq l. If γ(ti)FjDj\gamma(t_{i})\in F_{j}\setminus D_{j} for some j{1,,L}j\in\{1,\ldots,L\}, we choose ti±t^{\pm}_{i} as in Claim 4.2, and if γ(ti)Dj\gamma(t_{i})\in D_{j} for some j{1,,L}j\in\{1,\ldots,L\}, we set ti=ti+:=tit^{-}_{i}=t^{+}_{i}:=t_{i}\in\mathbbm{N}.

Claim 2: Let 1i<il1\leq i<i^{\prime}\leq l. Then titi+Nt^{-}_{i^{\prime}}-t^{+}_{i}\geq N.

Proof: By the second assertion of Claim 4.2 we have ti>ti+t^{-}_{i^{\prime}}>t^{+}_{i}. For any n{1,,N1}n\in\{1,\ldots,N-1\} with n+ti+kn+t^{+}_{i}\leq k, it holds by Corollary 3.2 that d¯(γ(n+ti+),φn(γ(ti+)))<η\overline{d}(\gamma(n+t^{+}_{i}),\varphi^{n}(\gamma(t^{+}_{i})))<\eta, and hence by (a)(a), γ(n+ti+)j=1LDjη\gamma(n+t^{+}_{i})\notin\bigcup_{j=1}^{L}D^{\eta}_{j}. The claim follows by the first assertion of Claim 4.2. ∎

We write l=T++Tl=T_{+}+T_{-}, where T+T_{+} is the number of i=1,,li=1,\ldots,l with γ(ti)j=1LDj\gamma(t_{i})\in\bigcup_{j=1}^{L}D_{j}, and TT_{-} the number of i=1,,li=1,\ldots,l with γ(ti)j=1LFjDj\gamma(t_{i})\in\bigcup_{j=1}^{L}F_{j}\setminus D_{j}. By Claims 4.2 and 4.2, Nn=kNT++3NT4NNn=k\geq NT_{+}+3NT_{-}-4N. By (23), T++TnχT++3T4χT_{+}+T_{-}\geq n-\chi\geq T_{+}+3T_{-}-4-\chi, hence 2Tχ+42T_{-}\leq\chi+4. It follows that

(25) T+n3χ22n2χ2.\displaystyle T_{+}\geq n-\frac{3\chi}{2}-2\geq n-2\chi-2.

Choose now t^[0,k]\hat{t}\in[0,k] with γ(t^)\gamma(\hat{t})\in\mathcal{L}, and let I[0,k]I\subset[0,k] be any interval with t^I\hat{t}\in I and length |I|=2N(2χ+5)=B2N|I|=2N(2\chi+5)=B-2N.

Claim 3: If γ(t^)j\gamma(\hat{t})\in\mathcal{L}_{j} for some j{1,,L}j\in\{1,\ldots,L\}, then the number of i{1,,l}i\in\{1,\ldots,l\} with tiIt_{i}\in I and γ(ti)j=1LDj\gamma(t_{i})\in\bigcup_{j=1}^{L}D_{j} is smaller or equal than 2χ+52\chi+5.

Proof: Write γ(t^)=φr(x)\gamma(\hat{t})=\varphi^{r}(x) with x𝒪Djx\in\mathcal{O}\cap D_{j}, 0r<2N0\leq r<2N, j{1,,L}j\in\{1,\ldots,L\}, and write t:=t^rt_{*}:=\hat{t}-r. By Corollary 3.2,

d¯(γ(t+r+t),φt+r(x))=d¯(γ(t+t^),φt(γ(t^)))<η,\overline{d}(\gamma(t+r+t_{*}),\varphi^{t+r}(x))=\overline{d}(\gamma(t+\hat{t}),\varphi^{t}(\gamma(\hat{t})))<\eta,

for all t[B,B]t\in[-B,B] with t+t^[0,k]t+\hat{t}\in[0,k]. Note that φq(x)v+jvj\varphi^{q}(x)\subset v^{j}_{+}\cup v^{j}_{-} if and only if q2Nq\in 2N\mathbbm{Z}. Hence, by (a)(a) and (d)(d), the points γ(t+r+t)\gamma(t+r+t_{*}) for t[B,B]t\in[-B,B] can only possibly lie in i=1LDj\bigcup_{i=1}^{L}D_{j} if t+r2Nt+r\in 2N\mathbbm{Z}. The claim follows. ∎

We are now in the situation to finish the proof of the proposition. By Claim 4.2 and by (25), the number T^\widehat{T} of i{1,,l}i\in\{1,\ldots,l\} with tiIt_{i}\notin I and γ(ti)j=1LDj\gamma(t_{i})\in\bigcup_{j=1}^{L}D_{j} satisfies

T^T+(2χ+5)n4χ7.\widehat{T}\geq T_{+}-(2\chi+5)\geq n-4\chi-7.

On the other hand, again by Claim 4.2, and with [0,k]I=:I1I2[0,k]\setminus I=:I_{1}\cup I_{2},

T^|I1|/N+|I2|/N<1N(k|I|)+2n4χ8.\widehat{T}\leq\lceil|I_{1}|/N\rceil+\lceil|I_{2}|/N\rceil<\frac{1}{N}(k-|I|)+2\leq n-4\chi-8.

This yields the desired contradiction. ∎

Remark 4.3.

It follows from the proof and by the possible choices of ϵ\epsilon in Lemma 3.1 and Corollary 3.2, that ϵ\epsilon as a function on (L0,L1)(L^{\prime}_{0},L^{\prime}_{1}) in the above proposition can be chosen to be constant in a C2C^{2}-small neighbourhood of any previously given, and not necessarily non-degenerate pair (L0,L1)(L^{\prime}_{0},L^{\prime}_{1}).

4.3. Isolation in the chain complex and length of bars

To deduce Theorem 3 from the crossing energy estimates above, we will additionally need a purely algebraic statement. It generalizes Prop. 3.8 in [13]. To this end, fix 𝒞=CF(L0,ψk(L1))\mathcal{C}=\mathrm{CF}(L_{0},\psi^{-k}(L_{1})), for some kk\in\mathbbm{N}. Consider now chain complexes (𝒞,)(\mathcal{C}^{\prime},\partial^{\prime}) for which 𝒞𝒞\mathcal{C}^{\prime}\subset\mathcal{C} is a subspace generated by a subset of 𝒫~k\widetilde{\mathcal{P}}_{k}. The remaining basis vectors generate the subspace (𝒞)(\mathcal{C}^{\prime})^{\perp} with 𝒞(𝒞)=𝒞\mathcal{C}^{\prime}\oplus(\mathcal{C}^{\prime})^{\perp}=\mathcal{C}. We say that a chain complex (𝒞,)(\mathcal{C}^{\prime},\partial^{\prime}) of this form is ϵ\epsilon-isolated in (𝒞,)(\mathcal{C},\partial) if the following holds:

  1. (i)

    if y𝒞{0}y\in\mathcal{C}^{\prime}\setminus\{0\}, then 𝒜(y)𝒜(()y)>ϵ\mathcal{A}(y)-\mathcal{A}((\partial-\partial^{\prime})y)>\epsilon;

  2. (ii)

    if y(𝒞){0},ζ𝒞y\in(\mathcal{C}^{\prime})^{\perp}\setminus\{0\},\zeta\in\mathcal{C}^{\prime} with ζ+y(𝒞)\zeta+\partial y\in(\mathcal{C}^{\prime})^{\perp}, then 𝒜(y)𝒜(ζ)>ϵ\mathcal{A}(y)-\mathcal{A}(\zeta)>\epsilon.

In the case that 𝒞\mathcal{C}^{\prime} is one-dimensional, this definition coincides with the definition given in [13]. Also, the following reduces to the statement of Prop. 3.8 in [13] if all complexes 𝒞i\mathcal{C}^{\prime}_{i} are one-dimensional.

Proposition 4.4.

Assume that (𝒞1,1),,(𝒞p,p)(\mathcal{C}_{1}^{\prime},\partial^{\prime}_{1}),\ldots,(\mathcal{C}^{\prime}_{p},\partial^{\prime}_{p}) are chain complexes with non-zero homology that are ϵ\epsilon-isolated in (𝒞,)(\mathcal{C},\partial), and for which 𝒞i𝒞j={0}\mathcal{C}^{\prime}_{i}\cap\mathcal{C}^{\prime}_{j}=\{0\} if iji\neq j. Then bϵ(𝒞)p/2b_{\epsilon}(\mathcal{C})\geq p/2.

Proof.

We give here a proof which closely follows that of Prop. 3.8 in [13]. We can assume for simplicity that the actions of the generators of 𝒞\mathcal{C} are all different, see the discussion in [13]. For i=1,,pi=1,\ldots,p, let ξi𝒞i\xi_{i}\in\mathcal{C}_{i}^{\prime} be a vector that represents a non-zero class in the homology of 𝒞i\mathcal{C}_{i}^{\prime} and such that the action 𝒜(ξi)\mathcal{A}(\xi_{i}) is minimal among all such vectors in 𝒞i\mathcal{C}^{\prime}_{i}. All ξi\xi_{i}, i=1,,pi=1,\ldots,p, are quasi ϵ\epsilon-robust: for any vectors y,τ𝒞y,\tau\in\mathcal{C} with 𝒜(τ)<𝒜(ξi)\mathcal{A}(\tau)<\mathcal{A}(\xi_{i}) and y=ξi+τ\partial y=\xi_{i}+\tau, it holds that 𝒜(y)𝒜(ξi)>ϵ\mathcal{A}(y)-\mathcal{A}(\xi_{i})>\epsilon. Indeed, assume the contrary, and let y,τy,\tau be as above such that 𝒜(y)𝒜(ξi)ϵ\mathcal{A}(y)-\mathcal{A}(\xi_{i})\leq\epsilon. By the ϵ\epsilon-isolation property (ii), y=u+vy=u+v for a non-zero vector u𝒞iu\in\mathcal{C}^{\prime}_{i} and some v(𝒞i)v\in(\mathcal{C}^{\prime}_{i})^{\perp}. Also by (ii), 𝒜(v)<𝒜(ξi)\mathcal{A}(\partial v)<\mathcal{A}(\xi_{i}), and hence by (i), 𝒜(iuξi)<𝒜(ξi)\mathcal{A}(\partial_{i}^{\prime}u-\xi_{i})<\mathcal{A}(\xi_{i}). By the minimality property of 𝒜(ξi)\mathcal{A}(\xi_{i}), we have that iuξi=0\partial_{i}^{\prime}u-\xi_{i}=0, contradicting ξiimi\xi_{i}\notin\mathrm{im}\partial_{i}^{\prime}. Note that also any ζ𝒞\zeta\in\mathcal{C} with 𝒜(ζ)=𝒜(ξi)\mathcal{A}(\zeta)=\mathcal{A}(\xi_{i}) is quasi ϵ\epsilon-robust, as well as any non-trivial linear combination of quasi ϵ\epsilon-robust vectors.

Let VV be the span of the vectors ξi\xi_{i}, i=1,,pi=1,\ldots,p, and write V=ker(|V)YV=\ker(\partial|_{V})\oplus Y, for some YVY\subset V. If now ζY\zeta\in Y, then by the ϵ\epsilon-isolation property (i) and since ξikeri\xi_{i}\in\ker\partial^{\prime}_{i}, i=1,,pi=1,\ldots,p, it holds that 𝒜(ζ)𝒜(ζ)>ϵ\mathcal{A}(\zeta)-\mathcal{A}(\partial\zeta)>\epsilon. Hence, if ζ\partial\zeta is not ϵ\epsilon-robust, then there is a vector uu with 𝒜(u)<𝒜(ζ)\mathcal{A}(u)<\mathcal{A}(\zeta) and w:=ζ+ukerw:=\zeta+u\in\ker\partial. Note that ww is quasi ϵ\epsilon-robust. One can now find by induction quasi ϵ\epsilon-robust vectors w1,,wrw_{1},\ldots,w_{r} and a subspace YYY^{\prime}\subset Y such that dimY+r+dim(ker(|V))=p\dim Y^{\prime}+r+\dim(\ker(\partial|_{V}))=p and such that (Y)\partial(Y^{\prime}) is ϵ\epsilon-robust. Since a subspace in im\mathrm{im}\partial that is generated by quasi ϵ\epsilon-robust vectors is ϵ\epsilon-robust, it follows as in [13] that bϵ(𝒞)max{dimY,r+dim(ker(|V))}p/2b_{\epsilon}(\mathcal{C})\geq\max\{\dim Y^{\prime},r+\dim(\ker(\partial|_{V}))\}\geq p/2. ∎

Proof of Theorem 3.

By Lemma 4.1 and the discussion in §4.2, there exist a compact hyperbolic set KK with isolating neighbourhood U=i=1LDiU=\bigcup_{i=1}^{L}{D}_{i}, a union of periodic orbits 𝒪M\mathcal{O}\subset M that contain the corners of DiD_{i}, and a choice of segments l0D0l_{0}\subset D^{0}, l1ψN(D1)l_{1}\subset\psi^{N}(D^{1}) for some connected components D0D^{0} and D1D^{1} of U{U} such that for any admissible pair (L0,L1)(L^{\prime}_{0},L^{\prime}_{1}) of closed curves in M𝒪M\setminus\mathcal{O} that intersect D0D^{0} and ψN(D1)\psi^{N}(D^{1}) in l0l_{0} and l1l_{1}, respectively, we have

lim supklog#𝒵1kk=htop(ψ|K)htop(ψ)e,\limsup_{k\to\infty}\frac{\log\#\mathcal{Z}_{1}^{k}}{k}=h_{\mathrm{top}}(\psi|_{K})\geq h_{\mathrm{top}}(\psi)-e,

where 𝒵1k\mathcal{Z}^{k}_{1} is the set of classes in 𝒵1+(l0,l1)\mathcal{Z}_{1}^{+}(l_{0},l_{1}) that are represented by chords of length kk. Let us call curves L0L^{\prime}_{0} and L1L^{\prime}_{1} adapted if they have the intersection property above. Choose homotopy classes α0\alpha_{0} and α1\alpha_{1} of closed curves in M𝒪M\setminus\mathcal{O} that have representing curves Γ0\Gamma_{0} and Γ1\Gamma_{1} that are adapted, and such that {0}×ΓiS1×M\{0\}\times\Gamma_{i}\in S^{1}\times M, i=1,2i=1,2, have zero algebraic intersection number with any of the surfaces F1,,FLF_{1},\ldots,F_{L} considered in §4.2. Let (L0,L1)(L_{0},L_{1}) be any admissible pair of curves that lie in M𝒪M\setminus\mathcal{O} such that [Li]=[αi][L_{i}]=[\alpha_{i}], i=0,1i=0,1. We can find two Hamiltonian isotopies supported in M𝒪M\setminus\mathcal{O} from LiL_{i} to adapted curves LiL^{\prime}_{i}, i=0,1i=0,1. One way to see this is as follows. Write the boundary of D0D^{0} as D0=v+vh+h\partial D^{0}=v_{+}\cup v_{-}\cup h_{+}\cup h_{-} as in §4.2. Note that the endpoints of v±v_{\pm} and h±h_{\pm} lie in 𝒪\mathcal{O}, and denote by v̊±\mathring{v}_{\pm}, h̊±\mathring{h}_{\pm} the open segments in the complement of 𝒪\mathcal{O}. We can find an isotopy in M𝒪M\setminus\mathcal{O} from L0L_{0} to a curve L0′′L_{0}^{\prime\prime} that has minimal intersection (in its homotopy class) with v̊±\mathring{v}_{\pm}, h̊±\mathring{h}_{\pm} (that is, it has one intersection with vv_{-} and with v+v_{+}, and none with h±h_{\pm}), by successively cancelling bigons whose first boundary component is contained in the curve and whose second is contained in v̊+\mathring{v}_{+}, v̊\mathring{v}_{-}, h̊+\mathring{h}_{+}, or h̊\mathring{h}_{-}. A cancellation of a bigon is obtained by “moving” the boundary segment of the bigon contained in the curve over the respective arc v̊+\mathring{v}_{+}, v̊\mathring{v}_{-}, h̊+\mathring{h}_{+}, or h̊\mathring{h}_{-}. By composing afterwards with an isotopy fixing the curve outside a neighbourhood of D0𝒪M𝒪D^{0}\setminus\mathcal{O}\subset M\setminus\mathcal{O} we can obtain an isotopy Φ:S1×[0,1]M𝒪\Phi:S^{1}\times[0,1]\to M\setminus\mathcal{O} from Φ(,0)=L0\Phi(\cdot,0)=L_{0} to an adapted curve Φ(,1)=L0\Phi(\cdot,1)=L_{0}^{\prime}. To be able to extend the isotopy Φ\Phi to a Hamiltonian isotopy, it needs to be exact, that is, (αs)s[0,1](\alpha_{s})_{s\in[0,1]}, αs:=ι|s(Φω)(s,t)\alpha_{s}:=\iota|_{\frac{\partial}{\partial_{s}}}(\Phi^{*}\omega)_{(s,t)}, needs to be a family of exact 11-forms on S1S^{1}. One can achieve this by modifying, during each step described above, the isotopy in a small disk in M(𝒪D0)M\setminus(\mathcal{O}\cup D^{0}) that intersects the curve but not the segment of it contained in the bigon to be cancelled in that step. (We might have to divide each step above into several smaller steps, and each time choose a different small disk). After these modifications, the second endpoint L0L_{0}^{\prime} of that isotopy will still be an adapted curve. The Hamiltonian isotopy from L1L_{1} to L1L^{\prime}_{1} is obtained similarly.

Since any ρ𝒵1k\rho\in\mathcal{Z}_{1}^{k} is represented by a unique chord, HFρ(L0,ψk(L1))0\mathrm{HF}^{\rho}(L^{\prime}_{0},\psi^{-k}(L^{\prime}_{1}))\neq 0. Assume first that (L0,L1)(L_{0},L_{1}) is additionally non-degenerate. Then, by (19) and (20), we have HFΥ(ρ)(L0,ψk(L1))0\mathrm{HF}^{\Upsilon(\rho)}(L_{0},\psi^{-k}(L_{1}))\neq 0. Here Υ=ΥL0,L0L1,L1\Upsilon=\Upsilon^{L^{\prime}_{1},L_{1}}_{L^{\prime}_{0},L_{0}}. By Proposition 4.2, there are almost complex structures and ϵ>0\epsilon>0 such that the chain complexes CFΥ(ρ)(L0,ψk(L1))\mathrm{CF}^{\Upsilon(\rho)}(L_{0},\psi^{-k}(L_{1})), ρ𝒵1+(l0,l1)\rho\in\mathcal{Z}^{+}_{1}(l_{0},l_{1}), are ϵ\epsilon-isolated in CF(L0,ψk(L1))\mathrm{CF}(L_{0},\psi^{-k}(L_{1})). It follows then by Proposition 4.4 that

(ψ;L0,L1)lim supllog#𝒵1kkhtop(ψ)e.\hbar(\psi;L_{0},L_{1})\geq\limsup_{l\to\infty}\frac{\log\#\mathcal{Z}_{1}^{k}}{k}\geq h_{\mathrm{top}}(\psi)-e.

By Remark 4.3, the inequality generalizes to degenerate pairs (L0,L1)(L_{0},L_{1}) that satisfy the assumptions of the theorem. ∎

5. Examples for which the strong barcode entropy is positive

In this last section we prove Theorem 6, which gives a class of examples for which the strong barcode entropy is positive. Let (Σ,σ)(\Sigma,\sigma) be a closed symplectic surface of genus at least 22, and let I1,I2:ΣΣI_{1},I_{2}:\Sigma\to\Sigma be two commuting anti-symplectic involutions such that ΣFix(I1)=ΣΣ′′\Sigma\setminus\mathrm{Fix}(I_{1})=\Sigma^{\prime}\cup\Sigma^{\prime\prime} has two components Σ\Sigma^{\prime} and Σ′′\Sigma^{\prime\prime} with positive genus, and such that the fixed point set Fix(f)\mathrm{Fix}(f) of f=I2I1f=I_{2}\circ I_{1} is non-empty and finite. Some examples can be already obtained by realizing Σ\Sigma as suitably embedded in 3\mathbbm{R}^{3} and the involutions as the restrictions of reflexions on coordinate planes. The problem of classification of triples (Σ,I1,I2)(\Sigma,I_{1},I_{2}) as above can be reduced to the corresponding problem about anti-holomorphic involutions on Riemann surfaces. Those involutions are well understood, see [30, §21]; for a description of all commuting pairs see [38].

In general, the fixed point set of anti-symplectic involutions on surfaces consists of a union of isolated points or of closed isolated circles. With our assumptions, as it is easy to see, Fix(I1)\mathrm{Fix}(I_{1}) and Fix(I2)\mathrm{Fix}(I_{2}) must both consist of a union of closed circles. Moreover, I1I_{1} flips Σ\Sigma^{\prime} and Σ′′\Sigma^{\prime\prime}, whereas I2I_{2} preserves each of them. It follows that Fix(f)=Fix(I1)Fix(I2)\mathrm{Fix}(f)=\mathrm{Fix}(I_{1})\cap\mathrm{Fix}(I_{2}). We fix some x^Fix(f)\hat{x}\in\mathrm{Fix}(f), and denote by SS the connected component in Fix(I1)\mathrm{Fix}(I_{1}) that contains x^\hat{x}. We also write Σ¯=ΣFix(I1)\overline{\Sigma}^{\prime}=\Sigma^{\prime}\cup\mathrm{Fix}(I_{1}) and Σ¯′′=Σ′′Fix(I1)\overline{\Sigma}^{\prime\prime}=\Sigma^{\prime\prime}\cup\mathrm{Fix}(I_{1}) for the surfaces that one obtains by adding the boundary Fix(I1)\mathrm{Fix}(I_{1}).

Let M=Σ×ΣM=\Sigma\times\Sigma, equipped with the symplectic form ω=σσ\omega=\sigma\oplus-\sigma, and let LML\subset M be the Lagrangian submanifold defined by L={(x,f(x))|xΣ}L=\{(x,f(x))\,|\,x\in\Sigma\}. The surface Σ\Sigma also embeds as a diagonal into MM, ϕ:ΣM,x(x,x)\phi:\Sigma\to M,x\mapsto(x,x), and we denote this Lagrangian submanifold by Δ=im(ϕ)\Delta=\mathrm{im}(\phi). By the Lagrangian neighbourhood theorem, see e.g. [33], there exist a neighbourhood 𝒩=𝒩(Σ0)\mathcal{N}=\mathcal{N}(\Sigma_{0}) of the zero section Σ0\Sigma_{0} in TΣT^{*}\Sigma, a neighbourhood VV of the diagonal ΔM\Delta\subset M, and a diffeomorphism Φ:𝒩V\Phi:\mathcal{N}\to V such that Φω=dλ\Phi^{*}\omega=-d\lambda, Φ(x)=ϕ(x)\Phi(x)=\phi(x) if xΣ0Σx\in\Sigma_{0}\cong\Sigma. Here λ\lambda is the canonical Liouville 11-form on TΣT^{*}\Sigma. We may assume, by choosing the neighbourhoods sufficiently small and by composition with a Hamiltonian isotopy fixing Δ\Delta, that every connected component of LVL\cap V is a disk, identified via Φ\Phi with the fibre of TΣT^{*}\Sigma in 𝒩\mathcal{N} over a fixed point of ff.666For 𝒩\mathcal{N} small, by the Lagrangian neighbourhood theorem applied to a fibre TxΣ𝒩T^{*}_{x}\Sigma\cap\mathcal{N} over some xFix(x)x\in\mathrm{Fix}(x), the Lagrangian disk Φ1(LV)\Phi^{-1}(L\cap V) through (x,x)(x,x) becomes a graph of a 11-form over TxΣ𝒩T^{*}_{x}\Sigma\cap\mathcal{N}. Then we can use local primitives to construct that isotopy. Let DD be the connected component of LVL\cap V that contains (x^,x^)(\hat{x},\hat{x}).

Before we give the definition of the Hamiltonian diffeomorphism ψ\psi on MM for the proof of Theorem 6, let us make some further topological observations. The diagonal embedding ϕ:ΣM\phi:\Sigma\hookrightarrow M, x(x,x)x\mapsto(x,x), induces an injective homomorphism π1(Σ,x^)\pi_{1}(\Sigma,\hat{x})\to π1(M,(x^,x^))\pi_{1}(M,(\hat{x},\hat{x})), and via the composition with the map π1(M,(x^,x^))π1(M;L,L)\pi_{1}(M,(\hat{x},\hat{x}))\to\pi_{1}(M;L,L) induced by inclusion, any 𝔞π1(Σ,x^)\mathfrak{a}\in\pi_{1}(\Sigma,\hat{x}) gives rise to a well-defined homotopy class 𝔠𝔞π1(M;L,L)\mathfrak{c}_{\mathfrak{a}}\in\pi_{1}(M;L,L) of paths in MM with endpoints in LL. Let 𝔄π1(Σ,x^)\mathfrak{A}\subset\pi_{1}(\Sigma,\hat{x}) be the subset of classes 𝔞\mathfrak{a} that have a representative curve in Σ¯\overline{\Sigma}^{\prime} and that cannot be represented by a multiple of a curve parametrizing SS.

Lemma 5.1.

Let x0,x1Fix(f)x_{0},x_{1}\in\mathrm{Fix}(f), and τ\tau any path in Σ\Sigma from x0x_{0} to x1x_{1}. Denote by 𝔠0\mathfrak{c}_{0} the homotopy class of paths in MM with boundary in LL defined by τ^=ϕτ\hat{\tau}=\phi\circ\tau. Then, for all 𝔞𝔄\mathfrak{a}\in\mathfrak{A} it holds that 𝔠0𝔠𝔞\mathfrak{c}_{0}\neq\mathfrak{c}_{\mathfrak{a}}, unless x0=x1=x^x_{0}=x_{1}=\hat{x}. Moreover, if x0=x1=x^x_{0}=x_{1}=\hat{x} and 𝔠0=𝔠𝔞\mathfrak{c}_{0}=\mathfrak{c}_{\mathfrak{a}} for some 𝔞𝔄\mathfrak{a}\in\mathfrak{A}, then any homotopy of paths in 𝔠0\mathfrak{c}_{0} from τ^\hat{\tau} to itself is homotopic among such homotopies to the constant homotopy.

Proof.

Let 𝔞𝔄\mathfrak{a}\in\mathfrak{A} and choose a representative in 𝔠𝔞\mathfrak{c}_{\mathfrak{a}} of the form γ^=ϕγ\hat{\gamma}=\phi\circ\gamma, where γ\gamma is a loop in Σ\Sigma with basepoint x^\hat{x}. Assume that there is a homotopy h:[0,1]2Mh:[0,1]^{2}\to M in MM relative LL between h(0,t)=τ^(t)h(0,t)=\hat{\tau}(t) and h(1,t)=γ^(t)h(1,t)=\hat{\gamma}(t). We obtain paths β^i:[0,1]L\hat{\beta}_{i}:[0,1]\to L, i=0,1i=0,1, from (xi,xi)(x_{i},x_{i}) to (x^,x^)(\hat{x},\hat{x}), defined by β^i(s):=h(s,i)\hat{\beta}_{i}(s):=h(s,i). We can write β^i(s)=(βi(s),f(βi(s)))M=Σ×Σ\hat{\beta}_{i}(s)=(\beta_{i}(s),f(\beta_{i}(s)))\in M=\Sigma\times\Sigma, s[0,1]s\in[0,1], i=0,1i=0,1. The concatenation of paths β^0#γ^#β^1¯\hat{\beta}_{0}\#\hat{\gamma}\#\overline{\hat{\beta}_{1}} is homotopic relative its endpoints to τ^\hat{\tau}. And projecting such a homotopy to the first as well as second component in M=Σ×ΣM=\Sigma\times\Sigma shows that τ\tau is homotopic in Σ\Sigma relative its endpoints both to the concatenation of paths β0#γ#β1¯\beta_{0}\#\gamma\#\overline{\beta_{1}} and to (fβ0)#γ#(fβ1¯)(f\circ\beta_{0})\#\gamma\#(\overline{f\circ\beta_{1}}). Hence (fβ0¯)#β0#γ#β1¯#(fβ1)(\overline{f\circ\beta_{0}})\#\beta_{0}\#\gamma\#\overline{\beta_{1}}\#(f\circ\beta_{1}) is homotopic to γ\gamma relative x^\hat{x}.

We claim that the loops θ0=(fβ0¯)#β0\theta_{0}=(\overline{f\circ\beta_{0}})\#\beta_{0} and θ1=β1¯#(fβ1)\theta_{1}=\overline{\beta_{1}}\#(f\circ\beta_{1}) must be contractible relative x^\hat{x}. Indeed, first, at least one is not homotopic to a non-trivial multiple of SS relative x^\hat{x}, since γ\gamma is not homotopic to such a multiple and π1(Σ¯,x^)\pi_{1}(\overline{\Sigma}^{\prime},\hat{x}) is free. Say that θ0\theta_{0} is such a loop, the other case is analogous. It follows that the image of [θ0]π1(Σ,x^)[\theta_{0}]\in\pi_{1}(\Sigma,\hat{x}) under the naturally induced map p:π1(Σ,x^)π1(Σ^Σ^′′,x^)=π1(Σ^,x^)π1(Σ^′′,x^)p:\pi_{1}(\Sigma,\hat{x})\to\pi_{1}(\hat{\Sigma}^{\prime}\cup\hat{\Sigma}^{\prime\prime},\hat{x})=\pi_{1}(\hat{\Sigma}^{\prime},\hat{x})\star\pi_{1}(\hat{\Sigma}^{\prime\prime},\hat{x}) is non-trivial, where Σ^\hat{\Sigma}^{\prime} and Σ^′′\hat{\Sigma}^{\prime\prime} are the spaces given by collapsing the boundary components of Σ¯\overline{\Sigma}^{\prime} resp. Σ¯′′\overline{\Sigma}^{\prime\prime} to a single point. Since ff flips Σ\Sigma^{\prime} and Σ′′\Sigma^{\prime\prime}, the element p([θ0])p([\theta_{0}]) can be written in reduced terms (with respect to some generators) as a word that has both a non-zero term in π1(Σ^,x^)\pi_{1}(\hat{\Sigma}^{\prime},\hat{x}) and one in π1(Σ^′′,x^)\pi_{1}(\hat{\Sigma}^{\prime\prime},\hat{x}). Hence p([θ0][γ][θ1])p([\theta_{0}]\cdot[\gamma]\cdot[\theta_{1}]) cannot be represented as a curve in Σ^\hat{\Sigma}^{\prime} and therefore [θ0][γ][θ1]=[γ]𝔄[\theta_{0}]\cdot[\gamma]\cdot[\theta_{1}]=[\gamma]\in\mathfrak{A} cannot be represented as a curve in Σ¯\overline{\Sigma}^{\prime}, a contradiction.

To show the first assertion of the lemma, we assume additionally, by contradiction, that xix^x_{i}\neq\hat{x} for some i{0,1}i\in\{0,1\}. That means that the contractible loop θ:=θi\theta:=\theta_{i} is a concatenation of an arc β\beta from x^\hat{x} to xix_{i} and the arc fβ¯\overline{f\circ\beta} from xix_{i} to x^\hat{x}. Hence β\beta is homotopic to fβf\circ\beta relative their endpoints. We can lift ff to a homeomorphism f~\widetilde{f} on the universal covering Σ~\widetilde{\Sigma} such that f~\widetilde{f} fixes a lift x~\widetilde{{x}} of x^\hat{x}. Note that f~2{\widetilde{f}}^{2} is a deck transformation with fixed point and hence necessarily equal to the identity. Let β~\widetilde{\beta} be the lift of β\beta starting at x~\widetilde{{x}}. By the above, also the second endpoint of β~\widetilde{\beta} is fixed by f~\widetilde{f}. Therefore f~\widetilde{f}, as any orientation preserving involution of the plane with at least two fixed points, must be the identity, which can be shown for example by elementary arguments involving the f~\widetilde{f}-invariant set β~f~(β~)\widetilde{\beta}\cup\widetilde{f}(\widetilde{\beta}). Hence ff is the identity, which contradicts our assumptions.

To conclude the second assertion of the lemma, let x0=x1=x^x_{0}=x_{1}=\hat{x} and assume that 𝔠0=𝔠𝔞\mathfrak{c}_{0}=\mathfrak{c}_{\mathfrak{a}} for some 𝔞𝔄\mathfrak{a}\in\mathfrak{A}. By an argument as in the previous paragraph, one sees that β0\beta_{0} and β1\beta_{1} are contractible loops, and hence β^0\hat{\beta}_{0} and β^1\hat{\beta}_{1} are contractible loops. Hence we can deform the homotopy hh among such homotopies to a homotopy h:[0,1]2Mh^{\prime}:[0,1]^{2}\to M such that h(s,0)=h(s,1)=x^h^{\prime}(s,0)=h^{\prime}(s,1)=\hat{x} for all s[0,1]s\in[0,1]. That hh^{\prime} is homotopic fixing the boundary h|[0,1]2h^{\prime}|_{\partial[0,1]^{2}} to h′′:[0,1]2Mh^{\prime\prime}:[0,1]^{2}\to M, h′′(s,t):=τ^(t)h^{\prime\prime}(s,t):=\hat{\tau}(t), s,t[0,1]s,t\in[0,1], follows from the homotopy exact sequence for the fibration ΩMΛMM\Omega M\hookrightarrow\Lambda M\to M of the free loop space ΛM\Lambda M with based loop space ΩM\Omega M as fibre, since MM is aspherical (π2(M)=0)(\pi_{2}(M)=0) and atoroidal (π1(ΛM)=0)(\pi_{1}(\Lambda M)=0). ∎

Proof of Theorem 6.

We keep the notation from above. Fix a hyperbolic metric gg on Σ\Sigma. We may choose the neighbourhoods 𝒩\mathcal{N} and VV constructed above in such a way that for some 0<δ<1/30<\delta<1/3 we have that 𝒩={vTΣ||v|g3δ}\mathcal{N}=\{v\in T^{*}\Sigma\,|\,|v|_{g}\leq 3\delta\}. Here |v|g|v|_{g} denotes the norm of vv in the cotangent fibre with respect to the to gg dual metric. In coordinates (r,x)𝒩Σ0(r,x)\in\mathcal{N}\setminus\Sigma_{0}, where r=|v|g(0,3δ),x=v/|v|gSgΣr=|v|_{g}\in(0,3\delta),x=v/|v|_{g}\in S^{*}_{g}\Sigma, the Liouville form λ\lambda can be written as λ=rα\lambda=r\alpha, with α=λ|SgΣ\alpha=\lambda|_{S^{*}_{g}\Sigma}. Here SgΣTMS^{*}_{g}\Sigma\subset T^{*}M denotes the unit co-sphere bundle with respect to gg.

Let C>2/δC>2/\delta, and choose a smooth function H0H_{0} on 𝒩\mathcal{N} that depends only on the rr-coordinate in 𝒩Σ0\mathcal{N}\setminus\Sigma_{0}, and, for some 0<ϵ<δ0<\epsilon<\delta satisfies the following:

  • H0(r,x)=H0(r)=CϵH_{0}(r,x)=H_{0}(r)=-C-\epsilon, for 0rδϵ0\leq r\leq\delta-\epsilon;

  • H0(r)H_{0}(r) is strictly convex, for δϵ<r<δ\delta-\epsilon<r<\delta;

  • H0(r)=Cδr2CH_{0}(r)=\frac{C}{\delta}r-2C, for δr2δ\delta\leq r\leq 2\delta;

  • H0(r)H_{0}(r) is strictly concave, for 2δ<r<2δ+ϵ2\delta<r<2\delta+\epsilon;

  • H0(r)=ϵH_{0}(r)=\epsilon, for r2δ+ϵr\geq 2\delta+\epsilon.

To simplify notation, we identify 𝒩\mathcal{N} with VMV\subset M and write for an element vVv\in V instead of v=Φ(r,x)v=\Phi(r,x) simply v=(r,x)v=(r,x). This defines H0:VH_{0}:V\to\mathbbm{R}, which moreover extends to a smooth function H:MH:M\to\mathbbm{R} as

H(z)={H0(z), if zV,ϵ, otherwise. H(z)=\begin{cases}H_{0}(z),&\text{ if }z\in V,\\ \epsilon,&\text{ otherwise. }\end{cases}

Every class 𝔞π1(Σ,x^)\mathfrak{a}\in\pi_{1}(\Sigma,\hat{x}) contains a unique geodesic arc γ𝔞:[0,T𝔞]Σ\gamma_{\mathfrak{a}}:[0,T_{\mathfrak{a}}]\to\Sigma, T𝔞0T_{\mathfrak{a}}\geq 0, from x^\hat{x} to itself, parametrized by arc length. Let η𝔞:[0,T𝔞]SgΣ\eta_{\mathfrak{a}}:[0,T_{\mathfrak{a}}]\to S^{*}_{g}\Sigma be its natural lift to the unit co-sphere bundle. We denote by ψ=ϕH1\psi=\phi_{H}^{1} the Hamiltonian diffeomorphism induced by HH. Let 𝔞π1(Σ,x^)\mathfrak{a}\in\pi_{1}(\Sigma,\hat{x}) be non-trivial. If nn is such that nC/δ>T𝔞nC/\delta>T_{\mathfrak{a}}, we obtain two distinguished intersection points y0y_{0} and y1y_{1} in Lψn(L)L\cap\psi^{-n}(L), given by yi=(ri,η𝔞(0))Vy_{i}=(r_{i},\eta_{\mathfrak{a}}(0))\in V, i=0,1i=0,1, where r0<r1r_{0}<r_{1} are such that rH(ri,x)=T𝔞/n\partial_{r}H(r_{i},x)=T_{\mathfrak{a}}/n. The Hamiltonian chords y¯i:[0,1]M\overline{y}_{i}:[0,1]\to M, i=0,1i=0,1, given by y¯i(t):=φHnt(yi)\overline{y}_{i}(t):=\varphi^{nt}_{H}(y_{i}), are, as chords in MM relative to LL, both of class 𝔠𝔞\mathfrak{c}_{\mathfrak{a}}. If 𝔞𝔄\mathfrak{a}\in\mathfrak{A}, then, by the first assertion of Lemma 5.1, there exists no other Hamiltonian chord in 𝔞\mathfrak{a}. Moreover, by the second assertion of Lemma 5.1, any homotopy between y¯0\overline{y}_{0} and y¯1\overline{y}_{1} relative to LL is homotopic to one that is contained in VV relative to DD.

There is a bijection of (a continuous family of) strips between y0y_{0} and y1y_{1} with boundaries in LL and ψn(L)\psi^{-n}(L), and (a continuous family of) homotopies relative to LL between y¯0\overline{y}_{0} and y¯1\overline{y}_{1}. It follows that the action difference 𝒜(y1,v1)𝒜(y0,v0)\mathcal{A}(y_{1},v_{1})-\mathcal{A}(y_{0},v_{0}) (with respect to the pair (L,ψn(L))(L,\psi^{-n}(L))) is independent of the cappings v0v_{0} and v1v_{1}, and is identical to vω-\int v^{*}\omega, where v:[0,1]2Vv:[0,1]^{2}\to V is a strip in VV with v(i,t)yiv(i,t)\equiv y_{i}, i=0,1i=0,1, and v(s,0)Dv(s,0)\in D, v(s,1)ψn(D)v(s,1)\in\psi^{-n}(D) for all s[0,1]s\in[0,1]. To calculate that action difference, consider the path z:[0,1]VMz:[0,1]\to V\subset M,

z(s):=(r(s),η𝔞(T𝔞))=(r1+s(r2r1),η𝔞(T𝔞)),z(s):=(r(s),\eta_{\mathfrak{a}}(T_{\mathfrak{a}}))=(r_{1}+s(r_{2}-r_{1}),\eta_{\mathfrak{a}}(T_{\mathfrak{a}})),

from ψn(y0)\psi^{n}(y_{0}) to ψn(y1)\psi^{n}(y_{1}) in DD. Consider a strip v:[0,1]2Mv:[0,1]^{2}\to M as above with v(s,0)=z(s)v(s,0)=z(s) and v(s,1)=ψn(z(s))v(s,1)=\psi^{-n}(z(s)) for all s[0,1]s\in[0,1]. Then,

𝒜(y1,v1)𝒜(y0,v0)\displaystyle\mathcal{A}(y_{1},v_{1})-\mathcal{A}(y_{0},v_{0}) =[0,1]2vω\displaystyle=-\int_{[0,1]^{2}}v^{*}\omega
=[zλ(ψnz)λ]\displaystyle=-\left[\int z^{*}\lambda-\int(\psi^{-n}\circ z)^{*}\lambda\right]
=(ψnz)λ\displaystyle=\int(\psi^{-n}\circ z)^{*}\lambda
=01r(s)α(r(s)Dψn(z(s))r)𝑑s\displaystyle=\int_{0}^{1}r(s)\alpha(r^{\prime}(s)D\psi^{-n}(z(s))\partial_{r})ds
=nr0r1rr2H(r,x)dr\displaystyle=-n\int_{r_{0}}^{r_{1}}r\partial^{2}_{r}H(r,x)dr
=n[rrH(r,x)]r0r1+nr0r1rH(r,x)dr\displaystyle=-n[r\partial_{r}H(r,x)]_{r_{0}}^{r_{1}}+n\int_{r_{0}}^{r_{1}}\partial_{r}H(r,x)dr
=(r1r0)T𝔞+n(H(r1,x)H(r0,x))\displaystyle=-(r_{1}-r_{0})T_{\mathfrak{a}}+n(H(r_{1},x)-H(r_{0},x))
>nCT𝔞.\displaystyle>nC-T_{\mathfrak{a}}.

For the fourth equation note that α(tϕHt(x,r))\alpha(\frac{\partial}{\partial_{t}}\phi^{t}_{H}(x,r)) is identical to rH(x,r)\partial_{r}H(x,r) for all tt\in\mathbbm{R}. Note that all intersections considered above are transverse and that (L,L)(L,L) can be perturbed to a non-degenerate pair, keeping those intersections unchanged. It follows that any non-trivial 𝔞𝔄\mathfrak{a}\in\mathfrak{A} with T𝔞T{T}_{\mathfrak{a}}\leq T gives rise to a finite bar of length greater than nCTnC-T, that is,

bnCT(ψ;L,L)#{𝔞𝔄|T𝔞T}=:N(T).b_{nC-T}(\psi;L,L)\geq\#\{\mathfrak{a}\in\mathfrak{A}\,|\,T_{\mathfrak{a}}\leq T\}=:N(T).

Since Σ\Sigma^{\prime} has positive genus and negative Euler characteristic, the number of elements in 𝔄π1(Σ,x^)\mathfrak{A}\subset\pi_{1}(\Sigma,\hat{x}) grows exponentially, and hence

N(T)eμTN(T)\geq e^{\mu T}

for some μ>0\mu>0. Altogether, with R:=C2R:=\frac{C}{2},

bRn(ψ;L,L)N(Rn)eμRn,b_{Rn}(\psi;L,L)\geq N(Rn)\geq e^{\mu Rn},

and therefore 𝖧R(ψ;)μR=:E.\mathsf{H}^{R}(\psi;\mathcal{L})\geq\mu R=:E.

References

  • [1] Alberto Abbondandolo, Marcelo R.R. Alves, Murat Sağlam, and Felix Schlenk. Entropy collapse versus entropy rigidity for Reeb and Finsler flows. Selecta Math. (N.S.), 29(5):Paper No. 67, 2023.
  • [2] Marcelo R.R. Alves. Growth rate of Legendrian contact homology and dynamics of Reeb flows. 2014. Thesis (Ph.D.)–Université Libre de Bruxelles.
  • [3] Marcelo R.R. Alves. Cylindrical contact homology and topological entropy. Geom. Topol., 20(6):3519–3569, 2016.
  • [4] Marcelo R.R. Alves. Positive topological entropy for Reeb flows on 3-dimensional Anosov contact manifolds. J. Mod. Dyn., 10:497–509, 2016.
  • [5] Marcelo R.R. Alves. Legendrian contact homology and topological entropy. J. Topol. Anal., 11(1):53–108, 2019.
  • [6] Marcelo R.R. Alves, Vincent Colin, and Ko Honda. Topological entropy for Reeb vector fields in dimension three via open book decompositions. J. Éc. polytech. Math., 6:119–148, 2019.
  • [7] Marcelo R.R. Alves and Matthias Meiwes. Dynamically exotic contact spheres of dimensions 7\geq 7. Comment. Math. Helv., 94(3):569–622, 2019.
  • [8] Marcelo R.R. Alves and Matthias Meiwes. Braid stability and the Hofer metric. Ann. Henri Lebesgue, page arXiv:2112.11351 - accepted for publication, 2021.
  • [9] Marcelo R.R. Alves and Abror Pirnapasov. Reeb orbits that force topological entropy. Ergodic Theory Dynam. Systems, 42(10):3025–3068, 2022.
  • [10] Paul Biran and Octav Cornea. Lagrangian cobordism. I. J. Amer. Math. Soc., 26(2):295–340, 2013.
  • [11] Rufus Bowen. Periodic points and measures for Axiom AA diffeomorphisms. Trans. Amer. Math. Soc., 154:377–397, 1971.
  • [12] Erman Çineli. A generalized pseudo-rotation with positive topological entropy. Preprint, arXiv:2310.14761, 2023.
  • [13] Erman Çineli, Viktor L. Ginzburg, and Başak Z. Gürel. Topological entropy of Hamiltonian diffeomorphisms: a persistence homology and Floer theory perspective. Preprint, arXiv:2111.03983, 2021.
  • [14] Erman Çineli, Viktor L. Ginzburg, and Başak Z. Gürel. Lower semi-continuity of Lagrangian volume. Preprint, arXiv:2210.04357, 2022.
  • [15] Erman Çineli, Viktor L. Ginzburg, and Başak Z. Gürel. On the growth of the Floer barcode. Preprint, arXiv:2207.03613, 2022.
  • [16] Erman Çineli, Viktor L. Ginzburg, Başak Z. Gürel, and Marco Mazzucchelli. On the barcode entropy of Reeb flows. Preprint, arXiv:2401.01421, 2024.
  • [17] Arnon Chor and Matthias Meiwes. Hofer’s geometry and topological entropy. Compos. Math., 159(6):1250–1299, 2023.
  • [18] Lucas Dahinden. Lower complexity bounds for positive contactomorphisms. Israel J. Math., 224(1):367–383, 2018.
  • [19] Michael Entov and Leonid Polterovich. Calabi quasimorphism and quantum homology. Int. Math. Res. Not., (30):1635–1676, 2003.
  • [20] Elijah Fender, Sangjin Lee, and Beomjun Sohn. Barcode entropy for Reeb flows on contact manifolds with Liouville fillings. Preprint, arXiv:2305.04770, 2023.
  • [21] Andreas Floer. Morse theory for Lagrangian intersections. J. Differential Geom., 28(3):513–547, 1988.
  • [22] John M. Franks and Michael Handel. Entropy and exponential growth of π1\pi_{1} in dimension two. Proc. Amer. Math. Soc., 102(3):753–760, 1988.
  • [23] Urs Frauenfelder and Felix Schlenk. Volume growth in the component of the Dehn-Seidel twist. Geom. Funct. Anal., 15(4):809–838, 2005.
  • [24] Urs Frauenfelder and Felix Schlenk. Fiberwise volume growth via Lagrangian intersections. J. Symplectic Geom., 4(2):117–148, 2006.
  • [25] Viktor L. Ginzburg, Başak Z. Gürel, and Marco Mazzucchelli. Barcode entropy of geodesic flows. Preprint, arXiv:2212.00943, 2022.
  • [26] Helmut Hofer. On the topological properties of symplectic maps. Proc. Roy. Soc. Edinburgh Sect. A, 115(1-2):25–38, 1990.
  • [27] Anatole Katok. Lyapunov exponents, entropy and periodic orbits for diffeomorphisms. Inst. Hautes Études Sci. Publ. Math., (51):137–173, 1980.
  • [28] Anatole Katok and Boris Hasselblatt. Introduction to the modern theory of dynamical systems, volume 54 of Encyclopedia of Mathematics and its Applications. Cambridge University Press, Cambridge, 1995. With a supplementary chapter by Katok and Leonardo Mendoza.
  • [29] Asaf Kislev and Egor Shelukhin. Bounds on spectral norms and barcodes. Geom. Topol., 25(7):3257–3350, 2021.
  • [30] Felix Klein. Über Riemann’s Theorie der algebraischen Functionen und ihrer Integrale. Teubner, Leipzig, 1882.
  • [31] Leonardo Macarini and Felix Schlenk. Positive topological entropy of Reeb flows on spherizations. Math. Proc. Cambridge Philos. Soc., 151(1):103–128, 2011.
  • [32] Pierre-Alexandre Mailhot. The spectral diameter of a Liouville domain. Preprint, arXiv:2205.04618, 2022.
  • [33] Dusa McDuff and Dietmar Salamon. Introduction to symplectic topology. Oxford Mathematical Monographs. The Clarendon Press, Oxford University Press, New York, second edition, 1998.
  • [34] Dusa McDuff and Dietmar Salamon. JJ-holomorphic curves and symplectic topology, volume 52 of American Mathematical Society Colloquium Publications. American Mathematical Society, Providence, RI, second edition, 2012.
  • [35] Matthias Meiwes. Rabinowitz Floer homology, leafwise intersections, and topological entropy. PhD thesis, 2018.
  • [36] Matthias Meiwes. Topological entropy and orbit growth in link complements. Preprint, arxiv:2308.06047, 2023.
  • [37] Al Momin. Contact homology of orbit complements and implied existence. Journal of Modern Dynamics, 5:409–472, 2010.
  • [38] Sergey M. Natanzon. Topological classification of pairs of commuting antiholomorphic involutions of Riemann surfaces. Uspekhi Mat. Nauk, 41(5(251)):191–192, 1986.
  • [39] Sheldon E. Newhouse. Entropy and volume. Ergodic Theory Dynam. Systems, 8(Charles Conley Memorial Issue):283–299, 1988.
  • [40] Yong-Geun Oh. Floer cohomology of Lagrangian intersections and pseudo-holomorphic disks. I. Comm. Pure Appl. Math., 46(7):949–993, 1993.
  • [41] Yong-Geun Oh. Construction of spectral invariants of Hamiltonian paths on closed symplectic manifolds. In The breadth of symplectic and Poisson geometry, volume 232 of Progr. Math., pages 525–570. Birkhäuser Boston, Boston, MA, 2005.
  • [42] Leonid Polterovich. Growth of maps, distortion in groups and symplectic geometry. Invent. Math., 150(3):655–686, 2002.
  • [43] Leonid Polterovich, Daniel Rosen, Karina Samvelyan, and Jun Zhang. Topological persistence in geometry and analysis, volume 74. American Mathematical Soc., 2020.
  • [44] Leonid Polterovich and Egor Shelukhin. Autonomous Hamiltonian flows, Hofer’s geometry and persistence modules. Selecta Math., 22(1):227–296, 2016.
  • [45] Leonid Polterovich and Jean-Claude Sikorav. A linear isoperimetric inequality for the punctured Euclidean plane. Preprint, arXiv:0106216, 2001.
  • [46] Matthias Schwarz. On the action spectrum for closed symplectically aspherical manifolds. Pacific J. Math., 193(2):419–461, 2000.
  • [47] Michael Usher. Hofer’s metrics and boundary depth. Ann. Sci. Éc. Norm. Supér. (4), 46(1):57–128, 2013.
  • [48] Michael Usher and Jun Zhang. Persistent homology and Floer-Novikov theory. Geom. Topol., 20(6):3333–3430, 2016.
  • [49] Jan B. van den Berg, Robert Ghrist, Robert C. Vandervorst, and Wojciech Wójcik. Braid Floer homology. J. Differential Equations, 259(5):1663–1721, 2015.
  • [50] Claude Viterbo. Symplectic topology as the geometry of generating functions. Math. Ann., 292(4):685–710, 1992.
  • [51] Yosef Yomdin. Volume growth and entropy. Israel J. Math., 57(3):285–300, 1987.