Quantum Steenrod powers and Hamiltonian maps

Assume on the contrary that {\it QSt}_p(\mu) = St_p(\mu). Let \phi be an {{\mathbb F}_p} Hamiltonian pseudorotation generated by a Hamiltonian H. Let \overline{x} \in \tilde {\mathcal O}(H) be provided by Proposition 4.6. Then by Lemma 4.3, for any k\geq 0, the iteration \overline{x}^{(p^k)} cannot be a symplectically degenerate maximum. Then for all k\geq 0, \Delta(\overline{x}^{(p^k)}) \neq 0.

By Proposition 4.6, when k is sufficiently large, HF^{{\rm loc}, 2n} (H^{(p^k)}, \overline{x}^{(p^k)}) \neq 0. It follows that the mean index \Delta(\overline{x}^{(p^k)}) is negative because HF^{\rm loc}(H^{(p^k)}, \overline{x}^{(p^k)}) is supported over degrees in the interval [-\Delta(\overline{x}^{(p^k)}), -\Delta(\overline{x}^{(p^k)}) + 2n] using our grading convention. By homogeneity of the mean index, we see that

However, just as above, note that HF^{\rm loc} (H^{(p^k)}, \overline{x}^{(p^k)}) is supported over degrees in the interval [-\Delta(\overline{x}^{(p^k)}), -\Delta(\overline{x}^{(p^k)}) + 2n], which contradicts the conclusion of Proposition 4.6 when k is sufficiently large. Therefore, {\it QSt}_p(\mu) \neq St_p(\mu).

Without loss of generality, we may assume that G = {\mathbb Z}/p' for a prime p', generated by \phi \in {\rm Ham}(M, \omega). We also assume by contradiction that for some prime p \neq p', {\it QSt}_p(\mu) = St_p(\mu), which can be translated to (?), i.e.

We focus on the case p\neq 2; the argument for the p=2 case is similar. Choose a Hamiltonian H generating \phi. By Proposition C.17, for all k \geq 0, one has a canonical decomposition

Let {\mathcal F}_1, \ldots, {\mathcal F}_m be all the connected components of {\rm Fix}_c(\phi). Then one can decompose

Proposition C.17 also provides a decomposition of \mu_{\mathbb{Z}/p} written as

where

The rest of the proof is completely parallel to that of Theorem 1.2. The key is an analogue of Proposition 4.6, which says that there exists a capped component \overline{\mathcal F} such that for all sufficiently large k,

Notice that the divisibility lemma holds in this case as the local equivariant pair-of-pants product coincides with the classical Steenrod operation (Proposition C.13) where the roles of Proposition 3.3 and Proposition 3.4 are replaced by Proposition C.15.

Let’s first consider the case when \phi^{p^k} = \mathrm{Id}. Using Proposition C.14 and the PSS isomorphism [Bai_Xu_2025, Theorem E], if H is the Hamiltonian generating \phi, we see that

where we only sum over contractible components of the fixed points. Using Proposition C.9 and the identification between local Floer cohomology under different cappings, we see that

On the other hand, by applying the classical Smith inequality iteratively, we have

If there were noncontractible components of {\mathcal O}(H), since \phi is Morse–Bott, the above two inequalities cannot hold simultaneously.

In general, we have \phi^{p_1^{k_1} \cdots p_l^{k_l}} = \rm{Id} where p_1, \dots, p_l are distinct prime numbers. Then, we can induct on l following [AS-torsion, Section 5.10]. This finishes the proof.

A standard argument (see [GG-generic, Sugimoto-generic]), which is a local argument, shows that for a C^2-generic Hamiltonian diffeomorphism either all of its periodic points are hyperbolic or it has infinitely many periodic points by a variant of the Birkhoff-Lewis theorem [Moser]. However, if all the periodic points of \phi are hyperbolic and there are finitely many of them, then an iteration \psi = \phi^{k_0} as above must be a pseudo-rotation with all periodic points hyperbolic. Theorem 1.6 now applies and finishes the proof.

We work over characteristic 2 and use quantum Steenrod squares in the monotone setting. Let \mu \in H^4 (X; {\mathbb F}_2) be the point class. By [covariant-constant, Theorem 1.7], one has

where \theta \in H^1(B {\mathbb Z}/2; {\mathbb F}_2) is the generator. On the other hand, let

be the usual Kunneth isomorphism. Consider

Then by the Kunneth property of the quantum Steenrod squares (Theorem 1.11) we have

On the other hand, the classical Steenrod square satisfies the same Kunneth property. Hence

Now we prove that a C^2-generic Hamiltonian diffeomorphism \phi on M has infinitely many periodic points. Recall that a C^2-generic \phi either has infinitely many periodic points or all periodic points are hyperbolic. If the claim is not true, then \phi has finitely many periodic points all of which are hyperbolic. Then the calculation {\it QSt}_2(u) = St_2(u) and the fact r = 8 > 6 = n contradict Theorem 1.6, see Remark 1.10.

We adapt Sugimoto’s argument in [Sugimoto] to the current, more general situation. Let H be a Hamiltonian as in the statement. Suppose, for a sufficiently large p, all 1-periodic orbits of H^{(p)} in class p \gamma are iterations of the generators of CF_\gamma(H). We prove that for the next prime p'>p, the Hamiltonian H^{(p')} has a 1-periodic orbit in homology class p\gamma. Since any such orbit must be simple (for p being sufficiently large relative to [\gamma]), this would prove Theorem 1.16.

Now by our assumption, there exists a \gamma-capped 1-periodic orbit x of H in class \gamma with HF_\gamma^{\rm loc}(H, x) \neq 0. Let \overline{x}^{(p)} be the obvious \gamma^{(p)}-capped 1-periodic orbit of H^{(p)}. Then for p sufficiently large, by Lemma 4.11, there holds

By adding a constant C_p to the Hamiltonian H^{(p)}, one may assume that

Claim. Let \inf \beta_{i,\gamma}(H) \geq 2\epsilon >0. Then there exists a (small) constant \delta>0 and a nondegenerate perturbation H_p of H^{(p)} such that the natural map

is nonzero.

Indeed, using the inequality (?), the argument of [Sugimoto, Proof of Theorem 2, Section 5] directly carries over: we have \inf \beta_{i,p\gamma}(H^{(p)}) \geq 2p \epsilon. If \delta > 0 is sufficiently small, we see that \overline{x}^{(p)} induces a well-defined cohomology class on both sides such that the version in the latter is the image of the version in the former.

Now consider the Hamiltonian H^{(p')} and a small nondegenerate perturbation H_{p'}. Notice that one can choose H_{p'} such that the pointwise difference between H_{p'} and H_p is close to (p'-p) \|H\| up to an error at most 2\delta. Hence by choosing homotopies between H_p and H_{p'}, one can build continuation maps

such that they naturally induce

and

However, as p'-p = o(p) as p\to +\infty, one has (p'-p)\| H \| \ll p\epsilon. Hence one has the following commutative diagram.

The above claim implies that the bottom item in the diagram is nonzero. It follows that H_{p'} has some 1-periodic orbit in class p\gamma. As H_{p'} can be arbitrarily close to H^{(p')}, it follows that H^{(p')} has a 1-periodic orbit in class p\gamma.

Indeed, the algebraic definition of C_{\mathbb{Z}/p}(CM(f)) uses a specific {\mathbb{Z}/p}-invariant cellular decomposition of S^\infty (see details in [covariant-constant]) and the {\mathcal{R}_p}-module structure (indeed only the cochain map \tilde\theta) depends on a {\mathbb{Z}/p}-equivariant cellular approximation of the diagonal map of S^\infty. The Morse-theoretic definition, which counts Y-shaped flow trees, gives another equivariant cellular approximation. As the cellular approximations are (chain) homotopic, the induced multiplications agree on the cohomology level.

The idea is the same as the comparison in the nonequivariant case as discussed in [Bai_Xu_2025, Theorem G]. At the level of moduli spaces, the construction in [SZhao-pants] differs from [Bai_Xu_2025] in that the former achieves transversality by choosing a certain family of time-dependent almost complex structures parametrized by S^\infty while the latter constructs the moduli spaces using a fixed almost complex structure, followed by abstract perturbations on Kuranishi charts of these moduli spaces. Nevertheless, the global Kuranishi chart construction in [Bai_Xu_2025, Section 25] applies equally well to the former situation. Then, following the argument in [Bai_Xu_2025, Theorem G], one sees that the induced Kuranishi sections on the derived orbifold charts are already FOP transverse. By passing to the {\mathbb Z}/p-invariant part of the resulting cochain complex, one can directly identify the differential with the formulas in [SZhao-pants, Section 6.1]. Therefore, we have shown that the equivariant cochain complex produced from the procedure in [Bai_Xu_2025] for this varying family of almost complex structures strictly agrees with CF_{\mathbb{Z}/p}^{\rm loc} (H_1^{(p)}, \overline{x}{}^{(p)})_{\rm SZ}. Finally, the desired chain map is obtained by a continuation map argument, which is a chain homotopy equivalence because we can also use continuation maps to construct a chain homotopy inverse.

The proof contains several steps. First, one uses a certain invariance property of the local (equivariant) Floer cohomology to reduce the consideration to certain normal forms of the germ of the Hamiltonian. Second, the normal form can be further deformed to a situation where the {\mathbb{Z}/p}-actions on the moduli spaces are all trivial.

We first consider the invariance property of the local (equivariant) Floer cohomology we will need. Let \phi_s, s \in [0, 1] be a continuous family of Hamiltonian diffeomorphisms and x \in {\rm Fix}(\phi_s) for all s. We say that x is uniformly isolated for the family \phi_s if there exists a fixed open neighborhood U\subset M of x such that for all s, {\rm Fix}(\phi_s) \cap U = \{x\}.

Now we follow a local argument of Ginzburg–Gurel [GG-local-gap]. Let x be an isolated and p-admissible fixed point of \phi. Let X_0 \subset T_x M be the generalized eigenspace of d\phi_x for eigenvalue 1 and X_1 \subset T_x M be the direct sum of other generalized eigenspaces. Both X_0 and X_1 are symplectic subspaces and one can identify a neighborhood of x with an open ball of X_0 \oplus X_1. Then by the argument of [GG-local-gap, Section 4.5], one can construct a family of Hamiltonian diffeomorphisms \phi_s defined in such a neighborhood satisfying

1. \phi_0 = \phi.

2. x is uniformly isolated for both the family \phi_s and the family \phi_s^p.

3. d\phi_s remains constant at x.

4. \phi_1 is the product of Hamiltonian diffeomorphisms \psi_0 on X_0 and \psi_1 on X_1 such that 0 \in X_0 is totally degenerate for \psi_0 and 0 \in X_1 is nondegenerate for \psi_1. Moreover, \psi_0 is autonomous, generated by a smooth function f_0: X_0 \to {\mathbb R}.

Then by the persistence of local (equivariant) Floer cohomology, one can reduce the proof to the case of \phi_1 which has the split form. This is the situation where equivariant transversality can be achieved by using a constant almost complex structure. Hence one can see that

One also does not need to perturb the Floer equation to achieve transversality for the moduli spaces defining the {\mathcal{R}_p}-module structure on the cohomology. Hence one has the isomorphism (2.16) as {\mathcal{R}_p}-modules.

The comparison of total bar-lengths is via the Tate bar-length spectrum of H^{(p)}. By Theorem 2.13, one can write the differential \widehat{d}_{{\mathbb{Z}/p}} on the Tate complex

in the form

where K does not decrease the (t, \theta)-degree. The leading 2\times 2 matrix is the cone of S_p, whose differential is denoted temporarily by d_{\rm Cone}. Then one can use the argument in [S-HZ]. More precisely, by [S-HZ, Proposition 17], one has

where the latter is the total bar-length of d_{\rm Cone}. Moreover, as S_p is null-homotopic (see Theorem 2.13) and strictly increases the energy filtration, by [S-HZ, Lemma 18], one has

Together with (?), one obtains (2.20).

By Lemma 2.21, one can use results from [SZhao-pants]. There, a local equivariant pair-of-pants coproduct is constructed. When x^{(p)} is nondegenerate, it has the form

It was also proved that

Note that H(CF^{\rm loc}(H, x)_{\mathbb{Z}/p}^{\otimes p}) is also 1-dimensional, generated by \overline{x}^{\otimes p}. Hence

where a_x \in {\mathcal{R}_p}\setminus \{0\} is a homogeneous divisor of t^{n(p-1)}. Hence a_x can only be a nonzero multiple of a power of t whose degree is at most n(p-1). Lastly, the quasi-Frobenius map is simply \overline{x} \mapsto \overline{x}^{\otimes p}.

See Appendix ?.

By definition, one has

which, by (2) of Proposition 3.1, is also equal to the dimension of the cohomology of C(H). Hence the differential on C(H) vanishes.

See Appendix ?.

See Appendix ?.

Consider a version of the previous construction in this section over the universal Novikov ring \Lambda_{0}^{\rm univ} as follows. Choose for each x \in {\mathcal O}(H) a capping \overline{x} and define

and

Then d_H canonically induces a \Lambda_0^{\rm univ}-linear differential on C(H; \Lambda_0^{\rm univ}). Similarly, one can define the Tate version

Then the cochain map \widehat{P} of (?) restricts to a \Lambda_{0, {\mathcal{K}_p}}^{\rm univ}-linear cochain map

As the local restrictions are chain homotopy equivalences, \widehat{P}_0 induces an isomorphism of \Lambda_{0, {\mathcal{K}_p}}^{\rm univ}-modules

Therefore, the Tate bar-length spectrum is computed by

Indeed, if d_{{\mathbb{Z}/p}} \neq 0, then the Tate version \widehat{C}_{{\mathbb{Z}/p}}(H^{(p)}) has a nontrivial differential. As d_{{\mathbb{Z}/p}} (which is also the differential of \widehat{C}_{\mathbb{Z}/p}(H^{(p)})) strictly increases the filtration, the Tate bar-length spectrum has a nonzero element \widehat{\beta}_1(H^{(p)}). By Corollary 3.5, one has \widehat{\beta}_1(H^{(p)}) = p \beta_1(H) > 0. However, this contradicts Corollary 3.2.

See Appendix ?.

We go through the homological perturbation in a slightly different way. Let G be a nondegenerate perturbation of H which is sufficiently close such that G^{(p)} is also nondegenerate. Consider the equivariant Floer complex CF_{{\mathbb{Z}/p}}(G^{(p)}). By the discussion prior to Theorem 2.13, one can identify

Moreover, by Theorem 2.13, one can write the equivariant differential as

If we ignore the terms S_p and tK, then via the homological perturbation construction given in the proof of Proposition 3.1, with H replaced by H^{(p)}, one obtains copies of the complex C(H^{(p)}) with a differential d_{H^{(p)}} which strictly increases the energy. One could, in addition, view the nilpotent term S_p and the term tK having positive t degrees as perturbations. Then one obtains a complex, denoted temporarily by

whose differential has the form

Further, it has the local part and “high energy” part, while the local part computes the local equivariant Floer cohomology. As d_{H^{(p)}} has vanishing local parts, the local part of \widetilde{d}_{{\mathbb{Z}/p}} reads

However, one knows that the local equivariant cohomology is isomorphic to HF^{\rm loc}(H^{(p)}, x^{(p)}) \otimes {\mathcal{R}_p}. Hence \widetilde{S}_p^{\rm loc} and \widetilde{K}^{\rm loc} both vanish. This implies that the complex \widetilde{C}_{{\mathbb{Z}/p}}(H^{(p)}) is indeed a case of the complex C_{{\mathbb{Z}/p}}(H^{(p)}). Moreover, the difference d_{{\mathbb{Z}/p}} - d_{H^{(p)}} strictly increases the energy filtration. Therefore, the bar-length spectrum of the complex C_{{\mathbb{Z}/p}}(H^{(p)}), if not empty, has no zero bars. Then by the forms of the ordinary and equivariant differentials, the assertion follows.

We first reduce the consideration to the case with a local normal form as in the proof of Proposition 2.16. Recall that the local (equivariant) Floer cohomology is invariant under deformation as long as the fixed point remains uniformly isolated through the deformation. Moreover, the identifications between different moments through the deformation are given by certain continuation maps. As the local equivariant pants product is compatible with continuation maps, one can assume that H has a product type. In other words, let \phi be the time-1 map of H. Then we may assume that M = X_1 \times X_2 and \phi = \phi_1 \times \phi_2 are products. If x = (x_1, x_2), then x_1 is a nondegenerate fixed point of \phi_1 and x_2 is a totally degenerate and isolated fixed point of \phi_2.

To save notations, abbreviate HF^{\rm loc}(\phi, x) by HF^{\rm loc}(\phi) (and in other similar occasions) as the critical points in the consideration are clear from the context. Let

be the local equivariant pants product and let

be their compositions with the quasi-Frobenius map y \mapsto y^{\otimes p}.

The basic strategy in the rest of the proof is to use the Kunneth property of the local equivariant pants product (Proposition 5.6) and the leading order bounds of the equivariant pants product on the two factors.

Then by Proposition 5.4 and Proposition 5.6 proved below, the following diagram commutes up to the Koszul sign.

Notice that the non-equivariant Kunneth map \kappa^{\rm loc} is an isomorphism. On the other hand, from the proof of Proposition 2.16 one knows that \kappa_{\mathbb{Z}/p}^{\rm loc} is also an isomorphism, although it wouldn’t be true if we drop the p-admissibility condition.

Now we prove that both P_{x_1}^{\rm loc} and P_{x_2}^{\rm loc} satisfy the bound we would like to prove. We first analyze the totally degenerate piece P_{x_2}^{\rm loc}. As x_2 is also p-admissible, by Proposition 2.16, one has an isomorphism of {\mathcal{R}_p}-modules

Claim. If we decompose

then

Proof of the claim. The basic idea is to reduce the consideration to local Steenrod operations on local Morse cohomology. By the argument of Proposition 2.16, the local equivariant cohomology can be computed using an autonomous Hamiltonian h and the isomorphism is provided by a continuation map. Therefore, the cochain level map {\it FSt}_{p, x}^{\rm loc} can also be defined using the autonomous h, denoted by

We know that the local Morse cohomology is identified with a relative cohomology H^*(B, N) with B being a 2n_2-dimensional ball while N \subset \partial B is a codimension zero submanifold with boundary.

Using an argument analogous to [wilk, Section 3.2] for general primes p, one can show that P_{\rm Morse}^{\rm loc} is identified with the classical Steenrod operation

given by [Seidel-formal, Equation (1.6)]. By the classical theory of Steenrod operations, the coefficient of t^\alpha \theta^\beta of St_p(z) for |z|=k vanishes when 2\alpha + \beta > k (p-1). As H^*(B, N) has non-vanishing degrees up to {\rm dim} B = 2n_2, the assertion of the claim follows. (The required conclusion also follows directly from an analysis of the Morse moduli spaces defining P_{\rm Morse}^{{\rm loc}}, as in the proof of [wilk, Proposition 3.3], specifically “Axiom 5".) End of the proof of the claim.

On the other hand, the nondegenerate piece is an easy consequence of Theorem 2.22. If e_1 \in HF^{\rm loc}(\phi_1) is a generator (which induces a corresponding generator e_1^{(p)} of HF_{\mathbb{Z}/p}^{\rm loc} (\phi_1^p)), then there exists a_1 \leq n_1 (p-1) and c_1 \neq 0 such that

Now we can use the Kunneth property of the local equivariant pants product (Proposition 5.6). Let e_1 \in HF^{\rm loc}(\phi_1; {\mathbb F}_p) be a generator and y_2 \in HF^{\rm loc}(\phi_2; {\mathbb F}_p). Suppose P^{\rm loc}( e_1 \otimes y_2 ) \neq 0. Then by (?) and the Kunneth property, one has

Here we used the property that \kappa_{\mathbb{Z}/p}^{\rm loc} is {\mathcal{R}_p}-bilinear.

Now let P^{\rm loc}_{x_2, i_2}(y_2) be the leading order term of P_{x_2}^{\rm loc} (y_2). Notice that \kappa_{\mathbb{Z}/p}^{\rm loc} does not decrease the (t, \theta)-grading and its leading order term is the non-equivariant Kunneth map \kappa^{\rm loc}, which is an isomorphism. Then the leading term of P^{\rm loc}(\kappa^{\rm loc}(e_1 \otimes y_2)) is a nonzero multiple of

which has (t, \theta)-degree no greater than 2n(p-1).

The assertion follows from the methods of proof of [Ginzburg-CC, Proposition 4.7], [GG-ai, Theorem 1.18], and [Hein-CCCY, Theorem 1.3]. Indeed, [Hein-CCCY] adapts the methods of [Ginzburg-CC] to show that on an arbitrary closed symplectically Calabi–Yau manifold, there holds

The argument in [Hein-CCCY] applies equally well for an arbitrary closed symplectic manifold, as it relies on a general decomposition theorem for filtered Floer homology and Ginzburg’s arguments [Ginzburg-CC] applied to the “local summand" of the decomposition.

Furthermore, Ginzburg proves that the above connecting map is non-trivial in the aspherical case, and this extends, by applying Hein’s decomposition, to the case of an arbitrary closed symplectic manifold.

Suppose on the contrary, for some k\geq 0, \tilde {\mathcal O}(H^{(p^k)}) contains some symplectically degenerate maximum. As \phi is an {{\mathbb F}_p} pseudo-rotation, \phi^{p^k} is also an {{\mathbb F}_p} pseudo-rotation. Hence we may assume that k = 0. By Lemma 4.2, if \overline{x} \in \tilde{\mathcal{O}}(H) is a symplectically degenerate maximum, then for every sufficiently small \epsilon > 0, there exists r_{\epsilon} \in \mathbb{Z}_{\geq 0} such that for all r \geq r_{\epsilon} we can find \delta \in (0,\epsilon) such that the connecting homomorphism (4.2) is nonzero. However, it follows from Corollary 3.2 that when r = p^k, the differential of C(H^{(p^k)}), as well as that of C(H^{(p^k)})^{(a,b)} for an arbitrary window (a,b) vanishes. Hence the connecting homomorphism (4.2) also vanishes, which is a contradiction. This finishes the proof.

We modify the pseudo-rotation to satisfy the condition of Lemma 4.1. As \phi is an {{\mathbb F}_p} Hamiltonian pseudo-rotation, we may replace \phi by \phi^{p^k} for any k\geq 0. Hence we may assume that for all x \in {\rm Fix}_c(\phi),

Now we translate the condition on the quantum Steenrod operation {\it QSt}_p(\mu) = t^{n(p-1)} \mu to one about the equivariant pants product. Using Theorem 2.18, one has

By abuse of notations, via {\it PSS}, we view \mu as its image {\it PSS}(\mu) \in HF(M). On the other hand, denote \mu_{\mathbb{Z}/p}:= {\it SSP}_{\mathbb{Z}/p}^{-1}(\mu) \in HF_{\mathbb{Z}/p}(M). Then one has

Now one can localize the above equation to periodic orbits. Denote

By Proposition 3.1, as \phi is an {{\mathbb F}_p} Hamiltonian pseudo-rotation, for all k\geq 0, one has

With respect to this decomposition, we can write

We consider a similar decomposition for the equivariant Floer cohomology. Applying Proposition 3.3 and Corollary 3.6, one has

So the class \mu_{{\mathbb{Z}/p}} can be decomposed as

Claim A. For all k\geq 0 and i, the leading order term \mu_{{\mathbb{Z}/p},i}^{(p^k)}(0,0) coincides with \mu_i^{(p^k)} up to higher energy terms.

Proof of Claim A. On the cochain level, \mu_{{\mathbb{Z}/p},i}^{(p^k)}(t, \theta) is realized by the inverse of the cochain-level equivariant SSP map. The involved moduli spaces contain the leading one corresponding to the non-equivariant SSP map, which gives \mu_i^{(p^k)} up to high energy terms by the results in [Bai_Xu_2025, Section 13.7]. End of the proof of Claim A.

One can also view \mu_{{\mathbb{Z}/p},i}^{(p^k)}(t, \theta) as a class in the Tate cohomology

What we do is to consider the ordinary and Tate spectral invariants and their “carriers” under the p^k-th iterations.

Claim B. For all k \geq 1, there holds

Proof of Claim B. The decomposition of \mu implies that

The Tate version implies

Claim A implies that

Moreover, as the equivariant pants product rescales the spectral invariants by p, one has

Then inductively, c(\mu, H^{(p^k)}) \geq p^k c(\mu, H). End of proof of Claim B.

Claim B has a local version. Suppose for some i, {\mathcal A}_{H^{(p^k)}}(\mu_i^{(p^k)}) = c(\mu, H^{(p^k)}).(One can roughly regard x_i, or its p^k-th iteration, as a spectral carrier.) Then

The proposition will follow from a specific quantitative statement and induction. For each k\geq 0, define

Claim C. There holds

We prove this assertion later. If it is true, then it follows that c(\mu, H^{(p^k)}) = p^kc (\mu, H) for all k. Choose i from the nonempty intersection. As all classes have a fixed degree 2n, by the definition of the group \Gamma and the Novikov ring, there exists a unique capping \overline{x}_i of x_i and a class

such that

Notice that the p^k-th iterations of \overline{x}_i still have degree 2n and action being p^k times the action of \overline{x}_i. It follows that one can write

Therefore, the assertion of Proposition 4.6 holds.

Now we prepare to prove Claim C. By induction, it is equivalent to

The inclusion I^{(p)} \subset I^{(1)} follows from (?). To prove the nonemptiness of I^{(p)}, one has to consider the Tate spectral numbers and use the decomposition of the equivariant pants product given by Proposition 3.4. Indeed, one can write

We know that P^{\rm big} strictly increases the energy filtration and is additive. Hence

On the other hand, since the (Tate version) operation \widehat{{\it FSt}}_p precisely rescales the energy filtration by p, we have

As the local equivariant pair-of-pants product rescales the local action by p times,

On the other hand, using (?), one has

Claim D. For each i \notin I^{(1)}, one has

Proof of Claim D. Choose i \notin I^{(1)}. So {\mathcal A}_H(\mu_i^{(1)}) > c(\mu, H). Notice that \mu_{{\mathbb{Z}/p},i}^{(p)}(t, \theta), if nonzero, either comes from P^{\rm loc}(\mu_i^{(1)}) or P_{\rm big}(\mu). Both of them have action strictly greater than p c(\mu, H). End of proof of Claim D.

Now we can prove Claim C. First, Claim D and (?) imply that

Hence there exists some i_0 \in I^{(1)} such that

Now the i_0-th component of (?) reads

As the P^{\rm big}(\mu) term has strictly higher energy, the first term on the right (ordered by (t, \theta)-degrees) which realizes the action, coincides with the leading order term of P^{\rm loc}(\mu_{i_0}^{(1)}). However, by Lemma 4.1, the leading degree can be at most 2n(p-1). Hence

By Claim A, this also agrees with {\mathcal A}_{H^{(p)}}(\mu_{i_0}^{(p)}). This finishes the proof of Claim C and hence the proposition.

The argument is very similar to that of Lemma 4.3. First, as the local Floer cohomology is defined over integers,

by the universal coefficient theorem, there exists p_0(\phi) such that for all primes p \geq p_0(\phi) and contractible fixed points x \in {\rm Fix}_c(\phi), there holds

Hence for p \geq p_0(\phi), one has

Further, by [GG-local-gap, Theorem 1.1], for p sufficiently large and all contractible fixed points x \in {\rm Fix}_c(\phi), one has

As \phi has only finitely many periodic points, when p is sufficiently large, one has {\rm Fix}_c(\phi) = {\rm Fix}_c(\phi^p ). Hence

If H is a 1-periodic Hamiltonian whose time-1 map is \phi, the above identity implies that the differential of the complex C(H^{(p)}) constructed in Proposition 3.1 vanishes. Replacing p^k in the proof of Lemma 4.3 by a sufficiently large p, one can obtain the same conclusion.

We can run the same argument as proving Proposition 4.6. Notice that one no longer has an induction from p^k to p^{k+1}. First, since \phi has only finitely many periodic points, by considering large primes p, one has

Then, the equation {\it QSt}_p(\mu) = {\it St}_p(\mu) can be translated, via the ordinary PSS map and the equivariant SSP map, to

Let x_1, \ldots, x_m be all contractible fixed points of \phi. Then by Proposition 3.1, one can decompose {\it PSS}(\mu) as

and

For p sufficiently large, one can also assume that each x_i is p-admissible. Using Proposition 3.3 and Proposition 2.16, one can decompose \mu_{{\mathbb{Z}/p}} as

Then one has the analogue of Claim A in the proof of Proposition 4.6, i.e.,

Claim A. For infinitely many primes p, the leading order term \mu_{{\mathbb{Z}/p}, i}^{(p)}(0,0) coincides with \mu_i^{(p)} up to higher energy terms.

The proof is similar. Then as a consequence, for infinitely many primes p, there holds

This is the analogue of Claim B in the proof of Proposition 4.6. Moreover, there exists i such that {\mathcal A}_H(\mu_i) = c(\mu, H) (i.e., x_i is a spectral carrier). Then for such i, one has

Then one can prove that for each p, there exists i_p such that

using the same argument. As there are infinitely many such p, by the pigeon-hole principle, one can choose i_0 such that for infinitely many p, i_0 = i_p.

Let \phi be the time-1 map of H. Then x = x(0) is an isolated fixed point of \phi. Notice that the local Floer cohomology HF_\gamma^{\rm loc}(H, \overline{x}; {\mathbb K}) only depends on the germ of the symplectomorphism \phi near x. Moreover, it is well-known that the germ of a symplectomorphism near a fixed point is Hamiltonian (see [GG-local-gap, Remark 4.1]). Then one can reduce the consideration to a local problem where the fixed point is contractible. Then the lemma follows from the contractible case [GG-local-gap, Theorem 1.1].

Fix an odd prime p. Suppose this is not the case. Then for all k \geq 1, the differential d_{H^{(p^k)}} vanishes. Recall that by Proposition 3.1, the cohomology of C(H^{(p^k)}) is isomorphic to the Floer cohomology of (M, \omega). Therefore one has

Then by Definition 1.1, \phi is an {{\mathbb F}_p} Hamiltonian pseudo-rotation.

This contradicts Lemma 4.3.

Choose classes a_i \in HM(f_i) and choose representing cocycles x_i \in CM(f_i). Recall that qF(a_i) \in H(CM(f_i)^{\otimes p}_{{\mathbb{Z}/p}}) is represented by the equivariant cocycle

Using the Morse model on S^\infty, let w_1, \ldots, w_p \in {\rm crit}(g) \subset S^\infty be the p critical points of g of index 0. Then the above cocycle is equivalent to the Morse cocycle

We claim that the equivariant Kunneth map

gives

Here the sign comes from switching certain copies of x_1 and x_2. This claim essentially follows from the fact that in this situation equivariant transversality holds. Hence the flow line equation in X_1 \times X_2 and the Y-shaped gradient tree equation in S^\infty are decoupled (see Remark 5.2). This verifies the commutativity (up to signs) of the diagram.

The proof is similar to that of Theorem 1.11. The idea is illustrated in Figure Proof.

We first choose data so that all objects and arrows appearing in the commutative diagram can be defined on the chain level.

1. Choose nondegenerate perturbations G_i of H_i and an almost complex structure J_i for which the local Floer cochain complexes CF^{\rm loc}(H_i, G_i, \overline{x}_i) are defined. In this way, the local Floer cochain complex CF^{\rm loc}(H_1 \times H_2, G_1 \times G_2, \overline{x}_1\times \overline{x}_2) is also defined using the product data, giving

   CF^{\rm loc}(H_1\times H_2, G_1 \times G_2, \overline{x}_1 \times \overline{x}_2) \cong CF^{\rm loc}(H_1, G_1, \overline{x}_1) \otimes CF^{\rm loc}(H_2, G_2, \overline{x}_2).

   This isomorphism coincides with the chain-level Kunneth map \kappa^{\rm loc}, which can be defined following the line in Section ?. From now on, the almost complex structures will be fixed and we omit the discussion on the dependence on them.

2. To define the local equivariant complexes CF_{{\mathbb{Z}/p}}^{\rm loc}(H_i^{(p)}, G_i^{(p)}, \overline{x}_i^{(p)}), we also choose a family of perturbations

   {\mathscr G}_{i,v}^{eq}: S^1 \times X_i \to {\mathbb R}, \ v \in S^\infty

   of G_i^{(p)} so the complex is defined.

3. From now on, we drop from the notations the dependence on H_1 and H_2 as all further constructions only depend on G_1 and G_2. To define the local equivariant complex for the product G_1^{(p)} \times G_2^{(p)}, we choose another perturbation

   {\mathscr G}_{\infty,v}^{eq}: S^1 \times X_1 \times X_2 \to {\mathbb R}, \ v \in S^\infty

   so the complex CF_{{\mathbb{Z}/p}}^{\rm loc}(G_1^{(p)} \times G_2^{(p)}, \overline{x}_1^{(p)} \times \overline{x}_2^{(p)}) is defined.

4. Choose a family of vector fields Y_s on S^\infty parametrized by s \in T_{2, 1} so that the multiplicative structure {\mathcal{R}_p} \otimes {\mathcal{R}_p} \to {\mathcal{R}_p} is defined. Then choose a family

   \tilde {\mathscr G}_{s, v}^{eq}: \tilde S \times X_1 \times X_2 \to {\mathbb R},\ s \in T_{2, 1},\ v \in S^\infty

   which connects {\mathscr G}_{1,v}^{eq} \times {\mathscr G}_{2, v}^{eq} and {\mathscr G}_{\infty, v}^{eq} so that the chain-level equivariant Kunneth map

   \kappa^{\rm loc}_{{\mathbb{Z}/p}}: CF_{{\mathbb{Z}/p}}^{\rm loc} (G_1^{(p)}, \overline{x}_1^{(p)}) \otimes CF_{{\mathbb{Z}/p}}^{\rm loc} (G_2^{(p)}, \overline{x}_2^{(p)} ) \to CF_{{\mathbb{Z}/p}}^{\rm loc} (G_1^{(p)}\times G_2^{(p)}, \overline{x}_1^{(p)} \times \overline{x}_2^{(p)})

   is defined.

5. Choose {\mathbb{Z}/p}-equivariant family of connections on the trivial {\rm Ham}(X_i)-bundle on \tilde S denoted by

   \sigma_{i, v}^{eq},\ v \in S^\infty

   which is equal to G_i dt on each of the incoming cylindrical ends and which is equal to {\mathscr G}_i^{eq} on the positive cylindrical end, to define the chain level equivariant pair-of-pants product

   \widetilde{FSt}{}_p^{{\rm loc}}: CF^{\rm loc}(G_i, \overline{x}_i)^{\otimes p}_{{\mathbb{Z}/p}} \to CF^{\rm loc}_{{\mathbb{Z}/p}}(G_i^{(p)}, \overline{x}_i^{(p)}).

6. Choose a {\mathbb{Z}/p}-equivariant family of connections on the trivial {\rm Ham}(X_1\times X_2)-bundle over \tilde S denoted by

   \sigma_{\infty, v}^{eq},\ v \in S^\infty

   which is equal to (G_1 \times G_2) dt on each of the incoming cylindrical ends and is equal to {\mathscr G}_\infty^{eq} dt on the positive cylindrical end, to define the chain-level equivariant pair-of-pants product

   \widetilde{FSt}{}_p^{X_1\times X_2}

Now we would like to choose a family of connections on the trivial {\rm Ham}(X_1)\times {\rm Ham}(X_2)-bundle over \tilde S, denoted by

satisfying

1. \tilde \sigma_{\rho, v_1, v_2}^{eq} is {\mathbb{Z}/p}-equivariant with respect to the diagonal action on S^\infty \times S^\infty \times \tilde S.

2. For all \rho, on each of the negative cylindrical ends of \tilde S, \tilde \sigma_{\rho, v_1, v_2}^{eq} coincides with (G_1 \times G_2) dt.

3. For all \rho, on the positive cylindrical end of \tilde S, \tilde \sigma_{\rho, v_1, v_2}^{eq}, when restricted to the diagonal of S^\infty, coincides with {\mathscr G}_{\infty, v}^{eq} dt.

4. When \rho \ll 0, \tilde \sigma_{\rho, v_1, v_2}^{eq} coincides with \sigma_{1, v_1}^{eq}\times \sigma_{2, v_2}^{eq}.

5. When \rho \gg 0, \tilde \sigma_{\rho, v_1, v_2}^{eq}, when restricted to the diagonal v = v_1 = v_2 of S^\infty, coincides with \sigma_{\infty, v}^{eq}.

We still use the canonical map T_{2,1} \to {\mathbb R}. For i = 1, 2, choose \overline{x}_{i, 1}, \ldots, \overline{x}_{i, p} \in \tilde {\mathcal O}(G_i, \overline{x}_i), \overline{y}_i \in \tilde {\mathcal O}(G_i^{(p)}, \overline{x}_i^{(p)}). Choose w_1, w_2, w_\infty \in {\rm crit}(g). Then for each fixed \rho, define the moduli space {\mathcal M}_\rho as follows. When \rho \leq 0 resp. \rho \geq 0, {\mathcal M}_\rho consists of pairs (u, \zeta) where u: \tilde S \to X_1 \times X_2 is a smooth map, \zeta = (\zeta_1, \zeta_2, \zeta_\infty): T_{2,1} \to S^\infty is a Y-perturbed gradient tree, such that

and such that u is asymptotic to \overline{x}_{1, k} \times \overline{x}_{2, k} along the k-th negative end and asymptotic to \overline{y}_1 \times \overline{y}_2 along the positive end, while \zeta is asymptotic to w_1, w_2, w_\infty respectively along the corresponding edges. Define

One can see that each {\mathcal M}_\rho is the zero locus of a smooth Fredholm section of a certain Banach vector bundle. Moreover, by choosing \tilde \sigma_{\rho, v_1, v_2}^{eq} generically, one can make {\mathcal M}_\pm together with {\mathcal M}_0 regular. One can also compactify the moduli space {\mathcal M}_\pm by suitably adding broken configurations (notice that in this local situation, sphere bubbling won’t happen). Then when the index of an individual moduli space {\mathcal M}_\rho is zero, the compactifications \overline{\mathcal M}_\pm are compact 1-dimensional topological manifolds with boundary. Then the union of \overline{\mathcal M}_\pm along {\mathcal M}_0 defines a chain homotopy between the two cochain maps. The first one, corresponding to the boundary \rho = -\infty, is the composition

The second one, corresponding to the boundary \rho = +\infty, is the composition

between the same pair of cochain groups. The chain homotopy is {\mathbb{Z}/p}-equivariant. Hence we can restrict to the {\mathbb{Z}/p}\times {\mathbb{Z}/p}-fixed subspace, which gives a chain homotopy between

and \widetilde{FSt}{}_p^{\rm loc} \circ \widetilde{\kappa}{}_{{\mathbb{Z}/p}}^{\otimes p} between the same pair of groups. Hence as maps between cohomology, one has the equality

Lastly, one has the following commutative diagram (which can be proved exactly the same as Lemma 5.3).

Then the proof this proposition is finished.

To simplify notations, we omit the superscript “Morse" in this proof. We may assume that \alpha is a single critical point; the case when \alpha is a linear combination of critical points only adds notational complexities. Consider the chain

As \alpha is a cocycle, one can see that \tilde \alpha^p is d_{{\mathbb{Z}/p}}-closed. Now by the definition of the chain map \widetilde{\widetilde{QSt}}_{p, A} and the {\mathbb{Z}/p}-equivariance of the counts, one has

As the cohomology classes of both sides are well-defined, the lemma follows.

Again, we assume that \alpha is a single critical point; the case when \alpha is a linear combination of critical points follows from the special case with only adding more notations. We can choose the inhomogeneous term \nu^{eq} such that the moduli spaces {\mathcal M}_Q(A; x_1, \ldots, x_p, y) are regular when Q is equal to faces of P_i, \Delta_i, or B_i. Then by using 1-dimensional moduli spaces of the form {\mathcal M}_{B_{i+1}}(A; \alpha, \ldots, \alpha, y), by Lemma A.4, one can see the following holds.

1. When i = 2k is even, one has

   \begin{split} &\ QSt^{\rm Morse}_{p,i,A}(\alpha \otimes \cdots \otimes \alpha) - QSt^{\rm cell}_{p,i,A}(\alpha \otimes \cdots \otimes \alpha) \\ = &\ \sum_{y \in {\rm crit}(f)} \left( \#_p {\mathcal M}_{P_i}(A; \alpha, \ldots, \alpha, y) - \#_p {\mathcal M}_{\Delta_i}(A; \alpha, \ldots, \alpha, y) \right) \cdot y\\ = &\ \pm d \Big( \sum_{z \in {\rm crit} (f)} \#_p {\mathcal M}_{B_{2k+1}}(A; \alpha, \ldots, \alpha, z) \cdot z \Big). \end{split}

2. When i = 2k+1 is odd, one has

   \begin{split} &\ QSt^{\rm Morse}_{p,i,A}(\alpha \otimes \cdots \otimes \alpha) - QSt^{\rm cell}_{p,i,A}(\alpha \otimes \cdots \otimes \alpha) \\ = &\ \sum_{y \in {\rm crit}(f)} \Big( \#_p {\mathcal M}_{P_i}(A; \alpha, \ldots, \alpha, y) - \#_p {\mathcal M}_{\Delta_i}(A; \alpha, \ldots, \alpha, y) \Big) \cdot y\\ = &\ \pm d \Big( \sum_{z \in {\rm crit} (f)} \#_p {\mathcal M}_{B_{2k+2}}(A; \alpha, \ldots, \alpha, z) \cdot z \Big) \\ &\ + \sum_{y \in {\rm crit}(f)} \Big( \#_p {\mathcal M}_{\tau(B_{2k+1})}(A; \alpha, \ldots, \alpha, y) - \#_p {\mathcal M}_{B_{2k+1}} (A; \alpha, \ldots, \alpha, y) \Big) \cdot y. \end{split}

   The last line vanishes by the {\mathbb{Z}/p}-invariance of the counts.

Therefore, the difference we are considering is a Morse coboundary.

\Delta_0 = P_0 is obvious and \Delta_1 = P_1 because they are both the unique positively oriented segment of S^1\subset S^\infty between \Delta_0 and \tau(\Delta_0). Hence we can choose B_1 = 0, B_2 = 0.

For any positive even degree i = 2k, by the differential of the Morse cochain complex CM(g), one has

Together with (?), one obtains

Notice that as chains, the sphere S^{2k-1} can be written as

It follows that \partial (P_{2k} - \Delta_{2k}) = 0. As S^\infty is contractible, there hence exists B_{2k+1} \in C_{2k+1}(S^{2k+1};{\mathbb Z}) such that

For any positive odd degree i = 2k+1, by the differential of the Morse cochain complex CM(g), one has

Then together with (?), one has

Again, by contractibility of S^\infty, there exists B_{2k+2} \in C_{2k+2}(S^{2k+2}, {\mathbb Z}) such that

The proof is based on a compactness argument and Lemma B.1. Suppose the statement is not true. Then one can find a sequence \sigma_i of Hamiltonian connection converging to H_0 dt and a sequence of domain-dependent almost complex structures J_i converging to J_0, both in C^\infty-topology, as well as a sequence of solutions u_i to the equation (\nabla^{\sigma_i} u_i)^{0,1} = 0 which do not satisfy the claims. By choosing a subsequence (still indexed by i), one can argue in the following two situations.

1. Suppose there exists x \in {\mathcal O}(H_0) such that for all i, u_i is asymptotic to x_{i, \pm} \in {\mathcal O}(H_{i, \pm}, x). Moreover, for a fixed capping \overline{x} of x with induced ones \overline{x}_{i, \pm}, the concatenation of \overline{x}_{i, -} with u_i coincides with \overline{x}_{i, +}. The condition on the integral of |F_{\sigma_i}| implies that E(u_i) \to 0. Then Lemma B.1 and the usual compactness argument imply that u_i converges uniformly to the constant Floer cylinder u_x. Hence for i sufficiently large, u_i is contained in a C^0-neighborhood of u_x.

2. Suppose for all i we are not in the above situation for any x \in {\mathcal O}(H_0). By contradiction, assume that E(u_i) \leq \frac{1}{2} \epsilon_0. Again one can apply the compactness argument. Lemma B.1 implies that any subsequential limit must be a constant Floer cylinder u_x and the convergence is uniform. This implies that for i sufficiently large, one is indeed in the first situation.

We prove by contradiction. For simplicity assume J_i = J_0; the general case can be treated with only notational complexity. Let G_i be a sequence of nondegenerate Hamiltonians converging to H_0 in C^\infty-topology. Let u_i: \Sigma_p^\circ \to M be a sequence of solutions to the Floer equation (\nabla^{\sigma_i} u)^{0,1} = 0 for \sigma_i being the Hamiltonian connection induced from G_i. We assume that we are in the situation of either (1) or (2) for all i.

Suppose we are in situation (1). By choosing a subsequence, we may assume that for the same x \in {\mathcal O}(H_0), u_i is asymptotic to an orbit y_{i, j} \in {\mathcal O}(G_i, x) for the j-th negative puncture and asymptotic to y_{i, \infty} \in {\mathcal O}(G_i^{(p)}, x^{(p)}) for the positive puncture. Then the condition together with the energy identity for solutions implies that E(u_i) \to 0. The Gromov compactness implies that a subsequence of u_i (still indexed by i) converges to a solution to (\nabla^{\sigma_0} u)^{0,1} = 0 on \Sigma_p^\circ, up to breakings at cylindrical ends. By Lemma B.2, the limit is one of u_{x}^p. Lemma B.1 (applied to both H_0 and H_0^{(p)}) also excludes breakings. Hence the convergence u_i \to u_x^p is uniform, implying that for i sufficiently large, u_i\in U(u_x^p).

Suppose we are in situation (2) for the sequence. If the conclusion is false, then for a subsequence of u_i (still indexed by i) one has

Then by the same compactness argument, a subsequence of u_i converges up to breaking to a u_x^p on \Sigma_p^\circ. Lemma B.1 and the energy bound exclude breakings at both positive and negative ends. The convergence is uniform. Hence for i being sufficiently large one is indeed in situation (1), which is a contradiction.

This lemma follows from Lemma B.3. Let \epsilon_H be \frac{1}{4} \epsilon_0 where \epsilon_0 is the one of Lemma B.1 with H_0 replaced by H. Choose a collection of sufficiently small C^0-neighborhoods U(u_{x_j}) for all x_j \in {\mathcal O}(H).

Then for each i, Lemma B.3 implies the dichotomy that each Floer cylinder of H_i is either contained in U(u_x) for some x \in {\mathcal O}(H) or has energy at least 2\epsilon_H. Hence one can define d_i^{\rm loc} resp. D_i to be the linear map coming from Floer cylinders belonging to the first resp. second category. By the definition of the modified action {\mathcal A}_H, one can see that d_i^{\rm loc} preserves it; moreover, as {\mathcal A}_H can be arbitrarily close to {\mathcal A}_{H_i}, one can see that when i is large, D_i increases the energy filtration by at least \epsilon_H. By this definition, one can see each CF(H_i, \overline{x}) is invariant under d_i^{\rm loc}. Applying the same dichotomy to Floer cylinders of index 1, one can see that d_i^{\rm loc} is a differential. Let d_{i, \overline{x}}^{\rm loc} be the restriction of d_i^{\rm loc} to CF(H_i, \overline{x}).

It remains to show that for each \overline{x}, the complex (CF(H_i, \overline{x}), d_{i,\overline{x}}^{\rm loc}) is chain homotopic to any complex CF^{\rm loc} (G, \overline{x}; {\mathbb Z}) computing the local Floer cohomology HF^{\rm loc}(H, \overline{x}; {\mathbb Z}). Indeed, one can view H_i as a local nondegenerate perturbation near the orbit x. Then the claim follows for the same argument as one shows that the local Floer cohomology is independent of the choices of local perturbations.

For each degree n, let Z^n \subset V^n be the kernel of differential and B^n = d(V^{n-1}). As H(V) is free, the exact sequence

splits. Choose a splitting

Then define \sigma_V to be the inclusion of H^n with respect to this splitting and \pi_V to be the corresponding projection. Then define \Theta_V to be the identification of B^n \subset V^n with B^n \subset V^{n-1}, extended by 0 on H^n \oplus B^{n+1}. It is then easy to see that all the conditions are satisfied.

When i is sufficiently large, consider a continuation map

induced by a homotopy from H_i to H_i' which remains to be C^\infty-close to H. Then, using Lemma B.3, we can decompose

where \Phi_{\rm small} is the contribution from the continuation Floer cylinders whose energy is strictly less than 2 \epsilon_H. One observes that, for each \overline{x} \in \tilde {\mathcal O}(H), \Phi_{\rm small} restricts to a chain homotopy equivalence between local Floer chain complexes. It follows from the construction of the modified action that \Phi_i preserves the action filtration.

The chain-level inclusion CF(H_i')^{{\mathcal A}_{H_i'} >a} \to CF(H_i)^{{\mathcal A}_H>a} induces the long exact sequence

By the exactness of the direct limit functor, it suffices to show that

Indeed, given any [c_i] \in HF(H_i)^{{\mathcal A}_{H_i'} \leq a < {\mathcal A}_H}, there exists \delta>0 such that

Moreover [c_i] can be represented by a cocycle

Then for j>i sufficiently large, the monotone continuation map from CF(H_i) to CF(H_j) sends c_i to a cocycle c_j with {\mathcal A}_{H_j}(c_j) > a + \delta_j \Longrightarrow {\mathcal A}_{H_j'}(c_j) > a. Hence [c_i] cannot survive the direct limit.

We prove by contradiction. Suppose there exists a sequence of solutions u_i: \Sigma_p^\circ \to M with 0 \neq E(u_i) \to 0, then one can apply the compactness argument. First, the energy bound implies that a subsequence of u_i (still indexed by i) converges in C_{\rm loc}^\infty to an energy zero solution u_\infty: \Sigma_p^\circ \to M, hence must coincide with some u_x^p. Corollary C.3 implies that the convergence is uniform on \Sigma_p^\circ without breaking. Hence for i sufficiently large, u_i is in a C^0-neighborhood of u_x^p. However, as the Hamiltonian connection is flat, the energy E(u_i) is topological. Hence E(u_i) = 0 which is a contradiction.

The proof can be carried out verbatim as that of Lemma B.3.

The proof can be carried out verbatim as that of Lemma B.4.

We prove that the two-way compositions of {\it PSS}_{\mathcal F}^{\rm loc} and {\it SSP}_{\mathcal F}^{\rm loc} are both identities up to homotopy. The composition {\it SSP}_{\mathcal F}^{\rm loc} \circ {\it PSS}_{\mathcal F}^{\rm loc} is about gluing at periodic orbits of the nondegenerate H. The smallness of the energy together with the crossing energy lemma for the cylinder (Lemma C.5) implies that this composition is homotopic to the map on CM({\mathcal F}; {\mathfrak o}) induced by a Morse continuation map. For the composition {\it PSS}_{\mathcal F}^{\rm loc} \circ {\it SSP}_{\mathcal F}^{\rm loc}, the argument contains gluing at the Morse–Bott periodic orbits of H_0. This can also be done by the standard method. The triviality of {\mathfrak o} was established in [AS-torsion, Section 5.2].

The proof is similar to that of Proposition C.9. Let (f_{\mathcal F}, h_{\mathcal F}) be a Morse–Smale pair on {\mathcal F} which induces a Morse complex CM({\mathcal F}; {{\mathbb F}_p}). Let G be a C^\infty-close nondegenerate perturbation of H such that G^{(p)} is also nondegenerate. Let \sigma be a Hamiltonian connection on {\mathbb R}\times S^1 interpolating from H to G, which induces a Hamiltonian connection \sigma^{(p)} interpolating from H^{(p)} to G^{(p)}. Then for each y \in {\rm crit}(f_{\mathcal F})\subset {\mathcal F} and \overline{x} \in \tilde {\mathcal O}(G^{(p)}), one can consider a similar moduli space as the one used in the proof of Proposition C.9, with \sigma replaced by \sigma^{(p)}.

Moreover, to count solutions equivariantly, couple the equation by parametrized gradient flow lines of g in S^\infty. Then for w_-, w_+ \in {\rm crit}(g) and y \in {\rm crit}(f_{\mathcal F}), \overline{x}\in \tilde {\mathcal O}(G^{(p)}, \overline{\mathcal F}{}^{(p)}), one has a moduli space

of pairs. As \sigma^{(p)} has small curvature, the usual energy identity together with Lemma C.5 implies that the moduli space can be compactified by only adding local broken configurations without sphere bubbles. Still, one does not have automatic {\mathbb Z}_{\geq 0} \times {\mathbb{Z}/p}-equivariant transversality. One could, however, achieve transversality by virtual perturbations which respect the symmetry. As there are no sphere bubbles, one can use single-valued perturbations and obtain integral counts. Then one obtains a {\mathbb{Z}/p}-equivariant cochain map

Its restriction to the {\mathbb{Z}/p}-invariant part is

and hence induces the PSS map

The standard method proves that this map is {\mathcal{R}_p}-linear.

One can also define a similar SSP map

To prove that the PSS and SSP maps have composition in either direction being invertible, one considers the two different gluings. As it is true for the non-equivariant versions, the equivariant version only differs by terms with positive (t, \theta)-degrees. Therefore, the claimed isomorphism can be induced by {\it PSS}_{p, {\mathcal F}}^{\rm loc}.

The proof is based on the crossing energy estimate and gluing (at nondegenerate Hamiltonian orbits). Let G be a nondegenerate perturbation of H. On the cochain level, one needs to verify the commutativity of

Here the unlabelled arrow, which is supposed to induce the classical Steenrod operation on {\mathcal F}, has not been explicitly defined yet.

One uses a typical TQFT+gluing argument to prove the commutativity. The triple composition

can be identified, via gluings at p negative ends resp. one positive end and at (nondegenerate) orbits of G resp. G^{(p)} as well as critical points of g, to the map defined via the count of configurations where one has p incoming gradient rays and one outgoing gradient ray on {\mathcal F}, between which is a solution to the Floer equation on \Sigma being {\mathbb C} removing p-th roots of unity with flat Hamiltonian connection equal to the pullback of H dt from {\mathbb R}\times S^1. The crossing energy estimate can guarantee the gluing happens “locally” and the Floer equation on \Sigma must be a constant one. Therefore, the configurations one finally obtains are those Morse gradient trees responsible for the classical Steenrod operation on {\mathcal F}. The original commutative diagram is then obtained by 1) restricting (Proof) to the {\mathbb{Z}/p}-invariant part, 2) reducing to cohomology, and 3) composing with the quasi-Frobenius maps.

Choose a nondegenerate Hamiltonian G satisfying the assumption of Lemma C.2. Then the chain complex CF(G; \Lambda_{\mathbb K}^\Gamma) can be defined with a differential d_G, which has a decomposition

into “local” part d_G^{\rm loc} and the big-energy part d_G^{\rm big}. Then the homological perturbation argument provides a construction of the desired complex C(H). The first two properties both follow from the construction directly. For the third one, the proof is exactly the same as the corresponding one of Proposition 3.1.

The construction of the differential d_{{\mathbb{Z}/p}} is very similar to the isolated case given in the proof of Proposition 3.3. The component-wise decomposition of C_{\mathbb{Z}/p}(H^{(p)}) follows from Proposition C.11. Lemma B.6 allows us to construct the chain homotopy equivalence between the local equivariant Floer complex and the local equivariant Floer cohomology, where the coefficient ring is {\mathcal{K}_{0, p}}. Then the first half of this proposition, i.e., the canonical isomorphism H(C_{\mathbb{Z}/p}(H^{(p)})) \cong HF_{\mathbb{Z}/p}(M) and the claim that d_{{\mathbb{Z}/p}} strictly increases the action by a positive constant, follows from the same argument as the isolated case.

The proof is very similar to that of Corollary 3.5. One can change the coefficient to the universal Novikov rings/fields. Then the equivariant pair-of-pants product induces an isomorphism of \Lambda_{0, {\mathcal{K}_p}}^{\rm univ}-modules

where the bar-lengths under consideration label the torsion parts.

We follow the argument of [AS-torsion, Section 6.3]. First, because \phi has order p', there holds

Now we make use of the quantitative Smith inequality (Theorem 2.20), which is established for nondegenerate Hamiltonians. As the barcodes vary continuously with the Hamiltonian diffeomorphism, together with (2) of Proposition C.13, for the degenerate Hamiltonian diffeomorphism \phi^i, one has

As k is arbitrary, we see that \beta_{\mathrm{tot}}(\phi^i) = 0 for all i \in \{1, \ldots, p'\}. In particular, the cochain complex C(H^{(p^k)}) does not admit any nontrivial differential.

For the complex C_{{\mathbb{Z}/p}}(H^{(p^k)}), suppose on the contrary it has a nontrivial differential d_{{\mathbb{Z}/p}}. Then the corresponding Tate complex \widehat{C}_{{\mathbb{Z}/p}}(H^{(p^k)}) also admits a nontrivial differential. As shown in Proposition C.15, d_{{\mathbb{Z}/p}} strictly increases the filtration. Then the Tate bar-length spectrum has a nonzero shortest bar-length \widehat{\beta}_1(H^{(p^k)}). It follows from Corollary C.16 that \widehat{\beta}_1(H^{(p^k)}) = p \beta_1( H^{(p^{k-1})}) \neq 0, which contradicts the first assertion we just proved.