Local Tate duality for syntomic cohomology in positive characteristic

The element fv-p is in the centre of \mathcal{A}^{(m)}, and the element \partial^{\langle p^m \rangle_{(m-1)}}-c_mv\partial^{[p^m]} commutes with f,v and \partial^{[p^m]}, so to prove that I is a centralising ideal it is enough to show that if P\in \widehat{\mathcal{D}}^{(m-1)}_C, then [\partial^{\langle p^m \rangle_{(m-1)}}-c_mv\partial^{[p^m]},P]\in I. But we can calculate

That \widehat{\mathcal{D}}^{(m-1,m),\star}_C\cong \mathcal{A}^{(m)}/I can be shown entirely similarly to the case m=1 handled in [Dod22, §3.1].

We reduce modulo p and replace M by its cohomology groups, which are necessarily finitely generated as C^\star-modules. Then f\colon M^i\rightarrow M^{i+1} is an isomorphism for i\gg 0, and so the map v\colon M^{i+1}\rightarrow M^i is zero for i\gg 0.

If we fix i_0, then for n large enough and x\in M^{i_0} we have

and so \left(\partial^{\langle p^m\rangle_{(m-1)}}\right)^n=0 on M^{i_0}. Since this is true for all i_0, we deduce that M is quasi-nilpotent.

If m\geq 0, this follows from suitably completing (?). If m=-1, this can be proved by working modulo p^n for all n, where we proceed more or less exactly as in [Ber00, Proposition 2.5.2], using Shiho’s calculations of the PD envelopes of level -1 from [Shi15, §2.1] as the key input. Of course, strictly speaking Shiho’s results do not apply in our setting, since k is not smooth over its residue field {\mathbb F}. However, the fact that any uniformiser t is a p-basis for k means that the relevant calculations of divided power envelopes still work in our setting.

The crucial difference we encounter when running through the proof of [Ber00, Proposition 2.5.2] is that we must work throughout with the full PD envelopes \mathcal{P}_{C/p^n,(0)} and \mathcal{P}_{C'/p^n,(-1)}, and we never pass to the quotients by the natural PD filtration. This gives the required answer because

and similarly at level -1. Note that we do not get the analogous isomorphism

for finite order differential operators, because the analogue of [Ber00, Lemme 2.3.4] does not hold at level -1. This means that the isomorphism

does not descend to an isomorphism of pro-objects

It suffices to prove the claim for the categories of objects without Frobenius structure, and in fact it suffices to prove the analogous claim for p-torsion gauges, that is, complexes of graded left modules over \mathcal{D}^{(m-1,m),\star}_k. Indeed, a straightforward p-completeness argument applied to {\bf R}{\rm Hom}’s then shows full faithfulness for p-complete gauges. For essential surjectivity, we see that the essential image of \sigma^* contains all p-torsion objects, hence all p^n-torsion objects by the already proved full faithfulness, and hence all p-complete objects.

We therefore wish to prove that the functor

is an equivalence of categories if m\geq 1, and induces an equivalence of categories

if m=0. Since both functors are clearly exact, it is enough to prove the corresponding result for static objects.

By Theorems 2.1 and 2.4 we know that, on each graded piece, the functor M^i\mapsto \sigma^*M^i is an equivalence (either for all objects if m\geq 1, or for quasi-nilpotent objects if m=0), so it suffices to show that the maps \partial^{[p^{m}]}\colon M^i\rightarrow M^{i+1} can be uniquely recovered from \partial^{[p^{m+1}]}\colon \sigma^*M^i\rightarrow\sigma^*M^{i+1}. If m\geq 1, this can be done in the usual way [Ber00, Proposition 2.2.4], since we have the explicit reconstruction

of each M^i from \sigma^*M^{i}, and the map \partial^{[p^{m}]}\colon M^i\rightarrow M^{i+1} is then simply the restriction of the map \partial^{[p^{m+1}]}\colon \sigma^*M^i\rightarrow\sigma^*M^{i+1} to the subspace M^i\subset \sigma^*M^i.

If m=0, then we use [GLSQ10, Theorem 5.8] instead to recover each M^i from \sigma^*M^{i}. One can then check by an explicit calculation that \partial^{[p]}\colon \sigma^*M^{i}\rightarrow \sigma^*M^{i+1} restricts to a map M^i\rightarrow M^{i+1}, where it is given by (1+g\theta)\partial for some g\in \widehat{\mathcal{D}}^{(-1)}_C. Since M^{i+1} is quasi-nilpotent, the map (1+g\theta) is invertible on M^{i+1}, thus \partial\colon M^i\rightarrow M^{i+1} can be recovered from \partial^{[p]}.

Since M\otimes_{C^\star}^{\mathbf{L}} commutes with cofibre sequences, we need to show that if \widehat{N}=0, then \left(M\otimes_{C^\star}^{\bf L}N\right)^\wedge=0. But vanishing of a derived p-complete complex can be checked on its mod p reduction, so we need to show that if N/^{\bf L}p=0, then (M\otimes^{\bf L}_{C^\star} N)/^{\bf L}p=0, which is clear.

Since p-adic completion creates colimits in {\bf D}(\widehat{\mathcal{D}}_C^{(m-1,m),\star}), and is (by construction) symmetric monoidal, it is enough to show that {\bf D}(\mathcal{B}^{(m)})^\otimes is presentably symmetric monoidal. Since the underlying category {\bf D}(\mathcal{B}^{(m)}) is stable and presentable, it is enough by [Lur09, Proposition 4.4.2.7] to show that for any given object M\in {\bf D}(\mathcal{B}^{(m)}), the functor M\otimes_{C^\star}^{\bf L} commutes with cofibre sequences and direct sums, both of which are clear.

As usual, it suffices to prove compatibility on the underlying abstract modules. In this case, using the definition of \widehat{\mathcal{D}}^{(m-1,m),i}_{C_2} as a submodule of \widehat{\mathcal{D}}^{(m)}_{C_2}, the proposition follows from applying [Ber00, Proposition 2.1.5] instead to \widehat{\mathcal{D}}^{(m)}_{C_2} as a module over itself, and then noting that Berthelot’s isomorphism \tau_{\beta,\beta'} is compatible with restriction of scalars along {\mathcal{D}}^{(m-1)}\rightarrow \mathcal{D}^{(m)}.

If G is a p-divisible group, then \mathbb{M}(G) is torsion-free, and hence gauge-like. If H is a finite flat group scheme, then by [BBM82, Théorème 3.1.1] there exists an injective map H\hookrightarrow G with G a p-divisible group. This induces a surjective map \mathbb{M}(G/H)\rightarrow \mathbb{M}(H), where G/H is p-divisible, and the fact that \mathbb{M}(H) is gauge-like follows from the fact that \mathbb{M}(G/H) is.

We immediately reduce to the case N=k and {\mathbb F}={\mathbb F}_p. Choosing a uniformiser t, the given pairing can be identified with the map

We therefore reduce to showing that the pairing

induces equivalences

The first of these follows more or less immediately from the definitions, using the fact that countable products of (light) condensed abelian groups are exact.

For the second, it follows from [RC26, proof of Theorem 3.2.4(9)] that the right hand side is concentrated in degree 0, and a straightforward calculation shows that the given map is an isomorphism in this degree.

Since W is p-adically complete, it is enough to show that \psi=\xi as maps C/p^n\rightarrow W/p^n, for all n. Next, writing C/p^n = \underset{m}{\mathrm{colim}}\, t^{-m} W/p^n\llbracket t \rrbracket it is enough to show that \psi=\xi as maps t^{-m} W/p^n\llbracket t \rrbracket\rightarrow W/p^n, for all m and n. But now t^{-m} W/p^n\llbracket t \rrbracket and W/p^n are both light profinite sets, and so it suffices to observe that W/p^n is Hausdorff and that t^{-m} W/p^n[t] \subset t^{-m} W/p^n\llbracket t \rrbracket is dense.

By Lemma 7.3 it is enough to show that if s\in C(*) is another such choice of uniformiser lift, then

If i\neq 0 then this is easy: we simply calculate

and then observe that t\partial(s^i) clearly has no constant term, and that W is torsion-free.

If i=0, we write s=tu_0u_1 where u_0\in W\llbracket t \rrbracket^\times and u_1\in 1+pC. Then

converges in C, and so we find that

and the second and third terms visibly have zero residue in t.

That \mathrm{Tr}_{W/{\mathbb Z}_p}\circ \mathrm{Res}\circ {\rm d}=0 is immediate, since we already have \mathrm{Res}\circ {\rm d}=0 by inspection. We next show that {\rm Res} intertwines \frac{\sigma}{p} with the usual Witt vector Frobenius \sigma on W. Indeed, by Lemma 7.3, it is enough to show that {\rm Res}(\frac{\sigma}{p}(t^i\,{\rm d}\!\log t))=0 for i\neq 0, and that {\rm Res}(\frac{\sigma}{p}({\rm d}\!\log t))=1. For the former, we calculate

which shows that ip\cdot {\rm Res}(\frac{\sigma}{p}(t^i\,{\rm d}\!\log t))=0. Thus {\rm Res}(\frac{\sigma}{p}(t^i\,{\rm d}\!\log t))=0 since W is torsion-free. For the latter, we write \sigma(t)=t^pu for some u\in 1+pC. As before,

converges in C, and so

which clearly has residue 1. To complete the proof, we now observe that, for any \omega\in \widehat{\Omega}^1_C,

and {\rm Tr}_{W/{\mathbb Z}_p} vanishes on the image of 1-\sigma by Galois theory for W/{\mathbb Z}_p.

This reduces via a p-completeness argument to Proposition 7.2.

Since de  Rham cohomology is a special case of Nygaard cohomology, by Corollary 5.5, and syntomic cohomology sits in a fibre sequence with Nygaard and de  Rham cohomology, it suffices to treat the case of Nygaard cohomology. We can then use Proposition 5.4 to compute the Nygaard cohomology of M as the double complex

We now simply note that M^0 and M^1 are perfect complexes over C, and hence reflexive by Proposition 7.7.

We need to show that for any P\in \widehat{\mathcal{D}}^{(-1,0),\star}_C, the map

commutes, for which it is sufficient to show that

for any \omega \in \widehat{\Omega}^1_C \otimes_C M^\lozenge, x\in M. We already know C-linearity, so it suffices to treat the three cases P=f,v,\partial. The first two are straightforward.

For the last, if we have \omega\in \widehat{\Omega}^1_C \otimes_C M^{\lozenge,i}=\widehat{\Omega}^1_C \otimes_C M^{-i,\vee}, and x\in M^{-(i+1)}, then the Leibniz rule for right \mathcal{D}-modules gives

and so

as required.

We may assume that M is static, in which case each graded piece is a finitely generated C-module. Note that by Propositions 5.4 and 7.9, the {\mathbb Z}_p-linear dual of

is computed by the totalisation of the complex

with M^{\lozenge,-1} sitting in horizontal cohomological degree -1, and the implicit pairing being given by the trace map. Again, by Proposition 5.4, we therefore see that the trace map induces an equivalence

and so replacing M by M^\lozenge(1)[1] and using (M^\lozenge)^\lozenge=M gives the result.

We may replace M^\circ first by its reduction modulo p, and then by its cohomology groups. It is then straightforward to reduce to the case {\mathbb F}={\mathbb F}_p, and we can treat separately the case where M^\circ is a finite free k^\circ\cong {\mathbb F}_p\llbracket t \rrbracket-module, and where M^\circ is t^\infty-torsion. We can reduce the former to the case M^\circ={\mathbb F}_p\llbracket t \rrbracket, and the latter to the case M^\circ=k^\circ/(t)={\mathbb F}_p.

If M^\circ=k^\circ, then the essential claim is that, under the duality pairing

the orthogonal complement of {\mathbb F}_p\llbracket t \rrbracket\subset {\mathbb F}_p(\!(t)\!) is {\mathbb F}_p\llbracket t \rrbracket\, {{\rm d} t}\subset {\mathbb F}_p(\!(t)\!)\, {\rm d}\!\log t, which is a routine calculation. If M^\circ={\mathbb F}_p, then the claim is that

which is, again, a routine calculation.

We immediately reduce to the case where M^\circ is static. Then

and so by Lemma 9.4 the cofibre of {\bf R}\Gamma_{\EuScript{N}}(k^\circ,M^\circ)\rightarrow {\bf R}\Gamma_{\EuScript{N}}(k,M) is given by the totalisation of the complex

with terms sitting in horizontal cohomological degrees 0 and 1. But we can identify this with

where the complex inside the brackets sits in horizontal cohomological degrees -1 and 0, and is hence nothing other than {\bf R}\Gamma_{\EuScript{N}}(k^\circ,M^{\circ,\lozenge}(1)[1]).