3
$\begingroup$

A few weeks ago, Bhatt and Scholze uploaded a preprint of their paper 'Prisms and Prismatic Cohomology' to arxiv.

In Theorem 6.3 they state their Hodge-Tate comparison. Recently, I started reading on Hodge decomposition and Hodge-Tate composition. However, all Hodge-Tate comparison theorems I encountered so far (see for example Theorem 1.3 here) are of a different form and neither could I relate the two to each other nor do I see a connection to Hodge-Tate composition. I would be happy if someone could give a (short) explanation for the name or point out references which I could consult.

$\endgroup$
10
$\begingroup$

tl;dr The Hodge-Tate comparison isomorphism relates the reduction mod $I$ of prismatic cohomology to something similar to the "Hodge-Tate cohomology" $\bigoplus_{i+j = k} H^i(X, \Omega^j_{X/K})$. Together with the étale comparison theorem relating prismatic cohomology away from $V(I)$ to étale cohomology, this gives the integral $p$-adic Hodge theory version of the usual Hodge-Tate comparison theorem.

Let $X$ be a smooth and proper variety over a $p$-adic field $K$ (i.e. a finite extension of $\mathbf{Q}_p$). Let $C = \mathbf{C}_p$ be the $p$-adic completion of an algebraic closure of $\overline{K}$.

The usual Hodge-Tate comparison theorem relates the $p$-adic étale cohomology of $X$ to its "Hodge-Tate cohomology".

On one side, we consider the étale cohomology $\mathrm{H}^k_{\mathrm{et}}(X_{\overline{K}}, \underline{\mathbf{Z}_p})$: this is a $\mathbf{Z}_p$-module with an action of $G_K$, the absolute Galois group of $K$.

On the other side, we consider the "Hodge-Tate cohomology" of $X$. This is: $$ \mathrm{H}^k_{\mathrm{HT}}(X/K) = \bigoplus_{i + j = k} \mathrm{H}^i(X, \Omega^j_{X/K}) $$ It is a graded $K$-vector space with $\mathrm{gr}^j \ \mathrm{H}^k_{\mathrm{HT}}(X/K) = \mathrm{H}^i(X, \Omega^j_{X/K})$.

Now, there is a Hodge-Tate spectral sequence, compatible with the Galois actions on both sides. $$ \mathrm{E}_2^{i,j} = \mathrm{H}^i(X, \Omega^j_{X/K}) \otimes_K C(-j) \Rightarrow \mathrm{H}^k_{\mathrm{et}}(X_{\overline{K}}, \underline{\mathbf{Z}_p}) \otimes_{\mathbf{Z}_p} C $$

This spectral sequence is defined more generally when $X$ is any rigid-analytic space over $C$, and is known to always degenerate at $\mathrm{E}_2$ (Thm. 1.7 in "Integral $p$-adic Hodge theory" by Bhatt, Morrow, and Scholze). In the case that $X$ is actually defined over $K$, there is even a canonical splitting, giving a canonical $G_K$-equivariant isomorphism $$ \mathrm{H}^k_{\mathrm{et}}(X_{\overline{K}}, \underline{\mathbf{Z}_p}) \otimes_{\mathbf{Z}_p} C \simeq \bigoplus_j \mathrm{H}^i(X, \Omega^j_{X/K}) \otimes_K C(-j) $$

Now, prismatic cohomology is about integral $p$-adic Hodge theory, meaning that we want to get comparison theorems between cohomology theories without having to invert $p$. More precisely, assume that $X$ is the generic fiber of a smooth proper (formal) scheme $\mathscr{X}$ over $\mathscr{O}_K$. Then we can ask for a relationship between the $\mathbf{Z}_p$-module $\mathrm{H}^k_{\mathrm{et}}(X_{\overline{K}}, \underline{\mathbf{Z}_p})$ and the $\mathscr{O}_K$-module $\mathrm{H}^k_\mathrm{HT}(\mathscr{X}/\mathscr{O}_K) := \bigoplus_{i+j =k} \mathrm{H}^i(\mathscr{X}, \Omega^j_{\mathscr{X}/\mathscr{O}_K})$. Unfortunately, a comparison theorem of the above form does not hold in this context.

Instead, prismatic cohomology constructs a "universal" cohomology object which "interpolates" between étale cohomology and Hodge-Tate cohomology (as well as de Rham cohomology, crystalline cohomology, etc).

Assume for ease of notation that $K$ is unramified over $\mathbf{Q}_p$. Let $A$ be the ring $\mathscr{O}_K[[u]]$, equipped with a natural Frobenius endomorphism sending $u$ to $u^p$. There is an $\mathscr{O}_K$-linear surjection $A \rightarrow \mathscr{O}_K$ defined by $u \mapsto p$, with kernel $I = (u - p)$. Then $(A, I)$ is an example of a prism. Cohomology of the structure sheaf on the prismatic site of $\mathscr{X}$ relative to $A$ gives a complex $\mathrm{R} \Gamma_{\mathrm{prism}}(\mathscr{X}/A)$ in the derived category of $A$-modules.

Loosely, $\mathrm{R} \Gamma_{\mathrm{prism}}(\mathscr{X}/A)$ recovers Hodge-Tate cohomology on the closed set $\mathrm{Spec }\ \mathscr{O}_K = V(u - p) \subseteq \mathrm{Spec }\ A$, and recovers $p$-adic étale cohomology of $\mathscr{X}$ on the open set $\mathrm{Spec }\ A[\frac{1}{u-p}] \subseteq \mathrm{Spec }\ A$. Bhatt and Scholze refer to the former as the Hodge-Tate comparison theorem, and the latter as the étale comparison theorem. Putting these two statements together describes the integral version of the relationship between Hodge-Tate and étale cohomologies.

More precisely, there is an object $\Delta_{\mathscr{X}/A}$ in the derived category of Zariski (or étale) sheaves of $A$-modules on $\mathscr{X}$ such that $\mathrm{R} \Gamma(\Delta_{\mathscr{X}/A}) = \mathrm{R} \Gamma_{\mathrm{prism}}(\mathscr{X}/A)$. We consider its (derived) restriction to $\mathrm{Spec} \ A/I$, given by $\overline{\Delta}_{\mathscr{X}/A} := \Delta_{\mathscr{X}/A} \otimes^L_A (A/I)$, so we have $\mathrm{R} \Gamma_{\mathrm{prism}}(\mathscr{X}/A) \otimes^L_A (A/I) \simeq \mathrm{R} \Gamma(\overline{\Delta}_{\mathscr{X}/A})$.

Now, the Hodge-Tate comparison theorem as stated in the Bhatt-Scholze paper gives a canonical isomorphism of $\mathscr{O}_K$-modules. Here, if $M$ is an $\mathscr{O}_K$-module, $M\{j\} = M \otimes_{\mathscr{O}_K} (I/I^2)^{\otimes j}$. $$ \mathrm{H}^j(\overline{\Delta}_{\mathscr{X}/A}) \simeq \Omega^j_{\mathscr{X}/\mathscr{O}_K}\{-j\} $$ This gives us a hypercohomology spectral sequence $$ \mathrm{E}_2^{i,j} = \mathrm{H}^i(\mathscr{X}, \Omega^j_{\mathscr{X}/\mathscr{O}_K})\{-j\} \Rightarrow \mathbf{H}^{i+j}(\mathscr{X}, \overline{\Delta}_{\mathscr{X}/A}) $$ The right side should be thought of as a "derived correction" of $\mathrm{H}^{i+j}_{\mathrm{prism}}(\mathscr{X}/A) \otimes_A (A/I)$.

To relate this to étale cohomology, we must pass to a bigger prism. Let $A_{\mathrm{inf}} = \mathrm{W}(\mathscr{O}_{C^\flat})$. This has a natural Frobenius automorphism $\varphi$ lifting the one on $\mathscr{O}_{C^\flat}$ and a surjection $\widetilde{\theta} \colon A_{\mathrm{inf}} \rightarrow \mathscr{O}_C$ with kernel $J = (d)$ for a certain element $d \in A_{\mathrm{inf}}$. ($\widetilde{\theta} = \varphi^{-1} \circ \theta$, where $\theta$ is the usual map as defined by Fontaine). There is a map of prisms $(A, I) \rightarrow (A_{\mathrm{inf}}, J)$ which lifts the inclusion $\mathscr{O}_K \hookrightarrow \mathscr{O}_C$.

Then we have $\mathrm{R} \Gamma_{\mathrm{prism}}(\mathscr{X}_{\mathscr{O}_C}/A_{\mathrm{inf}}) \simeq \mathrm{R} \Gamma_{\mathrm{prism}}(\mathscr{X}/A) \otimes_A^L A_{\mathrm{inf}}$. In particular, the above gives us a spectral sequence $$ \mathrm{E}_2^{i,j} = \mathrm{H}^i(\mathscr{X}, \Omega^j_{\mathscr{X}/\mathscr{O}_K}) \otimes_{\mathscr{O}_K} \mathscr{O}_C\{-j\} \Rightarrow \mathbf{H}^{i+j}(\mathscr{X}_{\mathscr{O}_C}, \overline{\Delta}_{\mathscr{X}_{\mathscr{O}_C}/A_\mathrm{inf}}) $$ where the right side should be thought of as a "derived correction" of $\mathrm{H}^{i+j}_{\mathrm{prism}}(\mathscr{X}_{\mathscr{O}_C}/A_{\mathrm{inf}}) \otimes_{A_{\mathrm{inf}}, \widetilde{\theta}} \mathscr{O}_C$

On the other hand, the étale comparison theorem gives an isomorphism, equivariant with respect to the natural $G_K$-actions on both sides: $$ \mathrm{R}\Gamma_{\mathrm{et}}(X_C, \underline{\mathbf{Z}_p}) \otimes_{\mathbf{Z}_p} A_{\mathrm{inf}}[1/d] \simeq \mathrm{R}\Gamma_{\mathrm{prism}}(\mathscr{X}_{\mathscr{O}_C}/A_{\mathrm{inf}})\otimes_{A_{\mathrm{inf}}}^L A_{\mathrm{inf}}[1/d] $$ Note that the right side is isomorphic to $\mathrm{R} \Gamma_{\mathrm{prism}}(\mathscr{X}/A) \otimes_A^L A_{\mathrm{inf}}[1/d]$.

Thus, the étale and Hodge-Tate comparison theorems together show that $\mathrm{R} \Gamma_{\mathrm{prism}}(\mathscr{X}/A)$ determines a deformation from $\mathrm{H}^{i+j}_{\mathrm{et}}(X, \underline{\mathbf{Z}_p})$ to an object related to the Hodge-Tate cohomology $\bigoplus_j \mathrm{H}^i(\mathscr{X}, \Omega^j_{\mathscr{X}/\mathscr{O}_K})\{-j\}$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.