# Describing the crystalline extension of $\mathbb{Q}_p$ by $\mathbb{Q}_p$

Let $K$ be a finite extension of $\mathbb{Q_p}$. The group $\ker H^1(G_K, \mathbb{Q}_p) \rightarrow H^1(G_K, B_{crys})$ is one-dimensional, which tells us that among all extensions of Galois modules

$$0 \rightarrow \mathbb{Q}_p \rightarrow E \rightarrow \mathbb{Q}_p \rightarrow 0,$$

which are classified by $H^1(G_K, \mathbb{Q}_p) \cong Hom_{cts}(K^*, \mathbb{Q}_p)$, there is a one-dimensional subspace corresponding to crystalline representations.

My question is basically: What homomorphism $K^* \rightarrow \mathbb{Q}_p$ does this crystalline cocycle correspond to?

Actually, I know the answer. It was explained to me that a theorem of Sen asserts that a crystalline representation with only $0$ as Hodge-Tate weight must have finite image of inertia, which in this case must be trivial, i.e. unramified. So the homomorphism in question is (a scalar multiple of) $x \mapsto val_p(x)$.

I would like to come to a more elementary and concrete understanding of this puzzle. The fact that this is a coboundary in $H^1(G_K, B_{crys})$ means that I can find $b \in B_{crys}$ such that inertia acts trivially on $b$ and Frobenius translates $b$ by $1$. Is it possible to describe this period explicitly in terms of the construction of $B_{crys}$?

As someone who has never really understood the definition of $B_{crys}$, it would be even more satisfying to me to have an explicit description in terms of elements which I believe should belong in $B_{crys}$ if my intuitive definition of the latter is merely the ring of periods for cohomology of varieties with good reduction''. In other words, I would love to see this period $b$ expressed in terms of periods coming from cohomology of familiar varieties, such as the period for the cyclotomic character.

• Both $p$-divisible groups and smooth proper formal schemes over the dvr give rise to crystalline representations (and similarly in the deRham case with smooth proper rigid-analytic spaces), so one gets "crystalline periods" for cohomology far beyond algebraic varieties. Your setup "is" an etale $p$-divisible group over the residue field $k$ (uniquely lifting over $O_K$), so its $W(\overline{k})[1/p]$-admissibility is the slope-0 case of the Dieudonne-Manin classification, illuminating the comment in Sam Derbyshire's answer concerning the need use an algebraic closure of the residue field. – nfdc23 Jan 12 '16 at 15:55

The representation $E$ in this case is not only crystalline, it is in fact unramified. This means we don't need much of the complicated machinery of $p$-adic Hodge theory to get a handle on the periods of $E$.

Whereas for general crystalline representations we need to use $\mathbf{B}_\text{cris}$ to find periods, for potentially unramified representations, we can work with $\mathbb{C}_p$ (potentially unramified representations are $\mathbb{C}_p$-admissible). In this case, we don't even need $\mathbb{C}_p$, as the representation is unramified (not just potentially so), and it suffices to work with $(K_0^\text{nr})^\vee$ instead (where $K_0$ is the maximal unramified intermediate extension $K / K_0 / \mathbb{Q}_p$).

To explicitly see that $E$ is $(K_0^\text{nr})^\vee$-admissible, we can start by taking $a$ to be a solution to the Artin–Schreier equation $x^q-x-1=0$ in $\mathcal{O}_K/\mathfrak{m} \cong \mathbb{F}_q$. We then have $\mathrm{Frob}([a]) = [a^q] = [a+1]$, where $[-]$ indicates Teichmüller lifts. This is nearly what we want, save for the fact that the Teichmüller lift is not additive. So you have to remedy that by hand by using the Witt addition polynomials; the upshot is that you'll obtain some element $b \in \mathrm{W}(\overline{\mathbb{F}_q})$ with $\varphi(b) = b+1$ as desired. It is necessary to go all the way up to $\mathrm{W}(\overline{\mathbb{F}_q})$; at any finite level $\mathrm{W}(\mathbb{F}_{q^n})$ there will always be the above issue of non-additivity. Indeed, only potentially trivial representations are detected at finite levels, and we need to pass to the completion $(K_0^\text{nr})^\vee$ to allow a Frobenius of infinite order.
At any rate, this means then that $\{1,b\}$ is a $(K_0^\text{nr})^\vee$-basis of $E \otimes_{\mathbb{Q}_p} (K_0^\text{nr})^\vee$. Using the inclusion $(\mathcal{O}_{K_0^\text{nr}})^\vee \subset \mathbf{A}_\text{cris}$ we can consider $b$ to be an element of $\mathbf{B}_\text{cris}$. It is a crystalline period of $E$; together with $1 \in \mathbf{B}_\text{cris}$ it provides a $\textbf{B}_\text{cris}$-basis of $E \otimes_{\mathbb{Q}_p} \mathbf{B}_\text{cris}$.

As for the motivic question, I think no such variety is expected to exist. We are in the crystalline situation, so we'd like to be able to find $E$ inside the cohomology of some (smooth, separated, finite type) scheme over $\mathcal{O}_K/\mathfrak{m}$. By analogy with the $\ell$-adic and complex situations, I believe such extensions as $E$ shouldn't occur there, because of weight filtration considerations (e.g. there are no non-split extensions of mixed Hodge structures of $\mathbb{Q}$ by $\mathbb{Q}$).

• Perfect! Exactly the sort of answer I was hoping for. – user84144 Jan 12 '16 at 17:01
• Two minor nitpicks: $\mathcal{O}_{\mathbb{C}_p}$ is not a subring of $\mathbf{A}_{\mathrm{cris}}$ -- if this were so, then every finite-image representation would be crystalline, which is not true. Moreover, $W(\overline{\mathbf{F}}_q)$ is bigger than $\mathcal{O}_{K^{\mathrm{nr}}}$ (the former is the completion of the latter). – David Loeffler Jan 12 '16 at 19:53
• Hang on. A more serious point: in the last paragraph it looks like you're trying to consider the p-adic etale cohomology of a mod p variety, which is generally rather pathological, and won't compare well to the generic fibre. Moreover, it definitely isn't true that the Galois action on the cohomology of the generic fibre of a smooth $\mathbf{Z}_p$-scheme is semisimple; you need the scheme also to be proper. For non-proper schemes there are lots of interesting extensions -- this is the whole point of the theory of mixed motives. – David Loeffler Jan 13 '16 at 12:38
• Sure, I was deliberately a bit vague in the last paragraph, and I should've been more specific. I didn't mean to consider the p-adic étale cohomology, but maybe instead a version of crystalline cohomology obtained through some simplicial resolution by smooth proper schemes. My impression is that there still won't be extensions of this form (although of course as you say other nontrivial extensions are possible); for instance there are no nontrivial extensions in the category of mixed Hodge structures of the trivial Hodge structure by itself. – Sam Derbyshire Jan 13 '16 at 13:11
• Yes, but in that case there are different weights; the weight filtration yields an extension of something of weight $1$ by something of weight $0$, in accordance with the sequence $0 \to W_0 \to W_1 \to W_1/W_0 \to 0$. I am arguing that such an extension shouldn't occur in geometry for the OP's situation because everything is in weight $0$. – Sam Derbyshire Jan 13 '16 at 14:11

For simplicity, let's take $K = \mathbf{Q}_p$.

One of the few things about $\mathbf{B}_{\mathrm{cris}}$ that one can straightforwardly prove directly from its definition is that it contains $\widehat{\mathbf{Q}}_p^{\mathrm{nr}}$, the completion of the maximal unramified extension of $\mathbf{Q}_p$ (equvalently, the field of fractions of the Witt vectors of $\overline{\mathbf{F}}_p$).

It's not too hard to check, using Hilbert 90 for $\overline{\mathbf{F}}_p$ and induction on $n$, that any unramified mod $p^n$ representation of $G_{\mathbf{Q}_p}$ is $W_n(\overline{\mathbf{F}}_p)$-admissible, and by passage to the limit ("devissage", as the French call it) one deduces that any unramified $\mathbf{Q}_p$-linear representation of $G_{\mathbf{Q}_p}$ is $\widehat{\mathbf{Q}}_p^{\mathrm{nr}}$-admissible.

From local class field theory, one knows that there's a unique unramfied extension of $\mathbf{Q}_p$ by $\mathbf{Q}_p$ (corresponding to the unramfied $\mathbf{Z}_p$-extension of $\mathbf{Q}_p$). So this representation is crystalline, and its periods lie in this much less scary subring $\widehat{\mathbf{Q}}_p^{\mathrm{nr}} \subset \mathbf{B}_{\mathrm{cris}}$.

I'm not so sure whether there's a nice way of exhibiting this extension in the etale cohomology of a variety. The variety would have to be either non-smooth or non-proper (since the cohomology of smooth proper varieties is conjectured to be semi-simple). Maybe someone with more of a geometric background than me can answer that.