11
$\begingroup$

If $A/k$ is a principally polarised ordinary abelian variety ($k$ a perfect field of characteristic $p$, we may assume it is finite for simplicity), we have a canonical lift $\hat{A}/W(k)$. Now if I take a deformation $A_{\epsilon}/k[\epsilon]$, does there still exist a canonical lift of this deformation to a deformation of (the generic fiber of) $\hat{A}$?

By the Kodaira-Spencer mapping deformations are essentially encoded by differentials of $A$, so the question boils down to whether differentials on $A$ lift canonically to differentials on $\hat{A}$. By Katz, Serre-Tate local moduli, Section 3, differentials on any lift $\tilde{A}$ correspond to points in $T_p(A^\vee)(k)$, and his main theorem 3.7.1 describes the compatibility of this identification with the Kodaira-Spencer map. Is there a way to use this to lift differentials canonically on the canonical lift?

$\endgroup$

1 Answer 1

15
$\begingroup$

No. The picture over a general base is this: let $A_0\to S_0$ be an ordinary abelian variety over a characteristic $p$ scheme $S_0$, and let $S_n$ ($n\geq 0$) be compatible flat liftings of $S_0$ over $\mathbf{Z}/p^{n+1}$. Let $F\colon S_0\to S_0$ be the Frobenius. Then the pull-back $(F^n)^* A_0$ has a canonical lifting $A_{n,\rm can}$ to $S_n$.

Thus if $S_0$ is perfect (in which case automatically $S_n = W_{n+1}(S_0)$), then we can undo the Frobenius twist $(F^n)^*$ above, and we obtain a compatible system of lifts $A_n = ((F^{-n})^* A_0)_{n, \rm can}$ to $S_n$, i.e. a formal abelian variety over the formal scheme $S_\infty = \varinjlim_n\, S_n$.

If you prefer, you can also say that $A_0$ has a canonical lifting $\tilde A_n$ to $W_{n+1}(S_0)$ for all $n$ (this is an equivalent point of view taken in a paper by Borger and Guerney). If $S_n$ is as above, there is a canonical map $\phi_n \colon S_n\to W_{n+1}(S_0)$ such that $\phi_n\circ i = j\circ F^n$ where $i\colon S_0\hookrightarrow S_n$ and $j\colon S_0\to W_{n+1}(S_0)$ are the canonical closed immersions. Then $\phi_n^* \tilde A_n \simeq A_{n, \rm can}$ as abelian schemes over $S_n$.

Unfortunately, in your case $S_0 = \operatorname{Spec} k[\varepsilon]/(\varepsilon^2)$ the Frobenius factors through $\operatorname{Spec} k$, so the above gives you nothing. But it does give you something e.g. for $S_0 = \operatorname{Spec} k[\![t]\!]$.

EDIT. Coming back to the heart of your question. Let $D=\operatorname{Def}_{A_0/W(k)}$ be the formal deformation space of $A_0$ over $W(k)$. The essence of Serre-Tate theory is that $D$ carries a canonical lifting of Frobenius $F\colon D\to D$, which in particularly chosen multiplicative coordinates $q_{ij}$ ($1\leq i,j\leq g$), i.e. $D\simeq \operatorname{Spf} W(k)[\![q_{ij}-1]\!]$, has the particularly simple form $F^*(q_{ij})=q_{ij}^p$. (These coordinates may only exist over a finite separable extension of $k$, so let us assume that $k$ is algebraically closed for simplicity.) The "canonical coordinates" $q_{ij}$ can be recovered back from $F^*$ once we fix a basis of the $p$-adic Tate module of $A_0$, and then the Hodge $F$-crystal of the universal formal abelian variety over $D$ is explicitly described as in Katz's article or the accompanying article by Deligne and Illusie (with an appendix by Katz) "Cristaux ordinaires et coordonnees canoniques".

So how does the Frobenius $F\colon D\to D$ give us canonical liftings? It is easy to check (see section 1 in Katz "Travaux de Dwork") that for every lifting of Frobenius $F$ on $R\simeq W(k)[\![x_1, \ldots, x_n]\!]$ there exists a unique homomorphism $\sigma_F\colon R\to W(k)$ over $W(k)$ which commutes with the Frobenius lifts, i.e. $\sigma_F F=F_{W(k)}\sigma_F$ where the second $F_{W(k)}$ is the unique Frobenius lift $F_{W(k)} = W(F_k)\colon W(k)\to W(k)$ on the Witt vectors. If $\operatorname{Def}_{A_0/W(k)} \simeq \operatorname{Spf} R$ and $F$ is the Serre-Tate lift of Frobenius, then $\sigma_F$ corresponds to an element $\tilde A_{\rm can}$ of $\operatorname{Def}_{A_0/W(k)}(W(k))$, i.e. a (formal) lifting of $A_0$ over $W(k)$.

What happens if we replace $W(k)$ with some $p$-adic $W(k)$-algebra $V$ (a quotient of $W(k)[\![x_1, \ldots, x_n]\!]$ for some $n$)? Do we get some canonical lift of $A_0$ over $\operatorname{Spf} V$? For this to make sense, we must assume that we are given some deformation $A_{V_0}$ of $V_0$ over $\operatorname{Spf} V_0$ where $V_0 = V/pV$, and that $V$ is endowed with a Frobenius lifting $F_V\colon V\to V$. Thus the question is whether there exists a natural (possibly unique) $\sigma\colon R\to V$ with $\sigma F=F_V\sigma$ lifting the $R_0\to V_0$ induced by $A_{V_0}$.

The answer to this is no. For example, let $V=W_2(k)[\varepsilon]/(\varepsilon^2)$, so $V_0 = k[\varepsilon]/(\varepsilon^2)$. Let $F_V\colon V\to V$ be a lifting of Frobenius, so $F_V(\varepsilon)=\varepsilon^p + p\beta = p\beta$ for some $\beta\in V_0$. Let $R \simeq W(k)[\![q_{ij}-1]\!]$ as in Serre--Tate theory, so $F(q_{ij})=q_{ij}^p$. Let $R_0\to V_0$ send $q_{ij}$ to $\varepsilon + 1$ for all $i$ and $j$ (just so that the tangent direction does not lie in any of the hyperplanes/subtori $q_{ij}=1$). Suppose that $\sigma\colon R\to V$ lifts $R_0\to V_0$, so $\sigma(q_{ij}) = \varepsilon + 1 + pg_{ij}$ for some $g_{ij}\in V_0$. The condition $\sigma F = F_V\sigma$ means that $(\varepsilon+1+pg_{ij})^p = p\beta + 1 + pg_{ij}^p$. But the left hand side is $(1+pg_{ij})^p + p\varepsilon(1+pg_{ij})^{p-1} = 1 + p\varepsilon$. We get $\beta + g_{ij}^p = \varepsilon$. Thus $\varepsilon - \beta$ has to be a $p$-th power in $V_0$, which is a non-trivial condition on $\beta$. (Perhaps this can be simplified)

What is true though is this. We have a canonical homomorphism $W_{n+1}(V_0)\to V_n$ ($V_n=V/p^{n+1}$), taking $(x_0, \ldots, x_n)$ to $y_0^{p^n} + py_1^{p^{n-1}} + \cdots + p^ny_n$, where $y_i\in V_n$ are lifts of $x_i\in V_0$. Moreover, the lifting of Frobenius $F$ on $R$ induces a homomorphism $t_F \colon R_n\to W_{n+1}(R_0)$ (called "Cartier's arrow" in Illusie "Complexe de de Rham-Witt...", chapitre 0). We can now form the composition $$ \tilde\sigma_0\colon R_n \to W_{n+1}(R_0)\xrightarrow{W_{n+1}(\sigma_0)} W_{n+1}(V_0)\to V_n $$ where $\sigma_0\colon R_0\to V_0$ is the given map. This map turns out to satisfy $\tilde\sigma_0 F = F_{V_n}\tilde\sigma_0$ for every lifting of Frobenius $F_{V_n}\colon V_n\to V_n$. However, modulo $p$ this map reduces to $F^n\sigma_0 = \sigma_0 F^n$. Therefore we obtain a lifting of $(F^n)^*A_{V_0}$, not of $A_{V_0}$, over $V_n$. This recovers what I said in the first paragraph.

$\endgroup$
3
  • $\begingroup$ Thanks a lot for this excellent answer. I would definitively appreciate if you have time to explain why there is no preferred lift! $\endgroup$ Commented Feb 3, 2021 at 15:56
  • $\begingroup$ Again, many thanks for your edit. I wish I could validate your answer twice! What you mentioned about iterating the Frobenius is really interesting, because this is one way to compute canonical lifts in practice. $\endgroup$ Commented Feb 3, 2021 at 21:02
  • $\begingroup$ @RandomMathUser Now you got me curious. What does "in practice" mean? Shoot me an email if you want to continue this thread. $\endgroup$ Commented Feb 3, 2021 at 21:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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