Loading [MathJax]/extensions/TeX/boldsymbol.js
Skip to main content

Section 1.14 The Honda Tate Theorem (Angus)

q = p^n\text{,} A a simple abelian variety over \FF_q\text{,} \pi_A the frobenius on A\text{,} \End^0(A) = \QQ\otimes \End(A)\text{,} f_A is the charpoly of A (i.e. of \pi_A).

Definition 1.14.3. q-Weil numbers.

A q-Weil number is an algebraic integer \pi s.t.

\begin{equation*} \forall \iota \colon \QQ\lb \pi\rb \hookrightarrow \CC,\,|\iota(\pi)| = \sqrt q \end{equation*}

we say that two q-Weil numbers are conjugate if they have the same minimal polynomial over \QQ\text{,} and write \pi \sim \pi'\text{.}

From the facts so far we have a map

\begin{equation*} \{\text{simple AVs}/\FF_q\}\to \{q\text{-Weil numbers}\} \end{equation*}
\begin{equation*} A \mapsto \pi_A \end{equation*}

We need to show this is well-defined, injectivity and surjectivity.

Subsection 1.14.1 Honda-Tate map

Recall:

By above \(V_l A \simeq V_lB\) for all \(l \ne p\) but \(f_A\) (resp. \(f_B\)) is the charpoly of \(\pi_a\) (\(\pi_B\)) on \(V_l A\) (\(V_l(B)\)).

The Galois modules \(V_lA\) and \(V_l B\) are semisimple. The Brauer-Nesbitt theorem says \(f_A = f_B \implies V_lA \simeq V_lB \) for \(l\ne p\text{.}\)

Recalling that f_A is a power of the minimal polynomial of \pi_A\text{,}

\begin{equation*} A \sim_{\text{isog}} B\implies f_A =f_B \implies \pi_A \sim \pi_B\text{.} \end{equation*}

So the Honda-Tate map is well defined.

This doesn't quite give injectivity because a priori f_A and f_B could be powers of the minpolys of \pi_A, \pi_B\text{.}

Subsection 1.14.2 Injectivity and Brauer groups

From last time:

Study the formula for \(r(f_A, f_A)\) Edixhoven-van der Geer-Moonen 16.22.

So the next step is to try and recover \End^0(A) from \pi\text{.}

Definition 1.14.9. Central simple algebras.

A central simple algebra B/k is a k-algebra B with no two-sided ideals and Z(B) = k\text{.}

Definition 1.14.11. Brauer groups.

The Brauer group of k \operatorname{Br}(k) is the set of central simple algebras under \otimes modulo the algebras M_n(k)\text{.}

Edixhoven-van der Geer-Moonen 16.30.

\(\Leftarrow\) done.

\(\Rightarrow\) Let \(D_{\pi_A} , D_{\pi_B}\) be the division rings with invariants specified as in Proposition 1.14.13. \(\pi_A\sim \pi_B \implies D_{\pi_A} \simeq D_{\pi_B} \implies f_A = \operatorname{minpoly}(\pi_A) ^d = f_B\text{.}\)

Subsection 1.14.3 Surjectivity and CM theory

We need to show that for \pi a q-Weil number there exists an abelian variety A/\FF_q such that \pi_A \sim \pi\text{.}

Definition 1.14.16.

Such a q-Weil number \pi is called effective.

\(\Rightarrow\) clear.

\(\Leftarrow\) By assumption we have \(A'/k\) a simple abelian variety s.t. \(\pi_{A'} \sim \pi^N\) for \(k\) a degree \(N\) extension of \(\FF_q\text{.}\) Let

\begin{equation*} A = \Res_{k/\FF_q}(A') \end{equation*}

on the rational Tate modules we have

\begin{equation*} V_l A = \Ind_{G_k}^{G_{\FF_q}} (V_lA') \end{equation*}

where

\begin{equation*} G_k = \Gal{\overline{\FF_q}}{k} \end{equation*}
\begin{equation*} G_{\FF_q} = \Gal{\overline{\FF_q}}{\FF_q} \end{equation*}

since \(G_k\text{,}\) \(G_{\FF_q}\) are abelian, by studying the induced action, one can see

\begin{equation*} \Ind_{G_k}^{G_{\FF_q}} (\pi_{A'}) = \pi_A^N \end{equation*}

in particular \(f_A(T) = f_{A'}(T^N)\text{.}\) Choosing a simple factor \(A_i\) one gets \(\pi_{A_i} \sim \pi\text{.}\)

So it is sufficient to show \pi^N is effective.

Strategy for proving surjectivity

  1. Construct a division algebra D_\pi\text{.}
  2. Choose a CM field L splitting D_\pi\text{.}
  3. Find an abelian variety A/\CC of type (L, \Phi)\text{.}
  4. In fact A is defined over a number field K and has good reduction at v|p\text{.}
  5. Apply the Shimura-Taniyama formula to relate \pi_A to \Phi\text{.}
  6. Choose \Phi wisely (in retrospect in 3) to relate \pi to \pi_A\text{.}
  7. Show \pi_A^N = \pi^{N'}\text{.}

D_\pi is given by the invariants described by \pi (and K = \QQ(\pi)).

Two cases:

  1. \(\QQ(\pi)\) is totally real, in which case \(\QQ(\pi) = \QQ\) or \(\QQ(\sqrt{p})\text{.}\)
  2. \(\QQ(\pi)\) is a CM field with totally real subfield \(\QQ(\pi + q/\pi)\text{.}\)

In the case

  1. Choose \(L = \QQ(\pi)(\sqrt{-p})\text{.}\)
  2. Let \(d= \lb D_\pi: \QQ(\pi)\rb^{1/2}\text{.}\) This \(L\) splits \(D_\pi\text{.}\)
Definition 1.14.19. CM types.

For a CM field L all the embeddings

\begin{equation*} \iota\colon L \hookrightarrow \CC \end{equation*}

come in complex conjugate pairs, choosing an embedding for each pair defines a subset \Phi\subseteq \Hom(L, \CC) such that

\begin{equation*} \Phi \cup \overline \Phi = \Hom(L,\CC) \end{equation*}
\begin{equation*} \Phi \cap \overline \Phi = \emptyset \end{equation*}

such a choice of \Phi is called a CM type.

Let A/\CC be an abelian variety with CM by L i.e.

\begin{equation*} L\hookrightarrow \End^0(A) \end{equation*}

then

\begin{equation*} \CC\otimes L = \prod_\iota \CC \end{equation*}

acts on the tangent space at the origin \Lie(A)\text{.}

Found in Shimura-Taniyama.

The fact that A is in fact defined over a number field K is also in Shimura-Taniyama.

Highly nontrivial, Neron models, Chevalley decomposition, Neron-Ogg-Shafarevich criterion, result of Grothendieck on potentially stable reduction.

After passing to a finite extension we will assume A/K has good reduction at places v|p\text{.} So we have a reduction A_{\FF_{q'}}/\FF_{q'}\text{.} For a place w|p of L let

\begin{equation*} \Sigma_w = \Hom(L_w, \CC_p) \end{equation*}
\begin{equation*} \Phi_w = \Phi \cap \Sigma_w\text{.} \end{equation*}

Tate has a proof using CM theory of \(p\)-divisible groups.

Recall we fixed \pi and from this we deterministically formed \QQ(\pi), D_{\pi}, L however we have no restriction on our choice of \Phi\text{.}

Let \(v = w|_{\QQ(\pi)}\) be the place of \(\QQ(\pi)\) below \(w\text{.}\) Let

\begin{equation*} n_w = \frac{w(\pi)}{w(q)}\#\Sigma_w = \frac{w(\pi)}{w(q)}[L_w :\QQ_p] \end{equation*}
\begin{equation*} = \frac{w(\pi)}{w(q)}[L_w :\QQ(\pi)_v ][ \QQ(\pi)_v:\QQ_p] \end{equation*}

by recalling the formula for the local invariants of \(D_\pi\) we get

\begin{equation*} n_w = \operatorname{inv}_w(D_\pi\otimes_{\QQ(\pi)} L)\text{.} \end{equation*}

But \(L\) splits \(D_\pi\) so \(n_w \in \ZZ\text{,}\) further

\begin{equation*} n_w + n_{\overline w} = \left( \frac{w(\pi)}{w(q)} + \frac{\overline w(\pi)}{\overline w(q)} \right) \#\Sigma_w \end{equation*}
\begin{equation*} = \left( \frac{w(\pi\overline \pi)}{w(q)} \right) \#\Sigma_w = \#\Sigma_w \end{equation*}

check the CM type \(\Phi = \bigcup_w \Phi_w\) where for each \(w\) \(\# \Phi_w = n_w\text{.}\) Then the formula follows.

Combining the previous result with the Shimura-Taniyama formula we get that for all places w|p

\begin{equation*} \frac{w(\pi_{A_{\FF_{q'}}})}{w(q')} = \frac{w(\pi)}{w(q)}\text{.} \end{equation*}

Taking the correct power,

\begin{equation*} w\left( \frac {\pi^m_{A_{\FF_{q'}}}}{\pi^{m'}}\right) = 0 \forall w|p \end{equation*}
\begin{equation*} \pi,\pi_{A_{\FF_{q'}}}|q^{m'} \end{equation*}
\begin{equation*} \implies w(\cdots) = 0 \forall w \nmid p \end{equation*}

since |\pi^{m'}|_w = |\pi_{A_{\FF_{q'}}}^m|_w = (q^{m'})^{1/2} \forall \text{ infinite places}

\begin{equation*} \pi_{A_{\FF_{q'}}}/ \pi_A^{m'} \end{equation*}

is a root of unity \pi^N_{A_{\FF_{q'}}} = \pi^{N'}\text{.}