1.6. ABELIAN VARIETIES OVER FINITE FIELDS 75
elliptic curve E, so Gκ αp as κ-groups (since Gκ is local-local of order p, so
= κ with F = V = 0). But OK,(p) is an absolutely unramified discrete
valuation ring, so there are no finite flat group schemes over OK,(p) with special
fiber αp (by the classification results in ). This contradiction shows that E
and E are not isogenous (so AK is not isotypic), as claimed. In fact, we have
proved something stronger: if Kp denotes the p-adic completion of K then AKp is
1.6.5. CM lifting after a field extension and isogeny. The proof of the sur-
jectivity aspect of the Honda-Tate theorem requires constructing abelian varieties
having prescribed properties over finite fields. The idea is to relate simple abelian
varieties over finite fields to simple factors of reductions of CM abelian varieties
over number fields, at least after some finite extension on the initial finite field.
(See  or [121, Lemme 3] for details.)
One can ask (as Honda implicitly did at the end of [50, §2]) to do better by
arranging simplicity to hold for the reduction of a CM abelian variety over a number
field (thereby eliminating the need to pass to a simple factor). Building on earlier
work of Honda, such an improved lifting theorem was proved by Tate [121, Thm. 2]
(and is really the starting point for the many lifting questions about CM abelian
varieties that we consider in this book):
188.8.131.52. Theorem (Honda, Tate). For any isotypic abelian variety A over a finite
field κ, there is a finite extension κ /κ such that Aκ is isogenous to the reduction
of a CM abelian variety with good reduction over a p-adic field with residue field κ .
Proof. By Corollary 184.108.40.206, there is a CM field L ⊂
with [L : Q] =
2 dim(A). The field L is its own centralizer in
so it contains an element π
which acts by the q-Frobenius endomorphism on A, where q = #κ. Let g = dim(A).
Since A is κ-isotypic, Tate’s work on isogenies among abelian varieties over finite
fields  gives two results for A: (i) the common characteristic polynomial over Q
for the action of π on the Tate modules of A is a power of an irreducible polynomial
fπ over Q (necessarily the minimal polynomial of π over Q), and (ii) A is κ-isogenous
to any g-dimensional isotypic abelian variety over κ whose q-Frobenius is a zero of
fπ. Moreover, these properties persist after replacing κ with any finite extension
κ (and replacing π with π[κ :κ]), due to Proposition 220.127.116.11.
By [121, §3, Thm. 2] (which is stated in the simple case but holds in the isotypic
case by the same proof), there exists a number field K ⊂ Qp, a g-dimensional
abelian variety B over K with good reduction at the induced p-adic place v of K, an
embedding of finite fields κ → κv, and an action of OL on B such that the reduction
B0 at v has qv-Frobenius in OL given by the action of πv = π[κv:κ] ∈ OL. (Here,
qv = #κv.) Since B0 admits a CM structure over κv by a field (namely, L), it is κv-
isotypic. Thus, since dim(B0) = dim(B) and Aκv satisfies FrAκv
it follows from the results (i) and (ii) in  recalled above that there exists a
κv-isogeny φ : B0 → Aκv .
18.104.22.168. Remark. The κv-isogeny φ : B0 → Aκv at the end of the proof of The-
orem 22.214.171.124 might not be L-linear, though it is Q(πv)-linear since it is compatible
with qv-Frobenius endomorphisms. We can exploit the Q(πv)-linearity of φ to find
an L-linear κv-isogeny B0 → Aκv as follows.