ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES

STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Abstract. We prove p-adic uniformization for Shimura curves attached to the group of uni-
tary similitudes of certain binary skew hermitian spaces V with respect to an arbitrary CM
field K with maximal totally real subfield F . For a place v|p of F that is not split in K and for
arXiv:2007.05211v1 [math.AG] 10 Jul 2020

which Vv is anisotropic, let ν be an extension of v to the reflex field E. We define an integral

model of the corresponding Shimura curve over Spec OE,(ν) by means of a moduli problem for
abelian schemes with suitable polarization and level structure prime to p. The formulation of
the moduli problem involves a Kottwitz condition, an Eisenstein condition, and an adjusted
invariant. The first two conditions are conditions on the Lie algebra of the abelian varieties;
the last condition is a condition on the Riemann form of the polarization. The uniformization
of the formal completion of this model along its special fiber is given in terms of the formal
Drinfeld upper half plane Ω b F for Fv . The proof relies on the construction of the contracting
v
functor which relates a relative Rapoport-Zink space for strict formal OFv -modules with a
Rapoport-Zink space of p-divisible groups which arise from the moduli problem, where the
OFv -action is usually not strict when Fv 6= Qp . Our main tool is the theory of displays, in
particular the Ahsendorf functor.

Contents
1. Introduction 1
2. Main local statements 10
3. Background on Display Theory 20
4. The contracting functor 45
5. The alternative moduli problem revisited 76
6. Moduli spaces of formal local CM-triples 96
7. Application to p-adic uniformization 109
8. Appendix: Adjusted invariants 135
References 144

1. Introduction
1.1. History of uniformization. One of the major results of the Mathematics of the 19th
century is the uniformization theorem. It states that any non-singular projective algebraic curve
X of genus g(X) ≥ 2 can be uniformized, i.e., can be written as
X ≃ Γ\ΩR , (1.1.1)
where ΩR = P1 (C) \ P1 (R) is the union of the upper and the lower half plane and Γ denotes a
discrete cocompact subgroup of PGL2 (R). This notation reinforces the analogy with the p-adic
uniformization discussed below. The history of this theorem is very complicated, and involves the
names of many mathematicians, among them Poincaré, Hilbert and Koebe, comp. [12]. Inspired
by the uniformization theorem, Poincaré gave a systematic construction of cocompact discrete
subgroups of PGL2 (R). For this he used the exceptional isomorphism between inner forms of
PGL2 and special orthogonal groups of ternary quadratic forms. In fact, for his construction, he
used arithmetic subgroups of the special orthogonal group of an indefinite anisotropic ternary
quadratic form over Q, cf. [12].

Date: July 13, 2020.

1
2 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Now let p be a prime number. The history of the p-adic uniformization of algebraic curves
starts with Tate’s uniformization theory of elliptic curves. It turns out that not all elliptic
curves over p-adic fields admit a p-adic uniformization, but only those with (split) multiplicative
reduction [30, §6].
The next step was Mumford’s p-adic uniformization theory of algebraic curves of higher
genus, [23]. Again, it turns out that not all such algebraic curves over p-adic fields admit a
p-adic uniformization by an admissible open subset of P1 , but only those with totally degenerate
reduction [23]. In view of Mumford’s results, it becomes interesting to single out classes of
algebraic curves with totally degenerate reduction. Such classes are exhibited by Cherednik’s
theorem [7].
Cherednik’s theorem states that Shimura curves associated to certain quaternion algebras
over a totally real field F admit p-adic uniformization. The quaternion algebra is required to be
split at precisely one archimedean place w of F (and ramified at all other archimedean places),
and to be ramified at a non-archimedean place v of residue characteristic p. In this case, the
reflex field can be identified with F . Then one obtains p-adic uniformization by the Drinfeld
halfplane associated to Fv , provided that the level structure is prime to v. Cherednik’s theorem
implies that if X is a connected component of the Shimura tower for such a level, considered as
an algebraic curve over F̄ , then there is an isomorphism of algebraic curves over F̄v ,

X ⊗F̄ F̄v ≃ (Γ̄\ΩFv ) ⊗Fv F̄v . (1.1.2)

Here ΩFv = P1Fv \ P1 (Fv ) denotes the Drinfeld upper halfplane for the local field Fv , and Γ̄
denotes a discrete cocompact subgroup of PGL2 (Fv ). Recall that ΩFv is a rigid-analytic space
over Fv . The isomorphism (1.1.2) is to be interpreted as follows: the rigid-analytic space Γ̄\ΩFv
is (uniquely) algebraizable by a projective algebraic curve over Fv . After extension of scalars
Fv −→ F̄v , there exists an isomorphism as in (1.1.2). We thus see that (1.1.2) allows us to pass
from the original complex uniformization X ⊗F̄ C ≃ Γ\ΩR , where Γ is a congruence subgroup
maximal at v, to p-adic uniformization.
Let us comment on the proof of Cherednik’s theorem. When F = Q, these Shimura curves
are moduli spaces of abelian varieties with additional structure, and Drinfeld [9] gave a moduli-
theoretic proof of Cherednik’s theorem in this special case. Furthermore, he proved an ‘integral
version’ of this theorem (which has the original version as a corollary). This integral version
describes a concrete model of the Shimura variety over Spec OF,(ν) and a description of the
formal completion along its special fiber in terms of a formal scheme version of ΩFν . It relies
on a theorem on formal moduli spaces of p-divisible groups, which is in fact the deepest part of
Drinfeld’s paper. When F 6= Q, Cherednik’s Shimura curves do not represent a moduli problem
of abelian varieties, and Cherednik’s method of proof is indirect and apparently uses Ihara’s
theory of elliptic elements in congruence monodromy problems.
There are also higher-dimensional versions of p-adic uniformization. Drinfeld’s method has
been generalized by Rapoport and Zink [27] to Shimura varieties associated to certain fake
unitary groups. These are associated to central division algebras over a CM-field equipped with
an involution of the second kind; for Rapoport-Zink uniformization, one has to assume that
the p-adic place of the totally real subfield splits in the CM-field. This higher-dimensional
generalization also includes integral uniformization theorems. In [27], these integral uniformiza-
tion theorems appear as a special instance of a general non-archimedean uniformization theorem,
which describes the formal completion of PEL-type Shimura varieties along a fixed isogeny class.
In the case of p-adic uniformization, the whole special fiber forms a single isogeny class.
The method of [27] has been applied by Boutot and Zink [5] to prove variants of Cherednik’s
original theorem by embedding Cherednik’s Shimura curves into Shimura curves obtained by the
Rapoport-Zink method; however, the integral uniformization theorems in [5] are rather weak in
that they only show that there exists some integral model of the Shimura curve for which one
has integral uniformization.
A variant of Cherednik’s method has been developed by Varshavsky [31, 32] to obtain p-
adic uniformization of certain higher-dimensional Shimura varieties associated to fake unitary
groups, again at a split place. We refer to Boutot’s Bourbaki talk [3] for an account of all these
developments.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 3

In the present paper, we deal with Shimura curves attached to unitary similitude groups
associated to anti-hermitian1 vector spaces V of dimension 2 over a CM-field K with totally
real subfield F of arbitrary degree. Our results generalize those in [18], where the case F = Q
is considered. Like Cherednik, we assume that V is split at precisely one archimedean place w
of F (and ramified at all other archimedean places). We also assume that V is ramified at a
non-archimedean place v of residue characteristic p of F . However, in contrast to the cases of
p-adic uniformization mentioned above, we assume that v does not split in K. Of course, these
Shimura curves are closely related to the Shimura curves considered by Cherednik (we refer to
[18] for a general discussion of the relation between quaternion algebras and two-dimensional
hermitian vector spaces). However, they are different. In particular, they have the enormous
advantage that they always represent a moduli problem of abelian varieties. Our uniformization
theorem is optimal when the level structure imposed is prime to p, in the sense that it extends
to an integral uniformization that allows an explicit interpretation of the points in the reduction
modulo p.
As in Drinfeld’s approach, our uniformization theorem relies on a theorem on formal moduli
spaces of p-divisible groups. In fact, the main work in proving our theorems is to establish an
isomorphism of our formal moduli spaces with the moduli space of Drinfeld. Such an isomorphism
is also constructed by Scholze and Weinstein [29]. Their construction relies on Scholze’s theory of
local Shimura varieties and his integral p-adic Hodge theory, as well as on results in a preliminary
(unpublished) version of the present paper on local models. Our construction here is more direct
and more elementary; it relies on the theory of displays, cf. [33]. It is not clear to us that the
isomorphism in [29] coincides with the one constructed here.
Drinfeld’s version of Cherednik’s theorem for F = Q has found numerous arithmetic ap-
plications, to level raising, level lowering and bounding Selmer groups, at the hands of Ribet,
Bertolini, Darmon, Nekovar and many others, comp. also the references in the introduction of
[18]. It is to be hoped that our direct construction is the basis of similar such applications for
general totally real fields F .
Our results are an expression of the exceptional isomorphism between an inner twist of the
adjoint group of GL2 and an inner twist of the adjoint group of U2 . Just as for Poincaré’s excep-
tional isomorphism of inner forms of PGL2 and special orthogonal groups of ternary quadratic
forms, there is no higher rank analogue.

1.2. Global results. Now let us state our global results. Let K be a CM-field, with totally
real subfield F . We denote the non-trivial F -automorphism of K by a 7→ ā. Let V be a two-
dimensional K-vector space, equipped with an alternating Q-bilinear form ψ : V × V −→ Q such
that
ψ(ax, y) = ψ(x, āy), x, y ∈ V, a ∈ K. (1.2.1)
There is a unique anti-hermitian form κ on V such that
TrK/Qp aκ(x, y) = ψ(ax, y), x, y ∈ V, a ∈ K. (1.2.2)
Conversely, the anti-hermitian form κ determines the alternating bilinear form ψ with (1.2.1).
We say that κ arises from ψ by contraction. Recall that anti-hermitian spaces V are determined
up to isomorphism by their signature at the archimedean places of F and their local invariants
invv (V ) at the non-archimedean places v of F . Let w be an archimedean place such that Vw has
signature (1, 1) and such that Vw′ is definite for all archimedean places w′ 6= w. Let us be more
precise. Let Φ = HomQ-Alg (K, C). Let r be a generalized CM-type of rank 2, special w.r.t. w,
i.e., a function
r : Φ −→ Z>0 , ϕ 7−→ rϕ , (1.2.3)
such that rϕ + rϕ̄ = 2 for all ϕ ∈ Φ, and such that for the extensions {ϕ0 , ϕ̄0 } of w we have
rϕ0 = rϕ̄0 = 1 and with rϕ ∈ {0, 2} for ϕ ∈
/ {ϕ0 , ϕ̄0 }, comp. [18]. Then we demand that the
signature of Vϕ = V ⊗K,ϕ C be equal to (rϕ , 2 − rϕ ).

1It turns out to be more natural to consider anti-hermitian forms, rather than hermitian forms, cf. below.
4 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

We denote the reflex field of r by E = Er . It is a subfield of Q, the algebraic closure of Q in

C. Note that F embeds via ϕ0 into E, and that the archimedean place of F induced by
ϕ0
F −→ E −→ C (1.2.4)
is equal to w. If F = Q, then E = F .
Associated to these data, there is a Shimura pair (G, {h}). Here G denotes the group of
unitary similitudes of V , with similitude factor in Gm , an algebraic subgroup of GSp(V, ψ) over
Q. For an open compact subgroup K ⊂ G(Af ), there is a Shimura variety ShK , with canonical
model over the reflex field E, whose complex points are given by
ShK (C) ≃ G(Q)\[ΩR × G(Af )/K].
Here ΩR is acted on by G(R) via the projection to GU(Vw )ad and a fixed isomorphism GU(Vw )ad ≃
PGL2 (R).
Consider the following moduli problem on (Sch/E). It associates to an E-scheme S the set
of isomorphism classes of tuples (A, ι, λ̄, η̄). Here
• A is an abelian scheme up to isogeny of dimension 2[F : Q] over S.
• ι : OK −→ End(A) is an action of OK on A such that
X
Tr(ι(a)| Lie A) = rϕ ϕ(a), for all a ∈ OK .
ϕ∈Φ

• λ̄ is a Q-homogeneous polarization of A such that its Rosati involution induces the conjugation
on K/F .
∼ b
• a K-orbit of K-linear similitudes η̄ : V ⊗ Af −
→ V (A).
Here the rational Tate module is equipped with its natural anti-hermitian form arising by con-
traction from its polarization form. In the case where S is connected, a more precise formulation
of the last datum is as follows. Fix a geometric point ω → S. Then η̄ is a K-orbit of K-linear
∼
similitudes η : V ⊗ Af → V̂ (Aω ) which is invariant by the action of π1 (S, ω) on V̂ (Aω ).
This moduli problem is represented by a quasi-projective scheme AK,E which is the canonical
model of ShK over E. It is a projective scheme when the existence of v as below is imposed.
Let p be a prime number and let v be a p-adic place of F which is non-split in K and such that
Vv is a non-split Kv /Fv -anti-hermitian space, i.e., invv (V ) = −1. We take the open compact
subgroup of the form K⋆ = K⋆p · Kp , where Kp is an arbitrary open compact subgroup of G(Apf ),
and where K⋆p has the following shape. Let
M
V ⊗ Qp = Vp
p|p

be the orthogonal decomposition according to the prime ideals of F over p. Note that the prime
ideal pv corresponding to v occurs as an index here. Then
Y
G(Qp ) ⊂ Gp (Qp ) ,
p

where Gp denotes the group of unitary similitudes of Vp with similitude factor in Gm . We take
K⋆p of the form
K⋆p = G(Qp ) ∩ Kv K⋆,v
p , (1.2.5)
⋆,v
Q
where Kv is the unique maximal compact subgroup of Gpv (Qp ), and where Kp ⊂ p6=pv Gp (Qp )
is an arbitrary open compact subgroup.
Let J be the inner form of G which is anisotropic at w and quasi-split at v, and which locally
coincides with G at all places 6= v, w of F . Then there exists an identification of the adjoint
group Jv,ad (Qp ) with PGL2 (Fv ) and an action of J(Q) on G(Af )/K⋆ (which is, however, not
induced by an action of J(Q) on G(Af )).
We now formulate our main theorem in the version over a p-adic field, cf. Corollary 7.5.2.
Recall the embedding (1.2.4) of F into E. We choose a place ν of E over v. Throughout the
paper, we always assume2 p 6= 2 if v is ramified in K.

2In the light of the results of Kirch [14], it should be possible to remove this blanket assumption.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 5

Theorem 1.2.1. Let K⋆ = K⋆p Kp , where K⋆p is of the form (1.2.5). Let Ĕν be the completion
of the maximal unramified extension of Eν in Qp . There is a finite abelian extension Ĕν⋆ of Ĕν
and an isomorphism of algebraic curves over Ĕν⋆ ,


AK⋆ ,E ×Spec E Spec Ĕν⋆ ≃ J(Q)\[ΩFv × G(Af )/K⋆ ] ×Spec Fv Spec Ĕν⋆
Here, as before, ΩFv denotes the Drinfeld halfplane relative to the local field Fv , and the
interpretation is as before that the scheme on the LHS is the algebraization of the rigid-analytic
variety on the RHS. If Kp is of the form (1.2.7) below, then we may take Ĕν⋆ = Ĕν , cf. Theorem
1.2.3 below; but in general, one needs a non-trivial extension, comp. Theorem 1.2.4.
From this theorem we deduce an analogue of Cherednik’s isomorphism (1.1.2), noting that
any geometric connected component X of ShK⋆ is defined over the maximal abelian extension
E ab of E,
X ⊗E ab Ĕνab ≃ (Γ̄\ΩFv ) ⊗Fv Ĕνab . (1.2.6)
Here Ĕνab denotes the maximal abelian extension of Ĕν , and Γ̄ is a cocompact discrete subgroup
of PGL2 (Fv ). Since the Cherednik Shimura datum is a central twist of (G, {h}), the geometric
connected components of ShK can be identified with those appearing in Cherednik’s theorem,
so that in fact (1.1.2) follows from Theorem 1.2.1.
By extending the moduli problem for ShK integrally over Spec OE,(pν ) , we obtain semi-global
integral models of these Shimura varieties. This gives us the possibility of formulating an ‘inte-
gral’ version of this theorem. Let us explain the moduli problem in question.
We first explain the level structure. For every p|p, we fix a lattice Λp in Vp . We assume that
Λp is a self-dual lattice when p is either split in K or ramified. When p is unramified in K, we
assume that Λp is selfdual when invp (V ) = 1, and almost selfdual when invp (V ) = −1. Let
Kp = {g ∈ G(Qp ) | gΛp = Λp , for all p|p}. (1.2.7)
We also fix an open compact subgroup K ⊂ p
G(Apf ) p
and set K = Kp K . We continue to assume
that for the distinguished p-adic place v we have invpv (V ) = −1.
We now define a functor AK on the category of OE,(pν ) -schemes. Let S ∈ (Sch/OE,(pν ) ).
Then a point of AK (S) consists of an equivalence class of quadruples (A, ι, λ̄, η̄ p ). Here
• A is an abelian scheme over S and ι : OK −→ End(A) ⊗ Z(p) is an action of OK on A.
• λ̄ is a Q-homogeneous polarization of A such that its Rosati involution induces the conjugation
on K/F .
∼ bp
• η̄ p : V ⊗ Ap −
f → V (A) is Kp -class of K-linear similitudes.
Here the prime-to-p-rational Tate module Vb p (A) of A is equipped with its natural anti-hermitian
form arising by contraction from its polarization form. Two quadruples (A, ι, λ̄, η̄ p ) and (A′ , ι′ , λ̄′ , η̄ p,′ )
are equivalent, if there exists an isogeny A −→ A′ of degree prime to p compatible with the re-
maining data.
We impose the following conditions on the quadruples (A, ι, λ̄, η̄ p ). First, for the action of
OK on Lie A induced by ι, we impose the Kottwitz condition (KCr ) relative to r, comp. [18]. In
addition, we demand that this action also satisfies the Eisenstein condition (ECr ) relative to r.
This condition is defined in section 2, and is a key novelty of this paper. The condition (ECr )
follows from the Kottwitz condition (KCr ) when S is an E-scheme but is quite subtle when p is
nilpotent in OS . Imposing this condition ensures the flatness of the moduli scheme.
Secondly, we demand that there exists a polarization λ ∈ λ̄ such that, for every p|p, the
localization of the kernel of the polarization λ at the place p satisfies
|(Ker λ)p | = [Λ∨
p : Λp ]. (1.2.8)
Thirdly, we impose that for each p|p, the r-adjusted invariant invrp (A, ι, λ)
coincides with the
invariant invp (V ) of the hermitian space V . Here the r-adjusted invariant of the triple (A, ι, λ),
defined in Appendix A, is another key novelty of this paper. This condition is automatically
satisfied when pv is the only prime ideal of F over p. In general, this condition cuts out the open
and closed part of the moduli scheme defined by the Shimura variety. The reason for the name
r-adjusted is that this adjusts the definition of the invariant in [18], where it was erroneously
6 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

asserted that the invariant is locally constant in families. We prove here that this local constancy
indeed holds for the r-adjusted invariant, cf. Proposition 8.2.1.
Proposition 1.2.2. Let r be a generalized CM-type of even rank n, with associated reflex field
E. Let S be an OE -scheme. Let (A, ι, λ) be a CM-triple over S which satisfies the Kottwitz
condition (KCr ). Let c ∈ {±1}. Then for every non-archimedean place v of F , the set of points
s ∈ S such that
invrv (As , ιs , λs ) = c
is open and closed in S.
We refer to the body of the text for the terminology used.
Here, now, is our main theorem in the context of schemes over p-adic integer rings, cf. Theorem
7.3.3.
Theorem 1.2.3. Let K = Kp Kp , where Kp is of the form (1.2.7).
(i) The functor AK is representable by a projective flat OE,(pν ) -scheme, with semi-stable reduc-
tion. Its generic fiber AK ⊗OE,(pν ) E is identified with AK,E , which is the canonical model of
ShK .
(ii) Let ÂK be the formal completion of AK along its special fiber, which is a formal scheme
over Spf OEν . Then there exists an isomorphism of formal schemes over Spf OĔν ,


ÂK ×Spf OEν Spf OĔν ≃ J(Q)\ Ω b Fv ×Spf OF Spf O
v Ĕν × G(Af )/K .

For varying Kp , this isomorphism is compatible with the action of G(Apf ) through Hecke corre-
spondences on both sides.
Here Ωb Fv denotes the formal scheme version of ΩFv over Spf OFv due to Deligne, Drinfeld
and Mumford, cf. [9]. In section 7 we give a variant of the RHS, which allows us to express
the descent datum from OĔν to OEν of the LHS. Theorem 1.2.3 is optimal in the sense that it
describes explicitly the scheme AK over OEν and its p-adic uniformization.
If we assume that there are prime ideals p|p different from pv , we may pass to deeper level
structures and still prove an integral version of p-adic uniformization. Let K⋆p ⊂ G(Qp ) be of
the form
K⋆p = G(Qp ) ∩ Kv K⋆,vp , (1.2.9)
where Kv isQthe stabilizer of Λpv , and where K⋆,v
p is an arbitrary open compact subgroup of
Gv (Qp ) = p6=pv Gp (Qp ). The system of such subgroups is stable under conjugation with
elements of G(Qp ). For such subgroups, we have the following version of our main theorem, cf.
Corollary 7.4.14.
Theorem 1.2.4. Let K⋆ = K⋆p Kp , for a choice of Kp ⊂ G(Apf ), where K⋆p is of the form (1.2.9).
There exists a normal scheme A⋆K⋆ over Spec OĔν such that for the p-adic completion of this
scheme there is an isomorphism
p
Â⋆K⋆ ≃ J(Q)\[(Ω̂Fv ×Spf OFv Spf OĔν ) × Gv (Qp )/K⋆,v p
p × G(Af )/K ].

For varying K⋆ , these schemes form a tower with an action of the group G(Qp ) × G(Apf ), where
the action of G(Qp ) factors through G(Qp ) → Gv (Qp ). The isomorphism of formal schemes is
compatible with these actions.
The general fibre of A⋆K⋆ is a Galois twist of AK⋆ ,E ×Spec E Spec Ĕν as follows. There is a
character χ0 : Gal(Ĕνab /Ĕν ) → G(Qp ) with values in the center of G(Qp ) and an isomorphism
AK⋆ ,E ×Spec E Spec Ĕνab → A⋆K⋆ ×Spec OĔν Ĕνab

such that, for σ ∈ Gal(Ĕνab /Ĕν ), the action of idA⋆K⋆ × Spec σ on the right hand side induces
on the left hand side χ0 (σ) × Spec σ, where χ0 (σ) acts as a Hecke operator. The Galois twist
respects the Hecke operators.
The scheme A⋆K⋆ represents a moduli problem of abelian varieties with additional structure
over OĔν , cf. section 7.4. We refer to section 7.6 for the explicit determination of χ0 .
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 7

It should be pointed out that Theorem 1.2.4 is not optimal since we cannot describe the
descent to Eν . Also, when v is ramified in K, we can only give the character χ0 explicitly after
restricting to a subgroup of index 2. This is in contrast to Theorem 1.2.3. The deeper reason for
this deficiency lies in the fact that the natural context for Theorem 1.2.4 is the class of Shimura
varieties appearing in [24]. Let Ψ ⊂ Φ be a CM-type for K/F such that

Ψ ∩ (Φ \ {ϕ0 , ϕ̄0 }) = {ϕ ∈ Φ \ {ϕ0 , ϕ̄0 } | rϕ = 2}. (1.2.10)

There are two possibilities for Ψ. Let EΨ be the reflex field of Ψ and let Ẽ be the composite of
EΨ and E = Er . Then Ẽ is an extension of degree one or two of Er . Associated to (V, ψ, Ψ),
there is a finite number of Shimura varieties ShK̃ (G̃, {h̃}) with reflex field Ẽ, cf. [24, §3]. Here
G̃ maps surjectively to G with kernel a central torus, hence the Shimura varieties ShK̃ (G̃, {h̃})
are central twists of the Shimura variety ShK (G, {h}). Each one represents a moduli problem
on (Sch/Ẽ). In a sequel to this paper, we will construct semi-global integral models of these
Shimura varieties over Spec OẼ,(ν̃) . These are described by moduli problems of abelian varieties
on (Sch/OẼ,(ν̃) ) and admit p-adic uniformization in the strong sense of Theorem 1.2.3, when
the congruence condition on the open compact subgroup K̃ is prime to the chosen place v. The
trade-off in comparison with our Shimura variety is that the corresponding reflex field Ẽ is larger
than the reflex field E of our Shimura variety (which, in turn, is larger than the reflex field F of
Cherednik’s Shimura variety).
Both Theorems 1.2.1 and 1.2.3 are proved in [18] when Fv = Qp . Most of the work in [18]
was local, and an essential ingredient was the alternative moduli interpretation of the Drinfeld
halfplane in [17]. Once this is accomplished, the proof of the global theorems follows in a
relatively straightforward way from the general non-archimedean uniformization theory of [27,
Chap. 6]. The same is true here. In [18], we expressed the hope that it might be possible to
eliminate the strong limitation Fv = Qp made there, and this hope is achieved in the present
paper. As explained in [18], the main issue is the contrast between the condition on the action
of OFv on the Lie algebras of the p-divisible groups in the local moduli problems. On the one
hand, for the moduli problem represented by the Drinfeld half-plane Ω b Fv , the action of OFv on
the Lie algebra is required to be strict, i.e., to factor through the structure morphism of the base
scheme S. On the other hand, in the global moduli problem, the Lie algebras of the relevant
abelian schemes are often free OF ⊗Z OS -modules. The main results of the present paper, and
in particular the contracting functor defined in section 4, provide the bridge between the two
types of moduli problems.

1.3. Local results. Let us now formulate our local results, referring to section 2 for more
details and more explanations of some terms used here. Let p be a prime number, and let F
be a finite extension of degree d = [F : Qp ] of Qp and let K/F be a quadratic extension. Let
Φ = HomQp -Alg (K, Q̄p ), and fix a pair {ϕ0 , ϕ̄0 } of conjugate elements in Φ. Here ϕ̄0 (a) = ϕ0 (ā).
Let r be a local CM-type of rank 2 which is special w.r.t {ϕ0 , ϕ̄0 }, i.e., a function

r : Φ −→ Z>0 , ϕ 7−→ rϕ , (1.3.1)

such that rϕ + rϕ̄ = 2 for all ϕ ∈ Φ, and rϕ0 = rϕ̄0 = 1 and rϕ ∈ {0, 2} for ϕ ∈ / {ϕ0 , ϕ̄0 }, comp.
[18]. We denote the reflex field of r by E. It is a subfield of Q̄p .
For an OE -scheme S, we consider triples (X, ι, λ), where X is a p-divisible group of height
4d and dimension 2d over S, where ι : OK −→ End(X) is an action of OK on X, and where
λ : X −→ X ∨ is a polarization of X such that its Rosati involution induces on OK the conjugation
involution over OF . We impose the Kottwitz condition (KCr ) and the Eisenstein conditions
(ECr ) on the action of OK on Lie X. Furthermore, we assume that λ is a principal polarization
if K/F is ramified, and that λ is an almost principal polarization if K/F is unramified.
We fix such a triple (X, ιX , λX ) over the algebraic closure k̄ of the residue field κE of E, and
refer to it as a framing object. When K/F is unramified, then any two such triples are isogenous
by an OK -isogeny of height zero which preserves the polarizations. The same is true when K/F
is ramified, provided we impose that the r-adjusted invariant invr (X, ιX , λX ) is −1 (this last
8 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

condition is automatic when K/F is unramified). In either case, the group J(Qp ) of OK -self-
quasi-isogenies of (X, ιX ) preserving the polarization λX can be identified with the unitary group
of a split K/F -hermitian space of dimension 2.
We consider the Rapoport-Zink space Mr over Spf OĔ representing the functor on (Sch/ Spf OĔ )
which associates to S ∈ (Sch/ Spf OĔ ) the set of isomorphism classes of 4-tuples (X, ι, λ, ρ), where
(X, ι, λ) is as above, and where ρ is a framing of height zero, with framing object (X, ιX , λX ). Our
main local result may now be formulated as follows. We fix an isomorphism J 1 (Qp ) ≃ SL2 (F ).
bF ⊗
Theorem 1.3.1. The RZ-space Mr is isomorphic to Ω ˆ OF ,ϕ0 OĔ . More precisely, there exists
a unique isomorphism of formal schemes
b F ×Spf OF Spf O ,
Mr ≃ Ω Ĕ

which is equivariant with respect to the fixed identification J 1 (Qp ) ≃ SL2 (F ). In particular, Mr
is flat over Spf OĔ with semi-stable reduction.
The proof of this theorem uses the contracting functor which is a key technical novelty of this
paper and is based on the theory of displays. Let R be a p-adic ring, and let W (R) be its ring of
Witt vectors. A display over R is a finitely generated W (R)-module with additional structures,
namely a Frobenius morphism, a Hodge filtration and a divided Frobenius morphism. A display
will be denoted by P and the underlying W (R)-module by P . Under suitable hypotheses, the
category of formal p-divisible groups over R is equivalent to the category of nilpotent displays
over R. Let R be an OE -algebra which we regard as a OF -algebra via ϕ0 . The contracting
functor is the composition of two functors. The first functor associates to a triple (X, ι, λ) as
above over R a new tuple (X ′ , ι′ , λ′ ), where X ′ is a p-divisible group of height 4d and dimension
2 and where ι′ is an OK -action such that its restriction to OF is strict, i.e., the induced action
on Lie X ′ coincides with the action via OF → R. This functor is defined on the level of displays,
i.e., we construct from the display P of X the display P ′ of X ′ . The third entry λ′ is a bilinear
form of displays
P ′ × P ′ → LR , (1.3.2)
where L is the display of a Lubin-Tate group associated to the local field F . The bilinear form
(1.3.2) is given by a bilinear form of W (R)-modules P ′ × P ′ → LR which satisfies additional
conditions. We call λ′ a polarization with values in the Lubin-Tate group or a polarization in the
sense of Faltings, comp. [11]. Note that because of the values for the height and the dimension
of X ′ , there cannot exist a polarization in the usual sense on X ′ .
The second functor is the Ahsendorf functor AOF /Zp ,R from [1]. This functor associates to
a display with strict OF -action a display over the ring of relative Witt vectors WOF (R). The
Ahsendorf functor is the analogue for displays of the Drinfeld functor which associates to a
Cartier module of a p-divisible group with strict OF -action its relative Cartier module, cf. [9,
§2]. We give here a new construction of the Ahsendorf functor which is more elementary and uses
as an intermediate step the Lubin-Tate frames as in Mihatsch [22]. However, we make a special
choice for these frames, cf. Definition 3.3.8. Then the Ahsendorf functor becomes compatible
with polarizations, as stated in the following Theorem.
Theorem 1.3.2. Let R be an OF -algebra such that p is nilpotent in R. The Ahsendorf functor
is a functor
   
W(R)-displays
AOF /Zp ,R : −→ WOF (R)-displays .
with strict OF -action
It induces an equivalence of categories
   
nilpotent W(R)-displays
AOF /Zp ,R : −→ nilpotent WOF (R)-displays .
with strict OF -action
Let P1 and P2 be W (R)-displays over R with a strict OF -action. We denote by P1,a and P2,a
their images by the Ahsendorf functor AOF /Zp ,R . Then there is a natural homomorphism between
groups of bilinear forms of displays,
BilOF -displays (P1 × P2 , LR ) −→ BilWOF (R)-displays (P1,a × P2,a , WOF (R)(π ef /pf )).
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 9

If the dual (P1,a )∨ of P1,a and P2,a are nilpotent WOF (R)-displays, then this homomorphism is
an isomorphism.
Here the ring WOF (R) with its Frobenius F ′ and Verschiebung V ′ is regarded as a display.
For a unit ε ∈ WOF (R)× we obtain a twisted display WOF (R)(ε) by replacing F ′ by εF ′ and V ′
by V ′ ε−1 . The underlying module of this display remains WOF (R). We apply this construction
to the image of the unit π ef /pf ∈ OF× in WOF (R)× .
Let us return to Theorem 1.3.1. It is more honest to formulate this theorem as follows. We
fix an extension ϕ̆0 : OF̆ −→ OĔ of ϕ0 : OF −→ OE . Let M be the relative RZ-space over
Spf OF̆ from [17]. It parametrizes tuples (X ′ , ι′ , λ′ , ρ′ ), where X ′ is a strict formal OF -module of
relative height 4 and dimension 2, and where ι′ is an OK -action on X, and where λ′ is a relative
polarization compatible with ι′ , which is principal if K/F is ramified and almost principal if K/F
is unramified. Also, ρ′ is a framing of height zero with the framing object (X′ , ι′X′ , λ′X′ ) which
is obtained from (X, ιX , λX ) by the contracting functor. Then the contracting functor defines a
morphism
Mr −→ M ×Spf OF̆ Spf OĔ . (1.3.3)
Using Theorem 1.3.2, we prove that this is an isomorphism of formal schemes over Spf OĔ .
Moreover, we prove that X ′ is the p-divisible group of a special formal OD -module in the sense of
[9]. Therefore, by Drinfeld’s local theorem [9], the right hand side of (1.3.3) has an interpretation
in terms of the Drinfeld halfplane Ω b F . This is the main result of [17], for which we give a new
proof, after it was already reproved by Kirch [14]. Theorem 1.3.1 follows.
Let us now put the local results of this paper in perspective. We address in our special case
the general problem of identifying a basic Rapoport-Zink space associated to the pair (G, {µ})
with a twist of the basic Rapoport-Zink space associated to the pair (G, {µ′ }), where µ′ differs
from µ by a central character, cf. the Introduction of [28]. This problem is also addressed by
Scholze in [29, Chap. 23], in both the case considered here and in the fake Drinfeld case of [28].
As mentioned above, Scholze’s proof uses in an essential way our formulation of the local moduli
problem, via the theory of local models (and hence implicitly the linear algebra lemma [28, Lem.
4.9]). One of the main reasons that we are successful in constructing the contracting functor in
the case treated here is that here we are able to develop a good understanding of the Kottwitz
condition (KCr ), even in unequal characteristic. Our failure to do the same in the fake Drinfeld
case is the essential reason that in [28] we only succeeded in defining the contracting functor
in the special fiber. The contracting functor is an expression of the exceptional isomorphism
between the quasi-split special unitary group in two variables and the special linear group in
two variables. We restricted ourselves here to the case of curves; it would have been possible to
prove a higher-dimensional version where the uniformizing space ΩFv is replaced by a product
of such spaces, comp. [18] and [27, §6].
1.4. Layout of the paper. We now explain the lay-out of the paper. The whole paper, with
the exception of section 7, is devoted to the local theory. In section 2 we explain in detail the
definition of the formal moduli spaces of (polarized) p-divisible formal groups, including the
Kottwitz conditions relevant here and the Eisenstein conditions; in particular, Subsection 2.6
contains the detailed statements of our main local results. Section 3 summarizes the relevant facts
on relative Dieudonné theory and relative display theory. The most important fact proved in this
section is the relation established by the Ahsendorf functor between the Lubin-Tate display and
the relative multiplicative display. In section 4 we first consider the relation between the Kottwitz
condition and the Eisenstein condition; this is used in the rest of the section to construct the
contracting functor. More precisely, we first consider the first step in its construction which we
call the pre-contracting functor, cf. above. After this, we complete the second step in the case of
a special generalized CM-type. In the final subsection of section 4, we consider the second step
in the case of a banal generalized CM-type. Section 5 is devoted to an alternative proof of the
main result of [17], based on the theory of displays. In section 6 we prove the main local results,
namely Theorem 1.3.1 and its banal counterpart. In the appendix, section 8, we give the correct
version of the sign factor of [18] by defining the adjusted invariant of a CM-triple of generalized
CM-type r of even rank n, and investigate its behaviour under the contracting functor. Section
7 deduces the global results from the local theory.
10 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

1.5. Acknowledgements. We are grateful to P. Scholze for very helpful discussions. In par-
ticular, he helped us locate the mistake in the definition of the invariant of a CM-triple (we had
discovered a discrepancy caused by this mistake in 2014, but it took us 10 months to locate the
error (e-mail from P. Scholze of 5 Oct. 2015)). We also thank the anonymous referee of [24]
whose report gave us the idea of defining the p-adic étale sheaf in Corollary 7.4.8.
We also acknowledge the hospitality of the MSRI during the fall of 2014 when the work on
this paper was begun.

1.6. Notation. • If R and R′ are Zp -algebras, we often write R ⊗ R′ for R ⊗Zp R′ . Also, we
often write X ⊗A B for X ×Spec A Spec B.
• If F is a finite extension of Qp , we write F̆ for the completion of a maximal unramified extension,
and F t for the maximal subfield unramified over Qp . We write d = ef , where d = [F : Qp ] and
f = [F t : Qp ] and e = [F : F t ]. We denote by OF , resp. OF t , resp. OF̆ the rings of integers.
• Let V be an C/R-anti-hermitian vector space. The signature of V is (a, b) if the anti-hermitian
form is equivalent to diag(i(a) , i(b) ), where i is the imaginary unit.
• Let F be a finite extension of Qp and let K/F be a quadratic extension. Let V be a K-vector
space, equipped with an alternating Q-bilinear form ψ : V × V −→ Q satisfying (1.2.1). Let Λ
be a OK -lattice in V . Then the dual OK -lattice is Λ∨ = {x ∈ V | ψ(x, y) ∈ Zp for all y ∈ Λ}.
The lattice Λ is called self-dual if Λ = Λ∨ ; it is called almost self-dual if Λ is contained in Λ∨
with colength one.
• If O is a discrete valuation ring with uniformizer π, we write NilpO for the category of O-
algebras R such that π is locally on Spec R nilpotent. Similarly, we denote by (Sch/ Spf O) the
category of O-schemes such that πOS is a locally nilpotent ideal sheaf.
• Given modules M and N over a ring R, we write M ⊂r N to indicate that M is an R-submodule
of N of finite colength r.
Warning. It is customary to denote a finite extension of Qp and the Frobenius by the same
symbol F . This should not lead to confusions.

2. Main local statements

In this section we formulate our main results in the local theory. We fix a prime number p
and an algebraic closure Qp of Qp . Let F be a finite field extension of Qp , with residue class
field κF . We set d = [F : Qp ], f = [κF : Fp ] and define e through d = ef . We let K/F be an
étale algebra of degree 2. We denote the non-trivial automorphism of Gal(K/F ) by a 7→ ā.
In the case where K/F is a ramified extension of local fields (ramified case) we choose a prime
element Π ∈ OK such that Π̄ = −Π. Then π = −Π2 is a prime element of F . In the case where
K/F is unramified extension of local fields (unramified case) or K ∼ = F ×F (split case) we choose
a prime element π ∈ F and we set Π = π.
Let Φ = ΦK = HomQp -Alg (K, Qp ) be the set of algebra homomorphisms.

2.1. Special and banal local CM-types. Let r be a generalized local CM-type of rank 2
(relative to K/F ) in the sense of [18, section 5], i.e., a function
r : Φ −→ Z>0 , ϕ 7−→ rϕ , (2.1.1)
such that rϕ + rϕ = 2 for all ϕ ∈ Φ. Here ϕ(a) = ϕ(a), where a 7→ a is the non-trivial
automorphism of K over F . The corresponding reflex field E = E(r) is the subfield of Qp fixed
by
Gal(Qp /E) := {τ ∈ Gal(Qp /Qp ) | rτ ϕ = rϕ , ∀ϕ}.
Let OE be the ring of integers of E.
When we fix an embedding ϕ0 : F −→ Qp , we denote by ϕ0 , ϕ0 the two extensions of ϕ0 to
K (by abuse of notation).
Definition 2.1.1. A local CM-type r of rank 2 is called special relative to the choice of embed-
ding ϕ0 : F −→ Qp if K/F is a field extension and
rϕ0 = rϕ0 = 1, and rϕ ∈ {0, 2}, for all ϕ ∈ Φ \ {ϕ0 , ϕ0 }.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 11

It is called banal non-split if K/F is a field extension and rϕ ∈ {0, 2}, for all ϕ ∈ Φ. It is called
banal split if K ≃ F ⊕ F and rϕ ∈ {0, 2}, for all ϕ ∈ Φ.
From now on, we will assume r to be either special (relative to a fixed choice of ϕ0 ) or banal
(non-split or split). We will consider p-divisible groups X with an action of OK over OE -schemes
S. We will want to impose certain conditions on the induced action of OK on Lie X.

2.2. The Kottwitz and the Eisenstein conditions. Let S be an OE -scheme, and let L be
a locally free OS -module, equipped with an action
ι : OK −→ EndOS L.
of OK .
We say that (L, ι) satisfies the Kottwitz condition (KCr ) relative to r if the identity of poly-
nomials with coefficients in OS holds
Y

char(T, ι(a)|L) = i (T − ϕ(a))rϕ , for all a ∈ OK , (2.2.1)
ϕ∈Φ

where i : OE −→ OS is the structure homomorphism (compare [28]).

We denote by F t ⊂ F the maximal subextension which is unramified over Qp . If K is the
field we use the same notation for K, and in the split case K ∼= F × F we set K t = F t × F t . We
t
set Ψ = ΨK = HomQp -Alg (K , Qp ). We call ψ ∈ Ψ banal if rϕ ∈ {0, 2} for each ϕ ∈ Φ such that
ϕ | ψ. If this is not the case we call ψ special. We use the notation
Φψ = {ϕ ∈ Φ | ϕ|K t = ψ}, ψ ∈ Ψ. (2.2.2)
We denote by E ′ be the compositum of E with the normal closure of the image of K t in Q̄p
under any embedding ψ ∈ Ψ. We remark that this normal closure is isomorphic to K t via any
ψ ∈ Ψ. In particular E ′ /E is an unramified extension of local fields.
Let S be an OE -scheme. Let α : S −→ Spec OE ′ be a morphism of OE -schemes. Then α
gives rise to an isomorphism of OK t ⊗Zp OS algebras
M
OK t ⊗Zp OS = OS , (2.2.3)
ψ∈Ψ

where the action of OK t on the ψ-th factor is via ψ. Hence for a locally free OS -module L with
action by OK , we obtain a decomposition into locally free OS -modules,
M
L= Lψ . (2.2.4)
ψ∈Ψ

If (L, ι) satisfies the Kottwitz condition we obtain from (2.2.1) applied to a ∈ OK t that
X
rank Lψ = rϕ (2.2.5)
ϕ∈Φψ

We say that (L, ι) satisfies the rank condition (RCr ), if (2.2.5) is satisfied for all ψ. The rank con-
dition does not depend on the α chosen above because a second α′ differs from α by an automor-
phism of E ′ over E if S is connected. If there is no α we use base change Spec OE ′ ×Spec OE S −→ S
to define the condition (RCr ). This agrees with the old definition if α exists.
We consider a pair (L, ι) that satisfies (RCr ). Then we will define the Eisenstein condition
(ECr ) (this definition is analogous to [28, section 2], but different). We introduce the notation
Aψ = {ϕ : K −→ Qp | ϕ|K t = ψ, and rϕ = 2}
(2.2.6)
Bψ = {ϕ : K −→ Qp | ϕ|K t = ψ, and rϕ = 0}.
We note that under the action of the non-trivial automorphism of K/F ,
Āψ = Bψ . (2.2.7)
Also, let aψ = |Aψ | and bψ = |Bψ |.
With this notation we may rewrite the rank condition (RCr )
rank Lψ = 2aψ + ǫψ , (2.2.8)
12 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

where 
 0, if ψ is banal

ǫψ = 1, if ψ is special and K/F is unramified

 2, if ψ is special and K/F is ramified.

In the case where K/F is ramified we have K t = F t , [K : K t ] = 2e, and for each ψ ∈ Ψ
ϕ|ψ ⇒ ϕ̄ | ψ.
Therefore aψ = bψ , and the rank condition reads, in the ramified case,
rank Lψ = 2e,
regardless of whether r is banal or not.
Consider the Eisenstein polynomial E(T ) of Π in OK t [T ]. We consider the image Eψ (T ) of
E(T ) in Qp [T ] under ψ, for ψ ∈ Ψ. In Qp [T ] this has a decomposition into linear factors,
Y
Eψ (T ) = (T − ϕ(Π)). (2.2.9)
ϕ∈Φψ

We define Y Y
EAψ (T ) = (T − ϕ(Π)), EBψ (T ) = (T − ϕ(Π)). (2.2.10)
ϕ∈Aψ ϕ∈Bψ

The action of Gal(Qp /E ′ ) stabilizes the corresponding subsets in the index set on the right hand
sides of (2.2.9) and (2.2.10). Therefore all three polynomials lie in OE ′ [T ].
If r is special we fix an embedding ϕ0 : K −→ Q̄p such that rϕ0 = 1. We denote by ψ0
the restriction of ϕ0 to K t . In the ramified case we have ψ0 = ψ¯0 and in the unramified case
ψ0 6= ψ¯0 .
We define Sψ by the following factorization in OE ′ [T ],
Eψ (T ) = Sψ (T ) · EAψ (T ) · EBψ (T ). (2.2.11)
Hence


 1, if ψ is banal

 (T − ϕ (Π))(T − ϕ (Π)),
0 0 if ψ = ψ0 and K/F is ramified
Sψ (T ) =

 T − ϕ0 (Π), if ψ = ψ0 and K/F is unramified


T − ϕ0 (Π), if ψ = ψ 0 and K/F is unramified.
Now using the structure morphism OE ′ −→ OS , each of the three factors in (2.2.11), when
evaluated on Π, defines an endomorphism of the OS -module Lψ . These endomorphisms are
denoted by EAψ (ι(Π)|Lψ ), resp. EBψ (ι(Π)|Lψ ), resp. Sψ (ι(Π)|Lψ ).
We say that (L, ι) satisfies the Eisenstein conditions if (RCr ) is fulfilled and if for each ψ


Sψ · EAψ (ι(Π) | Lψ ) = 0,
4−[K t :F t ]
^ (2.2.12)


EAψ (ι(Π) | Lψ ) = 0.
In the case where ψ is banal the first condition says
EAψ (ι(Π)|Lψ ) = 0, for all ψ ∈ Ψ. (2.2.13)
and the second condition follows from the first.
The Eisenstein conditions do not depend on the OE -morphism α : S −→ Spec OE ′ . Indeed,
if S is connected, any other choice of α differs by an automorphism ρ ∈ Gal(E ′ /E). In the
decomposition (2.2.4) Lψ is then replaced by Lρψ and Eψ is replaced by ρ(Eψ ) = Eρψ . Here the
last identity holds by the definition of the reflex field E. Therefore changing α does not change the
Eisenstein conditions (ECr ). If there exists no α, we use base change Spec OE ′ ×Spec OE S −→ S
to define the condition (ECr ). The same arguments apply to the condition (KCr ).
We first note the following statement.
Proposition 2.2.1. Let S be an OE -scheme and L a locally free OS -module with an OK -action
ι : OK −→ EndOS (L).
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 13

(i) The Eisenstein conditions (ECr ) are independent of the uniformizer Π.

(ii) When K/Qp is unramified, the Eisenstein conditions (ECr ) are implied by the Kottwitz
condition (KCr ). The same conclusion holds if F = Qp and K/F is ramified.
(iii) When S is an E-scheme, the Eisenstein conditions (ECr ) hold automatically.
Proof. Let us prove (i). Let Π′ be another uniformizer. It is enough to show that the elements
of OK ⊗OK t ,ψ OE ′ ,
EAψ (Π ⊗ 1), EAψ (Π′ ⊗ 1), resp.
Sψ (Π ⊗ 1)EAψ (Π ⊗ 1), Sψ (Π ⊗ 1)EAψ (Π′ ⊗ 1),
′

differ by a unit in OK ⊗OK t ,ψ OE ′ . Indeed, let Ẽ ′ be the normalization of E ′ in Q̄p . Since

OK ⊗OK t ,ψ OE ′ → OK ⊗OK t ,ψ OẼ ′ is a flat extension of local rings, we can replace E ′ by Ẽ ′ .
By the definitions (2.2.10) and (2.2.11), it suffices to show that the elements Π ⊗ 1 − 1 ⊗ ϕ(Π)
and Π′ ⊗ 1 − 1 ⊗ ϕ(Π′ ) differ by a unit in OK ⊗OK t ,ψ OE ′ . But by [25, Lem. 6.11] the elements
Π ⊗ 1 − 1 ⊗ Π and Π′ ⊗ 1 − 1 ⊗ Π′ of OK ⊗OK t OK differ by a unit, whence the assertion.
Now we prove (ii). Let us only treat the case where r is special; the banal case is similar.
When K = K t is unramified over Qp , then E(T ) = T − π is a linear polynomial. Furthermore,
Aψ has at most one element for ψ ∈ / {ψ0 , ψ 0 }, and Aψ0 = Aψ 0 = ∅. Let ψ ∈
/ {ψ0 , ψ 0 }. If Aψ = ∅,
then Lψ = (0) and the Eisenstein condition relative to the index ψ is empty; if Aψ has one
element, the Eisenstein condition relative to the index ψ is just equivalent to the definition of
the ψ-th eigenspace in the decomposition (2.2.4). Something analogous applies to the indices
ψ0 , ψ 0 . The case when F = Qp is handled in the same way.
Finally we prove (iii). Let K̃ be the normal closure of K in Qp . It suffices to prove the
assertion after replacing S by its base change S ×Spec E Spec K̃. Then we have a decomposition
M
OK ⊗Zp OS = OS .
ϕ∈Φ

Correspondingly, we have L = ⊕Lϕ , and the endomorphism ι(Π) is diagonal with respect to this
decomposition, with entries ϕ(Π)idLϕ . It is easy to see that (KCr ) is equivalent to the condition
rank Lϕ = rϕ , ∀ϕ ∈ Φ. (2.2.14)
The Eisenstein conditions (ECr ) involve endomorphisms of L which are products of endomor-
phisms of the form (ι(Π)|Lϕ − ϕ(Π)idLϕ ) ⊕ϕ′ 6=ϕ idLϕ′ . From this, the conditions follow triv-
ially. 
Let us make the Eisenstein conditions more explicit in the case where r is special. For this,
we distinguish between the case when K/F is ramified and the case when K/F is unramified.
Let S be an OE ′ -scheme and L be a locally free OS -module satisfying (RCr ).
• K/F ramified. In this case, we have K t = F t , and ψ = ψ for all ψ ∈ Ψ. Hence, (2.2.7) implies
in this case (
e, if ψ 6= ψ0
aψ = b ψ = (2.2.15)
e − 1, if ψ = ψ0 .
We have
Sψ0 (T ) = (T − ϕ0 (Π))(T − ϕ0 (Π)).
The Eisenstein conditions become in this case


Sψ0 · EAψ0 (ι(Π)|Lψ0 ) = 0,
3
^
(2.2.16)
EAψ0 (ι(Π)|Lψ0 ) = 0,
EAψ (ι(Π)|Lψ ) = 0, for all ψ 6= ψ0 .
• K/F unramified. In this case, [K t : F t ] = 2, and ψ 6= ψ for all ψ ∈ Ψ. Furthermore, aψ = bψ
and X
aψ = e(f − 1), aψ0 + aψ = e − 1. (2.2.17)
0
ψ∈Ψ\{ψ0 ,ψ 0 }
14 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

In this case, the Eisenstein conditions become

Sψ0 · EAψ0 (ι(π)|Lψ0 ) = 0,
2
^

EAψ0 (ι(π)|Lψ0 ) = 0,


Sψ 0 · EAψ (ι(π)|Lψ 0 ) = 0, (2.2.18)
0
2
^

EAψ (ι(π)|Lψ 0 ) = 0,
0

EAψ (ι(π)|Lψ ) = 0, for all ψ 6= ψ0 , ψ 0 .

2.3. Local CM-pairs and CM-triples. Let S be an OE -scheme such that p is locally nilpotent
i.e. a scheme over Spf OE . A local CM-pair of type r is a pair (X, ι) such that X is a p-divisible
group of height 4d and dimension 2d and ι is an Zp -algebra homomorphism
ι : OK −→ End X
such that the rank condition (RCr ) is satisfied for the induced action of OK on Lie X. In the
split case OK = OF × OF we require moreover that in the induced decomposition X = X1 × X2
each factor is a p-divisible group of height 2d.
Later we will introduce displays P, and these have a Lie algebra Lie P, cf. Definition 3.1.4.
Therefore, we can also speak of local CM-pairs (P, ι) of type r, where P is a display over S, cf.
section 3.
Let S = Spec k be a perfect field of characteristic p which is endowed with an OE ′ -algebra
structure. In this case, a display in the same thing as a Dieudonné module P = (P, F, V ), where
P is a finitely generated free module over the ring of Witt vectors W (k). If P is the Dieudonné
module of X, there is a canonical isomorphism of k-vector spaces Lie X ∼ = P/V P .
Via ι we regard P as a OK ⊗Zp W (k)-module. The homomorphisms
ψ : OK t −→ OE ′ −→ k, (2.3.1)
ψ ∈ Ψ lift uniquely to homomorphisms
ψ̃ : OK t −→ W (k). (2.3.2)
We obtain a ring isomorphism
Y
OK ⊗Zp W (k) = OK ⊗O W (k).
K t ,ψ̃
ψ∈Ψ

This induces a decomposition

P = ⊕ψ∈Ψ Pψ . (2.3.3)
More explicitly
Pψ = {x ∈ P | ι(a)x = ψ̃(a)x, for a ∈ OK t }.
Let us denote by σ the Frobenius automorphism of W (k). The operators F and V on P
induce σ-linear maps
F : Pψ −→ Pσψ , V : Pσψ −→ Pψ . (2.3.4)
Here σψ denotes the composite of (2.3.1) with the absolute Frobenius of k.
Lemma 2.3.1. Let (P, ι) be local CM-pair of type r over a perfect field k. Then P is a free
OK ⊗Zp W (k)-module of rank 2.
Proof. Since F V = p it follows that
rankOK ⊗O ,ψ̃ W (k)
Pψ = rankOK ⊗O ,σψ̃ W (k)
Pσψ . (2.3.5)
Kt Kt

Since rankW (k) P = 4d, and by the extra condition in the split case, this implies that the common
rank of (2.3.5) is 2. This proves the Lemma. 
To each local CM-pair (X, ι) we define the conjugate dual (X ∨ , ι∧ ). Here X ∨ is the dual
p-divisible group of X but we change the action dual to ι by the conjugation of K/F , i.e.,
ι∧ (a) = ι∨ (ā). We will denote the conjugate dual simply by X ∧ .
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 15

Lemma 2.3.2. The conjugate dual of a CM-pair (X, ι) of type r is again a local CM-pair of
type r. If (X, ι) satisfies the Kottwitz condition (KCr ), resp., the Eisenstein conditions (ECr ),
then so does its conjugate dual.

Proof. For the first assertion, we may assume that we are over an algebraically closed field. We
use the Dieudonné module P. We set

P ∨ = HomW (k) (P, W (k)).

We use the canonical pairing

h , i : P × P ∨ −→ W (k). (2.3.6)
The operators F and V on the dual Dieudonné module P ∨ are defined by the equations

h V x, V x∨ i = pσ −1 (h x, x∨ i), x ∈ P, x∨ ∈ P ∨
σ(h V x, x∨ i) = h x, F x∨ i.

One of these equations implies the other. It follows that V P/pP ⊂ P/pP and V P ∨ /pP ∨ ⊂
P ∨ /pP ∨ are orthogonal complements with respect to the non-degenerate pairing of k-vector
spaces,
P/pP × P ∨ /pP ∨ −→ k.
If we use the action ι∧ , we write for the decomposition (2.3.3)

P ∨ = ⊕ψ∈Ψ Pψ∧ = P ∧ .

Then Pψ1 and Pψ∧2 are for ψ1 6= ψ̄2 orthogonal with respect to (2.3.6) and

h , i : Pψ × Pψ̄∧ −→ W (k). (2.3.7)

is a perfect pairing. The k-vector spaces V Pσψ /pPψ and V Pσ∧ψ̄ /pPψ̄∧ are orthogonal complements
with respect to the induced non-degenerate k-bilinear form

Pψ /pPψ × Pψ̄∧ /pPψ̄∧ −→ k.

Let us assume that K/F is unramified or split. In this case Lemma 2.3.1 implies rankW (k) Pψ =
2e and by (2.3.7) rankW (k) Pψ∧ = 2e. Since P satisfies (RCr ) we find by the orthogonality above
X X X
rankk Pψ̄∧ /V Pσ∧ψ̄ = 2e − rankk Pψ /V Pσψ = 2e − rϕ = (2 − rϕ ) = rϕ̄ .
ϕ|ψ ϕ|ψ ϕ|ψ

This shows that the conjugate dual satisfies (RCr ). The case K/F ramified is similiar.
For the proof of the assertion concerning (KCr ), we refer to Proposition 4.2.13. For the proof
of the assertion concerning (ECr ), we refer to Corollary 4.2.8 in the case when K/F is unramified
or split, resp., Corollary 4.2.12 when K/F is ramified. 

The notion of a local CM-triple of type r over S was introduced in [18]. This is a triple
(X, ι, λ), where (X, ι) is a local CM-pair of type r and λ : X −→ X ∨ is an anti-symmetric
isogeny (also called a polarization) such that the corresponding Rosati involution induces the
non-trivial automorphism on K/F . In particular λ induces a morphism of local CM-pairs

λ : (X, ι) −→ (X ∧ , ι∧ ).

In the present paper, we will also say that (X, ι, λ) is a polarized local CM-pair (of type r). We
call the polarized local CM-pair (X, ι, λ) principal if Ker λ = 0; we call it almost principal if
Ker λ ⊂ X[ι(π)] and Ker λ has order p2f . We will distinguish these two cases by attaching the
integer h = 0 to the principal case, and h = 1 to the almost principal case, i.e., height λ = 2f h.
16 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

2.4. The invariant of a local CM-triple. Let K/F be a field extension. We recall from [18]
the definition of the invariant of a CM-triple, in a slightly more general context.
Let k be an algebraically closed field of characteristic p and let W (k) be the ring of Witt
vectors. We set W (k)Q = W (k) ⊗ Q. Let (M, F, V ) be a Dieudonné module of height 4d and
dimension 2d which is endowed with a Zp -algebra homomorphism
ι : K −→ End0 (M, F, V ).
We set MQ = M ⊗ Q. We assume that M is endowed with a non degenerate alternating bilinear
form
β : MQ × MQ −→ W (k)Q
Let us denote by σ the Frobenius automorphism of W (k). We require the following properties:
β(V x, V y) = pσ −1 (β(x, y)), x, y ∈ MQ
β(ι(a)x, y) = β(x, ι(ā)y), a ∈ K.
We will associate to such a set of data (M, ι, β) an invariant inv(M, ι, β) ∈ {±1}. We set
Ψ = HomQp -Alg (K t , W (k)Q ).
The ring OK ⊗Zp W (k) decomposes
Y
K ⊗Zp W (k) = K ⊗O t ,ψ̃ W (k). (2.4.1)
K
ψ

If ξ = (ξψ ) is an element of (2.4.1). Then we set

X
ordK⊗W (k) ξ = ordp NmK/Qp ξ = ordΠ ξψ ∈ Z. (2.4.2)
ψ

The Frobenius homomorphism σ acts via the second factor on K ⊗Zp W (k). The σ-conjugacy
class of an element ξ ∈ (K ⊗ W (k))× is uniquely determined by ordK⊗W (k) ξ.
We view MQ as a K ⊗Zp W (k)-module and suppress the notation ι. This is a free module of
rank 2. We define an anti-hermitian form κ = κβ ,
κ : MQ × MQ −→ K ⊗Zp W (k),
on the K ⊗Zp W (k)-module MQ by the formula
TrK/Qp (aκ(x, y)) = β(ax, y), x, y ∈ MQ , a ∈ K ⊗ W (k). (2.4.3)
Then κ satisfies
κ(V x, V y) = pσ −1 (κ(x, y)). (2.4.4)
2
V2
We write ∧ MQ := K⊗W (k) MQ for the exterior product as a K ⊗Zp W (k)-module. This is
a free K ⊗Zp W (k)-module of rank 1. According to (2.4.1) we have decompositions
M
MQ = MQ,ψ
ψ
M 2
^

∧2 MQ = MQ,ψ .
ψ
K⊗K t ,ψ WQ (k)

We choose an isomorphism ∧2 MQ ∼
= K ⊗Zp W (k). Then we can write
∧2 V (z) = γσ −1 (z).
We have
ordK⊗W (k) ∧2 V = ordp NmK/Qp det K⊗Zp W (k) (V |MQ )
= ordp detW (k) (V |MQ ) = dim M = 2d.
Therefore we find ordK⊗W (k) γ = 2d. Since ordK⊗W (k) p = 2d, the elements p, γ ∈ K ⊗Zp W (k)
are in the same σ-conjugacy class by the remark after (2.4.2). We conclude that there is a
generator x ∈ ∧2 MQ such that
∧2 V (x) = px. (2.4.5)
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 17

Note that the last equation is equivalent with ∧2 F (x) = px. Any other generator with this
property has the form ux, where u ∈ K × . We consider the hermitian form
h = ∧2 κ : ∧2 MQ × ∧2 MQ −→ K ⊗ W (k).
One deduces that h(x, x) is an element of F ⊂ K ⊗W (k) which is 6= 0 because β is non-degenerate
by assumption. We denote by
dK/F (M, ι, β) ∈ F × / NmK/F K × (2.4.6)
the class of h(x, x). This class is called the discriminant of (M, ι, β) and is independent of the
choice of x.
Let a ∈ F . Then aκ is again an anti-hermitian form which satisfies (2.4.4). We can replace
κ by aκ in the definition of (2.4.6) without changing the discriminant. We denote by
inv(M, ι, β) ∈ {±1} (2.4.7)
the image of dK/F (M, ι, β) by the canonical isomorphism F × / NmK/F K × ≃ {±1}.
Let r be a local CM-type of rank 2. Let E be the reflex field. Let OE −→ k an algebra
structure of the algebraically closed field k. Let (X, ι, λ) be a local triple of CM-type r over k.
Let (M, ι, β) be the associated Dieudonné module with its polarization β. Then we set
inv(X, ι, λ) := inv(M, ι, β).
For CM-triples of CM-type r, we use also the adjusted invariant invr (X, ι, λ) = invr (M, ι, β), cf.
section 8.2. In the case at hand we have

(−1)d−1 inv(M, ι, β), for r special,
invr (M, ι, β) = (2.4.8)
(−1)d inv(M, ι, β), for r banal.

2.5. Uniqueness of framing objects. In this subsection, we discuss the existence and unique-
ness of framing objects that are used in the formulation of the formal moduli problems. The
proofs of these statements are given later in the paper.
Let r be a generalized local CM-type of rank 2 for K/F . Let k be an algebraic closure of the
residue field κE of OE . Consider CM-triples (X, ι, λ) over k which satisfy (KCr ) and (ECr ).
(i) Assume that r is special. If K/F is ramified, then a local CM-triple of type r over k as above
such that the polarization is principal and with r-adjusted invariant −1 is isoclinic. When K/F
is unramified, then a local CM-triple of type r over k as above such that the polarization is almost
principal has r-adjusted invariant −1 and is isoclinic. In either case, any two such CM-triples
are isogenous by a OK -linear quasi-isogeny of height zero that preserves the polarizations.
Furthermore, the group of OK -linear self-isogenies of such a local CM-triple, preserving the
polarization, can be identified with the unitary group of the split K/F -hermitian space C of
dimension 2.
The assertions concerning slopes follow from Corollary 4.3.3. The uniqueness assertion is in
the ramified case the content of Proposition 5.2.12, and in the unramified case of Proposition
5.3.6. The last part of the assertion follows from the fact that the contraction functor is an
equivalence of categories.
(ii), a) Let r be banal non-split. Any local CM-triple of type r over k as above is isoclinic.
The group of OK -linear self-isogenies of such a local CM-triple, preserving the polarization, can
be identified with the unitary group of a K/F -hermitian space C of dimension 2. When K/F is
unramified and the polarization is principal, then the hermitian space C is split and the r-adjusted
invariant is 1; when K/F is ramified and the polarization is principal, then the hermitian space
C is non-split and the r-adjusted invariant is −1; when K/F is unramified and the polarization
is almost principal, then the hermitian space C is non-split and the r-adjusted invariant is −1.
The case K/F ramified and almost principal polarization does not occur. Any two CM-triples
with the same r-adjusted invariant are isogenous by a OK -linear quasi-isogeny of height zero that
preserves the polarizations.
The assertions concerning slopes follow from Corollary 4.3.3. The uniqueness assertion is the
content of Proposition 4.5.14. Similar arguments apply to the banal split case.
18 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

(ii), b) Let r be banal split. Then the p-divisible group underlying a local CM-triple of type r
is the direct product of two isoclinic p-divisible groups of slope λ, resp. 1 − λ, where λ depends
only on r. Any two local CM-triples of type r over k are isogenous by a OK -linear quasi-isogeny
of height zero that preserves the polarizations. The group of OK -linear self-isogenies of such a
local CM-triple, preserving the polarization, can be identified with ResF/Qp (GL2 ).
Here is a chart for the various possibilities.

type r K/F invr polarization type C

special ramified −1 principal split
special unramified −1 almost principal split
banal ramified −1 principal non-split
banal unramified 1 principal split
banal unramified −1 almost principal non-split
banal split 1 principal GL2 /F

Table 1. Framing objects

Remark 2.5.1. The statement (i) above generalizes [18, Prop. 5.4]. However, the proof of
the uniqueness assertion given there is incomplete. Note that for a local CM-type of the first
kind in the sense of loc. cit. we have imposed F = Qp ; therefore the condition in loc. cit. that
ε = inv(X, ι, λ) = −1 implies that the associated hermitian space (C, h) is split (in this case the
r-adjusted invariant coincides with the invariant).
Remark 2.5.2. The statement (i) is closely related to the fact that B(G, {µ}) has only one
element, cf. [16], §6. Here G = ResF/Qp (GU) is the linear algebraic group over Qp associated to
the group of unitary similitudes of the non-split hermitian space of dimension 2 over K, and {µ}
is the conjugacy class of cocharacters with component (1, 0) for ϕ0 and central component for
ϕ 6= ϕ0 . In fact, it seems that the essential contents of the calculations in section 8.3 is to show
that the Frobenius element of a local CM-triple of type r with r-adjusted invariant −1 over k
defines an element in B(G, µ).
2.6. Formal moduli spaces. In this subsection we are going to define RZ-spaces of formal
local CM-triples, and formulate our main results about them. We fix K/F as before.
First let r be special, so that K/F is a field extension. We fix a local CM-triple (X, ιX , λX ) of
type r over κ̄E as in (i) of subsection 2.5 (a framing object). We assume that if K/F is ramified,
then λX is principal and that, if K/F is unramified, then λX is almost principal. Then, in either
case, the r-adjusted invariant equals −1. We identify κ̄E with the residue class field of OĔ , the
ring of integers in the completion of the maximal unramified extension of E. Let (Sch/OĔ ) be
the category of OĔ -schemes S such that the ideal sheaf πOS is locally nilpotent.
Definition 2.6.1. We set h = 0 if K/F is ramified, and h = 1 if K/F is unramified.
We define a functor MK/F,r on (Sch/OĔ ). A point of MK/F,r (S) consists of an isomorphism
class of the following data:
(1) Two local CM-pairs (X0 , ι0 ), (X1 , ι1 ) of CM-type r over S which satisfy the Eisenstein
conditions (ECr ) relative to a fixed uniformizer π of F and the Kottwitz condition (KCr ).
(2) Two isogenies of p-divisible OK -modules
X0 −→ X1 −→ X0 ,
which have both height 2f h and such that the composite is ι0 (π)h idX0 .
(3) An isomorphism of p-divisible OK -modules
∼
κ : X1 −→ X0∧ .
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 19

∼
We require that the composite λ : X0 −→ X1 − → X0∧ is a polarization of X0 , i.e., this map
is anti-symmetric, and that this polarization is principal when K/F is ramified, and almost
principal when K/F is unramified.3
(4) A quasi-isogeny of height zero of p-divisible OK -modules
ρX : X0 ×S S̄ −→ X ×Spec κ̄E S̄,
such that the pullback quasi-isogeny ρ∗ (λX ) differs from λ|X0 ×S S by a scalar in F × , locally on
S. Here S = S ⊗OĔ κ̄E . We call ρX a framing.
We denote the data above simply by (X, ι, λ, ρ). Another datum (X ′ , λ′ , ρ′ ) defines the same
∼ ∼
point of MK/F,r (S) iff there are OK -isomorphisms X0 − → X0′ and X1 − → X1′ which commute
∼
with the data (2) and (4) above. This implies that the isomorphism X0 − → X0′ respects the
×
polarizations up to a factor in OF .
To ease the notation we write Mr = MK/F,r . If R is a p-adic OĔ -algebra we set Mr (R) =
limn Mr (Spec R/pn R).
←−
It follows by the methods of [27] that Mr is representable by a formal scheme which is locally
formally of finite type over Spf OĔ . Let J be the algebraic group over Qp of unitary K-linear
quasi-automorphisms of (X, ιX , λX ) which preserve the polarization up to a scalar in Q× p . Let J
1
1
denote the derived group of J. Then J (Qp ) acts on the functor Mr by changing the framing. It
follows from (i) in subsection 2.5 that J 1 can be identified with ResF/Qp (SU), where SU denotes
the quasi-split special unitary group in two variables over F . Note that SU is isomorphic to
SL2 /F .
The first main result in the local case can now be stated as follows.
bF ⊗
Theorem 2.6.2. Let r be special. Then the functor MK/F,r is represented by Ω ˆ OF ,ϕ0 OĔ .
More precisely, there exists a unique isomorphism of formal schemes
MK/F,r ≃ ΩbF ⊗
ˆ OF ,ϕ0 O , Ĕ
which is equivariant with respect to a fixed identification J 1 (Qp ) ≃ SL2 (F ). In particular,
MK/F,r is flat over Spf OĔ with semi-stable reduction.
Now let r be banal. Fix a local CM-triple (X, ιX , λX ) over k as in (ii) a) or (ii) b) in subsection
2.5. We write the height of λX as 2f h, where h ∈ {0, 1}. We assume that h = 0 when r is banal
split, or when r is banal and K/F is a ramified field extension. Recall from from (ii) a) in
subsection 2.5 that, when r is non-split, there is a hermitian space C = C(X, ιX , λX ) attached
to (X, ιX , λX ). By Proposition 4.5.14, the framing object (X, ιX , λX ) is uniquely defined up to
isogeny by the r-adjusted invariant invr (X, ιX , λX ) = inv(C) ∈ {±1} (see Proposition 8.3.6 for
this last identity). To make our statements uniform, we set invr (X, ιX , λX ) = 1 in the banal split
case.
We may now define a variant for banal r of the functor MK/F,r of Definition 2.6.1. Since
the functor depends not only on K/F but also on invr (X, ιX , λX ), we denote this functor by
MK/F,r,ε , where invr (X, ιX , λX ) = ε. When r is banal split, we have ε = 1; when r is banal
non-split and K/F is unramified, then ε = (−1)h , cf. Proposition 4.5.14.
Let S ∈ (Sch/OĔ ). A point of MK/F,r,ε (S) consists of an isomorphism class of exactly the
same data as in Definition 2.6.1.
Theorem 2.6.3. Let r be banal, and let ε ∈ {±1}. The formal scheme MK/F,r,ε is isomorphic
to (Spf OĔ ) × (J(Qp )o /CM ), where J(Qp )o denotes the subgroup of elements of J(Qp ) which
preserve the polarization up to a scalar in Z×
p . More precisely, there exists a unique isomorphism
of formal schemes
MK/F,r,ε ≃ (Spf OĔ ) × (J(Qp )o /CM ),
which is equivariant for the action of J(Qp )o . In particular, MK/F,r,ε is formally étale over
Spf OĔ .
3It can be proved that, when K/F is unramified, the fact that Ker λ ⊂ X [π] follows automatically from the
0
assumption that deg λ = p2f , cf. Proposition 5.3.7. We impose the condition Ker λ ⊂ X0 [π] in order to make
transparent that the moduli problem MK/F,r is of the kind considered in [27].
20 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Here, when r is non-split, J(Qp )o can be identified with the group of K-linear automorphisms
of C = C(X, ιX , λX ) preserving the hermitian form up to a factor in OF× , and CM is the stabilizer
in J(Qp )o of a lattice M in C which is self-dual when h = 0 and almost self-dual when h = 1.
When r is split, then J(Qp )o can be identified with the group of automorphisms of the two-
dimensional standard F -vector space of dimension 2 with determinant in OF× , and CM is the
stabilizer in J(Qp )o of the standard lattice M .
In the later part of the paper, we write simply J(Qp )o /CM for the formal scheme (Spf OĔ ) ×
(J(Qp )o /CM ) over Spf OĔ .

3. Background on Display Theory

In this section, K/Qp is an arbitrary finite field extension with ring of integers O = OK ,
and NilpO will denote the category of O-algebras R such that p is nilpotent in R. We recall
the classification of strict formal p-divisible O-modules over R ∈ NilpO proved in [1]. A main
ingredient is the Ahsendorf functor, which we present in a new form which is better suited for
our applications.
3.1. Displays. We fix a prime element π ∈ O. We denote by q the number of elements in the
residue class field κ of OK .
Definition 3.1.1. ([1, Def. 3.1], [20], [34]) Let R be an O-algebra. A frame F for R consists of
the following data:
(1) An O-algebra S and a surjective O-algebra homomorphism S → R. We denote the kernel
by I.
(2) An O-algebra endomorphism σ : S → S.
(3) A σ-linear map of S-modules σ̇ : I → S.
The following conditions are required.
(i) I + pS is contained in the radical of S.
(ii) σ(s) ≡ sq mod πS for all s ∈ S.
(iii) σ̇(I) generates S as an S-module.
We will denote a frame by F = (S, I, R, σ, σ̇) and we will sometimes make the identification
S/I = R.
A morphism of O-frames α : F = (S, I, R, σ, σ̇) −→ F ′ = (S ′ , I ′ , R′ , σ ′ , σ̇ ′ ) is an O-algebra
homomorphism α : S −→ S ′ such that α(I) ⊂ I ′ and such that
σ̇ ′ (α(a)) = α(σ̇(a)), a ∈ I.
The last equation implies that
σ ′ (α(s)) = α(σ(s)), s ∈ S.
Let F = (S, I, R, σ, σ̇) be an O-frame. Then there exists a unique element θ ∈ S in the radical
of S such that
σ(a) = θσ̇(a), for all a ∈ I, (3.1.1)
cf. [1, Lem. 3.2]. In the frames below we have θ = π.
Example 3.1.2. Let R be a p-adic O-algebra. Then the relative Witt ring WO (R) with respect
to the chosen uniformizer π ∈ O is a p-adic O-algebra. The relative Witt polynomials
n n−1 n−2
wO,n = X0q + πX1q + π 2 X2q q
+ . . . + π n−1 Xn−1 + π n Xn ,
define O-algebra homomorphisms wO,n : WO (R) −→ R. We denote by F and V the Frobenius
and the Verschiebung acting on WO (R), cf. [9]. In the case where k = R is a perfect field, the
ring WO (k) is the complete discrete valuation with residue class field k which is unramified over
O.
The relative Witt frame for R is the O-frame defined as
WO (R) = (WO (R), V WO (R), R, σ, σ̇). (3.1.2)
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 21

Here σ = F : WO (R) −→ WO (R) is the Frobenius endomorphism written as σ(ξ) = F ξ, and

σ̇(V ξ) = ξ, for ξ ∈ WO (R). We use also the notation Ḟ := σ̇ and IO (R) = V WO (R). If K = Qp
and π = p, we obtain the classical ring of Witt vectors W (R) = WZp (R). We write W(R) for
the Zp -frame WZp (R).
Example 3.1.3. Let S −→ R be a surjective homomorphism of p-adic O-algebras. We assume
that the kernel a is endowed with divided powers relative to O ([1], 1.2.2). They make sense out
of the expression ′′ aq /π ′′ . We also call this an O-pd-thickening. We denote by a[F n ] the ideal a
considered as an WO (S)-module via restriction of scalars relative to wO,n : WO (S) −→ S. The
divided powers give rise to divided Witt polynomials ẇO,n . They are homomorphisms of WO (S)-
modules ẇO,n : WO (a) −→ a[F n ] such that π n ẇO,n = wO,n . They give rise to an isomorphism
of WO (S)-modules
Y ∼
Y
ẇO,n : WO (a) −→ a[F n ] ,
n≥0 n≥0
cf. [1], 1.2.2. The inverse image in WO (a) of an element [a, 0, 0, . . .] from the right hand side
is called the logarithmic Teichmüller representative of a ∈ a. The logarithmic Teichmüller
representatives of elements of a form an ideal ã ⊂ WO (S). The ideal J = ã ⊕ IO (S) is the kernel
of the composition
wO,0
WO (S) −→ S −→ R.
Then Ḟ : IO (S) −→ WO (S) extends uniquely to a F -linear homomorphism Ḟ : J −→ WO (S)
such that Ḟ (ã) = 0. We define the relative Witt frame for S −→ R as
WO (S/R) = (WO (S), J , R, F, Ḟ ). (3.1.3)
This is an O-frame. Later we use the more precise notation
IO (S/R) = J = WO (a) + IO (S).
Definition 3.1.4 ([1], Def. 3.3). Let F = (S, I, R, σ, σ̇) be an O-frame. An F -display P =
(P, Q, F, Ḟ ) consists of the following data: a finitely generated projective S-module P , a sub-
module Q ⊂ P , and two σ-linear maps
F : P −→ P, Ḟ : Q −→ P.
The following conditions are required.
(i) IP ⊂ Q.
(ii) The factor module P/Q is a finitely generated projective R-module.
(iii) The following relation holds for a ∈ I and x ∈ P ,
Ḟ (ax) = σ̇(a)F (x).
(iv) Ḟ (Q) generates P as an S-module.
(v) The projective R-module Lie P = P/Q lifts to a finitely generated projective S-module. It
is called the Lie algebra of P.
If the rank of Lie P is constant, we call it the dimension of P. If the S-module P is of constant
rank, we call it the height of P. If we want to be precise, we say F -height.
F -displays form a category in the obvious way. In the case O = Zp and F = W(R) for a
p-adic ring R, we speak simply of a display over R. Displays for general frames F were originally
called F -windows, cf. [1, Def. 3.3]. We note that for the O-frames WO (R), the condition (v) of
Definition 3.1.4 is automatically satisfied, cf. [33, Lem. 2].
Example 3.1.5. For each O-frame F = (S, I, R, σ, σ̇) we have the multiplicative F -display
Pm = Pm,F = (S, I, σ, σ̇).
Example 3.1.6. Let P be an F -display. Let ε ∈ S be a unit. The display
P(ε) = (P, Q, εF, εḞ ).
is called the twist of P by ε.
22 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Recall the element θ from (3.1.1). The conditions in Definition 3.1.4 imply that
F (y) = θḞ (y), y ∈ Q. (3.1.4)
We can always find a direct sum decomposition P = T ⊕ L such that Q = IT ⊕ L. Such a
decomposition is called a normal decomposition of P . The σ-linear homomorphism
Φ := F|T ⊕ Ḟ|L : T ⊕ L −→ P (3.1.5)
is a σ-linear isomorphism, i.e., (3.1.5) corresponds to the linearization isomorphism,
∼
F ♯ ⊕ Ḟ ♯ : (S ⊗σ,S T ) ⊕ (S ⊗σ,S L) −→ P. (3.1.6)
Conversely, an arbitrary σ-linear isomorphism (3.1.5) defines an F -display in the obvious way.
For each display P there is a homomorphism of S-modules ([1, Def. 3.3])
V ♯ : P −→ S ⊗σ,S P (3.1.7)
which is uniquely determined by
V ♯ (sḞ y) = s ⊗ y, V ♯ (F x) = θ ⊗ x, x ∈ P, y ∈ Q, s ∈ S.
We have
V ♯ ◦ F ♯ = θ idS⊗σ,S P , F ♯ ◦ V ♯ = θ idP .
Any morphism of O-frames α : F −→ F ′ defines a base change functor α∗ from the category
of F -displays to the category of F ′ -displays as follows, cf. [1, Def. 3.8]. Let P be an F -display.
Then we define α∗ (P) = P ′ = (P ′ , Q′ , F ′ , Ḟ ′ ) as follows:


P ′ = S ′ ⊗S P, Q′ = Ker S ′ ⊗S P −→ R′ ⊗R (P/Q) , F ′ = σ ′ ⊗ F : P ′ −→ P ′ , (3.1.8)

and Ḟ ′ = σ ′ ⊗ Ḟ : Q′ −→ P ′ .
If P is given in terms of a normal decomposition (3.1.5), we obtain P ′ from the σ ′ -linear
extension of Φ,
Φ′ : (S ′ ⊗S T ) ⊕ (S ′ ⊗S L) −→ P ′ .
Example 3.1.7. The base change of the multiplicative display for the frame F under α : F −→
F ′ is the multiplicative display for F ′ .
If R is a perfect ring of characteristic p with an O-algebra structure, the category of WO (R)-
displays is equivalent with the more classical category of Dieudonné modules. We describe this
equivalence in its natural generality.
Definition 3.1.8. (a) A perfect O-frame is an O-frame F = (S, I, R, σ, σ̇) such that σ : S −→ S
is bijective.
If F = (S, I, R, σ, σ̇) is a perfect O-frame, then there is an element u ∈ I, such that σ̇(u) = 1.
Then σ(u) = θ is the element from (3.1.1). Furthermore, σ̇ : I −→ S is bijective, I = Su and
the elements u and θ are non zero divisors.
(b) A Dieudonné module (M, F, V ) for the perfect O-frame F consists of a finitely generated
projective S-module M and two additive maps F : M −→ M , V : M −→ M such that the
following conditions are satisfied.
(i) F (sx) = σ(s)F (x), V (sx) = σ −1 (s)V (x), x ∈ P, s ∈ S.
(ii) F ◦ V = θ idM , V ◦ F = u idM .
(iii) The R-module M/V M is projective and lifts to a finitely generated projective S-module.
V
If R is a perfect O-algebra, then F = WO (R) is a perfect O-frame and we have u = 1=
π = θ.
Proposition 3.1.9. Let F = (S, I, R, σ, σ̇) be a perfect O-frame. Let u, θ ∈ S as defined above.
Then the category of Dieudonné modules for F is equivalent to the category of F -displays.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 23

Proof. Let (M, F, V ) be a Dieudonné module. Since u and θ are not zero divisors, the maps
F : M −→ M and V : M −→ M are injective. Therefore we can define a display (P, Q, F, Ḟ ) by
setting
P = M, Q = V M, F = F, Ḟ = V −1 .
Conversely, let (P, Q, F, Ḟ ) be a display. We set (M, F ) := (P, F ). We have the bijective map
ν : S ⊗σ,S P −→ P, ν(s ⊗ x) := σ −1 (s)x.
Then we define V = ν ◦ V ♯ . More explicitly, we have
V (sḞ (y)) = σ −1 (s)y, y ∈ Q, s ∈ S.
This implies that V (P ) = Q. Moreover, we obtain
F V (sḞ y) = F (σ −1 (s)y) = sF (y) = θsḞ y
V F (x) = V (Ḟ (ux)) = ux.
Therefore (M, F, V ) is a Dieudonné module. 

In our basic example F = WO (R) for a perfect O-algebra R, we can replace the condition
(iii) above by the weaker condition that M/V M is a projective R-module. We note that for this
frame F ◦ V = π idM , V ◦ F = π idM .
We refer to [1, Def. 3.3] or [35] for the definition of a nilpotent F -display. If R is a perfect O-
algebra, a WO (R)-display is nilpotent iff for the corresponding Dieudonné module (M, F, V ) the
endomorphism V of M/πM is nilpotent. For an arbitrary O-algebra R such that π is nilpotent
in R, a WO (R)-display P is nilpotent iff for any homomorphism of O-algebras to a perfect field
R −→ k, the base change of P by the morphism of frames WO (R) −→ WO (k) is nilpotent.
Definition 3.1.10. Let R be an O-algebra. Let X be a p-divisible group over R endowed with
a Zp -algebra homomorphism ι : O −→ End X. We call the action ι strict if the induced action
on Lie X coincides with the O-action on this R-module given by restriction of scalars O −→ R.
We say that (X, ι) is a strict p-divisible O-module.
The following main result of [1] was known before for O = Zp [33], [21].
Theorem 3.1.11 ([1], Thm. 1.1). Let R ∈ NilpO . There is an equivalence of categories



nilpotent WO (R)-displays −→ strict formal p-divisible O-modules over R
which is functorial in R.
The theorem extends to p-adic R if we require the properties ”nilpotent” and ”formal” only
after base change to R/pR.
A nilpotent WO (R)-display gives rise to a crystal, as follows. Let S −→ R be a O-pd-
thickening, cf. Example 3.1.3. We assume that p is nilpotent in S. The ring homomorphism
WO (S) −→ WO (R) defines a morphism of O-frames
WO (S/R) −→ WO (R).
Theorem 3.1.12 ([20], [33]). Let S −→ R be an O-pd-thickening such that p is nilpotent in S.
The base change functor



nilpotent WO (S/R)-displays −→ nilpotent WO (R)-displays
is an equivalence of categories. 
Remark 3.1.13. In the case O = Zp , Lau [19] has defined a functor



p-divisible groups over R → W(S/R)-displays (3.1.9)
which gives a quasi-inverse of the functor in Theorem 3.1.12, when restricted to formal p-divisible
groups. In particular this functor associates to an arbitrary p-divisible group over R a display.
24 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Let P be a nilpotent WO (R)-display. Let P̃ be the unique WO (S/R)-display associated to P

by Theorem 3.1.12. Then we set
DP (S) = P̃ /IO (S)P̃ . (3.1.10)
This is a finitely generated projective S-module. It is a crystal in the following sense. If S ′ −→ R
is a another O-pd-thickening such that p in nilpotent in S ′ and S ′ −→ S is a morphism of O-pd-
thickenings, then there is a canonical isomorphism
S ⊗S ′ DP (S ′ ) ∼
= DP (S).
This crystal corresponds to the Grothendieck-Messing crystal of a p-divisible group via Theorem
3.1.11. From Theorem 3.1.12 one obtains the Grothendieck-Messing criterion for displays in the
following formulation.
Corollary 3.1.14. Let P be a nilpotent WO (R)-display. Let S −→ R be an O-pd-thickening.
Each WO (S)-display P̃ which lifts P defines a lifting Filf := Q̃/IO (S)P̃ ⊂ DP (S) of the Hodge
filtration Fil := Q/IO (R)P ⊂ DP (R).
For a fixed O-pd-thickening S −→ R, consider the category of pairs (P, Fil), f where P is a
f
nilpotent display and Fil ⊂ DP (S) is a lifting of the Hodge filtration associated to P. The functor
f is an equivalence of categories.
which maps a pair (P, P̃) to the pair (P, Fil) 
The following fact is well-known, but we give a proof.
Lemma 3.1.15. Let R be a p-adic O-algebra. Let P be a WO (R)-display. Let Õ be a discrete
valuation ring which is a finite extension of O. Let
Õ −→ End P,
be an O-algebra homomorphism. Then P is a locally on Spec R a free Õ ⊗O WO (R)-module.
Let S −→ R be an O-pd-thickening such that p is nilpotent in S. We assume that P is
nilpotent. Then DP (S) is locally on Spec S a free Õ ⊗Zp S-module.
Proof. We start with the case where S = R = k is a perfect field which contains the residue
class field of Õ. Let Õt be the maximal unramified extension of O contained in Õ. Let σ be the
Frobenius automorphism of Õt relative to O. To each O-algebra homomorphism ψ : Õt −→ k
there is a unique Frobenius equivariant O-algebra homomorphism
ψ̃ : Õt −→ WO (k)
which induces ψ when composed with wO,0 : WO (k) −→ k. This follows from the remark after
the definition of WO (R), cf. Example 3.1.2. The decomposition
Y
Õ ⊗O WO (k) = Õ ⊗Õt ,ψ̃ WO (k)
ψ

induces a decomposition
P = ⊕ψ Pψ .
Each Pψ is a free module over the discrete valuation ring Õ ⊗Õt ,ψ̃ WO (k). The Frobenius
F : P −→ P induces maps Pψ −→ Pψσ . This shows that all Pψ have the same rank as WO (k)-
modules. This proves the case where R = k is a perfect field containing the residue class field of
Õ.
Now let k be an arbitrary field of characteristic p. It suffices to show that P ⊗WO (k) k is a
free Õ ⊗O k-module. By base change α : k −→ k̄ this follows from the previous case because
P ⊗WO (k) k̄ = α∗ (P ) ⊗ k̄ is a free Õ ⊗O k̄-module.
If R is a local ring with residue class field of characteristic p we conclude by Nakayama’s lemma
that P ⊗WO (R) R is a free Õ ⊗O R-module. The generalization to arbitrary R is immediate. This
proves the first assertion of the Lemma.
If P is nilpotent, the crystal DP is defined. The case DP (R) = P/IO (R)P was proved above.
For arbitrary S −→ R we can apply again Nakayama’s lemma. 
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 25

Remark 3.1.16. Let R be a ring such that p is nilpotent in R. Let X be a p-divisible group
over R with a ring homomorphism
O −→ End X.
Let S −→ R be a nilpotent pd-thickening. Then the value of the Grothendieck-Messing crystal
DX (S) is locally on Spec S a free O ⊗Zp S-module. This can be shown by the same arguments
as above.
Finally we discuss isogenies of WO (R)-displays, where R is an O-algebra such that p is nilpo-
tent in R. We assume moreover that Spec R is connected. Let α : P1 −→ P2 be a morphism
of displays of the same height and dimension, cf. the remark after Definition 3.1.4. Locally on
Spec R the WO (R)-modules P1 and P2 are free of the same rank. We may choose a basis in each
of these modules and write det α ∈ WO (R). This is locally defined up to a unit in WO (R). More
invariantly one can write exterior powers.
Definition 3.1.17. A morphism of WO (R)-displays of the same height and dimension α :
P1 −→ P2 is called an isogeny if det α 6= 0.
Proposition 3.1.18 ([35], Prop. 17.6.2.). Let R be an O-algebra such that p is nilpotent in R
and such that Spec R is connected. Let α : P1 −→ P2 be an isogeny of WO (R)-displays. Then
there exists a natural number h ∈ Z≥0 such that locally on Spec R
det α = π h ǫ, ǫ ∈ WO (R)× .

We call h the O-height of α, and write h = heightO α. If O = Zp , we write simply height α.
An abbreviation for the Proposition is:
heightO α = ordπ det α.
Proposition 3.1.19 ([35], Prop. 17.6.4.). Assume that the ideal of nilpotent elements in R is
nilpotent and that Spec R is connected. Let α : P1 −→ P2 be an isogeny of O-height h. Then
there exists locally on Spec R a morphism of WO (R)-displays β : P2 −→ P1 such that
β ◦ α = π h idP1 , α ◦ β = π h idP2 .

Proposition 3.1.20. With the assumptions of Proposition 3.1.18, let a : X1 −→ X2 be a
morphism of strict formal p-divisible O-modules over R. Let α : P1 −→ P2 be the induced
morphism of the associated WO (R)-displays, cf. Theorem 3.1.11. The morphism a is an isogeny
of height h if and only if α is an isogeny of height h.
Proof. This can be reduced to the case of a perfect field R = k where it is well-known by
Dieudonné theory. 
Let R be an O-algebra and let a : X1 −→ X2 be a morphism of strict formal p-divisible O-
modules. By Theorem 3.1.11, there is an associated morphism α : P1 −→ P2 of WO (R)-displays.
We set
heightO a = heightO α, heightO X1 = heightWO (R) P1 . (3.1.11)
The last height was defined after Definition 3.1.4. It is equal to the O-height of the endomorphism
of P1 given by multiplication by π. We also write
heightO P1 = heightO (π|P1 ) = heightWO (R) P1 .
3.2. Bilinear forms of displays. Let F = (S, I, R, σ, σ̇) be an O-frame and let θ ∈ S be the
element from (3.1.1).
Definition 3.2.1. Let P1 , P2 , P be F -displays. A bilinear form of F -displays
β : P1 × P2 −→ P
is a bilinear form of S-modules
β : P1 × P2 −→ P (3.2.1)
with the following properties:
26 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

(i) The restriction of β to Q1 × Q2 takes values in Q.

(ii) For y1 ∈ Q1 and y2 ∈ Q2 ,
Ḟ β(y1 , y2 ) = β(Ḟ1 y1 , Ḟ2 y2 ).
We will denote the O-module of all bilinear forms by
Bil(P1 × P2 , P).
Lemma 3.2.2. The following equations hold
F β(x1 , y2 ) = β(F1 x1 , Ḟ2 y2 ), x1 ∈ P1 , y2 ∈ Q2 ,
F β(y1 , x2 ) = β(Ḟ1 y1 , F2 x2 ), y1 ∈ Q1 , x2 ∈ P2 ,

θF β(x1 , x2 ) = β(F1 x1 , F2 x2 ), x1 ∈ P1 , x2 ∈ P2 .
Proof. We omit the verification which is, for classical displays, contained in [33]. 
Let R be a perfect O-algebra and let F = WO (R). Then we may equivalently consider
Dieudonné modules (P, F, V ) and (Pi , Fi , Vi ) for i = 1, 2, cf. Proposition 3.1.9. We can reformu-
late the Definition 3.2.1 as follows: A bilinear form of Dieudonné modules is a bilinear form of
WO (R)-modules β : P1 × P2 −→ P such that
β(V1 x1 , V2 x2 ) = V β(x1 , x2 ). (3.2.2)
Proposition 3.2.3. Let β : P1 × P2 −→ P be a bilinear form of F -displays. Let α : F −→ F ′
be a morphism of frames. Denote by P1′ , P2′ , and P ′ the displays obtained by base change with
respect to α. Because Pi′ = S ′ ⊗S Pi , and P ′ = S ′ ⊗S P , there is an induced S ′ -bilinear form
β ′ : P1′ × P2′ −→ P ′ . This is a bilinear form of F ′ -displays
P1′ × P2′ −→ P ′ .
Proof. We omit the straightforward verification. 
Let P = (P, Q, F, Ḟ ) be an F -display. We are going to define the dual F -display P ∨ =
(P , Q∨ , F ∨ , Ḟ ∨ ). For an S-module M , we define M ∗ = HomS (M, S). We set P ∨ := P ∗ , and
∨

Q∨ = {ψ ∈ P ∨ | ψ(Q) ⊂ I}.
We note that we have a natural perfect pairing
P/IP × P ∨ /IP ∨ −→ R.
We deduce that Q∨ /IP ∨ is the orthogonal complement of Q/IP and is therefore a direct sum-
mand of P/IP . We claim that there are σ-linear maps
F ∨ : P ∨ −→ P ∨ , Ḟ ∨ : Q∨ −→ P ∨
which are uniquely determined by the following conditions. We denote by < , > : P × P ∨ −→ S
the natural perfect pairing. Then we require for x ∈ P, y ∈ Q, φ ∈ P ∨ , ψ ∈ Q∨ :
< Ḟ (y), F ∨ (φ) > = σ(< y, φ >), < F (x), F ∨ (φ) > = θσ(< x, φ >),
(3.2.3)
< Ḟ (y), Ḟ ∨ (ψ) > = σ̇(< y, ψ >), < F (x), Ḟ ∨ (ψ) > = σ(< x, ψ >).
Since Ḟ is a σ-linear surjection, the maps F ∨ and Ḟ ∨ are uniquely determined by these identities.
To verify the existence of these maps, we consider a normal decomposition,
P = T ⊕ L.
Let L ⊂ P be the orthogonal complement of L and T ∨ ⊂ P ∨ the orthogonal complement of
∨ ∨

T . Hence there are canonical isomorphisms

L∨ =∼ T ∗, T ∨ ∼
= L∗ .
We obtain the normal decomposition
P ∨ = T ∨ ⊕ L∨ , Q∨ = IT ∨ ⊕ L∨ .
For ψ ∈ L∨ = T ∗ , we set
Ḟ ∨ (ψ)(Ḟ (ℓ)) = 0, Ḟ ∨ (ψ)(Ḟ (t)) = σ(ψ(t)), ℓ ∈ L, t ∈ T.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 27

This definition makes sense because of the linearization isomorphism (3.1.6). Finally, we define
F ∨ (φ) for φ ∈ T ∨ = L∗ by the equations
F ∨ (φ)(Ḟ (ℓ)) = σ(φ(ℓ)), F ∨ (φ)(F (t)) = 0.
One verifies that, with these definitions, the identities (3.2.3) are satisfied. It follows from the
symmetry of the equations (3.2.3) that we have a natural isomorphism
P=∼ (P ∨ )∨ .
By the equations (3.2.3) we have a natural bilinear form of displays
P × P ∨ −→ Pm (3.2.4)
′
with values in the multiplicative display Pm = Pm,F . If P is another F -display, the bilinear
form (3.2.4) induces an isomorphism
∼
HomF -displays (P ′ , P ∨ ) −→ Bil(P ′ × P, Pm ). (3.2.5)
We deduce a variant of the Grothendieck-Messing criterion. Let P and P ′ be WO (R)-displays
such that P ∨ and P ′ are nilpotent. Let S −→ R be a O-pd-thickening in NilpO , cf. Example
∨ ′
3.1.3. We denote by Prel and Prel the associated WO (S/R)-displays, which exist by Theorem
3.1.12. We define Prel = (Prel ) , where the last ∨ denotes the dual in the category of WO (S/R)-
∨ ∨

displays. We set DP (S) = Prel /I(S)Prel . Then we obtain a crystal which is dual to the crystal
DP ∨ (S), cf. (3.1.10). This crystal agrees with DP (S) defined earlier, if P is nilpotent. It follows
form (3.2.5) that each bilinear from
β : P ′ × P −→ Pm,WO (R) (3.2.6)
induces a bilinear form
′
βrel : Prel × Prel −→ Pm,WO (S/R)
and, in particular, a S-bilinear form
βcrys : DP ′ (S) × DP (S) −→ S. (3.2.7)
Proposition 3.2.4. Let R ∈ NilpO and let S −→ R be an O-pd-thickening in NilpO . Let
P and P ′ be WO (R)-displays and assume that P ∨ and P ′ are nilpotent. Let P̃ and P̃ ′ be
′
f ⊂ DP (S) and Fil
liftings which correspond to liftings of the Hodge filtrations Fil f ⊂ DP ′ (S),
cf. Corollary 3.1.14. Then a bilinear form β : P ′ × P −→ Pm,WO (R) lifts to a bilinear form
β̃ : P̃ ′ × P̃ −→ Pm,WO (S) iff
′
f , Fil)
βcrys (Fil f = 0.

Proof. This is a consequence of Corollary 3.1.14 and (3.2.5). 
We go back to an arbitrary O-frame F and add a remark on the map V ♯ , cf. (3.1.7). If P
is an S-module we set P (σ) = S ⊗σ,S P . If P is projective and finitely generated, the perfect
pairing < , > induces a perfect pairing
< , >(σ) : P (σ) × (P ∗ )(σ) −→S
(s1 ⊗ x, s2 ⊗ φ) 7−→s1 s2 σ(φ(x)).
Let P be an F -display and let P be the dual display. The maps (F ∨ )♯ and V ♯ are dual in the
∨

following sense
< V ♯ x, s ⊗ φ >(σ) =< x, (F ∨ )♯ (s ⊗ φ) > .
Definition 3.2.5. A polarization of an F -display P is a bilinear form
β : P × P −→ Pm,F
such that the underlying bilinear form P ×P −→ S is alternating and its determinant is non-zero.
If F = WO (R), the height of β is the height of the associated isogeny P −→ P ∨ (cf. Propo-
sition 3.1.18). We write heightO β for the height of β. Then we have heightO β = ordπ det β in
the notation of Proposition 3.1.18. The polarization is called principal if heightO (β) = 0.
28 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Remark 3.2.6. Let X be a strict formal p-divisible O-module over R ∈ NilpO . Let P be the
WO (R)-display of X in the sense of Theorem 3.1.11. If the dual display P ∨ is nilpotent, it
corresponds to a strict formal p-divisible O-module X ∨ , called the O-dual of X. In this case, a
polarization is given by an anti-symmetric O-module homomorphism X −→ X ∨ .
3.3. The Ahsendorf functor. We will give here an alternative definition of the Ahsendorf
functor of [1] which is better suited to our purposes. One step of this definition is contained in
the Appendix of [22]. We use a Lubin-Tate frame introduced by Mihatsch in loc. cit., but for us
it will be important to make a specific choice, cf. Definition 3.3.8.
Let Qp ⊂ k ⊂ K be a subfield. We denote by o the ring of integers in k. We fix a prime
element ̟ ∈ o. If R is an o-algebra we denote by Wo (R) the Witt vectors relative to o and ̟.
The Frobenius and the Verschiebung will be denoted by f and v. We set [K : k] = ef where e is
the ramification index and f is the inertia index. Beginning with section 4 we will only consider
the case where k = Qp .
For an O-algebra R we have the Drinfeld homomorphism
µ : Wo (R) −→ WO (R), (3.3.1)
cf. [9, Prop. 1.2]. It is functorial in R and satisfies wO,n (µ(ξ)) = wo,f n (ξ), for ξ ∈ Wo (R). This
implies the following properties:

ff F ̟ V f −1
µ( ξ) = µ(ξ), µ( v ξ) =
(µ( f ξ)), µ([u]) = [u], (3.3.2)
π
for ξ ∈ Wo (R), u ∈ R. The last equation says that the Teichmüller representative [u] ∈ Wo (R),
is mapped by µ to the Teichmüller representative [u] ∈ WO (R).
We have µ(Io (R)) ⊂ IO (R). Therefore we may rewrite the second equation of (3.3.2) as
Ḟ ̟ f −1
(µ(η)) = µ( f ḟ η), η ∈ Io (R). (3.3.3)
π
The following definition extends Definition 3.1.10 to the relative case.
Definition 3.3.1. Let F = (S, I, R, σ, σ̇) be an o-frame, where R is a p-adic O-algebra. Let
P = (P, Q, F, Ḟ ) be an F -display. A strict O-action on P is a homomorphism of o-algebras
O −→ End P such that the induced action on the R-module P/Q coincides with the O-module
structure on P/Q obtained by restriction of scalars O −→ R.
For a p-adic O-algebra R we will define a functor
   
Wo (R)-displays
AO/o,R : −→ WO (R)-displays . (3.3.4)
with strict O-action
We call this functor the Ahsendorf functor. The image of a Wo (R)-display P as in Definition
3.3.1 will be denoted by Pa = AO/o,R (P). The main theorem on the Ahsendorf functor is:
Theorem 3.3.2. Let R be an O-algebra such that p is nilpotent in R. The Ahsendorf functor
induces an equivalence of categories
   
nilpotent Wo (R)-displays
AO/o,R : −→ nilpotent WO (R)-displays .
with strict O-action
Furthermore, the Ahsendorf functor canonically associates to a bilinear form
β : P ′ × P ′′ −→ P (3.3.5)
of Wo (R)-displays with a strict O-actions such that β is also O-bilinear, a bilinear form of
WO (R)-displays
Pa′ × Pa′′ −→ Pa .
Proof. The first statement is the main result of [1]. The second statement is shown in Proposition
3.3.15. 
Remark 3.3.3. By Theorem 3.3.2 we obtain a functor
   
strict formal p-divisible O-modules over R −→ WO (R)-displays , (3.3.6)
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 29

which is defined as follows. By [19], [35] there is a functor from the first category to the category
of W(R)-displays with a strict O-action. Composing this with AO/Zp ,R we obtain (3.3.6). In
particular this gives a quasi-inverse functor to the functor of Theorem 3.1.11.
We will now define the Ahsendorf functor. We denote by K t ⊂ K the maximal subextension
which is unramified over k. Let Ot be the ring of integers of K t . We consider the Witt vectors
WOt (R) with respect to the prime element ̟ ∈ Ot . The Frobenius resp. the Verschiebung
acting on WOt (R) will be denoted by F ′ and V ′ . We will define AO/o,R as the composite of two
functors
   
Wo (R) − displays WOt (R) − displays
AOt /o,R : −→ ,
with strict O-action with strict O-action
    (3.3.7)
WOt (R) − displays
AO/Ot ,R : −→ WO (R) − displays .
with strict O-action
We begin with the definition of AOt /o,R .
Lemma 3.3.4. Let S be an O-algebra which has no π-torsion. Let τ : S −→ S be a O-algebra
homomorphism such that
τ (s) ≡ sq ( mod π).
Let u0 , u1 , . . . , un , . . . ∈ S. Then there exists ξ ∈ WO (S) such that wO,n (ξ) = un for all n iff
τ (un−1 ) ≡ un mod π n S, for n ≥ 1.
The element ξ is uniquely determined.
Proof. The proof is up to obvious changes identical with the proof for the classical case O = Zp ,
cf. [2, IX, §1, 2, Lemme 2]. 
We denote by σ ∈ Gal(K t /k) the Frobenius automorphism. By Lemma 3.3.4, there is a
homomorphism λ : Ot −→ Wo (Ot ), defined by wo,n (λ(a)) = σ n (a) for a ∈ Ot and all n. We
obtain a ring homomorphism
λ
κ : Ot −→ Wo (Ot ) −→ Wo (R). (3.3.8)
t
We introduce the Ahsendorf frame with respect to the unramified extension O /o for a p-adic
Ot -algebra R,
Ao (R) = (Wo (R), Io (R), R, ff , ff −1 ḟ). (3.3.9)
t
This is an O -frame via κ.
Let P = (P, Q, F, Ḟ ) be a Wo (R)-display with a strict O-action. We set
Pm = {x ∈ P | ι(a)x = κ(σ m (a))x, for a ∈ Ot }, m ∈ Z/f Z.
The Wo (R)-module P decomposes as
P = ⊕m∈Z/f Z Pm . (3.3.10)
There is a similiar decomposition for Q. The maps F and Ḟ of P are graded of degree one,
F : Pm −→ Pm+1 , Ḟ : Qm −→ Pm+1 .
If the action ι is strict, we have Qm = Pm for m 6= 0. Then we define the Ao (R)-display Pua :
Pua = P0 , Qua = Q0 , Fua = Ḟ f −1 F, Ḟua = Ḟ f . (3.3.11)
It is clear that O acts strictly on Pua .
It follows from (3.3.2) that µ : Wo (R) −→ WOt (R) induces a morphism of Ot -frames
µ : Ao (R) −→ WOt (R). (3.3.12)
By base change we obtain from Pua a WOt (R)-display Pt = µ∗ (Pua ). The strict action of O
on Pua induces a strict action of O on Pt because the tangent space remains unchanged by this
base change.
Definition 3.3.5. The Ahsendorf functor AOt /o,R is the functor which associates to a Wo (R)-
display P with a strict O-action the WO (R)-display Pt defined above.
30 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

The Ahsendorf functor is compatible with bilinear forms as follows. Let β : P ′ × P ′′ −→ P

as in (3.3.5). Because β is Ot -bilinear, β induces for each m ∈ Z/f Z a pairing
′ ′′
β : Pm × Pm −→ Pm ,
and Pi′ and Pj′′ are orthogonal for i 6= j. For y ′ ∈ Q′0 and y ′′ ∈ Q′′0 we find
β((Ḟ ′ )f y ′ , (Ḟ ′′ )f y ′′ ) = Ḟ f β(y ′ , y ′′ ).
Therefore the restriction of β,
βua : P0′ × P0′′ −→ P0
induces a bilinear form of Ao (R)-displays
′ ′′
Pua × Pua −→ Pua .
Applying Proposition 3.2.3, we obtain a bilinear form in the category of WOt (R)-displays,
βt : Pt′ × Pt′′ −→ Pt . (3.3.13)
Now we define AO/Ot ,R . First we introduce the Lubin-Tate frame. We choose a finite normal
extension L of K t which contains K. We set Φ = HomK t -Alg (K, L). Let ϕ0 : K −→ L be the
identical embedding.
Let EK ∈ Ot [T ] be the Eisenstein polynomial of π ∈ O over Ot . In OL [T ] it decomposes as
Y
EK (T ) = (T − ϕ(π)).
ϕ∈Φ

We set Y
EK,0 (T ) = (T − ϕ(π)) ∈ OL [T ]
ϕ∈Φ,ϕ6=ϕ0
One sees easily that EK,0 ∈ O[T ]. We lift these polynomials via w0 to the ring of Witt vectors,
Q
ẼK (T ) = ϕ∈Φ (T − [ϕ(π)]) ∈ WOt (Ot )[T ],
Q (3.3.14)
ẼK,0 (T ) = ϕ∈Φ,ϕ6=ϕ0 (T − [ϕ(π)]) ∈ WOt (O)[T ].
The Frobenius F ′ and the Verschiebung V ′ act via the second factor on O ⊗Ot WOt (R). We set
Ḟ ′ = (V ′ )−1 : O ⊗Ot IOt (R) −→ O ⊗Ot WOt (R).
Proposition 3.3.6. The element
Ḟ ′
(ẼK (π ⊗ 1)) ∈ O ⊗Ot WOt (Ot )
is a unit of the form
πe
( ⊗ 1)δ, δ ∈ O ⊗Ot WOt (Ot ) (3.3.15)
̟
such that δ − (1 ⊗ 1) lies in the kernel of
O ⊗Ot WOt (Ot ) −→ O ⊗Ot WOt (κ),
and hence in the radical of O ⊗Ot WOt (Ot ).
Proof. The element is defined because id ⊗ wOt ,0 : O ⊗Ot WOt (Ot ) −→ O maps ẼK (π ⊗ 1) to 0.
In the following computation we pass to O ⊗Ot WOt (OL ). We find from the definitions:
′ Y
Ḟ ′
(ẼK (π ⊗ 1)) = Ḟ ( (π ⊗ 1 − 1 ⊗ [ϕ(π)])
ϕ∈Φ
e (3.3.16)
1 Y 1 X e−i F′
= (π ⊗ 1 − 1 ⊗ [ϕ(π)q ]) = π ⊗ si .
̟ ̟ i=0
ϕ∈Φ

Here we denote by si the elementary symmetric polynomial of degree i evaluated at the e

′
arguments [ϕ(π)]. By definition s0 = 1. We claim that for i > 0 the elements F si ∈ WOt (Ot )
are divisible by ̟. Clearly wOt ,0 (si ) is divisible by π. On the other hand, wOt ,0 (si ) ∈ Ot and
therefore is divisible by ̟. We find expressions in WOt (Ot ),
V′
si = [̟ci ] + ξi , ci ∈ Ot , ξi ∈ WOt (Ot ).
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 31

′
Therefore F si = [̟q ][cqi ] + ̟ξ i is divisible by ̟. Indeed, using Lemma 3.3.4, one shows as in
the proof of [33, Lem. 28] that ̟ divides [̟q ].
Now we may write the last term of (3.3.16) as
Xe F′
πe si
⊗1+ π e−i ⊗ .
̟ i=1
̟
F′
Finally, ̟si lies for i > 0 in the kernel of WOt (Ot ) −→ WOt (κ). Indeed, the elements [ϕ(π)] ∈
WOt (OL ) are mapped to zero in WOt (κL ) and therefore a fortiori the symmetric functions si .
′
We conclude that si and then F si become zero in WOt (κ) for i > 0. Because ̟ is not a zero
F′
si
divisor in WOt (κ) the elements ̟ are then also in the kernel. 

We write in the ring WO (O),

V
π − [π] = ε. (3.3.17)
One checks that ε ∈ WO (O) is a unit. If we apply F to the last equation, we obtain
π − [π q ] = πε. (3.3.18)
In particular [π q ] is divisible by π.
Ḟ ′

−1 F′
Lemma 3.3.7. The image of the element ẼK (π ⊗ 1) · ẼK,0 (π ⊗ 1) under the Drinfeld
homomorphism
µ : O ⊗Ot WOt (O) −→ WO (O)
equals ε−1 (̟/π).
Proof. It is enough to show the same assertion for O ⊗Ot WOt (OL ) −→ WO (OL ). The image of
Ḟ ′ Ẽ (π ⊗ 1) by the last map is ̟ −1
Q q
K ϕ (π − [ϕ(π) ]). Here we used that ̟ is not a zero divisor
in the participating rings. Our assertion is equivalent with the equation
Y ̟ Y
̟−1 (π − [ϕ(π)q ])ε−1 = (π − [ϕ(π)q ]).
ϕ
π
ϕ6=ϕ0

But this is a consequence of (3.3.18). 

The free WOt (R)-module O ⊗Ot WOt (R) has the basis
1 ⊗ 1, π ⊗ 1 − 1 ⊗ [π], . . . , π m ⊗ 1 − 1 ⊗ [π]m , . . . , π e−1 ⊗ 1 − 1 ⊗ [π]e−1 . (3.3.19)
To ease the notation, here [π] denotes the Teichmüller representative of the image of π by the
morphism O −→ R. Let


J = Ker O ⊗Ot WOt (R) −→ R ,
where the map is induced by w0 : WOt (R) −→ R. The ideal J is contained in the radical
of O ⊗Ot WOt (R). As a WOt (R)-module, J is the direct sum of O ⊗O IO (R) and the direct
summand generated by the last e − 1 elements of (3.3.19). In particular we obtain
J = O ⊗Ot IOt (R) + (π ⊗ 1 − 1 ⊗ [π])(O ⊗Ot WOt (R)). (3.3.20)
We define maps σlt : O ⊗Ot WOt (R) −→ O ⊗Ot WOt (R), σ̇lt : J −→ O ⊗Ot WOt (R) by
F′ Ḟ ′ Ḟ ′
σlt
ξ= ξ, σ̇lt
η=( ẼK )−1 (ẼK,0 η), ξ ∈ O ⊗Ot WOt (R), η ∈ J .
The map σ̇lt : J −→ O ⊗Ot WOt (R) is σlt -linear. Then we obtain
Ḟ ′ Ḟ ′
σ̇lt
(π ⊗ 1 − 1 ⊗ [π]) = ( ẼK )−1 (ẼK ) = 1. (3.3.21)
Definition 3.3.8. (comp. [22, Def. 2.7]) The Lubin-Tate frame for O is the O-frame
Flt (R) := (O ⊗Ot WOt (R), J , R, σlt , σ̇lt ).
32 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

This is indeed an O-frame: the only thing we need to check is

σlt
ξ ≡ ξ q ( mod (π ⊗ 1)O ⊗Ot WOt (R)),
F′
and this follows because a ≡ aq mod π for a ∈ O and η ≡ η q mod ̟ for η ∈ WOt (R).
We remark that by (3.3.21)
(π ⊗ 1 − 1 ⊗ [π q ]) σ̇lt
η= σlt
η, η ∈ J.
Now we start with a WOt (R)-display P with a strict O-action. The last condition can be
reformulated as
J P ⊂ Q. (3.3.22)
We refer to [1, Prop. 2.26] for the proof of the following lemma.
Lemma 3.3.9. Let P be a WOt (R)-module with an action of O, i.e., a homomorphism of O-
algebras
O −→ EndWOt (R) P.
Assume that locally on Spec R the WOt (R)-module P is free. Then P is locally on Spec R a
finitely generated free O ⊗Ot WOt (R)-module. 
Lemma 3.3.10. Let P be a WOt (R)-display with a strict O-action. Let x ∈ P . By (3.3.22)
(π ⊗ 1 − 1 ⊗ [π])x ∈ Q. The following equation holds,
′
−1 F ′

F x = Ḟ ẼK (π ⊗ 1) · ẼK,0 (π ⊗ 1) · Ḟ (π ⊗ 1 − 1 ⊗ [π])x .

Proof. From the definition of the polynomials ẼK and ẼK,0 , we find since ϕ0 (π) = π,
ẼK (π ⊗ 1) = ẼK,0 (π ⊗ 1) · (π ⊗ 1 − 1 ⊗ [π]).
Therefore
Ḟ (ẼK (π ⊗ 1)x) = Ḟ (ẼK,0 (π ⊗ 1)) · (π ⊗ 1 − 1 ⊗ [π])x)
F′
= (ẼK,0 (π ⊗ 1))Ḟ ((π ⊗ 1 − 1 ⊗ [π])x).
Because ẼK (π ⊗ 1) ∈ O ⊗Ot IOt (R), we obtain
Ḟ ′
Ḟ (ẼK (π ⊗ 1)x) = (ẼK (π ⊗ 1))F x.
We conclude by Lemma 3.3.6. 

We now associate to the WOt (R)-display P = (P, Q, F, Ḟ ) with a strict O-action a Flt -display
Plt = (Plt , Qlt , Flt , Ḟlt ). We set Plt = P , Qlt = Q, Ḟlt = Ḟ , and
Flt (x) = Ḟ ((π ⊗ 1 − 1 ⊗ [π])x), x ∈ P. (3.3.23)
Proposition 3.3.11. Plt is an Flt (R)-display.
Proof. The only thing we have not checked is the equation
σ̇lt
Ḟlt (ηx) = ηFlt x, η ∈ J. (3.3.24)
′
V
We begin with the case η = ξ. We apply Lemma 3.3.10
V′
Ḟlt (ηx) = Ḟ ( ξx) = ξF (x) = ξ( Ḟ ẼK )−1 · F
ẼK,0 · Ḟ ((π ⊗ 1 − 1 ⊗ [π])x). (3.3.25)
By definition
Ḟ ′ Ḟ ′ Ḟ ′ F′
σ̇lt
η=( ẼK )−1 (ẼK,0 V
ξ) = ( ẼK )−1 (ẼK,0 )ξ.
σ̇lt
Using the definition (3.3.23), we can write the right hand side of (3.3.25) as ηFlt (x), hence
we are done in this case.
Next we consider the case where η = (π ⊗ 1 − 1 ⊗ [π])ξ. Then we find
F
Ḟlt (ηx) = Ḟ ((π ⊗ 1 − 1 ⊗ [π])ξx) = ξFlt (x).
′
σ̇lt F
But by (3.3.21) we have ((π ⊗ 1 − 1 ⊗ [π])ξ) = ξ. Therefore in this case (3.3.24) is true as
well. 
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 33

We use the same symbol ε ∈ WO (R)× for the image of the element ε ∈ WO (O)× defined by
(3.3.17). We define the frame
ε
WO (R) = (WO (R), IO (R), F, ε−1 Ḟ ). (3.3.26)
ε
We note that the categories of displays over WO (R) and WO (R) are canonically isomorphic.
ε
Indeed, if P = (P, Q, F, Ḟ ) is a WO (R)-display, then P = (P, Q, εF, Ḟ ) is a WO (R)-display.
Recall that we denote the Frobenius and the Verschiebung acting on WO (R) by F ′ and V ′ .
t

We consider the Drinfeld homomorphism µ : WOt (R) −→ WO (R), cf. (3.3.1). This is a functorial
ring homomorphism such that wn′ (µ(ξ)) = wn (ξ) which has the following properties
F′ F V′ ̟ V
µ( ξ) = µ(ξ), µ( ξ) = µ(ξ), µ([a]) = [a], for a ∈ R. (3.3.27)
π
The Drinfeld homomorphism extends to a ring homomorphism
µ : O ⊗Ot WOt (R) −→ WO (R) (3.3.28)
which we denote by the same letter.
Proposition 3.3.12. The Drinfeld homomorphism induces a morphism of O-frames
ε
µ : Flt (R) −→ WO (R).
Proof. We have to check that the image of J ⊂ O ⊗Ot WOt (R) by µ is contained in IO (R). This
is immediate because µ(π ⊗ 1 − 1 ⊗ [π]) = π − [π] = V ε. It remains to prove the equations for
ξ ∈ O ⊗Ot WOt (R) and η ∈ J ,

µ( σlt
ξ) = F
µ(ξ), µ( σ̇lt
η) = ε−1 Ḟ
µ(η). (3.3.29)
The first equation follows from (3.3.27). To prove the second equation, it is enough to consider
′
the following two cases separately: η = V ξ and η = (π ⊗ 1 − 1 ⊗ [π])ξ. In the first case we have
Ḟ ′ Ḟ ′ V′ Ḟ ′ F′
σ̇lt
η=( ẼK )−1 (ẼK,0 ξ) = ( ẼK )−1 ( ẼK,0 )ξ.
Applying Lemma 3.3.7, we obtain
µ( σ̇lt
η) = ε−1 (̟/π)µ(ξ).
On the other hand, we have by (3.3.27)
′
ε−1 Ḟ
µ(V ξ) = ε−1 Ḟ
((̟/π) V
µ(ξ)) = ε−1 ((̟/π)µ(ξ),
as desired.
Now we consider the case η = (π ⊗ 1 − 1 ⊗ [π])ξ. We have
V V F
µ((π ⊗ 1 − 1 ⊗ [π])ξ) = (π − [π])µ(ξ) = εµ(ξ) = (ε µ(ξ)).
We obtain
ε−1 Ḟ
µ(η) = F
µ(ξ).
On the other hand, we find by (3.3.21) and (3.3.27)
σ̇lt F′ F
µ( η) = µ( ξ) = µ(ξ),
as desired. 

Starting now with a WOt (R)-display P = (P, Q, F, Ḟ ) with a strict O-action, we have the
associated Flt (R)-display Plt = (Plt , Qlt , Flt , Ḟlt ), cf. Proposition 3.3.11. After taking the base
ε
change by the morphism of frames of Proposition 3.3.12, we obtain a WO (R)-display Paε =
(Paε , Qεa , Faε , Ḟaε ). Then Pa = (Paε , Qεa , ε−1 Faε , Ḟaε ) is an WO (R)-display.
Definition 3.3.13. The Ahsendorf functor AO/Ot ,R is the functor which associates to the
WOt (R)-display P with a strict O-action the WO (R)-display Pa defined above.
34 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

From the construction we obtain that

Pa = WO (R) ⊗O⊗Ot WOt (R) P, (3.3.30)
and that Qa is the kernel of the natural map
WO (R) ⊗O⊗Ot WOt (R) P −→ P/Q.
We note that the canonical map P −→ Pa induces a map Q −→ Qa .
Proposition 3.3.14. Let P be a WOt (R)-display with a strict action of O. Let Pa = AO/Ot ,R (P)
be its image by the Ahsendorf functor. The following diagram is commutative
Ḟ
Q /P

 
Qa / Pa
Ḟa

Proof. This follows from the definition of Plt before Proposition 3.3.11 and the definition of base
change (via the morphism of Proposition 3.3.12). 
We note that this diagram determines the map Ḟa uniquely. Indeed, consider the following
equation in Pa under the identification (3.3.30),
V
ε ⊗ x = 1 ⊗ (π ⊗ 1 − 1 ⊗ [π])x.
Applying Ḟa , we obtain from the diagram that
εFa (1 ⊗ x) = 1 ⊗ Ḟ ((π ⊗ 1 − 1 ⊗ [π])x).
This shows that Fa is uniquely determined. Because the image of Q and IO (R)Pa generate Qa
as a WO (R)-module, the map Ḟa is then also uniquely determined.
We return to the notation that P is an o-display with a strict O-action. Applying the functors
AOt /o,R and AO/Ot ,R , we obtain first Pt and then Pa . We find by our definitions that, with the
notation of (3.3.10),
Pa = WO (R) ⊗
P0 = WO (R) ⊗
P. (3.3.31)
O⊗Ot ,κ Wo (R) O⊗o Wo (R)

We note that P0 = (O ⊗Ot ,κ Wo (R)) ⊗O⊗o Wo (R) P .

We already noted that the Ahsendorf functor AOt /o,R is compatible with bilinear forms.
Similar remarks are also valid for the Ahsendorf functor AO/Ot ,R : first one checks that the
functor P 7→ Plt is compatible with bilinear forms of displays, and then applies Proposition
3.2.3 for the compatibility of base change with bilinear forms. Taken together with (3.3.13), we
obtain the following property of the Ahsendorf functor AO/o,R .
Proposition 3.3.15. Consider a bilinear form of Wo (R)-displays,
β : P ′ × P ′′ −→ P,
which is also O-bilinear. Then the bilinear form β : P ′ ×P ′′ −→ P induces by (3.3.31) a WO (R)-
bilinear form βa : Pa′ × Pa′′ −→ Pa . The bilinear form βa is a bilinear form of WO (R)-displays,
βa : Pa′ × Pa′′ −→ Pa .

Remark 3.3.16. In the case where R = k is a perfect field, the description of the Ahsendorf
functor is very simple. We consider the functor AO/Zp ,k which is relevant for us. As a prime
element of Zp we choose p. The element ε ∈ WO (k) is 1. As above, we denote by f, resp. v, the
Frobenius, resp. the Verschiebung, of the ring of Witt vectors W (k). In this case the morphism
(3.3.12) of frames µ : AO/Zp ,k −→ WOt (k), and the morphism of frames µ : Flt (k) −→ WO (k)
of Proposition 3.3.12 are isomorphisms. Therefore we identify WO (k) with the frame
(O ⊗Ot W (k), πO ⊗Ot W (k), k, ff , ff π −1 ). (3.3.32)
This is a perfect frame with u = θ = π, cf. Definition 3.1.8.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 35

Let (P, F, V ) be a W(k)-Dieudonné module with a strict O-action. We have the decomposition
P = ⊕m Pm cf. (3.3.10). The summand P0 is an O ⊗Ot W (k)-module. Since the action of O is
strict, we find
πP0 ⊂ Q0 = V f P0 .
Therefore we can define
Va = V f , Fa = V −f π : P0 −→ P0 . (3.3.33)
Then (P0 , Fa , Va ) is a Dieudonné module for the frame (3.3.32). It is the image of (P, F, V ) by
the Ahsendorf functor AO/Zp ,k .
Proposition 3.3.17. Let R ∈ NilpO . We assume that Spec R is connected. Let P be a W(R)-
display with a strict O-action, and let Pa be the image by the Ahsendorf functor AO/Zp ,k . Then
height P = [O : Zp ] heightO Pa .
The right hand side denotes the height of the WO (R)-display Pa in the sense of Definition 3.1.4.
Let α : P1 −→ P2 be an isogeny of W(R)-displays with strict O-action, and let αa : P1,a −→
P2,a be the image by the Ahsendorf functor. Then
height α = [Ot : Zp ] heightO αa .
Let R = k be a perfect field. Let λ1 < . . . < λm be the slopes of P. Then the slopes of Pa are
[O : Zp ]λ1 , . . . , [O : Zp ]λm . The display P with its strict O-action is isogenous to a direct sum
of displays with a strict O-action ⊕m i=1 P(λi ) such that P(λi ) is isoclinic of slope λi .

Proof. It suffices to consider the case where R = k is a perfect field. Then it is a consequence of
the description of the Ahsendorf functor given above, cf. (3.3.33). 

In the end of this subsection, we relate explicitly the deformation theory of a display with
a strict O-action and its image by the Ahsendorf functor. Let S −→ R be an epimorphism
of O-algebras which are p-adic. We assume that the kernel a of this epimorphism is endowed
with divided powers γ relative to o. Then γ induces also divided powers γt on a relative to Ot .
Indeed, let qo be the number of elements in the residue class field of o. Then we set
′′
γt (a) = γ(a)aq−qo = ′′ aq /̟ , a∈a (3.3.34)
By setting γa (a) = γt (a)(̟/π), we obtain divided powers γa relative to O on a.
Let P = (P, Q, F, Ḟ ) be a Wo (S/R)-display with a strict action
ι : O −→ End P.
The definition of strictness is literally the same as Definition 3.3.1. Since (S −→ R, γa ) is an
O-pd-thickening, the O-frame WO (S/R) is defined, cf. Example 3.1.3. The Ahsendorf functor
generalizes to a relative Ahsendorf functor
   
Wo (S/R) − displays
AO/o,S/R : −→ WO (S/R) − displays . (3.3.35)
with strict O-action
The construction is the same but uses some additional arguments, which we will indicate now.
We define the relative Ahsendorf frame for S −→ R,
Ao (S/R) = (Wo (S), Io (S/R), R, ff , ff −1 ḟ), (3.3.36)
where ḟ : Io (S/R) −→ Wo (S) is defined as in Example 3.1.3. This is an Ot -frame by the
homomorphism κ : Ot −→ Wo (S).
From the Ot -action on the Wo (S)-module P we obtain a decomposition, comp (3.3.10),
P = ⊕m∈Z/f Z Pm , Q = ⊕m∈Z/f ZQm . (3.3.37)
We obtain an Ao (S/R)-display Pua = (Pua , Qua , Fua , Ḟua ) by the formulas (3.3.11).
Lemma 3.3.18. The Drinfeld homomorphism µ : Wo (S) −→ WOt (S) induces a morphism of
frames
Ao (S/R) −→ WOt (S/R). (3.3.38)
36 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Proof. We have to prove the formula

Ḟ ′ ff −1 ḟ
(µ(η)) = µ( η), η ∈ Io (S/R).
For η ∈ Io (S), this is (3.3.3). Therefore the formula follows if we show that µ : Wo (S) −→
WOt (S) maps logarithmic Teichmüller representatives of elements in a to logarithmic Teichmüller
representatives. Let ẇo,n be the divided Witt polynomials defined by γ and let ẇOt ,n be the
divided Witt polynomials defined by γt . It follows from the definition of the Drinfeld homomor-
phism (3.3.1) that
ẇOt ,n (µ(ξ)) = ̟(f −1)n ẇo,f n (ξ), ξ ∈ Wo (a). (3.3.39)
This is verified by reducing to a universal case where a is without p-torsion. If now ξ = ã ∈ Wo (a)
is a logarithmic Teichmüller representative, the right hand side of (3.3.39) is 0 for n 6= 0, and is a
for n = 0. This shows that µ(a) is the logarithmic Teichmüller representative of a in WOt (a). 

Applying now base change to Pua relative to (3.3.38), we obtain a WOt (S/R)-display Pt with
a strict O-action. The assignment P 7→ Pt defines the functor
   
Wo (S/R) − displays WOt (S/R) − displays
AOt /o,S/R : −→ .
with strict O-action with strict O-action
Next we define the functor
   
WOt (S/R) − displays
AO/Ot ,S/R : −→ WO (S/R) − displays . (3.3.40)
with strict O-action
We begin with the definition of the relative Lubin-Tate frame Flt (S/R). We start with the frame
WOt (S/R) = (WOt (S), IOt (S/R), F ′ , Ḟ ′ ). Recall that IOt (S/R) = ã ⊕ IOt (S), where the ideal ã
consists of the logarithmic Teichmüller representatives ã of elements a ∈ a, with respect to the
divided powers γt . We have by definition Ḟ ′ (ã) = 0. Tensoring with O⊗Ot , we obtain
F ′ : O ⊗Ot WOt (S) −→ O ⊗Ot WOt (S),
Ḟ ′ : O ⊗Ot IOt (S/R) −→ O ⊗Ot WOt (S).
We define an ideal in O ⊗Ot WOt (S),
J (S/R) = O ⊗Ot IOt (S/R) + (π ⊗ 1 − 1 ⊗ [π])(O ⊗Ot WOt (S)). (3.3.41)
For an element η ∈ J (S/R) we find
ẼK,0 (π ⊗ 1)η ∈ O ⊗Ot IOt (S/R).
Indeed, the factor ring O ⊗Ot WOt (S)/O ⊗Ot IOt (S/R) = O ⊗Ot R is annihilated by EK (π ⊗ 1).
As before in the definition of the Lubin-Tate frame, we define maps σlt : O ⊗Ot WOt (S) −→
O ⊗Ot WOt (S) and σ̇lt : J (S/R) −→ O ⊗Ot WOt (S) by
F′ Ḟ ′ Ḟ ′
σlt
ξ= ξ, σ̇lt
η= ẼK (π ⊗ 1)−1 (ẼK,0 (π ⊗ 1)η),
with ξ ∈ O ⊗Ot WOt (S), η ∈ J (S/R). The justification of the following definition is analogous
to the justification of Definition 3.3.8 of the Lubin-Tate frame.
Definition 3.3.19. The O-frame
Flt (S/R) := (O ⊗Ot WOt (S), J (S/R), σlt , σ̇lt )
is called the relative Lubin-Tate frame corresponding to the epimorphism of O-algebras S −→ R
and the divided powers γt relative to Ot on the kernel a.
Lemma 3.3.20. The Drinfeld homomorphism
µ : O ⊗Ot WOt (S) −→ WO (S)
defines a morphism of O-frames
ε
Flt (S/R) −→ WO (S/R). (3.3.42)
The last frame is defined by (3.3.26).
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 37

Proof. One can argue exactly as in the proof of Proposition 3.3.12, but we need that
µ : WOt (a) −→ WO (a)
maps logarithmic Teichmüller representatives ã ∈ WOt (a) with respect to the Ot -divided powers
γt to logarithmic Teichmüller representatives ã ∈ WO (a) with respect to the O-divided powers
γa . This is a consequence of the following relation of divided Witt polynomials,
 ̟ n
ẇO,n (µ(α)) = ẇOt ,n (α), α ∈ WOt (a).
π
Again we may restrict to the p-torsionfree case, where this formula follows immediately from the
definition of µ, cf. (3.3.1). 
Let Pt be a WOt (S/R)-display with a strict O-action. Then the R-module Pt /Qt is annihilated
by π ⊗ 1 − 1 ⊗ [π]. As in Proposition 3.3.11, we define Flt : P −→ P by
Flt (x) = Ḟt ((π ⊗ 1 − 1 ⊗ [π])x), x ∈ P.
We set Plt = Pt , Qlt = Qt , Ḟlt = Ḟt . Then we obtain a Flt (S/R)-display Plt = (Plt , Qlt , Flt , Ḟlt ).
ε
If we apply the base change by (3.3.42), we obtain a WO (S/R)-display Paε = (Paε , Qεa , Faε , Ḟaε ).
ε ε −1 ε ε
The assignement Pt 7→ Pa = (Pa , Qa , ε Fa , Ḟa ) is the desired relative Ahsendorf functor
AO/Ot ,S/R .
Proposition 3.3.21. Let P be a Wo (S/R)-display with a strict O-action. Let Pa be the
WO (S/R)-display associated to it by the relative Ahsendorf functor AO/o,S/R . Then there is
a canonical isomorphism
Pa /IO (S)Pa ∼
= S ⊗O⊗ S (P/Io (S)P ).
o (3.3.43)
Proof. This is an immediate consequence of (3.3.31) 
With the notation P = Pua of (3.3.37), we may write
Pa /IO (S)Pa = Pua /Io (S)Pua + (π ⊗ 1 − 1 ⊗ [π])Pua . (3.3.44)
To see this, one uses that π ⊗ 1 − 1 ⊗ π generates the kernel of the canonical map O ⊗ O −→ O Ot
as an ideal. We see that Pa /IO (S)Pa is the biggest quotient of Pua /Io (S)Pua such that the action
via ι and via the structure homomorphism O −→ S agree.
Let R be an O-algebra R such that π is nilpotent in R. Let P be a Wo (R)-display with strict
O-action. We assume that P is nilpotent. Then Pa is also nilpotent. Then there is a crystal
DP on the category of o-pd-thickenings and a crystal DPa on the category of O-pd-thickenings
associated to these displays.
Corollary 3.3.22. Let P be a nilpotent Wo (R)-display with a strict O-action. Then the image
Pa by the Ahsendorf functor is a nilpotent WO (R)-display. Let S −→ R be a surjective map
of O-algebras which are p-adic. Assume that the kernel a of this epimorphism is endowed with
divided powers γ relative to o. Let γa be the corresponding O-divided powers on a. There is a
canonical isomorphism
DPa (S, γa ) = S ⊗(O⊗o S) DP (S, γ).
Proof. Indeed, DP (S) is computed from a Wo (S/R)-display P̃ which lifts P and which is unique
up to isomorphism. But then the relative Ahsendorf functor applied to P̃ gives a WO (S/R)-
display P̃a which lifts Pa . We conclude by Proposition 3.3.21. 
Corollary 3.3.23. With the notation of Corollary 3.3.22, the Ahsendorf functor AO/o,S defines
a bijection between the liftings of P to a Wo (S)-display with a strict O-action and the liftings of
Pa to a WO (S)-display.
Proof. We show that each lifting of Pa is in the essential image of AO/o,S . A lifting P̃a of Pa
corresponds, by Grothendieck-Messing for nilpotent displays, to a direct summand Ua ⊂ DPa (S).
Let U ⊂ DP be the preimage of Ua by the natural epimorphism DP −→ DPa . Then U defines a
lifting of P which is mapped by the Ahsendorf functor to P̃a . 
It is straightforward to deduce from the last Corollary Ahsendorf’s Theorem 3.3.2 for an
artinian local ring with perfect residue class field, i.e., we reproved a special case of [1].
38 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

3.4. The Lubin-Tate display. Let K be a finite extension of Qp . Let K t ⊂ K be the maximal
subextension which is unramified over Qp . We denote by Ot ⊂ O the rings of integers and by κ
the common residue class field. We fix a prime element π ∈ O. Let L be a normal extension of
Qp which contains K. We set Lt = K t . Let Φ = HomQp -Alg (K, L) and Ψ = HomQp -Alg (K t , L).
We denote by ϕ0 ∈ Φ and ψ0 ∈ Ψ the identical embeddings. We denote by Φψ the preimage of
ψ by the restriction map Φ −→ Ψ. We define
Y
Eψ (T ) = (T − ϕ(π)) ∈ OL [T ].
ϕ∈Φψ

Clearly this polynomial has coefficients in OLt ⊂ OL . Let E ∈ Ot [T ] be the Eisenstein polynomial
of π in the extension K/K t . Then Eψ is the image of E by ψ in OLt [T ]. We consider the surjective
OL -algebra homomorphismus
OL [T ] −→ O ⊗Ot ,ψ OL ,
which maps T to π ⊗ 1. Then Eψ (π ⊗ 1) = 0.
We lift the polynomials Eψ to the Witt ring
Y
Ẽψ (T ) = (T − [ϕ(π)]) ∈ W (OLt )[T ].
ϕ∈Φψ

We consider the decomposition

Y
O ⊗Zp OLt = O ⊗Ot ,ψ OLt .
ψ∈Ψ

Let σ ∈ Gal(K t /Qp ) be the Frobenius automorphism. We have the morphism λ : Ot −→ W (Ot )
from (3.3.8). We define ψ̃ as the composite
λ W (ψ)
ψ̃ : Ot −→ W (Ot ) −→ W (OLt ).
Then we obtain the decomposition
Y
O ⊗Zp W (OLt ) = O ⊗Ot ,ψ̃ W (OLt ). (3.4.1)
ψ∈Ψ

Let Ẽψ (π ⊗ 1) be the image of Ẽψ by the homomorphism W (OLt )[T ] −→ O ⊗Ot ,ψ̃ W (OLt ) which
maps T to π ⊗ 1. Since Eψ (π ⊗ 1) = 0, we conclude that
Ẽψ (π ⊗ 1) ∈ O ⊗Ot ,ψ̃ I(OLt ). (3.4.2)
For an arbitrary OLt -algebra R, the decomposition (3.4.1) induces
Y
O ⊗Zp W (R) = O ⊗Ot ,ψ̃ W (R). (3.4.3)
ψ∈Ψ

The Frobenius and the Verschiebung act on the left hand side via the second factor, and this
induces on the right hand side the maps
F : O ⊗Ot ,ψ̃ W (R) −→ O ⊗Ot ,ψσ
f W (R)
F
a⊗ξ 7−→ a ⊗ ξ
(3.4.4)
V : O ⊗Ot ,ψσ
f W (R) −→ O ⊗Ot ,ψ̃ W (R)
V
a⊗ξ 7−→ a ⊗ ξ
f = ψ̃ ◦ σ. We will write Ḟ = V −1 : O ⊗ t I(R) −→ O ⊗ t f W (R).
We note that ψσ O ,ψ̃ O ,ψσ

Ḟ
Proposition 3.4.1. The element Ẽψ (π ⊗ 1) ∈ O ⊗Ot ,ψσ
f W (OLt ) is a unit of the form

Ḟ πe
Ẽψ (π ⊗ 1) = ( ⊗ 1)δ, δ ∈ O ⊗Ot ,ψσ
f W (OLt ), (3.4.5)
p
where δ − 1 ⊗ 1 is in the kernel of O ⊗Ot ,ψσ
f W (OLt ) −→ O ⊗Ot ,ψσ
f W (κLt ). In particular, δ − 1 ⊗ 1
lies in the radical of O ⊗Ot ,ψσ
f W (OLt ).

Proof. The proof is the same as that of Proposition 3.3.6. 

 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 39

The polynomial Eψ0 has the decomposition

Eψ0 (T ) = (T − π) · Eψ0 ,0 (T ).
These polynomials have coefficients in K ⊂ L. Recall that ϕ0 (π) = π. We set
Y
Ẽψ0 ,0 (T ) = (T − [ϕ(π)]) ∈ W (OL )[T ].
ϕ∈Φψ0 ,ϕ6=ϕ0

This polynomial lies in W (O)[T ] ⊂ W (OL )[T ]. We set

PL = O ⊗Zp W (O) = ⊕ψ∈Ψ O ⊗Ot ,ψ̃ W (O). (3.4.6)
We denote by Qψ0 ⊂ O ⊗Ot ,ψ˜0 W (O) the kernel of the map
0id⊗w mult.
O ⊗Ot ,ψ˜0 W (O) −−−−→ O ⊗Ot ,ψ0 O −−−→ O. (3.4.7)
Lemma 3.4.2.
Qψ0 = {x ∈ O ⊗Ot ,ψ˜0 W (O) | Ẽψ0 ,0 (π ⊗ 1)x ∈ O ⊗Ot ,ψ˜0 I(O)}
Proof. Let Q̄ψ0 be the kernel of the second map of (3.4.7). Then we can reformulate our assertion
as
Q̄ψ0 = {x ∈ O ⊗Ot ,ψ0 O | Eψ0 ,0 (π ⊗ 1)x = 0}.
We write

O ⊗Ot ,ψ0 O ≃ Ot [T ]/Eψ0 (T )Ot [T ] ⊗Ot ,ψ0 O ≃ O[T ]/Eψ0 (T )O[T ].
We see that Q̄ψ0 = (T − π)O[T ]/Eψ0 (T )O[T ] which coincides with Eψ0 ,0 (T )O[T ]/Eψ0 (T )O[T ].

We let F and V act via the second factor on O ⊗Zp W (O) and therefore on the right hand
side of (3.4.6) by the formulas (3.4.4).
Definition 3.4.3. The Lubin-Tate display is the W(O)-display L = (PL , QL , FL , ḞL ), defined
as follows. Let PL = O ⊗Zp W (O). Then PL = ⊕ψ Pψ,L with Pψ,L = O ⊗Ot ,ψ̃ W (O). Set
Qψ = Pψ,L for ψ 6= ψ0 and Qψ0 ⊂ Pψ0 ,L as above, and define
QL = ⊕ψ∈Ψ Qψ ⊂ PL .
The maps FL and ḞL are defined as the direct sum of the following maps for all ψ. For ψ 6= ψ0
we define
FL : O ⊗Ot ,ψ̃ W (O) −→ O ⊗Ot ,ψσ
f W (O), ḞL : O ⊗Ot ,ψ̃ W (O) −→ O ⊗Ot ,ψσ
f W (O).
F Ḟ
x 7−→ (Ẽψ (π ⊗ 1)x), y 7−→ (Ẽψ (π ⊗ 1)y)
For ψ0 we define
FL : O ⊗Ot ,ψ̃0 W (O) −→ O ⊗Ot ,ψg
0σ
W (O), ḞL : Qψ0 −→ O ⊗Ot ,ψg
0σ
W (O).
F Ḟ
x 7−→ (Ẽ0 (π ⊗ 1)x), y 7−→ (Ẽ0 (π ⊗ 1)y)
The action of O by multiplication via the first factor on O ⊗Zp W (O) defines a strict O-action
on L. If R is a p-adic O-algebra we denote by LR the base change of L via the morphism of
frames W(O) −→ W(R).
The tuple (PL , QL , FL , ḞL ) is indeed a W(O)-display. The only non-trivial point is that ḞL
is surjective. But this follows from Proposition 3.4.1.
We will now apply the Ahsendorf functor AO/Zp ,O to the Lubin-Tate display L. We use the
previous section in the case o = Zp and ̟ = p. We set ψi = ψ0 σ i : Ot −→ O, where ψ0 is the
identity. We have the decomposition
−1
PL = ⊕fi=0 PL,ψi , PL,ψi = O ⊗Ot ,ψ̃i W (O).
We denote the Frobenius on W (O) by F . In the last section we have associated to L an A(O)-
display Pua for the Ot -frame
AZp (O) = A(O) = (W (O), I(O), F f , F f −1 Ḟ ).
40 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

By definition Pua = PL,ψ0 , Qua = QL,ψ0 , and Ḟua = ḞLf . More explicitely, for y ∈ Qua ,
Ḟ F Ḟ F f −2 Ḟ F f −1 Ḟ
Ḟua (y) = Ẽψf −1 (π ⊗ 1) Ẽψf −2 (π ⊗ 1) · . . . · Ẽψ1 (π ⊗ 1) (Ẽ0 (π ⊗ 1)y).
We set
Ḟ F Ḟ F f −2 Ḟ F f −1 Ḟ
n= Ẽψf −1 (π ⊗ 1) Ẽψf −2 (π ⊗ 1) · . . . · Ẽψ1 (π ⊗ 1) Ẽψ0 (π ⊗ 1).
Then we may write
F f −1 Ḟ

−1 F f −1 Ḟ
Ḟua (y) = n Ẽψ0 (π ⊗ 1) (Ẽ0 (π ⊗ 1)y). (3.4.8)
To Pua we apply base change with respect to the Drinfeld morphism µ : A(O) −→ WOt (O). We
obtain the WOt (O)-display Pt , where
Pt = O ⊗Ot WOt (O)
and where Qt is the kernel of the homomorphism O ⊗Ot WOt (O) −→ O induced by wOt ,0 , cf.
(3.4.7). The polynomials Ẽψ0 and Ẽψ0 ,0 are mapped by µ to the polynomials ẼK and ẼK,0 of
(3.3.14). We denote by nlt the image of n by µ. Therefore we obtain
′
−1 Ḟ ′
Ḟt (y) = nlt Ḟ ẼK (π ⊗ 1) (ẼK,0 (π ⊗ 1)y), y ∈ Qt .
By Proposition 3.3.11, we associate to Pt a Flt (O)-display Plt . In terms of the Lubin-Tate frame
(comp. Definition 3.3.8), we may rewrite the last equation as
Ḟlt (y) = nlt σ˙lt (y), y ∈ Qt = Qlt = J .
Proposition 3.4.4. The Ahsendorf functor AO/Zp ,O maps the Lubin-Tate display L to a WO (O)-
display which is canonically isomorphic to Pm,WO (O) (π ef /pf ), i.e., the twist of the multiplicative
display by π ef /pf ∈ O ⊂ WO (O), cf. Example 3.1.6.
Proof. Let Pm,lt be the multiplicative Flt (O)-display. The above identities show that the display
Lt is equal to the Flt (O)-display Pm,lt (nlt ). Applying to this display the base change by the
Drinfeld morphism of frames, cf. Proposition 3.3.12, we obtain essentially (i.e., neglecting ε)
the image of L by the Ahsendorf functor. By Proposition 3.4.1 the image of the element n by
the map O ⊗Ot ,ψ̃0 W (O) −→ O ⊗Ot ,ψ0 W (κ) = O is π ef /pf . This implies that the image of
nlt ∈ O ⊗Ot WOt (O) in O ⊗Ot WOt (κ) is π ef /pf . By the following Lemma there is a uniquely
f
determined unit ξ ∈ O ⊗Ot WOt (O) such that ξnlt = (π ef /pf ⊗ 1) F ξ. This gives a canonical
∼
→ Pm,lt (π ef /pf ⊗ 1). The base change with respect to a morphism of
isomorphism Pm,lt (nlt ) −
frames maps the multiplicative display to the multiplicative display, cf. Example 3.1.7. Therefore
ε
the last display is mapped by the base change of Proposition 3.3.12 to the WO (O)-display


WO (O), IO (O), F, ε−1 Ḟ )(π ef /pf ⊗ 1 .
Here we denote by F the Frobenius acting on WO (O) and by Ḟ the inverse of the Verschiebung.
The element ε is defined by (3.3.17).
Therefore the definition of the Ahsendorf functor gives Pm,WO (O) (ε−1 (π ef /pf ⊗ 1)) as the
image of L. The image of ε by the homomorphism WO (O) −→ WO (κ) is 1. A variant of the
next Lemma shows that there is a unique ξ ∈ WO (O) such that F ξξ = ε. This shows that the
last display is canonically isomorphic to Pm,WO (O) (π ef /pf ⊗ 1). 

Lemma 3.4.5. Let α ∈ O ⊗Ot WOt (O) be a unit whose image in O ⊗Ot WOt (κ) is 1. Then
there exists a unique unit ξ ∈ O ⊗Ot WOt (O) whose image in O ⊗Ot WOt (κ) is 1 and such that
Ff
ξ · ξ −1 = α.
Proof. One proves this by induction on n for O ⊗Ot WOt (O/π n O). Alternatively one can use
Grothendieck-Messing for frames due to Lau and show that the multiplicative display of Flt (κ)
has no nontrivial deformation with respect to Flt (O) −→ Flt (κ). 

Remark 3.4.6. Let k a perfect field which is an extension of κ. We regard it as an O-algebra

via the residue class map O −→ κ. Then we can describe the Dieudonné module (PLk , FLk , VLk )
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 41

of Lk as follows. Let ψ̃0 : Ot = W (κ) −→ W (k) be the canonical map. The set Ψ consists of the
maps ψ̃0 σ i where σ is the Frobenius of the extension Ot /Zp . We have
Y
PLk = O ⊗Zp W (k) = O ⊗Ot ,ψ̃ W (k).
ψ̃∈Ψ

The Frobenius and the Verschiebung

k
FL
O ⊗Ot ,ψ̃ W (k) −→
←− O ⊗Ot ,ψ̃σ W (k) (3.4.9)
VLk

are defined as follows:

−1
FLk (xψ ) = π e F
xψ , VLk (xψσ ) = (p/π e ) F xψσ , for ψ 6= ψ0 ,
−1
FLk (xψ0 ) = π e−1 F
xψ0 , VLk (xψ0 σ ) = (p/π e−1 ) F xψ0 σ .
The upper left index F denotes the action of the Frobenius via the second factor on O ⊗Zp W (k).
This description follows easily because Ẽψ (T ) = T e , and therefore Ẽψ (π ⊗ 1) = π e ⊗ 1 ∈
O ⊗Ot ,ψ̃ W (k) and Ẽψ0 ,0 (π ⊗ 1) = π e−1 ⊗ 1. Moreover, we have Qψ0 = πO ⊗Ot ,ψ̃ W (k).
We have identified the frame WO (k) with


O ⊗Ot ,ψ̃0 W (k), πO ⊗Ot ,ψ̃0 W (k), k, F f , F f π −1 .

By Remark 3.3.16, the Ahsendorf functor associates to the Dieudonné module (PLk , FLk , VLk )
of Lk the WO (k)-Dieudonné module
π ef f pf

O ⊗Ot ,ψ̃0 W (k), F , ef −1 F −f . (3.4.10)
pf π
This is equal to the Dieudonné module of the twisted multiplicative display Pm,WO (k) (π ef /pf ),
in agreement with Proposition 3.4.4.
In the end of this subsection we discuss the Faltings dual of a display P = (P, Q, F, Ḟ ) with
a strict O-action over an O-algebra R. We begin with a recipe how to construct such P. We
consider the decomposition induced by (3.4.3),
P = ⊕ψ Pψ , Q = Qψ0 ⊕ (⊕ψ6=ψ0 Pψ ).
Let


Jψ0 = Ker O ⊗Ot ,ψ̃0 W (R) −→ R , (3.4.11)
where the map is induced by the structure homomorphism O −→ R and the homomorphism
w0 . Then
Jψ0 = O ⊗Ot ,ψ̃0 I(R) + (π ⊗ 1 − 1 ⊗ [π])(O ⊗Ot ,ψ̃0 W (R)),
cf. (3.3.20). To find a normal decomposition P = T ⊕ L, we start with
Pψ0 = Tψ0 ⊕ Lψ0 , Q ψ 0 = Jψ 0 T ψ 0 ⊕ L ψ 0 .
We define φTψ0 : Tψ0 −→ Pψ0 σ ,


φTψ0 (t) = Ḟ (π ⊗ 1 − 1 ⊗ [π])t .

Then we set T = Tψ0 ⊂ P and L = Lψ0 ⊕ (⊕ψ6=ψ0 Pψ ). Let φL : L −→ P be the restriction of Ḟ

to L and let φT = φTψ0 . The restriction to Lψ0 is denoted by φLψ0 .
Lemma 3.4.7. The map
φT ⊕ φL : T ⊕ L −→ P
is an F -linear O ⊗Zp W (R)-module isomorphism, where the Frobenius F acts on O ⊗Zp W (R)
via the second factor.
42 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Proof. For ψ 6= ψ0 the map Ḟψ : Pψ −→ Pψσ is an F -linear isomorphism. Therefore it suffices
to show that the map
φT ⊕ φLψ0 : Tψ0 ⊕ Lψ0 −→ Pψ0 σ (3.4.12)
is an F -linear isomorphism or, equivalently, an F -linear epimorphism. Since Pψ0 σ is generated
by Ḟ (Qψ0 ) it suffices to show that the following elements are in the image of the linearization
of (3.4.12):
V
Ḟ (ℓ), Ḟ ((π ⊗ 1 − 1 ⊗ [π])t), Ḟ ( ηt), ℓ ∈ Lψ0 , t ∈ T, η ∈ O ⊗Zp W (R).
For the first two elements this is obvious. For the last element this follows from the formula

−1 F

F x = Ḟ Ẽψ0 (π ⊗ 1) · Ẽ0 (π ⊗ 1) · Ḟ (π ⊗ 1 − 1 ⊗ [π])x , x ∈ Pψ0 , (3.4.13)
which is proved in the same way as Lemma 3.3.10. 
We omit the proof of the following proposition.
Proposition 3.4.8. Let R be a p-adic O-algebra. Let Tψ0 , Lψ0 be O⊗Ot ,ψ̃0 W (R)-modules and let
Pψ for ψ 6= ψ0 be O ⊗Ot ,ψ̃ W (R)-modules which are free locally on Spec R. Set Pψ0 = Tψ0 ⊕ Lψ0 .
Assume given F -linear isomorphisms
φTψ0 ⊕ φLψ0 : Tψ0 ⊕ Lψ0 −→ Pψ0 σ , φψ : Pψ −→ Pψσ , for ψ 6= ψ0 ,
cf. (3.4.4) for the meaning of F -linear.
Then there exists a unique display P = (P, Q, F, Ḟ ) with a strict action of O over R such
that P = ⊕ψ Pψ and Q = Jψ0 Tψ0 ⊕ (⊕ψ6=ψ0 Pψ ) with T = Tψ0 and L = Lψ0 ⊕ (⊕ψ6=ψ0 Pψ ) and
such that φT = φTψ0 : T −→ P and φL = φψ0 ⊕ (⊕ψ6=ψ0 φψ ) : L −→ P are given by the display
structure of P as in Lemma 3.4.7. 
Let P be a display with a strict action by O over R, as above. Then we set
P ∇ = HomO⊗Zp W (R) (P, O ⊗Zp W (R)) = ⊕ψ Pψ∇ , where
Pψ∇ = HomO⊗Ot ,ψ̃ W (R) (Pψ , O ⊗Ot ,ψ̃ W (R)).
We define
Q∇ ∇ ∇
ψ0 = {x̂ ∈ Pψ0 | x̂(Qψ0 ) ⊂ Jψ0 } ⊂ Pψ0 , Q∇ = Q∇ ∇
ψ0 ⊕ (⊕ψ6=ψ0 Pψ ).
Let
h , ican : P × P ∇ −→ O ⊗Zp W (R) (3.4.14)
be the canonical perfect O ⊗Zp W (R)-bilinear form. It induces pairings Pψ × Pψ∇ −→ O ⊗Ot ,ψ̃
W (R). Under the perfect R-bilinear form
Pψ0 /Jψ0 Pψ0 × Pψ∇0 /Jψ0 Pψ∇0 −→ (O ⊗Ot ,ψ̃ W (R))/Jψ0 ≃ R,
the R-submodules Qψ0 /Jψ0 Pψ0 and Q∇ ∇
ψ0 /Jψ0 Pψ0 are orthogonal complements.

Proposition 3.4.9. Let P be a display with a strict O-action over R. Let P ∇ and Q∇ as above.
Then there are unique F -linear homomorphisms of O ⊗Zp W (R)-modules
F ∇ : P ∇ −→ P ∇ , Ḟ ∇ : Q∇ −→ P ∇
such that P ∇ = (P ∇ , Q∇ , F ∇ , Ḟ ∇ ) becomes a display and such that the bilinear form (3.4.14)
defines a bilinear form of displays with a strict O-action,
h , ican : P × P ∇ −→ L. (3.4.15)
∇
We call P the Faltings dual of P.
Proof. It follows from the definition that
hQ, Q∇ ican ⊂ QL ⊂ O ⊗Zp W (R).
We will define a display P ∇ = (P ∇ , Q∇ , F ∇ , Ḟ ∇ ). Then we will show that
hḞ y, Ḟ ∇ ŷican = ḞL hy, ŷican y ∈ Q, ŷ ∈ Q∇ . (3.4.16)
This will show that (3.4.15) is a bilinear form of displays with an O-action. Since the pairing
(3.4.14) is perfect, the map Ḟ ∇ is uniquely determined by (3.4.16).
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 43

We begin by defining P ∇ . We chose a normal decomposition

Pψ0 = Tψ0 ⊕ Lψ0 , Q ψ 0 = Jψ 0 T ψ 0 ⊕ L ψ 0 (3.4.17)
Let Tψ∇0
be the orthogonal complement of Tψ0 and let L∇
ψ0 be the orthogonal complement of Lψ0
with respect to the perfect bilinear form
Pψ0 × Pψ∇0 −→ O ⊗Ot ψ̃0 W (R).
We consider the maps φTψ0 and φLψ0 . We define maps
φTψ∇ : Tψ∇0 −→ Pψ∇0 σ , φL∇
ψ
: L∇ ∇
ψ0 −→ Pψ0 σ
0 0

by the equations
Ḟ
hφTψ0 (t) + φLψ0 (ℓ), φTψ∇ (t̂)ican = Ẽψ0 (π ⊗ 1) F hℓ, t̂ican , t ∈ T ψ0 , ℓ ∈ L ψ0
0
(3.4.18)
hφTψ0 (t) + φLψ0 (ℓ), φL∇
ψ
(ℓ̂)ican = Ḟ
Ẽψ0 (π ⊗ 1) F ht, ℓ̂ican , t̂ ∈ Tψ∇0 , ℓ̂ ∈ L∇
ψ0 .
0

This definition makes sense because φTψ0 ⊕φLψ0 : Tψ0 ⊕Lψ0 −→ Pψ0 σ is an F -linear isomorphism.
For ψ 6= ψ0 the map Ḟ : Pψ −→ Pψσ is an F -linear isomorphism. Therefore we can define
Ḟ ∇ : Pψ∇ −→ Pψσ
∇
by the equation

hḞ z, Ḟ ∇ ẑican = Ḟ
Ẽψ (π ⊗ 1) F hz, ẑican , z ∈ Pψ , ẑ ∈ Pψ∇ .
We now apply Proposition 3.4.8 to the modules Tψ∇0 , L∇ ∇
ψ0 , and Pψ for ψ 6= ψ0 , and to the
maps φTψ∇ , φL∇
ψ
, and φψ = Ḟψ∇ for ψ 6= ψ0 . This concludes the definition of the display
0 0

P ∇ = (P ∇ , Q∇ , F ∇ , Ḟ ∇ ).
Now we verify (3.4.16). Let ψ 6= ψ0 . If y ∈ Pψ and ŷ ∈ Pψ∇ , the right hand side is by definition
Ḟ
Ẽψ (π ⊗ 1) · F(y, ŷ)can .
Therefore (3.4.16) holds in this case by the definition of Ḟψ∇ . For ψ0 we use the normal decompo-
sition (3.4.17) and the induced normal decomposition Q∇ ∇ ∇
ψ0 = Jψ0 Tψ0 ⊕ Lψ0 . Using the definition
(3.4.11) of Jψ0 , the identity (3.4.16) becomes for the ψ0 -components a series of equations:


(1) hḞ (π ⊗ 1 − 1 ⊗ [π])t , Ḟ ∇ ℓ̂ican = ḞL h(π ⊗ 1 − 1 ⊗ [π])t, ℓ̂ican ,
(2) hḞ ( V
ηt), Ḟ ∇ ℓ̂ican = ḞL h V
ηt, ℓ̂ican , η ∈ O ⊗Ot ,ψ̃0 σ W (R),
∇


(3) hḞ ℓ, Ḟ (π ⊗ 1 − 1 ⊗ [π])t̂ ican = ḞL hℓ, (π ⊗ 1 − 1 ⊗ [π])t̂ican ,
(4) hḞ ℓ, Ḟ ∇ ( V
η t̂)ican = ḞL hℓ, V
η t̂ican ,
∇
(5) hḞ ℓ, Ḟ ℓ̂ican = 0,
(6) hḞ (Jψ0 Tψ0 ), Ḟ ∇ (Jψ0 Tψ∇0 )ican = 0.
We compute the right hand side of equation (1):
RHS(1) =Ḟ Ẽ0 (π ⊗ 1)(π ⊗ 1 − 1 ⊗ [π]) F ht, ℓ̂ican = Ḟ
Ẽψ0 (π ⊗ 1) F ht, ℓ̂ican .
Therefore equation (1) is exactly the second equation of (3.4.18) for ℓ = 0. The equation (3)
follows in the same way.
We prove now the equation (2). For the right hand side we find:
F
RHS(2) = ηFL ht, ℓ̂ican = η Ẽ0 (π ⊗ 1) F ht, ℓ̂ican .
V
We compute the left hand side of (2) by applying (3.4.13) to Ḟ ( ηt) = ηF t:
hḞ ( V
ηt), Ḟ ∇ ℓ̂ican = η Ḟ
Ẽψ0 (π ⊗ 1)−1 F
Ẽ0 (π ⊗ 1)hḞ ((π ⊗ 1 − 1 ⊗ [π])t), Ḟ ∇ ℓ̂ican
=η Ḟ
Ẽψ0 (π ⊗ 1)−1 F
Ẽ0 (π ⊗ 1) Ḟ Ẽψ0 (π ⊗ 1) F ht, ℓ̂ican .
Here the last equation follows from (1). This proves (2). In the same way we obtain (4) from
equation (3). The equation (5) follows from the second equation of (3.4.18) for t = 0.
Finally we prove equation (6). The special case
hḞ ((π ⊗ 1 − 1 ⊗ [π])t), Ḟ ∇ ((π ⊗ 1 − 1 ⊗ [π])t̂)ican = 0 (3.4.19)
44 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

is exactly the first equation of (3.4.18) for ℓ = 0. We have to show that the same holds if we
replace the first argument of the bilinear form in (3.4.19) by Ḟ ( V ηt) or the second argument by
Ḟ ∇ ( V η t̂). But this may be reduced to (3.4.19) in the same way as in the proof of equation (2).
This finishes the proof of (3.4.16). 
Proposition 3.4.10. Let P1 and P2 be displays over R with a strict O-action. Then the natural
map
HomO-displays (P2 , P1∇ ) −→ BilO-displays (P1 × P2 , LR )
is an isomorphism. Here we consider bilinear forms of displays with a strict O-action which are
also O-bilinear.
Proof. We define the inverse map. Let β be an element from the right hand side. This is in
particular a O ⊗ W (R)-bilinear form
β : P1 × P2 −→ O ⊗ W (R).
On the other hand, we have the canonical perfect O ⊗ W (R)-bilinear form h , ican : P1 × P1∇ −→
O⊗W (R). Since this is perfect, we can define a O⊗W (R)-module homomorphism α : P2 −→ P1∇
by
β(x1 , x2 ) = hx1 , α(x2 )ican .
We omit the straightforward verification that α defines a morphism of displays. 
Theorem 3.4.11. Let R be an O-algebra such that p is nilpotent in R. Let P1 and P2 be displays
over R with a strict O-action. We denote by P1,a and P2,a their images by the Ahsendorf functor
AO/Zp ,R . Proposition 3.3.15 and Proposition 3.4.4 define a homomorphism
BilO-displays (P1 × P2 , LR ) −→ BilWO (R)-displays (P1,a × P2,a , Pm,WO (R) (π ef /pf )). (3.4.20)
If the displays P1∇ and P2 are nilpotent, the homomorphism (3.4.20) is an isomorphism.
Equivalently, (3.4.20) is an isomorphism if (P1,a )∨ and P2,a are nilpotent WOF (R)-displays.
Proof. We apply (3.4.20) to P2 = P1∇ and the canonical bilinear form. We obtain a bilinear
form
P1,a ⊗ (P1∇ )a −→ Pm,WO (O) (π ef /pf ),
which is perfect by Proposition 3.3.15. After twisting, we obtain also a perfect pairing of P1,a
and (P1∇ )a ((π ef /pf )−1 ) with values in Pm,WO (O) . Therefore we have an identification with the
dual display

(P1,a )∨ ∼
= (P1∇ )a (π ef /pf )−1 .
By Theorem 3.3.2 we have an isomorphism
∼
HomO-displays (P2 , P1∇ ) −→ HomWO (R)-displays (P2,a , (P1∇ )a ).
Here the left hand side agrees with the left hand side of (3.4.20) by Proposition 3.4.10. We have
seen that the right hand side is the same as
∼ BilW (R)-displays (P1,a × P2,a , Pm,W (R) (π ef /pf )).
HomW (R)-displays (P2,a , P ∧ (π ef /pf )) =
O 1,a O O

The last isomorphism follows from (3.2.5). 

Definition 3.4.12. Let R ∈ NilpO and let P be a display with a strict O-action. A relative
polarization of P with respect to O is a polarization of the WO (R)-display Pa obtained from P
by the Ahsendorf functor, cf. Definition 3.2.5.
Let Ŏ be the completion of the maximal unramified extension of O. We consider Theorem
3.4.11 in the case of an Ŏ-algebra R. We denote by τ ∈ Gal(Ŏ/O) the Frobenius automorphism.
Since π e /p ∈ Ŏ is a unit we find η0 ∈ Ŏ× such that τ (η0 )η0−1 = π e /p. By Lemma 3.3.4 there
is a τ − F -equivariant homomorphism Ŏ −→ WO (Ŏ) such that the composite with w0 is the
identity.
Let R ∈ NilpŎ . We denote by η0,R the image of η0 by the homomorphism
Ŏ −→ WO (Ŏ) −→ WO (R).
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 45

f
Then multiplication by η0,R defines an isomorphism of WO (R)-displays,
∼
Pm,WO (R) (π ef /pf ) −→ Pm,WO (R) . (3.4.21)
Therefore we may write Theorem 3.4.11 without the twist (π ef /pf ).

Corollary 3.4.13. Let R ∈ NilpŎ . Fix η0 ∈ Ŏ× with τ (η0 )η0−1 = π e /p, which defines the
isomorphism (3.4.21). Let P be a display with a strict O-action over R such that P and P ∇ are
nilpotent. Then a relative polarization on P with respect to O is the same thing as an isogeny
of displays with an O-action P −→ P ∇ such that the induced bilinear form
P × P −→ LR
is alternating. 
In the situation of the corollary, P is the display of a formal p-divisible group X with a strict
O-action and P ∇ is the display of a formal p-divisible group with a strict O-action which we
denote by X ∇ . We call X ∇ the Faltings dual of X. However, we do not relate our definition
to that of Faltings in [11], which operates directly in the realm of p-divisible groups. We can
consider a relative polarization as an isogeny of p-divisible groups with an O-action,
X −→ X ∇ . (3.4.22)

4. The contracting functor

We return to the notation of section 2. In particular, throughout this section, K/F denotes
an etale extension of degree two of a p-adic field, and r denotes a generalized CM-type of rank
2.
Let E be the reflex field of r, and let Ẽ ⊂ Q̄p be its normal closure. As in subsection 2.2,
E ′ ⊂ Ẽ is the composite of E and the normal closure of K t in Q̄p .

4.1. The aim of this section.

Definition 4.1.1. Let S be a scheme over Spec OE ′ such that p is locally nilpotent in OS . We
denote4 by Pr,S the category of local CM-pairs of type r over S which satisfy the Eisenstein
conditions (ECr ). If S = Spec R, we will also write Pr,R or simply PR .
The category of local CM-pairs (P, ι) of type r in the sense of displays which satisfy the
Eisenstein conditions will be denoted by dPr,S , resp., dPr,R .
We will define a functor C′r,R that associates to a CM-pair (P, ι) ∈ dPr,R a new display
(P ′ , ι′ ) over R with an action ι′ : OK −→ End P ′ . When r is special relative to ϕ0 : F −→ Qp ,
ϕ0
cf. Definition 2.1.1, then the restriction of ι′ to OF is strict with respect to OF −→ OE ′ −→ R.
′ ′
If r is banal, then P is étale. We will call the functor Cr,R the pre-contracting functor. Under
suitable hypotheses, the pre-contracting functor will be an equivalence of categories.
We will also describe what C′r,R does with polarizations, and define a functor C′,pol r,R on the
pol
category Pr,S , defined as follows.

Definition 4.1.2. We denote by Ppol r,S the category of polarized local CM-pairs (X, ι, λ) of type
r over S such that (X, ι) satisfies the Eisenstein conditions (ECr ). If S = Spec R, we also
write Ppol pol
r,R . We denote by dPr,S the corresponding category of local CM-triples in the sense
of displays. Explicitly, dPpol
r,R denotes the category of triples (P, ι, β) where (P, ι) ∈ dPr,R and
where β : P × P −→ Pm,W(R) is a polarization such that
β(ι(a)x1 , x2 ) = β(x1 , ι(ā)x2 ), a ∈ OK , x1 , x2 ∈ P.
In the sequel, we will abbreviate the last condition into saying that β is anti-linear for the
OK -action.

4The symbol P is to remind us that this is a category of local CM-pairs.

46 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

In a second step, we will use the functors C′r,R , resp. C′,pol

r,R , to define contracting functors.
Here, we make a distinction between the case when r is special and the case when r is banal. In
the case when r is special, the image of the contracting functor is a WOF (R)-display Pc , endowed
with an action ι of OK such that Pc is of height 4 and dimension 2 (and, in the polarized case,
with a polarization). When r is banal, the image is a p-adic étale sheaf G in Zp -flat modules of
rank 4d, endowed with an action of OK (and with a polarization form in the polarized case).

4.2. The Kottwitz condition for CM-pairs. The Kottwitz condition (KCr ) can be formu-
lated in terms of polynomial functions. Let L be a locally free R-module equipped with an action
of OK . If S is an R-algebra, we write LS = L ⊗R S.
Let us assume for a moment that R is endowed with an OẼ -algebra structure. For ϕ ∈ Φ =
HomQp -Alg (K, Ẽ) we consider the induced map
ϕ
ϕR : OK −→ OẼ −→ R.
Definition 4.2.1. We say that (L, ι) satisfies the Kottwitz condition relative to r if for each
OẼ -algebra S endowed with an OE -algebra homomorphism R −→ S
Y
detS (a | LS ) = ϕS (a)rϕ , for all a ∈ OK ⊗Zp S. (4.2.1)
ϕ∈Φ

Let A = V(OK ) be the affine space over Spec Zp . The right hand side of this equation may
be considered as a polynomial function on AOẼ . By base change to Ẽ, it is easily shown that
this function is defined on AOE . We note that each factor of the right hand side of (4.2.1) is a
linear polynomial function such that some coefficient is a unit in OẼ . Therefore these factors are
non-zero divisors in Γ(AS , OAS ) for each S. Here we remark that a polynomial in S[U1 , . . . , Ur ]
is a non-zero divisor if one of its coefficients is a non-zero divisor in S.
Because the right hand side of (4.2.1) is a polynomial function on AOE , the condition does
not depend on the OẼ -algebra structure on S. By a theorem of Amitsur, condition (4.2.1) is
equivalent to the Kottwitz condition (KCr ) of (2.2.1) (compare [28]).
For a OE ′ -algebra S we have a decomposition
Y
OK ⊗Zp S = OK ⊗OK t ,ψS S. (4.2.2)
ψ∈Ψ

ψ
Here ψS denotes the composite OK t −→ OE ′ −→ S. Let EψS the image of the Eisenstein
polynomial E ∈ OK t [T ] by the last homomorphism. We have a natural isomorphism
∼
S[T ]/EψS S[T ] −→ OK ⊗OK t ,ψS S, T 7−→ Π ⊗ 1. (4.2.3)
Therefore we may regard an OK ⊗OK t ,ψS S-module M as an S[T ]-module. If U ∈ S[T ] and
x ∈ M , we write U x = U (Π ⊗ 1)x. If U0 ∈ OE ′ [T ], with image U ∈ S[T ], then we write simply
U0 x = U x.
Returning to the R-module L with action by OK , the decomposition (4.2.2) induces a decom-
position
LS = ⊕LS,ψ .
By considering, for given ψ, an element a of (4.2.2) whose components are zero for ψ ′ 6= ψ,
we obtain Y
detS (a | LS,ψ ) = ϕS (a)rϕ , for all a ∈ OK ⊗OK t ,ψ S. (4.2.4)
ϕ|ψ

We call this condition (KCψ,r ). The condition (KCr ) holds iff the conditions (KCψ,r ) hold for
each ψ.
We will call ψ banal with respect to r if rϕ ∈ {0, 2} for each ϕ|ψ. We call ψ special with
respect to r if there exists ϕ|ψ such that rϕ = 1 and if for any other ϕ′ |ψ with rϕ′ = 1 we have
ϕ′ = ϕ̄. We note that another ϕ′ can only exist if ψ̄ = ψ.
We also use the conditions (ECψ,r ). The meaning is clear from (2.2.16) and (2.2.18) where
the two conditions (ECψ0 ,r ) and (ECψ,r ) are taken together.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 47

We consider CM-pairs (X, ι) of type r over R ∈ NilpOE , cf. section 2.3. Thus X is a p-divisible
group of height 4d and dimension 2d and ι is a Zp -algebra homomorphism
ι : OK −→ End X
such the rank condition (RCr ) is satisfied. If we speak about the Kottwitz or Eisenstein condition
we refer to the induced action on Lie X. We use a similar terminology when we consider CM-
pairs (P, ι) in the sense of displays. This means that P = (P, Q, F, Ḟ ) is a W(R)-display of
height 4d and dimension 2d endowed with a ring homomorphism
ι : OK −→ End P,
such that the rank condition (RCr ) is satisfied for the induced action on P/Q.
Display theory defines a functor from the category of CM-pairs (X, ι) of type r to the category
of CM-pairs (P, ι) of type r. We set DP = P/IR P and LP = P/Q. If P is the display of X, we
have the identifications
DP = DX (R), LP = Lie X.
Here DX (R) is the Grothendieck-Messing crystal evaluated at R. For an R-algebra S, we will
write DP,S := DP ⊗R S and LP,S := LP ⊗R S. If P is fixed, we write simply DS and LS . If S
is a OE ′ -algebra, (4.2.2) gives a decomposition
DS = ⊕ψ∈Ψ DS,ψ .
Proposition 4.2.2. Let ψ be banal with respect to r. Let (P, ι) be a CM-pair of type r over an
OE -algebra R.
Then the Eisenstein condition (ECψ,r ) is satisfied iff EAψ Dψ is the kernel of the canonical
map Dψ −→ LP,ψ . Moreover (ECψ,r ) implies
the Kottwitz condition (KCψ,r ).
Here EAψ denotes the operator EAψ ι(Π) on the module in question, for a fixed choice of Π,
cf. (2.2.12).
Proof. We reduce to the case where S is an R-algebra endowed with an OẼ -algebra structure.
Then EAψ,S ∈ S[T ] is defined as the image of EAψ by OẼ [T ] −→ S[T ]. It acts on any OK ⊗OK t ,ψ
S-module by (4.2.3).
Via ι, we view DS and LS as OK ⊗Zp S-modules. We consider the canonical surjective map
DS −→ LS .
The decomposition (4.2.2) induces decompositions,
DS = ⊕ ψ Dψ , LS = ⊕ψ Lψ . (4.2.5)
We allowed us to omit the index S on the right hand side of these equations.
The Eisenstein condition (ECψ,r ) for banal ψ says that Lψ is annihilated by EAψ , cf. (2.2.13).
Therefore it is clearly implied by the condition of the Proposition. If conversely (ECψ,r ) holds,
we obtain a surjective map
Dψ /EAψ Dψ −→ Lψ . (4.2.6)
By Lemma 3.1.15, Dψ is locally on Spec S a free OK ⊗OK t ,ψ S-module. It has rank 2 because
the height of P is 4d. We may assume that
Dψ ∼ = (OK ⊗OK t ,ψ S)2 = S[T ]2 /EψS S[T ]2 .
P
We see that both sides of (4.2.6) are locally free S-modules of the same rank rψ = ϕ|ψ rϕ .
Therefore this map is an isomorphism.
The condition (KCψ,r ) would follow from
Y
detS (a | S[T ]/EAψ S[T ]) = ϕS (a), a ∈ S[T ]. (4.2.7)
ϕ∈Aψ

We have Y
EAψ (T ) = (T − ϕS (Π ⊗ 1)), ϕS (a) = a(ϕS (Π ⊗ 1)).
ϕ∈Aψ

We obtain (4.2.7) from the following Lemma. 

48 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Lemma 4.2.3. Let R be a ring. Let

s
Y
E(T ) = (T − Πi ), Πi ∈ R
i=1

be a polynomial. A polynomial f (T ) ∈ R[T ] defines by multiplication an endomorphism of the

free R-module R[T ]/ER[T ]. Then
s
Y
detR (f (T ) | R[T ]/ER[T ]) = f (Πi ).
i=1

Proof. One can easily reduces the question to the case where R is a field of characteristic 0
such that E(T ) is a product of different linear factors. For the reduction one starts with a ring
homomorphism
Z[X, Y ] −→ R,
where for the first set of variables X = (X1 , . . . , Xs ), Xi is mapped to Πi and where the second
set of variables Y is mapped to the coefficients of f .
If now R is a field and E is separable, we have a canonical isomorphism of R-modules
∼
R[T ]/ER[T ] −→ ⊕si=1 R.
T 7−→ (Πi )i
Multiplication by f (T ) on the left hand side acts on the right hand side on the i-th factor by
multiplication by f (Πi ). This implies the assertion. 
Corollary 4.2.4. Let r be banal. Let (P, ι) be a CM-pair of type r over an OE -algebra R. Then
the Eisenstein condition (ECr ) implies the Kottwitz condition (KCr ). 
We consider next the case where ψ is special. This means by definition that there is exactly
one pair {ϕ, ϕ̄} such that ϕ|ψ and rϕ = rϕ̄ = 1. When K/F is ramified, we have E ′ = E and
when K/F is unramified, we have [E ′ : E] ≤ 2.
Proposition 4.2.5. Let ψ be special with respect to r. Let R be a OE -algebra such that p is
nilpotent in R. Let (P, ι) be a CM-pair of type r over R which satifies the Eisenstein condition
(ECψ,r ). Let S be a OE ′ -algebra which is endowed with an OE -algebra homomorphism R −→ S.
Then, with the notations of (4.2.5), the canonical map
Dψ /EAψ Dψ −→ Lψ /EAψ Lψ
is an isomorphism.
Proof. Clearly we may assume that S is a local ring with residue class field k. We postpone the
verification that the assertion holds for S = k (compare (4.3.17) and (4.3.20) below).
We begin with the case K/F unramified. Then rankS Lψ = rψ = 2aψ + 1. Let
f : Lψ −→ Lψ (4.2.8)
be the S-module homomorphism given by multiplication with EAψ . From the case of a field, we
deduce that dimk L/f (L) ⊗S k = 2aψ . By (ECψ,r ) we have
2
^
f = 0,
cf. (2.2.18). Therefore we can apply Lemma 4.9 of [28] with s = 1. This says that Lψ /f (Lψ ) is
a free S-module of rank 2aψ . Therefore the canonical map of the proposition is a surjection of
free S-modules of the same rank, and hence an isomorphism.
The argument in the case K/F ramified is similiar. In this case, we have rankS Lψ = rψ =
2aψ + 2 = 2e. The dimension dimk Lψ /f (Lψ ) ⊗S k = 2aψ , as before. In this case the condition
(ECψ,r ) says
^3
f = 0,
cf. (2.2.16). Therefore we may apply Lemma 4.9 loc.cit. with s = 2. We conclude as before. 
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 49

Proposition 4.2.6. Let r be special and K/F unramified. Let R be a OE -algebra such that p is
nilpotent in R. Let (P, ι) be a CM-pair of type r over R which satifies the Eisenstein condition
(ECr ). Then the condition (KCr ) is also satisfied.
Proof. We consider an algebra S as in the last proposition. We keep the notation of (4.2.5). We
need only to verify (KCψ0 ,r ) since the Kottwitz condition is satisfied for ψ banal by Proposition
4.2.2. We have to verify that
Y
detS (a | Lψ0 ) = ϕ0,S (a) · ϕS (a)2 , a ∈ OK ⊗OK t ,ψ0 S. (4.2.9)
ϕ∈Aψ0

Since Dψ0 is locally on S a free OK ⊗OK t ,ψ0 S-module of rank 2, we obtain from the isomorphism
of Proposition 4.2.5 Y
detS (a | Lψ0 /EAψ0 Lψ0 ) = ϕS (a)2 .
ϕ∈Aψ0
The proposition also shows that EAψ0 Lψ0 is a locally free S-module of rank 1. It follows from
the Eisenstein condition (2.2.18) that this module is annihilated by (T − ϕ0,S (Π ⊗ 1)). Hence
an element a ∈ OK ⊗OK t ,ψ0 S = S[T ]/Eψ0 S[T ] acts on EAψ0 Lψ0 as ϕ0,S (a). In particular
detS (a | EAψ0 Lψ0 ) = ϕ0,S (a).
The formula (4.2.9) follows. 
We reformulate the Eisenstein condition in the case where K/F is unramified.
Proposition 4.2.7. Let r be special and K/F unramified. Let ϕ0 , ϕ̄0 ∈ Φ be the two embeddings
such that rϕ0 = rϕ̄0 = 1, and let ψ0 , resp. ψ¯0 , the embeddings induced by ϕ0 , resp. ϕ̄0 . Let R
be a OE ′ -algebra, and let (P, ι) be a CM-pair of type r over R. Let D = DP (R). The CM-pair
(P, ι) satisfies the Eisenstein conditions iff the following conditions hold.
(1) If ψ ∈ Ψ is banal, then the canonical map
Dψ /EAψ Dψ −→ Lieψ X
is an isomorphism.
(2) If ψ ∈ {ψ0 , ψ¯0 }, then the canonical map
Dψ /EAψ Dψ −→ Lieψ X/EAψ Lieψ X
is an isomorphism.
(3) The R-modules EAψ0 Lieψ0 , resp. EAψ̄0 Lieψ̄0 , are locally free of rank 1 and OK acts on
them via
ϕ0,R : OK −→ OE ′ −→ R, resp. ϕ̄0,R : OK −→ OE ′ −→ R.
Proof. This is a consequence of Proposition 4.2.5 and the proof of Proposition 4.2.6. 
We next consider what happens to the Eisenstein conditions when passing from a display to
its conjugate dual, cf. Lemma 2.3.2. We note that we already checked in loc. cit. that the
condition (RCr ) is preserved. Recall that, if (P, ι) is a CM-pair, the conjugate dual (P ∧ , ι∧ ) is
defined by

P ∧ = P ∨ , ι∧ (a) = ι(ā)∧ .
Corollary 4.2.8. Let K/F be unramified and let r be arbitrary or let K/F be split. Let (P, ι) be
a CM-pair over an OE -algebra R such that p is nilpotent in R. If (P, ι) satisfies the Eisenstein
condition (ECr ), then the conjugate dual (P ∧ , ι∧ ) also satisfies (ECr ).
Proof. We have a canonical isomorphism D∧ = HomR (D, R). The resulting perfect pairing
h , i : D × D∧ −→ R (4.2.10)
satisfies
hax, x̂i = hx, āx̂i, a ∈ OK , x ∈ D, x̂ ∈ D∧ . (4.2.11)
This implies that for ψ1 6= ψ̄2 the modules Dψ1 and D∧ ψ2 are orthogonal and that for any ψ the
induced pairing
Dψ × D∧ψ̄ −→ R (4.2.12)
50 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

∧,1
is perfect. Let D1ψ ⊂ Dψ be the kernel of the map Dψ −→ Lψ := Lieψ P and let Dψ̄ ⊂ D∧
ψ̄
∧,1
be defined in the same way. By definition of the dual display, D1ψ and Dψ̄ are orthogonal
complements of each other with respect to (4.2.12). We consider the case ψ ∈ {ψ0 , ψ̄0 }. Recall
that this is possible only in the non-split case. The Eisenstein condition for P says that we have
a split filtration of direct summands of Dψ
Sψ EAψ Dψ ⊂ D1ψ ⊂ EAψ Dψ , (4.2.13)
such that the factor modules are locally free of rank 1. We claim that the orthogonal complement
of EAψ Dψ is Sψ̄ EAψ̄ D∧
ψ̄
. Indeed, by (4.2.11), we have
h EAψ x, x̂i = h x, EBψ̄ x̂i. (4.2.14)
This implies that EAψ Dψ and Sψ̄ EAψ̄ D∧
ψ̄
are orthogonal. Because (4.2.10) is perfect we obtain
a surjection of R-modules
Dψ /EAψ Dψ −→ HomR (Sψ̄ EAψ̄R D∧
ψ̄ , R). (4.2.15)
Recall that Dψ is locally on Spec R a free OK ⊗OK t ,ψ R-module of rank 2. It follows that both
sides of (4.2.15) are locally free R-modules of the same rank 2aψ = 2e − 2aψ̄ − 2. Therefore this
map is an isomorphism. This proves our claim about the orthogonal complement. By the same
argument, EAψ̄R D∧ ψ̄
is the orthogonal complement of Sψ EAψ Dψ .
We take the orthogonal complement of (4.2.13) and obtain the filtration
∧,1
Sψ̄ EAψ̄ D∧ ∧
ψ̄ ⊂ Dψ̄ ⊂ EAψ̄ Dψ̄ ,

and conclude that the factor modules are locally free of rank 1.
∧,1
Now let ψ be banal. We have to prove that EAψ D∧ ψ ⊂ Dψ . The right hand side is the
orthogonal complement of D1ψ̄ = EAψ̄ Dψ̄ . Therefore we have to prove that
h EAψ̄ x, EAψ x̂i = 0, for x ∈ Dψ̄ , x̂ ∈ D∧
ψ.

Using (4.2.14), we find for the right hand side

h x, EBψ EAψ x̂i = h x, Eψ x̂i = 0.

Before proving the analogue of Corollary 4.2.8 in the case when K/F is ramified, we further
analyze in this case the Eisenstein conditions.
Proposition 4.2.9. Let r be special and K/F ramified. Let R be a OE -algebra such that p is
nilpotent in R. Let (P, ι) be a CM-pair of type r over R. Since E ′ = E, the decomposition
(4.2.2) is defined for S = R. Then the Eisenstein condition (ECψ0 ,r ) holds iff the following
conditions are satisfied.
(1) The R-module EAψ0 Lψ0 ⊂ Lψ0 is a direct summand which is locally free of rank 2.
(2) The action of ι(π) on EAψ0 Lψ0 coincides with the action of ϕ0 (π) ∈ R, i.e., the action of
ϕ0
the image of π by the homomorphism OF −→ OE −→ R.
Furthermore, a CM-pair (P, ι) which satisfies (ECr ) also satisfies the Kottwitz condition
(KCr ) if and only if
TrR (ι(Π) | EAψ0 Lψ0 ) = 0. (4.2.16)
Proof. For the proof we may pass from R to an R-algebra S which is endowed with a OẼ -algebra
stucture. We continue with the notations of (4.2.5). The first assertion of the proposition is then
an immediate consequence of Proposition 4.2.5. By Proposition 4.2.6, we need only to consider
(KCψ0 ,r ). The condition reads
Y
detS (a | Lψ0 ) = ϕ0,S (a) · ϕ̄0,S (a) · ϕS (a)2 , a ∈ OK ⊗OK t ,ψ0 S.
ϕ∈Aψ0

In this case EAψ0 Lψ0 is a locally free S-module of rank 2. By Proposition 4.2.5, it is enough to
show
detS (a | EAψ0 Lψ0 ) = ϕ0,S (a) · ϕ̄0,S (a). (4.2.17)
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 51

In this case, the Eisenstein condition says that

(T − ϕ0,S (Π ⊗ 1))(T − ϕ̄0,S (Π ⊗ 1)) = T 2 + ψ0 (π) (4.2.18)
2
annihilates EAψ0 Lψ0 , cf (2.2.16). Note that the action of T on EAψ0 Lψ0 is by definition the
action of ι(Π2 ) = −ι(π). Therefore the action of OF on EAψ0 Lψ0 via ι coincides with the action
ϕ0
via OF −→ OẼ −→ S. The polynomial (4.2.18) is the characteristic polynomial of ι(Π) acting
on the S-module EAψ0 Lψ0 . This follows from the trace condition of the proposition. Therefore
the desired equation (4.2.17) is a consequence of Lemma 4.2.10 below.
Conversely, assume that the Kottwitz condition holds. By Proposition 4.2.5, this implies
Y Y
detS (a | EAψ0 Lψ0 ) · ϕS (a)2 = ϕ0,S (a) · ϕ̄0,S (a) · ϕS (a)2 , a ∈ OK ⊗ S.
ϕ∈Aψ0 ϕ∈Aψ0

We already remarked right after Definition 4.2.1 that ϕS is a non-zero divisor in the ring of
polynomial functions. Therefore we conclude
detS (a | EAψ0 Lψ0 ) = ϕ0,S (a) · ϕ̄0,S (a), for all a ∈ OK ⊗ S.
This implies that the characteristic polynomial of ι(Π) acting on EAψ0 Lψ0 is the polynomial
(4.2.18). Therefore the trace is 0. 
We state the needed Lemma without proof.
Lemma 4.2.10. Let S be a ring. Let L be a locally free S-module of rank 2. Let α : L −→ L be
an endomorphism with characteristic polynomial
detS (T idL − α | L) = T 2 − s1 T + s2 .
Then for all λ, µ ∈ S
detS (µidL − λα) = µ2 − s1 µλ + s2 λ2 .
Assume that the characteristic polynomial splits
T 2 − s1 T + s2 = (T − ρ1 )(T − ρ2 ).
Consider L as S[T ]-module, and let φi : S[T ] −→ S be the S-algebra homomorphism such that
φi (T ) = ρi . Then for each polynomial a ∈ S[T ]
detS (a | L) = φ1 (a) · φ2 (a).

c
Remark 4.2.11. Let A ⊂ B be R-modules. Then we write A ⊂ B if the factor module B/A is
a finitely generated projective R-module of rank c.
Let (P, ι) as in Proposition 4.2.9 such that (ECψ0 ,r ) is satisfied. We write D = P/IR P . Let
Q̄ψ0 the kernel of Dψ0 −→ Lψ0 . By Lemma 3.1.15, Dψ0 is a free OK ⊗OF t ,ψ0 R-module of rank
2. We obtain that
2(e−1)
E A ψ 0 Dψ 0 ⊂ Dψ 0 .
On the other hand, the condition (1) of Proposition 4.2.9 says
2 2(e−1)
Q̄ψ0 ⊂ EAψ0 Dψ0 + Q̄ψ0 ⊂ Dψ 0 .
This implies Q̄ψ0 ⊂ EAψ0 Dψ0 . Therefore we may reformulate the two conditions in Proposition
4.2.9 in one line:
2 2
Sψ0 EAψ0 Dψ0 ⊂ Q̄ψ0 ⊂ EAψ0 Dψ0 . (4.2.19)
Corollary 4.2.12. Let K/F be ramified. Let (P, ι) be a CM-pair over an OE -algebra R such
that p is nilpotent in R. If (X, ι) satisfies the Eisenstein condition (ECr ), then the conjugate
dual (P ∧ , ι∧ ) also satisfies (ECr ).
Proof. We use the notation of the last remark. The banal ψ are treated as in the unramified case.
We need only to check that the conjugate dual satisfies (ECψ0 ,r ). The orthogonal complement
of EAψ0 Dψ0 in HomR (Dψ0 , R) is Sψ0 EAψ0 HomR (Dψ0 , R), where in the last formula we use the
action via ι∨ . We obtain the result by taking the orthogonal complement of (4.2.19). 
52 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

To end this subsection, we check that the Kottwitz condition (KCr ) is preserved under passage
to the conjugate dual.
Proposition 4.2.13. Let K/F, r be arbitrary. Let (P, ι) be a CM-pair which satisfies (KCr ).
Then the conjugate dual (P ∧ , ι∧ ) satisfies (KCr ).
Proof. We may assume that R is endowed with the structure of an OẼ -algebra. We use the
notation of the proof of Corollary 4.2.8. In particular DP,R = D and DP ∧ ,R = D∧ , and we write
L and L∧ for the Lie algebras of P and P ∧ . We have to show that for each R-algebra S and for
each ψ ∈ Ψ Y
detS (a | L∧
S,ψ ) = ϕS (a)rϕ , for all a ∈ OK ⊗OK t ,ψR S.
ϕ|ψ
To show this, we may replace P by its base change PS . Therefore it is enough to consider the
case S = R. Since Dψ is locally on Spec R a free OK ⊗OK t ,ψR R-module of rank 2, we find
Y
det(a | Dψ ) = ϕR (a)2 , for a ∈ OK ⊗OK t ,ψR R.
ϕ∈Φψ

Since Lψ = Dψ /D1ψ we find

Y Y
det(ā | D1ψ̄ ) = ϕ̄R (ā)(2−rϕ̄ ) = ϕR (a)rϕ .
ϕ̄|ψ̄ ϕ∈Φψ

The perfect pairing (4.2.8) induces a perfect pairing

D1ψ̄ × L∧
ψ −→ R.

Therefore we obtain Y
det(a | L∧ 1
ψ ) = det(ā | Dψ̄ ) = ϕR (a)rϕ .
ϕ∈Φψ
∧
Therefore (KCψ̄,r ) for P implies (KCψ,r ) for P . 
4.3. The pre-contracting functor. Let (P, ι) be a CM-pair of type r over an OE ′ -algebra R
such that p is nilpotent in R. We assume that (P, ι) satisfies (ECr ). In other words (P, ι) ∈
dPr,R , cf. Definition 4.1.1. We will define a functor that associates to (P, ι) a new display P ′ of
the same height with an action
ι′ : OK −→ End P ′ .
In the case where r is banal, the display P ′ will be étale; in the case where r is special, the
restriction of the action ι′ to OF will be strict with respect to ϕ0,R : OF −→ OE −→ R. We will
call this functor the pre-contracting functor.
Let us first restrict our attention to the case where K/F is a field extension. The case
K = F × F will be treated separately because it needs different notations, see p. 56, starting
before eq. (4.3.22). Each ψ : K t −→ Q̄p induces a homomorphism
W (ψ)
ψ̃ : OK t −→ W (OK t ) −−−−→ W (OẼ ). (4.3.1)
For an OE ′ -algebra R we deduce a homomorphism ψ̃R : OK t −→ W (R) that is equivariant with
respect to the Frobenius homomorphisms on both sides. This induces decompositions
Y
OK t ⊗Zp W (R) ∼
= W (R),
ψ∈Ψ
Y (4.3.2)
OK ⊗Zp W (R) ∼= OK ⊗O t ,ψ̃R W (R)
K ψ∈Ψ

which lift the decomposition (4.2.2). Let σ ∈ Gal(K t /Qp ) be the Frobenius automorphism. The
operators F and V act via W (R) on the right hand side of (4.3.2). On the left hand side this
induces maps
F
OK ⊗O t ,ψ̃R W (R) −→←− OK ⊗OK t ,ψ̃R ◦σ W (R),
K
V
cf. (3.4.4). Recall that Y
EAψ (T ) = (T − ϕ(Π)) ∈ OE ′ [T ].
ϕ∈Aψ
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 53

We lift this to a polynomial with coefficients in W (OE ′ ) by taking the Teichmüller representatives
of the roots,
Y
ẼAψ (T ) = (T − [ϕ(Π)]) ∈ W (OE ′ )[T ]. (4.3.3)
ϕ∈Aψ

The image of this polynomial by the homomorphism W (OE ′ ) −→ W (R) is denoted by ẼAψ ,R (T ).
If we reduce with respect to w0 : W (OE ′ ) −→ OE ′ , we obtain the polynomial EAψ (T ). We note
that in the case where R is a κE ′ -algebra, we have
ẼAψ ,R (T ) = T aψ . (4.3.4)
We consider the ring homomorphism
W (OE ′ )[T ] −→ W (R)[T ] −→ OK ⊗O W (R).
K t ,ψ̃R
(4.3.5)
T 7−→ Π ⊗ 1

We denote by ẼAψ ,R (Π ⊗ 1) the image of ẼAψ (T ) under (4.3.5).

Let now (P, ι) ∈ dPr,R , where P = (P, Q, F, Ḟ ). We obtain decompositions of the OK ⊗Zp
W (R)-modules P and Q,
P = ⊕ψ∈Ψ Pψ , Q = ⊕ψ∈Ψ Qψ . (4.3.6)
For x ∈ Pψ , we write
ẼAψ x = ẼAψ ,R (Π ⊗ 1)x. (4.3.7)
On the left hand side we consider Pψ as a W (OE ′ )[T ]-module via (4.3.5).
We give first the recipe for the construction of P ′ for any R ∈ NilpOE′ . Then we will discuss
the case of a perfect field. This special case is then used to prove that P ′ is indeed a display.
We begin with the case where r is banal (and K/F is a field extension). Let (P, ι) as above.
We define
P ′ = ⊕ψ Pψ′ , Q′ = ⊕ψ Q′ψ
as follows: for all ψ we set
Pψ′ = Q′ψ = Pψ . (4.3.8)
By the Eisenstein condition (2.2.13), we have ẼAψ Pψ ⊂ Qψ . Then we may define
Ḟ ′ : Q′ψ −→ Pψσ
′
, Ḟ ′ (x) = Ḟ (ẼAψ x), x ∈ Pψ
′
(4.3.9)
F : Pψ′ ′
−→ Pψσ , F (x) = F (ẼAψ x), x ∈ Pψ .

We define F ′ : P ′ −→ P ′ and Ḟ ′ : Q′ −→ P ′ as the direct sum of the maps above. We have to

prove that P ′ = (P ′ , Q′ , F ′ , Ḟ ′ ) is a display. The only non-trivial property we have to check is
that Ḟ ′ : Pψ −→ Pψσ is an F -linear isomorphism. We postpone the verification, cf. p. 54, below
(4.3.16).
We now define the pre-contracting functor in the case where r is special and K/F unramified.
In this case we have ψ0 6= ψ̄0 . If ψ is banal, i.e., if ψ ∈ / {ψ0 , ψ̄0 }, we keep the definitions (4.3.8)
and (4.3.9). We set Pψ′ 0 = Pψ0 and we define Q′ψ0 as the kernel of the following map,
EAψ ,R
Pψ′ 0 = Pψ0 −→ Pψ0 /Qψ0 −−−−0−→ EAψ0 ,R (Pψ0 /Qψ0 ) ⊂ Pψ0 /Qψ0 . (4.3.10)
It follows from Proposition 4.2.5 that EAψ0 ,R (Pψ0 /Qψ0 ) is locally free of rank 1 and is a direct
summand of Pψ0 /Qψ0 . Therefore
Pψ′ 0 /Q′ψ0 ∼
= EAψ0 ,R (Pψ0 /Qψ0 ) (4.3.11)
is locally free of rank 1 and, as remarked at the end of the proof of Proposition 4.2.6, an element
a ∈ OK ⊗OK t ,ψ0 R acts on (4.3.11) by multiplication with ϕ0,R (a). This makes sense because
ϕ0 : OK −→ OẼ factors through OE ′ ⊂ OẼ . We define
F ′ : Pψ′ 0 −→ Pψ′ 0 σ , F ′ (x) = F (ẼAψ0,R x),
(4.3.12)
Ḟ ′ : Q′ψ0 −→ Pψ′ 0 σ , Ḟ ′ (y) = Ḟ (ẼAψ0,R y).
54 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

The last equation makes sense because, by definition, ẼAψ0 ,R Q′ψ0 ⊂ Qψ0 . The definitions of the
modules Pψ̄′ 0 , Q′ψ̄0 and the restrictions of F ′ and Ḟ ′ to these modules are defined by interchanging
the roles of ψ0 and ψ̄0 . This completes the definition of
F ′ : ⊕ψ Pψ′ −→ ⊕ψ Pψ′ , Ḟ ′ : ⊕ψ Q′ψ −→ ⊕ψ Pψ′ . (4.3.13)
′
Again we postpone the verification that P is a display, cf. below (4.3.17). The tangent space
L′ = P ′ /Q′ is a locally free R-module of rank 2. It has a decomposition
L′ = L′ψ0 ⊕ L′ψ̄0 ,
where an element a ∈ OK ⊗OK t ,ψ0 R acts on the first summand by multiplication with ϕ0,R (a)
and an element a ∈ OK ⊗OK t ,ψ̄0 R acts on the second summand by multiplication with ϕ̄0,R (a).
Next we define the pre-contracting functor in the case where r is special and K/F ramified.
In this case we have ψ0 = ψ¯0 . For banal ψ, we keep the definitions (4.3.8) and (4.3.9). The
R-module EAψ0,R (Pψ0 /Qψ0 ) ⊂ Pψ0 /Qψ0 is a direct summand which is locally free of rank 2. We
set Pψ′ 0 = Pψ0 and we define Q′ψ0 as the kernel of (4.3.10). We define F ′ and Ḟ ′ by (4.3.12). Then
Pψ′ 0 /Q′ψ0 ∼
= EAψ0,R (Pψ0 /Qψ0 ) is locally free and we define as before P ′ = (P ′ , Q′ , F ′ , Ḟ ′ ), with
its OK -action ι′ . It follows from Proposition 4.2.9 that the action of OF on EAψ0,R (Pψ0 /Qψ0 )
via ι coincides with the action of via ϕ0 , i.e., the action via ι′ on P ′ is strict. That P ′ is a display
is proved around (4.3.20).
Now we consider the case where R = k is a perfect field in more detail. We know that Pψ is
a free module of rank 2 over the discrete valuation ring OK ⊗O t ,ψ̃ W (k). Therefore Pψ /ΠPψ is
K
a k-vector space of dimension two. In the perfect field case, we have now also the operator V ,
F : Pψ −→ Pψσ , V : Pψσ −→ Pψ , V (Pψσ ) = Qψ . (4.3.14)
aψ
We will see that in all cases the Eisenstein condition implies that V (Pψσ ) ⊂ Π Pψ . Therefore
we may define operators F ′ and V ′ :
F ′ = Πaψ F : Pψ −→ Pψσ , V ′ = Π−aψ V : Pψσ −→ Pψ . (4.3.15)
The Dieudonné module of the display P ′ in the sense of Proposition 3.1.9 will then be (P, F ′ , V ′ ).
We begin with the case when r is banal. Now EAψ ,k (T ) = T aψ acts on Pψ as multiplication
by Πaψ . By the Eisenstein condition (2.2.13), Πaψ annihilates Pψ /Qψ . This implies Πaψ Pψ ⊂
V (Pψσ ). By the rank condition, the factor Pψ /V (Pψσ ) has length 2aψ as OK ⊗O t ,ψ̃ W (k)-
K
module. Since the same is true for the factor module Pψ /Πaψ Pψ , we obtain
Πaψ Pψ = V (Pψσ ). (4.3.16)
Therefore Ḟ ′ = Ḟ Πaψ = V −1 Πaψ : Pψ −→ Pψσ is bijective. This shows that P ′ is a display. We
set
F ′ = ⊕ψ F Πaψ , V ′ = ⊕ψ Π−aψ V.
Then (P, F ′ , V ′ ) is the Dieudonné module associated to P ′ .
We obtain from (4.3.16) in the ramified case that V 2f Pψ = Π2ef Pψ = pf Pψ for all ψ and in
the unramified case that V 2f Pψ = π ef Pψ . This implies that in both cases P is isoclinic of slope
1/2.
Now we can verify that P ′ is a display for r banal, for an arbitrary OE ′ -algebra R. Let
(P, ι) ∈ dPr,R . We must show that Ḟ : P −→ P is a Frobenius-linear isomorphism. We may
assume that P is a free W (R)-module. Let det Ḟ be the determinant of the matrix of Ḟ with
respect to any given basis of the W (R)-module P . We must verify that det Ḟ is a unit in W (R).
We have shown that, for each homomorphism R −→ k to a perfect field k, the image of det Ḟ by
W (R) −→ W (k) is a unit in W (k). In particular w0 (det Ḟ ) ∈ R has a nonzero image under any
homomorphism R −→ k. But then w0 (det Ḟ ) is a unit in R, and this implies that det Ḟ ∈ W (R)
is a unit. This finishes in the banal case the proof that P ′ is a display.
Next we consider the case when r is special and K/F unramified. By our conventions, Π = π
is the prime element of F . Let R = k be a perfect field and let P = (P, F, V ), regarded as
a Dieudonné module. If ψ ∈ Ψ is banal, we find as above that V Pψσ = π aψ Pψ . Now let
ψ ∈ {ψ0 , ψ̄0 }. Since π aψ +1 annihilates Lieψ X by the Eisenstein condition (2.2.18), we obtain
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 55

π aψ +1 Pψ ⊂ V Pψσ . We note that Pψ is a OK ⊗O t ,ψ̃ W (k)-module of rank 2. Therefore the

K
factor module of the last inclusion is, by the rank condition, a OK ⊗O t ,ψ̃ W (k)-module of length
K
1 and is therefore annihilated by π. This implies
π aψ +1 Pψ ⊂ V Pψσ ⊂ π aψ Pψ .
In particular
∼
Pψ /π aψ Pψ −→ Lieψ X/π aψ Lieψ X (4.3.17)
is an isomorphism, as claimed in the beginning of the proof of Proposition 4.2.5. By definition
(4.3.10) we have Q′ψ = π −aψ V Pψσ . The map Ḟ ′ = π aψ Ḟ : Q′ψ −→ Pψσ is therefore surjective.
Since we know this fact also for banal ψ we conclude that (P, Q′ , F ′ , Ḟ ′ ) is a display. The
associated Dieudonné module is (P, F ′ , V ′ ), where
Fψ′ = π aψ Fψ : Pψ −→ Pψσ ,
′
(4.3.18)
Vψσ = π −aψ Vψσ : Pψσ −→ Pψ .
Now we return to an arbitrary O-algebra R such that p is nilpotent in R. We note that the
definition of (P ′ , Q′ ) commutes with arbitrary base change because Q′ψ0 /I(R)Pψ′ 0 is defined as
the kernel of an epimorphism of projective R-modules,
Pψ0 /I(R)Pψ0 −→ EAψ0 (Pψ0 /Qψ0 ).
We choose a normal decomposition of (P ′ , Q′ ),
P ′ = T ′ ⊕ L′ ,
together with the Frobenius-linear endomorphism Φ′ : P ′ −→ P ′ of the W (R)-module P ′ such
that the restriction of Φ′ to T ′ is F ′ and the restriction to L′ is Ḟ ′ . We have to show that the
determinant of Φ in a locally chosen basis is a unit. Since we know that this is true after any
base change R −→ k with k a perfect field, this follows as in the banal case.
We can determine the possible slopes of P when r is special and K/F unramified. Let
P = (P, F, V ) over the perfect field k. By (4.3.15) we have (V ′ )2f = π −(ef −1) V f . Let
PQ = ⊕λ N (λ)
be the decomposition into isoclinic components. Fix λ = r/s. Then we find a W (k)-lattice
Λ ⊂ N (λ) such that V s Λ = pr Λ. From V 2f s Λ = p2f r Λ we obtain
(π ef −1 (V ′ )2f )s Λ = p2rf Λ, i.e., (V ′ )2f s Λ = π −ef s π s p2rf Λ.
We write the right hand side as p−sf ps/e p2rf Λ. This shows that N (λ) ⊂ PQ′ is an isoclinic
rational Dieudonné submodule of slope
−sf + (s/e) + 2f r 1 1
=− + + λ.
2f s 2 2d
Let us apply the Ahsendorf functor to P ′ . We obtain a WOF (k)-Dieudonné module (Pc , Fc , Vc )
of height 4 and dimension 2. The slopes of Pc are by Proposition 3.3.17
1 1
d(λ − ) + . (4.3.19)
2 2
The action of OK ⊗OF WOF (k) ∼ = WOF (k) × WOF (k) on Pc defines a decomposition Pc =
Pc,0 ⊕ Pc,1 such that Vc (Pc,0 ) ⊂ Pc,1 and Vc (Pc,0 ) ⊂ Pc,1 . The WOF (k)-module Pc,0 with
the semi-linear operator Vc2 is of height 2 and dimension 2. Therefore the possible slopes of
(Pc,0 , Vc2 ) are with multiplicities (1, 1) or (0, 2). We conclude that the slopes of (Pc , Vc ) are with
multiplicities (1/2, 1/2, 1/2, 1/2) or (0, 0, 1, 1). From (4.3.19) we find that in the first case all
slopes λ of P are 1/2, while in the second case we obtain the two slopes λ = 1/2 − 1/2d and
λ = 1/2 + 1/2d.
Now we consider the case where r is special and K/F is ramified. As in the last case, it is
enough to verify that P ′ is a display when R = k is a perfect field. Recall that aψ = e for ψ
banal and that aψ0 = e − 1. As above we find V Pψσ = Πe Pψ for ψ banal. By the Eisenstein
56 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

condition (2.2.16), Lψ0 is annihilated by Πe+1 and the k-vector space Πe−1 Lψ0 has dimension
at most 2. We consider the following filtration by subvector spaces,
Lψ0 ⊃ ΠLψ0 ⊃ Π2 Lψ0 ⊃ . . . ⊃ Πe Lψ0 ⊃ Πe+1 Lψ0 = 0.
We have dimk Πm Lψ0 /Πm+1 Lψ0 ≤ 2 for all m ≥ 0 since Lψ0 is a quotient of Pψ0 , which is a free
OK ⊗O t ,ψ̃0 W (k)-module of rank 2. Therefore we find
K

dimk Lψ0 = dimk Lψ0 /Πe−1 Lψ0 + dimk Πe−1 Lψ0 ≤ 2(e − 1) + 2 = 2e = dimk Lψ0 .
We must have equality
dimk Lψ0 /Πe−1 Lψ0 = 2(e − 1), dimk Πe−1 Lψ0 = 2.
The first equation shows that the natural map
Pψ0 /Πe−1 Pψ0 −→ Lψ0 /Πe−1 Lψ0 (4.3.20)
is an isomorphism of vector spaces, as asserted in the beginning of the proof of Proposition 4.2.5.
Finally we have by definition Q′ψ0 = Π−e+1 Qψ0 = Π−e+1 V Pψ0 σ . Therefore

Ḟ ′ = Πe−1 V −1 : Q′ψ0 −→ Pψ0 σ

is bijective. We conclude that (P ′ , Q′ , F ′ , Ḟ ′ ) is a display. The associated Dieudonné module is

(P, F ′ , V ′ ), where
V ′ = Π−e V : Pψσ −→ Pψ , ψ 6= ψ0
′ −e+1
(4.3.21)
V =Π V : Pψ0 σ −→ Pψ0 .
As in the unramified case we conclude for an arbitrary R ∈ NilpOE′ that our definitions (4.3.8),
(4.3.9), (4.3.12) give a display P ′ = (P ′ , Q′ , F ′ , Ḟ ′ ).
In the case of a perfect field k, the slopes of P ′ are computed in the same way as in the
unramified case. We have the equation (V ′ )f = Π−(ef −1) V f . Let N (λ) ⊂ PQ be an isoclinic
component. We find a lattice Λ ⊂ N (λ) such that V s Λ = pr Λ. We obtain that
(V ′ )sf Λ = Π−ef s Πs prf Λ.
Since Π2e and p differ by a unit, this implies that N (λ) ⊂ PQ′ is isoclinic of slope
−(f s/2) + (s/2e) + rf
= −1/2 + 1/2d + λ = (λ − 1/2) + 1/2d.
sf
If we apply the Ahsendorf functor, we obtain an WOF (k)-Dieudonné module (Pc , Fc , Vc ) with
slopes d(λ − 1/2) + 1/2. If we consider the OK ⊗OF WOF (k)-module Pc with the semi-linear
operator Vc the possible slopes with multiplicity are (1, 1) or (0, 2) because (Pc , Vc ) is of height 2
and dimension 2. If we regard (Pc , Vc ) over WOF (k), the heights are multiplied with 2 and then
the possible heights are (1/2, 1/2, 1/2, 1/2) or (0, 0, 1, 1). As in the unramified case we conclude
that P is either isoclinic of slope 1/2 or has exactly two slopes 1/2 − 1/2d and 1/2 + 1/2d.
Finally we consider the case where r is banal and K = F ×F . We set Θ = HomQp -Alg (F t , Q̄p ).
In this case σ will denote the Frobenius automorphism in Gal(F t /F ). If we compose θ ∈ Θ with
the first, resp. second, projection K t = F t × F t −→ F t we obtain θ1 , θ2 ∈ Ψ. Via the first, resp.
the second, projection, we obtain isomorphisms
(OF × OF ) ⊗(OF t ×OF t ),θi OE ′ ∼
= OF ⊗OF t ,θ OE ′ , i = 1, 2.
This leads to the decomposition
Y Y
Y

OK ⊗ OE ′ = OK ⊗OK t ,ψ OE ′ = OF ⊗OF t ,θ OE ′ × OF ⊗OF t ,θ OE ′ . (4.3.22)
ψ∈Ψ θ∈Θ θ∈Θ

Assume that ψ ∈ Ψ factors through θ ∈ Θ. We define ψ̃ as the composite

proj. W (θ)
OK t = OF t × OF t −−−→ OF t −→ W (OF t ) −−−→ W (OE ′ ). (4.3.23)
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 57

The first map is the projection to the first or second factor according to ψ. We denote by θ̃ the
composite of the last two arrows in (4.3.23). We obtain the decomposition
Y
OK ⊗Z W (OE ′ ) = OK ⊗O t ,ψ̃ W (OE ′ )
K
ψ∈Ψ
Y  Y  (4.3.24)
= OF ⊗O t ,θ̃
W (OE ′) × OF ⊗O t ,θ̃
W (OE ′) .
F F
θ∈Θ θ∈Θ

On the right hand side, the first set of factors correspond to those ψ which factor over the first
projection and the second set of factors correspond to those ψ which factor over the second
projection.
We consider a CM-pair (P, ι) over R ∈ NilpOE′ which satisfies the Eisenstein condition. By
(4.3.22) we obtain a decomposition
M M
P = P1 × P2 = ( P1,θ ) ⊕ ( P2,θ ). (4.3.25)
θ∈Θ θ∈Θ

This decomposition corresponds to the decomposition into displays P = P1 ⊕ P2 induced by the

OF × OF -action on P. By the definition of a CM-pair (at the beginning of subsection 2.3), the
displays P1 and P2 have both height 2d.
The maps F and Ḟ of the display P induce maps
F : Pi,θ −→ Pi,θσ , Ḟ : Qi,θ −→ Pi,θσ . (4.3.26)
The polynomial ẼAψ ∈ W (OE ′ )[T ] is defined as before, cf. (4.3.3). For i = 1, 2 we define the
displays Pi′ = (Pi′ , Q′i , Fi′ , Ḟi′ ) as follows
Pi′ = Q′i = Pi , Ḟi′ (x) = Ḟ (ẼAθi x), Fi′ (x) = F (ẼAθi x), x ∈ Piθ .

Here, by the convention (4.3.7), ẼAθi acts as the multiplication by ẼAθi ,R (π ⊗ 1) ∈ OF ⊗O t ,θ˜i
F
W (R). We set P ′ = P1′ ⊕P2′ . As in the unramified banal case, the verification that P ′ is a display
reduces to the case of a perfect field. However, when R is a κE ′ -algebra, then ẼAθi ,R (π ⊗ 1) =
π aθi ⊗ 1. If R = k is a perfect field, we consider the Dieudonné module (Pi , Fi , Vi ) of Pi . We
have
Vi (Pi,θσ ) = π aθi Pi,θ . (4.3.27)
We define
Fi′ = π aθi Fi : Pi,θ −→ Pi,θσ , Vi′ = π −aθi Vi : Pi,θσ −→ Pi,θ .
Then (Pi , Fi′ , Vi′ ) is the Dieudonné module of Pi′ . Finally we determine the slopes of P. If we
iterate (4.3.27) we find
P
V f Pi = π θ aθi Pi . (4.3.28)
P
We set gi = θ aθi . Then g1 + g2 = ef because aθ1 + aθ2 = e. We see that Pi is isoclinic of
slope λi = gi /d and that λ1 + λ2 = 1.
We summarize the properties of our constructions.
Definition 4.3.1. Let R ∈ NilpOE′ . We define categories dP′r,R and P′r,R as follows.
(1) If r is banal, then dP′r,R is the category of pairs (P ′ , ι′ ), where P ′ is an étale display (i.e.,
P ′ = Q′ ) of height 4d and where ι′ is an OK -action. In the split case OK = OF × OF , we
require in addition that in the induced decomposition P ′ = P1′ ⊕ P2′ both factors have height
2d.
(2) If r is special and K/F is unramified, then the category dP′r,R is the category of pairs (P ′ , ι′ ),
where P ′ is a display of height 4d and dimension 2 with an action ι′ : OK −→ End P ′ such
ϕ0
that the action of ι′ restricted to OF is strict with respect to ϕ0,R : OF −→ OE ′ −→ R and
′ ′ ′
such that Lie P = P /Q is locally on Spec R a free OK ⊗OF ,ϕ0,R R-module of rank 1.
(3) If r is special and K/F is ramified, then the category dP′r,R is the category of pairs (P ′ , ι′ ),
where P ′ is a display of height 4d and dimension 2 with an action ι′ : OK −→ End P ′ such
ϕ0
that the action of ι′ restricted to OF is strict with respect to ϕ0,R : OF −→ OE ′ −→ R.
58 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

The category P′r,R is the category of formal p-divisible groups X ′ with an OK -action ι′ such
that the associated display (P ′ , ι′ ) is an object of dP′r,R .
Let r be special. We call (P, ι) supersingular if (P ′ , ι′ ) satisfies the nilpotence condition. We
denote the full subcategory of supersingular objects of dPr,R by dPss r,R .

Theorem 4.3.2. Let R ∈ NilpOE′ be such that the ideal of nilpotent elements in R is nilpotent.
The construction above defines the pre-contracting functor5
C′r,R : dPr,R −→ dP′r,R
which commutes with arbitrary base change with respect to R. Furthermore,
(i) if r is banal, the functor C′r,R is an equivalence of categories.
(ii) if r is special and the ring R is reduced, the functor C′r,R is an equivalence of categories.
(iii) if r is special and R is arbitrary, C′r,R induces an equivalence of categories
′ nilp
C′r,R : dPss
r,R −→ dPr,R ,
where the right hand side is the full subcategory of nilpotent displays.
Let r be special and K/F be ramified. Let (P, ι) ∈ dPr,R and let (P ′ , ι′ ) be its image by C′r,R .
Then (P, ι) satisfies (KCr ) if and only if
TrR (ι′ (Π) | P ′ /Q′ ) = 0.
Before proving this, we state a Corollary which we already proved in the construction of C′r,R
above.
Corollary 4.3.3. Let k ∈ NilpOE′ be a perfect field. Let P ∈ dPr,k and let P ′ be its image by
the functor C′r,R .
(1) Let r be banal and K/F a field extension. Then the display P is isoclinic of slope 1/2 and
P ′ is étale.
(2) Let K/F be split (and then r is banal). Then P decomposes into P = P1 ⊕ P2 , where P1
is isoclinic of slope λ and P2 is isoclinic of slope 1 − λ. The number λ depends only on r. The
display P ′ is étale.
(3) Let r be special. Then P is either isoclinic of slope 1/2 (supersingular case) or it has the
two slopes 1/2 − 1/2d and 1/2 + 1/2d with the same multiplicity. In the first case P ′ is isoclinic
of slope 1/2d. In the second case it has the two slopes 0, 1/d with the same multiplicity.

Proof. We still have to prove the claimed equivalences of categories. We begin with the case
where R is reduced.
Let us consider first the case where r is special and K/F unramified. It is enough to in-
vert the construction of the functor C′r,R . For any ψ, Pψ /I(R)Pψ is locally on Spec R a free
(OK /pOK ) ⊗κK ,ψ R-module of rank 2, cf. Lemma 3.1.15. Let P ′ = (P ′ , Q′ , F ′ , Ḟ ′ ) be an object
of dP′r,R . We define as follows an object P = (P, Q, F, Ḟ ) of dPr,R such that P ′ is the image of
P by the functor C′r,R . We set Pψ = Pψ′ for ψ ∈ Ψ, and for ψ ∈ / {ψ0 , ψ̄0 } we set
Qψ = π aψ Pψ + I(R)Pψ . (4.3.29)
Since Pψ = Q′ψ , we have Ḟ ′ (I(R)Pψ ) ⊂ W (R)F ′ Pψ ⊂ pPψσ . By (4.3.29) we find Ḟ ′ (Qψ ) ⊂
π aψ Pψσ . Since p is not a zero divisor in W (R), the element π ∈ OK acts injectively on Pψ0 .
Therefore we may define
Ḟ = π −aψ Ḟ ′ : Qψ −→ Pψσ , F = π −aψ F ′ : Qψ −→ Pψσ .
If ψ ∈ {ψ0 , ψ̄0 } we consider the split homomorphism of R-modules
π aψ : Pψ /I(R)Pψ −→ Pψ /I(R)Pψ . (4.3.30)
It is split because Pψ /I(R)Pψ is a free (OK /pOK ) ⊗κK ,ψ R-module. We set
Qψ = π aψ Q′ψ + I(R)Pψ .
5Later we will also have a contracting functor C
r,R , which explains our notation.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 59

If we apply Ḟ ′ to the last equation we obtain that Ḟ ′ (Qψ ) ⊂ π aψ Pψσ . Indeed, because the action
of OF on P ′ is strict π annihilates Pψ /Q′ψ . We conclude that F ′ (πPψ ) ⊂ F ′ (Q′ψ ) = pḞ ′ (Qψ ) ⊂
pPψσ and therefore F ′ (Pψ ) ⊂ π e−1 Pψ ⊂ π aψ Pψ . This justifies the following definition:
Ḟ := π −aψ Ḟ ′ : Qψ −→ Pψσ , F := π −aψ F ′ : Pψ −→ Pψσ .
It is obvious that we obtain a display P = (P, Q, F, Ḟ ). We need to verify that the condition
(ECr ) is satisfied. We check the conditions (2) and (3) of Proposition 4.2.7. By definition of
dP′r,R , the R-module Pψ′ /Q′ψ is annihilated by π. The kernel of (4.3.30) is π e−aψ Pψ and therefore
contained in Q′ψ /I(R)Pψ . The image of the last module by (4.3.30) is therefore a direct summand
of Pψ /I(R)Pψ . This image is Qψ /I(R)Pψ . Therefore condition (2) holds. Moreover, we obtain
an isomorphism
∼
Pψ /Q′ψ −→ π aψ Pψ + I(R)Pψ /Qψ .
In particular, the last module is locally free of rank 1 and the action of π on this module coincides
with multiplication by ϕ0 (π) if ψ = ψ0 , resp., by ϕ̄0 (π) if ψ = ψ̄0 . Hence condition (3) holds.
In the split case the same arguments hold but we need only the easy part because ψ0 and ψ̄0
don’t exist.
Next we consider the case where r is special and K/F ramified. Again we reverse the con-
struction of the functor C′r,R . Let (P ′ , Q′ , F ′ , Ḟ ′ ) be an object of dP′r,R . We associate to it as
follows an object (P, Q, F, Ḟ ) ∈ dPr,R . We set Pψ = Pψ′ for all ψ ∈ Ψ. Assume that ψ 6= ψ0 .
We have Q′ψ = Pψ′ because the action of OF t is strict. We set
Qψ = Πe Pψ′ + I(R)Pψ′ .
It follows from Lemma 3.1.15 that Pψ /I(R)Pψ is locally on Spec R a free (OK /pOK ) ⊗κK ,ψ R-
module. Therefore Pψ /Qψ is a locally free R-module. From F ′ Pψ = pḞ ′ Pψ , we find that
Ḟ ′ Qψ ⊂ Πe Pψσ
′
. Since R is reduced, the map Πe : Pψσ −→ Πe Pψσ is bijective. Therefore we
may define
Ḟ := Π−e Ḟ ′ : Qψ −→ Pψσ .
It is clear that this map is a Frobenius-linear epimorphism. Next, we set
Pψ0 = Pψ′ 0 , Qψ0 = Πe−1 Q′ψ0 + I(R)Pψ′ 0 .
Since the action of OF on P ′ is strict, we find
Πe+1 Pψ′ 0 ⊂ Π2 Pψ′ 0 ⊂ Q′ψ0 . (4.3.31)
We consider the split homomorphism of R-modules,
Πe−1 : Pψ′ 0 /I(R)Pψ′ 0 −→ Pψ′ 0 /I(R)Pψ′ 0 . (4.3.32)
The kernel of this map is the image of Πe+1 . Therefore the kernel is contained in Q′ψ0 /I(R)Pψ′ 0 .
This implies that the image of Q′ψ0 /I(R)Pψ′ 0 under (4.3.32) is a direct summand of Pψ′ 0 /I(R)Pψ′ 0 .
Hence the cokernel Pψ0 /Qψ0 is a locally free R-module. We apply F ′ to (4.3.31) and obtain
Π2 F ′ Pψ′ 0 ⊂ F ′ Q′ψ0 = pḞ ′ Q′ψ0 ⊂ pPψ′ 0 σ .
Using this, we get


Ḟ ′ Qψ0 = Ḟ ′ Πe−1 Q′ψ0 + I(R)Pψ′ 0 ⊂ Πe−1 Pψ′ 0 σ + F ′ Pψ′ 0 ⊂ Πe−1 Pψ′ 0 σ .
It follows that the following definitions of maps Qψ0 −→ Pψ0 , resp. Pψ0 −→ Pψ0 , make sense:
Ḟ = (1/Πe−1 )Ḟ ′ , F = (1/Πe−1 )F ′ .
Therefore we have defined P = (P, Q, F, Ḟ ). It is clear that we obtain a display. We have to
verify the condition (ECr ). Only (ECψ0 ,r ) is not completely obvious. We prove the conditions
of Proposition 4.2.9. By the R-module homomorphism (4.3.32), Pψ′ 0 /I(R)Pψ′ 0 is mapped to
the direct summand (Πe−1 Pψ′ 0 + I(R)Pψ′ 0 )/I(R)Pψ′ 0 , and Q′ψ0 /I(R)Pψ′ 0 is mapped to the direct
summand Qψ0 /I(R)Pψ′ 0 . We obtain an isomorphism
∼
Pψ′ 0 /Q′ψ0 −→ (Πe−1 Pψ0 + I(R)Pψ0 )/Qψ0 .
60 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Therefore by the strictness of the OF -action, the right hand side is a locally free R-module
of rank 2 and ι(π) acts on the right hand side as ϕ0 (π). These are exactly the conditions of
Proposition 4.2.9. The rank condition is now obvious for (P, Q, F, Ḟ ).
Finally, in the case where r is banal, including the split case (and R is reduced), we can
reverse the functor C′r,R using the arguments for banal ψ given above.
Now we consider assertion (iii) of Theorem 4.3.2 when R is not reduced. It follows from
Corollary 4.3.3 (3) that P is isoclinic of slope 1/2 because P ′ is nilpotent. Therefore we may apply
Grothendieck-Messing for displays Corollary 3.1.14. We consider a surjective homomorphism
S −→ R of OE ′ -algebras and assume that the kernel a is endowed with a divided power structure.
We define the category dPr,S/R as the full subcategory of the category of pairs (P1 , ι1 ) where
P1 is a W(S/R)-display, cf. Example 3.1.3, and where
ι1 : OK −→ End P1
is an action such that the base change (P, ι) of such a pair by the morphism of frames W(S/R) −→
W(R) lies in the category dPr,R . We also say that (P1 , ι1 ) is a lift of (P, ι) to a relative display.
By Theorem 3.1.12, the lift (P1 , ι1 ) is uniquely determined by (P, ι) if P satisfies the nilpotence
condition.
In the same way we define the category dP′r,S/R . Then the functor C′r,R of Theorem 4.3.2
extends to a functor
C′r,S/R : dPr,S/R −→ dP′r,S/R . (4.3.33)
Indeed, the definition of C′r,S/R is essentially the same as that of C′r,R . We indicate it in the
case where K/F is unramified or split. By the OK -action, we have for the relative display P1 a
decomposition,
P1 = ⊕ψ P1,ψ , Q1 = ⊕ψ Q1,ψ .
′
We are going to define a W(S/R)-display P1′ = (P1′ , Q′1 , F1′ , Ḟ1′ ). We set P1,ψ = P1,ψ for all
ψ ∈ Ψ. Since P1 is a lifting of P, we have a natural isomorphism
P1,ψ /Q1,ψ = Pψ /Qψ . (4.3.34)
If ψ ∈ ′
/ {ψ0 , ψ̄0 }, we set Q′1,ψ = P1,ψ . By the condition (ECr ) for P we conclude that
ẼAψ P1,ψ ⊂ Q1,ψ . Therefore we can define

F1′ : P1,ψ
′
−→ P1,ψσ , F1′ (x) = F1 (ẼAψ x), x ∈ P1,ψ
′
(4.3.35)
Ḟ1′ : Q′1,ψ −→ P1,ψσ
′
, Ḟ1′ (x) = Ḟ1 (ẼAψ x), x ∈ Q′1,ψ .

In the split case this decribes P1′ already completely. Now we consider the case ψ ∈ {ψ0 , ψ̄0 }.
Then we define Q′1,ψ as the kernel of the map
EAψ
P1,ψ −→ Pψ /Qψ −−−→ Pψ /Qψ .
This implies ẼAψ Q′1,ψ ⊂ Q1,ψ . Therefore we can define

F1′ : P1,ψ
′ ′
−→ P1,ψσ , F1′ (x) = F1 (ẼAψ x), x ∈ P1,ψ
′
,
(4.3.36)
Ḟ1′ : Q′1,ψ −→ P1,ψ
′
0σ
, Ḟ1′ (y) = Ḟ1 (ẼAψ y), y ∈ Q′1,ψ .

We then define P1′ = (P1′ , Q′1 , F1′ , Ḟ1′ ), where P1′ = ⊕P1,ψ
′
and Q′1 = ⊕Q′1,ψ . In the ramified case
the same definition holds with slight modifications.
The functor C′r,S/R defines a natural isomorphism

DP (S) ∼
= DP ′ (S). (4.3.37)
This relates deformations of P and deformations of its image P ′ under Cr,R since P and P ′ are
nilpotent.
′ ′
Let (P, ι) ∈ dPss
r,R . It has a unique lift P1 ∈ dPr,S/R . The image P1 by the functor Cr,S/R is
′ ′
the unique lift of P to an object of dPr,S/R , cf. Theorem 3.1.12.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 61

Let us fix P. Let M be the set of all isomorphism classes of deformations of (P, ι) to an object
in dPr,S . Let M′ be the set of isomorphism classes of deformations of (P ′ , ι′ ) to an object of
dP′r,S/R . We claim that the functor C′r,S defines a bijection,

C′r,S : M −→ M′ . (4.3.38)

We indicate this when K/F is unramified. Let Q̄ ⊂ DP (R) = P/I(R)P be the image of Q, i.e.,
the Hodge filtration. The set M is identified with the set of liftings of Q̄ to a direct summand
Q̄1 ⊂ DP (S) = P1 /I(S)P1 which is a OK ⊗Zp S-submodule and such that the factor module
satisfies the Eisenstein condition. The OK -action gives a decomposition Q̄1 = ⊕Q̄1,ψ . For ψ
banal, we must have by Proposition 4.2.7 that

ẼAψ DP (S)ψ = Q̄1,ψ . (4.3.39)

We note that the left hand side is a direct summand of DP (S)ψ as an S-module. This follows
from the fact that P1,ψ /I(S)P1,ψ is a free module over S[T ]/Eψ S[T ]. Therefore, there is exactly
one possibility to lift the ψ-component of the Hodge filtration. We consider now liftings of Q̄ψ
when ψ is not banal. In this case the Eisenstein condition implies that

Sψ ẼAψ DP (S)ψ ⊂ Q̄1,ψ ⊂ ẼAψ DP (S)ψ .

By the freeness of DP (S)ψ just mentioned, the multiplication by ẼAψ gives an isomorphism

ẼAψ : DP (S)ψ /Sψ DP (S)ψ ∼

= ẼAψ DP (S)ψ /Sψ ẼAψ DP (S)ψ .

This shows that it is the same thing to lift Q̄ψ to a direct summand Q̄1,ψ ⊂ DP (S)ψ such
that the Eisenstein condition is satisfied or to lift Ẽ−1 ′
Aψ Q̄ψ to a direct summand Q̄1,ψ such that
DP (S)ψ /Q̄′1,ψ is annihilated by Sψ . The last condition means that the action of OF is strict
with respect to ϕ0 , resp., ϕ̄0 . In other words,

Q̄′1 = Q̄1,ψ0 ⊕ Q̄1,ψ̄0 ⊕ (⊕ψ6=ψ0 ψ̄0 DP (S)ψ )

is a lift of the Hodge filtration Q̄′ ⊂ DP ′ (R) = P/I(R)P to a Hodge filtration Q̄′1 ⊂ P1 /I(S)P1
such that the action of OF is strict, i.e., the Hodge filtration Q̄′1 defines a point of M′ . This shows
that (4.3.38) is bijective because the functor C′r,S maps the Hodge filtration Q̄1,ψ to E−1 Aψ Q̄1,ψ
when ψ is special by the definition (4.3.10). We leave the ramified case to the reader.
Finally we prove assertion (i) of Theorem 4.3.2, i.e., we assume that r is banal. We begin
with the case where K/F is a field extension. Then P is by Corollary 4.3.3 (1) of slope 1/2. By
(4.3.39) there is a unique way to lift the Hodge filtration and therefore the Grothendieck-Messing
criterion implies that there is a unique way to lift P to an object P1 ∈ dPr,S/R . On the other
hand P ′ is étale. Therefore it lifts obviously uniquely, and (i) follows. In the case where K/F is
split the same argument applies if P is local. If not, we consider the decomposition P = Pα ⊕ Pβ
induced by OK = OF × OF . By Corollary 4.3.3 (2), in each geometric point of Spec R one of the
factors of this decomposition is isoclinic of slope 0 and the other is isoclinic of slope 1. That Pα
is étale means that the locally free module Pα /Qα is zero. This is true on an open and closed
subset of Spec R. Therefore we may assume without loss of generality that Pα is étale. Then
Pα has a unique lift and Pβ has a unique lift by Grothendieck-Messing. Since P ′ is étale it has
also a unique lift. This completes the proof in the split case. 

Corollary 4.3.4. Let R ∈ NilpOE′ be such that the ideal of nilpotent elements in R is nilpotent.
′
We denote by Pss ss form
r,R the full subcategory of objects of Pr,R whose displays lie in dPr,R . Let Pr,R
′ ′
be the full subcategory of Pr,R whose objects are formal p-divisible groups. Then Cr,R induces
an equivalence of categories
′
C′r,R : Pss form
r,R −→ Pr,R .


62 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

4.4. The contracting functor in the case of a special CM-type. In this subsection, r will
denote a special CM-type. In this case, we will compose the functor C′r,R with the Ahsendorf
functor.
Definition 4.4.1. Let r be special. Let R ∈ NilpOF . We denote by dRR 6 the category of
WOF (R)-displays Pc endowed with a homomorphism of OF -algebras
ιc : OK −→ End Pc ,
such that Pc is of height 4 and dimension 2. In the case where K/F is unramified, we require
moreover that Lie Pc is locally on Spec R a free OK ⊗OF R-module of rank 1.
We note that in the ramified case, the OK ⊗OF R-module Lie Pc is in general not locally free
on Spec R.
Definition 4.4.2. Let r be special. Let R be a OE ′ -algebra. We regard R as a OF -algebra via
ϕ0
ϕ0,R : OF −→ OE −→ R. The contracting functor
Cr,R : dPr,R −→ dRR
is the composition of C′r,R with the Ahsendorf functor AOF /Zp ,R .
Theorem 4.4.3. Let r be special. Let R ∈ NilpOE′ be such that the ideal of nilpotent elements
of R is nilpotent. Then the functor Cr,R induces an equivalence of categories
nilp
Cr,R : dPss
r,R −→ dRR .

Here dRnilp
R denotes the full subcategory of nilpotent displays in dRR .
Proof. This follows from Proposition 4.3.2 and Theorem 3.3.2. 
Remark 4.4.4. Let R = k be a perfect field with an OE ′ -algebra structure. Then the construc-
tion of the functor Cr,k simplifies.
We begin with the unramified case. Let P = (P, F, V ) ∈ dPr,k viewed as a Dieudonné module.
The display P ′ = (P, F ′ , V ′ ) is described after (4.3.17). Applying the Ahsendorf functor to it,
we obtain the image Pc = (Pc , Fc , Vc ) of P by the functor Cr,k . The P0 of (3.3.33) is in our case
Pc = Pψ0 ⊕ Pψ̄0 and Vc = (V ′ )f is the Va of (3.3.33). We know that the restriction of V ′ to Pψσ
is
V ′ = π −aψ V : Pψσ −→ Pψ .
We conclude that (V ′ )f : Pψ0 −→ Pψ̄0 is equal to π −gψ̄0 V f where
gψ̄0 = aψ̄0 + aψ̄0 σ + . . . + aψ̄0 σf −1
(4.4.1)
= aψ0 σ−f + aψ0 σ−(f −1) + . . . + aψ0 σ−1 .
In the same way (V ′ )f : Pψ̄0 −→ Pψ0 is equal to π −gψ0 V f where
gψ0 = aψ0 + aψ0 σ + . . . + aψ0 σf −1 . (4.4.2)
From (2.2.17) we obtain gψ0 + gψ̄0 = ef − 1. In summary, Pc = Pψ0 ⊕ Pψ̄0 as a WOF (k) =
OF ⊗O t ,ψ̃0 W (k)-module, and Vc is given by the matrix
F
 
0 π −gψ0 V f
. (4.4.3)
π −gψ̄0 V f 0
Finally Fc is determined by the equation Fc Vc = π. For instance, Fc : Pψ̄0 −→ Pψ0 is equal to
(π gψ̄0 +1 /pf )F f . We obtain a Dieudonné module (Pc , Fc , Vc ) with respect to the perfect frame
WOF (k) = (OF ⊗O W (k), πOF ⊗O W (k), k, F f , F f π −1 ). (4.4.4)
F t ,ψ̃0 F t ,ψ̃0

In the ramified case we have Pc = Pψ0 as a module over WOF (k) = OF ⊗O t ,ψ̃0 W (k). If we
F
apply the Ahsendorf functor to P ′ , we obtain by (4.3.21)
Vc = Π−ef +1 V f : Pc −→ Pc . (4.4.5)
ef +1 f f
Fc is determined by the equation Fc Vc = π, i.e., Fc = −(Π /p )F . We obtain a Dieudonné
module (Pc , Fc , Vc ) for the frame (4.4.4)
6The symbol R is to remind us that this is a category of relative displays.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 63

We next add polarizations to the picture. We set t(a) = TrF/Qp ϑ−1 a where ϑ is the different
of F/Qp .
Proposition 4.4.5. Let r be special. Let R ∈ NilpOE′ . Let (P1 , ι1 ) and (P2 , ι2 ) be objects of
dPr,R . Let (P1′ , ι′1 ) and (P2′ , ι′2 ) be their images by the functor C′r,R . Assume given a bilinear
form of displays
β : P1 × P2 −→ Pm,R ,
where Pm is the multiplicative display of W(R). Assume that β is anti-linear for the OK -actions
ι1 , resp. ι2 , i.e,
β(ι1 (a)x1 , x2 ) = β(x1 , ι2 (ā)x2 ), x1 ∈ P1 , x2 ∈ P2 , a ∈ OK . (4.4.6)
Define
β̃ : P1 × P2 −→ OF ⊗Zp W (R)
by the equation
t(ξ β̃(x1 , x2 )) = β(ξx1 , x2 ), x1 ∈ P1 , x2 ∈ P2 , ξ ∈ OF ⊗Zp W (R).
Then β̃ is a OF -bilinear form of displays,
β̃ : P1′ × P2′ −→ LR ,
where LR is the Lubin-Tate display associated to the local field F and the algebra structure
ϕ0 : OF −→ OE ′ −→ R, cf. Definition 3.4.3. Furthermore, β̃ is anti-linear for the OK -actions
ι1 , resp. ι2 .
Proof. To avoid a conflict with the present notations, we adapt some of the notation of section
3.4 to our situation. What was K in section 3.4 is now F . We set Lt = ψ0 (F t ) ⊂ E ′ . We write
the polynomials of that section as follows:
Y
ẼF,ψ (Z) = (Z − [χ(π)]) ∈ W (OLt )[Z].
χ:F −→Ẽ, χ|F t =ψ

We stress that here ψ denotes an embedding of F t into E ′ , not as elsewhere in this section an
embedding of K t into E ′ . For ψ = ψ0 , we consider the decomposition ẼF,ψ0 (Z) = (Z − [ϕ0 (π)]) ·
ẼF,0 (Z) in W (ϕ0 (F ))[Z]. In particular all of these polynomials lie in W (OE ′ )[Z].
Let M be an OF ⊗OF t ,ψ W (R)-module. Then we write by our convention
ẼF,ψ m = ẼF,ψ (π ⊗ 1)m,
where ẼF,ψ (π ⊗ 1) ∈ OF ⊗OF t ,ψ W (R) is the evaluation at π ⊗ 1 in this W (OLt )-algebra.
We first consider the assertion of Proposition 4.4.5 in the ramified case. We have the de-
composition Pi = ⊕ψ Pi,ψ . By (4.4.6), we find β(P1,ψ , P2,ψ′ ) = 0 for ψ 6= ψ ′ . We consider the
restrictions of our bilinear forms
βψ :P1,ψ × P2,ψ −→ W (R)
β̃ψ :P1ψ × P2,ψ −→ OF ⊗OF t ,ψ W (R).
Lemma 4.4.6. Let K/F be ramified. Then:
β̃ψ (ẼAψ x1 , ẼAψ x2 ) = ẼF,ψ β̃ψ (x1 , x2 ), x1 ∈ P1,ψ , x2 ∈ P2,ψ , ψ 6= ψ0
β̃ψ0 (ẼAψ0 x1 , ẼAψ0 x2 ) = ẼF,0 β̃ψ (x1 , x2 ), x1 ∈ P1,ψ0 , x2 ∈ P2,ψ0 .
Proof. We can restrict ourselves to the case where R is a OẼ -algebra. Then we obtain



β̃ψ (Π ⊗ 1 − 1 ⊗ [ϕ(Π)])x1 , x2 = β̃ψ x1 , (−Π ⊗ 1 − 1 ⊗ [ϕ(Π)])x2


= −β̃ψ x1 , (Π ⊗ 1 − 1 ⊗ [ϕ̄(Π)])x2 .
In the case were ψ 6= ψ0 we deduce
β̃ψ (ẼAψ x1 , x2 ) = (−1)e β̃ψ (x1 , ẼBψ x2 ).
64 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

We find
Y
ẼBψ (Π ⊗ 1) · ẼAψ (Π ⊗ 1) = (Π ⊗ 1 − 1 ⊗ [ϕ(Π)])(Π ⊗ 1 − 1 ⊗ [ϕ̄(Π)])
ϕ∈Aψ
Y
= (Π ⊗ 1 − 1 ⊗ [ϕ(Π)])(Π ⊗ 1 + 1 ⊗ [ϕ(Π)])
ϕ∈Aψ
Y
= (−π ⊗ 1 + 1 ⊗ [ϕ(π)]) = (−1)e ẼF,ψ (π ⊗ 1).
ϕ∈Aψ

Therefore we obtain
β̃ψ (ẼAψ x1 , ẼAψ x2 ) = (−1)e β̃ψ (x1 , ẼBψ ẼAψ x2 ) = β̃ψ (x1 , ẼF,ψ x2 )
= ẼF,ψ β̃ψ (x1 , x2 ),
which finishes the proof for ψ 6= ψ0 .
We turn now to the case ψ0 . The polynomials ẼAψ0 , ẼBψ0 , and ẼF,0 are of degree e − 1. The
same computations yield for x1 ∈ P1,ψ0 and x2 ∈ P2,ψ0 ,
β̃ψ0 (ẼAψ0 x1 , x2 ) = (−1)e−1 β̃ψ0 (x1 , ẼBψ0 x2 ),
ẼBψ0 (Π ⊗ 1) · ẼAψ0 (Π ⊗ 1) = (−1)e−1 ẼF,0 (π ⊗ 1).
The assertion for ψ0 follows as before. 
We continue with the proof of Proposition 4.4.5 in the ramified case. We begin by showing
that
β̃ψ (Q′1,ψ , Q′2,ψ ) ⊂ QL,ψ . (4.4.7)
This is trivial for ψ 6= ψ0 . Let y1 ∈ Q′1,ψ0 and y2 ∈ Q′2,ψ0 . By Lemma 3.4.2, the inclusion (4.4.7)
is equivalent with
ẼF,0 β̃ψ0 (y1 , y2 ) ∈ OF ⊗OF t ,ψ0 I(R)
By Lemma 4.4.6 we find
ẼF,0 β̃ψ0 (y1 , y2 ) = β̃ψ0 (ẼAψ0 y1 , ẼAψ0 y2 ).
The elements u1 = ẼAψ0 y1 , resp., u2 = ẼAψ0 y2 , lie, by the definition of Q′1,ψ0 , in Q1,ψ0 , resp.,
by the definition of Q′2,ψ0 , in Q2,ψ0 . But for arbitrary elements u1 ∈ Q1,ψ0 and u2 ∈ Q2,ψ0 , we
have β(u1 , u2 ) ∈ I(R). By the definition of β̃ψ0 , we find
TrF/F t ϑ−1 ξ β̃ψ0 (u1 , u2 ) = β(ξu1 , u2 ) ∈ I(R),
for all ξ ∈ OF ⊗OF t ,ψ0 W (R). But this implies β̃ψ0 (u1 , u2 ) ∈ OF ⊗OF t ,ψ0 I(R), as desired.
Finally we have to check for y1 ∈ Q′1,ψ and y2 ∈ Q′2,ψ that
β̃ψ (Ḟ ′ y1 , Ḟ ′ y2 ) = ḞL β̃ψ (y1 , y2 ).
If ψ 6= ψ0 we find for the left hand side

Ḟ



β̃ψ Ḟ (ẼAψ )y1 , Ḟ (ẼAψ )y2 = ẼF,ψ β̃ψ (y1 , y2 ) = ḞL β̃ψ (y1 , y2 ) .
For ψ0 we obtain

Ḟ



β̃ψ0 Ḟ (ẼAψ0 )y1 , Ḟ (ẼAψ0 )y2 = ẼF,0 β̃ψ0 y1 , y2 ) = ḞL β̃ψ0 (y1 , y2 ) .
This ends the proof of Proposition 4.4.5 in the ramified case.
Now we consider the unramified case. We consider the decomposition (4.3.2). Let us denote
by eψ the idempotents corresponding to this decomposition. The conjugation of K/F maps eψ
to eψ̄ . We consider the corresponding decompositions Pi = ⊕Pi,ψ , for i = 1, 2. We obtain that
β(P1,ψ1 , P2,ψ2 ) = 0 for ψ2 6= ψ¯1 , and
β̃(P1,ψ , P2,ψ̄ ) ⊂ OF ⊗O W (R).
F t ,ψ̃

We note that there are natural isomorphisms

OK ⊗ W (R) ∼
OK t ,ψ̃ = OF ⊗ OF t ,ψ̃ W (R) ∼
= OK ⊗O ē W (R). (4.4.8)
K t ,ψ
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 65

Therefore β̃ induces an OF ⊗O W (R)-bilinear form

F t ,ψ̃

β̃ψ : P1,ψ × P2,ψ̄ −→ OF ⊗O W (R).

F t ,ψ̃

With the identification (4.4.8), we have ẼAψ̄ (π ⊗ 1) = ẼBψ (π ⊗ 1) ∈ OF ⊗O t ,ψ̃ W (R). The
F
analogue of Lemma 4.4.6 is
β̃ψ (ẼAψ x1 , ẼAψ̄ x2 ) = ẼF,ψ β̃ψ (x1 , x2 ), x1 ∈ P1,ψ , x2 ∈ P2,ψ̄ , ψ 6= ψ0 , ψ̄0 ,
(4.4.9)
β̃ψ (ẼAψ x1 , ẼAψ̄ x2 ) = ẼF,0 β̃ψ (x1 , x2 ), x1 ∈ P1,ψ , x2 ∈ P2,ψ̄ , ψ = ψ0 , ψ̄0 .
Here we recall again the notation introduced in the beginning of the proof: to be very precise,
the expression ẼF,ψ should be written as ẼF,ψ|F t . These identities follow from the identities
(
ẼF,ψ , ψ 6= ψ0 , ψ̄0 ,
ẼAψ ẼBψ =
ẼF,0 , ψ = ψ0 , ψ̄0 .
We need to check
β̃ψ (Q′1,ψ , Q′2,ψ̄ ) ⊂ QL,ψ . (4.4.10)
It suffices to consider the case ψ = ψ0 . By Lemma 3.4.2, the inclusion (4.4.10) is equivalent with
ẼF,0 β̃ψ0 (y1 , y2 ) ∈ OF ⊗OF t ,ψ0 I(R), y1 ∈ Q′1,ψ0 , y2 ∈ Q′2,ψ¯0 .
But, as in the ramified case, this is an immediate consequence of (4.4.9). Finally we have to
check that for y1 ∈ Q′1,ψ and y2 ∈ Q′2,ψ̄

β̃ψ (Ḟ ′ y1 , Ḟ ′ y2 ) = ḞL β̃ψ (y1 , y2 ).

For this we can repeat the last two formulas in the proof of the ramified case. 
Let R ∈ NilpOF . Let (P1 , ι1 ) be an object of dP′r,R .
We denote by (P1∆ , ι∆
the conjugate 1 )
Faltings dual. It is defined from the Faltings dual exactly as the conjugate dual from the dual.
Corollary 4.4.7. Let r be special. Let R ∈ NilpOE′ be such that the ideal of nilpotent elements
of R is nilpotent. We regard R as an OF -algebra via ϕ0 . Let (P, ι) be an object of dPr,R and
let (P ′ , ι′ ) ∈ dP′r,R be its image under the pre-contracting functor C′r,R . Then the image of the
conjugate dual (P ∧ , ι∧ ) under C′r,R is the Faltings dual ((P ′ )∆ , (ι′ )∆ ).
With the notation of Proposition 4.4.5, assume that P1∧ and P2 are in dPss r,R . Then the
canonical map
BilOK -anti-linear (P1 × P2 , Pm,R ) −→ BilOK -anti-linear (P1′ × P2′ , LR )
is bijective. Here these sets of bilinear forms Bil are meant as in Proposition 4.4.5.
Proof. We apply Proposition 4.4.5 to the canonical bilinear form βcan : P × P ∧ −→ Pm,R and
obtain
β̃can : P ′ × (P ∧ )′ −→ LR .
By Proposition 3.4.10, we obtain a morphism of displays
κ : (P ∧ )′ −→ (P ′ )∆ . (4.4.11)
By definition, β̃can is given by a perfect OF ⊗ W (R)-bilinear form
P × P ∗ −→ OF ⊗ W (R).
(Recall that P ∗ = HomW (R) (P, W (R)).) We obtain an isomorphism
∼
P ∗ −→ HomOF ⊗W (R) (P, OF ⊗ W (R)).
But this says exactly that the map which κ induces on the ”P -components” of the displays
(4.4.11) is an isomorphism. It is elementary to see that a morphism of displays κ : P1 −→ P2
which induces a W (R)-module isomorphism P1 −→ P2 is an isomorphism of displays.
Finally we prove the bijectivity of the last map in the corollary. The left hand side is, by
(3.2.5),
HomdPr,R (P1 , (P2 )∧ ).
66 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

This group is, by (iii) of Theorem 4.3.2, equal to

HomdP′ (P ′ , (P ∧ )′ ) ∼
r,R 1 2 = HomdP′ (P ′ , (P ′ )∆ ) ∼
r,R 1 = BilO
2 K -anti-linear (P1′ × P2′ , LR ).

We now combine the last corollary with Theorem 3.4.11.
Theorem 4.4.8. Let r be special. Let R ∈ NilpOE′ be such that the ideal of nilpotent elements
of R is nilpotent. Let (P1 , ι1 ) and (P2 , ι2 ) be objects of dPr,R , with images (P1′ , ι′1 ) and (P2′ , ι′2 )
under the pre-contracting functor C′r,R , cf. Proposition 4.3.2. Since the actions ι′i restricted
to OF are strict with respect to ϕ0 : OF −→ OE ′ −→ R, the Ahsendorf functor AOF /Zp ,R
may be applied to them. For i = 1, 2, let Pi,c = AOF /Zp ,R (Pi ), i = 1, 2, with its OF -algebra
homomorphism
ιi,c : OK −→ EndWOF (R) Pi,c .
If P1∧ and P2 are in dPss
r,R , then the natural homomorphism

BilOK -anti-linear (P1 × P2 , Pm,R ) −→ BilOK -anti-linear (P1,c × P2,c , Pm,WOF (R) (π ef /pf )).
is a bijection.
The twist Pm,WOF (R) (π ef /pf ) of the multiplicative display is defined in Example 3.1.6. More
precisely, this is the twist by the image of (π ef /pf ) under the canonical map OF −→ WOF (R).
Proof. This follows from Corollary 4.4.7 and Theorem 3.4.11. 

Remark 4.4.9. Let Ĕ ⊂ Q̄ ˆ be the completion of the maximal unramified extension of the
p
reflex field E of r. We extend ϕ0 : OF −→ OĔ to an embedding ϕ̆0 : OF̆ −→ OĔ . We denote
by τ ∈ Gal(F̆ /F ) the Frobenius automorphism. We apply the definition of η0 after Definition
3.4.12,
τ (η0 )η0−1 = π e /p, η0 ∈ OF̆× . (4.4.12)
Let R ∈ NilpOĔ . Via ϕ̆0 we consider R as an OF̆ -algebra. Therefore η0,R is defined, and
f
multiplication by η0,R defines an isomorphism
∼
Pm,WOF (R) (π ef /pf ) −→ Pm,WOF (R) , (4.4.13)
cf. (3.4.21). Therefore, if R ∈ NilpOĔ , we can ignore the twist by (π e /p) in Theorem 4.4.8.

We recall the definition of polarized CM-pairs Ppol

r,S , cf. Definition 4.1.2. We also introduce
the analogous category of polarized objects of dRR , as follows.
Definition 4.4.10. Let R ∈ NilpOF . We denote by dRpol R the category of triples (Pc , ιc , βc )
where (Pc , ιc ) ∈ dRR (cf. Definition 4.4.1) and where
βc : Pc × Pc −→ Pm,WOF (R)
is a polarization which is anti-linear for the OK -action ιc .
Let r be a special local CM-type with reflex field E. We regard an algebra R ∈ NilpOĔ as an
OF̆ -algebra via ϕ̆0 . We now define the contracting functor for polarized CM-pairs,
Cpol pol pol
r,R : dPr,R −→ dRR . (4.4.14)
Let (P, ι, β) ∈ dPpol
r,R . We apply the contracting functor Cr,R to (P, ι) and obtain (Pc , ιc ) ∈ dRR ,
cf. Definition 4.4.2. By Theorem 4.4.8, the polarization β : P × P −→ Pm,R induces an
alternating bilinear form
β̃c : Pc × Pc −→ Pm,WOF (R) (π ef /pf ). (4.4.15)
If we combine this with the chosen isomorphism (4.4.13), we obtain a polarization of the WOF (R)-
display Pc ,
βc : Pc × Pc −→ Pm,WOF (R) .
Then (Pc , ιc , βc ) is defined to be the image of (P, ι, β) by the functor Cpol
r,R .
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 67

Theorem 4.4.11. Let R ∈ NilpOĔ be such that the ideal of nilpotent elements is nilpotent. The
contracting functor Cpol
r,R induces an equivalence of categories

Cpol ss,pol
r,R : dPr,R −→ dRnilp,pol
R .
Let (Pc , ιc , βc ) the image of (P, ι, β) under the functor Cpol
r,R . Then
1
heightOF βc = height β,
f
cf. Definition 3.2.5.
Here the index ”ss” indicates the full subcategory of supersingular displays and the index
”nilp” the full subcategory of nilpotent displays.
Proof. We use the notation of Proposition 4.4.5. We have a commutative diagram
α̃
P ❯❯❯❯ − / HomO ⊗ W (R) (P, OF ⊗Zp W (R))
F Zp
❯❯❯❯
❯❯❯❯
❯❯❯❯ t∗
α ❯❯❯❯❯
❯❯❯* 
HomW (R) (P, W (R))

Here the map α̃ is induced by β̃ and the map α is induced by β. The vertical map is defined by
t∗ (ℓ) = t ◦ ℓ and is an isomorphism. The map α induces the isogeny P −→ P ∨ associated to β
and the map α̃ induces the isogeny P ′ −→ (P ′ )∆ . Therefore these isogenies have the same height.
If we apply the Ahsendorf functor to the last isogeny we obtain the map Pc −→ Pc∨ (π ef /pf )
which is associated to β̃c . By Proposition 3.3.17 we obtain
1
heightOF βc = heightOF β̃c = height β.
f

Remark 4.4.12. Let us explain how the bijection between bilinear forms of Theorem 4.4.8
simplifies when R = k is a perfect field in NilpOE′ . We take P1 = P2 .
We consider the Dieudonné module (P, F, V ) of P. We consider β : P × P −→ W (k) as a bi-
linear form of Dieudonné modules. Here we mean by W (k) the Dieudonné module (W (k), F, V ),
cf. (3.2.2). We define
β̃ : P × P −→ OF ⊗Zp W (k) (4.4.16)
as in Proposition 4.4.5. We know that β̃ induces a bilinear form of displays P × P −→ Lk . In
terms of Dieudonné modules, this means that the following equation holds,
β̃(V ′ x1 , V ′ x2 ) = VL β̃(x1 , x2 ). (4.4.17)
In terms of the decomposition (4.3.6), the operator V ′ is given by (4.3.15).
By (3.4.10), the Ahsendorf functor applied to Lk gives the WOF (k)-Dieudonné module
π ef f pf
(OF ⊗O W (k), F , ef −1 F −f ). (4.4.18)
F t ,ψ̃0 pf π
The bilinear form β̃ gives by restriction to Pc = P ⊗O t ,ψ̃0 W (k) ⊂ P the OF ⊗O t ,ψ̃0 W (k)-
F F
bilinear form
β̃c : Pc × Pc −→ OF ⊗O t ,ψ̃0 W (k). (4.4.19)
F

Because this is obtained by applying the Ahsendorf functor to (4.4.17), β̃c is a bilinear form of
WOF (k)-Dieudonné modules if we equip the right hand side with the WOF (k)-Dieudonné module
structure (4.4.18). Therefore we obtain
pf F −f
β̃c (Vc x1 , Vc x2 ) = β̃c (x1 , x2 ), x1 , x2 ∈ Pc . (4.4.20)
π ef −1
In the case where K/F is ramified, we have Pc = Pψ0 , and
Vc = Π−ef +1 V f : Pψ0 −→ Pψ0 ,
68 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

cf. (4.4.5). Note that (4.4.20) can be checked easily from these expressions.
In the case where K/F is unramified, we have Pc = Pψ0 ⊕ Pψ̄0 , and Vc is the endomorphism
of Pc = Pψ0 ⊕ Pψ̄0 given by the matrix
 
0 π −gψ0 V f
,
π −gψ̄0 V f 0
cf. (4.4.3). Before (4.4.8) we already remarked that β̃(Pψ0 , Pψ0 ) = 0 = β̃(Pψ̄0 , Pψ̄0 ). Again
(4.4.20) can be checked directly on these descriptions of Vc .
Now we assume moreover that k is a OĔ -algebra. We have the map (4.4.12),
OF̆ −→ OF ⊗O W (k) = WOF (k).
F t ,ψ̃0

We consider the image η0,k ∈ OF ⊗OF t ,ψ0 W (k) of η0 . We set

f
βc = η0,k β̃c : Pc × Pc −→ OF ⊗O W (k). (4.4.21)
F t ,ψ̃0

Then we find
f f pf F −f
βc (Vc x1 , Vc x2 ) = η0,k β̃c (Vc x1 , Vc x2 ) = η0,k β̃c (x1 , x2 )
π ef −1
f pf F −f −f F −f
= η0,k (η0,k βc (x1 , x2 )) = π βc (x1 , x2 ),
π ef −1
f −f −f
F
since η0,k (η0,k ) = π ef /pf . Indeed, the left hand side of the last identity is the image of
(η0 τ −1 (η0−1 ))f = π ef /pf .
This shows that βc is a bilinear form of WOF (k)-Dieudonné modules, if we consider on
OF ⊗O t ,ψ̃0 W (k) the WOF (k)-Dieudonné module structure which corresponds to Pm,WOF (R) ,
F
namely
(OF ⊗O t ,ψ̃0 W (k), F f , πF −f ).
F

Remark 4.4.13. Let us discuss the height identity in Theorem 4.4.11 in a more direct way. We
may assume that R is a perfect field. We may write the equation in the form
ordp detW (k) β = f ordπ detWOF (k) βc . (4.4.22)
On the left hand side the determinant is taken with respect to an arbitrary basis of the W (k)-
module P . After we take ordp , the result is independent of the choice of the basis. The right
hand side of this equation does not change if we replace βc by the form β̃c of (4.4.19). We
begin with the ramified case. The decomposition P = ⊕Pψ is orthogonal with respect to β. Let
βψ be the restriction to Pψ . Let ψ be banal. The map Π−e V : Pψσ −→ Pψ is a F −1 -linear
isomorphism. From the equation
F −1
βψ (Π−e V x, Π−e V y) = βψ (V (π −e x), V y) = βψσ (π −e px, y)
we conclude that ordp detW (k) βψ = ordp detW (k) βψσ . Therefore this value is independent of ψ.
In particular we obtain
ordp detW (k) β = f ordp detW (k) βψ0 = f ordπ detWOF (k) β̃c .
The last equation follows because
β̃c : Pψ0 × Pψ0 −→ WOF (k) = OF ⊗O ˜ W (k)
F t , ψ0

is obtained from βψ0 by the equation

TrWOF (k)/W (k) (ϑ−1 aβ̃c (x, y)) = βψ0 (ax, y), x, y ∈ Pψ0 , a ∈ OF ⊗O ˜ W (k),
F t , ψ0

and since the pairing

TrWOF (k)/W (k) (ϑ−1 a1 a2 ) : (OF ⊗O ˜ W (k)) × (OF ⊗O ˜ W (k)) −→ W (k)
F t , ψ0 F t , ψ0

is perfect.
In the unramified case we write ψσ f = ψ̄. The modules Pψ1 and Pψ2 are orthogonal for
ψ1 6= ψ¯2 . We denote by βψ the restriction of β to Pψ × Pψ̄ . We define ordp det βψ by taking
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 69

an arbitrary basis of Pψ and an arbitrary basis of Pψ̄ . Assume that ψ is banal; then ψ̄ is also
banal. We obtain two F −1 -linear isomorphisms
π −aψ V : Pψσ −→ Pψ , π −aψ̄ V : Pψ̄σ −→ Pψ̄ .
We have
−1
βψ (π −aψ V x, π −aψ̄ V y) = βψ (π −e V x, V y) = F βψσ (pπ −e x, y).
We conclude that ordp detW βψ = ordp detW βψσ . Because β is alternating, βψ0 and βψ̄0 have
the same order of determinant. We conclude that h := ordp detW βψ is independent of ψ ∈ Ψ.
We find ordp detW β = 2f h. The form β̃c is obtained from the restriction βψ0 by the equation
TrOF ⊗O ,ψ˜0 W (k)/W (k)
(aϑ−1 β̃c (x, y)) = βψ0 (ax, y), x ∈ Pψ0 , y ∈ Pψ̃0 , a ∈ OF .
Ft

Therefore we obtain
2h = 2 ordp detW βψ0 = ordπ detOF ⊗O ,ψ˜0 W (k)
β̃c .
Ft

4.5. The contracting functor in the case of a banal CM-type. In the banal case we will
associate to an object of the category dP′r,R (cf. Definition 4.3.1) an étale sheaf on Spec R. The
construction does not use the Ahsendorf functor AOF /Zp , which is not useful here.
Definition 4.5.1. Let R be a ring. An étale Frobenius module is a pair (M, Θ), where M is a
finitely generated W (R)-module which is locally on Spec R free and where Θ : M −→ M is a
Frobenius linear isomorphism, i.e., Θ : σ ∗ (M ) −→ M is an isomorphism.
The following proposition is a variant of a result of Drinfeld, comp. [10, Prop. 2.1]. It can also
be proved using the theory of displays. When R is an algebraically closed field, the proposition
is a theorem of Dieudonné.
Proposition 4.5.2. Let R be a ring such that p is nilpotent in R. There is a functor A from the
category of étale Frobenius modules over R to the category of locally constant p-adic étale sheaves
which are finitely generated and flat over Zp . The functor A is an equivalence of categories which
commutes with arbitrary base change. It is compatible with the tensor product of étale Frobenius
modules, resp., of p-adic étale sheaves.
Proof. We give a sketch of the proof which shows how this equivalence is constructed. By
[33, Lem. 42] it follows that the category of étale Frobenius modules over R and R/pR are
equivalent. Indeed, a Frobenius module lifts locally and by loc. cit. two liftings are canonically
isomorphic. Therefore we may assume pR = 0. Let (M, Θ) be an étale Frobenius module. We
set Mn = Wn (R) ⊗W (R) M . Because pR = 0, the Frobenius F on W (R) induces a Frobenius
F : Wn (R) −→ Wn (R). By base change we obtain a F -linear map Θn : Mn −→ Mn . We define
a functor An on the category of R-algebras,
An (S) = {x ∈ Wn (S) ⊗Wn (R) Mn | Θn,S (x) = x}.
One can show that An is representable by a finite étale scheme over Spec R. Clearly Z/pn Z =
Wn (Fp ) acts on An . We define the associated p-adic sheaf
A(M,Θ) = limn An .
←
Let W be the étale sheaf of Witt vectors. We have W ⊗Zp A(M,Θ) = M in the sense of étale
sheaves, where the action of Θ corresponds on the left hand side to the action of F ⊗ id.
Finally, we show the compatibility with tensor products. If (M ′ , Θ′ ) is a second étale Frobenius
module, we set (N, Ξ) = (M ⊗W (R) M ′ , Θ ⊗ Θ′ ). We obtain a natural homomorphism
A(M,Θ) ⊗Zp A(M ′ ,Θ′ ) −→ A(N,Ξ). (4.5.1)
To prove that this is an isomorphism, we may reduce by base change to the case where R is an
algebraically closed field. Then the assertion is clear by the theorem of Dieudonné. 
Definition 4.5.3. Let r be banal. Let R ∈ NilpOE′ . Let Et(OK )R be the category of locally
constant p-adic étale sheaves G over Spec R which are Zp -flat with rankZp G = 4d and which are
equipped with an action
ι : OK −→ EndZp G.
70 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

The contracting functor is the functor

Cr,R : dPr,R −→ Et(OK )R ,
which is the composite of the pre-contracting functor C′r,R of Theorem 4.3.2 and the functor A of
Proposition 4.5.2, applied to the étale Frobenius module (P ′ , Ḟ ′ ). The functor commutes with
arbitrary base change R −→ R′ .
Theorem 4.5.4. Let r be banal. Let R ∈ NilpOE′ be such that the ideal of nilpotent elements
of R is nilpotent. Then the contracting functor is an equivalence of categories,
Cr,R : dPr,R −→ Et(OK )R .
Proof. Since the objects in dP′r,R are étale, this is simply a combination of Proposition 4.3.2
and Proposition 4.5.2 

Remark 4.5.5. In the banal case there is a functor

Pr,R −→ dPr,R (4.5.2)
from p-divisible groups to displays which is an equivalence of categories. Indeed, in the case
when K/F is a field extension, the displays of objects in dPr,R are by Corollary 4.3.3 isoclinic
of constant slope 1/2 and therefore nilpotent. Therefore they are displays of formal p-divisble
groups, cf. Theorem 3.1.11. In the split case OK = OF × OF we have a corresponding decom-
position of a display P ∈ dPr,R : P = P1 ⊕ P2 . In the case where P is nilpotent we can argue as
before. If not, one of the summands is étale and the other is isoclinic of slope 1, cf. Corollary
4.3.3. But we have an equivalence between étale p-divisible groups over R and étale displays
over R, which is easily defined by the A-functor. Therefore we conclude also in this case that
the equivalence (4.5.2) exists.
We now add polarizations to the picture.
Lemma 4.5.6. Let r be banal. Let R ∈ NilpOE′ . Let (P1 , ι1 ) and (P2 , ι2 ) be in dPr,R . Let

β̄ : P1 /I(R)P1 × P2 /I(R)P2 −→ R
be an R-bilinear form such that
β̄(ι1 (a)x1 , x2 ) = β̄(x1 , ι2 (ā)x2 ), a ∈ OK .
Then the restriction of β̄ to Q1 /I(R)P1 × Q2 /I(R)P2 is zero.
Proof. We first consider the case where K/F is ramified. We consider Pi,ψ /I(R)Pi,ψ as an
OK ⊗OF t ,ψ R-module for i = 1, 2. Because of the isomorphism (4.2.6), it suffices to show that

β̄(EAψ x1 , EAψ x2 ) = 0, x1 ∈ (P1 /I(R)P1 )ψ , x2 ∈ (P2 /I(R)P2 )ψ .

We consider EAψ (Π ⊗ 1) ∈ OK ⊗OF t ,ψ OE ′ . The image of this element by the conjugation of
K/F is (−1)e EBψ (Π ⊗ 1), cf. the proof of Lemma 4.4.6. Therefore we find

β̄(EAψ x1 , EAψ x2 ) = (−1)e β̄(x1 , EBψ EAψ x2 ) = (−1)e β̄(x1 , Eψ x2 ) = 0.

Now we assume that K/F is unramified. Then the condition on β̄ implies that P1,ψ1 /I(R)P1,ψ1
and P2,ψ2 /I(R)P2,ψ2 are orthogonal with respect to β̄ if ψ1 6= ψ̄2 . Again by the isomorphism
(4.2.6), it suffices to show that
β̄(EAψ x1 , EAψ̄ x2 ) = 0, x1 ∈ (P1 /I(R)P1 )ψ , x2 ∈ (P2 /I(R)P2 )ψ̄ .

In this case the conjugation of K/F maps EAψ to EBψ̄ . Therefore the last equation follows from

EBψ̄ (π ⊗ 1)EAψ̄ (π ⊗ 1) = Eψ̄ (π ⊗ 1) = 0.

Exactly the same argument applies to the split case. 
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 71

Lemma 4.5.7. In the situation of the last lemma, assume that R is a reduced ring. Let β : P1 ×
P2 −→ W (R) be a W (R)-bilinear form such that β is anti-linear for the OK -actions ι1 , resp.
ι2 , and such that
F
β(F1 x1 , F2 x2 ) = p β(x1 , x2 ), x1 ∈ P1 , x2 ∈ P2 .
Then β induces a bilinear form of displays
β : P1 × P2 −→ Pm,R .
Proof. We must verify that β(Q1 , Q2 ) ⊂ I(R) and that
Ḟ
β(Ḟ1 y1 , Ḟ2 y2 ) = β(y1 , y2 ), y1 ∈ Q1 , y2 ∈ Q2 .
The inclusion is a consequence of Lemma 4.5.6. To verify the last equation, we may multiply
it by p2 because p is not a zero divisor in W (R). But then it follows from the assumptions on
β. 

Definition 4.5.8. Let ρ ∈ OF ⊗Zp W (R) be a unit. We define OF (ρ) as the p-adic étale sheaf
associated by Proposition 4.5.2 to the étale Frobenius module (OF ⊗Zp W (R), Θρ ), where
F
Θρ (a ⊗ ξ) = ρ · (a ⊗ ξ), a ∈ OF , ξ ∈ W (R). (4.5.3)
When ρ = 1 we obtain the constant p-adic étale sheaf OF = OF (1).
Let ρ = π e /p. Let R ∈ NilpOE′ and let (Pi , ιi ) ∈ Pr,R for i = 1, 2. We will associate to a
bilinear form of displays
β : P1 × P2 −→ Pm,R (4.5.4)
which is anti-linear for the OK -actions ι1 , resp. ι2 , a bilinear form of p-adic étale sheaves which
is anti-linear for the OK -actions on CP1 = Cr,R (P1 ), resp. CP2 = Cr,R (P2 ),
φ : CP1 × CP2 −→ OF (ρ). (4.5.5)
For the construction we may assume that R is a κE ′ -algebra because étale sheaves are insensitive
to nilpotent elements.
Let first K/F be ramified. Then we find for x1 ∈ P1 and x2 ∈ P2 that

β(Ḟ1′ x1 , Ḟ2′ x2 ) = β(Ḟ1 Πe x1 , Ḟ2 Πe x2 ) = Ḟ

β(Πe x1 , Πe x2 )
πe (4.5.6)
Ḟ
= β(π e x1 , x2 ) = F
β( x1 , x2 ).
p
We used the equation
Ḟ1′ x1 = Ḟ1 Πe x1 , (4.5.7)
which follows from (4.3.4) and (4.3.9). Recall the function t(a) = TrOF /Zp ϑ−1 a, for a ∈ OF ,
where ϑ ∈ OF is the different of F/Qp , cf. p. 63. We define β̃ : P1 × P2 −→ OF ⊗Zp W (R) by
the equation
t(ξ β̃(x1 , x2 )) = β(ξx1 , x2 ), ξ ∈ OF ⊗Zp W (R). (4.5.8)
Then β̃ is a bilinear form of OF ⊗Zp W (R)-modules. We conclude from (4.5.6) that

πe
β̃(Ḟ ′ x1 , Ḟ ′ x2 ) = · F
β̃(x1 , x2 ). (4.5.9)
p
Hence β̃ is a bilinear form of Frobenius modules. Since the functor A of Proposition 4.5.2
commutes with tensor products, it induces a bilinear form (4.5.5).
Now we consider the case where K/F is an unramified field extension. For each OE ′ -algebra
R we have the decomposition
Y
OK ⊗Zp W (R) = OK ⊗O W (R), (4.5.10)
K t ,ψ̃
ψ∈Ψ
72 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

which is induced by (4.3.2). The conjugation of K/F acts on OK ⊗Zp W (R) via the first factor.
We denote this by η 7→ η̄. We denote by F η the action of the Frobenius via the second factor.
On the right hand side of (4.5.10) these actions become
OK ⊗O W (OE ′ ) −→ OK ⊗O e f W (OE ′ )
K t ,ψ̃ K t ,ψσ

a ⊗ ξ 7−→ ā ⊗ ξ,
(4.5.11)
OK ⊗O W (R) −→ OK ⊗O W (R)
K t ,ψ̃ K t ,σ ψ̃
F
a ⊗ ξ 7−→ a⊗ ξ.
t
Here σ denotes the Frobenius automorphism of Gal(K /Qp ). Looking at the right hand side of
(4.5.10), we define
πr := (π aψ ⊗ 1)ψ∈Ψ ∈ OK ⊗Zp W (OE ′ ). (4.5.12)
It follows that
πr π̄r = π e ⊗ 1.
Let (P, ι) ∈ Pr,R . We note that, since R is a κE ′ -algebra, the definition of (P ′ , ι′ ) = C′r,R (P, ι)
in (4.3.9) takes the form
F˙ ′ (x) = Ḟ (πr x), F ′ (x) = F (πr x). (4.5.13)
Now let us start with a bilinear form
β : P1 × P2 −→ Pm
which is anti-linear for the OK -actions ι1 , resp. ι2 . We find
β(Ḟ1′ x1 , Ḟ2′ x2 ) = β(Ḟ1 πr x1 , Ḟ2 πr x2 ) = Ḟ
β(πr x1 , πr x2 )
πe (4.5.14)
Ḟ
= β(π e x1 , x2 ) = F
β( x1 , x2 ).
p
As before, β defines the OF ⊗Zp W (R)-bilinear form
β̃ : P1 × P2 −→ OF ⊗Zp W (R),
which by (4.5.14) satisfies

πe F
β̃(Ḟ1′ x1 , Ḟ2′ x2 ) =
· β̃(x1 , x2 ). (4.5.15)
p
Applying, as before, the A-functor to P ′ , we obtain the desired bilinear form (4.5.5).
Finally we consider the split case. In this case we consider in the decomposition (4.3.22) the
element


πr = πr,1 × πr,2 = (π aθ1 ⊗ 1)θ∈Θ × (π aθ2 ⊗ 1)θ∈Θ (4.5.16)
of (4.3.22). The conjugation acts on OK ⊗Zp W (OE ′ ) via the first factor. On the right hand side
of (4.3.22), the conjugation just interchanges the two factors in parentheses. This shows that
πr π̄r = π e ⊗ 1. Now starting with a bilinear form7 (4.5.4), the formulas (4.5.14), (4.5.15) from
the unramified case continue to hold, and this finishes the construction in the split case.
Proposition 4.5.9. Let r be banal. Let R ∈ NilpOE′ be such that the ideal of nilpotent elements
of R is nilpotent. Let (P1 , ι1 ) and (P2 , ι2 ) be objects of dPr,R . The construction above, which
associates to a bilinear form of displays (4.5.4) a bilinear form of p-adic étale sheaves (4.5.5) is
a bijection,
BilOK -anti-linear (P1 × P2 , Pm,R ) −→ BilOK -anti-linear (CP1 × CP2 , OF (ρ)).
Proof. We reduce the question to the case where R is reduced. Indeed, let S −→ R be a pd-
thickening in the category NilpOE′ . Assume that (Pi , ιi ) ∈ dPr,S for i = 1, 2. It follows from
Proposition 3.2.4 and Lemma 4.5.6 that any bilinear form
β̄ : P1,R × P2,R −→ Pm,R

7One should not confuse the notation P and P with the decomposition (4.3.25) which continues to exist,
1 2
e.g., P1 = P1,1 ⊕ P1,2 .
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 73

with the properties of (4.5.4) lifts uniquely to a bilinear form

β : P1 × P2 −→ Pm,S .
Since bilinear forms of étale sheaves have the same property, we can assume that R is reduced.
We begin with the ramified case. For i = 1, 2, let (Pi , ιi ) ∈ dPr,R , which correspond to (Pi′ , Ḟi′ )
under the pre-contracting functor C′r,R , cf. Theorem 4.3.2, and to CPi under the contraction
functor Cr,R . We start with a bilinear form of p-adic sheaves
πe
φ̃ : CP1 × CP2 −→ OF ( )
p
with the properties of (4.5.5). We have to construct a bilinear form of displays (4.5.4) which
induces φ. By Proposition 4.5.2, φ comes from a bilinear form of étale Frobenius modules
β̃ : P1′ × P2′ −→ OF ⊗Zp W (R)
which satisfies
πe F
β̃(Ḟ1′ x1 , Ḟ2′ x2 ) =
β̃(x1 , x2 ). (4.5.17)
p
After applying t we obtain a bilinear form β which satisfies
πe
β(Ḟ1′ x1 , Ḟ2′ x2 ) = F
β( x1 , x2 ). (4.5.18)
p
By (4.5.7) we may write
p
Fi = pḞi = Ḟi′
,
Πe
because multiplication by p is injective on Pi . We deduce from (4.5.18)
p p πe p p
β(F1 x1 , F2 x2 ) = β(Ḟ1′ e
x1 , Ḟ2′ e x2 ) = F
β( e
x1 , e x2 ) = p F
β(x1 , x2 ).
Π Π p Π Π
By Lemma 4.5.7, it follows that β is a bilinear form of displays β : P1 × P2 −→ Pm . This proves
the ramified case.
Now let K/F be an unramified field extension. We begin with a bilinear form of p-adic étale
sheaves (4.5.5) as before. This induces a bilinear form of étale Frobenius modules β̃ : P1′ ×P2′ −→
OF ⊗Zp W (R) which satisfies (4.5.17). Using (4.5.13), we rewrite this as
πe F
β̃(Ḟ1 πr x1 , Ḟ2 πr x2 ) = β̃(x1 , x2 ).
p
We multiply this equation with p2 and find for the left hand side
β̃(F1 πr x1 , F2 πr x2 ) = β̃( F πr F1 x1 , F
πr F2 x2 ) = β̃( F (π̄r πr )F1 x1 , F2 x2 ) = π e β̃(F1 x1 , F2 x2 ).
If we compare this to the right hand side multiplied with p2 , we obtain
F
β̃(F1 x1 , F2 x2 ) = p β̃(x1 , x2 ).
Setting now β = t ◦ β̃, the assumptions of Lemma 4.5.7 are satisfied. Therefore β induces a
bilinear form of displays β : P1 × P2 −→ Pm .
In the split case the argument is the same using the πr which appeared in this context. 

On Spec κ̄E we can choose a trivialization of the twisted constant étale sheaf,
∼
OF (π e /p) −→ OF , (4.5.19)
F −1 e
as follows. Choose η ∈ OF ⊗Zp W (κ̄E ) such that ηη = π /p (this is equivalent to the choice
of η0 in (4.4.12)). Then the multiplication by η
η : (OF ⊗Zp W (κ̄E ), (π e /p) ⊗ F ) −→ (OF ⊗Zp W (κ̄E ), 1 ⊗ F )
is an isomorphism of étale Frobenius modules, which induces (4.5.19) under the A-functor into
Et(OK )κ̄E , cf. Proposition 4.5.2.
74 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Definition 4.5.10. Let R ∈ NilpOĔ . Let Et(OK )pol R be the category of p-adic étale sheaves
(G, ι) ∈ Et(OK )R , equipped with a OF -linear alternating form
φ : G × G −→ OF , (4.5.20)
which is anti-linear for the OK -action.
Using the trivialization (4.5.19), and applying Proposition 4.5.9, we now obtain the contract-
ing functor with polarizations which is a functor from dPpol pol
r,R to Et(OK )R .

Theorem 4.5.11. Let r be banal. Let R ∈ Nilp OĔ be such that the ideal of nilpotent elements
in R is nilpotent. Then the contracting functor Cpol
r,R is an equivalence of categories,

Cpol pol pol

r,R : dPr,R −→ Et(OK )R .

Remark 4.5.12. In the split case, let CP = Cpol r,R (P). Let P = P1 × P2 be the decomposition
induced by OK = OF × OF . This induces a decomposition CP = CP,1 × CP,2 , where CP,i is
the étale sheaf associated to the Frobenius module (Pi , Ḟ πr,i ), i = 1, 2. Here the elements πr,i
are defined in (4.5.16). The subsheaves CP,i of CP are isotropic with respect to φ as in (4.5.20),
and hence φ corresponds to an OF -bilinear form
φ : CP,1 × CP,2 −→ OF .
Remark 4.5.13. Let k ∈ NilpOE′ be an algebraically closed field. Let (P, β) ∈ dPpol
r,k . We will
give a description of Cpol
r,R (P, β) = (CP , φ). We write P = (P, F, V ) as a Dieudonné module. The
image of P under the contracting functor, a sheaf CP on Spec k, is simply an OK -module.
Assume that K/F is ramified. From the definition of the pre-contracting functor (cf. Theorem
4.3.2) and the A-functor we have
CP = {x ∈ P | V −1 Πe x = x}.
To describe this further, with its bilinear form of displays, we extend the bilinear form β to
β̃ : P × P −→ OF ⊗ W (k),
cf. (4.5.8). The decomposition P = ⊕ψ∈Ψ Pψ is orthogonal with respect to β̃ and, by restriction,
we obtain for every ψ
β̃ψ : Pψ × Pψ −→ OF ⊗O W (k) ⊂ OF ⊗ W (k).
F t ,ψ̃

Let xψ , x′ψ ∈ Pψ . Since β is a polarization, we obtain

πe F
β̃ψσ (V −1 Πe xψ , V −1 Πe x′ψ ) =
β̃ψ (xψ , x′ψ ). (4.5.21)
p
The action of F on the right hand side is defined by (4.5.11). Fix ψa ∈ Ψ. The projection
x 7→ xψa is an isomorphism
CP ∼
= {xψ ∈ Pψ | V −f Πef xψ = xψ }.
a a a a (4.5.22)
In particular, we see that CP is indeed a free OK -module of rank 2. For x, x′ ∈ CP we obtain
from (4.5.21)
πe F
β̃ψσ (xψσ , x′ψσ ) = β̃ψ (xψ , x′ψ ).
p
Since ψa = ψa σ f , we obtain
 e f
π Ff
β̃ψa (xψa , x′ψa ) = β̃ψa (xψa , x′ψa ).
p
In the same way we may interpret the sheaf OF (π e /p): the projection
OF ⊗Zp W (k) −→ OF ⊗O W (k)
F t ,ψ̃a

defines an isomorphism
πe π e
f Ff
OF ( ) ∼
= {aψa ∈ OF ⊗OF t ,ψ̃a W (k) | aψa = aψa }. (4.5.23)
p p
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 75

f
For the last equation we may write ηaψa = F (ηaψa ) (cf. (4.5.19) for η) or, equivalently,
ηaψa ∈ OF . Therefore, using the expression (4.5.22) for CP , the restriction of β̃ to CP multiplied
by η gives the desired bilinear form
φ : CP × CP −→ OF
(4.5.24)
(xψa , x′ψa ) 7−→ η β̃ψa (xψa , x′ψa ).
Now let K/F be unramified. In this case, in the decomposition P = ⊕ψ∈Ψ Pψ , the summands
Pψ1 and Pψ2 are orthogonal, unless ψ1 = ψ̄2 . The OK -module CP is, in this case, given by
CP = {x = (xψ ) ∈ P | V −1 πr xψ = xψσ }, (4.5.25)
where we recall the element πr from (4.5.12). After fixing ψa , we can write
CP = {(xψa , xψ̄a ) ∈ Pψa ⊕ Pψ¯a | V −f π g xψa = xψ̄a , V −f π ḡ xψ̄a = xψa }, (4.5.26)
where g = aψa + aψa σ + · · · + aψa σf −1 and ḡ = aψ̄a + aψ̄a σ + · · · + aψ̄a σf −1 . Using the expression
(4.5.26) for CP , we may write
φ: CP × CP −→ OF (ρ)
(4.5.27)
(xψa + xψ̄a , yψa + yψ̄a ) 7−→ η β̃ψa (xψa , yψ̄a ) + η β̃ψ̄a (xψ̄a , yψa ).
We have for arbitrary elements xψ ∈ Pψ and yψ̄ ∈ Pψ̄ that
π ef F f
β̃ψ̄ (V −f π g xψ , V −f π ḡ yψ̄ ) = β̃ψ (xψ , yψ̄ ).
pf
If x = xψa + xψ̄a and y = yψa + yψ̄a in CP the last formula becomes
π ef Ff
β̃ψ̄ (xψ̄ , yψ ) = β̃ψ (xψ , yψ̄ ).
pf
By the formula (4.5.27) for φ we obtain
Ff
φ(x, y) = φ(x, y).
This shows again that φ(x, y) ∈ OF , cf. (4.5.23).
Finally let K = F × F . We use the notation of the last Remark. We obtain
CPi = {x ∈ Pi |V x = πr,i x },
where we recall the element πr,i from (4.5.15). We have the decomposition
CP = CP1 ⊕ CP2
t
If we fix θ0 ∈ Θ = HomQp -Alg (F , Q̄p ), the natural projection Pi → Pi,θ0 defines an isomorphism
CPi = {x ∈ Pi,θ0 |V f x = π gi x }, for i = 1, 2, (4.5.28)
P
where gi = θ aθi , cf. (4.3.28). The bilinear form β̃ induces by restriction
β̃θ0 : P1,θ0 × P2,θ0 → OF ⊗O W (k).
F t ,θ̃0

In the notation of (4.5.28) we obtain

φ : CP1 × CP2 → OF ,
(x1 , x2 ) 7−→ η β̃θ0 (x1 , x2 ).
This determines φ on CP since the subspaces CPi for i = 1, 2 are isotropic, cf. Remark 4.5.12.
Proposition 4.5.14. Let r be banal. Let k ∈ NilpOE′ be an algebraically closed field. Let
(P, ι, β) and (P + , ι+ , β + ) be two objects in dPpol
r,k .
(i) If K/F is split, then there exists a quasi-isogeny
(P, ι, β) → (P + , ι+ , β + ). (4.5.29)
(ii) Let K/F be a field extension. Then there exists a quasi-isogeny (4.5.29) iff inv(P, ι, β) =
inv(P + , ι+ , β + ), cf. (2.4.7).
(iii) Let K/F be an unramified field extension. If β is a polarization of height 2f h with h ∈ {0, 1}
then invr (P, ι, β) = (−1)h . For a given h, there exists (P, ι, β) with these properties.
76 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Proof. To prove the first assertion, we may apply the polarized contraction functor Cpol r,k of
Theorem 4.5.11. We choose an arbitrary isomorphism α1 of the F -vector spaces CP1 ⊗ Q and
CP + ⊗ Q. Since φ, resp. φ+ , define dualities of these spaces with CP2 ⊗ Q, resp. CP + ⊗ Q, we
1 2
can extend α1 to an isomorphism α : (CP , φ) ⊗ Q → (CP + , φ+ ) ⊗ Q.
If K/F is a field extension, we conclude by Proposition 8.3.6 that the equality inv(P, ι, β) =
inv(P + , ι+ , β + ) is equivalent to the equality inv((CP , ι, φ)⊗Q) = inv((CP + , ι+ , φ+ )⊗Q). There-
fore, by Definition 8.1.1, these anti-hermitian K-vector spaces are isomorphic, which proves our
assertion.
Finally we prove the last assertion. We consider the bilinear form βψ : Pψ ×Pψ̄ → W (k). If we
choose a W (k)-basis of Pψ and Pψ̄ , we can speak of ordp detW (k) βψ . This number is independent
of ψ and equals h. Let β(ψ) be the restriction of β to Pψ ⊕Pψ̄ . We obtain ordp detW (k) β(ψ) = 2h.
Recall β̃, cf. (4.5.8). Let β̃(ψ) the restriction of β̃,
β̃(ψ) : (Pψ ⊕ Pψ̄ ) × (Pψ ⊕ Pψ̄ ) → OF ⊗O W (k).
F t ,ψ̃

Then we have
ordp detW (k) β(ψ) = ordπ detOF ⊗O ,ψ̃ W (k)
β̃(ψ). (4.5.30)
Ft

Indeed, the function t(a) = TrOF /OF t ϑ−1 a, for a ∈ OF , where ϑ ∈ OF is the different of F/Qp ,
defines for an arbitrary OF ⊗O t ,ψ̃ W (k)-module U an isomorphism of OF ⊗O t ,ψ̃ W (k)-modules,
F F
∼
HomOF ⊗O ,ψ̃ W (k)
(U, OF ⊗O W (k)) −→ HomW (k) (U, W (k)), α̃ 7−→ α̃ ◦ t. (4.5.31)
Ft F t ,ψ̃

We apply this to U = Pψ ⊕ Pψ̄ . If we regard β̃ as a homomorphism of U to the left hand

side of (4.5.31) and β as a homomorphism from U to the right hand side, they correspond
to each other. Therefore the cokernels of these two homomorphisms are isomorphic and have
the same length. This shows (4.5.30). By Remark 4.5.13, we have for each ψ an isomorphism
CP ⊗OF (OF ⊗O t ,ψ̃ W (k)) ∼= P. Since φ coincides with the restriction of β̃(ψ) up to a unit,
F
we conclude that ordπ detOF φ = 2h. By Lemma 8.1.2, we have inv(CP , ι, φ) = (−1)h . By
Proposition 8.3.6 we are done. 

5. The alternative moduli problem revisited

In this section we give another proof of the main result of [17] which gives an alternative inter-
pretation of the Drinfeld moduli space of special formal OD -modules in the case of a quaternion
division algebra D over a p-adic local field F . We also prove a refinement concerning descent
data. The original proof was already simplified by Kirch [14], but the argument here is different
and is based on the theory of displays.

5.1. Special formal OD -modules. We fix the finite extension F of Qp , with uniformizer π and
residue field κF . Let R be an OF -algebra. Let (X, ι) be a p-divisible group over R with a strict
action ι : OF −→ End X. A relative polarization of X is a relative polarization of the display of
X. Here, by a relative polarization we mean one with respect to OF , cf. Definition 3.4.12.
If R = k is a perfect field, we may work with the associated WOF (k)-Dieudonné module
(M, F, V ) of X. It is obtained from the display of X by the Ahsendorf functor AOF /Zp ,R , cf.
Remark 3.3.16. In this language, a relative polarization is a WOF (k)-alternating pairing
ψ : M × M −→ WOF (k),
such that
F
ψ(F x, F y) = π ψ(x, y), x, y ∈ M, (5.1.1)
cf. (3.2.2). If the bilinear form ψ is perfect, we will say that the polarization is principal.
We denote by D the quaternion division algebra with center F . Let F ′ ⊂ D be a quadratic
unramified extension of F . Let OD ⊂ D be the ring of integers. Recall that a special formal
OD -module X over R is a p-divisible group X over R of height [D : Qp ] with an action ι : OD −→
End X such that the restriction of ι to OF is strict and such that Lie X is locally on Spec R a
free OF ′ ⊗OF R-algebra, cf. [9]. One can check that this condition is independent of the choice
of F ′ .
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 77

Proposition 5.1.1. Let k be an algebraically closed field of characteristic p with an OF -algebra

structure OF −→ k. Let F ′ be an unramified quadratic extension F ′ of F . We denote by F and
V the Frobenius and the Verschiebung acting on WOF (k).
Let (M, F, V ) be a WOF (k)-Dieudonné module (see Definition 3.1.8) of height 4 and dimension
2 which is endowed with an OF -algebra homomorphism
ι : OF ′ −→ End(M, F, V ).
Assume that ι makes M/V M into a free module of rank 1 over κF ′ ⊗κF k. Then there exists a
principal relative polarization ψ on (M, F, V ) such that
ψ(ι(u)x, y) = ψ(x, ι(u)y), for u ∈ OF ′ , x, y ∈ M (5.1.2)
Any other relative polarization φ of (X, ι) with the property (5.1.2) (with ψ replaced by φ) is of
the form
φ(x, y) = ψ(ι(c)x, y)
for some element c ∈ OF . ′

Proof. We choose an embedding OF ′ −→ WOF (k). We set, for i ∈ Z/2Z,

Fi
Mi = {x ∈ M | ι(u)x = ux, for u ∈ OF ′ }.
We have the decomposition
M = M0 ⊕ M1 . (5.1.3)
The operators F and V are of degree 1. The k-vector spaces M0 /V M1 and M1 /V M0 are by
assumption both of rank 1.
If ψ is a bilinear form with the properties (5.1.2), then the decomposition (5.1.3) is orthogonal.
We choose alternating perfect forms ψ̃0 resp. ψ̃1 on the free WOF (k)-modules M0 resp. M1 of
rank 2. These forms are unique up to a unit in WOF (k). By this uniqueness we find an equation
of the form
F −2
ψ̃0 (F 2 x0 , F 2 x′0 ) = ξπ 2 ψ̃0 (x0 , x′0 ), ξ ∈ WOF (k), for all x0 , x′0 ∈ M0 . (5.1.4)
2
By assumption we have ordπ det(F |M0 ) = 2. Comparing the determinants on both sides of
(5.1.4), we conclude that ξ is a unit. Since k is algebraically closed we may write
F −2
ξ= ηη −1 .
Replacing ψ̃0 by ψ0 := η ψ̃0 we may assume that we have ξ = 1 in equation (5.1.4).
With the same argument as before we find an equation
F −1
ψ0 (F x1 , F x′1 ) = ξ1 π ψ̃1 (x1 , x′1 ), ξ1 ∈ WOF (k), for all x1 , x′1 ∈ M1 .
Comparing the determinants we see that ξ1 ∈ WOF (k) is a unit. We set ψ1 = ξ1 ψ̃1 and
ψ = ψ0 ⊕ ψ1 (orthogonal sum). Then ψ satisfies (5.1.1). To prove (5.1.2) it suffices to show that
ψi (ι(u)xi , x′i ) = ψi (xi , ι(u)x′i ) for i = 0, 1.
This is trivial from the definition of (Mi , ψi ).
If we have a second φ satisfying (5.1.2), we find c ∈ WOF (k) such that
φ0 = cψ0 .
F2
Since both sides of this equation satisfy (5.1.4) with ξ = 1 we obtain c = c. Therefore we
have c ∈ OF ′ ⊂ WOF (k). We obtain that φ(x, y) = ψ(ι(c)x, y). 
Corollary 5.1.2. Let (N, ι) be a second WOF (k)-Dieudonné module of height 4 and dimension
2 with an action of OF ′ such that N/V N is a free κF ′ ⊗κF k-module of rank 1. Let ρ : N ⊗ Q −→
M ⊗ Q be a quasi-isogeny of height 0. Then
ψ(ρ(z), ρ(w)), z, w ∈ N (5.1.5)
is a perfect bilinear form on N .
Proof. Let ψN be a perfect alternating form on N given by Proposition 5.1.1, and let ψN0 be its
restriction to N0 . This form differs from the form (5.1.5) restricted to N0 by a factor in ζ ∈ F ′ .
Since ρ has height 0 we conclude that ζ is a unit. 
78 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Let K/F be a ramified quadratic extension of F generated by a prime element Π ∈ K such

that Π2 = −π. Let τ ∈ Gal(F ′ /F ) be the Frobenius automorphism. Let
OD = OF ′ [Π],
such that the following relations hold:
Πu = τ (u)Π, Π2 = −π, u ∈ OF ′ .
Then OD is the maximal order in the quaternion division algebra over F .
We have OK = OF [Π] ⊂ OD . We consider on OD the involution:
d = u + vΠ 7−→ d′ = u − Πv, u, v ∈ OF ′ . (5.1.6)
It is trivial on OF ′ and induces the conjugation of OK over OF .
Proposition 5.1.3. Let k be an algebraically closed field of characteristic p which is endowed
with an algebra structure OF −→ k. Let X be a special formal OD -module over k. Let (M, F, V )
be the WOF (k)-Dieudonné module of X. Then there exists a principal relative polarization
ψ : M × M −→ WOF (k),
on X such that
ψ(ι(d)x1 , x2 ) = ψ(x1 , ι(d′ )x2 ). (5.1.7)
Any other polarization with the property (5.1.7) is of the form uψ, with u ∈ OF .
Proof. We take ψ as in Proposition 5.1.1. Then we consider the alternating bilinear form
ψ1 (x, y) = ψ(ι(Π)x, ι(Π)y), x, y ∈ M.
Then ψ1 is by the uniqueness part of Proposition 5.1.1 of the form
ψ1 (x, y) = ψ(ι(c)x, y), c ∈ OF ′ .
If we apply the last equations to π ψ(x, y) = ψ(ι(Π)x, ι(Π)y), we obtain cτ (c) = π 2 . Therefore c
2

is divisible by π. We write c = aπ for some unit a ∈ OF ′ with aτ (a) = 1. We write a = uτ (u)−1

by Hilbert 90 and consider the form
ψ2 (x, y) = ψ(ι(u)x, y).
Then we have
ψ2 (ι(Π)x, ι(Π)y) = ψ(ι(u)ι(Π)x, ι(Π)y) = ψ(ι(Π)ι(τ (u))x, ι(Π)y) =
= ψ(ι(c)ι(τ (u))x, y) = ψ(ι(πu)x, y) = πψ2 (x, y).
Therefore ψ2 satisfies the requirements (5.1.7). The uniqueness assertion is proved as before. 
Definition 5.1.4. With the notation of Proposition 5.1.3 we call ψ a Drinfeld polarization on
the special formal OD -module X over k.
Proposition 5.1.5. Let X be a special formal OD -module over κ̄F . Let R be a OF̆ -algebra such
that p is nilpotent in R. Let (X, ι) be a special formal OD -module over R such that there exists
a quasi-isogeny of formal OD -modules
X ⊗κ̄F R/πR −→ X ⊗R R/πR. (5.1.8)
Then there is a principal relative polarization ψ on X which induces on OD the involution
d 7→ d′ and which is up to a factor in F × compatible with a Drinfeld polarization on X by the
quasi-isogeny (5.1.8).
Assume that R is noetherian and that Spec R is connected. Then any other relative polariza-
tion on X which induces the involution d 7→ d′ is of the form f ψ for some f ∈ OF .
Proof. We recall some generalities from [9] which are formulated there for Cartier modules.
We fix an embedding OF ′ −→ OF̆ . From this we obtain homomorphisms OF ′ −→ R and
λ : OF ′ −→ WOF (OF ′ ) −→ WOF (R). Let λ̄ be the composite with the conjugation of F ′ /F .
Let P be the WOF (R)-display of X. The action of OD on P is also denoted by ι. We have
the decompositions
P = P0 ⊕ P1 , Q = Q0 ⊕ Q1 (5.1.9)
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 79

such that for a ∈ OF ′ ⊂ OD the action of ι(a) on P0 is multiplication by λ(a) and the action
on P1 is multiplication by λ̄(a) and Qi = Q ∩ Pi . We regard (5.1.9) as a Z/2Z-grading. Then
F, Π, Ḟ are all homogenous of degree 1,
F : Pi −→ Pi+1 , Π : Pi −→ Pi+1 , Ḟ : Qi −→ Pi+1 . (5.1.10)
To show the existence of ψ, we begin with the case where R is reduced and Spec R is connected.
We remark that the set invariants of F : WOF (R) −→ WOF (R) is OF ⊂ WOF (R).
We assume that there exists a critical index i for X. We may suppose that i = 0, i.e., the
homomorphism
Π : P0 /Q0 −→ P1 /Q1
Π Ḟ
is zero. We consider the composite Φ : P0 −→ Q1 −→ P0 . We claim that Φ is a Frobenius
linear isomorphism. It is enough to show that det Φ ∈ WOF (R) is a unit. By base change we
may assume that R = k is a perfect field. Since i = 0 is critical we find ΠP0 ⊂ Q1 = V P0 .
Since P1 /V P0 and P1 /ΠP0 are k-vector spaces of dimension 1 we obtain ΠP0 = V P0 . Therefore
V −1 Π = Φ is bijective and therefore a Frobenius linear isomorphism.
For each n ∈ N we consider the functor on the category of R-algebras,
U0 (n) : S 7−→ (P0 ⊗WOF (R) WOF ,n (S))Φ , (5.1.11)
where the RHS denotes invariants of the Frobenius-linearly extended Φ. This functor is rep-
resentable by a scheme which is finite and étale over Spec R. Moreover the existence of the
quasi-isogeny (5.1.8) implies that this scheme is constant. The sheaf (5.1.11) is therefore with
its natural OF -module structure isomorphic to the constant OF -module (OF /π n OF )2 . We set
U0 = proj lim U0 (n) and obtain an isomorphism functorial in R-algebras S,
WOF (S) ⊗WOF (R) P0 ∼
= WOF (S) ⊗OF U0 .
We choose a perfect alternating pairing α : U0 × U0 −→ OF and extend it by base change to a
perfect bilinear pairing
ψ0 : P0 × P0 −→ WOF (R).
We extend this bilinear form to a form
ψ : (P ⊗ Q) × (P ⊗ Q) −→ WOF (R) ⊗ Q,
as follows. We note that π ∈ WOF (R) ⊗ Q is a unit. Therefore we may define a bilinear
alternating form on the WOF (R) ⊗ Q-module P1 ⊗ Q by the equation
1
ψ0 (Πx1 , Πx′1 ), x1 , x′1 ∈ P1 ⊗ Q.
ψ1 (x1 , x′1 ) =
π
We define ψ on P ⊗ Q as the orthogonal sum of ψ0 and ψ1 . Clearly ψ induces on OD the
involution d 7→ d′ . We show the equation
1 F
ψ(Ḟ x, Ḟ x′ ) =ψ(x, x′ ) = Ḟ ψ(x, x′ ), x, x′ ∈ P ⊗ Q. (5.1.12)
π
Since ψ0 is the linear extension of α, we find the equation
ψ0 (Ḟ Πx0 , Ḟ Πx′0 ) = F
ψ0 (x0 , x′0 ), x0 , x′0 ∈ P0 .
We note that Ḟ extends to an endomorphism of P ⊗ Q. This extension commutes with the
action of the field F . From the definition of ψ1 , we obtain
πψ1 (Ḟ x0 , Ḟ x′0 ) = ψ0 (ΠḞ x0 , ΠḞ x′0 ) = F
ψ0 (x0 , x′0 ), x0 , x′0 ∈ P0 ⊗ Q.
If we take x0 , x′0 ∈ Q0 we have ψ0 (x0 , x′0 ) ∈ IOF (R) because Q0 /IOF (R)P0 has rank 1 and ψ0 is
alternating. We deduce
1 F
ψ1 (Ḟ x0 , Ḟ x′0 ) =
ψ0 (x0 , x′0 ) = Ḟ ψ0 (x0 , x′0 ) ∈ WOF (R), x0 , x′0 ∈ Q0 .
π
This equation shows that ψ1 induces a pairing
ψ1 : P1 × P1 −→ WOF (R).
80 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

This pairing is perfect. Indeed, for the verification we may assume that R is a perfect field.
Then we already have a perfect pairing by Proposition 5.1.3, which agrees with the form ψ here
up to a unit constant.
To complete the verification of (5.1.12), we need to show that
ψ0 (Ḟ x1 , Ḟ x′1 ) = Ḟ
ψ1 (x1 , x′1 ), x1 , x′1 ∈ Q1 .
We may verify this in P ⊗ Q. We may assume that x1 = Πx0 , x′1 = Πx′0 . Then we find
1
ψ0 (Ḟ x1 , Ḟ x′1 ) = ψ0 (Ḟ Πx0 , Ḟ Πx′0 ) = F ψ0 (x0 , x′0 ) = 2 F ψ0 (Πx1 , Πx′1 )
π
1
= F ψ1 (x1 , x′1 ) = Ḟ ψ1 (x1 , x′1 ),
π
as desired.
We now drop the assumption that R is reduced but we continue to assume that the map
Π : P0 /Q0 −→ P1 /Q1 (5.1.13)
′
is zero. It suffices to show that, if R −→ R is a pd-thickening, then a polarization ψ on P lifts
to P ′ . Here P and P ′ are the displays of XR and XR′ . We apply the Grothendieck-Messing
lifting theorem for displays [1], cf. the end of subsection 3.1. Hence it suffices to see that
Q′ /I(R′ )P ′ is totally isotropic for the pairing on P ′ /I(R′ )P ′ × P ′ /I(R′ )P ′ induced by ψ. But
Q′ /I(R′ )P ′ = (Q′ /I(R′ )P ′ )0 ⊕ (Q′ /I(R′ )P ′ )1 as an orthogonal direct sum, and both summands
are free of rank one over R′ , locally on Spec R′ , whence the assertion.
For a general OF̆ -algebra R such that p is nilpotent in R we consider the closed subscheme
Spec R/a0 which represents the property that (5.1.13) is zero. In the same way Spec R/a1 is
defined from the map Π : P1 /Q1 −→ P0 /Q0 . We consider the exact sequence
0 −→ R/(a0 ∩ a1 ) −→ R/a0 × R/a1 −→ R/(a0 + a1 ) −→ 0.
Since we know the existence of ψ for R/a0 and R/a1 , a gluing argument shows the existence of ψ
for R/(a0 ∩ a1 ). Over a field one of the indices 0 or 1 is critical. This implies the V (a0 ) ∪ V (a1 ) =
Spec R, i.e. the ideal a0 ∩ a1 is nilpotent. We can as above apply the Grothendieck-Messing
criterion to R −→ R/(a0 ∩ a1 ) to obtain the existence of ψ for R.
For the last assertion we may assume by rigidity that R is an algebraically closed field. Let
ψ ′ be a second polarization. We may assume that the index 0 is critical. Then ψ ′ induces
by restriction an alternating pairing U0 × U0 −→ OF . This differs from α above by a factor
f ∈ OF . 
With the notation of the last Proposition we may regard ψ as an isomorphism X −→ X ∇ , cf.
opp opp
(3.4.22). By duality OD acts on X ∇ . If we compose this with the isomorphism OD −→ OD
defined by the involution (5.1.6), we obtain an action of OD on X ∇ . This gives a special formal
OD -module which we denote by X ∆ . The polarization ψ may be regarded as an isomorphism
of special formal OD -modules
λ : X −→ X ∆ . (5.1.14)
Corollary 5.1.6. Let ψX be a Drinfeld polarization on X. Let
ρ : X ⊗κ̄F R/πR −→ X ⊗R R/πR
be a quasi-isogeny of height 0. Then the relative quasi-polarization ψ̄ on X ⊗R R/πR induced by
ψX is a relative principal polarization that lifts to a relative polarization ψ on X.
Proof. We take a perfect relative polarization ψ on X which exists by Proposition 5.1.3. Then
the polarization induced from ψX must be of the form f ψ̄ for some f ∈ F × . Since ρ is of height
0, we conclude the f is a unit. Therefore the induced involution is perfect. 
Let us recall the Drinfeld moduli functor MDr on the category of schemes S over Spf OF̆ . We
will use the notation S = S ⊗Spf OF̆ Spec κF . We fix a special formal OD -module (Y, ιY ) over the
OF̆ -algebra κF . We call Y a framing object. By [9] there is a quasi-isogeny of height 0 between
any two choices. For a scheme S −→ Spf OF̆ , a point of MDr (S) consists of the following data
up to isomorphism:
(1) A special formal OD -module (Y, ι) over S.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 81

(2) A quasi-isogeny of OD -modules of height 0

ρ : Y ×S S −→ Y ×Spec κ̄F S. (5.1.15)
b F ×Spf OF Spf O .
The functor is representable by the p-adic formal OF̆ -scheme Ω F̆
We define the functor MDr (i) by replacing in (2) height 0 by the condition heightOF ρ = 2i.
We set a
M̃Dr = MDr (i).
i∈Z

Let (Y, ι) be a special formal OD -module. Let u ∈ D× . Then we define a new special formal
OD -module (Y u , ιu ) by setting
Y u = Y, ιu (d) = ι(u−1 du), for d ∈ OD .
u u
The multiplication ι(u) : (Y , ι ) −→ (Y, ι) is a quasi-isogeny of special formal OD -modules. We
obtain for each i ∈ Z an isomorphism of functors
∼
u : MDr (i) −→ MDr (i + ordD u), (Y, ρ) 7−→ (Y u , ιY (u)ρu ), (5.1.16)
×
This defines an action of D× on M̃Dr . If u ∈ OD , the multiplication by ι(u) defines an isomor-
phism ι(u) : (Y , ιY (u)ρ ) → (Y, ρ). Therefore the action of D× factors through ordD : D× → Z.
u u

We will call this action the translation.

We endow M̃Dr with a Weil descent datum relative to OF̆ /OF . Let τ ∈ Gal(F̆ /F ) be the
Frobenius automorphism. Let ε : OF̆ −→ R be an algebra in NilpOF̆ . We denote by R[τ ] the
ring R with the new OF̆ -algebra structure ε ◦ τ . The Frobenius τ induces τ̄ : κ̄F −→ κ̄F . We
have the Frobenius morphism
FY,τ : Y −→ τ∗ Y. (5.1.17)
f
For a κ̄F -algebra ε : κ̄F −→ R̄, we set φ(r) = rp for r ∈ R̄. This defines a κ̄F -algebra
homomorphism R̄ −→ R̄[τ ] . If we apply the functor Y we obtain (5.1.17). We will define a
morphism
ωMDr : MDr (i)(R) −→ MDr (i + 1)(R[τ ] ). (5.1.18)
Let (Y, ρ) ∈ MDr (i)(R). We define ρ′ as the composite
ρ ε∗ FY,τ
YR⊗O κ̄F −→ ε∗ Y −→ ε∗ τ∗ Y.
F̆

The image of (Y, ρ) under (5.1.18) is by definition (Y, ρ′ ). Since heightOF FY,τ = 2, we obtain
that heightOF ρ′ = 2i + 2. From (5.1.18) we obtain a Weil descent datum
ωMDr : M̃Dr (R) −→ M̃Dr (R[τ ] ) (5.1.19)
on the functor M̃Dr (compare [27]). We introduce the notation
(τ )
M̃Dr = M̃Dr ×Spf OF̆ ,Spf τ Spf OF̆ . (5.1.20)
(τ )
Then we have M̃Dr (R) = M̃Dr (R[τ ] ). We write (5.1.19) in the form
(τ )
ωMDr : M̃Dr −→ M̃Dr . (5.1.21)
The translation Π : MDr (i) → MDr (i + 1) is an isomorphism. We use it to identify these
functors. By Drinfeld’s theorem we obtain an isomorphism

M̃Dr ∼ b F ×Spf OF Spf O ) × Z.

= (Ω F̆ (5.1.22)
b F ×Spf OF Spf O .
We denote by ωτ the action of τ via the second factor on Ω F̆

Proposition 5.1.7. The Weil descent datum ωMDr induces on the right hand side of (5.1.22)
the Weil descent datum
ωMDr : (ξ, i) 7−→ (ωτ (ξ), i + 1). (5.1.23)
The translation functor is on the right hand side (ξ, i) 7→ (ξ, i + 1).
82 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Proof. Let (Y, ρ) ∈ MDr (R). Composing ωMDr with the translation we obtain a Weil-descent
datum on MDr ,
α : MDr (R) → MDr (Rτ ).
−1
It associates to (Y, ρ) the point (Y Π , ρ1 ), where ρ1 is the composite

Π −1 ρ −1 ε∗ FY,τ −1 ι(Π−1 )
YR⊗ O κ̄F −→ ε∗ YΠ −→ ε∗ τ∗ YΠ −→ ε∗ τ∗ Y. (5.1.24)
F̆

b F fits into a commutative diagram

Our assertion says that Drinfeld’s morphism MDr → Ω
α
MDr (R) / MDr (R[τ ] )
❏❏ s
❏❏
❏❏
❏❏ sssss
❏$ y ss
s
b F (R)
Ω

This is stated as an exercise in the proof of [27, Prop. 3.77], but we give the verification. We
have to go back to Drinfeld’s proof and therefore we use his notation. A point of Ω b F (R) is given
by data (η, T, u, r) ([9], §2, Thm.). Drinfeld constructs the data (η, T, u) entirely from a graded
−1
Cartier module M = ⊕Mi . The Cartier modules M and M[τΠ] are the same and the gradings
−1
are also the same because M 7→ M Π shifts the grading by 1 and M 7→ M[τ ] shifts the grading
by 1 in the opposite direction. Finally, we have to see that the rigidification r is not changed
by the application of αĞ . This can be checked on the geometric points of Spec R. But over an
algebraically closed field L, the rigidification is obtained as follows. We take the morphism of
rational Dieudonné modules
N →N
induced by (5.1.15) for S = S = Spec L. Then r is obtained by taking the invariants by V −1 Π
on both sides. We see from the definition (5.1.24) that αĞ does not change r. 

Let AutoD (Y) be the group of quasi-isogenies of Y which commute with the action of ιY . With
the notation of (5.1.9), let N = N0 ⊕ N1 = P ⊗ Q be the rational Dieudonné module of Y. The
natural map
−1
AutoD (Y) → GLF (N0V Π
)∼
= GL2 (F )

is an isomorphism. This group acts on the functor M̃Dr as follows. For g ∈ AutoD (Y) we define

g : MDr (i) → MDr (i + ord det g), (Y, ρ) 7−→ (Y, gρ).

The action commutes with the translation. Let JDr be the cokernel

Z → AutoD (Y) × Z → JDr → 0.

i 7−→ (π i , −2i)

The second group acts on M̃Dr such that the factor Z acts by translation. We obtain an action
of JDr on M̃Dr . We introduce the groups J ∗r and J ∗ur as cokernels

F× → AutoD (Y) × K × → J ∗r → 0,
f 7−→ (f, f −1 )
(5.1.25)
F× → AutoD (Y) × (F ′ )× → J ∗ur → 0.
f 7−→ (f, f −1 )

The homomorphisms ordK : K × → Z, resp. 2 ordF ′ : (F ′ )× → Z, induce homomorphisms

J ∗r → JDr , resp. J ∗ur → JDr . Therefore these groups act on M̃Dr .
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 83

5.2. The alternative theorem in the ramified case. Let K be a ramified quadratic extension
of F . We also assume that p 6= 2. We choose prime elements Π ∈ OK and π ∈ OF such that
Π2 = −π as in section 2. With the notation before Proposition 5.1.3, we regard OK as the
subring of OD .
For each i ∈ Z we define the functor N (i) = NK/F,r (i) on the category of schemes S −→
Spf OF̆ . We fix a special formal OD -module Y over κ̄F and we fix a Drinfeld polarization ψY .
We denote by λY : Y −→ Y∆ the isomorphism associated to ψY , cf. (5.1.14). We will consider
p-divisible groups X on S with an action ι : OK −→ End X such that the restriction of this
action to OF is strict. By duality we obtain an action of OK on the Faltings dual X ∇ . If we
compose this action with the conjugation of K/F we obtain ι∆ : OK −→ End X ∇ . We write
X ∆ = (X ∇ , ι∆ ) and call this the Faltings conjugate dual of (X, ι).
Definition 5.2.1. A point of N (i)(S) consists of the following data:
(1) A formal p-divisible group X over S with an action
ι : OK −→ End X,
such that the restriction of ι to OF is a strict action.
(2) An isomorphism of OK -modules λ : X −→ X ∆ which induces a relative polarization on X,
cf. Corollary 3.4.13.
(3) A quasi-isogeny of OK -modules
ρ : X ×S S −→ Y ×Spec κ̄F S.
We require that the following conditions are satisfied.
a) ρ respects the OK -actions. There is an element u ∈ OF× such that the following diagram of
quasi-isogenies is commutative
ρ
X ×S S̄ / Y ×Spec κ̄F S̄ (5.2.1)

uπ i λ λY
 
X ∆ ×S S̄ o Y∆ ×Spec κ̄F S̄.
ρ∆

b)
Tr(ι(Π) | Lie X) = 0. (5.2.2)
Two such data (X1 , ι1 , λ1 , ρ1 ) and (X2 , ι2 , λ2 , ρ2 ) define the same point of N (i)(S) iff there is
an isomorphism α : (X1 , ι1 ) −→ (X2 , ι2 ) which respects the polarizations up to a factor in OF×
and such that α commutes with ρ1 and ρ2 .
We note that changing λ by a factor in OF× does not alter the points of N (S). The existence
of ρ implies that dim X = 2 and that the OF -height of X is 4. The condition b) implies the
following Kottwitz condition for the characteristic polynomial,
char(ι(a) | Lie X) = (T − a)(T − ā), a ∈ OK . (5.2.3)
Clearly the functor N (i) does not depend on the choice of the Drinfeld polarization λY .
It follows from [27] that N (i) is representable by a formal scheme which is locally formally of
finite type over Spf OF̆ .
Let S = Spec R, R ∈ NilpOF̆ . Let PX be the WOF (R)-display associated to the p-divisible
∆
group X. The conjugate dual WOF (R)-display PX is nilpotent. It corresponds to X ∆ . We
denote by ψ : PX × PX −→ Pm,WOF (R) the bilinear form of displays which corresponds to λ.
We may reformulate the commutativity of the diagram (5.2.1) as follows: the quasi-polarization
ρ∗ ψY coincides with π i ψR/πR of (PX )R/πR up to a factor in OF× .
We obtain from (5.2.1) that
4i = 2 heightOF ρ.
As for the functors MDr (i) we have functor isomorphisms
∼
Π : N (i) −→ N (i + 1), (X, ρ) 7−→ (X, ι(Π)ρ), (5.2.4)
84 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

which we call the translations. Let τ ∈ Gal(F̆ /F ) be the Frobenius automorphism. Using the
Frobenius FY,τ : Y −→ τ∗ Y we define
ωN : N (R)(i) −→ N (i + 1)(R[τ ] ). (5.2.5)
exactly as ωMDr in (5.1.18). This defines a Weil descent datum ωN relative to OF̆ /OF on the
functor a
Ñ = N (i).
i∈Z
∗r
Lemma 5.2.2. The action of the group J from (5.1.25) on the OK -module Y gives an iso-
morphism
∼
J ∗r −→ J · ,
where
J · = {α ∈ AutoK (Y) | ψY (α(x), α(y)) = µ(α)ψY (x, y), for some µ(α) ∈ F × , x, y ∈ PY ⊗ Q}.

The group J ∗r acts on the functor Ñ by
(Y, ι, ρ) 7−→ (Y, ι, gρ), for g ∈ J ∗r .
We have a natural morphism of functors on NilpOF̆
MDr (i) −→ N (i). (5.2.6)
This is defined as follows. Let (Y, ι, ρ) ∈ MDr (i)(S) be a point. Let ψ be a Drinfeld polarization
on Y which is compatible with the quasi-isogeny ρ, cf. Proposition 5.1.5. It is uniquely deter-
mined up to a factor in OF× . Locally on S̄ we have ρ∗ ψY = f ψ for f ∈ F . Since heightOF ρ = 2i
we obtain ordπ f = i. Therefore (Y, ι|OK , ψ, ρ) ∈ N (i)(S).
The main result of [17] may now be formulated as follows. Note that in loc. cit. Weil descent
data were not considered.
Theorem 5.2.3 ([17]). Assume that p 6= 2. The functor morphisms (5.2.6) define a functor
isomorphism
∼
M̃Dr −→ Ñ
which commutes with the Weil descent data and the action of the group J · = J ∗r on both sides.
In particular it commutes with the translations.
It is clear that the morphism of functors M̃Dr −→ Ñ , is compatible with the Weil descent
data ωMDr and ωN relative to OF̆ /OF and with the translations (5.1.16) and (5.2.4). From this
we see that it commutes also with the actions of J ∗r . We need to prove that it is an isomorphism.
For the proof we need some preparations.
Let k be an algebraically closed field which is an OF̆ -algebra. We consider a WOF (k)-
Dieudonné module M of height 4 and dimension 2. We assume that an OK -action ι : OK −→
End M on M is given such that the restriction to OF is via OF −→ WOF (k).
Let
ψ : M × M −→ WOF (k) (5.2.7)
be a relative polarization, i.e., an alternating WOF (k)-bilinear form such that
F
ψ(F x1 , F x2 ) = π ψ(x1 , x2 ).
We require that
ψ(ι(a)x, y) = ψ(x, ι(ā)y), a ∈ OK .
Proposition 5.2.4. Let M be the WOF (k)-Dieudonné module of a special formal OD -module
with a Drinfeld polarization ψM , cf. Definition 5.1.4. Let (M, ι, ψ) be as above and such that ψ
is perfect. We assume that there exists an isomorphism of rational WOF (k)-Dieudonné modules
ρ : M ⊗ Q −→ M ⊗ Q, such that ρ is a homomorphism of OK -modules and such that ρ respects
the polarizations ψM and ψ up to a factor in F × .
Then there exists a unique OD -module structure on M such that M becomes the Dieudonné
module of a special formal OD -module and such that ρ is a quasi-isogeny of OD -modules.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 85

Proof. We will write ax := ι(a)x, for a ∈ OK and x ∈ M . For a ∈ OF this coincides by definition
with the action via OF −→ WOF (k).
We define W̃ = OK ⊗OF WOF (k). We extend the conjugation of K over F by linearity to
W̃ over WOF (k). We denote the traces of K/F and of W̃ /WOF (k) both by Tr. The Frobenius
endomorphism of WOF (k) extends OK -linearly to W̃ and is denoted by F . It will be impossible
to confuse this with the field F .
We define a hermitian form
h : M × M −→ W̃ ,
by requiring that
Tr ξΠ−1 h(x, y) = ψ(x, ξy) ξ ∈ W̃ , x, y ∈ M.
Then h is W̃ -linear in the second variable and hermitian,
h(x, y) = h(y, x).
The pairing h is perfect and satisfies the equation
F
h(F x, F y) = π h(x, y).
Since N := M ⊗ Q is the rational Dieudonné module of a special formal OD -module with its
Drinfeld polarization, we have a decomposition
N = N0 ⊕ N1 , (5.2.8)
which is orthogonal with respect to ψ (see (5.1.3)). One should note that N0 and N1 are not
W̃ -modules.
We note that for n0 , n′0 ∈ N0 , n1 , n′1 ∈ N1 we have

h(n0 , n′0 ) = 12 Πψ(n0 , n′0 ), h(n1 , n′1 ) = 12 Πψ(n1 , n′1 ),

(5.2.9)
h(n0 , n1 ) = 12 ψ(n0 , Πn1 ).
Indeed, the equation
Π
Trh(n0 , n′0 ) = ψ(n0 , Πn′0 ) = 0
Π
implies that Π−1 h(n0 , n′0 ) ∈ WOF (k) ⊗ Q. We obtain the first equation of (5.2.9):
2Π−1 h(n0 , n′0 ) = Tr Π−1 h(n0 , n′0 ) = ψ(n0 , n′0 ).
The proof of the next equation is the same. We have
Tr Π−1 h(n0 , n1 ) = ψ(n0 , n1 ) = 0.
This implies h(n0 , n1 ) ∈ WOF (k) ⊗ Q. We obtain the last equation of (5.2.9):
2h(n0 , n1 ) = Tr h(n0 , n1 ) = ψ(n0 , Πn1 ).
In particular we see from (5.2.9) that an element n0 ∈ N0 is isotropic for h.
We call an element x ∈ M primitive if it is not in ΠM . We find an element x ∈ M ∩ N0 such
that x ∈ / πM . Assume that x = Πy for some y ∈ M . Then y ∈ M ∩ N1 . Then it is clear that y
is a primitive element in M . Interchanging the role of the indices 0 and 1, we may assume that
x ∈ M ∩ N0 is primitive.
Since the pairing h is perfect and x is isotropic for h, we find an element y ′ ∈ M , such that
h(x, y ′ ) = 1. We can even choose y ′ to be isotropic for h. Indeed, we set y = y ′ + λx for some
λ ∈ W̃ . Then h(x, y) = 1. We compute:
h(y, y) = h(y ′ , y ′ ) + h(y ′ , λx) + h(λx, y ′ ) = h(y ′ , y ′ ) + λ + λ̄.
We choose λ = −(1/2)h(y ′, y ′ ) (which is legitimate, as p 6= 2) and obtain h(y, y) = 0. According
to (5.2.8) we write
y = y0 + y1 , y0 ∈ N0 , y1 ∈ N1 .
We write
1 = h(x, y) = h(x, y0 ) + h(x, y1 ).
86 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

We have h(x, y0 ) ∈ ΠWOF (k) ⊗ Q and h(x, y1 ) ∈ WOF (k) ⊗ Q by the formulas (5.2.9). This
implies h(x, y0 ) = 0 and h(x, y1 ) = 1. On the other hand, we find by (5.2.8)
0 = h(y, y) = h(y0 + y1 , y0 + y1 ) = h(y0 , y1 ) + h(y1 , y0 ).
Since h(y0 , y1 ) ∈ WOF (k) ⊗ Q, this implies h(y0 , y1 ) = 0 and then h(y, y0 ) = 0. The elements x
and y generate N as a W̃ ⊗ Q-vector space. Because we already proved that h(x, y0 ) = 0, we
conclude y0 = 0. Therefore y = y1 ∈ M ∩ N1 . We obtain
M = W̃ x + W̃ y,
because h is unimodular on the right hand side. Then the elements x, Πx, y, Πy are a basis of
the WOF (k)-module M . We have x, Πy ∈ M ∩ N0 and y, Πx ∈ M ∩ N1 and therefore
M = (M ∩ N0 ) ⊕ (M ∩ N1 ).
This shows that the OD -module structure on N induces an OD -module structure on M . 
Proof of Theorem 5.2.3. We consider the morphism (5.2.6) for i = 0 and denote it by
MDr −→ N . (5.2.10)
Clearly it is enough to show that this is an isomorphism. Proposition 5.2.4 shows that for
any algebraically closed field k which is an OF̆ -algebra, the induced map MDr (k) −→ N (k) is
bijective.
We note that the morphism (5.2.10) is formally unramified. Indeed, let S −→ R be a surjective
morphism in NilpOF̆ with nilpotent kernel. Let X be a p-divisible group over S with base change
XR over R. Then an OD -module structure on XR lifts by rigidity in at most one way to X.
We consider the underlying topological spaces in (5.2.6) with their induced structure of reduced
schemes. Then we obtain a formally unramified morphism of κ̄F -schemes
MDr,red −→ Nred (5.2.11)
These schemes are locally of finite type over κ̄F and have irreducible components which are
proper over κ̄F , cf. [27, Prop. 2.32]. Moreover, the morphism is bijective on geometric points.
Then the irreducible components of both schemes correspond bijectively to each other. We
consider a point x ∈ MDr (κ̄F ). Let X be the union of all irreducible components which pass
through x with the reduced scheme structure. Let y ∈ N (κ̄F ) be the image of x and define
Y ⊂ N in the same way as X. Then X −→ Y is a finite morphism. If we remove all points
in X resp. Y which belong to components not passing through x, resp. y, we obtain a finite
morphism of open neighbourhoods U −→ V of x ∈ MDr,red and y ∈ Nred . Therefore (5.2.11) is
a finite morphism of schemes locally of finite type over the algebraically closed field κ̄F . Since
this morphism is unramified and bijective on geometric points, it is an isomorphism.
Lemma 5.2.5. Let S −→ κ̄F be a surjective morphism in NilpOF̆ such that the kernel is nilpotent
and endowed with divided powers. Then the map
MDr (S) −→ N (S)
is bijective.
Let us assume that the lemma is proved. Then we consider points x and y as above. We
consider an open affine neighbourhood U of x. By the isomorphism (5.2.11) we regard U also
as a neighbourhood of y. Let n ∈ N. For a suitable ideal sheaf of definition J of N , we have a
homomorphism
(ON /J )(U ) −→ OMDr (U )/π n OMDr (U ).
This map is surjective modulo π by (5.2.11) and is therefore surjective. We note that by EGA0I ,
Prop. 7.2.4 the ring (ON /J )(U ) is π-adic. It follows that
ON (U ) −→ OMDr (U )
is surjective. Taking the inductive limit over U we obtain an epimorphism of local rings
Oy −→ Ox . (5.2.12)
By [27, Thm. 2.16] this is a homomorphism of noetherian adic rings (comp. EGA I, Prop. 10.1.6).
The ring Ox is, as a local ring of the scheme MDr , regular of dimension 2. Let my and mx be
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 87

the maximal ideals of the local rings. We remark that the squares of the ideals are open because
the topologies are adic.
We apply Lemma 5.2.5 to S = Oy /m2y . Then we obtain an oblique arrow which makes the
following diagram commutative,

Oy / Ox
✇✇
✇✇
✇✇✇
 {✇✇
Oy /m2y .

It follows that there is a surjective homomorphism mx /m2x −→ my /m2y . The epimorphism of

local rings also gives a surjection in the other direction. We conclude
dimκ̄F my /m2y = 2.
Therefore Oy is a regular local ring of dimension 2, and the map (5.2.12) is an isomorphism. It
follows that the map of sheaves
ON −→ OMDr
is an isomorphism. Finally let J be the maximal ideal sheaf of definition of ON . By the
isomorphism (5.2.12) we obtain an isomorphism
Oy /JOy −→ Ox /πOx .
Therefore JOy = πOy . Therefore J = πON is an ideal sheaf of definition. We obtain that
(5.2.10) is an isomorphism of formal schemes.
It remains to prove Lemma 5.2.5. We denote by m the kernel of S −→ κ̄F . Let ξ : Spec S −→
N be a morphism. We show that it lifts uniquely to Spec S −→ MDr . We denote by y ∈ N (κ̄F )
the point induced by ξ. Let x ∈ MDr (κ̄F ) be the unique point over y.
We denote by P the OF -display of the special formal OD -module over κ̄F which corresponds to
x. We denote by P̃ the unique WOF (S/κ̄F )-display which lifts P, cf. Theorem 3.1.12. We write
P̃ = (P̃ , Q̂, F, Ḟ ). The OD -action on P extends to P̃. Therefore we have the decompositions
P̃ = P̃0 ⊕ P̃1 , Q̂ = Q̂0 ⊕ Q̂1 .
We consider only the most interesting case where Π acts trivially on Lie P, i.e., Spec κ̄F −→
MDr is a singular point of the special fibre, cf. [9]. In this case we obtain Frobenius-linear
isomorphisms
Ḟ ◦ Π : P̃0 −→ Q̂1 −→ P̃0 , Ḟ ◦ Π : P̃1 −→ Q̂0 −→ P̃1 .
We set Ũi = {x ∈ P̃i | Ḟ ◦ Π(x) = x}. Then the canonical morphism WOF (S) ⊗OF Ũi −→ P̃i is
an isomorphism.
We can make the same construction with the display P. Then we obtain Ui ⊂ Pi such that
the canonical OF -module homomorphism Ũi −→ Ui is an isomorphism. Using our knowledge
about P we find elements ẽi ∈ Ui , for i = 0, 1 such that
ẽ0 , Πẽ1 ∈ P̃0 , ẽ1 , Πẽ0 ∈ P̃1 ,
are a basis of the WOF (S)-module P̃ . The natural polarization ψ on P extends to a polarization
ψ̃ on P̃ which is given by the conditions
ψ̃(ẽ0 , Πẽ1 ) = 1 = ψ̃(ẽ1 , Πẽ0 ),
and such that the decomposition P̃ = P̃0 ⊕ P̃1 is orthogonal with respect to ψ̃.
We classify now the liftings of Spec κ̄F −→ N to a point Spec S −→ N . We consider the
Hodge filtration L = Q/IOF (k)P ⊂ P/IOF (k)P . Since we compute now all the time modulo the
augmentation ideal IOF (k) ⊂ WOF (k), resp., IOF (S) ⊂ WOF (S), we continue to simply write
ẽ0 when we mean the residue class in P̃ /I(S)P̃ . The k-vector space L has the basis Πe0 , Πe1 .
Therefore a lifting of L to a direct summand L̃ ⊂ P̃ /I(S)P̃ has a unique basis of the form
f0 = Πẽ1 + γẽ0 + δẽ1 , f1 = Πẽ0 + αẽ0 + βẽ1 ,
88 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

because it is complementary to the module generated by ẽ0 , ẽ1 . Since we want a lifting of L we
have α, β, γ, δ ∈ m. The lifting L̃ determines a lifting of the display to S. The form ψ lifts to a
polarization of this display if and only if L̃ is isotropic under ψ. Therefore we must have
0 = ψ̃(Πẽ0 + αẽ0 + βẽ1 , Πẽ1 + γẽ0 + δẽ1 )
One obtains easily that the right hand side is
ψ̃(Πẽ0 , δẽ1 ) + ψ̃(αẽ0 , Πẽ1 ) = −δ + α.
Since the lifting L̃ should define a point of N , the condition 2) in the definition of points of N
implies
0 = Tr(Π | P̃ /L̃) = α + δ.
Because p 6= 2 we obtain α = δ = 0. This implies that L̃ = (L̃∩P̃0 )⊕(L̃∩P̃1 ). This shows that the
display over S defined by L̃ is the display of a special formal OD -module. Therefore the liftings
of Spec κ̄F −→ N to N (S) correspond via (5.2.6) bijectively to the liftings of Spec κ̄F −→ MDr
to a point of M(S). This proves Lemma 5.2.5 and Theorem 5.2.3.

The properties of Drinfeld’s moduli scheme MDr imply the following corollary, cf., e.g., [4].
Corollary 5.2.6. The formal scheme N is π-adic and has semi-stable reduction. The special
fiber N ⊗OF̆ κF of N is a reduced scheme. 
Finally we prove the uniqueness of the framing object, cf. (i) of subsection 2.5. We begin
with this question in the category dRpol
R , cf. Definition 4.4.10.

Proposition 5.2.7. Let r be special and let K/F be ramified. Let k ∈ NilpOF be an alge-
braically closed field. Let (Pc,1 , ιc,1 , βc,1 ) and (Pc,2 , ιc,2 , βc,2 ) be two objects in dRpol
k . Assume
that inv(Pc,i , ιc,i , βc,i )) = −1 for i = 1, 2. Then there exists a quasi-isogeny α : Pc,1 −→ Pc,2
which respects ιc,i and βc,i .
If the forms βc,i are perfect, then the actions ιc,i extend to actions ι̃c,i : OD −→ EndOF Pc,i
such that Pc,i becomes a special formal OD -module with Drinfeld polarization βi and such that
α becomes a homomorphism of OD -modules.
For the proof we need some preparations.
Lemma 5.2.8. Let K/F, r, k as in the last Proposition and let (Pc , ιc , βc ) ∈ dRpol
k . Assume that
inv(Pc , ιc , βc ) = −1, cf. Definition 8.3.1. Then the WOF (k)-display Pc is isoclinic of slope 1/2.
Proof. The K ⊗OF WOF (k)-vector space N = Pc ⊗ Q has dimension 2. The isoclinic decomposi-
tion of the WOF (k)-isocrystal N is invariant under the action of OK and has therefore at most
two summands. We have to show that there is only one summand. If not, we have N = N0 ⊕ N1 ,
where N0 is étale and N1 is dual to N0 . The dimension of each Ni as a K ⊗OF WOF (k)-vector
space is one. Therefore we find a generator e0 ∈ N0 such the Vc e0 = e0 . We use the notation
of before Definition 8.3.1. Let e1 ∈ N1 be the generator such that κc (e0 , e1 ) = 1. Let τ be the
Frobenius acting via the second factor on K ⊗OF WOF (k). From the equation
−1
κc (Vc e0 , Vc e1 ) = πκc (e0 , e1 )τ = π,
we conclude that Vc e1 = πe1 . Therefore Vc (e0 ∧ e1 ) = πe0 ∧ e1 . This implies that the invariant
of (Pc , ιc , βc ) is 1, which contradicts the assumption (Pc , ιc , βc ) = −1. 
Let (Pc , ιc ) ∈ dRk be isoclinic of slope 1/2. Then there is an WOF (k)-lattice Λ ⊂ Pc ⊗Q which
is invariant by π −1 Vc2 . Then there is also a lattice invariant by the ”square root” Π−1 Vc . One
deduces that the invariants C of Π−1 Vc acting on Pc ⊗ Q form a K-vector space of dimension 2
and
Pc ⊗ Q = C ⊗OF WOF (k). (5.2.13)
The anti-hermitian form κc associated to βc by (8.3.1) induces the anti-hermitian form on the
K-vector space C
κc : C × C −→ K. (5.2.14)
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 89

Indeed, for x, y ∈ C we find

F −1
πκc (x, y) = κc (Πx, Πy) = κc (Vc x, Vc y) = π κc (x, y).
This shows that κc (x, y) ∈ K ⊗OF WOF (k) is invariant by the Frobenius F acting on WOF (k),
and therefore this element is in K. The same argument shows that βc (x, y) ∈ F . The form κc
restricted to C is obtained from the restriction of βc to C by the formula
TrK/F (aκc (x, y)) = βc (ax, y), x, y ∈ C, a ∈ K.

Lemma 5.2.9. Let (Pc,1 , ιc,1 , βc,1 ) and (Pc,2 , ιc,2 , βc,2 ) be objects of dRpol
k such that Pc,1 and
Pc,2 are isoclinic of slope 1/2. Then the canonical map



Hom (Pc,1 , ιc,1 , βc,1 ), (Pc,2 , ιc,2 , βc,2 ) ⊗ Q −→ HomK (C1 , βc,1 ), (C2 , βc,2 )
is an isomorphism.
Proof. This is an immediate consequence of the isomorphism (5.2.13) because the K-action, βc ,
and Vc on Pc ⊗ Q can be recovered from the right hand side of the isomorphism. The map Vc is
induced from Π ⊗ F −1 on the right hand side. 
Lemma 5.2.10. There is the following relation between the invariants defined in Definition
8.3.1 and in Definition 8.1.1,
inv(Pc , ι, βc ) = −inv(C, βc ). (5.2.15)
We remark here that (C, βc ) determines (Pc , ι, βc ) up to isogeny.
Proof. Let x1 , x2 be a basis of the K-vector space C. Then the right hand side of (5.2.15) is
given by the 2 × 2-determinant
det(κc (xi , xj )).
By definition of C we have Vc xi = Πxi . We conclude that Vc (x1 ∧ x2 ) = −π(x1 ∧ x2 ) in ∧2K C.
From Lemma 8.3.3 we obtain that the determinant above gives −inv(Pc , ι, βc ). 
Lemma 5.2.11. Let (Psp , ιsp ) the WOF (k)-display of a special formal OD -module. We denote
by ψ a Drinfeld polarisation. Let ι′sp be the restriction of ιsp to OK ⊂ OD . Then (Psp , ι′sp , ψ) ∈
dRpol
k , and
inv(Psp , ι′sp , ψ) = −1.
Proof. We write M = Psp and consider it as a WOF (k)-Dieudonné module. Let N = M ⊗ Q.
Then ψ is a relative polarization that satisfies (5.1.2). By the decomposition (5.1.3) (or (5.1.9))
we obtain a decomposition
N = N0 ⊕ N1 ,
which is orthogonal with respect to ψ. As in the proof of Lemma 5.2.9, we consider the in-
−1
variants Csp = N V Π . Because V −1 Π is homogenous of degree zero, the decomposition of N
induces Csp = C0 ⊕ C1 . Each Ci is a F -vector space of dimension 2. The restriction of ψ is a
nondegenerate alternating pairing
ψ : Csp × Csp −→ F
Let κ : Csp × Csp −→ K be the anti-hermitian form associated to ψ as before Lemma 5.2.9. We
choose a basis e0 , f0 of the F -vector space C0 such that ψ(e0 , f0 ) = 2. We claim that
κ(e0 , e0 ) = κ(f0 , f0 ) = 0, κ(e0 , f0 ) = 1. (5.2.16)
Indeed, we write κ(e0 , e0 ) = a + Πb, a, b ∈ F . By definition of κ we find
TrK/F (κ(e0 , e0 )) = ψ(e0 , e0 ) = 0, TrK/F (Πκ(e0 , e0 )) = ψ(Πe0 , e0 ) = 0.
The last equation follows because C0 and C1 are orthogonal. This implies a = b = 0. Clearly it
is enough to verify the last equation of (5.2.16). Again we write κ(e0 , f0 ) = a + Πb, a, b ∈ F .
Then we find
TrK/F (κ(e0 , f0 )) = ψ(e0 , f0 ) = 2, TrK/F (Πκ(e0 , f0 )) = ψ(Πe0 , f0 ) = 0,
90 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

and therefore a = 1 and b = 0. Since e0 , f0 is a basis of the K-vector space Csp , the determinant
 
κ(e0 , e0 ) κ(e0 , f0 )
det =1
κ(f0 , e0 ) κ(f0 , f0 )
gives the invariant 1 = inv(Csp , ψ) = −inv(Psp , ι′sp , ψ) by the last Lemma. 

Proof. (of Proposition 5.2.7) By Lemma 5.2.8 we know that Pc,i is isoclinic of slope 1/2 for
i = 1, 2. Therefore Lemma 5.2.9 is applicable. By Lemma 5.2.10, the associated K-vector
spaces (Ci , βc,i ) have the same invariant 1 and are therefore isomorphic. Therefore we find the
quasi-isogeny α by Lemma 5.2.9.
We use the notations of Lemma 5.2.11. By what we just proved we find a quasi-isogeny
(Psp , ι′sp , ψ) −→ (Pc,1 , ιc,1 , βc,1 ). If βc,1 is perfect, this quasi-isogeny extends by Proposition
5.2.4 to a quasi-isogeny of special formal OD -modules and so does α. 

We can now prove the uniqueness of the framing object.

Proposition 5.2.12. Let r be special and K/F ramified. Let k be an algebraically closed field
in NilpOE . Let (P, ι, β) ∈ Ppol r,k be an object such that β is perfect, cf. Definition 4.1.2. Assume
that invr (P, ι, β) = −1. Then P is isoclinic of slope 1/2.
If moreover (P1 , ι1 , β1 ) is a second triple with the same properties, then there is a quasi-isogeny
of height zero
ρ : (P, ι, β) −→ (P1 , ι1 , β1 ),
such that there is an f ∈ OF× with
β1 (ρ(x), ρ(y)) = β(f x, y), x, y ∈ P.

Proof. We apply the functor Cpolr,k to (P, ι, β) and obtain (Pc , ιc , βc ), cf. (4.4.14). By the definition
of this functor, βc is perfect. We conclude from Proposition 8.3.2 that inv(Pc , ιc , βc ) = −1. By
Lemma 5.2.8, Pc is isoclinic of slope (1/2). By Corollary 4.3.3 and Proposition 3.3.17, P is
isoclinic of slope 1/2.
By Proposition 5.2.7, we find a quasi-isogeny α : (Pc , ιc , βc ) −→ (Pc,1 , ιc,1 , βc,1 ) which we can
make into a quasi-isogeny of special formal OD -modules. The height of α is then a multiple
of 2. Composing α with an endomorphism of the special formal OD -module (Pc , ιc ), we can
obtain a quasi-isogeny of height 0 of OD -modules ρc : (Pc , ιc ) −→ (Pc,1 , ιc,1 ). Then ρc respects
the Drinfeld polarizations βc and βc,1 up to a constant in OF× . By Theorem 4.4.11, we obtain a
quasi-isogeny of height zero as claimed in the proposition. 

Remark 5.2.13. We chose here the framing object for N as coming from the Drinfeld moduli
problem. It can also be characterized in terms of the moduli problem N , cf. [14]: it is a triple
(X, ι, λ) consisting of a p-divisible strict formal OF -module X over κF , with an action ι of OK
satisfying the Kottwitz condition (5.2.3), and a perfect relative polarization λ such that the
special automorphism group is isomorphic to SL2 (F ), comp. [14, Prop. 3.2].
5.3. The alternative theorem in the unramified case. In this subsection K denotes an
unramified quadratic extension of F . Let k be an algebraically closed field of characteristic p
which is endowed with an OF -algebra structure. We will sometimes write F ′ = K if we refer to
subsection 5.1. Let τ be the Frobenius of F ′ /F . We write τ (a) = ā for a ∈ OF ′ .
Let M be the WOF (k)-Dieudonné module of a special formal OD -module over k, as in Propo-
sition 5.1.3. In addition to ψ, we use another polarization of M ,
θ : M × M −→ WOF (k).
This is an alternating bilinear form of M which satisfies
F
θ(F x1 , F x2 ) = π θ(x1 , x2 ), x1 , x2 ∈ M
θ(ι(a)x1 , x2 ) = θ(x1 , ι(ā)x2 ), a ∈ OF ′ ,
(5.3.1)
θ(ι(Π)x1 , x2 ) = θ(x1 , ι(Π)x2 ),
ordπ det θ = 2.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 91

The polarization θ is unique up to a constant in OF× . It is constructed as follows: We choose

an element δ ∈ OF×′ , such that δ + τ (δ) = 0. We set Π1 = δΠ. Then Π1 is invariant under the
involution (5.1.6) and therefore we have
ψ(ι(Π1 )x, y) = ψ(x, ι(Π1 )y).
We define
θ(x, y) = ψ(ι(Π1 )x, y). (5.3.2)
We see that θ is alternating. It induces on D the involution given by
Π† = Π, u† = τ (u), for u ∈ F ′ . (5.3.3)
Conversely, assume that θ is a polarization with the properties (5.3.1). Let ψ1 (x, y) = θ(Π1 x, y).
Using Π1 Π = −ΠΠ1 we see that ψ1 satisfies the properties (5.1.7). By Proposition 5.1.3 this
shows the uniqueness of θ with the properties above.
Let (Y, ιY ) be a special formal OD -module over the OF -algebra κF and such that ι(Π) acts
as zero on Lie Y. We endow Y with the polarization θY defined above, cf. (5.3.2).
Definition 5.3.1. We define for each i ∈ Z the functor N (i) = NK/F (i) on the category
(Sch/ Spf OF̆ ). A point of N (i)(S) consists of the following data:
(1) A formal p-divisible group X over S with an action
ι : OK −→ End X,
such that the restriction of ι to OF is a strict action.
(2) A relative polarization θ on X such that the determinant of θ is π 2 up to a unit and such
that θ induces on OK the conjugation over OF .
(3) A quasi-isogeny of OK -modules
ρ : X ×S S −→ Y ×Spec κ̄F S.
Here, if S = Spec R, the condition in (2) means that the polarization of the corresponding OF -
display P of Y has determinant π 2 , up to a unit in WOF (R). We require that the following
conditions are satisfied.
a) ρ respects OK -actions. The relative quasi-polarization ρ∗ θY differs from π i θ by a factor in
OF× .
b) Lie X is locally on S a free OK ⊗OF OS -module of rank 1.
We note that, as in the ramified case, the OF -height of X is 4 and the dimension 2. The
condition b) implies the following Kottwitz condition for the characteristic polynomial:
char(ι(a) | Lie X) = (T − a)(T − ā), a ∈ OK .
It follows from [27] that N (i) is representable by a formal scheme which is locally formally of
finite type over Spec OF̆ . The functor N (0) will be also denoted by N .
We have a natural functor morphism
MDr (i) −→ N (i). (5.3.4)
Indeed, let (Y, ρ) ∈ MDr (i)(R). Then we have the Drinfeld polarization ψ of Y and we define
θY by the formula (5.3.2). This gives a point of N (R)(i).
The diagram similiar to (5.2.1) shows that
heightOF ρ = 2i.
We will define a translation functor isomorphism
∼
Π : N (i) −→ N (i + 1). (5.3.5)
Let (Y, ι) be a special formal OD -module over R ∈ NilpOF . We fix a Drinfeld polarization ψ.
This is also a Drinfeld polarization for (Y Π , ιΠ ). For the polarizations θ and θΠ derived by
(5.3.2), we obtain θΠ = −θ. We consider the morphism ιΠ : Y Π −→ Y . One easily checks that
θ(ι(Π)x, ι(Π)y) = πθΠ (x, y).
This is an identity of bilinear forms on the WOF (R)-display of Y .
92 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

If (X, ι) is a p-divisible OK -module, we define the conjugate p-divisible OK -module (X c , ιc )

by setting X c = X and ιc (a) = ι(ā) for a ∈ OK . For the special formal OD -module Y we have
(Y c , (ι|OK )c ) = (Y Π , ιΠ
|OK ).

Let R ∈ NilpOF̆ and let (X, ι, θ, ρ) ∈ N (i)(R). We define

ρ ι(Π)
ρc : XR̄
c
−→ YcR̄ = YΠ
R̄ −→ YR̄ .
We set θc = −θ. Then (X c , ιc , θc , ρc ) ∈ N (i + 1)(R). This defines the translation functor
morphism (5.3.5). It is clearly an isomorphism. With this definition, the functor morphism
(5.3.4) commutes with the translations on source and target.
Let τ ∈ Gal(F̆ /F ) be the Frobenius automorphism. Using the Frobenius FY,τ : Y −→ τ∗ Y,
we obtain a morphism
ωN : N (i) −→ N (i + 1)(τ )
with the same definition as (5.1.18). This induces a Weil descent datum ωN on
a
Ñ = N (i).
i∈Z
∗ur
Lemma 5.3.2. The action of the group J (cf. (5.1.25) for F ′ , which is now denoted by K)
on the OK -module Y gives an isomorphism
∼
J ∗ur −→ J · ,
where
J · = {α ∈ AutoK (Y) | θY (α(x), α(y)) = µ(α)θY (x, y), for some µ(α) ∈ F × , x, y ∈ PY ⊗ Q}

·
The group J acts via the rigidification ρ on the functor Ñ .
Theorem 5.3.3 ([17]). The morphisms of functors (5.3.4) extend to a functor isomorphism
∼
M̃Dr −→ Ñ
which commutes with the Weil descent data, the actions of J ∗ur = J · , and the translations on
both sides.
Proof. We already checked that (5.3.4) extends to a functor morphism which respects trans-
lations and Weil descent data on both sides. Therefore it suffices to see that (5.3.4) is an
isomorphism for i = 0,
∼
MDr −→ N . (5.3.6)
We begin with the case where R = k is an algebraically closed field. Let Y ∈ N (k). Let M be
the OF -Dieudonné module of Y and let M be the OF -Dieudonné module of Y. The quasi-isogeny
ρ induces an isomorphism M ⊗ Q ∼ = M ⊗ Q. The polarization ψM induces a polarization ψ1 on
M ⊗ Q. Since ρ is of height zero and ordπ det ψM = 0 we conclude that ordπ det ψ1 = 0. On the
other hand, we have by Proposition 5.1.1 a perfect pairing ψ on M which differs from ψ1 by an
element f ∈ F ′ . This shows that ψ1 is perfect on M . Then we define the action ι(Π1 ) by the
equation
θ(x, y) = ψ(ι(Π1 )x, y), x, y ∈ M. (5.3.7)
Therefore the morphism (5.3.6) evaluated at k is bijective.
Since both functors of (5.3.6) are representable by formal schemes locally of finite type,
it suffices now to check the following statement. Let S −→ R be a surjective OF̆ -algebra
homomorphism such that S and R are artinian local rings with algebraically closed residue class
field. Assume that (5.3.6) is bijective when evaluated at R. Then it is bijective when evaluated
at S. We may assume that the kernel of S −→ R is endowed with divided powers.
We consider a point Ye ∈ N (S) and we denote by Y ∈ N (R) its reduction. By our assumption,
Y carries the structure of an OD -module compatible with ρ. Therefore ψY induces a perfect
polarization on the OF -display P of Y . The OF -display P e of Ye is a lifting of P. By the
crystalline property of displays [1], cf. end of subsection 3.1, we obtain a perfect pairing
ψe : Pe × Pe −→ WO (S). F
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 93

The involution induced by ψe on OF ′ is trivial. It follows that the decomposition

Pe = Pe0 ⊕ Pe1
e The
according to the two OF -algebra embeddings OF ′ −→ OF̆ is orthogonal with respect to ψ.
Hodge filtration
Q̃i /IOF P̃i ⊂ P̃i /IOF P̃i , i = 0, 1
is isotropic with respect to ψ because these direct summands are of rank 1. Therefore ψe is
e
a polarization of the OF -display P.e Using the given polarization θe on P,
e we can define the
e
endomorphism ι(Π1 ) = ι(δΠ) of P by
e y) = ψ(ι(Π
θ(x, e 1 )x, y).

This gives the desired OD -module structure on Pe and therefore on Ye . 

The analogue of Corollary 5.2.6 follows as before from the properties of the Drinfeld moduli
space.
Corollary 5.3.4. The formal scheme N is π-adic and has semi-stable reduction. The special
fiber N ⊗OF̆ κF of N is a reduced scheme. 
We next prove the uniqueness of the framing object, cf. (i) of subsection 2.5. We start with
the following statement.
Proposition 5.3.5. Let k be an algebraically closed field which contains κF . Let M be a
WOF (k)-Dieudonné module of height 4 and dimension 2. Let ι be a homomorphisms of OF -
algebras
ι : OF ′ −→ End M.
Assume that M/V M is a free κF ′ ⊗κF k-module of rank 1. Let θ be a relative polarization on
M which satisfies
θ(ι(a)x1 , x2 ) = θ(x1 , ι(ā)x2 ), a ∈ OF ′ ,
ordπ det θ = 2.
Then the action ι extends to an action ι : OD −→ End M such that θ satisfies (5.3.1). In
particular M is isoclinic of slope 1/2. Furthermore, inv(M, ι, θ) = −1 (see Definition 8.3.1 for
this invariant).
If (M ′ , ι′ , θ′ ) is a second triple with the same properties, then there exists a quasi-isogeny
(M, ι) −→ (M ′ , ι′ ) of height 0 which respects the polarizations θ and θ′ up to a factor in OF× .
Proof. Let ψ be the principal relative polarization on M which exists by Proposition 5.1.1. We
define an endomorphism ρ : M −→ M by the equation
θ(x, y) = ψ(x, ρ(y)), x, y ∈ M.
One checks that ρ is an endomorphism of the Dieudonné module M such that
ρ(ι(a)x) = ι(ā)ρ(x), a ∈ OF ′ . (5.3.8)
As in the proof of Proposition 5.1.1, we choose an embedding λ : OF ′ −→ WOF (k) and obtain a
decomposition M = M0 ⊕ M1 . We note that
M1 = {x ∈ M | ι(a)x = λ(ā)x}.
It follows from (5.3.8) that ρ(M0 ) ⊂ M1 and ρ(M1 ) ⊂ M0 . We obtain a commutative diagram
ρ
M0 / M1

V V
 ρ 
M1 / M0 .
By our assumption on M/V M , the cokernels of both vertical maps have WOF (k)-length 1.
Therefore the cokernels of the horizontal maps have also the same length. This length must be
1 because ordπ det(ρ|M ) = ordπ det θ = 2.
We have
θ(ρ(x), y) = θ(x, ρ(y)),
94 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

because both sides are equal to ψ(ρ(x), ρ(y)). We consider the form
ψ1 (x, y) := ψ(ρ(x), ρ(y)).
This relative polarization satisfies the assumptions of the last part of Proposition 5.1.1. Therefore
there exists c ∈ OF ′ such that ψ1 (x, y) = ψ(ι(c)x, y). We find
ψ(ι(c)x, y) = θ(ρ(x), y) = −θ(y, ρ(x)) = −ψ(y, ρ2 (x)) = ψ(ρ2 (x), y).
This shows that
ρ2 = ι(c).
Since ρ commutes with the left hand side it commutes with ι(c). Comparing this with (5.3.8),
we obtain c ∈ OF . Since ρ2 has height 4 we obtain ordπ c = 1.
For α ∈ OF×′ we consider the endomorphism ρα (x) = ι(α)ρ(x) of M . We obtain ρ2α = ι(αᾱ)ρ2 .
Since each unit of F is a norm in the unramified extension F ′ /F we can arrange that ρ2α = −π.
We set Π = ρα . Then we obtain an action of OD = OF ′ [Π] on the Dieudonné module M . Since
M0 /ΠM1 and M1 /ΠM0 have length 1, we have obtained a special formal OD -module. The
equations (5.3.1) are satisfied for θ. Therefore θ is up to a factor in OF× uniquely determined by
the OD -action. This implies the first and the last assertion of the Proposition.
Because the invariant depends only on the isogeny class, it is enough to compute it for a
special formal OD -module with two critical indices and the canonical form θ from (5.3.1).
We use the isomorphism
OF ′ ⊗OF WOF (k) −→ WOF (k) × WOF (k), (5.3.9)
which maps a ⊗ ξ to (λ(a)ξ, λ(ā)ξ). Let σ = F be the Frobenius automorphism of WOF (k). It
acts on the left hand side of (5.3.9) via the second factor. This induces on the right hand side
the action σ : (ξ1 , ξ2 ) 7→ (σ(ξ2 ), σ(ξ1 )).
We set Ni = Mi ⊗ Q. This is a WOF (k)Q -vector space of dimension 2. In the decomposition
N = M ⊗Q = N0 ⊗N1 , the summands are isotropic with respect to θ. We consider the invariants
V −1 ι(Π)
Ui = Ni .
(In the notation of the proof of Proposition 5.1.5, this is Ui ⊗ Q.) The Ui are F -vector spaces of
dimension 2. Let
∧2 θ : ∧2 N0 × ∧2 N1 −→ WOF (k)Q
be the bilinear form defined by
 
′ ′ θ(n0 , n1 ) θ(n0 , n′1 )
θ(n0 ∧ n0 , n1 ∧ n1 ) = det ,
θ(n′0 , n1 ) θ(n′0 , n′1 )
for n0 , n′0 ∈ N0 , n1 , n′1 ∈ N1 . In the same way we can define ∧2 θ : ∧2 N1 × ∧2 N0 −→ WOF (k)Q .
Then we obtain ∧2 θ(x0 , x1 ) = ∧2 θ(x1 , x0 ) for x0 ∈ ∧2 N0 , x1 ∈ ∧2 N1 . From θ we pass to κ, cf.
(8.1.2) (there our F ′ is called K),
κ : N × N −→ F ′ ⊗F WO (k)Q ∼ = WO (k)Q × WO (k)Q .
F F F

Explicitly we have
κ(n0 + n1 , n′0 + n′1 ) = (θ(n0 , n′1 ), θ(n1 , n′0 )) ∈ WOF (k)Q × WOF (k)Q .
We take ∧2 κ on the F ′ ⊗F WOF (k)Q -module
2
^
N∼
= ∧2 N0 ⊕ ∧2 N1 .
F ′⊗ F WOF (k)Q

From the expression for κ we obtain

∧2 κ(x0 + x1 , x′0 + x′1 ) = (∧2 θ(x0 , x′1 ), ∧2 θ(x1 , x′0 )) ∈ WOF (k)Q × WOF (k)Q (5.3.10)
The restriction of θ to U0 × U1 induces a nondegenerate F -bilinear form
θ : U0 × U1 −→ δF ⊂ F ′ ⊂ WOF (k)Q , (5.3.11)
where δ was defined after (5.3.1). Indeed, for u0 ∈ U0 and u1 ∈ U1 we have by definition
θ(V u0 , V u1 ) = θ(ι(Π)u0 , ι(Π)u1 ) = θ(ι(Π)2 u0 , u1 ) = −πθ(u0 , u1 ).
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 95

Because θ is a polarization, we have on the other hand

θ(V u0 , V u1 ) = πσ −1 θ(u0 , u1 ) .
Therefore θ(u0 , u1 ) is anti-invariant by σ and (5.3.11) is proved.
We choose a nonzero element u0 ∈ U0 . Then we find u1 ∈ U1 such that θ(u0 , u1 ) = δ. We
remark that θ(ι(Π)n, n) = 0, for an arbitrary n ∈ N . This is clear from the third equation of
(5.3.1) because θ is alternating. Since θ(u0 , ι(Π)u0 ) = 0, the vectors u1 , ι(Π)u0 ∈ U1 are linearly
independent. We set
x = x0 + x1 := u0 ∧ ι(Π)u1 + u1 ∧ ι(Π)u0 ∈ ∧2F ′ ⊗F WO (k)Q N.
F

It satisfies ∧2 V x = πx. Indeed,

∧2 V (u0 ∧ι(Π)u1 ) = (V u0 ∧ι(Π)V u1 ) = (ι(Π)u0 ∧ι(Π)2 u1 ) = (ι(Π)u0 ∧(−π)u1 ) = π(u1 ∧ι(Π)u0 ).
The similiar equation holds for the second summand in the definition of x. By Definition 8.3.1
we obtain
2
inv(M, ι, θ) = (−1)ordπ ∧ κ(x,x) .
By (5.3.10) we have ordπ ∧2 κ(x, x) = ordπ ∧2 θ(x0 , x1 ). We compute
   
θ(u0 , u1 ) θ(u0 , ι(Π)u0 ) δ 0
∧2 θ(x0 , x1 ) = det = det = πδ 2 .
θ(ι(Π)u1 , u1 ) θ(ι(Π)u1 , ι(Π)u0 ) 0 δπ
This shows ordπ ∧2 θ(x0 , x1 ) = 1 and therefore inv(M, ι, θ) = −1. 

We can now prove the uniqueness of the framing object.

Proposition 5.3.6. Let r be special and K/F unramified. Let k be an algebraically closed field
in NilpOE . Let (P, ι, β) ∈ Ppol r,k be such that ordp detW (k) β = 2f , cf. Definition 4.1.2. Then P
is isoclinic of slope 1/2 and invr (P, ι, β) = −1.
If (P1 , ι1 , β1 ) is a second triple with the same properties, then there exists a quasi-isogeny of
height zero,
ρ : (P, ι, β) −→ (P1 , ι1 , β1 ),
such that there is an f ∈ OF× with
β1 (ρ(x), ρ(y)) = β(f x, y), x, y ∈ P.

Proof. We apply the functor Cpol r,k to (P, ι, β) and obtain (Pc , ιc , βc ), cf. (4.4.14). By Theorem
4.4.11 we find ordπ detWOF (k) βc = 2. Therefore we can apply Proposition 5.3.5 to (Pc , ιc , βc ).
We obtain that Pc is isoclinic of slope 1/2 and inv(Pc , ιc , βc ) = −1. By Corollary 4.3.3 and
Proposition 3.3.17, we find that P is isoclinic of slope 1/2, and by Proposition 8.3.2 we obtain
invr (P, ι, β) = −1.
By Proposition 5.3.5, there is a quasi-isogeny of height 0 between (Pc , ιc , βc ) and (P1,c , ι1,c , β1,c ).
It induces by Theorem 4.4.11 a quasi-isogeny of height zero as claimed in the Proposition. 

We end this section by justifying the footnote in Definition 2.6.1. Let S ∈ NilpOF̆ and let
(Y, ι, θ, ρ) ∈ N (S). Since there is an OD -module structure on Y such that θ is of the form (5.3.7),
the kernel of θ : Y −→ Y ∧ , considered as morphism to the dual relative to OF , is annihilated
by π. More generally we prove:
Proposition 5.3.7. Let K/F be an unramified quadratic field extension. Let R be an OK -
algebra. Let P and P ′ be WOF (R)-displays of height 4 with an action ι : OK −→ End P, resp.
ι′ : OK −→ End P ′ . Assume that Lie P, resp. Lie P ′ , is locally on Spec R a free OK ⊗OF R-
module of rank 2. Let α : P −→ P ′ be an isogeny of OF -height 2. Then there exists locally on
Spec R an isogeny β : P ′ −→ P such that
β ◦ α = πidP , α ◦ β = πidP ′ .
Let P and P ′ be the displays of formal p-divisible groups X and X ′ with an OK -action. Then
the kernel of any isogeny α : X −→ X ′ of height 2 is annihilated by π.
96 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Proof. The proof is a variant of the proof of Proposition 1.6.4 in [35]. We will use notation from
that proof. The OK -algebra structure on R induces a natural homomorphism OK −→ WOF (R)
which is equivariant with respect to the Frobenius τ ∈ Gal(K/F ) and the Frobenius on WOF (R).
The composition with τ gives a second homomorphism OK −→ WOF (R). We denote by Ψ the
set of these two homomorphism. We write ψ̄ = ψ ◦ τ for ψ ∈ Ψ.
The OK -action gives the usual decompositions,
P = ⊕ψ∈Ψ Pψ , P ′ = ⊕ψ∈Ψ Pψ′ .
We have the same kind of decompositions for Q ⊂ P and Q′ ⊂ P ′ . We choose normal decompo-
sitions
Pψ = Tψ ⊕ Lψ , Pψ′ = Tψ′ ⊕ L′ψ .
The T and L on the right hand sides are by assumption locally free of rank 1. Using these
decompositions, we write αψ : Pψ̄ −→ Pψ′ in matrix form
 
Xψ V Yψ
Mψ = . (5.3.12)
Uψ Zψ

The maps Ḟψ : IOF (R)Tψ ⊕ Lψ −→ Tψτ ⊕ Lψτ , resp. Ḟψ′ : IOF (R)Tψ′ ⊕ L′ψ −→ Tψτ
′
⊕ L′ψτ , are
given by invertible matrices
   ′ 
Aψ Bψ ′ Aψ Bψ′
Φψ = , Φψ = .
Cψ Dψ Cψ′ Dψ ′

The matrices (5.3.12) define a morphism α of displays iff

Mψτ Φψ = Φ′ψ s Mψ , for ψ ∈ Ψ. (5.3.13)
s
We will argue as in [35]. The meaning of the upper left indes is the same as there. Taking
the determinants we obtain
det Mψτ det Φψ = det Φ′ψ det s Mψ . (5.3.14)
2
F
In particular det Mψ and det Mψ differ by a unit in WOF (R). As in [35] we obtain that
det Mψ = π h ǫψ ,
for some units ǫψ ∈ WOF (R). By (5.3.14), h is independent of ψ. Since α is an isogeny of height
2 we conclude that h = 1. Now we pass to the adjucate matrices
ad
Φψ ad
Mψτ = ad
( s Mψ ) ad
Φ′ψ .
Since the matrices Φ are invertible, we conclude
(det Φψ ) ad
Mψτ Φ′ψ = (det Φ′ψ )Φψ ad
( s Mψ ). (5.3.15)
We consider first the case where R is reduced. Then π is not a zero divisor in WOF (R). In this
case, we conclude from (5.3.14)
det Φψ ǫψτ = det Φ′ψ F
ǫψ .
Thus we may rewrite equation (5.3.15) as
−1 F −1 ad s
ǫψτ ad
Mψτ Φ′ψ = Φψ ǫψ ( Mψ ).

This shows that the matrices ǫ−1

ψ
ad
Mψ define the desired morphism β : P ′ −→ P. In the case
where R is not reduced we may argue as in the proof of Proposition 1.6.4 in [35]. 

6. Moduli spaces of formal local CM-triples

In this section we prove Theorems 2.6.2 and 2.6.3. Recall that d = [F : Qp ], and that we
write d = ef .
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 97

6.1. The case r special and K/F ramified. Let (Y, ιY , λY ) be a special formal OD -module
over κ̄F with a Drinfeld polarization λY , cf. Definition 5.1.4 and section 5.2. We denote by F̆
resp. Ĕ, the completions of the maximal unramified extension of F , resp. of the reflex field
E. Their residue class fields are identified with κ̄F resp. κ̄E . We note that in the ramified
case E = E ′ . We extend the embedding ϕ0 : OF −→ OE to an embedding ϕ̆0 : OF̆ −→ OĔ .
The base change (Y, ιY , λY )κ̄E is the base change via ϕ̆0 . It is an object of dRκ̄nilp,pol
E
. We
refer to Theorem 4.4.11 for the latter notation. It is therefore isomorphic to the image of an
object (X, ιX , λX ) ∈ dPss,pol
r,R . The polarization λX is again principal. We consider the functor
Mr = MK/F,r of Definition 2.6.1 in the ramified case (where h = 0). By Proposition 4.2.9, this
Definition may be reformulated as follows.
Proposition 6.1.1. Let S be a scheme over Spf OĔ . A point of Mr (S) consists of
(1) a local CM-pair (X, ι) of CM-type r over S which satisfies the Eisenstein conditions (ECr )
relative to a fixed uniformizer Π of K and such that
Tr(ι(Π) | EAψ0 Lieψ0 X) = 0. (6.1.1)
(2) an isomorphism of p-divisible OK -modules
∼
λ : X −→ X ∧ ,
which is a polarization of X.
(3) a quasi-isogeny of height zero of p-divisible OK -modules
ρ : X ×S S̄ −→ X ×Spec κ̄E S̄,
such that the pullback quasi-isogeny ρ∗ (λX ) differs from λ|X×S S by a scalar in OF× , locally on S.
Here S = S ×Spec OĔ Spec κ̄E .
In Definition 2.6.1 we required that the scalar is in F × but because the polarizations λ and λX
are principal and ρ has height 0 the scalar is automatically in OF× . We remark that the condition
(6.1.1) depends only on the restriction of the structure morphism S −→ Spf OE .
We define for i ∈ Z the functor Mr (i) on the category of schemes S −→ Spf OĔ by replacing
(3) in Proposition 6.1.1 by
(3′ ) a quasi-isogeny of p-divisible OK -modules
ρ : X ×S S̄ −→ X ×Spec κ̄E S̄,
such that the pullback quasi-isogeny ρ∗ (λX ) differs from pi λ|X×S S̄ by a scalar in OF× , locally
on S̄.
It follows from the last condition that
2 height ρ = height(pi | X) = 4di. (6.1.2)
We have an isomorphism of functors
Mr −→ Mr (i),
which associates to a point (X, ι, λ, ρ) ∈ Mr (S) the point (X, ι, λ, Πei ρ) ∈ Mr (i)(S). We set
a
M̃r = Mr (i).
i∈Z

We define a Weil descent datum on M̃r relative to OĔ /OE . Let τE be the Frobenius of Ĕ/E.
It is enough to consider the functor M̃r for affine schemes S = Spec R. We write ε : OĔ −→ R
for the given algebra structure. We write R[τE ] for the ring R with the new algebra structure
ε ◦ τE . By base change to κ̄E , we obtain
ε̄ : κ̄E −→ R̄ := R ⊗OĔ κ̄E .
We consider a point (X, ι, λ, ρ) ∈ Mr (i)(R), where ρ is a quasi-isogeny
ρ : XR̄ −→ ε̄∗ X.
98 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Since the notion of a CM-triple depends only on the induced OE -algebra structure on R, we
may regard (X, ι, λ) as a CM-triple on R[τE ] . We set
ρ FX,τ
ρ̃ : XR̄ −→ ε̄∗ X −→E ε̄∗ (τE )∗ X.
The assignment (X, ι, λ, ρ) 7→ (X, ι, λ, ρ̃) defines a morphism
ωMr : Mr (i)(R) −→ Mr (i + fE )(R[τE ] ) (6.1.3)
where fE = [κE : Fp ]. Here we note that the inverse image of the polarization (τE )∗ λX on (τE )∗ X
by FX,τE : X −→ (τE )∗ X is pfE λX . From (6.1.3) we obtain the Weil descent datum

ωMr : M̃r −→ M̃r(τE ) , (6.1.4)

where the upper index (τE ) denotes the base change via Spec τE : Spf OĔ −→ Spf OĔ .
Let N (i) be the functor of Definition 5.2.1. Note that we took Y for the framing object
appearing in the definition of N (i). We consider a point (X, ι, λ, ρ) ∈ Mr (i)(R), where R ∈
NilpOĔ . Applying the contracting functor Cpolr,R , we obtain a quadruple (Xc , ιc , λXc , ρc ). It
follows from the isomorphism (4.3.11) (which also holds in the ramified case, cf. a few lines
below (4.3.11)) that the condition (6.1.1) implies
Tr(ιc (Π) | Lie Xc ) = 0.
By functoriality, the polarizations ρ∗c λY and pi λ(Xc )R̄ differ by a unit in OF . Hence (Xc , ιc , λXc , ρc )
defines a point of N (i). Therefore we obtain from Theorem 4.4.11 an isomorphism of functors,
∼
Mr (i) −→ N (ei) ×Spf OF̆ Spf OĔ . (6.1.5)
The base change on the right hand side is via ϕ̆0 : OF̆ −→ OĔ
We set
a
Ñ [e] = N (ei). (6.1.6)
i∈Z

We endow Ñ [e] with a Weil descent datum relative to OĔ /OE . Let R ∈ NilpOĔ . We consider
the map
f /f
Π(d−1)fE /f ωNE : N (ei)(R) −→ N (e(i + fE )(R[τE ] ).
Here on the left hand side appears the iterate of the Weil descent datum ωN : N (i) −→ N (i+1)(τ )
of Ñ relative to OF̆ /OF from (5.2.5) and the translation functor Π : N (i) −→ N (i + 1), cf.
(5.2.4). This defines a Weil descent datum relative to OĔ /OE ,
f /f
Π(d−1)fE /f ωNE : Ñ [e] −→ Ñ [e](τE ) . (6.1.7)
We define
J′ = {α ∈ AutoK X | α∗ λX = uλX , for u ∈ pZ OF× }
(6.1.8)
= {α ∈ AutoK Y | α∗ λY = uλY , for u ∈ pZ OF× }.
The last equation holds because of the contraction functor. This group acts via the rigidifications
ρ on the functors M̃r and Ñ [e]. By the last equation of (6.1.8), we may regard J ′ as a subgroup
of J ∗r of Lemma 5.2.2.
Proposition 6.1.2. There is an isomorphism of formal schemes over Spf OĔ

M̃r −→ Ñ [e] ×Spf OF̆ Spf OĔ ,

where the right hand side denotes the base change via ϕ̆0 : OF̆ −→ OĔ . This isomorphism is
compatible with the action of J ′ on both sides.
Under the isomorphism the Weil descent datum (6.1.4) on the left hand side corresponds to
the Weil descent datum
f /f
Π(d−1)fE /f ωNE : Ñ [e] −→ Ñ [e](τE )
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 99

on the right hand side. More explicitly, for any i ∈ Z, there is a commutative diagram
∼
Mr (i) / N (ei)O
Ĕ

ωMr f /f
Π(d−1)fE /f ωNE (6.1.9)
 
∼ / N (e(i + fE ))(τE ) .
Mr (i + fE )(τE ) O Ĕ

Proof. The isomorphism of formal schemes over Spf OĔ comes from (6.1.5). It remains to show
that the diagram is commutative. Let R ∈ NilpOĔ with structure morphism ε : OĔ −→ R.
Let ε̄ : κ̄E −→ R̄ = R ⊗OĔ κ̄E be the induced morphism. We start with a point (X, ι, λ, ρ) ∈
Mr (i)(R). If we apply ωMr , we change ρ to
FX,τ
ρ̃ : XR̄ −→ ε̄∗ X −→E ε̄∗ (τE )∗ X.
The lower horizontal arrow in (6.1.9) applies to ρ̃ the contracting functor Cr,R , cf. Definition
4.4.2. We have Xc = Y. Let PX be the W(κĒ )-Dieudonné module of X. Then the WOF (κĒ )-
Dieudonné module of Y is by definition the degree-zero component of PX′ defined by (4.3.21),
comp. Remark 4.4.4. From this definition we obtain
(V ′ )f = Π1−d V f .
In terms of Dieudonné modules, FX,τE is given by

PX −→ W (κĒ ) ⊗W (τE ),W (κĒ ) PX , x 7−→ 1 ⊗ V fE x.

In terms of the relative Dieudonné module, FY,τE is given by (V ′ )fE . Therefore the contracting
functor applied to FX,τE gives

ιY (Π(d−1)fE /f )FY,τE : Y −→ (τE )∗ Y,

since (V ′ )fE = Π(1−d)fE /f V fE . On the other hand ωN just multiplies ρc by FY,τE . Therefore
we obtain the commutativity of the diagram. 

Corollary 6.1.3. Let ωτE denote the action of τE on the formal scheme scheme Ω b F ×Spf OF
Spf OĔ via the second factor. There exists an isomorphism of formal schemes over Spf OĔ
∼
b F ×Spf OF Spf O ) × Z,
M̃r −→ (Ω Ĕ

such that the Weil descent datum ωMr induces on the right hand side the Weil descent datum
(ξ, i) 7−→ (ωτE (ξ), i + fE ).

In particular the formal scheme MK/F,r over Spf OĔ is a p-adic formal scheme which has semi-
stable reduction; hence it is also flat, with reduced special fiber.

Proof. We consider the isomorphism M̃Dr −→ Ñ of formal schemes over Spf OF̆ from Theorem
5.2.3. It is compatible with translations and the Weil descent data on both sides. Combining
this with the Drinfeld isomorphism (5.1.22), we obtain an isomorphism
b F ×Spf OF Spf O ) × Ze −→ Ñ [e]. ∼
(Ω F̆ (6.1.10)
f /f
We consider on the right hand side the Weil descent datum Π(d−1)fE /f ωNE which is a composite
of an iterate of the translation functor and an iterate of the Weil descent datum ωN . By (5.1.23)
f /f
we see that Π(d−1)fE /f ωNE induces on the left hand side of (6.1.10) the Weil descent datum
(ξ, ei) 7−→ (ωτE , e(i + fE )).
The assertion about descent data follows by forgetting e. The last assertion follows from Corol-
lary 5.2.6. 
100 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

6.2. The case r special and K/F unramified. Let ϕ0 , ϕ̄0 ∈ Φ be the special embeddings.
Their restrictions to F are the same. We extend the resulting embedding OF −→ OE to an
embedding OF̆ −→ OĔ . The two embeddings ϕ0 , ϕ̄0 then factor over the two OF -algebra
homomorphisms ϕ̆0 , ϕ̆¯0 : OK −→ OF̆ .
Let (Y, ιY ) be a special formal OD -module over κ̄F . We endow the WOF (κ̄F )-Dieudonné
module PY of Y with the polarization θ, cf. (5.3.1). (One should note that we call now K what
was F ′ in that section.) Then θ defines an isogeny to the Faltings dual,
λY : Y −→ Y∇ .
If we compose the action ι∇
Y of (OD )
opp
on Y∇ with the anti-involution † of (5.3.3), the isogeny
λY becomes an isogeny of special formal OD -modules. We indicate this by rewriting the isogeny
as
λY : Y −→ Y∆ . (6.2.1)
In section 5.3 we used (PY , θ) to define the functor N (i). Together with the restriction to
OK of the action of OD on PY , we obtain an object of dRpolκ̄F , cf. Definition 4.4.10. By Theorem
ss,pol
4.4.11, this object is the image of an object in dPr,κ̄E . The latter is the display of an object
(X, ιX , λX ) ∈ Ppol
r,κ̄E , cf. Definition 4.1.2. The height of the polarization λX is 2f , and the
associated Rosati involution induces on OK the conjugation over OF .
We consider the functor Mr = MK/F,r of Definition 2.6.1 with the framing object (X, ιX , λX )
as defined above. We can give an alternative description of that functor.
Proposition 6.2.1. Let S be a scheme over Spf OĔ . A point of Mr (S) consists of
(1) a local CM-pair (X, ι) of CM-type r over S which satisfies the Eisenstein conditions (ECr )
relative to the fixed uniformizer π of F .
(2) an isogeny of height 2f of p-divisible OK -modules
λ : X −→ X ∧ ,
which is a polarization of X.
(3) a quasi-isogeny of p-divisible OK -modules
ρ : X ×Spec R Spec R̄ −→ X ×Spec κ̄E Spec R̄,
such that the pullback quasi-isogeny ρ∗ (λX ) differs from λ|X×S Spec S̄ by a scalar in OF× , locally
on S̄, where S̄ = S ⊗Spf OĔ Spec κ̄E .
Proof. We may assume that S = Spec R. Let (X0 , X1 , λ, ρX ) ∈ Mr (R) be a point as in Definition
2.6.1. We obtain a point (X, λ, ρX ) as in the Proposition if we set X := X0 , keep ρX and redefine
λ to be the composite
λ
X0 −→ X1 −→ X0∧ .
We note that ρ is automatically of height zero because by the last condition ρ∗ (λX ) and λX have
the same height 2f .
Conversely, assume (X, λ, ρX ) is as in the Proposition. Then we set X0 = X and X1 = X ∧ .
By Corollary 4.2.8, X1 satisfies the condition (ECr ) and, by Proposition 4.2.6, X0 and X1 satisfy
(KCr ) . The polarization λ defines an isogeny a : X0 −→ X1 of p-divisible OK -modules which
has height 2f . To the morphism induced by a on the displays we apply the contracting functor
of Definition 4.4.2. We obtain an isogeny of WOF (R)-displays α : P0 −→ P1 of height 2. To
this isogeny we may apply Proposition 5.3.7. We find an isogeny β : P1 −→ P0 such that
β ◦ α = πidP0 .
The existence of ρ guarantees that X0 , X1 and a are defined over an OĔ -subalgebra R′ ⊂ R
which is of finite type over OĔ . Therefore we may apply Theorem 4.4.3. It gives us a morphism
b : X1 −→ X0 such that b ◦ a = πidX0 . We see that X0 , X1 , ρ together with the defining
∼
isomorphism X1 −→ X0 defines a point of the functor Mr of Definition 2.6.1. 

Definition 6.2.2. We define the functor Mr (i) in the same way as in Definition 6.2.1, but we
replace the data (4) by the following
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 101

(4′ ) A quasi-isogeny of p-divisible OK -modules

ρ : X ×S S̄ −→ X ×Spec κ̄E S̄,
such that the pullback quasi-isogeny ρ∗ (λX ) differs from pi λ|X×S S̄ by a scalar in OF× , locally
on S̄.
As in the ramified case (6.1.2) we conclude that height ρ = 2di. We set
a
M̃r = Mr (i).
i∈Z

Let R ∈ NilpOĔ . Exactly as in the ramified case we obtain a morphism

ωMr : Mr (i)(R) −→ Mr (i + fE )(R[τE ] ), (6.2.2)
where, as in the ramified case, fE = [κE : Fp ]. With the notation used in the ramified case, it
changes the datum ρ in point (4′ ) of Definition 6.2.2 to
ρ FX,τ
ρ̃ : (X0 )R̄ −→ ε̄∗ X −→E ε̄∗ (τE )∗ X.
From (6.2.2) we obtain the Weil descent datum,
ωMr : M̃r −→ M̃r(τE ) . (6.2.3)
We define an isomorphism of functors on NilpOĔ ,
∼
Mr (i) −→ N (ei). (6.2.4)
Let (X, ι, λ, ρ) ∈ Mr (i)(R). Applying the contracting functor Cpol
r,R ,
we obtain a quadruple
(Xc , ιc , λXc , ρc ), where ρc : Xc ⊗R , R̄ −→ Y ⊗κ̄E R̄. This gives a point of N (i)(R). The functor
Cpol
r,R is an equivalence of categories if the ideal of nilpotent elements of R is nilpotent, cf.
Theorem 4.4.11.Therefore we may reverse the construction of (6.2.4). Therefore Mr (i)(R) −→
N (ei)(R) is bijective for those R. For a general R we obtain the bijectivity as in the proof of
∼
Proposition 6.2.1. With the notation (6.1.6) we have a bijection M̃r (R) − → Ñ [e](R).
We define
J ′ = {α ∈ AutoK X | α∗ λX = uλX , for u ∈ pZ OF× }
(6.2.5)
= {α ∈ AutoK Y | α∗ λY = uλY , for u ∈ pZ OF× }.
The last equation holds because of the contraction functor. This group acts via the rigidifications
ρ on the functors M̃r and Ñ [e]. By the last equation of (6.2.5) we may regard J ′ as a subgroup
of J ∗ur of Lemma 5.3.2.
Proposition 6.2.3. There exists an isomorphism of formal schemes Spf OĔ
∼
M̃r −→ Ñ [e] ×Spf OF̆ Spf OĔ ,
where the right hand side denotes the base change via ϕ̆0 : OF̆ −→ OĔ . This isomorphism is
compatible with the action of J ′ on both sides, and the Weil descent datum (6.1.4) on the left
hand side corresponds to the Weil descent datum
f /f
π g ωNE : Ñ [e] −→ Ñ [e](τE ) .
on the right hand side. Here g = fE (d − 1)/2f . More explicitly, there is a commutative diagram
∼
Mr (i) / NO (ei)
Ĕ

ωMr f /f
π g ωNE (6.2.6)
 
∼
Mr (i + fE )(τE ) / NO (e(i + fE ))(τE ) .
Ĕ

The multiplication by π is the morphism N (j) → N (j + 2) which is obtained by multiplying ρ in

Definition 5.3.1 by π. Equivalently one can apply two times the translation functor (5.3.5).
Proof. We have already proved the isomorphism of formal schemes over Spf OĔ . The compati-
bility with the Weil descent datum follows from the following lemma. 
102 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Lemma 6.2.4. Let X be the framing object over κ̄E , with corresponding Frobenius morphism
FX,κE : X −→ (τE )∗ X. Let Y be its image under the contracting functor Cpol
r,κ̄E . Then the image
of FX,κE under Cpol
r,κ̄E is given as
π g FY,κE : Y −→ (τE )∗ Y.
Proof. Let M = PX be the W(κ̄E )-Dieudonné module of X. Then the WOF (κĒ )-Dieudonné
module M ′ of Y is by definition the ψ0 -component of PX′ defined by (4.3.18). In terms of
Dieudonné modules, FX,τE is given by
M −→ W (κĒ ) ⊗τE ,W (κĒ ) M, x 7−→ 1 ⊗ V fE x.
This induces
V fE : M ′ −→ W (κ̄E ) ⊗τE ,W (κ̄E ) M ′ . (6.2.7)
We consider the decomposition
M = ⊕Mψ .
We denote by σ the Frobenius on the Witt ring W (κ̄E ) and also the Frobenius of K t /Qp . We
note that
W (κ̄E ) ⊗τE ,W (κ̄E ) Mψ = (W (κ̄E ) ⊗τE ,W (κ̄E ) M )τE ψ = (W (κ̄E ) ⊗τE ,W (κ̄E ) M )ψσfE .
In terms of the relative Dieudonné module, FY,τE is given by (V ′ )fE . Our problem is to express
(6.2.7) in terms of V ′ . By definition we have
V ′ = π −aψ V : Mψσ −→ Mψ .
We consider
V V V
V fE : Mψ −→ Mψσ−1 −→ . . . −→ Mψσ−fE .
Therefore we obtain for this map
aψσ−fE
V fE = π aψσ−1 · . . . · π (V ′ )fE .
By definition of the reflex field E, we have aψ = aψσ−fE . This implies that the number
g = aψσ−1 + . . . + aψσ−fE
is independent of ψ. We add to the last equation
g = aψ̄σ−1 + . . . + aψ̄σ−fE .
Since aψ + aψ̄ is e or e − 1 we obtain
fE
2g = · (d − 1).
f
This proves the Lemma. 
Corollary 6.2.5. Let ωτE denote the action of τE on the formal scheme scheme Ω b F ×Spf OF
fE /f
Spf OĔ via the second factor, i.e., in the notation of (5.1.23), ωτE = ωτ . There exists an
isomorphism of formal schemes over Spf OĔ
∼b F ×Spf OF Spf O ) × Z,
M̃r −→ (Ω Ĕ
such that the Weil descent datum ωMr induces on the right hand side the Weil descent datum
(ξ, i) 7−→ (ωτE (ξ), i + fE ).
In particular the formal scheme MK/F,r over Spf OĔ is a p-adic formal scheme which has semi-
stable reduction; hence it is also flat, with reduced special fiber.
Proof. We consider the isomorphisms of functors
∼ ∼ ∼ b F ×Spf OF Spf O ) × Ze.
M̃r −→ ÑOĔ [e] ←− (M̃Dr [e])OĔ −→ (Ω Ĕ (6.2.8)
The last arrow is the isomorphism (5.1.22) and the left arrow in the middle is the isomorphism
f /f
of Theorem 5.3.3. We must see what the Weil descent data π g ωNE on ÑOĔ [e] does on the last
E f /f
functor. By Theorem 5.3.3, it induces on (M̃Dr [e])OĔ the Weil descent datum ωM Dr
multiplied
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 103

2g-times with the translation (cf. last sentence of Proposition 6.2.3). By (5.1.23), the induced
Weil descent datum on the last functor of (6.2.8) is
(ξ, ie) 7−→ (ωτE , ie + (fE /f ) + 2g).
But we have
ie + (fE /f ) + 2g = ie + (fE /f ) + (d − 1)(fE /f ) = ie + fE e.
This proves the Corollary. 

6.3. The case r banal and K/F ramified. Let ε ∈ {±1}. There is up to isomorphism a unique
anti-hermitian K-vector space (V, ψ) of dimension 2 such that inv(V, ψ) = ε, cf. Definition 8.1.1.
Let Λ ⊂ V be an OK -lattice such that ψ induces a perfect pairing on Λ, cf. Lemma 8.1.3. By
Theorem 4.5.11, (Λ, ψ) corresponds to a display (P, ι, β) ∈ dPpolr,κ̄E over the residue class field of
OĔ which is unique up to isomorphism. Let (X, ιX , βX ) be the corresponding polarized p-divisible
OK -module of type r. It is uniquely determined by the conditions that β is principal and that
invr (X, ιX , βX ) = ε. Then (CX ⊗ Q, φ) ≃ (V, ψ).
Definition 6.3.1. We define a functor Mr,ε (i) on the category NilpOĔ . For R ∈ NilpOĔ , a
point of Mr,ε (i)(R) is given by the following data:
(1) a local CM-pair (X, ι) of CM-type r over R which satisfies the Eisenstein conditions (ECr )
relative to a fixed uniformizer Π ∈ K.
(2) an isomorphism of p-divisible OK -modules
λ : X −→ X ∧ ,
which is a polarization of X.
(3) a quasi-isogeny of p-divisible OK -modules
ρ : XR̄ −→ X ×Spec κ̄E Spec R̄,
such that the pullback quasi-isogeny ρ∗ (λX ) differs from pi λ|X×S S̄ by a scalar in OF× , locally
on Spec R̄. Here R̄ = R ⊗OĔ κ̄E .
Two data (X, ι, λ, ρ) and (X1 , ι1 , λ1 , ρ1 ) define the same point iff there is an isomorphism of
OK -modules α : X → X1 such that ρ1 ◦ αR̄ = ρ. (This implies that α respects the polarizations
up to a factor in OF× ).
By Proposition 4.2.2, Mr,ε (0) is the functor MK/F,r,ε used in Theorem 2.6.3. We will also
consider the functor
a
M̃r,ε = Mr,ε (i).
i∈Z

Let i ∈ Z. We consider the following functor Gε (i) on the category NilpOĔ . A point of
Gε (i)(R) is given by the following data:
(1) a locally constant p-adic étale sheaf C on Spec R which is Zp -flat with rankZp C = 4d and
with an action
ι : OK −→ EndZp C.
(2) a perfect alternating OF -bilinear pairing
φ : C × C −→ OF ,
such that φ(ι(a)c1 , c2 ) = φ(c1 , ι(ā)c2 ) for c1 , c2 ∈ C and a ∈ OK .
(3) a quasi-isogeny of OK -module sheaves on Spec R
ρ : (C, ι) −→ (C X , ι) (6.3.1)
such that locally on Spec R there is an f ∈ OF× with
pi f φ(c1 , c2 ) = φX (ρ(c1 ), ρ(c2 )).
104 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

∼
Another set of data (C ′ , ι′ , λ′ , ρ′ ) defines the same point iff there is an isomorphism α : C −
→ C′
′ ′ ×
such that ρ ◦ α = ρ. Then α respects φ and φ up to a factor in OF .
We remark that in (6.3.1) we regard C X as the constant sheaf on Spec R. The existence of
the quasi-isogeny implies that C is locally constant for the Zariski topology. Therefore locally
on Spec R the sheaf C is the constant sheaf associated to an OK -submodule C ⊂ CX ⊗Zp Qp and
ρ is given by the last inclusion.
The polarized contraction functor Cpol r defines a morphism of functors
Mr,ε (i) −→ Gε (i). (6.3.2)
To describe the functor Gε (i), we may restrict to the case where the sheaf C is given by
an OK -submodule of CX ⊗Zp Qp . Then C defines a point of Gε (i)(R) iff (1/pi )φX is a perfect
alternating pairing on C. We define an algebraic group over Zp , and its Zp -rational points,
J ′ (Zp ) = {g ∈ GLOK (CX ) | φX (gc1 , gc2 ) = f · φX (c1 , c2 ) for some f ∈ OF× }.
By Lemma 8.1.3, there is an isomorphism g : (CX , φX ) −→ (C, p1i φX ). This means that gCX = C
and
φX (gc1 , gc2 ) = pi φX (c1 , c2 ), c1 , c2 ∈ CX ⊗Zp Qp . (6.3.3)
We define
J ′ (i) = {g ∈ GLK (CX ⊗Zp Qp ) | φX (gc1 , gc2 ) = pi f φX (c1 , c2 ), for some f ∈ OF× }.
This construction gives us a functor isomorphism
∼
Gε (i) −→ J ′ (i)/J ′ (Zp ),
where the right hand side is considered as the restriction of the constant sheaf to NilpOĔ . Let
J ′ ⊂ GL(CX ) be the union of the J ′ (i). Using the contraction functor we may write
J ′ = {α ∈ AutoK X | α∗ λX = µ(α)λX for some µ(α) ∈ pZ OF× }. (6.3.4)
Therefore J ′ acts via ρ on the functor M̃r,ε .
Proposition 6.3.2. The morphism of functors on NilpOĔ obtained from (6.3.2) is a J ′ -equivariant
isomorphism,
∼
M̃r,ε −→ J ′ /J ′ (Zp ). (6.3.5)
The left hand side is endowed with the Weil descent datum ωMr relative to OĔ /OE which is
defined exactly in the same way as (6.1.4). This Weil descent datum corresponds on the right
hand side to the Weil descent datum given by multiplication with ΠefE . Here we regard ΠefE as
an automorphism of the K-vector space CX ⊗Zp Qp .
Proof. We show that (6.3.2) is an isomorphism of functors. By the Grothendieck-Messing crite-
rion both are formally étale, hence we may restrict to the category of κ̄E -algebras. Both functors
commute with inductive limits of rings, i.e., they are locally of finite presentation. To see this,
we consider the special fiber Mr,ε,κ̄E (0). Let m ∈ N. We consider the subfunctor Um which
consists of points such that pm ρ−1 is an isogeny. Then X is the quotient of X ×Spec κ̄E Spec R by
a finite locally free subgroup scheme of X(4dm) ×Spec κ̄E Spec R. (We have denoted by X(4dm)
the kernel of the multiplication by p4dm .) This shows that Um is a scheme of finite presenta-
tion over Spec κ̄E . Therefore Um is locally of finite representation as a functor, cf. EGA IV,
Thm. (8.8.2). One easily deduces from this that Mr,ε,κ̄E (0) is locally of finite presentation as a
functor. In the same way we see that Gε (i) is locally of finite presentation. To show that (6.3.2)
is an isomorphism of functors we can therefore restrict to κ̄E -algebras R which are of finite type
over κ̄E . For such R, (6.3.2) induces a bijection by Theorem 4.5.11.
It remains to compare the Weil descent data on both sides of (6.3.5). It is enough to make
the comparison for the restriction of (6.3.5) to the category of κ̄E -algebras. Let ε : κ̄E −→ R
be a κ̄E -algebra. The restriction of the functor Mr,ε to κ̄E -algebras has a Weil descent datum
ωMr,ε ,Fp : Mr,ε (R) −→ Mr,ε (R[σ] ) over Fp , given by
ωMr,ε ,Fp ((X, ι, λ, ρ)) = (X, ι, λ, ρ[σ] ), (6.3.6)
where ρ[σ] is the composite
ρ ∗ Xε F
X −→ ε∗ X −→ ε∗ σ∗ X.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 105

Here σ denotes the Frobenius automorphism of κ̄E over Fp . To see that this makes sense, we
have to check that all p-divisible OK -modules above satisfy the rank condition (RCr ) and the
Eisenstein conditions (ECr ). The condition (ECr ) says that Πe Lie X = 0. This remains true
if we regard X as a p-divisible OK -module over R[σ] . For the condition (RCr ), the claim is
obvious.
Therefore it suffices to show that ωMr ,Fp induces on the right hand side of (6.3.5) the Weil
descent datum g 7→ Πe g. This follows from the following Lemma. 

Lemma 6.3.3. There is an identification CX = Cσ∗ X . The functor Cpol

r,κ̄E applied to the Frobenius
morphism FX : X −→ σ∗ X yields Πe : CX −→ CX .
Proof. The first assertion follows because the functor Cpol
r,κ̄E commutes with base change. Con-
sider the Dieudonné module PX of X. We have
CX = {c ∈ PX | V c = Πe c},
cf. Remark 4.5.13. The map
PX −→ W (κ̄E ) ⊗σ,W (κ̄E ) PX , c 7−→ 1 ⊗ c
defines the identification CX = Cσ∗ X . The Frobenius FX induces on the Dieudonné modules
V♯
PX −→ W (κ̄E ) ⊗σ,W (κ̄E ) PX , x 7−→ 1 ⊗ V x.
For c ∈ CX we obtain V ♯ c = 1 ⊗ V c = 1 ⊗ Πe c. 

We can reformulate a part of Proposition 6.3.2 as follows. We consider the algebraic group
over Zp such that
J(Zp ) = {g ∈ GLOK (CX ) | φX (gc1 , gc2 ) = u · φX (c1 , c2 ) for some u ∈ Z×
p }.

We define
J(i) = {g ∈ GLK (CX ⊗ Q) | φX (gc1 , gc2 ) = upi · φX (c1 , c2 ) for some u ∈ Z×
p }.

The union of the J(i) is the group J(Qp ) of unitary similitudes with similitude factor in Q×
p.

Corollary 6.3.4. Let J(Qp ) be the group of unitary similitudes of CX ⊗Zp Qp with similitude
factor in Q×p , and let J(Zp ) be its subgroup stabilizing the lattice CX . There are isomorphisms
of functors on NilpOĔ ,
∼ ∼
M̃r,ε −→ J(Qp )/J(Zp ), MK/F,r,ε −→ J(Qp )o /J(Zp ).
Here J(Qp )o denotes the group of unitary similitudes with similitude factor in Z×
p.

Proof. It is enough to show that Gε (i)(κ̄E ) is in bijection with J(i)/J ′ (Zp ). For this is enough
to show that for each C ∈ Gε (i)(κ̄E ) there exists g ∈ J(i) such that gCX = C. This we have
already shown before (6.3.3). 

In this reformulation it is less obvious what the Weil descent datum is.

6.4. The case r banal and K/F unramified. Let ε ∈ {±1}. We consider a CM-triple
(X, ι, λX ) over κ̄E such that λX is principal if ε = 1 and is almost principal if ε = −1. By
Proposition 4.5.14 such a CM-triple exists and invr (X, ι, λX ) = ε. In fact, by Lemma 8.1.2 and
Theorem 4.5.11, (X, ι, λX ) is unique up to isomorphism.
We recall the functor MK/F,r,ε from section 2.6.
Definition 6.4.1. Let (X, ι, λX ) be a framing object with an almost principal polarization. We
define a functor Mr− (i) on the category NilpOĔ . For R ∈ NilpOĔ , a point of Mr− (i)(R) is given
by the following data:
(1) a local CM-triple (X, ι, λ) of type r over R which satisfies the Eisenstein conditions (ECr )
relative to the fixed uniformizer π ∈ K.
(2) the polarization λ is almost principal.
 106 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

(3) a quasi-isogeny of p-divisible OK -modules

ρ : XR̄ −→ X ×Spec κ̄E Spec R̄,
such that the pullback quasi-isogeny ρ∗ (λX ) differs from pi λ|XR̄ by a scalar in OF× , locally
on Spec R̄. Here R̄ = R ⊗OĔ κ̄E .
Now let (X, ι, λX ) be a framing object with a principal polarization. Then we have h(λX ) = 0
and invr (X, ι, λX ) = 1. We define the functor Mr+ (i) by exactly the same data but we replace
the condition (2) above by the condition
(2′ ) the polarization λ is principal.
In the almost principal case there exists an isogeny X ∧ −→ X such that the composite
λ
X −→ X ∧ −→ X
is πidX . This follows from the following analogue of Proposition 5.3.7.
Proposition 6.4.2. Let α : P1 −→ P2 be an isogeny of CM-pairs of type r over R ∈ NilpOĔ
which both satisfy the Eisenstein condition (ECr ). Let αC : C1 −→ C2 be the morphism in
Et(OK )R associated by the contracting functor Cr,R . Then
height α = 2f · lengthOK Coker αC .
If height α = 2f then there exists a unique morphism β : P2 −→ P1 such that
α ◦ β = πidP2 , β ◦ α = πidP1 ,
Proof. To prove the first assertion we can assume that R = k is an algebraically closed field.
Then we can use that Coker α = Coker αC ⊗Zp W (k). If lengthOK Coker αC = 1, then βC clearly
exists. 
In the case where i = 0, the quasi-isogeny ρ is of height zero because the polarizations λX and
λ have the same degree. We set X0 = X and X1 = X ∧ . Since X and X ∧ satisfy the Eisenstein
condition (ECr ) and, by Proposition 4.2.2, also the Kottwitz condition (KCr ), we obtain a point
of the functor MK/F,r,−1 of Definition 2.6.1. We conclude that Mr− (0) = MK/F,r,1 . The
index r− on the left hand side indicates that we are in the case where the adjusted invariant of
the framing object is −1. Similarly, Mr+ (0) = MK/F,r,1 . The index r+ on the left hand side
indicates that we are in the case where the adjusted invariant of the framing object is 1.
We will describe the formal scheme which represents the functor MK/F,r,ε . More precisely,
consider the functors on NilpOĔ ,
a
M̃r± = Mr± (i).
i∈Z
These functors are endowed with a Weil descent datum ωMr± : Mr± (i) −→ Mr± (i + fE )τE
relative to Ĕ/E using exactly the same definition as in (6.1.4). Recall (CX , φX ) = (CX , ιX , φX ).
We define
J ′ (Zp ) = {g ∈ AutOK CX | φX (gc1 , gc2 ) = f · φX (c1 , c2 ) for some f ∈ OF× },
J ′ (i) = {g ∈ Aut CX ⊗Zp Qp | φX (gc1 , gc2 ) = pi f φX (c1 , c2 ), for some f ∈ OF× }.
As in the ramified case, we see that a point of Mr± (i)(R) is locally on Spec R given by a
lattice C ⊂ CX ⊗Zp Qp such that the restriction of (1/pi )φX to C induces a OF -bilinear form
1
φX : C × C −→ OF
pi
which is perfect in the case of r+ and such that ordπ p1i φX = 1 in the case of r− . For the case of
r− , we are using here Proposition 6.4.2.
We deduce that there is an isometry g : (CX , φX ) −→ (C, (1/pi )φX ), cf. Lemma 8.1.2. Con-
sequently we have g ∈ J ′ (i). Conversely, if g ∈ J ′ (i), the sublattice C = gCX with the bilinear
form (1/pi )φX gives rise to a point of Mr± (i)(R).
We will denote by J ′ ⊂ Aut CX the union of the J ′ (i). We can identify J ′ with a subgroup of
AutoK X exactly as in the ramified case, cf. (6.3.4). It acts via ρ on the functor M̃r± .
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 107

Proposition 6.4.3. There is a J ′ -equivariant isomorphism of functors on NilpOĔ ,

∼
M̃r± −→ J ′ /J ′ (Zp ). (6.4.1)
Here the right hand side is the constant sheaf on NilpOĔ . The Weil descent datum ωMr,±
on the left hand side corresponds on the right hand side to the Weil descent datum given by
multiplication with π efE /2 . Here we view π efE /2 as an automorphism of the K-vector space
CX ⊗Zp Qp by multiplication.
Proof. That (6.4.1) is an isomorphism of functors follows from Theorem 4.5.11 in the same way
as in the proof of Proposition 6.3.2.
Let us recall the definition of the Weil descent relative to OĔ /OE on the functor M̃r± . We
write
FX,τE : X −→ (τE )∗ X (6.4.2)
for the Frobenius relative to κE . Let ε : OĔ −→ R be an object of NilpOĔ . We write ε̄ = ε ⊗ κ̄E :
κ̄E −→ R̄. Let (X, ι, λ, ρ) ∈ Mr± (i)(R) be a point. We view (X, ι, λ) as a CM-triple on R[τE ]
and we endow it with the framing
ρ ε̄∗ FX,τE
ρ̃ : XR̄ −→ ε̄∗ X −→ ε̄∗ (τE )∗ X.
Then (X, ι, λ, ρ̃) defines a point of Mr± (i + fE )(R[τE ] ). Varying i ∈ Z, we obtain the Weil
descent datum
(τ )
ωMr± : M̃r± −→ M̃r±E .
We note that the inverse image of (τE )∗ λ by (6.4.2) is pfE λ. The compatibility of the Weil
descent data follows as in the proof of Proposition 6.3.2 from the following Lemma. 

Lemma 6.4.4. The contracting functor applied to the Frobenius morphism FX,τE : X −→ (τE )∗ X
yields the multiplication by π efE /2 : CX −→ CX .
Proof. We use the Dieudonné module P of X over κ̄E . The map FX,τE induces on the Dieudonné
modules the map
V fE ,♯ : P −→ W (κ̄E ) ⊗F fE ,κ̄E P, x 7−→ 1 ⊗ V fE x.
By definition we have
C := CX = {c ∈ P | V c = πr c},
where we recall πr from (4.5.12), cf. Remark 4.5.13. With the identification CX = C(τE )∗ X , the
restriction of V fE ,♯ to C gives
F −fE +1 F −1
πr · . . . · πr · πr : C −→ C. (6.4.3)
The right hand side is a module over
Y
OK ⊗Zp W (κ̄E ) = OK ⊗O W (κ̄E ).
K t ,ψ̃
ψ∈Ψ

On the right hand side, F −1 is given by the map

F −1
OK ⊗O ˜ W (κ̄E ) −→ OK ⊗O W (κ̄E ), a ⊗ ξ 7−→ a ⊗ ξ.
K t ,ψσ K t ,ψ̃

Therefore the components of the element on the left hand side of (6.4.3) are
aψσ(fE −1)
π · . . . · π aψσ · π aψ .
Since σ fE fixes κE we have aψσfE = aψ . It follows that the numbers
gψ := aψσ(fE −1) + . . . aψσ + aψ
are independent of ψ. We call this number g. We find:
2g = gψ + gψ̄ = efE
because aψ + aψ̄ = e. We conclude that (6.4.3) is the multiplication by π efE /2 . 
108 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Corollary 6.4.5. Let J(Qp ) be the group of unitary similitudes of CX ⊗Zp Qp with similitude
factor in Q×p , and let J(Zp ) be its subgroup stabilizing the lattice CX . There are isomorphisms
of functors on NilpOĔ ,
∼ ∼
M̃r,ε −→ J(Qp )/J(Zp ), MK/F,r,ε −→ J(Qp )o /J(Zp ).
Here J(Qp )o denotes the group of unitary similitudes with similitude factor in Z×
p. 
6.5. The banal split case. Let R = k be an algebraically closed field. There is up to isomor-
phism a unique hermitian OK -module (C, φ) of rank 2 with φ perfect. Hence there is a unique
CM-triple
(X, ιX , λX ) ∈ Ppol
r,κ̄E
with principal λX . We take this as framing object.
We define functors Mr (i) on the category NilpOĔ . For R ∈ NilpOĔ , a point of Mr (i)(R) is
given by the following data:
(1) a local CM-triple (X, ι, λ) of type r over R which satisfies the Eisenstein conditions (ECr )
relative to the fixed uniformizer π ∈ F .
(2) the polarization λ : X −→ X ∧ is principal.
(3) a quasi-isogeny of p-divisible OK -modules
ρ : XR̄ −→ X ×Spec κ̄E Spec R̄,
such that the pullback quasi-isogeny ρ∗ (λX ) differs from pi λ|XR̄ by a scalar in OF× , locally
on Spec R̄. Here R̄ = R ⊗OĔ κ̄E .
Note that Mr (0) = Mr,1 of before Theorem 2.6.3. Consider a point (X, ι, λ, ρ) as above. Let
(C, φ) be the p-adic étale sheaf associated to (X, ι, λ)R̄ on (Spec R̄)ét = (Spec R)ét . Let C X be
the constant sheaf on (Spec R)ét of the OK -module CX . The existence of ρ implies that C is
locally constant for the Zariski topology. Therefore, locally on Spec R, we may regard C as a
submodule of CX ⊗ Q. By the definition of Mr (i), we have
f pi φ(x, y) = φX (x, y), x, y ∈ CX ⊗ Q,
for some f ∈ OF× . We see by Theorem 4.5.11 that a point of Mr (i)(R) is the same as a OK -
sublattice C ⊂ (CX )R ⊗ Q such that the restriction of (1/pi )φX to C induces a perfect pairing
C × C −→ OF .
This is directly clear if the ideal of nilpotent elements of R is nilpotent, and follows from the
argument in the proof of Proposition 6.3.2 in the general case.
Again we set
J ′ (Zp ) = {g ∈ GLOK (CX ) | φX (gc1 , gc2 ) = f · φX (c1 , c2 ), for some f ∈ OF× },
J ′ (i) = {g ∈ GL(CX ⊗Zp Qp ) | φX (gc1 , gc2 ) = pi f φX (c1 , c2 ), for some f ∈ OF× }.
There is an isometry up to a constant in OF× ,
1
g : (CX , φX ) −→ (C, φX ).
pi
Then g ∈ J ′ (i) and gCX = C. Any other isometry of this type is of the form gh, with h ∈ J ′ (Zp ).
Therefore we have associated to the point (X, ι, λ, ρ) a section of the constant sheaf J ′ (i)/J ′ (Zp ).
We set a
M̃r = Mr (i), J ′ = ∪i∈Z J ′ (i).
i∈Z

As in the ramified case, the group J ′ acts via ρ on the functor M̃r , cf. (6.3.4). Moreover, this
functor is endowed with the Weil descent datum ωMr : Mr (i) −→ Mr (i + fE )(τE ) relative to
OĔ /OE .
Let σ ∈ Gal(F t /F ) be the Frobenius. We use the notation introduced below (4.3.22). We fix
θ ∈ Θ, with θ1 , θ2 ∈ Ψ. Set
a1,E = aθ1 σfE −1 + . . . + aθ1 σ + aθ1 (6.5.1)
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 109

This number is independent of the choice of θ because, by the definition of the reflex field E,
aθ1 σfE = aθ1 .
In the same way we define a2,E by using θ2 . If we add both definitions we find
a1,E + a2,E = efE .
The endomorphism
π aE := π a1,E ⊕ π a2,E : CX,1 ⊕ CX,2 −→ CX,1 ⊕ CX,2 (6.5.2)
is an element of J ′ (fE ).
Proposition 6.5.1. The polarized contraction functor defines an isomorphism of functors on
NilpOĔ ,
∼
M̃r −→ J ′ /J ′ (Zp ). (6.5.3)
The Weil descent datum ωMr relative to OĔ /OE corresponds on the right hand side to the Weil
descent datum given by multiplication with π aE ∈ J ′ (fE ).

We note that J ′ /J ′ (Zp ) = J ′ /J ′ (Zp )(τE ) because this is true for any constant sheaf. Propo-
sition 6.5.1 is the consequence of the definition of ωMr and the following Lemma.
Lemma 6.5.2. The Frobenius FX,τE : X −→ (τE )∗ X induces on CX the multiplication by π aE .

Proof. The statement needs an explanation. Because the functor Cpol

r,κ̄E commutes with base
change, we have a canonical isomorphism CX = C(τE )∗ X . Indeed, the inverse image of the
constant sheaf CX by Spec τE is the constant sheaf CX .
Let M = PX be the Dieudonné module of X. The Frobenius FX,τE is induced by the Ver-
schiebung
V fE : M −→ M.
We write in this proof C := CX . By definition we have
−1
C = M πr V = C1 ⊕ C2 ,
cf. Remark 4.5.13. Therefore the action of V fE on C coincides with the action of
F −fE +1 F −1
πr · . . . · πr · πr : C −→ C. (6.5.4)
We look at the components of the element on the left hand side in (4.3.22). Let us consider the
components of the first set of factors of (4.3.22) which act on C1 . The component of (6.5.4) in
the factor OF ⊗O t ,θ̃ W (OE ′ ) is
F

aθ
π f −1
1σ E · . . . · π aθ1 σ · π aθ1 ⊗ 1 = π a1,E ⊗ 1.
Therefore V fE induces on C1 the multiplication by π a1,E . By the same argument it induces on
C2 the multiplication by π a2,E . 

Corollary 6.5.3. Let J(Qp ) = GLK (CX,1 ⊗Zp Qp ) and J(Zp ) = GLOK (CX,1 ). There are iso-
morphisms of functors on NilpOĔ ,
∼ ∼
M̃r −→ J(Qp )/J(Zp ), MK/F,r,1 −→ J(Qp )o /J(Zp ).
×
Here J(Qp )o denotes the subgroup of elements with determinant in OK . 

7. Application to p-adic uniformization

In this section, we reap the global fruits from our local work in the preceding sections. This
section is modelled on the case of p-adic uniformization of the first kind of the previous paper
[18, section 4].
110 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

7.1. The Shimura variety and its p-integral model. In this section, K and F will be
number fields. Let K/F be a CM-field. We fix an archimedean place w0 of F . We denote by
a 7→ ā the conjugation acting on K.
Let V be a K-vector space of dimension 2. Let
ς( , ) : V × V −→ Q
be a non-degenerate alternating Q-bilinear form such that
ς(av1 , v2 ) = ς(v1 , āv2 ), a ∈ K, v1 , v2 ∈ V.
We define three algebraic groups over Q. For a Q-algebra R, the R-valued points are:
U (V, ς)(R) = {g ∈ GLK⊗R (V ⊗ R) | ςR (gx1 , gx2 ) = ςR (x1 , x2 )}
G(V, ς)(R) = {g ∈ GLK⊗R (V ⊗ R) | ςR (gv1 , gv2 ) = µ(g)ςR (v1 , v2 ), µ(g) ∈ R× }, (7.1.1)
×
Ġ(V, ς)(R) = {g ∈ GLK⊗R (V ⊗ R) | ςR (gv1 , gv2 ) = ςR (µ(g)v1 , v2 ), µ(g) ∈ (F ⊗ R) }.
If (V, ς) is fixed, we write U, G, Ġ.
Equivalently, we can replace the form ς by the anti-hermitian form
κ : V × V −→ K
on the K-vector space V which is defined by the equation
TrK/Q aκ(v1 , v2 ) = ς(av1 , v2 ), a ∈ K. (7.1.2)
Then κ is linear in the first argument and anti-linear in the second.
For each place w of F we obtain an anti-hermitian pairing
κw : V ⊗F Fw × V ⊗F Fw −→ K ⊗F Fw . (7.1.3)
Let w : F −→ R be an archimedean place. We choose an extension of w to ϕ : K −→ C. This
defines an isomorphism K ⊗F Fw ∼
= C and κw becomes an anti-hermitian pairing
κϕ : V ⊗K,ϕ C × V ⊗K,ϕ C −→ C.
Note that the space V is determined up to isomorphism by the signature at the archimedean
places of F and the local invariants invv (V ) at the non-archimedean places v of F . We will
impose the following signature condition on V . Let Φ = HomQ-Alg (K, C) and let r be a special
CM-type of rank 2 wrt. w0 , i.e., a function
r : Φ −→ Z>0 , ϕ 7−→ rϕ , (7.1.4)
such that rϕ + rϕ̄ = 2 for all ϕ ∈ Φ and such that there is exactly one pair {ϕ0 , ϕ̄0 } such that
rϕ0 = rϕ̄0 = 1. The archimedean place determined by ϕ0 and ϕ̄0 is supposed to be w0 .
We require that κϕ is isomorphic to the anti-hermitian form on C2 given by the matrix
 
iErϕ 0
, (7.1.5)
0 −iErϕ̄
for every ϕ. Here Erϕ is denotes the unit matrix of size rϕ , and i the imaginary unit. We note
that the last requirement is independent of the choice of ϕ above w. We endow
Y
V ⊗Q R = V ⊗F,w R,
w:F −→R
with the complex structure J such that κw (v1 , J v2 ) is hermitian and positive-definite for all w.
This defines a Shimura datum (G, h), resp., (Ġ, h), cf. [8, 4.9, 4.13] and an associated Shimura
variety Sh(G, h) with canonical model over the reflex field E of r.
Let p be a prime number. We choose a place v of F of residue characteristic p which does
not split in K and such that invv (V ) = −1. We choose an embedding E −→ Q̄p and denote by
ν the induced place of E, with corresponding prime ideal pν . We assume that the place vν of F
induced by
ϕ0
F −→ E −→ Q̄p
is equal to v, i.e, that pν induces the prime ideal pv corresponding to v.
We have an isomorphism
V ⊗ Qp ∼= ⊕p|p V ⊗F Fp . (7.1.6)
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 111

Here p runs over all prime ideals of F which divide p. We will write
Vp := V ⊗F Fp , Kp = K ⊗F Fp .
The decomposition (7.1.6) is orthogonal with respect to ς and, for each prime ideal p of F over
p, we obtain a bilinear form
ςp : Vp × Vp −→ Qp .
It is related to κp by
TrK⊗F Fp /Qp aκp (x1 , x2 ) = ςp (ax1 , x2 ), a ∈ K ⊗F Fp , x1 , x2 ∈ Vp .
One defines algebraic groups over Qp as in (7.1.1) above:
Up = U (Vp , ςp ), Gp = G(Vp , ςp ), Ġp = Ġ(Vp , ςp ).
Let p|p be such that p is unramified (and hence non-split) in K/F and such that inv(Vp , κp ) =
−1, cf. Definition 8.1.1. Then there is a OKp -lattice
Λp ⊂ Vp
such that ςp induces a bilinear form
ςp : Λp × Λp −→ Zp (7.1.7)
such that Λp is almost self-dual, i.e., h(Λp , ςp ) = 1 (compare (8.1.4)). Any other lattice with
these properties has the form gΛp where g ∈ U(Vp , ςp )(Qp ). This follows from Lemma 8.1.2.
In all other cases, we apply the following lemma.
Lemma 7.1.1. Assume that if p is unramified in K/F , then inv(Vp , κp ) = 1. Then there is an
OKp -lattice Λp ⊂ V ⊗F Fp such that ςp induces a perfect pairing
ςp : Λp × Λp −→ Zp .
Any other such lattice is of the form gΛp where g ∈ Up (Qp ).
Proof. Indeed, because of Lemmas 8.1.2 and 8.1.3, we need only a justification in the case where
p is split. In this case we have Kp = Fp × Fp and, accordingly, a decomposition V ⊗F Fp =
U1 ⊕ U2 . The vector spaces U1 and U2 are isotropic with respect to ςp and therefore ςp induces
an isomorphism U2 = HomQp (U1 , Qp ). The form ςp becomes
ςp (x + x∗ , y + y ∗ ) = x∗ (y) − y ∗ (x), x, y ∈ U1 , x∗ , y ∗ ∈ U2 .
The existence and uniqueness of Λp follows easily. 
To pass to a p-integral model over OE,(pν ) of Sh(G, h), we restrict the choice of the level
structure. To do this, we choose for each p|p a OK⊗F Fp -lattice Λp ⊂ Vp as above. We define
Kp = {g ∈ Gp | gΛp = Λp }
(7.1.8)
Kp = {g ∈ G(Qp ) | gΛp = Λp for all p|p}.
We choose an open compact subgroup Kp ∈ G(Apf ) and set K = Kp · Kp ⊂ G(Af ).
We extend the embedding ν : E −→ Q̄p to an embedding Q̄ −→ Q̄p . We obtain a decomposi-
tion a a
Φ = HomQ-Alg (K, Q̄) = HomQp -Alg (Kp , Q̄p ) = Φp . (7.1.9)
p p
The restriction of the function r to Φp will be denoted by rp . The group Gal(Q̄p /Eν ) acts on Φ
via the restriction
Gal(Q̄p /Eν ) −→ Gal(Q̄/E).
Therefore rτ ϕ = rϕ for ϕ ∈ Φ and τ ∈ Gal(Q̄p /Eν ). This implies that the local reflex fields
E(Kp /Fp , rp ) are all contained in Eν .
Let R be an OEν -algebra. Let L be an R-module with an OK -action. We have decompositions
Y
OK ⊗ R = OKp ⊗Zp R, L = ⊕p Lp .
p

We will say that L satisfies the Eisenstein condition (ECr ) if each Lp satisfies the Eisenstein
condition (ECrp ), cf. (2.2.12). We use a similar terminology for the Kottwitz condition (KCr ).
112 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Definition 7.1.2. We define the groupoid AK on the category of OE,(pν ) -algebras. A point of
AK (R) consists of the following data:
(1) An abelian scheme A over R, up to isogeny of degree prime to p, with an algebra homomor-
phism
ι : OK −→ End A ⊗ Z(p) .
such that Lie A ⊗OE,(pν ) OEν satisfies the conditions (KCr ) and (ECr ).
(2) A Q-homogeneous polarization λ̄ of A such that the Rosati involution induces the conjugation
of K/F .
(3) A class of OK -linear isomorphisms
η̄ p : V ⊗ Apf −→ V p (A) mod Kp
which respect the forms on both sides up to a constant in Apf (1)× .
We impose the following two conditions.
(i) There exists a polarization λ ∈ λ̄ such that the induced map to the dual variety λ : A → A∧
has the following property. Let Q (ker λ)p be the p-primary part of the kernel of λ. It has
the decomposition (ker λ)p = p|p (ker λ)p . We require that (ker λ)p is trivial, unless p is
unramified in K and inv(Vp , ςp ) = −1. In the latter case the height of (ker λ)p is 2fp .
(ii) There is an identity of invariants,
invrp (A, ι, λ) = invp (Vp , ςp ), for all p|p.
p
An isomorphism of such data (A, ι, λ̄, η̄ p ) −→ (A′ , ι′ , λ̄′ , η¯′ )
is given by a OK -linear quasi-isogeny
φ : A −→ A′ of degree prime to p compatible with the Q-homogeneous polarizations and the level
structures.
We will denote a point of AK (R) simply by (A, ι, λ̄, η̄ p ).
Remark 7.1.3. It is equivalent to consider in (2) a Z(p) -homogeneous polarization λ̄ of A such
that the elements λ ∈ λ̄ satisfy the condition (i) on the p-primary part of the kernel of λ.
Remark 7.1.4. Let (A, ι, λ̄, η̄ p ) ∈ AK (R). Let λ ∈ λ̄ be as in condition (i) of Definition 7.1.2.
For each geometric point ω of characteristic 0 of Spec R the pairing induced by λ on the p-adic
Tate module,
Ep : Tp (Aω ) × Tp (Aω ) −→ Zp (1) (7.1.10)
has the following properties. If p is ramified in K/F , the pairing is perfect and inv(Vp (Aω ), Ep ) =
inv(Vp , ςp ). If p is unramified, then (7.1.10) is perfect if inv(Vp , ςp ) = 1 and is almost perfect if
inv(Vp , ςp ) = −1. If p is split in K/F , then (7.1.10) is perfect.
For each geometric point ω of R of characteristic p, the polarization λ induces a pairing on
the Dieudonné module M of Aω and therefore for each prime p|p of F a pairing
Ep : Mp (Aω ) × Mp (Aω ) −→ W (κ(ω)), (7.1.11)
r
with the following properties. If p is ramified in K/F , the pairing is perfect and inv (Mp (Aω ), Ep ) =
inv(Vp , ςp ). If p is unramified, then (7.1.11) is perfect if inv(Vp , ςp ) = 1 and is almost perfect if
inv(Vp , ςp ) = −1. If p is split in K/F , then (7.1.11) is perfect.
Proposition 7.1.5. Assume that Kp is small enough. Then the functor AK is representable by
a projective scheme over Spec OE,(pν ) whose generic fiber is the Shimura variety ShK associated
to the Shimura datum (G, h). For general Kp , AK is a DM-stack proper over Spec OE,(pν ) whose
generic fiber is the Shimura variety ShK considered as an orbifold.
Proof. Let (A, ι, λ̄, η̄ p ) ∈ AK (R), and fix η p ∈ η̄ p . Let Λ ⊂ V be a OK -lattice on which ς is
integral. We find an abelian variety A1 in the class A up to isogeny prime to p such that for
each ℓ 6= p
η p (Tℓ (A1 )) = Λ ⊗ Zℓ .
In this way we obtain also a polarization on A1 whose degree is bounded in terms of ς and Λ. If
Kp is small enough we obtain a level structure on the m-division points for some m ≥ 3.
The fact that the moduli problem of abelian varieties with a polarization of given degree and a
m-level structure for m ≥ 3 is a quasi-projective scheme implies that the functor of (A, ι, λ̄, η̄ p ) as
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 113

in (1)–(3) of Definition 7.1.2, is representable by a quasi-projective scheme. Now the conditions

(i) and (ii) define open and closed subschemes (this is easy for condition (i), and follows from
Proposition 8.2.1 for condition (ii)). The representability by a Deligne-Mumford stack for general
Kp follows.
To compare the generic fiber of AK with ShK , recall from [15, §8] that ShK represents the
following functor AK,E on the category of E-algebras, comp. section 1.2. A point of AK,E (R)
consists of the following data.
(1) An abelian scheme A over R, up to isogeny, with an algebra homomorphism
ι : OK −→ End A ⊗ Q
such that the Kottwitz condition (KCr ) is satisfied.
(2) A Q-homogeneous polarization λ̄ of A such that the Rosati involution induces the conjugation
of K/F .
(3) A class of K-linear isomorphisms
η̄ : V ⊗ Af −→ V̂ (A) mod K
which respect the forms on both sides up to a constant in Af (1)× .
Here we are implicitly using the fact that G satisfies the Hasse principle, cf. [15, §7]. We define
a map AK,E (R) −→ AK (R). We fix a OK -lattice Λ in V whose localizations at p|p are the given
lattices Λp above. Let (A, ι, λ̄, η̄) ∈ AK,E (R), and fix η ∈ η̄. We find an abelian variety A1 in
the isogeny class A such that for each ℓ
η(Tℓ (A1 )) = Λ ⊗ Zℓ .
Then we obtain ι1 : OK −→ End(A1 ) ⊗ Z(p) . The Eisenstein condition (ECr ) is automatically
satisfied, cf. Proposition 2.2.1. We also find a polarization λ1 ∈ λ̄ which satisfies the condition
(i) in Definition 7.1.2. The existence of η implies Condition (ii). By forgetting the p-component
of η̄, we have associated to (A, ι, λ̄, η̄) a well-defined object (A1 , ι1 , λ̄1 , η̄ p ) of AK (R). By the
uniqueness property of the lattices Λp mentioned above, this map is bijective.
The properness of AK −→ Spec OE,(pν ) is a consequence of Proposition 7.1.7 below. 
Remark 7.1.6. If pv is the only prime ideal of F over p, then it follows from the product
formula that condition (ii) of Definition 7.1.2 is automatically satisfied. Indeed, condition (ii)
defines an open and closed subscheme which in this case has the same generic fiber.
Proposition 7.1.7. The morphism AK −→ Spec OE,(pν ) is proper.
For the proof of Proposition 7.1.7, we need two lemmas.
Lemma 7.1.8. Let K/F be a CM-field and let r be a generalized CM-type of rank 2. Let E ⊂ Q̄
be the reflex field. Let R be a complete discrete valuation ring with an OE -algebra structure. Let
L be the field of fractions of R. We assume that the residue characteristic of R is p > 0. Let
w be a finite place of F of residue characteristic ℓ, such that Kw /Fw is a field extension. We
assume that L is of characteristic zero or that ℓ 6= p.
Let (A, ι, λ) be a CM-triple of type r over L. The polarization induces on the rational Tate
module Vw (A) an alternating pairing
ψw : Vw (A) × Vw (A) −→ Qℓ (1). (7.1.12)
If inv(Vw (A), ψw ) = −1, then the abelian variety A has potentially good reduction.
Proof. We consider only the case ℓ = p. We may assume that A has semistable reduction. We
choose an isomorphism Qp ∼= Qp (1) over L̄. We obtain from ψw the anti-hermitian form
κw : Vw (A) × Vw (A) −→ Kw ,
cf. (8.1.2). Let T be the toric part of the special fiber of the Néron model of A. Then OK acts
on the character group X∗ (T ). If T is non-trivial, we obtain that
dim T = [K : Q] = dim A.
This implies that the toric part Vwt (A) ⊂ Vw (A) is a Kw -vector subspace of dimension 1. By the
orthogonality theorem [SGA7, Exp IX, Thm. 5.2], the anti-hermitian form κw is zero on this
114 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

subspace. Let u1 , u2 be a basis of Vw (A) such that u1 is a basis of Vwt (A). Then we obtain, in
the notation of (8.1.1),
dK/F (Vw (A), κw ) = −κw (u1 , u2 )κw (u2 , u1 ) = κw (u1 , u2 )κw (u1 , u2 ) ≡ 1
×
modulo NmKw /Fw Kw . This contradicts the assumption inv(Vw (A), κw ) = −1. 
With the notation of the last lemma, we consider the case where ℓ = p and where the
characteristic of L is also p. The OE -algebra structure on R factors
OE −→ κν −→ R,
where κν is the residue field of Eν . We fix a commutative diagram
EO ν / Q̄p
O

E / Q̄.
Let w be a p-adic place of F . By the last diagram we can restrict r to a local CM-type rw of
Kw /Fw . Then Eν is the composite of the subfields E(Kw /Fw , rw ), for w running over all places
of F over p. Q
Let (A, ι, λ) be a CM-triple of type r over L. The action of OF ⊗ Zp = w OFw , where w
runs over all p-adic places of F , induces a decomposition of the p-divisible group of A:
Y
X= Xw .
w

Then (Xw , ιw , λw ) is a local CM-triple with respect to Kw /Fw , rw over the field L.
Lemma 7.1.9. Let R be a discrete valuation ring of equal characteristic p > 0, and let L be the
field of fractions. Let R be an OE -algebra. Let ν be the p-adic place of E induced by this algebra
structure.
Let (A, ι, λ) be a CM-triple of type r over L. We assume that there is a p-adic place w of F
such that one of the following conditions is satisfied.
(1) Kw /Fw is a ramified field extension. The local CM-triple (Xw , ιw , λw ) satisfies the Eisenstein
condition (ECrw ), and invr (Xw , ιw , λw ) = −1.
(2) Kw /Fw is an unramified field extension. The local CM-triple (Xw , ιw , λw ) satisfies the Eisen-
stein condition (ECrw ), and λw is almost principal.
Then the abelian variety A has potentially good reduction over R.
Proof. We may assume that A has semistable reduction over R. Let à be the Néron model over
R, and let B be the identity component of the special fibre of Ã. Let us assume that the torus
part T ⊂ B is nontrivial. Since OK acts on T , we obtain that X∗ (T )Q is a K-vector space of
dimension one. Let Y be the p-divisible group of T . We obtain a decomposition
Y
Y = Yu
u

where u runs over all places of K over p and Yu is an OKu -module which is of height [Ku : Qp ]
and of multiplicative type. We pass from R to the completion R̂. Let X̂ = XK̂ be the p-divisible
group of AK̂ . By [SGA7, Exp IX, §5], the multiplicative group Yw lifts to a multiplicative group
Ỹw ⊂ X̂wf of the finite part of X̂w over R̂. If we pass to the general fibre of the last inclusion
we obtain a nontrivial multiplicative subgroup (Ỹw )L̂ ⊂ X̂w . But our assumption implies, by
Lemma 5.2.8 in the ramified case, and by Proposition 5.3.5 in the unramified case, that X̂w
is isoclinic of slope 1/2. This contradicts the existence of a nontrivial multiplicative part and
therefore the assumption that the torus part of B is nontrivial. 
Proof. (of Proposition 7.1.7) We check the valuative criterion. Let R a discrete valuation ring
with a OEν -algebra structure. Let L be the field of fractions of R. Let α : Spec L −→ AK be a
OEν -morphism. We have to show that α extends to Spec R −→ AK . It is enough to show that
for a discrete valuation ring R′ which dominates R, the morphism Spec L′ −→ AK induced by α
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 115

extends Spec R′ −→ AK . The map α gives a point (A, ι, λ̄, η̄ p ) ∈ AK (L). Since we may replace L
by L′ we may assume that A has semistable reduction. Let ω be a geometric point concentrated
in the generic point of Spec R. We are assuming that invr (Tpv (Aω ), Epv ) = inv(Vpv , ςpv ) = −1
when char L = 0, resp. invr (Mpv (Aω ), Epv ) = inv(Vpv , ςpv ) = −1, when char L = p. Hence we
conclude by the last two lemmas that A has good reduction. Let Ã/R denote the abelian scheme
which extends A. Then ι extends to an action ι̃ of OK on Ã. The Kottwitz condition and the
Eisenstein condition are closed conditions and hold therefore for Ã. The polarization λ extends
to λ̃ : Ã −→ Ã∧ . The condition (i) from Definition 7.1.2 extends from A to Ã. For a geometric
point ω0 concentrated in the closed point of Spec R we find
invrp (Ãω0 , ι̃ω0 , λω0 ) = invr (Mp (Ãω0 ), Ep ) = inv(Vp , ςpv )
because the left hand is by Proposition 8.2.1 equal to invrp (ÃL , ιL , λL ). Hence condition (ii) from
Definition 7.1.2 also extends from A to Ã. From this we obtain an extension of (A, ι, λ̄, η̄ p ) to a
point of AK (R). 
Remark 7.1.10. The scheme AK turns out to be flat over Spec OE,(pv ) , cf. Theorem 7.3.3, (i).
Hence its generic fiber is dense. It follows that it is enough to check the valuative criterion on
discrete valuation rings R with fraction field L of characteristic zero. Hence Lemma 7.1.9 is not
needed.
The following proposition shows that there is only one isogeny class in the special fiber of
AK . This is the underlying reason why there is p-adic uniformization.
Proposition 7.1.11. Let κν be the residue class field of Eν . Let (A1 , ι1 , λ̄1 , η̄1p ) and (A2 , ι2 , λ̄2 , η̄2p )
be two points of AK (κ̄ν ). Then there exists a quasi-isogeny
(A1 , ι1 , λ̄1 ) −→ (A2 , ι2 , λ̄2 ),
i.e., a quasi-isogeny which respects the actions ιi and the Q-homogeneous polarizations λ̄i . In
fact, there exists such a quasi-isogeny of degree prime to p.
Q
Proof. Let Xi be the p-divisible group of Ai , with its decomposition Xi = p|p Xi,p . It follows
from Proposition 5.2.7 (jointly with Lemma 5.2.8) and Proposition 5.3.6 that Xi,pv is isoclinic.
In the banal cases p 6= pv , the same follows from Lemma 4.3.3 for Xi,p . By [27, Cor. 6.29] we
find a quasi-isogeny
a : (A1 , ι1 ) −→ (A2 , ι2 ).
We choose λi ∈ λ̄i . We set λ = a∗ (λ2 ). We find an endomorphism u ∈ Endo (A1 ) such that
λ = λ1 u.
Since λ1 and λ induce the conjugation on K, we conclude that u ∈ EndoK A1 . Moreover u is
fixed by the Rosati involution ∗ induced by λ1 on D := EndoK A1 . It is enough to find an element
d ∈ D× such that
u = f d∗ d (7.1.13)
for some element f ∈ Q× . The solutions of these equations form a torsor under the algebraic
group J over Q such that
J(Q) = {e ∈ D× | e∗ e ∈ Q× }. (7.1.14)
By [15, §7], this group satisfies the Hasse principle. Therefore it is enough to find a solution of
the equation (7.1.13) in D ⊗ Q× w for all places w of Q. If w is a finite place w 6= p we have, by
[27, Cor. 6.29], that
D ⊗ Qw = EndK⊗Qw Vw (A1 )
such that the Riemann form Ewλ1 induces the involution ∗. A solution of (7.1.13) exists iff the
symplectic K ⊗ Qw -modules
(Vw (A1 ), Ewλ1 ), (Vw (A2 ), Ewλ2 )
p p
are similar up to a factor in Q× w . But this follows from the existence of η̄1 and η̄2 .
In the case w = p we can use Dieudonné modules. In this case we know, by condition
(ii) in Definition 7.1.2, that the rational Dieudonné modules of A1 and A2 together with their
polarizations are isomorphic.
116 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

If w is the infinite place, one can deduce the assertion from the fact that u in (7.1.13) is totally
positive.
Now let us prove the second assertion. We consider the Dieudonné modules M1 , resp. M2 ,
of A1 , resp. A2 . We choose the polarizations λ1 ∈ λ̄1 , resp. λ2 ∈ λ̄2 , as in condition (i) of
Definition 7.1.2. Using the contracting functor, it is clear that there is a quasi-isogeny of height
zero α : (M1 , λ1 ) −→ (M2 , λ2 ). Let ρ : (M1 , λ̄1 ) −→ (M2 , λ̄2 ) be an arbitrary quasi-isogeny.
Consider the morphism
α ◦ ρ−1 : (M2 , λ̄2 ) −→ (M2 , λ̄2 ).
We consider the group J for (A2 , ι2 , λ̄2 ) (compare (7.1.14)). Then α ◦ ρ−1 is an element of J(Qp )
by Tate’s theorem ([27, Cor. 6.29]). We approximate it by an element α1 ∈ J(Q). Then
ρ ◦ α1 : (A1 , ι1 , λ̄1 ) −→ (A2 , ι2 , λ̄2 )
is the desired quasi-isogeny of order prime to p. 
7.2. The RZ-space M̃r . We fix a point (A0 , ι0 , λ̄0 , η̄0p ) of AK (κ̄ν ). We also fix a polarization
λ0 ∈ λ̄0 which satisfies the condition (i) of Definition 7.1.2. We denote by X the p-divisible
group of A0 . The action ι0 induces an action ιX on X and λ0 induces a polarization λX on X.
We denote by qν the number of elements in κν = κEν .
Let R ∈ NilpOEν and let (X, ι) be a p-divisible group over Spec R with an action
ι : OK ⊗ Zp −→ End X.
The notion of a semi-local CM-triple (X, ι, λ) relative to K ⊗ Qp /F ⊗ Qp and r should be obvious
but we explain it more precisely: the decomposition
Y
OF ⊗ Zp = OFp
p

induces the decomposition Y

X= Xp .
Let λ be a polarization of X which induces the conjugation on K/F . Then the decomposition
extends to Y
(X, ι, λ) = (Xp , ιp , λp ). (7.2.1)
We call (X, ι, λ) a semi-local CM-triple of type r if each (Xp , ιp , λp ) is a local CM-triple of type
rp with respect to Kp /Fp . This makes sense because E(Kp /Fp , rp ) ⊂ Eν .
Definition 7.2.1. A semi-local CM-triple (X, ι, λ) of type (K ⊗ Qp /F ⊗ Qp , r) over an alge-
braically closed field with a κEν -algebra structure is said to be compatible with (V, ς) if, for each
p|p,
invr (Xp , ιp , λp ) = inv(Vp , ςp ),
and if λp is principal, except in the case where Kp /Fp is unramified and inv(Vp , ςp ) = −1. In
the latter case λp is almost principal.
We note that the CM-triple (X, ιX , λX ) over κ̄ν is compatible with (V, ς) and satisfies the
conditions (KCr ) and (ECr ), in the sense explained before Definition 7.1.2.
Definition 7.2.2. Let i ∈ Z. Let Mr (i) be the following functor on the category NilpOĔ . For
ν
an object R ∈ NilpOĔ , write R̄ = R ⊗OĔν κ̄Eν . A point of Mr (i)(R) is given by the following
ν
data:
(1) A CM-triple (X, ι, λ) of type (K ⊗ Qp /F ⊗ Qp , r) over Spec R which satisfies the conditions
(KCr ) and (ECr ) and is compatible with (V, ς).
(2) A OK ⊗ Zp -linear quasi-isogeny
ρ : X̄ := X ×Spec R Spec R̄ −→ X ×Spec κĔν Spec R̄
such that ρ respects the polarization pi λ on X and λX up to a factor in (OF ⊗ Zp )× .
We denote these data by (X, ι, λ, ρ). Two data (X, ι, λ, ρ) and (X ′ , ι′ , λ′ , ρ′ ) define the same
point of Mr (i) iff there is an isomorphism α : (X, ι) −→ (X ′ , ι′ ), such that ρ′ ◦ αR̄ = ρ. In
particular, α respects the polarizations λ and λ′ up to a factor in (OF ⊗ Zp )× .
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 117

Remark 7.2.3. In (2) we could replace the last condition on ρ by

(2′ ) The quasi-isogeny ρ respects the polarizations as follows,
pi λ = ρ∗ λX .
Then we obtain a functor which is naturally isomorphic to Mr (i). This follows because for
a ∈ (OF ⊗ Zp )× the points (X, ι, aλ, ρ) and (X, ι, λ, ρ) of Mr (R) are isomorphic. We could
also require upi λ = ρ∗ λX for some u ∈ Z× p without changing the functor. We use different
descriptions of the functor Mr (i) in order to describe better different group actions.
Let τEν ∈ Gal(Ĕν /Eν ) be the Frobenius automorphism and let fEν be the inertia index of
Eν /Qp , i.e., qν = pfEν . As earlier, the Frobenius FX,τEν defines a Weil descent datum on these
functors,
ωMr : Mr (i)(R) −→ Mr (i + fEν )(R[τEν ] ), (7.2.2)
cf. (6.1.3), (6.2.2). Since the degrees of the polarizations λ and λX are the same, it follows that
2 height ρ = height(pi | X) = 4[F : Q]i.
Q
More precisely, ρ = p ρp where p runs over the prime ideals of F over p. For each p we have
2 height ρp = height(pi | Xp ) = 4[Fp : Qp ]i.
We define a
M̃r = Mr (i). (7.2.3)
i∈Z

We describe the functor M̃r with its Weil descent datum. Let
J(Qp ) = {α ∈ EndoK⊗Qp X | α∗ (λX ) = cλX , for some c ∈ Q×
p }. (7.2.4)

This group acts naturally on M̃r via the rigidification ρ. We consider the decomposition (7.2.1)
for X. We set
Jp = {α ∈ EndoKp Xp | α∗ (λp ) = cλp , for some c ∈ Q×
p} (7.2.5)
For all p the groups Jp are subgroups of J ′ as introduced in section 6 in the local cases and they
agree with J introduced in the banal cases.
We will give an explicit description of these groups. For this, it is convenient to replace the
bilinear form ςp by the Fp -bilinear form
ς˜p : Vp × Vp → Fp ,
which is defined by
ςp (x1 , x2 )) = ςp (ax1 , x2 ),
t(a˜ a ∈ Fp ,
−1
for t(a) = TrFp /Qp ϑ a, where as usual ϑ ∈ OF is the different of F/Qp . The restriction to the
lattices Λp gives
ς˜p : Λp × Λp → OFp .
Let us consider the prime p = pv . We denote by Dv the quaternion division algebra over Fv .
We choose a two-dimensional Kv -vector space with an anti-hermitian form
ς¯pv : V̄pv × V̄pv → Fpv
of invariant +1. The contraction functor associates to (Xpv , ιXpv , λXpv ) a special formal ODv -
module Y with the relative polarization λv = ψv , resp. λv = θv , as in section 5.2 resp. 5.3.
Since the endomorphism ring is not changed by the contraction functor, it follows from Lemmas
5.2.2 and 5.3.2 that there is an isomorphism
Jpv = G(V̄pv , ς¯pv ). (7.2.6)
For a banal prime p|p of F , we consider the image (CXp , φXp ) by the polarized contraction
functor Cpol
r,k of Theorem 4.5.11. By Proposition 8.3.6, it follows from Condition (ii) in Definition
7.1.2 that there is an isomorphism
(CX , φX ) ∼
= (Λp , ς˜p ).
p p (7.2.7)
118 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

More precisely, Condition (ii) implies that the corresponding vector spaces are isomorphic; the
integral isomorphism follows from Lemmas 8.1.2 and 8.1.3. Therefore we obtain
Jp = G(Vp , ς˜p ) = Gp , for p 6= pv . (7.2.8)
Since we want a uniform notation, we set (V̄p , ς¯p ) = (Vp , ς˜p ) for p 6= pv . We set
Ḡp = G(V̄p , ς¯p ).
We now have fixed an isomorphism Jp = ∼ Ḡp for all p|p. For p banal, we have Gp = Ḡp .
Let
V̄p = ⊕p|p V̄p .
This is an K ⊗ Qp -module. Let
ς¯p : V̄p × V̄p → F ⊗ Qp
be the orthogonal sum of the forms ς¯p . We define
ςp (x, y), for some c ∈ Q×
Ḡ(Qp ) := G(V̄p , ς¯p ) := {g ∈ AutK⊗Qp (V̄p ) | ς¯p (gx, gy) = c¯ p }. (7.2.9)
We have shown that Ḡ(Qp ) = J(Qp ). In the description of the descent data, the following
slightly larger group will be needed. We define the group Ḡ′p ⊃ Ḡ(Qp ) via
ςp (x, y), for µp (g) ∈ pZ (OF ⊗ Zp )× },
Ḡ′p = {g ∈ AutK⊗Qp (V̄p ) | ς¯p (gx, gy) = µp (g)¯
and
ςp (x, y), for µp (g) ∈ pZ OF×p }.
Ḡ′p = {g ∈ AutKp (V̄p ) | ς¯p (gx, gy) = µp (g)¯
The groups Ḡ′p are isomorphic to the groups Jp′ = J ′ introduced in section 6 in the local cases.
We fix these isomorphisms which are associated to the framing objects. Therefore Q the groups
Ḡ′p act on the local moduli spaces M of section 6 and the subgroup Ḡ′p ⊂ p|p Ḡ′p acts on M̃r ,
cf. (7.2.3).
We define the group Ĝ′ (Qp ) as the union of the following sets for i ∈ Z,
Y
Ĝ′ (i) = {(c, gp ) ∈ pi OF×pv × Ḡ′p | µp (gp ) ∈ pi OF×p , for all p}. (7.2.10)
p banal

Let Ĝ (Zp ) ⊂ Ĝ (Qp ) be the subgroup of elements (c, gp ) such that c ∈ OF×pv and gp (Λp ) = Λp .
′ ′

The multiplicator µpv : Ḡ′pv → pZ OF×pv induces homomorphisms

Ḡ′p → Ĝ′ (Qp ) and G(Qp ) → Ĝ′ (Qp ). (7.2.11)

For the second map we used the identification Ḡp = Gp for p banal.
Definition 7.2.4. We consider the following element wr′ = (c, wp ) ∈ Ĝ′ (Qp ).
(1) c = pfEν .
(2) If Kp /Fp is ramified and hence λXp is principal, wp is the multiplication
e fEν
Πpp : V̄p −→ V̄p ,
see Proposition 6.3.2.
(3) If Kp /Fp is unramified, then both principal and almost principal λXp are allowed. In both
cases we define wp as the multiplication
e fEν /2
πpp : V̄p −→ V̄p ,
see Proposition 6.4.3.
(4) In the case where Kp = Fp ×Fp is split and hence λXp is principal, we have the decomposition
V̄p = V̄p,1 ⊕ V̄p,2 . Let E(rp ) be the reflex field of (Kp /Fp , rp ). Let frp be the inertia index
of E(rp ). We consider the numbers a1,E(rp ) and a2,E(rp ) as defined by (6.5.1). We set
f Eν f Eν
a1,Eν = a1,E(rp ) , a2,Eν = a2,E(rp ) .
f rp f rp
a
Then we have a1,Eν + a2,Eν = ep fEν . We define wp to be the multiplication by πp 1,Eν on
a
V̄p,1 and the multiplication by πp 2,Eν on V̄p,1 .
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 119

The element wr′ is clearly an element of the center of Ĝ′ (Qp ).

Proposition 7.2.5. There exists an isomorphism
∼ b Fv ×Spf OF Spf O ) × Ĝ′ (Qp )/Ĝ′ (Zp )
M̃r −→ (Ω v Ĕν

which is equivariant with respect to the action of Ḡ′p on both sides. This extends the action of
J(Qp ) = Ḡ(Qp ) ⊂ Ḡ′p .
The Weil descent datum ωMr relative to OĔν /OEν on the left hand side (7.2.2) corresponds
on the right hand side to
(ξ, g) 7−→ (ωτEν (ξ), wr′ g), g ∈ Ĝ′ (Qp ).
Proof. We use the decomposition
Y
Mr (i) = Mrp (i),
p|p

which follows immediately from (7.2.1). Then we conclude by the results of section 6, in partic-
ular Propositions 6.3.2, 6.4.3, 6.5.1. 
Remark 7.2.6. We may multiply each wp by a unit in Kp in the Definition 7.2.4 of wr′ . This
does not change the assertion of the last Proposition.
We introduce the group
Y
Ĝ(Qp ) = {(c, gp ) ∈ Q×
p × Gp | µp (gp ) = c, for all p banal}. (7.2.12)
p banal

There are natural homomorphisms

G(Qp ) → Ĝ(Qp ) and Ḡ(Qp ) → Ĝ(Qp ). (7.2.13)
For the second map, we used that in the definition of Ĝ(Qp ) we can replace Gp by Ḡp . In
particular the groups G(Qp ) and J(Qp ) act on Ĝ(Qp ). We denote by Ĝ(Zp ) ⊂ Ĝ(Qp ) the
subgroup of all (c, gp ) such that c ∈ Z×p and gp Λp = Λp . By Corollaries 6.3.4, 6.4.5, 6.5.3, we
obtain a bijection
∼
Ĝ(Qp )/Ĝ(Zp ) −→ Ĝ′ (Qp )/Ĝ′ (Zp ).
Corollary 7.2.7. There exists an isomorphism
∼ b Fv ×Spf OF Spf O ) × Ĝ(Qp )/Ĝ(Zp )
M̃r −→ (Ω v Ĕν

which is equivariant with respect to the action of J(Qp ) on both sides. 

Note that in this version of Proposition 7.2.5 we loose control of the descent data.
7.3. The p-adic uniformization. We will now define a uniformization morphism in the sense
of [27]. We fix a point (A0 , ι0 , λ̄0 , η̄0p ) of AK (κ̄ν ). The uniformization morphism will depend
on the choice of η0p ∈ η̄0p . This choice defines a point of the proscheme proj limKp AK for all
congruence subgroups K = Kp Kp as above. We also fix a polarization λ0 ∈ λ̄0 which satisfies the
condition (i) of Definition 7.1.2. Recall the p-divisible group with induced polarization (X, ιX , λX )
corresponding to A0 .
We denote by ÂK the restriction of AK to the category NilpOEν . The uniformization mor-
phism
Θ : M̃r × G(Apf )/Kp −→ ÂK ×Spf OEν Spf OĔν (7.3.1)
is defined as follows. Let (X, ι, λ, ρ) ∈ Mr (i)(R) and let g ∈ G(Apf ). Recall the notation
R̄ = R ⊗OĔν κ̄ν . The quasi-isogeny ρ extends uniquely to a quasi-isogeny of abelian varieties
ρ : Ā −→ A0 ×Spec κ̄ν Spec R̄. (7.3.2)
Because OK acts on X̄ = X ⊗R R̄, we obtain a map OK −→ End(Ā) ⊗ Z(p) . Moreover the
p
polarization λ0 : A0 −→ A∧ ′ ∧
0 induces on Ā a quasi-polarization λĀ : Ā −→ Ā and η̄0 induces
p
η̄Ā = V p (ρ−1 ) ◦ η0p : V ⊗ Apf −→ V p (Ā) mod Kp .
120 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

On the p-divisible groups, λ′Ā differs from pi λ by a factor from (OF ⊗ Zp )× and therefore
λĀ := p−i λ′Ā satisfies the condition (i) in the Definition 7.1.2 of the functor AK .
We associate to the pair (X, g) from the left hand side of (7.3.1) the point
p
(Ā, ιĀ , λĀ , η̄Ā g) ∈ AK (R̄). (7.3.3)
The CM-triple (X, ι, λ) over R defines by the Serre-Tate theorem a lifting of (7.3.3) to a point
of AK (R). This finishes the definition of the uniformization morphism Θ in (7.3.1).
Lemma 7.3.1. The uniformization morphism is compatible with the Weil descent data ωMr
acting on the first factor on the left hand side and the natural Weil descent data on ÂK ×Spf OEν
Spf OĔν .
Proof. This is essentially [27, Thm. 6.21] but we repeat the simple argument in our context.
By definition of the Weil descent data repeated below, it is enough to consider both sides of
(7.3.3) on the category of κ̄ν -algebras R. We will denote by ε : κ̄ν −→ R the algebra structure.
Consider a point (X, ι, λ, ρ) ∈ M̃r (R). The Weil descent datum ωMr is obtained by changing ρ
to ρ′ :
ρ ε∗ FX,τE
ρ′ : X −→ ε∗ X −→ ν ε∗ (τEν )∗ X.
This gives a point (X, ι, λ, ρ′ ) ∈ M̃r (R[τEν ] ). The point (X, ι, λ, ρ) defines a quasi-isogeny of
abelian varieties
ρ : A −→ ε∗ A0 ,
as explained in the definition of Θ. The point (X, ι, λ, ρ′ ) defines in the same way the quasi-
isogeny of abelian varieties over R[τEν ] ,
ε∗ FA0 ,τE
ν
A −→ ε∗ A0 −→ (ετEν )∗ A0 .
Here A with its additional structure is regarded as a point of ÂKp (R[τEν ] ). This makes sense
because to be a point of ÂKp (R) depends only on the κν -algebra structure on R. In other words
ÂKp (R) = ÂKp (R[τEν ] ). (7.3.4)
But this equation is the Weil descent datum on the right hand side of (7.3.1). 
We define the group
J(Q) = {γ ∈ EndoK A0 | γ ∗ λ0 = uλ0 , for some u ∈ Q× }, (7.3.5)
cf. (7.1.14). Regarded as an algebraic group over Q, the group J is an inner form of G. In the
proof of Proposition 7.1.11 we saw that the Qp -valued points of J coincide with the group J(Qp )
of (7.2.4). We proved in section 7.2 that Ḡ(Qp ) = J(Qp ). Let γ ∈ J(Q). With the chosen η0p ,
we define ω(γ) ∈ G(Apf ) by the equation
V p (γ) ◦ η0p = η0p ω(γ). (7.3.6)
This defines a homomorphism
ω : J(Q) −→ G(Apf ),
and an isomorphism J(Apf ) ∼ p
= G(Af ). Therefore J and G are isomorphic over the finite places
w 6= p of Q. At the infinite place J is anisotropic because the Rosati involution is positive.
The group J(Qp ) acts on M̃r ,
(X, ι, λ, ρ) 7−→ (X, ι, λ, γρ), γ ∈ J(Qp ).
Let ((X, ι, λ, ρ), g), with g ∈ G(Apf ) be a point from the left hand side of (7.3.1) and let
p
(A, ιA , λA , ηA g) be its image by Θ, cf. (7.3.3). If γ ∈ J(Q), the quasi-isogeny γρ extends
to the quasi-isogeny of abelian schemes
ρ γ
Ā −→ (A0 )R̄ −→ (A0 )R̄ .
In follows from (7.3.6) that the image of ((X, ι, λ, γρ), g) by the morphism Θ is
p
(A, ιA , λA , ηA ω(γ −1 )g)
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 121

We define an action of J(Q) on the left hand side of (7.3.1) by

((X, ι, λ, ρ), g) 7−→ ((X, ι, λ, γρ), ω(γ)g).
Proposition 7.3.2. The uniformization morphism (7.3.1) factors through an isomorphism
∼
Θ : J(Q)\(M̃r × G(Apf )/Kp ) −→ ÂK ×Spf OEν Spf OĔν .
This isomorphism is compatible with the Weil descent data relative to OĔν /OEν . Here the Weil
descent datum on the left is induced from ωMr , cf. Proposition 7.2.2.
Proof. We have just proved that the morphism is well-defined. The bijectivity follows from the
Proposition 7.1.11 and [27, Thm. 6.30]. 
By inserting Proposition 7.2.5 in this result, we obtain our main theorem about uniformization.
Theorem 7.3.3. (i) The OEν -scheme AK is projective and flat, with semi-stable reduction.
(ii) Let ÂK be the completion of AK along its special fiber, which is a formal scheme over
Spf OEν . There exists an isomorphism of formal schemes over Spf OĔν ,

p  ∼
b Fv ×Spf OF Spf O ′ ′ p
J(Q)\ Ω v Ĕν × Ĝ (Qp )/Ĝ (Zp )×G(Af )/K −→ ÂK ×Spf OEν Spf OĔν . (7.3.7)
For varying Kp , this isomorphism is compatible with the action of G(Apf ) through Hecke corre-
spondences on both sides.
Let wr′ the element in the center of Ĝ′ (Qp ) of Definition 7.2.4. We endow the left hand with
the Weil descent datum
(ξ, h, g) 7−→ (ωτEν (ξ), wr′ h, g), h ∈ Ĝ′ (Qp ), g ∈ G(Apf ).
Then the isomorphism (7.3.7) is compatible with the Weil descent data on both sides. 
There is a simpler version of this statement, as follows. When the inertia index fEν is even, this
simpler version can be used to describe the descent datum. We define Ĝ(Af ) = Ĝ(Qp ) × G(Apf ),
where we recall Ĝ(Qp ) from (7.2.12), and K̂ = Ĝ(Zp ) × Kp .
Corollary 7.3.4. There is a natural isomorphism of formal schemes

 ∼
b Fv ×Spf OF Spf O
J(Q)\ Ω v Ĕν × Ĝ(Af )/K̂ −→ ÂK ×Spf OEν Spf OĔν . (7.3.8)
Assume that the inertia index fEν is even. The multiplication by p on V ⊗ Qp defines an element
of G(Qp ). Let p̂ be the image in Ĝ(Qp ). We also denote by p̂ the element
(p̂, 1) ∈ Ĝ(Qp ) × G(Apf ) = Ĝ(Af ).
If we endow the left hand side of (7.3.8) with the Weil descent datum
(ξ, g) 7−→ (ωτEν (ξ), p̂fEν /2 g), g ∈ Ĝ(Af ),
then the morphism (7.3.8) is compatible with the Weil descent data. 
∼
Note that G(Qp )/G(Zp ) − → Ĝ(Qp )/Ĝ(Zp ), as follows from Kpv = ker(c : Gpv (Qp ) −→
Q× ×
p /Zp ). Hence
G(Af )/K ≃ Ĝ(Af )/K̂. (7.3.9)
Hence Corollary 7.3.4 implies Theorem 1.2.3 in the Introduction.
7.4. The uniformization for deeper level structures at p. We now pass to deeper level
structures. For each prime ideal p of OF with p|p we have the group
Gp = {g ∈ GLOKp (Vp ) | ςp (gx1 , gx2 ) = µ(g)ςp (x1 , x2 ), for some µp (g) ∈ Q×
p },

and the open compact subgroup Kp ⊂ Gp cf. (7.1.8). We will assume that there exist prime
ideals p which are banal since our deeper level structures exist only in this case. For each banal
p, we choose an open subgroup of K∗p ⊂ Kp . For a natural number M we consider the subgroup
Kp (pM ) ⊂ Kp which consists of the elements that act trivial on Λp /pM Λp . We will assume that
for some M
Kp (pM ) ⊂ K∗p (7.4.1)
122 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

For the special prime pv we set K⋆pv = Kpv . We set

K⋆p = {g = (gp ) ∈ G(Qp ) | gp ∈ K⋆p }.
This says that µp (gp ) is independent of p. We also introduce
Y
K⋆,ba
p = {(gp ) ∈ K⋆p | µp (gp ) = c ∈ Z×
p , independent of p}. (7.4.2)
p, banal

This is a subgroup of
Y
Gba (Qp ) = {(gp ) ∈ Gp | µp (gp ) = c ∈ Q×
p , independent of p}.
p, banal
ba
Q
Also, let OK = p, banal OKp .
We need with some generalities on p-divisible groups suited for our special case. Let X and
Y be p-divisible groups on a scheme S. We consider the category of étale morphisms U → S
with the étale topology.
Definition 7.4.1. Let n ∈ N. We define a sub-presheaf
Gpn ⊂ Homet (X(n), Y (n)),
where the right hand side denotes the Hom in the category of étale sheaves. A homomorphism
α : X(n)U → Y (n)U belongs to Gpn (U ) if there is a profinite étale covering Ũ → U and a
homomorphism of p-divisible groups α̃ : XŨ → YŨ such that the restriction of α̃ to X(n)Ũ is
αŨ .
We denote the sheafification of Gpn by Gn . We define the prosheaf
Homet (X, Y ) = ′′
lim′′ Gn .
←
The limit is taken with respect to the natural restriction maps Gn → Gm for n > m.
We note that a homomorphism of p-divisible groups α̃ : XŨ → YŨ defines a homomorphism
α : X(n)U → Y (n)U iff
pr⋆1 α̃ − pr⋆2 α̃ ∈ pn Hom(XŨ ×U Ũ , YŨ ×U Ũ ). (7.4.3)
We consider now a banal local CM-type (Kp /Fp , rp ). Let Ep = E(rp ) be the corresponding
reflex field. Let (X, ιX ) and (Y, ιY ) be local CM-pairs over S/ Spf OEp , which satisfy the Eisen-
stein condition. As above, we define Homet OKp (X, Y ) by replacing throughout homomorphisms by
homomorphisms of OKp -modules. The presheaf Gpn is now meant in this sense. The contracting
functor (cf. Definition 4.5.3) associates p-adic étale sheaves CX and CY with an OKp -module
structure. By Theorem 4.5.4,
HomO (X, Y ) ∼
Kp = HomO (CX , CY ).
Kp

n
We set Cn,X = CX /p CX . One checks easily by the remark after Definition 7.4.1 that
Gpn (U ) = HomOKp (CX , Cn,Y ) = HomOKp (Cn,X , Cn,Y ). (7.4.4)
In particular Gpn = Gn . We conclude that, for a scheme S/ Spf OEp , the pro-sheaf Homet
OKp (X, Y )
is a p-adic étale sheaf. Let ι : ω → S be a geometric point. Then we find for the fiber
Homet
OKp (X, Y )ω = HomOKp (Xω , Yω ),

where the right hand side is the Hom in the cateory of p-divisible OKp -modules.
Let us assume that S is a scheme over Spf OĔp . The contracting functor of Theorem 4.5.11
associates to a CM-triple (X, ιX , λX ) which satisfies the Eisenstein condition a p-adic étale sheaf
CX with an alternating form
φX : CX × CX → OFp . (7.4.5)
We set ξX = TrFp /Qp ϑ−1 φX . In particular there is a CM-triple (Xp , ιXp , λXp ) over κ̄Ĕp such that
(CXp , ξλXp ) ∼
= (Λp , ςp ), (7.4.6)
cf. (7.2.7).
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 123

The group Kp acts on the right hand side by similitudes. Therefore we obtain a homomor-
phism Kp → AutOKp Xp such that the automorphisms in the image respect the polarization λXp
up to a factor in Z×
p.

Definition 7.4.2. Let p be banal and let (X, ιX , λX ) be a CM-triple on S which satisfies the
Eisenstein condition as above. A CL-level structure on (X, ιX , λX ) as a class of isomorphisms
of p-adic étale sheaves
∼
(Λp , ςp ) −→ (CX , ξλX ) mod K⋆p , (7.4.7)
which respect the bilinear forms on both sides up to a factor in Z×
p . We will write
∼
Xp −→ X mod K⋆p . (7.4.8)
for a CL-structure.
More precisely this means the following. Let M ≥ 1 such that (7.4.1) holds. Then a CL-level
structure is a right K⋆p /(Kp (pM ))-torsor
T ⊂ IsomOKp (Λp ⊗ Z/pM Z, CX ⊗ Z/pM Z)

such that the inclusion is equivariant with respect to the right actions of K⋆p /(Kp (pM )) on both
sides and such that the local sections of T respect the bilinear forms on Λp and CX up to a
factor in (Z/pM Z)× . If S is connected and ω → S is a geometric point a CL-structure is given
by a K⋆p -orbit of an isomorphism Λp → (CX )ω which respects the bilinear forms on both sides
by a factor in Z× p and such that the orbit is preserved by the action of π1 (S, ω). This explains
the notation (7.4.8).
Let (X, ι, λ) be a semi-local CM-triple relative to (K ⊗ Qp /F ⊗ Q Qp , r) over a scheme S ∈
(Sch/ Spf OĔν ), cf. the beginning of section 7.2. We set X ba = p, banal Xp . We choose
(X, ιX , λX ) as in section 7.2. Then (7.4.6) holds. From this we obtain an action of K⋆,bap on Xba
Q ×
which respects the polarization p, banal λXp up to a factor in Zp .
∼
We define a CL-level structure on Xba modulo K⋆,ba p as a CL-level stuctures η̄p : Xp − → Xp
mod K⋆p , for each banal p which respect the bilinear forms up to a factor in Z×p that is independent
of p.
Definition 7.4.3. With the notations of Definition 7.2.2, let i ∈ Z. Let MK⋆p (i) be the following
functor on the category of schemes S over Spf OĔν . We will write S̄ = S ×Spf OĔν Spec κ̄Eν . A
point of MK⋆p (i)(S) is given by the following data:
(1) A CM-triple (X, ι, λ) of type (K ⊗ Qp /F ⊗ Qp , r) over S which satisfies the conditions (KCr )
and (ECr ) and is compatible with (V, ς).
(2) A OK ⊗ Zp -linear quasi-isogeny
ρ : X̄ := X ×S S̄ −→ X ×Spec κĔν S̄

such that ρ respects the polarization pi λ on X and λX up to a factor in Z×

p.
ba
Q
(3) Let X = p banal Xp . A CL-level structure
∼
η̄ : Xba −→ X ba mod K⋆,ba
p .
We set a
M̃K⋆p = MK⋆p (i).
i∈Z

We formulate a variant of Corollary 7.2.7. Since we assume that banal places exist, we do not
need the group Ĝ.
Proposition 7.4.4. There exists an isomorphism
∼
M̃K⋆p −→ (Ω̂Fv ×Spf OFv Spf OĔν ) × Gba (Qp )/K⋆,ba
p ,
which is equivariant with respect to the action of J(Qp ) on both sides.
124 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Proof. At the banal places we may use lisse p-adic étale sheaves to describe a point of M̃K⋆p (S).
A point consists of a CM-triple (Xpv , ιpv , λpv ) and a quasi-isogeny ρpv : Xpv ,S̄ → Xpv such that
ρ⋆pv (λXpv ) = upi λXpv for some u ∈ Z× p , i ∈ Z and an isomorphism of lisse p-adic étale sheaves
on S̄,
η
(CXba , ξXba ) −→ (C, ξ) ⊂ (CXba , ξXba ) ⊗ Q, (7.4.9)
where η respects the alternating forms up to a factor in Z× p and such that the restriction of ξXba
i
with respect to the last inclusion is equal to up ξ with the same u and i as above. By (7.2.7) we
have
(CXba , ξXba ) ∼= (Λba , ς ba ), (7.4.10)
where the right hand side is the orthogonal direct sum over all (Λp , ςp ) for p banal.
We denote by M̃ba K⋆p
the moduli functor described by the data (7.4.9). We claim that there is
a natural isomorphism
M̃ba ∼
= Gba (Qp )/K⋆,ba . (7.4.11)
K⋆p p

Indeed, the group Gba (Qp ) acts naturally on this functor: Let g ∈ Gba (Qp ) such that
φX (gx1 , gx2 ) = u′ pj ξX (x1 , x2 ).
Then g maps (7.4.9) to
gη
(CXba , ξXba ) −→ (gC, (1/u′ pj )ξ) ⊂ CXba ⊗ Q.
If we have an arbitrary point (7.4.9), then the composite of the arrow with the inclusion is an
element g ∈ Gba (Qp ) and therefore (7.4.9) is isomorphic to
g
CXba −→ (gC, (1/u′ pj )ξXba ) ⊂ CXba ⊗ Q.
We see that the action is transitive and that the stabilizer of the base point
id
(CXba , ξXba ) −→ (CXba , ξXba ) ⊂ CXba ⊗ Q
is K⋆,ba
p . This shows (7.4.11).
Now we fix i ∈ Z. We denote by Mrpv the functor of section 6 associated to the special local
CM-type (Kpv /Fpv , rpv ). There is the natural injection of functors
MK⋆p (i) → Mrpv (i) × M̃ba
K⋆p
(i).
We claim that this map is surjective. Indeed, assume we are given a point (Xpv , ιpv , λpv , ρ⋆pv ),
where ρ⋆pv (λXpv ) = u1 pi λpv , with u1 ∈ Z×
p , from the first factor on the right hand side, and a
point (C, ξ) ⊂ CXba (endowed with η), where u2 pi ξ = ξXba with u2 ∈ Z× p , from the second factor.
These two data form a point of MK⋆p (i) iff u1 = u2 . But in the point from the first factor we
can replace λpv by (u1 /u2 )λpv without changing the isomorphism class of this point. Therefore
the surjectivity holds and the proposition follows as Proposition 7.2.5. 
Let us fix an open and compact subgroup Kp ⊂ G(Apf ). We set K⋆ = K⋆p Kp and K = Kp Kp
as after (7.1.8). We choose (X, ιX , λX ) as above.
Definition 7.4.5. We define a functor Â⋆K⋆ on the category of schemes S over Spf OĔν . A point
of Â⋆K⋆ (S) consists of the following data:
(1) a point (A, ι, λ̄, η̄ p ) of AK (S),
(2) a CL-level structure
η̄p : Xba → A[p∞ ]ba mod K⋆,ba
p .
We denote here by A[p∞ ]ba the banal part of the p-divisible group of A with its structure of
a semi-local CM-triple. The morphism Â⋆K⋆ → ÂK is a finite étale covering of formal schemes.
Since we assume that Kp is small enough, AK is a proper scheme over Spec OĔν . By the
algebraization theorem, there is a unique finite étale morphism of schemes over Spec OĔν
A⋆K⋆ → AK ×Spec OE,(pν ) Spec OĔν , (7.4.12)
such that the p-adic completion of A⋆K⋆ is Â⋆K⋆ .
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 125

Recall the projective scheme AK⋆ ,E over E from section 1.2 (the canonical model of ShK⋆ ),
comp. the proof of Proposition 7.1.5. We will now relate AK⋆ ,E with the general fiber A⋆K⋆ ×Spec OĔν
Spec Ĕν . We start with a reformulation of the level structure η̄p in Definition 7.4.5.
We assume that S is a scheme over Spf OĔp , i.e., we pass to the completion of the maximal
unramified extension of Ep . We consider now a polarized local CM-pair (X, ιX , λX ) over S of
CM-type (Kp /Fp , rp ), cf. Definition 4.1.2. We will always assume that the Eisenstein conditions
are satisfied. By Theorem 4.5.11, λX is described by a OFp -bilinear form φλX , or also
ξλX : CX × CX → Zp ,
as defined after (7.4.5). Equivalently, we can consider the OKp -anti-hermitian form
κλX : CX × CX → Kp , (7.4.13)
which is defined by
TrKp /Fp aκλX (c1 , c2 ) = φλX (ac1 , c2 ), a ∈ OKp , c1 , c2 ∈ CX .
Then κλX is OKp -linear in the first variable and OKp -anti-linear in the second variable.
If we define (X, ιX , λX ) by (CX , ξλX ) ∼
= (Λp , ςp ) (cf. (7.4.6)), we can reformulate (7.4.7): A
CL-structure is a class of isomorphisms
(X, λX ) → (X, λX ) mod K⋆p
which respect the polarizations up to a factor in Z× p . This agrees with Definition 7.4.5.
Let κ̄Ep be the residue class field of Ĕp . We will consider CM-pairs (Z, ιZ ) of CM-type rp /2
over a scheme S/ Spf OĔp . Then Z is a p-divisible group of height 2dp and dimension dp , where
dp = [Kp : Fp ]. We will always assume that the Eisenstein condition is fulfilled. Proposition
4.2.2 continues to hold with the same polynomials EAψ . The functor CZ (cf. Definition 4.5.3)
exists for local CM-pairs of type (Kp /Fp , rp /2).
We will reformulate CL-level structures as suggested by [24]. There is up to isomorphism a
unique CM-pair (X̄0 , ῑ0 ) of CM-type rp /2 over κ̄Ep . It lifts uniquely to a CM-pair (X0 , ι0 ) over
OĔp , and
CX0 ∼= OKp (7.4.14)
is the constant p-adic sheaf. We consider biextensions
β : X0 × X0 → Ĝm
or, equivalently, bilinear forms of displays as in Proposition 4.5.9. They are in bijection with
bilinear forms
φ : CX0 × CX0 → OFp . (7.4.15)
Equivalently we use ξ = ξφ or κ = κφ as before (7.4.13).
We define κ0 : C0 × C0 → OKp using (7.4.14), by
κ0 (x, y) = xȳ. x, y ∈ OKp . (7.4.16)
We denote by s0 : X0 → X0∧ the homomorphism associated to φ0 . This homomorphism is
symmetric.
We note that there is a principal polarization λ on X0 . It is a generator of the free OKp -
module of rank one HomOKp (X0 , X0∧ ). In the case where Kp /Fp is ramified, the corresponding
form under the bijection (7.4.15) is
φλ (x, y) = TrKp /Fp Π−1 xȳ.
Then s0 = λΠ. In the case where Kp /Fp is unramified, we choose a unit ε ∈ OKp such that
ε + ε̄ = 0. Then the corresponding form under the bijection (7.4.15) is
φλ (x, y) = TrKp /Fp ε−1 xȳ.
Then s0 = λε.
Let S be a p-adic formal scheme over Spf OĔp . Let (X, ιX , λX ) be a polarized CM-pair
of type (Kp /Fp , rp ) which satisfies the Eisenstein condition as always required. We endow
HometOKp (X0 , X) with an OKp -anti-hermitian form with values in Kp . Let u1 , u2 ∈ Cn (U ). They
126 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

are given by homomorphisms ũ1 , ũ2 : X0 → X which are defined over a profinite étale covering
Ũ → U . We consider the homomorphism
ũ∧ ∧ ∧
2 λX ũ1 : X0 → X → X → X0 .

This element of HomOKp ((X0 )Ũ , (X0∧ )Ũ ) may be written as

ũ∧
2 λX ũ1 = δ̃(ũ1 , ũ2 )s0 , (7.4.17)

with some constant δ̃(ũ1 , ũ2 ) ∈ Kp . In the ramified case,

δ(u1 , u2 ) := δ̃(ũ1 , ũ2 ) mod pn Π−1 OOKp

is well defined. In the unramified case, the element δ̃(ũ1 , ũ2 ) is well-defined modulo pn OKp .
Varying n, we therefore obtain a bilinear form
δ : Homet et
OKp (X0 , X) × HomOKp (X0 , X) → Kp .

This is a OKp -anti-hermitian form. We set

e = TrFp /Qp TrKp /Fp ϑ−1

Fp /Qp δ.

Then e is an alternating form

e : Homet et
OKp (X0 , X) × HomOKp (X0 , X) → Zp .

which satisfies e(au1 , u2 ) = e(u1 , āu2 ), a ∈ OKp .

Proposition 7.4.6. A CL-level structure on a polarized CM-pair (X, ιX , λX ) of type (Kp /Fp , rp )
over the p-adic formal scheme S can equivalently be given as a class of isomorphisms of p-adic
étale sheaves
∼
η : (Λp , ςp ) −→ (Homet
OKp (X0 , X), e) mod K⋆p , (7.4.18)
which respect the bilinear forms on both sides up to a constant in Z×
p.

Proof. Indeed, we apply the contracting functor to the right hand side of the isomorphism
(7.4.18). We view ũ1 , ũ2 from (7.4.17) as homomorphisms
ũi : OKp = CX0 → CX .
Let κλX : CX × CX → Kp be the anti-hermitian form induced by λX . The definition (7.4.17) of
the sesqui-linear form δ which gives rise to e, reads in terms of the contracting functor as defined
by (7.4.17)
κλX (ũ1 (x), ũ2 (y)) = δ̃(ũ1 , ũ2 )xȳ. (7.4.19)
If we identify
Homet et
OKp (X0 , X) = HomOKp (OKp , CX ) = CX

by sending ũ to ũ(1), the form δ̃ is mapped to the form κλX . This is immediate by setting
x = y = 1 in (7.4.19). Therefore we have identified the right hand side of (7.4.18) with (CX , ξλX ).
This proves the assertion. 

Proposition 7.4.7. Let S be a flat proper scheme over Spec OĔp . Let (X, ιX ) and (Y, ιY ) be
CM-pairs of type (Kp /Fp , rp ) or (Kp /Fp , rp /2) over S. Let U → S be a finite étale covering,
and Û → Spf OĔp its formal completion along the special fibre. Set Uη = U ×Spec OĔp Spec Ĕp .
Then there is a natural bijective homomorphism
Gn (Û ) → HomOKp (X(n)Uη , Y (n)Uη ). (7.4.20)

In particular, the p-adic étale sheaf HomOKp (Tp (XSη ), Tp (YSη )) is unramified along the special
fibre of S.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 127

Proof. We consider the natural map Gn (Û ) → HomOKp (X(n)Û , Y (n)Û ). By Grothendieck’s ex-
istence theorem (EGA III, Thm. 5.1.4), the target of this arrow coincides with HomOKp (X(n)U , Y (n)U ).
If we restrict the last set of homomorphisms to the generic fibre we obtain the map (7.4.20).
The injectivity of (7.4.20) follows from the definition of Gn . To prove surjectivity, we can
assume that U is connected. By Grothendieck’s existence theorem we find a finite connected
étale covering U1 → U such that the sheaves Cn,XŜ and Cn,YŜ become trivial over Û1 .
We write the proof only in the case where X and Y are of CM-type rp . The cases where rp /2
appears will be obvious. By the choice of U1 , we deduce the isomorphism
∼ HomO ((OK /pn OK )2 , (OK /pn OK )2 ).
Gn (Û1 ) = Kp p p p p

We choose a geometric point ω of (U1 )η . Then we obtain injective homomorphisms

Gn (Û1 ) → HomOKp (X(n)U1,η , Y (n)U1,η ) =
HomOKp (Tp (Xη ) ⊗ Z/(pn ), Tp (Yη ) ⊗ Z/(pn ))(U1,η ) −→
HomOKp (Tp (Xω ), Tp (Yω )) ⊗ Z/(pn ) ∼
= HomO ((OK )2 , (OK )2 ) ⊗ Z/(pn ).
Kp p p

Since we have the same number of elements on both sides, the arrows are bijective. In particular
this shows that the étale sheaf HomOKp (Tp (Xη ), Tp (Yη )) ⊗ Z/(pn ) becomes trivial over the finite
étale covering U1,η → Sη . Therefore it is unramified along the special fibre of S.
Finally, we obtain the bijectivity of (7.4.20) by exploiting the sheaf property with respect to
the covering
U1 × U U1 → → U1 → U.

Corollary 7.4.8. With the assumptions of the last Proposition, there is a p-adic étale sheaf
HomOKp (X, Y ) on S whose restriction to the special fibre S×Spec OĔp Spec κ̄Ep is Homet
OKp (Xκ̄Ep , Yκ̄Ep )
and whose restriction to the general fibre S ×Spec OĔp Spec Ĕp is HomOKp (Tp (XĔp ), Tp (YĔp )).

Proof. The sheaves Gn over Ŝ are representable by finite étale morphisms of formal schemes.
They come therefore from finite étale morphisms Gal n → S. We have to compare the general
fibre of Gal
n with Hom OKp (T p (X η ), T p (Yη )) ⊗ Z/(p n
).
We have shown that both sheaves are trivialized by a finite étale covering S1 → S. The
homomorphism (7.4.20) gives a canonical isomorphism between these sheaves with constant
étale sheaves on S1 ×Spec OĔp Spec κĔp . Finally, we consider descent for the general fiber of the
covering
p1
S1 × S S1 →→ S1 → S
p2
We see that the descent data for the two sheaves agree since they are induced from the descent
datum on the étale sheaf Hom(X(n), Y (n)). 
We now go back to the Definition 7.4.5. We choose for each banal p a CM-pair (Xp,0 , ιp,0 )
of local CM-type (Kp /Fp , rp /2) over Spf OĔν . We may assume that CXp,0 = OKp . We endow
CXp,0 with the hermitian form (7.4.16) which corresponds to the symmetric homomorphism
∧
Q Q
sp,0 : Xp,0 → Xp,0 . We define X0ba = p, banal Xp,0 and we endow it with sba
0 = sp,0 . Then
by Proposition 7.4.6 we may replace (2) in Definition 7.4.5 by
(2′ ) A class η̄p of isomorphisms of p-adic étale sheaves,
∞ ba
ηp : Λba → HomOK ba
ba (X0 , A[p ] ) mod K⋆,ba
p ,
which respect the forms on both sides up to a constant in Z×
p.

The lisse p-adic sheaf on ÂK given by the right hand side of (2′ ) is the algebraization of a lisse p-
adic sheaf on AK which exists because this scheme is proper over Spec OĔν . We denote this sheaf
by the same symbol. Then the scheme A⋆K⋆ is given by the following functor on the category
of schemes S over Spec OĔν : A point of A⋆K⋆ (S) consists of a point (A, ι, λ̄, η̄ p ) of AK (S) and a
class η̄p as in (2′ ). We deduce the following description of A⋆K⋆ .
128 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Proposition 7.4.9. The scheme A⋆K⋆ ×Spec OĔν Spec Ĕν represents the following functor on the
category of Ĕν -schemes. A T -valued point is a point (A, ι, λ̄, η̄ p ) of AK (T ) and a class η̄p of
isomorphisms of p-adic étale sheaves
∼
ηp : Λba −→ HomOK ba
ba (Tp ((X0 )
ba ⋆,ba
Ĕν ), Tp (A) ) mod Kp , (7.4.21)
which respect the forms on both sides up to a constant in Z×
p.
The scheme A⋆K⋆ is the normalization of AK ×Spec OE,(pν ) Spec OĔν in A⋆K⋆ ×Spec OĔν Spec Ĕν
and is finite and étale over AK ×Spec OE,(pν ) Spec OĔν . 

Theorem 7.4.10. Let Ĕνab be the maximal abelian extension of Ĕν . Then there is an isomor-
phism
AK⋆ ,E ×Spec E Spec Ĕνab ∼
= A⋆K⋆ ×Spec OĔν Spec Ĕνab (7.4.22)
which is natural in K⋆ .
Proof. We make explicit what a level structure (7.4.21) means after base change to Ĕνab . Over
Ĕνab we may choose an isomorphism Zp (1) ∼ = Zp and therefore we do not need to worry about
Tate twists. The Tate module Tp (Xp,0 ) of Xp,0 over an algebraic closure of Ĕν is an OKp -module
which is free of rank 1. Therefore the Galois group of Ĕν acts on the Tate-module via its maximal
abelian quotient. We choose an isomorphism
Tp (Xp,0 ) ∼
= OK p (7.4.23)
such that the action of the Galois group of Ĕνab
on both sides is trivial. The symmetric map
∧
sp : Xp,0 → Xp,0 induces a hermitian form κp,0 on the Tate-module (7.4.23). We find
κp,0 (x, y) = cp,0 xȳ, x, y ∈ OK
for some constant cp,0 ∈ OF×p .
Note that in the ramified case two isomorphism classes are possible
for κp,0 .
p
Q We consider a T -valued point (A, ι, λ̄, η̄ , η̄p ) from the right hand side of (7.4.22). Let X =
Xp be the p-divisible group of A. A polarization from λ̄ induces an anti-hermitian pairing κp
on Tp (Xp ). The anti-hermitian form δp on Hom(Tp (Xp,0 ), Tp (Xp )) is given by
κp (u1 (x), u2 (y)) = δp (u1 , u2 )cp,0 xȳ, x, y ∈ OKp , (7.4.24)
where u1 , u2 ∈ Hom(Tp (Xp,0 ), Tp (Xp )) are sections.
For an OKp -lattice (Γ, κΓ ) with an anti-hermitian form κΓ : Γ × Γ → Kp , we write Γ[c] =
(Γ, cκΓ ). The equation (7.4.24) gives an isomorphism
(Hom(Tp (Xp,0 ), Tp (Xp ), cp,0 δp ) ∼
= (Tp (Xp ), κp ).
We see that a level structure (7.4.21) at the banal prime p is given by an isomorphism
Λp [cp,0 ] → (Tp (Xp ), κp ) mod K⋆p .
Choosing a fixed isomorphism Λp [cp,0 ] ∼
= Λp , we see that such a level structure at p is the same
as a class of isomorphisms
η̄p : Λp → (Tp (Xp ), κp ) mod K⋆p .
Since we want a level structure for Tp (A)ba , we require that the ηp must respect the bilinear
forms on both sides by the same factor u ∈ Z× p . For the special prime pv we take an arbitrary
isomorphism
ηpv : Λpv → Tp (Xpv )
which respects the bilinear forms on both sides up to the same factor u ∈ Z× p . This is possible
because by (ii) of Definition 7.1.2, the OKpv -lattices of both sides are isomorphic and since, by
Lemmas 8.1.2 and 8.1.3, there exist isomorphisms with an arbitrary multiplicator u ∈ Z× p . We
set
η̃p = ηpv ηp : Λ ⊗ Zp → Tp (A).
Finally we set
η = η̃p η p : V ⊗ Af → V̂ (A).
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 129

Let η̄ be the class of this isomorphism modulo K⋆ . Then (A, ι, λ̄, η̄) is a T -valued point of AK⋆ ,E .
Since the last construction can be reversed, we obtain the isomorphism of the theorem. 
We will formulate a more precise version of the last Theorem. Let Eνc be the algebraic closure
of Ĕν . The action of the Galois group Gal(Eνc /Ĕν ) on Tp (Xp,0 ) is given by a character
×
χp,0 : Gal(Eνc /Ĕν ) → OK p
, (7.4.25)

such that σ(t) = χp,0 (σ)t for t ∈ Tp (Xp,0 ) and σ ∈ Gal(Eνc /Ĕν ). Since the polarization of Xp,0
is defined over Ĕν , we obtain that NmKp /Fp χp,0 (σ) = 1. We define
Y
χba
0 (σ) = χp,0 (σ) ∈ Gba (Qp ).
p,banal

Finally we define χ0 : Gal(Eνc /Ĕν ) → G(Qp ) by setting

χ0 (σ) = 1 × χba ba
0 (σ) ∈ Gpv × G (Qp ).

We note that this element is in the center of the group G(Qp ). By definition of the functor
AK⋆ ,E before Remark 7.1.6, χ0 (σ) acts on AK⋆ ,Ĕν = AK⋆ ,E ×Spec E Spec Ĕν via the datum (3),
i.e. it acts by Hecke operators. We obtain the homomorphism
χh0 : Gal(Eνc /Ĕν ) → Autopp AK⋆ ,Ĕν . (7.4.26)
(We write here the opposite group because the Hecke operators act by definition from the right.)
Corollary 7.4.11. Let σ ∈ Gal(Eνc /Ĕν ). Then the action of idA⋆K⋆ × Spec σ on the right hand
side of (7.4.22) induces on the left hand side the automorphism χh0 (σ) × Spec σ.
Remark 7.4.12. In general, let X a quasi-projective scheme over Ĕν . Let χ : Gal(Eνc /Ĕν ) →
Autopp X be a continuous homomorphism. Then descent says that there is a unique quasi-
projective scheme X(χ) over Ĕν and an isomorphism
X ×Spec Ĕν Spec Eνc → X(χ) ×Spec Ĕν Spec Eνc

such that, for all σ ∈ Gal(Eνc /Ĕν ), the action of idX(χ) × Spec σ on the right hand side induces
on the left hand side the action χ(σ) × Spec σ. We will call X(χ) the Galois twist of X by χ.
Proof. (of Corollary 7.4.11) We take (7.4.22) over the algebraic closure Eνc . For σ ∈ Gal(Eνc /Ĕν ),
we write σ̂ := Spec σ. We consider the non-commutative diagram
AK⋆ ,Ĕν ×Spec Ĕν Spec Eνc / A⋆
K⋆ ,Ĕ
×Spec Ĕν Spec Eνc
ν

idA ×σ̂ idA⋆ ×σ̂ (7.4.27)

 
AK⋆ ,Ĕν ×Spec Ĕν Spec Eνc / A⋆
K⋆ ,Ĕ
×Spec Ĕν Spec Eνc .
ν

To understand how this does not commute we consider more generally a scheme S of finite type
over Ĕν and write SEνc = S ×Spec Ĕν Spec Eνc . The morphism σ̂S := idS × σ̂ : SEνc → SEνc induces
maps
σA : AK⋆ ,Ĕν (SEνc ) → AK⋆ ,Ĕν (SEνc ), σA⋆ : A⋆K⋆ ,Ĕ (SEνc ) → A⋆K⋆ ,Ĕ (SEνc ).
ν ν

Our task is to compare the effect of these maps on an element ξ ∈ AK⋆ ,Ĕν (SEνc ) = A⋆K⋆ ,Ĕ (SEνc ).
ν
The moduli interpretation describes ξ as a point of AK⋆ ,Ĕν by a point (A, ι, λ, η̄ p ) ∈ AK,Ĕν (SEνc )
and a rigidification η̄p : Λba → Tp (A)ba mod K⋆,ba . To make this more precise, we choose a
geometric point ω : Spec Eνc → S which extends naturally to a point ω : Spec Eνc → SE c . We
define ω ′ by the commutative diagram
σ̂S
SEνc / SE c
O Oν
ω′ ω

σ̂
Spec Eνc / Spec Eνc .
130 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

The rigidification is given by a homomorphism

ηp : Λba → Tp (Aω )ba .
There is an isomorphism
Tp ((σ̂S∗ A))ω′ = σ̂ ∗ (Tp (Aω )).
By the moduli interpretation, the point σA (ξ) is given by (σ̂S∗ A, σ̂S∗ ι, σ̂S∗ λ, σ̂S∗ η̄ p ) and the rigidi-
fication is given by
σ̂∗ (ηp )
Λba = σ̂ ∗ (Λba ) −→ σ̂ ∗ (Tp (Aω )). (7.4.28)
Now we consider σA⋆ (ξ). We can give the sheaf Tp ((X0ba )Ĕν )
in (7.4.21) equivalently by the
Gal(Eνc /Eν )-module Λba (χba0 ), where we indicate that the Galois group acts via the character
χba
0 . Then η̄p of (7.4.21) can be considered as a class of maps
Λba (χba
0 )Eνc → Tp (Aω )
ba
mod K⋆,ba . (7.4.29)
Since we are over Eνc ,the action via χba p
is trivial and therefore (A, ι, λ, η̄ , ηp ) describes also a
0
point of A⋆K⋆ ,Ĕ (SEνc ). But if we want to identify the inverse image of this point by σ̂S we must
ν
∗ p
take into account the twist χba ⋆ ⋆ ⋆
0 . This inverse image is again given by (σ̂S A, σ̂S ι, σ̂S λ, σ̂S η̄ ) as
before, but the new rigidification at the banal places is
σ̂∗ (ηp )
Λba (χba ∼ ∗ ba ba −→ σ̂ ∗ (Tp (Aω )).
0 )Eνc = σ̂ (Λ )(χ0 )Eνc (7.4.30)
The first isomorphism comes from the fact that both sides are the inverse image of Λba (χba
0 )
considered as a sheaf on Spec Ĕν . Therefore this isomorphism is the descent datum on the
constant sheaf, which is the multiplication by χba
0 (σ). We obtain
χba (σ)
∗ ∗ σ̂∗ (ηp )
Λba (χba 0 ba ba
0 ) −→ σ̂ (Λ )(χ0 ) −→ σ̂ (Tp (Aω )).

This proves that σA∗ = χh0 (σ)σA . If we apply this to the diagram (7.4.27), we obtain
idA∗ × σ̂ = χh0 (σ)(idA × σ̂).

We now drop the assumption on K⋆p that it be contained in Kp and come from a product of
K⋆p . More precisely, let K⋆p ⊂ G(Qp ) be of the form
K⋆p = G(Qp ) ∩ Kpv K⋆,ba
p , (7.4.31)
where K⋆,ba
p
ba
is an arbitrary open compact subgroup of G (Qp ). Since Kpv is a normal subgroup
of Gpv and G(Qp ), this class of subgroups is stable under conjugation by elements of G(Qp ).
Therefore, using the naturality of the construction in Proposition 7.4.9, we can extend the
definition of A⋆K⋆ to all such K⋆ = K⋆p Kp by first passing to a small enough normal subgroup
of finite index and then dividing out by the factor group.
We make this extension process more explicit by defining the functor Â⋆K⋆ without using the
choice of Λp . Thereby the action of the Hecke operators becomes more obvious.
Let θ : X → Y be an isogeny of p-divisible OF ⊗ Zp -modules. Let pv be a prime of OF over
p. We say that θ is an isogeny of order prime to pv if θpv is an isomorphism. We use a similar
terminology for abelian varieties with action by OF .
We consider a scheme S over Spf OĔν . We consider abelian schemes Ā over S up to isogeny of
order prime to pv which are endowed with an action ι : OK → End Ā and with a Q-homogeneous
polarization λ̄ such that the Rosati involution induces the conjugation on OK . Moreover, we
assume that there is a triple (A, ι, λ) as in Definition 7.1.2 which represents (Ā, ι, λ̄) such that
(A, ι) satisfies the conditions (KCr ) and (ECr ) and such that (A, ι, λ) satisfies the conditions
(i) andQ(ii) of Definition 7.1.2. Then call (Ā, ι, λ̄) an admissible prime-to-pv -isogeny class. Let
X = p Xp be the p-divisible group of A. Then (CX ba , ξλ ) makes sense and (CX ba ⊗ Q, ξλ )
depends only on (Ā, ι, λ̄). Let (X, ιX , λX ) as in Definition 7.4.5.
Definition 7.4.13. We define a functor Â⋆K⋆ on the category of schemes S over Spf OĔν . A
point of Â⋆K⋆ (S) consists of the following data:
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 131

(1) an admissible prime-to-pv -isogeny class (Ā, ι, λ̄) over S.

(2) a class of isomorphisms
η̄p : CXba ⊗ Q → CĀ[p∞ ]ba ⊗ Q mod K⋆,ba
p ,
which respects the bilinear forms on both sides up to a factor in Q×
p.

We explain more detailed what is meant by (2). We assume that S is connected and we choose
a geometric point ω of S. Then the meaning of (2) is that we have a class of isomorphisms
ηp : (CXba )ω ⊗ Q → (CĀ[p∞ ]ba )ω ⊗ Q mod K⋆,ba
p

which respects the bilinear forms on both sides up to a factor in Q× p and such that the class is
preserved by the action of π1 (S, ω).
Let K⋆p as in Definition 7.4.5. Then the functors of the Definitions 7.4.5 and 7.4.13 coincide.
Indeed, let us start with a point of Definition 7.4.13. Also, fix a triple (A, ι, λ) which represents
(Ā, ι, λ̄), as before Definition 7.4.13. The sublattice Λba
p ⊂ CXba )ω is fixed by Kp
⋆,ba
. Therefore
ba
the image C of Λp by ηp depends only on the class η̄p and is invariant by π1 (S, ω). Therefore
C defines a p-adic étale sheaf on S. We endow it with the polarization induced by ς ba , cf.
(7.4.6). Therefore, using the contracting functor C defines a p-divisible OK ba -module Y ba with
a polarization. Then Y := Xpv × Y ba is isogenous to the p-divisible group X of A. The
polarization on Y differs from the polarization induced from λ on A by a factor in Z× p , as we
see by comparing the degrees of the polarizations. Therefore we obtain a point (A1 , ι1 , λ̄1 ) of
Â⋆K⋆ (S) which is isogenous to (Ā, ι, λ̄). This proves that the point we started with comes from
a point of the functor in Definition 7.4.5. It is clear that we have a bijection.
We have an action of Gba (Qp ) on the tower Â⋆K⋆ for varying K⋆p . This action extends to the
algebraization A⋆K⋆ and coincides via Theorem 7.4.10 with the Hecke operators on the tower
AK⋆ ,E .
Corollary 7.4.14. For every K⋆ = K⋆p Kp with (7.4.31) , there exists a normal scheme A⋆K⋆
over Spec OĔν such that for the p-adic completion of this scheme there is an isomorphism

Â⋆K⋆ ≃ J(Q)\[(Ω̂Fv ×Spf OFv Spf OĔν ) × Gba (Qp )/K⋆,ba

p × G(Apf )/Kp ].
For varying K⋆ , these schemes form a tower with an action of the group G(Qp ) × G(Apf ), where
the action of G(Qp ) factors through G(Qp ) → Gba (Qp ). The isomorphism of formal schemes is
compatible with these actions.
The general fiber of A⋆K⋆ is a Galois twist of AK⋆ ,E ×Spec E Spec Ĕν by the character χh0 , cf.
(7.4.26) and Remark 7.4.12. The Galois twist respects the Hecke operators (cf. section 7.6 for
an explicit description of χh0 ).
Proof. This is a consequence of Proposition 7.4.4 and the general pattern of p-adic uniformiza-
tion, cf. (7.3.1). The last assertion follows because χ0 : Gal(Eνab /Eν ) → G(Qp ) factors through
the center. 

7.5. The rigid-analytic uniformization. Let Arig

K denote the rigid-analytic space over Sp Eν
associated to AK,E . Then Theorem 7.3.3 implies the following corollary concerning generic
fibers.
Corollary 7.5.1. Let K = Kp Kp as in (7.1.8). There exists an isomorphism of rigid-analytic
spaces over Sp Ĕν ,


Arig ′ ′ p
K ×Sp Eν Sp Ĕν ≃ J(Q)\ ΩFv ×Sp Fv Sp Ĕν × Ĝ (Qp )/Ĝ (Zp ) × G(Af )/K
p
.
For varying Kp , this isomorphism is compatible with the action of G(Apf ) through Hecke corre-
spondences on both sides. 
Here ΩFv = P1Fv \ P1 (Fv ) is Drinfeld’s p-adic halfspace corresponding to the p-adic field Fv .
Similarly, Corollary 7.4.14 implies the following corollary concerning generic fibers for deeper
level structures.
132 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Corollary 7.5.2. Assume that there are banal primes. Let K⋆ = K⋆p Kp with (7.4.31). Let Arig K⋆
denote the rigid-analytic space over Sp Eν associated to AK⋆ ,E . There exists an isomorphism of
rigid-analytic spaces over Sp Ĕνab ,
Arig ab ab ba ∗,ba
K∗ ×Sp Eν Sp Ĕν ≃ J(Q)\[(ΩFv ×Sp Fv Sp Ĕν ) × G (Qp )/Kp × G(Apf )/Kp ].
For variable K⋆ , this isomorphism is compatible with the Hecke correspondences by G(Qp ) ×
G(Apf ). 

7.6. Determination of the character χh0 . In this section we give an explicit description of
the character χh0 (7.4.26) which is used in Corollary 7.4.14. In the case where pv is ramified in
K/F , we only obtain the restriction of χh0 to the Galois group of a quadratic extension of Ĕν .
It is enough to describe χp,0 (7.4.25) for each banal prime p. This is done by Proposition 7.6.5
below.
Let K/F be a CM-field. Let Ξ ⊂ HomQ-Alg (K, C) be a CM-type. We denote the reflex field
by H. We define an algebraic torus over Q, with Q-valued points
T (Q) = {a ∈ K × | aā ∈ Q× }.
We use the notation V = K for K regarded as a K-vector space.
We recall the reciprocity law. We define the homomorphism
Y
µ : C× → (K ⊗Q C)× ∼= C× .
ϕ:K→C

The element µ(z), for z ∈ C, has component z for ϕ ∈ Ξ and has component 1 for ϕ ∈/ Ξ on the
right hand side. We find µµ̄ = 1 ⊗ z ∈ (K ⊗Q C)× . We obtain a homomorphism of algebraic tori
µ : Gm,C → TC .
This homomorphism is defined over H,
µ : Gm,H → TH .
From this we deduce the reciprocity map
µ NmH/Q
r : ResH/Q (Gm,H ) −→ ResH/Q (TH ) −→ T. (7.6.1)
We consider over the algebraic closure H̄ = Q̄ the set of tuples (A, ι, λ̄, κ), where (A, ι) is an
abelian variety over H̄ of CM-type Ξ, endowed with a Q-homogeneous polarization λ̄ which
induces on K the conjugation over F and an isomorphism κ : V̂ (A) → V ⊗ Af of K ⊗ Af -
modules. We call a second tuple (A′ , ι′ , λ̄′ , κ′ ) equivalent to (A, ι, λ̄, κ) if there is a quasi-isogeny
α : (A, ι, λ̄) → (A′ , ι′ , λ̄′ ) (7.6.2)
such that the following diagram
α / V̂ (A′ )
V̂ (A)
❍❍
❍❍ ✉✉
❍❍ ✉✉✉
κ ❍❍ ✉✉ ′
❍$ z✉✉ κ
V ⊗ Af

commutes. We also say that (A, ι, λ̄, κ) is quasi-isogenous to (A′ , ι′ , λ̄′ , κ′ ).

Let CΞ be the set of tuples (A, ι, λ̄, κ) up to equivalence. Let σ ∈ Gal(H̄/H). Taking the
inverse image of (A, ι, λ̄, κ) by σ̂ := Spec σ : Spec H̄ → Spec H̄ gives a left action of Gal(H̄/H)
on CΞ . We denote the inverse image by σ(A, ι, λ̄, κ).
We formulate the main theorem of complex multiplication of Shimura and Taniyama.
Theorem 7.6.1. ([8, Thm. 4.19]) The Galois group Gal(H̄/H) acts on CΞ via its maximal
abelian quotient Gal(H ab /H). Let e ∈ (H ⊗ A)× and let rec(e) ∈ Gal(H ab /H) be the automor-
phism given by the reciprocity law of class field theory. The following tuples are equivalent:
rec(e)(A, ι, λ̄, κ) ≡ (A, ι, λ̄, r(ef )κ),
where ef is the finite part of the idèle e.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 133

Remark 7.6.2. Let (H × )∧ ⊂ (H ⊗ Af )× be the closure of H × . We deduce a homomorphism

r
Gal(H̄/H) → (H ⊗ Af )× /(H × )∧ −→ T (Af )/T (Q), (7.6.3)
where the first arrow is deduced from class field reciprocity and the second arrow exists because
T (Q) = T (Q)∧ . To see this last fact, we note that the group of units in T (Q) is finite. Indeed,
the units are elements of K × with all absolute values equal to 1 at all places including the infinite
ones. Therefore T (Q) = T (Q)∧ by Chevalley’s theorem.
Theorem 7.6.1 says that the action of Gal(H̄/H) on CΞ is via (7.6.3). One can consider the
Shimura variety ShT . We may choose as usual a bijection
ShT (H̄) = ShT (C) = T (Af )/T (Q)∧ .
Then the theorem may be regarded as a consequence of Langlands’ description of the reduction
of this Shimura variety at good places [15].
We fix an embedding Q̄ −→ Q̄p . The p-adic place which is induced on a subfield of Q̄ will be
denoted by ν.
Proposition 7.6.3. Let L ⊂ Q̄ be a number field such that H ⊂ L. Let (A0 , ι0 , λ0 ) be an abelian
variety over L with an action ι0 : OK → End A0 which is of CM-type Ξ. We assume that A0
has good reduction at ν. The group Gal(L̄ν /Lν ) acts on the Tate module Tp (A0 ) via its maximal
abelian quotient Gal(Lab ab
ν /Lν ). Let Iν ⊂ Gal(Lν /Lν ) be the inertia group. The action of Iν on
the Tate module can be described as follows.
The inverse of the map (7.6.1) induces a homomorphism
NmLν /Hν r−1
ρ : L×
ν −→ Hν× ⊂ (H ⊗ Qp )× −→ (K ⊗ Qp )× .
Composing ρ with the reciprocity law of local class field theory yields
rec× ρ
Iν ≃ OL ν
−→ (OK ⊗ Zp )× .
The action of an element σ ∈ Iν on the Tate-module is the multiplication by the image in the
right hand side.
Proof. We set A = A0 ⊗H H̄ with the OK -action and the induced polarization. We set Λ =
∼
OK ⊂ V and Λ̂ = OK ⊗ Ẑ. We choose a rigidification κ : T̂ (A) → Λ̂. We consider the tuple
(A, ι, λ̄, κ). Let
σ ∈ I˜ν ⊂ Gal(H̄ν /Hν ) ⊂ Gal(H̄/H)
be an element of the inertia group at ν. The image in Gal(H ab /H) corresponds to an idèle in
×
(H ⊗ A)× which has components 1 outside ν and a component eν ∈ OH ν
at the place ν. We
denote the idèle also by eν . By Theorem 7.6.1, we have a quasi-isogeny
∼ (A, ι, λ̄, r(eν )κ).
σ̂ ∗ (A, ι, λ̄, κ) =
Let us moreover assume that σ fixes the elements of L. Since (A, ι, λ̄) is defined over L, it is not
changed by σ̂ ∗ . Now we consider the product of the Tate modules for all primes,
σ̂∗ (κ)
T̂ (A) = σ̂ ∗ (T̂ (A)) −→ Λ̂.
The first identification is due to the fact that T̂ (A) is a projective limit of étale sheaves on
Spec L.
Lemma 7.6.4. Denote by T̂ (σ) the action of σ on T̂ (A). Then
σ̂ ∗ (κ)T̂ (σ) = κ.
We postpone the proof of the Lemma. Because we have good reduction, the element T̂ (σ)
acts trivially on the Tate modules Tℓ (A) for ℓ 6= p. On Tp (A) it acts by multiplication with an
element u(σ) ∈ (OK ⊗ Zp )× . Therefore we have a quasi-isogeny
∼ (A, ι, λ̄, u(σ)−1 κ)
(A, ι, λ̄, r(eν )κ) = (7.6.4)
The quasi-isogeny giving this equivalence must be trivial on the Tate modules Vℓ (A) for ℓ 6= p.
It is therefore the identity. The proposition follows therefore from Lemma 7.6.4. 
134 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Proof. (of Lemma 7.6.4) We consider an étale sheaf G over Spec L where L is any field. Let
Ls be the separable closure of L. For σ ∈ Gal(Ls /L) we denote by G(σ) : G(Ls ) → G(Ls ) the
natural action. Let Γ be a constant sheaf on Spec L associated to a set Γ. Let
κ:G→Γ
be an isomorphism of sheaves on (Spec L )ét . There are canonical isomorphisms σ̂ ∗ (G) ∼
s
= G and
σ̂ ∗ (Γ) ∼
= Γ because both sheaves are defined over L. We must show that the map
∗
σ̂ (κ)
κ′ : G(Ls ) ∼
= σ̂ ∗ (G)(Ls ) −→ σ̂ ∗ (Γ)(Ls ) = Γ
coincides with κG(σ −1 ).
Let A be a finite étale algebra over Ls . By definition of the inverse image, we have σ̂ ∗ (G)(A) =
G(A[σ] ). Therefore the Ls -algebra isomorphism σ : Ls → Ls[σ] induces a natural map G[σ] : G(Ls ) →
σ̂ ∗ (G)(Ls ). Our assertion follows from the commutative diagram, in which the composition of
the two upper horizontal arrows is G(σ),
G[σ] ∼
G(Ls ) / σ̂ ∗ (G)(Ls ) / G(Ls )
rrr
κ σ̂∗ (κ) rrr′
r
  x rr κ
r
Γ /Γr
Γ[σ]=id


Let K/F, r, E → Q̄p , ν be as in section 7.1. Let p be a banal prime of K. Let (Xp,0 , ιp,0 ) be
the unique CM-pair of CM-type rp /2 over Spec OĔν . We set
Ξp = {ϕ ∈ Φp | rϕ = 2},
where Φp = HomQp -Alg (Kp , Q̄p ), as in (7.1.9). We consider the homomorphism
Y
× ∼
µp : Q̄×p → (Kp ⊗ Q̄p ) → Q̄×
p
Φp

such that the component of µp (a), a ∈ Q̄p , is equal to a for ϕ ∈ Ξp and is 1 for ϕ ∈
/ Ξp . This
morphism is defined over Eν . We define the local reciprocity law rp as
µp NmEν /Qp
rp : Eν× −→ (Kp ⊗Q Eν )× −→ Kp× . (7.6.5)
Let Iν ⊂ Gal(Ĕνab /Ĕν ) be the inertia group. As before (7.4.25), let Eνc be the algebraic closure
of Ĕν in the completion of Q̄p . By the reciprocity law of local class field theory, we define
× rec
× r−1
ρp : Gal(Eνc /Ĕν ) → Gal(Ĕνab /Ĕν ) ≃ OE
p
ν
−→ OK p
.
×
Proposition 7.6.5. Let p be a banal prime of K. Let χp,0 : Gal(Eνc /Ĕν ) → OK p
be the character
given by the action on Tp (Xp,0 ), compare (7.4.25). Then the restriction of this character to the
subgroup
Gal(Eνc /Ĕν ϕ0 (Kpv )) ⊂ Gal(Eνc /Ĕν )
coincides with the restriction of ρp to this subgroup.
We remark that Ĕν ϕ0 (Kpv ) equals Ĕν if pv is unramified in K/F and is a quadratic extension
of Ĕν if pv is ramified in K/F .
Proof. It follows from the functoriality of rec that the proposition implies the same statement
for a finite extension Eν′ of Eν .
We define a CM-type Ξ ⊂ Φ = HomQ-Alg (K, C) by choosing ϕ0 : K → C with rϕ0 = 1 and
setting
Ξ = {ϕ ∈ Φ | rϕ = 2} ∪ {ϕ0 }.
We denote by H the reflex field of Ξ. We find that Hϕ0 (K) = Eϕ0 (K). We claim that there
exists an extension of number fields L/H which is unramified at ν and a tuple (A, ι, λ̄, κ) which
is defined over L and such that A has good reduction à over OLν . Let Y be the p-divisible group
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 135

Q
of Ã, which we write as Y = p Yp , where p runs through the prime ideals of OK over p. Let p
be banal. Then (Yp , ι)⊗OLν OL̆ν is a CM-pair of type rp /2 which satisfies the Kottwitz condition
and the Eisenstein condition. Therefore it is isomorphic to (Xp,0 , ιXp,0 ) which is defined over
OH̆ν . Therefore the proposition follows from Proposition (7.6.3).
It remains to show the existence of L. We fix an open compact subgroup C ⊂ T (Af ) which
is maximal in p and is small enough. The Shimura variety ShΞ,C which is associated to (T, µ)
and C is representable by a moduli problem AΞ,C,H which is finite and étale over H. Moreover
it has a model AΞ,C over OHν . It is defined exactly in the same way as AK . Since for the
moduli problem AΞ,C each prime p of ` OK is banal, it is representable by a finite étale scheme
m
over OHν . We conclude the AΞ,C,H = i=1 Spec Li for some finite field extensions Li /H which
are unramified over ν. Restricting the universal abelian scheme over AΞ,C,H to some L = Li ,
we obtain a tuple as required. 

8. Appendix: Adjusted invariants

In this appendix we first collect some facts about anti-hermitian forms. Then we give a
correction to [18, Prop. 3.2], by introducing the r-adjusted invariant of a CM-triple. Finally, we
relate the r-adjusted invariant to the contracting functor of section 4.
8.1. Recollections on binary anti-hermitian forms over p-adic local fields. We first
recall the invariant of an anti-hermitian form in the case relevant to us. A good reference for
this material is [13].
Let K/F be a quadratic extension of fields of characteristic 0. We denote by a 7→ ā the
non-trivial automorphism of K over F . Let V be an 2-dimensional vector space over K. Let
κ : V × V −→ K,
be a sesquilinear form which is linear in the first argument and anti-linear in the second. We
assume that κ is anti-hermitian:
κ(x, y) = −κ(y, x).
We choose a basis {v1 , v2 } of V . Then det(κ(vi , vj ))i,j∈{1,2} ∈ F × . We denote by
dK/F (V, κ) ∈ F × / NmK/F K × (8.1.1)
the residue class of this element. It is independent of the choice of the basis and is called the
discriminant of (V, κ).
Definition 8.1.1. Let F be a p-adic local field and K/F a quadratic field extension. Let (V, κ)
be a K-vector space of dimension 2 with an anti-hermitian form κ which is nondegenerate.
We denote by inv(V, κ) ∈ {±1} the image of dK/F (V, κ) under the canonical isomorphism
F × / NmK/F K × ≃ {±1}. The invariant determines (V, κ) up to isomorphism, cf. [13].
We note that an anti-hermitian form κ can equivalently be given by an alternating non-
degenerate Qp -bilinear form
ψ : V × V −→ Qp (8.1.2)
such that
ψ(ax, y) = ψ(x, āy), x, y ∈ V, a ∈ K.
The anti-hermitian form κ is defined by the equation
TrK/Qp aκ(x, y) = ψ(ax, y).
In this case we set
inv(V, ψ) = inv(V, κ).
The invariant inv(V, ψ) determines (V, ψ) up to isomorphism.
Let Λ ⊂ V be an OK -lattice such that ψ induces a pairing
ψ : Λ × Λ −→ Zp , (8.1.3)
i.e., ψ is integral on Λ. We consider the map
Λ −→ HomZp (Λ, Zp )
(8.1.4)
y 7−→ ℓy , where ℓy (x) = ψ(x, y)
136 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

This is an anti-linear map of OK -modules. Therefore the image of this map is an OK -submodule.
We denote the length of the cokernel as an OK -module by h(Λ, ψ).
Lemma 8.1.2 ([13], Thm. 7.1). Let F be a local p-adic field and K/F an unramified field
extension. Let V be a 2-dimensional K-vector space. Let
ψ : V × V −→ Qp
as in (8.1.2). Then inv(V, ψ) = 1 iff there exists an OK -lattice Λ ⊂ V such that ψ is integral on
Λ and such that h(Λ, ψ) = 0, i.e., such that ψ|Λ×Λ is a perfect pairing. Moreover, Λ is uniquely
determined up to Aut(V, ψ).
Similarly, inv(V, ψ) = −1 iff there exists an OK -lattice Λ ⊂ V , such that ψ is integral on Λ
and such that h(Λ, ψ) = 1. Moreover, Λ is uniquely determined up to Aut(V, ψ). In this case,
ψ|Λ×Λ is called almost perfect.
Proof. This reduces to the analogous statement for the anti-hermitian form ψ̃ : V × V −→ K
defined by
t(ξ ψ̃(x1 , x2 )) = ψ(ξx1 , x2 ), x1 , x2 ∈ V, ξ ∈ K.
where t : K −→ Qp is defined by t(a) = trK/Qp (ϑ−1 a), where ϑ denotes the different of K/Qp .
Then it follows from loc. cit. 
Lemma 8.1.3 ([13], Prop. 8.1 a)). Let p = 6 2, and let F be a local p-adic field and K/F a
ramified quadratic field extension. Let V be a 2-dimensional K-vector space. Let
ψ : V × V −→ Qp .
as in (8.1.2). Then there exists an OK -lattice Λ ⊂ V such that ψ induces a perfect form
ψ : Λ × Λ −→ Zp .
Moreover Λ is unique up to Aut(V, ψ). 
8.2. The r-adjusted invariant. Let K be a CM-field, with totally real subfield F . We set
Φ = HomQ-Alg (K, Q̄). Let r be a generalized CM-type of rank n, i.e., rψ + rψ̄ = n. Let E = Er
be the reflex field, cf. [18, §2]. A CM-triple over an OE -algebra R is a triple (A, ι, λ) where A is
an abelian scheme over R with an action ι : OK → End A with satisfies the Kottwitz condition
(KCr ) and a polarization λ whose Rosati involution induces the conjugation of K/F . In the case
n = 2 this is a CM-triple with satisfies the Kottwitz condition. Let v be a place of F . We define
an r-adjusted invariant invrv (A, ι, λ) attached to a triple (A, ι, λ) of CM-type r, defined over a
field k that is at the same time an OE -algebra. When v is non-archimedean split in K, then
invrv (A, ι, λ) = invv (A, ι, λ) = 1. If v is archimedean, or non-archimedean non-split in K, with
residue characteristic of v different from the characteristic of k, then invrv (A, ι, λ) = invv (A, ι, λ),
i.e., the adjusted invariant coincides with the invariant of [18, §3]. Comp. section 2.4 for the
definition of the latter invariant for n = 2. The case of general n is substantially the same.
Now let v be non-split with residue characteristic equal to the characteristic p of k. We may
assume that k is algebraically closed. Let us first assume that the OE -algebra structure of k is
induced by a OQ -algebra structure. Let ν̃ be the induced p-adic place of Q. Let
Φv = {ϕ : K −→ Q | ν̃ ◦ ϕ induces v }. (8.2.1)
Then
Φv = HomQp (Kv , Qν̃ ).
Also let
rv = r|Φv .
Now define
invrv (A, ι, λ) = invv (A, ι, λ) sgn(rv ), (8.2.2)
with P P
(nd − + r ) 1
+ (r −r )
sgn(rv ) = (−1) 2 v ϕ∈Φv
ϕ
= (−1) 2 ϕ∈Φv ϕ ϕ̄ . (8.2.3)
+
Here Φv is a half-system of embeddings in Φv , which has cardinality dv = [Fv : Qp ]. Since
rϕ + rϕ = n for all ϕ ∈ Φv , and n is supposed to be even, (8.2.3) is independent of Φ+
v . Note
that sgn(rv ) only depends on the place ν of E induced by νe.
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 137

The correct version of [18, Prop. 3.2] is now as follows.

Proposition 8.2.1. Let S be an OE -scheme. Let (A, ι, λ) be a CM-triple over S which satisfies
(KCr ). Let c ∈ {±1}. Then for every place v of F , the set of points s ∈ S such that
invrv (As , ιs , λs ) = c
is open and closed in S.
Proof. Clearly we may assume that S is an OE -scheme of finite type. Further we can assume
that S is irreducible. Obviously the invariant is constant on the generic fiber of S. Also, we may
assume that v is non-archimedean non-split in K.
First we consider the case when S is an irreducible scheme of finite type over κEν . Since each
local ring of S is dominated by a discrete valuation ring R, it is enough to consider the case
S = Spec R. We may replace R by a discrete valuation ring that dominates R. Therefore we
can assume the R is complete with algebraically closed residue class field, i.e., R ∼ = k[[t]] for an
algebraically closed field
Q k. According to the action of F ⊗ Q p , the p-divisible group X of A is
isogenous to a product w|p Xw . We consider the factor Xv . Let P be the display of Xv over R,
cf. (3.1.9). We note that P is the value of the crystal of Xv at the pd-thickening W (R)/R. By
Lemma 8.2.2 below, there is an element x ∈ ∧nOK ⊗Z W (R) P such that F x = pn/2 x. We define
v p
the anti-hermitian form
κ : PQ × PQ → Kv ⊗Zp W (R)
as in (2.4.3). We consider the hermitian form h = ∧nOK ⊗Z W (R) κ on ∧nOK ⊗Z W (R) PQ . From
v p v p
the equation
h(F y1 , F y2 ) = pn F h(y1 , y2 ), y1 , y2 ∈ ∧nOKv ⊗Zp W (R) PQ ,
we obtain that h(x, x) lies in the invariants (Kv ⊗Zp W (R))F = Kv . Because h is hermitian, we
obtain h(x, x) ∈ Fv . The element x can be used to determine the invariant of the Dieudonné
module P ⊗W (R) W (L) obtained for arbitrary base change R → L to a perfect field. Therefore
invv (As , ιs , λs ) = invv (Aη , ιη , λη ) and invrv (As , ιs , λs ) = invrv (Aη , ιη , λη ), where s and η denote
the special and the generic point of Spec R. For the comparison with the definition of the
invariant of a Dieudonné module we should remark that the equations F x = pn/2 x and V x =
pn/2 x are equivalent because F V = pn on ∧nOK ⊗Z W (R) P .
v p
Now we consider the case when the function field of S has characteristic 0. This case can be
reduced to the case when S = Spec OL , where L is the completion of a subfield of Qp which
contains E and such that its ring of integers O = OL is a discrete valuation ring with residue
field Fp . We denote by AL the generic fiber of A, and by Ak its special fiber.
We decompose the rational p-adic Tate module of AL , resp. the rational Dieudonné module
of Ak , with respect to the actions of F ⊗ Qp ,
M M
Vp (AL ) = Vw (AL ), M (Ak )Q = M (Ak )Q,w .
w|p w|p

Here Vw (AL ) is a free K ⊗F Fw -module of rank n, and M (Ak )Q,w is a free K ⊗F Fw ⊗Qp W (k)Q -
module of rankV n. Set Q̆p = W (k)Q . V
Let Sv = nKv Vv (AL ) and NQ,v = nKv M (Ak )Q,v . Both are equipped with hermitian forms
(for the first module, cf. [18, section 3, case b)]; for the second module, cf. subsection 2.4). Also,
we have NQ,v = 1v ( n2 ), where 1v is a multiple of the unit object in the category of Dieudonné
modules, comp. (2.4.5), or Lemma 8.2.2. Let Uv be the image under the Fontaine functor of
NQ,v (− n2 ). We need to compare the two hermitian vector spaces Sv (− n2 ) and Uv .
Let T be the torus over Qp which is the kernel of the map defined by the norm of Kv /Fv ,
1 −→ T −→ ResKv /Qp Gm,Kv −→ ResFv /Qp Gm,Fv −→ 1.
Then H 1 (Qp , T ) = Fv× / Nm(Kv ). We may regard the isomorphisms of hermitian vector spaces
Isom(Uv , Sv (− n2 )) as an etale sheaf on Spec Fv . This is a T -torsor. Its class cl(Uv , Sv (− n2 )) is
calculated by [27, Prop. 1. 20].
To evaluate this formula, note that the first summand, κ(b), in loc. cit. is trivial. To evaluate
the second summand, µ♯ , we use the following description of the filtration on NQ,v ⊗Q̆p Q̆p . For
138 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

the filtration of M (Ak )Q,v ⊗Q̆p Q̆p = ⊕ϕ∈Φv M (Ak )Q,v,ϕ we have that the jumps are in degree 0
and 1, with
(0) ⊂ Fil1ϕ ⊂rϕ M (Ak )Q,v,ϕ . (8.2.4)
The upper index means that the cokernel has dimension rϕ . For the filtration of the one-
dimensional vector space NQ,v,ϕ , this means that the unique jump is in degree n − rϕ . We use
the identification

X∗ (T ) = Ker IndK Fv Fv
Fv (IndQp (Z)) −→ IndQp (Z) .
v

Then the corresponding filtration on N (A)Q,v,ϕ (− n2 ) is given by the cocharacter µ ∈ X∗ (T ) with

n
µϕ = − rϕ , ϕ ∈ Φv . (8.2.5)
2
We have to determine the image µ♯ of µ in X∗ (T )Γ . Under the identification X∗ (T )Γ =
H 1 (Qp , T ) = Z/2, we obtain
n X n X
cl(Uv , Sv (− )) = µ♯ = µϕ = dv − rϕ , (8.2.6)
2 +
2 +
ϕ∈Φv ϕ∈Φv

where we used the notation introduced for (8.2.3). We deduce invrv (Ak , ιk , λk ) = invv (AL , ιL , λL ),
as desired. 
In the proof of Proposition 8.2.1, we used the following lemma.
Lemma 8.2.2. Let F/Qp be a finite field extension of degree d and K/F be a quadratic field
extension. Let n be an even natural number. Let k be an algebraically closed field of characteristic
p. Let (X, ι) be a p-divisible group over k[[t]] of dimension nd and height 2nd with an action
OK → End X. Let (P, ι) be the display of X, cf. (3.1.9). Then there exists a non-zero element
x ∈ ∧nOK ⊗Z W (k)[[t]] P such that
p

∧n F (x) = pn/2 x.
Proof. We consider the Zp -frame Bk = (W (k)[[t]], pW (k)[[t]], k[[t]], σ, σ̇), where σ is the extension
of the Frobenius on W (k) to the power series ring given by σ(t) = tp and where σ̇ = (1/p)σ.
The evaluation P1 of the crystal of X at the pd-thickening W (k)[[t]]/k[[t]] has the structure of
a Bk -display. The display P is obtained by base change with respect to a morphism of frames
Bk → W(k[[t]]), cf. [34] and [21]. Therefore, it is enough to prove our assertion for the Bk -display
of X which we will now denote by P.
We consider first the case when K/F is ramified. When writing detW (k)[[t]] F , we mean
this with respect to an arbitrary W (k)[[t]]-basis of P . This determinant is well determined
up to multiplication with a unit in W (k)[[t]]. We know that detW (k)[[t]] F = phd u1 for some
u1 ∈ (W (k)[[t]])× .
We consider the decomposition P = ⊕Pψ according to
Y
OK ⊗ W (k)[[t]] = OK ⊗O t ,ψ̃ W (k)[[t]].
F
ψ

The Frobenius is graded, F : Pψ → Pψσ . We conclude that detW (k)[[t]] (F f |Pψ ) = pnd u2 for
some unit u2 ∈ (W (k)[[t]])× . We fix ψ. Up to a unit we have
NmK/F t detOK ⊗O ,ψ̃ W (k)[[t]]
(F f |Pψ ) = detW (k)[[t]] (F f |Pψ ). (8.2.7)
Ft

We fix a normal extension L of W (k)Q which contains K ⊗O t ,ψ̃ W (k). The left hand side of
F
(8.2.7) is the product of conjugates c1 , . . . , c2e ∈ OL [[t]] of detOK ⊗O ,ψ̃ W (k)[[t]] (F f |Pψ ). These
Ft
elements have the same order with respect to the prime element ωL of L which is a prime element
in the regular local ring OL [[t]]. We rewrite (8.2.7)
c1 · c2 · . . . · c2e = pnd · u3 ,
for some unit u3 . Since OL [[t]] is factorial, we find ci = pf n/2 µi for some units µi . We conclude
that
detOK ⊗O ,ψ̃ W (k)[[t]] (F f |Pψ ) = pf n/2 u4
Ft
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 139

for some unit u4 ∈ (OK ⊗O t ,ψ̃ W (k)[[t]])× . Since Pψ is a free OK ⊗O t ,ψ̃ W (k)[[t]]-module
F F
of rank n, we find that, for each element yψ ∈ ∧nOK ⊗ W (k)[[t]] Pψ , there is an equation
O t ,ψ̃
F
∧n F f yψ = pf n/2 u(yψ )yψ for some unit u(yψ ) ∈ OK ⊗O t ,ψ̃ W (k)[[t]]. On the last ring, σ f acts
F
via the second factor. There is a unit ζ ∈ OK ⊗O t ,ψ̃ W (k)[[t]] such that
F

f −1
σ (ζ)ζ = u(yψ ).
Indeed, consider the image ū of u(yψ ) in OK ⊗O t ,ψ̃ W (k) by setting t = 0. It is well-known
F
that in this ring σ f (ζ̄)ζ̄ −1 = ū is solvable. One can lift ζ̄ successively modulo tn to a solution ζ.
Then xψ = ζyψ satisfies
∧n F f xψ = pf n/2 xψ .
We define xψσi ∈ PQ by ∧n F i xψ = pin/2 xψσi for i = 1, . . . , f . Then x = (xψ ) ∈ P ⊗ Q satisfies
∧h F (x) = pn/2 x. Multiplying by a power of p we can arrange that x ∈ P .
The proof in the unramified case is almost the same. We indicate the differences. In this case
HomQp -Alg (F t , W (k)Q ) has 2f elements. Therefore we have the equation (8.2.7) with f replaced
by 2f ,
detOK ⊗O ,ψ̃ W (k)[[t]] (F 2f |Pψ ) = pf n u4 .
Ft
n i in/2
We define xψσi ∈ PQ by ∧ F xψ = p xψσi for i = 1, . . . , 2f . Then x = (xψ ) ∈ P ⊗ Q satisfies
∧h F (x) = pn/2 x. 
Remarks 8.2.3. (i) The remarks and results on a product formula at the end of §3 of [18] become
correct when the invariants invv (A, ι, λ) are replaced by the adjusted invariants invrv (A, ι, λ).
(ii) In the definition of Mr,h,V in [18, (4.3)], the invariants invv (A, ι, λ) have to be replaced
by the adjusted invariants invrv (A, ι, λ).
(iii) One defines in the obvious way the r-adjusted invariant invr (X, ι, λ) of a local CM-triple
of type r, (X, ι, λ), over a field of characteristic p.
8.3. r-adjusted invariant and the contracting functor. In this subsection, we return to
the situation in section 2.1. We assume that K/F is a field extension. Let k be an algebraically
closed field of characteristic p with an OF -algebra structure, i.e., k ∈ NilpOF .
We consider the case where r is special. Consider an object (Pc , ιc , βc ) ∈ dRpol
k cf. Definition
4.4.10. We write Pc = (Pc , Fc , Vc ) for the corresponding WOF (k)-Dieudonné module. To avoid
too many double notations we denote the Frobenius automorphism on WOF (k) by τ . The
Verschiebung on WOF (k) is then πτ −1 . For our purposes it is more convenient to allow quasi-
polarizations, i.e., βc is a WOF (k)-bilinear form
Pc ⊗ Q × Pc ⊗ Q −→ WOF (k)Q ,
t
dRpol
such that (Pc , ιc , p βc ) ∈ k for large enough t ∈ Z. Then βc is alternating and the following
equations hold:
βc (Fc u1 , Fc u2 ) = πτ (βc (u1 , u2 )), u1 , u2 ∈ Pc ⊗ Q,
βc (ιc (a)u1 , u2 ) = βc (u1 , ιc (ā)u2 ), a ∈ K.
The polarization βc defines an anti-hermitian form
κc : Pc ⊗ Q × Pc ⊗ Q −→ K ⊗OF WOF (k), (8.3.1)
by the formula
TrK/F aκc (u1 , u2 ) = βc (au1 , u2 ), a ∈ K ⊗OF WOF (k), u1 , u2 ∈ Pc ⊗ Q.
We note that, by Lemma 3.1.15, Pc ⊗ Q is a free K ⊗OF WOF (k)-module of rank two.
Since Lie Pc has dimension 2, we have ordπ det WOF (k)(Vc |Pc ) = 2. We recall that, for an
arbitrary K ⊗OF WOF (k)-linear map Vc♯ : Pc ⊗ Q −→ Pc ⊗ Q,
NmK/F det K⊗OF WOF (k) (Vc♯ |Pc ⊗ Q) = detWOF (k) (Vc♯ |Pc ⊗ Q).
We conclude that
ordΠ detK⊗OF WOF (k) (Vc |Pc ) = 2, K/F ramified,
(8.3.2)
ordπ detK⊗OF WOF (k) (Vc |Pc ) = 2, K/F unramified.
140 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

With our convention π = Π in the unramified case, this is the same formula.
Let
^2
Hc = Pc ⊗ Q.
K⊗OF WOF (k)

This is a free K ⊗OF WOF (k)-module of rank 1. There is an element xc ∈ Hc such that
∧2 Vc xc = πxc . (8.3.3)
The existence of xc follows from (8.3.2) and the fact that the τ -conjugacy class of an element
ξ ∈ K ⊗OF WOF (k) is determined by its order, compare (2.4.2).
The anti-hermitian form κc induces on Hc an hermitian form
hc = ∧2 κc : Hc × Hc −→ K ⊗OF WOF (k).
We find

hc (∧2 Vc x1 , ∧2 Vc x2 ) = π 2 τ −1 (hc (x1 , x2 )), (8.3.4)

where τ acts on K ⊗OF WOF (k) via the second factor. Using (8.3.3) this implies
hc (xc , xc ) ∈ F × ⊂ K ⊗OF WOF (k).
The following definition is analogous to (2.4.7).
Definition 8.3.1. The invariant inv(Pc , ιc , βc ) ∈ {±1} is defined as the image of hc (xc , xc ) by
the canonical map
∼
F × −→ F × / NmK/F K × −→ {±1},
The following proposition relates this invariant with the invariant (2.4.7) under the contracting
functor.
Proposition 8.3.2. Let K/F be a field extension and let r be special. Recall the reflex field E
associated to r. Let k ∈ N ilpOĔ be an algebraically closed field. Let (P, ι, β) ∈ dPpol
r,k and let
(Pc , ιc , βc ) ∈ dRpol
k be its image by the contracting functor Cpol
r,k , cf. (4.4.14). Then

invr (P, ι, β) = inv(Pc , ιc , βc ).

Here the r-adjusted invariant is given by
invr (P, ι, β) = (−1)d−1 inv(P, ι, β).
Proof. The second assertion follows from the definition of sgn(r), cf. (8.2.3). Let us prove the
first assertion.
We begin with the ramified case. We have the decomposition P = ⊕ψ Pψ , cf. (4.3.6). By the
definition of the contracting functor for Dieudonné modules, we have
Pc = Pψ0 , Vc = Π−ef +1 V,
cf. Remark 4.4.12. The bilinear form β̃c on Pc is the restriction of β̃ of Proposition 4.4.5. Since
we may change β̃ by a factor in F × without changing the invariant, we may replace β̃ by ϑ−1 β̃,
i.e., we may assume that TrF/Qp β̃ = β. We define the anti-hermitian form
κ : P ⊗ Q × P ⊗ Q −→ K ⊗Zp W (k)
by TrK/F κ = β̃. On
2
^
H= P ⊗Q (8.3.5)
K⊗W (k)
we obtain the hermitian form h = ∧2 κ. We have the decomposition
M 2
^ M
H= Pψ ⊗ Q = Hψ .
ψ K⊗O W (k) ψ
F t ,ψ̃

The hermitian form h is the orthogonal sum of the induced forms

hψ : Hψ × Hψ −→ K ⊗O W (k).
F t ,ψ̃
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 141

To determine inv(Pc , ιc , βc ), we consider κc defined by TrK/F κc = βc and the hermitian form

hc = ∧2 κc on Hc = Hψ0 . The form hψ0 coincides with the hermitian form deduced from the
f ˜
form β̃c above. By definition βc = η0.k βc , cf. (4.4.21). Hence we have
2f
hc = η0,k h ψ0 .
We choose an element x ∈ H such that
∧2 V (x) = px.
Let xψ0 be the ψ0 -component of x. Then inv(P, ι, β) is given by the element
hψ0 (xψ0 , xψ0 ) ∈ F × .
−f
We set zψ0 = η0,k xψ0 . Then we find
2f −f −f
hc (zψ0 , zψ0 ) = η0,k hψ0 (η0,k xψ0 , η0,k xψ0 ) = hψ0 (xψ0 , xψ0 ).
From Vc = Π−d+1 V f and ∧2 V f xψ0 = pf xψ0 , we obtain
∧2 Vc (xψ0 ) = (−1)d−1 π(p/π e )f xψ0
−f f
∧2 Vc (zψ0 ) = τ −1 (η0,k )(−1)d−1 π(p/π e )f η0,k zψ0 = (−1)d−1 πzψ0 .
By Lemma 8.3.3 below, hc (zψ0 , zψ0 ) ∈ F × defines (−1)d−1 inv(Pc , ιc , βc ).
We consider now the unramified case. As before, we have H with its hermitian form h, cf.
(8.3.5). We consider the decomposition
M 2
^  M
H= Pψ ⊗ Q = Hψ , (8.3.6)
ψ K⊗O ,ψ̃ W (k)
ψ
Kt

which has now 2f summands. Now Hψ1 and Hψ2 are orthogonal for ψ1 6= ψ̄2 . We denote by
hψ : Hψ × Hψ̄ −→ K ⊗Ot W (k)
K ,ψ̃

the sesquilinear form induced by h. Let x = (xψ ) ∈ H such that ∧2 V (x) = px or, equivalently,
∧2 V (xψ ) = pxψσ−1 for all ψ. The invariant of (P, ι, β) is the class in F × / NmK/F K × of
hψ (xψ , xψ̄ ) ∈ F ⊂ K ⊗O t ,ψ̃ W (k). This is independent of ψ. Equivalently, we can consider
K
ordπ hψ (xψ , xψ̄ ) ∈ Z/2Z. Note that ordπ makes sense for each element of K ⊗O t ,ψ̃ W (k).
K
The invariant of (Pc , ιc , βc ) is defined by Hc = Hψ0 ⊕ Hψ̄0 and hc , via
ordπ hc (xc,ψ0 , xc,ψ̄0 ), (8.3.7)
where xc = (xc,ψ0 , xc,ψ̄0 ) ∈ Hψ0 ⊕ Hψ̄0 is the element of (8.3.3). We note that we can change hc
and the elements xc,ψ0 , resp. xc,ψ̄0 , by a unit in K ⊗O t ,ψ̃ W (k) without changing (8.3.7). In
K
particular, (8.3.7) is equal to ordπ hψ0 (xc,ψ0 , xc,ψ̄0 ).
For an element y = (yψ0 , yψ̄0 ) ∈ Hc we obtain from (4.4.3)
∧2 Vc (yψ0 ) = π −2gψ̄0 ∧2 V (yψ0 ), ∧2 Vc (yψ̄0 ) = π −2gψ0 ∧2 V (yψ̄0 ).
We set
zψ0 = π −gψ0 xψ0 , zψ̄0 = π −gψ̄0 xψ̄0 .
We find
∧2 Vc (zψ0 ) = π −gψ0 π −2gψ̄0 pf xψ̄0 = π −gψ0 −gψ̄0 pf zψ̄0 ,
∧2 Vc (zψ̄0 ) = π −gψ̄0 −gψ0 pf zψ0 .
We have ordπ (π −gψ̄0 −gψ0 pf ) = 1. Therefore we obtain an element xc as in (8.3.3) if we change
zψ0 and zψ¯0 by a unit, cf. Lemma 8.3.5 below. Therefore the invariant of (Pc , ιc , βc ) is
ordπ hψ0 (zψ0 , zψ̄0 ) = (−gψ0 − gψ̄0 ) + ordπ hψ0 (xψ0 , xψ̄0 ) = (1 − d) + ordπ hψ0 (xψ0 , xψ̄0 ).
This proves the unramified case. 
In the previous proof, we used two lemmas which we state as Lemmas 8.3.3 and 8.3.5.
142 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

Lemma 8.3.3. Let K/F be ramified and let r be special. Let yc ∈ Hc be an element such that
∧2 Vc (yc ) = −πyc .
Then hc (yc , yc ) ∈ F × and the image of this element in {±1} is −inv(Pc , ιc , βc ).
Proof. We choose an element ζ ∈ WOF (k)× such that
τ −1 (ζ)ζ −1 = −1.
Then τ 2 (ζ) = ζ and therefore ζ ∈ OF ′ ⊂ WOF (k) where F ′ /F is the unramified extension of
degree 2. More explicitly, we take an element c ∈ κF ′ such that τ (c) = −c and define ζ = [c] to
be the Teichmüller representative.
We set xc = ζyc . Then the equation (8.3.3) is satisfied. We find
hc (xc , xc ) = ζ 2 hc (yc , yc ). (8.3.8)
Since ζ 2 mod π = c2 ∈ κF is not a square in this field, we conclude that ζ 2 is not in the image
×
of NmK/F : OK −→ OF× since the norm is the square on the residue fields. Therefore the image
of the right hand side of (8.3.8) in {±1} is different from the image of hc (yc , yc ). 

The last lemma has the following variant which we need in the banal case.
Lemma 8.3.4. Let K/F be ramified and let r be arbitrary. Let (P, ι, β) be a CM-triple of type
r over an algebraically closed field k. Let y ∈ ∧2K⊗W (k) PQ be an element such that

∧2 V (y) = −py.
Set h = ∧2 κ. Then h(y, y) ∈ F ⊂ F ⊗ W (k) and
h(y, y) ≡ (−1)f inv(P, ι, β) mod NmK/F K × .
Proof. We consider the decomposition
Y
OF ⊗ W (k) = OF ⊗O W (k).
F t ,ψ̃
ψ

We denote by σ the Frobenius acting on W (k). It induces via any of the embeddings ψ̃ the
Frobenius σ ∈ Gal(F t /Qp ). The decomposition induces a decomposition P = ⊕ψ Pψ and
∧2K⊗W (k) PQ = ⊕ψ ∧2K⊗ Pψ,Q ,
O t ,ψ̃ W (k)
F

which is orthogonal with respect to h. By restriction of h, we obtain

hψ : ∧2K⊗ Pψ,Q × ∧2K⊗ Pψ,Q −→ K ⊗O W (k).
O t ,ψ̃ W (k) O t ,ψ̃ W (k) F t ,ψ̃
F F

We find ζ ∈ OF ⊗ W (k) such that σ −1 (ζ)ζ −1 = −1 or equivalently σ(ζ) = −ζ. We set x = ζy.
Then we find
∧2 V (x) = σ −1 (ζ) ∧2 V (y) = −σ −1 (ζ)py = −σ −1 (ζ)ζ −1 px = px
Therefore inv(P, ι, β) is the class of
h(ζy, ζy) = ζ 2 h(y, y) mod NmK/F K × . (8.3.9)
This shows in particular that h(y, y) ∈ F × because ζ 2 ∈ F × . We can replace in (8.3.9) the left
hand side by ζψ hψ (y, y) which gives for all ψ the same element of F . The equation σ(ζ) = −ζ
may be written as σ(ζψ ) = −ζψσ . If we choose for a given ψ an element ζψ ∈ OF ⊗OF t W (k)
such that σ f (ζψ ) = (−1)f ζψ , we obtain from this element a unique ζ.
In the case where f is even, we can choose ζψ = 1, which proves the Lemma in this case. If
f is odd, we obtain that ζ ∈ F ′ \ F . This implies as in the last Lemma that ζ 2 ∈
/ NmK/F K × .
This proves the case where f is odd. 

The following fact is well-known.

 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 143

Lemma 8.3.5. Let K/F be unramified. Let u ∈ OK ⊗O t ,ψ̃0 W (k) be a unit. Then there exists
F
a unit ζ ∈ OK ⊗O t ,ψ̃0 W (k) such that
F

σ −f (ζ) · ζ −1 = u.

Proposition 8.3.6. Let r be banal, and let K/F be a field extension. Let R = k be an alge-
braically closed field. Let (CP , ι, φ) be the image of (P, ι, β) ∈ dPpol
r,k by the polarized contraction
functor Cpol
r,k , cf. Theorem 4.5.11. Then

inv(CP , ι, φ) = invr (P, ι, β).

Here the r-adjusted invariant is given by
invr (P, ι, β) = (−1)d inv(P, ι, β).
Proof. We begin with the ramified case. We choose κ̄E ⊂ k. Let F ηη −1 = π e /p = ρ, η ∈
OF ⊗ W (κ̄E ) as in (4.5.19). We define β̃ : P × P −→ OF ⊗ W (k) by (4.5.8) and the anti-
hermitian form κ : PQ × PQ −→ K ⊗ W (k) by Tr κ = β̃. This κ differs from the κ of (2.4.3) by
a constant in F . We can use it to compute invr (P, ι, β). We set
β̃ ′ = η β̃, κ ′ = ηκ.
We set V ′ = Π−e V . Then we have CP = {y ∈ P | V ′ y = y}, cf. Remark 4.5.13. From this, one
deduces
ρ · F β̃(y1 , y2 ) = β̃(y1 , y2 ), y1 , y2 ∈ CP ,
cf. (4.5.9). This implies
F ′
β̃ (y1 , y2 ) = β̃ ′ (y1 , y2 )
The restriction of β̃ ′ to CP is the form φ, cf. Remark 4.5.13.
We choose an element x ∈ ∧2 PQ := ∧2K⊗W (k) PQ such that ∧2 V (x) = (−1)e px. By Lemma
8.3.4 the class of ∧2 κ(x, x) ∈ F × / NmK/F K × = {±1} is (−1)f e inv(P, ι, β) = invr (P, ι, β).
We note that ∧2 V ′ = (−1)e π −e ∧2 V . We set
z = η −1 x ∈ ∧2 PQ .
Then we find
F −1 F −1 F −1
∧2 V ′ (z) = (η −1 )(−1)e π −e ∧2 V (x) = (η −1 )(−1)e π −e (−1)e px = (η −1 )ηπ −e pz = z.
Therefore z ∈ ∧2K CP ⊗ Q. The invariant inv(CP , ι, φ) is given by ∧2 κ ′ (z, z). Therefore the
equality of invariants follows from
∧2 κ ′ (z, z) = η 2 ∧2 κ(η −1 x, η −1 x) = ∧2 κ(x, x).
Now we consider the case where K/F is unramified. We use the notation H, h, Hψ , hψ from
(8.3.6). We have by (4.5.13) that
∧2OK CP = {z ∈ H | ∧2 V (z) = πr2 z}.
Using the decomposition (8.3.6), the condition for z = (zψ ) becomes
∧2 V (zψσ ) = π 2aψ zψ .
We choose z 6= 0. Then
inv(CP , ι, φ) = (−1)ordπ hψ (zψ ,zψ̄ ) ,
for any ψ. We set gψ = aψ + aψσ + . . . aψσf −1 . Then we obtain
gψσ − gψ = aψσf − aψ = aψ̄ − aψ = e − 2aψ .
We set uψ = π gψ zψ . Then u = (uψ ) satisfies
∧2 V (u) = π e u.
Indeed,
∧2 V (uψσ ) = π gψσ ∧2 V (zψσ ) = π gψσ π 2aψ zψ = π gψσ π 2aψ π −gψ uψ = π e uψ .
144 STEPHEN KUDLA, MICHAEL RAPOPORT, AND THOMAS ZINK

F −1
Let again η ∈ OF ⊗ W (κ̄E ) be the element defined after (4.5.19). It satisfies η −1 η = (p/π e ).
We set x = ηu. Then ∧2 V (x) = px. Therefore
inv(P, ι, β) = (−1)ordπ hψ (xψ ,xψ̄ ) ,
for any ψ. Therefore the unramified case follows from
ordπ hψ (xψ , xψ̄ ) = ordπ hψ (ηπ gψ zψ , ηπ gψ̄ zψ̄ ) = (gψ +gψ̄ )+ordπ hψ (zψ , zψ̄ ) = ef +ordπ hψ (zψ , zψ̄ ).


References
[1] T. Ahsendorf, Ch. Cheng, Th. Zink, O-displays and π-divisible formal O-modules, Journal of Algebra 457
(2016), 129-193. 8, 20, 21, 22, 23, 28, 32, 37, 80, 92
[2] N. Bourbaki, Éléments de mathématique. Algèbre commutative. Chapitre 10. Reprint of the 1998 original.
Springer-Verlag, Berlin, 2007. 29
[3] J.-F. Boutot, Uniformisation p-adique des variétés de Shimura, Séminaire Bourbaki, Vol. 1996/97.
Astérisque 245 (1997), Exp. No. 831, 307–322. 2
[4] J.-F. Boutot, H. Carayol, Uniformisation p-adique des courbes de Shimura: les théorèmes de Cherednik
et de Drinfeld, in: Courbes modulaires et courbes de Shimura (Orsay, 1987/1988). Astérisque 196-197
(1991), 45–158. 88
[5] J.-F. Boutot, T. Zink, The p-adic uniformization of Shimura curves, preprint 95–107, Univ. Bielefeld
(1995). 2
[6] H. Carayol, Sur la mauvaise réduction des courbes de Shimura, Compositio math. 59 (1986), 151-230.
[7] I. V. Cherednik, Uniformization of algebraic curves by discrete arithmetic subgroups of PGL2 (kw ) with
compact quotient spaces, (Russian) Mat. Sb. (N.S.) 100(142) (1976), no. 1, 59–88, 165. 2
[8] P. Deligne, Travaux de Shimura, Sém. Bourbaki 1970/71, exposé 389, Springer Lecture Notes 244 (1971).
110, 132
[9] V. G. Drinfeld, Coverings of p-adic symmetric domains, (Russian) Funkcional. Anal. i Prilozen. 10 (1976),
no. 2, 29–40. 2, 6, 8, 9, 20, 28, 76, 78, 80, 82, 87
[10] V. G. Drinfeld, Varieties of modules of F -sheaves, (Russian) Funkcional. Anal. i Prilozen. 21 (1987), no.
2, 23–41. 69
[11] G. Faltings, Group schemes with strict O-action, Mosc. Math. J. 2 (2002), no. 2, 249–279. 8, 45
[12] J. Gray, Poincaré’s uniformisation theorem, Aventures de l’analyse de Fermat à Borel, 595–625, Hist.
Géom., Presses Univ. Nancy/Ed. Univ. Lorraine, Nancy, 2012. 1
[13] R. Jacobowitz, Hermitian forms over local fields, Amer. J. Math. 84 (1962), 441–465. 135, 136
[14] D. Kirch, Construction of a Rapoport-Zink space for GU(1, 1) in the ramified 2-adic case, Pac. J. Math.
293 (2018), 341–389. 4, 9, 76, 90
[15] R. E. Kottwitz, Points on some Shimura varieties over finite fields, J. Amer. Math. Soc. 5 (1992), no. 2,
373–444. 113, 115, 133
[16] R. E. Kottwitz, Isocrystals with additional structure. II, Compositio Math. 109 (1997), no. 3, 255–339.
18
[17] S. Kudla, M. Rapoport, An alternative description of the Drinfeld p-adic half-plane, Annales de l’Institut
Fourier 64, no. 3 (2014), 1203–1228. 7, 9, 76, 84, 92
[18] S. Kudla, M. Rapoport, New cases of p-adic uniformization, Astérisque 370 (2015), 207–241. 3, 5, 7, 9,
10, 15, 16, 18, 109, 135, 136, 137, 139
[19] E. Lau, Smoothness of the truncated display functor, J. Amer. Math. Soc. 26 (2013), 129–165. 23, 29
[20] E. Lau, Frames and finite group schemes over complete regular local rings, Doc. Math. 15 (2010), 545–569.
20, 23
[21] E. Lau, Displays and formal p-divisible groups, Invent. Math. 171 (2008), 617–628. 23, 138
[22] A. Mihatsch, Relative unitary RZ-spaces and the Arithmetic Fundamental Lemma, arXiv:1611.06520 8,
28, 31
[23] D. Mumford, An analytic construction of degenerating curves over complete local rings, Compositio Math.
24 (1972), 129–174. 2
[24] M. Rapoport, B. Smithling, W. Zhang, Arithmetic diagonal cycles on unitary Shimura varieties,
arXiv:1710.06962 7, 10, 125
[25] M. Rapoport, B. Smithling, W. Zhang, On Shimura varieties for unitary groups, arXiv:1906.12346 13
[26] M. Rapoport, E. Viehmann, Towards a theory of local Shimura varieties, Münster J. of Math. 7 (2014),
273–326.
[27] M. Rapoport, Th. Zink, Period spaces for p-divisible groups. Annals of Mathematics Studies, 141, Prince-
ton University Press, Princeton, 1996. 2, 7, 9, 19, 81, 82, 83, 86, 91, 115, 116, 119, 120, 121, 137
[28] M. Rapoport, Th. Zink, On the Drinfeld moduli problem of p-divisible groups, Cambridge J. Math. 5
(2017), 229–279. 9, 11, 46, 48
[29] P. Scholze, J. Weinstein, Berkeley lectures on p-adic geometry, Annals of Mathematics Studies, 207,
Princeton University Press, Princeton, 2020. 3, 9
[30] J. Tate, The arithmetic of elliptic curves, Invent. Math. 23 (1974), 179–206. 2
 ON THE p-ADIC UNIFORMIZATION OF UNITARY SHIMURA CURVES 145

[31] Y. Varshavsky, P -adic uniformization of unitary Shimura varieties, Inst. Hautes Etudes Sci. Publ. Math.
87 (1998), 57–119. 2
[32] Y. Varshavsky, P -adic uniformization of unitary Shimura varieties. II, J. Differential Geom. 49 (1998),
75–113. 2
[33] Th. Zink, The display of a formal p-divisible group, in: Cohomologies p-adiques et applications
arithmétiques, I. Astérisque 278 (2002), 127–248. 3, 21, 23, 26, 31, 69
[34] Th. Zink, Windows for displays of p-divisible groups, in: Moduli of abelian varieties (Texel Island, 1999),
491–518, Progr. Math., 195, Birkhäuser, Basel, 2001. 20, 138
[35] Th. Zink, Displays and p-divisible groups, in: Spectral Structures and Topological Methods in Mathematics
(SFB701 report), eds. M. Baake, F. Götze, W. Hoffmann, 389–408, EMS Series of Congress Reports, Europ.
Math. Soc. Pub. House, Zürich 2019. 23, 25, 29, 96

Department of Mathematics, University of Toronto, 40 St. George St., BA6290, Toronto, ON

M5S 2E4, Canada.
E-mail address: [email protected]

Mathematisches Institut der Universität Bonn, Endenicher Allee 60, 53115 Bonn, Germany, and
University of Maryland, Department of Mathematics, College Park, MD 20742, USA
E-mail address: [email protected]

Fakultät für Mathematik, Universität Bielefeld, Postfach 100131, 33501 Bielefeld, Germany
E-mail address: [email protected]