Strickland Finite Subgroups of Formal Groups (1996) [sharethefiles com]

background image

Finite Subgroups of Formal Groups

Neil

P

.

Stric

kland

Trinity College, Cambridge CB2 1TQ, England

Abstract

We discuss various moduli problems involving the classi cation of nite subgroups

or related structures on formal groups of nite height

n

. We show that many moduli

schemes are smooth or at least Cohen-Macaulay. Moreover, many maps between

such schemes are nite and at, and their degrees can be predicted by thinking of

(

Q

p

=Z

p

)

n

as a \discrete model" for the formal group.

1 Introduction

In this paper we discuss various moduli problems involving the classi cation

of nite subgroups or related structures on formal groups of nite height.

Analogous problems for elliptic curves have of course been widely studied 9].

The moduli spaces which we consider turn out to be surprisingly well-behaved.

They are all Cohen-Macaulay, and most of them are smooth. The original

motivation for this work came from algebraic topology, in particular the study

of power operations in certain homology theories constructed by Morava. I

learnt most of what I know about these questions from Mike Hopkins, and a

great deal of the theory presented here was developed in discussions with him.

See Section 14 for a brief discussion of how moduli problems arise in algebraic

topology, and 15] for more details. Some readers may prefer a version of this

paper containing rather more exposition of standard algebraic material, which

is available from the author.

1.1 Synopsis

In Section 2 we set up our technical context by recalling various results about

complete local rings, interpreted in a geometric manner. In Sections 3 and 4,

1

partially supported by the N.S.F.

Preprint submitted to Elsevier Science

3 September 1996

background image

we establish some basic facts about formal groups and divisors. In Section 5

we show that the quotient of a formal group by a nite subgroup is again a

formal group. In Section 6 we reformulate the Lubin-Tate deformation theory

of formal groups in a coordinate-free way.
For the rest of this synopsis, we consider a formal group

G

0

of nite height

over a eld of positive characteristic, and let

G

=X

be its universal deformation.

The content of subsequent sections is as follows.

Section 7:

We de ne level-

A

structures on

G

(where

A

is a nite Abelian group). We prove

that the moduli space Level(

A

G

) is smooth, and that the map Level(

A

G

)

;

!

X

is nite and at.

Section 8:

We investigate maps of schemes over

X

between the schemes Level(

A

G

) they

all arise (contravariantly) from monomorphisms of nite Abelian groups, and

are nite and at.

Section 9:

We show how an epimorphism

u

:

A

;

!

B

gives rise to a nite at map

Level(

A

G

)

;

!

Level(

B

G

0

), for a dierent group

G

0

.

Section 10:

We show that the subgroups of

G

of degree

p

m

are classi ed by a scheme

Sub

m

(

G

) which is nite and at over

X

.

Section 11:

We consider ags 0 =

K

0

< K

1

< ::: < K

m

=

G

(1), where

G

(1) is the kernel

of multiplication by

p

on

G

and

K

i

has given degree

p

i

. We show that such

ags are classi ed by a smooth scheme Flag(

G

), and that there are nite at

maps Level(1

G

)

;

!

Flag(

G

)

;

!

X

, the rst of which is a Galois covering.

Section 12:

We consider the orbit scheme Type(

A

G

) = Level(

A

G

)

=

Aut(

A

) (suitably in-

terpreted). We show that Type(

A

G

) is smooth and the maps Level(

A

G

)

;

!

Type(

A

G

)

;

!

X

are nite and at. We also show that the normalisation

of Sub

m

(

G

) is a disjoint union of schemes of the form Type(

A

G

). More-

over, if the height of

G

0

is at most two (but not in general), then the maps

Type(

A

G

)

;

!

Sub

m

(

G

) are closed embeddings.

2

background image

Section 13:

We consider the problem of classifying deformations of a given isogeny of

formal groups over a eld.

Section 14:

We sketch the way in which the problems discussed above arise in algebraic

topology.

Section 15:

We derive some formulae which help one to construct and compute with formal

groups.

Section 16:

We present some detailed examples.

2 Geometry of Complete Local Rings

For brevity, a scheme will mean a formal scheme of the form spf(

A

), where

A

is a nite product of complete Noetherian local rings of residue characteristic

p >

0. We shall often assume that

A

is local, leaving trivial modi cations

for the semi-local case to the reader. We shall primarily think of schemes as

representable functors from the category of complete Noetherian local rings

and local homomorphisms to sets, via the de nition

spf(

A

)(

B

) = Hom(

AB

)

:

If

X

= spf(

A

) we write

O

X

for

A

. Given schemes

YZ

over a base scheme

X

(which is usually to be understood from the context), a point of

Y

dened

over

Z

will mean a map

a

:

Z

;

!

Y

of schemes over

X

. We write ;(

ZY

) for

the set of such points. If

f

2

O

Y

then we write

f

(

a

) for

a

f

2

O

Z

.

If

X

is connected (i.e.

O

X

is local) we write

X

=

O

X

=

m

X

for the residue

eld. We also write

X

0

= spf(

X

) and refer to this as the special bre of

X

.

We write dim(

X

) for the Krull dimension of

X

and embdim(

X

) for the em-

bedding dimension, that is

embdim(

X

) = dim

X

(

m

X

=

m

2
X

)

:

We shall say that

X

is integral if

O

X

is an integral domain, and smooth if

O

X

is a regular local ring. If

X

is integral we write

K

X

for the eld of fractions

3

background image

of

O

X

. Suppose we have a map of schemes

f

:

X

;

!

Y

, and thus a map

f

:

O

Y

;

!

O

X

, using which we consider

O

X

as a module over

O

Y

. As usual,

we say that

f

is at (resp. nite ) if

O

Y

is a at (resp. nitely generated)

O

X

-module. We also say that

f

is dominant (resp. epi) if the kernel of

f

is

nilpotent (resp. zero).
If

f

is nite and

Y

is connected we de ne the degree of

f

to be deg(

f

) =

dim

Y

(

Y

O

Y

O

X

). In our context, if

f

is also at then

O

X

is actually a free

module over

O

Y

(see 11, Theorem 7.10]).

We next state in this language some standard facts from commutative algebra.

Lemma 1

A nite dominant map of smooth schemes is at.

Lemma 2

Let

X

,

Y

and

Z

be integral schemes of the same dimension

d

, with

maps

X

f

;

!

Y

g

;

!

Z

. If

gf

is nite and dominant, then the same is true of

f

and

g

.

Lemma 3

If

X

is smooth and

f

:

X

;

!

Y

is nite and at then

Y

is smooth.

Proof.

Use the homological criterion 11, Theorem 19.2].

Let

X

be a scheme with an action of a nite group ;. We shall write

X=

; for

the scheme spf(

O

;X

). This is of course not a very good construction in general,

but it will turn out to be well-behaved in the cases we consider.

Denition 4

A nite extension

R

;

!

S

of integral domains is

Galois if

R

is the xed subring for the action of

Aut

R

(

S

) on

S

and

S

is free over

R

. I

am not sure whether this denition is standard. Similarly, we say that a map

X

;

!

Y

of integral schemes is a

Galois covering if

O

Y

;

!

O

X

is Galois.

The next two lemmas follow easily from the Galois theory of elds and the

fact that regular local rings are integrally closed.

Lemma 5

If

f

:

X

;

!

Y

is Galois then

K

X

is Galois over

K

Y

and

deg(

f

) =

O

X

:

O

Y

] =

K

X

:

K

Y

] =

j

Aut

O

Y

(

O

X

)

j

=

j

Aut

K

Y

(

K

X

)

j

:

Lemma 6

A nite dominant map

f

:

X

;

!

Y

of smooth schemes is a Galois

covering if and only if the extension of function elds

K

Y

;

!

K

X

is Galois.

Lemma 7

Let

X

be a smooth scheme with a faithful action of a nite group

;. Suppose that

f

:

X

;

!

Y

is a nite at map of degree

d

=

j

;

j

, such that

f

g

=

f

for all

g

2

;. Then

Y

is smooth,

f

is Galois, and the induced map

X=

;

;

!

Y

is iso.

4

background image

Proof.

By Lemma 3, we see that

Y

is smooth. Because

K

X

:

K

Y

] =

j

;

j

and

K

Y

K

;X

, Galois theory tells us that

K

Y

=

K

;X

and thus

K

Y

;

!

K

X

is Galois.

The claim now follows from Lemma 6.

We next record a few basic facts about norms. Suppose that

f

:

X

;

!

Y

is

a nite at map of degree

m

. We can then de ne a (nonadditive) norm map

N

f

=

N

X=Y

:

O

X

;

!

O

Y

, by letting

N

f

(

u

) be the determinant of multiplication

by

u

, considered as an

O

Y

-linear endomorphism of

O

X

.

Lemma 8

Suppose we have a pullback diagram

V

X

W

Y

w

a

u

u

g

u

u

f

w

b

If

f

is a nite free map then so is

g

, and

N

g

a

=

b

N

f

.

Proof.

For the square to be a pullback means that

O

V

=

O

W

O

Y

O

X

, and

with this identi cation we have

g

= 1

f

and

a

=

b

1. The claim follows

easily by choosing bases.

Lemma 9

Suppose that

s

:

Y

;

!

X

is a section of

f

, and that

s

u

= 0. Then

N

f

(

u

) = 0.

Proof.

We can write

O

X

=

f

O

Y

I

, where

I

= ker(

s

). As

u

2

I

, the image

of multiplication by

u

lies in the proper summand

I

of

O

X

, so the determinant

is zero.

3 Formal Groups

Let

X

= spf(

A

) be a scheme, and

G

a one-dimensional commutative formal

group over

X

. In more detail, we have

G

= spf(

B

) where

B

is isomorphic to

A

x

]] as an augmented

A

-algebra, and we are given a commutative group law

:

G

X

G

;

!

G

(which we shall write additively). In future, we shall simply

say \formal group" instead of \commutative one-dimensional formal group". A

coordinate

on a formal group

G

is a choice of generator

x

2

ker(0

:

O

G

;

!

O

X

)

such that

O

G

=

O

X

x

]]. (Here 0

is the map induced by the zero-section

0:

X

;

!

G

).

5

background image

A homomorphism (

fq

): (

G

X

)

;

!

(

H

Y

) of formal groups consists of maps

as follows, which make the obvious diagrams commute.

G

H

X

Y

w

q

u

u

w

f

We shall say that (

qf

) (or just

q

) is a brewise isomorphism if the diagram

above is a pullback.
Let

x

be a coordinate on

G

, and let

x

0

x

1

2

O

G

X

G

be obtained by pulling back

x

along the two projections

G

X

G

;

!

G

. Then

O

G

X

G

=

O

X

x

0

x

1

]], so we

can write

x

=

F

(

x

0

x

1

) for a unique power series

F

over

O

X

. Equivalently,

for any two points

ab

:

Y

;

!

G

we have

x

(

a

+

b

) =

F

(

x

(

a

)

x

(

b

)). This series

is called the formal group law associated to

G

and

x

. It is easily seen to have

the following properties:

F

(

yz

)=

y

+

z

(mod

xy

)

F

(

yz

)=

F

(

zy

)

F

(

F

(

xy

)

z

)=

F

(

xF

(

yz

))

We also write

x

+

F

y

for

F

(

xy

). Much of the literature on formal groups

is written in terms of formal group laws 6] 14, Appendix 2] 7]. However,

we shall nd it conceptually clearer to take a coordinate-free de nition the

primary one.
Let

X

0

= spf(

A=

m

) be the special bre, and put

G

0

=

G

X

X

0

. This is a

formal group over

X

0

.

The following lemma is well-known.

Lemma 10

Let

q

:

G

;

!

H

be a nonzero homomorphism of formal groups

over a base

X

, and let

x

and

y

be coordinates on

G

and

H

. Then there exist

a

2

O

X

and

n

2

N

such that

a

6

= 0 and

q

y

=

ax

p

n

(mod

x

p

n

+1

).

Denition 11

We shall call the integer

n

described above the

strict height of

q

, and dene the

height of

q

to be the strict height of

q

0

:

G

0

;

!

H

0

. We also

dene the (strict) height of

G

to be the (strict) height of the endomorphism

p

G

:

G

;

!

G

, which is just

p

times the identity map.

From now on, we shall take

G

to be a formal group of nite height

n

over a

connected base scheme

X

.

6

background image

Proposition 12

If

q

:

G

;

!

H

is nonzero, then

m

= height(

q

) is nite and

q

is at of degree

p

m

. Moreover,

height(

H

) = height(

G

).

Proof.

This is essentially standard. If

q

has in nite height then

q

= 0 this

is an easy generalisation of 6, p. 99] (in which

O

X

is assumed to be a DVR).

If there is a nonzero map

q

:

G

;

!

H

then height(

H

) = height(

G

) | this

is proved as a corollary of the last reference. The Weierstrass preparation

Theorem 6, Chapter 1] implies that

O

G

=q

y

is freely generated over

O

X

by

f

1

x::: x

p

m

;

1

g

. It is not hard to conclude that these elements also form a

basis for

O

G

over

q

O

H

.

4 Divisors

By a divisor on

G

we shall mean a closed subscheme

D

G

which is nite

and at over

X

. We briey record how things work out in our context. The

proofs are either easy or parallel to 9, Section 1.1] (for example).

Proposition 13

If

D

is a divisor of degree

d

on

G

and

x

is a coordinate,

then there is a unique monic polynomial

f

D

(

x

) of degree

d

such that

O

D

=

O

G

=f

D

(

x

). Moreover,

f

D

(

x

) is the characteristic polynomial of the

O

X

-linear

endomorphism of

O

D

'

O

dX

given by multiplication by

x

, and

f

D

(

x

) =

x

d

(mod

m

).

Denition 14

The

equation of a divisor

D

(with respect to a given coordinate

x

) is the polynomial

f

D

(

x

) as above. If

f

D

(

x

) =

P

dk=0

c

k

x

k

(with

c

d

= 1) then

the

parameters of

D

are the coecients

f

c

0

::: c

d

;

1

g

.

More generally, given a scheme

Y

over

X

, a divisor on

G

over

Y

will just mean

a divisor on the pulled-back group

G

X

Y

.

Because we are assuming that

G

has nite height

n

, the subscheme

G

(

m

) = ker(

p

m

G

:

G

;

!

G

) = spf(

O

G

=

(

p

m

G

)

x

)

is a divisor of degree

p

mn

.

If

a

is a section of

G

then we write

a

] for the associated divisor, which has

f

a]

(

x

) =

x

;

x

(

a

). The following lemma provides a useful alternative repre-

sentation:

Lemma 15

For any formal group law over any ring

A

, the power series

x

;

F

y

is a unit multiple of

x

;

y

in

A

xy

]].

7

background image

Proof.

x

;

F

y

clearly vanishes mod (

x

;

y

). By considering the rst few terms,

it is easy to see that (

x

;

F

y

)

=

(

x

;

y

) must be 1 modulo (

xy

), so it is a unit.

The functor from schemes over

X

to sets de ned by

Y

7!

f

divisors of degree

d

on

G

over

Y

g

is represented by a scheme Div

d

(

G

) =

G

dX

=

d

. If

x

is a coordinate on

G

then

the ring of functions on this scheme is just the symmetric subring

O

Div

d

(

G

)

=

O

X

1

:::

d

]]

O

X

x

1

::: x

d

]] =

O

G

dX

(where

k

is the

k

'th elementary symmetric function in the variables

x

i

). The

equation of the universal divisor over this ring is

f

D

(

x

) =

Y

k

(

x

;

x

k

) =

d

X

k=0

(

;

1)

k

k

x

d

;

k

:

The following proposition (taken from 9]) will help us to construct various

moduli schemes:

Proposition 16

Let

DD

0

be two divisors on

G

over

X

. There is then a

closed subscheme

Y

X

such that for any map

a

:

Z

;

!

X

we have

a

D

a

D

0

2

;(

Z

Div(

G

)) if and only if

a

factors through

Y

.

Proof.

Choose a coordinate

x

on

G

, and let

d

be the degree of

D

. Then there

are unique elements

b

k

2

O

X

such that

f

D

0

(

x

) =

d

;

1

X

k=0

b

k

x

k

(mod

f

D

(

x

))

:

Put

Y

= spf(

O

X

=

(

b

0

::: b

d

;

1

)), and check that this works.

5 Subgroups and Quotient Groups

By a nite subgroup of

G

we shall simply mean a divisor

K <

G

which is also

a subgroup scheme.

8

background image

Proposition 17

If

K

is a nite subgroup of

G

then the degree of

K

is a power

of

p

.

Proof.

It is enough to prove the proposition when

X

=

X

0

= spec(

), so that

f

K

(

x

) =

x

d

. It then follows from the general theory of nite group schemes

over a eld, but we shall give a direct proof. Let

F

be the formal group law

of

G

, so that

F

(

xy

) =

x

+

y

(mod

xy

). Because

K

is a subgroup, we have

f

K

(

F

(

xy

)) = 0 (mod

f

K

(

x

)

f

K

(

y

)). Reading this modulo (

xy

)

d+1

, we see

that the binomial coecient (

dk

) is divisible by

p

when 0

< k < d

. As in

Lemma 10, this implies that

d

is a power of

p

.

Proposition 18

If

K

is a nite subgroup of

G

of degree

p

m

then

p

mK

= 0 and

thus

K

G

(

m

) = ker(

p

m

G

).

Proof.

This is a special case of a result of Deligne | see 16]. A key point

is that one can give a purely equational proof that any element

u

of a nite

Abelian group

A

of order

d

has

u

d

= 1. Indeed,

Y

a

2

A

a

=

Y

a

2

A

ua

=

u

d

Y

a

2

A

a:

Let

K

be a nite subgroup of

G

of degree

p

m

. We would like to be able to

construct a quotient group

G

=K

. To do this, write

for the addition map

G

X

G

;

!

G

(or any of its restrictions) and

:

G

X

K

;

!

G

for the

projection. Let

O

G

=K

be the equaliser

O

G

=K

-

O

G

-

-

O

K

O

X

O

G

Let

x

be a coordinate on

O

G

, so that

y

=

N

x

2

O

G

.

Theorem 19

With

K

and

y

as above, we have

(i)

y

=

x

p

m

(mod

m

X

).

(ii)

O

G

=K

=

O

X

y

]].

(iii)

G

=K

has a natural structure as a formal group.

(iv) The projection

G

;

!

G

=K

is the categorical cokernel of

K

;

!

G

.

(v) For any map

f

:

Y

;

!

X

we have

f

G

=f

K

=

f

(

G

=K

).

Proof.

First, there is a map

:

G

X

K

;

!

G

X

K

de ned on points by

(

ab

) = (

a

;

bb

). This is an automorphism and satis es

=

. We know

9

background image

that

is a nite at map, so we conclude that the same is true of

. Next, let

0

:

G

X

K

X

K

;

!

G

X

K

be the projection on the rst two factors, and

consider the following commutative squares:

G

X

K

X

K

G

X

K

G

X

K

X

K

G

X

K

G

X

K

G

G

X

K

G

w

1

u

0

u

w

1

u

0

u

w

w

A diagram chase shows that they are both pullbacks. It follows that all the

maps involved are nite at maps, and that

N

=

N

0

(1

)

N

=

N

0

(

1)

:

Combining this with the fact that

(1

) =

(

1), we see that

y

=

N

x

2

O

G

actually lies in

O

G

=K

as claimed. The argument so far follows 3].

Next, write

j

:

K

;

!

G

for the inclusion. I claim that

j

y

= 0. To see this, let

i

:

K

;

!

G

X

K

be the map

a

7!

(0

a

), so that

i

= 0 and

i

=

j

. Thus

j

y

=

i

y

=

i

y

= 0

y

. Next consider the diagram

K

G

X

K

G

X

G

w

i

u

w

u

w

0

Using Lemma 8 we see that 0

y

=

N

K=X

i

x

=

N

K=X

j

x

. The claim now

follows from Lemma 9, using the zero section

X

;

!

K

.

Let

f

K

be the equation of the divisor

K

, so that

f

K

is a monic polynomial of

degree

p

m

and ker(

j

) = (

f

K

(

x

)). It thus follows that

y

is divisible by

f

K

(

x

).

From Proposition 13 we know that

O

K

=

m

X

=

x

]]

=x

p

m

, so that

O

G

X

K

=

m

X

=

xz

]]

=z

p

m

:

As working mod

z

is the same as restricting to

G

, we see that the image of

x

in this ring agrees with

x

mod

z

. Using the basis

f

1

z::: z

p

m

;

1

g

, we

conclude that the norm of

x

is just

x

p

m

. However, this norm is just the

image of

y

in

O

G

=

m

X

, which gives part (1) of the theorem. It follows easily

from this that

y

is a unit multiple of

f

K

(

x

).

10

background image

It is not hard to see that there is a commutative diagram

O

G

=K

O

G

O

K

O

G

O

X

O

K

O

K

O

K

w

u

0

u

j

-

-

u

1

j

w

-

-

Suppose

u

2

O

G

=K

O

G

. By the above diagram,

j

(

u

;

u

(0)) = 0. By the

last paragraph, we see that

u

;

u

(0) is divisible by

y

, say

u

=

u

(0) +

u

0

y

with

u

0

2

O

G

. Using the Weierstrass factorisation, we see that

y

is not a zero

divisor in

O

G

X

K

. Using this, we can check that

u

0

2

O

G

=K

. Extending this

inductively, we conclude that

O

G

=K

=

O

X

y

]], giving part (2) of the theorem.

For part (3), a diagram chase now shows that

G

X

G

;

!

G

q

;

!

G

=K

factors

as

G

X

G

q

q

;

;

!

G

=K

X

G

=K

;

!

G

=K

for a unique map

, which makes

G

=K

into a formal group with coordinate

y

. It is easy to see from the de nition of

O

G

=K

that

G

=K

is the categorical cokernel of

j

, in other words that a map

r

:

G

;

!

H

of formal groups factors through

q

if and only if

rj

= 0 this is

part (4). Finally, our construction of

y

clearly commutes with base change. It

follows that

f

(

G

=K

) =

f

G

=f

K

for any map

f

:

Y

;

!

X

, as claimed in (5).

6 The Universal Deformation

Suppose that

X

0

is a scheme of the form spf(

) (where

is a eld of character-

istic

p

), and that

G

0

is a formal group over

X

0

of nite height

n

. A deformation

of

G

0

consists of a formal group

H

over a scheme

Y

together with a pullback

diagram as follows, exhibiting the restriction of

H

to the special bre of

Y

as

a pullback of

G

0

.

H

0

G

0

Y

0

X

0

w

~f

0

u

u

w

f

0

Of course, if we start with a group

G

over

X

and de ne

X

0

to be the special

bre of

X

and

G

0

to be the restriction of

G

to

X

0

, then

G

is a deformation of

G

0

in a tautological way.

11

background image

A morphism of deformations is a pullback square as follows, which is compat-

ible in the obvious sense with the given maps

H

0

;

!

G

0

;

H

0

0

.

H

H

0

Y

Y

0

w

u

u

w

Recall that the pullback condition means (by de nition) that the map

H

;

!

H

0

is a brewise isomorphism.
In 10] Lubin and Tate construct universal deformations of formal group laws.

The following proposition is a simple translation of their result into coordinate-

free language.

Proposition 20

The category of deformations of

G

0

=X

0

has a terminal object

G

=X

. Moreover, coordinates can be chosen such that

O

X

=

W

u

1

::: u

n

;

1

]]

(where

W

is the Witt ring of

) and

p

G

x

=

u

m

x

p

m

(mod

u

0

::: u

m

;

1

x

p

m

+1

)

(where

u

0

=

p

). Note that

X

is a smooth scheme of dimension

n

, and

X

0

is

the special bre of

X

. We refer to

X

as the deformation space of

G

0

, and

G

as the universal deformation.
Suppose we start with a group

G

0

=X

0

, de ne

X

0

=

X

0

0

to be the special bre

of

X

0

and

G

0

=

G

0

0

the restriction of

G

0

over

X

0

0

. Let

G

=X

be the universal

deformation of

G

0

=X

0

. There is then a unique map

f

:

X

0

;

!

X

such that

G

0

'

f

G

. We shall often use only the much weaker corollary that

G

0

is

obtained by pulling back a group over a smooth scheme.

7 Level Structures

Let

A

be a nite Abelian

p

-group. It is not hard to see that the functor from

schemes over

X

to sets given by

Y

7!

Hom(

A

;(

Y

G

))

:

12

background image

is represented by a scheme Hom(

A

G

) over

X

. Indeed, we may write

A

in the

form

L

r

;

1

k=0

Z

=p

d

k

, and then Hom(

A

G

) is isomorphic to the closed subscheme

Q

k

G

(

d

k

) of

G

r

(where

G

(

d

) = ker(

p

d

G

) as usual). If

x

is a coordinate on

G

then

O

Hom(

A

G

)

=

O

X

x

0

::: x

r

;

1

]]

=

(

p

d

0

](

x

0

)

:::

p

d

r

;1

](

x

r

;

1

))

:

By Proposition 12, this is a nite free module over

O

X

, in other words the

map Hom(

A

G

)

;

!

X

is nite and at. The degree is

j

A

j

n

.

If

G

were a discrete group, it would be natural to write

Hom(

A

G

) =

a

B

A

Mon(

A=B

G

)

(where Mon denotes monomorphisms). However, except in trivial cases, there

is no scheme Mon(

A

G

) over

X

such that

Mon(

A

;(

Y

G

)) = ;(

Y

Mon(

A

G

))

:

Indeed, by considering the inclusion

;

!

Y

we see that the left hand side

is not even a functor of

Y

. The object of the theory of level structures is to

approximate the above decomposition as well as possible. The key idea is due

to Drinfel'd 5], and some of our results below are parallel to results in 9,

Section 1.6].
We write

A

(

k

) = ker(

p

k

:

A

;

!

A

). For any map

:

A

;

!

;(

Y

G

) we write

A

] for the divisor

P

a

2

A

(

a

)]. We also put = (

Q

p

=

Z

p

)

n

, so that (

m

) =

(

Z

=p

m

)

n

. Note that

j

Hom(

A

)

j

=

j

A

j

n

, which is the same as the degree

of Hom(

A

G

)

;

!

X

. This is the rst of many cases in which serves as a

\discrete approximation" to

G

.

Denition 21

A

level-

A

structure on

G

over an

X

-scheme

Y

is a map

:

A

;

!

;(

Y

G

) such that

A

(1)]

G

(1) as divisors. A level-

m

structure

means a level-

(

m

) structure.

The following is very similar to 9, Proposition 1.6.2].

Proposition 22

The functor from schemes over

X

to sets given by

Y

7!

f

level-

A

structures on

G

over

Y

g

is represented by a scheme Level

(

A

G

) over

X

. (We will write Level

(

m

G

) =

Level

((

m

)

G

)).

13

background image

Proof.

We can consider

G

X

Hom(

A

G

) as a formal group over Hom(

A

G

).

On this group we have a tautological map

:

A

;

!

;(Hom(

A

G

)

G

) and

thus a divisor

A

(1)]

2

;(Hom(

A

G

)

Div(

G

)). Applying Proposition 16 to

this divisor and the divisor

G

(1)

X

Hom(

A

G

), we get a closed subscheme

Level(

A

G

)

Hom(

A

G

). It is easy to see that this does the job.

Suppose that

u

:

A

;

!

B

is mono, and that

:

B

;

!

;(

Y

G

) is a level struc-

ture. Then

u

is clearly a level-

A

structure, and is thus classi ed by a

map

Y

;

!

Level(

A

G

). Applying this in the universal case, we obtain a map

u

!

: Level(

B

G

)

;

!

Level(

A

G

). This construction is contravariantly functo-

rial.

Theorem 23

Level

(

A

G

)

;

!

X

is a nite at map. If

G

is the universal

deformation of

G

0

then Level

(

A

G

) is smooth and has dimension

n

.

Theorem 24

If

:

A

;

!

G

is a level structure then

A

] is a subgroup of

G

and

A

(

k

)]

G

(

k

) for all

k

. If

A

= (

m

) then

A

] =

G

(

m

).

The proofs of these two theorems will follow after a number of intermediate

propositions. Similar results are proved in 5] and 9].

Notation

g

m

(

x

) is the Weierstrass polynomial (of degree

p

nm

) which is a

unit multiple of

(

p

m

G

)

(

x

).

Lemma 25

If rank

(

A

)

> n

then Level

(

A

G

) =

.

Proof.

In this case

A

(1)] has degree greater than that of

G

(1), so we can

never have

A

(1)]

G

(1).

From now on, we will always assume that

A

has rank at most

n

.

Proposition 26

If

O

Y

is an integral domain of characteristic zero, then a

map

:

A

;

!

;(

Y

G

) is a level structure if and only if it is injective.

Proof.

Observe that

is injective if and only if ker(

)

\

A

(1) = 0, so we may

assume that

A

=

A

(1). Write

x

a

=

x

(

(

a

)) and

y

=

f

A]

(

x

) =

Y

a

2

A

(

x

;

x

a

)

2

O

G

X

Y

:

14

background image

Suppose that

is a level structure. Then for

a

2

A

n

0 we have divisibility

relations:

x

(

x

;

x

a

)

j

y

j

p

G

x

=

px

+ higher terms

:

As

p

6

= 0 in

R

, we must have

x

a

6

= 0 and thus

(

a

)

6

= 0, as required. Con-

versely, suppose

is injective. The values

x

a

as

a

runs over

A

are distinct

roots of the Weierstrass polynomial

g

1

(

x

) de ned above. By the usual theory

of factorisation over a domain, the product

y

=

Y

a

2

A

(

x

;

x

a

)

divides

g

1

(

x

), so we have a level structure.

Proposition 27

Suppose that

X

is integral. Then there is a prime ideal

p

<

O

L

evel

(

A

G

)

with

p

\

O

X

= 0.

Proof.

Let

m

be large enough that

p

m

A

= 0. Let

K

be the eld of fractions

of

O

X

,

L

the splitting eld of

g

m

over

K

, and

O

Y

the subring of

L

generated

over

O

X

by the roots of

g

m

. These roots are the same as the zeros of (

p

m

G

)

x

, so

they form a group

A

0

under the operation +

F

. If we identify a map

a

:

Y

;

!

G

over

X

with the element

x

(

a

) =

a

(

x

)

2

O

Y

, then

A

0

= ;(

Y

G

(

m

)).

Write

y

= (

p

m

G

)

x

. Note that

y

=

p

m

x

(mod

x

2

), and

p

6

= 0 in

L

, so

y

vanishes

only to rst order at 0. Suppose that

ab

are points of

G

and

p

m

a

= 0. Then

y

(

a

+

b

) =

x

(

p

m

a

+

p

m

b

) =

y

(

b

). It follows that

y

vanishes only to rst order

at

a

. Thus, all the roots of

g

m

in

O

Y

are distinct.

Clearly

A

0

(1) is the set of roots of

g

1

, and thus has order

p

n

. Moreover,

p

m

A

0

=

0 and

j

A

0

j

=

p

nm

. It follows by the structure theory of nite Abelian groups

that

A

0

'

(

Z

=p

m

)

n

, and thus that we can choose an embedding

A

;

!

A

0

. By

Proposition 26, this is a level structure on

G

over

Y

. It is therefore classi ed

by a map

f

:

Y

;

!

Level(

A

G

) of schemes over

X

. We can take

p

to be the

kernel of

f

:

O

Level(

A

G

)

;

!

O

Y

.

Remark 28

In geometric language, this says that there is an irreducible com-

ponent

Y

of Level

(

A

G

) such that

Y

;

!

X

is nite and dominant. We shall

see shortly that

Y

= Level(

A

G

).

Corollary 29

If

X

is integral then

dim(Level(

A

G

))

dim(

X

)

Proposition 30

If

G

is the universal deformation of

G

0

then Level

(

A

G

) is

a smooth scheme of dimension

n

.

15

background image

Proof.

Write

E

=

O

X

=

W

u

1

::: u

n

;

1

]] (as in Section 6), and

D

A

=

O

Level(

A

G

)

. Suppose that

A

has rank

r

, say

A

=

h

a

0

::: a

r

;

1

i

. Write

x

k

=

x

(

(

a

k

))

2

D

A

. By the proof of Proposition 22, we know that these elements

topologically generate

D

A

over

E

, so that the

u

's and

x

's generate

m

D

A

. We

also put

I

= (

x

0

::: x

r

;

1

u

r

::: u

n

;

1

)

:

It is enough to show that

D

A

=I

=

, for then

embdimLevel(

A

G

)

n

= dim

X

dimLevel(

A

G

)

:

Write

Y

= spf(

D

A

=I

). The map

:

A

;

!

;(

Y

G

) sends the generators to zero,

so it is zero. Thus,

f

(

x

) =

f

A(1)]

(

x

) =

x

p

r

in (

D

A

=I

)

x

]. Because

is a level

structure, this divides

p

G

x

. Using the fact (see Section 6) that

p

G

x

=

u

m

x

p

m

(mod

u

0

::: u

m

;

1

x

p

m

+1

)

we nd that

u

0

::: u

r

;

1

vanish in

D

A

=I

. Of course, the rest of the

u

's also

vanish by de nition of

I

. As the

u

's and

x

's generate

m

D

A

, this implies that

D

A

=I

=

as claimed.

Proposition 31

For any

G

, the projection Level

(

A

G

)

;

!

X

is a nite at

map.

Proof.

We may assume that

G

is the universal deformation of

G

0

, as the

general case is easily recovered by base change. As Level(

A

G

) is a closed

subscheme of Hom(

A

G

) and Hom(

A

G

)

;

!

X

is a nite at map, we see

that Level(

A

G

)

;

!

X

is nite. By propositions 27 and 30, we see that it is

also dominant. A nite dominant map of smooth schemes is at (Lemma 1).

Proposition 32

If

is a level structure then

A

(

k

)] is a subgroup scheme

contained in

G

(

k

).

Proof.

We may assume that

A

=

A

(

k

), so that

p

k

A

= 0. Suppose for the mo-

ment that the base scheme

X

is integral. As in the proof of Proposition 27, we

see that the Weierstrass polynomial

g

k

has distinct roots. By Proposition 26,

is injective, so the polynomials

x

;

x

a

for

a

2

A

are distinct factors of

g

k

(

x

).

It follows that

Q

a

(

x

;

x

a

) divides

g

k

(

x

), so that

A

]

G

(

k

) as claimed.

16

background image

Next, write

z

=

Y

a

2

A

(

x

;

F

x

a

) = (unit)

Y

a

2

A

(

x

;

x

a

)

:

In other words, for any point

b

of

G

we have

z

(

b

) =

Y

a

2

A

x

(

b

;

(

a

))

:

Write

H

= spf(

O

X

z

]]), so

H

is a quotient scheme of

G

. By the Weierstrass

preparation theorem, we see that

q

:

G

;

!

H

is at, of degree

j

A

j

.

We can let

A

act on

G

, with

a

2

A

acting as translation by

(

a

). This action

is faithful, because

is injective. The map

q

:

G

;

!

H

clearly factors through

G

=A

. It follows from Lemma 7 that

H

=

G

=A

and that

q

is Galois. Similarly,

we nd that (

G

X

G

)

=

(

A

A

) =

H

X

H

. Using these descriptions, it is

easy to construct a (unique) multiplication on

H

such that

q

is an isogeny of

formal groups. The kernel of

q

is just spf(

O

G

=z

) =

A

], so

A

] is a subgroup

scheme as claimed.
Now suppose that

X

is not integral. Let

G

0

=X

0

be the universal deforma-

tion of

G

0

, and

0

the universal level structure de ned over Level(

A

G

0

). By

Proposition 30, we know that Level(

A

G

0

) is integral, so the above tells us

that

0

A

(

k

)] is a subgroup scheme contained in

G

0

(

k

). Using the de ning

properties of

X

0

and Level(

A

G

0

), we construct a map

f

:

X

;

!

Level(

A

G

0

)

such that the pullback of

G

0

is

G

and the pullback of

0

is

. It follows that

A

(

k

)] =

f

0

A

(

k

)] is a subgroup scheme of

G

(

k

), as claimed.

Corollary 33

It follows from the above proof that

z

=

Q

a

2

A

(

x

;

F

x

a

) is a

coordinate on the quotient group

G

=

A

].

8 Galois Theory of Level Structures

In this section, we assume that

G

is the universal deformation of

G

0

. We write

D

A

=

O

Level(

A

G

)

D

m

=

O

Level(

m

G

)

E

=

D

0

=

O

X

K

A

= eld of fractions of

D

A

:

17

background image

Theorem 34

Let

A

and

B

be nite Abelian

p

-groups of rank at most

n

, and

u

:

A

;

!

B

a monomorphism. Then:

(i)

Hom

E

(

D

A

D

B

) = Hom

K

(

K

A

K

B

) = Mon(

AB

).

(ii) The map

u

!

: Level(

B

G

)

;

!

Level

(

A

G

) (which sends

to

u

) is nite

and at.

(iii) If

B

'

(

m

) then

u

!

is a Galois covering.

(iv) The torsion subgroup of

;(Level(

A

G

)

G

) is

A

.

The proof will follow after a number of lemmas.

Lemma 35

Hom

K

(

K

A

K

B

) = Hom

E

(

D

A

D

B

) = Mon(

A

;(

D

B

G

))

:

Proof.

By applying Lemma 2 to a sequence of maps

Level(

B

G

)

;

!

Level(

A

G

)

;

!

X

we see that every

E

-algebra map

D

A

;

!

D

B

is a monomorphism, and thus

induces a map

K

A

;

!

K

B

of the elds of fractions. Conversely, consider a

K

-algebra map

K

A

;

!

K

B

. The image of

D

A

is integral over

E

, hence over

D

B

. As

D

B

is a unique factorisation domain (by Theorem 23), it is integrally

closed in its eld of fractions. Thus, the image of

D

A

is contained in

D

B

.

It follows that Hom

K

(

K

A

K

B

) = Hom

E

(

D

A

D

B

). On the other hand, the

de ning property of

D

A

implies that

Hom

E

(

D

A

D

B

) =

f

level-

A

structures on

G

over

D

B

g

:

By Proposition 26, this is just Mon(

A

;(

D

B

G

)).

Lemma 36

Aut((

m

)) acts transitively on Mon(

A

(

m

)).

Proof.

Without loss of generality, we may assume that there is a monomor-

phism

A

;

!

(

m

), and thus think of

A

as a subgroup of (

m

). Suppose that

:

A

;

!

(

m

) is a mononomorphism other than the inclusion. Our task is

then to nd an automorphism

of (

m

) extending

. By elementary linear

algebra over

F

p

, the map

1

:

A

(1)

;

!

(1) extends to an automorphism of

(1). This patches with

to give a map

0

:

A

+ (1)

;

!

(

m

)

. As is

a divisible group, it is injective, so

0

extends to a map

: (

m

)

;

!

. The

image must be killed by

p

m

, so we actually have

: (

m

)

;

!

(

m

). This is iso

on (1), so the kernel of

contains no points of order

p

, so

itself is mono.

By counting,

must be iso. Thus

2

Aut((

m

)) and

j

A

=

as required.

18

background image

Lemma 37

Let

B

be a subgroup of

(

m

), and set

; =

f

2

Aut((

m

)) such that

j

B

= 1

B

g

:

Then the group of xed points of

; is just

B

.

Proof.

Suppose that

u

62

B

. We need to construct

2

; with

(

u

)

6

=

u

.

Suppose that

u

generates a cyclic subgroup

C

of order

p

l

, and that

C

0

=

C

\

B

is generated by

p

k

u

and so has order

p

l

;

k

< p

l

. We can then construct an

automorphism

of

A

=

B

+

C

by de ning

(

b

+

c

) =

b

+ (1 +

p

l

;

k

)

c

when

b

2

B

and

c

2

C

(it is not hard to check that this is well-de ned). This

in turn extends to an automorphism of (

m

) by Lemma 36. It sends

u

to

(1 +

p

l

;

k

)

u

6

=

u

.

Note that

K

m

is generated over

K

by roots of the Weierstrass polynomial

g

m

,

which splits completely over

K

m

. Thus

K

m

is the splitting eld of

g

m

, and so

is Galois over

K

. We can also see that (

m

) is the full group of points of order

p

m

in ;(

D

m

G

), as

g

m

can only have deg(

g

m

) =

p

mn

=

j

(

m

)

j

roots. It follows

using Lemma 35 that Gal(

K

m

=K

) = Aut

K

(

K

m

) = Aut((

m

)). Moreover, if

p

m

A

= 0 then

K

A

is generated by a subset of the roots of

g

m

, so that

K

m

contains a normal closure of

K

A

. The degree of the extension

K

;

!

K

A

is just

the number of embeddings of

K

A

in the normal closure, so

K

A

:

K

] =

j

Hom

K

(

K

A

K

m

)

j

=

j

Mon(

A

(

m

))

j

:

Lemma 38

Hom

K

(

K

A

K

B

) = Mon(

AB

)

:

Proof.

The extension

K

m

=K

B

is Galois as

K

m

=K

is. Moreover, the subgroup

of Gal(

K

m

=K

) xing

K

B

is just the group ;

Aut((

m

)) of Lemma 37. By

the fundamental theorem of Galois theory, the xed eld of ; is just

K

B

. Thus

Hom

K

(

K

A

K

B

) = Hom

K

(

K

A

K

m

)

;

= Mon(

A

(

m

))

;

= Mon(

AB

)

the last equality coming from Lemma 37.

Corollary 39

The torsion subgroup of

;(

D

B

G

) is

B

.

Proof.

This follows easily by comparing Lemma 38 with Lemma 35

Proof of Theorem 34

Part (1) follows immediately from lemmas 35 and 38.

It follows that all the maps

u

!

: Level(

B

G

)

;

!

Level(

A

G

) are dominant. Both

19

background image

the source and target are smooth by Theorem 23, so

u

!

is at by Lemma 1.

This gives (2). Now suppose that

B

'

(

m

), so without loss of generality

A

B

= (

m

). We have seen above that

K

m

=K

A

is Galois, so Lemma 6 tells

us that

u

!

is a Galois covering. This gives (3). Finally, (4) is Corollary 39.

We can use the above theory to understand the structure of

D

A

more ex-

plicitly. For simplicity, assume that

A

=

Z

=p

k

Z

=p

l

, with

k

l

(the ex-

tension to groups of larger rank is evident). It is clear that

g

k

(

x

) is divisible

by

g

k

;

1

(

x

), say

g

k

(

x

) =

g

k

;

1

(

x

)

h

k

(

x

) for some Weierstrass polynomial

h

k

of

degree

p

nk

;

p

n(k

;

1)

. Let

D

0

be the subring of

D

A

generated over

E

by the

root

x

0

=

x

(

(

a

0

)) of

h

k

(

x

). Over

D

0

, the series

h

l

(

x

) has

p

l

;

p

l

;

1

roots,

f

x

(

jp

k

;

l

a

0

)

j

0

< j < p

l

j

6

= 0 (mod

p

)

g

. After dividing out the cor-

responding linear factors, we obtain a Weierstrass polynomial

f

(

x

) of degree

p

nl

;

p

n(l

;

1)

;

p

l

+

p

l

;

1

. The ring

D

A

is generated over

D

0

by a root

x

1

=

x

(

(

a

1

))

of this polynomial. It is not hard to see that the product of the degrees of

h

k

and

f

is the same as the rank

j

Mon(

A

)

j

of

D

A

over

E

. It follows that

D

A

=

E

x

0

x

1

]]

=

(

h

k

(

x

0

)

f

(

x

1

)).

9 Quotients by Level Structures

Let

G

0

be a formal group of height

n

over

X

0

= spec(

). For every

m

, the

divisor

p

m

0] is a subgroup of

G

0

. We write

G

0

h

m

i

for the quotient group

G

0

=p

m

0], and

G

h

m

i

;

!

X

h

m

i

for the universal deformation of

G

0

h

m

i

;

!

X

0

.

Note that

G

0

(1) = ker(

p

G

0

:

G

0

;

!

G

0

) =

p

n

0]

:

It follows that

p

G

0

induces an isomorphism

G

0

h

m

+

n

i

;

!

G

0

h

m

i

. We use this

to identify

X

h

m

+

n

i

with

X

h

m

i

and

G

h

m

+

n

i

with

G

h

m

i

.

Proposition 40

Let

u

:

A

;

!

B

be an epimorphism of nite Abelian

p

-groups,

with

j

ker(

u

)

j

=

p

l

say. Then

u

induces a map

u

!

: Level(

A

G

h

m

i

)

;

!

Level

(

B

G

h

m

+

l

i

)

:

Proof.

Let

be the universal level-

A

structure on

G

h

m

i

, which is de ned

over the scheme

Y

= Level(

A

G

h

m

i

). Write

K

=

ker(

u

)]. This is a subgroup

divisor on

G

h

m

i

, whose restriction to the special bre of

Y

is

K

0

=

p

l

0]. It

follows that

G

h

m

i

=K

is a deformation of

G

0

h

m

+

l

i

. The composite

ker(

u

)

;

!

A

;

!

;(

Y

G

h

m

i

)

;

!

;(

Y

G

h

m

i

=K

)

20

background image

is clearly zero. This gives a map

:

B

;

!

;(

Y

G

h

m

i

=K

)

:

I claim that this is a level structure. As

O

Y

is an integral domain of character-

istic zero, it is enough to show that

is injective (Proposition 26). Suppose

a

2

B

n

ker(

u

), and let

x

be a coordinate on

G

h

m

i

. In view of Corollary 33,

we need only check that that the following product does not vanish:

z

(

a

) =

Y

c

2

ker(

u)

x

(

a

;

c

) =

Y

c

2

ker(

u)

x

(

(

a

;

c

))

:

Because

is a level structure,

a

;

c

6

= 0, and

O

Y

is an integral domain, we

conclude that

x

(

(

a

;

c

))

6

= 0. It follows that

z

(

(

a

))

6

= 0 as required.

The level structure

is classi ed by a map Level(

A

G

h

m

i

)

;

!

Level(

B

G

h

m

+

l

i

), which we take to be

u

!

.

In particular, if

j

A

j

=

p

l

then the maps 0

0

;

!

A

0

;

!

0 induce maps

0

!

: Level(

A

G

h

m

i

)

;

!

Level(0

G

h

m

i

) =

X

h

m

i

0

!

: Level(

A

G

h

m

i

)

;

!

Level(0

G

h

m

+

l

i

) =

X

h

m

+

l

i

:

The rst of these is just the usual projection.

Theorem 41

(1)

u

!

is a covariant functor of

u

.

(2)

u

!

is a nite at map.

(3) The maps

0

0

;

!

(

l

)

0

;

!

0 give the same map

0

!

= 0

!

: Level(

l

G

h

m

i

)

;

!

X

h

m

i

=

X

h

m

+

n

i

:

(4) Consider a commutative square as follows.

A

B

C

D

v

w

r

u

u

u

u

u

v

v

w

s

This is a pullback if and only if it is a pushout. If so, then

s

!

v

!

=

u

!

r

!

: Level(

B

G

h

m

i

)

;

!

Level

(

C

G

h

m

+

l

i

)

21

background image

(where

j

ker(

u

)

j

=

p

l

=

j

ker(

v

)

j

).

(5) If

u

is iso then

u

!

= (

u

!

)

;

1

= (

u

;

1

)

!

.

(6) If

A

'

(

m

) then

u

!

is a Galois covering, with Galois group

; =

f

2

Aut(

A

)

j

u

=

u

g

.

(7) The maps

0

!

: Level(

B

G

h

m

i

)

;

!

X

h

m

+

l

i

and

0

!

: Level(

B

G

h

m

i

)

;

!

X

have the same degree.

Proof.

We can take

m

= 0 without loss of generality, and assume that

G

is

the universal deformation of

G

0

.

(1) This is clear.

(2) This will be proved after (3).

(3) The map 0

!

: Level(

l

G

)

;

!

X

is just the originally given projection, so

the group over Level(

l

G

) with which we implicitly start is just (0

!

)

G

.

We have a universal level structure

: (

l

)

;

!

(0

!

)

G

. By Theorem 24,

we have

(

l

)] =

G

(

l

). By the de nition of the map 0

!

, we have an

isomorphism

(0

!

)

G

=

G

(

l

) = (0

!

)

G

=

(

l

)]

;

!

0

!

G

h

nl

i

whose restriction to the special bre is just the identity map of

G

0

=p

nl

0].

After using our standard identi cation of

G

h

nl

i

with

G

, we get an iso-

morphism

(0

!

)

G

=

G

(

l

)

;

!

0

!

G

whose restriction to the special bre is the map

p

l

G

0

:

G

0

=p

nl

0]

;

!

G

0

:

Because

G

is the universal deformation of

G

0

, the map 0

!

is uniquely

characterised by the existence of such an isomorphism. On the other

hand, we have an isomorphism

p

l

G

: (0

!

)

G

=

G

(

l

)

;

!

(0

!

)

G

which has the same eect on the special bre. It follows that 0

!

= 0

!

.

(2) Choose an epimorphism

v

: (

l

)

;

!

A

(for some large

l

). We thus have a

chain of epimorphisms

(

l

)

v

;

!

A

u

;

!

B

0

;

!

0

:

This gives maps as follows, where the

's refer to various integers:

Level(

l

G

hi

)

v

!

;

!

Level(

A

G

hi

)

u

!

;

!

Level(

B

G

hi

)

0

!

;

!

X

hi

:

22

background image

By (3), we know that the full composite 0

!

u

!

v

!

is nite and dominant. All

the schemes involved are smooth (and thus integral) and have dimension

n

. Applying Lemma 2 twice, we see that

u

!

is nite and dominant. It is

therefore at, by Lemma 1.

(4) Consider the following sequence

A

(

ru

)

;

;

!

B

C

(

v

;

s)

;

;

;

;

!

D:

The left hand map is mono (because

r

is), the right hand map is epi

(because

v

is) and the composite is zero (because the diagram commutes).

It is easy to see from this that the sequence is exact if and only if the

square is a pushout, if and only if the square is a pullback. If so, we write

E

= ker(

u

) and observe that

r

:

E

'

ker(

v

). Write

Y

= Level(

B

G

), so

we have a level structure

:

B

;

!

;(

Y

G

). We de ne maps

and

by

requiring that the following diagrams commute:

A

B

;(

Y

G

)

C

D

;(

Y

G

=

rE

])

v

w

r

u

u

u

w

u

u

v

u

v

w

s

w

A

;(

Y

G

)

D

;(

Y

G

=

rE

])

w

r

u

u

u

u

w

The two maps

u

!

r

!

and

s

!

v

!

from

Y

to Level(

C

G

h

l

i

) classify the two level

structures

and

s

on

G

=

rE

]. On the other hand, it is clear from

the above diagrams that these level structures are the same.

(5) Apply (4) with

s

=

u

;

1

and

r

=

v

= 1.

(6) By (3) and Theorem 34, we know that the composite Level(

A

G

)

u

!

;

!

Level(

B

G

h

l

i

)

0

!

;

!

X

h

nm

i

is a Galois covering. Using Lemma 6 to reduce

to the case of eld extensions, we see that

u

!

is Galois. Referring again

to Theorem 34, we see that the Galois group is

; =

f

2

Aut(

A

)

j

u

!

!

=

u

!

g

:

Using (5) and the fact that level structures over integral domains in char-

acterstic zero are injective, we conclude that

; =

f

2

Aut(

A

)

j

u

=

u

g

:

23

background image

(7) Choose an epimorphism

u

:

A

= (

l

)

;

!

B

, and let ; be as above. By

the Pontrjagin dual of Lemma 36, we see that

deg(0

!

) =

j

Aut(

A

)

=

;

j

=

j

Mon(

A

(

l

)

)

j

=

j

Epi((

l

)

A

)

j

(where

A

= Hom(

A

Q

=

Z

) etc.). As there are (unnatural) isomorphisms

A

'

A

and (

l

)

'

(

l

)

, we see that

deg(0

!

) =

j

Mon(

A

(

l

))

j

=

j

Mon(

A

)

j

:

By Theorem 34, this is the same as the degree of deg(0

!

).

10 Classication of Subgroups

We now x

m

0 and study the classi cation of subgroups of

G

of degree

p

m

.

Theorem 42

The functor from schemes over

X

to sets given by

Y

7!

f

subgroups of

G

X

Y

of degree

p

m

g

is represented by a scheme Sub

m

(

G

) over

X

. Moreover, for any map

X

0

;

!

X

we have

X

0

X

Sub

m

(

G

) = Sub

m

(

X

0

X

G

)

:

The projection Sub

m

(

G

)

;

!

X

is a nite at map, of degree

d

=

j

Sub

m

()

j

= number of subgroups of of order

p

m

:

Formulae for

d

will be given later. The scheme Sub

m

(

G

) is Gorenstein.

The rest of this section will constitute the proof. The existence and behaviour

under pullback of Sub

m

(

G

) is given by 9, Corollary 1.3.7]. In Subsection 10.1,

we prove some combinatorial formulae. In 10.2, we apply these formulae in an

argument inspired by the theory of Gr obner bases to give an upper bound for

the degree of the map Sub

m

(

G

)

;

!

X

. In 10.3, we analyse what happens when

we invert

p

and thus make all nite subgroups !etale (cf. 9, Corollary 3.7.2]).

In 10.4, we assemble these results to prove the theorem, except for the fact

that Sub

m

(

G

) is Gorenstein, which is proved in 10.5.

24

background image

10.1 Combinatorics

We write

= Hom(

Q

p

=

Z

p

) =

Z

np

. Note that

d

(

mn

) is also the number of

lattices

L

of index

p

m

in

. Any lattice

L

has the form

M:

Z

np

for some

matrix

M

2

M

n

(

Z

p

) with det(

M

)

6

= 0, and two matrices give the same lattice

if and only if they are related by reversible column operations.

Notation

Given a sequence

(

k

k+1

:::

n

;

1

) we write

j

j

=

X

j

j

k

k

=

X

j

j

j

:

If

n

= 1 we allow the empty sequence, with

j

j

=

k

k

= 0.

Lemma 43

Every matrix in

M

2

M

n

(

Z

p

) with det(

M

)

6

= 0 can be reduced

by reversible column operations to a unique matrix of the following form (we

show the case

n

= 4, but the generalisation is evident). Equivalently, any

lattice

L

has a unique basis given by the columns of such a matrix.

0

B

B

B

B

B

B

B

B

@

p

0

0 0 0

a

01

p

1

0 0

a

02

a

12

p

2

0

a

03

a

13

a

23

p

3

1

C

C

C

C

C

C

C

C

A

0

a

kl

< p

l

The index of such a lattice is

p

j

j

.

Proof.

We nd an element of minimal valuation in the top row and move the

corresponding column to the left hand end. We then multiply this column by

a unit to get

p

0

at the top left, and then perform column operations to clear

the rest of the top row. By induction, we can reduce the (

n

;

1)

(

n

;

1)

submatrix at the bottom right to the desired form. A few more evident column

operations put the entries in the rst column into the stated range. It is now

not hard to see that

k

is the minimal valuation of

b

k

among vectors of the

form (0

:::

0

b

k

::: b

n

;

1

) in the lattice

L

=

M:

. Moreover, the (

k

+1)'th

column above is the unique such vector with

b

k

=

p

k

and 0

b

l

< p

l

for

l > k

. This gives uniqueness. It is not hard to see that the vectors

b

with

0

b

k

< p

k

form a transversal for

L

in

, so that

j

=L

j

=

p

j

j

.

25

background image

Corollary 44

If

runs over sequences

(

1

:::

n

;

1

) with

j

j

m

then

d

(

mn

) =

X

p

k

k

:

It follows that

Z

n

(

t

) =

X

l

d

(

mn

)

t

m

=

n

;

1

Y

k=0

1

1

;

p

k

t:

Proof.

The rst statement follows by counting the possible matrices in the

lemma in the obvious way. Next, write

(1

;

p

k

t

)

;

1

=

X

k

0

p

k

k

t

k

:

The second claim follows by expanding out the product.

We can also give a somewhat more explicit formula for

d

(

mn

). First, we recall

the de nition of the Gaussian binomial coecients:

2

6

4

k

l

3

7

5

p

=

l

;

1

Y

m=0

p

k

;

p

m

p

l

;

p

m

=

l

Y

m=1

p

k

;

l+m

;

1

p

m

;

1

:

This can also be interpreted as the number of

l

-dimensional subspaces of a

k

-dimensional vector space over

F

p

.

Lemma 45

The Gaussian binomial coecients satisfy

2

6

4

k

l

3

7

5

p

=

2

6

4

k

k

;

l

3

7

5

p

and

2

6

4

k

l

3

7

5

p

=

2

6

4

k

;

1

l

;

1

3

7

5

p

+

p

l

2

6

4

k

;

1

l

3

7

5

p

:

Proof.

Suppose dim

F

p

V

=

k

. There is an obvious bijection

W

7!

ann(

W

)

between

l

-dimensional subspaces of

V

and (

k

;

l

)-dimensional subspaces of

26

background image

V

, which gives the rst statement. Now write

V

=

L

U

with dim

L

= 1.

Let

W

V

have dimension

l

. If

W

contains

L

then it has the form

L

W

0

for a unique (

l

;

1)-dimensional subspace of

U

there are

2

6

4

k

;

1

l

;

1

3

7

5

p

such

W

's.

If

W

does not contain

L

then it is the graph of a unique linear map

W

00

;

!

L

for a unique

l

-dimensional subspace

W

00

U

this occurs

p

l

2

6

4

k

;

1

l

3

7

5

p

times.

The second claim follows.

Lemma 46

d

(

mn

) =

2

6

4

n

+

m

;

1

n

;

1

3

7

5

p

:

Proof.

We know that

Z

n

(

t

) =

X

m

d

(

mn

)

t

m

=

n

;

1

Y

k=0

1

1

;

p

k

t:

We want to show that this is the same as

W

n

(

t

) =

X

m

2

6

4

n

+

m

;

1

n

;

1

3

7

5

p

t

m

:

It is enough to show that

Z

1

(

t

) =

W

1

(

t

) (which is immediate) and that

(1

;

p

n

;

1

t

)

W

n

(

t

) =

W

n

;

1

(

t

)

:

This follows easily from the equation

2

6

4

m

+

n

;

1

n

;

1

3

7

5

p

;

p

n

;

1

2

6

4

m

+

n

;

2

n

;

1

3

7

5

p

=

2

6

4

m

+

n

;

2

n

;

2

3

7

5

p

which follows in turn from Lemma 45.

We shall need another auxiliary numerical function. Fix

m

and

n

. For 0

l

n

and 0

k

we de ne

e

(

kl

) =

X

p

k

k

where

= (

l

:::

n

;

1

) and

k

+

j

j

m:

27

background image

Of course, we have

e

(0

1) =

d

(

mn

). If

l

=

n

we again allow

to be the

empty sequence, with

j

j

=

k

k

= 0. Thus

e

(

kn

) = 1 provided that

k

m

.

Lemma 47

e

(

kl

) is the unique function with the properties

e

(

kn

) = 1

if

k

m

e

(

kl

) = 0

if

k > m

e

(

kl

) =

e

(

kl

+ 1) +

p

l

e

(

k

+ 1

l

) if

l < n:

Proof.

It is clear that

e

(

kl

) has the rst two properties. For the third, recall

that

e

(

kl

) =

X

f

p

k

k

j

= (

l

:::

n

;

1

) and

k

+

j

j

m

g

:

The sum of the terms with

l

= 0 is just

e

(

kl

+ 1). If

l

>

0 then we can

write

l

=

l

;

1 and

j

=

j

for

j > l

. Then

j

j

= 1+

j

j

and

k

k

=

l

+

k

k

.

The sum of the terms with

l

>

0 is thus

X

f

p

l

p

k

k

j

= (

l

:::

n

;

1

) and

k

+ 1 +

j

j

m

g

:

This is just

p

l

e

(

k

+ 1

l

). It is easy to see that these properties characterise

e

uniquely.

10.2 Innitesimal Theory

In this section, we study the scheme Sub

m

(

G

0

) (which is the same as

X

0

X

Sub

m

(

G

)). We rst give a lemma which is slightly more general than what we

need in this section, but the extra generality will be useful later.

Lemma 48

Let

B

be a nite-dimensional

-algebra, and suppose that ele-

ments

b

0

::: b

m

2

B

satisfy

Q

i

b

n

i

i

= 0 with

n

i

>

0 for all

i

. Then

dim

B

X

i

n

i

dim

B=b

i

:

Proof.

Write

b

0

=

Q

i>0

b

n

i

i

(so that

b

n

0

0

b

0

= 0). For 0

j < n

0

, the space

Bb

j

0

=Bb

j+1

0

is a cyclic module over

B=b

0

. Moreover,

Bb

n

0

0

is a cyclic module

for

B=b

0

. It follows that

dim

B

n

0

dim

B=b

0

+ dim

B=b

0

:

28

background image

By induction on

m

, we nd that

dim

B=b

0

X

i>0

n

i

dim

B=

(

b

0

b

i

) =

X

i>0

n

i

dim

B=b

i

:

The claim follows.

Proposition 49

deg Sub

m

(

G

)

;

!

X

]

d

(

mn

).

Proof.

Write

Y

= Sub

m

(

G

0

) and

E

0

=

O

Y

and

H

=

G

X

Y

. Note that

H

has strict height

n

. We have a universal subgroup

K <

H

of degree

p

m

and a

quotient map

q

:

H

;

!

H

=K

. Note that

q

has height

m

and

H

=K

has height

n

. Let

Y

(

kl

) be the closed subscheme of

Y

on which

q

has strict height at

least

k

and

H

=K

has strict height at least

l

, and write

E

0

(

kl

) =

O

Y (kl)

e

0

(

kl

) = dim

E

0

(

kl

)

:

Because

p

= 0 in

we have

Y

=

Y

(0

1). Clearly also

Y

(

kl

) =

and

e

0

(

kl

) = 0 if

k > m

.

Next, recall that

Y

= Sub

m

(

G

0

) is a closed subscheme of Div

p

m

(

G

0

). To see

what this means more explicitly, choose coordinates

x

and

y

on

H

and

H

=K

.

By the Weierstrass preparation theorem, we can write

q

y

=

f

(

x

)

u

(

x

) with

f

a

monic polynomial of degree

p

m

congruent to

x

p

m

modulo

m

Y

, and

u

invertible.

It follows that

f

is just the monic polynomial

f

K

classifying the divisor

K

, and

thus that the coecients of

f

(excluding the top one) generate the maximal

ideal of

E

0

. Over

Y

(

ml

) we have

q

y

= 0 (mod

x

p

m

) from which we see that

f

(

x

) =

x

p

m

. It follows that the maximal ideal of

E

0

(

ml

) is zero, so that

e

0

(

ml

)

1.

Next, I claim that

e

0

(

kl

)

e

0

(

kl

+1)+

p

l

e

0

(

k

+1

l

) when

l < n

. To see this,

we work temporarily over

E

0

(

kl

). There are elements

uva

2

E

0

(

kl

) with

u

invertible such that

p

H

x

=

ux

p

n

+

p

H

=K

y

=

vy

p

l

+

q

y

=

ax

p

k

+

Note that

E

0

(

kl

)

=a

=

E

0

(

k

+ 1

l

) and

E

0

(

kl

)

=v

=

E

0

(

kl

+ 1). The relation

p

H

=K

q

=

q

p

H

gives

va

p

l

x

p

k

+

l

+

=

au

p

k

x

p

k

+

n

+

:

29

background image

Because we assume that

l < n

, we nd that

va

p

l

= 0. The claim now follows

from Lemma 48.
We next consider the analogous situation with

l

=

n

and

k < m

, so that

a

lies

in the maximal ideal of

E

0

. By the same argument, we nd that

va

p

n

=

au

p

k

,

so that

a

(1

;

vu

;

p

k

a

p

n

;

1

) = 0

:

The term in parentheses is a unit, so

a

= 0. This shows that

E

0

(

kn

) =

E

0

(

k

+ 1

n

). By induction, we see that

E

0

(

kn

) =

E

0

(

mn

), so that

e

0

(

kn

) =

e

0

(

mn

)

1.

From the above and Lemma 47 we see that

e

0

(

kl

)

e

(

kl

). In particular,

dim(

E

0

) =

e

0

(0

1)

e

(0

1) =

d

(

mn

) as claimed.

10.3 Rational Theory

In this section, we shall assume that

O

X

is an integral domain in which

p

6

= 0,

that

K

is a subgroup of

G

of degree

p

m

, and that we are given a level-

m

structure

: (

m

)

;

!

G

. For

a

2

(

m

) we write

x

a

=

x

(

(

a

))

2

O

X

.

For any subgroup

A

of order

p

m

of (equivalently, of (

m

)), we have a

subgroup-scheme

A

] of

G

, with degree

p

m

. This corresponds to a section

A

:

X

;

!

Sub

m

(

G

). Putting these together, we get a map

Sub

m

()

X

;

!

Sub

m

(

G

)

or equivalently

O

Sub

m

(

G

)

;

!

F

(Sub

m

()

O

X

)

:

(The target ring

F

(Sub

m

()

O

X

) is the ring of functions from the nite set

Sub

m

() to the ring

O

X

, with pointwise operations.)

Proposition 50

The resulting map

O

Sub

m

(

G

)

1

p

]

;

!

F

(Sub

m

()

O

X

1

p

])

is surjective with nilpotent kernel.
The proof will be given after some intermediate results. We write

a

b

if and

only if

a

is a unit multiple of

b

.

30

background image

Firstly, recall from Theorem 24 that

(

m

)] =

G

(

m

), and hence that

(

p

m

G

)

x

Y

a

2

(

m)

(

x

;

x

a

)

:

On the other hand, for any formal group we have (

p

m

G

)

x

=

p

m

x

(mod

x

2

). It

follows that

Q

a

6

=0

x

a

p

m

.

Consider the discriminant

" =

Y

a

6

=

b

(

x

a

;

x

b

)

:

Write

N

=

j

(

m

)

j

=

p

mn

. It follows easily from the above and Lemma 15 that

"

Y

a

6

=

b

x

a

;

b

0

@

Y

a

6

=0

x

a

1

A

N

p

mN

:

In particular, " becomes invertible when we invert

p

.

Now write

E

0

=

O

Sub

m

(

G

)

, and

p

A

= ker(

A

)

C

E

0

. Note that

E

0

=

p

A

=

O

X

.

Lemma 51

If

A

6

=

B

then

p

is nilpotent mod

p

A

+

p

B

.

Proof.

We may suppose that there exists

a

2

A

n

B

. Modulo

p

B

we have

f

K

(

x

) =

Q

b

2

B

(

x

;

x

b

), and modulo

p

A

we have

f

K

(

x

a

) = 0. Thus, using

Lemma 15 we get

0 =

f

K

(

x

a

) =

Y

b

2

B

(

x

a

;

x

b

)

Y

b

x

a

;

b

(mod

p

A

+

p

B

)

:

Each term in the product

Q

b

2

B

(

x

a

;

x

b

) divides " and thus divides a power

of

p

. Thus

p

is nilpotent mod

p

A

+

p

B

as claimed.

Lemma 52

If

p

is a prime ideal in

E

0

and

p

62

p

then

p

p

A

for some

A

.

Proof.

We know that

K

G

(

m

), so

f

K

(

x

) divides

f

G

(

m)

(

x

) =

Q

a

2

(

m)

(

x

;

x

a

). Now work over the integral domain

E

0

=

p

. We know that the discriminant

of

f

G

(

m)

is nonzero, so the roots are distinct. We must therefore have

f

K

(

x

) =

Q

a

2

A

(

x

;

x

a

) for some uniquely determined subset

A

(

m

). As

K

is a

subgroup scheme of

G

, it must be invariant under translation by any of its

31

background image

points. From this, it follows easily that

A

is a subgroup of (

m

), and that

p

p

A

.

Proof of Proposition 50

The last two lemmas show that

p

A

1

p

] +

p

B

1

p

] =

E

0

1

p

] and

\

A

p

A

1

p

] =

\

p2

spec

E

0

1p]

p

=

p

0

:

The proposition follows by the Chinese remainder theorem.

Corollary 53

The map Sub

m

(

G

)

;

!

X

is at, with degree

d

(

mn

).

Proof.

Write

E

=

O

X

and

d

=

d

(

mn

). By Proposition 49 we know that

there is an

E

-linear epimorphism

E

d

;

!

E

0

. It follows that the induced map

f

0

:

E

1

p

]

d

;

!

E

0

1

p

]

;

!

E

0

1

p

]

=

p

0

is epi. By Proposition 50, this is an epimorphism between two free modules of

the same rank, so it is iso. By assumption

E

is an integral domain in which

p

6

= 0, so

E

d

;

!

E

1

p

]

d

is mono. By considering the following square, we see

that

f

is mono and thus iso (and thus that 0 =

p

0

C

E

0

).

E

d

E

0

E

1

p

]

d

E

0

1

p

]

=

p

0

'

E

1

p

]

d

w

w

f

v

u

u

w

w

f

0

10.4 General Theory

We can now prove Theorem 42, except for the Gorenstein property, which is

treated in the next subsection. Let

G

be a formal group of height

n

over an

arbitrary connected base

X

. Let

G

0

=X

0

be the universal deformation of

G

0

=X

0

,

and let

G

0

0

be the pullback of

G

0

to Level(

m

G

0

). We can apply Corollary 53

to conclude that the projection

Sub

m

(

G

0

0

) = Level(

m

G

0

)

X

0

Sub

m

(

G

0

)

;

!

Level(

m

G

0

)

32

background image

is at, with degree

d

(

mn

). As Level(

m

G

0

)

;

!

X

0

is faithfully at, it is not

hard to conclde that Sub

m

(

G

0

)

;

!

X

0

is also at, with degree

d

(

mn

). As

Sub

m

(

G

) =

X

X

0

Sub

m

(

G

0

), we see that Sub

m

(

G

)

;

!

X

is again at, with

the required degree.

10.5 The Gorenstein property

We next show that the scheme Sub

m

(

G

) is Gorenstein (I do not know whether

it is a complete intersection). We rst recall the de nition and basic facts about

Gorenstein rings.

Denition 54

A Noetherian local ring

R

is

Gorenstein if the

R

-module

R

admits a nite resolution by injective

R

-modules.

Proposition 55

(a) Let

R

be a Noetherian local ring, and

(

x

1

::: x

m

) a regular sequence in

R

generating an ideal

I

m

. Then

R

is Gorenstein if and only if

R=I

is

Gorenstein.

(b) A regular Noetherian local ring is Gorenstein.

(c) An Artinian local ring

R

is Gorenstein if and only if the socle

soc(

R

) =

f

a

2

R

j

a

m

= 0

g

has dimension one over

=

R=

m

.

Proof.

Part (a) is 2, Proposition 3.1.19(b)]. As a eld is visibly Gorenstein,

part (b) follows. Part (c) is essentially 2, Theorem 3.2.10].

Recall the notation used in the proof of Proposition 49. We write

Y

for the

scheme Sub

m

(

G

0

) and put

H

=

G

X

Y

. We have a universal subgroup

K <

H

of degree

p

m

and a quotient map

q

:

H

;

!

H

=K

. We choose coordinates

x

and

y

on

H

and

H

=K

. There is thus an element

a

2

O

Y

such that

q

y

=

ax

(mod

x

2

). We can also consider the equation

f

K

(

x

) of the divisor

K

, which is

a unit multiple of

q

y

in

O

Y

x

]]. Thus

f

K

(

x

) =

a

0

x

(mod

x

2

) for some element

a

0

2

O

Y

, which is a unit multiple of

a

.

Proposition 56

The ring

O

Sub

m

(

G

)

is Gorenstein. The socle of the quotient

ring

O

Sub

m

(

G

0

)

is generated by the element

a

"

(

p

+

:::

+

p

n

;

1

) (where

a

"

N

means

a

N

). It is also generated by

a

0

"

(

p

+

:::

+

p

n

;

1

).

Proof.

First, recall that

O

Sub

m

(

G

0

)

=

X

O

Sub

m

(

G

)

, which is obtained from

O

Sub

m

(

G

)

by killing the sequence (

u

0

::: u

n

;

1

). This sequence is regular (be-

cause Sub

m

(

G

)

;

!

X

is at). Thus, by part (a) of Proposition 55, it is enough

33

background image

to show that

O

Sub

m

(

G

0

)

is Gorenstein. Moreover, by part (c) of Proposition 55,

it is enough to show that the socle is generated by

a

"

(

p

+

:::

+

p

n

;

1

).

To do this, we reuse the notation and ideas of the proof of Proposition 49. It

was shown there that

e

0

(

kl

)

e

0

(

kl

+ 1) +

p

l

e

0

(

k

+ 1

l

)

(

l < n

)

:

There is a similar relation for the numbers

e

(

kl

), with the inequality replaced

by an equality. Moreover, we have

e

0

(

kn

)

1 =

e

(

kn

)

(

k

m

)

e

0

(

kl

) = 0 =

e

(

kl

)

(

k > m

)

:

Finally, Theorem 42 implies that

e

(0

1) =

d

(

mn

) =

e

0

(0

1). It follows eas-

ily that

e

0

(

kl

) =

e

(

kl

) when

k

m

and

l

n

, and thus that

e

0

(

kl

) =

e

0

(

kl

+ 1) +

p

l

e

0

(

k

+ 1

l

) when

l < n

. In particular, we have

e

0

(0

l

) =

e

0

(0

l

+1)+

p

l

e

0

(1

l

). Recall that the corresponding inequality was derived from

the relation

va

p

l

= 0, with

E

0

(0

l

)

=a

=

E

0

(1

l

) and

E

0

(0

l

)

=v

=

E

0

(0

l

+ 1).

Because we are considering the case

k

= 0, the element

a

here is the same as

that in the statement of the proposition. Using Lemma 57 below, we nd that

multiplication by

a

p

l

gives an isomorphism soc(

E

0

(0

l

+ 1))

'

soc(

E

0

(0

l

)).

It follows that multiplication by

b

=

a

"

(

p

+

:::

+

p

n

;

1

) gives an isomor-

phism soc(

E

0

(0

n

))

'

soc(

E

0

(0

1)). As

e

0

(0

n

) =

e

(0

n

) = 1, we see that

E

0

(0

n

) =

and thus that soc(

E

0

(0

n

)) =

. Moreover,

E

0

(0

1) =

O

Sub

m

(

G

0

)

.

It follows that soc(

O

Sub

m

(

G

0

)

) is generated by

b

, as claimed. The last statement

of the proposition is trivial.

We still owe the reader the following lemma.

Lemma 57

Let

R

be a nite-dimensional local algebra over a eld

, with

maximal ideal

m

. Suppose that

va

2

m

satisfy

va

k

= 0 (for some

k

) and

dim(

R

) = dim(

R=v

) +

k

dim(

R=a

). Then multiplication by

a

k

is a monomor-

phism

R=v

;

!

R

, whose image is the annihilator of

v

. Moreover, this map

induces an isomorphism

soc(

R=v

)

'

soc(

R

).

Proof.

As in the proof of Lemma 48, we lter

R

by the ideals

Ra

i

(where 0

i

k

). The quotients are cyclic modules over

R=a

, and the last ideal

Ra

k

is a

cyclic module over

R=v

, so that dim(

Ra

i

=Ra

i+1

)

dim(

R=a

) and dim(

Ra

k

)

dim(

R=v

). On adding up these inequalities, we obtain the inequality dim(

R

)

dim(

R=v

)+

k

dim(

R=a

), which is by assumption an equality. It follows that all

our inequalities are actually equalities, and thus that all our cyclic modules

34

background image

are actually free of rank one. This means that multiplication by

a

k

gives a

monomorphism

R=v

;

!

R

.

Let

J

be the annihilator of

v

, so that the image of the above map is clearly

contained in

J

. On the other hand, we have short exact sequences

J

;

!

R

v

;

!

Rv

Rv

;

!

R

;

!

R=v

which imply that dim(

J

) = dim(

R=v

). It follows that our monomorphism

a

k

:

R=v

;

!

J

is actually an isomorphism.

It is immediate that

a

k

soc(

R=v

)

soc(

R

). Conversely, suppose that

x

2

soc(

R

). As

v

2

m

, we have

xv

= 0, so that

x

2

J

. It follows that

x

=

a

k

y

for some

y

2

R=v

. Suppose that

z

2

m

. Then

a

k

(

yz

) =

xz

= 0 because

x

2

soc(

R

). As multiplication by

a

k

is a monomorphism, we see that

yz

=

0 in

R=v

. Thus

y

2

soc(

R=v

). This means that

a

k

gives an isomorphism

soc(

R=v

)

'

soc(

R

), as claimed.

11 Flags

Consider a sequence

= (

0

= 0

<

1

< ::: <

m

=

n

). A ag of type

on

G

will mean a sequence of subgroup divisors 0 =

K

0

< K

1

< ::: < K

m

=

G

(1)

such that

K

k

has degree

p

k

. Using Proposition 16 and Theorem 42, we see

that the functor from schemes over

X

to sets given by

Y

7!

f

ags of type

on

G

X

Y

g

is represented by a scheme Flag(

G

) over

X

. This is in fact a closed sub-

scheme of

Q

k

Sub

k

(

G

) and thus is nite over

X

.

Similarly, we let Flag(

) be the set of sequences 0 =

A

0

< A

1

< ::: < A

m

=

(1) such that

j

A

k

j

=

p

k

. Let

a

0

::: a

n

;

1

be the standard basis of (1) =

(

Z

=p

)

n

. Write

k

=

h

a

i

j

i <

k

i

, so that = (

0

:::

m

)

2

Flag(

).

Write ; = Aut((1)) and let ;(

) be the subgroup which preserves . Clearly

Flag(

)

'

;

=

;(

).

If

: (1)

;

!

;(

X

G

) is a level structure then by putting

K

k

=

k

] we

obtain a ag of type

. This construction gives rise to a map Level(1

G

)

;

!

Flag(

G

).

35

background image

Theorem 58

The maps

Level

(1

G

)

;

!

Flag

(

G

)

;

!

X

are at of degree

j

;(

)

j

and

j

Flag

(

)

j

respectively. If

G

is the universal

deformation of

G

0

then Flag

(

G

) is a smooth scheme and Level(1

G

)

;

!

Flag

(

G

) is Galois.

We will prove this theorem at the end of this section.
First, observe that whenever we have a ag of type

we have isogenies

G

=

G

=K

0

q

1

;

!

G

=K

1

q

2

;

!

:::

q

m

;1

;

;

;

!

G

=K

m

;

1

q

m

;

!

G

:

At the last stage, we have identi ed

G

=K

m

=

G

=

G

(1) with

G

via the isomor-

phism

p

G

:

G

=

G

(1)

'

G

. With this convention, we have

q

m

q

m

;

1

q

1

=

p

G

:

G

;

!

G

:

Moreover,

q

k

has height

k

;

k

;

1

.

11.1 Combinatorics

Consider a sequence

= (0 =

0

1

:::

m

n

). Choose linearly

independent subspaces

N

k

(1) of dimension

k

;

k

;

1

and put

M

k

=

L

kj=1

N

j

and

M

0

k

=

L

mj=k+1

N

j

.

Denition 59

d

(

) is the number of ags of type

such that

A

k

\

M

m

=

M

k

for

0

k

m

. This is clearly independent of the choices made. Note that

d

(0

:::

0) =

j

Flag

(

)

j

, that

d

(

) = 0 unless

k+1

;

k

k+1

;

k

for

0

k < m

, and that

d

(

) = 1 if

k

=

k

for

0

k < m

.

Denition 60

If

m

< n

and

0

< k

m

write

k

= (

0

:::

k

;

1

k

+ 1

:::

m

+ 1)

:

Lemma 61

If

m

< n

then

d

(

) =

m

X

k=1

p

m

;

k

d

(

k

)

:

36

background image

Proof.

Choose

L

(1) with dim

F

p

L

= 1 and

L

\

M

m

= 0. Let

A

be a ag

of type

such that

A

j

\

M

m

=

M

j

for all

j

. There is an exact sequence

0

;

!

M

j

;

!

A

j

\

(

M

m

L

)

j

;

!

L:

Note that

m

is epi. Let

k

be the least value of

j

for which

j

is epi (or

equivalently, nonzero). For

j < k

we have

A

j

\

(

M

m

L

) =

M

j

. For

j

k

,

there is a unique map

f

j

:

L

;

!

M

0

j

such that

A

j

\

(

M

m

L

) =

A

j

\

(

M

j

M

0

j

L

) =

M

j

graph(

f

j

)

:

Moreover,

f

j+1

is just the composite

L

f

j

;

!

M

0

j

=

N

j+1

M

0

j+1

;

!

M

0

j+1

It follows that all the

f

j

are determined by

f

k

2

Hom(

LM

0

k

). Moreover, any

homomorphism

L

;

!

M

0

k

can arise in this way. It follows that the number of

A

's with a given value of

k

is

j

Hom(

LM

0

k

)

j

d

(

k

) =

p

m

;

k

d

(

k

). The claim

follows.

11.2 Innitesimal Theory

We next study the scheme

Y

=

X

0

X

Flag(

G

). Consider a sequence

as

above. Write

Y

(

) for the closed subscheme of

Y

on which

q

k

has strict height

at least

k

;

k

;

1

for all

k

. Put

A

(

) =

O

Y ()

and

e

(

) = dim

A

(

). Note

that

e

(0

:::

0) = dim

A

. Recall that we have isogenies

G

=

G

=K

0

q

1

;

!

G

=K

1

q

2

;

!

q

m

;1

;

;

;

!

G

=K

m

;

1

q

m

;

!

G

with

q

m

q

m

;

1

q

1

=

p

G

:

G

;

!

G

:

Proposition 62

deg Flag(

G

)

;

!

X

] = dim

A

j

Flag

(

)

j

.

Proof.

Note that

e

(0

:::

0) = dim

A

= deg Flag(

G

)

;

!

X

]

37

background image

and that

d

(0

:::

0) =

j

Flag(

)

j

so it is enough to prove that

e

(

)

d

(

) for all

. It is clear that

e

(

) = 0

unless

k+1

;

k

k+1

;

k

for 0

k < m

. Next, suppose that

k

=

k

for

0

k < m

. It follows that over

Y

(

) we have

K

k

=

p

k

0] for 0

k < m

. By

construction, the ring

A

is generated over

by the parameters of the divisors

K

k

, so this implies that

;

!

A

(

) is epi and

e

(

)

1. We now need only

prove that when

m

< n

we have

e

(

)

m

X

k=1

p

k

;1

e

(

k

)

:

To see this, we work temporarily over

A

(

). We shall write

a

"

n

for

a

n

where typographically convenient. Choose coordinates

x

k

on the formal groups

G

=K

k

. There are elements

b

k

2

A

(

) such that

q

k

x

k

=

b

k

(

x

k

;

1

"

(

p

k

;

k

;1

)) + higher terms

:

Note that

A

(

)

=b

k

=

A

(

k

). We also have

q

1

q

2

:::q

m

x

m

= (

x

0

"

p

m

)

m

Y

k=1

(

b

k

"

p

m

;

k

) + higher terms

:

On the other hand,

q

1

:::q

m

x

m

=

p

G

x

0

, and this is divisible by

x

p

n

0

(because

Y

lies over

X

0

). By assumption

m

< n

, so we have

m

Y

k=1

(

b

k

"

p

m

;

k

) = 0

:

It follows by Lemma 48 that

dim

A

(

)

X

k

p

m

;

k

dim

A

(

)

=b

k

=

X

k

p

m

;

k

dim

A

(

k

)

:

In other words,

e

(

)

m

X

k=1

p

k

;1

e

(

k

)

:

as required.

38

background image

11.3 Galois Theory

In this subsection we assume that

G

is the universal deformation of

G

0

. As

mentioned earlier, there is a map Level(1

G

)

;

!

Flag(

G

), classifying the

ag

0

]

< ::: <

m

] this clearly factors through Level(1

G

)

=

;(

).

Given a subset

S

(1), we have a divisor

S

] =

P

a

2

S

(

a

)] on

G

over

Level(1

G

). Suppose we have maps Level(1

G

)

f

;

!

Y

g

;

!

X

such that

f

is

epi and

gf

is the usual map. We shall say that

S

is de ned over

Y

(or over

O

Y

) if and only if there is a (unique) divisor

D

on

Y

such that

S

] =

f

D

,

or equivalently if and only if the polynomial

Q

a

2

S

(

t

;

x

a

) lies in the subring

O

Y

t

]

O

Level(1

G

)

t

].

For example, if

S

is invariant under the action of ;(

) then

S

is de ned over

the quotient scheme Level(1

G

)

=

;(

). The following facts are evident:

(i) If

D

0

and

D

1

are de ned over

Y

then so is

D

0

+

D

1

.

(ii) If

D

0

and

D

0

+

D

1

are de ned over

Y

then so is

D

1

.

(iii) If the point

a

and the divisor

D

are de ned over

Y

then so is the translate

T

a

D

.

Lemma 63

Suppose we have epimorphisms Level

(1

G

)

;

!

Y

g

;

!

Y

0

;

!

X

.

Suppose that

a

2

S

A

, that

S

is dened over

Y

0

, and that

O

Y

is generated

by

x

a

over

O

Y

0

. Then

deg(

g

)

j

S

j

.

Proof.

Consider the polynomial

f

(

t

) =

Q

b

2

S

(

t

;

x

b

)

2

O

Y

0

t

], which is monic

and has degree

j

S

j

. Clearly

f

(

x

a

) = 0 so

O

Y

is a quotient of

O

Y

0

t

]

=f

(

t

) and

thus needs at most

j

S

j

generators over

O

Y

0

.

Remark 64

Geometrically, this means that

Y

embeds as a closed subscheme

of the divisor

S

] over

Y

0

.

Proposition 65

deg Level(1

G

)

;

!

Flag

(

G

)]

j

;(

)

j

.

Proof.

We again write

f

a

0

::: a

m

;

1

g

for the standard basis of (1), so that

k

=

h

a

i

j

i <

k

i

. Put

;

k

=

f

2

;(

)

j

(

a

i

) =

a

i

for

i < k

g

:

Clearly

j

;(

)

j

=

n

;

1

Y

k=0

j

;

k

=

;

k+1

j

=

n

Y

k=1

j

;

k

a

k

j

:

39

background image

Suppose that

j

k <

j+1

. A little linear algebra shows that

;

k

a

k

=

j+1

n

h

a

i

j

i < k

i

:

Write

Y

0

= spf(

C

0

) = image(Level(1

G

)

=

;(

)

;

!

Flag(

G

)

x

k

=

x

(

(

a

k

))

2

D

1

=

O

Level(1

G

)

C

k

=

C

0

x

j

j

j < k

]

D

1

Y

k

= spf(

C

k

)

:

We thus have maps

Level(1

G

) =

Y

n

;

!

Y

n

;

1

;

!

:::

;

!

Y

0

;

!

Flag(

G

)

the last of which is a closed embedding.
The map Level(1

G

)

;

!

Flag(

G

) is de ned so that the pullback of the

divisor

K

k

is precisely

k

]. It follows that

k

is de ned over

Y

0

for all

k

.

The point

a

k

of

G

over

Y

n

is actually de ned over

Y

k+1

. Using our description

of ;

k

a

k

above, we see that ;

k

a

k

is de ned over

Y

k

. Applying Lemma 63, we

see that deg

Y

k+1

;

!

Y

k

]

j

;

k

=

;

k+1

j

. It follows that

deg Level(1

G

)

;

!

Flag(

G

)]

Y

k

j

;

k

=

;

k+1

j

=

j

;(

)

j

:

Proof of Theorem 58

We may assume that

G

is the universal deformation

of

G

0

. Write

a

=

j

;(

)

j

and

b

=

j

Flag(

)

j

, and note that

ab

=

j

;

j

. Consider

the maps

Level(1

G

)

f

;

!

Level(1

G

)

=

;(

)

g

;

!

Flag(

G

)

h

;

!

X:

By propositions 62 and 65 we know that deg(

h

)

b

and deg(

gf

)

a

by Theorem 23 we know that

gfh

is at of degree

ab

and it is clear that

deg(

gfh

)

deg(

gf

)deg(

h

). It follows that deg(

gf

) =

a

and deg(

h

) =

b

.

Lemma 7 now tells us that Flag(

G

) is smooth,

f

is Galois and

g

is iso.

Finally, lemmas 2 and 1 tell us that

h

is at.

40

background image

12 Typed Subgroups

Let

A

be a nite Abelian

p

-group of order

p

m

and rank at most

n

. We write

Type(

A

)=

f

subgroups of isomorphic to

A

g

'

Mon(

A

)

=

Aut(

A

)

Sub

m

()

In this section, we investigate what it would mean to replace with

G

. We

shall de ne a scheme Type(

A

G

) of \ nite subgroups of

G

of type

A

".

Remark 66

This description is somewhat misleading, because it suggests that

Type

(

A

G

) should be a subscheme of Sub

m

(

G

). There is a natural map

Type

(

A

G

)

;

!

Sub

m

(

G

)

but it is not usually an embedding. The smallest case where it fails to be an

embedding is

n

= 3,

A

=

Z

=

4

Z

=

2 | I have proved this by elaborate calcula-

tion. See also Theorem 70 below, and the example at the end of Section 16. It

would be interesting to have a compelling moduli interpretation of Type

(

A

G

),

but I cannot at present oer one.

Denition 67

If

G

is the universal deformation of

G

0

, then we take

Type

(

A

G

) = Level(

A

G

)

=

Aut(

A

)

:

In the general case, we let

H

=Y

be the universal deformation of

G

0

=X

0

and

take

Type

(

A

G

) = Type(

A

H

)

Y

X:

Over the course of this section, we shall prove the following theorem.

Theorem 68

The maps

Level

(

A

G

)

;

!

Type

(

A

G

)

;

!

X

are at, of degree

j

Aut(

A

)

j

and

j

Type

(

A

)

j

respectively. If

G

is the universal

deformation of

G

0

then Type

(

A

G

) is smooth.

We next consider the scheme

`

(

A)

Type(

A

G

), where the coproduct runs over

the isomorphism classes of Abelian groups

A

of order

m

and rank at most

n

.

41

background image

Theorem 69

Let

f

be the evident map

f

:

a

(

A)

Type

(

A

G

)

;

!

Sub

m

(

G

)

:

Suppose that

G

is the universal deformation of

G

0

. Then the induced map

f

of rings is injective, and becomes iso after inverting

p

.

Theorem 70

Suppose that

A

has the form

Z

=p

m+l

(

Z

=p

l

)

n

;

1

. Then

Type

(

A

G

)

;

!

Sub

m+nl

(

G

)

is a closed embedding. In particular, when

n

= 2 (the case arising from a

deformation of a supersingular elliptic curve), this holds for all

A

.

We now start work on the proofs.
Let

A

be a nite Abelian

p

-group. De ne

U

k

(

A

)=

f

a

2

A

(

k

)

j

p

k

;

1

a

2

p

k

A

g

T

k

(

A

)=

A

(

k

)

=U

k

(

A

)

S

k

(

A

)=

A

(

k

)

n

U

k

(

A

)

:

Note that

U

k

and

T

k

are additive functors, and that if

A

'

Z

=p

l

is a cyclic

group then

T

k

(

A

) =

8

<

:

A=p

'

Z

=p

if

k

=

l

0

otherwise

Now choose an isomorphism

A

'

M

k

0

Z

=p

k

m

k

Write

a

kl

(with 0

l < m

k

) for the generators of the cyclic summands. We

shall order the pairs (

kl

) as follows:

(

ij

)

<

(

kl

) i

i > k

or (

i

=

k

and

j < l

)

:

Note that the ordering on the rst index is reversed, so that the generators

a

kl

of the largest cyclic summands come rst in the ordering.

42

background image

For 0

l < m

k

we de ne

V

(

kl

) =

h

a

k0

::: a

kl

;

1

i

+

U

k

(

A

)

A

(

k

)

V

(

kl

) =

V

(

kl

)

=U

k

(

A

)

T

k

(

A

)

and

S

(

kl

) =

A

(

k

)

n

V

(

kl

)

so

S

(

k

0) =

S

(

k

)

:

We also write

; = Aut(

A

)

and

;(

kl

) =

f

2

;

j

(

a

ij

) =

a

ij

if (

ij

)

<

(

kl

)

g

so

;(

km

k

) = ;(

k

;

1

0)

and

;(

kl

)

=

;(

kl

+ 1)

'

;(

kl

)

a

kl

:

Lemma 71

;(

kl

)

a

kl

=

S

(

kl

) and

j

;

j

=

Q

kl

j

S

(

kl

)

j

.

Proof.

As ;(

kl

)

=

;(

kl

+ 1)

'

;(

kl

)

a

kl

, the second statement will follow

easily from the rst.
Any

2

; preserves

S

(

k

) and induces an automorphism of

T

k

(

A

). If

2

;(

kl

) then it is easy to see that

(

a

kl

)

2

S

(

kl

), lest the induced endomor-

phism of

T

k

(

A

) fail to be iso. Conversely, suppose

b

2

S

(

kl

). Write

A

0

= (

Z

=p

k

)

m

k

A

00

=

M

l

6

=

k

(

Z

=p

l

)

m

l

:

43

background image

Let

b

0

and

b

00

be the components of

b

in

A

0

and

A

00

. By linear algebra over

F

p

we can nd an automorphism

0

of

T

k

(

A

) =

A

0

=p

with

0

(

a

kj

) =

a

kj

for

j < l

and

0

(

a

kl

) =

b

0

(mod

p

). We can then lift this to get an automorphism

1

of

A

0

with

1

(

a

kj

) =

a

kj

for

j < l

and

1

(

a

kl

) =

b

0

. Finally, we de ne an

endomorphism

of

A

=

A

0

A

00

, with components as follows:

A

0

;

!

A

0

1

A

0

;

!

A

00

a

kl

7!

b

00

other generators

7!

0

A

00

;

!

A

0

0

A

00

;

!

A

00

identity

It is easy to check that this is iso, and

(

a

kl

) =

b

. Thus

;(

kl

)

a

kl

=

S

(

kl

)

:

Remark 72

It is neither hard nor apparently helpful to write explicit formulae

for

j

S

(

kl

)

j

and

j

;

j

.

We assume until further notice that

G

is the universal deformation of

G

0

. As

in Subsection 11.3, we shall say that a subset

S

A

is de ned over a quotient

scheme

Y

of Level(

A

G

) if there is a divisor

D

on

G

over

Y

whose pullback

is

S

].

Proof of Theorem 68

We now write

D

A

=

O

Level(

A

G

)

and

C

=

D

;A

=

O

Type(

A

G

)

. De ne

x

ij

=

x

(

(

a

ij

))

C

(

kl

) =

C

x

ij

j

(

ij

)

<

(

kl

)]

D

A

Y

(

kl

) = spf(

C

(

kl

))

:

Note that

Y

(

km

k

) =

Y

(

k

;

1

0) and that we have maps

Level(

A

G

)

;

!

Y

(

kl

+ 1)

;

!

Y

(

kl

)

;

!

Type(

A

G

)

:

Because

U

k

(

A

) and

a

k0

::: a

kl

;

1

are de ned over

Y

(

kl

), we see that

S

(

kl

)

is de ned over

Y

(

kl

). Moreover,

a

kl

2

S

(

kl

) and

C

(

kl

+1) is generated over

C

(

kl

) by

x

kl

. It follows by Lemma 63 that

deg

Y

(

kl

+ 1)

;

!

Y

(

kl

)]

j

S

(

kl

)

j

:

44

background image

Combining this with Lemma 71, we see that

deg Level(

A

G

)

;

!

Type(

A

G

)]

j

;

j

:

It follows from Lemma 7 that Level(

A

G

)

;

!

Type(

A

G

) is Galois and that

Type(

A

G

) is smooth. Lemmas 2 and 1 now tell us that Type(

A

G

)

;

!

X

is

at. As deg Level(

A

G

)

;

!

X

] =

j

Mon(

A

)

j

, we see that

deg Type(

A

G

)

;

!

X

] =

j

Mon(

A

)

j

=

j

;

j

=

j

Type(

A

)

j

:

Proof of Theorem 69

Write

E

0

=

O

Sub

m

(

G

)

. In Subsection 10.3, we con-

structed a map

g

:

D

m

E

E

0

;

!

F

(Sub

m

()

D

m

)

:

We proved there that after inverting

p

,

g

becomes epi with nilpotent kernel.

In the light of Theorem 42, we conclude that

g

is mono, and becomes iso after

inverting

p

. We now put ; = Aut((

m

)), and note that

g

is equivariant for

the evident actions of ;. Passing to xed points, we obtain a map

g

;

: (

D

m

E

E

0

)

;

;

!

F

(Sub

m

()

D

m

)

;

which is again mono, and becomes iso after inverting

p

.

By Theorem 34, we know that

D

;m

=

E

. Because

E

0

is free over

E

, we conclude

that (

D

m

E

E

0

)

;

=

E

0

. Next, for each isomorphism class of Abelian

p

-groups

of order

m

and rank at most

n

, choose a representative subgroup

A

(

m

).

Write

S

for the set of chosen representatives. For

A

2

S

write ;

A

=

f

2

;

j

(

A

) =

A

g

. We have an isomorphism of ;-sets

Sub

m

()

'

a

A

2

S

;

=

;

A

:

It is not hard to conclude that

F

(Sub

m

()

D

m

)

;

=

Y

A

2

S

D

;

A

m

:

Lemma 36 implies that ;

A

;

!

Aut(

A

) is epi write ;

0

A

for the kernel. Using

Theorem 34 we nd that

D

;

A

m

= (

D

;

0

A

m

)

Aut(

A)

=

D

Aut(

A)

A

=

C

A

:

45

background image

Thus,

g

;

is a map

E

0

;

!

Y

A

2

S

C

A

:

One can check that this is the same map as that described in the statement

of the theorem.

Denition 73

Let

K

be the universal subgroup over Sub

m

(

G

). We write

p

;

r

K

for the kernel of the composite

G

p

r

G

;

!

G

q

;

!

G

=K:

Thus

p

;

r

K

is a subgroup divisor of degree

p

m+nr

dened over Sub

m

(

G

), and

its equation

f

p

;

r

K

(

x

) is a unit multiple of

f

K

((

p

r

G

)

x

). The subgroup

p

;

r

K

is

classied by a map

p

;

r

: Sub

m

(

G

)

;

!

Sub

m+nr

(

G

)

:

Lemma 74

The map

p

;

r

: Sub

m

(

G

)

;

!

Sub

m+nr

(

G

) is a closed embedding,

and identies Sub

m

(

G

) with the scheme of subgroups of degree

m

+

nr

which

contain

G

(

r

).

Proof.

Over Sub

m+nr

(

G

), we have an exact sequence of formal groups

G

(

r

)

;

!

G

p

r

G

;

!

G

:

We know by Proposition 16 that there is a closed subscheme

Z

Sub

m+nr

(

G

)

which is universal for subgroups containing

G

(

r

). Let

K

0

be the universal

subgroup over

Z

and

q

0

:

G

;

!

G

=K

0

the projection. Then the composite

G

(

r

)

;

!

G

;

!

G

=K

0

is zero so there is a map (of formal groups over

Z

)

q

:

G

;

!

G

=K

0

with

q

p

r

G

=

q

0

. The kernel

K

of

q

therefore has degree

p

m

, and is classi ed by a map

f

:

Z

;

!

Sub

m

(

G

). Clearly

p

;

r

: Sub

m

(

G

)

;

!

Sub

m+nr

(

G

) factors through

Z

, and the maps

Z

f

;

!

Sub

m

(

G

)

p

;

r

;

;

!

Z

are

mutually inverse.

Proposition 75

Suppose that

A

is such that

A

(

r

)

'

(

r

) (so that

A

has

rank

n

and every cyclic factor has length at least

r

). Write

A

0

=

A=A

(

r

) and

j

A

0

j

=

p

m

. Then there is a canonical isomorphism Type

(

A

G

)

'

Type

(

A

0

G

),

46

background image

and the following diagram commutes:

Type

(

A

G

)

Sub

m+nr

(

G

)

Type

(

A

0

G

)

Sub

m

(

G

)

w

u

'

w

u

p

;

r

Proof.

It is easy to see that multiplication by

p

r

induces a bijection between

Type(

A

) and Type(

A

0

). Thus

j

Epi(

A

)

j

j

Aut(

A

)

j

=

j

Type(

A

)

j

=

j

Type(

A

0

)

j

=

j

Epi(

A

0

)

j

j

Aut(

A

0

)

j

:

Let

u

:

A

;

!

A

0

be the projection. Any automorphism

of

A

induces an

automorphism

0

of

A

0

, with

0

u

=

u

. It follows easily that there is a map

u

0

!

making the following diagram commute:

Level(

A

G

)

Type(

A

G

)

Level(

A

0

G

)

Type(

A

0

G

)

w

f

u

u

!

u

u

0

!

w

f

0

By Theorem 41, we know that

u

!

is a nite at map of degree

j

Epi(

A

)

j

=

j

Epi(

A

0

)

j

:

On the other hand,

f

and

f

0

are nite at maps of degree

j

Aut(

A

)

j

and

j

Aut(

A

0

)

j

respectively. It follows (using lemmas 2 and 1) that

u

0

!

is a nite at

map of degree

j

Epi(

A

)

j

j

Epi(

A

0

)

j

j

Aut(

A

0

)

j

j

Aut(

A

)

j

= 1

:

In other words,

u

0

!

is iso as claimed. It is not hard to see that the diagram

commutes.

Proof of Theorem 70

We may as usual assume that

G

is the universal

deformation. First, suppose that

A

is cyclic, say

A

'

Z

=p

m

generated by

a

.

47

background image

We need to prove that

C

A

is generated by the parameters of the divisor

A

].

As discussed at the end of Section 8, we can write (

p

m

G

)

x

=

g

(

x

)(

p

m

;

1

G

)

x

for a

series

g

(

x

)

2

E

x

]] of Weierstrass degree

s

=

p

nm

;

p

n(m

;

1)

. Thus,

g

(

x

) is a unit

multiple of a monic polynomial

h

(

x

) of degree

s

with

h

(

x

) =

x

s

(mod

m

E

).

Moreover,

D

A

=

E

x

a

]

=h

(

x

a

), where

x

a

=

x

(

(

a

)). Write

D

0

A

=

D

A

=

m

E

and

C

0

A

=

C

A

=

m

E

, so that

D

0

A

is just

x

a

]

=x

sa

. We shall say that an element of

D

0

A

has valuation

t

if it is divisible by

x

ta

but not by

x

t+1

a

. For

j

2

Z

we have

x

ja

=

x

(

(

ja

)) =

jx

a

(mod

x

2a

), so this has valuation 1 if

j

6

= 0 (mod

p

) and

valuation greater than 1 otherwise. Now write

r

=

p

m

;

p

m

;

1

(which divides

s

)

and let

u

2

C

A

be the

r

'th symmetric polynomial in

f

x

ja

j

0

j < p

m

g

. This is

the

r

'th parameter of the divisor

A

], so it lies in the image of

O

Sub

m

(

G

)

;

!

C

A

.

Let

u

0

be the image of

u

in

D

0

A

. If we write

u

0

in the usual way as a sum of

r

-fold products, then there is one term

Q

f

x

ja

j

j

6

= 0 (mod

p

)

g

of valuation

r

, and the other terms have higher valuation. Thus

u

0

itself has valuation

r

,

and the subring of

D

0

A

generated by

u

0

has dimension

s=r

as a vector space

over

. On the other hand, one can see that

s=r

=

j

Type(

A

)

j

=

j

C

A

:

E

j

.

Consider the maps

E

s=r

=

E

f

1

u::: u

s=r

;

1

g

i

;

!

C

A

j

;

!

D

A

and the resulting maps

s=r

=

f

1

u

0

:::

(

u

0

)

s=r

;

1

g

i

0

;

!

C

0

A

j

0

;

!

D

0

A

We have just seen that

j

0

i

0

is mono, so

i

0

is mono. By dimension count,

i

0

is

iso. It follows that

i

is epi, and thus that

O

Sub

m

(

G

)

;

!

C

A

is epi as claimed.

Now suppose that

A

'

Z

=p

r+m

(

Z

=p

r

)

n

;

1

. Then

A

0

=

A=A

(

r

) is cyclic.

Proposition 75 gives a commutative diagram as follows:

Type(

A

G

)

Sub

m+nr

(

G

)

Type(

A

0

G

)

Sub

m

(

G

)

w

u

'

w

u

p

;

r

We have just shown that the bottom map is a closed embedding. The map

p

;

r

is a closed embedding by Lemma 74. It follows that the top map is also a

closed embedding, as claimed.

48

background image

13 Deformation of Isogenies

Consider two schemes

X

0

and

X

0

0

, each of which is spec of a eld of charac-

teristic

p

. Suppose we have a morphism of formal groups as follows:

G

0

G

0

0

X

0

X

0

0

w

q

0

u

u

w

f

0

We shall suppose that the induced map

G

0

;

!

f

0

G

0

0

is an isogeny of degree

p

m

. We would lose little generality by requiring

f

0

to be the identity, but in

algebraic topology we naturally encounter the case in which

f

0

is a power of

Frobenius. By a deformation of

q

0

we shall mean a morphism

q

:

H

;

!

H

0

of

formal groups over a scheme

Y

, where

H

and

H

0

are deformations of

G

0

and

G

0

0

respectively, and the restriction of

q

over

Y

0

is compatible with

q

0

. In more

detail, we de ne

H

0

and

H

0

0

to be the restrictions of

H

and

H

0

to

Y

0

, and we

require a commutative diagram as follows:

H

H

0

G

0

H

0

H

0

0

G

0

0

Y

Y

0

X

0

Y

Y

0

X

0

0

u

u

]

]

q

u

x

]

w

]

q

0

u

u

u

u

u

x

u

u

w

u

u

u

u

]

1

u

x

]

1

w

]

f

0

u

x

w

The diagram contains three parallel vertical squares. The middle one is by

de nition the restriction of the left-hand one over the special bre

Y

0

Y

.

The back face of the right hand cube is required to be a pullback square,

making

H

into a deformation of

G

0

. Similarly for the front face of the right

hand cube. This forces

q

to be an isogeny of degree

p

m

.

Our next task is to classify such deformations. First, let

G

=X

be the universal

deformation of

G

0

. Let

a

: Sub

m

(

G

)

;

!

X

be the usual projection, and let

K < a

G

be the universal example of a subgroup of degree

p

m

. As Sub

m

(

G

) is

a closed subscheme of Div

p

m

(

G

) and Div

p

m

(

G

)

0

=

X

0

, we see that Sub

m

(

G

)

0

=

X

0

. There is a unique subgroup of order

p

m

of

G

0

de ned over

X

0

, viz. the

49

background image

divisor

p

m

0] = spf(

O

G

0

=x

p

m

). In particular,

K

0

=

p

m

0] = ker(

q

0

). It follows

that there is a pullback diagram as shown below.

(

a

G

=K

)

0

G

0

=p

m

0]

G

0

0

Sub

m

(

G

)

0

X

0

X

0

0

u

w

u

w

q

0

u

w

a

0

w

f

0

Using this, we can consider the projection

a

G

;

!

a

G

=K

as a deformation

of

q

0

. It is not hard to check that it is the terminal object in the category of

deformations.
Now let

G

0

=X

0

be the universal deformation of

G

0

0

=X

0

0

. The above construction

also exhibits

a

G

=K

as a deformation of

G

0

0

. It is therefore classi ed by a map

b

: Sub

m

(

G

)

;

!

X

0

extending the map

b

0

=

f

0

a

0

: Sub

m

(

G

)

0

;

!

X

0

0

.

Proposition 76

b

is a nite at map, of degree

j

Sub

m

()

j

.

Proof.

Recall from Theorem 69 that we have a dominant map

f

:

a

(

A)

Type(

A

G

)

;

!

Sub

m

(

G

)

(where

A

runs over isomorphism types of nite Abelian

p

-groups of order

p

m

and rank at most

n

). Moreover,

f

becomes iso when we invert

p

. It is easy to

see that the composite

Level(

A

G

)

;

!

Type(

A

G

)

;

!

Sub

m

(

G

)

b

;

!

X

0

can be identi ed with the map 0

!

of Section 9. By part (7) of Theorem 41, this

map is at and has degree

j

Mon(

A

)

j

. Moreover, Level(

A

G

)

;

!

Type(

A

G

)

is a Galois covering with Galois group Aut(

A

). Using lemmas 2 and 1, we see

that Type(

A

G

)

;

!

X

0

is at. The degree is clearly

j

Mon(

A

)

j

=

j

Aut(

A

)

j

=

j

Type(

A

)

j

. It follows that

b

f

is a at map of degree

P

A

j

Type(

A

)

j

=

j

Sub

m

()

j

, and thus that

O

Sub

m

(

G

)

1

p

] is free of this rank over

O

X

0

1

p

].

We next let

Y <

Sub

m

(

G

) be the preimage of

X

0

0

under

b

, and write

0

=

O

X

0

0

and

E

0

=

O

Y

. I claim that dim

0

E

0

=

j

Sub

m

()

j

. The proof is similar to

that in Subsection 10.2. To recall some notation from there: Given a sequence

(

k

k+1

:::

n

;

1

) we write

j

j

=

X

j

j

50

background image

k

k

=

X

j

j

j

:

If

runs over sequences (

1

:::

n

;

1

) with

j

j

m

then

d

(

mn

) =

X

p

k

k

:

Note, however, that

E

0

and

Y

have dierent meanings here from those used

in Subsection 10.2. We write

H

for the restriction of

a

G

to

Y

, and

K

for the

evident subgroup of

H

, so that

H

=K

is a pullback of

G

0

0

and thus has strict

height

n

. Let

q

be the projection

H

;

!

H

=K

. Choose coordinates

x

and

y

on

H

and

H

=K

. Let

Y

(

kl

) be the largest closed subscheme of

Y

over which

H

has strict height at least

k

and

q

has strict height at least

l

, and write

E

0

(

kl

) =

O

Y (kl)

and

f

0

(

kl

) = dim

0

E

0

(

kl

). Working over

E

0

(

kl

) we see

that there are elements

uva

(with

v

invertible) such that

p

H

x

=

ux

p

k

+

p

H

=K

y

=

vy

p

n

+

q

y

=

ax

p

l

+

Thus

E

0

(

k

+ 1

l

) =

E

0

(

kl

)

=u

and

E

0

(

kl

+ 1) =

E

0

(

kl

)

=a

. Reading the

relation

p

H

=K

q

=

q

p

H

to lowest order, we nd that

va

p

n

x

p

l

+

n

+

=

au

p

l

x

p

k

+

l

+

:

If

k < n

then we conclude that

au

p

l

= 0, and then Lemma 48 tells us that

f

0

(

kl

)

f

0

(

kl

+ 1) +

p

l

f

0

(

k

+ 1

l

). If

k

=

n

then

u

is invertible and we have

a

(

u

p

l

;

va

p

n

;

1

) = 0. If also

l < n

then

a

is topologically nilpotent so

u

p

l

;

va

p

n

;

1

is invertible so

a

= 0, which shows that

E

0

(

nl

) =

E

0

(

nl

+ 1) =

E

0

(

nm

).

From the construction of Sub

m

(

G

) one can see that

0

;

!

E

0

(

nm

) is epi, and

that

E

0

(

kl

) = 0 if

k > n

or

l > m

. To summarise, we have

f

0

(

kl

)

f

0

(

kl

+ 1) +

p

l

f

0

(

k

+ 1

l

) if (

k < n

)

f

0

(

nl

)

1

f

0

(

kl

) = 0

if (

l > m

)

We next de ne a function

f

(

kl

) for 0

k

n

and 0

l

by

f

(

kl

) =

X

f

p

k

k

j

= (

1

:::

n

;

k

)

j

j

m

n

;

k

l

g

:

51

background image

We make the usual convention that

f

(

nl

) = 1 for

l

m

and

f

(

kl

) = 0 for

l > m

. I claim that

f

(

kl

)

f

(

kl

+ 1) +

p

l

f

(

k

+ 1

l

) when

k < n:

To see this, consider the sum which de nes

f

(

kl

). The sum of the terms with

n

;

k

> l

is

f

(

kl

+ 1). If

n

;

k

=

l

, we de ne

= (

1

:::

n

;

k

;

2

n

;

k

;

1

+

l

)

:

Note that

j

j

=

j

j

, that

n

;

(

k+1)

l

, and that

k

k

=

k

k

+

l

. Using this

construction, we see that the sum of

p

k

k

for these

's is

p

l

f

(

k

+ 1

l

), as

required. It follows that

f

0

(

kl

)

f

(

kl

), and in particular that

dim

0

(

E

0

) =

f

0

(1

0)

f

(1

0) =

d

(

mn

) =

j

Sub

m

()

j

:

By an argument very similar to that of Subsection 10.4, we conclude that the

map

b

: Sub

m

(

G

)

;

!

X

0

is at and has degree

d

(

mn

).

14 The Connection with Algebraic Topology

In this section, we make a few brief remarks about how formal groups and

moduli problems arise in algebraic topology. Good general references are 1]

and 14]. However, they are not written from an algebro-geometric point of

view, which was rst introduced by Morava 12,13]. This philosophy is de-

veloped in 15]. In the discussion below, we shall take a few liberties with

technical details.
Let

Z

be a topological space. A geometric chain in

Z

is a smooth manifold

(possibly with boundary)

M

equipped with a map

M

;

!

Z

. Geometric chains

form a graded Abelian monoid GC

Z

under disjoint union. Restriction to the

boundary gives a dierential

@

: GC

Z

;

!

GC

;

1

Z

. The homology of this

complex is a graded Abelian group

MO

Z

. If we require that all manifolds

have a given complex structure on the stable normal bundle, we obtain a

dierent group

MU

Z

, the complex bordism group of

Z

. We write

MU

for

MU

(point), which is a ring under cartesian product. By a related geometric

procedure, we can de ne a group

MU

Z

(which is often but not always the

52

background image

MU

-dual of

MU

Z

). In fact

MU

Z

has a natural ring structure. It can be

thought of as an analogue of the Chow ring, which some readers may nd

more familiar. The functor

Z

7!

MU

Z

has a number of formal properties

(Mayer-Vietoris sequences etc.), which can be summarised by saying that it is

a multiplicative generalised cohomology theory. It is the most powerful such

theory which one has any reasonable chance of computing for popular spaces

Z

. See 4] for some justi cation of this claim.

Let

C

P

1

denote the colimit of the nite-dimensional complex projective

spaces

C

P

k

, or equivalently the classifying space of the circle group. This

is itself (homotopy equivalent to) a topological Abelian group. Moreover,

MU

C

P

1

is isomorphic to

MU

x

]]. The group structure on

C

P

1

gives rise

to a coproduct on

MU

C

P

1

and hence a formal group law over

MU

. Lazard

showed that there is a universal example of a ring

L

equipped with a formal

group law

F

, and that

L

is a polynomial algebra over

Z

on countably many

generators. Quillen proved the fundamental theorem that the classifying map

L

;

!

MU

is an isomorphism. This is the source of all connections between

algebraic topology and formal group theory. If we consider instead manifolds

with a splitting of the stable normal bundle as a sum of two complex bundles,

we obtain a ring (

MU

^

MU

)

Z

. If we take

Z

to be a point, this is the univer-

sal example of a ring with two formal group laws and an isomorphism between

them. Using these ideas, we can construct descent data making

MU

Z

into a

sheaf over the stack of formal groups. The cohomology of the dual sheaf

MU

Z

is the

E

2

term of the Adams-Novikov spectral sequence, which converges to

the stable homotopy groups of

Z

.

A formal group

G

over a scheme

X

= spf(

E

) gives rise to a map from

X

to

this stack. By pulling back

MU

Z

, we obtain an algebra

E

Z

over

E

, and thus

a scheme

Z

E

= spf(

E

Z

) over

X

. If the map from

X

to the stack of formal

groups is at, then the functor

Z

7!

E

Z

de nes a generalised cohomology

theory. This atness condition is called Landweber exactness in the topological

literature. It holds in particular if

G

=X

is the universal deformation of a formal

group of nite height over a eld. If

A

is a nite Abelian group and

BA

is the

classifying space of its dual then (

BA

)

E

= Hom(

A

G

). For nite non-Abelian

groups

K

, the generalised character theory of 8] establishes a connection

between

BK

E

and schemes of the form Level(

B

G

). There is also a map from

a certain closed subscheme spf(

R

) of (

B

p

m

)

E

to Sub

m

(

G

), which turns out to

be an isomorphism. Theorem 42 is half of the proof of this fact, the other half

is topology. A topological calculation shows that the socle element is not in

the kernel of the map

O

Sub

m

(

G

0

)

;

!

R=

m

E

R

as the source is a Gorenstein ring

of dimension zero, we conclude that the map is injective. Another elaborate

topological argument computes the dimension of

R=

m

E

R

, showing that it

coincides with the dimension of

O

Sub

m

(

G

0

)

given by Theorem 42.

Using this fact, we obtain extra structure on

Z

E

. Suppose that

ab

are two

53

background image

points of

X

, that

G

a

and

G

b

are the bres of

G

over

a

and

b

, and that

q

:

G

a

;

!

G

b

is an isogeny. A certain topological construction then gives rise

to a map (

Z

E

)

a

;

!

(

Z

E

)

b

, which is functorial in

q

and natural in

Z

. This gives

powerful information about the schemes

Z

E

. It can be reformulated as saying

that a certain kind of generalised Hecke algebra acts on

E

Z

, and on certain

other topologically de ned rings. Certain

Ext

groups over this algebra form

the input to spectral sequences that compute homotopy groups of spaces of

maps of strictly commutative ring spectra, for example. A conjecture of Mike

Hopkins (inspired but not implied by the theory of the Bruhat-Tits building)

would mean that this Hecke algebra has nite global dimension this would be

very helpful for the applications. It should be possible to prove the conjecture

using the results and methods of this paper.

15 Explicit Formulae

Many of the above results were suggested by computer-assisted calculations

using explicit examples of formal group laws. In this section, we record some

useful formulae which will help the reader to carry out such experiments for

herself. We shall assume that

n >

1 and

p >

2 some small modi cations are

needed when in the exceptional cases, but they generally make things easier.

We write

a

b

for

a

=

b

(mod

p

).

We rst de ne

q

=

p

n

l

(

x

) =

X

k

0

x

q

k

=p

k

e

(

x

) =

l

;

1

(

x

)

( so

e

(

l

(

x

)) =

x

)

:

Note that

l

(

px

)

=p

2

Z

x

]], and in fact

l

(

px

)

=p

=

x

(mod

p

)

:

The composition inverse of this series is

e

(

px

)

=p

, so

e

(

px

)

=p

=

x

(mod

p

)

:

We de ne a formal group law

F

1

over

Q

by

F

1

(

xy

) =

e

(

l

(

x

)

l

(

y

))

:

54

background image

It is shown in 7] (for example) that this is actually de ned over

Z

p

(or even

Z

,

but this is irrelevant for us). We write

F

0

(

xy

) for the resulting formal group

over

F

p

. It is known that any formal group law of height

n

over a separably

closed eld of characteristic

p

is isomorphic to

F

0

(see 6, p. 72]).

For any

a

2

Z

p

we write

a

](

x

) =

e

(

al

(

x

))

:

In other words, if

G

1

is the formal group de ned by

F

1

we have

a

](

x

) =

a

G

1

x

.

By applying

l

one checks that

p

](

x

) =

e

(

px

) +

F

1

x

q

:

We next de ne

C

p

m

(

xy

) = (

x

p

m

+

y

p

m

;

(

x

+

y

)

p

m

)

=p:

Lemma 77

C

p

(

xy

)

;

P

p

;

1

k=1

x

k

(

;

y

)

p

;

k

=k

Proof.

This follows easily from Wilson's Theorem (that (

p

;

1)!

;

1), the

equation

(

p

;

k

)! =

p

;

1

Y

l=k

(

p

;

l

)

(

;

1)

p

;

k

(

p

;

1)!

=

(

k

;

1)!

and the binomial expansion of (

x

+

y

)

p

.

Lemma 78

C

q

(

xy

)

C

p

(

x

q=p

y

q=p

)

Proof.

We have

x

q=p

+

y

q=p

= (

x

+

y

)

q=p

+

pz

for some

z

. Raising this to the

p

'th power, we nd that (

x

q=p

+

y

q=p

)

p

= (

x

+

y

)

q

(mod

p

2

). We also have

x

q

+

y

q

= (

x

+

y

)

q

+

pC

q

(

xy

)

x

q

+

y

q

= (

x

q=p

+

y

q=p

)

p

+

pC

p

(

x

q=p

y

q=p

)

:

Subtracting these, we obtain

pC

q

(

xy

) =

pC

p

(

x

q=p

y

q=p

) (mod

p

2

)

:

The claim follows.

55

background image

Note that

xy

divides

C

p

(

xy

), so that

x

q=p

y

q=p

divides

C

q

(

xy

) mod

p

.

We write

v

p

(

n

) for the

p

-adic valuation of

n

.

Lemma 79

If

0

< l < q

k

then

v

p

0

B

@

q

k

l

1

C

A

=

nk

;

v

p

(

l

)

:

Proof.

We can write

0

B

@

q

k

l

1

C

A

=

q

k

l

l

;

1

Y

j=1

q

k

;

j

j :

The terms in the product are clearly

p

-adic units the claim follows.

Lemma 80

x

+

F

0

y

=

x

+

y

+

C

p

(

x

q=p

y

q=p

) (mod (

xy

)

q

p

)

:

Proof.

It is enough to prove this mod

y

q

, for then by symmetry and unique

factorisation it will hold mod (

xy

)

q

. For the rest of the proof, we put

y

q

= 0.

For some

w

2

Z

xy

]] we have

F

1

(

xy

) =

x

+

yw:

Applying

l

, we obtain

l

(

x

) +

l

(

y

) =

l

(

x

+

yw

) =

l

(

x

) +

q

;

1

X

j=1

l

(

j)

(

x

)(

yw

)

j

=j

!

:

Because

y

q

= 0, we have

l

(

y

) =

y

. We conclude that

y

=

q

;

1

X

j=1

l

(

j)

(

x

)(

yw

)

j

=j

!

:

Lemma 79 implies that the series

l

(

j)

(

x

)

=j

! for 0

< j < q

are all integral, and

moreover that

l

(

j)

(

x

)

j

!

1

p

0

B

@

q

j

1

C

A

x

q

;

j

(1

< j < q

)

56

background image

l

0

(

x

)

1

1 + 1

p

0

B

@

q

1

1

C

A

x

q

;

1

:

This implies that

y

yw

+

q

;

1

X

j=1

1

p

0

B

@

q

j

1

C

A

x

q

;

j

yw

;

C

q

(

xyw

)

yw

;

C

p

(

x

q=p

(

yw

)

q=p

)

:

Now de ne

z

by

yw

=

y

+

C

p

(

x

q=p

y

q=p

) +

z:

Note that

y

q=p

divides

C

p

(

x

q=p

y

q=p

). As

n

2, we nd that

y

q

2

=p

2

= 0 it

follows that

C

p

(

x

q=p

y

q=p

)

q=p

= 0, and thus that (

yw

)

q=p

y

q=p

+

z

q=p

. Feeding

this back into our previous equation, we nd

y

y

+

C

p

(

x

q=p

y

q=p

) +

z

;

C

p

(

x

q=p

y

q=p

+

z

q=p

)

so

z

C

p

(

x

q=p

y

q=p

+

z

q=p

)

;

C

p

(

x

q=p

y

q=p

)

:

It follows easily that

z

q=p

divides

z

mod

p

. On the other hand, the de nition

of

z

implies that

y

divides

z

, and thus that

z

is nilpotent. As

z

is nilpotent

and

z

q=p

divides

z

, we see that

z

= 0. The claim follows.

Corollary 81

There is a power series

G

over

F

p

such that

F

0

(

xy

) =

x

+

y

+

G

(

x

q=p

y

q=p

)

:

Proof.

Work in

F

p

xyz

]]

=z

q=p

. Let

k

be an integer, and write

k

=

p

r

l

with

l

6

0. We then have

(

y

+

z

)

k

= (

y

p

r

+

z

p

r

)

l

=

y

k

+

ly

k

;

p

r

z

p

r

+

:

It follows that (

y

+

z

)

k

=

y

k

if

q=p

divides

k

. Next, write

F

0

(

xy

) =

x

+

y

+

X

k>0

a

k

(

x

)

y

k

:

57

background image

The associativity law says that

F

0

(

xF

0

(

yz

)) =

F

0

(

F

0

(

xy

)

z

)

:

By the previous lemma, we have

F

0

(

uz

) =

u

+

z

for all

u

. It follows that

F

0

(

xy

+

z

) =

F

0

(

xy

) +

z

so

X

k>0

a

k

(

x

)(

y

+

z

)

k

=

X

k>0

a

k

(

x

)

y

k

:

We conclude that

a

k

(

x

) = 0 unless

q=p

divides

k

, so that

F

0

(

xy

)

;

x

;

y

is a

function of

y

q=p

. By symmetry, it is also a function of

x

q=p

. The claim follows.

Next, we write

W

=

W

F

q

for the Witt ring of

F

q

. Recall that for each

a

2

F

q

there is a unique element ^

a

2

W

(the Teichm uller representative) such

that ^

a

q

= ^

a

and ^

a

a

. Any element

a

2

W

can be written uniquely as

P

i

0

a

i

p

i

, with

a

q

i

=

a

i

. There is a Frobenius automorphism

F

of

W

de ned

by

F

(

P

a

i

p

i

) =

P

a

p

i

p

i

.

For any

a

2

F

q

we have

l

(^

ax

) = ^

al

(

x

)

:

It follows easily that

e

(^

ax

) = ^

ae

(

x

)

F

1

(^

ax

^

ay

) = ^

aF

1

(

xy

)

:

If we write

e

(

x

) =

X

k

m

k

x

k

F

1

(

xy

) =

X

kl

a

kl

x

k

y

l

then we can conclude that

a

kl

= 0 unless

k

+

l

= 1 (mod

q

;

1), and similarly

for

e

(

x

) etc.

58

background image

Now suppose that

a

2

W

is arbitrary. We de ne

a

](

x

) =

e

(

al

(

x

))

:

By expressing

a

in the form

P

i

^

a

i

p

i

, we can show that this lies in

W

x

]]. The

map

a

7!

a

](

x

) gives a ring homomorphism

W

;

!

End(

F

1

).

We next put

E

=

W

u

1

::: u

n

;

1

]]

u

0

=

p u

n

= 1

:

We also let

denote the endomorphism of

E

which is the Frobenius on

W

and sends

u

k

to

u

p

k

for 0

< k < n

, so that

a

=

a

p

(mod

p

) for all

a

.

Over

E

1

p

] we have a formal power series log

F

(

x

) =

x

+

O

(

x

2

) characterised

uniquely by either of the two following equations:

log

F

(

x

) =

x

+ 1

p

n

X

k=1

log

F

(

u

k

x

p

k

)

log

F

(

x

) =

x

+ 1

p

n

X

k=1

u

k

log

k

F

(

x

p

k

)

:

From either of these equations, one can see that

F

=

F

1

(mod

u

1

::: u

n

;

1

).

The following more explicit formulae are given in 14, Section 4.3]. Consider

a (possibly empty) sequence

I

= (

i

1

::: i

m

) with 0

< i

k

n

for all

k

. Write

j

I

j

=

m

and

k

I

k

=

P

k

i

k

and

j

k

=

P

l<k

i

l

. Finally, put

u

I

=

Y

k

u

p

jk

i

k

:

We then have

log

F

(

x

) =

X

I

u

I

p

j

I

j

x

p

k

I

k

:

We write exp

F

(

x

) for the inverse of log

F

(

x

) (under composition) and set

F

(

xy

) =

x

+

F

y

= exp

F

(log

F

(

x

) + log

F

(

y

))

:

One can show that this lies in

E

xy

]] (rather that

E

1

p

]

xy

]]), so it de nes a

formal group law over

E

.

59

background image

By applying log

F

, one can check that

p

]

F

(

x

) = exp

F

(

px

) +

F

u

1

x

p

+

F

+

F

u

n

;

1

x

p

n

;1

+

F

x

p

n

:

Lemma 82

If

k >

0 then

x

+

F

y

=

x

+

y

+

u

k

C

p

k

(

xy

)

mod (

u

1

::: u

k

;

1

) + (

xy

)

p

k

+1

:

Proof.

We work mod (

xy

)

p

k

+1

and (

u

1

::: u

k

;

1

). This leaves us with a

torsion-free ring, so we can still use the logarithm. In this setting, for

w

2

(

xy

)

we have log

F

(

w

) =

w

+

u

k

w

p

k

=p

and

wC

p

k

(

xy

) = 0. For some

z

2

(

xy

) we

have

F

(

xy

) =

x

+

y

+

u

k

C

p

k

(

xy

) +

z:

Applying log

F

, we get

log

F

(

x

) + log

F

(

y

) = log

F

(

x

+

y

+

u

k

C

p

k

(

xy

) +

z

)

so

x

+

y

+

u

k

x

p

k

=p

+

u

k

y

p

k

=p

=

x

+

y

+

u

k

C

p

k

(

xy

) +

z

+

u

k

(

x

+

y

+

z

)

p

k

=p

z

=

u

k

(

x

+

y

)

p

k

=p

;

u

k

(

x

+

y

+

z

)

p

k

=p:

Thus

z

2

(

xyz

)

z

= (

xy

)

z

. As the ideal (

xy

) is nilpotent, we have

z

= 0 as

claimed.

It follows from this (see 10]) that

F

is the universal deformation of

F

0

.

Lemma 83

F

(

xy

) =

x

+

y

mod

(

xy

) or mod (

y

q=p

u

0

yu

1

y::: u

n

;

1

y

).

Proof.

The rst statement holds for any formal group law. Now work mod

the second ideal. By the rst statement,

F

(

xy

)

;

F

1

(

xy

) is divisible by

xy

.

From the de nitions, all the coecients also lie in the ideal

J

= (

u

1

::: u

n

;

1

).

Because

Jy

= 0, we see that

F

(

xy

) =

F

1

(

xy

). Now consider

F

1

(

xy

)

;

x

;

y

.

Using Lemma 80, we see that every term is divisible either by

py

or by

y

q=p

,

both of which are zero. Thus

F

1

(

xy

) =

x

+

y

as required.

60

background image

Next, observe that (

x

+

F

0

y

)

p

k

=

x

p

k

+

F

0

y

p

k

. It follows that (

x

+

F

1

y

)

p

k

;

F

1

(

x

p

k

+

F

1

y

p

k

) is divisible by

p

. Thus, there are unique series

k

(

xy

)

2

Z

p

xy

]]

such that

(

x

+

F

1

y

)

p

k

=

x

p

k

+

F

1

y

p

k

+

F

1

p

k

(

xy

)

:

Lemma 84

F

(

xy

) =

F

1

(

xy

)

;

P

n

;

1

k=1

u

k

k

(

xy

) mod (

u

0

::: u

n

;

1

)

2

Proof.

Write

J

= (

u

1

::: u

n

;

1

) and

I

= (

p

) +

J

. For the moment, we work

mod

J

2

. In this setting we have

log

F

(

x

) =

x

+

n

X

k=1

u

k

l

(

x

p

k

)

=p

=

l

(

x

) +

n

;

1

X

k=1

u

k

l

(

x

p

k

)

=p:

Next, observe that

x

+

F

y

;

F

(

x

+

F

1

y

) vanishes mod

J

, so there are unique

series

k

(

xy

)

2

Z

p

xy

]] such that

x

+

F

y

;

F

(

x

+

F

1

y

) =

n

;

1

X

k=1

u

k

k

(

xy

) (mod

J

2

)

:

By Lemma 83 we can rewrite the right hand side as a formal sum. Thus

x

+

F

y

= (

x

+

F

1

y

) +

F

u

1

1

(

xy

) +

F

+

F

u

n

;

1

n

;

1

(

xy

)

:

We now apply log

F

to this equation. The left hand side becomes

l

(

x

) +

l

(

y

) +

n

;

1

X

k=1

u

k

p

(

l

(

x

p

k

) +

l

(

y

p

k

))

:

The right hand side becomes

l

(

x

+

F

1

y

) +

X

u

k

p l

((

x

+

F

1

y

)

p

k

) +

X

u

k

k

(

xy

)

:

Recall that

l

(

x

+

F

1

y

) =

l

(

x

) +

l

(

y

) and

l

((

x

+

F

1

y

)

p

k

) =

l

(

x

p

k

) +

l

(

y

p

k

) +

l

(

p

k

(

xy

))

:

Using this to simplify the right hand side further, and rearranging, we conclude

that

X

u

k

k

(

xy

) +

X

k

u

k

p l

(

p

k

(

xy

)) = 0

:

61

background image

Finally, we recall that

l

(

px

)

=p

=

x

(mod

p

). This implies that

k

(

xy

) =

;

k

(

xy

) (mod

p

), which proves the lemma.

16 Examples

In this section we let

G

1

be the formal group over

Z

2

with logarithm log

F

(

x

) =

P

k

x

4

k

=

2

k

, so that

G

0

is a formal group of height 2 over

=

F

2

. Computer

calculations give results as follows.

(i) Let

A

'

Z

=

2 be generated by

a

. Given a level-

A

structure

, we shall

identify

a

with

x

(

(

a

)). We then have

Level(

A

G

0

) = spf(

a

]

=a

3

)

:

A coordinate on the quotient group

G

0

=

A

] is given by

y

=

ax

+

x

2

+

a

2

x

3

:

The associated formal group law is

y

+

F

0

z

=

y

+

z

+

ayz

+

y

2

z

2

+

a

(

y

2

z

3

+

y

3

z

2

) +

a

2

(

y

2

z

4

+

y

3

z

3

+

y

4

z

2

) +

(ii) Now take

A

'

(

Z

=

2)

2

, with generators

a

and

b

. Then

Level(

A

G

0

) = spf(

ab

]

=

(

a

3

b

3

a

2

+

ab

+

b

2

))

:

Because

A

= (1), the corresponding subgroup divisor is just

G

0

(1), and

a coordinate on the quotient group is just given by 2](

x

) =

x

4

.

(iii) Take

A

=

Z

=

4 generated by

a

. Then

Level(

A

G

0

) = spf(

a

]

=a

12

)

Type(

A

G

0

) = spf(

b

]

=b

6

)

b

=

a

2

+

a

5

+

a

8

+

a

11

:

The equation of the subgroup divisor is

Y

c

2

A

(

x

;

c

) =

x

4

+

b

5

x

3

+

bx

2

+

b

3

x:

(iv)

Sub

2

(

G

0

) = spf(

b

]

=b

7

)

:

The equation of the subgroup divisor is

x

4

+

b

5

x

3

+

bx

2

+ (

b

3

+

b

6

)

x:

The map of rings induced by Type(

A

G

0

)

;

!

Sub

2

(

G

0

) just sends

b

to

b

.

62

background image

Now let

G

1

be the standard height 3 formal group over

Z

2

, with logarithm

log

F

(

x

) =

P

k

x

8

k

=

2

k

. Let

A

'

Z

=

4

Z

=

2 be generated by

a

and

b

. Then

Level(

A

G

0

) = spf(

ab

]

=

(

a

56

b

6

+

b

5

a

8

+

b

4

a

16

+

+

a

48

))

Type(

A

G

0

) = spf(

cd

]

=

(

c

15

c

14

dc

13

d

2

+

c

14

d

3

+

c

8

d

+

c

12

))

:

where

c

=

a

4

+

a

18

+

a

32

+

a

10

b

+

a

17

b

+

a

38

b

+

a

2

b

2

+

a

9

b

2

+

a

16

b

2

+

a

37

b

2

+

b

4

+

a

14

b

4

+

a

21

b

4

+

a

35

b

4

+

a

34

b

5

+

a

41

b

5

d

=

a

8

b

+

b

2

+

a

32

b

5

:

The equation of the group divisor is

f

(

x

)=

x

8

+

(

dc

7

+

c

3

d

2

+

c

10

d

2

)

x

6

+

c

5

d

2

x

5

+

cx

4

+

(

dc

13

+

c

9

d

2

)

x

3

+

(

cd

+

c

5

)

x

2

+

(

dc

3

+

c

14

)

x:

Note that

f

(

x

) =

x

8

(mod

c

) it follows that

d

does not lie in the subring

generated by the coecients of

f

, which shows that the map Type(

A

G

)

;

!

Sub

8

(

G

) is not a closed embedding.

References

1] J. F. Adams. Stable Homotopy and Generalised Homology. University of

Chicago Press, Chicago, 1974.

2] W. Bruns and J. Herzog. Cohen-Macaulay Rings, volume 39 of Cambridge

Studies in Advanced Mathematics

. Cambridge University Press, 1993.

3] M. Demazure and P. Gabriel. Groupes Algebriques. North-Holland, 1970.
4] E. S. Devinatz, M. J. Hopkins, and J. H. Smith. Nilpotence and stable

homotopy theory I. Annals of Mathematics, 128:207{242, 1988.

63

background image

5] V. G. Drinfel'd. Elliptic modules. Math. USSR Sb., 23:561{592, 1974.
6] A. Frolich. Formal Groups, volume 74 of Lecture Notes in Mathematics.

Springer{Verlag, 1968.

7] M. Hazewinkel. Formal Groups and Applications. Academic Press, 1978.
8] M. J. Hopkins, N. J. Kuhn, and D. C. Ravenel. Generalised group characters

and complex oriented cohomology theories. To appear.

9] N. M. Katz and B. Mazur. Arithmetic Moduli of Elliptic Curves, volume 108

of Annals of Mathematics Studies. Princeton University Press, 1985.

10] J. Lubin and J. Tate. Formal moduli for one parameter formal lie groups. Bull.

Soc. Math. France

, 94:49{60, 1966.

11] H. Matsumura. Commutative Ring Theory, volume 8 of Cambridge Studies in

Advanced Mathematics

. Cambridge University Press, 1986.

12] J. Morava. Completions of complex cobordism. Lecture Notes in Mathematics,

658:349{361, 1978.

13] J. Morava. Noetherian localisations of categories of cobordism comodules.

Annals of Mathematics

, 121:1{39, 1985.

14] D. C. Ravenel. Complex Cobordism and Stable Homotopy Groups of Spheres.

Academic Press, 1986.

15] N. P. Strickland. Functorial philosophy for formal phenomena. in preparation,

1994.

16] J. Tate and F. Oort. Group schemes of prime order. Ann. scient. Ecoles Norm.

Sup.

, 3:1{21, 1970.

64


Wyszukiwarka

Podobne podstrony:
Serre Finite Subgroups of Lie Groups (1998)
Borovik Mirrors And Reflections The Geometry Of Finite Reflection Groups (2000) [sharethefiles com
Lusztig Some remarks on supercuspidal representations of p adic semisimple groups (1979) [sharethef
Elkies Combinatorial game Theory in Chess Endgames (1996) [sharethefiles com]
Turbiner Lie Algebraic Approach 2 the Theory of Polynomial Solutions (1992) [sharethefiles com]
Lasenby GA a Framework 4 Computing Invariants in Computer Vision (1996) [sharethefiles com]
Howe Counterexample to the generalized Ramanujan conjecture for quasi split groups (1979) [sharethe
Knutson Universal enveloping algebras, verma modules and degrees of a lie group (2002) [sharethefil
Olver Non Associative Local Lie Groups (2003) [sharethefiles com]
Marsden et al Poisson Structure & Invariant Manifolds on Lie Groups (2000) [sharethefiles com]
Burstall Harmonic Tori in Lie Groups (1991) [sharethefiles com]
Gull & Doran Multilinear Repres of Rotation Groups within GA (1997) [sharethefiles com]
[Martial arts] Physics of Karate Strikes [sharethefiles com]
Olver Lie Groups & Differential Equations (2001) [sharethefiles com]
Kollar The Topology of Real & Complex Algebraic Varietes [sharethefiles com]
Bradley Numerical Solutions of Differential Equations [sharethefiles com]
Pavsic Clifford Algebra of Spacetime & the Conformal Group (2002) [sharethefiles com]
The incorporation of carboxylate groups into temperature responsive poly(N isopropylacrylamide) base
Weiss Lie Groups & Quantum Mechanics [sharethefiles com]

więcej podobnych podstron