Skip to main content

Notes on class field theory

Section 6.1 Adèles

Reference.

[37]; [38], VI.1 and VI.2; [34], VII.

Subsection Lattices of number fields

The basic idea is that we want some sort of “global completion” of a number field K. Let us first recall an older version of this idea: Minkowski’s construction of the Euclidean lattice associated to a number field. We follow [38], I.5.

Definition 6.1.1.

Let K be a number field of degree n. It then has n distinct embeddings τ:KC. The product embedding
j:KτC,a(τ(a))τ
induces an isomorphism of KC=KQC with τC.
The ring KC=KQC admits an involution F which fixes K and acts on C via complex conjugation. The corresponding action on τC is
(zτ)τ(zτ)τ
where τ denotes the composition of τ with complex conjugation on C. The fixed subring under F is KR=KQR.
Equip KCτC with the standard Hermitian inner product, that is,
z1,z2=τz1,τz2,τ.
This restricts to a positive definite inner product on KR.
Via the embedding of K into KR, oK corresponds to a lattice in KR, i.e., a discrete cocompact subgroup. Similarly, any fractional ideal of K corresponds to a lattice in KR.

Subsection Profinite completions

Let us put aside the Minkowski construction for the moment and turn to some more arithmetic considerations. We have already used in multiple places the fact that the profinite completion Z^ of the group Z can be identified, via the Chinese remainder theorem, with the product pZp. This generalizes to an arbitrary number field as follows.

Remark 6.1.2.

Before continuing, we should clarify our use of notation like oK^ to denote the profinite completion of oK for K a number field. We originally defined this as an inverse limit over finite group quotients of oK. However, remember that we can define the same inverse limit using any smaller collection of quotients which is cofinal (that is, any finite quotient factors through some chosen quotient). In particular, if G is a subgroup of oK of some finite index n, then noKG and so the quotient map oKoK/G factors through the ring quotient oK/noK. That is, oK^ can be identified with the inverse limit limnoK/noK, and hence also carries the structure of a topological ring.

Proof.

As in Remark 6.1.2, we identify oK^ with limnoK/noK. This ring maps to oK/pm for each prime p and each positive integer m; putting these maps together gives us a map oK^limmoK/pm for each p, and hence a map to the product.
To see that this map is a bijection, factor the ideal noK as p1e1prer for some primes p1,,pr and some positive integers e1,,er. By the Chinese remainder theorem for ideals in a Dedekind domain, the natural map
oK/noKi=1roK/piei
is an isomorphism. This immediately implies that the original map is injective. To see that the original map is surjective, we must also observe that for each prime p and each positive integer m, there exists a positive integer n such that noK is divisible by pm; for instance, we may take n to be the absolute norm of pm.

Remark 6.1.4.

We cannot help mentioning a variant of Remark 6.1.2 that plays a key role in p-adic Hodge theory. Let Cp be a completed algebraic closure of Qp. Consider the inverse system
xxpoCpxxpoCp.
Since the maps are multiplicative but not additive, the inverse limit only appears to carry the structure of a multiplicative monoid. However, it was originally observed by Fontaine that the natural map from this inverse system to the inverse system
xxpoCp/poCpxxpoCp/poCp
is an isomorphism. In this inverse system, the maps upgrade to ring homomorphisms because (x+y)p=xp+yp in any ring in which p=0; consequently, the original inverse limit is upgraded to a ring! This then implies that the inverse limit of the system
xxpCpxxpCp
is again a ring; it is in fact an algebraically closed field which is complete with respect to a certain nonarchimedean absolute value. This construction has come to be known as forming the tilt of Cp, and generalizes to a large class of fields which are complete with respect to nonarchimedean absolute values (the perfectoid fields). See [3] for an introduction to this circle of ideas.

Subsection The adèles (rational case)

Our next step is to put the Minkowski construction together with profinite completion to define the ring of adèles. Let us do this first in the case of the rational numbers.

Definition 6.1.5.

We define the ring of finite adèles AQfin as any of the following isomorphic objects:
  • the tensor product Z^ZQ;
  • the direct limit of 1nZ^ over all nonzero integers n;
  • the restricted direct product pQp, where we only allow tuples (αp) for which αpZp for almost all p. See Definition 6.1.6.
This is a locally compact topological ring, with the groups 1nZ^ forming a fundamental system of neighborhoods of 0 consisting of compact subgroups. The natural group homomorphism
Q/ZAQfin/Z^
is an isomorphism.
In preparation for the definition of adèles associated to a general number field, we introduce the formalism of restricted products.

Definition 6.1.6.

Let I be an index set. For each iI, let Gi be a set and let Hi be a set of Gi. The restricted (direct) product G of the pairs (Gi,Hi) is the set of tuples (gi)i=1 such that giHi for all but finitely many indices i. Another way to say this is to define, for each finite subset SI, the set
GS=iSGi×iSHi
and take G=SGS.
We upgrade this construction from sets to richer categories as follows.
  • If each Gi is a group and each Hi is a subgroup, then G admits a group structure.
  • If each Gi is a ring and each Hi is a subring, then G admits a ring structure. (However, if each Gi is a field, then G cannot be a field unless I is a singleton set.)
  • If each Gi is a locally compact topological space and each Hi is a compact subspace, then G may be viewed as a locally compact topological space. One way to see this is to use a system of neighborhoods of the identity given by taking products of compact neighborhoods SiGi in which Si=Hi for all but finitely many i. (Remember that by Tikhonov’s theorem, any product of compact topological spaces is compact.) Another way is to equip each subset GS with the product topology, then declare a subset UG to be open if its intersection with each GS is an open subset of GS.
  • Likewise, if each Gi is a locally compact topological group/ring and each Hi is a compact subgroup/subring, then G may be viewed as a locally compact topological group/ring.

Definition 6.1.7.

Define the ring of adèles over Q as AQ=R×AQfin. Then AQ is a locally compact topological ring with a canonical embedding QAQ. We refer to the elements of Q as principal adèles within AQ.
We may also view AQ as a restricted direct product of the pairs
(R,{0}),(Q2,Z2),(Q3,Z3),;
note that taking the subgroup {0} of R has no real effect because the definition of the restricted product involves checking membership in the chosen subgroup for all but finitely many indices.

Remark 6.1.8.

Note that AQ contains the neighborhood U of 0 consisting of tuples (x)v where |x|<1 and |x|p1 for all primes p. Any element of the intersection UQ must be an integer (because of the condition at primes), but cannot be a nonzero integer (due to the condition at the real place); hence UQ={0}. That is, just as Z sits inside R as a discrete subgroup, Q sits inside AQ as a discrete subgroup.
In fact, we can do somewhat better. Just as the quotient group R/Z is covered by the compact subset [0,1] of R (and therefore is compact: a continous map from a compact topological space to Hausdorff topological space has compact image), the quotient group AQ/Q is covered by a compact subset
[0,1]×pZp

Remark 6.1.9.

In general, the restricted direct product of pairs (Gi,Hi) does not carry the subspace topology from the ordinary product iGi. We may see a concrete example using AQfin: for each prime p, let ep be the element consisting of 1p in the place v=p and 0 elsewhere. Then the sequence {ep} converges to zero in iGi but not in the restricted product: no cofinite subset of the sequence belongs to any neighborhood of 0.

Subsection The adèles (general case)

We now put the Minkowski construction together with profinite completion to define the ring of adèles of a number field.

Definition 6.1.10.

Let K be a number field. By Lemma 6.1.3, the profinite completion oK^ is canonically isomorphic to poKp. We may thus define the ring of finite adèles AKfin as any of the following isomorphic objects:
  • the tensor product oK^oKK;
  • the direct limit of 1αoK^ over all nonzero αoK;
  • the restricted direct product of the pairs (Kp,oKp) over all primes p of K.
The natural homomorphism
K/oKAKfin/oK^
is an isomorphism.
The ring of adèles AK is the product KR×AKfin. In other words, this is the restricted product of the pairs (Kv,{0}) for infinite places v and (Kv,oKv) for finite places v. We again have a diagonal embedding KAK; we again refer to the elements of the image of this embedding as principal adèles.

Definition 6.1.11.

For each place v of K, let ||v be the absolute value on the completion Kv normalized as follows.
  • For v real, take the usual real absolute value.
  • For v complex, take the square of the usual absolute value. (This does not satisfy the triangle inequality; sorry.)
  • For v a finite place above the prime p, normalize so that |p|v=p1.
We then have a well-defined function || on AK given by
|x|K=v|x|v;
this makes sense because by virtue of the definition of a restricted direct product, all but finitely many of the values |x|v are equal to 1.

Proof.

The normalizations have been chosen so that for each place v of Q, for each αK, the product of |α|w over all places w of K above p equals |NormL/K(α)|v. Taking the product over v, we deduce that |α|K=|NormL/K(α)|Q. That is, the product formula reduces to the case K=Q, which we may check directly: for α=±p1e1prer,
|α|v={p1e1prerif v=pieiif v=pi for some i0otherwise.

Subsection Adelic S-integers

Definition 6.1.14.

For any finite set S of places, let AK,S (resp. AK,Sfin) be the subring of AK (resp. AKfin) consisting of those adèles which are integral at all finite places not contained in S. The elements of AS might be thought of as “adelic S-integers”.
We can formulate an adelic analogue of the Chinese remainder theorem.

Proof.

We end up with an adelic analogue of the Minkowski embedding, but with the role of oK played by the entire field K!

Proof.

Choose a compact subset T of the Minkowski space M containing a fundamental domain for the lattice oK. Then every element of M×AKfin is congruent modulo oK to an element of T×AKfin. However, we can also interpret M×AKfin as AK,S where S is the set of infinite places of K. We may thus apply Proposition 6.1.15 to see that
M/oK×AKfin(M×AKfin)/oKAK,S/oK=AK,S/(KAK,S)=(K+AK,S)/KAK/K
is surjective. Consequently, the compact set T×AKfin surjects onto AK/K; the latter is Hausdorff by Corollary 6.1.13, so it being covered by a compact set makes it also compact.

Remark 6.1.17.

We mention in passing that just as the various completions of Q are “rigid” in the sense that they have no nontrivial automorphisms even if you ignore the topology (Exercise 3), the ring AQ also has no nontrivial automorphisms even if you ignore the topology (Exercise 6).

Subsection The approximation theorem

We already mentioned one analogue of the Chinese remainder theorem (Proposition 6.1.15). Here is another one.

Proof.

Exercises Exercises

1.

Prove that the map from [0,1]×pZp to AQ/Q is surjective.

3.

Let K be a number field and let v be a place of K which is not complex. Prove that every automorphism of the field Kv (as a ring without topology) is continuous. (See Remark 6.3.3 for discussion of how this breaks down when v is complex.)
Hint.
For a real place, note that any automorphism of R preserves the perfect squares and hence the order relation. For a finite place, let q be the cardinality of the residue field of v. Show first that an element of Kv belongs to oKv if and only if it has an m-th root for every positive integer m coprime to p(q1). Then note that an element of Kv belongs to oKv iff it is a difference of two elements of oKv.

4.

Let S be a finite set of places of a number field K, none of which is complex. Prove that every automorphism of vSKv (as a ring without topology) is continuous.
Hint.
Using Exercise 3, reduce to checking that for two noncomplex places v and w of K, lying over distinct places of Q, the completions Kv and Kw are not isomorphic as underlying rings. To prove this, consider the set of xK which are squares in Kv, and similarly for w.

5.

Let K be a number field and let v be a place of K which is not complex. Let Q(x,y,z) be a quadratic form over K defined as follows.
  • If v is real, put Q(x,y,z):=x2+y2+z2.
  • If v is finite lying over the rational prime p, choose aKoKv whose image in the residue field of v is not a quadratic residue, and put Q(x,y,z):=x2ay2+pz2.
Let T be the intersection of the images of the maps cQ:AK3AK over all cK. Prove that T=ker(AKwSKw) for some finite set S of places of K containing v.
Hint.
Use Hensel’s lemma to show that for w a finite place not lying above 2, a,b,coKw, and tKw, the equation ax2+by2+cz2=t always has a solution with a,b,cKw.

6.

Prove that every automorphism of the ring AQ, not necessarily continuous, is trivial.
Hint.
Use Exercise 4 and Exercise 5 to prove that the map AQvQv is equivariant for any automorphism of AQ and the trivial action on vQv. Luckily Q has no complex places!

7.

Hint.
Prove by induction on n that given any pairwise distinct places v1,,vn, we can find xK with
|x|v1>1,|x|v2<1,,|x|vn<1.
Then make a careful linear combination of powers of such elements. For more details, see [38], Theorem II.3.4.