Romyar SharifiLECTURE NOTES
READING EDITIONPDF

LECTURE NOTES / Chapter 10

Algebraic Number Theory

Romyar Sharifi

Chapter 10 Global class field theory via idèles

Book contents

Chapter 10
Global class field theory via idèles

10.1. Restricted topological products

We begin by defining the notion of a restricted topological product.

Definition 10.1.1.

Let I be an indexing set. For each i ∈ I, let Xi be a topological space and Ai be an open subset. The restricted topological product of the spaces Xi relative to the open subsets Ai is the set

∏i∈I(Xi,Ai) = {(xi)i∈I ∈∏i∈IXi∣xi ∈ Ai for all but finitely many i ∈ I},

endowed with the topology which has as a (standard) basis the open sets of the form

{(xi)i∈I∣xi ∈ Ui for i ∈ J and xi ∈ Ai for i ∈ I −J},

where J ⊆ I is finite and Uj is an open subset of Xj for each j ∈ J. This topology on the set ∏ ⁡i∈I(Xi,Ai) is referred to as the restricted product topology.

Remark 10.1.2.

The standard basic open sets of a restricted topological product ∏ ⁡i∈I(Xi,Ai) have the form

∏j∈JUj×∏i∈I−JAi

with J ⊂ I finite and Uj open in Xj for each j ∈ J. The subspace topology on these open sets is exactly the product topology for the subspace topology from the Xi on the sets that form its product.

Remark 10.1.3.

The restricted topological product ∏ ⁡i∈I(Xi,Ai) does not in general have the subspace topology of the product topology on ∏ ⁡i∈IXi. That is, a standard basic open neighborhood of ∏ ⁡i∈IXi has the form

∏k∈KVk×∏i∈I−KXi

with K ⊂ I finite and Vk open in Xk for each k ∈ K. Any such set contains a basic open neighborhood in the restricted product topology on ∏ ⁡i∈I(Xi,Ai). On the other hand, no such set will have its intersection with ∏ ⁡i∈I(Xi,Ai) contained in any set of the form

∏j∈JUj×∏i∈I−JAi

with J ⊂ I finite and Uj open in Xj for each j ∈ J so long as there are infinitely many i ∈ I for which Ai≠Xi. In other words, the restricted product topology on ∏ ⁡i∈I(Xi,Ai) is finer and can be strictly finer than the subspace topology for the product topology.

The following simple lemma is quite useful.

Lemma 10.1.4.

Let I be an indexing set, and for each i ∈ I, let Xi and Yi be topological spaces, let Ai be an open subset of Xi, let Bi be an open subset of Yi, and let fi: Xi → Yi be a continuous function. Suppose that fi(Ai) ⊆ Bi for all but finitely many i ∈ I. Then the product of the maps fi restricts to a continuous function

∏i∈I(Xi,Ai) →∏i∈I(Yi,Bi).
Proof.

For x = (xi)i∈I ∈ X, we have that xi ∈ Ai for almost all i, so fi(xi) ∈ Ai for almost all i, and therefore ∏ ⁡ifi(x) ∈∏i∈I(Yi,Bi). Thus, it makes sense to let f denote the restriction of ∏ ⁡ifi to a map between the restricted topological products.

Consider a finite subset J of I and open subsets Vj of Yj for each j ∈ J, and suppose f(x) ∈∏ ⁡j∈JVj×∏i∈I−JYi. For each j ∈ J, there exists an open neighborhood Uj of xj in X such that fj(Uj) ⊆ Vj, as fj is continuous. Consequently, we have that f(∏ ⁡j∈JUj×∏i∈I−JXi) is contained in ∏ ⁡j∈JVj×∏i∈I−JYi. Thus, f is continuous. □

Lemma 10.1.5.

Let I be an indexing set. For each i ∈ I, let Xi be a Hausdorff topological space, and let Ai be an open subset of Xi. Then the restricted topological product ∏ ⁡i∈I(Xi,Ai) is Hausdorff.

Proof.

This is a consequence of either the discussion of Remark 10.1.3 or Lemma 10.1.5 applied to the case that Yi = Bi = Xi. The product ∏ ⁡i∈IXi is Hausdorff, and if distinct x,y ∈∏ ⁡i∈I(Xi,Ai) are contained in disjoint open neighborhoods in ∏ ⁡i∈IXi, then the intersection of these neighborhoods with ∏ ⁡i∈I(Xi,Ai) are disjoint open neighborhoods of x and y in the latter space. □

We are often interested in the case that our sets are topological groups or rings. In the proof, we treat only the case of groups, the case of rings being analogous.

Lemma 10.1.6.

Let I be an indexing set, and for each i ∈ I, let Gi be a locally compact, Hausdorff topological group (resp., ring), and let Ki an open subgroup (subring) of Gi that is compact for almost all i. Then the restricted topological product ∏ ⁡i∈I(Gi,Ki) is a locally compact, Hausdorff topological group (resp., ring).

Proof.

That 𝒢 = ∏ ⁡i∈I(Gi,Ki) is a group is straightforward. That is, clearly 1 = (1)i ∈ G, and if a = (ai)i and b = (bi)i are elements of 𝒢, then ai,bi ∈ Ki for all but finitely many i ∈ I, so 𝑎𝑏 ∈𝒢. Similarly, a−1 ∈𝒢 since ai−1 ∈ Ki if ai ∈ Ki. It is then a topological group by applying Lemma 10.1.4 to the multiplication and inverse maps on 𝒢.

That 𝒢 is Hausdorff is Lemma 10.1.5. Let J be a finite subset of I such that Ki is compact for i ∈ I −J, and let Uj be a compact neighborhood of 1 in Gj for j ∈ J. Then ∏ ⁡j∈JUj×∏i∈I−JKi is an compact neighborhood of 1 in 𝒢 by Tychonoff’s theorem, so 𝒢 is locally compact. □

10.2. Adeles

We now define the ring of adeles of a global field K.

Definition 10.2.1.

Let K be a global field. The ring of adeles (or adele ring) 𝔸K of K is the restricted topological product

𝔸K = ∏v∈VK(Kv,𝒪v),

where VK is the set of all places of K, where Kv is the completion of K at the place v, and where 𝒪v is the valuation ring of Kv, which we take to be Kv if v is archimedean. An element of 𝔸K is referred to as an adele.

Remark 10.2.2.

Let α be an adele in a global field K. Then αv for some v ∈ VK shall denote the v-coordinate of α.

Lemma 10.2.3.

Let a ∈ K. Then a ∈𝒪v for all but finitely many places v of K.

Proof.

It suffices to check this on the cofinite subset of finite places of K in VK. Only those finitely many finite primes 𝔭 occur in the factorization of a𝒪K, so only finitely many have negative valuation v𝔭(a). The result follows. □

As an immediate consequence of Lemma 10.2.3, we see that every element of a global field gives rise to an element of its adele ring (since units are in particular integers).

Definition 10.2.4.

The diagonal embedding δK: K →𝔸K is the homomorphism δK(a) = (a)v∈VK.

We note also that we have embeddings of adele rings into adele rings of extension fields.

Definition 10.2.5.

For L∕K finite, the canonical embedding of 𝔸K in 𝔸L is the map ιL∕K: 𝔸K →𝔸L given by ιL∕K(α)w = αv for every place w of L and the place v of w lying below it.

We will show that the diagonal embedding has discrete image. This is rather straightforward for K = ℚ, for example, so we proceed by reduction to this case, using the following result. Note that the diagonal embedding provides 𝔸K with the structure of a K-vector space.

Proposition 10.2.6.

Let L∕K be a finite, separable extension of global fields. Then there is a canonical isomorphism of topological L-algebras

κ : L⊗K𝔸K →∼𝔸L

given on simple tensors of b ∈ L and α ∈𝔸K by

κ(b⊗α) = δL(b)ιL∕K(α) = (bαv)w,

where v is used to denote the place of K lying below a place w of L. Here, a choice of K-basis of L provides an isomorphism 𝔸K⊗K𝐿≅𝔸K[L:K] of 𝔸K-modules, and we give 𝔸K⊗KL the topology induced via this isomorphism from the product topology on 𝔸K[L:K].

Proof.

Recall that Proposition 6.1.6 says that there is, for each place v of K, an isomorphism

κv: L⊗KKv →∼∏w∣vLw,

and the map κ as defined is the restriction of the product of these to L⊗K𝔸K, which has image in 𝔸L since both δL and ιL∕K do.

Since the product of the maps κv is an isomorphism, the map κ is an injection that we claim is surjective. For this, let b1,…,bn be a K-basis for L. For all but finitely many finite primes v of K, we have for all places w of L over v both that bi ∈𝒪w for 1 ≤ i ≤ n and that w(D(b1,…,bn)) = 0. For any such v, the map κv restricts to a map

κv′: ⨁ i=1n(1⊗𝒪 v)(bi⊗1) →∏w∣v𝒪w

of free 𝒪v-modules of rank n, which by Proposition 6.1.10 is an isomorphism.

Given this, choose a finite set S of places of K containing the infinite places and those finite places v with bi∉𝒪w for some i or w(D(b1,…,bn))≠0 for some w dividing v, and let T be the finite set of places of L above it. Since κv is surjective for all v ∈ S and κv′ is surjective for all v∉S, the image of κ contains

∏w∈T Lw×∏w∈VL−T 𝒪w.

Since any arbitrary finite set of places of L is contained in some such set T , it follows that κ is surjective.

For continuity of κ, note that each fi: 𝔸K →𝔸L defined by fi(α) = δL(bi)ιL∕K(α) is continuous, noting that ιL∕K is continuous by Lemma 10.1.4. Then κ viewed as a map

𝔸K[L:K] →𝔸 L

using this basis is continuous as a sum of the continuous maps fi, since 𝔸L is a topological group under addition. □

We next show that K sits discretely in its adele ring.

Proposition 10.2.7.

The diagonal embedding δK: K →𝔸K has discrete image.

Proof.

If K is a number field, Proposition 10.2.6 for the extension K∕ℚ identifies κ−1 ∘δK for the map κ therein with the map

id ⁡ K⊗δℚ: K ⊗ℚℚ → K ⊗ℚ𝔸ℚ.

Therefore, it suffices to show that δℚ has discrete image. (Similarly, for function fields of characteristic p, it suffices to consider ι𝔽p(t), for which we omit the proof.)

We identify ℚ with its image in 𝔸ℚ under δℚ and endow it with the subspace topology. The intersection of the open neighborhood

∏p primeℤp×{a ∈ℝ∣|a| < 1}

of 0 in 𝔸ℚ with ℚ consists of elements with nonnegative valuation at p for every p, which to say integers, that also have absolute value less than 1. In other words, the intersection is {0}. Since 𝔸ℚ is a topological ring and ℚ is a subring, we have by translation that ℚ has discrete image. □

Notation 10.2.8.

We use the diagonal embedding δK to identify a global field K with a subring of 𝔸K, denoted also by K.

We also have the following.

Proposition 10.2.9.

Let K be a global field. Then there exists a finite set S of places of K including the archimedean places and positive real numbers 𝜖v for each v ∈ S such that

X = ∏v∈VK−S𝒪v×∏v∈S{a ∈ Kv∣|a|≤ 𝜖v}. (10.2.1)

satisfies 𝔸K = K +X.

Proof.

First consider the case K = ℚ. In this case, we claim that the set

Y = ∏p primeℤp×{a ∈ℝ∣|a|≤12}

satisfies 𝔸ℚ = ℚ+Y. Let α ∈𝔸ℚ. Let p1,…,pn be the finite list of prime numbers p such that αp∉ℤp. Let ki = −vpi(αpi) for each i with 1 ≤ i ≤ n. Let b = p1k1⋯pnkn. Let a ∈ℤ be the unique integer such that

−b2 < a−bα∞≤b2

and

a ≡ bαpi modpikiℤ pi

for each i. Then α −ab ∈ Y, as desired.

For an arbitrary number field, choose a basis b1,…,bn of K as a ℚ-vector space, and note that Proposition 10.2.6 tells us that every element β ∈𝔸K has the form

β = ∑i=1nb iιK∕ℚ(αi)

for some α1,…,αn ∈𝔸ℚ. In turn, each αi may be written as αi = yi+ci with yi ∈ Y and ci ∈ℚ. We then have

β ∈ K +∑i=1nb iιK∕ℚ(Y ),

and we let X′ denote the latter sum. We let S be the set consisting of the archimedean places of K and the nonarchimedean places of K such that bi∉𝒪v for some i, and we note that every element α ∈ X′ satisfies αv ∈𝒪v for all v∉S. For each nonarchimedean (resp., archimedean) v ∈ S, let 𝜖v be such that |bi|v ≤ 𝜖v (resp., |bi|v ≤ 2𝜖v) for all 1 ≤ i ≤ n. Then X as defined by (10.2.1) contains X′ and satisfies 𝔸K = K +X.

The case of a function field is similarly derived from the case K = 𝔽p(t), and the proof is left to the reader. □

Corollary 10.2.10.

Let K be a global field. Then the quotient K-vector space 𝔸K∕K is compact Hausdorff.

Proof.

The compact (Hausdorff) set X of Proposition 10.2.9 maps onto the quotient 𝔸K∕K under the continuous projection map from 𝔸K. The space 𝔸K∕K is Hausdorff as K is closed in 𝔸K, so 𝔸K∕K is compact as well. □

Since an adele necessarily has valuation less than or equal to 1 in all but finitely many places, the infinite product in the following definition converges.

Definition 10.2.11.

Let K be a global field. Let α ∈𝔸K. The content cK(α) of α is defined to be

cK(α) = ∏v∈VK∥αv∥v,

where we recall that ∥αv∥v = |αv|v unless v is complex, in which case ∥αv∥v = |αv|v2.

Lemma 10.2.12.

Let K be a global field. Then there exists a positive real number C such that for every α ∈𝔸K with cK(α) > C, there exists an element a ∈ K× with |a|v ≤|αv|v for all places v of K.

Sketch of proof.

Since 𝔸K is a locally compact abelian group, it has an invariant Haar measure. Letting S denote the set of archimedean places of K, we set

Z = ∏v∈VK−S𝒪v×∏v∈SB¯1∕2(0)

where B1∕2(0) denotes the closed ball of radius 1∕2 around 0 under the usual absolute value corresponding to v. We normalize our Haar measure so that Z has volume 1. As 𝔸K∕K is compact, it has finite quotient measure, and we take C to be this measure.

Now let α be as in the statement. The set 𝛼𝑍 has measure cK(α) > C (which we leave to the reader to verify, using uniqueness of Haar measure and noting that local Haar measures will scale by the multiplicative valuation used in defining the content), so it follows that there exist two distinct elements β, β′ of 𝛼𝑍 with the same image in 𝔸K∕K, which is to say that the difference a = β −β′ lies in K. For each v ∈ VK, we clearly have |β −β′|v ≤|α|v by choice of Z, so the result holds. □

The lemma has the following corollary.

Corollary 10.2.13.

Let K be a global field and u be a place of K. Let S be a finite set of places of K not including u, and choose a real number 𝜖v > 0 for each v ∈ S. Then there exists an element a ∈ K× such that |a|v ≤ 𝜖v for all v ∈ S and |a|v ≤ 1 for all v∉S with v≠u.

Proof.

Let C be as in Lemma 10.2.12. Choose elements αv ∈ Kv for each v with |αv|v ≤ 𝜖v for each v ∈ S and |αv|v = 1 for all other places v≠u of K. Let αu ∈ Ku be such that

∥αu∥u > C∏v∈S∥αv∥v−1.

Setting α = (αv)v ∈𝔸K, we then have cK(α) > C, so there exists an element a ∈ K× with |a|v ≤|αv|v for all v and therefore |a|v ≤ 𝜖v for all v ∈ S and |av|v ≤ 1 for v∉S∪{u}. □

While any global field K sits discretely in 𝔸K, if we exclude one prime from 𝔸K, the result is very different.

Theorem 10.2.14 (Strong Approximation).

Let K be a global field, and let u be a place of K. Set

𝔸K≠u = ∏ v∈VK−{u}(Kv,𝒪v).

Then K is embedded diagonally as a dense subset of 𝔸K≠u.

Proof.

Let α ∈𝔸K≠u, let δ > 0, and let T be a finite subset of Vk−{u} that includes its archimedean places and places with αv∉𝒪v. We claim that there exists a ∈ K such that |a−αv|v ≤ δ for all v ∈ T and |a|v ≤ 1 for all v∉T , which will prove the result.

By Proposition 10.2.9, we have a set X of the form in (10.2.1) such that 𝔸K = K +X. We use the notation S and 𝜖v for v ∈ S found therein. Setting 𝜖v = 1 for v∉S, there exists by Corollary 10.2.13 an element b ∈ K× with

|b|v ≤{𝜖v−1δif v ∈ T, 𝜖v−1 if v∉T ∪{u}.

Note that

𝔸K = b𝔸K = 𝑏𝑋 +K.

Viewing 𝔸K≠u as the subset of 𝔸K of elements with u-coordinate 0, we may then write α = 𝑏𝑥+a for some x ∈ X and a ∈ K. Since |bxv|v ≤ δ for all v ∈ T and |bxv|v ≤ 1 for all v∉T ∪{u}, the element a = αv−bxv for any v≠u has the desired properties. □

Let us record the rephrasing of the strong approximation theorem found in its proof. The statement is more clearly a direct generalization of weak approximation.

Corollary 10.2.15.

Let K be a global field, and let u be a place of K. Let αv ∈ Kv for each place v≠u of K and suppose that αv ∈𝒪v for all but finitely many such v. Then for every 𝜖 > 0 and finite set of primes S of K with u∉S, there exists a ∈ K such that |a−αv|v < 𝜖 for all v ∈ S and |a|v ≤ 1 for all finite places v∉S with v≠u.

We next investigate how adele rings behave in extensions. For a finite extension of L of K, we identify 𝔸K with a closed subgroup of 𝔸L via the canonical embedding.

Lemma 10.2.16.

Let L∕K be a finite Galois extension of global fields. Then 𝔸LGal ⁡ (L∕K) = 𝔸K.

Proof.

The Galois group G = Gal ⁡ (L∕K) permutes the places of L lying over a place v of K. Let

(αw)w ∈∏w∈VL w∣v Lw

be G-invariant. Since the decomposition group Gw at w∣v preserves the w-coordinate, it fixes αw. Thus, αw ∈ Kv, and this holds for all w∣v. Moreover, G acts on ∏ ⁡w∣vKv by permuting the coordinates, and the action of G on the set of places is transitive, so all of the αw must be equal. That is, (αw)w is in the image of some a ∈ Kv in the product ∏ ⁡w∣vLw. If, moreover, (αw)w ∈∏ ⁡w∣v𝒪w, then clearly a ∈𝒪v.

With respect to inclusion maps, we have

𝔸K = lim →S⊂VK (∏v∈SKv×∏v∈VK−S𝒪v),

where S runs over the finite sets of places of K. For such a set S, let SL be the subset of VL of places lying over places in S. Every finite set of places of L is contained in some SL, so

𝔸LG = lim → S⊂VK (∏w∈SLLw×∏w∈VL−SL𝒪w)G = lim → S⊂VK(∏v∈S(∏w∈VL w∣v Lw)G×∏ v∈VK−S(∏w∈VL w∣v 𝒪w)G) = 𝔸 K.

□

Let us consider norm and trace maps on adeles.

Definition 10.2.17.

Let L∕K be an extension of global fields.

a.

The norm map NL∕K: 𝔸L →𝔸K is the multiplicative function defined by

NL∕K(β) = (∏w∣vNL∕K(βw))v

on β ∈𝔸L.

b.

The trace map Tr ⁡ L∕K: 𝔸L →𝔸K is the homomorphism

Tr ⁡ L∕K(β) = (∑w∣vTr ⁡ L∕K(βw))v.

on β ∈𝔸L.

Remarks 10.2.18.

Let L∕K be a finite extension of global fields.

a.

That the norm and trace for L∕K on adeles have images inside the adeles follows from the fact that β ∈𝔸L has βw in the valuation ring of Lw for all but finitely many w, and hence for all w dividing v for all but finitely many places v of K, and therefore NL∕K(βw) and Tr ⁡ L∕K(βw) lie in the valuation ring of Kv for all w dividing v for all but finitely many v.

b.

In the notation of Definition 6.1.8, the norm and trace maps on adeles have v-coordinates on β ∈𝔸L given by

NL∕K(β)v = NL∕Kv((β w)w) and Tr ⁡ L∕K(β)v = Tr ⁡ L∕Kv((β w)w),

where w runs over the places dividing v.

c.

It follows form Lemma 10.1.4 that NL∕K and Tr ⁡ L∕K are continuous maps on adeles.

10.3. Idèles

In this section, we define the idèles, the elements of which are the units in the adeles. We continue to let K denote a global field.

Definition 10.3.1.

Let K be a global field. The group of idèles (or idèle group) 𝕀K of K is the restricted topological product

𝕀K = ∏v∈VK(Kv×,𝒪 v×),

where VK is the set of all places of K, where Kv is the completion of K at the place v, and where 𝒪v is the valuation ring of Kv, for which we set 𝒪v× = Kv× if v is archimedean. An element of 𝕀K is referred to as an idèle.

Remark 10.3.2.

Note that 𝕀K = 𝔸K× as sets, but 𝕀K does not have the subspace topology from 𝔸K. For each finite prime v, fix a uniformizer πv in Kv, and let αv be the adele that is πv in its v-coordinate and 1 in every other coordinate. Then every open neighborhood of 1 in 𝔸K contains all but finitely many αv. On the other hand, the basic open neighborhood

∏v∈VK𝒪v×

of 1 in 𝕀K contains not a single αv. On the other hand, the intersection of a basic open neighborhood of 𝔸K with 𝕀K is an open neighborhood of 𝕀K, so the topology on 𝕀K is strictly finer than the subspace topology from 𝔸K.

We leave it to the reader to check the following.

Lemma 10.3.3.

The restricted product topology on the idèle group 𝕀K of a global field agrees with the subspace topology induced by the injection

𝕀K →𝔸K×𝔸K,α↦(α,α−1).

Remark 10.3.4.

Note that the content cK(α) of an idèle α is a positive real number, as all but finitely many coordinates of α will have multiplicative valuation 1 and the valuations of the other coordinates will be nonzero. In fact, the property that of having nonzero content characterizes the idèles as a subset of the adeles, as the reader may quickly check.

We may then make the following definition.

Definition 10.3.5.

Let K be a global field. The content homomorphism cK: 𝕀K →ℝ>0 is the function that takes an idèle α to its content

cK(α) = ∏v∈VK∥αv∥v.

Let 𝕀K1 denote the kernel of cK.

Proposition 10.3.6.

Let K be a global field. Then the content homomorphism cK: 𝕀K →ℝ>0 is continuous.

Proof.

We leave it to the reader to verify the following simple claim, which implies the statement. For any 𝜖 > 0, there exists a sufficiently small δ > 0 such that cK−1((1−𝜖,1+𝜖)) contains ∏ ⁡v∈VK−S𝒪v××∏v∈SBδ(1), where S is the set of archimedean places of K and Bδ(1) ⊂ Kv× is a ball of radius δ about 1. □

Corollary 10.3.7.

The group 𝕀K1 of idèles of content 1 is a closed subgroup of 𝕀K

We have the following result on the kernel of the content homomorphism.

Lemma 10.3.8.

The topology on 𝕀K1 from 𝕀K agrees with its subspace topology from 𝔸K.

Proof.

By Remark 10.3.2, the subspace topology on 𝕀K1 from 𝕀K is finer than the subspace topology from 𝔸K, so we need only show that the intersection of a basic open neighborhood of 1 in 𝕀K with 𝕀K1 contains the intersection of an open neighborhood of 1 in 𝔸K with 𝕀K1.

Let S be a finite set of places of K containing the archimedean places, and for each v ∈ S, let Uv be an open subset of Kv× containing 1. Then

U = ∏v∈SUv×∏v∈VK−S𝒪v×

is a basic open in 𝕀K containing 1. We may suppose that the sets Uv are chosen to be balls of sufficiently small radius such that the products of the (modified) valuations of any elements in ∏ ⁡v∈SUv is less than 2. Since every element 𝒪v−𝒪v× has valuation at most 12, we then have

U ∩𝕀K1 = (∏ v∈SUv×∏v∈VK−S𝒪v)∩𝕀K1,

and ∏ ⁡v∈SUv×∏v∈VK−S𝒪v is a basic open neighborhood of 1 in 𝔸K. □

Note that we may think of K× as a subgroup of 𝕀K via the diagonal embedding. By the product formula for valuations on global fields, every element of K× lies in 𝕀K1.

Proposition 10.3.9.

The image of K× in 𝕀K1 is discrete, and 𝕀K1∕K× is compact Hausdorff.

Proof.

The first statement follows from Corollary 10.2.10, since Lemma 10.3.8 tells us that 𝕀K1 has the subspace topology from 𝔸K. For the second statement, let α ∈𝔸K be an adele with cK(α) > C, where C is as in the statement of Lemma 10.2.12. We define a compact subset of 𝔸K by

X = {β ∈𝕀K1∣|β v|v ≤|αv|v for all v ∈ VK},

where here we use the fact that 𝕀K1 is closed in 𝔸K. For an arbitrary γ ∈𝕀K1, Lemma 10.2.12 tells us that there exists a ∈ K× with |a|v ≤|γv−1αv|v for all places v of K. We then have 𝛾𝑎 ∈ X, so X surjects onto the Hausdorff space 𝕀K1∕K×, and therefore the latter quotient is compact. □

Definition 10.3.10.

The principal idèles of a global field K are the elements of 𝕀K that lie in K× (under its diagonal embedding).

Definition 10.3.11.

Let K be a global field. Then the idèle class group ℂK of K is the quotient topological group 𝕀K∕K×. The image [α] of α ∈𝕀K in ℂK is the idèle class of α.

Notation 10.3.12.

For a global field K, we shall use VK,f to denote its set of finite places.

Definition 10.3.13.

Let K be a global field. The fractional ideal of 𝒪K defined by an idèle α of K is the finite product

∏v∈VK,f𝔭vv(α),

where 𝔭v denotes the prime corresponding to a finite place v of K.

Proposition 10.3.14.

Let K be a number field. Let πK: 𝕀K → IK be the homomorphism that takes an idèle to the fractional ideal it defines. Then πK(𝕀K1) = IK, and πK is continuous if we endow IK with the discrete topology.

Proof.

For the first statement, we need only show that every nonzero prime 𝔭 is the image of an element of 𝕀K1. We may take an idèle α of content 1 that is a uniformizer π𝔭 in the coordinate corresponding to 𝔭, that in a fixed archimedean place w satisfies ∥αw∥w = |π𝔭|𝔭−1, and which is 1 in all other coordinates.

For the second statement, we need only note that

πK−1({(1)}) = ∏ v∈VK𝒪v×

is open in 𝕀K. □

Remark 10.3.15.

Proposition 10.3.14 enables us to give a second proof that Cl ⁡ K is finite. That is, since πK in the proposition takes K× onto PK, we have an induced continuous, surjective map 𝕀K1∕K×→ Cl ⁡ K. It follows that Cl ⁡ K is both compact as the continuous image of a compact space and discrete as a quotient of IK, and therefore Cl ⁡ K is finite.

Notation 10.3.16.

For a finite extension L∕K of global fields, we use ιL∕K: ℂK →ℂL also to denote the map induced by the canonical embedding ιL∕K: 𝕀K →𝕀L.

Lemma 10.3.17.

For a finite extension L∕K of global fields, the map ιL∕K: ℂK →ℂL is injective.

Proof.

Identifying 𝕀K and L× with their images in 𝕀L, we need only see that 𝕀K∩L× = K×. Let M be a finite Galois extension of K containing L. We claim that 𝕀K∩M× = K×, which will prove the result. But Lemma 10.2.16 gives us the first equality in

𝕀K∩M× = 𝕀 MGal ⁡ (M∕K) ∩M× = (M×)Gal ⁡ (M∕K) = K×,

the second equality following from the compatibility of the Galois action on M× and 𝕀M under the diagonal embedding. □

Notation 10.3.18.

We identify ℂK with a (closed) subgroup of ℂL via the embedding ιL∕K, noting Lemma 10.3.17.

As a consequence of Lemma 10.2.16, we have that 𝕀LGal ⁡ (L∕K) = 𝕀K for any finite Galois extension L∕K. We claim that the same holds for idèle class groups.

Lemma 10.3.19.

Let L∕K be a finite Galois extension of global fields. Then ℂLGal ⁡ (L∕K) = ℂK.

Proof.

We have an exact sequence of modules for G = Gal ⁡ (L∕K) given by

0 → L×→𝕀 L →ℂL → 0,

and this gives rise to a long exact sequence starting

0 → K×→𝕀 K →ℂLG → H1(G,L×).

Since the latter group is zero by Hilbert’s Theorem 90, the resulting short exact sequence yields the result. □

Since the idèles are the units in the adele ring, the norm map on adele ring is immediately seen to define a norm map on the idèle group. Continuity of the norm follows from Lemma 10.1.4, as with adeles.

Definition 10.3.20.

Let L∕K be an extension of global fields. The norm map NL∕K: 𝕀L →𝕀K is the homomorphism that is the restriction of NL∕K: 𝔸L →𝔸K.

Since the norms of principal idèles are principal, we may make the following definition.

Definition 10.3.21.

The norm map NL∕K: ℂL →ℂK is the map induced on quotient groups by the corresponding norm map on idèle groups.

Remark 10.3.22.

The norm map is continuous on idèle groups and idèle class groups.

10.4. Statements

The reciprocity map in the idèle-theoretic approach to global class field theory is constructed out of the local reciprocity maps of the completions of the global fields in question. We provide the preliminary results to its construction.

Lemma 10.4.1.

Let K be a global field, let L be a finite abelian extension of K, and let α ∈𝕀K. Then ρLw∕Kv(αv) = 1 for all places w lying over v for all but finitely many places v of K.

Proof.

All but finitely many v are unramified in L∕K and for all but finitely many v, the valuation v is nonarchimedean and αv is a unit in its valuation ring 𝒪v. Since ρLw∕Kv(𝒪v×) is contained in the inertia subgroup of Gal ⁡ (Lw∕Kv) for any v and this inertia subgroup is trivial in an unramified extension, we have the result. □

Remark 10.4.2.

Suppose we start with a global field K and a place v of K. Consider the canonical map from GKvab to a decomposition group Dv in GKab at a place w over v. (Note that, while the map GKv↪GK identifying GKv with the decomposition group at a place of Ksep is injective, the map GKvab → GKab it induces may not be.) The resulting map GKvab → GKab is independent of the choice of w since conjugation by an element of GKab is a trivial automorphism of GKab. We may then view the local reciprocity map as producing global elements via the composition

Kv×→ρ KvGKvab →∼D v↪GKab

that takes α ∈ Kv× to ρKv(α)|Kab, and since all of the maps in the composition are independent of w, this map is as well.

The following lemma is now an immediate consequence of Remark 10.4.2.

Lemma 10.4.3.

Let L∕K be a finite abelian extension of global fields, let v be a place of K, and let α ∈ Kv×. The quantity

ρLw∕Kv(α)|L ∈ Gal ⁡ (L∕K)

for a place w of L lying over v is independent of w.

Lemmas 10.4.1 and 10.4.3 allow us to define the reciprocity map for a finite abelian extension of global fields.

Definition 10.4.4.

Let L∕K be a finite abelian extension of global fields. The (global) reciprocity map for L∕K is the homomorphism ΦL∕K: 𝕀K → Gal ⁡ (L∕K) defined by

ΦL∕K(α) = ∏v∈VKρLw∕Kv(αv)|L,

where for each valuation v of K, we have chosen a valuation w of L lying over v.

We note the following compatibility among the global reciprocity maps.

Lemma 10.4.5.

Let K be a global field, and let L and M be finite abelian extensions of K with L ⊆ M. For every α ∈𝕀K, we have ΦL∕K(α) = ΦM∕K(α)|L.

Proof.

For each v ∈ VK, we choose w ∈ VL lying over v and u ∈ PM lying over w. By property (ii) of local reciprocity, we have

ρLw∕Kv(αv) = ρMu∕Kv(αv)|Lw,

and the result is then an immediate consequence of the definition of the global reciprocity map. □

Corollary 10.4.6.

Let K be a global field. Then for each α ∈𝕀K, the quantity

lim ←LΦL∕K(α),

with the inverse limit taken over finite abelian extensions L of K with respect to restriction maps, is well-defined.

We may therefore make the following definition.

Definition 10.4.7.

Let K be a global field. The (global) reciprocity map for K is the homomorphism ΦK: 𝕀K → GKab given by

ΦK = lim ←LΦL∕K,

where the inverse limit is taken over finite abelian extensions L of K with respect to restriction maps.

Remark 10.4.8.

For α ∈𝕀K, we have

ΦK(α) = lim ←LΦL∕K(α) = lim ←L∏v∈VKρLw∕Kv(αv)|L = ∏v∈VK(lim ←LwρLw∕Kv(αv))|Kab = ∏v∈VKρKv(αv)|Kab,

where the first two inverse limits run over the finite abelian extensions of K and the third runs over the completions of the finite abelian extensions of K at a fixed prime of Kab over v.

Our key result is now a reworking of Artin reciprocity.

Theorem 10.4.9 (Global reciprocity).

Let K be a global field.

a.

We have ΦK(a) = 1 for all a ∈ K×.

b.

For every finite abelian extension L of K, the reciprocity map ΦL∕K is surjective with kernel K×NL∕K(𝕀L).

In other words, the global reciprocity map factors through the idèle class group.

Definition 10.4.10.

Let K be a global field.

a.

The global reciprocity map ϕK: ℂK → GKab is the homomorphism induced on the quotient ℂK of 𝕀K by the global reciprocity map ΦK on idèles.

b.

The global reciprocity map ϕL∕K: ℂK → Gal ⁡ (L∕K) for L∕K is the composition of ϕK with restriction to Gal ⁡ (L∕K).

The following is then just a rewording of global reciprocity.

Theorem 10.4.11.

Let L∕K be a finite abelian extension of global fields. The global reciprocity map ϕL∕K induces an isomorphism

ℂK∕NL∕KℂL →∼Gal ⁡ (L∕K).

Remark 10.4.12.

For a number field K, the reciprocity map ϕK is surjective, and its kernel is the connected component ℂK∘ of 1 in ℂK. This connected component is the closure of the image of the subgroup of 𝕀K consisting of idèles that are zero in all nonarchimedean coordinates and positive in all real coordinates.

We have the following compatibilities between reciprocity maps, which are quickly derived from the analogous result in local reciprocity.

Proposition 10.4.13.

Let K be a global field, and let L∕K be a finite separable extension. Then we have the following commutative diagrams:

a.
Global reciprocity: norm and restriction. A full diagram description follows.
Diagram description: Global reciprocity: norm and restriction

The structural squares and triangles displayed here commute.

Objects, listed by row and column:

  • Row 1, from left to right: column 1: blackboard C subscript (L); column 2: G subscript (L) superscript (ab).
  • Row 2, from left to right: column 1: blackboard C subscript (K); column 2: G subscript (K) superscript (ab).

Arrows and lines:

  1. An arrow from blackboard C subscript (L) to G subscript (L) superscript (ab), labelled phi subscript (L).
  2. An arrow from blackboard C subscript (L) to blackboard C subscript (K), labelled N subscript (L / K).
  3. An arrow from G subscript (L) superscript (ab) to G subscript (K) superscript (ab), labelled R subscript (L / K).
  4. An arrow from blackboard C subscript (K) to G subscript (K) superscript (ab), labelled phi subscript (K).

where RL∕K is the restriction map on Galois groups,

b.
Global reciprocity: inclusion and transfer. A full diagram description follows.
Diagram description: Global reciprocity: inclusion and transfer

The structural squares and triangles displayed here commute.

Objects, listed by row and column:

  • Row 1, from left to right: column 1: blackboard C subscript (K); column 2: G subscript (K) superscript (ab).
  • Row 2, from left to right: column 1: blackboard C subscript (L); column 2: G subscript (L) superscript (ab).

Arrows and lines:

  1. An arrow from blackboard C subscript (K) to G subscript (K) superscript (ab), labelled phi subscript (K).
  2. An arrow from blackboard C subscript (K) to blackboard C subscript (L), labelled iota subscript (L / K).
  3. An arrow from G subscript (K) superscript (ab) to G subscript (L) superscript (ab), labelled V subscript (L / K).
  4. An arrow from blackboard C subscript (L) to G subscript (L) superscript (ab), labelled phi subscript (L).

where the map ιL∕K is induced by the natural injection map 𝕀K →𝕀L, and

c.

for any embedding σ : L↪Ksep,

Global reciprocity: conjugation. A full diagram description follows.
Diagram description: Global reciprocity: conjugation

The structural squares and triangles displayed here commute.

Objects, listed by row and column:

  • Row 1, from left to right: column 1: blackboard C subscript (L); column 2: G subscript (L) superscript (ab).
  • Row 2, from left to right: column 1: blackboard C subscript (sigma (L)); column 2: G subscript (sigma (L)) superscript (ab).

Arrows and lines:

  1. An arrow from blackboard C subscript (L) to G subscript (L) superscript (ab), labelled phi subscript (L).
  2. An arrow from blackboard C subscript (L) to blackboard C subscript (sigma (L)), labelled sigma.
  3. An arrow from G subscript (L) superscript (ab) to G subscript (sigma (L)) superscript (ab), labelled sigma superscript (star).
  4. An arrow from blackboard C subscript (sigma (L)) to G subscript (sigma (L)) superscript (ab), labelled phi subscript (sigma (L)).

where the map σ∗ is induced by conjugation by σ.

Proof.

We prove only part a, in that parts b and c is similar. For β ∈ℂL, we have

ϕL(β)|Kab = ∏w∈VLρLw(βw)|Kab = ∏v∈VK∏w∣vρLw(βw)|Kab = ∏v∈VK∏w∣vρKv(NLw∕Kvβw)|Kab = ∏v∈VKρKv(NL∕Kv((β w)w∣v))|Kab = ϕK(NL∕Kβ),

where Remark 10.4.8 is used in the first and last equality and Proposition 9.2.7 is used in the third equality. □

Much as in the case of local class field theory, we have a one-to-one correspondence between norm subgroups of ℂK and finite abelian extensions of K. That norm subgroups are open follows by the same arguments as in the case of local fields.

Theorem 10.4.14 (Existence theorem of global CFT).

The open subgroups of ℂK of finite index are exactly the norm subgroups NL∕KℂL with L a finite abelian extension of K.

The following consequences of the global reciprocity law and existence theorem can then be obtained much as before.

Proposition 10.4.15.

Let K be a global field. For finite abelian extensions L and M of K, we have the following:

a.

NL∕KℂL∩NM∕KℂM = N𝐿𝑀∕Kℂ𝐿𝑀,

b.

NL∕KℂL⋅NM∕KℂM = N(L∩M)∕KℂL∩M, and

c.

NM∕KℂM ⊆ NL∕KℂL if and only if L ⊆ M.

Theorem 10.4.16 (Uniqueness theorem of global CFT).

Let L and M be distinct finite abelian extensions of a global field K. Then NL∕KℂL≠NM∕KℂM.

10.5. Comparison of the approaches

In this section, we compare the ideal-theoretic and idèle-theoretic approaches to global class field theory for number fields. For this, we begin by comparing ideal groups with groups of idèles. This requires a good deal of notation.

Notation 10.5.1.

Let K be a number field, and let 𝔪 be a modulus for K. For a finite place v such that the associated prime ideal 𝔭 divides 𝔪f, set mv = v(𝔪f). For a real place v such that the associated absolute value divides 𝔪∞, set mv = 1, and let U1(Kv) denote the positive real numbers in Kv = ℝ. For exactly those v of either sort, we say that v divides 𝔪 and write v∣𝔪.

Definition 10.5.2.

Let K be a number field and 𝔪 a modulus for K.

a.

The 𝔪-idèle group is the open subgroup 𝕀K𝔪 of 𝕀K given by

𝕀K𝔪 = {α ∈𝕀K∣αv ∈ Umv(Kv) for all v∣𝔪}.
b.

The congruence subgroup of 𝕀K𝔪 with modulus 𝔪 is the open subgroup

W𝔪 = ∏v∈VK v∣𝔪 Umv(Kv)×∏ v∈VK v∤𝔪 𝒪v×.

Proposition 10.5.3.

Let K be a number field and 𝔪 be a modulus for K.

a.

The homomorphism πK𝔪: 𝕀K𝔪 → IK𝔪 given by

πK𝔪(α) = ∏ v∈VK,f v∤𝔪 𝔭vv(αv).

is surjective with kernel W𝔪.

b.

The inclusion of 𝕀K𝔪 in 𝕀K induces an isomorphism

ιK𝔪: 𝕀K𝔪∕K𝔪,1 →∼ℂK.
Proof.

That πK𝔪 is surjective is immediate from the definitions, since elements of 𝕀K𝔪 have arbitrary coodinates (with all but finitely many unit coordinates) for v ∤ 𝔪. The kernel is also clearly W𝔪, as the requirement that an element of 𝕀K𝔪 lie in the kernel is exactly that it be a unit in all finite v not dividing 𝔪. This proves part a.

As for part b, note that K×∩𝕀K𝔪 = K𝔪,1, since the condition that a ∈ K× lie in 𝕀K𝔪 is exactly that it lie in each Umv(Kv) for v∣𝔪, which is to say that it lies in K𝔪,1. For α ∈𝕀K, we may choose b ∈ K× such that αvb−1 ∈ Umv(Kv) for all v dividing 𝔪 by weak approximation. It follows that αb−1 ∈𝕀K𝔪 and therefore that α ∈𝕀K𝔪K×. Since this tells us that 𝕀K = 𝕀K𝔪K×, the map ιK𝔪 is onto. □

Notation 10.5.4.

Let us use ηK𝔪 to denote the composition

ηK𝔪 = π¯K𝔪 ∘(ιK𝔪)−1: ℂ K → Cl ⁡ K𝔪,

where π¯K𝔪: 𝕀K𝔪∕K𝔪,1 → Cl ⁡ K𝔪 is the surjection induced by πK𝔪, and where πK𝔪 and ιK𝔪 are as in Proposition 10.5.3.

The following now gives the comparison between the Artin map ψL∕K𝔪: Cl ⁡ K𝔪 → Gal ⁡ (L∕K) for an abelian extension L∕K with defining modulus 𝔪 and the global reciprocity map ϕL∕K: ℂK → Gal ⁡ (L∕K).

Theorem 10.5.5.

Let L∕K be a finite abelian extension of number fields and 𝔪 a defining modulus for L∕K. Then

ϕL∕K = ψL∕K𝔪 ∘ηK𝔪

as maps from ℂK to Gal ⁡ (L∕K).

Proof.

Choose an idèle class in ℂK, let α ∈𝕀K𝔪 represent it, and let 𝔞 ∈ IK𝔪 be the product

𝔞 = ∏ v∈VK,f v∤𝔪 𝔭vv(αv),

which implies that [𝔞]𝔪 = ηK𝔪([α]). We have

ψL∕K𝔪([𝔞]𝔪) = ∏ v∈VK,f v∤𝔪 ψL∕K𝔪([𝔭v]𝔪)v(αv) = ∏ v∈VK,f v∤𝔪 (𝔭v,L∕K)v(αv) = ∏ v∈VK,f v∤𝔪 ρLw∕Kv(αv). (10.5.1)

Now, note that the conductor 𝔣L∕K divides 𝔪 and that, by Proposition 7.4.21, we have 𝔣L∕K,f𝒪v = 𝔣Lw∕Kv. Since for any finite v dividing 𝔪 and w a place of v lying over it we have by choice of α that αv ≡ 1mod𝔣Lw∕Kv, the definition of the local conductor tells us that ρLw∕Kv(αv) = 1. Moreover, if τ is a real embedding such that ||τ divides 𝔪, we have that τ(αv) > 0, so ρLw∕Kv(αv) = 1 as well. Therefore, we have

∏ v∈VK,f v∤𝔪 ρLw∕Kv(αv) = ∏v∈VKρLw∕Kv(αv) = ΦL∕K(α),

and this together with (10.5.1) yields the result. □

Remark 10.5.6.

The global reciprocity map is defined a product of local reciprocity maps and computes the Artin maps for all finite abelian extensions. The Artin maps avoid much of the difficulty of ramification, as they arise from ideal groups that exclude ramified primes. The connection with local reciprocity is then much weaker, as one only sees the local maps for unramified extensions, whereby any uniformizer is taken to the unique Frobenius in the Galois group of the local extension. In that sense, the Artin maps then miss much of the complexity of the maps of local class field theory, which on the other hand is seen in the idèlic viewpoint.

Let us end this section by examining the case of class field theory over ℚ.

Example 10.5.7.

Let us verify the global reciprocity law for ℚ via our computation of the local reciprocity map over ℚp. The computation is given in Proposition 9.5.8, and we use it repeatedly.

Let p be a prime and k ≥ 1. By the Kronecker-Weber theorem, it suffices to demonstrate that each Φℚ carries −1 and each prime number ℓ to Galois elements that act trivially on a primitive pkth root of unity ζpk.

For all primes q∉{p,ℓ}, we have that ρℚq(ℓ) fixes ζpk. Also, since ℓ > 0, we have ρℝ(ℓ) = 1. If ℓ = p, then we have

Φℚ(ℓ)(ζpk) = ρℚℓ(ℓ)(ζpk) = ζpk.

On the other hand, if ℓ≠p, then we have

Φℚ(ℓ)(ζpk) = ρℚℓ(ℓ)⋅ρℚp(ℓ)(ζpk) = ρℚℓ(ℓ)(ζpkℓ−1) = ζ pk,

As for −1, we have that ρℚq(−1) fixes ζpk for all q≠p, that ρℚp(−1) inverts ζpk, and that ρℝ(−1) is complex conjugation, so

Φℚ(−1)(ζpk) = ρℝ(−1)⋅ρℚp(−1)(ζpk) = ρℝ(−1)(ζpk−1) = ζ pk.

Remark 10.5.8.

Recall that we did not actually prove Theorem 9.5.9 that the map constructed in Proposition 9.5.8 equals the local reciprocity map ρℚp. Via the argument of the last lemma, noting that local reciprocity maps for unramified extensions will be trivial on units and take uniformizers to the Frobenius element, we can actually use the global reciprocity law to give a short proof of this theorem.

Example 10.5.9.

We refer to the subspace

𝕀ℚf = ∏ p prime(ℚp×,ℤ p×)

of 𝕀ℚ as the group of finite idèles. Note that

𝕀ℚ = 𝕀ℚf×ℝ×.

Since ℝ>0 is the kernel of ρℝ, the map Φℚ factors through 𝕀ℚf×{±1}. We have

𝕀ℚ∕ℚ×ℝ >0≅(𝕀ℚf×{±1})∕ℚ×≅𝕀ℚf∕ℚ >0.

Now, consider the map 𝕀ℚf →ℚ>0 that takes a finite idèle to the unique positive generator of the ideal to which it gives rise: that is, α is taken to ∏ ⁡ppvp(αp). This is surjective with kernel

ℤ^× = ∏ pℤp×

and splits the natural inclusion ℚ>0 →𝕀ℚf. In other words, we have an identification

𝕀ℚ∕ℚ×ℝ >0≅ℤ^×,

and the global reciprocity map ϕℚ: ℂℚ → Gℚab factors through ℤ^×.

The resulting map ϕ : ℤ^×→ Gℚab is the inversion of the inverse map to the cyclotomic character χ : Gℚab →ℤ^×. That is, we have

ϕ(a)(ζ) = ζa−1

for all a = (ap)p ∈ℤ^× and roots of unity ζ. To see this, note that if ζpk is a primitive pkth root of unity for some prime p and k ≥ 1, then Theorem 9.5.9 and our definition of ϕ imply that

ϕ(a)(ζpk) = ρℚp(ap)(ζpk) = ζpka−1,

where we may make sense of a−1 modulo pk. In particular, ϕ is an isomorphism, so the image of ℝ>0 is the kernel of ϕℚ.

10.6. Cohomology of the idèles

Let L∕K be a finite Galois extension of global fields with Galois group G. For a place w of L, let Gw denote the decomposition group of w in G.

Lemma 10.6.1.

Let v be a place of K and u denote a fixed place of L lying over it. We have isomorphisms of G-modules

Ind ⁡ GuG(L u×)≅∏ w∣vLw× and Ind ⁡ GuG(𝒪 u×)≅∏ w∣v𝒪w×.
Proof.

The elements σ ∈ G induce isomorphisms σ : Lu → Lσ(u) of K-algebras that in particular preserve valuations. The maps are then the restriction of the map

Ind ⁡ GuG(L u) = ℤ[G]⊗ℤ[Gu]Lu →∏w∣vLw

induced by the biadditive map taking (σ,β) for σ ∈ G and β ∈ Lu to σ(β) ∈ Lσ(u). That this is well-defined follows from the fact that if σ = σ′τ for some τ ∈ Gu, then (σ,β) and (σ′,τ(β)) map to the same element. It is then easily seen to be an isomorphism of G-modules, and then the restrictions are isomorphisms as well. □

By Lemma 10.6.1 and Shapiro’s lemma, we have isomorphisms

H^i(G u,Lu×)≅H^i(G,∏ w∣vLw×) and H^i(G u,𝒪u×)≅H^i(G,∏ w∣v𝒪w×)

for all i ∈ℤ. Together, these enable us to describe the Tate cohomology groups of the idèle group 𝕀L.

Notation 10.6.2.

For a finite set of places S of K, we set

𝕀L,S = ∏v∈S∏w∣vLw××∏ v∉S∏w∣v𝒪w×.

Remark 10.6.3.

We have 𝕀L = lim →S𝕀L,S, where S runs over all finite subsets of VK, with the injective maps induced by inclusions of sets of places. We can of course use any cofinal set of finite sets of places here.

Proposition 10.6.4.

Let S be a finite set of places of K containing the archimedean places and those places that ramify in L∕K. We have isomorphisms

H^i(G,𝕀 L,S)≅⨁ v∈SH^i(G w,Lw×)

for all i ∈ Z, where w denotes a choice of place over v ∈ S.

Proof.

As Tate cohomology commutes with products by construction, we have

H^i(G,𝕀 L,S) ≅∏v∈SH^i(G,∏ w∣vLw×)×∏ v∉SH^i(G,∏ w∣v𝒪w×) ≅∏v∈SH^i(G w,Lw×)×∏ v∉SH^i(G w,𝒪w×),

where in the latter step we have fixed a place w over each place v ∈ VK. Since each v∉S is unramified in L, we have H^i(Gw,𝒪w×) = 0 for all i ∈ℤ by Lemma 9.1.6. The result follows. □

By (9.2.1), Hilbert’s Theorem 90 and via the isomorphisms given by the local invariant maps, we have the following.

Corollary 10.6.5.

Let S be a finite set of places of K containing the archimedean places and those places that ramify in L∕K. We have |H^0(G,𝕀L,S)| = ∏ ⁡v∈S|Gw|, the group H1(G,𝕀L,S) is trivial, and

H2(G,𝕀 L,S) = ⨁ v∈S 1 |Gw|ℤ∕ℤ

for any choice w of place of L over v ∈ S.

Proposition 10.6.6.

We have

H^i(G,𝕀 L)≅⨁ v∈VKH^i(G w,Lw×)

for all i ∈ℤ, where w denotes a choice of place over v ∈ VK.

Proof.

It follows from Remark 10.6.3 that

H^i(G,𝕀 L)≅lim →SH^i(G,𝕀 L,S),

where S runs over the finite sets of places of K containing the archimedean places of K and those places that ramify in L∕K. By Proposition 10.6.4, we then have

H^i(G,𝕀 L)≅lim →S∏v∈SH^i(G w,Lw×)≅⨁ v∈VKH^i(G w,Lw×).

□

Again, we have the following.

Corollary 10.6.7.

We have H1(G,𝕀L) = 0 and

H2(G,𝕀 L)≅⨁ v∈VK 1 |Gw|ℤ∕ℤ.

10.7. The first inequality

We will restrict our proofs of the reciprocity laws of global class field theory to number fields. We remark that the proofs we give carry over with little-to-no change to the function field setting for extensions of degree prime to the characteristic, but the proof of the second inequality in the case of equal characteristic requires some additional work.

So, we now let L∕K be a finite Galois extension of number fields, still with Galois group G. Our interest is in the G-cohomology of ℂL, in that we would like to define invariant maps that make it into a class formation. Recall that we have a surjection 𝕀L → IL taking an idèle to the fractional ideal it defines, and this map induces a surjection ℂL → Cl ⁡ L.

We will use S to denote a finite set of places of K, which we consistently suppose contains the archimedean places of K. (In the function field setting, S should be taken to be nonempty and the class group considered below should be replaced by a certain divisor class group.)

Lemma 10.7.1.

Suppose that S contains a set of finite places generating the ideal class group of 𝒪L. Then 𝕀K = 𝕀K,SK×.

Proof.

The kernel of the surjection 𝕀K → IK is generated by the product of the local units at finite places and local multiplicative groups at infinite places, so the kernel of ℂK → Cl ⁡ K is as well. We then note that the class group is generated by the classes of the chosen set of finite representatives 𝔭, and these are the images of idèles in 𝕀K,S that are 1 in places but that for 𝔭 and the uniformizer at the prime. Thus 𝕀K∕𝕀K,SK× = 0. (Since we take 𝒪w× = Lw× for archimedean places w, it is not strictly necessary to include these places in our set.) □

Definition 10.7.2.

The ring of S-integers 𝒪K,S of K is the set of elements of K that lie in the valuation ring at all nonarchimedean places of K not in S. The S-unit group of K is 𝒪K,S×.

Remark 10.7.3.

We have 𝕀K,S∩K× = 𝒪K,S×.

Notation 10.7.4.

We use 𝒪L,S to denote the SL-integer ring of 𝒪L, where SL denotes the set of places of L lying over those in S.

We have the following extension of Dirichlet’s unit theorem. It also holds for function fields, though we restrict to the case of number fields.

Proposition 10.7.5.

For a finite set S of places of a number field K containing its archimedean places, we have

rank ⁡ ℤ𝒪K,S× = |S|−1.
Proof.

By Dirichlet’s unit theorem, we know that rank ⁡ ℤ𝒪K = r1(K)+r2(K)−1, one less than the number of archimedean places of K. We have an exact sequence

1 →𝒪K×→𝒪 K,S×→∑ v∈Sfv⨁ v∈Sfℤ,

where Sf denotes the set of finite places in S. It then suffices to exhibit an S-unit with nonzero additive valuation at a given v ∈ Sf and trivial valuation at all other finite places of K. For this, note that some power of the prime 𝔭 corresponding to v is principal, and any generator is then an S-unit with the desired property. □

As with local class field theory, much can be gained from the study of cyclic extensions.

Theorem 10.7.6.

If L∕K is cyclic, then h(ℂL) = [L : K], for G = Gal ⁡ (L∕K).

Proof.

Let S contain the ramified places in L∕K and a set of finite places lying below primes generating the ideal class group of 𝒪L. We have ℂL≅𝕀L,S∕𝒪L,S× by Lemma 10.7.1. Thus, we have

h(ℂL) = h(𝕀L,S) h(𝒪L,S×).

Now, consider the ℝ-vector space V with basis the elements of the set SL of places of L over those in S. Let G act on V by its canonical permutation of the standard basis. Consider its ℤ[G]-submodule A generated over ℤ by the standard basis of V. We have 𝐴≅⊕ ⁡ v∈SInd ⁡ GwG(ℤ), where w is again used to denote a place over v. Then

h(A)≅∏v∈Sh(Gw,ℤ) = ∏v∈Snv,

where nv denotes the local degree of L∕K at a prime over v.

We define a second lattice as follows. We have the homomorphism ℓL,S: 𝒪L,S×→ V given by

ℓL,S(β) = (log⁡∥β∥w)w∈SL.

By Corollary 4.4.2, we have that ker⁡ℓL,S is finite and, by the product formula, the image B0 of ℓL,S is contained in the hyperplane V0 of elements that sum to zero. The 𝒪L,S× is the rank of 𝒪L× plus the number of finite places in S, as some power of any finite prime is principal, and from this and Theorem 10.7.5, we see that B0 must be a complete lattice in the hyperplane V0.

Let x = (1)w∈SL ∈ VG, and set

B = ℤ𝑥+B0,

which is a complete lattice in V. We have an exact sequence of G-modules

0 → B0 → B →ℤ𝑥 → 0,

so h(B) = h(B0)h(ℤ) = 𝑛h(𝒪L,S×). On the other hand, any two complete lattices in a finite-dimensional ℝ-vector space are isomorphic upon tensor product with ℚ, from which one can see that their Herbrand quotients are equal. Thus we have h(A) = h(B), which tells us upon application of Corollary 10.6.5 that

h(𝒪L,S×) = 1 nh(A) = 1 n∏v∈Snv.

Combining this with our computation of h(𝕀L,S) yields the theorem. □

As a corollary of this, we obtain what is known as the first inequality of global class field theory.

Corollary 10.7.7 (The first inequality).

For any finite cyclic extension L∕K, we have [ℂK : NL∕KℂL] ≥ [L : K].

Proof.

The quantity on the left-hand side of the inequality is the order of H^0(G,ℂL), which is a multiple of the Herbrand quotient. □

The following is a simple consequence of the much stronger Čebotarev density theorem. We prove it using the first inequality.

Corollary 10.7.8.

Let L∕K be finite abelian and S be a finite set of places of K containing the archimedean places and the ramified places in L∕K. Then G is generated by the Frobenius elements in G of places not in S.

Proof.

We may by enlarging S suppose that it contains a set of representatives of the class group of K. Let E be the fixed field of the subgroup of G generated by the Frobenius elements of nonarchimedean places not in S. Then for any v∉S and place w lying over v in G, we have that the local extension Ew∕Kv is trivial, and in particular that NEw∕KvEw× = Kv×. Thus, we have that NE∕K𝕀E,S = 𝕀K,S, and by Lemma 10.7.1 we have that K×𝕀K,S = 𝕀K, so NE∕KℂE = ℂK. This implies the same equality with E replaced by any cyclic subextension, and then by the first inequality, such an extension must be trivial, so E = K. □

We leave it to the reader to prove the following additional consequences in a similar fashion, using Corollary 10.7.8.

Corollary 10.7.9.

Let L∕K be cyclic of prime power degree. Then there exist infinitely many primes of K that remain inert in L∕K.

Corollary 10.7.10.

Let L1,…,Lt be cyclic extensions of K of prime degree p such that each Li is disjoint from the compositum of the Lj for j≠i. Then there are infinitely many primes of K that are inert in L1 and split completely in Li for i ≥ 2.

10.8. The second inequality

We turn to the opposite inequality, known as the second inequality, for general Galois extensions of number fields, beginning with Kummer extensions of prime exponent.

For now, let us fix n ≥ 1. For a subset T of S, we set

IT = ∏v∈S−T Kv××∏ v∈T Kv×n×∏ v∉S𝒪v×.

Lemma 10.8.1.

Suppose that μn ⊂ K and S contains the primes dividing n. Let T be a finite subset of S, and set Δ = K×∩IT and L = K(Δ1∕n). Then L∕K is unramified outside of the places in S−T and completely split at the places in T .

Proof.

It suffices to see that for all a ∈Δ, the extension Kv(a1∕n)∕Kv is unramified if v ∈ S−T and trivial if v ∈ T . If v∉S, then a ∈𝒪v×, so Kv(a1∕n)∕Kv is tamely ramified, being of prime-to-p degree. Its is moreover unramified as the group ⟨a⟩Kv×n contains no mth power of a uniformizer of Kv for m properly dividing n. If v ∈ T , then a ∈ Kv×n, so clearly Kv(a1∕n) = Kv. □

The following simple group-theoretic lemma will be of use to us shortly.

Lemma 10.8.2.

For subgroups A, B, and C of a group with A ≤ B of finite index, we have

[A : B] = [𝐴𝐶 : 𝐵𝐶][A∩C : B∩C].
Proof.

By the second and third isomorphism theorems and the fact that A ≤ B, we have

𝐴𝐶 𝐵𝐶≅𝐴𝐶∕C 𝐵𝐶∕C≅ A∕(A∩C) (A∩𝐵𝐶)∕(A∩C)≅ A (A∩C)B≅ A∕B (A∩C)B∕B,

and (A∩C)B∕𝐵≅(A∩C)∕(B∩C). □

Lemma 10.8.3.

Suppose that μn ⊂ K and S contains the primes over n and a set of representatives for Cl ⁡ K. Let S1 be a subset of S and S2 = S−S1. Let Δi = K×∩ISi and Li = L(Δi1∕n) for i ∈{1,2}. Then

a.

IS1 ⊆ NL2∕K𝕀L2 and IS2 ⊆ NL1∕K𝕀L1, and

b.

[ℂK : K×IS1][ℂK : K×IS2] = [L1 : K][L2 : K].

Proof.

For part a, let α = (αv)v ∈ISi with i ∈{1,2}. Let j be such that {i,j} = {1,2}, and let E = Lj for brevity. For v ∈ Si, we have αv ∈ Kv×n. By the the local reciprocity law, the quotient of Kv× by the norm group NEw∕KvEw× for w∣v has exponent dividing n, so therefore αv lies in it. Any v ∈ Sj splits completely in E∕K, so αv is automatically a local norm for all places w∣v. For all v∉S, the extension E∕K is unramified at v, and αv ∈𝒪v×. Since the local extension is unramified, its norm group contains 𝒪v× (and in fact is generated by it and the uniformizer of Kv to the power of the residue degree of the extension). Thus, α ∈ NE∕K𝕀E. That is, we have ISi ⊆ NE∕K𝕀E.

As for part b, by Lemma 10.7.1 and the group-theoretic equality of indices that is Lemma 10.8.2, we have

[𝕀K : K×I S1] = [K×𝕀 K,S : K×I S1] = [𝕀K,S : IS1] [K×∩𝕀K,S : K×∩IS1] = [𝕀K,S : IS1] [𝒪K,S× : Δ1].

Note that

𝕀K,S∕IS1≅∏v∈S1Kv×∕K v×n,

For a place v over p we have

[Kv× : K v×n] = n2 ⋅∥n∥ v−1,

since Kv×≅ℤ×μ(Kv)×ℤp[Kv:ℚp] by Propositions 6.3.4 and 6.3.9 and μn ⊆ μ(Kv). In fact, for any place v, we have we still have the equality, noting that the only archimedean places for n ≥ 3 are complex. Letting s = |S|, we then have

[𝕀K,S : IS1][𝕀K,S : IS2] = n2s∏ v∈S∥n∥v−1 = n2s∏ v∉S∥n∥v = n2s

by the product formula and the fact that every place that divides n and all archimedean places lie in S.

We also have [𝒪K,S : 𝒪K,S×n] = ns by Proposition 10.7.5 (which says that 𝒪K,S×≅ℤs−1 ×μ(K)), and [Δ1 : 𝒪K,S×n] = [L1 : K] by Kummer theory. Thus,

[𝒪K,S× : Δ1][𝒪 K,S× : Δ2] = n2s([L1 : K][L2 : K])−1,

and we then have

[𝕀K : K×I S1][𝕀K : K×I S2] = [𝕀K,S : IS1][𝕀K,S : IS2] [𝒪K,S× : Δ1][𝒪K,S× : Δ2] = [L1 : K][L2 : K],

as claimed. □

We now specialize to the case that n equals a prime p.

Proposition 10.8.4.

Let K be a number field that contains the pth roots of unity for a prime p. Let L∕K be finite abelian of exponent p. Then [ℂK : NL∕KℂL] divides [L : K].

Proof.

Let k be such that [L : K] = pk, and let a1,…,ak ∈ K× be such that L = K(a11∕p,…,ak1∕p). We aim to construct finite disjoint sets S1 and S2 of primes such that S = S1 ∪S2 satisfies the conditions of Lemma 10.8.3 with K×IS2 = 𝕀K and L2 = L. We will then have [ℂK : NL1∕KℂL1] = 1, which by the first inequality forces L1 = K, and Lemma 10.8.3 then implies the desired divisibility.

To start, choose S1 to consist of the archimedean places, the primes over p, a set of representatives of Cl ⁡ K, and every finite place v such that v(ai)≠0 for 1 ≤ i ≤ k. Then K×𝕀K,S1 = 𝕀K by Lemma 10.7.1, and ai ∈𝒪K,S1× for 1 ≤ i ≤ k. Let b1,…,bt ∈𝒪K,S1× be such that the images of a1,…,ak,b1,…,bt form a basis of 𝒪K,S1×∕𝒪K,S1×p. Now, by Corollary 10.7.10, we may choose S2 = {v1,…,vt}, where for each vi splits completely in L∕K, remains inert in K(bi1∕p)∕K, and splits completely in K(bj1∕p)∕K for j≠i.

Recalling that we have set n = p, we have

𝕀K,S1 ∩IS2 = ∏v∈S1Kv××∏ v∈S2𝒪v×p×∏ v∉S𝒪v×.

We then have

𝕀K,S1∕(𝕀K,S1 ∩IS2)≅∏i=1t𝒪 vi×∕𝒪 vi×p,

and since the residue characteristic of Kvi is not p and μp ⊂ Kvi×, we have 𝒪vi×∕𝒪vi×p≅ℤ∕𝑝ℤ. Note that bi ∈𝕀K,S1, and for 1 ≤ i,j ≤ t, we have bi∉𝒪vj×p if and only if i = j. Thus, the images of the bi generate 𝕀K,S1∕(𝕀K,S1 ∩IS2). Since the bi lie in K, this tells us that K×IS2 = K×𝕀K,S1 = 𝕀K, as desired.

Next, note that

IS2 ⊆𝕀K,S1𝕀Kp = 𝕀 K,S1(K×𝕀 K,S1)p = 𝕀 K,S1K×p

by what we have just shown. In particular,

Δ2 = IS2 ∩K×⊆𝒪 K,S1×K×p,

and Δ2K×p∕K×p is generated by the S1-units that are locally pth powers at all v ∈ S2. Recall that 𝒪K,S1×K×p∕K×p is generated by the images of a1,…,ak,b1,…,bt. Since each vj splits completely in the subfield K(ai1∕p) of L, we have that ai ∈𝒪vj×p, so ai ∈Δ2 for 1 ≤ i ≤ k. On the other hand, any non-pth-power in ⟨b1,…,bt⟩ has nontrivial image in 𝒪vj×∕𝒪vj×p for some j, so does not lie in Δ2K×p. Thus, the images of the ai generate Δ2K×p∕K×p, which is to say that L2 = L. □

We now turn to more general extensions, no longer supposing μn ⊂ K.

Lemma 10.8.5.

For any finite extension L∕K, the index [ℂK : NL∕KℂL] is finite and divisible only by primes dividing [L : K].

Proof.

Once we have finiteness, the divisibility statement follows from the fact that for any α ∈ℂK, we have α[L:K] ∈ NL∕KℂL. For finiteness, we may suppose that L∕K is Galois, since the norms of idèle classes from the Galois closure of L to K will also be norms from L. By Lemma 10.7.1, we may find a finite set of primes S of K containing the primes that ramify in L such that 𝕀L = 𝕀L,SL× and 𝕀K = 𝕀K,SK×. Then

[ℂK : NL∕KℂL] = [𝕀K : K×N L∕K𝕀L] = [K×𝕀 K,S : K×N L∕K𝕀L,S] ≤ [𝕀K,S : NL∕K𝕀L,S] = ∏v∈S[Lw : Kv],

where w is any place of L over v, with the last equality by Proposition 10.6.5. □

For brevity, for a finite extension E∕F of number fields, we let nE∕F = [ℂF : NE∕F ℂE].

Lemma 10.8.6.

Let M∕K be a finite Galois extension and L an intermediate field. Then nM∕K divides nM∕LnL∕K.

Proof.

Note that

nM∕K = [NL∕KℂL : NM∕KℂM]nL∕K.

The map NL∕K induces a surjective map

ℂL∕NM∕LℂM → NL∕KℂL∕NM∕KℂM

so [NL∕KℂL : NM∕KℂM] divides nM∕L. □

The following is then immediate from the multiplicativity of degrees of field extensions.

Corollary 10.8.7.

Let M∕K be a finite Galois extension and L an intermediate field. If nM∕L∣[M : L] and nL∕K∣[L : K], then nM∕K∣[M : K].

Theorem 10.8.8.

Let L∕K be a finite Galois extension of number fields with Galois group G. Then H^0(G,ℂL) and H2(G,ℂL) have order dividing [L : K], and H1(G,ℂL) = 0.

Proof.

By Lemma 9.1.12 applied in the cases (i,r) = (0,1),(1,0),(2,1) in that order, the result follows for arbitrary Galois extensions from the case of cyclic extensions of prime degree. So, suppose that L∕K is cyclic of degree a prime p. If we can show that nL∕K = |H^0(G,ℂL)| divides p, then the 2-periodicity of Tate cohomology and Theorem 10.7.6 give the result.

By Lemma 8.2.3, we have that nL∕K∣nL(μp)∕K, and by Lemma 10.8.6, we have that

nL(μp)∕K∣nL(μp)∕K(μp)nK(μp)∕K.

Since nK(μp)∕K is prime to p and nL∕K is a power of p by Lemma 10.8.5, we have that nL∕K divides nL(μp)∕K(μp), which is p by Proposition 10.8.4. □

Corollary 10.8.9 (The second inequality).

For any finite Galois extension L∕K, we have

[ℂK : NL∕KℂL] ≤ [L : K].

From the fact that H1(G,ℂL) = 0, we obtain the interesting consequence that in cyclic extensions, global elements that are local norms everywhere are global norms.

Corollary 10.8.10.

Suppose that L∕K is cyclic. If a ∈ K× and a ∈ NLw∕KvLw× for some w∣v for all places v ∈ VK, then a ∈ NL∕KL×.

Proof.

Since H1(G,ℂL) = 0, the map

H2(G,L×) → H2(G,𝕀 L)

is an injection. Since G is cyclic, we have that the corresponding map on 0th Tate cohomology groups is injective as well. In other words, the map

K×∕N L∕KL×→⨁ v∈VKK×∕N Lw∕KvLw×

is injective, noting that the norm group for Lw∕Kv is independent of the choice of w∣v (as L∕K is Galois). This is exactly what was claimed. □

10.9. The reciprocity law

We continue to let K denote a number field and S a finite set of places of K containing the archimedean places. In this section, we use L to denote a finite abelian extension of K with Galois group G.

As noted in the proof of the Corollary 10.8.10, the triviality of H1(G,ℂL) implies that the map

Br ⁡ (L∕K) = H2(G,L×) → H2(G,𝕀 L)≅⨁ v∈SBr ⁡ (Lw∕Kv)

is injective. In the direct limit over all Galois extensions L of K, we obtain an injective map

Br ⁡ (K)↪⨁ v∈VKBr ⁡ (Kv).

Let us use inv ⁡ v: Br ⁡ (K) →ℚ∕ℤ to denote the composition of the map Br ⁡ (K) → Br ⁡ (Kv) with the local invariant map inv ⁡ Kv. (Here, inv ⁡ ℝ is the unique injection of the group Br ⁡ (ℝ) of order 2 in ℚ∕ℤ.) We see from the fact that Br ⁡ (K) maps to the direct sum that

∑v∈VKinv ⁡ v: Br ⁡ (K) →ℚ∕ℤ

is well-defined. We will show that this map is zero.

In the following, we also use the notation inv ⁡ v to denote the composition

inv ⁡ v: H2(G,𝕀 L) → Br ⁡ (Kv) →inv ⁡ Kvℚ∕ℤ.

Lemma 10.9.1.

For α ∈𝕀K and χ ∈ Hom ⁡ (G,ℚ∕ℤ), we have

∑v∈VKinv ⁡ v(α¯∪𝛿𝜒) = χ(ΦL∕K(α)),

where δ : H1(G,ℚ∕ℤ) → H2(G,ℤ) is the connecting homomorphism for 0 →ℤ →ℚ →ℚ∕ℤ → 0, and α¯ denotes the class of α in H^0(G,𝕀L).

Proof.

By definition and the compatibility of cup products with restriction, we have

inv ⁡ v(α¯∪𝛿𝜒) = inv ⁡ Kv(αv¯∪δχv),

where αv¯ denotes the image of αv in H^0(Gv,Lw×) for a place w over v, where χv ∈ H1(Gv,ℚ∕ℤ) is the restriction of χ to the decomposition group Gw, and where δ continues to denote the corresponding connecting homomorphism. By Proposition 8.1.10, we have that

inv ⁡ Kv(αv¯∪δχv) = χv(ρLw∕Kv(αv)).

By definition of ΦL∕K and the fact that χ is a homomorphism, we have that

χ(ΦL∕K(α)) = ∑v∈VKχv(ρLw∕Kv(αv)),

hence the result. □

From the global recirpocity law for ℚ, we may easily prove the global reciprocity law for cyclotomic extensions.

Lemma 10.9.2.

Let L be an extension of K contained in K(μN) for some N ≥ 1. Then ΦL∕K(a) = 1 for all a ∈ K×.

Proof.

For E = ℚ(μN), we have ΦL∕K(a)|E = ΦE∕ℚ(NK∕ℚ(a)) by part a of Proposition 10.4.13. Since the restriction map G → Gal ⁡ (E∕ℚ) is injective, we are reduced to the already proven reciprocity law for ℚ. □

Lemma 10.9.3.

For any n ≥ 1, let S denote the set of places of K dividing n and all real places. Then there exists a cyclic extension L of K contained in K(μN) for some N ≥ 1 such that the local degree of L∕K is divisible by n for all v ∈ Sf and is equal to 2 for all real places of K.

Proof.

Without loss of generality, we may suppose that 2 divides n. Let n = p1r1⋯pkrk be the prime factorization of n. If pi is odd, let Li be the maximal pro-p subextension of the field given by adjoining to K all pi-power roots of unity. Otherwise, let Li be the extension of K given by adjoining ζ −ζ−1 for all 2-power roots of unity ζ. The unique degree 2 subextension of the latter field has no real places. For each i, the completions of the fields Li at places v ∈ Sf are infinite pro-pi procyclic extensions. In particular, the compositum L of the fields Li is a procyclic extension of K that contains a finite degree extension that is the desired subfield. □

We require the following simple cohomological lemma, the proof of which is left to the reader.

Lemma 10.9.4.

Let G be a finite cyclic group of order n and χ : G →ℚ∕ℤ be an injective character. Let δ be the connecting map for 0 →ℤ →ℚ →ℚ∕ℤ → 0. Let g be a generator of G, and let ug ∈H^−2(G,ℤ) be as in Proposition 1.10.3. Viewing H^0(G,ℤ) and ℤ∕𝑛ℤ, and letting χ~: G →ℤ∕𝑛ℤ be the isomorphism obtained from χ by multiplication by n, we have

ug∪𝛿𝜒 = χ~(g) ∈ℤ∕𝑛ℤ.

In other words, we have the following, the map being inverse to cup product with ug for a generator g ∈ G with χ(g) = 1 n.

Corollary 10.9.5.

Let G be a finite cyclic group of order n and χ : G →ℚ∕ℤ be an injective character. Let δ be the connecting map for 0 →ℤ →ℚ →ℚ∕ℤ → 0. For any G-module A and and i ∈ℤ, the map

H^i(G,A) →H^i+2(G,A),c↦𝛿𝜒 ∪c

is an isomorphism.

Proposition 10.9.6.

The map ∑ ⁡v∈VKinv ⁡ v: Br ⁡ (K) →ℚ∕ℤ is trivial.

Proof.

Let β ∈ Br ⁡ (K), and let n be the least common multiple of the orders of the elements inv ⁡ v(β) for v ∈ VK. Let S contain the places where inv ⁡ v(β) is nonzero, and let L be as in Lemma 10.9.3. Then for each v ∈ S, the group Br ⁡ (Lw∕Kv) sits in Br ⁡ (Kv) as as a cyclic subgroup of order a multiple of n, and hence it contains the image of β. It follows that β ∈ Br ⁡ (L∕K). Since L∕K is cyclic, we have an injective character χ : G →ℚ∕ℤ. Corollary 10.9.5 then tells us that there exists b ∈ K× such that such that b∪𝛿𝜒 = β. By Lemma 10.9.1, we have

∑v∈VKinv ⁡ v(β) = χ(ΦL∕K(b)),

and ΦL∕K(b) = 0 by Lemma 10.9.2. □

We can now prove that the global reciprocity map factors through ℂK.

Corollary 10.9.7.

We have ΦK(a) = 1 for all a ∈ K×.

Proof.

It suffices to show that ΦL∕K(a) = 1 for all finite abelian extensions L∕K, and for this, it suffices to show that χ(ΦL∕K(a)) = 0 for all characters χ : G →ℚ∕ℤ for all such L. By Lemma 10.9.1, the latter quantity equals ∑ ⁡v∈VKinv ⁡ v(a¯∪𝛿𝜒), but this is zero by Proposition 10.9.6. □

Note that for any finite extension L of K, we have an injection ℂK↪ℂL by Lemma 10.3.17. We aim to construct an invariant map inv ⁡ : H2(GK,ℂKsep) →ℚ∕ℤ to show that ℂKsep = lim →LℂL together with the invariant maps associated for finite separable extensions of K forms a class formation.

Since since H1(G,ℂL) = 0 and

H2(G,𝕀 L)≅⨁ v∈VKBr ⁡ (Lw∕Kv),

the latter by Proposition 10.6.6, we have an exact sequence

0 → Br ⁡ (L∕K) →⨁ v∈VKBr ⁡ (Lw∕Kv) → H2(G,ℂ L).

Recall that we have an isomorphism

inv ⁡ Lw∕Kv: Br ⁡ (Lw∕Kv) →∼ 1 |Gw|ℤ∕ℤ,

where Gw is the decomposition group at any w∣v. The sum of these local invariant maps

∑v∈VKinv ⁡ Lw∕Kv: ⨁ v∈SBr ⁡ (Lw∕Kv) → 1 |G|ℤ∕ℤ

is surjective due to the existence of an inert prime w over some v ∈ VK. Let

inv ⁡ ~L∕K: H2(G,𝕀 L) ↠ 1 |G|ℤ∕ℤ

denote the composite map. By Proposition 10.9.6, we have that Br ⁡ (L∕K) is contained in the kernel of inv ⁡ ~L∕K. That is, inv ⁡ ~L∕K factors through a surjective global invariant map

inv ⁡ L∕K: BL∕K ↠ 1 |G|ℤ∕ℤ

from the image BL∕K of H2(G,𝕀L) → H2(G,ℂL). We aim to show that H2(G,𝕀L) → H2(G,ℂL) is surjective so BL∕K = H2(G,ℂK), and inv ⁡ L∕K is an injective. We start with cyclic extensions.

Lemma 10.9.8.

Let L∕K be finite cyclic. Then H2(G,𝕀L) → H2(G,ℂL) is surjective, and the invariant map

inv ⁡ L∕K: H2(G,ℂ L) →∼ 1 |G|ℤ∕ℤ

is an isomorphism.

Proof.

We have H3(G,L×)≅H1(G,L×) = 0 by the periodicty of Tate cohomology, so BL∕K = H2(G,ℂL). Since H2(G,ℂL) has order |G| by the first and second inequalities, the result follows. □

The next lemma shows that global invariant maps behave as expected under restriction.

Lemma 10.9.9.

Let E be a finite extension of K contained in L, and set H = Gal ⁡ (L∕E). The diagram

Restriction of global invariant maps. A full diagram description follows.
Diagram description: Restriction of global invariant maps

The structural squares and triangles displayed here commute.

Objects, listed by row and column:

  • Row 1, from left to right: column 1: H superscript (2)(G,blackboard I subscript (L)); column 2: fraction (1) over (|G|) blackboard Z / blackboard Z.
  • Row 2, from left to right: column 1: H superscript (2)(H,blackboard I subscript (L)); column 2: fraction (1) over (|H|) blackboard Z / blackboard Z.

Arrows and lines:

  1. An arrow from H superscript (2)(G,blackboard I subscript (L)) to fraction (1) over (|G|) blackboard Z / blackboard Z, labelled tilde of (inv) subscript (L / K).
  2. An arrow from H superscript (2)(G,blackboard I subscript (L)) to H superscript (2)(H,blackboard I subscript (L)), labelled Res.
  3. An arrow from fraction (1) over (|G|) blackboard Z / blackboard Z to fraction (1) over (|H|) blackboard Z / blackboard Z, labelled [E:K].
  4. An arrow from H superscript (2)(H,blackboard I subscript (L)) to fraction (1) over (|H|) blackboard Z / blackboard Z, labelled tilde of (inv) subscript (L / E).

commutes.

Proof.

Since the global invariant maps are sums of local invariant maps, this reduces to the commutativity of the diagram

Restriction and sums of local invariant maps. A full diagram description follows.
Diagram description: Restriction and sums of local invariant maps

The structural squares and triangles displayed here commute.

Objects, listed by row and column:

  • Row 1, from left to right: column 1: H superscript (2)(G, product subscript (w divides v) L subscript (w) superscript (times)); column 2: fraction (1) over (|G|) blackboard Z / blackboard Z.
  • Row 2, from left to right: column 1: H superscript (2)(H, product subscript (w divides v) L subscript (w) superscript (times)); column 2: fraction (1) over (|H|) blackboard Z / blackboard Z.

Arrows and lines:

  1. An arrow from H superscript (2)(G, product subscript (w divides v) L subscript (w) superscript (times)) to fraction (1) over (|G|) blackboard Z / blackboard Z, labelled sum subscript (w divides v) inv subscript (L subscript (w) / K subscript (v)).
  2. An arrow from H superscript (2)(G, product subscript (w divides v) L subscript (w) superscript (times)) to H superscript (2)(H, product subscript (w divides v) L subscript (w) superscript (times)), labelled Res.
  3. An arrow from fraction (1) over (|G|) blackboard Z / blackboard Z to fraction (1) over (|H|) blackboard Z / blackboard Z, labelled [E:K].
  4. An arrow from H superscript (2)(H, product subscript (w divides v) L subscript (w) superscript (times)) to fraction (1) over (|H|) blackboard Z / blackboard Z, labelled sum subscript (w divides v) inv subscript (L subscript (w) / K subscript (v)).

where v ∈ VK and w denotes a place of L over v. By Lemma 10.6.1 and Shapiro’s lemma, this reduces to the commutativity of

Restriction on relative local Brauer groups. A full diagram description follows.
Diagram description: Restriction on relative local Brauer groups

The structural squares and triangles displayed here commute.

Objects, listed by row and column:

  • Row 1, from left to right: column 1: Br (L subscript (w subscript (0)) / K subscript (v)); column 2: fraction (1) over (|G|) blackboard Z / blackboard Z.
  • Row 2, from left to right: column 1: direct sum subscript (u divides v) Br (L subscript (w) / E subscript (u)); column 2: fraction (1) over (|H|) blackboard Z / blackboard Z.

Arrows and lines:

  1. An arrow from Br (L subscript (w subscript (0)) / K subscript (v)) to fraction (1) over (|G|) blackboard Z / blackboard Z, labelled inv subscript (L subscript (w subscript (0)) / K subscript (v)).
  2. An arrow from Br (L subscript (w subscript (0)) / K subscript (v)) to direct sum subscript (u divides v) Br (L subscript (w) / E subscript (u)), labelled Res.
  3. An arrow from fraction (1) over (|G|) blackboard Z / blackboard Z to fraction (1) over (|H|) blackboard Z / blackboard Z, labelled [E:K].
  4. An arrow from direct sum subscript (u divides v) Br (L subscript (w) / E subscript (u)) to fraction (1) over (|H|) blackboard Z / blackboard Z, labelled sum subscript (u divides v) inv subscript (L subscript (w) / E subscript (u)).

where u runs over the places over v in E, we use w to denote a place over u, and w0 denotes a fixed place of L over v. Here, each w is conjugate to w0 over K, and the u-coordinate fo the restriction map Res ⁡ is induced by conjugation by σ ∈ G with σ(w0) = w followed by restriction. We remark that inv ⁡ Lw∕Kv ∘σ∗ = inv ⁡ Lw 0∕Kv by definition. The local invariant maps have the property that

inv ⁡ Lw∕Eu ∘Res ⁡ Eu∕Kv = [Eu : Kv]inv ⁡ Lw∕Kv,

so we have

(∑u∣vinv ⁡ Lw∕Eu) ∘Res ⁡ Eu∕Kv = ∑u∣v[Eu : Kv]inv ⁡ Lw∕Kv = [E : K]inv ⁡ Lw∕Kv,

the latter step as the sum of local degrees is the global degree. □

We next treat the general case.

Proposition 10.9.10.

The map H2(G,𝕀L) → H2(G,ℂL) is surjective, and the invariant map

inv ⁡ L∕K: H2(G,ℂ L) →∼ 1 |G|ℤ∕ℤ

is an isomorphism.

Proof.

Since H1(G,AL) = 0, where AL ∈{L×,𝕀L,ℂL}, the inflation maps

Inf ⁡ : H2(G,A L) → H2(G K,AKsep)

are injective, being part of the inflation-restriction sequences. The direct limit of the maps inv ⁡ ~L∕K over Galois extensions L∕K provide a surjective map

inv ⁡ ~K: H2(G K,𝕀Ksep) →ℚ∕ℤ.

which factors through a surjective map

inv ⁡ K: BK →ℚ∕ℤ,

where BK = lim →LBL∕K. By Lemma 10.9.3,

H2(G K,𝕀Ksep)≅⨁ v∈VKBr ⁡ (Kv)

is the union of its subgroups H2(Gal ⁡ (F∕K),𝕀F ), where F runs over the set E of cyclic cyclotomic extensions of K. From the map of exact sequences

Brauer groups and cohomology of ideles and idele classes. A full diagram description follows.
Diagram description: Brauer groups and cohomology of ideles and idele classes

The two displayed rows are exact, and the squares commute.

Objects, listed by row and column:

  • Row 1, from left to right: column 1: 0; column 2: union subscript (F in script E) Br (F / K); column 3: union subscript (F in script E) H superscript (2)( Gal (F / K),blackboard I subscript (F)); column 4: union subscript (F in script E) H superscript (2)( Gal (F / K),blackboard C subscript (F)); column 5: 0.
  • Row 2, from left to right: column 1: 0; column 2: Br (K); column 3: H superscript (2)(G subscript (K),blackboard I subscript (K superscript (sep))); column 4: H superscript (2)(G subscript (K),blackboard C subscript (K superscript (sep))).

Arrows and lines:

  1. An arrow from 0 (row 1, column 1) to union subscript (F in script E) Br (F / K), without a label.
  2. An arrow from union subscript (F in script E) Br (F / K) to union subscript (F in script E) H superscript (2)( Gal (F / K),blackboard I subscript (F)), without a label.
  3. A hooked arrow from union subscript (F in script E) Br (F / K) to Br (K), without a label.
  4. An arrow from union subscript (F in script E) H superscript (2)( Gal (F / K),blackboard I subscript (F)) to union subscript (F in script E) H superscript (2)( Gal (F / K),blackboard C subscript (F)), without a label.
  5. An arrow from union subscript (F in script E) H superscript (2)( Gal (F / K),blackboard I subscript (F)) to H superscript (2)(G subscript (K),blackboard I subscript (K superscript (sep))), labelled isomorphism symbol.
  6. An arrow from union subscript (F in script E) H superscript (2)( Gal (F / K),blackboard C subscript (F)) to 0 (row 1, column 5), without a label.
  7. A hooked arrow from union subscript (F in script E) H superscript (2)( Gal (F / K),blackboard C subscript (F)) to H superscript (2)(G subscript (K),blackboard C subscript (K superscript (sep))), without a label.
  8. An arrow from 0 (row 2, column 1) to Br (K), without a label.
  9. An arrow from Br (K) to H superscript (2)(G subscript (K),blackboard I subscript (K superscript (sep))), without a label.
  10. An arrow from H superscript (2)(G subscript (K),blackboard I subscript (K superscript (sep))) to H superscript (2)(G subscript (K),blackboard C subscript (K superscript (sep))), without a label.

we see that Br ⁡ (K) = ⋃ ⁡ F∈EBr ⁡ (F∕K) is the kernel of inv ⁡ ~K and BK = ⋃ ⁡ F∈EH2(G,ℂF ). In particular, we have an induced isomorphism inv ⁡ K: BK →ℚ∕ℤ.

Since BL∕K injects into BK≅ℚ∕ℤ with image 1|G|ℤ∕ℤ, we see that BL∕K has order [L : K]. On the other hand, BL∕K divides the order of H2(G,ℂL), which divides [L : K] by Theorem 10.8.8. So BL∕K = H2(G,ℂL) is mapped isomorphically to 1|G|ℤ∕ℤ by inv ⁡ L∕K, as asserted. □

We have now constructed our global class formation.

Theorem 10.9.11.

Given a number field K, the pair (ℂKsep,inv ⁡ ), where inv ⁡ is the collection of invariant maps

inv ⁡ E: H2(G E,ℂKsep) →ℚ∕ℤ

for finite separable extensions E of K in Ksep forms a class formation for K. Moreover, the reciprocity map defined by this class formation is the map ϕK: ℂK → GKab induced by the product ΦK of local reciprocity maps on the idèles.

Proof.

That we have a global class formation is an immediate corollary of Proposition 10.9.9 and Lemma 10.9.10. Let ϕL∕K be the resulting reciprocity map. We know by Proposition 8.1.10 that

inv ⁡ L∕K(a¯∪δ(χ)) = χ(ϕL∕K(a))

for all χ ∈ Hom ⁡ (G,ℚ∕ℤ) and a ∈ℂK. On the other hand, we showed in Lemma 10.9.1 that

inv ⁡ ~L∕K(α¯∪𝛿𝜒) = χ(ΦL∕K(α)),

for all such χ and α ∈𝕀K. Since ϕL∕K and ΦL∕K are completely determined by their compositions with characters of G, we must have that ΦL∕K factors through ℂK and induces ϕL∕K on the quotient. □

10.10. Power reciprocity laws

In this section, we consider higher reciprocity laws that generalize quadratic reciprocity. For this, we introduce the notion of an nth power residue symbol for a number field.

Notation 10.10.1.

In this section, n will denote a positive integer, and K will denote a number field that contains the full group μn of nth roots of unity.

Definition 10.10.2.

The nth power residue symbol for K is a function with values

(a 𝔟 )n,K ∈ μn

defined on pairs (a,𝔟) consisting of a nonzero element a of 𝒪K and a nonzero ideal 𝔟 of 𝒪K such that 𝔟 is relatively prime to (𝑛𝑎) defined as follows. For a prime 𝔭 not dividing (𝑛𝑎), its value is the unique nth root of unity in K× satisfying the congruence

(a 𝔭 )n,K ≡ a(𝑁𝔭−1)∕nmod𝔭,

and for an arbitrary 𝔟 with prime factorization 𝔟 = 𝔭1r1⋯𝔭krk, its value is given by

(a 𝔟 )n,K = ∏i=1k ( a 𝔭i )n,Kri.

Notation 10.10.3.

If a and b are nonzero elements of 𝒪K such that (b) is relatively prime to (𝑛𝑎), then we set

(a b )n,K = ( a (b) )n,K.

Remark 10.10.4.

The 2nd power residue symbol for ℚ is none other than the Jacobi symbol, since N(p) = p for a prime (p) of ℤ.

We will derive an nth power reciprocity law for the power residue symbols, generalizing quadratic reciprocity. To begin with, we have the following.

Lemma 10.10.5.

Suppose that 𝔭 is a nonzero prime of K not dividing n and that a ∈𝒪K with v𝔭(a) = 0. Let π𝔭 be a unifomizer of K𝔭. Then

(a 𝔭 )n,K = (a,π𝔭)n,K𝔭.
Proof.

Since 𝑁𝔭 is the order of the residue field of K𝔭 and a is a unit in the valuation ring of K𝔭, that the two sides are equal are an immediate consequence of our formula for the tame symbol. □

Corollary 10.10.6.

Let a,b ∈𝒪K be nonzero, and suppose that (b) is relatively prime to (𝑛𝑎). Then

(a b )n,K = ∏𝔭∣(b)(a,b)n,K𝔭,

where the product is over nonzero primes of 𝒪K dividing (b).

Proof.

Write (b) = 𝔭1r1⋯𝔭krk for distinct primes 𝔭1,…,𝔭k and positive integers ri for some k ≥ 0. Letting π𝔭i denote a uniformizer for K𝔭i, we have by Lemma 10.10.5 that

(a b )n,K = ∏i=1k ( a 𝔭i )n,Kri = ∏ i=1k(a,π𝔭 iri) n,K𝔭i.

Since b is π𝔭iri times a unit in the valuation ring of K𝔭i, we have (for instance by the formula for the tame symbol) that

(a,π𝔭iri) n,K𝔭i = (a,b)n,K𝔭i

for each i, and the result follows. □

Note also that global reciprocity gives us the following product formula for norm residue symbols.

Lemma 10.10.7.

For every a,b ∈ K×, we have

∏v∈VK(a,b)n,Kv = 1.
Proof.

We have ρKv(a1∕n)∕Kv(b) = 1 outside of a finite set S = {v1,…,vt} of places of K, so the product is finite, and global reciprocity then says that

∏v∈SρKv(a1∕n)∕Kv(b) = 1.

It follows that

∏v∈VK(a,b)n,Kv = ∏v∈S(a,b)n,Kv = ∏v∈SρKv(b)(a1∕n) a1∕n = ∏i=1tρ Kv1(b)⋯ρKvi−1(b)(ρKv i(b)(a1∕n) a1∕n ) = (∏v∈SρKv(a1∕n)∕Kv(b))(a1∕n) a1∕n = 1.

□

We are now in a position to prove the nth power reciprocity law for K.

Theorem 10.10.8 (Higher reciprocity law).

Let K be a number field containing the group μn of nth roots of unity. Let a,b ∈𝒪K elements relatively prime to each other and to n. We then have

(a b )n,K (b a )n,K−1 = ∏ v∣𝑛∞(b,a)n,Kv,

where the product is over the places of K extending a prime dividing n or the real infinite place of ℚ. Moreover, if c ∈𝒪K is relatively prime to a and divisible only by primes dividing n, then

( c b )n,K = ∏v∣𝑛∞(b,c)n,Kv.
Proof.

Corollary 10.10.6 tells us that

(a b )n,K (b a )n,K−1 = ∏ 𝔭∣(b)(a,b)n,K𝔭 ⋅∏𝔭∣(a)(b,a)n,K𝔭−1 = ∏ 𝔭∣(𝑎𝑏)(a,b)n,K𝔭.

Note that (a,b)n,K𝔭 = 1 unless the prime 𝔭 of 𝒪K divides one of a, b, or n, since otherwise the extension K𝔭(a1∕n)∕K𝔭 is unramified and b is a unit in K𝔭. Applying Lemma 10.10.7, we then have

∏𝔭∣(𝑎𝑏)(a,b)n,K𝔭 = (∏v∣𝑛∞(a,b)n,Kv)−1 = ∏ v∣𝑛∞(b,a)n,Kv,

finishing the proof in this case. In the remaining case, the same argument, but now noting that (a,b)n,K𝔭 = 1 unless 𝔭 divides b or n, we have

( c b )n,K = ∏𝔭∣(b)(c,b)n,K𝔭 = ∏v∣𝑛∞(b,c)n,K𝔭.

□

Example 10.10.9.

Take the case that K = ℚ and n = 2. Let a and b be positive, odd integers. Then Theorem 10.10.8 implies that

(a b ) (b a ) = (b,a)2,ℚp(b,a)2,ℝ.

The first symbol is (−1)(a−1)(b−1)∕4 by Proposition 9.3.6, and the second symbol is trivial by Remark 9.3.9. Similarly, we have

( −1 b ) = (−1)(b−1)∕2 and (2 b ) = (−1)(p2−1)∕8.

Thus, the power reciprocity law for K = ℚ and n = 2 is simply quadratic reciprocity.

The following special case of a result of the author serves as an entertaining example of the use of higher reciprocity laws.

Proposition 10.10.10.

Let p be an odd prime number, and let m be an integer. Suppose that ℓ = Φp(𝑝𝑚) is a prime number. Then every divisor of m is a pth power residue modulo ℓ.

Proof.

Let a be a divisor of m. Note that

ℓ = Nℚ(μp)∕ℚ(1−𝑝𝑚ζp)

and can only be prime if m is nonzero. The definition of the pth power residue symbol says that

( a 1−𝑝𝑚ζp )p,ℚ(μp) ≡ a(ℓ−1)∕pmod(1−𝑝𝑚ζ p), (10.10.1)

Since a is an integer, this implies that the symbol in (10.10.1) is the unique pth root of unity congruent to a(ℓ−1)∕p modulo ℓ. Thus, it will be trivial if and only if a is a pth power residue modulo ℓ.

So, we compute the symbol. We have

( a 1−𝑝𝑚ζp )p,ℚ(μp) = (1−𝑝𝑚ζp a )p,ℚ(μp)(1−𝑝𝑚ζp,a)p,ℚp(μp) = (1−𝑝𝑚ζp,a)p,ℚp(μp).

If p divides m, then 1−𝑝𝑚ζp is a pth power in ℚp(μp), and we are done. If a is a pth power in ℤp×, we are done as well. So, we may assume that a is not a pth power in ℤp×. Then ℚp(ζp,a1∕p) = ℚp(ζp,(1−p)1∕p), and the conductor of the latter extension of ℚp(μp) is (1−ζp)2 by Proposition 9.6.16. Since 1−𝑝𝑚ζp ∈ Up−1(ℚp(μp)) and p−1 ≥ 2, we then have that (1−𝑝𝑚ζp,a)p,ℚp(μp) = 1, which completes the proof. □

Find in the notes