Patrick Lei

All notes

Class field theory

Lectures by Chao Li, Columbia, Spring 2021.

Algebraic Number Theory
Spring 2021
Notes by Patrick Lei
Lectures by Chao Li

Columbia University

Disclaimer

These notes were taken during lecture using the vimtex package of the editor neovim. Any errors are mine and not the instructor’s. In addition, my notes are picture-free (but will include commutative diagrams) and are a mix of my mathematical style and that of the instructor. If you find any errors, please contact me at plei@math.columbia.edu.

1 Motivation

Here is a very classical question (that the ancients were interested in):

Question 1.0.1.

Which prime numbers p can be written as p=x2+y2 for integers x,y?

We can try to answer this by experiment. Clearly, 2=1+1,5=1+4,13=4+9,17=1+16 and the other primes below 20 cannot be written as a sum of two squares. Then, because any square is congruent to 0 or 1 modulo 4, we see that if p is an odd prime, then

Theorem 1.0.2 (Fermat, Christmas Day, 1640).

An odd prime p can be written as a sum of two squares if and only if p≡1(mod4).1 Similarly, we have:

  • •

    2≠p=x2+2⁢y2 if and only if p≡1,3(mod8).

  • •

    3≠p=x2+3⁢y2 if and only if p≡1(mod3).

In the modern day, we should reinterpret p=x2+y2 as a factorization problem in the number field k=ℚ⁡(i). Now we write our problem as p=(x+i⁢y)⁢(x−i⁢y). Similarly, the second problem can be written as p=(x+−2⁢y)⁢(x−−2⁢y) in ℚ⁡(−2) and the third problem can be expressed in the field ℚ⁡(−3). More generally, we can consider k=ℚ⁡(d) for an arbitrary d, called the discriminant of k. For a general quadratic extension, the ring of integers is not a UFD, but it is Dedekind, so we have unique factorization of prime ideals. Therefore we can write

What we want to know is when the ideal (p) splits, and this behavior is governed by the Legendre symbol. This symbol satisfies the miraculous identity (due to Gauss)

Theorem 1.0.3 (Quadratic Reciprocity).

Let p,q be odd primes. Then we have the identity

Recall that if p is odd and p∤d, then (dp)=1 if and only if x2≡d(modp) has solutions. Quadratic reciprocity tells us that the equation x2≡q(modp) is highly related to the equation x2≡p(modq). This ability to change the modulus is very helpful in solving these classical problems.

Example 1.0.4.

A prime p splits in ℚ⁡(−3) if and only if (−3p)=1 if and only if (p3)=1 if and only if p≡1(mod3).

Now we can generalize our question about splitting in quadratic fields to more general fields:

Question 1.0.5.

Is there a criterion of the form p splits in k if and only if p≡∗(modN) for some N? If so, we can generalize quadratic reciprocity. This is one of the main questions of class field theory.

Example 1.0.6.

We have some examples of splitting behavior:

  • •

    p splits in k=ℚ⁡(−5) if and only if p≡1,3,7,9(mod2)0.

  • •

    p splits in k=ℚ⁡(−5,i) if and only if p≡1,9(mod2)0.

  • •

    p splits in k=ℚ⁡(ζ5) if and only if p≡1(mod5).

However, there is no congruence condition for splitting in k=ℚ⁡(23) for any modulus N. The question is what is different about the last example. First, the fields ℚ⁡(−5),ℚ⁡(−5,i),ℚ⁡(ζ5) are all Galois extensions of ℚ with Galois groups ℤ/2,ℤ/2×ℤ/2,ℤ/4. On the other hand, k=ℚ⁡(23) is not Galois and its Galois closure ℚ⁡(23,ζ3) has Galois group S3.

Definition 1.0.7.

A field extension L/K is called abelian if L/K is Galois and Gal⁡(L/K) is abelian.

This gives us the following slogan: given a number field k, class field theory

  1. 1.

    classifies all abelian extensions L/K in an accessible way;

  2. 2.

    describes factorization of primes of K in L in terms of groups intrinsic to K (for example the class group of K).

1.1 A Special Case of CFT

Here, we will classify all unramified abelian extensions L/K. Recall that the class group Cl⁡(K) of a number field K is

This is always a finite abelian group.

Definition 1.1.1.

Let H⊂Cl⁡(K) be a subgroup. Then a finite unramified abelian extension L/K is a class field for a subgroup H⊂Cl⁡(K) if 𝔭 splits in L/K if and only if [𝔭]∈H⊂Cl⁡(K).

Theorem 1.1.2 (Unramified CFT).

Given H⊂Cl⁡(K), the class field for H exists and is unique. Moreover, each finite unramified abelian extension arises as a class field. This gives us a bijection

Moreover, Gal⁡(L/K)≅Cl⁡(K)/H.

Definition 1.1.3.

Note that the class field for H=0 is the maximal unramified abelian extension HK of K, called the Hilbert class field. This gives a canonical isomorphism Gal⁡(HK/K)≅Cl⁡(K). Also, we see that 𝔭 splits in HK if and only if 𝔭 is a principal ideal.

Example 1.1.4.

For the fields K=ℚ,ℚ⁡(i),ℚ⁡(−2),ℚ⁡(−3), we have Cl⁡(K)=0 and thus HK=K.

Example 1.1.5.

The simplest example of a number field with nontrivial class group is K=ℚ⁡(−5). Here, Cl⁡(K)=ℤ/2 and HK=ℚ⁡(−5,i).

Remark 1.1.6.

More generally, class field theory will add ramification on both sides of the correspondence to obtain a correspondence

1.2 Back to Fermat

Consider the equation p=x2+5⁢y2. Recall that we have 2,5≠p=x2+5⁢y2 if and only if p≡1,9(mod2)0. Note that this is different from the splitting behavior in K=ℚ⁡(−5). This happens because Cl⁡(ℚ⁡(−5))=ℤ/2 is not trivial, and so we need both the condition that (p)=𝔭1⁢𝔭2 and that the 𝔭i are principal. By unramified CFT, we know that 𝔭i are principal if and only if they split in HK=ℚ⁡(−5,i), and therefore p=x2+5⁢y2 if and only if p splits in ℚ⁡(−5,i). In this case, HK/ℚ is abelian, so we have a nice answer.

Example 1.2.1.

The primes of the form p=x2+14⁢y2 cannot be described in terms of a congruence condition. The field K=ℚ⁡(−14) has Cl⁡(K)=ℤ/4 and Gal⁡(HK/ℚ)≅D4 is non-abelian.

Remark 1.2.2.

We can study non-abelian extensions to get some nice answers that involve modular forms, and this is called the Langlands program, which is beyond the scope of this course.

Our outline for the semester is to prove local CFT, then prove global CFT, then do applications if time permits. This will be done using group cohomology.

2 Local Fields

Recall that it is very difficult to detect whether a polynomial equation over a global field like ℚ has solutions. However, we can embed ℚ into the local field ℝ and then checking whether the polynomial has real solutions is very easy because we can do analysis. To try to recover all information about ℚ, we can embed ℚ↪ℚp for a prime p. We then have the following slogan, known as the local-to-global principle:

We will study problems in ℚ by studying problems in all the local fields ℝ and ℚp.

2.1 Absolute Values

Definition 2.1.1.

Let K be a field. An absolute value on K is a function |−|:K→ℝ such that

  1. 1.

    |−| sends K× to ℝ>0 and 0 to 0.

  2. 2.

    We have |x⁢y|=|x|⋅|y| for all x,y∈K.

  3. 3.

    For all x,y∈K, we have |x+y|≤|x|+|y|.

Example 2.1.2.

The usual absolute value on ℝ defines an absolute value in this semse. This induces an absolute value on ℚ⊆ℝ usually denoted by |−|∞. This is known as the archimedean absolute value on ℚ.

Similarly, any embedding K↪𝜎ℝ or K↪𝜎ℂ induces an absolute value on K defined by |x|σ≔|σ⁡(x)|.

There is a different kind of absolute value that is not archimedian. Here, we will strengthen the triangle inequality.

Definition 2.1.3.

If |−| satisfies the ultrametric inequality

then we say |−| is nonarchimedean.

Remark 2.1.4.

Recall that ℝ satisfies the archimedean property: If 0≠x∈ℝ there exists n∈ℤ such that |n⁢x|>1. This property fails for nonarchimedean absolute values because |n⁢x|≤|x| for all n∈ℤ. In fact, |−| is nonarchimedean if and only if the set {|n⋅1|}n∈ℤ is bounded.

Example 2.1.5.

Let a∈ℚ× and p be a prime. Then define ordp⁢(a)∈ℤ such that

Now for any c<1, we define

Then we simply need to check that |−|p is a nonarchimedean absolute value on ℚ. Here, it is easy to check the ultrametric inequality, and this absolute value is called the p-adic absolute value. By convention, we will choose c=p−1 and this is the normalized p-adic absolute value on ℚ.

Example 2.1.6.

For any number field K and prime ideal 𝔭⊆𝒪K. Then we have a normalized 𝔭-adic absolute value

where N⁢𝔭=#⁢𝒪K/𝔭.

Definition 2.1.7.

An absolute value |−| is discrete if |K×|⊂ℝ is discrete under the usual topology on ℝ.

Example 2.1.8.

For a number field K and prime 𝔭, the p-adic absolute value |−|𝔭 is discrete. On the other hand, |−|∞ is not discrete.

Definition 2.1.9.

Suppose |−| be nonarchimedean. Then define

  1. 1.

    A≔{a∈K∣|a|≤1}. This is a subring of K.

  2. 2.

    Now define A×={a∈K∣|a|=1}. This is a subgroup of A of invertible elements.

  3. 3.

    Set 𝔪={a∈K∣|a|<1}. This is the unique maximal ideal of A.

Then |−| is discrete if and only if 𝔪 is principal. In this case, a generator π of 𝔪 is called a uniformizer. Then every a∈K can be uniquely written as a=πr⋅u for some r∈ℤ,u∈A×.

Example 2.1.10.

(Non-example) Consider the field ℚ⁡({p1/n},n∈ℤ) with p-adic absolute value. This is not a discrete absolute value.

Definition 2.1.11.

An absolute value defines a metric on K by d⁡(a,b)=|a−b| for all a,b∈K. This induces a topology on K where a basis of open neighborhoods of a∈K is given by open balls

Example 2.1.12.

In the p-adic topology, we see that a,b∈ℚ are closer under |−|p if and only if |a−b|p is smaller, which is equivalent to ordp⁢(a−b) being larger, which is equivalent to a−b being divisible by a large power of p. In other words, a≡b(modp)N for N large.

Definition 2.1.13.

We say two absolute values on K are equivalent, or |−|⁢sin⁡|−|′ if they induce the same topology on K.

Theorem 2.1.14 (Ostrowski, 1916).

Let |−| be an absolute value on ℚ.

  1. 1.

    If |−| is archimedean, then |−|∼|−|∞.

  2. 2.

    If |−| is nonarchimedean, then |−|∼|−|p for a unique p.

Remark 2.1.15.

Similarly, absolute values on a number field K are given by

  1. 1.

    |−|𝔭 for a prime ideal 𝔭⊆𝒪K (𝔭-adic place);

  2. 2.

    |−|σ for some σ:K↪ℝ (real place);

  3. 3.

    |−|σ for some complex embeddings σ:K↪ℂ (complex place). Here, note that complex embeddings come in conjugate pairs.

Definition 2.1.16.

An equivalence class of absolute values on K is called a place (or prime) of K.

Remark 2.1.17.

When v is a complex place it corresponds to a pair of complex embeddings σ,σ¯, so we define

and this is the normalized absolute value for v.

One reason for this normalization is

Theorem 2.1.18 (Product Formula).

Let K be a number field. Then for all a∈K×, we have

Remark 2.1.19.

When K=ℚ, let a=mn for m,n∈ℤ. Then all but finitely many terms in the product are finite. Now it suffices to check this formula for a=p and a=−1.

When a=p, we see that |a|p=p−1 and for primes ℓ≠p, we have |a|ℓ=1. Finally, we see that |a|∞=p, so the formula holds. When a=−1, all absolute values are 1, so the product of all absolute values is trivial.

For a general number field, we can simply take the norm map NK/ℚ to ℚ and check that it behaves well with respect to the places.

Theorem 2.1.20 (Weak Approximation).

Let |−|1,…,|−|n be inequivalent absolute values on a field K. Let a1,…,an∈K. Then for all ε>0, there exists a∈K such that |a−ai|i<ε for all i=1,…,n.

Remark 2.1.21.

This allows us to approximate any finite collection ai∈K for inequivalent |−|i.

Remark 2.1.22.

As a sanity check, consider K=ℚ and suppose |−|i=|−|pi. Then given a1,…,an, we simply find a∈ℚ such that ai≡a(modp)iN, which is possible by the Chinese remainder theorem.

Remark 2.1.23.

More generally, if |−|1≁|−|2 then one can choose a∈K such that |a|1>1 and |a|2<1. Then if we consider ar1+ar as r→∞, the absolute value under |−|1 approaches 1 and under |−|2 it approaches 0.

2.2 Completions

Consider the field ℚ equipped with the the absolute value |−|∞. Then we can complete ℚ as a metric space to obtain the field ℝ. More generally, if (K,|−|) is a field equipped with an absolute value (a valued field), then we will produce a general completion K̂. Our aim is to produce a field that contains the original field and whose arithmetic is easier to understand.

Definition 2.2.1.

Let (K,|−|) be a valued field. Then a sequence {an} of elements in K is called Cauchy if for all ε>0, there exists N≥1 such that |an−am|<ε for all n,m>N. We say that K is complete if any Cauchy sequence in K has a limit in K.

Example 2.2.2.

Consider the sequence of integers {an=2n}=2,4,8,16,32,…. Clearly, this is not Cauchy under the usual absolute value on ℚ, but then if m>n, we see that

so {an} is Cauchy in (ℚ,|−|2). We then see that |an−0|2=(12)n→0 and thus the limit of the sequence is 0.

Example 2.2.3.

Consider {an}={4,34,334,3334,…}. Then if m>n, we have

and therefore {an} is Cauchy in (ℚ,|−|2). We then see that

and therefore an→23.

Remark 2.2.4.

In general, the limit of a Cauchy sequence may not exist.

Theorem 2.2.5.

Let (K,|−|) be a valued field. Then there exists a complete valued field (K̂,|−|) and an embedding K↪K̂ of valued fields such that any other embedding K↪L of K into a complete valued field factors uniquely through K̂. In particular, K̂ is unique up to isomorphism and is called the completion of (K,|−|).

Proof.

Let K̂ be the set of all Cauchy sequences in K under the equivalence relation where {an}∼{bn} where limn→∞|an−bn|=0. Then K̂ is equipped with termwise addition and multiplication and absolute value |{an}|=limn→∞|an|. Thus K̂ is a complete valued field.

To verify the universal property, we see that x↦(x,x,…,x) embeds K↪K̂ and this satisfies the desired universal property. □

Definition 2.2.6.

Let K be a number field and v a place of K. Denote by Kv≔(K,|−|v). When v is a finite place, denote by 𝒪Kv=𝒪K,v=𝒪v the valuation ring {x∈Kv:|x|≤1}⊆Kv. When v is an infinite place, we see that Kv≅ℝ,ℂ.

Example 2.2.7.

Let K=ℤ. Then we see that ℚ∞=ℝ and ℚp has a subring ℤp, which is a discrete valuation ring. Here, elements of ℤp have a nonnegative lowest power of p.

Example 2.2.8.

If K=ℚ⁡(i), then K∞=ℂ and K𝔭 for 𝔭⊂𝒪K=ℤ⁡[i] prime ideals are the completions of K.

Remark 2.2.9.

Let K be a nonarchimedean discrete valued field. Then K̂ is a complete discrete valued field and the valuation ring Â⊂K̂ is the closure of A⊆K in K̂. Also, the maximal ideal 𝔪̂⊆Â is the closure of 𝔪⊆A in K̂. Finally, a uniformizer π of K is also a uniformizer of K̂.

We also see that the natural map A/𝔪n→Â/𝔪̂n is an isomorphism. This tells us that we can approximate elements in  up to πn⁢A using elements in A.

Proposition 2.2.10.

Let K be a discrete valued field. Let S be a complete set of representatives of A/𝔪 and π be a uniformizer of K. Then any element of K̂ can be uniquely written as ak⁢πk+ak+1⁢πk+1+⋯ where ai∈S and k∈ℤ.

Corollary 2.2.11.

Let x∈ℚp. Then x has a p-adic expansion x=∑i≥kai⁢pi, where ai∈S=ℤ/p⁢ℤ={0,1,…,p−1} and k∈ℤ.

Remark 2.2.12.

The main term of the p-adic expansion is the lowest term pk. This is completely unlike the situation with the decimal digits of x∈ℝ, where the highest power of 10 is the main term.

Remark 2.2.13.

ℚp resembles 𝔽p⁢((T)) but is more complicated arithmetically. When we add two power series, we simply add the coefficients, but addition in ℚp requires carrying. In addition, we see that 𝔽p⁢[[T]]/Tn↪𝔽p⁢[[T]] but ℤp/pn does not embed in ℤp.

Corollary 2.2.14.

We have an isomorphism

For example, we have

Proof of Proposition 2.2.10.

Let x∈K̂×. Then x=πk⋅y for some k∈ℤ and y∈Â×. Then we find the first digit a0 by considering y≡a0 in A/𝔪=S. Then we replace π−1⁢(y−a0)=a1+a2⁢π2+⋯ and repeat the process. Repeating this, we simply use the completeness of K̂ to obtain the desired expansion. □

The advantage of completeness is that it is much easier to solve equations. Here, we take solutions modulo a high power of p and then take the limit.

Theorem 2.2.15 (Hensel’s Lemma).

Let K be a complete discrete valued field and k=A/𝔪 be the residue field. Now let f⁡(x)∈A⁡[x] be a monic polynomial and let f¯⁢(x)≔f⁡(x)mod𝔪∈k⁡[x]. Assume that f¯⁢(x)=g0⁢(x)⁢h0⁢(x) in k⁡[x] where g0,h0 are monic and coprime. Then there exist unique g,h∈A⁡[x] such that f⁡(x)=g⁡(x)⁢h⁡(x) and g¯≡g0,h¯=h0.

Corollary 2.2.16.

Suppose f¯ has a simple root α0∈k. Then f⁡(x) has a unique zero α∈A such that α¯=α0.

Corollary 2.2.17.

If k=𝔽q for q=pt, then f⁡(x)=xq−x has q distinct roots in k=𝔽q and hence q distinct roots in K. In particular, K× contains all (q−1)-th roots of unity, so we have a map 𝔽q×↪K×, called the Teichmuller lift.

Proof of Hensel’s lemma.

Let g0,g0∈A⁡[x] be arbitrary monic lifts. Then f−g0⁢h0∈π⁢A⁢[x]. Now inductiely we assume that there exist gn,hn∈A⁡[x] monic such that f−gn⁢hn∈πn+1⁢A⁢[n]. We simply write gn+1=gn+πn+1⁢u for some u∈A⁡[x] such that deg⁡u<deg⁡gn and hn+1=hn+πn+1⁢v where deg⁡v<deg⁡hn. Then f−gn+1⁢hn+1∈πn+2⁢A if and only if

but we can find such u,v by Bezout’s lemma. Thus the desired gn+1,hn+1 exist, and we obtain g,h by taking the limit. □

2.3 Extension of Absolute Values and Unramified Extensions

Let K be a complete discrete valued field and L be a finite separable extension of K. Suppose [L:K]=n. The main result is

Theorem 2.3.1.

Let K,L be as above. Then

  1. 1.

    |−|K extends uniquely to a discrete absolute value |−|L on L;

  2. 2.

    L is complete with respect to |−|L;

  3. 3.

    For all β∈L, we have

Remark 2.3.2.

To perform a sanity check, if β∈K, we have |β|L=|NL/K⁢(β)|K1/n=|βn|K1/n=|β|K.

Proof.
  1. 1.

    We first need to prove that a unique extension exists. We know that |−|K comes from a discrete valuation (hence is nonarchimedean), so let A⊆K be the valuation ring. Then A is a Dedekind domain. Then let B be the integral closure of A in L. Then B is also a Dedekind domain. But then any absolute value on L extending |−|K comes from a maximal iddeal of B lying above the unique maximal ideal 𝔭⊆A. Therefore, we need to show that B is a local ring.

    To see this, assume not. Suppose there exist two prime ideals 𝔓1,𝔓2⊆B lying above 𝔭. Let β∈𝔓 but β∉𝔓2. This implies that A⁡[β]∩𝔓1≠A⁡[β]∩𝔓2. Let f⁡(x)∈A⁡[x] be the minimal polynomial of β. Then f¯⁢(x)=f⁡(x)mod𝔭∈A/𝔭⁡[x]=k⁡[x] must satisfy f¯⁢(x)=h⁢(x)m for an irreducible h⁡(x)∈k⁡[x] (otherwise it has two distinct irreducible factors and Hensel tells us that the factorization can be lifted to A⁡[x]). This implies that

    has a unique prime ideal, generated by h⁡(x), which contradicts our original assumption that A⁡[β]∩𝔓1,A⁡[β]∩𝔓2 were distinct prime ideals.

  2. 2.

    Now we show that L is complete. Let {ak} be a Cauchy sequence in L. Choose a K-basis {e1,…,en} of L and write

    But then each sequence {ai,k}k forms a Cauchy sequence in K. By completeness of K, we can take ai≔limk→∞ai.k∈K and so we have

    and thus L is complete.

  3. 3.

    Let L~ be the Galois closure of L/K. Then we know that |−|K also extends uniquely to L~. For any σ∈Gal⁡(L~/K), the map L∈β↦|σ⁡(β)|L~ is also an absolute value on L extending |−|K. Therefore, by the uniqueness of the extension, we see that |β|L=|σ⁡(β)|L~. This implies that

    as desired. □

Corollary 2.3.3.

If L/K is merely an algebraic and separable extension, then |−|K also extends uniquely to an absolute value on L, but |−|L may fail to be discrete or complete.

Proof.

Note that L is the union of all of its finite subextensions. □

Definition 2.3.4.

Let K be a complete discrete valued field and L be a finite separable extension. Let 𝒪K⊆K and 𝒪L⊆L be the valuation rings and 𝔭⊆𝒪K,𝔓⊆𝔒L be the maximal ideals. Next, let k=𝒪K/𝔭,ℓ=𝔒L/𝔓 be the residue fields.

Define the ramification index e⁡(L/K) to be the e≥1 such that 𝔭⁢𝒪L=𝔓e. Next, define the residual degree f⁡(L/K) to be f≥1 such that f=[ℓ:k]. Then n=e⁢f.

Definition 2.3.5.

The extension L/K is called

  1. 1.

    Unramified if e⁡(L/K)=1 (which implies f⁡(L/K)=n and 𝔭=𝔓);

  2. 2.

    Totally ramified if e⁡(L/K)=n (which implies ℓ=k and 𝔭=𝔓n).

2.4 Unramified Extensions

We will study unramified extensions. Here, we will try to understand L/K via ℓ/k.

Proposition 2.4.1.

If L/K is unramified, write ℓ=k⁡(α0),α0∈ℓ. Then for any α∈𝒪L such that α¯=α0 we have L=K⁡(α).

Proof.

Let f⁡(x)∈𝒪K⁢[x] be the minimal polynomial of α. Then deg⁡f¯=deg⁡f=[K⁡(α):K]≤[L:K]. But then we know that deg⁡f¯≥[k⁡(α0):k]=[ℓ:k]=[L:K]. But this implies that deg⁡f=[L:K], so K⁡(α)=L. □

Proposition 2.4.2.

If L=K⁡(α) with minimal polynomial of α given by f⁡(x) such that f¯⁢(x) has no repeated roots over k¯, then L/K is unramified.

Proof.

If f⁡(x) is irreducible, then by Hensel’s lemma, we have f¯⁢(x)=h⁢(x)m where h⁡(x)∈k⁡[x] is irreducible. But then because f¯⁢(x) has no repeated roots, we see that m=1. But then we see that [ℓ:k]=[L:K] and thus L/K is unramified. □

Proposition 2.4.3.
  1. 1.

    Let K⊂L⊂M be a tower of field extensions. Then M/K is unramified if and only if M/L and L/K are unramified.

  2. 2.

    Assume k is perfect. If L/K is unramified and L′/K is finite, then L⁢L′/L′ is unramified.

  3. 3.

    Assume k is perfect. Then if L/K and L′/K are unramified, then L⁢L′/K is unramified.

Proof.
  1. 1.

    Note that M/K is unramified if and only if e⁡(M/K)=1, which is equivalent to e⁡(M/L)=e⁡(L/K)=1 by multiplicativity of the ramification index.

  2. 2.

    Suppose L/K is unramified. Then let L=K⁡(α) and let f⁡(x)∈𝒪K⁢[x] be the minimal polynomial of α. Then the reduction f¯⁢(x)∈k⁡[x] is irreducible and ℓ=k⁡(α¯). Because k is perfect, f¯⁢(x) has no repeated roots.

    Because L⁢L′/L′=L′⁢(α)/L′, let g⁡(x)∈𝒪L′⁢[x] be the minimal polynomial of α. Then g¯⁢(x)∣f¯⁢(x) and thus g¯⁢(x) has no repeated roots, so L⁢L′/L′ is unramified.

  3. 3.

    Consider the tower K⊆L′⊆L⁢L′. Because L/K,L′/K are unramified, we know L⁢L′/L is unramified. This implies that L⁢L′/K is unramified. □

Theorem 2.4.4.

Assume that k is perfect. Then there is an inclusion-preserving bijection

Moreover, L/K is Galois if and only if ℓ/k is Galois and Gal⁡(L/K)≃Gal⁡(ℓ/k) in this case.

Proof.

We prove surjectivity. Let ℓ/k be a finite extension. Write ℓ=k⁡(α0) and let f¯⁢(x)=k⁡[x] be the minimal polynomial of α0. Then any monic lift f⁡(x)∈𝒪K⁢[x] of f¯⁢(x) has a root α such that α¯=α0 by Hensel’s Lemma. Then L=K⁡(α) has residue field ℓ=k⁡(α0). Because f¯ is irreducible and k is perfect, we know L/K is unramified.

Now we will prove injectivity. Let L/K,L′/K be unramified with the same residue field ℓ. Then L⁢L′/K is also unramified with residue field ℓ. But this implies that

so we must have L=L⁢L′=L′.

Now we will show the statements about Galois extensions. If L/K is Galois, then Gal⁡(L/K) preserves 𝒪L and 𝔭L⊆𝒪L and acts trivially on 𝒪K and 𝔭K⊆𝒪K. This implies that any σ∈Gal⁡(L/K) induces σ¯∈Aut⁡(ℓ/k). If L=K⁡(α) and α0=α¯, then L/K is Galois if and only if it contains α, but this is equivalent to ℓ containing all conjugates of α0, which is equivalent to ℓ/k being Galois. Then the natural map Gal⁡(L/K)→Gal⁡(ℓ/k) is an isomorphism because the permutation on conjugates of α induces the same permutation on the conjuages of α0. □

Corollary 2.4.5.

If L/K is an algebraic extension (possibly infinite), then there exists a largest unramified subextension K0/K of L/K. Moreover, L/K0 is totally ramified.

Proof.

Let K0 be the compositum of all finite unramified subextensions of L/K. Then the residue field of K0 is equal to the residue field of L (otherwise, we can create an even larger unramified extension). This implies L/K0 is totally ramified. □

Corollary 2.4.6.

Assume k=𝔽q. Then for all n≥1 there is a unique unramified extension L/K of degree n and Gal⁡(L/K)=Gal⁡(𝔽qn/𝔽q)≅ℤ/n⁢ℤ.

Proof.

There is a unique degree n extension of 𝔽q. □

Definition 2.4.7.

Define the Frobenius element σ∈Gal⁡(L/K) when k=𝔽q to be the generator of Gal⁡(L/K) corresponding to the Frobenius map under Gal⁡(L/K)→≃Gal⁡(𝔽qn/𝔽q). We will call this element FrobL/K, and FrobL/K(α)≡αq(mod)pL for all α∈𝒪L.

Corollary 2.4.8.

When k=𝔽q, the maximal unramified extension of K is

In particular, we have

Proof.

We know 𝔽qn=𝔽q⁢(ζqn−1), so 𝔽¯q is given by adjoining all coprime-to-p roots of unity. □

2.5 Totally Ramified Extensions

Definition 2.5.1.

A polynomial f⁡(x)∈K⁡[x] is Eisenstein if

such that |an|=1,|ai|<1,|a0|=|π|.

Example 2.5.2.

The polynomial xn−π is Eisenstein. Note that K⁡[x]/f⁡(x)=K⁡(πn) is totally ramified.

Proposition 2.5.3.

A finite extension L/K is totally ramified if and only if L=K⁡(α) where α is a root of an Eisenstein polynomial.

Proof.

Let α be the root of an Eisenstein polynomial. Then |αn|=∏σ:L↪L~|σ⁡(α)|=|a0|=|π|. Therefore e⁡(L/K)≥n and thus L/K is totally ramified.

Now suppose L/K is totally ramified. Let α be a uniformizer of L. Therefore (αn)=(π) and thus |α|L=|π|L1/n. Then 1,α,…,αn−1 have absolute values representing different cosets in |L×|/|K×|. Thus the minimal polynomial of α has degree n. Moreover, if we write the minimal polynomial as

then |α|n=|a0|=|π| and |ai|<1 so that the required cancellation happens. But this implies that α is a root of an Eisenstein polynomial. □

Proposition 2.5.4.

Assume k=𝔽q. Then there are only finitely many totally ramified extensions of K.

Proof.

Recall Krasner’s lemma: Let f⁡(x)=∑i=0nai⁢xi,g⁡(x)=∑i=0nbi⁢xi∈K⁡[x] and assume |ai−bi| is sufficiently small for all i. If f⁡(x) is irreducible, then so is g⁡(x), and

Therefore a totally ramified extension depends only on a small neighborhood of (a0,…,an−1) in the set

which is compact, so it can be covered by finitely many such small neighborhoods. □

Recall that if K is a complete discrete valued field with residue field k=𝔽q, there exists a unique unramified extension L/K of degree n. Together with the proposition, there exist finitely many totally unramified extensions L/K of degree n. This is of course false for number fields; for example, ℚ has infinitely many quadratic extensions.

Remark 2.5.5.

Krasner (1966) gave an explicit formula for the number of extensions of p-adic fields of degree n and an algorithm to construct the set of generating polynomials of degree n. More desirable is a way to organize all these extensions, and local class field theory achieves this for abelian extensions of all local fields.

Definition 2.5.6.

A local field is a valued field K that is locally compact under the topology induced by the absolute value.

Remark 2.5.7.

Recall that

  1. 1.

    A topological space is compact if and only if open cover has a finite subcover.

  2. 2.

    A topological space is locally compact if every point has compact neighborhood.

  3. 3.

    A metric space is compact if and only if it is complete and totally bounded.

  4. 4.

    A metric space is compact if and only if all closed balls are compact.

  5. 5.

    This tells us that local fields are always complete. To find a limit for a Cauchy sequence, everything is contained in a closed ball, which is complete and thus has a limit.

Example 2.5.8.
  1. 1.

    The easiest examples of local fields are ℝ,ℂ.

  2. 2.

    If K is archimedean and complete, then K≃ℝ or ℂ.

Lemma 2.5.9.

Let K be a complete discrete valued field. Then K is locally compact if and only if the residue field k is finite.

Proof.

Let K be locally compact. Then 𝒪K={x∈K∣|x|≤1} is a closed ball. This means that 𝒪K is compact. If we consider an open cover

this has a finite subcover. But all of the x+𝔭K are disjoint, so k is finite.

Now suppose k is finite. We show that every x∈K has a compact neighborhood. In particular, we will show that x+𝒪K is compact and therefore that 𝒪K is compact. To do this, we need to show that 𝒪K is totally bounded. Choose r>0 and consider the open balls Ba,r give a cover of 𝒪K as long as a∈𝒪K/𝔭Kn and n is sufficiently large. By finiteness of k, we know that 𝒪K/𝔭Kn is finite, as desired. □

Theorem 2.5.10.

Every local field is one of the following:

  1. 1.

    ℝ or ℂ;

  2. 2.

    A finite extension of ℚp;

  3. 3.

    𝔽q⁢((t)) for a prime power q.

Proof.

Suppose char⁡K=0. Then ℚ⊆K. If K is archimedean, then K=ℝ or ℂ. Otherwise, ℚp⊆K and by local compactness, K/ℚp must be finite.

If char⁡K=0, then 𝔽p⊆K. Let k=𝔽q be the residue field. Then

By Hensel’s lemma, we have 𝔽q×↪K× and we thus have 𝔽q⊆K. Therefore K≅𝔽q⁢((π)), as desired. □

2.6 Statement of Local Class Field Theory

Recall that a field extension L/K is abelian if it is Galois and Gal⁡(L/K) is abelian.

Exercise 2.6.1.

If L1/K,L2/K are abelian, then L1⁢L2/K is also abelian.

Define Kab to be the maximal abelian extension of K. Equivalently, this is the compositum of all finite extensions of K. Then Kab/K has infinite degree, and classifying abelian extensions of K is equivalent to understanding Gal⁡(Kab/K).

Definition 2.6.2.

Let Ω/K be a possible infinite extension. We call Ω/K Galois if it is algebraic, separable, and normal. Equivalently, Ω is the union of all its finite Galois subextensions. In particular, we have

is an inverse limit of finite groups, known as a profinite group. Then Gal⁡(Ω/K) has a profinite topology with a basis of open neighborhoods of 1 given by Gal⁡(Ω/L)⊆Gal⁡(Ω/K) for finite subextensions L/K.

Example 2.6.3.

Consider Ω/K=F¯q/𝔽q. Then

The open neighborhoods of 1 are given by Gal⁡(F¯q/𝔽qn)≅n⁢ℤ̂⊆ℤ̂.

Remark 2.6.4.

If L/K is finite, then Gal⁡(Ω/L) is an open subgroup. In addition, Gal⁡(Ω/L) is also closed, so it is clopen.

If L/K is any extension, then

is a closed subgroup.

Theorem 2.6.5 (Galois correspondence).

Let Ω/K be Galois. Then there is a Galois correspondence

Moreover, L/K is Galois if and only if the corresponding closed subgroup H⊆Gal⁡(Ω/K) is normal.

Remark 2.6.6.

If H⊆Gal⁡(Ω/K) is not necessarily closed (for example, ℤ⊂ℤ̂ is not closed and its closure is ℤ̂), then ΩH corresponds to H¯ under the Galois correspondence. In particular, Gal⁡(Kab/K)=G/[G,G]¯≕Gab, where G=Gal⁡(K¯/K).

This gives us the slogan, that when K is a local field, Gal⁡(Kab/K) can be understood in terms of K×.

Theorem 2.6.7 (Local Artin reciprocity).

There exists a unique ϕK:K×→Gal⁡(Kab/K) (local Artin reciprocity map) such that

  1. 1.

    For any finite abelian L/K, there is a commutative diagram

  2. 2.

    For any finite unramified L/K and uniformizer π of K, we have ϕL/K⁢(π)=FrobL/K∈Gal⁡(L/K).

2.7 Norm Subgroups

Let K be a nonarchimedean local field.

Definition 2.7.1.

A subgroup of K× is a norm subgroup if it is of the form N⁡(L×) for some finite abelian extension L/K.

Proposition 2.7.2.

Assume local CFT I.

  1. 1.

    N⁡(L1×)∩N⁡(L2×)=N⁡((L1⁢L2)×).

  2. 2.

    N⁡(L1×)⊆N⁡(L2×) if and only if L1⊇L2.

  3. 3.

    A subgroup of K× containing a norm subgroup is also a norm subgroup.

  4. 4.

    N⁡(L1×)⁢N⁡(L2×)=N⁡((L1∩L2)×).

Proof.

Recall that there exists a unique local Artin reciprocity map K×→ϕKGal⁡(Kab/K).

  1. 1.

    Note that if K⊆L2⊆L1, then N⁡(L1×)⊆N⁡(L2×), so clearly for L1,L2, we see that N⁡((L1⁢L2)×)⊆N⁡(L1×)∩N⁡(L2×). Conversely, if a∈N⁡(L1×)∩N⁡(L2×), then by local Artin reciprocity, we see that a∈ker⁡ϕL1/K∩ker⁡ϕL2/K. But this means that ϕK⁢(a)|L1=ϕK⁢(a)|L2=1, and thus ϕK⁢(a)|L1⁢L2=1. But this implies that a∈ker⁡ϕL1⁢L2/K=N⁡((L1⁢L2)×).

  2. 2.

    One direction is obvious. Assume that N⁡(L1×)⊆N⁡(L2×). Therefore

    However, we know that

    which implies that L1=L1⁢L2, so L1⊇L2.

  3. 3.

    Assume H⊇N⁡(L×). Let M=LϕL/K⁢(H)⊆Gal⁡(L/K). Then by local Artin reciprocity, we have a commutative diagram

    This tells us that H=N⁡(M×) is also a norm subgroup.

  4. 4.

    Note that L1∩L2 is the largest subextension contained in both L1,L2. On the other hand, N⁡((L1∩L2)×) is the smallest subgroup containing both N⁡(L1×),N⁡(L2×), and the desired result follows. □

Corollary 2.7.3.

The map L↦N⁡(L×) defines a bijection

The idea of local Artin reciprocity was to understand extrinsic data about extensions using intrinsic data about the group K×. However, the notion of a norm subgroup still still extrinsic, so we want a more intrinsic characterization of norm subgroups.

Lemma 2.7.4.

Let L/K be a finite extension. If N⁡(L×) has finite index in K×, it must be open.

Proof.

Note that N:L×→K× is continuous and 𝒪L× is compact. Then N⁡(𝒪L×)⊆K× ic compact and hence closed. But then 𝒪K×/N⁡(𝒪L×)→K×/N⁡(L×), and thus N⁡(𝒪L×)⊆𝒪K× is open (and closed). But this implies that N⁡(𝒪L×)⊆K× is open (because 𝒪K×⊆K× is open), and thus N⁡(L×) must be open. □

Corollary 2.7.5.

If L/K is finite abelian, then N⁡(L×)⊆K× is a finite index open subgroup.

Theorem 2.7.6 (Local CFT II: local existence).

Every finite index open subgroup of K× is a norm subgroup.

Corollary 2.7.7.

We have a bijection

Remarks 2.7.8.
  1. 1.

    This bijection also holds for archimedean local fields. For K=ℝ, the two extensions ℝ,ℂ correspond to ℝ×,ℝ>0, while ℂ is algebraically closed and ℂ× is the only finite-index open subgroup of itself.

  2. 2.

    If K is a finite extension of ℚp, then any finite index subgroup of K× is automatically open. However, this is not true for K=𝔽q⁢((t)). In fact, if H⊆K× has finite index n, then (K×)n⊆H. Therefore, it suffices to show that (K×)n is open. It is easy to see that (K×)n⊇1+𝔭Km for some m≫0. Therefore the equation xn−a=0 has solutions in K for a∈1+𝔭Km by Hensel for p∤n and a stronger version for p∣n (that does not always hold).

Now we want a reformulation of local CFT using the norm topology.

Definition 2.7.9.

The norm topology on K× is given by declaring a basis of open neighborhoods of 1 to be the norm subgroups of K×, which are the same as finite index open subgroups in the usual topology.

Example 2.7.10.

Note that 𝒪K× is open under the usual topology, but is not open under the norm topology.

Remark 2.7.11.

The norm topology has fewer open sets and is therefore coarses than the usual topology.

Definition 2.7.12.

Define K̂× to be the completion of K× under the norm topology:

Then the Artin reciprocity map induces an isomorphism K̂×≃Gal⁡(Kab/K).

Proposition 2.7.13.

We have an isomorphism K̂×≅𝒪K××Ẑ as topological groups.

Proof.

Choose a uniformizer π of K. Then K×≅𝒪K××πℤ. Then a basis of finite index open subgroups is given by (q+𝔭km)×πn⁢ℤ for some m,n≥1. This implies that

Corollary 2.7.14.

There is an isomorphism Gal⁡(Kab/K)≅𝒪K××ℤ̂ as topological groups for any choice of uniformizer π. Therefore, we have a decomposition Kab=Kπ⋅Kur, where Kπ=(Kab)ϕK⁢(π) and Kur=(Kab)ϕK⁢(𝒪K×) is the maximal unramified extension. This means that Kπ is the totally ramified part of Kab.

Remark 2.7.15.

More canonically, consider the short exact sequence

If we consider the profinite completion of this, we obtain an exact sequence

Because ℤ↪ℤ̂ is dense, so is K×↪K̂×.

Now, it remains to prove local class field theory. First, we will construct the local Artin reciprocity map ϕK using Galois cohomology. After we prove that ϕK has the desired property, we will prove the local existence theorem by constructing enough norm subgroups using cyclic extensions.

3 Group Cohomology

Definition 3.0.1.

Let G be a group. Then a G-module is a (left) module over the ring ℤ⁡[G], or in other words, an abelian group with a linear (left) G-action.

Example 3.0.2.

Let L/K be a finite Galois extension of fields and G=Gal⁡(L/K). Then M=L and M=L× are both G-modules.

Example 3.0.3.

Any abelian group M can be regarded as a G-module under the trivial action.

Definition 3.0.4.

A homomorphism of G-modules α:M→N is a G-equivariant group homomorphism, or equivalently a morphism of ℤ⁡[G]-modules. The set of such morphisms is denoted HomG⁡(M,N).

We will denote the category of G-modules with G-linear maps by 𝖬𝗈𝖽G. Because 𝖬𝗈𝖽G=𝖬𝗈𝖽ℤ⁡[G], it is an abelian category with enough injectives and projectives. This allows us to develop the full theory of homological algebra in a concrete way.

Definition 3.0.5.

M∈𝖬𝗈𝖽G is injective if the functor HomG⁡(−,M) is exact. Dually, M∈𝖬𝗈𝖽G is projective if HomG⁡(M,−) is exact.

Definition 3.0.6.

An abelian category has enough injectives if any object can be embedded in an injective object. Similarly, an abelian category has enough projectives if any object has a surjection from a projective object.

Example 3.0.7.

The free ℤ⁡[G]-module M=ℤ⁡[G] is projective. In fact, HomG⁡(ℤ⁡[G],M)=M.

3.1 Definition of Cohomology

Definition 3.1.1.

Let M∈𝖬𝗈𝖽G. Define its G-invariants by

to be the largest submodule with trivial G-action.

Example 3.1.2.

Let G=Gal⁡(L/K). Then if M=L, Galois theory tells us that MG=K. Similarly, if M=L×, then MG=K×.

Remark 3.1.3.

In other words, we have MG=Homℤ⁡[G]⁡(ℤ,M), where ℤ has the trivial action. This implies that the functor M↦MG is always left-exact. Therefore, for a short exact sequence

we have an exact sequence

However, this may fail to be right-exact.

We may resolve this failure of right-exactness by constructing the derived functor of (−)G. This will give us a long exact sequence.

Definition 3.1.4.

The group cohomology Hr⁢(G,M) for any r≥0 is defined by be the functor Extr⁡(G,M). This is the right derived functor of Homℤ⁡[G]⁡(ℤ,M). It is characterized by

  1. 1.

    H0⁢(G,M)=MG.

  2. 2.

    A short exact sequence 0→A→B→C→0 induces a long exact sequence in group cohomology

  3. 3.

    If I∈𝖬𝗈𝖽G is injective, then Hr⁢(G,I)=0 for all r≥1.

Remark 3.1.5.

More concretely, we can compute Hr⁢(G,M) using an injective resolution of M. If we consider an injective resolution

then we apply (−)G to obtain a complex

and then we have Hr⁢(G,M)=Hr⁢((I∙)G).

Remark 3.1.6.

We can also compute Hr⁢(G,M) using a projective resolution of ℤ. If we consider a projective resolution

then we apply HomG⁡(−,M) to this and obtain a complex

and then H∙⁢(G,M) is just the cohomology of this complex.

Now we will give a description of the cohomology in low degree. In particular, what we will do is use an explicit free resolution of ℤ.

Definition 3.1.7.

Define Pr=ℤ⁡[G×⋯⁢G⏟r+1] where G acts on Pr by

Note that Pr is a free ℤ⁡[G]-module with basis {(1,g1,…,gr)}. Now we define the morphism Pr→Pr−1 by

Lemma 3.1.8.

The previous definition gives a free resolution of ℤ in 𝖬𝗈𝖽G.

Definition 3.1.9.

This is clearly a complex. To prove exactness, let

Then we can check that dr∘kr+kr−1∘dr−1=id. Then taking the image of both sides of ker⁢dr−1, we obtain dr∘kr⁢(ker⁢dr−1)=ker⁢dr−1 and thus ker⁢dr−1⊆Im⁢dr.

Corollary 3.1.10.

We can compute Hr⁢(G,M)=Hr⁢(HomG⁡(P∙,M)).

Definition 3.1.11.

We have an identification

These are called the homogeneous r-cochains of G with values in M and are denoted by C~r⁢(G,M). Then the differentials are given by

Then we have an explicit cochain description

Note that homogeneous r-cocycles φ:Gr+1→M are determined by their values on elements of the form (1,g1,…,gr) for gi∈G, or equivalently on elements of the form (1,g1,g1⁢g2,…,g1⁢⋯⁢gr). Therefore we may eliminate one degree of freedom.

Definition 3.1.12.

Define the group of inhomogeneous r-cochains to be the group

Now we have an isomorphism C~r⁢(G,M)≃Cr⁢(G,M) given by

The differentials Cr⁢(G,M)→dr+1Cr+1⁢(G,M) are given by

Now we can define the r-cocycles Zr⁢(G,M) and r-coboundaries Br⁢(G,M) and the corresponding cohomology groups Hr⁢(G,M).

Example 3.1.13.

Suppose r=1. Then we have

Such functions are usually called crossed homomorphisms.

On the other hand, we have

These functions are called principal crossed homomorphisms. Therefore, we have

Example 3.1.14.

If M acts trivially on G, then Z1⁢(G,M)=Hom𝖦𝗋𝗉⁡(G,M) and B1⁢(G,M) is trivial. Thus H1⁢(G,M)=Hom𝖦𝗋𝗉⁡(G,M).

Remark 3.1.15.

There is an explicit cochain description for Hr⁢(G,M), but computing using this definition is extremely tedious. Instead, we will try to break G and M into smaller pieces and then piece them back together.

3.2 Change of Groups

Definition 3.2.1.

Let H⊆G be a subgroup and let M∈𝖬𝗈𝖽H. Define the induced module

with the action of G given by

Example 3.2.2.

For any M∈𝖬𝗈𝖽G, we have IndGG⁡M=M.

Definition 3.2.3.

Let M∈𝖬𝗈𝖽G. Then define the restriction ResHG⁡M to be the same M but viewed as an H-module.

Proposition 3.2.4.
  1. 1.

    (Frobenius reciprocity) For any M∈𝖬𝗈𝖽G,N∈𝖬𝗈𝖽H, we have

  2. 2.

    The functor IndHG is an exact functor.

  3. 3.

    The functor IndHG preserves injections.

Proof.
  1. 1.

    We will construct explicit mutual inverses. We will define

    It is easy to check that these are inverse to each other.

  2. 2.

    Now we need to show that induction is right exact. Suppose M↠N is a surjective map of H-modules. Now let φ∈IndHG⁡N. Note that φ is uniquely determined by its values on a complete set of representatives s∈H\G. Then we lift φ⁡(s)∈N to φ⁡(s)~∈M using the surjection M↠N. Now define φ~⁢(h⁢s)=h⁢φ~⁢(s) for all h∈H, and now we obtain an element φ~∈IndHG⁡M mapping to φ.

  3. 3.

    Let I∈𝖬𝗈𝖽H be injective. Then IndHG⁡I is injective if and only if HomG⁡(−,IndHG⁡I) is exact. By Frobenius reciprocity, this is equivalent to exactness of HomH⁡(ResHG−,I), which is obvious. □

Proposition 3.2.5 (Shapiro’s lemma).

Let N∈𝖬𝗈𝖽H. Then Hr⁢(G,IndHG⁡N)≃Hr⁢(H,N) for all r≥0.

Proof.

Choose an injective resolution N→I∙ in 𝖬𝗈𝖽H. But then exactness of induction and preservation of injectives imply that IndHG→IndHG⁡I∙ is an injective resolution in 𝖬𝗈𝖽G. But now we see that

Definition 3.2.6.

A module M∈𝖬𝗈𝖽G is called induced if M=Ind1G⁡M0 for some abelian group M0.

Corollary 3.2.7.

If M is induced, then for all r≥1, Hr⁢(G,M)=0.

Proof.

By Shapiro, this reduces to computing cohomology Hr⁢(1,M0), but then we know that Hom⁡(ℤ,−)=(−) is exact, so all higher derived functors vanish. □

Now we will consider functorial properties of group cohomology with respect to change of groups. Given H⊆G and M∈𝖬𝗈𝖽G, we will define

  1. 1.

    The restriction functor Res:Hr⁢(G,M)→Hr⁢(H,M).

  2. 2.

    The corestriction functor Cor:Hr⁢(H,M)→Hr⁢(G,M) whenever [G:H]<∞.

  3. 3.

    The inflation functor Inf:Hr⁢(G/H,MH)→Hr⁢(G,M) when H is a normal subgroup of G.

Suppose we are given α:G′→G and 𝖬𝗈𝖽G∋M→𝛽M′∈𝖬𝗈𝖽G′ that are compatible in the sense that

for all g′∈G′,m∈M. Then we obtain a morphism of cochain complexes

This is compatible with the differentials, so we obtain a morphism Hr⁢(G,M)→Hr⁢(G′,M′). Now using this generality, we can define the three functors.

  1. 1.

    (Restriction) We will set α:H↪G and β:M→idM.

  2. 2.

    (Corestriction) We will set α:G→idG and

    This gives us the corestriction map by Shapiro.

  3. 3.

    (Inflation) We take α:G↠G/H and β:MH↪M.

Remark 3.2.8.

Suppose r=0. Then the functors are

Proposition 3.2.9.

The map Cor∘Res:Hr⁢(G,M)→Hr⁢(H,M)→Hr⁢(G,M) is given by multiplication by [G:H].

Proof.

Consider M→IndHG⁡M→M. Then this map is given by

Corollary 3.2.10.

If G is finite, then Hr⁢(G,M) is killed by |G|.

Proof.

Take H=1 and apply the previous proposition together with the fact that all higher cohomology vanishes for H=1. □

Corollary 3.2.11.

If G is finite and M is a finitely generated abelian group, then Hr⁢(G,M) is finite.

Proof.

If M is finitely generated, then so is Hr⁢(G,M) because the cochain complex is finitely generated. But then Hr⁢(G,M) is torsion, so it must be finite. □

Theorem 3.2.12 (Inflation-restriction exact sequence).

There is an exact sequence

Proof.

We will check this by hand.

First map is injective:

Let φ∈Z1⁢(G/H,MH). Now assume that

is a coboundary. Thus there exists m∈M such that Inf⁡(φ)⁢(g)=g⁢m−m for all g∈G. But then φ⁡(g¯)=g⁢m−m, but then for all h∈H, we see that φ⁡(h¯)=h⁢m−m=0 because h fixes MH. Therefore φ is a coboundary.

Composition is zero:

Clearly we have Res∘Inf=0 because the composition

is trivial because it passes through G/H.

Exactness on right:

Assume that Res⁡(φ)∈B1⁢(H,M). Then there exists m∈M such that φ⁡(h)=h⁢m−m for all h∈H. Now define

Now φ′,φ are cohomologous. We will now write φ′ as an inflation. Because φ′⁢(h)=φ⁡(h)−(h⁢m−m)=0, we know that φ′ factors through G/H. Moreover, we see that

This implies that φ′ is valued in MH, so it must have been an inflation. □

Theorem 3.2.13 (Inflation-restriction, general case).

Let r≥1. Assume Hi⁢(H,M)=0 for all 1≤i<r. Then we have an exact sequence

Proof.

We will proceed by induction on r using a dimension-shifting argument. Consider the induced G-module M∗=Ind1G⁡M. Then look at the exact sequence

where we have the map m↦(g↦g⁢m). But then the higher cohomology of M∗ vanishes, so we have isomorphisms Hi⁢(G,M′)≃Hi+1⁢(G,M) for all i≥1. But then M∗=IndHG⁡Ind1H⁡M, so we have isomorphisms Hi⁢(H,M′)≃Hi+1⁢(H,M) for all i≥1. Similarly, because H1⁢(H,M)=0, we have an exact sequence

and therefore M∗H is also an induced G/H-module. Therefore Hi⁢(G/H,(M′)H)≃Hi+1⁢(G/H,MH) for all i≥1.

Now, by assumption, we have Hi⁢(H,M)=0 for all 1≤i≤r−1. Therefore, Hi⁢(H,M′)=0 for all 1≤i≤r−2 and now we can apply the inductive hypothesis to obtain an exact sequence

This implies that

is also exact. □

Remark 3.2.14.

We have a more general version of the dimension shift. Given an exact sequence

where each Ai is induced, then Hr⁢(G,M)≃Hr+1⁢(G,N) for all r≥1.

Remark 3.2.15.

The inflation-restriction exact sequence is a special case of the Hochschild-Serre spectral sequence. We have

In our case, only the rows q=0,q=r are nontrivial.

3.3 Group Homology

Definition 3.3.1.

Define the G-coinvariants of M by

This is the largest quotient module of M where G acts trivially.

Definition 3.3.2.

The augmentation ideal is defined to be

Then we see that MG=M/IG⁢M=M⊗ℤ⁡[G]ℤ⁡[G]/IG=M⊗ℤ⁡[G]ℤ. In particular, the coinvariants functor is right-exact. Applying the derived functor package, we obtain

Definition 3.3.3.

Define the group homology by

to be the left derived functor Torrℤ⁡[G]⁡(ℤ,M) of M↦MG.

Remarks 3.3.4.

Group homology is characterized by

  1. 1.

    H0⁢(G,M)=MG;

  2. 2.

    A short exact sequence 0→A→B→C→0 in 𝖬𝗈𝖽G gives a long exact sequence

  3. 3.

    If P∈𝖬𝗈𝖽G is projective, then Hr⁢(G,P)=0 for all r≥1.

Remark 3.3.5.

Let P∙→M be a projective resolution. Then Hr⁢(G,M)=Hr⁢((P∙)G).

Proposition 3.3.6.

For any G, we have H1⁢(G,ℤ)≃Gab=G/[G,G].

Proof.

Consider the short exact sequence

This gives a long exact sequence

This becomes

But this tells us that H1⁢(G,ℤ)≃IG/IG2. Now define the map

By the identity

we see that this is a group homomorphism. Because IG/IG2 is abelian, the group homomorphism factors through Gab→IG/IG2. One can easily check that have an inverse IG/IG2→Gab given by g−1↦g. □

Remark 3.3.7.

This gives a homological interpretation of Gal⁡(L/K)ab as H1⁢(Gal⁡(L/K),ℤ).

Remark 3.3.8.

This is analogous to the topological fact that if X is path-connected, then H1⁢(X,ℤ)=π1⁢(X)ab. In fact, for any group G, we can define a space B⁢G=K⁡(G,1) such that π1⁢(B⁢G)=G and πi⁢(B⁢G)=0 for i≠1. We have Hr⁢(B⁢G,ℤ)=Hr⁢(G,ℤ), where the LHS is singular homology and the RHS is group homology. If we choose r=1, we recover H1⁢(G,ℤ)=Gab.

Example 3.3.9.
  1. 1.

    If G=ℤ, then we have B⁢ℤ=ℝ/ℤ=S1.

  2. 2.

    If G=ℤ∗ℤ∗⋯∗ℤ⏟n, then B⁢G=S1∨⋯∨S1⏟n.

3.4 Tate Cohomology

The idea is that for a finite group G, we will patch both group cohomology and group homology into a single theory.

Definition 3.4.1.

Let G be a finite group and M∈𝖬𝗈𝖽G. Define the norm map

By definition, the image of the norm map lies in the invariants and NmG⁡(g⁢m−m)=0, so the kernel always contains IG⁢M. This gives us a commutative diagram

and an exact sequence

In particular, this gives us

where Ĥ just means a modified version of cohomology.

Definition 3.4.2.

For r∈ℤ, define the Tate cohomology groups

Example 3.4.3.

Given a short exact sequence

in 𝖬𝗈𝖽G, we obtain a long exact sequence

and by the snake lemma, these can be patched to form a very long exact sequence

Remark 3.4.4.
  1. 1.

    If M∈𝖬𝗈𝖽G is induced, then Ĥr⁢(G,M)=0 for all r∈ℤ.

  2. 2.

    The dimension shift argument works for Ĥr in both directions. If M↪Ind1G⁡M, then Ĥr⁢(M1)=Ĥr+1⁢(M), where M1 is the cokernel. Similarly, if Ind1G⁡M↠M, then Ĥr⁢(M)=Ĥr+1⁢(M1), where M1 is the kernel.

  3. 3.

    Shapiro’s lemma still holds for Ĥr.

  4. 4.

    The property that CorHG∘ResHG is multiplication by [G:H] holds for Ĥr.

Remark 3.4.5.

Take P∙→ℤ to be a free resolution in 𝖬𝗈𝖽G. Taking the dual M↦M∗=Homℤ⁡(M,ℤ) where (g⁢φ)⁢(m)=φ⁡(g−1⁢m). This gives a resolution ℤ→P∙. If we take P−n≔Pn−1∗, we obtain

Then we can compute Ĥr⁢(G,M)=Hr⁢(HomG⁡((P∙),M)).

Now we will compute Tate cohomology explicitly for finite cyclic groups G=⟨σ⟩. We can very explicitly write down the norm map, invariants, and coinvariants:

This implies that

Remark 3.4.6.

All remaining Ĥr⁢(G,M) are determined by Ĥ0,Ĥ−1.

Theorem 3.4.7 (Period 2).

Let G=⟨σ⟩. Then we have isomorphisms Ĥr⁢(G,M)≃Ĥr+2⁢(G,M).

Proof.

Note that we have a free resolution of ℤ in 𝖬𝗈𝖽G given by

This computes the Tate cohomology and has period 2, so the Tate cohomology also must have period 2. □

Definition 3.4.8.

Define the Herbrand quotient

if both Ĥ0,Ĥ1 are finite.

Proposition 3.4.9.

If 0→A→B→C→0 is an exact sequence, then h⁡(B)=h⁡(A)⋅h⁡(C).

Proof.

Consider the long exact sequence

This implies that

From this, we obtain

so it suffices to show that |Q|=|K|. But here, we see that

because Tate cohomology has period 2. □

Proposition 3.4.10.

If M is finite, then h⁡(M)=1.

Proof.

Note that we have the exact sequence

This tells us that |MG|=|MG|. Next, we note that

is exact, so |Ĥ−1⁢(G,M)|=|Ĥ0⁢(G,M)|. □

Corollary 3.4.11.

If α:M→N has finite kernel and cokernel, then h⁡(M)=h⁡(N).

Proof.

Consider the exact sequence

This tells us that

Because α has finite kernel and cokernel, h⁡(ker⁡α)=h⁡(coker⁡α)=1, so h⁡(M)=h⁡(N). □

Now we want to compute Tate cohomology for more general G.

Theorem 3.4.12.

Let G be a finite group. If for any subgroup H⊆G we have H1⁢(H,M)=H2⁢(H,M)=0, then Ĥr⁢(G,M)=0 for all r∈ℤ.

Proof.

If G is cyclic, then it follows from the period 2 isomorphism. In the general case, if G is solvable, we can induct on |G|. If H⊲G is normal such that G/H is cyclic. By the induction hypothesis, we know Ĥr⁢(H,M)=0 for all r∈ℤ. Applying inflation-restriction, we have an exact sequence

By assumption, Hr⁢(G,M)=0 for r=1,2. By exactness, we know Hr⁢(G/H,MH)=0 for r=1,2, so Ĥr⁢(G/H,MH)=0 for all r∈ℤ because G/H is cyclic. By exactness and the inductive hypothesis, we know Ĥr⁢(G,M)=0 for all r≥1. But now the hypotheses are invariant under dimension shift, so we obtain the desired result.

Finally, if G is an arbitrary finite group, reduce to the solvable case. If we consider the Sylow subgroups Gp⊆G, we know Gp is solvable. Therefore Ĥr⁢(Gp,M)=0 for all r∈ℤ. By an exercise from the homework, we see Ĥr⁢(G,M)=0 for all r∈ℤ. □

Unfortunately, H2 generally does not vanish, but we can still gain insight about the behavior of Tate cohomology in this case.

Theorem 3.4.13 (Tate).

Let G be a finite group. If for any subgroup H⊆G we have

  1. 1.

    H1⁢(H,M)=0;

  2. 2.

    H2⁢(H,M)=ℤ/|H|⁢ℤ,

then Ĥr⁢(G,ℤ)≃Ĥr+2⁢(G,M) for all r∈ℤ.

Example 3.4.14.

We will apply Tate’s theorem to construct the local Artin map. Let G=Gal⁡(L/K) and M=L× and set r=−2. Then both conditions of the theorem are satisfied, and so

Now we have defined the inverse of the local Artin map!

Remark 3.4.15.

Chao says that Tate was very brave to go from the object Ĥ0⁢(G,M) that we want to understand to an object that did not exist at the time!

Before we prove Tate’s theorem, we will give an explicit description of H2⁢(G,M). Recall that

We also know that

Proposition 3.4.16.

The cohomology group H2⁢(G,M)=Z2⁢(G,M)B2⁢(G,M) parameterizes extensions

such that the conjugation action of E/M≅G on M agrees with the G-module structure. Moreover, 0∈H2⁢(G,M) corresponds to the split extension.

Proof.

Given such an extension, choose a splitting of E=G×M as sets. This gives us a section s:E→G Then for all g1,g2∈G, we have s⁡(g1)⋅s⁡(g2)=φ⁡(g1,g2)⁢s⁡(g1⁢g2) for some φ:G2→M.

We check that φ is actually a cocycle. To see this, note that

by associativity, which gives us

and therefore φ is a cocycle. We can check that any φ gives such an extension together with a section s and that different choices of the section s give coboundary relations. □

Definition 3.4.17.

For any φ∈Z2⁢(G,M), define its splitting module to be

with the action of G given by

for any choice of g1,g2∈G. By convention, we write X1≔φ⁡(1,1)∈M.

Remark 3.4.18.

By construction, we have an exact sequence

where Xg↦g−1.

Lemma 3.4.19.

The image of φ under H2⁢(G,M)→H2⁢(G,M⁡(φ)) is 0.

Proof.

Note that φ⁡(g1,g2)=g1⁢Xg2−Xg1⁢g2+Xg1 is the coboundary of g↦Xg. □

Remark 3.4.20.

The extension E′ associated to φ

of M⁡(φ) is a split extension.

Proof of Tate’s theorem.

There are two steps to the proof.

  1. 1.

    We know that H2⁢(G,M)=ℤ/|G|⁢ℤ=⟨φ⟩ with splitting module M⁡(φ). We also know that H1⁢(H,M)=0. We will show that H1⁢(H,M⁡(φ))=0 and H2⁢(H,M⁡(φ))=0. Consider the short exact sequence

    This gives a long exact sequence

    However, we know that H1⁢(H1,M)=0,H1⁢(H,IG)=ℤ/|H|⁢ℤ, and H2⁢(H,IG)=0. We also know that H2⁢(H,M)→H2⁢(H,M⁡(φ)) is the zero map (the restriction of φ to H2⁢(H,M) is a generator), so the map H1⁢(H,IG)→H2⁢(H,M) must be an isomorphism and therefore H1⁢(H,M⁡(φ))=H2⁢(H,M⁡(φ))=0.

  2. 2.

    Now we know that all Tate cohomology Ĥr⁢(G,M⁡(φ))=0 for all r∈ℤ. Now we use the four term exact sequence

    Because Ĥr⁢(G,M⁡(φ))=0 and Ĥr⁢(G,ℤ⁡[G])=0 for all r∈ℤ, By dimension shift, we have Ĥr⁢(G,ℤ)≅Ĥr+2⁢(G,M). □

Remark 3.4.21.

There is an alternative description of Ĥr⁢(G,ℤ)≃Ĥr+2⁢(G,M) as a cup product with some generator φ∈H2⁢(G,M). Here, if M,N∈𝖬𝗈𝖽G, we can consider the tensor product M⊗N as a G-module with the action g⁡(m⊗n)=g⁢m⊗g⁢n.

Now a cup product is a family of ℤ-bilinear maps

given by the following properties:

  1. 1.

    The cup product is functorial in both M,N.

  2. 2.

    For r=s=0, it is given by the natural map MG⊗NG→(M⊗N)G.

  3. 3.

    If 0→M′→M→M′′→0 is a short exact sequence such that

    is also exact, then for all m′′∈Hr⁢(G,M′′) and n∈Hs⁢(G,N), we have δ⁢m′′∪n=δ⁡(m′′∪n)∈Hr+s+1⁢(G,M′⊗N).

  4. 4.

    If 0→N′→N→N′′→0 is a short exact sequence such that

    is also exact, then for all m∈Hr⁢(G,M),n′′∈Hs⁢(G,N′′), we have m∪δ⁢n′′=(−1)r⁢δ⁢(m∪n′′).

The cup product exists and is unique. Given cocycles φ,ψ, we write

and this is unique by dimension shift. If G is finite, then the cup product extends to Tate cohomology.

4 Local Class Field Theory

Now our goal is to understand Hi⁢(Gal⁡(L/K),L×) for i=1,2. This will allow us to construct the local Artin reciprocity map.

4.1 Vanishing of first cohomology

This result was proved by Hilbert in 1894 Zahlbericht for cyclic extensions and then by Noether for general extensions.

Theorem 4.1.1 (Hilbert Theorem 90).

Let L/K be a finite Galois extension of fields. Then

Remark 4.1.2.

Note that this applies to extensions of arbitrary fields, not just local fields.

Proof.

Choose φ∈Z1⁢(Gal⁡(L/K),L×). We want φ⁡(g)=g⁢mm for some m∈L×. For any a∈L×, construct

Because g:L×→L× are distinct characters of L× for g∈G≔Gal⁡(L/K), we know that they are L-linearly independent, so the map

is not the zero map and thus there exists a∈L× such that m≠0. But now we compute

and thus φ⁡(g)=ng⁢m=g⁢m−1m−1, so φ∈B1⁢(G,L×). □

Example 4.1.3.

In the case where L/K is a cyclic extension and G=⟨σ⟩, then Ĥ−1⁢(G,L×)=H1⁢(G,L×)=0, and thus ker⁡NmG=Im(σ−1), which means that if a∈L× satisfies NmL/K⁡(a)=1, then a=σ⁢bb for some b∈L×.

Example 4.1.4.

Let L/K=ℚ⁡(i)/ℚ be the simplest quadratic extension. Let a=x+i⁢y∈L× with x,y=ℚ. Then we know NmL/K⁡(a)=x2+y2. If b=m+i⁢n∈L× for m,n∈ℚ, we see that

By Hilbert 90, if x2+y2=1, then there exist m,n∈ℤ such that

In particular, this tells us how to classify all Pythagorean triples.

Remark 4.1.5.

We may also consider infinite Galois extensions L/K, for example L=Ksep,Kab. Then we know

This allows us to define continuous Galois cohomology

The Hctsr can be computed using continuous cochains.

Remark 4.1.6.

More generally, if G=lim←H⁡G/H is a profinite group and M is a discrete G-module (which means G×M→M is continuous with respect to the profinite topology on G and the discrete topology on M), we may consider the continuous group cohomology

where the maps in the directed system are inflation.

Notation 4.1.7.

Let G=Gal⁡(L/K). We will simply write

for the continuous Galois cohomology.

Example 4.1.8.

Hilbert Theorem 90 implies that H1⁢(K,(Ksep)×)=0. Equivalently, we have Hét1⁢(Spec⁡K,𝔾m)≅Pic⁡Spec⁡K=0. In other words, all line bundles on a point are trivial.

4.2 Second Cohomology

Let L/K be a finite unramified extension of local fields. Recall that

where ℓ/k is the extension of residue fields. Our goal is to compute H2⁢(L/K,L×)≅ℤ/n⁢ℤ, where n=[L:K]. The idea is to break L× into pieces, so choose a uniformizer πL. Then we have L×≅𝒪L××πLℤ. Because L/K is unramified, we may choose πL=πK∈K×. Therefore, as a Galois module, we have L×≅𝒪L×⊕ℤ.

Definition 4.2.1.

Let UK=𝒪K×. Then write UK(i)≔1+𝔪Ki for all i≥1.

This gives us a natural filtration

with short exact sequences

and similarly

for i≥1.

Lemma 4.2.2.

We have Ĥr⁢(G,ℓ×)=Ĥr⁢(G,ℓ)=0 for all r∈ℤ.

Proof.

By Hilbert Theorem 90, we know H1⁢(G,ℓ×)=0. Because G is cyclic and ℓ× is finite, we can understand the Herbrand quotient. Therefore h⁡(ℓ×)=1, so H2⁢(G,ℓ×)=0. Because G is cyclic, we may apply period 2 to obtain the desired result.

Now we consider ℓ. We know that ℓ is actually an induced module by the homework, so all Tate cohomology vanishes by period 2. □

Corollary 4.2.3.

The norm map Nmℓ/k:ℓ×→k× and the trace map Trℓ/k:ℓ→k are surjective.

Proof.

We know that Ĥ0⁢(G,ℓ×)=k×Nmℓ/k⁡(ℓ×)=0, so the norm map is surjective. Similarly, we know Ĥ0⁢(G,ℓ)=kTrℓ/k⁡(ℓ)=0, so the trace is surjective. □

Lemma 4.2.4.

The norm map NmL/K:UL→UK is surjective.

Proof.

Consider the commutative diagram

Given any a∈UK, we want b∈UL such that Nm⁡(b)=a. Because Nm:ℓ×→k×, we may find b0∈UL such that Nm⁡(b0)≡a(modπK). Now let a1≔aNm⁡(b0)∈UK(1). Because Tr:ℓ→k is surjective, we may find b1∈UL(1) such that a2≔a1Nm⁡(b1)∈UK(2). Repeating this to infinity, we can find bi∈UL(i) such that ai+1≔aiNm⁡(bi)∈UK(i+1). Setting

we know

so a=Nm⁡(b). □

Corollary 4.2.5.

For all r∈ℤ, we have Ĥr⁢(G,UL)=0.

Proof.

By the Lemma, we know that Nm:UL→UK is surjective, so Ĥ0⁢(G,UL)=0. But then we know H1⁢(G,UL)⊆H1⁢(G,L×)=0 is a direct factor, so H1⁢(G,UL)=0. The desired result is simply an application of period 2. □

Theorem 4.2.6.

If L/K is an unramified extension, then H2⁢(G,L×)≅Hom⁡(G,ℚ/ℤ)≅1n⁢ℤ/ℤ.

Proof.

We know H2⁢(G,L×)≅H2⁢(G,UL)⊕H2⁢(G,ℤ)=H2⁢(G,ℤ)=Hom⁡(G,ℚ/ℤ) by the homework. The final isomorphism is obvious. □

Definition 4.2.7.

Define the invariant map

Taking the limit over all unramified extensions, we obtain a map

To simplify our notation, we will write H2⁢(L/K)≔H2⁢(L/K,L×), so we have

Our goal now will be to extend this to H2⁢(Ksep/K)≃ℚ/ℤ.

Remark 4.2.8.

We always have an injective inflation map H2⁢(Kur/K)→InfH2⁢(Ksep/K). The map exists, and injectivity is by Hilbert 90 and inflation-restriction. We want to prove that this is indeed an isomorphism.

Lemma 4.2.9.

Let L/K be a finite extension of degree n. Then the diagram

commutes.

Remark 4.2.10.

Because Lur=L⋅Kur, we may view Gal⁡(Lur/L) as a subgroup of Gal⁡(Kur/K), and the inclusion is comparible with (Kur)×→(Lur)×, so restriction makes sense.

Proof.

Recall that invK is computed by H2⁢(Kur/K)≃H2⁢(Kur/K,ℤ)≃H1⁢(Kur/K,ℚ/ℤ)≃ℚ/ℤ. Now we need to see the effect of restriction on the groups on the right of the chain of isomorphisms. Note that if e=e⁡(L/K) and f=f⁡(L/K), then πLe=πK, so the diagram

commutes. Therefore, the maps on H2⁢(Kur/K,ℤ)→H2⁢(Lur/L,ℤ) are given by multiplication by e. Finally, for φ∈H1⁢(Kur/K,ℚ/ℤ)=Hom⁡(Gal⁡(Kur/K),ℚ/ℤ), we see that φ↦φ⁡(FrobK). Because FrobL=FrobKf, we see that Res⁡(φ)⁢(FrobL)=f⁢φ⁢(FrobK). Therefore, the final map is simply f⋅e⋅Res=n⋅Res. □

Corollary 4.2.11.

Let L/K be a finite Galois extension of degree n. Then H2⁢(L/K) contains a subgroup of 1n⁢ℤ/ℤ.

Proof.

By Hilbert 90 and inflation-restriction, we have a commutative diagram

Therefore, 1n⁢ℤ/ℤ injects into H2⁢(L/K). □

Proposition 4.2.12.

We have |H2⁢(L/K)|=n. In particular, H2⁢(L/K)≅1n⁢ℤ/ℤ.

Proof.

We know this is true with L/K is cyclic. We know h⁡(L×)=n and H1⁢(L/K)=0, so |H2⁢(L/K)|=n. In general, we will reduce to the cyclic case because L/K is always solvable (because K is a local field). We will induct on [L:K]. Choose a tower of fields K⊂K′⊂L such that K′/K is cyclic. Applying inflation-restriction, we have an exact sequence

We know that |H2⁢(K′/K)|=[K′:K] because it is cyclic. We also know that |H2⁢(L/K′)|=[L:K′] by the inductive hypothesis, so |H2⁢(L/K)|≤[K′:K]⋅[L:K′]=[L:K]=n. But we know that H2⁢(L/K) contains a subgroup of order n, so |H2⁢(L/K)|=n. □

Corollary 4.2.13.

We have a commutative diagram

In particular, we may view H2⁢(L/K)→InfH2⁢(Kur/K).

Theorem 4.2.14.

There is an isomorphism H2⁢(Kur/K)→InfH2⁢(Ksep/K).

Proof.

Recall that H2⁢(L/K)↪H2⁢(Kur/K)↪H2⁢(Ksep/K). Taking the limit over finite Galois extensions L/K, we have an injection H2⁢(Ksep/K)↪H2⁢(Kur/K)↪H2⁢(Ksep/K), but the composition is the identity, so all inclusions are equalities. □

Definition 4.2.15.

We can extend the invariant map to invK:H2⁢(Ksep/K)≃ℚ/ℤ.

Remark 4.2.16.

For any field K, define the Brauer group Br⁡(K)≔H2⁢(Ksep/K,(Ksep)×). This is in fact Hét2⁢(Spec⁡K,𝔾m). The goal of local class field theory is to compute Br⁡(K)≃ℚ/ℤ for a local field K.

Remark 4.2.17.

Classically, the group Br⁡(K) classifies central simple algebras over K. In fact, it classifies central simple algebras up to the relation A∼B if and only if A⊗KMn⁢(K)≃B⊗Mm⁢(K) for some n,m. The group structure is defined by [A]⊗[B]=[A⊗KB] with identity 0=[Mn⁢(k)].

Theorem 4.2.18 (Wedderburn).

Each class of Br⁡(K) is represented by a central division algebra over K.

Example 4.2.19.

If K is a local field, then we may define the invariant invK⁢(B) for any central division algebra B over K. If B=K, then invK⁢(K)=0. Now if B is the quaternion algebra over K, we have invK⁢(B)=12. In fact there is an explicit construction of a central division algebra associated to 1n for every n. If L/K is a quadratic extension, then

so B⊗KL=M2⁢(L).

Example 4.2.20.

Call the element uL/K∈H2⁢(L/K) with invK⁢(uL/K)=1n the fundamental class of L/K.

Remark 4.2.21.

The local Artin reciprocity isomorphism Ĥ−2⁢(G,ℤ)→Ĥ0⁢(G,L×) can be realized as the cup product by the fundamental class uL/K.

4.3 Proof of Local Class Field Theory

Our goal is to prove the following result:

Theorem 4.3.1 (Local class field theory).
  1. 1.

    (Local Artin reciprocity) There exists a unique homomorphism

    such that

    1. (a)

      For any finite Galois extension L/K, we have a commutative diagram

    2. (b)

      For any finite unramified extension, ϕL/K⁢(π)=FrobL/K, where π is an uniformizer of K.

  2. 2.

    (Local existence theorem) There is a bijection between norm subgroups of K× and finite index open subgroups of K×. In addition, if char⁡K=0, then the finite index open subgroups of K× are precisely the finite index subgroups of K×.

4.3.1 Proof of local Artin reciprocity

To construct ϕK, we will use Tate’s theorem. Recall that by Hilbert 90, we have H1⁢(L/K,L×)=0. We also computed H2⁢(L/K,L×)=1n⁢ℤ/ℤ, where n=[L:K]. By Tate’s theorem, we have an isomorphism Ĥr⁢(L/K,ℤ)≃Ĥr+2⁢(L/K,L×) for all r∈ℤ. We will specialize to the case where r=−2, and this gives us an isomorphism

Then we define ϕL/Kab→∼Gal⁡(L/K)ab to be its inverse. Taking the limit over all finite Galois extensions L/K, we obtain a map

By construction, the property a) is satisfied.

To finish the proof, we need to check property b). Let L/K be a finite unramified extension. Our goal is to explicitly describe the shift by 2 isomorphism produced by Tate’s theorem. Recall that Tate’s theorem comes from M=L× and G=Gal⁡(L/K). Then we construct the splitting module for some φ∈H2⁢(G,M)=H2⁢(L/K,L×)=1n⁢ℤ/ℤ by defining

Then we have short exact sequences

and

Now we need to trace through the isomorphisms

The first isomorphism is given by FrobL/K≕σ↦σ−1. Therefore it remains to write down the connecting homomorphism Ĥ−1⁢(G,IG)→Ĥ0⁢(G,M). This is described using the exact sequences

by lifting (σ−1)∈IG/IG2 to M⁢(φ)G and then taking Nm. This will automatically land in MG. An obvious list of σ−1 to M⁢(φ)G is Xσ, so we need to compute Nm⁡(Xσ). Therefore we have

Now it remains to describe this explicitly for φ∈H2⁢(G,M)=H2⁢(L/K,L×)=1n⁢ℤ/ℤ.Recall that this comes from H2⁢(L/K,L×)→∼H1⁢(L/K,ℚ/ℤ)=Hom⁡(G,ℚ/ℤ) coming from the exact sequence

Now if f∈Hom⁡(G,ℚ/ℤ) such that f⁡(σ)=1n, then we can lift f to f~∈Z1⁢(G,ℚ) and take df~. This gives us f~⁢(σi)=in∈ℚ. If φ=df~, then we have

This gives us

and thus Nm⁡(Xσ)=π∈K×/Nm⁡(L×). □

4.3.2 Proof of Local Existence

We will assume that char⁡K=0. We will establish a bijection

Recall that any subgroup of K× containing a norm subgroup is also a norm subgroup. Also note that any finite index subgroup of index n contains (K×)n. Therefore it suffices to show that (K×)n is a norm subgroup. Our goal is to study the n-th power may using the Kummer sequence

This is an exact sequence in 𝖬𝗈𝖽G for G=Gal⁡(Ksep/K), so it induces a long exact sequence in cohomology

This gives us the Kummer isomorphism K×/(K×)n≅H1⁢(G,μn).

  1. 1.

    In the simplest case, when μn⊆K×, then the isomorphism becomes

    where L/K is the maximal abelian extension of exponent n. Now the pairing

    is in fact a perfect pairing. Therefore |K×/(K×)n|=|Gal⁡(L/K)|, so by local Artin reciprocity, we have the map

    Also, we know Nm⁡(L×)⊇(K×)n, and therefore (K×)n=Nm⁡(L×).

  2. 2.

    In general, if μn is not contained in K×, then take K1=K⁡(μn). Now we can find L1/K1 such that Nm⁡(L1×)=(K1×)n. Now take L/K Galois such that L1⊆L. Then

    Therefore (K×)n contains a norm subgroup, and is thus also a norm subgroup. □

5 Global class field theory

Recall that a number field K is a finite extension of ℚ. The goal of global class field theory is to understand Gal⁡(K¯/K)ab, or equivalently to understand finite abelian extensions of K in terms of intrinsic data associated to K. Analogously to the local case, we would like to construct a global Artin reciprocity map

which induces isomorphisms

Remark 5.0.1.

When K is a local field, we took CK=K×. When K is global, we can again try to take CK=K×, but in fact this fails.

Example 5.0.2.

Choose L/K=ℚ⁡(i)/ℚ. Then Gal⁡(L/K)=ℤ/2, but

For example, any p≡3(mod4) cannot be written as a2+b2, and in fact the quotient is infinite.

Instead of K×, we will take CK to be a generalization of the ideal class group ClK, which is the quotient of all fractional ideals by the principal ones. Thus the class group measures the failure of unique factorization in K.

5.1 Idèles

Our generalization of ideals will be the idèle class group of K, which is again given by

for some group 𝕀K. We need to define 𝕀K and NmL/K:CL→CK.

Notations 5.1.1.

We will denote a (finite or infinite) place of K by v and |−|v the absolute value associated to v such that

for all a∈K×. Then we will denote the completion of K with respect to |−|v by Kv. When v is a finite place, we write 𝒪V⊆KV for the ring of integers, 𝔭v⊆𝒪K the corresponding maximal ideal, and 𝔭̂v⊆𝒪v for the completion of the maximal ideal.

Remark 5.1.2.

We may attempt to take 𝕀K=∏vKv×. The problem is that this group is too large in the sense that it is not locally compact.

Instead of this, we will allow only finitely many coordinates to be not a unit in 𝒪v. This will make our group locally compact.

Definition 5.1.3.

The group of idèles (from the French élément idéal for ideal element) is

This is the restricted product of ∏vKv× with respect to ∏v𝒪V×.

Remark 5.1.4.

The idèles are equipped with the restricted product topology with a basis of open neighborhoods given by

where Uv is open in Kv× for all v and Uv=𝒪V× for all but finitely many v.

Let S be a finite set of places containing S∞={v⁢infinite}. Then define

Note the first factor is locally compact and the second product is compact, so 𝕀K,S is locally compact. Then we can write

and each 𝕀K,S is open in 𝕀K, so in particular 𝕀K is locally compact.

Remark 5.1.5.

The group 𝕀K has a basis of open neighborhoods of 1 given by

Remark 5.1.6.

For all places v, there is a natural injection

where a goes in the coordinaate corresponding to v. The induced topology on Kv× agrees with the usual topology.

Remark 5.1.7.

If IK is the group of fractional ideals of K, then there is a natural surjection

with kernel given by

In particular, 𝕀K can be thought of as an enlargement of IK with extra information from the infinite places and units at finite places.

Remark 5.1.8.

The group 𝕀K is completely local.

Proposition 5.1.9.

The natural map

has discrete image.

Proof.

It suffices to show that 1∈K× is open in its imaage. Equivalently, for some S⊇S∞ and some ε>0, we have K×∩U⁡(s,ε)={1}. Let a∈K×∩U⁡(S,ε). Then |a−1|v<ε for all v∈S and |a|v=1 for all v∉S. Note that |a−1|v≤1 for all v∉S. Therefore

On the other hand, by the product formula, we know ∏v|a−1|v=1 unless a−1=0. In particular, for S sufficiently large and ε sufficiently small, we must have a=1. □

Remark 5.1.10.

Note that K×↪Kv× has dense image for any single v. However, the product formula implies that for a∈K×, its absolute values |a|v at different places v “repel” each other and makes K×↪𝕀K have discrete image.

Example 5.1.11.

Consider the ring ℤ⁡[2]↪ℝ. Because 2 is irrational, this has dense image. However, the embedding

has discrete image. In fact, the image is a lattice in ℝ2.

Similarly, ℤ⁢[1p]↪ℝ,ℚp both have dense image, but

has discrete image.

Now it should not be surprising that K×↪𝕀K has discrete image. This is simply an infinite dimensional generalization of this phenomenon.

Definition 5.1.12.

Define the idèle class group Ck≔𝕀K/K× and equip it with the quotient topology. Because K× is discrete, this group is locally compact.

Remark 5.1.13.

The 𝕀K↠IK descends to CK↠ClK. In particular, CK can be viewed as an enlargement of ClK.

Definition 5.1.14.

Let L/K be a finite extension. Recall that

given Kv⁢[x]/f⁡(x)=∏(Kv⁢[x]/fi⁢(x)), where fi are the irreducible factors of f after the extension. Therefore, for all α∈L×, we have

Now we can define

Definition 5.1.15.

By definition, we have a commutative diagram

Therefore, we have an induced map NmL/K:CL→CK.

5.2 Statement of global class field theory

Our goal is now to construct the global Artin reciprocity map. We hope to do this using the local Artin reciprocity maps. First, we recall some properties from algebraic number theory.1 If L/K is a finite Galois extension of number fields, v a place of K, and w∣v a place of L above v, we can define the decomposition group

Remark 5.2.1.

Suppose that w′∣v is another place of L above v. Then D⁡(w),D⁡(w′) are conjugate in Gal⁡(L/K). In particular, if Gal⁡(L/K) is abelian, then D⁡(w)=D⁡(w′) only depends on v. By local class field theory, we obtain

which only depends on v.

Proposition 5.2.2.

There exists a unique continuous homomorphism

such that for any finite abelian extension L/K and any place v of K, the diagram

commutes.

Proof.

Let a=(av)∈𝕀K. If av∈𝒪v× and Lw/Kv is unramified (true for all but finitely many places), then ϕv⁢(av)=1∈Gal⁡(Lw/Kv). Now ϕK is uniquely determined by a finite product

where v runs over the places where av∉𝒪v× or Lv/Kv is ramified. Therefore, varying L/K and taking a limit, this determines ϕK uniquely.

Now it remains to check that ϕK is continuous. In particular, we need the kernel to be open. Let S be the set of ramified places of L/K. Then by local class field theory, the diagram

commutes. Therefore ker⁡(ϕL/K)⊇NmL/K⁡(𝕀L,S), which is open. Therefore the kernel itself is also open. □

Theorem 5.2.3 (Global class field theory).
  1. 1.

    (Global Artin reciprocity) The homomorphism ϕK:𝕀K→Gal⁡(Kab/K) satisfies

    1. (a)

      ϕK⁢(K×)=1. Therefore we obtain a global Artin map ϕK:Ck→Gal⁡(Kab/K).

    2. (b)

      For any finite abelian extension L/K, ϕK induces an isomorphism

  2. 2.

    (Global existence theorem) For any finite index open subgroup N⊆CK, there exists a finite abelian extension L/K such that N=Nm⁡(CL).

Corollary 5.2.4.

We have a bijection between finite abelian extensions L/K and finite index open subgroups of CK.

Before we turn to the proof of global class field theory, we will describe some consequences of this bijection. We will begin with ray class fields.

Definition 5.2.5.

A modulus of K is a function

such that

  1. 1.

    𝔪⁡(v)=0 for all but finitely many places v.

  2. 2.

    𝔪⁡(v)=0,1 if v is a real place.

  3. 3.

    𝔪⁡(v)=0 if v is a complex place.

Definition 5.2.6.

For a modulus 𝔪, define the principal congruence subgroup

where Uv,0=𝒪v× and Uv,i=1+𝔭vi for i≥1 for finite places. For infinite places, we have Kv,0×=Kv×, and Kv,1×=ℝ>0.

Definition 5.2.7.

For a modulus 𝔪, define

Then define the ray class group to be

Here, the term “ray” comes from a positivity condition ℝ>0⊂ℝ× at real places, which geometrically means specifying a ray.

Definition 5.2.8.

Write 𝔪=∏v𝔭v𝔪⁡(v)=𝔪0⋅𝔪∞. Then we have a description

In particular, if 𝔪⁡(v)≡0, then 𝔪=∏v𝔭v0=1 and Cl1=ClK. On the other hand, if 𝔪=𝔪∞, then

This is called the narrow class group.

Definition 5.2.9.

The abelian extension L/K associated to Cl𝔪≅Gal⁡(L/K) under global class field theory is called the ray class field of 𝔪. The case of 𝔪=1 is called the Hilbert class field and is denoted by H with ClK=Gal⁡(H/K). The case where 𝔪=𝔪∞ gives us the narrow Hilbert class field and is denoted by H+. This is characterized by ClK+=Gal⁡(H+/K). In particular, H⊆H+.

Remark 5.2.10.

By local class field theory, the ramified places of the ray class field of m is contained in 𝔪. In particular, H/K is the maximal abelian abelian extension unramified everywhere, and H+/K is the maximal abelian extension unramified at all finite places.

Example 5.2.11.

If K=ℚ, then ClK=ClL+≅{1} and thus H=H+=ℚ.

Example 5.2.12.

If K=ℚ⁡(3), then ClK={1}. However, we have

Therefore H=K=ℚ⁡(3) and H+=K⁡(i)=ℚ⁡(3,i).

Example 5.2.13.

Let K=ℚ and let 𝔪=(m) for some m∈ℤ. Then

If 𝔪=(m)⋅∞, then Cl(m)⋅∞=(ℤ/m)×.

Now the ray class fields L/ℚ for either 2∤m or 4∣m are given by Cl(m)⋅∞≅Gal⁡(L/ℚ), where L=ℚ⁡(ζm), and the field given by Cl(m) is L=ℚ⁡(ζm+ζm−1). In particular, by varying m, we obtain the Kronecker-Weber theorem.

Theorem 5.2.14 (Kronecker-Weber).

We have ℚab=⋃m≥1ℚ⁡(ζm) and thus

Remark 5.2.15.

Hilbert’s 12th problem asks for an explicit description of the ray class fields for a general number field. For K=ℚ, we have the Kronecker-Weber theorem. For imaginary quadratic fields, we have modular functions and elliptic curves with complex multiplication. In general, this problem is still open.

We will use the following strategy to prove global Artin reciprocity:

  1. 1.

    We will prove the first inequality2 that [CK:Nm⁡(CL)]≥[L:K]. This will be a consequence of the cohomology of 𝕀K,CK. In particular, we will show that h⁡(CK)=[L:K]. This will imply that |Ĥ0⁢(G,CL)|≥[L:K], but this is precisely the index we need.

  2. 2.

    Next, we will prove the surjectivity of 𝕀K→ϕL/KGal⁡(L/K). This is an application of the first inequality and the weak Chebotarev density theorem.

  3. 3.

    We will prove the second inequality3 [CK:Nm⁡(CL)]≤[L:K]. This is a consequence of Chebotarev density and we will use the analytic tool of L-functions.4

  4. 4.

    We will prove that ϕK⁢(K×)=1. This will follow from the determination of the Brauer groups of K.

5.3 Cohomology of Idèles and first inequality

Let L/K be a finite Galois extension of number fields. Recall that if v is a place of K, then L⊗KKv=∏w∣vLw. The action of Gal⁡(L/K) on L⊗Kv induces an action on the product by permuting the places above v. If α=(αw)w∣v and σ∈Gal⁡(L/K), then (σ⁢α)σ⁢w=σ⁡(αw)∈Lσ⁢w.

Proposition 5.3.1.

Fix w0∣v. Then there exists an isomorphism of G≔Gal⁡(L/K)-modules

where Gw0=D⁡(w0) is the decomposition group.

Proof.

Recall that IndGw0G⁡Lw0≔{f:G→Lw0∣f(τσ)=τf(σ) for all τ∈Gw0,σ∈G}. For any α=(αw)w, we define

Then it is easy to compute

Conversely, for f∈IndGw0G⁡Lw0, we define (αf)σ⁢w0≔σ⁢f⁢(σ−1). This is well-defined by Gw0-equivariance, and we can check that these two assignments are inverse to each other. □

Corollary 5.3.2.

By Shapiro’s lemma, we have Hr⁢(G,∏w∣vLw)=Hr⁢(Gw0,Lw0) for all r≥0.

Similarly, we have

Corollary 5.3.3.

For all r≥0, Hr⁢(G,∏w∣vLw×)=Hr⁢(Gw0,Lw0×) and Hr⁢(G,∏w∣vUw)=Hr⁢(Gw0,Uw0).

In particular, all of these are independent of the choice of w0∣v. To stress the independence, we write Gv≔Gw0,Lv≔Lw0, and so on and so forth.

Proposition 5.3.4.

We have the following cohomology of the idèles:

  1. 1.

    H0⁢(G,𝕀L)=𝕀K.

  2. 2.

    For all r∈ℤ, Hr⁢(G,𝕀L)=⨁vĤr⁢(Gv,(Lv)×).

Proof.
  1. 1.

    We know that a=(aw)∈𝕀L is G-invariant if and only if (aw)w∣v is G-invariant for all places v of K if and only if aw∈Kv× is independent of the choice of w∣v for all v, and this is equivalent to a=(av)v∈𝕀K.

  2. 2.

    Let S be a finite set of places of K containing S∞. Write

    Then 𝕀L=lim→⁡𝕀L,S and therefore

    If S contains all places of K that are ramified in L, then for all v∉S, Lv/Kv is unramified. By local class field theory, Ĥr⁢(Gv,Uv)=0, so

Corollary 5.3.5.

We have H1⁢(G,𝕀L)=0 and H2⁢(G,𝕀L)=⨁v1nv⁢ℤ/ℤ, where nv=[Lv:Kv].

These statements follow from collecting all of the local information.

Corollary 5.3.6.

Assume L/K is cyclic. Assume S contains all places of K that are ramified in L and the infinite places. Then h⁡(𝕀L,S)=∏v∈Snv.

Proof.

By local class field theory, we have

Definition 5.3.7.

Let T≔{w∣v:v∈S} be a finite set of places of L. Define the group of T-units of L to be

Our next goal is to compute h⁡(U⁡(T)) and use this to prove the first inequality. First, we must compute the cohomology of units. Let L/K be a finite Galois extension of number fields and S be a finite set of places containing S∞. Let T be the set of places of L above S.

Proposition 5.3.8.

Assume L/K is cyclic. Then

where nv=[Lv:Kv],n=[L:K].

Proof.

We need to use the following comparison result for Herbrand quotients from the homework: let G be a cyclic group and V an ℝ⁡[G]-module. If M,N⊆V are two G-stable lattices, then h⁡(M)=h⁡(N). We will apply this to the case where V=ℝ|T|=Hom𝖲𝖾𝗍⁡(T,ℝ).

  1. 1.

    Consider N=Hom𝖲𝖾𝗍⁡(T,ℤ). This is clearly a G-stable lattice. As a G-module, we have

    In particular, we have

  2. 2.

    Now consider λ:U⁡(T)→V=ℝ|T| given by α↦(log⁡|α|w)w∈T. If M0=Im(λ), Dirichlet’s unit theorem for U⁡(T) says that M0 is a lattice of rank |T|−1 in the hyperplane {∑w∈Txw=0}. Then M=M0⊕ℤ⋅(1,…,1) is a G-stable lattice.

    Now we know that h⁡(M)=h⁡(N), so h⁡(U⁡(T))=h⁡(M0) because the kernel of λ is the set of roots of unity in L, which is finite. Now we obtain

    as desired. □

Now we are able to compute the cohomology of the idèle class group CK.

Lemma 5.3.9.

For L/K, we have H0⁢(G,CL)=CK.

Proof.

The short exact sequence in 𝖬𝗈𝖽G

gives a long exact sequence

where the last equality is by Hilbert 90. This implies that H0⁢(G,CL)=𝕀K/K×=CK. □

We need to compute the Herbrand quotient of the idèle class group, so first we will rewrite 𝕀K in terms of S-units.

Lemma 5.3.10.

Choose S containing a generating set of prime ideals in ClK. Then 𝕀K=K×⋅𝕀K,S.

Proof.

Recall that we have a surjection

with kernel 𝕀K,S∞. In particular, we can write

By the choice of S, we have

Theorem 5.3.11.

Assume L/K is cyclic. Then h⁡(CL)=[L:K]. In particular, [CK:Nm⁡(CL)]≥[L:K].

Proof.

Choose S such that T a generating set of prime ideals for ClL. Then we can rewrite

Now we have

As an application of the first inequality, we will prove

Lemma 5.3.12.

Asssume L/K is solvable. If there exists D⊆𝕀K such that D⊆Nm⁡(𝕀L) and K×⋅D⊆𝕀K is dense, then L=K.

Proof.

Assume L≠K. Then we may find a tower K⊆K′⊆L such that K′/K is cyclic. Then we know D⊆Nm⁡(𝕀K′). By local class field theory, Nm⁡(𝕀K′)⊆𝕀K is an open subgroup, so K×⋅Nm⁡(𝕀K′) is also an open subgroup. Therefore K×⋅Nm⁡(𝕀K′) is also closed. By density, we see that K×⁢Nm⁡(IK′)=𝕀K. In particular, [CK:Nm⁡(CK′)]=1, so [K′:K]=1, which is a contradiction. Thus L=K. □

Proposition 5.3.13.

Assume L/K is solvable. If all but finitely many places v of K split completely in L, then L=K. Equivalently, if L≠K, then there are infinitely many places v of K that do not split completely in L.

Proof.

Choose S to contain the set of all v that do not split completely in L. Also choose

Then if v∉S, it splits completely in L, so Lv/Kv is a trivial extension. Therefore D⊆Nm⁡(𝕀L). On the other hand, by weak approximation, any (av)v∈S can be approximated by a global element a∈K×, so K×⋅D is dense in 𝕀K. By the previous lemma, L=K. □

Proposition 5.3.14.

Assume that L/K is solvable. Then {Frobw/v}v⁢unramified generate Gal⁡(L/K).

Proof.

Take H=⟨Frobw/v⟩⊆Gal⁡(L/K). Choose E=LH. If v is unramified in L, then v splits completely in LH. But this means that E=K because only finitely many places are ramified, so H=Gal⁡(L/K). □

Corollary 5.3.15.

Assume L/K is abelian. Then the Artin map ϕK:𝕀K→Gal⁡(L/K) is surjective.

Proof.

By local class field theory, if v is unramified in L, then ϕK⁢(1,1,…,πv,1,…)=Frobw/v. We know that the Frobw/v generate Gal⁡(L/K), so ϕK is surjective onto Gal⁡(L/K). □

5.4 Analytic aspects and second inequality

Recall that if L/K is solvable and L≠K, then there exist infinitely many places of K that do not split completely in L. For example, if L/K=ℚ⁡(i)/ℚ, then p splits in L if and only if p≡1(mod4). Thus there are infinitely many primes p≡3(mod4). This result is sometimes called the weak Chebotarev density theorem, so what is the strong Chebotarev density theorem? In our example, we expect that half of all primes are congruent to 1 modulo 4 and that the other half are congruent to 3 modulo 4. Of course, there are infinitely many primes, so we need to understand what is meant by “half of all primes” here.

Definition 5.4.1.

Let P be a set of primes in ℤ. Define the natural density of P to be

if this limit exists.

Definition 5.4.2.

Let P be a set of finite places of a number field K. Define the natural density of P to be

if this limit exists.

Of course, this definition is a terrible way to do analysis because we are throwing away so much information, so we would like a definition better suited to analytic methods.

Definition 5.4.3.

Define the Dirichlet density

if this limit exists.

The fundamental fact, proved in any first course in analytic number theory, is that if μ⁡(P) exists, then δ⁡(P) exists and μ⁡(P)=δ⁡(P). Thus natural density is a stronger notion than Dirichlet density. The proof of this fact uses the prime number theorem and is omitted here because this is not a course in analytic number theory.

Remark 5.4.4.

It is possible that the Dirichlet density exists, but the natural density does not. For a∈{1,…,9}, choose Pa to be the set of primes starting with the digit a. Then μ⁡(Pa) does not exist, but δ⁡(Pa)=log10⁡(1+1a), a result due to Bombieri.5

Dirichlet density is easier to study because we can use the tool of L-functions, which generalize the Riemann zeta function.

Definition 5.4.5.

The Riemann zeta function is defined by

For a general number field K, the Dedekind zeta function is defined by

These series converge when Re(s)≫0. When K=ℚ, ζℚ⁢(s)=ζ⁡(s).

Definition 5.4.6.

Let χ:(ℤ/m)×→ℂ× be a character. The Dirichlet L-function of χ is given by

where χ⁡(n)=0 if (n,m)≠1. When χ=𝟙, then L⁡(s,𝟙)=ζ⁡(s).

Definition 5.4.7.

Let K be a number field and 𝔪 be a modulus of K. Let χ:Cl𝔪→ℂ× be a character. The Weber L-function of χ is given by

Again, we write χ⁡(𝔞)=0 if (𝔞,𝔪)≠1. These series converge absolutely when Re(s)≫0. When K=ℚ, 𝔪=(m)⋅∞ and we have Cl(m)⋅∞=(ℤ/m)×, so the Weber L-function recovers the Dirichlet L-function. Of course, when χ=𝟙, we recover the Dedekind zeta function.

Theorem 5.4.8 (Analytic properties of Weber L-functions).
  1. 1.

    If χ≠𝟙, then L⁡(s,χ) has an analytic continuation to all s∈ℂ and L⁡(1,χ)≠0.

  2. 2.

    If χ=𝟙, then L⁡(s,χ)=ζK⁢(s) has an analytic continuation to all s∈ℂ∖{1} and a simple pole at s=1. The residue Ress=1⁡ζK⁢(s) is given by the class number formula for K.

  3. 3.

    The function L⁡(s,χ) satisfies the functional equation relating the values at s and 1−s.

Remark 5.4.9.

In general, there are more complicated complicated notions of L-functions, although this is a vaguely defined notion. In general, L-functions are series of the form ∑n≥1anns satisfying

  1. 1.

    The series has an Euler product ∏p1polynomial⁢(p−s).

  2. 2.

    The series has an analytic continuation to ℂ except for poles for ζ⁡(s) at s=1.

  3. 3.

    The analytic continuation satisfies a functional equation relating the values at s and 1−s.

Corollary 5.4.10.

For any number field K, log⁡ζK⁢(s)∼log⁡1s−1 (this means the difference is analytic at s=1).

Proof.

By the theorem, ζK⁢(s)⁢(s−1) is analytic at s=1, so log⁡ζK⁢(s)=log⁡ζK⁢(s)⁢(s−1)−log⁡(s−1). Because log⁡ζK⁢(s)⁢(s−1) is analytic at s=1, so we obtain the desired result after rearranging. □

Proposition 5.4.11.

We have the identity

Proof.

We compute

The second term is analytic at s=1, so we obtain the desired result. □

Corollary 5.4.12.

We can rewrite

Theorem 5.4.13.

Let L/K be a finite Galois extension and P be the set of primes of K splitting completely in L. Then

Example 5.4.14.

Let K=ℚ,L=ℚ⁡(i). Then we have δ⁡(p≡1(mod4))=δ⁡(p≡3(mod4))=12.

Proof.

We can rewrite

because the terms with f⁡(𝔮)≥2 contribute to an analytic function and the second item on the right hand side is ζL⁢(s). Now we obtain

Now we want to use these tools to prove the second inequality. The idea is to show that

We will reinterpret the right-hand side as the density of primes of K that split completely in L. Now we will reinterpret the left-hand side as a certain density also. Recall that for 𝔪=𝔪0⋅𝔪∞ a modulus of K, we have the ray class group Cl𝔪.

Theorem 5.4.15.

Let K𝔪⊆H⊆IK𝔪. Then let A∈IK𝔪/H be an ideal class. Then

Proof.

Let χ be a character of 𝕀K𝔪/H. This is a quotient of Cl𝔪, so it induces a quotient of Cl𝔪. Now let L⁡(s,χ) be the Weber L-function. But now

But now we want to pick out a single ideal class, so we need to use some Fourier analysis. We use the fact that

for any g∈G. Multiplying χ⁢(A)−1 and summing over all χ, we see that

and thus δ⁡(𝔭∈A)=1[IK𝔪:H], as desired. □

Remark 5.4.16.

Every finite abelian extension L/K admits a modulus 𝔪 such that

is a quotient of the ray class group Cl𝔪. In particular, we may reinterpret the global Artin map as a map ϕL/K:IK𝔪→Gal⁡(L/K) given by 𝔭↦Frob𝔭.

Theorem 5.4.17 (Second inequality).

We have [IK𝔪:K𝔪⋅Nm⁡(IL𝔪)]≤[L:K].

Proof.

Apply the previous theorem to H=K𝔪⋅Nm⁡(IL𝔪) and A=[0]. Then we know

On the other hand, we know

But now we know 𝔭 splits completely in L only if 𝔭∈Nm⁡(IL𝔪), which is the same thing as 𝔭∈A. Thus δ⁡(𝔭∈A)≥δ⁡(𝔭⁢ splits completely in ⁢L), as desired. □

Before we continue with the proof of global class field theory, we will prove an important result in analytic number theory.

Theorem 5.4.18 (Chebotarev density theorem).

Let L/K be a finite Galois extension of number fields. Let σ∈G≔Gal⁡(L/K) and Cσ be the conjugacy class of σ. Then

Proof.

We will use global Artin reciprocity, so in the future we will not be using the Chebotarev density theorem. By global Artin reciprocity for abelian extensions, we have an isomorphism Ik𝔪/H→Gal⁡(L/K) for some modulus 𝔪. But then it is easy to see that

because G is abelian. Now when L/K is nonabelian, we will reduce to the abelian case. Let σ∈G and write M=L⟨σ⟩. Applying the abelian case to the abelian extension L/M, write

We know δ⁡(s1)⁢1|σ|. Now consider a subset of degree 1 primes of M and write

We know that δ⁡(S1)=δ⁡(S2). Finally, consider

Now we have a surjection S2↠S3 and the fiber has size |ZG⁢(σ)||σ|. This is because if 𝔮↦𝔭 with Frob𝔭=σ, then τ⁢𝔮↦𝔭 if and only if Frobτ⁢𝔮=τ⁢Frob𝔮⁡τ−1 if and only if τ⁢σ=σ⁢τ. Therefore τ⁢𝔮=𝔮 if and only if τ∈D⁡(𝔮)=⟨σ⟩=Gal⁡(L/M). Then we know that

and therefore

Example 5.4.19.

Suppose σ=1. Then Cσ={1}, so Chebotarev density tells us that

Example 5.4.20.

Choose K=ℚ and L=ℚ⁡(ζN), where either N is odd or 4∣N. Then we know G=(ℤ/N)×. If we choose σ=a∈(ℤ/N)×, then Cσ={a}, so by Chebotarev, we see that

Thus we have recovered a strengthening of Dirichlet’s theorem on primes in arithmetic progression!

Remark 5.4.21.

The original proof of Chebotarev density does not rely on global Artin reciprocity. Instead, it reduces to the cyclotomic case, where global Artin reciprocity is known explicitly. In fact, Artin’s proof of global Artin reciprocity was inspired by Chebotarev’s proof (will be reduced to the cyclotomic case).

Example 5.4.22 (Simplest nonabelian example).

Let K=ℚ and L=ℚ⁡[x]/(x3−x2+1). This is the cubic extension with minimal |dL|, and here dL=−23. Then Gal⁡(L~/ℚ)≅S3. Now S3 has three distinct conjugacy classes (1),(12),(123). These correspond to f⁡(x) splitting completely in 𝔽p, f⁡(x) splitting into a quadratic and a linear factor in 𝔽p, and f⁡(x) being irreducible in 𝔽p, respectively. Now we can compute (possibly using a computer) the table

Table 5.1: Primes in each conjugacy class
σ p
(1) 59,…
(12) 5,7,11,17,19,37,43,53,…
(123) 2,3,13,29,31,41,47,…

Chebotarev predicts that the three classes will occur in a ratio of 1:3:2.

Remark 5.4.23.

There is no simple congruence condition on p to determine Frobp. However, there is a criterion on p in terms of modular forms. For example, for z∈ℋ, we may consider the function

Now some of the coefficients are

Table 5.2: Coefficients at primes
p 2 3 5 7 11 13 17 19 23 29 31 37 41 43 47 53 59
ap −1 −1 0 0 0 −1 0 0 1 −1 −1 0 −1 0 −1 0 2

So we see that the conjugacy classses corresppond to the coefficients being 2,0,−1. In fact, these are the character values for the two-dimensional irreducible representation ρ of S3. By a miracle, we have Tr⁡ρ⁡(Frobp)=ap!

Remark 5.4.24.

Recall that abelian extensions can be understood in terms of the idèle class group using Artin reciprocity. But if we write 𝕀ℚ=𝔸ℚ×, we can consider the idèle class group as a space related to G⁢L1. But now abelian extensions now correspond to 1-dimensional representations of Gal⁡(ℚ¯/ℚ), and the Langlands program gives a conjectural correspondence between n-dimensional representations of Gal⁡(ℚ¯/ℚ) and automorphic forms on [G⁢Ln]=G⁢Ln⁢(𝔸ℚ)/G⁢Ln⁢(ℚ). For example, when n=2, the 2-dimensional representations correspond to modular forms relating Tr⁡ρ⁡(Frobp) and coefficients ap of the modular form f⁡(z).

5.5 Brauer groups and proof of global Artin reciprocity

We have already proved the first inequality that [CK:Nm⁡(CL)]≥[L:K], at least when L/K is cyclic. We have also proved the second inequality that [CK:Nm⁡(CL)]≤[L:K] in general.

Corollary 5.5.1 (Global version of Hilbert 90).

Let L/K be finite Galois. Then H1⁢(G,CL)=0.

Proof.

We begin in the case when G is cyclic. Here, [CK:Nm⁡(CL)]=[L:K]. However, h⁡(CL)=[CK:Nm⁡(CL)]|H1⁢(G,CL)|=[L:K], so H1⁢(G,CL)=0.

Now it remains to reduce the general case to the cyclic case. First, we will consider G solvable and induct on |G|. Consider H⊲G such that G/H is cyclic. Then inflation-restriction gives us an exact sequence

Now the two outside terms vanish by G/H being cyclic and the inductive hypothesis, so H1⁢(G,CL)=0.

Finally, if G is an arbitrary group, we simply consider the restriction to all Sylow subgroups

However, this restriction is injective by a past homework, so we are done. □

Now our goal is to compute the Brauer group H2⁢(K). We will do this in terms of the local Brauer groups.

Proposition 5.5.2.

The natural restriction map H2⁢(L/K)→⨁vH2⁢(Lv/Kv) is injective.

Proof.

The short exact sequence

gives a long exact sequence

But now from the computation of the cohomology of idèles and the previous corollary, we have an injection

Corollary 5.5.3.

Given β∈H2⁢(K), there exists a cyclic extension L/K with L⊆K⁡(ζm) for some m such that β is split by L (β∈H2⁢(L/K)=ker⁡(H2⁢(K)→H2⁢(L))).

Proof.

By the proposition, we know that β∈H2⁢(K) is completely determined by its local restrictions {βv}v. We also know that all but finitely many βv vanish. By local class field theory, we know that β∈H2⁢(Kv)=ℚ/ℤ (the isomorphism is given by the invariant map). But now there exists m≥1 such that m⋅invv⁢(βv)=0 for all v. However, by functoriality of invv for w∣v, we obtain

by one of the homeworks. Once we prove the following lemma, we will be done. □

Lemma 5.5.4.

Let S be a finite set of finite places of K and m≥1. Then there exists a cyclic cyclotomic extension L/K such that m∣[Lv:Kv] for all v∈S.

Proof.

First we reduce to the case K=ℚ. Here, we can just take the composite with K and replace m with m⋅[K:ℚ]. Also, we will reduce to the case where m=ℓs, wheere ℓ is some prime. Here, we can just take the composite of the extensions for all prime power factors of m. We need ℓs∣[Lp:ℚp] for some cyclic cyclotomic extension L/ℚ. Now

Now we can form a cyclic cyclotomic extension

Now we can compute for all p∈S that

Where t is the smallest integer such that ℓr∣|𝔽pt×|=pt−1. In particular, as we increase r, [ℚp⁢(ζpr):ℚp]→∞, so [Lp:ℚp]→∞ as we increase r, so we may choose r≫0 such that ℓs∣[Lp:ℚp]. □

We have already proved that ϕL/K:𝕀K→Gal⁡(L/K) is surjective, so it remains to show that it factors through CK/Nm⁡(CL). We need to prove that

Theorem 5.5.5.

K× is contained in the kernel of ϕL/K. This means ϕL/K⁢(K×)=1.

This will imply that the global Artin map is an isomorphism. To prove this, we will use a global relation from Brauer groups.

Theorem 5.5.6.

For all β∈H2⁢(L/K), we have

Equivalently, ∑vinvv⁢(β)=0.

Our strategy will contain the following steps:

  1. 0.

    We will prove Theorem 5.5.5 for L/K=ℚ⁡(ζm)/ℚ.

  2. 1.

    Next, we will prove Theorem 5.5.5 for L/K cyclic cyclotomic.

  3. 2.

    Next, we will prove Theorem 5.5.6 for L/K cyclic cyclotomic.

  4. 3.

    Next, we will prove Theorem 5.5.6 for L/K finite abelian.

  5. 4.

    Finally, we will prove Theorem 5.5.5 for L/K finite abelian.

5.5.1 Step 0

Choose a∈ℚ×. We need to show that ϕ⁡(a)=1∈Gal⁡(ℚ⁡(ζm)/ℚ)=(ℤ/m⁢ℤ)×. It suffices to treat the case where m=ℓr is a prime power, so we compute explicitly ϕ=∏pϕp. There are several cases:

  1. 1.

    p≠ℓ is finite. Then p is unramified, so ϕp⁢(a)=(ζℓr↦ζℓrps)=ps∈(ℤ/ℓr)×, where a=u⋅ps.

  2. 2.

    p=ℓ. Here p is ramified, so ϕp⁢(a)=(ζℓr↦ζℓru−1)=u−1∈(ℤ/ℓr)×, where a=u⁢ℓs.

  3. 3.

    p=∞. Here, ϕ∞⁢(a)=sign⁡(a)∈{±1}=Gal⁡(ℂ/ℝ).

To show that ϕ⁡(a)=1, it suffices to consider a=p,ℓ,−1.

  1. 1.

    If p≠ℓ, then ϕp⁢(p)⁢ϕℓ⁢(p)=p⋅p−1=1.

  2. 2.

    If p=ℓ, then ϕ⁡(ℓ)=ϕℓ⁢(ℓ)=1−1=1.

  3. 3.

    Finally, ϕ⁡(−1)=ϕℓ⁢(−1)⁢ϕ∞⁢(−1)=(−1)−1⁢(−1)=1.

5.5.2 Step 1

We will prove the following result. Because we already proved Theorem 5.5.5 for cyclotomic extensions, this will give us the result for cyclic cyclotomic extensions.

Lemma 5.5.7.

If Theorem 5.5.5 holds for L/K, then it also holds for subextensions M/K of L/K and for composites L′/K′, where L′=L⋅K′.

Proof.

By functoriality of the local Artin maps, ϕM/K=𝕀K→ϕL/KGal⁡(L/K)↠Gal⁡(M/K). Therefore if ϕL/K⁢(K×)=1, ϕM/K⁢(K×)=1 also.

For the composites, we use functoriality of the local Artin maps to obtain a commutative diagram

But now the desired result follows from diagram chasing with (K′)×. □

5.5.3 Step 3

Every β∈H2⁢(K) is split by a cyclic cyclotomic extension by Corollary 5.5.3. This means there exists L/K cyclic cyclotomic such that β∈H2⁢(L/K). This reduces the finite abelian case to the cyclic cyclotomic case.

5.5.4 Steps 2 and 4

We will relate the global Artin maps with Brauer groups. Recall that H2⁢(G,ℤ)=H1⁢(G,ℚ/ℤ)=Hom⁡(G,ℚ/ℤ). Denote this map by δχ↦χ. Now the cup product with δχ gives us a map H0⁢(G,M)→H2⁢(G,M). If L/K is an extension of local fields, then for M=L×, we have a commutative diagram

If L/K is an extension of global fields, we consider M=𝕀L. Then we have a commutative diagram

5.5.5 Step 2

Assume that Theorem 5.5.5 holds for L/K cyclic cyclotomic. Then the top row is trivial, so because G is cyclic, the bottom row is also trivial (in the cyclic case, the shift in degree map is an isomorphism). But this means that β∈H2⁢(L/K) satisfies ∑invv⁢(β)=0, which is the same thing as Theorem 5.5.6 holding for L/K cyclic cyclotomic.

5.5.6 Step 4

Assume that Theorem 5.5.6 holds for L/K finite abelian. But this means the bottom row of (5.1) is trivial, so in particular, χ⁡(ϕL/K⁢(K×))=0. But χ is arbitrary in Hom⁡(G,ℚ/ℤ), so ϕL/K⁢(K×)=1. This gives Theorem 5.5.5 for all finite abelian extensions.

5.5.7 Brauer groups

As a consequence of global Artin reciprocity, we can determine the Brauer group of an arbitrary number field.

Theorem 5.5.8.

Let K be a number field. Then

Proof.

By Theorem 5.5.6, we know that H2⁢(K) is contained in the kernel. It remains to prove the reverse inclusion. Consider (βv)∈ker. We know there exists L/K cyclic cyclotomic such that (βv)v∈⨁vH2⁢(Lv/Kv). Because L/K is cyclic, we have an isomorphism of exact sequences

By definition, the top row is exact, so the bottom row is also exact. But this means that for (βv)∈ker⁡(∑invv), there exists β∈H2⁢(L/K) such that (βv)=β. □

Example 5.5.9.

Consider K=ℚ. Then we have a map

This is given by Bv=B⊗ℚv. Now by the theorem, we note that (Bv)v comes from B/ℚ if and only if ∑invv⁢(Bv)=0. For example, we have

This means that quaternion algebras B/Q are in bijection with the set of (Bv)v where an even number of Bv are not M2⁢(ℚv).

5.6 Proof of global existence

We will now complete the proof of global class field theory. We need to prove the following statement:

Theorem (Global existence).

For any finite index open subgroup U⊆CK, there exists a finite abelian extension L/K such that U=Nm⁡(CL).

Definition 5.6.1.

A subgroup of CK is a norm subgroup if it is of the form Nm⁡(CL) for some finite abelian extension L/K.

Now global existence is the same thing as every finite index open subgroup being a norm subgroup.

Lemma 5.6.2.

Any subgroup containing a norm subgroup is also a norm subgroup.

Proof.

Assume U=Nm⁡(CL) and V⊇U. Then global Artin reciprocity implies that CK/U≃Gal⁡(L/K). Now we have a commutative diagram

by the fundamental theorem of Galois theory. By construction, V=ker⁡(CK→Gal⁡(M/K)), so in particular, V=Nm⁡(CM) by global Artin reciprocity. □

Now it suffices to produce a enough norm subgroups such that every finite index open subgroup contains a norm subgroup. Here, we will use Kummer extensions. Let K be a field and μn⊆K. Then if Ln is the maximal abelian extension of K with exponent n, then by considering the Kummer sequence

we obtain a perfect pairing Gal⁡(Ln/K)⊗K×/(K×)n→μn given by (σ,b)↦σ⁡(bn)bn. This gives a bijection between finite subgroups of K×/(K×)n and finite abelian extensions of K of exponent n. Here, to a subgroup B, we associate field L=K⁡(Bn) given by adjoining all n-th roots of elements in B and to an extension L/K we simply take the n-th powers of all elements in L×. Here, it is easy to see that |B|=[L:K].

Example 5.6.3.

Let K be a local field and assume μn⊆K. Then Ln/K is finite and

Here, we know K×=𝒪K××ℤ, so K×/(K×)n=𝒪K×/(𝒪K×)n×ℤ/n. Because 𝒪K× contains a finite index subgroup isomorphic to 𝒪K, we have |𝒪K×/(𝒪K×)n|=|μn⁢(𝒪K×)|⋅|𝒪K/n⁢𝒪K|=n⁢1|n|K, as desired.

Now assume K is a number field.

Lemma 5.6.4.

Assume μn⊆K. Assume that S⊇S∞∪{v∣n}. Assume S is large enough so that 𝕀K=K×⋅𝕀K,S. Let a∈K× such that a∈(Kv×)n for all v∈S and a∈Uv for all v∉S. Then a∈(K×)n.

Proof.

Let L=K⁡(an). Then we see that Lv=Kv for all v∈S and Lv/Kv is unramified for all v∉S. By local class field theory, in the first case we have Nm⁡((Lv)×)=Kv×, and in the second case Nm⁡(Uv)=Uv. Therefore Nm⁡(𝕀L)⊇𝕀K,S, so K×⁢Nm⁡(𝕀L)⊇K×⁢𝕀K,S=𝕀K, and therefore Nm⁡(CL)⊇CK, so L=K. □

Proposition 5.6.5.

Assume μn⊆K. Let U⊆CK be a finite index open subgroup and assume CK/U has exponent n. Then U is a norm subgroup.

Proof.

Let S⊇S∞∪{v∣n} and

Then let V≔K×⋅E/K×⊆CK. Because CK/U has exponent n, we know U⊇V for S sufficiently large. It remains to show that V is a norm subgroup. Enlarging S we may assume that 𝕀K=K×⁢𝕀K,S.

Let U⁡(S)=K×∩𝕀K,S be the S-units. By the previous lemma, K×∩E=U⁡(S)∩E=U⁢(S)n. Set L=K⁡(U⁡(S)n). This is a finite abelian extension of exponent n unramified away from S. By local class field theory, we know Nm⁡((Lv)×)⊇(Kv×)n for all s∈S and Nm⁡(Uv)=Uv for all v∉S. But now E⊆Nm⁡(𝕀L) and thus Nm⁡(CL)⊇V. Now we show that Nm⁡(CL)=V.

We will show that [CK:Nm⁡(CL)]=[CK:V]. But now we have

Now we compute

We also compute

by Dirichlet’s unit theorem. Thus [CK:V]=n|S|. On the other hand, we have

using Kummer theory and the Dirichlet unit theorem. This implies that V=Nm⁡(CL). □

To finish the proof of global existence, we reduce to the case of Kummer extensions.

Lemma 5.6.6.

Let U⊆CK be a finite index open subgroup and V=NmK′/K−1⁡(U) for some finite extension K′/K. If V is a norm subgroup, then so is U.

Proof.

Write V=NmL/K′⁡(CL) for some finite abelian L/K′. Now let M be the maximal abelian subextension of L/K. Then

Therefore U is a norm subgroup. □

We are now able to complete the proof of global existence. We will induct on [CK:U] and assume p∣[CK:U]. If μp⊆K, choose U1⊆CK containing U such that [CK:U1]=p. Then U1 is a norm subgroup, so write U1=Nm⁡(CL). Now NmL/K:CL→CK/U has image U1/U with kernel V≔NmL/K−1⁡(U). Then [CL:V]=|U1/U|=[CK:U]p, and by the inductive hypothesis, V is a norm subgroup, so U is also a norm subgroup.

If μp⊈K, then take K′=K⁡(μp). Now U′=NmK′/K−1⁡(U). Then U′ is a norm subgroup by the above, so by the lemma, U is also a norm subgroup.

5.7 Primes of the form x2+n⁢y2

In this section we discuss a classical application of class field theory. This was discussed at the beginning of these notes. Recall that if K=ℚ⁡(d) with discriminant dK, then the splitting behavior of

is determined by the Legendre symbol (dKp).

Example 5.7.1.

Consider K=ℚ⁡(−1) with dK=4. Then p splits in K if and only if (−4p)=1, which is equivalent to (−1p)=1, is equivalent to p≡1(mod4) by basic arithmetic.

On the other hand, 𝒪K=ℤ⁡[i] is a PID, so p splits in K if and only if p=(x+i⁢y)⁢(x−i⁢y), which is equivalent to p=x2+y2 for integers x,y∈ℤ. Combining the two equivalences, we see that p=x2+y2 if and only if p≡1mod4.

Example 5.7.2.

Consider K=ℚ⁡(−5) with dK=−40. Then p splits in K if and only if (−5p)=1, and by quadratic reciprocity this is equivalent to p≡1,3,7,9(mod20). However, 𝒪K=ℤ⁡[−5] is no longer a PID, so we have a different criterion for p=x2+5⁢y2.

Theorem 5.7.3.

If p≠2,5, then p=x2+5⁢y2 if and only if p≡1,9(mod20).

Some examples are given below.

Table 5.3: Primes of the form x2+5⁢y2
p (x,y)
29 (3,2)
41 (6,1)
61 (4,3)
89 (3,4)

Similarly, we have

Theorem 5.7.4.

If p≠2,3, then p=x2+6⁢y2 if and only if p≡1,7(mod24).

Some examples are given below.

Table 5.4: Primes of the form x2+6⁢y2
p (x,y)
7 (1,1)
31 (5,1)
73 (7,2)

Our goal is to explain the proof of these results using class field theory.

Lemma 5.7.5.

Let 𝔭 be a finite prime for a number field K. Then 𝔭 is principal if and only if 𝔭 splits completely in HK, where HK is the Hilbert class field of K.

Proof.

By global Artin reciprocity, ClK≅Gal⁡(HK/K). But then 𝔭 is principal if and only if [𝔭]∈ClK is trivial. However, this is equivalent to Frob𝔭 being trivial, which is equivalent to p splitting completely in HK. □

Proposition 5.7.6.
  1. 1.

    If p≠2,5, then p=x2+5⁢y2 if and only if p splits completely in ℚ⁡(−5,−1).

  2. 2.

    If p≠2,3, then p=x2+6⁢y2 if and only if p splits completely in ℚ⁡(−6,−3).

Proof.
  1. 1.

    If p≠2,5, then p=x2+5⁢y2 if and only if (p)=𝔭1⁢𝔭2 splits as a product of principal ideals in K=ℚ⁡(−5). By the lemma, this is equivalent to 𝔭1,𝔭2 splitting completely in HK=ℚ⁡(−5,−1), and combining, we see this is equivalent to p splitting completely in HK.

  2. 2.

    This is the same argument, except HK=ℚ⁡(−6,−3) when K=ℚ⁡(−6). □

Therefore, to prove the main theorems, it remains to prove the following result.

Proposition 5.7.7.
  1. 1.

    p splits completely in ℚ⁡(−5,−1) if and only if p≡1,9(mod20).

  2. 2.

    p splits completely in ℚ⁡(−6,−3) if and only if p≡1,7(mod20).

The key to this result is that in these cases, HK/ℚ is an abelian extension, so we may apply global Artin reciprocity. Recall that for ℚ, Cl(N)⋅∞=Gal⁡(ℚ⁡(ζN)/ℚ)=(ℤ/N)×. Then under the Artin reciprocity map, p≡1(modN) if and only if p splits completely in ℚ⁡(ζN).

Example 5.7.8.

When N=4, we have ℚ⁡(ζ4)=ℚ⁡(i), so we recover the result that p≡1(mod4) if and only if p splits in ℚ⁡(i).

More generally, the global existence theorem gives a bijection between finite index open subgroup of Cℚ and finite abelian extensions of ℚ. In particular, we obtain a bijection between finite subgroups of (ℤ/N)× and finite abelian subextensions K/ℚ of ℚ⁡(ζN). This is given by Gal⁡(K/ℚ)=(ℤ/N)×/H. Thus we see that p∈H if and only if p splits completely in K.

Example 5.7.9.

Let N=5. Then (ℤ/5)×≅ℤ/4, and so the possibilities for H are H={1},{±1},(ℤ/5)×. Clearly the trivial subgroup corresponds to ℚ⁡(ζ5) and (ℤ/5)× corresponds to ℚ, and {±1} corresponds to ℚ⁡(ζ5+ζ5−1)=ℚ⁡(5). By global class field theory, this is the only abelian subextension K/ℚ of ℚ⁡(ζ5).

Note that because quadratic extensions are abelian, every quadratic extension is contained in a cyclotomic extension. For example, 5=ζ5−ζ52−ζ53+ζ54.

Example 5.7.10.

Consider N=20. Then (ℤ/20)×≅ℤ/4×ℤ/2. Now the subgroups and corresponding subextensions are given below.

Table 5.5: Subgroups and subfields
H⊆(ℤ/20)× K
{1} ℚ⁡(ζ20)
{±1} ℚ⁡(ζ20+ζ20−1)
{1,11} ℚ⁡(ζ5)
{1,9,11,19} ℚ⁡(5)
{1,9,13,17} ℚ⁡(ζ4)=ℚ⁡(i)
{1,9} ℚ⁡(5,i)
{1,3,7,9} ℚ⁡(−5)
(ℤ/20)× ℚ

This gives us the following diagram of field extensions:

In particular, we see that p≡1,9(mod20) if and only if p splits completely in ℚ⁡(−5,−1).

Example 5.7.11.

When N=24, then (ℤ/24)×=ℤ/2×ℤ/2×ℤ/2. Then we see that ℚ⁡(−6) corresponds to {1,5,7,11} and ℚ⁡(−3) corresponds to {1,7,13,23}, so ℚ⁡(−6,−2) corresponds to {1,7}. Therefore p≡1,7(mod24) if and only if p splits completely in ℚ⁡(−6,−3).

Remark 5.7.12.

When K/ℚ is abelian, class field theory gives us a congruence condition for primes splitting completely in K. On the other hand, when K/ℚ is nonabelian, then such a congruence condition does not exist.

Lemma 5.7.13.

Denote by S⁢p⁢l⁢(L/K) to be the set of primes 𝔭 of K splitting completely in L. Then L1⊆L2 if and only if S⁢p⁢l⁢(L1/K)⊇S⁢p⁢l⁢(L2/K). In particular, L1=L2 if and only if S⁢p⁢l⁢(L1/K)=S⁢p⁢l⁢(L2/K).

Proof.

One direction is clear. In the other direction, suppose S⁢p⁢l⁢(L1/K)⊇S⁢p⁢l⁢(L2/K). Then S⁢p⁢l⁢(L1⁢L2/K)⊇S⁢p⁢l⁢(L2/K). But now Chebotarev density tells us that [L1⁢L2:K]≤[L2:K], and this is only possible if L1⁢L2=L2, which means L1⊆L2. □

Proposition 5.7.14.

If K/ℚ is nonabelian, then there does not exist a congruence condition for p splitting completely in K.

Proof.

If there exists such a congruence condition, then choose p∈S⁢p⁢l⁢(K⁡(ζN)/ℚ). Thus p∈ℚ⁡(ζN/ℚ), and therefore p≡1(modN). By the criterion, {p≡1(modN)}⊆S⁢p⁢l⁢(K/ℚ). But this means that S⁢p⁢l⁢(ℚ⁡(ζN)/ℚ)⊆S⁢p⁢l⁢(K/ℚ), so ℚ⁡(ζN)⊇K. This implies K/ℚ is abelian. □

Example 5.7.15.

If K=ℚ⁡(−14), then ClK=ℤ/4, so HK=ℚ⁡(2⁢2−1) with Gal⁡(HK/ℚ)≅D4. Thus there is no congruence condition for p=x2+14⁢y2.

Remark 5.7.16.

Even though there is no congruence condition, non-abelian reciprocity gives us a condition using coefficients of modular forms.