Class field theory
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 can be written as for integers ?
We can try to answer this by experiment. Clearly, and the other primes below cannot be written as a sum of two squares. Then, because any square is congruent to or modulo , we see that if is an odd prime, then
Theorem 1.0.2 (Fermat, Christmas Day, 1640).
An odd prime can be written as a sum of two squares if and only if .1 Similarly, we have:
- •
if and only if .
- •
if and only if .
In the modern day, we should reinterpret as a factorization problem in the number field . Now we write our problem as . Similarly, the second problem can be written as in and the third problem can be expressed in the field . More generally, we can consider for an arbitrary , called the discriminant of . 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 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 be odd primes. Then we have the identity
Recall that if is odd and , then if and only if has solutions. Quadratic reciprocity tells us that the equation is highly related to the equation . This ability to change the modulus is very helpful in solving these classical problems.
Example 1.0.4.
A prime splits in if and only if if and only if if and only if .
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 splits in if and only if for some ? 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:
- •
splits in if and only if .
- •
splits in if and only if .
- •
splits in if and only if .
However, there is no congruence condition for splitting in for any modulus . The question is what is different about the last example. First, the fields are all Galois extensions of with Galois groups . On the other hand, is not Galois and its Galois closure has Galois group .
Definition 1.0.7.
A field extension is called abelian if is Galois and is abelian.
This gives us the following slogan: given a number field , class field theory
- 1.
classifies all abelian extensions in an accessible way;
- 2.
describes factorization of primes of in in terms of groups intrinsic to (for example the class group of ).
1.1 A Special Case of CFT
Here, we will classify all unramified abelian extensions . Recall that the class group of a number field is
This is always a finite abelian group.
Definition 1.1.1.
Let be a subgroup. Then a finite unramified abelian extension is a class field for a subgroup if splits in if and only if .
Theorem 1.1.2 (Unramified CFT).
Given , the class field for exists and is unique. Moreover, each finite unramified abelian extension arises as a class field. This gives us a bijection
Moreover, .
Definition 1.1.3.
Note that the class field for is the maximal unramified abelian extension of , called the Hilbert class field. This gives a canonical isomorphism . Also, we see that splits in if and only if is a principal ideal.
Example 1.1.4.
For the fields , we have and thus .
Example 1.1.5.
The simplest example of a number field with nontrivial class group is . Here, and .
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 . Recall that we have if and only if . Note that this is different from the splitting behavior in . This happens because is not trivial, and so we need both the condition that and that the are principal. By unramified CFT, we know that are principal if and only if they split in , and therefore if and only if splits in . In this case, is abelian, so we have a nice answer.
Example 1.2.1.
The primes of the form cannot be described in terms of a congruence condition. The field has and 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 has solutions. However, we can embed into the local field and then checking whether the polynomial has real solutions is very easy because we can do analysis. To try to recover all information about , we can embed for a prime . We then have the following slogan, known as the local-to-global principle:
We will study problems in by studying problems in all the local fields and .
2.1 Absolute Values
Definition 2.1.1.
Let be a field. An absolute value on is a function such that
- 1.
sends to and to .
- 2.
We have for all .
- 3.
For all , we have .
Example 2.1.2.
The usual absolute value on defines an absolute value in this semse. This induces an absolute value on usually denoted by . This is known as the archimedean absolute value on .
Similarly, any embedding or induces an absolute value on defined by .
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 satisfies the archimedean property: If there exists such that . This property fails for nonarchimedean absolute values because for all . In fact, is nonarchimedean if and only if the set is bounded.
Example 2.1.5.
Let and be a prime. Then define such that
Now for any , we define
Then we simply need to check that is a nonarchimedean absolute value on . Here, it is easy to check the ultrametric inequality, and this absolute value is called the -adic absolute value. By convention, we will choose and this is the normalized -adic absolute value on .
Example 2.1.6.
For any number field and prime ideal . Then we have a normalized -adic absolute value
where .
Definition 2.1.7.
An absolute value is discrete if is discrete under the usual topology on .
Example 2.1.8.
For a number field and prime , the -adic absolute value is discrete. On the other hand, is not discrete.
Definition 2.1.9.
Suppose be nonarchimedean. Then define
- 1.
. This is a subring of .
- 2.
Now define . This is a subgroup of of invertible elements.
- 3.
Set . This is the unique maximal ideal of .
Then is discrete if and only if is principal. In this case, a generator of is called a uniformizer. Then every can be uniquely written as for some .
Example 2.1.10.
(Non-example) Consider the field with -adic absolute value. This is not a discrete absolute value.
Definition 2.1.11.
An absolute value defines a metric on by for all . This induces a topology on where a basis of open neighborhoods of is given by open balls
Example 2.1.12.
In the -adic topology, we see that are closer under if and only if is smaller, which is equivalent to being larger, which is equivalent to being divisible by a large power of . In other words, for large.
Definition 2.1.13.
We say two absolute values on are equivalent, or if they induce the same topology on .
Theorem 2.1.14 (Ostrowski, 1916).
Let be an absolute value on .
- 1.
If is archimedean, then .
- 2.
If is nonarchimedean, then for a unique .
Remark 2.1.15.
Similarly, absolute values on a number field are given by
- 1.
for a prime ideal (-adic place);
- 2.
for some (real place);
- 3.
for some complex embeddings (complex place). Here, note that complex embeddings come in conjugate pairs.
Definition 2.1.16.
An equivalence class of absolute values on is called a place (or prime) of .
Remark 2.1.17.
When is a complex place it corresponds to a pair of complex embeddings , so we define
and this is the normalized absolute value for .
One reason for this normalization is
Theorem 2.1.18 (Product Formula).
Let be a number field. Then for all , we have
Remark 2.1.19.
When , let for . Then all but finitely many terms in the product are finite. Now it suffices to check this formula for and .
When , we see that and for primes , we have . Finally, we see that , so the formula holds. When , all absolute values are , so the product of all absolute values is trivial.
For a general number field, we can simply take the norm map to and check that it behaves well with respect to the places.
Theorem 2.1.20 (Weak Approximation).
Let be inequivalent absolute values on a field . Let . Then for all , there exists such that for all .
Remark 2.1.21.
This allows us to approximate any finite collection for inequivalent .
Remark 2.1.22.
As a sanity check, consider and suppose . Then given , we simply find such that , which is possible by the Chinese remainder theorem.
Remark 2.1.23.
More generally, if then one can choose such that and . Then if we consider as , the absolute value under approaches and under it approaches .
2.2 Completions
Consider the field equipped with the the absolute value . Then we can complete as a metric space to obtain the field . More generally, if is a field equipped with an absolute value (a valued field), then we will produce a general completion . Our aim is to produce a field that contains the original field and whose arithmetic is easier to understand.
Definition 2.2.1.
Let be a valued field. Then a sequence of elements in is called Cauchy if for all , there exists such that for all . We say that is complete if any Cauchy sequence in has a limit in .
Example 2.2.2.
Consider the sequence of integers . Clearly, this is not Cauchy under the usual absolute value on , but then if , we see that
so is Cauchy in . We then see that and thus the limit of the sequence is .
Example 2.2.3.
Consider . Then if , we have
and therefore is Cauchy in . We then see that
and therefore .
Remark 2.2.4.
In general, the limit of a Cauchy sequence may not exist.
Theorem 2.2.5.
Let be a valued field. Then there exists a complete valued field and an embedding of valued fields such that any other embedding of into a complete valued field factors uniquely through . In particular, is unique up to isomorphism and is called the completion of .
Proof.
Let be the set of all Cauchy sequences in under the equivalence relation where where . Then is equipped with termwise addition and multiplication and absolute value . Thus is a complete valued field.
To verify the universal property, we see that embeds and this satisfies the desired universal property. □
Definition 2.2.6.
Let be a number field and a place of . Denote by . When is a finite place, denote by the valuation ring . When is an infinite place, we see that .
Example 2.2.7.
Let . Then we see that and has a subring , which is a discrete valuation ring. Here, elements of have a nonnegative lowest power of .
Example 2.2.8.
If , then and for prime ideals are the completions of .
Remark 2.2.9.
Let be a nonarchimedean discrete valued field. Then is a complete discrete valued field and the valuation ring is the closure of in . Also, the maximal ideal is the closure of in . Finally, a uniformizer of is also a uniformizer of .
We also see that the natural map is an isomorphism. This tells us that we can approximate elements in up to using elements in .
Proposition 2.2.10.
Let be a discrete valued field. Let be a complete set of representatives of and be a uniformizer of . Then any element of can be uniquely written as where and .
Corollary 2.2.11.
Let . Then has a -adic expansion , where and .
Remark 2.2.12.
The main term of the -adic expansion is the lowest term . This is completely unlike the situation with the decimal digits of , where the highest power of is the main term.
Remark 2.2.13.
resembles but is more complicated arithmetically. When we add two power series, we simply add the coefficients, but addition in requires carrying. In addition, we see that but does not embed in .
Corollary 2.2.14.
We have an isomorphism
For example, we have
Proof of Proposition 2.2.10.
Let . Then for some and . Then we find the first digit by considering in . Then we replace and repeat the process. Repeating this, we simply use the completeness of 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 and then take the limit.
Theorem 2.2.15 (Hensel’s Lemma).
Let be a complete discrete valued field and be the residue field. Now let be a monic polynomial and let . Assume that in where are monic and coprime. Then there exist unique such that and .
Corollary 2.2.16.
Suppose has a simple root . Then has a unique zero such that .
Corollary 2.2.17.
If for , then has distinct roots in and hence distinct roots in . In particular, contains all -th roots of unity, so we have a map , called the Teichmuller lift.
Proof of Hensel’s lemma.
Let be arbitrary monic lifts. Then . Now inductiely we assume that there exist monic such that . We simply write for some such that and where . Then if and only if
but we can find such by Bezout’s lemma. Thus the desired exist, and we obtain by taking the limit. □
2.3 Extension of Absolute Values and Unramified Extensions
Let be a complete discrete valued field and be a finite separable extension of . Suppose . The main result is
Theorem 2.3.1.
Let be as above. Then
- 1.
extends uniquely to a discrete absolute value on ;
- 2.
is complete with respect to ;
- 3.
For all , we have
Remark 2.3.2.
To perform a sanity check, if , we have .
Proof.
- 1.
We first need to prove that a unique extension exists. We know that comes from a discrete valuation (hence is nonarchimedean), so let be the valuation ring. Then is a Dedekind domain. Then let be the integral closure of in . Then is also a Dedekind domain. But then any absolute value on extending comes from a maximal iddeal of lying above the unique maximal ideal . Therefore, we need to show that is a local ring.
To see this, assume not. Suppose there exist two prime ideals lying above . Let but . This implies that . Let be the minimal polynomial of . Then must satisfy for an irreducible (otherwise it has two distinct irreducible factors and Hensel tells us that the factorization can be lifted to ). This implies that
has a unique prime ideal, generated by , which contradicts our original assumption that were distinct prime ideals.
- 2.
Now we show that is complete. Let be a Cauchy sequence in . Choose a -basis of and write
But then each sequence forms a Cauchy sequence in . By completeness of , we can take and so we have
and thus is complete.
- 3.
Let be the Galois closure of . Then we know that also extends uniquely to . For any , the map is also an absolute value on extending . Therefore, by the uniqueness of the extension, we see that . This implies that
as desired. □
Corollary 2.3.3.
If is merely an algebraic and separable extension, then also extends uniquely to an absolute value on , but may fail to be discrete or complete.
Proof.
Note that is the union of all of its finite subextensions. □
Definition 2.3.4.
Let be a complete discrete valued field and be a finite separable extension. Let and be the valuation rings and be the maximal ideals. Next, let be the residue fields.
Define the ramification index to be the such that . Next, define the residual degree to be such that . Then .
Definition 2.3.5.
The extension is called
- 1.
Unramified if (which implies and );
- 2.
Totally ramified if (which implies and ).
2.4 Unramified Extensions
We will study unramified extensions. Here, we will try to understand via .
Proposition 2.4.1.
If is unramified, write . Then for any such that we have .
Proof.
Let be the minimal polynomial of . Then . But then we know that . But this implies that , so . □
Proposition 2.4.2.
If with minimal polynomial of given by such that has no repeated roots over , then is unramified.
Proof.
If is irreducible, then by Hensel’s lemma, we have where is irreducible. But then because has no repeated roots, we see that . But then we see that and thus is unramified. □
Proposition 2.4.3.
- 1.
Let be a tower of field extensions. Then is unramified if and only if and are unramified.
- 2.
Assume is perfect. If is unramified and is finite, then is unramified.
- 3.
Assume is perfect. Then if and are unramified, then is unramified.
Proof.
- 1.
Note that is unramified if and only if , which is equivalent to by multiplicativity of the ramification index.
- 2.
Suppose is unramified. Then let and let be the minimal polynomial of . Then the reduction is irreducible and . Because is perfect, has no repeated roots.
Because , let be the minimal polynomial of . Then and thus has no repeated roots, so is unramified.
- 3.
Consider the tower . Because are unramified, we know is unramified. This implies that is unramified. □
Theorem 2.4.4.
Assume that is perfect. Then there is an inclusion-preserving bijection
Moreover, is Galois if and only if is Galois and in this case.
Proof.
We prove surjectivity. Let be a finite extension. Write and let be the minimal polynomial of . Then any monic lift of has a root such that by Hensel’s Lemma. Then has residue field . Because is irreducible and is perfect, we know is unramified.
Now we will prove injectivity. Let be unramified with the same residue field . Then is also unramified with residue field . But this implies that
so we must have .
Now we will show the statements about Galois extensions. If is Galois, then preserves and and acts trivially on and . This implies that any induces . If and , then is Galois if and only if it contains , but this is equivalent to containing all conjugates of , which is equivalent to being Galois. Then the natural map is an isomorphism because the permutation on conjugates of induces the same permutation on the conjuages of . □
Corollary 2.4.5.
If is an algebraic extension (possibly infinite), then there exists a largest unramified subextension of . Moreover, is totally ramified.
Proof.
Let be the compositum of all finite unramified subextensions of . Then the residue field of is equal to the residue field of (otherwise, we can create an even larger unramified extension). This implies is totally ramified. □
Corollary 2.4.6.
Assume . Then for all there is a unique unramified extension of degree and .
Proof.
There is a unique degree extension of . □
Definition 2.4.7.
Define the Frobenius element when to be the generator of corresponding to the Frobenius map under . We will call this element , and for all .
Corollary 2.4.8.
When , the maximal unramified extension of is
In particular, we have
Proof.
We know , so is given by adjoining all coprime-to- roots of unity. □
2.5 Totally Ramified Extensions
Definition 2.5.1.
A polynomial is Eisenstein if
such that .
Example 2.5.2.
The polynomial is Eisenstein. Note that is totally ramified.
Proposition 2.5.3.
A finite extension is totally ramified if and only if where is a root of an Eisenstein polynomial.
Proof.
Let be the root of an Eisenstein polynomial. Then . Therefore and thus is totally ramified.
Now suppose is totally ramified. Let be a uniformizer of . Therefore and thus . Then have absolute values representing different cosets in . Thus the minimal polynomial of has degree . Moreover, if we write the minimal polynomial as
then and so that the required cancellation happens. But this implies that is a root of an Eisenstein polynomial. □
Proposition 2.5.4.
Assume . Then there are only finitely many totally ramified extensions of .
Proof.
Recall Krasner’s lemma: Let and assume is sufficiently small for all . If is irreducible, then so is , and
Therefore a totally ramified extension depends only on a small neighborhood of in the set
which is compact, so it can be covered by finitely many such small neighborhoods. □
Recall that if is a complete discrete valued field with residue field , there exists a unique unramified extension of degree . Together with the proposition, there exist finitely many totally unramified extensions of degree . This is of course false for number fields; for example, has infinitely many quadratic extensions.
Remark 2.5.5.
Krasner (1966) gave an explicit formula for the number of extensions of -adic fields of degree and an algorithm to construct the set of generating polynomials of degree . 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 that is locally compact under the topology induced by the absolute value.
Remark 2.5.7.
Recall that
- 1.
A topological space is compact if and only if open cover has a finite subcover.
- 2.
A topological space is locally compact if every point has compact neighborhood.
- 3.
A metric space is compact if and only if it is complete and totally bounded.
- 4.
A metric space is compact if and only if all closed balls are compact.
- 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.
The easiest examples of local fields are .
- 2.
If is archimedean and complete, then or .
Lemma 2.5.9.
Let be a complete discrete valued field. Then is locally compact if and only if the residue field is finite.
Proof.
Let be locally compact. Then is a closed ball. This means that is compact. If we consider an open cover
this has a finite subcover. But all of the are disjoint, so is finite.
Now suppose is finite. We show that every has a compact neighborhood. In particular, we will show that is compact and therefore that is compact. To do this, we need to show that is totally bounded. Choose and consider the open balls give a cover of as long as and is sufficiently large. By finiteness of , we know that is finite, as desired. □
Theorem 2.5.10.
Every local field is one of the following:
- 1.
or ;
- 2.
A finite extension of ;
- 3.
for a prime power .
Proof.
Suppose . Then . If is archimedean, then or . Otherwise, and by local compactness, must be finite.
If , then . Let be the residue field. Then
By Hensel’s lemma, we have and we thus have . Therefore , as desired. □
2.6 Statement of Local Class Field Theory
Recall that a field extension is abelian if it is Galois and is abelian.
Exercise 2.6.1.
If are abelian, then is also abelian.
Define to be the maximal abelian extension of . Equivalently, this is the compositum of all finite extensions of . Then has infinite degree, and classifying abelian extensions of is equivalent to understanding .
Definition 2.6.2.
Let be a possible infinite extension. We call Galois if it is algebraic, separable, and normal. Equivalently, 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 has a profinite topology with a basis of open neighborhoods of given by for finite subextensions .
Example 2.6.3.
Consider . Then
The open neighborhoods of are given by .
Remark 2.6.4.
If is finite, then is an open subgroup. In addition, is also closed, so it is clopen.
If is any extension, then
is a closed subgroup.
Theorem 2.6.5 (Galois correspondence).
Let be Galois. Then there is a Galois correspondence
Moreover, is Galois if and only if the corresponding closed subgroup is normal.
Remark 2.6.6.
If is not necessarily closed (for example, is not closed and its closure is ), then corresponds to under the Galois correspondence. In particular, , where .
This gives us the slogan, that when is a local field, can be understood in terms of .
Theorem 2.6.7 (Local Artin reciprocity).
There exists a unique (local Artin reciprocity map) such that
- 1.
For any finite abelian , there is a commutative diagram
- 2.
For any finite unramified and uniformizer of , we have .
2.7 Norm Subgroups
Let be a nonarchimedean local field.
Definition 2.7.1.
A subgroup of is a norm subgroup if it is of the form for some finite abelian extension .
Proposition 2.7.2.
Assume local CFT I.
- 1.
.
- 2.
if and only if .
- 3.
A subgroup of containing a norm subgroup is also a norm subgroup.
- 4.
.
Proof.
Recall that there exists a unique local Artin reciprocity map .
- 1.
Note that if , then , so clearly for , we see that . Conversely, if , then by local Artin reciprocity, we see that . But this means that , and thus . But this implies that .
- 2.
One direction is obvious. Assume that . Therefore
However, we know that
which implies that , so .
- 3.
Assume . Let . Then by local Artin reciprocity, we have a commutative diagram
This tells us that is also a norm subgroup.
- 4.
Note that is the largest subextension contained in both . On the other hand, is the smallest subgroup containing both , and the desired result follows. □
Corollary 2.7.3.
The map defines a bijection
The idea of local Artin reciprocity was to understand extrinsic data about extensions using intrinsic data about the group . 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 be a finite extension. If has finite index in , it must be open.
Proof.
Note that is continuous and is compact. Then ic compact and hence closed. But then , and thus is open (and closed). But this implies that is open (because is open), and thus must be open. □
Corollary 2.7.5.
If is finite abelian, then is a finite index open subgroup.
Theorem 2.7.6 (Local CFT II: local existence).
Every finite index open subgroup of is a norm subgroup.
Corollary 2.7.7.
We have a bijection
Remarks 2.7.8.
- 1.
This bijection also holds for archimedean local fields. For , the two extensions correspond to , while is algebraically closed and is the only finite-index open subgroup of itself.
- 2.
If is a finite extension of , then any finite index subgroup of is automatically open. However, this is not true for . In fact, if has finite index , then . Therefore, it suffices to show that is open. It is easy to see that for some . Therefore the equation has solutions in for by Hensel for and a stronger version for (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 is given by declaring a basis of open neighborhoods of to be the norm subgroups of , which are the same as finite index open subgroups in the usual topology.
Example 2.7.10.
Note that 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 to be the completion of under the norm topology:
Then the Artin reciprocity map induces an isomorphism .
Proposition 2.7.13.
We have an isomorphism as topological groups.
Proof.
Choose a uniformizer of . Then . Then a basis of finite index open subgroups is given by for some . This implies that
Corollary 2.7.14.
There is an isomorphism as topological groups for any choice of uniformizer . Therefore, we have a decomposition , where and is the maximal unramified extension. This means that is the totally ramified part of .
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 is dense, so is .
Now, it remains to prove local class field theory. First, we will construct the local Artin reciprocity map using Galois cohomology. After we prove that 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 be a group. Then a -module is a (left) module over the ring , or in other words, an abelian group with a linear (left) -action.
Example 3.0.2.
Let be a finite Galois extension of fields and . Then and are both -modules.
Example 3.0.3.
Any abelian group can be regarded as a -module under the trivial action.
Definition 3.0.4.
A homomorphism of -modules is a -equivariant group homomorphism, or equivalently a morphism of -modules. The set of such morphisms is denoted .
We will denote the category of -modules with -linear maps by . Because , 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.
is injective if the functor is exact. Dually, is projective if 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 -module is projective. In fact, .
3.1 Definition of Cohomology
Definition 3.1.1.
Let . Define its -invariants by
to be the largest submodule with trivial -action.
Example 3.1.2.
Let . Then if , Galois theory tells us that . Similarly, if , then .
Remark 3.1.3.
In other words, we have , where has the trivial action. This implies that the functor 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 . This will give us a long exact sequence.
Definition 3.1.4.
The group cohomology for any is defined by be the functor . This is the right derived functor of . It is characterized by
- 1.
.
- 2.
A short exact sequence induces a long exact sequence in group cohomology
- 3.
If is injective, then for all .
Remark 3.1.5.
More concretely, we can compute using an injective resolution of . If we consider an injective resolution
then we apply to obtain a complex
and then we have .
Remark 3.1.6.
We can also compute using a projective resolution of . If we consider a projective resolution
then we apply to this and obtain a complex
and then 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 .
Definition 3.1.7.
Define where acts on by
Note that is a free -module with basis . Now we define the morphism by
Lemma 3.1.8.
The previous definition gives a free resolution of in .
Definition 3.1.9.
This is clearly a complex. To prove exactness, let
Then we can check that . Then taking the image of both sides of , we obtain and thus .
Corollary 3.1.10.
We can compute .
Definition 3.1.11.
We have an identification
These are called the homogeneous -cochains of with values in and are denoted by . Then the differentials are given by
Then we have an explicit cochain description
Note that homogeneous -cocycles are determined by their values on elements of the form for , or equivalently on elements of the form . Therefore we may eliminate one degree of freedom.
Definition 3.1.12.
Define the group of inhomogeneous -cochains to be the group
Now we have an isomorphism given by
The differentials are given by
Now we can define the -cocycles and -coboundaries and the corresponding cohomology groups .
Example 3.1.13.
Suppose . 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 acts trivially on , then and is trivial. Thus .
Remark 3.1.15.
There is an explicit cochain description for , but computing using this definition is extremely tedious. Instead, we will try to break and into smaller pieces and then piece them back together.
3.2 Change of Groups
Definition 3.2.1.
Let be a subgroup and let . Define the induced module
with the action of given by
Example 3.2.2.
For any , we have .
Definition 3.2.3.
Let . Then define the restriction to be the same but viewed as an -module.
Proposition 3.2.4.
- 1.
(Frobenius reciprocity) For any , we have
- 2.
The functor is an exact functor.
- 3.
The functor preserves injections.
Proof.
- 1.
We will construct explicit mutual inverses. We will define
It is easy to check that these are inverse to each other.
- 2.
Now we need to show that induction is right exact. Suppose is a surjective map of -modules. Now let . Note that is uniquely determined by its values on a complete set of representatives . Then we lift to using the surjection . Now define for all , and now we obtain an element mapping to .
- 3.
Let be injective. Then is injective if and only if is exact. By Frobenius reciprocity, this is equivalent to exactness of , which is obvious. □
Proposition 3.2.5 (Shapiro’s lemma).
Let . Then for all .
Proof.
Choose an injective resolution in . But then exactness of induction and preservation of injectives imply that is an injective resolution in . But now we see that
Definition 3.2.6.
A module is called induced if for some abelian group .
Corollary 3.2.7.
If is induced, then for all , .
Proof.
By Shapiro, this reduces to computing cohomology , but then we know that is exact, so all higher derived functors vanish. □
Now we will consider functorial properties of group cohomology with respect to change of groups. Given and , we will define
- 1.
The restriction functor .
- 2.
The corestriction functor whenever .
- 3.
The inflation functor when is a normal subgroup of .
Suppose we are given and that are compatible in the sense that
for all . Then we obtain a morphism of cochain complexes
This is compatible with the differentials, so we obtain a morphism . Now using this generality, we can define the three functors.
- 1.
(Restriction) We will set and .
- 2.
(Corestriction) We will set and
This gives us the corestriction map by Shapiro.
- 3.
(Inflation) We take and .
Remark 3.2.8.
Suppose . Then the functors are
Proposition 3.2.9.
The map is given by multiplication by .
Proof.
Consider . Then this map is given by
Corollary 3.2.10.
If is finite, then is killed by .
Proof.
Take and apply the previous proposition together with the fact that all higher cohomology vanishes for . □
Corollary 3.2.11.
If is finite and is a finitely generated abelian group, then is finite.
Proof.
If is finitely generated, then so is because the cochain complex is finitely generated. But then 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 . Now assume that
is a coboundary. Thus there exists such that for all . But then , but then for all , we see that because fixes . Therefore is a coboundary.
- Composition is zero:
-
Clearly we have because the composition
is trivial because it passes through .
- Exactness on right:
-
Assume that . Then there exists such that for all . Now define
Now are cohomologous. We will now write as an inflation. Because , we know that factors through . Moreover, we see that
This implies that is valued in , so it must have been an inflation. □
Theorem 3.2.13 (Inflation-restriction, general case).
Let . Assume for all . Then we have an exact sequence
Proof.
We will proceed by induction on using a dimension-shifting argument. Consider the induced -module . Then look at the exact sequence
where we have the map . But then the higher cohomology of vanishes, so we have isomorphisms for all . But then , so we have isomorphisms for all . Similarly, because , we have an exact sequence
and therefore is also an induced -module. Therefore for all .
Now, by assumption, we have for all . Therefore, for all 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 is induced, then for all .
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 are nontrivial.
3.3 Group Homology
Definition 3.3.1.
Define the -coinvariants of by
This is the largest quotient module of where acts trivially.
Definition 3.3.2.
The augmentation ideal is defined to be
Then we see that . 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 of .
Remarks 3.3.4.
Group homology is characterized by
- 1.
;
- 2.
A short exact sequence in gives a long exact sequence
- 3.
If is projective, then for all .
Remark 3.3.5.
Let be a projective resolution. Then .
Proposition 3.3.6.
For any , we have .
Proof.
Consider the short exact sequence
This gives a long exact sequence
This becomes
But this tells us that . Now define the map
By the identity
we see that this is a group homomorphism. Because is abelian, the group homomorphism factors through . One can easily check that have an inverse given by . □
Remark 3.3.7.
This gives a homological interpretation of as .
Remark 3.3.8.
This is analogous to the topological fact that if is path-connected, then . In fact, for any group , we can define a space such that and for . We have , where the LHS is singular homology and the RHS is group homology. If we choose , we recover .
Example 3.3.9.
- 1.
If , then we have .
- 2.
If , then .
3.4 Tate Cohomology
The idea is that for a finite group , we will patch both group cohomology and group homology into a single theory.
Definition 3.4.1.
Let be a finite group and . Define the norm map
By definition, the image of the norm map lies in the invariants and , so the kernel always contains . This gives us a commutative diagram
and an exact sequence
In particular, this gives us
where just means a modified version of cohomology.
Definition 3.4.2.
For , define the Tate cohomology groups
Example 3.4.3.
Given a short exact sequence
in , 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.
If is induced, then for all .
- 2.
The dimension shift argument works for in both directions. If , then , where is the cokernel. Similarly, if , then , where is the kernel.
- 3.
Shapiro’s lemma still holds for .
- 4.
The property that is multiplication by holds for .
Remark 3.4.5.
Take to be a free resolution in . Taking the dual where . This gives a resolution . If we take , we obtain
Then we can compute .
Now we will compute Tate cohomology explicitly for finite cyclic groups . We can very explicitly write down the norm map, invariants, and coinvariants:
This implies that
Remark 3.4.6.
All remaining are determined by .
Theorem 3.4.7 (Period ).
Let . Then we have isomorphisms .
Proof.
Note that we have a free resolution of in given by
This computes the Tate cohomology and has period , so the Tate cohomology also must have period . □
Definition 3.4.8.
Define the Herbrand quotient
if both are finite.
Proposition 3.4.9.
If is an exact sequence, then .
Proof.
Consider the long exact sequence
This implies that
From this, we obtain
so it suffices to show that . But here, we see that
because Tate cohomology has period . □
Proposition 3.4.10.
If is finite, then .
Proof.
Note that we have the exact sequence
This tells us that . Next, we note that
is exact, so . □
Corollary 3.4.11.
If has finite kernel and cokernel, then .
Proof.
Consider the exact sequence
This tells us that
Because has finite kernel and cokernel, , so . □
Now we want to compute Tate cohomology for more general .
Theorem 3.4.12.
Let be a finite group. If for any subgroup we have , then for all .
Proof.
If is cyclic, then it follows from the period isomorphism. In the general case, if is solvable, we can induct on . If is normal such that is cyclic. By the induction hypothesis, we know for all . Applying inflation-restriction, we have an exact sequence
By assumption, for . By exactness, we know for , so for all because is cyclic. By exactness and the inductive hypothesis, we know for all . But now the hypotheses are invariant under dimension shift, so we obtain the desired result.
Finally, if is an arbitrary finite group, reduce to the solvable case. If we consider the Sylow subgroups , we know is solvable. Therefore for all . By an exercise from the homework, we see for all . □
Unfortunately, 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 be a finite group. If for any subgroup we have
- 1.
;
- 2.
,
then for all .
Example 3.4.14.
We will apply Tate’s theorem to construct the local Artin map. Let and and set . 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 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 . Recall that
We also know that
Proposition 3.4.16.
The cohomology group parameterizes extensions
such that the conjugation action of on agrees with the -module structure. Moreover, corresponds to the split extension.
Proof.
Given such an extension, choose a splitting of as sets. This gives us a section Then for all , we have for some .
We check that is actually a cocycle. To see this, note that
by associativity, which gives us
and therefore is a cocycle. We can check that any gives such an extension together with a section and that different choices of the section give coboundary relations. □
Definition 3.4.17.
For any , define its splitting module to be
with the action of given by
for any choice of . By convention, we write .
Remark 3.4.18.
By construction, we have an exact sequence
where .
Lemma 3.4.19.
The image of under is .
Proof.
Note that is the coboundary of . □
Remark 3.4.20.
The extension associated to
of is a split extension.
Proof of Tate’s theorem.
There are two steps to the proof.
- 1.
We know that with splitting module . We also know that . We will show that and . Consider the short exact sequence
This gives a long exact sequence
However, we know that , and . We also know that is the zero map (the restriction of to is a generator), so the map must be an isomorphism and therefore .
- 2.
Now we know that all Tate cohomology for all . Now we use the four term exact sequence
Because and for all , By dimension shift, we have . □
Remark 3.4.21.
There is an alternative description of as a cup product with some generator . Here, if , we can consider the tensor product as a -module with the action .
Now a cup product is a family of -bilinear maps
given by the following properties:
- 1.
The cup product is functorial in both .
- 2.
For , it is given by the natural map .
- 3.
If is a short exact sequence such that
is also exact, then for all and , we have .
- 4.
If is a short exact sequence such that
is also exact, then for all , we have .
The cup product exists and is unique. Given cocycles , we write
and this is unique by dimension shift. If is finite, then the cup product extends to Tate cohomology.
4 Local Class Field Theory
Now our goal is to understand for . 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 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 . We want for some . For any , construct
Because are distinct characters of for , we know that they are -linearly independent, so the map
is not the zero map and thus there exists such that . But now we compute
and thus , so . □
Example 4.1.3.
In the case where is a cyclic extension and , then , and thus , which means that if satisfies , then for some .
Example 4.1.4.
Let be the simplest quadratic extension. Let with . Then we know . If for , we see that
By Hilbert 90, if , then there exist such that
In particular, this tells us how to classify all Pythagorean triples.
Remark 4.1.5.
We may also consider infinite Galois extensions , for example . Then we know
This allows us to define continuous Galois cohomology
The can be computed using continuous cochains.
Remark 4.1.6.
More generally, if is a profinite group and is a discrete -module (which means is continuous with respect to the profinite topology on and the discrete topology on ), we may consider the continuous group cohomology
where the maps in the directed system are inflation.
Notation 4.1.7.
Let . We will simply write
for the continuous Galois cohomology.
Example 4.1.8.
Hilbert Theorem 90 implies that . Equivalently, we have . In other words, all line bundles on a point are trivial.
4.2 Second Cohomology
Let be a finite unramified extension of local fields. Recall that
where is the extension of residue fields. Our goal is to compute , where . The idea is to break into pieces, so choose a uniformizer . Then we have . Because is unramified, we may choose . Therefore, as a Galois module, we have .
Definition 4.2.1.
Let . Then write for all .
This gives us a natural filtration
with short exact sequences
and similarly
for .
Lemma 4.2.2.
We have for all .
Proof.
By Hilbert Theorem 90, we know . Because is cyclic and is finite, we can understand the Herbrand quotient. Therefore , so . Because is cyclic, we may apply period to obtain the desired result.
Now we consider . We know that is actually an induced module by the homework, so all Tate cohomology vanishes by period . □
Corollary 4.2.3.
The norm map and the trace map are surjective.
Proof.
We know that , so the norm map is surjective. Similarly, we know , so the trace is surjective. □
Lemma 4.2.4.
The norm map is surjective.
Proof.
Consider the commutative diagram
Given any , we want such that . Because , we may find such that . Now let . Because is surjective, we may find such that . Repeating this to infinity, we can find such that . Setting
we know
so . □
Corollary 4.2.5.
For all , we have .
Proof.
By the Lemma, we know that is surjective, so . But then we know is a direct factor, so . The desired result is simply an application of period . □
Theorem 4.2.6.
If is an unramified extension, then .
Proof.
We know 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 , so we have
Our goal now will be to extend this to .
Remark 4.2.8.
We always have an injective inflation map . 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 be a finite extension of degree . Then the diagram
commutes.
Remark 4.2.10.
Because , we may view as a subgroup of , and the inclusion is comparible with , so restriction makes sense.
Proof.
Recall that is computed by . Now we need to see the effect of restriction on the groups on the right of the chain of isomorphisms. Note that if and , then , so the diagram
commutes. Therefore, the maps on are given by multiplication by . Finally, for , we see that . Because , we see that . Therefore, the final map is simply . □
Corollary 4.2.11.
Let be a finite Galois extension of degree . Then contains a subgroup of .
Proof.
By Hilbert 90 and inflation-restriction, we have a commutative diagram
Therefore, injects into . □
Proposition 4.2.12.
We have . In particular, .
Proof.
We know this is true with is cyclic. We know and , so . In general, we will reduce to the cyclic case because is always solvable (because is a local field). We will induct on . Choose a tower of fields such that is cyclic. Applying inflation-restriction, we have an exact sequence
We know that because it is cyclic. We also know that by the inductive hypothesis, so . But we know that contains a subgroup of order , so . □
Corollary 4.2.13.
We have a commutative diagram
In particular, we may view .
Theorem 4.2.14.
There is an isomorphism .
Proof.
Recall that . Taking the limit over finite Galois extensions , we have an injection , but the composition is the identity, so all inclusions are equalities. □
Definition 4.2.15.
We can extend the invariant map to .
Remark 4.2.16.
For any field , define the Brauer group . This is in fact . The goal of local class field theory is to compute for a local field .
Remark 4.2.17.
Classically, the group classifies central simple algebras over . In fact, it classifies central simple algebras up to the relation if and only if for some . The group structure is defined by with identity .
Theorem 4.2.18 (Wedderburn).
Each class of is represented by a central division algebra over .
Example 4.2.19.
If is a local field, then we may define the invariant for any central division algebra over . If , then . Now if is the quaternion algebra over , we have . In fact there is an explicit construction of a central division algebra associated to for every . If is a quadratic extension, then
so .
Example 4.2.20.
Call the element with the fundamental class of .
Remark 4.2.21.
The local Artin reciprocity isomorphism can be realized as the cup product by the fundamental class .
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.
(Local Artin reciprocity) There exists a unique homomorphism
such that
- (a)
For any finite Galois extension , we have a commutative diagram
- (b)
For any finite unramified extension, , where is an uniformizer of .
- (a)
- 2.
(Local existence theorem) There is a bijection between norm subgroups of and finite index open subgroups of . In addition, if , then the finite index open subgroups of are precisely the finite index subgroups of .
4.3.1 Proof of local Artin reciprocity
To construct , we will use Tate’s theorem. Recall that by Hilbert 90, we have . We also computed , where . By Tate’s theorem, we have an isomorphism for all . We will specialize to the case where , and this gives us an isomorphism
Then we define to be its inverse. Taking the limit over all finite Galois extensions , we obtain a map
By construction, the property a) is satisfied.
To finish the proof, we need to check property b). Let be a finite unramified extension. Our goal is to explicitly describe the shift by isomorphism produced by Tate’s theorem. Recall that Tate’s theorem comes from and . Then we construct the splitting module for some by defining
Then we have short exact sequences
and
Now we need to trace through the isomorphisms
The first isomorphism is given by . Therefore it remains to write down the connecting homomorphism . This is described using the exact sequences
by lifting to and then taking . This will automatically land in . An obvious list of to is , so we need to compute . Therefore we have
Now it remains to describe this explicitly for .Recall that this comes from coming from the exact sequence
Now if such that , then we can lift to and take . This gives us . If , then we have
This gives us
and thus . □
4.3.2 Proof of Local Existence
We will assume that . We will establish a bijection
Recall that any subgroup of containing a norm subgroup is also a norm subgroup. Also note that any finite index subgroup of index contains . Therefore it suffices to show that is a norm subgroup. Our goal is to study the -th power may using the Kummer sequence
This is an exact sequence in for , so it induces a long exact sequence in cohomology
This gives us the Kummer isomorphism .
- 1.
In the simplest case, when , then the isomorphism becomes
where is the maximal abelian extension of exponent . Now the pairing
is in fact a perfect pairing. Therefore , so by local Artin reciprocity, we have the map
Also, we know , and therefore .
- 2.
In general, if is not contained in , then take . Now we can find such that . Now take Galois such that . Then
Therefore contains a norm subgroup, and is thus also a norm subgroup. □
5 Global class field theory
Recall that a number field is a finite extension of . The goal of global class field theory is to understand , or equivalently to understand finite abelian extensions of in terms of intrinsic data associated to . Analogously to the local case, we would like to construct a global Artin reciprocity map
which induces isomorphisms
Remark 5.0.1.
When is a local field, we took . When is global, we can again try to take , but in fact this fails.
Example 5.0.2.
Choose . Then , but
For example, any cannot be written as , and in fact the quotient is infinite.
Instead of , we will take to be a generalization of the ideal class group , which is the quotient of all fractional ideals by the principal ones. Thus the class group measures the failure of unique factorization in .
5.1 Idèles
Our generalization of ideals will be the idèle class group of , which is again given by
for some group . We need to define and .
Notations 5.1.1.
We will denote a (finite or infinite) place of by and the absolute value associated to such that
for all . Then we will denote the completion of with respect to by . When is a finite place, we write for the ring of integers, the corresponding maximal ideal, and for the completion of the maximal ideal.
Remark 5.1.2.
We may attempt to take . 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 . 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 with respect to .
Remark 5.1.4.
The idèles are equipped with the restricted product topology with a basis of open neighborhoods given by
where is open in for all and for all but finitely many .
Let be a finite set of places containing . Then define
Note the first factor is locally compact and the second product is compact, so is locally compact. Then we can write
and each is open in , so in particular is locally compact.
Remark 5.1.5.
The group has a basis of open neighborhoods of given by
Remark 5.1.6.
For all places , there is a natural injection
where goes in the coordinaate corresponding to . The induced topology on agrees with the usual topology.
Remark 5.1.7.
If is the group of fractional ideals of , then there is a natural surjection
with kernel given by
In particular, can be thought of as an enlargement of with extra information from the infinite places and units at finite places.
Remark 5.1.8.
The group is completely local.
Proposition 5.1.9.
The natural map
has discrete image.
Proof.
It suffices to show that is open in its imaage. Equivalently, for some and some , we have . Let . Then for all and for all . Note that for all . Therefore
On the other hand, by the product formula, we know unless . In particular, for sufficiently large and sufficiently small, we must have . □
Remark 5.1.10.
Note that has dense image for any single . However, the product formula implies that for , its absolute values at different places “repel” each other and makes have discrete image.
Example 5.1.11.
Consider the ring . Because is irrational, this has dense image. However, the embedding
has discrete image. In fact, the image is a lattice in .
Similarly, both have dense image, but
has discrete image.
Now it should not be surprising that has discrete image. This is simply an infinite dimensional generalization of this phenomenon.
Definition 5.1.12.
Define the idèle class group and equip it with the quotient topology. Because is discrete, this group is locally compact.
Remark 5.1.13.
The descends to . In particular, can be viewed as an enlargement of .
Definition 5.1.14.
Let be a finite extension. Recall that
given , where are the irreducible factors of after the extension. Therefore, for all , we have
Now we can define
Definition 5.1.15.
By definition, we have a commutative diagram
Therefore, we have an induced map .
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 is a finite Galois extension of number fields, a place of , and a place of above , we can define the decomposition group
Remark 5.2.1.
Suppose that is another place of above . Then are conjugate in . In particular, if is abelian, then only depends on . By local class field theory, we obtain
which only depends on .
Proposition 5.2.2.
There exists a unique continuous homomorphism
such that for any finite abelian extension and any place of , the diagram
commutes.
Proof.
Let . If and is unramified (true for all but finitely many places), then . Now is uniquely determined by a finite product
where runs over the places where or is ramified. Therefore, varying and taking a limit, this determines uniquely.
Now it remains to check that is continuous. In particular, we need the kernel to be open. Let be the set of ramified places of . Then by local class field theory, the diagram
commutes. Therefore , which is open. Therefore the kernel itself is also open. □
Theorem 5.2.3 (Global class field theory).
- 1.
(Global Artin reciprocity) The homomorphism satisfies
- (a)
. Therefore we obtain a global Artin map .
- (b)
For any finite abelian extension , induces an isomorphism
- (a)
- 2.
(Global existence theorem) For any finite index open subgroup , there exists a finite abelian extension such that .
Corollary 5.2.4.
We have a bijection between finite abelian extensions and finite index open subgroups of .
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 is a function
such that
- 1.
for all but finitely many places .
- 2.
if is a real place.
- 3.
if is a complex place.
Definition 5.2.6.
For a modulus , define the principal congruence subgroup
where and for for finite places. For infinite places, we have , and .
Definition 5.2.7.
For a modulus , define
Then define the ray class group to be
Here, the term “ray” comes from a positivity condition at real places, which geometrically means specifying a ray.
Definition 5.2.8.
Write . Then we have a description
In particular, if , then and . On the other hand, if , then
This is called the narrow class group.
Definition 5.2.9.
The abelian extension associated to under global class field theory is called the ray class field of . The case of is called the Hilbert class field and is denoted by with . The case where gives us the narrow Hilbert class field and is denoted by . This is characterized by . In particular, .
Remark 5.2.10.
By local class field theory, the ramified places of the ray class field of is contained in . In particular, is the maximal abelian abelian extension unramified everywhere, and is the maximal abelian extension unramified at all finite places.
Example 5.2.11.
If , then and thus .
Example 5.2.12.
If , then . However, we have
Therefore and .
Example 5.2.13.
Let and let for some . Then
If , then .
Now the ray class fields for either or are given by , where , and the field given by is . In particular, by varying , we obtain the Kronecker-Weber theorem.
Theorem 5.2.14 (Kronecker-Weber).
We have 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 , 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.
We will prove the first inequality2 that . This will be a consequence of the cohomology of . In particular, we will show that . This will imply that , but this is precisely the index we need.
- 2.
Next, we will prove the surjectivity of . This is an application of the first inequality and the weak Chebotarev density theorem.
- 3.
- 4.
We will prove that . This will follow from the determination of the Brauer groups of .
5.3 Cohomology of Idèles and first inequality
Let be a finite Galois extension of number fields. Recall that if is a place of , then . The action of on induces an action on the product by permuting the places above . If and , then .
Proposition 5.3.1.
Fix . Then there exists an isomorphism of -modules
where is the decomposition group.
Proof.
Recall that . For any , we define
Then it is easy to compute
Conversely, for , we define . This is well-defined by -equivariance, and we can check that these two assignments are inverse to each other. □
Corollary 5.3.2.
By Shapiro’s lemma, we have for all .
Similarly, we have
Corollary 5.3.3.
For all , and .
In particular, all of these are independent of the choice of . To stress the independence, we write , and so on and so forth.
Proposition 5.3.4.
We have the following cohomology of the idèles:
- 1.
.
- 2.
For all , .
Proof.
- 1.
We know that is -invariant if and only if is -invariant for all places of if and only if is independent of the choice of for all , and this is equivalent to .
- 2.
Let be a finite set of places of containing . Write
Then and therefore
If contains all places of that are ramified in , then for all , is unramified. By local class field theory, , so
Corollary 5.3.5.
We have and , where .
These statements follow from collecting all of the local information.
Corollary 5.3.6.
Assume is cyclic. Assume contains all places of that are ramified in and the infinite places. Then .
Proof.
By local class field theory, we have
Definition 5.3.7.
Let be a finite set of places of . Define the group of -units of to be
Our next goal is to compute and use this to prove the first inequality. First, we must compute the cohomology of units. Let be a finite Galois extension of number fields and be a finite set of places containing . Let be the set of places of above .
Proposition 5.3.8.
Assume is cyclic. Then
where .
Proof.
We need to use the following comparison result for Herbrand quotients from the homework: let be a cyclic group and an -module. If are two -stable lattices, then . We will apply this to the case where .
- 1.
Consider . This is clearly a -stable lattice. As a -module, we have
In particular, we have
- 2.
Now consider given by . If , Dirichlet’s unit theorem for says that is a lattice of rank in the hyperplane . Then is a -stable lattice.
Now we know that , so because the kernel of is the set of roots of unity in , which is finite. Now we obtain
as desired. □
Now we are able to compute the cohomology of the idèle class group .
Lemma 5.3.9.
For , we have .
Proof.
The short exact sequence in
gives a long exact sequence
where the last equality is by Hilbert 90. This implies that . □
We need to compute the Herbrand quotient of the idèle class group, so first we will rewrite in terms of -units.
Lemma 5.3.10.
Choose containing a generating set of prime ideals in . Then .
Proof.
Recall that we have a surjection
with kernel . In particular, we can write
By the choice of , we have
Theorem 5.3.11.
Assume is cyclic. Then . In particular, .
Proof.
Choose such that a generating set of prime ideals for . Then we can rewrite
Now we have
As an application of the first inequality, we will prove
Lemma 5.3.12.
Asssume is solvable. If there exists such that and is dense, then .
Proof.
Assume . Then we may find a tower such that is cyclic. Then we know . By local class field theory, is an open subgroup, so is also an open subgroup. Therefore is also closed. By density, we see that . In particular, , so , which is a contradiction. Thus . □
Proposition 5.3.13.
Assume is solvable. If all but finitely many places of split completely in , then . Equivalently, if , then there are infinitely many places of that do not split completely in .
Proof.
Choose to contain the set of all that do not split completely in . Also choose
Then if , it splits completely in , so is a trivial extension. Therefore . On the other hand, by weak approximation, any can be approximated by a global element , so is dense in . By the previous lemma, . □
Proposition 5.3.14.
Assume that is solvable. Then generate .
Proof.
Take . Choose . If is unramified in , then splits completely in . But this means that because only finitely many places are ramified, so . □
Corollary 5.3.15.
Assume is abelian. Then the Artin map is surjective.
Proof.
By local class field theory, if is unramified in , then . We know that the generate , so is surjective onto . □
5.4 Analytic aspects and second inequality
Recall that if is solvable and , then there exist infinitely many places of that do not split completely in . For example, if , then splits in if and only if . Thus there are infinitely many primes . 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 modulo and that the other half are congruent to modulo . 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 be a set of primes in . Define the natural density of to be
if this limit exists.
Definition 5.4.2.
Let be a set of finite places of a number field . Define the natural density of 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 exists, then exists and . 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 , choose to be the set of primes starting with the digit . Then does not exist, but , a result due to Bombieri.5
Dirichlet density is easier to study because we can use the tool of -functions, which generalize the Riemann zeta function.
Definition 5.4.5.
The Riemann zeta function is defined by
For a general number field , the Dedekind zeta function is defined by
These series converge when . When , .
Definition 5.4.6.
Let be a character. The Dirichlet L-function of is given by
where if . When , then .
Definition 5.4.7.
Let be a number field and be a modulus of . Let be a character. The Weber L-function of is given by
Again, we write if . These series converge absolutely when . When , and we have , so the Weber L-function recovers the Dirichlet L-function. Of course, when , we recover the Dedekind zeta function.
Theorem 5.4.8 (Analytic properties of Weber L-functions).
- 1.
If , then has an analytic continuation to all and .
- 2.
If , then has an analytic continuation to all and a simple pole at . The residue is given by the class number formula for .
- 3.
The function satisfies the functional equation relating the values at and .
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 satisfying
- 1.
The series has an Euler product .
- 2.
The series has an analytic continuation to except for poles for at .
- 3.
The analytic continuation satisfies a functional equation relating the values at and .
Corollary 5.4.10.
For any number field , (this means the difference is analytic at ).
Proof.
By the theorem, is analytic at , so . Because is analytic at , 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 , so we obtain the desired result. □
Corollary 5.4.12.
We can rewrite
Theorem 5.4.13.
Let be a finite Galois extension and be the set of primes of splitting completely in . Then
Example 5.4.14.
Let . Then we have .
Proof.
We can rewrite
because the terms with contribute to an analytic function and the second item on the right hand side is . 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 that split completely in . Now we will reinterpret the left-hand side as a certain density also. Recall that for a modulus of , we have the ray class group .
Theorem 5.4.15.
Let . Then let be an ideal class. Then
Proof.
Let be a character of . This is a quotient of , so it induces a quotient of . Now let 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 . Multiplying and summing over all , we see that
and thus , as desired. □
Remark 5.4.16.
Every finite abelian extension admits a modulus such that
is a quotient of the ray class group . In particular, we may reinterpret the global Artin map as a map given by .
Theorem 5.4.17 (Second inequality).
We have .
Proof.
Apply the previous theorem to and . Then we know
On the other hand, we know
But now we know splits completely in only if , which is the same thing as . Thus , 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 be a finite Galois extension of number fields. Let and be the conjugacy class of . 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 for some modulus . But then it is easy to see that
because is abelian. Now when is nonabelian, we will reduce to the abelian case. Let and write . Applying the abelian case to the abelian extension , write
We know . Now consider a subset of degree primes of and write
We know that . Finally, consider
Now we have a surjection and the fiber has size . This is because if with , then if and only if if and only if . Therefore if and only if . Then we know that
and therefore
Example 5.4.19.
Suppose . Then , so Chebotarev density tells us that
Example 5.4.20.
Choose and , where either is odd or . Then we know . If we choose , then , 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 and . This is the cubic extension with minimal , and here . Then . Now has three distinct conjugacy classes . These correspond to splitting completely in , splitting into a quadratic and a linear factor in , and being irreducible in , respectively. Now we can compute (possibly using a computer) the table
Chebotarev predicts that the three classes will occur in a ratio of .
Remark 5.4.23.
There is no simple congruence condition on to determine . However, there is a criterion on in terms of modular forms. For example, for , we may consider the function
Now some of the coefficients are
So we see that the conjugacy classses corresppond to the coefficients being . In fact, these are the character values for the two-dimensional irreducible representation of . By a miracle, we have !
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 , we can consider the idèle class group as a space related to . But now abelian extensions now correspond to -dimensional representations of , and the Langlands program gives a conjectural correspondence between -dimensional representations of and automorphic forms on . For example, when , the -dimensional representations correspond to modular forms relating and coefficients of the modular form .
5.5 Brauer groups and proof of global Artin reciprocity
We have already proved the first inequality that , at least when is cyclic. We have also proved the second inequality that in general.
Corollary 5.5.1 (Global version of Hilbert 90).
Let be finite Galois. Then .
Proof.
We begin in the case when is cyclic. Here, . However, , so .
Now it remains to reduce the general case to the cyclic case. First, we will consider solvable and induct on . Consider such that is cyclic. Then inflation-restriction gives us an exact sequence
Now the two outside terms vanish by being cyclic and the inductive hypothesis, so .
Finally, if 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 . We will do this in terms of the local Brauer groups.
Proposition 5.5.2.
The natural restriction map 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 , there exists a cyclic extension with for some such that is split by ().
Proof.
By the proposition, we know that is completely determined by its local restrictions . We also know that all but finitely many vanish. By local class field theory, we know that (the isomorphism is given by the invariant map). But now there exists such that for all . However, by functoriality of for , we obtain
by one of the homeworks. Once we prove the following lemma, we will be done. □
Lemma 5.5.4.
Let be a finite set of finite places of and . Then there exists a cyclic cyclotomic extension such that for all .
Proof.
First we reduce to the case . Here, we can just take the composite with and replace with . Also, we will reduce to the case where , wheere is some prime. Here, we can just take the composite of the extensions for all prime power factors of . We need for some cyclic cyclotomic extension . Now
Now we can form a cyclic cyclotomic extension
Now we can compute for all that
Where is the smallest integer such that . In particular, as we increase , , so as we increase , so we may choose such that . □
We have already proved that is surjective, so it remains to show that it factors through . We need to prove that
Theorem 5.5.5.
is contained in the kernel of . This means .
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 , we have
Equivalently, .
Our strategy will contain the following steps:
- 0.
We will prove Theorem 5.5.5 for .
- 1.
Next, we will prove Theorem 5.5.5 for cyclic cyclotomic.
- 2.
Next, we will prove Theorem 5.5.6 for cyclic cyclotomic.
- 3.
Next, we will prove Theorem 5.5.6 for finite abelian.
- 4.
Finally, we will prove Theorem 5.5.5 for finite abelian.
5.5.1 Step 0
Choose . We need to show that . It suffices to treat the case where is a prime power, so we compute explicitly . There are several cases:
- 1.
is finite. Then is unramified, so , where .
- 2.
. Here is ramified, so , where .
- 3.
. Here, .
To show that , it suffices to consider .
- 1.
If , then .
- 2.
If , then .
- 3.
Finally, .
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 , then it also holds for subextensions of and for composites , where .
Proof.
By functoriality of the local Artin maps, . Therefore if , 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 . □
5.5.3 Step 3
Every is split by a cyclic cyclotomic extension by Corollary 5.5.3. This means there exists cyclic cyclotomic such that . 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 . Denote this map by . Now the cup product with gives us a map . If is an extension of local fields, then for , we have a commutative diagram
If is an extension of global fields, we consider . Then we have a commutative diagram
| (5.1) |
5.5.5 Step 2
Assume that Theorem 5.5.5 holds for cyclic cyclotomic. Then the top row is trivial, so because is cyclic, the bottom row is also trivial (in the cyclic case, the shift in degree map is an isomorphism). But this means that satisfies , which is the same thing as Theorem 5.5.6 holding for cyclic cyclotomic.
5.5.6 Step 4
Assume that Theorem 5.5.6 holds for finite abelian. But this means the bottom row of (5.1) is trivial, so in particular, . But is arbitrary in , so . 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 be a number field. Then
Proof.
By Theorem 5.5.6, we know that is contained in the kernel. It remains to prove the reverse inclusion. Consider . We know there exists cyclic cyclotomic such that . Because 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 , there exists such that . □
Example 5.5.9.
Consider . Then we have a map
This is given by . Now by the theorem, we note that comes from if and only if . For example, we have
This means that quaternion algebras are in bijection with the set of where an even number of are not .
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 , there exists a finite abelian extension such that .
Definition 5.6.1.
A subgroup of is a norm subgroup if it is of the form for some finite abelian extension .
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 and . Then global Artin reciprocity implies that . Now we have a commutative diagram
by the fundamental theorem of Galois theory. By construction, , so in particular, 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 be a field and . Then if is the maximal abelian extension of with exponent , then by considering the Kummer sequence
we obtain a perfect pairing given by . This gives a bijection between finite subgroups of and finite abelian extensions of of exponent . Here, to a subgroup , we associate field given by adjoining all -th roots of elements in and to an extension we simply take the -th powers of all elements in . Here, it is easy to see that .
Example 5.6.3.
Let be a local field and assume . Then is finite and
Here, we know , so . Because contains a finite index subgroup isomorphic to , we have , as desired.
Now assume is a number field.
Lemma 5.6.4.
Assume . Assume that . Assume is large enough so that . Let such that for all and for all . Then .
Proof.
Let . Then we see that for all and is unramified for all . By local class field theory, in the first case we have , and in the second case . Therefore , so , and therefore , so . □
Proposition 5.6.5.
Assume . Let be a finite index open subgroup and assume has exponent . Then is a norm subgroup.
Proof.
Let and
Then let . Because has exponent , we know for sufficiently large. It remains to show that is a norm subgroup. Enlarging we may assume that .
Let be the -units. By the previous lemma, . Set . This is a finite abelian extension of exponent unramified away from . By local class field theory, we know for all and for all . But now and thus . Now we show that .
We will show that . But now we have
Now we compute
We also compute
by Dirichlet’s unit theorem. Thus . On the other hand, we have
using Kummer theory and the Dirichlet unit theorem. This implies that . □
To finish the proof of global existence, we reduce to the case of Kummer extensions.
Lemma 5.6.6.
Let be a finite index open subgroup and for some finite extension . If is a norm subgroup, then so is .
Proof.
Write for some finite abelian . Now let be the maximal abelian subextension of . Then
Therefore is a norm subgroup. □
We are now able to complete the proof of global existence. We will induct on and assume . If , choose containing such that . Then is a norm subgroup, so write . Now has image with kernel . Then , and by the inductive hypothesis, is a norm subgroup, so is also a norm subgroup.
If , then take . Now . Then is a norm subgroup by the above, so by the lemma, is also a norm subgroup.
5.7 Primes of the form
In this section we discuss a classical application of class field theory. This was discussed at the beginning of these notes. Recall that if with discriminant , then the splitting behavior of
is determined by the Legendre symbol .
Example 5.7.1.
Consider with . Then splits in if and only if , which is equivalent to , is equivalent to by basic arithmetic.
On the other hand, is a PID, so splits in if and only if , which is equivalent to for integers . Combining the two equivalences, we see that if and only if .
Example 5.7.2.
Consider with . Then splits in if and only if , and by quadratic reciprocity this is equivalent to . However, is no longer a PID, so we have a different criterion for .
Theorem 5.7.3.
If , then if and only if .
Some examples are given below.
Similarly, we have
Theorem 5.7.4.
If , then if and only if .
Some examples are given below.
Our goal is to explain the proof of these results using class field theory.
Lemma 5.7.5.
Let be a finite prime for a number field . Then is principal if and only if splits completely in , where is the Hilbert class field of .
Proof.
By global Artin reciprocity, . But then is principal if and only if is trivial. However, this is equivalent to being trivial, which is equivalent to splitting completely in . □
Proposition 5.7.6.
- 1.
If , then if and only if splits completely in .
- 2.
If , then if and only if splits completely in .
Proof.
- 1.
If , then if and only if splits as a product of principal ideals in . By the lemma, this is equivalent to splitting completely in , and combining, we see this is equivalent to splitting completely in .
- 2.
This is the same argument, except when . □
Therefore, to prove the main theorems, it remains to prove the following result.
Proposition 5.7.7.
- 1.
splits completely in if and only if .
- 2.
splits completely in if and only if .
The key to this result is that in these cases, is an abelian extension, so we may apply global Artin reciprocity. Recall that for , . Then under the Artin reciprocity map, if and only if splits completely in .
Example 5.7.8.
When , we have , so we recover the result that if and only if splits in .
More generally, the global existence theorem gives a bijection between finite index open subgroup of and finite abelian extensions of . In particular, we obtain a bijection between finite subgroups of and finite abelian subextensions of . This is given by . Thus we see that if and only if splits completely in .
Example 5.7.9.
Let . Then , and so the possibilities for are . Clearly the trivial subgroup corresponds to and corresponds to , and corresponds to . By global class field theory, this is the only abelian subextension of .
Note that because quadratic extensions are abelian, every quadratic extension is contained in a cyclotomic extension. For example, .
Example 5.7.10.
Consider . Then . Now the subgroups and corresponding subextensions are given below.
This gives us the following diagram of field extensions:
In particular, we see that if and only if splits completely in .
Example 5.7.11.
When , then . Then we see that corresponds to and corresponds to , so corresponds to . Therefore if and only if splits completely in .
Remark 5.7.12.
When is abelian, class field theory gives us a congruence condition for primes splitting completely in . On the other hand, when is nonabelian, then such a congruence condition does not exist.
Lemma 5.7.13.
Denote by to be the set of primes of splitting completely in . Then if and only if . In particular, if and only if .
Proof.
One direction is clear. In the other direction, suppose . Then . But now Chebotarev density tells us that , and this is only possible if , which means . □
Proposition 5.7.14.
If is nonabelian, then there does not exist a congruence condition for splitting completely in .
Proof.
If there exists such a congruence condition, then choose . Thus , and therefore . By the criterion, . But this means that , so . This implies is abelian. □
Example 5.7.15.
If , then , so with . Thus there is no congruence condition for .
Remark 5.7.16.
Even though there is no congruence condition, non-abelian reciprocity gives us a condition using coefficients of modular forms.