Learning Algebraic Number Theory: Sam Ruth March 31, 2010
Learning Algebraic Number Theory: Sam Ruth March 31, 2010
Learning Algebraic Number Theory: Sam Ruth March 31, 2010
Sam Ruth
March 31, 2010
1 Introduction
After multiple conversations with all levels of mathematicians (undergrads, grad
students, and professors), I’ve discovered that I’m confused about learning mod-
ern algebraic number theory. Both to clarify what I need to do for myself and to
help younger mathematicians (which at this point, is only undergrads) to have
more focus than I had, I’ve decided to compile a list of what I think is the rel-
evant background. I am obviously not as valuable a resource on this subject as
an actual researcher in algebraic number theory, but as much as grad students
are a bridge between undergrads and professors, I intend to fulfill that purpose.
I know less, but I’m more didactic. If you are someone that knows better and
disagrees with what I’ve written, I invite your comments. In each section, I’ve
listed some jargon that shepherds the area, as well as texts that I’ve read or
intend to read.
2 Basic Math
As I’m writing this essay primarily for advanced undergraduates or beginning
graduates, I’m assuming most of the material I discuss here has already been
encountered. I will use this section merely to highlight which things deserve
extra focus, especially if you make the mistake I did of thinking that being a
number theorist means there are other branches of math you can safely ignore.
1
theorem is fun and important. The more you know about quadratic forms, the
better. I really enjoyed continued fractions and Pell’s equation, though I haven’t
reencountered them in grad school. w. And of course, quadratic reciprocity is
the crowning glory of this all
Texts: Hardy and Wright is a good read, though it’s hard to find a copy.
You’d have a strong background if you read that whole book. Apostol’s Analytic
Number Theory is a little dry, but it will tell you about connections between
the arithmetic functions, the Riemann hypothesis, and prime numbers.
2
2.4 Undergraduate Topology
It’s unbelievable to me how important topology is in number theory consid-
erations. First of all, a course in abstract point set topology will be helpful,
because when doing number theory, and especially algebraic geometry, you’ll be
dealing with some weird topologies (i.e. non-Haussdorf). Secondly, in the class
field theory, cohomology becomes the natural way to define maps. I’ll say more
about this in the graduate topology section; just know that you’ll get plently of
mileage out of your semester of topology.
Texts: Everyone loves Munkers, and I don’t know any sensible alternative.
3
Let me begin by laying out what I see as the broad picture, before breaking
it down into where I studied each individual pixel. Some of the terms I use here
may seem unfamiliar, but I intend to reference them all again in the appropri-
ate subsection below. My rough understanding of Langlands is this. There are
two kinds of L-functions. One kind immediatly comes from number theory or
sometimes geometry-Dirichlet and Dedekind zeta functions, L-functions count-
ing points on elliptic curves, or more generally L-functions that count some-
thing. The other kind come from representation theory. For instance, modular
forms are some kind of representation of SL(2, Z). Associated to these modular
forms are L-functions. There are lots of special functions to consider, and you
certainly need to look at groups bigger than SL(2, Z). Here I’m running up
against the edge of what I know, but you actually move beyond matrix groups
and Lie groups, at least as far as algebraic groups, and almost certainly past
that. Then the conjecture is that these two kinds of L-functions are the same.
Every number theory L-function corresponds to some representation theory one.
The connection is more sophisticated than that, however. This is what the term
functoriality means. You can combine L-functions on one side, and that ought
to somehow correspond to combining them on the other side. For instance, if
you have L-functions corresponding to field extensions (Dedekind), you can take
the composite field extension and now you have a new zeta function that is in
some sense a combination of the previous two. Then there ought to be three
corresponding L-functions on the other side, with the third function somehow a
combination of the first two. Anyway, to even make sense of all these functions,
there seems to be a fair amount of necessary background, at least enough to
keep me busy well into the foreseeable future.
4
In general, it is hard problem to find the ring of integers (though it is easy to
do so for quadratic extensions). What is true though is that the ring of integers
is a free module over Z of dimension n = [K : Q]. In fact, much more can be
said about the ring of integers as a ring.
A commonly asked test or generals
√ question is to name a ring that is not a
UFD. The standard answer is Z[ −5]. This is a standard mathematical theme:
we can gain something by looking at a bigger object, but we lose something as
well. For instance, by moving from R to C, we gain algebraically closed, but
we lose our ordering. Similarly, in ring of integers over Z, we gain solutions
to a polynomial, but we lose unique factorization. Note that factorization is a
problem dealing with our other theme, the structure of the primes. While we
can no longer uniquely factor elements into prime elements, what we can do is
uniquely factor ideals into prime ideals. If we start with a prime ideal (p) in Z,
it may or not stay prime-that is, there may be bigger prime ideals containing
it, and it may be contained in more than one prime ideal. For instance, in the
ring of integers Z[i], there are exactly three things that can happen to a prime
p of Z. Note that here by (p) I mean pZ[i], that is,the ideal generated by p in
this bigger ring.
In the first case, p is said to ramify, in the second case it splits, and in
the third case it is inert. Note that what happens to p in Z[i] has to do with
whether −1 is a square (mod p). To know whether −1 is a square (mod p),
√
we use quadratic reciprocity. In general, the behavior of (p) in Z[ q] is related
2
to whether x − q splits (mod p).
The idea of primes not remaining prime in a ring extension is an idea that is
useful in much more generality than just finite extension of Q, which are called
number fields. This is an example of a map between rings (in this case, the
inclusion map Z ⊂ Ok ) giving rise to a map between prime ideals. These maps
are fundamental to algebraic geometry. This also forms the beginning of the
connection between number fields and function fields. A (relatively) concrete
example of this is the case of p-adic numbers Qp , where p is a prime number.
These can be defined in several ways, but I think the most staightforward is to
think of them as formal power series, that is they look like
∞
X
α= αi pi ,
i=−n
5
called a discrete valuation ring. One way of thinking about the p-adic numbers
is that powers of p are getting close to 0-this is some explanation for why the
power series should be thought of as converging, because the high powers of p
are small. One reason we study p-adics is because they can tell us information
about Q. Here’s an example that’s encountered early in the study of p-adics.
a1 x21 + · · · + an x2n = c
This result characterizes the kinds of results we aim for with p-adics. We
can reduce global questions (solutions over Q) by looking locally (solutions over
R and Qp ). With a bit more development of the finite extensions of Qp , we can
say even more about number fields. A finite extension of Qp is called a p-adic
number field.
The heart of late 19th, early 20th century number theory is the classification
of abelian extensions of Q. An abelian extension is a Galois extension whose
Galois group is abelian. There are multiple approaches to developing this theory,
but in any case, it goes by the name of class field theory, and has a local and
global component. In the local case, it’s easy to state, and hard to say what the
maps are explicitly.
Theorem 3.2. (Local Class Field Theory) There is a bijection between abelian
extensions of Qp and open subgroups of Q∗p .
6
Connected to the class group, although it’s not at first obvious why, is the
idele class group. The adeles of a number field K. are the elements of the direct
product Πp Kp , where the product is over all prime ideals, and the coordinate in
the p place is a integer for all but finitely many p. Addition and multiplication
are defined componentwise. The ideles are a subring of the adeles where almost
every (all but finitely many) of the coordinates are units.
Note that K sits inside all of its completions; there is a embedding of K
into Kp for every p. Under this embedding, an element α ∈ K is a unit for
almost every p (because there are only finitely many primes that divide the
numerator of α). This means that every element of K gives rise to an idele.
The image of K in AK is called the principal ideles. The ideles modded out by
the principal ideles form a group, called the idele class group. The ideal class
group is isomorphic to a quotient of the idele class group.
Theorem 3.4 (Global Class Field Theory). There is a bijection between open
subgroups of the idele class group of K and finite abelian extensions of K.
Texts: I started by trying to read Lang’s Algebraic Number Theory, which
was painful, though I think it helped. In retrospect, I should have started with
Marcus’ Number Fields or Samuel’s Algebraic Integers. These both explain,
concretely, why these algebraic constructs should be of any interest to a number
theorist, and Samuel has lots of exercises to familiarize yourself with decompo-
sition group, inertia group, and the like. I eventually read Neukirch’s Algebraic
Number Theory (I think that’s the title), which I would highly recommend. He
does a lot of things with commutative diagrams and exact sequences, which is
annoying, but supposedly better generalizes to higher dimension. I started with
almost no algebraic number theory background and I finished this book in one
semester, so it’s not too serious of a time commitment.
7
homomorphism and every element in G has finite order, the image is actually
contained in µφ(n) , the φ(n)th roots of unity. Extend this function to all of
Z/nZ by letting χ(m) = 0 if (m, n) > 1. Now we can turn this into a function
of Z by composing Z → Z/nZ → C. Now define the function
∞
X χ(n)
L(s, χ) = .
n=1
ns
If we picked the right χ (we want it to be primitive), then this function will be
nice. Specifically, if n = p, some prime p, then all the χ are primitive. By being
clever about adding together the different χ corresponding to a fixed p, we can
get a sum that only picks out terms n that are congruent to a (mod p). This is
how one shows that X X 1
L(χ, s) ∼ s
,
χ
q
q≡a (mod p)
where the sum on the right is over prime q. Then since the sum on the left
diverges to ∞ as s → 1+ (it includes a term that is essentially the Riemann
zeta function), there must be infinitely many q ≡ a (mod p). This is how one
proves
Theorem 3.5 (Dirichlet’s Theorem on Primes in Arithmetic Progressions). If
(s, r) = 1, then there are infinitely many primes of the form s + nr, and
This can be reinterpreted as summing over all ideals in Z. The reason for this
reinterpretation is that it allows us to define a L-function for a number field
X 1
ζK (s) = .
a
N as
This requires some work to define the norm of an ideal. A prime ideal p ⊂ OK
sits over some prime ideal (p) ∈ Z, and has some inertia degree, f = [OK /p :
Z/(p)] Define N p to be pf . Now as the set of ideals is the free abelian group
generated by the prime ideals, this definition can be extended to an arbitrary
8
ideal. Note that information about this ideal, the inertia degree, is included in
the definition of the Dedekind zeta function. Now, the study of the Riemann
zeta function really began when Euler noticed it prime factorization
Y −1
ζQ (s) = 1 − p−s .
pprime
The Dedekind zeta function can be split into sums among the elements of the
class group; the fact the these sums turn out to be roughly equal, regardless
of the choice of element of the class group, allows one to say that the primes
in OK are evenly distibuted among the cosets of the principal ideals. Like the
Riemann zeta function, the Dedekind zeta function has a simple pole at s = 1.
The residue of the subsums at s = 1 can be computed in terms of arithmetic
information about the ring of integers; then the sum of the residues is equal
to hK , the class number, times this data. In this way, the L-function can be
used to compute the class number. The L-functions somehow can tell you a
lot about the field then. In the case of quadratic fields, their is a connection
between the Dirichlet, Dedekind,
√ and Riemann functions. A quadratic extension
is necessarily of the form Q( d) for some square-free d ∈ Z. Let χ be the unique
quadratic character defined (mod d). (Quadratic means that χ2 = 1, so χ is
real-valued). Then
ζK (s) = ζQ (s)L(s, χ).
The L-functions quickly become more complicated. As already mentioned,
there are the Hecke and Artin L-functions. Dirichlet L-functions arise from
characters on the groups Z/nZ. Hecke L-functions arise from characters on
the ray class group (Fix an ideal m. Define Jm to be the set of ideals coprime
to p, that is, their prime factorizations do not include any of the prime ideals
in m. Define Pm to be the set of principal ideals generated by α such that
α ≡ 1 (mod m). The group Jm /Pm is called the ray class group with respect to
m). The Hecke L-functions are defined based on these characters. There is (at
least) a hard way and an easy way to prove these have analytic continuations
and functional equations. The hard way is how it was originally done. The new
easier way is one of the main results of Tate’s thesis, which explains how these
global L-functions are products of local L-functions.
We also have Artin L-functions; these are given by taking representations
of the Galois group. In case the Galois group is abelian, these coincide with
the Hecke L-functions. From a basic theorem in representation theory, Artin
L-functions can be written as rational functions of simpler L-functions, which
allows one to show they have a functional equation and they are meromorphic.
It remains an open problem to show they are in fact entire.
9
Beyond these, there are the L-functions coming from other special functions.
These include L-functions from counting the points on an elliptic curve, the
L-functions coming from the Fourier coefficients of modular forms, and the L-
functions coming from the eigenvalues of the Laplacian on hyperbolic space. I’m
not entirely clear about these L-functions, so I’ll say a little here about modular
forms and devote a later section to elliptic curves.
A modular form is a holomorphic function on the upper half plane of the
complex numbers that acts nicely under the action of fractional linear transfor-
mations. More precisely, a modular form of weight k is a holomorphic function
on H = {z ∈ C|Imz > 0} such that
az + b
f = (cz + d)k f (z),
cz + d
for a, b, c, z ∈ Z with ad − bc 6= 0, which is also meromorphic at infinity. I’m still
learning the various interpretations of these conditions in terms of differential
forms and differential operators and functions on lattices and elliptic curves, so
I can’t really get into that. There are however several interesting facts about
modular forms. For one, the space of modular forms for any fixed weight k is
finite dimensional (in fact, 0 dimensional if k is odd). The easiest example of a
modular form of weight k are the Eisenstein series
X 1
Ek (z) = .
(m + nz)k
−∞<m,n<∞,(m,n)6=(0,0)
What is in fact true is that the space of modular forms of any given weight
is generated by E4 and E6 . Furthermore, since by the modularity P condition
f (z + 1) = f (z), the modular forms has Fourier series, f (z) = an e2πinz .
−s
P
We use the Fourier coefficients then to make a L-function an n . Again,
because of the modularity conditions and other properties satisfied by the mod-
ular forms (including their relationshisps with the Hecke operators), these L-
functions satisfy certain properties similar to those encountered before (such as
a multiplicativity of the Fourier coefficients). I’ll update this as I learn more
about it.
TALK ABOUT MODULAR FORMS FOR OTHER CONGRUENCE SUB-
GROUPS
Texts: I cheated by learning the analytic theory the algebraic way. I think
again that Serre’s A Course in Arithmetic is helpful, as he works through the
baisc facts about modular forms. The Neukirch book I mentioned earlier will
tell you about the different kind of number theory L-functions. At some point,
you’ll have to read Tate’s thesis.
10
The more involved way is the representation theory of different kinds of Lie
groups. As I discussed earlier, I think this plays a large role in modern Langlands
philosophy, but I don’t know enough about specific examples to say how exactly
it comes up. What I do know is that there are two special cases that are worked
out first, compact connected Lie groups and simple Lie algebras. Here again
I’m probably not qualified to say any more than that.
Texts: For the finite stuff, I think Serre’s Linear Representations of Finite
Groups is great. For Lie groups, I’ve read three different texts that have com-
plemented each other well. Humphreys is very pure in its algebraic study of
the Lie algebras. It proves everything but it can be difficult to follow what’s
going on. Fulton & Harris is the opposite extreme-it pushes all the proofs off to
the appendix (or doesn’t do them at all), but is loaded with examples and tells
you the general algorithm for studying the representations of a semi simple Lie
algebra. I don’t think the phrase Lie group appears at all in Humphreys, and
Fulton & Harris is pretty cavalier that taking care of the Lie algebras should
be enough. To study the Lie groups, Adams’ Lectures on Lie groups is great.
It’s a little archaic in some of its terminology, and some parts of proofs (like the
Peter-Weyl theorem) are pushed out to other sources. However, it’s really the
groups that you care about, so it’s necessary to read.
11
Also, like other texts in algebraic geometry, you have to be careful to notice
when he’s using a piece of commutative algebra. The first volume deals with
varieties, and I think is pretty helpful for that. I also read the beginning of the
second volume, where he begins discussing schemes. A big asset of these books is
that they have exercises. I benefited from reading Miranda’s Riemann Surfaces,
which lays out the definitions quite clearly in the case of curves. I would advise
anyone interested in learning their algebraic geometry about curves first, before
trying to learn it in the abstract. Everyone says eventually you have to read
Hartshorne. Here’s a three step process to mastering Hartshorne
• You’re screwed.
This book is extremely difficult to read, and it should definitely be read with
company and other texts. I’m now reading through Ravi Vakil’s online notes on
Algebraic Geometry, which I’m finding very helpful. In particular, he develops
the category theory and the commutative algebra as he needs it.
12
can graph over the real numbers-we get a circle. It’s also an equation where we
can easily find a solution, for instance (−1, 0). Now the clever part. Suppose
(x, y) is another rational solution to the equation. Then the line connecting
(−1, 0) and (x, y) has rational slope. It is also true that if we take a line with
rational slope, it will intersect the circle at a rational point. This is because
the finding the intersection of a line and a circle will give us a quadratic in one
variable, which we can solve with the quadratic equation. Since we know one
solution is rational, x = −1, the other one must be as well. Therefore, we have
a bijection between the set of rational slopes Q, and the rational points on the
curve. Note that this bijection depends on the 2 choice of
initial point. It we
t −1
take the line with slope t, we find the point 1+t2 , 1+t2 . Letting t = m
2t
n , we
come back to a well-known parametrization of the integer Pythagorean triples,
x = m2 − n2 , y = 2mn, z = m2 + n2 .
In this analysis, there was something special about the fact the our equation
was quadratic-namely, it let us say that we had one rational point, we could find
the rest of them. Increasing the degree by 1 makes this question much harder.
As before, we assume we have one rational point. In fact, this is usually given as
the other condition of an elliptic curve-that a base point be given. If our curve
has a point over Q, we make perform a change of variables and move this point
to (0, 0) (here already we can apply to algebraic geometry, saying this map is a
rational map and therefore doesn’t change any of the essential properties of the
curve). In fact, over Q, we can write the equation as y 2 = x3 − g2 x − g3 . Again
we want to find the rational points on this curve. However, simply drawing
a line with rational slope won’t work this time-most lines should intersect the
curve at three points, but there’s no guarantee the other two will be rational
(or even real).
The rational points on an elliptic curve have an algebraic property, however,
that we can exploit. We can in fact make a group law on the points that turns
them into an abelian group. Again, the most natural way to define this law is
through algebraic geometry, but it can be defined, albeit mysteriously, through
simple planar geometry. For any two points P, Q on the curve, draw the secant
line connecting them (if P = Q, take the tangent line through P ) This line
will intersect the curve at a third point. Because of the equation, the curve
is symmetric with respect to the x-axis. Take the reflection of the intersection
point with respect to the x-axis. This point will be defined as P + Q.
FINISH DEFINING GROUP LAW ( IDENTITY) From the equation of
the elliptic curve, we can compute explicitly what P + Q is in terms of the
coordinates of P and Q. In particular, we find that P + Q has coordinates that
are rational functions of the coordinates of P and Q, so if P, Q are rational
points, so is P + Q. Thus our operation is on the set of rational points.
To make this a group, we need to identify the identity, additive inverse, and
verify the associative law. Again, these seem more natural when given from the
standpoint of algebraic geometry, but can still be explicitly discussed without
it. The identity is the point at infinity of the curve; intuitively, you can think of
the point at infinity as being at x = 0, and y = ∞. Drawing the secant line from
13
a point P and the point at infinity O is then a vertical line going through P ;
the group operation described then gives you back the point O. It is also clear
what the inverse should be-you take −P to be the reflection of P with respect
to the x-axis. Finally, with the explicit formula, it is straightforward, though
computationally heavy, to verify the operation is associative. It is obviously
commutative.
Now that we know the rational points form a group, we can ask what sort
of group they are. Though there are some positive results in this area, there are
also some frustrations. To begin with,
14
Texts: Tate and Silverman’s Rational Points on Elliptic Curves is a great
introduction-it barely assumes you know Galois theory, and gives some back-
ground. After that, Silverman’s The Arithmetic of Elliptic Curves is good-it
develops the theory much more fully, and again gives you something to do with
all that algebraic geometry you’ve been learning I’m not sure what I’m going
to read next-perhaps the other Silverman book.
3.7 Cohomology
It’s extremely surprising to me how important topology considerations are to
number theorists. Then again, I think this was surprising to everyone until
about 60 years ago. This comes up in a wide variety of ways, which seem
fairly disjoint to me. I’ll point some of them out here, and add that to call my
knowledge of this area rudimentary would be to pay me too great a compliment.
Algebraic Number Theory-The way to define the bijection in the case of
local class field theory turns out to be through the group cohomology of the
Galois group. The first time I saw this, I didn’t realize it was a topological
construction, but then after I learned some more homological algebra I found
the construction more natural.
Algebraic Geometry-I’m told that when considering Spec of a ring, and
schemes, their cohomological properties are important. So I’ll have to learn
more about that. I think this is where étale cohomology comes in. I do know
that part of the proof of the Weil conjectures is relating points over finite fields
to cohomology of function fields. There are many functors in algebraic geometry
that are only left exact or right exact, so we try to compute the cohomology.
Representation Theory-When you make the move from representations of
finite groups to representations of compact groups, you suddenly restrict your
attention to unitary representations. My understanding is this is largely because
you need the extra Hilbert space structure to get any nice theorems about the
representations. I believe you care about the topology of something involved
here as well; in particular, I think for parts of the Langlands program you look
at the cohomology of Lie groups.
Texts: Hatcher still seems to be indispensable for learning about homology
and cohomology. I found it easier to read after I already knew what I would be
using it for; otherwise it seemed too abstract. I didn’t get much out of it on
my first read, so I’ll try again over the summer. Dummit and Foote give you
some material on group cohomology, including interpretations of the first two
cohomology groups.
15