Number Theory II: What is an algebrai

number
and why should I are?

Kiran Kedlaya
Berkeley Math Cir le, De ember 10, 2000

Basi fa ts

I'll answer half of the title question immediately. For the purposes of this talk, an algebrai
number is a omplex number whi h is the root of a polynomial with integer oe ients. An
algebrai integer is an algebrai number whi h is the root of a moni polynomial with integer
oe ients.
I'll say more about why you should are later. For now, let me just give a bun h of
examples to onvin e you that algebrai numbers rop up all over the pla e.

A rational number is an algebrai number; a rational number is an algebrai integer if

and only if it is an integer. For larity, I'll refer to the usual integers as the rational
integers.
For any rational number p=q , the root of unity e2ip=q satises the polynomial xq 1 = 0,
and so is an algebrai integer.




A Gaussian integer, a number of the form a + bi, is an algebrai integer.

Given a re urren e relation xn+k = a1 xn+k 1 +


+ ak xn with (rational) integer oe ients, all solutions an be expressed in terms of some algebrai integers. For example,
the n-th Fibona i number an be written as

For any rational number p=q , the numbers os(2p=q ) and sin(2p=q ) are algebrai
integers, and tan(2p=q ) is an algebrai number. Can you expli itly write down polynomials that these are roots of? (These polynomials turn out to have lots of interesting
properties.)

p1
5


 1 + p 5 n  1 p 5 n 
2

The eigenvalues of a matrix with (rational) integer entries are algebrai integers. This
is one way algebrai numbers ome up in topology, group theory, algebrai geometry,
ombinatori s, et .

Here are some basi fa ts about algebrai numbers. These may not be obvious at rst;
I'll mention two ways to prove them in a moment.
1. The set of algebrai numbers is losed under addition, subtra tion, multipli ation and
division. The set of algebrai integers is losed under addition, subtra tion and multipli ation, but not division.
1

2. The root of a polynomial whose oe ients are algebrai numbers (resp., algebrai
integers) is one also.
The rst method involves symmetri polynomials, whi h are interesting enough in their
own right that I'll dis uss them in some detail.
Let P be a symmetri polynomial (with integer oe ients) in x1 ; : : : ; xn . Then
P an be expressed as a polynomial (with integer oe ients) in the elementary symmetri
fun tions 1; : : : ; n given by

Theorem 1.

tn + 1 tn 1 +


+ n = (t + x1)


(t + xn ):

For example, 1 = x1 +


+ xn , 2 = i<j xi xj , and so on.

The proof of this is a su essive elimination argument of a form quite ommon in omputational algebrai geometry. It's related to something alled a \Grobner basis".
Proof. The idea is to deal with the terms of P \from the outside in". That is, we rst deal
with terms whi h are as \unbalan ed" as possible.
We an write
X X
a1 ;:::;an xa11


xann :
P=
a1 


an sym
We put an ordering on n-tuples by saying that (a1; : : : ; an ) > (b1; : : : ; bn ) if and only if
na1 + (n 1)a2 +


+ an > nb1 + (n 1)b2 +


+ bn . Now sort the terms in de reasing
order by na1 + (n 1)a2 +


+ an . Choose the biggest term (a1; : : : ; an ) and noti e that
the polynomial
1a1 a2


nan 11 an nan
has the same largest term, and the oe ient of that term is 1. So subtra t o a1 ;:::;an times
this produ t, and repeat.

I'll demonstrate why this is a useful fa t by showing that the produ t of two algebrai
integers is an algebrai integer. Given two algebrai integers whi h are the roots of the moni
polynomials P and Q with rational integer oe ients, let 1 ; : : : ; m and 1; : : : ; n be the
roots of P and Q, respe tively. Now onsider the polynomial
m Y
n
Y

i=1 j =1

(x

i j ):

Now nj=1 (x i j ) an be viewed as a symmetri polynomial in 1; : : : ; n , if I treat x

and i as onstant. (More pre isely, if I look at all terms with a xed power of x and i ,
these together form a symmetri polynomial in the i). By the theorem, this polynomial
is a polynomial in the elementary symmetri fun tions of the i, whi h by assumption are
integers. So we now have a polynomial in x and i with integer oe ients; we now take
the produ t over the i and repeat the argument.
2

The se ond, more modern method, uses some linear algebra. The main idea is that  is
an algebrai number if and only if 1; ; 2 ; : : : lie in a nite dimensional ve tor spa e over
Q. So to show that  is algebrai given that  and  are, suppose  satises a polynomial
of degree m and  satises a polynomial of degree n over Q. Then all of the produ ts i  j
lie in the ve tor spa e spanned by k  l for k < m; l < n. In parti ular, all of the ( )i lie
in a nite dimensional ve tor spa e, so  is algebrai .
2

Unique and nonunique fa torization

It's easier to study algebrai numbers as part of a larger stru ture than on their own. So
we dene a number eld to be the smallest set ontaining Q plus some nite set 1 ; : : : ; n
of algebrai numbers, whi h is also losed under addition, subtra tion, multipli ation and
division. We'll usually denote this number eld Q(1 ; : : : ; n ). We dene a ring of integers
to be the set of algebrai integers in a number eld.
WARNING:pit's not always obvious what the ring of integers in a numberpeld is. Take
the example Q( D ) (D positive or negative). If D  1 (mod 4), then (1 + pD )=2 is an
algebrai integer; more generally, the integers in the number eld are (a + b D )=2 for
a; b rational integers of the same parity.pIf D  2; 3 (mod 4), then the only integers in the
number eld are the obvious ones a + b D for a; b rational integers.
One ni e property about the rational integers is unique fa torization. Is unique fa torization true for other rings of integers? Sometimes yes, sometimes no.
To make that pre ise, we'll need some more denitions. dene a unit to be an algebrai
integer whose re ipro al is also an algebrai integer. (For example, roots of unity are units,
but there are other units too; we'll see some later.) We all an element  of a ring of integers
irredu ible if whenever you write  =  as the produ t of two elements of the ring, one of 
or is a unit. (I didn't say \prime" be ause I'm saving that word for later). We say a ring
of integers has unique fa torization if whenever an element of a ring of integers is expressed
as a produ t of irredu ible elements, that expression is unique up to hanging the order and
multiplying by units.
For example, the Gaussian integers have unique fa torization, be ause they admit an
analogue of the Eu lidean division algorithm.
Given Gaussian integers p and q with q =
6 0, there exist Gaussian integers r
and s with p = qr + s and jsj < jq j.

Theorem 2.

Proof. Draw the square with verti es 0; q; iq; (1 + i)q . Then p is ongruent to a Gaussian
integer z inside (or on the boundary of) the square. Also, the open dis s of radius jq j entered
at 0; q; iq; (1 + i)q over the square ompletely, so z is within jq j of one orner of the square,
say w. Now take s = z w; then jsj < jq j and s  p (mod q ), so we an set r = (p s)=q
and we're done.
Every (rational) prime ongruent to 1 modulo 4 is the sum of two squares;
moreover, this expression is unique up to order and signs.

Corollary 1.

Proof. If p  1 (mod 4), then there exist x and y su h that x2 + y 2 is divisible by p but not
by p2 . Apply the Eu lidean algorithm in the Gaussian integers (left for you to write down!)
to x + iy and p; the result will be a Gaussian integer r + si with r2 + s2 = p. Uniqueness is
also left to you.

EXERCISE:p Find some other

rings of integers whi h have unique fa torization. (For
p
starters, try Z[ 2 and Z[(1 +
3)=2.
p
On the other hand, onsider this example in Z[ 5:

6 = 2
3 = (1 +

5)(1

5):

None of 2, 3, or 1 
5 an be written as a nontrivial produ t of two elements of Z[ 5,
so this ring doesn't have unique fa torization. What to do?
Kummer realized that one ould nd an algebrai integer in a bigger ring p
that would
allow you to break
up
su h
problem
fa torizations.
(For
example,
if
we
toss
in
2, then it
p
divides both 1 +
5 and 2.) However, it turns out that it's a little better to work not with
these \ideal numbers", as Kummer alled them, but with the olle tion of their multiples.
Denition: an ideal in a ring of integers R is a subset S su h that
1. for x; y 2 S , x + y 2 S ;
2. if x 2 S and r 2 r, then xr 2 S .
Example: If R = Z, then an ideal is an arithmeti progression ontaining 0. More general
example: the prin ipal ideal generated by r 2 R onsists of all multiples of r. But not all
ideals have this form!
Be ause of the way an ideal is dened, we an work \modulo" an ideal, that is, it makes
sense to write a  b (mod I ) be ause this equivalen e respe ts addition and multipli ation.
If I is nonzero, then the number of equivalen e lasses modulo I is nite; we all this number
the norm of the ideal.
An ideal I is prime if xy 2 I implies x 2 I or y 2 I . For example, if R = Zand I = (n),
then I is prime if and only if n is prime. In general, if I has prime norm, it is a prime ideal,
but the onverse is not true; we only know that I has prime power norm. For example, the
ideal (3) in Z[i is prime, but its norm is 9.
The arithmeti on Z[i=(3) is not the same as on Z=(9), though! The main distin tion is
that in Z[i=(3), everything not ongruent to 0 mod (3) has a multipli ative inverse. (Thus
Z[i=(3) is an example of a nite eld.)
The big theorem about prime ideals is the re overy of unique fa torization.
Theorem 3.

Every nonzero ideal in a ring of integers has a unique prime fa torization.

If every ideal of a ring of integers R is prin ipal, then R has unique fa torization. (Note: the onverse is also true.)
Corollary 2.

Two ideals I and J in the ring of integers R of a number eld K are equivalent if there
exists k 2 K su h that kI = J . (Note that k need not lie in R. If you prefer a denition
within R: I and J are equivalent if there exist nonzero i; j 2 R su h that iI = jJ .) This
equivalen e is respe ted by multipli ation.
Theorem 4 (Minkowski).

is nite.

The number of equivalen e lasses of ideals in a ring of integers

This number is alled the lass number of the number eld.

For n > 3 not divisible by 4, the number of primitive (having no
ommon fa tor) triplesp(x; y; z ) of integers su h that x2 + y 2 + z 2 = n is equal to 12 ptimes
the lass number of Q(
n) if n  1; 2 (mod 4), or 24 times the lass number of Q( n)
if n  3 (mod 4).

Theorem 5 (Gauss).

Note: Gauss didn't express this theorem in terms of number elds, but in terms of binary
quadrati forms ax2 + bxy + y 2 whose dis riminant b2 4a equals 4n, if n  1; 2 (mod 4),
or n, if n  3 (mod 4). Two forms are equivalent if you an get from one to the other
by making a variable substitution of the form u = px + qy; v = rx + sy where p; q; r; s are
integers with ps qr = 1.
EXERCISE:
Prove that the number of equivalen e lasses of forms equals the lass
p
number of Q( n).
3

Diophantine equations

One important use of algebrai numbers is to answer questions about Diophantine equations.
We have already seen one example of this (representing an integer as the sum of two squares);
let's onsider a few more examples.
p
The equation x2 Dy 2 = 1 is (mis)named \Pell's equation". Over Q( D), we an
fa tor the left side and rewrite the equation as

(x + y D )(x

y D ) = 1:

This makes lear the multipli ative stru ture of the set of solutions: if (a; b) and ( ; d) are
solutions, then

1 = (a + b D )( + d D )(a b D )( d D )
p
p
= [(a + bdD) + (ad + b ) D [(a + bdD) (ad + b ) D
= (a + bdD)2 D(ad + b )2:

Moreover, we an sort all solutions into in reasing order by x + y D (whi h is an in reasing

fun tion of x for x  0, given that x2 Dy 2 = 1). Now it's easy to see that all solutions in
positive integers are \powers" of the smallest solution, assuming that any solutions exist.
5

There are several ways to show that solutions exist. One method uses ontinued fra tions
and has been known at least for 1000 years (it o urs in an old Indian text); it is probably
the best method for expli itly omputing solutions.
ASIDE: What does this have to do with p
algebrai numbers? What we've done is to
lassify the algebrai integers in the eld Q( D ) whose produ ts with their onjugates
equal 1. An analogous lassi ation an be made for an arbitrary number eld, whi h solves
the Pell equation along the way.
What about the equation x2 Dy 2 = n when n 6= 1? The situation is more ompli ated,
p
so one needs to know a bit more to make progress. For example, given that Q( 2) has
unique fa torization, one an prove the following. (Note the resemblan e to the proof that
a prime p  1 (mod 4) is the sum of two squares.)
Theorem 6. For n a squarefree integer, the equation x2 2y 2 = n has a solution in integers
if and only if it has a solution modulo n.
Proof. By multipli ativity, it su es to show that x2 2y 2 = n has a solution for n = 1,
n = 2, and n = p for p an odd prime su h that 2 is ongruent to a square modulo p. For
n = 1, use 12 2
12 = 1; for n = 2, use 22 2
12 = 2.
Now suppose p is an odd prime su h that 2 is ongruent to a square modulo p. Find x; y
su h that x2 2y 2 is divisible by p but not by
p p2 (if it is divisible by p2,2x that2 by repla ing
x with x + p). Now form the ideal (x + y D; p). Its norm divides p and x 2y 2, so it
must be p.

In identally, one an repla e 2 by any integer D su h that Q( D ) has unique fa torization, provided that x2 Dy 2 = 1 has a solution. It turns out (but is by no means obvious!)
that
p unique fa torization implies that D is prime, and it is believed (but not proved) that
Q( D) has unique fa torization for about 75% of the primes D. Moreover, existen e of a
solution of x2 Dy 2 = 1 then implies D  1 (mod 4), but not every prime ongruent to 1
modulo 4 will work (try D = 5).
For an example
p of a dierent avor, let us nd the solutions of the equation x2 + 2 = y3.
In the ring Z[ 2, whi h has unique fa torization, this fa tors as
p
p
2)(x
2) = y 3:
(x +

Note that x must be odd: if x were even, then x2 + 2 would

be divisiblepby 2 but not by 4,
p
so ould not be a perfe t ube. Therefore the ideals (x + p
2) and (x
p2) are relatively
prime, and ea h must be the ube of an ideal. That is, x +
2 and x
2 are equal to
a unit (whi h an only be 1) times a a ube. In parti ular, we have
p
p
p
x + 2 = (a + b 2)3 = (a3 6ab2) + (3a2 b 2b3 ) 2:
In parti ular, 3a2 b b3 = 1. Sin e this is a multiple of b, we must have b = 1. If b = 1,
then 3a2 2 = 1, so a = 1 and x = 5. If b = 1, then 3a2 + 2 = 1, whi h is impossible.
ASIDE: We didn't a tually need unique fa torization: the argument still would go
through if we just knew that the number eld had lass number not divisible by 3.
Additional examples:
6

1. One an prove the law of quadrati re ipro ity by working with number elds ontaining
roots of unity. (Quadrati re ipro ity will be des ribed in Oaz's talk.)
2. Lame gave a proof of Fermat's Last Theorem for p-th powers assuming that the number
eld Q(e2i=p ) has unique fa torization. Unfortunately, this only holds for nitely many
primes p. Fortunately, Kummer gave a proof that also works if the lass number of
Q(e2i=p) is not divisible by p. Unfortunately, no one has proved that there are innitely
many su h p. Fortunately, numeri al eviden e and heuristi s suggest that about 60%
of primes have this property. (More fortunately, Fermat's Last Theorem has now been
proved by Wiles et al.)
4

Read all about it!

There are tons of books on algebrai number theory out there (some of whi h don't assume
very mu h from lassi al number theory). Some titles that ome to mind (with ommentary):

Esmonde and Murty, Problems in Algebrai Number Theory (beware of the many small
errors, hopefully to be orre ted in a future edition)

Ireland and Rosen, A Classi al Introdu tion to Modern Number Theory (a pretty good
read, I'm told)





Lang, Algebrai Number Theory (not an easy read, assumes undergraduate algebra)

Mar us, Number Fields (mostly does examples)

Neukir h, Algebrai Number Theory (not to be onfused with his other books, whi h
are very di ult reading)
Pollard and Diamond, The Theory of Algebrai Numbers (the avoidan e of abstra t
algebra makes this an easy read but obs ures ertain points)