Course notes for PMATH 641

Christopher Hawthorne
Lectures by David McKinnon

Contents
1 Dedekind domains 2

2 Geometry of numbers 10

3 Factorization of primes in extensions 17

4 InterludeFinite elds 23

5 Slightly less nite elds 23

Preliminaries
My thanks to Bahaa Khaddaj for the use of his notes for the lectures I missed.
All rings are commutative and have unity.
Denition 0.1. A number eld is a nite extension of Q. An algebraic number is an element of a number
eld.
Denition 0.2. An algebraic integer is an C such that Z[] is a nitely generated Z-module.
Denition 0.3. Suppose K is a number eld. We dene the ring of integers of K, denoted OK , to be the
set of algebraic integers lying in K.
In fact, OK is a ring. That 0 OK is obvious; closure under addition, multiplication, and additive
inverses will follow from the following theorem.
Theorem 0.4. Suppose C. Then is an algebraic integer if and only if p() = 0 for some monic
p Z[x].
Proof.
( = ) Suppose is an algebraic integer. Then Z[] is a nitely generated Z-module, say by f1 (), . . . , fn ().
Let k = max{ deg(fi ) : i { 1, . . . , n } } + 1. Then

k = a1 f1 () + + an fn ()

So if
p(x) = xk + a1 f1 (x) + + an fn (x)
then p Z[x] is non-zero and monic, and p() = 0.
( = ) Suppose
k + ak1 k1 + + a0 = 0
where a0 , . . . , ak1 Z. Then { k1 , . . . , } generates Z[] as a Z-module, and Z[] is nitely
generated as a Z-module.

1
 Theorem 0.4
Remark 0.5. Qalg is an algebraic integer if and only if its monic minimal polynomial over Q has integer
coecients.
Hence if , OK , then Z[, ] is nitely generated. But Z[ + ], Z[], and Z[] are submodules of
Z[, ], and Z is Noetherian; so Z[ + ], Z[], and Z[] are nitely generated, and + , , OK .
So OK is a ring.

1 Dedekind domains
Denition 1.1. A number ring is the ring of integers of a number eld.
It turns out that for number rings, being a UFD is equivalent to being a PID. Not all rings satisfy this
property:

Example 1.2. Z[ 10] is not a UFD: 10 = 10 10 = 2 5, and all of 10, 2, and 5 are irreducible.
We dene a kind of ring that better corresponds to number rings:
Denition 1.3. Suppose D is a domain, T D is a subring, and D is an element. Then is integral
over T if and only if p() = 0 for some monic p T [x].
Fact 1.4. If T is Noetherian, then is integral over T if and only if T [] is a nitely-generated T -module.
Denition 1.5. The integral closure of T in D is { D : is integral over T }.
Fact 1.6. The integral closure of T in D is a ring.

So the ring of integers OK in a number eld K is the integral closure of Z in K.

Denition 1.7. We say D is integral over T if every element of D is integral over T .
Fact 1.8. Suppose A B C are domains. If B is integral over A and C is integral over B, then C is
integral over A.

Hence the integral closure of OK in K is OK , since every element of the integral closure of OK in K is
integral over O, and is therefore already in OK .
Denition 1.9. A Dedekind domain is a domain D that is not a eld satisfying the following:
1. D is Noetherian.

2. Every non-zero prime ideal of D is maximal.

3. D is integrally closed in its eld of fractions.
Proposition 1.10. Suppose K is a number eld of degree d over Q; let OK be the ring of integers in K.
Suppose I OK is a non-zero ideal. Then I
= Zd as additive grapes.
Proof. We rst check the case I = OK .
Claim 1.11. For any K there is some n Z such that n OK .

Proof. Let p Q[x] be the monic minimal polynomial for over Q. Then for any n N, we have na p( nx ) is
the monic minimal polynomial for n over Q (where a = deg(p)). If we choose n to cancel the denominators
of the coecients of p(x), we get that na p( nx ) Z. Claim 1.11

2
 Now, let 1 , . . . , d be a basis for K as a vector space over Q. By the claim, we may assume that
1 , . . . , n OK . So OK contains an additive subgrape additively isomorphic to Zd ; i.e. Z1 + + Zd .
We now nd non-zero A Z such that AOK Z1 + + Zd . Choose any b OK . Since 1 , . . . , n
are a basis for K over Q, we may write
b = a1 1 + + ad d
for some (unique) a1 , . . . , ad Q. For simplicity, we assume that K is a Galois extension of Q. Then for
every Gal(K/Q) we get a new equation
i (b) = a1 i (1 ) + + ad i (d )
This yields a d d system of linear equations in a1 , . . . , ad :

1 (b) a1
.. ..
. =M.
d (b) ad
But M is invertible: since the a1 , . . . , ad are unique, there is a unique solution. So, by Cramers rule, we get
that
a1 1 (b) 1 (b)
.. 1 .
M 0 ...
1 1
. = M .. = Od

det(M ) det(M ) K
ad d (b) d (b)
But det(M ) OK ; so there is an integer A Z such that det(M
A
) OK . So each Aai is in OK . But Aai Q;
so each ai Z, and Ab Z1 + + Zd . Since A doesnt depend on b, we get that
AOK Z1 + + Zd
So OK is additively isomorphic to a subgrape of Zd that contains a copy of Zd ; so OK
= Zd as an additive
grape.
Now, let I OK be any non-zero ideal; let I be non-zero. Then OK I, and OK = OK = Zd
as additive grapes; better yet, we have I OK = Z as an additive grape. So I = Z as an additive grape.
d d

Proposition 1.10
This then implies that every non-zero prime of OK is maximal: for any prime ideal P of OK , we note
that the quotient OK /P is an integral domain and it must have nite rank, and is thus a eld. So OK is a
Dedekind domain for every number eld K.
In general, every PID that is not a eld is a Dedekind domain. We also have that every number ring is
a Dedekind domain. In general, we can ask how close a number ring is to being a PID.
Denition 1.12. Suppose D is a domain with fraction eld K. A fractional ideal of D is a D-submodule
I of K such that I D for some K \ { 0 }.
Remark 1.13. Fractional ideals of D are 1
I for D \ { 0 } and some integral ideal I D (i.e. an ideal in
the usual ring-theoretic sense).
Denition 1.14. A fractional ideal I is invertible if there is some fractional ideal J such that IJ = D = (1).
The ideal grape of D is the grape of invertible fractional ideals under .
Denition 1.15. Suppose I is a non-zero fractional ideal of D. We dene I 1 = { K : I D }.
Theorem 1.16. I 1 is a fractional ideal, and II 1 = D if and only if I is invertible; furthermore, in this
case we have that I 1 is the unique fractional ideal J with IJ = D.
Proof. We rst check that I 1 is a fractional ideal. It is immediate that 0 I 1 . To check closure under
addition, suppose , I 1 ; then ( + )I = I + I D, and + I 1 . To see closure under
multiplication by arbitrary elements of D, suppose I 1 and D; then I D D, and I 1 .
So I 1 is a fractional ideal.
We now check the equivalence above.

3
( = ) Immediate.
( = ) Suppose I is invertible; then there is some fractional ideal J of D with IJ = D. Hence J I 1 ;
then IJ II 1 D = IJ, and II 1 = D.
For the furthermore, we note that

J = J(II 1 ) = (JI)I 1 = DI 1 = I 1

Theorem 1.16

Example 1.17. I = (2, 1 + 5) Z[ 5] is not invertible.
To compute I , we
1
note that a + b 5 I
1
is
equivalent to requiring that (1 + 5)(a + b 5) Z[ 5] and 2(a + b 5) Z[ 5]. So
n n m o
I 1 = + 5 : m n (mod 2)
2 2
n n + 2k

= + 5 : n, k Z
2 2
( )
1+ 5
= n + k 5 : n, k Z
2
!
1+ 5
= , 5
2

But
! !
1+ 5
(2D + (1 + 5)D) D + 5D = (1 + 5)D + 2 5D + (3 + 5)D + (5 + 5)D 2Z + 5Z $ D
2

Denition 1.18. Suppose D is a domain. The class grape of D is the following quotient:

Cl(D) = { invertible fractional ideals }/{ principal fractional ideals }

If every fractional ideal of D is invertible, then Cl(D) measures how close D is to being a PID; in
particular, D is a PID if and only if Cl(D) = { 1 } (under the above assumption).
We will show that every non-zero fractional ideal of a Dedekind domain is invertible.
Denition 1.19. Suppose D is a domain with fraction eld K; suppose P D is a prime ideal. We dene
the local ring of D at P to be n a o
DP = K : = , b /P
b
One checks that this is a ring whose unique maximal ideal is
na o
P DP = : a P, b
/P
b
Example 1.20. Consider D = Z, P = (2). Then
n a o n a o
/ (2) = Q : = , b odd
DP = Z(2) = Q : = , b
b b
Now, given a
b Q, we can write
a a0
= 2n 0
b b
where 2 - a b ; this n is called the (additive) 2-adic valuation of
0 0
b,
a
denoted v2 ( ab ). Then Z(2) = { Q :
v2 () 0 }.
A result from algebra:
Theorem 1.21. If D is a Dedekind domain, then DP is a PID.

4
Denition 1.22. A local ring is a ring that has a unique maximal ideal.
Denition 1.23. A discrete valuation ring (or DVR) is a local ring that is also a PID (and is not a eld).
Now, if DP is a DVR with maximal ideal P DP , then P DP = uDP ; we call u a uniformizing parameter
or uniformizer for DP .
Let K be the fraction eld of DP (which is also the fraction eld of D). If K \ { 0 }, we dene
vP () = max{ n Z : (P DP )n } where (P DP )0 = DP and (P DP )1 = u1 DP .
Example 1.24. Using D = Z and P = (2) as above, we note that v2 (47) = 0 since 2 is a uniformizing
parameter and 47 Z(2) but 47
/ 2Z(2) .
In general we have that vP () is the largest n such that = un r where r DP ; i.e. r = ab with b / P.
To achieve the maximum, we demand that a, b / P ; i.e. vP () is the unique n such that there is some unit
v of DP with = un v.
Remark 1.25. P DP is always principal by the above; if P = (r1 , . . . , rn ) and none of r1 , . . . , rn generated
P DP as an ideal of DP , then they would all have valuation greater than 2. Hence if u is a uniformizing
parameter, then u2 | ri for all i { 1, . . . , n }; so u2 | r for all r r1 DP + + rn DP = P DP , and in
particular we get that u2 | u. So u is a unit, a contradiction. So P DP = ri DP for some i { 1, . . . , n }.
Theorem 1.26. Every non-zero fractional ideal in a Dedekind domain is invertible.
Proof. Suppose D is a Dedekind domain; suppose I is a fractional ideal of D. Then there is some K
(where K is the fraction eld of D) such that I D is an integral ideal of D. Better yet, I is invertible
if and only if I is invertible; we may thus assume that I D. Now, II 1 D by denition of I 1 .
If II 1 6= D, then there is some non-zero prime ideal P of D with II 1 P D; consider the ideal
IP = IDP of DP . Also consider the ideal (I 1 )P = I 1 DP of DP . Finally, consider (IP )1 , the inverse of
the fractional ideal of IP of DP .
Now, DP is a DVR, and in particular is a PID; so IP , (IP )1 , and (I 1 )P are all principal, and thus
invertible. So IP (IP )1 = DP .
Claim 1.27. (IP )1 = (I 1 )P .
Proof. We rst note that
n o n o
(I 1 )P = :b
/ P, I D : IP DP = (IP )1
b b b
For the converse, write I = a1 D + + an D. Then x (IP )1 implies that xai DP for all i; in particular,
there is some ci D \ P such that ci xai D. Now, if c = c1 cn , then (cx)ai D for all i. So cx I 1 ,
and c / P ; so x (I 1 )P . Claim 1.27
So (II 1 )P = IP (I 1 )P = IP (IP )1 = DP . But II 1 P D, so (II 1 )P PP $ DP , a contradiction.
Theorem 1.26
Theorem 1.28. Every non-zero fractional ideal of a Dedekind domain D can be written uniquely (up to
permutation) as a product of prime ideals and their inverses.
Proof. Let I be a non-zero fractional ideal of D; then I = a1 J for some integral ideal J of D and some a D.
If J + (a) can be factorized, then I can as well; we may thus assume that I D.
Now, if I = D, then I is the empty product of prime ideals, and were done. If I is prime, were also
done. Otherwise, there is some maximal ideal P with I P D. Then I = P (IP 1 ), and P and IP 1 are
both integral ideals of D. Iterating, we get

I IP11 IP11 P21 D

Since D is Noetherian, we get that this process must terminate. Hence IP11 Pn1 = D, and I = P1 Pn .
Uniqueness follows straightforwardly from the invertibility of all of the Pi . Theorem 1.28
So every ideal in a Dedekind domain factors uniquely into prime ideals. If we want to understand all
ideals, it then suces to understand prime ideals.

5
Example 1.29. What do the prime ideals of Z[i] look like? Well, if Z[i] is prime, then Z[i]/ is a nite
eld (by Proposition 1.10). So Z[i]/ Fp for some prime p; in other words, p . i.e. Every prime of
Z[i] contains some prime p of Z; hence (p). Hence if we factor (p) inside Z[i], we will nd as a factor.
The upshot is that if we want to nd all of the primes of Z[i], all we have to do is start with the primes
p Z and factor (p) in Z[i]; every will show up as a factor for some p.
Case 1. By homework, if p 3 (mod 4) then (p) is prime in Z[i].
Case 2. If p = 2, we note that 2 = (1 + i)(1 i) = (1 i)2 i.
Case 3. Suppose p 1 (mod 4). Note that

Z[i]/(p)
= Z[x]/(p, x2 + 1)
= Fp /(x2 + 1)

Fact 1.30. x2 + 1 = (x a)(x b) in Fp [x] for some a, b Fp .

Claim 1.31. a 6= b.
d
Proof. We simply note that gcd(x2 + 1, dx (x2 + 1)) = 1, and hence x2 + 1 has no multiple roots.
Alternatively, note that p 6= 2, so b = a 6= a. Claim 1.31

Then (x a) and (x b) are comaximal, and

Fp /(x2 + 1)
= Fp [x]/(x a)Fp [x]/(x b)

= Z[x]/(x2 + 1, x a, p) Z[x]/(x2 + 1, x b, p)

= Z[x]/(x a, p) Z[x]/(x b, p)

One then checks (by hand) that (p) = (p, i a)(p, i b). (Or one follows the maps through and veries
that the above guarantees equality.)
Hence all the primes in Z[i] are
(p) for p 3 (mod 4)
(1 + i) = (1 i)
(p, i a) where p 1 (mod 4) and a2 1 (mod p)
The above algorithm also works for any quadratic number eld; well end up with
Some (p) already prime.
Some that are of the form (p) = 2 .
The rest of the form (p) = (p, x a)(p, x b).
We will later show that the class grape is nite.
Question 1.32. How do we tell if weve written down all generators of the class grape?
Some denitions:
Denition 1.33. Suppose K is a number eld; suppose K and [K : Q] = n. We then have n embeddings
f1 , . . . , fn : K , C. We dene the trace of to be
n
trK () =
X
fi ()
i=1

We dene the norm of to be

n
Y
NK () = fi ()
i=1

6
Remark 1.34. Note that trK (), NK () Q, as they are both Galois-invariant. Further note that if OK ,
then tr(), N () Z; this is because the characteristic polynomial of is

xn tr()xn1 + N ()

and the characteristic polynomial is a power of the minimal polynomial, which has integer coecients if
OK .
The converse is false: there are K \ OK with tr(), N () Z.
Example 1.35. Let K = Q() where satises x3 + x2 + 14
67 x + 4. Then K \ OK , but

tr() = 1
N () = 4

Remark 1.36. tr( + ) = tr() + tr() and N () = N ()N ().

Denition 1.37. We say 1 , . . . , n K is an integral basis if n = [K : Q] and

OK = Z1 + + Zn

i.e. if 1 , . . . , n OK , they form a Q-basis for K, and OK = Z1 Zn . (I think this is just a Z-

module basis.)
Remark 1.38. Integral bases always exist, since by Proposition 1.10 we know OK
= Zn .
Question 1.39. How do we nd an integral basis in practice?
Denition 1.40. Let n = [K : Q]. If 1 , . . . , n K, we dene the discriminant of 1 , . . . , n to be
2
f1 (1 ) f1 (n )
.. .. ..
det(fj (i )) = det .
2
. .

fn (1 ) fn (n )
Example 1.41. A special case: we would like to nd OK such that OK = Z[]. This is equivalent to
1, , . . . , n1 being an integral basis. Computing the discriminant, we nd
2
1 f1 () f1 ()2 f1 ()n1

disc(1, , . . . , n1 ) = det ... ..

.
..
.
..
.
..
.


1 fn () fn ()2 fn ()n1
Y
= (fi () fj ())2
i<j
n
!
= disc
Y
(x fi ())
i=1
= the characteristic polynomial of

Remark 1.42. If 1 , . . . , n K is not a vector space basis for K over Q then disc(1 , . . . , n ) = 0. This is
because in this case we have c1 , . . . , cn Q not all 0 such that
n
X
ci i = 0
i=1

So !
n
X n
X
0 = fj ci i = ci fj (i )
i=1 i=1

for all j. Hence the columns of (fj (i ))ij are linearly dependent; hence the determinant is 0, and hence the
discriminant is 0.

7
 If, on the other hand, 1 , . . . , n is a basis, then the discriminant is non-zero. This is because if
disc(1 , . . . , n ) = 0, then the columns of (fj (i ))ij are linearly dependent. So there are c1 , . . . , cn Q
such that for all j we have !
X n Xn
fj ci i = ci fj (i ) = 0
i=1 i=1
But the fj are embeddings; so
n
X
ci i = 0
i=1
and the i are linearly dependent.
Remark 1.43. If { 1 , . . . , n } and { 1 , . . . , n } are both bases, how do disc(1 , . . . , n ) and disc(1 , . . . , n )
relate? Well, there is a unique Q-linear T : K K such that i 7 i . In this case, we get T (fj (i ))ij =
(fj (i ))ij ; hence disc(1 , . . . , n ) = det(T )2 disc(1 , . . . , n ).
If in fact { 1 , . . . , n } is an integral basis and 1 , . . . , n OK , then T Mnn (Z); hence
disc(1 , . . . , n ) = (det(T ))2 disc(1 , . . . , n )
where (det(T ))2 is a square integer. Hence if { 1 , . . . , n } and { 1 , . . . , n } are both integral bases, then
T is invertible. But T Mn (Z); so det(T ) = 1, and det(T )2 = 1. So in this case we get disc(1 , . . . , n ) =
disc(1 , . . . , n ).
This allows the following denition:
Denition 1.44. Suppose K is a number eld. We dene the discriminant of K to be disc(1 , . . . , n )
for any integral basis 1 , . . . , n . We just showed that this is well-dened and independent of the choice of
integral basis.
Remark 1.45. If 1 , . . . , n OK and disc(1 , . . . , n ) 6= 0 is not divisible by the square of an integer, then
{ 1 , . . . , n } is an integral basis.
Unfortunately, the converse fails.
Denition 1.46. Say K is a number eld with OK its ring of integers; say I OK is a non-zero ideal. We
dene the norm of I to be N (I) = |OK /I|.
(Recall that the norm of an element is given by
Y
N () = fi ()
f : K,C

We will see later how these relate.)

disc(I)
Theorem 1.47. N (I)2 = disc(K) .

(I think we dene disc(I) to be the discriminant of an integral basis for I; i.e. 1 , . . . , d I such that
I = Z1 + + Zd , where d = [K : Q]. Recall by Proposition 1.10 that I
= Zd as additive grapes.)
Proof. Let { a1 , . . . , ad } and { b1 , . . . , bd } be integral bases for OK and I, respectively; let T : OK I be
the unique Q-linear transformation with T (ai ) = bi for all i.
Recall from linear algebra that
(volume of the polytope spanned by { a1 , . . . , ad }) det(T ) = (volume of the polytope spanned by { b1 , . . . , bd })
But then
volume of the polytope spanned by { a1 , . . . , ad }
|det(T )| = = [OK : I] = |OK /I| = N (I)
volume of the polytope spanned by { b1 , . . . , bd }
(The third equality follows by noting that (OK /I) = (K/I)/(K/OK ), with K/I the fundamental polytope
of I and K/OK the fundamental polytope of OK . Here the fundamental domain of a grape action is a set
that every orbit intersects exactly once.)
disc(I)
But disc(K) = (det(T ))2 = N (I)2 by Remark 1.43. Theorem 1.47

8
 We can thus extend the denition of norm to fractional ideals as follows:
Denition 1.48. Suppose I is a fractional ideal of OK . We dene
s
disc(I)
N (I) =
disc(K)

Denition 1.49. A sequence A C is exact at B if im(f ) = ker(g).

f g
B
TODO 1. Swapped denition of norm of ideals?
Theorem 1.50. Given non-zero fractional ideals I and J of OK , we have N (I)N (J) = N (IJ).
Proof. If I = I 0 , then playing with integral bases we get
s s s
disc(K) disc(K) 1 disc(K)
N (I) = = =
disc(I) disc(I 0 ) |N ()| disc(I 0 )
So if we choose OK such that I, J OK , then
N ()2 N (I)N (J) = N (I)N (J)
and
N ()2 N (IJ) = N (IJ)
Hence if we can prove the theorem for I and J, it will follow for I and J; i.e. we may assume I, J OK .
Now, since OK is a Dedekind domain, we can factor I and J into primes
I = P1a1 Prar
J = P1b1 Prbr
Then N (I) = N (P1a1 Prar = N (P1a1 ) N (Prar ) by Chinese remainder theorem. (Pairwise comaximality
follows by noting that any prime in the factorization of Piai + Pj j must contain both Piai and Pj j , and hence
a a

must be both Pi and Pj , a contradiction if I 6= J; so the prime factorization is empty, and Pi + Pj j = OK .

ai a

Alternatively, a general theorem states that powers of distinct maximal ideals are comaximal.) So
N (I) = N (P1a1 ) N (Prar )
N (J) = N (P1b1 ) N (Prbr )
N (J) = N (P1a1 +b1 ) N (Prar +br )
It then suces to show that N (Piai ) = N (Pi )ai . We proceed by induction on ai ; the claim is clear if ai = 1.
For the induction step, we note that
0 Piai /Piai +1 OK /Piai +1 OK /Piai 0
| {z } | {z }
N (Pi ) elements N (Pi )ai elements

is exact. The latter claim about sizes is just the induction hypothesis; for the former, we use the following:
Claim 1.51. P a /P a+1
= P a DP /P a+1 DP where D = OK , P = Pi , and a = ai .
Proof. We dene : P a /P a+1 P a DP /P a+1 DP by a + P a+1 7 + P a+1 DP . This is clearly an injective
homomorphism. For surjectivity, we note that P a /P a+1 is a vector space over OK /P via
( + P )( + P a+1 ) = + P a+1
It has dimension 1, as it is spanned by ua (where P DP = (u)); surjectivity then follows because because
P a /P a+1 is a 1-dimensional vector space over DP /P DP and P a 6= P a+1 . Claim 1.51
In particular, for a = 0, we get that OK /P = DP /P DP , which has N (P ) elements. But Piai /Piai +1
is a 1-dimensional vector space over OK /Pi , as noted above; so by the short exact sequence, we get that
OK /Piai +1 has N (Pi )N (Pi )ai = N (Pi )ai +1 elements. Theorem 1.50
Theorem 1.52. If 6= 0 then N (()) = |N ()|.
Proof. If K = Q(), then the characteristic polynomial of T : K K given by x 7 x is the monic minimal
polynomial for over Q. Its constant term is both N () and det(T ) = N (()). Theorem 1.52

9
2 Geometry of numbers
Denition 2.1. Suppose K is a number eld. Let f1 , . . . , fr1 be the distinct embeddings of K , R; let
fr1 +1 , gr1 +1 , . . . , fr1 +r2 , gr1 +r2 be the complex conjugate pairs of embeddings K , C (i.e. with fi = gi ).
Then [K : Q] = r1 + 2r2 = d. Dene
r1
! r2
!
Y Y
K : K R C
i=1 i=1

by
7 (f1 (), . . . , fr1 (), fr1 +1 (), . . . , fr1 +r2 ())
This K is the Minkowski map. Its codomain is Minkowski space, usually considered as a vector space over
R.
The choice of which embedding is fi and which is gi is not important.
Fact 2.2. The image of OK under K is a lattice.
Example 2.3. Let K = Q(i). We pick f1 (a + bi) = a + bi) and g1 (a + bi) = a bi. So r1 = 0 and r2 = 1.
Then OK = Z[i], and its image under K is Z[i].

2.4. Let K = Q( 2). Here as well we have r1 = 2; we pick f1 (a+b 2) = a+b 2 and f2 (a+b 2) =
Example
a b 2. We note that Q gets mapped to the diagonal, with Z mapped to the integer points on the diagonal.
We also note that 2Q gets mapped to the antidiagonal. We nally note that the unit ball (in which
|N ()| 1) is a hyperbola, since it is dened by 1 = |N ()| = |f1 ()f2 ()|, and hence is given by y = x1 . In
particular, it isnt compact.
Remark 2.5. If the unit ball is compact, then there are nitely many units in OK , since units have norm 1.
Remark 2.6. K is a Q-linear map, since if Q and x K then fi (x) = fi ()fi (x) = fi (x).
We think of Minkowski space as unfolding the Q-vector space so it no longer lies on the real line.
Aside 2.7 (How to compute OK from K). Find some algebraic integers 1 , . . . , n K, and consider
R = Z[1 , . . . , n ]; compute the discriminant of R. By Remark 1.43 we have disc(OK )(det(T ))2 = disc(R).
But all of these are integers; so if is square-free, then [K : R] = 1, so R = OK .
N.B. It might be the case that disc(OK ) has a square factor.
Example 2.8. Find OK (with proof) if K = Q() where is a root of x3 + 3x + 7 = 0.
disc(D)
We rst make a guess; we let D = Z[], and we guess that D = OK . We know that disc(O K)
= [OK : D]2
(where the index is taken as additive grapes).
Lets compute disc(D). If disc(D) is square-free, then since [OK : D]2 | disc(D), we get that [OK : D] = 1
and D = OK .
Fact 2.9. |disc(Z[])| = |disc(m(x))| where m(x) is a monic minimal polynomial for over Q.
The discriminant of a monic polynomial is
Y
(ri rj )2
i<j

where the ri are the roots of the polynomial. It also coincides with the resultant of the polynomial and its
derivative, given by

1 an1 a0
1 an1 a0
. .. .. .. ..

. . .



1 an1 a0
Res(f, f 0 ) = det


n (n 1)an1 a1

n (n 1)an1 a1
. .. .. ..

.. . . .


n (n 1)an1 a1

10
where f (x) = xn + an1 xn1 + + a0 . For example,

1 0 2
disc(x2 2) = det 2 0 0 = 8
0 2 0
In our case, we end up with
 

 1 0 3 7 0 
 0 1 0 3 7 
|disc(Z[])| = |disc(x3 + 3x + 7)| = det 0 = 1431 = 33 53
 
3 0 3 0
 0 3 0 3 0

 0 0 3 0 0 

Thus [OK : D] is 3 or 1. Now, if every local ring DQ (for Q a prime ideal of D) is a DVR, then D is a
Dedekind domain (since D is already Noetherian and one-dimensional (i.e. every non-zero prime ideal is
maximal)). Conversely, if any DQ is not a DVR, then D is not a Dedekind domain.
How does this help? Well, Dedekind domains are integrally closed in their eld of fractions, and K is the
fraction eld of D; hence if D is a Dedekind domain then OK D, and OK = D.
Now, let Q D be a prime ideal. How do we check if DQ is a DVR? Well, DQ is a DVR if and only if
QDQ is a principal ideal. But every non-zero prime ideal Q of D contains a unique positive prime integer q.
Case 1. If q 6= 3 then for any prime ideal P of OK with Q P we have DQ = (OK )P , which is a local
ring.

Proof. If D = OK , were done. Otherwise, we have [OK : D] = 3, so for all a OK we have 3a D;

hence 3a DQ , and a DQ . So OK DQ , and (OK )P DQ . Hence (O)P = DQ .

Case 2. We must now check all the prime ideals Q of D containing 3. By the correspondence theorem,
these are in bijection with prime ideals of

D/(3)
= Z[x]/(x3 + 3x + 7, 3)
= (Z/3Z)[x]/(x3 + 3x + 7)
= (Z/3Z)[x]/(x + 1)3
Fact 2.10 (IMPORTANT). In general the prime ideals of F [x]/(g) correspond to the irreducible factors
of g.

In our case, we get that the only prime ideal of D/(3) is (+1), which corresponds to the ideal (+1, 3)
of D. So Q = ( + 1, 3) is the only prime ideal of D = Z[] containing 3.
We now check if QDQ is principal. (This is equivalent to checking that Q is a DVR.) Well, QDQ =
(3, + 1)DQ is a principal ideal if and only if QDQ = (3)DQ or QDQ = ( + 1)DQ , which occurs if
and only if +1
3
DQ or +13 DQ . We easily get that +1
3
/ DQ ; on the other hand, we have

( + 1)3 = 3 + 32 + 3 + 1 = 3 7 + 32 + 3 + 1 = 32 6 = 3(2 2)

Hence
3 ( + 1)2
= 2 DQ
+1 2
But 2 2
/ Q (since ( + 1)( 1) = 2 1 Q, and hence otherwise we would have 1 Q). Hence
+1 DQ . So QDQ is principal.
3

So D is a Dedekind domain, and D = OK .

Theorem 2.11 (Convex body). Suppose L is a lattice in Rn ; i.e. L is a subgrape of Rn isomorphic to Zn
with the property that a basis for L (over Z) is also a basis for Rn as a vector space. Let F be the fundamental
domain of L; i.e. a subset of Rn such each coset of L intersects in exactly one point. (We will need F to
have a sensible volume, so we will require that it be measurable, and in practice we will imagine it to be
the parallelepiped given by the basis for L.) Let S be a symmetric convex subset of Rn . (Symmetric here
means that if x S then x S.) If vol(S) > 2n vol(F ) then S L contains a non-zero vector.

11
Proof. We may assume S is bounded. For any

v L we dene
1
1
v = S+ v ={
T
x +v :
x S}
2 2
Then
vol(S)
vol(T
v)=
> vol(F )
2n
Hence for some

v 6=

w L, we must have T
v T



w 6= . Then for some x , y S, we have
1 1
x +

v = y +
w
2 2
and thus

1
v

w = (
y
x)S
2
So

v

w (L S) \ { 0 }. Theorem 2.11

Example
2.12. Consider K = Q[ 15], in which OK = Z[ 15]. Draw the lattices OK , (2, 3 15), and
(3, 15) in Minkowski space. Compute the fundamental domain volumes, discriminants, and smallest non-
zero vectors.
Theorem 2.13. Suppose K is a number eld of degree n = r1 + 2r2 (where r1 is the number of real
embeddings of K and r2 is the number of complex conjugate pairs of complex embeddings). Let A OK be
an additive subgrape of nite index m. Then there is some A with
  r2
4 n!m p
|N ()| disc(OK )
nn
(Note here that disc(OK ) is the volume of the fundamental domain of OK , and m disc(OK ) is the volume
p p

of the fundamental domain of A.)

Proof. For B R dene

r1
X r2
X
SB = (1 , . . . , r1 , 1 , . . . , r2 ) : |j | + 2 |j | B Rr1 Cr2

j=1 j=1

For example, if r1 = 0 and r2 = 1 then we end up with the disc 2|z| B; if r1 = 2 and r2 = 0 then we end
up with the diamond |x| + |y| B. It turns out SB are roughly products of the things in the above two
examples. In particular, SB is bounded, symmetric, and convex, with volume
  r2 B n
2r1
2 n!

 n1
Pick B > 4 n!m disc(OK ) . Then
r 2
p

  r2 B n  
vol(SB ) = 2r1 > 2n 2r2 m disc(OK )
p
2 n!
But the volume of a fundamental polytope of A is 2r2 (m disc(OK )); hence by Theorem 2.11 there is non-
p

zero 1 ((A) SB ).
Now, let f1 , . . . , fr1 +r2 be the embeddings of K into R and C (up to complex conjugation). Then
r1 r2  r2  n1
4
n!m disc(OK )
X X p
|fj ()| + 2 |fr1 +j ()| +
j=1 j=1

n
r1 r2  r2
4
= |fj ()| + 2 n!m disc(OK ) + 0
X X p
|fr1 +j ()|
j=1 j=1


r1 r2  r2
1X 2X 4 n! p
= |fj ()| + |fr1 +j ()|
m disc(OK ) + 00
n j=1 n j=1 nn

12
   n1
(Here = B 4 r2
n!m disc(OK ) , and 0 and 00 can be computed from ; in particular, they approach

p

0 as approaches 0.)
Hence, by the arithmetic-geometric mean inequality, we get that
r1 r2  r2
4 n! p
m disc(OK + 00
Y Y 2
|N ()| = |fj ()| |fr1 +j ()| n
j=1 j=1
n

We somehow concluded that an can be chosen to work for all suciently small . Hence this is our desired
. Theorem 2.13
In particular, if A = I is an ideal, then m = N (I); so we have I with
  r2
4 n! p
|N ()| disc(OK )N (I)
nn

Exercise 2.14. Verify the above theorem for K = Q( 15) with ideals Z[ 15], (2, 3 15), and (3, 15).
When working in a xed K with r1 real embeddings and r2 pairs of complex embedings, we typically
denote  r2
4 n! p
M= disc(OK )
nn
Theorem 2.15. Suppose K is a number eld. Then Cl(OK ) is nite.
Proof. Well, for any B R we have that there are only nitely many fractional ideals I with N (I) B; it
then suces to show that

r2 n!there
p is some B such that every ideal class contains a representative of norm B.
Let B = M = 4 nn disc(OK ). We will show that for any fractional ideal I there is some K
with N (I) M .
Suppose I is a non-zero fractional ideal; then there is an integral ideal J OK in the ideal class of I 1 .
Then by Theorem 2.13 there is some J \ { 0 } with |N ()| M N (J). But I J 1 in Cl(OK ), and
|N ()|
N (J 1 ) = M
|N (J)|
as desired. Theorem 2.15
Example 2.16. Compute the ideal class grape of OK in K = Q() where 3 + 3 + 7 = 0. (See Example 2.8.)
Our steps:
1. We compute OK . In our case, Example 2.8 yields that OK = Z[].
2. We compute  r2
4 n! p
M= disc(OK ) 10.7 < 11
nn
3. Pick representatives of each of the residue classes of Z/dM eZ, apply f , and factor the results; ideally
pick representatives n that make f (n) small.

5 133 = 7 19
4 69 = 3 23
3 29
2 7
1 3
0 7
1 11
2 21 =37
3 43
4 83
5 147 = 3 72

13
 Why is this useful? Well, for any n we have |N ( n)| is the absolute value of the constant coecient
of the minimal polynomial for n. But the minimal polynomial for n is f (x + n), which has
constant term f (n); hence |N ( n)| = |f (n)|.
4. List all the ideals of OK of norm < M .

(2) is prime since Z[]/(2)

= (Z/2Z)[x]/(x3 + x + 1) is a eld.
(3) = (3, + 1)3 , as computed in Example 2.8; we let P3 = (3, + 1).
(5) is prime.
(7) = (7, )(7, + 2)(7, + 5) = P7 Q7 R7 .
But any prime P of norm < 11 would have to satisfy |OK /P | = pe for some p { 2, 3, 5, 7 }, and hence
would have to contain one such p. So these are all the primes of norm < 11.
5. The above primes thus generate Cl(K); in fact we can omit the principal ideals, so we get Cl(K) =
hP3 , P7 , Q7 , R7 i.
We immediately get that P33 1 and P7 Q7 R7 1. Using the table, we see that |N ()| = |f (0)| = 7,
and hence that |OK /()| = N (()) = |N ()| = 7; in particular we get that 7 0 in OK /(), and that
7 (), and thus that P7 = (7, ) = (). So P7 is principal, and P7 1.
We likewise get that P3 R7 1 (since |N ( 2)| = 21 implies that P3 R7 = ( 2)) and that P3 Q27 1
(since |N ( 5)| = 3 72 implies ( 5) = P3 Q27 since 5 Q7 and 5
/ R7 ).
Continuing, we nd P3 1 and Q7 1. So Cl(K) is trivial.

Exercise 2.17. Compute Cl(Q( 46)). (It will be the trivial grape.)
We showed last time how to compute the ideal class grape, modulo being able to gure out if a given
ideal is principal. We do an example of the last question.

Example 2.18. Say K = Q( 10); then OK = Z[ 10]. Consider P = (2, 10). Is P principal?
Well, if P is principal, then P = () for some , and |N ()| = N (P ) = 2. But = a + b 10 for some
a, b Z. So
a2 10b2 = 2
Reducing this equation modulo 5, we nd that a2 2 (mod 5), which has no solutions for a. So P is not
principal.
This strategy works for all quadratic number elds.
Example 2.19. Let K = Q() where 3 2 + 5 = 0.
1. It turns out then that OK = Z[]. (One checks that the discriminant is 643, which is prime, and in
particular is square-free.)
2. The Minkowski constant is then
  r2  
4 n! p 4 6
M= |disc(K)| = 643 8
nn 27

(Observe that making the Minkowski constant a bit bigger wont be problematic; it may just make a
bit more work.)
3.
3 16 = 24 = N ( + 3)
2 1 = N ( + 2)
1 6 =23 = N ( + 1)
0 5 = N ()
1 4 = 22 = N ( 1)
2 9 = 32 = N ( 2)
3 26 = 2 13 = N ( 3)

14
 4.

(2) = P2 Q2
|{z} |{z}
norm 2 norm 4
(3) = P3 Q3
|{z} |{z}
norm 3 norm 9
(5) = P5 Q5
|{z} |{z}
norm 5 norm 25
(7)

5.

P2 Q2 1
P3 Q3 1
P5 Q5 1
P2 P3 1
P5 1
P32 1

Manipulating, we nd that P2 P3 and P2 Q2 . So Cl(K) = hP2 i with P22 1. Is P2 principal?

Well, P2 = (2, + 1). Our plan is to nd a small box B such that if there is an element of norm 2 then
there is an element of norm 2 in B; we will then look in B. Now,

K (a) = (f1 (a), f2 (a)) R C

Checking by hand, we nd that

10
1 < |f2 ( + 2)| <
3
If |N (u)| = 2, then by multiplying and dividing by appropriate powers of + 2 (which has norm 1),
we can ensure that
10
1 |f2 (u)|
3
Since |N (u)| = 2, we get that |f1 (u)||f2 (u)| = 2; hence
3
|f1 (u)| 2
5
So our box is

|f1 (u)| 2
10
|f2 (u)|
3
This box contains 1 , , 1, 2 , and other elements that obviously dont have norm 2. (e.g.
2 2 has a factor of 2, and hence its norm has a factor of 4, and is not 2.)
What does the grape OK

look like? Well, it contains elements of nite order, namely the roots of unity
in K.
Denition 2.20. Let ! !
r1
Y r2
Y
VK = R C
i=1 i=1

be Minkowski space. We dene a map

r1Y
+r2
log : VK0
i=1

15
(where VK0 = VK \ { (x1 , . . . , xr1 , zr1 +1 , . . . , zr1 +r2 ) : x1 xr1 zr1 +1 zr1 +r2 = 0 }) by

log(x1 , . . . , xr1 , zr1 +1 , . . . , zr1 +r2 ) = (log(|x1 |), . . . , log(|xr1 |), 2 log(|zr1 +1 |), . . . , 2 log(|zr1 +r2 |))

(The base doesnt particular matter; we can choose e.)

The point of log is to turn the multiplicative set OK

into the additive subset log(K (OK

)).
Remark 2.21. One notes that log and K are homomorphisms; what is ker(log K )? Its the grape of
roots of unity. Indeed, if w is a root of unity, then |fi (w)| = 1 for all i, and log(K (w)) = 0. Conversely, if
log(K (w)) = 0, then |fi (w)| = 1 for all i; hence every power of w is contained in the nite set K (OK ) B
(where B = { v : |vi | 1 }). Hence wa = wb for some a 6= b, and w is a root of unity.
Hence
i log K
0W OK 0
is a short exact sequence, where W is the grape of roots of unity in K and = log(K (OK

)); i.e. log K
induces an isomorphism OK /W

= .
Sadly, is not a full lattice in Rr1 +r2 : if OK

, then |N ()| = 1; hence
r1
Y r2
Y 2
|fi ()| |fr1 +i ()| = 1
i=1 i=1

and thus
r1 r2
log(|fi ()| + 2 log(|fr1 +i ()|) = 0
X X

i=1 i=1

So log(K ()) H where H Rr1 +r2 is the hyperplane given by

rX
1 +r2

xi = 0
i=1

(where the xi are the coordinates in Rr1 +r2 ).

Theorem 2.22 (Dirichlet Unit Theorem). Let K be a number eld of degree d with r1 real embeddings and
r2 pairs of complex embeddings. Then OK
= W Zr1 +r2 1 where W is the grape of roots of unity in K.

Exercise 2.23. Compute OK
, where K = Q( 2).

It turns out OK
is generated by 1 and 1 + 2. By the Dirichlet unit theorem, we get that the torsion-
free
part of OK
is cyclic; hence, since 1 + 2 has innite order, it must be a power of the generator. But if
1 + 2 is a power of some OK
, then in particular every coordinate
of 1 + 2 in Minkowski spaceis a
power of the corresponding coordinate of ; i.e. fi ()n = fi (1 + 2) for all i. But then |fi ()| |fi (1 + 2)|
for all i, which
yields a box in Minkowski space; checking every element of the box, we nd that none of
them has 1 + 2 as a power. (In principal this would involve checking every power of everything in the box;
however, for any in the box we note that once n becomes large enough that n lands outside the box, we
can stop checking.)
Proof of Theorem 2.22. We have maps
r1 r2

Y Y log
K K
R C Rr1 +r2
i=1 i=1

Let = log|K (OK

)|. We will prove that is a full lattice in the (r1 + r2 1)-dimensional space H =
{ (x1 , . . . , xr1 +r2 ) : x1 + + xr1 +r2 = 0 }.
1. Modulo units, there are only nitely many elements of OK of norm N , for any xed N . Indeed, if
N () = N , then N (()) = |N |, and N (); so there are only nitely many choices for (), and hence
only nitely many choices for up to units.

16
 2. Note that is a discrete subset of Rr1 +r2 because K (OK ) is discrete in Minkowski space.
3. Let A be the fundamental domain of OK in Minkowski space. Pick C > 2d vol(A). Choose a1 , . . . , aN
OK such that every OK with 1 |N ()| C satises = uai for some u OK
and some
i { 1, . . . , N }.
4. Write K = (f1 , . . . , fr1 , fr1 +1 , . . . , fr1 +r2 ). Choose c1 , . . . , cr1 +r2 R>0 with c1 cr1 +r2 = C.
5. Dene
r1
Y r2
Y
X = { (x1 , . . . , xr1 , zr1 +1 , . . . , zr1 +r2 ) : |xi | < ci , |zi | < ci } R C
i=1 i=1

Set
N
[
T = [K (ai )]1 X
i=1

(where the reciprocal is taken component-wise).

6. Note that T is bounded. We will show that H is covered by the translates of log|T | by .
If

v H then the inverse image of log|X|

v contains a non-zero vector K () for some OK by


the convex body theorem. Hence if v = log(y) then () = xy 1 for some x X. Then N () < C
K
by denition of X and H, so u = ai for some i. So y = xK ()1 = xK (a1
i u); so


v = log|xK (a1 )| + log|K (u)|
|{z} | {z i } | {z }
H log|T |

So H is covered by -translates of log|T |. But log|T | has nite volume; so there is a fundamental
domain of of nite volume, and is a full lattice.
Theorem 2.22

Exercise 2.24. Compute Z[ 3] .
Fact 2.25. Any nite subgrape of F is cyclic if F is a eld.

3 Factorization of primes in extensions

Given an extension L/K of number elds and a non-zero prime ideal P of OK , how does P OL factor into
prime ideals of OL ?
Fact 3.1. If P is a prime ideal of OL , then Q = P OK is prime.
In this case we say that P lies over Q. Now, QOL is an ideal of OL , though not necessarily prime; we
also have that QOL P . So
Yr
QOL = P a Piai
i=1

where Pi 6= P for any i.

Denition 3.2. We set

e(P /Q) = a
f (P /Q) = [OL /P : OK /Q]

The former is called the ramication index of P over Q, and the latter is called the inertia degree of P over
Q (or sometimes the residue degree of P over Q).
Example 3.3. Let L = Q(i) and K = Q. Let Q = (2) and P = (1 + i). From previous work we know that
QOL = P 2 ; hence e(P /Q) = 2. We also have f (P /Q) = 1 because |OL /P | = |OK /Q| = 2.

17
Remark 3.4. In general we can compute inertia degrees just from the norms of P and Q, since the quotients
are always nite.
Example 3.5. Let L = Q(i) and K = Q. Let Q = (3) and P = (3). Then e(P /Q) = 1 and f (P /Q) =
[Z[i]/(3) : Z/(3)] = 2.
Example 3.6. Let L = Q(i) and K = Q. Let Q = (5) and P = (2 + i) or (2 i). (It doesnt matter which
because they are conjugate over K, and hence the ramication indices and inertia degrees are the same.)
Then e(P /Q) = 1 since QOL = (2 + i)(2 i), and f (P /Q) = [Z[i]/(2 + i) : Z/5] = 1 (since N (2 + i) = 5).
Remark 3.7. Suppose K L M are number elds; suppose P OM is prime, and let Q = P OL and
R = P OK = Q OK . Then
e(P /R) = e(P /Q)e(Q/R)
f (P /R) = f (P /Q)f (Q/R)
(This last is because [OM /P : OK /R] = [OM /P : OL /Q][OL /Q : OK /R].)
Theorem 3.8. Suppose L/K be number elds; suppose Q OK is a non-zero prime ideal. Factor QOL =
P1e1 Prer . Then
Xr
e(Pi /Q)f (Pi /Q) = [L : K]
i=1

Proof. We check the case K = Q. Note that

r
Y
N (QOL ) = N (Pi )ei
i=1

Since K = Q, we have that

N (QOL ) = |OL /QOL |
and we may write Q = (q). But by the Chinese remainder theorem we have
OL /QOL
= OL /P1e1 OL /Prer
where ei = e(Pi /Q). Hence
[L:K]
q [L:K] = |OK /Q|
= |OL /QOL |
= |OL /P1e1 | |OL /Prer |
= N (P1 )e1 N (Prer )
= (q f (P1 /Q) )e(P1 /Q) (q f (Pr /Q) )e(Pr /Q)
as desired. Theorem 3.8
Denition 3.9. Suppose L/K be an extension of number elds; suppose a L. Consider T : L L given
by Ta (x) = ax; then Ta is a K-linear transformation. We then dene
trL/K (a) = tr(Ta )
NL/K (a) = det(Ta )
Remark 3.10.
r
trL/K (a) =
X
fi (a)
i=1
Yr
NL/K (a) = fi (a)
i=1

where f1 , . . . , fr : L K are the embeddings of L into K. That is, x an embedding : K , C and let
f1 , . . . , fr be the embeddings L , C such that fi  K = . Equivalently, regard K and L as K-algebras,
and let f1 , . . . , fr be the K-embeddings (that is, embeddings of K-algebras) L , K.

18
 h i
Exercise 3.11. Let L = Q( 2, 3) and K = Q( 2). One checks that OL = Z 2, 2+
2
6
and that
disc(OL ) = 28 32 . Factor (2), (3), and (5) in OL and compute all of the ramication indices and inertia
degrees that come up.

Example 3.12. Let K = Q( 2) and L = Q( 2, 5). Let a = 2 + 5. Then

trL/K (a) = ( 2 + 5) + ( 2 5)

=2 2

NL/K (a) = ( 2 + 5)( 2 5)
= 3

(using Remark 3.10). Alternatively, we can use the denitions: we use the basis { 1, 5 } for L over K. Then

Ta (1) = 2 + 5

( 2, 1)

Ta ( 5) = 2 5 + 5

(5, 2)

Hence  
2 5
[Ta ] =
1 2
and

trL/K (a) = 2 2
NL/K (a) = 3

Remark 3.13. Suppose K L M is a tower of elds. Then

NL/R (NM /L (a)) = NM /K (a)

trL/K (trM /L (a)) = trM /K (a)

We also have NL/K (a), trL/K (a) K, and if a OL then NL/K (a), trL/K (a) OK .
Remark 3.14. The converse to the last statement is false; there are a with NL/K (a), trL/K (a) OK but
a/ OL . (Note that NL/K (a) and trL/K (a) are, up to sign, coecients in (the monic minimal polynomial of
a over K raised to the power [L : K(a)]).)
We also want to dene NL/K (I) for an ideal I OL . Factor

I = P1a1 Prar

for prime ideals Pi OL . Now, if K = Q, then

NL/K (I) = (p1 )a1 f (Pi /pi ) (pr )ar f (Pr /pr )

(where Pi Z = (pi )).

Denition 3.15. We set
a f (P1 /Q1 )
NL/K (I) = Q1 1 Qar r f (Pr /Qr )
where Qi = Pi OK .

h 3.16.
Exercise
i
(Continuation of Exercise 3.11.) Let K = Q( 2) and L = Q( 2, 3). Then OL =
Z 2, 6+ 2
2
with disc(OL ) = 28 32 . Factor (2), (3), and (5), in OL , and compute all relevant e and
   
f values that arise. Further compute NL/K 6+ 2
2
and trL/K 6+ 2
2
.
Suppose now that L/K is a Galois extension. Say P OK is prime; factor P OK = Qe11 Qgg .
e

19
Claim 3.17. e1 = = eg and f1 = = fg .

Example 3.18. This isnot necessarily true if L/K is not Galois. Consider

3
L = Q( 2), K = Q, and P = (5).
Then OL contains Z[ 2] as a subring of nite index, and 5 - disc(Z[ 2]). So (Z[ 3 5])QZ[ 3 2] = (OL )Q and
3 3

= OL /Q for any Q containing 5. Hence we can do our computations in Z[ 5] instead of
Z[ 3 2]/(Q Z[ 3 2]) 3

OK . Now

Z[ 2]/(5)
= Z[x]/(x3 2, 5)
3

= (Z/5Z)[x]/(x3 2)

= (Z/5Z)[x]/(x + 2)(x2 2x 1)

Hence
3 3 3
(5) = ( 2 + 2, 5) ( 4 2 2 1, 5)
| {z }| {z }
f =1 f =2

Proof of Claim 3.17. This follows from the fact that if Qi OK = Qj OK for prime ideals Qi , Qj OL ,
then there is some element Gal(L/K) satisfying (Qi ) = Qj ; it remains to check this fact.
Fix i; choose Qi such that 1 (mod Qj ) for all j 6= i. (Possible by the Chinese remainder
theorem.) Then NL/K () P OK . But
Y
NL/K () = () Qj
Gal(L/K)

for all j, because P Qj for all j. Hence for each j there is some j in the Galois grape with j () Qj .
But j () 1 (mod QK ) for all other k; so j ()
/ Qk for k 6= j. So j (Qi ) = Qj . Claim 3.17
Given Gal(L/K), can we reduce modulo P for some prime ideal P OL ?
We cannot!

Example 3.19. Consider L = Q( 2, K = Q, and P = (7, 2 3). Then

OL /P
= Z[ 2]/(7, 2 3)

= Z[x]/(x2 2, x 3, 7)

= (Z/7Z)[x]/(x2 2, x 3)
= (Z/7Z)[x]/(x 3)

= Z/7Z

But if is the non-trivial automorphism of L over K, we would be trying to ll in the following diagram:

OL OL
q
q q

OL /P OL /P

The universal property of quotients tells us that Gal((OL /P )/(OK /P K)) exists exactly when P
ker(q ) = 1 (P ); i.e. when P = (P ).
Denition 3.20. For xed P , we have that DP = { Gal(L/K) : (P ) = P } is a subgrape of Gal(L/K),
called the decomposition grape of P . We then get a homomorphism P : DP Gal((OL /P )/(OK /P K)),
called the decomposition homomorphism. This homomorphism need not be injective; we thus dene the
inertia grape of P to be

IP = ker(P ) = { DP : = 1 in Gal((OL /P )/(OK /P K)) }

20
 h i
Exercise 3.21. As in Exercise 3.16, we let L = Q( 2, 3), so OL = Z 2, 6+
2
2
, and we let K = Q so
that OK = Z. We saw that
!
6+ 2
P2 = + 1, 2
2
!
6+ 2
P3 = + 2, 3
2
!
6+ 2
P5 = 3 2 + 3, 5
2

were prime. Find DP and IP for all of these ideals.

Denition 3.22. Suppose L/K is an extension of number elds; suppose I OL is an ideal. The codierent
of I is
I = { x L : trL/K (xI) OK }
This turns out to be a fractional ideal of OL . The codierent of L/K is the codierent of OL over K. The
dierent of I is (I )1 ; the dierent of L/K is DL/K = (OL ) .
1

Example 3.23. We compute the codierent of (1) in Z[ 2] over Q; i.e. the codierent
of Q( 2) overQ. Note
that tr(a + b 2) = 2a; we thus require that 2a Z. Further note that tr((a + b 2) 2) = tr(2b + a 2) = 4b;
we thus require that 4b Z. Hence
( ) !
k ` 2 2
(1) = + : k, ` Z =
2 4 4

We thus also get that the dierent is DQ(2)/Q = (I )1 = (2 2).
Fact 3.24. II = OL

.
Proof. It is easily seen that II OL

; hence

I I 1 OL

I

and I 1 OL

= I . Fact 3.24
Fact 3.25. If I OL then (I )1 OL .
Proof. This is just because if I J then J I . Fact 3.25
h i
Example 3.26. Compute the dierent of L = Q( 5) over Q. Well, OL = Z 1+2 5 . We want tr(a + b 5) =
  
2a Z and tr (a + b 5) 1+2 5 = a + 5b Z for a, b Q. We thus get that a + b 5 = 2` + 15 m 2` 5.

Hence ( ! ! ) ! !
5 5 5 5 5 5 5
OL = ` +m : `, m Z = , =
10 5 10 5 5
 
(since 510 5 = 55 1+2 5 ). Hence DL/K = ( 5).

Fact 3.27. DM /K = DM /L DL/K . Also, if L/K is Galois, then for all Gal(L/K) we have (DL/K ) =
DL/K .

Denition 3.28. The discriminant of L/K is L/K = NL/K (DL/K ).

Theorem 3.29. (disc(OK )) = K/Q .

21
Proof. We will show that disc(OK )2 generates 2K/Q . For any ideal I OK , we have that I = a1 Z+ +an Z
for some a1 , . . . , an OK . The ideal I is a1 Z + + an Z where ai K is given by tr(ai aj = ij . So OK

(the codierent of OK ) is a1 Z + + an Z where OK = a1 Z + + an Z. But OK is a fractional ideal of

OK ; so there is some non-zero m Z with mOK

OK . Dene I = mOK
; then

1 m2n m2n K K
NK/Q (DK/Q )2 = = = =

NK/Q (OK ) 2 NK/Q (I)2 disc(ma1 , . . . , man ) disc(a1 , . . . , an )

(since OK ) = DK/Q ). (Here we regard NK/Q (DK/Q ) as an integer by identifying it with a generator of
1

this ideal.) (Here K refers to disc(K) = disc(a1 , . . . , an ).)

Claim 3.30. disc(a1 , . . . , an ) = disc(a1 ,...,an ) .
1

Proof. We note that

2
f1 (a1 ) fn (a1 )
.. .. ..
disc(a1 , . . . , an ) = det . . .
f1 (an ) fn (an )
2
f1 (a1 ) f1 (an )

disc(a1 , . . . , an ) = det ... ..

.
..
.

fn (a1 ) fn (an )

But the (i, j) entry of the product of these two matrices is

f1 (ai )f1 (aj ) + + fn (ai )fn (aj ) = trK/Q (ai aj ) = ij

Hence disc(a1 , . . . , an ) = disc(a1 ,...,an ) .

1
Claim 3.30

Theorem 3.29
Fact 3.31. M /K =
[M :L]
L/K NL/K (M /L ).

Some hard facts:

Fact 3.32.
1. Say P OL is prime; let Q = P OK and e = e(P /Q). Then P e1 | DL/K and if gcd(e, NL/Q (P )) = 1
then P e - DL/K .
2. Suppose n Z; suppose S is a nite set of prime ideals of OK . Then the set of extensions L/K with
[L : K] n and L ramied only over primes in S is nite.
3. (Due to Hermite.) Suppose n Z. Then there are nitely many number elds with discriminant at
most n.
h i
Example 3.33. Let L = Q( 2, 3) so that OL = Z 2, 6+ 2
2
; let K = Q( 2) so that OK = Z[ 2].
Compute L/K .
Steps:

1. Compute OL

(codierent, not grape of units) over K.
2. Invert it to get DL/K .
3. Compute NL/K (DL/K ) = L/K .

22
4 InterludeFinite elds
Suppose we have a eld extension K of Fp with pd elements; so [K : Fp ] = d. Then K has pd 1 elements
and is a grape; so K is the splitting eld of xp 1. One checks that it is also a separable extension; hence
d

every extension of nite elds is Galois. But splitting elds are unique up to isomorphism; hence up to
isomorphism there is exactly one eld with pd elements.
How do they relate? In partial diagram:

Fp4 Fp6

Fp2 Fp3 Fp5

Fp

In general, the lattice of nite elds of characteristic p is isomorphic to the lattice of positive integers under
the divisibility relation.
What of the Galois theory? What is Gal(Fpk /Fp )? There is a natural automorphism Frobp : Fpk Fpk
given by Frobp () = p . By Fermats little theorem we get that Frobp xes Fp pointwise. Is it true that
Gal(Fpk /Fp ) = hFrobp i?
Well, |Gal(Fpk /Fp )| = k. What is |hFrobp i|? Well, (Frobp )n () = p ; so (Frobp )n is trivial if and only
n

if = p for all Fpk . But Fpk is the splitting eld of xp x; so the order of Frobp divides k. Can
n k

it be smaller? Well, any xed by (Frobp )n satises xp x = 0, and there are no more than pn such ;
n

hence the order of Frobp is k.

So Gal(Fpk /Fp ) = hFrobp i. So Gal(Fpk /Fp ) is cyclic; thus Gal(Fpn /Fpm ) is also cyclic, and is generated
by (Frobp )m .
Exercise 4.1. Let a F53 = F5 (a) be a root of x3 + 3x + 7. Express the other two roots as x + ya + za2 for
some x, y, z F5 .

5 Slightly less nite elds

Suppose now we have a Galois extension of number elds L K; suppose we have P OL lying over
Q OK . Recall that

DP = { Gal(L/K) : (P ) = P }
IP = { DP : id (mod P ) }

Denition 5.1. Suppose L/K be a Galois extension of number elds; suppose P OL is a prime ideal
lying over Q OK . The decomposition eld of P over K, denoted ZP is the xed eld of DP . The inertia
eld of P over K, denoted FP , is the xed eld of IP . (This notation is not standard.)
Remark 5.2. If P1 and P2 both lie over Q, then DP1 and DP2 are conjugate; so ZP1 and ZP2 are isomorphic.

Theorem 5.3. Suppose L/K is a Galois extension of number elds; suppose P OL is prime and Q =
P OK . Let Z = ZP be the decomposition eld of P ; let P = P OZ . Then
1. P is the only prime ideal of OL that lies over PZ .
2. [L : Z] = |DP | = e(P /Q)f (P /Q).

3. e(PZ /Q) = f (PZ /Q).

Proof.

23
1. Well, Gal(L/Z) = DP ; so Gal(L/Z) xes P . But Gal(L/Z) acts transitively on the primes lying over
PZ (see the proof of Claim 3.17).
2. Well,
e(P /Q)f (P /Q)[L : K]
[L : K] = e(P /Q)f (P /Q)(index of DP in Gal(L/K)) =
|DP |
Hence |DP | = e(P /Q)f (P /Q).

3. Well, [L : Z] = e(P /PZ )f (P /PZ ); so [L : K] = e(P /PZ )f (P /PZ )[Z : K]. But [L : K] = e(P /Q)f (P /Q)[Z :
K]; so e(P /PZ )f (P /PZ ) = e(P /Q)f (P /Q). So e(PZ /Q) = f (PZ /Q) = 1. Theorem 5.3

24