Algebraic Number Theory
Andrew Kobin Spring 2016 Contents Contents
Contents
0 Introduction 1 0.1 Attempting Fermat’s Last Theorem ...... 1
1 Algebraic Number Fields 4 1.1 Integral Extensions of Rings ...... 4 1.2 Norm and Trace ...... 5 1.3 The Discriminant ...... 6 1.4 Factorization of Ideals ...... 11 1.5 Ramification ...... 15 1.6 Cyclotomic Fields and Quadratic Reciprocity ...... 21 1.7 Lattices ...... 24 1.8 The Class Group ...... 26 1.9 The Unit Theorem ...... 31
2 Local Fields 34 2.1 Discrete Valuation Rings ...... 34 2.2 The p-adic Numbers ...... 37 2.3 Absolute Values ...... 42 2.4 Local Fields ...... 49 2.5 Henselian Fields ...... 52 2.6 Ramification Theory ...... 54 2.7 Extensions of Valuations ...... 61 2.8 Galois Theory of Valuations ...... 65 2.9 Higher Ramification Groups ...... 69 2.10 Discriminant and Different ...... 75
i 0 Introduction
0 Introduction
These notes follow a course on algebraic number theory taught by Dr. Andrew Obus at the University of Virginia in Spring 2016. The main topics covered are:
Algebraic number fields (the global case)
The ideal class group
Structure of the unit group
The p-adic numbers (the local case)
Hensel’s Lemma
Ramification theory
Further topics, including adeles and ideles
The main companion for the course is Neukirch’s Algebraic Number Theory. Other great ref- erences include Cassels and Frohlich’s Algebraic Number Theory, Janusz’s Algebraic Number Fields, Lang’s Algebraic Number Theory, Marcus’s Number Fields and Weil’s Basic Number Theory.
0.1 Attempting Fermat’s Last Theorem
Algebraic number theory was developed primarily as a set of tools for proving Fermat’s Last Theorem. We recall the famous (infamous?) theorem here.
Fermat’s Last Theorem. The equation xn + yn = zn has no solutions in positive integers for n ≥ 3.
In attempting to prove the theorem, we first remark that the n = 4 case is elementary; it’s just a matter of parametrizing the Pythagorean triples (x, y, z) that solve x2 + y2 = z2 and noticing that not all three can be perfect squares. With this, we can reduce to the case when n = p, an odd prime. There are two cases:
Case 1: x, y, z are all relatively prime to p.
Case 2: p divides exactly one of x, y, z.
We will show a proof for the first few primes in Case 1; the other case uses similar tech- niques. Let ζ be a primitive pth root of unity (e.g. ζ = e2πi/p) and assume Z[ζ] is a unique factorization domain (UFD). This was the classical approach, but number theorists quickly realized that Z[ζ] is not always a UFD. In fact, it is an open question whether there are an infinite number of primes p for which Z[e2πi/p] is a UFD. In any case, the assumption that Z[ζ] is a UFD holds for p < 23 so we will have proven a number of cases of Fermat’s Last Theorem with the following proof.
1 0.1 Attempting Fermat’s Last Theorem 0 Introduction
Proof. Suppose x, y, z are positive integers satisfying xp + yp = zp. We may assume x, y, z are relatively prime in Z. The equation above may be factored as
p Y (x + ζiy) = zp (∗) i=1
For p = 3, the only cubes mod 9 are ±1 and 0 so there are no solutions for (*) where 3 - xyz. So we may assume p ≥ 5. We need the following lemmas: p−1 Y Lemma 0.1.1. p = (1 − ζi). i=1
tp−1 Proof. Consider expanding t−1 in two ways: tp − 1 (t − ζ) · (t − ζp−1) = = tp−1 + ... + t + 1. t − 1 Then plugging in t = 1 gives the result. Lemma 0.1.2. For any 0 ≤ i < j ≤ p − 1, the elements x + ζiy and x + ζjy are coprime in Z[ζ]. Proof. Suppose that π ∈ Z[ζ] is a prime which divides x + ζiy and x + ζjy. Then π divides ζiy(1 − ζj−i). Notice that ζi is a unit and p - y by assumption, but 1 − ζj−i | p. So in particular, π | y and thus π | yp. Since π is a prime, π | y or π | p. Repeating the argument for x shows that π | x or π | p. Since x and y are coprime in Z, we cannot have π | x and π | y simultaneously, so π | p. By assumption we have that π divides xp + yp and therefore also zp in Z, but (p, z) = 1 so the Euclidean algorithm implies that π | 1. Therefore x + ζiy and x + ζjy are relatively prime in Z[ζ]. Now, each factor x + ζiy must be a pth power in Z[ζ], possibly multiplied by a unit. Write x + ζy = utp for u ∈ Z[ζ]∗ and t ∈ Z[ζ]. Lemma 0.1.3. u/u¯ is a pth root of unity.
Proof. It is simple to show that u/u¯ and all of its Galois conjugates have modulus 1 in C; this is then true for all powers of u/u¯ as well. Then the degree of u/u¯ and all of its powers is bounded. Since all of these are algebraic integers, there are only finitely many possible choices for their minimal polynomials. Hence the set {(u/u¯)k : k ∈ N} is finite. This proves u/u¯ is a root of unity in Z[ζ]. In particular, (u/u¯)2p = 1 but we want to show it is a pth root of unity. Suppose (u/u¯)p = −1. Then up = −u¯p. Since u ∈ Z[ζ] we may write
2 p−2 u = a0 + a1ζ + a2ζ + ... + ap−2ζ
for unique ai ∈ Z; this follows from unique factorization in Z[ζ]. Now
p p p p u ≡ a0 + a1 + ... + ap−2 (mod p)
≡ a0 + a1 + ... + ap−2 (mod p) by Fermat’s Little Theorem.
2 0.1 Attempting Fermat’s Last Theorem 0 Introduction
In particular, up is conjugate to a real number mod p. Likewise, we can write −u¯ as −u¯ = p−1 2 −(a0 + a1ζ + ... + ap−2ζ ) so
p −u¯ ≡ −a0 − a1 − ... − ap−2 (mod p).
p This implies a0 + a1 + ... + ap−2 ≡ 0 (mod p) so p | u . However, this is impossible if u is a unit. Therefore (u/u¯)p = 1. Putting these results together, we can now write
x + ζy = ζjut¯ p ≡ ζju¯t¯p ≡ ζj(x + ζy¯ ) (mod p).
Expanding this out gives us
x + ζy − ζjx − ζj−1y ≡ 0 (mod p). (∗∗)
∼ p−1 ∼ p−1 Now Z[ζ]/(p) = Z[x]/(p, x + ... + x + 1) = Fp[x]/(x + ... + x + 1). Thus the images p−2 p−1 of 1, x, . . . , x are Fp-linearly independent in this ring. This implies 1, ζ, . . . , ζ are Z- linearly independent in Z[ζ]/(p). Since x, y ∈ Z, the only possibilities in (**) for j are j = 0, 1, 2, p − 1. If p = 0, 2, p − 1, it is easy to simplify (**) and produce a nontrivial ζ2 term, which is impossible. If j = 1, (**) becomes
(x − y)(1 − ζ) ≡ 0 (mod p).
Qp−1 i Thus i=2 (1 − ζ ) divides x − y but since x − y ∈ Z, it must be that p | (x − y). Rearranging the equation xp + yp = zp to read xp + (−z)p = yp and repeating the argument so far shows that p | (x + z) as well. Thus y ≡ x ≡ −z (mod p). But then
0 = xp + yp − zp ≡ 3xp (mod p)
which implies p | x, contradicting the assumption that p 6= 3. Therefore no solutions exist to xp + yp = zp for p > 5 such that p - xyz. This proof fails for general primes p in two places: as we mentioned, not every ring Z[e2πi/p] is a UFD; moreover, there can be many more units than just the roots of unity in Z[e2πi/p]. This motivates the study of ideal class groups – which measure how far from being a PID (and a UFD) a ring of integers is – and unit groups in algebraic number theory.
3 1 Algebraic Number Fields
1 Algebraic Number Fields
1.1 Integral Extensions of Rings
Let A ⊆ B be rings.
Definition. An element x ∈ B is integral over A if it is a root of a monic polynomial with coefficients in A. We say B is integral over A if every element of B is integral over A.
Definition. The integral closure of A in B is the set of all x ∈ B which are integral over A. If A is equal to its integral closure in B then we say A is integrally closed in B. In particular, if A is a domain and B is the fraction field of A then we simply say that A is integrally closed.
Lemma 1.1.1. x ∈ B is integral over A if and only if A[x] is a finitely generated A-module.
n n−1 n Pn−1 i Proof. ( =⇒ ) If x + an−1x + ... + a0 for ai ∈ A then x ∈ M := i=1 Ax which is a finitely generated A-module. By induction, for all m ≥ n, xm ∈ M. This implies A[x] = M, so in particular A[x] is finitely generated. ( ⇒ = ) Suppose A[x] is generated by f1(x), . . . , fn(x) where fi are polynomials in a single n variable over A. Let d ≥ max{deg fi}i=1. Then
n d X x = aifi(x) i=1
d Pn for some choice of ai ∈ A. This shows that x is a root of the polynomial t − i=1 aifi(t) so x is integral over A.
Theorem 1.1.2. The integral closure of A in B is a ring.
Proof. It suffices to prove that the integral closure A¯ is closed under the addition and mul- tiplication of B. If x, y ∈ A¯, Lemma 1.1.1 shows A[x, y] is finitely generated. This implies that the submodules A[x + y] and A[xy] are also finitely generated, so x + y, xy ∈ A¯. Hence A¯ is a ring. Let A ⊂ B be a subring. We will make use of the following facts about integral extensions of rings:
Every UFD is integrally closed.
If A is a domain, B is finite over A if and only if B is integral over A and B is finitely generated as an A-module.
Suppose C ⊇ B ⊇ A are all rings. If C is integral over B and B is integral over A then C is integral over A.
If B is integral over A then S−1B is integral over S−1A for any multiplicatively closed subset S ⊂ A.
4 1.2 Norm and Trace 1 Algebraic Number Fields
The two most important objects in global algebraic number theory are defined next.
Definition. K is a number field if K is a finite field extension of Q. Definition. For a number field K ⊃ Q, the integral closure of Z in K is called the ring of integers of K, written OK . Examples.
1 The ring of integers of Q is Z. √ √ h 1+ −3 i 2 For K = Q( −3), the ring of integers is OK = Z 2 .
2πi/p 3 For a prime p, the cyclotomic field K = Q(ζp) = Q(e ) has ring of integers OK = Z[ζp].
It turns out that OK is always a free Z-module of rank [K : Q]. Thus we can think of OK as a lattice embedded in the vector space K.
1.2 Norm and Trace
Two important maps for understanding number fields are introduced in this section. Let L/K be a finite field extension and fix x ∈ L.
Definition. The norm of x is the element NL/K (x) = det Tx ∈ K, where Tx : L → L is the K-linear map Tx(`) = x`.
Definition. The trace of x is TrL/K (x) = tr Tx, where tr denotes the trace. Note that the norm and trace are defined for any finite extension L/K, not just number fields. We will often drop the subscript and write N(x) and Tr(x) when the extension is understood.
× × Lemma 1.2.1. The norm map NL/K : L → K is a homomorphism of multiplicative groups, and the trace map TrL/K :(L, +) → (K, +) is a homomorphism of abelian groups.
Theorem 1.2.2. Suppose L/K is a finite, separable extension of fields. Let σ1, . . . , σn be the distinct embeddings L,→ K where K is the algebraic closure of K. Then for all x ∈ L,
n n Y X NL/K (x) = σi(x) and TrL/K (x) = σi(x). i=1 i=1
n−1 Proof. Assume σi(x) 6= σj(x) when i 6= j. A basis of L/K is 1, x, . . . , x and the matrix for Tx in this basis is 0 0 ··· 0 −a0 1 0 ··· 0 −a1 . .. 0 1 0 −a2 . . . . . ...... 0 0 ··· 1 −an−1
5 1.3 The Discriminant 1 Algebraic Number Fields
n where f(x) = a0 + a1x + ... + anx is the minimal polynomial of x over K. In this case f is also the characteristic polynomial of x, so by linear algebra, Tr(x) is equal to the sum of the roots of f and N(x) is equal to the product of the roots of f. This implies the result. √ Example 1.2.3. Let K = Q( d) for d a squarefree√ integer√ (this means d = ±p1p2 ··· pr in its prime factorization). Then an element x = a + b d ∈ Q( d) has norm N(x) = a2 − b2d and trace Tr(x) = 2a.
1.3 The Discriminant
In this section let L/K be a finite, separable extension of fields and let {α1, . . . , αn} be a K-basis of L, so that [L : K] = n. Also denote by σ1, . . . , σn : L,→ K the n distinct K-embeddings of L into the algebraic closure of K.
Definition. The discriminant of the basis {α1, . . . , αn} is
2 dL/K (α1, . . . , αn) = [det(σi(αj))] .
Proposition 1.3.1. Let A = [TrL/K (αiαj)]. Then dL/K (α1, . . . , αn) = det A. In particular, dL/K (α1, . . . , αn) lies in K. Pn Proof. By Theorem 1.2.2, TrL/K (αiαj) = k=1 σk(αi)σk(αj). Thus A = BC, where
T B = (σk(αi)) and C = (σk(αj)).
2 Taking the determinant gives us det A = (det B)(det C) = (det C) = dL/K (α1, . . . , αn). One case of interest is when L = K(α) is a simple extension and {1, α, α2, . . . , αn−1} is a basis for L as a K-vector space. Then the discriminant of α is defined to be
2 n−1 dL/K (α) := dL/K (1, α, α , . . . , α ).
Lemma 1.3.2. For any algebraic element α over K, dL/K (α) equals the discriminant of the minimal polynomial of α.
Proof. Set L = K(α) and let αi = σi(α) for each embedding σi : L,→ K. Then
n−1 1 α1 ··· α1 1 α ··· αn−1 2 2 dL/K (α) = det . . .. . . . . . n−1 1 αn ··· αn This is a Vandermonde determinant, which evaluates to
Y Y 2 dL/K (α) = (αi − αj) = (αi − αj) . 1≤i,j≤n 1≤i Since K(α)/K is separable, dL/K (α) 6= 0. In fact, the product formula above is precisely the discriminant of f, the minimal polynomial of α over K. 6 1.3 The Discriminant 1 Algebraic Number Fields Proposition 1.3.3. For any K-basis {α1, . . . , αn} of L, dL/K (α1, . . . , αn) 6= 0. Proof. Since L/K is finite and separable, L = K(θ) for some θ ∈ L by the primitive element n−1 theorem. Then by Lemma 1.3.2, dL/K (1, θ, . . . , θ ) 6= 0. Let A ∈ GLn(K) be the change n−1 of basis matrix from {α1, . . . , αn} to {1, θ, . . . , θ }. Then for each 1 ≤ i, j ≤ n, j−1 det(σi(αj)) = (det A)(det(σi(θ )). Both determinants on the right are nonzero, so det(σi(αj)) 6= 0 which implies finally that dL/K (α1, . . . , αn) 6= 0 by the definition of disciminant. The proof of Proposition 1.3.3 gives us the following useful formula: If A, B are two K-bases for L with change of basis matrix A, then 2 dL/K (A) = (det A) dL/K (B). √ Example 1.3.4. √Take our favourite example, K = Q( d) over Q, where d is a squarefree integer. Then {1, d} is a basis for L, and its discriminant is √ 2 √ 1 d √ d (1, d) = det √ = (−2 d)2 = 4d. K/Q 1 − d This matches the fact that the discriminant of x2 − d is 4d. Suppose A ⊆ K is integrally closed with fraction field K. Let B be the integral closure of A in L. Observe that if x ∈ B then all conjugates of x in K are integral over K. Thus NL/K (x) ∈ A and TrL/K (x) ∈ A since A is integrally closed. × × Lemma 1.3.5. If x ∈ B then NL/K (x) ∈ A . Proof. By Lemma 1.2.1, NL/K is a homomorphism of groups. Lemma 1.3.6. Suppose α1, . . . , αn ∈ B form a K-basis of L. Let d = dL/K (α1, . . . , αn). Then dB ⊆ Aα1 + ... + Aαn. Pn Proof. Let a1, . . . , an ∈ K such that α := i=1 aiαi ∈ B. Then (a1, . . . , an) is a solution to the system of linear equations n X TrL/K (αiα) = TrL/K (αiαj)xj, 1 ≤ i ≤ n. j=1 The matrix corresponding to this system has determinant d by Proposition 1.3.1. Thus 1 each aj can be written as d times an A-linear combination of Tr(αiα). Since αi, α ∈ B, Tr(αiα) ∈ A so dαj ∈ A for each j. Thus n X dα = dajαj ∈ Aα1 + ... + Aαn. j=1 Since α ∈ B was arbitrary, we have shown that dB ⊆ Aα1 + ... + Aαn. 7 1.3 The Discriminant 1 Algebraic Number Fields Proposition 1.3.7. If A is a PID and M ⊆ L is a finitely generated B-module, then M is free of rank n = [L : K] as an A-module. In particular, B is free of rank n as an A-module. Proof. Let {α1, . . . , αn} ⊂ B be a basis for L/K. We know the rank of B, which is well- defined over a PID, is at most n. On the other hand, since the αi are linearly independent, the rank of B is at least n. Thus the rank of B equals n. Now suppose M is finitely generated as a B-module, say by elements µ1, . . . , µr. Then there exists an a ∈ A such that aµi ∈ B for each i (by a homework problem). Thus daµi ∈ M0 := Aα1 + ... + Aαn by Lemma 1.3.6. So daM ⊆ M0. By the structure theory of modules over a PID, since M0 is free, daM is also free, so M is free of rank at most n. On the other hand, rank M ≥ rank B = n by assumption so rank M = n. The second statement follows from taking M = B. Definition. In the situation above, an A-basis for B is called an integral basis. The most important case of Proposition 1.3.7 is when K = Q, A = Z and B ⊂ L is the integral closure of A in some number field L. Proposition 1.3.8. Let {α1, . . . , αn} and {β1, . . . , βn} be two integral bases for B/Z. Then dL/Q(α1, . . . , αn) = dL/Q(β1, . . . , βn). 0 2 0 Proof. Let d = dL/Q(α1, . . . , αn) and d = dL/Q(β1, . . . , βn). Then d = (det M) d for some 0 M ∈ GLn(Z). Thus det M = ±1 so d = d . This allows us to define: Definition. The discriminant of a number field K/Q is dK := dK/Q(α1, . . . , αn) for any Z-basis {α1, . . . , αn} of OK . √ √ Example 1.3.9. The quadratic field K = Q(√ 2) has integral basis {1, 2}. Then dK = 8. In general, for a quadratic extension K = Q( d), the discriminant is given by ( 4d, d ≡ 2, 3 (mod 4) dK = d, d ≡ 1 (mod 4) Example 1.3.10. Let ζ be a primitive prth root of unity and let K = Q(ζ). We know that r r−1 r [K : Q] = ϕ(p ) = p (p − 1). Set n = ϕ(p ). We will show that OK = Z[ζ] for every r prime power p . First, we compute the discriminant dK = dK/Q(ζ). To do so, we need the following lemma. Lemma 1.3.11. If f is the minimal polynomial of α and K = Q(α), 0 dK/Q(α) = ±NK/Q(f (α)). 8 1.3 The Discriminant 1 Algebraic Number Fields Proof. Let σ1, . . . , σn be the Q-embeddings of K into Q. We may assume σ1 = id : K,→ Q. Then by Lemma 1.3.2, Y 2 dK/Q(α) = (σi(α) − σj(α)) . 1≤i n 0 Y f (α) = (α − σi(α)). i=2 Finally, compute the norm of f 0(α): n ! 0 Y Y NK/Q(f (α)) = σj (α − σi(α)) 1≤j≤n i=2 Y = (σi(α) − σj(α)) 1≤i6=j≤n Y 2 = ± (σi(α) − σj(α)) = dK/Q(α). 1≤i (Note: in fact, Lemma 1.3.11 holds for any finite separable extension L/K.) Now, to compute dK/Q(ζ) in our example, let f(x) be the minimal polynomial of ζ over Q. We may write this in two ways: pr x − 1 r−1 r f(x) = or (xp − 1)f(x) = xp − 1. xpr−1 − 1 Differentiating the second expression gives us f 0(x)(xpr−1 − 1) + f(x)(pr−1xpr−1−1) = prxpr−1. Then plugging in ζ and solving for f 0(ζ) produces prζr−1 f 0(ζ) = . ζpr−1 − 1 Take the norm of this expression: N(pr) N(f 0(ζ)) = ± = ±pa for some a ∈ . N(ζpr−1 − 1) Z a r r Thus by Lemma 1.3.11, dK/Q(ζ) = ±p for some a ∈ Z, where a ≤ ϕ(p )p . It turns out that it’s easier to work with 1 − ζ in this example. In general this creates no obstacles, since dK/Q(1 − ζ) = dK/Q(ζ). In our case, we observe that Y 2 Y 2 dK/Q(1 − ζ) = (1 − σi(ζ) − (1 − σj(ζ))) = (σi(ζ) − σj(ζ)) = dK/Q(ζ). 1≤i a Thus dK/Q(1 − ζ) = ±p . To proceed, we need the following generalization of Lemma 0.1.1. 9 1.3 The Discriminant 1 Algebraic Number Fields Y Lemma 1.3.12. (1 − ζk) = p. p-k 1≤k≤pr Proof. Consider pr x − 1 r−1 r−1 r−1 f(x) = = 1 + xp + x2p + ... + x(p−1)p . xpr−1 − 1 Plugging in x = 1 gives the result. Observe that for any two k1, k2 ∈ N not divisible by p, 1 − ζk1 ∈ Z[ζ]. 1 − ζk2 k Then by symmetry, 1−ζ 1 is a unit in [ζ] for all such k , k . We will now show O = [1−ζ]. 1−ζk2 Z 1 2 K Z 2 n−1 Consider the basis {1, 1 − ζ, (1 − ζ) ,..., (1 − ζ) } for K/Q. If x ∈ OK , we can write x in the following manner by Lemma 1.3.6: n−1 X bi x = (1 − ζ)i for b ∈ , pa i Z i=0 a bi using the fact that dK/Q(1 − ζ) = ±p . If pa ∈ Z for each i, then we’re done. If not, multiply c bi c 1 c by some p so that all pa p ∈ p Z but not all of them lie in Z. Note that p x ∈ OK , so we may replace x with pcx and write n−1 X bi x = (1 − ζ)i, b ∈ . p i Z i=0 Suppose x 6∈ Z[1 − ζ]. Subtracting off the terms where p | bi if necessary, we may assume bi = 0 whenever p | bi. Let j be the smallest index with p - bj. Then n−1 X bi x = (1 − ζ)j, p b . p - j i=j p n The element (1−ζ)j+1 lies in Z[1 − ζ] since j + 1 ≤ n and (1 − ζ) | p by Lemma 1.3.12. p Therefore we may multiply the expression for x by (1−ζ)j+1 to obtain b x = j + (terms in [1 − ζ]). 1 − ζ Z n n bj bj bj bj Note that N 1−ζ = N(1−ζ) = p is not divisible by p, Thus N 1−ζ 6∈ Z but this con- bj tradicts the fact that 1−ζ ∈ OK . Hence x ∈ Z[1 − ζ] which finally proves the claim that OK = Z[1 − ζ] = Z[ζ]. 10 1.4 Factorization of Ideals 1 Algebraic Number Fields The following theorem allows us to generalize Example 1.3.10 to all Q(ζ) where ζ is a primitive nth root of unity. Theorem 1.3.13. Let A be an integrally closed integral domain with field of fractions K and suppose L/K and M/K are finite separable extensions with ω1, . . . , ωn an integral basis for L with respect to A and α1, . . . , αm an integral basis for M with respect to A. Further suppose dL/K (ω1, . . . , ωn) and dM/K (α1, . . . , αm) are relatively prime in A. Then {ωiαj} is an integral basis for the compositum LM over A and m n dLM/K (ωiαj) = dL/K (ωi) dM/K (αj) . Corollary 1.3.14. If ζm is an mth root of unity then OQ(ζm) = Z[ζm]. Q a1 ar Proof. Factor m = p ··· p . Then (ζm) = (ζ a1 ) ··· (ζ ar ). Moreover, for distinct 1 r Q Q p1 Q pr primes p 6= q, dQ(ζp)/Q and dQ(ζq)/Q are relatively prime. Therefore by Theorem 1.3.13, the ring of integers of Q(ζm) is O (ζ ) = [ζ a1 , . . . , ζ ar ] = [ζm]. Q m Z p1 pr Z 1.4 Factorization of Ideals Let K be a number field. We have seen that unique factorization may fail in OK , as we recall in the example below. √ √ Example 1.4.1. The quadratic field K = Q( −5) has ring of integers OK = Z[ −5]. In this ring, 6 has two different factorizations: √ √ 6 = 2 · 3 = (1 + −5)(1 − −5). √ Therefore unique factorization fails√ in Z[ −5]. To√ see that these are the only two factor- izations of 6, observe that N(1 +√ −5) = N(1 − −5) = 6, but there are no solutions in integers to the equation N(a + b −5) = a2 + 5b2 = 2, 3. It is our goal in this section to in some fashion repair the failure of unique factorization in OK , and in an arbitrary Dedekind domain A (to be defined below). Then we will further study the problem of determining all factorizations of an element in an integral extension. For the unique factorization√ problem, it would be nice (even ‘ideal’) if there were some objects p1, p2, p3, p4 ∈ Z[ −5] such that √ 2 = p1p2 1 + −5 = p1p3 √ 3 = p3p4 1 − −5 = p2p4. In fact, the exact objects we are looking for are prime ideals in OK . In order to describe a unique factorization into prime ideals, recall that for ideals I,J ⊂ A, their ideal product is ( n ) X IJ = xiyi : xi ∈ I, yi ∈ J . i=1 11 1.4 Factorization of Ideals 1 Algebraic Number Fields Definition. An integral domain A is a Dedekind domain if (1) A is integrally closed. (2) A is Noetherian. (3) All nonzero prime ideals of A are maximal. The main theorem we will prove is: Theorem 1.4.2. If A is a Dedekind domain, then every nonzero ideal I ⊂ A has a factor- Qn ai ization I = i=1 pi for distinct prime ideals pi ⊂ A which are unique up to ordering. Theorem 1.4.3. For every number field K, OK is a Dedekind domain. Proof. (1) OK is integrally closed since by definition it is the integral closure of Z in K. (2) We have seen (Prop. 1.3.7) that OK lies inside a finitely generated Z-module. By commutative algebra, this is sufficient to conclude that OK is Noetherian. (3) The property of nonzero prime ideals being maximal is alternatively known as Krull dimension 1. It is known that finite integral extensions which are integrally closed preserve Krull dimension, e.g. by the going up theorem. Since Z is Dedekind, integer unique factorization can be captured by Theorem 1.4.2 by associating a prime√ integer p ∈ Z with the principal prime ideal it generates: (p) ⊂ Z. For example, in Z[ −5] we have √ √ √ √ (6) = (2, 1 + −5)(2, 1 − −5)(3, 1 + −5)(3, 1 − −5). Lemma 1.4.4. If A is Dedekind, every nonzero ideal I ⊂ A contains a finite product of prime ideals. Proof. Let M be the set of nonzero ideals of A not divisible by a finite product of primes. Since A is Noetherian, there exists a maximal element a ∈ M. Then a must not be prime, so there exist elements b1, b2 ∈ A r a such that b1b2 ∈ a. Consider the ideals a + (b1) and a + (b2). Since a is maximal in M, each of these contains a finite product of prime ideals. Then (a + (b1))(a + (b2)) ⊆ a contains a product of primes, a contradiction. Hence M is empty. The classic proof of unique factorization of integers relies on being able to cancel out primes (by dividing), so to mimic this in our proof of Theorem 1.4.2, we define an analogy of inverses for ideals. Definition. If J ⊂ A is an ideal, the fractional ideal generated by J is the A-module J −1 := {x ∈ K | xJ ⊆ A}. Lemma 1.4.5. For every ideal J ⊂ A, J −1 is an A-submodule of K. Notice that J −1 ⊇ A so for any proper ideal J ( A, J −1 is not an ideal of A. 12 1.4 Factorization of Ideals 1 Algebraic Number Fields Lemma 1.4.6. If p ⊂ A is a prime ideal, p−1 6= A. Proof. Let x ∈ p. Then By Lemma 1.4.4, (x) ⊇ p1 ··· pr for prime ideals pi ⊂ A. Assume r is minimal among such products of primes contained in (x). We claim that p = pi for some 1 ≤ i ≤ r. If not, there exists an ai ∈ pi r p for each i, by maximality of prime ideals. Then a1 ··· ar ∈ p1 ··· pr ⊆ (x) ⊆ p, a contradiction. Thus p = pi for some i. Assume p = p1. −1 Then by minimality of r, we know (x) ) p2 ··· pr. Let b ∈ p2 ··· pr r (x). Then x b 6∈ A, but x−1bp ⊆ A. So x−1b ∈ p−1 r A. Lemma 1.4.7. If a ⊂ A is an ideal and p ⊂ A is a prime ideal, then ap−1 ) a. Proof. Certainly ap−1 ⊇ a since p−1 ⊃ A. Suppose ap−1 = a and let x ∈ p−1. Then xa ⊆ a, so in particular, left multiplication by x is an element of the A-algebra EndA(a). Since A is Noetherian, EndA(a) is finitely generated. Clearly EndA(a) is also a faithful A-module, so by a well-known characterization of integrality (cf. Atiyah-Macdonald), x is integral over A. Then since A is integrally closed, x ∈ A. We have shown p−1 = A, but this contradicts Lemma 1.4.6. Therefore ap−1 ) a. Corollary 1.4.8. For any prime ideal p ⊂ A, pp−1 = A. Proof. By Lemma 1.4.7, we have p ( pp−1 ⊆ A, but primes are maximal in a Dedekind domain, so pp−1 = A. Corollary 1.4.9. For an ideal a ⊂ A and a prime ideal p ) a, ap−1 ( A. Proof. If ap−1 = A then p = app−1 = a, a contradiction. We are now prepared to prove the unique factorization theorem for nonzero ideals in a Dedekind domain. Proof. (of Theorem 1.4.2) Let M be the collection of nonzero, non-unital ideals in A that do not have a factorization into prime ideals. Since A is Noetherian, M has a maximal element a. As before, a cannot be prime so it is contained in a prime ideal p. By Lemma 1.4.7, ap−1 ) a so ap−1 6∈ M. On the other hand, by Corollary 1.4.9 we have ap−1 6= (1) so −1 −1 ap = p1 ··· pr. Multiplying by p gives us a = app = pp1 ··· pr which shows a has a prime factorization. Thus M must be empty. This proves the existence part of the theorem. For uniqueness, suppose a = p1 ··· pr = q1 ··· qs for prime ideals pi, qj ⊂ A. Then by the Q proof of Lemma 1.4.6, p1 ⊃ j qj implies p1 = qj for some 1 ≤ j ≤ s. Multiplying both −1 sides by p1 cancels out terms, yielding shorter prime factorizations of a which are equal by induction. The base case of this induction is easy: if a is prime then it only has the trivial factorization a = a. This finishes the proof of unique factorization of ideals in a Dedekind domain. Remark. For an ideal a ⊂ A and a prime ideal p ⊂ A, we will use the expressions p ⊇ a (‘p contains a’) and p | a (‘p divides a’) to mean the same thing: p appears in the prime factorization of a. If I,J ⊂ A are ideals, we will write (I,J) = 1 if I + J = (1), that is, if I and J are relatively prime in A. 13 1.4 Factorization of Ideals 1 Algebraic Number Fields Definition. A fractional ideal of A is any finitely generated A-submodule of K. Example 1.4.10. For any ideal J ⊂ A, J −1 is a fractional ideal. Proposition 1.4.11. The nonzero fractional ideals of A form a group under ideal multipli- cation, with identity (1). Q ai Proof. By Theorem 1.4.2, fractional ideals of the form pi , with ai ∈ Z and pi ⊂ A prime, form a group which is isomorphic to a direct sum of copies of Z. Let M be any fractional ideal. Then M is finitely generated, so there exists an element x ∈ K such that xM ⊂ A is an ideal. Since (x) and xM have prime factorizations, so does M = (x)−1xM. Hence all fractional ideals form a group under multiplication. Corollary 1.4.8 shows that (1) is the identity element in this group. Proposition 1.4.12. If I ⊂ A is an ideal then II−1 = (1). −1 −1 Proof. Suppose I = p1 ··· pr is the prime factorization of I. Then J = p1 ··· pr is a fractional ideal, and by Corollary 1.4.8, IJ = (1). It remains to show J = I−1. First, since IJ = (1) we have J ⊆ I−1. If x ∈ I−1 then xI ⊂ A so −1 −1 −1 −1 xIp1 ··· pr ⊆ p1 ··· pr = J. Thus Ax ⊆ J so x ∈ J. This proves J = I−1 as required. Corollary 1.4.13. If A is a Dedekind domain and a unique factorization domain, then A is also a PID. Definition. For a Dedekind domain A, let JA denote the group of fractional ideals of A. The ideal class group of A is defined as the quotient group CA = JA/PA where PA is the subgroup of JA consisting of the principal fractional ideals of A. Clearly |CA| = 1 if and only if A is a PID (and therefore a UFD), so the ideal class group is a direct measure of the failure of unique factorization in A. Moreover, the ideal class group corresponds to an exact sequence of groups × × 1 → A → K → JA → CA → 1. We will study this further when we characterize the unit group K× in Section 1.9. Lemma 1.4.14. Every class in CA can be represented by an ideal I ⊂ A. Example 1.4.15. The ring A = C[x, y]/(y2 − x3 − x) is a Dedekind domain. It turns out that the ideal class group CA has cardinality equal to |C|, so this example shows that ideal class groups can be particularly bad. In particular, unique factorization fails in A: x3 − x = y2 = x(x − 1)(x + 1). One of the most important results in algebraic number theory is the following theorem, which we will prove in Section 1.8. 14 1.5 Ramification 1 Algebraic Number Fields Theorem. For a number field K, the class group CK := COK is finite. √ √ √ Example 1.4.16. Let K = Q( −5) and√ recall that 6 = 2 · 3 = (1 + −5)(1 − −5). What do 2 and 3 split into as ideals in OK = Z[ −5]? It turns out that √ √ √ √ 2OK = (2, 1 + −5)(2, 1 − −5) and 3OK = (3, 1 + −5)(3, 1 − −5). The underlying principle√ governing this splitting behavior is the fact that the minimal poly- nomial x2 + 5 of −5 splits differently mod 2 and 3: x2 + 5 ≡ (x + 1)2 (mod 2) and x2 + 5 ≡ (x + 1)(x − 1) (mod 3). 1.5 Ramification In this section let L/K be a finite separable field extension, let OK be a Dedekind domain with field of fractions K and let OL be the integral closure of OK in L. Put n = [L : K]. Lemma 1.5.1. OL is a Dedekind domain. Proof. This is the same proof as for Theorem 1.4.3. Lemma 1.5.2. If p ⊂ OK is a prime ideal then pOL 6= OL. −1 Proof. Take x ∈ p r OK , which exists by Lemma 1.4.6. Then xp ⊆ OK so xpOL ⊆ OL. If pOL = OL then we have xpOL = xOL ( OL, a contradiction. Therefore pOL 6= OL. Now fix a nonzero prime ideal p ⊂ OK . By Theorem 1.4.2, p considered as an ideal of OL has a unique factorization e1 eg pOL = P1 ··· Pg where the Pi ⊂ OL are distinct primes and each ei > 1. Note that for each i, OL/Pi is a finite dimensional OK /p-vector space. (This follows from the fact that Pi ∩ OK = p.) We say the Pi are the primes of OL lying over p. By unique factorization, these are the only primes lying over p. Definition. For a prime Pi in the factorization of pOL, the index fi = [OL/Pi : OK /p] is called the inertial degree of Pi (over p) and the exponent ei is called the ramification index of Pi (over p). We say the prime p is totally split if ei = fi = 1 for all 1 ≤ i ≤ g; p is totally ramified if g = 1 and f1 = 1; and p is inert if g = 1 and e1 = 1. Definition. If any ei > 1 or (OL/Pi)/(OK /p) is inseparable, we say the prime p is ramified (in OL). Otherwise p is unramified. Qg ei Theorem 1.5.3. For any prime p ⊂ OK with prime factorization pOL = i=1 Pi , we have Pg i=1 eifi = n = [L : K]. 15 1.5 Ramification 1 Algebraic Number Fields Proof. By the Chinese remainder theorem, we can write g g Y ei ∼ M ei OL/pOL = OL/ Pi = OL/Pi . i=1 i=1 ei To prove the theorem, we show that [OL/pOL : OK /p] = n and [OL/Pi : OK /p] = eifi for each 1 ≤ i ≤ g. For the first equality, take {ω1,..., ωm} to be a basis for OL/pOL as an OK /p-vector space. Lift these elements to ω1, . . . , ωm ∈ OL. Suppose a1ω1 + ... + amωm = 0 for coefficients ai ∈ OK . −1 Let a = (a1, . . . , am) ⊂ OK and let x ∈ a r ap; such an element exists by Lemma 1.4.7. Then xai ∈ OK for all i, but xai 6∈ p for some i. Replacing ai with xai and reducing mod p gives us a linear dependence, contradicting the assumption that ω1,..., ωm are a basis of OL/pOL. Hence ω1, . . . , ωm must be linearly independent in OK . To show they span OK , let M = ω1OK + ... + ωmOK ⊆ OL. Since the ωi generated OL/pOL, we get M + pOL = OL. In other words, p(OL/M) = OL/M. By Nakayama’s Lemma, this means OL/M is killed by some x ∈ 1 + p. In particular, such an x is necessarily ωm ω1 ωm nonzero so xOL ⊆ M. Thus ω1xOK +...+ x OK ⊇ OL. This implies that x K+...+ x K = L so the ωi span OL as an OK -vector space. This of course is only possible if m = n, so we have the first equality. Now consider the sequence ei ei 2 ei ei−1 ei OL/Pi ⊇ Pi/Pi ⊇ Pi /Pi ⊇ · · · ⊇ Pi /Pi ⊇ 0. ν ν+1 Taking each quotient in the chain yields something of the form Pi /Pi , and by unique ν ν+1 factorization, each of these quotients is nontrivial. Thus we can choose x ∈ Pi r Pi . Consider the map ν ν+1 ϕ : OL −→ Pi /Pi α 7−→ xα. Certainly ker ϕ = Pi since Pi ⊆ ker ϕ and primes are maximal in OL. Also, ϕ is surjective ν+1 ν+1 ν ν+1 ν ∼ ν ν+1 since Pi ( (x)+Pi ⊆ Pi which implies (x)+Pi = Pi . Therefore OL/Pi = Pi /Pi as OK /p-vector spaces. Adding these up gives us ei ei X dimOK /p OL/Pi = dimOK /p OL/Pi = eifi. j=1 Pg This proves both claims, and this is of course enough to conclude that n = i=1 eifi. Let θ ∈ OL be a primitive element of L/K, that is, L = K(θ). It is not always guaranteed that OK [θ] = OL. However, we have a way of measuring how far off from the whole ring OL the submodule OK [θ] really is. Definition. The conductor of the extension L/K is the ideal f := {α ∈ OK | αOK ⊆ OK [θ]} ⊂ OK where L = K(θ). 16 1.5 Ramification 1 Algebraic Number Fields Example 1.5.4. If OK [θ] = OL, then f = (1). √ √ Example 1.5.5. For K = Q and L = Q( −3), the conductor is f = (2, 1 + −3). Note that f is always nonzero. Theorem 1.5.6. Let L/K be a finite separable extension with L = K(θ). Suppose p ⊂ OK is prime and pOL + f = (1), where f is the conductor of the extension L/K. Let ϕ(x) be the minimal polynomial of θ over K. If ϕ(x) factors completely in (OK /p)[x] as e1 eg ϕ(x) = ϕ1(x) ··· ϕg(x) mod p Qg ei with deg ϕi = fi, then the factorization of p in OL is pOL = i=1 Pi where for each i, Pi is a prime ideal with ramification index ei and inertia degree fi, given explicitly by Pi = ϕi(θ)OL + pOL for any lift ϕi(x) of ϕi(x) in OK [x]. 0 Proof. Set O = OK [θ]. We will prove the following isomorphisms: ∼ 0 0 ∼ OL/pOL = O /pO = (OK /p)[x]/ϕ(x) 0 where again ϕ(x) is the minimal polynomial of θ over K. Clearly O ⊆ OL so we have a map 0 0 0 O /pO → OL/pOL. By assumption, pOL + f = OL but f ⊆ OL so we have pOL + O = OL. 0 0 Hence the map is surjective. On the other hand, pO ⊆ pOL ∩ O and 0 0 0 0 pOL ∩ O = (pOL ∩ O )(pOL + f) ⊆ pO + fpOL ⊆ pO . This proves injectivity, so the first isomorphism is proven. The second isomorphism is im- mediate from the fact that 0 0 ∼ ∼ O /pO = OK [x]/(ϕ(x), p) = (OK /p)[x]/ϕ(x). Now by the Chinese remainder theorem, we may write g ∼ ∼ M ei OL/pOLL = (OK /p)[x] = (OK /p)[x]/ϕ(x) . i=1 The prime ideals on the right are just the ideals (ϕi(x)). Set R = (OK /p)[x]/ϕ(x) and Qg ei notice that [R/(ϕi(x)) : OK /p] = fi = deg ϕi and i=1 ϕi(x) = 0. The primes in OL/pOL corresponding under the above isomorphism to the ϕi(θ) are Pi := ϕi(θ)OL + pOL. Notice Qg ei that i=1 Pi ⊆ pOL, but since g g X Y ei dimOK /p OL/pOL = eifi = dimOK /p OL/ Pi , i=1 i=1 Qg ei we have i=1 Pi = pOL. This proves the theorem. 17 1.5 Ramification 1 Algebraic Number Fields Example 1.5.7. Let OK = Z[i] be the Gaussian integers. Here the conductor is f = (1). Consider how x2 + 1 splits mod 13: x2 + 1 ≡ (x − 5)(x + 5) (mod 13). Then by Theorem 1.5.6, the ideal (13) splits in Z[i] in the following way: 13Z[i] = (13, 5 + i)(13, −5 + i) = (3 + 2i)(3 − 2i). For a prime p ⊂ OK and a prime P ⊂ OL lying over p, write kp = OK /p and `P = OL/P. Proposition 1.5.8. If p ⊂ OK is a nonzero prime such that p + f = (1) in OL, then p is unramified if and only if p does not divide the principal ideal (dL/K (θ)) generated by the discriminant of L/K. Q 2 Proof. We know dL/K (θ) = i ⇐⇒ p is relatively prime to each θi − θj in OM for some normal closure M of L/K ⇐⇒ p - (dL/K (θ)) in OK . Hence p is unramified precisely when p - (dL/K (θ)). We now discuss Hilbert’s program for ramification theory. Assume that L/K is Galois and let G = Gal(L/K). Note that σ(OL) = OL for all σ ∈ G. If p ⊂ OK is a prime and Qg ei pOL = i=1 Pi , then each σ ∈ G acts on the primes lying over p: σ(Pi) = Pj for some 1 ≤ j ≤ g. The key observation is that this action is transitive. Proposition 1.5.9. For any prime p ⊂ OK , G = Gal(L/K) acts transitively on the set of primes of OL lying over p. Proof. Suppose not. Then there is some pair of primes Pi, Pj lying over p such that σPj 6= Pi for all σ ∈ G. By the Chinese remainder theorem, pick x ∈ Pj such that x ≡ 1 (mod σPi) Q for all σ ∈ G. Then NL/K (x) ∈ Pj ∩ OK = p. On the other hand, NL/K (x) = σ∈G σ(x) but σ(x) 6∈ Pi for any σ, so NL/K (x) 6∈ p. This is impossible, so there is some σ ∈ G such that σPj = Pi. Corollary 1.5.10. When L/K is Galois, for any prime p ⊂ OK , all ramification indices ei and all inertia degrees fi for primes over p are equal, and therefore [L : K] = efg, where e = ei and f = fi for any prime Pi | p. 18 1.5 Ramification 1 Algebraic Number Fields ν ν Proof. An ideal Pi divides pOL if and only if σPi divides pOL for all σ ∈ G, which by ν Proposition 1.5.9 is equivalent to Pj dividing pOL for all 1 ≤ j ≤ g. Therefore the ramifi- cation indices are all equal; let e denote any one of them. Now given 1 ≤ i, j ≤ g, suppose σ ∈ G is a permutation taking Pj to Pi, that is, Pi = σPj. Then σ determines an isomor- phism OL/Pj → OL/Pi. Therefore fi = fj. Let f denote any of the inertial degrees. Then Pg finally, by Theorem 1.5.3 we have [L : K] = i=1 ef = efg. Fix a prime P ⊂ OL lying over p. Definition. The subgroup DP = {σ ∈ G | σ(P) = P} of G is called the decomposition group of P. Clearly by the orbit-stabilizer theorem, |DP| = ef where e and f are the ramification index and inertia degree of p, respectively. By Galois theory, there is a field extension ZP/K DP corresponding to the subgroup DP ≤ G, which is explicitly the fixed field ZP = L . Definition. For a prime P | p, the field ZP is called the decomposition field of P. L DP ZP G K 0 0 −1 Lemma 1.5.11. If σP = P for two primes P, P lying over p, then DP0 = σDPσ for some σ ∈ G. Proof. This is a more general fact about the stabilizers of a transitive group action. Note that for σ, τ ∈ Gal(L/K), −1 −1 τ στ ∈ DP ⇐⇒ τ στP = P ⇐⇒ στP = τP ⇐⇒ σ ∈ DτP −1 −1 which implies that σ ∈ DP ⇐⇒ τστ ∈ DP. Hence τDPτ = DτP. The ramification index and inertia degree are transitive in any tower of Galois field extensions: Lemma 1.5.12. For a Galois tower of number fields M ⊃ L ⊃ K and a prime p ⊂ OK , let Q ⊂ OM be a prime lying over p and set P = Q ∩ L. Then e(Q | p) = e(Q | P)e(P | p) and f(Q | p) = f(Q | P)f(P | p). Proof. Clearly P is a prime lying over p in OL, so e(P | p) and f(P | p) are defined. Then e(Q | p) = e(Q | P)e(P | p) is immediate by unique factorization in Dedekind domains, and f(Q | p) = f(Q | P)f(P | p) follows from Corollary 1.5.10 and the fact that [M : K] = [M : L][L : K]. The decomposition field is characterized by the following proposition. 19 1.5 Ramification 1 Algebraic Number Fields Proposition 1.5.13. Let a = P ∩ ZP be a prime below P in ZP. Then (1) P is the only prime in OL lying over a. (2) If e = e(P | p) and f = f(P | p) then e(P | a) = e, f(P | a) = f and e(a | p) = 1 = f(a | p). L P e f ZP a 1 1 K p Proof. (1) For all σ ∈ DP = Gal(L/ZP), σP = P. By Proposition 1.5.9, DP acts transitively on the primes over a, so P must be the unique one. (2) Since |DP| = ef, e(P | a)f(P | a) = ef but by Lemma 1.5.12, e(P | a) divides e and f(P | a) divides f. Therefore e(P | a) = e and f(P | a) = f, and the others are 1 by Lemma 1.5.12. Remark. Every σ ∈ DP induces an automorphism ϕσ : OL/P → OL/P which fixes kp = OK /p ⊆ `P = OL/P. Thus we get a map ϕ : DP −→ Aut(`P/kp) σ 7−→ ϕσ. Proposition 1.5.14. ϕ : DP → Aut(`P/kp) is surjective and `P/kp is a normal extension. Proof. By Proposition 1.5.13, ka = kp for any prime ideal a in the ring of integers of the decomposition field, so we can replace K with Z = ZP and G with DP. Thus P is the ¯ only prime lying over p. Take θ ∈ `P and let θ ∈ OL be any lift, with minimal polynomials ¯ ¯ ¯ g¯(x) ∈ kp[x] and f(x) ∈ K[x], respectively. Certainly f(θ) = 0 mod p sog ¯ | f in kp[x]. ¯ Since L/K is normal (it is a Galois extension), f splits over L. This implies f splits over `P, sog ¯ splits as well. This proves `P/kp is a normal extension. ¯ Now choose θ generating the separable closure of kp in `P. Letσ ¯ ∈ Aut(`P/kp). Then σ¯θ¯ is a root ofg ¯ and thus of f¯. Since f splits in L, there exists a root α ∈ L of f such ¯ thatα ¯ =σ ¯θ in `P. Choose σ ∈ G = DP such that σθ = α, which is possible since L/K is ¯ normal. Then ϕ(σ) =σ ¯ because θ generates the separable closure of kp in `P. This proves ϕ is surjective. Definition. The kernel IP = ker ϕ ≤ DP is called the inertia group of P. Explicitly, IP = {σ ∈ G | σ(x) ≡ x mod P for all x ∈ OL}. IP Definition. The fixed field TP = L is called the inertia field of P (over p). 20 1.6 Cyclotomic Fields and Quadratic Reciprocity 1 Algebraic Number Fields By Proposition 1.5.14, we have an exact sequence 1 → IP → DP → Gal(`P/kp) → 1. Proposition 1.5.15. Let b = P ∩ TP and a = P ∩ ZP = b ∩ ZP be prime ideals in the inertia and decomposition fields, respectively. Set e = e(P | p) and f = f(P | p). Then e(P | b) = e, f(b | a) = f and e(b | a) = f(P | b) = 1. L P e 1 TP b 1 f ZP a 1 1 K p Proof. Let Z = ZP and T = TP. In light of Proposition 1.5.13, it’s enough to show `P = OT /b and |DP/IP| = f. By the exact sequence 1 → IP → DP → Gal(`P/kp) → 1, ∼ DP/IP = Gal(`P/kp) so if `P = OT /b, then |DP/IP| = | Gal((OT /b)/kp)| = | Gal((OT /b)/(OZ /a))| = f. Therefore it suffices to prove the former statement, that is, `P = OT /b. The decomposi- tion/inertia group exact sequence for the extension L/T is 1 → IP → IP → Gal(`P/(OT /b)) → 1 which implies `P = OT /b as claimed. 1.6 Cyclotomic Fields and Quadratic Reciprocity Recall that when ζm is a primitive mth root of unity and K = Q(ζm), the ring of integers of this cyclotomic number field is OK = Z[ζm]. This was proven in Corollary 1.3.14. In this section, we further elaborate on the properties of Q(ζm) and Z[ζm] and use algebraic number theory to prove Gauss’s celebrated quadratic reciprocity law. Recall the following definition from elementary number theory. Definition. If p is an odd prime and a ∈ Z, the Legendre symbol of a mod p is 1 p - a and a is a square (mod p) a = −1 a is not a square (mod p) p 0 p | a. 21 1.6 Cyclotomic Fields and Quadratic Reciprocity 1 Algebraic Number Fields a Proposition 1.6.1. Suppose a, p ∈ Z with p prime and (2a, p) = 1. Then p = 1 if and √ only if the prime ideal (p) splits completely in OQ( a). Proof. The conductor f√ divides 2, so (p) splitting in O √ is equivalent to (p) splitting in √ a Q( a) Z[ a]. This, in turn, is equivalent to x2 − a splitting mod p, by Theorem 1.5.6, i.e. a ≡ x2 √ (mod p) for some x ∈ Z. Hence (p) splits in OQ( a) if and only if a is a square mod p. × a (p−1)/2 Remark. Since Fp is cyclic, it’s easy to show that p ≡ a (mod p) for any a ∈ Z — this is called Euler’s criterion. In particular, for a = −1 we have ( −1 1, p ≡ 1 (mod 4) = p −1, p ≡ −1 (mod 4). Most elementary proofs of quadratic reciprocity exploit this characterization of the Legendre symbol in some fashion. Here we prove the reciprocity law by considering the factorization of (p) in the ring Z[ζq]. Theorem 1.6.2 (Quadratic Reciprocity). Let p and q be distinct, odd primes. Then p q = (−1)(p−1)(q−1)/4. q p ∗ (q−1)/2 q∗ q (p−1)(q−1)/4 First, set q = (−1) q so that p = p (−1) by Euler’s criterion. The p q∗ statement we must then prove is that q = p . Example 1.6.3. The beauty of the quadratic reciprocity law is that it allows for fast com- putations of the Legendre symbol. For example, is 15 a square mod 37? Rather than trying 2 to compute all squares mod 37, or trying to factor x − 37 in F37, we can use reciprocity. Since 37 ≡ 1 (mod 4), we have: 15 5 3 37 37 2 1 = = = = (−1)(1) = −1. 37 37 37 5 3 5 3 So 15 is not a square mod 37. Lemma 1.6.4. Suppose n ≥ 2 is an integer with prime factorization n = Q pν(p), where the product is over all primes p and ν(p) ≥ 0 for all p. For each prime p, let fp be the ν(p) ϕ(pν(p)) multiplicative order of p mod n/p . Then in R = Z[ζn] we have pR = (p1 ··· pr) for distinct prime ideals p1,..., pr ⊂ R such that f(pi | p) = fp for each 1 ≤ i ≤ r. Proof. Fix a prime p and set m = n/pν(p) so that n = pν(p)m. Consider the number field K = Q(ζn). We know the conductor of ζn in OK is f = 1. Let γn be the nth cyclotomic ν(p) polynomial and let {αi} be the primitive p th roots of unity and {βj} be the primitive mth roots of unity. Then by the Chinese remainder theorem, × ∼ ν(p) × × (Z/nZ) = (Z/p Z) × (Z/mZ) 22 1.6 Cyclotomic Fields and Quadratic Reciprocity 1 Algebraic Number Fields so we can write Y γn(x) = (x − αiβj). i,j Note that all the αi are 1 in any field of characteristic p. Thus, modulo p, Y ϕ(pν(p)) ϕ(pν(p)) γn(x) ≡ (x − βj) = γm(x) . j ν(p) This allows us to reduce to the case when m = n, that is, the case when p = 1. Letγ ¯m(x) m denote the factorization of γm(x) mod p. Since x − 1 is separable over Fp (m is relatively m prime to p) andγ ¯m(x) | x − 1, we have thatγ ¯m(x) is also separable over Fp. The smallest extension of Fp containing a primitive mth root of unity (and thus all of them) is Fpfp . Thus γ¯m splits over Fpfp and each irreducible factor ofγ ¯m over Fp is the minimal polynomial of some primitive mth root of unity, each of which having degree fp. This impliesγ ¯m is a product of degree fp irreducible polynomials over Fp. By Theorem 1.5.6, we have ϕ(pν(p)) pR = (p1 ··· pr) . ϕ(n) Remark. In general, Theorem 1.5.10 implies that r = ν(p) . ϕ(p )fp Corollary 1.6.5. An odd prime integer p is ramified in Q(ζn)/Q if and only if p | n, and p = 2 is ramified if and only if 4 | n. p (q−1)/2 Lemma 1.6.6. If q is an odd prime integer, then (−1) q ∈ Q(ζq). Proof. Set X a τ = ζa. q q a∈(Z/qZ)× 2 (q−1)/2 Then τ ∈ Q(ζq) and τ = (−1) q. We are now able to prove quadratic reciprocity (Theorem 1.6.2). Proof. Let p and q be distinct odd primes and set q∗ = (−1)(q−1)/2q. Consider the tower of √ ∗ number fields Q(ζq) ⊃ Q( q ) ⊃ Q, with Galois groups as shown: Q(ζq) Z/((q − 1)/2)Z √ Z/(q − 1)Z Q( q∗) Z/2Z Q 23 1.7 Lattices 1 Algebraic Number Fields q∗ Then we determine the reciprocity law for p as follows: q∗ √ = 1 ⇐⇒ (p) splits in ( q∗) by Proposition 1.6.1 p Q √ ∗ ⇐⇒ Q( q ) ⊆ ZP, the decomposition field for any prime P over (p) ⇐⇒ there exist an even number of primes in Z[ζq] lying over (p) q − 1 ⇐⇒ is even, where fp is the multiplicative order of p mod q fp q − 1 ⇐⇒ f divides p 2 q−1 ⇐⇒ p 2 ≡ 1 (mod q) p ⇐⇒ = 1. q Thus quadratic reciprocity is proven. Corollary 1.6.7. If q is an odd prime, then ( 2 1, q ≡ 1, 7 (mod 8) = q −1, q ≡ 3, 5 (mod 8). ∗ q−1 ∗ Proof. Set q = (−1) 2 q, so that q ≡ 1 (mod 4). Then √ ∗ 2 √ 1 + q = 1 ⇐⇒ (2) splits in O ∗ = q Q( q ) Z 2 1 − q∗ ⇐⇒ f(x) = x2 − x + splits mod 2 4 ⇐⇒ q∗ ≡ 1 (mod 8) ⇐⇒ q = 1, 7 (mod 8). 1.7 Lattices One perspective on rings of algebraic integers is to view them as lattices. For example, Z[i] is very clearly a lattice in C spanned by the vectors 1 and i. We will show that any ring n of integers OK in a number field K/Q is a lattice in some R . This is the beginning of Minkowski’s so-called theory of geometry of numbers. n Definition. A Z-module Γ ⊆ R is a lattice of rank m if Γ = Zv1 + ldots + Zvm for R-linearly independent vectors v1, . . . , vm. If m = n then we say Γ is a complete lattice, or has full rank in Rn. 24 1.7 Lattices 1 Algebraic Number Fields Definition. For a lattice Γ ⊆ Rn, the set Φ = {x1v1 + ... + xmvm | 0 ≤ xi < 1} is called the fundamental domain of Γ, also sometimes called the fundamental paral- lelopiped. Observe that Γ is a complete lattice in Rn if and only if Γ + Φ = Rn. Definition. A subgroup W ⊆ Rn is said to be discrete if every point in W is open in the topology on Rn, that is, if every point x ∈ W has a neighborhood U in Rn such that U ∩ W = {x}. Proposition 1.7.1. If Γ ⊆ Rn is a subgroup, then Γ is discrete if and only if Γ is a lattice. Proposition 1.7.2. If Γ ⊆ Rn is a lattice, then Γ is complete if and only if there exists a bounded set M such that Γ + M = Rn. Proof. ( =⇒ ) When Γ is complete, M = Φ works. ( ⇒ = ) If Γ is not complete, let V (Rn be the R-span of Γ. Then V lies in some hyperplane H in Rn. Choose d > 0. Then for any bounded set of diameter diam(M) < d, all points further than d from H do not lie in Γ + M ⊆ H + M. Hence Γ + M 6= Rn. n Definition. If Γ = Zv1 + ... Zvn is a complete lattice in R , we define the volume of Γ to be the volume of the parallelopiped spanned by v1, . . . , vn: | | vol(Γ) := vol(Φ) = | det A| where A = v1 ··· vn . | | Note that since det(AT A) = (det A)2, we can write the volume formula as q vol(Γ) = det(vivj). Definition. A region Ω ⊆ Rn is centrally-symmetric if x ∈ Ω implies −x ∈ Ω. Minkowski’s theorem is the key result in the geometry of numbers which allows us to describe lattices like OK and UK , the ring of integers and unit group, respectively, in a number field K/Q. Theorem 1.7.3 (Minkowski). If Γ is a complete lattice in Rn and X is a centrally-symmetric, convex region of Rn such that vol(X) > 2n vol(Γ), then X contains a nonzero point of Γ. Proof. By a linear change of variables, we may assume Γ = Zn. Then vol(Γ) = det(I) = 1. n 1 Suppose X is as described, with vol(X) > 2 . Then vol( 2 > 1. We claim that there exist 25 1.8 The Class Group 1 Algebraic Number Fields