Romyar SharifiLECTURE NOTES
READING EDITIONPDF

LECTURE NOTES / Chapter 1

Algebraic Number Theory

Romyar Sharifi

Chapter 1 Abstract algebra

Book contents

Part 1 Algebraic number theory

Chapter 1
Abstract algebra

In this chapter, we introduce the many of the purely algebraic results that play a major role in algebraic number theory, pausing only briefly to dwell on number-theoretic examples. When we do pause, we will need the definition of the objects of primary interest in these notes, so we make this definition here at the start.

Definition 1.0.1.

A number field (or algebraic number field) is a finite field extension of .

We have the following names for extensions of of various degrees.

Definition 1.0.2.

A quadratic (resp., cubic, quartic, quintic, ...) field is a degree 2 (resp., 3, 4, 5, ...) extension of .

1.1. Tensor products of fields

Proposition 1.1.1.

Let K be a field, and let f K[x] be monic and irreducible. Let M be a field extension of K, and suppose that f factors as i=1mfiei in M[x], where the fi are irreducible and distinct and each ei is positive. Then we have an isomorphism

κ : K[x](f)KM i=1mM[x](f iei)

of M-algebras such that if g K[x], then κ((g+(f))1) = (g+(fiei))i.

Proof.

Note that we have a canonical isomorphism K[x]KM M[x] that gives rise to the first map in the composition

K[x](f)KM M[x](f) i=1mM[x](f iei),

the second isomorphism being the Chinese remainder theorem. The composition is κ.

We have the following consequence.

Lemma 1.1.2.

Let LK be a finite separable extension of fields, and let M be an algebraically closed field containing K. Then we have an isomorphism of M-algebras

κ : LKM σ : LMM,

where the product is taken over field embeddings of L in M fixing K, such that

κ(β 1) = (𝜎𝛽)σ

for all β L.

Proof.

Write L = K(𝜃), and let f K[x] be the minimal polynomial of 𝜃. Then we define κ as the composition

LKM σ : LM M[x] (xσ(𝜃)) σ : LMM,

where the first isomorphism is that of Proposition 1.1.1 and the second takes x to σ(𝜃) in the coordinate corresponding to σ. Any β L has the form g(𝜃) for some g K[x], and since any σ : LM fixing K fixes the coefficients of g, we have κ(β 1) is as stated.

Remark 1.1.3.

If we compose κ of Lemma 1.1.2 with the natural embedding LLKM that takes α L to α 1, then the composition

ιM: L σ : LMM

is the product of the field embeddings σ of L in M fixing K.

Definition 1.1.4.

Let K be a field and L and M be extensions of K both contained in some field Ω. We say that L and M are linearly disjoint over K if every K-linearly independent subset of L is M-linearly independent.

Lemma 1.1.5.

Let K be a field and L and M be extensions of K both contained in some field Ω. If L and M are linearly disjoint over K, then LM = K.

Proof.

If x LM with xK, then x and 1 are elements of L that are K-linearly independent but not M-linearly independent, so L and M are not linearly disjoint over K.

From the definition, it may not be clear that the notion of linear disjointness is a symmetric one. However, this follows from the following.

Proposition 1.1.6.

Let K be a field and L and M be extensions of K both contained in some field Ω. Then L and M are linearly disjoint over K if and only if the map φ : LKM 𝐿𝑀 induced by multiplication is an injection.

Proof.

Suppose that γ1,,γs M are L-linearly dependent, and write i=1sβiγi = 0 for some βi L. If φ is injective, then we must have i=1sβiγi = 0, which means that the γi are K-linearly dependent.

Conversely, let L and M be linearly disjoint over K. Suppose that we have a nonzero

x =i=1sβ iγi kerφ

for some βi L and γi M, with s taken to be minimal. If x0, then the γi are L-linearly dependent, so they are K-linearly dependent. In this case, without loss of generality, we may suppose that

γs+i=1s1α iγi = 0

for some αi in K. Then

x =i=0s1(β iαiβs)γi,

contradicting minimality. Thus kerφ = 0.

Corollary 1.1.7.

Let K be a field and L and M be extensions of K both contained in a given algebraic closure of K. Then L and M are linearly disjoint over K if and only if LKM is a field.

Proof.

Note that 𝐿𝑀 is a union of subfields of the form K(α,β) with α L and β M. Since α and β are algebraic over K, we have K(α,β) = K[α,β], and every element of the latter ring is a K-linear combination of monomials in α and β. Thus φ of Proposition 1.1.6 is surjective, and the result follows from the latter proposition.

Corollary 1.1.8.

Let K be a field and L and M be finite extensions of K inside a given algebraic closure of K. Then [𝐿𝑀 : K] = [L : K][M : K] if and only if L and M are linearly disjoint over K.

Proof.

Again, we have the surjection φ : LKM 𝐿𝑀 given by multiplication which is an injection if and only if L and M are linearly disjoint by Proposition 1.1.6. As LKM has dimension [L : K][M : K] over K, the result follows.

Remark 1.1.9.

Suppose that L = K(𝜃) is a finite extension of K. To say that L is linearly disjoint from a field extension M of K is by Propostion 1.1.1 exactly to say that the minimal polynomial of 𝜃 in K[x] remains irreducible in M[x].

We prove the following in somewhat less generality than possible.

Lemma 1.1.10.

Let L be a finite Galois extension of a field K inside an algebraic closure Ω of K, and let M be an extension of K in Ω. Then L and M are linearly disjoint if and only if LM = K.

Proof.

We write L = K(𝜃) for some 𝜃 L, and let f K[x] be the minimal polynomial of 𝜃. As Gal(𝐿𝑀M)Gal(L(LM)) by restriction, we have LM = K if and only if [𝐿𝑀 : M] = [L : K]. Since 𝐿𝑀 = M(𝜃), this occurs if and only if f is irreducible in M[x]. The result then follows from Remark 1.1.9.

1.2. Integral extensions

Definition 1.2.1.

We say that BA is an extension of commutative rings if A and B are commutative rings such that A is a subring of B.

Definition 1.2.2.

Let BA be an extension of commutative rings. We say that β B is integral over A if β is the root of a monic polynomial in A[x].

Examples 1.2.3.

a.

Every element a A is integral over A, in that a is a root of xa.

b.

If LK is a field extension and α L is algebraic over K, then α is integral over K, being a root of its minimal polynomial, which is monic.

c.

If LK is a field extension and α L is transcendental over K, then α is not integral over K.

d.

The element 2 of (2) is integral over , as it is a root of x2 2.

e.

The element α = 15 2 of (5) is integral over , as it is a root of x2 x1.

Proposition 1.2.4.

Let BA be an extension of commutative rings. For β B, the following conditions are equivalent:

i.

the element β is integral over A,

ii.

there exists n 0 such that {1,β,,βn} generates A[β] as an A-module,

iii.

the ring A[β] is a finitely generated A-module, and

iv.

there exists a finitely generated A-submodule M of B that such that 𝛽𝑀 M and which is faithful over A[β].

Proof.

Suppose that (i) holds. Then β is a root of a monic polynomial g A[x]. Given any f A[x], the division algorithm tells us that f = 𝑞𝑔+r with q,r A[x] and either r = 0 or degr < degg. It follows that f(β) = r(β), and therefore that f(β) is in the A-submodule generated by {1,β,,βdegg1}, so (ii) holds. Since this set is independent of f, it generates A[β] as an A-module, so (iii) holds. Suppose that (iii) holds. Then we may take M = A[β], which being free over itself has trivial annihilator.

Finally, suppose that (iv) holds. Let

M =i=1nAγ i B

be such that 𝛽𝑀 M, and suppose without loss of generality that β0. We have

βγi =j=1na 𝑖𝑗γj

for some a𝑖𝑗 A with 1 i n and 1 j n. Consider A-module homomorphism T : Bn Bn represented by (a𝑖𝑗). The characteristic polynomial f(x) A[x] of T is monic, and f(β) acts as zero on M. Since M is a faithful A[β]-module, we must have f(β) = 0. Thus, β is integral.

Example 1.2.5.

The element 12 is not integral over , as [1,21,,2n] for n 0 is equal to [2n], which does not contain 2(n+1).

Definition 1.2.6.

Let BA be an extension of commutative rings. We say that B is an integral extension of A if every element of B is integral over A.

Example 1.2.7.

The ring [2] is an integral extension of . Given α = a+b2 with a,b , note that α is a root of x2 2𝑎𝑥+a2 2b2.

Lemma 1.2.8.

Suppose that BA is an extension of commutative rings such that B is finitely generated as an A-module, and let M be a finitely generated B-module. Then M is a finitely generated A-module.

Proof.

Let {m1,,mn} be a set of generators of M as a B-module, and let {β1,,βk} be a set of generators of B as an A-module. We claim that {βimj1 i k,1 j n} is a set of generators of M as an A-module. To see this, let m M and write

m =j=1nb jmj

with bj B for 1 j n. For 1 j n, we then write

bj =i=1ka 𝑖𝑗βi

with a𝑖𝑗 A for 1 i k. We then have

m =i=1k j=1na 𝑖𝑗βimj,

as desired.

We now give a criterion for a finitely generated algebra over a ring to be finitely generated as a module.

Proposition 1.2.9.

Let BA be an extension of commutative rings and suppose that

B = A[β1,β2,,βk]

for some k 0 and βi B with 1 i k. Then the following are equivalent.

i.

the ring B is integral over A,

ii.

each βi with 1 i k is integral over A, and

iii.

the ring B is finitely generated as an A-module.

Proof.

Clearly, (i) implies (ii), so suppose that (ii) holds. By definition, each βi is then integral over any commutative ring containing A. By Proposition 1.2.4, each A[β1,,βj] with 1 j k is a finitely generated A[β1,,βj1]-module, generated by {1,βj,,βjnj} for some nj 0. Assuming recursively that A[β1,,βj1] is finitely generated as an A-module, Lemma 1.2.8 implies that A[β1,,βj] = A[β1,,βj1][βj] is finitely generated as an A-module as well. Therefore, (iii) holds. Finally, if (iii) holds and β B, then since 𝛽𝐵 B, the element β is integral over a by Proposition 1.2.4. Thus (i) holds.

We derive the following important consequence.

Proposition 1.2.10.

Suppose that CB and BA are integral extensions of commutative rings. Then CA is an integral extension as well.

Proof.

Let γ C, and let f B[x] be a monic polynomial which has γ as a root. Let B be the subring of B generated over A by the coefficients of f, which is integral over A as B is. By Proposition 1.2.9, the ring B is then finitely generated over A. As B[γ] is finitely generated over B as well, we have B[γ] is finitely generated over A. Hence, B[γ] is itself an integral extension of A. By definition of an integral extension, the element γ is integral over A. Since γ C was arbitrary, we conclude that C is integral over A.

Definition 1.2.11.

Let BA be an extension of commutative rings. The integral closure of A in B is the set of elements of B that are integral over A.

Proposition 1.2.12.

Let BA be an extension of commutative rings. Then the integral closure of A in B is a subring of B.

Proof.

If α and β are elements of B that are integral over A, then A[α,β] is integral over A by Proposition 1.2.9. Therefore, every element of A[α,β], including α +β and α β, is integral over A as well. That is, the integral closure of A in B is closed under addition, additive inverses, and multiplication, and it contains 1, so it is a ring.

Example 1.2.13.

The integral closure of in [x] is , since if f [x] is of degree at least 1 and g [x] is nonconstant, then g(f(x)) has degree deggdegf in x, hence cannot be 0.

Definition 1.2.14.

a.

The ring of algebraic integers is the integral closure ¯ of inside .

b.

An algebraic integer is an element of ¯.

Definition 1.2.15.

Let BA be an extension of commutative rings. We say that A is integrally closed in B if A is its own integral closure in B.

Definition 1.2.16.

We say that an integral domain A is integrally closed if it is integrally closed in its quotient field.

Example 1.2.17.

Every field is integrally closed.

Proposition 1.2.18.

Let A be an integrally closed domain, let K be the quotient field, and let L be a field extension of K. If β L is integral over A with minimal polynomial f K[x], then f A[x].

Proof.

Since β L is integral, it is the root of some monic polynomial g A[x] such that f divides g in K[x]. As g is monic, every root of g in an algebraic closure K¯ containing K is integral over K. As every root of f is a root of g, the same is true of the roots of f. Write f = i=1n(xβi) for βi K¯ integral over A. As the integral closure of A in K¯ is a ring, it follows that every coefficient of f is integral over A, being sums of products of the elements βi. Since f K[x] and A is integrally closed, we then have f A[x].

The following holds in the case of UFDs.

Proposition 1.2.19.

Let A be a UFD, let K be the quotient field of A, and let L be a field extension of K. Suppose that β L is algebraic over K with minimal polynomial f K[x]. If β is integral over A, then f A[x].

Proof.

Let β L be integral over A, let g A[x] be a monic polynomial of which it is a root, and let f K[x] be the minimal polynomial of β. Since f divides g in K[x] and A is a UFD with quotient field K, there exists d K such that 𝑑𝑓 A[x] and 𝑑𝑓 divides g in A[x]. Since f is monic, d must be an element of A (and in fact may be taken to be a least common denominator of the coefficients of f). The coefficient of the leading term of any multiple of 𝑑𝑓 will be divisible by d, so this forces d to be a unit, in which case f A[x].

Corollary 1.2.20.

Every unique factorization domain is integrally closed.

Proof.

The minimal polynomial of an element a of the quotient field K of a UFD A is xa. If aA, it follows from Proposition 1.2.19 that a is not integral over A.

Examples 1.2.21.

The ring is integrally closed.

Example 1.2.22.

The ring [17] is not integrally closed, since α = 1+17 2 is a root of the monic polynomial x2 x4. In particular, [17] is not a UFD.

Proposition 1.2.23.

Let BA be an extension of commutative rings, and suppose that B is an integrally closed domain. Then the integral closure of A in B is integrally closed.

Proof.

Let A¯ denote the integral closure of A in B, and let Q denote the quotient field of A¯. Let α Q, and suppose that α is integral over A¯. Then A¯[α] is integral over A¯, so A¯[α] is integral over A, and therefore α is integral over A. That is, α is an element of A¯, as desired.

Example 1.2.24.

The ring ¯ of algebraic integers is integrally closed.

Proposition 1.2.25.

Let A be an integral domain with quotient field K, and let L be an algebraic extension of K. Then the integral closure B of A in L has quotient field equal to L inside L. In fact, every element of L may be written as bd for some d A and b B.

Proof.

Any β L is the root of a monic polynomial f = i=0naixi K[x]. Let d A be such that 𝑑𝑓 A[x]. Then

dnf(d1x) = i=0na idnixi A[x]

is both monic and has 𝑑𝛽 as a root. In other words, 𝑑𝛽 is contained in B, as desired.

Example 1.2.26.

The quotient field of ¯ is ¯.

Definition 1.2.27.

The ring of integers (or integer ring) 𝒪K of a number field K is the integral closure of in K.

The prototypical examples of rings of integers arise in the setting of quadratic fields.

Theorem 1.2.28.

Let d1 be a square-free integer. The ring of integers of (d) is

𝒪(d) = { [1+d 2 ]if d 1mod4, [d] if d 2,3mod4.
Proof.

Suppose that α = a+bd is integral for a,b . If b = 0, then we must have a . If b0, then the minimal polynomial of α is f = x2 2𝑎𝑥+a2 b2d. Since α is integral, we must have f [x], so 2a . If a , then since a2 b2d and d is square-free, we have b as well. If a, then 2a = a and 2b = b for some odd a,b, and (a)2 (b)2dmod4. As (4)2 = {0,1}, this is impossible if d1mod4. If d 1mod4, then a+bd lies in the claimed ring, since it contains d, and clearly (1+d)2 is integral.

1.3. Norm and trace

Definition 1.3.1.

Let LK be a finite extension of fields. For α L, let mα: L L denote the linear transformation of K-vector spaces defined by left multiplication by α.

a.

The norm map NLK: L K is defined by NLK(α) = detmα for α L.

b.

The trace map TrLK: L K is defined by TrLK(α) = trmα for α L.

Remark 1.3.2.

For a finite field extension LK, the trace map TrLK is a homomorphism, and the norm map NLK is a homomorphism to K× upon restriction to L×.

Proposition 1.3.3.

Let LK be a finite extension of fields, and let α L. Let f K[x] be the minimal polynomial of α over K, let d = [K(α) : K], let s = [L : K(α)], and let K¯ be an algebraic closure of K. Suppose that f factors in K¯[x] as

f =i=1d(xα i)

for some α1,,αd K¯. Then the characteristic polynomial of mα is fs, and we have

NLK(α) =i=1dα is and Tr LK(α) = si=1dα i.
Proof.

We claim that the characteristic polynomial of the K-linear transformation mα is fs. First suppose that L = K(α). Note that {1,α,,αd1} forms a K-basis of K(α), and with respect to this basis, mα is given by the matrix

A = ( 0 a0 1 0 a1 1 0 a d2 1 ad1 ),

where ai K for 1 i d are such that

f = xd+ i=0d1a ixi.

Expanding the determinant of 𝑥𝐼 A using its first row, we see that

charmα = det(𝑥𝐼 A) = xdet(𝑥𝐼 A)+(1)d1a0det ( 1 x 1 x 1 ) = xdet(𝑥𝐼 A)+a0,

where Ais the (1,1)-minor of A. By induction on the dimension of A, we may assume that

det(𝑥𝐼 A) = xd1 + i=0d2a i+1xi,

so charmα = f. Since

f = xdtr(m α)xd1 ++(1)ddet(m α),

we have by expanding out the factorization of f in K¯[x] that NLKα and TrLKα are as stated in this case.

In general, if {β1,,βs} is a basis for LK(α), then {βiαj1 i s,0 j d1}is a basis for LK. The matrix of mα with respect to this basis (with the lexicographical ordering on the pairs (i,j)) is the block diagonal matrix consisting of s copies of A. In other words, charmα is the fs, from which the result now follows easily.

We can also express the norm as a power of a product of conjugates and the trace as a multiple of a sum of conjugates.

Proposition 1.3.4.

Let LK be a finite extension of fields, and let m = [L : K]i be its degree of inseparability. Let 𝔖 denote the set of embeddings of L fixing K in a given algebraic closure of K. Then, for α L, we have

NLK(α) =σ𝔖σαm and Tr LK(α) = mσ𝔖𝜎𝛼.
Proof.

The distinct conjugates of α in a fixed algebraic closure K¯ of K are exactly the 𝜏𝛼 for τ in the set 𝔗 of distinct embeddings of K(α) in K¯. These 𝜏𝛼 are the distinct roots of the minimal polynomial of α over K, each occuring with multiplicity the degree [K(α) : K]i of inseparability of K(α)K. Now, each of these embeddings extends to [L : K(α)]s distinct embeddings of L into K¯, and each extension σ 𝔖 of τ sends α to τ(α). By Proposition 1.3.3, we have

NLKα =τ𝔗(𝜏𝛼)[L:K(α)][K(α):K]i = σ𝔖σα[L:K]i,

and similarly for the trace.

We have the following immediate corollary.

Corollary 1.3.5.

Let LK be a finite separable extension of fields. Let 𝔖 denote the set of embeddings of L fixing K in a given algebraic closure of K. Then, for α L, we have

NLK(α) =σ𝔖𝜎𝛼 and TrLK(α) =σ𝔖𝜎𝛼.

We also have the following.

Proposition 1.3.6.

Let MK be a finite field extension and L be an intermediate field in the extension. Then we have

NMK = NLKNML and TrMK = TrLKTrML.
Proof.

We prove this for norm maps. Let 𝔖 denote the set of embeddings of L into K¯ that fix K, let 𝔗 denote the set of embeddings of M into K¯ that fix L, and let 𝔘 denote the set of embeddings of M into K¯ that fix K. Since [M : K]i = [M : L]i[L : K]i, it suffices by Proposition 1.3.4 to show that

δ𝔘𝛿𝛼 =σ𝔖σ (τ𝔗𝜏𝛼).

We extend each σ to an automorphism σ~ of K¯ fixing K. We then have

σ𝔖σ (τ𝔗𝜏𝛼) =σ𝔖τ𝔗(σ~τ)α.

For the trace map, we simply replace the products by sums.

We claim that the subset X = {σ~τσ 𝔖,τ 𝔗} of 𝔘 is exactly 𝔘, which will finish the proof. Let σ,σ𝔖 and τ,τ𝔗, and suppose that

σ~τ = σ~τ. (1.3.1)

Since σ~τ|L = σ|L, we have that σ = σ. Since σ~ is an automorphism, we then apply its inverse to (1.3.1) to obtain τ = τ. As there are then |𝔖||𝔗| = |𝔘| elements of X, we have the result.

Example 1.3.7.

The norm for the extension (d), where d is a square-free integer, is given by

N(d)(x+yd) = (x+yd)(xyd) = x2 dy2

for x,y .

Example 1.3.8.

For a,b,c , we have

N(23)(a+b23+c43) = (a+b23+c43)(a+𝑏𝜔23+cω243)(a+bω223+𝑐𝜔43) = a3 +2b3 +4c3 6𝑎𝑏𝑐,

for ω a primitive cube root of unity. The trace is simpler:

Tr(23)(a+b23+c43) = 3a.

Definition 1.3.9.

A K-valued linear character of a group G is a group homomorphism χ : G K×, where K is a field.

Definition 1.3.10.

We say that a set of K-valued linear characters X of a group G is K-linearly independent if it is linearly independent as a subset of the K-vector space of functions G K.

Theorem 1.3.11.

Any set of K-valued linear characters G K× of a group G is K-linearly independent.

Proof.

Let X be a set of linear characters G K×. Suppose by way of contradiction that m 2 is minimal such that there m distinct, linearly dependent elements of G. Choose ai K and χi X with 1 i m for which a10 and

i=1ma iχi = 0.

Also, let h G be such that χ1(h)χm(h). Set bi = ai(χi(h)χm(h)) for 1 i m1. For any g G, we then have

i=1m1b iχi(g) =i=1ma i(χi(h)χm(h))χi(g) =i=1ma iχi(h𝑔)χm(h)i=1ma iχi(g) = 0.

Since b10 and i=1m1biχi has only m1 terms, this contradicts the existence of m.

In the case of cyclic extensions, the kernel of the norm map bears a simple description.

Theorem 1.3.12 (Hilbert’s Theorem 90).

Let LK be a finite cyclic extension of fields, and let σ be a generator of its Galois group. Then

kerNLK = {σ(β) β β L×}.
Proof.

Set n = [L : K]. Let β L, and note that

NLK (σ(β) β ) =i=0n1σi+1(β) σi(β) = NLK(β) NLK(β) = 1.

Next, suppose that α kerNLK, and set

xγ = γ +𝛼𝜎(γ)+𝛼𝜎(α)σ2(γ)++𝛼𝜎(α)σn2(α)σn1(γ)

for γ L. The elements of Gal(LK), which is to say the powers of σ, are distinct L-valued characters on L×, and therefore they are L-linearly independent. Thus, there exists γ L× such that xγ0. We then note that

𝛼𝜎(xγ) = 𝛼𝜎(γ)+𝛼𝜎(α)σ2(γ)++𝛼𝜎(α)σn2(α)σn1(γ)+N LK(α)γ = xγ,

so α1 = σ(xγ)xγ1, finishing the proof.

There is also an additive form of Hilbert’s Theorem 90, which describes the kernel of the trace. We leave the proof to the reader.

Proposition 1.3.13 (Additive Hilbert’s Theorem 90).

Let LK be a finite cyclic extension of fields, and let σ be a generator of its Galois group. Then

kerTrLK = {σ(β)ββ L}.

Lemma 1.3.14.

Let BA be an integral extension of domains, and suppose that A is integrally closed in its quotient field K. Let L denote the quotient field of B, and suppose that LK is finite. Then NLK(β) and TrLK(β) are elements of A for every β B.

Proof.

Since β is integral over A, so are all of its conjugates in an algebraic closure L¯ of L, since they are also roots of the monic polynomial of which β is a root. It follows from Proposition 1.3.4 and the fact that the integral closure of A in L¯ is a ring that NLK(β) and TrLK(β) are elements of K integral over A, so A is integrally closed.

1.4. Discriminants

Definition 1.4.1.

Let K be a field and V a finite-dimensional K-vector space. A K-bilinear form (or simply, bilinear form) ψ : V ×V K on V is a function satisfying

ψ(v+v,w) = ψ(v,w)+ψ(v,w) and ψ(v,w+w) = ψ(v,w)+ψ(v,w)

and

ψ(𝑎𝑣,w) = 𝑎𝜓(v,w) = ψ(v,𝑎𝑤)

for all a K and v,v,w,w V.

Definition 1.4.2.

A K-bilinear form ψ on a K-vector space V is said to be symmetric if

ψ(v,w) = ψ(w,v)

for all v,w V.

Example 1.4.3.

Given a matrix Q Mn(K), we can define a bilinear form on Kn by

ψ(v,w) = vT 𝑄𝑤

for v,w Kn, where we use a superscript T to denote the transpose. It is symmetric if and only if Q is.

Example 1.4.4.

If LK is a finite extension of fields, then ψ : L×L K defined by

ψ(α,β) = TrLK(𝛼𝛽)

for α,β L is a symmetric K-bilinear form on L.

Definition 1.4.5.

The discriminant of a bilinear form ψ on a finite dimensional K-vector space V relative to an ordered basis (v1,,vn) of V is the determinant of the matrix (ψ(vi,vj))i,j.

Lemma 1.4.6.

Let ψ : V ×V K be a K-bilinear form on a finite-dimensional vector space V of dimension n 1. Let v1,,vn V, and let T : V V be a linear transformation. Then

det(ψ(T vi,T vj))i,j = (detT )2 det(ψ(v i,vj))i,j.
Proof.

Suppose first that the vi form a basis of V. Let A = (a𝑖𝑗) denote the matrix of T with respect to the ordered basis (v1,,vn). for each i. We may then write

ψ(T vi,T vj) =k=1na 𝑖𝑘l=1na 𝑗𝑙ψ(vk,vl).

As matrices, we then have

(ψ(T vi,T vj)) = A(ψ(vi,vj))AT ,

and the result follows as detT = detA = detAT .

If the vi do not form a basis of V, then there is an ordered basis (e1,,en) of V and a linear transformation U : V V of with U(ei) = vi for all i. As detU = 0, we have

det(ψ(vi,vj))i,j = (detU)2det(ψ(e i,ej))i,j = 0.

As the T v1,,T vn cannot be a basis, we have that both sides in the formula are zero.

Remarks 1.4.7.

Let ψ : V ×V K be a K-bilinear form on a finite-dimensional vector space V of dimension n 1. Then Lemma 1.4.6 implies the following.

a.

The discriminant of ψ to a basis is independent of its ordering, since a permutation matrix has determinant ±1.

b.

We have det(ψ(vi,vj))i,j = 0 if v1,,vn V are linearly dependent.

Definition 1.4.8.

Let LK be a finite extension of fields. The discriminant of LK relative to a basis of L as a K-vector space is the discriminant of the bilinear form

(α,β)TrLK(𝛼𝛽)

relative to the basis.

Notation 1.4.9.

If LK is a finite extension of fields and β1,,βn L are arbitrary, we set

D(β1,,βn) = det(TrLK(βiβj))i,j.

If (β1,,βn) is an ordered basis of LK, then D(β1,,βn) is its discriminant.

Proposition 1.4.10.

Let LK be a finite separable extension of fields. Then for any β1,,βn L, we have

D(β1,,βn) = (det(σiβj)i,j)2,

where {σ1,,σn} is the set of embeddings of L in an algebraic closure of K that fix K.

Proof.

Note that

TrLK(βiβj) =k=1nσ k(βi)σk(βj),

so the matrix (TrLK(βiβj)) equals QT Q, where Q Mn(L) satisfies Q𝑖𝑗 = σi(βj).

Definition 1.4.11.

Let K be a field, and let α1,,αn K. Then the matrix

Q(α1,,αn) = ( 1 α1 α1n1 1 α2 α2n1 1 αn αnn1 )

is called the Vandermonde matrix for α1,,αn.

Lemma 1.4.12.

Let K be a field, and let Q(α1,,αn) be the Vandermonde matrix for elements α1,,αn of K. Then

detQ(α1,,αn) =1i<jn(αjαi).
Proof.

We work by induction on n 1, the case n = 1 asserting the obvious fact that detQ(α) = 1 for any α K. To compute the determinant of Q = Q(α1,,αn), subtract α1 times its ith column from its (i+1)th column for each 1 i n1, which leaves the determinant unchanged. We then obtain

detQ = | 1 0 0 1 α2 α1 α2n2(α2 α1) 1 αnα1 αnn2(αnα1) | = | α2 α1 α2n2(α2 α1) αnα1 αnn2(αnα1) | =i=2n(α iα1)detQ(α2,,αn),

and the result now follows by induction.

Proposition 1.4.13.

Suppose that LK is a separable extension of degree n, and let α L be such that L = K(α). Then

D(1,α,,αn1) =1 i<jn(αjαi)20,

where α1,,αn are the conjugates of α in an algebraic closure of K.

Proof.

By Proposition 1.4.10, we have that D(1,α,,αn1) is the square of the determinant of the Vandermonde matrix Q(α1,,αn), and the result then follows from Lemma 1.4.12.

Example 1.4.14.

Let d be a square-free integer with d1. Consider the basis {1,d} of (d) as a -vector space. Since the distinct conjugates of d are ±d, we have D(1,d) = 4d.

The following is basically a rephrasing of Proposition 1.4.13.

Corollary 1.4.15.

Suppose that LK is a separable extension of degree n, and let α L be such that L = K(α), and let f K[x] be the minimal polynomial of α. Then

D(1,α,,αn1) = (1)n(n1) 2 NLK(f(α)),

where f K[x] is the derivative of f.

Proof.

Let α1,,αn be the conjugates of α in an algebraic closure K¯ of K. Then

f(x) = i=1n j=1 ji n(xα j),

so we have

f(α i) =j=1 ji n(α iαj)

for each i, and the conjugates of f(α) in K¯ are the f(αi). We then have

NLK(f(α)) = i=1n j=1 ji n(α iαj) = (1)n(n1)21 i<jn(αjαi)2.

Corollary 1.4.16.

Let LK be a finite separable extension of fields. Then the discriminant of LK relative to an ordered basis (β1,,βn) of L is nonzero.

Proof.

Since LK is separable, there exists α L such that L = K(α). Then (1,α,,αn1) is an ordered basis of LK, and there exists an invertible K-linear transformation T : L L with T (αi1) = βi for 1 i n. By Lemma 1.4.6, we have that

D(β1,β2,,βn) = (detT )2D(1,α,,αn1).

It follows Proposition 1.4.13 that D(1,α,,αn1)0, so we have the result.

Remark 1.4.17.

Together, Lemma 1.4.6 and Corollary 1.4.16 tell us that the discriminant of a finite separable field extension LK (relative to an ordered basis) reduces to an element of K×K×2 that is independent of the choice of basis.

Definition 1.4.18.

Let BA be an integral extension of domains such that A is integrally closed, and suppose that B is free of rank n as an A-module. Let (β1,,βn) be an ordered basis of B as a free A-module. The discriminant B over A relative to the basis (β1,,βn) is D(β1,,βn).

Lemma 1.4.19.

Let A be an integrally closed domain with quotient field K. Let L be a finite separable extension of K, and let B denote the integral closure of A in L. Let (α1,,αn) be any ordered basis of L as a K-vector space that is contained in B. Let β L be such that TrLK(𝛼𝛽) A for all α B. Then

D(α1,,αn)β i=1nAα i.
Proof.

Since β L, we may write

β =i=1na iαi

for some ai K for 1 i n. For any i, we have that

TrLK(αiβ) =j=1na jTrLK(αiαj). (1.4.1)

The right-hand side of (1.4.1) is the ith term of the product of the matrix Q = (TrLK(αiαj)) times the column vector with ith entry ai. Since the determinant of Q is d = D(α1,,αn), letting Q Mn(A) denote the adjoint matrix to Q, we have QQ = dIn. Thus, we have dai A for each i. In other words, 𝑑𝛽 lies in the A-module generated by the αi, so we are done.

Proposition 1.4.20.

Let A be an integrally closed domain with quotient field K. Let L be a finite separable extension of K, and let B denote the integral closure of A in L. There exists an ordered basis (α1,,αn) of L as a K-vector space contained in B. Moreover, for any such basis, we have

i=1nAα i B i=1nAd1α i,

where d = D(α1,,αn).

Proof.

First, take any ordered basis (β1,,βn) of LK. By Proposition 1.2.25, there exists a A{0} such that αi = aβi B for each 1 i n. Clearly, (α1,,αn) is a basis of LK, so in particular, the A-module generated by the αi is free and contained in B. The other containment is simply a corollary of Lemma 1.4.19 and the fact that TrLK(B) A.

The following notion of rank is most interesting for finitely generated modules, though we shall have occasion to use it without this assumption.

Definition 1.4.21.

The rank of a module M over a domain A is

rankA(M) = dimK(K AM).

Corollary 1.4.22.

Let A be an integrally closed Noetherian domain with quotient field K. Let L be a finite separable extension of K, and let B denote the integral closure of A in L. Then B is a finitely generated, torsion-free A-module of rank [L : K].

Proof.

By Proposition 1.4.20, we have free A-modules M and M of rank n = [L : K] such that M B M. Since M has no A-torsion, neither does B. We have

K AM K AB K AM.

As M and M are both isomorphic to An, their tensor products over A with K are n-dimensional K-vector spaces, which forces K AB to have K-dimension n as well. Moreover, B is finitely generated being a submodule of a finitely generated module over A, as A is Noetherian.

Proposition 1.4.23.

Let A be an integrally closed Noetherian domain with quotient field K. Let L be a finite separable extension of K, and let B denote the integral closure of A in L. Then any finitely generated, nonzero B-submodule of L is a torsion-free A-module of rank [L : K].

Proof.

Let M be a finitely generated, nonzero B-submodule of L. If β L×, then the multiplication-by-β map B 𝐵𝛽 is an isomorphism of B-modules, so 𝐵𝛽 has rank [L : K] as an A-module. In particular, rankA(M) rankA(B), taking β M. Since M is B-finitely generated and contained in the quotient field of B, there exists α B such that 𝛼𝑀 B. Since multiplication by α is an isomorphism, rankA(M) rankA(B). The result now follows from Corollary 1.4.22.

Corollary 1.4.24.

Let A be a PID with quotient field K, let L be a finite separable extension of K, and let B denote the integral closure of K in L. Then any finitely generated, nonzero B-submodule of L is a free A-module of rank [L : K].

Proof.

By the structure theorem for modules over a PID, any torsion-free rank n module over A is isomorphic to An. The result is then immediate from Proposition 1.4.23.

We have the following application to number fields.

Lemma 1.4.25.

Let K be a number field. Then the discriminant of 𝒪K over is independent of the choice of ordered basis of 𝒪K as a free -module.

Proof.

By Corollary 1.4.24, the ring 𝒪K is free of rank n = [K : ] over . If β1,,βn and α1,,αn are bases of 𝒪K as a free -module, then there exists a -linear homomorphism T : K K such that T (αi) = βi for all i. Then

D(β1,,βn) = det(T )2D(α1,,α n),

and det(T ) is a unit in , so in {±1}, which is to say that det(T )2 = 1.

Definition 1.4.26.

If K is a number field, the discriminant disc(K) of K is the discriminant of 𝒪K over relative to any basis of 𝒪K as a free -module.

Noting Theorem 1.2.28, the case of quadratic fields is immediately calculated as in Example 1.4.14.

Proposition 1.4.27.

Let K = (d), where d1 is a square-free integer. Then

disc(K) = { d d 1mod4, 4d d 2,3 mod 4.

We end with the following general result.

Proposition 1.4.28.

Let A be an integrally closed domain with quotient field K. Let L and L be finite separable extensions of K that are linearly disjoint, and let B and B denote the integral closures of A in these fields, respectively. Suppose that B is A-free with basis β1,,βn and that B is A-free with basis γ1,,γm. Set d = D(β1,,βn) and d = D(γ1,,γm). Then the K-basis of LL consisting of the elements βiγj for 1 i m and 1 j n has discriminant dm(d)n. Moreover, if we let C denote the integral closure of A in LL and C denote the A-algebra that is the A-span of the βiγj, then (d,d)C C.

Proof.

Since L and L are linearly disjoint, LL is a field extension of K of degree [L : K][L : K] with K-basis the βiγj. Let α C, and write

α =i=1n j=1ma 𝑖𝑗βiγj

for some a𝑖𝑗 K. Set

δj =i=1na 𝑖𝑗βi L

for each 1 j m, so α = j=1mδjγj.

Let K¯ be an algebraic closure of K containing L, and let τ1,,τm be the distinct field embeddings of LL in K¯ that fix L. Let M = (τiγj) Mm(K¯), and let w = (δ1,,δm) Lm. Then 𝑀𝑤 = v, where

v = (τ1α,,τmα).

Let M be the adjoint matrix to M, so we have Mv = det(M)w. As the entries of M and v are contained in the integral closure A¯ of A in K¯, the vector det(M)w K¯n has entries in A¯ as well. Note that det(M)2 = d by Proposition 1.4.10, so in fact dw has entries in A¯L, which is to say that dδj B for each j. Since the βi form a basis for B over A, we have that da𝑖𝑗 A for all i and j. The analogous argument tells us that da𝑖𝑗 A for all i and j as well, so (d,d)α C, as we set out to prove.

Finally, we compute the discriminant of the basis βiγj. Let σ1,,σn be the field embeddings of LL in an algebraic closure K¯ of K that fix L, so

{σiτj1 i n,1 j m}

is the set of field embeddings of LL in K¯ that fix K. Let Q be the matrix in M𝑛𝑚(K¯) which we think of as consisting of n2 square blocks of size m by m each, the (i,j)th of which is the matrix σiβjM. Then Q is a product of two matrices, the first of which is block diagonal with m copies of N = (σiβj)i,j, and the second of which consists of n2 square blocks, the (i,j)th of which is the identity times τiγj. Letting M be as before, a simple computation then tells us that

D(βiγj1 i n,1 j m) = det(Q)2 = det(N)2mdet(M)2n = dm(d)n.

1.5. Normal bases

Definition 1.5.1.

A normal basis of a finite Galois extension LK is a basis of L as a K-vector space of the form {σ(α)σ Gal(LK)} for some α L.

The goal of this section is to prove E. Noether’s theorem that every finite Galois extension has a normal basis. We start with the following lemma.

Lemma 1.5.2.

Let LK be a finite Galois extension with Galois group {σ1,,σn}, where n = [L : K]. Let {α1,,αn} be a basis of L as a K-vector space. Then the set

{(σ1(αj),,σn(αj))1 j n}

is an L-basis of Ln.

Proof.

By Corollary 1.4.16, the discriminant D(α1,,αn) is nonzero. By Proposition 1.4.10, this discriminant is the square of det(σiαj)i,j. Since the latter determinant is therefore nonzero, the vectors in question are linearly independent over L.

Lemma 1.5.3.

Every finite cyclic extension of fields has a normal basis.

Proof.

Let LK be finite cyclic of degree n, generated by an element σ. Then K[Gal(LK)] is isomorphic to K[x](xn1) via the unique K-algebra homomorphism that takes σ to x. As L is a K[Gal(LK)]-module, it becomes a K[x]-module annihilated by xn1. If f = i=0n1cixi K[x] annihilates L, then i=0n1ciσi(α) = 0 for all α L, which by the linear independence of the σi forces f to be zero. Thus, the annihilator of L is (xn1), and by the structure theorem for finitely generated modules over the PID K[x], this means that L has a K[x]-summand isomorphic to K[x](xn1), generated by some α L. Since the latter module has K-dimension n, as does L, the elements {α,σ(α),,σn1(α)} form a K-basis of L.

Theorem 1.5.4 (Normal basis theorem).

Every finite Galois extension of fields has a normal basis.

Proof.

Let LK be a finite Galois extension of degree n. Since any finite extension of finite fields is cyclic, we may by Lemma 1.5.3 suppose that K is infinite. Write Gal(LK) = {σ1,,σn} and σ1 = 1. Let {α1,,αn} be a basis of L as a K-vector space. It suffices to find β L with D(σ1(β),,σn(β))0 by Corollary 1.4.16.

Define an element p L[x1,,xn] by

p(x1,,xn) = det(k=1nσ j1σ i(αk)xk)2.

Note that the coefficients of p are fixed by the elements of Gal(LK), since they permute the columns of the matrix. By Lemma 1.5.2, we can find βj L for 1 j n be such that

j=1nβ j(σ1(αj),σ2(αj),,σn(αj)) = (1,0,,0).

Then for all 1 i,j n, we have

k=1nσ j1σ i(αk)βk = δi,j,

so p(β1,,βn) = det(In)2 = 1, so p0. Since K is infinite, there exist a1,,an K with p(a1,,an)0. For γ = j=1naiαi, we have by Proposition 1.4.10 the first equality in

D(σ1(γ),,σn(γ)) = det(σj1σ i(γ))2 = p(a1,,a n)0.

Find in the notes