10 Dedekind domains and extensions

Let LK be a finite extension. For xL, we write TrLK(x)K for the trace of the K-linear map LL, yxy.

If LK is separable of degree n and σ1,,σn:LK¯denotes the set of embeddings of L into an algebraic closure K¯, then TrLK(x)=i=1nσi(x)K.

Lemma 10.1. Assuming that:

  • LK a finite separable extension of fields

Then the symmetric bilinear pairing (,)K(x,y)TrLK(xy)

is non-degenerate.

Proof. LK separable tells us that L=K(α) for some αL. Consider the matrix A for (,) in the K-basis for L given by 1,α,αn1.

Then Aij=TrLK(αi+j)=[BB]ij where

B=(111σ1(α)σ2(α)σn(α)σ1(αn1)σ2(αn1)σn(αn1))

So

detA=det(B)2=[1i<jn(σi(α)σj(α))]2

(Vandermonde determinant), which is non-zero since σi(α)σj(α) for ij by separability.

Exercise: On Example Sheet 3 we will show that a finite extension LK is separable if and only if the trace form is non-degenerate.

Theorem 10.2. Assuming that:

Then the integral closure OL of OK in L is a Dedekind domain.

Proof. OL a subring of L, hence OL is an integral domain.

Need to show:

  • (i) OL is Noetherian.
  • (ii) OL is integrally closed in L.
  • (iii) Every 0 prime ideal P in OL is maximal.

Proofs:

  • (i) Let e1,,enL be a K-basis for L. Upon scaling by K, we may assume eiOL for all i.

    Let fiL be the dual basis with respect to the trace form (,). Let xOL, and write x=i=1nλifi,λiK. Then λi=TrLK(xei)OK.

    (For any zOL, TrLK(z) is a sum of elements in K¯ which are integral over OK. Hence TrLK(z)K is integral over OK, hence TrLK(z)OK.)

    Thus OLOKf1++OKfnL. Since OK is Noetherian, OL is finitely generated as an OK-module, hence OL is Noetherian.

  • (ii) Example Sheet 2.
  • (iii) Let P be a non-zero prime ideal of OL, and p:=POK be a prime ideal of OK. Let 0xP. Then x satisfies an equation
    xn+an1xn1++a0=0,aiOK,

    with a00. Then a0POK is a non-zero element of p, hence p is non-zero, hence p is maximal.

    We have OKpOLP, and OLP is a finite dimensional vector space over OKp. Since OLP is an integral domain and finite, it is a field.

Remark. Theorem 10.2 holds without the assumption that LK is separable.

Corollary 10.3. The ring of integers of a number field is a Dedekind domain.

Convention: OK is the ring of integers of a number field – pOK a non-zero prime ideal. We normalise ||p (absolute value associated to vp, as defined in Definition 9.6) by |x|p=(Np)vp(x), where Np=|OKp|.

In the following theorems and lemmas we will have:

Lemma 10.4. Assuming that:

  • 0xO)K

Then
(x)=p0prime idealpvp(x).

Proof. xOK,(p)=(pOK,(p))vp(x) by definition of vp(x).

Lemma follows from property of localisation

I=JIOK,(p)=JOK,(p)

for all prime ideals p.

Notation. POL, pOK non-zero prime ideals. We write P|p if pOL=P1e1Prer and P{P1,,Pr} (ei>0, P distinct).

Theorem 10.5. Assuming that:

  • OK, OL, K, L as usual

  • for p a non-zero prime ideal of OK, we write pOLP1e1Prer

Then the absolute values on L extending ||p (up to equivalence) are precisely ||P1,,||PL.

Proof. By Lemma 10.4 for any 0xOK and i=1,,r we have vPi(x)=eivp(x). Hence, up to equivalence, ||Pi extends ||p.

Now suppose || is an absolute value on L extending ||p. Then || is bounded on , hence is non-archimedean. Let R={xL||x|1}L be the valuation ring for L with respect to ||. Then OKR, and since R is integrally closed in L (Lemma 6.8), we have OLR. Set

P:={xOL||x|<1}=mROL

(where mR is the maximal ideal of R).

Hence P a prime ideal in OL. It is non-zero since pP. Then OL,(p)R, since sOLP|s|=1.

But OL,(p) is a discrete valuation ring, hence a maximal subring of L, so OL,(p)=R. Hence || is equivalent to ||p. Since || extends ||p, POK=p so P1e1PrerP, so P=Pi for some i.

Let K be a number field. If σ:K, is a real or complex embedding, then x|σ(x)| defines an absolute value on K (Example Sheet 2) denoted ||σ.

Corollary 10.6. Let K be a number field with ring of integers OK. Then any absolute value on K is equivalent to either

  • (i) ||p for some non-zero prime ideal of OK.
  • (ii) ||σ for some σ:K,.

Proof. Case 1: || non-archimedean. Then ||| is equivalent to ||p for some prime p by Ostrowski’s Theorem. Theorem 10.5 gives that || is equivalent to ||p for some 𝔭OK a prime ideal with 𝔭|p.

Case 2: || archimedean. See Example Sheet 2.

10.1 Completions

OK a Dedekind domain, LK a finite separable extension.

Let 𝔭OK, POL be non-zero prime ideals with P|𝔭.

We write K𝔭 and LP for the completions of K and L with respect to the absolute values ||𝔭 and ||P respectively.

Lemma 10.7.

  • (i) The natural πP:LKK𝔭LP is surjective.
  • (ii) [LP:KP][L:K].

Proof. Let M=LK𝔭=Im(πP)LP.

Write L=K(α) then M=K𝔭(α). Hence M is a finite extension of K𝔭 and [M:K𝔭][L:K]. Moreover M is complete (Theorem 6.1) and since LMLP, we have M=LP.

Lemma 10.8 (Chinese remainder theorem). Assuming that:

  • R a ring

  • I1,,InR ideals

  • Ii+Ij=R for all ij

Then
  • (i) i=1n=i=1nIi (=I say).
  • (ii) RII=1nRIi.

Proof. Example Sheet 2.

Theorem 10.9. The natural map

LKK𝔭P|𝔭LP

is an isomorphism.

Proof. Write L=K(α) and let f(X)K[X] be the minimal polynomial of α. Then we have

f(X)=f1(X)fr(X)K𝔭[X]

where fi(X)K𝔭[X] are distinct irreducible (separable). Since LK[X]f(X),

LKK𝔭K𝔭[X]fi(X)i=1rK𝔭[X]fi(X).

Set Li=K𝔭[X]fi(X) a finite extension of K𝔭. Then Li contains both K𝔭 and L (use K[X]f(x)K𝔭[X]fi(X) injective since morphism of fields). Moreover L is dense inside Li (approximate coefficients of K𝔭[X]fi(X) with an element of K[X]fi(X)).

The theorem follows from the following three claims:

  • (1) LiLP for some prime P of OL dividing 𝔭.
  • (2) Each P appears at most once.
  • (3) Each P appears at least once.

Proof of claims:

  • (1) Since [Li:K𝔭]<, there is a unique absolute value on Li extending ||𝔭. Theorem 10.5 gives us that |||L is equivalent to ||P for some P|𝔭. Since L is dense in L and Li is complete, we have LiLP.
  • (2) Suppose φ:LiLj is an isomorphism preserving L and K𝔭; then
    φ:K𝔭[X]fi(X)K𝔭[X]fi(X)

    takes x to x and hence fi=fi.

  • (3) By Lemma 10.7, the natural map πP:LKK𝔭LP is surjective for any prime P|𝔭.

    Since LP is a field, πP factors through Li for some i, and hence LiLP by surjectivity of πP.

Example. K=, L=(i), f(X)=X2+1. Hensel’s Lemma version 1 gives us that 15. Hence (5) splies in (i), i.e. 5OL=𝔭1𝔭2.

Corollary 10.10. Let 0𝔭OK a prime ideal. For xL we have

NLK(x)=P|𝔭NLPL𝔭(x).

Proof. Let B1,,Br be bases for LP1,,LPr as K𝔭-vector spaces. Then B=iBi is a basis for LKK𝔭 over K𝔭. Let [mult(x)]B (respectively [mult(x)]Bi) denote the matrix for mult(x):LKK𝔭LKK𝔭 (respectively LPiLPi) with respect to the basis B (respectively Bi). Then

[mult(x)]B=([mult(x)]B1[mult(x)]Br)

hence

NLK(x)=det([mult(x)]B)=i=1rdet[mult(x)]Bi=i=1rNLPiK𝔭(x)