11 Decomposition groups

Definition 11.1 (Ramification). Let 0≠𝔭 be a prime ideal of OK, and

𝔭OL=P1e1β‹―Prer

with Pi distinct prime ideals in OL, and ei>0.

  • (i) ei is the ramification index of Pi over 𝔭.
  • (ii) We say 𝔭 ramifies in L if some ei>1.

Example. OK=β„‚[t], OL=β„‚[T]. OKβ†’OL sends t↦Tn. Then tOL=TnOL, so the ramification index of (T) over (t) is n.

Corresponds geometrically to the degree n of covering of Riemann surfaces β„‚β†’β„‚, x↦xn.

Definition 11.2 (Residue class degree). fi:=[OLβˆ•Pi:OKβˆ•π”­] is the residue class degree of Pi over 𝔭.

Theorem 11.3. βˆ‘β‘i=1reifi=[L:K].

Proof. Let S=OKβˆ–(𝔭). Exercise (properties of localisation):

  • (1) Sβˆ’1OL is the integral closure of Sβˆ’1OK in L.
  • (2) msup𝔭Sβˆ’1OLβ‰…Sβˆ’1P1e1β‹―Sβˆ’1Prer.
  • (3) Sβˆ’1OLβˆ•Sβˆ’1Piβ‰…OLβˆ•Pi and Sβˆ’1OKβˆ•Sβˆ’1𝔭≅OKβˆ•π”­.

In particular, (2) and (3) imply ei and fi don’t change when we replace OK and OL by Sβˆ’1OK and Sβˆ’1OL.

Thus we may assume that OK is a discrete valuation ring (hence a PID). By Chinese remainder theorem, we have

OLβˆ•π”­OLβ‰…βˆi=1rOLβˆ•Piei.

We count dimension as k:=OKβˆ•π”­ vector spaces.

RHS: for each i, there exists a decreasing sequence of k-suibspaces

0βŠ†Pieiβˆ’1βˆ•PieiβŠ†β‹―βŠ†Piβˆ•PieiβŠ†OLβˆ•Piei.

Thus dim⁑kOLβˆ•Piei=βˆ‘β‘j=0eiβˆ’1dim⁑k(Pijβˆ•Pij+1). Note that Pijβˆ•Pij+1 is an OLβˆ•Pi-module and x∈Pijβˆ–Pij+1 is a generator (for example can prove this after localisation at Pi).

Then dim⁑kPijβˆ•Pij+1=fi and we have

dim⁑kOLβˆ•Piei=eifi,

and hence

dim⁑k∏i=1rOLβˆ•Piei=βˆ‘i=1reifi.

LHS: Structure theorem for finitely generated modules over PIDs tells us that OL is a free module over OK of rank n.

Thus OLβˆ•π”­OLβ‰…(OKβˆ•π”­)n as k-vector spaces, hence dim⁑kOLβˆ•π”­OL=n. β–‘

Geometric analogue:

f:Xβ†’Y a degree n cover of compact Riemann surfaces. For y∈Y:

n=βˆ‘x∈fβˆ’1(y)eβˆ’x

where ex is the ramification index of x. Now assume Lβˆ•K is Galois. Then for any ΟƒβˆˆGal⁑(Lβˆ•K), Οƒ(Pi)∩OK=𝔭 and hence Οƒ(Pi)∈{P1,…,Pr}.

Proposition 11.4. The action of Gal⁑(Lβˆ•K) on {P1,…,Pr} is transitive.

Proof. Suppose not, so that there exists iβ‰ j such that Οƒ(Pi)β‰ Pj for all ΟƒβˆˆGal⁑(Lβˆ•K).

By Chinese remainder theorem, we may choose x∈OL such that x≑0modPi, x≑1modΟƒ(Pi) for all ΟƒβˆˆGal⁑(Lβˆ•K). Then

NLβˆ•K(x)=βˆΟƒβˆˆGal⁑(Lβˆ•K)Οƒ(x)∈OK∩Pi=π”­βŠ†Pj.

Since Pj prime, there exists Ο„βˆˆGal⁑(Lβˆ•K) such that Ο„(x)∈Pj. Hence xβˆˆΟ„βˆ’1(Pj), i.e. x≑0modΟ„βˆ’1(Pj), contradiction. β–‘

Corollary 11.5. Suppose Lβˆ•K is Galois. Then e1=β‹―=er=e, f1=β‹―=fr=f, and we have n=efr.

Proof. For any ΟƒβˆˆGal⁑(Lβˆ•K) we have

  • (i) 𝔭OL=Οƒ(𝔭)OL=Οƒ(P1)e1β‹―Οƒ(Pr)er, hence e1=β‹―=er.
  • (ii) OLβˆ•Piβ‰…OLβˆ•Οƒ(Pi) via Οƒ. Hence f1=β‹―=fr. β–‘

If Lβˆ•Kis an extension of complete discretely valued fields with normalised valuations vL, vK and uniformisers Ο€L,Ο€K, then the ramification index is e=eLβˆ•K=vL(Ο€K). The residue class degree is f:=fLβˆ•K=[kL:k].

Corollary 11.6. Let Lβˆ•K be a finite separable extension. Then [L:K]=ef.

OK a Dedekind domain:

Definition 11.7 (Decomposition). Let Lβˆ•K be a finite Galois extension. The decomposition at a prime P of OL is the subgroup of Gal⁑(Lβˆ•K) defined by

GP={ΟƒβˆˆGal⁑(Lβˆ•K)|Οƒ(P)=P}.

Proposition 11.8. Assuming that:

  • Lβˆ•K a finite Galois extension

  • 0β‰ PβŠ†OL a prime ideal

  • P|π”­βŠ†OK

Then
  • (i) LPβˆ•K𝔭 is Galois.
  • (ii) There is a natural map
    res⁑:Gal⁑(LPβˆ•K𝔭)β†’Gal⁑(Lβˆ•K)

    which is injective and has image GP.

Proof.

  • (i) Lβˆ•K Galois implies that L is a splitting field of a separable polynomial f(X)∈K[X]. Hence LP is the splitting field of f(X)∈K[X], hence LPβˆ•K𝔭 is Galois.
  • (ii) Let ΟƒβˆˆGal⁑(LPβˆ•K𝔭), then Οƒ(L)=L since Lβˆ•K is normal, hence we have a map res⁑:Gal⁑(LPβˆ•K𝔭)β†’Gal⁑(Lβˆ•K), σ↦σ|L. Since L is dense in LP, res⁑ is injective. By LemmaΒ 8.2, we have
    |Οƒ(x)|P=|x|P

    for all ΟƒβˆˆGal⁑(LPβˆ•K𝔭) and x∈LP. Hence Οƒ(P)=P for all ΟƒβˆˆGal⁑(LPβˆ•K𝔭) and hence res⁑(Οƒ)∈GP for all ΟƒβˆˆGal⁑(LPβˆ•K𝔭).

    To show surjectivity, it suffices to show that

    |GP|=ef=[LP:K𝔭].

    Write 𝔭OL=P1e1β‹―Prer, f=[OLβˆ•P:OKβˆ•π”­]. Then

    • |GP|=|Gal⁑(Lβˆ•K)|r=efrr=ef (using CorollaryΒ 11.5).

    • [LP:K𝔭]=ef. Apply CorollaryΒ 11.6 to LPβˆ•K𝔭, noting that e,f don’t change when we take completions. β–‘