Chapter V: The generalized P -scheme algorithm
5.7 A P -collection C ˜ induced from C and auxiliary elements
The idempotent decompositionsIK produced in Section 5.4 define a P-scheme of double cosetsCrather than an (ordinary)P-scheme. Section 5.7–5.9 are devoted to turning it to aP-schemeC˜. In particular, this section focuses on the definition ofC˜ as aP-collection.
We assume p > deg(f) in Section 5.7–5.9. As mentioned in Section 5.1, this assumption implies that the wild inertia groupWQ0 ⊆GofQ0 overK0is trivial.
Suppose the partitions inCall have locally constant ramification indices and inertia degrees (with respect to(DQ0,IQ0)). Then forK ∈ F andδ∈IK, the (nonempty) set of maximal idealsPofOK satisfying
δ ≡1 (mod ¯P) where P¯ := P/pOK
Rad( ¯OK) ∩RK
all have the same ramification indexe(P) and the same inertia degreef(P). We denote e(P) by eδ and f(P) by fδ. Note that eδ and fδ are coprime to p by Theorem 5.3 and the assumptionp > deg(f).
Recall that for a finite extensionK ofK0 andi ∈ N+, we denote byAK,i the ring ( ¯OK/Rad( ¯OK))⊗Fq Fqi. To defineC˜, we need an auxiliary collection of elements in ringsO¯K orAK,i. We call such a collection of elements anI-advice:
Definition 5.7. Suppose I = {IK : K ∈ F } is a collection of idempotent de- compositions of the rings RK, K ∈ F, that defines to a P-collection of double cosetsC(with respect toDQ0), such that all the partitions inChave locally constant ramification indices and inertia degrees (with respect to(DQ0,IQ0)). AnI-advice {S,T }consists of the following data:
• S = {sδ : δ ∈ IK, eδ > 1}, where each sδ ∈ S is an element of O¯K such that sδ ∈ m− m2 for all the maximal ideals m of O¯K satisfying δ ≡ 1 (mod m/Rad( ¯OK)).
• T = {tδ : δ ∈ IK, fδ >1}, where eachtδ ∈ T is an element of AK,fδ such thattδ6∈mfor all the maximal idealsmofAK,fδ satisfyingδ ≡1 (mod m), andσK,fδ(tδ) =ξ·tδ, whereξ ∈Fqfδ is a primitivefδth root of unity.13 AnI-advice can be computed fromIby the following lemma. Its proof is deferred to Appendix C.
Lemma 5.17. Under GRH, there exists a subroutine ComputeAdvice that given I = {IK : K ∈ F } as in Definition 5.7, either properly refines some idempotent decompositionIK ∈ I, or computeseδ, fδfor K ∈ F, δ ∈IK and anI-advice.14 Moreover, the subroutine runs in time polynomial inlogpand the size ofF.
We also need the following notations: recall that forH ∈ P, we chose an isomor- phismτH : K →LH overK0 whereK is the unique field inF isomorphic toLH overK0. The induced isomorphismO¯K ∼= ¯OLH identifies eachsδ ∈ S (whereS is as in Definition 5.7) with an element inO¯LH, which we denote bysδ,H. Similarly, we identify eachtδ∈ T with an element inALH,fδ, denoted bytδ,H.
Next we define aP-collectionC˜usingIand anI-advice:
Definition 5.8. Let I = {IK : K ∈ F } be as in Definition 5.7 and {S,T } be an I-advice. LetC = {CH : H ∈ P}be the P-collection of double cosets with respect to DQ0 associated with I (see Section 5.4). For H ∈ P, let K be the unique field inF isomorphic toLH overK0, and define the partitionC˜H ofH\G so that Hg, Hg0 ∈ H\G are in the same block of C˜H iff the following conditions are satisfied:
13We regardδ∈RK ⊆O¯K/Rad( ¯OK)as an element ofAK,fδviaδ7→δ⊗1, andξ∈Fqfδ as an element ofAK,fδviaξ7→1⊗ξ.
14We need to compute the ringsAK,fδ before computing the elementstδ ∈AK,fδ. These rings will be computed before the call of the subroutineComputeAdvice. See Section 5.9.
1. HgDQ0 andHg DQ0 are in the same blockB ofCH.
2. Letδbe the unique idempotent inIKsuch thatτ˜H(δ) =δB(see Definition 5.4), where B ∈ CH is as in the previous condition. If eδ > 1, the order of the unique elementcinκ×Q
0 satisfying
g−1sδ,H +I =c·(g0−1sδ,H +I) is coprime toeδ, whereI = (Q0/pOL)e(Q0)/eδ+1.
3. Letδ ∈IK be as in the previous condition. Let m0 be an arbitrary maximal ideal ofAL,fδ containing Rad( ¯Q0/pOOL
L). Iffδ > 1, the order of the unique element cin(AL,fδ/m0)×satisfying
g−1
tδ,H +m0 =c·(g0−1tδ,H +m0) is coprime tofδ.
DefineC˜={C˜H : H ∈ P}, which is aP-collection. We sayC˜is theP-collection associated withIand{S,T }.
We check thatC˜is well defined:
Lemma 5.18. TheP-collectionC˜in Definition 5.8 is well defined.
The proof of Lemma 5.18 is routine and can be found in Appendix C.
5.8 (C,D)-separatedP-collections
We continue the discussion in the previous section. Our goal is to compute I = {IK : K ∈ F } and an I-advice {S,T } such that the associated P-collection C˜ is a strongly antisymmetric P-scheme. To achieve this goal, we introduce another property ofP-collections called(C,D)-separatedness:
Definition 5.9. Let P be a subgroup system over a finite group G, and let C = {CH :H ∈ P}be aP-collection of double cosets with respect to a subgroupDof G. We say aP-collectionC˜= {C˜H : H ∈ P} is(C,D)-separatedif the following conditions are satisfied:
1. All the partitionsC˜H ∈C˜are invariant under the action ofDby inverse right translation, i.e. for allB ∈C˜Handg ∈ D, the setgB ={Hhg−1 :Hh∈B}
is also inC˜H.
2. ForH ∈ P, the mapπH : H\G → H\G/D sendingHg ∈ H\Gto HgD maps each block ofC˜H bijectively to a block ofCH.
It is worth noting that ifC˜is(C,D)-separated, then all the partitions inC automati- cally have locally constant ramification indices and inertia degrees:
Lemma 5.19. Suppose C˜ = {C˜H : H ∈ P} is a (C,D)-separated P-collection where P, C, D are as in Definition 5.9. LetI be a normal subgroup of D. Then all the partitions inChave locally constant ramification indices and inertia degrees with respect to(D,I).
Proof. FixH ∈ P,B ∈CH, andB˜ ∈C˜H such thatπH( ˜B) = B, whereπH is as in Definition 5.9. LetD0 be a subgroup ofD. Consider arbitraryHgD0, Hg0D0 ∈ B and lift them to Hg, Hg0 ∈ B˜ respectively. Choose h1, . . . , hk ∈ D0 such that the D0-orbit ofHgis{Hgh1, . . . , Hghk}and the cosetsHghiare all distinct. We claim Hg0h1, . . . , Hg0hk are also distinct. Assume to the contrary thatHg0hi1 =Hg0hi2 holds for distinct i1, i2 ∈ [k]. Then Hg0hi1 and Hg0hi2 are in the same block of C˜H. It follows by the first condition in Definition 5.9 that Hghi1 and Hghi2 are also in the same block. ButHghi1 6=Hghi2 and they are both mapped toHgDby πH, contradicting the second condition in Definition 5.9. This proves the claim. So the cardinality of theD0-orbit of anyHg ∈H\Gonly depends on the block inCH containingHgD. In particular, this holds forD0 =DandD0 =I. The lemma then follows from Definition 5.2.
The following lemma provides a criterion for a(C,D)-separatedP-collection to be a strongly antisymmetricP-scheme.
Lemma 5.20. Let P be a subgroup system over a finite group G, and let C = {CH : H ∈ P}be aP-scheme of double cosets with respect to D ⊆ G. Suppose C˜ = {C˜H : H ∈ P} is a compatible, invariant, (C,D)-separated P-collection.
Then it is actually a P-scheme. Moreover, if C is antisymmetric (resp. strongly antisymmetric), so isC˜.
Proof. For the first claim, we just need to showC˜is regular. Consider H, H0 ∈ P withH⊆H0. LetπH :H\G→H\G/Dbe the map sendingHg ∈H\GtoHgD,
and defineπH0 similarly. Then the following diagram commutes.
H\G H0\G
H\G/D H0\G/D
πH,H0
πH πH0
πD
H,H0
For B ∈ C˜H and B0 ∈ C˜H0 containing πH,H0(B), we need to show the map πH,H0|B : B → B0 has the constant degree, i.e., the cardinality ofπH,H−1 0(y)∩B is independent ofy ∈B0. AsC˜is(C,D)-separated, the mapπH sendsBbijectively to πH(B)∈CH, and similarlyπH0 sendsB0 bijectively toπH0(B0)∈CH0. The claim then follows from regularity ofC.
Note that the conjugations also commute with the maps πH, i.e.,πhHh−1 ◦cH,h = cDH,h◦πH for H ∈ P andh ∈ G. Assume C˜is not strongly antisymmetric. Then there exists a nontrivial permutation τ of a blockB ∈ C˜H for some H ∈ P that arises as a composition of mapsσi : Bi−1 → Bi, i = 1. . . , k whereBi is a block of C˜Hi, Hi ∈ P, and σi is of the form cHi−1,h|Bi−1 (whereh ∈ G), πHi−1,Hi|Bi−1, or (πHi,Hi−1|Bi)−1 (see Definition 2.7). As the maps πHi|Bi : Bi → πHi(Bi) are bijective and commute with projections and conjugations, we seeτ0 :=σk0 ◦ · · · ◦σ10 is a nontrivial permutation ofπH(B) ∈ CH, where each mapσi0 := πHi|Bi ◦σi ◦ (πHi−1|Bi−1)−1is of the formcDHi−1,h|Bi−1,πHDi−1,Hi|Bi−1, or(πHDi,Hi−1|Bi)−1. SoC is not strongly antisymmetric. The proof of antisymmetry is the same except that we only consider mapsσi that are conjugations.
We need to compute I = {IK : K ∈ F } and an I-advice {S,T } such that the associated P-collectionC˜is (C,DQ0)-separated. The following lemma states that for P-collections arising from Definition 5.8, the first condition of(C,DQ0)- separatedness is in fact automatic.
Lemma 5.21. LetI,{S,T },C andC˜be as in Definition 5.8. Then all the partitions inC˜are invariant under the action ofDQ0 by inverse right translation.
To prove it, we need the following observation.
Lemma 5.22. Let m0 be a maximal ideal of AL,fδ containing Rad( ¯Q0/pOOL
L). For all x ∈ AL,fδ, ω ∈ IP, and σ ∈ DQ0 such that the image of σ in Gal(κQ0/O¯K0) is the Frobenius automorphismx 7→xq overFq, it holds thatωx ≡x (mod m0)and
σx≡σL,fδ(x) (mod m0).
Proof. By bilinearity, we may assume x = a⊗b where a ∈ O¯L/Rad( ¯OL) and b ∈ Fqfδ. Asω ∈ IP, it holds thatωa ≡ a (mod Rad( ¯Q0/pOOL
L))and hence ω(a⊗b) ≡
ωa⊗b≡a⊗b (mod m0). Similarly, we haveσa≡aq (mod Rad( ¯Q0/pOOL
L))by definition and henceσ(a⊗b)≡σa⊗b≡aq⊗b ≡σL,fδ(a⊗b) (mod m0).
Now we are ready to prove Lemma 5.21.
Proof of Lemma 5.21. ConsiderH ∈ P andHg, Hg0 in the same block ofC˜H. Fix h ∈ DQ0. We prove Hgh−1, Hg0h−1 are also in the same block by verifying the three conditions in Definition 5.8. LetBbe the block ofCH containing bothHgDQ0 andHg0DQ0. The first condition in Definition 5.8 obviously holds forHgh−1 and Hg0h−1 sinceHgh−1DQ0 =HgDQ0 ∈B andHg0h−1DQ0 =Hg0DQ0 ∈B. LetK be the field inF isomorphic toLH overK0. Letδ be the idempotent inIK satisfyingτ˜H(δ) =δB(see Definition 5.4). Supposeeδ >1. By Definition 5.8, the order of the unique elementcinκ×Q
0 satisfying
g−1
sδ,H +I =c·(g0−1sδ,H +I)
is coprime toeδ, where I = (Q0/pOL)e(Q0)/eδ+1. We havehI =I sinceh∈ DQ0. Therefore
hg−1sδ,H +I =hc·(hg0−1sδ,H +I), wherehc∈κ×Q
0 has the same order asc. So the second condition in Definition 5.8 is satisfied byHgh−1andHg0h−1.
Now supposefδ >1. Letm0 be a maximal ideal ofAL,fδ containing Rad( ¯Q0/pOOL
L). By Definition 5.8, the order of the unique elementcin(AL,fδ/m0)×satisfying
g−1tδ,H +m0 =c·(g0−1tδ,H +m0)
is coprime to fδ. Fix σ ∈ DQ0 whose image in Gal(κQ0/O¯K0) is the Frobenius automorphismx7→xqoverFq. Chooseω ∈ IQ0 andi∈Zsuch thath =ωσi. By Lemma 5.22, we have
hg−1tδ,H ≡ωσ
i
(g−1tδ,H)≡σ
i
(g−1tδ,H)≡σL,fi
δ(g−1tδ,H)≡g−1 σiL,f
δ(tδ,H)
≡g−1(ξi·tδ,H)≡ξi·g−1tδ,H (mod m0),
whereξ is the primitivefδth root of unity satisfyingσK,fδ(tδ) =ξ·tδas in Defini- tion 5.7. The same argument showshg0−1tδ,H ≡ ξi ·g0−1tδ,H (mod m0). It follows that
hg−1
tδ,H +m0 =c·(hg0−1tδ,H +m0).
So the third condition in Definition 5.8 is also satisfied byHgh andHg h . We also show thatP-collections arising from Definition 5.8 always satisfy a weaken- ing of the second condition of(C,DQ0)-separatedness, where bijectivity is replaced by injectivity:
Lemma 5.23. LetI,{S,T },C andC˜be as in Definition 5.8. Then forH ∈ P, the mapπH : H\G→H\G/DQ0 sendingHg ∈H\GtoHgDQ0 maps each block of C˜H injectively to a block ofCH.
Proof. ConsiderH ∈ P andg ∈Gandh∈ DQ0 such thatHg 6=Hgh−1. We want to prove thatHgandHgh−1are in different blocks ofC˜H.
LetB be the block ofCH containingHgDQ0 =Hgh−1DQ0. LetK be the field in F isomorphic toLH overK0. Letδbe the idempotent inIK satisfyingτ˜H(δ) =δB (see Definition 5.4). Fixσ ∈ DQ0 whose image inGal(κQ0/O¯K0)is the Frobenius automorphismx7→xq overFq.
As we assumep >deg(f), the wild inertia groupWQ0 ⊆GofQ0overK0is trivial.
SoIQ0 is a cyclic group of ordere(Q0). Fix a generatorωofIQ0. By Theorem 5.3 and Definition 5.2, we knoweδis the smallest positive integerksatisfyingHgω−k= Hg, and fδ is the smallest positive integer k satisfying Hgσ−kIQ0 = HgIQ0. So there exist unique i ∈ {0, . . . , fδ − 1} and j ∈ {0, . . . , eδ − 1} such that Hgh−1 = Hgσ−iω−j. AsHg 6= Hgh−1, we have (i, j)6= (0,0). By replacing h withωjσi if necessary, we may assumeh=ωjσi.
First assume i 6= 0. Then fδ > 1. Let m0 be a maximal ideal ofAL,fδ containing
Q0/pOL
Rad( ¯OL). As shown in the proof of Lemma 5.21, we have
hg−1tδ,H ≡ξi·g−1tδ,H (mod m0),
whereξis a primitivefδth root of unity. The order ofξi isfδ/gcd(fδ, i)>1and is a divisor of fδ. So the third condition in Definition 5.8 is not satisfied byHg and Hgh−1. It follows thatHg andHgh−1 are in different blocks ofC˜H, as desired.
Now assumei= 0andj 6= 0. Theneδ >1. Letme=Q0/pOLandk=e(Q0)/eδ. As shown in the proof of Lemma 5.18, we have g−1sδ,H ∈ mke −mk+1e . Choose πL ∈ me−m2e. We have a group homomorphism IQ0 → κ×Q
0 sending g ∈ IQ0 to the unique elementcg ∈ κ×Q
0 satisfyinggπL+m2e =cg(πL+m2e). This map is
injective since its kernel is WQ0 = {e}. In particular, we know cω is a primitive e(Q0)th root of unity inκ×Q
0. Choosec∈κ×Q
0 such that
g−1sδ,H +mk+1e =c(πLk+mk+1e ),
which exists sinceg−1sδ,H andπLk are both inmke−mk+1e . Then we have
hg−1sδ,H +mk+1e =ω
j
(g−1sδ,H +mk+1e ) = ω
j
(c(πLk+mk+1e ))
=c·cjkω ·(πLk +mk+1e ) =cjkω ·(g−1sδ,H +mk+1e ).
The order ofcjkω ∈ κ×Q
0 ise(Q0)/gcd(e(Q0), jk) = eδ/gcd(eδ, j) >1, which is a divisor ofeδ. So the second condition in Definition 5.8 is not satisfied byHg and Hgh−1. It follows thatHgandHgh−1are in different blocks ofC˜H, as desired.
In the next section, we give subroutines that refine the idempotent decompositions IK so that C˜is eventually a compatible, invariant, (C,D)-separated P-collection, and hence a strongly antisymmetricP-scheme.