4.1 ±Construction of H1r(K; )
LEMMA4.1.1. Let T be a crystallineZp-representation of GK. The inclusion H1f(K;T)H1(K;T)induces, for every integer n1, an injection
H1f(K;T)=pnH1f(K;T),!H1(K;T=pnT):
(4:1:1:1)
PROOF. The mapH1f(K;T)=pnH1f(K;T)!H1(K;T=pnT) factors through H1f(K;T)=pnH1f(K;T)!H1(K;T)=pnH1(K;T)!H1(K;T=pnT):
The second morphism is injective since it is part of the long exact sequence asso-ciated to 0!T!pn T!T=pnT!0.
To show that the first one is injective we have to prove that H1f(K;T)\ pnH1(K;T)pnH1f(K;T). Lethbe a class of a cocycle inH1f(K;T)\pnH1(K;T).
By definition the image of h in H1(K;TZpBcris) is trivial and since H1(K;TZpBcris) is aQp-vector space the same is true for the image ofh=pn. This
exactly means thathbelongs topnH1f(K;T). p
DEFINITION4.1.2. For every integersn1,i2 f0;1;2g and 0rp 2 we define functors
Hir(K; ):RepZp GK0[0;r]crys !Mod(Zp);
Hir(K; ):RepZ=pnZ GK0[0;r]crys !Mod(Z=pnZ);
by putting, for everyTin RepZp GK0[0;r]crysandTin RepZ=pnZ GK0[0;r]crys,
Hir(K;T):Hi(FilrC(M(Dcris(T)))); Hir(K;T):Hi(FilrC(M(D1cris(T))));
where M( ),Dcris( ) andD1cris( ) are the functors defined in 3.1.3 and 3.2.1, and Mod(Zp) denotes the category ofZp-modules.
REMARK4.1.3. For everyrmaxrp 2 we haveHir(K;T)Hirmax(K;T) and Hir(K;T)Hirmax(K;T). Indeed by Remark 3.1.4, we have
FilrM(D[0;r]cris(T))Filr(D[0;r]cris(T)WS)
Xr
j0
Filj(D[0;r]cris(T))WFilr jS
Xr
j0
Filjrmax r(D[0;rcrismax](T))WFilr jS
Xrmax
jrmax r
Filj(D[0;rcrismax](T))WFilrmax jS
Xrmax
j0
Filj(D[0;rcrismax](T))WFilrmax jS
FilrmaxM(D[0;rcrismax](T));
and soH1r(K;T)Hi(FilrC(M(D[0;r]cris(T))))Hi(FilrmaxM(D[0;rcrismax](T)))H1rmax(K;T).
The case ofTis the same.
PROPOSITION4.1.4. Let T be inRepZp GK0[0;r]crys, with0rp 2. The short exact sequence
0!T p!n T!T=pnT!0
induces two long exact sequences of abelian groups connected as in the commutative diagram below.
(4:1:4:1)
PROOF. The right vertical sequence is the long exact sequence of continous Galois cohomology. Set TT=pnT, DDcrys(T), DD1cris(T), MM(Dcrys(T)) and MM(D1cris(T)). We have a short exact sequence 0!D!pn D!D!0, cf. (3.1.3.1), and by applyingM( ) another short exact sequence 0!M!pn M!M!0, cf. 3.2.1.
Therefore, by using Lemma 3.3.2, we have a short exact sequence of complexes 0!FilrC(M)!pn FilrC(M)!FilrC(M)!0:
(4:1:4:2)
The left vertical sequence in (4.1.4.1) is the long exact sequence of cohomology as-sociated to the short exact sequence (4.1.4.2).
To constructt1st:H1r(K;T)!H1(K;T=pnT) we proceed as follow.
H1r(K;T)H1(FilrC(M))
Ext1ModFI(S=pnS)r(1;M) (by Prop:3:3:4)
!Ext1Rep
Z=pnZ GK[0;r]crys(Z=pnZ;Tst1(M)) (by exactness of T1st; cf:2:3:7)
Ext1Rep
Z=pnZ GK[0;r]crys(Z=pnZ;TjGK) (by Prop:3:2:2) Ext1RepZ=pnZ GK(Z=pnZ;TjGK)
H1(K;T=pnT)
To construct the morphismtstwe proceed similarly.
H1r(K;T)H1(FilrC(M))
Ext1Mod(S)r(1;M) (by Prop:3:3:4)
! Ext1Rep
Zp GK[0;r]crys(Zp;Tst(M)) (by Cor:2:3:5)
Ext1Rep
Zp GK[0;r]crys(Zp;TjGK) (by Prop:3:2:2)
H1f(K;T) (by definition)
Ext1RepZp GK(Zp;TjGK)H1(K;T) (by definition)
We get a morphism ofZ-modulestst:H1r(K;T)!H1(K;T) with imageH1f(K;T).
Finally the commutativity for the upper left square in the diagram follows from the additivity ofTst, for the upper right square is evident and for the lower square it follows from the the commutativity of the diagram (2.3.7.1). p COROLLARY 4.1.5. The injection H1f(K;T)=pnH1f(K;T),!H1(K;T=pnT), cf.(4.1.1.1), factors canonically through t1st:H1r(K;T=pnT)!H1(K;T=pnT). We denote by
e:H1f(K;T)=pnH1f(K;T),!H1r(K;T=pnT) the monomorphism obtained.
PROOF. The morphism e is induced by the compositionp(1)r tst1. The factor-isation throught1st follows by the commutativity of the diagram (4.1.4.1). p
We now state and prove the main result of this article.
THEOREM 4.1.6. Let p3 be a prime integer and we fix an integer r with 0rp 2. We consider a crystallineZp-representations of GK0, T, with Hodge-Tate weights in[0;r]and assume e(rmax 1)p 1. Then for every integer n1, H1f(K;T)=pnH1f(K;T)is isomorphic to H1r(K;T=pnT)viae.
PROOF. LetTbe as in Theorem 4.1.6. To show thate:H1f(K;T)=pnH1f(K;T)! H1r(K;T=pnT) is an isomorphism we have to prove thatp(1)r in the diagram (4.1.4.1) is surjective. It is clear that ifH2r(K;T)0 (or more generally ifH2r(K;T) had not p-torsion) thenp(1)r would be surjective. We will prove below thatH2r(K;T)0 ife1 orr1 (cf. Remark 4.1.11) but in general this does not seem to be true. What we do instead is:
(1) we prove that if H2r(K;T=pnT)0 thenp(1)r is surjective, see Proposition 4.1.7;
(2) we prove H2r(K;T=pnT)0 under the hypothesis e(rmax(T) 1)p 1, see Proposition 4.1.10.
PROPOSITION4.1.7. Let T be inRepZp GK0[0;r]cryswith rp 2. If H2r(K;T=pnT)0 then p(1)r :H1r(K;T)!H1r(K;T=pnT)
is surjective.
PROOF. Set MM(Dcris(T)) andMM=pnM. It is enough to show that the reduction modpn mapZ1(FilrC(M))!Z1(FilrC(M)) on 1-cocycles is surjective.
Leta2Mandg2Filr 1MV1be such that rM(a)CM(g);
where CM(g)(WMr 1W1)(g) g. Let a02M, (resp. g02Filr 1MV1) be any lifting of a (resp. g). We have rM(a0) CM(g0)pnj1, for some j12MV1. Denote byj1the class ofj1inMV1. SinceH2r(K;T=pnT)0, there exist z12M andv12Filr 1MV1, such that
rM(z1) CM(v1) j1:
Take any lifting z12M (resp. v12Filr 1MV1) of z1 (resp. v1) and set a1a0pnz1 and g1g0pnv1. We have rM(z1) CM(v1) j1pnj2 for somej22MV1and
rM(a1) CM(g1)pnj1 pnj1p2nj2p2nj2:
SinceMand Filr 1MV1arep-adically complete and separated, it is clear that we
can finish by induction. p
4.1.8
-Let us introduce some notations on the ring S=pS. We have E(u)uemodp.
For any integersj0 and 0de 1 we denote byuhejdithee-partial divided power of u(it satisfies j!uhejdiuejd). We set V1S=pSS=pSSV1SS=pSdu.
The canonical differential d:S!V1S composed with the projection V1S!V1S=pS factors through a derivation S=pS!V1S=pS that we denote still by d. For any f 2S=pSwe will writedf dud (f)du.
LEMMA 4.1.9. Let f :auhejdi be in S=pS, with j0, 0de 1 and a2k. Then f is integrable in S=pS, i.e. there is g2S=pS such thatdud (g)f , if and only if:
- ford6e 1, p does not divide ejd1;
- forde 1, p does not divide e.
PROOF. It is obvious that the conditions are sufficient. Indeed ifd6e 1 andp does not divideejd1, a primitive offis ejd1a uhejd1i; whende 1 andp6 je, a primitive isaeuhe(j1)i.
The conditions of the lemma are also necessary because if they are not satisfied,
we havedud(uhejd1i)0. p
PROPOSITION4.1.10. Let T be inRepZp GK0[0;r]crys, w ith0rp 2. For every n1, theZ=pnZ-module H2r(K;T=pnT) is of finite type and if e(rmax(T) 1) p 1, then H2r(K;T=pnT)0.
PROOF. Thanks to Remark 4.1.3 we may assumerrmax(T). Then1 case implies the casen2 by induction onnanddeÂvissageon the exact sequence
0!pnT=pn1T!T=pn1T!T=pnT!0:
Let us compute H2r(K;T=pT). Set TT=pT, DDcris(T) and DD1cris(T) D=pD, cf. 3.1.3. By construction the smallest jump in the filtration FiliD is rmin(D)0 and the biggest jump is rmax(D)r rmin(T), where rmin(T) is the smallest Hodge-Tate weight of T. Put M:M(D)=pM(D)2ModFI(S=pS)r and C:CM:(WrM1W1) Id. The idea is to show first that
C:Coker(C:Filr 1MSV1!MSV1)
is a finite dimensionalFp-vector space, which implies of course thatH2r(K;T=pT) is
also finite. Then, we show that, under the hypothesis e(r 1)p 1, every ele-ment in this cokernelCis in the image ofrM(see the diagram below), which means H2r(K;T=pT)0.
In the particular case r1, the cokernel C is zero. Indeed, in this case, Filr 1MM so that the operatorWr 1W1:MV1!MV1 can be iterated.
The divided FrobeniusW1:V1!V1is nilpotent:
Wm1(du)upm 1duhpme1i
!upmpm11
e !du; thereforeWr 1W1is also nilpotent,Cis invertible andCis zero.
For the rest of the proof we supposer2. Let us prove dimFpCe(r 1)dimFp(T=pT):
(4:1:10:1) We write
MSV1DWS=pSSSdu:
Anyg inMSV1 is a finite sum of elements of the formvuhejdidu, where v2D,j0 and 0de 1. We want to find conditions such thatgbelongs to the image ofC. It is enough to treat the casegvuhejdidu, andv60. Letsbe the weight such thatv2FilsDnFils1D; thengbelongs to Filr 1MSV1if and only if jr 1 s.
We compute
(Wr 1W1)(g)Ws(v)Wr 1 s(uhejdi)up 1du
Ws(v)1Wr 1 s uejdj! up 1du
Ws(v)1 up(ejd)
j!pr 1 sup 1du
Ws(v)uhp(ejd)i
p(ejd) e
h i
!
j!pr 1 s up 1du:
Sincehp(ejd)e i
where, for every m,sm denotes the sum ofp-adic digits ofm. Clearly, we have spjsjand so conclusion the element gvuhejdidu belongs to the image ofC for every jr 1 and the inequality (4.1.10.1) follows.
To proveH2r(K;T=pT)0 it is enough to show that everygvuhejdidu inMV1, withv2D,j(r 2) and 0de 1, is in the image ofrM. Recall that everyv2Dis a horizontal section (by definition ofrM) andeis coprime withp (because ofr2 ande(r 1)p 1). By Lemma 4.1.9 we may assume moreover that 0de 2 andejd1pmfor some integerm1. We have
pmejd1e(r 2)d1(p 1) ed1p 2;
which shows that there are no such elements. p
REMARK4.1.11. It is clear by the proof of Proposition 4.1.10 that if e1 or r1, then H2r(K;T)0. Indeed, if e1, every element in S is integrable, and so rM is surjective. If r1, then Filr 1MM, which implies that CM:Wr 1W1 Id is an isomorphism: the series P1
i0 (Wr 1W1)i converges to an inverse of CM. This was the idea of the proof in [13]. We do not know if under the hypothesis of Proposition 4.1.10, we have H2r(K;T)0 or not.
End of the proof of Theorem 4.1.6. p COROLLARY4.1.12. Let p3 and let T1 and T2 be two crystalline Zp -representations of GK0with Hodge-Tate weights in[0;r][0;p 2]and assume e(r 1)p 1. Then for every morphism (resp. isomorphism) i:T1=pnT1! T2=pnT2ofZ=pnZ[GK0]-modules, there exists a morphism (resp. isomorphism)~i ofZ=pnZ-modules making the following diagram commutative.
(4:1:12:1)
PROOF. By the constructions above (cf. (4.1.4.1) and Corollary 4.1.5) we have the following commutative diagram
wheree1ande2are isomorphisms by Theorem 4.1.6. Set~ie21H1r(K;i)e1. p
4.2 ±Some complements
Let us finish this section with a variant of Corollary 4.1.12 and some ex-amples.
PROPOSITION4.2.1. Let T1and T2be two crystallineZp-representations of GK0
with Hodge-Tate weights in [0;r][0;p 2] and assume erp 2. Then for every morphism (resp. isomorphism)i:T1=pnT1!T2=pnT2 ofZ=pnZ[GK ]-mod-ules, there exists a morphism (resp. isomorphism)~iofZ=pnZ-modules making the following diagram commutative.
(4:2:1:1)
REMARK4.2.2.
(1) The hypothesiserp 2 is more restrictive than that of Corollary 4.1.12 (it implies (e;p)1 even forr1), but Proposition 4.2.1 gives functoriality forGK-morphismes.
(2) The hypothesis erp 2 is necessary. Indeed consider the following counterexample for ep 1 and r1. Take KQp(mp(Qp)), T1Zp(1) andT2Zp. Clearly T1 and T2 are congruent modulop but H1f(K;T1)ZpFp and H1f(K;T2)ZpFp do not even have the same di-mension over Fp.
PROOF. The proof is similar but simpler than that of Corollary 4.1.12 so we only sketch it. Fori1;2 setDiDcris(Ti) andMiM(Di); we haveTi=pnTi Tst1(Mi=pnMi). Under the hypothesiserp 2, the functorT1st is fully faithful (cf. [8, TheeÂoreÁme 1.0.4]) therefore the morphism i is induced by a unique
morphism i0:M1=pnM1!M2=pnM2 in ModFI(S=pnS)r and we have a commu-tative diagram
wheree1ande2are isomorphisms by Theorem 4.1.6. Set~ie21H1(FilrC(i0))e1. p For an abelian groupAwe denote byA[n] then-torsion subgroup ofA.
PROPOSITION4.2.3. Let2rp 2be an integer and assume ep, then:
(1) The groups H2r(K;Zp(r))[pn]and H2r(K;Z=pnZ(r))are non-zero.
(2) The monomorphism e:H1f(K;Zp(r))=pnH1f(K;Zp(r)),!H1r(K;T=pnT) de-fined in Corollary4:1:5is not surjective.
PROOF. The statement (2) follows from (1) by using (4.1.4.1). Let us prove (1):
it is enough to treat the case n1. Set DDcris(Zp(r)). We have D Dcris(Zp( r)(r))HomZp[GK0] Zp;Acris
W, Fil0DD, Fil1D0, andWD0 s.
HenceM(D)DWSS, FilrM(D)FilrSasS-modules, andM(D)1(r), cf.
2.3.6. We haveH1r(K;Zp(r))H1(FilrC(1(r))) andH1r(K;Fp(r))H1(FilrC(1(r))), where1(r)1(r)=pn1(r). Denote byjthe class ofup 1SduinH2(FilrC(1(r))).
We haved(up)p(up 1Sdu), sopj0. We claim that the class ofup 1duin H2(FilrC(1(r))) is not zero, which implies at the same timej60,H2r(K;Fp(r))60 and H2r(K;Zp(r))[p]60. Let us prove this claim by contradiction: let a be in 1(r)S=pSandgdube in Filr 11(r)V1Filr 1(S=pS)SV1, such that
d
du(a)du (Wr 1W1)(gdu)gduup 1du;
or equivalently
d
du(a) up 1Wr 1(g)gup 1: Since r 11 and ep, by writing gP
jr 1gjuhejdi (with gj2k and 0de 1), and by taking reduction modulo Filp(S=pS), we get that there exists a2S=(pSFilpS)k[u]=up, such that
d
du(a)up 1
which is impossible. p
REMARK4.2.4. Let 2rp 2 be an integer, assumeep and that the residue field ofK is finite. We have
H1f(K;Zp(r))=pH1f(K;Zp(r)) ,!6e H1r(K;Fp(r)) t!1st H1(K;Fp(r));
and by [1, Example 3.9, pg. 359]H1f(K;Zp(r))H1(K;Zp(r)); in particulart1st is surjective but not injective and H1f(K;Zp(r))=pH1f(K;Zp(r)) can be identified to t1st(H1r(K;Fp(r))). We suspect that this phenomenon happens more generally.
PROPOSITION 4.2.5. Let T be a crystalline Zp-representation of GK0 with Hodge-Tate weights in[0;r],0rp 2. We have an isomorphism
t1st:H0r(K;T=pnT)! (T=pnT)GK:
PROOF. SetDDcris(T) andMM(D). A direct computation shows H0r(K;T=pnT)Hom0Mod(S)r 1=pn1;M=pnM
Filr((S=pnS)r0WnD=pnD)WrId Filr(D=pnD)WrId(T=pnT)GK0:
The last inclusion is strict in general. Indeed we have the exact sequence 0!H0r(K;T) p!n H0r(K;T)!H0r(K;T=pnT)!d H1r(K;T) p!n H1r(K;T)! induced by the short exact sequence 0!M p!n M!M=pnM!0:By the theo-rem of Liu (2.3.4)Tstis an equivalence of categories and therefore we have
H0r(K;T)Hom0Mod(S)r 1;M
HomMod(S)r 1;M
! TGK
andH1r(M)H1f(K;T), cf. (4.1.4.1). By definitionH1f(K;T) contains all the torsion of H1(K;T), so the kernel ofH1r(K;T) p!n H1r(K;T) is equal to thepn-torsion subgroup H1(K;T)[pn] ofH1(K;T). As in the proof of (4.1.4.1), the exact functorsTstandTst1 induce the commutative diagram
where the rows are exact. This proves the claim. p