Betti number estimates in p-adic cohomology

arXiv:1411.2933v1 [math.AG] 11 Nov 2014

Daniel Caro
Abstract
Within the framework of Berthelots theory of arithmetic D-modules, we prove the p-adic analogue of Betti
number estimates and we give some standard applications.

Contents
1

Lagrangianity in the context of arithmetic D-modules

1.1 Convention and preliminaries on filtered modules . . . . . . . . . . . . . . . . . .
1.2 Purity of the codimension for holonomic modules . . . . . . . . . . . . . . . . . .
1.3 Preliminaries on tangent spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4 Lagrangianity for log-extendable overconvergent isocrystals and the case of curves

.
.
.
.

2
2
4
6
7

Relative generic O-coherence

2.1 Inverse and direct images of complexes of arithmetic D-modules and characteristic varieties . . . . .
2.2 Some kind of Frobenius descent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 The result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9
9
11
13

Betti numbers estimates

3.1 Preliminary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 The result and some applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

15
15
17

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

Introduction
Let k be a perfect field of characteristic p and l be a prime number different to p. When k is algebraically closed,
within the framework of Grothendiecks l-adic etale cohomology of k-varieties, Bernstein, Beilinson and Deligne
in their famous paper on perverse sheaves, more precisely in [BBD82, 4.5.1] (or see the p-adic translation here in
3.2.2), established some Betti number estimates. The goal of this paper is to get the same estimates in the context
of Berthelots arithmetic D-modules. We recall that this theory of Berthelot gives a p-adic cohomology stable under
six operations (see [CT12]) and admitting a theory of weights (see [AC13b]) analogous to that of Deligne in the ladic side (see [Del80]). This allows us to consider Berthelots theory as a right p-adic analogue of Grothendiecks
l-adic etale cohomology. By trying to translate the l-adic proof of Betti number estimates in [BBD82, 4.5.1] into the
framework of arithmetic D-modules, two specific problems appear. The first one is that we do not have a notion of
local acyclicity within the theory of arithmetic D-modules. We replace the use of this notion by another one that we
might call relative generic O-coherence (more precisely, see Theorem 2.3.3). The goal of the chapter two is to prove
this property. The proof of this relative generic O-coherence uses the lagrangianity of the characteristic variety of a
unipotent overconvergent F-isocrystals. This lagrangian result is the main purpose of the first chapter. In the case
of curves, this yields easily the lagrangianity of the characteristic variety of a holonomic arithmetic D-modules. We
recall that, following Kedlayas semistable reduction theorem of [Ked11], any overconvergent F-isocrystal becomes
unipotent after the pullback by some generically etale alteration. This semistable reduction theorem of Kedlaya will
be very useful in the proof of the relative generic O-coherence to reduce by descent to the case of a unipotent Fisocrystal and then to be able to apply the lagrangian result of the first chapter (see the step II.1. of 2.3.3). In this first
1

chapter, adapting Kashiwaras original proof in [Kas77] to our context, we also check the purity of the dimension of
Berthelots characteristic variety of a holonomic arithmetic D-module endowed with a Frobenius structure (see 1.2.9).
The second emerging problem when we follow the original l-adic proof of Betti number estimates is that we still do not
have vanishing cycles theory as nice as within the l-adic framework (so far, following [AC13a] we only have a p-adic
analogue of Beilinsons unipotent nearby cycles and vanishing cycles). Here, we replace successfully in the original
proof on Betti number estimates the use of vanishing cycles by that of some Fourier transform and of Abe-Marmora
formula ([AM, 4.1.6.(i)]) relating the irregularity of an isocrystal with the rank of its Fourier transform. We conclude
this paper by the remark that these Betti number estimates allow us to state that the results of [BBD82, chapters 4 and
5] are still valid (except [BBD82, 5.4.78] because the translation is not clear so far).

Acknowledgment
I would like to thank Ambrus Pl for his invitation at the Imperial College of London. During this visit, we noticed that
the p-adic analogue of this Gabbers proof of purity theorem published in [BBD82] was not obvious since contrary to
the l-adic case we did not have p-adic vanishing cycles. I would like to thank him for the motivation he inspired to
overcome this problem.

Convention, notation of the paper

Let V be a complete discrete valued ring of mixed characteristic (0, p), be a uniformizer, K its field of fractions, k its
residue field which is supposed to be perfect. If l is a field, an l-variety (or simply variety is there is no ambiguity with
the field l) is a separated reduced scheme of finite type over l. We recall that a k-variety X is realizable if there exists
an immersion of the form X P, where P is a proper smooth formal scheme over V. In this paper, k-varieties will
always be supposed to be realizable. We will denote formal schemes by curly or gothic letters and the corresponding
straight roman letter will mean the special fiber (e.g. if X is a formal scheme over V, then X is the k-variety equal
to the special fiber of X). The topological space underlying to a k-variety X is denoted by |X|. When M is a Vb its -adic completion and by MQ := M V K. Concerning the notion of cohomological
module, we denote by M
operations in the theory of arithmetic D-modules, we follow the usual notation (for instance, see the beginning of
[AC13b]). Moreover, is f : P P is a morphism of formal schemes over V, we denote by L f the functor defined by
L f (M) = OP ,Q Lf 1 O f 1 M, for any complex M of OP,Q -modules. When f is flat, we remove L in the notation.
P,Q
If T S is a morphism of schemes and f : X Y is a S-morphism, then we denote by fT : XT YT or simply by
f : XT YT the base change of f by T S.
If j : U X is an open immersion of (realizable) varieties, the functor j! : Dbovhol (X/K) Dbovhol (U/K) (or the
functor j! : Ovhol(X/K) Ovhol(U/K)) will simply be denoted by |Y (in other papers, to avoid confusion, it was
sometimes denoted by kY but, here, there is no risk of ambiguity since we do not work partially).

1 Lagrangianity in the context of arithmetic D-modules

1.1 Convention and preliminaries on filtered modules
We use here the terminology of Laumon in [Lau85, A.1]. For instance, a filtered ring (D, Di ) is a ring D, unitary, non
necessary commutative, with an increasing filtration indexed by Z. If there is no risk of confusion, we will simply say
complete for ind-pro-complete. In this section, (D, Di ) will be a complete filtered ring such that gr(D, Di ) is a left
and right noetherian ring.
Definition 1.1.1.
1. A filtered (D, Di )-module (M, Mi ) is said to be good if (Mi )iZ is a good filtration (see the
definition given in [Lau85, A.1.0]). From [Lau85, A.1.2], a filtered (D, Di )-module is complete.
2. Following [Gro61, 2.1.2], we say that a filtered (D, Di )-module (M, Mi ) is free if (M, Mi ) is a direct sum of some
filtered (D, Di )-modules of the form (D, Di )(n) (where n is some integer and (D, Di )(n) := (D, Di+n )).

1.1.2. We remark that Lemma [Lau83, 3.3.2] are still valid in our context if we suppose that the filtered (D, Di )f

modules are complete. In particular, let 0 (M , Mi ) (M, Mi ) (M , Mi ) 0 be a sequence of good filtered

(D, Di )-modules such that g f = 0. Then 0 grM grM grM 0 is exact if and only if, for any i Z, the
f

sequences 0 Mi Mi Mi 0 are exact. When this property is satisfied, we say that 0 (M , Mi )

g
(M, Mi ) (M , Mi ) 0 is an exact sequence of good filtered (D, Di )-modules.
Let u : (M, Mi ) (N, Ni ) be a morphism of good filtered (D, Di )-modules. From [Lau85, A.1.1.2], we notice
that the filtered (D, Di )-modules ker u and Cokeru are good. Moreover, we have the exact sequences of good filtered
(D, Di )-modules:
0 ker u (M, Mi ) Coimu 0,
0 Imu (M, Mi ) Cokeru 0.
Hence, the category of good filtered (D, Di )-modules is exact and we get a notion of strict morphism of good filtered
(D, Di )-modules (see [Lau83, 1.0]). A morphism u : (M, Mi ) (N, Ni ) of good filtered (D, Di )-modules is strict if and
only if the canonical morphism Coimu Imu is an isomorphism of filtered (D, Di )-modules if and only if, for any
i Z, we have u(Mi ) = u(M) Ni if and only if ker gr(u) = gr ker(u) and coker gr(u) = gr coker(u).
Remark 1.1.3. Let (M, Mi ) be a filtered (D, Di )-module. We remark that (M, Mi ) is a good filtered (D, Di )-module if
and only if there exists a epimorphism of the form u : (L, Li ) (M, Mi ), where (L, Li ) is a free filtered (D, Di )-module
of finite type, such that u(Li ) = Mi (i.e. such that u is strict).
Notation 1.1.4 (Localisation). Let f be a homogeneous element of grD. We denote by D[ f ] the complete filtered ring
of (D[ f ] , D[ f ],i ) relatively to S1 ( f ) := { f n , n N} grD (see the definition after [Lau85, Corollaire A.2.3.4]).
Let (M, Mi ) be a good filtered (D, Di )-module. We put
(M[ f ] , M[ f ],i ) := (D[ f ] , D[ f ],i ) (D,Di ) (M, Mi ),

(1.1.4.1)

the localized filtered module of (M, Mi ) with respect to S1 ( f ). We recall that (M[ f ] , M[ f ],i ) is also a good filtered

(D[ f ] , D[ f ],i )-module (see [Lau85, A.2.3.6]) and grM[ f ] grD[ f ] grD grM (see [Lau85, A.1.1.3]).
The results and proofs of Malgrange in [Mal76, IV.4.2.3] (we can also find the proof in the book [HTT08, D.2.2])
can be extended without further problem in the context of complete filtered ring:
Lemma 1.1.5. Let (M, Mi ) be a good filtered (D, Di )-module. Then there exist some free filtered (D, Di )-modules of
finite type (Ln , Ln,i ) with n N and strict morphisms of good filtered (D, Di )-modules (Ln+1 , Ln+1,i ) (Ln , Ln,i ) and
(L0 , L0,i ) (M, Mi ) such that L M is a resolution of M (in the category of D-modules).
In that case, we say that (L , L,i ) is a good resolution of (M, Mi ).
Proof. This is almost the same than [Mal76, IV.4.2.3.2]. For the reader, we recall the construction: with the remark
1.1.3, there exists a strict epimorphism of good filtered (D, Di )-modules of the form 0 : (L0 , L0,i ) (M, Mi ), with
(L0 , L0,i ) a free filtered (D, Di )-module of finite type. Let (M1 , M1,i ) be the kernel of 0 (in the category of good filtered
(D, Di )-modules: see 1.1.2). Since (M1 , M1,i ) is good, there exists a strict epimorphism of the form 1 : (L1 , L1,i )
(M1 , M1,i ), with (L1 , L1,i ) a free filtered (D, Di )-module of finite type. Hence, the morphism (L1 , L1,i ) (L0 , L0,i ) is
strict. We go on similarly.
Remark 1.1.6. Let (L , L,i ) be a good resolution of (M, Mi ). Then gr(L , L,i ) is a resolution of gr(M, Mi ) by free
gr(D, Di )-modules of finite type.
Lemma 1.1.7. Let K be a complex. Let (Fi K )iZ be an increasing filtration of K . We put
Fi H r (K ) := Im(H r (Fi K ) H r (K )).
Then gri (H r (K )) is a subquotient of H r (gri K ).
3

Proof. For instance, we just have to copy the end of the proof of [HTT08, D.2.4] (or also Malgranges description of
the corresponding spectral sequence in [Mal76, IV.4.2.3.2]).
Proposition 1.1.8. Let (M, Mi ) be a good filtered (D, Di )-module. Let (N, Ni ) be a filtered (D, Di )-module. For any
integer r, there exists a filtration F of ExtrD (M, N) such that
1. ExtrD (M, N) = iZ Fi ExtrD (M, N),
2. 0 = iZ Fi ExtrD (M, N)
3. grF ExtrD (M, N) is a subquotient of ExtrgrD (grM, grN).
In particular, we have the implication
ExtrgrD (grM, grN) = 0 ExtrD (M, N) = 0.
Proof. We can follow the usual arguments: from 1.1.5, there exists a good resolution (L , L,i ) of (M, Mi ). We put
K := HomD (L , N). Since L is a resolution of M by projective D-modules, we get H r (K ) = ExtrD (M, N).
Let Fi K n be the subset of the elements of HomD (Ln , N) such that, for any integer j Z, (Ln, j ) Ni+ j . Since
Ln is a D-module of finite type, we get iZ Fi K n = K n , iZ Fi K n = 0. With the canonical induced filtration on
H r (K ) = ExtrD (M, N) (see 1.1.7), this yields the first two properties. Since Ln is a free filtered (D, Di )-modules
of finite type, we check that the canonical morphism grK n HomgrD (grLn , grN) is an isomorphism. Since grL is
a resolution of grM by projective grD-modules, this implies H r (grK ) = ExtrgrD (grM, grN). We conclude by using
Lemma 1.1.7.

1.2 Purity of the codimension for holonomic modules

(0)

Lemma 1.2.1. Let X be an affine smooth variety over k, D := (X, DX/k ), (Di )iN be its filtration by the order of the
operators, f be an homogeneous element of gr D. Let (M, M i ) and (N, N i ) be two good filtered (D[ f ] , D[ f ],i )-modules
and r be an integer.
1. We have codim ExtrgrD (grM, grN) r.
[f]

2. If r < codimgrM then ExtrgrD (grN, grD[ f ] ) = 0.

[f]

Proof. By construction (see [Lau85, A.2]), we get grD[ f ] (gr D) f . Hence this is [Vir00, III.2.1.2], whose proof by
the way comes from [Mal76, 4.2.4].
1.2.2 (Localisation and -adic completion). Let S := Spf V. Let X be an affine smooth V-formal scheme, Xn the
(0)
(0)
reduction of X modulo n+1 . We put D := (X, DX/S ) and Dn := (X, DXn /Sn ). These rings are canonically filtered
by the order of the differential operators ; we denote by (D, Di ) and (Dn , Dn,i ) the (ind-pro) complete filtered rings.
Let f be an homogeneous element of gr D and fn its image in gr Dn . With the notation of 1.1.4, by using the same
arguments than in the proof of [Abe14, 2.3], we get the canonical isomorphism of (ind-pro) complete filtered rings

(D[ f ] , D[ f ],i ) V V/n+1 (Dn[ fn ] , Dn[ fn ],i ).

(1.2.2.1)

b [ f ] as the -adic completion of D[ f ] , i.e. D

b [ f ] := lim D[ f ] /n+1 D[ f ] limDn,[ f ] . Using [Lau85, A.1]
We put D
n

b[ f ] .
and [Ber96, 3.2.2.(iii)], we get the notherianity of D
b
b Q -module), we set M[ f ] := D
b [ f ] b M (resp.
Finally, if M (resp. N) is a coherent D-module (resp. coherent D
D
b [ f ] b N).
N[ f ] := D
D
bD
b [ f ] is flat.
Lemma 1.2.3. With the notation of 1.2.2, the homomorphism D
4

Proof. This is a consequence of [Ber96, 3.2.3.(vii)], [Lau85, A.2.3.4.(ii)] and 1.2.2.1.

b
Remark 1.2.4. With the notation of 1.2.2, let M be a coherent D-module.
We put Mn := M/n+1 M. From [Ber02,
5.2.3.(iv)], there exists a good filtration (Mn,i )iN of Mn indexed by N. We recall (see notation 1.1.4.1) that we get a
good filtered (Dn,[ fn ] , Dn,[ fn ],i )-module by putting (Mn,[ fn ] , Mn,[ fn ],i ) := (Dn,[ fn ] , Dn,[ fn ],i ) (Dn ,Dn,i ) (Mn , Mn,i ). Moreover,

b [ f ] /n+1D
b [ f ]
Dn,[ fn ] (use 1.2.2.1), then
since D

M[ f ] /n+1 M[ f ] Mn,[ fn ] .

(1.2.4.1)

In particular, this means that the good filtration (Mn,i )iN induces canonically a good filtration (Mn,[ fn ],i )iZ (this is
now indexed in Z) on M[ f ] /n+1 M[ f ] .
b [ f ] -module of finite type, then M[ f ] is complete for the -adic topology.
From [Ber96, 3.2.3.(v)], since M[ f ] is a D

b [ f ] -modules M[ f ]
lim
Hence, we get the canonical isomorphism of D
Mn,[ fn ] .
n

b (0) -module, Car(0) (N) its characteristic variety of level

Lemma 1.2.5. We keep notation 1.2.2. Let N be a coherent D
X,Q
0 (see the definition in [Ber02, 5.2.5]). We put N := (X, N). The following assertion are equivalent
/
1. D( f0 ) Car(0) (N) = 0.
2. N[ f ] = 0.

b
bD
b [ f ] is flat
Proof. Let M be a coherent D-module
without p-torsion such that MQ N. Since the extension D
b and D
b [ f ] ). This yields that N[ f ] = 0 if
(see 1.2.3), we get that M[ f ] is also without p-torsion (p is in the center of D

and only if M[ f ] = 0. Let M := M/M. From 1.2.4.1, we have M[ f ] /M[ f ] M [ f0 ] . Hence, M[ f ] = 0 if and only if
M [ f0 ] = 0 (e.g. see [Ber96, 3.2.2]). From [Ber02, 5.2.3.(iv)], there exists a good filtration (M i )iN of M indexed by
N. From the remark 1.2.4, this induces canonically the (ind-pro) complete (D[ f0 ] , D[ f0 ],i )-module (M [ f0 ] , M [ f0 ],i ). Since
M [ f0 ] is (ind-pro) complete, then the equalities M [ f0 ] = 0 and gr (M [ f0 ] , M [ f0 ],i )) = 0 are equivalent. Also, (gr M) f0 = 0

/ Since (gr M) f0 gr (M [ f0 ] ) (see [Lau85, A.1.1.3]) and since by definition

if and only if D( f0 ) Supp(gr M) = 0.
Car(0) (N) = Supp(gr (M, M i )), we conclude the proof.

Remark 1.2.6. Let A = iN Ai be a graded ring. Let I be a graded ideal. Let a1 , . . . , ar be homogeneous generators
of I. We notice that |SpecA| \ V (I) = ri=1 D(ai ).
The following proposition is the analogue of [Kas77, 2.11]:
b (0) -module and V be an irreducible
Proposition 1.2.7. Let X be a smooth V-formal scheme. Let N be a coherent D
X,Q
component of codimension r of Car(0) (N), the characteristic variety of level 0 of N (see [Ber02, 5.2.5]). Then,
b (0) )) contains V .
Car(0) (Ext rb (0) (N, D
X,Q
DX,Q

Proof. It is enough to follow the proof of [Kas77, 2.11]. More precisely, we can suppose X affine with local coordi(0)
(0)
b
without
nates. We set S := Spf V, D := (X, DX/S ), N := (X, N), D := (X, DX/S ). Let M be a coherent D-module

p-torsion such that MQ N. Let M := M/M. From [Ber02, 5.2.3.(iv)], there exists a good filtration (M i )iN
of M indexed by N. By definition, we have Car(0) (N) = Supp(gr (M, M i )) (we recall that this is independent on the
choice of the good filtration). Let be the generic point of V . With [Mat80, 7.D and 10.B.i)], we check that the
irreducible components of Supp(gr M) are of the form V (J) with J a homogeneous ideal. Let Z be the union of the
irreducible components of Supp(gr M) which do not contain . Then, we get from the remark 1.2.6 that there exists a
/ i.e. D( f0 ) Car(0) (N) = D( f0 ) V .
homegeneous element f gr D such that D( f0 ) and D( f0 ) Z = 0,
(0)
b
Now, suppose absurdly that 6 Car(0) (Ext rb (0) (N, D
X,Q )). Using the same arguments than above, there exists
DX,Q

(0)

b
/ We put h = f g.
a homogeneous element g gr D such that D(g0 ) and D(g0 ) Car(0) (Ext rb (0) (N, D
X,Q )) = 0.
DX,Q

(0)

b
/
Hence, we have D(h0 ) and D(h0 ) Car(0) (N) = D(h0 ) V and D(h0 ) Car(0) (Ext rb (0) (N, D
X,Q )) = 0.
DX,Q

b (0) ) is coherent (right) D

b (0) -module, from Theorem A and B of Berthelot, we get the
1) Since Ext rb (0) (N, D
X,Q
X,Q
DX,Q

b (0) )) = Extr (N, D

bQ ). From 1.2.5, this implies Extr (N, D
bQ ) b D
b [h],Q = 0. Since the
equality (X, Ext rb (0) (N, D
X,Q
b
b
D
D
D
DX,Q

bQ D
b [h],Q is flat (see 1.2.3), we get Extr
extension D
b
D

[h],Q

b [h],Q )
bQ ) b D
b
(N[h] , D
ExtrDb (N, D
DQ [h],Q = 0.
Q

2) a) Since Car(0) (N) = Supp(gr M), then D(h0 ) Car(0) (N) = Supp((gr M)h0 ). Since we have also D(h0 )
(0)
Car (N) = D(h0 ) V , then in particular we get Codim(gr M)h0 = r. Since (gr M)h0 = gr (M [h0 ] ), then from 1.2.1 for
i
any i < r we obtain ExtigrD (gr(M [h0 ] ), grD[h0 ] ) = 0. Then it follows from 1.1.8 that for i < r, ExtD
(M [h0 ] , D[h0 ] ) = 0.
[h0 ]

On the other hand, from 1.2.1 we get for any i > r the inequality Codim(ExtigrD

[h0 ]

[h0 ]

(gr(M [h0 ] ), grD[h0 ] )) > r. Hence, by

reducing D(h0 ) if necessary (use again the remark 1.2.6), for any i > r we get ExtigrD
i
then ExtD

[h0 ]

[h0 ]

(gr(M [h0 ] ), grD[h0 ] ) = 0 and

(M [h0 ] , D[h0 ] ) = 0. To sum up, we have found an homogeneous element h gr D such that D(h0 ) and

i
for i 6= r, ExtD

[h0 ]

(M [h0 ] , D[h0 ] ) = 0.

b[h] ) L D[h ]
2) b) Now, since M[h] is without p-torsion, RHomDb (M[h] , D
RHomD[h ] (M [h0 ] , D[h0 ] ). From the
b
0
D
[h]

[h]

exact sequence of universal coefficients (e.g. see the beginning of the proof of [Vir00, I.5.8]), we get the inclusion
b [h] ) b D[h ] Exti (M [h ] , D[h ] ). Hence, for any i 6= r, from the step 2) a) of the proof, we obtain
ExtiDb (M[h] , D
0
0
0
D
D
[h]

[h]

[h0 ]

b [h] ) b D[h ] = 0. By using [Ber96, 3.2.2.(ii)], since Exti (M[h] , D

b[h] ) is a coherent
the vanishing ExtiDb (M[h] , D
b
D
0
D
[h]

[h]

[h]

b [h] -module, for i 6= r we get Exti (M[h] , D

b [h] ) = 0 and then Exti
D
b
b
D
D
[h]

[h],Q

3) From steps 1) and 2), we have checked that RHomDb

[h],Q

b[h],Q ) = 0 (because D
b [h] D
b [h],Q is flat).
(N[h] , D

b [h],Q ) = 0. By using the biduality isomorphism

(N[h] , D

(see [Vir00, I.3.6]), we get N[h] = 0, which is absurd following Lemma 1.2.5 because D(h0 ).

b (0) -module such that

Theorem 1.2.8. Let X be a smooth V-formal scheme. Let r be an integer, N be a coherent D
X,Q
b (0) ) = 0 for any s 6= r. Then, the characteristic variety Car(0) (N) of N is purely of codimension r.
Ext sb (0) (N, D
X,Q
DX,Q

(0)

b
Proof. If V is an irreducible component of Car(0) (N) of codimension s, then from 1.2.7 we get Ext sb (0) (N, D
X,Q ) 6= 0
DX,Q

since it contains V . Hence s = r.

Corollary 1.2.9. Let X be a smooth integral V-formal scheme of dimension d. Let N 6= 0 be a holonomic F-DX,Q module. Then, the characteristic variety Car(N) of N is purely of codimension d.
Proof. The is a consequence of Virrions holonomicity characterization (see Theorem [Vir00, III.4.2] and of Theorem
1.2.8.

1.3 Preliminaries on tangent spaces

Notation 1.3.1. Let f : X Y be a morphism of smooth k-varieties. We denote by T X (resp. T Y ) the cotangent
space of X (resp. Y ), and by X : T X X, f : X Y T Y T Y the projections and by f : X Y T Y T X the

canonical morphism. We set TXY := 1

f (TX X). We denote by U f the function from the set of subvarieties of T X to
1
the set of subvarieties of T Y defined by posing, for any subvariety V of T X, U f (V ) := f ( f (V )). The application
U : f 7 U f is transitive (with respect to the composition), i.e. we have the equality Ug U f = Ug f for any g : Y Z
(see [HTT08, I.2.4.1]).
Lemma 1.3.2. Let u : Z X be an immersion of smooth k-varieties and f : X P be a morphism of smooth kvarieties.

1. Then we the equality 1

f (TZ X) = TZ P. In particular, if f is a immersion we get the identification U f (TZ X) =

TZ P (as subspaces of T P).

2. Moreover, 1
u (TX X) = Z X TX X TZ X. When u is an open immersion, we have Z X TX X = TZ X.

Proof. The equality 1

f (TZ X) = TZ P can be checked by considering the following diagram with cartesian squares:

o
TZ Z
_

o
TZ X
_


T Z o

TZ P
_


(1.3.2.1)

Z P T
_ P

Z X T
_ X o
u

T X o


X P T P,

where, from [HTT08, I.2.4.1], the composition of the second row is f u (which justifies the cartesianity of the upper
right square).
Concerning the second assertion, both Z X TX X and TZ X are subvarieties of Z X T X. So, the inclusion is local
in X and Z and this is checked by computations in local coordinates.
Lemma 1.3.3. Let f : X P be a morphism of smooth k-varieties and (Xi )1ir be a family of smooth subvarieties
of X (resp. open subvarieties of X) such that X ri=1 Xi . Then, TX P ri=1 TXi P (resp. TX P = ri=1 TXi P).
Proof. From the first equality of Lemma 1.3.2, we reduce to the case X = P. Using the second part of Lemma 1.3.2,
we get the inclusions Xi X TX X TXi X (resp. the equalities Xi X TX X = TXi X), which yields the desired result when
X = P.
Proposition 1.3.4. Let f : X Y be a morphism of smooth varieties.

1. If f is tale, then f is an isomorphism, TXY = 1

f (TY Y ), U f (TX X) = f (TX Y ) TY Y .

2. If f is smooth then U f (TX X) TYY .

3. When f is a smooth surjective morphism, we have the equality U f (TX X) = TYY .
Proof. First, suppose that f is etale. Using [Gro61, 1.7.11.(iv)], we check that f is an isomorphism. By computation

in local coordinates, we get TXY = 1

f (TY Y ), which implies the end of the first assertion. Let us prove the second
assertion. Since this is local in X (indeed, for instance use Lemma 1.3.3 and the transitivity of the application U ), we
can suppose that there exists an etale morphism of the form X AYn whose composition with the projection AYn Y
gives f . By transitivity of U with respect to the composition (see 1.3.1) and by using the tale case, we reduce to the
case where f is the projection AYn Y . Since this is local relatively to Y , we can suppose that Y has local coordinates.
Then, this is an easy computation. When f is a smooth surjective morphism, we get the last assertion because in that
case Y (U f (TX X)) = Y .

1.4 Lagrangianity for log-extendable overconvergent isocrystals and the case of curves
1.4.1. Let P be a smooth formal scheme over V. We recall that for any exact sequence 0 E E E 0 of
coherent F-DP,Q -modules, we have the equality Car(E) = Car(E ) Car(E ) ([Ber02, 5.2.4.(i) and 5.2.7]). Let E be
a coherent F-DX,Q -module. We say that E is Lagrangian (resp. isotropic) if Car(E) is Lagrangian (resp. isotropic).
From 1.2.9, E is Lagrangian if and only if E is holonomic and isotropic.
Let E be a complex of F-Dbcoh (DP,Q ). We say that E is Lagrangian if for any integer n, the characteristic variety
of Hn (E) is Lagrangian. If E E E E [1] is a exact triangle of F-Dbcoh (DP,Q ) then E and E are Lagrangian
if and only if E is Lagrangian.

1.4.2. Let X be an affine smooth variety over k admitting local coordinates t1 , . . . ,td . We denote by X (m) the base
change of X under the mth power of Frobenius of S := Spec k, by F m : X X (m) the relative Frobenius morphism.
(m)
From the equalities [Ber96, 1.1.3.1, 2.2.4.(iii)], we compute that for any j < m, we have in DX/S the equality
<p j >(m) p
)

= 0. From [Ber96, 1.1.3.(ii), 2.2.4.(iii)], for any k and l, we compute that there exists u Zp such that
<pm kl>(m)
<pm k>(m) l
<pm k>(m)
(m)
. Hence, with the formula [Ber96, 2.2.5.1], we check that (DX/S )red = kNd i
) = ui
(i
OX .
<pm >(m)
[pm ]
(m)
(m)
(m)
(m) k
= i in (gr DX/S )red . Then we get (gr DX/S )red = kNd (i ) OX .
Let i be the class of i
(m)
(0)
(m)
From [Ber02, 5.2.2], the canonical morphism (gr DX/S )red F m gr DX (m) /S induced by the morphism DX/S
(0)
(m)
F m DX (m) /S of left DX/S -modules is an isomorphism. We remark that this is a consequence of above computations
<k>(m)
(m)
(0)
k/pm
if pm divides k and
via DX/S F m DX (m) /S is 1 i
and of [Ber00, 2.2.4.3] which states that the image of i
(m)
(0)
otherwise is 0 (and then i is sent to 1 i, where i is the class of i in gr DX (m) /S ).
(i

Proposition 1.4.3. Let X be an integral separated smooth formal V-scheme, Z be a strict normal crossing divisor of
X. We put X# := (X, Z) the corresponding smooth log formal scheme. Let G be a coherent DX# ,Q -module which is
also a locally projective OX,Q -module of finite type (i.e., following [Car09, 4.9 and 4.15], G is a convergent isocrystal
on X# ). Let E := ( Z)(G) be the induced isocrystal on X \ Z overconvergent along Z. We suppose that G is endowed
with a Frobenius structure. Let Z1 , . . . , Zr be the irreducible components of Z. For any subset I of {1, . . . , r}, we put
ZI := iI Zi . Then we have the inclusion
|Car(E)| I{1,...,r} TZI X,

(1.4.3.1)

where TZI X is the standard notation (see 1.3.1).

Proof. This is local so we can suppose X affine and with local coordinates t1 , . . . ,td such that Zi = V (ti ) for i = 1, . . . , r.
We proceed by induction on r. For r = 0 (i.e. Z is empty), this is already known (see the example after [Ber02, 5.2.7]).
b (0)# -module G(0) which is also OX -coherent and
Suppose now r 1. From [Car09, 4.12], there exists a coherent D
X
(0)

such that GQ G. We can suppose that G(0) has no p-torsion (indeed, from [Ber96, 3.4.4], the subsheaf of G(0) of
b (0)# -module and also a coherent OX -module). With the notation [Car09, 5.1], G(Z)
p-torsion elements is a coherent D
X

is also a coherent DX# ,Q -module and a locally projective OX,Q -module of finite type. Hence, from [Car09, 4.14], we

b (m)# (0) (G(0) (Z)) by its p-torsion part.

b (m)# (0) G(Z)
G(Z). We denote by H(m) the quotient of D
get D
X ,Q

b
D
X# ,Q

(m)

b
D
X#

The latter isomorphism implies that HQ G(Z). By using [Ber96, 3.4.5] and [Car09, 4.12], it follows that H(m)
b (m)# -module which is also OX -coherent. Since (X, OX ) is noetherian and H(m) has no
is isogeneous to a coherent D
X
b (m)# -module, this
p-torsion, we get that (X, H(m) ) is a (X, OX )-module of finite type. Since H(m) is a coherent D
X

yields that H(m) is also OX -coherent. (this is a log-variation of [Car06, 2.2.13]).

b (m) (m) H(m) and by N(m) the quotient of M(m) by its p-torsion part. We put H(m) :=
We denote by M(m) := D
b
X
D #
X

H(m) /H(m) ,

(m)

:=

M(m) /M(m)
(m)

and N

(m)

(m)

(m)

:= N(m) /N(m) . Since we have the epimorphism M

(m)

(m)

(m)

, from

[Ber02, 5.2.4.(i)] we get Car (N ) Car (M ) (see the notation [Ber02, 5.2.5]). Since G has a Frobenius

structure, it follows from [Car09, 5.24.(ii)] and [CT12, 2.2.9] that we have the isomorphism DX,Q D G(Z) E.
X# ,Q

Hence,

DX,Q b (m)
D

X,Q

(m)
NQ

E. By definition of Car(E) (see [Ber02, 5.2.7]), by using [Ber96, 3.6.2.(i)] this implies

that for m large enough we have the equality Car(E) = Car(m) (N

and T X of [Ber02, 5.2.2.1]).

(m)

) (modulo the homeomorphism between T (m) X

(m)

(m)

Let Mn be the image of DX,n H

is a good filtration of M

(m)

(m)

. Since H

(m)

(m)

(m)

is DX # -coherent and above all OX -coherent, then Mn

(see [Ber02, 5.2.3]). By definition, this implies Car(m) (M

<pm >

For i > r, we remark that i

(m)
di=r+1V (i )

(m)

(m)

(m)

(m)

(m)

(m)

DX # . Hence, we get in T (m) X = Spec(gr DX/S )red the inclusion Car(m) (M

(see the notation of 1.4.2). Since, modulo the homeomorphism between

(m)
di=r+1V (i )

(m)

) = Supp(nN (Mn /Mn1 )).

T (m) X

and

T X

(m)

of [Ber02,

di=r+1V (i )

5.2.2.1], the closed variety

corresponds to
(see the description given in 1.4.2 of the homeomorphism [Ber02, 5.2.2.1]), then we get Car(E) di=r+1V (i ).
Let Zmin := ri=1 Zi . Then, we get 1 (Car(E)) 1 (di=r+1V (i )) = TZmin X, where : Zmin X T X T X is the
canonical immersion.
For any i = 1, . . . , r, we put Xi := X \ Zi . From the induction hypothesis, we get the first inclusion Car(E|Xi )
Ii TXi ZI Xi = Xi X Ii TZI X, where the union runs through subsets Ii of {1, . . . , r} which does not contain i. Since
i

(Xi ) is a open covering of X \ Zmin and since 1 (Car(E)) TZmin X, we conclude.

In the case of curves, the above proposition (and independent results checked later in this paper) implies the
Lagrangianity for overholonomic arithmetic D-modules.
Proposition 1.4.4. Let P be a smooth formal scheme over V of dimension 1. Let E be a complex of F-Dbcoh (DP,Q ).
Then E is Lagrangian.
Proof. By definition (see 1.4.1), we can suppose that E is a coherent F-DX,Q -module. There exists a divisor T of
P such that putting U := P \ T we get ( T )(E) F-Isoc (U, P/K). Since RT (E) is Lagrangian, by devissage (see
1.4.1), we reduce to the case where E F-Isoc (U, P/K).
From Kedlayas semistable reduction theorem (more precisely the global one, i.e. [Ked11, 2.4.4]), there exists a
finite generically finite and tale morphism a : P P such that, putting U := a1 (U) and T := a1 (T ), P is smooth
and the object a! (E) F-Isoc (U , P /K) comes from a log convergent isocrystal on the smooth log scheme (P , T )
(i.e. the underlying scheme is P and the log structure comes from T ). Since P is a curve, there exists a smooth
formal scheme P over V whose special fiber is (isomorphic to) P . Since a is a finite morphism of normal varieties,
there exists (canonically) a universal homeomorphism morphism b : P P and a finite etale morphism c : P P
such that a = c b. From 1.3.4 and 2.1.3, we check that E is Lagrangian if and only if c! (E) is Lagrangian. Hence,
we can suppose that a is a universal homeomorphism morphism. From the remark 2.2.3 and its notation, for s large
s : P P . Since the morphism
enough, there exists a universal homeomorphism f : P P such that a f = FP/k

f induces the morphism of smooth log-varieties (P, T ) (P , T ), this implies that f ! (a ! (E )) is an object of
F-Isoc (U, P/K) which comes from a log convergent isocrystal on the log scheme (P, T ). From 1.4.3, we get that

s (E )
FP/k
f ! (a ! (E )) is Lagrangian. Since E has a Frobenius structure, we conclude.

2 Relative generic O-coherence

2.1 Inverse and direct images of complexes of arithmetic D-modules and characteristic varieties
Every results of [Lau85, 2] concerning the inverse and direct images of complexes of filtered D-modules are still valid
for arithmetic D-modules at level 0 without new arguments for the check. For the reader convenience, below via the
following two propositions we translate into the context of arithmetic D-modules of level 0 the corollaries [Lau85,
2.5.1 and 2.5.2] that we will need.
(0)

(0)
Proposition 2.1.1 (Laumon). With the notation of 1.3.1, for any F Dbcoh (DY ) such that the restriction f |1
f (|Car (F)|)
(0)

(0)

(0)

(0)
is proper, we have f ! (F) Dbcoh (DX ) and |Car(0) ( f ! F)| f (1
f (|Car (F)|)).
(0)

(0)
Proposition 2.1.2 (Laumon). With the notation of 1.3.1, for any E Dbcoh (DX ) such that the restriction f |1
f (|Car (E)|)
(0)

(0)
(0)
is proper, we have f+(0) (E) Dbcoh (DX ) and |Car(0) ( f+(0) E)| f (1
f (|Car (E)|)) =: U f (|Car (E)|).

(0)

Corollary 2.1.3. Let f : X Y be an tale morphism of integral smooth k-varieties. Let E Dbcoh (DX ), F
(0)

Dbcoh (DY ).
(0)

(0)
1. We have the equality |Car(0) ( f ! (F))| = f (1
f (|Car (F)|)).

2. If f is moreover finite and surjective of degree d, then

|Car(0) ( f+(0) (E))| = U f (|Car(0) (E)|);
(0)

ZCar ( f+(0) f

!(0)

(2.1.3.1)

(0)

(F)) = dZCar (F).

(2.1.3.2)

Proof. First, let us check the first equality. Since this is local in X and Y , we can suppose that X and Y are affine with
(0)
(0)
local coordinates. We have f gr DY gr DX (i.e. f is an isomorphism) and the etale homomorphism of rings
(0)

(0)

(0)

(0)

(0)

(0)

(Y, gr DY ) (X, gr DX ). We get (X, gr f ! (F)) = (X, gr DX )

!(0)

(0)

(Y,gr DY )

(Y, gr F). Since the extension

(Y, gr DY ) (X, gr DX ) is flat, we get Supp((X, gr f (F))) = 1

f (Supp((Y, gr F))), which gives the desired
equality.
Since the second part of the lemma is local in Y , we can suppose that X and Y are affine with local coordinates and
(0)
(0)
f (OX ) is a free OY -module of rank d. We have (Y, f+(0) (E)) = (X, E) and since (Y, DY ) (X, DX ) is finite,
(0)

we get (Y, gr f+(0) (E)) = (X, gr E), which gives 2.1.3.1. Morevoer, (X, gr f+(0) f ! (F)) = ((Y, gr F))d .
2.1.4. Let f : P P be a proper morphism of smooth formal V-schemes. Let (E , ) be a coherent F-DP ,Q -module.
b (0) -module
From the equivalence of categories of [Ber00, 4.5.4], there exist (unique up to isomorphism) a coherent D
P ,Q

b (1) (0) F(0)

F F(0) which induced (E , ) by extension. Fix an integer
F(0) and an isomorphism (0) : D
b
P ,Q
DP ,Q

i Z. From the isomorphisms

b (1) (0) Hi f (0) (F (0) )
D
b
+
P,Q D
P,Q

(1)

[Ber02,3.5.3.1]

b (0) F (0) ) Hi f (1) (F F (0) )

Hi f+(1) (D
b
+
P ,Q D
P ,Q

(0)

[Ber02,3.5.4.1]

F Hi f+(0) (F (0) ),

b (0) -module without p-torsion E(0) such that

we get Car(Hi f+ (E )) = Car(0) (Hi f+(0) (F(0) )). Choose a coherent D
P
(0)

EQ F(0) . Since Hi f+(0) (F(0) ) (Hi f+(0) (E(0) ))Q , then by putting G(0) equal to the quotient of Hi f+(0) (E(0) )
by its p-adic torsion part, Car(0) (Hi f+(0) (F(0) )) = Car(0) (G(0) ) Car(0) (Hi f+(0) (E(0) )). Hence,
|Car(Hi f+ (E ))| |Car(0) (Hi f+(0) (E(0) ))| := |Car(0) (k V Hi f+(0) (E(0) ))|.
By using a spectral sequence (the result is given in the beginning of the proof of [Vir00, I.5.8]), we obtain the
monomorphism k V Hi f+(0) (E(0) ) Hi (k LV f+(0) (E(0) )). Hence, |Car(0) (k V Hi f+(0) (E(0) ))| |Car(0) (Hi (k LV

f+(0) (E(0) )))|. We have k LV f+(0) (E(0) ) f+(0) (E

(0)

), where E

(0)

:= k V E(0) k LV E(0) . Finally we get

(0)

|Car(Hi f+ (E ))| |Car(0) (Hi f+(0) (E

))|.

(0)

(0)

From 2.1.2, since f is proper then |Car(0) (Hi f+(0) (E

caracterisc variety of

E ,

we have

Car(E ) =

Car

(0)

(0)

(E

))| U f (|Car(0) (E

(2.1.4.1)

)|). By Berthelots definition of the

). Hence, we have checked the following proposition.

Proposition 2.1.5. Let f : P P be a proper morphism of smooth formal V-schemes. Let (E , ) F-Dbcoh (DP ,Q ).
We have the inclusion
|Car( f+ (E ))| U f (|Car(E )|).
2.1.6. Let f : P P be a finite tale surjective morphism of smooth formal V-schemes.

10

1. In that case, since f+(0) = f (and then preserves the property of p-torsion freeness) and f+ = f , the inclusion
2.1.4.1 is an equality. In fact, with the usual notation of Abe (see [Abe10]), we get the equality
(0)

ZCar(Hi f+ (E )) = ZCar(0) (Hi f+(0) (E

)).

2. Moreover, let (E, ) be a coherent F-DP,Q -module. From the equivalence of categories of [Ber00, 4.5.4],
b (0) -module F(0) and an isomorphism (0) : D
b (1) (0)
there exist (unique up to isomorphism) a coherent D
P,Q

F(0)

such that F(0)

(0)

P,Q

b
D
P,Q

b (0) -module without p-torsion E(0)

which induced (E, ) by extension. Choose a coherent D
P

(0)
(0)
b (0) , we check that
b = f D
b (0) = f !(0) D
EQ . Since DP ,Q = f DP,Q = f ! DP,Q and D
P
P
P

F F(0)

(0)

(0)

f ! E(0) = f E(0) has no p-torsion and that E DP ,Q b (0) f ! E(0) . Moreover, putting E
DP

we get

(0)

(0)

ZCar( f ! (E)) = ZCar( f ! E ).

:= k V E(0) ,

(2.1.6.1)

Proposition 2.1.7. Let f : P P be a finite tale surjective morphism of degree d of integral smooth formal Vschemes. Let E F-Dbcoh (DP,Q ). Then we get
ZCar( f+ f ! E) = dZCar(E);

(2.1.7.1)

(P, f+ f (E)) = d (P, E).

(2.1.7.2)

Proof. From 2.1.3.2 and 2.1.6, we get ZCar( f+ f ! E) = dZCar(E). Then, the equality 2.1.7.2 is a consequence of
2.1.7.1 and of Berthelots index theorem [Ber02, 5.4.4].

2.2 Some kind of Frobenius descent

We will need some kind of Frobenius descent so we introduce the following notation.
2.2.1. Let Fk : k k be the Frobenius map and 0 : V V be a lifting of Frobenius. Let s be a positive integer and
= s0 the corresponding lifting of the sth power of the Frobenius map. Let X be a k-variety and P be a formal
V-scheme. We denote by X and P the corresponding k-variety (resp. formal V-scheme) induced by the base change
s : X X and F s

by . We will denote by FX/k

P/V : P P the corresponding relative Frobenius.
Lemma 2.2.2. Let f : X P be a dominant morphism of smooth integral k-varieties. Let Z X be a proper closed
subset. Then there exists a dense open subvariety U of P, a universal homeomorphism g : U U with U normal, a
e is smooth over U ,
projective generically finite and etale U -morphism of the form f : Ve (X P U )red such that V
1

e
f (Z P U )red is a strict normal crossing divisor in V relatively to U .
Proof. Let l be the field of fraction of P, X(l) := X P Spec (l), Z(l) := Z X X(l) . Let l be an algebraic closure
of l, L := l Gal(l/l) the fixed field by Gal(l/l). We put Y(L) := X(l) Spec (l) Spec (L), X(L) := (Y(L) )red and Z(L) :=
(Z(l) Spec (l) Spec (L))red . From desingularisation de Jongs theorem (see [dJ96] or [Ber97, 4.1]), since L is per X

fect, there exists a projective and generically finite and etale morphism L : X(L)
(L) such that X(L) is smooth
:= 1 (Z ) is a strict normal crossing divisor in X . By using [Gro66, 8.8.2.(ii)], there exists a fiand Z(L)
(L)
L
(L)

nite (radicial) extension l of l included in L such that there exists l -varieties X(l ) and X(l ) satisfying X(L)

X(l ) Spec(l ) Spec(L) and X(L)

X(l ) Spec (l ) Spec (L). By increasing l is necessary, we can suppose that X(l )
and X(l ) are reduced. (Indeed, since X(L) is reduced then so is X(l ) Spec (l ) Spec(L). From [Gro60, 5.1.8], this yields
X(l ) Spec(l ) Spec (L) = (X(l ) Spec (l ) Spec(L))red = ((X(l ) )red Spec (l ) Spec(L))red . Hence, the closed immersion
(X(l ) )red Spec(l ) Spec (L) X(l ) Spec (l ) Spec (L) is in fact an isomorphism. By increasing l is necessary, it follows
from [Gro66, 8.10.5.(i)] that (X(l ) )red X(l ) is an isomorphism.)
We put Y(l ) := X(l) Spec (l) Spec (l ). By increasing l is necessary, it follows from [Gro66, 8.8.2.(i)] that there exists
a morphism l := X(l ) X(l ) (resp. X(l ) Y(l ) ) inducing L (resp. the surjective closed immersion X(L) Y(L) ).

11

By using [Gro67, 17.7.8] and [Gro66, 8.10.5] (by increasing l is necessary), we can suppose that X(l ) Y(l ) is a
surjective closed immersion (i.e. X(l ) = (Y(l ) )red since X(l ) is reduced), that l is projective and generically finite and

etale morphism, that X(l ) is smooth and Z(l ) := 1

l (Z(l ) ) is a strict normal crossing divisor in X(l ) .

Let P be the normalization of P in l . Then the canonical morphism g : P P is a universal homeomorphism,

i.e. is finite (e.g. use [Liu02, 4.1.27]), surjective and radicial (e.g. use the exercise [Liu02, 5.3.9.(a)]). By using
[Gro66, 8.8.2.(ii)] (this times, we consider the projective system of open affine dense subvarieties of P ), there exists

a dense open affine subvariety U of P , two morphisms V U and Ve U such that X(l ) V U Spec(l )

e U Spec (l ). With the same arguments as above, we can suppose that V and Ve are reduced and
and X(l ) V
that V = (X P U )red . By shrinking U is necessary, from [Gro66, 8.8.2.(i)], there exists a U -morphism f : Ve V
which induces l . By shrinking U is necessary, by using [Gro67, 17.7.8] and [Gro66, 8.10.5] we get the desired
properties.
Remark 2.2.3. Let g : U U be a universal homeomorphism of normal integral k-varieties. Then, for s large enough,
s : U U (we can have also h g = F s ).
there exists a universal homeomorphism h : U U such that g h = FU/k
U /k
Proposition 2.2.4. Let f : X P be a dominant morphism of smooth integral k-varieties. Let Z X be a proper
closed subset. Then, for s large enough, there exists a dense open subvariety U of P, such that, by putting W :=
s : U U with the open
(X P U)red (where X P U means the base change of X by the composition of FU/k

immersion U P ), there exists a projective generically finite and etale U-morphism of the form : W W such
that W is smooth over U, Z := 1 (Z P U)red is a strict normal crossing divisor in W relatively to U.
Proof. Using the lemma 2.2.2 and the remark 2.2.3, we get with their notation the diagram of morphisms of k-varieties
/ Ve


/ (X P U )red
_


T _



/ X P U


X P U



U

/ X P U


/ U

/ X



/ P


/ U

e U U. Since W is smooth over U and U is

where T := (X P U) (X P U ) (X P U )red and W := V

smooth, then W is smooth (and in particular reduced). Since W T is projective, generically finite and etale (i.e.
there exists a dense open subvariety T of T such that its base change by the open immersion T T becomes finite
and etale), this implies that the factorization : W Tred is also projective, generically finite and etale. We remark
that Tred = (X P U)red . Finally Z := 1 ((Z P U)red ) is a strict normal crossing divisor relatively to U.
Lemma 2.2.5. Let f : P P be a finite, surjective morphism of smooth formal V-schemes. Let E be a coherent
DP,Q -module. Then E is OP,Q -coherent if and only if f ! (E) is OP ,Q -coherent.
Proof. Recall that we denote by f the functor defined by f (M) = OP ,Q f 1 OP,Q f 1 M, for any complex M of

OP,Q -modules (remark that f is flat). If E is OP,Q -coherent then f ! (E) f (E) is OP ,Q -coherent. Conversely,
suppose f ! (E) is OP ,Q -coherent. First, suppose that f is finite and tale. Since f+ = f , we get that f+ f ! (E) is
OP,Q -coherent. Since E is a direct factor of f+ f ! (E), then E is also OP,Q -coherent. By splitting f by a universal
homeomorphism followed by a finite etale morphism, then we reduce to the case where f is a universal homeomorphism. In that case, for s large enough, FPs /V : P P factors throught f . Hence, since from [Ber00] the functor
(FPs /V ) induces an equivalence of categories (for coherent DQ -modules) and using the easy implication of the lemma,

12

we can suppose that f = FPs /V (and P = P ). In that case, from the isomorphism given just after Theorem [Ber00,

4.2.4] and from the definition of f ! given in [Ber02, 4.3.2.2], we get that f ! (E) f (E) (this is harder to check when
E is only DP,Q -coherent). We can suppose P affine. Since E is a coherent DP,Q -module, this is sufficient to check
that (P, E) is of finite type over (P, OP,Q ) (see [Car06, 2.2.13]). Since the extension (P, OP,Q ) (P , OP ,Q ) is

faithfully flat (because f : P P is faithfully flat), since (P , f ! E) (P , OP ,Q ) (P,OP,Q) (P, E) is of finite

type over (P , OP ,Q ), we conclude.

2.3 The result

e Db (X/K)).
e
Lemma 2.3.1. Let : Xe X be a finite tale surjective morphism. Let E Dbovhol (X/K) (resp. E
ovhol
e Db (X/K)
e Db (X/K). The property E Db (X/K) is
e
The property E
is equivalent to the property + (E)
isoc
isoc
isoc
+
b
e
equivalent to the property (E) Disoc (X/K).
Notation 2.3.2. Let X be an integral variety and E F-Ovhol(X/K). Then, there exists a smooth dense open subvariety Y of X such that E|Y F-Isoc(Y /K) (see the notation [AC13b, 1.2.14] and use [Car07, 2.3.2]). Then, by
definition, rk(E) means the rank of E|Y (which does not depend on the choice of such open dense subvariety Y ).
Theorem 2.3.3. Let P1 be a smooth separated formal V-scheme, P2 be a proper smooth formal V-scheme, P :=
P1 P2 and pr : P P1 be the projection. Let E be a complex of F-Dbovhol (DP,Q ).
Then there exists an open dense formal subscheme U1 of P1 such that, for any finite tale surjective morphisms of
e 1 P1 and 2 : P
e 2 P2 , putting P
e=P
e1 P
e2, : P
e P and E
e := + (E), we have (pr )+ (E)|U
e 1
the form 1 : P
b
Dcoh (OU1 ,Q ).
Proof. I) We can suppose that 1 = Id. Indeed, consider the following diagram
e1 P
e2
P
1


e2
P1 P

e 1 P2
/P
1


2


/ P1 P2

pr

e1
/P
1

pr

/ P1 .

b
Suppose there exists an open dense formal subscheme U1 of P1 such that pr+ 2+ (+
2 (E))|U1 Dcoh (OU1 ,Q ). By
base change theorem we get

+
+ +
e
e
1+ +
1 pr+ 2+ (2 (E)) 1+ pr+ 2+ 1 (2 (E)) 1+ pr+ 2+ (E) pr+ + (E).

e 1 Db (OU ,Q ).
From Lemma 2.3.1, this implies that (pr )+ (E)|U
1
coh
II) We proceed by induction on the dimension of the support X of E. There exists a smooth dense open subvariety
Y of X such that E|Y F-Dbisoc (Y, P/K) (see the notation [AC13b, 1.2.14] and use [Car07, 2.3.2]). Let j : Y X

be the inclusion. By devissage, we can suppose Y integral, that E is a module and that E j+ j! (E) with j! (E)

F-Isoc (Y, P/K). We put Z := X \ Y . By abuse of notation (to simplify them), P1 will mean a dense open set U1 of
P1 (be careful that the open set has to be independent of the choice of 2 ), X and P will mean the base change of X
and P by the inclusion U1 P1 .
1) From Kedlayas semistable reduction theorem (more precisely the global one, i.e. [Ked11, 2.4.4]), there exists a
surjective, projective, generically finite tale morphism a : X X, with X integral and smooth such that Z := a1 (Z)
is the support of a strict normal crossing divisor of X and, putting Y := a1 (Y ), a (E)|Y extends to a log-F-isocrystal
with nilpotent residues on the smooth log scheme (X , Z ) (i.e. the underlying scheme is X and the log structure
comes from the strict normal crossing divisor Z ). Since a is projective, there exists a closed immersion of the form
b N P (this is the product in the category of formal scheme over V) such that the composition of u with the
u : X P
bN P P induces a. Since a+ a! (E) is a direct factor of E, we reduce to the case where X is smooth,
projection f : P
Z is a strict normal crossing divisor of X and E|Y extends to a log-F-isocrystal with nilpotent residues on (X, Z).
2) From Proposition 2.2.4, by replacing P1 by an open dense formal subscheme if necessary and for s large enough,
putting W := (X P1 P1 )red , there exists a projective generically finite and etale P1 -morphism of the form : W W
13

such that W is smooth over P1 , Z := 1 (Z P1 P1 )red is a strict normal crossing divisor in W relatively to P1 . We
set Y := 1 ((Y P1 P1 )red ). Hence, we get the commutative diagram:
Z 

/ W

(Z P1 P1 )red 


 
Z 

/ P1


/ W = (X P P1 )red  
1
red

pr

/ P = P3 P1

red

pr


/ P P1
2


/ X 

FPs /V
1


pr


/ P P
2
1

/ P1

/ P ,
1

where P3 is a proper smooth integral formal scheme over V, pr : P3 Spf V is the structural morphism, f is in fact
induced by a morphism of the form f : P3 P2 (since is projective, this f is more precisely chosen to be equal
bN P P for some integer N). We put G := a! (E ) = (F s ) (E ) and G := ! (G) =
to the projection f : P
2
2
P1 /V

!



RW
f (G). Since a induces the morphism of smooth log-schemes (W , Z ) (X , Z ), since E |Y extends to a
log-F-isocrystal with nilpotent residues on (X , Z ) then G |Y extends to a log-F-isocrystal with nilpotent residues
on (W , Z ).
3) Let Z1 , . . . , Zr be the irreducible components of Z . For any subset I of {1, . . . , r }, we set ZI := i I Zi . Then
|Car(G )| I {1,...,r } TZ P . Indeed, since the check is local on P , we can suppose P affine with local coordinates
I

t1 , . . . ,td of P inducing local coordinates t 1 , . . . ,t n of W and such that Zi = V (t i ) for i = 1, . . . , r . From [Elk73],
there exists a smooth affine formal V-scheme W whose special fiber is W . Let u : W P be a lifting of W P

and let F := u! (G ). From 1.4.3.1, |Car(F )| I {1,...,r } TZ W . Since G u+ (F ), from [Ber02, 5.3.3], we get
I

|Car(G )| = Uu (|Car(F )|). Using 1.3.2, we get Uu (TZ W ) = TZ P , which gives the desired result.
I

e 3 P3 be the finite etale morphism of integral proper smooth formal V-schemes induced from
4) Let 3 : P

e P
e := + (G ), we get
bN P
bN P ). Putting G
2 by base change by f : P3 P2 (hence, 3 = id 2 : P
2
2
3

e
|Car(pr+ 3+ (G ))| TP1 P1 .
Proof: Since pr is proper, from 2.1.5, we get the inclusion
e ))|).
e ))| Upr (|Car(3+ (G
|Car(pr+ 3+ (G

Using 2.1.7.1, we get |Car(3+ +

3 (G )| = |Car(G )|. Hence,

e ))|) = Upr (|Car(G )|) Upr (I {1,...,r } T P ) = I {1,...,r } (Upr (T P )),

Upr (|Car(3+ (G
Z
Z
I

where the inclusion comes from the step 3). Moreover, from 1.3.2, UuI (TZ ZI ) = TZ P , where uI : ZI P is the
I

closed immersion. By transitivity of the application U (see 1.3.1), Upr (TZ P ) = Upr (UuI (TZ ZI )) = U pr uI (TZ ZI ).
I

Since pr uI : ZI P1 is smooth, using 1.3.4.2, we get the inclusion U pr uI (TZ ZI ) TP1 P1 , which yields the deI

sired result.
5) G is a direct factor of + (G ) (which is by definition f+ (G )).
a) Put := Y (Y P1 P1 )red and b : (Y P1 P1 )red Y the morphisms induced respectively by and a. We

have the isomorphism b! b+ . Indeed, from [AC13b, 1.3.12], since b is a universal homeomorphism, the functors
b! and b+ induce quasi-inverse equivalence of categories (for categories of overholonomic complexes). Since b is
proper, then b+ = b! (i.e., via the biduality isomorphism, b+ commutes with dual functors). Hence, we get that b!
commutes also with dual functors.

b) Since := b is a morphism of smooth varieties and E |Y is an isocrystal, then ! (E |Y ) + (E |Y ).

Hence, since is proper, we get the morphisms by adjunction E |Y + (E |Y ) + (E |Y ) E |Y .

The composition is an isomorphism. Indeed, since E |Y is an isocrystal, we reduce to check it on a dense open
14

subset of Y . Hence, we can suppose that and b are morphisms of smooth varieties. We reduce to check this kind
of property separately for b and . The case of b can be checked using [AC13b, 1.3.12]. Since is generically finite
and etale, the second case is already known.
c) We have just checked that E |Y is a direct factor of + ! (E |Y ). This implies that b! (E |Y ) is a direct

factor of b! + ! (E |Y ) + ! (E |Y ) + (G |Y ). Since (Y P1 P1 )red = a1 (Y ), with the convention

of the paper, we get the morphism j : (Y P1 P1 )red (X P1 P1 )red = W . We have also j : Y W . We

b! (E |Y ) is a direct factor of j (G |Y ). By base change isomorphism, since E j j ! E we obtain

get j+
+ +
+

!
G j+
j G . This yields + (G ) + j+
(G |Y ) j+
+ (G |Y ) and G a! j+
(E |Y ) j+
b (E |Y ),
which gives the desired result.
e := (G) (and with the abuse of notation f : P
e 3 P1 P
e 2 P1 of the paper), the step 5) im6) Putting G
2+

e is a direct factor of + (G
e ) = f+ (G
e ). Since pr 3+ (G
e )
plies (using some base change isomorphism), that G
+
e ), we obtain |Car(pr (G))|
e |Car(pr 3+ (G
e ))|. From part 4), this yields |Car(pr (G))|
e
pr+ 2+ + (G
+ 2+
+
+ 2+

e 2 Spf V be the structural morphism. Since pr (G)

e = pr
e this inclusion is equivalent to
e: P
e + (G),
TP1 P1 . Let pr
+ 2+
e
b
e + (G) Dcoh (OP1 ,Q ) (recall that in the proof of the theorem, by abuse of notation, P1 means a dense open
say that pr

e
e ), we obtain pr
e
e ))
e ). From Lemma 2.2.5, this
e (G)
e (a! (E
e (E
subscheme). Since G
a! (E
pr
(F s )! pr

implies that

e )
e + (E
pr

Dbcoh (OP1 ,Q ),

e Db (OP ,Q ).
e + (E)
which yields pr
coh
1

P1 /V

3 Betti numbers estimates

3.1 Preliminary
Lemma 3.1.1. Let U be an affine smooth variety. Let F F-Isoc (U/K). Then we have the inequalities:
dim H d pU+ (F) rk(F);

(3.1.1.1)

dim H pU! (F) rk(F).

(3.1.1.2)

Proof. Since pU! (F) (pU+ (F )) , where means the O-linear dual functor. Hence, we reduce to check the first
inequality. Since U is affine, H d pU+ (F) corresponds to F=0 , the horizontal sections of F (i.e. the 0th cohomological
space of the Monsky-Washnitzer cohomology of F). Since F=0 K OU is included in F (where OU means the constant
object of F-Isoc (U/K)), we get the desired inequality.
Lemma 3.1.2. Let X be a smooth irreducible curve, j : U X an open immersion such that Z := X \ U is a closed
point. Let F F-Isoc (U/K). Let i : Z X be the closed immersion. Then, for any integer n {0, 1},
dim H n i! j! (F) rk(F).

(3.1.2.1)

Proof. Let j : j! j+ be the canonical morphism. Since i! j+ = 0, since the cone of j has its support in Z, then

by using Berthelot-Kashiwara theorem (see [Ber02, 5.3.3]) we get ker( j ) i+ H 0 i! j! and coker( j ) i+ H 1 i! j! .

0
!
1
!
0
!
1
!
From [AC13b, 1.4.3], we obtain DX i+ H i j! (F) i+ H i j! (DU (F)). Hence, H i j! (F) DZ H i j! (DU (F)) and
we reduce to check 3.1.2.1 for n = 1.
We have the exact sequence 0 j!+ (F) j+ (F) i+ H 1 i! j! (F) 0. From Kedlayas semistable theorem
[Ked03], there exists a finite surjective morphism f : P X, with P smooth integral, such that f ! (F) comes from
a convergent isocrystal on P with logarithmic poles along f 1 (Z). Since P and X are smooth and f is finite and
surjective then f is flat (e.g. see [Gro67, IV.15.4.2]). Let X be an open dense subset of P such that Z := f 1 (Z) X
is a closed point. We get the (quasi-finite) flat morphism a : X X and the open immersion j : U := a1 (U) U.
Since a is flat, then a! is exact. Hence, we get the exact sequence 0 a! j!+ (F) a! j+ (F) a! i+ H 1 i! j! (F) 0.

From [AC13b, 1.4.8], the inclusion j!+ (a! (F)) j+ (a! (F)) factors through the composition a! j!+ (F) a! j+ (F)
!
j+ a (F). Then, we get the epimorphism

i+ H 1 i! j! (a! F) j+ (a! (F))/ j!+ (a! (F)) a! j+ (F)/a! j!+ (F) a! i+ H 1 i! j! (F) i+ H 1 i! j! (F)
15

(we have the last isomorphism because a induces the isomorphism a1 (Z) Z). By applying i! (and by using
Berthelot-Kashiwara theorem), we get dim H 1 i! j! (F) dim H 1 i! j! (a! F). Since a! F is log-extendable, then from
[AC13b, 3.4.12] we obtain the inequality dim H 1 i! j! (a! F) rk(a! F) = rk(F).
3.1.3. Let S be a k-variety. Let us briefly review the geometric Fourier transform [NH04], but only in case of A1S /S.
Fix Q p such that p1 = p. We assume that K. We denote by L the Artin-Schreier isocrystal in
F-Isoc (A1k /K) (see [Ber84]).
We denote by : A1k (A1k ) A1k the canonical duality bracket given by t 7 xy. We denote the composition by
S : A1S S (A1S ) A1k (A1k ) A1k .
p2
p1
Now, consider the following diagram: (A1S )
A1S S (A1S )
A1S , where (A1S ) is the dual affine space over S,
1
which is nothing but AS . Similarly than Katz and Laumon in [KL85, 7.1.4, 7.1.5] (in fact, this paragraph describe the
particular case where r = 1), for any E F-Dbovhol (A1S ), the geometric Fourier transform F (E) is defined to be
e A2 !S L
F (E) := p2+ p!1 E
S

e is compatible with Laumons notation (see [KL85, 7.0.1, page 192]) and was defined in
(cf. [NH04, 3.2.1]1). Here
the context of arithmetic D-modules in [AC13b, 1.1.6].
3.1.4. An important property for us of Fourier transform is the following. The functor F is acyclic, i.e. if E
F-Ovhol(A1S /K) then F (E) F-Ovhol((A1S ) /K) (cf. [NH04, Theorem.3.1]).
Lemma 3.1.5. Let f : T S be a morphism of k-varieties. Let E F-Dbovhol (A1S /K) and F F-Dbovhol (A1T /K). We
have the canonical isomorphisms

f ! F (E) F ( f ! E);

(3.1.5.1)

f+ F (F) F ( f+ F).

(3.1.5.2)

Proof. From some base change isomorphism, we get the first isomorphism:






e A2 f ! !S L .
e A2 !S L
e A2 !S L
p2+ f ! p!1 E
p2+ f ! p!1 E
f ! F (E) = f ! p2+ p!1 E
T

e A2 !T L = F ( f ! E), which gives 3.1.5.1.

e A2 f ! !S L p2+ p!1 ( f ! E)
Since T = S f : , we obtain p2+ f ! p!1 E
T
T
Moreover, similarly and by using the projection formula (see [AC13b, A.6])






e A2 !S L
e A2 f ! !S L
e A2 !T L
p2+ f+ p!1 (F)
p2+ f+ p!1 (F)
f+ F (F) = f+ p2+ p!1 (F)
T
T
S

e A2 !S L = F ( f+ F).
p2+ p!1 ( f+ F)
S

Since the proof of the main result on Betti estimate (see 3.2.2) in the case of curves is easier (e.g. remark that we
do not need in this case the Lemma 2.3.3) and since its proof is made by induction, we first check separately this curve
case via the following lemma.
Lemma 3.1.6 (Curve case). Suppose k is algebraically closed. Let X1 be a projective, smooth and connected curve,
E F-D0 (X1 /K) (see the notation of [AC13b, 1.2]). There exists a constant c(E) such that, for any finite tale
e := + (E), we have
morphism of degree d1 of the form : Xe1 X1 with Xe1 connected, by putting E
e c(E) ;
1. dim H 1 pXe1 + (E)
e c(E)d1 .
2. For any r 0, dim H r pXe1 + (E)
1 Notice

that our twisted tensor product and hers are the same.

16

Proof. There exists an open dense affine subvariety U1 of X1 such that E|U1 F-Dbisoc (U1 /K), Let Z1 be the closed
e1 := 1 (U1 ), Ze1 := 1 (Z1 ) i.e we get
subvariety X1 \ U1 , j : U1 X1 and i : Z1 X1 be the immersions. We put U
1
1
the cartesian squares:
Ze1

1 


Z1

/ Xe1 o

e1
U
1

1 


/ X1 o

U1 .

By considering the exact triangle j! j! (E) E i+ i+ (E) +1 we reduce to check the proposition for E = j! j! (E)
or E = i+ i+ (E) (and because the functors j! j! and i+ i+ preserve D0 ).
1) In the case where E = i+ i+ (E), we can suppose that Z1 is a point. We put G := i+ (E). Since Ze1 is d1 copies of

d1
d1
d1
e
Z1 , then we get 1+ +
1 G G and then pXe1 + (E) pX1 + (i+ G ) G , which gives the desired result.
e = j! (E).
e By devissage, we can suppose that F is a module, i.e.
2) Suppose now E = j! j! (E). We put F = j! (E), F

e
e
F F-Isoc (U1 /K) (and then F F-Isoc (U1 /K)).

e
e for any r 6 {0, 1}, we get H r p e (E)
e
e = 0.
Since U1 is affine, since j! is exact (and E
j! (F)),
H r p e (F)
X1 +

U1 !

e = dim H 1 p e (F)
e rk(F)
e = rk(F). It remains to estimate |(Xe1 , E)|.
e
From 3.1.1.2, we get dim H 1 pXe1 + (E)
From
U1 !
+
e
e
Lemma 2.1.7 we get the second equality (X1 , E) = (X1 , 1+ 1 (E)) = d1 (X1 , E). Hence, we can choose in that
case c(E) = max{|(X1 , E)|; rk(F)}.

3.2 The result and some applications

We will use the following remark during the proof of the main theorem.
Remark 3.2.1. Let M be a solvable differential module on the Robba ring over K. We get the differential slope
decomposition M = M , where M is purely of slope . By definition Irr(M) := 0 rk(M ). Hence, we get that
Irr(M) rk(M) + Irr(M]1,[ ),

(3.2.1.1)

where M]1,[ := ]1,[ M .

Theorem 3.2.2. Suppose k is algebraically closed. Let (Xa )1an be projective, smooth and connected curves, X =
na=1 Xa , E F-D0 (X/K) (see the notation of [AC13b, 1.2]). There exists a constant c(E) such that, for any finite
tale morphism of degree da of the form a : Xea Xa with Xea connected, by putting Xe = na=1 Xea , : Xe X and
e := + (E), we have
E
n
e
1. For any r, dim H r pX+
e (E) c(E) a=1 da .

e
2. For any r 1, dim H r pX+
e (E) c(E) max{aA da | |A| = n r}.
Proof. We proceed by induction the integer n 1. The case n = 1 has already been checked in 3.1.6. Suppose
n 2. Let a : Xea Xa be some finite tale morphism of degree da with Xea connected, Xe = na=1 Xea , : Xe X and
e := + (E). We put Y := a6=1 Xa , Ye := a6=1 Xea , : Ye Y . Let pr : Y Spec k and pr
e : Ye Spec k be the projections
E
e means also the projection pr
e : Xe = Xe1 Ye Xe1 etc.).
(recall that from the convention of this paper, for instance, pr
From Lemma 2.3.3, there exists an affine open dense subvariety U1 (independent of the choice of i ) of X1 such that
e 1 Db (U1 /K). Let Z1 be the closed subvariety X1 \U1 , U
e1 := 1 (U1 ), Ze1 := 1 (Z1 ). Let j : U1 X1
pr+ (+ E)|U
1
1
isoc
i : Z1 X1 be the inclusions.
e = pr
e U
e
e1 F-Db (U
e
e + (E)
e + (E)|
e + (E))|U
We have j! pr
1
isoc 1 /K). Indeed, from 2.3.1, this is equivalent to prove 1+ (pr

b
e
e
e + (E)) pr+ (+ E).
Disoc (U1 /K). Then, we get the desired property from the isomorphism 1+ (pr

Step I) With the notation 2.3.2, we check that there exists a constant c (only depending on E) such that
17

e c n db .
For any s, rk Hs pr
e + (E)
b=2
e c max{bB db | B {2, . . . , n} and |B| = n 1 s}.
For any s 1, rk Hs pr
e + (E)

e1 such that 1 (e
Proof. Let t be a closed point of U1 , e
t be a closed point of U
t ) = t. Let it : t X1 , iet : e
t Xe1 ,

!
b
!
!
e
e
t U1 be the closed immersions. Since the functor et [1] is acyclic on F-Disoc (U1 /K), since iet [1] (et [1]) j! ,
et : e

e
e
e = dim i! [1]Hs pr
e
we obtain i! [1](Hs pr
e (E))
Hs (i! pr
e (E[1])).
Moreover, for such e
t , we have rk(Hs pr
e (E))
e (E).
e
t

e
t

e
t

e1 := + (E1 ). Since t Y is a smooth divisor of X, then i! [1] is right exact. Hence

We put E1 := it! (E)[1] and E
t
E1 F-D0 (Y /K). Since + = ! and iet = it : e
t Ye X, we get the isomorphism

e = i! + (E)[1] + i! (E)[1] = + (E1 ) = E

e1 .
iet! E[1]
t
e
t

e pr
e Hence,
By base change isomorphism, we get: iet! pr
e + (E)
e + iet! (E).

e
e1 ) = pe (E
e
iet! pr
e + (E)[1]
pr
e + (E
Y + 1 ),

e
e1 ) and then
where we have identified Ye with e
t Ye . By composition, we obtain iet! [1](Hs pr
e + (E))
Hs pYe+ (E
s
s
e = dim H pe (E
e
rk(H pr
e + (E))
Y + 1 ). We conclude by applying the induction hypothesis to E1 (notice the fact that U1 is
independent on the choice of i is fundamental, i.e. we do need Theorem 2.3.3).

Step II)
1) By considering the exact triangle j! j! (E) E i+ i+ (E) +1 we reduce to check the proposition for E = j! j! (E)
or E = i+ i+ (E) (and because the functors j! j! and i+ i+ preserve D0 ).

2) Suppose E = i+ i+ (E). We can suppose that Z1 is a irreducible (i.e. since k is algebraically closed, Z1 Speck).
Consider the diagram with cartesian squares:
1

Ze1  _ Ye
i


Xe


1

/ Z1  Ye
_
i

/ Z1  Y
_


/ X1 Ye


/X

pr

/ Z1
_


pr

/ X1 .

e e d1
e := + (G). Since Ze1 is d1 copies of Z1 , then we get 1+ + G
We put G := i+ (E), G
1 (G) . This implies

+
+ +
+ e
e
e d1
pX+
e (E) = pX+
e ( (i+ G)) pX+
e (i+ 1 G) pYe + 1+ 1 (G) (pYe + (G)) ,

(3.2.2.1)

where in the last isomorphism we have identified Ye with Z1 Ye . We conclude by applying the induction hypothesis to
G.
3) Suppose now E = j! j! (E), i.e. i+ (E) = 0.
a) We check that there exists a constant c (only depending on E) such that
e c n db .
For any s, dim H 1 pXe1 + (Hs pr
e + (E))
b=2
e c max{bB db | B {2, . . . , n} and |B| = n 1 s}.
For any s 1, dim H 1 pXe1 + (Hs pr
e + (E))
e
e
e := + (F). By using base change isomorphism, we get j! F
E. Moreover,
Proof: We put F = j! (E), F

!
!
b
e j pr
e j pr
e D (U
e pr
e
e
e1 /K). Moreover, j! pr
pr
e + (F)
e + ( j! F)
e + (E)
e
(
F)
e
j
(
F)

pr
e
(
E).
Since j! is
+
+ !
+
isoc

s
s
e
e
exact, we get j! H pr
e + (F) H pr
e + (E). From 3.1.1.2, this implies

e = dim H 1 p e (Hs pr
e rkHs pr
e = rkHs pr
e
dim H 1 pXe1 + (Hs pr
e + (E))
e + (F))
e + (F)
e + (E).
U1 !
From the step I), we get the desired estimate.
18

b) We have the spectral sequence

e H r+s p e (E).
e
E2r,s = H r pXe1 + Hs pr
e + (E)
X+

e
e we get E r,s = 0 when r 6 {0, 1}.
e1 is affine of dimension 1 and since p e (Hs pr
Since U
e + (E))
pUe1 ! (Hs pr
e + (F)),
2
X1 +
Hence, by using the step a) (we can also vary the order of X1 , . . . , Xn ), it remains to check that there exists a constant
c(E) such that

e c(E) n da .
For any s, |(Xe1 , Hs pr
e + (E))|
a=1

e c(E) max{aA da | |A| = n s}.

For any s 1, |(Xe1 , Hs pr
e + (E))|

i) In this step, we reduce to the case d1 = 1. For this purpose, consider the following diagram
Xe1 Ye
pr
e


Xe1

/ X1 Ye

/X

pr
e
pr

 }
/ X1 .


1

e = (X1 , 1+ Hs pr
e By base change, we get the second isomorphism: 1+ pr
e
We have (Xe1 , Hs pr
e + (E))
e + (E)).
e + (E)

+
+
s e (E)
s e (+ (E)).
e
1+ pr
e + +
e + (+ (E)). Since 1+ and +
1+ +
+
+
1 ( (E)) 1+ 1 pr
1 are exacts, we get 1+ H pr
1 H pr
s e (+ (E))) =
e = (X1 , 1+ + Hs pr
Hence, (X1 , 1+ Hs pr
e + (E))
e + (+ (E))). From Lemma 2.1.7, we have (X1 , 1+ +
+
1
1 H pr
e = d1 (X1 , Hs pr
d1 (X1 , Hs pr
e + (+ (E))). Hence, we have checked that (Xe1 , Hs pr
e + (E))
e + (+ (E))), which yields the
desired result.

ii) We suppose from now that d1 = 1. We prove that we can reduce to the case where X1 = P1k and U1 = A1k . Indeed,
from Kedlayas main theorem of [Ked05], by shrinking U1 is necessary, there exists a finite morphism f : X1 P1k
such that U1 = f 1 (A1k ) and the induced morphism f : U1 A1k is etale. We get the cartesian squares:
pr
e

Xe


P1k Ye

/X
f

pr



/ P1 Y
k

pr

/* X1 o
f


4/ P1k o

U1


A1k

(3.2.2.2)

pr
e

e := + (E )
e We have already seen in Step II.3.a) that pr
e
Set E := f+ (E). Then we get that E
f+ (E).
e + (E)

e Since f+ j!
j! g+ , g+ is exact (because g is finite and etale), and j! is exact, we get
j! pr
e + (F).

e
e
e
e
e
Hs f+ pr
e + (E)
Hs f+ j! pr
e + (F)
j! g+ Hs pr
e + (F)
f+ j! Hs pr
e + (F)
f+ Hs pr
e + (E).

e ) Hs f+ pr
e we obtain (X1 , Hs pr
e = (P1 , f+ Hs pr
e = (P1 , Hs pr
e )).
Morever, since Hs pr
e + (E
e + (E),
e + (E))
e + (E))
e + (E
k
k
iii) We suppose from now X1 = P1k and U1 = A1k . From 3.1.2, for any m {0, 1}, we have
e rk(Hs pr
e = rk(Hs pr
e
Hm (i! j! Hs pr
e + (F))
e + (F))
e + (E)).
e = (X1 , j! Hs pr
e = (X1 , i+ i! j! Hs pr
e
From the step I, this latter is well estimated. Since (X1 , Hs pr
e + (E))
e + (F))
e + (F))+
s
s
e
e
(X1 , j+ H pr
e + (F)), we reduce to estimate (X1 , j+ H pr
e + (F)).
From Christol-Mebkhouts Theorem [CM01, 5.0-10] (as described in the introduction), we have the following
p-adic Euler-Poincare formula:
s
e = (U1 , Hs pr
e = rk (Hs pr
e
e
(X1 , j+ Hs pr
e + (F))
e + (F))
e + (F))(U
e + (F)),
1 ) Irr (H pr

19

where is the complement of U1 in X1 , i.e. of A1K in P1k .

From Abe-Marmoras formula [AM, 2.3.2 and 4.1.6.(i)], with the inequality 3.2.1.1, we get the following one:
e rk(F (Hs pr
e + 2rk(Hs pr
e
Irr (Hs pr
e + (F))
e + (F)))
e + (F)),
where F means the Fourier transform (see 3.1.3). Hence, we reduce to check the (last) step iv).
e
iv) In this step, we estimate rk(F (Hs pr
e + (F))).
From the step I) applied to M := j+ F ( j! E), there exists a constant c (only depending on E) such that
For any s, rk(Hs pr
e + (+ M)) c nb=2 db .

For any s 1, rk(Hs pr

e + (+ M)) c max{bB db | B {2, . . ., n} and |B| = n 1 s}.
e = rk(Hs pr
It remains to check that rk(F (Hs pr
e + (F)))
e + (+ M)). Since = (i.e. d1 = 1), we get the diagram
pr
e

XeO

/X
O

pr

e
U

/U

pr

/) X1
O

(3.2.2.3)

/5 U1

pr
e

By base change (recall that ! = + ) and next by using 3.1.5, we have

e j+ (F ( pr
e
pr
e + (+ M) = pr
e + (+ j+ F (F)) pr
e + j+ + (F (F)) j+ pr
e + + (F (F)) j+ pr
e + (F (F))
e + F)).

e
Since the Fourier transform F is acyclic (see 3.1.4 ) and so is j+ , we get Hs pr
e + (+ M) j+ (F (Hs pr
e + F)).
s
s
e
e
Since rk(F (H pr
e + F) = rk( j+ (F (H pr
e + F)) (recall the notation of 2.3.2), we can conclude.
From Theorem 3.2.2, the reader can check the p-adic analogues of corollaries [BBD82, 4.5.25] by copying
the proofs. Moreover, from [AC13b], we have a theory of weight within the framework of arithmetic D-modules.
For instance, we have checked the stability properties under Grothendieck six operations, i.e. the p-adic analogue
of Deligne famous work in [Del80]), which is also explained in [BBD82, 5.1.14]. In [BBD82, 5.2.1], a reverse
implication was proved. The reader can check that we can copy the proof without further problems (i.e., we only
have to check that we have nothing new to check, e.g. we already have 3.1.1 or the purity of the middle extension of
some pure unipotent F-isocrystal as given in [AC13b, 3.6.3]). For the reader, let us write this p-adic version and its
important corollary [BBD82, 5.3.1] (this corollary is proved in [AC13b] in another way, but Theorem 3.2.3 below is a
new result).
Theorem 3.2.3 ([BBD82, 5.2.1]). We suppose k = F ps is finite and that F means the sth power of Frobenius. Choose

an isomorphism of the form : Q p C. Let X be a k-variety and E F-Ovhol(X/K). We suppose that, for any etale
morphism U X with U affine, the K-vector space H 0 (pU+ (+ (E)) is -mixed of weight w. Then E is -mixed of
weight w.
Corollary 3.2.4 ([BBD82, 5.3.1]). With the notation 3.2.3, if E is -mixed of weight w (resp. w), then any
subquotient of E is -mixed of weight w (resp. w).
Finally, except [BBD82, 5.4.78], the reader can check easily the other results of the chapter 5 of [BBD82] by
translating the proofs in our p-adic context.

20

References
[Abe10] T. A BE Comparison between Swan conductors and characteristic cycles, Compos. Math. 146 (2010), no. 3, p. 638
682.
[Abe14] T. A BE Rings of microdifferential operators for arithmetic D-modules, Rend. Sem. Mat. Univ. Padova 131 (2014),
no. 1, p. p.89149.
[AC13a] T. A BE et D. C ARO On beilinsons equivalence for p-adic cohomology, (2013).
[AC13b] , Theory of weights in p-adic cohomology, (2013).
[AM]

T. A BE et A. M ARMORA Product formula for p-adic epsilon factors, to appear in J. Inst. Math. Jussieu.

[BBD82] A. A. B E ILINSON , J. B ERNSTEIN et P. D ELIGNE Faisceaux pervers, Analysis and topology on singular spaces, I
(Luminy, 1981), Astrisque, vol. 100, Soc. Math. France, Paris, 1982, p. 5171.
[Ber84]

P. B ERTHELOT Cohomologie rigide et thorie de Dwork: le cas des sommes exponentielles, Astrisque (1984),
no. 119-120, p. 3, 1749, p-adic cohomology.

[Ber96]

, -modules arithmtiques. I. Oprateurs diffrentiels de niveau fini, Ann. Sci. cole Norm. Sup. (4) 29 (1996), no. 2,
p. 185272.

[Ber97]

, Altrations de varits algbriques (daprs A. J. de Jong), Astrisque (1997), no. 241, p. Exp. No. 815, 5, 273311,
Sminaire Bourbaki, Vol. 1995/96.

[Ber00]

, D-modules arithmtiques. II. Descente par Frobenius, Mm. Soc. Math. Fr. (N.S.) (2000), no. 81, p. vi+136.

[Ber02]

, Introduction la thorie arithmtique des D-modules, Astrisque (2002), no. 279, p. 180, Cohomologies padiques et applications arithmtiques, II.

[Car06]

D. C ARO Fonctions L associes aux D-modules arithmtiques. Cas des courbes, Compositio Mathematica 142
(2006), no. 01, p. 169206.

[Car07]

, Overconvergent F-isocrystals and differential overcoherence, Invent. Math. 170 (2007), no. 3, p. 507539.

[Car09]

, Overconvergent log-isocrystals and holonomy. (Log-isocristaux surconvergents et holonomie.), Compos. Math.

145 (2009), no. 6, p. 14651503.

[CM01] G. C HRISTOL et Z. M EBKHOUT Sur le thorme de lindice des quations diffrentielles p-adiques. IV, Invent.
Math. 143 (2001), no. 3, p. 629672.
[CT12]

D. C ARO et N. T SUZUKI Overholonomicity of overconvergent F-isocrystals over smooth varieties, Ann. of Math.
(2) 176 (2012), no. 2, p. 747813.

[Del80]

P. D ELIGNE La conjecture de Weil. II, Inst. Hautes tudes Sci. Publ. Math. (1980), no. 52, p. 137252.

[dJ96]

A. J. DE J ONG Smoothness, semi-stability and alterations, Inst. Hautes tudes Sci. Publ. Math. (1996), no. 83,
p. 5193.

[Elk73]

R. E LKIK Solutions dquations coefficients dans un anneau henslien, Ann. Sci. cole Norm. Sup. (4) 6 (1973),
p. 553603 (1974).

[Gro60] A. G ROTHENDIECK lments de gomtrie algbrique. I. Le langage des schmas, Inst. Hautes tudes Sci. Publ.
Math. (1960), no. 4, p. 228.
[Gro61] , lments de gomtrie algbrique. II. tude globale lmentaire de quelques classes de morphismes, Inst. Hautes
tudes Sci. Publ. Math. (1961), no. 8, p. 222.
[Gro66] , lments de gomtrie algbrique. IV. tude locale des schmas et des morphismes de schmas. III, Inst. Hautes
tudes Sci. Publ. Math. (1966), no. 28, p. 255.
[Gro67] , lments de gomtrie algbrique. IV. tude locale des schmas et des morphismes de schmas IV, Inst. Hautes
tudes Sci. Publ. Math. (1967), no. 32, p. 361.
[HTT08] R. H OTTA , K. TAKEUCHI et T. TANISAKI D-modules, perverse sheaves, and representation theory, Progress in
Mathematics, vol. 236, Birkhuser Boston, Inc., Boston, MA, 2008, Translated from the 1995 Japanese edition by
Takeuchi.
[Kas77] M. K ASHIWARA B-functions and holonomic systems. Rationality of roots of B-functions, Invent. Math. 38 (1976/77),
no. 1, p. 3353.

21

[Ked03] K. S. K EDLAYA Semistable reduction for overconvergent F-isocrystals on a curve, Math. Res. Lett. 10 (2003),
no. 2-3, p. 151159.
[Ked05] , More tale covers of affine spaces in positive characteristic, J. Algebraic Geom. 14 (2005), no. 1, p. 187192.
[Ked11] , Semistable reduction for overconvergent F-isocrystals, IV: local semistable reduction at nonmonomial valuations,
Compos. Math. 147 (2011), no. 2, p. 467523.
[KL85]

N. M. K ATZ et G. L AUMON Transformation de Fourier et majoration de sommes exponentielles, Inst. Hautes tudes
Sci. Publ. Math. (1985), no. 62, p. 361418.

[Lau83] G. L AUMON Sur la catgorie drive des D-modules filtrs, Algebraic geometry (Tokyo/Kyoto, 1982), Lecture Notes
in Math., vol. 1016, Springer, Berlin, 1983, p. 151237.
[Lau85] , Transformations canoniques et spcialisation pour les D-modules filtrs, Astrisque (1985), no. 130, p. 56129,
Differential systems and singularities (Luminy, 1983).
[Liu02]

Q. L IU Algebraic geometry and arithmetic curves, Oxford Graduate Texts in Mathematics, vol. 6, Oxford University
Press, Oxford, 2002, Translated from the French by Reinie Ern, Oxford Science Publications.

[Mal76] B. M ALGRANGE Caractrisation homologique de la dimension, Sminaire Oprateurs diffrentiels et pseudodiffrentiels, Grenoble (1975-1976).
[Mat80] H. M ATSUMURA Commutative algebra, second d., Mathematics Lecture Note Series, vol. 56, Benjamin/Cummings
Publishing Co., Inc., Reading, Mass., 1980.
[NH04]

C. N OOT-H UYGHE Transformation de Fourier des D-modules arithmtiques. I, Geometric aspects of Dwork theory.
Vol. I, II, Walter de Gruyter GmbH & Co. KG, Berlin, 2004, p. 857907.

[Vir00]

A. V IRRION Dualit locale et holonomie pour les D-modules arithmtiques, Bull. Soc. Math. France 128 (2000),
no. 1, p. 168.

Daniel Caro
Laboratoire de Mathmatiques Nicolas Oresme
Universit de Caen Campus 2
14032 Caen Cedex
France.
email: daniel.caro@unicaen.fr

22