Documente Academic
Documente Profesional
Documente Cultură
Editors:
J.-M. Morel, Cachan
F. Takens, Groningen
B. Teissier, Paris
Yukiyoshi Nakkajima Atsushi Shiho
ABC
Yukiyoshi Nakkajima Atsushi Shiho
Department of Mathematics Graduate School of Mathematical Sciences
Tokyo Denki University University of Tokyo
2-2 Nishiki-cho Kanda Chiyoda-ku 3-8-1, Komaba Meguro-ku
Tokyo 101-8457 Tokyo 153-8914
Japan Japan
nakayuki@cck.dendai.ac.jp shiho@ms.u-tokyo.ac.jp
The use of general descriptive names, registered names, trademarks, etc. in this publication does not
imply, even in the absence of a specic statement, that such names are exempt from the relevant protective
laws and regulations and therefore free for general use.
987654321
springer.com
Preface
The main goal of this book is to construct a theory of weights for the log
crystalline cohomologies of families of open smooth varieties in character-
istic p > 0. This is a p-adic analogue of the theory of the mixed Hodge
structure on the cohomologies of open smooth varieties over C developed
by Deligne in [23]. We also prove the fundamental properties of the weight-
ltered log crystalline cohomologies such as the p-adic purity, the functori-
ality, the weight-ltered base change theorem, the weight-ltered K unneth
formula, the convergence of the weight ltration, the weight-ltered Poincare
duality and the E2 -degeneration of p-adic weight spectral sequences. One can
regard some of these results as the logarithmic and weight-ltered version of
the corresponding results of Berthelot in [3] and K. Kato in [54].
Following the suggestion of one of the referees, we have decided to state
some theorems on the weight ltration and the slope ltration on the rigid
cohomology of separated schemes of nite type over a perfect eld of char-
acteristic p > 0. This is a p-adic analogue of the mixed Hodge structure on
the cohomologies of separated schemes of nite type over C developped by
Deligne in [24]. The detailed proof for them is given in another book [70] by
the rst-named author.
We have to assume that the reader is familiar with the basic premises
and properties of log schemes ([54], [55]) and (log) crystalline cohomologies
([3], [11], [54]). We hope that the ndings in this book will serve as a role
as a rst step to understanding the rich structures which p-adic cohomology
theory should have.
v
Acknowledgements
We would like to thank K. Bannai, as the genesis of this work was a discussion
with him in Azumino in 2000. Three lectures of P. Berthelot in Dwork
trimester in 2001 [8] were also very helpful for us. Most key ideas in 1.1
and 1.2 can be attributed to him. We are deeply grateful to him for his
permission to include some of his ndings in this book. We would also like
to thank to K. Sato for a useful discussion. We gave some proofs in this
book during both authors stay at Padova University and during the rst-
named authors stay at Universite de Rennes 1. Finally, we are very grateful
to B. Chiarellotto, B. Le Stum and the universities for their hospitality.
vii
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
ix
x Contents
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Introduction
Though they are vague, we have the following dreams as in [37], [22] and [25]:
(6) In some cases, they grow and they are rigid as Grothendieck said for
crystalline sheaves.
Here we assume that the family is the complement of a relative simple
normal crossing divisor on a family of smooth varieties.
Before explaining our results, we recall Delignes result on the weight l-
tration on the higher direct image of Q by a morphism from a family of open
smooth algebraic varieties with good compactications to a base scheme over
the complex number eld C ([23], [25]).
Let U be a smooth variety over C. Let X be a smooth variety over C with
a simple normal crossing divisor D such that U = X \ D. Let j : U X
(0)
be the natural open immersion. Set D := X and, for a positive integer
k, let D(k) be the disjoint union of all k-fold intersections of the dierent
irreducible components of D. Let P := {Pk }kZ be the weight ltration on
the sheaf iX/C (log D) (i N) of the logarithmic dierential forms on X which
Y. Nakkajima, A. Shiho, Weight Filtrations on Log Crystalline Cohomologies 1
of Families of Open Smooth Varieties. Lecture Notes in Mathematics 1959,
c Springer-Verlag Berlin Heidelberg 2008
2 Introduction
Here (k) (D/C) is the orientation sheaf of D(k) /C; (k) (D/C)(k) := kZ
in [23]. By using the isomorphism (0.0.0.1), we have the following spectral
sequence
(0.0.0.2)
E1k,h+k = H hk (D(k) , D(k) /C Z (k) (D/C))(k) = H h (X, X/C (log D)).
(cf. [58, (1.5.1)]). By the same calculation as that in [69, (3.3)], the following
morphism
(k)
(0.0.0.5) aan (C(Dan )(k) Z (k) (Dan /C))(k)
(0.0.0.4)C
Rk jan (CUan )
(0.0.0.3) +k
= grP
k Xan /C (log Dan )
Res
aan ((Dan )(k) /C Z (k) (Dan /C))(k)
(k)
(k)
= aan (C(Dan )(k) Z (k) (Dan /C))(k)
(0.0.0.6)
(0.0.0.4) (1)k
(k) (k)
Rk jan (ZUan ) aan ((k) (Dan /C))(k) aan ((k) (Dan /C))(k)
Introduction 3
(X, M (D)) by (X, D). Once we obtain the topos ((X, D)/S)log
crys , we can use
powerful techniques of [42] and many techniques of [3] (cf. [54]). Let O(X,D)/S
be the structure sheaf in ((X, D)/S)log and O (k) the structure sheaf
crys D /S
natural morphisms of topoi. Then one of our main results in this book is to
show the existence of the following functorial spectral sequence:
Then (0.0.0.8) is compatible with the Frobenius action. We call (0.0.0.8) the
preweight spectral sequence of (X, D)/(S, I, ). Here, as noted in [68], we use
the terminology preweight instead of the terminology weight since OS
is a sheaf of torsion modules (and hence Rh f(X,D)/S (O(X,D)/S ) is also). If
S0 is the spectrum of a perfect eld of characteristic p > 0 and if S is the
spectrum of the Witt ring Wn of nite length n > 0 of , then (0.0.0.8) is
canonically isomorphic to the following preweight spectral sequence
h
= Hlog-crys ((X, D)/Wn ),
essentially constructed in [65] and [68].
Let V be a complete discrete valuation ring of mixed characteristics with
perfect residue eld of characteristic p > 0. Then we can also construct the
spectral sequence (0.0.0.8) when S is a p-adic formal V -scheme in the sense
of [74]. In this case, we call (0.0.0.8) the p-adic weight spectral sequence of
(X, D)/S and the induced ltration on Rh f(X,D)/S (O(X,D)/S ) by (0.0.0.8)
the weight ltration on Rh f(X,D)/S (O(X,D)/S ).
Let us return to the case where (S, I, ) is a PD-scheme as above;
especially, a prime number p is locally nilpotent on S. Let OX/S be
the structure sheaf in the classical crystalline topos (X/S) crys . Let
u(X,D)/S : ((X, crys Xzar (resp. uX/S : (X/S)crys Xzar )
D)/S)log
be the natural (resp. classical) projection. Then uX/S induces a mor-
phism u crys , O
: ((X/S) ) (X zar , f 1 (OS )) of ringed topoi. Let
X/S X/S
uD(k) /S : (D
crys D
(k) /S) (k)
zar be also the classical projection. Let
(X,D)/S : ((X, crys (X/S)crys be the forgetting log morphism
D)/S)log
induced by the morphism (X, M (D)) (X, OX ) of log schemes. Then
uX/S (X,D)/S = u(X,D)/S . Let QX/S : (X/S)Rcrys (X/S) crys be a
morphism of topoi dened in [3, IV (2.1.1)]. Then we have a morphism
Rcrys , Q (OX/S )) ((X/S)
QX/S : ((X/S) crys , OX/S ) of ringed topoi.
X/S
Rcrys , Q (O
Set uX/S := uX/S QX/S : ((X/S) 1
X/S )) (Xzar , f
X/S (OS ))
as in [3].
Introduction 5
(0.0.0.9)
(Ecrys (O(X,D)/S ), P ) := (Ecrys (O(X,D)/S ), {Pk Ecrys (O(X,D)/S )}kZ )
D+ F(OX/S ),
(0.0.0.10)
(Ezar (O(X,D)/S ), P ) := (Ezar (O(X,D)/S ), {Pk Ezar (O(X,D)/S )}kZ )
D+ F(f 1 (OS ))
(0.0.0.11)
(CRcrys (O(X,D)/S ), P ) := (CRcrys (O(X,D)/S ),{Pk CRcrys (O(X,D)/S )}kZ )
D+ F(QX/S (OX/S )),
(0.0.0.12)
(Czar (O(X,D)/S ), P ) := (Czar (O(X,D)/S ), {Pk Czar (O(X,D)/S )}kZ )
D+ F(f 1 (OS )).
Here D+ F(OX/S ), D+ F(QX/S (OX/S )) and D+ F(f 1 (OS )) are the ltered de-
rived categories of the bounded below ltered OX/S -modules, QX/S (OX/S )-
modules and f 1 (OS )-modules, respectively.
The denitions of (Ecrys (O(X,D)/S ), P ) and (Ezar (O(X,D)/S ), P ) are as fol-
lows:
(Ecrys (O(X,D)/S ), P ) := (R(X,D)/S (O(X,D)/S ), ),
(Ezar (O(X,D)/S ), P ) := RuX/S (Ecrys (O(X,D)/S ), P ).
(Here denotes the canonical ltration (2.7).) Note that they are functorial
with respect to (X, D). (This is not the case for (CRcrys (O(X,D)/S ), P ).)
In a simple case we soon give the denition of (CRcrys (O(X,D)/S ), P ) in
(0.0.0.14) below. In the general case we give it in the text. The ltered
complex (Czar (O(X,D)/S ), P ) is, by denition, RuX/S (CRcrys (O(X,D)/S ), P ).
We call (Ecrys (O(X,D)/S ), P ) and call (Ezar (O(X,D)/S ), P ) the preweight-
ltered vanishing cycle crystalline complex and the preweight-ltered van-
ishing cycle zariskian complex of (X, D)/(S, I, ), respectively. We also call
(CRcrys (O(X,D)/S ), P ) the preweight-ltered restricted crystalline complex of
(X, D)/(S, I, ) and call (Czar (O(X,D)/S ), P ) the preweight-ltered zariskian
complex of (X, D)/(S, I, ), respectively. The main theme of this book is to
investigate fundamental properties of (Ecrys (O(X,D)/S ), P ), (Ezar (O(X,D)/S ),
P ), (CRcrys (O(X,D)/S ), P ) and (Czar (O(X,D)/S ), P ). They enjoy the following
properties:
6 Introduction
(0.0.0.13): (CRcrys (O(X,D)/S ), P ) QX/S (Ecrys (O(X,D)/S ), P ); {Pk Ecrys (O
(X,D)/S )}kZ is an increasing ltration on Ecrys (O(X,D)/S ) which is nite
locally on X such that P1 Ecrys (O(X,D)/S ) = 0, QX/S P0 Ecrys (O(X,D)/S )
QX/S (OX/S ) and CRcrys (O(X,D)/S ) QX/S R(X,D)/S (O(X,D)/S ).
where LX/S is the classical linearization functor for OX -modules ([3, IV 3],
[11, 6]).
(0.0.0.15): (Czar (O(X,D)/S ), P ) (Ezar (O(X,D)/S ), P ).
(Hence (Czar (O(X,D)/S ), P ) is functorial with respect to (X, D).) In particu-
lar,
(0.0.0.16): {Pk Czar (O(X,D)/S )}kZ is an increasing ltration on Czar (O
(X,D)/S ) which is nite locally on X such that P1 Czar (O(X,D)/S ) = 0,
P0 Czar (O(X,D)/S ) RuX/S (OX/S ), and Czar (O(X,D)/S ) Ru(X,D)/S
(O(X,D)/S ), and
(0.0.0.17): If (X, D) has an admissible immersion (X, D) (X , D) over S
with respect to = {D } ,
where D is the PD-envelope of the immersion X X over (S, I, ).
(k) (k)
(0.0.0.18): grPk (CRcrys (O(X,D)/S )) = QX/S acrys (OD (k) /S Z crys (D/S))
{k}, where {k} is the shift which will be dened in the Convention (1)
below (note that we do not consider the Tate twist (k) on the right hand
side of (0.0.0.18) because the functor QX/S appears on the right hand side
(In [3, IV (2.5)] Berthelot has noted that the restricted crystalline topos does
not have the functoriality in general).).
(k) (k)
k (Czar (O(X,D)/S )) = azar (RuD (k) /S (OD (k) /S ) Z zar (D/S0 ))
(0.0.0.19): grP
(k)
(k){k}, where zar (D/S0 ) is the zariskian orientation sheaf of D/S0
which will be dened in 2.2. Here, see 2.9 for the meaning of the Tate
twist (k).
(0.0.0.20): (CRcrys (O(X,D)/S ), ) (CRcrys (O(X,D)/S ), P ), where is the
canonical ltration.
Introduction 7
(0.0.0.22)
/C crystal
log
Uan , (Xan , Dan )
((X log
an , Dan ))et ([51])
((X, D)/S)log
crys
Xan , X an
(X/S) crys
jan : Uan Xan
top : (Xan , Dan )log Xan
an : ((X log
an , Dan ))et Xan
(X,D)/S : ((X,
crys (X/S)crys
D)/S)log
Rjan (Z) = Rtop (Z) ([58]),
Rtop (Z/n) = Ran (Z/n) QX/S R(X,D)/S (O(X,D)/S )
(n Z) ([72])
Xan X crys X
uX/S : (X/S) zar
Z(Xan ,Dan )log
(Z/n)(Xan ,Dan )log , (Z/n)((X
,D log O(X,D)/S
an an ))et
(n Z)
ZXan
(Z/n)Xan (n Z) OX/S
(Xan /C (log Dan ), ) (CRcrys (O(X,D)/S ), )
= (Xan /C (log Dan ), P ) = (CRcrys (O(X,D)/S ), P )
(X/C (log D), P ) (Czar (O(X,D)/S ), P )
Here (Xan , Dan )log is the real blow up of (Xan , Dan ) dened in [58] and
top is the natural morphism of topological spaces which is denoted by in
[loc. cit.], and X an is the topos dened by the local isomorphisms to Xan and
an is the natural morphism forgetting the log structure.
To construct (CRcrys (O(X,D)/S ), P ) and (Czar (O(X,D)/S ), P ), we use local
admissible immersions of (X, D) over S, which are local exact closed immer-
sions. On the other hand, in the case where S0 is the spectrum of a perfect
eld of characteristic p > 0 and where S = Spec(Wn ()), Mokrane has
used local lifts of (X, D) over S in [64] and [65] in order to construct the
ltered log de Rham-Witt complex (Wn X (log D), P ). Our guiding principle
8 Introduction
which will be proved in 2.7. The reason why we obtain the equality (0.0.0.20)
is that (CRcrys (O(X,D)/S ), P ) exists in the world of the classical restricted
crystalline topos; we can consider divided powers a[n] = an /n! (n N) in
the restricted crystalline topos and we can use a Poincare lemma in it. By
(0.0.0.23) and the Poincare lemma for R(X,D)/S (O(X,D)/S ) which will be
proved in the text, we obtain (0.0.0.13) and then an important property of
(Czar (O(X,D)/S ), P ): it is functorial, that is, for another smooth scheme X
with a relative simple normal crossing divisor D over S0 and for a morphism
g : (X, D) (X , D ) of log schemes over S0 in the sense of Fontaine-Illusie-
Introduction 9
Finally in this rough explanation of the book, we remark that the following
naive p-adic purity
(k)
Rk (X,D)/S (O(X,D)/S ) = acrys (OD(k) /S Z crys
(k)
(D/S))(k) (k N)
does not hold in general, which will be proved in Remark 2.7.11. For this
reason, we have to consider the undesirable functor QX/S .
Here the numbers under the objects above in A mean the degrees.
For a morphism f : (E , dE ) (F , dF ) of complexes of objects of A,
f {n} denotes a natural morphism (E {n}, dE {n}) (F {n}, dF {n}) in-
duced by f . A morphism f : (E , dE ) (F , dF ) in the derived category
D (A) ( = b, +, , nothing) of the complexes of objects in A naturally in-
duces a morphism f {n} : (E {n}, dE {n}) (F {n}, dF {n}) in D (A).
12 Introduction
proj. (1)
(G , dG ) (E [1], dE [1]) (F , dF ) (E [1], dE [1]) (E [1], dE [1]).
(0.0.0.24)
0 (I , dI ) (J , dJ ) (K , dK ) 0
0 (E , dE ) (F , dF ) (G , dG ) 0
commutative; in this book, we consider only the case where the ltration on
A is trivial and A is commutative; hence many facts in this section are special
cases of results in [8]. Let MF(A) be the category of ltered A-modules. Let
CF(A) be the category of ltered complexes of A-modules, and let C+ F(A),
C F(A) and Cb F(A) be the categories of the bounded below, bounded above
and bounded ltered complexes of A-modules, respectively.
Let KF(A) be the category of ltered complexes of A-modules modulo
ltered homotopies, and let K+ F(A), K F(A) and Kb F(A) be the cate-
gories of bounded below, bounded above and bounded ltered complexes
of A-modules modulo ltered homotopies, respectively. We dene the direct
image and the inverse image of an object of CF(A) or KF(A) by a morphism
of ringed topoi in an obvious way. Since MF(A) is an additive category,
K F(A) ( = +, , b, nothing) is a triangulated category. For a ltered
complex E = (E , {Ek }) CF(A) and for an integer l, we dene the shift
E l = (E , {Ek })l := (E , {E lk }) by E lk := El+k
([23, (1.1)]).
We say that a ltered complex (E , {Ek }) CF(A) is strictly exact if
Hq (E ) = Hq (Ek ) = 0 (q, k Z). It is easy to see that, if a ltered complex
(E , {Ek }) is strictly exact and (E , {Ek }) (F , {Fk }) in KF(A), then
(F , {Fk }) is also strictly exact (a special case of [78, Lemma 1.2.12, Remark
1.2.13]); we can dene the notion of the strict exactness of a ltered complex
(E , {Ek }) KF(A). We say that a ltered morphism f : (E , {Ek })
(F , {Fk }) in CF(A) is a ltered quasi-isomorphism if the mapping cone of f
is strictly exact. It is easy to check that f is a ltered quasi-isomorphism if and
only if f induces isomorphisms Hq (E ) Hq (F ) and Hq (Ek ) Hq (Fk )
for all q, k Z; we can dene the notion of ltered quasi-isomorphism in
KF(A).
Let us consider the set of morphisms (FQis) whose elements are the ltered
quasi-isomorphisms in KF(A). Then (FQis) forms a saturated multiplicative
system in the sense of [89, II (2.1.1)]: the conditions (SM1) and (SM5) are
obviously satised; (SM2) and (SM3) can be proved as in [44, I (4.2)]; (SM6)
follows from the ve lemma; it is easy to check the saturatedness (SM4).
Hence we can consider the localization of K F(A) ( = +, , b, nothing)
by the multiplicative system (FQis)|K F(A) and we obtain a derived category
D F(A) := K F(A)(FQis)|K F(A) .
Let K ((T , A)) ( = +, , b, nothing) be the category of com-
plexes of (T, A)-modules modulo homotopies for = +, , b, nothing. Let
D ((T , A)) := K ((T , A))(Qis) be its derived category. Let MF((T , A))
and C F((T , A)) be the category of ltered (T , A)-modules and that of
ltered complexes of (T , A)-modules, respectively. Let K F((T , A)) be
the category of ltered complexes of (T , A)-modules modulo ltered ho-
motopies and let D F((T , A)) be the derived category K F((T , A))(FQis)
localized by the set of the ltered quasi-isomorphisms in K F((T , A)).
For two objects (E, {El }), (F, {Fl }) MF(A), we dene
in a well-known way: HomA ((E, {El }), (F, {Fl })) is, by denition, a (T , A)-
module HomA (E, F ) with ltration
HomA ((E, {El }), (F, {Fl }))k := {f HomA (E, F ) | f (El ) Fl+k (l Z)}
(k Z).
An equality
(1.1.0.1)
HomA ((E, {El }), (F, {Fl }))k = HomMF(A) ((E, {El }), (F, {Fl })k)
(E, {El }) = ( (E, {El }), ( (E, {El }))k ).
A A A
In [8] Berthelot has called A(E, {El }) the special module of (E, {El }). A
morphism f : (E, {Ek }) A (F, {Fk })
induces a morphism
E F
and a composite morphism E/Ek1 A (F, {Fl })/( A (F, {Fl }))k1 =
proj.
Fl Fk . The following formula in [8] is easily veried but very useful:
lk
(1.1.0.2)
HomMF(A) ((E, {Ek }), (F, {Fk })) = HomA (E, F ) HomA (E/Ek1 , Fk ).
A kZ
Denition 1.1.1 ([8]). Let (J, {Jk }) be an object of MF(A). We say that
(J, {Jk }) is strictly injective if J and Jk are injective A-modules and if, for a
strict monomorphism (E, {Ek }) (F, {Fk }), the induced morphism
HomA ((F, {Fk }), (J, {Jk })) HomA ((E, {Ek }), (J, {Jk }))
is a strict epimorphism.
18 1 Preliminaries on Filtered Derived Categories and Topoi
We obviously see that Istinj (A) Iinj (A) Ias (A). We also see that
Ispinj (A) Istinj (A), which is due to Berthelot:
Proposition
1.1.2 ([8]). Let A (I, {Ik }) be an object of Ispinj (A). Then
A (I, {Ik }) Istinj (A).
a proof. Let (E, {Ek })
Proof. For the completeness of this book, we give
(F, {Fk }) be a strict monomorphism. Since I k Ik is an injective A-module,
the morphism
HomA (F, I Ik ) HomA (E, I Ik )
k k
(1.1.2.1)
HomMF(A) ((F, {Fk }), (I, {Ik })) HomMF(A) ((E, {Ek }), (I, {Ik }))
A A
The category MF(A) has enough special injectives in the following sense:
proj.
Two morphisms E I and E E/El1 Il induce a monomor-
phism E J; furthermore, two composite morphisms Ek E I and
proj.
Ek E E/El1 Il induce a monomorphism Ek Jk .
It remains to prove that the morphism (E, {Ek }) (J, {Jk }) is strict.
Set Nk := Im(E J) Jk . Then Nk is isomorphic to the kernel of a
composite morphism E J Il . This kernel is nothing but Ek by
l>k
the denition of Il (l > k). Hence the morphism (E, {Ek }) (J, {Jk }) is
strict.
Remark 1.1.6. Let (T , A) be a ringed topos which has enough points. Let
(F , {Fk }) be an object of C+ F(A). For an integer p, let (I pq , {Ikpq }; dpq )qZ0
be the Godement resolution of (F p , {Fkp }). Then the sequence
Proof. (cf. [44, I (4.6) 1)]) We may assume that E q = 0 for q < 0. Assume
that we are given (J 0 , {Jk0 }), (J 1 , {Jk1 }), . . . , (J q , {Jkq }) Ispinj (A). We con-
sider A-modules J q E q E q+1 and Jkq Ekq Ekq+1 . Using the strictness of the
morphism E q J q , we can easily check that Jkq Ekq Ekq+1 J q E q E q+1
is a monomorphism. Hence {Jkq Ekq Ekq+1 } is a ltration on J q E q E q+1 . The
natural morphism E q+1 s
(0, s) J q E q E q+1 induces a ltered mor-
phism (E q+1 , {Ekq+1 }) (J q E q E q+1 , {Jkq Ekq Ekq+1 }). It is immediate to
check that this ltered morphism is strict. Let I q+1 and Ikq+1 be injective
A-modules such that there exist the following monomorphisms of A-modules:
20 1 Preliminaries on Filtered Derived Categories and Topoi
(1.1.7.1) J q E q E q+1 I q+1 , J q E q E q+1 /Jkq Ekq Ekq+1 Ik+1
q+1
.
Set J q+1 := I q+1 Ilq+1 and Jkq+1 := I q+1 Ilq+1 . Then (J q+1 , {Jkq+1 })
lZ lk
Ispinj (A). By (1.1.0.2) and (1.1.7.1), we have a natural monomorphism
(1.1.7.2) (J q E q E q+1 , {Jkq Ekq Ekq+1 }) (J q+1 , {Jkq+1 }).
q+1
Let J q E q E q+1 IE be a monomorphism into an injective A-module.
q+1
Let IF be the push-out of the following diagram:
q+1
J q E q E q+1 IE
g q f q f q+1
K q F q F q+1 .
Proof. By the denition of the strict injectivity, the same argument as that
in the classical case works.
Proof. (1): Let (G , {Gk }) be the mapping cone of s. Since (G , {Gk })[n]
(G , {Gk }){n}, we have
Remark 1.1.11. There is a mistake in signs in the proof of [44, I (4.5)]. Let
the notations be as in [loc. cit.]. Then the homotopy operator
(k, t) : T (I ) Y I
in [loc. cit.] is not right because the boundary operators of T (I ) are twisted
by 1; strictly speaking, we have a homotopy operator T (I ) Y
T (I ){1}. Thus we have an equation v = (idI , 0) = (k, t)dZ dI (k, t)
and two formulas idI = dk kd + ts and dt = td. Anyway, ts is homotopic
to idI .
1.1 Filtered Derived Category. I 23
(2) Set I := Ias (A), Iinj (A), Istinj (A) or Ispinj (A). Let f : (T , A)
(T , A ) be a morphism of ringed topoi. Then there exists the right derived
functor
Rf : D+ F(A) D+ F(A )
of f such that Rf [(I , {Ik })] = [(f (I ), f (Ik ))] for an object (I , {Ik })
K+ F(I). Here [ ] is the localization functor. (Because we shall use the symbol
Q for an object of MF(A), we do not use Q as the localization functor. We
sometimes omit the notation [ ] for simplicity.)
(3) Let f : (T , A) (T , A ) and g : (T , A ) (T , A ) be morphisms
of ringed topoi. Then R(gf ) = Rg Rf .
Proof. (1): The rst two equalities follow from (1.1.7) and the proof of [44, I
(5.1)]. The last two equalities follow from the proof of [44, I (4.7)] and (1.1.10)
(2).
(2): (2) follows from [44, I (5.1)].
(3): (3) follows by setting I := Ias (A) in (2).
We also need to recall the cospecial module and the strictly at resolution
([8]).
For two objects (M, {Mk }), (N, {Nk }) MF(A), we dene the ltered
tensor product (M A N, {(M A N )k }) of (M, {Mk }) and (N, {Nk }) as usual:
we dene the ltration by the following: (M A N )k := Im( Mj A Nl
j+l=k
M A N ).
Following
[8], for A-modules E, El (l Z), we
set A (E, {El }) :=
E El with ltration (A (E, {El }))k := El , which Berthelot
lZ lk
has called the cospecial module of (E, {El }). Then A (E, {El }) =
(A (E, {El }), (A (E, {El }))k ) is an object of MF(A). A formula
(1.1.12.1)
HomMF(A) (A (E, {El })), (F, {Fl })) = HomA (E, F ) HomA (El , Fl )
lZ
(1.1.13.1)
Q (A) := {(Q, {Qk }) | Q and Q/Qk (k Z) are at A-modules },
(1.1.13.3)
Qsp (A) := {A (Q, {Qk }) | Q and Qk are at A-modules (k Z)}.
Proposition 1.1.16 ([8]). For a ltered A-module E = (E, {Ek }), there
exists a strict epimorphism A (Q, {Ql }) E with A (Q, {Ql }) Qsp (A).
(U
free (U, A)-module with basis (U, E) \ {0}).
The morphisms Q E and Ql El E induce an epimorphism
R E. The morphism Ql El Ek (l k) induces an epimorphism
Rk Ek .
It is trivial to check that the morphism R E is strict. Thus (1.1.16)
follows.
Proof. (1) and (2) are obvious; (3) follows by setting Q := Q (A) in (2).
(1.1.20.1)
d d d d
L3 (E 2 ) L2 (E 2 ) L1 (E 2 ) L0 (E 2 )
d d d d
L3 (E 1 ) L2 (E 1 ) L1 (E 1 ) L0 (E 1 )
d d d d
L3 (E 0 ) L2 (E 0 ) L1 (E 0 ) L0 (E 0 )
d d d d
(1) RHom.
As in [44, p. 63], we set
HomnA ((E , {Ek }), (F , {Fk })) := HomA ((E q , {Ekq }), (F q+n , {Fkq+n }))
qZ
(1.2.0.1)
H n (HomA ((E , {El }), (F ,{Fl }))k ) =
HomKF(A) ((E , {El }), (F , {Fl })[n]k).
In particular,
(1.2.0.2)
H 0 (HomA ((E , {El }), (F , {Fl }))0 ) = HomKF(A) ((E , {El }), (F , {Fl })).
Similarly we obtain a ltered complex Hom A ((E , {Ek }), (F , {Fk })) of
A-modules.
To dene the derived functor of the functor
Proof. (1): By the denition of the strict injectivity, there exist A-modules
J n and Jkn (n, k Z) satisfying the following conditions:
(i) Jkn J n , Jk1
n
Jkn ,
(ii) (I , {Ik }) (J n1 , {Jkn1 }) (J n , {Jkn }),
n n
As in [11, 7.7 Proposition, 7.8 Theorem], we shall need the following ad-
junction formula for the ltered base change theorem in 2.10.
28 1 Preliminaries on Filtered Derived Categories and Topoi
(1.2.2.1)
HomA (f (Q , {Qk }), (I , {Ik })) = HomA ((Q , {Qk }), f (I , {Ik }))
HomA ((Q , {Qk }), (J , {Jk })) HomA ((E , {Ek }), (J , {Jk })).
The last quasi-isomorphism follows from (1.2.1) (2).
As in [3, V Proposition 3.3.1], by the transitive condition, we have only to
prove that (1.2.2) holds for a morphism f : (T , A) (T , B) of ringed topoi
such that f = idT as a morphism of topoi. As in the trivial ltered case,
consider the following functor f ! :
Here we endow f (A) with the trivial ltration. The functor f ! is the right
adjoint functor of f :
(1.2.2.2)
HomA ((M, {Mk }), f ! (K, {Kk })) = HomB (f (M,{Mk }), (K, {Kk }))
((M, {Mk }) MF(A)).
(1.2.2.5)
HomB ((Qq , {Qqk }), f f ! (K , {Kk })) HomB ((Qq , {Qqk }), (J , {Jk }))
by (1.2.2.4).
Let (C , {Ck }) be the mapping cone of the morphism f f ! (K , {Kk })
(J , {Jk }). Then we have a triangle
HomB ((Q , {Qk }), f f ! (K , {Kk })) HomB ((Q , {Qk }), (J , {Jk }))
By (1.2.2.5), the ltered complex HomB ((Qq , {Qqk }), (C , {Ck })) is strictly
exact. As in [3, p. 327], by noting that (Q , {Qk }) is bounded above, one can
easily check that
HomB ((Q , {Qk }), (C , {Ck }))
is also strictly exact. Therefore we obtain
HomB ((Q , {Qk }), f f ! (K , {Kk })) HomB ((Q , {Qk }), (J , {Jk })),
ExtqA ((E , {Ek }), (F , {Fk })) := HomDF(A) ((E , {Ek }), (F , {Fk })[q]).
(1.2.3.1)
H q (RHomA ((E , {Ek }), (F , {Fk }))0 ) = ExtqA ((E , {Ek }), (F , {Fk })).
In particular,
(1.2.3.2)
H 0 (RHomA ((E , {Ek }), (F , {Fk }))0 ) = HomDF(A) ((E , {Ek }), (F , {Fk })).
Here RHomA ((E , {Ek }), (F , {Fk }))0 is an object of D+ ((Ab)), which is,
by denition, 0 (RHomA ((E , {Ek }), (F , {Fk }))) in the case (T , A) =
(1-point, Z) in (1.3.4.4) below.
Proof. Let : (F , {Fk }) (I , {Ik }) be a ltered quasi-isomorphism into
an object of K+ F(Istinj (A)). Then
ExtqA ((E , {Ek }), (F , {Fk })) = ExtqA ((E , {Ek }), (I , {Ik }))
= HomDF(A) ((E , {Ek }), (I , {Ik })[q]).
where the two vertical equalities and the lower horizontal one are obtained
from (1.1.10) (1). Hence the morphism f
t g is well-dened. Moreover
we also have a natural morphism
HomKF(A) ((E , {Ek }), (I , {Ik })[q]) HomDF(A) ((E , {Ek }), (I , {Ik })[q]).
Consequently we have
1.2 Filtered Derived Category. II.RHom and L 31
ExtqA ((E , {Ek }), (F , {Fk })) = HomDF(A) ((E , {Ek }), (I , {Ik })[q])
= HomKF(A) ((E , {Ek }), (I , {Ik })[q])
= H q (HomA ((E , {Ek }), (I , {Ik }))0 )
= H q (RHomA ((E , {Ek }), (F , {Fk }))0 ).
(2) L .
Next we dene the ltered derived tensor product. First we need the following
lemma:
Lemma 1.2.4. (1) (cf. [13, I Proposition 7]) Let E and F be A-modules. Let
E and F be A-submodules of E and F , respectively. Assume that F and
F/F are at A-modules. Then
(E A F ) (EA F ) = E A F .
is an isomorphism.
Proof. (1): The A-module F is a at A-module by the assumptions. (1)
immediately follows from the following commutative diagram with exact rows
0 E A F E A F E A (F/F ) 0
0 EA F EA F EA (F/F ) 0,
Assume that f were not a monomorphism. Let (sl )l l+m=k grl EA grm
F be a non-zero local section such that f ((sl )l ) = 0. Let l0 be the minimum
integer such that sl0
= 0. Set m0 := k l0 . Let g be the following composite
morphism
f
grl0 EA grm0 F grl EA grm F (EA F )k /(EA F )k1
l+m=k
Then we have g(sl0 ) = 0 since f ((sl )l ) = 0 and f ((sl )l>l0 ) (EA Fm0 1 )
(EA F )k .
Let s El0 A Fm0 be a local lift of sl0 . Then we have
(1.2.4.1)
t (El0 1 A F ) (EA Fm0 ) = El0 1 A Fm0
= s
by (1). Then u t El0 A Fm0 . Hence we have
induces a functor
: KF(A) KF(A) KF(A).
As in [44, II (4.1)], we need the following lemma to dene a ltered derived
functor ([8])
LA : D F(A) D F(A) D F(A).
E2pq = HII
p
HIq (G
k ) = H
p+q
((E A F )k ).
(1.2.5.1)
Im((E p1 A F q )k (E p A F q )k ) Ker((E p A F q )k (E p+1 A F q )k ).
(E p1 A F q )k (E p A F q )k (E p+1 A F q )k
(1.2.5.2)
p1
E A F
q
E
p
A F
q
p+1
E A F
q
(1.2.5.3)
(K p A F q )k (E p A F q )k (K p+1 A F q )k
0 K p A F q E p A F q K p+1 A F q 0.
We claim that, to prove (1.2.5.1), we have only to prove that the following
sequence
(1.2.5.4)
0 grk (K p A F q ) grk (E p A F q ) grk (K p+1 A F q )( 0)
is exact. Indeed, let s be a local section of the sheaf on the right hand side
of (1.2.5.1). Then, by the lower exact sequence of (1.2.5.3) and by the as-
sumptions E = E
and F q = Fq
, there exists an integer k k such that
s (K A F )k . If k = k, there is nothing to prove. If k > k, then (1.2.5.4)
p q
(1.2.5.5) 0 grl K p A grm F q grl E p A grm F q
l+m=k l+m=k
grl K p+1 A grm F q ( 0)
l+m=k
tion (a).
Next, assume that (b) holds. Since F is bounded above, E p
A F is
q q q+1
exact. We prove that (E A F )k is exact. Set L := Ker(F F )
p
by (1.2.5).
Remark 1.2.6. If (F q , Fkq ) Qst (A) (q Z) and if (a) in (1.2.5) holds, then
the proof in (1.2.5) for the case (a) is simpler: indeed, let q be a xed integer.
Since F q is strictly at, the ltered complex (E A F q , (E A F q )k ) is
strictly exact. Hence HIp (G
k ) = 0 (p Z); therefore H ((E A F )k ) = 0
p
(p Z).
This section is a complement of 1.1 and 1.2. Let the notations be as in 1.1.
First we consider a larger category than Ias (A), which is convenient in
later sections.
Let f : (T , A) (T , A ) be a morphism of ringed topoi. Consider the
following full subcategory If -acyc (A) of MF(A) (cf. [23, (1.4.5)]):
We can consider the dual notion of the above: consider the following full
subcategory Qf -acyc (A) of MF(A ):
of categories and the left derived functor Lf is calculated by the following for-
mula Lf [(P , {Pl })] = [f (P , {Pl })] ((P , {Pl }) K F(Qf -acyc (A ))).
36 1 Preliminaries on Filtered Derived Categories and Topoi
Next, we dene the gr-functor. For an integer k, there exists the following
functor
(1.3.2.1)
grk : K F(A) (E , {Ek })
grk E := Ek /Ek1
K (A)
( = +, , b, nothing),
gr
D F(A)
k
D (A)
Lf
Lf
(1.3.4.2)
gr
D F(A )
k
D (A ).
(1.3.4.3)
: K F(A) (E , {El })
E K (A) ( = +, , b, nothing)
induces a morphism
D F(A) D (A)
Lf
Lf
(1.3.4.6)
D F(A ) D (A ).
(1.4.0.1): For any object (E, {Ek }) MF(A), there exists an object (I, {Ik })
I with a strict monomorphism (E, {Ek }) (I, {Ik }).
(1.4.0.2): In any short strictly exact sequence
in MF(A) is a short strictly exact sequence with (I, {Ik }), (J, {Jk }), (K, {Kk })
I, then, for any morphism f : (T , A) (T , A ) of ringed topoi, the
sequence
is strictly exact.
Proposition 1.4.1. Let I be Ias (A), Iinj (A) or Istinj (A). Then I satises
the conditions (1.4.0.1), (1.4.0.2) and (1.4.0.3).
Proof. By (1.1.7), I satises the condition (1.4.0.1); I also satises the con-
dition (1.4.0.3). It is easy to check that the categories Ias (A) and Iinj (A)
satisfy the condition (1.4.0.2). Consider the exact sequence in (1.4.0.2) with
(I, {Ik }), (J, {Jk }) Istinj (A). Then, by the denition of Istinj (A), there
exists a splitting of the strict monomorphism (I, {Ik }) (J, {Jk }). Hence
(J, {Jk }) (I, {Ik }) (K, {Kk }). Now it is easy to see that (K, {Kk })
Istinj (A).
Remark 1.4.2. We do not know whether Ispinj (A) satises the condition
(1.4.0.2).
Let I be Ias (A), Iinj (A) or Istinj (A). Let N+ F(I) be a full subcategory
of K+ F(I) which consists of the strictly exact sequences of K+ F(I). We can
prove the following as in the classical case ([78, (1.3.4)]):
Corollary 1.4.3. The canonical functor
is an equivalence of categories.
Let Q be an additive full subcategory of MF(A). Next, let us consider the
conditions [78, Denition 1.3.2 (a), (b), (c)] in our situation:
(1.4.3.1): For any object (E, {Ek }) MF(A), there exists an object (Q, {Qk })
Q with a strict epimorphism (Q, {Qk })(E, {Ek }).
(1.4.3.2): In any short strictly exact sequence
in MF(A) is a short strictly exact sequence with (Q, {Qk }), (R, {Rk }), (S,
{Sk }) Q, then, for any morphism f : (T , A ) (T , A) of ringed topoi,
the sequence
is strictly exact.
0 EA Q EA R EA S 0
0 F A Q F A R F A S 0,
where the two horizontal sequences are exact, the left vertical morphism is a
monomorphism and the middle vertical morphism is a strict monomorphism.
By this diagram we can reduce (1.4.5) to the following:
Proposition 1.4.6. Let
be a strictly exact sequence in MF(A). Assume that (R, {Rk }) Q (A) and
that (S, {Sk }) Q (A). Then, for an object (E, {Ek }) MF(A), the induced
monomorphism EA Q EA R is strict.
Proof. For a ltered A-module (G, {Gk }), set G := kZ Gk .
Let s be a local section of (EA Q) (EA R)k . We have to prove that
s (EA Q)k .
First we claim the following weaker statement: s E A Q . Indeed, the
image of s by the following composite morphism
EA Q EA R (E/E )A R
40 1 Preliminaries on Filtered Derived Categories and Topoi
is the zero. Obviously the morphism above is equal to the following composite
morphism
EA Q (E/E )A Q(E/E )A R.
Because S is a at A-module, the morphism (E/E )A Q(E/E )A R
is a monomorphism. Hence the image of s in (E/E )A Q is the zero; thus
s E A Q. Moreover we obtain s E A Q similarly by noting that a
composite morphism
E A Q E A R E A (R/R ),
E A Q E A (Q/Q )E A (R/R ),
and by noting that S/S is a at A-module since S/S = limk S/Sk . Thus
we may assume that E = E , Q = Q , R = R . In this case, it suces to
prove that a canonical morphism
is a monomorphism.
Let T be Q or R. Then we have a natural isomorphism
fT : grl EA grm T grk (EA T )
l+m=k
Corollary 1.4.7. Let Q be Q (A) or Qst (A). Then the canonical functor
is an equivalence of categories.
1.5 The Topos Associated to a Diagram of Topoi. I 41
We quickly review the general theory of the diagram of topoi in [3, V 3.4.1]
(=the D-topos in [42, Vbis (1.2)]) (cf. [24, 5]), and we apply it to ltered
derived categories.
Let I be a small category. Assume that we are given a contravariant functor
Then we have
K+ (I(T ,A ;f ) )(Qis) = K+ (I(T ,A ;f ) )/N+ (I(T ,A ;f ) ) D+ (A ).
I(T
l
,A ;f )
:= {(Ei ,{(Ei )k })iI
| (Ei , {(Ei )k }) : ltered fi -acyclic Ai -module (i I)}.
Thus we have the following two commutative diagrams by (1.1.12) (3) and
(1.1.19) (3), respectively:
Rg
D+ F(A ) D+ F(A )
R
R
(1.5.0.8)
Rg
D+ F(A) D+ F(A ),
44 1 Preliminaries on Filtered Derived Categories and Topoi
Lg
D F(A ) D F(A )
(1.5.0.9)
L L
Lg
D F(A) D F(A ).
Let (T , A) be a ringed topos. Let {Ui0 }i0 I0 and {Uj0 }j0 J0 be two open
coverings of the nal object of T . As above, we x total orders on I0 and J0 .
Then we have two sets I and J as above. For an object i j I J, set
Uij = Ui Uj , Tij := T |Uij and Aij := A|Uij . Then we have an I J-ringed
topos, which is denoted by (TIJ , AIJ ) by abuse of notation, and we have
the ringed topos (T , A ) := (Tij , Aij )ijIJ associated to the diagram
of ringed topoi (TIJ , AIJ ). If we x an object i I, we have a ringed topos
(Ti , Ai ). Varying i I, we have an I-ringed topos, which is denoted by
(TI , AI ) by abuse of notation. Similarly we have an analogous J-ringed
topos (TJ , AJ ).
We have a morphism of ringed topoi
(1.5.0.10) : (T , A ) (T , A )I
F R 1 (F )
is an isomorphism in D+ (A).
(2) Let (F , {Fk }) be an object of K+ F(A). Then the natural morphism
(F , {Fk }) R 1 (F , {Fk })
1.6 The Topos Associated to a Diagram of Topoi. II 45
is an isomorphism in D+ F(A).
(1.5.1.1) Hi (F ) Ri 1 (F )
is an isomorphism. One can calculate the right hand side by the following
spectral sequence
In this section we apply the theory in 1.5 to (restricted) log crystalline topoi.
Let S be a ne log scheme with a PD-structure on a quasi-coherent
ideal sheaf I of OS . Assume that a power of a prime number p kills OS . Let
f : Y S be a morphism of ne log schemes such that extends to Y .
log
crystalline topos (Y /S)crys associated to Y /(S, I, ) ([54,
Then we have a log
5]). Let Y = Yi0 be an open covering, where I0 is a (not necessarily
i0 I0
nite) set. We x a total order on I0 . Let I be the category dened in 1.5.
For i = (i0 , . . . , ir ) I (ij I0 ), set Yi := Yi0 Yir (each Y i has a log
structure induced by that of Y ). If there is a morphism : i i, the open
immersion Yi Yi induces a morphism
: ((Y log
i /S)crys , OYi /S ) ((Yi /S)crys , OYi /S )
log
of ringed topoi. Then we have the ringed topos ((Y /S)crys , OY /S ) as-
log
log
crys : (Y log log
/S)crys (Y /S)crys be the morphism of topoi dened in 1.5.
log 1
Note that (crys ) (OY /S ) = OY /S ; crys
log
induces a morphism
log
crys : ((Y log log
/S)crys , OY /S ) ((Y /S)crys , OY /S )
log
of ringed topoi. The morphism crys induces a morphism of ltered derived
categories:
(1.6.0.1) log
Rcrys : D+ F(OY /S ) D+ F(OY /S ).
categories:
(1.6.0.2) log
Rcrys : D+ F(OY /S ) D+ F((OY /S )I ),
log
(1.6.0.3) Ri,crys : D+ F(OYi /S ) D+ F(OYi /S ).
We also have analogous objects for the Zariski topos Yzar : by using the
f
composite morphisms f i : Y i Y S (i I), we also have an analogous
ringed topos (Y zar , f1 (OS )) and an analogous morphism
(1.6.0.4) zar : (Y zar , f1 (OS )) (Y zar , f 1 (OS ))
1
(1.6.0.6) Rzar : D+ F(f (OS )) D+ F((f1 (OS ))I ),
1
(1.6.0.7) Ri,zar : D+ F(fi (OS )) D+ F(fi1 (OS )).
Let
(1.6.0.8) uY /S : ((Y crys , OY /S ) (Y zar , f
/S)log 1
(OS ))
(1.6.0.9) uY /S : ((Y 1
/S)crys , OY /S ) (Y zar , f (OS ))
log
1.6 The Topos Associated to a Diagram of Topoi. II 47
(1.6.0.10) uY /S : ((Y 1
/S)crys , OY /S ) (Y zar , f (OS )).
log
and
log
Rcrys
D+ F(OY /S ) D+ F(OY /S )
(1.6.0.12)
RuY /S
Ru
Y /S
1 Rzar
D+ F(f (OS )) D+ F(f1 (OS )).
Let
g
Y Y
f
f
(S, I, ) (S , I , )
be a commutative diagram of ne log schemes such that the PD-structure
extends to Y . Assume that we are given open coverings {Yi0 }i0 I0 of Y
and {Yi0 }i0 I0 of Y such that g induces a morphism gi0 : Yi0 Yi0 of
log schemes. Let Ylogcrys and Ylog crys be morphisms of (ringed) topoi dened
above for Y /(S, I, ) and Y /(S , I , ), respectively. Let Y zar and Y zar be
obvious analogous morphisms of (ringed) topoi. Then the family {gi0 }i0 I0
induces morphisms satisfying the conditions (1.5.0.5) and (1.5.0.6) for the log
crystalline topoi and the Zariski topoi of Y and Y ; we obtain the following
commutative diagrams:
log
Rgcrys
D+ F(OY /S ) D+ F(OY /S )
Rlog
(1.6.0.13) crys Y crys
log
RY
log
Rgcrys
D+ F(OY /S ) D+ F(OY /S ),
Rgzar
D+ F(f1 (OS )) D+ F(f 1
(OS ))
RY zar
R
(1.6.0.14) Y zar
Rgzar
D+ F(f 1 (OS )) D+ F(f 1 (OS )).
48 1 Preliminaries on Filtered Derived Categories and Topoi
We leave the analogous formulations of the left derived functors to the reader.
Now, assume that f is locally of nite presentation. Let (Y /S)log Rcrys be the
restricted log crystalline site of Y /S ([85, (6.2)], (cf. [3, IV Proposition 1.5.5,
Denition 1.7.1])): (Y /S)log log
Rcrys is a full subcategory of (Y /S)crys whose objects
are isomorphic to triples
(1.6.0.16) QY /S : (Y
/S)log log
Rcrys (Y /S)crys
be the natural morphism of topoi which is the log version of the morphism
in [3, IV (2.1.1)]. Then we have a morphism
(1.6.0.17) QY /S : ((Y
/S)log log
Rcrys , QY /S (OY /S )) ((Y /S)crys , OY /S )
of ringed topoi. Let j : U Y be an open immersion over S. For a
sheaf E in (Y /S)log
Rcrys and for an open log subscheme VU of U and a log
smooth scheme VU over S which contains VU as a locally closed log scheme,
log 1
set (jRcrys (E))(VU ,DVU (VU ),[ ]) := E(VU ,DVU (VU ),[ ]) . Then we have a sheaf
log 1 log 1
jRcrys (E) and we see that the functor jRcrys commutes with nite inverse
limits (in fact, it commutes with inverse limits and direct limits). Hence we
have a morphism
(1.6.0.18) log
jRcrys log (Y
: (U/S) /S)log
Rcrys Rcrys
of ringed topoi. The morphisms (1.6.0.18) and (1.6.0.19) t into the following
commutative diagrams of (ringed) topoi, respectively:
log
log jRcrys
log
(U/S) Rcrys (Y /S)Rcrys
QU/S
Q
(1.6.0.20) Y /S
log
log (Y
(U/S)
jcrys
/S)log
crys crys
1.6 The Topos Associated to a Diagram of Topoi. II 49
and
log
log , Q (O log jRcrys
((U/S)Rcrys U/S U/S )) ((Y /S)Rcrys , QY /S (OY /S ))
QU/S
Q
(1.6.0.21) Y /S
log
log , O jcrys
((Y
((U/S)crys U/S ) crys , OY /S ).
/S)log
log 1
(1.6.0.22) jRcrys ((V, DV (V), [ ])) = (V U, DV (V)|U , [ ]|U ).
(1.6.0.23) QY /S jcrys
log log
= jRcrys QU/S .
(QY /S jcrys
log
(E))(V,DV (V),[ ])
log
= (jcrys (E))(V,DV (V),[ ])
= log
(jRcrys QU/S (E))(V,DV (V),[ ]) .
is exact.
Proof. The proof is the same as that in the proof of [3, IV Proposition 2.3.2].
(1.6.1.2)
uY /S := uY /S QY /S : ((Y
/S)log log
Rcrys , QY /S (OY /S )) ((Y /S)crys , OY /S )
(1.6.1.3) (Y zar , f 1 (OS )).
50 1 Preliminaries on Filtered Derived Categories and Topoi
Proof. By using the exactness of the sequence (1.6.1.1), the proof is the same
as that of [3, IV Corollaire 2.3.5].
Proof. One can reduce (1.6.3) in the general case to (1.6.3) in the non-ltered
case. In this non-ltered case, one can prove (1.6.3) in the same way as [3, V
Corollaire 1.3.3], by using (1.6.1), (1.6.2), the analogue of [3, IV Proposition
2.2.5] and the Cech-Alexander complex computing RuY /S and RuY /S .
Let the notations be as in the beginning of this section. Using the mor-
phisms (1.6.0.18) and (1.6.0.19), we have (ringed) topoi (Y log
/S)Rcrys and
((Y
/S)
log
, Q (OY /S )), and natural morphisms of (ringed) topoi
Rcrys Y /S
log
Rcrys : (Y log log
/S)Rcrys (Y /S)Rcrys
and
log
Rcrys : ((Y log log
/S)Rcrys , QY /S (OY /S )) ((Y /S)Rcrys , QY /S (OY /S )).
(1.6.3.2) log
RRcrys : D+ F(QY /S (OY /S )) D+ F(QY /S (OY /S )).
(1.6.3.4) log
Ri,Rcrys : D+ F(QYi /S (OYi /S )) D+ F(QYi /S (OYi /S )).
log
RRcrys
D+ F(QY /S (OY /S )) D+ F(QY /S (OY /S ))
RQY /S
RQ
(1.6.3.5) Y /S
log
Rcrys
D+ F(OY /S ) D+ F(OY /S ).
and
log
RiRcrys
D+ F(QYi /S (OYi /S )) D+ F(QYi /S (OYi /S ))
RQYi /S
RQ
(1.6.3.7) Yi /S
log
Ricrys
D+ F(OYi /S ) D+ F(OYi /S ).
QY /S Rcrys QY /S .
= log
(1.6.4.1) log
RRcrys
(1.6.4.2) QY /S Rcrys
log
QY /S Rcrys
log
RQY /S QY /S
= QY /S RQY /S RRcrys
log
QY /S RRcrys
log
QY /S .
Hence it suces to prove that the two morphisms in (1.6.4.2) are isomor-
phisms. Moreover, it suces to prove that these are isomorphisms as mor-
phisms between functors on non-ltered derived categories.
For an object I in D+ (QY /S (OY /S )) consisting of asque QY /S (OY /S )-
modules, we have
QY /S RQY /S (I ) = QY /S QY /S (I ) = I ,
where the second equality follows from the direct calculation. From this, we
easily see that the second morphism in (1.6.4.2) is an isomorphism.
We prove that the rst morphism in (1.6.4.2) is also an isomorphism. Let
us say that a sheaf F of OY /S -modules (resp. OY /S -modules) is parasitic if
52 1 Preliminaries on Filtered Derived Categories and Topoi
log
Hence we are reduced to proving that Rs crys Ht (Cone(F RQY /S QY /S (F )))
is parasitic for any s, t Z. If we take a asque resolution QY /S (F ) I
of QY /S (F ), we have
(Y
/S)crys Dzar
log
log
(1.6.4.3) crys D
(Y
crys Dzar ,
/S)log
(1.6.4.4) uY /S : ((Y log 1
/S)Rcrys , QY /S (OY /S )) (Y zar , f (OS )),
(1.6.4.5) uY /S : ((Y log 1
/S)Rcrys , QY /S (OY /S )) (Y zar , f (OS )).
1.6 The Topos Associated to a Diagram of Topoi. II 53
and
log
Rcrys
D+ F(QY /S (OY /S )) D+ F(QY /S (OY /S ))
(1.6.4.7)
RuY /S
Ru
Y /S
1 Rzar
D+ F(f (OS )) D+ F(f1 (OS )).
In this section we give some results on exact closed immersions. After that,
we dene a relative simple normal crossing divisor (=:relative SNCD) and a
key notion admissible immersion of a smooth scheme with a relative SNCD.
log
be a natural morphism of topoi: QERY /S (E) for an object E (Y /S)Rcrys is
the natural restriction of E and QER
Y /S commutes with inverse limits. We also
have a morphism
(2.1.1.2)
log ER log
Y /S : ((Y /S)ERcrys , QY /S QY /S (OY /S )) ((Y /S)Rcrys , QY /S (OY /S ))
QER
of ringed topoi.
i i,i
= QER ER
Y /S QY /S F (D).
Next we prove the second fundamental exact sequence for exact closed
immersions of ne log schemes and using this, we give a local description of
exact closed immersions of ne log schemes under certain assumption.
J /J 2 (1Y /S ) 1Z/S 0
(2.1.3.1)
is exact for any OZ -module E. The question is local. Assume that the re-
striction of an element of g HomOZ ( (1Y /S ), E) to (J ) is the zero.
Let t be a section of J such that 1 + t OY . Then g(d log(1 + t)) =
g(dt/(1 + t)) = g(dt) = 0. Let : 1 (MY ) 1 (OY ) OY /J be
the natural morphism. Since MZ is the push-out of the following diagram
1 ((OY /J ) ) 1 (MY )
(OY /J ) ,
s
s1 (MZ ) MY |Y 1
mo
|
s1 (Z ) Y Y1
s
s1 (OZ ) ri OY 1 ,
HomOY (1Y /S , F) h
(h d, h d log) DerS (Y, F).
In particular,
HomOZ ( (1Y /S ), J /J 2 ) = HomOY (1Y /S , J /J 2 ) DerS (Y, J /J 2 ).
Lemma 2.1.4. Let the notations be as in (2.1.3) with Y, Z log smooth over S.
Let AnS (n N) be a log scheme whose underlying scheme is An and whose
S
log structure is the pull-back of that of S by the natural projection An S.
S
Let z be a point of Z and assume that there exists a chart (Q MS , P
MZ , Q P ) of Z S on a neighborhood of z such that is injective, such
that Coker(gp ) is torsion free and that the natural homomorphism OZ,z Z
(P gp /Qgp ) 1Z/S,z is an isomorphism. Then, on a neighborhood of z,
there exist a nonnegative integer c and the following cartesian diagram:
Z Y
(2.1.4.1)
(S Z[Q] Z[P ], P a ) (S Z[Q] Z[P ], P a ) S AcS .
2.1 Exact Closed Immersions, SNCDs and Admissible Immersions 59
Here the vertical morphisms are strict etale and the lower horizontal mor-
phism is the base change of the zero section S AcS .
Proof. Assume that Y is ane. By (2.1.3) we have the following split exact
sequence
0 J /J 2 (1Y /S ) 1Z/S 0.
(2.1.4.2)
the same argument as that in [54, p. 205], we have compatible etale morphisms
Z S Z[Q] Z[P ] and Y (S Z[Q] Z[P ]) Spec (O [xd+1 , . . . , xd+c ])
S S S
in the classical sense.
Corollary 2.1.5. Let S0 S be a closed immersion of ne log schemes.
Let Z0 (resp. Y ) be a log smooth scheme over S0 (resp. S), which can be
considered as a log scheme over S. Let : Z0 Y be an exact closed im-
mersion over S. Let z be a point of Z 0 and assume that there exists a chart
(Q MS , P MZ0 , Q P ) of Z0 S0 S on a neighborhood of
gp
z such that is injective, such that Coker( ) is torsion free and that the
natural homomorphism OZ0 ,z Z (P gp /Qgp ) 1Z/S0 ,z is an isomorphism.
Then, on a neighborhood of z, there exist a nonnegative integer c and the
following cartesian diagram
(2.1.5.1)
Z0 Y Y
(S0 Z[Q] Z[P ], P a ) (S Z[Q] Z[P ], P a ) (S Z[Q] Z[P ], P a ) S AcS ,
where the vertical morphisms are strict etale and the lower second horizontal
morphism is the base change of the zero section S AcS and Y := Y AcS S.
Proof. Set Y0 := Y S S0 and let 0 : Z0 Y0 be the closed immer-
sion induced by . Apply (2.1.4) for 0 . Then we have a cartesian diagram
(2.1.4.1) for Z0 /S0 and Y0 /S0 around any point z Z 0 . By the same ar-
gument as in the proof of (2.1.4) using the isomorphism (MY /OY )(z)
(MY0 /OY 0 )0 (z) , we see that the chart P MY0 extends to a chart
60 2 Weight Filtrations on Log Crystalline Cohomologies
Remark 2.1.6. By a similar argument to the proof of (2.1.4) and (2.1.5) and
using [54, (3.5), (3.13)], we see that the diagrams as in (2.1.4.1), (2.1.5.1)
always exist etale locally (for some Q P ) even if we drop the condition
on the existence of a nice chart which we assumed in (2.1.4), (2.1.5).
(for some positive integers s and d such that s d), where the morphism g
is etale.
By abuse of notation, we denote the log scheme (X, M (D)) by (X, D).
Set U := X \ D and let j : U X be the natural open immersion. Set
N (D) := OX j (OU ). We remark that M (D) N (D) in general; indeed,
the stalks of N (D)/OX are not even nitely generated in general (see (A.0.9)
below).
62 2 Weight Filtrations on Log Crystalline Cohomologies
Let S0 S be a closed immersion of schemes dened by a quasi-coherent
ideal sheaf I of OS . We can consider the scheme X as a scheme over S by the
closed immersion S0 S. Let (X , D)(= (X , M (D))) be a smooth scheme
with a relative SNCD over S. Let : X X be a closed immersion over S
dened by a quasi-coherent ideal sheaf of OX .
Denition 2.1.10. Let := {D } be a decomposition of D by smooth
components of D. Let : (X, D) (X , D) be an exact (closed) immersion
into a smooth scheme with a relative SNCD over S. Then we call (or a pair
(X , D)/S by abuse of terminology) an admissible (closed) immersion over S
with respect to if D admits a decomposition := {D } by smooth
components of D such that induces an isomorphism D D X X
of schemes over S0 for all . We say that is compatible with .
We sometimes denote the admissible (closed) immersion by : (X, D; )
(X , D; ).
Remark 2.1.11. If the underlying topological spaces of S 0 and S are the same
and if (X , D) is a lift of (X, D) with a decomposition of D by smooth
components of D, we obtain the decomposition of D by smooth components
of D canonically.
be an admissible immersion. Let V be an
Let : (X, D; ) (X , D; )
open subscheme of X. If we set V := X \ (X \ V ) (here X is the closure of X
in X ), the restriction of to (V, D V )
(2.1.11.1) V : (V, D V ) (V, ( D ) V)
V
(2.1.13.2)
V V
g
SpecS (OS0 [x1 , . . . , xd ]/(xd+1 , . . . , xd )) SpecS (OS [x1 , . . . , xd ]),
0
Denition 2.1.14. We call the site ((X, D)/S)log ARcrys (resp. the topos
((X, log
D)/S)ARcrys ) the admissible restricted log crystalline site (resp. admis-
sible restricted log crystalline topos) of (X, D)/(S, I, ).
Let
log log
(2.1.14.1) (X,D)/S : ((X, D)/S)ARcrys ((X, D)/S)Rcrys
QAR
(((X, D)/S)log
Rcrys , Q(X,D)/S (O(X,D)/S ))
of ringed topoi.
Proposition 2.1.15. The morphism (2.1.14.1) (resp. (2.1.14.2)) gives an
equivalence of topoi (resp. ringed topoi).
Proof. Let : (X, D) P be an exact closed immersion into a log smooth
scheme over S. Let P be an exact closed log subscheme of P locally obtained
64 2 Weight Filtrations on Log Crystalline Cohomologies
In this section we recall the log version of the linearization functor in [11, 6]
(cf. [54, (6.9)]) and the log HPD dierential operators. After that, we show
some properties of the log linearization functor for a smooth scheme with a
relative SNCD.
be a morphism of topoi dened by the following: for an object E in (Y /S)log
crys ,
jT : (Y log
crys |T (Y /S)crys
/S)log
(2.2.0.2) : ((Y
crys |T , OY /S |T ) (Tzar , OT )
/S)log
be a morphism of ringed topoi dened by the following (cf. [11, 5.26 Propo-
sition]): for an OT -module E, the sections of (E) at (T , ) is (T , (E));
for an OY /S -module E in (Y
crys |T , (E) is dened as follows: let T be
/S)log
an open log subscheme of T . Let T also denote the object (T T V
(T T T = T )) in (Y /S)log
crys . Then we have a natural morphism : T T
in (Y /S)log
crys ; the section of (E) is, by denition, (T , (E)) := E((T , )).
By the log version of the ringed topos version of [11, 5.26 Proposition], we
have the following diagram of ringed topoi
2.2 The Log Linearization Functor 65
(2.2.0.3)
((Y log uY /S
jT
crys |T , OY /S |T ) ((Y /S)crys , OY /S ) (Yzar , f
/S)log 1
(OS ))
(Tzar , OT ) (Vzar , OV ) (Vzar , f 1 (OS )|V )
(Y log
jT
crys |T (Y /S)crys
/S)log
(2.2.0.4)
uY /S
Tzar = Vzar Yzar ,
Now let us recall the log linearization functor briey (cf. [11, 6.10 Propo-
sition], [54, (6.9)]).
Let : Y Y be a closed immersion into a log smooth scheme over
S such that extends to Y. Let DY (Y) be the log PD-envelope of over
(S, I, ). Let
(2.2.1.1) : ((Y crys |DY (Y) , OY /S |DY (Y) ) (DY (Y)zar , ODY (Y) )
/S)log
(2.2.1.3)
{the category of ODY (Y) -modules and ODY (Y) -linear morphisms}
{OY /S -modules}.
crys , let DU (T S Y) be the PD-envelope of U T S Y
For (U, T, ) (Y /S)log
compatible with and and let pT : DU (T S Y) T , pY : DU (T S Y)
DY (Y) be natural morphisms. Then the sheaf L(E)(U,T,) on Tzar induced by
L(E) is given by L(E)(U,T,) = pT pY E = ODU (T S Y) ODY (Y) E.
66 2 Weight Filtrations on Log Crystalline Cohomologies
Set L (E) := ODY (Y 2 ) ODY (Y) E. Then, as in the classical crystalline case,
there is a canonical log HPD stratication
ODY (Y 2 ) ODY (Y) L (E) L (E) ODY (Y) ODY (Y 2 ) .
{the category of ODY (Y) -modules and log HPD dierential operators}
Proposition 2.2.3. For an ODY (Y) -module E, there exists a canonical iso-
morphism
L (E) L(E).
2.2 The Log Linearization Functor 67
{the category of ODY (Y) -modules and log HPD dierential operators}
(2.2.3.1)
id
ODY (T S Y) ODY (Y) E
T
ODY (T S Y) ODY (Y) ODY (Y 2 ) ODY (Y) E
idu
ODY (T S Y) ODY (Y) F,
{the category of ODY (Y) -modules and ODY (Y) -linear morphisms}
Remark 2.2.5. (cf. [3, IV Remarque 1.7.8]) The functor L is not left exact as
a functor (2.2.1.3) in general. Indeed, let be a perfect eld of characteristic
p > 0 and let Wn (n Z2 ) be the Witt ring of of length n. Set S :=
(Spec(Wn ), Wn , pWn , [ ]), Y := (Spec(), ), Y := S and E := Wn . Then,
though a sequence
pn1
0 pE E E
of Wn -modules is exact, the following sequence
pn1
0 L(pE) L(E) L(E)
0 0
0 pWn /p2 Wn .
Let Ji be the dening ideal sheaf of the closed immersion Ui Ti . Let
Y1 Y2 be an exact closed immersion which induces an exact closed im-
mersion U1 U2 . Let g : T1 T2 be an exact closed immersion of ne
log PD-schemes over S tting into the following commutative diagram
U1 U2
g
T1 T2 .
: (Y log
1 /S)crys |T1 (Y2 /S)crys |T2
log
Here we omit to write the direct images. We claim that the left vertical com-
posite morphism induces a PD-morphism (OT1 , J 1 ) (OD (T T2 T1 ) , J 12,T ).
Indeed, by the denition of , the composite morphism (OT2 , J 2 )
(OT , J ) (OD (T T2 T1 ) , J 12,T ) is a PD-morphism. Let s be a local sec-
tion of Ker(g : J2 J1 ). Then the image of s in OD (T T2 T1 ) by the right
vertical composite morphism is the zero. Hence the claim follows because
g : g (J 2 ) J 1 is surjective by the assumption. Consequently we actually
have a natural morphism
q : (U U2 U1 , D (T T2 T1 ), [ ]) (U1 , T1 , 1 )
= E((U U2 U1 , D (T T2 T1 ), [ ], q)).
OT /J OU
1 (OT2 /J2 ) 1 (OU2 ),
we have J 1 (J12 )OT = 1 (J2 J12 )OT . It is easy to see that the ideal
sheaf J12 J2 is a sub PD-ideal sheaf of J2 . Hence, by the same proof of
[11, 3.5 Lemma], the PD-structure denes a unique PD-structure 12 on
J (OT /J12 OT ). Moreover, it is easy to see that extends to OT T2 T1 . Hence
(OT /J12 OT , J (OT /J12 OT ), 12 ) is a sheaf of the universal PD-algebras of
(OT T2 T1 , J12,T ) over (OT , J , ), that is, we have (2).
to Y . Let DY (Y) be the log PD-envelope of over (S, I, ). Then the natural
morphism
is a quasi-isomorphism.
be the morphism in [75, p. 14]. Then, by the formula [75, (1.1.4.2)], (uj ) =
uj uj . Hence (sj ) = sj sj + sj 1 + 1 sj (the last formula in [75, p. 16]).
Hence the natural connection
is given by
[i ]
n
[i ] [i 1]
(2.2.7.2) (as1 1 s[i
n )
n]
= a( s1 1 sj j s[i
n (sj +1)d log tj
n]
j=1
[i ] q
+s1 1 s[i
n d) (a ODY (Y) , i1 , . . . , in N, Y/S )
n]
2.2 The Log Linearization Functor 71
(2.2.7.3)
m
m
T ( an s[n] ) = (an + (n 1)an1 )s[n1] d log t + mam s[m] d log t
n=0 n=1
(m N, an OT (0 n m)).
Hence Ker(T ) = OT . Because p is locally nilpotent on S, we may assume
that pN apN = 0 if N is suciently large. Hence we see that Coker(T ) = 0 by
the formula (2.2.7.3). Therefore we have checked that the morphism (2.2.7.1)
is a quasi-isomorphism for the case n = 1.
The rest of the proof is the same as that of [11, 6.12 Theorem].
Proof. The proof is the same as that in [11, 6.14 Theorem]: we have the
following equalities in D+ (OY /S ):
Let : Z Y be an exact closed immersion of ne log schemes over S to
which extends. Assume that there exists the following cartesian diagram
72 2 Weight Filtrations on Log Crystalline Cohomologies
Z Y
(2.2.8.2)
Y,Z
Z Y,
where Y,Z is an exact closed immersion of ne log schemes over S and the
vertical two morphisms are closed immersions. Let DZ (Z) and DY (Y) be
the log PD-envelopes of the closed immersions Z Z and Y Y over
(S, I, ), respectively. Then we have the following diagram of ringed topoi:
(2.2.8.3)
D (Z)
(Zzar , OZ ) Z (D
g
Z (Z)zar , OD Z (Z) ) Z log
((Z/S) crys |D Z (Z) , OZ/S |D Z (Z) )
Y,Z
Y,Z
PD log,loc
crys
Y ((Y
gY D (Y)
zar , OY )
(Y
(DY (Y)zar , OD Y (Y) ) /S)log
crys |D Y (Y) , OY /S |D Y (Y) )
jD (Z)
Z
log
((Z/S) crys , OZ/S )
crys
log
jD (Y)
Y
((Y
/S)log
crys , OY /S ).
Let J Z (resp. J Y ) be the PD-ideal sheaf of DZ (Z) (resp. DY (Y)). Let JY,Z
be the ideal sheaf of the closed immersion Y,Z .
Lemma 2.2.9. Assume that DZ (Z) = Z Y DY (Y). Then the diagram
g
(Zzar , OZ ) Z (D
Z (Z)zar , ODZ (Z) )
Y,Z
PD
(2.2.9.1) Y,Z
(Yzar , OY ) (D
gY
Y (Y)zar , ODY (Y) ).
is commutative for a quasi-coherent ODZ (Z) -module E, that is, the natural
morphism DY (Y) PD log,loc
Y,Z (E)crys DZ (Z) (E) is an isomorphism.
(2.2.10.3) DY (Y) PD PD
Y,Z (E)(T, ) = (T, Y,Z (E)).
log | jD (Z)
log
(Z/S) crys DZ (Z)
Z (Z/S)crys
log,loc
(2.2.11.1) crys log
crys
(Y Y (Y
jD (Y)
crys |DY (Y)
/S)log /S)log
crys .
is commutative.
Proof. Let T = (U, T, MT , ) be an object of (Y /S)log crys . Let be the PD-
structure of Ker(OT OU ) + IOT which is an extension of and .
Let D(T ) := DU Z, (T ) be the log PD-envelope of the closed immersion
U Z T over (T, MT , ). By the log version of [11, 6.2.1 Lemma],
crys (T ) = (U Z, D(T )). Hence jDZ (Z) crys (T ) = (D(T ) DZ (Z), p2,Z )
log log
log |
Here we consider the last equality as sheaves in (Z/S)crys DZ (Z) . Hence
(2.2.11.1) is commutative.
Corollary 2.2.12. Assume that DZ (Z) = Z Y DY (Y). Let LPD Z/S (resp.
LPD
Y /S ) be the linearization functor of ODZ (Z) -modules (resp. ODY (Y) -mod-
ules). Then there exists a canonical isomorphism of functors
for quasi-coherent ODZ (Z) -modules. Set LY /S := LPD Y /S gY and LZ/S :=
Z/S gZ . Then there also exists a canonical isomorphism of functors
LPD
is a log HPD dierential operator of ODY (Y) -modules and we see easily that
the diagram
induced by u (resp. u) in the way described in (2.2.3.1) and the vertical mor-
phisms are the homomorphism (2.2.12.3) for E and F. Therefore we see the
compatibility of (2.2.12.1) with log HPD dierential operators.
Remark 2.2.13. In the case where Y , Z are trivial log smooth schemes over
a trivial log scheme S, we can also prove (2.2.12) by an analogous proof of
[3, IV Proposition 3.1.7]. In the case where Y , Z are ne log (not necessarily
smooth) schemes over a ne log scheme S, we can also prove (2.2.12) by
the second fundamental exact sequence of log dierential forms on ne log
smooth schemes ((2.1.3)) and by the log version of an analogous proof of [3,
IV Proposition 3.1.7].
(2) Now let us study some properties of log linearization functors for a smooth
scheme with a relative SNCD.
Let S0 S be a closed immersion of schemes(=trivial log schemes)
dened by a quasi-coherent ideal sheaf. Let f : X S0 be a smooth
scheme with a relative SNCD D on X over S0 . Let Z be a relative SNCD
on X over S0 which intersects D transversally over S0 . Let D := {D }
(resp. Z := {Z } ) be a decomposition of D (resp. Z) by smooth compo-
nents of D (resp. Z). Then := {D , Z }, is a decomposition of D Z by
smooth components of D Z. Let (X, D Z) (X , D Z) be an admissi-
ble closed immersion over S with respect to . Let := {D , Z }, be the
decomposition of D Z which is compatible with .
Set
(2.2.13.3) D := X
Set
immersions.
As usual, we dene the preweight ltration PD on the sheaf of the log
dierential forms iX /S (log(D Z)) (i N) in Xzar with respect to D as
follows:
0 (k < 0),
Im(kX /S (log(D Z))OX ik
X /S
(log Z) iX /S (log(D Z))) (0 k i),
i
X /S
(log(D Z)) (k > i).
Now, assume that the dening ideal sheaf I of the closed immersion
S0 S is a PD-ideal sheaf with a PD-structure .
Let the right objects in the following table be the log PD-envelopes of the
left exact closed immersions over (S, I, ):
2.2 The Log Linearization Functor 77
(X, D Z) (X , D Z) DD
(X, Z) (X , Z) D
(D(k) , Z|D(k) ) (D(k) , Z|D(k) ) D(k)
Lemma 2.2.16. (1) The natural morphism (D(k) , Z|D(k) ) (D(k) , Z|D(k) )
(X ,Z) (X, Z) is an isomorphism.
(2) The natural morphism D(k) D (X ,Z) (D(k) , Z|D(k) ) is an isomor-
phism.
(k)
(3) Let J (resp. J ) be the PD-ideal sheaf of OD (resp. OD(k) ). Then
(k)
the natural morphism c(k) : c(k) (J ) J is surjective.
As usual, we denote the left objects in the following table by the right ones
for simplicity of notation:
((X, D Z) DD ) ((X, D Z)/S)log
crys DD
((X, Z) D) ((X, Z)/S)log
crys D
((D (k)
, Z|D(k) ) D (k)
) ((D (k)
, Z|D(k) )/S)log
crys D(k)
(X, D Z) (X , D Z) S (X , D Z) DD (1)
(X, Z) (X , Z) S (X , Z) D(1)
(D , Z|D(k) ) (D , Z|D(k) ) S (D , Z|D(k) ) D(k) (1)
(k) (k) (k)
Let
Z)/S)log
crys |DD ((X, D Z)/S)crys
log
jDD : ((X, D
jD : ((X, Z)/S)log |D ((X, Z)/S)log
crys crys
D : (((X, D Z)/S)log
crys |D D , O(X,DZ)/S |D D ) (Dzar , OD )
: (((X, Z)/S)log
crys |D , O(X,Z)/S |D ) (Dzar , OD )
(k) : (((D(k) ,
Z|D(k) )/S)log
crys |D (k) , O(D (k) ,Z|
)/S |D (k) ) (D
(k)
zar , OD (k) )
D (k)
g : (Dzar , OD ) (X zar , OX )
g (k) : (D(k) zar , OD ) (D zar , OD (k) )
(k)
be natural morphisms.
For an OX -module E, set
L(X,DZ)/S (E) := jDD D g (E) ((X, D Z)/S)log
crys
and
L(X,Z)/S (E) := jD g (E) ((X, Z)/S)log
crys .
(2.2.16.1) O(X,DZ)/S L(X,DZ)/S (X /S (log(D Z))).
(2.2.16.3) O(D(k) ,Z| )/S L(k) (D(k) /S (log Z|D(k) )).
D (k)
is injective.
(2) The natural morphism
(2.2.17.2)
Q(X,Z)/S PkD L(X,Z)/S (X /S (log(D Z)))
Q(X,Z)/S L(X,Z)/S (X /S (log(D Z)))
is injective.
Proof. (1): The question is local. We may have cartesian diagrams (2.1.13.1)
and (2.1.13.2) for SNCD DZ on X ; we assume that D (resp. Z) is dened by
an equation x1 xt = 0 (resp. xt+1 xs = 0). Set J := (xd+1 , . . . , xd )OX ,
X := SpecX (OX /J ) and X := SpecS (OS [xd+1 , . . . , xd ]). Then X is
smooth over S. Let D (resp. Z ) be a closed subscheme of X dened by
an equation x1 xt = 0 (resp. xt+1 xs = 0). Because p is locally nilpo-
tent on S, we may assume that there exists a positive integer N such that
J N OD = 0. Since X is smooth over S, there exists a section of the surjection
OX /J N OX . Hence, as in [11, 3.32 Proposition], we have a morphism
OX [xd+1 , . . . , xd ] OX /J N
80 2 Weight Filtrations on Log Crystalline Cohomologies
and
OD OX PkD X /S (log(D Z)) s(PkD X /S (log(D Z ))OS
OS xd+1 , . . . , xd OX X /S ).
k
0 k1 zar (D/S0 ) := ZE
D Dk1
0
the zariskian orientation sheaf (resp. crystalline orientation sheaf , log crystal-
line orientation sheaf) of D(k)/S0 (resp. D(k) /(S, I, ), (D(k), Z|D(k) )/(S, I, )).
(k) (k) (k)log
Remark 2.2.19. The sheaves zar (D/S0 ), crys (D/S) and crys (D/S; Z)
are dened by the local nature of D; they are independent of the choice of
the decomposition by smooth components of D.
(k)log
Proof. (2.2.20) immediately follows from the denition of crys (D/S; Z).
82 2 Weight Filtrations on Log Crystalline Cohomologies
(2.2.21.1)
D
0 OD OX Pk1 X /S (log(D Z)) OD OX PkD X /S (log(D Z))
(2.2.21.2)
Q PD
grk (X,Z)/S Q(X,Z)/S L(X,Z)/S (X /S (log(D Z)))
Q(X,Z)/S acrys L(k) (D(k) /S (log Z|D(k) ) Z zar
(k)log (k)
(D/S)){k}
Q(X,Z)/S acrys (O(D(k) ,Z|
(k)log
)/S Z crys
(k)log
(D/S; Z)){k}.
D (k)
Proof. (1): By the Poincare residue isomorphism with respect to D (cf. [21,
3.6]), we have the following isomorphism
(2.2.21.3)
D
ResD : grP
k X /S (log(D Z))
b (D(k) /S (log Z|D(k) ) Z zar
(k) (k)
(D/S){k}).
Q D
(X,Z)/S L(X,Z)/S (Res )
Q(X,Z)/S L(X,Z)/S (b (D(k) /S (log Z|D(k) )
(k)
=
Z zar
(k)
(D/S))){k}.
which is equal to
by (2.2.20).
By (2.2.7) we obtain the second quasi-isomorphism in (2.2.21.2).
2.2 The Log Linearization Functor 83
(2.2.22.2)
Ru(X,Z)/S (Q(X,Z)/S L(X,Z)/S (X /S (log(D Z))), Q(X,Z)/S P D )
= (OD OX X /S (log(D Z)), P D )
by (2.2.0.4). Hence
0 1 + IT NTinv,gp N gp |U 0.
log structure.
2.3 Forgetting Log Morphisms and Vanishing Cycle Sheaves 85
(G)((U, T, NT , , )) = Hom((Y,M
)/S)log
( (T ), G).
crys
Denition 2.3.2. We call the morphism (Y,M,N )/S in (2.3.0.1) and the mor-
phism (Y,M,N )/S in (2.3.0.2) the forgetting log morphism of log schemes
over S along M \ N and the forgetting log morphism of log crystalline topoi
along M \ N , respectively. When N is trivial, we call the two (Y,M,N )/S s
the forgetting log morphisms of Y /S. When Y is a smooth scheme X over
S0 := SpecS (OS /I), M = M (D Z) and N = M (Z), where D and Z are
transversal relative SNCDs on X/S0 , we call the two (Y,M,N )/S s the forget-
ting log morphisms along D and denote them by (X,DZ,Z)/S .
(2.3.2.1) : ((Y
, M )/S)log
crys ((Y , N )/S)crys ,
log
((Y
crys ((Y , N )/S)crys
, M )/S)log log
log
(2.3.2.2) crys N crys
log
M
((Y, log
crys ((Y, N )/S)crys .
M )/S)log
log log
Here M crys and N crys are morphisms of topoi dened in 1.6; we have
written the symbols M and N in subscripts for clarity. Let u(Y,L)/S , u(Y ,L )/S
and u(Y ,L )/S (L := M, N ) be the projections in (1.6.0.8), (1.6.0.9) and
(1.6.0.10) for (Y, L), respectively. Since u(Y,N )/S = u(Y,M )/S and
u(Y ,N )/S = u(Y ,M )/S , we have the following two equations
as morphisms of topoi.
Let the notations be as in 1.6. Then we have the following commutative
diagram:
86 2 Weight Filtrations on Log Crystalline Cohomologies
((Y
crys ((Y , N )/S)crys
, M )/S)log log
log
(2.3.2.4) crys N crys
log
M
((Y
crys ((Y , N )/S)crys .
, M )/S)log log
(((Y, (((Y,
crys , O(Y,M )/S )
M )/S)log crys , O(Y,N )/S )),
N )/S)log
(2.3.2.8)
(((Y
crys , O(Y ,M )/S ) (((Y , N )/S)crys , O(Y ,N )/S )
, M )/S)log log
log
crys N crys
log
M
(((Y (((Y
crys , O(Y ,M )/S )
, M )/S)log crys , O(Y ,N )/S ).
, N )/S)log
(2.3.2.12)
|F : (((Y, log
crys |FM , O(Y,M )/S |FM ) (((Y, N )/S)crys |FN , O(Y,N )/S |FN ).
M )/S)log
of ringed topoi.
Lemma 2.3.3. Let the notations be as above. Then the functor |F is exact.
crys |FN . Let : T
Proof. Let (U, T, NT , , , N ) be an object in ((Y, N )/S)log
TF be the underlying morphism of schemes of N . Set MT := (MF ). Let
M : (U, T, MT , , ) (UF , TF , MF , F , F )
Indeed, let U : U UF be the open immersion. Then we have the following:
1
U (M |UF /OUF ) 1 (N |U )
/OU
N |U /OU
U F F
= M |U /OU (MF )/OT .
(2.3.3.2)
((T, NT , N ), (|F ) (E)) = Hom((Y,M
)/S)log
((|F ) ((T, NT , N )), E)
crys |FM
= E((T, MT , M )).
Lemma 2.3.4. Let the notations be as above. Then the following diagram of
topoi is commutative:
((Y, log
jF
crys |FM ((Y, M )/S)crys
M )/S)log M
|F
(2.3.4.1)
((Y, log
jFN
crys |FN ((Y, N )/S)crys .
N )/S)log
88 2 Weight Filtrations on Log Crystalline Cohomologies
canonical morphism
Though is not exact in general (see (2.7.1) below), the following holds:
Corollary 2.3.6. Let : (Y, M ) (Y, M) be a closed immersion into a log
smooth scheme over S. Let DY (Y) be the log PD-envelope of over (S, I, ).
(Y,M )/S (E) be the linearization of E with
Let E be an ODY (Y) -module. Let LPD
respect to . Then the canonical morphism
gM
DM (Y, M)
(2.3.7.2) h
gN
DN (Y, N )
Assume that the underlying morphism h of schemes is the identity. Then
there exist natural isomorphisms
(Y,N )/S (Y,M,N )/S L(Y,M )/S
LPD PD
(2.3.7.3)
and
(Y,N )/S g N (Y,M,N )/S L(Y,M )/S g M
LPD PD
(2.3.7.4)
|D : (((Y,
M )/S)log log
crys |D M , O(Y,M )/S |D M ) (((Y, N )/S)crys |D N , O(Y,N )/S |D N )
be the natural morphism. Then, using the formula (2.3.3.2), we can immedi-
ately check that (|D ) M = N . Hence we have the following commutative
diagram
g
(Yzar , OY ) M (DMzar , ODM ) M
(2.3.7.5)
g
(Yzar , OY ) N (DN zar , ODN ) N
(((Y, log
M
jD
crys |DM , O(Y,M )/S |DM ) (((Y, M )/S)crys , O(Y,M )/S )
M )/S)log
|D (Y,M,N )/S
(((Y, log
jDN
crys |DN , O(Y,N )/S |DN ) (((Y, N )/S)crys O(Y,N )/S ),
N )/S)log
and this implies the isomorphisms (2.3.7.3), (2.3.7.4).
Finally we check the functoriality of the isomorphism (2.3.7.3) with re-
spect to log HPD dierential operators. To show this, it suces to prove the
required functoriality for the morphism
and it is easy to see that this homomorphism is functorial with respect to log
HPD dierential operators (see (2.2.3.1)). Hence we nish the proof of the
lemma.
Denition 2.3.9. For an OY /S -module E, we call R(Y,M,N )/S (E) the van-
ishing cycle sheaf of E along M \ N . We call R(Y,M,N )/S (O(Y,M )/S ) the
vanishing cycle sheaf of (Y, M )/(S, I, ) along M \ N . If N is trivial, we omit
the word along M \ N .
(2.3.10.1)
R(Y,M,N )/S (Y,M,N )/S (E) LPD
(Y,N )/S (E OY Y/S (log M/MS ))
in D+ (O(Y,N )/S ).
Applying R(Y,M,N )/S to both hands of (2.3.10.2) and using (2.3.6) and
(2.3.7), we obtain
2.3 Forgetting Log Morphisms and Vanishing Cycle Sheaves 91
T
XU
g
U T.
Let D be the log PD-envelope of and let h : D T be the composite mor-
phism D T T . Then we have Ri fXU /T (E) = Hi (h (E(XU ,D) OT
T /T )) = 0 for i > a + b. Hence we have proved the claim and consequently
we nish the proof of (2.3.11).
Remark 2.3.13. In the proof of (2.3.11), we used the convention that the log
structures in this book are dened on a Zariski site. However, if we assume
that f is log smooth, we can prove the statement of (2.3.11) also in the case
where the log structures are dened on an etale site. Indeed, in this case, by
(2.3.14) below, if we assume that X and Y are ane, then we have always a
log smooth lift g : T T of XU U for any (U, T, ) (Y /S)log
crys such
that T is ane. Then we have
We give a proof of a lemma which has been used in (2.3.13), which is useful
also in later sections.
such that Z = Z S S0 for a large (cf. [40, 3 (8.8.2) (ii)], [40, 4 (17.7.8)],
[86, 4.11]). By [40, 3 (8.10.5) (viii)], we may assume that Z is ane. Since
I is nilpotent, the existence of Z follows from [54, (3.14) (1)].
Let Z be another lift of Z over S. Since the structural morphism Z S0
is quasi-separated, the structural morphisms Z S and Z S are quasi-
separated by [40, 1 (1.2.5)]. Set Z := Z S S and Z := Z S S . Then
Z and Z are quasi-compact, quasi-separated and of nite presentation over
S . Because lim Z = Z = lim Z , there exists an isomorphism Z Z
over S for a large which induces the identity of Z (cf. [40, 3 (8.8.2) (i)],
[86, 4.11.3]). Since I is nilpotent, there exists an isomorphism Z Z over
S which induces the isomorphism Z Z ([54, (3.14) (1)]).
The rest we have to prove is that Z is ane. Let Z be the ane ne
log scheme above. Because I is nilpotent, we may assume that I2 = 0.
Let J be a coherent ideal sheaf of OZ . By the proof in [45, III (3.7)] of
Serres theorem on the criterion of the aneness of a scheme, we have only
to prove that H 1 (Z, J ) = 0 (the assumption noetherianness in [loc. cit.] is
unnecessary). Consider the following exact sequence
0 I J J J /I J 0.
Because Z is ane, H 1 (Z, J /I J ) = H 1 (Z , J /I J ) = 0. Similarly,
H 1 (Z, I J ) = 0. Hence H 1 (Z, J ) = 0. Hence we nish the proof.
Assume that, for any element i0 of I0 , there exists a smooth scheme Xi0
with a relative SNCD Di0 on Xi0 over S such that there exists an admissible
immersion
(Xi0 , Di0 ) (Xi0 , Di0 )
with respect to i0 := {D(;i0 ) } . By (2.3.14), if {Xi0 }i0 I0 is an ane
open covering of X, we can assume that (Xi0 , Di0 ) is, in fact, a lift of
(Xi0 , Di0 ): (Xi0 , Di0 ) S S0 = (Xi0 , Di0 ).
We wish to construct the following object:
(2.4.0.2): A diagram (X , D ) (X , D ) ( I) of admissible immersions
into a diagram of smooth schemes with relative SNCDs over S with respect
to , where i := {D(;i) }Xi (i I).
i = {D(;i ) }
Let be a decomposition of Di0 which is compatible
0
0 Xi
0
with i0 : Di0 = X D(;i0 ) and D(;i0 ) Xi0 Xi0 = D(;i0 ) ( Xi0 ).
i0
(i ) (i i0 ) (i0 )
(xt ut xt | 1 r, 1 t s),
(i ) (i ) (i i ) (i i )
where (0 r) and u1 0 , . . . , us 0 (1 r) are
x1 , . . . , xdi
independent variables over OS and s is a positive integer. The exceptional
(i ) (i )
divisor Di is dened by an equation x1 0 xs 0 = 0.
Proof. The problem is etale local. We may assume that there exists an isomor-
(i ) (i )
phism Xi SpecS (OS [x1 , . . . , xdi ]). Assume, furthermore, that D(i ,i)
(i ) (i )
is dened by an equation x1 xs = 0 (1 s min{di | 0 r}).
Here a positive integer s is independent of .
(i ) (i )
Set A := OS [x1 , . . . , xdi | 0 r]. Let It A (1 t s) be the
ideal sheaf of a closed subscheme
(i ) (i ) (i )
(xt 0 = 0) (xt 1 = 0) (xt r = 0).
are equivalent to
(i ) (i )
ut
= 0, xt
= 0 (1 t s, 0 r),
OXi is isomorphic to
(i ) (i ) (i i )1
OS [x1 , . . . , xdi | 0 r][ut | 0
= r, 1 t s]/
(i ) (i i ) (i ) (i i ) (i i ) (i i ) (i i ) (i i )
(xt ut xt , ut ut 1, ut ut ut
| 0
=
=
= r, 1 t s).
The last sheaf of rings is isomorphic to
(i ) (i ) (i i0 )1
OS [x1 , . . . , xdi | 0 r][ut | 1 r, 1 t s]/
(i ) (i i0 ) (i0 )
(xt ut xt | 1 r, 1 t s).
Now the claim on the exceptional divisor is obvious.
We think that the reader is ready to read the following theorem which
tells us that (Xi , Di ) (Xi , Di ) is, indeed, an admissible immersion with
respect to i .
( )
Here u (i ,i) (i0 ,i) s are independent variables. The exact locally closed im-
mersion (Xi , Di ) (Xi , Di ) is an admissible immersion with respect to
{D(;i) }i .
of the admissible immersion (Xi , Di ) (Xi , Di ) with respect to i
(0 r).
Set
are equivalent to
((i ,i) ) ((i ,i) )
u
= 0, x
= 0 (0 r),
PkDi L(Xi ,Zi )/S (Xi /S (log(Di Zi ))) := L(Xi ,Zi )/S (PkDi Xi /S (log(Di Zi ))),
(Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log(Di Zi ))), Q(Xi ,Zi )/S P Di )
C+ F(O(Xi ,Zi )/S ),
: 1 (Q(Xi ,Zi )/S PkDi L(Xi ,Zi )/S (Xi /S (log(Di Zi ))))
Q(Xj ,Zj )/S Pk j L(Xj ,Zj )/S (Xj /S (log(Dj Zj ))),
D
(Q(Xi ,Zi )/S PkDi L(Xi ,Zi )/S (Xi /S (log(Di Zi ))))iI
C+ (Q(X ,Z )/S (O(X ,Z )/S ))
(Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log(Di Zi ))), Q(Xi ,Zi )/S P Di )iI
and
log,Z
(Czar (O(X ,D Z )/S ), P D ) := (ODi OXi Xi /S (log Di ), P Di )iI .
Remark 2.4.5. Once we are given the data (2.4.0.1) and (2.4.0.2) for (X, D
Z) with respect to = {D , Z }, , we can obtain two ltered complexes
log,Z
(CRcrys log,Z
(O(X ,D Z )/S ), P D ) and (Czar (O(X ,D Z )/S ), P D ).
Let
(2.4.5.1) log
(X,Z)/SRcrys : (((X
, Z )/S)log
Rcrys , Q(X ,Z )/S (O(X ,Z )/S ))
100 2 Weight Filtrations on Log Crystalline Cohomologies
(((X, Z)/S)log
Rcrys , Q(X,Z)/S (O(X,Z)/S ))
and
(2.4.6.2) log,Z
Rzar (Czar (O(X ,D Z )/S ), P D ) D+ F(f 1 (OS ))
(2.4.6.3) (X,DZ,Z)/S : (((X, D Z)/S)log
crys ,O(X,DZ)/S )
(((X, crys , O(X,Z)/S )
Z)/S)log
(2.4.7.2)
log,Z
Ru(X,Z)/S (CRcrys (O(X,DZ)/S )) Ru(X,DZ)/S (O(X,DZ)/S ),
2.4 Preweight-Filtered Restricted Crystalline and Zariskian Complexes 101
(2.4.7.3) Ru(X,DZ)/S (O(X,DZ)/S ) Czar
log,Z
(O(X,DZ)/S ).
Proof. Let
(2.4.7.4) log
(X,DZ)/Scrys
: ((X , D log
Z )/S)crys ((X, D Z)/S)crys
log
and
log
(2.4.7.5) (X,DZ)/SRcrys
: ((X , D log log
Z )/S)Rcrys ((X, D Z)/S)Rcrys
=Q(X,Z)/S R(X,Z)/Scrys
log
R(X ,D Z ,Z )/S (O(X ,D Z )/S )
=Q(X,Z)/S R(X,Z)/Scrys
log
L(X ,Z )/S (X /S (log(D Z )))
log
=R(X,Z)/SRcrys Q(X ,Z )/S L(X ,Z )/S (X /S (log(D Z )))
log,Z
=CRcrys (O(X,DZ)/S ).
(2.4.7.3) is a special case of [46, (2.20)], which follows from the cohomo-
logical descent.
and
log,Z
(Czar (O(X,DZ)/S ), P D ) D+ F(f 1 (OS ))
are independent of the data (2.4.0.1) and (2.4.0.2) for (X, D Z) if we x a
decomposition of D and Z by their smooth components, and then, in 2.7,
we shall prove that they are independent of the choice of the decompositions
of D and Z by their smooth components. Once we know that the denitions
log,Z
of (CRcrys log,Z
(O(X,DZ)/S ), P D ) and (Czar (O(X,DZ)/S ), P D ) are well-dened,
we know that
102 2 Weight Filtrations on Log Crystalline Cohomologies
(2.4.8.1)
log,Z
Ru(X,Z)/S (CRcrys log,Z
(O(X,DZ)/S ), P D ) = (Czar (O(X,DZ)/S ), P D )
in (2.4.6.2) are independent of the data (2.4.0.1) and (2.4.0.2). To prove this
independence, we need not make local explicit calculations of PD-envelopes;
the notion of the admissible immersion enables us to use the classical crys-
talline Poincare lemma implicitly; see (2.5.1), (2.5.2) and (2.5.3) below for
the detail.
Let S0 S be a PD-closed immersion dened by a quasi-coherent ideal
sheaf I. Let (X, DZ), D , Z and be as in the previous section. Consider
the following commutative diagram
(X, D Z) (X1 , D1 Z1 )
(X, D Z) (X2 , D2 Z2 ),
where the horizontal morphisms above are admissible immersions with re-
spect to a decomposition ; assume that the horizontal morphisms in-
duce admissible immersions (X, D) (Xi , Di ) with respect to D and
(X, Z) (Xi , Zi ) with respect to Z (i = 1, 2). Let Di (i = 1, 2) be the
log PD-envelope of the admissible immersion (X, Z) (Xi , Zi ). Then the
following holds:
Lemma 2.5.1. The induced morphisms
Q PD
Then gk := grk (X,Z)/S {(2.5.1.1)} ts into the following commutative dia-
gram:
Remark 2.5.2. To compare our straight method with previous works, assume
that Z = and consider two admissible immersions (X, D) (Xi , Di )
(i = 1, 2) with respect to a decomposition = {D } of D by smooth
components of D. As in 2.4, we make the following operation. Set X12 :=
X1 S X2 . Let Di = D(;i) (i = 1, 2) be the union of smooth components
of Di . Blow up X12 along (D(;1) S D(;2) ). Let X12 be the complement
of the strict transform of
{(D(;1) S X2 ) (X1 S D(;2) )}
104 2 Weight Filtrations on Log Crystalline Cohomologies
in this blow up. Let D12 be the exceptional divisor on X12 . By consider-
ing the strict transform of X in X12 , we have an admissible immersion
(X, D) (X12 , D12 ) with respect to , and we have the following com-
mutative diagram:
(X, D) (X12 , D12 )
(2.5.2.1)
(X, D) (Xi , Di ),
and
which are ltered quasi-isomorphisms by (2.5.1). Thus the proof for (2.5.2.3)
gives a simpler proof of a ltered version of the last lemma in [47] (cf. [48,
(1.7)], [64, 3.4]). Because we allow not only local lifts of (X, D) but also
local admissible immersions in the constructions of (CRcrys (O(X,D)/S ), P ) and
(Czar (O(X,D)/S ), P ), we can use the Poincare lemma implicitly for the proof
of the quasi-isomorphism (2.5.2.3). We can also use a complicated version of
[64, 3.4] to prove that (2.5.2.3) is a ltered quasi-isomorphism; however we
omit this proof because this proof is lengthy.
log,Z
Next we prove that (CRcrys log,Z
(O(X,DZ)/S ), P D ) and (Czar (O(X,DZ)/S ),
P ) are independent of the data (2.4.0.1) and (2.4.0.2) for D Z and .
D
Let the notations be as in 2.4. Let {Xi0 }i0 I0 and {Xj0 }j0 J0 be two open
coverings of X, where I0 and J0 are two sets. Let I and J be two sets in 1.5.
By 1.6 we have a diagram of ringed topoi (((X crys , O(X ,Z )/S )
, Z )/S)log
1
and (Xzar , f (OS )).
Let i and j be arbitrary elements of I and J, respectively. For simplic-
ity of notation, set E := D Z. Let {D } and {Z } be decomposi-
tions of D and Z by smooth components of D and Z, respectively. Set
:= {E } := {D , Z }, . Then is a decomposition of E by smooth
components of E. Assume that there exist two diagrams of admissible im-
i )iI and (Xj , Ej ; |X )jJ
mersions (Xi , Ei ; |Xi )iI (Xi , Ei ; j
j )jJ over S. Set Xij := Xi Xj and Eij := Ei Ej . Let X(i,ij) :=
(Xj , Ej ;
2.5 Well-Denedness of the Preweight-Filtered Complexes 105
Xi\(X i \ Xij ) (resp. X(j,ij) := Xj \ (X j \ Xij )) and set Xij := X(i,ij) S X(j,ij) .
i
Then we have a locally closed immersion Xij Xij . Set {E(;i) } :=
and {E(;j) } := j . Set also E(i,ij) := Ei X(i,ij) , E(j,ij) := Ej X(j,ij) ,
E(;i,ij) := E(;i) X(i,ij) and E(;j,ij) := E(;j) X(j,ij) . Blow up Xij along
(E(;i,ij) S E(;j,ij) ). Let Xij be the resulting scheme. Let Xij be the com-
in Xij . Let Eij be the exceptional divisor on Xij . Then Eij is a relative SNCD
on Xij by (2.4.2). Considering the strict transform of the image of Xij in
Xij , we have a locally closed immersion Xij Xij , in fact, an admissible
immersion (Xij , Eij ) (Xij , Eij ) by (2.4.2). Let {E(;ij) } be the resulting
decomposition of Eij by smooth components of Eij . We also have a relative
SNCD Zij on Xij /S by using Z instead of E. Let Dij be the log PD-envelope
of the locally closed immersion (Xij , Zij ) (Xij , Zij ).
Let
log
RRcrys : D+ F(Q(X ,Z )/S (O(X ,Z )/S )) D+ F((Q(X ,Z )/S (O(X ,Z )/S ))I )
and
log
Ri,Rcrys : D+ F(Q(Xi ,Zi )/S (O(Xi ,Zi )/S )) D+ F(Q(Xi ,Zi )/S (O(Xi ,Zi )/S ))
Theorem 2.5.3.
(2.5.3.1)
log
RRcrys (Q(X ,Z )/S L(X ,Z )/S (X /S (log E )), Q(X ,Z )/S P D )
=(Q(X ,Z )/S L(X ,Z )/S (X /S (log E )), Q(X ,Z )/S P D )I .
Proof. Because (2.5.3.2) follows from (2.5.3.1) by (2.2.22) and by the com-
mutative diagram (1.6.4.7), we have only to prove (2.5.3.1).
106 2 Weight Filtrations on Log Crystalline Cohomologies
Dj
Ker{ j0 Rcrys (Q(X
j0 ,Zj0 )/S
L(Xj ,Zj )/S (Xj /S (log Ej0 )), Q(Xj ,Zj )/S P 0 )
0 0 0 0 0
j0
j0 j1 Rcrys (Q(X L(Xj ,Zj j )/S (Xj j /S (log Ej0 j1 )),
j0 j1 ,Zj0 j1 )/S 0 j1 0 1 0 1
j0 <j1
Dj j
Q(X P 0 1 )} (j0 , j1 J0 ).
j0 j1 ,Zj0 j1 )/S
For i I, let
ei : (((Xi
, Zi )/S)log log
Rcrys , O(Xi ,Zi )/S ) (((X , Z )/S)Rcrys , O(X ,Z )/S )
be a morphism dened in 1.5. Let (I , {(I )k }) be a ltered asque reso-
lution of (Q(X ,Z )/S L(X ,Z )/S (X /S (log E )), Q(X ,Z )/S P D ) such
that, for each i, (Ii , {(Ii )k }) is a ltered asque resolution of
(Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log Ei )), Q(Xi ,Zi )/S P Di ).
(2.5.3.3) (Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log Ei )),Q(Xi ,Zi )/S P Di )
log
i,Rcrys (Ii , {(Ii )k })
(2.5.3.4) log,1
i,Rcrys (Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log Ei )), Q(Xi ,Zi )/S P Di ) =
(Q(Xi ,Zi )/S L(Xi ,Zi )/S (X(i,i) /S (log E(i,i) )), Q(Xi ,Zi )/S P Di ).
Next, we construct two morphisms (2.5.3.5) and (2.5.3.6) below (cf. [47],
[48, (1.7)], [64, 3.4]). Blow up Xi S X(i,i) along (E(;i) S E(;i,i) ). Let
Wi be the complement of the strict transform of
2.5 Well-Denedness of the Preweight-Filtered Complexes 107
((E(;i) S X(i,i) ) (Xi S E(;i,i) ))
(2.5.3.5) (Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log Ei )), Q(Xi ,Zi )/S P Di )
(Q(Xi ,Zi )/S L(Xi ,Zi )/S (Wi /S (log Fi )), Q(Xi ,Zi )/S P Di )
and
(2.5.3.6)
(Q(Xi ,Zi )/S L(Xi ,Zi )/S (X(i,i) /S (log E(i,i) )), Q(Xi ,Zi )/S P Di )
(Q(Xi ,Zi )/S L(Xi ,Zi )/S (Wi /S (log Fi )), Q(Xi ,Zi )/S P Di ).
Because there exists the following commutative diagram
(Xij , Eij ) (Wij , Fij )
(Xij , Eij ) (Xij , Eij )
such that the horizontal arrows are admissible immersions, we see that
(2.5.3.5) is a ltered quasi-isomorphism by (2.5.1). By the same proof, we
see that (2.5.3.6) is a ltered quasi-isomorphism.
Now we can prove that (2.5.3.3) is a ltered quasi-isomorphism. Indeed,
let (Ji , {(Ji )k }) be a ltered asque resolution of
(Q(Xi ,Zi )/S L(Xi ,Zi )/S (Wi /S (log Fi )), Q(Xi ,Zi )/S P Di ).
= (Q(Xi ,Zi )/S L(Xi ,Zi )/S (X(i,i) /S (log E(i,i) )), Q(Xi ,Zi )/S P Di )
(Q(Xi ,Zi )/S L(Xi ,Zi )/S (Wi /S (log Fi )), Q(Xi ,Zi )/S P Di )
(Ji , {(Ji )k }).
Hence, by the ltered cohomological descent (1.5.1) (2), the following com-
posite morphism
108 2 Weight Filtrations on Log Crystalline Cohomologies
(2.5.3.7)
(Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log Ei )), Q(Xi ,Zi )/S P Di )
log
i,Rcrys log,1
i,Rcrys (Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log Ei )), Q(Xi ,Zi )/S P Di )
log
i,Rcrys (Ji , {(Ji )k })
Since (Ii , {(Ii )k }) and (Ji , {(Ji )k }) are ltered asque resolutions of the
same complex (Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log Ei )), Q(Xi ,Zi )/S P Di ),
log log
we have an isomorphism i,Rcrys (Ji , {(Ji )k }) i,Rcrys (Ii , {(Ii )k }) in
+
D F(Q(Xi ,Zi )/S (O(Xi ,Zi )/S )) which makes the diagram of the triangle above
commutative. Hence the composite morphism
2.5 Well-Denedness of the Preweight-Filtered Complexes 109
(Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log Ei )), Q(Xi ,Zi )/S P Di )
log log
i,Rcrys (Ji , {(Ji )k }) i,Rcrys (Ii , {(Ii )k })
is an isomorphism in D+ F(Q(Xi ,Zi )/S (O(Xi ,Zi )/S )). Therefore we have proved
that the morphism (2.5.3.3) is a ltered quasi-isomorphism. We nish the
proof of (2.5.3).
log,Z
As a result, (Czar (O(X,DZ)/S ), P D ) is independent of the data (2.4.0.1)
and (2.4.0.2).
log,Z
Proof. (1): By (2.5.3), (CRcrys (O(X,DZ)/S ), P D ) is equal to
log log,Z
R(X,Z)/SRcrys (CRcrys (O(X ,D Z )/S ), P D )I
log
=R(X,Z)/SRcrys log
RRcrys (Q(X ,Z )/S L(X ,Z )/S (X /S (log E )),
Q(X ,Z )/S P D )
log log,Z
=R(X,Z)/SRcrys (CRcrys (O(X ,D Z )/S ), P D )J .
(2): We have
log,Z
(2.5.4.2) Ru(X,Z)/S (CRcrys (O(X,DZ)/S ), P D )
=Rzar Ru(X ,Z )/S (Q(X ,Z )/S L(X ,Z )/S (X /S (log E )),
Q(X ,Z )/S P D )
=Rzar (OD OX X /S (log E ), P D )
log,Z
=(Czar (O(X,DZ)/S ), P D ).
Here the rst (resp. second) equality follows from (1.6.4.6) (resp. (2.2.22.2)).
The fact that the isomorphism (2.5.4.1) is independent of the data (2.4.0.1)
and (2.4.0.2) immediately follows from (2.5.3.1) and (2.5.3.2).
log,Z
Remark 2.5.5. In 2.7 we shall prove that (CRcrys (O(X,DZ)/S ), P D ) is inde-
pendent of the choice of the decompositions of D and Z by their smooth
log,Z
components. As a result, (Czar (O(X,DZ)/S ), P D ) is also independent of
the choice above.
110 2 Weight Filtrations on Log Crystalline Cohomologies
Corollary 2.5.6. Let : (X, D Z) (X , D Z) be an admissible im-
mersion over S with respect to the union of decompositions D and Z of
D and Z by smooth components of D and Z, respectively. Let D be the log
PD-envelope of the locally closed immersion (X, Z) (X , Z) over (S, I, ).
Then the following hold:
(1)
log,Z
(2.5.6.1) (CRcrys (O(X,DZ)/S ), P D )
=(Q(X,Z)/S L(X,Z)/S (X /S (log(D Z))), Q(X,Z)/S P D ).
(2.5.6.2) log,Z
(Czar (O(X,DZ)/S ), P D ) = (OD OX X /S (log(D Z)), P D )
log log
Rcrys : (((Xn , Zn )/S)Rcrys , Q(Xn ,Zn )/S (O(Xn ,Zn )/S ))nN
(((X, Z)/S)log
Rcrys , Q(X,Z)/S (O(X,Z)/S ))
(2.5.7.1)
log,Z
(CRcrys (O(X,E)/S ), P D ) =
R log
(X,Z)Rcrys ((Q(Xn ,Zn )/S L(Xn ,Zn )/S (Xn /S (log En )), Q(Xn ,Zn )/S P
Dn
)nN ).
log
(X,Z)/SRcrys : (((X
, Z )/S)log
Rcrys , Q(X ,Z )/S (O(X ,Z )/S ))I
(((X, Z)/S)log
Rcrys , Q(X,Z)/S (O(X,Z)/S ))
log,Z
by the denition of (CRcrys
(O(X,E)/S ), P D ). Because Cech complexes are cal-
culated by alternating cochains as in [80, 3], the right hand side is canonically
isomorphic to
R log
(X,Z)/SRcrys
((Q(Xn ,Zn )/S L(Xn ,Zn )/S (Xn /S (log En )), Q(Xn ,Zn )/S P Dn )nN ).
112 2 Weight Filtrations on Log Crystalline Cohomologies
Corollary 2.5.8. With the notation of (2.5.7), let Dn be the log PD-envelope
of the locally closed immersion (Xn , Zn ) (Xn , Zn ) over (S, I, ). Let
zar : (Xn )nN X be a natural morphism of topoi. Then the following
holds:
(2.5.8.1)
log,Z
(Czar (O(X,E)/S ), P D ) = Rzar ((ODn OXn Xn /S (log En ), P Dn )nN ).
Let the notations be as in 2.4 and 2.5. Recall the projections u(X,Z)/S
and u(X,DZ)/S ((2.2.22.1), (2.4.6.4)). Set f(X,Z)/S := f u(X,Z)/S and
f(X,DZ)/S := f u(X,DZ)/S . Then we have the log crystalline coho-
mology sheaf Rh f(X,DZ)/S (O(X,DZ)/S ) (h Z). We also have the
log crystalline cohomology sheaf Rh f(D(k) ,Z| (k) )/S (O(D(k) ,Z| (k) )/S ) of
D D
(D(k) , Z|D(k) )/(S, I, ). In this section we construct the following spectral
sequence of OS -modules:
(2.6.0.1)
E1k,h+k = Rhk f(D(k) ,Z| )/S (O(D (k) ,Z|D(k) )/S Z crys
(k)log
(D/S; Z))
D (k)
= Rh f(X,DZ)/S (O(X,DZ)/S ).
and
azar : D+ ((f a(k) )1 (OS )) D+ (f 1 (OS ))
(k)
and
2.6 The Preweight Spectral Sequence 113
D
(2.6.1.2) grP log,Z
k (Czar (O(X,DZ)/S ))
(k)
=azar Ru(D(k) ,Z| )/S (O(D (k) ,Z|D(k) )/S Z zar
(k)
(D/S0 )){k}.
D (k)
log (k)
(2.6.1.3) (D (k) ,Z| )/Scrys
: (((D , Z |D(k) )/S)log
crys , O(D (k) ,Z | )
D (k) (k) )/S
D
be the natural morphism of ringed topoi (1.6). Then we have the following
equalities:
(2.6.1.4)
D
log,Z
grP
k (CRcrys (O(X,DZ)/S ))
D
log
k R(X,Z)/SRcrys (Q(X ,Z )/S L(X ,Z )/S (X /S (log(D Z ))))
=grP
Q P D
log
=R(X,Z)/SRcrys grk (X ,Z )/S (Q(X ,Z )/S L(X ,Z )/S
(X /S (log(D Z ))))
Q(X ,Z )/S acrys (O(D(k) ,Z |
log (k)log
=R(X,Z)/SRcrys
(k) )/S
D
Z crys
(k)log
(D /S; Z )){k}
=Q(X,Z)/S R(X,Z)/Scrys
log (k)log
acrys (O(D(k) ,Z |
(k) )/S
D
Z crys
(k)log
(D /S; Z )){k}
=Q(X,Z)/S acrys R(D
(k)log
log
(k) ,Z| )/Scrys
(O(D(k) ,Z |
D (k) (k) )/S
D
Z crys
(k)log
(D /S; Z )){k}
=Q(X,Z)/S acrys (O(D(k) ,Z|
(k)log
)/S Z crys
(k)log
(D/S; Z)){k}.
D (k)
Here the second, the third, the fourth and the fth equalities follow from
(1.3.4.1), (2.2.21.2), (1.6.4.1) and (1.6.0.13), respectively. The last equality
follows from the cohomological descent.
Next we prove that the isomorphism (2.6.1.4) is independent of the choice
of the data (2.4.0.1) and (2.4.0.2). Assume that we are given the other data
(2.4.0.1) and (2.4.0.2) as in 2.5. By the trivially ltered version of (2.5.3),
we have
114 2 Weight Filtrations on Log Crystalline Cohomologies
Z crys
(k)log
(D /S; Z )))
log P D P D log
Since RRcrys grk = grk RRcrys by (1.3.4.1), we have the following
commutative diagram
Q P D
grk (X ,Z )/S (Q(X ,Z )/S L(X ,Z )/S
(X /S (log(D Z ))))
Q P D
log
RRcrys grk (X ,Z )/S (Q(X ,Z )/S L(X ,Z )/S
(X /S (log(D Z ))))
Z crys
(k)log
(D /S; Z ))
Q(X ,Z )/S Rcrys (D(k) /S (log Z |D(k) )){k}
log (k)log
acrys (L(D(k) ,Z |
(k) )/S
D
Z crys
(k)log
(D /S; Z )).
Hence we see that the isomorphism (2.6.1.1) (and hence (2.6.1.2)) is inde-
pendent of the choice of the data (2.4.0.1) and (2.4.0.2).
0 (k > k ).
Set f (X,Z)/S := f u(X,Z)/S . Then there exists the following spectral sequence
(2.6.2.2)
E1k,h+k = E1k,h+k ((X, D Z)/S)
= Rhk f(D(k) ,Z| )/S (O(D (k) ,Z|D(k) )/S Z crys
(k)log
(D/S; Z))
D (k)
= Rh f(X,DZ)/S (O(X,DZ)/S ).
log,Z
Obviously we have Hh (f (X,Z)/S Ik ) = Rh f (X,Z)/S (PkD CRcrys (O(X,DZ)/S )).
By the proof of (1.3.4.1), grk (Ik ) is a asque resolution of grP log,Z D
k (CRcrys (O(X,
DZ)/S )) for k k . Hence, for k k , we have
Here, in the last equality, we have used the commutativity of the diagram
(1.6.3.1) for the trivially ltered case. Therefore we obtain (2.6.2.1). By using
(2.4.7.2), we obtain (2.6.2.2) similarly.
Corollary 2.6.3. Fix decompositions D and Z of D and Z by their
smooth components, respectively. Let : (X, D Z) (X , D Z) be an ad-
missible immersion over S with respect to D and Z . Let f : (X, DZ)
S0 and fS : (X , D Z) S be the structural morphisms. Let D be the log
PD-envelope of the locally closed immersion (X, Z) (X , Z) over (S, I, ).
(k)
Let fS : D(k) S be the PD-envelope of the locally closed immersion
D(k) D(k) over (S, I, ). Let k be a nonnegative integer. For integers k
and h, set
(2.6.3.2)
E1k,h+k = E1k,h+k ((X , D Z)/S; k ) = Rh fS (PkD X /S (log(D Z)))
such that
log,Z
CRcrys (O(X,DZ)/S ) = Q(X,Z)/S R(X,DZ,Z)/S (O(X,DZ)/S )
is the forgetting log morphism along D ((2.3.2)). Let j : U := X \ D X
be the natural open immersion. Let n be a positive integer. Let (X, D Z)
be as above or an analogous log scheme over C or an algebraically closed eld
of characteristic p > 0. Then we have the following translation if Z = :
2.7 The Vanishing Cycle Sheaf and the Preweight Filtration 117
(2.7.0.1)
/C l-adic crystal
Uan et
U ?
(Xan , Dan )log , (X log
an , Dan )et
(X, D)log et
((X, D)/S)logcrys
Xan , X an
et
X
(X/S) crys
et X
et ?
jan : Uan Xan jet : U
top : (Xan , Dan )log Xan
an : (X log
an , Dan )et Xan et :
(X, D)log
et
(X,D)/S : ((X, D)/S)log
crys
Xet
(X/S) crys
Rjan (Z) = Rtop (Z) Rjet (Z/ln ) = ?
Rtop (Z/n) = Ran (Z/n) Ret (Z/ln ) R(X,D)/S (O(X,D)/S )
Xan X et X
X zar crys X
uX/S : (X/S) zar
Z(Xan ,Dan )log
(Z/n)(Xan ,Dan )log (Z/n) log O(X,D)/S
(X,D)et
(Z/n)(X
,D log (p n)
an an )et
ZXan
(Z/n)Xan (n Z) et (p n)
(Z/n)X OX/S
(X/C (log D), P ) ? (Czar (O(X,D)/S ), P )
(Xan /C (log Dan ), P ) ? (CRcrys (O(X,D)/S ), P )
(Xan /C (log Dan ), ) ? (CRcrys (O(X,D)/S ), )
Here (Xan , Dan )log is the real blow up of (Xan , Dan ) ([58, (1.2)]) and top is
the natural morphism of topological spaces, (X log
an , Dan )et is the analytic log
etale topos of (Xan , Dan ) ([51]) and an is the forgetting log morphism to
the topos X an dened by the local isomorphisms to Xan ; the morphism et
in the middle column is the forgetting log morphism ([30], cf. [67, (1.1.2)]);
the upper (resp. lower) equality in the left column has been obtained in [58,
(1.5.1)] (resp. [72]), and the equality in the middle column ([30, (3.6)]) follows
from the following composite equality
(2.7.0.2)
h
gp
Rh et (Z/ln ) = (MD /OX ) Z Z/ln (h) = Rh jet (Z/ln ) (h Z, n Z>0 ).
Here the rst equality follows from [58, (2.4)] and the second equality is
Gabbers purity ([33]) which has solved Grothendiecks purity conjecture.
Recall that, in the crystalline case, Rjcrys (OU/S ) is not a good object ([3,
VI Lemme 1.2.2]).
The purpose of this section is to give another intrinsic description of the
log,Z
preweight-ltered restricted crystalline complex (CRcrys (O(X,DZ)/S ), P D )
and, as a corollary, to obtain the spectral sequence (2.6.2.2) in a dierent
way.
118 2 Weight Filtrations on Log Crystalline Cohomologies
(2.7.1.1)
Q(X,Z)/S Rk (X,DZ,Z)/S (O(X,DZ)/S )
= Q(X,Z)/S acrys (O(D(k) ,Z|
(k)log
)/S Z crys
(k)log
(D/S; Z)).
D (k)
log,Z log,Z
Proof. The increasing ltration {PkD CRcrys (O(X,DZ)/S )}kZ on CRcrys (O
(X,DZ)/S ) gives us the following spectral sequence
(2.7.1.2)
E1k,h+k = Hh (grP
D log,Z log,Z
k CRcrys (O(X,DZ)/S )) = H (CRcrys (O(X,DZ)/S )).
h
log,Z
Hh (CRcrys (O(X,DZ)/S )) = Hh (Q(X,Z)/S R(X,DZ,Z)/S (O(X,DZ)/S ))
= Hh (Q(X,Z)/S (X,DZ,Z)/S (I ))
= Q(X,Z)/S Hh ((X,DZ,Z)/S (I ))
= Q(X,Z)/S Rh (X,DZ,Z)/S (O(X,DZ)/S ),
this is equal to Q(X,Z)/S acrys (O(D(k) ,Z| (k) )/S Z crys (D/S; Z)), 0 for
(k)log (k)log
D
k = h and k
= h, respectively. Hence (2.7.1.2) degenerates at E1 ; thus we
have a canonical isomorphism
By the Leray spectral sequence for the functor (X,DZ,Z)/S : ((X, D Z)/S)
log log
crys ((X, Z)/S)crys and f(X,Z)/S : ((X, Z)/S)crys Xzar , we obtain the
log
(2.7.1.4) E2st = Rs f(D(t) ,Z| )/S (O(D (t) ,Z|D(t) )/S Z crys
(t)log
(D/S; Z)) =
D (t)
in D+ F(A ).
Proof. Let E I be a quasi-isomorphism into a complex of asque
A-modules. Let (I , ) (J , {Jk }) be a ltered asque resolution of
(I , ). Then, by applying the functor f to the morphism of this resolution,
we obtain a morphism
By (1.1.12) (2), the right hand side of (2.7.2.2) is equal to Rf ((E , )). On
the other hand, there exists a natural morphism f (k I ) k f (I ); in
fact, by the left exactness of f , we have f (k I ) k f (I ). Hence the
left hand side of (2.7.2.2) is equal to (f (I ), {k f (I )}) = (Rf (E ), ). It
is easy to check that the induced morphism in D+ F(A ) by the morphism
(2.7.2.2) is independent of the choice of I and (J , {Jk }). Therefore we have
a canonical morphism (2.7.2.1).
log,Z
Now we give another description of (CRcrys (O(X,DZ)/S ), P D ).
120 2 Weight Filtrations on Log Crystalline Cohomologies
In particular,
log,Z log,Z
(2.7.3.3) (CRcrys (O(X,DZ)/S ), ) = (CRcrys (O(X,DZ)/S ), P D ).
Proof. Fix the data (2.4.0.1) and (2.4.0.2) for D Z. Then, as usual, there
exists a natural morphism of ltered O(X ,Z )/S -modules:
log
R(X,Z)/SRcrys (Q(X ,Z )/S L(X ,Z )/S (X /S (log(D Z ))), ).
log
By composing (2.7.3.5) with the morphism R(X,Z)/SRcrys ((2.7.3.4)), we ob-
tain a morphism
(2.7.3.6)
log
(R(X,Z)/SRcrys Q(X ,Z )/S L(X ,Z )/S (X /S (log(D Z ))), )
log
R(X,Z)/SRcrys (Q(X ,Z )/S L(X ,Z )/S (X /S (log(D Z ))), Q(X ,Z )/S P D )
which is nothing but a morphism
(2.7.3.7)
(Q(X,Z)/S R(X,DZ,Z)/S (O(X,DZ)/S ), ) (CRcrys
log,Z
(O(X,DZ)/S ), P D )
by (1.6.4.1). (We have not yet claimed that the morphism (2.7.3.7) is in-
dependent of the data (2.4.0.1) and (2.4.0.2).) To prove that the morphism
(2.7.3.7) is a ltered quasi-isomorphism, it suces to prove that the induced
morphism
(2.7.3.8)
grk Q(X,Z)/S R(X,DZ,Z)/S (O(X,DZ)/S ) grP
D log,Z
k CRcrys (O(X,DZ)/S )
2.7 The Vanishing Cycle Sheaf and the Preweight Filtration 121
(2.7.3.9)
0 (i
= k).
D
log,Z
By the proof of (2.7.1) again, Hi (grP k CRcrys (O(X,DZ)/S )) is also equal
to the last formulas in (2.7.3.9). Hence the morphism (2.7.3.7) is a quasi-
isomorphism.
Finally we show that the morphism (2.7.3.7) is independent of the data
(2.4.0.1) and (2.4.0.2). Indeed, let the notations be as in 2.5. Using (2.5.3.1),
we have the following commutative diagram:
log
(R(X,Z)/SRcrys Q(X ,Z )/S L(X ,Z )/S
(X /S (log(D Z ))), )
log
(R(X,Z)/SRcrys log
RRcrys Q(X ,Z )/S L(X ,Z )/S
(X /S (log(D Z ))), )
log
(R(X,Z)/SRcrys Q(X ,Z )/S L(X ,Z )/S (X /S (log(D Z ))),
Q(X ,Z )/S P D )
log
(R(X,Z)/SRcrys log
RRcrys Q(X ,Z )/S L(X ,Z )/S (X /S (log(D Z ))),
Q(X ,Z )/S P D ).
Thus the independence in question follows.
log,Z
Denition 2.7.4. We call (Ecrys (O(X,DZ)/S ), P D ) D+ F(O(X,Z)/S ) the
preweight-ltered vanishing cycle crystalline complex of (X, D Z)/S with
respect to D. Set
log,Z
(Ezar log,Z
(O(X,DZ)/S ), P D ) := Ru(X,Z)/S (Ecrys (O(X,DZ)/S ), P D )
122 2 Weight Filtrations on Log Crystalline Cohomologies
Proof. By [23, (1.4.8)], the spectral sequence (2.7.1.4) is obtained from the
log,Z
increasing ltration {k CRcrys (O(X,DZ)/S )}kZ ; this ltration is equal to
D log,Z
{Pk CRcrys (O(X,DZ)/S )}kZ by (2.7.3). Hence (2.7.6) follows.
log,Z
Corollary 2.7.7. (1) The ltered complex (CRcrys (O(X,DZ)/S ), P D ) is in-
dependent of the choice of the decompositions of D and Z by their smooth
components. The spectral sequence (2.6.2.2) is also independent of the choice
of them.
(2) Let the assumptions be as in (2.5.6). Then the right hand sides of
(2.5.6.1) and (2.5.6.2) are independent of the choice of the decompositions of
D and Z by their smooth components.
R log
(X,Z)/SRcrys (Q(Xn ,Zn )/S L(Xn ,Zn )/S (Xn /S (log En ))nN )
By the same proof as that for the formula (2.7.3.2), we also have
R log
(X,Z)/SRcrys ((Q(Xn ,Zn )/S L(Xn ,Zn )/S (Xn /S (log En )),
in general. More specially, in this remark, we prove that the natural morphism
Hence, by the snake lemma and (2.7.11.3), we obtain the following exact
sequence
Now set X := SpecS (OS [x]) and let D be a relative smooth divisor on
X dened by an equation x = 0. Set (X, D) := (X , D) S S0 . In this
case, Coker(d) = 0 by the crystalline Poincare lemma. Hence, to prove
that (2.7.11.2) is not an isomorphism in this case, it suces to prove that
Ker()
= 0. Set A0 := OS0 [x, y]/(xy). Let f : A0 OS0 [x] be a morphism
of sheaves of rings over OS0 dened by equations f (x) = x and f (y) = 0.
Let APD
0 be the PD-envelope of A0 with respect to Ker(f ). Let be the PD-
structure on Ker(f ) and let f PD : APD
0 OS0 [x] be the induced morphism
of sheaves of rings over OS0 by f . Set T := SpecS (APD0 ). Then f induces a
0
PD closed immersion X T ; the triple (X, T, ) is an object of (X/S)crys .
0 OS0 OS0 [x] OS0 [x] be a morphism of sheaves of rings over
Let g : APD
OS0 dened by g(s t) := f PD (s)t (s APD 0 , t OS0 [x]) and let B be the
PD-envelope of APD 0 OS0 OS0 [x] with respect to Ker(g). Then, by the proof
of [11, (6.10)], the value LX/S (1X /S )T of LX/S (1X /S ) at T is given by the
following formula
while the value LX/S (1X (log D))T is given by the following formula
Let T : LX/S (1X /S )T LX/S (1X /S (log D))T be the value of at T . Then
T (dx) = (1 x)d log x.
To prove that T is not injective, it suces to prove that a morphism
B B given by multiplication by 1 x is not injective. Here we denote the
0 OS0 OS0 [x] in B by the same symbol s by
image of a local section s of APD
abuse of notation. We check
(A) y 1
= 0 in B
and
(B) (1 x)p (y 1) = 0 in B.
First we check (A). Consider the following commutative diagram
2.7 The Vanishing Cycle Sheaf and the Preweight Filtration 125
f
A0 OS0 [x]
OS0 [y] OS0 ,
where the vertical morphisms are dened by sending x to 0 and the lower
horizontal morphism is dened by sending y to 0. By taking the PD-envelopes
with respect to the kernels of the horizontal morphisms, we obtain the fol-
lowing commutative diagram:
f PD
APD
0 OS0 [x]
(2.7.11.5)
OS0 y OS0 .
Denote by the left vertical morphism in (2.7.11.5) and let : APD 0 OS0
OS0 [x] OS0 y be a morphism dened by (st) := (s)(t mod xOS0 [x])
0 , t OS0 [x]). Then the diagram (2.7.11.5) gives the following com-
(s APD
mutative diagram
g
APD
0 OS0 OS0 [x] OS0 [x]
OS0 y OS0
B OS0 [x]
(2.7.11.6)
OS0 y OS0 .
(1 x)p (y 1) = xp y 1 + (1 xp xp 1)(y 1)
= 0 + (1 x x 1)p (y 1)
= p!(1 x x 1)[p] (y 1) = 0
because p = 0 in B.
Now we have proved that the morphism (2.7.11.2) is not an isomorphism
in general.
126 2 Weight Filtrations on Log Crystalline Cohomologies
Remark 2.7.12. (1) Let (X, D) be a smooth analytic variety with (not neces-
sarily simple) NCD over the complex number eld. Let U be the complement
(0) be the nor-
of D in X and let j be the natural inclusion U X. Let D
(k)
malization of D and for a positive integer k, dene D in the way described
in (2.2.15) from D (0) . Let (k) X be the natural morphism. Then,
a(k) : D
in [23, (3.1.8)], Deligne has proved that
log (X/C)
where : (X/C) inf is the forgetting log morphism of inn-
inf
log
itesimal topoi, OX/C (resp. O (k) ) is the structure sheaf in (X/C)
D /C inf
Here the rst equality follows from (2.7.5.1). The third equality follows from
our assumption. The fourth equality follows from (2.4.7.3). However it is
practically well-known that the equality (2.7.12.3) does not hold in gen-
eral. Indeed, let be a eld of characteristic p > 0 and let (X, D) be a
smooth scheme with an SNCD over . Assume that S = S0 = Spec(). Then
(2.7.12.3) is an isomorphism
In this section we dene the log cycle class of a smooth divisor which intersects
the log locus transversally (cf. [29, 2]).
As an application, we give the description of the boundary morphism be-
tween the E1 -terms of the spectral sequence (2.6.2.2).
Let f : (X, Z) S0 be a smooth scheme with a relative SNCD over a
scheme S0 . Let D be a smooth divisor on X which intersects Z transver-
sally over S0 ; for a decomposition = {Z } of Z by smooth components
of Z, (D) := {D, Z } is a decomposition of D Z by smooth compo-
nents of D Z. The closed subscheme Z|D := Z D in D is a relative
SNCD on D/S0 ; |D := {Z |D } be a decomposition of Z|D by smooth
components of Z|D . Let a : (D, Z|D ) (X, Z) be the natural closed im-
mersion over S0 . Let azar : (D zar , OD ) (X zar , OX ) be the induced mor-
phism of Zariski ringed topoi. Let a log
crys : (((D,Z|D )/S)log , O(D,Z| )/S )
crys D
(((X, crys , O(X,Z)/S ) be also the induced morphism of log crystalline
Z)/S)log
ringed topoi. Let
(2.8.0.1)
ResD : X/S0 (log(D Z)) azar (D/S0 (log(Z|D )) Z zar
(1)
(D/S0 )){1}
be the Poincare residue morphism with respect to D/S0 . Then we have the
following exact sequence:
ResD
azar (D/S (log(Z|D )) Z zar
(1)
(D/S0 )){1} 0.
Let (S, I, ) and S0 be as in 2.4. As in 2.4, by abuse of notation, we also
denote by f the composite morphism (X, Z) S0 S.
As in 2.4, we have the following data:
r
(2.8.0.3): An open covering X = i0 I0 Xi0 with Xi = s=0 Xis (i =
(i0 , . . . , ir )). The family {(Xi , Di Zi )}iI (Di := D Xi , Zi := Z Xi )
of log schemes form a diagram of log schemes over (X, D Z), which we
denote by (X , D Z ). That is, (X , D Z ) is a contravariant functor
0 X /S (log Z ) X /S (log(D Z ))
Res
(2.8.0.5)
(2.8.0.6)
0 Q(X ,Z )/S L(X ,Z )/S (X /S (log Z ))
Q(X ,Z )/S L(X ,Z )/S (X /S (log(D Z )))
Q(X ,Z )/S alog
crys (L(D ,Z |D )/S (D /S (log(Z |D )))
Z crys
(1)log
(D /S; Z )){1}) 0.
log log
Recall the morphisms (X,Z)/Scrys and (D,Z| D )/Scrys
of ringed topoi in
(2.4.7.4) for the case D = and (2.6.1.3). By (1.6.0.23) we have the following
triangle
2.8 Boundary Morphisms 129
(2.8.0.7)
Q(X,Z)/S R(X,Z)/Scrys
log
L(X ,Z )/S (X /S (log Z ))
Q(X,Z)/S R(X,Z)/Scrys
log
L(X ,Z )/S (X /S (log(D Z )))
Q(X,Z)/S alog log
crys R(D,Z|D )/Scrys L(D ,Z |D )/S (D /S (log(Z |D ))
+1
Z crys
(1)log
(D /S; Z )){1} .
Q(X,Z)/S alog
+1
crys (O(D,Z|D )/S Z crys (D/S; Z)){1} .
(1)log
(2.8.0.9)
d : Q(X,Z)/S alog
crys (O(D,Z|D )/S Z crys (D/S; Z))){1}
(1)log
(2.8.0.10)
d : Q(X,Z)/S alog
crys (O(D,Z|D )/S Z crys (D/S; Z)))
(1)log
(2.8.0.11) GD/(X,Z) := d.
Z crys
(1)log
(D/S; Z)), Q(X,Z)/S (O(X,Z)/S )[1]{1}).
(1)log
Since crys (D/S; Z) is canonically isomorphic to Z and since there exists
a natural morphism Q(X,Z)/S (O(X,Z)/S ) Q(X,Z)/S alog
crys (O(D,Z|D )/S ), we
have a cohomology class
130 2 Weight Filtrations on Log Crystalline Cohomologies
X /S (log(D Z )) d log t
|D bzar (D /S (log(Z |D ))
Z zar
(1)
(D /S; Z ))[1]
Here we have used the Convention (4). Hence, by the Convention (2), we
have a Gysin morphism
Q(X,Z)/S (O(X,Z)/S ).
Proposition 2.8.2. The morphism GD/(X,Z) and the class c(X,Z)/S (D) are
independent of the data (2.8.0.3) and (2.8.0.4).
Proof. Use notations in 2.5. Assume that we are given two data in (2.8.0.3)
and two data in (2.8.0.4). Because the question is local, we may assume that
the two admissible immersions are admissible immersions with respect to the
same decompositions of D and Z by their smooth components. As in 2.5 we
have two morphisms
and
(D,Z|D )/S : (((D , Z |D )/S)crys , O(D ,Z |D )/S )
log
(((D ,
Z |D )/S)log
crys , O(D ,Z |D )/S )
(2.8.2.1)
Q(X ,Z )/S R(X,Z)/S L(X ,Z )/S (X /S (log Z ))
Q(X ,Z )/S L(X ,Z )/S (X /S (log Z ))
Remark 2.8.3. We can also construct c(X,Z)/S (D) by using the vanishing cycle
sheaf as follows.
Z)/S)log log
Let (X,DZ,Z)/S : ((X, D crys ((X, Z)/S)crys be the forgetting
log morphism along D ((2.3.2)). By (2.3.2.9), there exists a natural morphism
Set HDi
(O(X,Z)/S ) := Hi (RD (O(X,Z)/S )) (i Z). Then we have the follow-
ing exact sequence
(2.8.3.3) HD
i
(O(X,Z)/S ) Hi (O(X,Z)/S )
Ri (X,DZ,Z)/S (O(X,DZ)/S ) .
Here we have used the Convention (4) and (5). By (2.7.1), we have
(2.8.3.4) Q(X,Z)/S HD
i
(O(X,Z)/S )
Q(X,Z)/S alog
(1)log
crys (O(D,Z|D )/S Z crys (D/S; Z)) (i = 2),
=
0 (i
= 2).
132 2 Weight Filtrations on Log Crystalline Cohomologies
Q(X,Z)/S RD (O(X,Z)/S )
=Q(X,Z)/S alog
crys (O(D,Z|D )/S Z crys (D/S; Z)){1}[1].
(1)log
Q(X,Z)/S (O(X,Z)/S ).
log
crys (D/S; Z) := log
0 k1 crys
(D/S; Z)
and
log
j crys
(D/S; Z) := log j k1 crys (D/S; Z).
0
We x an isomorphism
(2.8.4.2) log
j crys
(D/S; Z) Z log
j crys
log
(D/S; Z) crys (D/S; Z)
We identify log
j crys
(D/S; Z)Z log
j crys
log
(D/S; Z) with crys (D/S; Z) by this
isomorphism. We also have the following composite morphism
(2.8.4.3)
j log
(1)j Gj : Q(D ,Z|D )/S crys (O(D ,Z|D )/S
log
Z crys (D/S; Z))
j j
j log
Q(D ,Z|D )/S crys (O(D ,Z|D )/S Z log
j crys
(D/S; Z) Z log
j crys
(D/S; Z))
j j
Gj 1
Q(D ,Z|D )/S (O(Dj ,Z|Dj )/S Z log
j crys
(D/S; Z))[1]{1}
j j
134 2 Weight Filtrations on Log Crystalline Cohomologies
dened by
(2.8.4.4)
j k1 ).
x (0 k1 )
(1)j Gj (x) (0
Here we have used the Convention (6). If D{0 ,...,k1 } = , set (1)j Gj :
= 0.
Denote by a (resp. aj ) the natural exact closed immersion (D , Z|D )
(X, Z) (resp. (Dj , Z|Dj ) (X, Z)).
Q
(X ,D )S P
D
grk (Q(X ,D )/S L(X ,Z )/S (X /S (log(D Z )))) 0.
log Q
(X ,D )/S P
D
R(X,Z)/SRcrys grk1 (Q(X ,D )/S L(X ,Z )/S (X /S (log(D Z ))))
log
R(X,Z)/SRcrys ((Q(X ,D )/S PkD /Q(X ,D )/S Pk2
D
)
log Q
(X ,D )S P
D
+1
R(X,Z)/SRcrys grk (Q(X ,D )/S L(X ,Z )/S (X /S (log(D Z )))) .
and D
ResD P
: grk X /S (log(D Z ))
D(;) /S (log Z |D(j ;) ){k} Z zar (D /S; Z ).
As in (2.8.4.2), we x an isomorphism
(2.8.5.1) j zar (D /S; Z ) Z j zar (D /S; Z ) zar (D /S; Z )
dened by
(2.8.5.3) j k1 )
(1)j ResD (x) (0 k1 ).
x (0 j
It is easy to check that (1)j Resj is well-dened. The morphism (1)j Resj
induces a morphism
(2.8.5.4)
L(X ,Z )/S ((1)j Resj ) :
L(X ,Z )/S (D( /S (log(D(;) Z |D(j ;) )) Z log
j crys
(D/S; Z ))
j ;)
(2.8.5.5)
Q P D
Q (
(X ,D )/S
0
gr L (log(D Z )))
k1 (X ,D )/S (X ,Z )/S X /S
D
Q (Res )
j
L
(X ,D )/S (X ,Z )/S
Q (
log
0
L (log Z |D )){(k 1)} Z (D /S; Z )
(X ,D )/S (X ,Z )/S D( ;) /S (j ;) j crys
j
(Q P /Q P )Q (
D D
L (D Z ))))
(X ,D )/S k (X ,D )/S k2 (X ,D )/S (X ,Z )/S X /S
D
Q (Res )
j ,
L
(X ,D )/S (X ,Z )/S
Q (
log
L (log(D(;) Z |D ))){(k 1)} Z (D /S; Z )
(X ,D )/S (X ,Z )/S D( ;) /S (j ;) j crys
j
Q L ((1)j Resj )
(X ,D )/S (X ,Z )/S
Q P D
Q (
(X ,D )/S
gr L (log D ))
0
k (X ,D )/S (X ,Z )/S X /S
D
Q L (Res )
(X ,D )/S (X ,Z )/S
Q (
log
L (log Z |D ;) )){k} Z (D /S; Z )
0.
(X ,D )/S (X ,Z )/S D(;) /S ( crys
ResD
j , (yd log x0 d log xk1 ) = (1) yd log xj (0 j k1 ),
j
Z)/S)log log
: ((X, D crys ((X, Z)/S)crys
and
: ((X , D
Z )/S )log ((X
crys
, Z )/S )log
crys
log
gcrys
: (((X, D Z)/S)log
crys ,O(X,DZ)/S )
(((X , D
Z )/S )log , O
crys (X ,D Z )/S )
Here the rst and the last equalities follow from (2.7.5.1); the rst arrow is
log
induced by gcrys and the second arrow is obtained from (2.7.2).
Corollary 2.9.2. Let Ess ((X, D Z)/S) (resp. Ess ((X , D Z )/S )) be the
spectral sequence (2.6.2.2) (resp. (2.6.2.2) for (X , D Z )/S ). Then the
log
morphism gcrys induces a morphism
(2.9.2.1) log
gcrys : Ess ((X , D Z )/S ) Ess ((X, D Z)/S)
of spectral sequences.
(2.9.2.2)
log
grP
k (g(X,Z)crys ) :
log
In the following, we make the morphism grPk (g(X,Z)crys ) in (2.9.2.2) explicit
in certain cases by using a notion which is analogous to the D-twist in [71].
Assume that the following two conditions hold:
(2.9.2.5)
Ru(X ,Z )/S a log log
crys (gcrys ) :
Proposition 2.9.3. Let the notations and the assumptions be as above. Let
log
be the induced morphism by g. Then the morphism Ru(X ,Z )/S (gcrys ) in
k log log
(2.9.2.5) is equal to ( j=1 ej )Ru(X ,Z )/S a crys (g(D ,Z|D )crys ) for k 0.
k
for each i0 I0 and from these data, let us construct the diagrams of admis-
sible immersions
(X , D Z ) (X , D Z ), (X , D Z ) (X , D Z )
D
(2.9.3.2) grP
k
log,Z
Czar (O(X ,D Z )/S )
D
=Rzar Rg(X ,Z )zar (OD OX grP
k X /S (log(D Z ))),
140 2 Weight Filtrations on Log Crystalline Cohomologies
where D (resp. D ) denotes the log PD-envelope of (X , Z ) (X , Z )
(resp. (X , Z ) (X , Z )). Because (Xi0 , Di0 Zi0 ) is log smooth over S
and the exact closed immersion (Xi0 , Di0 Zi0 ) (Xi0 , Di0 Zi0 ) is dened
by the nil-ideal sheaf IOXi0 , there exists a morphism gi0 : (Xi0 , Di0 Zi0 )
(Xi0 , Di0 Zi0 ) which is a lift of g|(Xi0 ,Di0 Zi0 ) (cf. [11, N.B. in 5.27]). The
family { gi0 }i0 I0 induces a morphism
(2.9.3.3) g : (X , D Z ) (X , D Z )
Di
Pk1 Xi /S (log(Di Zi )). So, if we put
we have the following commutative diagram (the vertical arrows are Poincare
residue morphisms with respect to D and D ):
D
D log,Z grP g )
(
grP
k (Czar (O(X ,D Z )/S ))
k
X /S (log(D Z )))
D
(2.9.3.4) (OD OX grP
k
Res
D
k
( e j ) h
(OD OX (a |(D ,Z |D ) )zar (;) ){k}
j=1 (;)
2.9 The Functoriality of the Preweight-Filtered Zariskian Complex 141
D
g(X ,Z )zar grP log,Z
k (Czar (O(X ,D Z )/S ))
D
g(X ,Z )zar (OD OX grP
k X /S (log(D Z )))
D
Res
Corollary 2.9.6. Assume that e s for all s are equal. Let Ess ((X, D
Z)/S) be the following spectral sequence
142 2 Weight Filtrations on Log Crystalline Cohomologies
= Rh f(X,DZ)/S (O(X,DZ)/S )
and let Ess ((X , D Z )/S ) be the obvious analogue of the above for (X , D
Z )/S . Then there exists a morphism
(2.9.6.1) log
gcrys : Ess ((X , D Z )/S ) Ess ((X, D Z)/S)
of spectral sequences.
F : (X, D Z) (X , D Z )
F(X,Z) : (X, Z) (X , Z )
and
F (k) : (D(k) , Z|D(k) ) (D(k) , Z |D(k) ).
Let
a(k) : (D(k) , Z|D(k) ) (X, D Z)
and
a(k) : (D(k) , Z |D(k) ) (X , D Z )
be the natural morphisms. We dene the relative Frobenius action
(k) log
(D /S; Z ) Fcrys
(k)log log (k)log
(k)log
(D(k) ,Z| )/S : acrys crys acrys crys (D/S; Z)
D (k)
(2.9.6.2)
E1k,h+k ((X, D Z)/S) = Rhk f(D(k) ,Z| )/S (O(D (k) ,Z|D(k) )/S
D (k)
Z crys
(k)log
(D/S; Z))(k)
= Rh f(X,DZ)/S (O(X,DZ)/S ).
2.10 The Base Change Theorem and the K
unneth Formula 143
Denition 2.9.7. We call the sequence (2.9.6.2) the preweight spectral se-
quence of (X, D Z)/(S, I, ) with respect to D. If Z = , then we call it the
preweight spectral sequence of (X, D)/(S, I, ).
(2.9.7.1)
G : Rhk f(D(k) ,Z| )/S (O(D (k) ,Z|D(k) )/S Z crys
(k)log
(D/S; Z)))(k)
D (k)
Z crys
(k1)log
(D/S; Z))((k 1)).
(2.9.7.2)
E2st := Rs f(D(t) ,Z| )/S (O(D (t) ,Z|D(t) )/S Z crys
(t)log
(D/S; Z))(t)
D (t)
g
Y Y
f
(2.10.1.1) f
u
(T , J , ) (T, J , )
in DF(OT ).
a morphism
in DF(OT ).
Proposition 2.10.2. (1) Let f : (X, D Z) S0 ( S) and (S, I, ) be
as in 2.4. Assume moreover that S is quasi-compact and that f : X S0 is
quasi-separated and quasi-compact. Let f(X,Z) : (X, Z) S0 ( S) be the
log,Z
induced morphism by f . Then Rh f(X,Z)/S PkD (Ecrys (O(X,DZ)/S ) (h, k
Z) are quasi-coherent OS -modules and Rf(X,Z)/S (Ecrys
log,Z
(O(X,DZ)/S ), P D )
is isomorphic to a bounded ltered complex of OS -modules.
(2) Let (S, I, ) and S0 be as in 2.4. Let Y be a quasi-compact smooth
scheme over S0 (with trivial log structure). Let f : (X, D Z) Y be a
morphism of log schemes such that f : X Y is smooth, quasi-compact and
quasi-separated and such that DZ is a relative SNCD over Y . (In particular,
D Z is also a relative SNCD on X over S0 .) Let f(X,Z) : (X, Z) Y be
2.10 The Base Change Theorem and the K
unneth Formula 145
log log,Z
the induced morphism by f . Then Rf(X,Z)crys (Ecrys (O(X,DZ)/S ), P D ) is
isomorphic to a bounded ltered complex of OY /S -modules.
By [11, 7.6 Theorem] and by the spectral sequences (2.6.2.2) and (2.10.2.1),
Hh ((f(X,Z) u(X,Z)/S ) (Ik )) (h, k Z) are quasi-coherent OS -modules
and there exists an integer h0 such that, for all h h0 and for all k Z,
Hh ((f(X,Z) u(X,Z)/S ) (Ik )) = 0. Hence Rh f(X,Z)/S PkD (Ecrys
log,Z
(O(X,DZ)/S )
(h, k Z) are quasi-coherent OS -modules and Rf(X,Z)/S (Ecrys log,Z
(O(X,DZ)/S ),
P D ) = ((f(X,Z) u(X,Z)/S ) (I ), (f u(X,Z)/S ) (Ik )) is isomorphic to a
bounded ltered complex of OS -modules.
(2): (2) immediately follows from (1) and from the proof of [3, V Corollaire
3.2.3] (cf. the proof of [11, 7.11 Corollary]).
Theorem 2.10.3 (Base change theorem). Let f : (X, D Z) S0 (
S) and (S, I, ) be as in (2.10.2). Let u : (S , I , ) (S, I, ) be a mor-
phism of PD-schemes. Assume that I is a quasi-coherent ideal sheaf of OS .
Set S0 := SpecS (OS /I ). Let f : (X , D Z ) := (X S0 S0 , (D Z) S0
S0 ) S0 be the base change morphism of f with respect to u|S0 . Then there
exists a canonical isomorphism
(2.10.3.1) Lu Rf(X,Z)/S (Ecrys
log,Z
(O(X,DZ)/S ), P D )
log,Z
Rf(X ,Z )/S (Ecrys (O(X ,D Z )/S ), P D )
log
g(X,DZ)crys : (((X , D
Z )/S )log , O
crys (X ,D Z )/S )
(((X, D Z)/S)log
crys , O(X,DZ)/S )
to the morphism i i , we have a ringed topos (((Xj , Dj Zj )/Sj )crys , O(Xj ,
log
Dj Zj )/Sj )jI . Let (Ij )jI be a asque resolution of (O(Xj ,Dj Zj )/Sj )jI
Z)/S)log log
((1.5.0.2)). Let : ((X, D crys ((X, Z)/S)crys and : ((X , D Z
)/S )log
log
crys ((X , Z )/S )crys be the forgetting log morphisms along D and
log,Z
D , respectively. Then (Ecrys log,Z
(O(X,DZ)/S ), P D ) and (Ecrys (O(X ,D Z )
D
/S ), P ) are represented by ( (I
i ), ) and ( (I
i ), ), respectively. Since
log 1 log 1 log 1
g(X,Z)crys is exact, g(X,Z)crys ( (Ii ), ) = (g(X,Z)crys (Ii ), ). By the follow-
ing commutative diagram
g(X,DZ)
(X , D Z ) (X, D Z)
g(X,Z)
(X , Z ) (X, Z),
log 1 log 1
we have a natural morphism (g(X,Z)crys (Ii ), ) ( g(X,DZ)crys (Ii ), ).
log 1
By the denition of (Ij )jI , we have the morphism g(X,DZ)crys (Ii ) Ii .
Hence we have a composite morphism
log 1
(g(X,Z)crys (Ii ), ) ( (Ii ), ).
and we have the same formula for (X , D Z )/S . We claim that f (X /S (log
(D Z)))/f (PkD ) is a at OS -module for any k. Indeed, the ltration PkD
on X /S (log(D Z)) is nite and f (X /S (log(D Z))) is a at OS -module.
Because X is ane over S, we have the following exact sequence
2.10 The Base Change Theorem and the K
unneth Formula 147
(2.10.3.2)
D
D
0 f (grP
k X /S (log(D Z))) f (X /S (log(D Z)))/f (Pk1 )
Lv ((Rf(X,Z)crys
log log,Z
(Ecrys (O(X,DZ)/S ), P D ))T )
log
log,Z
Rf(X ,Z )crys (Ecrys (O(X ,D Z )/S ), P D )T
is an isomorphism, where (X , D Z ) := (X , D Z ) U U .
log,Z
Rh f(X,Z)/Y (PkD Ecrys (O(X,DZ)/S ))OY 1Y/S (k Z)
making the following diagram commutative for any two nonnegative integers
kl:
(2.10.5.2)
k
Rh f(X,Z)/Y (Pk
D E log,Z (O
(X,DZ)/S )) Rh f(X,Z)/Y (Pk
D E log,Z (O
(X,DZ)/S ))O 1
crys crys Y Y/S
l
Rh f(X,Z)/Y (PlD Ecrys
log,Z (O
(X,DZ)/S )) Rh f(X,Z)/Y (PlD Ecrys
log,Z (O
(X,DZ)/S ))O 1 .
Y Y/S
148 2 Weight Filtrations on Log Crystalline Cohomologies
Lhcrys Rf(X,Z)crys
log log,Z
(Ecrys (O(X,DZ)/S ), P D )
Rf log log,Z
(X ,Z )crys (Ecrys (O(X ,D Z )/S ), P
D
)
is an isomorphism.
A-modules, we say that (E, {Ek }) is the direct sum of {Tl }lZ if Ek =
of
lk Tl (k Z).
Proof. Roughly speaking, the proof is dual to that of (1.1.7) with some ad-
ditional calculations.
We have only to prove the if part. Let k be an integer such that k0 <
k k1 . By the assumption, we may assume that E q = 0 (q > 0). Since
H 0 (Ek ) is nitely generated, there exists a free A-module Tk0 of nite rank
with a morphism Tk0 Ek0 such that the induced morphism Tk0 H 0 (Ek )
is surjective. Set Tk0 := 0 for k k0 or k > k1 . Let (Q0 , {Q0k }) be the direct
sum of {Tk0 }. Then we have a natural ltered morphism (Q0 , {Q0k })
(E 0 , {Ek0 }).
Assume that, for a nonpositive integer q, we are given a morphism
(Qq , {Qq
k }) (E q , {Ekq })
Ekq1 Ekq Ker(Qqk Qq+1 q q+1 q
k ) Ker(Qk Qk ) Qk .
Since A is noetherian, Ikq is nitely generated. Let {yi }iI be a system of nite
generators of Ikq . Take an element (xi , yi ) Ekq1 Ekq Ker(Qqk Qq+1 k ).
Because H q1
(Ek ) is nitely generated, we can take a family {zj }jJ of nite
elements of Ker(Ekq1 Ekq ) whose images in H q1 (Ek ) form a system of
generators of H q1 (Ek ).
Now consider a nitely generated A-module Skq1 generated by {(xi , yi )}iI
and {(zj , 0)}jJ in Ekq1 Ekq Ker(Qqk Qq+1 k ). Let Tk
q1
be a free A-
module of nite rank such that there exists a surjection Tkq1 Skq1 .
Set Tkq1 := 0 for k k0 or k > k1 . Let (Qq1 , {Qq1 k }) be the
direct sum of {Tkq1 }kZ . Then we have a natural ltered morphism
(Qq1 , {Qq1
k }) (E
q1
, {Ekq1 }).
By assumption, Ker(Qk Qq+1
q
k ) H (Ek ) is a surjection. Moreover,
q
Ker(Qqk Qq+1 k ) H (Ek ) induces an isomorphism H (Qk )
q q
q1
H (Ek ). Moreover, it is easy to see that Ker(Qk Qk ) H
q q q1
(Ek )
is surjective. Hence the induction works well and so we have constructed a
ltered complex (Q , {Qk }) such that Qq = 0 (q > 0), such that Qk0 = 0
and Qk1 = Q , such that (Qq , {Qqk }) (q Z) is the direct sum of a family
{Tkq }kZ of free A-modules of nite rank and such that there exists a l-
tered quasi-isomorphism (Q , {Qk }) (E , {Ek }). Because Ek (k) has
nite tor-dimension, grk E (k) also has it. Since (Q , {Qk }) is lteredly
quasi-isomorphic to (E , {Ek }), grk Q (k) also has it. Since the ltration
on Q is nite, there exists a nonpositive integer r and a complex Fk of at
A-modules for each k Z satisfying the following properties:
(a) Fk is quasi-isomorphic to grk Q ,
(b) Fk = 0 for > 0 or r.
Set Bkq := Im(Qq1k Qqk ). Let l k1 k0 be a positive integer. Set
0 (q < r l + 1 or q > 0),
Rkq 0 +l = Qqk0 +l /(Qqk0 +rq + Bkq0 +rq+1 ) (r l + 1 q r),
q
Qk0 +l (r < q 0).
TorA rl+1
i (Rk0 +l , M )
Hence Rkrl+1
0 +l
is a at A-module. The rest for showing the claim is to prove
that Rkr 0 +1 is a at A-module. As above, we can prove this using the following
resolution
k0 +1 Qk0 +1 Rk0 +1 0.
Qr1 r r
2.10 The Base Change Theorem and the K
unneth Formula 151
grk0 +l Qrl+2 .
rl+2
(2) K
unneth formula.
zar , OS )
(S zar , OS ),
(S
152 2 Weight Filtrations on Log Crystalline Cohomologies
log,Z2
Rf(X2 ,Z2 )/S (Ecrys (O(X2 ,D2 Z2 )/S ), P D2 )
Rf(X3 ,Z3 )/S (Ecrys
log,Z3
(O(X3 ,D3 Z3 )/S ), P D3 ).
For simplicity of notation, set j := (Xj ,Dj Zj ,Zj )/S (j = 1, 2, 3). We have to
construct a morphism
(2.10.11.3)
in D F(A).
Proof. Let Q E be a quasi-isomorphism from a complex of at A -
modules. Let (R , {Rk }) (Q , ) be a ltered at resolution of (Q , ).
Then, by applying the functor f to the morphism of this resolution, we
obtain a diagram
f (Rk ) f (k Q )
(2.10.12.2)
f (R ) f (Q ).
By (1.1.19) (2), the left hand side of (2.10.12.2) is equal to Lf ((E , )).
On the other hand, there exists a natural morphism f (k Q ) k f (Q ).
Hence there exists a natural diagram
f (k Q ) k f (Q )
(2.10.12.3)
f (Q ) f (Q ).
(2.10.13.1) (E , ) L L
A (F , ) (E A F , ).
(E , ) L
A (F , )
= (Q , {Qk }) A (F , )
= (Q A F , {Im( Ql A m F Q A F )}kZ )
l+m=k
(P A F , {Im( l P A m F P A F )}kZ )
l+m=k l+m=k
(E L
A F , ).
Rf(X2 ,Z2 )/S (R2 (O(X2 ,D2 Z2 )/S ), )}, (R3 (O(X3 ,D3 Z3 )/S ), ))].
First we have the following morphism
Lf(X3 ,Z3 )/S
{Rf(X1 ,Z1 )/S (R1 (O(X1 ,D1 Z1 )/S ), )L
OS
Lplog L
1crys (R1 (O(X1 ,D1 Z1 )/S ), )O(X 3 ,Z3 )/S
Lplog
2crys (R2 (O(X2 ,D2 Z2 )/S ), ).
Lplog log
jcrys (Rj (O(Xj ,Dj Zj )/S ), ) (Lpjcrys Rj (O(Xj ,Dj Zj )/S ), )
log
(R3 Lqjcrys (O(Xj ,Dj Zj )/S ), )
= (R3 (O(X3 ,D3 Z3 )/S ), )
Here we have obtained the rst morphism by (2.10.12), and the second mor-
phism by the commutative diagram (2.10.11.1) and the adjunction morphism.
Thus we have only to construct a canonical morphism
log,Z2
Rf(X2 ,Z2 )/S (Ecrys (O(X2 ,D2 Z2 )/S ), P D2 )
2.10 The Base Change Theorem and the K
unneth Formula 155
Rf(X3 ,Z3 )/S (Ecrys
log,Z3
(O(X3 ,D3 Z3 )/S ), P D3 ).
(2) Let Y and fj : (Xj , Dj Zj ) Y (j = 1, 2) be as in (2.10.2) (2). Set
f3 := f1 Y f2 . Then there exists a canonical isomorphism
log log,Z1
(2.10.14.2) Rf(X 1 ,Z1 )crys
(Ecrys (O(X1 ,D1 Z1 )/S ), P D1 )L
OY /S
log log,Z2
Rf(X 2 ,Z2 )crys
(Ecrys (O(X2 ,D2 Z2 )/S ), P D2 )
log
Rf(X 3 ,Z3 )crys
log,Z3
(Ecrys (O(X3 ,D3 Z3 )/S ), P D3 ).
and
log,Z
Hi := Rf(Xi ,Zi )/S (Ecrys i (O(Xi ,Di Zi )/S ), P Di )
(i = 1, 2, 3). Then the following diagram is commutative:
Lu (K)
Lu H1 L
OS Lu H2 Lu H3
(2.10.15.1)
K
H1 L
OS H2 H3 .
156 2 Weight Filtrations on Log Crystalline Cohomologies
Proof. We leave the proof of (2.10.15) to the reader because the proof is a
straightforward (but long) exercise by recalling the constructions of the base
change isomorphism and the K unneth isomorphism (cf. [3, V Proposition
4.1.3]).
Let the notations be as in 2.4. Let us dene a variant of a special case of the
denition of the log crystalline cohomology sheaf with compact support in [85,
5] briey (cf. [29, 2]). Let (U, T, , MT , ) be an object of the log crystalline
site ((X, D Z)/S)logcrys = ((X, M (D Z))/S)crys . Set MU := M (D Z)|U .
log
Denition 2.11.1. We call the higher direct image sheaf Rh f(X,DZ)/S (I(X,
D
DZ)/S ) in Szar the log crystalline cohomology sheaf with compact support
with respect to D and denote it by Rh f(X,DZ)/S,c (O(X,DZ;Z)/S ).
The local description of Rh f(X,DZ)/S,c (O(X,DZ;Z)/S ) is as follows; as-
sume that there exists an exact closed immersion : (X, DZ) (X , DZ)
into a smooth scheme with a relative SNCD over S such that induces exact
closed immersions (X, D) (X , D) and (X, Z) (X , Z). Let D be the log
PD-envelope of the exact closed immersion (X, Z) (X , Z) over (S, I, )
with structural morphism fS : D S. Let u(X,DZ)/S : ((X, D Z)/S)log
crys
X zar be the canonical projection. Let F be the crystal on ((X, D
crys corresponding to the integrable log connection OD OX OX (D)
Z)/S)log
OD OX OX (D)1X /S (log(D Z)). Then there exists a natural mor-
phism F I(X,DZ)/S
D
and it induces an isomorphism Q(X,DZ)/S (F)
=
Q(X,DZ)/S (I(X,DZ)/S
D
) by [85, (5.3)]. Hence we have the following
formula:
D
(2.11.1.1) Ru(X,DZ)/S (I(X,DZ)/S )
= Ru(X,DZ)/S Q(X,DZ)/S (I(X,DZ)/S
D
)
= Ru(X,DZ)/S Q(X,DZ)/S (F)
= Ru(X,DZ)/S (F) = OD OX X /S (log(Z D)),
2.11 Log Crystalline Cohomology with Compact Support 157
log
(2.11.1.2) (1)j crys
j
: O(Dj ,Z|D )/S Z log
j crys
(D/S; Z)
j
log log
j
crys (O(D ,Z|D )/S ) Z crys (D/S; Z)
j k1 )
(1)j j log (x) (0 k1 ). It is
dened by x (0 crys
log
easy to check that the morphism (1)j crys
j
is well-dened. Set
(2.11.1.3)
k1
log
(k1)log
crys := alog
j crys ((1) crys ) :
j j
(k1)log
acrys (O(D(k1) ,Z| )/S Z crys
(k1)log
(D/S; Z))
D (k1)
(k)log
acrys (O(D(k) ,Z| )/S Z crys
(k)log
(D/S; Z)).
D (k)
(k)log (k1)log
The composite morphism crys crys is the zero. Indeed, the question
is local. By taking trivializations of orientation sheaves, we can reduce this
vanishing to the usual well-known case.
In this section we start with the following:
(k1)log
Lemma 2.11.2. The morphism crys is independent of the choice of the
decomposition of {D } by smooth components of D/S0 .
Z)/S)log log
Theorem 2.11.3. Let : ((X, D crys ((X, Z)/S)crys be the for-
getting log morphism along D ((2.3.2)). Set
log,Z
(2.11.3.1) Ecrys,c (O(X,DZ)/S )
(0)log
:= (O(X,Z)/S Z crys (D/S; Z)
(0)log crys
(1)log (1)log
Z crys (D/S; Z))
(1)log crys
acrys (O(D(1) ,Z| )/S
D (1)
158 2 Weight Filtrations on Log Crystalline Cohomologies
(2)log (2)log
Z crys (D/S; Z)) ).
(2)log crys
acrys (O(D(2) ,Z| )/S
D (2)
Z)/S)log log
Rcrys : ((X, D Rcrys ((X, Z)/S)Rcrys
Z)/S)log log
Rcrys
((X, D Rcrys ((X, Z)/S)Rcrys
Q(X,DZ)/S
Q
(2.11.4.1) (X,Z)/S
Z)/S)log
log
((X, D crys ((X, Z)/S)crys .
Proof. First we show the existence of Rcrys . To show this, it suces to see
that, for an object T := (U, T, MT , , ) ((X, D Z)/S)log Rcrys , the object
log
(U, T, NT , , ) constructed in 2.3 belongs to ((X, Z)/S)Rcrys Zariski locally
inv
on T . (Then we can dene the exact functor Rcrys in the same way as in
2.3.) Let us assume that T is the log PD-envelope of the closed immersion
i : (U, (D Z)|U ) (U, MU ), where (U, MU ) is log smooth over S. Since the
log structure MU is dened on the Zariski site of U, we have a factorization
(U, (D Z)|U ) (U , MU ) (U, MU )
where CA(F
) = FD() is the log version of the Cech-Alexander complex of
F ([3, V 1.2.3]). Since F is parasitic, we have FD(n) = 0 for any n. Now we
have (Rq (F ))T = 0.
Proof (of Theorem 2.11.3). Assume that we are given the data (2.4.0.1) and
(k) (k)
(2.4.0.2) for (X, D Z). Let b : D X be the natural morphism. Let
log log
(X,DZ)/Scrys be the morphism of topoi dened in (2.4.7.4). Let (X,Z)/Scrys
be the morphism of topoi dened in (2.4.7.4) for the case D = . Let F be
the crystal on (X , D Z )/S corresponding to the integrable log connection
OD OX OX (D ) OD OX OX (D )1X /S (log(D Z )), where
D denotes the log PD-envelope of (X , D Z ) in (X , D Z ). Then we
have
Q(X,Z)/S R (I(X,DZ)/S
D
)
Q(X,Z)/S R R(X,DZ)/Scrys
= log log,1 D
(X,DZ)/Scrys (I(X,DZ)/S )
Q(X ,D Z )/S (X,DZ)/Scrys
= log log,1
RRcrys R(X,DZ)/SRcrys D
(I(X,DZ)/S )
Q(X ,D Z )/S (F )
= log
RRcrys R(X,DZ)/SRcrys
Q(X ,Z )/S R (F )
= log
R(X,Z)/SRcrys
Q(X ,Z )/S R L(X ,D Z )/S (X /S (log(Z D )))
= log
R(X,Z)/SRcrys
Q(X ,Z )/S L(X ,Z )/S (X /S (log(Z D ))).
= log
R(X,Z)/SRcrys
160 2 Weight Filtrations on Log Crystalline Cohomologies
By the same argument as that in [27, (4.2.2) (a), (c)], the following se-
quence
(0) (1)
b (D(1) /S (log Z |D(1) ) Z zar
,zar (1) ,zar
(1)
(D /S))
(0)
(0) ,zar (1) (1)
2X (log Z ) Z zar (D /S) b (2 (1) (log Z | (1) ) Z zar (D /S))
/S D /S D
d d
(0)
(0) ,zar (1) (1)
1X (log Z ) Z zar (D /S) b (1 (1) (log Z | (1) ) Z zar (D /S))
/S D /S D
d d
(0)
(0) ,zar (1) (1)
OX Z zar (D /S) b (O (1) Z zar (D /S))
D
d
(1) (2)
,zar (2) (2) ,zar
b (2 (2) (log Z | (2) ) Z zar (D /S))
D /S D
d
(1) (2)
,zar (2) (2) ,zar
b (1 (2) (log Z | (2) ) Z zar (D /S))
D /S D
d
(1) (2)
,zar (2) (2) ,zar
b (O (2) Z zar (D /S)) .
D
(2.11.5.3)
Q (1)
(X ,Z )/S (,zar )
Z zar
(1)
(D /S)))
is exact. Indeed, the question is local and we have only to prove that the
sequence (2.11.5.3) for = i is exact for a xed i I. As in (2.2.17), we have
only to prove that the following sequence
(2.11.5.4)
is exact. The following argument is the same as that in the proof of (2.2.17)
(1). We may have cartesian diagrams (2.1.13.1) and (2.1.13.2) for SNCD
Di Zi on Xi ; we assume that Di (resp. Zi ) is dened by an equation
x1 xt = 0 (resp. xt+1 xs = 0). Set Ji := (xd+1 , . . . , xd )OXi . We
may assume that there exists a positive integer N such that JiN ODi = 0.
Set Xi := SpecX (OXi /Ji ) and X := SpecS (OS [xd+1 , . . . , xd ]). Let Di
i
(resp. Zi ) be the closed subscheme of Xi dened by an equation x1 xt = 0
(resp. xt+1 xs = 0). As in [11, 3.32 Proposition], we may assume that there
exists a morphism
such that the induced morphism OXi [xd+1 , . . . , xd ]/J0iN OXi /JiN is an
isomorphism, where J0i := (xd+1 , . . . , xd ). By [11, 3.32 Proposition], ODi
is locally isomorphic to the PD-polynomial algebra OXi xd+1 , . . . , xd . Let
b i (k Z>0 ) and i,zar (k Z0 ) be analogous morphisms to bi
(k) (k) (k)
and
i,zar , respectively, for Xi , Di and Zi . Then we have an exact sequence
(k)
i,zar i,zar
(0) (1)
Z zar
(1)
(D /S)), d) }.
log
Applying the direct image R(X,Z)/SRcrys to (2.11.5.6), we have
log
(2.11.5.7) R(X,Z)/SRcrys Q(X ,Z )/S L(X ,Z )/S (X /S (log(Z D )))
log
{R(X,Z)/SRcrys Q(X ,Z )/S L(X ,Z )/S (X /S (log Z ) Z zar
(0)
(D /S))
Z zar
(1)
(D /S), d) }.
RRcrys Q (
log log
R L (log(Z D )))
(X,Z)/SRcrys (X ,Z )/S (X ,Z )/S X /S
Q (
log (0)
{R L (log Z ) Z zar (D /S))
(X,Z)/SRcrys (X ,Z )/S (X ,Z )/S X /S
RRcrys Q (
log log (0)
{R L (log Z ) Z zar (D /S))
(X,Z)/SRcrys (X ,Z )/S (X ,Z )/S X /S
2.11 Log Crystalline Cohomology with Compact Support 163
log (1)log
(log Z | (1) )
(X,Z)/SRcrys (X ,Z )/S crys (D (1) ,Z |
(R( Q a ) L ( (1)
)/S D /S D
D (1)
(1)
Z zar (D /S)), d) }
log log (1)log
(R RRcrys Q(X a L ( (1) (log Z | (1) )
(X,Z)/SRcrys ,Z )/S crys (D (1) ,Z
| (1) )/S D /S D
D
(1)
Z zar (D /S)), d) }.
Remark 2.11.6. Let the notation be as in the proof of (2.11.3) and let
L(X ,Z )/S be the complex
D
(2.11.6.1) R (I(X,DZ)/S )
= log log,1
R R(X,DZ)/Scrys (X,DZ)/Scrys D
(I(X,DZ)/S )
log
R R(X,DZ)/Scrys (F )
= log
R(X,Z)/Scrys R (F )
= log
R(X,Z)/Scrys R L(X ,D Z )/S (X /S (log(Z D )))
= log
R(X,Z)/Scrys L(X ,Z )/S (X /S (log(Z D )))
log
R(X,Z)/Scrys L(X ,Z )/S
= log log,1
R(X,Z)/Scrys (X,Z)/Scrys log,Z
Ecrys,c (O(X,DZ)/S )
=
Ecrys,c
log,Z
(O(X,DZ)/S ).
Note that the arrows in the above diagram without = are not necessarily
isomorphisms: they become isomorphic only after we apply Q(X,Z)/S . Note
also that they become isomorphic if we apply Ru(X,Z)/S or Rf(X,Z)/S be-
cause Ru(X,Z)/S = Ru(X,Z)/S Q(X,Z)/S and Rf(X,Z)/S = Rf (X,Z)/S
Q(X,Z)/S .
D+ F(O(X,Z)/S ).
164 2 Weight Filtrations on Log Crystalline Cohomologies
log,Z
Denition 2.11.7. We call (Ecrys,c (O(X,DZ)/S ), PcD ) the preweight-ltered
vanishing cycle crystalline complex with compact support of O(X,DZ)/S (or
(X, D Z)/S) with respect to D. Set
log,Z
(Ccrys,c (O(X,DZ)/S ), PcD ) := Q(X,Z)/S (Ecrys,c
log,Z
(O(X,DZ)/S ), PcD ).
log,Z
We call (Ccrys,c (O(X,DZ)/S ), PcD ) the preweight-ltered crystalline complex
with compact support of O(X,DZ)/S (or (X, D Z)/S) with respect to D.
Set
log,Z
(Ezar,c log,Z
(O(X,DZ)/S ), PcD ) := Ru(X,Z)/S (Ecrys,c (O(X,DZ)/S ), PcD ).
log,Z
We call (Ezar,c (O(X,DZ)/S ), PcD ) the preweight-ltered vanishing cycle zari-
skian complex with compact support of O(X,DZ)/S (or (X, D Z)/S) with
respect to D.
log,Z
By the denition of (Ezar,c (O(X,DZ)/S ), PcD ), there exists the following
canonical isomorphism in D+ (f 1 (OS )) :
(2.11.7.1)
log,Z
Ezar,c (O(X,DZ)/S )
{Ru(X,Z)/S (O(X,Z)/S Z crys
(0)log
(D/S; Z))
(1)
azar (Ru(D(1) ,Z| )/S (O(D (1) ,Z|D(1) )/S Z crys
(1)log
(D/S; Z)), d)
D (1)
}.
Remark-Denition 2.11.8. Because the notation for the right hand side
of (2.11.7.1) is only suggestive, we have to give the strict denition of it. Let
I be a double complex of O(X,Z)/S -modules such that, for each nonnegative
(k)log
integer k, I k is a u(X,Z)/S -acyclic resolution of (acrys (O(D(k) ,Z| )/S Z
D (k)
(k)log
crys (D/S; Z)), (1)k d). Then the right hand side of (2.11.7.1) is, by de-
nition, an object in D (f 1 (OS )) which is given by the single complex
+
of u(X,Z)/S (I ). Let PcD := {PcD,k }kZ be the stupid ltration with re-
spect to the rst degree of u(X,Z)/S (I ). Then (Ezar,c
log,Z
(O(X,DZ)/S ), PcD ) =
D + 1
(u(X,Z)/S (I ), Pc ) in D F(f (OS )).
log,Z D
Corollary 2.11.9. Ezar,c (O(X,DZ)/S ) = Ru(X,DZ)/S (I(X,DZ)/S ).
Proof. We have only to apply the direct image Ru(X,Z)/S to (2.11.3.2) and
to use the commutative diagram (1.6.3.1) for the case of the trivial ltration.
(2.11.9.1)
Rf(X,DZ)/S,c (O(X,DZ;Z)/S ) {Rf(X,Z)/S (O(X,Z)/S Z crys
(0)
(D/S; Z))
log,Z
Next we prove the base change theorem of (Ecrys,c (O(X,DZ)/S ), PcD ).
Proof. This immediately follows from the spectral sequence (2.10.2.1) and
[11, 7.6 Theorem], [11, 7.13 Corollary].
Theorem 2.11.11 (Base change theorem). Let the notations and the
assumptions be as in (2.10.3). Then there exists the following canonical
isomorphism
Because Rf(D(k) ,Z| )/S (O(D (k) ,Z|D(k) )/S ) is bounded by (2.10.2.1) and
D (k)
[11, 7.6 Theorem], and because D = if k 0 (since X is quasi-
(k)
compact), if r is large enough, the natural inclusions r (f u(X,Z)/S ) (I k )
(f u(X,Z)/S ) (I k ) are quasi-isomorphisms for all k. Hence the natural mor-
phism
s(r (f u(X,Z)/S ) (I )) s((f u(X,Z)/S ) (I ))
is a quasi-isomorphism. Let
and
d : (f u(X,Z)/S ) (I k ) (f u(X,Z)/S ) (I k,+1 )
be the boundary morphisms. Using the functor L0 in [11, 7], we have a at
resolution Qk of r (f u(X,Z)/S ) (I k ) for a xed r 0. The morphisms d
and d induce morphisms dQ : Qjl Qj,+1,l and dQ : Qjk Qjk,+1 ,
respectively. We x the boundary morphisms as follows: (1)j dQ : Qjl
166 2 Weight Filtrations on Log Crystalline Cohomologies
Qj,+1,l and (1)j dQ : Qjk Qjk,+1 . We also have a natural boundary
morphism Qkl Q+1,k,l . By these three boundary morphisms, we have
a triple complex Q . Then Lu PcD,k Rf(X,DZ)/S (Ecrys,c
log,Z
(O(X,DZ)/S )) =
k ,k+1,
(u Q u Q ){k}. By the base change theorem of Kato
([54, (6.10)]), this complex is isomorphic to
(k)log
{(Rf(D(k) ,Z | )/S (O(D (k) ,Z | (k) )/S Z crys (D /S ; Z ))){k}, (1)k d)
D (k) D
log,Z
} = Rf(X ,Z )/S PcD ,k
(Ecrys,c (O(X ,D Z )/S )).
T
XU
(2.11.13.1) g
U T
2.11 Log Crystalline Cohomology with Compact Support 167
in the same way as in the proof of (2.3.11) such that is an exact closed
immersion and that g is log smooth. Then we can form a crystal F on
(XU /T )log D
crys satisfying the equality QXU /S (I(X,DZ)/S ) = QXU /S (F). Then
i D i
we have R fXU /T (I(X,DZ)/S ) = R fXU /T (F) and it vanishes for i > a + b
log D
by (2.3.11). Now we have proved that Rfcrys (I(X,DZ)/S ) is bounded.
D
Let us prove that R (I(X,DZ)/S ) is bounded. It suces to prove that
there exists a positive integer r such that, for any (U, T, ) ((X, Z)/S)log
crys ,
Ri fXU /T (I(X,DZ)/S
D
) = 0 (where XU := U (X,Z) (X, D Z) = (U, (D
Z)|U )) for i > r. We may also assume that X is suciently small. Hence
we may assume that the log structure M (D) associated to D admits a chart
of the form : Nb M (D). Let us denote the log structure on X as-
sociated to D Z by MX . If we put 1 := idOX and 2 := idNb , we can
construct the commutative diagram (2.11.13.1) in the same way as (2.3.11)
and then we can form a crystal F on (XU /T )log crys which satises the equality
QXU /S (I(X,DZ)/S
D
) = QXU /S (F). Then we have Ri fXU /T (I(X,DZ)/S
D
) =
i
R fXU /T (F) and it vanishes for i > b by (2.3.11). Hence we have also proved
D
that R (I(X,DZ)/S ) is bounded.
Proof. (1) follows from (2.11.12) in the same way as [3, V], [11, 7] (see also
17).
Let us prove (2). One can construct the morphism (2.11.14.2) in usual
log D
way ([3, V Theor`eme 3.5.1]), using the boundedness of Rfcrys (I(X,DZ)/S )
which has been proved in (2.11.13). We can take data (X , D Z )
(X , D Z ) as (2.4.0.1), (2.4.0.2) for (X, D Z). If we put (X , D Z ) :=
(X , D Z ) S S and (X , D Z ) := (X , D Z ) S S , we obtain the
data (X , D Z ) (X , D Z ) as (2.4.0.1), (2.4.0.2) for (X , D Z ).
Then we see from the diagram (2.11.6.1) that there exists a diagram of base
change morphisms
168 2 Weight Filtrations on Log Crystalline Cohomologies
Lhcrys Rfcrys
log D
(I(X,DZ)/S ) Lhcrys Rf(X
log
,D Z )crys
(F )
Rf crys (I(X Rf (X ,D Z )crys (F )
log log
D
,D Z )/S )
Lhcrys Rf(X,Z)crys
log log,Z
(Ecrys,c (O(X,DZ)/S ))
Rf (X ,Z )crys (Ecrys,c
log log,Z
(O(X ,D Z )/S )),
where f(X ,D Z ) (resp. f(X ) denotes the composite morphism of
,D Z )
(X , D Z ) (X, DZ) with f (resp. (X , D Z ) (X , D Z ) with
f ) and F is the crystal on (X , D Z )/S dened in the same way as F. To
prove (2), it suces to prove that the horizontal arrows are isomorphisms. We
are reduced to showing (as in [3, V 3.5.5]) that, in the situation in (2.10.3),
the horizontal arrows in the following diagram of base change morphisms
Lu Rf(X,DZ)/S (I(X,DZ)/S
D
) Lu Rf(X ,D Z )/S (F )
D
Rf(X ,D Z )/S (I(X ,D Z )/S ) Rf(X ,D Z )/S (F )
Lu Rf(X,Z)/S (Ecrys,c
log,Z
(O(X,DZ)/S ))
Rf(X ,Z )/S (Ecrys,c
log,Z
(O(X ,D Z )/S ))
are isomorphisms. This follows from (2.11.6) because the arrows in (2.11.6.1)
become isomorphic if we apply Rf(X,Z)/S . Hence we have proved (2).
log,Z
By using the ltered complex (Ecrys,c (O(X,DZ)/S ), PcD ), by (2.11.9) and
by the Convention (6), we have the following spectral sequence
k,hk
(2.11.14.3) E1,c ((X, D Z)/S)
= Rhk f(D(k) ,Z| )/S (O(D (k) ,Z|D(k) )/S Z crys
(k)log
(D/S; Z)))
D (k)
= Rh f(X,DZ)/S,c (O(X,DZ;Z)/S ).
0 (k < k).
2.11 Log Crystalline Cohomology with Compact Support 169
(k)log
Rhk f(X,DZ)/S ((acrys (O(D(k) ,Z| )/S Z crys
(k)log
(D/S; Z)), (1)k d)
D (k)
(k+1)log
acrys (O(D(k+1) ,Z| )/S Z crys
(k)log
(D/S; Z)), (1)k+1 d) ).
D (k+1)
(2.11.15.1)
D
Ph,c Rh f(X,DZ)/S,c (O(X,DZ;Z)/S )
= PcD, Rh f(X,DZ)/S,c (O(X,DZ;Z)/S ).
(2.11.16.1)
k,h+k
{E1,c ((X, D Z)/S) OS OS
= Rh f(X,DZ)/S,c (O(X,DZ;Z)/S ) OS OS }
k,h+k
{(E1,c ((X , D Z )/S )
= Rh f(X ,D Z )/S ,c (O(X ,D Z ;Z )/S )}
of OS -modules.
Proof. Though the proof is the same as that of [68, (5.1)], we give the proof
here.
We have the following triangle
(2.11.17.1) Rf(D(k+1) ,Z| )/S (O(D (k+1) ,Z|D(k+1) )/S )[(k + 1)]
D (k+1)
+1
PcD,k /PcD,k+2 ((2.11.9.1)) Rf(D(k) ,Z| )/S (O(D (k) ,Z|D(k) )/S )[k] .
D (k)
170 2 Weight Filtrations on Log Crystalline Cohomologies
by the Convention (4) and (5). By the Convention (3), (4), (6) and by
(l)log
taking the Godement resolution of the complex acrys (O(D(l) ,Z| (l) )/S Z
D
(l)log
crys (D/S; Z)) (l = k, k + 1) , we can check that dk,hk
1 is equal to the
(k)log
morphism induced by crys .
log
Rf(X 3 ,D3 Z3 )crys
D3
(I(X3 ,D3 Z3 )/S
)
which is compatible with the isomorphism (2.11.19.1).
(3) The isomorphisms (2.11.19.1), (2.11.19.2) are compatible with the base
change isomorphism (cf. (2.10.15)).
2.11 Log Crystalline Cohomology with Compact Support 171
(2.11.19.3)
log
qjcrys
log log
(((Xj , D jZj )/S)crys , O(X j ,Dj Zj )/S
) (((X3 , D 3Z3 )/S)crys , O(X 3 ,D3 Z3 )/S
)
j
3
log
pjcrys
(((Xj
, Zj )/S)log
crys , O(Xj ,Zj )/S ) (((X3
, Z3 )/S)log
crys , O(X3 ,Z3 )/S )
f(X f
j ,Zj )/S (X 3 ,Z3 )/S
zar , O )
(S zar , O )
(S
S S
log log
R3 (Lq1crys (F1 ) L
O(X Lq2crys (F2 )).
3 ,D3 Z3 )/S
(2.11.19.6)
Lplog
1crys R1 (F1 ) O(X
L
Lplog
2crys R2 (F2 ) R3 (F3 ).
3 ,Z3 )/S
172 2 Weight Filtrations on Log Crystalline Cohomologies
for k N, where
(k) log (k)
aj : (Dj , Zj |D(k) ) (Xj , Zj )
j
which is constructed in the same way as (2.11.19.6) and we also have natural
morphisms
(2.11.19.8) Lplog
1crys (L1 ) O(X
L
Lplog
2crys (L2 ) L3 ,
3 ,Z3 )/S
log
(2.11.19.9) Lplog
1crys (L1 ) O(X3 ,Z3 )/S Lp2crys (L2 ) L3 .
(2.11.19.6) (2.11.19.7),
=
the isomorphism Rj (Llog
j ) Lj induces the isomorphism
=
(2.11.19.7) (2.11.19.8)
(2.11.19.8) (2.11.19.9).
log
By applying R(X 3 ,Z3 )/Scrys
to the diagram (2.11.19.10) and by using the
adjunction formula, we obtain the commutative diagram
(2.11.19.11)
Lplog log
1crys R1 R(X (F1 ) L
O Lplog log
2crys R2 R(X (F2 )
1 ,D1 Z1 )/Scrys (X3 ,Z3 )/S 2 ,D2 Z2 )/Scrys
log
R3 R(X (F3 )
3 ,D3 Z3 )/Scrys
Lplog
log
1crys R(X (L
1 ) O(X Lplog log
2crys R(X (L
2 )
1 ,Z1 )/Scrys 3 ,Z3 )/S 2 ,Z2 )/Scrys
log
R(X (L
3 ).
3 ,Z3 )/Scrys
log log
(Note that, by (2.3.11), Rj R(X j ,Dj Zj )/Scrys
(Fj ) and R(X j ,Zj )/Scrys
(Lj ) are bounded.) Let us put
(k)log
Ojk := ajcrys (O(D(k) ,Zj | Z crys
(k)log
(Dj /S; Zj )),
j (k) )/S
Dj
where
(k) log (k)
aj : (Dj , Zj |D(k) ) (Xj , Zj )
j
(2.11.19.12) Lplog
1crys (O1 ) O(X
L
Lplog
2crys (O2 ) O3 ,
3 ,Z3 )/S
=
log
and the isomorphism R(X j ,Zj )/Scrys
(Lj ) Oj induces the isomorphism
=
(the right column of (2.11.19.11)) (2.11.19.12).
(2.11.19.13)
Lplog
1crys R1 (I(X1 ,D1 Z1 )/S ) O(X
D1 L
Lplog D2
2crys R2 (I(X2 ,D2 Z2 )/S )
3 ,Z3 )/S
R3 (I(X
D3
3 ,D3 Z3 )/S
),
D
(note that Rj (I(Xjj ,Dj Zj )/S ) is bounded by (2.11.13)) and the morphism
D log
I(Xjj ,Dj Zj ) R(X j ,Dj Zj )/Scrys
(Fj ) induces the morphism
(2.11.19.14)
D D
Lplog
1crys R1 I(X
1 L
O Lplog
2crys R2 I(X
2
1 ,D1 Z1 )/S (X3 ,Z3 )/S 2 ,D2 Z2 )/S
D
R3 (I(X3 )
3 ,D3 Z3 )/S
Lplog log
1crys R1 R(X (F1 ) L
O Lplog log
2crys R2 R(X (F2 )
1 ,D1 Z1 )/Scrys (X3 ,Z3 )/S 2 ,D2 Z2 )/Scrys
log
R3 R(X (F3 )
3 ,D3 Z3 )/Scrys
Lplog
1crys (O1 ) L
O(X ,Z )/S Lplog
2crys (O2 )
3 3
O3 .
log
By applying Rf(X 3 ,Z3 )
to the diagram (2.11.19.14) and by using the adjunc-
tion formula, we obtain the diagram
(2.11.19.15)
log D log D2
Rf (I 1 ) L
OY /S Rf(X2 ,D2 Z2 )crys (I(X2 ,D2 Z2 )/S )
(X1 ,D1 Z1 )crys (X1 ,D1 Z1 )/S
log D
Rf (I 3 )
(X3 ,D3 Z3 )crys (X3 ,D3 Z3 )/S
log log
Rf R (F3 )
(X3 ,D3 Z3 )crys (X3 ,D3 Z3 )/Scrys
Rf
log ) L
(O1
log
(X1 ,Z1 )crys OY /S Rf(X2 ,Z2 )crys (O2 )
Rf
log ).
(O3
(X3 ,Z3 )crys
(2) These isomorphisms for various n Z>0 are compatible with transition
morphisms with respect to n.
However, as pointed out in [68, 7], the proofs of these two claims have
gaps: especially we cannot nd a proof of (2) in the proof of [46, (4.19)];
in [68, (7.19)], we have completed the proof of [46, (4.19)]. Hence we can use
[46, (4.19)]. In addition, we have to note one more point as in [68, (7.20)]
for the completeness of this book; in the denition of the embedding system
in [46, p. 237], we allow the (not necessarily closed) immersion as in [82,
Denition 2.2.10].
Now we come back to our situation. We keep the notations in 2.4. For
example, the morphism f : X Spec() is smooth and D Z is a transver-
sal SNCD on X; by abuse of notation, we also denote by f the composite
morphism X Spec() Spec(Wn ) (n Z>0 ). Because the morphism
(X, D Z) (Spec(), ) of log schemes is of Cartier type, we can ap-
ply the general theory of the log de Rham-Witt complexes in [46, 4] and
[68, 6, 7] (cf. [48]) to our situation above. In particular, we have a canoni-
cal isomorphism
(2.12.0.1) Ru(X,DZ)/Wn (O(X,DZ)/Wn ) Wn X (log(D Z))
Z zar
(k)
(Dn /Wn )(k)) 0 (q Z).
Proof. (cf. [64, 1.2]) The problem is Zariski local. In the following, we x an
(k)
isomorphism zar (Dn /Wn ) Z.
Let u : Xet X zar be a canonical morphism of topoi. For a coherent
ODX (Xn ) -module (resp. a coherent OYn -module) F on DX (Xn )zar Xzar
(resp. Ynzar Xzar ), let Fet be the corresponding coherent ODX (Xn ) -module
(resp. a coherent OYn -module) on DX (Xn )et Xet (resp. Yn et Xet ). Let
us consider the following diagram
Dn
(2.12.2.2) 0 Hq ((ODX (Xn ) OXn Pk1 Xn /Wn (log(Dn Zn )))et )
Hq ((ODX (Xn ) OXn PkDn Xn /Wn (log(Dn Zn )))et )
Hqk ((ODX (Xn ) OXn D(k) /W (log Zn |Dn(k) ))et ) 0
n n
(q Z),
which is the etale analogue of the diagram (2.12.2.1). We prove that the dia-
gram (2.12.2.1) is exact for any k, q Z if and only if the diagram (2.12.2.2)
is exact for any k, q Z.
By the Zariski analogue of [54, (6.4)], both
and
Hq (E (k) /W (log(Wn(k) |En(k) ))) = Wn qD(k) (log(Z|D(k) ))
n n
calculate Rq u(D(k) ,Z| )/Wn (O(D (k) ,Z|D(k) )/Wn ). Hence we have
D (k)
be the morphism of topoi which is dened in the same way as the morphism
u(D(k) ,Z| (k) )/Wn . Then, by [54, (6.4)], both
D
and
Hq ((E (k) /W (log(Wn(k) |En(k) )))et ) = Wn qD(k) (log(Z|D(k) ))
n n
calculate R u(D(k) ,Z| (k) )/Wn ,et (O(D(k) ,Z| (k) )/Wn ). Hence we have Hq ((ODX
q
D D
(Xn ) OXn (k) (log(Zn |Dn(k) )))et ) = Wn qD(k) (log(Z|D(k) )) on Xet and it
Dn /Wn
is the quasi-coherent Wn (OX )-module on Xet corresponding to Hq (ODX (Xn )
OXn (k) (log(Zn |Dn(k) ))). Hence there exists the canonical isomorphism
Dn /Wn
and for any etale morphism : X X, there exists the following canonical
isomorphism
(2.12.2.4)
Wn (OX ) 1 (Wn (OX )) Hq (ODX (Xn ) OXn D(k) /W (log(Zn |Dn(k) )))
n n
Now let us assume that the diagram (2.12.2.2) is exact for any k, q Z.
Then, by (2.12.2.3) and the induction on k, we see that each term of (2.12.2.2)
is u -acyclic and that u ((2.12.2.2)) gives the exact sequence (2.12.2.1). On
the other hand, assume that the diagram (2.12.2.1) is exact for any k, q Z.
In this case, note that the morphisms in the diagram (2.12.2.1) and the long
exact sequence
Dn
Hq ((ODX (Xn ) OXn Pk1 Xn /Wn (log(Dn Zn )))et )
Hq ((ODX (Xn ) OXn PkDn Xn /Wn (log(Dn Zn )))et )
Hqk ((ODX (Xn ) OXn D(k) /W (log Zn |Dn(k) ))et )
n n
Wn (OX ) 1 (Wn (OX )) Hq (ODX (Xn ) OXn PkDn Xn /Wn (log(Dn Zn )))
Hq ((ODX (Xn ) OXn PkDn Xn /Wn (log(Dn Zn )))et )|Xzar
=
2.12 Filtered Log de Rham-Witt Complex 179
(2.12.2.5)
Rq u(X,DZ)/Wn (O(X,DZ)/Wn ) Hq (OD X (Xn ) OXn X (log(Dn Zn )))
n /Wn
Hq (f )
Rq u(X,DZ)/Wn (O(X,DZ)/Wn ) Hq (Y (log(En Wn ))).
n /Wn
ODX (Xn ) OXn PkDn Xn /Wn (log(Dn Zn )) PkEn Yn /Wn (log(En Wn )).
(k)
By noting that DX (Xn )Xn Dn is the PD-envelope of the closed immersion
(k)
D(k) Dn ((2.2.16) (2)), we see that the morphism above is an isomor-
phism by [11, 7.1 Theorem].
180 2 Weight Filtrations on Log Crystalline Cohomologies
Proof. (1): We can show the well-denedness by the standard method in,
e.g., [64, 3.4] and by (2.12.3). The latter statement is obvious by (2.12.3).
(2): (2) is obvious by the denition (2.12.0.3).
(3): (2.12.4.2) is an isomorphism of complexes of Wn -modules by (2.6.1.1)
and the denition of the boundary morphism of the two log de Rham Witt
complexes in (2.12.4.2). The compatibility with the Frobenius endomor-
phisms is obtained by the same argument as that in [68, (9.3) (1)].
Let g : Y := (Y , M ) s be as in the beginning of this section.
By abuse of notation, we denote Y by Y . Let : (Y, M ) (Y, M)
be a closed immersion into a ne formally log smooth scheme over
(Spf(W ), W (L)), where W (L) is the canonical lift of L over Spf(W )
(cf. [46, (3.1)]). Let g : (Y, M) (Spf(W ), W (L)) be the structural
morphism. Let (DY (Y), MDY (Y) ) be the log PD-envelope of the closed
immersion (Y, M ) (Y, M). Set (Yn , Mn ) := (Y, M) W Wn ,
(DY (Yn ), MDY (Yn ) ) := (DY (Y), MDY (Y) ) W Wn and gn := g W Wn
(n Z>0 ). Let n : (Y, M ) (Yn , Mn ) be the induced natural closed
immersion. Let Wn (M ) be the canonical lift of M over Wn (Y ). Assume that
there exists an endomorphism of (Y, M) which is a lift of the Frobenius
morphism of (Y1 , M1 ). Then there exists a morphism
of log schemes which has been constructed in ([68, (7.17)]) by using a log
version of a lemma of Dieudonne-Cartier ([68, (7.10)]). In this book we only
review the denition of the morphism Wn (). As a morphism of schemes,
2.12 Filtered Log de Rham-Witt Complex 181
(2.12.4.4) Wn () (Mn ) m
(n (m), (s0 , . . . , sn1 ))
M (1 + V Wn1 (OY )) = Wn (M ).
Following [31], let us denote by iYn /Wn the sheaf of log dierential forms
of degree i on (Yn , Mn )/(Spec(Wn ), Wn (L)), and by Wn iY the Hodge-Witt
sheaf of log dierential forms of degree i on (Y, M )/s. The morphism Wn ()
induces a morphism
(2.12.5.1) s : OY x
(s0 , s1 , . . . , sn1 , . . .) W (OY ),
182 2 Weight Filtrations on Log Crystalline Cohomologies
m1 m2
where si s satisfy the following equations sp0 + psp1 + + pm1 sm1 =
m1
(x) (m Z>0 ) (e.g., [50, 0 (1.3.16)]). The morphism (2.12.4.5) for = 0
is induced by the following composite morphism
s mod V n W (OY1 )
OYn Wn (OY1 ) Wn (OY ).
n n1
tp0 + p
tp1 + + pn1
tpn1 H0 (ODY (Yn ) OYn Yn /Wn ),
p1 d log (m)
= p1 d log(m + p1 d log(1 + pa)
p (1 + pa)) = d log m
+ d( (1)i1 (pi1 /i)ai ).
= d log m
i=1
Hence
This immediately follows from (2.2.16) (2) and [46, (4.19)]=[68, (7.19)].
Proof. Let [ ] : OY a
(a, 0, . . . , 0) Wn (OY ) be the Teichmuller lift. By
noting that V Wn (OU ) is a nilpotent ideal sheaf of Wn (OU ) ([50, 0 (1.3.13)]),
we have a formula Wn (OU ) = [OU
] Ker(Wn (OU ) OU
). We claim
that
The inclusion is obvious. Let a be a local section on the left hand side.
Then the image of a in OY is 1 since OY j (OU ) is injective. Hence we
have a Wn (OY ) since V Wn (OY ) is a nilpotent ideal sheaf of Wn (OY ).
Therefore
Let us also consider the case of the log crystalline cohomology with com-
pact support.
Assume that Z = for the time being. Fix a total order on s only
in (2.12.7.1) below. In [64, Lemma 3.15.1], it is claimed that the following
sequence
in [64, 1.3.3] (we have to say that the turn of the tensor product in [64, 1.3.3]
is not desirable) is necessary for the proof of [64, Lemma 3.15.1]. However
the proof of the second isomorphism in [64, 1.3.3] is too sketchy. In [68, 6] we
have given a precise proof of the second isomorphism in [64, 1.3.3]. Hence we
can use [64, Lemma 3.15.1] without anxiety, and we identify Wn X ( log D)
with the following complex
in D+ (f 1 (Wn )).
We generalize the exact sequence (2.12.7.2) to the case Z
= as follows.
First assume that X is ane. Let (X , D Z) be a formal lift of (X, D Z)
over Spf(W ) with a lift : (X , D Z) (X , D Z) of the Frobenius of
(X, D Z). Let f: X Spf(W ) be the structural morphism. Set :=
X /W (log(Z D)), 1 := /p and = : . Then ( , )
satises the axioms of (6.0.1) (6.0.5) in [68], that is,
2.12 Filtered Log de Rham-Witt Complex 185
mod p
i i1
1
pi C
mod p
i Hi (1 ).
(2.12.8.1) Zp idM : Zp M Zp M
is a quasi-isomorphism.
(3) (cf. [52, III (1.5)]) Let i (resp. n) be a nonnegative (resp. positive)
integer. Then
(2.12.8.2)
pi { i | d pn+1 i+1 }
{ i | d pn i+1 }
.
pi+n { | d p } + p d
i i+1 i i1 pn i + pdi1
Proof. (1): We only remark that the proof is the same as that in [11, 8.8
Theorem].
(2): By the assumption, the complex MC( ) Zp M is acyclic.
(3): Set M := Z/pn in (2). Let be any gauge such that (i 1) = 1 and
(i) = 0. Then (2.12.8.1) at the degree i is equal to (2.12.8.2).
186 2 Weight Filtrations on Log Crystalline Cohomologies
Set
(2.12.8.3)
Zni := { i | d pn i+1 }, Bni := pn i + di1 , Wn i := Zni /Bni .
(2.12.8.4)
proj.
Wn+1 i = Zn+1
i i
/Bn+1 Zn+1
i
/(pn Z1i + di1 )
(pi )1
Zni proj. i
Zn /Bni = Wn i .
pn i + pdi1
d2 = 0, F dV = d, F V = V F = p ,
F p = pF, V p = pV, dp = pd, p = p = p.
We recall the following (cf. [50, I (3.31)], [50, I (3.21.1.5)], [62, (1.20)], [52,
II (1.2)], [62, (2.16)]):
Proof. Here we only remark that (3) is a formal consequence of (1) and (2)
(see [52, II (1.2)]).
2.12 Filtered Log de Rham-Witt Complex 187
(0) (1)
Wn 2X (log Z) Wn 2D(1) (log Z|D(1) ) Z zar (D/)
(2.12.10.1) d d
(0) (1)
Wn 1X (log Z) Wn 1D(1) (log Z|D(1) ) Z zar (D/)
d d
(0) (1)
Wn 0X (log Z) Wn 0D(1) (log Z|D(1) ) Z zar (D/)
d
C+ F(f 1 (Wn )), K+ F(f 1 (Wn )) and D+ F(f 1 (Wn )), and we denote it by
If one forgets the ltrations of both hands of (2.12.11.2), one can identify the
isomorphism (2.12.11.2) with the isomorphism
(2.12.11.3) D
Ru(X,DZ)/Wn (I(X,DZ)/W n
) Wn X (log(Z D))
Proof. (1): Let {Xi0 }i0 I0 be an ane open covering of X. Set Di0 := DXi0
and Zi0 := Z Xi0 . Then there exists an ane formal log scheme (Xi0 , Di0
Zi0 )i0 I0 over Spf(W ) such that each (Xi0 , Di0 Zi0 ) is a lift of the log scheme
(Xi0 , Di0 Zi0 ). The Frobenius morphism (Xi0 , Di0 Zi0 ) (Xi0 , Di0 Zi0 )
lifts to a morphism i0 : (Xi0 , Di0 Zi0 ) (Xi0 , Di0 Zi0 ). Then, using
{(Xi0 , Di0 Zi0 )}i0 I0 , we have a diagram of log schemes (X , D Z )I over
Spf(W ) as in 2.4. Using {i0 }i0 I0 , we have an endomorphism : (X , D
Z ) (X , D Z ) of a diagram of log schemes; i is a lift of the Frobenius
of (Xi , Di Zi ) W (i I). Let DX (X ) be the PD-envelope of the locally
closed immersion X X over (Spec(Wn ), pWn , [ ]). Then the morphism
induces a natural morphism DX (X ) DX (X ).
Set (X,n , D,n Z,n )I := (X W Wn , (D W Wn ) (Z W Wn ))I
and set ,n := mod pn . Then there exists a morphism (Wn (X ), Wn (D )
Wn (Z )) (X,n , D,n Z,n ) of diagrams of log schemes, where (Wn (X ),
Wn (D ) Wn (Z )) is a log scheme dened in (2.12.7). By (2.12.4.7), this
morphism induces a morphism
(2.12.11.4)
ODX (X,n ) OX,n X,n /Wn (log(D,n Z,n )) Wn X (log(D Z )).
2.12 Filtered Log de Rham-Witt Complex 189
(Note that (Wn (Xi ), Wn (Di ) Wn (Zi )) is the canonical lift of (Xi , Di Zi )
over Wn by (2.12.7); thus, by applying the ltered higher direct image of
the natural morphism zar : (X zar , f 1 (Wn )) (Xzar , f 1 (Wn )) to the
morphism in (2.12.11.4), we obtain a morphism which is equal to a special
case of a morphism dened in [46, (4.19)].)
The morphism (2.12.11.4) induces a ltered quasi-isomorphism with re-
spect to preweight ltrations. Indeed, the problem is local; in this case, it
follows from (2.12.6). By applying the ltered higher direct image of zar to
(2.12.11.4), we have an isomorphism (2.12.11.1). As in the proof of (2.6.1),
we can check that the morphism (2.12.11.1) is independent of the choice of
the open covering of X and the lift of each open scheme.
Let g : (X1 , D1 Z1 ) (X2 , D2 Z2 ) be a morphism of smooth schemes
with SNCDs over which induces morphisms (X1 , D1 ) (X2 , D2 ) and
(X1 , Z1 ) (X2 , Z2 ). Then, by the proof of [68, (9.3) (2)], g induces a
morphism
Using the diagram of log schemes, we see that the proof of the functoriality
of (2.12.11.1) is reduced to the local question on (Xi , Di Zi ) (i = 1, 2). In
this case, by the functoriality of the morphisms (2.12.4.3) and (2.12.4.6) and
by (2.12.6), we obtain the functoriality of (2.12.11.1).
In [68, (7.18)] we have proved that the morphism (2.12.4.6) is compatible
with two projections; as a result, the morphism (2.12.4.7) is also compatible
with them. In particular, we have the following commutative diagram
(2.12.11.4)
(ODX (X,n+1 ) OX,n+1 X,n+1 /Wn+1 (log(D,n+1 Z,n+1 )), P D )
proj.
(2.12.11.4)
(ODX (X,n ) OX,n X,n /Wn (log(D,n Z,n )), P D )
(2.12.11.5)
ODX (X,n ) OX,n X,n /Wn (log(Z,n D,n )) Wn X (log(Z D )).
(0)
(0)
(Rzar (OD (0) O (log Z,n |D(0) )) Z zar (D/) )
(0) (D,n ) (0)
D,n D,n /Wn ,n
D
(Rzar (Wn X (log Z )) Z zar (D/) ).
(0)
D
Res
j (Wn i1
D (log Z|D ))(1) 0.
The spectral sequences (2.12.15.1) for ns are compatible with the projections.
The boundary morphism
192 2 Weight Filtrations on Log Crystalline Cohomologies
(2.12.15.2) dk,hk
1 : H hik (D(k) , Wn iD(k) (log Z|D(k) ) Z zar
(k)
(D/))
E1k,h+k = H hi (D(k) , W ik
D (k)
Z zar
(k)
(D/))(k)
= H hi (X, W iX (log D)),
So far we have worked over a base scheme whose structure sheaf is killed by
a power of p. We can also work over a (not necessarily ane) P -adic base
in the sense of [11, 7.17 Denition], and the analogues of results in previous
sections hold in this case.
Let V be a complete discrete valuation ring of mixed characteristics with
perfect residue eld of characteristics p > 0. Let W be the Witt ring
of with fraction eld K0 . Let K be the fraction eld of V . For a V -
module M , MK denotes the tensor product M V K. Unless otherwise
stated, from this section to 2.19, S denotes a p-adic formal V -scheme in
the sense of [74, 1], i.e., S is a noetherian formal scheme over V with the
p-adic topology such that, for any ane open formal subscheme U , there ex-
ists a surjective morphism V {x1 , . . . , xn } (U, OU ) of topological rings
for some n. Let f : (X, D Z) S denote a proper smooth morphism
of p-adic formal V -schemes (e.g., V /p-schemes) of nite type with relative
transversal SNCD. Following [74], for a p-adic formal scheme T /Spf(V ), set
T1 := SpecT (OT /pOT ).
By virtue of results in previous sections, we can give the compatibil-
ity of the weight ltrations on log crystalline cohomologies as convergent
F -isocrystals with some canonical operations, e.g., the base change, the
Kunneth formula, the functoriality. Later, in 2.19, we shall give the compat-
ibility of them with the Poincare duality.
2.13 Filtered Convergent F -isocrystal 193
Proof. The base change theorem (2.10.3) and the argument in [74, (3.1)] show
the existence of a p-adically convergent isocrystal Ekh .
As in the proof of [74, (3.7)], we may assume that V = W ; furthermore,
by the log version of [74, (3.4)], we may assume that pOS = 0. The spectral
sequence in (2.9.6.3) for
DT 1 log,ZT1
Rh f(XT1 ,ZT1 )/T (Pk Ecrys (O(XT1 ,DT1 ZT1 )/T ))
Corollary 2.13.5. Let k, h be two nonnegative integers. For any p-adic en-
largement T of S/V ,
DT 1
(2.13.5.1) Pk Rh f(XT1 ,DT1 ZT1 )/T (O(XT1 ,DT1 ZT1 )/T )K :=
DT 1 log,ZT1
Im(Rh f(XT1 ,ZT1 )/T (Pk Ecrys (O(XT1 ,DT1 ZT1 )/T ))K
(2) K
unneth formula
Theorem 2.13.7. Let (Xj , Dj Zj ) (j = 1, 2) be a log scheme stated in the
beginning of this section. Let (X3 , D3 Z3 ) be the product (X1 , D1 Z1 ) S
(X2 , D2 Z2 ) in the category of ne log schemes. Then the there exists the
following canonical isomorphism
(2.13.7.1) Ri f (O(X1 ,D1 Z1 )/K )OS/K Rj f (O(X2 ,D2 Z2 )/K )
i+j=h
Using (2.11.11) and (2.11.19), we obtain the following as in (1) and (2).
Theorem 2.13.8. Let k, h be two nonnegative integers.
h
(1) There exists a convergent F -isocrystal Ek,c on S/V such that
h DT 1
(Ek,c )T = Pk Rh f(XT1 ,DT1 ZT1 )/T ,c (O(XT1 ,DT1 ZT1 ;ZT1 )/T )K
Rh f,c (O(X,DZ;Z)/K )T = Rh f(XT1 ,DT1 ZT1 )/T ,c (O(XT1 ,DT1 ZT1 ;ZT1 )/T )K
Rh f,c (O(X3 ,D3 Z3 ;Z3 )/K )
(4) Functoriality
Proof. (1) and (2) immediately follow from (2.9.1) and (2.11.18), respectively.
= Rh f (O(X,DZ)/K ),
k,hk
(2.13.11.2) E1,c ((X, D Z)/K)
= Rhk f (O(D(k) ,Z| )/K Z (k)log (D/K; Z))
D (k)
= Rh f,c (O(X,DZ;Z)/K ).
Let p be a prime number. Let T be a noetherian formal scheme with
an ideal sheaf of denition aOT , where a is a global section of (T , OT ).
Assume that there exists a positive integer n such that pOT = an OT . Let
T be a ne formal log scheme with underlying formal scheme T . Assume
that OT is a-torsion-free, that is, the endomorphism a idOT EndOT (OT )
is injective, and that the ideal sheaf aOT has a PD-structure . We call
T = (T, aOT , ) above an adic ne formal log PD-scheme. We dene the
notion of a morphism g : T T of adic ne formal log PD-schemes in the
following way: the morphism g is nothing but a morphism of formal ne log
PD-schemes, and T is a -adically complete and separated and a -torsion-free,
where a := g (a). In this section we assume that, for each ane open set
Spf(R) of T , aR is a prime ideal and that the localization ring Ra at the
ideal aR is a discrete valuation ring.
Let H be an OT -module of nite type. Since Ra is a PID, there exists
a non-empty open log formal subscheme T of T such that there exists an
isomorphism H|T OTr Htor , where Htor is a direct sum of OT -modules
OT /ae (e Z>0 ) (Delignes remark ([49, (3.10)])). Let E be an a-torsion-free
OT -module. Then, as in [68, (3.1)], it is easy to see that
(2.14.0.1) Tor O
r
T (H| , E| ) = 0
T T (r Z>0 )
and
1
(2.14.0.2) Tor gr (OT )
(g 1 (H|T ), OT ) = 0 (r Z>0 )
(2.14.0.3) Tor O
r
T (Rh f
XT /T (OXT /T ), E|T ) = 0 (r Z>0 )
1 1
these facts, we see that the log version of the proofs of [11, (7.8), (7.13), (7.16),
(7.24)] and [74, (3.3)] work for the coecient IX/S . Hence Rh fX/T ,c (OX/T )
is a perfect complex of OT -modules and it satises the base change property.
Therefore, if T is suciently small, we have
(2.14.0.4) Tor O
r
T (Rh f
XT /T ,c (OXT /T ), E|T ) = 0 (r Z>0 , h Z).
1 1
is an isomorphism of OT -modules.
(2) The canonical morphism
is an isomorphism of OT -modules.
DT
Pk 1
Rh f(X,DZ)T /T (O(X,DZ)T /T )
1 1
is an isomorphism.
(2) The canonical morphism
DT
g Pk 1
Rh f(X,DZ)T /T ,c (O(X,DZ;Z)T /T )
1 1
2.15 The E2 -degeneration of the p-adic Weight Spectral Sequence of a Variety 199
DT
Pk 1
Rh f(X,DZ)T /T ,c (O(X,DZ;Z)T /T )
1 1
is an isomorphism
(2.14.2.1)
E1k,h+k = Rhk f(D(k) ,Z| ) /T (O(D(k) ,Z| ) /T
D (k) T1 D (k) T1
Z crys
(k)log
(DT1 /T ; ZT1 ))(k)
= Rh f(X,DZ)T /T (O(X,DZ)T /T ),
1 1
(2.14.2.2)
E1k,h+k = Rhk f(D(k) ,Z| ) /T (O(D(k) ,Z| ) /T
D (k) T1 D (k) T1
Z crys
(k)log
(DT1 /T ; ZT ))(k)
= Rh f(X,DZ)T /T (O(X,DZ)T /T ).
1 1
T or O
r
T (Rh f
(D (k) ,Z| ) /T (O(D(k) ,Z| ) /T ), E|T ) = 0
D (k) T1 D (k) T1
for any OT -module E without a-torsion and for all r Z>0 . Hence we have
an isomorphism
g 1 E1k,h+k ((X, DZ)T 1 /T )g1 (OT ) OT E1k,h+k ((X, DZ)T1 /T )
have used a somewhat tricky argument in [68, (5.2)] (cf. [68, (3.2), (3.4), (3.5),
(3.6)]) based on Delignes remark ([49, 3.10]). Though we also use Delignes
remark in this book, the proof in this section is not tricky by virtue of the
existence of the weight spectral sequences (2.9.6.2) and (2.11.14.3) over a
general base (cf. [68, (3.7)]).
Let (X, D) be a proper smooth scheme with an SNCD over . By [40,
3, (8.9.1) (iii), (8.10.5)] and [40, 4, (17.7.8)] , there exist a smooth ane
scheme S1 over a nite eld Fq and a model (X , D) of (X, D) over S1 .
By a standard deformation theory ([41, III (6.10)]), there exists a formally
smooth scheme S such that S W (Fq ) Fq = S1 . Let T be an ane open
subscheme of S, and set T1 := T W (Fq ) Fq . Take a closed point t of T1 .
The point t is the spectrum of a nite eld t . We x a lift FT : T T
of the Frobenius(=p-th power morphism) FT1 of T1 . Then we have the Te-
ichm uller lift (T, OT ) W (t ) (resp. (T, OT ) W ) of the morphism
(T1 , OT1 ) t (resp. (T1 , OT1 ) ) (e.g., [50, 0 1.3]). The rings W (t )
and W become (T, OT )-algebras by these lifts.
To prove the E2 -degenerations, we prove some elementary lemmas.
Let A be a p-adically complete and separated p-torsion-free ring with a
lift f of the Frobenius endomorphism of A1 := A/p. Then there exists a
unique section : A W (A) of the projection W (A) A such that
f = F , where F is the Frobenius of W (A) (e.g., [50, 0 (1.3.16)]). This
morphism induces morphisms : A W (A1 ) and n : A/pn Wn (A1 ).
Then the following holds:
Proof. For a W (Fq )-module M , MK0 (Fq ) denotes M W (Fq ) K0 (Fq ). First
we prove (2.15.4) for (2.9.6.2). Replace T by a suciently small ane
open sub log formal scheme in order that, for any h, k Z, r Z>0 ,
Erk,h+k ((XT1 , DT1 )/T ) has the form OTn N (n N), where N is a di-
rect sum of modules of type OT /pe (e Z>0 ). Then we have
1
Tor gs (OT )
(g 1 Erk,h+k ((XT1 , DT1 )/T ), OT ) = 0 (s Z>0 )
dk,h+k
r ((XT1 , DT1 )/T )K0 (Fq ) : Erk,h+k ((XT1 , DT1 )/T )K0 (Fq )
Erk+r,h+kr+1 ((XT1 , DT1 )/T )K0 (Fq )
dk,h+k
r ((Xt , Dt )/W (t ))
=dk,h+k
r ((XT1 , DT1 )/T ) OT W (t )
HomW (t ) (W (t )n , W (t )n ) (N (T,OT ) W (t )).
By the purity of the weight [15, (1.2)] or [68, (2.2) (4)], we have
dk,h+k
r ((Xt , Dt )/W (t ))K0 (Fq ) = 0 for any closed point t of T1 , that is,
dk,h+k
r ((Xt , Dt )/W (t )) is contained in N (T,OT ) W (t ). From this
and (2.15.3), we see that dk,h+k r ((XT1 , DT1 )/T ) is contained in N . Hence
dk,h+k
r ((X T1 , D T1 )/T )K0 (Fq ) = 0.
The proof of the degeneration of (2.11.14.3) is the same as the above. (One
may use the duality between (2.9.6.2) W K0 and (2.11.14.3) W K0 for the
case Z = and S = Spf(W ).)
(Rf(X,DZ)/T (O(X,DZ)/T ) L
Z Q, P ),
D
(2.16.1.2) gc : (Rf(X
,D Z )/T ,c (O(X ,D Z ;Z )/T ) Z Q, Pc
L D
)
(Rf(X,DZ)/T ,c (O(X,DZ;Z)/T ) L
Z Q, Pc ),
D
(Rf(X,DZ)/T (O(X,DZ)/T ) L
Z Q, P ) =
D
(Rf(X ,D Z )/T (O(X ,D Z )/T ) L
Z Q, P
D
)
and
(Rf(X,DZ)/T ,c (O(X,DZ;Z)/T ) L
Z Q, Pc ) =
D
(Rf(X ,D Z )/T ,c (O(X ,D Z ;Z )/T ) L
Z Q, Pc ).
D
(2.16.4.1)
h
Hlog-crys ((X , D Z )/W ())K H h (XK , XK /K (log(DK ZK )))
h
= HdR (UK /K),
(2.16.4.2)
h
Hlog-crys,c ((X , D Z ; Z )/W ())K H h (XK , XK /K (log(ZK DK )))
which are compatible with the weight ltrations with respect to D and DK .
See also [17] for analogous statements by the rigid analytic method in the
case Z = .
(2) Let (X, D) be a proper smooth scheme with a relative SNCD
over . Set U := X \ D. By the nite base change theorem ([5, (1.8)])
and by Shihos comparison theorems [82, Theorem 2.4.4, Corollary 2.3.9,
Theorem 3.1.1]), there exists a canonical isomorphism Hrig h
(U/K)
h
Hlog-crys ((X, D)/W ) W K. As a result, Hrig
h
(U/K) has a weight ltration.
2.17 The E2 -degeneration of the p-adic Weight Spectral Sequence of a Family 205
By [85], [82, Theorem 2.4.4, Corollary 2.3.9, Theorem 3.1.1] and [6, (2.4)],
h
we obtain Hlog-crys,c ((X, D)/W ) W K = Hrig,c h
(U/K). In particular,
h
Hrig,c (U/K) has a weight ltration.
If (X, D) is the special ber of (X , D) in (1), there exists a weight-ltered
isomorphism Hrig h
(U/K) HdR h
(UK /K). An analogous statement can be
found in [17].
(3) Let U be a separated scheme of nite type over . Let Z/ be a closed
subscheme of U . In [70] the rst-named author has dened a nite increasing
h
ltration on Hrig,Z (U/K) which deserves the name weight ltration. In
h
particular, the weight ltration on Hrig (U/K) dened in (2) is independent
of the choice of (X, D). See 3.4 below for more details. In [loc. cit.] he has
h
also dened a nite increasing ltration on Hrig,c (U/K) which deserves the
name weight ltration in the case where U is embeddable into a smooth
scheme over as a closed subscheme. See also 3.6 below for more details.
(2.17.2.3) k,h+k
l(Rh f(XS ,DS )/S (O(XS ,DS )/S )K ) = l( E2 ((XS1 , DS1 )/S )K ).
1 1 1 1 k
(2.17.2.4)
dimK (Rh f(XS ,DS )/S (O(XS ,DS )/S )K )
k,h+k
1 1 1 1
(2.17.2.5)
dimK (Rh f(XV ,DV )/V (O(XV ,DV )/V )K )
1 1 1 1
k,h+k
= dimK ( E2 ((XV1 , DV1 )/V )K ).
k
(2.17.2.7)
E2k,h+k ((XV1 , DV1 )/V ) V K E2k,h+k ((X , D )/W )W K .
dk,h+k
r,(n) : Erk,h+k ((X(S(n) )1 , D(S(n) )1 )/S(n) )K
Erk+r,h+kr+1 ((X(S(n) )1 , D(S(n) )1 )/S(n) )K
d
r BK (lim B /mn ) = lim(d B /mn ) = 0.
n K r BK K
n
Since (BK )m is a Zariski ring, limn (BK )m /mn (BK )m is faithfully at over
(BK )m ([13, III 3 Proposition 9]). Therefore d r BK (BK )m = 0. Since m is
an arbitrary maximal ideal of BK , d r = 0 (r 2). Hence we have proved
(2.17.2) for (2.9.6.2) V K.
Next we prove (2.17.2) for (2.11.14.3)V K for the case Z = and S0 = S1 .
As we remarked before (2.14.0.4), we have the base change property for
Rq f(X,D)/S,c (O(X,D)/S )K = (Rq f(X,D)/S I(X,D)/S ) V K. Hence the proof
is analogous to the proof of (2.17.2) for (2.9.6.2)V K for the case Z = and
S0 = S1 : we have only to use (2.16.2) for Rf(X,D)/S,c (O(X,D)/S )K , (2.11.17)
and use [68, (5.2) (2)] (=(2.15.4) for (2.11.14.3)).
2.17 The E2 -degeneration of the p-adic Weight Spectral Sequence of a Family 209
(XT ,DT )/T )K . The claim on the F -isocrystal follows as in [74, (3.7)].
(3): (3) immediately follows from (2.17.2).
We can reprove (2.13.8) (1) and (2) in the case Z = and more:
(Pk Rh f,c (O(X,D)/K ))T = Pk Rh f(XT1 ,DT1 )/T ,c (O(XT1 ,DT1 )/T )K
k,hk
(2.17.4.1) E1,c ((X, D)/K) = Rhk f (OD(k) /K Z (k) (D/K))
= Rh f,c (O(X,D)/K ).
Proof. (1), (2), (3): We obtain (1), (2) and (3) as in (2.17.3).
degenerates at E3 .
(2.18.1.1) log
gcrys h
: Hlog-crys (X/W )K Hlog-crys
h
(X /W )K
(2.18.1.2) log
gcrys,c h
: Hlog-crys,c (X/W )K Hlog-crys,c
h
(X /W )K
Proof. (1): In this proof, for the sake of clarity, denote by P and P the weight
h
ltrations on Hlog-crys h
(X/W )K0 and Hlog-crys (X /W )K0 , respectively.
Since Pk Hlog-crys (X/W )K0 K0 K = (Pk Hlog-crys (X/W ))K (k Z {}),
h h
(2.18.1.4) Pk P
= Pk .
By [40, 3, (8.9.1) (iii), (8.10.5)] and [40, 4, (17.7.8)], there exists a model of
g, that is, there exists a morphism g : (X , D ) (X , D) of proper smooth
schemes with relative SNCDs over the spectrum S1 := Spec(A1 ) of a smooth
algebra A1 ( ) over a nite eld Fq such that g A1 = g. By a standard
deformation theory ([41, III (6.10)]), there exists a formally smooth scheme
S = Spf(A) over Spf(W (Fq )) such that S W (Fq ) Fq = S1 . We x a lift
F : S S of the Frobenius of S1 . Then, as in 2.15, W is an A-algebra. Let
P and P be the analogous ltrations on Rh fX /S (OX /S ) W (Fq ) K0 (Fq ),
where K0 (Fq ) is the fraction eld of W (Fq ). By (2.14.2), in order to prove
(2.18.1.4), it suces to prove that
(2.18.1.5) Pk P
= Pk
(2.18.1.6) (Pk P
)s = (Pk )s
for any closed point s S by [74, (3.17)]. In this case, (2.18.1.6) immediately
follows from the purity of the weight of the crystalline cohomologies ([15,
(1.2)] or [68, (2.2) (4))]) and by the spectral sequence (2.9.6.2). Thus we have
proved (1).
(2): By the assumption of g, the analogue of (2.18.1.3) for the log crys-
talline cohomology with compact support holds. Using (2.13.8) instead of
(2.13.3), we obtain (2) in a similar way.
(2.18.2.1)
g : Rh f(X,D)/S (O(X,D)/S )K Rh f(X
,D )/S (O(X ,D )/S )K (h Z)
Proof. Since the proofs of (1) and (2) are similar, we give only the proof of
(1).
By (2.13.3) (or (2.17.3)) and by the proof of [74, (3.17)], we may assume
that S is the formal spectrum of a nite extension V of V . Let be the
residue eld of V . As mentioned in the proof of (2.17.2), V /p is an -algebra;
the two pairs (X, D) and ((X, D) V ) V /p are two deformations of
(X, D) V ; the obvious analogue for (X , D ) also holds. Hence, by the
deformation invariance of log crystalline cohomologies with weight ltrations
((2.16.2)), we may assume that S = Spf(W ( )) and that (X, D) and (X , D )
are smooth schemes with SNCDs over a perfect eld of characteristic
p > 0. Hence (1) follows from (2.18.1) (1).
Proof. We may assume that S is at over Spf(V ). By the second proof of [65,
(2.2)] and by [loc. cit., (2.4)], the induced morphism
is an isomorphism (cf. [43, VII (3.5)], (2.18.7) below). By the log version of
a triangle in the proof of [11, 7.16 Theorem] and by the log version of [11,
7.22.2], the induced morphism
tration.
2.18 Strict Compatibility 213
Let n be a positive integer, and set Sn := SpecS (OS /pn ). Then we have
an exact sequence
By using the base change theorem of the log crystalline cohomology sheaf
with compact support ((2.11.11.1)), we have the following triangle as in [11,
7.16 Theorem]:
Remark 2.18.7. The following example (=a very special case of [65, (2.3)])
shows that the strictness of the induced morphism on sheaves of log dieren-
tial forms by a morphism of smooth schemes with relative SNCDs does not
hold.
Let S be a scheme and let X be an ane plane A2S = SpecS (OS [x, y]).
Let D be a relative SNCD on X/S dened by xy = 0. Let g : X X
be the blow up of X along the center (0, 0). Let D be the union of
the strict transform of D and the exceptional divisor of g; then D is a
relative SNCD on X /S. Let i be an integer. Then Mokrane has proved that
214 2 Weight Filtrations on Log Crystalline Cohomologies
Applying the higher direct image functor R f X/S to (2.18.8.1), we have the
following exact sequence
(2.18.8.2)
Rh2 fD/S (OD/S )(1) Rh fX/S (OX/S )
Rh f(X,D)/S (O(X,D)/S ) .
(2.18.8.3)
Rh fX/S (OX/S ) Rh fD/S (OD/S )
Rh+1 f(X,D)/S,c (O(X,D)/S ) .
Now assume that S is a p-adic formal V -scheme (in the sense of [74, 1])
over a complete discrete valuation ring V of mixed characteristics with perfect
residue eld. Assume also that S0 = SpecS (OS /p), that X is projective over
S0 of pure relative dimension d and that D is a smooth hypersurface section.
Let K be the fraction eld of V . Then the induced morphism
2.19 The Weight-Filtered Poincare Duality 215
Then Rd1 fD/S,ev (OD/S )K (1) is the orthogonal part of the image of the
injective morphism Rd1 fX/S (OX/S )K Rd1 fD/S (OD/S )K . Therefore
we have the following direct decomposition:
(2.18.8.5)
Rd1 fD/S (OD/S )K = Rd1 fD/S,ev (OD/S )K Rd1 fX/S (OX/S )K .
which is strictly compatible with the weight ltration. That is, the natural
morphism
(2.19.1.2)
Rh f,c (O(X,D)/K ) Hom OS/K (R2dh f (O(X,D)/K ), OS/K (d))
(2.19.1.3)
Rh f,c (O(X,D)/K ) R2dh f (O(X,D)/K ) R2d f,c (O(X,D)/K )
Tr
Let S be a scheme. Let (X, D)/S be a proper smooth scheme with a relative
SNCD. Set U := X \ D and let f : U S be the structural mor-
phism. Let f (k) : D(k) S (k Z0 ) be the structural morphism and
a(k) : D(k) X also the natural morphism. Let l be a prime number
(k)
which is invertible on S. Let et (D/S)(k) (k N) be the etale orien-
k
tation sheaf of D(k) : et (D/S)(k) := {u1 ( (M (D)/OX
(k)
))}|D(k) , where
et
u is the canonical morphism Xet Xzar of topoi. Here note that we do
(k)
not dene et (D/S). If S is of characteristic p > 0, then the Frobenius
(k)
of (X, D) acts on et (D/S)(k) by the multiplication by pk . Almost all
the results in the previous sections have l-adic analogues. For example, the
excision spectral sequence
(k) (k)
(2.20.0.1) E1k,hk = Rhk f (Ql (k) Z et (D/S)(k)) = Rh f,c (Ql ).
(k) (k)
(2.20.0.2) Rk j (Ql ) a (Ql,D(k) Z et (D/S)(k))
(1)k
(k) (k)
a (Ql,D(k) Z et (D/S)(k)).
In the rest of this book, following the suggestion of one of the referees, we
state some results to answer natural questions arising from results in previous
sections: we outline the existence of the weight ltration on the rigid cohomol-
ogy of a separated scheme of nite type over a perfect eld of characteristic
p > 0 without proofs. In this section we gather general objects needed in the
later sections.
First we have to x a convention on the sign of the boundary morphism of
the single complex of a multicosimplicial complex in [24] and [70]. We follow
the convention in [70], which is dierent from that in [24].
Let (T , A) be a ringed topos. For a positive integer r, let (Tt1 tr , At1 tr )t1 ,
...,tr N be a constant r-simplicial ringed topos dened by (T , A): Tt1 tr
= T , At1 tr = A. Let M be an object of the category C(A ) (r-
points) of complexes of A -modules. For nonnegative integers t1 , . . . , tj
(1 j r), set tj := t1 + + tj and t0 := 0. Set also t := (t1 , . . . , tr )
t
1 +1 t
2 +1
(3.1.0.3) s : D+ (A ) D+ (A)
Then we have
(P )
(3.1.0.7) grk (s(M )) = grP t
tr +k M [tr ].
t0
3.1 Preliminaries for Later Sections 221
t0 t0
Assume that the two ltrations (P )s on s(M ) and s(N ) are exhaustive and
complete, that grP tr +k M
t
and grPtr +k N
t
are quasi-isomorphic to objects of
+
C (A ) and that the spectral sequence arising from the two ltrations (P )s
on s(M ) and s(N ) are regular and bounded below. Here we say that (P )
is complete if s(M ) = limk s(M )/((P )k (s(M ))) (cf. [90, (5.4.4)]). Then
we have the following convergent spectral sequence by the Convention (6)
(cf. [90, (5.5.10)], [24, (8.1.15)]):
(3.1.0.12)
E1k,h+k = Hhtr (grP
tr +k M )
t
Hhtr 1 (grP t
tr +k+1 N )
t0 t0
= Hh (s(MF())).
Let
d(M ) : Hhtr (grP
tr +k M )
t
Hhtr +1 (grP t
tr +k1 M )
t0 t0
Let
d(N ) : Hhtr 1 (grP
tr +k+1 N )
t
Hhtr (grP t
tr +k N )
t0 t0
d : E1k,h+k E1k+1,h+k
Hhtr +1 (grP
tr +k1 M )
t
Hhtr (grP t
tr +k N ).
t0 t0
the set of the morphisms in D+ F(A ). Then we have the following functor
(3.1.0.14)
(s(MF), MF ) : Mor(D+ F(A )) [( : (M, P ) (N, P ))]
[(s(MF()), MF (P ))] D+ F(A).
: (M, ) (N, )
Then
(3.1.0.16) gri s(MF()) = M ti [tr ]{i} N ti [tr 1]{i}
t0 t0
(3.1.0.18)
e1
n1 n2 nu1 nu : D (A ) [M ]
(3.1.0.20) XN CN (Y )N
(F , P ) F ((F , P ))
(3.1.5.1)
s(F , P ) F (s(F , P )).
226 3 Weight Filtrations and Slope Filtrations on Rigid Cohomologies (Summary)
and
h
(3.2.0.3) Hrig (U/K) := H h (Rrig (U/K)).
h h
We call Hrig (U/K, C) and Hrig (U/K) the rigid cohomology of C and the
rigid cohomology of U/K, respectively. By [88, (6.4.1)] there exists an integer
c such that
h
(3.2.0.4) Hrig (U/K, C) = 0 for all h > c.
s e1
D+ (K) D+ (K N )
t
D+ (K).
N
Then there exists an object Crig ((UN /U )/K, C) in D+ (K N ) such that
there exist canonical isomorphisms
N
(3.2.1.1) Rrig (U/K, C) c s(Crig ((UN /U )/K, C))
and
(3.2.1.2)
N
Rrig (Ut /K, C t ) e1
t (Crig ((UN /U )/K, C)) (0 t N ).
N
The object Crig ((UN /U )/K, C) is functorial for a morphism of aug-
mented simplicial schemes UN U s and a morphism of overconvergent
isocrystal Cs. Furthermore, for N N , there exists a canonical
isomorphism
(3.2.1.3)
N N
s{(Crig ((UN /U )/K, C))N } s(Crig ((UN /U )/K, C)).
Here N on the left hand side of (3.2.1.3) means the N -truncation of the
rst cosimplicial degree.
N
The denition of Crig ((UN /U )/K, C) is as follows.
By (3.1.5) there exists an N -truncated proper hypercovering (UN , U N )
of (U, U ). The pair (UN U Z, U N U Z) over (Z, Z) is an N -truncated
proper hypercovering of (Z, Z). Let (VN , V N ) be a renement ([87,
(4.2.1)]) of the proper hypercovering (UN U Z, U N U Z) of (Z, Z)
such that there exists a closed immersion V N PN into a separated
formally smooth p-adic formal V-scheme ([87, (4.2.3)]). Consider the N -
truncated triple
(3.2.1.4) (VN , V N , V
N (PN )N ).
(3.2.1.5)
V
(Vmn , V mn , Qmn ) :=(coskU
0 (Vm )n , cosk0 (V m )n , cosk0 (Qm V Z)n )
m Um
(0 m N, n N)
Let V, , K, K0 , W and j : U U be as in 3.2. In this section we state
comparison theorems between Rrig (U/K), the log convergent cohomologial
complex of a certain split proper hypercovering of (U, U ) and the log crys-
talline cohomological complex of it.
The following is an easy corollary of de Jongs alteration theorem ([28,
(4.1)]):
Using (3.3.1) and a general formalism in [42, Vbis ], we have the following:
3.3 Comparison Theorems 229
([42, Vbis (5.1.1)], [24, (6.2.2)]). Here [n] [m] means a surjection in .
(Un , Yn ) is split.
(3.3.4.6) R(((Y, conv , K(Y,E)/V ) Rrig (V /K).
E)/V)log
R(((Y,
(3.3.4.7) conv , K(Y,E)/W ) Rlog-crys ((Y, E)/W )K0 .
E)/W )log
R(((X
(3.3.5.1) conv , K(X ,D )/V ) Rrig (U/K).
, D )/V)log
(V , Y ) (U , X )
(V, V ) (U, U ),
Set lm := coskV
0 (l )m . Then we have a closed immersion
which produces log naive rigid cohomologies which are trisimplicial version
of the log naive rigid cohomologies dened in [82, Denition 2.2.12] (cf. [4,
p. 14]) (the terminology naive rigid cohomology has been used in [4, p. 14],
while the terminology log analytic cohomology has been used in [82] instead
of the terminology log naive rigid cohomology). Then, by the cohomological
descent and by the proof of [82, Corollary 2.3.9] and [82, Theorem 2.4.4], there
exists a natural morphism
R((XN, , DN, )/K, LDR(jN, O]XN, [PN, ))
:= R(]XN, [PN, , jN, ]XN, [P ).
N,
= h R(((XN
conv , K(XN ,DN )/V ) h Rrig (U/K).
, DN )/V)log
By (3.2.0.4) for the trivial coecient, there exists an integer c such that
i
Hrig (U/K) = 0 for i > c. Hence, by (3.3.5.7),
(3.3.5.8)
R(((X
conv , K(X ,D )/V ) = c R(((X , D )/V)conv , K(X ,D )/V ).
, D )/V)log log
E1ij = Hcrys
j i+j
((Xi , Di )/W ) = Hlog-crys ((X , D )/W )).
h
Remark 3.3.7. (1) In [5] Berthelot has proved that Hrig (U/K) is nite di-
mensional if U is smooth over . In [36] Groe-Kl onne has rst proved that
h
Hrig (U/K) is nite dimensional. In [87] Tsuzuki has given an alternative proof
h
of the nite dimensionality of Hrig (U/K) by using his proper cohomological
descent in rigid cohomology and by reducing the niteness to Berthelots
result above.
(2) The right hand side of (3.3.6.1) depends only on U/ and K; this solves
a problem raised in [28, Introduction] for the split case in a stronger form.
(3) In [1], in a dierent method from ours, Andreatta and Barbieri-Viale
1
have proved that Hlog-crys ((X , D )/W ) depends only on U/ for the case
where p 3 and where the augmentation morphism X0 \ D0 U is gener-
ically etale, even for the non-split case.
We can generalize (3.3.6) as follows.
Let : V U be a morphism of separated schemes of nite type over .
It is easy to see that there exists an open immersion V V into a proper
234 3 Weight Filtrations and Slope Filtrations on Rigid Cohomologies (Summary)
(3.4.0.2)
(X ,D Z ,Z )/S : (((X , D Z )/S)crys , O(X ,D Z )/S )
log
(3.4.0.3) log,Z
(Ecrys (O(X ,D Z )/S ), P D ) := ((X ,D Z ,Z )/S (I ), )
(3.4.0.4) log,Z
(Ezar (O(X ,D Z )/S ), P D ) := u(X ,Z )/S ((I , P D ))
in (2.7.3.1) and (2.7.4) in D+ F(O(Xt ,Zt )/S ) and D+ F(ft1 (OS )), respectively,
where ft : Xt S is the structural morphism.
In the case Z = , we denote
log,Z
(Ecrys (O(X ,D Z )/S ), P D ) and log,Z
(Ezar (O(X ,D Z )/S ), P D )
by
(Ecrys (O(X ,D )/S ), P ) and (Ezar (O(X ,D )/S ), P ),
respectively.
Now, let the notations be as in (3.3.7.2). Let f : X Spf(W ) and
g : Y Spf(W ) be the structural morphisms. From this section, assume
that is perfect unless stated otherwise. Consider the natural morphism
O(X ,D )/W logcrys (O(Y ,E )/W ). By the nal note in 3.1, we have
asque resolutions I(X ,D )/W
and I(Y ,E )/W
of O(X ,D )/W and O(Y ,E )/W ,
respectively, with a natural morphism
I(X ,D )/W
log
crys (I(Y ,E )/W ).
(3.4.0.5)
((X ,D )/W (I(X ,D )/W
), ) ((X ,D )/W log
crys (I(Y ,E )/W ), )
(crys (Y ,E )/W (I(Y ,E )/W
), )
crys ((Y ,E )/W (I(Y ,E )/W
), )
crys ((J , P )).
236 3 Weight Filtrations and Slope Filtrations on Rigid Cohomologies (Summary)
(s(MF(
)), MF (P )).
(3.4.0.8)
grkMF s(MF(
(P )
))
= t+k Ezar (O(Xt ,Dt )/W )[t]
grP t+k+1 Ezar (O(Yt ,Et )/W )[t 1]
grP
t0 t0
(t+k)
= at,zar RuD(t+k) /W (OD(t+k) /W Z crys
(t+k)
(Dt /W ))
t t
t0
{t k}[t]((t + k))
(t+k+1)
bt,zar RuE (t+k+1) /W (OE (t+k+1) /W Z crys
(t+k+1)
(Et /W ))
t t
t0
(3.4.1.1)
E1k,h+k =
(t+k)
H h2tk ((Dt /W )crys , OD(t+k) /W
t
t0
Z crys
(t+k)
(Dt /W ))((t + k))K
3.4 Weight Filtration on Rigid Cohomology 237
(t+k+1)
H h2tk2 ((Et /W )crys , OE (t+k+1) /W
t
t0
Z crys
(t+k+1)
(Et /W ))((t + k + 1))K
=H h (MF(rig )).
Remark 3.4.3. By using (3.1.0.13), we can give the explicit description of the
boundary morphism dk,h+k
1 of the spectral sequence (3.4.1.1).
V U
v
u
V U
be a commutative diagram of separated schemes of nite type over . Then
the induced morphism
238 3 Weight Filtrations and Slope Filtrations on Rigid Cohomologies (Summary)
(u , v ) : H h (MF(rig )) H h (MF(rig )) (h Z)
and
h
(3.4.5.2) Hrig,Z (U/K) := H h (Rrig,Z (U/K)).
We call Rrig,Z (U/K) the rigid cohomological complex of U/K with support
h
in Z, and call Hrig,Z (U/K) the rigid cohomology of U/K with support in Z
([5, (2.3)]).
As an immediate application of (3.4.1) and (3.4.4), we obtain the following:
Corollary 3.4.6 ([70, (16.1)]). (1) Let U U be an open immersion into
a proper scheme over . Let V be the closure of V in U . Let the notations be
as in (3.3.7.2). Then there exists the following spectral sequence
(3.4.6.1)
E1k,h+k =
(t+k)
H h2tk ((Dt /W )crys , OD(t+k) /W
t
t0
Z crys
(t+k)
(Dt /W ))((t + k))K
(t+k+1)
H h2tk2 ((Et /W )crys , OE (t+k+1) /W
t
t0
Z crys
(t+k+1)
(Et /W ))((t + k + 1))K
h
=Hrig,Z (U/K).
(3.4.6.2) u : Hrig,Z
h
(U/K) Hrig,Z
h
(U /K)
h
Denition 3.4.7. (1) We call (3.4.6.1) the weight spectral sequence of Hrig,Z
h
(U/K). We call the ltration P the weight ltration on Hrig,Z (U/K).
h
(2) Let k be an integer. We say that Hrig,Z (U/K) is of weight k (resp.
h h h
k) if Pk1 Hrig,Z (U/K) = 0 (resp. Hrig,Z (U/K) = Pk Hrig,Z (U/K)). We say
that Hrig,Z (U/K) is of pure weight k if Hrig,Z (U/K) is of weight k and k.
h h
h
We call a vector v of Hrig,Z (U/K) is of weight k if v Pk Hrig,Z
h
(U/K).
If is a nite eld, then the denitions in (3.4.7) are usual ones using the
eigenvalues of the Frobenius endomorphism by the spectral sequence (3.4.6.1)
and by the purity of the weight ([57], [15, (1.2)], [68, (2.2) (4)]).
We list some fundamental properties of the weight ltration on rigid
cohomology.
h
Proposition 3.4.8 ([70, (16.1)]). The weight ltration on Hrig,Z (U/K) is
compatible with the extension of complete discrete valuation rings of mixed
characteristics with perfect residue elds.
(3.4.9.1) Hrig,Z
h
(U/K) Hrig
h
(U/K) Hrig
h
(V /K)
(cf. [5, (2.3.1)]) is strictly exact with respect to the weight ltration.
(2) Let U be an open subscheme of U which contains Z as a closed sub-
scheme. Then the isomorphism
(3.4.9.2) h
Hrig,Z (U/K) Hrig,Z
h
(U /K)
(3.4.9.3) h
Hrig,T (U/K) Hrig,Z
h
(U/K)
(cf. [5, (2.5.1)]) is strictly compatible with respect to the weight ltration.
(4) If Z = Z1 Z2 , then the isomorphism
(3.4.9.4) h
Hrig,Z 1
(U/K) Hrig,Z
h
2
(U/K) Hrig,Z
h
(U/K)
The following (1) is a generalization of [14, (2.3)] (We reduce the proof of
the following to [loc. cit.] as in (3.4.2.2), (3.4.2.3) and (3.4.2.4).).
h
(1) The weights of Hrig,Z (U/K) lie in [h, 2(h c)].
h
(2) The weights of Hrig,Z (U/K) is less than or equal to 2d.
Let the notations be as in 3.3. In this section, assume that is perfect and
that V = W .
By using [68, (9.1), (9.3) (2)], we have a complex Wn X (log D ) (n
Z1 ) on X , which is the cosimplicial version of the log de Rham-Witt com-
plex Wn X (log D) in 2.12. We have a projection : Wn+1 X (log D )
Wn X (log D ). The family {Wn X (log D )}n=1 forms an inverse system.
Set W X (log D ) := limn Wn X (log D ).
As in (3.3.4), in [70] we have proved that there exists a split sim-
plicial log scheme (X , D ) over Spec() with a natural morphism
(X , D ) (X , D ) of simplicial log schemes such that Xm
(m N)
is the disjoint union of ane open subschemes which cover Xm . Set
(Xmn , Dmn ) := (coskX Dm
0 (Xm )n , cosk0 (Dm )n ). Then we have a nat-
m
(3.5.0.4)
Rlog-crys ((XN , DN )/Wn ) Rlog-crys ((XN, , DN, )/Wn )
R(XN, , Wn XN, (log DN, ))
R(XN , Wn XN (log DN )),
which is the N -truncated split simplicial version of [46, (4.19)] and [68, (7.19)].
The isomorphism (3.5.0.4) is compatible with respect to ns.
Let c be an integer in (3.2.0.4). Let N be a suciently large integer (e.g.,
N > 21 (c + 1)(c + 2)). By (3.3.6.1) and (3.2.0.4), we have
Because
and
i
(3.5.1.1) E1i,hi = H hi (MF(log h
dRW ))K0 = H (MF(rig )).
h
i
(3.5.1.3) H h (MF(rig )) = H hi (MF(log
dRW ))K0 .
i=0
Denition 3.5.2. We call the spectral sequence (3.5.1.1) the slope spectral
sequence of H h (MF(rig )). We call the direct sum decomposition (3.5.1.3) the
slope decomposition of H h (MF(rig )).
By (3.5.1.1) we obtain the following:
Proposition 3.5.3 ([70, (15.8)]). The slopes of H h (MF(rig )) lie in [0, h].
Now consider the case Y = E = . Then (3.5.1) is the following:
Theorem 3.5.4 (Slope decomposition). (1) There exists the following
spectral sequence
h
h
(3.5.4.3) Hrig (U/K0 ) = H hi (X , W iX (log D ))K0 .
i=0
is a functorial isomorphism.
We conclude this section by stating ranges of slopes on rigid cohomology
with closed support (The following (3) is a generalization of the latter part
of [16, (3.1.2)].):
Theorem 3.5.7 ([70, (16.11)]). Let the notations be as in (3.4.10). Then
the following hold:
h
(1) The slopes of Hrig,Z (U/K0 ) lie in [0, h].
h
(2) If U is smooth over , the slopes of Hrig,Z (U/K0 ) lie in [c, h c].
h
(3) Under the assumption in (2), the slopes of Hrig,Z (U/K0 ) lie in [hd, d].
Consequently, the slopes of Hrig,Z (U/K0 ) lie in [max{c, h d},
h
min{h c, d}].
(3.6.0.2) 2d
TrU : Hrig,c (U/K) K(d)
by the proof of [16, (2.1.1)] ([70, (16.9)]). Endow K(d) with an increasing
2dh
ltration P : P2d K(d) = K(d), P2d1 K(d) = 0. Because Hrig (U/K)
has the weight ltration, under the following identication
(3.6.0.3) h
Hrig,c (U/K) HomK (Hrig
2dh
(U/K), K(d)),
h
Hrig,c (U/K) has the weight ltration P .
3.6 Weight Filtration on Rigid Cohomology with Compact Support 245
The following are relative versions of [82, Theorem 2.4.4, Corollary 2.3.9,
Theorem 3.1.1] for the trivial coecients. In [83] and [73], we have proved that
the morphism (3.6.1.1) below is a functorial isomorphism; in [83] the second-
named author has also proved that the morphism (3.6.1.2) is a functorial
isomorphism.
and
(3.6.1.2) Rh f(X,D)/SK0 (K(X,D)/SK0 ) sp Rh frig (U/SK0 )
induces an isomorphism
(3.6.2.2) 2dh
Hrig,Z (U/K) HomK (Hrig,c
h
(Z/K), K(d))
into a separated smooth scheme of nite type over of pure dimension d and
that Z is of pure codimension c in U .
By the proof of [6, (2.4)], we have the following isomorphism
(3.6.4.2) h
Hrig,c (Z/K) HomK (Hrig,Z
2dh
(U/K), K).
2dh 2dh
Because Hrig,Z (U/K) has the weight ltration, HomK (Hrig,Z (U/K), K(d))
has the weight ltration. Thus, under the following identication
h 2dh
(3.6.4.3) Hrig,c (Z/K) = HomK (Hrig,Z (U/K), K(d)),
h
we have a weight ltration P on Hrig,c (Z/K).
If U are Z are reduced, then there exist dense open smooth subschemes
U sm and Z sm of U and Z, respectively. Using induction on dim Z and using
(3.6.2), we can prove the following:
h
Theorem 3.6.5 ([70, (17.11)]). The weight ltration P on Hrig,c (Z/K) by
the formula (3.6.4.3) is independent of the choice of U .
h
Denition 3.6.6. We call the weight ltration on Hrig,c (Z/K) given by the
h
formula (3.6.4.3) the weight ltration on Hrig,c (Z/K).
h
We list some properties of the weight ltration on Hrig,c (Z/K).
h h
(3.6.7.2) Hlog -crys,c ((X, D)/W )K = Hrig,c (U/K).
h
Proposition 3.6.8 ([70, (17.4)]). The weight ltration on Hrig,c (U/K) in
h
this section is equal to the weight ltration on Hlog -crys,c ((X, D)/W )K in
(2.11.15.1) under the canonical isomorphism (3.6.7.2).
(3.6.9.1) Hrig,c
h
(V /K) Hrig,c
h
(U/K) Hrig,c
h
(Z/K)
([4, (3.1) (iii)]) is strictly exact with respect to the weight ltration.
The following immediately follows from [6, (2.2) (ii)] and (3.6.10).
(3.6.12.1) h
Hrig,c (Z/K) Hrig
h
(Z/K)
249
250 A Relative SNCD
m
s
(A.0.1.1) div(di ) = div(yi )
i=1 i=1
(A.0.1.3)
yj
0 B B B/yj B 0
yj
0 (BK )w (BK )w (BK )w /yj (BK )w 0.
A Relative SNCD 251
s
m
div(yj ) = div(yj(i) ) in Div(Xz /S)0 .
j=1 i=1
m
Set Y := i=1 div(yj(i) ) Xz . Since z is contained in Y and since Y is at
over S, there exists a point w Y such that w is sent to by the structural
morphism Xz S. Hence w Y . If we dene (BK )w as before, we have the
s m
equality j=1 div(yj ) = i=1 div(yj(i) ) in Div(Spec((BK )w )/K)0 . Since
(BK )w is a UFD and since yj
(BK )w (1 j m), we obtain s = m and
div(di ) = div(yi ) up to some renumbering of the indexes.
Case III: Next consider the case where S is a noetherian scheme.
Let t be the image of z by f . Then we may assume that S is the spectrum
of a noetherian local ring A with closed point t. Let m be the maximal ideal
of A. Let p be a prime ideal of A. By the result in the Case II, we have
s = m and di (B/pB) = yi (B/pB) (1 i s) up to some renumbering of the
indexes. Note that the correspondence di yi is independent of the choice
of p. Indeed, di (B/pB) = yi (B/pB) implies di (B/mB) = yi (B/mB), and
hence yi is uniquely determined by the result in the Case I.
Now let : B B/pB be the projection and by the induced morphism
B/yi B (B/pB)/yi (B/pB) by . Consider the following diagram
(A.0.1.4)
yi
0 B B B/yi B 0
yi
0 B/pB B/pB (B/pB)/yi (B/pB) 0.
for any 1 i sand for any prime ideal p of A. Let n be the nilpotent radical
q
of A and let n = k=1 pk (q Z>0 ) be the primary decomposition of n, where
q of A. Then, by (A.0.1.5) for p = pk (1 k q), we obtain
pk is a prime ideal
di (B/yi B) k=1 pk (B/yi B) = n(B/yi B), where the last equality follows
from the atness of B/yi B over A. Hence we obtain di B yi B + nB for
1 i s.
Now we prove the inclusion di B yi B (1 i s). To prove this, it
suces to prove the inclusion di B yi B + ne B (1 i s) for any positive
integer e since ne = 0 for some e. We prove this inclusion by induction on e.
We have already proved the inclusion for the case e = 1. Assume that we
have the inclusion di B yi B + ne B (1 i s) for a positive integer e. Then
there exists an element ui (resp. i ) of B (resp. ne B) such that di = ui yi + i .
By the diagram (2.1.7.2), we have
s
s
s
(A.0.1.6) ( yi )B = ( di )B = ( (ui yi + i ))B.
i=1 i=1 i=1
s
s
s
( yi )(B/nB) = ( ui )( yi )(B/nB).
i=1 i=1 i=1
s
Since i=1 yi is a non-zero divisor in B/nB, each ui is invertible in B/nB,
and hence so is in B. Replacing di by u1
i di , we may assume that ui = 1 for
any i. Fix an index i0 (1 i0 s) and consider the
equation (A.0.1.6) (with
ui = 1) modulo yi0 B + ne+1 B. Then we see that i0 i=i0 yi yi0 B + ne+1 B.
Because the following diagram
( i=i yi )
B/n e+1
0 B/ne+1
yi0
yi
(A.0.1.7) 0
( i=i yi )
B/ne+1 0 B/ne+1
V V
S S
satisfying the following conditions:
(1) S is a noetherian scheme and V S is smooth.
(2) There exists an eective Cartier divisor Di = div(di ) (1 i m) on
m
V which is smooth over S satisfying Di S S = Di |V . Set D := i=1 Di .
(3) There exists a cartesian diagram
D V
g
SpecS (OS [y1 , . . . , yd ]/(y1 ys )) SpecS (OS [y1 , . . . , yd ]),
(, z) = (, z0 ) = = (, zn ) = (, w).
A Relative SNCD 255
Thus we have proved the claim. We denote this index by (). Then we have
D = (D Uz ) = (D () Uz ) D () ,
zD zD
Using (A.0.5), we dene the operation for the elements in DivD (X/S)0
as follows:
Proposition A.0.6. Let the notations be as above. Then the following hold:
(1) For two elements E1 and E2 in DivD (X/S)0 , there exists a unique
element E1 E2 DivD (X/S)0 satisfying the following equality:
(A.0.8.1) y + = uy n .
q
q+s
ni mi
(A.0.9.1) v (y + ) = (y ni + )mi
i=1 i=q+1
in B. q q+s
Dene integers P and Q by P := i=1 mi ni and Q := i=q+1 mi ni ,
respectively. Since 2 = 0, we obtain the following equality
q
q+s
(A.0.9.2) v{y P + ( mi y P ni )} = y Q + ( mi y Qni ).
i=1 i=q+1
Let [y]loc be the localization of the polynomial ring [y] at the prime ideal
(y). Let be the composite morphism A[y, y2 , . . . , yd ] [y] [y]loc ,
where the rst homomorphism sends a A (resp. y, yi (2 i d)) to
a mod m (resp. y,0) and the second homomorphism is the localization. Then
the ring C := B A[y,y2 ,...,yd ], [y]loc is a local ring which is ind-etale over
[y]loc . Hence C is a discrete valuation ring with uniformizer y. Denote the
natural homomorphism B C by and apply to the equality (A.0.9.2).
Then we obtain P = Q since (v) C . Hence we obtain the equality
260 A Relative SNCD
q
q+s
P ni
(A.0.9.3) P
v{y + ( mi y P
)} = y + ( mi y P ni ).
i=1 i=q+1
The proposition (A.0.9) tells us that N (D) is far from nice in general. This
is the reason why we have considered only M (D) in the text.
References
261
262 References
19. Chiarellotto, B., Tsuzuki, N.: Cohomological descent of rigid cohomology for etale
coverings. Rend. Sem. Mat. Univ. Padova 109, (2003), 63215.
20. Deligne, P.: Theor`eme de Lefschetz et crit` eres de d
egenerescence de suites spectrales.
Publ. Math. 35, (1968), 259278.
IHES
21. Deligne, P.: Equations dierentielles a
` points singuliers r eguliers. Lecture Notes in
Math. 163, Springer-Verlag, (1970).
22. Deligne, P.: Theorie de Hodge I. Actes du Congr`es International des Mathematiciens
(Nice, 1970), Tome 1, pp. 425430. Gauthier-Villars, Paris, (1971).
23. Deligne, P.: Theorie de Hodge, II. IHES Publ. Math. 40, (1971), 557.
24. Deligne, P.: Theorie de Hodge, III. IHES Publ. Math. 44, (1974), 577.
25. Deligne, P.: Poids dans la cohomologie des vari et
es algebriques. Proceedings of the
International Congress of Mathematicians (Vancouver, B. C., 1974), Vol. 1, pp. 7985.
26. Deligne, P.: La conjecture de Weil, II. IHES Publ. Math. 52, (1980), 137252.
27. Deligne, P., Illusie, L.: Rel`evements modulo p2 et d ecompositon du complexe de de
Rham. Invent. Math. 89, (1987), 247270.
28. de Jong, A. J.: Smoothness, semi-stability and alterations. IHES Publ. Math. 83,
(1996), 5193.
29. Faltings, G.: F -isocrystals on open varieties: Results and conjectures. The
Grothendieck Festschrift II, Progr. Math. 87, Birkh auser, (1990), 219248.
30. Fujiwara, K., Kato, K.: Logarithmic etale topology theory. Preprint.
31. Friedman, R.: Global smoothings of varieties with normal crossings. Ann. of Math. 118,
(1983), 75114.
32. Fujiwara, K.: Theory of tubular neighborhood in etale topology. Duke Math. J. 80,
(1995), 1557.
33. Fujiwara, K.: A proof of the absolute purity conjecture (after Gabber). Advanced Stud-
ies in Pure Math. 36, (2002), Azumino, 153183.
34. Gabber, O.: Sur la torsion dans la cohomologie ladique dune vari et
e.
C. R. Acad. Sci. Paris Ser. I, Math. 297, (1983), 179182.
35. Guillen, F., Navarro Aznar, V.: Sur le th eor`
eme local des cycles invariants. Duke
Math. J. 61, (1990), 133155.
36. Groe-Kl onne, E.: Finiteness of de Rham cohomology in rigid analysis. Duke
Math. J. 113, (2002), 5791.
37. Grothendieck, A.: R ecoltes et semailles: R eexions et t emoinage sur un pass e de
math ematicien, Parties I, II, IV. Gendai-Sugaku-sha, Japanese translation by Y. Tsuji,
(1989), (1990), unpublished.
38. Grothendieck, A., Dieudonne, J.: El ements de g eom etrie algebrique. I. IHES
Publ. Math. 4, (1960).
39. Grothendieck, A., Dieudonne, J.: El ements de g eom etrie algebrique. II. IHES
Publ. Math. 8, (1961).
40. Grothendieck, A., Dieudonne, J.: El ements de g eom etrie alg
ebrique IV-1, 3, 4. IHES
Publ. Math. 20, 28, 32, (1964), (1966), (1967).
41. Grothendieck, A., Raynaud, Mme M.: Rev etements etales et groupe fondamental. Lec-
ture Notes in Math. 224, Springer-Verlag, (1971).
42. Grothendieck, A. et al.: Th eorie des topos et cohomologie etale des schemas-1, 2.
Lecture Notes in Math. 269, 270, Springer-Verlag, (1972).
43. Grothendieck, A. et al.: Th eorie des intersections et th eor`eme de Riemann-Roch. Lec-
ture Notes in Math. 225, Springer-Verlag, (1971).
44. Hartshorne, R.: Residues and duality. Lecture Notes in Math. 20, Springer-Verlag,
(1966).
45. Hartshorne, R.: Algebraic geometry. Graduate Texts in Math. 52, Springer-Verlag,
(1977).
46. Hyodo, O., Kato, K.: Semi-stable reduction and crystalline cohomology with loga-
rithmic poles. Periodes p-adiques (Bures-sur-Yvette, 1988). Asterisque 223, (1994),
221268.
References 263
47. Hyodo, O.: A cohomological construction of Swan representations over the Witt ring. I.
Proc. Japan Acad. Ser. A Math. Sci. 64, (1988), 300303.
48. Hyodo, O.: On the de Rham-Witt complex attached to a semi-stable family.
Comp. Math. 78, (1991), 241260.
49. Illusie, L.: Report on crystalline cohomology. Proc. of Symposia in Pure
Math. Vol. XXIX, (1975), 459478.
50. Illusie, L.: Complexe de de Rham-Witt et cohomologie cristalline. Ann. Sci-
Norm. Sup. 4e s
ent. Ec. erie 12, (1979), 501661.
51. Illusie, L., Kato, K., Nakayama, C.: Quasi-unipotent logarithmic Riemann-Hilbert cor-
respondences. J. Math. Sci. Univ. Tokyo 12, (2005), 166.
52. Illusie, L., Raynaud, M.: Les suites spectrales associ
ees au complexe de de Rham-Witt.
IHES Publ. Math. 57, (1983), 73212.
53. Kato, F.: Log smooth deformation theory. T ohoku Math. J. 48, (1996), 317354.
54. Kato, K.: Logarithmic structures of Fontaine-Illusie. In: Algebraic analysis, geometry,
and number theory, Johns Hopkins Univ. Press, (1989), 191224.
55. Kato, K.: Toric Singularities. Amer. J. of Math. 116, (1994), 10731099.
56. Katz, N., Mazur, B.: Arithmetic moduli of elliptic curves. Annals of Mathematics
Studies, 108, Princeton University Press, (1985).
57. Katz, N., Messing, W.: Some consequences of the Riemann hypothesis for varieties
over nite elds. Invent. Math. 23, (1974), 7377.
58. Kato, K., Nakayama, C.: Log Betti cohomology, log etale cohomology, and log de Rham
cohomology of log schemes over C. Kodai Math. J. 22, (1999), 161186.
59. Kedlaya, K. S.: Finiteness of rigid cohomology with coecients. Duke Math. J. 134,
(2006), 1597.
60. Laumon, G.: Sur la cat egorie d ee des D-modules ltr
eriv es.. Algebraic geometry
(Tokyo/Kyoto, 1982), Lecture Notes in Math. 1016, Springer-Verlag, (1983), 151237.
61. Lazard, M.: Commutative formal groups. Lecture Notes in Math. 443, Springer-Verlag,
(1975).
62. Lorenzon, P.: Logarithmic Hodge-Witt forms and Hyodo-Kato cohomology. J. Algebra
249, (2002), 247265.
63. Matsumura, H.: Commutative algebra. Second edition. Mathematics Lecture Note Se-
ries, 56. Benjamin/Cummings Publishing Co., Inc., Reading, Mass., (1980).
64. Mokrane, A. La suite spectrale des poids en cohomologie de Hyodo-Kato. Duke
Math. J. 72, (1993), 301337.
65. Mokrane, A.: Cohomologie cristalline des vari et
es ouvertes. Rev. Maghrebine Math. 2,
(1993), 161175.
66. Nagata, M.: Imbedding of an abstract variety in a complete variety. J. Math. Kyoto
Univ. 2, (1962), 110.
67. Nakayama, C.: Logarithmic etale cohomology. Math. Ann. 308, (1997), 365404.
68. Nakkajima, Y.: p-adic weight spectral sequences of log varieties. J. Math. Sci. Univ.
Tokyo 12, (2005), 513661.
69. Nakkajima, Y.: Signs in weight spectral sequences, monodromyweight conjectures, log
Hodge symmetry and degenerations of surfaces. Rend. Sem. Mat. Univ. Padova 116,
(2006), 71185.
70. Nakkajima, Y.: Weight ltration and slope ltration on the rigid cohomology of a
variety in characteristic p > 0. To appear in Mem. Soc. Math. France.
71. Nakkajima, Y.: Zeta functions of log varieties. Preprint.
72. Nakkajima, Y.: Kato-Nakayamas comparison theorem and analytic log etale topoi.
Preprint.
73. Nakkajima, Y., Shiho, A.: Weight-ltered convergent complex. Preprint.
74. Ogus, A.: F -isocrystals and de Rham cohomology. II. Convergent isocrystals. Duke
Math. J. 51, (1984), 765850.
75. Ogus, A.: F -crystals, Griths transversality, and the Hodge decomposition. Ast erisque
221, (1994).
264 References
76. Ogus, A.: F -crystals on schemes with constant log structure. Compositio Math. 97,
(1995), 187225.
77. Raynaud, M., Gruson, L.: Crit` eres de platitude et de projectivite. Techniques de
platication dun module. Invent. Math. 13, (1971), 189.
78. Schneiders, J.-P.: Quasi-abelian categories and sheaves. M em. Soc. Math. Fr. (N.S.)
76, (1999).
79. Serre, J.-P.: Corps locaux. Hermann, Paris, (1968).
80. Serre, J.-P.: Faisceaux alg
ebriques coh erents. Ann. of Math. 61, (1955), 197278.
81. Shiho, A.: Crystalline fundamental groups IIsocrystals on log crystalline site and
log convergent site. J. Math. Sci. Univ. Tokyo 7, (2000), 509656.
82. Shiho, A.: Crystalline fundamental groups IILog convergent cohomology and rigid
cohomology. J. Math. Sci. Univ. Tokyo 9, (2002), 1163.
83. Shiho, A.: Relative log convergent cohomology and relative rigid cohomology I, II, III.
Preprint.
84. Tate, J.: Rigid analytic spaces. Invent. Math. 12, (1971), 257289.
85. Tsuji, T.: Poincar e duality for logarithmic crystalline cohomology. Compositio
Math. 118, (1999), 1141.
86. Tsuji, T.: p-adic etale cohomology and crystalline cohomology in the semi-stable re-
duction case. Invenr. Math. 137, (1999), 233411.
87. Tsuzuki, N.: Cohomological descent of rigid cohomology for proper coverings. In-
vent. Math. 151, (2003), 101133.
88. Tsuzuki, N.: Cohomological descent in rigid cohomology. In: Geometric aspects of
Dwork theory Vol. II (Adolphson, A., Baldassari, F., Berthelot, P., Katz, N., Loeser,
F. eds.), Walter de Gruyter, (2004), 931981.
89. Verdier, J.-L.: Des categories d
erivees des cat
egories ab
eliennes. Ast
erisque No. 239,
(1996).
90. Weibel, C. A.: An introduction to homological algebra. Cambridge Studies in Advanced
Mathematics, 38. Cambridge University Press, Cambridge, (1994).
Index
adic ne formal log PD-scheme, 197 Gysin morphism, 129, 130, 195
adjunction formula, 28 Gysin morphism in log Hodge-Witt
admissible (closed) immersion, 62 cohomologies, 191
265
266 Index
Vol. 1776: K. Behrend, C. Gomez, V. Tarasov, G. Tian, Vol. 1800: O. Gabber, L. Ramero, Almost Ring Theory
Quantum Comohology. Cetraro 1997. Editors: P. de Bar- (2003)
tolomeis, B. Dubrovin, C. Reina (2002) Vol. 1801: J. Azma, M. mery, M. Ledoux, M. Yor
Vol. 1777: E. Garca-Ro, D. N. Kupeli, R. Vzquez- (Eds.), Sminaire de Probabilits XXXVI (2003)
Lorenzo, Osserman Manifolds in Semi-Riemannian Vol. 1802: V. Capasso, E. Merzbach, B. G. Ivanoff,
Geometry (2002) M. Dozzi, R. Dalang, T. Mountford, Topics in Spatial
Vol. 1778: H. Kiechle, Theory of K-Loops (2002) Stochastic Processes. Martina Franca, Italy 2001. Editor:
Vol. 1779: I. Chueshov, Monotone Random Systems E. Merzbach (2003)
(2002) Vol. 1803: G. Dolzmann, Variational Methods for Crys-
Vol. 1780: J. H. Bruinier, Borcherds Products on O(2,1) talline Microstructure Analysis and Computation
and Chern Classes of Heegner Divisors (2002) (2003)
Vol. 1781: E. Bolthausen, E. Perkins, A. van der Vaart, Vol. 1804: I. Cherednik, Ya. Markov, R. Howe, G.
Lectures on Probability Theory and Statistics. Ecole d Lusztig, Iwahori-Hecke Algebras and their Representa-
Et de Probabilits de Saint-Flour XXIX-1999. Editor: tion Theory. Martina Franca, Italy 1999. Editors: V. Bal-
P. Bernard (2002) doni, D. Barbasch (2003)
Vol. 1782: C.-H. Chu, A. T.-M. Lau, Harmonic Functions Vol. 1805: F. Cao, Geometric Curve Evolution and Image
on Groups and Fourier Algebras (2002) Processing (2003)
Vol. 1783: L. Grne, Asymptotic Behavior of Dynamical Vol. 1806: H. Broer, I. Hoveijn. G. Lunther, G. Vegter,
and Control Systems under Perturbation and Discretiza- Bifurcations in Hamiltonian Systems. Computing Singu-
tion (2002) larities by Grbner Bases (2003)
Vol. 1784: L. H. Eliasson, S. B. Kuksin, S. Marmi, J.-C. Vol. 1807: V. D. Milman, G. Schechtman (Eds.), Geomet-
Yoccoz, Dynamical Systems and Small Divisors. Cetraro, ric Aspects of Functional Analysis. Israel Seminar 2000-
Italy 1998. Editors: S. Marmi, J.-C. Yoccoz (2002) 2002 (2003)
Vol. 1785: J. Arias de Reyna, Pointwise Convergence of
Vol. 1808: W. Schindler, Measures with Symmetry Prop-
Fourier Series (2002)
erties (2003)
Vol. 1786: S. D. Cutkosky, Monomialization of Mor-
Vol. 1809: O. Steinbach, Stability Estimates for Hybrid
phisms from 3-Folds to Surfaces (2002)
Coupled Domain Decomposition Methods (2003)
Vol. 1787: S. Caenepeel, G. Militaru, S. Zhu, Frobenius
and Separable Functors for Generalized Module Cate- Vol. 1810: J. Wengenroth, Derived Functors in Functional
gories and Nonlinear Equations (2002) Analysis (2003)
Vol. 1788: A. Vasilev, Moduli of Families of Curves for Vol. 1811: J. Stevens, Deformations of Singularities
Conformal and Quasiconformal Mappings (2002) (2003)
Vol. 1789: Y. Sommerhuser, Yetter-Drinfeld Hopf alge- Vol. 1812: L. Ambrosio, K. Deckelnick, G. Dziuk,
bras over groups of prime order (2002) M. Mimura, V. A. Solonnikov, H. M. Soner, Mathemat-
Vol. 1790: X. Zhan, Matrix Inequalities (2002) ical Aspects of Evolving Interfaces. Madeira, Funchal,
Vol. 1791: M. Knebusch, D. Zhang, Manis Valuations Portugal 2000. Editors: P. Colli, J. F. Rodrigues (2003)
and Prfer Extensions I: A new Chapter in Commutative Vol. 1813: L. Ambrosio, L. A. Caffarelli, Y. Brenier,
Algebra (2002) G. Buttazzo, C. Villani, Optimal Transportation and its
Vol. 1792: D. D. Ang, R. Goreno, V. K. Le, D. D. Trong, Applications. Martina Franca, Italy 2001. Editors: L. A.
Moment Theory and Some Inverse Problems in Potential Caffarelli, S. Salsa (2003)
Theory and Heat Conduction (2002) Vol. 1814: P. Bank, F. Baudoin, H. Fllmer, L.C.G.
Vol. 1793: J. Corts Monforte, Geometric, Control and Rogers, M. Soner, N. Touzi, Paris-Princeton Lectures on
Numerical Aspects of Nonholonomic Systems (2002) Mathematical Finance 2002 (2003)
Vol. 1794: N. Pytheas Fogg, Substitution in Dynamics, Vol. 1815: A. M. Vershik (Ed.), Asymptotic Combi-
Arithmetics and Combinatorics. Editors: V. Berth, S. natorics with Applications to Mathematical Physics.
Ferenczi, C. Mauduit, A. Siegel (2002) St. Petersburg, Russia 2001 (2003)
Vol. 1795: H. Li, Filtered-Graded Transfer in Using Non- Vol. 1816: S. Albeverio, W. Schachermayer, M. Tala-
commutative Grbner Bases (2002) grand, Lectures on Probability Theory and Statistics.
Vol. 1796: J.M. Melenk, hp-Finite Element Methods for Ecole dEt de Probabilits de Saint-Flour XXX-2000.
Singular Perturbations (2002) Editor: P. Bernard (2003)
Vol. 1797: B. Schmidt, Characters and Cyclotomic Fields Vol. 1817: E. Koelink, W. Van Assche (Eds.), Orthogonal
in Finite Geometry (2002) Polynomials and Special Functions. Leuven 2002 (2003)
Vol. 1798: W.M. Oliva, Geometric Mechanics (2002) Vol. 1818: M. Bildhauer, Convex Variational Problems
Vol. 1799: H. Pajot, Analytic Capacity, Rectiability, with Linear, nearly Linear and/or Anisotropic Growth
Menger Curvature and the Cauchy Integral (2002) Conditions (2003)
Vol. 1819: D. Masser, Yu. V. Nesterenko, H. P. Schlick- Vol. 1844: K.F. Siburg, The Principle of Least Action in
ewei, W. M. Schmidt, M. Waldschmidt, Diophantine Geometry and Dynamics (2004)
Approximation. Cetraro, Italy 2000. Editors: F. Amoroso, Vol. 1845: Min Ho Lee, Mixed Automorphic Forms,
U. Zannier (2003) Torus Bundles, and Jacobi Forms (2004)
Vol. 1820: F. Hiai, H. Kosaki, Means of Hilbert Space Vol. 1846: H. Ammari, H. Kang, Reconstruction of Small
Operators (2003) Inhomogeneities from Boundary Measurements (2004)
Vol. 1821: S. Teufel, Adiabatic Perturbation Theory in Vol. 1847: T.R. Bielecki, T. Bjrk, M. Jeanblanc, M.
Quantum Dynamics (2003) Rutkowski, J.A. Scheinkman, W. Xiong, Paris-Princeton
Vol. 1822: S.-N. Chow, R. Conti, R. Johnson, J. Mallet- Lectures on Mathematical Finance 2003 (2004)
Paret, R. Nussbaum, Dynamical Systems. Cetraro, Italy Vol. 1848: M. Abate, J. E. Fornaess, X. Huang, J. P.
2000. Editors: J. W. Macki, P. Zecca (2003) Rosay, A. Tumanov, Real Methods in Complex and CR
Vol. 1823: A. M. Anile, W. Allegretto, C. Ring- Geometry, Martina Franca, Italy 2002. Editors: D. Zait-
hofer, Mathematical Problems in Semiconductor Physics. sev, G. Zampieri (2004)
Cetraro, Italy 1998. Editor: A. M. Anile (2003) Vol. 1849: Martin L. Brown, Heegner Modules and Ellip-
Vol. 1824: J. A. Navarro Gonzlez, J. B. Sancho de Salas, tic Curves (2004)
C Differentiable Spaces (2003) Vol. 1850: V. D. Milman, G. Schechtman (Eds.), Geomet-
Vol. 1825: J. H. Bramble, A. Cohen, W. Dahmen, Mul- ric Aspects of Functional Analysis. Israel Seminar 2002-
tiscale Problems and Methods in Numerical Simulations, 2003 (2004)
Martina Franca, Italy 2001. Editor: C. Canuto (2003) Vol. 1851: O. Catoni, Statistical Learning Theory and
Vol. 1826: K. Dohmen, Improved Bonferroni Inequal- Stochastic Optimization (2004)
ities via Abstract Tubes. Inequalities and Identities of Vol. 1852: A.S. Kechris, B.D. Miller, Topics in Orbit
Inclusion-Exclusion Type. VIII, 113 p, 2003. Equivalence (2004)
Vol. 1827: K. M. Pilgrim, Combinations of Complex Vol. 1853: Ch. Favre, M. Jonsson, The Valuative Tree
Dynamical Systems. IX, 118 p, 2003. (2004)
Vol. 1828: D. J. Green, Grbner Bases and the Computa- Vol. 1854: O. Saeki, Topology of Singular Fibers of Dif-
tion of Group Cohomology. XII, 138 p, 2003. ferential Maps (2004)
Vol. 1829: E. Altman, B. Gaujal, A. Hordijk, Discrete- Vol. 1855: G. Da Prato, P.C. Kunstmann, I. Lasiecka,
Event Control of Stochastic Networks: Multimodularity A. Lunardi, R. Schnaubelt, L. Weis, Functional Analytic
and Regularity. XIV, 313 p, 2003. Methods for Evolution Equations. Editors: M. Iannelli,
Vol. 1830: M. I. Gil, Operator Functions and Localiza- R. Nagel, S. Piazzera (2004)
tion of Spectra. XIV, 256 p, 2003. Vol. 1856: K. Back, T.R. Bielecki, C. Hipp, S. Peng,
Vol. 1831: A. Connes, J. Cuntz, E. Guentner, N. Hig- W. Schachermayer, Stochastic Methods in Finance, Bres-
son, J. E. Kaminker, Noncommutative Geometry, Mar- sanone/Brixen, Italy, 2003. Editors: M. Fritelli, W. Rung-
tina Franca, Italy 2002. Editors: S. Doplicher, L. Longo galdier (2004)
(2004) Vol. 1857: M. mery, M. Ledoux, M. Yor (Eds.), Smi-
Vol. 1832: J. Azma, M. mery, M. Ledoux, M. Yor naire de Probabilits XXXVIII (2005)
(Eds.), Sminaire de Probabilits XXXVII (2003) Vol. 1858: A.S. Cherny, H.-J. Engelbert, Singular Sto-
Vol. 1833: D.-Q. Jiang, M. Qian, M.-P. Qian, Mathemati- chastic Differential Equations (2005)
cal Theory of Nonequilibrium Steady States. On the Fron- Vol. 1859: E. Letellier, Fourier Transforms of Invariant
tier of Probability and Dynamical Systems. IX, 280 p, Functions on Finite Reductive Lie Algebras (2005)
2004. Vol. 1860: A. Borisyuk, G.B. Ermentrout, A. Friedman,
Vol. 1834: Yo. Yomdin, G. Comte, Tame Geometry with D. Terman, Tutorials in Mathematical Biosciences I.
Application in Smooth Analysis. VIII, 186 p, 2004. Mathematical Neurosciences (2005)
Vol. 1835: O.T. Izhboldin, B. Kahn, N.A. Karpenko, Vol. 1861: G. Benettin, J. Henrard, S. Kuksin, Hamil-
A. Vishik, Geometric Methods in the Algebraic Theory tonian Dynamics Theory and Applications, Cetraro,
of Quadratic Forms. Summer School, Lens, 2000. Editor: Italy, 1999. Editor: A. Giorgilli (2005)
J.-P. Tignol (2004) Vol. 1862: B. Helffer, F. Nier, Hypoelliptic Estimates and
Vol. 1836: C. Nastasescu, F. Van Oystaeyen, Methods of Spectral Theory for Fokker-Planck Operators and Witten
Graded Rings. XIII, 304 p, 2004. Laplacians (2005)
Vol. 1837: S. Tavar, O. Zeitouni, Lectures on Probabil- Vol. 1863: H. Fhr, Abstract Harmonic Analysis of Con-
ity Theory and Statistics. Ecole dEt de Probabilits de tinuous Wavelet Transforms (2005)
Saint-Flour XXXI-2001. Editor: J. Picard (2004) Vol. 1864: K. Efstathiou, Metamorphoses of Hamiltonian
Vol. 1838: A.J. Ganesh, N.W. OConnell, D.J. Wischik, Systems with Symmetries (2005)
Big Queues. XII, 254 p, 2004. Vol. 1865: D. Applebaum, B.V. R. Bhat, J. Kustermans,
Vol. 1839: R. Gohm, Noncommutative Stationary J. M. Lindsay, Quantum Independent Increment
Processes. VIII, 170 p, 2004. Processes I. From Classical Probability to Quantum
Vol. 1840: B. Tsirelson, W. Werner, Lectures on Probabil- Stochastic Calculus. Editors: M. Schrmann, U. Franz
ity Theory and Statistics. Ecole dEt de Probabilits de (2005)
Saint-Flour XXXII-2002. Editor: J. Picard (2004) Vol. 1866: O.E. Barndorff-Nielsen, U. Franz, R. Gohm,
Vol. 1841: W. Reichel, Uniqueness Theorems for Vari- B. Kmmerer, S. Thorbjnsen, Quantum Independent
ational Problems by the Method of Transformation Increment Processes II. Structure of Quantum Lvy
Groups (2004) Processes, Classical Probability, and Physics. Editors: M.
Vol. 1842: T. Johnsen, A. L. Knutsen, K3 Projective Mod- Schrmann, U. Franz, (2005)
els in Scrolls (2004) Vol. 1867: J. Sneyd (Ed.), Tutorials in Mathematical Bio-
Vol. 1843: B. Jefferies, Spectral Properties of Noncom- sciences II. Mathematical Modeling of Calcium Dynam-
muting Operators (2004) ics and Signal Transduction. (2005)
Vol. 1868: J. Jorgenson, S. Lang, Posn (R) and Eisenstein Vol. 1896: P. Massart, Concentration Inequalities and
Series. (2005) Model Selection, Ecole dt de Probabilits de Saint-
Vol. 1869: A. Dembo, T. Funaki, Lectures on Probabil- Flour XXXIII-2003. Editor: J. Picard (2007)
ity Theory and Statistics. Ecole dEt de Probabilits de Vol. 1897: R. Doney, Fluctuation Theory for Lvy
Saint-Flour XXXIII-2003. Editor: J. Picard (2005) Processes, Ecole dt de Probabilits de Saint-Flour
Vol. 1870: V.I. Gurariy, W. Lusky, Geometry of Mntz XXXV-2005. Editor: J. Picard (2007)
Spaces and Related Questions. (2005) Vol. 1898: H.R. Beyer, Beyond Partial Differential Equa-
Vol. 1871: P. Constantin, G. Gallavotti, A.V. Kazhikhov, tions, On linear and Quasi-Linear Abstract Hyperbolic
Y. Meyer, S. Ukai, Mathematical Foundation of Turbu- Evolution Equations (2007)
lent Viscous Flows, Martina Franca, Italy, 2003. Editors: Vol. 1899: Sminaire de Probabilits XL. Editors:
M. Cannone, T. Miyakawa (2006) C. Donati-Martin, M. mery, A. Rouault, C. Stricker
Vol. 1872: A. Friedman (Ed.), Tutorials in Mathemati- (2007)
cal Biosciences III. Cell Cycle, Proliferation, and Cancer Vol. 1900: E. Bolthausen, A. Bovier (Eds.), Spin Glasses
(2006) (2007)
Vol. 1873: R. Mansuy, M. Yor, Random Times and En- Vol. 1901: O. Wittenberg, Intersections de deux
largements of Filtrations in a Brownian Setting (2006) quadriques et pinceaux de courbes de genre 1, Intersec-
Vol. 1874: M. Yor, M. mery (Eds.), In Memoriam Paul- tions of Two Quadrics and Pencils of Curves of Genus 1
Andr Meyer - Sminaire de Probabilits XXXIX (2006) (2007)
Vol. 1875: J. Pitman, Combinatorial Stochastic Processes. Vol. 1902: A. Isaev, Lectures on the Automorphism
Ecole dEt de Probabilits de Saint-Flour XXXII-2002. Groups of Kobayashi-Hyperbolic Manifolds (2007)
Editor: J. Picard (2006) Vol. 1903: G. Kresin, V. Mazya, Sharp Real-Part Theo-
Vol. 1876: H. Herrlich, Axiom of Choice (2006) rems (2007)
Vol. 1877: J. Steuding, Value Distributions of L-Functions Vol. 1904: P. Giesl, Construction of Global Lyapunov
(2007) Functions Using Radial Basis Functions (2007)
Vol. 1878: R. Cerf, The Wulff Crystal in Ising and Percol- Vol. 1905: C. Prvot, M. Rckner, A Concise Course on
ation Models, Ecole dEt de Probabilits de Saint-Flour Stochastic Partial Differential Equations (2007)
XXXIV-2004. Editor: Jean Picard (2006) Vol. 1906: T. Schuster, The Method of Approximate
Vol. 1879: G. Slade, The Lace Expansion and its Applica- Inverse: Theory and Applications (2007)
tions, Ecole dEt de Probabilits de Saint-Flour XXXIV- Vol. 1907: M. Rasmussen, Attractivity and Bifurcation
2004. Editor: Jean Picard (2006) for Nonautonomous Dynamical Systems (2007)
Vol. 1880: S. Attal, A. Joye, C.-A. Pillet, Open Quantum Vol. 1908: T.J. Lyons, M. Caruana, T. Lvy, Differential
Systems I, The Hamiltonian Approach (2006) Equations Driven by Rough Paths, Ecole dt de Proba-
Vol. 1881: S. Attal, A. Joye, C.-A. Pillet, Open Quantum bilits de Saint-Flour XXXIV-2004 (2007)
Systems II, The Markovian Approach (2006) Vol. 1909: H. Akiyoshi, M. Sakuma, M. Wada,
Vol. 1882: S. Attal, A. Joye, C.-A. Pillet, Open Quantum Y. Yamashita, Punctured Torus Groups and 2-Bridge
Systems III, Recent Developments (2006) Knot Groups (I) (2007)
Vol. 1883: W. Van Assche, F. Marcelln (Eds.), Orthogo- Vol. 1910: V.D. Milman, G. Schechtman (Eds.), Geo-
nal Polynomials and Special Functions, Computation and metric Aspects of Functional Analysis. Israel Seminar
Application (2006) 2004-2005 (2007)
Vol. 1884: N. Hayashi, E.I. Kaikina, P.I. Naumkin, Vol. 1911: A. Bressan, D. Serre, M. Williams,
I.A. Shishmarev, Asymptotics for Dissipative Nonlinear K. Zumbrun, Hyperbolic Systems of Balance Laws.
Equations (2006) Cetraro, Italy 2003. Editor: P. Marcati (2007)
Vol. 1885: A. Telcs, The Art of Random Walks (2006) Vol. 1912: V. Berinde, Iterative Approximation of Fixed
Vol. 1886: S. Takamura, Splitting Deformations of Dege- Points (2007)
nerations of Complex Curves (2006) Vol. 1913: J.E. Marsden, G. Misioek, J.-P. Ortega,
Vol. 1887: K. Habermann, L. Habermann, Introduction to M. Perlmutter, T.S. Ratiu, Hamiltonian Reduction by
Symplectic Dirac Operators (2006) Stages (2007)
Vol. 1888: J. van der Hoeven, Transseries and Real Dif- Vol. 1914: G. Kutyniok, Afne Density in Wavelet
ferential Algebra (2006) Analysis (2007)
Vol. 1889: G. Osipenko, Dynamical Systems, Graphs, and Vol. 1915: T. Bykoglu, J. Leydold, P.F. Stadler,
Algorithms (2006) Laplacian Eigenvectors of Graphs. Perron-Frobenius and
Vol. 1890: M. Bunge, J. Funk, Singular Coverings of Faber-Krahn Type Theorems (2007)
Toposes (2006) Vol. 1916: C. Villani, F. Rezakhanlou, Entropy Methods
Vol. 1891: J.B. Friedlander, D.R. Heath-Brown, for the Boltzmann Equation. Editors: F. Golse, S. Olla
H. Iwaniec, J. Kaczorowski, Analytic Number Theory, (2008)
Cetraro, Italy, 2002. Editors: A. Perelli, C. Viola (2006) Vol. 1917: I. Veselic, Existence and Regularity Prop-
Vol. 1892: A. Baddeley, I. Brny, R. Schneider, W. Weil, erties of the Integrated Density of States of Random
Stochastic Geometry, Martina Franca, Italy, 2004. Editor: Schrdinger (2008)
W. Weil (2007) Vol. 1918: B. Roberts, R. Schmidt, Local Newforms for
Vol. 1893: H. Hanmann, Local and Semi-Local Bifur- GSp(4) (2007)
cations in Hamiltonian Dynamical Systems, Results and Vol. 1919: R.A. Carmona, I. Ekeland, A. Kohatsu-
Examples (2007) Higa, J.-M. Lasry, P.-L. Lions, H. Pham, E. Tain,
Vol. 1894: C.W. Groetsch, Stable Approximate Evalua- Paris-Princeton Lectures on Mathematical Finance 2004.
tion of Unbounded Operators (2007) Editors: R.A. Carmona, E. inlar, I. Ekeland, E. Jouini,
Vol. 1895: L. Molnr, Selected Preserver Problems on J.A. Scheinkman, N. Touzi (2007)
Algebraic Structures of Linear Operators and on Function Vol. 1920: S.N. Evans, Probability and Real Trees. Ecole
Spaces (2007) dt de Probabilits de Saint-Flour XXXV-2005 (2008)
Vol. 1921: J.P. Tian, Evolution Algebras and their Appli- Vol. 1945: F. Brauer, P. van den Driessche, J. Wu (Eds.),
cations (2008) Mathematical Epidemiology (2008)
Vol. 1922: A. Friedman (Ed.), Tutorials in Mathematical Vol. 1946: G. Allaire, A. Arnold, P. Degond, T.Y. Hou,
BioSciences IV. Evolution and Ecology (2008) Quantum Transport. Modelling, Analysis and Asymptot-
Vol. 1923: J.P.N. Bishwal, Parameter Estimation in ics. Cetraro, Italy 2006. Editors: N.B. Abdallah, G. Fros-
Stochastic Differential Equations (2008) ali (2008)
Vol. 1924: M. Wilson, Littlewood-Paley Theory and Vol. 1947: D. Abramovich, M. Marino, M. Thaddeus,
Exponential-Square Integrability (2008) R. Vakil, Enumerative Invariants in Algebraic Geo-
Vol. 1925: M. du Sautoy, L. Woodward, Zeta Functions metry and String Theory. Cetraro, Italy 2005. Editors:
of Groups and Rings (2008) K. Behrend, M. Manetti (2008)
Vol. 1926: L. Barreira, V. Claudia, Stability of Nonauto- Vol. 1948: F. Cao, J-L. Lisani, J-M. Morel, P. Mus,
nomous Differential Equations (2008) F. Sur, A Theory of Shape Identication (2008)
Vol. 1927: L. Ambrosio, L. Caffarelli, M.G. Crandall, Vol. 1949: H.G. Feichtinger, B. Helffer, M.P. Lamoureux,
L.C. Evans, N. Fusco, Calculus of Variations and Non- N. Lerner, J. Toft, Pseudo-Differential Operators. Quan-
Linear Partial Differential Equations. Cetraro, Italy 2005. tization and Signals. Cetraro, Italy 2006. Editors: L.
Editors: B. Dacorogna, P. Marcellini (2008) Rodino, M.W. Wong (2008)
Vol. 1928: J. Jonsson, Simplicial Complexes of Graphs Vol. 1950: M. Bramson, Stability of Queueing Networks,
(2008) Ecole dEt de Probabilits de Saint-Flour XXXVI-2006
Vol. 1929: Y. Mishura, Stochastic Calculus for Fractional (2008)
Brownian Motion and Related Processes (2008) Vol. 1951: A. Molt, J. Orihuela, S. Troyanski,
M. Valdivia, A Non Linear Transfer Technique for
Vol. 1930: J.M. Urbano, The Method of Intrinsic Scaling.
Renorming (2008)
A Systematic Approach to Regularity for Degenerate and
Vol. 1952: R. Mikhailov, I.B.S. Passi, Lower Central and
Singular PDEs (2008)
Dimension Series of Groups (2008)
Vol. 1931: M. Cowling, E. Frenkel, M. Kashiwara,
Vol. 1953: K. Arwini, C.T.J. Dodson, Information Geo-
A. Valette, D.A. Vogan, Jr., N.R. Wallach, Representation
metry (2008)
Theory and Complex Analysis. Venice, Italy 2004.
Vol. 1954: P. Biane, L. Bouten, F. Cipriani, N. Konno,
Editors: E.C. Tarabusi, A. DAgnolo, M. Picardello
N. Privault, Q. Xu, Quantum Potential Theory. Editors:
(2008)
U. Franz, M. Schuermann (2008)
Vol. 1932: A.A. Agrachev, A.S. Morse, E.D. Sontag, Vol. 1955: M. Bernot, V. Caselles, J.-M. Morel, Optimal
H.J. Sussmann, V.I. Utkin, Nonlinear and Optimal Con- Transportation Networks (2008)
trol Theory. Cetraro, Italy 2004. Editors: P. Nistri, G. Ste- Vol. 1956: C.H. Chu, Matrix Convolution Operators on
fani (2008) Groups (2008)
Vol. 1933: M. Petkovic, Point Estimation of Root Finding Vol. 1957: A. Guionnet, On Random Matrices: Macro-
Methods (2008) scopic Asymptotics, Ecole dEt de Probabilits de Saint-
Vol. 1934: C. Donati-Martin, M. mery, A. Rouault, Flour XXXVI-2006 (2008)
C. Stricker (Eds.), Sminaire de Probabilits XLI (2008) Vol. 1958: M.C. Olsson, Compactifying Moduli Spaces
Vol. 1935: A. Unterberger, Alternative Pseudodifferential for Abelian Varieties (2008)
Analysis (2008) Vol. 1959: Y. Nakkajima, A. Shiho, Weight Filtrations
Vol. 1936: P. Magal, S. Ruan (Eds.), Structured Popula- on Log Crystalline Cohomologies of Families of Open
tion Models in Biology and Epidemiology (2008) Smooth Varieties (2008)
Vol. 1937: G. Capriz, P. Giovine, P.M. Mariano (Eds.),
Mathematical Models of Granular Matter (2008)
Vol. 1938: D. Auroux, F. Catanese, M. Manetti, P. Seidel,
B. Siebert, I. Smith, G. Tian, Symplectic 4-Manifolds
and Algebraic Surfaces. Cetraro, Italy 2003. Editors:
F. Catanese, G. Tian (2008)
Recent Reprints and New Editions
Vol. 1939: D. Bof, F. Brezzi, L. Demkowicz, R.G. Vol. 1702: J. Ma, J. Yong, Forward-Backward Stochas-
Durn, R.S. Falk, M. Fortin, Mixed Finite Elements, tic Differential Equations and their Applications. 1999
Compatibility Conditions, and Applications. Cetraro, Corr. 3rd printing (2007)
Italy 2006. Editors: D. Bof, L. Gastaldi (2008) Vol. 830: J.A. Green, Polynomial Representations of
Vol. 1940: J. Banasiak, V. Capasso, M.A.J. Chap- GLn , with an Appendix on Schensted Correspondence
lain, M. Lachowicz, J. Miekisz, Multiscale Problems in and Littelmann Paths by K. Erdmann, J.A. Green and
the Life Sciences. From Microscopic to Macroscopic. M. Schoker 1980 2nd corr. and augmented edition
Bedlewo, Poland 2006. Editors: V. Capasso, M. Lachow- (2007)
icz (2008) Vol. 1693: S. Simons, From Hahn-Banach to Monotonic-
Vol. 1941: S.M.J. Haran, Arithmetical Investigations. ity (Minimax and Monotonicity 1998) 2nd exp. edition
Representation Theory, Orthogonal Polynomials, and (2008)
Quantum Interpolations (2008) Vol. 470: R.E. Bowen, Equilibrium States and the Ergodic
Vol. 1942: S. Albeverio, F. Flandoli, Y.G. Sinai, SPDE in Theory of Anosov Diffeomorphisms. With a preface by
Hydrodynamic. Recent Progress and Prospects. Cetraro, D. Ruelle. Edited by J.-R. Chazottes. 1975 2nd rev.
Italy 2005. Editors: G. Da Prato, M. Rckner (2008) edition (2008)
Vol. 1943: L.L. Bonilla (Ed.), Inverse Problems and Imag- Vol. 523: S.A. Albeverio, R.J. Hegh-Krohn, S. Maz-
ing. Martina Franca, Italy 2002 (2008) zucchi, Mathematical Theory of Feynman Path Integral.
Vol. 1944: A. Di Bartolo, G. Falcone, P. Plaumann, 1976 2nd corr. and enlarged edition (2008)
K. Strambach, Algebraic Groups and Lie Groups with Vol. 1764: A. Cannas da Silva, Lectures on Symplectic
Few Factors (2008) Geometry 2001 Corr. 2nd printing (2008)
LECTURE NOTES IN MATHEMATICS 123
Edited by J.-M. Morel, F. Takens, B. Teissier, P.K. Maini
1. Lecture Notes aim to report new developments in all areas of mathematics and their
applications - quickly, informally and at a high level. Mathematical texts analysing new
developments in modelling and numerical simulation are welcome.
Monograph manuscripts should be reasonably self-contained and rounded off. Thus
they may, and often will, present not only results of the author but also related work
by other people. They may be based on specialised lecture courses. Furthermore, the
manuscripts should provide sufcient motivation, examples and applications. This clearly
distinguishes Lecture Notes from journal articles or technical reports which normally are
very concise. Articles intended for a journal but too long to be accepted by most journals,
usually do not have this lecture notes character. For similar reasons it is unusual for
doctoral theses to be accepted for the Lecture Notes series, though habilitation theses may
be appropriate.
2. Manuscripts should be submitted either to Springers mathematics editorial in Heidelberg,
or to one of the series editors. In general, manuscripts will be sent out to 2 external referees
for evaluation. If a decision cannot yet be reached on the basis of the rst 2 reports, further
referees may be contacted: The author will be informed of this. A nal decision to publish
can be made only on the basis of the complete manuscript, however a refereeing process
leading to a preliminary decision can be based on a pre-nal or incomplete manuscript.
The strict minimum amount of material that will be considered should include a detailed
outline describing the planned contents of each chapter, a bibliography and several sample
chapters.
Authors should be aware that incomplete or insufciently close to nal manuscripts
almost always result in longer refereeing times and nevertheless unclear referees recom-
mendations, making further refereeing of a nal draft necessary.
Authors should also be aware that parallel submission of their manuscript to another
publisher while under consideration for LNM will in general lead to immediate rejection.
3. Manuscripts should in general be submitted in English. Final manuscripts should contain
at least 100 pages of mathematical text and should always include
a table of contents;
an informative introduction, with adequate motivation and perhaps some historical re-
marks: it should be accessible to a reader not intimately familiar with the topic treated;
a subject index: as a rule this is genuinely helpful for the reader.
For evaluation purposes, manuscripts may be submitted in print or electronic form, in
the latter case preferably as pdf- or zipped ps-les. Lecture Notes volumes are, as a rule,
printed digitally from the authors les. To ensure best results, authors are asked to use
the LaTeX2e style les available from Springers web-server at:
ftp://ftp.springer.de/pub/tex/latex/svmonot1/ (for monographs).
Additional technical instructions, if necessary, are available on request from:
lnm@springer.com.
4. Careful preparation of the manuscripts will help keep production time short besides en-
suring satisfactory appearance of the nished book in print and online. After acceptance
of the manuscript authors will be asked to prepare the nal LaTeX source les (and also
the corresponding dvi-, pdf- or zipped ps-le) together with the nal printout made from
these les. The LaTeX source les are essential for producing the full-text online version
of the book (see www.springerlink.com/content/110312 for the existing online volumes
of LNM).
The actual production of a Lecture Notes volume takes approximately 12 weeks.
5. Authors receive a total of 50 free copies of their volume, but no royalties. They are entitled
to a discount of 33.3% on the price of Springer books purchased for their personal use, if
ordering directly from Springer.
6. Commitment to publish is made by letter of intent rather than by signing a formal contract.
Springer-Verlag secures the copyright for each volume. Authors are free to reuse material
contained in their LNM volumes in later publications: a brief written (or e-mail) request
for formal permission is sufcient.
Addresses:
Professor J.-M. Morel, CMLA,