Weight Filtrations

Lecture Notes in Mathematics 1959
Editors:
J.-M. Morel, Cachan
F. Takens, Groningen
B. Teissier, Paris
Yukiyoshi Nakkajima Atsushi Shiho
Weight Filtrations on Log

Crystalline Cohomologies
of Families of
Open Smooth Varieties
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
ISBN: 978-3-540-70564-2 e-ISBN: 978-3-540-70565-9

DOI: 10.1007/978-3-540-70565-9
Lecture Notes in Mathematics ISSN print edition: 0075-8434

ISSN electronic edition: 1617-9692
Library of Congress Control Number: 2008932186
Mathematics Subject Classication (2000): 14F30 (Primary), 14F40, 13K05 (Secondary)
c 2008 Springer-Verlag Berlin Heidelberg

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,
reproduction on microlm or in any other way, and storage in data banks. Duplication of this publication
or parts thereof is permitted only under the provisions of the German Copyright Law of September 9,
1965, in its current version, and permission for use must always be obtained from Springer. Violations
are liable to prosecution under the German Copyright Law.
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.
Cover design: SPi Publishing Services
Printed on acid-free paper
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.
Tokyo Yukiyoshi Nakkajima

January 2008 Atsushi Shiho
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
1 Preliminaries on Filtered Derived Categories and Topoi . . 15

1.1 Filtered Derived Category. I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2 Filtered Derived Category. II.RHom and L . . . . . . . . . . . . . . 26
1.3 Filtered Derived Category. III . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.4 Some Remarks on Filtered Derived Categories . . . . . . . . . . . . . 37
1.5 The Topos Associated to a Diagram of Topoi. I . . . . . . . . . . . . 41
1.6 The Topos Associated to a Diagram of Topoi. II . . . . . . . . . . . . 45
2 Weight Filtrations on Log Crystalline Cohomologies . . . . . . 55

2.1 Exact Closed Immersions, SNCDs and Admissible
Immersions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.2 The Log Linearization Functor . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.3 Forgetting Log Morphisms and Vanishing Cycle Sheaves . . . . . 84
2.4 Preweight-Filtered Restricted Crystalline and Zariskian
Complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.5 Well-Denedness of the Preweight-Filtered Restricted
Crystalline and Zariskian Complexes . . . . . . . . . . . . . . . . . . . . . . 102
2.6 The Preweight Spectral Sequence . . . . . . . . . . . . . . . . . . . . . . . . . 112
2.7 The Vanishing Cycle Sheaf and the Preweight Filtration . . . . . 116
2.8 Boundary Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
2.9 The Functoriality of the Preweight-Filtered Zariskian
Complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
2.10 The Base Change Theorem and the K unneth Formula . . . . . . . 143
2.11 Log Crystalline Cohomology with Compact Support . . . . . . . . 156
2.12 Filtered Log de Rham-Witt Complex . . . . . . . . . . . . . . . . . . . . . 175
2.13 Filtered Convergent F -isocrystal . . . . . . . . . . . . . . . . . . . . . . . . . 192
2.14 Specialization Argument in Log Crystalline Cohomology . . . . . 196
2.15 The E2 -degeneration of the p-adic Weight Spectral Sequence
of an Open Smooth Variety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
ix
x Contents
2.16 The Filtered Log Berthelot-Ogus Isomorphism . . . . . . . . . . . . . 202

2.17 The E2 -degeneration of the p-adic Weight Spectral Sequence
of a Family of Open Smooth Varieties . . . . . . . . . . . . . . . . . . . . . 205
2.18 Strict Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
2.19 The Weight-Filtered Poincare Duality . . . . . . . . . . . . . . . . . . . . . 215
2.20 l-adic Weight Spectral Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . 216
3 Weight Filtrations and Slope Filtrations on Rigid

Cohomologies (Summary) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
3.1 Preliminaries for Later Sections . . . . . . . . . . . . . . . . . . . . . . . . . . 219
3.2 Rigid Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
3.3 Comparison Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
3.4 Weight Filtration on Rigid Cohomology . . . . . . . . . . . . . . . . . . . 234
3.5 Slope Filtration on Rigid Cohomology . . . . . . . . . . . . . . . . . . . . . 240
3.6 Weight Filtration on Rigid Cohomology
with Compact Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
A Relative SNCD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Introduction
Though they are vague, we have the following dreams as in [37], [22] and [25]:
(1) Cohomologies in characteristic 0 of algebro-geometric objects in any

characteristic have remarkable increasing ltrations, which are called weight
ltrations.
(2) They are motivic.
(3) They are constructible sheaves. Sometimes they are, in fact, smooth
sheaves.
(4) They are compatible with canonical operations, e.g., base change,
Kunneth formula, Poincare duality.
(5) They are functorial: certain classes of morphisms (e.g., the induced
morphisms by the morphisms of algebro-geometric objects) are strictly com-
patible with them.
In this book, for a family of open smooth varieties in characteristic p > 0,
we discuss the p-adic aspects of (1), (3), (4), (5) and the following new aspect:
(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
is obtained by counting the number of the logarithmic poles of local sections

of iX/C (log D). Let a(k) : D(k) X (k N) be the natural morphism of
schemes over C. Then we have the following isomorphism of complexes (the
Poincare residue isomorphism):
k (k)
k X/C (log D) a (D (k) /C Z
Res : grP (k)
(0.0.0.1) (D/C)(k)).
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)).
Moreover we have the following isomorphisms in the ltered derived category

of bounded below ltered complexes of CXan -modules:

(0.0.0.3) (Xan /C (log Dan ), P ) (Xan /C (log Dan ), )

(jan (Uan /C ), ) (Rjan (CUan ), ) = (Rjan (ZUan ) Z C, ),
where := {k }kZ is the canonical ltration. By using the exponential
sequence on Uan and the cup product, we have the purity isomorphism
(k)
(0.0.0.4) Rk jan (ZUan ) aan ((k) (Dan /C))(k) (k N)
(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)
is equal to the multiplication by (1)k . Hence we use the following isomor-

phism
(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
instead of (0.0.0.4). The isomorphism (0.0.0.6) is equal to the isomorphism

in [23, (3.1.9)]. Using the Leray spectral sequence for two functors jan and
(Xan , ?), using the isomorphism (0.0.0.6) and renumbering Erk,h+k :=
hk,k
Er+1 , we have the following spectral sequence
(0.0.0.7) E1k,h+k = H hk ((Dan )(k) , (k) (Dan /C))(k) = H h (Uan , Z).
If X is proper, one obtains the weight ltration on H h (Uan , Z) by the spectral

sequence (0.0.0.7); if X is proper, (0.0.0.2) is equal to (0.0.0.7)Z C by GAGA
and the Poincare lemma.
In fact, the existence of the weight ltration above modulo torsion has been
generalized to the case of a family in characteristic 0 by Deligne as follows.
Let f : X S be a proper smooth morphism of schemes of nite type
over C. Let D be a relative simple normal crossing divisor on X over S. Set
U := X \ D, and let f also denote the structural morphism f : U S. Then
Rh fan (QUan ) is a local system ([21, II (6.14)]), and there exists a ltration P
on Rh fan (QUan ) by sub local systems such that the induced ltration Ps on
the stalk Rh fan (QUan )s = H h ((Uan )s , Q) (s San ) ([loc. cit.]) is obtained
from the spectral sequence (0.0.0.7) ([25]).
Now let us turn to the case of characteristic p > 0.
Let (S, I, ) be a PD-scheme with a quasi-coherent ideal sheaf I. Set
S0 := SpecS (OS /I). Assume that p is locally nilpotent on S. Let f : X S0
be a smooth scheme with a relative simple normal crossing divisor D on X
over S0 (by abuse of notation, we denote by the same symbol f the composite
f
morphism X S0 S). Then the pair (X, D) of S0 -schemes denes an
fs(=ne and saturated) log scheme (X, M (D)) over S0 (2.1 below, cf. [54])
in the sense of Fontaine-Illusie-Kato and (X, D)/S denes a log crystalline

topos ((X, M (D))/S)log ([54], cf. [29]). By abuse of notation, we often denote
crys
(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
in the classical crystalline topos (D(k) /S)

crys . (See (2.2.13.2) and (2.2.15)
below for the precise denition of D(k) for a nonnegative integer k.) Let

crys Szar and fD (k) /S : (D
D)/S)log crys Szar be the
f(X,D)/S : ((X, (k) /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:
(0.0.0.8) E1k,h+k = Rhk fD(k) /S (OD(k) /S Z crys

(k)
(D/S))(k)
= Rh f(X,D)/S (O(X,D)/S ).
Here crys (D/S) is the crystalline orientation sheaf of D/S in (D

(k) (k) /S)
crys
which will be dened in 2.2. If S0 is of characteristic p > 0, then the relative
4 Introduction
Frobenius morphism F : (X, D) (X , D ) over S0 induces the relative

Frobenius morphism F (k) : D(k) D(k) = D(k) . Let a(k) : D(k) X

and a(k) : D(k) X be the natural morphisms. We dene the relative
Frobenius action
(k)
(D /S) Fcrys acrys crys
(k)
(k) : acrys crys
(k) (k)
(D/S)
as the identity under the natural identication

(D /S) Fcrys crys
(k) (k) (k)
crys (D/S).
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
E1k,h+k = H hk ((D n crys , OD (k) /Wn Z crys (D/Wn ))(k)

(k) /W ) (k)
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
To construct (0.0.0.8) and to prove the functoriality of it, we dene two

ltered complexes
(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 ))
and construct two other ltered complexes
(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 ).
{D } be a decomposition of D by smooth compo-

(0.0.0.14): Let :=
nents of D: D = D and each D is smooth over S0 . If (X, D) has

an admissible immersion (X, D) (X , D) over S with respect to (see
(2.1.10) below for the denition of the admissible immersion), then
(CRcrys (O(X,D)/S ), P ) (QX/S LX/S (X /S (log D)),

{QX/S LX/S (Pk X /S (log D))}kZ ),
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 } ,
(Czar (O(X,D)/S ), P ) (OD OX X /S (log D), {OD OX Pk X /S (log D)}kZ ),

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
(0.0.0.21): If S0 is the spectrum of a perfect eld of characteristic p and if

S = Spec(Wn ()) (n > 0), then (Czar (O(X,D)/S ), P ) is canonically isomorphic
to the ltered complex (Wn X (log D), P ) := (Wn X (log D), {Pk Wn X (log
D)}kZ ) in [65].
Thus we obtain the following translation which let us recall Grothendiecks
project to unify algebra, geometry and analysis ([37]):
(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
is: if we can do something by using local lifts, we can do something analogous

and more by using admissible immersions. Since a standard exactication
of the product of two local lifts of (X, D) is not a local lift of (X, D) at
all, the notion local lift is not exible for the construction of the spectral
sequence (0.0.0.8). Moreover, because we do not take a local lift of (X, D),
we can give a simple proof of the independence of (CRcrys (O(X,D)/S ), P ) and
(Czar (O(X,D)/S ), P ) of the choice of the open covering of (X, D) and that of
the admissible immersion of each open log scheme: we do not need a concrete
(slightly) laboring key calculation in [47]; in a future paper we shall develop
analogous theory for a family of simple normal crossing log varieties over log
points, and we shall show that a concrete key calculation in [48] and [64] is not
necessary. Furthermore, we come to know that the ltered log de Rham-Witt
complex of an open variety is not a necessary ingredient for the construction
of (0.0.0.8) (in the special case S = Spec(Wn ())). We are sure that it is
natural to capture something producing (0.0.0.8) as an object in a ltered
derived category; to capture it as a real ltered complex is not exible. How-
ever, because it is anyway possible to capture it as a real ltered complex in
the case above, it gives us something deep in a special case. Indeed, in [70],
we use the simplicial version of the ltered log de Rham-Witt complex above
for the proof of a variant of the Serre-Grothendieck formula on the virtual
Betti numbers of a separated scheme of nite type over , which has been
conjectured in [37].
From the ltered object (Czar (O(X,D)/S ), P ), we immediately obtain the
spectral sequence (0.0.0.8) by (0.0.0.19). If one wishes to obtain only (0.0.0.8),
the object (Czar (O(X,D)/S ), P ) is enough; however, it is an important fact that
something producing (0.0.0.8) exists not only in D+ F(f 1 (OS )) but also in
a higher stage D+ F(QX/S (OX/S )).
By (0.0.0.17), it is well-known that the canonical ltration and the
preweight ltration P on Czar (O(X,D)/S ) do not coincide in general; how-
ever, impressively, and P on CRcrys (O(X,D)/S ) coincide ((0.0.0.20)). The
equality (0.0.0.20) follows from the following p-adic purity
(0.0.0.23) QX/S Rk (X,D)/S (O(X,D)/S )

= QX/S acrys (OD(k) /S Z crys
(k) (k)
(D/S))(k) (k N),
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
Kato, we have a natural morphism
g : (Czar (O(X ,D )/S ), P ) Rg (Czar (O(X,D)/S ), P ).
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 .
Now we outline the contents of this book.

In Chapter 1, we prove preliminary results which we use in later chapters.
From 1.1 to 1.4, we show some facts on ltered modules in a ringed
topos which are necessary for later sections. Many key notions and many
results are due to P. Berthelot. Especially, the notions (co)special modules and
strictly injective resolutions are due to him. The notion strictly at resolutions
is also due to him. The ltered adjunction formula, which is also due to
him, is a key ingredient for the proof of the ltered base change theorem of
(Ecrys (O(X,D)/S ), P ).
In 1.5 and 1.6 we review general facts on diagrams of topoi for this book
and future papers.
In Chapter 2, which is the main body of this book, we construct the theory
of the weight ltration of the log crystalline cohmologies of families of open
smooth varieties.
In 2.1 we give the denition of a relative simple normal crossing divisor
and a key notion admissible immersion and give a local description of the
admissible immersion.
In 2.2 we recall a (log) linearization functor and we calculate the graded
pieces of QX/S LX/S (X /S (log D)).
In 2.3 we prove a crystalline Poincare lemma for RY /S (E) for a mor-
phism g : Y S0 of ne log schemes which can be embedded into a ne log

smooth scheme Y/S whose underlying scheme Y/S is also smooth and for a
crystal E of OY /S -modules.
In 2.4 we construct (CRcrys (O(X,D)/S ), P ) and (Czar (O(X,D)/S ), P ) for an
open covering of X and an admissible immersion of each open log subscheme
of (X, D). In 2.5 we prove the independence of (CRcrys (O(X,D)/S ), P ) and
(Czar (O(X,D)/S ), P ) of the choice of the open covering of X and that of the
admissible immersion of each open log subscheme of (X, D).
In 2.6 we prove (0.0.0.18) and (0.0.0.19). In 2.7 we calculate the re-
striction of the vanishing cycle sheaves QX/S Rk (X,D)/S (O(X,D)/S ) (k Z)
and we prove the p-adic purity (0.0.0.23). Then we prove (0.0.0.13) and
(0.0.0.20); as a corollary, we immediately see that (CRcrys (O(X,D)/S ), P ) and
10 Introduction
(Czar (O(X,D)/S ), P ) are independent of the choice of the decomposition of D

by smooth components of D.
In 2.8 we give the description of the boundary morphism of the E1 -terms
of the spectral sequence (0.0.0.8).
In 2.9 we prove the functoriality of (Czar (O(X,D)/S ), P ).
In 2.10 we prove the ltered base change theorem of (Ecrys (O(X,D)/S ), P ),
and we prove the ltered K unneth formula of (Ecrys (O(X,D)/S ), P ). As in
[3] and [11], we have some important corollaries of the ltered base change
theorem.
In 2.11 we develop the analogous theory for a log crystalline cohomol-
ogy with compact support, and, especially, we obtain the preweight spectral
sequence of a log crystalline cohomology with compact support.
In 2.12 we prove that, if S0 is the spectrum of a perfect eld of character-
istic p > 0 and if S = Spec(Wn ()), then (Czar (O(X,D)/S ), P ) is canonically
isomorphic to (Wn X (log D), P ).
In 2.13 we prove the convergence of the weight ltration as a corollary
of the ltered base change theorem in 2.10. This is a ltered version of the
convergence of a (log) crystalline cohomology in [74], [76] and [29].
In 2.14 we give a specialization argument of Deligne-Illusie ([49]) in our
situation.
In 2.15 we prove the E2 -degeneration of (0.0.0.8) modulo torsion in the
case where S0 := Spec() and S = Spf(W ()). The proof in this book is
another, more natural proof than the proof in [68]; In 2.16 we prove that a
ltered log Berthelot-Ogus isomorphism is strictly compatible with the weight
ltration. In 2.17 we generalize the E2 -degeneration of (0.0.0.8) modulo
torsion to the case where the base scheme S is a p-adic formal V -scheme
in the sense of [74], by using the results in the previous two sections. Using
this generalized result, we reprove the convergence of the weight ltration by
reducing it to a result in [74].
In 2.18 we prove the strict compatibility of the weight ltration with
respect to the induced morphism of log schemes by using the specialization
argument in 2.14 and the convergence of the weight ltration.
In 2.19 we prove the strict compatibility of the weight ltration with the
Poincare duality.
In 2.20 we give a remark on the corresponding l-adic weight spectral
sequence.
Following the suggestion of one of the referees, we state some results ob-
tained in [70] in Chapter 3 to answer natural questions arising from results
in Chapter 2.
For a (separated) scheme of nite type U over C, P. Deligne has endowed
H h (Uan , Q) (h Z) with the mixed Hodge structure in [24] by using results
in [23]. In [70] the rst-named author has succeeded in dening the weight
ltration on the rigid cohomology of a separated scheme U of nite type over
a perfect eld of characteristic p > 0 by using de Jongs alteration theorem
Introduction 11
([28]), Tsuzukis proper descent ([87]), Shihos comparison theorems ([82])

and results up to Chapter 2.
In 3.1 we give preliminaries for later sections: the single complex of a
complex in a multisimplicial topos in [24], the diagonal ltration in [loc. cit.],
and the key functor in [19], [87] and [88].
In 3.2 we give a reformulation of Tsuzukis proper descent on rigid coho-
mology.

Let U U be an open immersion into a proper scheme over . In 3.3
we state comparison theorems between the rigid cohomology of U , the log
convergent cohomology of a certain split proper hypercovering of (U, U ) and
the log crystalline cohomology of the proper hypercovering. As an application
of the comparison theorems, we can endow the rigid cohomology with closed
support with the weight ltration (3.4). As another application, we can cal-
culate the slope ltration on the rigid cohomology by the log Hodge-Witt
sheaves of the proper hypercovering (3.5). In 3.6 we state the existence
theorem of the weight ltration on the rigid cohomology with compact sup-
port.
Finally we add the Appendix: we give some results on relative simple
normal crossing divisors which are used in the text or related to it.
In a future paper we shall construct an analogous theory for a family of
semistable varieties, which is a generalization of [64] and [68].

Notations. (1) For a log scheme Y , Y denotes the underlying scheme of

Y . For a morphism f : Y Z of log schemes, f denotes the underlying

morphism Y Z of schemes.
(2) (S)NCD=(simple) normal crossing divisor.
Conventions. We assume that the log structures on (formal) schemes are

dened on Zariski site unless otherwise stated.
Also, we make the following conventions about signs. Let A be an additive
exact category.
(1) For a complex (E , d ) of objects in A and for an integer n,
+n
(E , d+n ) or (E {n}, d {n}) denotes the following complex:
dq1+n dq+n dq+1+n

E q1+n E q+n E q+1+n .
q1 q q+1
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
(2) For a complex (E , d ) of objects in A and for an integer n,

(E [n], d [n]) denotes the following complex as usual: (E [n])q := E q+n
with boundary morphism d [n] = (1)n d+n .
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]) induced
by f without change of signs. This operation is well-dened in the derived
category as in (1).
(3) ([10, 0.3.2], [18, (1.3.2)]) For a short exact sequence
f g
0 (E , dE ) (F , dF ) (G , dG ) 0
of bounded below complexes of objects in A, let (E [1], dE [1]) (F , dF ) be

the mapping cone of f . We x an isomorphism (E [1], dE [1]) (F , dF )
(x, y)
g(y) (G , dG ) in the derived category D+ (A).
(4) ([10, 0.3.2], [18, (1.3.3)]) Under the situation (3), the boundary mor-
phism (G , dG ) (E [1], dE [1]) in D+ (A) is the following composite mor-
phism
proj. (1)
(G , dG ) (E [1], dE [1]) (F , dF ) (E [1], dE [1]) (E [1], dE [1]).
(5) Let A be an abelian category with enough injectives. Let F : A B

be a left exact functor of abelian categories. Then, in the situation (3), the
boundary morphism : Rq F((G , dG )) Rq+1 F((E , dE )) of cohomolo-
gies is, by denition, the induced morphism by the morphism (G , dG )
(E [1], dE [1]) in (4). By taking injective resolutions (I , dI ), (J , dJ ) and
(K , dK ) of (E , dE ), (F , dF ) and (G , dG ), respectively, which t into the
following commutative diagram
(0.0.0.24)
0 (I , dI ) (J , dJ ) (K , dK ) 0

0 (E , dE ) (F , dF ) (G , dG ) 0
of complexes of objects in A such that the upper horizontal sequence is exact,

it is easy to check that the boundary morphism above is equal to the
usual boundary morphism obtained from the upper short exact sequence of
(0.0.0.24). (For a short exact sequence in (3), the existence of the commutative
diagram (0.0.0.24) has been proved in, e.g., (1.1.7) below, as a very special
case.)
(6) For a complex (E , d ) of objects in A, the identity id : E q E q

(q Z) induces an isomorphism Hq ((E , d )) Hq ((E , d )) (q Z)
of cohomologies. We sometimes use this convention.
(7) We often denote a complex (E , d ) simply by (E , d) or E as usual
when there is no risk of confusion.
Introduction 13
(8) Let r 2 be a positive integer. As usual, an r-uple complex of objects

in A is, by denition, a pair (E , {di }ri=1 ) such that E m1 mr (mi Z) is an
object of A with morphisms di : E ,mi , E ,mi +1, satisfying
the following relations d2i = 0 and di dj + dj di = 0 (i = j).
Chapter 1
Preliminaries on Filtered Derived
Categories and Topoi
In this chapter, we prove some preliminary results on ltered derived

categories and topoi which we need in later chapters. In the rst four
sections, we recall several notions and basic properties concerning ltered de-
rived categories: ltered complex, ltered quasi-isomorphism, ltered derived
category, ltered injective resolution, ltered at resolution, ltered derived
functor and ltered adjunction formula. It is important for us because our
weight-ltered complexes will be dened as objects in certain ltered derived
categories. In the last two sections, we recall the notion of topoi associated to
diagrams of topoi and prove several basic properties of the topoi associated
to diagrams of (restricted) log crystalline topoi.
1.1 Filtered Derived Category. I
In this section we recall the denition of a ltered derived category and

calculate derived functors for complexes of ltered modules in a ringed topos
(cf. [24, 7], [60], [78], [8]). Many facts in this section are due to Berthelot ([8]).
Though many facts in this section hold for abelian categories with enough
injectives by using a standard technique reducing the general case to the case
of the category of usual modules over a ring with unit element (e.g., [90,
Remark on p. 12]), we are content with the case of a ringed topos.
Let (T , A) be a ringed topos. Let E be an A-module in T . In this book, a
ltration on E is, by denition, a family {Ek } = {Ek }kZ of A-submodules of
E such that Ek Ek+1 for all k Z; in this book we consider only increasing
ltrations indexed by Z except stupid decreasing ltrations in later sections.
As in [8], ltrations are not necessarily exhaustive nor separated. We denote
by (E, {Ek }) an A-module E with ltration {Ek }. We sometimes denote
(E, {Ek }) only by E if there is no risk of confusion. For an A-module E, we
mean by the trivial ltration on E a ltration {Ek } such that E0 = E and
E1 = 0. In [8], A may have a nontrivial ltration and A is not necessarily

16 1 Preliminaries on Filtered Derived Categories and Topoi
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
HomA ((E, {El }), (F, {Fl })) MF((T , A))

1.1 Filtered Derived Category. I 17
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)
is immediate to verify. Let f : (E, {El }) (F, {Fl }) be a morphism in

MF(A). We say that f is a strict morphism if the inclusion morphism

f (El ) (Im f ) Fl (l Z) is an isomorphism. We say that f is a
strict monomorphism (resp. strict epimorphism) if f is strict and if f is a
monomorphism (resp. epimorphism).
We obtain a similar object Hom A ((E, {El }), (F, {Fl })) MF(A) in a
similar way.
Following [8], let us dene the special module of (E, {El }) for A-
modules E, El (l Z): we set A (E, {El }) := E El with ltration
lZ
( A (E, {El }))k := E El . Then we have a ltered A-module
lk

(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
We recall the following denition due to Berthelot ([8]):
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.
Next, we calculate the ltered derived functor of a morphism of ringed

topoi. To do this, we consider several additive full subcategories of MF(A):
Ias (A) := {(J, {Jk }) | J and Jk (k Z) are asque A-modules},

Iinj (A) := {(J, {Jk }) | J and Jk (k Z) are injective A-modules},
Istinj (A) := {(J, {Jk }) | (J, {Jk }) is a strictly injective A-module},

Ispinj (A) := { (I, {Ik }) | I and Ik are injective A-modules (k Z)}.
A
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
is surjective. By (1.1.0.1), it suces to prove that the morphism
(1.1.2.1)
HomMF(A) ((F, {Fk }), (I, {Ik })) HomMF(A) ((E, {Ek }), (I, {Ik }))
A A
is surjective. For each integer k, the induced morphism E/Ek1 F/Fk1

is a monomorphism. Hence (1.1.2.1) is surjective by (1.1.0.2).

Denition 1.1.3. We say that a ltered module (J, {Jl }) MF(A) is

lteredly asque, lteredly injective, and specially injective if (J, {Jl })
Ias (A), (J, {Jl }) Iinj (A) and (J, {Jl }) Ispinj (A), respectively.
The category MF(A) has enough special injectives in the following sense:
Proposition 1.1.4 ([8]). For a ltered module E = (E, {Ek }) MF(A),

there exists a strict monomorphism E A (I, {Il }) with A (I, {Il })
Ispinj (A).

Proof. Let E I andE/El1 Il be
monomorphisms into injective A-
modules. Set J := I Il and Jk := I Il . There are natural monomor-
lZ lk

phisms Jk Jk+1 and Jk J. Since I and Ik (k Z) are injective
A-modules, (J, {Jk }) is an object of Ispinj (A).
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.

Denition 1.1.5. Let (E , {Ek }) be an object of K+ F(A).

(1) ([8]) We say that a ltered complex (J , {Jk }) K+ F(A) with a
ltered morphism (E , {Ek })(J , {Jk }) is a strictly injective resolution
of (E , {Ek }) if (J q , {Jkq }) Istinj (A) for all q Z and if the morphism
(E , {Ek })(J , {Jk }) is a ltered quasi-isomorphism which is a strict
monomorphism for all = q Z.
(2) We say that a ltered complex (J , {Jk }) K+ F(A) with a l-
tered morphism (E , {Ek })(J , {Jk }) is a ltered asque resolution, l-
tered injective resolution and specially injective resolution of (E , {Ek }) if
(J q , {Jkq }) Ias (A), Iinj (A) and Ispinj (A), respectively, for all q Z
and if the morphism (E , {Ek })(J , {Jk }) is a ltered quasi-isomorphism
which is a strict monomorphism for all = q Z.
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
(1)p dp0 (1)p dp1

0 (F p , {Fkp }) (I p0 , {Ikp0 }) (I p1 , {Ikp1 }) (p Z)
gives a ltered asque resolution s(I , {Ik }) of (F , {Fk }).
We prove the following (cf. (1.1.4) and [78, Lemma 1.3.3]):
Proposition 1.1.7. For a ltered complex (E , {Ek }) K+ F(A), there ex-

ists a specially injective resolution (J , {Jk }) of (E , {Ek }).
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:

(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 }).
Since E q J q is a monomorphism, the morphism E q+1 J q E q E q+1

is a monomorphism, and so is the following composite morphism

E q+1 J q E q E q+1 J q+1 .
In fact, we have a morphism (E q+1 , {Ekq+1 }) (J q+1 , {Jkq+1 }). Using a

morphism J q s
(s, 0) J q E q E q+1 and the morphism (1.1.7.2), we
have a morphism (J q , {Jkq }) (J q+1 , {Jkq+1 }). By the proof of (1.1.4) and
by the strictness of the morphism (E q+1 , {Ekq+1 }) (J q E q E q+1 , {Jkq Ekq
Ekq+1 }), the morphism (E q+1 , {Ekq+1 }) (J q+1 , {Jkq+1 }) is strict. We ob-
tain (J , {Jk }) inductively by the argument above.
We claim that (J , {Jk }) is lteredly quasi-isomorphic to (E , {Ek }). Let
be nothing or an integer k Z and let q be an integer. Because Ker(Jq
Jq+1 ) = Ker(Jq Jq Eq Eq+1 ), the morphism Ker(Eq Eq+1 )
Ker(Jq Jq+1 ) is an epimorphism; furthermore, the morphism Jq1
Jq factors through Jq1 Jq1 Eq1 Eq . Note that Jq1 Eq1 Eq Jq
is a monomorphism. Because the inverse image of Im(Jq1 Jq ) in Eq
is equal to the inverse image of Im(Jq1 Jq1 Eq1 Eq ), the morphism
Ker(Eq Eq+1 )/Im(Eq1 Eq ) Hq (J ) is an isomorphism. There-

fore we have Hq (E ) Hq (J ).

Proposition 1.1.8. Let f : (E , {Ek }) (F , {Fk }) be a morphism in

C+ F(A). Then there exists a morphism g : (J , {Jk }) (K , {Kk }) in
C+ F(A) such that (J , {Jk }) (resp. (K , {Kk })) is a specially injective reso-
lution of (E , {Ek }) (resp. (F , {Fk })), and such that g makes the following
diagram commutative:

(E , {Ek }) (J , {Jk })

f g

(F , {Fk }) (K , {Kk }).
Proof. Assume that we are given morphisms g 0 , . . . , g q . Then we have the

following commutative diagram
(E q+1 , {Ekq+1 }) (J q E q E q+1 , {Jkq Ekq Ekq+1 })

q q q+1
f q+1 g f f
(F q+1 , {Fkq+1 }) (K q F q F q+1 , {Kkq Fkq Fkq+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 .
Note that the morphism K q F q F q+1 IF q+1 is a monomorphism. Let

IF q+1 IFq+1 be a monomorphism into an injective A-module. Obviously
we have the following commutative 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 IFq+1 .
q+1
Similarly there exists a morphism (IE )k+1 (IFq+1 )k+1 of injective
A-modules tting into the following commutative diagram:

(J q E q E q+1 )/(Jkq Ekq Ekq+1 ) (IE
q+1
)k+1

g q f q f q+1

(K q F q F q+1 )/(Kkq Fkq Fkq+1 ) (IFq+1 )k+1 .
q+1 q+1 q+1
Set J q+1 := IE (IE )k , Jkq+1 := IE q+1
(IE )l , K q+1 := IFq+1
kZ lk
q+1
(IF )k and Kkq+1 := IFq+1 (IFq+1 )l . Obviously we have the following
kZ lk
commutative diagram in C+ F(A):
(E q+1 , {Ekq+1 }) (J q+1 , {Jkq+1 })

(F q+1 , {Fkq+1 }) (K q+1 , {Kkq+1 }).

By the proof of (1.1.7), we have Hq (E ) Hq (J ) and Hq (F )
Hq (K ). Repeating these processes, we have the desired ltered complexes.

For an additive full subcategory I of MF(A), let K+ F(I) be the category

of the bounded below ltered complexes whose components belong to I.
We prove the following two lemmas for (1.1.12) below and for later sections.
Lemma 1.1.9. Let (E , {Ek }) be a ltered complex of A-modules and let

(I , {Ik }) be an object of K+ F(Istinj (A)). Assume that (E , {Ek }) is strictly
exact. Let f : (E , {Ek }) (I , {Ik }) be a morphism of ltered complexes.
Then f is lteredly homotopic to zero.
Proof. By the denition of the strict injectivity, the same argument as that
in the classical case works.

Lemma 1.1.10. (1) Let (I , {Ik }) be an object of K+ F(Istinj (A)). Let

s : (E , {Ek }) (F , {Fk }) be a ltered quasi-isomorphism. Then s induces
an isomorphism

s : HomKF(A) ((F , {Fk }), (I , {Ik })) HomKF(A) ((E , {Ek }), (I , {Ik })).
(2) If a morphism s : (I , {Ik }) (E , {Ek }) is a ltered quasi-isomor-

phism from an object of K+ F(Istinj (A)) to a ltered complex of A-modules,
then s has a ltered homotopy inverse.
Proof. (1): Let (G , {Gk }) be the mapping cone of s. Since (G , {Gk })[n]
(G , {Gk }){n}, we have
(1.1.10.1) HomKF(A) ((G , Gk )[n], (I , {Ik }))

HomKF(A) ((G , Gk ){n}, (I , {Ik }))
=HomKF(A) ((G , Gk ), (I , {Ik }){n})
for any integer n. By (1.1.9) we have HomKF(A) ((G , Gk ), (I , {Ik }){n}) = 0.

By [44, I (1.1)], HomKF(A) (, (I , {Ik })) is a cohomological functor. Hence s
is an isomorphism.
(2): The proof is the same as that of [44, I (4.5)] by using (1.1.9) (cf. (1.1.11)
below).

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 .
Corollary 1.1.12. (1) The following equalities hold:
D+ F(A) = K+ F(Ias (A))(FQis) = K+ F(Iinj (A))(FQis)

= K+ F(Istinj (A)) = K+ F(Ispinj (A)).
(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
in [8] is easily veried.

Denition 1.1.13 ([8]). We say that a ltered module (Q, {Qk }) MF(A)
is strictly at if it satises the following two conditions:
(1) Q and Q/Qk are at A-modules

(2) For a strict monomorphism (E, {Ek }) (F, {Fk }), the induced mor-
phism
(QA E, {(QA E)k }) (QA F, {(QA F )k })

is a strict monomorphism.
Let us consider the following additive full subcategories of MF(A):
(1.1.13.1)
Q (A) := {(Q, {Qk }) | Q and Q/Qk (k Z) are at A-modules },
(1.1.13.2) Qst (A) := {(Q, {Qk }) | (Q, {Qk }) is a strictly at A-module },
(1.1.13.3)
Qsp (A) := {A (Q, {Qk }) | Q and Qk are at A-modules (k Z)}.
We obviously have Qst (A) Q (A).
Denition 1.1.14. We say that a ltered module (Q, {Qk }) MF(A) is

lteredly at (resp. specially at) if (Q, {Qk }) Q (A) (resp. (Q, {Qk })
Qsp (A)).
Lemma 1.1.15 ([8]). Qsp (A) Qst (A).

Proof. Let A (Q, {Ql }) be an object of Qsp (A) and let : (E, {El })
(F, {Fl }) be a strict monomorphism. Let k be an integer. Denote by the same

symbol the induced morphism A (Q, {Ql }) A E A (Q, {Ql }) A F .
We have to prove that an A-submodule (A (Q, {Ql }) A E) (A (Q, {Ql })
A F )k is equal to ((A (Q, {Ql })A E)k ). Let s be a local section of (A (Q,
{Ql }) A E) (A (Q, {Ql }) A F )k . By the denition of the ltration on the
ltered tensor product, s is a nite sum of local sections of Qi A Fj (i + j
k). Because
(Qi A Fj ) (Qi A (E)) = Qi A (Fj (E))

= Qi A (Ej ),
s is a local section of ((A (Q, {Ql }) A E)k ).

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).
Proof. Recall the functor L0 : {A-modules} {at A-modules} ([11, 7]):

for an A-module, L0 (E) is, by denition, the sheacation of the presheaf
(U
free (U, A)-module with basis (U, E) \ {0}).
The natural morphism L0 (E) E is an epimorphism.

Let Q
E and Ql E
l be epimorphisms from at A-modules.
Set R := Q Ql and Rk := Ql ; (R, {Rk }) is an object of Qsp (A).
lZ lk

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.

Denition 1.1.17. Let (E , {Ek }) be an object of K F(A).

(1) ([8]) We say that a ltered complex (Q , {Qk }) K F(A) with a
ltered morphism (Q , {Qk }) (E , {Ek }) is a strictly at resolution
of (E , {Ek }) if (Qq , {Qqk }) Qst (A) for all q Z and if the morphism
(Q , {Qk }) (E , {Ek }) is a ltered quasi-isomorphism which is a strict
epimorphism for all = q Z.
(2) We say that a ltered complex (Q , {Qk }) K F(A) with a l-
tered morphism (Q , {Qk }) (E , {Ek }) is a ltered at resolution
(resp. specially at resolution) of (E , {Ek }) if (Qq , {Qqk }) Q (A)
(resp. (Qq , {Qqk }) Qsp (A)) for all q Z and if the morphism
(Q , {Qk }) (E , {Ek }) is a ltered quasi-isomorphism which is a strict
epimorphism for all = q Z.
Proposition 1.1.18 ([8]). For a ltered complex (E , {Ek }) K F(A),

there exists a specially at resolution (Q , {Qk }) of (E , {Ek }).
Proof. (1.1.18) follows from (1.1.16) and [78, Lemma 1.3.3].

For an additive full subcategory Q of MF(A), let K F(Q) be the category

of the bounded above ltered complexes whose components belong to Q.
Corollary 1.1.19. (1) The following equalities hold:
D F(A) = K F(Q (A))(FQis) = K F(Qst (A))(FQis) = K F(Qsp (A))(FQis) .
(2) Set Q := Q (A ), Qst (A ) or Qsp (A ). Let f : (T , A)

(T , A ) be a morphism of ringed topoi. Then there exists the left de-

rived functor Lf : D F(A ) D F(A) such that Lf [(Q , {Qk })] =

[(f (Q ), {f (Qk )})] for an object (Q , {Qk }) K F(Q ).
(3) Let f : (T , A) (T , A ) and g : (T , A ) (T , A ) be morphisms
of ringed topoi. Then L(gf ) = Lf Lg .
Proof. (1) and (2) are obvious; (3) follows by setting Q := Q (A) in (2).

Remark 1.1.20. In the case of the trivial ltration, a at resolution of a com-

plex E K (A) has been given in [11, 7] by the single complex of the
double complex L (E ), though we have to change some signs of bound-
ary morphisms of L (E ) in order to make L (E ) a double complex (see
Convention (8)). We make the following convention:
(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
1.2 Filtered Derived Category. II.RHom and L
In this section we prove the adjoint property of Lf and Rf for a morphism

of topoi f ([8]). The notion of strict injectivity in (1.1.1) due to Berthelot is a
key notion for the denition of RHom and the proof for the adjoint property.
We also dene the derived functor of the ltered tensor product.
(1) RHom.
As in [44, p. 63], we set

HomnA ((E , {Ek }), (F , {Fk })) := HomA ((E q , {Ekq }), (F q+n , {Fkq+n }))
qZ
for (E , {Ek }), (F , {Fk }) CF(A). Then we have a ltered complex of

(T , A)-modules HomA ((E , {Ek }), (F , {Fk })) CF((T , A)); the bound-
ary morphism
HomnA ((E , {Ek }), (F , {Fk })) Homn+1

A ((E , {Ek }), (F , {Fk }))
is dened as in [10, p. 4] and [18, p. 10]:

dn := ((1)n+1 dqE + dq+n
F ).
qZ
This boundary morphism is dierent from that in [44, p. 64].

As in the classical case, an n-cocycle of HomA ((E , {El }), (F , {Fl }))k
corresponds to a ltered morphism E F [n]k. (Recall that k is the
shift of the ltration dened in 1.1.) An n-coboundary of HomA ((E , {El }),
1.2 Filtered Derived Category. II.RHom and L 27
(F , {Fl }))k corresponds to a morphism E F [n]k which is homotopic

to zero: dn1 (((1)n f q )q ) = (f q+1 dqE + (1)n dn+q1
F f q )q ((f q )q Homn1
A

((E , {El }), (F , {Fl }))k ). Hence
(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
HomA (, ) : KF(A) K+ F(A) KF((T , A)),
we have to check the following:
Lemma 1.2.1. Let (E , {Ek }) KF(A) be a ltered complex of A-modules

and let (I , {Ik }) be an object of K+ F(Istinj (A)). Assume that one of the
following two conditions holds.
(1) (I , {Ik }) is strictly exact.
(2) (E , {Ek }) is strictly exact.
Then HomA ((E , {Ek }), (I , {Ik })) is strictly exact.
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
(iii) the boundary morphism d : (I n , {Ikn }) (I n+1 , {Ikn+1 }) is identied

id
with the induced morphism by the morphisms J n1 0 and J n J n .
By (1.1.0.1) and (1.2.0.1), we have only to construct a ltered homotopy
for a morphism f HomCF(A) ((E , {Ek }), (I , {Ik })), which is easy.
(2): By (1.2.0.1) and by the denition of the strict injectivity, the same
argument as that in the classical case works.

By (1.2.1) we obtain the following derived functor
RHomA : DF(A) D+ F(A) DF(Ab).
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.
Theorem 1.2.2 ([8], Adjunction formula). Let f : (T , A) (T , A )

be a morphism of ringed topoi. Let (E , {Ek }) (resp. (F , {Fk })) be an object
of K F(A ) (resp. K+ F(A)). Then there exists a canonical isomorphism
RHomA (Lf (E , {Ek }), (F ,{Fk }))

RHomA ((E , {Ek }), Rf (F , {Fk }))
=
in D+ F((T , A )). The isomorphism above satises the transitive condition

(cf. [3, V Proposition 3.3.1]).
Proof. (The proof below is a ltered version of that in [3, V Proposition

3.3.1].) Let (I , {Ik }) be a strictly injective resolution of (F , {Fk }). Let
(Q , {Qk }) be a ltered at resolution of (E , {Ek }). Let (J , {Jk }) K+ F
(Istinj ) be a strictly injective resolution of f (I , {Ik }). Then we have a mor-
phism
(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 ! :
f ! : MF(B) (K, {Kk })

Hom B (f (A), (K, {Kk })) MF(A).
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)).
By (1.2.2.2), we see that, if (K, {Kk }) is a strictly injective B-module,

then f ! (K, {Kk }) is a strictly injective A-module. Moreover, for a strict

monomorphism f (M, {Mk }) (K, {Kk }) of B-modules, the correspond-

ing morphism (M, {Mk }) f ! (K, {Kk }) is a strict monomorphism of
A-modules, which is easily checked. Hence, by the same proof as that of
(1.1.7) (especially, by noting that the functor f ! commutes with the direct
product), we can take f ! (K , {Kk }) as (I , {Ik }), where (K , {Kk }) is a
bounded below complex of strictly injective B-modules.
Let R be a at resolution of f (A) with the trivial ltration. Since the
ltration on R is trivial, it is obvious that the morphism (Qq , {Qqk })B R
(Qq , {Qqk })B f (A) is a ltered quasi-isomorphism (q Z). By (1.2.1) (2),

we have
(1.2.2.3) HomB ((Qq , {Qqk })B f (A), (K , {Kk }))

HomB ((Qq , {Qqk })B R , (K , {Kk })).
(1.2.2.3) is equal to the following:
(1.2.2.4) HomB ((Qq , {Qqk }), Hom B (f (A), (K , {Kk })))

HomB ((Qq , {Qqk }), Hom B (R , (K , {Kk }))).
Here Hom B (f (A), (K , {Kk })) is considered as a ltered B-module, which is

nothing but f f ! (K , {Kk }). It is easy to check that Hom B (Rq , (K q+n , {Kkq+n
})) is a strictly injective B-module; so is Hom nB (R , (K , {Kk })) (n Z).
Therefore Hom B (R , (K , {Kk })) is a strictly injective resolution of f f ! (K ,
{Kk }) by the sheacation of (1.2.1) (2). Hence we can take Hom B (R , (K ,
{Kk })) as (J , {Jk }), and we have a ltered quasi-isomorphism
(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 }))
HomB ((Q , {Qk }), (C , {Ck })) .

+1
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 })),
which enables us to nish the proof of (1.2.2).

Let (E , {Ek }) (resp. (F , {Fk })) be an object of KF(A) (resp. K+ F(A)).
Set
ExtqA ((E , {Ek }), (F , {Fk })) := HomDF(A) ((E , {Ek }), (F , {Fk })[q]).
The following lemma is a ltered version of a classical lemma [44, I (6.4)],

and it will be used in 2.10 below.
Lemma 1.2.3. The following formula holds:
(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]).
Let f : (E , {Ek }) (I , {Ik })[q] be a morphism in DF(A). Then f is

g s
represented by a diagram (E , {Ek }) (G , {Gk }) (I , {Ik })[q] in
KF(A), where s is a ltered quasi-isomorphism. By (1.1.10) (2), s has a
homotopy inverse t : (G , {Gk }) (I , {Ik })[q] in KF(A). We prove that
the morphism f
t g is well-dened. Let u : (G , {Gk }) (G , {Gk })
be a ltered quasi-isomorphism. Let t : (G , {Gk }) (I , {Ik })[q] be a
homotopy inverse of u s. We have to show that t u = t in KF(A). This
equality immediately follows from the following commutative diagram
HomKF(A) ((I , {Ik })[q], (I , {Ik })[q]) HomKF(A) ((I , {Ik })[q], (I , {Ik })[q])

(us) ,=
s ,=
u ,=
HomKF(A) ((G , {Gk }), (I , {Ik })[q]) HomKF(A) ((G , {Gk }), (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]).
Therefore we have an isomorphism

HomDF(A) ((E , {Ek }), (I , {Ik })[q]) HomKF(A) ((E , {Ek }), (I , {Ik })[q]).
Consequently we have
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 ).
(The third equality above follows from (1.2.0.1).)

(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 .
(The intersection above is considered in EA F.)

(2) Let (E, {Ek }) and (F, {Fk }) be ltered A-modules. Assume that (F,
{Fk }) Q (A) ((1.1.13.1)). Then the natural morphism

grl EA grm F grk (E A F )
l+m=k
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,
where the three vertical morphisms are monomorphisms.

(2): For two A-modules L and M and for two A-submodules L and M
of L and M , respectively, it is an elementary exercise to check that
Im((L A M LA M ) LA M ) = Ker(LA M (L/L )A (M/M )).
Hence we have a natural morphism

grl EA grm F grk (EA F ),
l+m=k
which we denote by f . The morphism f is an epimorphism by the denition

of the tensor product of two ltered A-modules.

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
(EA F )k /((EA F )k1 + (EA Fm0 1 ) (EA F )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
s (EA F )k1 + (EA Fm0 1 ) (EA F )k El0 1 A F + EA Fm0 1 .
Take two sections EA Fm0 1 such that s =

t El0 1 A F , u . Then
t+u

t = s u
EA Fm0 . Hence 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
(1.2.4.2) (EA Fm0 1 ) (El0 A Fm0 ) = El0 A Fm0 1

u
again by (1). By (1.2.4.1) and (1.2.4.2) we have
s = El0 1 A Fm0 + El0 A Fm0 1 .

t+u
Hence we have sl0 = 0 and this contradicts the denition of l0 . Therefore f

is a monomorphism.

As in the non-ltered case, we need a ltered derived tensor product in

the preweight-ltered K unneth formula in 2.10.
Let (E , {Ek }) and (F , {Fk }) be two objects of CF(A). Set

(1.2.4.3) (E A F )nk := Im( Elp A Fmq
E p A F q ).
p+q=n l+m=k p+q=n
Then we have a ltered complex (E , {Ek }) A (F , {Fk }) := ((E A

F ) , {(E A F )k }) of A-modules; the boundary morphism is dened by
the following formula
(1.2.4.4) d|E p A F q = (dpE 1) + (1)p (1 dqF ).
It is a routine work to check that the functor
: CF(A) CF(A) CF(A)

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).
Lemma 1.2.5. Let E := (E , {Ek }) be a ltered complex of A-modules and

let F := (F , {Fk }) be an object of K F(Q (A)). Assume that either
(a) E is strictly exact,
or
(b) F is strictly exact,
and assume also that either
(c) E is bounded above,
or
(d) F is bounded below.
Then E A F is strictly exact.
Proof. By [44, II (4.1)], E A F is exact. We have only to prove that

(E A F )k is exact for all k Z.
Let G be a ltered
double complex dened by Gpq := E p A F q with
pq p
ltration Gk := El A Fm q
in Gpq . Then we have the following two
l+m=k
spectral sequences
E2pq = HII
p
HIq (G
k ) = H
p+q
((E A F )k ).
E2pq = HIp HII

q
(G
k ) = H
p+q
((E A F )k ).
The assumptions (c), (d) imply that the two spectral sequences above are
bounded and regular.
p
First, assume that (a) holds. Set E := kZ Ekp and F q
:= kZ Fkq
q
(p, q Z). Then F is a at A-module since Fk (k Z) is so. Because E
q

is exact by (a), E A F is exact. We prove that (E A F )k is exact; we
q q
have only to prove that
(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 ).
Since the lower horizontal sequence of the following commutative diagram
(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
is exact, we may assume that E = E

, F q = F
q
. Set Kp := Ker(Ep
E ) ( = k Z or nothing). Then the sequence 0 Kp Ep
p+1
Kp+1 0 is exact. Moreover, we have the following commutative diagram

with lower exact row:
(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
for k implies that s (K p A F q )k 1 . Repeating this process k k times,

we see that s (K p A F q )k ; thus we have shown the claim.
Now let us prove the exactness of (1.2.5.4).
By (1.2.4) (2), we have only to prove that the following sequence

(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
is exact. Because grm F is a at A-module, (1.2.5.5) is exact by the assump-

q
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
( = k Z or nothing). Then Lq is a at A-module since F is bounded

above. As in the case (a), we have only to prove that the following sequence

(1.2.5.6) 0 grl E p A grm Lq grl E p A grm F q
l+m=k l+m=k

grl E p A grm Lq+1 0
l+m=k
is exact. Because grm L q+1

is a at A-module, (1.2.5.6) is exact.
We nish the proof.

1.3 Filtered Derived Category. III 35
As in [44, II 4], we have a ltered derived functor ([8])

(1.2.5.7) L
A : D F(A) D F(A) D F(A)
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).
1.3 Filtered Derived Category. III
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)]):
If -acyc (A) := {(J, {Jl }) | J and Jl are f -acyclic}
Then the following holds by (1.1.7):
Proposition 1.3.1. The canonical morphism K+ F(If -acyc (A)) D+ F(A)

induces an equivalence

K+ F(If -acyc (A))(FQis) D+ F(A)
of categories and the right derived functor Rf is calculated by the following

formula Rf [(I , {Il })] = [f (I , {Il })] ((I , {Il }) K+ F(If -acyc (A))).
We can consider the dual notion of the above: consider the following full
subcategory Qf -acyc (A) of MF(A ):
Qf -acyc (A) := {(Q, {Ql }) | Q and Q/Ql are f -acyclic}
Then the following holds by (1.1.18):
Proposition 1.3.2. The canonical morphism K F(Qf -acyc (A )) D F

(A ) induces an equivalence

K F(Qf -acyc (A ))(FQis) D F(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 ))).
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),
which we call the gr-functor.

Lemma 1.3.3. If f : (E , {Ek }) (F , {Fk }) is a ltered quasi-isomorph-
ism in KF(A), then grk f : grk E grk F is a quasi-isomorphism.
Proof. (1.3.3) follows from the ve-lemma.

By (1.3.3), the gr-functor (1.3.2.1) induces a functor
(1.3.3.1) grk : D F(A) D (A).
We also call this functor the gr-functor.

Lemma 1.3.4. For a morphism f : (T , A) (T , A ) of ringed topoi, the
following diagrams are commutative:
gr
D+ F(A)
k
D+ (A)

Rf
Rf
(1.3.4.1)
gr
D+ F(A )
k
D+ (A ),
gr
D F(A)
k
D (A)

Lf
Lf
(1.3.4.2)
gr
D F(A )
k
D (A ).
Proof. Let (E , {Ek }) be an object of K+ F(A) and let (I , {Ik }) be a ltered

asque resolution of (E , {Ek }). Then grk I = Ik /Ik1

is a asque resolution
q
of grk E by (1.3.3); hence Rf (grk E ) = f (grk I ). Since R1 f (Ik1 ) = 0
(q Z), the following sequence
q
0 f (Ik1 ) f (Ikq ) f (grk I q ) 0 (q Z)
is exact, and f (Ik )/f (Ik1

) = f (grk I ). Hence grk Rf (E , {Ek }) =

f (grk I ) = Rf (grk E ).
The commutativity of (1.3.4.2) follows from the same proof as that of the
commutativity of (1.3.4.1).

Lastly we give the denition of the functor forgetting the ltration and
the functor taking the ltration k .
1.4 Some Remarks on Filtered Derived Categories 37
Let be an integer k or nothing. A morphism
(1.3.4.3)
: K F(A) (E , {El })
E K (A) ( = +, , b, nothing)
induces a morphism
(1.3.4.4) : D F(A) D (A) ( = +, , b, nothing).
It is easy to check the following diagrams are commutative:

D+ F(A) D+ (A)

Rf
Rf
(1.3.4.5)

D+ F(A ) D+ (A ),

D F(A) D (A)

Lf
Lf
(1.3.4.6)

D F(A ) D (A ).
1.4 Some Remarks on Filtered Derived Categories
In this section we prove some properties of ltered complexes of modules in a

ringed topos with an emphasis on the relation with the formulation of derived
categories in [78].
Let the notations be as in 1.1. Let I be an additive full subcategory
of MF(A). First recall the conditions in our situation which are dual to
[78, Denition 1.3.2 (a), (b), (c)].
(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
0 (I, {Ik }) (J, {Jk }) (K, {Kk }) 0
in MF(A), if (I, {Ik }) I and (J, {Jk }) I, then (K, {Kk }) I.

(1.4.0.3): If
0 (I, {Ik }) (J, {Jk }) (K, {Kk }) 0

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
0 f (I, {Ik }) f (J, {Jk }) f (K, {Kk }) 0
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
K+ F(I)/N+ F(I) D+ F(A)
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
0 (Q, {Qk }) (R, {Rk }) (S, {Sk }) 0
in MF(A), if (R, {Rk }) Q and (S, {Sk }) Q, then (Q, {Qk }) Q.

(1.4.3.3): If a sequence
0 (Q, {Qk }) (R, {Rk }) (S, {Sk }) 0

1.4 Some Remarks on Filtered Derived Categories 39
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
0 f (Q, {Qk }) f (R, {Rk }) f (S, {Sk }) 0
is strictly exact.
Proposition 1.4.4. The category Q (A) satises the conditions (1.4.3.1),

(1.4.3.2) and (1.4.3.3).
Proof. By (1.1.16), we obtain (1.4.4) immediately.

It is easy to check that Qst (A) satises the condition (1.4.3.3), and
Qst (A) satises (1.4.3.1) by (1.1.16); however it is not so easy to show that
Qst (A) satises (1.4.3.2).
Theorem 1.4.5. The category Qst (A) satises the condition (1.4.3.2).
Proof. Consider the strictly exact sequence in (1.4.3.2). Assume that (R,
{Rk }), (S, {Sk }) Qst (A). Obviously (Q, {Qk }) satises the condition

in (1.1.13) (1). We check the condition in (1.1.13) (2). Let (E, {Ek })
(F, {Fk }) be a strict monomorphism. Then we have a commutative diagram
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
0 (Q, {Qk }) (R, {Rk }) (S, {Sk }) 0
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
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 ),
is the zero, which is equal to a composite morphism
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
grk (EA Q) grk (EA R)
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
by (1.2.4) (2) and (1.4.4). Consider the following diagram:
grk (EA Q) grk (EA R)

f Q ,
f ,
R

l+m=k grl EA grm Q l+m=k grl EA grm R.
Because grm S is at, the lower horizontal morphism above is a monomor-

phism. Hence the upper horizontal morphism is a monomorphism. Now we
have proved (1.4.6).

Let N F(Q) be a full subcategory of K F(Q) which consists of the strict

exact sequences of K F(Q). By (1.4.4) and (1.4.5), we have the following:
Corollary 1.4.7. Let Q be Q (A) or Qst (A). Then the canonical functor
K F(Q)/N F(Q) D F(A)
is an equivalence of categories.
1.5 The Topos Associated to a Diagram of Topoi. I 41
1.5 The Topos Associated to a Diagram of Topoi. I
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
F : I o (2-category of ringed topoi).
The functor F is called a diagram of ringed topoi. To give F is equivalent to

giving the I-ringed topos in the sense of [42, Vbis (1.2.1)]. Set (Ti , Ai ) := F(i)
(i I) and : (Ti , Ai ) (Ti , Ai ) be the corresponding morphism to
: i i. Then we have a ringed topos (T , A )I ; an object of (T , A )I
is a system ((Ei )iI , ( )Mor(I) ), where Ei is an Ai -module, : i i is
a morphism in I and : 1 (Ei ) Ei is a morphism of abelian sheaves
which is compatible with the morphism 1 (Ai ) Ai such that idi = id
and = 1 ( ) for a morphism : i i . We call the ringed topos
(T , A )I the ringed topos associated to F. If no confusion seems likely to
occur, we denote (T , A )I simply by (T , A ). Let ei : (Ti , Ai ) (T , A )
be the morphism dened in [42, Vbis (1.2.8)] (cf. [3, V 3.4.2]): e1
i is the nat-
ural restriction and ei (E) = (( (E))i I , canonical morphisms)
HomI (i ,i)
for an Ai -module E. Let

(1.5.0.1) I(T ,A ) := { ei (Ii ) | each Ii is a asque Ai -module}
iI
be an additive full subcategory of (T , A ). Then, by [42, Vbis (1.3.10.1)], for

any object F I(T ,A ) , the inverse image e1
i (F ) is a asque Ai -module.
For an additive full subcategory I of the category of A -modules, let K+ (I)
(resp. N+ (I)) be the category of the bounded below (resp. exact) complexes
whose components belong to I. By [42, Vbis (1.3.10)],

(1.5.0.2) K+ (I(T ,A ) )(Qis) = K+ (I(T ,A ) )/N+ (I(T ,A ) ) D+ (A ).
By the proof of [3, V Proposition 3.4.4], if is at for all , that is,

is exact for all , then we can replace the condition each Ii is a asque
Ai -module in (1.5.0.1) by the condition each Ii is an injective Ai -module
without changing the property (1.5.0.2).
Let f : (T , A )I (T , A )I be a morphism of the ringed topoi
associated to diagrams of ringed topoi. Consider the following additive full
subcategory:
I(T ,A ;f ) := {(Ei )iI | Ei : fi -acyclic Ai -module (i I)}.

Then we have

K+ (I(T ,A ;f ) )(Qis) = K+ (I(T ,A ;f ) )/N+ (I(T ,A ;f ) ) D+ (A ).
By using the proof of (1.1.7), it is straightforward to generalize the above

for ltered modules. Let
(1.5.0.3)

I(T
l
,A )
:= { ei (Ii ,{(Ii )k })
iI
| each (Ii , {(Ii )k }) is a ltered asque Ai -module}
be an additive full subcategory of MF(A ). For an additive full subcategory I

of MF(A ), let K+ F(I) (resp. N+ F(I)) be the bounded below (resp. strictly
exact) complexes whose components belong to I. Then we have an equiva-
lence of categories
(1.5.0.4)

K+ F(I(T
l
)
,A ) (FQis)
= K+ F(I(T
l
,A )
)/N+ F(I(T
l
,A )
) D+ F(A )
by the proof of [3, V Proposition 3.4.4] and that of (1.1.7).

By the proof of [3, V Proposition 3.4.4] and by the proof of (1.1.7), if the
morphism : (Ti , Ai ) (Ti , Ai ) of ringed topoi is at for all : i i,
then we can replace the condition each (Ii , {(Ii )k }) is a ltered asque
Ai -module in (1.5.0.3) by the condition each (Ii , {(Ii )k }) is a strictly injec-
tive Ai -module without changing the property (1.5.0.4) because the product
of strictly injective modules is strictly injective.
Let f : (T , A )I (T , A )I be a morphism of the ringed topoi
associated to diagrams of ringed topoi. Let
I(T
l
,A ;f )
:= {(Ei ,{(Ei )k })iI
| (Ei , {(Ei )k }) : ltered fi -acyclic Ai -module (i I)}.
Then we have the following by the proof of (1.1.7):

K+ F(I(T
l
) = K+ F(I(T
,A ;f ) Qis
l
,A ;f )
)/N+ F(I(T
l
,A ;f )
) D+ F(A ).
In particular, the following holds: let (E , {(E )k }) be an object of K+ F(A ).

Assume that (Eiq , {(Eiq )k }) is ltered fi -acyclic for all i I and for all q Z.
Then the following formula holds (cf. [24, (5.2.5)]):
Rf ((E , {(E )k )}) = (fi (Ei , {(Ei )k }))iI .
Let (T , A) and (T , A ) be two ringed topoi. Let g : (T , A) (T , A ) be

a morphism of ringed topoi. Let I0 be a (not necessarily nite) set. We x
a total order on I0 . Let {Ui0 }i0 I0 and {Ui0 }i0 I0 be open coverings of the
nal objects of T and T , respectively. Let I be a category whose objects are
1.5 The Topos Associated to a Diagram of Topoi. I 43
i = (i0 , . . . , ir )s (i0 < i1 < < ir , r Z0 ). Set {i} := {i0 , . . . , ir }. For

two objects i , i I, a morphism from i to i is, by denition, the inclusion

{i } {i}. As in [3, V 3.4.6], we set Ui := Ui0 Uir , Ui := Ui0 Uir ,
Ti := T |Ui and Ti := T |Ui . Let Ai (resp. Ai ) be the pull-back of A (resp. A )
to Ti (resp. Ti ). Let gi : (Ti , Ai ) (Ti , Ai ) (i I) be morphisms of ringed
topoi. Assume that the following two conditions hold:
(1.5.0.5): If there exists a morphism i i in I, then the following diagram
is commutative:
gi
(Ti , Ai ) (Ti , Ai )

g
(Ti , Ai ) i (Ti , Ai ).
(1.5.0.6): The following diagram is commutative for all i0 I0 :
gi
(Ti0 , Ai0 )
0
(Ti0 , Ai0 )

g
(T , A) (T , A ).
Let (T , A ) be the ringed topos associated to a diagram of ringed topoi

obtained from the open covering {Ui0 }i0 I0 . Let : (T , A ) (T , A)
be a morphism of ringed topoi characterized by the following: 1 (E) (E:
A-module) is obtained from restrictions of E to the ringed topos (Ti , Ai ) (i
I) (e.g., [3, V 3.4.6]). We also have an analogous morphism : (T , A )
(T , A ) of ringed topoi. Then we have the following commutative diagram of
ringed topoi by (1.5.0.5) and (1.5.0.6):
g
(T , A ) (T , A )

(1.5.0.7)
g
(T , A) (T , A ).
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 ),
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
which is characterized by the following: 1 is the natural restriction. For

each object i I, there also exists a natural morphism of ringed topoi
(1.5.0.11) i : (Ti , Ai ) (Ti , Ai )
which is characterized as above. The morphisms and i induce morphisms

of ltered derived categories:
(1.5.0.12) R : D+ F(A ) D+ F((A )I ),
(1.5.0.13) Ri : D+ F(Ai ) D+ F(Ai ).
The following (ltered) cohomological descent is often used in this book:

Lemma 1.5.1 (cf. [3, V Proposition 3.4.8]).
Let (T , A) be a ringed topos and let (T , A ) be the ringed topos associated
to a diagram of ringed topoi which is obtained from a covering of the nal
object of T . Let : (T , A ) (T , A) be the natural morphism. Then the
following hold:
(1) Let F be an object of K+ (A). Then the natural morphism
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).
Proof. (1): Let I be a double complex of asque A -modules such that

there exists a asque resolution 1 (F ) s(I ), where s means the single
complex, and such that 1 (F i ) I i (i) is also a asque resolution; for
example we can take the Cartan-Eilenberg resolution of 1 (F ) as I . It
suces to prove that the following natural morphism
(1.5.1.1) Hi (F ) Ri 1 (F )
is an isomorphism. One can calculate the right hand side by the following
spectral sequence
E1ij = Hj ( I i ) = Hi+j ( s(I )) = Ri+j 1 (F ).
The usual cohomological descent in [3, V Proposition 3.4.8] tells us that

E1i0 = F i and E1ij = 0 if j > 0. Hence (1.5.1.1) follows.
(2): Let 1 (F , {Fk }) (I , {Ik }) be a ltered quasi-isomorphism into
a complex of ltered asque A -modules. Since I (resp. Ik ) is a asque
resolution of 1 (F ) (resp. 1 (Fk )), we see that the morphism
(F , {Fk }) ( (I ), { (Ik )}) = (I , {Ik }) = R 1 (F , {Fk })
is a ltered quasi-isomorphism by (1).

1.6 The Topos Associated to a Diagram of Topoi. II
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
crys , OYi /S )}iI . Let

sociated to the diagram of the ringed topoi {((Yi /S)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 ).
If we are given another open covering {Yj0 }j0 J0 of Y , we have a ringed

topos ((Y /S)crys , OY /S ) and we have two morphisms of ltered derived
log
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 ))
of ringed topoi. We have three morphisms of ltered derived categories:
(1.6.0.5) Rzar : D+ F(f1 (OS )) D+ F(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 ))
be the projection. We also have natural projections

(1.6.0.9) uY /S : ((Y 1
/S)crys , OY /S ) (Y zar , f (OS ))
log

(1.6.0.10) uY /S : ((Y 1
/S)crys , OY /S ) (Y zar , f (OS )).
log
We have the following commutative diagrams:

log
Rcrys
D+ F(OY /S ) D+ F(OY /S )

(1.6.0.11)
RuY /S
Ru
Y /S
R
D+ F(f1 (OS )) zar
D+ F(f 1 (OS ))
and
log
Rcrys

(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

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 )).
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
(1.6.0.15) (V, DV (V), [ ])s,

where V is an open log subscheme of Y , : V V is a locally closed
immersion into a log smooth scheme over S and DV (V) is the log PD-envelope
of over (S, I, ) ([54, (5.3)]); the topology of (Y /S)log
Rcrys is the induced
log
topology by that of (Y /S) . Let (Y /S) log
be the topos associated to
crys Rcrys
(Y /S)log
Rcrys . Let
(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 topoi and a morphism

log log , Q (O log
(1.6.0.19) jRcrys : ((U/S)Rcrys U/S U/S )) ((Y /S)Rcrys , QY /S (OY /S ))
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
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
Let (V, DV (V), [ ]) be an object (1.6.0.15), which is considered as a repre-

sentable sheaf in (Y /S)log log 1
Rcrys . By the denition of jRcrys ,
log 1
(1.6.0.22) jRcrys ((V, DV (V), [ ])) = (V U, DV (V)|U , [ ]|U ).
Here, because DV (V)|U = DV U (V \ (U \ V )), we can consider (V

U, DV (V)|U , [ ]|U ) as an object of (U/S)log
Rcrys . By using this equality, we have
the following equality
(1.6.0.23) QY /S jcrys
log log
= jRcrys QU/S .
log , we have the following equalities:

Indeed, for a sheaf E in (U/S)crys
(QY /S jcrys
log
(E))(V,DV (V),[ ])
log
= (jcrys (E))(V,DV (V),[ ])
= E(V U,DV (V)|U ,[ ]|U )
= log
(jRcrys QU/S (E))(V,DV (V),[ ]) .
Lemma 1.6.1 (cf. [3, IV Proposition 2.3.2]). Let E be a sheaf in (Y

/S)log
crys .

Let U be an open log subscheme of Y . Let U P be an immersion into a log
smooth scheme over S. Let DU (P 2 ) be the log PD-envelope for the diagonal

immersion U P S P over (S, I, ). Then the following sequence
(1.6.1.1) uY /S (E)|U EDU (P) EDU (P 2 )
is exact.
Proof. The proof is the same as that in the proof of [3, IV Proposition 2.3.2].

As in [3, IV (2.3.2)], set
(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 )).
Corollary 1.6.2. For a sheaf E in (Y

/S)log
crys ,
(1.6.2.1) uY /S (E) = uY /S (QY /S (E)).
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].

Corollary 1.6.3. The morphism (1.6.0.17) denes a well-dened functor
QY /S : D+ F(OY /S ) D+ F(QY /S (OY /S ))
tting into the following commutative diagram

Q
D+ F(QY /S (OY /S ))
Y /S
D+ F(OY /S )

(1.6.3.1)
RuY /S
Ru
Y /S
D+ F(f 1 (OS )) D+ F(f 1 (OS )).
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 )).
Hence we have a morphism of ltered derived categories:
(1.6.3.2) log
RRcrys : D+ F(QY /S (OY /S )) D+ F(QY /S (OY /S )).
Furthermore, we have morphisms

log
(1.6.3.3) RRcrys : D+ F(QY /S (OY /S )) D+ F((QY /S (OY /S ))I ),
(1.6.3.4) log
Ri,Rcrys : D+ F(QYi /S (OYi /S )) D+ F(QYi /S (OYi /S )).
By the commutative diagram (1.6.0.21), we have the following commutative

diagram:
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 ).
We also have the following commutative diagrams

log
RRcrys
D+ F(QY /S (OY /S )) D+ F((QY /S (OY /S )I ))

RQY /S
RQ
(1.6.3.6) Y /S
log
Rcrys
D+ F(OY /S ) D+ F((OY /S )I )
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 ).
We also have the following equality of functors, whose proof is slightly

complicated:
Proposition 1.6.4. There exists the following equality of functors
QY /S Rcrys QY /S .
= log
(1.6.4.1) log
RRcrys
Proof. Note that the morphism (1.6.4.1) factors as
(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
QY /S (F ) = 0 (resp. QY /S (F ) = 0). Then, to prove that the rst morphism in

(1.6.4.2) is an isomorphism, it suces to prove that, for any sheaf F of OY /S -
modules and for any q Z, the sheaf Rq crys log
Cone(F RQY /S QY /S (F ))
is parasitic. Note that we have the spectral sequence
log
E2st = Rs crys Ht (Cone(F RQY /S QY /S (F )))
log
= Rs+t crys Cone(F RQY /S QY /S (F )).
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
QY /S Ht (Cone(F RQY /S QY /S (F )))

= QY /S Ht (Cone(F QY /S (I )))
= Ht (Cone(QY /S (F ) QY /S QY /S (I )))
= Ht (Cone(QY /S (F ) I )) = 0 (t),
that is, Ht (Cone(F RQY /S QY /S (F ))) is parasitic for any t Z. Hence

it suces to prove that, for any parasitic sheaf F of OY /S -modules and for
log
any s Z, Rs crys (F ) is a parasitic sheaf of OY /S -modules. To prove this, it
log
suces to prove the equality (Rq crys (F ))D = 0 for any D := (V, DV (V), [ ])
in (1.6.0.15). Let us put D := (V Y Y , DV (V)|Y , [ ]|Y ) and denote the
natural morphism D D by D . Then we have the following commutative
diagram
(Y
/S)crys Dzar
log

log
(1.6.4.3) crys D
(Y
crys Dzar ,
/S)log
where the horizontal morphisms are the canonical restrictions to Zariski

topoi. Then the horizontal morphisms in (1.6.4.3) are exact and they send in-
jective sheaves to injective sheaves (cf. the proof of [11, 7.21]). Hence we have,
log
for a parasitic sheaf F of OY /S -modules, (Rs crys (F ))D = Rs D (FD ) = 0.
Hence we have nished the proof of the proposition.

As in (1.6.1.2), we have natural morphisms

(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 )).
We have the following commutative diagrams by (1.6.3.5) and (1.6.0.11), and

by (1.6.3.6) and (1.6.0.12), respectively:
log
RRcrys

(1.6.4.6)
RuY /S
Ru
Y /S
R
D+ F(f1 (OS )) zar
D+ F(f 1 (OS ))
and
log
Rcrys

(1.6.4.7)
RuY /S
Ru
Y /S
1 Rzar
D+ F(f (OS )) D+ F(f1 (OS )).
Set Z := Y or Y and g := f or f , respectively. Then, by (1.6.2),
uZ/S (E) = uZ/S (QZ/S (E))
log . As in (1.6.3), we have the following commutative

for a sheaf E in (Z/S)crys
diagram
Q
D+ F(QZ/S (OZ/S ))
Z/S
D+ F(OZ/S )

(1.6.4.8)
RuZ/S
Ru
Z/S
D+ F(g 1 (OS )) D+ F(g 1 (OS )).
Chapter 2
Weight Filtrations on Log Crystalline
Cohomologies
In this chapter, we construct a theory of weights of the log crystalline coho-

mologies of families of open smooth varieties in characteristic p > 0, by
constructing four ltered complexes. We prove fundamental properties of
these ltered complexes. Especially we prove the p-adic purity, the functori-
ality of three ltered complexes, the convergence of the weight ltration, the
weight-ltered K unneth formula, the weight-ltered Poincare duality and the
E2 -degeneration of p-adic weight spectral sequences. We also prove that our
weight ltration on log crystalline cohomology coincides with the one dened
by Mokrane in the case where the base scheme is the spectrum of a perfect
eld of characteristic p > 0.
2.1 Exact Closed Immersions, SNCDs and Admissible

Immersions
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.
(1) Let the notations be as in 1.6. Consider triples
(2.1.0.1) (V, DV (V), [ ])s,

where V is an open log subscheme of Y , : V V is an exact immersion
into a log smooth scheme over S and DV (V) is the log PD-envelope of over
(S, I, ). Let (Y /S)log log
ERcrys be a full subcategory of (Y /S)crys whose objects
log
are the triples (2.1.0.1). We dene the topology of (Y /S)ERcrys as the induced
topology by that of (Y /S)logcrys .

56 2 Weight Filtrations on Log Crystalline Cohomologies
Denition 2.1.1. We call the site (Y /S)log log

ERcrys (resp. the topos (Y /S)ERcrys )
the exact restricted log crystalline site (resp. exact restricted log crystalline
topos) of Y /(S, I, ).
Let
log log
(2.1.1.1) Y /S : (Y /S)ERcrys (Y /S)Rcrys
QER
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.
Proposition 2.1.2. The morphism (2.1.1.1) (resp. (2.1.1.2)) gives an equiv-

alence of topoi (resp. ringed topoi).
=
Proof. One can check easily the isomorphism F QER ER
Y /S QY /S F for any
F (Y/S)log
ERcrys .
On the other hand, let D := (V, DV (V), [ ]) be an object of (Y /S)log
Rcrys . By

[54, (5.6)], DV (V) is constructed locally by a local exactication V V ex

of V V. Hence there exists a covering D = i Di such that each Di is an
object in (Y /S)log log
ERcrys . Note that Di D Di is also an object in (Y /S)ERcrys .
Then, for any F (Y /S)log
, we have
Rcrys

F (D) = Ker( F (Di ) F (Di D Di ))
i i,i

Y /S QY /S F (Di )
= Ker( QER Y /S QY /S F (Di D Di ))
ER
QER ER
i i,i
= QER ER
Y /S QY /S F (D).
Hence we have F = QER ER

Y /S QY /S F . Thus the equivalences follow.

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.
Lemma 2.1.3 (Second fundamental exact sequence).

Let : Z Y be an exact closed immersion of ne log schemes over a
ne log scheme S dened by a coherent ideal J of OY . Then the following
sequence
2.1 Exact Closed Immersions, SNCDs and Admissible Immersions 57
J /J 2 (1Y /S ) 1Z/S 0

(2.1.3.1)
is exact. Here is the composite morphism : J /J 2 (1Y /S ) (1Y /S ).

If Z/S is log smooth, then is injective. If Z/S is log smooth and if Y is
ane, then (2.1.3.1) is split.
Proof. Let MY and MZ be the log structures of Y and Z with structural

morphisms Y : MY OY and Z : MZ OZ , respectively. Let MS be
the log structure of S. Because the natural morphisms (1 ) 1
Y /S Z/S
and 1 (MY /OY ) MZ /OZ
are surjective, so is (1Y /S ) 1Z/S . To
prove the exactness of the middle term of (2.1.3.1), it suces to prove that
the following sequence
(2.1.3.2) HomOZ (J /J 2 , E) HomOZ ( (1Y /S ), E) HomOZ (1Z/S , E)
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 ) ,
we may assume that a local section of MZ is represented by (u, m) (u

(OY /J ) , m 1 (MY )). Let g : 1Z/S E be a morphism dened by
g () = g(
) ( 1 ) and g ([(u, m)]) = g(d log u) + g(d log m) ([(u, m)]
Z/S
denotes any lift of to (1
MZ ), where ). It is straightforward to check
Y /S
that g is well-dened and that g induces g. Thus (2.1.3.2) is exact.

Next assume that Z/S is log smooth and that Y is ane. Let Y 1 be the

rst log innitesimal neighborhood of the exact closed immersion Z Y .
1 1
For two sections m (MY ) and a (OY ), let [m] and [a] be the images

in MZ and OZ , respectively. Because Z/S is log smooth and Y is ane, there

exists a section s : Y 1 Z of the exact closed immersion Z Y 1 induced
1
by . In particular, there exist morphisms smo : s (MZ ) MY |Y 1 and
sri : s1 (OZ ) OY 1 such that smo ([m]) = m(1 + t) (t J /J 2 , 1 + t
OY 1 ) and sri ([a]) = a + t (t J /J 2 ); moreover, smo and sri t into the
following commutative diagram:
s
s1 (MZ ) MY |Y 1
mo

|
s1 (Z ) Y Y1
s
s1 (OZ ) ri OY 1 ,
where the vertical morphisms above are structural morphisms.

To prove the existence of the local splitting of (2.1.3.1), we need the module
of the log derivations, e.g., in [53, (5.1)].
Let F be an OY -module. Let f : Y S be the structural morphism.

Let DerS (Y, F) be a (S, OS )-module whose elements are the pairs (, )s
satisfying the following conditions:

(1) is a derivation OY F over S,
(2) is a morphism MY F of monoids,

(3) Y (m)(m) = (Y (m)) (m MY ),
(4) (f 1 (n)) = 0 (n MS ).
Then, by [53, (5.3)], we have an isomorphism
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 ).
Let be the isomorphism (1+J )/(1+J 2 ) 1+t

t J /J 2 of abelian

sheaves. It is easy to check that the morphisms : OY a
a sri ([a])
J /J 2 and : MY m
(m/smo ([m])) J /J 2 satisfy (1) (4) and
give a local splitting of (2.1.3.1).

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 .
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)
Let s be the image of z in S. Since Coker(gp ) is torsion free, there ex-

gp
ists a homomorphism P gp MY,(z) which is compatible with the monoid
homomorphisms Q MS,s MY,(z) and P MZ,z . Since we have

(MZ /OZ )z = (MY /OY )(z) , the homomorphism P gp MY,(z) gp
induces the
homomorphism P MY,(z) , which induces a chart of Y S on a neigh-
borhood of (z). By the exact sequence (2.1.4.2), there exist local sections
xr+1 , . . . , xr+c J and elements m1 , . . . , mr P such that {d log mi }ri=1 is
a basis of 1Z/S,z and {{d log mi }ri=1 , {dxj }d+cj=r+1 } is a basis of Y /S,(z) . By
1
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
P MY around (z). Let J0 (resp. J ) be the ideal sheaf of 0 (resp. ).

(0) (0)
Let {{d log mi }ri=1 , {dxj }r+c
j=r+1 } (mi P, xj J0 ) be a basis of 1Y0 /S0 .
(0)
Let xj be any lift of xj in J . Then, using [13, Corollaire to II 3 Propo-
sition 6], we see that {{d log mi }ri=1 , {dxj }r+c
j=r+1 } is a basis of Y /S,(z)
1
(cf.[40, 4 (17.12.2)]). Hence we have a strict etale morphism Y (S Z[Q]

Z[P ], P a ) S AcS . Now we obtain the diagram (2.1.5.1).

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).
(2) Let Y be a scheme over a scheme T . Let Div(Y /T )0 be the integral

monoid of eective Cartier divisors on Y over T (e.g., [56, (1.1.1)]). We say
that a family {E } of non-zero elements in Div(Y /T )0 has a locally nite
intersection if, for any point z Y , there exists a Zariski open neighborhood
V of z such that V := { | E |V = 0} is a nite set. If {E } has a
locally nite intersection, then we can dene a sum n E (n N) in
Div(Y /T )0 .
Let f : X S0 be a smooth morphism of schemes.
Denition 2.1.7. We call an eective Cartier divisor D on X/S0 is a relative

simple normal crossing divisor (=:relative SNCD) on X/S0 if there exists a
family := {D } of non-zero eective Cartier divisors on X/S0 of locally
nite intersection which are smooth schemes over S0 such that

(2.1.7.1) D= D in Div(X/S0 )0

and, for any point z of D, there exist a Zariski open neighborhood V of z in

X and the following cartesian diagram:
(2.1.7.2)

D|V V

g

SpecS (OS0 [y1 , . . . , yd ]/(y1 ys )) SpecS (OS0 [y1 , . . . , yd ])
0 0
(for some positive integers s and d such that s d), where the morphism g
is etale.
Note that we do not require a relation a priori between {D |V }V and

the family {yi = 0}si=1 of closed subschemes in V in the diagram (2.1.7.2).
However, by (A.0.1) below, we obtain {D |V }V = {{yi = 0}}si=1 in the
diagram (2.1.7.2) if V is small.
Denition 2.1.8. We call a smooth divisor on X/S0 contained in D a smooth

component of D. We call = {D } a decomposition of D by smooth
components of D over S0 .
Note that the decomposition of a relative SNCD by smooth components is

not unique.
Let DivD (X/S0 )0 be a submonoid of Div(X/S0 )0 consisting of eective
Cartier
divisors Es on X/S0 such that there exists an open covering X =
iI Vi (depending on E) of X such that E|Vi is contained in the submonoid
of Div(Vi /S0 )0 generated by D |Vi ( ). By (A.0.1) below, we see that
the denition of DivD (X/S0 )0 is independent of the choice of .
The pair (X, D) gives a natural fs(=ne and saturated) log structure in

Xzar as follows (cf. [54, p. 222223], [29, 2]).
Let M (D) be a presheaf of monoids in X zar dened as follows: for an open
subscheme V of X,
(2.1.8.1) (V, M (D) ) := {(E, a) DivD|V (V /S0 )0 (V, OX )|

a is a generator of (V, OX (E))}
with a monoid structure dened by (E, a) (E , a ) := (E + E , aa ). The nat-

ural morphism M (D) OX dened by the second projection (E, a)
a
induces a morphism M (D) (OX , ) of presheaves of monoids in X zar .
The log structure M (D) is, by denition, the associated log structure to
the sheacation of M (D) . Because DivD|V (V /S0 )0 is independent of the
choice of the decomposition of D|V by smooth components, M (D) is inde-
pendent of the choice of the decomposition of D by smooth components of D.
Proposition 2.1.9. Let the notations be as above. Let z be a point of D

and let V be an open neighborhood
s of z in X which admits the diagram
(2.1.7.2). Assume that z i=1 {yi = 0}. If V is small, then the log struc-

ture M (D)|V OV is isomorphic to OV y1N ysN OV . Consequently
M (D)|V is associated to the homomorphism NsV ei
yi M (D)|V
(1 i s) of sheaves of monoids on V , where {ei }si=1 is the canonical basis
of Ns . In particular, M (D) is fs.
Proof. By the denition of M (D) and by (A.0.1) below, the homomorphism

M (D)|V OV factors through OV y1N ysN if V is small. Hence there
exists a natural morphism M (D)|V OV y1N ysN of log structures on V .
By taking the stalks, one can easily check that the morphism above is an
isomorphism.

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).

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
is an admissible immersion with respect to {D }V .

Denition 2.1.12. We call the admissible immersion V the restriction of
to V , and |V := {D }V the restriction of to V .
By (2.1.5) and (A.0.1) below, we have the following:
be an admissible immersion.
Lemma 2.1.13. Let : (X, D; ) (X , D; )
Then, for any point z of X, there exist Zariski open neighborhoods V of z
and V of (z), positive integers s d d and the following two cartesian
diagrams:
(2.1.13.1)

D|V V

g

SpecS (OS [x1 , . . . , xd ]/(x1 xs )) SpecS (OS [x1 , . . . , xd ]),
(2.1.13.2)

V V

g

SpecS (OS0 [x1 , . . . , xd ]/(xd+1 , . . . , xd )) SpecS (OS [x1 , . . . , xd ]),
0
where g is etale and {D |V }V = {{xi = 0}}si=1 in the diagram (2.1.13.1).

Let (S, I, ) be a PD-scheme and let (X, D) be a smooth scheme with a
relative SNCD over S0 := SpecS (OS /I). Let be a decomposition of D by
smooth components of D. Consider triples
(2.1.13.3) ((U, D|U ), D(U,D|U ) ((U, D)), [ ])s,

where U is an open subscheme of X, (U, D|U ) (U, D) is an admissible
immersion over S with respect to U and D(U,D|U ) ((U, D)) is the log PD-
envelope of the immersion above over (S, I, ). Let ((X, D)/S)logARcrys be a
full subcategory of ((X, D)/S)log
crys whose objects are the triples (2.1.13.3).
log
We dene the topology of ((X, D)/S)ARcrys as the induced topology by
that of ((X, D)/S)log . Let ((X, D)/S)log be the topos associated to
crys ARcrys
((X, D)/S)log
ARcrys .
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
be a natural morphism of topoi: For an object E ((X, D)/S)log

Rcrys ,
QAR
(X,D)/S (E) is the natural restriction of E and QAR
(X,D)/S commutes with
inverse limits. We also have a morphism
(2.1.14.2) QAR log AR

(X,D)/S : (((X, D)/S)ARcrys , Q(X,D)/S Q(X,D)/S (O(X,D)/S ))

(((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
in (2.1.5) for . Then is locally an admissible immersion with respect to the

restriction of to an open subscheme of X since P is a local lift of (X, D).
Hence we obtain (2.1.15) by (2.1.2) and by the proof of (2.1.2).

2.2 The Log Linearization Functor
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.
(1) Let (S, I, ) and f : Y S be as in 1.6. For an object (V, T, MT , , )

of the log crystalline site (Y /S)logcrys , we sometimes denote it simply by
(V, T, MT , ), (V, T, ) or even T as usual. We also denote by T the rep-
resentable sheaf in (Y /S)log log
crys dened by T . Let F be an object of (Y /S)crys .
Let (Y/S)log |F be the localization of the topos (Y
crys /S)log at F : the objects
crys
in (Y log
crys |F are the pairs (E, )s, where E is an object in (Y /S)crys and
/S)log
is a morphism E F in (Y /S)log . As usual, let
crys
(2.2.0.1) jF : (Y log

crys |F (Y /S)crys
/S)log

be a morphism of topoi dened by the following: for an object E in (Y /S)log
crys ,
j (E) is a pair (E F, E F F ); for an object (E, ) in (Y

proj.
F /S)log |F , crys
jF ((E, )) is the sheaf of the sections of .
Let (V, T, MT , ) be an object of the log crystalline site (Y /S)log
crys . Let
jT : (Y log
crys |T (Y /S)crys
/S)log
be the localization morphism in (2.2.0.1) for F = T . Let
(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 )
and the following commutative diagram of topoi
(Y log
jT
crys |T (Y /S)crys
/S)log

(2.2.0.4)

uY /S

Tzar = Vzar Yzar ,
where is dened as follows: ((T , ), 1 (E)) := (T , 1 (E)) for E Tzar

and (T , ) (Y crys |T .
/S)log
By the log version of [11, 5.27 Corollary], we have the following:
Proposition 2.2.1. Let the notations be as above. Assume that V = Y . Then
the following hold:
(1) The functors jT is exact.
(2) For an abelian sheaf E in (Y
/S)log
crys , jT (E) is uY /S -acyclic.
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
be the morphism (2.2.0.2) for T = DY (Y). For an ODY (Y) -module E, we

dene L(E) as follows:
(2.2.1.2) L(E) := jDY (Y) (E) (Y

/S)log
crys .
As in the classical crystalline case, L denes a functor:
(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.
As in the classical crystalline case, another denition of the log lineariza-

tion functor is known. To state it, we need to recall the denition of a log
HPD stratication (cf. [11, 4.3H Denition]; however there is a mistype in
[loc.cit., 1)]: DX/S -linear should be replaced by DX/S (1)-linear).

Let DY (Y 2 ) be the log PD-envelope of the locally closed immersion Y
Y S Y over (S, I, ). Let J be the PD-ideal sheaf dening the exact locally

closed immersion Y DY (Y 2 ).
Denition 2.2.2. Let E and F be two ODY (Y) -modules.

(1) An isomorphism : ODY (Y 2 ) ODY (Y) E E ODY (Y) ODY (Y 2 ) is called
a log HPD stratication if is ODY (Y 2 ) -linear, if mod J is the identity and
if the cocycle condition holds.
(2) ([75, (1.1.3)]) An ODY (Y) -linear morphism u : ODY (Y 2 ) ODY (Y)
E F is called a log HPD dierential operator.
(3) ([75, (1.1.3)]) For a positive integer n, an ODY (Y) -linear morphism
u : (ODY (Y 2 ) /J [n+1] ) ODY (Y) E F
is called a log PD dierential operator of order n.
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 ) .
Hence L (E) denes a crystal of OY /S -modules (cf. [54, (6.7)]), which we

denote by the same symbol L (E). L denes a functor
{the category of ODY (Y) -modules and log HPD dierential operators}
{the category of crystals of OY /S -modules} :

For a log HPD dierential operator u : ODY (Y 2 ) ODY (Y) E F, L (u) :
L (E) L (F) is given by the composite
id
(2.2.2.1) ODY (Y 2 ) ODY (Y) E ODY (Y 2 ) ODY (Y) ODY (Y 2 ) ODY (Y) E
idu
ODY (Y 2 ) ODY (Y) F,
where : ODY (Y 2 ) ODY (Y 2 ) ODY (Y) ODY (Y 2 ) = ODY (Y 3 ) is the map

induced by the projection Y 3 Y 2 to the rst and the third factors. By
the log version of [11, 6.10 Proposition], the following holds:
Proposition 2.2.3. For an ODY (Y) -module E, there exists a canonical iso-
morphism

L (E) L(E).
Hence L also denes the functor
{the category of ODY (Y) -modules and log HPD dierential operators}
{the category of crystals of OY /S -modules}.

By (2.2.2.1) and (2.2.3), we see the following: For a log HPD dierential
operator u : ODY (Y 2 ) ODY (Y) E F and (U, T, ) (Y /S)log
crys , L(u)(U,T,) :
L(E)(U,T,) L(F)(U,T,) is given by the composite
(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,
where T : ODY (T S Y) ODY (T S Y) ODY (Y) ODY (Y 2 ) = ODY (T S Y 2 )

(the equality follows from the log version of [11, 6.3, proof of 6.10]) is the
map induced by the projection T S Y 2 T S Y to the rst and the third
factors. It is easy to obtain the following lemma from the denition of L .
Lemma 2.2.4. The functor L, regarded as the functor
{the category of ODY (Y) -modules and ODY (Y) -linear morphisms}
{the category of crystals of OY /S -modules}

is exact.
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)
in OY /S -modules is not exact since the value of the sequence above at Y is
0 0
0 pWn /p2 Wn .
The following is analogous to [11, 6.2 Proposition].
Lemma 2.2.6. (1) Let Yi (i = 1, 2) and (S, I, ) be as in 1.6. Let Ti =

(Ui , Ti , i ) = (Ui , Ti , MTi , i ) (i = 1, 2) be an object of the log crystalline site
(Yi /S)log log
crys , which is considered as a representable sheaf in the topos (Yi /S)crys .

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 .

Assume that g induces a surjective morphism g : g (J2 ) J1 . Let
: (Y log
1 /S)crys |T1 (Y2 /S)crys |T2
log
be the induced morphism of topoi. Let (U, T, , ) = (U, T, MT , , ) be a

representable object in (Y
2 /S)crys |T2 . Let J be the dening ideal sheaf of
log

the closed immersion U T . Set J := J + IOT and let be the ex-
tension of and on J . Let D (T T2 T1 ) be the log PD-envelope of the

closed immersion U U2 U1 T T2 T1 over (T, J , ) with natural mor-
phism q : (U U2 U1 , D (T T2 T1 ), [ ]) (U1 , T1 , 1 ) in (Y1 /S)log
crys . Then
((U, T, , )) is representable by an object (U U2 U1 , D (T T2 T1 ), [ ], q)
crys |T1 ; the functor is exact.
(Y1 /S)log
(2) Let the notations and the assumptions be as in (1). Then D (T T2
T1 ) = T T2 T1 .
Proof. (1): We have to check that (U U2 U1 , D (T T2 T1 ), [ ], q) is actually
crys |T1 .
an object of (Y1 /S)log
Since U U2 U1 is an open subscheme of U1 , extends to OU U2 U1 . Since
the image J in OU U2 U1 is IOU U2 U1 , actually extends to OU U2 U1 (cf. [11,
6.2.1 Lemma]). Since extends to OU U2 U1 , the exact closed immersion U U2

U1 D (T T2 T1 ) is a PD closed immersion by [11, 3.20 Remarks 4)].
Set J i := Ji +IOTi (i = 1, 2) and let i be the extension of i and on J i .

Set J12,T := Ker(OT T2 T1 OU U2 U1 ). Let J 12,T be the PD-ideal sheaf

of OD (T T2 T1 ) obtained from J12,T . Set J 12,T := J 12,T + IOD (T T2 T1 ) .
Consider the following commutative diagram
OD (T T2 T1 ) OD (T T2 T1 )

OT T2 T1 OT

OT1 OT2 .
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 )
of log PD-schemes over (S, I, ).

By using the universality of the log PD-envelope, it is straightforward to
see that
(2.2.6.1) ((U, T, , )) = (U U2 U1 , D (T T2 T1 ), [ ], q).
Therefore, for an object E in (Y

2 /S)crys |T2 , we have
log
(2.2.6.2) E((U, T, , )) = Hom(Y

/S)log
( ((U, T, , )), E)
1 crys |T1
= E((U U2 U1 , D (T T2 T1 ), [ ], q)).
Using the formula (2.2.6.2) and noting that D (T T2 T1 ) T T2 T1 is a

closed set of T as a topological space, we can easily see that the functor is
exact.
(2): Set J12 := Ker(OT2 g (OT1 )). The structure sheaf of T T2 T1 is
equal to OT /J12 OT . By the following commutative diagram
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).

Following [31], let us denote by iY /S the sheaf of log dierential forms of

Y /S of degree i (i N). The following is a log version of [11, 6.12 Theorem]:

Proposition 2.2.7. Let : Y Y be a closed immersion of ne log
schemes over S. Assume that Y is log smooth over S and that extends

to Y . Let DY (Y) be the log PD-envelope of over (S, I, ). Then the natural
morphism
(2.2.7.1) OY /S L(ODY (Y) OY Y/S )
is a quasi-isomorphism.
Proof. Let DY (Y i ) (i Z>0 ) be the log PD-envelope of the composite im-

mersion Y Y Y i over S, where Y Y i is the diagonal immersion.
i-th proj.
Let pi : DY (Y 2 ) Y 2 Y (i = 1, 2) be a natural morphism and let
J be the ideal sheaf of the locally exact closed immersion Y DY (Y 2 ).
The problem is local as in [11, 6.12 Theorem]; we may assume that 1Y/S has
a basis {d log tj }nj=1 , where tj is a local section of the log structure of Y. Let

uj be a local section of Ker(OD 2
Y (Y )
OY ) such that p2 (tj ) = p1 (tj )uj .
Then, by [54, (6.5)], the following morphism
[n]
OY s1 , . . . , sn sj
(uj 1)[n] ODY (Y 2 )
is an isomorphism, where sj s are independent indeterminates. We identify

ODY (Y 2 ) with OY s1 , . . . , sn by this isomorphism. By the log version of [11,
6.2 Proposition], log
crys (OY /S ) is a crystal of OY/S -modules. Hence, as in [11,

6.3 Corollary], we obtain a canonical isomorphism ODY (Y) OY ODY (Y 2 )
ODY (Y 2 ) . Consequently we can identify ODY (Y 2 ) with ODY (Y) s1 , . . . , sn
(cf. [54, (6.5)]). Moreover, by [54, (5.8.1)] and [81, Proposition 3.2.5], there

exists an isomorphism 1Y/S d log tj
uj 1 J /J 2 = J /J [2] of
OY -modules.
Let p13 : DY (Y 3 ) DY (Y 2 ) be the induced morphism by the product
of the rst and the third projections Y 3 Y 2 . Let
p
: ODY (Y 2 )
13
ODY (Y 3 ) ODY (Y 2 ) ODY (Y) ODY (Y 2 )
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
: ODY (Y 2 ) OY qY/S ODY (Y 2 ) OY q+1

Y/S
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]
as in [11, 6.11 Lemma]. Let (U, T, ) be an object of (Y /S)log

crys . Because the
problem is local, we may assume that there exists the following commutative
diagram:

U T

Y Y.
Then we have a natural morphism (U, T, ) (Y, DY (Y), [ ]) in (Y /S)log crys
and a natural complex OT s1 , . . . , sn OY Y/S , which is equal to the com-
plex L(ODY (Y) OY Y/S )(U,T,) .
Now, consider the case n = 1 and set s1 = s and t1 = t. Then the
T
complex OT s1 , . . . , sn OY Y/S is equal to OT s OT sd log t. Because
T (s ) = s
[n] [n1] [n]
(s + 1)d log t = (ns + s [n1]
)d log t for a positive integer
n, we have the following formula
(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].

Proposition 2.2.8 ([54, the proof of (6.9)]).

Let : Y Y, Y and DY (Y) be as in (2.2.7). Let E be a crystal of OY /S -
modules. Let (E, ) be the corresponding ODY (Y) -module with integrable con-
nection. Then there exists a natural quasi-isomorphism
(2.2.8.1) E L(E OY Y/S ).
Proof. The proof is the same as that in [11, 6.14 Theorem]: we have the
following equalities in D+ (OY /S ):
E = E OY /S L(ODY (Y) OY Y/S )

= L(E OY Y/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

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 OZ -module E, that is, the natural mor-

phism gY Y,Z (E)PD
Y,Z gZ (E) is an isomorphism.

Proof. Since Y,Z is ane, (2.2.9) immediately follows from the ane base
change theorem ([39, (1.5.2)]).

Y,Z induces a surjection Y,Z (J Y ) J Z .

Lemma 2.2.10. Assume that PD PD
Then the diagram

(D
Z (Z)zar , ODZ (Z) )
D (Z)
log |
Z ((Z/S)crys DZ (Z) , OZ/S |DZ (Z) )

log,loc
(2.2.10.1) Y,Z
PD crys

(D Y ((Y
D (Y)
Y (Y)zar , ODY (Y) ) crys |DY (Y) , OY /S |DY (Y) )
/S)log
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.
Proof. Let E be a quasi-coherent ODZ (Z) -module. Let (T, ) = (U, T, MT , , )

crys |DY (Y) . Then, by (2.2.6) (1) and (2),
be an object of (Y /S)log

(2.2.10.2) crys DZ (Z) (E)(T, ) = (T DY (Y) DZ (Z), p2 (E)),
log,loc
where p2 : T DY (Y) DZ (Z) DZ (Z) is the second projection. On the other

hand,
(2.2.10.3) DY (Y) PD PD
Y,Z (E)(T, ) = (T, Y,Z (E)).
Since DZ (Z) DY (Y) is a closed immersion, in particular, an ane mor-

phism, the ane base change theorem tells us that both right hand sides of
(2.2.10.2) and (2.2.10.3) are the same. This completes the proof of (2.2.10).

Lemma 2.2.11. Assume that PD Y,Z induces a surjection Y,Z (J Y ) J Z .
PD
Then the following diagram of topoi
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
as a sheaf, where p2,Z : D(T ) DZ (Z) DZ (Z) is the second projection.

Analogously, let p2,Y : T DY (Y) DY (Y) be the second projection. Let
Z be the PD-structure of DZ (Z) and let Z be the extension of the and
on Ker(ODZ (Z) OZ ) + IODZ (Z) . Let D(T S DY (Y)) be the double
log PD-envelope of T and DY (Y) (cf. [11, 5.12 Lemma]) over (S, I, ). Let
D() be the PD-structure of D(T S DY (Y)) and D() the extension of D()
and . Then we have
log,loc
crys log,loc
jDY (Y) (T ) = crys (T DY (Y), p2,Y )
log,loc
= crys (D(T S DY (Y)), p2,Y )
= DD() (D(T S DY (Y)) DY (Y) DZ (Z)) (2.2.6)
= D(T ) DZ (Z) (the universality of D(T S DY (Y))).
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
Y /S Y,Z crys LZ/S

LPD PD log PD
(2.2.12.1)

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
(2.2.12.2) LY /S Y,Z log

crys LZ/S
for quasi-coherent OZ -modules. Moreover, the isomorphism (2.2.12.1) is

functorial with respect to log HPD dierential operators of quasi-coherent
ODZ (Z) -modules.
Proof. Because DZ (Z) = Z Y DY (Y) and because the diagram (2.2.8.2) is

cartesian, the natural morphism PD Y,Z : Y,Z (J Y ) J Z is surjective. The
PD
rst statement of (2.2.12) immediately follows from (2.2.1.2), (2.2.10) and

(2.2.11). The second statement follows from the former and (2.2.9).
Let us prove the last statement. For a quasi-coherent ODZ (Z) -module E
and (U, T, ) (Y /S)log
crys , the isomorphism
Y /S Y,Z (E)T crys LZ/S (E)T

LPD PD log PD
induced by (2.2.12.1) is given by the natural homomorphism
(2.2.12.3) ODU (T S Y) ODY (Y) PD

Y,Z (E) ODU Y Z (T S Z) ODZ (Z) E.
If we are given a log HPD dierential operator u : ODZ (Z 2 ) ODZ (Z) E F

of ODZ (Z) -modules, the composite morphism
u
: ODY (Y 2 ) ODY (Y) PD
u Y,Z (E) ODZ (Z 2 ) ODZ (Z) E F
is a log HPD dierential operator of ODY (Y) -modules and we see easily that
the diagram
ODU (T S Y) ODY (Y) PD

Y,Z (E) ODU (T S Y) ODY (Y) Y,Z (F)
PD

ODU Y Z (T S Z) ODZ (Z) E ODU Y Z (T S Z) ODZ (Z) F
is commutative for any T = (U, T, ) (Y /S)log

crys , where the upper horizon-
tal morphism (resp. the lower horizontal morphism) is the homomorphism
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
(2.2.13.1) D{1 ,2 ,...k } := D1 D2 Dk (i = j if i = j)
for a positive integer k, and set

X (k = 0),
(2.2.13.2) D(k) =
D{1 ,2 ,...,k } (k 1)
{1 ,...,k | i =j (i=j)}
for a nonnegative integer k. Set
(2.2.13.3) D := X
for later convenience.

The following proposition says that a decomposition of a relative SNCD
by smooth components is locally unique:
Proposition 2.2.14. Let and be decompositions of D by smooth com-
ponents. Then, for any z X, there exists an open neighborhood V of z in
X such that V = V .
Proof. If V is small enough, we can take the diagram (2.1.7.2) such that
(A.0.1) below holds for both and . Then V = {yi = 0}si=1 = V .

Proposition 2.2.15. D(k) is independent of the choice of the decomposition

of D by smooth components of D.
Proof. Obviously we may assume that k is positive.

First we prove (2.2.15) for the case k = 1. Let D = {D } and
D = {D } be two decompositions of D by smooth components of D. By
(2.2.14) there exists an open covering {Xi }i of X such that D |Xi = D |Xi .

Hence we have an isomorphism ( D ) X Xi ( D ) X Xi . This

local isomorphism is compatible with the open
immersions Xi Xi Xi ;

therefore we have the global isomorphism D D .
Let D[k] be the k-fold ber product of D(1) over X; D[k] admits the action
of the symmetric group Sk of degree k. For a positive integer k, denote
the set {1, 2, . . . , k} by [1, k]. For a surjective map : [1, k] [1, l], we
have the corresponding morphism D[l] D[k] , which we denote by s .
Let Sk be the set of surjective morphisms [1, k] [1, k 1]. Set D{k} :=
D[k] \ Sk s (D[k1] ); D{k} is an open subscheme of D[k] . The scheme
D{k} also admits the action of Sk . Then we can check D(k) = D{k} /Sk by
the construction of D{k} . Consequently D(k) is independent of the choice of
the decomposition of D by smooth components of D.

Set
(2.2.15.1) Z|D(k) := Z X D(k) .
The scheme Z|D(k) is a relative SNCD on D(k) . We use analogous notations

D(k) and Z|D(k) (k N) for DZ with . Let a(k) : (D(k) , Z|D(k) ) (X, Z)
and b : (D , Z|D(k) ) (X , Z) be morphisms induced by natural closed
(k) (k)
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:
(2.2.15.2) PkD iX /S (log(D Z)) =

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, ):

(X, D Z) (X , D Z) DD

(X, Z) (X , Z) D

(D(k) , Z|D(k) ) (D(k) , Z|D(k) ) D(k)
Let gD : DD (X , DZ), g : D (X , Z) and g (k) : D(k) (D(k) , Z|D(k)

) be natural morphisms. Note that the underlying schemes of the log schemes
DD and D are the same. Let c(k) : D(k) D be a morphism induced by
b(k) : (D(k) , Z|D(k) ) (X , Z).
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.
Proof. Apply (2.1.13) to the SNCD D Z and assume that D (resp. Z) is

dened by an equation x1 = = xt = 0 (resp. xt+1 = = xs = 0)
(1 t s).
(1): (1) is obvious.
(2): By the universality of the log PD-envelope, this is a local question. We
may have two cartesian diagrams in (2.1.13) for D Z; we may assume that
k t. Let D1k be a closed subscheme dened by an equation x1 = =
xk = 0. Then OD OX OD1k = OX xd+1 , . . . , xd OX (OX /(x1 , . . . , xk )) =
OX xd+1 , . . . , xd /(x1 , . . . , xk ).
Set D1k := D1k X X. Then the structure sheaf of the PD-envelope of

the closed immersion D1k D1k is
OD1k xd+1 , . . . , xd = OX xd+1 , . . . , xd /(x1 , . . . , xk ).
Furthermore it is immediate to see that there exists a natural isomorphism

D(k) D (X ,Z) (D(k) , Z|D(k) ) as log schemes. Thus (2) follows.
(3): The proof of (3) is evident by the local description of OD and OD(k) .

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)
Furthermore, as usual, we identify the representable sheaf by DD in ((X,

Z)/S)log
crys with DD . Let ((X, D Z)/S)crys |DD be the localization of
log
D
Z)/S)log log (k) ,

((X, D crys at DD . Let ((X, Z)/S)crys |D and ((D Z|D(k) )/S)log
crys
| (k) be obvious analogues. Let acrys : ((D(k) ,

(k)log
D Z| (k) )/S)log ((X,
D Z)/S)
crys
log
crys be a morphism of topoi induced by the morphism a(k) . By the log version
(k)log
of [11, 6.2 Proposition], the functor acrys is exact.
Let the right objects in the following table be the log PD-envelope of the
locally closed immersion of the left ones:

(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
jD(k) : ((D(k) , crys |D(k)

Z|D(k) )/S)log ((D(k) ,
Z|D(k) )/S)log
crys
be localization functors (2.2.0.1) and let

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)
be morphisms of ringed topoi dened in (2.2.1.1) and let

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 .
For an OD(k) -module E, set also
L(k) (E) := jD(k) (k) g (k) (E) ((D(k) ,

Z|D(k) )/S)log
crys .
As usual, we have a complex L(X,DZ)/S (X /S (log(D Z))) of O(X,DZ)/

S -modules. By (2.2.7) we have a natural quasi-isomorphism

(2.2.16.1) O(X,DZ)/S L(X,DZ)/S (X /S (log(D Z))).
Similarly we have two quasi-isomorphisms:

(2.2.16.2) O(X,Z)/S L(X,Z)/S (X /S (log Z)),

(2.2.16.3) O(D(k) ,Z| )/S L(k) (D(k) /S (log Z|D(k) )).
D (k)
Let {PkD X /S (log(D Z))}kZ be the ltration on X /S (log(D Z))

dened in (2.2.15.2). Then PkD X /S (log(D Z)) forms a subcomplex of
X /S (log(D Z)) and the boundary morphisms of PkD X /S (log(D Z))
are log HPD dierential operators of order 1 with respect to (X , Z)/S.
Set
PkD L(X,Z)/S (X /S (log(DZ))) := L(X,Z)/S (PkD X /S (log(DZ))) (k Z).
Lemma 2.2.17. (1) The natural morphism
(2.2.17.1) OD OX PkD X /S (log(D Z)) OD OX X /S (log(D Z))
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
such that the induced morphism OX [xd+1 , . . . , xd ]/J0N OX /J N is an

isomorphism, where J0 := (xd+1 , . . . , xd ). By [11, 3.32 Proposition], OD is
isomorphic to the PD-polynomial algebra OX xd+1 , . . . , xd . Hence we have
the following isomorphisms

OD OX X /S (log(D Z)) s(X /S (log(D Z ))OS
OS xd+1 , . . . , xd OX X /S )
and

OD OX PkD X /S (log(D Z)) s(PkD X /S (log(D Z ))OS
OS xd+1 , . . . , xd OX X /S ).
Since the complex OS xd+1 , . . . , xd OX X /S consists of free OS -

modules, we obtain the desired injectivity.
(2): By (1) and (2.2.4), the natural morphism
(2.2.17.3) PkD L(X,Z)/S (X /S (log(D Z))) L(X,Z)/S (X /S (log(D Z)))
is injective in the category of crystals of O(X,Z)/S -modules. As in [3, IV

Proposition 2.1.3], the functor
{the category of crystals of O(X,Z)/S -modules}
{the category of Q(X,Z)/S (O(X,Z)/S )-modules}

is exact. Hence (2.2.17.2) is injective.

By (2.2.17) (2), a family {Q(X,Z)/S PkD L(X,Z)/S (X /S (log(D Z)))}kZ of

complexes of Q(X,Z)/S (O(X,Z)/S )-modules denes a ltration on the complex
Q(X,Z)/S L(X,Z)/S (X /S (log(D Z))). Hence we obtain an object
(Q(X,Z)/S L(X,Z)/S (X /S (log(D Z))),

{Q(X,Z)/S PkD L(X,Z)/S (X /S (log(D Z)))}kZ )
in C+ F(Q(X,Z)/S (O(X,Z)/S )).

Now we consider the Poincare residue isomorphism with respect to D.
Though a relative divisor in this book is a union of smooth divisors, we
consider the orientation sheaf of it for showing that our theory in this book
is independent of the choice of the numbering of the smooth components of
a relative SNCD.
First, let us recall the orientation sheaf in [23, (3.1.4)].
k E
Let E be a nite set with cardinality k 0. Set E := Z if k 1
and E := Z if k = 0.
Let k be a positive integer. Let P be a point of D(k) . Let D0 , . . . , Dk1

be dierent smooth components of D such that D0 Dk1 contains
P . Then the set E := {D0 , . . . , Dk1 } gives an abelian sheaf

k
0 k1 zar (D/S0 ) := ZE
D Dk1
0
on a local neighborhood of P in D(k) . The sheaf 0 k1 zar (D/S0 ) is glob-

alized on D(k) ; we denote this globalized abelian sheaf by the same symbol
0 k1 zar (D/S0 ). We denote a local section of 0 k1 zar (D/S0 ) by the
following way: m(0 k1 ) (m Z). Set

(k)
zar (D/S0 ) := 0 k1 zar (D/S0 ).
{0 ,...k1 }
(k) (k) (k)

By abuse of notation, we often denote a zar (D/S0 ) simply by zar (D/S0 ).
(0) (k)
Set zar (D/S0 ) := ZX . The sheaves 0 k1 zar (D/S0 ) and zar (D/S0 )
(k)log
are extended to abelian sheaves log
0 k1 crys
(D/S; Z) and crys (D/S; Z),
respectively, in the log crystalline topos ((D(k) , Z| (k) )/S)log since, for
D crys
an object (U, T, MT , , ) ((D(k) ,

Z|D(k) )/S)log
crys , the closed immersion

: U T is a homeomorphism of topological spaces. If Z = , then de-
(k)log
note log
0 k1 crys
(D/S; Z) and crys (D/S; Z) by 0 k1 crys (D/S) and
(k)
crys (D/S), respectively.
Denition 2.2.18. We call

(k) (k) (k)log
zar (D/S0 ) (resp. crys (D/S), crys (D/S; Z))
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.
Lemma 2.2.20. Let E be an OD(k) -module. Then there exists a canonical

isomorphism

(2.2.20.1) L(k) (E Z zar
(k)
(D/S)) L(k) (E) Z crys
(k)log
(D/S; Z).
(k)log
Proof. (2.2.20) immediately follows from the denition of crys (D/S; Z).

Proposition 2.2.21. (1) There exists the following exact sequence:
(2.2.21.1)
D
0 OD OX Pk1 X /S (log(D Z)) OD OX PkD X /S (log(D Z))
OD OX b (D(k) /S (log Z|D(k) ) Z zar

(k) (k)
(D/S){k}) 0.
(2) There exist quasi-isomorphisms
(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}).
Hence (1) follows from (2.2.17) (1).

(2): By the isomorphism (2.2.21.3), (2.2.17) (1) and (2.2.4), we have
Q PD
grk (X,Z)/S Q(X,Z)/S L(X,Z)/S (X /S (log(D Z)))
D
= Q(X,Z)/S L(X,Z)/S (grP
k X /S (log(D Z)))
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}.
By (2.2.12) and (2.2.16) (1), (2), this complex is equal to
Q(X,Z)/S acrys L(k) (D(k) /S (log Z|D(k) ) Z zar

(k)log (k)
(D/S)){k},
which is equal to
Q(X,Z)/S acrys (L(k) (D(k) /S (log Z|D(k) )) Z crys

(k)log (k)log
(D/S; Z)){k}
by (2.2.20).
By (2.2.7) we obtain the second quasi-isomorphism in (2.2.21.2).

For simplicity of notation, set

(Q(X,Z)/S L(X,Z)/S (X /S (log(D Z))), Q(X,Z)/S P D ) :=
(Q(X,Z)/S L(X,Z)/S (X /S (log(D Z))), {Q(X,Z)/S PkD L(X,Z)/S (X /S (log(D Z)))}kZ )

and
(OD OX X /S (log(D Z)), P D ) :=
(OD OX X /S (log(D Z)), {OD OX PkD X /S (log(D Z))}kZ ).

(2.2.22.1)

u(X,Z)/S : (((X, Z)/S)log zar , f 1 (OS ))
(X
Rcrys , Q(X,Z)/S (O(X,Z)/S ))
be the morphism in (1.6.1.2). Then
(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 )
in D+ F(f 1 (OS )).

Proof. By (1.6.3.1), by (2.2.1) (2) and by (1.3.1), the left hand side of
(2.2.22.2) is equal to
(u(X,Z)/S L(X,Z)/S (X /S (log(D Z))), u(X,Z)/S L(X,Z)/S (PkD X /S (log(D Z)))).
For an OX -module F, we have
u(X,Z)/S L(X,Z)/S (F) = u(X,Z)/S jD g (F) = g (F) = OD OX F
by (2.2.0.4). Hence
u(X,Z)/S (L(X,Z)/S (PkD X /S (log(D Z)))) = OD OX PkD X /S (log(D Z)).
Thus (2.2.22) follows.

Remark 2.2.23. For simplicity, we assume that Z = in this remark. By the

proof of (2.2.7), the dierential operator of OD X PkD X /S (log D) is not a
log HPD dierential operator in general since the log HPD dierential oper-
ator ODD (1) OX X /S (log D) OD X +1 X /S (log D) induces a morphism
ODD (1) OX Pk X /S (log D) OD X Pk+1 +1
D D
X /S (log D), but does not in-
duce a morphism ODD (1) OX PkD X /S (log D) OD X PkD +1 X /S (log D)
D
in general; there does not exist a complex L(X,D)/S (Pk X /S (log D)) in
C+ (O(X,D)/S ) in general.
2.3 Forgetting Log Morphisms and Vanishing Cycle

Sheaves
In this section we investigate some properties of the forgetting log morphism

of log crystalline topoi.
Let the notations be as in 1.6. However, in this section, we denote the
underlying scheme of the log scheme Y also by Y by abuse of notation. Let
M be the log structure of Y . Let N M be also a ne log structure on Yzar .
Then we have a natural morphism
(2.3.0.1) = (Y,M,N )/S : (Y, M ) (Y, N )
of log schemes over S. The morphism induces a morphism of topoi which

is denoted by the same notation:
(2.3.0.2) = (Y,M,N )/S : ((Y, log

crys ((Y, N )/S)crys .
M )/S)log
When N is trivial, we denote (Y,M,N )/S by Y /S ; the morphism Y /S is a

p-adic analogue of the l-adic forgetting log morphism in [30] and [67, (1.1.2)].
In this section, let us assume the following condition on the log structure
N unless otherwise stated:
(2.3.0.3)
Locally on Y , there exists a chart P N such that P gp has no p-torsion.
Then we have the following lemma:

Lemma 2.3.1. Let the notation be as above and let (U, T, MT , , ) be an

object of ((Y, M )/S)log inv
crys , let NT be the inverse image of N |U /OU by the

proj.
following morphism: MT MT /OT M |U /OU
. Then NTinv is a ne log
structure on T (under the assumption (2.3.0.3)).
Proof. It is easy to see that NTinv is a log structure on T such that NTinv /OT =

N |U /OU . Set IT := Ker(OT OU ). Then we have the exact sequence
0 1 + IT NTinv,gp N gp |U 0.
Shrink U and take a chart : P N |U such that P gp has no p-torsion.

Then, since any element of 1 + IT is killed by some power of p, we have
Ext1 (P gp , 1 + IT ) = 0. Hence we have a homomorphism : P gp NTinv,gp
lifting gp locally on T and it induces the homomorphism of monoids
P NTinv , which we also denote by . If we denote the log structure

associated to P NTinv OT by P a , induces a homomorphism of
a : P a NTinv such that the induced homomorphism
log structures
a : P a /OT NTinv /OT is nothing but the identity on N |U /OU

. Hence
is an isomorphism, that is,
a
is a chart of NT . Therefore NT is a ne
inv inv
log structure.

2.3 Forgetting Log Morphisms and Vanishing Cycle Sheaves 85
Under the assumption (2.3.0.3), the explicit description of = ( , )

is given as follows: for an object F of ((Y, N )/S)log
crys and an object
(U, T, MT , , ) ((Y, M )/S)crys ,
log
(F )((U, T, MT , , )) = F ((U, T, NTinv , , ));
for an object G of ((Y,

M )/S)crys and an object (U, T, NT , , )((Y, N )/S)log
crys ,
(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 .
Let {Yi }iI be an open covering of Y . Let Mi (resp. Ni ) be the pull-back

of M (resp. N ) to Yi . Then we also have an analogous morphism of topoi
(2.3.2.1) : ((Y
, M )/S)log
crys ((Y , N )/S)crys ,
log
and we have the following commutative diagram
((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
(2.3.2.3) u(Y,N )/S = u(Y,M )/S , u(Y ,N )/S = u(Y ,M )/S
as morphisms of topoi.
Let the notations be as in 1.6. Then we have the following commutative
diagram:
((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
Let O(Y,L)/S (L := M, N ) be the structure sheaf in ((Y, L)/S)log

crys . Since

there is a morphism (O(Y,N )/S ) O(Y,M )/S , there is a morphism
(2.3.2.5) O(Y,N )/S (O(Y,M )/S ).
The morphism also induces a morphism
(2.3.2.6) : (((Y, log

crys , O(Y,M )/S ) (((Y, N )/S)crys , O(Y,N )/S )
M )/S)log
of ringed topoi. We have the analogues of the commutative diagrams (2.3.2.2)

and (2.3.2.4) for the ringed topoi:
(2.3.2.7)
(((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
(2.3.2.8)
(((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
The morphism (2.3.2.5) gives a morphism
(2.3.2.9) O(Y,N )/S R (O(Y,M )/S ).
Using (2.3.2.3), we have a morphism
(2.3.2.10) Ru(Y,N )/S (O(Y,N )/S ) Ru(Y,M )/S (O(Y,M )/S ).
Next we dene the localization of . Let FM = (UF , TF , MF , F , F ) be a

representable sheaf in ((Y,M )/S)log . Set FN := (UF , TF , N inv , F , F ),
crys F
where NFinv is the inverse image of N |UF by the morphism MF

M |UF /OU F
as before. Then we have a morphism
(2.3.2.11) |F : ((Y, log

crys |FM ((Y, N )/S)crys |FN
M )/S)log
of topoi and a morphism

(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 )
be the natural morphism. Then (U, T, MT , , , M ) is an object in ((Y, M )/S)

crys |FM .
log
Let (TF , MF )(TF ,NFinv ) (T, NT ) be the ber product of (TF , MF ) and
(T, NT ) over (TF , NFinv ) in the category of ne log schemes. We claim that
(2.3.3.1) (TF , MF ) (TF ,NFinv ) (T, NT ) = (T, MT ).

Indeed, let U : U UF be the open immersion. Then we have the following:
( (MF ) (NFinv ) NT )/OT = (MF )/OT (NFinv )/OT NT /OT

= 1 (MF /OT F ) 1 (NFinv /OT ) NT /OT
F
1
U (M |UF /OUF ) 1 (N |U )
/OU

N |U /OU
U F F

= M |U /OU (MF )/OT .
Hence the natural morphism (MF ) (MF ) (NFinv ) NT is an iso-

morphism and we have shown the claim. Denote (U, T, LT , , , L ) (L :=
M, N ) by (T, LT , L ) for simplicity of notation. By the formula (2.3.3.1),
(|F ) ((T, NT , N )) is represented by (T, MT , M ). Therefore, for an object
E in ((Y, M )/S)log |F , we have
crys M
(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 )).
Using this formula, we see that the functor |F is exact.

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
The obvious analogue of (2.3.4.1) for ringed topoi also holds.
Proof. Let G be an object of ((Y, N )/S)log

crys . By the proof of (2.3.3), (|F )

(FN ) = FM . Hence (|F ) jFN (G) = (|F ) (G FN ) = (G) FM =
jF M (G). Hence the former statement follows.
The latter statement immediately follows.

Lemma 2.3.5. Let FM = (Y, T, MT , , ) be a representable sheaf in ((Y, M )/
log
S)crys . Let E ((Y, M )/S)crys |FM be an O(Y,M )/S |FM -module. Then the
log
canonical morphism
jFM (E) R jFM (E)
is an isomorphism in the derived category D+ (O(Y,N )/S ).

Proof. Indeed, we have
(2.3.4) (2.3.3)
jFM (E) = jFN (|F ) (E) = jFN R(|F ) (E)
(2.2.1) (1)
= RjFN R(|F ) (E) = R(jFN |F ) (E)
(2.3.4) (2.2.1) (1)
= R(jFM ) (E) = R RjFM (E) = R jFM (E).

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
(Y,M )/S (E) R L(Y,M )/S (E)

LPD PD
(2.3.6.1)
is an isomorphism in the derived category D+ (O(Y,N )/S ).

Proof. (2.3.6) immediately follows from (2.2.1.2) and (2.3.5).

Lemma 2.3.7. Let (Y, M) be a log smooth scheme over S. Let N be a ne
sub-log structure of M on Y such that (Y, N ) is also log smooth over S. Let
M
(Y, M ) (Y, M)

(2.3.7.1) (Y,M,N )/S

(Y,M,N )/S

(Y, N ) N (Y, N )
be a commutative diagram whose horizontal morphisms are closed immer-

sions. Let DM and DN be the log PD-envelopes of M and N over (S, I, ),
respectively, with the natural following commutative diagram:
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)
of functors. Moreover, the functor (2.3.7.3) is functorial with respect to log

HPD dierential operators.

Proof. Let M : (((Y, crys |DM , O(Y,M )/S |DM ) (DM zar , ODM ) be
M )/S)log
the morphism of ringed topoi in (2.2.1.1). Let N be the analogue of M for
(Y, N ). Let
|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
(2.3.7.6) (Y,M,N )/S LPD

(Y,N )/S L(Y,M )/S .
PD
For TM := (U, T, MT , , ) in ((Y, M )/S)log inv

crys , let TN := (U, T, NT , , ) be as
above. Then, for an ODN -module E, the homomorphism
((Y,M,N )/S LPD

(Y,N )/S (E))TM (L(Y,M )/S (E))TN
PD
induced by (2.3.7.6) is given by the canonical homomorphism
ODU (TN S (Y,N )) ODN E ODU (TM S (Y,M)) ODM E,
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.

Remark 2.3.8. In (2.3.7), we do not have to assume the condition (2.3.0.3)

on the log structure N . The reason why we imposed the condition (2.3.0.3)
was to assure that the log structure NTinv is always ne. However, in the
situation in (2.3.7), the neness of NTinv for any T = (U, T, MT ) follows from
the assumption. Indeed, we have a morphism : (T, MT ) DM etale
locally on T and one can see that NTinv is isomorphic to the pull-back of the
log structure of DN by .
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 .
The following theorem is the crystalline Poincare lemma of a vanishing

cycle sheaf:
Theorem 2.3.10 (Poincar e lemma of a vanishing cycle sheaf ). Let

MS be the log structure of S. Let E be a crystal of O(Y,N )/S -modules and
let (E, ) be the ODM -module with integrable log connection corresponding to
(Y,M,N )/S (E). Assume that we are given the commutative diagram (2.3.7.1)

and that h in (2.3.7) is the identity. Then there exists a canonical isomor-
phism
(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 ).
Proof. By (2.2.8.1), we have an isomorphism

(2.3.10.2) (Y,M,N )/S (E) LPD
(Y,M )/S (E OY Y/S (log M/MS )).
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
R(Y,M,N )/S (Y,M,N )/S (E)

R(Y,M,N )/S LPD
(Y,M )/S (E OY Y/S (log M/MS ))

(Y,M,N )/S LPD
(Y,M )/S (E OY Y/S (log M/MS ))

(Y,N )/S (E OY Y/S (log M/MS )).
= LPD

We prove the boundedness of log crystalline cohomology in a general sit-

uation.
Proposition 2.3.11. Let (S, I, ) be the log PD-scheme in 1.6. Set S0 :=

SpecS (OS /I). Let f : X Y be a morphism of ne log schemes over S0 .

Assume that X and Y are quasi-compact and that f : X Y is quasi-
separated morphism of nite type. Let E be a quasi-coherent crystal of OY /S -
log
modules. Then Rfcrys (E) is bounded.
Proof. For (U, T, ) (Y /S)log

crys , put XU := XY U and denote the morphism
of topoi (f |XU ) uXU /T by fXU /T . By the same argument as [3, V Theorème
3.2.4], we are reduced to proving the following claim: there exists a positive
integer r such that, for any (U, T, ) (Y /S)log
crys and for any quasi-coherent
log i
crystal E on (XU /T )crys , we have R fXU /T (E) = 0 for i > r. Again, by the
same argument as [3, V Theorème 3.2.4, Proposition 3.2.5], we are reduced

to showing the above claim in the case where X and Y are suciently small
ane schemes. Hence we may assume that X admits a chart : P MX .
(Note that, in this book, log structures are dened on a Zariski site.) Let us
take surjections 1 : OY [Na ] OX and 2 : Nb P (a, b N). For (U, T,
MT , , ) (Y /S)log
crys , let us dene T := (T , MT ) by
T := SpecO (OT [Na Nb ]),

T
MT := the log structure associated to MT Nb OT [Na Nb ],
where the map MT Nb OT [Na Nb ] is induced by MT OT and the

natural inclusion Nb OT [Na Nb ]. Then we have the canonical ane log
smooth morphism g : T T . Let 1 be the morphism OT [Na Nb ] OXU
1 | U 2 |X
induced by Na OU [Na ] OXU and Nb P U MXU OXU . Let
2 be the morphism MT Nb MXU induced by MT MU MXU
|X
and Nb P U MXU . Then we have the closed immersion : XU T
2
of log schemes induced by 1 , 2 and we have the commutative diagram of

log schemes
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).

Corollary 2.3.12. Let (S, I, ) and S0 be as in (2.3.11). Let (Y, M ) be a

ne log smooth scheme over S0 such that Y is quasi-compact. Let N be a
ne sub log structure of M on Y . Then, for a quasi-coherent crystal E of
O(Y,M )/S -modules, the complex R(Y,M,N )/S (E) is bounded.
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
Ri fXU /T E = Hi (g (E(XU ,T ) OT T /T )) = 0
for i > r, where r is the maximum of the rank of 1X/Y,x (x X).
We give a proof of a lemma which has been used in (2.3.13), which is useful
also in later sections.
Lemma 2.3.14. Let S be a ne log scheme and let I be a quasi-coherent

nil-ideal sheaf of OS . Let S0 be an exact closed log subscheme of S dened

by I. Assume that S is ane. Let Z be a log smooth scheme over S0 . Then,

if Z is ane, there exists a unique log smooth lift Z (up to an isomorphism)
of Z over S and Z is also ane.
Proof. Let (P) be a property of a scheme or a morphism of schemes. In

this proof, for simplicity, we say that a log scheme W (resp. a morphism

f : W W of log schemes) has the property (P) if W (resp. f ) has the
property (P). Though the unique existence of Z seems more or less well-
known, we give a proof as follows (cf. [54, (3.14) (1)], [11, N.B. in 5.28]).
Express I as the inductive limit of the inductive system {I } of quasi-
coherent nilpotent ideal sheaves of OS : I = lim I . Let S be an exact

closed log subscheme of S dened by I . Since S0 = lim S and since Z is of

nite presentation over S0 , there exists a ne log smooth scheme Z over S
2.4 Preweight-Filtered Restricted Crystalline and Zariskian Complexes 93
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.

Let X be a smooth scheme over S0 and let D and Z be relative SNCDs

on X/S0 which meets transversally over S0 . In 2.7 below, we investigate
important properties of R(Y,M,N )/S (O(Y,M )/S ) for the case where (Y, M ) =
(X, D Z) and (Y, N ) = (X, Z).
2.4 Preweight-Filtered Restricted Crystalline

and Zariskian Complexes
Let (S, I, ) be a PD-scheme such that OS is killed by a power of a prime

number p and such that I is quasi-coherent. Set S0 := SpecS (OS /I). Let
f : X S0 be a smooth morphism and D a relative SNCD on X over S0 .

Let f : (X, D) S0 be the natural morphism of log schemes. By abuse of

notation, we also denote by f the composite morphism (X, D) S0 S.
The aim in this section is to construct two fundamental objects in
D+ F(QX/S (OX/S )) and in D+ F(f 1 (OS )) which we call the preweight-
ltered restricted crystalline complex of (X, D)/(S, I, ) and preweight-ltered
zariskian complex of (X, D)/(S, I, ), respectively. In fact, we construct these
complexes in a more general setting.
As explained in 2.1, X has the fs log structure M (D) dened by D.

As in 2.1, we denote this log scheme by (X, D). Let = {D } be
a decomposition of D by smooth components of D over S0 . Let X =
Xi0 be an open covering, where I0 is a set. Set Di0 := D Xi0
i0 I0
and D(;i0 ) := D Xi0 . Fix a total order on I0 and let I be the cat-
egory
r dened in1.5. For an object i= (i0 , . . . , ir ) I, set Xi :=
r r
s=0 is , Di :=
X s=0 Dis and D(;i) := s=0 D(;is ) . As explained in 1.6,
we have two ringed topoi (((X, D )/S)log
, Q
Rcrys (O(X ,D )/S )) and
(X ,D )/S
zar , f1 (OS )).

(X
Thus we have the following datum:

(2.4.0.1): An open covering X = i0 I0 Xi0 and the family {(Xi , Di )}iI of
log schemes which form a diagram of log schemes over the log scheme (X, D),
which we denote by (X , D ). That is, (X , D ) is nothing but a contravariant
functor
I o {smooth schemes with relative SNCDs over S0

which are augmented to (X, D)}
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
Let i = (i0 , . . . , ir ) be an object of I. Set X(i ,i) := Xi \ (X i \ Xi ) (0

r), where X i denotes the closure of Xi in Xi . Since X i \ Xi is
a closed subscheme of Xi , X(i ,i) is an open subscheme of Xi . It is easy

to see that the morphism Xi X(i ,i) is a closed immersion. Denote by
D(;i ,i) (resp. D(i ,i) ) the closed subscheme D(;i ) X(i ,i) (resp. Di

X(i ,i) ) of X(i ,i) . Set Xi := r=0 X(i ,i) . The closed immersions Xi
S

X(i ,i) ( = 0, . . . , r) induce an immersion Xi Xi . Blow up Xi along
r
=0 D(;i ,i) . Denote this scheme by Xi . We consider the complement
S
Xi of the strict transform of

r
(X(i0 ,i) S S X(i1 ,i) S D(;i ,i) S X(i+1 ,i) S S X(ir ,i) )
=0
in Xi . Let Di be the exceptional divisor on Xi . Then Di is a relative SNCD

on Xi by (2.4.2) below. Considering the strict transform of the image of Xi

of the diagonal embedding in Xi , we have an immersion Xi Xi , in fact,

an admissible immersion (Xi , Di ) (Xi , Di ) with respect to i by (2.4.2)

below. Let Di be the log PD-envelope of the immersion (Xi , Di ) (Xi , Di )
over (S, I, ) with structural morphism fi : Di S.
First we give the local description of OXi at a point of Di (cf. [47, 2],
[48, (1.7)], [64, 3.4]) for the warm up for the general description of OXi in
(2.4.2) below.
Lemma 2.4.1. Let i = (i0 , . . . , ir ) be an element of I. Then, Zariski locally

at the image of a point of Di in Xi , the structure sheaf OXi of Xi is etale
over the following sheaf of rings
(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),
(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).
Set U0 := Xi and let Ut (1 t s) be a scheme dened inductively as

follows: Ut is the blowing up of Ut1 with respect to the ideal sheaf It OUt1 .
Then, by [77, (5.1.2) (v)], Us = Xi . By the construction of Us , Us is covered
by the following spectrums over S of the following sheaves of rings:
(i1 ) (i1 ) (i ) (i )
A[u1 /u1 , . . . , us s /us s | 0 1 , . . . , s r]/
(i1 ) (i1 ) (i1 ) (i1 ) (i ) (i1 ) (i ) (i )

(x1 (u1 /u1 )x1 , . . . , xs s (us /us s )xs s ) (0 1 , . . . , s r).
Since the following equations

(i )
xt = 0 (1 t s, 0 r)
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 .
Theorem 2.4.2. Fix i = (i0 , . . . , ir ) I. Let A := r=0 OX(i ,i) be the

structure sheaf of Xi . Set i := { | D(;i ,i) = (0 r)}. (Then
we have i = Xi .)
Let J(;i ,i) ( i ) be the ideal sheaf of OX(i ,i) dening the closed im-

mersion D(;i ,i) X(i ,i) . Let X(i ,i) = (i ,i) X(i ,i) be an open covering
of X(i ,i) such that the restriction of J(;i ,i) to X(i ,i) is generated by a local
( )
section x (i ,i) for all i (such open covering exists by the commutative
diagram (2.1.7.2) for (X(i ,i) , D(i ,i) )). Set
(r) (r)
i := i ((i0 ,i) , . . . , (ir ,i) )
:= { i | D(;i ,i) X(i ,i) = , (0 r)}.
Then Xi is covered by the spectrums over S of the following sheaves of rings

((i ,i) (i0 ,i) ) 1 (r)
A[(u ) | i , 1 r]/
((i ,i) ) ((i ,i) (i0 ,i) ) ((i0 ,i) ) (r)
(x u x | i ) ((i0 ,i) , . . . , (ir ,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 .
Proof. We have the restriction

(Xi , Di ) (X(i ,i) , D(;i ,i) )
i

of the admissible immersion (Xi , Di ) (Xi , Di ) with respect to i
(0 r).
Set
M () := {((i0 ,i) , . . . , (ir ,i) ) | D(;i ) X(i ,i) = (0 r)}
and let M1 () be the set of the (is ,i) s (0 s r) appearing in an element

of M (). Then, by [77, (5.1.2) (v)], Xi is covered by the spectrums over S
of the following sheaves of rings:
((i ,i) ) ((i ,i) ) (r)
A[u
/u
| 0 r, i , (i ,i) M1 ()]/
((i ,i) ) ((i ,i) ) ((i ,i) ) ((i ,i) )

(x
(u
/u
)x
) (0 r, (i ,i) M1 ()).
Since the following equations
((i ,i) )
x = 0 (0 r)
are equivalent to
((i ,i) ) ((i ,i) )
u = 0, x = 0 (0 r),
Xi is covered by the spectrums of the quotient sheaves of

((i ,i) (i ,i) ) 1 (r)
A[(u ) | 0 = r, i , (i ,i) , (i ,i) M1 ()]
divided by ideal sheaves generated by

((i ,i) ) ((i ,i) (i ,i) ) ((i ,i) )
x u x ,
((i ,i) (i ,i) ) ((i ,i) ,(i ,i) )
u u 1,
and
((i ,i) (i ,i) ) ((i ,i) (i ,i) ) ((i ,i) (i ,i) )
u u u
(r)
(0 = = = r, i , (i ,i) , (i ,i) , (i ,i) M1 ()).
This quotient sheaf is isomorphic to

((i ,i) (i0 ,i) ) 1 (r)
A[(u ) | i ]/
((i ,i) (i0 ,i) ) ((i0 ,i) )
((i ,i) ) (r)
(x u x | i ).
r
Let D(;i) be the strict transform of =0 D(;i ,i) in Xi . Now we see that,
(r)
for i , the intersection of D(;i) and the inverse image of r=0 X(i ,i) in
S
( )
Xi is dened by an equation x (i0 ,i) = 0. Hence D(;i) is a smooth divisor on
Xi over S and Di is a relative SNCD on Xi over S, and D(;i) Xi Xi = D(;i) .
Therefore we obtain (2.4.2).

Now we change notations. Let X be a smooth scheme and let D and Z be
transversal relative SNCDs on X/S0 . Let D := {D } (resp. Z := {Z } )
be a decomposition of D (resp. Z) by smooth components of D (resp. Z).
Then D and Z give a decomposition := {D , Z }, of DZ by smooth
components of D Z. We can construct the objects in (2.4.0.1) and (2.4.0.2)

for D Z and : (X , D Z ) (X , D Z ). Let Di be the log PD-

envelope of the admissible immersion (Xi , Zi ) (Xi , Zi ) with respect to
(k) (k) (k)
Z |Xi . Set Zi |D(k) := Zi Di and Zi |D(k) := Zi Di (k N), where Di
i i
is a scheme over S dened in (2.2.13.2) for Di .
(k)
Lemma 2.4.3. The log scheme Di (Xi ,Zi ) (Di , Zi |D(k) ) is the log PD-enve-
i
(k) (k)
lope of the locally closed immersion (Di , Zi |D(k) ) (Di , Zi |D(k) ).
i i
Proof. (2.4.3) is a special case of (2.2.16) (2).

Let {PkDi }kZbe the ltration on Xi /S (log(Di Zi )) dened in (2.2.15.2).
As in 2.2, we set
PkDi L(Xi ,Zi )/S (Xi /S (log(Di Zi ))) := L(Xi ,Zi )/S (PkDi Xi /S (log(Di Zi ))),
PkDi (ODi OXi Xi /S (log Di )) := ODi OXi PkDi Xi /S (log(Di Zi )).

By (2.2.17) (1) and (2), we have two ltered complexes
(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 ),
(ODi OXi Xi /S (log(Di Zi )), P Di ) C+ F(fi1 (OS )).

Lemma 2.4.4. For a morphism : i j in I, let : (Xj , Dj Zj )
(Xi , Di Zi ) be the natural morphism. Then {(Xi , Di Zi ), }iI,Mor(I)
denes a diagram of smooth schemes with relative SNCDs over S :
I o {smooth schemes with relative SNCDs over S},

that is, for another morphism : j l in I, = , and idi =

id. Moreover, {Di }iI is a diagram of log schemes. In particular, there are
natural morphisms
: 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
: 1 (PkDi (ODi OXi Xi /S (log(Di Zi ))))

D
Pk j (ODj OXj Xj /S (log(Dj Zj )))
such that idi = id and = 1 ( ).

Proof. The open immersion Xj Xi induces a morphism Xj Xi . By
the universality of the blow ups, we have a morphism Xj Xi and this
morphism induces morphisms (Xj , Dj Zj ) (Xi , Di Zi ), (Xj , Dj )
(Xi , Di ) and (Xj , Zj ) (Xi , Zi ). The universality of the log PD-envelope
induces a morphism Dj Di . Thus (2.4.4) follows.

By (2.4.4), we obtain a complex
(Q(Xi ,Zi )/S PkDi L(Xi ,Zi )/S (Xi /S (log(Di Zi ))))iI
C+ (Q(X ,Z )/S (O(X ,Z )/S ))
of Q(X ,Z )/S (O(X ,Z )/S )-modules and a complex
(PkDi (ODi OXi Xi /S (log(Di Zi ))))iI C+ (f1 (OS ))
of f1 (OS )-modules. Now we have the following ltered complex of Q(X ,Z )

1
/S (O(X ,Z )/S )-modules and the following ltered complex of f (OS )-mod-
ules:
log,Z
(CRcrys (O(X ,D Z )/S ), P D ) :=
(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 ))

(((X, Z)/S)log
and
(2.4.5.2) zar , f1 (OS )) (X

zar : (X zar , f 1 (OS ))
be natural morphisms of ringed topoi dened in 1.5 and 1.6.

Denition 2.4.6. Assume that we are given the data (2.4.0.1) and (2.4.0.2)
for (X, D Z) with respect to = {D , Z }, .
(1) We call
(2.4.6.1)
log
R(X,Z)/SRcrys log,Z
(CRcrys (O(X ,D Z )/S ), P D ) D+ F(Q(X,Z)/S (O(X,Z)/S ))
the preweight-ltered restricted crystalline complex of O(X,DZ)/S (or (X, D

log,Z
Z)/S) with respect to D. We denote it by (CRcrys (O(X,DZ)/S ), P D ). If
log,Z
Z = , then we call (CRcrys (O(X,DZ)/S ), P ) the preweight-ltered re-
D
stricted crystalline complex of O(X,D)/S or (X, D)/S and we denote it by

(CRcrys (O(X,D)/S ), P ).
(2) We call
(2.4.6.2) log,Z
Rzar (Czar (O(X ,D Z )/S ), P D ) D+ F(f 1 (OS ))
the preweight-ltered zariskian complex of O(X,DZ)/S (or (X, D Z)/S)

log,Z
with respect to D. We denote it by (Czar (O(X,DZ)/S ), P D ). If Z = , then
log,Z D
we call (Czar (O(X,DZ)/S ), P ) the preweight-ltered zariskian complex of
O(X,D)/S or (X, D)/S and we denote it by (Czar (O(X,D)/S ), P ).
Let

(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
be the forgetting log morphism along D ((2.3.2)) and let

(2.4.6.4)
Z)/S)log 1
u(X,DZ)/S : (((X, D crys , O(X,DZ)/S ) (Xzar , f (OS ))
be the canonical projection ((1.6.0.8)).

Proposition 2.4.7. There exists the following canonical isomorphisms

(2.4.7.1) Q(X,Z)/S R(X,DZ,Z)/S (O(X,DZ)/S ) CRcrys
log,Z
(O(X,DZ)/S ),
(2.4.7.2)
log,Z
Ru(X,Z)/S (CRcrys (O(X,DZ)/S )) Ru(X,DZ)/S (O(X,DZ)/S ),

(2.4.7.3) 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
be natural morphisms of topoi dened in 1.6.

The isomorphism (2.4.7.1) follows from the cohomological descent [42,
log,Z
Vbis ], (2.3.2.2), (2.3.10.1), (1.6.4.1) and the denition of CRcrys (O(X,DZ)/S ).
Indeed, the left hand side of (2.4.7.1) is equal to
Q(X,Z)/S R(X,DZ,Z)/S R(X,DZ)/Scrys

log log,1
(X,DZ)/Scrys (O(X,DZ)/S )
=Q(X,Z)/S R(X,Z)/Scrys
log
R(X ,D Z ,Z )/S (O(X ,D Z )/S )
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 ).
By the trivially ltered case of (1.6.3.1) and by (2.4.7.1),

log,Z
Ru(X,Z)/S (CRcrys (O(X,DZ)/S ) = Ru(X,Z)/S R(X,DZ,Z)/S (O(X,DZ)/S )
= Ru(X,DZ,Z)/S (O(X,DZ)/S ).
(2.4.7.3) is a special case of [46, (2.20)], which follows from the cohomo-
logical descent.

Remark 2.4.8. (1) In the next section we shall prove that

log,Z
(CRcrys (O(X,DZ)/S ), P D ) D+ F(Q(X,Z)/S (O(X,Z)/S ))
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
(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 )
by the constructions of them.

(2) The complex Czar (O(X,D)/S ) in (2.4.6) is dierent from that dened in
[46, (2.19)]. Because the latter depends on an embedding system of (X, D),
it should be called a crystalline complex with respect to an embedding system.
2.5 Well-Denedness of the Preweight-Filtered

Restricted Crystalline and Zariskian Complexes
In this section we prove that the preweight-ltered restricted crystalline

complex
log,Z
in (2.4.6.1) and the preweight-ltered zariskian complex

log,Z
(Czar (O(X,DZ)/S ), P D ) D+ F(f 1 (OS ))
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
(2.5.1.1) (Q(X,Z)/S L(X,Z)/S (X2 /S (log(D2 Z2 ))), Q(X,Z)/S P D )
(Q(X,Z)/S L(X,Z)/S (X1 /S (log(D1 Z1 ))), Q(X,Z)/S P D ),

2.5 Well-Denedness of the Preweight-Filtered Complexes 103
(2.5.1.2) (OD2 OX2 X2 /S (log(D2 Z2 )), P D )
(OD1 OX1 X1 /S (log(D1 Z1 )), P D )

are ltered quasi-isomorphisms.
Q PD
Proof. Apply the gr-functor grk (X,Z)/S (k N) to (2.5.1.1). Then, by
(2.2.21.2), we obtain the following morphism:
Q PD
grk (X,Z)/S {(2.5.1.1)} :
Q(X,Z)/S acrys L(k) (D(k) /S (log Z2 |D(k) )){k}
(k)log
Z crys
(k)log
(D/S; Z)
2 2
Q(X,Z)/S acrys L(k) (D(k) /S (log Z1 |D(k) )){k} Z crys

(k)log (k)log
(D/S; Z).
1 1
Q PD
Then gk := grk (X,Z)/S {(2.5.1.1)} ts into the following commutative dia-
gram:
Q(X,Z)/S acrys L(k) ( (k)

(k)log (k)log
(log Z2 |D(k) )){k} Z crys (D/S; Z)
D2 /S
2

gk
Q(X,Z)/S acrys L(k) ( (k)

(k)log (k)log
(log Z1 |D(k) )){k} Z crys (D/S; Z)
D1 /S 1
Q(X,Z)/S acrys O(D(k) ,Z|

(k)log (k)log
)/S {k} Z crys (D/S; Z)
D (k)

Q(X,Z)/S acrys O(D(k) ,Z|
(k)log (k)log
D (k)
)/S {k} Z crys (D/S; Z),
where the horizontal morphisms are quasi-isomorphisms. Hence gk is also a
quasi-isomorphism and so is (2.5.1.1).
Applying the ltered direct image Ru(X,Z)/S to (2.5.1.1), we immedi-
ately see that (2.5.1.2) is a ltered quasi-isomorphism by the log version of
[11, 5.27.2, (7.1.2)].

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) )}

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 ),
Let Di (i = 1, 2) and D12 be the log PD-envelope of the admissible immersions

(X, D) (Xi , Di ) and (X, D) (X12 , D12 ), respectively.
Then the induced morphisms (X12 , D12 ) (Xi , Di ) (i = 1, 2) induce
morphisms of ltered complexes
(2.5.2.2) (QX/S LX/S (Xi /S (log Di )), QX/S P )

(QX/S LX/S (X12 /S (log D12 )), QX/S P ),
and
(2.5.2.3) (ODi OXi Xi /S (log Di ), P ) (OD12 OX12 X12 /S (log D12 ), P ),
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 ;
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-
plement of the strict transform of

[ E(;i,ij) S X(j,ij) )] [ X(i,ij) S E(;j,ij) ]

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 ))
be the natural morphisms dened in (1.6.0.2) and (1.6.0.3), respectively.

Let
1
Rzar : D+ F(f (OS )) D+ F((f1 (OS ))I ),
1
Ri,zar : D+ F(fi (OS )) D+ F(fi1 (OS ))
be the natural morphisms dened in (1.6.0.6) and (1.6.0.7), respectively. Then
we have the following:
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 .
(2.5.3.2) Rzar (OD OX X /S (log E ), P D )

=(OD OX X /S (log E ), 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).
Let ij : Xij Xi (i I, j J) be the natural morphism. Then

log D
Rcrys (Q(X ,Z )/S L(X ,Z )/S (X /S (log E )), Q(X ,Z )/S P )=
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
Thus there exists a natural composite morphism
(Q(X ,Z )/S L(X ,Z )/S (X /S (log E )), Q(X ,Z )/S P D )

log
Rcrys (Q(X ,Z )/S L(X ,Z )/S (X /S (log E )), Q(X ,Z )/S P D )
log
RRcrys (Q(X ,Z )/S L(X ,Z )/S (X /S (log E )), Q(X ,Z )/S P D ).
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 ).
Obviously we have e1 log log

i (Rcrys (I , {(I )k }) = i,Rcrys (Ii , {(Ii )k }). Hence
it suces to prove that the morphism
(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 })
is a ltered quasi-isomorphism. Henceforth we x i I in this proof.

If there exists a morphism j j in J, then there exists the natural open
log
immersion (X(i,ij) , E(i,ij) ) (X(i,ij ) , E(i,ij ) ). By the denition of i,Rcrys ,
we obtain an equality
(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

((E(;i) S X(i,i) ) (Xi S E(;i,i) ))

in this blowing up. Let Fi be the exceptional divisor on Wi . By considering

the strict transform of the image of Xi in Wi , we have a locally closed

immersion Xi Wi .
The two projections Wi Xi and Wi X(i,i) induce two mor-
phisms
(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 ).
Because (2.5.3.6) is a ltered quasi-isomorphism, so is the following composite

morphism
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 )

(Ji , {(Ji )k }).
Hence, by the ltered cohomological descent (1.5.1) (2), the following com-
posite morphism
(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
log
i,Rcrys (Ji , {(Ji )k })
is a ltered quasi-isomorphism. Because (2.5.3.5) is a ltered quasi-isomor-

phism, so is the following composite morphism
(2.5.3.8)
(Ji , {(Ji )k }).
The ltered quasi-isomorphism (2.5.3.8) induces a morphism

log
log
i,Rcrys (Ji , {(Ji )k }).
By the denition of the composite morphisms (2.5.3.7) and (2.5.3.8), the

following diagram is commutative:
(Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log Ei )),
log , {(J ) })
i,Rcrys (Ji i k
Q(Xi ,Zi )/S P Di )

log
i,Rcrys (Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log Ei )),
log , {(J ) }).
i,Rcrys (Ji i k
Q(Xi ,Zi )/S P Di )
We also have the following diagram

log
i,Rcrys (Q(Xi ,Zi )/S L(Xi ,Zi )/S (Xi /S (log Ei )),
log , {(J ) })
i,Rcrys (Ji i k
Q(Xi ,Zi )/S P Di )

log , {(I ) })
i,Rcrys (Ii i 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
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).

Corollary 2.5.4. Fix decompositions of D and Z by their smooth compo-

nents. Then the following hold:
log,Z
(1) (CRcrys (O(X,DZ)/S ), P D ) is independent of the data (2.4.0.1) and
(2.4.0.2).
(2) The following formula holds in D+ F(f 1 (OS )) :
(2.5.4.1)
log,Z
Ru(X,Z)/S (CRcrys log,Z
(O(X,DZ)/S ), P D ) = (Czar (O(X,DZ)/S ), P D ).
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.

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 ).
In particular, the ltered complex (Q(X,Z)/S L(X,Z)/S (X /S (log(D Z))),

Q(X,Z)/S P D ) is independent of the choice of the admissible immersion of
(X, D Z) over S if one xes D and Z .
(2)
(2.5.6.2) log,Z
(Czar (O(X,DZ)/S ), P D ) = (OD OX X /S (log(D Z)), P D )
in D+ F(f 1 (OS )). In particular, the ltered complex (OD OX X /S (log(D

Z)), P D ) is independent of the choice of the admissible immersion of (X, D
Z) over S if one xes D and Z .
Proof. By (2.5.4) (1), we have

log,Z
(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 ).
Hence (1) follows. The proof of (2) is the same.

Proposition 2.5.7. Let (S, I, ) and f : (X, E) := (X, D Z) S0

be as in 2.4. Let be a decomposition of E by smooth components of
D and Z. Let X = Xi0 be an ane open covering of X, where
i0 I0
I0 is a set. Set (X0 , E0 ) := ( i0 Xi0 , i0 (E Xi0 )) and (Xn , En ) :=
0 (X0 )n , cosk0 (E0 )n ) (n N). Let (Xn , Zn ) and (X
(coskX E
n , Dn ) be the ana-
logues of (Xn , En ) for Z and D, respectively. Set 0 := i0 |Xi0 ((2.1.12))
and let n (n Z>0 ) be the induced decomposition of En of smooth compo-
nents of En . Let
log log
Rcrys : (((Xn , Zn )/S)Rcrys , Q(Xn ,Zn )/S (O(Xn ,Zn )/S ))nN

(((X, Z)/S)log
be a natural morphism of ringed topoi. Then there exists an admissible immer-

sion (Xn , En )nN (Xn , En )nN of simplicial smooth schemes with simpli-
cial relative SNCDs over S with respect to (n )nN . Moreover,
(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 ).
Proof. Let I be a category whose objects are (i0 , . . . , ir )s (r N, i0 , . . . , ir

I0 ) and the morphism from i := (i0 , . . . , ir ) j := (j0 , . . . , js ) is one
point if {i0 , . . . , ir } {j
r0 , . . . , js } and empty
r otherwise. For an object i =
(i0 , . . . , ir ), set Xi := s=0 Xis , Ei := s=0 (E Xis ). Then we have the
following contravariant functor:
(X , E ) : I o {smooth schemes with relative SNCDs over S0 }.
The construction in 2.4 shows the existence of a diagram of admissible im-

mersions into a diagram of smooth schemes with relative SNCDs over S:

(X , E ) (X , E ) ( I ) with respect to , where is the induced
decomposition of E by ((2.1.12)). For an element j1 , j2 I0 , there ex-
ists two natural morphisms k : (X(j1 ,j2 ) , E(j1 ,j2 ) ) (Xjk , Ejk ) (k = 1, 2).
Using these morphisms, we have natural face morphisms m : (Xn , En )
(Xn1 , En1 ) (m = 0, . . . , n). Moreover, note that X(i,i) (i I0 ) is an open
scheme of the blow up of Xi S Xi by a closed subscheme of it. By considering

the strict transform of the diagonal immersion Xi Xi S Xi , we have a
natural morphism s : Xi X(i,i) . Using this morphism, we have natural
degeneracy morphisms sm : (Xn1 , En1 ) (Xn , En ) (m = 0, . . . , n 1).
The morphisms sm and m (m N) satisfy the standard relations in [90,
(8.1.3)]. Hence we have a desired simplicial log scheme (Xn , En )nN .
Fix a total order < on I0 . Let I be a subcategory of I whose objects are
(i0 , . . . , ir )s (r N, i0 < < ir , ij I0 ). Let
log
(X,Z)/SRcrys : (((X
, Z )/S)log
Rcrys , Q(X ,Z )/S (O(X ,Z )/S ))I

(((X, Z)/S)log
be a natural morphism of ringed topoi. Then we have

log,Z
(CRcrys (O(X,E)/S ), P D )
log
=R(X,Z)/Scrys ((Q(X ,Z )/S L(X ,Z )/S (X /S (log E )), Q(X ,Z )/S P D )I )
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 ).

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 ).
Proof. We immediately have (2.5.8) since we have the analogue of (2.5.4.1)

for
R log
(X,Z)/SRcrys
((Q(Xn ,Zn )/S L(Xn ,Zn )/S (Xn /S (log En )), Q(Xn ,Zn )/S P Dn )nN ).

2.6 The Preweight Spectral Sequence
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 ).
Theorem 2.6.1. Let a(k) : (D(k) , Z|D(k) ) (X, Z) (k N) be the natural

morphism. Let
(k)log
acrys : D+ (O(D(k) ,Z| )/S ) D+ (O(X,Z)/S )
D (k)
and
azar : D+ ((f a(k) )1 (OS )) D+ (f 1 (OS ))
(k)
be the induced morphisms by a(k) . Fix decompositions of D and Z by their

smooth components. Then there exist the following canonical isomorphisms
Dlog,Z
(2.6.1.1) grP
k (CRcrys (O(X,DZ)/S ))
=Q(X,Z)/S acrys (O(D(k) ,Z|

(k)log
)/S Z crys
(k)log
(D/S; Z)){k}
D (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)
Proof. Let the notations be as in 2.4. By applying Ru(X,Z)/S to both hands

of (2.6.1.1), we immediately have (2.6.1.2) by (1.3.4.1) and (2.5.4.1); hence
we have only to prove (2.6.1.1).
Let
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
(((D(k) , crys , O(D (k) ,Z|

Z|D(k) )/S)log )/S )
D (k)
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}
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}
(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
(Q(X ,Z )/S acrys L(X ,Z )/S (D(k) /S ( log Z |D(k) )

log (k)log
RRcrys

Z crys
(k)log
(D /S; Z )))
= Q(X ,Z )/S acrys (L(X ,Z )/S (D(k) /S (log Z |D(k) )) Z crys

(k)log (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 ))))
Q(X ,Z )/S acrys (L(D(k) ,Z | (D(k) /S (log Z |D(k) )){k}

(k)log
(k) )/S
D
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).

Corollary 2.6.2. Let k be a nonnegative integer. For integers k and h, set
E1k,h+k ((X, D Z)/S; k )

Rhk f(D(k) ,Z| (k) )/S (O(D(k) ,Z| (k) )/S Z crys (D/S; Z)) (k k ),
(k)log
:= D D
0 (k > k ).
Set f (X,Z)/S := f u(X,Z)/S . Then there exists the following spectral sequence
(2.6.2.1) E1k,h+k = E1k,h+k ((X, D Z)/S; k )

log,Z
= Rh f (X,Z)/S (PkD CRcrys (O(X,DZ)/S )).
In particular, there exists the following spectral sequence

2.6 The Preweight Spectral Sequence 115
(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)
Proof. Let (Ik , {Il }lk ) K+ F(Q(X,Z)/S (O(X,Z)/S )) be a ltered asque

log,Z log,Z
resolution of a representative of (PkD CRcrys (O(X,DZ)/S ), {PlD CRcrys (O(X,

)}
DZ)/S lk ) D +
F(Q(X,Z)/S (O(X,Z)/S )). Consider the following spectral
sequence
E1k,h+k = Hh (f (X,Z)/S grk (Ik )) = Hh (f (X,Z)/S Ik ).
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
E1k,h+k =Rh f (X,Z)/S (grP log,Z

D
k (CRcrys (O(X,DZ)/S )))
(k)log (k)log
=Rh f (X,Z)/S (Q(X,Z)/S (acrys (O(D(k) ,Z| )/S {k} Z crys (D/S; Z))))
D (k)
(k)log
=Rhk f(D(k) ,Z| )/S (O(D (k) ,Z| (k) )/S Z crys (D/S; Z)).
D (k) D
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
E1k,h+k ((X , D Z)/S; k )

Rhk fS (OD(k) OD(k) D(k) /S (log Z|D(k) ) Z zar (D/S)) (k k ),
(k) (k)
:=
0 (k > k ).
Then the following spectral sequence
(2.6.3.1) E1k,h+k := E1k,h+k ((X , D Z)/S; k )

= Rh fS (OD OX PkD X /S (log(D Z)))
is isomorphic to (2.6.2.1), and hence it is independent of the choice of the

admissible immersion fS . In particular, if fS : (X , D Z) S is a lift of
f : (X, D Z) S0 , then the following spectral sequence
(2.6.3.2)
E1k,h+k = E1k,h+k ((X , D Z)/S; k ) = Rh fS (PkD X /S (log(D Z)))
is independent of the choice of the lift. Here
E1k,h+k ((X , D Z)/S; k )

Rhk fS (D(k) /S (log Z|D(k) ) Z zar (D/S)) (k k ),
(k) (k)
=
0 (k > k ).
Proof. (2.6.3) immediately follows from (2.5.4.1) and (2.6.2.1).

Remark 2.6.4. In 2.9 below, we consider the functoriality of (2.6.2.2); in

particular, in the case where S0 is of characteristic p, we shall consider the
compatibility of (2.6.2.2) with the relative Frobenius F : (X, D) (X , D )
over S0 .
2.7 The Vanishing Cycle Sheaf and the Preweight

Filtration
Let S, S0 and f : (X, D Z) S0 be as in 2.4. Let a(k) : (D(k) , Z|D(k) )

(X, Z) be as in 2.2 (2). In 2.4 and 2.5, we have constructed the preweight-
ltered restricted crystalline complex
log,Z
such that
log,Z
CRcrys (O(X,DZ)/S ) = Q(X,Z)/S R(X,DZ,Z)/S (O(X,DZ)/S )
in D+ (Q(X,Z)/S (O(X,Z)/S )). Here

Z)/S)log log
(X,DZ,Z)/S : ((X, D crys ((X, Z)/S)crys

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.
We start with the following, which includes a crystalline analogue of Gab-

bers purity.
Theorem 2.7.1 (p-adic purity). Let k be a nonnegative integer. Then
(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
Let I be a asque resolution of O(X,DZ)/S . By (2.4.7.1) and by the exact-

ness of Q(X,Z)/S , we have
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 ),
and by (2.6.1.1) we have

D log,Z
Hh (grP
k CRcrys (O(X,DZ)/S ))
=Hhk (Q(X,Z)/S acrys (O(D(k) ,Z|

(k)log
)/S Z crys
(k)log
(D/S; Z)));
D (k)
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
Q(X,Z)/S Rk (X,DZ,Z)/S (O(X,DZ)/S )

(k)log
)/S Z crys
(k)log
(D/S; Z)).
D (k)

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
following spectral sequence
(2.7.1.3) E2st := Rs f(X,Z)/S Rt (X,DZ,Z)/S (O(X,DZ)/S ) =

Rs+t f(X,DZ)/S (O(X,DZ)/S ).

Set f (X,Z)/S := f(X,Z)/S Q(X,Z)/S : ((X, Z)/S)log log
Rcrys ((X, Z)/S)crys
Xzar . Because Rf (X,Z)/S Q
(X,Z)/S = Rf(X,Z)/S , (2.7.1.3) is equal to the
(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)
Rs+t f(X,DZ)/S (O(X,DZ)/S )

by (2.7.1).
log,Z
Using (2.7.1), we can give another simpler expression of (CRcrys (O(X,DZ)
D
/S ), P ). To do this, let us recall the canonical ltration of a complex.
Let (T , A) be a ringed topos and let E be an object in C(A). Then the
canonical ltration := {k E }kZ of E is dened as follows: k E i := E i
(i < k), := Ker(E k E k+1 ) (i = k), := 0 (i > k). Let E and F be objects
in C+ (A). Then a homotopy h between two morphisms f, g : E F also
gives a ltered homotopy between two morphisms f, g : (E , ) (F , ) of
ltered complexes. Furthermore, a quasi-isomorphism f : E F induces
a ltered quasi-isomorphism f : (E , ) (F , ); thus a functor C+ (A)
E
(E , ) C+ F(A) induces a functor D+ (A) D+ F(A), which is
also denoted by E
(E , ).
We prove the following lemma for a main result (2.7.3) below in this
section:
Lemma 2.7.2. Let f : (T , A) (T , A ) be a morphism of ringed topoi.
Then, for an object E in D+ (A), there exists a canonical morphism
(2.7.2.1) (Rf (E ), ) Rf ((E , ))
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
(2.7.2.2) (f (I ), {f (k I )}) (f (J ), {f (Jk )}).
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 ).
Theorem 2.7.3 (Comparison theorem). Let S0 , S, X, D and Z be as in

2.4. Set
log,Z
(2.7.3.1) (Ecrys (O(X,DZ)/S ), P D )
:=(R(X,DZ,Z)/S (O(X,DZ)/S ), ) D+ F(O(X,Z)/S ).
Then there exists a canonical isomorphism

(2.7.3.2)

Q(X,Z)/S (Ecrys
log,Z log,Z
(O(X,DZ)/S ), P D ) (CRcrys (O(X,DZ)/S ), P D ).
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:
(2.7.3.4) Q(X ,Z )/S (L(X ,Z )/S (X /S (log(D Z ))), )
(Q(X ,Z )/S L(X ,Z )/S (X /S (log(D Z ))), Q(X ,Z )/S P D ).

By (2.7.2) there exists a canonical morphism
(2.7.3.5)
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 ))), ).
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 )
is a quasi-isomorphism for each k Z. By the denition of the canonical

ltration and by the proof of (2.7.1), we have
(2.7.3.9)
Hi (grk Q(X,Z)/S R(X,DZ,Z)/S (O(X,DZ)/S ))

Q(X,Z)/S Rk (X,DZ,Z)/S (O(X,DZ)/S ) (i = k)
=
0 (i = k)

Q(X,Z)/S acrys (O(D(k) ,Z| (k) )/S Z crys (D/S; Z))
(k)log (k)log
(i = k),
= D
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 )
and we call it the preweight-ltered vanishing cycle zariskian complex of

(X, D Z)/S with respect to D.
Corollary 2.7.5. There exists a canonical isomorphism

(2.7.5.1) log,Z
(Ezar (O(X,DZ)/S ), P D ) (Czar
log,Z
(O(X,DZ)/S ), P D ).
Proof. The left hand side of (2.7.5.1) is equal to
Ru(X,Z)/S Q(X,Z)/S (Ecrys

log,Z
(O(X,DZ)/S ), P D )
log,Z
=Ru(X,Z)/S (CRcrys log,Z
(O(X,DZ)/S ), P D ) = (Czar (O(X,DZ)/S ), P D ).
Here we have used (2.5.4.1) for the last equality.

Corollary 2.7.6. The spectral sequence (2.7.1.4) is equal to (2.6.2.2) if we

make the renumbering Erk,h+k = Er+1
hk,k
(r 1).
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.
Proof. The proof is obvious.

Corollary 2.7.8. The isomorphism (2.6.1.1) is independent of the choice of

the decompositions of D and Z by their smooth components. Consequently the
isomorphism (2.6.1.2) and the spectral sequences (2.6.2.1), (2.6.2.2), (2.6.3.1)
and (2.6.3.2) are also independent of the choice.
Proof. Since both hands of (2.6.1.1) is independent of the choice by (2.7.3)

and (2.2.15), the problem is local. By (A.0.1) below, we may assume that two
choices of the decompositions of D and Z by their smooth components are
the same. Now the independence follows from the proof of (2.5.1) and the
argument in (2.5.3).

The following is another proof of (2.5.7):
Corollary 2.7.9. (2.5.7) and (2.5.8) hold.

Proof. By (1.6.4.1), (2.3.10.1) and the cohomological descent, we have
R log
(X,Z)/SRcrys (Q(Xn ,Zn )/S L(Xn ,Zn )/S (Xn /S (log En ))nN )
=Q(X,Z)/S R(X,DZ,Z)/S (O(X,DZ)/S ).
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 )),
Q(Xn ,Zn )/S PkDn )nN )

=(R log
(X,Z)/SRcrys (L(Xn ,Zn )/S (Xn /S (log En ))nN ), ).
Hence we have (2.5.7) and (2.5.8).

We shall use the following for the preweight-ltered K

unneth formula:
Proposition 2.7.10. Assume that X is quasi-compact. Then the ltered

log,Z
complex (Ecrys (O(X,DZ)/S ), P D ) is bounded.
Proof. By (2.3.11), R(X,DZ,Z)/S (O(X,DZ)/S ) is bounded. Hence (2.7.10)

immediately follows.

Remark 2.7.11. In this remark we show an unexpected nonequality

(2.7.11.1)
(k)
Rk (X,D)/S (O(X,D)/S ) = acrys (OD(k) /S Z crys
(k)
(D/S)) (k N)
in general. More specially, in this remark, we prove that the natural morphism
(2.7.11.2) OX/S (X,D)/S (O(X,D)/S ) = R0 (X,D)/S (O(X,D)/S )
is not surjective in general if p = 0 on S0 .

Let (X, D) (X , D) be an exact closed immersion into a smooth scheme
with a relative SNCD over S. Let : LX/S (1X /S ) LX/S (1X /S (log D)) be
a natural morphism of OX/S -modules, and let d : LX/S (OX ) LX/S (1X /S )
be the natural boundary morphism. By the crystalline Poincare lemma and
the Poincare lemma of a vanishing cycle sheaf ((2.3.10)), we have the follow-
ing:
(2.7.11.3)
=
OX/S Ker(d : LX/S (OX ) LX/S (1X /S ))

=
(X,D)/S (O(X,D)/S ) Ker( d : LX/S (OX ) LX/S (1X /S (log D)))
Consider the following commutative diagram of exact sequences:

0 0 LX/S (OX ) LX/S (OX ) 0

d d
0 Ker LX/S (1X /S ) Im 0.
Hence, by the snake lemma and (2.7.11.3), we obtain the following exact
sequence
(2.7.11.4) 0 OX/S (X,D)/S (O(X,D)/S ) Ker() Coker(d).
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
LX/S (1X /S )T = B OS [x] OS [x]dx = Bdx,
while the value LX/S (1X (log D))T is given by the following formula
LX/S (1X /S (log D))T = Bd log x.
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
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
and then the following commutative diagram:
B OS0 [x]

(2.7.11.6)
OS0 y OS0 .
Since the image of y 1 B by the left vertical morphism in (2.7.11.6) is

equal to y OS0 y, y 1 = 0 in B.
Next we check (B). It is clear that 1 x x 1 B is a local section of
the PD-ideal sheaf of B. Hence we have the following equalities in B
(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.
We also remark the following.

Rcrys ((2.7.1)), (O
By the p-adic purity in (X/S) X/S )X = ((X,D)/S (O(X,D)
/S ))X . Hence the exact sequence (2.7.11.4) tells us that (X,D)/S (O(X,D)/S )
is not a crystal of OX/S -modules in general.
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
(2.7.12.1) (X/C (log D), ) (X/C (log D), P )
is a quasi-isomorphism by using the Poincare residue isomorphism and the

Poincare lemma
Res

a (D (k) /C {k} Z
(k)
k X/C (log D)
grP (k) (D/C)(k))
(k)
a (CD (k) {k} Z
= (k) (D/C)(k)),
where (k) (Since we have used the

(k) (D/C) is the orientation sheaf of D
notation as a forgetting log morphism, we cannot use the notation in
[23]). Note that, in (2.7.1), (2.7.3) and (2.7.12.1), the graded pieces grP
k is
isomorphic to the complex which consists of one component; this property is
a key point for (2.7.1) and the quasi-isomorphism (2.7.12.1). It is reasonable
to expect that, if D is an SNCD, if we use the log innitesimal topos and if
we develop analogous theory for this topos by the same method as that in
this book, we will be able to prove that
(k)
(2.7.12.2) Rk (OX/C ) = a (OD(k) /C Z (k) (D/C)(k)),
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
(resp. (D (k) /C) ), a(k) :=

inf a(k) : D(k) = D (k) X and (k) (D/C) :=
(D/C).
(k)
(2) The morphism (2.7.2.1) is not a ltered isomorphism in general. Indeed,

if it were so, we would have the following contradiction.
Assume that Z = and that it were an isomorphism. Then, by applying
RuX/S to (2.7.3.2), we would have
2.8 Boundary Morphisms 127
(2.7.12.3) (Czar (O(X,D)/S ), P ) = RuX/S (Ecrys (O(X,D)/S ), P )

= RuX/S (R(X,D)/S (O(X,D)/S ), )
= (Ru(X,D)/S (O(X,D)/S ), )
= (Czar (O(X,D)/S ), ).
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
(X/ (log D), ) = (X/ (log D), P ).
If we take X := A1 , D: the origin of X and k = 0, we have a contradiction.

Hence (2.7.2.1) is not a ltered isomorphism in general.
2.8 Boundary Morphisms
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:
(2.8.0.2) 0 X/S0 (log Z) X/S0 (log(D Z))

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
I o {smooth schemes with relative SNCDs over S0

which are augmented to (X, D Z)}.
We have a diagram (D ) of a decomposition of D Z by a diagram of

smooth components of D Z .

(2.8.0.4): A family (X , D Z ) (X , D Z ) ( I) of admissible
immersions into a diagram of smooth schemes with relative SNCDs over S
with respect to (D ).
Let b : D X be a diagram of the natural closed immersions. By using

the Poincare residue isomorphism with respect to D , we have the following
exact sequence ([29, 2]):
0 X /S (log Z ) X /S (log(D Z ))
Res
(2.8.0.5)
bzar (D /S (log(Z |D )) Z zar

(1)
(D /S)){1} 0.
Let L(X ,Z )/S (resp. L(D ,Z |D )/S ) be the log linearization functor with

respect to the diagram of the locally closed immersions (X , Z )

(X , Z ) (resp. (D , Z |D ) (D , Z |D )). By (2.2.12) and (2.2.16),
L(X ,Z )/S bzar = alog
crys L(D ,Z |D )/S . Hence we have the following exact
sequence by (2.2.17) (2) and (2.2.21.2):
(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
(2.8.0.7)
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} .
By (2.2.7), (2.3.10.1) and by the cohomological descent, we have the following

triangle:
(2.8.0.8)
Q(X,Z)/S (O(X,Z)/S ) Q(X,Z)/S R(X,DZ,Z)/S (O(X,DZ)/S )
Q(X,Z)/S alog
+1
crys (O(D,Z|D )/S Z crys (D/S; Z)){1} .
(1)log
Using the Convention (4), we have the boundary morphism
(2.8.0.9)
d : Q(X,Z)/S alog
crys (O(D,Z|D )/S Z crys (D/S; Z))){1}
(1)log
Q(X,Z)/S (O(X,Z)/S )[1]
in D+ (Q(X,Z)/S (O(X,Z)/S )). Equivalently, we have the following morphism
(2.8.0.10)
d : Q(X,Z)/S alog
crys (O(D,Z|D )/S Z crys (D/S; Z)))
(1)log
Q(X,Z)/S (O(X,Z)/S )[1]{1}
in D+ (Q(X,Z)/S (O(X,Z)/S )). Set
(2.8.0.11) GD/(X,Z) := d.
and call GD/(X,Z) the Gysin morphism of D. Then we have a cohomology

class
(2.8.0.12)
c(X,Z)/S (D) := GD/(X,Z) Ext 0Q (Q(X,Z)/S alog
crys (O(D,Z|D )/S
(X,Z)/S (O(X,Z)/S )
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
c(X,Z)/S (D) Ext 0Q

(O(X,Z)/S ) (Q(X,Z)/S (O(X,Z)/S ), Q(X,Z)/S (O(X,Z)/S )[1]{1})
(X,Z)/S
=: Q(X,Z)/S H2log-crys ((X, Z)/S).
As usual, if Z = , we denote GD/(X,Z) and c(X,Z)/S (D) simply by GD/X

and cX/S (D), respectively.
Remark 2.8.1. (cf. [35, (1.6)]) Let t = 0 be a local equation of D in X . If

we use a Poincare residue morphism
X /S (log(D Z )) d log t
|D bzar (D /S (log(Z |D ))
Z zar
(1)
(D /S; Z ))[1]
instead of the Poincare residue morphism in (2.8.0.5), then we have a Gysin

morphism
GD/(X,Z) : Q(X,Z)/S alog

crys (O(D,Z|D )/S Z crys (D/S; Z))[1]
(1)log
Q(X,Z)/S (O(X,Z)/S )[1].
Here we have used the Convention (4). Hence, by the Convention (2), we
have a Gysin morphism
(2.8.1.1) GD/(X,Z) : Q(X,Z)/S alog

crys (O(D,Z|D )/S Z crys (D/S; Z))[2]
(1)log
Q(X,Z)/S (O(X,Z)/S ).
However we do not use this Gysin morphism in this book.
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
(X,Z)/S : (((X crys , O(X ,Z )/S )

, Z )/S)log
(((X crys , O(X ,Z )/S ),
, Z )/S)log
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 )
of ringed topoi. Then we have the following commutative diagram of triangles:

(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 ))
Q(X ,Z )/S R(X,Z)/S L(X ,Z )/S (X /S (log(D Z )))

Q(X ,Z )/S L(X ,Z )/S (X /S (log(D Z )))
Q(X ,Z )/S alog

crys R(D,Z|D )/S L(D ,Z |D )/S
+1

(D /S (log(Z |D )){1})

+1
Q(X ,Z )/S alog
crys L(D ,Z |D )/S (D /S (log(Z |D )){1}) .
By the proof of (2.5.3), the three vertical morphisms above are isomorphisms.
Hence (2.8.2) follows.

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
(2.8.3.1) O(X,Z)/S R(X,DZ,Z)/S (O(X,DZ)/S )
in D+ (O(X,Z)/S ). Let RD (O(X,Z)/S ) be the mapping ber of (2.8.3.1). Then

we have a triangle
(2.8.3.2)
+1
RD (O(X,Z)/S ) O(X,Z)/S R(X,DZ,Z)/S (O(X,DZ)/S ) .
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).
Let E be a representative of Q(X,Z)/S RD (O(X,Z)/S ). Then we have an

isomorphism

2 E E
and we can take an isomorphism

(2.8.3.5) 2 E Q(X,Z)/S HD
2
(O(X,Z)/S ){1}[1].
Therefore we have a canonical isomorphism
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
Since there exists a natural morphism RD (O(X,Z)/S ) O(X,Z)/S by the

denition of RD (O(X,Z)/S ), we have a canonical morphism
(2.8.3.6) 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 ).
By (2.8.0.8), we see that the morphism (2.8.3.6) is equal to GD/(X,Z) .

If we take the canonical isomorphism

(2.8.3.7) 2 E Q(X,Z)/S HD
2
(O(X,Z)/S )[2].
instead of (2.8.3.5), we obtain the Gysin morphism (2.8.1.1) again.
Proposition 2.8.4. Let u : (S , I , ) (S, I, ) be a morphism of PD-

schemes. Set S0 := SpecS (OS /I ). Let h : Y S0 be a smooth morphism
of schemes tting into the following commutative diagram
g
Y X

f
h
S0 S0 .
Set E := D X Y and W := Z X Y . Assume that E W is a relative SNCD

on Y over S0 . Let b : (E, W |E ) (Y, W ) be a natural closed immersion of
1
log schemes. Then the image of gzar Ru(X,Z)/S (c(X,Z)/S (D)) in (2.8.0.12) by
the natural morphism
1
gzar Ext 0f 1 (OS ) (Ru(X,Z)/S alog
crys (O(D,Z|D )/S Z crys (D/S; Z)),
(1)log
Ru(X,Z)/S (O(X,Z)/S )[1]{1})

Ext 0h1 (OS ) (Ru(Y,W )/S blog
crys (O(E,W |E )/S Z crys
(1)log
(E/S ; W )),
Ru(Y,W )/S (O(Y,W )/S )[1]{1})
is equal to Ru(Y,W )/S (c(Y,W )/S (E)).
Proof. (2.8.4) immediately follows from the functoriality of the construction

given in (2.8.3).

Finally we prove that the boundary morphism d 1 of (2.6.2.2) is expressed

by summation of Gysin morphisms with signs.
Henceforth D denotes a (not necessarily smooth) relative SNCD on X
over S0 which meets Z transversally. First, x a decomposition {D }
of D by smooth components of D over S0 . Assume that D{0 ,...,k1 } = .
j , . . . , k1 }, D := D{ ,..., } ,
Set := {0 , . . . , k1 }, j := {0 , . . . , 0 k1
and Dj := D{0 ,..., j ,...,k1 } for k 2 and D0 := X. Here
means the elimination. Then D is a smooth divisor on Dj over S0 . Let

j : (D , Z|D ) (Dj , Z|Dj ) be the closed immersion. Set
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
By (2.8.0.11) we have a morphism

(2.8.4.1) Gj := GD /(Dj ,Z|D ) :
j
j log
Q(D ,Z|D )/S crys
(1)log
(O(D ,Z|D )/S Z j crys (D/S; Z))
j j
Q(D ,Z|D )/S (O(Dj ,Z|Dj )/S )[1]{1}.

j j
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)
by the following morphism

j k1 )
(1)j (0 k1 ).
(j ) (0
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
dened by
(2.8.4.4)

j k1 ).
x (0 k1 )
(1)j Gj (x) (0
The morphism (2.8.4.3) induces a morphism of log crystalline cohomologies:

(2.8.4.5)
log
(1)j Gj : Rhk f(D ,Z|D )/S (O(D ,Z|D )/S Z crys (D/S; Z))
Rhk+2 f(Dj ,Z|D )/S (O(Dj ,Z|D )/S Z log

j crys
(D/S; Z)).
j j

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)).
Proposition 2.8.5. Let dk,h+k

1 : E1k,h+k E1k+1,h+k be the boundary
k1
morphism of (2.6.2.2). Set G := {0 ,...,k1 | i =j (i=j)} j=0 (1)j Gj .
Then dk,h+k
1 = G.
Proof. (cf. [64, 4.3]) Assume that we are given the data (2.4.0.1) and (2.4.0.2)
for D Z. Consider the following exact sequence
Q
(X ,D )/S P
D
0 grk1 (Q(X ,D )/S L(X ,Z )/S (X /S (log(D Z ))))
(Q(X ,D )/S PkD /Q(X ,D )/S Pk2

D
)(Q(X ,D )/S L(X ,Z )/S (X /S (log(D Z ))))
Q
(X ,D )S P
D
grk (Q(X ,D )/S L(X ,Z )/S (X /S (log(D Z )))) 0.
Then the boundary morphism dk,h+k

1 is induced by the boundary morphism
of the following triangle
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
)
(Q(X ,D )/S L(X ,Z )/S (X /S (log(D Z )))))
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 )))) .
Here we have used the Convention (4).

Assume that D(;) := D(0 ;) D(k1 ;) = . Set D(j ;) :=
D(0 ;) D(j1 ;) D(j+1 ;) D(k1 ;) . We use a shorter notation
zar (D /S; Z ) for a zariskian orientation sheaf 0 k1 zar (D /S; Z ) and
so on as for crystalline orientation sheaves.
The Poincare residue morphisms with respect to Dj (0 j k 1) and
D induce the following morphisms
D
ResD
j : grk1 X /S (log(D Z ))
P
D( /S (log Z |D(j ;) ){(k 1)} Z j zar (D /S; Z )

j ;)
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 )
by the following morphism

j k1 )
(1)j (0 k1 ).
(j ) (0
We identify j zar (D /S; Z ) Z j zar (D /S; Z ) with zar (D /S; Z ) by

this isomorphism. Let Resj be the Poincare residue morphism
(2.8.5.2) D( /S (log(D(;) Z |D(j ;) ))

j ;)
D(;) /S (log Z |D(;) ){1} Z j zar (D /S; Z )

with respect to the divisor D(;) on D(j ;) . Then we have a composite
morphism
(1)j Resj : D( /S (log(D(;) Z |D(j ;) )) Z j zar (D /S; Z )

j ;)
D(;) /S (log Z |D(;) ){1} Z j zar (D /S; Z ) Z j zar (D /S; Z )

D(;) /S (log Z |D(;) ){1} 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 ;)
L(X ,Z )/S (D(;) /S (log Z |D(;) ){1} Z crys

log
(D /S; Z )).
As in [64, 4.3], the morphism Q(X ,D )/S L(X ,Z )/S (ResD

j ) uniquely extends
to a morphism Q(X ,D )/S L(X ,Z )/S (ResD
j , ) tting into the following com-
mutative diagram:
(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
Here the morphism ResD

j , is dened by a formula
ResD
j , (yd log x0 d log xk1 ) = (1) yd log xj (0 j k1 ),
j
where xi = 0 (xi OX ) is a local equation of D(i ;) in X and y is a

local section of X (log Z ) (the formula ResIIq () = dxiq /xiq |DIq in [64,
p. 323, l. -9] have to be replaced by ResIIq () = (1)q1 dxiq /xiq |DIq ). By
the formulas (2.8.4.4) and (2.8.5.3), by the denition of the Gysin morphism
for smooth divisors ((2.8.0.11)) and by the Convention (4) and (5), we see

that (1)j (Gj ) is the boundary morphism of the lower exact sequence.
Hence we obtain (2.8.5).

2.9 The Functoriality of the Preweight-Filtered

Zariskian Complex
Let S0 , S and (X, D Z) be as in 2.4. In this section we prove the functo-

log,Z
riality of (Czar (O(X,DZ)/S ), P D ); (2.7.3) is indispensable for the proof of
the functoriality.
2.9 The Functoriality of the Preweight-Filtered Zariskian Complex 137
Let (S , I , ) be another PD-scheme satisfying the same conditions in the

beginning of 2.4. Set S0 := SpecS (OS /I ). Let u : (S, I, ) (S , I , )
be a morphism of PD-schemes. Let u0 : S0 S0 be the induced morphism
by u. Let (X , D Z ) be a smooth scheme with a relative SNCD over S0 .
Let
g
(X, D Z) (X , D Z )

(2.9.0.1)
u
S0 0 S0
be a commutative diagram of log schemes. Assume that the morphism g

induces g(X,D) : (X, D) (X , D ) and g(X,Z) : (X, Z) (X , Z ) over
u0 : S0 S0 . Let
Z)/S)log log
: ((X, D crys ((X, Z)/S)crys
and
: ((X , D
Z )/S )log ((X
crys
, Z )/S )log
crys
be the forgetting log morphisms along D and D , respectively.
Theorem 2.9.1 (Functoriality). Let the notations be as above. Then the

following hold:
(1) There exists a canonical morphism
log
log,Z
(2.9.1.1) g(X,Z)crys : (Ecrys (O(X ,D Z )/S ), P D )
log
Rg(X,Z)crys log,Z
(Ecrys (O(X,DZ)/S ), P D ).
(2) There exists a canonical morphism

(2.9.1.2)
log,Z
gzar : (Czar (O(X ,D Z )/S ), P D ) Rgzar (Czar
log,Z
(O(X,DZ)/S ), P D ).
Proof. (1): (1) is clear.

(2): Let
log
gcrys
: (((X, D Z)/S)log
crys ,O(X,DZ)/S )
(((X , D
Z )/S )log , O
crys (X ,D Z )/S )
be the morphism of log crystalline ringed topoi induced by g. Then we con-

struct a desired morphism in the following way:

log,Z
(Czar (O(X ,D Z )/S ), P D )

log,Z
= Ru(X ,Z )/S (Ecrys (O(X ,D Z )/S ), P D )
= Ru(X ,Z )/S (R (O(X ,D Z )/S ), )

Ru(X ,Z )/S (R Rgcrys
log
(O(X,DZ)/S ), )
log
= Ru(X ,Z )/S (Rg(X,Z)crys R (O(X,DZ)/S ), )
log
Ru(X ,Z )/S Rg(X,Z)crys (R (O(X,DZ)/S ), )
log log,Z
= Ru(X ,Z )/S Rg(X,Z)crys (Ecrys (O(X,DZ)/S ), P D )
log,Z
= Rgzar Ru(X,Z)/S (Ecrys (O(X,DZ)/S ), P D )
log,Z
= Rgzar (Czar (O(X,DZ)/S ), P D ).
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.
Proof. The proof is straightforward.

Let a(k) : (D(k) , Z |D (k) ) (X , Z ) be a natural morphism. Assume

that g induces a morphism gD(k) : (D(k) , Z|D(k) ) (D(k) , Z |D (k) ) for any
log
k N. By (2.6.1.1), (2.9.1) and (1.3.4.1), the morphism g(X,Z)crys induces
the following morphism
(2.9.2.2)
log
grP
k (g(X,Z)crys ) :
Ru(X ,Z )/S a crys (O(D (k) ,Z | (D /S ; Z )){k}

(k)log
)/S Z crys
(k)log
D (k)
Ru(X ,Z )/S a crys RgD

(k)log log
(k) crys (O(D (k) ,Z| )/S Z crys
(k)log
(D/S; Z)){k}.
D (k)
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:
(2.9.2.3): there exists the same cardinality

of smooth components
of D and
D over S0 and S0 , respectively: D = D , D = D , where D
and D are smooth divisors over S0 and S0 , respectively.
(2.9.2.4): there exist positive integers e ( ) such that e D = g (D ).
As in the previous section, set := {1 , . . . , k } (j , (i = j (i =

j))). Let a : (D , Z|D ) (X, Z) and a : (D , Z |D ) (X , Z ) be
natural morphisms. Consider the following direct factor of the morphism
(2.9.2.2):
(2.9.2.5)
Ru(X ,Z )/S a log log
crys (gcrys ) :
Ru(X ,Z )/S a log log

crys (O(D ,Z|D )/S Z crys (D /S ; Z )){k}

Ru(X ,Z )/S a log log log

crys Rgcrys (O(D ,Z|D )/S Z crys (D/S; Z)){k}.
Proposition 2.9.3. Let the notations and the assumptions be as above. Let
g(D ,Z|D ) : (D , Z|D ) (D , Z |D )
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
Here we dene j=1 ej as 1 for k = 0.
Proof. We may assume

that k 1. Let us take ane open coverings X =

i0 I0 Xi0 , X = i0 I0 Xi0 of X, X by the same index set I0 satisfying

g(Xi0 ) Xi0 (i0 I0 ) and let us form diagrams of log schemes (X , D Z )
and (X , D Z ) indexed by I as in (2.4.0.1). Then we have a morphism
g : (X , D Z ) (X , D Z ) of diagrams of log schemes over g. Next
let us take log smooth lifts

(Xi0 , Di0 Zi0 ) (Xi0 , Di0 Zi0 ), (Xi0 , Di0 Zi0 ) (Xi0 , Di0 Zi0 )
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 )
by the method explained in 2.4 before (2.4.1). Let g(X ,Z ) : (X , Z )

(X , Z ) be the morphism induced by g , which exists by assumption on g
and let zar be the morphism dened in (2.4.5.2). Then we have
(2.9.3.1)
D
P D
k Czar (O(X,DZ)/S ) = Rzar (OD OX grk
grP X /S (log(D Z ))),
log,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 ))),


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 )
of diagrams of log schemes by the universality of blow-up. Let

h(;) : (D(;) , Z |D(;) ) (D(;) , Z |D(;)
)
k
be the induced morphism. (Here we put D(;) := i=1 D(i ;) , D(;) :=
k
i=1 D(i ;) , where D(i ;) , D(i ;) are as in 2.4 before (2.4.1).)
For i0 I0 , Let x(j;i0 ) = 0 (resp. x(j;i0 ) = 0) be a local equation

of D(j ;i0 ) in Xi0 (resp. D( j ;i0 )
in Xi0 ) (1 j k). Then we have
e
gi0 (x(j;i0 ) ) = u(j;i0 ) x(j;ij 0 ) for some unit u(j;i0 ) . For i = (i0 , ..., ir ) I, let
us put x(j;i) := x(j;i0 ) , x(j;i) := x(j;i0 ) , u(j;i) := u(j;i0 ) . Then, by denition

of D(j ;i) , D( j ;i)
(via the blow-up construction), x(j;i) = 0 (resp. x(j;i) = 0)

is a local equation of D(j ;i) in Xi (resp. D( j ;i)
in Xi ) (1 j k) and
e
we have the equality gi (x(j;i) ) = u(j;i) x(j;i)
j
. So, for a local section =

ad log x(1;i) d log x(k;i) of PkD X /S (log(D Z )) (a k
X /S (log Z )),

we have gi () = ( j=1 ej ) gi (a)d log x(1;i) d log x(k;i) + , where
k
Di
Pk1 Xi /S (log(Di Zi )). So, if we put
(;) := D(;) /S (log Z |D(;) ) Z zar (D /S),
(;) := D (log Z |D(;)

) Z zar (D /S),
(;) /S
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 (;)

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
g(X ,Z )zar (OD OX (a |(D ,Z|D ) )zar (;) ){k}.

Now, by (2.9.3.1), (2.9.3.2), (2.9.3.4) and log crystalline Poincare lemma
for (D , Z|D ), (D , Z |D ), (2.9.3) is reduced to the following obvious lemma.

Lemma 2.9.4. Let F : A B be a left exact functor of abelian categories.

Let M and M (resp. N and N ) be objects of K+ (B) (resp. K+ (A)). Let
f
M F (N )

f
M F (N )
be the commutative diagram in K+ (B). Assume that A has enough injectives.

Then the following diagram is commutative:
f
M RF (N )

f
M RF (N ).

Denition 2.9.5. (1) We call {e } Z >0 the multi-degree of g with

respect to a decomposition := {D } and := {D } of D and D ,
respectively. We denote it by deg, (g) Z >0 . If e s for all s are equal,
we also denote e Z>0 by deg, (g) Z>0 .
(2) Assume that e s for all s are equal. Let u : E F be a morphism
of OS -modules. Let k be a nonnegative integer. The k-twist
u(k) : E(k; g; , ) F(k; g; , )
of u with respect to g, and is, by denition, the morphism

deg, (g)k u : E F.
Corollary 2.9.6. Assume that e s for all s are equal. Let Ess ((X, D
Z)/S) be the following spectral sequence
E1k,h+k ((X, D Z)/S)

= Rhk f(D(k) ,Z| )/S (O(D (k) ,Z|D(k) )/S )(k; g; , )
D (k)
= 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.
Proof. (2.9.6) immediately follows from (2.9.3).

Assume that S0 is a scheme of characteristic p > 0. Let FS0 : S0 S0

be the p-th power endomorphism. Let (X , D Z ) be the base change of
(X, D Z) by FS0 . The relative Frobenius morphism
F : (X, D Z) (X , D Z )
over S0 induces the relative Frobenius morphisms
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)
as the identity under the natural identication

(D /S; Z ) Fcrys crys
(k)log (k)log
(k)log
crys (D/S; Z).
When g is the relative Frobenius F : (X, D Z) (X , D Z ), we denote

(2.9.6.1) by
(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)
2.10 The Base Change Theorem and the K
unneth Formula 143
((2.9.6.2) is equal to (2.6.2.2)+(the compatibility with Frobenius).) (2.9.6.2)

is generalized to the following spectral sequence
(2.9.6.3) E1k,h+k = E1k,h+k ((X, D Z)/S; k )(k)

log,Z
= Rh f (X,DZ)/S (PkD CRcrys (O(X,DZ)/S ))
log,Z
= Rh f(X,DZ)/S (PkD Ecrys (O(X,DZ)/S ))
by (2.6.2.1) and (2.7.3.2).
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, ).
By the proof of (2.8.5) and (2.9.3), the morphism G in (2.8.5) is a morphism
(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)
Rhk+2 f(D(k1) ,Z| )/S (O(D (k1) ,Z|D(k1) )/S

D (k1)
Z crys
(k1)log
(D/S; Z))((k 1)).
By (2.7.6) we also have the following Leray spectral sequence
(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)
= Rs+t f(X,DZ)/S (O(X,DZ)/S ).

unneth
Formula
In this section we prove the base change theorem of a preweight-ltered van-

ishing cycle crystalline complex and the Kunneth formula of it. (2.7.5) plays
an important role in this section.
We keep the notations in 2.4. In this section we assume that X is quasi-
compact. Hence we can assume that the cardinality of the family {Xi0 }i0 I0
of an open covering of X is nite.
(1) Base change theorem.

g
Y Y

f
(2.10.1.1) f
u
(T , J , ) (T, J , )
be a commutative diagram of ne log schemes, where a PD-structure

(resp. ) on a PD-ideal sheaf J (resp. J ) of OT (resp. OT ) extends to
Y (resp. Y ) and u is a PD-morphism of PD-log schemes. Let (E , {Ek }) be
a bounded below ltered complex of OY /T -modules. Assume that RfY /T (E ,
{Ek }) is bounded above. Then there exists a canonical morphism
(2.10.1.2) Lu RfY /T (E , {Ek }) RfY /T gcrys

log1
(E , {Ek })
in DF(OT ).
Proof. By (1.2.3.2) we have only to nd an element in
H0 [{RHomOT (Lu RfY /T (E , {Ek }), RfY /T gcrys

log1
(E , {Ek })}0 ].
Using (1.2.2), we have the following formula
(2.10.1.3) RHomOT (Lu RfY /T (E , {Ek }), RfY /T gcrys

log1
(E , {Ek }))
= RHomOT (RfY /T (E , {Ek }), Ru RfY /T gcrys

log1
(E , {Ek }))
= RHomOT (RfY /T (E , {Ek }), RfY /T Rgcrys

log log1
gcrys (E , {Ek })).
The adjunction morphism (E , {Ek }) gcrys

log log1
gcrys (E , {Ek }) induces a
log
morphism (E , {Ek }) Rgcrys gcrys (E , {Ek }). This morphism induces
log1
a morphism
RfY /T (E , {Ek }) RfY /T Rgcrys

log log1
gcrys (E , {Ek })
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
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.
Proof. (1): Let (I , {Ik }) be a ltered asque resolution of (Ecrys

log,Z
(O(X,DZ)

/S ), P D
). Then Rf (X,Z)/S (E log,Z
crys (O(X,DZ)/S ), P D
) = (f u (X,Z)/S ) (I ,

{Ik }).
Now, x a decomposition {D } of D by its smooth components and give a
(k)log
total order on s. Then there exists an isomorphism Z crys (D/S; Z).
Furthermore, for each k, x a decomposition {(Z|D(k) ) } of Z|D(k) by its
smooth components and give a total order on s. Because X is quasi-
compact, the sets s and s are nite. By (2.6.2.2) we have the following
spectral sequence
(2.10.2.1) E1l,h+l = Rhl fZ (l) | /S (OZ (l) |D(k) /S )

D (k)
= Rh f(D(k) ,Z| )/S (O(D (k) ,Z|D(k) )/S ).

D (k)
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 )
in the ltered derived category DF(f 1 (OS )).

Proof. Let g(X,Z) : (X , Z ) (X, Z) and g(X,DZ) : (X , D Z )
(X, D Z) be the natural morphisms of log schemes. First we use the general
theory in 1.5 as follows.
Consider a small category I := {i, i } consisting of two elements. The mor-
phisms in I, by denition, consist of three elements idi , idi and a morphism
i i . By corresponding the natural morphism
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 ), ).
Therefore we have a canonical morphism (2.10.3.1) by (2.10.1) and

(2.10.2) (1).
We prove that (2.10.3.1) is an isomorphism. By the ltered cohomological
descent (1.5.1) (2) and by the same argument as that in the proof of [3, V
Proposition 3.5.2] ([11, 7.8 Theorem]), we may assume that S is ane and
that X is an ane scheme over S0 . Then (X, D Z) has a lift (X , D Z)/S
(D = D X X, Z = Z X X) by (2.3.14). In this case, we may assume

that the morphism (2.4.5.1) is the identity of (((X, Z)/S)log , O(X,Z)/S ). Let
crys
f : (X , D Z) S be the lift of f . Set f (PkD ) := f (PkD X /S (log(D Z)))
(k Z) and f (P D ) := {f (PkD )}kZ for simplicity of notation. Then, by
(2.7.5), we have
log,Z
Rf(X,Z)/S (Ecrys (O(X,DZ)/S ), P D ) = (f (X /S (log(D Z))), f (P D )).
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
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 )
f (X /S (log(D Z)))/f (PkD ) 0.
By the Poincare residue isomorphism, the left term of (2.10.3.2) is isomorphic

to f (b D(k) /S (log Z|D(k) ) Z zar (D/S)){k}, where b(k) : D(k) X
(k) (k)
is the natural morphism. Hence, the descending induction on k shows the

claim. Therefore the left hand side of (2.10.3.1) is equal to u f (X /S (log(D
Z)), P D ). Since f : X S is an ane morphism, we obtain (2.10.3) by the
ane base change theorem ([39, (1.5.2)]) as in the classical case ([11, 7.8
Theorem]).

As in [3, V] and [11, 7], we have some important consequences of (2.10.3).
Corollary 2.10.4. Let f : (X, D Z) Y be as in (2.10.2) (2). Then

log log,Z
Rf(X,Z)crys (Ecrys (O(X,DZ)/S ), P D )
is a ltered crystal in DF(OY /S ). That is, for a morphism v : (U , T , )

(U, T, ) of the crystalline site (Y /S)crys , the canonical morphism
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 .
Corollary 2.10.5. Let f : (X, D Z) Y be as in (2.10.2) (2). Assume

that Y has a smooth lift Y over S. Let h be an integer. Then the following
holds:
(1) There exists a quasi-nilpotent integrable connection

(2.10.5.1) log,Z
Rh f(X,Z)/Y (PkD Ecrys (O(X,DZ)/S ))
k
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
(2) For k Z, set
PkD Rh f(X,DZ)/Y (O(X,DZ)/S ) :=

log,Z
Im(Rh f(X,Z)/Y (PkD Ecrys (O(X,DZ)/S )) Rh f(X,DZ)/Y (O(X,DZ)/S )).
Then there exists a quasi-nilpotent connection
PkD Rh f(X,DZ)/Y (O(X,DZ)/S )

PkD Rh f(X,DZ)/Y (O(X,DZ)/S )OY 1Y/S .
Corollary 2.10.6. Let f : (X, D Z) Y be as in (2.10.2) (2). Let

g
(X , D Z ) (X, D Z)

f
f
h
Y Y

(S , I , ) (S, I, )
be a commutative diagram such that the upper rectangle is cartesian. As-

sume that Y is a quasi-compact smooth scheme over S . Then the natural
morphism
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.
Corollary 2.10.7. Let the notations and the assumptions be as in (2.10.2)

log
(1). Then Rf(X,Z)/Scrys log,Z
(PkD Ecrys (O(X,DZ)/S )) (k N) has nite tor-
dimension. Moreover, if S is noetherian and if f is proper, then Rf(X,Z)/S
log,Z
(PkD Ecrys (O(X,DZ)/S )) is a perfect complex of OS -module.
Denition 2.10.8. Let A be a noetherian commutative ring. Let (E , {Ek })

CF(A) be a ltered complex of A-modules. We say that (E , {Ek }) is
lteredly strictly perfect if it is bounded, if the ltration {Ekq } is nite for any
q and if all Ekq s are nitely generated projective A-modules.
Denition 2.10.9. Let A be a commutative ring with unit element. For a

ltered A-module (E, {Ek }) whose ltration is nite and for a family {Tl }lZ
A-modules, we say that (E, {Ek }) is the direct sum of {Tl }lZ if Ek =
of
lk Tl (k Z).
The following is a nontrivial ltered version of [11, 7.15 Lemma]:

unneth Formula 149
Theorem 2.10.10. Let A be a noetherian commutative ring. Let (E , {Ek })

be a ltered complex of A-modules. Assume that there exist integers k0 k1
such that Ekq1 = E q and Ekq0 = 0 for all q Z. Then (E , {Ek }) is quasi-
isomorphic to a lteredly strictly perfect complex if and only if Ek (k) has
nite tor-dimension and nitely generated cohomologies.
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 })
of ( q)-truncated ltered complexes such that the induced morphism

H (Qk ) H (Ek ) is an isomorphism for > q, Ker(Qqk Qq+1 k )
H q (Ek ) is surjective, Q = 0 for 0, Q = Qk1 , Qk0 = 0 (q 0) and
that (Qr , {Qrk }) (r q) is the direct sum of some family {Tkr }kZ of free
A-modules of nite rank.
For an integer k0 < k k1 , consider the ber product Ekq1 Ekq
Ker(Qqk Qq+1 q
k ). Let Ik be the image of the following composite morphism

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
if the image of an element of Ker(Qk Qk ) vanishes in H q (Ek ), then

q q+1
this element belongs to Im(Tkq1 Qqk ) by the denition of Tkq1 . In partic-

ular, this element belongs to Im(Qq1 k Qqk ). Hence the natural morphism

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).
Then we claim that Rkq 0 +l is a at A-module. We proceed on induction on

l. Unusually we assume that the initial case l = 1 holds and that l 2.
Consider the following exact sequence
0 Rkq 0 +l1 Rkq 0 +l grk0 +l Qq 0 (r l + 1 < q r).
By the induction hypothesis, we may assume that Rkq 0 +l1 (r l + 1 < q r)

is a at A-module. Since grk0 +l Qq is a at A-module, so is Rkq 0 +l (r l + 1 <
q r). Now we show that Rkrl+1
0 +l
is a at A-module. By the properties (a)
and (b), we have the following exact sequence
grk0 +l Qrl grk0 +l Qrl+1 Rkrl+1

0 +l
0.
For a positive integer i and for any A-module M ,
TorA rl+1
i (Rk0 +l , M )
=H i ( grk0 +l Qrl A M grk0 +l Qrl+1 A M 0)

=H rl+1i (grk0 +l Q A M ) = H rl+1i (Fk0 +l A M ) = 0.
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
unneth Formula 151
Set R := Rk := Rk1 for k k1 and Rk := 0 for k k0 . Then {Rk }kZ is

an increasing ltration on R since the natural morphism Rk0 +l1 Rk0 +l
is injective. Note that R is a bounded complex of projective A-modules.
Finally we claim that the natural morphism (Q , {Qk }) (R , {Rk })
is a ltered quasi-isomorphism. Indeed, for a positive integer l k1 k0 ,
grk0 +l R is the following complex
0 grk0 +l Qrl+1 /Im(grk0 +l Qrl grk0 +l Qrl+1 )

rl+1
grk0 +l Qrl+2 .
rl+2
This complex is isomorphic to grk0 +l Q by the properties (a) and (b).

Hence we have nished the proof of (2.10.10).

Corollary 2.10.11. Let the notations and the assumptions be as in (2.10.7).

log,Z
Then the ltered complex Rf(X,Z)/S (Ecrys (O(X,DZ)/S ), P D ) is a ltered
perfect complex of OS -modules, that is, locally on Szar , lteredly quasi-
isomorphic to a ltered strictly perfect complex.
Proof. (2.10.11) immediately follows from (2.10.7) and (2.10.10).

(2) K
unneth formula.
Next, we give the K unneth formula of preweight-ltered vanishing cycle

crystalline complexes.
Let Xj /S0 (j = 1, 2) be a smooth scheme with transversal relative SNCDs
Dj and Zj over S0 . Set X3 := X1 S0 X2 , D3 = (D1 S0 X2 ) (X1 S0 D2 )
and Z3 = (Z1 S0 X2 ) (X1 S0 Z2 ). Let fj : (Xj , Dj Zj ) S0
(j = 1, 2, 3) be the structural morphism. Assume that S is quasi-compact

and that f j (j = 1, 2) is quasi-compact and quasi-separated. We denote
log,Z log,Z
Rfj(Xj ,Zj )/S (Ecrys j (O(Xj ,Dj Zj )/S ), P Dj ) simply by Rf(Xj ,Zj )/S (Ecrys j
(O(Xj ,Dj Zj )/S ), P Dj ). We have the following commutative diagram of ringed
topoi for j = 1, 2:
(2.10.11.1)
log
qjcrys

(((Xj , D log log
j Zj )/S)crys , O(Xj ,Dj Zj )/S ) (((X3 , D3 Z3 )/S)crys , O(X3 ,D3 Z3 )/S )

(X ,D Z ,Z )/S (X ,D Z ,Z )/S
j j j j 3 3 3 3
log
pjcrys
(((Xj
, Zj )/S)log
crys , O(Xj ,Zj )/S ) (((X3
, Z3 )/S)log
crys , O(X3 ,Z3 )/S )

f(X ,Z )/S

f
(X3 ,Z3 )/S
j j
zar , OS )
(S zar , OS ),
(S
where qj : (X3 , D3 Z3 ) (Xj , Dj Zj ) and pj : (X3 , Z3 ) (Xj , Zj ) are

the projections. We shall construct a canonical morphism
log,Z1
(2.10.11.2) Rf(X1 ,Z1 )/S (Ecrys (O(X1 ,D1 Z1 )/S ), P D1 )L
OS
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)
Rf(X1 ,Z1 )/S (R1 (O(X1 ,D1 Z1 )/S ), )L

OS Rf(X2 ,Z2 )/S (R2 (O(X2 ,D2 Z2 )/S ), )
Rf(X3 ,Z3 )/S (R3 (O(X3 ,D3 Z3 )/S ), ).
To construct it, we need the following two lemmas:
Lemma 2.10.12 (cf. (2.7.2)). Let f : (T , A) (T , A ) be a morphism
of ringed topoi. Then, for an object E in D (A ), there exists a canonical
morphism
(2.10.12.1) Lf ((E , )) (Lf (E ), )
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 ).
Composing (2.10.12.2) with (2.10.12.3), we have a morphism (2.10.12.1).

Lemma 2.10.13. Let (T , A) be a ringed topos. Let E and F be two com-
plexes of A-modules. Assume that E is bounded above. Then there exists a
canonical morphism
unneth Formula 153
(2.10.13.1) (E , ) L L
A (F , ) (E A F , ).
Proof. Let P E be a at resolution of E . Let (Q , {Qk }) (P , )

be a ltered at resolution of (P , ). Then we have the following:
(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 , ).

Now we construct the canonical morphism (2.10). We need a canonical

element in
(2.10.13.2)

H0 [RHomO(X1 ,Z1 )/S (Lf(X3 ,Z3 )/S
{Rf(X1 ,Z1 )/S (R1 (O(X1 ,D1 Z1 )/S ), )L
OS
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
Rf(X2 ,Z2 )/S (R2 (O(X2 ,D2 Z2 )/S ), )}

=Lf(X3 ,Z3 )/S
Rf(X1 ,Z1 )/S (R1 (O(X1 ,D1 Z1 )/S ), )L
O(X3 ,Z3 )/S

Lf(X3 ,Z3 )/S
Rf(X2 ,Z2 )/S (R2 (O(X2 ,D2 Z2 )/S ), )

=Lplog L
1crys Lf(X1 ,Z1 )/S Rf(X1 ,Z1 )/S (R1 (O(X1 ,D1 Z1 )/S ), )O(X 3 ,Z3 )/S

Lplog
2crys Lf(X2 ,Z2 )/S Rf(X2 ,Z2 )/S (R2 (O(X2 ,D2 Z2 )/S ), )
Lplog L
1crys (R1 (O(X1 ,D1 Z1 )/S ), )O(X 3 ,Z3 )/S
Lplog
2crys (R2 (O(X2 ,D2 Z2 )/S ), ).
Note that Rj (O(Xj ,Dj Zj )/S ) (j = 1, 2, 3) is bounded above by (2.7.10).

Therefore it suces to construct a canonical morphism
(2.10.13.3)
Lplog L
1crys (R1 (O(X1 ,D1 Z1 )/S ), )O(X Lplog
2crys (R2 (O(X2 ,D2 Z2 )/S ), )
3 ,Z3 )/S
(R3 (O(X3 ,D3 Z3 )/S ), ).

We also have the following composite morphism
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
(R3 (O(X3 ,D3 Z3 )/S ), )L

O(X (R3 (O(X3 ,D3 Z3 )/S ), )
3 ,Z3 )/S
(R3 (O(X3 ,D3 Z3 )/S ), ).
By (2.10.13), it suces to construct a canonical morphism
(R3 (O(X3 ,D3 Z3 )/S )L

O(X R3 (O(X3 ,D3 Z3 )/S ), )
3 ,Z3 )/S
(R3 (O(X3 ,D3 Z3 )/S ), )
and, furthermore, to construct a canonical morphism
R3 (O(X3 ,D3 Z3 )/S ) L

O(X R3 (O(X3 ,D3 Z3 )/S )
3 ,Z3 )/S
R3 (O(X3 ,D3 Z3 )/S ).
Hence we have only to have a canonical element of
(2.10.13.4) H0 [RHomO(X3 ,D3 Z3 )/S (L3 {R3 (O(X3 ,D3 Z3 )/S )L

O(X 3 ,Z3 )/S
R3 (O(X3 ,D3 Z3 )/S )}, O(X3 ,D3 Z3 )/S )].

The source of [ ] in (2.10.13.4) is
L3 R3 (O(X3 ,D3 Z3 )/S ) L

O(X L3 R3 (O(X3 ,D3 Z3 )/S ).
3 ,D3 Z3 )/S
Using the adjunction, we have a composite morphism
L3 R3 (O(X3 ,D3 Z3 )/S ) L

O(X L3 R3 (O(X3 ,D3 Z3 )/S )
3 ,D3 Z3 )/S
O(X3 ,D3 Z3 )/S L

O(X O(X3 ,D3 Z3 )/S = O(X3 ,D3 Z3 )/S .
3 ,D3 Z3 )/S
Thus we have a morphism (2.10.11.2).
Theorem 2.10.14 (K unneth formula). (1) Let the notation be as above.

log,Z1
(2.10.14.1) Rf(X1 ,Z1 )/S (Ecrys (O(X1 ,D1 Z1 )/S ), P D1 )L
OS
log,Z2
Rf(X2 ,Z2 )/S (Ecrys (O(X2 ,D2 Z2 )/S ), P D2 )
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 ).
Proof. (1): By virtue of the ltered cohomological descent (1.5.1) (2), we

may assume that Xj (j = 1, 2) and S are ane as in the proof of [3, V
Corollary 4.2.2], and hence that (Xj , Dj Zj ) (j = 1, 2) has a log smooth lift
(Xj , Dj Zj ) over S. Let (X3 , D3 Z3 ) be the ber product of (X1 , D1 Z1 )
and (X2 , D2 Z2 ) over S. Let gj : (Xj , Dj Zj ) S (j = 1, 2, 3) be the
structural morphism. In this case, by (2.7.5), the proof of (1) is reduced to
showing an isomorphism
(g1 X1 /S (log(D1 Z1 )) OS g2 X2 /S (log(D2 Z2 )),

{ g1 PlD1 X1 /S (log(D1 Z1 )) OS g2 Pm
D2
X2 /S (log(D2 Z2 ))}kZ )
l+m=k
g3 (X3 /S (log(D3 Z3 )), P D3 ),

which is easily veried.
(2): (2) follows from (1) as in [3, V Theorem 4.2.1].

The following is the compatibility of the preweight-ltered K
unneth for-
mula with the base change formula.
Proposition 2.10.15. Let u be the morphism in (2.10.3). Let mean the base
change of an object over S by u|S0 . Let K be the preweight-ltered K
unneth
isomorphism (2.10.14.1) and K the preweight-ltered K
unneth isomorphism
for (Xi , Di Zi ) (i = 1, 2, 3). Set
log,Zi
Hi := Rf(Xi ,Zi )/S (Ecrys (O(Xi ,Di Zi )/S ), P Di )
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 .
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]).

2.11 Log Crystalline Cohomology with Compact

Support
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
Because : (U, MU ) (T, MT ) is an exact closed immersion, MT /OT =

MU /OU on Uzar = Tzar . Hence the dening local equation of the relative
SNCD D U on U lifts to a local section t of MT . We dene an ideal
sheaf I(X,DZ)/S
D
O(X,DZ)/S by the following: I(X,DZ)/S
D
(T )= the ideal
generated by the image of t by the structural morphism MT OT . One
can prove that Q(X,DZ)/S (I(X,DZ)/S
D
) is a crystal on the restricted log
crystalline site ((X, D Z)/S)log
Rcrys in the same way as [85, (5.3)].
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
where X /S (log(Z D)) := OX (D)X /S (log(D Z)). As a result, we have
Rh f(X,DZ)/S,c (O(X,DZ;Z)/S ) = Rh fS (OD OX X /S (log(Z D))).
Let {D } be a decomposition of D by smooth components of D. Let

the notations be as in 2.8. The exact closed immersion j : (D , Z|D )
(Dj , Z|Dj ) induces the morphism
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
{0 ,1 , ,k1 | i =l (i=l)} j=0
(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 .
Proof. The question is local. Let and be two decompositions of D by

smooth components of D. Let x be a point of X. By (A.0.1) below, there
exists an open neighborhood U of x such that |U = |U . Thus we have
(2.11.2).

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)
(2)log (2)log
Z crys (D/S; Z)) ).
(2)log crys
acrys (O(D(2) ,Z| )/S
D (2)
Then there exists the following canonical isomorphism in D+ (Q(X,Z)/S

(O(X,Z)/S )) :

(2.11.3.2) Q(X,Z)/S R (I(X,DZ)/S
D
) Q(X,Z)/S Ecrys,c
log,Z
(O(X,DZ)/S ).
Before the proof of (2.11.3), we prove two lemmas.

Lemma 2.11.4. There exists a morphism of topoi
Z)/S)log log
Rcrys : ((X, D Rcrys ((X, Z)/S)Rcrys
tting into the following commutative diagram of topoi:
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 )
of i Zariski locally on U such that the rst morphism is an exact closed

immersion and that the second morphism is log etale. Then T is the log
PD-envelope of the rst morphism. Hence we may suppose that i is an exact
closed immersion. Then, by (2.1.5), we may assume that i is an admissible

closed immersion (U, (D Z)|U ) (U, D Z). In this case, the log struc-
ture NTinv on T is nothing but the pull-back of the log structure on U dened
by Z. Hence (T, NTinv ) is the log PD-envelope of the exact closed immer-

sion (U, Z|U ) (U, Z). Hence (U, T, NTinv , , ) belongs to ((X, Z)/S)log
Rcrys
Zariski locally on T . Now it is clear that we have the morphism Rcrys of
topoi. It is easy to see that we have the commutative diagram (2.11.4.1).
Lemma 2.11.5. Let the notations be as in (2.11.4). Then the following nat-
ural morphism of functors
Q(X,Z)/S R RRcrys Q(X,DZ)/S

for O(X,DZ)/S -modules is an isomorphism.

Proof. By the same argument as that in the proof of (1.6.4), we are reduced to
showing that, for any parasitic O(X,DZ)/S -module F of ((X, D Z)/S)log crys ,
Rq (F ) is also parasitic for any q 0. To see this, it suces to prove that, for
any object T := (U, T, MT , , ) ((X, Z)/S)log Rcrys with T suciently small,
the sheaf (Rq (F ))T on Tzar induced by Rq (F ) is equal to zero. Hence we

may assume that there exists a closed immersion i : (U, Z|U ) X into an
ane log smooth scheme over S such that (T, MT ) is the log PD-envelope of

i. On the other hand, let us take a closed immersion i : (U, (D Z)|U ) Y
into an ane log scheme which is log smooth over S. Then, for any n Z1 ,

we have the closed immersion in : (U, (D Z)|U ) X S Y n induced by

i (|U ) and i . Let D(n) be the log PD-envelope of the closed immersion
in over (S, I, ). Then it is isomorphic to the log PD-envelope of the closed

immersion (U, (D Z)|U ) (T, MT ) S Y n (induced by the composite
(|U ) and i ) compatible with , where is the PD-structure on Ker(OT
OU ) + IOT extending and . By the log version of [3, V 1.2.5], we have

(Rq (F ))T = Rq f(U,(DZ)|U )/T F = Rq ( (|U )) CA(F ),

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
By the same argument as that in [27, (4.2.2) (a), (c)], the following se-
quence
(2.11.5.1) 0 X /S (log(Z D )) X /S (log Z ) Z zar

(0)
(D /S)
(0) (1)
b (D(1) /S (log Z |D(1) ) Z zar
,zar (1) ,zar
(1)
(D /S))

similarly as for crys . Hence X /S (log(Z

(k) (k)log
is exact. Here we dene ,zar
D )) is quasi-isomorphic to the single complex of the following double
complex
(2.11.5.2)

d d
(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
We claim that the following sequence
(2.11.5.3)
0 Q(X ,Z )/S L(X ,Z )/S (X /S (log(Z D )))

Q(X ,Z )/S L(X ,Z )/S (X /S (log Z )
Q (0)
(X ,Z )/S (,zar )
Z zar
(0)
(D /S))
Q(X ,Z )/S L(X ,Z )/S (b (D(1) /S (log Z |D(1) )

(1)

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)
0 ODi OXi Xi /S (log(Zi Di ))

ODi OXi Xi /S (log Zi ) Z zar
(0)
(Di /S)
(0) (1)
i,zar i,zar
ODi OXi bi (D(1) /S (log Zi |D(1) ) Z zar
(1) (1)
(Di /S))
i i
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
OXi [xd+1 , . . . , xd ] OXi /JiN
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)
(2.11.5.5) 0 Xi /S (log(Zi Di )) Xi /S (log Zi ) Z zar

(0)
(Di /S)
i,zar i,zar
(0) (1)
b i (D (1) /S (log Zi |D (1) ) Z zar (Di /S))

(1) (1)
i i
Since OS xd+1 , . . . , xd OX qX /S

(q N) is a free OS -module, applying
the tensor product OS OS xd+1 , . . . , xd OX qX /S (q N) to the exact
sequence (2.11.5.5) preserves the exactness. Because
ODi OXi Xi /S (log(Zi Di )) Xi /S (log(Zi Di ))OS

OS xd+1 , . . . , xd OX X /S
and because the similar formulas for ODi Xi bi ( (k)

(k)
(log Zi |D(k) ) Z
Di /S i
(k)
zar (Di /S))(k N) hold, we have the exactness of (2.11.5.4).
By (2.2.12) and (2.11.5.3), we have the following quasi-isomorphism
(2.11.5.6) Q(X ,Z )/S L(X ,Z )/S (X /S (log(Z D )))

{Q(X ,Z )/S L(X ,Z )/S (X /S (log Z ) Z zar
(0)
(D /S))
(Q(X ,Z )/S acrys L(D(1) ,Z | (D(1) /S (log Z |D(1) ))
(1)log
(1) )/S
D
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))
Q(X ,Z )/S acrys L(D(1) ,Z | (D(1) /S (log(Z |D(1) ))

log (1)log
(R(X,Z)/SRcrys )/S
D (1)
Z zar
(1)
(D /S), d) }.
(See (2.11.8) below.) By (1.6.4.1) and (2.2.20.1), the isomorphism (2.11.5.7)

is nothing but an isomorphism (2.11.3.2).
Now we show that the isomorphism (2.11.3.2) is independent of the data
(2.4.0.1) and (2.4.0.2).
Let the notations be as in the proof of (2.5.3). Let
log
RRcrys : D+ F(Q(X ,Z )/S (O(X ,Z )/S ))
D+ F((Q(X ,Z )/S (O(X ,Z )/S ))I )
be a morphism of ltered derived categories in 2.5. Then we have the fol-

lowing commutative diagram by the cohomological descent:

Q (
log
R L (log(Z D )))

(X,Z)/SRcrys (X ,Z )/S (X ,Z )/S X /S

RRcrys Q (
log log
R L (log(Z D )))

Q (
log (0)
{R L (log Z ) Z zar (D /S))
RRcrys Q (
log log (0)
{R L (log Z ) Z zar (D /S))
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) }.
Hence the isomorphism (2.11.3.2) is independent of the data (2.4.0.1) and

(2.4.0.2).

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
{L(X ,Z )/S (X /S (log Z ) Z zar

(0)
(D /S))
(D(1) /S (log Z |D(1) )) Z zar

(1)log
(acrys L(D(1) ,Z | (1)
(D /S)), d) }.
(1) )/S
D
Then, by the proof of (2.11.3), we see that the isomorphism (2.11.3.2) is

obtained by applying Q(X,Z)/S to the following diagram:
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 .
Let PcD := {PcD,k }kZ be the stupid ltration on Ecrys,c

log,Z
(O(X,DZ)/S ).
Then, by (2.11.3), we have a ltered complex (Ecrys,c (O(X,DZ)/S ), PcD )
log,Z
D+ F(O(X,Z)/S ).
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.

By applying Rf to both hands of (2.11.7.1) (cf. (2.11.8)), we have a

canonical isomorphism
(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))
(Rf(D(1) ,Z| )/S (O(D (1) ,Z|D(1) )/S Z crys

(1)
(D/S; Z)), d) }.
D (1)
log,Z
Next we prove the base change theorem of (Ecrys,c (O(X,DZ)/S ), PcD ).
Proposition 2.11.10. Let the notations and the assumptions be as in

log,Z
(2.10.2) (1). Then Rh f(X,Z)/S (PcD,k Ecrys,c (O(X,DZ)/S )) (h, k Z) is a
quasi-coherent OS -module and Rf(X,Z)/S (PcD,k Ecrys,c
log,Z
(O(X,DZ)/S )) (k Z)
has nite tor-dimension.
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
(2.11.11.1) Lu Rf(X,Z)/S (Ecrys,c

log,Z
(O(X,DZ)/S ), PcD )

Rf(X ,Z )/S (Ecrys,c
log,Z
(O(X ,D Z )/S ), PcD ).
Proof. Let I be a double complex of O(X,Z)/S -modules such that, for

(k)log
each k N, I k is an injective resolution of (acrys (O(D(k) ,Z| )/S Z
D (k)
(k)log
crys (D/S; Z)), (1)k d). Then we have a double complex ((f u(X,Z)/S )
(I 0 ) (f u(X,Z)/S ) (I 1 ) ). This double complex is a representative
log,Z
of Rf(X,DZ)/S (Ecrys,c (O(X,DZ)/S )). For a nonnegative integer r, let r (f
u(X,Z)/S ) (I ) be the canonical ltration of the complex (f u(X,Z)/S ) (I k ).
k
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
d : (f u(X,Z)/S ) (I l ) (f u(X,Z)/S ) (I +1,l )
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
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 )).

Proposition 2.11.12. Let the notations and the assumptions be as those in

(2.10.2) (1). Assume moreover that f : X S0 is proper. Then
log,Z
Rf(X,Z)/S (Ecrys,c (O(X,DZ)/S ), PcD )
is a lteredly strictly perfect complex.
Proof. We use the criterion (2.10.10); we have checked the condition as to

the tor-dimension in (2.11.10) and we obtain the niteness from the spectral
sequence (2.10.2.1) and [11, 7.16 Theorem].

We prove the boundedness property of the log crystalline cohomology for

the coecient I(X,DZ)/S :
Proposition 2.11.13. Let f : (X, D Z) Y be as in (2.10.2) (2) and

let : (X, D Z) (X, Z) be the forgetting log morphism along D. Then
log D D
Rfcrys (I(X,DZ)/S ) and R (I(X,DZ)/S ) are bounded.
log D
Proof. Let us rst prove that Rfcrys (I(X,DZ)/S ) is bounded. Let the no-
tations be as that in the proof of (2.3.11). By the same argument as [3, V
Theorème 3.2.4], we are reduced to proving the following claim: there ex-
ists a positive integer r such that, for any (U, T, ) (Y /S)log
crys , we have
i D
R fXU /T (I(X,DZ)/S ) = 0 for i > r. Again by the same argument as that
in the proof of [3, V Theorème 3.2.4, Proposition 3.2.5], we are reduced to
showing the above claim in the case where X and Y are suciently small
ane schemes. Hence we may assume that the log structure M (D Z) as-
sociated to D Z admits a chart of the form Nb M (D Z). If we take
=
a surjection 1 : OY [Na ] OX and if we set 2 := id : Nb Nb , we can
construct a commutative diagram
T

XU

(2.11.13.1) g

U T
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.

Using (2.11.11) and (2.11.13), we can prove the following:
Proposition 2.11.14. Let the notations and the assumptions as in (2.10.6).

(1) The natural morphism
(2.11.14.1) Lhcrys Rf(X,Z)crys

log log,Z
(Ecrys,c (O(X,DZ)/S ), PcD )

Rf (X ,Z )crys (Ecrys,c
log log,Z
(O(X ,D Z )/S ), PcD )
is an isomorphism.
(2) There exists a natural isomorphism

Lhcrys Rfcrys ) Rf crys (I(X
log D Dlog
(2.11.14.2) (I(X,DZ)/S ,D Z )/S )
which is compatible with the isomorphism (2.11.14.1).
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 Theorème 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
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 ).
Let k be a xed integer. Set

k ,hk
E1,c ((X, D Z)/S) =
(k )log
Rhk f(D(k ) ,Z| (k ) )/S
(O(D(k ) ,Z| (k ) )/S
Z crys (D/S; Z)) (k k),
D D
0 (k < k).
We shall also need the following spectral sequence later

(2.11.14.4) E1k ,hk = E1,c
k ,hk
((X, D Z)/S) =
(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)
Denition 2.11.15. We call the spectral sequence (2.11.14.3) the preweight

spectral sequence of (X, D Z)/(S, I, ) with respect to D for the log crys-
talline cohomology with compact support. If Z = , then we call (2.11.14.3)
the preweight spectral sequence of (X, D)/(S, I, ) for the log crystalline co-
homology with compact support.
Let PcD, be the ltration on Rh f(X,DZ)/S,c (O(X,DZ;Z)/S ) induced from

the spectral sequence (2.11.14.3). Since PcD, is the decreasing ltration, we
D
also consider the following increasing ltration P,c :
(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 ).
Proposition 2.11.16. Let the notations be as in (2.10.3). There exists a

canonical morphism of spectral sequences
(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. (2.11.16) immediately follows from the construction of (2.11.14.3).

Proposition 2.11.17. The boundary morphism dk,hk

1
k,hk
: E1,c ((X, D Z)
k+1,hk (k)log
/S) E1,c ((X, DZ)/S) is equal to the morphism induced by crys .
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)
Hence the boundary morphism dk,hk

1 is equal to the boundary morphism
Rf(D(k) ,Z| )/S (O(D (k) ,Z|D(k) )/S )[k]

D (k)
Rf(D(k+1) ,Z| )/S (O(D (k+1) ,Z|D(k+1) )/S )[k]

D (k+1)
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 .

Proposition 2.11.18. Let u be as in (2.10.3). Let u0 : S0 S0 be the

induced morphism by u. Let (Y, E W ) and (X, D Z) be smooth schemes
with relative SNCDs over S0 and S0 , respectively. Let
g
(Y, E W ) (X, D Z)

(2.11.18.1)
u
S0 0 S0
be a commutative diagram of log schemes such that the morphism g in-

duces morphisms g (k) : (E (k) , W |E (k) ) (D(k) , Z|D(k) ) of log schemes over
u0 : S0 S0 for all k N. Then the isomorphism in (2.11.11.1) and the
log
spectral sequence (2.11.14.3) are functorial with respect to gcrys .

The following is the K

unneth formula for the log crystalline cohomology
sheaf with compact support Rh f(X,DZ)/S,c (O(X,DZ;Z)/S ).
Theorem 2.11.19 (K unneth formula). Let the notations be as in those

in (2.10.14) (2). Then the following hold:
log log,Zi
(1) Set Hi.c := Rf(X i ,Zi )crys
(Ecrys,c (O(Xi ,Di Zi )/S ), PcDi ) (i = 1, 2, 3).

(2.11.19.1) H1,c L
OY /S H2,c H3,c .
(2) There exists a canonical isomorphism

(2.11.19.2)
log log
Rf(X 1 ,D1 Z1 )crys
D1
(I(X1 ,D1 Z1 )/S
) L D2
OY /S Rf(X2 ,D2 Z2 )crys (I(X2 ,D2 Z2 )/S )
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)).
(k) (i) (j)

Proof. (1): Let k be a nonnegative integer. Then D3 = i+j=k D1 S0 D2 .
Hence (1) follows from the usual K unneth formula.
(2): We have to check that the diagram (2.11.6.1) is compatible with log
Kunneth morphisms.

Let (Xj , Dj Zj )I (Xj , Dj Zj )I (j = 1, 2) be the data
(2.4.0.1) and (2.4.0.2) with j for (Xj , Dj Zj )/S0 /S. Here note that we
may assume that the index set I is independent of j = 1, 2 since I0 in 2.4 can
be assumed to be independent of j = 1, 2 by considering the product of two
index sets. Set X3 := X1 S X2 , D3 := (D1 S X2 ) (X1 S D2 ) and
Z3 := (Z1 S X2 )(X1 S Z2 ). Then we have a natural datum (X3 , D3

Z3 )I (X3 , D3 Z3 )I with 3 . Set j := (Xj ,Dj Zj ,Zj )/S
(j = 1, 2, 3). Then we have the following diagram
(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
as (2.10.11.1) (j = 1, 2). Let Fj (j = 1, 2, 3) be the crystal F in the proof of

(2.11.3) for the admissible immersion (Xj , Dj Zj ) (Xj , Dj Zj ).
Then we have a natural morphism
log log
q1crys (F1 ) O(X3 ,D3 Z3 )/S q2crys (F2 ) F3
and hence a natural morphism

log log
(2.11.19.4) Lq1crys (F1 ) L
O(X Lq2crys (F2 ) F3 .
3 ,D3 Z3 )/S
Using the adjunction formula, we have a natural morphism

(2.11.19.5) Lplog
1crys R1 (F1 ) O(X
L
Lplog
2crys R2 (F2 )
3 ,D3 Z3 )/S
log log
R3 (Lq1crys (F1 ) L
O(X Lq2crys (F2 )).
3 ,D3 Z3 )/S
Here, note that Rj (Fj ) (j = 1, 2) is bounded above by (2.3.12). Compos-

ing (2.11.19.5) with (2.11.19.4), we have a morphism
(2.11.19.6)

Lplog
1crys R1 (F1 ) O(X
L
Lplog
2crys R2 (F2 ) R3 (F3 ).
3 ,Z3 )/S
Now let us set

Llog
j := L(Xj ,Dj Zj )/S (Xj /S (log(Zj Dj ))),
Lj := L(Xj ,Zj )/S (Xj /S (log(Zj Dj )))

and
(D(k) /S (log Zj |D(k) )Z crys

(k) log
Lkj := ajcrys L(D(k) ,Zj | (k)log
(Dj /S; Zj ))
j (k) )/S j j
Dj
for k N, where
(k) log (k)
aj : (Dj , Zj |D(k) ) (Xj , Zj )
j
is a natural morphism. Then we have a morphism

(2.11.19.7)

Lplog log
1crys R1 (L1 ) O(X
L
Lplog log log
2crys R2 (L2 ) R3 (L3 )
3 ,Z3 )/S
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 .
We can check that the canonical morphism Fj Llog

j induces the morphism
(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)
and the morphism Lj Lj induces the morphism
(2.11.19.8) (2.11.19.9).
Hence we have the commutative diagram

Lplog
1crys R1 (F1 ) O(X
L
Lplog
2crys R2 (F2 )

3 ,Z3 )/S

(2.11.19.10)
R3 (F3 )
log
Lplog
1crys (L1 ) O(X3 ,Z3 )/S Lp2crys (L2 )

L3 .
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
is a natural morphism. Then we have a natural morphism

(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).
On the other hand, we have a natural morphism
(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
(2.11.19.13) (the left column of (2.11.19.11)).

Hence we obtain the diagram
(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 log log

Rf R (F ) L
OY /S Rf(X2 ,D2 Z2 )crysR(X2 ,D2 Z2 )/Scrys(F2 )
(X1 ,D1 Z1 )crys (X1 ,D1 Z1 )/Scrys 1
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
The left vertical morphism in (2.11.19.15) is the morphism in the statement

of (2) and the right vertical morphism is (the non-ltered version of) the
morphism (2.11.19.1). Therefore, to prove (2), it suces to prove that the
horizontal morphisms in (2.11.19.15) are isomorphisms. We can check this in
the same way as (2.11.14).
(3): (3) immediately follows from [3, V Corollary 4.1.4], (2.11.14) and (2).

2.12 Filtered Log de Rham-Witt Complex 175
2.12 Filtered Log de Rham-Witt Complex
Let be a perfect eld of characteristic p > 0. Let W (resp. Wn ) be the

Witt ring of (resp. the Witt ring of length n Z>0 ). Let K0 be the
fraction eld of W . Let (X, D) be a smooth scheme with an SNCD over .
In this section, as a special case, we prove that (Czar (O(X,D)/S ), P ) in the
case S = Spec(Wn ) is canonically isomorphic to the ltered log de Rham-
Witt complex (Wn X (log D), P ) := (Wn X (log D), {Pk Wn X (log D)}kZ )
constructed by Mokrane ([64, 1.4]).
Before proceeding on our way, we have to give the following remarks. Let

s = (Spec(), L) be a ne log scheme. Let g : Y := (Y , M ) s be a log
smooth morphism of Cartier type. Let Wn Y be the reverse log de Rham-
Witt complex dened in [46, (4.1)] and denoted by Wn Y in [loc. cit.]. Then,
in [46, (4.19)], we nd the following statements:
(1) There exists a canonical isomorphism

RuY /Wn (OY /Wn ) Wn Y (n Z>0 ).
(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))
by the Zariski analogue of [46, (4.19)]=[68, (7.19)]. In other words, we have

a canonical isomorphism

(2.12.0.2) log,Z
Czar (O(X,DZ)/Wn ) Wn X (log(D Z))
Let (Yn , En Wn ) be a lift of (X, D Z) over Wn . Then we have

Wn iX (log(D Z)) = Hi (Yn /Wn (log(En Wn ))). Set
(2.12.0.3) PkD Wn iX (log(D Z)) = Hi (PkEn Yn /Wn (log(En Wn ))).
Denition 2.12.1. We call the ltration P D := {PkD Wn iX (log(DZ))}kZ

the preweight ltration on Wn iX (log(D Z)) with respect to D.
We shall prove, in (2.12.4) below, that PkD Wn iX (log(D Z)) is independent

of the choice of the lift (Yn , En Wn ). If Z = , PkD Wn iX (log(D Z)) is the
preweight ltration dened in [64, (1.4.1)]. Here, as noted in [68, (4.3)], we
use the terminology preweight ltration instead of the terminology weight
ltration since Wn iX (log(D Z)) is a sheaf of torsion W -modules in X zar .
To prove a ltered version of (2.12.0.2), we need some lemmas (cf. [64, 1.2,
1.4.3]).
Let D := {D } (resp. Z := {Z } ) be a decomposition of D
(resp. Z) by smooth components of D (resp. Z). Set := {D , Z }, .

Let : (X, D Z) (Xn , Dn Zn ) be an admissible immersion over

Wn with respect to which induces an admissible immersion (X, D)

(Xn , Dn ) (resp. (X, Z) (Xn , Zn )) with respect to D (resp. Z ). Let

: (X, D Z) (Yn , En Wn ) be a lift of (X, D Z) over Wn such

that induces a lift (X, D) (Yn , En ) (resp. (X, Z) (Yn , Wn )). As-
sume that (Yn , En Wn ) and (Xn , Dn Zn ) are ane log schemes. Be-
cause (Xn , Dn Zn ) is log smooth over Wn , there exists a morphism of
log schemes f : (Yn , En Wn ) (Xn , Dn Zn ) over Wn such that f in-
duces morphisms (Yn , En ) (Xn , Dn ) and (Yn , Wn ) (Xn , Zn ) and
such that f = . Let DX (Xn ) be the PD-envelope of the closed immer-
sion : X Xn over (Spec(Wn ), pWn , [ ]). The morphism f also induces a
morphism f : (Yn , pOYn ) (DX (Xn ), Ker(OXn OX )ODX (Xn ) ) of PD-
schemes. Hence f induces a morphism f : ODX (Xn ) OXn Xn /Wn (log(Dn
Zn )) Yn /Wn (log(En Wn )) of complexes. By (2.2.17) (1), we have the
following exact sequence
Dn
(2.12.1.1) 0 ODX (Xn ) OXn Pk1 Xn /Wn (log(Dn Zn ))
ODX (Xn ) OXn PkDn Xn /Wn (log(Dn Zn ))
ODX (Xn ) OXn k

D (k) /W
(log Zn |Dn(k) ) Z zar
(k)
(Dn /Wn )(k) 0
n n
by using the Poincare residue isomorphism with respect to Dn ((2.2.21.3)).

(The compatibility of the Poincare residue isomorphism with the Frobenius
can be checked as in [68, (9.3) (1)].) Note that the derivative
d : PkDn Xn /Wn (log(Dn Zn )) PkDn +1

Xn /Wn (log(Dn Zn ))
extends to a derivative of ODX (Xn ) OXn PkDn Xn /Wn (log(Dn Zn ))

(cf. [50, 0 (3.1.4)], [54, (6.7)]).
Lemma 2.12.2. The long exact sequence associated to (2.12.1.1) is decom-
posed into the following short exact sequences:
Dn
(2.12.2.1) 0 Hq (ODX (Xn ) OXn Pk1 Xn /Wn (log(Dn Zn )))
Hq (ODX (Xn ) OXn PkDn Xn /Wn (log(Dn Zn )))
Hqk (ODX (Xn ) OXn D(k) /W (log Zn |Dn(k) )
n n
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
Hq (ODX (Xn ) OXn D(k) /W (log(Zn |Dn(k) )))

n n
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)
Hq (ODX (Xn ) OXn D(k) /W (log(Zn |Dn(k) ))) = Wn qD(k) (log(Z|D(k) ))

n n
and it is a quasi-coherent Wn (OX )-module on Xzar . On the other hand, let

((D(k) , Z|D(k) )/Wn )log
crys,et be the log crystalline site of (D
(k)
, Z|D(k) ) over Wn
with respect to the etale topology and let
u(D(k) ,Z| )/Wn ,et : ((D(k) ,

Z|D(k) )/Wn )log
crys,et Xet
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
Hq ((ODX (Xn ) OXn D(k) /W (log(Zn |Dn(k) )))et )

n n
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
(2.12.2.3) Hq (ODX (Xn ) OXn D(k) /W (log(Zn |Dn(k) )))

n n
Ru Hq ((ODX (Xn ) OXn D(k) /W (log(Zn |Dn(k) )))et )

=
n n

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
Hq ((ODX (Xn ) OXn D(k) /W (log(Zn |Dn(k) )))et )|Xzar

=
.
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
are Wn (OX )-linear with respect to the natural action of Wn (OX ) = H0

(ODX (Xn ) OXn Xn /Wn ). Then, by (2.12.2.4) and the induction on k, we see
that there exists the canonical isomorphism
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
=

for any etale morphism : X X. Hence the diagram (2.12.2.2)|Xzar
is exact for any : X X as above, and this implies the exactness of

(2.12.2.2) for any k, q Z. Hence the exactness of (2.12.2.1) for any k, q Z
is equivalent to the exactness of (2.12.2.2) for any k, q Z.
By the claim we have shown in the previous paragraph, we may work etale
locally to prove the lemma. Hence we may assume that Xn is the scheme

Spec(Wn [x1 , . . . , xd ]) and that Dn Xn is the closed immersion dened by
the ideal (x1 xs ) for some 0 s d. In this case, by the proof of [64, 1.2],
the morphism
Z(ODX (Xn ) OXn PkDn qXn /Wn (log(Dn Zn )))

Z(ODX (Xn ) OXn qk
(k) (log Zn |Dn(k) ))
Dn /Wn
is surjective on Xzar . Hence we obtain the exactness of (2.12.2.1).

By the Zariski analogue of [54, (6.4)] we have the following commutative
diagram:
(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
Lemma 2.12.3. Let k be a nonnegative integer. Then Hq (f ) induces an

isomorphism
Hq (ODX (Xn ) OXn PkDn Xn /Wn ( log(Dn Zn )))

Hq (PkEn Yn /Wn (log(En Wn ))).
Proof. We have two proofs.

First proof: The morphism f induces a morphism
ODX (Xn ) OXn PkDn Xn /Wn (log(Dn Zn )) PkEn Yn /Wn (log(En Wn )).
By using the Poincare residue isomorphisms with respect Dn , by (2.12.2) and

by induction on k, it suces to prove that f induces an isomorphism

Hqk (ODX (Xn ) OXn D(k) /W (log Zn |Dn(k) ) Z zar
(k)
(Dn /Wn )(k))
n n
Hqk (E (k) /W (log Wn |En(k) ) Z zar

(k)
(En /Wn )(k)).
n n
(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].
Second proof: (2.12.3) immediately follows from (2.5.4) (2).

The following is a generalization of the preweight ltration on Wn iX (log

D) ([64, (1.4.1)]) for an admissible closed immersion (X, DZ) (Xn , Dn
Zn ) over (Spec(Wn ), pWn , [ ]) even if Z = :
Corollary 2.12.4. (1) The preweight ltration on Wn X (log(D Z)) with

respect to D is well-dened. More generally, {H (ODX (Xn ) OXn PkDn Xn /Wn
(log(Dn Zn )))}kN induces the preweight ltration on Wn X (log(D Z)).
(2) Let i be a nonnegative integer. Then
(2.12.4.1) PkD Wn iX (log(D Z)) = Hi (PkD Czar

log,Z
(O(X,DZ)/Wn ))
(3) There exists the following canonical isomorphism

(2.12.4.2)
D

k Wn X (log(D Z)) Wn D (k) (log Z|D (k) ) Z zar (D/)(k)
ResD : grP (k)
which is compatible with the Frobenius endomorphisms.
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
(2.12.4.3) Wn () : (Wn (Y ), Wn (M )) (Yn , Mn )
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,
Wn () is well-known (e.g., [50, 0 (1.3.21), II (1.1.4)]). Let m be a local section

of M with image m Mn . Let zj (1 j n 1) be a unique local section
of 1 + pOY satisfying an equality j (m)
j
=m p zj . Let {
sj }n1
j=1 be a family
of local sections of OY satisfying the following equalities
j1
sp1
1 + p + + pj sj = zj .
(The existence of { sj }n1

j=1 has been proved in the proof of the log version of a
lemma of Dieudonne-Cartier ([68, (7.10)]) by using the argument in [61, VII
4].) Set sj := (
sj ) (1 j n 1) and s0 := 1. Then Wn () as a morphism
of log structures is, by denition, the following morphism:
(2.12.4.4) Wn () (Mn ) m
(n (m), (s0 , . . . , sn1 ))
M (1 + V Wn1 (OY )) = Wn (M ).
Here we denote Wn () (m) simply by m.

By the universality of the log PD-envelope, induces a natural morphism
DY (Y) : (DY (Y), MDY (Y) ) (DY (Y), MDY (Y) ).
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
(2.12.4.5) ODY (Yn ) OYn Yn /Wn Wn (Y )/(Wn ,Wn (L)),[ ]
of complexes of gn1 (Wn )-modules, where Wn (Y )/(Wn ,Wn (L)),[ ] is dened in

the proof of [46, (4.19)] and denoted by W n (Y )/(Wn ,Wn (L)),[ ]
in [loc. cit.]. By
[46, (4.9)] there exists a canonical morphism
(2.12.4.6) Wn (Y )/(Wn ,Wn (L)),[ ] H (ODY (Yn ) OYn Yn /Wn )(= Wn Y ).
Composing (2.12.4.5) with (2.12.4.6), we have a morphism
(2.12.4.7) ODY (Yn ) OYn Yn /Wn H (ODY (Yn ) OYn Yn /Wn ).
As usual ([50, II (1.1)]), the induced morphism by (2.12.4.7) in the derived

category is independent of the choice of Y and .
Lemma 2.12.5. Set := DY (Y) . Then the morphism (2.12.4.7) is equal to

the morphism (/p )n mod pn .
Proof. First consider the case = 0. Because OY is p-torsion-free, the fol-

lowing morphism s is well-dened:
(2.12.5.1) s : OY x
(s0 , s1 , . . . , sn1 , . . .) W (OY ),
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 ).
The morphism (2.12.4.6) for = 0 is dened by
(2.12.5.2) Wn (OY ) (t0 , t1 , . . . , tn1 )
n n1

tp0 + p
tp1 + + pn1
tpn1 H0 (ODY (Yn ) OYn Yn /Wn ),
where tj ODY (Yn ) (1 j n 1) is a lift of tj OY ([46, (4.9)]). Since

ni ni+1
(sp ) sp mod pni+1 OYn (s OYn , i {0, 1, . . . , n}), (2.12.4.7)
for = 0 is induced by the morphism x
n (x) (x OYn ).
Next, consider the case = 1. Because the image of (2.12.4.5) is contained
in the image of Wn (OY )Z Wn (M )gp in 1Wn (Y )/(Wn ,Wn (L)),[ ] , consider the
following composite morphism
(2.12.5.3) Wn (OY )Z Wn (M )gp 1Wn (Y )/(Wn ,Wn (L)),[ ]

(2.12.4.6)
H1 (ODY (Yn ) OYn Yn /Wn ).
The morphism (2.12.5.3) is dened by the morphisms (2.12.5.2) and d log m
d log m mod pn (m M ), where m MDY (Y) is a lift of m. Since

: MDY (Y) MDY (Y) is a lift of the Frobenius endomorphism, there
=m
exists a section a of ODY (Y) such that (m) p (1 + pa). Then
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
mod pn = p1 d log (m)

d log m mod pn
=
= pn d log n (m)
mod pn .
in H1 (ODY (Yn ) OYn Yn ). Furthermore the image of 1 (1 + V Wn1 (OY ))

by the morphism (2.12.5.3) is the zero.
Let m and {sj }n1
j=1 be local sections in (2.12.4.4). Then the image of d log m
n1 j pnj
by the morphism (2.12.4.7) is the class of d log m + d log(1 + j=1 p sj ),
where sj is a lift of sj in ODY (Yn ) . As in the argument above, the second
form is exact. Hence the morphism (2.12.4.7) for = 1 is induced from
(/p)n mod pn .
When 2, (2.12.5) follows from the denition of (2.12.4.6), from [46,

(4.9)] and from the calculation above.

Corollary 2.12.6. Let : (X, D Z) (X , D Z) be an admissible immer-
sion into a formally smooth scheme over Spf(W ) with a relative transversal
SNCD over Spf(W ). Set (Xn , Dn Zn ) := (X , D Z) W Wn . Assume
that there exists a lift : (X , D Z) (X , D Z) of the Frobenius en-
domorphism of (X1 , D1 Z1 ). Let DX (X ) be the PD-envelope of the closed

immersion X X over (Spf(W ), pW, [ ]). Set DX (Xn ) := DX (X ) W Wn .
Then the morphism
(2.12.6.1) ODX (Xn ) OXn Xn /Wn (log(Dn Zn )) Wn X (log(D Z))
dened in (2.12.4.7) induces an isomorphism

(2.12.6.2)
(ODX (Xn ) OXn Xn /Wn (log(Dn Zn )), P Dn ) (Wn X (log(D Z)), P D )
in D+ F(f 1 (Wn )).
Proof. The endomorphism induces an endomorphism DX (X ) : DX (X )

DX (X ). Set := DX (X ) . By the denition of Wn X (log(D Z)) ([46,
(4.1)]), we have Wn X (log(D Z)) = H (ODX (Xn ) OXn Xn /Wn (log(Dn
Zn ))). By (2.12.5), the morphism (2.12.6.1) is induced by n := (/p )mod pn .
By a calculation in [68, (8.1), (8.4)], n preserves the preweight ltrations
with respect to Dn :
n (ODX (X ) OX PkD X /W (log(D Z))/pn )

H (ODX (X ) OX PkD X /W (log(D Z))/pn ).
Hence, by using the Poincare residue isomorphism and by (2.12.2), it suces

to prove that n induces an isomorphism

(2.12.6.3) ODX (Xn ) OXn D(k) /W (log Zn |Dn(k) ) Z zar
(k)
(Dn /Wn )(k)
n n
H (ODX (Xn ) OXn D(k) /W (log Zn |Dn(k) ) Z zar

(k)
(Dn /Wn )(k)).
n n
This immediately follows from (2.2.16) (2) and [46, (4.19)]=[68, (7.19)].

Lemma 2.12.7. Let Y be a scheme over Fp with a closed subscheme E.

Let U be the complement of E in Y and j : U Y the open immersion.

Denote by (Y, E) a log scheme (Y, j (OU ) OY ). Let (Wn (Y ), Wn (E))
be a similar log scheme over Wn (Fp ) = Z/pn : (Wn (Y ), Wn (E)) :=
(Wn (Y ), Wn (j) (Wn (OU ) ) Wn (OY )). Assume that the natural morphism
OY j (OU ) is injective. Then (Wn (Y ), Wn (E)) is the canonical lift of
(Y, E) in the sense of [46, (3.1)].
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
Ker{Wn (j) (Wn (OU ) ) Wn (OY ) j (OU

)} = Ker(Wn (OY ) OY ).
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
Wn (j) (Wn (OU ) ) Wn (OY ) = [j (OU

) OY ] Ker(Wn (OY ) OY ).
This equality shows (2.12.7).

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
(2.12.7.1) 0 Wn X ( log D) Wn X Wn D(1)
is exact. Let R be the Cartier-Dieudonne-Raynaud algebra over ([52, I

(1.1)]). Set Rn := R/(V n R + dV n R). The second isomorphism

Rn L
R W X ( log D) Wn X ( log D)
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
(2.12.7.2) Wn X (Wn D(1) Z zar

(1)
(D/), d)
Wn D(2) Z zar
(2)
(D/)
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,
(2.12.7.3): i = 0 for i < 0.

(2.12.7.4): i (i N) are p-torsion-free, p-adically complete Zp -modules in
C+ (f1 (Zp )).
(2.12.7.5): (i ) { pi i | d pi+1 i+1 } (i N).
(2.12.7.6): There exists an Fp -linear isomorphism

C 1 : i1 Hi (1 ) (i N).
((19.7.6) is an isomorphism in [27, (4.2.1.3)].)

(2.12.7.7): A composite morphism (mod p) pi : i i i1 factors
through Ker(d : i1 i+1
1 ), and the following diagram is commutative:
mod p
i i1

1
pi C
mod p
i Hi (1 ).
Theorem 2.12.8 ([68, (6.2), (6.3), (6.4)]). (1) For a gauge : Z N

([11, 8.7 Denition]), let : Z N be the associated cogauge to dened by

(i) + i (i 0),
(i) :=
(0) (i 0).
Let (resp. ) be the largest complex of whose i-th degree is contained

in p(i) i (resp. p(i) i ). Then the morphism : induces a quasi-
isomorphism : .
(2) Assume that and are bounded above and that they consist of at
Zp -modules. Let M be an f1 (Zp ) = Zp -module. Then the morphism
(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).

Set
(2.12.8.3)
Zni := { i | d pn i+1 }, Bni := pn i + di1 , Wn i := Zni /Bni .
As usual (e.g., [68, 6]), we can dene the following operators:
F : Wn+1 i Wn i , V : Wn i Wn+1 i , d : Wn i Wn i+1 ,
p : Wn i Wn+1 i and : Wn+1 i Wn i .

We only remark that p is an injective morphism induced by p(i1) : i
i (note that (i1) is positive if i = 0) and that is the following composite
surjective morphism ([68, (6.5)]):
(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
Here the isomorphism pi in (2.12.8.4) is given by (2.12.8.2). As usual, one

can endow Wn i with a natural Wn (OX )-module structure, and the following
formulas hold:
d2 = 0, F dV = d, F V = V F = p ,
F p = pF, V p = pV, dp = pd, p = p = p.
Set W = lim Wn . Then W is a complex of sheaves of W (OX )-modules

and torsion-free W -modules in C+ (f1 (W )). In fact, W (resp. Wn ) is
naturally an R-module (resp. Rn -module). Set

Ker(Wi Wr i ) (r > 0),
Fil W :=
r i
Wi (r 0).
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)]):
Proposition 2.12.9 ([68, 6, (A), (B), (C)]). The following formulas

hold:
(1) Filr Wi = V r Wi + dV r Wi1 (i Z, r Z0 ).
(2) d1 (pn W ) = F n W .

(3) Rn L R W = Wn (n Z>0 ).
Proof. Here we only remark that (3) is a formal consequence of (1) and (2)
(see [52, II (1.2)]).

Let us come back to the general case.

Recall the ideal sheaf I(X,DZ)/S
D
of O(X,DZ)/S in 2.11. Set
(2.12.9.1) Wn iX (log(Z D)) := Ri u(X,DZ)/Wn (I(X,DZ)/W

D
n
) (i N).

Zariski locally on X, we have an isomorphism Wn iX (log(Z D)) Wn i .
It is a routine work to check that the family {Wn X (log(Z D))}nZ>0 of
complexes has the operators F , V , d, p and (cf. [46, (4.1), (4.2)]) (especially
one can check that p and are well-dened by considering embedding systems
of (X, D Z) over W ); in fact, W X (log(Z D)) is naturally an R-module.
Then the following holds:
Proposition 2.12.10. The complex Wn X (log(Z D)) (n Z>0 ) is quasi-
isomorphic to the single complex of the following double complex:

d d
(0) (1)
Wn 2X (log Z) Wn 2D(1) (log Z|D(1) ) Z zar (D/)

(2.12.10.1) d d
(0) (1)

d d
(0) (1)

d
(1) (2) (2)

Wn 2D(2) (log Z|D(2) ) Z zar (D/)

d
(1) (2) (2)

Wn 1D(2) (log Z|D(2) ) Z zar (D/)

d
(1) (2) (2)

Wn 0D(2) (log Z|D(2) ) Z zar (D/) .
Proof. The proof is the same as that of [64, Lemma 3.15.1]: by using (2.12.9)
(3) and Ekedahls Nakayama duality, we can reduce the exactness to that for
the case n = 1, and in this case, we obtain the exactness by the argument of
[27, (4.2.2) (a), (c)] (cf. (2.11.5.1)).

The complex (2.12.10.1) has a stupid ltration k (k Z) with respect to
the columns and we set PcD,k := k . Hence we obtain a ltered complex in
C+ F(f 1 (Wn )), K+ F(f 1 (Wn )) and D+ F(f 1 (Wn )), and we denote it by
(2.12.10.2) (Wn X (log(Z D)), PcD ).
The following is the main result in this section:
Theorem 2.12.11 (Comparison theorem). (1) In D+ F(f 1 (Wn )), there

exists the following canonical isomorphism:

(2.12.11.1) log,Z
(Czar (O(X,DZ)/Wn ), P D ) (Wn X (log(D Z)), P D ).
The isomorphisms (2.12.11.1) for ns are compatible with two projections of

both hands of (2.12.11.1).
(2) In D+ F(f 1 (Wn )), there exists the following canonical isomorphism:

(2.12.11.2) log
(Ezar,c (O(X,DZ)/Wn ), PcD ) (Wn X (log(Z D)), PcD ).
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))
induced by the isomorphism (2.12.0.2). The isomorphisms (2.12.11.2) for ns

are compatible with two projections of both hands of (2.12.11.2). The iso-
morphism (2.12.11.2) is functorial for the commutative diagram (2.11.18.1)
for the case S0 = Spec(), S = Spec(Wn ), S0 = Spec( ) and S =
Spec(Wn ( )), where is a perfect eld of characteristic p.
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 )).
(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
g : (Wn iX2 (log(D2 Z2 )), P D2 ) (Wn iX1 (log(D1 Z1 )), P D1 ).
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 )
(Wn+1 X (log(D Z )), P D )

(Wn X (log(D Z )), P D ).
Applying the direct image Rzar , we obtain the compatibility with two
projections.
(2): The morphism (2.12.11.4) induces a morphism
(2.12.11.5)
ODX (X,n ) OX,n X,n /Wn (log(Z,n D,n )) Wn X (log(Z D )).
By (2.2.16) (2), DX (X ) X D(k) is the PD-envelope of the locally closed

(k) (k) (k) (k)
immersion D D . Set DD(k) (D,n ) := (DX (X ) X D ) W Wn .

By (2.11.9) and (2.12.10), we have the following commutative diagram:

Rzar (ODX (X,n ) OX,n X,n /Wn (log(Z,n D,n )))

(2.12.11.6)

Rzar (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)
By the cohomological descent, the lower vertical morphism in (2.12.11.6) is

equal to
{Ru(X,Z)/Wn (O(X,Z)/Wn ) Z zar

(0)
(D/) }
{Wn X (log Z) Z zar

(0)
(D/) }.
By [46, (4.19)]=[68, (7.19)], this is an isomorphism. The claim as to the
compatibility of the ltrations is obvious by the denitions. As usual (cf. [50,
II (1.1)], 2.5), we see that the lower vertical morphism in (2.12.11.6) is
independent of the choice of the open covering of X, that of the lift of each
open subscheme and that of the lift of the Frobenius.
The compatibility with respect to two projections follows from the follow-
ing commutative diagram:

Ru(D() ,Z| )/Wn+1 (O(D () ,Z|D() )/Wn+1 ) Wn+1 D() (log Z|D() )
D ()

proj.

Ru(D() ,Z| )/Wn (O(D () ,Z|D() )/Wn ) Wn D() (log Z|D() ),
D ()
which we can prove in the same way as [46, (4.19)]=[68, (7.19)].

The functoriality claimed in (2) is obvious by the proof above.

Let i be a nonnegative integer. We conclude this section by constructing
the preweight spectral sequences of Wn iX (log(DZ)) and Wn iX (log(ZD))
with respect to D and describing the boundary morphisms between the E1 -
terms of the spectral sequences.
The following is a generalization of [68, (5.7.1;n)]:
Proposition 2.12.12. Let i be a nonnegative integer. Then there exists the
(2.12.12.1) E1k,h+k = H hi (D(k) , Wn ik

D (k)
(k)
(log Z|D(k) )Z zar (D/))(k)
= H hi (X, Wn iX (log(D Z))).

The spectral sequences (2.12.12.1) for ns are compatible with the projections.
Proof. (2.12.12.1) immediately follows from (2.12.4.2). The compatibility

with the projection immediately follows from the same proof as that of [68,
(8.4) (2)].

Next we describe the boundary morphism between the E1 -terms of the

spectral sequence (2.12.12.1).
Let the notations be before (2.8.5). Consider the following exact sequence
(2.12.12.2) 0 Wn iD (log Z|Dj ) Wn iD (log(Z D ))

j j
D
Res
j (Wn i1
D (log Z|D ))(1) 0.
We have the boundary morphism

log
(2.12.12.3) Gj : j D (log ZD )(1) Wn D (log Z|Dj )[1].
Wn i1 i
j
of (2.12.12.2). Here we have used the Convention (4). As in (2.8.4.5), the

morphism (2.12.12.3) induces the following morphism
(2.12.12.4)
log
(1)j Gj : H hi (D , Wn ik (log Z|D ) Z ,zar (D/))(k)
H hi+1 (Dj , Wn i+1k (log Z|Dj ) Z log

j ,zar
(D/))((k 1)).
Denition 2.12.13. We call the morphism (2.12.12.4) the Gysin mor-

phism in log Hodge-Witt cohomologies associated to the closed immersion

(D , Z|D ) (Dj , Z|Dj ).
k1 j j
Proposition 2.12.14. Set G := {0 ,...,k1 | i =j (i=j)} j=0 (1) G .
Then the boundary morphism dk,h+k
1 : E1k,h+k E1k+1,h+k of
(2.12.12.1) is equal to G.
Proof. The proof is the same as that of (2.8.5).

Proposition 2.12.15. Let i be a nonnegative integer. Then there exists the

(2.12.15.1) E1k,hk = H hik (D(k) , Wn iD(k) (log Z|D(k) ) Z zar

(k)
(D/))
= H hi (X, Wn iX (log(Z D))).
The spectral sequences (2.12.15.1) for ns are compatible with the projections.
The boundary morphism
(2.12.15.2) dk,hk
1 : H hik (D(k) , Wn iD(k) (log Z|D(k) ) Z zar
(k)
(D/))
H hik (D(k+1) , Wn iD(k+1) (log Z|D(k+1) ) Z zar

(k+1)
(D/))
is equal to (k) .
Proof. (2.12.15) immediately follows from (2.12.10.1). (The compatibility

with the projection is easy to check.)

Remark 2.12.16. If X is proper over and if Z = , the rst-named author

has proved the E2 -degeneration of the following spectral sequences modulo
torsion ([68, (5.9)]):
E1k,h+k = H hi (D(k) , W ik
D (k)
Z zar
(k)
(D/))(k)
= H hi (X, W iX (log D)),
E1k,hk = H hik (D(k) , W iD(k) Z zar

(k)
(D/))
= H hi (X, W iX ( log D)).
2.13 Filtered Convergent F -isocrystal
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.
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
(1) Base change theorem
Theorem 2.13.1. Let k, h be two nonnegative integers. Then there exists a

convergent F -isocrystal Ekh on S/V such that
DT 1 log,ZT1
(Ekh )T = Rh f(XT1 ,ZT1 )/T (Pk Ecrys (O(XT1 ,DT1 ZT1 )/T ))K
for any p-adic enlargement T of S/V . In particular, there exists a convergent

F -isocrystal Rh f (O(X,DZ)/K ) on S/V such that
Rh f (O(X,DZ)/K )T = Rh f(XT1 ,DT1 ZT1 )/T (O(XT1 ,DT1 ZT1 )/T )K
for any p-adic enlargement T of S/V .
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 ))
shows that the Frobenius action FS (Ekh ) Ekh is an isomorphism. Thus

Ekh prolongs to a convergent F -isocrystal as in [74, (3.7)].

Remark 2.13.2. The existence of the convergent F -isocrystal Rh f (O(X,DZ)

/K ) is a special case of [76, Theorem 4] and [29, 2 (e), (f)]. This existence
also follows from the log base change theorem ([54, (6.10)]), the bijectivity of
the Frobenius [46, (2.24)], and the same proof of [74, (3.1), (3.7)].
Corollary 2.13.3. The weight ltration on Rh f (O(X,DZ)/K ) with respect

to D is a convergent F -isocrystal on S/V . That is, the image PkD Rh f (O(X,D
Z)/K ) := Im(Ek R f (O(X,DZ)/K )) (k N) is a convergent F -isocrystal.
h h
Proof. The category of the convergent isocrystals on S/V is abelian ([74,

(2.10)]); hence the image Im(Ekh Rh f (O(X,DZ)/K )) is a convergent
isocrystal.
Now, by [74, (2.18), (2.21)], we have only to prove that PkD Rh f (O(X,DZ)
/K gives a p-adically convergent F -isocrystal for the case V = W . The exis-
)
tence of the Frobenius on PkD Rh f (O(X,DZ)/K ) is clear by the functoriality
which will be stated in (2.13.9) below soon. Because the Frobenius F s on the
E1 -terms of (2.9.6.3)V K for a p-adic formal V -scheme T are isomorphisms,
the Frobenius on PkD Rh f (O(X,DZ)/K ) is also an isomorphism. This com-
pletes the proof of (2.13.3).

Remark 2.13.4. We can also develop theory of weight ltrations by virtue of

theory of log convergent topoi ([82]). See [73] for details.
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
Rh f(XT1 ,DT1 ZT1 )/T (O(XT1 ,DT1 ZT1 )/T )K )

is a at OT V K-module.
Proof. (2.13.5) follows from [74, (2.9)] and (2.13.3).

Remark 2.13.6. The atness of Rh f(XT1 ,DT1 ZT1 )/T (O(XT1 ,DT1 ZT1 )/T )K is
a special case of [76, Lemma 36] and [29, 2 (e), (f)].
(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
Rh f (O(X3 ,D3 Z3 )/K )
of convergent F -isocrystals on S/V which is compatible with the weight l-

trations with respect to D1 , D2 and D3 .
Proof. The existence of the canonical isomorphism in (2.13.7.1) as weight-
ltered convergent F -isocrystals immediately follows from (2.10.15).

(3) Log crystalline cohomology sheaf with compact support
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
for any p-adic enlargement T of S/V . In particular, there exists a convergent

F -isocrystal Rh f,c (O(X,DZ;Z)/K ) on S/V such that
Rh f,c (O(X,DZ;Z)/K )T = Rh f(XT1 ,DT1 ZT1 )/T ,c (O(XT1 ,DT1 ZT1 ;ZT1 )/T )K
for any p-adic enlargement T of S/V .

2.13 Filtered Convergent F -isocrystal 195
(2) The OT V K-module

DT 1
Pk Rh f(XT1 ,DT1 ZT1 )/T ,c (O(XT1 ,DT1 ZT1 ;ZT1 )/T )K
is at for any p-adic enlargement T of S/V .

(3) Let (Xj , Dj Zj ) (j = 1, 2) be as in (2.13.7). Then there exists the
following canonical isomorphism

Ri f,c (O(X1 ,D1 Z1 ;Z1 )/K ) OS/K Rj f,c (O(X2 ,D2 Z2 ;Z2 )/K )
i+j=h

Rh f,c (O(X3 ,D3 Z3 ;Z3 )/K )
of convergent F -isocrystals on S/V which is compatible with the weight l-

trations with respect to D1 , D2 and D3 .
(4) Functoriality
Theorem 2.13.9. Let f : (X, D Z) S be as in the beginning of this

section. Let k, h be nonnegative integers. Then the following hold:
(1) The convergent F -isocrystal PkD Rh f (O(X,DZ)/K ) (k Z) is
functorial.
(2) The convergent F -isocrystal PkD Rh fc (O(X,DZ;Z)/K ) (k Z) is func-
torial with respect to the obvious analogue of the morphism in (2.11.18).
Proof. (1) and (2) immediately follow from (2.9.1) and (2.11.18), respectively.

(5) Gysin morphisms
Proposition 2.13.10. The Gysin morphism (2.8.4.5) induces the following

morphism

(2.13.10.1) (1)j Gj : Rhk f (O(D ,Z|D )/K Z log (D/K; Z))(k)
Rhk+2 f (O(Dj ,Z|D )/K Z log

j
(D/K; Z))((k 1)).
j
of convergent F -isocrystals on S/V . Here Rh f (O(D ,Z |D )/K Z log (D/K;

Z)) is a convergent F -isocrystal on S/V such that Rh f (O(D ,Z |D )/K Z
log (D/K; Z))T = Rh fXT1 /T (O((D )T1 ,(Z )T1 |(D )T /T ) Z crys
log
(DT1 /T ;
1
ZT1 )) for a p-adic enlargement T of S/V .
Proof. (2.13.10) immediately follows from (2.8.4).

Using (1), (3), (4) and (5), we obtain the following:

Theorem 2.13.11. Let Rh f (O(D(k) ,Z| (k) )/K Z (k)log (D/K; Z)) be a con-
D
vergent F -isocrystal on S/V such that
Rh f (OD(k) /K Z (k)log (D/K; Z))T =
Rh fXT1 /T (O(D(k) ,Z| Z crys

(k)log
(DT1 /T ; ZT1 ))
T1 (k) )/T
DT
1
for any p-adic enlargement T of S/V . Then the following hold:

(1) There exist the following weight spectral sequences of convergent
F -isocrystals
(2.13.11.1) E1k,h+k ((X, D Z)/K)

= Rhk f (O(D(k) ,Z| )/K Z (k)log (D/K; Z))(k)
D (k)
= 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 ).
The boundary morphism of (2.13.11.1) (resp. (2.13.11.2)) is given by G

(resp. (k) ) induced by the morphism in (2.8.5) (resp. (2.11.1.3)).
(2) The spectral sequences (2.13.11.1) and (2.13.11.2) are functorial with
respect to the obvious analogue of the morphism in (2.9.0.1) and (2.11.18),
respectively.
Proof. (1): (1) follows from (2.9.6.2) and (2.11.14.3).
(2): Obvious.

Denition 2.13.12. In the case Z = , we call (2.13.11.1) (resp. (2.13.11.2))
the p-adic weight spectral sequence of Rh f (O(X,D)/K ) (resp. Rh f,c
(O(X,D)/K )).
2.14 Specialization Argument in Log Crystalline

Cohomology
Let us recall a specialization argument of Deligne-Illusie in log crystalline

cohomologies (cf. [49, (3.10)], [68, 3]) for later sections 2.15 and 2.18.
2.14 Specialization Argument in Log Crystalline Cohomology 197

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 )
for any morphism g : T T of adic ne formal log PD-schemes.

Set T1 := SpecT (OT /a), and set T1 := SpecT (OT /a) for an open log
formal subscheme T of T . Let f : X T1 be a proper log smooth integral
morphism. By the niteness of log crystalline cohomologies (cf. [11, 7.24
Theorem]), there exists a non-empty open log formal subscheme T of T such
that
(2.14.0.3) Tor O
r
T (Rh f
XT /T (OXT /T ), E|T ) = 0 (r Z>0 )
1 1
for any a-torsion-free OT -module E and for any h Z. Assume furthermore

that the log structures on X, T are fs. Let IX/T be the ideal sheaf on OX/T

dened in [85, 5]. (In [85, 5], IX/T is dened under the condition that T
is equal to Spec Wm () ( is a perfect eld of characteristic p > 0), the log
structure on T is associated to the morphism N 1
b Wm () for some
b and that the morphism f is universally saturated. However, for the den-
ition of IX/T , we do not need these assumptions.) Set Rh fX/T ,c (OX/T ) :=
Rh fX/T (IX/T ). One can see that IX/S is a crystal on the restricted log crys-
talline site (X/T )log
Rcrys as in [85, (5.3)] and that, for any log smooth integral
lift X T of f , the sheaf (IX/T )X is at over OT by [85, (2.22)]. By using
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
Proposition 2.14.1. Let T = (T, aOT , ) be as above. Let g : T T be

a morphism from an adic ne formal log scheme into an open log formal
subscheme of T . If T is small enough, then the following hold:
(1) The canonical morphism
g Rh fXT /T (OXT /T )Rh fXT /T (OXT /T )

1 1 1 1
is an isomorphism of OT -modules.
g Rh fXT /T ,c (OXT /T )Rh fXT /T ,c (OXT /T )

1 1 1 1
is an isomorphism of OT -modules.
Proof. We may assume that (2.14.0.2), (2.14.0.3) and (2.14.0.4) hold.

(1): As in [68, (3.2)], we immediately obtain (1) using the existence of a
strictly perfect complex of OT -modules representing R (XT1 /T , OXT /T )
1
(cf. [11, 7.14 Denition, 7.24.3 Theorem]), using (2.14.0.2) and (2.14.0.3), and
using the log base change theorem ([54, (6.10)], cf. [74, (3.3)]).
(2): By the facts described before (2.14.0.4), the same proof as that of (1)
works.

We will use the following proposition in 2.18 below.
Proposition 2.14.2. Let T be an adic formal scheme. Let g : T T be

a morphism from an adic scheme into an open formal subscheme of T . Let
f : (X, D Z) T1 be a proper smooth scheme with a relative SNCD over
T1 . If T is small enough, then the following hold:
DT
g Pk 1
Rh f(X,DZ)T /T (O(X,DZ)T /T )
1 1
DT
Pk 1
Rh f(X,DZ)T /T (O(X,DZ)T /T )
1 1
is an isomorphism.
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
Proof. By (2.9.6.2), there exist the following two spectral sequences
(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
By (2.9.1) (2), there exists a canonical morphism
g 1 ((2.14.2.1)) g1 (OT ) OT (2.14.2.2).
Then, by (2.14.0.3), there exists a non-empty open formal subscheme T such

that
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 )
as in the proof of (2.14.1) (1), and therefore the morphism in (1) is an

isomorphism.
The proof of (2) is the same as that of (1).

2.15 The E2 -degeneration of the p-adic Weight Spectral

Sequence of an Open Smooth Variety
Let be a perfect eld of characteristic p > 0. Let W be the Witt ring of

. Let K0 be the fraction eld of W . In [68, (5.2)] we have proved the E2 -
degenerations modulo torsion of the weight spectral sequences (2.9.6.2) and
(2.11.14.3) when Z = and S = Spf(W ). To prove the degenerations, we
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 ).
Lemma 2.15.1. If A1 is reduced, then the morphism : A W (A1 ) is

injective.
Proof. Let Fn (A1 ) be the restriction of scalars of A1 by the n-th power

of the Frobenius endomorphism of A1 . By the assumption, the morphism
F n : A1 Fn (A1 ) is injective. (2.15.1) follows from the following commu-
tative diagram in [50, 0 (1.3.22)]:
Fn
A1 Fn (A1 )

pn V n (n N)
n
gr n+1
pn A/pn+1 A V n W (A1 )/V n+1 W (A1 ).

Lemma 2.15.2. (1) Let B be a commutative ring whose Jacobson radical

rad(B) is the zero. Let M(B)
be the set of the maximal ideals of B. Then
the morphism W (B) W (B/m) is injective.
mM(B)
(2) Let C be a commutative ring with unit element and let D be a smooth
C-algebra. If rad(C) = 0, then rad(D) = 0.
2.15 The E2 -degeneration of the p-adic Weight Spectral Sequence of a Variety 201

Proof. (1): By the assumption, the natural morphism B B/m is
mM(B)

injective. Thus W (B) W ( B/m) = W (B/m) is injective.
mM(B) mM(B)

(2): Let {fi }i be a family of elements of D such that Spec(D) = i Spec
(Dfi
). Then the natural morphism D i Dfi is injective since D
Dm is injective. Thus the problem is local; we may assume that there
mM(D)

exists a nite etale morphism C[X1 , . . . , Xm ] D. Let ( 0)C and ( 0)D be
nilpotent radicals of C and D, respectively. Since ( 0)C rad(C) = 0,
the
( 0)C = 0. Hence C isa Jacobson ring and D is also by [13, V 3, n 4,
Theorem 3]. Therefore ( 0)D = rad(D). Since C[X1 , . . . , Xm ] is reduced, D
is also by [41, I Proposition 9.2]. Hence ( 0)D = 0.

Corollary 2.15.3. Let be a perfect eld of characteristic p > 0. Let A

be a p-adically complete and separated formally smooth algebra over W ( )
witha lift of the Frobenius morphism of A1 . Then the morphism A
W (A1 /m) is injective.
mM(A1 )
Proof. (2.15.3) follows from (2.15.1) and (2.15.2).

Theorem 2.15.4 ([68, (5.2)]). If Z = and S = Spf(W ), then (2.9.6.2)

and (2.11.14.3) degenerate at E2 modulo torsion.
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 )
for any morphism g : T T of p-adic ne log PD-schemes and for any

h, k Z, r Z>0 . Then we have
g Erk,h+k ((XT1 , DT1 )/T ) = Erk,h+k ((XT1 , DT1 )/T )
for any morphism g : T T of p-adic ne log PD-schemes and for any

h, k Z, r Z>0 . Indeed, for r = 1, it is nothing but (2.14.1) (1); for general
r, it follows from the functoriality of the spectral sequence (2.9.6.2) and
induction. Hence, to prove the theorem for the spectral sequence (2.9.6.2),
we have to only to prove that the morphism
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 )
is zero for any r 2. Let us express

Erk,h+k ((XT1 , DT1 )/T ) = OTn N ,

Erk+r,h+kr+1 ((XT1 , DT1 )/T ) = OTn N ,
where N , N are direct sums of modules of type OT /pe (e Z>0 ). Then we
have

dk,h+k
r ((XT1 , DT1 )/T ) HomOT (OTn N , OTn N )

= HomOT (OTn , OTn ) N,
where N is a direct sum of modules of type (T, OT )/pe (e Z>0 ). Then,

for any closed point t of T1 , we have
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 ).)

2.16 The Filtered Log Berthelot-Ogus Isomorphism
In this section we prove a ltered version of Berthelot-Ogus isomorphism.

Because the proof of this isomorphism is almost the same as that in [12] and
[74], we give only the sketch of the proof.
Proposition 2.16.1. Let S be a scheme of characteristic p > 0 and let

S0 S be a nilpotent immersion. Let S T be a PD-closed immersion
into a formal scheme with p-adic topology such that OT is p-torsion-free. Let
f : (X, D Z) S and f : (X , D Z ) S be smooth schemes with rel-
ative transversal SNCDs. Assume that X, X , S and T are noetherian. Set
(X0 , D0 Z0 ) := (X, D Z) S S0 and (X0 , D0 Z0 ) := (X , D Z ) S S0 .
Let g : (X0 , D0 Z0 ) (X0 , D0 Z0 ) be a morphism of log schemes over S0
which induces morphisms (X0 , D0 ) (X0 , D0 ) and (X0 , Z0 ) (X0 , Z0 ).
(1) There exists a canonical ltered morphism
2.16 The Filtered Log Berthelot-Ogus Isomorphism 203

(2.16.1.1) g : (Rf(X

,D Z )/T (O(X ,D Z )/T ) Z Q, P
L D
)
(Rf(X,DZ)/T (O(X,DZ)/T ) L
Z Q, P ),
D
which is compatible with compositions. If g has a lift g : (X, D Z)

(X , D Z ), then g = gcrys
log
.
(2) Assume that g induces a morphism g (k) : (D0 , Z0 |D(k) ) (D 0 , Z0
(k) (k)
0
|D (k) ) for all k N. Then there exists a canonical ltered morphism
0

(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
which is compatible with compositions. If g has a lift g : (X, D Z)

(X , D Z ), then gc = gcrys
log
.
Proof. (1): The relative Frobenius F(X,DZ)/S : (X, D Z) (X (p) , D(p)

Z (p) ) over S induces an isomorphism
(p)
PkD Rf(X (p) ,D(p) Z (p) )/T (O(X (p) ,D(p) Z (p) )/T ) L
Z Q

PkD Rf(X,DZ)/T (O(X,DZ)/T ) L
Z Q (k Z)
by (2.9.6.3) and (2.10.2.1) because the relative Frobenius induces an isomor-

phism of the classical iso-crystalline cohomology of a smooth scheme over S
([12, (1.3)]). Hence the same proof as that in [12, (2.1)] shows that we have
the morphism (2.16.1.1).
(2): The proof for (2.16.1.2) is the same as that for (2.16.1.1) by using
(2.11.14.4) instead of (2.9.6.3) and using (2.11.18).

Corollary 2.16.2. If (X0 , D0 Z0 ) = (X0 , D0 Z0 ), then
(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
Proof. Obvious (cf. [12, (2.2)]).

Theorem 2.16.3 (Filtered log Berthelot-Ogus isomorphism). Let V

be a complete discrete valuation ring of mixed characteristics with perfect
residue eld . Let p be the characteristic of . Set K := Frac(V ). Let S be
a p-adic formal V -scheme in the sense of [74, 1]. Let (X, D Z) S be a
proper formally smooth scheme with a relative transversal SNCD over S. Let
T be an enlargement of S with morphism z : T0 := (SpecT (OT /p))red S.

Set T1 := SpecT (OT /p). Let f0 : (X0 , D0 Z0 ) := (X, D Z) S,z T0 T0
be the base change of f : (X, D Z) S. Then the following hold:
(1) If there exists a log smooth lift f1 : (X1 , D1 Z1 ) T1 of f0 , then
there exist the following canonical ltered isomorphisms
T : (Rh f(X1 ,D1 Z1 )/T (O(X1 ,D1 Z1 )/T )K , P D1 )

(Rh f (O(X,DZ)/K )T , P D ),
T,c : (Rh f(X1 ,D1 Z1 )/T ,c (O(X1 ,D1 Z1 ;Z1 )/T )K , PcD1 )

(Rh f,c (O(X,DZ;Z)/K )T , PcD ).
(2) If there exists a log smooth lift f : (X , D Z) T of f0 , then there
exist the following canonical ltered isomorphisms

log
crys,T : (Rh f (X /T (log(D Z)))K , P D ) (Rh f (O(X,DZ)/K )T , P D ),

log
crys,T,c : (Rh f (X /T (log(ZD)))K , PcD ) (Rh f,c (O(X,DZ;Z)/K )T , PcD ).
Proof. The proof is the same as that of [74, (3.8)].

Remark 2.16.4. Let V , and p be as in (2.16.3). Then V /p is a -algebra by

[79, II Proposition 8].
(1) Let (X , D Z) a proper smooth scheme over Spec(V ) with an (S)NCD.
Set UK := XK \(DK ZK ). Then, by (2.16.3) and (2.16.2) and the base change
theorem of the log crystalline cohomology ([54, (6.10)]), there are canonical
isomorphisms:
(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 The E2 -degeneration of the p-adic Weight Spectral

Sequence of a Family of Open Smooth Varieties

perfect residue eld of characteristic p > 0. Let B be a topologically nitely
generated ring over V . For a V -module M , MK denotes the tensor product
M V K. In particular, BK = B V K. Let m be a maximal ideal of BK . By
the proof of [84, (4.5)], BK /m is a nite extension of K. Set K := BK /m.
Let C be the image of B in BK /m = K . Let V be the integer ring of K .
Then the following is well-known (cf. [74, the proof of (4.2)]):
Lemma 2.17.1. V C V .
Proof. The inclusion V C is obvious. Let be a uniformizer of V . Let v
be a normalized valuation of V . Let e be the ramication index of V /V . By
the denition of B, there exists a surjection V {x1 , . . . , xr } B. It suces
to show that the image yi (1 i r) of xi in K belongs to V . If not,
v(yi ) < 0 for some i. Set

n/(e+1) (n N, e + 1|n),
an =
0 (n N, e + 1 n).

Then the image of an element n=0 an xni V {x1 , . . . , xr } in K does not
converge in K . This is a contradiction.

We keep the notations in 2.4 except that S is a p-adic formal V -scheme
in the sense of [74, 1] and that X is a proper smooth scheme with a relative
SNCD D over S1 := SpecS (OS /p). The main result in this section is the
following:
Theorem 2.17.2 (E2 -degeneration). Assume that S is a p-adic formal

V -scheme and that X is a proper smooth scheme over S1 . Then (2.9.6.2)V K
and (2.11.14.3) V K degenerate at E2 in the case Z = and S0 = S1 .
Proof. (Compare the following proof with [20, (5.5)].)
We rst prove the theorem for (2.9.6.2)V K for the case Z = and
S0 = S1 . We may assume that S is a p-adic ane at formal scheme Spf(B)
over Spf(V ). Consider the following boundary morphism:
(2.17.2.1)
dk,h+k
r : Erk,h+k ((X, D)/S)K Erk+r,h+kr+1 ((X, D)/S)K (r 2).
We prove that dk,h+k

r = 0 (r 2).
Case I: First we consider a case where B is a topologically nitely gen-
erated ring over V such that BK is an artinian local ring. Let m be the
maximal ideal of BK . Then m is nilpotent. Set K := BK /m. Consider the
following ideal of B: I := Ker(B BK /m). Then C = B/I, CK = K and

V C V ((2.17.1)). Let : Spf(C) Spf(B) be the nilpotent closed im-
mersion. Since the characteristic of K is 0, the morphism Spec(CK )
Spec(K) is smooth and hence there exists a section sK : Spec(BK )
Spec(CK ) of the nilpotent closed immersion Spec(CK ) Spec(BK ). By
[74, (1.17)], there exists a nite modication : Spf(B ) Spf(B), a nilpo-

tent closed immersion : Spf(C) Spf(B ) with = and a morphism
s : Spf(B ) Spf(C) such that s induces sK and that s = id. Set
S := Spf(B ). Because the boundary morphisms {dk,h+k
1 } are summations
of Gysin morphisms (with signs) ((2.8.5)), the E2 -terms of (2.9.6.2) V K
are convergent F -isocrystals by [74, (3.7), (3.13), (2.10)]. Hence we have
E2k,h+k ((X, D)/S)K = E2k,h+k ((XS1 , DS1 )/S )K since BK
= BK . Let

{dr } (r 1) be the boundary morphism of (2.9.6.2)V K for (XS1 , DS1 )/S .

Because {d r } (r 2) are functorial with respect to a morphism of p-adic

enlargements, we have the following commutative diagram for r 2:
Erk,h+k ((X, D)/S)K Erk,h+k ((XS1 , DS1 )/S )K

k,h+k
dk,h+k
r dr
Erk+r,h+kr+1 ((X, D)/S)K Erk+r,h+kr+1 ((XS1 , DS1 )/S )K .
Here, if r = 2, then the two horizontal morphisms above are isomorphisms.

By induction on r 2, we see that d
r vanishes if dr does. Hence it suces
to prove that the boundary morphism
(2.17.2.2) dr k,h+k : Erk,h+k ((XS1 , DS1 )/S )K

Erk+r,h+kr+1 ((XS1 , DS1 )/S )K (r 2)

is the zero. Let l(M ) be the length of a nitely generated BK = BK -module

M . Furthermore, to prove the vanishing of dr , it suces to prove that
(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
Set S := Spf(C). Then we have the morphism (XS1 , DS1 ) S . Let us

denote the pull-back of the morphism (XS1 , DS1 ) S by s : S S
by (XS , DS ) S . Then, since we have = and s = id, both
1 1
(XS1 , DS1 ) and (XS , DS ) are deformations of (XS1 , DS1 ) to S1 . Hence, by
1 1
(2.16.2), the spectral sequence (16.6.2)V K for (XS1 , DS1 )/S and that for
(XS , DS )/S are isomorphic. Therefore we have
1 1
E2k,h+k ((XS1 , DS1 )/S )K = E2k,h+k ((XS 1 , DS 1 )/S )K

= B C E2k,h+k ((XS1 , DS1 )/S )K .
Hence, to prove (2.17.2.3), it suces to prove that
(2.17.2.4)
dimK (Rh f(XS ,DS )/S (O(XS ,DS )/S )K )
k,h+k
1 1 1 1
= dimK ( E2 ((XS1 , DS1 )/S )K ).

k
Set V1 := V /p. Because there exists a morphism Spf(V ) Spf(C) of

p-adic enlargements of S, it suces to prove that
(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
We reduce (2.17.2.5) to a result of [68, (5.2) (1)](=(2.15.4) for (2.9.6.2) in

this book) by using (a log version of) a result of Berthelot-Ogus ([12, 2]) as
follows.
Let be the residue eld of V . Since is perfect and since is a nite
extension of , is also perfect. Let W be the Witt ring of . The ring V1
is an artinian local -algebra with residue eld ([79, II Proposition 8]).
Set X := XV1 V1 and D := DV1 V1 . Then (X V1 , D V1 ) and
(XV1 , DV1 ) are two log deformations of (X , D ). Therefore, by (2.16.2), the
spectral sequence (2.9.6.2)V K for (X V1 , D V1 )/V and that for
(XV1 , DV1 )/V are isomorphic. From this fact, the log base change theorem
([54, (6.10)]) and the compatibility of Gysin morphisms with base change ([3,
VI Theorème 4.3.12]), we have
(2.17.2.6) Rh f(XV ,DV )/V (O(XV ,DV )/V ) V K

1 1 1 1

Rh f(X ,D )/W (O(X ,D )/W )W K ,
(2.17.2.7)

E2k,h+k ((XV1 , DV1 )/V ) V K E2k,h+k ((X , D )/W )W K .
Hence it suces to prove that
E2k,h+k ((X , D )/W )W K = E

k,h+k
((X , D )/W )W K .
We have already proved this in [68, (5.2) (1)](=(2.15.4)).

Case II: Next, we consider the general case. Let m be a maximal ideal
of BK . Consider the following ideal I (n) and the following ring B(n) in
[74, p. 780]:
I (n) := Ker(B BK /mn ), B(n) := B/I (n) (n N).
The ring B(n) denes a p-adic enlargement S(n) of S. Let
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
be the boundary morphism. Because {d

r } is functorial, we have the following
commutative diagram:
Erk,h+k ((X, D)/S) (B

k,h+k ((X
B (n) )K Er (S(n) )1 , D(S(n) )1 )/S(n) )K

k,h+k
dk,h+k
r BK (B )
(n) K dr,(n)
k+r,h+kr+1 ((X, D)/S) (B k+r,h+kr+1 ((X

Er B (n) )K

Er (S(n) )1 , D(S(n) )1 )/S(n) )K .
Because E2k,h+k ((X, D)/S)K is a convergent F -isocrystal, the two horizon-

tal morphisms are isomorphisms if r = 2. By induction on r and by the
proof for the Case I, the boundary morphism d r BK (B(n) )K (r 2) van-
ishes. Thus limn (d B /m n
) = 0. Because B K is a noetherian ring and
r BK K
k,h+k
E2 ((X, D)/S)K is a nitely generated BK -module, we have
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)).

We can reprove (2.13.3) in the case Z = and more:

Corollary 2.17.3. Let k be a nonnegative integer. Then the following hold:
(1) There exists a convergent F -isocrystal E2k,h+k ((X, D)/K) such that
E2k,h+k ((X, D)/K)T = grP h

h+k R f(XT1 ,DT1 )/T (O(XT1 ,DT1 )/T )K
for any p-adic enlargement T of S over Spf(V ).

(2) There exists a convergent F -isocrystal Pk Rh f (O(X,D)/K ) such that
Pk Rh f (O(X,D)/K )T = Pk Rh f(XT1 ,DT1 )/T (O(XT1 ,DT1 )/T )K

(3) There exists a spectral sequence of convergent F -isocrystals on (X, D)
/S over Spf(V ) :
(2.17.3.1) E1k,h+k ((X, D)/K) = Rhk f (OD(k) /K Z (k) (D/K))(k)

= Rh f (O(X,D)/K ).
This spectral sequence degenerates at E2 .
Proof. (1): By (2.8.5), the boundary morphism d 1 of (2.9.6.2) V K is a

summation (with signs) of Gysin morphisms, and thus d 1 is a morphism of
convergent F -isocrystals by [74, (3.13)]. By [74, (3.1)] and by (2.17.2), we
obtain (1).
(2): By (1), for a morphism g : T T of p-adic ane enlargements of S
over Spf(V ), Pk Rh f(XT ,DT )/T (O(XT ,DT )/T )K = g Pk Rh f(XT1 ,DT1 )/T (O
1 1 1 1
(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:
Corollary 2.17.4. Let k be a nonnegative integer. Then the following hold:

k,hk
(1) There exists a convergent F -isocrystal E2,c ((X, D)/K) such that
k,hk
E2,c ((X, D)/K)T = grP h
hk R f(XT1 ,DT1 )/T ,c (O(XT1 ,DT1 )/T )K

(2) There exists a convergent F -isocrystal Pk Rh f,c (O(X,D)/K ) on S/
Spf(V ) such that
(Pk Rh f,c (O(X,D)/K ))T = Pk Rh f(XT1 ,DT1 )/T ,c (O(XT1 ,DT1 )/T )K
for any p-adic enlargement T of S/Spf(V ).

(3) There exists a spectral sequence of convergent F -isocrystals on X/S
over Spf(V ) :
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).

As in [11, 7], for a p-adic formal V -scheme S, we have a log crystalline

topos ((X, D)/S)log
crys and the forgetting log morphism (X,D)/S : ((X, D)/S)

crys (X/S)crys . The following is nothing but a restatement of a part of
log
(2.17.2) by the p-adic version of (2.7.6):
Corollary 2.17.5. The following Leray spectral sequence

(2.17.5.1) E2k,hk = Rhk f (X,D)/S Rk (X,D)/S (O(X,D)/S )K
= Rh f(X,D)/S (O(X,D)/S )K
degenerates at E3 .
2.18 Strict Compatibility
In this section, using a specialization argument of Deligne-Illusie (2.14) and

by using the convergence of the weight ltration (2.13, 2.17), we prove
the strictness of the induced morphism of log crystalline cohomologies by a
morphism of log schemes with respect to the weight ltration.
perfect residue eld of characteristic p > 0 and with fraction eld K. Let
g : (X , D ) (X, D) be a morphism of two proper smooth schemes with
SNCDs over . Let W be the Witt ring of and K0 the fraction eld of W .
Theorem 2.18.1. Let h be an integer. Then the following hold:
(1) The induced morphism
(2.18.1.1) log
gcrys h
: Hlog-crys (X/W )K Hlog-crys
h
(X /W )K
is strictly compatible with the weight ltration.

(2) Assume that g induces morphisms g (k) : D(k) D(k) for all k N.
Then the induced morphism
(2.18.1.2) log
gcrys,c h
: Hlog-crys,c (X/W )K Hlog-crys,c
h
(X /W )K

2.18 Strict Compatibility 211
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
we may assume that V = W . By (2.9.1) the morphism g induces a morphism

(2.18.1.3)
log
gcrys h
: Pk Hlog-crys (X/W )K0 Pk Hlog-crys
h
(X /W )K0 (k Z {}).
Let Pk Hlog-crys

h
(X /W )K0 be the image of Pk Hlog-crys
h log
(X/W )K0 by gcrys .
Then we prove that
(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
by shrinking S. Here, note that the extension /Frac(A1 ) of elds may be

innite and transcendental. Because Pk and P
are convergent isocrystals

((2.13.3) or (2.17.3)), so is Pk P by [74, (2.10)]. Since two inclusions
(Pk P

) Pk Pk and (Pk P
) Pk Pk P

are morphisms of
convergent isocrystals, it suce to prove that
(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.

Theorem 2.18.2 (Strict compatibility). Let S be a p-adic formal V -

scheme. Let f : (X, D) S1 and f : (X , D ) S1 be proper smooth
schemes with relative SNCDs over S1 . Let g : (X , D ) (X, D) be a mor-
phism of log schemes over S1 . Let h be an integer. Then the following hold:
(1) The induced morphism
(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)

(2) Assume that g induces morphisms g (k) : D(k) D(k) for all k N.
Then the induced morphism
(2.18.2.2)
gc : Rh f(X,D)/S,c (O(X,D)/S )K Rh f(X

,D )/S,c (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).

Corollary 2.18.3. Let the notations be as in (2.18.2). Let g : (X , D )

(X, D) be a log etale morphism such that Rg (OX ) = OX (e.g., the blowing
up along center a smooth component of D(k) ). Then g in (2.18.2.1) is a
ltered isomorphism.
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
Rf (X/S1 (log D)) Rf (X /S1 (log D ))
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
g : RfX/S (O(X,D)/S ) Rf(X

,D )/S (O(X ,D )/S )
is an isomorphism; in particular, g : Rh f(X,D)/S (O(X,D)/S )K Rh f(X ,D )

/S (O(X ,D )/S )K is an isomorphism. (2.18.3) follows from (2.18.2) (1).

Remark 2.18.4. Let the notations be as in (2.18.2). We do not know an ex-

ample such that the induced morphism g : (Rh f(X,D)/S (O(X,D)/S ), P )

(Rh f(X ,D )/S (O(X ,D )/S ), P ) is not strictly compatible with the weight l-
tration.
2.18 Strict Compatibility 213
Theorem 2.18.5. Let the notations be as in (2.18.2). Assume that g induces

morphisms g (k) : D(k) D(k) for all k N. Assume, moreover, that g is
log etale, that Rg (OX ) = OX and that g (OX (D)) = OX (D ). Then gc
in (2.18.2.2) is a ltered isomorphism.
Proof. We may assume that S is at over Spf(V ). Because g is log etale, we

have g (iX/S1 (log D)) = iX /S1 (log D ) (i N). Hence, by the assumption,
we have g (iX/S1 ( log D)) = iX /S1 ( log D ). By using the projection
formula as in [65, p. 168], we have iX/S1 ( log D) = Rg (iX /S1 ( log D )).
Consequently, as in [65, (2.4)], we have X/S1 ( log D) = Rg (X /S1 ( log
D )) by using the spectral sequence
E1ij = Rj g (iX /S1 ( log D )) = Ri+j g (X /S1 ( log D )).
Let n be a positive integer, and set Sn := SpecS (OS /pn ). Then we have
an exact sequence
0 pn OS /pn+1 OS OSn+1 OSn 0.
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]:
(2.18.5.1) Rf(X,D)/S1 ,c (O(X,D)/S ) L

OS1 p OS /p
n n+1
OS
Rf(X,D)/Sn+1 ,c (O(X,D)/Sn+1 )
+1
Rf(X,D)/Sn ,c (O(X,D)/Sn ) .
Hence, by induction on n and by (2.11.7.1) and [11, 7.22.2], we have
Rh f(X ,D )/S,c (O(X ,D )/S ) = Rh f(X,D)/S,c (O(X,D)/S ).
In particular, gc is an isomorphism of OS V K-modules. Moreover, by

(2.18.2) (2), gc is a ltered isomorphism.

Remark 2.18.6. It is straightforward to generalize (2.18.2), (2.18.3), (2.18.5)

into the framework of convergent F -isocrystals.
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
Rj g (iX /S (log D )) = 0 (j Z>0 ) and g (iX /S (log D )) = iX/S (log D) (a

very special case of [65, (2.2)]; however, note that in the notations in [loc. cit.],

the condition that the closed immersion Y X is a regular embedding is
necessary for [loc. cit.] because the fact Rf (OX ) = OX in [43, VII (3.5)]
has been shown under this assumption.). The pull-back morphism
g : (2X/S (log D), P ) g (2X /S (log D ), P )
is a morphism of ltered sheaves; however, as remarked in [loc. cit.], g is not

strict. (Consequently g does not induce an isomorphism of ltered sheaves
of log dierential forms.)
Note that the number of smooth components of D is more than those of
D; the log structure of (X , D ) is bigger than that of (X, D).
Remark 2.18.8. The following remark is the crystalline analogue of a part of
results in [24, (9.2)].
Let (S, I, ) and S0 be as in 2.4. Let f : (X, D) S0 be a smooth scheme

with a smooth relative divisor over S0 . Let a : D X be the natural closed
immersion. Then, by (2.6.1.1), we have the following exact sequence
(2.18.8.1) 0 QX/S (OX/S ) QX/S CRcrys (O(X,D)/S )

QX/S acrys (OD/S )(1){1} 0.
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 ) .
The spectral sequence (2.9.6.2) degenerates at E2 in this case since E2ij = 0

if i = 0 or i = 1. It is easy to check that the exact sequence (2.18.8.2) is
strictly compatible with the preweight ltration.
Using (2.11.7.1), we also have the following exact sequence which is strictly
compatible with the preweight ltration
(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
(2.18.8.4) Rh fX/S (OX/S )K Rh fD/S (OD/S )K

by the closed immersion D X is an isomorphism for h d 2 and an
injection for h = d 1 (cf. [2, Theorème]). Indeed, rst consider the case h
d 2. Then we can assume that S is the formal spectrum of a nite extension
of V by [74, (3.17)]. In this case, the argument in the proof of (2.18.2) and
the specialization argument of Deligne-Illusie ([49, 3.10], cf. the argument in
(2.18.1)) show that the hard Lefschetz theorem holds for Rh fX/S (OX/S )K
(cf. [49, 3.8]). Hence the proof of [57, p. 76 Corollary] shows that (2.18.8.4) is
an isomorphism for h d 2. As to the case h = d 1, the same proof works
by considering the image of Rd1 fX/S (OX/S )K in Rd1 fD/S (OD/S )K . By
the Poincare duality ([74, (3.12)]), the Gysin morphism
Gh : Rh2 fD/S (OD/S )K (1) Rh fX/S (OX/S )K
is an isomorphism for h d + 2 and a surjection for h d + 1. Set
Rd1 fD/S,ev (OD/S )K (1) := Ker Gd+1 .
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 .
2.19 The Weight-Filtered Poincar

e Duality
The following is the Poincare duality:

Theorem 2.19.1 (Weight-ltered Poincar e duality). Let V be a com-
plete discrete valuation ring of mixed characteristics with perfect residue eld
of characteristic p > 0. Let S be a p-adic formal V -scheme. Let (X, D) be
a formally smooth scheme with a relative SNCD over S. Assume that X/S
is projective and that the relative dimension of X/S is of pure dimension d.
Then there exists a perfect pairing of convergent F -isocrystal on S/Spf(V )
(2.19.1.1) Rh f,c (O(X,D)/K ) R2dh f (O(X,D)/K ) OS/K (d),
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))
is an isomorphism of weight-ltered convergent F -isocrystals on S/V .

Proof. By (2.11.3), there exists a canonical morphism R2d f,c (O(X,D)/K )

R2d f (OX/K ) of convergent isocrystals on S/Spf(V ), which is constructed
from natural morphisms R2d f,c (O(XT1 ,DT1 )/T ) R2d f (OXT1 /T ) for p-
adic enlargements T of S/Spf(V ). Using the cup product, we have the fol-
lowing composite morphism
(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
R2d f (OX/K ) OS/K (d).

f
by [74, (3.12.1)]. The morphism (2.19.1.2) is an isomorphism. Indeed, by [74,

(3.17)], we may assume that S is the spectrum of a perfect eld of nite
characteristic. In this case Tr is the classical trace map ([74, pp. 809810]),
f
Therefore (2.19.1.2) for S = Spec() is an isomorphism by [85, (5.6)], and
hence we have an isomorphism (2.19.1.2).
By using the arguments in (2.18.1) and (2.18.2), we obtain the strict com-
patibility of the isomorphism (2.19.1.2) with the weight ltration.

2.20 l-adic Weight Spectral Sequence
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 ).
calculates Rh f,c (Ql ).

Let j : U X be the open immersion. By Grothendiecks absolute pu-

rity, which has been solved by O. Gabber ([33]), we obtain Rk j (Ql )
(k) (k)
a (Ql,D(k) Z et (D/S)(k)). As in the Introduction, we use the following
isomorphism
2.20 l-adic Weight Spectral Sequence 217
(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)).
Then we have the following spectral sequence:

(k) (k)
(2.20.0.3) E2k,hk = Rhk f (Ql Z et (D/S)(k)) = Rk f (Ql ).
The spectral sequence (2.20.0.1) (resp. (2.20.0.3)) degenerates at E2

(resp. E3 ) by the standard specialization argument (e.g., [34]) and the
Weil conjecture ([26, (3.3.9)]).
Chapter 3
Weight Filtrations and Slope
Filtrations on Rigid Cohomologies
(Summary)
In this chapter, we give an outline of the construction of the weight ltration

and the calculation of the slope ltration on the rigid cohomology of a sepa-
rated scheme of nite type over a perfect eld of characteristic p > 0. They are
constructed and calculated by using de Jongs alteration theorem, Tsuzukis
proper cohomological descent for rigid cohomology, the comparison theorem
between log crystalline cohomology and rigid cohomology by the second-
named author and the results in the previous chapters in this book. It is
analogous to the fact that the mixed Hodge structure on the cohomology
of a separated scheme of nite type over C is constructed from the mixed
Hodge complex on open smooth schemes over C via the technique of proper
hypercovering. The detailed proof of the results in this chapter is given by
another book [70] by the rst-named author.
3.1 Preliminaries for Later Sections
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 )

220 3 Weight Filtrations and Slope Filtrations on Rigid Cohomologies (Summary)
and := (r-points). The object M denes an (r + 1)-uple complex

M = (M t1 tr )t1 ,...,tr N of A-modules whose boundary morphisms will be
xed in (3.1.0.1) below. Let us consider the following single complex s(M )
with the following boundary morphism (cf. [24, (5.1.9.1), (5.1.9.2)]):

(3.1.0.1) s(M )n = M ts = M t1 tr s ;
tr +s=n t1 ++tr +s=n
t
1 +1 t
2 +1
d(xts ) = (1)i 1i (xts ) + (1)t1 (1)i 2i (xts ) +

i=0 i=0
t
r +1
+ (1)tr1 (1)i ri (xts ) + (1)tr dM (xts ) (xts M ts ),

i=0
where dM : M ts M t,s+1 is the boundary morphism arising from the

boundary morphism of the complex M and ji : M t1 tj tr s M t1 tj1 ,
tj +1,tj+1 tr s
(1 j r, 0 i tj + 1) is a standard coface morphism. We
can check that the functor
(3.1.0.2) s : C(A ) C(A)
induces the following functor
(3.1.0.3) s : D+ (A ) D+ (A)
(cf.[42, Vbis (2.3.2.2)]).

Next we consider the ltered version of the above.
Let (M, P ) := (M, {Pk M }) be

an object CF(A ). Then (M, P ) denes
of
an (r+1)-uple ltered complex ( t0,s M , { t0,s Pk M ts }) of A-modules.

ts
Here t 0 means that tj 0 (1 j r). Then we have the following

functors
(3.1.0.4) s : C+ F(A ) (M, P )

(s(M ), {s(Pk M )}) C+ F(A),
(3.1.0.5) s : D+ F(A ) [(M, P )]

[(s(M ), {s(Pk M )})] D+ F(A).
Let (P ) = {(P )k }kZ be the diagonal ltration on s(M ) dened by the

following formula (For the case r 2, this is equal to [24, (7.1.6.1), (8.1.22)].):

(3.1.0.6) (P )k (s(M )) = Ptr +k M ts .
t0,s
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
As in [24, (7.1.6.3)], we have the following functor
(3.1.0.8) (s, ) : D+ F(A ) [(M, P )]

[(s(M ), (P ))] D+ F(A).
Next we generalize the above to the mapping ber of a morphism in

D+ F(A ).
Let : (M, P ) (N, P ) be a morphism in CF(A ). Let MF() := (M, P )
(N, P )[1] be the mapping ber of . Then s(MF()) = s(M, P )
(s(N, P )[1]). We dene the diagonal ltration MF (P ) on s(MF()) as
follows:

(3.1.0.9) MF (P )k s(MF()) := Ptr +k M ts Ptr +k+1 N t,s1 .
t0,s t0,s
By the denition of MF (P ), we have

(P )

(3.1.0.10) grkMF s(MF()) = grP
tr +k M
ts
grP
tr +k+1 N
t,s1
.
t0,s t0,s
Since the morphism : grP

tr +k M
t
grP
tr +k+1 N
t
is the zero,
(3.1.0.11)
(P )
tr +k M [tr ]
grP tr +k+1 N [tr 1].
t
grkMF s(MF()) = grP t
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
be the boundary morphism arising from the following exact sequence

(P ) (P )
0 grk1 s(M ) ((P )k /(P )k2 )s(M ) grk s(M ) 0.
Let
d(N ) : Hhtr 1 (grP
tr +k+1 N )
t
Hhtr (grP t
tr +k N )
t0 t0
be the analogous boundary morphism for (N, P ). The boundary morphism
d : E1k,h+k E1k+1,h+k
between the E1 -terms in (3.1.0.12) is equal to

(3.1.0.13)

d(M )
: Hhtr (grP
tr +k M )
t
Hhtr 1 (grP
tr +k+1 N )
t
0 d(N )
t0 t0

Hhtr +1 (grP
tr +k1 M )
t
Hhtr (grP t
tr +k N ).
t0 t0
Let [] be the morphism in D F(A ) induced by . Let Mor(D+ F(A )) be

+
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).
Next we consider the decreasing stupid ltration on the mapping ber.

Let : M N be a morphism in C+ (A ). Then induces a morphism
: (M, ) (N, )
of ltered complexes, where s are decreasing stupid ltrations. Then we

dene a decreasing stupid ltration on s(MF()) as follows:

(3.1.0.15) i s(MF()) = M ts N t,s1 (i N).
t0,si t0,s1i
Then

(3.1.0.16) gri s(MF()) = M ti [tr ]{i} N ti [tr 1]{i}
t0 t0
= sMF( {i} : M {i} N i {i}),

i i
where i : M i N i is the degree i-part of . Hence we have the following

spectral sequence:
(3.1.0.17) E1i,hi = Hhi (s(MF(i ))) = Hh (s(MF())).
Let (T , A ) be an r-simplicial ringed topos. For a family {nj }uj=1 (1

u r) of nonnegative integers, we also need the following natural restriction
functor in later sections:
(3.1.0.18)
e1
n1 n2 nu1 nu : D (A ) [M ]
[M n1 n2 nu1 nu ] D+ (An1 n2 nu1 nu ).
Let N be a nonnegative integer. Next we recall the key functor N in [19],

[87] and [88].
Let in this section be the standard simplicial category: an object of
is denoted by [n] := {0, . . . , n} (n N); a morphism in is a non-decreasing
function [n] [m] (n, m N).
Let C be a category in which there exist nite inverse limits. Let l
be a nonnegative integer. Following [19, 11], dene a set Homl ([n], [m])
(n, m N): the set Homl ([n], [m]), by denition, consists of the morphisms
: [n] [m] in such that ([n]) l. Let us recall the denition of a
simplicial object := CN (X)l in C for an object X C ([19, 11], [88,
(7.3.1)]): for an object [m], set

m := X
l
Hom ([N ],[m])
with X = X; for a morphism : [m ] [m] in , : m m is

dened to t the following commutative diagram for any : [N ] [m ]
such that ([N ]) l:

m m

proj.
proj.

id
X X
X .
In fact, we have a functor
(3.1.0.19) CN (?)l : C Hom(o , C) =: {simplicial objects in C}.
Set CN (X) = CN (X)N . Then CN (X)l = CN (X) for l N .

Let XN be an N -truncated simplicial object of C and let f : XN Y
be a morphism in C. Then we have a morphism
(3.1.0.20) XN CN (Y )N
dened by the following commutative diagram:

Xm Hom ([N ],[m]) Y

X()
proj
(3.1.0.21)
f
XN Y = Y.
Assume, furthermore, that C has nite disjoint sums. Let N be a full

subcategory of consisting of [k]s (0 k N ). An object of Hom(oN , C)
is, by denition, an N -truncated simplicial object in C. Let us recall the
skelton:
skCN : Hom(o , C) X
XN Hom(oN , C).
The coskelton
coskCN : Hom(oN , C) Hom(o , C)
is, by denition, the right adjoint functor of skCN . Let [r] (r N) be the cat-
egory of objects of augmented to [r] and let N [r] be a full subcategory of
[r] whose objects are [q] [r]s such that q N . The explicit description
of the coskelton is the following ([42, Vbis (3.0.1.2)]):
coskCN (XN )r = lim Xq (r N).

N [r]
As in the proof of [19, (11.2.5)], we have the following by the adjointness

of the skelton and the coskelton:
Lemma 3.1.1 ([19, (11.2.5)], [88, (7.3.2)]). coskCl (CN (X)l ) = CN (X)l .
Let S be a (formal) scheme (resp. a ne log (formal) scheme). Let CS be
the category of (formal) schemes (resp. ne log (formal) schemes) over S.
The inverse limit of a nite inverse system exists in the category of ne log
(formal) schemes (over a ne log (formal) schemes) (cf. [54, (1.6), (2.8)]). Set
SN (?) := CNS (?) and coskSN (?) := coskCNS (?).
The following are immediate generalizations of [19, (11.2.4), (11.2.6)]:
Lemma 3.1.2 ([70, (6.5)]). Let N be a nonnegative integer. Let S be a ne
log (formal) scheme. Then the following hold:
(1) Let XN be a ne N -truncated simplicial log (formal) scheme over S.
If XN Y is a closed immersion of ne log schemes over S, then the mor-
phism XN SN (Y )N in (3.1.0.20) is an immersion of N -truncated

ne log (formal) schemes over S. Moreover, if X N and Y are separated

over S, then the morphism XN SN (Y )N is a closed immersion.
(2) Let X S be a morphism of ne log (formal) schemes. Assume that
the morphism X S satises a condition (P) which is stable under the
base change X S Y Y for any Y CS . Then the natural morphism
coskSl (SN (X)l )m coskSl (SN (X)l )m
satises (P) for l < l and for any m.

Let U be a separated scheme of nite type over a eld . We recall the
following denition:
Denition 3.1.3 ([24, (5.3.8)]). Let N be a nonnegative integer (resp. ).
Let UN be a separated N -truncated simplicial scheme (resp. a separated
simplicial scheme) of nite type over . The N -truncated simplicial scheme

UN is said to be an N -truncated proper hypercovering (resp. a proper
hypercovering) of U if the natural morphism Un+1 coskU n (Un )n+1
(1 n N 1) is proper and surjective. Here we set coskU
1 (U1 )0 := U .

Let U U be an open immersion of separated schemes of nite type
over . As in [19] and [87], we call (U, U ) a pair over . Let n be a nonnegative
integer. We recall the following denition due to Tsuzuki:
Denition 3.1.4 ([87, (2.1.1) (2)]). Let N be a nonnegative integer
(resp. ). Let

UN U N

U U
be a commutative diagram of (N -truncated simplicial) separated schemes
of nite type over , where the horizontal morphisms are open immersions.
(As in [19] and [87], we call (UN , U N ) a pair over (U, U ).) The pair
(UN , U N ) is said to be an N -truncated proper hypercovering (resp. a
proper hypercovering) of (U, U ) if UN is an N -truncated proper hypercov-
ering (resp. a proper hypercovering) of U , if U N is proper over U and if the
morphism (UN , U N ) (U, U ) is strict, that is, UN = U N U U .
Lemma 3.1.5 ([70, (10.5)]). Let N be a nonnegative integer and let (U, U )
be a pair. Let UN be an N -truncated proper hypercovering of U . Then there
exists an N -truncated proper hypercovering (UN , U N ) of (U, U ).
Finally we note the following, which is an immediate generalization of the
technique used in the proof of (2.10.3).
Let F : (T , A) (T , A ) be a morphism of ringed topoi. Consider a
small category I := {i, i } consisting of two elements. The morphisms in
I, by denition, consist of three elements idi , idi and a morphism i i.
Set (Ti , Ai ) := (T , A) and (Ti , Ai ) := (T , A ). By corresponding the natural
morphism F to the morphism i i, we have a ringed topos (Tj , Aj )jI . Let
(F , P ) and (F , P ) be a ltered complex of A-modules and a ltered com-
plex of A -modules, respectively, with a morphism (F , P ) F ((F , P ))
of ltered complexes of A -modules. Set (Fi , Pi ) := (F , P ) and (Fi , Pi ) :=
(F , P ). Then we have a ltered complex (Fj , Pj )jI of (Aj )jI -modules.
Let (Fj , Pj )jI be a ltered asque resolution of (Fj , Pj )jI . Then the com-
plex s(Fj , Pj )jI gives us asque resolutions of s(F , P ) and s(F , P )
of (F , P ) and (F , P ), respectively, tting into the following commutative
diagram
(F , P ) F ((F , P ))

(3.1.5.1)
s(F , P ) F (s(F , P )).
3.2 Rigid Cohomology

(not necessarily perfect) residue eld . Let W be a Cohen ring of in V.
Let K (resp. K0 ) be the fraction eld of V (resp. W ). Let U be a separated
scheme of nite type over . By Nagatas embedding theorem ([66]), there

exists an open immersion j : U U into a proper scheme over . Let C be
an overconvergentm isocrystal on U/K ([7, (2.2), (2.3)], [19, (10.3.1)]).
Let Z := i=1 Z i (m Z>0 ) be the disjoint union of ane open sub-
schemes which cover U . Set Z := Z U U . The scheme Z can be embedded
into a separated formally smooth p-adic formal V-scheme Z. Then we have
V
a Cech diagram Z := (Z , Z , Z ) := (coskU U
0 (Z), cosk0 (Z), cosk0 (Z)). Let
: (Z , Z ) (U, U ) be the natural augmentation. Denote the open im-

mersion Z Z by j . Then, by [19, (10.1.4)], the Cech diagram is a
universally de Rham descendable hypercovering of (U, U ) over Spf(V). Then
we have, by denition,
(3.2.0.1) Rrig (U/K, C) := R(]Z [Z , DR( (C)))
([19, (10.4)]). Here DR( (C)) is the complex (C)Z j O j ]Z

]Z [Z [ Z

on ]Z [Z . For simplicity of notation, we denote DR( (C)) only by DR(C).
The complex Rrig (U/K, C) is a well-dened object in the derived category
D+ (K) of bounded below complexes of K-vector spaces ([19, (10.4.3)]). We
call Rrig (U/K, C) the rigid cohomological complex of C (in [19, p. 185]
Chiarellotto and Tsuzuki have called it the rigid cohomology complex of C).
When C is the trivial coecient, we call Rrig (U/K, C) the rigid cohomolog-
ical complex of U/K and we denote it by Rrig (U/K). Set
h
(3.2.0.2) Hrig (U/K, C) := H h (Rrig (U/K, C))
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.
The following is a reformulation of the proper cohomological descent of

rigid cohomology in Tsuzukis theory ([87] and [88]):
Theorem 3.2.1 ([70, (10.9)]). Let U be a proper hypercovering of U . Let

C t be the pull-back of C by the structural morphism Ut U (t N). Let c
3.2 Rigid Cohomology 227
be an integer in (3.2.0.4). Let N be a positive integer satisfying the inequality

N > 21 (c + 1)(c + 2). Let D+ (K N ) be the derived category of N -truncated
cosimplicial complexes of K-vector spaces. Let s (resp. e1 t (0 t N ))
be the N -truncated cosimplicial version of the functor (resp. the cosimplicial
version of the functor) dened in (3.1.0.3) (resp. (3.1.0.18)) for D+ (K N ) :
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 ).
The simplicial formal V-scheme V N (PN )N contains V N as an N -truncat-

ed simplicial closed subscheme. Set Q := VN (PN ).
Following the idea in [87, (4.4)], we consider the triple (VN, , V N, ,
QN, ) of (N, )-truncated bisimplicial (formal) schemes dened by
(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)
with the natural morphisms which make (VN, , V N, , QN, ) a triple of

(N, )-truncated bisimplicial schemes.
Set
N
(3.2.1.6) Crig ((UN /U )/K, C) := RN (]V N, [QN, , DR(C)).
Here the N -truncated cosimplicial degree in RN means the rst

N -truncated cosimplicial degree. The natural morphism (VN, , V N, ,
QN, ) (Z , Z , Z ) of triples is shown to induce an isomorphism
(3.2.1.1) ([87, pp. 126127], [70, (10.9)]). For a nonnegative integer t N ,
(Vt , V t , Qt ) is a universally de Rham descendable covering of (Ut , U t ) by
the proof of [87, (4.4.1) (2)]. Hence
(3.2.1.7) Rrig (Ut /K) = R(]V t [Qt , DR(C)),
which is nothing but the isomorphism (3.2.1.2).
3.3 Comparison Theorems

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)]):
Theorem 3.3.1 ([28, Introduction]). There exists the following cartesian

diagram

U0 X0

(3.3.1.1)

U U
such that the left vertical morphism is proper and surjective and such that U0
is the complement of an SNCD D0 in a projective regular scheme X0 over .
Using (3.3.1) and a general formalism in [42, Vbis ], we have the following:
3.3 Comparison Theorems 229
Proposition 3.3.2 ([70, (9.2)]). There exists a proper hypercovering (U ,

U )N of (U, U ) ((3.1.4)). Moreover one can take the pair (U , U )N over
(3.3.2.1): X := U is regular,
(3.3.2.2): D := X \ U is a simplicial SNCD on X ,

(3.3.2.3): (U , X ) is split, that is, there exists an open immersion N (U )m
N (X)m (m N) of schemes such that

(Un , Xn ) = ( N (U )m , N (X)m )
0mn [n][m] 0mn [n][m]
([42, Vbis (5.1.1)], [24, (6.2.2)]). Here [n] [m] means a surjection in .
Denition 3.3.3. We say that a proper hypercovering of (U, U ) satisfying

(3.3.2.1) and (3.3.2.2) is good. We say that a proper hypercovering of (U, U )
satisfying (3.3.2.1) (3.3.2.3) is gs(=good and split).
We can prove the following without diculty.

Lemma 3.3.4 ([70, (9.6)]). Let U Y be an open immersion of simpli-

cial schemes over an open immersion U Y of schemes over . Assume
that (U , Y ) is split. Then there exists a pair (U , Y ) of split simplicial
schemes over (U, Y ) with a natural morphism (U , Y ) (U , Y ) of the
pairs of the simplicial schemes satisfying the following conditions:

(3.3.4.1): Um (m N) is the disjoint union of open subschemes of Um which
are open subschemes of ane open subschemes of Ym ; these open subschemes
cover Um .
(3.3.4.2): Ym (m N) is the disjoint union of ane open subschemes which
cover Ym and whose images in Y are contained in ane open subschemes
of Y .

(3.3.4.3): Um = Um Ym Ym (m N).
(3.3.4.4): If (U , Y ) is strict over (U, Y ), that is, U = Y Y U , then (U , Y )
is strict over (U, Y ).
(3.3.4.5): If Ym (m N) is quasi-compact, then the number of the open
subschemes in (3.3.4.1) and (3.3.4.2) can be assumed to be nite.
In particular, there exists a natural morphism (U , Y ) (U , Y ) by set-

ting (Umn , Ymn ) := (coskU0 (Um )n , cosk0 (Ym )n ). Moreover, for each n N,
m Ym
(Un , Yn ) is split.
Let Y be a proper smooth scheme over and E an SNCD on Y /. Set

V := Y \E. Let K(Y,E)/V be the isostructure sheaf in ((Y, E)/V)log
conv . Then, by
[82, Corollary 2.3.9] and [82, Theorem 2.4.4], we have a canonical isomorphism

(3.3.4.6) R(((Y, conv , K(Y,E)/V ) Rrig (V /K).
E)/V)log
By [82, Theorem 3.1.1] we also have a canonical isomorphism
R(((Y,

(3.3.4.7) conv , K(Y,E)/W ) Rlog-crys ((Y, E)/W )K0 .
E)/W )log
Let (U , X ) be a gs proper hypercovering of (U, U ). Assume that X is

smooth over . Set D := X \ U .
Now we state the following comparison theorems. ((1) and (2) in (3.3.5)
are generalizations of (3.3.4.6) and (3.3.4.7), respectively. The main part of
the proof of (3.3.5) is the reduction to the facts (3.3.4.6) and (3.3.4.7).)
Theorem 3.3.5 (Comparison theorems ([70, (11.6)])). (1) There ex-

ists an isomorphism
R(((X

(3.3.5.1) conv , K(X ,D )/V ) Rrig (U/K).
, D )/V)log
The isomorphism (3.3.5.1) is functorial: for a morphism (V, V ) (U, U ) of

pairs over and for gs proper hypercoverings (U , X ) of (U, U ) and (V , Y )
of (V, V ), respectively, tting into the following commutative diagram
(V , Y ) (U , X )

(V, V ) (U, U ),
the following diagram

(3.3.5.2)
R(((Y
conv , K(Y ,E )/V ) R(((X , D )/V)conv , K(X ,D )/V )
, E )/V)log log

Rrig (V /K) Rrig (U/K)
is commutative, where E := Y \ V . In particular, R(((X conv , K

, D )/V)log
(X ,D )/V ) depends only on U/ and K.
(2) There exists a functorial isomorphism
(3.3.5.3)
R(((X

, D )/W )log , K(X ,D )/W ) Rlog-crys (((X , D )/W )K .
conv 0
The construction of the morphism (3.3.5.1) is as follows.

Let h be a nonnegative integer. Let N be a suciently large integer (e.g.,
N > 21 (h + 1)(h + 2); the reader does not need to mind this inequality).
Let (U , X ) be a pair of bisimplicial schemes over in (3.3.4) for the pair
(U , X ). Let (Z, Z, Z) be the triple in the beginning of 3.2. Consider the
ber product (U U Z, X U Z) and a pair of schemes dened by the

following:
(Xlmn , Dlmn ) := (coskX

0
lm
(Xlm U Z)n , coskD
0
lm
(Dlm U Z)n )
(l, m, n N).
Then we have the pair (X , D ) of a smooth trisimplicial scheme with

trisimplicial SNCD over . Set U := X \ D . Because XN 0 is
ane, there exists a lift (XN 0 , DN 0 ) of (XN 0 , DN 0 ) over Spf(V) such that
XN 0 is a formally smooth scheme over Spf(V) and such that DN 0 is a
relative SNCD on XN 0 /Spf(V). By (3.1.2) (1) we have a closed immer-
sion (XN,0 , DN,0 ) VN ((XN 0 , DN 0 ))N into a formally log smooth
N -truncated simplicial scheme over Spf(V). Set l := V N ((XN 0 , DN 0 ))l
(l N). Then we have a closed immersion
V
0 (Xl0 )m , cosk0 (Dl0 )m ) cosk0 (l )m
(Xlm , Dlm ) = (coskX (0 l N ).
l Dl
Set lm := coskV
0 (l )m . Then we have a closed immersion
(Xlm , Dlm ) = (coskX

0
lm
(Xlm U Z), coskD
0
lm
(Dlm U Z))

coskV
0 (lm V Z) (m, n N).
In this way we have a closed immersion

(3.3.5.4) (XN, , DN, ) (PN, , MN, )
into a formally log smooth (N, , )-truncated trisimplicial p-adic ne

log formal V-scheme such that the underlying trisimplicial formal scheme
PN, is formally smooth over Spf(V); moreover there exists a morphism
PN, Z over V tting into the following commutative diagram

XN, PN,

Z Z .

Let jN, : UN, XN, be the open immersion. The morphism
(3.3.5.4) induces the following morphism
(3.3.5.5) Rrig (U/K) = R(]Z [Z , j ]Z [Z )

R(]XN, [PN, , jN, ]XN, [P ).
N,
Using the triple (VN, , V N, , QN, ) of (N, )-truncated bisimplicial

(formal) schemes in the previous section, we can prove that H h ((3.3.5.5)) is
an isomorphism ([70, (11.6)]).
Let R((XN, , DN, )/K, LDR(O]XN, [log )) be the complex

PN,
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
(3.3.5.6) R(((XN conv , K(XN ,DN )/V )

, DN )/V)log
= R(((XN, conv , K(XN, ,DN, )/V )

, DN, )/V)log
= R((XN, , DN, )/K, LDR(O]XN, [log ))
PN,

R((XN, , DN, )/K, LDR(jN, O]XN, [PN, ))

:= R(]XN, [PN, , jN, ]XN, [P ).
N,
We can check that the induced morphism H h ((3.3.5.6)) is an isomorphism

by Shihos comparison theorems ([82, Corollary 2.3.9, Theorem 2.4.4]) ([70,
(11.6)]).
Thus, for any nonnegative integer h, there exists a suciently large positive
integer N such that there exists an isomorphism
(3.3.5.7) h R(((X conv , K(X ,D )/V )

, D )/V)log
= 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
Using (3.3.5.7) for h = c, we have the isomorphism (3.3.5.1).

We can check that the isomorphism (3.3.5.1) is independent of the choice
of the lift (XN 0 , DN 0 ) and the choice of (U , X ). We can also check that
the isomorphism (3.3.5.1) is independent of the choice of (Z , Z , Z ) and
the choice of N satisfying N > 21 (c + 1)(c + 2). Moreover we can prove the
functoriality of the isomorphism (3.3.5.1). See [70, (11.6)] for details.
The construction of the morphism (3.3.5.3) is as follows.
Let AN, be the structure sheaf of the log PD-envelope of the closed

immersion (XN, , DN, ) (PN, , MN, ). By the proof of [82,
Theorem 3.1.1], we have a morphism
(3.3.5.9) R((XN, , DN, )/K0 , LDR(O]XN, [log ))

PN,
R(XN, , AN, OPN, PN, /W (log MN, ))K0 .
By the usual cohomological descent, the canonical ltration h on the right

hand side is nothing but h Rlog-crys ((X , D )/W )K0 .
As in the case of (3.3.5.6), by Shihos comparison theorem ([82, Theorem
3.1.1]), we see that the induced morphism H h ((3.3.5.9)) is an isomorphism
([70, (11.6)]).
Now, by (3.3.5.8), we have the isomorphism (3.3.5.3).
We can prove that the isomorphism (3.3.5.3) is independent of the choice
of the lift (XN 0 , DN 0 ). We can also check that the isomorphism (3.3.5.3) is
independent of the choice of N satisfying N > 21 (c + 1)(c + 2). Moreover
we can prove the functoriality of the isomorphism (3.3.5.3) ([70, (11.6)]).
Corollary 3.3.6. The following hold:
(1) Let the notations and assumptions be as in (3.3.5) (1). Then there

(3.3.6.1) Rrig (U/K) Rlog-crys ((X , D )/W )K .
In particular, Rlog-crys ((X , D )/W )K depends only on U and K, and there


(3.3.6.2) h
Hrig (U/K) Hlog-crys
h
((X , D )/W )K (h Z).
h
In particular, Hrig (U/K) is nite dimensional (because we have the following
spectral sequence
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
scheme tting into the following commutative diagram

V V

(3.3.7.1)

U U
over (e.g., [5, p. 340]). Using the general formalism in [42, Vbis ], we have gs
proper hypercoverings (V , Y ) and (U , X ) of (V, V ) and (U, U ), respectively,
tting into the following commutative diagram:

(V , Y ) (U , X )

(3.3.7.2)

(V, V ) (U, U ).
Set D := X \ U and E := Y \ V . For a morphism f : (F , dF )

(G , dG ) of complexes in an additive category A, let MF(f ) := (F , dF )
(G [1], dG [1]) be the mapping ber of f . This operation is well-dened in
the derived category D+ (A).
Then we have the following by the functoriality in (3.3.5) (3).
Corollary 3.3.8. There exists the following canonical isomorphism

(3.3.8.1) MF(rig : Rrig (U/K) Rrig (V /K))
MF( : Rlog-crys ((X , D )/W )K Rlog-crys ((Y , E )/W )K ).
This isomorphism is independent of the choice of the commutative diagram

(3.3.7.2).
Let (S, I, ) be a PD-scheme on which a prime number p is locally nilpotent

or a (not necessarily ane) P -adic base in the sense of [11, 7.17 Denition].
Set S0 := SpecS (OS /I). Let X be a smooth simplicial scheme over S0
and let D and Z be transversal simplicial relative SNCDs on X /S0 . Let
f : X S be the structural morphism. As in 2.3, we have the following
forgetting log morphisms of simplicial log schemes over S and log crystalline
ringed topoi:
(3.4.0.1) (X ,D Z ,Z )/S : (X , D Z ) (X , Z ),
(3.4.0.2)
(X ,D Z ,Z )/S : (((X , D Z )/S)crys , O(X ,D Z )/S )
log
(((X crys , O(X ,Z )/S ).

, Z )/S)log
3.4 Weight Filtration on Rigid Cohomology 235
Let I be a asque resolution of O(X ,D Z )/S , which exists by [42, Vbis

(1.3.10)]. Then, set
(3.4.0.3) log,Z
(Ecrys (O(X ,D Z )/S ), P D ) := ((X ,D Z ,Z )/S (I ), )
in D+ F(O(X ,Z )/S ). Let (I , P D ) be the ltered asque resolution of the

right hand side above. Set
(3.4.0.4) log,Z
(Ezar (O(X ,D Z )/S ), P D ) := u(X ,Z )/S ((I , P D ))
in D+ F(f 1 (OS )). It is clear that

log,Zt
(Ecrys (O(Xt ,Dt Zt )/S ), P Dt ) and log,Zt
(Ezar (O(Xt ,Dt Zt )/S ), P Dt ) (t N)
in this section are isomorphic to

log,Zt
(Ecrys (O(Xt ,Dt Zt )/S ), P Dt ) and log,Zt
(Ezar (O(Xt ,Dt Zt )/S ), P Dt )
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 ).
Let (J , P ) be a ltered asque resolution of ((Y ,E )/W (I(Y

,E )/W
), ).
Then, by using the morphism (2.7.2.1), we have the following composite
morphism
(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 )).
The morphism (3.4.0.5) in C+ F(OX /W ) gives us a morphism
(3.4.0.6) (Ecrys (O(X ,D )/W ), P ) Rcrys (Ecrys (O(Y ,E )/W ), P )
in D+ F(f 1 (W )). As a result, we have a natural morphism
(3.4.0.7) (Ezar (O(X ,D )/W ), P ) R (Ezar (O(Y ,E )/W ), P )
in D+ F(f 1 (W )). We can also take a representative of the morphism (3.4.0.7)

in C+ F(f 1 (W )).
Consider the following morphism

: R (Xzar , (Ezar (O(X ,D )/W ), P ))
R (Yzar , (Ezar (O(Y ,E )/W ), P ))
of ltered complexes of W -modules induced by (3.4.0.7) and the following

ltered mapping ber of
:
(s(MF(
)), MF (P )).
(k) (k) (k) (k)

Let at : Dt Xt (k, t N) and bt : Et Yt (k, t N) be the
natural morphisms. Then (3.1.0.11) for (s(MF(
)), MF (P )) is
(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
{t k 1}[t 1]((t + k + 1))
in D+ (f 1 (W )) by (2.7.5.1) and (2.6.1.2).

By (3.1.0.12) and (3.3.8.1), we have the following spectral sequences:
Theorem 3.4.1 ([70, (16.1)]). Denote MF(rig : Rrig (U/K) Rrig
(V /K)) simply by MF(rig ). Then there exists the following spectral sequence:
(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

(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 )).
The spectral sequence (3.4.1.1) degenerates at E2 . The induced ltration P

on H h (MF(rig )) is independent of the choice of the commutative diagram
(3.3.7.2).
Denition 3.4.2. We call (3.4.1.1) the weight spectral sequence of MF(rig ).

We call the ltration P the weight ltration on H h (MF(rig )).
We can prove the E2 -degeneration of the spectral sequence (3.4.1.1) as in

(2.17.2). We can prove the independence of P by using Grothendiecks idea
for the reduction of geometric problems to arithmetic problems ([40, 3], [37])
in the following way:
(3.4.2.1): for two commutative diagrams (3.3.7.2)s over , we may assume

that there exists a morphism between them (a general formalism in [42, Vbis ]),
(3.4.2.2): for any nonnegative integer N , the existence of a model of N over

the spectrum of a smooth ring of nite type over a nite eld (the standard
log deformation theory (cf. the proof of (2.18.1)),
(3.4.2.3): the reduction to the case of nite elds (the specialization argument
of Deligne-Illusie ([49], [68], 2.14)),
(3.4.2.4): the purity of the eigenvalues of the Frobenius on the classical crys-
talline cohomology of a proper smooth variety over a nite eld ([57], [15,
(1.2)], [68, (2.2) (4)]).
See [70, (12.5), (15.2)] for the more detailed proof of the independence
of P .
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).
Theorem 3.4.4 (Strict compatibility ([70, (15.5)])). Let

V U

v
u

V U
be a commutative diagram of separated schemes of nite type over . Then
the induced morphism
(u , v ) : H h (MF(rig )) H h (MF(rig )) (h Z)
Proposition 3.4.5 ([70, (15.4)]). The weight ltration on H h (MF(rig )) is

compatible with the extension of complete discrete valuation rings of mixed
characteristics with perfect residue elds.
Let Z be a closed subscheme of U . Let V be the complement of Z in U

and let : V U be the open immersion. Set
(3.4.5.1) Rrig,Z (U/K) := MF(rig : Rrig (U/K) Rrig (V /K))
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).
The spectral sequence (3.4.6.1) degenerates at E2 . The induced ltration P

h
on Hrig,Z (U/K) is independent of the choice of the commutative diagram
(3.3.7.2).
(2) Let u : (U , Z ) (U, Z) be a morphism of separated schemes of nite
type over with closed subschemes. Then the induced morphism
(3.4.6.2) u : Hrig,Z
h
(U/K) Hrig,Z
h
(U /K)
is strictly compatible with P s.

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.
Theorem 3.4.9 ([70, (16.6)]). The following hold:

(1) The exact sequence
(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)
(cf. [5, (2.4.1)]) is an isomorphism of weight-ltered K-vector spaces.

(3) Let T be a closed subscheme of Z. Then the natural morphism
(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)
(cf. [5, (2.4.2)]) is an isomorphism of weight-ltered K-vector spaces.
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).).
Theorem 3.4.10 ([70, (16.7)]). Set d := dim U , c := codim(Z, U ) and

dZ := dim Z. Assume that U and Z are of pure dimensions and that U is
smooth over . (Then d = c + dZ ([45, II Exercise, (3.20) (d)]).) Then the
following hold:
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.
The following is a generalization of Berthelots K

unneth formula ([6, (3.2)])
and Kedlayas Kunneth formula ([59, (1.2.4)]):
Theorem 3.4.11 (cf. [24, (8.1.25)]). Let U i (i = 1, 2) be a separated

scheme of nite type over a not necessarily perfect eld of characteristic

p > 0. Let Z i be a closed subscheme of U i . Let U i U i be an open immer-
sion into a proper scheme over . Set U 12 := U 1 U 2 and Z 12 := Z 1 Z 2 .
Then the canonical morphism
(3.4.11.1) Rrig,Z 1 (U 1 /K0 ) K0 Rrig,Z 2 (U 2 /K0 ) Rrig,Z 12 (U 12 /K0 )
is an isomorphism. Assume, furthermore, that is perfect. Then the induced

isomorphism on the cohomologies by the isomorphism (3.4.11.1) is compatible
with the weight ltration.
h
From the denition of the weight ltration on Hrig (U/K), we obtain the
following:
Proposition 3.4.12 ([70, (12.9) (2)]). If U is the complement of an SNCD

D on a proper smooth scheme X over , then the weight ltration P on
h h
Hrig (U/K) = Hlog-crys ((X, D)/W )K coincides with the weight ltration de-
ned by the ltered zariskian complex (Czar (O(X,D)/W ), P ).
3.5 Slope Filtration on Rigid Cohomology
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
ural morphism (X , D ) (X , D ). Let g : X Spec(Wn )

(n Z1 ) be the structural morphism. By using [68, (9.1), (9.3) (2)]
again, we have a complex Wn X (log D ) on X in C+ (g 1 (Wn )),
which is the bicosimplicial version of Wn X (log D) in 2.12. We have
3.5 Slope Filtration on Rigid Cohomology 241
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 ).
Let N be a nonnegative integer. Then we have a closed immersion

(XN , DN ) (XN , DN
) into a formally smooth scheme over Spf(W )
with a relative SNCD. We can assume that there exists an endomorphism
: (XN , DN
) (XN , DN
) lifting the Frobenius of (XN , DN

) mod p.
By using the functor (3.1.0.19) and (3.1.2) (1), we have an immersion

(XN , DN ) YN into a formally log smooth scheme over Spf(W ).

Note that YN also has an endomorphism lifting the Frobenius of YN

(Yl )m
Spf(W )
modp. Set YN mod pn := YN W Wn . Set Ylm := cosk0
(0 l N ) and Ylm mod p := Ylm W Wn . By the obvious bicosimpli-
n
cial version of (3.4.0.3) (resp. (3.4.0.4)), we have a complex
(3.5.0.1) Ecrys (O(X ,D )/Wn ) (resp. Ezar (O(X ,D )/Wn ))
in C+ (OX /Wn ) or D+ (OX /Wn ) (resp. C+ (g 1 (Wn )) or D+ (g 1 (Wn ))).

Let DN, be the log PD envelope of the (N, )-truncated simplicial im-
mersion

(XN, , DN, ) YN,
over (Spf(W ), pW, [ ]). Set DN, mod pn := DN, W Wn . Then, in
D+ (g 1 (Wn )), we have the following explicit description
(3.5.0.2) Ezar (O(XN, ,DN, )/Wn )

RuXN, /Wn (Ecrys (O(XN, ,DN, )/Wn ))

uXN, /Wn LXN, /Wn (YN, mod pn /Wn )
= ODN, mod pn OYN, mod pn

YN, mod pn /Wn
by the Poincare lemma of a vanishing cycle sheaf (2.3.10) and by (2.2.1)

(cf. the proof of (2.2.22)). Because
Wn iXN, (log DN, ) = Hi (ODN, mod pn OYN, mod pn

YN, /Wn ),
we have a natural morphism
(3.5.0.3) Ezar (O(XN, ,DN, )/Wn ) Wn XN, (log DN, )
in D+ (g 1 (Wn )) as in 2.12. Indeed, this morphism is an isomorphism by

(2.12.11.1). The isomorphisms (3.5.0.3) for ns are compatible with two pro-
jections of both hands of (3.5.0.3). As a corollary, we have the following
composite canonical isomorphism
(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
Rlog-crys ((X , D )/W )K0 = c Rlog-crys ((X , D )/W )K0 .
Because
c Rlog-crys ((X , D )/Wn ) = c Rlog-crys ((XN , DN )/Wn )
and
c R(X , Wn X (log D )) = c R(XN , Wn XN (log DN )),
we have the following canonical isomorphism by (3.5.0.4):

(3.5.0.5) Rlog-crys ((X , D )/W )K0 R(X , W X (log D ))K0 .
(The isomorphism (3.5.0.5) is independent of the choice of N .) Hence we have

the following isomorphism by (3.3.8.1):

(3.5.0.6) MF(rig ) MF(log
dRW : R(X ,W X (log D ))K0
R(Y , W Y (log E ))K0 ).
For each nonnegative integer i, we have the following morphism by the

functoriality of log Hodge-Witt sheaves ([68, (9.3) (2)]):
i
(3.5.0.7) log
dRW : R(X , W X (log D )) R(Y , W Y (log E )).
i i
Here we calculate R(X , W iX (log D )) and R(Y , W iY (log E )) by the

Godement resolutions of W iX (log D ) and W iY (log E ), respectively.
By (3.4.1.1) we know that H h (MF(rig )) is an F -isocrystal, which is a
generalization of [87, (5.1.1)]. As a consequence, H h (MF(rig )) has the slope
ltration. We have the following spectral sequence (3.5.1.1) by (3.1.0.17) and
(3.5.0.6):
Theorem 3.5.1 (Slope decomposition ([70, (15.7)])). The following

hold:
(1) There exists the following spectral sequence
3.5 Slope Filtration on Rigid Cohomology 243
i
(3.5.1.1) E1i,hi = H hi (MF(log h
dRW ))K0 = H (MF(rig )).

(2) There exists the following isomorphism:
i
(3.5.1.2) H hi (MF(log
dRW ))K0 (H (MF( )))[i,i+1) .
h
(3) There exists the following slope decomposition

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
(3.5.4.1) E1ij = H j (X , W iX (log D ))K0 = Hrig

i+j
(U/K0 ).

(2) There exists the following canonical isomorphism:

i+j
(3.5.4.2) H j (X , W iX (log D ))K0 Hrig (U/K0 )[i,i+1) .
In particular, H j (X , W iX (log D ))K0 depends only on U/.

(3) There exists the following slope decomposition

h
h
(3.5.4.3) Hrig (U/K0 ) = H hi (X , W iX (log D ))K0 .
i=0
P. Berthelot, S. Bloch and H. Esnault have given an impressive description

h
of Hrig (X/K0 )[0,1) by the cohomology of Witt rings on a proper scheme X
over :
Theorem 3.5.5 ([9]). Let X be a proper scheme over . Then there exists
a functorial morphism
h
Hrig (X/K0 ) H h (X, W (OX ))K0 (h N)

h
which induces an isomorphism Hrig (X/K0 )[0,1) H h (X, W (OX ))K0 .
By (3.5.4) and (3.5.5), we have the following corollary:

Corollary 3.5.6 ([70, (13.7)]). Let X be a proper scheme over . Let
X X be a gs proper hypercovering of X. Then the natural morphism
H h (X, W (OX ))K0 H h (X , W (OX ))K0 (h N)
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}].

with Compact Support
Let the notations be as in 3.3. In this section, assume that is perfect.

First we note that there is an announcement in [4, pp. 2123] that the rigid
cohomology with compact support of a separated scheme U of nite type
over is dened.
Let U be a separated smooth scheme of nite type over of pure dimen-
sion d. Then, by the same proof as that of [6, (2.4)], we have a canonical
isomorphism

(3.6.0.1) h
Hrig,c (U/K) HomK (Hrig
2dh
(U/K), K).
2d
The trace morphism TrU : Hrig,c (U/K) K ([6]) is an underlying mor-
phism of the following morphism
(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.
Theorem 3.6.1 ([83] ([73])). Let S be a p-adic formally smooth formal

W -scheme. Set S1 := S W . Endow S with the canonical PD structure
(pOS , [ ]). Let f : (X, D) S1 be a proper smooth scheme with a rela-
tive SNCD over S1 . Set U := X \ D. By abuse of notation, denote also by
f the structural morphism U S1 . Let Rh f(X,D)/SK0 (K(X,D)/SK0 ) and
Rh f(X,D)/S (O(X,D)/S ) (h Z) be the relative log convergent cohomology of
(X, D)/S and the relative log crystalline cohomology of (X, D)/S, respec-
tively. Let Rh frig (U/SK0 ) be the relative rigid cohomology of U/SK0 ([19,
(10.6)]). (It is an OSK0 -module, where SK0 denotes the rigid analytic space
associated to S.) Let sp : SK0 S be the specialization map dened in [8,
(0.2.3)]. Then the canonical morphisms
(3.6.1.1) Rh f(X,D)/SK0 (K(X,D)/SK0 ) Rh f(X,D)/S (O(X,D)/S )K0
and
(3.6.1.2) Rh f(X,D)/SK0 (K(X,D)/SK0 ) sp Rh frig (U/SK0 )
are functorial isomorphisms.
Using (3.6.1) and similar arguments to (3.4.2.2), (3.4.2.3) and (3.4.2.4),

we can obtain the following:
Theorem 3.6.2 ([70, (17.6)]). Let U be a separated smooth scheme of nite
type over of pure dimension d. Let Z be a smooth closed subscheme of U .
Then the following composite morphism of the cup product and the trace
morphism
Tr
(3.6.2.1) h
Hrig,c (Z/K) K Hrig,Z
2dh
(U/K) Hrig,c
2d
(U/K)
U
K(d)
induces an isomorphism

(3.6.2.2) 2dh
Hrig,Z (U/K) HomK (Hrig,c
h
(Z/K), K(d))
of weight-ltered vector spaces over K.

Using Berthelots duality (3.6.0.3) and (3.6.2), we obtain the following:
Corollary 3.6.3 ([70, (17.7)]). Assume that Z is of pure codimension c.

Then the Gysin isomorphism

(3.6.3.1) h2c
GZ/U : Hrig (Z/K)(c) Hrig,Z
h
(U/K)
is an isomorphism of weight-ltered K-vector spaces.

By (3.4.6) and (3.6.2), we obtain the following:

Corollary 3.6.4 ([70, (17.9)]). Let f : U V be a nite etale morphism
of smooth ane schemes over . Then the trace morphism
h
Trf : Hrig,c (U/K) Hrig,c
h
(V /K)
[6, (1.4)] is strictly compatible with the weight ltration.

Let Z be a separated scheme of nite type over . Until (3.6.6), assume
that there exists a closed immersion

(3.6.4.1) Z U
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).
Proposition 3.6.7 ([70, (17.1)]). Let U i (i = 1, 2) be a separated smooth

scheme of nite type over . Then Berthelots K unneth isomorphism

(3.6.7.1) h1
Hrig,c (U 1 /K) K Hrig,c
h2
(U 2 /K) Hrig,c
h
(U 1 U 2 /K)
h1 +h2 =h
is compatible with the weight ltration.

Let (X, D) be a proper smooth scheme with an SNCD over . Set U :=
X \ D. By [85], [82, Theorem 2.4.4, Corollary 2.3.9, Theorem 3.1.1] and [6,
(2.4)], we obtain
3.6 Weight Filtration on Rigid Cohomology with Compact Support 247
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).
Proposition 3.6.9 ([70, (17.13)]). Let Z and U be as in (3.6.4.1). Set

V := U \ Z. Then the following sequence
(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.
Proposition 3.6.10 ([70, (17.19)]). Let f : U V be a proper mor-

phism of separated schemes of nite type over . Assume that U and V are
closed subschemes of separated smooth schemes of nite type over . Then
the pull-back
f : Hrig,c
h
(V /K) Hrig,c
h
(U/K)
The following immediately follows from [6, (2.2) (ii)] and (3.6.10).
Corollary 3.6.11 ([70, (17.20)]). Let f : U V be a nite etale mor-

phism of ane smooth schemes over . Then the trace morphism
h
Trf : Hrig (U/K) Hrig
h
(V /K)
([6, (1.4)]) is strictly compatible with the weight ltration.
Proposition 3.6.12 ([70, (17.21)]). Let Z be as in (3.6.4.1). Then the

canonical morphism
(3.6.12.1) h
Hrig,c (Z/K) Hrig
h
(Z/K)
([4, (3.1) (i)]) is strictly compatible with the weight ltration.
Let Z be as in (3.6.4.1). Finally we give ranges of weights and slopes of

h
Hrig,c (Z/K).
The following is a p-adic version of a special case of [26, III (3.3.4)]; see
also [37, Partie II, III 3].
Proposition 3.6.13 ([70, (17.14)]). Let Z be a separated scheme of nite

type over of pure dimension dZ . Assume that there exists a closed immer-
sion (3.6.4.1). Let h [0, 2dZ ] be an integer. Then the weights of Hrig,c
h
(Z/K)
lie in [0, h].
The following is [16, (3.1.2)] (see also [26, III (3.3.8)]).

Proposition 3.6.14 ([16, (3.1.2)], [70, (17.3)]).

Let h [0, 2d] be an integer. Let U be a separated scheme of nite type
h
over of dimension d. Then the slopes of Hrig,c (U/K) lie in [max{0, h d},
min{h, d}].
Appendix A
Relative SNCD
Let f : X S be a smooth morphism of schemes and let D be a relative

SNCD on X/S. Set U := X \ D and let j : U X be the natural open

immersion. Set N (D) := OX j (OU ) as in 2.1. In this appendix we prove
basic properties of D, M (D) and N (D):
(A) We determine the local behavior of the decomposition of D by smooth
components ((A.0.1)).
(B) We rene decompositions of D by smooth components as possible
((A.0.3), (A.0.7)).
(C) We prove the equality N (D) = M (D) if S is reduced at any point of
f (D) ((A.0.8) (1)).
(D) We show the inequality N (D) = M (D) if S is not reduced at f (x) for

a point x X ((A.0.8) (2)). In fact, we prove that the stalk (N (D)/OX )x is
not nitely generated ((A.0.9)).
For a morphism Y T of schemes, recall Div(Y /T )0 in 2.1. If T is
the spectrum of a commutative ring A with unit element, we often denote
Div(Y /T )0 simply by Div(Y /A)0 in this section. For a non-zero divisor
section y (Y, OY ) such that SpecY (OY /yOY ) is at over T , we denote
by div(y) the eective Cartier divisor dened by the ideal sheaf yOY of OY .
Let us take a diagram (2.1.7.2) (with S0 replaced by S) and a point z in
s
V X. Then, we may assume that z i=1 {yi = 0} by shrinking s V and
replacing s if necessary. Henceforth we always assume that z i=1 {yi = 0}
when we take a diagram like (2.1.7.2) and a point z in V .
In (2.1.7), there is no relation a priori between a decomposition of D
by smooth components and the diagram (2.1.7.2). Though the uniqueness
of the decomposition does not necessarily hold, the local decomposition is
determined by the diagram (2.1.7.2):
Proposition A.0.1. Let f : (X, D) S be as in the beginning of this

section. Let := {D } be a decomposition of D by smooth components.
Let z be a point of D and assume that we are given
s a cartesian diagram
(2.1.7.2) (with S0 replaced by S, such that z i=1 {yi = 0}). Then, by
249
250 A Relative SNCD
shrinking V , for any 1 i s, there exists a unique element i satisfy-

ing Di |V = div(yi ) in Div(V /S)0 .
Proof. Set B := OX,z and Xz := Spec(B). Let D1 , . . . , Dm be the elements

in which contain z and let di (1 i m) be elements of B such that
Di Xz = div(di ). Then we have the following equality

m
s
(A.0.1.1) div(di ) = div(yi )
i=1 i=1
in Div(Xz /S)0 by the diagram (2.1.7.2). We have to prove that s = m and

that yi B = di B (1 i s) up to some renumbering of the indexes.
Case I: First consider the case S = Spec(), where is a eld.
In this case, B is a regular local ring. By Auslander-Buchsbaums theorem
[63, 19, Theorem 48], B is a UFD. Since ODi ,z = B/di B is also a regular
localring, the ideal di B of B is prime. Since yi B (1 i s) (because
s
z i=1 {yi = 0}), the ideal yi B of B is also prime. Since B is a UFD, the
equality (A.0.1.1) implies the equalities s = m and di B = yi B (1 i s) up
to some renumbering of the indexes. This completes the proof in the Case I.
Case II: Next consider the case S = Spec(A), where A is an integral
domain.
Set K := Frac(A) and let be the generic point of S. For a scheme
T over S, set T := T S . Set BK := B A K and for a point w in
Xz, := Spec(BK ), denote by (BK )w the localization of BK at the prime
ideal corresponding to w. Then Spec((BK )w ) is nothing but the localization
of X at w, where we identify w with its image in X .
Fix an index i (1 i m) for the moment. Since Di Xz is at over S,
there exists a point w Di Xz such that w is sent to by the structural
morphism Xz S (Indeed, we may assume that S is local; in this case the
morphism Di Xz S is faithfully at.). Then w Di, . By the result in
the Case I, there exists an index j (1 j s) such that
(A.0.1.2) di (BK )w = yj (BK )w .
Denote by the composite morphism B BK (BK )w and by

the induced morphism B/yj B (BK )w /yj (BK )w by . Then is at.
Consider the following commutative diagram
(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
Because the upper horizontal sequence in (A.0.1.3) is exact and because is

at, the lower horizontal sequence is also exact. The morphisms B BK
and B/yj B BK /yj BK are injective. Because BK and BK /yj BK are
regular, they are integral domains. Hence the morphisms BK (BK )w and
BK /yj B (BK )w /yj (BK )w are also injective. Thus the vertical morphisms
in the diagram (A.0.1.3) are injective. By an easy diagram-chasing, we have
an equality yj B = 1 (yj (BK )w ). Similarly we have an equality di B =
1 (di (BK )w ). Hence di B = yj B by (A.0.1.2).
For each i (1 i m), we may take j(i) with 1 j(i) m such that
di B = yj(i) B. Then we obtain the following equality

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.
Since yi is a non-zero divisor in B and in B/pB, the horizontal sequences

are exact. Because Ker() = (pB + yi B)/yi B, an easy diagram-chasing
gives an equality 1 (yi (B/pB)) = yi B + pB. Similarly we have an equality
1 (di (B/pB)) = di B +pB, and hence we have an inclusion di B yi B +pB,
that is,
(A.0.1.5) di (B/yi B) p(B/yi B)

252 A Relative SNCD
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
By reducing the equality (A.0.1.6) modulo n, we obtain

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
is cartesian (because B/ne+1 B is at over (A/ne+1 )[y1 , . . . , yd ]), i0 yi0 B +

ne+1 B. Hence, for any i (1 i s), i yi B + ne+1 B and consequently
di yi B + ne+1 B. Therefore we obtain the inclusion di B yi B + ne+1 B,
infact, the inclusion
s di B yi B. By the last inclusion and the equality
s
( i=1 di )B = ( i=1 yi )B, we can easily deduce the equality di B = yi B. We
complete the proof in the Case III.
Case IV: Finally we prove the proposition in the general case. By shrinking
V , the structural morphism V S ts into the following cartesian diagram
A Relative SNCD 253
V V

S S
(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 ]),
over S which is compatible with the diagram (2.1.7.2), where g is an etale

morphism. Let z be the image of z in V . Then, by the result in the Case III
to (V , D), s = m and div(di ) = div(yi ) (1 i s) in Div(Spec(OV ,z )/S) up
to some renumbering of the indexes. Then div(di ) = div(yi ) (1 i s) in
Div(Xz /S). This equality completes the proof of the proposition.

To study the global behavior of decompositions of a relative SNCD

by smooth components, we introduce the notion of the renement of a
decomposition by smooth components.
Denition A.0.2. (1) Let f : (X, D) S be as in the beginning of this

section and let := {D } and := {D } be decompositions of
D by smooth components. Then we say that is a renement
of if, for
any , there exists a subset of with = such that
D = D .
(2) We say that a closed and open set of a topological space is clopen.
If X is locally noetherian as a topological space, we can prove the existence

of the nest (hence canonical) decomposition of a relative SNCD by smooth
components:
Proposition A.0.3. Let f : (X, D) S be as in the beginning of this sec-

tion. If X is locally noetherian as a topological space, then there exists a
unique decomposition of D by smooth components which is a renement of
any decomposition of D by smooth components.

Proof. Let T be a locally noetherian topological space and let T = iI Ti
(Ti Tj = for i = j) be the decomposition into the connected components
of T . Then we claim that Ti is open. Let O beany noetherian open subset
of T . Set I := {i I | O Ti = }. Then O = iI (O Ti ). We claim that
I is nite. Indeed, if not, we have a union O = O1 Or (r Z>0 ) of the
254 A Relative SNCD
irreducible components of O since O is noetherian. Then there exist a positive

integer s (s r) and i = i (i, i I ) such that Os Ti = and Os Ti = .
Since Os is irreducible, Os is connected. Hence Os Ti Ti is also connected.
However, this union contains both Ti and Ti ; this contradicts the denition
of Ti . Thus we see that I is nite and that Ti O = O\( i I \{i} Ti ) is open
in T . Now we see thatTi is open since T is locally noetherian. In conclusion,
the disjoint sum T = iI Ti is the disjoint sum of clopen subspaces of T .
Take a decomposition = {D } of D by smooth components and
let D = D, be the decomposition of D into the connected com-
ponents. Since X is locally noetherian, so is D . Hence, by the previous
paragraph, we have the decomposition D = D, into clopen sub-
schemes. In particular, each D, is smooth over S and we have an equality
D = D, in Div(X/S)0 . Therefore 0 := {D, }, is a
renement of .
We say that a decomposition of D by smooth components is very ne if
each member of the decomposition is connected as a topological space.
It is clear that 0 is very ne. Hence we have shown that any decompo-
sition of D by smooth components admits a renement by a very ne one.
To prove (A.0.3), it suces to prove that there exists only one very ne
decomposition of D by smooth components.
Let = {D } and = {D } be very ne decompositions of D
by smooth components. Fix an index . For each point z in D , take an
open neighborhood Uz of z in X such that Uz = Uz ((A.0.1)). Then there
exists a unique (, z) such that D Uz = D (,z) Uz .
We claim that (, z) does not depend on z. Let V be a subset of D
dened as follows:

sequence of points z = z0 , z1 , . . . , zn = w in D
V := w D .
satisfying D Uzj Uzj+1 = (0 j n 1)
Then, if w is a point of V , D Uw is contained in V ; V is an open set

of D . On the other hand, if w D is not contained in V , the points in
Uw are not contained in V ; V is a closed set of D . Since D is connected
and V is non-empty (since z V ), D = V . Let w be a point of D . Then
there exists a sequence of points z = z0 , z1 , . . . , zn = w in D satisfying
D Uzj Uzj+1 = for all 0 j n 1. Since D Uzj = D (,zj ) Uzj
and D Uzj+1 = D (,zj+1 ) Uzj+1 ,
D Uzj Uzj+1 = D (,zj ) Uzj Uzj+1 = D (,zj+1 ) Uzj Uzj+1 .
This implies the equality (, zj ) = (, zj+1 ) since Uzj Uzj+1 is non-empty

and that Uzj Uzj+1 = Uz Uz . Hence we have
j j+1
(, 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
and we see that D is an open subscheme of D () . On the other hand, D is

a closed subscheme of D () since it is closed in X. Since D () is connected,
D = D () .

map
() is injective. This fact and the equality
It is clear that the
D = D = D imply the bijectivity of the correspondence

().

In the case where X is not locally noetherian as a topological space, there

exists an example of a relative SNCD which does not have the nest decom-
position by smooth components:

Example A.0.4. Let k be a eld and let A := nN k be the countable product
of k. Set S := Spec(A), X := Spec(A[x]) and let D be a relative SNCD on X

over S dened by the ideal (x) A[x]. Then D S. Let := {D }
be any decomposition of D by smooth components. Then each D is closed

in D and it is open in D because the composite morphism D D S
is smooth. Since D is closed in D Spec(A), D is quasi-compact.
For a subset T of N, dene eT := (eT,n )nN A by eT,n = 1 (resp. 0) if
n T (resp. n / T ). Then the sets UT := {p Spec(A) | eT / p} (T N)
forms an open basis of Spec(A)(= S D). It is easy to see that, for a
l
nite number of subsets T1 , T2 , . . . , Tl of N, we have i=1 UTi = Ul Ti ; if
i=1
T1 T2 = , then UT1 UT2 = .
Because D is open in D and quasi-compact, D is a union of a nite
numbers of open sets of the form UT . By the fact in the previous paragraph,

D UT for a non-empty subset T of N.
If T = 1 ( ), then we can deduce an equality Spec(A) =
nN U{n} . However, a maximal ideal of A containing the ideal
{(xn )nN A | xn = 0 for except nite numbers of ns}

does not belong to nN U{n} and it is a contradiction. Thus, for some ,
thecardinality of T is greater than 1. Then we have a disjoint sum T =
T T of T with T = and T = . Hence we have a renement of by
factorizing D as D = UT + UT . Thus is not the nest decomposition.
Hence, in this case, there does not exist the nest decomposition of D by
smooth components.
As we have seen in the above, we cannot have the nest decomposition of

D in general. However, we can prove that any two decompositions admit a
common renement in the following way.
256 A Relative SNCD
Let be a partial order in Div(X/S)0 such that E1 E2 if and only if

there exists an eective Cartier divisor F on X over S such that E2 = E1 +F .
Then we have the following lemma:
Lemma A.0.5. Let f : X S be as in the beginning of this section and let

= {D } be a family of smooth eective Cartier divisors
on X/S. For
nonnegative integers n1, and n2, ( ), n1, D n2, D if
and only if n1, n2, for all .
For a point t S and for a

Proof. It suces to prove the only if part.
scheme T over S, set Tt := T S t. Then t n1, D,t t n2, D,t ,
where t := { | D,t = }. Since the assertion is well-known in the case
where S is the spectrum of a eld, n1, n2, for t . Since any
belongs to t for some point t S, n1, n2, for all .

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:
E1 E2 = max{F DivD (X/S)0 | F E1 and F E2 }.
(2) Let E and F be elements in DivD (X/S)0 . If E is smooth over S,

then E F is clopen in E. In particular, E F is smooth over S.
(3) Let E be an element in DivD (X/S)0 . Let {F } be a family of
elements in DivD (X/S)0 of locally nite intersection such that F F = 0
for any , with = . Then

E( F ) = (E F ).

Proof. (1): Take a decomposition {D } of D by smooth components.

Then,
by the denition of DivD (X/S)
0 , there exists an open covering X =
jJ X j of X such that E i |Xj
= X ni,j, (D |Xj ) (i = 1, 2) for some
j
Xj is the set { | D |Xj = }. Dene
nonnegative integers ni,j, , where
Fj DivD|Xj (Xj /S)0 by Fj := X min{n1,j, , n2,j, }(D |Xj ). Then it
j
is easy to see (by using (A.0.5)) that Fj Ei |Xj (i = 1, 2) and that Fj is the
maximum element among the elements F s in DivD|Xj (Xj /S)0 satisfying
F Ei |Xj (i = 1, 2). By using this characterization, one can see that there
exists a unique element F DivD (X/S)0 such that F |Xj = Fj for any
j J. Set E1 E2 := F . Then E1 E2 satises the equality in (1).
(2): Since E F is closed in E, it suces to prove that
E F is open in E.
Since the problem is local, we may assume that F = n D (n N).
Moreover, we may assume, by the smoothness of E, that there exists an
A Relative SNCD 257
element 0 satisfying E = D0 . Then, E F = E (resp. 0) if n0 1

(resp. n0 = 0). In both cases, E F is open in E.
(3): Since the problem is local, we may assume that is a nite set and
that
E= n D , F = m, D ,

where n , m, are nonnegative integers and is a subset of satisfying

= for = . Then

E( F ) = min{n , m, }D = (E F ).

Using (A.0.6), we can prove that any two decompositions of a relative

SNCD by smooth components have a common renement:
Proposition A.0.7. Let f : (X, D) S be as in the beginning of this

section. Let = {D } and = {D } be two decompositions of D
by smooth components. Then there exists a decomposition of D by smooth
components which is a renement of and .
Proof. By (A.0.6), D D is smooth over S (possibly empty) and we have

the following equalities

D = D D , D = D D .
=0
,D D =0
,D D

Set := {(, ) | D D = 0}. Then it is easy to see from the

above equalities that := {D D }(, ) has a desired property.
Lastly we discuss log structures on X associated to D. Though we always

consider log structures in the Zariski topos of X, (A.0.8) and (A.0.9) below
remain valid if we consider log structures in the etale topos of X.
By (2.1.9), (X, M (D)) is a ne log scheme; in fact, it is log smooth over S.

Let us recall N (D) := OX j (OU ) with structural morphism N (D)
OX . From the local expression of M (D) given in (2.1.9), there exists a natural

inclusion M (D) N (D). Both M (D) and N (D) have U as the maximal
open subscheme where they are trivial. By [55, (8.2), (11.6)], they coincide in
the case where S is regular. However we prove M (D) N (D) in general, in
fact, we give an equivalent condition for the equality M (D) = N (D) ((A.0.8)
below) and show that N (D) is not a ne log structure in general ((A.0.9)
below):
Proposition A.0.8. Let f : (X, D) S be in the beginning of this section.

(1) If S is reduced at the points of f (D), then M (D) = N (D).
258 A Relative SNCD
(2) If S is not reduced at some point of f (D), then M (D) = N (D).

Consequently, the log structures M (D) and N (D) coincide if and only if
S is reduced at the points of f (D).
Proof. Let z be a point of X. Set B := OX,z and Xz := Spec(B). Take an
open neighborhood V of z which admits the diagram (2.1.7.2).

(1): It suces to prove the surjectivity of the homomorphism M (D)
N (D). To prove this, we may work locally. By (2.1.9), it suces to prove
(Xz , N (D)) = B y1N ysN . Here we denote the pull-back of N (D) to Xz by
the same symbol. We may assume that S is noetherian because we can reduce
the general case to the noetherian case by the well-known technique similar
to that of Case IV in the proof of (A.0.1). Furthermore, we may assume that
S is the spectrum of a noetherian local ring A and that t = f (z) is the closed
point of S. By the assumption, A is reduced.
Let g be an element of (Xz , N (D)) = B B[(y1 ys )1 ] . Because yi
is a non-zero divisor of B (1 i s), there exist an element h B and a
nonnegative integer a satisfying gh = (y1 ys )a in B.
For a prime ideal p of A, consider the equation gh = (y1 ys )a in the
ring B/pB. Then, since yi (B/pB)s (1 i s) are dierent prime ideals of
an integral domain B/pB, there exist unique integers bp,i (1 i s) with
b ab
0 bp,i a satisfying g yi p,i B + pB and h yi p,i B + pB for any prime
b
ideal p of A. Let m be the maximal ideal of A. Then g yi p,i B + mB and h
abp,i
yi B + mB. By the uniqueness of bm,i , we can conclude that bp,i is inde-

pendent of p. Set bi := bp,i . Then g pSpec(A) (yibi B+pB). Because B/yibi B

is at over A and because A is a reduced noetherian ring, pSpec(A) (yibi B +
s
pB) is equal to yibi B. Because B is at over A[y1 , . . . , ys ], i=1 (yibi B) =
s s s
( i=1 yibi )B. Hence g ( i=1 yibi )B. Similarly, h ( i=1 yiabi )B. Dene

g := g/( i=1 yibi ) and h := h/( i=1 yiabi ). Then g h = 1. Consequently
s s
N N
(Xz , N (D)) = B y1 ys .
(2): Let t be a point of f (D) where S is not reduced at t. To prove (2),
we may assume that S is the spectrum of a nonreduced local ring A with the
closed point t. Denote by m the maximal ideal of A and by the residue eld
A/m.
Consider a point z in Dt such that D is smooth at z. Then there
exists an open neighborhood V of z which admits the cartesian diagram
(2.1.7.2) for the case where s = 1, S0 = S = Spec(A) and g(z) is equal
to the maximal ideal mA[y1 , . . . , yd ] + (y1 , . . . , yd ). The induced morphism
g : A[y1 , . . . , yd ] B by g is ind-etale. In the following, we denote the
image g (a) of an element a A[y1 , . . . , yd ] by a by abuse of notation.
To prove (2), it suces to prove that (Xz , N (D)) = B y1N . Set y := y1
for simplicity of notation. Take an element of A satisfying = 0 but 2 = 0.
Then we have (y + )(y ) = y 2 in B. Hence y + (Xz , N (D)). Assume
that y + B y N . Then there exists a nonnegative integer n and an element
u B such that
A Relative SNCD 259
(A.0.8.1) y + = uy n .
Apply the projection B B/(y) = OD,z to the equality (A.0.8.1). If n = 0,

= u in B/(y); this is a contradiction. If n 1, = 0 in B/(y); this
contradicts the choice of since the natural ring homomorphism A B/(y)
is faithfully at. In conclusion, y + (Xz , N (D)) does not belong to
(Xz , M (D)).

Proposition A.0.9. Let the notations be as in the proof of (A.0.8) above.

Then (Xz , N (D))/OX,z is not a nitely generated monoid.
Proof. Keep the notation in the proof of (A.0.8) (2). Assume that
(Xz , N (D))/B were nitely generated. Since (y n + )(y n ) = y 2n
(n Z>0 ), y n + is an element in (Xz , N (D)). Fix a surjective homomor-
phism : Nr (Xz , N (D))/B of monoids for some r and for each posi-
tive integer n, take an element hn Nr satisfying (hn ) = y n + . Then there
exist elements an Q (1 n r + 1) with (a1 , a2 , . . . , ar+1 ) = (0, 0, . . . , 0)
r+1
such that n=1 an hn = 0 in Qr . By multiplying an s by some integers, we
may assume that an s are integers with gcd{ai | ai = 0, (1 i r + 1)} = 1.
Set
{n | 1 n r + 1, an > 0} =: {n1 , n2 , . . . , nq },
n r + 1, an < 0} =: {nq+1 , nq+2 , . . . , nq+s },
{n | 1
ani , (1 i q),
mi :=
ani , (q + 1 i q + s).
q q+s
Then we obtain the equality i=1 mi hni = i=q+1 mi hni in Nr . By applying
to this equality, we see that there exists an element v B such that

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
Let p 0 be the characteristic of . For a positive integer L, let (C)L be

the claim that mi is a multiple of p for is with ni L. We prove that (C)L
holds for any L by descending induction. If L > ni for any 1 i q +s, (C)L
obviously holds. Assume that (C)L+1 is true. Then, if there is no integer i with
ni = L, (C)L obviously holds since (C)L is equivalent to (C)L+1 . Assume that
there exists an integer i0 with ni0 = L. We may assume that 1 i0 q. Let
be the homomorphism A[y, y2 , . . . , yd ] (A/pA)[y]/(y P L+1 ) which
sends a A (resp. y, yi (2 i d)) to the class of a (resp. y, 0). Then the
ring R := B A[y,y2 ,...,yd ], (A/pA)[y]/(y P L+1 ) is a local ring which is ind-
etale over the local ring (A/pA)[y]/(y P L+1 ). Denote the homomorphism
B R by and apply to the equality (A.0.9.3). Then we obtain an
equality mi0 y P L = 0 in R by the induction hypothesis. Since R is faithfully
at over (A/pA)[y]/(y P L+1 ), the equality mi0 y P L = 0 in R implies that
in (A/pA)[y]/(y P L+1 ). Hence we have mi0 = 0 in A/pA. This implies
that there exists an element a A such that (mi0 pa) = 0 in A. Hence mi0
is equal to the zero in , and consequently mi0 is a multiple of p. Therefore
the claim (C)L holds. Now, by the descending induction, (C)1 holds and all
mi s (1 i r) are multiples of p. This contradicts gcd{mi | mi = 0, (1

i r)} = 1. In conclusion, (Xz , N (D))/OX,z is not a nitely generated
monoid.

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
1. Andreatta, F., Barbieri-Viale, L.: Crystalline realizations of 1-motives. Math. Ann.

331, (2005), 111-172.
2. Berthelot, P.: Sur le th eorème de Lefschetz faible en cohomologie cristalline. C.
R. Acad. Sci. Paris Ser. A-B277, (1973), A955A958.
3. Berthelot, P.: Cohomologie cristalline des sch emas de caract
eristique p > 0. Lecture
Notes in Math. 407, Springer-Verlag, Berlin-New York, (1974).
4. Berthelot, P.: G eometrie rigide et cohomologie des vari etes algebriques de car-
act
eristique p. In: Introductions aux cohomologies p-adiques (Luminy, 1984).
Mem. Soc. Math. France 23, (1986), 732.
5. Berthelot, P. Finitude et puret e cohomologique en cohomologie rigide, avec un appen-
dice par Aise Johan de Jong. Invent. Math. 128, (1997), 329377.
6. Berthelot, P.: Dualit
e de Poincar e et formule de Kunneth en cohomologie rigide. C.
R. Acad. Sci. Paris Serie. I Math. t. 325, (1997), 493498.
7. Berthelot, P.: Cohomologie rigide et cohomologie rigide a ` supports propres.
Premi` ere partie, available from Berthelots homepage: http://name.math.univ-
rennes1.fr/pierre.berthelot/.
8. Berthelot, P.: Three lectures Cohomology operators, D-modules and crystalline
sheaves in Dwork Trimester 2001.
9. Berthelot, P., Bloch S., Esnault, H.: Hodge type over the complex numbers and rational
points over nite elds. Preprint, arXiv:math.AG/0510349v1.
10. Berthelot, P., Breen, L., Messing, W., Th eorie de Dieudonne cristalline. II. Lecture
Notes in Math. 930, Springer-Verlag, (1982).
11. Berthelot, P., Ogus, A.: Notes on crystalline cohomology. Princeton Univ. Press,
(1978).
12. Berthelot, P., Ogus, A.: F -isocrystals and de Rham cohomology. I. Invent. Math. 72,
(1983), 159-199.
13. Bourbaki, N.: Algèbre commutative. Masson, Paris, (1985).
14. Chiarellotto, B.: Weights in rigid cohomology. Applications to unipotent F -isocrystals.

Ann. Sci. Ecole Norm. Sup. (4) 31, (1998), 683715.
15. Chiarellotto, B., Le Stum, B.: Sur la puret e de la cohomologie cristalline.
C. R. Acad. Sci. Paris, Serie I, 326, (1998), 961963.
16. Chiarellotto, B., Le Stum, B.: Pentes en cohomologie rigide et F -isocristaux unipo-
tents. Manuscripta Math. 100, (1999), 455468.
17. Chiarellotto, B., Le Stum, B.: A comparison theorem for weights. J. reine
angew. Math. 546, (2002), 159176.
18. Conrad, B.: Grothendieck duality and base change. Lecture Notes in Math. 1750,
Springer-Verlag, (2000).
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.: Theorème 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 Congrès 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.: Relèvements 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 eorème 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
base change theorem, 145, 165, 193 K

unneth formula, 154, 170, 194
comparison theorem, 188, 230 locally nite intersection, 60

compatible, 62 log crystalline cohomology sheaf with
cospecial module, 23 compact support, 156, 194
crystalline orientation sheaf, 81 log crystalline orientation sheaf, 81
log HPD dierential operator, 66
decomposition of an SNCD by its smooth log HPD stratication, 66
components, 61 log PD dierential operator of nite order,
diagonal ltration, 220 66
direct sum, 148
multi-degree, 141
E2 -degeneration, 206
p-adic purity, 118
ltered asque resolution, 19 p-adic weight spectral sequence, 196
ltered at resolution, 25 pair, 225
ltered injective resolution, 19 Poincare duality, 215
ltered log Berthelot-Ogus isomorphism, Poincare lemma of a vanishing cycle sheaf,
203 90
ltered perfect complex, 151 preweight ltration, 76, 176
ltered quasi-isomorphism, 16 preweight spectral sequence, 143, 169
lteredly asque, 18 preweight-ltered restricted crystalline
lteredly at, 24 complex, 100
lteredly injective, 18 preweight-ltered vanishing cycle
lteredly strictly perfect, 148 crystalline complex, 121
forgetting log morphism, 85 preweight-ltered vanishing cycle crys-
functor forgetting the ltration, 36 talline complex with compact
functor taking the ltration, 36 support, 164
functoriality, 137, 195 preweight-ltered vanishing cycle zariskian
complex, 122
good proper hypercovering, 229 preweight-ltered zariskian complex, 100
gr-functor, 36 proper hypercovering, 225
gs proper hypercovering, 229 pure weight, 239
265
266 Index
renement, 253 strict compatibility, 211, 237

relative simple normal crossing divi- strict epimorphism, 17
sor(=:relative SNCD), 60 strict monomorphism, 17
restriction, 62 strict morphism, 17
rigid cohomological complex, 226 strictly exact, 16
rigid cohomological complex with support, strictly at, 23
238 strictly at resolution, 25
rigid cohomology, 226 strictly injective, 17
rigid cohomology with compact support, strictly injective resolution, 19
244
rigid cohomology with support, 238 truncated proper hypercovering, 225
ringed topos associated to a diagram of twist, 141
ringed topoi, 41
slope decomposition, 242, 243 vanishing cycle sheaf, 90

slope spectral sequence, 243 very ne, 254
smooth component, 61
special module, 17 weight, 239
specially at, 24 weight ltration, 237, 239, 246
specially at resolution, 25 weight spectral sequence, 237, 239
specially injective, 18
specially injective resolution, 19 zariskian orientation sheaf, 81
Lecture Notes in Mathematics
For information about earlier volumes
please contact your bookseller or Springer
LNM Online archive: springerlink.com
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
Editorial Policy (for the publication of monographs)
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,
Ecole Normale Superieure de Cachan,

61 Avenue du President Wilson, 94235 Cachan Cedex, France
E-mail: Jean-Michel.Morel@cmla.ens-cachan.fr
Professor F. Takens, Mathematisch Instituut,
Rijksuniversiteit Groningen, Postbus 800,
9700 AV Groningen, The Netherlands
E-mail: F.Takens@math.rug.nl
Professor B. Teissier, Institut Mathematique de Jussieu,
UMR 7586 du CNRS, Equipe Geometrie et Dynamique,
175 rue du Chevaleret
75013 Paris, France
E-mail: teissier@math.jussieu.fr
For the Mathematical Biosciences Subseries of LNM:

Professor P.K. Maini, Center for Mathematical Biology,
Mathematical Institute, 24-29 St Giles,
Oxford OX1 3LP, UK
E-mail: maini@maths.ox.ac.uk
Springer, Mathematics Editorial I, Tiergartenstr. 17
69121 Heidelberg, Germany,
Tel.: +49 (6221) 487-8259
Fax: +49 (6221) 4876-8259
E-mail: lnm@springer.com

Weight Filtrations

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Weight Filtrations

Încărcat de

Drepturi de autor:

Formate disponibile

Lecture Notes in Mathematics 1959

Weight Filtrations on Log

ISBN: 978-3-540-70564-2 e-ISBN: 978-3-540-70565-9

Lecture Notes in Mathematics ISSN print edition: 0075-8434

Library of Congress Control Number: 2008932186

Mathematics Subject Classication (2000): 14F30 (Primary), 14F40, 13K05 (Secondary)

c 2008 Springer-Verlag Berlin Heidelberg

Cover design: SPi Publishing Services

Printed on acid-free paper

Tokyo Yukiyoshi Nakkajima

1 Preliminaries on Filtered Derived Categories and Topoi . . 15

2 Weight Filtrations on Log Crystalline Cohomologies . . . . . . 55

2.16 The Filtered Log Berthelot-Ogus Isomorphism . . . . . . . . . . . . . 202

3 Weight Filtrations and Slope Filtrations on Rigid

A Relative SNCD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

(1) Cohomologies in characteristic 0 of algebro-geometric objects in any

is obtained by counting the number of the logarithmic poles of local sections

Moreover we have the following isomorphisms in the ltered derived category

is equal to the multiplication by (1)k . Hence we use the following isomor-

instead of (0.0.0.4). The isomorphism (0.0.0.6) is equal to the isomorphism

(0.0.0.7) E1k,h+k = H hk ((Dan )(k) , (k) (Dan /C))(k) = H h (Uan , Z).

If X is proper, one obtains the weight ltration on H h (Uan , Z) by the spectral

in the classical crystalline topos (D(k) /S)

(0.0.0.8) E1k,h+k = Rhk fD(k) /S (OD(k) /S Z crys

Here crys (D/S) is the crystalline orientation sheaf of D/S in (D

Frobenius morphism F : (X, D) (X  , D ) over S0 induces the relative

as the identity under the natural identication

E1k,h+k = H hk ((D n crys , OD (k) /Wn Z crys (D/Wn ))(k)

To construct (0.0.0.8) and to prove the functoriality of it, we dene two

and construct two other ltered complexes

{D } be a decomposition of D by smooth compo-

(CRcrys (O(X,D)/S ), P )  (QX/S LX/S (X /S (log D)),

(Czar (O(X,D)/S ), P )  (OD OX X /S (log D), {OD OX Pk X /S (log D)}kZ ),

(0.0.0.21): If S0 is the spectrum of a perfect eld of characteristic p and if

is: if we can do something by using local lifts, we can do something analogous

(0.0.0.23) QX/S Rk (X,D)/S (O(X,D)/S )

Kato, we have a natural morphism

g : (Czar (O(X ,D )/S ), P ) Rg (Czar (O(X,D)/S ), P ).

Now we outline the contents of this book.

(Czar (O(X,D)/S ), P ) are independent of the choice of the decomposition of D

([28]), Tsuzukis proper descent ([87]), Shihos comparison theorems ([82])

Conventions. We assume that the log structures on (formal) schemes are

dq1+n dq+n dq+1+n

(2) For a complex (E , d ) of objects in A and for an integer n,

of bounded below complexes of objects in A, let (E [1], dE [1]) (F , dF ) be

(5) Let A be an abelian category with enough injectives. Let F : A B

of complexes of objects in A such that the upper horizontal sequence is exact,

(8) Let r 2 be a positive integer. As usual, an r-uple complex of objects

In this chapter, we prove some preliminary results on ltered derived

1.1 Filtered Derived Category. I

In this section we recall the denition of a ltered derived category and

Y. Nakkajima, A. Shiho, Weight Filtrations on Log Crystalline Cohomologies 15

HomA ((E, {El }), (F, {Fl })) MF((T , A))

is immediate to verify. Let f : (E, {El }) (F, {Fl }) be a morphism in

We recall the following denition due to Berthelot ([8]):

Next, we calculate the ltered derived functor of a morphism of ringed

Ias (A) := {(J, {Jk }) | J and Jk (k Z) are asque A-modules},

is surjective. By (1.1.0.1), it suces to prove that the morphism

is surjective. For each integer k, the induced morphism E/Ek1 F/Fk1

Denition 1.1.3. We say that a ltered module (J, {Jl }) MF(A) is

Proposition 1.1.4 ([8]). For a ltered module E = (E, {Ek }) MF(A),

Denition 1.1.5. Let (E , {Ek }) be an object of K+ F(A).

(1)p dp0 (1)p dp1

gives a ltered asque resolution s(I , {Ik }) of (F , {Fk }).

Frobenius morphism F : (X, D) (X , D ) over S0 induces the relative

(CRcrys (O(X,D)/S ), P ) (QX/S LX/S (X /S (log D)),

(Czar (O(X,D)/S ), P ) (OD OX X /S (log D), {OD OX Pk X /S (log D)}kZ ),

Note that the morphism K q F q F q+1 IF q+1 is a monomorphism. Let

s is a local section of ((A (Q, {Ql }) A E)k ).

Proof. (1.1.18) follows from (1.1.16) and [78, Lemma 1.3.3].

(2) Set Q := Q (A ), Qst (A ) or Qsp (A ). Let f : (T , A)

rived functor Lf : D F(A ) D F(A) such that Lf [(Q , {Qk })] =

(F , {Fl }))k corresponds to a morphism E F [n]k which is homotopic

Theorem 1.2.2 ([8], Adjunction formula). Let f : (T , A) (T , A )

in D+ F((T , A )). The isomorphism above satises the transitive condition

which enables us to nish the proof of (1.2.2).

(The third equality above follows from (1.2.0.1).)