arXiv:1410.3079v1 [math.AG] 12 Oct 2014
METRIZATION OF DIFFERENTIAL PLURIFORMS ON
BERKOVICH ANALYTIC SPACES
MICHAEL TEMKIN
Abstract. We introduce a general notion of a seminorm on sheaves of rings
or modules and provide each sheaf of relative differential pluriforms on a
Berkovich k-analytic space with a natural seminorm, called Kähler seminorm.
If the residue field e
k is of characteristic zero and X is a quasi-smooth k-analytic
space, then we show that the maximality locus of any global pluricanonical
form is a PL subspace of X contained in the skeleton of any semistable formal
model of X. This extends a result of Mustaţă and Nicaise, because the Kähler
seminorm on pluricanonical forms coincides with the weight norm defined by
Mustaţă and Nicaise when k is discretely valued and of residue characteristic
zero.
1. Introduction
1.1. Motivation. It often happens that a Berkovich space X possesses natural
skeletons, which are, in particular, deformational retracts of X of finite topological
type, see [Ber90, Section 4.3] and [Ber99]. In some special cases, such as the case
of curves or abelian varieties, there is a canonical (usually, minimal) skeleton. As
a rule, skeletons are obtained from nice formal models, e.g. poly-stable or, more
generally, log-smooth ones. Although, we should mention for completeness that
a different and very robust method of constructing skeletons was developed very
recently by Hrushovski and Loeser, see [HL12].
In [KS06], Kontsevich and Soibelman constructed a canonical skeleton of analytic
K3 surfaces over k = C((t)) by use of a new method: the skeleton is detected as the
extremality locus of the canonical form. In [MN13], this method was extended by
Mustaţă and Nicaise as follows. If k is discretely valued and X is the analytification
of a smooth and proper k-variety, they constructed norms on the pluricanonical
⊗n
sheaves ωX
and showed that the maximality locus of any non-zero pluricanonical
form φ is contained in the skeleton associated with any semistable formal model
of X. The union of the maximality loci of non-zero pluricanonical forms is called
the essential skeleton of X in [MN13]. It is an important ”combinatorial” subset
of X, see [NX13], though we should aware the reader that it does not have to be a
skeleton of X.
An advantage of the above approach is that it constructs a valuable combinatorial
subset, the essential skeleton, in a canonical way. In particular, the only input is the
metrization of pluricanonical sheaves, and no choice of a formal model is involved.
⊗n
Slightly ironically, one heavily exploits formal models to metrize ωX
. On the one
hand, existence of nice formal models is not needed, and the construction works
Key words and phrases. Berkovich analytic spaces, Kähler seminorms, pluriforms, skeletons.
This work was supported by the European Union Seventh Framework Programme (FP7/20072013) under grant agreement 268182.
1
2
MICHAEL TEMKIN
fine when char(e
k) > 0 (and existence of nice models is a dream). But on the other
hand, this still leads to technical restrictions, including the assumption that k is
discretely valued. In addition, the construction of the norm is not so geometric:
first one defines it at so-called divisorial points by use of formal models, and then
extends to the whole X by continuity.
The original aim of this project was to provide a natural local analytic construction of Mustaţă-Nicaise norm, which applies to all points on equal footing and
eliminates various technical restrictions of their method. The basic idea is very simple: provide ΩX with the maximal seminorm making the differential d : OX → ΩX
⊗n
a contracting map, and induce from it a seminorm on ωX
= (∧d ΩX )⊗n . Moreover,
the same definition makes sense for any morphism f : X → S, so the construction
generalizes to the relative situation and no assumption on k and f is needed. This
increases the flexibility even in the original setting; for example, one obtains a way
to work with analytic families of proper smooth varieties. Unfortunately, the simplicity of the basic idea is partially compensated by lack of various foundations. So,
a large part of this project is devoted to developing of basic topics, including a theory of Kähler seminorms on modules of differentials, its application to real-valued
fields, metrization of sheaves of modules, topological realization of analytic and PL
G-topologies, etc. On the positive side, we think that this foundational work will
be useful for other projects in non-archimedean geometry and related areas.
Once Kähler seminorms will have been defined, we will study the maximality lok) = 0 seems
cus of pluricanonical forms. Unfortunately, the assumption that char(e
unavoidable with current technique, but we manage to treat the non-discrete case
as well. In particular, we only use a result a la de Jong (see [Har03]) instead of the
existence of semistable model (a result a la Hironaka). Finally, under the assumptions of [MN13], we compare the Kähler seminorm to the Mustaţă-Nicaise norm on
the pluricanonical sheaves. Surprisingly, they coincide only when char(e
k) = 0, and
in general they are related by a factor which up to a constant coincides with the
log different of H(x)/k.
In order to get to the definition of Kähler seminorm on ΩX/S as soon as possible,
we split the project into two papers. Only a necessary minimum of foundations is
developed here, and a more systematic approach will be worked out in [Tem14].
1.2. Methods. At few places in the paper, including the definition of Kähler seminorms, we have to choose one method out of few possibilities. Let us discuss briefly
what these choices are, what is the way we choose in this paper, and which material
is postponed until [Tem14].
1.2.1. An approach via unit balls. One way to define a seminorm on a k-vector
space V is by using its unit ball V ⋄ , which is a k ◦ -module. Technically, this leads
⋄
to the easiest way to metrize coherent OX -sheaves: just choose a k ◦ -submodule OX
(subject to simple restrictions). For example, if the valuation is not discrete one can
simply define the Kähler seminorm k kΩ on ΩX/S as the seminorm associated with
◦
◦
◦
◦
the sheaf OX
dX/S (OX
) – the minimal OX
-submodule of ΩX/S containing dX/S OX
.
Unfortunately, this definition is problematic when the valuation is discrete and
breaks down completely when the valuation is trivial. In order to consider ground
fields with discrete or trivial valuations on the equal footing, we have to develop
all basic constructions in terms of seminorms themselves. For example, k kΩ can
be characterized as the maximal seminorm such that the map d : OX → ΩX/S is
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 3
contracting. The approach with unit balls will be established in [Tem14], although
some implicit use of unit balls is made in Section 5, see Theorem 5.1.11.
1.2.2. Seminormed algebras versus Banach algebras. Most seminormed rings in
Berkovich geometry are Banach. Nevertheless, more general seminormed rings
also show up, with the main example being the local rings OX,x and their residue
fields κ(x). For this reason, it is technically much more convenient to work with
seminormed rings and modules rather than their Banach analogues throughout the
paper. In addition, it turns out to be important to consider only contracting homomorphisms, while classical Banach categories contain all bounded homomorphisms.
1.2.3. Metrization of sheaves. There exist two ways to define seminorms on sheaves
of rings or modules. An ad hoc method, which is worked out in this paper, is to
provide each stalk with a seminorm so that the following condition (a version of
a sheaf condition) is satisfied: for a section s ∈ F (U ) let |s| : U → R+ denote
the function sending x ∈ U to |s|x , then all functions |s| should be upper semicontinuous. We stress that it is the semi-continuity that encodes the sheaf condition
correctly, and even the function kdtkΩ on the disc M(k{t}) is not continuous;
in fact, kdtkΩ is the radius function, see 6.2.1. Note for comparison, that the
definition of metrization in [CLD12] requires that all sections |s| are continuous.
As a drawback of our definition, we can only metrize sheaves on topological spaces,
so instead of working with the G-topology XG on an analytic space X, one has
to consider the associated topological space |XG |. This is a new tool in Berkovich
geometry, so some foundational work in that direction is also required.
Alternatively, one can develop a sheaf theoretic approach to metrization of
sheaves, which applies to any site. One should consider sheaves of seminormed
rings and modules, presheaves of seminorms on modules and their sheafifications,
etc. We make a first step in this direction in Section 3.2, but many details are
postponed until [Tem14].
1.3. Overview of the paper and main results. We collect basic results about
k-analytic spaces in Section 2. Our description of the topological space |XG | seems
to be new, see Section 2.2, though its strictly analytic analogue has been well
known for a long time. In particular, we interpret the points of |XG | in terms of
the graded reductions of germs. Also, we interpret points of the PL topologies in
terms of combinatorial valuations (or valuations on lattices), see Section 2.6, and
for a PL subspace P ⊆ X describe the embedding |PG | ֒→ |XG |. This is used to
prove a strong result on the structure of a PL subspace P : locally P possesses an
unramified chart, see Theorem 2.6.8. It seems that our usage of the space |PG |
is more-or-less equivalent to the use of model theory in [Duc12] and [DT14], see
Remark 2.6.3.
Section 3 is devoted to metrization of sheaves of rings and modules. In particular,
a seminormed sheaf of modules is defined by providing each stalk with a seminorm,
see 3.2.1, a sheaf of seminormed modules is defined in 3.2.3, and equivalence of the
two notions is established in Lemma 3.2.5.
In Section 4.1, we extend the theory of Kähler differentials to seminormed rings.
In particular, given a contracting homomorphism of seminormed rings A → B
we show that dB/A : B → (ΩB/A , k kΩ ) is the universal contracting A-derivation,
where the Kähler seminorm k kΩ is defined as the maximal seminorm making the
homomorphism dB/A contracting.
4
MICHAEL TEMKIN
In Section 5 we study the Kähler seminorm on a vector space ΩK/A , where K is
a real-valued field. In Theorem 5.1.11 we show that Ωlog
K ◦ /A◦ modulo its torsion is
the unit ball of the Kähler seminorm on ΩK/A . Thus, study of Kähler seminorms
is tightly related to study of Ωlog
K ◦ /A◦ and the ramification theory. Although this is
classical in the discretely-valued case, [GR03, Chapter 6] is the only reference for
such material when |K × | is dense. Unfortunately, even loc.cit. does not cover all
our needs, so we have to dig into the theory of real-valued fields, that makes this
section the most technical in the paper. One of our main results there is that if
b K/A f
b L/A , see Theorem 5.5.5 and its
K is dense in a real-valued field L then Ω
→Ω
corollary.
In Section 6 we define Kähler seminorm k kΩ on ΩX/S for any morphism X →
b κ(x)/κ(s) = Ω
b H(x)/H(s) by Theorem 5.5.5,
S. The main ingredient here is that Ω
and hence the Kähler seminorms at the stalks ΩX/S,x satisfy the semicontinuity
condition. Some examples of Kähler seminorms are described in Section 6.2. In
particular, we show that Kähler seminorm behaves weird when k possesses wildly
ramified extensions. Also, we study compatibility of k kΩ with base changes in 6.3.
In particular, it is compatible with restricting onto the fibers and tame extensions
of the ground field (Theorem 6.3.9 and its corollaries), but is incompatible with
wild extensions of the ground field.
Finally, we use k kΩ to metrize the sheaves of pluriforms on a quasi-smooth
analytic space X. The obtained seminorms are also called Kähler, and we study
their properties in Section 7. In Corollary 7.1.3 we obtain a very simple formula that
evaluates Kähler seminorms at monomial points, and we deduce in Theorem 7.1.6
that the restriction of Kähler seminorms onto PL subspaces of X are PL. The main
ingredient here is Theorem 2.6.8. In Theorem 7.2.7 we establish the connection
between Kähler norm on pluricanonical sheaves and the weight norm of Mustaţă
and Nicaise. Finally, in Section 7.3 we study the maximality loci of a non-zero
pluricanonical form with respect to the geometric Kähler norm. We prove that it is
contained in the essential skeleton of X, see Theorem 7.3.4, and, if char(e
k) = 0, it is
e
a PL subspace of X, see Theorem 7.3.9. When char(k) = 0 this extends the results
of Mustaţă and Nicaise to the case of non-discrete |k × | and arbitrary quasi-smooth
X, not necessarily algebraizable or even strictly analytic. (If char(e
k) > 0, our norm
differs from the weight norm.)
1.4. Acknowledgements. The author is grateful to J. Nicaise for answering questions related to [MN13], to A. Ducros for showing his work with Thuillier [DT14]
and answering various questions, and to V. Berkovich, I. Tyomkin and Y. Varshavsky for useful discussions.
Contents
1.
2.
3.
4.
5.
6.
7.
Introduction
k-analytic spaces
Seminorms ad hoc
Differentials of seminormed rings
Kähler seminorms of real-valued fields
Metrization of ΩX/S
Metrization of pluricanonical forms
1
5
18
24
27
39
45
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 5
51
References
2. k-analytic spaces
In this section we fix notation and summarize various material on k-analytic
spaces that will be used throughout the paper.
2.1. Conventiones.
2.1.1. The ground field. Throughout this paper, k is a non-archimedean analytic
field, i.e., k is provided with a non-archimedean real valuation | | : k → R+ with
respect to which k is complete. Trivial valuation is allowed. If the valuation is not
trivial, π may denote any element of k ◦◦ \ {0}. If the valuation is trivial then π = 0.
In particular, the natural topology on K ◦ is the (π)-adic one.
2.1.2. The residue characteristic. By p = exp.char(k) we denote the exponential
characteristic of e
k, i.e., p = 1 when e
k contains Q.
2.1.3. H-strict k-analytic spaces. All analytic spaces we will consider are k-analytic
spaces in the sense of [Ber93, §1]. In addition we fix a divisible subgroup H ⊆ R×
+
such that H 6= 1 and |k × | ⊆ H and consider only H-strict analytic spaces in the
sense of [CT]. For shortness, we will often call them analytic spaces.
2.2. G-topology and its topological realization.
2.2.1. The G-topology. The usual topology of an analytic space X can be used
for working with coherent sheaves only when X is good. In general, one has to
work with the G-topology of analytic domains: objects of the site XG are H-strict
analytic domains of X and coverings are set-theoretical coverings U = ∪i∈I Ui such
that {Ui } is a quasi-net on U in the sense of [Ber90, §1.1]. Note that the G-topology
depends on H, but since H is fixed, we will not mention it in the notation.
2.2.2. The space |XG |. We refer to [vdPS95, p. 83, (p1)–(p3),(p4)’] for the definition of prime filters on G-topological spaces. By a point x of XG we mean a
prime filter {Ui } of analytic domains. Informally speaking, this is the prime filter
of all analytic domains ”containing” x. We denote by |XG | the set of all points of
XG . For any analytic domain U ⊆ X we have an obvious embedding |UG | ֒→ |XG |.
We provide |XG | with the topology whose base is formed by all sets of the form
|UG |. Obviously, any sheaf F on XG extends to a sheaf F ′ on |XG | by setting
∼
F ′ (|UG |) = F (U ) and sheafifying, so we obtain a functor αX : XG
→ |XG |∼ between the associated topoi, where, as in [sga72a], given a site C we denote by
C ∼ the topos of sheaves of sets on C. The stalk of F ′ at x ∈ |XG | is simply
colimx∈|(Ui )G | F (Ui ). For shortness, we will denote this stalk as Fx .
Remark 2.2.3. We refer to [sga72b, IV.6] for the definition of points of a general
topos in terms of fibred functors. It is easy to see that for G-topological spaces this
definition agrees with our definition given in terms of prime filters.
6
MICHAEL TEMKIN
2.2.4. Abundance of points. Since any point of X possesses a compact neighborhood
and any compact analytic space is quasi-compact in the G-topology, the site of X is
∼
has enough points
locally coherent in the sense of [sga72b, VI.2.3]. Therefore, XG
by Deligne’s theorem, see [sga72b, VI.9.0].
Theorem 2.2.5. For any k-analytic space X the topological space |XG | is sober
∼
and the functor αX : XG
→ |XG |∼ is an equivalence of categories.
Proof. In view of [sga72a, IV.7.1.9] and [sga72a, VI.7.1.6], it suffices to show that
∼
XG
is generated by subsheaves of the final sheaf 1X . For any analytic subdomain
U ⊆ X, let LU ⊆ 1X denote the extension of 1U , i.e. LU (V ) = {1} if V ⊆ U and
LU (V ) = ∅ otherwise. If F is a sheaf on X then any section s ∈ F`(U ) induces a
morphism LU → F . In particular, we obtain an epimorphism φ : s,U LU → F ,
where U runs over all analytic subdomains and s runs over F (U ).
2.2.6. Ultrafilters. Our next aim is to classify the points of XG and we start with
those corresponding to the maximal prime filters.
Lemma 2.2.7. The prime filter Px associated with a point x ∈ X is maximal, and
any maximal prime filter is of the form Px . In particular, we obtain an embedding
X ֒→ |XG | whose image consists of all points that have no non-trivial generization.
Proof. If Px is not maximal then it can be increased, and hence there exists an
affinoid domain not contained in Px but having non-empty intersections with all
elements of Px , which is an absurd. Conversely, assume that P is a maximal prime
filter and fix an affinoid V ∈ P. If P is not of the form Px with x ∈ V then for any
x ∈ V we can find an affinoid domain Vx ⊂ V with x ∈
/ Vx . Then ∩x∈V Vx = ∅, and
hence already the intersection of finitely many sets Vx is empty. This contradicts
P being a filter.
In the sequel we will identify X with its image in |XG |.
Corollary 2.2.8. Any point z ∈ |XG | possesses a unique generization r(z) lying
in X.
Proof. We should prove that any prime filter is contained in a single filter of the
form Px . One such Px exists by Lemma 2.2.7. Assume that P is contained in
Px and Py with x 6= y. Choose any V ∈ P and find an admissible covering
V = Vx ∪ Vy such that x ∈
/ Vx and y ∈
/ Vy . Since either Vx or Vy lies in P, we obtain
a contradiction.
2.2.9. The retraction. By Corollary 2.2.8 we obtain a retraction rX : |XG | → X
given by x 7→ r(x). For each x ∈ X, the fiber r−1
X (x) is the set of all specializations
of x in |XG |, equivalently, r−1
(x)
is
the
closure
of x in |XG |.
X
Theorem 2.2.10. Let X be a k-analytic space, U ⊆ X a k-analytic subdomain
and x ∈ U a point. Then,
(i) rX : |XG | → X is a topological quotient map and X is the maximal Hausdorff
quotient of |XG |.
−1
(ii) U is a neighborhood of x if and only if r−1
U (x) = rX (x). In particular, U is
−1
open if and only if |UG | = rX (U ).
Proof. We start with (ii). For an analytic domain x ∈ V ⊆ X let (V, x) denote the
germ of V at x. Define a presheaf of abelian groups on |XG | as follows: F (V ) is
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 7
either 0 or Z, and the second case takes place if and only if x ∈ V and (V, x) is not
contained in (U, x), i.e. for any neighborhood W of x one has that W ∩ V * W ∩ U .
The restriction maps are either isomorphisms or the map Z → 0. If V1 , . . . ,Vn are
domains containing x and satisfying ∪ni=1 (Vi , x) = (V, x), then (V, x) * (U, x) if
and only if (Vi , x) * (U, x) for some i. It follows that the presheaf F is separated
and hence the sheafification map F → F = αF is injective by [sga72a, II.3.2]. In
particular, F = 0 if and only if F = 0. Obviously, F = 0 if and only if U is a
neighborhood of x.
On the other hand, the stalks of F are given by Fz = colimW F (W ) where W
run through the domains with z ∈ |WG |. In particular, Fz is either 0 or Z and the
−1
second possibility holds if and only if z ∈ r−1
X (x) \ rU (x). Thus, F = 0 if and only
−1
−1
if rU (x) = rX (x), and we obtain (ii).
−1
Let us prove (i). If U is open then r−1
X (U ) = |UG | by (ii). Thus, rX (U ) is
open and hence rX is continuous. Conversely, assume that U is not open. Then
U is not a neighborhood of a point x ∈ U , and using (ii) we can choose a point
−1
−1
z ∈ r−1
X (x) \ rU (x). We claim that rX (U ) is not a neighborhood of z. Indeed, if
it is a neighborhood then there exists an analytic domain W such that z ∈ |WG | ⊂
r−1
X (U ). Then x ∈ W and (W, x) * (U, x), in particular, there exists a point
y ∈ W \ U . Thus, y ∈ |WG | and y ∈
/ r−1
X (U ), a contradiction. This proves that
−1
rX (U ) is not open, and hence rX is a topological quotient map. Furthermore, X is
the maximal Hausdorff quotient of |XG | because any Hausdorff quotient |XG | → Z
should identify each point x ∈ X with any its specialization z ∈ r−1
X (x).
2.2.11. Notation XG . Starting with this point we will not distinguish the site XG
and the topological space |XG |. In particular, we will write x ∈ XG instead of
x ∈ |XG |.
2.2.12. Non-analytic points. Our next aim is to describe the non-analytic or infinitesimal points of XG , i.e. the points of XG \ X. For this we have to recall some
results about reductions of germs.
2.2.13. Germ reduction. By AeH we denote the H-graded reduction ⊕h∈H Aeh . In
[CT, Section 7] and [Duc14, Section 1.5], to any germ (X, x) of an H-strict an^
alytic space at a point one associates an H-graded reduction (X,
x)H , which is
an H-graded Riemann-Zariski space associated to the extension of H-graded fields
] /e
]
H(x)
H kH . In particular, any such point induces a graded valuation on H(x)H and
if (X, x) is separated, it is also determined by this valuation.
Theorem 2.2.14. Let X be an H-strict k-analytic space and x ∈ X a point. Then
^
the closure of x is canonically homeomorphic to (X,
x)H .
Proof. Let L denote the set-theoretical lattice of subdomains (U, x) of (X, x).
Clearly, the closure of x is a sober topological space and L is its topology base.
On the other hand, subdomains of (X, x) are in a one-to-one correspondence with
^
quasi-compact open subspaces of (X,
x)H by [Tem04, Theorem 4.5] and [CT, Theorem 7.5]. So, L is also the lattice of a topology base of the sober topological
^
space (X,
x)H . Since, a sober topological space is determined by such a lattice (it
can be reconstructed as points of the corresponding topos), we obtain the asserted
homeomorphism.
8
MICHAEL TEMKIN
◦
2.3. Stalks of OXG and OX
. Our next aim is to describe the stalks of the sheaves
G
◦
OXG and OXG at non-analytic points. In particular, this will lead to an explicit
description of the homeomorphism from Theorem 2.2.14.
2.3.1. Spectral seminorm. We provide each stalk OXG ,x with the spectral seminorm
| |x , i.e. |s|x = inf x∈UG |s|U , where | |U is the spectral seminorm of U .
Lemma 2.3.2. Let X be an analytic space and let x ∈ XG be a point. Then,
(i) The seminorm | |x is a semivaluation and OXG ,x is a local ring whose maximal
ideal is the kernel of | |x .
(ii) If y ∈ XG generizes x then the generization homomorphism φx,y : OXG ,x →
OXG ,y is an isometry with respect to | |x and | |y . In particular, φ is local.
For the sake of comparison we note that in the case of schemes any non-trivial
generization is not local.
Proof. Let z = rX (x) be the maximal generization of x. We claim that |s|x = |s(z)|
for any s ∈ OXG ,x . If U is a domain with x ∈ UG then z ∈ U and so |s|x ≥ |s(z)|.
Conversely, set r = |s(z)| and note that Xε = X{|s| ≤ r + ε} is a neighborhood
of z for any ε > 0. Since x ∈ (Xε )G we obtain that |s|x ≤ r + ε for any ε, and
so |s|x ≤ |s(z)|. In other words, φx,z is an isometry. In the same way, φy,z is an
isometry, and therefore φx,y is an isometry.
It remains to prove (i). Since φx,z is an isometry and | |z is multiplicative, it
follows that | |y is multiplicative. If |s|x = 0 for s ∈ OXG ,x then the image of s in
OXG ,z is not invertible by the above, and hence s is not invertible. Conversely, if
|s|x = r > 0 then |s(z)| > 0 and hence there exists a neighborhood U of z such
that s ∈ OXG (U )× . Since x ∈ UG , we obtain that s is invertible.
2.3.3. Residue fields. Once we know that OXG ,x is local, we denote its maximal
ideal by mG,x . The residue field will be denoted κG (x) = OXG ,x /mG,x. Since mG,x
is the kernel of | |x , the residue field acquires a real valuation and by H(x) we denote
its completion. Let us compare these notions with their analogues introduced by
Berkovich in [Ber90] and [Ber93].
Remark 2.3.4. (i) For a good space X, one uses the usual topology of X to define
the local ring OX,x , its maximal ideal mx and the residue field κ(x). In addition,
one defines the completed residue field H(x) for an arbitrary space X using the
d is preserved when passing to an affinoid
fact that if X is affinoid then H(x) = κ(x)
subdomain containing x.
(ii) We use the notation mG,x and κG (x) to avoid a confusion with mx and κ(x)
when X is good. It is easy to see that H(x) = κ\
G (x), so the notation HG (x) is
ambiguous and will not be used.
2.3.5. Generization homomorphisms. Recall that by Lemma 2.3.2(ii), any generization homomorphism φx,y : OXG ,y → OXG ,x induces an embedding of real-valued
fields κG (y) ֒→ κG (x).
Lemma 2.3.6. If y ∈ XG generizes x ∈ XG then the induced embedding κG (x) ֒→
κG (y) has dense image and so H(x) = H(y).
Proof. Note that we can replace X with an analytic domain X ′ such that x ∈
′
XG
. In particular, we can assume that X is affinoid. Let z ∈ X be the maximal
generization of y and x. The local embeddings OX,z ֒→ OXG ,x ֒→ OXG ,y ֒→ OXG ,z
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 9
induce embeddings of fields κ(z) ֒→ κG (x) ֒→ κG (y) ֒→ κG (z). It remains to use
that κ(z) ֒→ κG (z) has a dense image.
◦
2.3.7. Stalks of OX
. As in [Tem11, Section 2.1], by a semivaluation ring A with
G
semifraction ring B we mean the following datum: a local ring (B, m) and a subring
A ⊆ B such that m ⊂ A and A/m is a valuation ring of k = B/m. Such a datum
defines an equivalence class of semivaluations ν : B → Γ ∪ {0} whose kernel is m,
and A = ν −1 (Γ≤1 ∪ {0}) is the ring of integers of ν.
◦
Lemma 2.3.8. For any x ∈ XG , the stalk OX
is a semivaluation ring with
G ,x
◦
semifraction ring OXG ,x . In particular, OXG ,x /mG,x is a valuation ring of H(x).
Proof. If s ∈ mG,x then |s|x = 0 and hence |s| < 1 in a neighborhood of x. In
×
◦
particular, s ∈ OX
. It remains to show that if u ∈ OX
then either u or u−1
G ,x
G ,x
◦
lies in OXG ,x . Shrinking X we can assume that u ∈ OXG (X)× . Then X is the
union of Y = X{u} and Z = X{u−1 }, hence x lies in either YG or ZG , and then
◦
◦
u ∈ OX
or u−1 ∈ OX
, respectively.
G ,x
G ,x
◦
2.3.9. Valuation νx . Let νx denote both the semivaluation induced by OX
on
G ,x
OXG ,x and the valuation induced on H(x). If x ∈ X then an element s ∈ OXG ,x lies
◦
◦
in OX
if and only if |s|x ≤ 1. Thus, OX
/mG,x = H(x)◦ , i.e. νx is the usual
G ,x
G ,x
◦
real valuation of H(x). If x is arbitrary, we only have an inclusion OX
/mG,x ֒→
G ,x
◦
H(x) , so νx is composed from the real valuation of H(x) and the residue valuation
]
νex on H(x).
Remark 2.3.10. (i) The construction of νex can be extended by associating to
] whose degree-1 component is O◦
x a graded valuation ring of H(x)
XG ,x /mG,x ,
H
and the argument is essentially the same. Since we will not need this in the
sequel, let us just formulate the result. Set O = OXG for shortness and let
Or◦ and Or◦◦ be the subsheaves of O whose sections on U satisfy |s|U ≤ r and
|s|U < r, respectively. Then Ax = ⊕h∈H (Oh◦ )x /(Oh◦◦ )x is a graded valuation ring
] = ⊕h∈H (Ox )◦ /(Ox )◦◦ that coincides with the graded valuation ring
of H(x)
h
h
H
^
induced by the image of x under the homeomorphism r−1
X (y) = (X, y)H from Theorem 2.2.14, where y ∈ X is the maximal generization of x. In particular, x ∈ X if
] , i.e. the graded valuation is trivial.
and only if Ax = H(x)
H
p
(ii) Assume that H = |k × | and hence XG is the usual strictly analytic Gtopology. Then the H-graded reduction coincides (as a topological space) with the
^
ungraded reduction (X,
x) because taking the degree-1 components provides a one] and valuation
to-one correspondence between graded valuation e
kH -rings in H(x)
H
] In particular, x is determined by its maximal generization y and
e
k-rings in H(x).
^
a point of (X,
y), or, that is equivalent, x is determined by the valuation νx . This
implies that XH coincides with the Huber’s adic space X ad corresponding to X,
and x ∈ X if and only if νx is of height one. (The latter is not true for a larger
H because the graded residue valuation can be non-trivial even when its degree-1
component νex is trivial.)
2.3.11. Topologically unramified locus. Assume that f : Y → X is a quasi-étale
morphism. We say that f is topologically unramified or topologically semi-tame at
y ∈ Y if the extension H(y)/H(f (y)) is unramified or tame, respectively.
10
MICHAEL TEMKIN
Remark 2.3.12. This is not essential for this paper, but we reserve the notion of
topological tameness to the case when [H(y) : H(f (y))] is invertible in e
k.
Theorem 2.3.13. Assume that f : Y → X is a quasi-étale morphism, y ∈ Y is a
point and y ′ ∈ YG is a specialization of y. Set x = f (y) and x′ = f (y ′ ) and provide
] and K = H(x)
] with the valuations νey′ and νex′ corresponding to
the fields L = H(y)
y ′ and x′ . Assume that the extensions H(y)/H(x) and L/K are unramified. Then
there exists an analytic domain U ⊆ Y such that y ′ ∈ UG and f is topologically
unramified at any point of U .
Proof. The basic idea is to use Chevalley’s description of étale homomorphisms, see
[Gro67, IV4 , 18.4.6].
Step 1. We can assume that X = M(A) is affinoid and f is finite étale. The
question is G-local at x′ and y ′ hence we can assume that X and Y are affinoid.
Also, it follows easily from [Ber93, Theorem 3.4.1] that replacing X and Y with
neighborhoods of x and y, we can achieve that f factors as Y ֒→ Y → X, where
Y is an affinoid domain in Y and Y → X is finite étale (see [DT14, 3.12]). So,
replacing Y with Y we can assume that f is finite étale.
Step 2. We can assume that Y = M(B) where B = A[u] and if v ∈ L denotes
the image of u in L then L◦ is a localization of K ◦ [v]. By our assumption, L◦ /K ◦
is étale, hence [Gro67, IV4 , 18.4.6] implies that L◦ is the localization of a ring
A = K ◦ [T ]/(g(T )) at its maximal ideal m such that g(T ) is a monic polynomial
and g ′ (v) ∈ (Am )× , where v ∈ Am is the image of T . Since L is a localization of
K[T ]/(g(T )), we have that g = g1 g2 where (g1 , g2 ) = 1 and L = K[T ]/(g1(T )).
Clearly, we can replace g with g1 making it irreducible over K.
g = K and OX,x ։ κ(x), we can choose a monic lift φ[T ] ∈ OX,x [T ]
Since κ(x)
of g(T ) ∈ K[T ]. (Note that X is good and we use the usual local ring OX,x
and residue field κ(x).) Shrinking X around x we can assume that φ(T ) ∈ A[T ].
Set B = A[T ]/(φ(T )) and Y ′ = M(B). Then Y ′ → X is a finite map of degree
d = deg(φ) and the preimage of x is a single point y ′ such that H(y ′ )/H(x) is the
unramified extension whose residue field extension is L/K. Therefore, Y ′ → X
is étale at y ′ and the maps of germs (Y ′ , y ′ ) → (X, x) and (Y, y) → (X, x) are
isomorphic by [Ber93, Theorem 3.4.1]. In particular, after shrinking X around x
the morphism Y ′ → X becomes étale, and then we can replace Y with Y ′ . Then
the image u ∈ B of T is as required.
Step 3. The domain U = Y {φ′ (u)−1 } is as required. Set w = φ′ (u). By the
] equals to g ′ (v) ∈ L. Thus,
construction, w(y) lies in H(y)◦ and its reduction w(y)
◦
−1
] is invertible in L , hence w(y) lies in the valuation ring of H(y) defined by
w(y)
y ′ and therefore w−1 ∈ OY◦ G ,y′ . This proves that y ′ ∈ UG .
Now, let z ∈ U , F = H(z), q = f (z) and E = H(q). Clearly, F = E[u] and the
minimal polynomial of u over E is a divisor ψ(t) of φ(t), say φ(t) = ψ(t)λ(t). By the
definition of U , the element φ′ (u) is invertible in F ◦ . Using that ψ(u) = 0 we obtain
that φ′ (u) = ψ ′ (u)λ(u), and hence ψ ′ (u) is invertible in F ◦ and C = E ◦ [u, ψ ′ (u)−1 ]
is an E ◦ -subalgebra of F ◦ . By [Gro67, IV4 , 18.4.2(ii)], C is étale over E ◦ . It follows
that C is normal, and since C ⊆ F ◦ and Frac(C) = F we necessarily have that
F ◦ = C. Thus, F/E is unramified, as required.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 11
Corollary 2.3.14. Keep the assumptions of Theorem 2.3.13 except the condition
on H(y)/H(x), and assume only that the latter extension is tame. Then there exists
U ⊆ Y such that y ′ ∈ UG and f is topologically semi-tame at any point of U .
Proof. [Ber93, Theorem 3.4.1] implies that after shrinking Y at y ′ we can split
g
h
f into a composition Y → Z → X such that H(y)/H(z) is purely ramified and
H(z)/H(x) is unramified, where z = g(y). It then suffices to prove the assertion for
g and h. The second case is covered by Theorem 2.3.13, so we are reduced to the
case when H(y)/H(x) is purely ramified. It is enough to establish this case after an
unramified extension of k, hence we can assume that e
k is separably closed. Then,
H(y) = H(x)[t]/(tn − a), where a ∈ H(x) and n is invertible in e
k. As in the proof
of Theorem 2.3.13, we can assume that f is finite. Then [Ber93, Theorem 3.4.1]
implies that shrinking X around x we can achieve that X = M(A), Y = M(B),
B = A[t]/(tn − φ) and φ ∈ A× satisfies φ(x) = a. Obviously, f is topologically
semi-tame at any point of Y .
2.4. Invariants of a point. In this section we recall basic results about invariants
t and s associated to points of k-analytic spaces, see [Ber90, §9]. To stress the
valuative-theoretic origin of these invariants we prefer to denote them F and E,
the transcendental analogues of e and f .
2.4.1. Invariants of extensions of valued fields. To any extension of valued fields
L/K one can associate two cardinals that measure the ”transcendence size” of the
e
extension: the residual transcendence degree FL/K = tr.deg.K
e (L) and the rational
×
×
rank EL/K = dimQ (|L |/|K | ⊗Z Q). Obviously, both invariants are additive in
towers of extensions L′ /L/K.
2.4.2. Invariants of points. For a point x of a k-analytic space X we define the
residual transcendence degree FX,x = FH(x)/k and the rational rank EX,x = EH(x)/k .
Usually, we will omit the space X in this notation. Additivity of the invariants can
now be expressed as follows. Assume that f : X → Y is a morphism of k-analytic
spaces, x ∈ X is a point with y = f (x), and Z = X ×Y M(H(x)) is the fiber over
y. Then FX,x = FY,y + FZ,x and EX,x = EY,y + EZ,x . In particular, if f is finite
then FX,x = FY,y and EX,x = EY,y .
2.4.3. Classification of points on a curve. Starting with [Ber90, §1.4.4] and [Ber93,
ca , (2)
§3.6], points of k-analytic curves are classified to four types: (1) H(x) ⊆ k
Fx = 1, (3) Ex = 1, (4) the rest. In all cases, it is easy to see that Ex + Fx ≤ 1,
i.e., Ex = 0 for type 2 points, Fx = 0 for type 3 points, and Ex = Fx = 0 for type
1 and 4 points.
] is finitely
If x is of type 2 or 3 then the following three claims hold: H(x)
generated over e
k, |H(x)× | is finitely generated over |k × |, if k is stable then H(x)
is stable. The first two are simple, and we refer to [Tem10, Theorem 6.3.1(iii)] for
the stability theorem. Note that all three claims fail for types 1 and 4.
2.4.4. Monomial points. Fibering X by curves and using the additivity of E and
F and induction on dimension, it is easy to see that any point x ∈ X satisfies the
inequality Ex + Fx ≤ dimx (X). A point x ∈ X is called monomial or Abhyankar if
Fx + Ex = dimx (X). The set of all monomial points of X will be denoted X mon.
12
MICHAEL TEMKIN
Remark 2.4.5. (i) Monomial points are adequately controlled by the invariants
Ex and Fx . This often makes the work with them much easier than with general
points. For example, see Corollary 2.5.5 below.
(ii) Analogues of monomial points in the theory of Riemann-Zariski spaces are
often called Abhyankar valuations. They are much easier to work with too; for
example, local uniformization is known for such points.
2.5. PL subspaces. The set X mon is huge and, at first glance, may look a total
mess when dim(X) > 1. Nevertheless, it possesses a natural structure of an ind-PL
space that we are going to recall.
2.5.1. The model case. We will use the underline to denote tuples, e.g.
(r1 , . . . ,rn ). Recall that points of the affine space
r =
Ank = ∪r M(k{r−1 t1 , . . . ,r−1 tn })
with coordinates t1 , . . . ,tn can be identified with real semivaluations on k[t1 , . . . ,tn ]
that extend the valuation of k. The semivaluations that do not vanish at t1 , . . . ,tn
n
form the open subspace T = Gnm . For each tuple r ∈ (R×
+ ) the formula
X
|
ai ti |r = max |ai |r i
i∈Nn
i
defines a valuation pr called generalized Gauss valuation, and the correspondence
n
r 7→ pr provides a topological embedding i : (R×
+ ) ֒→ T. Let S be the image of i.
Remark 2.5.2. (i) One often calls S the skeleton of T; this terminology is justified
by (ii) and (iii) below. Sometimes one refers to the points of S as t-monomial
Qn valuations because they are determined by their restriction onto the monoid i=1 tN
i .
(ii) Any semivaluation x ∈ T is dominated by the t-monomial valuation p|t(x)| .
The map x 7→ p|t(x)| is a retraction r : T ։ S. Moreover, Berkovich constructs in
[Ber90, §6] a deformational retraction of T onto S whose level at 1 is the above
map.
(iii) All t-monomial valuations are distinguished by invertible functions (in fact,
by monomials ta ), hence they are incomparable with respect to the domination.
Thus, S is the set of all points of T that are maximal with respect to the domination.
In particular, it is independent of the coordinates.
2.5.3. Monomial charts. By a monomial chart of an analytic space X we mean a
morphism f : U → Gnm such that U is an analytic domain in X and f has zerodimensional fibers. Clearly, f is determined by invertible functions t1 , . . . ,tn ∈
×
OX
(U ). A point x ∈ U of the chart is called t-monomial or f -monomial if the
restriction of | |x onto k[t1 , . . . ,tn ] is monomial. In this case, we also say that
d is
t1 , . . . ,tn is a family of monomial parameters at x. Note that H(f (x)) = k(t)
provided with a generalized Gauss valuation and H(x) is its finite extension.
In addition, we say that the chart f and the family of monomial parameters
t1 , . . . ,tn are tame or unramified at x if the extension H(x)/H(f (x)) is so. The
following result shows that monomial charts adequately describe the whole X mon .
Lemma 2.5.4. A point x ∈ X is monomial if and only if there exists a monomial
chart f : U → Gnm such that x is f -monomial. Moreover, if k is algebraically closed
then there exists a monomial chart which is unramified at x.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 13
Proof. Set L = H(x). If a monomial chart exists then the induced map f : U → Gnm
has zero-dimensional fibers and hence L is finite over K = H(y), where y = f (x).
Since Ey + Fy = n, we obtain that Ex + Fx = n, i.e. x is monomial.
Conversely, assume that the sum of E = Ex and F = Fx equals to n = dimx (X).
e e
Choose t1 , . . . ,tF ∈ κG (x) so that e
t1 , . . . ,e
tF is a transcendence basis of L/
k, and
choose {tF +1 , . . . ,tn } so that its image in (|L× |/|k × |) ⊗ Q is a basis. Then the
valuation on K = k(t1\
, . . . ,tn ) is a generalized Gauss valuation. Take an analytic
domain x ∈ U ⊆ X such that dim(U ) = n and t induces a morphism f : U → Gnm .
Then y = f (x) is a monomial point of Gm
n (even a point of its skeleton). Consider
the fiber Uy = f −1 (y). By [Duc14, Corollary 8.4.3] dimy (Uy ) = 0, hence using
[Duc07, Theorem 4.9] we can shrink U around y so that f has zero-dimensional
fibers.
e is perfect and |L× | is divisible and
Finally, if k is algebraically closed then L
×
×
n
hence |L |/|k | = Z . Thus, we can choose ti so that e
t1 , . . . ,e
tF is a separable
e e
transcendence basis of L/
k and the images of tF +1 , . . . ,tn form a basis of |L× |/|k × |.
e K
e is separable, and since K is stable (see Remark 2.4.5),
Then |L× | = |K × | and L/
it follows that L/K is unramified.
Corollary 2.5.5. Assume that x ∈ X is a monomial point, then
(0) The ring OXG ,x is Artin. In particular, if X is reduced then mG,x = 0 and
OXG ,x = κG (x).
] k
e is finitely generated.
(1) The extension H(x)/
(2) The group |H(x)× |/|k × | is finitely generated.
(3) If k is stable then H(x) is stable.
Proof. By Lemma 2.5.4 there exists a chart f : U → Gnm that takes x to a point
y corresponding to a generalized Gauss valuation. Note that OXG ,x is finite over
OYG ,y . All four properties are satisfied by y, hence they also hold for x.
2.5.6. Skeletons of monomial charts. The set of all f -monomial points of U will be
denoted S(f ) and called the skeleton of the chart. Note that S(f ) is nothing else
but the preimage of the skeleton of Gnm under f . By Lemma 2.5.4, X mon = ∪f S(f ),
where f runs over all monomial charts of X. We aware the reader that, in general,
S(f ) does not have to be a retract of U .
2.5.7. RS -PL structures. We will usually abbreviate piecewise linear as PL. We
refer to [Ber04, §1] for the definition of an RS -PL spaces Q for a ring R ⊆ R and
its exponential module S ⊆ R×
+ . Here we only recall that Q has an atlas {Pi }
e1
n
en
of RS -PL polytopes, i.e. polytopes in (R×
+ ) given by inequalities st1 . . . tn with
s ∈ S, ei ∈ R and provided with the family of RS -PL functions. In fact, we will
mainly use RS from the following list: R = ZH , RQ = QH , R0 = Z|k× | .
2.5.8. Rational PL-subspaces. Absolute values of the coordinates of Gnm induce
coordinates ti : S → R×
up to the action of
+ on its skeleton S. The latter are unique
Q
GL(Z, n) ⋉ |k × | combined from the action of GL(Z, n) on ni=1 tZ
i and rescaling the
coordinates by elements of |k × |. Therefore, S acquires a canonical R0 -PL structure.
The following facts were proved by Ducros: (1) for any monomial chart f : U →
Gnm the skeleton S(f ) possesses a unique RQ -PL structure such that the map
S(f ) → S is RQ -PL, see [Duc03, Th. 3.1], (2) for any two monomial charts f and
g the intersection S(f ) ∩ S(g) is RQ -PL in both S(f ) and S(g), see [Duc12, Th.
14
MICHAEL TEMKIN
5.1], and so S(f ) ∪ S(g) acquires a natural RQ -PL structure and the whole X mon
acquires a natural structure of an ind-RQ -PL space. By a rational RQ -PL subspace
of X we mean a subset of the form ∪ni=1 S(fi ) with its induced RQ -PL structure.
2.5.9. Integral structure. One may wonder if the RQ -PL structure on a rational PL
subspace P of X can be refined to an integral one. Ducros and Thuillier showed in
[DT14] that this is indeed the case: there is an R-PL structure on P such that a
function f : P → R×
+ is R-PL if and only if G-locally it is of the form r|f |, where f
is an analytic function on X and r ∈ H (see [DT14, 3.7], and note that H Q = H).
Obviously, such a structure is unique but consistency of the definition requires an
argument. In addition, they show that the associated RQ -PL space (obtained by
adjoining integral roots of all R-PL functions on P ) coincides with the original
RQ -PL space. In the sequel, we provide P with this R-PL structure and call it the
R-PL subspace of X.
2.6. Topological realization of PL spaces and skeletons.
2.6.1. The n-dimensional affine RS -PL space. Consider the RS -PL space A =
n
(R×
+ ) with coordinates t1 , . . . ,tn . It is provided with the G-topology AG of RS polyhedrons whose objects are finite unions of (bounded) RS -polytopes and whose
coverings are the finite one. For any polytope P ⊂ A the topological realization
|PG | is a quasi-compact topological space. As in the case of analytic spaces (see
2.2.4), Deligne’s theorem implies that AG has enough points, and, moreover, (AG )∼
is equivalent to |AG |∼ . For shortness, we will not distinguish AG (resp. PG ) and
its topological realization |AG | (resp. |PG |).
2.6.2. G-skeletons. If X is an analytic space with an R-PL subspace then the embedding i : P ֒→ X is continuous with respect to the G-topologies, see [Ber04,
Theorem 6.3.1]. Since the functor that associates to a sober topological space the
lattice of its open subsets is fully faithful, this implies that i extends to a continuous
embedding iG : PG ֒→ XG and we say that PG is an R-PL subspace of XG .
Remark 2.6.3. (i) The main advantage of working with PG is that it is a locally
quasi-compact topological space, so many proofs reduce to local arguments. For
this, one has to describe the new points, but, as we will see, this is simple: points
of PG correspond to valuations on abelian groups, and points of iG (PG ) correspond
to t-monomial valuations.
(ii) It seems that the use of model theory in [DT14] is mainly needed for the
same aim. One interprets PG in terms of definable sets and types in the theory of
ordered groups, and Deligne’s theorem on points of locally coherent sites is replaced
with Goedel’s completeness theorem. This is not so surprising, since it is known
that Goedel’s theorem and Deligne’s theorem are equivalent, when appropriately
translated.
2.6.4. Monomiality. Notions of t-monomial and generalized Gauss valuations naturally extend to general valuations. Namely, let L/l be an extension of valued fields
and (t1 , . . . ,tn ) a tuple of elements of L. We say that the valuation on
P l(t) is a generalized Gauss valuation (with respect to l) if for any polynomial a = i∈Nn ai ti ∈ l[t]
the equality |a| = maxi |ai ti | holds. Such a valuation
Qn on l(t) is uniquely determined
by its restrictions onto l and the monoid tZ := j=1 tZ
j . If, in addition, L is finite
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 15
over the closure of l(t) in L then we say that L and its valuation are t-monomial.
The following result is proved in [DT14, Proposition 1.8.3].
Lemma 2.6.5. Assume that X is an analytic space, f : U → Gnm is a monomial
×
chart given by t1 , . . . ,tn ∈ OX
(U ), and x ∈ XG is a point whose maximal generG
]
ization y is contained in the R-PL subspace P = S(f ). Let l denote the field H(y)
k).
provided with the valuation νex (see Remark 2.3.10(i)), and let d = tr.deg.(l/e
Assume that |ti |x = 1 for 1 ≤ i ≤ d, and set si = e
ti for 1 ≤ i ≤ d. Then x ∈ PG if
and only if the extension l/e
k is s-monomial.
×
(U ). Let
Consider a monomial chart f : U → Gnm given by u1 , . . . ,un ∈ OX
G
Q
lij
n
×
n
g : U → Gm be given by ti = ci j=1 ui , where ci ∈ k and (lij ) is an n-by-n
integer matrix with non-zero determinant. Then it is easy to see that g is another
monomial chart and S(f ) = S(g). Note also that if x ∈ UG is a point whose maximal
generization y is monomial then choosing monomials ti appropriately we can achieve
] e
that |ti |x = 1 for 1 ≤ i ≤ Fy = tr.deg.(H(x)/
k) (in terms of [DT14, Section 1.8], ti
are well presented at x). Therefore, Lemma 2.6.5 implies the following corollary.
Corollary 2.6.6. Assume that P is an R-PL subspace of X and x ∈ PG is a point,
] with the valuation νex . Then the extension
and let l denote the residue field H(x)
e
l/k is Abhyankar.
2.6.7. Unramified monomial charts. A monomial chart U → Gnm is called unramified (resp. tame) if it is so at all points x ∈ U .
Theorem 2.6.8. Assume that k is algebraically closed. Let X be a k-analytic space
and P a compact R-PL subspace of X. Then there exist finitely many unramified
monomial charts fi : Ui → Gnmi such that P = ∪i S(fi ).
Proof. First, we observe that it suffices to show that for any point x ∈ PG there
exists an unramified monomial chart f : U → Gnm such that x ∈ S(f )G . Indeed,
intersecting this chart with a chart g such that x ∈ S(g)G and S(g) ⊆ P we can also
achieve that S(f ) ⊆ P , and then the assertion follows from the quasi-compactness
of PG .
Now, fix a point x ∈ PG . The remaining local argument is analogous to that in
] e
Lemma 2.5.4, but this time we should choose the transcendence basis of H(x)/
k
]
more carefully. Let y ∈ X be the maximal generization of x and provide l = H(y)
k is Abhyankar by Corollary 2.6.6.
with the valuation νex . Then y is monomial and l/e
Set E = Ey and F = Fy . Choose elements aF +1 , . . . ,aF +E such that their images
in (|H(y)× |/|k × |)⊗ Q form a basis. Next, choose a transcendence basis b1 , . . . ,bF of
l/e
k such that l/e
k(b1 , . . . ,bF ) is unramified; this is possible by [Kuh10, Theorem 1.3].
Remark 2.6.9. Existence of such a basis is the valuation-theoretic ingredient of
local uniformization of Abhyankar valuations. It is an immediate consequence of the
deep theorem that l is stable, see [Kuh10, Theorem 1.1] or [Tem13, Remark 2.1.3].
In fact, one can take b1 , . . . ,bF to be any basis such that b1 , . . . ,bEe is mapped to
e = E e , and b e , . . . ,bF is mapped to a separable
a basis of |l× | ⊗ Q, where E
l/k
E+1
transcendence basis of e
l/e
k.
Choose any lifts a1 , . . . ,aF ∈ κG (x) of b1 , . . . ,bF and then choose any lifts
t1 , . . . ,tn ∈ OXG ,x of a1 , . . . ,an , where n = E +F = dimx (X). Let U be such that ti
16
MICHAEL TEMKIN
are defined on U and x ∈ UG . Then t induces a monomial chart f : U → Gnm , and,
e =e
by our choice, the induced valuations on K = k(a1 , . . . ,an ) and K
k(b1 , . . . ,bF )
are generalized Gauss valuations. The first implies that y ∈ S(f ), hence the second
implies that x ∈ S(f )G by Lemma 2.6.5. In addition, the morphism f satisfies the
assumptions of Theorem 2.3.13 at the points x and y. Indeed, H(y) is unramified
over H(f (y)) = k(a1\
, . . . ,aF ) because the fields are stable, have the same group
of values and the residue field extension l/e
k(b1 , . . . ,bF ) is unramified by the construction (in particular, it is separable). By Theorem 2.3.13, there exists a domain
V ⊆ U such that x ∈ VG and f is unramified at any point of V ; in particular,
f : V → Gnm is an unramified monomial chart as required.
2.6.10. Structure sheaf of A. In the remaining part of Section 2.6 we provide the
promised valuation-theoretic description of the points of PG . This will not be used
in the sequel, so the reader can skip to Section 2.7.
Until the end of Section 2.6, we fix arbitrary R and S, and L = S ⊕ (⊕ni=1 tR
i )
◦
denotes the group of RS -monomial functions on A. We provide A with the sheaf OA
◦
such that if P is an RS -polyhedron then OA (P ) is the monoid of RS -PL functions
with values in (0, 1].
2.6.11. Combinatorial valuations. By an RS -valuation on L we mean any homomorphism | | : L → Γ to an ordered group such that if r ∈ R+ = R ∩ R+ and x ∈ L
with |x| ≤ 1 then |xr | ≤ 1, and the restriction of | | onto S ⊆ L is equivalent to
the embedding S ֒→ R×
+ . We say that a valuation is bounded if for any x ∈ L there
exists s ∈ S with |x| ≤ |s|.
2.6.12. Valuation monoids. Analogously to ring valuations, the valuation is determined up to an equivalence by the valuation monoid L◦ consisting of all elements
x ∈ L with |x| ≤ 1. It is bounded if and only if SL◦ = L. In addition, (L◦ )gp = L
and L◦ is an (RS )◦ -monoid, i.e. it contains S ◦ = S ∩ (0, 1] and is closed under the
action of R+ . In this case, we say that L◦ is a bounded valuation R+ -monoid of L.
2.6.13. Valuative interpretation of points. Similarly to the points of XG , points of
PG admit a simple valuative-theoretic interpretation.
Theorem 2.6.14. Let A and L be as above. Then for any point x ∈ AG the stalk
◦
OA
is a bounded valuation R+ -monoid of L, and this establishes a one-to-one
G ,x
correspondence between points of PG and bounded RS -valuations on L.
◦
Proof. Choose an RS -polytope P with x ∈ PG . Clearly, M = OA
is an R+ G ,x
gp
monoid. Also, M
= L = SM because these equalities hold already for the
◦
monoid M ′ = OA
(P ). To prove that M is a valuation monoid it suffices to show
G
that if a ∈ L then either a ∈ M or a−1 ∈ M , but this follows from the fact that
P = P {a ≤ 1} ∪ P {a−1 ≤ 1}.
It remains to show that any bounded valuation R+ -monoid L◦ of L equals to
◦
OAG ,x for a unique point x. Consider the set F of all RS -polytopes given by
finitely many inequalities ai ≤ 1 with ai ∈ L◦ . Then F is a prime filter and the
stalk at the corresponding point x is L◦ . Uniqueness of F is also clear from the
construction.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 17
2.6.15. G-skeleton of the torus. In the sequel, RS = R and so L = H ⊕ tZ . Set
T = Gnm , and consider the embedding i : A ֒→ T sending s to the generalized
Gauss valuation | |s and the retraction r : T → A from Remark 2.5.2(ii). Both are
continuous in the G-topology, see [Ber04, Theorems 6.3.1 and 6.4.1], hence extend to
continuous maps iG : AG ֒→ TG and rG : TG → AG . Furthermore, the description
of the maps i and r can be naturally extended to iG and rG . For simplicity, we
explain this only in the case when H = |k × |Q , and the reduction is ungraded.
Given a point x ∈ TG consider its maximal generization y ∈ T and provide H(x)
with the valuation νx , see Remark 2.3.10. Then the restriction of νx onto |k × | ⊕ tZ
is a bounded R-valuation, so we obtain a map rG . Conversely, given a bounded
R-valuation µ : L →
P Γ, we extend it to a generalized Gauss valuation on k[t] by
the max formula | i ai ti |µ = maxi |ai |µ(ti ). Since | |µ is composed from a real
valuation and a valuation on its residue field, we obtain a point of TG . Clearly,
rG ◦ iG = Id, so rG is a retraction onto S(T)G := iG (A).
2.7. Semistable formal models.
2.7.1. Semistable formal schemes. We say that a formal k ◦ -scheme is strictly semistable if locally it admits an étale morphism to a formal scheme of the form Zn,a =
Spf(k ◦ {T0 , . . . ,Tn }/(T0 . . . Tn − a)) with 0 6= a ∈ k ◦ . A formal k ◦ -scheme is called
semistable if it possesses a strictly semistable étale covering.
Remark 2.7.2. Sometimes one does not require that a 6= 0, thereby obtaining a
wider class of semistable formal schemes. If X is semistable in this sense then Xη
is quasi-smooth if and only if one can find charts with a 6= 0. Since we will only
be interested in formal models of quasi-smooth spaces, we use the definition that
includes the condition a 6= 0.
2.7.3. Skeletons associated to semistable formal models. To any semistable formal
k ◦ -scheme X with generic fiber X = Xη Berkovich associated in [Ber99, Section 5]
the skeleton S(X) ⊂ X and constructed a deformational retraction X ։ S(X).
Moreover, these constructions are compatible with any étale morphism φ : Y →
X, i.e. S(Y) = φη−1 (S(X)) and the retraction is compatible with φη , see [Ber99,
Theorem 5.2(vii)].
Since any semistable formal k ◦ -scheme is connected with model semistable formal schemes Zn,a by a zigzag of two étale morphisms, description of the skeleton
and the retraction reduces to the model case, and the latter is induced from Gnm .
Namely, let Gnm be the n-dimensional torus with coordinates T1 , . . . ,Tn . The affinoid subdomain given by |T1 . . . Tn | ≥ |a| and |Ti | ≤ 1 for 1 ≤ i ≤ n equals to
Z = M(k{T0 , . . . ,Tn }/(T0 . . . Tn − a) and hence can be identified with the generic
fiber of Z = Zn,a . Then T1 , . . . ,Tn give rise to the monomial chart f : Z ֒→ Gnm
with skeleton S(f ) = S(Gm
n ) ∩ Z, and S(Z) coincides with S(f ). The retraction of
m
Z onto S(Z) is the restriction of the retraction of Gm
n onto S(Gn ).
Remark 2.7.4. (i) Slightly more generally, Berkovich makes the above constructions for poly-stable models, i.e. models étale-locally isomorphic to products of
semistable ones. In fact, this can be extended further to log smooth formal models
with trivial generic log structure – these are formal schemes that étale-locally admit
smooth morphisms to formal schemes of the form Spf(k ◦ {P/Π}), where Π and P
are sharp fs monoids, α : Π ֒→ k ◦ \ {0} is an embedding such that the composition
18
MICHAEL TEMKIN
Π → |k ◦ \ {0}| is injective, φ : Π → P is an injective homomorphism with no ptorsion in the cokernel and such that Πgp P = P gp , and k ◦ {P/Π} is the quotient of
k ◦ {P } by the ideal generated by elements α(π) − φ(π) for π ∈ Π. The details will
be worked out elsewhere.
(ii) The main motivation for considering formally smooth formal models is that
it is believed (at least by the author) that any quasi-smooth compact strictly analytic space possesses a formally smooth formal model. This is absolutely open
when char(e
k) > 0, but one may hope to prove this by current techniques when
char(e
k) = 0. Currently, the latter is only known for a discretely-valued k: desingularization of excellent formal schemes implies that there even exists a semistable
formal model. If the Q-rank of |k × | is larger than one, it is easy to give examples
when semistable models do not exists, e.g. Spf(k{at−1 , t}) × Spf(k{bt−1 , t}), where
|b| ∈
/ |a|Q . The situation with poly-stable models is unclear: it is still an open combinatorial question (in dimension at least 4) whether any formally smooth formal
model possesses a poly-stable refinement. An equivalent question was raised by
Abramovich and Karu in [AK00, Section 8]: formally smooth (resp. poly-stable)
models are analogues of weakly semistable (resp. semistable) morphisms in the
sense of [AK00, Section 0].
3. Seminorms ad hoc
In this section we study seminorms on modules and sheaves in an ad hoc way.
This requires less preparatory work and suffices for applications to analytic spaces.
A more general categorical approach will be developed in [Tem14].
3.1. Seminormed rings and modules. In Section 3.1 we recall well-known facts
and definitions concerning seminorms. All seminorms we consider are non-archimedean.
Ring seminorms will be denoted | |, | |′ , etc., and module seminorms will be denoted
k k, k k′ , etc.
3.1.1. Seminormed abelian groups. Throughout this paper, a seminorm on an abelian
group A is a function k k : A → R+ such that
(0) k0k = 0,
(1) k − xk = kxk for any x ∈ A,
(2) the strong triangle inequality kx + yk ≤ max(kxk, kyk) holds for any x, y ∈
A.
The pair (A, k k) will be called a seminormed group. If k k has trivial kernel then
it is called a norm.
3.1.2. Bounded and contracting homomorphisms. A homomorphism φ : (A, k kA ) →
(B, k kB ) is called bounded (with respect to k kA and k kB ) if there exists C = C(φ)
such that kak ≤ Ckφ(a)k for any a ∈ A. If C = 1 then φ is called contracting.
Remark 3.1.3. It would be more precise to say ”non-expanding” instead of ”contracting”, but we follow the usual terminology.
3.1.4. The non-expanding category. As in [BBK13, Section 5.1], the category of
seminormed abelian groups with contracting homomorphism will be called the nonexpanding category (of abelian groups).
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 19
Remark 3.1.5. (i) Often one works with the larger category whose morphisms
are arbitrary bounded morphisms; let us call it the bounded category. In fact, it is
equivalent to the localization of the non-expanding category by bounded contracting
isomorphisms.
(ii) Working with the bounded category is natural when one wants to study
seminorms up to equivalence; for example, this is the case in the theory of Banach
spaces. On the other side, working with the non-expanding category is natural
when one distinguishes equivalent seminorms, so this fits the goals of the current
paper.
(iii) A serious advantage of working with the non-expanding category is that one
can describe limits and colimits in a simple way, see [BBK13, Section 5.1].
3.1.6. Seminormed rings. A seminorm on a ring A is a seminorm | | on the underlying group (A, +) that satisfies |xy| ≤ |x||y| for any x, y. Ring with a fixed
seminorm is called a seminormed ring. Usually it will be denoted by calligraphic
letters and the seminorm will be omitted from the notation, e.g. A = (A, | |A ). An
important example, of a normed ring is a real-valued field.
3.1.7. Seminormed modules. A (non-expanding) seminormed A-module M is an
A-module provided with a seminorm k kM such that kamkM ≤ |a|A kmkM for any
a ∈ A and m ∈ M . The notions of contracting homomorphisms of seminormed
rings and modules are defined in the obvious way.
3.1.8. Residue seminorms and cokernels. If φ : A → B is a surjective homomorphism of seminormed abelian groups, rings or modules then the residue seminorm
on B is defined by kxkB = inf y∈φ−1 (x) kykA . It is the maximal seminorm such that
φ is contracting, hence, in the case of groups and modules, (B, k kB ) is the cokernel
of any contracting homomorphism C → A whose image is Ker(φ).
3.1.9. Strictly admissible homomorphisms. Recall that a homomorphism ψ : C →
D of seminormed abelian groups (resp. rings or modules) is called admissible if
the residue seminorm on ψ(C) is equivalent to the seminorm induced from D. In
the non-expanding category, it is natural to consider the following more restrictive
notion: ψ is strictly admissible if the residue seminorm on ψ(C) equals to the
seminorm induced from D.
3.1.10. The completion along a seminorm. The completion of A with respect to
b Note that k k extends to A
b by continuity and
the seminorm topology is denoted A.
b
the completion map α : A → A is an isometry (i.e. kxk = kα(x)k) whose kernel is
the kernel of k k. Completion is functorial and, in the case of groups or modules,
it takes exact sequences to semiexact sequences, i.e. sequences in which Im(di+1 )
is dense in Ker(di ). Moreover, if all morphisms of an exact sequence are strictly
admissible then the completion is exact and strictly admissible.
3.1.11. Tensor products. Given a seminormed ring A, by tensor product of seminormed A-modules M and N we mean the module L = M ⊗A M provided with the
standard tensor seminorm k k⊗ . Note that k k⊗ is the maximal seminorm such that
the bilinear map φ : M ×N → L is contracting, i.e. satisfies kφ(m, n)k⊗ ≤ kmk·knk.
Obviously, M × N → (L, k k⊗ ) is the universal contracting bilinear map.
20
MICHAEL TEMKIN
3.1.12. Exterior and
Vnsymmetric powers. If M is a seminormed A-module then the
M acquire a natural seminorm as follows: both are quotients
modules S n M and
V
⊗n M , so we consider the tensor seminorm on ⊗n M and provide S n M and n M
with the residue seminorms. The latter can be characterized as the maximal
Qn seminorms such that km1 ⊗ · · · ⊗ mn k and km1 ∧ · · · ∧ mn k do not exceed i=1 kmi k
for any m1 , . . . ,mn ∈ M .
3.1.13. Filtered colimit. Assume that {(Aλ , k kλ )}λ∈Λ is a filtered family of seminormed abelian rings (resp. rings or A-modules) with contracting transition homomorphisms, A = colimλ Aλ is the filtered colimit, and fλ : Aλ → A are the natural
maps. We provide A with the colimit seminorm given by kak = inf λ∈Λ,b∈f −1 (a) kbkλ .
λ
Obviously, this is the maximal seminorm making each homomorphism fλ contractible, and hence A is the colimit of Aλ in the category of seminormed rings.
3.1.14. Banach rings and modules. A seminormed ring or module is called Banach
if it is complete; in particular, the seminorm is a norm. Note that a Banach Ab
module M is automatically a Banach A-module.
3.1.15. Unit balls. The unit ball of a seminormed ring A will be denoted A⋄ ; it is
a subring of A. The unit ball of a seminormed A-module M will be denoted M ⋄ ;
it is an A⋄ -module.
Remark 3.1.16. (i) Perhaps, M ◦ would be a better notation, but the sign ◦ is
traditionally reserved for the unit ball of the spectral seminorm of a Banach algebra.
(ii) If k is a valued field, 0 6= π ∈ k ◦◦ , and V is a seminormed k-space then the
induced topology on V ⋄ is the π-adic one. In this case, V is Banach if and only V ⋄
is π-adic.
3.1.17. Bounded categories. For the sake of completeness, we make some remarks
on the categories of seminormed abelian groups (resp. rings or A-modules) with
bounded homomorphisms. Usually, seminormed A-modules in our sense are called
non-expanding and a general seminormed A-module M is defined to be an Amodule provided with a seminorm k kM such that there exists C = C(M ) satisfying
kamkM ≤ C|a|A kmkM for any a ∈ A and m ∈ M . Some results below can be
extended to bounded homomorphisms using the following lemma, but we will not
pursue this direction in the sequel.
Lemma 3.1.18. Assume that φ : S → R is a bounded homomorphism of seminormed abelian groups, rings or modules. Then replacing the norm of S with an
equivalent norm one can make φ contracting.
Proof. Define a new seminorm k k′S by kak′S = max(kakS , kφ(a)kR ).
3.2. Sheaves. Throughout Section 3.2 X is a topological space. We will define
seminorms on sheaves and various operations with them using stalks. See [Tem14]
for a categorical approach, which is both more conceptual and general.
3.2.1. Seminormed sheaves. Assume that X is a topological space and A is a sheaf
of rings on X. By a seminorm | | on A we mean the family of seminorms | |x : AX →
R+ on the stalks of A such that for any open U ⊆ X and a section s ∈ A(U ) the
function fs : U → R+ given by fs (x) = |sx |x is upper semicontinuous. We will
denote fs as |s|. The pair (A, | |) will be called seminormed sheaf of rings, and | |
will be often omitted in the notation.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 21
Similarly, if A is a seminormed sheaf of rings and M is a sheaf of A-modules
then by an A-seminorm on M we mean a family of Ax -seminorms on Mx such
that the same semicontinuity condition as above holds for the sections of M. A
homomorphism of seminormed sheaves (M, k k) → (N , k k′ ) is contracting if so are
all its stalks (Mx , k kx ) → (Nx , k k′x ).
3.2.2. Quasi-norms. The sup seminorm on non-compact sets can be infinite, so it
is technically convenient to introduce the following notion. Let R+ = R+ ∪ {∞} be
the one-point compactification of R+ with addition and multiplication satisfying
all natural rules and the rule 0 · ∞ = 0. By a quasi-norm on a ring A we mean a
function | | : A → R+ such that |0| = 0, |a + b| ≤ max(|a|, |b|) and |ab| ≤ |a||b| for
a, b ∈ A. Given an A-module M we define an A-quasi-norm on M similarly.
3.2.3. Sheaves of seminormed rings and modules. By a sheaf of seminormed rings
(A, | |) on S we mean a sheaf of rings A provided with quasi-norms | |U : A(U ) → R+
for each open U so that the following two conditions hold for any open U and
s ∈ A(U ):
S
(1) Boundedness: there exists an open covering U = i Ui such that |si |Ui < ∞
for any i, where si denotes s|Ui .
S
(2) The sheaf condition: for any open covering U = i Ui the equality |s|U =
supi |si |Ui holds.
Similarly, if A is a sheaf of seminormed rings then by a sheaf of seminormed
A-modules we mean a sheaf of A-modules M provided with an A(U )-quasi-norm
k kU : M(U ) → R+ for each open U so that the above two conditions hold for the
family of quasi-norms k kU .
3.2.4. Equivalence of the two notions. Given a seminormed sheaf of rings A, define the supremum quasi-norm | |U on A(U ) by |s|U = supx∈U |s|x . Clearly, the
supremum seminorms satisfy the sheaf condition. Also, it follows from the semicontinuity of |s| that for any x ∈ U there exists a neighborhood x ∈ V such that
|s|V ≤ |s|x + ε. In particular, the boundedness condition is satisfied and we obtain
a sheaf of seminormed rings α(A). The same construction applies to seminormed
sheaves of A-modules.
Conversely, if B is a sheaf of seminormed rings then for any s ∈ B(U ) and x ∈ U
we set |s|x = inf x∈V |s|V where V runs through the open neighborhoods of x.
Since the family of neighborhoods of x is filtered, taking the infimum preserves the
subadditivity and the submultiplicativity, and so | |x is a quasi-norm. Moreover, it
follows from the boundedness condition that |s|x < ∞ for any s, and hence | |x is
a seminorm. In addition, if |s|x < r then already |s|V < r for a neighborhood of
x and hence the family of stalk seminorms satisfies the semicontinuity condition.
In particular, we obtain a seminormed sheaf of rings β(B). The same construction
applies to sheaves of seminormed A-modules.
Lemma 3.2.5. The constructions α and β are inverse to each other and, thereby,
establish a one-to-one correspondence between the isomorphism classes of seminormed sheaves of rings on X and sheaves of seminormed rings on X. Furthermore,
for each seminormed sheaf of rings A they provide essentially inverse equivalences
between the categories of seminormed sheaves of A-modules and sheaves of seminormed α(A)-modules.
22
MICHAEL TEMKIN
Proof. We will deal only with the sheaves of rings, since the case of modules is
similar. Assume that A = (A, | |) is a seminormed sheaf of rings and let β(α(A)) =
(A, | |′ ). Clearly, | |x ≤ | |′x for any x ∈ X and we should prove that the equality
holds. Fix x ∈ U and s ∈ A(U ). By the semicontinuity of |s|, for any ε > 0 there
exists open V such that x ∈ V ⊆ U and |s|y < |s|x + ε for any y ∈ V . Therefore,
|s|′x ≤ |s|V ≤ |s|x + ε.
Conversely, assume that B = (B, | |) is a sheaf of seminormed rings and let
α(β(B)) = (B, | |′ ). Clearly, | |U ≥ | |′U for any open U and we should prove that the
equality holds. Assume, conversely, that |s|U > |s|′U and fix r with |s|U > r > |s|′U .
Then |s|x < r for any x ∈ U , where | |x is the x-stalk of β(B). By the definition of
β, there exists an open Vx such that x ∈ Vx ⊆ U and |s|Vx < r. In particular, we
obtain an open covering U = ∪x Vx such that supx |s|Vx ≤ r < |s|U , violating the
sheaf condition.
Remark 3.2.6. (i) In the sequel we will not distinguish the notions of seminormed
sheaves of rings/modules and sheaves of seminormed rings/modules. The seminorm
will be denoted | | or analogously (e.g. k k or | |′ ) and its sections and stalks will
be denoted | |U and | |x .
(ii) Similarly to the classical sheaves, the definition with stalks is often more convenient for direct computations and constructions. However, the second definition
is more conceptual and can be extended to arbitrary topoi in a sheaf-theoretic way.
In particular, one can define presheaves of seminormed objects, sheafifications, etc.,
see [Tem14].
3.2.7. Operations on seminormed sheaves. Quotients, tensor products, symmetric
powers and exterior powers of seminormed sheaves are defined stalkwise; in particular, the underlying sheaf is the usual quotient, tensor product, etc. For example,
assume that A is a seminormed sheaf of rings on X, φ : M → N is an epimorphism of A-modules and M is provided with an A-seminorm k k. Then each stalk
Mx → Nx is an epimorphism, and we define the residue seminorm k k′ on N by
taking k k′x to be the residue seminorm of k kx for any x ∈ X. Locally on X any
section of N comes from a section of M and this easily implies that k k′x satisfies
the semicontinuity condition. All other constructions are similar, so we skip the
details.
Remark 3.2.8. Let K denote the kernel of φ. In general, the map M(U )/K(U ) →
N (U ) is not isomorphism, and even if it is an isomorphism, k k′U does not have to
be the residue norm of k kU . However, quite naturally U 7→ M(U )/K(U ) can be
viewed as a presheaf of seminormed modules whose sheafification coincides with N .
This approach is worked out in [Tem14].
3.2.9. Pullbacks. If f : Y → X is a continuous map and (F , | |) is a seminormed
sheaf of rings or modules on X, we define its pullback as follows. Consider the
usual pullback G = f −1 F ; then Fx = Gy for any x ∈ X and y ∈ f −1 (x), and we
set | |x = | |′y . Locally, any section s of G is induced from a section t of F , hence
any function |s|′ locally decomposes as |t| ◦ f . This implies that the seminorms
| |′y satisfy the semicontinuity condition and f −1 (F , | |) := (G, | |′ ) is a seminormed
sheaf of rings.
Remark 3.2.10. Definition of pushforwards is more problematic. If (F , | |) is a
seminormed sheaf of rings on Y , a natural definition is to set (G, | |′ ) := f∗ (F , | |),
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 23
where G = f∗ F and |s|′x = supy∈f −1 (x) |s|y . However, this requires some assumptions on f ; for example, it suffices to assume that the fibers of f are quasi-compact
and any neighborhood of a fiber f −1 (x) contains the preimage of a neighborhood
of x.
3.3. Sheaves on analytic spaces.
3.3.1. The spectral seminorm. If X is an analytic space then the semivaluations | |x
of the stalks OXG ,x provide the structure sheaf OXG with the spectral seminorm
| |. For any open U ∈ XG and f ∈ OXG (U ) we have that |f |U = supx∈U |f (x)|, so
if U = M(A) then | |U is the spectral seminorm of A.
3.3.2. OXG -seminorms. Assume that (F , k k) is a seminormed OXG -module. Then
for any point x ∈ XG the kernel of k kx contains mx Fx and hence k kx is induced
from the residue seminorm on the fiber F (x) = Fx /mx Fx . We call the latter the
fiber seminorm and denote it k k(x) . The completion of F (x) with respect to k k(x)
[
will be called completed fiber and denoted F
(x).
Remark 3.3.3. Since kskU = maxx∈U ks(x)k(x) for any open U ⊆ XG and s ∈
F (U ), one can view k k as the family of fiber seminorms on the vector spaces F (x)
[
or F
(x).
3.3.4. Pullbacks. If f : Y → X is morphism of Berkovich spaces and F is a seminormed OXG -module then we define the pullback as f ∗ F = f −1 F ⊗f −1 OXG OYG ,
where both f −1 and the tensor product are taken in the sense of seminormed
sheaves.
3.3.5. The case of invertible sheaves. If K is a real-valued field and V is a onedimensional vector space with basis v then for any r ∈ R+ there exists a unique
K-seminorm | |r on V such that |v|r = r. This obvious fact easily extends to
invertible sheaves.
Lemma 3.3.6. Assume that X is an analytic space and F is a free OXG -module
of rank one with basis s. Then the correspondence k k 7→ ksk establishes a bijection
between OXG -seminorms on F and upper semicontinuous functions r : XG → R+ .
Proof. Let us construct the opposite correspondence. So, assume that r is an upper
semicontinuous function on XG . For any x ∈ XG define the seminorm k k(x) on
F (x) by the condition ks(x)k(x) = r(x) and let k kx be the induced seminorm
on Fx . The stalk seminorms Fx satisfy the semicontinuity condition. Indeed, if
s′ ∈ F (U ) then s′ = f s for f ∈ OXG (U ) and hence ks′ kx = |f |x r(x). It remains to
use that r is semicontinuous and |f | is even continuous. Thus, the seminorms k kx
give rise to a seminorm k kr on F and, obviously, the correspondence r 7→ k kr is
opposite to k k 7→ ksk.
3.3.7. Analytic seminorms. Assume that (F , k k) is a seminormed OXG -module.
We say that the seminorm k k is analytic if for any point y ∈ XG with a generization
x ∈ XG (see 2.2.8), the seminorm k ky is induced from k kx , i.e. the generization
map Fy → Fx is an isometry. For example, the spectral norm on OXG is analytic.
Analytic seminorms on F are determined by their restriction onto the set of analytic
points X ⊂ XG . In addition, we have the following relation with the retraction
XG → X.
24
MICHAEL TEMKIN
Lemma 3.3.8. Assume that an OXG -module F is provided with an analytic OXG seminorm. Then for any analytic domain U ⊆ X and s ∈ F (UG ) the map
ksk : UG → R+ factors through the retraction UG → U and the induced map
U → R+ is upper semicontinuous.
Proof. The first claim is obvious. Since ksk is upper semicontinuous and UG → U
is a topological quotient by Theorem 2.2.10, we obtain the second claim.
Combining the above lemma with Lemma 3.3.6 we obtain the following result.
Corollary 3.3.9. Assume that X is an analytic space and F is a free OXG -module
of rank one with basis s. Then the correspondence k k 7→ ksk establishes a bijection between analytic OXG -seminorms on F and upper semicontinuous functions
r : X → R+ .
3.3.10. Analytic seminorms on coherent sheaves. If F is a coherent OXG -module,
y ∈ XG and x ∈ X is the maximal generization of y then Fx = Fy ⊗OXG ,y OXG ,x .
In particular, F (y) is dense in F (x) = F (y) ⊗κG (y) κG (x).
Lemma 3.3.11. Assume that coherent OXG -modules F and G are provided with
OXG -seminorms. Then,
(i) The seminorm of F is analytic if and only if for any point y ∈ XG with the
[
[
maximal generization x ∈ X the induced map of completed fibers F
(y) → F
(x) is
an isometric isomorphism.
(ii) If the seminorms on F and G are analytic then the induced seminorms on
F ⊗ G, S n F and ∧n F are analytic.
Proof. Let y ∈ XG and let x ∈ X be the maximal generization of y. As we noted
above, φ : F (y) → F (x) has a dense image, hence φ is isometry if and only if the
[
[
completion F
(y) → F
(x) is an isometric isomorphism. This implies (i).
d = F
d and similarly for x, hence
[
Set E = F ⊗ G. Then E(y)
(y) ⊗H(y) G(y)
d = E(x)
d as normed vector spaces, and the seminorm on E is analytic by (i).
E(y)
The other claims of (ii) are proved similarly. For example, if L = ∧n F then
d = ∧n (F
[
L(y) = ∧n (F (y)) and hence L(y)
(y)) depends only on the completed fiber
of F .
4. Differentials of seminormed rings
4.1. Kähler seminorms.
4.1.1. The definition. Given a contracting homomorphism A → B of seminormed
rings we provide ΩB/A with the Kähler seminorm k kΩ given by the formula
kxkΩ =
x=
inf
P
ci dbi
max |ci |B |bi |B ,
i
P
where x ∈ ΩB/A and the infimum is over all representations of x as
ci dB/A (bi )
with ci , bi ∈ B. This seminorm depends on the seminorm of B but is independent
of the seminorm of A.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 25
4.1.2. Universal properties. Both the Kähler seminorm and the seminormed module
(ΩB/A , k kΩ ) can be characterized by appropriate universal properties.
Lemma 4.1.3. Assume that φ : A → B is a contracting homomorphism of seminormed rings. Then,
(i) k kΩ is the maximal B-seminorm that makes dB/A : B → ΩB/A a contracting
A-homomorphism.
(ii) dB/A is the universal contracting A-derivation of B with values in a seminormed B-module:
HomB,contr (ΩB/A , M )f
→DerA,contr (B, M )
for any seminormed B-module M .
Proof. The first claim is obvious. In (ii), we should only prove that any contracting
A-derivation d : B → M with values in a seminormed B-module M factors into a
composition of dB/A and a contracting homomorphism h : ΩB/A → M . By the usual
universal property, we have a unique such factoring with h being a homomorphism
of modules, and it remains to show that h is contracting. Let k k′Ω be defined by
kxk′Ω = max(kxkΩ , kh(x)kM ). Then it immediately follows that (ΩB/A , k k′Ω ) is a
seminormed B-module and dB/A is contracting with respect to k k′Ω . So, k kΩ = k k′Ω
by (i), and hence h is contracting.
4.1.4. An alternative definition. Similarly to the case of usual rings, one can define
the seminormed module ΩB/A in terms of the kernel of B ⊗A B → B. The proof
reduces to repeating the classical argument (see [Mat80, Section 26.C]) and checking
that all relevant maps are contracting.
Lemma 4.1.5. Assume that f : A → B is a contracting homomorphism of seminormed rings and I is the kernel of the induced homomorphism of seminormed rings
B ⊗A B → B. Then the classical isomorphism φ : ΩB/A f
→I/I 2 is an isometry.
Proof. Recall that φ is induced by the derivation d : B → I/I 2 given by db = b ⊗ 1 −
1⊗b. By Lemma 4.1.3, it suffices to show that any contracting A-derivation ∂ : B →
M factors uniquely into a composition of d and a contracting homomorphism of
B-modules h : I/I 2 → M . In fact, a homomorphism h exists and is unique by the
classical theory, so we should only check that it is contracting.
Let B ∗ M denote the seminormed B-module B ⊕ M provided with the multiplication (b, m)(b′ , m′ ) = (bb′ , bm′ + b′ m). The A-bilinear map B × B → B ∗ M sending
(b, b′ ) to (bb′ , b∂(b′ )) is contracting, hence it induces a contracting homomorphism
λ : B ⊗A B → B ∗ M . By the classical argument, λ vanishes on I 2 and takes I to M ,
in particular, it induces a contracting homomorphism h : I/I 2 → M . It remains to
notice that h ◦ d = ∂.
4.2. Basic properties of Kähler seminorms.
4.2.1. Fundamental sequences. First and second fundamental sequences extend to
the context of seminormed rings.
Lemma 4.2.2. Assume that A → B → C are contracting homomorphisms of
seminormed rings. Then,
(i)
f
ΩB/A ⊗B C → ΩC/A → ΩC/B → 0
26
MICHAEL TEMKIN
is a semi-exact sequence of contracting maps and f is strictly admissible.
(ii) If the homomorphism B → C is onto and J is its kernel then ΩC/B = 0 and
the sequence of (i) extends to a semi-exact sequence of contracting maps
J/J 2 → ΩB/A ⊗B C → ΩC/A → 0.
f
Proof. Consider the classical first fundamental sequence ΩB/A ⊗B C → ΩC/A →
ΩC/B → 0 and note that all homomorphisms are contracting by the definition of
their seminorms, and f is strictly admissible. The second claim is proved similarly.
Remark 4.2.3. We aware the reader that even if the map g : ΩB/A ⊗B C → ΩC/A
is an isomorphism it does not have to be strictly admissible. A simple example can
be obtained when A ⊂ B ⊂ C are real-valued fields and the extension C/B is finite,
separable and wildly ramified. This situation will be studied deeper in [Tem14],
but see also Theorem 5.4.4 below.
4.2.4. Base change. Similarly to the classical case, Kähler differentials of seminormed modules are compatible with base changes.
Lemma 4.2.5. Assume that A → B and A → A′ are contracting homomorphism
of seminormed rings and set B ′ = B ⊗A A′ . Then φ : ΩB/A ⊗A′ B ′ → ΩB′ /A′ is an
isomorphism of seminormed modules.
Proof. Since φ is an isomorphism of modules we should prove that it is an isometry.
The seminorm k ks on the source is the maximal one for which kad(b) ⊗ cks ≤
|a|B |b|B |c|B′ for any a, b ∈ B and c ∈ B ′ . The seminorm k kt on the target is the
maximal one for which kxd(y)kt ≤ |x|B′ |y|B′ for any choice of x, y ∈ B ′ . Since
ad(b) ⊗ c goes to acd(b), it follows that any inequality defining k ks holds also for
k kt , and so φ is contracting.
Conversely, let z = xdy ∈ ΩB′ /A′ be an element with x, y ∈ B ′ . Fix r > 1 and
Pn
find a representation y = i=1 bi ⊗P
a′i with bi ∈ B, a′i ∈ A′ , and |bi |B |a′i |A′ ≤ r|y|B′
n
−1
for any 1 ≤ i ≤ n. Then φ (z) = i=1 d(bi ) ⊗ a′i x and hence
kφ−1 (z)ks ≤ max |bi |B |a′i |A′ |x|B′ ≤ r|y|B′ |x|B′ .
i
Thus kφ−1 (xdy)ks ≤ |y|B′ |x|B′ for any x, y ∈ B ′ , and so any inequality defining k kt
holds also for k ks . Thus, φ−1 is contracting and hence φ is an isometry.
4.2.6. Density.
Lemma 4.2.7. If φ : B0 → B is a homomorphism of seminormed rings with a
dense image then the Kähler seminorm of ΩB/B0 vanishes.
Proof. For any b ∈ B and ε > 0 there exists b0 ∈ B0 such that |b − b0 | < ε. Hence
kdbk = kd(b − b0 )k ≤ ε.
Corollary 4.2.8. If A → B0 → B are homomorphism of seminormed rings and
the image of B0 → B is dense then the homomorphism φ : ΩB0 /A ⊗B0 B → ΩB/A
has a dense image.
Proof. By Lemma 4.2.2(ii), K = Im(φ) is the kernel of the admissible surjection
ΩB/A → ΩB/B0 . Since the seminorm on ΩB/B0 is trivial by Lemma 4.2.7, it follows
that any element of ΩB/A can be approximated by an element of K with any
precision, i.e. K is dense.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 27
4.2.9. Filtered colimits. We will also need that Kähler seminorms are compatible
with filtered colimits.
Lemma 4.2.10. Assume that A is s seminormed ring and {Bλ }λ∈Λ is a filtered
family of seminormed A-algebras with contractible transition homomorphisms. Let
B be the colimit seminormed algebra, see 3.1.13. Then ΩB/A = colimλ ΩBλ /A as
seminormed A-modules.
Proof. It is a classical result that the modules are isomorphic, so we should compare
the seminorms. Let k k be the colimit seminorm on ΩB/A . By Lemma 4.1.3(i) and
3.1.13, k k is the maximal seminorm such that the composed A-homomorphisms
φλ
d
d
λ
Bλ →
ΩBλ /A → ΩB/A are contracting. The latter decompose as Bλ → B → ΩB/A
and it follows from the definition of | |B that d : B → ΩB/A is contracting and k k is
the maximal seminorm for which this happens. Thus, k k coincides with the Kähler
seminorm.
4.3. Completed differentials.
b B/A .
4.3.1. The module Ω
Banach B-module and we
derivation will be denoted
b B/A we denote the completion of ΩB/A . It is a
By Ω
call its norm the Kähler norm. The corresponding Ab B/A .
dbB/A : B → Ω
4.3.2. The universal property. Lemma 4.1.3 and the universal property of the completions imply the following result.
Lemma 4.3.3. If φ : A → B is a contracting homomorphism of seminormed rings
then dbB/A is the universal contracting A-derivation of B with values in Banach
B-modules. Namely,
b B/A , M )f
HomB,contr (Ω
→DerA,contr (B, M )
for any Banach B-module M .
b B/A in
Remark 4.3.4. (i) In the case of k-affinoid algebras, Berkovich defines Ω
2
b
[Ber93, §3.3] as J/J , where J = Ker(B → B ⊗A B). Note that the notation in
[Ber93] does not use hat because uncompleted modules of Kähler differentials are
never considered there, but we have to distinguish them in our paper. It follows from
[Ber93, Prop. 3.3.1(ii)] and Lemma 4.3.3 that Berkovich’s definition is equivalent
b B/A .
to ours. In particular, if X = M(B) and S = M(A) then Γ(ΩX/S ) = Ω
(ii) Alternatively, one could deduce the equivalence of the two definitions from
[2 = J/J 2
b B/A = J/J
Lemma 4.1.5. Even more generally, Lemma 4.1.5 implies that Ω
for an arbitrary homomorphism of Banach rings A → B, where J = Ker(B →
b A B) and J 2 is the closure of J 2 in J.
B⊗
5. Kähler seminorms of real-valued fields
Our next aim is to study Kähler seminorms on the vector spaces ΩK/A , where
K is a real-valued field and φ : A → K is a homomorphism of rings. In this case we
will use the notation A◦ = φ−1 (K ◦ ). In fact, we are mainly interested in the cases
when A = Z or A is a field, but we will consider an arbitrary A when possible.
5.1. Log differentials. The aim of this section is to express the Kähler seminorm
on ΩK/A in terms of modules of log differentials, so we start with recalling some
well known facts about the latter.
28
MICHAEL TEMKIN
5.1.1. Log rings. A log structure on a ring A is a homomorphism of monoids αA : MA →
(A, ·) inducing an isomorphism MA× f
→A× . The triple (A, MA , αA ) is called a log
ring. Usually we will denote it as (A, MA ). Homomorphisms of log rings are defined
in the natural way.
5.1.2. Log differentials. If (A, MA ) → (B, MB ) is a homomorphism of log rings
then we denote by Ω(B,MB )/(A,MA ) the module of log differentials. Recall that
the latter is defined as the quotient of ΩB/A ⊕ (B ⊗ MBgp ) by the relations of the
form (dB/A (b), −αB (b) ⊗ b) for b ∈ MB , see [Kat89, §1.7]. For any b ∈ MB we
denote by δB/A the image of 1 ⊗ b in Ω(B,MB )/(A,MA ) . It satisfies the equality
bδB/A (b) = dB/A (b). Intuitively, one may view δb as the log differential d log b and
Ω(B,MB )/(A,MA ) is the universal module obtained from ΩB/A by adjoining the log
differentials of the elements of MB .
5.1.3. The universal log derivative. A log (A, MA )-derivation of (B, MB ) with values in a B-module N consists of an A-derivation d : B → N and a homomorphism
δ : MB → N such that dm = mδm for any m ∈ MB and δm = 0 for any m coming
from MA . The B-module of all log derivations is denoted Der(A,MA ) ((B, MB ), N ).
Lemma 5.1.4. If (A, MA ) → (B, MB ) is a homomorphism of log rings then
dB/A : B → Ω(B,MB )/(A,MA ) , δB/A : MB → Ω(B,MB )/(A,MA )
is a universal log (A, MA )-derivation of (B, MB ). Namely, for any B-module N
the induced homomorphism
HomB (Ω(B,MB )/(A,MA ) , N )f
→Der(A,MA ) ((B, MB ), N )
is an isomorphism.
This fact is proved similarly to its classical analogue for Kähler differentials, so
we skip the proof (see also [GR12, 9.3.4]). As an immediate corollary one obtains
the first fundamental sequence (see also [Ogu, Proposition IV.2.3.1]).
Corollary 5.1.5. Assume that (A, MA ) → (B, MB ) → (C, MC ) are homomorphisms of log rings. Then the sequence
Ω(B,MB )/(A,MA ) ⊗B C → Ω(C,MC )/(A,MA ) → Ω(C,MC )/(B,MB ) → 0
is exact.
5.1.6. The integral log structure. Given a valued field K and a homomorphism
φ : B → K ◦ , we will use the notation
Ωlog
K ◦ /B = Ω(K ◦ ,K ◦ \{0})/(B,B\Ker(φ))
log
and Ωlog
K ◦ = ΩK ◦ /Z .
5.1.7. Torsion. If M is a K ◦ -module then we denote by tor(M ) and Mtf = M/tor(M )
the submodule of torsion elements and the maximal torsion-free quotient, respectively. We say that an element x ∈ M is almost zero if for any r < 1 there exists
π ∈ K ◦ such that r < |π| and πx = 0, and x is called essential otherwise. If |K × |
is discrete then any non-zero element is essential. As in [GR03], we say that a
module is almost zero if so are all its elements, and a homomorphism is an almost
isomorphism if its kernel and cokernel almost vanish.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 29
5.1.8. Adic seminorm. If x ∈ M is an element we define its adic seminorm as
kxkadic =
inf
a∈K ◦ | x∈aM
|a|.
In particular, kxkadic = 0 if and only if x is divisible. Here and in the sequel by
saying divisible we mean infinitely π-divisible, where π is as in 2.1.1. For example,
if the valuation is trivial then k kadic is trivial, i.e. kxkadic = 1 whenever x 6= 0,
and so kxkadic = 0 if and only if x is 0-divisible.
Lemma 5.1.9. Any homomorphism φ : M → N between torsion-free K ◦ -modules
is contracting with respect to the adic seminorms. Furthermore, φ is an isometry if
and only if Ker(φ) is divisible and Coker(φ) contains no essential torsion elements.
Proof. Obviously, φ is contracting. If K = Ker(φ) is not divisible, then it contains
an element x with kxkadic 6= 0, so φ is not an isometry. So, we can assume that
K is divisible and we should prove that in this case φ is an isometry if and only
if Coker(φ) contains no essential torsion elements. Since K is divisible, M/K is
torsion-free and the map M → M/K is an isometry. Thus, replacing M with M/K
we can assume that φ is injective. In this case the assertion is obvious.
5.1.10. The Kähler seminorm on ΩK/A . The following result expresses the Kähler
seminorm in terms of a module of log differentials.
Theorem 5.1.11. Assume that K is a real-valued field with a subring A and provide ΩK/A with the Kähler seminorm k kΩ . Then,
log
(i) The image of the map h : Ωlog
K ◦ /A◦ → ΩK/A is (ΩK ◦ /A◦ )tf .
(ii) k kΩ is the maximal K-seminorm such that kIm(h)kΩ ≤ 1.
Proof. Part (i) follows from the fact that Kähler differentials are compatible with
localizations, in particular,
◦
◦
Ωlog
K ◦ /A◦ ⊗K K = Ω(K,K × )/(A◦ ,A◦ \{0}) = ΩK/A = ΩK/A .
To prove (ii) we note that k kΩ is determined by the inequalities kdxkΩ ≤ |x| with
x ∈ K. For any seminorm, kdxk ≤ x if and only kd(x−1 )k = k − x−2 dxk ≤ |x−1 |.
Hence k kΩ is the maximal K-seminorm whose unit ball contains the K ◦ -span of
all elements x−1 dx with x ∈ K ◦ . The latter is the image of h since Ωlog
K ◦ /A◦ is
generated by the elements δx.
As an immediate corollary we obtain the following formula for k kΩ .
Corollary 5.1.12. Keep the assumptions of Theorem 5.1.11. Then the restriction
of Kähler seminorm onto (Ωlog
K ◦ /A◦ )tf coincides with the adic seminorm of the latter,
i.e. kxkΩ = kxkadic for any x lying in (Ωlog
K ◦ /A◦ )tf ⊆ ΩK/A .
Remark 5.1.13. (i) We have just seen that the Kähler seminorm on ΩK/A is
tightly related to the torsion-free part of Ωlog
K ◦ /A◦ . Nevertheless, we will later see
(e.g. in Theorem 5.4.4) that the torsion part of the modules Ωlog
K ◦ /A◦ does affect
subtle issues related to Kähler seminorms.
(ii) It follows from Corollary 5.2.3 below that if A itself is a field and |A× | is
dense then one can replace Ωlog
K ◦ /A◦ with ΩK ◦ /A◦ in Theorem 5.1.11 and its corollary.
However, in the discretely-valued case the discrepancy between the two modules is
essential, and one has to stick to logarithmic differentials (e.g, see Section 7.2).
30
MICHAEL TEMKIN
(iii) Although any serious investigation of ramification theory of valued fields
should investigate the modules ΩL◦ /K ◦ and Ωlog
L◦ /K ◦ and their torsion, it seems that
[GR03, Chapter 6] is the only source where this topic was systematically studied.
5.2. The torsion part of Ωlog
K ◦ /A◦ . In the next two sections, we will extract from
[GR03, Chapter 6] a few results on Ωlog
K ◦ /A◦ we will need later. We are especially
interested in a control on the torsion part of these modules.
5.2.1. A relation between ΩL◦ /K ◦ and Ωlog
L◦ /K ◦ . Let λK/A denote the natural map
ΩK ◦ /A◦ → Ωlog
K ◦ /A◦ , and let λK = λK/Z .
Lemma 5.2.2. For any real-valued field K the map λK is injective and its cokernel
e where K ◦ acts through K.
e
is isomorphic to |K × | ⊗ K,
Proof. This is a particular case of [GR03, 6.4.15] obtained when ∆ = |K × | and
N = |K ◦◦ \ {0}|.
Corollary 5.2.3. If L/K is an extension of real-valued fields then Ker(λL/K ) is
annihilated by K ◦◦ and Coker(λL/K ) is annihilated by L◦◦ . In particular, if |K × |
is dense then λL/K is an almost isomorphism.
log
◦
Proof. Let N be the image of the map Ωlog
K ◦ ⊗K ◦ L → ΩL◦ and let µ be the
◦
composition of λK ⊗K ◦ L◦ and the map Ωlog
K ◦ ⊗K ◦ L ։ N . Applying the snake
lemma to the commutative diagram
ΩK ◦ ⊗ K ◦ L ◦
/ ΩL ◦
_
µ
0
/N
/ ΩL◦ /K ◦
/0
λL/K
λL
/ Ωlog◦
L
/ Ωlog◦ ◦
L /K
/0
we obtain an exact sequence
0 → Ker(λL/K ) → Coker(µ) → Coker(λL ) → Coker(λL/K ) → 0.
(1)
By Lemma 5.2.2, Coker(λL ) is annihilated by L◦◦ , hence the same is true for
Coker(λL/K ). Similarly, Coker(λK ) is annihilated by K ◦◦ , hence the same is true
for Coker(λK ) ⊗K ◦ L◦ = Coker(λK ⊗K ◦ L◦ ), for its quotient Coker(µ), and for the
submodule Ker(λL/K ) of Coker(µ).
Using Lemma 5.2.2 and its corollary can one easily verify the following examples.
Example 5.2.4. (i) If K is discretely-valued with uniformizer π then Coker(λK ) =
e with generator δπ. In particular, λK is not an almost isomorphism in this case.
K
e = p > 0 and |K × | is p-divisible then λK is an isomorphism.
(ii) If char(K)
(iii) Assume that L/K is a tamely ramified finite extension of discretely-valued
fields with uniformizers πK and πL . Note that Ωlog
L◦ /K ◦ = 0 by Lemma 5.3.7 below
(in fact, L◦ /K ◦ is log étale), but ΩL◦ /K ◦ f
→πK L◦ /πL L◦ is a cyclic module of length
eL/K −1. In particular, λL/K is not injective if the extension L/K is not unramified.
5.3. Almost tame fields and extensions.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 31
5.3.1. Almost unramified extensions. For any separable algebraic extension of reallog
◦
valued fields L/K the module Ωlog
L◦ /K ◦ is torsion because ΩL◦ /K ◦ ⊗L L = ΩL/K = 0.
Recall that L/K is called almost unramified if ΩL◦ /K ◦ almost vanishes.
Lemma 5.3.2. Assume that F/L/K is a tower of separable algebraic extensions
of real-valued fields. Then F/K is almost unramified if and only if both F/L and
L/K are so.
Proof. By [GR03, Theorem 6.3.23], the first fundamental sequence
0 → ΩL◦ /K ◦ ⊗K ◦ L◦ → ΩF ◦ /K ◦ → ΩF ◦ /L◦ → 0
starts with an injection. The lemma follows.
5.3.3. Almost tame extensions. In the discrete valuation case, almost unramified
is the same as unramified, but in general there even are wildly ramified almost
unramified extensions. On the other hand, if the valuation is not discrete then any
tame extension is almost unramified, so the class of almost unramified extensions
is rather weird: a tame extension is almost unramified if and only if the valuation
is not discrete.
It is natural to seek for a natural enlargement of the class of almost unramified
extensions that includes all tame ones, and this can be achieved very simply: one
should replace ΩL◦ /K ◦ with Ωlog
L◦ /K ◦ . So, we say that a separable algebraic extension
L/K is almost tame if Ωlog
L◦ /K ◦ almost vanishes.
Remark 5.3.4. In fact, it is proved in [Tem14] that Ωlog
L◦ /K ◦ is almost zero if and
only if it is zero.
5.3.5. The non-discrete case. In the non-discrete case, there is no difference between almost tame and almost unramified because the kernel and cokernel of
ΩL◦ /K ◦ → Ωlog
L◦ /K ◦ almost vanish by Corollary 5.2.3.
5.3.6. Tame extensions. In this case, we can easily show that the log differentials
vanish. This is rather natural, since if L/K is tame then, ignoring the finite presentation issue, L◦ /K ◦ should behave similarly to a log étale extension.
Lemma 5.3.7. Assume that L/K is a tamely ramified extension of real-valued
fields. Then Ωlog
L◦ /K ◦ = 0.
Proof. Since Ω is compatible with filtered colimits, it suffices to deal with the case
when L/K is finite. By étale descent, it suffices to prove the lemma in the case
when K ◦ is strictly henselian. By the classical ramification theory, in this case L/K
splits into a tower of elementary extensions of the form L = K(π 1/l ), where π ∈ K
e By the first fundamental sequence, it suffices to establish
and l is invertible in K.
the case of an elementary extension. Note that L◦ = ∪K ◦ [aπ 1/l ] and the union is
over all a ∈ K such that |aπ 1/l | ≤ 1. It follows that Ωlog
L◦ /K ◦ is the filtered colimit of
the modules Ωlog
. But the latter modules vanish since they are generated
K ◦ [aπ 1/l ]/K ◦
by δ(aπ 1/l ) = 1l aδπ = 0.
32
MICHAEL TEMKIN
5.3.8. The discrete case. When the valuation of K is discrete, the opposite of
Lemma 5.3.7 is also true.
Lemma 5.3.9. Assume that L/K is a separable algebraic extension of real-valued
fields and the valuation on K is discrete. Then L/K is almost tame if and only if
it is tame.
Proof. The opposite implication is covered by Lemma 5.3.7, so it suffices to prove
that if L/K is not tame then Ωlog
L◦ /K ◦ contains an essential torsion element. Furthermore, since L is discretely-valued it suffices to show that Ωlog
L◦ /K ◦ 6= 0. By def×
×
e K
e is not separable or |L |/|K | contains a non-trivial p-torsion
inition, either L/
e
for p = exp.char(K). In the first case,
◦ e
Ωlog
eL
e × )/(K,
e K
e × ) = ΩL/
e K
e 6= 0.
L◦ /K ◦ ⊗L L = Ω(L,
In the second case, we will use Lemma 5.2.2. It is easy to see that Coker(λL/K )
is the cokernel of the map h : Coker(λK ) ⊗K ◦ L◦ → Coker(λL ), e.g. use the exact
sequence (1). By Lemma 5.2.2, we can represent h as the homomorphism
e
e ⊗K ◦ L◦ = |K × | ⊗ L◦ /mK L◦ → |L× | ⊗ L.
(|K × | ⊗ K)
e → |L× |⊗ L.
e Since p divides
So, Coker(h) coincides with the cokernel of g : |K × |⊗ L
×
×
e is killed by p, the homomorphism g is not surjective.
the order of |L |/|K | and L
Hence Coker(λL/K ) = Coker(g) 6= 0 and we obtain that Ωlog
L/K 6= 0.
5.3.10. The defectless case. Recall that a finite extension of real valued fields L/K
is called defectless if eL/K fL/K = [L : K], and an algebraic extension is defectless
if all its finite subextensions are so.
Lemma 5.3.11. Assume that L/K is a defectless separable extension of real-valued
fields. Then L/K is almost tame if and only if it is tame.
Proof. We will need the following particular case: if L/K is defectless wildly ramified of degree p then ΩL◦ /K ◦ is not essentially zero. We will establish here only
the case when fL/K = p. The other case can be dealt with similarly (and will be
established in [Tem14]), so we leave it to the interested reader. Fix x ∈ L◦ with
e If f (t) = tp + Pp−1 ai ti is the minimal polynomial of x then |a0 | = 1 and
x
e∈
/ K.
i=0
|ai | < 1 for i > 0 because fe(t) is the minimal polynomial of x
e, which is inseparable
e
over K.
e the elements 1, x, . . . ,xp−1
Since 1, x
e, . . . ,e
xp−1 are linearly independent over K,
◦
form an orthonormal K-basis of L and so L = K ◦ [x]. Therefore, ΩL◦ /K ◦ =
L◦ dx/L◦ f ′ (x)dx and since
p−1
X
iai xi−1 ≤ max(|p|, max (|ai |)) < 1
|f ′ (x)| = pxp−1 +
0<i<p
i=1
we obtain that dx is an essential torsion element.
Now, let us return to the general case. By Lemma 5.3.9, it suffices to consider the
case when |K × | is dense. By Lemma 5.3.7 and Section 5.3.5, it suffices to show that
if L/K is not tame then L/K is not almost unramified. Note that L/K contains a
finite non-tame subfield L′ /K, and by Lemma 5.3.2, it suffices to show that L′ /K
is not almost unramified. Thus, we can assume that L/K is finite. Furthermore,
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 33
by Lemmas 5.3.7 and 5.3.2, it suffices to show that F/K is not almost unramified,
where F/L is a tame extension (in particular, it is defectless). Enlarging L in this
way we can achieve that L/K splits into a tower L/E/K, where E/K is tame
and L/E is Galois and totally wildly ramified. Then L/E splits into a tower of pextensions, hence L/K splits as L/K ′ /K, where L/K ′ is wildly ramified of degree
p. Clearly, L/K ′ is defectless, and so it is not almost unramified by the particular
case we considered first, and so L/K is not almost unramified.
5.3.12. Almost tame extensions: the summary. We summarize the results of Section
5.3.5 and Lemmas 5.3.7, 5.3.9 and 5.3.11 as follows.
Theorem 5.3.13. Assume that L/K is a separable algebraic extension of realvalued fields, then
(i) L/K is almost tame if and only if it is either tame or almost unramified.
(ii) Assume that L/K is defectless; in particular, this is the case when L/K is
tame or |K × | is discrete. Then L/K is almost tame if and only if it is tame.
(iii) If |K × | is dense then L/K is almost tame if and only if L/K is almost
unramified.
5.3.14. Almost tame fields. We refer the reader to [GR03, Section 6.6] for the definition and basic properties of deeply ramified fields. Recall that a real-valued field
K is deeply ramified if and only any its finite separable extension is almost unramified. We say that a real-valued K is almost tame if any its finite separable extension
is almost tame.
Theorem 5.3.15. A real-valued field K is almost tame if and only if at least one
e = 0, (ii) K is deeply ramified.
of the following is true: (i) char(K)
e = 0 then any algebraic extension of K is tamely ramified, so
Proof. If p = char(K)
K is almost tame. In the sequel we assume that p > 0. If K is discretely-valued then
it possesses wildly ramified extensions, so K is not almost tame by Lemma 5.3.9.
If |K × | is dense then, by Section 5.3.5, there is no difference between almost tame
and almost unramified extensions. In particular, in this case K is almost tame if
and only if it is deeply ramified.
5.3.16. Arbitrary extensions of almost tame fields. The defining property of almost
tame fields extends to non-algebraic extensions as follows.
Theorem 5.3.17. Let K be a real-valued field. Then K is almost tame if and
only if for any separable extension of real-valued fields L/K the module Ωlog
L◦ /K ◦ is
almost torsion-free.
Proof. The inverse implication is obvious, so assume that K is almost tame. If
e = 0 then Ωlog◦ ◦ is torsion-free by [GR03, Lemma 6.5.16]. Assume
p = char(K)
L /K
that p > 0, then |K × | is dense and hence λL/K is an almost isomorphism by
Corollary 5.2.3. So, it suffices to show that ΩL◦ /K ◦ is torsion-free. Recall that K
is deeply ramified by Theorem 5.3.15 and hence ΩK ◦ is divisible.
Case 1: char(k) = p. Then ΩL◦ = ΩL◦ /Fp is torsion-free by [GR03, Claim 6.5.21].
Thus ΩL◦ /K ◦ is the cokernel of the map ΩK ◦ ⊗K ◦ L◦ → ΩL◦ whose source is divisible
and target is torsion-free. Hence ΩL◦ /K ◦ is torsion-free.
Case 2: char(k) = 0. Consider the algebraic closure F = K a provided with
an extension of the valuation. Let E be the composite valued field F L. Then
34
MICHAEL TEMKIN
ΩL◦ /K ◦ ⊗L◦ E ◦ ֒→ ΩE ◦ /K ◦ by [GR03, Lemma 6.3.32(ii)], hence it suffices to show
that ΩE ◦ /K ◦ is torsion-free. The latter follows from the facts that ΩF ◦ /K ◦ = 0 since
K is deeply ramified and ΩE ◦ /F ◦ is torsion-free by [GR03, Lemma 6.5.20(i)].
5.4. Kähler seminorms and field extensions. Assume that L/K is an extension of real-valued fields and A → K is a homomorphism of rings. In this section
we will compare the Kähler seminorms | |Ω,K/A and | |Ω,L/A on ΩL/A and ΩK/A ,
respectively.
5.4.1. The map ψL/K/A . The two seminorms are related by the contracting map
ψL/K/A : ΩK/A ⊗K L → ΩL/A ,
where the seminorm on the source is the base change of k kΩ,K/A . Naturally, we
say that k kΩ,K/A and k kΩ,L/A agree if ψL/K/A is an isometry. For shortness,
the seminorms on both sides of ψL/K/A will be denoted k k. This is safe since we
consider only one seminorm on each vector space.
To study ψL/K/A we will use Theorem 5.1.11 and the following commutative
diagram
χ
◦
◦
Ωlog
K ◦ /A◦ ⊗K L
/ Ωlog◦
L /A◦
/ Ωlog◦ ◦
L /K
ε
ζ
◦
◦
(Ωlog
K ◦ /A◦ ⊗ K L )tf
_
φ
/ (Ωlog◦ ◦ )tf
L /A
_
α
ΩK/A ⊗K L
ψL/K/A
(2)
λ
/ Coker(φ)
/0
γ
β
/0
/ ΩL/A
/ ΩL/K
/ 0,
where the top and the bottom rows are first fundamental sequences. It may happen
that Coker(φ) contains a non-trivial torsion and γ is not injective. However, Ker(λ)
is a torsion module, hence
Coker(φ)tf = (Ωlog
L◦ /K ◦ )tf ֒→ ΩL/K .
Lemma 5.4.2. Keep the above notation. Then ψL/K/A is an isometry if and only
if Ker(φ) is divisible and Coker(φ) contains no essential torsion elements.
Proof. Note that ψL/K/A = φ ⊗L◦ L. So, Corollary 5.1.12 implies that ψL/K/A is
an isometry if and only if φ is an isometry with respect to the adic seminorm. It
remains to use Lemma 5.1.9.
5.4.3. Separable algebraic extensions. Note that ψL/K/A is an isomorphism whenever L/K is separable and algebraic.
Theorem 5.4.4. Assume that L/K is a separable algebraic extension of real-valued
fields and A → K is a homomorphism of rings.
(i) If L/K is almost tame then the isomorphism ψL/K/A : ΩK/A ⊗K L → ΩL/A
is an isometry.
(ii) Assume that Ωlog
L◦ /A◦ is torsion-free. If ψL/K/A is an isometry then L/K is
almost tame.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 35
Proof. Note that Ωlog
L◦ /K ◦ is a torsion module because it vanishes after tensoring
with L.
(i) Being a quotient of Ωlog
L◦ /K ◦ , the module Coker(φ) in diagram (2) is almost
zero. In addition, Ker(φ) ⊆ Ker(ψL/K/A ) = 0. Thus, (i) follows from Lemma 5.4.2.
(ii) The assumption on Ωlog
L◦ /A◦ implies that ε in diagram (2) is an isomorphism.
Since ζ is onto, this implies that λ is also an isomorphism. By Lemma 5.4.2,
Coker(φ) contains no essential torsion, hence the torsion module Ωlog
L◦ /K ◦ almost
vanishes, i.e. L/K is almost tame.
Remark 5.4.5. A more detailed study of modules Ωlog
L◦ /K ◦ and their relations to the
seminorms will be undertaken in [Tem14]. In particular, we will show that if ΩL◦ /A◦
is torsion-free then Ωlog
L◦ /K ◦ measures the discrepancy between the seminorms of
ΩK/A ⊗K L and ΩL/A . In fact, the ratio of the volumes of the unit balls equals to
the log different of L/K. (See also Section 7.2.)
5.5. Dense extensions. Another case that will be very important in the sequel is
when K is dense in L, for example K = κG (x) and L = H(x).
5.5.1. The six-term sequence. In this case, L/K can be inseparable (even for an
extension of the form H(x)/κG (x)) so we will have to consider the first homology
H1 (LL◦ /K ◦ ) of the cotangent complex of L◦ /K ◦ , that will be denoted ΥL◦ /K ◦ .
Since, it is difficult to reach ΥL◦ /K ◦ directly, we will also have to study ΩL◦ /K ◦ .
Let us first recall what is proven by Gabber and Ramero. By [GR03, 6.5.12],
ΥL◦/K ◦ is torsion free and Hi (LL◦ /K ◦ ) vanish for i > 1. In particular, for a tower
L/F/K of real-valued fields we have the six-term exact sequence
0 → ΥF ◦ /K ◦ ⊗F ◦ L◦ → ΥL◦/K ◦ → ΥL◦ /F ◦ → ΩF ◦ /K ◦ ⊗F ◦ L◦ → ΩL◦ /K ◦ → ΩL◦ /F ◦ → 0.
5.5.2. Divisibility. We say that an L◦ -module M is a vector space if it is an Lmodule. Equivalently, M is divisible and torsion free.
Lemma 5.5.3. Assume that L/K is an extension of real-valued fields and K is
dense in L. Then both ΩL◦ /K ◦ and ΥL◦ /K ◦ are L-vector spaces.
Proof. We start with three basic types of extensions, and then the general case will
be deduced in three more steps.
Case 1. Assume that L/K is separable algebraic. Since K is dense in L, it follows
that L lies in the henselization of K, i.e. L◦ /K ◦ is étale. In this case, the cotangent
complex vanishes.
Case 2. Assume that L/K is purely inseparable of degree p. In this case, L =
K(x) with a = xp ∈ K \ K p and there exist ci ∈ K such that limi |x − ci | = 0.
Clearly, f (t) = tp − a is the minimal polynomial of x. It is easy to see that L◦ is
i
the filtered colimit of its subrings K ◦ [xi ], with xi = x−c
πi where πi ∈ K are such
that limi |πi | = 0 (same argument as in the proof of [GR03, 6.3.13(i)]). It follows
that LL◦ /K ◦ = colimi LK ◦ [xi ]/K ◦ . Since K ◦ [xi ] = K ◦ [t]/Ii where Ii = (fi ) and
a−cp
fi (t) = tp − πp i is the minimal polynomial of xi , the homologies of LK ◦ [xi ]/K ◦ are
i
easily computable (cf. the proof of [GR03, 6.3.13(iv)]): ΩK ◦ [xi ]/K ◦ is the invertible
module with basis dxi and ΥK ◦ [xi ]/K ◦ = Ii /Ii2 is the invertible module with basis
◦
◦
fi . Since dxi = dx
πi , we obtain that ΩL /K = Ldx.
36
MICHAEL TEMKIN
To describe the map ΥK ◦ [x]/K ◦ → ΥK ◦ [xi ]/K ◦ we consider compatible presentations K ◦ [t] ։ K ◦ [x] and K ◦ [ti ] ։ K ◦ [xi ], where the connecting maps take t
and x to πi ti + ci and πi xi + ci , respectively. This induces the map I/I 2 → Ii /Ii2
that sends f = tp − a to πip tpi + cpi − a = πip fi (ti ). This completely determines the
filtered family ΥK ◦ [xi ]/K ◦ , and since limi |πip | = 0, its colimit is the one-dimensional
L-vector space with basis f .
Case 3. Assume that L = K(x) is purely transcendental. In this case ΥL◦/K ◦ ⊗L◦
L = ΥL/K = 0, and since ΥL◦/K ◦ is torsion free, it actually vanishes. The module
ΩL◦ /K ◦ is computed as in Case 2. First, one shows that L◦ is the filtered colimit
i
of localizations of the K ◦ -algebras K ◦ [xi ] where xi = x−c
πi and limi |πi | = 0 (same
argument as in [GR03, 6.3.13(i)] or [GR03, 6.5.9]). It then follows that ΩL◦ /K ◦ is
◦
◦
the colimit of invertible modules L◦ dxi and dxi = dx
πi . Thus, ΩL /K = Ldx.
Case 4. Assume that L/K is finite. We induct on [L : K]. If L/K is as in Cases
1 or 2 then we are done. Otherwise, it can be split into a tower K ( F ( L, and
the claim holds true for F/K and L/F by the induction. Therefore, in the six-term
exact sequence the terms corresponding to F/K and L/F are vector spaces. It
follows easily that the remaining terms are vector spaces too.
Case 5. Assume that L/K is finitely generated. This time we induct on d =
tr.deg.(L/K). If d = 0 then are in Case 4. Otherwise choose a purely transcendental
subextension F = K(x) ⊆ L and note that the assertion holds for F/K by Case
3 and for L/F by the induction. The same argument with the six-term sequence
completes Case 5.
Case 6. The general case. Obviously, L is the filtered colimit of its finitely
generated K-subsfields Li and L◦ = colimi L◦i . It remains to use the the cotangent
complex and homology are compatible with filtered colimits, and colimit of vector
spaces is a vector space.
5.5.4. Equivalence of Kähler seminorms. Now we apply the above computation
to Kähler seminorms. The following proof could be simplified if one would define the logarithmic cotangent complex and establish a logarithmic analogue of
Lemma 5.5.3.
Theorem 5.5.5. Assume that L/K is an extension of real-valued fields such that
K is dense in L, and let A → K be a homomorphism of rings. Then the map
ψL/K/A is an isometry with a dense image.
Proof. Density of the image follows from Corollary 4.2.8. To prove that ψL/K/A is
an isometry we will use Lemma 5.4.2. Consider the following commutative diagram,
whose rows are exact by Lemma 5.2.2:
0
/ ΩK ◦ ⊗ K ◦ L ◦
/ Ωlog◦ ⊗K ◦ L◦
K
α
0
/ ΩL ◦
β
/ Ωlog◦
L
e ⊗K ◦ L◦
/ (|K × | ⊗ K)
/0
/ |L× | ⊗ L
e
/0
γ
Note that Ker(α) and Coker(α) are vector spaces by Lemma 5.5.3. The map γ is an
e = L,
e hence we obtain that Ker(β) = Ker(α)
isomorphism since |K × | = |L× | and K
log
◦
◦
and Coker(β) = Coker(α) are vector spaces. The map χ : Ωlog
K ◦ /A◦ ⊗K L → ΩL◦ /A◦
is obtained from β by dividing both the source and the target by the image of
◦
Ωlog
A◦ ⊗A◦ L , hence Coker(χ) = Coker(β) is a vector space and Ker(χ) is divisible
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 37
since it is a quotient of Ker(β). Let χtor and χtf be the maps χ induces between
the torsion parts and maximal torsion-free quotients of its arguments. Note that
χtf is nothing else but the map φ from diagram (2).
Since Coker(χ) is torsion free, χtor is surjective and hence Coker(φ) = Coker(χ) is
◦
◦
a vector space. Let x be any element of Ker(φ). Choose a lifting y ∈ Ωlog
K ◦ /A◦ ⊗K L
of x, then χ(y) is torsion, say πχ(y) = 0 for 0 6= π ∈ L◦ . Thus, πy ∈ Ker(χ) and
hence πy is divisible. Therefore, πx is divisible in Ker(φ), and since Ker(φ) is
torsion-free we obtain that x is divisible too. Thus, Ker(φ) is a divisible module
(hence even a vector space), and ψL/K/A : ΩK/A ⊗K L → ΩL/A is an isometry by
Lemma 5.4.2.
Corollary 5.5.6. In the situation of Theorem 5.5.5, the completion of ψbL/K/A is
b K/A f
b L/A .
an isometric isomorphism. In particular, Ω
→Ω
5.6. Kähler seminorms and monomial valuations. In this section we study
finitely generated extensions L/K such that L◦ /K ◦ behaves similarly to log smooth
extensions, though it does not have to be finitely presented.
5.6.1. Orthonormal bases of ΩL/K . Note that if t1 , . . . , tn is a separable transcendence basis of a field extension L/K then dt1 , . . . ,dtn is a basis of ΩL/K . The
following result is an immediate corollary of Theorem 5.1.11.
Lemma 5.6.2. Given an extension of real-valued fields L/K with a separable transcendence basis t1 , . . . ,tn consider the following conditions:
◦
(i) Ωlog
L◦ /K ◦ is a free L -module with basis δt1 , . . . ,δtn .
(ii)
dtn
dt1
t1 , . . . , tn
is an orthonormal basis of ΩL/K .
Then (i) =⇒ (ii) and the conditions are equivalent whenever Ωlog
L◦ /K ◦ is torsionfree.
5.6.3. Generalized Gauss valuations. Assume that K is a real-valued field and A =
K[t] for t = (t1 , . . . ,tn ). For any tuple r = (r1 , . . . ,rn ) of positive numbers by | |r we
P
denote the generalized Gauss valuation | i∈Nn ai ti |r = maxi |ai |r i on A and the
normed ring (A, | |r ) will be denoted K[t]r . Since | |r is multiplicative it extends to
a norm on K(t) that will be denoted by the same letter, and we use the notation
(K(t), | |r ) = K(t)r . The following lemma indicates that (K(t)r )◦ /K ◦ behaves as
a log smooth extension.
Lemma 5.6.4. Assume that K is a real-valued field, r1 , . . . ,rn > 0 and L =
◦
K(t1 , . . . ,tn )r . Then Ωlog
L◦ /K ◦ is a free L -module with basis δt1 , . . . ,δtn .
Proof. By Lemma 5.1.4 it suffices to show that for any L◦ -module M and elements
m1 , . . . ,mn there exists a unique log K ◦ -derivation
(d, δ) : (L◦ , L◦ \ {0}) → M such
P
◦
that δti = mi . If u ∈ K[t] ∩ L then u = l∈Nn al tl ∈ L with |al tl | ≤ 1 and we set
!
n
X X
l i a l tl m i .
δu = u−1
l∈Nn i=1
An arbitrary element z ∈ L◦ is of the form atl u/v for a ∈ K and u, v ∈ K[t] ∩ L◦ ,
so we set δz = lδt + δu − δv and dz = zδz. It is a direct check that the so defined
(d, δ) is a log K ◦ -derivation. Any other log K ◦ -derivation should satisfy the same
formulas for δu, δz and dz, so uniqueness is clear.
38
MICHAEL TEMKIN
5.6.5. A characterization of Gauss valuations. Under mild technical assumptions,
one can also characterize generalized Gauss valuations in terms of Ωlog
L◦ /K ◦ .
Lemma 5.6.6. Let L = K(t1 , . . . ,tn )/K be a purely transcendental extension of
e is perfect and |K × | is p-divisible,
real-valued fields, ri = |ti |, and assume that K
dtn
dt1
e
where p = exp.char(K). If t1 , . . . , tn is an orthonormal basis of ΩL/K then L =
K(t1 , . . . ,tn )r as a valued field.
Proof. Assume that, conversely, the valuation | | on L is not generalized Gauss with
respect to t. Then the valuation on K[t] is strictly dominated by | |r , in particular,
P
there exists a polynomial f (t) = l∈Nn al tl in L◦ such that |f | < |f |r = maxl |al |r l .
Removing all monomials of absolute value strictly smaller than |f |r we can achieve
that the following condition holds: (*) |f | < |f |r and |al tl | = |f |r for each al 6= 0.
Choose f satisfying (*) and of minimal possible degree. We claim that if p > 1
then P
not all monomials are p-th powers. Indeed, assume that the claim fails, say
f = l∈pNn al tl . By our assumption on K, for any al there exists bl such that
P
P
|bpl −al | < |al |. Clearly, l∈pNn bpl tl satisfies (*) and hence l∈Nn bpl tl also satisfies
(*), but its degree is smaller than that of f . It follows that there exists l ∈ Nn and
1 ≤ i ≤ n such that al 6= 0 and |li | = 1.
dtn
1
Since the basis dt
t1 , . . . , tn of ΩL/K is orthonormal, |f | ≥ kdf kΩ , and
df =
n
X X
l∈Nn i=1
l i a l tl
dti
,
ti
the inequality |li al tl | ≤ |f | holds for any choice of l and i. However, |li al tl | =
|al tl | = |f |r > |f | for the choice with al 6= 0 and |li | = 1, a contradiction.
5.6.7. t-monomial valuations. Let L/K be an extension of real-valued fields and
t = (t1 , . . . ,tn ) a tuple of elements of L. We say the valuation on L is t-monomial
with respect to K if the induced valuation on k[t] is a generalized Gauss valuation.
This happens if and only if L contains the subfield K(t)r where ri = |ti |. All results
proved in Section 5.6 can be summarized as follows.
Theorem 5.6.8. Assume that L/K is an extension of real-valued fields with a
separable transcendence basis t1 , . . . ,tn . Consider the following conditions:
(i) The valuation on L is t-monomial and Ωlog
L◦ /K(t)◦ = 0.
◦
(ii) Ωlog
L◦ /K ◦ is a free L -module with basis δt1 , . . . ,δtn .
dtn
1
(iii) dt
t1 , . . . , tn is an orthonormal basis of ΩL/K .
dtn n
n
1
(iv) | dt
t1 ∧ · · · ∧ tn |Ω = 1 in ΩL/K .
Then (i) =⇒ (ii) =⇒ (iii)⇐⇒(iv). Furthermore, if K is almost tame then all
four conditions are equivalent.
Proof. (i) =⇒ (ii) Set F = K(t). By the first fundamental sequence, we have a
log
◦
◦
surjective map ψ ◦ : Ωlog
F ◦ /K ◦ ⊗F L → ΩL◦ /K ◦ , which becomes the isomorphism
ψ : ΩF/K ⊗F Lf
→ΩL/K after tensoring with L. In particular, the kernel of ψ ◦ is
torsion. By Lemma 5.6.4, the source of ψ ◦ is a free module with basis δt1 , . . . ,δtn .
So, the kernel of ψ ◦ is trivial, and hence ψ ◦ is an isomorphism.
(ii) =⇒ (iii) This is covered by Lemma 5.6.2.
i
(iii)⇐⇒(iv) Using that k dt
ti kΩ ≤ 1, this reduces to a simple multilinear algebra.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 39
Finally, assume that K is almost tame, in particular, Ωlog
L◦ /K ◦ is torsion-free.
Then the implication (iii) =⇒ (ii) follows from Lemma 5.6.2. Assume, now, that
(ii) holds; in particular, ψ ◦ is surjective and hence Ωlog
L◦ /F ◦ = 0. Furthermore, the
i
isomorphism ψ is contracting and k dt
ti kΩ ≤ 1 in its source. Therefore, ψ is an
dt1
dtn
isometry and t1 , . . . , tn is an orthonormal basis of ΩF/K . By Lemma 5.6.6, the
valuation on F is generalized Gauss, i.e. L is t-monomial.
Remark 5.6.9. (i) In fact, conditions (i) and (ii) of Theorem 5.6.8 are always
equivalent, but proving this requires to analyse the torsion of Ωlog
L◦ /K ◦ .
(ii) The restriction on K in the last part of the theorem is necessary even when
ep
n = 1. For example, assume that p = char(K) > 0 and a ∈ K ◦ is such that e
a∈
/K
p
and define the norm on L = K(t) so that |t − a| = r for some r ∈ (0, 1) and the
norm is (tp − a)-monomial. Then a simple computation shows that Ωlog
L◦ /K ◦ is a
◦
p
direct sum of L δt and a non-trivial torsion submodule generated by δ(t − a), and
log
hence dt
t is of norm one. Note that ΩL◦ /K ◦ is not free in this case, although its
torsion-free quotient is free with basis δx.
6. Metrization of ΩX/S
Throughout Section 6, f : X → S denotes a morphism of k-analytic spaces. In
the case of sheaves, | | will always denote spectral seminorms on the structure
sheaves and their stalks, and k k will always denote Kähler seminorms (defined
below) on sheaves of pluriforms and their stalks.
6.1. Kähler seminorm on ΩX/S . The aim of this section is to define a seminorm
k k = k kΩ,X/S on ΩX/S (we will mention Ω and X/S in the notation only when a
confusion is possible). For this we should define seminorms k kx on the stalks and
establish their semicontinuity. The basic idea is very simple: given x ∈ XG and
s = f (x) we simply induce the stalk seminorm at x ∈ X from the Kähler norm on
b H(x)/H(s) .
Ω
6.1.1. The stalks. Given x ∈ XG with s = f (x), fix an affinoid domain U = M(A)
with s ∈ UG and let Vλ = M(Bλ ) be the family of affinoid domains in X such that
x ∈ (Vλ )G and f (Vλ ) ⊆ U . By Remark 4.3.4(i), ΩX/S,x is the filtered colimit of
b B /A , and we provide it with the colimit seminorm,
the seminormed A-modules Ω
λ
that will be denoted k kx . If U ⊆ M(A′ ) is a larger affinoid domain then the
b B /A = Ω
b B /A′ respects the Kähler norms since they depend only
isomorphism Ω
λ
λ
on the seminorm of Bλ . In particular, k kx is independent of the choice of U , and
we have constructed intrinsic seminorms on the stalks of ΩX/S .
6.1.2. Fibers. Our next aim is to relate ΩX/S,x and its seminorm to modules of
differentials with Kähler seminorms. Set Ox = OXG ,x and Os = OSG ,s for shortness. In fact, ΩX/S,x is an (uncompleted) filtered colimit of completed modules of
differentials, so it can be informally thought of as a partial completion of ΩOx /Os .
40
MICHAEL TEMKIN
Consider the following commutative diagram of seminormed modules
ΩBλ /A
αλ
βλ
ΩOx /Os
αx
//Ω
b B /A
λ
(3)
❑❑❑
❑❑❑ψλ
γλ
❑❑❑
❑%
b H(x)/H(s) ,
/ ΩX/S,x ψx / Ω
where ψx is the contracting A-homomorphism induced by the A-homomorphisms
b B /A in the category of seminormed modules (see 3.1.13).
ψλ since ΩX/S,x = colimλ Ω
λ
Theorem 6.1.3. Keep the above notation. The homomorphism αx and ψx are
b H(x)/H(s) is the completion of both
isometries with dense images. In particular, Ω
ΩX/S,x and ΩOx /Os .
Proof. By Lemma 4.2.10, ΩOx /Os = colimλ ΩBλ /Aλ as seminormed modules. In
particular, αx is the colimit of isometries with dense images αλ , and hence αx
itself is an isometry with a dense image. Therefore, it suffices to show that the
b H(x)/H(s) is the completion homomorphism. Since
homomorphism ΩOx /Os → Ω
|mx |x = 0, the surjection
ΩOx /Os → ΩOx /Os /mx ΩOx /Os = ΩκG (x)/κG (s)
b κ (x)/κ (s) = Ω
b H(x)/H(s) by Corolis an isometry and it remains to use that Ω
G
G
lary 5.5.6.
Corollary 6.1.4. In the situation of Theorem 6.1.3, ψx identifies the completed
b
fiber Ω\
X/S (x) with ΩH(x)/H(s) .
Remark 6.1.5. (i) Corollary 6.1.4 provides an alternative way to define k kx .
b H(x)/H(s)
Instead of the colimit definition in 6.1.1, one can simply induce k kx from Ω
by the rule kφkx = kψx (φ)kΩ,H(x)/H(s)
for φ ∈ ΩX/S,x .
b
(ii) Even more importantly, the corollary provides a convenient way to comb H(x)/H(s) are often pretty
pute the values of k kx since the normed vector spaces Ω
explicit. For comparison, we note that the fiber ΩX/S (x) is not so manageable;
it is a (usually huge) vector space lying somewhere between ΩκG (x)/κG (s) and its
b H(x)/H(s) .
completion Ω
6.1.6. Semicontinuity. For any analytic domain U ⊆ X and a section φ ∈ ΩX/S (U )
we denote by kφk : UG → R+ the function that associates kφkx to x ∈ UG . The
following lemma asserts that the seminorms k kx provide ΩX/S with a seminorm
k k, that will be called the Kähler seminorm.
Lemma 6.1.7. For any choice of U ⊆ X and φ ∈ ΩX/S (U ), the function kφkΩ is
upper semicontinuous.
Proof. It suffices to show that kφk is upper semicontinuous locally at x ∈ UG .
Shrinking S and X at s = f (x) and x, respectively, we can make them affinoid, say
S = M(A) and X = M(B). Also, we can assume that U = X. Consider diagram
b B/A is a finite B-module, the
(3) with removed subscripts λ (i.e. B = Bλ ). Since Ω
completion map α is surjective and we can choose a preimage φ′ ∈ ΩB/A . Since the
maps in the bottom row of (3) are isometries by Theorem 6.1.3, kβ(φ′ )k = kγ(φ)k =
kφkx , where each seminorm is the Kähler seminorm of the appropriate module. Fix
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 41
Pn
any r with r > kφkx . Then there exist bi , ci ∈ Ox such that β(φ′ ) = i=1 bi dci
and maxi |bi ||ci | < r. Shrinking P
X at x we can achieve that bi , ci ∈ B, |ci |B |bi |B < r
for any i, and the equality φ′ = i bi dci holds in ΩB/A . Then kφk = kφk′ < r, and
b B/A → Ω
b H(y)/H(f (y)) is contracting, we obtain that
since for any y ∈ UG the map Ω
kφky < r. This completes the proof.
6.1.8. Kähler seminorms on pluriforms. By the constructions of Section 3.2.7, k kΩ
induces seminorms on the sheaves obtained from ΩX/S by tensor products, symmetric powers and exterior powers. In particular, it induces a canonical seminorm
k k(ΩlX )⊗m on the sheaf of pluriforms (ΩlX )⊗m , that will be called Kähler seminorm
too.
Of particular interest will be the situation when X → S isVquasi-smooth of relan
tive dimension n. Then the relative canonical sheaf ωX/S =
ΩX/S is invertible,
⊗m
as well as the relative pluricanonical sheaves ωX/S . The corresponding seminorms
will be denoted k kω and k kω⊗m
6.1.9. Analyticity of the seminorms. We have already used Theorem 5.5.5 when
proving the semicontinuity of the stalk seminorms. As another application, let us
show that all seminorms we have constructed are analytic.
Lemma 6.1.10. The seminorm k kΩ and the induced seminorms on the sheaves
obtained from ΩX/S by tensor products, symmetric powers and exterior powers are
analytic.
Proof. Assume that x ∈ X is a generization of y ∈ XG . By Lemma 2.3.6 the
embedding κG (y) ֒→ κG (x) induces an isomorphism of the completions, hence
b H(y)/H(f (y)) f
b H(x)/H(f (x)) .
h: Ω
→Ω
It follows from the construction of the maps ψx and ψy in 6.1.2 that they are
compatible with the generization g : ΩX/S,y → ΩX/S,x in the sense that ψx ◦ g =
h ◦ ψy . By Corollary 6.1.4, the seminorms on the stalks are induced via ψ’s, hence g
is an isometry and we obtain that k kΩ is analytic. By Lemma 3.3.11, this implies
that all other seminorms mentioned in the lemma are analytic as well.
6.2. Examples. In this section, we compute k kΩ and its completed fibers in few
basic cases. We try to choose simple examples that illustrate the general situation.
In particular, we will see that k kΩ discovers a rather subtle behaviour even in the
one-dimensional case.
6.2.1. The case of a disc. Assume that k = k a and X = M(k{T }) is a unit disc.
Then ΩX is a free sheaf with basis dT , so we can identify it with OX by sending
dT to 1. Let r(x) be the radius function, i.e. r(x) is the infimum of radii of
subdiscs of X containing x. We claim that k k = k kΩ is the seminorm, in fact a
norm, corresponding to r(x) in the sense of Lemma 3.3.8 and Corollary 3.3.9. In
particular, the maximality locus of kdT k consists of a single point – the maximal
point of X.
By Corollary 6.1.4, the claim reduces to checking that kdT kx = r(x). If x is
contained in a disc of radius s with center at a then |T − a|x ≤ s and hence
kdT kx = kd(T − a)kx ≤ s. The function kdT k : X → R+ is semicontinuous by
Lemma 6.1.7, therefore it suffices to check that kdT kx ≥ r(x) at a type 2 or 3 point
x. In such case, replacing T by a suitable T − a with a ∈ k we can achieve that
42
MICHAEL TEMKIN
|T |x = r(x) and the valuation on H(x) is T -monomial. But then k dT
T kΩ,H(x)/k = 1
by Theorem 5.6.8(i) =⇒ (iii), and so kdT kx = |T |x = r(x).
Remark 6.2.2. The radius function r(x) is upper semicontinuous but not continuous, and this is a typical behaviour for functions of the form kφk. This indicates
that the Kähler seminorm on ΩX/S is very different from the spectral seminorm on
OXG even when ΩX/S is invertible. For example, if X is a curve and f ∈ Γ(OXG )
is a global function then |f | is locally constant outside of a finite graph. On the
other hand, if φ ∈ Γ(ΩX ) then for any type two point x the value of kφk decreases
in almost all directions leading from x. This property is tightly related to the fact
that the maximality locus of such φ is a finite graph.
6.2.3. Rigid points: perfect ground field. Assume that k is perfect. Let x ∈ X be
ca , for example, a rigid point (i.e. a point with [H(x) :
any point with H(x) ⊆ k
k] < ∞), or a type 1 point on a curve. Since k a ∩ H(x) is dense in H(x) by Ax-Sen
b H(x)/k = 0. Therefore, any differential form φ satisfies kφkΩ,x = 0 by
theorem, Ω
Corollary 6.1.4.
6.2.4. Rigid points: non-perfect ground field. If k is not perfect then the situation
is different. For example, assume that x is a rigid point with l = H(x) inseparable
over k; a disc contains a plenty of such points. One can easily give examples when
Kähler seminorm on the finite dimensional vector space Ωl/k is a norm and hence
b l/k = Ωl/k 6= 0. (Probably, this is always the case since k is complete.) On the
Ω
b l/k is the completion of ΩX/k,x , so the seminorm on the latter does
other hand, Ω
not vanish.
Let us outline a concrete particular case. Assume that l = k(α) is a purely inseparable extension of k of degree p; in particular, Ωl/k = ldα. Then r = inf c∈k |α − c|
is positive since k is complete. Since dα = d(α − c), we obviously have that
kdαkΩ,l/k ≤ r. One can check straightforwardly that the choice of kdαk = r
makes the map dl/k contracting, so, in fact, kdαkΩ,l/k = r. In particular, if T is
the coordinate on X = A1k and x ∈ X is the type 1 point given by T p = a, then
kdT kx = s1/p , where s = inf c∈k |a − cp |.
e is not
6.2.5. Disc over non-perfect field. Assume that char(k) = p > 0 and k
◦
p
e
perfect and let X = M(k{T }). Choose any a ∈ k with e
a ∈
/ k and let x be
the rigid point given by T p = a. Then kdT kx = 1 by 6.2.4, and we claim that,
more generally, kdT k = 1 on the whole line connecting x with the maximal point
of X. In particular, the maximum locus of kdT k contains a huge subgraph, whose
combinatorial cardinality (i.e. the cardinality of the set of vertices and edges) is
easily seen to be equal to the cardinality of k.
To verify our claim, let q be the maximal point of the disc around x given by
[ ⊆ H(q), then the valuation on L is
|T p − a| ≤ r. Set S = T p − a and L = k(S)
S-monomial and hence inf c∈L |c + S| = 1. Since T = (a + S)1/p we obtain by 6.2.4
that kdT kΩ,H(q)/L = 1, and hence kdT kΩ,H(q)/k ≥ 1. The opposite inequality is
obvious, so kdT kq = kdT kΩ,H(q)/k = 1.
Remark 6.2.6. In the mixed characteristic case the situation is even weirder. On
the one hand, the Kähler seminorm vanishes at rigid points, but on the other hand,
if e
k is not perfect, say e
a ∈
/ e
k p , and I is the interval connecting x = (T p − a)
with the maximal point q then a similar argument shows that kdT k = 1 on some
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 43
neighborhood of q in I. In fact, the maximality locus of kdT k is a huge tree with
root q but its leaves are not rigid points.
6.3. Kähler seminorms and base changes.
6.3.1. Domination. For general base changes, Kähler seminorms are related as follows.
Lemma 6.3.2. Assume that X → S and S ′ → S are morphisms of k-analytic
spaces and X ′ = X ×S S ′ . Let φ ∈ Γ(ΩX/S ) and let φ′ ∈ Γ(ΩX ′ /S ′ ) be the pullback
of φ. Then for any point x′ ∈ X ′ with image x ∈ X one has that kφ′ (x′ )k ≤ kφ(x)k.
In other words, the pullback of k kΩ,X/S (see 3.3.4) dominates k kΩ,X ′ /S ′ .
Proof. Let s ∈ S and s′ ∈ S ′ be the images of x′ . Unrolling the definitions of
the Kähler seminorms and the pullback operation we see that the assertion of the
b H(x)/H(s) ⊗H(x) H(x′ ) → Ω
b H(x′ )/H(x) is
lemma reduces to the claim that the map Ω
contracting. The latter follows straightforwardly from the definition of the seminorm on the source: similarly to the argument in Lemma 4.2.5, all inequalities
defining the seminorm of the source hold in the target.
Remark 6.3.3. (i) The domination of seminorms from Lemma 6.3.2 is not an
equality in general. For example, assume that k = k a , S = M(k) and X = S ′ is
the unit disc over k, and let x = s′ be a point. It is easy to see that the fiber over
(x, s′ ) in X ′ contains a type 1 point x′ . So, k kx′ = 0 by Section 6.2.3. On the
other hand, if r(x) > 0 then k kx 6= 0.
(ii) At first glance, the above example is surprising because in the context of seminormed rings, Kähler seminorms are compatible with base changes by Lemma 4.2.5.
However, we use structure sheafs provided with the spectral seminorms in the definition of Kähler seminorms, while the tensor seminorms do not have to be spectral. For example, if E/k and F/k are finite extensions of analytic fields such that
K = E ⊗k F is a field, the tensor norm on K can be strictly larger than the valuation. As we will prove below, this phenomenon only happens when the extensions
are wild.
6.3.4. Universally spectral norms. Remark 6.3.3(ii) motivates the following definition. We say that a Banach k-algebra A is spectral if its norm is power-multiplicative.
In other words, | |A coincides with the spectral seminorm. We say that a Banach kb k l provided with the tensor seminorm is spectral
algebra is universally spectral if A⊗
for any extension of analytic fields l/k.
Remark 6.3.5. (i) The condition that a Banach algebra is spectral is an analogue
of reducedness in the usual ring theory. Thus, universal spectrality can be viewed
as an analogue of geometric reducedness over a field.
(ii) For comparison, we note that a point x ∈ X is called universal if the tensor
b k l is multiplicative for any l/k (originally, universal norms were
norm on H(x)⊗
called peaked, see [Ber90, Section 5.2]). The algebraic analogue of the property
that a Banach field is universal over k is geometrical integrality.
6.3.6. Defectless case. One can show that an algebraic extension is universally spectral if and only if it is almost tame, but this will be worked out elsewhere. Here we
only check this for defectless extensions.
44
MICHAEL TEMKIN
Lemma 6.3.7. Assume that K/k is a finite defectless extension of analytic fields.
Then K is universally spectral over k if and only if K/k is tame. In particular,
any tame extension is universally spectral.
Proof. Throughout the proof, given an analytic k-field l we set L = K ⊗k l and
kr be the R×
provide it with the tensor seminorm | |L . Let e
kgr = ⊕r>0 e
+ -graded
e gr and e
e ′gr be the R×
reduction of k, and define K
lgr similarly. Also, let L
+ -graded
e
e gr
ring associated with the filtration on L induced by | |L ; in particular, L′gr = L
only when | |L coincides with the spectral seminorm. Note that | |L is not spectral
if and only if |xn |L < |x|nL for some x and large n, and this happens if and only if
x
e is a homogeneous nilpotent element of L. Thus, | |L is spectral if and only if the
e ′gr is reduced in the sense of graded commutative algebra of [Tem04,
graded ring L
Section 1].
Since K/k is defectless, K possesses an orthogonal k-basis a1 , . . . ,an . Then, it
e gr
is also an orthogonal basis of L over l and hence e
a1 , . . . ,e
an is a basis of both K
e
e
e
e
e
e
over kgr and Lgr over lgr . In particular, we obtain that Lgr = Kgr ⊗ekgr lgr .
e e
Now, let us prove the lemma. The extension K/k is tame if and only if K/
k is
×
×
separable and |K |/|k | has no p-torsion. By [Duc13, Proposition 2.10 ] the latter
e gr /e
happens if and only if the extension of graded fields K
kgr is separable (in graded
setting). This implies that for any extension of analytic fields l/k, the graded e
lgr e gr ⊗e e
l
is
reduced,
and
as
we
saw
above
this
happens
if
and
only
if
the
algebra K
kgr gr
tensor seminorm on K ⊗k l is spectral.
e gr /e
Conversely, if K/k is wild then we showed in the above paragraph that K
kgr
ca (in fact, l = K would suffice). Then
is not separable. For simplicity take l = k
e
e gr ⊗e e
lgr is an algebraically closed graded field and K
kgr lgr is an inseparable finite
e
e
lgr is not reduced, and
lgr -algebra. By [Duc13, 1.14.3], this implies that Kgr ⊗e e
kgr
hence | |L is not spectral.
6.3.8. Compatibility with base changes. In view of Remark 6.3.3(i), one has to restrict base change morphisms in order to ensure compatibility of Kähler seminorms
with the base change. Here is a natural way to impose such a restriction.
Theorem 6.3.9. Let f : X → S and g : S ′ → S be two morphisms of analytic kspaces, and assume that for any point s′ ∈ S ′ with s = g(s′ ) the field H(s′ ) is finite
and universally spectral over H(s). Then the pullback of the Kähler seminorm on
ΩX/S coincides with the Kähler seminorm on ΩX×S S ′ /S ′ .
Proof. Fix a point x′ ∈ X ′ = X ×S S ′ and let x ∈ X, s′ ∈ S ′ and s ∈ S be its
images. Set also K = H(s), K ′ = H(s′ ), L = H(x), and L′1 = H(x′ ). We should
prove that if φ ∈ ΩX/S,x and φ′ ∈ ΩX ′ /S ′ ,x′ is its pullback then kφkx = kφ′ kx′ . By
b L/K and Ω
b L′ /K ′ ,
Corollary 6.1.4, the values of the seminorms can be computed at Ω
1
b L/K ⊗L L′1 →
respectively. This reduces the question to proving that the map ψ : Ω
b L′ /K ′ is an isometry.
Ω
1
b L/K ⊗L L′ f
b L′ /K ′ is an
Consider the seminormed ring L′ = L ⊗K K ′ , then λ : Ω
→Ω
′
isometry
Qnby Lemma 4.2.5. Since L is a finite spectral L-algebra, there is an isometry
L′ = i=1 L′i , where L′i are valued extensions of L, and L′1 is as defined earlier.
Qn
b L′ /K ′ ⊗L′ L′1 → Ω
b L′ /K ′ is
In particular, ΩL′ /K ′ = i=1 ΩL′i /K ′ and hence ψ ′ : Ω
1
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 45
an isometric isomorphism. Therefore, the composition ψ = ψ ′ ◦ (λ ⊗L′ L′1 ) is an
isometry.
Let us record two important particular cases of the theorem.
Corollary 6.3.10. Assume that X → S is a morphism of k-analytic spaces, s ∈ S
is a point and Xs is the fiber over s. Then k kXs /s equals to the restriction of
k kX/S onto the fiber.
Corollary 6.3.11. Assume that f : X → S is a morphism of k-analytic spaces and
l/k is a finite extension such that l is universally spectral over k (e.g., l/k is tame).
Set Xl = X ⊗ l and Sl = Y ⊗ l. Then k kΩ,Xl /Sl is the pullback of k kΩ,X/S .
Remark 6.3.12. (i) In fact, instead of finiteness of l/k it suffices to assume that
the extension is algebraic. We leave this to the interested reader.
(ii) In particular, the Kähler seminorm is preserved when we replace k with its
completed tame closure. In principle, this reduces the study of Kähler seminorms
to the case of a tamely closed ground field k.
6.3.13. Geometric Kähler seminorm. Given a morphism f : X → S, we denote
by f : X → S its ground field extension with respect to kca /k. Let g denote the
morphism X → X. If φ ∈ ΩX/S (U ), x ∈ U and x′ ∈ X is a preimage of x then
kg ∗ φkx′ does not depend on the choice of x′ and we set kφkΩ,x = kg ∗ φkx′ . It is
easy to see that the seminorms k kΩ,x satisfy the semicontinuity condition and hence
give rise to a seminorm k kΩ on ΩX/S , that we call the geometric Kähler seminorm.
Absolutely in the same way one defines geometric Kähler seminorm k k(Ωl )⊗m on
X/S
l
the sheaf of pluriforms (ΩX/S )⊗m .
Remark 6.3.14. (i) In fact, k kΩ,X/S is induced via the embedding ΩX/S ֒→
g∗ ΩX/S from the pushforward of the Kähler seminorm k kΩ,X/S in the sense of
Remark 3.2.10. Note also that k kΩ,X/S is the pullback of k kΩ,X/S .
(ii) If k possesses non-trivial wild extensions, Kähler seminorms on k-analytic
spaces can behave rather weird (see 6.2.4 and 6.2.5). So, in this case it is often
more useful to work with the geometric Kähler seminorm.
7. Metrization of pluricanonical forms
Until the end of the paper X is assumed to be quasi-smooth of pure dimension
n. In particular, ΩX = Ω1X/k is a locally free sheaf of rank n and the pluricanonical
Vn
⊗m
sheaves ωX
=(
ΩX )⊗m are invertible.
7.1. Monomiality of Kähler seminorms.
7.1.1. Stalks at monomial points. Recall that any monomial point possesses a family of tame monomial parameters by Lemma 2.5.4. This allows to describe k kω as
follows.
Theorem 7.1.2. Assume that k is algebraically closed. Let X be a quasi-smooth
k-analytic space of dimension n with a point x ∈ X, and let t1 , . . . ,tn be elements
dtn
1
of OXG ,x . Then k dt
t1 ∧ · · · ∧ tn kω,x ≤ 1 and the following conditions are equivalent:
dtn
1
(i) k dt
t1 ∧ · · · ∧ tn kω,x = 1.
b H(x)/k .
(ii) dt1 , . . . , dtn form an orthonormal basis of Ω
t1
tn
(iii) t is a family of tame monomial parameters at x.
46
MICHAEL TEMKIN
dtn
1
Proof. Since kdti kx ≤ |ti |x , it follows that k dt
t1 ∧ · · · ∧ tn kx ≤ 1 and (i)⇐⇒(ii).
We will complete the proof by showing that (ii)⇐⇒(iii).
Set l = k(t) and L = b
l. First, assume that ti form a tame monomial family.
dtn
1
b l/k by Theorem 5.6.8. It remains to
Then dt
,
.
.
.
,
is
an
orthonormal
basis of Ω
t1
tn
b l/k = Ω
b L/k by Corollary 5.5.6, and Ω
b L/k = Ω
b H(x)/k because H(x)/L is
use that Ω
tame.
dtn
1
b
Conversely, assume that dt
t1 , . . . , tn is an orthonormal basis of ΩH(x)/k . Since the
b L/k ⊗L H(x) → Ω
b H(x)/k is contracting and k dti kΩ,L/k ≤
isomorphism ψH(x)/L/k : Ω
ti
dtn
1
1, we obtain that ψH(x)/L/k is an isometry and dt
t1 , . . . , tn is an orthonormal basis
b L/k . The module Ωlog ◦ ◦ is almost torsion-free by Theorem 5.3.17, hence
of Ω
H(x) /k
H(x)/L is almost tame by Theorem 5.4.4. But L is stable by Remark 2.4.5(i), and
hence H(x)/L is tame by Theorem 5.3.13(ii).
It remains to show that t1 , . . . ,tn is a family of monomial parameters. Note
that t1 , . . . ,tn are algebraically independent over k because dt1 , . . . ,dtn are linearly
b H(x)/k , hence t1 , . . . ,tn is a separable transcendence basis of l/k.
independent in Ω
b l/k is surjective, and
Since Ωl/k is finite-dimensional, the completion Ωl/k → Ω
comparing the dimensions we see that it is an isomorphism. Thus, Ωl/k ⊗l L →
b L/k is an isometric isomorphism by Corollary 5.5.6. This implies that dt1 , . . . , dtn
Ω
t1
tn
is an orthonormal basis of Ωl/k , and hence the valuation on l is t-monomial by
Theorem 5.6.8.
Corollary 7.1.3. Assume that k is algebraically closed, X is quasi-smooth, t1 , . . . ,tn ∈
OXG ,x is a family of tame monomial parameters at a point x and F = (ΩlX )⊗m .
Then
dtj1
dtil
dtjl
dti1
⊗ ···⊗
∧ ···∧
∧··· ∧
B=
ti1
til
tj1
tjl
l ⊗m
is an orthonormal basis
P of Fx . In particular, any pluriform φ ∈ Γ((ΩX ) ) can be
represented as φ = e∈B φe e locally at x, and the following equality holds
kφk(ΩlX )⊗m ,x = max |φe |x .
e∈B
dtn
1
Proof. By Theorem 7.1.2, dt
t1 , . . . , tn is an orthonormal basis of ΩX,x . This reduces
the claim to a simple multilinear algebra.
Corollary 7.1.4. Keep the assumptions of Corollary 7.1.3 and assume that F =
⊗m
dtn ⊗m
1
ωX
. Then e = ( dt
is a basis of Fx , and if φ = f e is the represent1 ∧ · · · ∧ tn )
tation of a pluricanonical form φ at x then kφkω⊗m ,x = |f |x .
7.1.5. Piecewise monomiality. Now, we can prove that norms of pluriforms induce
RQ -PL functions on RQ -PL subspaces of X (2.5.8). In particular, when restricted
to RQ -PL subspaces of X spectral seminorm and Kähler seminorms demonstrate
similar behaviour, although their global behaviour is very different.
Theorem 7.1.6. Assume that X is quasi-smooth and φ ∈ Γ((ΩlX )⊗m ) is a pluriform on X, and consider the function kφk : X → R+ that sends x to the value
kφk(Ωl )⊗m ,x of the geometric Kähler seminorm. Then for any R-PL subspace
X
P ⊂ X the restriction of kφk onto P is an R-PL function.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 47
Proof. We start with the case when k = k a . By Theorem 2.6.8, we can cover S by
finitely many skeletons of tame monomial charts, hence it suffices to consider the
case when P itself is the skeleton of a tame monomial chart f : U → Gnm given by
×
t1 , . . . ,tn ∈ Γ(OU
). By Corollary 7.1.3, locally at a point x ∈ P we can represent
kφk as the maximum of RQ -PL functions |φi |. Hence kφk is RQ -PL too.
b k k a and let φ ∈ Γ((Ωl )⊗m ) be
Assume, now, that k is arbitrary. Set X = X ⊗
X
the pullback of φ. The preimage P ⊂ X of P is an R-PL subspace: just take the
charts of P and extend the ground field to kca . Also, it follows from the description
with charts that the map P → P is R-PL. Since the pullback of kφk(Ωl )⊗m ,x to
X
P is kφk(Ωl )⊗m ,x and the latter function is R-PL by the case of an algebraically
X
closed ground field, kφk(Ωl )⊗m ,x is an R-PL function as well.
X
Remark 7.1.7. (i) An analytic space is quasi-smooth at a monomial point if and
only if it is reduced at that point. In particular, the assumption that X is quasismooth in Theorem 7.1.6 can be replaced with the assumption that X is reduced.
(ii) Most probably, the assumption that X is reduced can also be removed. Also,
it seems probable that the theorem holds for Kähler seminorm too.
7.2. Comparison with weight norm of Mustaţă-Nicaise. In Section 7.2 we
assume that k is discretely-valued, and our aim is to compare k kω with the weight
norm a la Mustaţă and Nicaise, see [MN13].
7.2.1. Content of modules. Assume that M is torsion k ◦ -module. We define its
content cont(M ) ∈ [0, 1] as follows: if M is not finitely generated then cont(M ) = 0
and if M is finitely generated then cont(M ) = |a|, where (a) = F 0 (M ) is the zeroth
Fitting ideal of M .
Remark
7.2.2. (i) If M is finitely generated then M = ⊕ni=1 k ◦ /πi k ◦ and cont(M ) =
Qn
i=1 |πi |.
(ii) It is easy to extend the content function to the non-discrete case, see [Tem14].
Also, using the almost mathematics of Gabber-Ramero, one can generalize the
whole formation of Fittings ideals to the non-discrete case, see [GR03, Section 2.3].
7.2.3. Different. Assume that K/k is a geometrically reduced extension of realvalued fields. Let M be the torsion part of ΩK ◦ /k◦ . Then δK/k := cont(M ) ∈
log
[0, 1] will be called the different of K/k. We define the log different as δK/k
=
−1
δK/k |πK πk |, where πk is the uniformizer of k and similarly for πK . For completelog
ness, we also set δK/k
= δK/k = ∞ for any geometrically non-reduced extension.
Remark 7.2.4. (i) If k is discretely-valued and K/k is finite then this gives the
classical notions of the (absolute value of) different and log different. However, the
same definition applies to the non-discrete case once the content is defined. In the
log
latter case one should replace |πk | and |πK | with 1, hence δK/k
= δK/k .
log
(ii) In fact, δK/k
equals to the content of Ωlog
K ◦ /k◦ . We could not find this fact in
the literature in full generality, but it will be established in [Tem14].
(iii) Classically, one defines the different only for algebraic extensions, but it is
a reasonable invariant in the transcendental case too. For example, one can show
that if x ∈ A1k is the maximal point of a disc E then (the non-classical) δH(x)/k is
the minimum of δH(z)/k where z runs over rigid points in E. In the algebraic case,
48
MICHAEL TEMKIN
the different measures how wild the extension K/k is. This also provides a good
intuition for meaning of the different when K/k is not algebraic.
7.2.5. Weight seminorm. Assume, now, that K/k is a separable finitely generated
extension of transcendence degree n and K is discretely-valued, and let us recall
how a norm on ωK/k = ΩnK/k , is defined in [MN13]. We will call it the weight
e k)
e = n. Fix t1 , . . . ,tn ∈ K ◦ such that
norm and denote k kwt . Note that tr.deg.(K/
e
e
e
e
t1 , . . . ,tn is a transcendence basis of K/k. Then t1 , . . . ,tn is a transcendence basis
of K/k and replacing ti with elements of ti + K ◦◦ if needed we can also achieve
that K is separable over l = k(t1 , . . . ,tn ). Note that the valuation on l is Gauss,
and so l◦ is a localization of k ◦ [t1 , . . . ,tn ].
Since l◦ ֒→ K ◦ is a finite lci homomorphism, K ◦ = l◦ [s1 , . . . ,sm ]/(f1 , . . . ,fm )
and hence K ◦ is a localization of k ◦ [t, s]/(f ). By [MN13, 4.1.4], the canonical
∂fi
.
module ωK ◦ /k◦ is generated by ∆−1 φ, where φ = dt1 ∧ · · · ∧ dtn and ∆ = det ∂s
j
To describe k kwt it suffices to compute the norm of φ. The definitions of [MN13,
, where
4.2.3–4.2.5] introduce a log-norm that we call weight: wt(φ) = νk (∆)+1
e
e = eK/k and νk : k × → Z is the additive valuation of k (the weight function wtφ
considered in loc.cit. is a function of a point x ∈ X where K appears as the residue
field of x; it is the logarithmic analogue of the function kφk on X). So, we define
the weight norm by kφkwt = |∆πK |, where πK is a uniformizer of K. (The factor
1/e is only needed in the additive setting to make the group of values of K equal
⊗m
to e−1 Z, so that νk agrees with νK .) The weight norm on ωK/k
is defined via
⊗m
m
kφ kwt⊗m = (kφkwt ) .
Note that ΩK ◦ /l◦ is the cokernel of the map
d
where dfi =
Pn
m
◦
◦
⊕m
i=1 fi K → ⊕j=1 dsj K ,
∂fi
j=1 ∂sj .
It follows that ∆ = cont(ΩK ◦ /l◦ ) = δK/l .
7.2.6. Comparison. The norms k kω and k kwt differ by a factor, so to compare them
it suffices to evaluate the Kähler norm at φ. Since |ti | = 1, Theorem 5.6.8 implies
◦
◦
that dt1 , . . . ,dtn is a basis of the K ◦ -module Ωlog
l◦ /k◦ ⊗l K . Also, by Theorem 5.1.11,
(Ωlog
K ◦ /k◦ )tf is the unit ball of k kω,K/k . It follows by simple linear algebra that
kφkω,K/k = cont(M ), where
◦
◦
.
/ Ωlog
M = Ωlog
l◦ /k◦ ⊗L l
K ◦ /k◦
tf
We will show in [Tem14] that since K/l is finite separable, the first fundamental
sequence starts with an embedding
log
log
◦
◦
0 → Ωlog
l◦ /k◦ ⊗l K → ΩK ◦ /k◦ → ΩK ◦ /l◦ → 0.
(For usual differentials this follows from [GR03, Theorem 6.3.23] but the logarithmic
case is uncovered by loc.cit.) It follows that
log
log
log
= δK/l
/δK/k
cont(M ) = cont Ωlog
K ◦ /l◦ /cont tor ΩK ◦ /k◦
log
log
log
and we obtain that kφkω = δK/l
/δK/k
. Since δK/l
= δK/l |πK πl−1 | and |πl | = |πk |, we
log −1
log
obtain that kφkω = (δK/k
) ∆|πK πk−1 |. Thus, k kwt = |πk |δK/k
k kω and twisting
by m we obtain the following comparison result.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 49
Theorem 7.2.7. If k is discretely-valued, X is quasi-smooth and x ∈ X is a
divisorial point, i.e. a monomial point with discretely-valued K = H(x), then the
Kähler and the weight norms on m-canonical forms are related by
m
log
k kω⊗m .
k kwt⊗m = |πk |m δK/k
Remark 7.2.8. The constant factor |πk | in the formula for k kwt is analogous to
the −1 shift in [MN13, 4.5.3], while the different factor is subtle. We will show
log
in [Tem14] that the function δ log (x) = δH(x)/k
is upper semicontinuous on X. In
log
particular, the seminorms k kwt,x = |πk |δH(x)/k
k kω,x define a seminorm k kwt,X on
ωX , that we call the weight seminorm. Furthermore, we will show that k kwt,X is
R-PL, hence it is obtained from its restriction onto the divisorial points as a unique
PL extension. In particular, it coincides with the weight norm of Mustaţă-Nicaise.
The advantage of our definition is that it is local analytic, and applies to any ground
field k (if k is not discretely-valued, one should replace |πk | with 1).
7.3. Maximality locus of pluricanonical forms. In this section we study the
maximality loci M φ of pluricanonical forms φ with respect to the geometric Kähler
seminorm. Recall that M φ is closed by Lemmas 6.1.10 and 3.3.8. The main results
of this section, including Theorems 7.3.4 and 7.3.9, do not hold for the maximality
locus Mφ of the Kähler seminorm of φ, at least when the residue field is not perfect;
the counterexamples being as in Section 6.2.5 and Remark 6.2.6.
7.3.1. The torus case. We start with studying the standard pluricanonical form on
a torus.
Lemma 7.3.2. Let X be the k-analytic torus Gn,an
with coordinates t1 , . . . ,tn .
m
dtn ⊗m
1
∧
·
·
·
∧
)
. Then kφkω⊗m ,x ≤ 1 for
Consider the pluricanonical form φ = ( dt
t1
tn
any x ∈ X and the equality takes place if and only if x is a generalized Gauss point.
n
In particular, the maximum locus of kφkω⊗m is the skeleton (R×
+ ) of X.
Proof. This immediately follows from Theorem 7.1.2.
7.3.3. The semistable case. Recall that the skeleton S(X) ⊂ X associated with a
strictly semistable formal model has a natural structure of a simplicial complex.
Theorem 7.3.4. Assume that X is a quasi-smooth compact strictly k-analytic
⊗m
space, φ ∈ Γ(ωX
) is a pluricanonical form on X, X is a strictly semistable formal
model of X, and S(X) ⊂ X is the skeleton associated with X. Then the maximality
locus M φ is a union of faces of S(X).
Proof. Set k k = k kω⊗m for shortness. First, let us prove that M φ ⊆ S(X).
Working locally on X we can assume that there exists an étale morphism
g : X → Y = Spf(k ◦ {t0 , . . . ,tn }/(t0 . . . tn − a)),
where 0 6= a ∈ k ◦ . Since Y = Yη is a domain in Gnm with coordinates t1 , . . . ,tn
dtn ⊗m
1
is a nowhere vanishing pluricanonical form on Y
(see 2.7.3), e = ( dt
t1 ∧ · · · ∧ tn )
and therefore φ = he for a function h ∈ Γ(OX ). By [Ber99, Theorem 5.3.2(ii)], the
retraction r : X → S(X) is compatible with domination, in particular, |h|x ≤ |h|r(x) .
Since kφkx = |h|x kekx, it remains to prove that kekx ≤ kekr(x) and the equality
holds if and only if x = r(x), i.e. x ∈ S(X).
50
MICHAEL TEMKIN
When working with e, we can also view it as a form on Y . Let g : X → Y
be the generic fiber of g. Since g is étale, if x ∈ X and y = g(x) then the extension H(x)/H(y) is unramified by [Ber99, Lemma 1.6]. So, Theorem 6.3.9 and
Lemma 6.3.7 imply that kekx = keky . By Corollary 7.3.2, keky ≤ 1 for any point
y ∈ Y and the equality holds precisely for the points of S(Y). So, kekx ≤ 1 for any
x ∈ X and the equality holds precisely for the points of g −1 (S(Y)) = S(X).
It remains to show that if ∆ is a face of S(X) then either ∆ ⊂ M φ or the interior
∆◦ is disjoint from M φ . This claim is local on X so we can assume that X = Spf(A◦ )
is affine and there is an étale morphism g : X → Y as above. In particular, we can
assume that φ = he, as above, and so kφkx = |h|x at any point x ∈ S(X). Since
A is strictly affinoid and |k × | is divisible, |h|A ∈ |k × |. So, multiplying φ by an
element of k × we can achieve that |h|A = 1, and so h ∈ A◦ is a function on X.
Note that ∆◦ is the preimage in S(X) of a point x ∈ X. If e
h(x) = 0 then |h|x < 1 at
◦
any point in the preimage of x in X, and hence ∆ ∩ M φ = ∅. Otherwise, |h|x = 1
for any x as above, hence ∆◦ ⊆ M φ and by the closedness of M φ we obtain that
∆ ⊆ M φ.
Remark 7.3.5. (i) Theorem 7.3.4 is an analogue of [MN13, Th. 4.5.5], though it
applies to a wider context (e.g., ground field is arbitrary). Remind, however, that
these results consider different seminorms when char(e
k) > 0.
(iii) The lemma can be extended to general semistable models at cost of considering generalized simplicial complexes. Moreover, it should extend to arbitrary log
smooth formal models, once the foundations are set (see Remark 2.7.4).
7.3.6. Tame coverings. Recall that topological tameness was defined in Section 2.3.11.
Lemma 7.3.7. Assume that X is a quasi-smooth compact strictly k-analytic space
admitting a topologically tame étale covering Y → X such that Y possesses a strictly
⊗m
semistable formal model. Then for any pluricanonical form φ ∈ Γ(ωX
) on X the
geometric maximality locus M φ is a compact RQ -PL subspace of X.
Proof. Let ψ ∈ Γ(ωY ) be the pullback of φ. By Theorem 6.3.9, the norm functions
kφkω⊗m and kψkω⊗m are compatible, and hence M φ = f (M ψ ). It remains to recall
that M ψ is a compact RQ -PL subspace by Theorem 7.3.4, and hence its image
under f is an RQ -PL subspace by [DT14, Proposition 2.1].
7.3.8. Residue characteristic zero. In order to use the previous lemma, one should
construct an appropriate covering Y → X. The author conjectures that any quasismooth strictly analytic space possesses a quasi-net of analytic subdomains that
admit a semistable model (this is an analogue of the local uniformization conjecture
in the desingularization theory). However, this seems to be out of reach when
char(e
k) > 0. Although the case of char(e
k) = 0 is relatively simple, it is missing in
the literature and deserves a short separate paper. We can avoid proving it here
in view of the following theorem of Hartl, see [Har03, Theorem 1.4]: there exist a
finite extension l/k and an étale morphism Y → Xl = X ⊗k l such that Y possesses
a strictly semistable formal model. The theorem does not provide any control on
tameness of f , but it is automatic whenever char(e
k) = 0.
k) = 0 and X is a quasi-smooth compact
Theorem 7.3.9. Assume that char(e
⊗m
strictly k-analytic space. Then for any pluricanonical form φ ∈ Γ(ωX
) on X,
the maximality locus Mφ is a compact RQ -PL subspace of X.
METRIZATION OF DIFFERENTIAL PLURIFORMS ON BERKOVICH ANALYTIC SPACES 51
Proof. In this case, there is no difference between k kω⊗m and k kω⊗m . Take Y and l
as in Hartl’s theorem then the composition Y → X ⊗k l → X is an étale morphism,
which is automatically topologically tame. It remains to use Lemma 7.3.7.
References
[AK00]
D. Abramovich and K. Karu, Weak semistable reduction in characteristic 0, Invent.
Math. 139 (2000), no. 2, 241–273. MR 1738451 (2001f:14021)
[BBK13] Oren Ben-Bassat and Kobi Kremnizer, Non-archimedean analytic geometry as relative
algebraic geometry, ArXiv e-prints (2013).
[Ber90] Vladimir G. Berkovich, Spectral theory and analytic geometry over non-Archimedean
fields, Mathematical Surveys and Monographs, vol. 33, American Mathematical Society,
Providence, RI, 1990. MR 1070709 (91k:32038)
[Ber93]
, Étale cohomology for non-Archimedean analytic spaces, Inst. Hautes Études
Sci. Publ. Math. (1993), no. 78, 5–161 (1994). MR 1259429 (95c:14017)
[Ber99]
, Smooth p-adic analytic spaces are locally contractible, Invent. Math. 137
(1999), no. 1, 1–84. MR 1702143 (2000i:14028)
[Ber04]
, Smooth p-adic analytic spaces are locally contractible. II, Geometric aspects of
Dwork theory. Vol. I, II, Walter de Gruyter GmbH & Co. KG, Berlin, 2004, pp. 293–370.
MR 2023293 (2005h:14057)
[CLD12] Antoine Chambert-Loir and Antoine Ducrois, Formes diffrentielles relles et courants
sur les espaces de Berkovich, ArXiv e-prints (2012).
[CT]
Brian Conrad and Michael Temkin, Descent for non-archimedean analytic spaces,
http://www.math.huji.ac.il/~ temkin/papers/Descent.pdf.
[DT14] Antoine Ducros and Amaury Thuillier, Squelettes et valuations monomiales, in preparation, 2014.
[Duc03] Antoine Ducros, Image réciproque du squelette par un morphisme entre espaces de
Berkovich de même dimension, Bull. Soc. Math. France 131 (2003), no. 4, 483–506.
MR 2044492 (2004m:14042)
, Variation de la dimension relative en géométrie analytique p-adique, Compos.
[Duc07]
Math. 143 (2007), no. 6, 1511–1532. MR 2371379 (2008j:14046)
[Duc12]
, Espaces de Berkovich, polytopes, squelettes et théorie des modèles, Confluentes
Math. 4 (2012), no. 4, 1250007, 57. MR 3020334
[Duc13]
, Toute forme modérément ramifiée d’un polydisque ouvert est triviale, Math.
Z. 273 (2013), no. 1-2, 331–353. MR 3010163
[Duc14]
,
Families
of
berkovich
spaces,
in
preparation,
2014,
https://www.imj-prg.fr/~ antoine.ducros/Families-Berko.pdf.
[GR03] Ofer Gabber and Lorenzo Ramero, Almost ring theory, Lecture Notes in Mathematics,
vol. 1800, Springer-Verlag, Berlin, 2003. MR 2004652 (2004k:13027)
[GR12] O. Gabber and L. Ramero, Foundations for almost ring theory, ArXiv e-prints (2012),
math.univ-lille1.fr/~ ramero/hodge.pdf.
[Gro67] A. Grothendieck, Éléments de géométrie algébrique., Inst. Hautes Études Sci. Publ.
Math. (1960-1967).
[Har03] Urs T. Hartl, Semi-stable models for rigid-analytic spaces, Manuscripta Math. 110
(2003), no. 3, 365–380. MR 1969007 (2004a:14029)
[HL12]
Ehud Hrushovski and Fransua Loeser, Non-archimedean tame topology and stably dominated types, ArXiv e-prints (2012).
[Kat89] Kazuya Kato, Logarithmic structures of Fontaine-Illusie, Algebraic analysis, geometry,
and number theory (Baltimore, MD, 1988), Johns Hopkins Univ. Press, Baltimore, MD,
1989, pp. 191–224. MR 1463703 (99b:14020)
[KS06]
Maxim Kontsevich and Yan Soibelman, Affine structures and non-Archimedean analytic
spaces, The unity of mathematics, Progr. Math., vol. 244, Birkhäuser Boston, Boston,
MA, 2006, pp. 321–385. MR 2181810 (2006j:14054)
[Kuh10] Franz-Viktor Kuhlmann, Elimination of ramification I: the generalized stability theorem, Trans. Amer. Math. Soc. 362 (2010), no. 11, 5697–5727.
MR 2661493
(2011e:12013)
52
[Mat80]
[MN13]
[NX13]
[Ogu]
[sga72a]
[sga72b]
[Tem04]
[Tem10]
[Tem11]
[Tem13]
[Tem14]
[vdPS95]
MICHAEL TEMKIN
Hideyuki Matsumura, Commutative algebra, second ed., Mathematics Lecture Note
Series, vol. 56, Benjamin/Cummings Publishing Co., Inc., Reading, Mass., 1980.
MR 575344 (82i:13003)
Mircea Mustaţă and Johannes Nicaise, Weight functions on non-archimedean analytic
spaces and the Kontsevich-Soibelman skeleton, ArXiv e-prints (2013).
Johannes Nicaise and Chenyang Xu, The essential skeleton of a degeneration of algebraic varieties, ArXiv e-prints (2013).
Arthur
Ogus,
Lectures
on
logarithmic
algebraic
geometry,
math.berkeley.edu/~ ogus/preprints/log_book/logbook.pd.
Théorie des topos et cohomologie étale des schémas. Tome 1: Théorie des topos,
Springer-Verlag, Berlin, 1972, Séminaire de Géométrie Algébrique du Bois-Marie 1963–
1964 (SGA 4), Dirigé par M. Artin, A. Grothendieck, et J. L. Verdier. Avec la collaboration de N. Bourbaki, P. Deligne et B. Saint-Donat, Lecture Notes in Mathematics,
vol. 269.
Théorie des topos et cohomologie étale des schémas. Tome 2, Springer-Verlag, Berlin,
1972, Séminaire de Géométrie Algébrique du Bois-Marie 1963–1964 (SGA 4), Dirigé par
M. Artin, A. Grothendieck et J. L. Verdier. Avec la collaboration de N. Bourbaki, P.
Deligne et B. Saint-Donat, Lecture Notes in Mathematics, vol. 270.
Michael Temkin, On local properties of non-Archimedean analytic spaces. II, Israel J.
Math. 140 (2004), 1–27. MR 2054837 (2005c:14030)
, Stable modification of relative curves, J. Algebraic Geom. 19 (2010), no. 4,
603–677. MR 2669727 (2011j:14064)
, Relative Riemann-Zariski spaces, Israel J. Math. 185 (2011), 1–42.
MR 2837126
, Inseparable local uniformization, J. Algebra 373 (2013), 65–119. MR 2995017
, Metrization of differential pluriforms on berkovich analytic spaces II, in preparation, 2014.
M. van der Put and P. Schneider, Points and topologies in rigid geometry, Math. Ann.
302 (1995), no. 1, 81–103. MR 1329448 (96k:32070)
Einstein Institute of Mathematics, The Hebrew University of Jerusalem, Giv’at Ram,
Jerusalem, 91904, Israel
E-mail address: temkin@math.huji.ac.il