Burnside rings and volume forms with logarithmic

poles
Antoine Chambert-Loir
Université Paris Cité, Institut de Mathématiques de Jussieu-Paris Rive Gauche, F-75013,
Paris, France
arXiv:2301.02899v2 [math.AG] 13 Jan 2023

E-mail: antoine.chambert-loir@ u-paris.fr

Maxim Kontsevich
Institut des Hautes Études Scientifiques, 35 route de Chartres, 91440 Bures-sur-Yvette,
France
E-mail: maxim@ ihes.fr

Yuri Tschinkel
Courant Institute, NYU, 251 Mercer St. New York, NY 10012, USA
Simons Foundation, 160 5th Av., New York, NY 10010, USA
E-mail: tschinkel@ cims.nyu.edu

Abstract. — We develop a theory of Burnside rings in the context of birational equivalences

of algebraic varieties equipped with logarithmic volume forms. We introduce a residue
homomorphism and construct an additive invariant of birational morphisms. We also define
a specialization homomorphism.
Résumé. — Nous proposons une théorie d’anneaux de Burnside dans le contexte de la
géométrie birationnelle des variétés algébriques munies d’une forme volume à pôles logarith-
miques. Nous introduisons un homomorphisme « résidu », construisons un invariant additif
des morphismes birationnels. Nous définissons aussi un homomorphisme de spécialisation.

2000 Mathematics Subject Classification. — 14E08, 14E07, 14D06.

Key words and phrases. — Birational geometry, Burnside rings, logarithmic volume
forms, specialization.

Contents
1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Logarithmic differential forms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3. Burnside rings for logarithmic forms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
4. Residues. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
5. A complex of Burnside rings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
6. Algebraic structure of Burn(k) after localization at 2 . . . . . . . . . . . 17
7. Birational morphisms preserving volume forms. . . . . . . . . . . . . . . . . . 20
8. Specialization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

1. Introduction
The study of birationality of algebraic varieties is a classical and well studied
subject, with many open problems. In some cases, it is interesting to study birational
maps preserving additional structure, for example group actions, symplectic forms,
or volume forms. Such a study is already implicit in many questions of birational
geometry, eg, in the notion of crepant resolution of singularities.
In this paper, we consider the case of varieties endowed with volume forms with
logarithmic poles and develop a formalism of Burnside rings along the lines of their
counterpart introduced by Kontsevich & Tschinkel (2019) to establish the spe-
cialization of rationality, and its equivariant version by Kresch & Tschinkel
(2022b).
Let k be a field of characteristic zero. For each integer n, we define
Burnn (k)
as the free abelian group on birational equivalence classses of pairs (X, ω) consisting
of an integral smooth proper k-variety X of dimension n equipped with an n-form ω
with at most logarithmic poles.
The graded abelian group
M
Burn(k) = Burnn (k)
n∈N

carries a ring structure, induced by taking products of varieties, decomposed into ir-
reducible components, and equipped with the external product of the volume forms.
In section 4, we define morphisms of abelian groups
∂ : Burnn (k) → Burnn−1 (k).
When X is smooth and the polar divisor of ω has strict normal crossings, the image
of the class [X, ω] is given by the following formula. Let (Dα )α∈A be the family
of irreducible components
T of the polar divisor of ω. For each subset A of A , the
intersection DA = α∈A Dα is a union of integral smooth varieties of codimension |A|;
taking iterated residues, we may equip it with a volume form with logarithmic
poles ωA . Then
X
∂([X, ω]) = (−1)|A|−1[DA , ωA ] · T|A|−1 ,
∅6=A⊆A

where
T = [P1 , dt/t].
In particular, the existence of the map ∂ relies on the birational invariance of this
expression, see theorem 4.7.
This construction is reminiscent of the boundary map in polar homology
(Khesin & Rosly, 2003; Khesin et al , 2004; Gorchinskiy & Rosly, 2015).
However, apart from the obvious difference that we only record birational types of
strata, rather than the strata themselves, our formula takes into account strata of
all codimensions, rather than those of codimension one.
 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 3

The map ∂ is additive. Furthermore, we prove in theorem 4.10 that

∂(a · b) = εn · ∂(a) · b + a · ∂(b) − T · ∂(a) · ∂(b),
when a ∈ Burnm (k) and b ∈ Burnn (k). Here ε is the class of the point Spec(k)
equipped with the volume form equal to −1.
In theorem 5.1, we show that
∂ ◦ ∂ = 0.
These formulas may look complicated. However, as we explain in §6, they simplify
significantly after inverting 2.
Inspired by the constructions of Lin et al (2020); Lin & Shinder (2022);
Kresch & Tschinkel (2022a), we define in §7 a homomorphism
c : Bir(X, ω) → Burnn−1 (k),
from the group of birational automorphisms of the pair (X, ω), where X is an n-
dimensional integral proper smooth variety equipped with a logarithmic volume
form ω. As in the above references, our map c is defined at the groupoid level of
birational maps preserving logarithmic volume forms.
When the birational isomorphism ϕ : (X, ω) 99K (Y, η) is described by a diagram
W
←

p q
←
→

ϕ
X → Y
←

of smooth proper integral k-varieties, with birational morphisms p and q, the two
logarithmic volume forms p∗ ω and q ∗ η on W are equal, and the element c(ϕ) ∈
Burnn−1 (k) is given by
X X
c(ϕ) = [E, p∗ ωE ] − [D, q ∗ ηD ]
E∈Exc(q) D∈Exc(p)

where Exc(p) is the set of irreducible components of the exceptional divisor of p,

and where, for each such component D, the logarithmic volume form p∗ ωD on D is
obtained by taking the residue of p∗ ω along D (and similarly for q).
Finally, consider a discrete valuation ring with residue field k and field of frac-
tions K and let t be a uniformizing element. In this context, we define a specialization
map
ρt : Burnn (K) → Burnn (k).
The image of the class [X, ω] involves the combinatorics of a good model (X , ω)
over the valuation ring, and a certain subcomplex of the Clemens complex of the
special fiber. In the particular case where X is smooth, the polar divisor of ω is
a relative divisor with normal crossings, and denoting by ωo the restriction of ω to
the special fiber Xo , one has
ρt ([X, ω]) = [Xo , ωo].
Note that the existence of such a specialization map implies, as in theorem 1
of Kontsevich & Tschinkel (2019), or as in (Nicaise & Shinder, 2019), that
4 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

birational equivalence of varieties with logarithmic volume forms is preserved under

“good specializations”.
In the geometric case, where the valuation is the local ring of a curve C at point o,
the construction of the specialization map can be viewed as a restriction to the
special fiber of a normalization of a global residue map ∂ that takes place on a proper
model whose special fiber is a divisor with normal crossings. The normalization
procedure extracts a subcomplex of the Clemens complex of the special fiber. A
similar situation appeared in the study of Tamagawa measures on analytic manifolds
(Chambert-Loir & Tschinkel, 2010).
Related constructions emerged from the seminal work of Kontsevich & Soibelman
(2006) inspired by mirror symmetry, and its subsequent developments, eg, by
Mustaţă & Nicaise (2015); Nicaise & Xu (2016); Boucksom & Jonsson
(2017); Jonsson & Nicaise (2020).
Our constructions use essentially only formal properties of the residue maps. Con-
sequently, one can envision analogous theories for logarithmic forms of smaller de-
gree, Milnor K-theory, or even for the cycle modules of Rost (1996).

Acknowledgments. — We are grateful to Hsueh-Yung Lin and Evgeny Shinder

for having pointed out a lapsus in §7. The third author was partially supported by
NSF grant 2000099.

2. Logarithmic differential forms

2.1. Kähler differentials. — Let k be a field of characteristic zero and let K be
a finitely generated extension of k; let n be its transcendence degree. The space of
Kähler differentials ΩK/k is the K-vector space generated by symbols da, for a ∈ K,
subject to the relations:
(1) For a ∈ k, one has da = 0;
(2) For a, b ∈ K, one has d(a + b) = da + db and d(ab) = adb + b(da).
For any integer m > 0, we may consider its mth exterior power Ωm K/k , which is
n


a K-vector space of dimension m ; in particular, it vanishes if m > n, Ω1K/k has
dimension n, and ΩnK/k has dimension one. One has Ω0K/k = k, canonically.
Elements of ΩnK/k , for n = tr. degk (K), are also called volume forms.
For a ∈ K× , we also write dlog a = da/a ∈ ΩK/k .

2.2. Models. — Let m be an integer and let ω ∈ Ωm K/k . A model of K is an

integral k-scheme X together with a k-isomorphism K ≃ k(X); we say that this
model is proper, resp. smooth if X is proper, resp. smooth over k. Given such a
model, ω induces a meromorphic global section ωX of Ωm X/k . The polar ideal of ωX
is the subsheaf of OX whose local sections are the a ∈ OX such that aωX is induced
by a regular m-form. Let D be the zero-locus of this ideal. Its complement U is
the largest open subscheme of X such that ωX is induced by a regular m-form on U.
 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 5

If X is smooth, then ωX is locally free, hence the scheme D is an effective divisor

(Hartogs’s principle); we call it the polar divisor of ω on X.

2.3. Logarithmic forms. — By Hironaka’s theorem on embedded resolution of

singularities, there exist smooth projective models (X, ωX ) of (K, ω) such that the
polar divisor D of ωX has normal crossings.
Following (Deligne, 1970, chap. 2, §3), we then say that ωX has at most loga-
rithmic poles, or that ω has at most logarithmic poles on X, if both ωX and dωX
have at most simple poles along D.
The following lemma implies that this condition is essentially independent of the
choice of X such that the polar divisor of ωX has normal crossings.

Lemma 2.4. — Let g : X′ → X be a morphism of smooth k-varieties, let D be a

divisor with normal crossings in X and let D′ be a divisor with normal crossings
in X′ such that D′ = g −1 (D). Let ω be a regular m-form on X D and let ω ′ = g ∗ ω.
(1) If ω has at most logarithmic poles along D, then ω ′ has at most logarithmic
poles along D′.
(2) The converse holds if g is proper and surjective.
Proof. — The first assertion is (Deligne, 1970, chap. II, prop. 3.2, (iv)). Let us
prove the second one.
Consider the generic point η of X and a point η ′ ∈ X′ D′ which is algebraic
over k(η). The Zariski closure X′1 of η ′ is proper and generically finite over X, and
D′1 = D′ ∩ X′1 is a divisor. There is a proper modification h : X′2 → X′1 such that
D′2 = h−1 (D′1 ) has normal crossings. By the first part, the form h∗ ω ′|X′1 has at most
logarithmic poles along D′2 . Replacing g by g ◦h, we may assume that g is generically
finite.
Since the sheaf of forms with at most logarithmic poles along D is locally free and
X is smooth, we can delete from X a subset of codimension at least 2. Thus, we
may assume that g is flat, D is smooth and irreducible, and g is étale outside of D.
It suffices to argue étale locally at the generic point of D. By the local description
of ramified morphisms, there are étale local coordinates (z1 , . . . , zn ) on X such that
Dred = V(z1 ), local coordinates (z1′ , . . . , zn′ ) on X′ such that g ∗ z1 = (z1′ )e , g ∗ z2 = z2′ ,
etc., where e is the ramification index of g along D. Let d be the order of the pole
of ω along D; write ω = α/z1d + β ∧ dz1 /z1d , where α, β are regular forms which do
not involve dz1 . Then

ω ′ = g ∗ ω = g ∗ α/(z1′ )de + e g ∗β ∧ dz1′ /(z1′ )1+(d−1)e .

Assume, by contradiction, that d > 2, so that de > 2 and 1 + (d − 1)e > 2. Since ω ′
has at most logarithmic poles along D, we get g ∗ α = 0 and g ∗β = 0. This implies
that both α and β are multiples of z1 , contradicting the hypothesis that d was the
order of the pole of ω along D. Therefore, d 6 1. This concludes the proof.
6 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

2.5. — We say that an m-form ω ∈ Ωm K/k is logarithmic if for all proper smooth
models X of K such that the polar divisor of ωX has normal crossings, the meromor-
phic differential form ωX has at most logarithmic poles.
By resolution of singularities (Hironaka, 1964), two models are dominated by a
third one, hence lemma 2.4, implies that it suffices that this condition is satisfied on
some proper smooth model for which the polar divisor of ωX has normal crossings.
Analogously, if X is a reduced k-variety, then we say that a meromorphic m-form ω
on X is logarithmic “everywhere” if for all proper birational models (X′ , ω ′ ) of (X, ω),
the meromorphic m-form ω ′ on X′ has at most logarithmic poles. It suffices that
this holds on one such model.

3. Burnside rings for logarithmic forms

3.1. Burnside rings. — Let k be a field of characteristic zero and n an integer
such that n > 0. Kontsevich & Tschinkel (2019) defined the Burnside group
Burnn (k) as the free abelian group on isomorphism classes of finitely generated
extensions of k of transcendence degree n.
Any integral k-variety X of dimension n has a class [X] in Burnn (k). This gives rise
to alternative useful presentations of Burnn (k), for example involving only classes
of integral projective smooth varieties.
The group
M
Burn(k) = Burnn (k)
n≥0

carries a natural commutative ring structure, with multiplication defined by taking

products of (smooth projective) k-varieties:
[X] · [X′ ] = [X × X′ ].

3.2. Definition of a Burnside group for volume forms. — Let k be a field

of characteristic zero and let n be an integer > 0. We define Burnn (k) to be the
free abelian group on isomorphisms classes of pairs (K, ω), where
– K is a finitely generated extension of k of transcendence degree n and
– ω ∈ ΩnK/k is a logarithmic volume form.
We write
[K, ω] ∈ Burnn (k)
for the class of a pair (K, ω).

Remark 3.3. — This definition has obvious more geometric formulations. For
example, we can take for generators equivalence classes of pairs (X, ω), where
– X is a smooth integral k-scheme of dimension n, and
– ω a regular volume form on X which is logarithmic “everywhere”,
 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 7

modulo the smallest equivalence relation that identifies (X, ω) and (X′ , ω ′ ) if there
exists an open immersion f : X′ → X such that ω ′ = f ∗ ω.
Alternatively, we can assume that X is proper, smooth and integral, the form ω is
a logarithmic volume form on X, and consider the smallest equivalence relation that
identifies (X, ω) and (X′ , ω ′) if there exists a proper birational morphism f : X′ → X
such that ω ′ = f ∗ ω. By the weak factorization theorem of (Abramovich et al ,
2002), this equivalence relation is generated by such morphisms f which are blowing-
ups along smooth centers in good position with respect to the polar divisor of X.
In both contexts, if X is an n-dimensional k-variety and ω is a meromorphic n-form
on X which is logarithmic “everywhere”, then we define [X, ω] to be the sum, over
all irreducible components Y of X which have dimension n, of the classes [Y, ω|Y ].

Example 3.4. — Finitely generated extensions of k of transcendence degree 0 are

finite extensions of k. Let K be such an extension. Since k has characteristic zero,
one has Ω1K/k = 0. However, Ω0K/k , which is its 0th exterior power, is canonically
isomorphic to K. Consequently, Burn0 (k) is the free abelian group on isomorphism
classes of pairs (K, λ), where K is a finite extension of k and λ ∈ K.
We will let 1 = [Spec(k), 1] and ε = [Spec(k), −1].

Example 3.5. — Let K = k(t). The differential form dt/t is a logarithmic volume
form; indeed X = P1k is a model of K and this form has poles of order 1 at 0 and ∞,
and no other poles. We write T for the class of (k(t), dt/t).
Note that the k-isomorphism of K that maps t to 1/t maps dt/t to its opposite;
consequently, we also have T = [k(t), −dt/t] = ε · T.
In the context of birational geometry in presence of logarithmic volume forms,
“rational varieties” would have class in Tn , and similarly for stable birationality.

3.6. Multiplicative structure. — We view the direct sum

M
Burn(k) = Burnn (k)
n∈N

as a graded abelian group. It is endowed with a multiplication such that

[X, ω] · [X′ , ω ′] = [X × X′ , ω ∧ ω ′]
when X, X′ are proper, smooth and integral and ω, resp. ω ′ are logarithmic volume
forms on X, resp. X′ , and Y ranges over the set of irreducible components of X × X′ .
Let s : X′ × X → X × X′ be the isomorphism exchanging the two factors. One has
′
s∗ (ω ∧ ω ′ ) = (−1)nn ω ′ ∧ ω,
if n = dim(X), n′ = dim(X′ ), ω is a volume form on X and ω ′ is a volume form
on X′ . Consequently,
′
a · b = εnn · b · a
for a ∈ Burnn (k) and b ∈ Burnn′ (k). In particular, classes in Burnn (k), for even n,
are central in Burn(k).
8 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

We remark that the element T ∈ Burn1 (k) is central as well. Let indeed a ∈
Burnn (k). If n is even, then a · T = T · a. Otherwise, we have a · T = ε · T · a, but
we have seen in example 3.5 that T = ε · T. As a consequence, a · T = T · a.
However, the ring Burn(k) is not commutative. Indeed, consider curves X, X′
without automorphisms and no nonconstant morphism between them. Then the
switch is the only isomorphism from X′ × X to X × X′ . Take nonzero logarithmic
1-forms ω, ω ′ on X, X′ respectively. The classes [X × X′ , ω ∧ ω ′ ] and [X′ × X, ω ′ ∧ ω]
are then distinct.

3.7. Functoriality. — Let k ′ be an extension of k. Then there is a natural ring

homomorphism
Burn(k) → Burn(k ′ )
described as follows. Let (X, ω) be an integral k-variety of dimension n equiped with
a logarithmic q-form. Let X′ = X ⊗k k ′ be its base change to k ′ , and let ω ′ be the
volume form on X′ deduced from ω by base change. Then the class of (X, ω) maps to
the sum of classes (Y, ω ′ |Y ), where Y runs the (finite) set of irreducible components
of X′ .
If k ′ is a finite extension of k, we also have a trace map
Trk′ /k : Burn(k ′ ) → Burn(k)
obtained by averaging over a set of representatives of automorphisms of the Galois
closure of k ′ over k modulo those preserving k ′ .

3.8. Relation with the classical Burnside group. — Forgetting the form ω
gives a ring morphism
π : Burn(k) → Burn(k).
On the other hand, if K is a finitely generated extension of k of transcendence
degree n, we can endow it with the zero n-form. The resulting map
̟ : Burn(k) → Burn(k)
identifies Burn(k) with an ideal of Burn(k). One has π ◦ ̟ = id.

3.9. Variations on the theme. — The construction of the Burnside ring Burn(k)
admits several natural variants that are relevant in more specific contexts. Some of
them will be used in later sections.
3.9.1. A relative ring. — Let n be an integer. For any k-scheme S, we define
Burnn (S/k) as the free abelian group on triples (X, ω, u) where X is an integral
smooth n-dimensional k-scheme, ω ∈ ΩnX/k is a regular volume form which is loga-
rithmic “everywhere”, and u : X → S is a morphism, modulo the smallest equivalence
relation that identifies (X, ω, u) and (X′ , ω ′, u′) if there exists an open immersion
f : X′ → X such that ω ′ = f ∗ ω and u′ = u ◦ f .
Let h : S → T be a morphism of k-schemes. It induces a morphism of abelian
groups
h∗ : Burnn (S/k) → Burnn (T/k)
 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 9

such that h∗ ([X, ω, u]) = [X, ω, h ◦ u] for any triple (X, ω, u) as above.
3.9.2. Pluriforms. — One can replace volume forms with volume r-pluriforms, that
is, elements of (ΩnK/k )⊗r , for some given integer r. The corresponding logarithmic
condition requires that the pluriform has poles of order at most r on an adequate
model. Note that when r is even, the obtained ring is commmutative.
3.9.3. Forms up to scalars. — In the construction, we may wish to identify (K, ω)
and (K′ , ω ′ ) if there exists λ ∈ k × , resp. λ ∈ {±1}, and a k-isomorphism f : K → K′
such that f ∗ ω ′ = λω. These variants also give rise to a commutative ring.
3.9.4. Group actions. — Let G be a profinite group scheme over k. One can also
consider pairs (K, ω), where the field K is endowed with an action of G leaving the
form ω invariant. The obtained ring will be denoted by BurnG (k).

4. Residues
4.1. Residue of a volume form. — Let X be an equidimensional smooth k-
variety of dimension n.
Let D be a smooth divisor on X. We denote by Ωm X/k (log D) the sheaf of m-forms
on X with logarithmic poles along D, locally of the form η ∧ d log f + η ′ , where η and
η ′ are regular and f is a local equation of D. The residue map is the homomorphism
of OX -modules
ρD : Ωm m−1
X/k (log D) → ΩD/k ,
characterized by the relation
ρD (η ∧ d log f + η ′ ) = η|D
for every local sections η ∈ Ωm−1 ′ m
X/k and η ∈ ΩX/k , and any local generator f of the
ideal of D.
If ω is a logarithmic m-form on X, there is an open subset U of X such that
U ∩ D 6= ∅ and such that ω|U belongs to ΩmX/k (log D). Its residue ρD (ω|U ) is then a
m−1
meromorphic section of ΩD/k .

Lemma 4.2. — Let ω be a logarithmic differential form of degree m on X. Then

ρD (ω) is a logarithmic (m − 1)-form on D.
Proof. — We may assume that the sum of D and of the polar divisor of ω has strict
normal crossings. The assertion is then evident in local coordinates.

4.3. Blowing-ups and normal bundles. — Let Y be a smooth closed sub-

scheme of X. The blow-up BlY (X) of X along Y is a smooth k-variety. The blowing-
up morphism bY : BlY (X) → X is an isomorphism over the complement of Y. If Y
is nowhere dense and nonempty, then EY = b−1 Y (Y) is a smooth divisor in BlY (X).
In general, EY = b−1
Y (Y) identifies, as an Y-scheme, with the projectivization of
the normal bundle NY (X) of Y in X.
10 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

Let W be a closed smooth subscheme of X. Assume that W and Y are transversal.

Then the Zariski closure of b−1Y (W (Y ∩ W)) is called the strict transform of W
in BlY (X). It identifies with BlY∩W (W).
Let now ω be a logarithmic m-form on X. Then the form b∗Y ω on BlY (X) is
logarithmic; assuming that Y is nonempty and nowhere dense, we can consider the
residue ρY (ω) of b∗Y ω along EY . It is a logarithmic (n − 1)-form on P(NY (X )).

Definition 4.4. — Let X be an irreducible proper smooth k-variety, let n be its

dimension and let ω ∈ ΩnX/k be logarithmic volume form whose polar divisor D has
strict normal crossings.
T Let (Dα )α∈A be the family of its irreducible components; for
A ⊆ A , we let DA = α∈A Dα. We then define an element ρ(X, ω) in Burnn−1 (X/k)
by the formula: X
ρ(X, ω) = (−1)|A|−1ρDA (X, ω).
∅6=A⊆A

(In this formula and all similar ones below, it is always implicit that the terms
where DA = ∅ are omitted.)

4.5. Iterated residues. — We retain the notation of definition 4.4

Fix a logarithmic volume form ω on X and a nonempty subset A of A such
that DA 6= ∅. It will be useful to compute inductively the logarithmic volume form
ρDA (ω) that appears in definition 4.4.
Let bA : X̃ → X be the blowing-up of X along DA and let E be its exceptional
divisor.
When A = {α} has a single element, DA is the divisor Dα , the blowing-up mor-
phism bA is an isomorphism and the exceptional divisor identifies with DA . Then
ρDA (X, ω) = [Dα , ρDα (ω), jα],
where jα is the immersion of Dα into X.
This construction can be pursued in higher codimension, using iterated residues.
Fix a total order on A . There is a unique, strictly increasing sequence (α1 , . . . , αm )
in A such that A = {α1 , . . . , αm }. Given the chosen order on A , we may apply the
iterated residues construction and obtain a logarithmic form of degree n − m
ρDA (ω) = ρDα1 ◦ · · · ◦ ρDαm (ω).
On a nonempty open subset U of X that meets DA , we may write
ω = η ∧ dlog(fα1 ) ∧ . . . dlog(fαm ),
for a regular form η, and then one has ρDA (ω) = η|U∩DA .
Denote by bA the blowing-up of X along DA and by EA its exceptional divisor;
recall that EA identifies with the projectivized normal bundle NDA (X) of DA in X.
Using local equations for the divisors Dα , for α ∈ A, we trivialize NDA (X) on a dense
open subscheme of DA ; this gives a birational isomorphism of EA with DA × Pm−1
(with m = |A|), and a local computation gives the formula
ρDA (X, ω) = [DA , ρDA (ω)] · Tm−1
 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 11

in Burnn−1 (DA /k).

When m > 2, the definition of ρDA actually depends on the chosen order of A , but
only up to a sign, so that the class [DA , ρDA (ω)] is well defined up to multiplication
by the class ε ∈ Burn0 (k). On the other hand, it is multiplied by Tm−1 and we
have observed that ε · T = T.

Proposition 4.6. — Let (X, ω), D, and (Dα )α∈A be as in definition 4.4. Let Y be
a strict irreducible
P subvariety of X; let AY be the set of all α ∈ A such that Y 6⊆ Dα ;
we assume that α∈AY Dα meets Y transversally.
Let g : X′ → X be the blowing-up of X along Y and let ω ′ = g ∗ω; it is a logarithmic
form, its polar divisor has strict normal crossings, and we have
g∗ ρ(X′ , ω ′) = ρ(X, ω)
in Burn(X/k).
Proof. — Let E = g −1 (Y) be the exceptional divisor; for each α ∈ A ,P
let D′α be the
strict transform of Dα . The blow-up X′ is smooth; the divisor E + α∈A D′α has
strict normal crossings and contains the polar divisor of ω ′ .
Let B be the set of all β ∈ A such that Y ⊆ Dβ , so that DB is the minimal
stratum containing Y.
We now split the discussion into two cases.
(1) Assume that dim(Y) < dim(DB ). Since g is ramified along E, its Jacobian
vanishes along E. Since ω has poles of order at most one, the form ω ′ = g ∗ω is
regular at the generic point of E. Consequently, the polar divisor of ω ′ does not
contain E and we have to compare
X
(−1)|A|−1ρD′A (ω ′)
∅6=A⊆A

with X
(−1)|A|−1 ρDA (ω).
∅6=A⊆A

Since g is a local isomorphism around the P generic points of Dα , for α ∈ A , we

see that the polar divisor of ω ′ is equal to α∈A D′α . For every nonempty subset A
of A , one has
g∗ ρDA (X′ , ω ′) = ρDA (X, ω)
for every nonempty subset A of A , which implies the desired formula in this case.
(2) Assume that dim(Y) = dim(DB ). In this case, Y is an irreducible component
of DB . Since D∅ = X and Y 6= X, we have B 6= ∅. We have to compare the expression
X X
(−1)|A|−1ρD′A (ω ′ ) + (−1)|A| ρE∩D′A (ω ′)
∅6=A⊆A A⊆A

with X
(−1)|A|−1 ρDA (ω).
∅6=A⊆A
12 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

The argument takes place in a neighborhood of Y, which allows us to assume that

Y = DB .
Let A be a nonempty subset of A . One has D′A = ∅ whenever B ⊆ A, and the
corresponding terms are absent from the second expression. On the other hand,
if B 6⊆ A, the morphism g identifies D′A with the blow-up of DA along DA ∩ Y =
DA∪B . In particular, g induces a birational isomorphism from D′A to DA , so that
g∗ ρD′A (X′ , ω ′ ) = ρDA (X, ω). Moreover, E ∩ D′A is the projectivized normal bundle
PNDA∪B (DA ), and
g∗ ρE∩D′A (X′ , ω ′) = ρDA∪B (X, ω).
Similarly, one has
g∗ ρE (X′ , ω ′) = ρDB (X, ω).
This gives a formula of the form
X X
g∗ ρ(X′ , ω ′) = (−1)|A|−1 ρDA (X, ω) + (−1)|A| ρDA∪B (X, ω)
∅6=A⊆A A⊆A
B6⊆A B6⊆A
X
= n′A ρDA (X, ω),
∅6=A⊆A

where 
(−1)|A|−1 if B 6⊆ A,
P
n′A = (−1)|C| if B ⊆ A.
 C⊆A
B6⊆C
C∪B=A
It suffices to prove that n′A = nA for any nonempty subset A of A . This is obvious
when B 6⊆ A, so let us assume that B ⊆ A. In the sum that defines n′A , we write
C = (C B) ∪ C′ , where C′ = C ∩ B is a subset of B; the condition C ∪ B = A means
C B = A B; the condition B 6⊆ C means C′ 6= B. Consequently, we have
X ′
n′A = (−1)|A B| (−1)|C |
C′ ⊆B
C′ 6=B
!
X
|C′ |
= (−1)|A B|
(−1) − (−1)|B|
C′ ⊆B


= (−1)|A B|
(1 − 1)|B| − (−1)|B|
= (−1)|A|−1,
since |B| > 1. This concludes the proof of the proposition.

Theorem 4.7. — Let (X, ω) be as in definition 4.4. If X is proper, then the image
of ρ(X, ω) in Burnn−1 (k) only depends on the class [X, ω] ∈ Burnn (k). It gives rise
to a morphism of abelian groups
∂n : Burnn (k) → Burnn−1 (k).
 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 13

Proof. — By the definition of Burnn (k) involving pairs (X, ω) where X is proper,
it suffices to consider two pairs (X, ω) and (X′ , ω ′) as in definition 4.4 which are
related by a proper birational morphism g : X′ → X such that g ∗ω = ω ′ . By the
weak factorization theorem of Abramovich et al (2002), in order to prove the
theorem, we may assume that g is a blowing-up of X along a smooth subvariety
which is transversal to the polar divisor of ω. In this case, proposition 4.6 asserts
that g∗ ρ(X′ , ω ′) = ρ(X, ω) in Burn(X/k). In particular, the images in Burn(k) of
ρ(X′ , ω ′ ) and ρ(X, ω) are equal.

Example 4.8. — The meromorphic differential form dt/t on P1k has residues 1
and −1 at 0 and ∞ respectively. By construction, we thus have
∂1 (T) = [Spec(k), 1] + [Spec(k), −1] = 1 + ε.
Let n be an integer such that n > 2 and let us compute ∂n (Tn ). We view Tn as
the class of Pn , with homogeneous coordinates [1 : x1 : . . . : xn ], and with the toric
differential form
ωn = (dx1 /x1 ) ∧ . . . (dxn /xn ).
Its divisor is the sum of the toric hyperplanes D0 , . . . , Dn . Each of these hyperplanes
identifies with Pn−1 , and ρDj (ωn ) is (−1)n−j ωn−1 . Let A = {0, . . . , n}. If A = A ,
then DA = ∅. Otherwise, we see by induction that DA is isomorphic to Pn−|A| and
ρDA (ωn ) identifies with ±ωn−|A| , so that

[DA , ρDA (ωn )] · T|A|−1 = [Gm n−1 , ±ωn−1 ] = Tn−1 ,

since n − 1 > 1. Then,
X
∂n (Tn ) = (−1)|A|−1 [DA , ρDA (ωn )] · T|A|−1
∅6=A⊆A
X
= (−1)|A|−1 Tn−1 .
∅6=A(A

Now,
(
X 2 if n is odd;
(−1)|A|−1 = 1 − (1 − 1)n+1 + (−1)n+1 =
∅6=A(A
0 if n is even.

We get ∂n (Tn ) = 2Tn−1 if n is odd and ∂n (Tn ) = 0 if n is even. (Remind that

n > 2.) Since T = ε · T, the following formula unifies the various cases: for n > 1,
we have
∂n (Tn ) = (1 + (−1)n−1 ε) · Tn−1 .

Proposition 4.9. — For every class b ∈ Burnn (k), we have

∂n+1 (b · T) = −∂n (b) · T + b · ∂1 T.
14 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

Proof. — We may assume that b = [X, ω], where X is a proper integral smooth
variety of dimension n, and ω is a logarithmic volume form on X whose polar divisor
has strict normal crossings. Let (Dα )α∈A be the family of its irreducible components.
We view b · T as the class of [X × P1 , ω ∧ dt/t]. The polar divisor of ω ∧ dt/t is equal
to X
Dα × P1 + X × {0} + X × {∞}.
α∈A

It has strict normal crossings, and its strata are of the form DA × P1 , for nonempty
A ⊆ A , or DA × {0}, or DA × {∞}, for A ⊆ A . This decomposes ∂n+1 (b × T) as
the sum of three terms.
The first one is
X
[DA × P1 , ρDA ×P1 (ω ∧ dt/t)] · T|A|−1.
∅6=A⊆A

For any nonempty subset A of A , one has

ρDA ×P1 (ω ∧ dt/t) = ±ρDA (ω) ∧ dt/t,
so that
[DA × P1 , ρDA ×P1 (ω ∧ dt/t)] · T|A|−1 = [DA , ρDA (ω)] · T · T|A|−1 .
Consequently, the first term equals ∂n (b) × T.
Write D0 = X × {0} and D∞ = X × {∞}, and identify both divisors to X. For a
subset A of A , we have
ρDA∪{0} (ω ∧ dt/t) = ρDA ◦ ρD0 (ω ∧ dt/t) = ρDA (ω).
Consequently, the second term is equal to
X
(−1)|A| [DA , ρDA (ω)] · T|A| = [X, ω] − ∂n (b) · T.
A⊆A

Similarly, the third term is equal to

[X, −ω] − ∂n (b) · T.
Summing up these three terms, we get
∂n+1 (b × T) = −∂n (b) · T + [X, ω] + [X, −ω].
We now recall that ∂1 (T) = [Spec(k), 1] + [Spec(k), −1], so that
[X, ω] + [X, −ω] = [X, ω] · ∂1 (T) = b · ∂1 (T).
This concludes the proof.

Theorem 4.10. — Let a ∈ Burnm (k) and b ∈ Burnn (k); we have

∂m+n (a · b) = εn · ∂m (a) · b + a · ∂n (b) − T · ∂m (a) · ∂n (b)
in Burnm+n−1 (k).
 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 15

Proof. — It suffices to treat the case where a and b are classes of proper integral
smooth varieties (X, ω), (Y, η), endowed with meromorphic volume forms whose
polar divisors have strict normal crossings and no multiplicities. Let (Dα )α∈A be
the irreducible components of the polar divisor of ω, let (Eβ )β∈B be the irreducible
components of the polar divisor of η. Then [X, ω] · [Y, η] is the class of [X × Y, ω ∧ η];
the polar divisor of ω ∧ η is equal to
X X
Dα × Y + X × Eβ .
α∈A β∈B

We fix a total order on the disjoint union of A and B such that the elements of A
are smaller than those of B. For any subsets A, B of A and B, observe that we
have
ρDA∪B (ω ∧ η) = ±ρDA (ω) ∧ ρEB (η),
where ρDA has to be understood as the identity when A is empty, and similarly
for ρEB . The sign is 1 when A = ∅; when B = ∅, it is equal to (−1)|A|n ; we won’t
need to use its explicit value in the other cases. Then we can write ∂([X, ω] · [Y, η])
as X
(−1)|A|+|B|−1[DA × EB , ±ρA (ω) ∧ ρEB (η)] · T|A∪B|−1
A⊆A
B⊆B
A∪B6=∅
and we split it into the sum of three terms, according to which B = ∅, or A = ∅, or
none of them is empty. The first two terms are respectively equal to
X
(−1)|A|−1[DA × Y, (−1)n|A| ρDA (ω) ∧ η] · T|A|−1 = ∂([X, (−1)n ω]) · [Y, η]
∅6=A⊆A

and X
(−1)|B|−1[X × EB , ω ∧ ρEB (η)] · T|B|−1 = [X, ω] · ∂([Y, η]),
∅6=B⊆B
since T belongs to the center of Burn(k). As for the third one, we obtain
X X
− (−1)|B|−1 (−1)|A|−1 [DA , ρDA (ω)] · [EB , ρEB (η)] · T|A|+|B|−2
∅6=B⊆B ∅6=A⊆A

which equals
−∂([X, ω]) · ∂([Y, η]) · T.
Finally, we get
∂m+n (a · b) = ∂m+n ([X, ω] · [Y, η])
= ∂m ([X, (−1)n ω) · [Y, η] + [X, ω] · ∂n ([Y, η])
− T · ∂m ([X, ω]) · ∂n ([Y, η])
= εn · ∂m (a) · b + a · ∂n (b) − T · ∂m (a) · ∂n (b)
as was to be shown.
In particular, using the computation of example 4.8, we obtain the following
generalization of proposition 4.9.
16 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

Corollary 4.11. — For any a ∈ Burnm (k) and any integer n, we have
(
∂m (a) · Tn if n is even;
∂m+n (a · Tn ) =
−∂m (a) · T + a · ∂n (T ) if n is odd.
n n

Remark 4.12. — For the variant of Burn(k) where we consider forms up to sign,
the formula of theorem 4.10 simplifies to
∂m+n (a · b) = ∂m (a) · b + a · ∂n (b) − T · ∂m (a) · ∂n (b).

5. A complex of Burnside rings

Theorem 5.1. — For any integer n > 2, we have
∂n−1 ◦ ∂n = 0.
In other words, the residue morphisms of Burnside groups give rise to a complex
· · · → Burnn (k) → Burnn−1 (k) → · · · → Burn1 (k) → Burn0 (k)
Proof. — It suffices to prove the following result: Let (X, ω) be an integral
proper smooth variety of dimension n equipped with a meromorphic volume
form ω whose polar divisor has strict normal crossings and no multiplicities; then
∂n−1 (∂n ([X, ω])) = 0.
Let (Dα )α∈A be the family of irreducible components of the polar divisor of ω
in X. By definition, one has
X
∂n ([X, ω]) = (−1)|A|−1 ρDA (X, ω).
∅6=A⊆A

Fix a total order on A . Let (α1 , . . . , αm ) be a strictly increasing sequence in A

and let A = {α1 , . . . , αm }. We have seen in §4.5 that ρDA (X, ω) can be defined via
iterated residue maps:
ρDA ([X, ω]) = [DA , ρDα1 ◦ · · · ◦ ρDαm (ω)] · T|A|−1 = [DA , ωA ] · T|A|−1
where we wrote ωA for the composition ρDα1 ◦ · · · ◦ ρDαm (ω). When |A| is odd, we
have
∂(ρDA ([X, ω])) = ∂([DA , ωA ]) · T|A|−1,
while when |A| is even, we have
∂(ρDA ([X, ω])) = −∂([DA , ωA ]) · Ta−1 + [DA , ωA ] · ∂(T|A|−1 ).
Consequently, we have
X X
∂ ◦ ∂([X, ω]) = ∂([DA , ωA ]) · T|A|−1 − [DA , ωA ] · ∂(T|A|−1 ).
∅6=A⊆A ∅6=A⊆A
|A| even
 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 17

P
The polar divisor of the form ωA on DA is equal to β6∈A Dβ ∩ DA , so that, by
definition (and computation of ∂ via iterated residues),
X
∂([DA , ωA ]) = (−1)|B|−1 [DAB , ωA∪B ] · T|B|−1.
∅6=B⊆∁A

Also, when A is nonempty and of even cardinality, ∂(T|A|−1 ) = 2T|A|−2. When we

put these two formulas into the antepenultimate one and collect the various terms,
we obtain X
∂ ◦ ∂([X, ω]) = nC [DC , ωC ] · T|C|−2,
C⊆A
26|C|

where X
nC = − (−1)|B| − 2δ|C| is even .
∅6=A,B
A∪B=C,A∩B=∅

In the first sum, the terms A = ∅ or B = ∅ are omitted, while if we put them in, we
obtain
|C|  
X X |C|
|B|
(−1) = (−1)b = (1 − 1)|C| = 0
A∪B=C b=0
b
A∩B=∅
since |C| > 1. Consequently,
nC = 1 + (−1)|C| − 2δ|C| is even = 0.
This concludes the proof.

6. Algebraic structure of Burn(k) after localization at 2

In this section, we study the algebraic structure of the Burnside ring Burn(k),
endowed with its elements ε, T and the operator ∂.
L
6.1. — By construction, Burn(k) = n>0 Burnn (k) is an associative unital Z>0 -
graded ring, ε ∈ Burn0 (k), T ∈ Burn1 (k) and ∂ is a homogeneous additive map
of degree −1. They satisfy the following relations, for homogeneous elements a, b ∈
Burn(k):
(1) b · a = ε|a||b| · a · b (§3.6);
(2) ε2 = 1 (example 3.4);
(3) T = ε·T (example 3.5);
(4) ∂(T) = 1 + ε (example 4.8);
(5) ∂(a · b) = ε|b| · ∂(a) · b + a · ∂(b) − T · ∂(a) · ∂(b) (theorem 4.10);
(6) ∂(∂(a)) = 0 (theorem 5.1).
18 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

By (1), the element ε is central, and by (2), we may view Burn(k) as an algebra over
Z[ε]/(ε2 − 1). After inverting 2, the algebra Burn(k) splits into two components
Burnε=1 (k) and Burnε=−1 (k), one over which ε = 1, and the other over which
ε = −1.
In the rest of this section, we implicitly assume that 2 is inverted, without changing
the notation.

6.2. Sector ε = −1. — Here, we have T = −T, hence T = 0 since 2 is invertible.

As a consequence, after replacing ∂ with ∂ ′ : a 7→ (−1)1+|a| ∂(a), one gets from (5)
the usual graded Leibniz rule
∂ ′ (a · b) = ∂ ′ (a) · b + (−1)|a| a · ∂ ′ (b)
and therefore Burnε=−1 (k) is a classical differential graded (super-)commutative
algebra, similar to, eg, the de Rham complex.

6.3. Sector ε = 1. — The algebra Burnε=1 is now commutative (and not graded
commutative). This reflects the intuition in our constructions that they speak about
volume forms (as opposed to top-degree differential forms) for which we have com-
mutativity (as reflected by the change of order of integration in multiple integrals).

Lemma 6.4. — The map F : a 7→ a − T · ∂(a) is a ring endomorphism of

Burnε=1 (k), and F2 = id. Moreover, one has F ◦ ∂ = ∂ = −∂ ◦ F.
Proof. — This map is additive. One has F(1) = 1 − T · ∂(1) = 1. Let us show
multiplicativity. Indeed, for a, b ∈ Burnε=1 (k), one has
F(a) · F(b) = (a − T · ∂(a)) · (b − T · ∂(b))
= a · b − T · ∂(a) · b − T · a · ∂(b) + T2 · ∂(a) · ∂(b)
= a · b − T · (∂(a) · b + a · ∂(b) − T · ∂(a) · ∂(b))
= a · b − T · ∂(a · b) (using (5))
= F(a · b).
Since ∂ 2 = 0, one has
F(∂(a)) = ∂(a) − T · ∂(∂(a)) = ∂(a).
On the other hand,
∂(F(a)) = ∂(a − T · ∂(a))
= ∂(a) − ∂(T · ∂(a))
= ∂(a) − ∂(T) · ∂(a) − T · ∂(∂(a)) + T · ∂(T) · ∂(∂(a))
= −∂(a)
using that ∂(T) = 2 and ∂ 2 = 0.
Consequently, for a ∈ Burnε=1 (k), we have
F2 (a) = F(a) − T · ∂(F(a)) = a − T · ∂(a) + T · ∂(a) = a
 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 19

since ∂ ◦ F = −∂.

6.5. — To simplify the notation, write B = Burnε=1 (k). Since F2 = id and 2 is

invertible, the algebra B splits as a direct sum
B = B+ ⊕ B− ,
such that F acts as id on B+ and as − id on B− . Moreover, B+ is a subalgebra.
Since the operator ∂ anticommutes with F, it induces maps
∂± : B+ → B− , ∂∓ : B− → B+ .
Note that
F(T) = T − T · ∂(T) = −T,
so that T ∈ B− . Consequently, the multiplication by T map induces two maps
t± : B+ → B− , t∓ : B− → B+ .

Lemma 6.6. — The map ∂ vanishes on B+ . Equivalently, ∂± = 0.

The maps 12 ∂∓ and t± are inverses the one of the other.
Proof. — For a ∈ B+ , one has ∂(a) = −∂(F(a)) = −∂(a), since ∂ ◦ F = −∂, hence
∂(a) = 0.
On the other hand, for a ∈ B+ , one has
∂(T · a) = 2 · a + T · ∂(a) − 2T · ∂(a) = 2a − T · ∂(a) = a + F(a) = 2a
while for a ∈ B− , we have
T · ∂(a) = a − F(a) = 2a.
This concludes the proof of the lemma.
In particular, we see that the cohomology of the differential ∂ vanishes in the
sector Burnε=1 (k) = B.

6.7. — It follows from the lemma that we have a ring isomorphism

B = B+ [t](t2 − T2 ),
from which we see that all the algebraic structure of B+ (namely δ, T, F) can be
canonically reconstructed from a unital commutative associative Z>0 -graded ring B+
endowed with an element in degree +2 (namely, the element T2 ).

Remark 6.8. — The situation clarifies even more if we invert the class T. Then
we can write ∂(a) = (a − F(a))/T, and all relations happen to follow from the fact
that F is an involution such that F(T) = −1. Indeed,
 
2 ∂(a) − F(∂(a)) 1 a − ∂(a) a − ∂(a)
∂ (a) = = − F( =0
T T T T
20 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

explains that ∂ 2 = 0. Moreover, for a, b ∈ B, we have

a · b − F(a · b) a · b − F(a) · F(b)
∂(a · b) = =
T T
a − F(a) b − F(b) a − F(a) b − F(b)
= ·b+a· −T· ·
T T T T
= ∂(a) · b + a · ∂(b) − T · ∂(a) · ∂(b).

7. Birational morphisms preserving volume forms

7.1. — Let (X, ωX ) be a smooth integral k-variety of dimension n equipped with a
logarithmic volume form, and let f : Y → X be a proper birational morphism.
Let E be an exceptional divisor in Y, that is, such that dim(f (E)) < dim(E).
By lemma 2.4, and lemma 4.2, the residue ρE (f ∗ ωX ) along E of the meromorphic
form f ∗ ω is a logarithmic volume form on E.
We define c(f ; X, ω) to be the sum of all such classes [E, ρE (f ∗ ω)] in the free
abelian group Burnn−1 (k).

Lemma 7.2. — Let g : Z → Y be a proper birational morphism of smooth integral

varieties of dimension n. Then g ◦ f is a proper birational morphism and one has

c(g ◦ f ; X, ω) = c(g; Y, f ∗ω) + c(f ; X, ω)

in Burnn−1 (k).
Proof. — An integral divisor F in Z is exceptional for g ◦ f if and only if one of the
two mutually excluding situations happen:
– The divisor F is exceptional for g;
– Or g(F) is a divisor in Y which is exceptional for f .
Moreover, any divisor E in Y which is exceptional for f appears once and only as
a divisor of the form g(F). The contribution of F to c(g ◦ f ; X, ω) is given by the
volume form ρF ((g ◦f )∗ ω). In the first case, we write ρF ((g ◦f )∗ω) = ρF (g ∗ (f ∗ ω)), so
that the contribution of F coincides with its contribution to the term c(g; Y, f ∗ ω).
In the second case, g induces a birational isomorphism from F to E = g(F); writing
ρF ((g ◦ f )∗ ω) = g ∗ (ρF (f ∗ ω)), we see that the contribution of F coincides with the
contribution of E to c(f ; X, ω). This concludes the proof.

7.3. — Let (X, ωX ) and (Y, ωY ) be proper smooth k-varieties equipped with loga-
rithmic volume forms and let

ϕ : (X, ωX ) 99K (Y, ωY )

 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 21

be a birational map preserving the volume forms. By definition, this means that
there exists a diagram
W

←
p q

←
→

→
ϕ
X → Y
←

of integral k-varieties such that that p and q are proper and birational, and such
that p∗ ω = q ∗ ω ′ on W. In this situation, we may assume that W is smooth.

Lemma 7.4. — With this notation, the element

c(ϕ) = c(q) − c(p) ∈ Burnn−1 (k)

only depends on the birational map ϕ, and not on the choice of the triple (W, p, q).
p q r s
Proof. — Consider two possible diagrams X ← −V− → Y and X ← −W− → Y describ-
ing ϕ. Considering for example a resolution of singularities U of V ×X W, we can fit
these two diagrams in a common commutative diagram of the following form:

U
←

u v
←
→

V ← q r W
←
←

p s
←
→

→
ϕ →
X → Y
←

The equalities p∗ ωX = q ∗ ωY and r ∗ ωX = s∗ ωY imply that

(p ◦ u)∗ ωX = u∗ p∗ ωX = u∗ q ∗ ωY = (q ◦ u)∗ ωY = (s ◦ v)∗ ωY .

By lemma 7.2, we then have

c(p) − c(q) = c(p ◦ u) − c(q ◦ u) = c(r ◦ v) − c(s ◦ v) = c(r) − c(s).

This concludes the proof.

Theorem 7.5. — If ψ : (Y, ωY ) 99K (Z, ωZ) is another birational map preserving
volume forms, then one has

c(ψ ◦ ϕ) = c(ψ) + c(ψ).

p q r s
Proof. — Consider two diagrams X ← − V − → Y and Y ← − W − → Y describing ϕ
and ϕ. Considering for example a resolution of singularities U of V ×Y W, we can
22 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

fit these two diagrams in a common commutative diagram of the following form:

←
u v

←
→

→
V W

←
←
p q r s

←
→

→
→

→
ϕ ψ
X → Y → Z
← ←

p◦u s◦v
and the diagram X ←−− U −−→ describes the birational map ψ ◦ ϕ. Since q ◦ u = r ◦ v,
we then have
c(ψ ◦ ϕ) = c(p ◦ u) − c(s ◦ v)
= c(p ◦ u) − c(q ◦ u) + c(r ◦ v) − c(s ◦ v)
= c(p) − c(q) + +c(r) − c(s)
= c(ϕ) + c(ψ),
as was to be shown.

Corollary 7.6. — Let Bir(X, ω) be the set of birational automorphisms of X pre-

serving ω. The map c induces a homomorphism of abelian groups
Bir(X, ω) → Burnn−1 (k).
Its kernel contains the group of automorphisms of X that preserve ω.

8. Specialization
Let K be the field of fractions of a discrete valuation ring R with residue field k.
Fix a uniformizer t ∈ R.
In this context, Kontsevich & Tschinkel (2019) have defined two (distinct)
specialization morphisms
ρt : Burnn (K) → Burnn (k),
relating the Burnside groups of K and k (see 3.1), one of which is a ring homo-
morphism. (The latter homomorphism actually depends on the choice of t, see
example 6.2 of (Kresch & Tschinkel, 2022b).)
The goal of this section is to define a similar homomorphism
ρt : Burn(K) → Burn(k)
for varieties with logarithmic volume forms.
 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 23

8.1. — Let X be an integral proper scheme over R, of relative dimension n, whose

special fiber ∆ is a divisor with strict normal crossings.
Let (∆α )α∈A be the family of irreducible components of the special fiber ∆; for
α ∈ A , let eα be the multiplicity of ∆α in ∆. For every nonempty subset A of A ,
let ∆A be the intersection of all divisors ∆α , for α ∈ A and eA be theSgreatest
common divisor of the eα , for α ∈ A; let also ∆◦A be the complement ∆A α6∈A ∆α .
The first specialization morphism of (Kontsevich & Tschinkel, 2019) is de-
fined by
X
(8.2) ρt ([XK ]) = (−1)|A|−1[∆A ]L|A|−1 ,
∅6=A⊆A

where L ∈ Burn(k) is the class of the affine line.

Although this map is not multiplicative, it proved sufficient for many applications
to rationality problems.
To ensure multiplicativity, a more delicate construction was necessary, valued in
the Burnside ring
Burnµb (k)
of varieties endowed with an action of the profinite group µ b, limit of finite groups of
roots of unity.
Fix a nonempty subset A of A . We identify the normal bundle of ∆A in X as a
direct sum of line bundles:
M
N∆A (X ) ≃ N∆α (X )|∆A .
α∈A

Let us consider its open subscheme N∆◦A (X ) obtained by restricting to ∆◦A and
taking out all “coordinate” hyperplanes. This furnishes a morphism
O
νA : N∆◦A (X ) → N∆α (X )⊗eα |∆◦A .
α∈A
P
Since the uniformizer t has divisor − α∈A eα ∆α on X , it trivializes the line bundle
on the target of νA . We set ∆′A = νA−1 (t). By construction, the projection ∆′A → ∆A
is a torsor with group µeA .
With this notation, the correct, multiplicative, specialization map of (Kontsevich & Tschinkel,
2019) is given by the formula
X
ρbt (X) = (−1)|A|−1 [∆′A ]L|A|−1
∅6=A⊆A

in Burnµb (k).

Remark 8.3. — The relation between the two specialization morphisms is as fol-
lows. Fix a nonempty subset A of A . The group Gm acts diagonally on N∆◦A (X )
(the factors of index α ∈
/ A don’t act), and this induces an action of the finite group
of roots of unity of order eA on ∆′A , hence an action of µ
b, so that ρbt (X) naturally
µb
lives in the equivariant Burnside ring Burn (k). Moreover, taking the µ b-invariants
24 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

of ∆′A , we get ∆◦A , so that the specialization map ρt is the composition of ρbt with
the map
Burnµb (k) → Burn(k)
obtained by taking µ b-invariants.
Taking invariants does not commute with taking products, in general, so that ρt
is not multiplicative.

8.4. — Let us explain how to define analogous specialization homomorphisms in

our context of Burnside groups with volume forms.
For simplicity, we only consider the case where K has transcendence degree 1
over k, in which case the idea can be explained geometrically as follows. We assume
that there exists an smooth integral curve C together with a k-point o ∈ C(k) such
that K = k(C) and R = OC,o. We fix a local parameter t ∈ R such that V(t) = o.
Let us consider a pair (X, ω) consisting of an integral proper K-variety X of di-
mension n and a logarithmic n-form ω on X.

8.5. — Consider a regular flat proper model X is of X over C, let ∆ = (Xo )red
be its reduced special fiber, and consider a divisor D with relative normal cross-
ings on X . We assume that the divisor D + ∆ has normal crossings. In this
situation, Deligne (1970, §3.3.2) says that a meromorphic relative differential m-
form on X /C is logarithmic with respect to D + ∆ if it is (locally) the image of
a logarithmic m-form ω̃ in Ωm X /k with poles D + ∆ under the natural morphism
m m
ΩX /k → ΩX /C .
Consider a logarithmic relative n-form ω on X /C. We consider an associated
volume form ω ′ on X , defined locally by
ω ′ = ω̃ ∧ dt/t,
where ω̃ is any local lift of ω. This form ω ′ is logarithmic and we can compute its
“residue along ∆” as in §4, only taking into account the strata of the polar divisor
of ω ′ which are contained in the special fiber ∆.
There exists a subset Ao of A and a subset Bo of B such that the polar divisor
of ω ′ is given by X X
∆α + Dβ .
α∈Ao β∈Bo
We thus set X
ρt (X , ω) = (−1)|A|+|B|−1ρ∆A ∩DB (X , ω).
∅6=A⊆Ao
B⊆Bo
This is an element of Burnn (Xo /k).

Proposition 8.6. — Let Y be an irreducible closed subscheme of X which is

transverse to D + ∆ and let g : X ′ → X be the blowing-up of X along Y . The
form g ∗ ω on X ′ is logarithmic and we have
g∗ ρt (X ′ , g ∗ω) = ρt (X , ω)
 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 25

in Burnn (Xo /k).

Proof. — With the notation of §4, the difference
ρ(X , ω̃) − ρt (X , ω)
is exactly the part of ρ(X , ω̃) which lies over the complement of the special fiber Xo
in X . We have seen in theorem 4.7 that
g∗ ρ(X ′ , ω̃ ′ ) = ρ(X , ω),
and a similar formula holds over X Xo . This implies the proposition.

8.7. — Starting from a smooth proper K-variety X and a logarithmic volume

form ω on X, we can define a model X /C, with D and ∆ as above, but the form ω
will not necessarily extend to a logarithmic relative form with respect to D + ∆, nor
does the volume form ω̃ on X . However, this can be achieved by multiplying ω by
a suitable power of the uniformizing element.
Let us write the polar divisor of ω̃ on X as
X X
divX (ω̃) = D + ∆ = dα ∆α + dβ Dβ .
α∈A β∈B

With this notation, the condition for ω̃ to be logarithmic on X is just that

dα > −1, dβ > −1.
In particular, while the conditions at the horizontal components follow from their
counterparts on the generic fiber, those for the vertical components are not auto-
matic. On the other hand, for any κ ∈ Z, the form tκ ω̃ is logarithmic if and only
if
κeα + dα > −1
for all α ∈ A , that is, if and only if κ > κ(ω), where κ(ω) is defined by
1 − dα
κ(ω) = inf .
α∈A eα
Since the rational number κ(ω) is defined in terms of logarithmic forms, it only
depends on the class of (X, ω) in Burnn (K), and not on the actual model which is
chosen to compute it.

8.8. — We assume for the moment that κ(ω) ∈ Z. This holds in particular if the
special fiber Xo is reduced. Let then Ao be the subset of A consisting of all α such
that
κ(ω)eα + dα = −1,
and let Bo be the subset of B consisting of all β such that dβ = −1. The polar
divisor of tκ ω̃ is equal to
X X
∆α + Dβ ,
α∈Ao β∈Bo
26 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

and we set
ρt (X , ω) = ρt (X , tκ(ω) ω)
in Burnn (Xo /k).
In the particular case where D is empty, the strata of the Clemens complex of
the special fiber that actually appear in the definition of this class are those defined
by Kontsevich & Soibelman (2006), more precisely, by the adjustment provided
by Mustaţă & Nicaise (2015).

8.9. — In the general case, the rational number κ(ω) is not an integer. Let us
consider the finite ramified extension Kd = K(t1/d ) of K, whose ramification index d
is a multiple of the denominator of κ(ω), but which induces an isomorphism on
the residue field. Geometrically, this furnishes a morphism π : Cd → C which is
ramified at the point o, together with a lift of o in Cd (k) (still denoted by o), and a
distinguished uniformizing element t1/d .
We consider the extension of (X, ω) to Kd and introduce a model (Xd , ωd ) as
above, over Cd . Now, the corresponding κ-parameter is integral, so that any choice
of a uniformizing element t1/d in Rd induces a class ρt1/d (Xd , ωd ) in Burn(k). In
fact, we can assume that the scheme Xd carries an action of the group scheme µd of
dth roots of unity induced by its action on Spec(Rd ), leaving the logarithmic form ωd
invariant. In other words, we obtain a class in the group Burnµb (k).
Combinining these classes, we obtain the desired group homomorphism
ρbt : Burn(K) → Burnµb (k).
In fact, as explained in (Nicaise, 2013, §2.3), especially proposition 2.3.2, one
can compute the normalisation of X ⊗ Rd in terms of the given model X . This
gives an explicit decomposition of ρbt (X, ω) as a sum
X
(−1)|A|−1 [D′A , νA∗ ωA ] · T|A|−1,
∅6=A⊆Ao

where νA : D′A → DA is the µdA -torsor introduced in §8.1 for the definition of the
classical specialization map.

Remark 8.10. — In the case of specialization of rationality, it has proved fruitful

to consider models with singularities on the special fiber, mild enough so that the
special fiber computes the specialization of the birational type of the generic fiber.
This is in particular the case for rational double points.
A parallel study can be developped in the context of varieties with logarithmic
forms.
Following (Kontsevich & Tschinkel, 2019) and keeping track of the various
logarithmic volume forms on the strata, we have:

Theorem 8.11. — The morphism ρbt is a ring homomorphism.

 BURNSIDE RINGS AND VOLUME FORMS WITH LOGARITHMIC POLES 27

References
D. Abramovich, K. Karu, K. Matsuki & J. Włodarczyk (2002), “Torifica-
tion and factorization of birational maps”. Journal of the American Mathematical
Society, 15 (3), pp. 531–572 (electronic).
S. Boucksom & M. Jonsson (2017), “Tropical and non-Archimedean limits of
degenerating families of volume forms”. Journal de l’École polytechnique - Math-
ématiques, 4, pp. 87–139.
A. Chambert-Loir & Yu. Tschinkel (2010), “Igusa integrals and volume asymp-
totics in analytic and adelic geometry”. Confluentes Mathematici, 2, pp. 351–429.
P. Deligne (1970), Equations différentielles à points singuliers réguliers, Lect.
Notes Math. 163, Springer, Cham.
S. Gorchinskiy & A. Rosly (2015), “A Polar Complex for Locally Free Sheaves”.
International Mathematics Research Notices, 2015 (10), pp. 2784–2829.
H. Hironaka (1964), “Resolution of singularities of an algebraic variety over a
field of characteristic zero. I, II”. Annals of Mathematics. Second Series, 79, pp.
109–203, 205–326.
M. Jonsson & J. Nicaise (2020), “Convergence of p-adic pluricanonical measures
to Lebesgue measures on skeleta in Berkovich spaces”. Journal de l’École poly-
technique — Mathématiques, 7, pp. 287–336.
B. Khesin & A. Rosly (2003), “Polar Homology”. Canadian Journal of Mathe-
matics, 55 (5), pp. 1100–1120.
B. Khesin, A. Rosly & R. Thomas (2004), “A polar de Rham theorem”. Topology,
43 (5), pp. 1231–1246.
M. Kontsevich & Y. Soibelman (2006), “Affine structures and non-Archimedean
analytic spaces”. The Unity of Mathematics, Progr. Math. 244, pp. 321–385,
Birkhäuser Boston, Boston, MA.
M. Kontsevich & Y. Tschinkel (2019), “Specialization of birational types”.
Inventiones mathematicae, 217 (2), pp. 415–432.
A. Kresch & Y. Tschinkel (2022a), “Burnside groups and orbifold invariants of
birational maps”. arXiv:2208.05835 .
A. Kresch & Y. Tschinkel (2022b), “Equivariant birational types and Burnside
volume”. Annali Scuola Normale Superiore - Classe Di Scienze, 23 (2), pp. 1013–
1052.
H.-Y. Lin & E. Shinder (2022), “Motivic invariants of birational maps”.
arXiv:2207.07389 .
H.-Y. Lin, E. Shinder & S. Zimmermann (2020), “Factorization centers in di-
mension two and the Grothendieck ring of varieties”. arXiv:2012.04806 .
M. Mustaţă & J. Nicaise (2015), “Weight functions on non-Archimedean analytic
spaces and the Kontsevich–Soibelman skeleton”. Algebraic Geometry, 2 (3), pp.
365–404.
J. Nicaise (2013), “Geometric criteria for tame ramification”. Mathematische
Zeitschrift, 273 (3-4), pp. 839–868.
28 ANTOINE CHAMBERT-LOIR, MAXIM KONTSEVICH & YURI TSCHINKEL

J. Nicaise & E. Shinder (2019), “The motivic nearby fiber and degeneration of
stable rationality”. Inventiones mathematicae, 217 (2), pp. 377–413.
J. Nicaise & C. Xu (2016), “The essential skeleton of a degeneration of algebraic
varieties”. American Journal of Mathematics, 138 (6), pp. 1645–1667.
M. Rost (1996), “Chow Groups with Coefficients”. Documenta Mathematica,
1 (16), pp. 319–393.