This paper provides a counting formula for twisted Fourier-Mukai partners of projective K3 surfaces and characterizes all such partners for surfaces with Picard number 1.
Contribution
It introduces a new counting formula for twisted Fourier-Mukai partners and classifies these partners for K3 surfaces with Picard number 1.
Findings
01
Derived a counting formula for twisted Fourier-Mukai partners.
02
Classified all twisted Fourier-Mukai partners for K3 surfaces with Picard number 1.
Abstract
We give a counting formula for the twisted Fourier-Mukai partners of a projective K3 surface. As an application, we describe all twisted Fourier-Mukai partners of a projective K3 surface of Picard number 1.
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAlgebraic Geometry and Number Theory · Geometry and complex manifolds · Commutative Algebra and Its Applications
Full text
Twisted Fourier-Mukai number of a K3 surface
Shouhei Ma
Graduate School of Mathematical Sciences, University of Tokyo
We give a counting formula for
the twisted Fourier-Mukai partners of
a projective K3 surface.
As an application,
we describe
all twisted Fourier-Mukai partners
of a projective K3 surface
of Picard number 1.
1 Introduction
Fourier-Mukai (FM) partners of a projective K3 surface S
have been studied since Mukai’s work [8].
The basic results due to
Mukai and Orlov ([10]) are as follows :
(1)
If a 2-dimensional moduli space M of stable sheaves on S
is non-empty and compact,
then M is a K3 surface.
When M is fine,
a universal sheaf induces an equivalence
Db(M)≃Db(S).
(2)
Every FM partner of S is isomorphic to a certain 2-dimensional
fine moduli space of stable sheaves on S.
(3)
A projective K3 surface S′ is a FM partner of S
if and only if their Mukai lattices are Hodge isometric.
Using Mukai-Orlov’s theorem (especially (3)),
Hosono-Lian-Oguiso-Yau ([5])
derived a counting formula for the FM partners of S.
On the other hand,
it has been recognized that
studying twisted FM partners
as well as usual (i.e, untwisted) FM partners of S
is important
when analyzing the derived category Db(S) of S.
Recall that
a twisted FM partner of S is
a twisted K3 surface (S′,α′) such that
there is an equivalence
Db(S′,α′)≃Db(S).
Căldăraru ([1], [2]),
Huybrechts-Stellari ([6], [7]),
and Yoshioka ([13])
generalized Mukai and Orlov’s results to twisted situation.
When S is untwisted,
their results can be stated as follows :
(1)
When a 2-dimensional coarse moduli space M of stable sheaves on S
is a K3 surface,
a twisted universal sheaf induces an equivalence
Db(M,α)≃Db(S),
where α is the obstruction to the existence of a universal sheaf.
(2)
Every twisted FM partner of S is isomorphic to a certain 2-dimensional
coarse moduli space of stable sheaves on S endowed with a natural twisting.
(3)
A twisted K3 surface (S′,α′) is a twisted FM partner of S
if and only if the twisted Mukai lattice of (S′,α′)
and the Mukai lattice of S are Hodge isometric.
Let
FM\/d(S)
be the set of
isomorphism classes of twisted FM partners (S′,α′) of S
with
ord\/(α′)=d.
The number
#FM\/d(S)
is an invariant of the category Db(S)
representing the number of certain geometric origins of Db(S).
The aim of this paper
is to present an explicit formula for
#FM\/d(S).
It enables us to calculate
#FM\/d(S)
from
lattice-theoretic informations about the Neron-Severi lattice NS(S)
and
the knowledge of the group OHodge(T(S))
of the Hodge isometries of the transcendental lattice T(S).
Let
Id(DNS(S))
be the set of
order\/=d isotropic elements of the discriminant form DNS(S).
From an element x∈Id(DNS(S))
we can construct overlattices
Mx, Tx of NS(S), T(S) respectively,
and a homomorphism
αx:Tx↠Z/dZ.
After defining certain subsets
G1(M), G2(M) of
the genus G(M) of a lattice M
such that
G(M)=G1(M)⊔G2(M),
our formula is stated as follows.
For a projective K3 surface S
the following formula holds.
[TABLE]
Here
x runs over the set OHodge(T(S))\Id(DNS(S))
and the lattices M, M′ run over the sets
G1(Mx), G2(Mx) respectively.
The natural number
ε(d) is defined by
ε(d)=1 if d=1,2, and
ε(d)=2 if d≥3.
When d=1,
Theorem 1.1 is Hosono-Lian-Oguiso-Yau’s formula.
Even though Theorem 1.1 seems complicated in appearance,
it is rather adequate for calculations for the following reasons :
•
The genus G(M),
and the homomorphism O(M)→O(DM)
have been studied in lattice theory,
and much is known (cf. [9]).
•
If we can identify the discriminant form DNS(S),
it is easy to calculate Id(DNS(S)).
•
The group OHodge(T(S)) and its subgroup OHodge(Tx,αx) are cyclic,
and the Euler function of the order of OHodge(T(S))
must divides rk\/(T(S)) (cf. [5]).
For example,
OHodge(T(S))={±id\/}
whenever rk\/(T(S)) is odd.
In fact,
we shall derive more simple formulas for
#FM\/d(S)
for several classes of K3 surfaces :
Jacobian K3 surfaces,
in particular K3 surfaces with rk\/(NS(S))≥13,
K3 surfaces with 2-elementary NS(S),
and
K3 surfaces with rk\/(NS(S))=1.
As a by-product of the proof of Theorem 1.1,
we obtain an upper bound for the twisted FM number of
a twistedK3 surface (Proposition 4.3).
The sharpness of the estimate
is related with a Căldăraru’s problem stated in [1]
(Remark 3.7).
Note that
if a twisted K3 surface (S′,α′)
has an untwisted FM partner,
e.g, if rk\/(NS(S′))≥12 ([6]),
then
Theorem 1.1
gives a counting formula for the twisted FM number of (S′,α′).
As an application of Theorem 1.1,
we shall describe all twisted FM partners of a K3 surface of Picard number 1
as follows.
Let
S be a projective K3 surface with
NS(S)=ZH, (H,H)=2n.
We have
FM\/d(S)=ϕ
if and only if
d2∣n.
For a natural number d with d2∣n,
let
{vσ,k∣(σ,k)∈Σ×(Z/dZ)×}
be the primitive isotropic
vectors in NS(S)
defined by (\refeqn:Mukaivector).
Let
Mσ,k be
the moduli space of Gieseker stable sheaves on S with Mukai vector vσ,k,
endowed with
the obstruction
ασ,k∈Br\/(Mσ,k)
to the existence of a universal sheaf.
(1)*
If d2<n or d≤2, then*
[TABLE]
(2)*
Assume that
d2=n and d≥3.
Choose a set
{j}⊂(Z/dZ)×
of representatives of
(Z/dZ)×/{±id\/}.
Then*
[TABLE]
Theorem 1.2 is a twisted generalization of a result of [4].
This paper is organized as follows.
In Sect.2.1,
we prepare some lattice theory
following [9].
In Sect.2.2,
we recall from [1] and [6]
several facts about twisted K3 surfaces.
In Sect.3.1,
we describe a quotient set of FM\/d(S)
in terms of lattice theory.
In Sect.3.2 and 3.3,
we describe the fibers of the quotient map
in terms of lattice theory.
In Sect.4,
we derive Theorem 1.1 and its corollaries.
In Sect.5,
we prove Theorem 1.2.
Notation**.**
By an even lattice,
we mean
a free Z-module L
equipped with a non-degenerate symmetric bilinear form
(,):L×L→Z
satisfying
(x,x)∈2Z
for all
x∈L.
The group of isometries of L is denoted by O(L).
For two lattices L and M,
Emb\/(L,M) is the set of the primitive embeddings of L into M.
An element l∈L is said to be
isotropic
if (l,l)=0.
The hyperbolic planeU
is the even indefinite unimodular lattice
Ze+Zf,(e,e)=(f,f)=0,(e,f)=1.
By a K3surface,
we mean a projective K3 surface over C.
The Neron-Severi (resp. transcendental) lattice of a K3 surface S
is denoted by NS(S) (resp. T(S)).
Set
NS(S):=H0(S,Z)⊕NS(S)⊕H4(S,Z),
ΛK3:=U3⊕E82,
and
ΛK3:=U⊕ΛK3=U4⊕E82.
Acknowledgement.
The author would like to
thank Professor Shinobu Hosono
for conversations that inspired this work.
His thanks also go to
Professors Keiji Oguiso and Ken-Ichi Yoshikawa
for valuable comments.
The idea of constructing moduli spaces in Section 5
is due to Professor Oguiso.
2 Preliminaries
2.1 Even lattices
For an even lattice L,
we can associate
a finite Abelian group DL:=L∨/L
equipped with a quadratic form
qL:DL→Q/2Z,qL(x+L)=(x,x)+2Z
for x∈L∨.
We call (DL,qL)
the discriminant form of L.
There is a natural homomorphism
rL:O(L)→O(DL,qL),
whose kernel is denoted by O(L)0.
We often write just
(DL,q) or DL (resp. r)
instead of (DL,qL) (resp. rL).
Set
Let M be a primitive sublattice of an even unimodular lattice L
with the orthogonal complement M⊥.
Then
(1)* There is an isometry λ:(DM,qM)→≃(DM⊥,−qM⊥).*
(2)* For two isometries
γM∈O(M) and γM⊥∈O(M⊥),
there is an isometry γL∈O(L)
such that γL∣M⊕M⊥=γM⊕γM⊥
if and only if
λ∘rM(γM)=rM⊥(γM⊥)∘λ.*
Two even lattices L and L′ are said to be isogenus
if
L⊗Zp≃L′⊗Zp
for every prime number p
and sign\/(L)=sign\/(L′).
By [9],
these are equivalent to the conditions that
(DL,q)≃(DL′,q) and
sign\/(L)=sign\/(L′).
The set of the isometry classes of the lattices isogenus to L is denoted by G(L).
Let l(DL) be the minimal number of the generators of DL.
If L is indefinite and rk\/(L)≥l(DL)+2,
then G(L)={L}
and the homomorphism rL is surjective.
Let L be a lattice of
sign\/(L)=(b+,b−), b+>0.
We denote by ΩL
the set of oriented positive-definite b+-planes in L⊗R,
which is an open subset of the oriented Grassmannian
and has two connected components.
A choice of a component of ΩL,
which is equivalent to
a choice of an orientation for a positive-definite b+-plane in L⊗R,
is sometimes called an
orientation of L.
For example,
for a K3 surface S
the lattice NS(S) equipped with the Mukai pairing
is of sign\/(NS(S))=(2,ρ(S)).
We define
the orientation of NS(S)
so that
R(1,0,−1)⊕R(0,H,0) is of positive orientation,
where H∈NS(S) is an ample class.
For a subgroup
Γ⊂O(L),
let Γ+ be the subgroup of Γ
consisting of orientation-preserving isometries in Γ.
The group Γ+
is of index at most 2 in Γ.
We define the subsets
G1(L),G2(L)⊂G(L)
by
[TABLE]
We have the obvious decomposition
G(L)=G1(L)⊔G2(L).
When b+ is odd,
we have
O(L)0+=O(L)0
if and only if
−id\/DL is contained in rL(O(L)+).
2.2 Twisted K3 surfaces
Let (S,α) be a twisted K3 surface,
where α is an element of the Brauer group Br\/(S).
Via the exponential sequence
[TABLE]
there is a canonical isomorphism
Br\/(S)≃Hom\/(T(S),Q/Z).
We identify
a twisting α∈Br\/(S)
with
a surjective homomorphism α:T(S)→Z/ord\/(α)Z,
whose kernel is denoted by T(S,α).
So
the identity of Br\/(S) is denoted by [math]
and the inverse of α∈Br\/(S)
is denoted by −α.
Definition 2.3**.**
Set
[TABLE]
where
OHodge(T(S)) is the group of the Hodge isometries of T(S),
and
λ:O(DT(S))→≃O(DNS(S))
is the isomorphism induced from the isometry
λH(S,Z):(DT(S),q)≃(DNS(S),−q).
From the inclusions
T(S,α)⊂T(S)⊂T(S,α)∨,
we have the isomorphism
[TABLE]
where 1ˉ∈Z/ord\/(α)Z.
In generic case,
OHodge(T(S),α)={id\/}
if ord\/(α)≥3
and
OHodge(T(S),α)={±id\/}
if ord\/(α)≤2.
We need the following corollary of Huybrechts-Stellari’s theorem.
Let S be a K3 surface
and let (S′,α′) be a twisted K3 surface.
Then
there exists an equivalence Db(S′,α′)≃Db(S)
if and only if
there exists a Hodge isometry T(S)≃T(S′,α′).
Proof.
By the theorem of Canonaco-Stellari ([3]),
every equivalence between derived categories of twisted K3 surfaces
is of Fourier-Mukai type.
Thus only ”if” part requires an explanation.
Assume that
there is a Hodge isometry T(S)≃T(S′,α′).
Let B′∈H2(S′,Q) be a B-field lift of α′.
Since there is a Hodge isometry
T(S′,α′)≃T(S′,B′),
we obtain a Hodge isometry g:T(S)≃T(S′,B′).
By Proposition 2.1,
NS(S)
and
NS(S′,B′)
are isogenus.
Because
NS(S) admits an embedding of the hyperbolic plane U,
it follows from Proposition 2.2 that
g
extends to a Hodge isometry
Φ:H(S,Z)≃H(S′,B′,Z).
Composing Φ
with the isometry
−id\/H0(S)+H4(S)⊕id\/H2(S)
if necessary,
we may assume that
Φ is orientation-preserving.
Now we have
Db(S′,α′)≃Db(S)
by Huybrechts-Stellari’s theorem.
∎
When α′=0,
Proposition 2.4 is
Mukai-Orlov’s theorem ([8], [10]).
Definition 2.5**.**
For a twisted K3 surface (S,α), set
[TABLE]
A member (S′,α′)∈FM\/d(S,α)
is called a
twisted Fourier-Mukai (FM) partner of (S,α).
It is easily seen that
FM\/d(S,α) is empty
unless
d2 divides det\/(T(S,α)).
Definition 2.6**.**
Set
[TABLE]
where
two twisted FM partners
(S1,α1),
(S2,α2)∈FM\/d(S,α)
are equivalent
if there exists a Hodge isometry
g:T(S1)≃T(S2)
with
g∗α2=α1.
Denote by
[TABLE]
the quotient map.
We remark that
for a twisted K3 surface (S,α)
with a B-field lift B∈H2(S,Q),
the period of S is determined by the natural Hodge isometry
[TABLE]
In Section 5,
we shall calculate twistings in terms of discriminant group.
Lemma 2.7**.**
Let (S,α) be a twisted K3 surface with a B-field lift B∈H2(S,Q).
We denote
d:=ord\/(α)
and
λ:=λH(S,B,Z):(DNS(S,B),q)≃(DT(S,B),−q).
Then
(0,0,d−1)∈NS(S,B)∨
and
we have
the Hodge isometry
[TABLE]
The twisting
α
is given by
[TABLE]
Proof.
Since
α:T(S)→Z/dZ
is surjective,
we can take a transcendental cycle
l∈T(S)
such that
(l,B)=d1+k with k∈Z.
Setting
l′:=e\/B(l)=l+(0,0,d1+k)∈T(S,B)∨,
we have
e\/B:T(S)≃⟨l′,T(S,B)⟩.
Since
l′−(0,0,d1)∈H(S,B,Z),
it follows that
((0,0,d−1),m)∈Z
for all m∈NS(S,B).
By the relation
λ((0,0,d−1))=l′∈DT(S,B),
we have (1).
Then (2)
follows from the relation
α(l)=1ˉ∈Z/dZ.
∎
More precisely,
the divisibility of
(0,0,1)∈NS(S,B)
is equal to d,
but
we do not need this fact in the following.
3 Lattice-theoretic descriptions
3.1 Isotropic elements of the discriminant form
Let
(S,α)
be a twisted K3 surface.
We write T:=T(S,α),
which is an even lattice of sign\/(T)=(2,20−ρ(S))
equipped with a period.
For a twisted FM partner
(S1,α1)∈FM\/d(S,α)
there exists a Hodge isometry
g1:T(S1,α1)≃T
by Huybrechts-Stellari’s theorem ([6]).
Then
g1 and α1
induce an isomorphism
g1(T(S1))/T→≃Z/dZ.
The subgroup
g1(T(S1))/T⊂DT
is an isotropic cyclic group of order=d.
Definition 3.1**.**
Define the map
[TABLE]
by
μ([(S1,α1)]):=[g1(α1−1(1ˉ))].
Lemma 3.2**.**
The map μ is well-defined and is injective.
Proof.
The ambiguity of the choice of a Hodge isometry
g1:T(S1,α1)≃T
comes from the action of OHodge(T)
so that
the class
[g1(α1−1(1ˉ))]∈OHodge(T)\Id(DT)
is well-defined from the partner (S1,α1).
Since
the map μ is defined
in terms of T(S1) and α1,
it is clear that
μ is well-defined as a map from FMd(S,α).
We prove the injectivity of μ.
For two partners
(Si,αi)∈FM\/d(S,α),
i=1,2,
and Hodge isometries
gi:T(Si,αi)≃T,
write
xi:=gi(αi−1(1ˉ))∈Id(DT).
Assume
the existence of a Hodge isometry
φ∈OHodge(T) such that
x2=r(φ)(x1).
Since
r(φ)(⟨x1⟩)=⟨x2⟩⊂DT,
φ extends to the Hodge isometry
φ:T(S1)≃T(S2).
By the equality
r(φ)(g1(α1−1(1ˉ)))=g2(α2−1(1ˉ))
we have
φ∗α2=α1.
Hence
[(S1,α1)]=[(S2,α2)]∈FMd(S,α).
∎
Proposition 3.3**.**
If
the twisted K3 surface (S,α)
has an untwisted FM partner,
then the map μ is bijective.
Proof.
It suffices to show the surjectivity.
Let
S′ be an untwisted FM partner of (S,α).
Since there exists a Hodge isometry
T(S′)≃T,
we may assume from the first that
(S,α) itself is untwisted.
Take an isotropic element
x∈Id(DT)=Id(DT(S)).
Via the isometry
λ:(DT(S),q)≃(DNS(S),−q),
we obtain an isotropic element
λ(x)∈Id(DNS(S)).
Set
[TABLE]
which are even overlattices of
NS(S), T(S) respectively.
Via the isometry
[TABLE]
we obtain an embedding
Mx⊕Tx↪ΛK3
with
both Mx and Tx embedded primitively.
Since
NS(S)⊂Mx,
the lattice
Mx
admits an embedding
φ of the hyperbolic plane U.
Then the lattice
Λφ:=φ(U)⊥∩ΛK3
is isometric to the K3 lattice ΛK3
and has the period induced from Tx.
Now
the surjectivity of the period map ([12], [14])
assures the existences of
a K3 surface Sφ
and a Hodge isometry
Φ:H2(Sφ,Z)→≃Λφ.
Pulling back the homomorphism
[TABLE]
by
Φ∣T(Sφ),
we obtain a twisted K3 surface
(Sφ,αφ).
Since there exists a Hodge isometry
T(Sφ,αφ)≃Ker\/(αx)=T(S),
it follows from Proposition 2.4 that
(Sφ,αφ)∈FM\/d(S).
By the construction,
we have
μ([(Sφ,αφ)])=[x].
∎
In general,
there is a condition on the image of μ.
Let
B∈H2(S,Q) be a B-field lift of α∈Br\/(S).
There is a natural isometry
λ:(DT,q)≃(DNS(S,B),−q).
We define
[TABLE]
One can verify that
the set Jd(DT) is
independent of the choice of a lift B.
Proposition 3.4**.**
We have the inclusion
[TABLE]
Proof.
For a twisted FM partner (S1,α1)∈FM\/d(S,α),
write [x]:=μ([(S1,α1)]).
By Proposition 2.2,
it suffices to show that
NS(S1) and
M:=⟨λ(x),NS(S,B)⟩
are isogenus.
We have the isometry
[TABLE]
∎
We make the following identification
tacitly in Section 4.
For
x∈Id(DT)
such that
[x]=μ([(S1,α1)]),
set
Tx:=⟨x,T⟩.
For a Hodge isometry
g1:T(S1,α1)≃T
with
g1(α1−1(1ˉ))=x,
by Proposition 2.2
there exists an isometry
[TABLE]
Here
the Hodge isometry
g1ˉ:T(S1)≃Tx
is induced from g1,
and
the isometry
λˉ:(DTx,q)≃(D⟨λ(x),NS(S,B)⟩,−q)
is induced from
λ.
3.2 Embeddings of the hyperbolic plane
Let
(S1,α1)∈FM\/d(S,α)
be a twisted FM partner of (S,α).
Recall from Section 2.2 that
we have defined the finite-index subgroup
Γ(S1,α1)⊂O(NS(S1)).
The subgroup
Γ(S1,α1)+⊂Γ(S1,α1)
consists of
orientation-preserving isometries in
Γ(S1,α1) (see Section 2.1).
We define the map
[TABLE]
as follows.
For a twisted FM partner
(S_{2},\alpha_{2})\in\pi^{-1}\Bigl{(}\pi((S_{1},\alpha_{1}))\Bigr{)}
there exists a Hodge isometry
g:T(S2)→≃T(S1)
with
g∗α1=α2.
By Proposition 2.2,
g can be extended to a Hodge isometry
Φ:H(S2,Z)→≃H(S1,Z)
such that
the isometry
Φ∣NS(S2):NS(S2)→≃NS(S1)
is orientation-preserving.
Then,
by considering
Φ:H0(S2,Z)+H4(S2,Z)↪NS(S1),
we obtain an embedding
φ:U↪NS(S1).
Here
we identify
H0(S2,Z)+H4(S2,Z)
with
U by
identifying (1,0,0) with e,
and (0,0,−1) with f.
Lemma 3.5**.**
The assignment
(S2,α2)↦[φ]
defines an injective map
[TABLE]
Proof.
Firstly we prove that ν is well-defined.
Let us given two Hodge isometries
g,g′:T(S2)→≃T(S1)
with
g∗α1=(g′)∗α1=α2.
Let
γ,γ′:NS(S2)→≃NS(S1)
be orientation-preserving isometries
such that
γ⊕g and γ′⊕g′
extend to Hodge isometries
H(S2,Z)≃H(S1,Z).
Then
the isometry
γ′∘γ−1∈O(NS(S1))
preserves the orientation
and satisfies
λ∘r(γ′∘γ−1)=r(g′∘g−1)∘λ,
so that we have
γ′∈Γ(S1,α1)+⋅γ.
Hence
the map ν is well-defined.
We prove the injectivity of ν.
For two partners
(S_{i},\alpha_{i})\in\pi^{-1}\Bigl{(}\pi((S_{1},\alpha_{1}))\Bigr{)},
i=2,3,
set
[φi]:=ν((Si,αi))
and assume the existence of an isometry
γ∈Γ(S1,α1)+
such that
φ2=γ∘φ3.
We can extend γ to a Hodge isometry
Φ:H(S1,Z)→≃H(S1,Z)
with
(Φ∣T(S1))∗α1=α1.
By restricting Φ to
φ2(U)⊥∩H(S1,Z),
we obtain a Hodge isometry
Φ:H2(S2,Z)→≃H2(S3,Z)
with
(Φ∣T(S2))∗α3=α2.
Since
γ preserves the orientation,
Φ maps
the positive cone NS(S2)+ to
the positive cone NS(S3)+.
Composing Φ with an element of the Weyl group W(S3),
we may assume that Φ is effective.
Then
it follows from the Torelli theorem ([11], [14]) that
(S3,α3)≃(S2,α2).
∎
Proposition 3.6**.**
If (S,α) has an untwisted FM partner,
then the map ν
for every twisted FM partner
(S1,α1)∈FM\/d(S,α)
is bijective.
Proof.
We may assume that
(S,α) itself is untwisted.
To prove the surjectivity,
we shall construct a twisted K3 surface
(Sφ,αφ)
from an arbitrary embedding
φ:U↪NS(S1).
The lattice
Λφ:=φ(U)⊥∩H(S1,Z)
is isometric to the K3 lattice ΛK3
and possesses the period induced from T(S1).
On the other hand,
the lattice
Mφ:=φ(U)⊥∩NS(S1)
has the orientation
induced from φ and the orientation of NS(S1).
That is,
we can choose a connected component Mφ+
of the open set
{v∈Mφ⊗R,(v,v)>0}
so that
for each vector
v∈Mφ+
the oriented positive-definite two-plane
Rφ(e+f)⊕Rv⊂NS(S1)⊗R
is of positive orientation.
By the surjectivity of the period map,
there exist a K3 surface Sφ
and a Hodge isometry
Φ:H2(Sφ,Z)→≃Λφ
such that
Φ(NS(Sφ)+)=Mφ+.
Pulling back α1 by Φ∣T(Sφ),
we obtain a twisted K3 surface
(Sφ,αφ).
Since there is a Hodge isometry
[TABLE]
we have an equivalence
Db(Sφ,αφ)≃Db(S)
by Proposition 2.4.
Hence
we have
(S_{\varphi},\alpha_{\varphi})\in\pi^{-1}\Bigl{(}\pi((S_{1},\alpha_{1}))\Bigr{)}.
Since
the direct sum of
the Hodge isometry
Φ∣T(Sφ):T(Sφ)→≃T(S1)
satisfying
(Φ∣T(Sφ))∗α1=αφ
and
the orientation-preserving isometry
[TABLE]
can be extended to the Hodge isometry
φ⊕Φ,
we have
ν((Sφ,αφ))=[φ].
∎
Remark 3.7**.**
In his thesis [1],
Căldăraru proposed the following problem:
Let (S1,α1) be a twisted K3 surface.
For each untwisted FM partner
S2∈FM\/(S1)
and each Hodge isometry
g:T(S2)→≃T(S1),
we obtain a twisted K3 surface
(S2,g∗α1).
Is it true that
(S2,g∗α1)∈FM\/d(S1,α1) ?
From the construction of the twisted K3 surface
(Sφ,αφ)
in the proof of Proposition 3.6,
it is immediately verified that
the map ν in Lemma 3.5 is bijective
if and only if
the answer to Căldăraru’s question is positive
for the twisted K3 surface (S1,α1)
and each S2, g.
3.3 Union of the two orbits
Given an embedding
φ:U↪NS(S1),
we have the decomposition
[TABLE]
From this we obtain
[TABLE]
Lemma 3.9**.**
Let
(Sφ,αφ)
be the twisted K3 surface
constructed in the proof of Proposition 3.6.
Then
−φ∈Γ(S1,α1)+⋅φ
if and only if
(Sφ,αφ)≃(Sφ,−αφ).
Proof.
It suffices to show that
(S−φ,α−φ)≃(Sφ,−αφ).
Since
M−φ+=−Mφ+,
we have a Hodge isometry
Φ:H2(S−φ,Z)→≃H2(Sφ,Z)
with
(Φ∣T(S−φ))∗αφ=α−φ
and
Φ(NS(S−φ)+)=−NS(Sφ)+.
Then
−Φ:H2(S−φ,Z)→≃H2(Sφ,Z)
satisfies
(−Φ∣T(S−φ))∗(−αφ)=α−φ
and
−Φ(NS(S−φ)+)=NS(Sφ)+.
∎
When
ord\/(α1)≤2,
we have
αφ=−αφ
so that
Γ(S1,α1)⋅φ=Γ(S1,α1)+⋅φ
for every embedding φ.
Hence we obtain the identification
[TABLE]
Proposition 3.10**.**
Assume that
ord\/(α1)≥3.
For an embedding
φ:U↪NS(S1)
the following conditions are equivalent.
(i\/)*
(Sφ,αφ)≃(Sφ,−αφ).*
(ii\/)*
Sφ admits an anti-symplectic automorphism.*
(i) ⇒ (ii) :
Assume the existence of an automorphism
f:(Sφ,αφ)≃(Sφ,−αφ).
Since
ord\/(αφ)≥3,
ord\/(f∗∣T(Sφ)) must be even,
say 2n.
As OHodge(T(Sφ)) is a cyclic group
(cf. [5]),
then fn is an anti-symplectic automorphism of Sφ.
(ii) ⇒ (iii) :
For an anti-symplectic automorphism f,
we have
−f∗∣NS(Sφ)∈O(NS(Sφ))0−O(NS(Sφ))0+.
(iii) ⇒ (ii) :
Assume the existence of an isometry
γ∈O(NS(Sφ))0−O(NS(Sφ))0+.
Then we can extend
the isometry
−γ⊕−id\/T(Sφ)
to the anti-symplectic Hodge isometry
Φ:H2(Sφ,Z)≃H2(Sφ,Z)
which preserves the positive cone.
Composing Φ with an element of the Weyl group W(Sφ),
we may assume that Φ is effective.
∎
4 The counting formula
Proposition 4.1**.**
For a twisted K3 surface (S1,α1)
there exists a bijection
[TABLE]
Proof.
We can decompose the set
Emb\/(U,NS(S1)) as
[TABLE]
The isometry group
O(NS(S1))=O(M⊕U)
acts
transitively on each component
{φ∈Emb\/(U,NS(S1)),φ(U)⊥≃M}
with the stabilizer subgroup
O(M)⊂O(M⊕U).
From the surjectivity of
r:O(M⊕U)→O(DM⊕U)≃O(DM)
(Proposition 2.2)
and the inclusion
O(M⊕U)0⊂Γ(S1,α1),
we have
[TABLE]
∎
Let
(S,α)
be a twisted K3 surface
with a B-field lift B∈H2(S,Q).
For an isotropic element
x∈Id(DT(S,α))
we define the lattice Tx
by Tx=⟨x,T(S,α)⟩
and define the homomorphism
αx:Tx↠Z/dZ
by
[TABLE]
For a pair (x,M)
such that
x∈Id(DT(S,α))≃λId(DNS(S,B))
and
⟨λ(x),NS(S,B)⟩≃U⊕M,
we define the natural number
τ(x,M) by
[TABLE]
where the group OHodge(Tx,αx)
acts on O(DM) via
r:OHodge(Tx,αx)→O(DTx)
and
λˉ:O(DTx)≃O(DM⊕U)≃O(DM).
For a natural number d
we define the natural number
ε(d) by
[TABLE]
From
Section 3 and Proposition 4.1,
we deduce the formula for
#FM\/d(S).
Theorem 4.2**.**
For a K3 surface S
the following formula holds.
[TABLE]
Here
x runs over the set
OHodge(T(S))\Id(DT(S)),
and the lattices M, M′
run over the sets
G1(⟨λ(x),NS(S)⟩),
G2(⟨λ(x),NS(S)⟩)
respectively.
We also obtain the following inequality.
Proposition 4.3**.**
For a twisted K3 surface (S,α) the following inequality holds.
[TABLE]
Here
x
runs over the set
OHodge(T(S,α))\Jd(DT(S,α)),
where
Jd(DT(S,α)) is the set defined in (\refeqn:Jd(D)).
The lattices
M, M′
run over the sets
G1(Mφ),
G2(Mφ)
respectively,
where Mφ is a lattice satisfying
⟨λ(x),NS(S,B)⟩≃U⊕Mφ.
Corollary 4.4**.**
When NS(S,B) contains U⊕U,
then
[TABLE]
Proof.
By considering
U⊥∩H(S,B,Z),
one observes the existence of
an untwisted FM partner
S′∈FM\/1(S,α).
We have a Hodge isometry
T(S,α)≃T(S′).
Since
NS(S′) contains U⊕U,
then
NS(S′) contains U.
Now we apply Theorem 4.2 to S′.
For
x∈Id(DT(S′)),
the lattice
Mx:=⟨λ(x),NS(S′)⟩
admits an embedding of U
so that
G(Mx)={Mx} and
τ(x,Mx)=1
by Proposition 2.2.
Then
G(Mx)=G1(Mx)
due to the existence of the isometry
−id\/U⊕id\/U⊥.
∎
By [9],
NS(S,B) admits an embedding of U⊕U
if rk\/(NS(S))≥13.
Corollary 4.5**.**
When the lattice NS(S) is 2-elementary,
i.e, DNS(S)≃(Z/2Z)a,
then
#FM\/d(S)=0
unless
d=1,2.
If d=1,2,
we have
[TABLE]
Proof.
Because the lattice
Mx:=⟨λ(x),NS(S)⟩
is also 2-elementary,
G(Mx)={Mx} and
τ(x,Mx)=1
by [9].
∎
5 Twisted Fourier-Mukai partners of a K3 surface of Picard number 1
Let S be a K3 surface with NS(S)=ZH, (H,H)=2n.
In this section
we shall describe all twisted FM partners of S
(up to isomorphism)
as moduli spaces of stable sheaves on S with explicit Mukai vectors,
endowed with natural twistings.
First of all,
we calculate the twisted FM numbers by Theorem 4.2.
Proposition 5.1**.**
For a natural number d,
we have
#FM\/d(S)=0
unless
d2∣n.
If d2∣n, then
[TABLE]
Here
φ is the Euler function
and
τ(m) is the number of the prime factors of m.
We define
τ(1)=1.
Proof.
We have
DNS(S)=⟨2nH⟩≃Z/2nZ
and
OHodge(T(S))={±id\/}
(cf. [5]).
Since
[TABLE]
it follows that
#FM\/d(S)=0
unless
d2∣n.
Assume that d2∣n.
Then we have
Id(DNS(S))={kdH,k∈(Z/dZ)×}
so that
[TABLE]
For
d≥3 and k∈(Z/dZ)×,
we have
⟨ZH,kdH⟩=ZdH
and
[TABLE]
Because
OHodge(TkdH,αkdH)={id\/},
O(ZdH)={±id\/}
and
Let d≤2.
Since
OHodge(TdH,αdH)=O(ZdH)={±id\/}
and the isomorphisms (4) still hold,
we have
#FM\/d(S)=2τ(d−2n)−1=φ(d)⋅2τ(d−2n)−1
by Theorem 4.2.
∎
Now we shall construct the moduli spaces.
Since the primitive embeddings of the lattice ZH
into the K3 lattice ΛK3 are unique modulo O(ΛK3),
we fix an isometry
H2(S,Z)≃ΛK3=U3⊕E82
so that
H is written as H=nu+v,
where
{u,v} is a standard basis of the first U.
Then
the transcendental lattice T(S) is written as
T(S)=Z(nu−v)⊕U2⊕E82.
Set
G:=nu−v∈T(S).
For
a natural number d with d2∣n,
let
d−2n=∏i=1τ(d−2n)piei
be the prime decomposition.
When d2=n,
we put p1=e1=1.
Set
[TABLE]
We have #Σ=2τ(d−2n)−1.
For
σ={i1,⋯,ia}∈Σ,
set
rσ:=∏i∈σpiei
and
sσ:=rσ−1d−2n.
Then
rσ is coprime to sσ.
We have
rσ>sσ if d2<n.
Let
k∈(Z/dZ)×.
By the arithmetic progression,
we can find a natural number
k~∈N coprime to 2n with k~≡k∈Z/dZ.
For example,
we may take k~=1 if k=1ˉ.
Fix such k~ for each k∈(Z/dZ)×.
For
(σ,k)∈Σ×(Z/dZ)×,
we define
[TABLE]
Then
vσ,k is a primitive isotropic vector in NS(S)
with
div\/(vσ,k)=d.
Let
Mσ,k:=MH(vσ,k)
be the coarse moduli space of
Gieseker stable sheaves on S with Mukai vector vσ,k.
It follows from [8] that
Mσ,k is non-empty and is a K3 surface.
Let
ασ,k∈Br\/(Mσ,k)
be the obstruction to the existence of a universal sheaf ([2])
with a B-field lift Bσ,k∈H2(Mσ,k,Q).
By [6],
there exists an orientation-preserving Hodge isometry
[TABLE]
which
induces a Hodge isometry
[TABLE]
(We remark that the Hodge isometry (7) was proved first by [8].)
We have
NS(Mσ,k)=ZHσ,k
and
k~Hσ,k=[(rσ,0,−sσk~2)].
Since
the Hodge isometry (6)
is orientation-preserving,
Hσ,k
is the ample generator of NS(Mσ,k).
For
the transcendental lattice
we have
[TABLE]
Here
L means
the primitive hull of the lattice L
in H(S,Z).
Since
λH(S,Z)(dvσ,k)=dk~G∈DT(S),
it follows from
Lemma 2.7 and
the Hodge isometry (6)
that
the twisting ασ,k is given by
[TABLE]
Via
the Hodge isometry (7),
we obtain a Hodge isometry
[TABLE]
Set
Gσ,k:=gσ,k−1(dG),
which is a primitive vector in
T(Mσ,k).
From
(9),
for
(σ,k),(σ′,k′)∈Σ×(Z/dZ)×
we have
a Hodge isometry
[TABLE]
Therefore
we have an equivalence
Db(Mσ,k)≃Db(Mσ′,k′)
by Proposition 2.4.
Note that
the Hodge isometries
from T(Mσ,k)
to T(Mσ′,k′)
are just
{±gσ,kσ′,k′}.
Now
we calculate λσ,k
as follows.
We have
[TABLE]
Firstly
assume that rσsσ is odd.
With respect to orthogonal decompositions
[TABLE]
we write
[TABLE]
Each
xσ,ki,
yσ,ki
are
generators of Z/pieiZ.
Since
we have
[TABLE]
the isometry
λσ,k=λH2(Mσ,k,Z)
is
determined
by
[TABLE]
If
rσsσ is even,
put p1=2 and write
[TABLE]
Then
λσ,k is determined by
the same equations as (\refeqn:patching).
Proposition 5.2**.**
For the twisted K3 surfaces
(Mσ,k,ασ,k)
the followings hold.
(1)*
The underlying K3 surfaces
{Mσ,k,(σ,k)∈Σ×(Z/dZ)×}
are derived equivalent to each other.*
(2)*
We have
Mσ,k≃Mσ′,k′
if and only if
σ=σ′.*
(3)*
We have
(Mσ,k,ασ,k)≃(Mσ,1,k−1ασ,1).*
Proof.
We already proved the assertion (1)
using Mukai-Orlov’s theorem.
Assume the existence of an effective Hodge isometry
Φ:H2(Mσ,k,Z)≃H2(Mσ′,k′,Z)
for
(σ,k),(σ′,k′)∈Σ×(Z/dZ)×.
We must have
Φ(Hσ,k)=Hσ′,k′
and
Φ(Gσ,k)=±Gσ′,k′.
Since
the identity
λσ′,k′∘r(Φ∣NS(Mσ,k))=r(Φ∣T(Mσ,k))∘λσ,k
holds,
it follows from (\refeqn:patching)
that
σ=σ′or\/(σ′)c.
By the definition of Σ
we have σ=σ′.
Conversely
let σ=σ′.
If we define
γσ,kσ,1:NS(Mσ,k)≃NS(Mσ,1)
by
γσ,kσ,1(Hσ,k)=Hσ,1,
then
γσ,kσ,1⊕gσ,kσ,1
extends to an effective Hodge isometry
H2(Mσ,k,Z)≃H2(Mσ,1,Z).
Since
the twistings
ασ,k
are given by (8),
we have
(gσ,kσ,1)∗ασ,1=kασ,k,
or equivalently,
(gσ,kσ,1)∗(k−1ασ,1)=ασ,k.
∎
Finally,
we obtain
the following theorem.
Theorem 5.3**.**
Let S be a K3 surface
with
NS(S)=ZH, (H,H)=2n,
and let d be a natural number satisfying d2∣n.
(1)*
If d2<n or d≤2,
then*
[TABLE]
(2)*
Assume d2=n and d≥3.
Choose a set
{j}⊂(Z/dZ)×
of representatives of
(Z/dZ)×/{±id\/}.
then*
[TABLE]
Proof.
That
(Mσ,k,ασ,k)∈FM\/d(S)
is a direct consequence of Căldăraru’s theorem ([2]).
By Proposition 5.1,
it suffices to show that
the twisted K3 surfaces in the right hand sides are not isomorphic to each other.
(1)
By Proposition 5.2,
the right hand side is identified with the set
[TABLE]
We have
(Mσ,1,kασ,1)≃(Mσ′,1,k′ασ′,1)
if
σ=σ′.
If
d2<n,
then
Aut(Mσ,1)={id\/}
so that
(Mσ,1,kασ,1)≃(Mσ,1,k′ασ,1)
if
k=k′.
If d≤2,
there remains nothing to prove.
(2)
Let
d2=n and d≥3.
The right hand side is identified with the set
[TABLE]
Similarly as (1),
We have
(Mσ,1,kασ,1)≃(Mσ′,1,k′ασ′,1)
if
σ=σ′.
Now
the K3 surface Mσ,1 admits
an anti-symplectic involution ισ,1
and
Aut(Mσ,1)={id\/,ισ,1}.
Therefore
(Mσ,1,kασ,1)≃(Mσ,1,k′ασ,1)
if
k=±k′.
∎
4[4] Hosono, S.; Lian, B. H.; Oguiso, K.; Yau, S.-T. Fourier-Mukai partners of a K 3 𝐾 3 K 3 surface of Picard number one. Vector bundles and representation theory , 43–55, Contemp. Math., 322 , 2003.
5[5] Hosono, S.; Lian, B. H.; Oguiso, K.; Yau, S.-T. Fourier-Mukai number of a K 3 surface. Algebraic structures and moduli spaces, 177–192, CRM Proc. Lecture Notes, 38 , 2004
6[6] Huybrechts, D. ; Stellari, P. Equivalences of twisted K 3 𝐾 3 K 3 surfaces. Math. Ann. 332 (2005), no. 4, 901–936.
7[7] Huybrechts, D.; Stellari, P. Proof of Căldăraru’s conjecture. Appendix: ”Moduli spaces of twisted sheaves on a projective variety” by K. Yoshioka. Moduli spaces and arithmetic geometry, 31–42, Adv. Stud. Pure Math., 45 , 2006.
8[8] Mukai, S. On the moduli space of bundles on K 3 𝐾 3 K 3 surfaces. I. Vector bundles on algebraic varieties, 341–413, Tata Inst. Fund. Res. Stud. Math., 11 , 1987.