We study twisted derived equivalence between genus 1 curves over an arbitrarly perfect field.
Introduction
In this paper we study twisted Fourier–Mukai partners of genus 1 curves over a perfect field k.
We recall that a twisted variety over k is given by a pair (X,α), with X a smooth projective geometrically connected variety over k and α∈Br(X) an element of its Brauer group. Given a twisted variety, there is a Gm-gerbe π:X→X with class α, unique up to isomorphism. We write Db(X,α)=Db(X)(1) for the derived category of 1-twisted sheaves. We say that two varieties X and Y are twisted derived equivalent if there are two Brauer classes α∈Br(X) and β∈Br(Y) and an exact equivalence
Φ:Db(X,α)→Db(Y,β).
We say that two varieties X and Y are derived isogenous if there are a finite sequence of varieties
X≃X0,X1,…,Xn≃Y
and Brauer classes βi∈Br(Xi) and αi+1∈Br(Xi+1) such that Db(Xi,βi)≃Db(Xi+1,αi+1) for 0≤i<n. Investigating the hidden relationships between twisted derived equivalent or derived isogenous varieties is a question that has raised, both recently and less recently, much interest (see for example [2 LLT25, 6, 5]).
In this paper we completely characterize derived isogenous genus 1 curves, generalizing the results of [1] to the twisted setting.
Main Theorem. Let k be a perfect field and let C,C′ be genus 1 curves over k. The following are equivalent:
C and C′ are derived isogenous;
the Jacobians J(C) and J(C′) are isomorphic over k;
C and C′ are connected by a chain (of length two) of twisted derived equivalences passing through J(C).
Ramachandran and Rosenberg in [14] show that a genus 1 curve C with Jacobian J satisfies Db(C)≃Db(J,αC) for a Brauer class αC determined by the class of C in H1(k,J).
Hence any two genus 1 curves with isomorphic Jacobians are derived isogenous. From this, the implications (2) ⇒ (3) ⇒ (1) in the above statement follow easily. Our main input is the proof that derived isogenous genus 1 curves have the same Jacobian, which will be a consequence of the following:
Theorem. If C and C′ are twisted derived equivalent genus 1 curves, then there is an isomorphism between their Jacobians J(C)≃J(C′).
Our strategy is to leverage the main construction of [8], where, starting from a twisted variety (X,α), the author provides a group algebraic space O(X,α) which is a twisted derived invariant. More precisely, given a twisted equivalence Φ:Db(X,α)→Db(Y,β), Olsson constructs an isomorphism of group algebraic spaces
RΦ:O(X,α)⟶O(Y,β).
When X and Y are genus 1 curves, using that the classes α and β trivialize when passing to the algebraic closure of k, we construct, in § 3, isomorphisms
θτ:J(X)×J(X)⟶O(X,α),θτ′:J(Y)×J(Y)⟶O(Y,β)
such that the composition θτ′−1∘RΦ∘θτ is symplectic in the sense of [11] (see also Subsection 2.1 for more details). We can then conclude thanks to Proposition 2.2, which shows that a symplectic isomorphism E×E⟶E′×E′, where E and E′ are elliptic curves, yields an isomorphism E≃E′.
Notation. We work over a perfect field k with algebraic closure k and absolute Galois group Gk=Gal(k/k). Given X a reduced scheme over k we denote by X its base extension over k. For γ∈Gk and an object or morphism Z defined over k we write γZ for its Galois conjugate.
A genus 1 curve is a smooth projective geometrically connected curve of genus 1; its Jacobian J(C)=PicC0 is an elliptic curve.
Given an abelian variety A over k, we denote by A its dual, and we identify A=A. Given a homomorphism f:A→B of abelian varieties over k, we denote by f:B→A its dual. For an integer m we denote by [m]:A→A the multiplication by m. For an elliptic curve E we write λE:E→E for its canonical principal polarization; recall that λE=λE.
Acknowledgments. The second and third authors would like to thanks D. Rydh for valuable mathematics conversation and M. Olsson for a very nice e-mail exchange. The second author was partially supported by the grant number VR2023-03837 of the Vetenskapsrådet and by the grant number 2023.0395 of the Knut and Alice Wallenberg foundation.
AI Usage. Claude Opus 5.5 was used to check for correand polish the exposition.
Preliminary results and Background Material
Abelian varieties and Symplectic Isomorphisms
Let A and B be abelian varieties over k. A morphism f:A×A→B×B can be represented by a matrix of morphisms:
f=(xzyw),
with x:A→B, y:A→B, z:A→B and w:A→B. We recall the following definition.
Definition 2.1. An isomorphism f:A×A→B×B is said to be symplectic (or isometric) if
f−1=(w−z−yx).
The fact that two abelian varieties A and B are tied by a symplectic isomorphism A×A≃B×B is very strong: over an algebraically closed field of characteristic not 2, this is equivalent to A and B being derived equivalent ([Orl02, Pol96, Pol03]). Since for elliptic curves derived equivalence implies isomorphism ([1]), we directly get that elliptic curves tied by a symplectic isomorphism are isomorphic over (most) algebraically closed fields. The next result says that the same is true over any field.
Proposition 2.2. Let k be a field and let E1, E2 be elliptic curves over k. If there is a symplectic isomorphism f:E1×E1→E2×E2 over k, then E1 and E2 are isomorphic over k.
Proof. Write f in matrix form, that is f=(xzyw), and set a:=degx, b:=degy, c:=degz and d:=degw. Using the principal polarizations λEi:Ei→Ei we identify Ei with Ei. Then x,y,z,w become elements of Homk(E1,E2), and the dual of a homomorphism u:E1→E2 becomes its dual isogeny λE1−1∘u∘λE2:E2→E1, which we still denote by u; recall that u∘u=[degu] and u∘u=[degu].
Since f is symplectic and the isomorphisms λEi are symmetric, f−1 can be expressed as in (2.1). Comparing the entries of the first row of f∘f−1=IdE2×E2 gives
x∘w−y∘z=1E2,y∘x=x∘y.(1)
If x=0, the first equation gives −y∘z=1E2, so y is an isomorphism and E1 and E2 are isomorphic. Similarly, if z=0, then x∘w=1E2 and x is an isomorphism. Therefore we can assume that x,z=0, and thus a,c=0.
As the dual of y∘x is x∘y, the second equation in (2.2) implies that y∘x is self-dual, so it must be multiplication by an integer n. Indeed, the map u↦u is the Rosati involution of the canonical principal polarization, whose fixed points in End(E2) are the integers (see e.g. [Sil09, III.6.2 and §III.9]).
By comparing degrees, we have n2=ab; in particular ab is a perfect square. Now we compose both sides of the second equation of (2.2) on the right with x. Using that x∘x=[a], we get
ay=nx.(2)
Now we turn our attention to the second row of f∘f−1, which gives the equations
z∘w=w∘z,w∘x−z∘y=1E2.(3)
As before, z∘w is self-dual, hence multiplication by an integer m with m2=cd. Composing on the right with z, and using z∘z=[c], we get
cw=mz.(4)
By (2.3) and (2.5) we have y=anx in Hom(E1,E2)⊗Q, and, taking duals, w=cmz in Hom(E2,E1)⊗Q. Substituting into the first equation of (2.2), we get the following equalities in End(E2)⊗Q:
1E2=x∘(cmz)−anx∘z=(cm−an)x∘z.(5)
This means that x∘z is a rational multiple h of the identity. Comparing degrees, we get h2=ac, and therefore h is an integer. Now observe that, from (2.6), we get
1=chm−ahn.
Put p:=hm/c and q:=hn/a. These are rational numbers with
p2=c2h2m2=c2ac⋅cd=ad,q2=a2h2n2=a2ac⋅ab=bc,
so p and q are integers. Since p−q=1, the integers p and q are relatively prime. Therefore the same is true for their squares p2=ad and q2=bc. Any common divisor of a and b divides both ad and bc. Thus we deduce that a and b are coprime. As ab=n2 is a perfect square, a and b are themselves perfect squares. Write a=r2 with r a positive integer. Since r2=a divides n2=ab, the integer r divides n. Put s:=n/r∈Z, so that s2=b. Then (2) reads r2y=rsx. Since Homk(E1,E2) is torsion free and r=0, we get
ry=sx.
Moreover r and s are coprime, because r2=a and s2=b are. Choose u,v∈Z with ur+vs=1. Then
x=(ur+vs)x=urx+vsx=urx+vry=r(ux+vy).
Again comparing degrees we have
a=degx=r2deg(ux+vy)=adeg(ux+vy).
Therefore deg(ux+vy)=1 because a=0. Hence ux+vy:E1→E2 is an isomorphism defined over k, which concludes the proof. □
We conclude by remarking that a variant of the proof of Proposition 2.2 works for principally polarized abelian varieties A and B over k, with polarizations λA and λB, whose only endomorphisms fixed by the Rosati involution are the integers, i.e. with Homksym(A,A)=ZλA and Homksym(B,B)=ZλB. In that case the self-dual composites y∘x and x∘x are integers by hypothesis, the second one nonzero when x=0 by positivity of the Rosati involution. This shows that x and y are proportional, and one concludes by comparing degrees in the first equation of (1), using dimA=dimB.
Olsson groups
In this paragraph we review the main construction in [8]. Before beginning, we observe that the introduction of [8] is phrased over an algebraically closed field, but the constructions and proofs of the results we use are carried out over an arbitrary field; see [8 3.1 and §5].
Let X be a smooth projective geometrically connected variety over k and let π:X→X be a Gm-gerbe with class α∈Br(X). As in the introduction, we write Db(X,α)=Db(X) for the derived category of 1-twisted sheaves. Let AutX be the stack which to any k-scheme T associates the groupoid of isomorphisms XT→XT inducing the identity on the stabilizer groups Gm. By [8], AutX is a Gm-gerbe over a group algebraic space AutX, locally of finite type over k. We set
O(X,α):=AutX0,
the identity component, and we call it the Olsson group of (X,α). Up to isomorphism, it only depends on α, since gerbes associated to the same class are isomorphic.
By [8 Theorem 1.1(2)] there is an exact sequence of group algebraic spaces
1⟶PicXιXAutXrXAutX;
Equation not extracted
(6)
more precisely, when the set is nonempty, the automorphisms of XT lying over a given automorphism of XT form a torsor under PicX(T)[8 3.5 and Remark 3.6].
The map ιX sends a line bundle L to the automorphism tL of X over idX given by twisting by L. Explicitly, if ξ is an object of X over a T-point x of X, then tL(ξ)=ξ⊗π∗L, where ⊗ denotes the action of Gm-torsors on X given by its banding. (Depending on conventions, [8 3.5] may use L−1 instead of L; this plays no role below.) The map rX sends σ to the automorphism σˉ of X that it induces. Since X is the rigidification of X, every morphism from X to an algebraic space factors uniquely through π. So σˉ is the unique automorphism of X with π∘σ≃σˉ∘π.
Restricting to identity components we obtain an exact sequence
0→KX,α→O(X,α)→AutX0,KX,α:=PicX∩O(X,α).(7)
Since PicX0 is connected, PicX0⊆KX,α, but equality need not hold. On the other hand, O(X,α) is of finite type, being the identity component of a group algebraic space locally of finite type over k. So KX,α has finitely many connected components, and its image in NS(X)=Pic(X)/Pic0(X) is a finite group. Therefore
PicX0⊆KX,α⊆PicXτ.
In particular KX,α=PicX0 whenever NS(X) is torsion free, for instance when X is a curve or an abelian variety.
Olsson group for trivial Brauer classes
Let X=BGm,X be the trivial gerbe. Then
Db(X)∼Db(BGm,X)(1)=Db(X,α),M↦M⊠χ.
where χ is the universal invertible sheaf on BGm (its pullback along the T-point of BGm given by a line bundle N on T is N). The stabilizers act on χ through the standard character. By [8 Example 1.2] there is an isomorphism of stacks AutX×PicX→AutBGm,X sending a pair (a,M), with a∈AutX(T) and M∈PicX(T), to the automorphism σ(a,M) which sends a T-point (x,N) of X×BGm to (a(x),N⊗x∗M). Under this isomorphism the composition law reads
σ(a,M)∘σ(a′,M′)=σ(aa′,M′⊗a′∗M).(8)
so that AutBGm,X≃AutX⋉PicX. If PicX0 is an abelian variety, then AutX0 acts trivially on PicX0 (a connected group scheme acts trivially on an abelian variety by group automorphisms), and therefore
O(X,0)≃AutX0×PicX0
as group algebraic spaces over k. In any case, the identity component of AutX×PicX has underlying space AutX0×PicX0. So, for the trivial gerbe the kernel in (7) is KX,0=PicX0, and the map to AutX0 is surjective. We will need the following two lemmas. The first one describes the action of O(X,0)(T) on Db(XT).
Lemma 2.3. Let (a,M)∈AutX(T)×PicX(T) and σ=σ(a,M). Under the identification Db(BGm,XT)(1)≃Db(XT), the functor σ∗ is F↦a∗F⊗M.
Proof. For a=id this is simply [8 Proposition 6.1] in the case of the trivial gerbe. For the general case, let p and q be the projections from XT×BGm to first and second factor, respectively. By definition, σ sends a T-point (x,N) to (a(x),N⊗x∗M). Hence p∘σ=a∘p. Moreover, the pullback of (q∘σ)∗χ along (x,N) is N⊗x∗M, which is the pullback of q∗χ⊗p∗M along the same point; since this holds functorially in (x,N), we get (q∘σ)∗χ≃q∗χ⊗p∗M. Therefore
σ∗(F⊠χ)=(p∘σ)∗F⊗(q∘σ)∗χ≃p∗a∗F⊗q∗χ⊗p∗M=(a∗F⊗M)⊠χ.
The second lemma is about the kernel of the action on k-valued points when k is an algebraically closed field.
Lemma 2.4. Assume that k is algebraically closed. If (a,M)∈AutX(k)×PicX(k) and a∗(−)⊗M≃id as functors on Db(X), then a=idX and M≃OX.
Proof. This is a special case of [8 Remark 1.8], which deduces the faithfulness of the action of AutX(k) on Db(X,α) from the uniqueness of Fourier–Mukai kernels [3 Theorem 1.1]. Directly: evaluating on the skyscraper sheaves k(x) gives a−1(x)=x for all closed points x, so a=idX, and evaluating on OX gives M≃OX.
Twisted Rouquier isomorphisms
The main result of [8] is that the Olsson group is a twisted derived invariant. Let Y be another smooth projective geometrically connected variety over k, let Y→Y be a Gm-gerbe with class β∈Br(Y), and let Φ:Db(X,α)→Db(Y,β) be an exact equivalence. By [3 Theorem 1.1], Φ is of Fourier–Mukai type, and by [8 Theorems 1.7 and 1.9] it induces an isomorphism of group algebraic spaces
RΦ:O(X,α)⟶O(Y,β),
which we call the twisted Rouquier isomorphism associated to $\Phi.
Let an automorphism σ of a gerbe act on the twisted derived category by σ∗, as in [8 Example 1.5]. Then, on k-points, RΦ is given by conjugation [8 Remark 7.2]:
Φk∘σ∗∘Φk−1≃RΦ(σ)∗.(9)
When α and β are trivial and O(X,0) and O(Y,0) are reduced, then the twisted Rouquier isomorphism is the isomorphism constructed by Rouquier in [13]. In particular, when X and Y are abelian varieties with an equivalence Φ:Db(X)→Db(Y), we have O(A,0)=X×X and O(B,0)=Y×Y, up to some negligible conventions discussed in Remark 2.5. Then the twisted Rouquier isomorphism is exactly the isomorphism fΦ constructed by Orlov [Orl02]. In particular, RΦ is symplectic by [Orl02, Corollary 2.13].
Remark 2.5 (Conventions). The comparison above depends on two conventions: whether an automorphism a acts on Db(X) through a∗ or through a∗=(a−1)∗, and whether a line bundle L acts through −⊗L or through −⊗L−1 (see the parenthetical remark after (2.7)). These choices do not affect symplecticity. Indeed, let A and B abelian varieties with Φ:Db(A)≃Db(B) an equivalence between their untwisted derived categories. The isomorphism fΦ:A×A→B×B constructed by Orlov, is such that, over k, we have fΦ(a,ξ)=(b,η) if, and only if
Φ∘(ta∗(−)⊗Pξ)∘Φ−1≃tb∗(−)⊗Pη,(10)
where ta denotes the translation by a and Pξ denotes the line bundle in Pic0(A) corresponding to ξ∈A(k). Now, let f′ be the isomorphism characterized by the analogue of (2.11) in which ta∗ and tb∗ are replaced by ta∗=t−a and tb∗=t−b. Then f′(a,ξ)=(b,η) if and only if f(−a,ξ)=(−b,η), that is,
f′=ϵB∘f∘ϵA,where ϵA:=(−1001):(a,ξ)⟼(−a,ξ).
Similarly, replacing Pξ and Pη by Pξ−1=P−ξ and Pη−1=P−η replaces f by ϵB′∘f∘ϵA′, where ϵA′:=−ϵA; which is again ϵB∘f∘ϵA (and changing both conventions leaves f unchanged). In matrix form f′′ is represented as follows
ϵB∘(xzyw)∘ϵA=(x−z−yw).
If f is symplectic, then by (2.1) the inverse of this isomorphism is
ϵA∘f−1∘ϵB=ϵA∘(w−z−yx)∘ϵB=(wzyx),
which is exactly the matrix that Definition 2.1 prescribes for (x−z−yw), because −u=−u^. Hence eB∘f∘eA is symplectic whenever f is, and we may ignore these conventions. □
We now describe two twisted Rouquier isomorphisms that play a central role in § 3.
Example 2.6 (Conjugation by an isomorphism of gerbes). Let X and X′ be Gm-gerbes over X, and let τ:X→X′ be an isomorphism of Gm-gerbes over X, that is, an isomorphism of stacks which commutes with the projections to X and induces the identity on stabilizer groups. Then τ∗:Db(X′)(1)→Db(X)(1) is an equivalence and thus induces an isomorphism Rτ∗:O(X,β)→O(X,α), where α and β are the Brauer classes associated to X→X and X′→X (of course α=β, since the two gerbes are isomorphic). By (9) and the faithfulness of the action [8 Remark 1.8], we see that Rτ∗ is the isomorphism induced after rigidification by the isomorphism of group stacks AutX′→AutX given by conjugation by τ: σ↦τ−1∘σ∘τ.
To lighten the notation we will, in the sequel, denote Rτ∗ simply by Rτ.
We see that Rτ behaves well with respect to the exact sequences (6) for X′ and X, i.e. the diagram
1
⟶
PicX
ιX′
AutX′
rX′
AutX
∥
↓Rτ
∥
1
⟶
PicX
ιX
AutX
rX
AutX
Table 2.12.
commutes. Let π:X→X and π′:X′→X be the structure maps. Saying that τ is an isomorphism of gerbes over X means that π′∘τ≃π. Hence, for every automorphism σ of X′,
π∘(τ−1στ)≃π′∘σ∘τ≃σˉ∘π′∘τ≃σˉ∘π.
By the uniqueness in the definition of r, this gives rX(τ−1∘σ∘τ)=σˉ=rX′(σ), so the right square commutes. For the left square, observe that τ induces the identity on stabilizers, so it is compatible with the action of Gm-torsors: τ(ξ⊗P)≃τ(ξ)⊗P, functorially in the object ξ and in the Gm-torsor P. Taking P=x∗L gives τ∘tL≃tL∘τ, that is, Rτ(ιX′(L))=ιX(L).
The case of main interest for us will be when, over kˉ, the gerbe Xkˉ→X is associated to the trivial Brauer class and τ:Xkˉ→BGm,X is a trivialization. Applying the above over kˉ and passing to identity components, by Subsection 2.2.1 the top row of (2.12) restricts to the split exact sequence 0→PicX0→O(X,0)kˉ→AutX0→0. We therefore obtain an isomorphism of group algebraic spaces Rτ:O(X,0)kˉ→O(X,α)kˉ fitting in the commutative diagram
0
⟶
PicX0
⟶
O(X,0)kˉ
⟶
AutX0
⟶
0
∥
↓Rτ
∥
0
⟶
PicX0
⟶
O(X,α)kˉ
⟶
AutX0
⟶
0
Table 2.13.
The bottom row is exact, because the top row is and Rτ is an isomorphism. When PicX0 is an abelian variety, O(X,0)=AutX0×PicX0.
Example 2.7 (The automorphisms Ad(Q)). Let Q be a line bundle on X. Then σ(id,Q) is an automorphism of the trivial gerbe BGm,X, with inverse σ(id,Q)−1=σ(id,Q−1) by (8). Applying Example 2.6 to τ=σ(id,Q−1), we obtain the automorphism
Ad(Q):=Rσ(id,Q−1):ρ⟼σ(id,Q)∘ρ∘σ(id,Q)−1
of AutBGm,X, that is, conjugation by σ(id,Q). By Example 2.6 and Lemma 2.3, it is the twisted Rouquier isomorphism of the autoequivalence σ(id,Q−1)∗=−⊗Q−1 of Db(X). Note that Ad(Q) is not an automorphism of BGm,X itself: it acts on automorphisms of BGm,X. It preserves the identity component O(X,0), and we denote the induced automorphism again by Ad(Q).
By (2.9) we have Ad(Q⊗Q′)=Ad(Q)∘Ad(Q′), and
Ad(Q)(σ(a,L))=σ(a,L⊗a∗Q⊗Q−1)(11)
Assume now that PicX0 is an abelian variety, so that O(X,0)=AutX0×PicX0. Then Ad(Q) acts on it by (a,L)↦(a,L⊗a∗Q⊗Q−1). If Q∈Pic0(X), this automorphism is the identity: the morphism a↦a∗Q⊗Q−1 from AutX0 to PicX0 is trivial, because AutX0 acts trivially on the abelian variety PicX0. Consequently, on AutX0×PicX0 the automorphism Ad(Q) only depends on the class of Q in NS(X). In Section 3 we use this over k.
Olsson groups for Algebraic Brauer classes
The main result of this section is the following
Proposition 3.1. If C is a genus 1 curve and α∈Br(C), then
O(C,α)≃J(C)×J(C).
We say that α∈Br(X) is algebraic when it sits in the kernel of the map Br(X)→Br(X). Note that every class α∈Br(C) becomes trivial over k: since C is a smooth projective curve over an algebraically closed field, Br(C)=0 by Tsen’s theorem. Thus every α∈Br(C) is algebraic.
We will put ourselves in a much more general situation and work with X, a smooth geometrically connected projective variety over a perfect field k such that PicX0 is an abelian variety. The idea is that, when α is algebraic, over k there is an isomorphism
O(X,α)k≃(PicX0×AutX0)k,
since α is trivial by hypothesis and PicX0 is an abelian variety (see Subsection 2.2.1). So we want to define a morphism AutX0×PicX0→O(X,α) over k whose base change to k is the isomorphism Rτ of Example 2.6.
If α is algebraic, then Xk is a trivial gerbe, so there is an isomorphism of Gm-gerbes
τ:Xk∼BGm,X,
which, by a limit argument, is defined over some finite Galois extension k′/k. Let γ∈Gk. The projections Xk→X and BGm,X→X are base changes of morphisms defined over k, hence Galois equivariant. So the conjugate γτ again commutes with the projections to X and induces the identity on stabilizers. Consider the composite
ργ:=γτ∘τ−1:BGm,X→BGm,X.
Clearly, it induces the identity on stabilizers, so it is an object of AutBGm,X(k). In addition, it commutes with the projection to X, so its image under the map AutBGm,X→AutX of [8 Theorem 1.1(2)] is idX. Hence, by [8 Example 1.2] (see Subsection 2.2.1), it is of the form σ(id,Mγ) for a line bundle Mγ on X. By Lemma 2.4 this line bundle is unique up to isomorphism.
Lemma 3.2. In the notation above we have the following:
(1) for γ∈Gal(k/k′), Mγ≃OX;
(2) for all γ,δ∈Gk, Mγδ≃Mγ⊗γMδ;
(3) every other trivialization of Xk is of the form τ′=σ(id,Q)∘τ with Q∈Pic(X). In addition, if we denote by Mγ′ the line bundle associated to γτ′∘τ′−1 then
Mγ′≃Mγ⊗γQ⊗Q−1.
Proof. The first claim is clear since τ is defined over k, and thus left fixed by conjugation by γ. We turn our attention to (2). Since the isomorphism AutX×PicX≃AutGm,X is defined over k, we have that
where the last step is (2.9), and the statement is proven.
By Lemma 3.2(2), the map γ↦[Mγ]∈NS(X) is a 1-cocycle, so it defines a class in H1(k,NS(X)). By Lemma 3.2(3), replacing τ by another trivialization σ(id,Q)∘τ changes this cocycle by γ↦[γQ⊗Q−1]=γ[Q]−[Q], which is a coboundary; hence the class does not depend on the choice of τ. It does not depend on the gerbe X either, but only on α. In fact, given another gerbe X′ representing α and an isomorphism i:X′→X over X, we have
γ(τ∘i)∘(τ∘i)−1=γτ∘τ−1.
Thus we can define the following
Definition 3.3. For α∈Br(X) algebraic we define the class
ob(α):=[γ↦MγmodPic0(X)]∈H1(k,NS(X)).
Remark 3.4. If NS(X) is torsion free with trivial Galois action, then
H1(k,NS(X))=0,
hence ob(α)=0. This applies to genus 1 curves, for which NS(C)=Z via the degree.
From now on, until the end of this section, we work under the following assumptions.
Setting 3.5. Let X be a smooth projective geometrically connected variety over k such that PicX0 is an abelian variety, and let α∈Br(X) be algebraic.
Let τ:Xk→Gm,X be a trivialization. By Example 2.6, conjugation by τ gives an isomorphism of group algebraic spaces over k
Rτ:(AutX0×PicX0)k⟶O(X,α)k,
fitting in the commutative diagram (2.13) with exact rows. Here we use that O(X,0)≃AutX0×PicX0, because PicX0 is an abelian variety (Subsection 2.2.1). For a line bundle Q on X we also use the automorphism Ad(Q) of (AutX0×PicX0)k of Example 2.7. It is the identity if Q∈Pic0(X), and in general it only depends on the class of Q in NS(X).
Lemma 3.6.Under the assumption in Setting 3.5, we have the following:
If ob(α)=0, there is a trivialization τ:Xk→BGm,X with Mγ∈Pic0(X) for all γ∈Gk.
Let τ and τ′ be two trivializations of Xk, and write τ′=σ(id,Q)∘τ with Q∈Pic(X) as in Lemma 3.2(3). Then Rτ′=Rτ∘Ad(Q)−1. If moreover ob(α)=0 and both τ and τ′ are as in (1), then the class of Q in NS(X) is Galois invariant.
Proof. (1) Let τ0 be any trivialization, with associated line bundles Mγ0. Since ob(α)=0, there is ν∈NS(X) such that the class of Mγ0 is γν−ν for all γ∈Gk. Let Q∈Pic(X) be a lift of −ν and put τ=σ(id,Q)∘τ0. Denote the associated line bundles by Mγ. By Lemma 3.2(3), the class of Mγ in NS(X) is (γν−ν)−(γν−ν)=0, that is, Mγ∈Pic0(X).
If both τ and τ′ are as in (1), Lemma 3.2(3) gives γQ⊗Q−1≃Mγ′⊗Mγ−1∈Pic0(X) for all γ∈Gk, so the class of Q in NS(X) is Galois invariant. □
The next result shows that O(X,α)≃AutX0×PicX0 when the obstruction ob(α) vanishes. Proposition 3.1 then follows immediately from Remark 3.4 and the classical facts that, for a genus 1 curve C, we have PicC0≃J(C) and AutC0≃J(C).
Proposition 3.7.Under the hypothesis of Setting 3.5, assume in addition that ob(α)=0, and let τ:Xk→BGm,X be a trivialization as in Lemma 3.6(1). Then Rτ descends to an isomorphism of group algebraic spaces over k
θτ:AutX0×PicX0∼O(X,α),
*compatible with the inclusions of PicX0 and with the maps to AutX0; that is, θτ fits in the analog of (2.13) over k. In particular, in (2.8) the map O(X,α)→AutX0 is surjective and KX,α=PicX0, so that O(X,α) is a split extension of AutX0 by PicX0.
Proof. Once we know that Rτ descends to an isomorphism θτ of group algebraic spaces over k, we are done. In fact, the commutativity of the analog of (2.13) over k can be checked over k, where it is (2.13) itself. The surjectivity of O(X,α)→AutX0 and the equality KX,α=PicX0 can also be checked over k, where they follow from the exactness of the bottom row of (2.13). Finally, θτ itself gives the splitting.
To construct θτ, we use Galois descent. We first observe that the trivialization τ is defined over some finite Galois extension k′ of k, and thus so is Rτ.
Now, for γ∈Gk, we have γ(Rτ)=Rγτ. In addition, by definition of Mγ we have
γτ=ργ∘τ=σ(id,Mγ)∘τ.
Hence γ(Rτ)=Rτ∘Ad(Mγ)−1 by Lemma 3.6(2).
Since τ is as in Lemma 3.6(1), we have Mγ∈Pic0(X) for all γ∈Gk. Hence, as observed after (2.14), the automorphism Ad(Mγ) of (AutX0×PicX0)k is the identity. Therefore Rτ is invariant under the action of Gk, which factors through Gal(k′/k). So Rτ descends to a morphism θτ over k. The same applies to its inverse, and thus θτ is an isomorphism. It is a homomorphism of group algebraic spaces, since this can be checked over k, where it holds for Rτ. □
We now specialize to the case relevant for the main result: torsors under abelian varieties. To this aim, we let X be a torsor under an abelian variety A over k. Our main example will be a genus 1 curve, which is a torsor under its Jacobian. The action of A on X by translations identifies AutX0 with A[4 Lemma A.3]. Moreover, PicX0 is canonically isomorphic to PicA0=A[12 XIII, 1.1] (see also [14 §4]).
In particular PicX0 is an abelian variety. If α∈Br(X) is an algebraic class, we are under the assumptions of Setting 3.5. Thus, if ob(α)=0, Proposition 3.7 gives isomorphisms
θτ:A×A∼O(X,α).
Proof of the main theorem
In this section we prove the Main Theorem from the introduction.
To this aim, let C and C′ be two genus 1 curves over k with Jacobians J and J′ respectively. If (2) hold, then they are torsors over the same abelian variety J, and there are two classes α and α′ in H1(k,J) such that [C]=α and [C′]=α′. Regarding them as Brauer classes on J via Br(J)≃Br(k)⊕H1(k,J), by [14] we have twisted equivalences
Db(C)≃Db(J,α),Db(J,α′)≃Db(C′).
We deduce that (3) holds, which implies directly (1). To conclude the proof we need to show that two derived isogenous genus 1 curves have isomorphic Jacobians. Suppose that C and C′ are derived isogenous. Then there is a sequence of varieties C=X0,…,Xn=C′ and with twisted equivalences
Db(Xi,βi)≃Db(Xi+1,αi+1).
Since twisted derived equivalence preserve dimension and Kodaira dimension (or order of the canonical bundle), we can show by induction that Xi are curves of genus 1. Therefore, the last implication of the main theorem will be a consequence of Proposition 2.2 and the following result.
Proposition 4.1. Let C and C′ be as before, and let α∈Br(C) and β∈Br(C′). If there is an exact equivalence Db(C,α)≃Db(C′,β), then there is a symplectic isomorphism J×J≃J′×J′ defined over k.
Proof. Let Φ:Db(C,α)→Db(C′,β) be an equivalence, and let RΦ:O(C,α)→O(C′,β) be its twisted Rouquier isomorphism (Subsection 2.2.2). Let X→C and Y→C′ be the gerbes associated to α and β, choose trivializations τ:Xk→BGm,C and τ′:Yk→BGm,C′ as in Lemma 3.6(1), and put
F:=θτ′−1∘RΦ∘θτ:J×J⟶J′×J′.
We want to show that F is symplectic. This can be checked after base change to k. By construction, (θτ)k is simply Rτ and similarly (θτ′)k=Rτ′. In addition, the formation of RΦ commutes with extension of the base field (see [8 7.1–7.6]). Thus we have
Fk=Rτ′−1∘RΦk∘Rτ.
Recall (see Example 2.6) that the trivialization τ induces an equivalence τ∗:Db(C)=Db(BGm,C)(1)→Db(C,α), and Rτ is simply the twisted Rouquier isomorphism associated to τ∗. Similarly Rτ′∗=Rτ′. If we consider the equivalence
Φ0:=(τ′∗)−1∘Φk∘τ∗:Db(C)→Db(C′),
since the construction of twisted Rouquier isomorphisms is functorial, we have
RΦ0=R(τ′∗)−1∘RΦk∘Rτ∗=Rτ′−1∘RΦk∘Rτ=Fk.
Now it is immediate to see that Fk is symplectic, as it is the (untwisted) Rouquier isomorphism associated to an equivalence between abelian varieties and so symplectic by the work of Orlov [Orl02] (see Subsection 2.2.2).
Department of Mathematics, Indian Institute of Technology Gandhinagar, Palaj Campus, Gandhi-Nagar, Gujarat, India
[1]Benjamin Antieau, Daniel Krashen, and Matthew Ward. Derived categories of torsors for abelian schemes. Advances in Mathematics, 306:1–23, 2017.
[2]Nicolas Addington and Andrew Wray. Twisted Fourier–Mukai partners of Enriques surfaces. arXiv preprint arXiv:1803.03250, 2018.DOI
[3]Alberto Canonaco and Paolo Stellari. Twisted Fourier–Mukai functors. Adv. Math., 212(2):484–503, 2007.
[4]Katrina Honigs. Derived equivalence, Albanese varieties, and the zeta functions of 3-dimensional varieties. Proc. Amer. Math. Soc., 146(3):1005–1013, 2018. With an appendix by Jeffrey D. Achter, Sebastian Casalaina-Martin, Katrina Honigs, and Charles Vial.arxiv.org/abs/1604.00042
[5]Tyler Lane. Twisted derived equivalences between abelian varieties, 2026. arXiv:2601.16871.arxiv.org/abs/2601.16871
[6]Ruoxi Li. Point objects and derived equivalences of twisted derived categories of abelian varieties, 2025. arXiv:2511.19282.arxiv.org/abs/2511.19282
[8]Martin Olsson. Twisted derived categories and Rouquier functors. International Mathematics Research Notices, 2025(18):rnaf290, 2025.arxiv.org/abs/2501.05599
[11]Alexander Polishchuk. Abelian Varieties, Theta Functions and the Fourier Transform. Cambridge University Press, April 2003.DOI
[12]Michel Raynaud. Faisceaux amples sur les schémas en groupes et les espaces homogènes, volume 119 of Lecture Notes in Mathematics. Springer-Verlag, Berlin-New York, 1970.DOI
[13]Raphaël Rouquier. Automorphismes, graduations et catégories triangulées. J. Inst. Math. Jussieu, 10(3):713–751, 2011.DOI
[14]Niranjan Ramachandran and Jonathan Rosenberg. Derived categories of curves of genus one and torsors over abelian varieties. Math. Res. Lett., 32:293–318, 2025. arXiv:2212.14497.DOI