betaFeedback

WorksMathematics

Generic vanishing and singularities of theta divisors

Abstract

We study the connection between the generic vanishing conjecture for subvarieties of principally polarized abelian varieties and various predictions about the singularities of theta divisors. We deduce new cases of a conjecture on the multiplicities of points on irreducible theta divisors, as well as Beauville's conjecture characterizing Jacobians by the size of their theta singularity loci.

Introduction

All throughout we work over C\mathbb{C}. The main technical result of the paper is a stronger version of a conjecture made by Pareschi and the second author, about the equivalence of two notions of regularity for ideal sheaves on abelian varieties, designed according to [21] to correspond to the notion of Castelnuovo-Mumford 2-regularity in projective space, namely the implication (3) ⟹\Longrightarrow (2) in [23 Conjecture A].1 Note that while geometric nondegeneracy was an assumption in the conjecture, here we see that it is not needed, and in fact follows from the hypothesis.

Theorem A. Let (A,Θ)(A,\Theta) be an indecomposable principally polarized abelian variety, and X⊂AX \subset A a proper closed subscheme. If IX(2Θ)\mathcal{I}_X(2\Theta) satisfies IT0IT_0, then IX(Θ)\mathcal{I}_X(\Theta) is a GVGV-sheaf. Moreover, XX is geometrically nondegenerate and integral.

The case when XX is a divisor was obtained earlier in [16]. For the technical notions in the theorem above, please see §1. If a coherent sheaf F\mathcal{F} on an abelian variety is GVGV, then F(Θ)\mathcal{F}(\Theta) satisfies IT0IT_0; see [24 Proposition 3.1]. The converse however is not true for arbitrary F\mathcal{F}; see [23 Example 3.2]. Therefore the conjecture/theorem above is a property that is specific to twisted ideal sheaves. The (a priori stronger) GVGV property for IX(Θ)\mathcal{I}_X(\Theta) is more useful for applications, as it is related to a Schottky-type conjecture formulated in [23] (namely (2) ⟹\Longrightarrow (5) in that same Conjecture A in loc. cit.), which has been verified in a few cases in [23], [6], [7].

Conjecture B. Let (A,Θ)(A,\Theta) be an indecomposable principally polarized abelian variety of dimension gg. If X⊂AX \subset A is a geometrically nondegenerate closed subscheme of pure dimension 1≤d≤g−21 \le d \le g-2, then IX(Θ)\mathcal{I}_X(\Theta) is a GVGV-sheaf if and only if one of the following holds:

  1. (A,Θ)(A,\Theta) is the Jacobian of a smooth curve CC, and XX is ++ or −- a translate of Wd(C)W_d(C).

  2. (A,Θ)(A,\Theta) is the intermediate Jacobian of a smooth cubic threefold YY, and XX is ++ or −- a translate of the Fano surface FF of lines on YY.

Although not directly relevant to this paper, it is worth mentioning that both items in this conjecture should also be equivalent to XX having minimal class, i.e. [X]=θg−d/(g−d)![X] = \theta^{g-d}/(g-d)!, by a well-known conjecture of Debarre [8], which is the most significant conjecture in this direction.

It is shown in [23] that every GVGV-subscheme, i.e. one satisfying the conditions in Conjecture B, has minimal class. Therefore the conjecture on minimal classes can be considered as input in the results below; note that besides the Matsusaka-Ran criterion, which implies that it holds for curves, it is also known to hold in dimension up to four, by [26]. For more in this direction see e.g. the Introduction of [6].

While results about Conjecture B do rely specifically on the GVGV property, the IT0IT_0 property for IX(2Θ)I_X(2\Theta) is easier to verify in practice, and as a consequence, Theorem A unlocks a number of significant applications regarding the singularities of theta divisors. The main one is related to the following well-known conjecture of the multiplicities of points on irreducible theta divisors. The first part has been folklore for some time, see for instance [11 Conjecture 5.12], while the stronger second statement is formulated as [4 Question 4.7].

Conjecture C. Let (A,Θ)(A,\Theta) be an indecomposable principally polarized abelian variety of dimension gg. Then:

  1. for every x∈Θx \in\Theta we have mult⁡x(Θ)≤g+12\operatorname{mult}_x(\Theta) \le\frac{g+1}{2}.

  1. more generally, we have

dim⁡Σm(Θ)≤g−2m+1,\dim\Sigma_m(\Theta) \le g - 2m + 1,

where Σm(Θ):={x∈Θ∣mult⁡x(Θ)≥m}\Sigma_m(\Theta) := \{x \in\Theta\mid\operatorname{mult}_x(\Theta) \ge m\}.

Remark (Equality case). Note that (2) implies that in (1) we can also add that the upper bound mult⁡x(Θ)=g+12\operatorname{mult}_x(\Theta) = \frac{g+1}{2}, if attained at all, can only be attained at finitely many points. In fact we will see that we can conjecture something more precise: the upper bound is attained at exactly one point, and only on the Jacobian of a hyperelliptic curve of odd genus, at the multiple of a g21g^1_2, or at the origin of the intermediate Jacobian of a smooth cubic threefold.

Conjecture C is known in dimension up to 5 by Casalaina-Martin’s work on singularities of theta divisors on Prym varieties [5], and it is also known when Θ\Theta has isolated singularities by [19 Theorem I]. Using a strategy outlined by the second author in a 2018 Oberwolfach lecture [25], based on the use of Hodge ideals as in [19], and which was conditional on a result like Theorem A, we show:

Theorem D. Conjecture B implies Conjecture C.

Thanks to results in [23], [6] and [7], we then get:

Corollary E. Conjecture C holds unconditionally for g≤6g \le6, or for arbitrary gg when dim⁡Sing⁡(Θ)≤2\dim\operatorname{Sing}(\Theta) \le2 or dim⁡Sing⁡(Θ)=g−3\dim\operatorname{Sing}(\Theta) = g - 3.

Recall that an irreducible theta divisor is normal, hence in any case

dim⁡Sing⁡(Θ)≤g−3.\dim\operatorname{Sing}(\Theta) \le g - 3.

For a more precise statement about the unconditional cases, please see Corollary 8.3. This also gives an alternative proof for the case g≤5g \le5.

An upper bound on the dimension of Σm(Θ)\Sigma_m(\Theta) that goes in the direction of Conjecture C(2) is given in Theorem 8.1 (which is independent of Theorem A or Conjecture B) and Corollary 8.2. A special case of the latter that is easy to state is:

Theorem F. Conjecture C(2) holds in arbitrary dimension for m≤3m \le3, or when Θ\Theta is 11-Du Bois.3

The method we use for the results above also leads to a proof of the following Schottky-type conjecture originating with Beauville [2]; see also Debarre’s [9 Conjecture 3] and the surrounding discussion. It was verified in dimension at most 5 in [2 Proposition 6.6], using the theory of generalized Prym varieties.

Theorem G. If a ppav (A,Θ)(A,\Theta) of dimension gg satisfies dim⁡Sing⁡(Θ)=g−3\dim\operatorname{Sing}(\Theta)=g-3, then (A,Θ)(A,\Theta) is the Jacobian of a hyperelliptic curve.

It is of course known that all hyperelliptic Jacobians satisfy this property. It is also not hard to show that if Θ\Theta is reducible, then dim⁡Sing⁡(Θ)=g−2\dim\operatorname{Sing}(\Theta)=g-2. (See [10 Corollary 2] for a converse to this statement, and more.) Therefore the conjecture/theorem fully describes the case of maximal dimension for singular loci of irreducible theta divisors. As shown in [9 §4.3] (see also [12] for more details), it also implies a conjecture of Debarre [9 Conjecture 2] on the base loci of polarizations; cf. also [1] for related results.4

Corollary H. If (A,L)(A,L) be a polarized abelian variety, then Bs⁡(L)\operatorname{Bs}(L) is generically reduced along any component of codimension 2.

The main technical tool used in the paper is the zero-scheme of the first Hodge ideal I1(Θ)I_1(\Theta) of [19]. The method applies however to the center of minimal exponent of [29] as well, answering a problem raised there (and shown in dimension one) in a few additional cases; see Corollary 10.1 for the precise statement.

Acknowledgements

The first author would like to thank G. Pareschi for many conversations over the years. His ideas have shaped many of the questions and techniques in this paper. We thank J. McKernan for helpful discussions and comments. We would also like to thank S. Huang and Z. Jiang, who have shared their work [12] written at the same time as ours, where they also show Theorem A, and Beauville’s conjecture Theorem G, using the same general strategy. AI was used for cleaning up and catching small errors; note however that some of the unconditional statements here, as well as in [12], rely on the recent preprint [7], which provides an AI proof of Conjecture B for surfaces in ppavs of dimension g≥6g \ge6.

Preliminaries

Vanishing properties

Let (A,Θ)(A,\Theta) be a complex principally polarized abelian variety (ppav), and let A^=Pic⁡0(A)\widehat{A}=\operatorname{Pic}^{0}(A). The cohomological support loci of a coherent sheaf F\mathcal{F} on AA are

Vi(F):={α∈A^∣Hi(A,F⊗α)≠0}.V^{i}(\mathcal{F}):=\{\alpha\in\widehat{A}\mid H^{i}(A,\mathcal{F}\otimes\alpha)\ne0\}.

We say that F\mathcal{F} satisfies IT0IT_{0} if

Hj(A,F⊗α)=0for all α∈A^ and all j>0,H^{j}(A,\mathcal{F}\otimes\alpha)=0\quad\text{for all }\alpha\in\widehat{A}\text{ and all }j>0,

i.e. Vi(F)=∅V^{i}(\mathcal{F})=\varnothing for all i>0i>0. We say that F\mathcal{F} is a GVGV-sheaf if

codim⁡A^Vi(F)≥ifor all i>0.\operatorname{codim}_{\widehat{A}}V^{i}(\mathcal{F})\ge i\quad\text{for all }i>0.

A closed subscheme X⊆AX\subseteq A is called a GVGV-subscheme if IX(Θ)\mathcal{I}_{X}(\Theta) is a GVGV-sheaf.

Theta-dual

For a reduced closed subscheme Y⊆AY \subseteq A, we denote its theta-dual by V(Y)V(Y). Set-theoretically,

V(Y)={a∈A∣Y⊆Θ+a}.V(Y)=\{a\in A\mid Y\subseteq\Theta+a\}.

In particular Y−V(Y)⊆ΘY-V(Y)\subseteq\Theta. For a detailed study of theta-duals, please see [23]; in particular, they are endowed there with a scheme structure, using the Fourier-Mukai transform.

Pontrjagin product

Let p1,p2:A×A→Ap_1,p_2:A\times A\to A be the natural projections, and let

m:A×A→A,m(x,y)=x+ym:A\times A\to A,\quad m(x,y)=x+y

be the addition map. For coherent sheaves E,G\mathcal{E},\mathcal{G} on AA, denote E⊠G=p1∗E⊗p2∗G\mathcal{E}\boxtimes\mathcal{G}=p_1^*\mathcal{E}\otimes p_2^*\mathcal{G}, and define

E∗G=m∗(E⊠G)\mathcal{E}*\mathcal{G}=m_*(\mathcal{E}\boxtimes\mathcal{G})

to be their Pontrjagin product, with the derived version E∗RG\mathcal{E}*^{\mathrm{R}}\mathcal{G}. Thus the ordinary and derived Pontrjagin products agree whenever Rjm∗(E⊠G)=0R^jm_*(\mathcal{E}\boxtimes\mathcal{G})=0 for j>0j>0. More about sheaf-theoretic Pontrjagin products can be found in [22 §5].

By interchanging the factors via τ(x,y)=(y,x)\tau(x,y)=(y,x) and the equality m∘τ=mm\circ\tau=m, we get a natural isomorphism

φE,G:E∗G→∼G∗E.\varphi_{\mathcal{E},\mathcal{G}}:\mathcal{E}*\mathcal{G}\xrightarrow{\sim}\mathcal{G}*\mathcal{E}.

If a:E→E′a:\mathcal{E}\to\mathcal{E}' and b:G→G′b:\mathcal{G}\to\mathcal{G}' are morphisms, it is immediate to check that

φE′,G′∘(a∗b)=(b∗a)∘φE,G.(1)\varphi_{\mathcal{E}',\mathcal{G}'}\circ(a*b)=(b*a)\circ\varphi_{\mathcal{E},\mathcal{G}}. \tag*{(1)}

Example 3.2. (1) Let a∈Aa\in A and M=O(Θa)M=\mathcal{O}(\Theta_a). We note that

Rim∗(M⊠M)=0for all i>0,R^i m_*(M\boxtimes M)=0\quad\text{for all }i>0,

which implies that M∗M≃M∗RMM*M\simeq M*^{\mathrm{R}}M. (This will be useful later on.) Indeed, for t∈At\in A we identify m−1(t)m^{-1}(t) with AA by x↦(x,t−x)x\mapsto(x,t-x), and we write σt(x)=t−x\sigma_t(x)=t-x. Under this identification we have

(M⊠M)∣m−1(t)≃M⊗σt∗M.(M\boxtimes M)|_{m^{-1}(t)}\simeq M\otimes\sigma_t^*M.

This is an ample line bundle on AA for each tt, hence it has vanishing higher cohomology, and therefore the cohomology and base change theorem applies to imply the claim.

(2) In addition to the notation in (1), let now XX be a closed subscheme of AA, and let F=IX⊗M\mathcal{F}=\mathcal{I}_X\otimes M. Assume in addition that IX(2Θ)\mathcal{I}_X(2\Theta) satisfies IT0IT_0. We again have

Rim∗(F⊠M)=0for all i>0,R^i m_*(\mathcal{F}\boxtimes M)=0\quad\text{for all }i>0,

and consequently F∗M≃F∗RM\mathcal{F}*M\simeq\mathcal{F}*^{\mathrm{R}}M. This time we have

(F⊠M)∣m−1(t)≃F⊗σt∗M.(\mathcal{F}\boxtimes M)|_{m^{-1}(t)}\simeq\mathcal{F}\otimes\sigma_t^*M.

Since this is of the form IX(2Θb)\mathcal{I}_X(2\Theta_b) for some bb, it has vanishing higher cohomology by our assumption, and therefore the same argument applies.

Hodge ideals

We continue to consider a complex ppav (A,Θ)(A, \Theta). Recall from [19] that there is a Hodge filtration F∙OA(∗Θ)F_\bullet\mathcal{O}_A(*\Theta) on the DA\mathcal{D}_A-module OA(∗Θ)\mathcal{O}_A(*\Theta) of meromorphic functions on AA with poles of arbitrary order. In each degree it is contained in the pole order filtration PkOA(∗Θ)=OA((k+1)Θ)P_k\mathcal{O}_A(*\Theta)=\mathcal{O}_A((k+1)\Theta), hence for each k≥0k \ge0 we have

FkOA(∗Θ)=OA((k+1)Θ)⊗Ik(Θ).F_k\mathcal{O}_A(*\Theta)=\mathcal{O}_A((k+1)\Theta)\otimes I_k(\Theta).

where Ik(Θ)I_k(\Theta) is the kk-th Hodge ideal of Θ\Theta. For k=0k=0 we have

I0(Θ)≃I((1−ϵ)Θ)I_0(\Theta)\simeq\mathcal{I}((1-\epsilon)\Theta)

the multiplier ideal of the Q\mathbb{Q}-divisor (1−ϵ)Θ(1-\epsilon)\Theta, where 0<ϵ≪10<\epsilon\ll1 is a rational number; see [19 Proposition 10.1]. This holds for any reduced divisor on a smooth variety, but in our case a well known result of Kollár [14 Theorem 17.13] says then that I0(Θ)≃OAI_0(\Theta)\simeq\mathcal{O}_A, i.e. the pair (A,Θ)(A,\Theta) is log canonical.

In this paper we will deal with the Hodge ideal I1(Θ)I_1(\Theta). We will need:

Theorem 4.1. We have:

  1. For every irreducible closed subset W⊂AW\subset A of codimension rr, if m=mult⁡W(Θ)m=\operatorname{mult}_W(\Theta), then I1(Θ)⊆IW(q)I_1(\Theta)\subseteq I_W^{(q)}, where q=min⁡{m−1,2m−r}q=\min\{m-1,2m-r\}. (Here IW(q)I_W^{(q)} is the qq-th symbolic power of the ideal sheaf IWI_W.)

  1. Hi(A,I1(Θ)⊗OA(2Θ)⊗α)=0H^i(A,I_1(\Theta)\otimes\mathcal{O}_A(2\Theta)\otimes\alpha)=0 for all i>0i>0 and all α∈A^\alpha\in\widehat{A}, i.e. I1(Θ)⊗OA(2Θ)I_1(\Theta)\otimes\mathcal{O}_A(2\Theta) satisfies IT0IT_0.

The first is a special case of [19 Theorem E], a local property that holds on any smooth variety, while the second is a special case of [19 Theorem 28.2], which is in turn the better behaved abelian varieties version of the Hodge ideal analogue of Nadel’s vanishing theorem.

GV-subschemes

Vanishing away from theta-dual

The first thing to show is that all the cohomological support loci of IX(Θ)I_X(\Theta) are subsets of the theta-dual of XredX_{\mathrm{red}} (see § 3.2), under the IT0IT_0 assumption. (Eventually we will see that XX must in fact be integral.)

Proposition 5.1. Let (A,Θ)(A,\Theta) be a ppav, and let XX be a closed subscheme of AA. If IX(2Θ)I_X(2\Theta) satisfies IT0IT_0, then we have

Hi(A,IX(Θa))=0for all a∉V(Xred) and all i≥0.H^i(A,I_X(\Theta_a))=0\quad\text{for all }a\notin V(X_{\mathrm{red}})\text{ and all }i\ge0.

In other words, for all i≥0i\ge0 we have

Vi(IX(Θ))⊆V(Xred).V^i(I_X(\Theta))\subseteq V(X_{\mathrm{red}}).

Proof. Let a∉V(Xred)a\notin V(X_{\mathrm{red}}), and set

M=OA(Θa)andF=IX⊗M.M=\mathcal{O}_A(\Theta_a)\qquad\text{and}\qquad\mathcal{F}=I_X\otimes M.

We are aiming to show that

Hi(A,F)=0for all i≥0.H^i(A,\mathcal{F})=0\quad\text{for all }i\ge0.

By the choice of aa, there exist a closed point p∈Xp\in X, and a section s∈H0(A,M)s\in H^0(A,M), such that s(p)≠0s(p)\ne0. By abuse of notation, we denote by M(p)M(p) the fiber of MM at pp, thought of as a skyscraper sheaf at pp or a vector space over C\mathbb{C} depending on the specific setting. Consider

f:F↪M,g:M⟶M(p),f:\mathcal{F}\hookrightarrow M,\qquad g:M\longrightarrow M(p),

i.e. the natural inclusion and evaluation maps. Clearly g∘f=0g \circ f = 0.

We consider the composition of Pontrjagin products

F∗M→f∗idM∗M→id∗gM∗M(p),(2)\mathcal{F} \ast M \xrightarrow{f\ast\mathrm{id}} M \ast M \xrightarrow{\mathrm{id}\ast g} M \ast M(p), \tag*{(2)}

Claim: this composition is equal to 0.

Let us first conclude the proof assuming the Claim. It is immediate to check that the composed morphism above also factors as

F∗M→id∗gF∗M(p)→f∗idM∗M(p).\mathcal{F} \ast M \xrightarrow{\mathrm{id}\ast g} \mathcal{F} \ast M(p) \xrightarrow{f\ast\mathrm{id}} M \ast M(p).

Here the second map is injective; it suffices to have the injectivity of the map F⊠M(p)→M⊠M(p)\mathcal{F}\boxtimes M(p) \to M\boxtimes M(p) on A×AA \times A, but this can be identified on A×{p}A \times\{p\} with the inclusion F↪M\mathcal{F} \hookrightarrow M tensored over C\mathbb{C} with a vector space, which stays injective. Therefore we deduce that

id∗g:F∗M⟶F∗M(p)\mathrm{id}\ast g:\mathcal{F}\ast M \longrightarrow\mathcal{F}\ast M(p)

is the 0 map. On the other hand, by Example 3.2(2), for all ii we have

Hi(A,F∗M)≃Hi(A×A,F⊠M)≃Hi(A,F)⊗H0(A,M),H^i(A,\mathcal{F}\ast M) \simeq H^i(A\times A,\mathcal{F}\boxtimes M) \simeq H^i(A,\mathcal{F})\otimes H^0(A,M),

as the higher cohomology of MM vanishes. Similarly,

Hi(A,F∗M(p))≃Hi(A×A,F⊠M(p))≃Hi(A,F)⊗M(p).H^i(A,\mathcal{F}\ast M(p)) \simeq H^i(A\times A,\mathcal{F}\boxtimes M(p)) \simeq H^i(A,\mathcal{F})\otimes M(p).

Moreover, the map on cohomology induced by id∗g\mathrm{id}\ast g is identified with

Hi(A,F)⊗H0(A,M)→1⊗evpHi(A,F)⊗M(p),H^i(A,\mathcal{F})\otimes H^0(A,M) \xrightarrow{1\otimes\mathrm{ev}_p} H^i(A,\mathcal{F})\otimes M(p),

where evp\mathrm{ev}_p is the evaluation of global sections at pp. Now H0(A,M)=CsH^0(A,M)=\mathbb{C}s and M(p)M(p) are both one-dimensional, and evp(s)=s(p)≠0\mathrm{ev}_p(s)=s(p)\ne0, so evp\mathrm{ev}_p is an isomorphism. It follows that Hi(A,F)=0H^i(A,\mathcal{F})=0 for all i≥0i\ge0, as required.

It remains to prove the Claim, i.e. the vanishing of the composition in (2). To this end, we first show that the involution φM,M\varphi_{M,M} on M∗MM\ast M induced by the exchange map (x,y)↦(y,x)(x,y)\mapsto(y,x) on A×AA\times A is equal to the identity. To see this, we first note that E=M∗ME=M\ast M is a simple vector bundle on AA. Indeed, by Example 3.2(1) we have E≃M∗RME\simeq M\ast_{\mathrm{R}}M, and by [17 §3, formula (3.7)] we know that the Fourier-Mukai transform

Φ:Db(A)⟶Db(A^),\Phi:D^b(A)\longrightarrow D^b(\widehat{A}),

interchanges Pontrjagin and tensor product. Therefore

ΦA(E)≃M^⊗2.\Phi_A(E)\simeq\widehat{M}^{\otimes2}.

Here M^\widehat{M} is the total Fourier-Mukai transform of MM, which is well known to be a line bundle on A^\widehat{A} (the line bundle associated to a translate of −Θ-\Theta). The endomorphism group of a line bundle is C\mathbb{C}, and since Φ\Phi is an equivalence [17 Theorem 2.2], we obtain End⁡A(E)≃C\operatorname{End}_A(E)\simeq\mathbb{C}.

Since EE is simple, we have that φM,M=λ⋅idE\varphi_{M,M}=\lambda\cdot\mathrm{id}_E for some λ∈C\lambda\in\mathbb{C}. Moreover, φM,M\varphi_{M,M} fixes the nontrivial section s⊠ss\boxtimes s of EE, since this is preserved by the exchange involution. Hence λ=1\lambda=1.

We use this to deduce our claim, namely that the morphism

Ψ=(id∗g)∘(f∗id):F∗M⟶M∗M(p)\Psi=(\mathrm{id}\ast g)\circ(f\ast\mathrm{id}):\mathcal{F}\ast M\longrightarrow M\ast M(p)

is 0. By applying the commutation relation (1) to idM\mathrm{id}_M and gg, we get

φM,M(p)∘(id∗g)=(g∗id)∘φM,M.\varphi_{M,M(p)}\circ(\mathrm{id}\ast g)=(g\ast\mathrm{id})\circ\varphi_{M,M}.

Using φM,M=id⁡E\varphi_{M,M}=\operatorname{id}_{E}, we finally get

φM,M(p)∘Ψ=(g∗id⁡M)∘(f∗id⁡M)=(g∘f)∗id⁡M=0.\varphi_{M,M}(p)\circ\Psi=(g*\operatorname{id}_{M})\circ(f*\operatorname{id}_{M})=(g\circ f)*\operatorname{id}_{M}=0.

But φM,M(p)\varphi_{M,M}(p) is an isomorphism, so Ψ=0\Psi=0.

Corollary 5.3. Under the hypothesis of Proposition 5.1, if XX is geometrically nondegenerate, then IX(Θ)I_{X}(\Theta) is a GVGV-sheaf.

Proof. Note that for i>0i>0 and any a∈Aa\in A we have

Hi+1(A,IX(Θa))≃Hi(X,OX(Θa)),H^{i+1}(A,I_{X}(\Theta_{a}))\simeq H^{i}(X,\mathcal{O}_{X}(\Theta_{a})),

which vanishes for i≥d+1i\ge d+1, where d=dim⁡Xd=\dim X. Therefore we only need to concern ourselves with Vi(IX(Θ))V^{i}(I_{X}(\Theta)) for i≤d+1i\le d+1, and show that their dimension is at most g−ig-i.

Now since XX is geometrically nondegenerate, so is XredX_{\mathrm{red}} and by [23 Lemma 4.4] we have

dim⁡V(Xred)≤g−d−1.\dim V(X_{\mathrm{red}})\le g-d-1.

Combined with Proposition 5.1, this immediately implies what we want.

Geometric nondegeneracy will be treated in the next section, but it is automatic in the 00-dimensional case, and this already has an interesting consequence.

Corollary 5.4. Let Y⊆AY\subseteq A be a 00-dimensional closed subscheme such that IY(2Θ)I_{Y}(2\Theta) satisfies IT0IT_{0}. Then YY is exactly one reduced point.

Proof. According to Corollary 5.3, IX(Θ)I_{X}(\Theta) is a GVGV-sheaf. Thus for general a∈Aa\in A we have H1(A,IX(Θa))=0H^{1}(A,I_{X}(\Theta_{a}))=0, and therefore the restriction map

H0(A,OA(Θa))⟶H0(X,OX(Θa))H^{0}(A,\mathcal{O}_{A}(\Theta_{a}))\longrightarrow H^{0}(X,\mathcal{O}_{X}(\Theta_{a}))

is surjective. This implies that the length of XX is 1.

This provides in particular a different approach to [19 Theorem 29.2] in loc. cit., which is a proof of the multiplicity Conjecture C when Θ\Theta has isolated singularities. A careful reading of the proof in loc. cit. shows that it relies precisely on the statement of Corollary 5.4, proved there using Theorem 4.1, plus the fact that the linear series ∣2Θ∣|2\Theta| separates neither points nor tangent vectors on AA.

Geometric nondegeneracy

The second ingredient for the proof of Theorem A is a result that combined with the work of Schreieder [28] ensures the geometric nondegeneracy of a closed subscheme XX for which IX(2Θ)I_{X}(2\Theta) satisfies the IT0IT_{0} property.

Proposition 6.1. Let (A,Θ)(A,\Theta) be an indecomposable principally polarized abelian variety of dimension gg, and X⊂AX\subset A a proper closed subscheme of dimension dd. Assume that IX(2Θ)I_{X}(2\Theta) satisfies IT0IT_{0}. Then for every integral closed subscheme YY of XredX_{\mathrm{red}} of dimension dd, there exists an integral closed subscheme Z⊆V(Xred)Z\subseteq V(X_{\mathrm{red}}) of dimension g−d−1g-d-1 such that

Y−Z=Θ.Y-Z=\Theta.

In particular YY, hence also XX, is geometrically nondegenerate.

Proof. The last assertion follows for YY from the first. If 1≤d≤g−21 \le d \le g - 2, it follows by [28 Theorem 1.1(1)]. If d=g−1d = g - 1, we get that YY is a translate of Θ\Theta, in which case it is clear. If d=0d = 0, geometric nondegeneracy is automatic. Now since YY is a geometrically nondegenerate irreducible component of XredX_{\mathrm{red}}, we deduce that the latter is also geometrically nondegenerate by [28 Lemma 4]. Hence so is XX, because this property only depends on the reduced scheme structure.

We now focus on the first assertion. Let P\mathcal{P} be the normalized Poincaré bundle on A×A^A \times\widehat{A}. Under the isomorphism A≃A^A \simeq\widehat{A} induced by the principal polarization, we can regard P\mathcal{P} as a line bundle on A×AA \times A.

Set

θ=c1(OA(Θ))∈H2(A,Q)andℓ=c1(P)∈H2(A×A,Q).\theta= c_1(\mathcal{O}_A(\Theta)) \in H^2(A,\mathbb{Q}) \quad\text{and} \quad\ell= c_1(\mathcal{P}) \in H^2(A \times A,\mathbb{Q}).

Let

δ:A×A⟶A,(a,b)↦a−b\delta: A \times A \longrightarrow A,\quad(a,b) \mapsto a-b

be the difference map. By the construction of P\mathcal{P} in [18 p.78], our choice of the isomorphism A≃A^A \simeq\widehat{A}, and the fact that P\mathcal{P} is trivial when restricted to A×{0}A \times\{0\} or {0}×A\{0\} \times A, we have that

ℓ=δ∗θ−p1∗θ−p2∗θ∈H1(A,Q)⊗QH1(A,Q)⊆H2(A×A,Q).\ell= \delta^*\theta- p_1^*\theta- p_2^*\theta\in H^1(A,\mathbb{Q}) \otimes_{\mathbb{Q}} H^1(A,\mathbb{Q}) \subseteq H^2(A \times A,\mathbb{Q}).

Denote the two natural projections by p1,p2:A×A⟶Ap_1,p_2 : A \times A \longrightarrow A, and consider the Fourier–Mukai transform of IX(Θ)\mathcal{I}_X(\Theta), namely

C:=Φ(IX(Θ))=Rp2∗(p1∗IX(Θ)⊗P).C := \Phi(\mathcal{I}_X(\Theta)) = Rp_{2*}(p_1^*\mathcal{I}_X(\Theta) \otimes\mathcal{P}).

By Proposition 5.1, we have C∣A∖V(Xred)=0C|_{A \setminus V(X_{\mathrm{red}})} = 0.

We now proceed to some Riemann–Roch calculations. Recall that there is an induced Fourier–Mukai transform at the level of cohomology with rational coefficients, defined on H∙(A,Q)H^\bullet(A,\mathbb{Q}) by

F(α)=p2∗(p1∗α∪eℓ).\mathcal{F}(\alpha) = p_{2*}(p_1^*\alpha\cup e^\ell).

Using the fact that θg=g!\theta^g = g!, we have

F(eθ)=p2∗(eδ∗θ−p2∗θ)=e−θ.\mathcal{F}(e^\theta) = p_{2*}(e^{\delta^*\theta-p_2^*\theta}) = e^{-\theta}.

Using the triviality of the tangent bundle of AA, as well as the standard exact sequence involving IX\mathcal{I}_X and OX\mathcal{O}_X, Grothendieck–Riemann–Roch implies

ch⁡(C)=e−θ−F(ch⁡(OX(Θ))).\operatorname{ch}(C) = e^{-\theta} - \mathcal{F}(\operatorname{ch}(\mathcal{O}_X(\Theta))).

Now since the support of OX(Θ)\mathcal{O}_X(\Theta) has dimension dd, we have ch⁡r(OX(Θ))=0\operatorname{ch}_r(\mathcal{O}_X(\Theta)) = 0 for r<g−dr < g-d. Moreover, since the bidegree of ℓ\ell is (1,1)(1,1) and the reverse Fourier–Mukai transform is the inverse of F\mathcal{F} up to sign and the standard involution, for all jj we have

F(H2j(A,Q))=H2(g−j)(A,Q).\mathcal{F}(H^{2j}(A,\mathbb{Q})) = H^{2(g-j)}(A,\mathbb{Q}).

Thus for every j>dj > d we deduce that

ch⁡j(C)=(−1)jθjj!.\operatorname{ch}_j(C) = (-1)^j \frac{\theta^j}{j!}.

and in particular ch⁡d+1(C)=(−1)d+1θd+1(d+1)!\operatorname{ch}_{d+1}(C) = (-1)^{d+1} \frac{\theta^{d+1}}{(d+1)!}.

Since C∣A∖V(Xred)=0C|_{A \setminus V(X_{\mathrm{red}})} = 0, the Chern character ch⁡(C)\operatorname{ch}(C) in the Chow ring of AA restricts to 00 on this open set, and in particular ch⁡d+1(C)\operatorname{ch}_{d+1}(C) is represented by a rational cycle class with support in V(Xred)V(X_{\mathrm{red}}). Passing to cohomology, we obtain an expression

θd+1(d+1)!=∑jrj[Zj],\frac{\theta^{d+1}}{(d+1)!} = \sum_j r_j[Z_j],

where rj∈Qr_j \in\mathbb{Q} and each Zj⊆V(Xred)Z_j \subseteq V(X_{\mathrm{red}}) is a closed subvariety of dimension g−d−1g-d-1; here [W][W] denotes the cohomology class of a closed subvariety WW.

Now fix a dd-dimensional component Y⊆XredY \subseteq X_{\mathrm{red}} as in the hypothesis, and set

γ=δ∗(p1∗[Y]∪p2∗θd+1(d+1)!)=∑jrjδ∗[Y×Zj]∈H2(A,Q).\gamma=\delta_*\left(p_1^*[Y]\cup p_2^*\frac{\theta^{d+1}}{(d+1)!}\right)=\sum_j r_j\delta_*[Y\times Z_j]\in H^2(A,\mathbb{Q}).

For the Pontrjagin product u∗v=m∗(p1∗u∪p2∗v)u*v=m_*(p_1^*u\cup p_2^*v) of cohomology classes, we have the standard identity

θd+1(d+1)!∗θg−1(g−1)!=(g−d)θdd!.\frac{\theta^{d+1}}{(d+1)!}*\frac{\theta^{g-1}}{(g-1)!}=(g-d)\frac{\theta^d}{d!}.

See for instance [3 Corollary 16.5.8]. The automorphism f:A×A→A×Af:A\times A\to A\times A defined by (a,b)↦(b,a−b)(a,b)\mapsto(b,a-b) satisfies m∘f=p1m\circ f=p_1, p1∘f=p2p_1\circ f=p_2, and p2∘f=δp_2\circ f=\delta. Thus we deduce that

∫Aγ∪θg−1(g−1)!=∫A×Ap1∗[Y]∪p2∗θd+1(d+1)!∪δ∗θg−1(g−1)!=∫A×Am∗[Y]∪p1∗θd+1(d+1)!∪p2∗θg−1(g−1)!=∫A[Y]∪(θd+1(d+1)!∗θg−1(g−1)!)=g−dd!∫Yθd>0.\begin{aligned} \int_A\gamma\cup\frac{\theta^{g-1}}{(g-1)!} &=\int_{A\times A}p_1^*[Y]\cup p_2^*\frac{\theta^{d+1}}{(d+1)!}\cup\delta^*\frac{\theta^{g-1}}{(g-1)!}\\ &=\int_{A\times A}m^*[Y]\cup p_1^*\frac{\theta^{d+1}}{(d+1)!}\cup p_2^*\frac{\theta^{g-1}}{(g-1)!}\\ &=\int_A[Y]\cup\left(\frac{\theta^{d+1}}{(d+1)!}*\frac{\theta^{g-1}}{(g-1)!}\right)=\frac{g-d}{d!}\int_Y\theta^d>0. \end{aligned}

Thus we deduce that δ∗[Y×Zj]≠0\delta_*[Y\times Z_j]\ne0 for some jj and therefore

dim⁡(Y−Zj)=dim⁡(Y×Zj)=g−1.\dim(Y-Z_j)=\dim(Y\times Z_j)=g-1.

On the other hand,

Y−Zj⊆Xred−V(Xred)⊆Θ.Y-Z_j\subseteq X_{\mathrm{red}}-V(X_{\mathrm{red}})\subseteq\Theta.

Since the theta divisor of an indecomposable ppav is integral, we deduce that Y−Zj=ΘY-Z_j=\Theta. Taking Z=ZjZ=Z_j concludes the proof. □

Proof of Theorem A

We assume that XX is a proper closed subscheme of an indecomposable ppav (A,Θ)(A,\Theta) such that IX(2Θ)\mathcal{I}_X(2\Theta) satisfies IT0IT_0. By Proposition 6.1 we know that XX is geometrically nondegenerate. Then by Corollary 5.3 we deduce that IX(Θ)\mathcal{I}_X(\Theta) is a GV-sheaf. As XX is a geometrically nondegenerate closed GV-subscheme of AA, [28 Theorem 2(1)] implies that XX is integral.

Remark 7.1. Note that once it is shown that XX is a GV-scheme, and therefore automatically geometrically nondegenerate by Theorem A, then various other nice properties of XX follow. We have already seen that XX is integral by [28]. It is also known that XX is normal, and has even rational singularities when dim⁡X=2\dim X=2, by [13]. Further generation properties relying only on the GV-condition were studied in [15].

Singularities of theta divisors

Multiplicities of points, and multiplicity loci

Let (A,Θ)(A,\Theta) be a complex ppav. We will make use of the Hodge ideal I1(Θ)\mathcal{I}_1(\Theta), see § 3.4. From here on, write

I1(Θ)=IZ,\mathcal{I}_1(\Theta)=\mathcal{I}_Z,

where ZZ is the closed subscheme of AA determined by this ideal.

We start with the following result towards Conjecture C, which is independent of the generic vanishing conjecture or Theorem A. Note first that when Θ\Theta is irreducible, it has rational singularities by [10 Theorem 1], and in particular it is normal. Therefore we have

dim⁡Z≤dim⁡Sing⁡(Θ)≤dim⁡Θ−2=g−3.\dim Z\leq\dim\operatorname{Sing}(\Theta)\leq\dim\Theta-2=g-3.

For this reason, Conjecture C is immediate for m=2m=2.

Theorem 8.1. If (A,Θ)(A,\Theta) is an indecomposable ppav of dimension gg, then for all m≥2m\geq2 we have

dim⁡Σm(Θ)≤max⁡{g−2m+1,dim⁡Sing⁡(Z)},\dim\Sigma_m(\Theta)\leq\max\{g-2m+1,\dim\operatorname{Sing}(Z)\},

with the convention that dim⁡∅=−∞\dim\varnothing=-\infty.5

Proof. For m=2m=2 the result is clear by the discussion above, so we may assume m≥3m\geq3. Recall that we consider Σm(Θ)\Sigma_m(\Theta) with its reduced scheme structure, and let WW be an irreducible component. Denote

r=codim⁡AWande=mult⁡W(Θ).r=\operatorname{codim}_A W \quad\text{and}\quad e=\operatorname{mult}_W(\Theta).

In particular we have e≥m≥3e\geq m\geq3. If r≥2m−1r\geq2m-1, or equivalently dim⁡W≤g−2m+1\dim W\leq g-2m+1, we are in good shape already. So we may assume r≤2m−2r\leq2m-2.

With this data, we apply Theorem 4.1(1) to obtain

IZ⊆IW(q),for q=min⁡{e−1,2e−r},\mathcal{I}_Z\subseteq\mathcal{I}_W^{(q)},\qquad\text{for }q=\min\{e-1,2e-r\},

where IW(q)\mathcal{I}_W^{(q)} is the qq-th symbolic power of IW\mathcal{I}_W. Given our numerical assumptions, we have q≥2q\geq2, so that at the very least

IZ⊆IW(2).\mathcal{I}_Z\subseteq\mathcal{I}_W^{(2)}.

Let x∈Wx\in W be a general point, at which WW is smooth. Since the symbolic and usual powers agree at smooth points, we have

IZ,x⊆IW,x2⊆mx2,\mathcal{I}_{Z,x}\subseteq\mathcal{I}_{W,x}^2\subseteq\mathfrak{m}_x^2,

where the latter is the maximal ideal of xx in AA. This implies that xx is a singular point of ZZ, and therefore W⊆Sing⁡(Z)W\subseteq\operatorname{Sing}(Z). □

Studying the case dim⁡Z=0\dim Z=0 relies on Corollary 5.4, which says that ZZ is a reduced point. Next we begin to make use of Theorem A, if dim⁡Z≥1\dim Z\geq1. The sheaf IZ(2Θ)\mathcal{I}_Z(2\Theta) satisfies IT0IT_0 by Theorem 4.1(2), and consequently ZZ is a GV-scheme by Theorem A. It follows that ZZ is itself normal by [13 Lemma 7.8], and so overall

dim⁡Sing⁡(Z)≤dim⁡Z−2≤g−5.\dim\operatorname{Sing}(Z)\leq\dim Z-2\leq g-5.

Putting everything together, we get a conclusion that includes Theorem F.

Corollary 8.2. Conjecture C(2)C(2) holds in arbitrary dimension for

m≤g+1−dim⁡Sing⁡(Z)2m \le\frac{g+1-\dim\operatorname{Sing}(Z)}{2}

Hence it always holds for m≤3m \le3. It holds for all mm if ZZ is smooth or empty.

The condition Z=∅Z=\varnothing is equivalent to having I1(Θ)=OA\mathcal{I}_1(\Theta)=\mathcal{O}_A, which is in turn equivalent to the condition α~(Θ)≥2\widetilde{\alpha}(\Theta)\ge2 for the minimal exponent of Θ\Theta by [27]. This is known as the 1-Du Bois condition; equivalently, the pair (A,Θ)(A,\Theta) is 1-log canonical in the terminology of [19]. This is stronger than rationality; for instance, in the case of Jacobians it holds for Brill–Noether general curves, by [29 Theorem 5.1].

We next prove the main result in this direction.

Proof of Theorem D. We need to show that the generic vanishing Conjecture B implies the strong multiplicity Conjecture C.

If ZZ is smooth or empty, the conjecture holds by Corollary 8.2. In particular this applies when dim⁡Z=0\dim Z=0, as we may apply Corollary 5.4 to deduce that ZZ is smooth. Assume now that dim⁡Z>0\dim Z>0. Then by Theorem A we have that IZ(Θ)\mathcal{I}_Z(\Theta) is a GVGV-sheaf, and ZZ is integral and geometrically nondegenerate. Conjecture B then leads to two possibilities.

On one hand, (A,Θ)(A,\Theta) could be the intermediate Jacobian of a smooth cubic threefold; this however cannot be the case, since its theta divisor has only one singular point, contradicting the assumption on dim⁡Z\dim Z. On the other hand, (A,Θ)(A,\Theta) could be the Jacobian of a smooth curve of genus gg. Then the bound in Conjecture C(2)C(2) holds by Martens’ theorem. □

The following is a slightly stronger version of Corollary E.

Corollary 8.3. Conjecture CC holds unconditionally when:

  1. g=dim⁡A≤6g=\dim A\le6

  2. More generally, when dim⁡Z≤2\dim Z\le2 or dim⁡Z≥g−3\dim Z\ge g-3, in any dimension gg; this holds in particular when dim⁡Sing⁡(Θ)≤2\dim\operatorname{Sing}(\Theta)\le2 or dim⁡Sing⁡(Θ)≥g−3\dim\operatorname{Sing}(\Theta)\ge g-3.

Proof. It suffices to show (2). The case dim⁡Z=0\dim Z=0 is dealt with by Corollary 5.4. Next note that the case dim⁡Z≤2\dim Z\le2 is technically equivalent to the case dim⁡Z≥g−3\dim Z\ge g-3. This is because by theta-duality, see [23 Theorem 5.2], if ZZ is a GVGV-scheme of dimension dd, its theta-dual WW is a GVGV-scheme of dimension g−d−1g-d-1, and the argument in the proof of Theorem D works for either.

Hence all we need to know is that the GVGV-conjecture holds for all positive dimensional closed subschemes XX of dimension at most two. If dim⁡X=1\dim X=1, this is known by [23 Theorem C]. If dim⁡X=2\dim X=2, it is known if g≤5g\le5 by [6 Theorem A], and if g≥6g\ge6 by [7 Corollary 1.3].

Finally, if dim⁡Sing⁡(Θ)≤2\dim\operatorname{Sing}(\Theta)\le2, then clearly dim⁡Z≤2\dim Z\le2. On the other hand, if dim⁡Sing⁡(Θ)≥g−3\dim\operatorname{Sing}(\Theta)\ge g-3, then in fact dim⁡Sing⁡(Θ)=g−3\dim\operatorname{Sing}(\Theta)=g-3 since Θ\Theta is normal, in which case we also have dim⁡Z=g−3\dim Z=g-3 by Lemma 9.1 below. □

Remark 8.4. The first case when we do not have the multiplicity conjecture unconditionally is when g=7g=7 and dim⁡Z=dim⁡Sing⁡(Θ)=3\dim Z=\dim\operatorname{Sing}(\Theta)=3. This is the only case left to deal with in dimension 7, and the conjecture is that it should not be possible (as it is not hard to see that it cannot happen on Jacobians, with Z=W3Z=W_3 up to sign and translation).

Remark 8.5 (Equality case). This refers to the Remark right after Conjecture C in the Introduction, regarding the maximal multiplicity case

mult⁡x(Θ)=g+12.\operatorname{mult}_{x}(\Theta)=\frac{g+1}{2}.

On the Jacobian of a smooth curve it is well known that the multiplicity of each point on Θ\Theta is at most (g+1)/2(g+1)/2, by the Riemann Singularity Theorem combined with Clifford’s Theorem. Moreover, Clifford also tells us that the equality case happens only on hyperelliptic Jacobians, at a unique point (the multiple of the g21g^1_2) when gg is odd, and not at all when gg is even. It is also well known that on the intermediate Jacobian of a smooth cubic threefold the only singular point is 0∈Θ0 \in\Theta, which also satisfies the equality case mult⁡0(Θ)=3\operatorname{mult}_{0}(\Theta)=3. It it natural to conjecture that these are the only situations when equality holds.

Characterization of hyperelliptic Jacobians

The methods discussed above also lead to a proof of Beauville’s conjecture characterizing hyperelliptic Jacobians; see Theorem G and the preceding paragraph. We start with a technical result:

Lemma 9.1. If dim⁡Sing⁡(Θ)=g−3\dim\operatorname{Sing}(\Theta)=g-3, then dim⁡Z=g−3\dim Z=g-3. Moreover, Sing⁡(Θ)\operatorname{Sing}(\Theta) has a unique irreducible component of maximal dimension g−3g-3, equal to ZZ.

Proof. Clearly we have Z⊂Sing⁡(Θ)Z \subset\operatorname{Sing}(\Theta). By Theorem A, we also know that ZZ is integral. It suffices therefore to show that if WW is a (g−3)(g-3)-dimensional component of Sing⁡(Θ)\operatorname{Sing}(\Theta), then W⊂ZW \subset Z.

Denote by mm the generic multiplicity of Θ\Theta along WW. Since the pair (A,Θ)(A,\Theta) is log canonical (see § 3.4), it is standard that, for any k≥2k \ge2, each irreducible component of Σk(Θ)\Sigma_k(\Theta) has codimension at least kk in AA; moreover, by [10 Corollary 2], the existence of a component of codimension precisely kk implies that Θ\Theta is reducible. Therefore in our case we have the inequality

g−3=dim⁡W≤g−m−1,g-3=\dim W\le g-m-1,

and therefore m=2m=2. We now use the relationship between Hodge ideals and multiplicity loci in Theorem 4.1(1); in our case, with codimension r=3r=3 and multiplicity m=2m=2, it implies that

I1(Θ)⊆IW,I_1(\Theta)\subseteq I_W,

which is precisely what we want. □

Proof of Theorem G. It is a standard fact that if Θ\Theta is reducible, then dim⁡Sing⁡(Θ)=g−2\dim\operatorname{Sing}(\Theta)=g-2. Therefore under our assumption Θ\Theta is irreducible.

As discussed in the Introduction, Beauville [2] proved this result for g≤5g\le5, hence we may assume g≥6g\ge6. By Lemma 9.1, we see that if ZZ is the zero-scheme of the Hodge ideal I1(Θ)I_1(\Theta), then ZZ itself has dimension g−3g-3. Moreover, as before Theorem A implies that IZ(Θ)I_Z(\Theta) is a GVGV-sheaf. Using the main result of [7], the GVGV-conjecture holds for ZZ, and therefore (A,Θ)(A,\Theta) is a Jacobian. But on the Jacobian of a curve CC it is well known that dim⁡Sing⁡(Θ)=g−3\dim\operatorname{Sing}(\Theta)=g-3 if and only if CC is hyperelliptic. □

Centers of minimal exponent

Another closed subvariety that is amenable to the same techniques as above is the center of minimal exponent YY of Θ\Theta, defined by Schnell-Yang [29 §4.2]. From the point of view of this paper, YY plays a role similar to that of the zero-scheme ZZ of the ideal I1(Θ)I_1(\Theta), but it is not hard to check that in general they are different. This is also the place to note that some of the arguments in this paper can be equally applied to the zero-schemes of the higher multiplier ideals of loc.cit.⁡\operatorname{loc. cit.}; they are indeed considered in [12].

It is shown in [29 Theorem 1.12] that if dim⁡Y=1\dim Y = 1, then YY must be a hyperelliptic curve, and if moreover there exists a component of the maximal allowed dimension g−2m+1g - 2m + 1 in some Σm(Θ)\Sigma_m(\Theta), then AA is the Jacobian of YY. Please see the interesting discussion in §1.6 of loc. cit. for the significance of such a statement. Using the discussion in the paragraph above, we have:

Corollary 10.1. Assume that dim⁡Y≥1\dim Y \ge1, and the minimal exponent of Θ\Theta satisfies 1<α~(Θ)<21 < \widetilde{\alpha}(\Theta) < 2. If Conjecture B holds, then (A,Θ)(A,\Theta) is the Jacobian of a hyperelliptic curve.6

In particular, this holds unconditionally if 1≤dim⁡Y≤21 \le\dim Y \le2, or if dim⁡Y=g−3\dim Y = g - 3.

Proof. In the proof of [29 Lemma 5.5] it is shown that under this minimal exponent hypothesis IY(2Θ)\mathcal{I}_Y(2\Theta) satisfies IT0IT_0, so the methods we developed here apply: by Theorem A, IY(Θ)\mathcal{I}_Y(\Theta) is GVGV, and YY is a geometrically nondegenerate (integral, normal) closed subvariety of AA.

We obviously have Y⊆Sing⁡(Θ)Y \subseteq\operatorname{Sing}(\Theta), and therefore 1≤dim⁡Y≤g−31 \le\dim Y \le g - 3. Assuming Conjecture B, we deduce that since YY is positive dimensional, (A,Θ)(A,\Theta) must be the Jacobian of a curve CC. The conjecture also says that YY must be ++ or −- a translate of Wd(C)W_d(C) where d=dim⁡Yd = \dim Y. But we may also assume that Y=−YY = -Y, by choosing Θ\Theta to be symmetric, as the construction of the center of minimal exponent is canonical. It is however well known that if this property holds for some Wd(C)W_d(C), then CC must be hyperelliptic; see also [8 Lemma 5.5] for a more general statement.

The unconditional statement refers to the cases of the GVGV-conjecture we know to hold, as everywhere else in the paper. □

References

Department of Mathematics, University of California San Diego, 9500 Gilman Drive # 0112, La Jolla, 92093-0112, USA

Email address: [email protected]

Department of Mathematics, Harvard University, 1 Oxford Street, Cambridge, MA 02138, USA

Email address: [email protected]

References

  1. [1][ACP] E. Arbarello, G. Codogni, G. Pareschi, Characterizing Jacobians via the KP equation and via flexes and degenerate trisecants to the Kummer variety: an algebro-geometric approach J. reine angew. Math. 777 (2021), 251–271.arxiv.org/abs/2009.14324
  2. [2][Be] A. Beauville, Prym varieties and the Schottky problem, Invent. Math. 41 (1977), 149–196.DOI
  3. [3][BL] Ch. Birkenhake and H. Lange, Complex abelian varieties, Second, Grundlehren der mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 302, Springer-Verlag, Berlin, 2004.
  4. [4][Ca1] S. Casalaina-Martin, Singularities of theta divisors in algebraic geometry, In Curves and abelian varieties, Contemp. Math. 465, 25–43. Amer. Math. Soc., Providence, RI, 2008.
  5. [5][Ca2] S. Casalaina-Martin, Singularities of the Prym theta divisor, Ann. of Math. (2) 170 (2009), no.1, 162–204.
  6. [6][CMPS] S. Casalaina-Martin, M. Popa, and S. Schieder, Generic vanishing and minimal cohomology classes on abelian fivefolds, J. Algebraic Geom. 27 (2018), 553–581.
  7. [7][Ch] Y. Chen, Generic vanishing subschemes of codimension three, preprint arXiv:2609.15391.
  8. [8][De1] O. Debarre, Minimal cohomology classes and Jacobians, J. Algebraic Geom. 4 (1995), 321–335.
  9. [9][De2] O. Debarre, Update on: Singularities of divisors on abelian varieties, lecture notes at https://perso.imj-prg.fr/olivier-debarre/exposes-talks/.
  10. [10][EL] L. Ein and R. Lazarsfeld, Singularities of theta divisors and the birational geometry of irregular varieties, J. Amer. Math. Soc. 10 (1997), no.1, 243–258.arxiv.org/abs/alg-geom/9603017
  11. [11][Gr] S. Grushevsky, The Schottky problem, In Current developments in algebraic geometry, Math. Sci. Res. Inst. Publ. 59, 129–164. Cambridge Univ. Press, Cambridge, 2012.arxiv.org/abs/1009.0369
  12. [12][HJ] S. Huang and Z. Jiang, GV-subschemes and hyperelliptic Jacobians, preprint.
  13. [13][JP] Z. Jiang and G. Pareschi, Cohomological rank functions on abelian varieties, Ann. Sci. Éc. Norm. Supér. (4) 53 (2020), no. 4, 815–846.arxiv.org/abs/1707.05888
  14. [14][Ko] J. Kollár, Shafarevich maps and automorphic forms, M.B. Porter Lectures, Princeton Univ. Press, 1995.DOI
  15. [15][LT] L. Lombard and S. Tirabassi, GV-subschemes and their embeddings in principally polarized abelian varieties, Math. Nachr. 288 (2015), no. 11-12, 1405–1412.
  16. [16][Me] F. Meng, Generic vanishing conjecture for divisors, preprint arXiv:2609.11377.
  17. [17][Muk] S. Mukai, Duality between \(D(X)\) and \(D(\widehat{X})\) with its application to Picard sheaves, Nagoya Math. J. 81 (1981), 153–175.
  18. [18][Mu] D. Mumford, Abelian varieties, Tata Institute of Fundamental Research Studies in Mathematics, vol. 5, Tata Institute of Fundamental Research, Bombay; by Hindustan Book Agency, New Delhi, 2008. With appendices by C. P. Ramanujam and Yuri Manin, Corrected reprint of the second (1974) edition.
  19. [19][MP] M. Mustaţă and M. Popa, Hodge ideals, Mem. Amer. Math. Soc. 262 (2019), no.1268.
  20. [20][Pa] G. Pareschi, Singularities of base loci on abelian varieties, Atti Accad. Naz. Lincei Rend. Lincei Mat. Appl. 36 (2025), no. 2, 365–375.
  21. [21][PP1] G. Pareschi and M. Popa, Regularity on abelian varieties I, J. Amer. Math. Soc. 16 (2003), 285–302.DOI
  22. [22][PP2] G. Pareschi and M. Popa, Regularity on abelian varieties II: Basic results on linear series and defining equations, J. Algebraic Geom. 13 (2004), no. 1, 167–193.DOI
  23. [23][PP3] G. Pareschi and M. Popa, Generic vanishing and minimal cohomology classes on abelian varieties, Math. Ann. 340 (2008), 209–222.arxiv.org/abs/math/0610166
  24. [24][PP4] G. Pareschi and M. Popa, Regularity on abelian varieties III: relationship with Generic Vanishing and applications, in Grassmannians, Moduli Spaces and Vector Bundles 2008, Clay Mathematics Proceedings 14, Amer. Math. Soc., Providence RI, 2011, 141–167.
  25. [25][Po] M. Popa, Connections between some conjectures on subvarieties of abelian varieties, Oberwolfach Report 33/2018, 1993–1995.
  26. [26][Ra] Z. Ran, On subvarieties of abelian varieties, Invent. Math. 62 (1981), 459–479.DOI
  27. [27][Sa] M. Saito, Hodge ideals and microlocal \(V\)-filtration, preprint arXiv:1612.08667.arxiv.org/abs/1612.08667
  28. [28][Sch] S. Schreieder, Decomposable theta divisors and generic vanishing, Int. Math. Res. Not. IMRN 16 (2017), 4984–5009.DOI
  29. [29][SY] C. Schnell and R. Yang, Higher multiplier ideals, J. reine angew. Math. 832 (2026), 153–197.arxiv.org/abs/2309.16763

Paper details

vlnc:W_jwzzkvtq
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 9

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

AI

View statement
Looking for related papers…
Contents