This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

\newaliascnt

propthm \newaliascntcorthm \newaliascntlemthm \newaliascntclaimthm \newaliascntdefnthm \newaliascntquesthm \newaliascntconjthm \newaliascntfactthm \newaliascntremthm \newaliascntexthm \aliascntresettheprop \aliascntresetthecor \aliascntresetthelem \aliascntresettheclaim \aliascntresetthedefn \aliascntresettheques \aliascntresettheconj \aliascntresetthefact \aliascntresettherem \aliascntresettheex

Higher syzygies on general polarized abelian varieties of type (1,,1,d)(1,\dots,1,d)

Atsushi Ito Graduate School of Mathematics, Nagoya University, Nagoya, Japan [email protected]
Abstract.

In this paper, we show that a general polarized abelian variety (X,L)(X,L) of type (1,,1,d)(1,\dots,1,d) and dimension gg satisfies property (Np)(N_{p}) if di=0g(p+2)id\geqslant\sum_{i=0}^{g}(p+2)^{i}. In particular, the case p=0p=0 affirmatively solves a conjecture by L. Fuentes García on projective normality.

Key words and phrases:
Syzygy, Abelian variety, Basepoint-freeness threshold
2010 Mathematics Subject Classification:
14C20,14K99

1. Introduction

Throughout this paper, we work over the complex number field \mathbb{C}.

For an ample line bundle LL on an abelian variety XX of dimension gg, we can associate a sequence of positive integers (d1,,dg)(d_{1},\dots,d_{g}) with d1|d2||dgd_{1}|d_{2}|\cdots|d_{g}, called the type of LL. It is well known that LL is basepoint free if d12d_{1}\geqslant 2 and projectively normal if d13d_{1}\geqslant 3. On the other hand, in the case d1=1d_{1}=1, equivalently the case when LL is not written as some multiple of another line bundle, basepoint freeness or projective normality of LL is more subtle. In [DHS94], the authors investigate general LL of type (1,,1,d)(1,\dots,1,d) and prove the following theorem.

Theorem 1.1 ([DHS94, Proposition 2, Proposition 6, Corollary 25]).

Let (X,L)(X,L) be a general polarized abelian variety of type (1,,1,d)(1,\dots,1,d) and dimension gg. Then

  1. (1)

    LL is base point free if and only if dg+1d\geqslant g+1.

  2. (2)

    The morphism defined by |L||L| is birational onto the image if and only if dg+2d\geqslant g+2.

  3. (3)

    LL is very ample if d>2gd>2^{g}.

On the other hand, L. Fuentes García investigates projective normality based on the work [Iye03] of J. N. Iyer, and conjectures the following:

Conjecture \theconj ([FG05, Conjecture 4.7]).

Let (X,L)(X,L) be a general polarized abelian variety of type (1,,1,d)(1,\dots,1,d) and dimension gg. Then LL is projectively normal if d2g+11d\geqslant 2^{g+1}-1.

For g=2g=2, this conjecture follows from [Laz90], [FG04] or [Iye99]. Fuentes García proves this conjecture for g=3,4g=3,4 using results in [Iye03] and some calculations of the ranks of suitable matrices using computer. We note that d=h0(L)2g+11d=h^{0}(L)\geqslant 2^{g+1}-1 is a necessary condition for the projective normality of LL since dimSym2H0(L)dimH0(L2)\dim\operatorname{Sym}^{2}H^{0}(L)\geqslant\dim H^{0}(L^{\otimes 2}) must hold for such LL. Hence § 1 states that it is a sufficient condition as well for general (X,L)(X,L).

In this paper, we prove § 1 affirmatively. In fact, we prove not only projective normality but also property (Np)(N_{p}) as follows:

Theorem 1.2.

Let p1p\geqslant-1 be an integer and let (X,L)(X,L) be a general polarized abelian variety of type (1,,1,d)(1,\dots,1,d) and dimension gg. Then LL satisfies property (Np)(N_{p}) if

di=0g(p+2)i={g+1 if p=1((p+2)g+11)/(p+1) if p0.d\geqslant\sum_{i=0}^{g}(p+2)^{i}=\begin{cases}g+1&\text{ if }\ p=-1\\ ((p+2)^{g+1}-1)/(p+1)&\text{ if }\ p\geqslant 0.\end{cases}

In particular, LL is projectively normal if and only if d2g+11d\geqslant 2^{g+1}-1. In this case, the homogeneous ideal of XX embedded by |L||L| is generated by quadrics and cubics.

We refer the readers to [Laz04, Chapter 1.8.D], [Eis05] for the definition of property (Np)(N_{p}). We just note here that (Np)(N_{p})’s consist an increasing sequence of positivity properties. For example, (N0N_{0}) holds for LL if and only if LL defines a projectively normal embedding, and (N1N_{1}) holds if and only if (N0N_{0}) holds and the homogeneous ideal of the embedding is generated by quadrics. Usually (Np)(N_{p}) is considered for p0p\geqslant 0, but we add the basepoint freeness in the sequence of positivity properties as (N1)(N_{-1}), as in [Loz18], [Jia21]. Hence the case p=1p=-1 of Theorem 1.2 recovers Theorem 1.1 (1) and the case p=0p=0 of Theorem 1.2 proves § 1 affirmatively.

For an abelian surface XX, it is known that a very ample line bundle LL of type (1,d)(1,d) with d7d\geqslant 7 is projectively normal and the homogeneous ideal of XX embedded by |L||L| is generated by quadrics and cubics by [Laz90], [FG04], [Ago17]. The last statement of Theorem 1.2 generalizes this result to higher dimensions at least for general (X,L)(X,L). Furthermore, Theorem 1.2 gives a better bound of dd than the bounds obtained by [KL19], [Ito18] in dimension two and [Iye03], [LPP11], [Ito20], [Jia21] in higher dimensions. See Remarks 4, 5 for details.

In the rest of Introduction, we explain the idea of the proof. In [JP20], Z. Jiang and G. Pareschi introduce an invariant β(l)=β(X,l)(0,1]\beta(l)=\beta(X,l)\in(0,1], called the basepoint-freeness threshold, for the class lPicX/Pic0Xl\in\operatorname{Pic}X/\operatorname{Pic}^{0}X of an ample line bundle LL. By [JP20], [Cau20a], a suitable upper bound of β(l)\beta(l) implies property (Np)(N_{p}) as follows:

Theorem 1.3 ([JP20, Theorem D, Corollary E],[Cau20a, Theorem 1.1]).

Let p1p\geqslant-1 and let (X,L)(X,L) be a polarized abelian variety. Then LL satisfies (Np)(N_{p}) if β(l)<1/(p+2)\beta(l)<1/(p+2).

Hence the following theorem implies Theorem 1.2.

Theorem 1.4.

Let d,g1d,g\geqslant 1 be integers and set m:=dgm:=\lfloor\sqrt[g]{d}\rfloor. Let (X,l)(X,l) be a general polarized abelian variety of type (1,,1,d)(1,\dots,1,d) and dimension gg. Then

  1. (1)

    1/dgβ(l)1/m1/\sqrt[g]{d}\leqslant\beta(l)\leqslant 1/m holds.

  2. (2)

    1/dgβ(l)<1/m1/\sqrt[g]{d}\leqslant\beta(l)<1/m holds if dmg++m+1=(mg+11)/(m1)d\geqslant m^{g}+\cdots+m+1=(m^{g+1}-1)/(m-1).

Remark \therem.

Upper bounds of β(l)\beta(l) imply not only properties (Np)(N_{p}) but also jet ampleness and vanishings of suitable Koszul cohomologies: [Cau20b, Theorem D], [Ito20, Proposition 2.5] and Theorem 1.4 imply that LL is p+1p+1-jet ample and the Koszul cohomology group Kp,q(X,L;kL)=0K_{p,q}(X,L;kL)=0 for any q,k1q,k\geqslant 1 if (X,L)(X,L) is a general polarized abelian variety of type (1,,1,d)(1,\dots,1,d) and dimension gg with di=0g(p+2)id\geqslant\sum_{i=0}^{g}(p+2)^{i}.

In [Ito20], the author observes a similarity between β(l)1\beta(l)^{-1} and Seshadri constants. Since Seshadri constants are lower-semicontinuous, it is natural to ask whether β(l)\beta(l) is upper-semicontinuous or not. In fact, this is the case as we see in § 3. Hence, Theorem 1.4 is reduced to finding an example (X0,L0)(X_{0},L_{0}) of type (1,,1,d)(1,\dots,1,d) such that β(l0)1/m\beta(l_{0})\leqslant 1/m or <1/m<1/m.

Theorem 1.1 (3) is proved in [DHS94] by degenerating polarized abelian varieties to a polarized variety (X0,L0)(X^{\prime}_{0},L^{\prime}_{0}) whose normalization is a (1)g1(\mathbb{P}^{1})^{g-1}-bundle over an elliptic curve and showing that L0L^{\prime}_{0} is very ample. Contrary to very ampleness, β(l)\beta(l) is defined only for abelian varieties. Hence we do not use such degenerations but find (X0,L0)(X_{0},L_{0}) as a polarized abelian variety. In fact, we construct such (X0,L0)(X_{0},L_{0}) as a suitable polarization on a product of elliptic curves.

We note that Theorem 1.3 is also used to show (Np)(N_{p}) in [Ito20], [Jia21], where techniques to cutting minimal log canonical centers are used to bound β(l)\beta(l) from above. In this paper, we do not need such techniques.

This paper is organized as follows. In § 2, we recall some notation. In § 3, we show the upper-semicontinuity of β(l)\beta(l). In § 4, we study β(l)\beta(l) of polarized abelian surfaces and show Theorem 1.4 for g=2g=2. In § 5, we prove Theorems 1.2, 1.4 in any dimension. In Appendix, we compute β(l)\beta(l) of general polarized abelian surfaces (X,l)(X,l) of type (1,d)(1,d) for some dd.

Acknowledgments

The author would like to express his gratitude to Professor Zhi Jiang for sending drafts of [Jia21] to the author. He also thanks Professor Victor Lozovanu for valuable comments. The author was supported by JSPS KAKENHI Grant Number 17K14162, 21K03201.

2. Preliminaries

Let XX be an abelian variety of dimension gg. We denote the origin of XX by oXo_{X} or oXo\in X. For bb\in\mathbb{Z}, we denote the multiplication-by-bb isogeny by

μb=μbX:XX,pbp.\mu_{b}=\mu^{X}_{b}:X\rightarrow X,\quad p\mapsto bp.

For an ample line bundle LL on XX, we call (X,L)(X,L) or (X,l)(X,l) a polarized abelian variety, where lNS(X)=Pic(X)/Pic0(X)l\in\mathrm{NS}(X)=\operatorname{Pic}(X)/\operatorname{Pic}^{0}(X) is the class of LL in the Neron-Severi group of XX. Let

K(L)={pX|tpLL},K(L)=\{p\in X\,|\,t_{p}^{*}L\simeq L\},

where tp:XXt_{p}:X\rightarrow X is the translation by pp on XX. It is known that there exist positive integers d1|d2||dgd_{1}|d_{2}|\cdots|d_{g} such that K(L)(i=1g/di)2K(L)\simeq(\bigoplus_{i=1}^{g}\mathbb{Z}/d_{i}\mathbb{Z})^{\oplus 2} as abelian groups. The vector (d1,,dg)(d_{1},\dots,d_{g}) is called the type of LL. Since K(L)K(L) depends only on the class ll of LL, (d1,,dg)(d_{1},\dots,d_{g}) is called the type of ll as well. It is known that χ(l)=i=1gdi\chi(l)=\prod_{i=1}^{g}d_{i} holds.

For a coherent sheaf \mathcal{F} on XX and xx\in\mathbb{Q}, a \mathbb{Q}-twisted coherent sheaf xl\mathcal{F}\langle xl\rangle is the equivalence class of the pair (,xl)(\mathcal{F},xl), where the equivalence is defined by

(Lm,xl)(,(x+m)l)(\mathcal{F}\otimes L^{m},xl)\sim(\mathcal{F},(x+m)l)

for any line bundle LL representing ll and any mm\in\mathbb{Z}.

Recall some notions of generic vanishing: a coherent sheaf \mathcal{F} on XX is said to be IT(0) if hi(X,Pα)=0h^{i}(X,\mathcal{F}\otimes P_{\alpha})=0 for any i>0i>0 and any αX^=Pic0(X)\alpha\in\widehat{X}=\operatorname{Pic}^{0}(X), where PαP_{\alpha} is the algebraically trivial line bundle on XX corresponding to α\alpha. It is said to be GV if codimX^{αX^|hi(X,Pα)>0}i\operatorname{codim}_{\widehat{X}}\{\alpha\in\widehat{X}\,|\,h^{i}(X,\mathcal{F}\otimes P_{\alpha})>0\}\geqslant i for any i>0i>0.

In [JP20], such notions are extended to the \mathbb{Q}-twisted setting. A \mathbb{Q}-twisted coherent sheaf xl\mathcal{F}\langle xl\rangle for x=a/bx=a/b with b>0b>0 is said to be IT(0) or GV if so is μbLab\mu_{b}^{*}\mathcal{F}\otimes L^{ab}. We note that this definition does not depend on the representation x=a/bx=a/b nor the choice of LL representing ll. By [JP20, Theorem 5.2], xl\mathcal{F}\langle xl\rangle is GV if and only if (x+x)l\mathcal{F}\langle(x+x^{\prime})l\rangle is IT(0) for any rational number x>0x^{\prime}>0.

In [JP20], an invariant 0<β(l)10<\beta(l)\leqslant 1 is introduced for a polarized abelian variety (X,l)(X,l). It is defined using cohomological rank functions, which are also defined in [JP20], but β(l)\beta(l) is characterized by the notion IT(0) as follows:

Lemma \thelem ([JP20, Section 8],[Cau20a, Lemma 3.3]).

Let (X,l)(X,l) be a polarized abelian variety and xx\in\mathbb{Q}. Then β(l)<x\beta(l)<x if and only if pxl\mathcal{I}_{p}\langle xl\rangle is IT(0) for some (and hence for any) pXp\in X, where p𝒪X\mathcal{I}_{p}\subset\mathcal{O}_{X} is the ideal sheaf corresponding to pXp\in X.

Remark \therem.

For a rational number x=a/b>0x=a/b>0, β(l)x\beta(l)\leqslant x if and only if pxl\mathcal{I}_{p}\langle xl\rangle is GV for some (and hence for any) pXp\in X since

β(l)x\displaystyle\beta(l)\leqslant x\ β(l)<x+x for any rational number x>0\displaystyle\Longleftrightarrow\ \beta(l)<x+x^{\prime}\text{ for any rational number }x^{\prime}>0
p(x+x)l is IT(0) for any rational number x>0\displaystyle\Longleftrightarrow\ \mathcal{I}_{p}\langle(x+x^{\prime})l\rangle\text{ is IT(0) for any rational number }x^{\prime}>0
pxl is GV.\displaystyle\Longleftrightarrow\ \mathcal{I}_{p}\langle xl\rangle\text{ is GV}.

Fix a representative LL of ll. By the exact sequence

0μbpLabPαLabPα(𝒪X/μbp)LabPα00\rightarrow\mu_{b}^{*}\mathcal{I}_{p}\otimes L^{ab}\otimes P_{\alpha}\rightarrow L^{ab}\otimes P_{\alpha}\rightarrow(\mathcal{O}_{X}/\mu_{b}^{*}\mathcal{I}_{p})\otimes L^{ab}\otimes P_{\alpha}\rightarrow 0

and hi(LabPα)=hi((𝒪X/μbp)LabPα)=0h^{i}(L^{ab}\otimes P_{\alpha})=h^{i}((\mathcal{O}_{X}/\mu_{b}^{*}\mathcal{I}_{p})\otimes L^{ab}\otimes P_{\alpha})=0 for i1i\geqslant 1, we have hi(μbpLabPα)=0h^{i}(\mu_{b}^{*}\mathcal{I}_{p}\otimes L^{ab}\otimes P_{\alpha})=0 for any αX^\alpha\in\widehat{X} and i2i\geqslant 2. Hence pxl\mathcal{I}_{p}\langle xl\rangle is GV if and only if h1(μbpLabPα)=0h^{1}(\mu_{b}^{*}\mathcal{I}_{p}\otimes L^{ab}\otimes P_{\alpha})=0 for some α\alpha. Equivalently, pxl\mathcal{I}_{p}\langle xl\rangle is GV if and only if h1(μbpLab)=0h^{1}(\mu_{b}^{*}\mathcal{I}_{p^{\prime}}\otimes L^{ab})=0 for some pXp^{\prime}\in X.

We use the following lemma to estimate β(l)\beta(l).

Lemma \thelem ([Ito20, Lemmas 3.4, 4.3]).

Let (X,l)(X,l) be a polarized abelian gg-fold. Then

  1. (i)

    β(l)1/χ(l)g\beta(l)\geqslant 1/\sqrt[g]{\chi(l)}.

  2. (ii)

    For an abelian subvariety ZXZ\subset X, it holds that β(l)β(l|Z)\beta(l)\geqslant\beta(l|_{Z}). Furthermore,

    β(l|Z)β(l)max{β(l|Z),g(lg1.Z)(lg)}=max{β(l|Z),χ(l|Z)χ(l)}\beta(l|_{Z})\leqslant\beta(l)\leqslant\max\left\{\beta(l|_{Z}),\frac{g(l^{g-1}.Z)}{(l^{g})}\right\}=\max\left\{\beta(l|_{Z}),\frac{\chi(l|_{Z})}{\chi(l)}\right\}

    holds if the codimension of ZXZ\subset X is one.

3. Semicontinuity of basepoint-freeness thresholds

In this section, we prove the upper-semicontinuity of β(l)\beta(l) as follows:

Theorem 3.1.

Let f:𝒳Tf:\mathcal{X}\rightarrow T be an abelian scheme over a variety TT and let \mathcal{L} be a line bundle on 𝒳\mathcal{X} which is ample over TT. Set Xt:=f1(t),Lt:=|XtX_{t}:=f^{-1}(t),L_{t}:=\mathcal{L}|_{X_{t}} for tTt\in T and let ltl_{t} be the class of LtL_{t}. Take a point 0T0\in T and a rational number xx such that β(X0,l0)x\beta(X_{0},l_{0})\leqslant x. Then β(Xt,lt)x\beta(X_{t},l_{t})\leqslant x holds for general tTt\in T.

In particular, the function T:tβ(Xt,lt)T\rightarrow\mathbb{R}:t\mapsto\beta(X_{t},l_{t}) is upper-semicontinuous in Zariski topology.

Proof.

Note that xx is positive since 0<β(X0,l0)x0<\beta(X_{0},l_{0})\leqslant x. By § 2, there exists p0X0p_{0}\in X_{0} such that h1(μbp0L0ab)=0h^{1}(\mu_{b}^{*}\mathcal{I}_{p_{0}}\otimes L_{0}^{ab})=0, where x=a/bx=a/b. By taking a suitable base change of 𝒳T\mathcal{X}\rightarrow T by a finite cover TTT^{\prime}\rightarrow T, we may assume that there exists a section σ:T𝒳\sigma:T\rightarrow\mathcal{X} such that σ(0)=p0\sigma(0)=p_{0}. Then h1(μbσ(t)Ltab)=0h^{1}(\mu_{b}^{*}\mathcal{I}_{\sigma(t)}\otimes L_{t}^{ab})=0 for general tTt\in T by the semicontinuity of cohomology. Hence β(Xt,lt)x\beta(X_{t},l_{t})\leqslant x holds for general tTt\in T by § 2 again.

By definition, the upper-semicontinuity of tβ(Xt,lt)t\mapsto\beta(X_{t},l_{t}) is equivalent to the openness of {tT|β(Xt,lt)<y}\{t\in T\,|\,\beta(X_{t},l_{t})<y\} in Zariski topology for any yy\in\mathbb{R}. We already show that {tT|β(Xt,lt)x}\{t\in T\,|\,\beta(X_{t},l_{t})\leqslant x\} is open for any xx\in\mathbb{Q}. Hence

{tT|β(Xt,lt)<y}=x,x<y{tT|β(Xt,lt)x}\{t\in T\,|\,\beta(X_{t},l_{t})<y\}=\bigcup_{x\in\mathbb{Q},x<y}\{t\in T\,|\,\beta(X_{t},l_{t})\leqslant x\}

is open as well. ∎

Remark \therem.

By Theorem 3.1, β(Xt,lt)β(X0,l0)\beta(X_{t},l_{t})\leqslant\beta(X_{0},l_{0}) holds for general tTt\in T if β(X0,l0)\beta(X_{0},l_{0}) is rational. If β(X0,l0)\beta(X_{0},l_{0}) is irrational, though we do not know such examples yet, we can only say that β(Xt,lt)β(X0,l0)\beta(X_{t},l_{t})\leqslant\beta(X_{0},l_{0}) holds for very general tTt\in T.

4. On general polarized abelian surfaces of type (1,d)(1,d)

Let k1k\geqslant 1 be an integer and take an isogeny f:EEf:E\rightarrow E^{\prime} between elliptic curves with kerf/k\ker f\simeq\mathbb{Z}/k\mathbb{Z}, e.g. we take f:/(+kτ)/(+τ)f:\mathbb{C}/(\mathbb{Z}+k\tau\mathbb{Z})\rightarrow\mathbb{C}/(\mathbb{Z}+\tau\mathbb{Z}) induced from +kτ+τ\mathbb{Z}+k\tau\mathbb{Z}\subset\mathbb{Z}+\tau\mathbb{Z}. For the dual isogeny f^:E=E^E^=E\hat{f}:E^{\prime}=\widehat{E^{\prime}}\rightarrow\widehat{E}=E, it is well-known that f^f=μkE\hat{f}\circ f=\mu^{E}_{k} and ff^=μkEf\circ\hat{f}=\mu^{E^{\prime}}_{k} hold (see e.g. [Sil09, Chapter III, Theorem 6.2]).

Set X=E×EX=E\times E^{\prime} and

F1={oE}×E,F2=E×{oE},Γ={(p,f(p))E×E|pE}.F_{1}=\{o_{E}\}\times E^{\prime},\quad F_{2}=E\times\{o_{E^{\prime}}\},\quad\Gamma=\{(p,f(p))\subset E\times E^{\prime}\,|\,p\in E\}.
Lemma \thelem.

Let a,b0a,b\geqslant 0 be integers such that (a,b)(0,0)(a,b)\neq(0,0) and let l=la,bl=l_{a,b} be the class of D=aF1+bF2+ΓD=aF_{1}+bF_{2}+\Gamma. Then

  1. (i)

    (l.F1)=1+b(l.F_{1})=1+b.

  2. (ii)

    ll is a polarization of type (1,a+ab+bk)(1,a+ab+bk).

  3. (iii)

    1/(1+b)β(l)max{1/(1+b),(1+b)/(a+ab+bk)}1/(1+b)\leqslant\beta(l)\leqslant\max\{1/(1+b),(1+b)/(a+ab+bk)\}.

  4. (iv)

    K(l)={(p,q)E×E|f^(q)=(a+k)p,f(p)=(1+b)q}K(l)=\{(p,q)\in E\times E^{\prime}\,|\,\hat{f}(q)=(a+k)p,f(p)=(1+b)q\}.

Proof.

(i) It is easy to check (F12)=(F22)=(Γ2)=0(F_{1}^{2})=(F_{2}^{2})=(\Gamma^{2})=0 and (F1.F2)=(F1.Γ)=1,(F2.Γ)=k(F_{1}.F_{2})=(F_{1}.\Gamma)=1,(F_{2}.\Gamma)=k. Hence (l.F1)=1+b(l.F_{1})=1+b holds.

(ii) Since (l2)=2(a+ab+bk)>0(l^{2})=2(a+ab+bk)>0 and (l.F1+F2)=a+b>0(l.F_{1}+F_{2})=a+b>0, ll is ample by [BL04, Corollary 4.3.3]. Since kerf/k\ker f\simeq\mathbb{Z}/k, fHom(E,E)f\in\operatorname{Hom}(E,E^{\prime}) is primitive, that is, ff is not written as cλc\lambda for some integer c2c\geqslant 2 and some λHom(E,E)\lambda\in\operatorname{Hom}(E,E^{\prime}). Hence lNS(X)l\in\mathrm{NS}(X) is primitive as well by [BL04, Theorem 11.5.1]. Since 2χ(l)=(l2)=2(a+ab+bk)2\chi(l)=(l^{2})=2(a+ab+bk), the type of ll is (1,a+ab+bk)(1,a+ab+bk).

(iii) holds by applying § 2 (ii) to S=F1S=F_{1} since β(l|F1)=1/deg(l|F1)=1/(1+b)\beta(l|_{F_{1}})=1/\deg(l|_{F_{1}})=1/(1+b).

(iv) By EE^:p𝒪E(poE)E\stackrel{{\scriptstyle\sim}}{{\rightarrow}}\widehat{E}:p\mapsto\mathcal{O}_{E}(p-o_{E}) and EE^:q𝒪E(qoE)E^{\prime}\stackrel{{\scriptstyle\sim}}{{\rightarrow}}\widehat{E^{\prime}}:q\mapsto\mathcal{O}_{E^{\prime}}(q-o_{E^{\prime}}), we may identify X^=E^×E^\widehat{X}=\widehat{E}\times\widehat{E^{\prime}} with XX. For (p,q)X(p,q)\in X, the point in X^=X\widehat{X}=X corresponding to t(p,q)𝒪X(F1)𝒪X(F1)=pr1𝒪E((p)oE)t_{(p,q)}^{*}\mathcal{O}_{X}(F_{1})\otimes\mathcal{O}_{X}(-F_{1})=\operatorname{pr}_{1}^{*}\mathcal{O}_{E}((-p)-o_{E}) is (p,oE)(-p,o_{E^{\prime}}), where pr1:XE\operatorname{pr}_{1}:X\rightarrow E is the natural projection. Similarly, t(p,q)𝒪X(F2)𝒪X(F2)t_{(p,q)}^{*}\mathcal{O}_{X}(F_{2})\otimes\mathcal{O}_{X}(-F_{2}) corresponds to (oE,q)(o_{E},-q).

Claim \theclaim.

t(p,q)𝒪X(Γ)𝒪X(Γ)t_{(p,q)}^{*}\mathcal{O}_{X}(\Gamma)\otimes\mathcal{O}_{X}(-\Gamma) corresponds to (f^(q)kp,f(p)q)E×E(\hat{f}(q)-kp,f(p)-q)\in E\times E^{\prime}.

Proof of § 4.

For a line bundle or a divisor NN on XX, we can define a group homomorphism

φN:XX^=X:xtxNN1.\varphi_{N}:X\rightarrow\widehat{X}=X\quad:\quad x\mapsto t_{x}^{*}N\otimes N^{-1}.

We need to show φΓ(p,q)=(f^(q)kp,f(p)q)\varphi_{\Gamma}(p,q)=(\hat{f}(q)-kp,f(p)-q) for (p,q)X=E×E(p,q)\in X=E\times E^{\prime}.

Let ΔE×E\Delta^{\prime}\subset E^{\prime}\times E^{\prime} be the diagonal. For (q1,q2)E×E(q_{1},q_{2})\in E^{\prime}\times E^{\prime}, we have

t(q1,q2)Δ(E×{oE})={q2q1},Δ(E×{oE})={oE}t_{(q_{1},q_{2})}^{*}\Delta^{\prime}\cap(E^{\prime}\times\{o_{E^{\prime}}\})=\{q_{2}-q_{1}\},\quad\Delta^{\prime}\cap(E^{\prime}\times\{o_{E^{\prime}}\})=\{o_{E^{\prime}}\}

under the identification E=E×{oE}E^{\prime}=E^{\prime}\times\{o_{E^{\prime}}\}. Hence the algebraically trivial line bundle t(q1,q2)𝒪(Δ)𝒪(Δ)|E×{oE}t_{(q_{1},q_{2})}^{*}\mathcal{O}(\Delta^{\prime})\otimes\mathcal{O}(-\Delta^{\prime})|_{E^{\prime}\times\{o_{E^{\prime}}\}} on EE^{\prime} corresponds to q2q1E=E^q_{2}-q_{1}\in E^{\prime}=\widehat{E^{\prime}}. Similarly, t(q1,q2)𝒪(Δ)𝒪(Δ)|{oE}×Et_{(q_{1},q_{2})}^{*}\mathcal{O}(\Delta^{\prime})\otimes\mathcal{O}(-\Delta^{\prime})|_{\{o_{E^{\prime}}\}\times E^{\prime}} on EE^{\prime} corresponds to q1q2E=E^q_{1}-q_{2}\in E^{\prime}=\widehat{E^{\prime}}. Thus the map

φΔ:E×EE×E^:(q1,q2)t(q1,q2)𝒪(Δ)𝒪(Δ)\varphi_{\Delta^{\prime}}:E^{\prime}\times E^{\prime}\rightarrow\widehat{E^{\prime}\times E^{\prime}}\quad:\quad(q_{1},q_{2})\mapsto t_{(q_{1},q_{2})}^{*}\mathcal{O}(\Delta^{\prime})\otimes\mathcal{O}(-\Delta^{\prime})

is written as φΔ(q1,q2)=(q2q1,q1q2)E×E=E×E^\varphi_{\Delta^{\prime}}(q_{1},q_{2})=(q_{2}-q_{1},q_{1}-q_{2})\in E^{\prime}\times E^{\prime}=\widehat{E^{\prime}\times E^{\prime}}.

Since Γ\Gamma is the pullback of Δ\Delta^{\prime} by (f,idE):E×EE×E(f,\operatorname{id}_{E^{\prime}}):E\times E^{\prime}\rightarrow E^{\prime}\times E^{\prime}, we have a commutative diagram

E×E\textstyle{E\times E^{\prime}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}φΓ\scriptstyle{\varphi_{\Gamma}}(f,idE)\scriptstyle{(f,\operatorname{id}_{E^{\prime}})}E×E^=E×E\textstyle{\widehat{E\times E^{\prime}}=E\times E^{\prime}}E×E\textstyle{E^{\prime}\times E^{\prime}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}φΔ\scriptstyle{\varphi_{\Delta^{\prime}}}E×E^=E×E.\textstyle{\widehat{E^{\prime}\times E^{\prime}}=E^{\prime}\times E^{\prime}\ignorespaces\ignorespaces\ignorespaces\ignorespaces.}(f^,idE)\scriptstyle{(\hat{f},\operatorname{id}_{E^{\prime}})}

Hence (p,q)E×E(p,q)\in E\times E^{\prime} is mapped by φΓ=(f^,idE)φΔ(f,idE)\varphi_{\Gamma}=(\hat{f},\operatorname{id}_{E^{\prime}})\circ\varphi_{\Delta^{\prime}}\circ(f,\operatorname{id}_{E^{\prime}}) as

(p,q)(f(p),q)(qf(p),f(p)q)(f^(q)\displaystyle(p,q)\mapsto(f(p),q)\mapsto(q-f(p),f(p)-q)\mapsto(\hat{f}(q) f^(f(p)),f(p)q).\displaystyle-\hat{f}(f(p)),f(p)-q).

Since f^(f(p))=kp\hat{f}(f(p))=kp by f^f=μkE\hat{f}\circ f=\mu_{k}^{E}, we have φΓ(p,q)=(f^(q)kp,f(p)q)E×E\varphi_{\Gamma}(p,q)=(\hat{f}(q)-kp,f(p)-q)\in E\times E^{\prime} and this claim follows. ∎

By § 4, t(p.q)𝒪X(D)𝒪X(D)t_{(p.q)}^{*}\mathcal{O}_{X}(D)\otimes\mathcal{O}_{X}(-D) corresponds to

a(p,oE)+b(oE,q)+(f^(q)kp,f(p)q)=(f^(q)(a+k)p,f(p)(1+b)q)E×E.a(-p,o_{E^{\prime}})+b(o_{E},-q)+(\hat{f}(q)-kp,f(p)-q)=(\hat{f}(q)-(a+k)p,f(p)-(1+b)q)\in E\times E^{\prime}.

Hence K(l)={(p,q)E×E|f^(q)=(a+k)p,f(p)=(1+b)q}K(l)=\{(p,q)\in E\times E^{\prime}\,|\,\hat{f}(q)=(a+k)p,f(p)=(1+b)q\} and (iv) follows. ∎

Now we can show Theorem 1.4 for abelian surfaces:

Proposition \theprop (== Theorem 1.4 for g=2g=2).

Let d1d\geqslant 1 be an integer and set m:=dm:=\lfloor\sqrt{d}\rfloor. Let (X,l)(X,l) be a general polarized abelian surface of type (1,d)(1,d). Then

  1. (1)

    1/dβ(l)1/m1/\sqrt{d}\leqslant\beta(l)\leqslant 1/m holds.

  2. (2)

    1/dβ(l)(m+1)/d<1/m1/\sqrt{d}\leqslant\beta(l)\leqslant(m+1)/d<1/m holds if dm2+m+1d\geqslant m^{2}+m+1.

Proof.

The lower bound 1/dgβ(l)1/\sqrt[g]{d}\leqslant\beta(l) follows from § 2 (i). Thus it suffices to find an example (X,l)(X,l) of type (1,d)(1,d) which satisfies the upper bound in (1) or (2) by Theorem 3.1. We construct such examples as (E×E,la,b)(E\times E^{\prime},l_{a,b}) by choosing suitable k,a,bk,a,b in § 4.

(1) Since (1) is true for m=1m=1, we may assume m2m\geqslant 2. Write d=(m1)q+rd=(m-1)q+r for integers q,rq,r with 1rm11\leqslant r\leqslant m-1 and set

k=qr,a=r,b=m1.k=q-r,\quad a=r,\quad b=m-1.

We note that kk is positive since qm+1>rq\geqslant m+1>r by dm2d\geqslant m^{2}, and

a+ab+bk=r+r(m1)+(m1)(qr)=r+(m1)q=d.a+ab+bk=r+r(m-1)+(m-1)(q-r)=r+(m-1)q=d.

Hence by § 4, (E×E,la,b)(E\times E^{\prime},l_{a,b}) for these k,a,bk,a,b is of type (1,d)(1,d) and

β(la,b)\displaystyle\beta(l_{a,b}) max{11+b,1+bd}=max{1m,md}1m\displaystyle\leqslant\max\left\{\frac{1}{1+b},\frac{1+b}{d}\right\}=\max\left\{\frac{1}{m},\frac{m}{d}\right\}\leqslant\frac{1}{m}

since dm2d\geqslant m^{2}.

(2) Write d=mq+rd=mq+r for integers q,rq,r with 1rm1\leqslant r\leqslant m and set

k=qr,a=r,b=m.k=q-r,\quad a=r,\quad b=m.

We note that kk is positive since qm+1>rq\geqslant m+1>r by dm2+m+1d\geqslant m^{2}+m+1, and

a+ab+bk=r+rm+m(qr)=r+mq=d.a+ab+bk=r+rm+m(q-r)=r+mq=d.

Hence by § 4, (E×E,la,b)(E\times E^{\prime},l_{a,b}) for these k,a,bk,a,b is of type (1,d)(1,d) and

β(la,b)\displaystyle\beta(l_{a,b}) max{11+b,1+bd}=max{1m+1,m+1d}=m+1d<1m\displaystyle\leqslant\max\left\{\frac{1}{1+b},\frac{1+b}{d}\right\}=\max\left\{\frac{1}{m+1},\frac{m+1}{d}\right\}=\frac{m+1}{d}<\frac{1}{m}

since m2+m+1d<(m+1)2m^{2}+m+1\leqslant d<(m+1)^{2}. ∎

As stated in Introduction, a very ample line bundle LL of type (1,d)(1,d) with d7d\geqslant 7 on an abelian surface XX satisfies (N0)(N_{0}), that is, LL is projectively normal by [Laz90], [FG04]. Furthermore, in this case the homogeneous ideal of XX embedded by |L||L| is generated by quadrics and cubics by [Ago17]. For d10d\geqslant 10, a general polarized abelian surface of type (1,d)(1,d) satisfies (N1)(N_{1}), that is, the homogeneous ideal of XX embedded by |L||L| is generated by quadrics by [GP98].

By § 4, we can show Theorem 1.2 for abelian surfaces, which partially recovers and generalizes the above results to higher syzygies. Partial means that we cannot give explicit conditions for the generality of (X,L)(X,L) and the bound d13d\geqslant 13 for (N1)(N_{1}) is larger than the bound d10d\geqslant 10 in [GP98].

Corollary \thecor (== Theorem 1.2 for g=2g=2).

Let p1p\geqslant-1 be an integer and let (X,L)(X,L) be a general polarized abelian surface of type (1,d)(1,d). Then

  1. (1)

    (Np)(N_{p}) holds for LL if d(p+2)2+(p+2)+1=p2+5p+7d\geqslant(p+2)^{2}+(p+2)+1=p^{2}+5p+7.

  2. (2)

    The homogeneous ideal of Xd1X\subset\mathbb{P}^{d-1} embedded by |L||L| is generated by quadrics and cubics if d7d\geqslant 7.

Proof.

(1) follows from Theorem 1.3 and § 4. For (2), it suffices to see that the Koszul cohomology group K1,q(X;L)=K1,q(X,𝒪X;L)K_{1,q}(X;L)=K_{1,q}(X,\mathcal{O}_{X};L) vanishes for any q3q\geqslant 3 (see [EL12, p. 606] for example), which follows from β(l)<1/2\beta(l)<1/2 for d7d\geqslant 7 and [Ito20, Proposition 2.5]. ∎

Remark \therem.

(1) Applying [Ito18, Theorem 1.2], which is a slight generalization of [KL19, Theorem 1.1], to a general polarized abelian surface (X,L)(X,L) of type (1,d)(1,d), we see that (Np)(N_{p}) holds for LL if d>2(p+2)2d>2(p+2)^{2}. § 4 improves the bound by a factor of approximately 22 for p1p\gg 1.
(2) Though the bound in § 4 (1) is a quadratic of pp, M. Gross and S. Popescu conjecture the following linear bound: [GP98, Conjecture (b)] states that LL satisfies (Nd/24)(N_{\lfloor d/2\rfloor-4}) for a general polarized abelian surface (X,L)(X,L) of type (1,d)(1,d) with d10d\geqslant 10. Equivalently, (Np)(N_{p}) holds for such LL if d2p+810d\geqslant 2p+8\geqslant 10.

5. On general polarized abelian varieties of type (1,,1,d)(1,\dots,1,d)

In this section, we prove Theorem 1.4. Although the argument becomes a little complicated, the essential idea is the same as the surface case. In the following examples, it is not difficult to bound β(l)\beta(l). To show that the type is of the form (1,,1,d)(1,\dots,1,d), we use § 4.

Let g2,k1,,kg11g\geqslant 2,k_{1},\dots,k_{g-1}\geqslant 1 be integers and set kg:=1k_{g}:=1. Let EgE_{g} be an elliptic curve and take an elliptic curve EiE_{i} and an isogeny fi:EiEgf_{i}:E_{i}\rightarrow E_{g} for each 1ig11\leqslant i\leqslant g-1 with kerfi/ki\ker f_{i}\simeq\mathbb{Z}/k_{i}\mathbb{Z}. Then we have f^ifi=μkiEi,fif^i=μkiEg\hat{f}_{i}\circ f_{i}=\mu^{E_{i}}_{k_{i}},f_{i}\circ\hat{f}_{i}=\mu^{E_{g}}_{k_{i}} for the dual isogeny f^i:EgEi\hat{f}_{i}:E_{g}\rightarrow E_{i}. Let X=E1×E2××EgX=E_{1}\times E_{2}\times\dots\times E_{g} and let Fi=pri(oEi)F_{i}=\operatorname{pr}_{i}^{*}(o_{E_{i}}), where pri:XEi\operatorname{pr}_{i}:X\rightarrow E_{i} is the projection to the ii-th factor. A divisor Γ\Gamma on XX is defined by

Γ={(p1,,pg1,i=1g1fi(pi))X|piEi for 1ig1}.\Gamma=\left\{\Big{(}p_{1},\dots,p_{g-1},\sum_{i=1}^{g-1}f_{i}(p_{i})\Big{)}\in X\,\Big{|}\,p_{i}\in E_{i}\text{ for }1\leqslant i\leqslant g-1\right\}.

The following proposition is a generalization of § 4 (ii), (iii) to higher dimensions:

Proposition \theprop.

Under the above setting, let l=la,bl=l_{a,b} be the class of

Da,b=aF1+i=2g1Fi+bFg+ΓD_{a,b}=aF_{1}+\sum_{i=2}^{g-1}F_{i}+bF_{g}+\Gamma

for integers a,b0a,b\geqslant 0 with (a,b)(0,0)(a,b)\neq(0,0). Set Ni:=j=i+1gkjN_{i}:=\sum_{j=i+1}^{g}k_{j} for 1ig11\leqslant i\leqslant g-1 and Ng:=0N_{g}:=0. Then

  1. (1)

    The type of ll is (1,,1,d)(1,\dots,1,d) for d=a+abN1+bk1d=a+abN_{1}+bk_{1}.

  2. (2)

    It holds that

    11+bβ(l)max{max1ig11+bNi+11+bNi,1+bN1d}.\frac{1}{1+b}\leqslant\beta(l)\leqslant\max\left\{\max_{1\leqslant i\leqslant g-1}\frac{1+bN_{i+1}}{1+bN_{i}},\frac{1+bN_{1}}{d}\right\}.
Example \theex.

Before proving § 5, we give an example. If g=3g=3, § 5 state that la,bl_{a,b} is of type (1,1,a+ab+abk2+bk1)(1,1,a+ab+abk_{2}+bk_{1}) and

(5.1) 11+bβ(la,b)max{11+b,1+b1+b+bk2,1+b+bk2a(1+b+bk2)+bk1}.\displaystyle\frac{1}{1+b}\leqslant\beta(l_{a,b})\leqslant\max\left\{\frac{1}{1+b},\frac{1+b}{1+b+bk_{2}},\frac{1+b+bk_{2}}{a(1+b+bk_{2})+bk_{1}}\right\}.

For example, if we take a=1,b=3,k1=9,k2=3a=1,b=3,k_{1}=9,k_{2}=3, then l1,3l_{1,3} is of type (1,1,40)(1,1,40) and 1/4β(l1,3)max{1/4,4/13,13/40}=13/40<1/31/4\leqslant\beta(l_{1,3})\leqslant\max\{1/4,4/13,13/40\}=13/40<1/3 holds.

To show § 5, we prepare two lemmas. We denote by oiEio_{i}\in E_{i} the origin of EiE_{i}. By EiE^i:pi𝒪Ei(pioi)E_{i}\stackrel{{\scriptstyle\sim}}{{\rightarrow}}\widehat{E}_{i}:p_{i}\mapsto\mathcal{O}_{E_{i}}(p_{i}-o_{i}), we may identify X^=E^1××E^g\widehat{X}=\widehat{E}_{1}\times\dots\times\widehat{E}_{g} with XX as in the proof of § 4. Under this identification, the point in X^=X\widehat{X}=X corresponding to tx𝒪X(Fi)𝒪X(Fi)t_{x}^{*}\mathcal{O}_{X}(F_{i})\otimes\mathcal{O}_{X}(-F_{i}) for x=(p1,,pg)Xx=(p_{1},\dots,p_{g})\in X is (o1,,oi1,pi,oi+1,,og)(o_{1},\dots,o_{i-1},-p_{i},o_{i+1},\dots,o_{g}) since

(5.2) tx𝒪X(Fi)𝒪X(Fi)=pri𝒪Ei((pi)oi).\displaystyle t_{x}^{*}\mathcal{O}_{X}(F_{i})\otimes\mathcal{O}_{X}(-F_{i})=\operatorname{pr}_{i}^{*}\mathcal{O}_{E_{i}}((-p_{i})-o_{i}).

The following is a generalization of § 4 to higher dimensions:

Lemma \thelem.

For x=(p1,,pg)Xx=(p_{1},\dots,p_{g})\in X, tx𝒪X(Γ)𝒪X(Γ)t_{x}^{*}\mathcal{O}_{X}(\Gamma)\otimes\mathcal{O}_{X}(-\Gamma) corresponds to

(f^1(A),,f^g1(A),A)X=X^,(\hat{f}_{1}(A),\dots,\hat{f}_{g-1}(A),-A)\in X=\widehat{X},

where A=pgi=1g1fi(pi)EgA=p_{g}-\sum_{i=1}^{g-1}f_{i}(p_{i})\in E_{g}.

Proof.

Set Ei={(q1,,qg)|qj=og for ji}Y:=Eg××EgE^{\prime}_{i}=\{(q_{1},\dots,q_{g})\,|\,q_{j}=o_{g}\text{ for }j\neq i\}\subset Y:=E_{g}\times\dots\times E_{g} and consider a divisor

Δ={(q1,,qg1,i=1g1qi)|qiEg for 1ig1}Y=Eg××Eg.\Delta^{\prime}=\left\{\Big{(}q_{1},\dots,q_{g-1},\sum_{i=1}^{g-1}q_{i}\Big{)}\,\Big{|}\,q_{i}\in E_{g}\text{ for }1\leqslant i\leqslant g-1\right\}\subset Y=E_{g}\times\dots\times E_{g}.

For y=(q1,,qg)Yy=(q_{1},\dots,q_{g})\in Y, we see that

tyΔEi={A} for 1ig1,tyΔEg={A},ΔEi={og} for 1igt_{y}^{*}\Delta^{\prime}\cap E^{\prime}_{i}=\{A^{\prime}\}\text{ for }1\leqslant i\leqslant g-1,\quad t_{y}^{*}\Delta^{\prime}\cap E^{\prime}_{g}=\{-A^{\prime}\},\quad\Delta^{\prime}\cap E^{\prime}_{i}=\{o_{g}\}\text{ for }1\leqslant i\leqslant g

for A=qgi=1g1qiEgA^{\prime}=q_{g}-\sum_{i=1}^{g-1}q_{i}\in E_{g} under the natural identification EiEgE^{\prime}_{i}\simeq E_{g}. Hence ty𝒪Y(Δ)𝒪Y(Δ)t_{y}^{*}\mathcal{O}_{Y}(\Delta^{\prime})\otimes\mathcal{O}_{Y}(-\Delta^{\prime}) corresponds to (A,,A,A)Y^=Y(A^{\prime},\dots,A^{\prime},-A^{\prime})\in\widehat{Y}=Y.

By definition, Γ\Gamma is the pullback of Δ\Delta^{\prime} by (f1,,fg1,idEg):XY(f_{1},\dots,f_{g-1},\operatorname{id}_{E_{g}}):X\rightarrow Y. Hence this lemma follows from the same argument as in § 4 using the commutative diagram

X\textstyle{X\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}φΓ\scriptstyle{\varphi_{\Gamma}}(f1,,fg1,idEg)\scriptstyle{(f_{1},\dots,f_{g-1},\operatorname{id}_{E_{g}})}X^=X\textstyle{\widehat{X}=X}Y\textstyle{Y\ignorespaces\ignorespaces\ignorespaces\ignorespaces}φΔ\scriptstyle{\varphi_{\Delta^{\prime}}}Y^=Y.\textstyle{\widehat{Y}=Y.\ignorespaces\ignorespaces\ignorespaces\ignorespaces}(f^1,,f^g1,idEg)\scriptstyle{(\hat{f}_{1},\dots,\hat{f}_{g-1},\operatorname{id}_{E_{g}})}

In fact, AA^{\prime} for (q1,,qg)=(f1(p1),,fg1(pg1),pg)(q_{1},\dots,q_{g})=(f_{1}(p_{1}),\dots,f_{g-1}(p_{g-1}),p_{g}) is nothing but A=pgi=1g1fi(pi)A=p_{g}-\sum_{i=1}^{g-1}f_{i}(p_{i}). Thus (p1,,pg)X(p_{1},\dots,p_{g})\in X is mapped as

(p1,,pg)(f1(p1),,fg1(pg1),pg)(A,,A,A)(f^1(A),,f^g1(A),A)(p_{1},\dots,p_{g})\mapsto(f_{1}(p_{1}),\dots,f_{g-1}(p_{g-1}),p_{g})\mapsto(A,\dots,A,-A)\mapsto(\hat{f}_{1}(A),\dots,\hat{f}_{g-1}(A),-A)

by φΓ=(f^1,,f^g1,idEg)φΔ(f1,,fg1,idEg)\varphi_{\Gamma}=(\hat{f}_{1},\dots,\hat{f}_{g-1},\operatorname{id}_{E_{g}})\circ\varphi_{\Delta^{\prime}}\circ(f_{1},\dots,f_{g-1},\operatorname{id}_{E_{g}}). ∎

Lemma \thelem.

Under the setting of § 5, it holds that

K(la,b){(p1,pg)E1×Eg|ap1=bf^1(pg),f1(p1)=(1+bN1)pg}.K(l_{a,b})\simeq\left\{(p_{1},p_{g})\in E_{1}\times E_{g}\,\Big{|}\,ap_{1}=-b\hat{f}_{1}(p_{g}),f_{1}(p_{1})=\left(1+bN_{1}\right)p_{g}\right\}.
Proof.

By 5.2 and § 5, tx𝒪X(Da,b)𝒪X(Da,b)t_{x}^{*}\mathcal{O}_{X}(D_{a,b})\otimes\mathcal{O}_{X}(-D_{a,b}) for x=(p1,,pg)x=(p_{1},\dots,p_{g}) corresponds to the point

(5.3) (ap1+f^1(A),p2+f^2(A),,pg1+f^g1(A),bpgA)X.\displaystyle(-ap_{1}+\hat{f}_{1}(A),-p_{2}+\hat{f}_{2}(A),\dots,-p_{g-1}+\hat{f}_{g-1}(A),-bp_{g}-A)\in X.

Hence (p1,,pg)(p_{1},\dots,p_{g}) is contained in K(la,b)K(l_{a,b}) if and only if

ap1=f^1(A),pi=f^i(A) for  2ig1,A=bpg\displaystyle ap_{1}=\hat{f}_{1}(A),\quad p_{i}=\hat{f}_{i}(A)\ \text{ for }\ 2\leqslant i\leqslant g-1,\quad A=-bp_{g}
\displaystyle\Leftrightarrow\quad ap1=f^1(bpg),pi=f^i(bpg) for  2ig1,A=bpg\displaystyle ap_{1}=\hat{f}_{1}(-bp_{g}),\quad p_{i}=\hat{f}_{i}(-bp_{g})\ \text{ for }\ 2\leqslant i\leqslant g-1,\quad A=-bp_{g}
\displaystyle\Leftrightarrow\quad ap1=bf^1(pg),pi=bf^i(pg) for  2ig1,A=bpg.\displaystyle ap_{1}=-b\hat{f}_{1}(p_{g}),\quad p_{i}=-b\hat{f}_{i}(p_{g})\ \text{ for }\ 2\leqslant i\leqslant g-1,\quad A=-bp_{g}.

Under the condition pi=bf^i(pg)p_{i}=-b\hat{f}_{i}(p_{g}) for 2ig12\leqslant i\leqslant g-1, it holds that

bpg+A=bpg+pgi=1g1fi(pi)\displaystyle bp_{g}+A=bp_{g}+p_{g}-\sum_{i=1}^{g-1}f_{i}(p_{i}) =bpg+pgi=2g1fi(bf^i(pg))f1(p1)\displaystyle=bp_{g}+p_{g}-\sum_{i=2}^{g-1}f_{i}(-b\hat{f}_{i}(p_{g}))-f_{1}(p_{1})
=bpg+pg+bi=2g1fi(f^i(pg))f1(p1)\displaystyle=bp_{g}+p_{g}+b\sum_{i=2}^{g-1}f_{i}(\hat{f}_{i}(p_{g}))-f_{1}(p_{1})
=bpg+pg+bi=2g1kipgf1(p1)\displaystyle=bp_{g}+p_{g}+b\sum_{i=2}^{g-1}k_{i}p_{g}-f_{1}(p_{1})
=(1+b(1+i=2g1ki))pgf1(p1)\displaystyle=\left(1+b\left(1+\sum_{i=2}^{g-1}k_{i}\right)\right)p_{g}-f_{1}(p_{1})
=(1+bN1)pgf1(p1),\displaystyle=\left(1+bN_{1}\right)p_{g}-f_{1}(p_{1}),

where the fourth equality follows from fif^i=μkiEgf_{i}\circ\hat{f}_{i}=\mu^{E_{g}}_{k_{i}} and the last one follows from kg=1k_{g}=1 and N1=i=2gkiN_{1}=\sum_{i=2}^{g}k_{i}. Hence we have

K(la,b)\displaystyle K(l_{a,b}) ={(p1,,pg)|ap1=bf^1(pg),pi=bf^i(pg) for 2ig1,f1(p1)=(1+bN1)pg}\displaystyle=\{(p_{1},\dots,p_{g})\,|\,ap_{1}=-b\hat{f}_{1}(p_{g}),p_{i}=-b\hat{f}_{i}(p_{g})\text{ for }2\leqslant i\leqslant g-1,f_{1}(p_{1})=\left(1+bN_{1}\right)p_{g}\}
{(p1,pg)|ap1=bf^1(pg),f1(p1)=(1+bN1)pg}.\displaystyle\simeq\{(p_{1},p_{g})\,|\,ap_{1}=-b\hat{f}_{1}(p_{g}),f_{1}(p_{1})=\left(1+bN_{1}\right)p_{g}\}.

Proof of § 5.

(1) Since Fi,ΓXF_{i},\Gamma\subset X are abelian subvarieties of codimension one, Fi2=Γ2=0F_{i}^{2}=\Gamma^{2}=0 as cycles on XX. By the definition of Γ\Gamma, it is easy to see that

(F1Fg)=(F1Fg1.Γ)=1,(F1Fi1.Fi+1Fg.Γ)=ki for  1ig1.(F_{1}\cdots F_{g})=(F_{1}\cdots F_{g-1}.\Gamma)=1,\quad(F_{1}\cdots F_{i-1}.F_{i+1}\cdots F_{g}.\Gamma)=k_{i}\ \text{ for }\ 1\leqslant i\leqslant g-1.

Hence we have

χ(l)=(lg)g!\displaystyle\chi(l)=\frac{(l^{g})}{g!} =b(F2Fg.Γ)+abi=2g1(F1Fi1.Fi+1Fg.Γ)\displaystyle=b(F_{2}\cdots F_{g}.\Gamma)+ab\sum_{i=2}^{g-1}(F_{1}\cdots F_{i-1}.F_{i+1}\cdots F_{g}.\Gamma)
+a(F1Fg1.Γ)+ab(F1Fg)\displaystyle\hskip 170.71652pt+a(F_{1}\cdots F_{g-1}.\Gamma)+ab(F_{1}\cdots F_{g})
=bk1+abi=2g1ki+a+ab\displaystyle=bk_{1}+ab\sum_{i=2}^{g-1}k_{i}+a+ab
=a+ab(1+i=2g1ki)+bk1=a+abN1+bk1=d.\displaystyle=a+ab\left(1+\sum_{i=2}^{g-1}k_{i}\right)+bk_{1}=a+abN_{1}+bk_{1}=d.

For simplicity, set N:=N1N:=N_{1}. By § 5, we may identify K(l)K(l) with

{(p1,pg)E1×Eg|ap1=bf^1(pg),f1(p1)=(1+bN)pg}.\displaystyle\left\{(p_{1},p_{g})\in E_{1}\times E_{g}\,\Big{|}\,ap_{1}=-b\hat{f}_{1}(p_{g}),f_{1}(p_{1})=(1+bN)p_{g}\right\}.

Consider a group

K={(p1,pg)E1×Eg|aNp1=bNf^1(pg),f1(p1)=(1+bN)pg}.\displaystyle K^{\prime}=\left\{(p_{1},p_{g})\in E_{1}\times E_{g}\,\Big{|}\,aNp_{1}=-bN\hat{f}_{1}(p_{g}),f_{1}(p_{1})=(1+bN)p_{g}\right\}.

By definition, we have

NK:={(Np1,Npg)|(p1,pg)K}K(l)K.\displaystyle NK^{\prime}:=\{(Np_{1},Np_{g})\,|\,(p_{1},p_{g})\in K^{\prime}\}\subset K(l)\subset K^{\prime}.

Under the condition f1(p1)=(1+bN)pgf_{1}(p_{1})=(1+bN)p_{g},

(5.4) aNp1=bNf^1(pg)aNp1+f^1(f1(p1))=bNf^1(pg)+f^1((1+bN)pg)aNp1+k1p1=bNf^1(pg)+(1+bN)f^1(pg)(aN+k1)p1=f^1(pg).\displaystyle\begin{aligned} aNp_{1}=-bN\hat{f}_{1}(p_{g})\ &\Leftrightarrow\ aNp_{1}+\hat{f}_{1}(f_{1}(p_{1}))=-bN\hat{f}_{1}(p_{g})+\hat{f}_{1}((1+bN)p_{g})\\ &\Leftrightarrow\ aNp_{1}+k_{1}p_{1}=-bN\hat{f}_{1}(p_{g})+(1+bN)\hat{f}_{1}(p_{g})\\ &\Leftrightarrow\ (aN+k_{1})p_{1}=\hat{f}_{1}(p_{g}).\end{aligned}

Hence we have

K={(p1,pg)E1×Eg|f^1(pg)=(aN+k1)p1,f1(p1)=(1+bN)pg}.K^{\prime}=\left\{(p_{1},p_{g})\in E_{1}\times E_{g}\,\Big{|}\,\hat{f}_{1}(p_{g})=(aN+k_{1})p_{1},f_{1}(p_{1})=(1+bN)p_{g}\right\}.

By § 4 (iv), we have K=K(l)K^{\prime}=K(l^{\prime}) for l:=laN,bNl^{\prime}:=l_{aN,bN} on E×EE\times E^{\prime} with E=E1,E=Eg,k=k1E=E_{1},E^{\prime}=E_{g},k=k_{1} in § 4. By § 4 (ii), ll^{\prime} is of type (1,aN+aNbN+bNk1)=(1,Nd)(1,aN+aNbN+bNk_{1})=(1,Nd) since d=a+abN+bk1d=a+abN+bk_{1}. Hence K=K(l)(/Nd)2K^{\prime}=K(l^{\prime})\simeq(\mathbb{Z}/Nd\mathbb{Z})^{\oplus 2}. Since

K(l)NK(N/Nd)2(/d)2K(l)\supset NK^{\prime}\simeq(N\mathbb{Z}/Nd\mathbb{Z})^{\oplus 2}\simeq(\mathbb{Z}/d\mathbb{Z})^{\oplus 2}

and |K(l)|=χ(l)2=d2|K(l)|=\chi(l)^{2}=d^{2}, we have K(l)=NK(/d)2K(l)=NK^{\prime}\simeq(\mathbb{Z}/d\mathbb{Z})^{\oplus 2}. Thus the type of ll is (1,,1,d)(1,\dots,1,d).

(2) Set a flag X=X0X1Xg1X=X_{0}\supset X_{1}\supset\dots\supset X_{g-1} of abelian subvarieties of XX as

Xi={(p1,,pg)X|pj=oj for 1ji}=F1FiX.X_{i}=\{(p_{1},\dots,p_{g})\in X\,|\,p_{j}=o_{j}\text{ for }1\leqslant j\leqslant i\}=F_{1}\cap\dots\cap F_{i}\subset X.

Applying § 2 (ii) repeatedly, we have

β(l)max{1χ(l|Xg1),χ(l|Xg1)χ(l|Xg2),,χ(l|X1)χ(l|X0)}\beta(l)\leqslant\max\left\{\frac{1}{\chi(l|_{X_{g-1}})},\frac{\chi(l|_{X_{g-1}})}{\chi(l|_{X_{g-2}})},\dots,\frac{\chi(l|_{X_{1}})}{\chi(l|_{X_{0}})}\right\}

since β(l|Xg1)=1/deg(l|Xg1)=1/χ(l|Xg1)\beta(l|_{X_{g-1}})=1/\deg(l|_{X_{g-1}})=1/\chi(l|_{X_{g-1}}) for the elliptic curve Xg1X_{g-1}. Then the upper bound in (2) follows from

χ(l|Xi)=(l|Xigi)(gi)!=(F1Fi.lgi)(gi)!=1+bNi\chi(l|_{X_{i}})=\frac{(l|_{X_{i}}^{g-i})}{(g-i)!}=\frac{(F_{1}\cdots F_{i}.l^{g-i})}{(g-i)!}=1+bN_{i}

for 1ig11\leqslant i\leqslant g-1 and χ(l|X0)=χ(l)=d\chi(l|_{X_{0}})=\chi(l)=d. The lower bound β(l)1/(1+b)\beta(l)\geqslant 1/(1+b) holds by applying § 2 (ii) to Z=Xg1Z=X_{g-1} since β(l|Xg1)=1/χ(l|Xg1)=1/(1+b)\beta(l|_{X_{g-1}})=1/\chi(l|_{X_{g-1}})=1/(1+b). ∎

Now we can show Theorem 1.4:

Proof of Theorem 1.4.

The lower bound 1/dgβ(l)1/\sqrt[g]{d}\leqslant\beta(l) follows from § 2 (i). If g=1g=1, this theorem holds since β(l)=1/deg(l)\beta(l)=1/\deg(l) in this case. Thus it suffices to find an example (X,l)(X,l) of type (1,,1,d)(1,\dots,1,d) such that β(l)1/m\beta(l)\leqslant 1/m or β(l)<1/m\beta(l)<1/m for m=dgm=\lfloor\sqrt[g]{d}\rfloor and g2g\geqslant 2 by Theorem 3.1. We construct such examples as (E1××Eg,la,b)(E_{1}\times\dots\times E_{g},l_{a,b}) for suitable k1,,kg1,a,bk_{1},\dots,k_{g-1},a,b.

We set ki=mgik_{i}=m^{g-i} for 2ig12\leqslant i\leqslant g-1. In this case, we have

Ni=j=i+1gkj=j=i+1gmgj=n=0gi1mnN_{i}=\sum_{j=i+1}^{g}k_{j}=\sum_{j=i+1}^{g}m^{g-j}=\sum_{n=0}^{g-i-1}m^{n}

for 1ig11\leqslant i\leqslant g-1 since kg=1k_{g}=1. We take k1,a,bk_{1},a,b as follows.

(1) Write d=(m1)q+rd=(m-1)q+r for integers q,rq,r with 1rm11\leqslant r\leqslant m-1 and set

k1=qrN1,a=r,b=m1.k_{1}=q-rN_{1},\quad a=r,\quad b=m-1.

Since dmgd\geqslant m^{g}, we have qmg1++m+1q\geqslant m^{g-1}+\dots+m+1. Hence k1k_{1} is positive by N1=mg2++m+1N_{1}=m^{g-2}+\dots+m+1 and rm1r\leqslant m-1. Furthermore,

1+bNi\displaystyle 1+bN_{i} =1+(m1)n=0gi1mn=mgi,\displaystyle=1+(m-1)\sum_{n=0}^{g-i-1}m^{n}=m^{g-i},
a+abN1+bk1\displaystyle a+abN_{1}+bk_{1} =r+r(m1)N1+(m1)(qrN1)=r+(m1)q=d.\displaystyle=r+r(m-1)N_{1}+(m-1)(q-rN_{1})=r+(m-1)q=d.

Hence by § 5, (E1××Eg,la,b)(E_{1}\times\dots\times E_{g},l_{a,b}) for these k1,,kg1,a,bk_{1},\dots,k_{g-1},a,b is of type (1,,1,d)(1,\dots,1,d) and

β(la,b)\displaystyle\beta(l_{a,b}) max{max1ig11+bNi+11+bNi,1+bN1d}\displaystyle\leqslant\max\left\{\max_{1\leqslant i\leqslant g-1}\frac{1+bN_{i+1}}{1+bN_{i}},\frac{1+bN_{1}}{d}\right\}
=max{1m,mm2,,mg2mg1,mg1d}=max{1m,mg1d}1m\displaystyle=\max\left\{\frac{1}{m},\frac{m}{m^{2}},\dots,\frac{m^{g-2}}{m^{g-1}},\frac{m^{g-1}}{d}\right\}=\max\left\{\frac{1}{m},\frac{m^{g-1}}{d}\right\}\leqslant\frac{1}{m}

since 1+bNi=mgi1+bN_{i}=m^{g-i} and dmgd\geqslant m^{g}.

(2) Write d=mq+rd=mq+r for integers q,rq,r with 1rm1\leqslant r\leqslant m and set

k1=qrN1,a=r,b=m.k_{1}=q-rN_{1},\quad a=r,\quad b=m.

Since dmg++m+1d\geqslant m^{g}+\dots+m+1, we have qmg1++m+1q\geqslant m^{g-1}+\dots+m+1. Hence k1k_{1} is positive by N1=mg2++m+1N_{1}=m^{g-2}+\dots+m+1 and rmr\leqslant m. Furthermore,

1+bNi\displaystyle 1+bN_{i} =1+mn=0gi1mn=n=0gimn,\displaystyle=1+m\sum_{n=0}^{g-i-1}m^{n}=\sum_{n=0}^{g-i}m^{n},
a+abN1+bk1\displaystyle a+abN_{1}+bk_{1} =r+rmN1+m(qrN1)=r+mq=d.\displaystyle=r+rmN_{1}+m(q-rN_{1})=r+mq=d.

Hence by § 5, (E1××Eg,la,b)(E_{1}\times\dots\times E_{g},l_{a,b}) for these k1,,kg1,a,bk_{1},\dots,k_{g-1},a,b is of type (1,,1,d)(1,\dots,1,d) and

β(la,b)\displaystyle\beta(l_{a,b}) max{max1ig11+bNi+11+bNi,1+bN1d}<1m\displaystyle\leqslant\max\left\{\max_{1\leqslant i\leqslant g-1}\frac{1+bN_{i+1}}{1+bN_{i}},\frac{1+bN_{1}}{d}\right\}<\frac{1}{m}

since

1+bNi+11+bNi=mgi1++m+1mgi++m+1<1m\frac{1+bN_{i+1}}{1+bN_{i}}=\frac{m^{g-i-1}+\dots+m+1}{m^{g-i}+\dots+m+1}<\frac{1}{m}

for 1ig11\leqslant i\leqslant g-1 and

1+bN1d=mg1++m+1dmg1++m+1mg++m+1<1m\frac{1+bN_{1}}{d}=\frac{m^{g-1}+\dots+m+1}{d}\leqslant\frac{m^{g-1}+\dots+m+1}{m^{g}+\dots+m+1}<\frac{1}{m}

by the assumption dmg++m+1d\geqslant m^{g}+\dots+m+1. ∎

Proof of Theorem 1.2.

The proof is the same as that of § 4: The statement about (Np)(N_{p}), in particular projective normality, follows from Theorem 1.3 and Theorem 1.4. The last statement about generators of homogenous ideal of XX follows from the vanishing K1,q(X;L)=0K_{1,q}(X;L)=0 for any q3q\geqslant 3, which follows from β(l)<1/2\beta(l)<1/2 for d2g+11d\geqslant 2^{g+1}-1 and [Ito20, Proposition 2.5]. ∎

Example \theex.

If we choose a,b,kia,b,k_{i} carefully in § 5, we might obtain a better bound than that in Theorem 1.4. For example, consider the case g=3g=3 and 4d74\leqslant d\leqslant 7. By Theorem 1.4, we know that 1/2<1/d3β(l)<11/2<1/\sqrt[3]{d}\leqslant\beta(l)<1 for general (X,l)(X,l) of type (1,1,d)(1,1,d). In fact, we have the following bound by taking a=b=1a=b=1 and suitable k1,k2k_{1},k_{2} in 5.1:

  • d=4d=4 : β(l)3/4\beta(l)\leqslant 3/4 by taking (k1,k2)=(1,1)(k_{1},k_{2})=(1,1).

  • d=5d=5 : β(l)2/3\beta(l)\leqslant 2/3 by taking (k1,k2)=(2,1)(k_{1},k_{2})=(2,1).

  • d=6d=6 : β(l)2/3\beta(l)\leqslant 2/3 by taking (k1,k2)=(3,1)(k_{1},k_{2})=(3,1) or (2,2)(2,2).

  • d=7d=7 : β(l)4/7\beta(l)\leqslant 4/7 by taking (k1,k2)=(3,2)(k_{1},k_{2})=(3,2).

Remark \therem.

We recall some related results and compare them to Theorem 1.2.

(1) For p=0p=0, Iyer [Iye03, Theorem 1.2] proves that an ample line bundle LL on a simple abelian variety of dimension gg is projectively normal if χ(L)>2gg!\chi(L)>2^{g}\cdot g!. Although Theorem 1.2 gives no explicit condition on the generality of (X,L)(X,L), [Iye03] gives an explicit condition as XX is simple. Furthermore, [Iye03] has no assumption on the type of LL. On the other hand, the bound χ(L)=d2g+11\chi(L)=d\geqslant 2^{g+1}-1 in Theorem 1.2 is smaller than the bound in [Iye03] by a factor of approximately g!/2g!/2.

(2) For p1p\geqslant-1, R. Lazarsfeld, G. Pareschi and M. Popa [LPP11, Corollary B] prove that LL satisfies (Np)(N_{p}) if χ(L)>(4g)g2g!(p+2)g\chi(L)>\frac{(4g)^{g}}{2g!}(p+2)^{g} and (X,L)(X,L) is very general. [LPP11] also has no assumption on the type of LL. On the other hand, the bound χ(L)((p+2)g+11)/(p+1)\chi(L)\geqslant((p+2)^{g+1}-1)/(p+1) in Theorem 1.2 is smaller than the bound in [LPP11] by a factor of approximately (4g)g2g!p+1p+2\frac{(4g)^{g}}{2g!}\cdot\frac{p+1}{p+2}.

(3) For p1p\geqslant-1, the author [Ito18, Question 4.2] asks if LL satisfies (Np)(N_{p}) when (LdimZ.Z)>((p+2)dimZ)dimZ(L^{\dim Z}.Z)>((p+2)\dim Z)^{\dim Z} holds for any abelian subvariety ZXZ\subset X. This question is answered affirmatively by [Ito18], [Ito20] for g=2,3g=2,3 (see also [KL19], [Loz18]). In arbitrary dimension, Z. Jiang [Jia21, Theorem 1.5] proves that (Np)(N_{p}) holds for LL under the assumption (LdimZ.Z)>(2(p+2)dimZ)dimZ(L^{\dim Z}.Z)>(2(p+2)\dim Z)^{\dim Z}.

If this question has an affirmative answer for any g1g\geqslant 1, (Np)(N_{p}) holds for LL if XX is simple and (Lg)>((p+2)g)g(L^{g})>((p+2)g)^{g}, equivalently χ(L)>ggg!(p+2)g\chi(L)>\frac{g^{g}}{g!}(p+2)^{g}. The bound χ(L)((p+2)g+11)/(p+1)\chi(L)\geqslant((p+2)^{g+1}-1)/(p+1) in Theorem 1.2 is smaller than the bound χ(L)>ggg!(p+2)g\chi(L)>\frac{g^{g}}{g!}(p+2)^{g} by a factor of approximately ggg!p+1p+2\frac{g^{g}}{g!}\cdot\frac{p+1}{p+2}.

(4) Jiang also gives a numerical condition for a very general abelian variety (X,L)(X,L) to satisfy (Np)(N_{p}) in [Jia21, Theorem 2.9]. As a special case, he [Jia21, Theorem 1.6] proves that (Np)(N_{p}) holds for LL if (X,L)(X,L) is very general of type (1,,1,d)(1,\dots,1,d) and d>ggg!(p+2)gd>\frac{g^{g}}{g!}(p+2)^{g}. In particular, [Ito18, Question 4.2] has an affirmative answer for very general (X,L)(X,L) of type (1,,1,d)(1,\dots,1,d). In fact, the condition that (X,L)(X,L) is very general is explicit there, that is, [Jia21, Theorems 1.6, 2.9] just require the space of Hodge classes to be of dimension one in each degree. Furthermore, [Jia21, Theorem 2.9] treats LL of any type.

On the other hand, the bound in Theorem 1.2 is smaller than the bound in [Jia21, Theorem 1.6] by a factor of approximately ggg!p+1p+2\frac{g^{g}}{g!}\cdot\frac{p+1}{p+2} as in (3).

Appendix A Computation of β(l)\beta(l) of general (X,l)(X,l) of type (1,d)(1,d) for some dd

Let (X,l)(X,l) be a general polarized abelian surface of type (1,d)(1,d). By § 4 (1), we have β(l)=1/m\beta(l)=1/m if d=m2d=m^{2} for some integer m1m\geqslant 1. In the appendix, we show that the upper bounds of β(l)\beta(l) in § 4 are sharp when mm is odd and dd is equal to m2+mm^{2}+m or (m+1)22(m+1)^{2}-2 or (m+1)21(m+1)^{2}-1.

Lemma \thelem.

Let m1m\geqslant 1 be an odd integer. Let (X,l)(X,l) be a polarized abelian surface of type (1,m2+m)(1,m^{2}+m). Then β(l)1/m\beta(l)\geqslant 1/m holds.

In particular, β(l)=1/m\beta(l)=1/m holds if (X,l)(X,l) is general.

Proof.

Let φl:XX^\varphi_{l}:X\rightarrow\widehat{X} be the isogeny obtained by ptpLL1X^p\mapsto t_{p}^{*}L\otimes L^{-1}\in\widehat{X}.

Claim \theclaim.

There exists σ~X\tilde{\sigma}\in X of order 2m2m such that the order of φl(σ~)X^\varphi_{l}(\tilde{\sigma})\in\widehat{X} is 2m2m as well.

Proof of Appendix A.

Let X2m={xX| 2mx=oX}(/2m)4X_{2m}=\{x\in X\,|\,2mx=o_{X}\}\simeq(\mathbb{Z}/2m\mathbb{Z})^{\oplus 4} and consider the exact sequence

0kerφlX2mX2mφl(X2m)0.0\rightarrow\ker\varphi_{l}\cap X_{2m}\rightarrow X_{2m}\rightarrow\varphi_{l}(X_{2m})\rightarrow 0.

Since kerφl=K(l)(/(m2+m))2\ker\varphi_{l}=K(l)\simeq(\mathbb{Z}/(m^{2}+m)\mathbb{Z})^{\oplus 2}, the subgroup kerφlX2mkerφl\ker\varphi_{l}\cap X_{2m}\subset\ker\varphi_{l} is generated by at most two elements. By considering elementary divisors, we see that there exists a basis σ1,,σ4\sigma_{1},\dots,\sigma_{4} of X2mX_{2m} as a free /2m\mathbb{Z}/2m\mathbb{Z}-module such that the subgroup kerφlX2mX2m\ker\varphi_{l}\cap X_{2m}\subset X_{2m} is contained in σ1,σ2X2m\langle\sigma_{1},\sigma_{2}\rangle\subset X_{2m}. Then σ~:=σ3\tilde{\sigma}:=\sigma_{3} satisfies the condition in this claim. ∎

Take σ~X\tilde{\sigma}\in X as in Appendix A and set σ:=φl(σ~)X^\sigma:=\varphi_{l}(\tilde{\sigma})\in\widehat{X}. Let π:YX\pi:Y\rightarrow X be the dual isogeny of the quotient X^X^/σ\widehat{X}\rightarrow\widehat{X}/\langle\sigma\rangle, where /2mσX^\mathbb{Z}/2m\mathbb{Z}\simeq\langle\sigma\rangle\subset\widehat{X} is the subgroup generated by σ\sigma. By a similar argument as the proof of [GP98, Lemma 2.6], we can check that K(πl)(/2m/(m2+m))2K(\pi^{*}l)\simeq(\mathbb{Z}/2m\mathbb{Z}\oplus\mathbb{Z}/(m^{2}+m)\mathbb{Z})^{\oplus 2}. Since mm is odd by assumption, 2m|m2+m2m|m^{2}+m holds and hence πl\pi^{*}l is of type (2m,m2+m)(2m,m^{2}+m). Thus there exists a polarization ll^{\prime} on YY of type (2,m+1)(2,m+1) such that πl=ml\pi^{*}l=ml^{\prime}. By [Ito20, Lemma 2.6],

β(l)<1m\displaystyle\beta(l)<\tfrac{1}{m}\ o1ml is IT(0)\displaystyle\Longleftrightarrow\ \mathcal{I}_{o}\left\langle\tfrac{1}{m}l\right\rangle\text{ is IT(0)}
πo1mπl=π1(o)l is IT(0)\displaystyle\Longleftrightarrow\ \pi^{*}\mathcal{I}_{o}\left\langle\tfrac{1}{m}\pi^{*}l\right\rangle=\mathcal{I}_{\pi^{-1}(o)}\left\langle l^{\prime}\right\rangle\text{ is IT(0)}
π1(o)L is IT(0),\displaystyle\Longleftrightarrow\ \mathcal{I}_{\pi^{-1}(o)}\otimes L^{\prime}\text{ is IT(0)},

where LL^{\prime} is a line bundle on YY representing ll^{\prime} with characteristic 0 with respect to some decomposition for ll^{\prime}. The last condition is equivalent to

h0(π1(o)+pL)=h0(L)#π1(o)=2(m+1)2m=2\displaystyle h^{0}(\mathcal{I}_{\pi^{-1}(o)+p}\otimes L^{\prime})=h^{0}(L^{\prime})-\#\pi^{-1}(o)=2(m+1)-2m=2

for any pYp\in Y, where π1(o)+p\pi^{-1}(o)+p is the parallel translation of π1(o)\pi^{-1}(o) by pp. Hence to show β(l)1/m\beta(l)\geqslant 1/m, it suffices to find a point pYp\in Y such that h0(π1(o)+pL)>2h^{0}(\mathcal{I}_{\pi^{-1}(o)+p}\otimes L^{\prime})>2.

Since LL^{\prime} is symmetric, /2\mathbb{Z}/2\mathbb{Z} acts on H0(L)H^{0}(L^{\prime}). Since LL^{\prime} is of type (2,m+1)(2,m+1), the dimension of the invariant part H0(L)+H^{0}(L^{\prime})^{+} is h0(L)/2+2=m+3h^{0}(L^{\prime})/2+2=m+3 by [BL04, Corollary 4.6.6]. Take εY\varepsilon^{\prime}\in Y such that 2εY2\varepsilon^{\prime}\in Y is a generator of kerπ/2m\ker\pi\simeq\mathbb{Z}/2m\mathbb{Z}. Then we have

π1(o)(2m1)ε=i=1m{(2i1)ε,(2i1)ε}.\pi^{-1}(o)-(2m-1)\varepsilon^{\prime}=\bigsqcup_{i=1}^{m}\{(2i-1)\varepsilon^{\prime},-(2i-1)\varepsilon^{\prime}\}.

For a section sH0(L)+s\in H^{0}(L^{\prime})^{+}, ss vanishes at (2i1)ε(2i-1)\varepsilon^{\prime} if and only if so does at (2i1)ε-(2i-1)\varepsilon^{\prime}. Hence sH0(L)+s\in H^{0}(L^{\prime})^{+} is contained in H0(π1(o)(2m1)εL)H^{0}(\mathcal{I}_{\pi^{-1}(o)-(2m-1)\varepsilon^{\prime}}\otimes L^{\prime}) if ss vanishes at mm points ε,3ε,,(2m1)ε\varepsilon^{\prime},3\varepsilon^{\prime},\dots,(2m-1)\varepsilon^{\prime}. Thus h0(π1(o)(2m1)εL)h^{0}(\mathcal{I}_{\pi^{-1}(o)-(2m-1)\varepsilon^{\prime}}\otimes L^{\prime}) is greater than or equal to

dimH0(L)+H0(π1(o)(2m1)εL)h0(L)+m=3\dim H^{0}(L^{\prime})^{+}\cap H^{0}(\mathcal{I}_{\pi^{-1}(o)-(2m-1)\varepsilon^{\prime}}\otimes L^{\prime})\geqslant h^{0}(L^{\prime})^{+}-m=3

and β(l)1/m\beta(l)\geqslant 1/m follows.

The last statement follows from § 4 (1). ∎

Remark \therem.

In the first version of this paper, the author wrote that it might be natural to guess β(1,d)β(1,d)\beta(1,d)\geqslant\beta(1,d^{\prime}) for ddd\leqslant d^{\prime}. However, this naive expectation is not true. In fact, a recent paper [Roj21] investigates β(l)\beta(l) for abelian surfaces using stability conditions, and improves the bound in § 4. For example, β(1,11)10/33\beta(1,11)\leqslant 10/33 holds by [Roj21, Theorem A]. Hence we have β(1,11)<1/3=β(1,32+3)=β(1,12)\beta(1,11)<1/3=\beta(1,3^{2}+3)=\beta(1,12).

Lemma \thelem.

Let m1m\geqslant 1 be an odd integer. Let (X,l)(X,l) be a polarized abelian surface of type (1,d)(1,d) such that d=(m+1)21d=(m+1)^{2}-1 or d=(m+1)22d=(m+1)^{2}-2. Then β(l)(m+1)/d\beta(l)\geqslant(m+1)/d holds.

In particular, β(l)=(m+1)/d\beta(l)=(m+1)/d holds if (X,l)(X,l) is general.

Proof.

Since (X,l)(X,l) is of type (1,d)(1,d), there exist an isogeny π:YX\pi:Y\rightarrow X and a principal polarization θ\theta on YY such that πl=dθ\pi^{*}l=d\theta and π1(o)=kerπ/d\pi^{-1}(o)=\ker\pi\simeq\mathbb{Z}/d\mathbb{Z}. As in the proof of Appendix A,

β(l)<m+1d\displaystyle\beta(l)<\tfrac{m+1}{d}\ om+1dl is IT(0)\displaystyle\Longleftrightarrow\ \mathcal{I}_{o}\left\langle\tfrac{m+1}{d}l\right\rangle\text{ is IT(0)}
πom+1dπl=π1(o)(m+1)θ is IT(0)\displaystyle\Longleftrightarrow\ \pi^{*}\mathcal{I}_{o}\left\langle\tfrac{m+1}{d}\pi^{*}l\right\rangle=\mathcal{I}_{\pi^{-1}(o)}\left\langle(m+1)\theta\right\rangle\text{ is IT(0)}
π1(o)𝒪Y((m+1)Θ) is IT(0),\displaystyle\Longleftrightarrow\ \mathcal{I}_{\pi^{-1}(o)}\otimes\mathcal{O}_{Y}((m+1)\Theta)\text{ is IT(0)},

where 𝒪Y(Θ)\mathcal{O}_{Y}(\Theta) is a line bundle representing θ\theta with characteristic 0 with respect to some decomposition for θ\theta. Hence to show β(l)(m+1)/d\beta(l)\geqslant(m+1)/d, it suffices to find pYp\in Y such that

(A.1) h0(π1(o)+p𝒪Y((m+1)Θ))>h0(𝒪Y((m+1)Θ))#π1(o)=(m+1)2d={1 if d=(m+1)212 if d=(m+1)22.\displaystyle\begin{aligned} h^{0}(\mathcal{I}_{\pi^{-1}(o)+p}\otimes\mathcal{O}_{Y}((m+1)\Theta))&>h^{0}(\mathcal{O}_{Y}((m+1)\Theta))-\#\pi^{-1}(o)\\ &=(m+1)^{2}-d=\begin{cases}1&\text{ if }d=(m+1)^{2}-1\\ 2&\text{ if }d=(m+1)^{2}-2.\end{cases}\end{aligned}

By [BL04, Corollary 4.6.6], the dimension of the /2\mathbb{Z}/2\mathbb{Z}-invariant part H0(𝒪Y((m+1)Θ)+H^{0}(\mathcal{O}_{Y}((m+1)\Theta)^{+} is (m+1)2/2+2(m+1)^{2}/2+2 since mm is odd. Let επ1(o)\varepsilon\in\pi^{-1}(o) be a generator of π1(o)/d\pi^{-1}(o)\simeq\mathbb{Z}/d\mathbb{Z}.

Case 1 : d=(m+1)21d=(m+1)^{2}-1. Since mm is odd, dd is odd and

π1(o)={oY}i=1d12{iε,iε}.\pi^{-1}(o)=\{o_{Y}\}\sqcup\bigsqcup_{i=1}^{\frac{d-1}{2}}\{i\varepsilon,-i\varepsilon\}.

As in the proof of Appendix A, a section sH0(𝒪Y((m+1)Θ)+s\in H^{0}(\mathcal{O}_{Y}((m+1)\Theta)^{+} is contained in H0(π1(o)𝒪Y((m+1)Θ))H^{0}(\mathcal{I}_{\pi^{-1}(o)}\otimes\mathcal{O}_{Y}((m+1)\Theta)) if ss vanishes at d+12=(m+1)22\frac{d+1}{2}=\frac{(m+1)^{2}}{2} points oY,ε,2ε,,d12εo_{Y},\varepsilon,2\varepsilon,\dots,\frac{d-1}{2}\varepsilon. Thus

dimH0(𝒪Y((m+1)Θ))+H0(π1(o)𝒪Y((m+1)Θ))\displaystyle\dim H^{0}(\mathcal{O}_{Y}((m+1)\Theta))^{+}\cap H^{0}(\mathcal{I}_{\pi^{-1}(o)}\otimes\mathcal{O}_{Y}((m+1)\Theta))\hskip 93.89409pt
h0(𝒪Y((m+1)Θ))+(m+1)22=2.\displaystyle\geqslant h^{0}(\mathcal{O}_{Y}((m+1)\Theta))^{+}-\frac{(m+1)^{2}}{2}=2.

Hence A.1 holds for p=oYp=o_{Y} and we have β(l)(m+1)/d\beta(l)\geqslant(m+1)/d.

Case 2: d=(m+1)22d=(m+1)^{2}-2. Since mm is odd, dd is even. Take εY\varepsilon^{\prime}\in Y such that 2ε=ε2\varepsilon^{\prime}=\varepsilon. Then

π1(o)(d1)ε=i=1d2{(2i1)ε,(2i1)ε}.\pi^{-1}(o)-(d-1)\varepsilon^{\prime}=\bigsqcup_{i=1}^{\frac{d}{2}}\{(2i-1)\varepsilon^{\prime},-(2i-1)\varepsilon^{\prime}\}.

Hence sH0(𝒪Y((m+1)Θ)+s\in H^{0}(\mathcal{O}_{Y}((m+1)\Theta)^{+} is contained in H0(π1(o)(d1)ε𝒪Y((m+1)Θ))H^{0}(\mathcal{I}_{\pi^{-1}(o)-(d-1)\varepsilon^{\prime}}\otimes\mathcal{O}_{Y}((m+1)\Theta)) if ss vanishes at d2=(m+1)222\frac{d}{2}=\frac{(m+1)^{2}-2}{2} points ε,3ε,5ε,,(d1)ε\varepsilon^{\prime},3\varepsilon^{\prime},5\varepsilon^{\prime},\dots,(d-1)\varepsilon^{\prime}. Thus

dimH0(𝒪Y((m+1)Θ))+H0(π1(o)(d1)ε𝒪Y((m+1)Θ))\displaystyle\dim H^{0}(\mathcal{O}_{Y}((m+1)\Theta))^{+}\cap H^{0}(\mathcal{I}_{\pi^{-1}(o)-(d-1)\varepsilon^{\prime}}\otimes\mathcal{O}_{Y}((m+1)\Theta))\hskip 56.9055pt
h0(𝒪Y((m+1)Θ))+(m+1)222=3.\displaystyle\geqslant h^{0}(\mathcal{O}_{Y}((m+1)\Theta))^{+}-\frac{(m+1)^{2}-2}{2}=3.

Hence A.1 holds for p=(d1)εp=-(d-1)\varepsilon^{\prime} and we have β(l)(m+1)/d\beta(l)\geqslant(m+1)/d.

If dm2+m+1d\geqslant m^{2}+m+1, the last statement of this proposition follows from the first statement of this proposition and § 4 (2). Since d=(m+1)21=m2+2md=(m+1)^{2}-1=m^{2}+2m or d=(m+1)22=m2+2m1d=(m+1)^{2}-2=m^{2}+2m-1, the condition dm2+m+1d\geqslant m^{2}+m+1 does not hold only when m=1m=1 and d=2d=2. For m=1m=1 and d=2d=2, we have β(l)=1=(m+1)/d\beta(l)=1=(m+1)/d by the case p=1p=-1 of Theorem 1.3 since ll is not basepoint free. Thus the last statement of this lemma holds. ∎

Example \theex.

By § 4 and Appendix A, we have the following computation or estimation of β(1,d)\beta(1,d) for small dd.

dd 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
β(1,d)\beta(1,d) 1 1 23\dfrac{2}{3} 12\dfrac{1}{2} 12\dfrac{1}{2} 12\dfrac{1}{2} 37\leqslant\dfrac{3}{7} 38\leqslant\dfrac{3}{8} 13\dfrac{1}{3} 13\leqslant\dfrac{1}{3} 13\leqslant\dfrac{1}{3} 13\dfrac{1}{3} 413\leqslant\dfrac{4}{13} 27\dfrac{2}{7} 415\dfrac{4}{15} 14\dfrac{1}{4}
Table 1. β(1,d)\beta(1,d) for d16d\leqslant 16

For d=1,2d=1,2, β(1,d)=1\beta(1,d)=1 follows from the case p=1p=-1 of Theorem 1.3 since ll is not basepoint free. For d=3,14,15d=3,14,15, β(1,d)\beta(1,d) is computed by Appendix A. For d=4,9,16d=4,9,16, β(1,d)=1/d\beta(1,d)=1/\sqrt{d} follows from § 4 (1). For d=5,6d=5,6, we have β(1,d)1/2\beta(1,d)\leqslant 1/2 by § 4 (1). Furthermore, β(1,d)1/2\beta(1,d)\geqslant 1/2 follows from the case p=0p=0 of Theorem 1.3 since ll is not projectively normal for d6d\leqslant 6 by dimH0(X,L2)>dimSym2H0(X,L)\dim H^{0}(X,L^{\otimes 2})>\dim\operatorname{Sym}^{2}H^{0}(X,L). For d=12d=12, we have β(1,d)=1/3\beta(1,d)=1/3 by Appendix A. For the rest dd, the upper bounds of β(1,d)\beta(1,d) follow from § 4.

References

  • [Ago17] Daniele Agostini, A note on homogeneous ideals of abelian surfaces, Bull. Lond. Math. Soc. 49 (2017), no. 2, 220–225. MR 3656291
  • [BL04] Christina Birkenhake and Herbert Lange, Complex abelian varieties, second ed., Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 302, Springer-Verlag, Berlin, 2004. MR 2062673
  • [Cau20a] Federico Caucci, The basepoint-freeness threshold and syzygies of abelian varieties, Algebra Number Theory 14 (2020), no. 4, 947–960. MR 4114062
  • [Cau20b] Federico Caucci, The basepoint-freeness threshold, derived invariants of irregular varieties, and stability of syzygy bundles, Ph.D. thesis, 2020.
  • [DHS94] O. Debarre, K. Hulek, and J. Spandaw, Very ample linear systems on abelian varieties, Math. Ann. 300 (1994), no. 2, 181–202. MR 1299059
  • [Eis05] David Eisenbud, The geometry of syzygies, Graduate Texts in Mathematics, vol. 229, Springer-Verlag, New York, 2005, A second course in commutative algebra and algebraic geometry. MR 2103875
  • [EL12] Lawrence Ein and Robert Lazarsfeld, Asymptotic syzygies of algebraic varieties, Invent. Math. 190 (2012), no. 3, 603–646. MR 2995182
  • [FG04] Luis Fuentes García, Projective normality of abelian surfaces of type (1,2d)(1,2d), Manuscripta Math. 114 (2004), no. 3, 385–390. MR 2076454
  • [FG05] by same author, Some results about the projective normality of abelian varieties, Arch. Math. (Basel) 85 (2005), no. 5, 409–418. MR 2181770
  • [GP98] Mark Gross and Sorin Popescu, Equations of (1,d)(1,d)-polarized abelian surfaces, Math. Ann. 310 (1998), no. 2, 333–377. MR 1602020
  • [Ito18] Atsushi Ito, A remark on higher syzygies on abelian surfaces, Comm. Algebra 46 (2018), no. 12, 5342–5347. MR 3923760
  • [Ito20] by same author, Basepoint-freeness thresholds and higher syzygies on abelian threefolds, arXiv:2008.10272v2, 2020.
  • [Iye99] Jaya N. Iyer, Projective normality of abelian surfaces given by primitive line bundles, Manuscripta Math. 98 (1999), no. 2, 139–153. MR 1667600
  • [Iye03] by same author, Projective normality of abelian varieties, Trans. Amer. Math. Soc. 355 (2003), no. 8, 3209–3216. MR 1974682
  • [Jia21] Zhi Jiang, Cohomological rank functions and syzygies of abelian varieties, Mathematische Zeitschrift (2021).
  • [JP20] Zhi Jiang and Giuseppe Pareschi, Cohomological rank functions on abelian varieties, Ann. Sci. Éc. Norm. Supér. (4) 53 (2020), no. 4, 815–846. MR 4157109
  • [KL19] Alex Küronya and Victor Lozovanu, A Reider-type theorem for higher syzygies on abelian surfaces, Algebr. Geom. 6 (2019), no. 5, 548–570. MR 4009173
  • [Laz90] Robert Lazarsfeld, Projectivité normale des surfaces abéliennes, (written by O. Debarre), preprint Europroj 14, Nice, 1990.
  • [Laz04] by same author, Positivity in algebraic geometry. I, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics], vol. 48, Springer-Verlag, Berlin, 2004, Classical setting: line bundles and linear series. MR 2095471
  • [Loz18] Victor Lozovanu, Singular divisors and syzygies of polarized abelian threefolds, arXiv:1803.08780, 2018.
  • [LPP11] Robert Lazarsfeld, Giuseppe Pareschi, and Mihnea Popa, Local positivity, multiplier ideals, and syzygies of abelian varieties, Algebra Number Theory 5 (2011), no. 2, 185–196. MR 2833789
  • [Roj21] Andrés Rojas, The basepoint-freeness threshold of a very general abelian surface, arXiv:2107.01896, 2021.
  • [Sil09] Joseph H. Silverman, The arithmetic of elliptic curves, second ed., Graduate Texts in Mathematics, vol. 106, Springer, Dordrecht, 2009. MR 2514094