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

Involutions of curves in abelian surfaces and their Jacobians

Katrina Honigs  and  Pijush Pratim Sarmah Department of Mathematics
Simon Fraser University
8888 University Drive
Burnaby, BC, V5A 1S6
Canada
Abstract.

We examine étale covers of genus two curves that occur in the linear system of a polarizing line bundle of type (1,d)(1,d) on a complex abelian surface. We give results counting fixed points of involutions on such curves as well as decomposing their Jacobians into isogenous products.

In [ES93], Ekedahl and Serre examine smooth, projective complex curves whose Jacobians are completely decomposable, that is, isogenous to a product of elliptic curves. They ask if it is possible to find such curves of each genus, and if the genus of such curves is bounded. Much progress has been made on this question — see for example [PS25] for recent updates — but it remains open.

In their paper, Ekedahl and Serre give examples of curves that are unramified abelian covers of genus 22 curves HH, where JHJH is isogenous to a product of elliptic curves. We examine more generally the decomposability of the Jacobians of curves CC in abelian surfaces AA that are unramified abelian covers of genus 2 curves, though we limit our scope to curves in the linear system of a line bundle LL defining a primitive polarization (1.1). We prove the following two general theorems on decomposing the Jacobians of these curves:

Theorem 0.1 (4.2).

Let CC be as in 1.1 for X=xK(L)X=\langle x\rangle\leq K(L) a cyclic group. We have the following isogeny relations:

  1. (a)

    When dd is odd: JCA×JC/12J_{C}\sim A\times J^{2}_{C/\langle-1\rangle}.

  2. (b)

    When dd is even: JCA×JC/1×JC/1txJ_{C}\sim A\times J_{C/\langle-1\rangle}\times J_{C/\langle-1\circ t_{x}\rangle}.

Theorem 0.2 (4.3).

Let pp be an odd prime and d=p2d=p^{2}. Let CC be as in 1.1 for

(/p)2XK(L).(\mathbb{Z}/p\mathbb{Z})^{2}\cong X\leq K(L).

Then we have the following isogeny relations, where each CiC_{i} is a curve of genus p12\frac{p-1}{2} that is a quotient of CC by [1][-1] and a translation:

JCA×i=1p+1JCi2,JC/1i=1p+1JCi.J_{C}\sim A\times\prod_{i=1}^{p+1}J_{C_{i}}^{2},\quad J_{C/\langle-1\rangle}\sim\prod_{i=1}^{p+1}J_{C_{i}}.

0.2 also gives a new proof of the main result in [CLR21], which demonstrates that there exist curves whose Jacobian is isogenous to an arbitrarily high number of Jacobians of the same dimension. We also give a class of examples where JCJ_{C} contains an elliptic factor.

Our primary method for decomposing Jacobians is applying [KR89, Theorem B] of Kani and Rosen to the automorphism groups of these curves. In order to obtain our results on Jacobian decompositions, we prove general results on the numbers of fixed points of involutions [1]tx[-1]\circ t_{x} on CC, which control how low the genus of a quotient curve can be.

Theorem 0.3 (2.5).

Let dd be odd and CC a curve as in 1.1. Any involution [1]tx[-1]\circ t_{x} on CC has 66 fixed points.

In [BO19], Borówka and Ortega completed an explicit construction of all possible smooth hyperelliptic curves in (generic) abelian surfaces by producing a curve that is a Klein cover of a genus 22 curve on a (1,4)(1,4)-polarized abelian surface. We generalize this construction by giving criteria for the existence of an involution [1]tx[-1]\circ t_{x} that has 1212 fixed points.

Theorem 0.4 (Propositions 3.1, 3.2).

Let dd be even and CC a curve as in 1.1, which is an XX-covering of a smooth genus 22 curve for XK(L)X\leq K(L). There is an xXx\in X so that the involution [1]tx[-1]\circ t_{x} on CC has 1212 fixed points if and only if the Sylow 22-subgroup of XX is not cyclic.

Outline

In section 1, we begin by presenting the curves CC that we will study. In section 2, we relate fixed points of involutions [1]tx[-1]\circ t_{x} on CC to translations of the line bundle LL, and conclude by proving 0.3. In section 3, we prove 0.4. We then take the opportunity to briefly review the smooth hyperelliptic curves in |L||L|; when d=2,4d=2,4, they arise from curves made with 1.1. In section 4, we give our results on Jacobian splittings, and recover some of the results of Ekedahl and Serre using [KR89, Theorem B] and our results on automorphisms.

Notation

The polarization isogeny given by LL is denoted by φL:AA\varphi_{L}:A\to A^{\vee}, ataLL1a\mapsto t_{a}^{*}L\otimes L^{-1} and we write its kernel as:

K(L):=kerφL(/d)2.K(L):=\ker\varphi_{L}\cong(\mathbb{Z}/d\mathbb{Z})^{2}.

We write the involution given by the inverse group law map on AA as [1][-1]. We denote the set of fixed points of an involution ι\iota on CC by FixC(ι)\operatorname{Fix}_{C}(\iota) and their number by #FixC(ι)\#\operatorname{Fix}_{C}(\iota).

Acknowledgements

The authors thank Jen Paulhus and Nils Bruin for helpful conversations, as well as P. Borówka, particularly for showing us how to extend our methods to some of the examples in section 4.2. K. H. is supported by an NSERC Discovery Grant.

1. Construction

We now construct the curves that we will analyze in this paper. Let AA be an abelian surface that is (1,d)(1,d)-polarized by a Néron–Severi class ll.

Ample line bundles that have the same Néron–Severi class differ by translation. However, because we wish to identify the number of ramification points of specific automorphisms of curves in our exposition, we will fix a particular line bundle representative. Considering AA as a complex torus V/ΛV/\Lambda, we fix a decomposition V1V2V_{1}\oplus V_{2} for ll and select the line bundle LL of characteristic 0 (see [BL04, §3.1]) with respect to this decomposition.

Construction 1.1.

Let XX be a subgroup of K(L)K(L) of order dd and consider the quotient π:AA/X\pi:A\to A/X. By Lemma 1.3 below, A/XA/X is principally polarized by the line bundle πL=:M\pi_{*}L=:M. Furthermore by [BL04, Prop. 6.5.2], we may choose a decomposition of A/XA/X for the Néron–Severi class of MM that is compatible with the decomposition of AA for ll (replacing the decomposition of AA if needed) and so that both LL and MM have characteristic 0 with respect to these decompositions.

If A/XA/X is the Jacobian of a smooth genus 22 curve (for instance if AA is simple this will be the case for any choice of XX), then the curve we construct is the preimage C:=π1(H)C:=\pi^{-1}(H) in |L||L|.

Remark 1.2.

For each xXx\in X, the translation txt_{x} is an automorphism of the curve CC. Since [1][-1] is an involution of HH, it is an involution of CC as well. Thus Aut(C)\operatorname{Aut}(C) contains a subgroup isomorphic to 1X\langle-1\rangle\rtimes X whose elements are of the form [1]tx[-1]\circ t_{x} and txt_{x} for xXx\in X.

Lemma 1.3.

Let XX be a subgroup of K(L)K(L) of order dividing dd and consider the quotient isogeny π:AA/X\pi:A\to A/X. Then A/XA/X is (1,d|X|)(1,\frac{d}{|X|})-polarized by the pushforward of the polarization φL\varphi_{L} by π\pi.

Proof.

First, we note ker(π)ker(φL)\ker(\pi)\subseteq\ker(\varphi_{L}). In order to take the pushforward of φL\varphi_{L} by π\pi, we must check that XX is isotropic with respect to the commutator pairing eLe^{L} of the theta group of LL. We may write a decomposition of K(L)K(L) with respect to this pairing as /d×μd\mathbb{Z}/d\mathbb{Z}\times\mu_{d}, noting μdHom(/d,)\mu_{d}\simeq\operatorname{Hom}(\mathbb{Z}/d\mathbb{Z},\mathbb{C}^{*}), with generators 11 and ζ\zeta (see [BL04, §6.3], [EvdGM, (8.21)]). Then:

eL((x,χ),(x,χ))=χ(x)χ(x)1.e^{L}((x,\chi),(x^{\prime},\chi^{\prime}))=\chi^{\prime}(x)\chi(x^{\prime})^{-1}.

As a subgroup of K(L)K(L), XX is generated by some (a,ζα)(a,\zeta^{\alpha}), (b,ζβ)(b,\zeta^{\beta}) of orders d1d_{1}, d2d_{2} so that d1d2=|X|d_{1}d_{2}=|X|. The pairing of these generators is ζaβbα=1\zeta^{a\beta-b\alpha}=1: since d1ad_{1}a and d2βd_{2}\beta are both multiples of dd, aβa\beta is a multiple of dd (as is bα)b\alpha). We may use similar reasoning to show that any two elements of XX pair trivially.

The pushforward polarization πφL:A/X(A/X)\pi_{*}\varphi_{L}:A/X\to(A/X)^{\vee} fills in the following diagram and corresponds to the line bundle πL\pi_{*}L [Mum70, Corollary, p. 231] (moreover, LππLL\simeq\pi^{*}\pi_{*}L):

A\textstyle{A\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}φL\scriptstyle{\varphi_{L}}π\scriptstyle{\pi}A\textstyle{A^{\vee}}A/X\textstyle{A/X\ignorespaces\ignorespaces\ignorespaces\ignorespaces}φπL\scriptstyle{\varphi_{\pi_{*}L}}(A/X)\textstyle{(A/X)^{\vee}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}π\scriptstyle{\pi^{\vee}} (1.1)

That φπL\varphi_{\pi_{*}L} is a (1,d|X|)(1,\frac{d}{|X|})-polarization follows, for instance, from an analysis of the kernels in the above diagram. ∎

Remark 1.4.

Lemma 1.3 does not hold for XX where |X|>d|X|>d since any maximal groups isotropic with respect to eLe^{L} is of order dd [Mum70, p. 233, Thm. 4].

2. Symmetric linear systems and ramification of [1][-1]

In this section, we analyze the number of fixed points of involutions [1]tx[-1]\circ t_{x} by relating them to properties of symmetric line bundles. We begin by reviewing properties of such line bundles.

Let AA be an abelian surface that is (1,d)(1,d)-polarized by a symmetric line bundle LL.

Since LL is symmetric, there is an isomorphism [1]LL[-1]^{*}L\simeq L and thus [1][-1]^{*} is a linear involution on H0(A,L)H^{0}(A,L). Its eigenvalues are 11 and 1-1 and we denote the associated eigenspaces by H0(A,L)+H^{0}(A,L)^{+} and H0(A,L)H^{0}(A,L)^{-}, respectively. The involution [1][-1] of AA restricts to an involution of each curve whose class is contained in H0(A,L)±H^{0}(A,L)^{\pm}.

The dimensions of H0(A,L)±H^{0}(A,L)^{\pm} are affected by the parity of LL and whether it has a symmetric theta structure (sts).

The Weil pairing on AA gives a non-degenerate strictly alternating form on A[2]A[2] as an 𝔽2\mathbb{F}_{2} vector space. The quadratic forms associated to this pairing form a principal homogeneous space over A[2]A[2] and correspond to the symmetric line bundles with Néron–Severi class [L][L]. These line bundles are called even or odd depending on the parity of the corresponding quadratic form. We examine further specifics on determining this parity in section 2.0.1.

A symmetric theta structure is an isomorphism between the extended theta and extended Heisenberg groups. Such an isomorphism exists if and only if a constraint on the dimension of h0(A,L)+h^{0}(A,L)^{+} or h0(A,L)h^{0}(A,L)^{-} is satisfied (maximality, see [BL04, Exercise 6.10(10)]).

A line bundle chosen to have characteristic 0 with respect to a decomposition, as in 1.1 has an sts and is even.

The following proposition summarizes the dimensions and base loci of H0(A,L)±H^{0}(A,L)^{\pm}. The base loci are contained in A[2]A[2] and we call them A[2]A[2]^{\mp}, named for the points where the quadratic form associated to LL takes the values 1-1 and +1+1, respectively. For any smooth curve CH0(A,L)±C\in H^{0}(A,L)^{\pm}, FixC[1]=A[2]\operatorname{Fix}_{C}[-1]=A[2]^{\mp}.

Proposition 2.1 ([BL04, Prop. 4.7.5, Exercise 6.10(10)]).

Let AA be an abelian surface that is (1,d)(1,d)-polarized by a symmetric line bundle LL.

If dd is odd, then LL has an sts and the linear eigensystems of LL under the action of [1][-1]^{*} have the following properties:

LL even: h0(A,L)+\displaystyle h^{0}(A,L)^{+} =d+12,\displaystyle=\tfrac{d+1}{2}, h0(A,L)\displaystyle h^{0}(A,L)^{-} =d12,\displaystyle=\tfrac{d-1}{2},
#A[2]\displaystyle\#A[2]^{-} =6,\displaystyle=6, #A[2]+\displaystyle\#A[2]^{+} =10.\displaystyle=10.
LL odd: h0(A,L)+\displaystyle h^{0}(A,L)^{+} =d12,\displaystyle=\tfrac{d-1}{2}, h0(A,L)\displaystyle h^{0}(A,L)^{-} =d+12,\displaystyle=\tfrac{d+1}{2},\hskip 42.67912pt
#A[2]\displaystyle\#A[2]^{-} =10,\displaystyle=10, #A[2]+\displaystyle\#A[2]^{+} =6.\displaystyle=6.

If dd is even, we distinguish more cases, depending on whether LL has an sts.

LL sts, even: h0(A,L)+\displaystyle h^{0}(A,L)^{+} =d2+1,\displaystyle=\tfrac{d}{2}+1, h0(A,L)\displaystyle h^{0}(A,L)^{-} =d21.\displaystyle=\tfrac{d}{2}-1.
#A[2]\displaystyle\#A[2]^{-} =4,\displaystyle=4, #A[2]+\displaystyle\#A[2]^{+} =12.\displaystyle=12.
LL sts, odd: h0(A,L)+\displaystyle h^{0}(A,L)^{+} =d21,\displaystyle=\tfrac{d}{2}-1, h0(A,L)\displaystyle h^{0}(A,L)^{-} =d2+1,\displaystyle=\tfrac{d}{2}+1,
#A[2]\displaystyle\#A[2]^{-} =12,\displaystyle=12, #A[2]+\displaystyle\#A[2]^{+} =4.\displaystyle=4.
LL no sts: h0(A,L)+\displaystyle h^{0}(A,L)^{+} =d2,\displaystyle=\tfrac{d}{2}, h0(A,L)\displaystyle h^{0}(A,L)^{-} =d2,\displaystyle=\tfrac{d}{2},
#A[2]\displaystyle\#A[2]^{-} =8,\displaystyle=8, #A[2]+\displaystyle\#A[2]^{+} =8.\displaystyle=8.

Given a curve CC as in 1.1, for each xXx\in X, [1]tx[-1]\circ t_{x} is an involution of CC. The following lemma shows that the fixed locus of [1]tx[-1]\circ t_{x} is in bijection with the fixed locus of [1][-1] acting on a translated curve C+yC+y, which lies in the linear system of tyLt_{-y}^{*}L. Subsequent results in this section allow us to determine the properties of tyLt_{-y}^{*}L so that we may apply 2.1 to find #FixC([1]tx)\#\operatorname{Fix}_{C}([-1]\circ t_{x}).

Lemma 2.2.

Let CC be a curve contained in an abelian surface AA. Suppose there are x,aAx,a\in A so that [1]ta+x[-1]\circ t_{a+x} is an involution of CC. For any yAy\in A such that 2y=x2y=x, [1]ta[-1]\circ t_{a} is an involution of the translated curve C+yC+y and there is a bijection of sets:

FixC(1tx+a)FixC+y(1ta).\operatorname{Fix}_{C}(-1\circ t_{x+a})\leftrightarrow\operatorname{Fix}_{C+y}(-1\circ t_{a}).
Proof.

For any point cCc\in C, by our assumptions, cxaC-c-x-a\in C. Any point in C+yC+y may be written as c+yc+y for some cCc\in C. Applying [1]ta[-1]\circ t_{a} gives cya-c-y-a, and we note (cya)y=caxC(-c-y-a)-y=-c-a-x\in C, and thus cya-c-y-a is in C+yC+y.

Any point cCc\in C is fixed by [1]tx+a[-1]\circ t_{x+a} if c=cxac=-c-x-a and c+yc+y is fixed by [1]ta[-1]\circ t_{a} if c+y=cyac+y=-c-y-a. Since 2y=x2y=x, these conditions are identical. Thus the bijection between these fixed loci is given by translation by yy. ∎

Lemma 2.3.
  1. (a)

    For any yAy\in A, tyLt_{y}^{*}L is symmetric if and only if 2yK(L)2y\in K(L).

  2. (b)

    If LL has an sts, then for any yAy\in A, tyLt_{y}^{*}L has an sts if and only if tyLtzLt_{y}^{*}L\simeq t_{z}^{*}L (i.e. φL(y)=φL(z)\varphi_{L}(y)=\varphi_{L}(z)) for some zA[2]z\in A[2].

Proof.

(a) Since LL is symmetric and 2y=x2y=x, then tyLt_{y}^{*}L is symmetric:

tytyLtxLLtyLtyL[1]tyL[1]tyLty[1]LtyLt_{y}^{*}t_{y}^{*}L\simeq t_{x}^{*}L\simeq L\Rightarrow t_{y}^{*}L\simeq t_{-y}^{*}L\Rightarrow[-1]^{*}t_{y}^{*}L\simeq[-1]^{*}t_{-y}^{*}L\simeq t_{y}^{*}[-1]^{*}L\simeq t_{y}^{*}L

The elements of the quotient group y2yK(L)/K(L)\langle y\!\mid\!2y\in K(L)\rangle/K(L) correspond to distinct line bundles tyLt_{y}^{*}L and has order |A[2]||A[2]|. Thus we obtain all 16=|A[2]|16=|A[2]| symmetric line bundles in this way.

(b) See the proof of [BL04, Theorem 6.9.5]. Moreover, the symmetric line bundles admitting symmetric theta structures correspond to the cosets A[2]/(K(L)A[2])A[2]/(K(L)\cap A[2]). ∎

Lemma 2.4.

Let CC be a curve produced using 1.1 with the group XK(L)X\leq K(L).

  1. (a)

    [C]H0(A,L)[C]\in H^{0}(A,L) has eigenvalue 11 with respect to action of [1][-1]^{*}.

  2. (b)

    Let aAa\in A such that 2aX2a\in X and let MM be the principal polarization of A/XA/X produced in 1.1. The eigenvalue of [Ca]H0(A,taL)[C-a]\in H^{0}(A,t_{a}^{*}L) is 11 or 1-1 if tπ(a)Mt_{\pi(a)}^{*}M is even or odd, respectively.

Proof.

(a) Let CC be a curve produced using 1.1 with the subgroup XK(L)X\leq K(L). Let π:AA/X\pi:A\to A/X be the quotient isogeny. Then M:=πLM:=\pi_{*}L principally polarizes A/XA/X and let HA/XH\subset A/X be the genus 22 curve so that M𝒪(H)M\simeq\mathcal{O}(H).

Both LL and MM both have characteristic 0 with respect to compatible decompositions. In particular, MM is even. Since π\pi commmutes with [1][-1], and the following pullback maps [H][H] to [C][C], [C][C] must also have an eigenvalue of 11 with respect to [1][-1]^{*}:

π:H0(JH,M)H0(A,L)\pi^{*}:H^{0}(J_{H},M)\to H^{0}(A,L) (2.1)

(b) Let aAa\in A be such that 2aX2a\in X. Then tπ(a)Mt_{\pi(a)}^{*}M is symmetric and since XK(L)X\leq K(L), by Lemma 2.3(a), taLt_{a}^{*}L is symmetric. The following pullback maps [Hπ(a)][H-\pi(a)] to [Ca][C-a] and is compatible with [1][-1]^{*}:

π:H0(A/X,tπ(a)M)H0(A,taL).\pi^{*}:H^{0}(A/X,t_{\pi(a)}^{*}M)\to H^{0}(A,t_{a}^{*}L). (2.2)

Thus, the eigenvalue of [Ca][C-a] under the action of [1][-1]^{*} is 11 if tπ(a)Mt_{\pi(a)}^{*}M is even and 1-1 if it is odd. ∎

Let CC be any curve in 1.1. Combining 2.1 and Lemma 2.2, we may find #FixC([1]tx)\#\operatorname{Fix}_{C}([-1]\circ t_{x}) by producing yAy\in A so that 2y=x2y=x and determining whether tyLt_{-y}^{*}L has an sts and if so, determining the parity of tyLt_{-y}^{*}L and the eigenvalue of [Cy]H0(A,L)[C-y]\in H^{0}(A,L) under the action of [1][-1]^{*}. We may determine whether tyLt_{-y}^{*}L has an sts using the criteria in Lemma 2.3(b) and by Lemma 2.4, we may determine the eigenvalue of [Cy][C-y] given the parity of tπ(a)Mt_{\pi(a)}^{*}M. We use this line of reasoning to distinguish several cases in the following propositions.

Proposition 2.5.

Let dd be odd and CC be a curve as in 1.1. For any xXx\in X, the action of [1]tx[-1]\circ t_{x} on CC has 66 fixed points.

Proof.

Since the order of xx is odd, we may choose yxy\in\langle x\rangle of the same order as xx so that 2y=x2y=x. Then, yXK(L)y\in X\leq K(L), tyLLt_{-y}^{*}L\simeq L and π(y)=0\pi(-y)=0, so tπ(y)MMt_{\pi(y)}^{*}M\simeq M. Both LL and MM are even, so

#FixC+y[1]=#FixC([1]tx)=6.\#\operatorname{Fix}_{C+y}[-1]=\#\operatorname{Fix}_{C}([-1]\circ t_{x})=6.\hfill\qed
Proposition 2.6.

Let dd be even and CC be a curve as in 1.1. For any xXx\in X, one of the following mutually exclusive conditions applies to Fix([1]tx)\operatorname{Fix}([-1]\circ t_{x}).

Let nn be the order xXx\in X and 2m2^{m} the highest power of 22 dividing dd.

  1. (a)

    Suppose that for all yAy\in A such that 2y=x2y=x, yK(L)y\not\in K(L). Then, #Fix([1]tx)=8\#\operatorname{Fix}([-1]\circ t_{x})=8. This case applies, for instance, if 2m|n2^{m}|n.

Now, suppose there exists yAy\in A such that 2y=x2y=x and yK(L)y\in K(L).

  1. (b)

    If tπ(y)Mt_{\pi(-y)}^{*}M is even, then #Fix([1]tx)=4\#\operatorname{Fix}([-1]\circ t_{x})=4. This case applies, for instance, if yXy\in X (hence tπ(y)M=Mt_{\pi(y)}^{*}M=M), which must occur if nn is odd or if nn is even and xx is contained in a cyclic subgroup of XX with order 2n2n.

  2. (c)

    If tπ(y)Mt_{\pi(-y)}^{*}M is odd, then #Fix([1]tx)=12\#\operatorname{Fix}([-1]\circ t_{x})=12.

Proof.

(a) Let yAy\in A such that 2y=x2y=x. The condition for part (a) is equivalent to the statement that for any zA[2]z\in A[2], zyK(L)z-y\not\in K(L). By Lemma 2.3(b), tyLt_{-y}^{*}L does not have a symmetric theta structure, hence #Fix([1]tx)=8\#\operatorname{Fix}([-1]\circ t_{x})=8.

This condition applies if 2m|n2^{m}|n: For any yAy\in A such that 2y=x2y=x, yy has order 2n2n and therefore cannot be contained in K(L)K(L).

(b) Since yK(L)y\in K(L), tyLLt_{-y}^{*}L\simeq L has an sts and is even, and the result follows.

If nn is odd, then there is some yxy\in\langle x\rangle such that 2y=x2y=x, so yXy\in X. If nn is even but xx is contained in a cyclic subgroup of XX of order 2n2n then we may take yy to be its generator.

(c) Under the above conditions, tyLLt_{-y}^{*}L\simeq L has an sts and is even, so the result follows. ∎

2.0.1. Even and odd

To apply 2.6, we examine how to determine the parity of tπ(y)Mt_{\pi(-y)}^{*}M for yK(L)y\in K(L) such that 2yX2y\in X.

In 1.1, we have fixed a decomposition of A/XA/X for the Néron–Severi class of MM. The induced decomposition on the 22-torsion subgroup is a sum of maximal isotropic subgroups of (A/X)[2](A/X)[2] with respect to the Weil pairing:

(A/X)[2]Z1Z2.(A/X)[2]\cong Z_{1}\oplus Z_{2}.

This decomposition allows us to determine the parity of tzMt_{z}^{*}M for any z(A/X)[2]z\in(A/X)[2], which is even or odd if the quadratic form qM(z)q_{M}(z) is 11 or 1-1, respectively. We may choose dual symplectic bases {u1,u2}\{u_{1},u_{2}\} and {v1,v2}\{v_{1},v_{2}\} of Z1Z_{1} and Z2Z_{2} as 𝔽2\mathbb{F}_{2}-vector spaces so that ui,vj=(1)δij\langle u_{i},v_{j}\rangle=(-1)^{\delta_{ij}}. We may write z=u+vz=u+v for uZ1u\in Z_{1}, vZ2v\in Z_{2}, and then qM(z)=u,vq_{M}(z)=\langle u,v\rangle.

However, if z=π(y)z=\pi(-y) for some yK(L)y\in K(L) (as in the cases we treat in 2.6), then it is contained in a smaller subspace of (A/X)[2](A/X)[2]. The decomposition of AA for ll induces a decomposition of K(L)K(L) as follows:

K(L)=K1K2=k1k2.K(L)=K_{1}\oplus K_{2}=\langle k_{1}\rangle\oplus\langle k_{2}\rangle.

The image of K(L)K(L) in A/XA/X is π(K(L))=π(k1)π(k2)\pi(K(L))=\langle\pi(k_{1})\rangle\oplus\langle\pi(k_{2})\rangle, which, by the compatibility assumptions of 1.1, respects the decomposition of A/XA/X with respect to MM. Since π(K(L))\pi(K(L)) is generated by two cyclic groups, its intersection with (A/X)[2](A/X)[2] (if nontrivial) is contained in two cyclic groups:

π(K(L))(A/X)[2]w1w2\pi(K(L))\cap(A/X)[2]\subseteq\langle w_{1}\rangle\oplus\langle w_{2}\rangle

for some wiw_{i} such that /2wiZi\mathbb{Z}/2\mathbb{Z}\cong\langle w_{i}\rangle\leq Z_{i} and w1,w2=1\langle w_{1},w_{2}\rangle=-1. Then we have the following result:

If π(y)\pi(-y) is 0, w1w_{1}, or w2w_{2}, then tπ(y)Mt^{*}_{\pi(-y)}M is even.
If π(y)\pi(-y) is w1+w2w_{1}+w_{2}, then tπ(y)Mt^{*}_{\pi(-y)}M is odd.

3. Counting fixed points of [1]tx[-1]\circ t_{x}

In this section, let CC be as in 1.1 and dd be even. We give further results on #FixC([1]tx)\#\operatorname{Fix}_{C}([-1]\circ t_{x}) for xXx\in X, particularly investigating the unusual case where the number of fixed points is 1212, and then apply these results to review the number of smooth hyperelliptic curves in LL.

3.1. When can #FixC([1]tx)=12\#\operatorname{Fix}_{C}([-1]\circ t_{x})=12?

In the first result in this section, we see a situation where such an involution cannot occur, and in our second result, we characterize such involutions. Let 2m2^{m} be the highest power of 22 dividing dd. Let X2X_{2} be the Sylow 22-subgroup of XX. We refer to the projection of any xXx\in X to X2X_{2} as the 22-primary part of xx.

Proposition 3.1.

If X2X_{2} is cyclic, then, for any xXx\in X:

  1. (a)

    #FixC([1]tx)=8\#\operatorname{Fix}_{C}([-1]\circ t_{x})=8 if the 22-primary part of xx generates X2X_{2}.

  2. (b)

    #FixC([1]tx)=4\#\operatorname{Fix}_{C}([-1]\circ t_{x})=4 if the 22-primary part of xx does not generate X2X_{2}.

Proof.

The order of X2X_{2} is 2m2^{m}, which is also the highest order of any elements in the 22-primary part of K(L)K(L).

(a) If the 22-primary part of xx generates X2X_{2}, we may choose yy such that 2y=x2y=x where yK(L)y\not\in K(L) (we may choose the non 22-primary part of yy to be in XX). Thus 2.6(a) applies.

(b) Choose yXy\in X so that 2y=x2y=x and apply 2.6(b). ∎

Proposition 3.2.

If X2X_{2} is not cyclic, then there is an xXx\in X so that #FixC([1]tx)=12\#\operatorname{Fix}_{C}([-1]\circ t_{x})=12.

Moreover, for any xXx\in X, #FixC([1]tx)\#\operatorname{Fix}_{C}([-1]\circ t_{x}) is either 44 or 1212.

Proof.

We may write the decomposition of the 22-primary part of K(L)K(L) for ll as follows:

K(L)2:=K(L)A[2m]K1K2=k1k2.K(L)_{2}:=K(L)\cap A[2^{m}]\cong K_{1}\oplus K_{2}=\langle k_{1}\rangle\oplus\langle k_{2}\rangle.

Since we have chosen compatible decompositions in 1.1, we may also write X2X_{2} as a sum with respect to this decomposition for some 0<m1m20<m_{1}\leq m_{2} and m1+m2=mm_{1}+m_{2}=m:

X2/2m1/2m2.X_{2}\simeq\mathbb{Z}/2^{m_{1}}\mathbb{Z}\oplus\mathbb{Z}/2^{m_{2}}\mathbb{Z}.

Since m1,m2<mm_{1},m_{2}<m, for any xXx\in X, there exists yK(L)y\in K(L) so that 2y=x2y=x, and thus 2.6(a) cannot apply and #FixC([1]tx)\#\operatorname{Fix}_{C}([-1]\circ t_{x}) cannot be 88, leaving 44 and 1212 as the possibile numbers of fixed points.

Again, by examining orders of elements, there is a yK(L)Xy\in K(L)\setminus X so that 2y=xX2y=x\in X and π(y)\pi(y) is not contained in either group π(ki)\langle\pi(k_{i})\rangle. As shown in section 2.0.1, tπ(y)t_{\pi(-y)}^{*} must be odd. ∎

3.2. Smooth hyperelliptic curves in (1,d)(1,d)-polarizations

In this section, we review the number of smooth hyperelliptic curves in a fixed symmetric linear system of type (1,d)(1,d) on a general abelian surface. These values are nonzero for 1d41\leq d\leq 4 and in these cases are 11, 66, 99 and 44, respectively (see [BOPY18, Table 1]).

These curves are symmetric and have hyperelliptic involution [1][-1] or [1]tx[-1]\circ t_{x} for some xK(L)x\in K(L) or are translations of such a curve: by Lemma 2.2, for any curve CAC\subset A with involution [1]tx[-1]\circ t_{x} and bK(L)b\in K(L), #FixC([1]tx)=#FixC+b([1]tx2b)\#\operatorname{Fix}_{C}([-1]\circ t_{x})=\#\operatorname{Fix}_{C+b}([-1]\circ t_{x-2b}).

Smooth curves in |L||L| have genus d+1d+1 and thus a hyperelliptic involution on such a curve must have 2d+42d+4 fixed points. For d=2d=2 or 44, we produce these curves using 1.1. In those cases, we may then produce |K(L)/X||K(L)/X| distinct translations of each curve.

3.2.1. dd odd

In the principally polarized case (d=1d=1), the abelian surface is the Jacobian of a genus 22 curve. Any symmetric polarizing line bundle on such a surface has its linear system given by (a translation of) this curve with hyperellipitic involution [1][-1], making the total number of hyperelliptic curves in a fixed linear system |K(L)|=1|K(L)|=1.

When d=3d=3, a hyperelliptic involution must have 1010 fixed points. In 2.1, we see that the systems H0(A,L)±H^{0}(A,L)^{\pm} contain a single smooth curve where [1][-1] has 1010 fixed points. By 2.5, this curve cannot occur as a cover of a genus 22 curve using 1.1. Borówka and Sankaran [BS17] exhibit it in terms of theta functions and, show that, up to translation, it is the unique hyperelliptic curve in a general (1,3)(1,3)-polarized abelian surface.

Given this curve CC, for each bK(L)b\in K(L), the translation C+bC+b is a distinct hyperelliptic curve in |L||L|. The total number of hyperelliptic curves in |L||L| produced in this way is:

|K(L)|=9.|K(L)|=9.

3.2.2. (1,2)(1,2)-polarizations

In this case, a hyperelliptic involution must have 88 fixed points. Any curve CC as in 1.1 is formed using an order two cyclic group x=XK(L)\langle x\rangle=X\leq K(L). By 3.1, #FixC[1]=4\#\operatorname{Fix}_{C}[-1]=4 and #FixC([1]tx)=8\#\operatorname{Fix}_{C}([-1]\circ t_{x})=8, making [1]tx[-1]\circ t_{x} a hyperelliptic involution.

A detailed treatment of the (1,2)(1,2)-polarized case where LL has a symmetric theta structure is given by Barth in [Bar87]. Our argument here recovers his result that, for instance, a hyperelliptic involution of the curves in |L||L| must be distinct from [1][-1].

The total number of hyperelliptic curves in |L||L| that we can produce from this construction, as well as translating, is thus:

/2XK(L)|K(L)/X|=32=6.\sum_{\mathbb{Z}/2\mathbb{Z}\simeq X\leq K(L)}|K(L)/X|=3\cdot 2=6.

3.2.3. (1,4)(1,4)-polarizations

In this case a hyperelliptic involution must have 1212 fixed points. Curves can be obtained using 1.1 using a choice of order 44 subgroup XK(L)(/4)2X\leq K(L)\simeq(\mathbb{Z}/4\mathbb{Z})^{2}.

By 3.1, if XK(L)X\leq K(L) is cyclic, then #Fix([1]tx)\#\operatorname{Fix}([-1]\circ t_{x}) is 44 or 88, so 1.1 will not produce a hyperelliptic involution.

Now, let XK(L)X\leq K(L) be the Klein group. We may write K(L)K(L) as a sum with respect to the decomposition of AA for [L][L]:

k1k2K1K2.\langle k_{1}\rangle\oplus\langle k_{2}\rangle\cong K_{1}\oplus K_{2}.

Using this notation, X=2k12k2X=\langle 2k_{1}\rangle\oplus\langle 2k_{2}\rangle. By 3.2,

#FixC[1]=#FixC([1]t2k1)=#FixC([1]t2k2)=4,\displaystyle\#\operatorname{Fix}_{C}[-1]=\#\operatorname{Fix}_{C}([-1]\circ t_{2k_{1}})=\#\operatorname{Fix}_{C}([-1]\circ t_{2k_{2}})=4,
#FixC([1]t2k1+2k2)=12,\displaystyle\#\operatorname{Fix}_{C}([-1]\circ t_{2k_{1}+2k_{2}})=12,

making [1]t2k1+2k2[-1]\circ t_{2k_{1}+2k_{2}} a hyperelliptic involution for CC.

Since there is only one way to embed the Klein group V4V_{4} into K(L)K(L), we have found one hyperelliptic curve using 1.1. The total number of hyperelliptic curves in |L||L| that we may then produce by translating is:

|K(L)/V4|=4.|K(L)/V_{4}|=4.
Remark 3.3.

This construction coincides with that of Borówka and Ortega [BO19], and so we may think of 2.6 as a generalization of their results. However, in 1.1 we have specified additional data by choosing decompositions.

The authors prove their curve is the unique hyperelliptic curve (up to translation) on a general (1,4)(1,4)-polarized abelian surface. It is constructed as a cover of a genus 22 curve, C~H\widetilde{C}\to H with respect to the Klein group {0,η1,η2,η1+η2}=GJH[2]\{0,\eta_{1},\eta_{2},\eta_{1}+\eta_{2}\}=G\leq J_{H}[2]. The result [BO19, Theorem 4.7] states that if C~\widetilde{C} is hyperelliptic then GG is non-isotropic with respect to the Weil pairing on JH[2]J_{H}[2] and that, conversely, if GG is non-isotropic and η1\eta_{1} and η2\eta_{2} may be written as the difference of two Weierestrass points, then C~\widetilde{C} is hyperelliptic.

In the notation of this paper π(K(L))\pi(K(L)) plays the role of the group GG generated by π(k1)\pi(k_{1}), π(k2)\pi(k_{2}). It’s precisely this non-isotropy condition that allows us to pick an element of zπ(K(L))z\in\pi(K(L)) where tzMt_{z}^{*}M is odd. The Weierstrass points of zHz\in H are those where qM(z)=1q_{M}(z)=-1, but since we have arranged that qM(π(k1))=qM(π(k2))=1q_{M}(\pi(k_{1}))=q_{M}(\pi(k_{2}))=1, we can see that π(k1)\pi(k_{1}) and π(k2)\pi(k_{2}) must be differences of Weierstrass points.

4. Decomposing Jacobians using subgroups

By Poincaré reducibility, every abelian variety is isogenous to a product of simple abelian varieties that are unique, up to isogeny. In this section, we decompose the Jacobians of the curves CC produced using 1.1 into, up to isogeny, products of smaller abelian varieties.

In section 4.1, we give some general consequences of [KR89, Theorem B] for decomposing JCJ_{C}. In section 4.2, we use our methods to examine several cases where JCJ_{C} is isogenous to a product of AA and elliptic curves, recovering some results of [ES93]. Finally in section 4.3 we give some examples of curves that cover an elliptic curve.

4.1. General results on decomposing Jacobians

Kani and Rosen proved the following theorem by showing that partitions of the automorphism group of CC induce idempotent relations in the endomorphism algebra of the Jacobian JCJ_{C}. The automorphisms of the curves CC produced using 1.1 contain the group 1X\langle-1\rangle\rtimes X (1.2), but not every such group admits a partition. The only subgroups admitting a partition are dihedral groups or certain pp-groups (see [Sch94, Theorem 3.5.10]), hence the structure of this group affects the results that may be gained from applying this theorem.

Theorem 4.1 ([KR89, Theorem B]).

Let CC be a (smooth, projective, geometrically connected) curve and GG be a finite subgroup of Aut(C)\operatorname{Aut}(C) such that G=H1HtG=H_{1}\cup\cdots\cup H_{t} where the subgroups HiGH_{i}\leq G satisfy HiHj=1H_{i}\cap H_{j}=1 if iji\neq j. Then we have the isogeny relation

JCt1×JC/GgJC/H1h1××JC/HthtJ_{C}^{t-1}\times J_{C/G}^{g}\sim J_{C/H_{1}}^{h_{1}}\times\cdots\times J_{C/H_{t}}^{h_{t}} (4.1)

where |G|=g|G|=g and |Hi|=hi|H_{i}|=h_{i}.

We deduce two general results on the decomposition of Jacobians of curves CC produced as in 1.1 in the cases where XK(L)X\leq K(L) is cyclic or it is a product of two cyclic groups of odd prime order.

Theorem 4.2.

Let CC be as in 1.1 for X=xK(L)X=\langle x\rangle\leq K(L) a cyclic group (of order dd). We have the following isogeny relations:

  1. (a)

    When dd is odd: JCA×JC/12J_{C}\sim A\times J^{2}_{C/\langle-1\rangle}.

  2. (b)

    When dd is even: JCA×JC/1×JC/1txJ_{C}\sim A\times J_{C/\langle-1\rangle}\times J_{C/\langle-1\circ t_{x}\rangle}.

Proof.

Aut(C)\operatorname{Aut}(C) contains the dihedral group G:=1,txG:=\langle-1,t_{x}\rangle of order 2d2d, which admits the following partition:

G=tx11tx1t(d1)x.G=\langle t_{x}\rangle\cup\langle-1\rangle\cup\langle-1\circ t_{x}\rangle\cup\cdots\cup\langle-1\circ t_{(d-1)x}\rangle.

Applying 4.1 gives the following isogeny relation:

JCd×JC/G2dJC/txd×JC/12×JC/1tx2××JC/1t(d1)x2.J_{C}^{d}\times J_{C/G}^{2d}\sim J_{C/\langle t_{x}\rangle}^{d}\times J_{C/\langle-1\rangle}^{2}\times J_{C/\langle-1\circ t_{x}\rangle}^{2}\times\cdots\times J_{C/\langle-1\circ t_{(d-1)x}\rangle}^{2}.

We now analyze these quotients further. The quotient C/GC/G is isomorphic to the quotient of the genus two curve C/XHC/X\cong H of 1.1 by its hyperelliptic involution, and thus C/G1C/G\cong\mathbb{P}^{1}, making JC/GJ_{C/G} trivial.

Furthermore, quotienting a curve by conjugate subgroups yields isomorphic quotient curves. When dd is odd, the order two subgroups of GG are precisely the 22-Sylow subgroups of GG and hence are conjugate, proving (a).

When dd is even, the order two subgroups generated by 1,1t2x,,1t(d2)x-1,-1\circ t_{2x},\ldots,-1\circ t_{(d-2)x} are conjugate to one another, and the the remaining order two subgroups are all conjugate to 1tx\langle-1\circ t_{x}\rangle, proving result (b). ∎

Theorem 4.3.

Let pp be an odd prime and d=p2d=p^{2}. Let CC be as in 1.1 for:

(/p)2X=x1×x2K(L).(\mathbb{Z}/p\mathbb{Z})^{2}\cong X=\langle x_{1}\rangle\times\langle x_{2}\rangle\leq K(L).

Then we have the following isogeny relations, where each CiC_{i} is a curve of genus p12\frac{p-1}{2} that is a quotient of CC by [1][-1] and a translation:

JCA×i=1p+1JCi2,JC/1i=1p+1JCi.J_{C}\sim A\times\prod_{i=1}^{p+1}J_{C_{i}}^{2},\quad J_{C/\langle-1\rangle}\sim\prod_{i=1}^{p+1}J_{C_{i}}.
Proof.

The subgroup G:=tx1,tx2G:=\langle t_{x_{1}},t_{x_{2}}\rangle of Aut(C)\operatorname{Aut}(C) has the following partition:

tx1tx2tx1+x2tx1+2x2tx1+(p1)x2.\langle t_{x_{1}}\rangle\cup\langle t_{x_{2}}\rangle\cup\langle t_{x_{1}+x_{2}}\rangle\cup\langle t_{x_{1}+2x_{2}}\rangle\cup\cdots\cup\langle t_{x_{1}+(p-1)x_{2}}\rangle.

Then 4.1 gives the following isogeny relation:

JCp×JC/Gp2JC/tx1××JC/tx1+(p1)x2.J_{C}^{p}\times J_{C/G}^{p^{2}}\sim J_{C/\langle t_{x_{1}}\rangle}\times\cdots\times J_{C/\langle t_{x_{1}+(p-1)x_{2}}\rangle}. (4.2)

Each curve on the right-hand side in the above isogeny relation is a quotient of CC by a cyclic group of order pp, which is in turn a cyclic cover of degree pp of a genus 22 curve.

Since the quotient C/txC/\langle t_{x}\rangle is in the linear system of a (1,p)(1,p)-polarization on A/x1A/\langle x_{1}\rangle, we may apply the argument used in 4.2(a):

JC/tx1A/x1×JC12A×JC12J_{C/\langle t_{x_{1}}\rangle}\sim A/\langle x_{1}\rangle\times J_{C_{1}}^{2}\sim A\times J_{C_{1}}^{2}

where C1=C/tpx1,1C_{1}=C/\langle t_{px_{1}},-1\rangle is of genus p12\frac{p-1}{2}. Performing an analogous decomposition for each cyclic quotient of CC and observing that JC/GAJ_{C/G}\sim A, we simplify (4.2) to

JCpAp×i=1p+1JCi2pJCA×i=1p+1JCi2.J_{C}^{p}\sim A^{p}\times\prod_{i=1}^{p+1}J_{C_{i}}^{2p}\ \ \Rightarrow\ \ J_{C}\sim A\times\prod_{i=1}^{p+1}J_{C_{i}}^{2}. (4.3)

Now, let G:=tx1,tx2,[1]Aut(C)G:=\langle t_{x_{1}},t_{x_{2}},[-1]\rangle\leq\operatorname{Aut}(C), which has the following partition:

G=(0m,np1[1]tmx1+nx2)tx1,tx2.G=\biggl{(}\bigcup_{0\leq m,n\leq p-1}\langle[-1]\circ t_{mx_{1}+nx_{2}}\rangle\biggr{)}\cup\langle t_{x_{1}},t_{x_{2}}\rangle.

Applying 4.1 gives the isogeny relation:

JCp2×JC/Gp2JC/tx1,tx2p2×0m,np1JC/1tmx1+nx22.J_{C}^{p^{2}}\times J_{C/G}^{p^{2}}\sim J_{C/\langle t_{x_{1}},t_{x_{2}}\rangle}^{p^{2}}\times\prod_{0\leq m,n\leq p-1}J^{2}_{C/\langle-1\circ t_{mx_{1}+nx_{2}}\rangle}. (4.4)

Now, in the notation of 1.1, JC/tx1,tx2JHAJ_{C/\langle t_{x_{1}},t_{x_{2}}\rangle}\cong J_{H}\sim A, and hence C/G1C/G\cong\mathbb{P}^{1}. All order 22 subroups of GG are Sylow, hence conjugate to 1\langle-1\rangle. Simplifying (4.4) and combining with (4.3) yields:

JC/1i=1p+1JCi.J_{C/\langle-1\rangle}\sim\prod_{i=1}^{p+1}J_{C_{i}}.\qed

4.2. Completely decomposable Jacobians

In [ES93, §4], Ekedahl and Serre prove the following result using characters. In this section, we examine some cases where this theorem applies using 4.1 and our results on automorphisms and ramification.

Theorem 4.4 ([ES93, Prop. 3 and Corollaire]).

Let HH be a curve of genus 22 and let CHC\to H be a finite, unramified abelian covering of curves whose Galois group has exponent dividing 44 or 66. If JHJ_{H} is isogenous to a product of elliptic curves, then JCJ_{C} is completely decomposable.

This result implies that if the group XX of 1.1 has exponent dividing 44 or 66 and AA is isogenous to a product of elliptic curves, then JCJ_{C} is completely decomposable. We now examine the cases where XX is cyclic as well as the smallest non-cyclic case, the Klein cover of [BO19].

XX cyclic

When d=2d=2 or 33, 4.2 directly implies complete decomposability of JCJ_{C}. When d=2d=2, as shown in section 3.2.2, JC/1J_{C/\langle-1\rangle} is elliptic and JC/1txJ_{C/\langle-1\circ t_{x}\rangle} is trivial. When d=3d=3, by 2.5, JC/1J_{C/\langle-1\rangle} is an elliptic curve (cf. [BL95, §5]).

When d=4d=4 and X/4X\cong\mathbb{Z}/4\mathbb{Z}, 4.2 and 3.1 imply that JCJ_{C} may be decomposed into the product of AA with a surface JC/1J_{C/\langle-1\rangle} and an elliptic curve JC/1txJ_{C/\langle-1\circ t_{x}\rangle}. However, C/t2x{C/\langle t_{2x}\rangle} is a curve in the linear system of a (1,2)(1,2)-polarization and JC/t2x,1J_{C/\langle t_{2x},-1\rangle} is an elliptic curve. Since C/1{C/\langle-1\rangle} covers an elliptic curve, JC/1J_{C/\langle-1\rangle} is isogenous to a product of elliptic curves.

When d=6d=6 and X/6X\cong\mathbb{Z}/6\mathbb{Z}, 4.2 and 3.1 imply that JCJ_{C} may be decomposed into the product of AA with a threefold JC/1J_{C/\langle-1\rangle} and a surface JC/1txJ_{C/\langle-1\circ t_{x}\rangle}. Similarly to the previous case, C/t3xC/\langle t_{3x}\rangle is in a (1,3)(1,3)-polarization and JC/t3x,1J_{C/\langle t_{3x},-1\rangle} is an elliptic curve. Also, C/t2xC/\langle t_{2x}\rangle is in a (1,2)(1,2)-polarization and JC/t2x,1J_{C/\langle t_{2x},-1\rangle} is an elliptic curve. Since C/1t3x{C/\langle-1\circ t_{3x}\rangle} covers C/t3x,1{C/\langle t_{3x},-1\rangle} and JC/1txJC/1t3xJ_{C/\langle-1\circ t_{x}\rangle}\cong J_{C/\langle-1\circ t_{3x}\rangle}, JC/1txJ_{C/\langle-1\circ t_{x}\rangle} is isogenous to a product of elliptic curves. The threefold JC/1JC/1t2xJ_{C/\langle-1\rangle}\cong J_{C/\langle-1\circ t_{2x}\rangle} covers the distinct elliptic curves JC/t2x,1J_{C/\langle t_{2x},-1\rangle} and JC/1t3xJ_{C/\langle-1\circ t_{3x}\rangle}, so is completely decomposable.

Klein cover

Let d=4d=4 and XX be the Klein group. We may write X=x1x2(/2)2X=\langle x_{1}\rangle\oplus\langle x_{2}\rangle\cong(\mathbb{Z}/2\mathbb{Z})^{2}. In this notation, [1]tx1+x2[-1]\circ t_{x_{1}+x_{2}} is the hyperelliptic involution (see section 3.2.3). Analogously to the proof of 4.3, the group G:=tx1,tx2Aut(C)G:=\langle t_{x_{1}},t_{x_{2}}\rangle\leq\operatorname{Aut}(C) admits a partition:

G=tx1tx2tx1+x2.G=\langle t_{x_{1}}\rangle\cup\langle t_{x_{2}}\rangle\cup\langle t_{x_{1}+x_{2}}\rangle.

The isogeny relation given by 4.1 simplifies to:

JC×A2JC/tx1×JC/tx2×JC/tx1+x2.J_{C}\times A^{2}\sim J_{C/\langle t_{x_{1}}\rangle}\times J_{C/\langle t_{x_{2}}\rangle}\times J_{C/\langle t_{x_{1}+x_{2}}\rangle}. (4.5)

There is a cover C/tx1HC/\langle t_{x_{1}}\rangle\to H given by taking the quotient by tx2t_{x_{2}}. Thus the automorphism group of C/tx1C/\langle t_{x_{1}}\rangle contains the order 44 dihedral group 1,tx2\langle-1,t_{x_{2}}\rangle. Applying 4.1, we have:

JC/tx1A/x1×JC/tx1,1×JC/tx1,1tx2.J_{C/\langle t_{x_{1}}\rangle}\sim A/\langle{x_{1}}\rangle\times J_{C/\langle t_{x_{1}},-1\rangle}\times J_{C/\langle t_{x_{1}},-1\circ t_{x_{2}}\rangle}.

Since tx1,1tx2\langle t_{x_{1}},-1\circ t_{x_{2}}\rangle contains the hyperelliptic involution, JC/tx1,1tx2J_{C/\langle t_{x_{1}},-1\circ t_{x_{2}}\rangle} is trivial, meaning that JC/tx1,1J_{C/\langle t_{x_{1}},-1\rangle} is elliptic. Arguing similarly for C/tx2C/\langle t_{x_{2}}\rangle and C/tx1+x2C/\langle t_{x_{1}+x_{2}}\rangle, the isogeny relation (4.5) reduces to the following relation between JCJ_{C} and the product of AA with three elliptic curves (see [BO19, Theorem 5.5]), which are each the Jacobian of a quotient of CC by a subgroup of Aut(C)\operatorname{Aut}(C) that does not contain its hyperelliptic involution:

JCA×JC/tx1,1×JC/tx2,1×JC/tx1+x2,1tx1.J_{C}\sim A\times J_{C/\langle t_{x_{1}},-1\rangle}\times J_{C/\langle t_{x_{2}},-1\rangle}\times J_{C/\langle t_{x_{1}+x_{2}},-1\circ t_{x_{1}}\rangle}.

Finally, we note that 4.3 directly implies complete decompsability in the case where X=(/3)2X=(\mathbb{Z}/3\mathbb{Z})^{2} (see [BS25] for analysis of this example).

4.3. Curves that cover an elliptic curve

We can use our methods to examine the following weaker condition: does the curve CC of 1.1 cover an elliptic curve? The following result gives some cases where this question has a positive answer.

Proposition 4.5.

When d=2kd=2k, 3k3k, or 4k4k for some kk\in\mathbb{N} and X/j×/kX\cong\mathbb{Z}/j\mathbb{Z}\times\mathbb{Z}/k\mathbb{Z} for j=2j=2, 33 or 44, the curve CC of 1.1 covers an elliptic curve. Further the degree of the cover is dd and JCJ_{C} has an elliptic factor in its isogeny class.

Proof.

Let d=2kd=2k and let x1,x2K(L)x_{1},x_{2}\in K(L) be elements of orders kk and 22 so that X=x1,x2K(L)X=\langle x_{1},x_{2}\rangle\leq K(L) is a subgroup of order 2k2k.

The curve C~:=C/tx1\widetilde{C}:=C/\langle t_{x_{1}}\rangle has genus 33 and quotienting by tx2t_{x_{2}} gives a cover C~H\widetilde{C}\to H, and thus the automorphism group of C~\widetilde{C} contains tx2,1D4\langle t_{x_{2}},-1\rangle\cong D_{4}. Applying 4.1, we have:

JC~A/x1×JC~/1×JC~/1tx2.J_{\widetilde{C}}\sim A/\langle x_{1}\rangle\times J_{\widetilde{C}/\langle-1\rangle}\times J_{\widetilde{C}/\langle-1\circ t_{x_{2}}\rangle}.

Comparing dimensions, we see that one of C~/1\widetilde{C}/\langle-1\rangle or C~/1tx2\widetilde{C}/\langle-1\circ t_{x_{2}}\rangle must be an elliptic curve EE and the covering map CEC\to E has degree 2k2k.

Since JCJC~×P(C/C~)J_{C}\sim J_{\widetilde{C}}\times P(C/\widetilde{C}), where P(C/C~)P(C/\widetilde{C}) is the Prym variety, EE appears as an isogenous factor in the decomposition of JCJ_{C}.

For d=3kd=3k, we make the same argument with x2x_{2} an element of order 33. Then C~:=C/tx1\widetilde{C}:=C/\langle t_{x_{1}}\rangle is a curve of genus 44 whose automorphism group contains 1,tx2D6\langle-1,t_{x_{2}}\rangle\cong D_{6} and JC~A/x1×JC/12J_{\widetilde{C}}\sim A/\langle x_{1}\rangle\times J_{C/\langle-1\rangle}^{2}, thus C~/1=:E\widetilde{C}/\langle-1\rangle=:E is elliptic, CEC\to E has degree 3k3k, and EE is an isogenous factor of JCJ_{C}.

For d=4kd=4k, we let x2x_{2} have order 44. The curve C~:=C/tx1\widetilde{C}:=C/\langle t_{x_{1}}\rangle has genus 55 and its automorphism group contains D8D_{8}. The Jacobian JC~J_{\widetilde{C}} thus also has an elliptic factor EE where CEC\to E has degree 4k4k. ∎

References

  • [Bar87] Wolf Barth. Abelian surfaces with (1,2)(1,2)-polarization. In Algebraic geometry, Sendai, 1985, volume 10 of Adv. Stud. Pure Math., pages 41–84. North-Holland, Amsterdam, 1987.
  • [BL95] Ch. Birkenhake and H. Lange. Moduli spaces of abelian surfaces with isogeny. In Geometry and analysis (Bombay, 1992), pages 225–243. Tata Inst. Fund. Res., Bombay, 1995.
  • [BL04] Ch. Birkenhake and H. Lange. Complex abelian varieties, volume 302 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, second edition, 2004.
  • [BO19] Paweł Borówka and Angela Ortega. Hyperelliptic curves on (1,4)(1,4)-polarised abelian surfaces. Math. Z., 292(1-2):193–209, 2019.
  • [BOPY18] Jim Bryan, Georg Oberdieck, Rahul Pandharipande, and Qizheng Yin. Curve counting on abelian surfaces and threefolds. Algebr. Geom., 5(4):398–463, 2018.
  • [BS17] P. Borówka and G. K. Sankaran. Hyperelliptic genus 4 curves on abelian surfaces. Proc. Amer. Math. Soc., 145(12):5023–5034, 2017.
  • [BS25] Paweł Borówka and Anatoli Shatsila. Pryms of 3×3\mathbb{Z}_{3}\times\mathbb{Z}_{3} coverings of genus 2 curves, 2025.
  • [CLR21] Angel Carocca, Herbert Lange, and Rubí E. Rodríguez. Decomposable Jacobians. Ann. Sc. Norm. Super. Pisa Cl. Sci. (5), 22(4):1673–1690, 2021.
  • [ES93] Torsten Ekedahl and Jean-Pierre Serre. Exemples de courbes algébriques à jacobienne complètement décomposable. C. R. Acad. Sci. Paris Sér. I Math., 317(5):509–513, 1993.
  • [EvdGM] Bas Edixhoven, Gerard van der Geer, and Ben Moonen. Abelian varieties. online; accessed 15 April 2025.
  • [KR89] E. Kani and M. Rosen. Idempotent relations and factors of Jacobians. Math. Ann., 284(2):307–327, 1989.
  • [Mum70] D. Mumford. Abelian varieties, volume 5 of Tata Institute of Fundamental Research Studies in Mathematics. Published for the Tata Institute of Fundamental Research, Bombay by Oxford University Press, London, 1970.
  • [PS25] Jennifer Paulhus and Andrew V. Sutherland. Completely decomposable modular Jacobians, 2025.
  • [Sch94] Roland Schmidt. Subgroup lattices of groups, volume 14 of De Gruyter Expositions in Mathematics. Walter de Gruyter & Co., Berlin, 1994.