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

Abelian varieties with Real multiplication :
classification and isogeny classes over finite fields

Tejasi Bhatnagar, Yu Fu
Abstract.

In this paper, we provide a classification of certain points on Hilbert modular varieties over finite fields under a mild assumption on Newton polygon. Furthermore, we use this characterization to prove estimates for the size of isogeny classes.

1. Introduction

Throughout this paper, 𝔽q{\mathbb{F}}_{q} will denote the finite field with qq elements where qq is a power of an odd prime pp. We denote by kk the algebraic closure of 𝔽q{\mathbb{F}}_{q}. In [De69], Deligne provides a classification of ordinary abelian varieties over finite fields in terms of certain {\mathbb{Z}}-modules along with an action of a “Frobenius map”. A crucial step in this classification requires the existence of the “Serre-Tate canonical lift” of an ordinary abelian variety to characteristic zero. Using the “canonical lift” of an ordinary abelian variety, Deligne provides a characterisation of such abelian varieties by associating to them the integral homology of their lift. A similar classification for simple almost ordinary abelian varieties over finite fields is proved in [OS20]. In order to extend Deligne’s work, Oswal and Shankar construct canonical lift(s) of almost ordinary abelian varieties using Grothendieck-Messing theory. Moreover, the lifts are characterized by the property that all of the endomorphisms lift to characteristic zero. Using these canonical lift(s), they classify simple almost ordinary abelian varieties, and as an application, they give a lower bound of the size of almost-ordinary isogeny classes. This paper aims to generalise their work to certain abelian varieties, (not necessarily simple) with real multiplication over finite fields with some assumptions on pp, as described below. (See Theorem 1.4)

1.0.1. The Hilbert moduli space

Let LL be a totally real number field of degree gg over {\mathbb{Q}}. Let 𝒪L{\mathcal{O}}_{L} be its ring of integers. We denote by L,𝔞\mathcal{H}_{L,\mathfrak{a}} the coarse Hilbert moduli space over 𝔽q{\mathbb{F}}_{q} associated to LL. This moduli space parametrises abelian varieties with real multiplication with additional data. More precisely, we describe the moduli problem as follows.

Definition 1.1.

Fix a fractional ideal 𝔞\mathfrak{a} of LL. Define L,𝔞\mathcal{H}_{L,\mathfrak{a}} to be the functor such that for any 𝔽q{\mathbb{F}}_{q}-scheme SS, L,𝔞(S)\mathcal{H}_{L,\mathfrak{a}}(S) is the set of isomorphism classes of triples,(A,ι,λ)(A,\iota,\lambda) where:

  • (1)

    ASA\to S is a gg-dimensional abelian scheme over SS;

  • (2)

    ι:𝒪LEndA\iota:{\mathcal{O}}_{L}\hookrightarrow\operatorname{End}A is an embedding into the endomorphism ring;

  • (3)

    λ:𝔞Hom𝒪L,SSym(A,A)\lambda:\mathfrak{a}\stackrel{{\scriptstyle\sim}}{{\smash{\longrightarrow}\rule{0.0pt}{1.72218pt}}}\operatorname{Hom}_{{\mathcal{O}}_{L},S}^{\operatorname{Sym}}(A,A^{\vee}) is an isomorphism of 𝒪L{\mathcal{O}}_{L} modules. Here λ\lambda identifies the set of polarizations with the set of totally positive element 𝔞+𝔞\mathfrak{a}^{+}\subset\mathfrak{a}. For every a𝔞+a\in\mathfrak{a}^{+}, λ(a)\lambda(a) is an 𝒪L\mathcal{O}_{L}-linear polarization of AA, and the homomorphism A𝒪L𝔞AA\otimes_{\mathcal{O}_{L}}\mathfrak{a}\stackrel{{\scriptstyle\sim}}{{\smash{\longrightarrow}\rule{0.0pt}{1.72218pt}}}A^{\vee} induced by λ\lambda is an isomorphism of abelian schemes.

See [Goren, Chapter 3.6.1] for a complete description of the moduli problem. For the purpose of this paper, we will work with the coarse moduli space and classify 𝔞\mathfrak{a}-polarised abelian varieties with RM, for a fixed fractional ideal 𝔞\mathfrak{a} of L.L. In particular, when the polarisations are principal, the corresponding fractional ideal class is the inverse different of LL, denoted by 𝒟L/1\mathcal{D}_{L/{\mathbb{Q}}}^{-1}. We restrict to this case while counting the size of isogeny classes.

Definition 1.2.

Two abelian varieties (A,λA,ιA)(A,\lambda_{A},\iota_{A}) and (B,λB,ιB)(B,\lambda_{B},\iota_{B}) in L,𝔞(𝔽q)\mathcal{H}_{L,\mathfrak{a}}({\mathbb{F}}_{q}) are said to be isogenous if there is an isogeny φ:AB\varphi:A\rightarrow B over 𝔽¯q\overline{\mathbb{F}}_{q} that is compatible with the action of 𝒪L{\mathcal{O}}_{L}, that is φιA=ιBφ\varphi\circ\iota_{A}=\iota_{B}\circ\varphi. Moreover, the isogeny preserves polarization up to scaling by L×L^{\times}.

Remark 1.1.

From this section onwards we fix a totally real field LL of degree gg over {\mathbb{Q}} and a fractional ideal 𝔞\mathfrak{a} of LL. When we say that an abelian variety AA over 𝔽q{\mathbb{F}}_{q} has RM, we mean that AA over 𝔽q{\mathbb{F}}_{q} is 𝔞\mathfrak{a}-polarised and has RM by 𝒪L.\mathcal{O}_{L}. That is, AA is an 𝔽q{\mathbb{F}}_{q}-point of L,𝔞.\mathcal{H}_{L,\mathfrak{a}}.

1.0.2. The main theorem: classification

Let AA be a geometrically simple 𝔽q{\mathbb{F}}_{q}-point of L,𝔞\mathcal{H}_{L,\mathfrak{a}}. We assume that the pp-rank111We recall that the pp-rank of an abelian variety is defined to be the integer f=dim𝔽pHom(μp,A[p])f=\operatorname{dim}_{{\mathbb{F}}_{p}}\operatorname{Hom}(\mu_{p},A[p]). of AA is gag-a for some 0ag1.0\leq a\leq g-1. Furthermore, we assume that pp is totally split in 𝒪L{\mathcal{O}}_{L}. This assumption guarantees that the pp-divisible group splits into either ordinary or local-local factors (see Lemma 3.1), implying that the endomorphism algebra of AA is, in fact, a CM field K.K. Consequently, the extension KpK\otimes{\mathbb{Q}}_{p}, splits as KordKssK_{\operatorname{ord}}\oplus K_{\operatorname{ss}} where KordK_{\operatorname{ord}} is a direct sum of 2(ga)2(g-a) factors of p{\mathbb{Q}}_{p}, while KssK_{\operatorname{ss}} splits as a sum of aa quadratic extensions of p{\mathbb{Q}}_{p}. We write Kss=1iaKssiK_{\operatorname{ss}}=\oplus_{1\leq i\leq a}K^{i}_{\operatorname{ss}}.

Terminology 1.3.

We call the abelian variety AA (or its isogeny class) totally ramified, if all the factors of KssK_{\operatorname{ss}} are ramified, inert if all the factors are inert, and ramified if some of the factors are inert and some are ramified. We note that the above terminology extends to non-simple abelian varieties as well, and this result is true in generality, thus giving us a classification for RM abelian varieties with non-commutative endomorphism ring as well.

In what follows, we denote by 𝒞h\mathcal{C}_{h} the category of abelian varieties over 𝔽q{\mathbb{F}}_{q} with RM such that the characteristic polynomial of their relative Frobenius is hh. Its objects are 𝔽q{\mathbb{F}}_{q}-points of L,𝔞\mathcal{H}_{L,\mathfrak{a}}, see 4.2 for a precise definition. The morphisms in this category are morphisms of abelian varieities over 𝔽q{\mathbb{F}}_{q} that are compatible with the action of 𝒪L\mathcal{O}_{L} and preserve polarisations up to scaling by L×L^{\times}. We classify the objects of 𝒞h\mathcal{C}_{h} in terms of certain {\mathbb{Z}}-modules with a “Frobenius” map and an action by 𝒪L\mathcal{O}_{L}. These objects along with the defined morphisms form a category of Deligne modules with RM which we denote by h.\mathcal{L}_{h}. See 4.1 for the precise definition. We prove the following classification result in this paper.

Theorem 1.4.

Let 𝒞h\mathcal{C}_{h} and h\mathcal{L}_{h} be defined as above. We assume that the pp-rank of abelian varieties in 𝒞h\mathcal{C}_{h} equals to gag-a for some 0ag1.0\leq a\leq g-1.

  1. (1)

    If 𝒞h\mathcal{C}_{h} is totally ramified, then there exists 2a2^{a} canonical functors i:𝒞hh\mathcal{F}_{i}:\mathcal{C}_{h}\rightarrow\mathcal{L}_{h} with 1i2a.1\leq i\leq 2^{a}. Each of these functors induce equivalences between two categories.

  2. (2)

    If 𝒞h\mathcal{C}_{h} is inert, then there are 2a2^{a} full subcategories 𝒞h,i\mathcal{C}_{h,i} of 𝒞h\mathcal{C}_{h} where 1i2a.1\leq i\leq 2^{a}. The functors i\mathcal{F}_{i} from each of these subcategories to h\mathcal{L}_{h} induce an equivalence between the categories.

  3. (3)

    If 𝒞h\mathcal{C}_{h} is ramified, analogous to (2)(2), then there are 2ak2^{a-k} subcategories of 𝒞h\mathcal{C}_{h} where kk is the number of inert factors of Kss.K_{\operatorname{ss}}. The functors i\mathcal{F}_{i} from each of these subcategories to h\mathcal{L}_{h} for 1i2ak1\leq i\leq 2^{a-k}, induce an equivalence between the categories.

1.1. Strategy to prove the classification theorem.

We first prove that AA lifts to characteristic zero along with all its endomorphism algebra. More precisely, when AA is totally ramified, it has exactly 2a2^{a} lifts to characteristic zero. When AA is inert, then it has one canonical lift, and finally, in the case when AA is ramified, its number of lifts depend on the number of ramified quadratic extensions in the direct sum KssK_{\operatorname{ss}}. Fix a lift A~\widetilde{A} of AA, we associate it with the integral homology H1(A~i,)H_{1}(\widetilde{A}\otimes_{i}{\mathbb{C}},{\mathbb{Z}}). The Frobenius map on the homology lattice comes from the lift of the Frobenius of A.A. This gives us a functor from the category of RM-abelian varieties to the category of {\mathbb{Z}}-modules with a “Frobenius” map. When all is said and done, however, there is no unique choice of a functor between the categories 𝒞h\mathcal{C}_{h} and h\mathcal{L}_{h}. As in [OS20], this ambiguity comes from possible CM-types on K.K. Suppose σ1,σ1,σ2,σ2,σa,σa\sigma_{1},\sigma^{\prime}_{1},\sigma_{2},\sigma^{\prime}_{2},\dots\sigma_{a},\sigma^{\prime}_{a} are the embeddings corresponding to the slope 1/21/2 part of a,a, then to give a possible CM type on KK, we could choose either one of σi\sigma_{i} or σi\sigma_{i}^{\prime} for 1ia1\leq i\leq a together with the embeddings corresponding to the slope 0 part. In each of the cases, this relates to the functoriality of the association between the two categories. For example, in the totally ramified case, we get a functorial map between 𝒞h\mathcal{C}_{h} and h\mathcal{L}_{h} once we choose a particular lift of AA out of 2a2^{a} possibilities. On the other hand, in the inert case, even though we have one canonical lift of A,A, the association of AA with the integral homology of its lift is not functorial. Once we restrict to one of 2a2^{a} subcategories of 𝒞h\mathcal{C}_{h} described in section 4, we get a functorial association.

1.2. Application to the size of isogeny classes.

The classification point of view of abelian varieties over finite fields has been utilised in many directions, many computational in nature. For example, Howe in [howe] studies polarisation building on Deligne’s work to show that there exists a principally polarised abelian variety in certain isogeny classes. In [ST18] Shankar and Tsimerman formulate and give evidence of certain conjectures regarding intersections of irreducible varieties and isogeny classes in 𝒜g\mathcal{A}_{g}, the moduli space of principally polarised abelian varieties. One of the applications of classification results to answer such questions is to estimate size of isogeny classes.

In this section, using the characterization in Theorem 1.4, we estimate the size of isogeny classes of a 𝔽q{\mathbb{F}}_{q} point, AL,𝔞(𝔽q)A\in\mathcal{H}_{L,\mathfrak{a}}({\mathbb{F}}_{q}) of the Hilbert modular variety. We define I(A,qn)I(A,q^{n}) to be the set of abelian varieties in L,𝔞(𝔽qn)\mathcal{H}_{L,\mathfrak{a}}({\mathbb{F}}_{q^{n}}) such that there exists an isogeny to AA over 𝔽¯q\overline{\mathbb{F}}_{q}. Denote by N(A,qn)N(A,q^{n}) the size of the set I(A,qn)I(A,q^{n}). We prove the following estimate for N(A,qn)N(A,q^{n}).

Theorem 1.5.

For all but finitely many nn, we have N(A,qn)=qn2(ga+o(1))N(A,q^{n})=q^{\frac{n}{2}(g-a+o(1))}.

We note that we get the leading term for all but finitely many nn, while the error is an asymptotic as nn goes to infinity.

In [AC02], Achter and Cunningham prove a precise formula for the size of isogeny classes of ordinary abelian varieties with RM. Their paper uses a different technique of orbital integrals. The estimate in [AC02, Theorem 3.1] agrees with our Theorem 1.5, when the pp-rank of the abelian variety is gg, that is, for ordinary abelian varieties with RM.

1.3. Strategy to prove the bounds for the size of the isogeny classes.

In order to find the lower bound for N(A,qn)N(A,q^{n}), we count the number of points in L,𝔞(𝔽qn)\mathcal{H}_{L,\mathfrak{a}}({\mathbb{F}}_{q^{n}}) with endomorphism by the smallest order Rn𝒪KR_{n}\subset\mathcal{O}_{K} that occurs as an endomorphism ring in the isogeny class of A.A. We characterize this ring in proposition 3.2. The classification theorem gives us a bijection between 𝒞(Rn)\mathcal{C}\ell(R_{n}) and the 𝔽qn{\mathbb{F}}_{q^{n}} points of L,𝔞\mathcal{H}_{L,\mathfrak{a}} with endomorphism ring RnR_{n}. Finding a lower bound therefore reduces to estimating the size of the class group. Furthermore, in section 5.2, we prove that this bound is sharp. We compute the upper bound by estimating size of the class group of all over orders of Rn.R_{n}. We note that RnR_{n} and all its over orders are Gorenstein, a property that gives a unique characterization of its over-orders in terms of the subgroups of 𝒪K/Rn{\mathcal{O}}_{K}/R_{n}, a cyclic group in our case (see Proposition 5.10).

1.4. Organization of the paper

Our first step is to construct canonical lift(s) of abelian varieties with RM in section 22. We then give the main theorem of classification in section 33, and finally, section 44 provides estimates for the size of isogeny classes.

2. Acknowledgements.

Our approach owes a substantial intellectual debt to the work of Oswal and Shankar. We are very grateful to Ananth Shankar for introducing this question to us and for his help throughout the project. We wish to thank Nathan Kaplan and Ziquan Yang for helpful conversations. We thank Jordan Ellenberg for many useful comments on the introduction. We were partially supported by NSF grant DMS-2100436.

3. Canonical lift

3.1. The simple case.

For this section, we fix a geometrically simple gg-dimensional abelian variety AA over 𝔽q{\mathbb{F}}_{q} that has RM by 𝒪L.{\mathcal{O}}_{L}. Let gag-a be the pp-rank of AA where 1ag1.1\leq a\leq g-1. We assume that pp is totally split in 𝒪L.{\mathcal{O}}_{L}. Let A[p]A[p^{\infty}] be the pp-divisible group of A.A. Let End0(A)\operatorname{End}^{0}(A) be the endomorphism algebra of AA defined over 𝔽q{\mathbb{F}}_{q}. Let K=(π)K={\mathbb{Q}}(\pi) be the field generated by the relative Frobenius of AA over 𝔽q{\mathbb{F}}_{q}. We denote by W(𝔽q)W({\mathbb{F}}_{q}), the ring of Witt vectors over 𝔽q{\mathbb{F}}_{q} and let W=W(k).W=W(k).

Our first step is to prove that the endomorphism algebra of AA is commutative. To that end, we state a useful lemma that describes the decomposition of A[p]A[p^{\infty}] over kk based on the splitting of p.p.

Lemma 3.1.

[AG04, Lemma 5.2.1] Let xL,𝔞(k)x\in\mathcal{H}_{L,\mathfrak{a}}(k). Then its pp-divisible group 𝒢x{\mathscr{G}}_{x} decomposes as product of pp-divisible groups with factors 𝒢x,𝔭{\mathscr{G}}_{x,{\mathfrak{p}}} corresponding to the primes lying above pp in 𝒪L{\mathcal{O}}_{L}. The Dieudonné module of 𝒢x,𝔭{\mathscr{G}}_{x,{\mathfrak{p}}} is an 𝒪L𝔭W(k){\mathcal{O}}_{L_{{\mathfrak{p}}}}\otimes W(k) module where 𝒪L𝔭{\mathcal{O}}_{L_{\mathfrak{p}}} is the completion of 𝒪L{\mathcal{O}}_{L} at 𝔭.\mathfrak{p}. Corresponding to each prime 𝔭{\mathfrak{p}}, the associated Dieudonné module is either ordinary or local-local. Moreover, the Dieudonné module of each local-local factor is free of rank two.

Let RR and SS be the endomorphism rings of AA and A[p]A[p^{\infty}] respectively over 𝔽q.{\mathbb{F}}_{q}. We know by Tate’s theorem that Rp=SR\otimes{\mathbb{Z}}_{p}=S.

Proposition 3.2.

The endomorphism ring RR and (hence) the p{\mathbb{Z}}_{p}-algebra SS is commutative. Moreover, SS admits a decomposition over 𝔽q{\mathbb{F}}_{q} into SétStorSssiS_{\text{\'{e}t}}\oplus S_{\operatorname{tor}}\oplus S^{i}_{\operatorname{ss}} for 1ia1\leq i\leq a where SssiS^{i}_{\operatorname{ss}} is rank two p{\mathbb{Z}}_{p}-algebra. Further, for each such i,i, the algebra SssiS^{i}_{\operatorname{ss}} is maximal in its field of fractions.

Proof.

We show that End0(A)=K\operatorname{End}^{0}(A)=K. Since the endomorphisms by 𝒪L{\mathcal{O}}_{L} are defined over 𝔽q{\mathbb{F}}_{q}, we get a splitting of the pp-divisible group and hence of its Dieuodonné module over 𝔽q{\mathbb{F}}_{q} into rank two modules. By Lemma 3.1, we get a decomposition into ordinary and supersingular factors 𝔻ord𝔻ss\mathbb{D}_{\operatorname{ord}}\oplus\mathbb{D}_{\operatorname{ss}} where 𝔻ss\mathbb{D}_{\operatorname{ss}} is a product of aa supersingular rank. Since pp is completely split in 𝒪L{\mathcal{O}}_{L}, we write 𝒪Lp=pga×pa{\mathcal{O}}_{L}\otimes{\mathbb{Z}}_{p}={\mathbb{Z}}_{p}^{g-a}\times{\mathbb{Z}}_{p}^{a}, where the pa{\mathbb{Z}}_{p}^{a} factor acts on 𝔻ss\mathbb{D_{\operatorname{ss}}}. Choose a basis of idempotents e1,e2,,eae_{1},e_{2},\dots,e_{a} of pa{\mathbb{Z}}_{p}^{a}. Since e1,e2,,eae_{1},e_{2},\dots,e_{a} are idempotent elements, we see that 𝔻ss{\mathbb{D}}_{\operatorname{ss}} is isomorphic to the direct sum e1𝔻ssea𝔻sse_{1}\cdot{\mathbb{D}}_{\operatorname{ss}}\oplus\cdots\oplus e_{a}\cdot{\mathbb{D}}_{\operatorname{ss}} such that the action of pa{\mathbb{Z}}_{p}^{a} respects the direct sum. Let f(x)f(x) denote the characteristic polynomial of π\pi over {\mathbb{Z}}. Since AA is geometrically simple, we have that f(x)=r(x)mf(x)=r(x)^{m} where r(x)r(x) is an irreducible polynomial over {\mathbb{Q}}. We have a decomposition of f(x)=ford(x)fss(x)f(x)=f_{\text{ord}}(x)f_{\operatorname{ss}}(x) over p{\mathbb{Z}}_{p} corresponding to the decomposition of the Dieudonné module. We note that fssf_{\operatorname{ss}} is product of factors fssif^{i}_{\operatorname{ss}} for 1ia1\leq i\leq a. We have a similar decomposition of r(x)r(x) as well. We know that the exponent mm is either 11 or 22 since each rssir^{i}_{\operatorname{ss}} is degree 2.2. Suppose that f(x)=r(x)2.f(x)=r(x)^{2}. Then each rssir^{i}_{\operatorname{ss}} must be a degree 11 polynomial. Let α\alpha be the root of say, rss1(x).r^{1}_{\operatorname{ss}}(x). Then, fss1(x)=(xα)2f^{1}_{\operatorname{ss}}(x)=(x-\alpha)^{2}. However, this implies that the minimal polynomial is reducible over {\mathbb{Q}}, that is, A𝔽q2A_{{\mathbb{F}}_{q^{2}}} is not simple, which is a contradiction. This proves that End0(A)\operatorname{End}^{0}(A) equals its center. Therefore, LL is the degree gg totally real field of K.K. Let ι\iota denote the complex conjugation of KK over L.L. Consider the gag-a ordinary primes in 𝒪L{\mathcal{O}}_{L} over p.p. These further split into two different primes 𝔭{\mathfrak{p}} and 𝔭{\mathfrak{p}}^{\prime} in KK such that 𝔭=ι𝔭{\mathfrak{p}}^{\prime}=\iota{{\mathfrak{p}}}. On the other hand, the supersingular primes either stay inert or ramified as they correspond to a degree 22 extension of p.{\mathbb{Q}}_{p}. This describes the Newton polygon of AA and hence gives a decomposition of the pp-divisible group over 𝔽q{\mathbb{F}}_{q} into 𝒢ét×𝒢tor×𝒢ss{\mathscr{G}}_{\text{\'{e}t}}\times{\mathscr{G}}_{\operatorname{tor}}\times{\mathscr{G}}_{\operatorname{ss}}. Therefore, we get the corresponding decomposition of SS as well. Finally, in order to show that SssiS^{i}_{\operatorname{ss}} is maximal, we follow a similar argument given at the end of [OS20]. Without loss of any generality, we do this for i=1.i=1. Suppose 𝒢{\mathscr{G}}^{\prime} is a pp-divisible group isogenous to 𝒢ss1{\mathscr{G}}^{1}_{\operatorname{ss}} such that End(𝒢)\operatorname{End}({\mathscr{G}}^{\prime}) is maximal. Let j:𝒢0𝒢ss1j:{\mathscr{G}}_{0}\rightarrow{\mathscr{G}}^{1}_{\operatorname{ss}} be any isogeny. We will show that End(𝒢)\operatorname{End}({\mathscr{G}}^{\prime}) preserves kerj,\ker j, proving that End(𝒢)End(𝒢ss1).\operatorname{End}({\mathscr{G}}^{\prime})\subset\operatorname{End}({\mathscr{G}}^{1}_{\operatorname{ss}}). Note that 𝒢0{\mathscr{G}}_{0} is a one dimensional pp-divisible group and so for each rr, it has a unique subgroup of order prp^{r} corresponding to the unique quotient of order prp^{r} of its Dieudonné module. Therefore, End(𝒢ss1)\operatorname{End}({\mathscr{G}}^{1}_{\operatorname{ss}}) preserves the kernel of jj which is a unique subgroup of order prp^{r} in 𝒢0{\mathscr{G}}_{0}. ∎

Let WW^{\prime} denote the rings of integers of a slightly ramified extension W(1/p)W^{\prime}(1/p) of W(1/p)W(1/p). That is, [W(1/p):W(1/p)][W^{\prime}(1/p):W(1/p)] is at most p1.p-1. The following proposition in [OS20] uses Grothendieck-Messing theory to construct lifts of AA to WW^{\prime}.

Lemma 3.3.

Let 𝒢ss{\mathscr{G}}_{\operatorname{ss}} denote a one dimensional pp-divisible group over 𝔽¯p.\overline{{\mathbb{F}}}_{p}. Suppose 𝒪End(𝒢ss){\mathcal{O}}\subset\operatorname{End}({\mathscr{G}}_{\operatorname{ss}}) is an integrally closed rank two p{\mathbb{Z}}_{p}-algebra. Then

  1. (1)

    If 𝒪{\mathcal{O}} is unramified, there exists a unique lift 𝒢ss{\mathscr{G}}_{\operatorname{ss}} to WW such that the action of 𝒪{\mathcal{O}} lifts.

  2. (2)

    If 𝒪{\mathcal{O}} is ramified, then there exists two lifts of 𝒢ss{\mathscr{G}}_{\operatorname{ss}} to WW^{\prime} such that the action of 𝒪{\mathcal{O}} lifts.

The proof of this proposition is given in [OS20]. Oswal and Shankar construct lifts of AA to WW^{\prime} by lifting its pp-divisible group. More precisely, Grothendieck-Messing theory gives a bijection between the lifts of 𝒢ss{\mathscr{G}}_{\operatorname{ss}} to WW^{\prime} with rank-one filtrations of the Dieudonné module 𝔻WW{\mathbb{D}}\otimes_{W}W^{\prime} that reduce to the kernel of the Frobenius on 𝔻{\mathbb{D}} modulo p.p. Moreover, in order to lift the endomorphisms, we need the filtration to be invariant under the action of the endomorphism algebra 𝒪{\mathcal{O}} of 𝒢ss{\mathscr{G}}_{\operatorname{ss}}. We can explicitly compute such filtrations depending on whether 𝒪{\mathcal{O}} is inert or ramified. We either have a unique choice of such a filtration or two such choices that correspond to the lift(s) of 𝒢ss{\mathscr{G}}_{\operatorname{ss}} to WW^{\prime}.

Therefore, in the geometrically simple case, Lemma 3.3 guarantees the following theorem.

Theorem 3.4.

We assume the notation described at the beginning of this section. The super-singular part A[p]ssA[p^{\infty}]_{\operatorname{ss}} of the pp-divisible group admits at-most 2a2^{a} lifts to W.W^{\prime}. Moreover, the lifts are characterized by the property that all of the endomorphisms of A[p]ssA[p^{\infty}]_{\operatorname{ss}} lift as well.

3.2. The non-simple case

In this section, we extend the lifting theorem to the non-simple case. As in the simple case, it is enough to establish the result for the supersingular part of the pp-divisible group.

Lemma 3.5.

Suppose AA is a non-simple abelian variety of dimension 2g2g with RM by 𝒪L{\mathcal{O}}_{L} over 𝔽q{\mathbb{F}}_{q}. Then AA is isogenous to BmB^{m}, an mm-fold product of a simple RM abelian variety BB defined over a possible field extension of 𝔽q{\mathbb{F}}_{q}.

Proof.

We show that if A=B1×B2A=B_{1}\times B_{2}, a product of two geometrically simple RM abelian varieties then B1B_{1} is isogenous to B2B_{2}. Let g1g_{1} be the dimension of B1B_{1} and g2g_{2} be the dimension of B2B_{2}. Without any loss of generality, we assume g1g2.g_{1}\geq g_{2}. Then by the definition of RM abelian varieties, B1B_{1} has RM by a totally real field, say L1L_{1}, of degree g1g_{1}, while B2B_{2} has RM by a totally real field L2L_{2} of degree g2.g_{2}. Now, if B1B_{1} were not isogenous to B2B_{2}, then End0A=End0(B1×B2)=End0(B1)×End0(B2)\operatorname{End}^{0}A=\operatorname{End}^{0}(B_{1}\times B_{2})=\operatorname{End}^{0}(B_{1})\times\operatorname{End}^{0}(B_{2}). We note that LL is a degree gg field in End0A\operatorname{End}^{0}A. However, End0(B1)×End0(B2)\operatorname{End}^{0}(B_{1})\times\operatorname{End}^{0}(B_{2}), a direct sum of CM fields can at most have dimension g1g_{1} totally real field embedded in it, which is a contradiction. ∎

Proposition 3.6.

Let AA be a non-simple abelian variety with RM by 𝒪L{\mathcal{O}}_{L} over 𝔽q.{\mathbb{F}}_{q}. Then the supersingular part of its pp-divisible group A[p]ssA[p^{\infty}]_{\operatorname{ss}} lifts to characteristic zero with all of its endomorphisms.

Proof.

We work over the algebraic closure kk. By the previous lemma, we know that AA is isogenous to BmB^{m} for some simple abelian variety BB that is RM by 𝒪L\mathcal{O}_{L^{\prime}} over kk. Let φ:BmA\varphi:B^{m}\rightarrow A be an isogeny and let φ[p]:B[p]mA[p]\varphi[p^{\infty}]:B[p^{\infty}]^{m}\rightarrow A[p^{\infty}] be the corresponding isogeny on the pp-divisible groups. Note that since pp is split in 𝒪L{\mathcal{O}}_{L}, it is completely split in 𝒪L.{\mathcal{O}}_{L^{\prime}}. Hence, we write B[p]ssm=1ia𝒢ssi××𝒢ssiB[p^{\infty}]_{\operatorname{ss}}^{m}=\prod_{1\leq i\leq a}{\mathscr{G}}^{i}_{\operatorname{ss}}\times\cdots\times{\mathscr{G}}^{i}_{\operatorname{ss}}, the mm-fold product of 𝒢ssi{\mathscr{G}}^{i}_{\operatorname{ss}}, that is, the factor of the super-singular part of B[p]B[p^{\infty}]. We know that pm{\mathbb{Z}}_{p}^{m} acts on (𝒢ssi)m({\mathscr{G}}^{i}_{\operatorname{ss}})^{m} for each 1ia.1\leq i\leq a. As in Proposition 3.2, we can choose a basis of idempotents e1,e2,,eme_{1},e_{2},\dots,e_{m} of pm{\mathbb{Z}}_{p}^{m}. Without loss of any generality consider i=1i=1 and let (𝔻1)m({\mathbb{D}}^{1})^{m} be the corresponding Dieudonné module of the pp-divisible group (𝒢ss1)m({\mathscr{G}}^{1}_{\operatorname{ss}})^{m}. Since e1,e2,,eme_{1},e_{2},\dots,e_{m} are idempotent elements, we see that (𝔻1)m({\mathbb{D}}^{1})^{m} is isomorphic to the direct sum e1(𝔻1)mem(𝔻1)me_{1}\cdot({\mathbb{D}}^{1})^{m}\oplus\cdots\oplus e_{m}\cdot({\mathbb{D}}^{1})^{m}. This gives us an isomorphism of (𝒢ss1)m({\mathscr{G}}^{1}_{\operatorname{ss}})^{m} with 1××m\mathscr{H}_{1}\times\cdots\times\mathscr{H}_{m} such that the action of p,{\mathbb{Z}}_{p}^{,} respects the direct sum. Here j\mathscr{H}_{j} is the pp-divisible groups corresponding to the Dieudonné module ej(𝔻1)e_{j}\cdot({\mathbb{D}}^{1}) for 1jm1\leq j\leq m. Now, let GG be the kernel of φ\varphi such that the order of GG is a power of p.p. Since the action of pm{\mathbb{Z}}_{p}^{m} respects the direct sum, therefore, GG1××GmG\simeq G_{1}\times\cdots\times G_{m} where GjjG_{j}\subset\mathscr{H}_{j} is of pp-power order for each 1jm1\leq j\leq m. Hence, A[p]1/G1m/Gm.A[p^{\infty}]\simeq\mathscr{H}_{1}/G_{1}\oplus\cdots\oplus\mathscr{H}_{m}/G_{m}. Since, each j/Gj\mathscr{H}_{j}/G_{j} is isogenous to j\mathscr{H}_{j} their endomorphism algebra is commutative, and therefore they lift to WW^{\prime}, along with all their endomorphisms by Lemma 3.3.

Theorem 3.7.

The pp-divisible group A[p]A[p^{\infty}] of AA has at most 2a2^{a} lifts to WW^{\prime} where gag-a is the pp-rank of the abelian variety. Furthermore, each lift 𝒢~\widetilde{{\mathscr{G}}} of A[p]A[p^{\infty}] is characterized by the property that every endomorphism of AA lifts.

Proof.

We know that for ordinary abelian varieties, there exists a Serre-Tate “canonical lift” of their pp-divisible group to W(k)W(k). We denote the corresponding lift of 𝒢ét×𝒢tor{\mathscr{G}}_{\text{\'{e}t}}\times{\mathscr{G}}_{\operatorname{tor}} by 𝒢ét~×𝒢tor~.\widetilde{{\mathscr{G}}_{\text{\'{e}t}}}\times\widetilde{{\mathscr{G}}_{\operatorname{tor}}}. Since each of the super-singular factors of A[p]A[p^{\infty}] has at most 22 lifts, we see that 𝒢ss{\mathscr{G}}_{\operatorname{ss}} has at most 2a2^{a} lifts. Moreover, the Serre-Tate canonical lift is characterized by the fact that every endomorphism also lifts. This is guaranteed by the previous Lemma 3.3 for the lift of the supersingular part as well. ∎

Definition 3.8.

We will refer to the lift(s) of A[p]A[p^{\infty}] (and hence of A) constructed in Theorem 3.7 as canonical lift(s).

4. Classification

Analogous to the work of Deligne for ordinary abelian varieties in [De69] and Oswal-Shankar [OS20] for (simple) almost ordinary abelian varieties over finite fields, we provide a classification for abelian varieties with RM by 𝒪L{\mathcal{O}}_{L}, not necessarily simple. However, for the sake of simplicity, we first consider AA to be a geometrically simple abelian variety with RM over 𝔽q{\mathbb{F}}_{q}. Let A~\widetilde{A} be one of the canonical lifts of AA. We fix an embedding i:Wi:W^{\prime}\hookrightarrow{\mathbb{C}} so that we can consider A~\widetilde{A} to be an abelian variety over {\mathbb{C}} by base changing. Consider its integral homology T(A)=H1(A~i,)T(A)=H_{1}(\widetilde{A}\otimes_{i}{\mathbb{C}},{\mathbb{Z}}) which is a free {\mathbb{Z}} module of rank 2g.2g. Since all the endomorphisms of AA lift, we have a Frobenius morphism FAF_{A} of T(A)T(A) that is the lift of the Frobenius of A.A. The pair (T(A),FA,λA,ιA)(T(A),F_{A},\lambda_{A},\iota_{A}) satisfies the following properties.

  1. (1)

    FAF_{A} acts semi-simply on T(A)T(A)\otimes{\mathbb{Q}}.

  2. (2)

    There exists VEnd(T(A))V\in\operatorname{End}_{\mathbb{Z}}(T(A)) such that FV=q=VF.F\circ V=q=V\circ F.

  3. (3)

    The characteristic polynomial h(x)[x]h(x)\in{\mathbb{Z}}[x] of FAF_{A} is a Weil qq-polynomial which is irreducible over .{\mathbb{Q}}. Furthermore, h(x)h(x) factors as hord1iahssih_{\operatorname{ord}}\prod_{1\leq i\leq a}h^{i}_{\operatorname{ss}} over p{\mathbb{Z}}_{p} where 1ia1\leq i\leq a. Each hssih^{i}_{\operatorname{ss}} is a degree 22 polynomial. The polynomial h(x)h(x) has gag-a roots that are qq-adic units, gag-a roots that have qq-adic valuation 11 and 2a2a roots with valuation 1/21/2 in ¯p.\overline{{\mathbb{Q}}}_{p}. Here gag-a is the pp-rank of the associated abelian variety. Each of the factors hssih^{i}_{\operatorname{ss}} is irreducible over p{\mathbb{Q}}_{p}. Otherwise, the supersingular factor of AA has endomorphism by p×p{\mathbb{Q}}_{p}\times{\mathbb{Q}}_{p}, but that is clearly not the case. Furthermore, hssih^{i}_{\operatorname{ss}} does not have ±q\pm{\sqrt{q}} as its roots, otherwise A×𝔽q𝔽q2A\times_{{\mathbb{F}}_{q}}{\mathbb{F}}_{q^{2}} won’t be simple.

  4. (4)

    Since all the endomorphisms of AA lift, we get an embedding ιA:𝒪LEnd(T(A)).\iota_{A}:{\mathcal{O}}_{L}\hookrightarrow\operatorname{End}_{{\mathbb{Z}}}(T(A)).

  5. (5)

    We note that T(A)T(A) decomposes over p{\mathbb{Q}}_{p} as TordTss1TssaT_{\operatorname{ord}}\oplus T^{1}_{\operatorname{ss}}\oplus\dots\oplus T^{a}_{\operatorname{ss}} where Tord=ker(hord(FA))T_{\operatorname{ord}}=\ker(h_{\operatorname{ord}}(F_{A})) and Tssi=ker(hss(FA))T^{i}_{\operatorname{ss}}=\ker(h_{\operatorname{ss}}(F_{A})) for each 1ia.1\leq i\leq a. Furthermore by the argument in 3.2, it follows that the endomorphism ring of T1/2T_{1/2} is the maximal order 𝒪ss{\mathcal{O}}_{\operatorname{ss}}.

  6. (6)

    The map λA\lambda_{A} identifies 𝔞\mathfrak{a} with a set of Riemann forms on T(A)T(A).

We note that for an abelian variety AA with RM isogenous to a power of a simple abelian variety BmB^{m}, since fA=(fB)mf_{A}=(f_{B})^{m}, its Deligne module T(A)T(A) is isogenous to the product i=1mT(B)\oplus_{i=1}^{m}T(B).

Definition 4.1.

(Deligne Module with RM) Let 𝔞\mathfrak{a} be a fractional ideal of L.L.

  1. (1)

    A pair (T,F,λ,ι)(T,F,\lambda,\iota) satisfying (1), (2), (3), (4), (5) and (6) is said to be a simple Deligne module with RM by 𝒪L{\mathcal{O}}_{L}.

  2. (2)

    An arbitrary Deligne module with RM is isogenous to a direct sum of simple Deligne modules.

  3. (3)

    A morphism of two Deligne modules with RM is given by a map φ:(T,F,λ,ι)(T,F,λ,ι)\varphi:(T,F,\lambda,\iota)\rightarrow(T^{\prime},F^{\prime},\lambda^{\prime},\iota^{\prime}) such that φ\varphi is compatible by the action of 𝒪L{\mathcal{O}}_{L} on both the modules and φF=Fφ.\varphi\circ F=F^{\prime}\circ\varphi. Analogous to abelian varieties, we have φλ=αλ\varphi^{*}\lambda^{\prime}=\alpha\lambda for some αL×\alpha\in L^{\times}.

  4. (4)

    An isogeny of Deligne modules with RM is defined to be a morphism φ:(T,F,λ,ι)(T,F,λ,ι)\varphi:(T,F,\lambda,\iota)\rightarrow(T^{\prime},F^{\prime},\lambda^{\prime},\iota^{\prime}) such that φ:TT\varphi\otimes{\mathbb{Q}}:T\otimes{\mathbb{Q}}\rightarrow T^{\prime}\otimes{\mathbb{Q}} is an isomorphism and φλ=αλ\varphi^{*}\lambda^{\prime}=\alpha\lambda for some αL×\alpha\in L^{\times}.

Definition 4.2.

(The category h\mathcal{L}_{h} and 𝒞h\mathcal{C}_{h}).

  1. (1)

    Let h(x)[x]h(x)\in{\mathbb{Z}}[x] denote the polynomial in property (3).(3). The we define h\mathcal{L}_{h} to be the category of simple Deligne modules with RM with Frobenius polynomial hh. Similarly, we define 𝒞h\mathcal{C}_{h} to be the category of polarized abelian varieties with RM over 𝔽q{\mathbb{F}}_{q} with Frobenius polynomial h(x).h(x). Morphisms in 𝒞h\mathcal{C}_{h} are also polarized, i.e., they preserve polarizations up to scaling by elements in L×L^{\times}.

  2. (2)

    We say 𝒞h\mathcal{C}_{h} or h\mathcal{L}_{h} is totally ramified if KssK_{\operatorname{ss}} (respectively inert) is a direct sum ramified (respectively) inert extensions of p{\mathbb{Q}}_{p}. We will use the term ramified for either of the categories when KssK_{\operatorname{ss}} has at least one ramified factor.

Proposition 4.3.

Every Deligne module (T,F,λ,ι)(T,F,\lambda,\iota) with RM by 𝒪L{\mathcal{O}}_{L} in h\mathcal{L}_{h} arises from an abelian variety with RM by 𝒪L{\mathcal{O}}_{L}.

Proof.

We first prove the result for a simple Deligne module, and the non-simple case follows by taking the product of the corresponding simple abelian varieties. So assume (T,F)(T,F) is a simple Deligne module. Since FF satisfies a Weil-qq polynomial, using the theorem of Honda and Tate, we can find a simple abelian variety AA over 𝔽q{\mathbb{F}}_{q} such that the characteristic polynomial of πA\pi_{A} is the same as that of F.F. We note that up to isogeny, we can always assume that AA has endomorphism by the maximal order in KK; hence, it has RM by 𝒪L{\mathcal{O}}_{L}. Therefore, (T(A),πA)(T,F).(T(A)\otimes{\mathbb{Q}},\pi_{A})\simeq(T\otimes{\mathbb{Q}},F). We wish to find an abelian variety BB,that is 𝒪L{\mathcal{O}}_{L}-isogenous to AA such that (T(B),πB)(T,F)(T(B),\pi_{B})\simeq(T,F). It is sufficient to assume that T(A)T.T(A)\subseteq T. Let H~A~\widetilde{H}\subset\widetilde{A} be the subgroup corresponding to T/T(A)T/T(A). Since TT and T(A)T(A) are both 𝒪L{\mathcal{O}}_{L}-modules, H~\widetilde{H} is stable under the action of 𝒪L{\mathcal{O}}_{L} and πA.\pi_{A}. Let HH be the subgroup obtained by reducing H~\widetilde{H} modulo p.p. If H~\widetilde{H} is of order prime-to-pp, then H~\widetilde{H} is a lift of HH as there exists a unique prime-to-pp subgroup in A~[n]\widetilde{A}[n] for some (n,p)=1(n,p)=1 that will reduce to HH modulo p.p. In the case when H~\widetilde{H} has order which is a power of p,p, we write H=Hét×Htor×HssH=H_{\text{\'{e}t}}\times H_{\operatorname{tor}}\times H_{\operatorname{ss}} where Hss=i=1aHiH_{\operatorname{ss}}=\prod_{i=1}^{a}H^{i}. This decomposition follows because HH is stable under the action of 𝒪L.{\mathcal{O}}_{L}. We note that the étale part and its dual lift uniquely to H~\widetilde{H}. For the supersingular components, say, Hss1H^{1}_{\operatorname{ss}}, we know that TpT\otimes{\mathbb{Z}}_{p} and T(A)pT(A)\otimes{\mathbb{Z}}_{p} both have endomorphisms by the maximal order 𝒪ss.{\mathcal{O}}_{\operatorname{ss}}. Therefore, T(A)=ω¯rTT(A)=\overline{\omega}^{r}T where ω¯\overline{\omega} is the uniformiser of 𝒪ss{\mathcal{O}}_{\operatorname{ss}} and rr is a positive integer. By construction, the subgroup HH corresponds to the kernel of ω¯r.\overline{\omega}^{r}. Since this endomorphism lifts uniquely to A~[p]\widetilde{A}[p^{\infty}], we see that H~ss1\widetilde{H}^{1}_{\operatorname{ss}} must be the kernel of the lift of ω¯r\overline{\omega}^{r}. Finally, if we consider A~/H~=B~,\widetilde{A}/\widetilde{H}=\widetilde{B}, then its reduction corresponds to A/H=BA/H=B such that (T(B),πB)=(T,F).(T(B),\pi_{B})=(T,F). By construction the polarisation data on Deligne modules is the same as the data of polarisation on RM-abelian varieties. That is, the data on both are parametrized by the fixed fractional ideal 𝔞\mathfrak{a}. This completes the proof of proposition 4.3. ∎

4.1. Totally ramified classes

Let 𝒞h\mathcal{C}_{h} denote a totally ramified isogeny class. As mentioned before, because any abelian variety in 𝒞h\mathcal{C}_{h} has 2a2^{a} lifts to W(k),W(k), the functor from 𝒞h\mathcal{C}_{h} to h\mathcal{L}_{h} depends on the choice of the lift of A.A. However, we will show that once we choose a lift A~\widetilde{A} of AA, then we can canonically determine a lift of every member in the isogeny class. This is because we can lift every subgroup canonically to A~\widetilde{A} as we prove below in proposition 4.4. Therefore, up to a choice of the lift, the association from 𝒞h\mathcal{C}_{h} to h\mathcal{L}_{h} is functorial.

Proposition 4.4.

Let 𝒞h\mathcal{C}_{h} be a totally ramified isogeny class. Suppose A𝒞hA\in\mathcal{C}_{h} and A~\widetilde{A} be one of its lifts to W(k).W(k). Let GAG\subset A be the kernel of an isogeny in 𝒞h\mathcal{C}_{h}. Then there exists a canonical subgroup G~A~\widetilde{G}\subset\widetilde{A} lifting GG.

Proof.

Since the isogenies in L,𝔞\mathcal{H}_{L,\mathfrak{a}} are compatible with the action of 𝒪L,{\mathcal{O}}_{L}, we see that GG remains fixed by its action and therefore, it splits as Get×Gtor×i=1aGssiG_{\operatorname{et}}\times G_{\operatorname{tor}}\times\prod_{i=1}^{a}G_{\operatorname{ss}}^{i}. It suffices to prove this result for each component. The result is certainly true for prime-to-pp subgroups of AA. Therefore we assume that the order of GG is a power of pp. We check this on the level of pp-divisible groups. We know that the canonical lift of A[p]=𝒢ét×𝒢tor×i=1a𝒢ssA[p^{\infty}]={\mathscr{G}}_{\operatorname{\text{\'{e}t}}}\times{\mathscr{G}}_{\operatorname{tor}}\times\prod_{i=1}^{a}{\mathscr{G}}_{\operatorname{\operatorname{ss}}} to be the product of the canonical lifts of its étale, multiplicative and the local-local part. The étale and multiplicative subgroups lift uniquely by the definition of the canonical lift, so we are left to check that every local-local finite flat subgroup of AA lifts uniquely to A~\widetilde{A}. Since Gss𝒢ssG_{\operatorname{\operatorname{ss}}}\subset{\mathscr{G}}_{\operatorname{\operatorname{ss}}}, it suffices to check that 𝒢ss{\mathscr{G}}_{\operatorname{\operatorname{ss}}} lifts to a subgroup G~ss𝒢~ss\widetilde{G}_{\operatorname{\operatorname{ss}}}\subset\widetilde{{\mathscr{G}}}_{\operatorname{\operatorname{ss}}}. For each 1ia1\leq i\leq a, the pp-divisible group 𝒢ssi{\mathscr{G}}_{\operatorname{ss}}^{i} is a connected and 11-dimensional group. Therefore it has a unique order prp^{r} subgroup for each positive integer rr. Since each 𝒢ss{\mathscr{G}}_{\operatorname{ss}} is ramified, we see that this has to be the kernel of the endomorphism ω¯r\overline{\omega}^{r} where ω¯\overline{\omega} is the uniformizer of 𝒪ssi{\mathcal{O}}_{\operatorname{ss}}^{i}. The canonical lift G~ssi\widetilde{G}_{\operatorname{ss}}^{i} has the property that every endomorphism of 𝒢ssi{\mathscr{G}}_{\operatorname{ss}}^{i} lifts and hence it follows that G~=G~ssi[ω¯r]\widetilde{G}=\widetilde{G}_{\operatorname{ss}}^{i}[\overline{\omega}^{r}] is the required lift of GG. ∎

Definition 4.5.

(Lifts of objects in the totally ramified isogeny class)

  1. (1)

    Let GAG\subset A be a finite flat subgroup as in Proposition 4.3. We define G~A~\widetilde{G}\subset\widetilde{A} to be the canonical lift as in the same proposition.

  2. (2)

    Let B𝒞hB\in\mathcal{C}_{h} be an abelian variety over 𝔽q{\mathbb{F}}_{q} such that there exists an isogeny in 𝒞h\mathcal{C}_{h} φ:AB\varphi:A\to B with kernel GG, i.e. it respects the 𝒪L\mathcal{O}_{L}-actions. We define B~\widetilde{B} to be the lift of BB given by A~/G~\widetilde{A}/\widetilde{G}.

Proposition 4.6.

The lift B~\widetilde{B} is the canonical lift of BB such that all the endomorphisms also of BB also lift to B~\widetilde{B}. Furthermore, B~\widetilde{B} does not depend on the choice of the isogeny φ\varphi.

Proof.

We check this on the level of pp-divisible groups. We have B~[p]=𝒢~B,et×𝒢~B,tor×i=12a𝒢~B,ssi\widetilde{B}[p^{\infty}]=\widetilde{{\mathscr{G}}}_{B,\operatorname{et}}\times\widetilde{{\mathscr{G}}}_{B,\operatorname{tor}}\times\prod_{i=1}^{2^{a}}\widetilde{{\mathscr{G}}}^{i}_{B,\operatorname{\operatorname{ss}}} where 𝒢~B,\tilde{{\mathscr{G}}}_{B,\heartsuit} is the canonical lift of 𝒢B,{\mathscr{G}}_{B,\heartsuit} for \heartsuit either et,tor\operatorname{et},\operatorname{tor} or ss\operatorname{ss}. By Proposition 4.4, the subgroup G~\widetilde{G} is the canonical lift of GG which is preserved under the action of 𝒪ss{\mathcal{O}}_{\operatorname{ss}}, which implies that it acts on B~[p]\widetilde{B}[p^{\infty}] as well. Therefore B~\widetilde{B} is a canonical lift of BB.

For the second claim, we take φ1\varphi_{1} and φ2\varphi_{2} to be two isogenies from AA to BB with kernels G1G_{1} and G2G_{2} respectively. Let B~i=A~/G~i\widetilde{B}_{i}=\widetilde{A}/\widetilde{G}_{i} for i=1,2i=1,2 be the corresponding abelian varieties isogenous to A~.\widetilde{A}. Without loss of generality we may assume that φ2=αφ1\varphi_{2}=\alpha\circ\varphi_{1} where αEnd(B)\alpha\in\operatorname{End}(B), by replacing φ2\varphi_{2} with an integer scalar multiple. Since every endomorphism in End(B)\operatorname{End}(B) lifts to End(B~)\operatorname{End}(\widetilde{B}) as a canonical lift, the result follows. ∎

Proposition 4.7.

Let BB and CC be abelian varieties in the totally ramified isogeny class 𝒞h\mathcal{C}_{h} of AA. Then every 𝔽q{\mathbb{F}}_{q}-isogeny φ:BC\varphi:B\to C lifts uniquely to an isogeny φ~:B~C~\widetilde{\varphi}:\widetilde{B}\to\widetilde{C}.

Proof.

Let ψ:AB\psi:A\to B be an 𝔽q{\mathbb{F}}_{q}-isogeny. Denote by GG and HH, the kernel of ker(ψ)\ker(\psi) and ker(φψ)\ker(\varphi\circ\psi) respectively. By Proposition 4.4, there are unique lifts of BB and CC respectively, namely B~=A~/G~\widetilde{B}=\widetilde{A}/\widetilde{G} and C=A~/H~C=\widetilde{A}/\widetilde{H}. Therefore we take φ~:A~/G~A~/H~.\widetilde{\varphi}:\widetilde{A}/\widetilde{G}\to\widetilde{A}/\widetilde{H}.

Hence for an 𝔽q{\mathbb{F}}_{q} isogeny φ:BC\varphi:B\to C, we have a map on the integral homology

(φ):H1(B~ϵ,)H1(C~ϵ,)\mathcal{F}(\varphi):H_{1}(\widetilde{B}\otimes_{\epsilon}{\mathbb{C}},{\mathbb{Z}})\to H_{1}(\widetilde{C}\otimes_{\epsilon}{\mathbb{C}},{\mathbb{Z}})

which is compatible with the Frobenius action. This gives a functor on the two categories in definition 4.2,

:𝒞hh\mathcal{F}:\mathcal{C}_{h}\to\mathcal{L}_{h}
AH1(A~ϵ,)A\mapsto H_{1}(\widetilde{A}\otimes_{\epsilon}{\mathbb{C}},{\mathbb{Z}})
Proposition 4.8.

Let AA be an abelian variety in the totally ramified class 𝒞h\mathcal{C}_{h}. Fix a lift A~\widetilde{A} of A.A. Then the functor \mathcal{F} mentioned above is an equivalence of categories.

Proof.

By Proposition 4.3, the functor \mathcal{F} is essentially surjective. The proof of [OS20, Proposition 3.4] shows that \mathcal{F} is fully faithful as well. ∎

4.2. Inert isogeny classes

We next consider the case where 𝒞h\mathcal{C}_{h} is inert. That is, the p{\mathbb{Z}}_{p}-algebra 𝒪ssi{\mathcal{O}}_{\operatorname{ss}}^{i} is inert for all 1ia1\leq i\leq a. Let A,BOb(𝒞h)A,B\in\operatorname{Ob}(\mathcal{C}_{h}) and φ:AB\varphi:A\to B be an 𝔽q{\mathbb{F}}_{q} isogeny with kernel GG. Let φ[p]:A[p]B[p]\varphi[p^{\infty}]:A[p^{\infty}]\to B[p^{\infty}] be the induced map of pp-divisible groups. We know that, the map φ\varphi commutes with actions by 𝒪L{\mathcal{O}}_{L} and therefore GG splits as a product Gord×1iaGissG_{\operatorname{ord}}\times\prod_{1\leq i\leq a}{G^{i}}_{\operatorname{ss}}. As a consequence, φ[p]\varphi[p^{\infty}] and its kernel also decompose accordingly.

Definition 4.9.

We define an equivalence relation \sim on the set of objects of Ob(𝒞h)\operatorname{Ob}(\mathcal{C}_{h}) as follows. For A,BOb(𝒞h)A,B\in\operatorname{Ob}(\mathcal{C}_{h}) we say ABA\sim B when some (hence every) 𝔽q{\mathbb{F}}_{q}-isogeny f:ABf:A\to B is such that if f[p]:A[p]B[p]f[p^{\infty}]:A[p^{\infty}]\to B[p^{\infty}] has all its supersingular components fssj[p]:Assj[p]Bssj[p]f_{\operatorname{ss}}^{j}[p^{\infty}]:A_{\operatorname{ss}}^{j}[p^{\infty}]\to B_{\operatorname{ss}}^{j}[p^{\infty}] such that the kernel ker(fssj[p])\ker(f_{\operatorname{ss}}^{j}[p^{\infty}]) has order equal to an even power of p.p.

The equivalence relation partitions Ob(𝒞h)\operatorname{Ob}(\mathcal{C}_{h}) into 2a2^{a} equivalence classes and we denote the corresponding full subcategories by 𝒞1,h,,𝒞2a,h\mathcal{C}_{1,h},\cdots,\mathcal{C}_{2^{a},h}. Therefore Ob(𝒞h)=i=12aOb(𝒞i,h)\operatorname{Ob}(\mathcal{C}_{h})=\bigcup_{i=1}^{2^{a}}\operatorname{Ob}(\mathcal{C}_{i,h}). With this notation, we make the following claim.

Proposition 4.10.

By restricting \mathcal{F} to each equivalence class 𝒞i,h\mathcal{C}_{i,h} for 1i2a1\leq i\leq 2^{a}, the association

i:A(T(A),F(A))\mathcal{F}_{i}:A\mapsto(T(A),F(A))

is functorial. Moreover, it induces an equivalence of categories.

The proof of this proposition follows verbatim as in [OS20, Proposition 3.4].

4.3. Ramified isogeny classes

Finally, the results in subsections 4.1 and 4.2 can be used to similarly prove that for a ramified isogeny class 𝒞h\mathcal{C}_{h}, the association A(T(A),F(A))A\mapsto(T(A),F(A)) gives an equivalence of categories once we fix a lift of an abelian variety AOb(𝒞h)A\in\operatorname{Ob}(\mathcal{C}_{h}) and define an equivalence relation on the objects as in the inert case. More precisely, suppose we have kk ramified components in KssK_{\operatorname{ss}} out of the aa supersingular components. Let A~\tilde{A} denote one of the 2k2^{k} canonical lifts of A.A. We call A~\tilde{A} the canonical lift of AA. Furthermore, we partition 𝒞h\mathcal{C}_{h} into 2ak2^{a-k} equivalence classes using the relation \sim defined in 4.2. Analogous to the inert case, we write Ob(𝒞h)=j=12akOb(𝒞j,h)\operatorname{Ob}(\mathcal{C}_{h})=\bigcup_{j=1}^{2^{a-k}}\operatorname{Ob}(\mathcal{C}_{j,h}).

Following the same proofs as in subsection 4.1, we can show that for each 1jak1\leq j\leq a-k, fixing a lift of an object AjOb(𝒞j,h)A_{j}\in\operatorname{Ob}(\mathcal{C}_{j,h}), fixes the lift of every element in the isogeny class every isogeny between the objects in 𝒞j,h.\mathcal{C}_{j,h}. Now, analogous to the inert case, we restrict \mathcal{F} to each subcategories 𝒞j,h\mathcal{C}_{j,h}. This gives a functor

j:𝒞j,hh\mathcal{F}_{j}:\mathcal{C}_{j,h}\to\mathcal{L}_{h}

for all jj such that 1j2ka.1\leq j\leq 2^{k-a}. Finally, as is proved before, we have the following proposition in the ramified case as well.

Proposition 4.11.

The functor j:𝒞j,hh\mathcal{F}_{j}:\mathcal{C}_{j,h}\to\mathcal{L}_{h} is an equivalence of categories for all 1j2ka.1\leq j\leq 2^{k-a}.

5. Size of isogeny classes

The goal of this section is to prove Theorem 1.5. Throughout the section, we fix a geometrically simple abelian variety AA over 𝔽q{\mathbb{F}}_{q} and define RnR_{n} as the smallest order in (πA){\mathbb{Q}}(\pi_{A}) containing πAn\pi_{A}^{n}, qn/πAnq^{n}/\pi_{A}^{n} and 𝒪L{\mathcal{O}}_{L} such that RnpR_{n}\otimes{\mathbb{Z}}_{p} contains 𝒪ss{\mathcal{O}}_{\operatorname{ss}}. By Theorem 1.4, the set of 𝔽qn{\mathbb{F}}_{q^{n}}-points in L,𝔞\mathcal{H}_{L,\mathfrak{a}} isogenous to AA with the endomorphism ring exactly RnR_{n} is in bijection with the set of isomorphism classes of finitely generated RnR_{n}-submodules of (πA){\mathbb{Q}}(\pi_{A}). In fact, the following lemma showsthat RnR_{n} is a Gorenstein ring. Hence, every such RnR_{n}-sub-module is invertible.

Lemma 5.1.

Let KK be a CM field and LL be its totally real field. Let DD be a Dedekind domain with the field of fractions LL. For any DD-subalgebra RR inside the integral closure of DD in KK such that R=KR\otimes{\mathbb{Q}}=K, RR is a Gorenstein ring.

Proof.

We know that a ring RR is Gorenstein if and only if all its localizations are Gorenstein. Thus, without loss of generality, we assume that DD is a principal ideal domain. Since RR is a free DD-module of rank 22, there exists a basis {1,α}\{1,\alpha\} such that R=D[α]R=D[\alpha]. Let fD[X]f\in D[X] be the minimal polynomial of α\alpha such that R=D[X]/fD[X]R=D[X]/fD[X]. Write α=(X mod f)R\alpha=(X\text{ mod }f)\in R then the complementary module R=f(α)1RR^{\dagger}=f^{\prime}(\alpha)^{-1}R is invertible. Therefore, [BL94, Proposition 2.7] implies that RR is a Gorenstein ring. ∎

Corollary 5.2.

RnR_{n} is a Gorenstein order inside 𝒪K{\mathcal{O}}_{K}.

Proof.

This comes directly from Lemma 5.1. One can also get proof by checking that RR\otimes{\mathbb{Z}}_{\ell} is monogenic for every p\ell\neq p in the context of [KK18] and applying the fact that monogenic orders are Gorenstein. ∎

Let IKI\subset K be an RnR_{n}-submodule. By Corollary 5.2, we see that II is invertible if there exists some fractional ideal JJ with IJ=RnIJ=R_{n}. We say two fractional ideals, I1I_{1} and I2I_{2} are equivalent if and only if there exists some aK×a\in K^{\times} such that I1=aI2.I_{1}=aI_{2}. We define the equivalence classes of all such fractional ideals of RnR_{n} to be the class group of RnR_{n}. We use the standard notation 𝒞(Rn)\mathcal{C}\ell(R_{n}) to denote the class group.

5.1. The lower bound

We now have a bijection between 𝔽qn{\mathbb{F}}_{q^{n}}-isomorphism classes of abelian varieties in 𝒞h\mathcal{C}_{h} having endomorphism ring exactly RnR_{n} and the ideal class group 𝒞(Rn)\mathcal{C}\ell(R_{n}). The class group 𝒞(Rn)\mathcal{C}\ell(R_{n}) is well approximated by the square root of its discriminant as nn approaches infinity. Therefore, it suffices to compute the (square root) of the discriminant of 𝒪L[πAn]{\mathcal{O}}_{L}[\pi_{A}^{n}] and the index of 𝒪L[πAn]{\mathcal{O}}_{L}[\pi_{A}^{n}] inside Rn.R_{n}. The discriminant of RnR_{n} is then approximated by dividing the two estimates. We recall that πA\pi_{A} is the Weil qq-integer of AA. Let α1,α2,,αg,α1¯,,αg¯\alpha_{1},\alpha_{2},\cdots,\alpha_{g},\overline{\alpha_{1}},\cdots,\overline{\alpha_{g}} be the image of πA\pi_{A} under the 2g2g complex embeddings of LL. Let θ1,,θg\theta_{1},\cdots,\theta_{g} denote the arguments of α1,,αg\alpha_{1},\cdots,\alpha_{g} respectively. Note that for each ii, αi¯=q/αi\overline{\alpha_{i}}=q/\alpha_{i} is the complex conjugate of αi\alpha_{i}. We have the following lemma which points out the fact that for all 1ig1\leq i\leq g and for all but finitely many nn, nθin\theta_{i} (hence sinnθi\sin n\theta_{i}) cannot be too small.

Lemma 5.3.

For every ϵ>0,\epsilon>0, and 1ig1\leq i\leq g, let ωn,i[0,2π)\omega_{n,i}\in[0,2\pi) be the unique real number such that nθiωn,i(mod2π)n\theta_{i}\equiv\omega_{n,i}\pmod{2\pi}. We have |ωn,i|>1qnϵ|\omega_{n,i}|>\frac{1}{q^{n\epsilon}} for all but finitely many nn.

Proof.

For each 1ig1\leq i\leq g, the argument θi\theta_{i} can be viewed as the logarithm of βi:=αi/αi¯\beta_{i}:=\alpha_{i}/\overline{\alpha_{i}}, which is same as log(αi)log(α¯i)\log(\alpha_{i})-\log(\overline{\alpha}_{i}). We first prove that log(αi)\log(\alpha_{i}) and log(α¯i)\log(\overline{\alpha}_{i}) are linearly independent over \mathbb{Q}. Suppose we have a,b0a,b\neq 0 such that

alog(αi)+blog(α¯i)=0.a\log(\alpha_{i})+b\log(\overline{\alpha}_{i})=0.

This is,

αia=α¯ib.\alpha_{i}^{a}=\overline{\alpha}_{i}^{-b}.

Comparing the qq-power we see that a=ba=-b. This only happens when bθi=kπb\theta_{i}=k\pi for some integer kk. Indeed, θi/π\theta_{i}/\pi\in{\mathbb{Q}} if and only if βi=αi/αi¯\beta_{i}=\alpha_{i}/\overline{\alpha_{i}} is a root of unity for each ii if and only if the Frobenius angle rank of AA equals zero. This happens and only if AA is supersingular. Since we have assumed that AA is geometrically simple, this is true only when g=1g=1 and AA is a supersingular elliptic curve. See section 3 of [DKZ] for details.

Using Baker’s theorem for the absolute value of linear combinations of logarithms of algebraic numbers, we see that there exists some absolute constant cc such that for every ϵ>0\epsilon>0,

|ωn,i|=|nlog(αi)n(logq/αi)|>1nc>1qnϵ|\omega_{n,i}|=|n\log(\alpha_{i})-n(\log q/\alpha_{i})|>\frac{1}{n^{c}}>\frac{1}{q^{n\epsilon}}

for all but finitely many nn. ∎

Lemma 5.4.

We have disc(𝒪L[πAn])=qn(g+o(1))\operatorname{disc}({\mathcal{O}}_{L}[\pi_{A}^{n}])=q^{n(g+o(1))} as n.n\to\infty.

Proof.

Consider the tower of extensions of rings 𝒪L𝒪L[πAn]{\mathbb{Z}}\subset{\mathcal{O}}_{L}\subset{\mathcal{O}}_{L}[\pi_{A}^{n}]. We know that

disc(𝒪L[πAn])=disc(𝒪L)2N𝒪L/(disc𝒪L(𝒪L[πAn]))\operatorname{disc}({\mathcal{O}}_{L}[\pi_{A}^{n}])=\operatorname{disc}({\mathcal{O}}_{L})^{2}N_{{\mathcal{O}}_{L}/{\mathbb{Z}}}(\operatorname{disc}_{{\mathcal{O}}_{L}}({\mathcal{O}}_{L}[\pi_{A}^{n}]))

See, for example, [Se79, Prop. III.8]. Consider a basis {1,πAn}\{1,\pi_{A}^{n}\} of 𝒪L[πAn]{\mathcal{O}}_{L}[\pi_{A}^{n}] over 𝒪L{\mathcal{O}}_{L}. We compute,

N𝒪L/(Disc𝒪L(𝒪L[πAn]))\displaystyle N_{{\mathcal{O}}_{L}/{\mathbb{Z}}}(\operatorname{Disc}_{{\mathcal{O}}_{L}}({\mathcal{O}}_{L}[\pi_{A}^{n}])) =N𝒪L/((πAnqn/πAn)2)\displaystyle=N_{{\mathcal{O}}_{L}/{\mathbb{Z}}}((\pi_{A}^{n}-q^{n}/\pi_{A}^{n})^{2})
=j=1g(αjnqn/αjn)2\displaystyle=\prod_{j=1}^{g}(\alpha_{j}^{n}-q^{n}/\alpha_{j}^{n})^{2}
=j=1g4qn(isinnθj)2\displaystyle=\prod_{j=1}^{g}4q^{n}(i\sin n\theta_{j})^{2}

This along with the Lemma 5.3, completes the proof.

Lemma 5.5.

For all but finitely many nn, we have 𝒞(Rn)=qn2(ga+o(1))\mathcal{C}\ell(R_{n})=q^{\frac{n}{2}(g-a+o(1))}.

Proof.

In order to estimate the index of 𝒪L[αn]{\mathcal{O}}_{L}[\alpha^{n}] inside RnR_{n}, we compute the corresponding index locally at pp. Without loss of any generality, we do the computation for n=1n=1. Let p(x)p(x) denote the minimal polynomial of α\alpha over 𝒪L{\mathcal{O}}_{L} which we know is degree 2.2. Let σ1,,σg\sigma_{1},\cdots,\sigma_{g} denote the restrictions of gg embeddings L¯pL\hookrightarrow\overline{\mathbb{Q}}_{p} to 𝒪L.{\mathcal{O}}_{L}. Then we have

𝒪L[αn]p\displaystyle{\mathcal{O}}_{L}[\alpha^{n}]\otimes{\mathbb{Z}}_{p} =𝒪L[x]h(x)p\displaystyle=\frac{{\mathcal{O}}_{L}[x]}{h(x)}\otimes{\mathbb{Z}}_{p}
=p[x]σ1(p(x))p[x]σg(p(x))\displaystyle=\frac{{\mathbb{Z}}_{p}[x]}{\sigma_{1}(p(x))}\oplus\cdots\oplus\frac{{\mathbb{Z}}_{p}[x]}{\sigma_{g}(p(x))}

since pp splits completely in 𝒪L{\mathcal{O}}_{L}. Let f(x)f(x) be the minimal polynomial of α\alpha over p{\mathbb{Z}}_{p}. We know that f(x)=σ1(p(x))σg(p(x))f(x)=\sigma_{1}(p(x))\cdots\sigma_{g}(p(x)) gives the decomposition of p(x)p(x) over 𝒪Lp.{\mathcal{O}}_{L}\otimes{\mathbb{Z}}_{p}. Without loss of any generality, we take σ1(p(x)),,σa(p(x))\sigma_{1}(p(x)),\cdots,\sigma_{a}(p(x)) to be the aa supersingular factors of p(x)p(x). Let δ1,1,δ1,2,,δa,1,δa,2\delta_{1,1},\delta_{1,2},\cdots,\delta_{a,1},\delta_{a,2} to be the corresponding roots of the polynomials. Let gi(x)g_{i}(x) denote the polynomial with roots δi,1/q12\delta_{i,1}/q^{\frac{1}{2}} and δi,2/q12\delta_{i,2}/q^{\frac{1}{2}}. The index of 𝒪F[α]p{\mathcal{O}}_{F}[\alpha]\otimes{\mathbb{Z}}_{p} inside RnpR_{n}\otimes{\mathbb{Z}}_{p} is approximated by dividing the product of the discriminants of σi(p(x)\sigma_{i}(p(x) and gi(x)g_{i}(x) which gives the following expression.

i=1a(δi,1δi,2)(δi,1/q12δi,2/q12)=qa2\prod_{i=1}^{a}\frac{{(\delta_{i,1}-\delta_{i,2})}}{(\delta_{i,1}/q^{\frac{1}{2}}-\delta_{i,2}/q^{\frac{1}{2}})}=q^{\frac{a}{2}}

The lemma now follows by putting together the above calculation with Lemma 5.4. ∎

5.2. The upper bound

In this section, we show that the lower bound computed in the above section is indeed sharp. We will do this by approximating the class groups of all over orders of RnR_{n} in order to count the 𝔽qn{\mathbb{F}}_{q^{n}}-isomorphism classes of abelian varieties in 𝒞h\mathcal{C}_{h} with endomorphism ring 𝒪{\mathcal{O}} where 𝒪{\mathcal{O}} is any arbitrary over-order of RnR_{n}. We recall some standard notation first.

Definition 5.6.

Let RR be an order in a number field KK. Let (R)\mathcal{I}(R) be the set of all fractional ideals of RR The ideal class monoid of RR, denoted by ICM(R) is defined to be

ICM(R)=(R)/\operatorname{ICM}(R)=\mathcal{I}(R)/\simeq

where IJI\simeq J if and only if there exists an αK\alpha\in K^{*} such that I=αJI=\alpha J.

Definition 5.7.

For all invertible factional ideals, we analogously define the Picard group of RR to be

𝒞(R)={invertible fractional ideal I(R)}/𝒫(R)\mathcal{C}\ell(R)=\{\text{invertible fractional ideal }I\in\mathcal{I}(R)\}/\mathcal{P}(R)

where 𝒫(R)\mathcal{P}(R) is the group of of principal fractional RR-ideals.

If two fractional RR-ideals are isomorphic, then they have the same multiplicator ring. It follows that

R𝒪𝒪K𝒞(𝒪)ICM(R)\bigsqcup_{R\subset{\mathcal{O}}\subset{\mathcal{O}}_{K}}\mathcal{C}\ell(\mathcal{O})\subseteq\operatorname{ICM}(R)

where the disjoint union is taken over all over-orders 𝒪{\mathcal{O}} of RR.

Definition 5.8.

An order RR is called a Bass order if every over-order of RR is Gorenstein, or equivalently, if 𝒪K/R{\mathcal{O}}_{K}/R is cyclic, which is to say 𝒪K=R+xR{\mathcal{O}}_{K}=R+xR for some x𝒪Kx\in{\mathcal{O}}_{K}.

We have the following lemma, which gives another useful characterization for Bass orders.

Lemma 5.9.

The following statements are equivalent:

  • a.

    RR is a Bass order,

  • b.

    R𝒪𝒪K𝒞(𝒪)=ICM(R).\bigsqcup_{R\subset{\mathcal{O}}\subset{\mathcal{O}}_{K}}\mathcal{C}\ell(\mathcal{O})=\operatorname{ICM}(R).

For proof of this proposition, see [Mar20, Proposition 3.7]. Finally, we have the following proposition that gives us an upper bound on the size of the isogeny classes.

Proposition 5.10.

For any positive integer nn, we have

N(A,qn)(qn2)(ga+o(1))N(A,q^{n})\leq(q^{\frac{n}{2}})^{(g-a+o(1))}
Proof.

From Lemma 5.1 we deduce that every order between 𝒪F[αn]{\mathcal{O}}_{F}[\alpha^{n}] and 𝒪K{\mathcal{O}}_{K} is Gorenstein, hence every element of ICM(Rn)\operatorname{ICM}(R_{n}) lies in 𝒞(𝒪)\mathcal{C}\ell({\mathcal{O}}) for some unique order 𝒪{\mathcal{O}} over RnR_{n}. Therefore by Lemma 5.9 we have

Rn𝒪𝒪K𝒞(𝒪)=ICM(Rn).\bigsqcup_{R_{n}\subset{\mathcal{O}}\subset{\mathcal{O}}_{K}}\mathcal{C}\ell(\mathcal{O})=\operatorname{ICM}(R_{n}).

Let h(𝒪)h({\mathcal{O}}) be the class number of 𝒪{\mathcal{O}}. We note that the class group of the smallest order RnR_{n} has the largest size. By the lattice isomorphism theorem, subgroups of 𝒪K{\mathcal{O}}_{K} containing RnR_{n} are in bijection with subgroups of 𝒪K/Rn{\mathcal{O}}_{K}/R_{n}. Since RnR_{n} is Bass, we have that 𝒪K/Rn{\mathcal{O}}_{K}/R_{n} is cyclic. Therefore its subgroups are indexed by dind\mid i_{n} where in=|𝒪K/Rn|.i_{n}=|{\mathcal{O}}_{K}/R_{n}|. Putting everything together, we see that

N(A,qn)=dinh(𝒪d)ino(1)h(Rn)=(qn2)(ga+o(1))N(A,q^{n})=\sum_{d\mid i_{n}}h({\mathcal{O}}_{d})\leq i_{n}^{o(1)}h(R_{n})=(q^{\frac{n}{2}})^{(g-a+o(1))}

where 𝒪d{\mathcal{O}}_{d} is the order corresponding to the subgroup of order dd in 𝒪K/Rn{\mathcal{O}}_{K}/R_{n}. ∎

5.3. The proof of Theorem 1.5

Let Rn+=RnLR_{n}^{+}=R_{n}\cap L be the totally real part of RnR_{n}. Since RnR_{n} contains 𝒪L,{\mathcal{O}}_{L}, we have Rn+=𝒪L.R_{n}^{+}={\mathcal{O}}_{L}.

Proposition 5.11.

The subset I(A,qn)I(A,q^{n}) with endomorphism ring equal to RnR_{n} is either empty, or admits a bijective map if 𝒞h\mathcal{C}_{h} is totally ramified, or admits a 2k2^{k}-to-one map if there are kk inert factors, onto the kernel of the norm map

N:𝒞(Rn)𝒞+(Rn+),N:\mathcal{C}\ell(R_{n})\to\mathcal{C}\ell^{+}(R_{n}^{+}),

where 𝒞+(Rn+)\mathcal{C}\ell^{+}(R_{n}^{+}) is the narrow class group of the totally real order Rn+R_{n}^{+}.

Proof.

The proof follows verbatim as the proof of [ST18, Proposition 3.5]. ∎

One last thing to finish the proof of Theorem 1.5 is to prove that there exists some principally polarized abelian varieties with endomorphism ring equal to RnR_{n}.

Let 𝔞\mathfrak{a}, 𝔟\mathfrak{b} be fractional ideals of 𝒪L\mathcal{O}_{L} and z+z\in\mathcal{H}^{+} . For a lattice Λz\Lambda_{z} given by

Λz=𝔞z𝔟1,\Lambda_{z}=\mathfrak{a}\cdot z\oplus\mathfrak{b}\cdot 1,

recall that a polarization on Az=g/ΛzA_{z}=\mathbb{C}^{g}/\Lambda_{z} is given by a Riemann form on 𝔞𝔟\mathfrak{a}\oplus\mathfrak{b}:

Er,z((x1,y1),(x2,y2)):=TrL/(r(x1y2x2y1))E_{r,z}((x_{1},y_{1}),(x_{2},y_{2})):=\operatorname{Tr}_{L/{\mathbb{Q}}}(r(x_{1}y_{2}-x_{2}y_{1}))

for some r(𝒟L/𝔞𝔟)1r\in(\mathcal{D}_{L/{\mathbb{Q}}}\mathfrak{a}\mathfrak{b})^{-1}. Therefore we have

Hom𝒪L(Az,Az)symm=(𝒟L/𝔞𝔟)1\operatorname{Hom}_{\mathcal{O}_{L}}(A_{z},A_{z}^{\vee})^{symm}=(\mathcal{D}_{L/{\mathbb{Q}}}\mathfrak{a}\mathfrak{b})^{-1}

and the set of polarization is identified with elements in (𝒟L/𝔞𝔟)1,+(\mathcal{D}_{L/{\mathbb{Q}}}\mathfrak{a}\mathfrak{b})^{-1,+}. Moreover, if this holds, then the degree of the polarization on AzA_{z} is given by Er,zE_{r,z} is Norm(r𝒟L/𝔞𝔟)(r\mathcal{D}_{L/{\mathbb{Q}}}\mathfrak{a}\mathfrak{b}). In particular, there exists an rr such that Er,zE_{r,z} is principal if and only if 𝔞𝔟=𝒟L/1\mathfrak{a}\mathfrak{b}=\mathcal{D}_{L/{\mathbb{Q}}}^{-1} in the narrow class group 𝒞(L)+\mathcal{C}\ell(L)^{+}. In other words, we may take 𝔞𝔟=𝒟L/1𝒪L.\mathfrak{a}\oplus\mathfrak{b}=\mathcal{D}^{-1}_{L/{\mathbb{Q}}}\oplus\mathcal{O}_{L}. [Goren, Chapter 2; Corollary 2.10].

Proposition 5.12.

For every positive integer nn, there exists some principally polarized abelian varieties with endomorphism ring equal to RnR_{n}.

Proof.

Let 𝔦\mathfrak{i} be a fractional ideal of 𝒪L\mathcal{O}_{L}. A polarization on the abelian variety corresponding to RnR_{n} can be realized as a {\mathbb{Z}}-valued bilinear form on RnR_{n}, which is of the form:

x,yTraceK/(aixy¯),\langle x,y\rangle\mapsto\operatorname{Trace}_{K/{\mathbb{Q}}}(aix\bar{y}),

where,

  • (1)

    y¯\bar{y} is the complex conjugation of yy.

  • (2)

    ,\langle-,-\rangle restricted to 𝔞\mathfrak{a} is integral.

  • (3)

    aa is totally positive element in 𝔦\mathfrak{i}.

In order to construct a principally polarized abelian variety, the argument above requires that 𝔦=𝒟L/1.\mathfrak{i}=\mathcal{D}_{L/{\mathbb{Q}}}^{-1}. Let RnR_{n}^{\vee} be the dual of RnR_{n} under the bilinear form induced by the trace map (x,y)TraceK/(x,y).(x,y)\mapsto\operatorname{Trace}_{K/{\mathbb{Q}}}(x,y). Let Rn+=RnLR_{n}^{+}=R_{n}\cap L be the totally real part of RnR_{n}. We know that Rn+=𝒪L.R_{n}^{+}={\mathcal{O}}_{L}. Consider RnR_{n} as a imaginary extension of degree 22 over 𝒪L{\mathcal{O}}_{L}. Therefore Rn=2bi(𝒪L)R_{n}^{\vee}=2bi({\mathcal{O}}_{L})^{\vee} for some positive integer bb. Now, (𝒪L)=𝒟L/1({\mathcal{O}}_{L})^{\vee}=\mathcal{D}_{L/{\mathbb{Q}}}^{-1}. Consider the bilinear form on RnR_{n} given by

(x,y)TraceK/(λnxy¯),(x,y)\mapsto\operatorname{Trace}_{K/{\mathbb{Q}}}(\lambda_{n}x\bar{y}),

where λnRn.\lambda_{n}\in R_{n}^{\vee}. Therefore, we can get a totally positive λn\lambda_{n} once we choose an element cc from 𝒟L/1\mathcal{D}_{L/{\mathbb{Q}}}^{-1} such that φ(2bci)/i\varphi(2bci)/i is totally positive for all φΦ\varphi\in\Phi. This gives a principal polarization on the abelian variety with endomorphism ring equal to RnR_{n}. ∎

Proof of Theorem 1.5.

The theorem now follows from Lemma 5.5, Proposition 5.10, Proposition 5.11, Proposition 5.12, and the fact that the size of 𝒞+(Rn+)\mathcal{C}\ell^{+}(R_{n}^{+}) does not grow with nn in our case because Rn+=𝒪L.R_{n}^{+}={\mathcal{O}}_{L}.

References