Traverso's Isogeny Conjecture for p-Divisible Groups.
MARC-HUBERTNICOLE(*) - ADRIANVASIU(**)
To Carlo Traverso, for his 60th birthday
ABSTRACT - Let k be an algebraically closed field of characteristic p>0. Let c;d2N. Letbc;d1be the smallest integer such that for any twop-divisible groupsHandH0overkof codimensioncand dimensiondthe following assertion holds: IfH[pbc;d] andH0[pbc;d] are isomorphic, thenHandH0are isogenous. We show thatbc;dl cd
cd
m. This proves Traverso's isogeny conjecture forp-di- visible groups overk.
1. Introduction.
Let p2N be a prime. Letk be an algebraically closed field of char- acteristic p. Let c;d2N. Let H be a p-divisible group over k of codi- mensioncand dimensiond; its height iscd. Itis well-known (see [Ma], [Tr1, Thm. 3], [Tr2, Thm. 1], [Va, Cor. 1.3], and [Oo3, Cor. 1.7]) that there exists a minimal numbernH2Nsuch thatHis uniquely determined up to isomorphism byH[pnH] (i.e., ifH0is a p-divisible group overksuch that H0[pnH] is isomorphic toH[pnH], thenH0is isomorphic toH). For instance, we have nHcd1 (cf. [Tr1, Thm. 3]). This implies that there exists a minimal numberbH2Nsuch that theisogeny classofHis uniquely de- termined by H[pbH]. We call bH the isogeny cutoff of H. We have 1bHnH. Traverso speculated the following (cf. [Tr3, § 40, Conj. 5]):
(*) Indirizzo dell'A.: University of Tokyo, Department of Mathematical Sciences, Komaba, 153-8914, Tokyo, Japan.
E-mail: nicole@ms.u-tokyo.ac.jp
(**) Indirizzo dell'A.: University of Arizona, Department of Mathematics, 617 N. Santa Rita Ave., P.O. Box 210089, Tucson, AZ 85721-0089, U.S.A.
E-mail: adrian@math.arizona.edu
CONJECTURE 1.1. The isogeny cutoff bH is bounded from above by l cd
cd
mi.e., we havebHl cd cd
m.
The goal of this paper is to prove an optimal variant of the Conjecture:
THEOREM1.2. Let c;d2Nand let bc;d1be the smallest integer such that for any two p-divisible groups H and H0over k of codimension c and dimension d the following assertion holds:If H[pbc;d] and H0[pbc;d] are isomorphic, then H and H0are isogenous. Then bc;dl cd
cd
mand there- fore we have bHl cd
cd m.
In Section 2, we introduce notations and basic invariants which pertain top-divisible groups overkand which allow us to obtain a practical upper bound ofbH(see Corollary 2.13). In Section 3, we prove Theorem 1.2; see Example 2.10 for a simpler proof in the particular case whenHis isosimple.
If thea-number ofHis at most 1, then the inequalitybHl cd cd
mis in essence due to Traverso (cf. [Tr3, Thm. of § 21, and § 40]; see Theorem 3.1).
We refer to Theorem 2.2 and Example 3.2 for concrete examples with bH1 and withbHl cd
cd
m(respectively).
2. Estimates of the isogeny cutoffbH. Letr:cdandj:lcd
r
m2N. LetW(k) be the ring of Witt vectors with coefficients ink. Lets be the Frobenius automorphism of W(k) in- duced fromk. Let(M;f) be the (contravariant) Dieudonne module ofH.
ThusMis a freeW(k)-module of rankrandf:M!Mis as-linear en- domorphism such that we havepMf(M). We havecdimk(f(M)=pM) andddimk(M=f(M)). Let w:M!M be the s 1-linear Verschiebung map off; we havefwwfp1M.
The DieudonneÂ-Manin classification ofF-isocrystals over k(see [Di, Thms. 1 and 2], [Ma, Ch. II, §4], [De, Ch. IV]) states that:
there exists m2N such that we have a direct sum decomposition Mh1
p i;f
m
i1(Mi;f) into simpleF-isocrystals overk, and
there exist numbers ci;di2N[ f0g which satisfy the inequality ri:dici>0, which are relative prime (i.e., g:c:d:(ci;di)1), and for which there exists aB(k)-basis forMiformed by elements fixed byp difri.
The unique slope of (Mi;f) is ai:di
ri2[0;1]\Q. To fix ideas, we assume thata1a2 am. Eachi2 f1;. . .;mggives a slopeaiwith multiplicityri. The Newton polygonNHof His a continuous, piecewise linear, upward convex function NH:[0;r]![0;d] which for all i2 f1;. . .;mghas slopeaion the interval iP1
`1r`;Pi
`1r`
. As t he fieldkis algebraically closed, the isogeny class of His uniquely determined by the Newton polygonNH. The finite setNc;dof Newton polygons we thus obtain by varyingH, can be partially ordered as follows: we say thatN1
lies above (resp. strictly above) N2 if for all t2[0;r] we have N1(t) N2(t) (resp. if for all t2[0;r] we have N1(t) N2(t) and moreoverN16 N2). This partial order is convenient when studying the variation of the Newton polygon in families. We will use the notationN to denote the Newton polygon of, whereis ap-divisible group over a field that contains k.
Let Hci;di be the p-divisible group over kwhose Dieudonne module (Mci;di;fci;di) has the property that there exists aW(k)-basisfe0;. . .;eri 1g for Mci;di such that we have fci;di(el)edil if l2 f0;. . .;ci 1g and fci;di(el)pedilifl2 fci;. . .;ri 1g(here the lower right indices ofeare taken modulo ri). Itis well-known thatHci;di has the following two properties (see [dJO, Lem. 5.4]): (i) it has codimension ci, dimension di, and unique Newton polygon slope ai, and (ii) its endomorphism ring End(Hci;di) is the maximal order in the simple Qp-algebra End(Hci;di)
Zp
Qpof invariantai2Q=Z. In addition, these two properties determine Hci;di up to isomorphism (see [Ma] and [Oo4, § 1.1]). Let H0:Qm
i1Hci;di; we have NH0 NH. The Dieudonne module of H0 is (M0;f0):Qm
i1(Mci;di;fci;di).
DEFINITION 2.1 (Oort). We say that a p-divisible group H over k is minimal if it is isomorphic to the p-divisible group H0 determined un- iquely from the Newton polygonNHof H.
Thep-divisible groupH0is theunique(up to isomorphism) minimalp- divisible overkthat is isogenous toH.
THEOREM2.2. A minimal p-divisible group H over k is determined up to isomorphism by H[p](thus we have bHnH1).
PROOF. For the isoclinic case (i.e., when the pairs (ci;di) do notdepend oni) we refer to [Va1, Exa. 3.3.6]. For the general case we refer to either
[Oo4, Thm. 1.2] or [Va2, Thm. 1.8]. p
DEFINITION2.3. By the minimal height qHof a p-divisible group H over k we mean the smallest non-negative integer qHsuch that there exists an isogenyc0 :H!H0to a minimal p-divisible group, whose kernelKer(c0) is annihilated by pqH.
DEFINITION2.4. By the a-number aHof a p-divisible group H over k we mean the number dimk(ap;H[p])dimk(M=f(M)w(M))2N[ f0g, whereapis the local-local group scheme of order p.
See [Oo2, Prop. 2.8] for a proof of the following specialization trick.
PROPOSITION2.5. For everyp-divisible groupHoverk, there exists ap- divisible groupHoverk[[x]]that has the following two properties:
(i) its fibre over k is H;
(ii) if k((x))is an algebraic closure of k((x)), thenHk((x))has the same Newton polygon as H and its a-number is at most1.
DEFINITION2.6. By the weak isogeny cutoff of a p-divisible group H over k we mean the smallest number~bH2Nsuch that the following two properties hold:
(i) if H0 is a p-divisible group over k such that H0[p~bH] is iso- morphic to H[p~bH], then its Newton polygonNH0is not strictly aboveNH; (ii) there exists a p-divisible groupHover k[[x]]that has the fol- lowing three properties:
(ii.a) its fibre over k is H;
(ii.b) the fibreHk((x))over k((x))has the same Newton polygon as H;
(ii.c) the isogeny cutoff of Hk((x)) is at most~bH and the a-number of Hk((x))is at most1.
Note that property (i) holds for any level greater or equal tobH. Thus the existence of the number~bH2Nis implied by Proposition 2.5.
FACT2.7. If Htis the Cartier dual of H, then qHqHt, bHbHt, and
~bH~bHt.
PROOF. This follows from Cartier duality (the Cartier dual of Hci;diis
Hdi;ci). p
LEMMA2.8. The isogeny cutoff bHis the smallest natural number such that for every element g2GLM(W(k)) congruent to 1M modulo pbH, the Dieudonne module(M;gf)is isogenous to(M;f).
PROOF. Lett2N. The arguments proving either [Va1, Cor. 3.2.3] or [NV, Thm. 2.2 (a)] show that: (i) ifg2GLM(W(k)) is congruentto 1Mmodulo pt, then every p-divisible group Hg over k whose Dieudonne module is isomorphic to (M;gf), has the property thatHg[pt] is isomorphic toH[pt], and (ii) if ap-divisible groupH0overkhas the property thatH0[pt] is iso- morphic toH[pt], then there existsg2GLM(W(k)) congruentto 1Mmodulo ptand such that the Dieudonne module ofH0is isomorphic to (M;gf). The Lemma follows from these two statements and the very definition of the
isogeny cutoffbH. p
LEMMA2.9. For every p-divisible group H over k, the isogeny cutoff bHis bounded from above by the minimal height qH plus 1 i.e., we have bHqH1.
PROOF. Letc:H!H0 be an isogeny whose kernel is annihilated by pqH. Let(M0;f0),!(M;f) be the monomorphism of Dieudonne modules associated to the isogenyc; we will identifyM0 with its image under this monomorphism. We have pqHMM0M. Let g2GLM(W(k)) be con- gruentto 1MmodulopqH1. We writeg1MpqH1e, wheree2End(M).
We have pqH1e(M0)pqH1e(M)pqH1MpM0. Thus we have g2GLM0(W(k)) and moreovergis congruentto 1M0 modulop. Therefore the reductions modulopof the two triples (M0;f;w) and (M0;gf;wg 1) co- incide. As the Dieudonne module (M0;f) is minimal (being isomorphic to (M0;f0)), from Theorem 2.2 we get that it is isomorphic to (M0;gf). The isogeny class of (M;gf) (resp. of (M;f)) is the same as of (M0;gf) (resp. of (M0;f)). From the last two sentences we get that the Dieudonne modules (M;gf) and (M;f) are isogenous. From this and Lemma 2.8, we get that
bHqH1. p
EXAMPLE2.10. Suppose thatm1 i.e.,His an isosimplep-divisible group. Let u0:H0!H0 be an endomorphism such that End(H0)
W(Fpr)[u0] and ur0p, cf. [Ma, Ch. III, § 4, 1] and [dJO, Lem. 5.4].
From [dJO, 5.8] we get that there exist inclusions (M0;f0),!(M;f),!
,!(u0(c 1)(d 1)M0;f0) between Dieudonne modules over k. Let c0:H!H0 be the isogeny defined by the first inclusion (M0;f0),! ,!(M;f). Its kernel is annihilated by u(c0 1)(d 1) p(c 1)(dr 1) and therefore we haveqHl(c 1)(d 1)
r
mj 1. ThusbHj, cf. Lemma 2.9.
REMARK2.11. The functionf :(0;1)(0;1)!(0;1) defined by the rulef(x;y) xy
xyis subadditive i.e., for allx1;x2;y1;y22(0;1), we have an inequality f(x1;y1)f(x2;y2)f(x1x2;y1y2). But the function g(x;y): df(x;y)eis not subadditive. Due to this and the plus 1 part of Lemma 2.9, our attempts to use Lemma 2.9 in order to prove Theorem 1.2 by induction onm2N, failed. On the other hand, in most cases Lemma 2.9 provides better upper bounds ofbHthan those provided by the following Proposition.
PROPOSITION2.12. For everyp-divisible groupH overk, the isogeny cutoffbHis bounded from above by the weak isogeny cutoffb~Hi.e., we have bH b~H.
PROOF. Letk((x)) be an algebraic closure ofk((x)). LetHbe ap-divisible group overk[[x]] of constant Newton polygon such that its fibre overkisH and the isogeny cutoffbofHk((x))is atmost~bH, cf. property (ii) of Definition 2.6. We recall that the statement thatHhas constant Newton polygon means that the Newton polygons ofHandHk((x))coincide i.e.,NH NHk((x)). For the proof of this Proposition it is irrelevant what thea-number ofHk((x))is.
LetH0be ap-divisible group overksuch thatH0[p~bH]H[p~bH]. Based on [Il, Thm. 4.4. f)] we getthatfor alln2N, there exists ap-divisible group H0n over Spec(k[[x]]=(xn)) that lifts H0 and such that H0n[p~bH]
H[p~bH]k[[x]]k[[x]]=(xn). Based on loc. cit., we can assume that H0n1 restricted to Spec(k[[x]]=(xn)) isH0n. Therefore theH0n's glue together to define ap-divisible groupH0fover the formal scheme Spf(k[[x]]). We recall that the categories ofp-divisible groups over Spf(k[[x]]) and Spec(k[[x]]) are naturally equivalent, cf. [dJ, Lem. 2.4.4]. Thus letH0be thep-divisible group over Spec(k[[x]]) that corresponds naturally toH0f.
Thep-divisible groupH0liftsH0and we have
H0[p~bH]proj:lim:n!1H0n[p~bH] H[p~bH]:
AsH0[p~bH] H[p~bH] andb~bH, from the very definition ofb, we getthat
H0k((x))has the same Newton polygon asHk((x)); thusNH0
k((x)) NH. As t he Newton polygons go up under specialization (see [De, Ch. IV, §7, Thm.]), we conclude that the Newton polygon NH0 of H0 is above the Newton polygon NH of H. But NH0 is notstrictly above NH, cf. property (i) of Definition 2.6. From the last two sentences we get that NH0 NH. This
implies thatbH~bH. p
From Lemma 2.10 and Proposition 2.12 we get:
COROLLARY2.13. For everyp-divisible group H overk, we have the following inequalitybHminfb~H; qH1g.
3. The proof of Theorem 1.2.
In this Section we prove Theorem 1.2. We begin by proving the fol- lowing particular case of Conjecture 1.1 which in essence is due to Traverso (cf. [Tr3, Thm. of § 21, and § 40]).
THEOREM3.1. Suppose that aH1. Then we have bHj:
PROOF. We first recall how to compute the Newton polygonNHofH(cf.
[Ma], [De, Ch. IV, Lem. 2, pp. 82-84], [Tr3, §21, Thm.], and [Oo1, Lem. 2.6]).
Let vp:W(k)n f0g !N[ f0;1g be the normalized valuation of W(k);
thusvp(p)1. Letx2Mbe such that its reduction modulopgenerates the k-vector space M=f(M)w(M). Let a0;a1;. . .;ac;b1;. . .;bd2W(k) be such that the map
c:Xc
i0
ac ifiXd
`1
b`w`:M!M
annihilates x. It is easy to see that we can choose x such that vp(a0)vp(bd)0 andfx;f(x);. . .;fc 1(x);w(x);. . .;wd(x)gis aW(k)-basis for M. For `2 f1;. . .;dg let ac`:p`b`; thus ai is well defined for all i2 f0;. . .;rg. Let
Qx(t):tr X
i2f1;...;rg ai60
vp(ai)tr itr X
i2f1;...;rg ai60
vp(sd(ai))tr i2Z[t]:
We recall that the Newton polygon ofQx(t) (more precisely, of the (r1)- tuple (a0;. . .;ar)) is the greatest continuous, piecewise linear, upward convex function Nx:[0;r]![0;d] with the property that for alli2 f0;. . .;rgwe
haveNx(i)vp(ai). Then we have
NH Nx: (1)
To check Formula (1) we viewMas aW(k)[F]-module, whereFllsF and whereFacts onMasfdoes. TheW(k)[F]-moduleMis isogenous to the W(k)[F]-moduleM0:W(k)[F]=W(k)[F]c0, where
c0Fdc:Xr
i0
sd(ai)Fr i:
But the Newton polygon of theW(k)[F]-moduleM0isNx, cf. [De, Ch. IV, Lem. 2, pp. 82-84]. Thus Formula (1) holds.
We recall thatjlcd r
m. Asc;d>0, we havej1. Letg2GLM(W(k)) be congruentto 1Mmodulopj. LetHgbe ap-divisible group overkwhose Dieudonne module is isomorphic to (M;gf). Asj1,Hg[p] is isomorphic to H[p] and therefore aHg aH1. Based on Lemma 2.8, to prove the Theorem it suffices to show that the Newton polygons NHg and NH
coincide. By replacing (f;w) with (gf;wg 1), the map c:M!M which annihilatesxgets replaced by another map
cg:Xc
i0
ag;c ifiXd
`1
bg;`w`:M!M
which annihilatesx, whereag;c iis congruenttoac imodulopjand wherebg;`
is congruentto b` modulo pj. For `2 f1;. . .;dg let ag;c`:p`bg;`; t hus ag;i2W(k) is well defined for all i2 f0;. . .;rg and itis congruentto ai
modulo pjminf0;i cg. The pair (Qx(t);Nx) gets replaced by the pair (Qg;x(t);Ng;x), whereQg;x(t):tr P
i2f1;...;rg ai60
vp(ag;i)tr iand whereNg;xis the Newton polygon of Qg;x(t). As Nx is upward convex and as Nx(0)0 and Nx(r)d, we have Nx(t)dt
r for all t2[0;r]. Thus jlcd r
m
dNx(c)e Nx(c) Nx(i) for all i2 f0;. . .;cg. Asr>d andjcd r, for i2 fc1;. . .;rg we have ji cdi
r. From the last two sentences we getthatjminf0;i cg di
r Nx(i) for all i2 f0;. . .;rg. From these inequalities and the fact that for all i2 f0;. . .;rg the elementsag;iand aiare congruentmodulopjminf0;i cg, we easily getthatNg;x Nx. From this identity and Formula (1) we get thatNHg NH. p
3.1 ±End of the proof of Theorem1.2.
Based on Proposition 2.12, to prove the inequalitybHjitsuffices to show that~bHj. We recall thatNc;dis the set of Newton polygons ofp- divisible groups overkof codimensioncand dimensiond. LetDH be the class ofp-divisible groups of codimensioncand dimensiondoverkwhose Newton polygons are strictly aboveNH.
We prove the inequality~bHjby decreasing induction onNH2 Nc;d. Thus we can assume that for every 2 DHwe have~b j; asb ~b (cf.
Proposition 2.12) we also have bj. To show that~bHj, letHbe ap- divisible group over k[[x]] such that the properties (ii.a) to (ii.c) of Defi- nition 2.6 hold. Letbbe the isogeny cutoff of Hk((x)). From the very defi- nition of~bHwe getthat
~bHmaxfb;bj 2 DHg:
Asbj(cf. Theorem 3.1 applied toHk((x))) and asbjfor all 2 DH(cf.
the inductive assumption), we have~bHj. This ends the induction.
Thus~bHjand thereforebHj. AsHis an arbitraryp-divisible group overkof codimensioncand dimensiond, we getthatbc;dj. Ifj1, then we obviously have bc;d1. If j2, then the below Example shows that there existp-divisible groupsHoverkof codimensioncand dimensiond and such that we havebHj. This implies that forj2, we havebc;dj.
We conclude that for all values ofj2Nwe havebc;djlcd r
m. p
EXAMPLE3.2. Suppose thatj2. Lets2Nbe the smallest number such thatrdividescd s; we have j 1cd s
r . We assume that there exists an element x2M such that the r-tuple (e1;. . .;er):
:(x;f(x);. . .;fc 1(x);wd(x);. . .;w(x)) is an ordered W(k)-basis for M and we have an equality wd(x)fc(x). Thus f(ei)ei1 if i2 f1;. . .;cg andf(ei)pei1ifi2 fc1;. . .;rg. We havefr(ei)pdeiand therefore all Newton polygon slopes ofHared
r; thusmg:c:d:(c;r). Replacing the equality wd(x)fc(x) by the equality w~d(x)~fc(x) pj 1x, we geta p- divisible group H~ over k whose Dieudonne module (M;~f) is such that f(e~ i)ei1 ifi2 f1;. . .;c 1g,~f(ec)pj 1e1ec1, and f(e~ i)pei1 if i2 fc1;. . .;rg. As (~f;w) and (f;~ w) are congruentmodulopj 1,H[p~ j 1] is isomorphic to H[pj 1]. We have NH(t)dt
r butthe piecewise linear functionNH~(t) changes slope atc; more precisely we haveNH~(c)j 1
(cf. Formula (1) applied to H, t o~ x2M, to the function
~
c: pj 1f~0~fc w~d:M!M, and to the polynomial Q~x(t):tr
(j 1)tdd). Thus the Newton polygon slopes of H~ are j 1 c (with multiplicity c) and 1 j 1
d (with multiplicity d) and therefore they are differentfromd
r. This implies thatbHjand therefore (cf. the inequality bHjproved in the previous paragraph) we havebHj.
EXAMPLE 3.3. Suppose that cd and that His as in Example 3.2.
Then qHc 1
2 (cf. [NV, Rmk. 3.3]) and jlc 2
m. If c is odd, then
bHjc1
2 qH1; moreover, for H~ :(Qp=Zp)Hmp1 we also havebH~ jc1
2 qH~ 1. Thus forcd, the inequalitybHqH1 (see Lemma 2.9) is optimal in general.
Acknowledgments. The first author has been supported by the Japa- nese Society for the Promotion of Science (JSPS PDF) while working on this paper at the University of Tokyo. The second author would like to thank University of Arizona for good conditions in which to write this note.
REFERENCES
[dJ] J.DEJONG,Crystalline Dieudonne module theory via formal and rigid geometry, Inst. Hautes EÂtudes Sci. Publ. Math., Vol.82(1995), pp. 5-96.
[dJO] J.DEJONG- F. OORT,Purity of the stratification by Newton polygons, J. of Amer. Math. Soc.,13, no. 1 (2000), pp. 209-241.
[De] M. DEMAZURE,Lectures on p-divisible groups, Lecture Notes in Math., Vol.302, Springer-Verlag, 1972.
[Di] J. DIEUDONNEÂ,Groupes de Lie et hyperalgeÁbres de Lie sur un corps de caracteÂrisque p>0(VII), Math. Annalen,134(1957), pp. 114-133.
[Il] L. ILLUSIE, DeÂformations des groupes de Barsotti-Tate (d'apreÁs A.
Grothendieck), Seminar on arithmetic bundles:the Mordell conjecture (Paris, 1983/84), pp. 151-198, J. AsteÂrisque,127, Soc. Math. de France, Paris, 1985.
[Ma] Y. I. MANIN, The theory of formal commutative groups in finite char- acteristic, Russian Math. Surv.,18, no. 6 (1963), pp. 1-83.
[NV] M.-H. NICOLE - A. VASIU, Minimal truncations of supersingular p- divisible groups, manuscript, June 2006 (see math.NT/0606777).
[Oo1] F. OORT,Newton polygons and formal groups:conjectures by Manin and Grothendieck, Ann. of Math. (2)152, no. 1 (2000), pp. 183-206.
[Oo2] F. OORT,Newton polygon strata in the moduli of abelian varieties, Moduli of abelian varieties(Texel Island, 1999), pp. 417-440, Progr. Math.,195, BirkhaÈuser, Basel, 2001.
[Oo3] F. OORT, Foliations in moduli spaces of abelian varieties, J. of Amer.
Math. Soc.,17, no. 2 (2004), pp. 267-296.
[Oo4] F. OORT,Minimal p-divisible groups, Ann. of Math. (2)161, no. 2 (2005), pp. 1021-1036.
[Tr1] C. TRAVERSO,Sulla classificazione dei gruppi analitici di caratteristica positiva, Ann. Scuola Norm. Sup. Pisa,23, no. 3 (1969), pp. 481-507.
[Tr2] C. TRAVERSO, p-divisible groups over fields, Symposia Mathematica XI (Convegno di Algebra Commutativa, INDAM, Rome, 1971), pp. 45-65, Academic Press, London, 1973.
[Tr3] C. TRAVERSO,Specializations of Barsotti-Tate groups, Symposia Mathe- matica XXIV (Sympos., INDAM, Rome, 1979), pp. 1-21, Acad. Press, London-New York, 1981.
[Va1] A. VASIU,Crystalline Boundedness Principle, Ann. Sci. EÂcole Norm. Sup., 39, no. 2 (2006), pp. 245-300.
[Va2] A. VASIU, Reconstructing p-divisible groups from their truncations of small level, manuscript, April 2006 (see math.NT/0607268).
Manoscritto pervenuto in redazione il 20 maggio 2006.