by
Christophe Ritzenthaler
Abstrat. In this survey, we disuss the problem of the maximum numberof points of
urvesofgenus
1, 2
and3
overniteelds.Résumé (Courbes optimales de genre
1, 2
et3
). Nous examinons la question dunombremaximumdepointspourlesourbesdegenre
1, 2
et3
surlesorpsnis.1. Introdution
The foundations of the theory of equations over nite elds were laid, among others, by
mathematiianslike Fermat,Euler, GaussandJaobi(see [Di66℄). Subsequently,therewas
little ativityinthe eld at leastuntil theend of the19thentury andthe studyof thezeta
funtion of a urve. Initiated by Dedekind, Weber, Artin and Shmidt, this work led to an
analogue ofthe Riemann hypothesiswhihwasprovedbyHasseintheaseofellipti urves
andthen byWeil ingeneral in1948 (see[Wei48 ℄). Thethird,modern, periodstarts in1980
with the work of Goppa [Gop77, Gop88℄. His onstrution of error-orreting odes with
good parameters fromurvesovernite eldsrenewed theinterest inthis theory.
Withthis appliationinmind,thetheoryhasfousedonthemaximumnumberofpointsofa
(projetive,geometriallyirreduible,nonsingular)urveofgenus
g
overaniteeldk = F q,
denoted
N q (g)
. Asymptoti results, i.e. values ofN q (g)/g
wheng
goes to innity andq
isxed,drewattentionrst,butSerre,inhisleturesatHarvard[Ser85℄,gaveequaltreatment
to the `dual'ase, i.e. values of
N q (g)
wheng
is xedandq
varies. It quiklyappearedthatdetermining
N q (g)
was a hard problem and as soon asg ≥ 3
, only sparse values are known(see forinstane the web page www.manypoints.orgfor thebestestimates when
q
issmall).Inthissurvey,we aregoingtodesribe themainideas thathave been developedtodealwith
2000 MathematisSubjetClassiation. Primary11G20,11G10Seondary14K25,14H45.
Key wordsand phrases. Optimalurve,isogenylass,indeomposablepolarization,hermitianmodule,
Serre'sobstrution,planequarti, Siegelmodularform,Hasse-Weil-Serrebound.
TheauthoraknowledgespartiallysupportedbygrantMTM2006-11391fromtheSpanishMECandbygrant
ANR-09-BLAN-0020-01fromtheFrenhANR..
theases
1 ≤ g ≤ 3
. Itisinterestingtonotethatforeahvalueofg
,wewillbeonfrontednotonlytoharderomputationsbutalsotoaompletelynewkindofissue. Inordertoemphasize
thisprogression,wewillnotonsidertheatualvalueof
N q (g)
butthefollowingsub-problem.As we shall reall in Setion 2.1 ,
N q (g) ≤ 1 + q + g ⌊ 2 √ q ⌋
and we an wonder whenN q (g)
reahes this bound. If it does, a urve with this number of points is alled optimal and we
aregoingto askfor whih valuesof
q
suh urvesexist.When
g ≤ 3
,thelassial gameto prove or disprove the existeneof optimalurvesis1. to prove theexistene(ornot)ofan abelian variety
A/k
witha`good'Weil polynomial(Setion 2). Thisisgoingto ontrol thenumberofpointson apossibleurve
C/k
suhthat
Jac C ≃ A
.2. to put a good polarization
a
on A suh that(A, a)/k
is geometrially (i.e. over¯ k
) theJaobian ofa urve
C ¯
withits anonial polarization (Setion3).3. to see if
C ¯
admitsa modelC/k
suh that(Jac C, j) ≃ (A, a)
(wherej
is the anonialpolarizationof
C
). WewillseethatifC ¯
isnonhyperellipti,thereanbeanobstrution to this desent (see Setion 4) and forg = 3
,we will propose solutions to address theomputation ofthe obstrution (see Setion 5).
Mostideas we aregoing to present here are alreadyontained in[Ser85℄but our proofsfor
g = 1
and2
aresometimes dierent fromthe original onesandtake advantageof subsequentsimpliationsof thetheory.
Conventions and notation. In the following
g ≥ 1
is an integer andq = p n with p
a
primeand
n > 0
an integer. The letterk
denotes the niteeldF q and K
any perfet eld.
When we speak about a genus
g
urve we mean that the urve is projetive, geometrially irreduible and non-singular. IfA
andB
are varieties over a eldK
, when we speak of amorphismfrom
A
toB
wealways meanamorphism dened overK
. So,forinstaneEnd(A)
isthering ofendomorphisms denedover
K
,A ∼ B
meansA
isogenous toB
overK
,et. If(A, a)
and(B, b)
arepolarizedabelian varieties,byanisomorphismbetween them,wealways mean`aspolarizedabelian varieties'.Aknowledgements. I would like to thank Christian Maire for suggesting me to write
this survey. This is part of my `habilitation' thesis [Rit09℄ whih was defended during the
workshopTheory of Numbers and Appliations whih was organizedbyKarim Belabas and
Christian Maire in Luminy in Deember 2009. I am really grateful to Detlev Homann for
thereferenes of Remark3.5and to the Number TheoryList ommunityand partiularly to
SamirSiksek for helping mewithRemark 3.10 .
2. Control of the isogeny lass
2.1. Bounds. Let
C/k
beagenusg
urve. We reallthatitsWeilpolynomialχ C isthe
Weilpolynomialof
Jac C/k
,i.e. theharateristipolynomialoftheationofthek
-Frobeniusendomorphismon an
ℓ
-adiTatemoduleforanyprimeℓ 6 = p
. It iswellknownthatitanbewritten
χ C = Y g i=1
(X 2 + x i X + q) ∈ Z [X]
with
x i ∈ R
and| x i | ≤ 2 √ q
. Sine#C(k) = q + 1 + X g
i=1
x i ,
it islear that
#C(k) ≤ 1 + q + ⌊ 2g √ q ⌋
,whih is known asHasse-Weil bound [Wei48 ℄ andso
N q (g)
is less than this bound too. It is possible to improve this bound as the followinglemmashows.
Lemma 2.1 (Hasse-Weil-Serre bound [Ser83b℄). Let
m = ⌊ 2 √ q ⌋
. ThenN q (g) ≤ 1 + q + gm.
Proof. It isenough to usethearithmeti-geometri meaninequality:
1 g
X g i=1
(m + 1 − x i ) ≥ Y g i=1
(m + 1 − x i )
! 1/g
≥ 1,
thelast inequalityoming from the fatthattheprodutis anon-zero integer.
Thismotivates us to give the following denition.
Denition 2.2. We saythata genus
g
urveC/ F q isoptimal if
#C( F q ) = q + 1 + gm.
Inthat ase
N q (g) = q + 1 + gm
.Note that the previous denition is not universally aepted. Some authors all maximal
(or
F q-maximal) what we all optimal by referene to the historial ases with n
even and
N q (g) = q + 1 + gm
. Wepreferto keepthewordmaximal forurveswhihnumbersof points
isequal to
N q (g)
andour terminology is oherent withthehistorial one aswell.Remark 2.3. If
g ≥ (q − √ q)/2
, the bound an be improved, thanks to the expliitmethods of Oesterlé (and is known as Oesterlé bound [Ser83b℄). It uses the fat that the
number of plaes of eah degree on the urve is non negative. As we will mainly deal with
small values of
g
ompared toq
,theHasse-Weil-Serrebound will be our referene.2.2. Existene of the isogeny lass. Equality in the arithmeti-geometri mean in-
equality is equivalent to the fat that all terms in the sum are equal and so
x i = m
for all1 ≤ i ≤ g
. Hene, ifanoptimal urveC
exists, itsWeilpolynomialhasthepartiular simpleexpression
χ C = (X 2 + mX + q) g .
Honda-Tatetheory asexplained in[Tat66℄, [Hon68℄,[Wat69℄, [MW71 ℄ or [Tat71 ℄shows
that if
p ∤ m
(resp.n
is even) thenJac C
is isogenous toE g where E
is an ordinary (resp.
supersingular)elliptiurvewithtrae
− m
. However,ifp | m
andn
isodd,this might not betrue(seetheproofofProposition2.5below)andthereisfor instaneasimpleabelianvariety
ofdimension
9
overF 5 9 withsuhWeil polynomial. Ifwe restrit to g ≤ 3
,it an be proved
thatthisneverhappens (seefor instane the proofof Corollary4.2 of[NR10℄).
Lemma 2.4. If
C/ F q is an optimal urve of genus g ≤ 3
then Jac C
is isogenous to E g
where
E
is an ellipti urve of trae− m
.Therstneessaryondition isthento seewhether suh anellipti urve existsor not.
Proposition 2.5 (Deuring [Deu41℄). There does not exist an ellipti urve with trae
− m
if and onlyifn ≥ 3
,n
isoddandp | m
.Proof. Let
F = X 2 + mX + q
. Sinem < 2 √ q
if and only ifq
is not a square,F
isirreduibleover
Q
whenn
isoddandF = (X + √ q) 2 whenn
iseven.
If
n
is odd, by [Wat69 , p.527℄, the minimale
for whihχ = F e is the Weil polynomial of
an abelian variety of dimension
e
overk
is the least ommon denominator ofv p (F ν (0))/n
where
F ν denotes thefatorsof F
inQ p [t]
andv p thep
-adi valuationof Q p. Hene F
isthe
p
-adi valuationofQ p. Hene F
isthe
Weilpolynomialofanelliptiurve ifand onlyif
n | v p (F ν (0))
forallfators. Thisisofoursesatisedif
n = 1
. LookingattheNewtonpolygonofF
,weseethatifp ∤ m
thenv p (F ν (0)) = n
or
0
, soe = 1
. Withthe same tehnique, ifn > 1
odd andp | m
, thenv p (F ν (0)) < n
and soe > 1
.If
n
is even,we apply the previous arguments toF = X + √ q
. Sinev p ( √ q)/n = 1/2
,e = 2
so
F 2 = X 2 + mX + q
isthe Weil polynomialof anellipti urve.Atually,theonly valuesof
q = p
for whihp | m
areq = 2
orq = 3
.Remark 2.6. For any value of
− m ≤ t ≤ m
, one knows if an ellipti urve with traet
exists (see [Wat69, Th.4.1℄). Also, thepossible Weil polynomials of the isogeny lasses of abelian surfaes(resp. threefolds) an be found in [MN02 , Lem.2.1,Th.2.9℄ (resp. [Xin96,Hal10℄).
3. Existene of an indeomposable prinipal polarization
The Jaobian of a genus
g
urveC/K
is naturally equipped witha prinipal polarizationj
induedbytheintersetionpairingontheurve
C
. SinethethetadivisorSym g−1 C ֒ → Jac C
assoiated to
j
is geometrially irreduible,(Jac C, j)
is geometrially indeomposable, i.e.theredoesnot existsanabelian subvariety
B ⊂ Jac C
dened overK ¯
suh thatj
indues onB
a prinipalpolarization. Conversely,starting withA = E g where E
is an elliptiurve,it
islearthat
A
alwaysadmitsaprinipalpolarizationa 0 given bytheprodutoftheprinipal
polarizations on eah fator. As
a 0 is deomposable, (A, a 0 )
is not (even geometrially) a
Jaobian. Hene `good' prinipal polarizations on A
(or on abelian varieties in the isogeny
lassof
A
) have to be moresubtle. Lukily, equivalenesofategory have been developed to translatetheexisteneofanindeomposablepolarizationintotheexisteneofpurelyalgebraiobjets. As far asI know several points of view o-existand it is not lear to seehowto go
fromone to the other. IshalluseSerre's one andmentionothers inremark.
Remark 3.1. Howe [How95℄,[How96 ℄hasdevelopedapowerfulmahineryto provethe
existene of a prinipally polarized abelian variety inthe isogenylass of an abelian variety
A/k
. ButonlywhenA
issimple,itiseasyto seethatthepolarizationisindeomposable (see [Ryb08℄fortheaseE × B
whereE
isanelliptiurveandB
ageometriallysimpleabelian surfae).3.1. The equivalenes. Let us start with
E
ordinary. LetE/k
be anordinary elliptiurve with trae
t
. Ifπ
denotes theF q-Frobenius endomorphism of the urve E
, then the
ring
R := Z [X]/(X 2 − tX + q)
is isomorphi toZ [π] ⊂ End(E)
. Serre [Ser85, Se.50-53℄,[Lau02 , Appendix℄ denes an equivalene of ategory
T
between the ategory of abelianvarietieswhih are isogenous to a power of
E
andR
-modules of nite type without torsion.Thefuntor
T
mapsan objetA
to theR
-moduleL = Hom(E, A)
. Obviously,therank ofL
isequalto thedimensionof
A
. Thisfuntoralsobehavesnielywithrespettoduality: ifwedenote
L ˆ
thering ofanti-linear homomorphismf : L → R
(i.e.f(rx) = ¯ rf (x)
for allr ∈ R
and
x ∈ L
)thenT ( ˆ A) = ˆ L
. Thus amorphisma : A → A ˆ
denesa morphismh : L → L ˆ
andheneanhermitian form
H : L × L → R
. Serreprovesthata
isapolarizationifandonlyifH
is positive denite, that
a
is prinipal if(L, H )
is unimodular (i.e.h(L) = ˆ L
) and moreover(geometrially)indeomposableifandonlyif
(L, H)
isindeomposable,i.e. annotbewritten asasum of orthogonalsub-modules. Theouple(L, H )
isalleda hermitian module.Remark 3.2. This equivalene is inspired by the lassial theory over
C
, whih is notsurprisingsineordinaryabelianvarietiesanbeliftedanoniallyandthisisusedin[Del69℄.
When
A = E g and End(E) ≃ R
,a more expliit point of viewan be onsidered looking at
thehermitian matrix
M := a − 0 1 a ∈ End(A) = M g (End(E)) ≃ M g (R)
(see [Rit10℄,[Lan06 ℄).For
g = 2
, Kani's onstrution [Kan97℄ also gives neessary and suient onditions for`gluing' twoelliptiurvesalongtheir
n
-torsionforn > 1
. BothpointsofviewarerelatedbytheCholewskydeomposition of
M
.A lassiation of rank
2
and3
hermitian modules wasahieved in [Hof91, Th.8.1,8.2℄ (see also [Sh98℄for further omputations) andtranslates into the following result.Proposition 3.3. Let
E
be an ordinary ellipti urve with traet
. There is no abeliansurfae (resp. threefold)withageometrially indeomposableprinipalpolarizationinthelass
of
E 2 (resp. E 3) ifand onlyif t 2 − 4q ∈ {− 3, − 4, − 7 }
(resp. t 2 − 4q ∈ {− 3, − 4, − 8, − 11 }
).
t 2 − 4q ∈ {− 3, − 4, − 7 }
(resp.t 2 − 4q ∈ {− 3, − 4, − 8, − 11 }
).Remark 3.4. For
g = 2
,theresultanbetraedbakto[HN65,p.14℄,wheretheauthorsprove the existene of genus
2
urves whih Jaobian is isomorphi toE 2 by onstruting
free indeomposable hermitian modules (in [Hay68 ℄, the preise number of isomorphism
lasses of suh urves is omputed). For
g = 2
or3
, it ould also be dedued from themassformulae(i.e. numberof weightedlasses bytheorder oftheirautomorphismgroup)of
[HK86,HK89℄ (although, aordingto Homann (lo. it. p.400) thereis aminor mistake
inthese omputations).
Remark 3.5. For
g > 3
, there have been several partial answers on the existene ofindeomposable unimodular positive denite hermitian modules of rank
g
over the ring ofintegers of an imaginary quadrati eld
Q ( √
− d)
. It seems that in[Zhu97℄ and [WL01 ℄ aomplete answer is given: there always exists one, exept when
d = 1
andg = 5
ord = 3
and
g = 4, 5, 7
. Oneshould be areful sine, aordingto the Mathsinet reviewof [Zhu97℄by Homann, the proofs ontain several mistakes. Also, I do not know if the ase of non
maximalorders hasbeen onsidered.
Assumenowthat
E
issupersingular. Morepreisely,letE/ F p beanelliptiurvewithtrae
0
,sothatE
issupersingular,all the geometri automorphisms ofE
aredened overF p 2 and
Tr(E/ F p 2 ) = − 2p = − m
. OnesaysthatanabelianvarietyA
(resp. aurveC
)issuperspeialif
A
(resp.Jac C
)isgeometriallyisomorphito aprodutofsupersingularelliptiurves. A resultofDeligne (see[Shi79,Th.3.5℄)shows thatwheng > 1
,asuperspeialabelian variety ofdimensiong
isgeometriallyisomorphitoE g (whereasforg = 1
therearenon-isomorphi
supersingularelliptiurvesassoonasp > 7
). However,thedesriptionoftheisogenylassis
mademoreompliated thaninthe ordinary asebytheexisteneof `ontinuous' familiesof
isogenies. For instane, already when
g = 2
(see[Oor75℄), a supersingular abelian surfae is eithergeometriallyisomorphitoE 2 (andsosuperspeial)oroftheformE 2 /α p whereα p is
α p is
theunique loal-loal groupshemeover
F p,theinjetionofα p inE 2 beingparametrizedby
E 2 beingparametrizedby
P 1 (¯ F p ) \ P 1 ( F p 2 )
. Inthe latter,itanbeshownthatA
isnotsuperspeialandthedesription ofthepolarizations on this objetis moreevolved. For this reason, we will onentrate onlyonexisteneresults and limit ourselvesto thesuperspeialase.
Remark 3.6. Note that it is still possible to obtain a omplete desription for
g = 2
inthe non-superspeial ase like in [IKO86℄ or [HNR09, Part.2℄ where the mass formula of
[Ibu89℄wereused.
As in Remark 3.2, we desribe the polarizations on
A = E g by matries M := a − 0 1 a
in
End(A) = M g (End(E))
. Now, End(E)
is a quaternion algebra, so we need results on the
number
n g of positive denite quaternion hermitian forms. Then, to obtain the number of
(geometrially)indeomposablepolarizationson
E g,theideaistosubtratton g thenumber
ofpolarizationsomingfromombinationsoflowerdimensionalabelianvarieties. Inthisway,
one gets
Proposition 3.7 ([Eke87 ,Prop.7.5℄). There is no geometrially indeomposable prini-
pal polarization on
E g ifand onlyif g = 2
and p = 2
or 3
, or g = 3
andp = 2
.
Remark 3.8. Morepreisely,Ekedahlgivesin[Eke87 ,Prop.7.2℄themassofindeompos-
able prinipal polarizations on
E g. However, Brok[Bro93,Th3.10.℄ orrets a mistake in
thease
g = 3
. He alsoompletes and reoversseveral resultsobtained in[HI83,I℄, [KO87℄for
g = 2
and in [Has83 ℄, [Oor91℄ forg = 3
. For instane, in [Bro93, Th.3.14,Th.3.15℄, he gives the number of genus2
and genus3
superspeial urves for eah possible group of automorphisms.3.2. Appliation. We annowanswerthequestion oftheexisteneofa goodpolariza-
tionwhen
g ≤ 3
.Theorem 3.9. Let
E
be an ellipti urve with trae− m
. There is no abelian surfae(resp. threefold) with a geometrially indeomposable prinipal polarization in the isogeny
lass of
E 2 (resp. E 3) if and only if q = 4
or 9
or m 2 − 4q ∈ {− 3, − 4, − 7 }
(resp. q = 4
or
16
or m 2 − 4q ∈ {− 3, − 4, − 8, − 11 }
).
q = 4
or9
orm 2 − 4q ∈ {− 3, − 4, − 7 }
(resp.q = 4
or16
orm 2 − 4q ∈ {− 3, − 4, − 8, − 11 }
).Proof. When
p ∤ m
,E
is ordinarysowe an useProposition3.3 .When
n
isoddandp | m
,E
existsifandonly ifq = 2
or3
,whih leavesthese twoases tobetreated apart(for instane by extensive omputer researh of urvesusing Theorem 4 or by
Remark3.10 ).
When
n
iseven thenp | m
. We distinguish several ases.When
p > 3
andg = 2
(resp.p > 2
and g= 3
), Proposition 3.7 shows that there is alwaysanindeomposableprinipalpolarizationonE 2 (resp. E 3). Notethanwhen4 | n
,
4 | n
,thepresent
E
isthe quadratitwist oftheellipti urve inProposition3.7.When
p = 2
andg = 2
,expliitonstrutionsasin[MN07 ℄ora`gluing'argumentasin [Ser85,Se.32℄,[Sha01,Prop.30℄showsthatonean getaurveC/ F 2 n suhthatJac C
isisogenous to
E 2 assoonasn > 2
.
When
p = 2
,g = 3
andn > 4
, An expliit non hyperellipti urveC/ F 2 n suh that
Jac C ∼ E 3 an be onstruted. (see [Rit09, Lem.2.3.8℄). Note that in [NR08℄, the
moregeneralquestionoftheexisteneofaJaobianintheisogenylassofasupersingular
abelianthreefold inharateristi
2
is addressed.Finally when
p = 3
andg = 2
, one an nd an expliit onstrution in [Kuh88℄ (seealso[Sha01,Cor.37℄ whereamistakeisorreted)assoonas
n > 2
. Thisworkwasalsogeneralized to all supersingular abelian surfaesinharateristi
3
in[How08℄.Remark 3.10. Theases
m 2 − 4q ∈ {− 3, − 4 }
analsobeexludedthankstoaproofdueto Beauville [Sha01,Th.16℄, [Ser85,Se.13℄ without any hypothesis onthe
p
-rankofE
(andthen
q = 2, 3
areovered).Asonjeturally, there is innitely many
p
in theformsp = x 2 + 1
andp = x 2 + x + 1
theequations
m 2 − 4p n ∈ {− 3, − 4 }
have innitely many solutions withn = 1
. Forn > 1
odd,one knows that theset of solutions is nite. For instane, in[Ser83a ℄, we nd that there is
onlyone solution tothe equation
q = x 2 + x + 1
namelyq = 7 3 and noneto q = x 2 + 1
.
Similarly,theaseofdisriminant
− 7
orrespondstotheequationq = x 2 + x + 2
withuniquesolutions
q ∈ { 2 3 , 2 5 , 2 13 }
whenn > 1
isodd.Thease ofdisriminant
− 8
orrespondstoq = x 2 + 2
,whih,whenn > 1
isodd,hasq = 3 3
forunique solution. Thisis proved usingthe same argumentsasthelast asebelow.
Finally, the ase of disriminant
− 11
orresponds toq = x 2 + x + 3
. Whenn > 1
is odd,q = 3 5 is the unique solution thanks to the following argument due to Samir Siksek. The
equationan be rewritten
(2x + 1) 2 + 11 = 4p nand fatorsinK = Q ( √
− 11)
as2x + 1 + √
− 11 2
2x + 1 − √
− 11 2
= p n .
Sine
n
is odd andO K is a prinipal domain, there exists α = (a + b √
− 11)/2 ∈ O K suh
that
α n = (2x + 1 + √
− 11)/2
andαβ = p
withβ = ¯ α
. Now,notethatα n − β n = √
− 11
andsine
(α n − β n )/(α − β) = 1/b ∈ O K,we seethatb = ± 1
. Hene, ifwe xn
,we an ndthe
niteset of integer solutions of this polynomial equationin
a
. However, to solve itfor alln
we have to invoke the muh deepertheorem from [BHV01℄ whih tells us that ifthere is a
solutionthen
n < 4
,n = 5
orn = 12
. Indeed, withthe terminology andnotation of lo. it., one seesthatu n = (α n − β n )/(α − β) = ± 1
is a Luasnumber withoutprimitive divisor, sotheLuas pair
(α, β)
isn
-defetive.4. Optimal urves
Aswehave seeninSetion 3,thestrategy we have applied sofar worksinanydimension. If
wenowhavetorestritourselvesto thedimensionslessthanorequalto
3
isbeause,intheseases,theondition`has anindeomposable prinipal polarization'is geometrially suient
to be theJaobian of a urve. Thisis not truewhen thedimension is bigger, asitis proved
simplybynoting that the dimension of the moduli spaeof urvesof genus
g
,3g − 3
,is lessthanthedimensionofthe modulispaeofprinipally polarizedabelianvarietiesofdimension
g
,g(g + 1)/2
. However,Proposition 4.1 ([OU73℄). For
g ≤ 3
,any geometriallyindeomposable prinipally po- larized abelian variety(A, a)/K
is theJaobian of a urveC ¯
overK ¯
.So,given
(A, a)/K
asin Proposition 4.1 , the question boilsdown to know whether one an desend the urveC ¯
to a urveC
overK
suh that(Jac C, j) ≃ (A, a)
. Surprisingly the answeris `notalways'.Theorem 4.2 (Arithmeti Torelli theorem). Thereisauniquemodel
C/K
ofC ¯
suhthat:
1. If
C ¯
ishyperellipti, there is an isomorphism(Jac C, j) −−−−→ ∼ (A, a).
2. If
C ¯
is not hyperellipti, there is a unique quadrati haraterε
ofGal( ¯ K/K )
, and anisomorphism
(Jac C, j) −−−−→ ∼ (A, a) ε
where
(A, a) ε is the quadrati twist of A
by ε
.
Remark 4.3. Itistrikyto ndtheright origin ofthepreviousresult. In[Ser85,Se.69℄,
Gouvéa indiates `Oort +...' as a referene. One an indeed nd in [Oor91, Lem.5.7℄ a
similarresult (butthisis 1991). Sekiguhialsoworked on thisquestion but aftertwo errata,
he gives in[Sek86℄ only the existene of the model
C/K
but doesnot speak aboutε
. Onean alsond this resultin[Maz86 ,p.236℄.
The notation
(A, a) ε = (A ε , a ε )
should be understood asfollows. The varietyA ǫ isuniquely
denedupto isomorphismbythefollowingproperty: there existsaquadratiextension
L/K
and an isomorphism
φ : A → A ǫ dened over L
suh that for all σ ∈ Gal( ¯ K/K)
one has
φ σ = ε(σ)φ
. Thepolarizationa ε isthepull-bak ofa
byφ − 1.
a
byφ − 1.
This result is a onsequene of Weil's desent as explained in [Ser68, 4.20℄ and of Torelli
theorem [Mat58 , p.790-792℄. The shism whih appears between the hyperellipti and non
hyperellipti aseisdue to thefatthat
Aut(Jac C, j) ≃
( Aut(C)
ifC
ishyperellipti,Aut(C) × {± 1 }
ifC
isnon hyperellipti.Denition 4.4. The harater
ε
(or the disriminant of the extensionL/K
) is alledSerre's obstrution. Byextension, inthehyperellipti ase, we saythat
ε
istrivial.Letus emphasizewhythisobstrutionisanissueinourstrategy. Sofarwehavebeenableto
prove inertain ases the existene of a geometrially indeomposable prinipally polarized
abelian variety
(A, a)/k
withWeil polynomial(X 2 + mX + q) g. Thanks to Proposition 4.1,
we knowthatitis geometriallytheJaobian ofa urve C ¯
. Iftheobstrutionis trivial,then
C ¯
desendsto aurveC/k
suhthatJac C ≃ A
and soC
isoptimal. Ontheontrary, ifthe
obstrution is not trivial, then
C ¯
desendsto a urveC/k
suh thatJac C
is isomorphi tothe(unique)quadratitwistof
A
andsoitsWeilpolynomialis(X 2 − mX + q) g. Inpartiular
#C(k) = q + 1 − gm
andC
isnotoptimal(andatuallyC
hastheminimumnumberofpointsagenus
g
urve overk
an have).4.1. The end of the genus
1
and2
ases. Sine a genus1
urve over a nite eldalwayshasarationalpoint,itisanelliptiurveandProposition2.5tellsuswhenanoptimal
genus
1
urve exists(forthevalueofN q (1)
see[Deu41℄ or [Ser83a℄).When
g = 2
, all genus2
urvesare hyperellipti sotheobstrution isalways trivialand the resultis similarto Theorem 3.9,namelyTheorem 4.5 (Serre). There isno optimalurve of genus
2
overF q ifandonlyif q = 4
or
9
orm 2 − 4q ∈ {− 3, − 4, − 7 }
.In[Ser83a℄, a losed formulafor the value of
N q (2)
is given. Morereently,ompleting thework started by many authors, we obtained in [HNR09℄ the omplete piture for abelian
surfaes, i.e. we determined whih isogeny lasses ontain theJaobian of a genus
2
urvesinterms oftheoeients ofthe Weilpolynomial.
5. The genus
3
aseAs there exist non hyperellipti genus
3
urves (the non-singular plane quartis), Serre's obstrution may not be trivial. One hope is that, for eahq
, there would be an optimalhyperellipti urve but this possibilityhasto bedisarded: for instane, theredoesnot exist
anyoptimalhyperelliptigenus
3
urvesoverF 2 n withn
evensinesupersingularhyperellipti
urvesdonot existsinharateristi 2
[Oor91℄. Other ounterexamples an be foundinodd
harateristi aswell asit will be apparent in Proposition 5.6. Therefore,it is important to
beabletoomputeSerre'sobstrution. Currently,thereisnoperfetsolutiontothisproblem
butwewill summarizesome of the ideas andpartial answers whihhave beenobtained.
5.1. Speial families. The key-idea is to use some families of urves with non trivial
automorphisms,suhthattheir Jaobian isaprodutofelliptiurvesexpliitlyobtained as
quotientbyertainautomorphismsubgroups. Thenonetriestoreversetheproessandseeif
oneangluegivenelliptiurvestogethertogetaurveinthefamily. Thepossiblequadrati
extension one has to make during the onstrution is Serre's obstrution. Let us illustrate
thisproedure withanexample.
Example 5.1. The following family represents genus
3
non hyperellipti urves in har- ateristi2
withautomorphism groupontaining( Z /2 Z ) 2
C : (a(x 2 + y 2 ) + cz 2 + xy + ez(x + y)) 2 = xyz(x + y + z), ac(a + c + e) 6 = 0.
Theinvolutions are
(x : y : z)
maps to(y : x : z), (x + z : y + z : z)
or(y + z : x + z : z)
. Togettheequation ofthe urve
E 1 = C/ h (x : y : z) 7→ (y : x : z) i
, one introduesthe invariantfuntions
X = x + y, Y = xy
andndsE 1 : (aX 2 + c + Y + eX ) 2 = Y (X + 1).
Doing similarly with the other involutions and rewriting the equations of theellipti urves
(see[NR10℄ for details) one getsthat
Jac C ∼ E 1 × E 2 × E 3 where
E 1 : y 2 + xy = x 3 + ex 2 + a 2 (a + c + e) 2 ,
E 2 : y 2 + xy = x 3 + ex 2 + c 2 (a + c + e) 2 ,
E 3 : y 2 + xy = x 3 + ex 2 + c 2 a 2 .
Conversely,wenowwant toglueordinaryelliptiurves
E i withj
-invariant j i 6 = 0
. Theyan
always be written
E i : y 2 + xy = x 3 + ex 2 + 1/j i where,ifq > 2
,Tr F q / F 2 (e) = 0
ifandonly if
Tr(E i ) ≡ 1 (mod 4)
. Lets 4 i = 1/j i,then
a = s 1 s 3 s 2
, c = s 2 s 3
s 1 , e = s 1 s 3
s 2 + s 2 s 3
s 1 + s 1 s 2
s 3 .
(1)
Now, for instane, assume that
m ≡ − 1 (mod 8)
. This happens forn = 35, 37, 63, . . .
. Wehoose
E = E 1 = E 2 = E 3 an ordinary ellipti urve with trae − m
and j
-invariant j
(E
existssine
2 ∤ m
). Sine we an assume thatq > 4
,the urveE
hasan8
-torsion point anditis not diultto hekthatthis implies(atuallyis equivalent to)
Tr F q / F 2 1/j = 0
. HeneTr F q / F 2
s 1 s 3 s 2
+ s 2 s 3 s 1
+ s 1 s 2 s 3
= Tr F q / F 2 (1/j) = 0.
On theother hand,sine
Tr(E) ≡ 1 (mod 4)
, we haveTr F q / F 2 (e) = 0
aswell, sothere is noobstrution to (1) . Atually,we getan expliitequation
C : (j −1/4 (x 2 + y 2 + z 2 + xz + yz) + xy) 2 = xyz(x + y + z)
for theoptimalurve.
Exploitingother familiesof urvesinharateristi
2
,we getthefollowing result.Theorem 5.2 ([NR08 , NR10℄). If
n
is even, there existsan optimal urve overF 2 n if
andonly if
n ≥ 6
.If
n
isodd andm ≡ 1, 5, 7 (mod 8)
, there isan optimalurve overF 2 n.
When
n > 1
isoddandm
iseven,thereisofoursenooptimal urvesinethereisnoelliptiurve with trae
− m
. So only the asem ≡ 3 (mod 8)
ismissing to get a omplete answerwhen
p = 2
.Morereently,Mestre[Mes10 ℄hasworked withafamily ofurveswithautomorphismgroup
S 3 and showed that if p = 3
(resp. p = 7
), 3 ∤ m
(resp. 3 | m
) and − m
is a non-zero square
modulo
7
(resp.n ≥ 7
),thenthere existsanoptimal urve overF 3 n (resp. F 7 n).
To onlude on this approah, let us point out that one ould use the family with auto-
morphism group
( Z /2 Z ) 2 (alled Ciani quartis) also in harateristi greater than 2
, sine
Serre's obstrution hasbeen worked out in[HLP00℄. Unfortunately,one doesnot see when
this obstrution is trivial knowing only
m
(one needs theequations of theellipti fators todeide).
5.2. Serre's analyti strategy. InspiredbyresultsofKlein[Kle90 ,Eq.118,p.462℄ and
Igusa [Igu67, Lem.10,11℄, in a 2003 letter to Jaap Top [LR08℄, Serre stated a strategy to
ompute the obstrution when the harateristi is dierent from
2
. Roughly speaking, hisideawasthataertainSiegelmodularformevaluatedata`modulipoint'
(A, a)/K
isasquarein
K
ifand only if the obstrution is trivial. In a series of three papers, it was shown thatthisisaurate(rst forCianiquartis,theningeneral)andhowtoompute theobstrution
inthe ase of thepowerof a CMellipti urve. Let us state the general result without any
omments on the proof whih would lead us to far from our initial purpose (see however
[Rit09,Chap.4℄ fordetails).
Theorem 5.3 ([LRZ10 ℄). Let
A = (A, a)/K
be a prinipally polarized abelian threefolddenedoveraeld
K
withchar K 6 = 2
. Assumethata
isgeometriallyindeomposable. There existsa unique primitive geometriSiegel modular form of weight18
dened overZ
, denotedχ 18, suh that
i)
(A, a)
isa hyperellipti Jaobian if and onlyifχ 18 (A, a) = 0
.ii)
(A, a)
isa non hyperellipti Jaobian if andonlyifχ 18 (A, a)
isa non-zero square.Moreover, if
K ⊂ C
, let
(ω 1 , ω 2 , ω 3 )
be a basis of regular dierentials onA
;
γ 1 , . . . γ 6 be a sympleti basis (fora
)of H 1 (A, Z )
;
Ω a := [Ω 1 Ω 2 ] = [ R
γ j ω i ] be a period matrix withτ a := Ω − 2 1 Ω 1 ∈ H 3
a Riemann matrix.
Then
(A, a)
isa Jaobian if and onlyif(2)
χ 18 ((A, a), ω 1 ∧ ω 2 ∧ ω 3 ) := (2π) 54 2 28 ·
Q
[ε] θ[ε](τ a ) det(Ω 2 ) 18
isa square in
K
.Letus reallthatthe Thetanullwerte
θ[ε](τ )
arethe36
onstants suh that[ε] =
ǫ 1 ǫ 2
∈ { 0, 1 } 3 ⊕ { 0, 1 } 3 ,
with
ǫ 1 t ǫ 2 ≡ 0 (mod 2)
andforτ ∈ H 3
θ ǫ 1
ǫ 2
(τ ) = X
n∈Z 3
exp(iπ(n + ǫ 1 /2)τ t (n + ǫ 1 /2) + iπ(n + ǫ 1 /2) t ǫ 2 ).
Remark 5.4. For adierentapproah onthis result, see[Mea08℄.
The initial aim of Serre's letter was of ourse the existene of optimal urves of genus
3
.However, one does not know how to ompute diretly the value of
χ 18 over nite elds.
Therefore, as Serre suggested, when
A
is ordinary, we lift(A, a)
anonially over a numbereld and there, we use formula (2). Doing the omputation with enough preision, we an
reognizethis value asan algebrai number. Finallywe redue itto theinitial nite eldto
seeifitisa square.
As the Jaobian of an optimal urve is isogenous to the power of an ellipti urve
E
, in[Rit10℄,we worked out this proedure expliitinthe partiular ase
A = E 3. Leta 0 be the
produt prinipal polarization on
E 3 and M = a − 0 1 a ∈ M 3 (End(E))
. When End(E)
is an
order in an imaginary quadrati eld, it is well known that
M
is the matrix of a prinipalpolarizationon
E 3 ifandonly ifM
isa positive denite hermitian matrix with determinant
1
(see Remark3.2 and [Mum08,p.209℄). Moreover, whenE
is dened overa number eld,weshowhowto translatethe data
(E 3 , a 0 M)
intoaperiodmatrixoftheorrespondingtorus inorder to ompute theanalyti expressionofχ 18. Letus illustrate thisproedure withthe
following example.
Example 5.5. Does there exist an optimal urve
C
of genus3
overk = F 47 ? If so,
by Lemma 2.4 we know that
Jac C
is isogenous toE 3 where E
is an ellipti urve with
trae
−⌊ 2 √
47 ⌋ = − 13
. The urveE
is then an ordinary ellipti urve andEnd(E)
on-tains
Z [π] ≃ Z [(13 + √
13 2 − 4 · 47)/2] = Z [τ ]
(whereπ
is thek
-Frobeniusendomorphismandτ = (1 + √
− 19)/2
). HeneEnd(E) = Z [π]
istheringofintegersO L ofL = Q ( √
− 19)
. SineO Lisprinipal,E
isuniqueuptoisomorphism. Usingtheworkof[Sh98℄,one anseethat,
up to automorphism, there is a unique positive denite hermitian matrix M ∈ M 3 ( O L )
of
determinant
1
whihisindeomposable. InthelanguageofSetion 3.1,thismeansthatthere exists a unique positive denite unimodular indeomposable rank3
hermitianO L-module.
The abelian threefold
(E 3 , a 0 M )
isthenthe unique prinipally polarized geometrially inde- omposable abelian threefoldwithWeil polynomial(X 2 + 13X + 47) 3,upto isomorphism.
Lifting
E
anonially overQ
asE ¯ : y 2 = x 3 − 152x − 722
we an onsider the prinipallyabelianthreefold
( ¯ E 3 , a 0 M)
sineEnd( ¯ E) = O Laswell. Let[w 1 w 2 ]
beaperiodmatrix ofE
withrespetto the anonialregulardierential
dx/(2y)
. Ifwe letΩ 0 =
w 1 0 0 0 w 1 0 0 0 w 1
w 2 0 0 0 w 2 0 0 0 w 2
,
C 3 /Ω 0 Z 6 ≃ E ¯ 3 ( C )
withthe produtpolarizationa 0. Wethenneed tond asympletibasis
of
Ω 0 Z 6 forthepolarizationa 0 M
. ItisnotdiulttoprovethattherstChernlassofa 0 M
with respet to the pull-bak
ω i of the dierentials dx/(2y)
on eah urve is represented by
thematrix
H = 1 w 1 w 2
t M .
The alternated form
T
lassially assoiated toH
on thelattieΩ 0 Z 6 is T = Im( t Ω 0 HΩ 0 )
.
Onethennds a matrix
B ∈ GL 6 ( Z )
suh thatBT t B =
0 I 3
− I 3 0
and
Ω = Ω 0 t B
isaperiodmatrixfor thepolarizationa 0 M
. Finally,one omputesanapprox-imation of
χ = χ 18 (( ¯ E 3 , a 0 M ), ω 1 ∧ ω 2 ∧ ω 3 )
thankstotheanalytiformula(2 )andwereognizeitasanelementof
L
. Wendinouraseχ = (2 19 · 19 7 ) 2 .
Thevalue
χ
isanon-zero squareoverF 47 sobyTheorem5.3(ii )Serre'sobstrutionis trivial
andthere isa nonhyperellipti optimal urve ofgenus
3
overk
.Similaromputationsshowthatthereisanoptimalurveover
F q forq = 61, 137, 277
butnot
for
q = 311
. Notethatthisresultforq = 47
andq = 61
hasalreadybeen obtainedin[Top03℄usingexpliitmodelsand the othershave been onrmed by[AAMZ09℄. In [Rit10℄,tables
of values of
χ
as the one from Example 5.5 are given for( ¯ E 3 , a 0 M)
whereE ¯
is an elliptiurvewithlassnumber
1
andM
istakenfrom[Sh98℄. Fromthem,weangetforinstane:Proposition 5.6. Assume that
q = p n is suh that 4q = m 2 + d
with d = 7
(resp. 19
).
Thenthere exists an (expliit) genus
3
optimal urve overF q if and onlyif
m ≡ 1, 2
or4 (mod 7)
(resp. m
19
− 2 p
= 1
).
Moreover if this urve exists,it isnon hyperellipti.
Assume that
q = p n is suh that 4q = m 2 + 43
. In partiular 43
is a square in F p, let say
43 = r 2 with r ∈ F p. Thenthere existsa genus 3
optimalurve over F q if and onlyif
43 = r 2 with r ∈ F p. Thenthere existsa genus 3
optimalurve over F q if and onlyif
3
optimalurve overF q if and onlyif
m 43
α p
= 1
where
α
iseither− 2 · 3 · 7, − 487, − 47 · 79 · 107 · 173
or− 15156 ± 8214r
. Moreover ifthis urveexists,it isnon hyperellipti.
Remark 5.7. The term `expliit' inProposition 5.6omes from thefatthat for ertain
( ¯ E 3 , a 0 M )
of [Rit10℄ we were able to give the equation of a urveC ¯
suh thatJac ¯ C
isisomorphito
( ¯ E 3 , a 0 M)
using [Guà09℄. Henefor these ases,we have a `universal' family ofexpliitequations for theoptimal urve.Thefatthatthepreviousstatementisembarrassedlyumbersomerevealseithertheintrinsi
diultyoftheproblemorawrongattakangle. Moreover,thelimitsofthisstrategyalready
appearintheexample: theomputationoftheanoniallift,ofthematries
M
andofaperiodmatrix make it algorithmi in nature. Worse, the omputation of an approximation of
χ
istime-onsuming sine one has to reognize it as an algebrai number (atually for a good
hoie of themodel
E ¯
,χ
isan algebrai integer). Therefore large values of thedisriminant ofEnd(E)
seemout of reah.Itmight thenbe interesting totry tounderstand theprimedeomposition of
χ
algebraially.Klein's formula linking
χ 18 to the squareof the disriminant of plane quartis (see [Kle90 , Eq.118,p.462℄ and [LRZ10 , Th.2.23℄) makesus think about an analogue of the Néron-Ogg-
Shafarevih formulafor ellipti urve [Sil92,Appendix C,16℄. We shalltheninterpret
p | χ
intermsofthenatureof
(A, a) := ( ¯ E 3 , a 0 M ) (mod p)
. For instane, using[Ih95,p.1059℄,p | χ
ifandonlyif
(A, a)
isgeometriallydeomposable orahyperelliptiJaobian. Unfortunately, wedonot knowhowto detetalgebraiallythis lastpossibility(seethedisussionin[Rit09,Se.4.5.1℄).
Remark 5.8. We have not spoken yetabout thease
q
squarewhenp > 2
. First, whenp ≡ 3 (mod 4)
, one knows [Ibu93, p.2℄ that there exists an optimal genus3
urve. Thisurveiseven hyperellipti [Oor91℄butnot expliit(seehowever [KTW09℄for someexpliit
sub-ases). Also Fermat urve
x 4 + y 4 + z 4 = 0
is optimal ifn ≡ 2 (mod 4)
. Then, whenp ≡ 1 (mod 4)
andn ≡ 2 (mod 4)
, Ibukiyama (lo. it.) shows that there is an optimalurve. Ibukiyama'sstrategy uses a massformulaon quaternion hermitian forms to showthe
desentofanindeomposableprinipal polarizationonamodelover
F p ofE 3 where E/ F p 2 is
E/ F p 2 is
an ellipti urve withtrae
− 2p = − m
. The abelian threefoldand its quadrati twist beingisomorphiover
F p 2,he avoids theissue ofomputing Serre'sobstrution.
5.3. The geometri approah. Following aonstrution of Reillas[Re74℄,we were
able to give in [BR10℄ a geometri haraterization of Serre's obstrution. For the sake of
simpliity,let usassume that
char k 6 = 2
and that(A, a)/k
is geometrially theJaobian ofa nonhyperellipti genus3
urve. Sinek
isaniteeld,thereexistsasymmetrithetadivisorΘ
(forthepolarizationa
)dened overk
. LetΣ
betheunionof2 ∗ Θ
andoftheuniquedivisorin
| 2Θ |
withmultipliity greaterthan or equal to4
at0
.Proposition 5.9. Let
α ∈ A(¯ k) \ { 0 }
. The urveX ˜ α = Θ ∩ (Θ + α)
is smooth andonneted if and onlyif
α ∈ A(¯ k) \ Σ
.Hene, thedivisor
Σ
representsabad lousthatneedsto beavoidedinthesequel. Assumingthat
α / ∈ Σ
,theinvolution(z 7→ α − z)
ofX ˜ α isxedpoint freeandsoX α = ˜ X α /(z 7→ α − z)
isa smooth genus
4
urve.Proposition 5.10. The urve
X α is non hyperellipti and its anonial model in P 3 lies
on a quadri
Q α whihis smooth.
To gofurther,weneed to assumethat
α
isrational. Whenk
isbigenough,suh anα
alwaysexists. Wethenobtain the following result.
Theorem 5.11. Assume there exists
α ∈ A(k) \ Σ
. Then(A, a)
isa Jaobian ifand onlyif
δ = Disc Q α isa square ink ∗.
Let us sketh the proof. A non hyperellipti genus
4
urveX
lies anonially inP 3 on the
intersetion of a unique quadri
Q
and a ubi surfaeE
. If we assume thatQ
is smooth,then
X
hastwog 3 1 oming fromthetwo rulingsofQ
byintersetingthemwithE
. Moreover,
an easy omputation shows that
Disc Q
is a square if and only if these twog 1 3 are dened
over
k
. NowReillas' onstrution,whihanbeusedwhen(A, a)
istheJaobianof aurve,shows that
X α hastwo (ratherexpliit) rational g 1 3. To onlude, itis then enough to show
thata quadrati twist of
(A, a)
(whih isno morea Jaobian)leads to two onjugateg 3 1.
The advantage of this approah is that it stays over the nite eld
k
and is ompletelyalgebrai. Unfortunately, sofar, we do not seehow to ompute