• Aucun résultat trouvé

An extension of Kedlaya's algorithm to superelliptic curves

N/A
N/A
Protected

Academic year: 2021

Partager "An extension of Kedlaya's algorithm to superelliptic curves"

Copied!
16
0
0

Texte intégral

(1)

HAL Id: inria-00514826

https://hal.inria.fr/inria-00514826

Submitted on 3 Sep 2010

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

An extension of Kedlaya’s algorithm to superelliptic curves

Pierrick Gaudry, Nicolas Gürel

To cite this version:

Pierrick Gaudry, Nicolas Gürel. An extension of Kedlaya’s algorithm to superelliptic curves. Asi- acrypt, 2001, Gold Coast, Australia. pp.480-494, �10.1007/3-540-45682-1_28�. �inria-00514826�

(2)

Algorithm to Superellipti Curves

PierrikGaudryandNiolasGurel

LIX,

Eolepolytehnique,91128Palaiseau Cedex,Frane

fgaudry, gurelglix.polytehnique.fr

Abstrat. Wepresentanalgorithmforountingpointsonsuperellipti

urvesy r

=f(x) overa nite eld F

q

of smallharateristi dierent

fromr.Thisisanextensionofanalgorithmforhyperelliptiurvesdue

toKedlaya.Inthisextension,theomplexity,assumingr andthegenus

are xed, is O(log 3+"

q) in time and spae, just like for hyperellipti

urves.Wegivesomenumerialexamplesobtainedwithourrstimple-

mentation,thusprovingthatryptographisizesarenowreahable.

1 Introdution

Inthepastfewyearsalotofandidateshavebeenproposedtoenlargethesetof

groupsthatanbeusedinprotoolsbasedonthedisretelogarithmproblemlike

DiÆe{Hellman orElGamal. Beside the lassialmultipliativegroups of nite

elds, the most famous are ertainly the systems based onellipti urves [21,

26℄.Indeed,forthesesystemstheonlygeneralattaksknownarevariantsofthe

PollardRhomethodwhihrequireexponentialtimeomputation;in pratieit

meansthatthekeysizeismuhshorterthaninasystemthat usesniteelds.

Thereafter,systemsbasedonhyperelliptiurveswereproposed[22℄.Theyseem

to havethesameadvantages asellipti urveryptosystems(at leastwhenthe

genusislessthan4[1,14℄).

Morereently, systemsbasedon the disrete logarithm problem in the Ja-

obians of other urves were designed. Namely, in the literature, we an now

ndalgorithms forworkingin Jaobiansofsuperelliptiurves[13℄ andof C

ab

urves[2℄. Several works relatedto these urveshave already beenpublished,

onerningseurityissues[4℄,eÆieny[17,6℄,buildingurveswithknownnum-

berof points[3℄, orpossible usein aWeil restrition attakon ellipti urves

[5℄. The next step for studying the possible ryptographiuse of these urves

is to oneive an algorithm for ounting points of the Jaobian of a random

urve.Indeed, this is thoughtto be oneof themost seure ways ofbuilding a

ryptosystembyalargepartoftheommunity.

Intheaseofelliptiurves,thisproblemofpointountinghasbeenahal-

lenge of thepast 15yearsand nowadayswehavesatisfatory solutions.When

the harateristi of thebase eld is large thebest known method is Shoof's

algorithmandalltheimprovementsleadingtotheso-alledShoof{Elkies{Atkin

algorithm.Wereferto[7℄or[23℄forsurveysofthesetehniquesandtotherefer-

(3)

to be enough for ryptographi sizes. The situation is quite dierent in small

harateristi: twoyearsago, Satoh[32℄ showedthat p-adi methodsusing the

anonialliftouldleadtoanalgorithmasymptotiallyfasterthanSEA.Some

work hasbeendone onsequentlyon the subjetto extendit to harateristi

2 [33,9℄, to implement it and obtainnew reords[9℄, to use less memory [34℄,

andtoombineitwithanearly-abortstrategyforgeneratingseureurves[10℄.

Mestre,HarleyandGaudryreentlyproposedarelatedalgorithm,basedonthe

arithmeti-geometrimean,forellipti urvesandhyperelliptiurvesof genus

2in haratersti2; aniefeature ofthis tehniqueis that it does notexpli-

itly make useof j-invariants, ofmodular equationsnor of Velu-typeformulae,

and these had previously been the main obstrutions to generalizing beyond

theellipti ase.However,theAGM method doesnotseemtoextendeasily to

non-hyperelliptiurves.Anotherapproah,also usingp-adi methodsbut not

basedonanoniallifting, hasbeenproposed by Kedlaya[18℄.Hismethod ap-

pliestohyperelliptiurvesinsmalloddharateristi.Theomplexityin time

isO(log 3+"

q),forurvesoverF

q

ofxedgenus,i.e.thesameasallthevariants

ofSatoh'smethodandtheomplexityinspaeisO(log 3

q)whihisthesameas

Satoh'soriginalalgorithm,butbadomparedtothealgorithmof[34℄orAGM.

Theontributionofthispaperistwofold:rstly weshowthatKedlaya'sal-

gorithmanbeextendedinaratherstraightforwardwaytosuperelliptiurves;

seondly we report some results obtained with our rst implementation writ-

tenin Magma.Toourknowledge, these aretherstpublished pointounting

omputationsforrandomhyperelliptiandsuperelliptiurvesofryptographi

sizes.

Thepaperis organizedas follows: after realling somebasisabouturves

andp-adinumbers,wedesribeKedlaya'soriginalalgorithmandshowhowto

adapt it for superellipti urves. Then we give some moredetails on the way

thesealgorithmsanbehandledinpratieandweestimatetheomplexity.We

onludeby numerialexamplesand remarks about theuse of these urvesin

ryptography.

2 Bakground on Algebrai Curves and p-adi Number

Rings

Inthissetion,wereallsomebasifatsaboutalgebraiurvesoverniteelds

andp-adinumbers.Weshallnotgivepreisedenitionsandwereferthereader

to lassialbooks onthesubjet([12,20,19,24℄forinstane).

2.1 Hyperelliptiand SuperelliptiCurves

LetF

q

beaniteeld with q=p n

elements. Weshallonsider onlytwotypes

ofurvesoverF ,namelyhyperelliptiandsuperelliptiurves.

(4)

equation ofthe form

y r

=f(x);

whererisaprimedierentfrompandf ismoni,squarefreeofdegreedoprime

tor.

Withsuhadenition,Cisnon-singularinitsaÆnepart,andadmitsaunique

plaeofdegree1at innity.Moreoveritsgenusisgivenbyg=

(d 1)(r 1)

2 .

Denition2 In harateristi dierent from 2, a hyperellipti urve isa su-

perellipti urve whose equation is of the form y 2

=f(x), with r=2and f of

degree 2g+1.

Note that there exists a more general denition of hyperellipti urves whih

donotexludetheaseof harateristi2.Butthealgorithmswewilldesribe

workonlyforthispartiular ase.

LetC beasuperelliptiurveof genus g. Assoiatedto this urve, onean

dene itsJaobian, notedJ(C),whihis anite abeliangroup.Inthepastfew

years,severalalgorithmsweredevelopedtoomputeexpliitlyinthisgroup[13,

2,17,6℄. The next step is to study the order of J(C). For this the q-th power

Frobenius endomorphismand itsharateristipolynomial (T)are keytools.

Morepreisely,(T)anbewritten as

(T)= 2g

X

i=0 a

i T

i

; with a

2g

=1; a

i

=q g i

a

2g i

fori=0;:::;g 1;

andallitsrootshaveabsolutevalue p

q.ThisisessentiallytheRiemannHypoth-

esisforzetafuntionsofurves.Forus,theinterestingfatisthat#J(C)=(1).

Ourgoalinthispaperistoompute(T)andtoobtain#J(C)asabyprodut.

2.2 The Ring Z

q

Let K bethe (unique upto isomorphism) unramiedextension of degreen of

Q

p

;itsresidualeld isF

q

.Wedenote byZ

q

thering ofintegersof K. Inorder

to onstrut it, we anstart with the polynomial P(t)whih denes F

q asan

algebraiextensionofF

p

;wethenonsidertheextension

Z

q :=Z

p

[t℄=(P(t));

where the polynomial P(t) is obtained from P(t) by lifting triviallyits oeÆ-

ientsto p-adiintegers.In pratie,anelement z of Z

q

anberepresented as

apolynomialz=z

n 1 t

n 1

+z

n 2 t

n 2

++z

1 t+z

0

takenmodulo P(t)and

where thez

i

areintegersmoduloapowerof palled thepreisionatwhihthe

omputationisdone.

ItanbeshownthattheGaloisgroupofKoverQ

p

isyli.Wewilldenote

(5)

q

writing z

for an element z in Z

q

expressed on a polynomial basis asabove.

Later on, we will desribe how to preompute t

and then z

is obtained as

follows:

z

= n 1

X

i=0 z

i t

i

!

= n 1

X

i=0 z

i (t

) i

:

3 Kedlaya's Algorithm and its Extension

3.1 Overview ofKedlaya's Algorithmfor Hyperellipti Curves

LetC beahyperelliptiurveof genus g given byits equationy 2

=f(x) over

F

q

. Followingthe onstrutionof Kedlaya(see also [20℄,page72), we onsider

the urveC 0

obtainedfrom C by removingthe point at innity andthe points

withvertialtangent(i.e.y=0).

ThereisawaytolifttheoordinateringofC 0

alledtheweakompletion[27℄,

withtheniepropertythatitsohomologyveriesa\Lefshetz traeformula"

[28℄andhenegivesinformationabouttheardinalitiesoftheinitialurveC.

Taking alowbrowpoint ofviewin whihweanforgetabouttheurveC 0

,

weshall workonthevetorspaegenerated overthep-adinumbereldK by

thefollowingdierentialforms:

D=

x i

dx

y

; i2[0;2g 1℄

;

in whih we have the relations oming from the equation of the urve and

d'(x;y) 0for every rational funtion '. On the dierential forms onean

dene a Frobenius ation whih is ompatible with the p-th power Frobenius

onC: takex

=x p

,y

givenby(y

) 2

=f(x)

and (dx)

=px p 1

dx. Kedlaya

showsinaonstrutivewaythatthespaeDisstableundertheationofthis.

Hene isanendomorphismofavetorspaeofdimension2g;andeverything

is done in order for its harateristi polynomial to be losely related to the

(T) weare looking for. The heartof Kedlaya'salgorithm isthen to ompute

thematrixof forthegivenbasisofD.

Foreahiin[0;2g 1℄,

x i

dx

y

= 1

y

px ip+p 1

dx;

thereforethetrikypartistheomputationof 1

y

.Thisisnotdenedinalifted

oordinatering beauseit involvesa square root and that is areasonwhy we

usetheweakompletion.Fromapratialpointofview,itmeansthatweshall

beableto expand 1

as apower seriesin = 1

2

: startingwith the denition

(6)

(y ) =f(x) , wehave

1

y

=(f(x)

) 1=2

=(f(x)

f(x) p

+f(x) p

) 1=2

=(f(x) p

) 1=2

1+ f(x)

f(x) p

f(x) p

1=2

= 1

y p

( 1+ p

(f(x)

f(x) p

)) 1=2

:

Bytheusualpowerseriesexpansionof(1+X) 1=2

wegetanexpressionofthe

form

1

y

=y p

X

k 0 P

k (x)

pk

=y 1

(p 1)=2

X

k 0 P

k (x)

pk

:

Notethatpdivides(f(x)

f(x) p

)sothatthepowerofpdividingP

k

(x)tends

to innity ask grows(atually this is what is expeted due to the theoretial

onstrutionoftheweakompletion).Weannowwrite

x i

dx

y

= 0

X

k 0 Q

k (x)

k 1

A dx

y

;

where Q

k

(x) are polynomials. The algorithm proeeds asfollows:weompute

this expression up to somepreisionin , and then we use the relationsin D

desribedabovetoreduetheexpressiontoapolynomialofdegreeatmost2g 1,

times dx

y

.InthiswayweshallprovethatDisindeed -stableandmoreoverwe

obtainanexplitdesriptionoftheationofonthebasis.Forthiswewilluse

threestrategiesofredution:

Red 1. Firstofall,usingtheequationoftheurve,oneanwrite

Q

k (x)

k

=(

k

(x)f(x)+

k (x))

k

=

k (x)

k 1

+

k (x)

k

;

where

k and

k

arethe quotientand theremainder in thedivision of Q

k

byf.ThereforeoneanassumethatQ

k

(x)isofdegreeatmost2gforallk,

exeptforQ

0

(x)forwhihoneanshowthatthedegreeisatmost2pg 1.

Red 2. Then we use the relations of ohomology to rewrite the series in the

form Q(x) dx

y

. Fixk 1 and onsider the term Q

k (x)

kdx

y

. Let U(x) and

V(x)besuh thatQ

k

(x)=U(x)f(x)+V(x)f 0

(x) (theydoexist beausef

issquarefree).Using

d

V(x)

y 2k 1

0;

oneobtains

Q

k (x)

k dx

y

U(x)+ 2

2k 1 V

0

(x)

k 1

dx

y :

Repeatingthisfordereasingk's,weanrewriteeverythingontheonstant

(7)

Red 3. Finally,intheexpressionQ(x)

y

thatweobtained,oneanreduethe

degreeÆofQtoatmost2g 1inthefollowingway.Assume Æ2g:using

d(x Æ 2g

y)0;

onegetsapolynomialofdegreeÆthatanbesubtratedfromQ.

Atthispoint,wehaveomputeda2g2gmatrixM suhthat

0

B

B

dx

y

.

.

.

x 2g 1

dx

y 1

C

C

A

= M 0

B

B

dx

y

.

.

.

x 2g 1

dx

y 1

C

C

A :

MostoftheoperationsdoneduringtheomputationinvolveelementsofZ

q ,but

at theendwemayhaveto dividebysmall powersofp.FinallytheoeÆients

ofM liein p s

Z

q

withasmall,preditables, whihdependsonlyonpandg.

Thenalstepisthentoomputetheharateristipolynomialofthematrix

MM

M

n 1

;

whihhasoeÆientsin Z

2

andisap-adiapproximationof (T).

3.2 SuperelliptiCurves

LetC beasuperelliptiurvegivenbyitsequationy r

=f(x)withf ofdegreed

overF

q

.Thetheoryisexatlythesameasforhyperelliptiurves.Inthepresent

ase,thespaeofdierentialforms weonsideris

x i

dx

y j

; i2[0;d 2℄; j2[1;r 1℄

:

TheFrobeniusationliftingthep-thpowerFrobenius onC isdened similarly:

takex

=x p

,y

givenby(y

) r

=f(x)

and(dx)

=px p 1

dx.

Again,thespaeof dierentialforms hasbeen hosensuh that itis stable

under theationof ;wewillnowdesribetheredutionproesswhihallows

us to rewrite

x i

dx

y j

overthebasis. Fixan i2[0;d 2℄and aj 2[1;r 1℄.

Weanwrite

1

y j

asapowerseries

1

y j

=y jp

1+ f(x)

f(x) p

y rp

j=r

=y jp

X

k 0 P

k (x)

pk

;

wherewehaveset =y r

.Hene weanwrite

x i

dx

y j

= 0

X

k 0 Q

k (x)

k 1

A dx

y jpmodr

:

In the following, we let ` = jpmodr. We now proeed with three redution

(8)

k

areofdegreeatmostd 1,exeptfortherstone.

Red 2. Then,rewritethetermin k

asatermin k 1

.Fork1,letU(x)and

V(x)besuhthat Q

k

(x)=U(x)f(x)+V(x)f 0

(x),onehas

Q

k (x)

k dx

y

`

U(x)+ r

r(k 1)+` V

0

(x)

k 1

dx

y

` :

Red 3. Finally,weareleftwithanexpressionoftheformQ(x) dx

y

`

,whereQ(x)is

apolynomialofdegreeÆthatweanreduetodegreeatmostd 2:assume

Æ d 1, the exat dierential d(x Æ d+1

y r l

) 0 gives apolynomial of

degreeÆthatanbesubtratedfromQ(x).

Weobtaina2g2gmatrixM andweonludeasbeforebytakingthehara-

teristipolynomialofits\norm".

Notethatthedierentialformsin dx

y j

aresentbytothesubspaegenerated

by forms in dx

y

`

with` =jpmodr. As aonsequene, M isa matrixthat an

be viewed in bloks of size d 1, with the property that there is exatlyone

non-zeroblokin eahrowblokand eaholumnblok.

4 Details and Complexity

4.1 Preision ofthe Computation

Theintermediateresultobtainedfrom thealgorithmof setion3isanapprox-

imation of thepolynomial (T)that weare looking for, and by omputing to

suÆient preision we an determine it exatly. Two parameters have to be

tuned, to ensure that at theend we get enoughinformation to onlude.The

rstisthep-adipreisionp

atwhihwetrunateelementsofZ

p

.Theseond

isthe-adipreisionatwhihwetrunatetheseries.

Boundson theoeÆients of (T)an be dedued from the bounds on its

roots: ja

i j

2g

i

q i=2

fori 2[1;g℄. We assumethat q is largeompared to the

genus, so that a

g

determines the required preision. Hene we need to know

(T)modulod2 2g

g

q g=2

etobesuretoreoveralltheoeÆients.Thereforethe

workingpreisionshouldbeatleast

=

log

p

2

2g

g

q g=2

:

Thepreisionin is moreproblemati:at rstsightitis notlearthat we

do notneed all theterms ofthe series to get aresultwhih makessense even

modulop.Atuallyinthepowerseriesexpansion,oneanseethattheoeÆient

in k

(whih is apolynomial overZ

q

)isdivisible by apowerofp whih grows

toinnityatthespeedofk=p.Heneitappearsthatthepreisionin should

beatleast ptimes thep-adipreision. Moreover,theredutionproessalso

perturbsthings:startingwith atermQ

k (x)

kdx

`

, withp m

dividing Q

k

(x),one

Références

Documents relatifs

The notion of p-rationality of number elds naturally appears in several branches of number the- ory. In Iwasawa theory, the study of Galois groups of innite towers of number elds,

This thesis treats diraction of time harmonic electromagnetic elds by nite size crys- tals and quasi-crystals for large wavelengths and propagation of modes in photonic crystal

Abstract We prove in this paper an uniform surjectivity result for Galois representations associated with non-CM Q -curves over imaginary quadratic elds, using various tools for

Earlier work on Iwasawa main conjecture includes the work of Mazur-Wiles [42], Wiles [71] for p -adic families of Hecke characters of totally real elds using the Eisenstein

The new method can prun- ing most of the invalid roots as the geometric methods do, but the computation time for each pruning step is much less than the geometric methods, which

exponential Young funtions for stationary real random elds whih are bounded.. or satisfy some nite exponential

partiular when q is not a square), there exists a maximal asymptotially exat sequenes of algebrai funtions elds namely attaining the Generalized Drinfeld-Vladut bound.. Then, we

An hexahedral fi nite element is developed with a classical nodal formulation for mechanical fi elds while the electromagnetic ones are expressed by an edge formulation with vector fi