• Aucun résultat trouvé

Construction of secure random curves of genus 2 over prime fields

N/A
N/A
Protected

Academic year: 2021

Partager "Construction of secure random curves of genus 2 over prime fields"

Copied!
19
0
0

Texte intégral

(1)

HAL Id: inria-00514121

https://hal.inria.fr/inria-00514121

Submitted on 1 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.

prime fields

Pierrick Gaudry, Eric Schost

To cite this version:

Pierrick Gaudry, Eric Schost. Construction of secure random curves of genus 2 over prime fields. Eu- rocrypt, 2004, Interlaken, Switzerland. pp.239-256, �10.1007/978-3-540-24676-3_15�. �inria-00514121�

(2)

PierrikGaudry 1

and

EriShost 2

1

LaboratoireLIX,

Eolepolytehnique,Frane

gaudrylix.polytehnique.fr

2

LaboratoireSTIX,

Eolepolytehnique,Frane

Eri.Shostpolytehnique.fr

Abstrat. For ountingpointsof Jaobians of genus 2urvesdened

overlargeprimeelds,thebestknownmethodis avariantof Shoof's

algorithm.Wepresentseveralimprovementsonthealgorithmsdesribed

by GaudryandHarley in2000.Inpartiular werebuild thesymmetry

that hadbeen brokenbythe use ofCantor'sdivision polynomials and

design afaster division by 2 and a division by 3. Combined with the

algorithm by Matsuo, Chao and Tsujii,our implementationan ount

thepointsonaJaobianofsize164bitswithinaboutoneweekonaPC.

1 Introdution

Genus2hyperelliptiurvesprovideaninterestingalternativetoelliptiurves

forthedesignofdisrete-logbasedryptosystems.Indeed,forasimilarseurity,

thekeyorsignaturelengthsarethesameasforelliptiurvesandfurthermore

thesizeof thebaseeld inwhihtheomputationstakeplaeistwiesmaller.

Duringthelastyears,eortsinimprovingthegrouplawalgorithmsmadethese

ryptosystemsquiteompetitive[19,25℄.

Toensuretheseurityofthesystem,it isrequiredto haveagroupoflarge

prime order. Until reently, for the Jaobian of a genus 2 urve, only spei

onstrutionsprovidedurveswithknownJaobianorder,namelytheomplex

multipliation (CM) method [34℄ andthe Koblitz urves.These urveshavea

veryspeial struture;although nobody knowsif theyareweakerthangeneral

urves,itispertinenttoonsiderrandomurvesaswell.Thisraisestheproblem

ofpoint-ounting:givenarandomurve,ndthegrouporderofitsJaobian.

With today's stateof theart, theomplexityof thepoint ounting taskin

genus2highlydependsonthesizeoftheharateristiofthebaseeld:inshort,

thesmallertheharateristi,theeasierthetaskofpointounting(\easy"means

fastanddoesnotmeanthatthetheoretialtoolsaresimple).

In the ase of genus 2 urves in small harateristi p, the point ounting

problem was reently solved using p-adi methods [31,23,20℄. The partiular

asewherep=2isinfattreatedalmostasquiklyasingenus1.Unfortunately,

thesedramatiimprovementsdonotapplywhenpbeomestoolarge(say,afew

(3)

For largep, the best known algorithms are variantsof Shoof's algorithm,

theoretialdesriptionsofwhihanbefoundin[26,18,1,16℄.In2000,Gaudry

and Harley [11℄ designed and implemented the rst pratial genus 2 Shoof

algorithm,makinguseofCantor'sdivisionpolynomials[8℄.Toreahreasonable

sizes,however,itwasneessaryto ombinetheShoofapproahwithaPollard

lambda method. Their reord wasa random genus 2 urve overa prime eld

of size about 10 19

, thus too small to be used in aryptosystem. For\medium

harateristi", they also proposed to use the Cartier-Manin operator to get

additionalinformationthatanbeombinedwithothers.Therefore,formedium

harateristip(say10 9

,see[5℄),pointountingiseasierthanforverylargep.

Wementionedthat in thenon-small harateristi ase,onethe groupor-

derhasbeenomputedmodulo somelargeinteger,theomputation isnished

usingaPollardlambdamethod.Matsuo,ChaoandTsujii[21℄proposedaBaby-

step/Giant-stepalgorithmthat speedsupthislastphase.With thisdevie and

usingtheCartier-Manintrik,theyperformedapointountingomputationof

ryptographialsizeforamediumharateristield.

Inthispaper,weimproveonthemethodsof[11℄,sothat,ombinedwiththe

algorithm of [21℄,weanreah ryptographialsize overprime elds. Ourim-

provementsareonernedwiththeonstrutionandthemanipulationoftorsion

elementsintheShoof-likealgorithmof[11℄.Theimpatoftheseimprovements

is asymptotially by a onstant fator, but they yield signiant speed-up in

pratieforthesize ofinterestin ryptography.Wenowsummarizethem:

Our rst ontribution is the reintrodution of symmetries that were lost

in [11℄. Indeed, the use of Cantor's division polynomials to onstrut torsion

elementsis veryeÆient,but theresulting divisor is given asasum of points

insteadofin Mumfordrepresentation. Thereforeafatorof2in thedegreesof

the polynomials that are manipulated is lost.In Setions 3.2 and 3.3, wegive

algorithmstosavethisfatorof2inthedegrees.

In[11℄,itisproposedtobuild2 k

-torsionelementsusingahalvingalgorithm

basedonGrobnerbasisomputations.Ourseondontributionis afasterdivi-

sionby2,usingabetterrepresentationofthesystem;inthesamespiritweshow

that adivision by 3an also be done: this is desribed in Setion 4. Another

pratial improvement is theubiquitous use of an expliitation on the roots

omingfromthegrouplawtospeed-upthefatorizationsthatouratdierent

stages.Weexplainitin detailsintheaseofthedivision by2inSetion 3.4.

To illustrate and to test the performane of our improvements, we imple-

mentedtheminMagmaorNTLandmixedthemwiththealgorithmof[21℄and

an early abort strategy. Our main outome is the rst onstrution of seure

randomurvesofgenus2overaprimeeld,asweobtainedJaobiansofprime

orderofsize about2 164

.

2 Generalities

Inthiswork,pdenotesaxedoddprime,F

p

istheniteeldwithpelements,

2

(4)

squarefreemonipolynomialin F

p

[X℄ofdegree5.Themainobjetweonsider

is theJaobianJ(C) ofC. Wehandle elementsof J(C) throughtheirMumford

representation: eah element of J(C)an be uniquelyrepresented bya pair of

polynomialshu(x);v(x)i,whereuismoniofdegreeatmost2,visofdegreeless

than thedegreeof u,and udivides v 2

f.The degreeof theu-polynomialin

Mumford'srepresentationisalled theweightofadivisor.IfK isanextension

eld of F

p

, we may distinguish the urves dened on K and F

p

, by denoting

themC=K andC=F

p

;theJaobiansareorrespondinglydenotedbyJ(C=K)and

J(C=F

p

). Forpreise denitions and algorithms for the group law, we refer to

[22℄and[7,19℄.

LetF

p

beanalgebrailosureofF

p

andletusonsidertheFrobeniusendo-

morphismonJ(C=F

p

)denotedby.ByWeil'stheorem(see[24℄),theharater-

istipolynomial(T)of hastheform(T)=T 4

s

1 T

3

+s

2 T

2

ps

1 T+p

2

,

where s

1 and s

2

are integerssuhthat js

1 j 4

p

p andjs

2

j 6p. Furthermore

#J(C)=(1)=p 2

+1 s

1

(p+1)+s

2 .

In point-ounting algorithms based on Shoof's idea [27℄, the torsion ele-

ments ofJ(C) play animportant role. IfN is apositiveinteger, thesubgroup

ofN-torsionelementsofJ(C=F

p

)isanitegroupdenoted byJ(C)[N℄; itisiso-

morphito(Z=NZ) 4

andhasthestrutureofafreeZ=NZ-mo duleofdimension

4(see[24℄).Furthermore,theharateristipolynomialoftherestritionof to

J(C)[N℄is(T)modN.Applyingthistodierentsmallprimesorprimepowers

leadstothegenus2ShoofalgorithmthatisskethedinAlgorithm1.

Algorithm1Sketh ofagenus2Shoofalgorithm

1. ForsuÆientlymanysmallprimesorprimepowers`:

(a) LetL=f(s

1

;s

2 ); s

1

;s

2

2[0;` 1℄g.

(b) While#L>1do

{ Construtanew`-torsiondivisorD;

{ Eliminatethoseelements(s

1

;s

2

)inLsuhthat

4

(D) s1 3

(D)+s2 2

(D) (ps1mod`)(D)+(p 2

mod`)D6=0

() Dedue(T)mod`fromtheremainingpairinL.

2. Dedue(T)fromthepairs(`;(T)mod`)byChineseremaindering,orusingthe

algorithmof[21℄.

Ourontributionistoimprovetherstpartofthealgorithm,theonstru-

tionof `-torsiondivisors;theomputations forsmall primes andprime powers

arerespetivelydesribedinSetions3and4.

WewillfrequentlymakegeneriityassumptionsontheurveCanditstorsion

divisors.WeassumethatCishosenrandomlyamonggenus2urvesdenedover

alargeeldF

p

,soweanexpetthatwithhighprobability,suhassumptionsare

satised.Theaseswhenourassumptionsfailshouldrequirespeialtreatments,

whiharenotdevelopedhere.

Fortheomplexityestimates,wedenotebyM(d)thenumberofF

p

-operations

requiredtomultiplytwopolynomialsofdegreeddenedoverF

p

.Wemakethe

(5)

if no preise referene is given for an algorithm, then it an be found in [32℄,

togetherwithaomplexityanalysisintermsofM.

3 Computation modulo a small prime `

In thelassialShoof algorithm forellipti urves,aformal `-torsionpoint is

used: theomputations aremade withapoint P =(x;y),where x anelsthe

`-thdivisionpolynomial

`

andyislinkedtoxbytheequationoftheurve.In

otherwords,weworkinarank2polynomialalgebraquotientedbytworelations:

F

p

[x;y℄=(

` (x);y

2

(x 3

+ax+b)).

Ingenus2,weimitatethisstrategy.Aordingto[18℄,itisenoughtoonsider

the`-torsiondivisorsofweight2(thisisnotsurprisingsineageneridivisorhas

weight2).LetthusDbeaweight2divisorgiveninMumfordrepresentation,D=

hx 2

+u

1 x+u

0

;v

1 x+v

0

i:ThenthereexistsaradialidealI

` ofF

p [U

1

;U

0

;V

1

;V

0

suhthat

D 2J(C)[`℄ () '(u

1

;u

0

;v

1

;v

0

)=0; 8'2I

` :

Byanalogywithelliptidivisionpolynomials,thisidealI

`

isalledthe`-thdivi-

sionideal.Thereare` 4

1non-zero`-torsionelements,sothatI

`

hasdimension

0anddegreeat most` 4

1;generially,bytheManin-Mumfordonjeture[15,

p.435℄,allnon-zerotorsiondivisorshaveweight2,sothedegreeofI

`

isexatly

` 4

1.

Fromtheomputationalpointofview,agoodhoieforageneratingsetofI

`

isaGrobnerbasisforalexiographiorder.UsingtheorderU

1

<U

0

<V

1

<V

0 ,

we an atually predit the shape of this Grobner basis. Indeed, if D is an

`-torsiondivisor, then its opposite D is also `-torsion,so it hasthe sameu-

oordinates, and opposite v-oordinates. Furthermore,we make the generiity

assumptionthatallthepairsfD; Dgof`-torsiondivisorshavedierentvalues

foru

1

.Then,theGrobnerbasisfortheidealI

`

takestheform

I

`

= 8

>

>

<

>

>

: V

0 V

1 S

0 (U

1 )

V 2

1 S

1 (U

1 )

U

0 R

0 (U

1 )

R

1 (U

1 );

where R

1

is a squarefree polynomial of degree (`

4

1)=2 and R

0

;S

1

;S

0 are

polynomials of degree at most (`

4

1)=2 1. If suh a Grobner basis for I

`

is known, then it is not diÆult to imitate Shoof's algorithm, by working in

the quotient algebra F

p [U

1

;U

0

;V

1

;V

0

℄=I

`

: Unfortunately, no easy omputable

reurreneformulaeareknownthatrelateGrobnerbasesof`-divisionidealsfor

dierentvaluesof`,justlikefordivisionpolynomialsofelliptiurves.Therefore

weshallstartwiththeapproahof[11℄usingCantor'sdivisionpolynomialsand

showthat weanderiveeÆientlyamultipleofR

1 .

3.1 Cantor's divisionpolynomials

Letusxaprime`.Cantor'sdivisionpolynomials[8℄arepolynomialsinF

p [X℄,

denoted byd , d , d ,e ,e ,,with thefollowingproperty:foradivisorP =

(6)

hx x

P

;y

P

iofweight1,themultipliationofP by`in J(C)isgivenby

[`℄P =

x 2

+ d

1 (x

P )

d

2 (x

P )

x+ d

0 (x

P )

d

2 (x

P )

; y

P

e

1 (x

P )

(x

P )

x+ e

0 (x

P )

(x

P )

:

Thesepolynomialshaverespetivedegrees2`

2

1,2`

2

2,2`

2

3,3`

2

2,3`

2

3,

3`

2

2andareeasilyomputedbymeansofreurreneformulae.Evenifanaive

method is used, theost oftheir omputationis byfar negligibleompared to

thesubsequentoperations.

Now,letD=hx 2

+U

1 x+U

0

;V

1 x+V

0

ibeageneridivisorofweight2,where

U

1 ,U

0 , V

1 , V

0

areindeterminates,subjetto theonditionthat x 2

+U

1 x+U

0

divides(V

1 x+V

0 )

2

f.ThedivisorDanbewrittenasthesumoftwoweight

1divisors P

1

=hx X

1

;Y

1

iand P

2

=hx X

2

;Y

2

i,where U

1

= (X

1 +X

2 ),

U

0

=X

1 X

2

,and whereY

1 andY

2

satisfyV

1 X

1 +V

0

=Y

1 andV

1 X

2 +V

0

=Y

2 .

SineD=P

1 +P

2

,thenD is`-torsionifandonlyif[`℄P

1

= [`℄P

2 .

Rewriting this equation using Cantor's division polynomials, we get four

equationsthat mustbesatisedforD to be`-torsion.Someofthese equations

aremultiplesofX

1 X

2

:thisisanartifatduetothesplittingofDintodivisors

of weight1 and ifthis is the aseone should divide out this fator. Hene we

obtainthefollowingsystem:

8

>

>

<

>

>

: E

1 (X

1

;X

2

) =(d

1 (X

1 )d

2 (X

2 ) d

1 (X

2 )d

2 (X

1 ))=(X

1 X

2 )=0;

E

2 (X

1

;X

2

) =(d

0 (X

1 )d

2 (X

2 ) d

0 (X

2 )d

2 (X

1 ))=(X

1 X

2 )=0;

F

1 (X

1

;X

2

;Y

1

;Y

2

)= Y

1 e

1 (X

1 )e

0 (X

2 )+Y

2 e

1 (X

2 )e

0 (X

1

) =0;

F

2 (X

1

;X

2

;Y

1

;Y

2

)= Y

1 e

2 (X

1 )e

0 (X

2 )+Y

2 e

2 (X

2 )e

0 (X

1

) =0:

Considernowthenite-dimensionalF

p

-algebra

B=F

p [X

1

;X

2

;Y

1

;Y

2

℄=(E

1

;E

2

;F

1

;F

2

;Y 2

1 f(X

1 );Y

2

2 f(X

2 )):

In a generi situation, the minimal polynomial of (X

1 +X

2

) in B is then

preisely the polynomial R

1

that appears in the Grobner basis of I

`

(failures

ouldour,e.g.,ifthereexistsan`-torsiondivisorD=P

1 +P

2

,suhthat[`℄P

1

isnotofweight2).Wewill seebelowthat thewholeGrobnerbasisof I

` isnot

neessarytothepoint-ountingappliationwehaveinmind.Thus,weanstart

byworkingwiththersttwoequationsE

1

;E

2

,whihinvolveX

1

;X

2 only.

Thesepolynomialswerealreadyonsideredin[11℄.Thestrategyusedinthat

paperonsisted in omputing theresultantof E

1

;E

2

with respet to X

2 fora

start,fromwhihitwaspossibletodeduetheoordinatesof[`℄-torsiondivisors.

Thisapproahdidnottakeintoaountthesymmetryin(X

1

;X

2

);wenowshow

howto work diretlyin Mumford'soordinatesU

1

= (X

1 +X

2 );U

0

=X

1 X

2 ,

soastoomputeresultantsoflowerdegrees.

3.2 Resymmetrisation

The polynomialsE

1 (X

1

;X

2

)and E

2 (X

1

;X

2

) aresymmetri polynomials.It is

(7)

polynomialsX

1 X

2 andX

1 +X

2

.TheheartofMumford'srepresentationisthe

useofthisexpression,butthishadbeenbrokeninordertoapplyCantor'sdivi-

sionpolynomials.Weallresymmetrisationthemethodthat wepresentnowto

ome baktoarepresentationofbivariatepolynomialsin termsoftheelemen-

tary symmetri polynomials. This is notas trivialasit seems, sinethe naive

shoolbookmethodtosymmetrizeapolynomialyieldsaomplexityjumpinour

ase.

Let us onsider the unique polynomialsE

1 and E

2 in F

p [U

0

;U

1

suh that

E

1 (X

1 X

2

; X

1 X

2 ) = E

1 (X

1

;X

2

) and E

2 (X

1 X

2

; X

1 X

2 ) = E

2 (X

1

;X

2 )

andletR

1 2F

p [U

1

betheirresultantwithrespetto U

0

;thenR

1

dividesR

1 .

Wewanttousethefollowingevaluation/interpolationtehniquestoompute

R

1

:evaluatethevariableU

1

atsuÆientlymanysalarsu

1

, omputetheresul-

tants of E

1 (U

0

;u

1

) and E

2 (U

0

;u

1

), and interpolate theresults. Unfortunately,

omputing with E

1 and E

2

themselves has prohibitive ost, as these polyno-

mials haveO(`

4

) monomials.However,theirspei shapeyieldsthe following

workaround.

LethbeapolynomialinF

p

[X℄andX

1 andX

2

betwoindeterminates.Then

thedivideddierenesofharethebivariatesymmetripolynomials

A

0 (h)=

h(X

1 ) h(X

2 )

=(X

1 X

2

) and A

1 (h)=

X

1 h(X

2 ) X

2 h(X

1 )

=(X

1 X

2 ):

We let A

0

(h) and A

1

(h) be the unique polynomials in F

p [U

0

;U

1

suh that

A

0 (h)(X

1 X

2

; X

1 X

2 )=A

0

(h) andA

1 (h)(X

1 X

2

; X

1 X

2 )=A

1

(h). Then

adiretomputationshowsthat

E

1

=A

0 (d

1 )A

1 (d

2 ) A

0 (d

2 )A

1 (d

1 );

E

2

=A

0 (d

0 )A

1 (d

2 ) A

0 (d

2 )A

1 (d

0

) inF

p [U

0

;U

1

℄:

Given an arbitrary polynomial h in F

p

[X℄ and u

1 2 F

p

, we show in the last

paragraphshowtoomputethepolynomialsA

0

(h)andA

1

(h)evaluatedatU

1

=

u

1

eÆiently. Taking this operation for granted, we dedue Algorithm 2 for

omputingtheresultantR

1 ofE

1 andE

2 .

Algorithm2ComputationoftheresultantR

1

1. Fordeg (R1)+1dierentvaluesofu12Fp,do

(a) ComputeA

0 (d

0 ),A

1 (d

0 );A

0 (d

1 );A

1 (d

1 );A

0 (d

2 );A

1 (d

2

)evaluatedatU

1

=u

1 .

(b) DedueE1 andE2,evaluatedatU1=u1.

() ComputeR

1 (u

1

)astheresultantinU

0 ofE

1 andE

2 .

2. InterpolateR

1

fromthepairs(u

1

;R

1 (u

1 )).

Thelassial estimates for the degrees of resultantsimply that the degree

of R

1 is 6`

4

17`

2

+12; thus to be able to perform the interpolation, it is

neessarytotakeatleast6`

4

17`

2

+13dierentvaluesofu

1

.Inpratie,itis

reommendedtotakeafewmorevaluesofu

1

,inordertohektheomputation.

Notethat theresultantofE

1

;E

2

hasdegree8`

4

22`

2

+15.

We nish this subsetion by detailing our solution to the problem raised

above:givenu in F and hin F [X℄,howto ompute thepolynomialsA (h)

Références

Documents relatifs

Note however that the product theorems of [8, 9, 5, 19] are not used in the proof of Theorem 1, at least not directly; in their stead we use the assumed spectral gap of the

of disrete logarithms in the Jaobian of hyperellipti urves dened.. over

This leads us to an accurate form of the connection between the spectral radius of finite subgraphs of the geometric graph, connected components of these subgraphs and the number

Indeed, the number of additions performed by the addition-thread de- pends on the Hamming weight of the NAF representation of the scalar (i.e. the non-zero digits). As

In this section, we show how to construct irreducible polynomials using elliptic curves.. Let K be a field and let Ω be an algebraic closure

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

of class number one and for a rational prime p which splits or ramifies in k, we have isogenies of degree p which are represented by elliptic curves.. with

familiar modular curves, give a modular interpretation of their Fq-defined points, and use Weil’s estimate to approximate the number of their Fq-defined points...