20.11.2013 Views

Lattices and Their Application to Cryptography.pdf - BeKnowledge

Lattices and Their Application to Cryptography.pdf - BeKnowledge

Lattices and Their Application to Cryptography.pdf - BeKnowledge

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

<strong>Lattices</strong><strong>and</strong><strong>Their</strong><strong>Application</strong><strong>to</strong>Cryp<strong>to</strong>graphy LectureNotes:<br />

StanfordUniversity,SpringQuarter,1998 -Version<br />

IBMAlmadenResearchCenter CynthiaDwork June13,1998


Thesenotesareforthecourse<strong>Lattices</strong><strong>and</strong><strong>Their</strong><strong>Application</strong><strong>to</strong>Cryp<strong>to</strong>graphy,taughtatStanford duringtheSpringQuarter,1998.ThematerialonroundingnumbersistakenmostlyfromLovasz' undalgorithmischeGeometrie,ReduktionvonGitterbasenunPolynomidealen",deliveredatJohannWolfgangGoetheUniversity,Frankfurt/MainduringtheSummersemesterof1994<strong>and</strong>the<br />

Wintersemesterof1994/95,compiledbyRogerFischlin,whohasgraciouslysharedwithmehis LATEXles.Inparticular,thePreliminaries<strong>and</strong>Chapters2{6aretranslationsofthesenotes(with ThematerialonlatticetheoryfollowsthenotesfromClausSchnorr'sLectures\Gittertheorie monographAnAlgorithmicTheoryofNumbers,Graphs<strong>and</strong>Convexity[Lovasz86].<br />

somemodications).<br />

CynthiaDwork dwork@almaden.ibm.com


Contents<br />

Preface Preliminaries 5<br />

1OnRoundingNumbers 1.1LengthsofRationals...................................13 13 9<br />

2Complexity,NP-Completeness 1.2DiophantineApproximation<strong>and</strong>RelatedProblems..................14 2.1NP-Completeness....................................21 2.2HardAlgorithmicLatticeProblems...........................23 21<br />

3Introduction<strong>to</strong>LatticeTheory 3.2Fundamentals<strong>and</strong>Properties..............................28 3.1Terminology........................................27<br />

4Succ.Minima,Minkowski'sTheorems,HermiteConstant 3.3Length-<strong>and</strong>Weight-ReducedLatticeBases......................48 3.4Examples.........................................51<br />

4.2HermiteConstant<strong>and</strong>CriticalLattice.........................61 4.1SuccessiveMinima<strong>and</strong>theFirstMinkowskiTheorem................55<br />

5Gauss'BasisReductionProcedure 4.3GaugeFunctions<strong>and</strong>Minkowski'sTheorems.....................69 5.1ReducedBasis......................................73 5.2Algorithms........................................76<br />

3


46LLLReducedLatticeBasis 6.1Denition<strong>and</strong>Properties................................79 CONTENTS<br />

6.2TheLLLReductionAlgorithm.............................85<br />

7Babai'sApproximation<strong>to</strong>CVP 7.1ApproximateCVP....................................95 7.2InhomogeneousDiophantineApproximation......................98<br />

8BreakingtheLinearCongruentialGenera<strong>to</strong>r 8.2AGeneralReconstructionResult............................102 8.1LinearCongruentialSequences.............................101 101<br />

8.3<strong>Application</strong><strong>to</strong>theLinearCongruentialGenera<strong>to</strong>r...................106 Bibliography 8.4ModernDenitionsofPseudo-R<strong>and</strong>omness......................107 109


Preface<br />

Alatticeisaregulararrangemen<strong>to</strong>fpointsinspace.Inparticular,forlinearlyindependent abasisofthelattice.Aswewillsee,therearemanydierentbasesforanygivenlattice.The b1;b2;:::;bn2Rn,thelatticeL=L(b1;b2;:::;bn)isthese<strong>to</strong>fallintegerlinearcombinations a1b1++anbn,a1;a2;:::;an2Z,oftheelmentsb1;b2;:::;bn.Thevec<strong>to</strong>rsb1;b2;:::;bnform lengthofabasisisthelengthofthelongestbasisvec<strong>to</strong>r.<br />

ticular, Copper,CFJP,H,LaOd85,Shamir82].Wewillcoverseveralofthese\negative"results;inpar-<br />

Lenstra,<strong>and</strong>Lovasz,wereusedincryp<strong>to</strong>graphyprincipally<strong>to</strong>provecryp<strong>to</strong>graphicinsecurity[Adel83, Prior<strong>to</strong>1996,lattices,<strong>and</strong>inparticular,thelatticebasisreductionalgorithmofLenstra,<br />

1.breakingofknapsack-basedcryp<strong>to</strong>systems<br />

3.breakingthesupposedsemanticsecurityofpaddedRSA. 2.breakingthelinearcongruentialpseudo-r<strong>and</strong>omgenera<strong>to</strong>r<br />

N=pq,wherep<strong>and</strong>qarelargeprimes.Ifanadversarycanfac<strong>to</strong>rNthenthesystemisinsecure. reliesontheintractabilityofar<strong>and</strong>ominstanceoftheproblemonwhichtheconstructionisbased. Forexample,inthecaseoftheRSApublickeycryp<strong>to</strong>system,thepublickeycontainsamodulus akeyweredeterministic,thenthekeycouldnotbesecret.Thus,thesecurityoftheconstruction Cryp<strong>to</strong>graphicconstructionsnecessarilyrequirer<strong>and</strong>omchoices:if,forexample,thechoiceof<br />

particular,therelativedicultyofthehardestinsancesoffac<strong>to</strong>ring<strong>and</strong>r<strong>and</strong>ominstancesof fac<strong>to</strong>ringisnotknown.Indeed,eveniffac<strong>to</strong>ringwereNP-hard(itprobablyisnot),<strong>and</strong>evenif Pwereknown<strong>to</strong>bedierentfromNP,thiswouldsaynothingaboutthehardnessofr<strong>and</strong>om Whenauserpicksar<strong>and</strong>omN,itisnotenoughthatsometwo-primemoduliarehard<strong>to</strong>fac<strong>to</strong>r:<br />

instancesoffac<strong>to</strong>ring. theuserwantsthatar<strong>and</strong>ominstanceshouldbehard<strong>to</strong>fac<strong>to</strong>r.Nosuchresultisknown;in<br />

onecanestablishanexplicitconnectionbetweenthehardnessofr<strong>and</strong>ominstances<strong>and</strong>thehardness ofthehardest,orworst-case,instances.Suchaconnectionisthecontributionofthecelebrated thepaperpresentsar<strong>and</strong>omprobleminvolvingacertainclassofr<strong>and</strong>omlattices,whosesolution paperofAjtai,\GeneratingHardInstancesofLatticeProblems"[Ajtai96](1996).Specically, Ithasthereforebeenalongst<strong>and</strong>inggoalincryp<strong>to</strong>graphy<strong>to</strong>nda\hard"problemforwhich<br />

wouldimplythesolutionofthreefamousworst-caseproblems: 5


61.Findthelengthofashortestnonzerovec<strong>to</strong>rinann-dimensionallatticeapproximately,up<br />

<strong>to</strong>apolynomialfac<strong>to</strong>r. CONTENTS<br />

3.Findabasisb1;:::;bninthen-dimensionallatticeLwhoselength,denedasmaxni=1kbik, 2.Findtheshortestnonzerovec<strong>to</strong>rinann-dimensionallatticeLwheretheshortestvec<strong>to</strong>rv<br />

isthesmallestpossibleup<strong>to</strong>apolynomialfac<strong>to</strong>r. wherecisasucientlylargeabsoluteconstant. isuniqueinthesensethananyothervec<strong>to</strong>rwhoselengthisatmostnckvkisparallel<strong>to</strong>v,<br />

lattices<strong>to</strong>cryp<strong>to</strong>graphy: problems.Thiseorthasbeenfruitful.Wecoveratleastthefollowing\positive"applicationsof tionofcryp<strong>to</strong>graphicprimitivesontheassumedhardnessofsolvingtheabove-mentionedlattice MotivatedbyAjtai's1996paper,peoplebegan<strong>to</strong>explorethepossibilitybasingtheconstruc-<br />

1.Ajtai'sproofshowsthatthatcertaincryp<strong>to</strong>graphichashfunctionsenjoyworst-case/averagecaseequivalence.<br />

3.Theconstructionofthecryp<strong>to</strong>systemyieldsanaturalpseudo-r<strong>and</strong>omgenera<strong>to</strong>rwithworstcase/average-caseequivalence.<br />

2.Thereexistsapublickeycryp<strong>to</strong>systemwithworst-case/average-caseequivalence.<br />

RoughOutlineoftheCourse<br />

ity[Lovasz86]. inChapter1ofLovasz'monographAnAlgorithmicTheoryofNumbers,Graphs<strong>and</strong>Convex-<br />

Wewillbeginwithsomeclassicalmotivationfromthegeometryofnumbers,basedonthematerial<br />

1994<strong>and</strong>theWintersemesterof1994/95.Inparticular,Ihavetranslatedfromthesenotesthe deliveredatJohannWolfgangGoetheUniversity,Frankfurt/MainduringtheSummersemesterof gorithm,followingtheRogerFischlin'scompilationofthenotesfromClausSchnorr'sLectures \GittertheorieundalgorithmischeGeometrie,ReduktionvonGitterbasenundPolynomidealen", Wethendiscussthefundamentalsoflatticetheory,<strong>and</strong>theLLLlatticebasisreductional-<br />

followingmaterial.<br />

2.SuccessiveMinima<strong>and</strong>TwoTheoremsofMinkowski 1.Introduction<strong>to</strong>LatticeTheory:terminology,basicpropertiesofalattice,lengthreduction,<br />

3.Gauss'BasisReductionProcedure(for2-dimensionallattices) weightreduction;<br />

onproposedcryp<strong>to</strong>graphicprimitives.Followingourstudyoflatticetheory,wewillcoversome oftheseattacks. 4.LLLLatticeBasisReduction Asmentionedearlier,theLLLlatticebasisreductionalgorithmisatthehear<strong>to</strong>fseveralattacks


CONTENTS withthelastdayreservedforoneortwooutst<strong>and</strong>ingnewerresultsrelated<strong>to</strong>thematerialofthe Thelastpar<strong>to</strong>fthecoursewilldiscussthepositiveapplicationsoflattices<strong>to</strong>cryp<strong>to</strong>graphy, 7<br />

course.


8 CONTENTS


Preliminaries<br />

Notation WeletMm;n(S)denotethese<strong>to</strong>fallmnmatriceswithentriesfromthesetS.Forexample, Mm;n(Z)isthese<strong>to</strong>fallintegermnmatrices.WemayalsousethealternatenotationSmn.<br />

fx2Rjx>0gforthese<strong>to</strong>fpositiverealnumbers. Zn,Rn,etc.,denotecolumnvec<strong>to</strong>rs. Forrealnumbersrweletdrc:=r12denotetheintegerclosest<strong>to</strong>r.WewriteR+= FormatrixBweletBTdenotethetransposeofB.Unlessotherwiseindicated,elementsof<br />

ScalarProduct Ascalarproducth;i:RnRn!Risamappingwithfollowingproperties:Forallu;v;w2Rn <strong>and</strong>2R: h;iisbilinear:<br />

hu;v+wi=hu;vi+hu;wi hu+w;vi=hu;vi+hw;vi hu;vi=hu;vi hu;vi=hu;vi<br />

h;iissymmetric:<br />

h;iispositivedenite: hu;vi=hv;ui<br />

hu;ui>0 9 foru6=0


10 Foru=0itfollowsfromthelinearityineachcomponentthathu;ui=0.Thest<strong>and</strong>ardscalar productisdenedas:D(u1;u2;:::;un)T;(v1;v2;:::;vn)TE:=nXi=1uivi canbewrittenas: Mostapplicationsuseonthest<strong>and</strong>ardscalarproduct.Everyscalarproducth;i:RnRn!R<br />

productthematrixSistheidentitymatrix. onlyifSissymmetric<strong>and</strong>xTSx>0forallx2Rnnf0g.)Inthecaseofthest<strong>and</strong>ardscalar forsomeapositivedenitematrixS2Mn;n(R).(AnnnmatrixSispositivedeniteif<strong>and</strong> hu;vi:=uTSv<br />

Norms Amappingkk:Rn!Riscalledanorm,ifforallu;v2Rn<strong>and</strong>2R:<br />

Therealnumberkukiscalledthenorm(orlength)ofthevec<strong>to</strong>ru=(u1;u2;:::;un).Forevery ku+vkkuk+kvk kvk=jjkvk kuk>0 foru6=0 (positivehomogeneous)<br />

scalarproductweobtainacorrespondingEuclideannormasfollows: (triangleinequality)<br />

kuk:=phu;ui(positivedeniteness)<br />

The`1normis:<br />

The`2normisobtainedfromthest<strong>and</strong>ardscalarproduct: (u1;u2;:::;un)T1:=nXi=1juij<br />

Ingeneral,the`pNormis: (u1;u2;:::;un)T2:=phu;ui= nXi=1u2i!12<br />

Thesupnorm,maximumnormor`1normis: (u 1;u2;:::;un)Tp:= nXi=1juijp!1p<br />

(u1;u2;:::;un)T1:=max i=1;2;:::;njuij


Inequalities 11<br />

Forthesup-,`1-<strong>and</strong>`2-norm,ofavec<strong>to</strong>ru2Rnwehave: kuk1kuk2pnkuk1 kuk1kuk1nkuk1 kuk2kuk1pnkuk2<br />

ity(foru;v2Rn): Everyscalarproduct<strong>and</strong>correspondingnormkuk:=phu;uisatisfytheCauchy-SchwarzInequal-<br />

Equalityholdsonlywhenthetwovec<strong>to</strong>rsarelinearlyindependent. Letb1;b2;:::;bn2Rnbethecolumnvec<strong>to</strong>rs(orrowvec<strong>to</strong>rs)ofthematrixB2Mn;n(R). jhu;vijkukkvk<br />

Hadamard'sInequalitysays:<br />

Equalityholdswhenthevec<strong>to</strong>rsb1;b2;:::;bnareorthogonal. jdetBjnYi=1kbik2


Chapter1<br />

OnRoundingNumbers<br />

ThematerialherefollowsSections1.0<strong>and</strong>1.1of[Lovasz86]<strong>and</strong>Chapter5of[GLLS88].We discusstwonaturalhardware-independentmodelsofnumbers<strong>and</strong>ase<strong>to</strong>fproblemsinvolvingthe<br />

1.1 rounding,orapproximation,ofrationals.<br />

Wedenethebinaryencodinglengthofniteobjectsasfollows: LengthsofRationals<br />

`(0):=1<br />

`(A)=Pi;j`(aij)forA=[aij]2Mm;n(Q) `(n):=1+dlog2(jnj+1)eforn2Z `pq:=`(p)+`(q)wherep;q2Z,q6=0,<strong>and</strong>gcd(p;q)=1<br />

Lemma1.1.1 2.Foreveryvec<strong>to</strong>rx2Qn,1+kxk1+kxk12`(x)n 1.Foreveryrationalnumberr,1+jrj2`(r)1<br />

allu2Rnwehavekuk2kuk1,inparticularkxk2kxk1.Thenfrom1.wehave: Proof.1.followsdirectlyfromthedenition.Toprove2.,letx=(x1;x2;:::;xn)T.Sincefor 3.ForeverymatrixD2Qnn,jdetDj2`(D)n21<br />

1+kxk1=1+nXi=1jxijnYi=1(1+jxij)nYi=12`(xi)1=2`(x)n: 13


14Letd1;d2;:::;dnbetherowsofD.ByHadamard'sInequality<strong>and</strong>2.wehave:<br />

CHAPTER1.ONROUNDINGNUMBERS<br />

1+jdetDj1+nYi=1kdik2nYi=1(1+kdik2)nYi=12`(di)n=2`(D)n2<br />

contributesone<strong>to</strong>thelengthoftheinput.Thebinaryencodingallowstheinputs<strong>to</strong>\appear" Incontrast<strong>to</strong>thebinaryencoding,wecouldusethearithmeticencoding,inwhicheachinteger <br />

arithmeticencodingthelengthsoftheargumentsdonotgrowduringexecutionofthealgorithm. thearithmeticencodinglength(thatis,itdoesnotrunintimepolynomialin2).Conversely,inthe commondivisoroftwointegersrunsintimepolynomialinthebinaryencodinglength,butnotin functionoftheinputlength).Forexample,theEuclideanalgorithmforcomputingthegreatest longerthantheydointhearithmeticencoding,makingalgorithmsappear<strong>to</strong>runfaster(asa<br />

polynomial.Everystronglypolynomialalgorithmispolynomial. takespolynomialtimeinthearithmeticsensewhilethelengthofthebinaryencodingremains butthelengthoftheargumentgrowsexponentially.Analgorithmisstronglypolynomialifit Forexample,nrepeatedsquaringsofannbitnumberonlyrequiresnarithmeticoperations<br />

1.2 Incomputationsinvolvingrealnumberswereplacethenumbersbyrationalapproximations.For example,ifwexanintegerqthenthebestchoiceofpsuchthatpqapproximatesisp=bqc DiophantineApproximation<strong>and</strong>RelatedProblems<br />

orp=dqe.Theresultingerrorisboundedby<br />

Supposeweallowthedenomina<strong>to</strong>rq<strong>to</strong>vary,subject<strong>to</strong>theconstraintqQ.Wehave: pq12q<br />

Proposition1.2.1(Dirichelet) Forall2R<strong>and</strong>integerQ1,thereexistp;q2Z,0


1.2.DIOPHANTINEAPPROXIMATIONANDRELATEDPROBLEMS p=jidisaninteger.Thenforq=jiwehavejqpj=d1 Q+10,arepositive:<br />

)1< )0ixi


16 wherek>0isaninteger<strong>and</strong>(ri;qi)=1.Thenbydenition CHAPTER1.ONROUNDINGNUMBERS<br />

Thus,pi+qi>pi+1+qi+1>0,soeventuallytheprocedureterminates. i+1= ixi=qi 1 ri=pi+1<br />

Toseehowquicklyitterminates,denetwoauxiliarysequences,gk<strong>and</strong>hk,asfollows. qi+1<br />

gi=xigi1+gi2,fori=0;1;::: hi=xihi1+hi2,fori=0;1;::: g2=0,g1=1,h2=1,h1=0<br />

Notethat1=h0h1


1.2.DIOPHANTINEAPPROXIMATIONANDRELATEDPROBLEMS Definition1.2.3(BestApproximationProblem) Given:2Q<strong>and</strong>integerQ>0 17<br />

Theproblemissometimesgiveninanequivalentbut\reverse"form: Find:rationalp=qsuchthat00<br />

Form: Toproveequivalence,wewillprovethatbothformsareequivalent<strong>to</strong>thefollowingGeneral Findintegersp;qsuchthatq>0,pq


183711=0:6363:::<br />

CHAPTER1.ONROUNDINGNUMBERS<br />

be<strong>to</strong>ndaroundingprocedureresultinginapproximationsp1 While1+2+3=1,thisisnottrueofthesumoftherespectiveapproximations.Thetrickwill<br />

Definition1.2.6(SimultaneousDiophantineApproximationProblem) denomina<strong>to</strong>r,q. q;p2 q;:::;pnqwithasinglecommon<br />

Given1;2;:::;n2Q,">02Q,<strong>and</strong>Q2Z,Q>0,ndintegersp1;p2;:::;pn<strong>and</strong>q suchthat0


1.2.DIOPHANTINEAPPROXIMATIONANDRELATEDPROBLEMS Thus 19<br />

foranychoiceofp1<strong>and</strong>q. Kroneneckergaveageneralconditionforthesolvabilityofthisproblem(see[Cassels71]): j(q1p1)1j16<br />

Proposition1.2.9 Forany2nrealnumbers1;2;:::;n;1;2;:::;n,either 1.Foreach">0thereexistintegersp1;p2;:::;pn;qsuchthatq>0<strong>and</strong> 2.Thereexistintegersu1;u2;:::;unsuchthatPi=1nuiiisanintegerwhilePi=1nuiiis not. jqipiij":<br />

Wewillreturn<strong>to</strong>thisproblem,<strong>and</strong>inparticularthespecialcaseinwhichthei<strong>and</strong>iare rationals,afterweseetheLLLlatticebasisreductionalgorithm.


20 CHAPTER1.ONROUNDINGNUMBERS


Chapter2<br />

Complexity,NP-Completeness<br />

Wereviewthebasicconceptsofcomplexitytheoryrelating<strong>to</strong>latticetheory,especiallyNPcompleteness.<br />

2.1 Recallthatwehavedenedthelengthofthebinaryencodinglengthofniteobjectsasfollows: NP-Completeness<br />

`(0):=1<br />

`(A)=Pi;j`(aij)forA=[aij]2Mm;n(Q) `(n):=1+dlog2(jnj+1)eforn2Z<br />

Wedenetherunningtimeofanalgorithmasafunctionofthelengthsoftheinputs.Weare `pq:=`(p)+`(q)wherep;q2Z,q6=0,<strong>and</strong>gcd(p;q)=1<br />

Definition2.1.1(PolynomialTime) interestedinpolynomialtime:<br />

operations)ispolynomiallyboundedinthelengthoftheinputs: Analgorithmrunsinpolynomialtimeifthenumberofsteps(Turingmachineornumberofbit<br />

Intheoreticalcomputersciencepolynomialtimealgorithmsaresometimesreferred<strong>to</strong>asecient. NumberofSteps(Inputs)=poly(`(Inputs))<br />

<strong>and</strong>onlyifa2A. Definition2.1.2(CharacteristicFunction) ForasetAf0;1gthecharacteristicfunctionA:f0;1g!f0;1gisdenedas:A(a)=1if 21


Usingcharacteristicfunctions,wedenetheclassofpolynomialtimelanuages: 22 CHAPTER2.COMPLEXITY,NP-COMPLETENESS<br />

Definition2.1.3(ClassPofPolynomialTimeLanguages)<br />

TheclassNPcontainsthelanguagesforwhich,foreachwordinthelanguagethereisashort, characteristicfunctionAispolynomialtimecomputable. TheclassPofpolynomialtimelanguagesisthese<strong>to</strong>flanguagesAf0;1gforwhichthe<br />

ecientlyveriableproofofmembershipinthelanguage. Definition2.1.4(ClassNP) TheclassNPofnondeterministicpolynomialtimelanguagesAf0;1gisdenedas:<br />

Let(x;y)2B.Thenyiscalledawitnessforx2A. A2NP() 9B2f0;1gf0;1g;B2P:<br />

Cook'sthesisisthatP6=NP,thatis,thereexistsalanguageinNPthatisnotrecognizablein A=x2f0;1g9y2f0;1gpoly(`(x))with(x;y)2B<br />

deterministicpolynomialtime. Definition2.1.5(KarpReduction) LetA;Bf0;1g:<br />

IfApolB<strong>and</strong>BpolCthenApolC.ApolBthenitispossibleinpolynomialtime<strong>to</strong> ApolB() 9polynomialtimetransformationhwith:<br />

decideifx2Awithasinglequery<strong>to</strong>anoracleforB. 8x2f0;1g:x2A,h(x)2B<br />

theoracleforB,providedthatthe<strong>to</strong>talcomputationtime,includingthesettingupofthecalls <strong>to</strong>theoracle(eachoraclecallitselfhasunitcost)<strong>and</strong>anysubsequentanalysis,ispolynomially bounded. Amoregeneraltypeofpolynomialtimereduction(aCookreduction),allowsmultiplecalls<strong>to</strong><br />

Definition2.1.6(NP-Complete) timemachineMthat,usinganoracleforB,c<strong>and</strong>ecidemembershipinA. Wewillalsobeinterestedinr<strong>and</strong>omizedreductions,inwhichthereisaprobabilisticpolynomial<br />

Af0;1gissaid<strong>to</strong>beNP-complete,if:<br />

IfthereisapolynomialtimealgorithmforanyNP-completeproblem,thenP=NP.Thiswould 2.8B2NP:BpolA 1.A2NP<br />

contradictCook'sThesis.Hence,theNP-completeproblemsarethehardestproblemsinNP.


2.2.HARDALGORITHMICLATTICEPROBLEMS HardAlgorithmicLatticeProblems 23<br />

Programming): Inthissectionwementionproblemsrelated<strong>to</strong>latticetheorythatareeitherNP-completeorfor<br />

Definition2.2.1(IntegerLinearProgramming) whichnoecientalgorithmisknown.Onesuchproblemisintegerlinearprogramming(Integer<br />

Theproblemofinteger,linearprogrammingis: Given:m;n2N,A2Mm;n(Z)<strong>and</strong>b2Zm<br />

Integerlinerarprogrammingis\hard".WeshowinProposition2.2.5thatthecorresponding decisionproblemisNP-complete: Findx2ZnwithAxborshowthatnosuchvec<strong>to</strong>rexists.<br />

Definition2.2.2(DecisionProblemforIntegerLinearProgramming) Thedecisionproblemforintegerlinearprogrammingis: Given:m;n2N,A2Mm;n(Z)<strong>and</strong>b2Zm<br />

polynomialtimealgorithmfortheanalogousproblemofrationallinearprogramming: IfP6=NP,thennopolynomialtimealgorithmsolvesthisproblem.Incontrast,thereisa Decideifthereexistsx2ZnwithAxb.<br />

Theproblemofrationallinearprogrammingis: Definition2.2.3(RationalLinearProgramming) Given:m;n2N,A2Mm;n(Z)<strong>and</strong>b2Qm<br />

Therstpolynomialtimealgorithmforrationallinearprogrammingistheellipsoidmethodof L.G.Khachiyan[Khach79,Khach80].Thismethodis,however,impractical.Aprovablypolynomialtimealgorithmthatalsoappears<strong>to</strong>bepracticalwasdevelopedbyM.Karmarkar[Karma84],<br />

whosestartingpointwastheclassicalinteriorpointmethod.Anotherpolynomialtimealgorithmis<br />

Findx2QnwithAxborshowthatnosuchvec<strong>to</strong>rexists.<br />

due<strong>to</strong>Y.Ye[Ye91].Asimple,practicalprocedureisthesimplexalgorithm[Dantzig63,Schrijver86] ofG.B.Dantzig,whichhasexponentialrunningtimeintheworstcase.Thefollowingproblem<br />

Givenm;n2N,A2Mm;n(Z)<strong>and</strong>b2Zm,inpolynomialtimeonecan: Proposition2.2.4(Sieveking1976) canbesolvedinpolynomialtime:


24a)SolveAx=b,x2Znorshowthatnosolutionexists.<br />

b)FindaZ-Basisb1;b2;:::;bkforfx2ZnjAx=0g,theZ-kernel.AZ-Basisisase<strong>to</strong>f CHAPTER2.COMPLEXITY,NP-COMPLETENESS<br />

linearlyindependentvec<strong>to</strong>rsb1;b2;:::;bk,where: fx2ZnjAx=0g=(kXi=1tibit1;t2;:::;tk2Z)<br />

Proposition2.2.5 [KaBa79]. Proof.ModicationofGaussianelimination(M.Sievekingin[SpStr76]).Alternateproofin<br />

ThefollowinglanguagesareNP-complete: <br />

1.Integer-Programming:IP:=(m;n;A;b)A2Mm;n(Z);b2Zm; 2.KnapsackorSubsetSum: 9x2Zn:Axb <br />

SubsetSum:=((n;a1;a2;:::;an;b)2Nn+29x2f0;1gn:nXi=1aixi=b)<br />

3.f0;1g-Integer-Programming:<br />

4.WeakDependence: f0;1g-IP:=(m;n;A;b)A2Mn;m(Z);b2Zm; 9x2f0;1gn:Axb<br />

((n;a1;a2;:::;an)2Nn+19(x1;x2;:::;xn)2f0;1gnnf0ng:nXi=1aixi=0)<br />

Proof.For1,2,3see[GaJo79,SpStr76],for4see[EmBoas81].InProposition2.2.6weprove thatintegerprogramminghasapolynomiallengthwitness<strong>and</strong>henceIP2NP. Proposition2.2.6(vonzurGathen,Sieveking1978) IP2NP. <br />

ifsuchanxexists,then(m;n;A;b)2IP.Weneedonlyshowthatthewitnesshaspolynomial Proof.Thewitnessfor(m;n;A;b)2IPwillbeasuitablex2ZnwithAxb.Obviously, length.LetA=:(aij)<strong>and</strong>b=:(b1;b2;:::;bm)T.SetM:=maxi;jfjaijj;jbijg.VonzurGathen <strong>and</strong>Sieveking[GaSi78]prove: (9x2Zn:Axb)()9x2Zn:Axb;kxk1(n+1)nn2Mn


thelengthofAundb.Since`(m;n;A;b)nm+log2Mwehave: 2.2.HARDALGORITHMICLATTICEPROBLEMS Theupperboundonkxk1impliesthatthelengthofthewitnessxispolynomiallyboundedin 25<br />

`(x)=On2(logn+logM=O`(m;n;A;b)3<br />

Wemakethefollowingdenitions,whichwillbeneededinthenextChapter. <br />

Definition2.2.7(Lattice,Basis,Dimension,Rank) Letb1;b2;:::;bn2Rmbelinearlyindependentvec<strong>to</strong>rs.Wecalltheadditivesubgroup<br />

ofRmalatticewithbasisb1;b2;:::;bn.Whenthesequenceofthebasisvec<strong>to</strong>rsisxed,speakof L(b1;b2;:::;bn):=nXi=1biZ=(nXi=1tibit1;t2;:::;tm2Z)<br />

anorderedbasis.Therankorthedimensionofthelatticeisrank(L):=n. Weconsideranexample:<br />

A2Mm;n(Z)thesetfx2ZnjAx=0gisalatticeofranknrank(A);byProposition2.2.4we Example2.2.8(Lattice) canconstructabasisinpolynomialtime. Zmisalatticeofrankm,thest<strong>and</strong>ardunitvec<strong>to</strong>rse1;:::;emformabasis.Giventhematrix<br />

supnormthiscannotbedoneecientlyundertheassumptionthatP6=NP: Throughlatticereductionwesearchforashortest,non-trivial,latticevec<strong>to</strong>r.Inthecaseofthe <br />

Corollary2.2.9 Theproblemofndingtheshortestlatticevec<strong>to</strong>rwithrespect<strong>to</strong>thekk1norm:<br />

isNP-complete. L1-SVP:=(m;n;b1;b2;:::;bn)m;n2N;b1;b2;:::;bn2Zm; 9x2L(b1;b2;:::;bn):kxk1=1<br />

bershipisavec<strong>to</strong>rx2L(b1;b2;:::;bn)nf0gwithkxk1=1.TheNP-completeproblem\weak Proof.Theproblemofndingthekk1-shortestlatticevec<strong>to</strong>risinNP:Thewitnessofmem-<br />

dependence"fromProposition2.2.5canbereducedinpolynomialtime<strong>to</strong>thekk1-shortestlattice<br />

k,decidewhetherthereisalatticevec<strong>to</strong>rz2L(b1;b2;:::;bn)withz6=0<strong>and</strong>kzk2pk. Theproblemoftheshortestlatticevec<strong>to</strong>rinthe`2normis,givenalatticebasisb1;b2;:::;bn<strong>and</strong> vec<strong>to</strong>r.


Definition2.2.10(ShortestVec<strong>to</strong>rProblemSVP) 26 Thelanguageoftheshortestlatticevec<strong>to</strong>rsinthe`2-normis: CHAPTER2.COMPLEXITY,NP-COMPLETENESS<br />

L2-SVP:=(k;m;n;b1;b2;:::;bn)k;m;n;2N;b1;b2;:::;bn2Zm; 9x2L(b1;b2;:::;bn)nf0g:kxk2k<br />

Thestatusofthisproblemisopen.Eorts<strong>to</strong>showthatL2-SVPisNP-hard,incontrast<strong>to</strong>the sup-norm-SVP(seeCorollary2.2.9)havefailed(see[Kannan87]).However,Ajtai[Ajtai98]has shownthatthisproblemisNP-hardwithrespect<strong>to</strong>r<strong>and</strong>omizedreductions.<br />

Proposition2.2.11(ClosestVec<strong>to</strong>rProblemCVP) theclosestlatticevec<strong>to</strong>r,which,however,isknown<strong>to</strong>beNP-completeforanynorm: Theproblemoftheshortestlatticevec<strong>to</strong>risthehomogeneousspecialcaseoftheproblemof<br />

Theproblemofthe`2-closestlatticevec<strong>to</strong>r<br />

isNP-complete. L2-CVP:=(k;m;n;b1;b2;:::;bn;z)k;m;n;2N;b1;b2;:::;bn;z2Zm; 9x2L(b1;b2;:::;bn):kzxk2k<br />

avec<strong>to</strong>rinthelatticethatisnearestwithinafac<strong>to</strong>rofcn,foraxedconstantc. Proof.See[Kannan87,Theorem6.2]. Laterwewillseeanapproximatesolutiondue<strong>to</strong>Babai,basedontheLLLalgorithm,yielding <br />

hardalgorithmiclatticeproblems: Findashortnon-triviallatticevec<strong>to</strong>r. Summarizing:givenalatticebasisb1;b2;:::;bn2Zm,thefollowingtasksarethought<strong>to</strong>be<br />

Incontrast,givenasystemofgenera<strong>to</strong>rsb1;b2;:::;bn2ZmforalatticeL,nrank(L),itis Findabasiscomprisedofshortlatticevec<strong>to</strong>rs.<br />

possible<strong>to</strong>constructabasisforLinpolynomialtime. Findforagivenz2span(b1;b2;:::;bn)theclosestlatticevec<strong>to</strong>r.


Chapter3<br />

Introduction<strong>to</strong>LatticeTheory<br />

integermatrixisunique.Wecharacterizelatticesasdiscrete,addititve,subgroupsofRm.We discussthese<strong>to</strong>fallbasesofalattice,primitivesystemsoflatticevec<strong>to</strong>rs,thelatticedeterminant, <strong>and</strong>theGram-Schmid<strong>to</strong>rthogonalizationofalattice.Wedenelengthreduction<strong>and</strong>weight WedenetheHermitenormalformofamatrix<strong>and</strong>showthattheHermitenormalformofan<br />

bases. reductionofalatticebasis,<strong>and</strong>showthateverylatticehasalengthreduced<strong>and</strong>aweightreduced<br />

phx;xiiscalledthelengthofthevec<strong>to</strong>rx.Forlinearlyindependentvec<strong>to</strong>rsb1;b2;:::;bn2Rm Leth;i:RmRm!Rbeanarbitraryscalarproduc<strong>to</strong>nthevec<strong>to</strong>rspaceRm.Thenkxk= 3.1 Terminology<br />

welet<br />

denotethelatticewithbasisb1;b2;:::;bn.Forarbitraryvec<strong>to</strong>rsb1;b2;:::;bnlet L(b1;b2;:::;bn):=nXi=1biZ<br />

bethespacespannedbyb1;b2;:::;bn<strong>and</strong>let span(b1;b2;:::;bn):=nXi=1biR<br />

denotetheorthogonalcomplemen<strong>to</strong>fthisspaceinRm. Anintegermatrixwithdeterminant1issaid<strong>to</strong>beunimodular.These<strong>to</strong>fallunimodular span(b1;b2;:::;bn)?:=fy2Rmjhy;bii=0fori=1;2;:::;ng<br />

matricesisdenotedGLn(Z): 27


Definition3.1.1(GLn(Z)) 28 GLn(Z)isthegroupofintegernnmatriceswithdeterminant1: CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

S;T2GLn(Z)thendet(ST)=detSdetT,weseethattheproductSTisaunimodularmatrix. WearguethatGLn(Z)isagroup.Theidentitymatrixisunimodular<strong>and</strong>fromthatfactthatif GLn(Z):=fA2Mn;n(Z)jdetA=1g<br />

LetT2GLn(Z).From<br />

wehavethatdetT1=1,<strong>and</strong>byCramer'srulethe(i;j)entryinthematrixT1is: detT1=1<br />

(1)i+jdetTijdetT<br />

followsthatdetTij2Z.Thus,forT2GLn(Z)wehaveT12GLn(Z). whereTijdenotesTwiththeithrow<strong>and</strong>j-thcolumndeleted.SinceTijisanintegermatrix,it<br />

=detTij;<br />

withanappropriatelychosenunimodularmatrix: Thefollowingelementarycolumnoperationscanbeperformedonamatrixbyright-multiplication Exchangetwocolumns Mulitiplicationofacolumnby1<br />

operations. multiplicationbyaunimodularmatrixcorresponds<strong>to</strong>carryingoutase<strong>to</strong>felementarycolumn Itcanbeshownthateveryunimodularmatrixistheproduc<strong>to</strong>fthesethreematrixtypes.So Additionofanintegermultipleofonecolumn<strong>to</strong>another<br />

3.2 Inthissectionwedenethefundamentalsoflatticetheory<strong>and</strong>showsomeelmentaryproperties. Fundamentals<strong>and</strong>Properties<br />

Definition3.2.1(DiscreteSet) 3.2.1Discrete,AdditiveSubgroupsofRm<strong>and</strong><strong>Lattices</strong><br />

Wehave: AsetSRmiscalleddiscrete,whenShasnolimitpointinRm.<br />

Lemma3.2.2 LetGRmbeanadditivegroup.Thenthefollowingstatementsareequivalent:


3.2.FUNDAMENTALSANDPROPERTIES a)Gisdiscrete. b)0isisnotalimitpoin<strong>to</strong>fG. 29<br />

d)inffkxyk:x6=y;x;y2Gg>0;thatis,thereisapositive,real,suchthat8x;y2G, c)fx2G:kxk0.<br />

Proposition3.2.3 LRmisalatticeif<strong>and</strong>onlyifLisadiscrete,additive,subgroupofRm. kxyk.<br />

Proof.Weshowbothdirections: independent,isdiscrete.Let':Rn!span(L)thelinearmapping \)"WemustshowthateverylatticeL:=L(b1;b2;:::;bn)Rm,whereb1;b2;:::;bnarelinearly<br />

Znisdiscrete<strong>and</strong>'1iscontinuousonspan(L),itfollowsthatLisdiscrete. 'isanisomorphismwith'(Zn)=L.Thus,intuitively,'<strong>and</strong>'1preservelocalstructure.Since '(t1;t2;:::;tn)=nXi=1tibi<br />

\("LetLRmbeadiscrete,additive,subgroup.Letnbethemaximumnumberoflinearly rankn.NotethatthisimpliesthattherankofalatticeListhemaximalnumberoflinearly independentvec<strong>to</strong>rsinL.Thennm.ByinductiononnwewillshowthatLisalatticeof<br />

n=1<br />

n>1 ofL).Thenitisnothard<strong>to</strong>verifythatL(b)=L. Letb2Lbeashortestvec<strong>to</strong>rwithb6=0(suchavec<strong>to</strong>rexists,since0isnotalimitpoint<br />

Chooseb12Lnf0gwith1kb162Lforallk2.Then Theorthogonalprojection:Rm!span(b1)?isdenedby: (3.1)<br />

(b)=bhb1;bi L(b1)=L\span(b1)<br />

Theinductivestepfollowsfromthefollowingassertions: 1.(L)isdiscrete<strong>and</strong>isalatticeofrankn1. hb1;b1ib1<br />

2.Foreverybasis(b2);(b3);:::;(bn)for(L)withb2;b3;:::;bn2Lwehave: L=L(b1;b2;:::;bn)


30vec<strong>to</strong>rs.Thesevec<strong>to</strong>rsarenecessarilytheimagesofvec<strong>to</strong>rsinL,say,b2;:::;bn.Solet<br />

Therstassertionsaysthatthereexistsabasisfor(L)ofn1linearlyindependent CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

(b2);:::;(bn)beabasisfor(L).Thesecondassertionsaysthatforeverysuchbasisfor (L),ifweaddb1<strong>to</strong>these<strong>to</strong>fpre-images,thenL=L(b1;b2;:::;bn)<strong>and</strong>sobydenition isalattice.Proofofthetwoassertions: 1.Weshowthat0isnotalimitpoin<strong>to</strong>f(L).Forthesakeofcontradiction,assumethat arepairwisedistinct<strong>and</strong>lim 0isalimitpoin<strong>to</strong>f(L).Lety(i)i2NbeasequenceinL,sothatthevec<strong>to</strong>rsy(i) i!1y(i)=0.Forthesevec<strong>to</strong>rs<br />

wecomputey(i)denedby:y(i):=y(i)&y(i);b1 y(i)=y(i)y(i);b1<br />

hb1;b1i% hb1;b1ib1<br />

| integer {z } b1 Theny(i)2Ly(i)=y(i),<strong>and</strong>:<br />

Sincelim i!1y(i)=0,thereareinnitelymanyvec<strong>to</strong>rsy(i)2Lwith: y(i)y(i)12kb1k<br />

ThiscontradictsthefactthatLisdiscrete,<strong>and</strong>hencetheassumptionthaty(i)are pairwisedistinct<strong>and</strong>y(i)=y(i).Thus0isnotalimitpoin<strong>to</strong>f(L),<strong>and</strong>by y(i)kb1k<br />

2.Let(b2);(b3);:::;(bn)beabasisfor(L)withb2;b3;:::;bn2L.Wewillshow in(L)isn1.Bytheinductivehypothesis(L)isalatticeofrankn1. thatLL(b1;b2;:::;bn).Letb2L.From Lemma3.2.2(L)isdiscrete.Themaximumnumberoflinearlyindependentvec<strong>to</strong>rs<br />

thereisab2L(b2;b3;:::;bn)with(b)=b.Wehavebb2span(b1).Bythe choiceofb1<strong>and</strong>from(b)2(L)=L((b2);(b3);:::;(bn))<br />

itfollowsthatbb2L(b1).Thus,b2L(b1;b2;:::;bn). bb2(L\span(b1))(3.1) =L(b1)


3.2.FUNDAMENTALSANDPROPERTIES pendentlatticevec<strong>to</strong>rs.ItfollowsfromthenextPropositionthateverylatticehasmanydierent Theaboveproofalsoshowsthattherankofalatticeisthemaximumnumberoflinearlyinde-<br />

31<br />

Proposition3.2.4 bases.Let[b1;b2;:::;bn]bethematrixwithcolumnvec<strong>to</strong>rsb1;b2;:::;bn.RecallthatGLn(Z)is<br />

Thevec<strong>to</strong>rsb1;b2;:::;bn2RmformabasisofthelatticeL(b1;b2;:::;bn)if<strong>and</strong>onlyifthere thegroupofintegernnmatriceswithdeterminant1.<br />

existsamatrixT2GLn(Z)suchthat: Proof.Weprovebothdirections: b1;b2;:::;bn=[b1;b2;:::;bn]T<br />

\)"Sinceb1;b2;:::;bn2L(b1;b2;:::;bn)thereisaT2Mn;n(Z)with: Sinceb1;b2;:::;bnarelinearlyindependent,wehavedetT6=0.Itfollowsthat b1;b2;:::;bnT1=[b1;b2;:::;bn]<br />

b1;b2;:::;bn=[b1;b2;:::;bn]T<br />

\("LetussupposeforsomeT2GLn(Z)that Sincebi2Lb1;b2;:::;bnfori=1;2;:::;n,T1hasintegerentries.SincedetT detT1=1<strong>and</strong>detT,detT1arebothinteger,itfollowsthatjdetTj=1.<br />

Itfollowsthatb1;b2;:::;bn2L(b1;b2;:::;bn).Similarly,itfollowsfrom b1;b2;:::;bnT1=[b1;b2;:::;bn];<br />

b1;b2;:::;bn=[b1;b2;:::;bn]T<br />

thatb1;b2;:::;bn2Lb1;b2;:::;bn.Thus, Lb1;b2;:::;bn=L(b1;b2;:::;bn)<br />

byaunimodularmatrix,thissaysthatifwemodifybasisB=[b1;b2;:::;bn]by Sinceelementarycolumnoperationscanbeachievedbyrightmultiplicationofthebasismatrix <br />

1.reorderingthecolumns<br />

thentheresultingmatrixisabasismatrixforthesamelattices. 3.addingintegermultiplesofsomecolumns<strong>to</strong>othercolumns 2.multiplyinganynumberofcolumnsby1


3.2.2DefinitionofHermiteNormalForm 32 CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

Definition3.2.5(HermiteNormalForm) WeintroducetheHermiteNormalFormofamatrix<strong>and</strong>proveitsuniqueness.InChapter??on<br />

Amatrix[aij]2Mm;n(R)withmnisinHermiteNormalForm(HNF)when: Page??wewillshowanalgorithmforobtainngtheHermiteNormalFormofanarbitrarymatrix.<br />

a)aij=0forj>i,i.e.,Aislowertriangular. b)aii>0fori=1;2;:::;m.<br />

Itisarbitrary<strong>to</strong>requirethattheHermiteNormalFormwillbelowertriangularratherthanupper triangular(seeCorollary3.2.8).AlternativesforPointc)appearintheliterature.In[DKT87] c)0aij


3.2.FUNDAMENTALSANDPROPERTIES HermiteNormalFormsforAcanbeobtainedfromeachotherbymultiplicationwithunimodular LetB1;B2;:::;Bn<strong>and</strong>C1;C2;:::;Cnbethecolumnvec<strong>to</strong>rsforB<strong>and</strong>C,respectively.Sinceboth 33<br />

conversely,socmm=bmm. L(B).Sincethediagonalelementsareallnon-zero,BmmustbeanintegermultipleofCm<strong>and</strong> matrices,wehavebyProposition3.2.4thatL(B)=L(C).Inparticular,Bm2L(C)<strong>and</strong>Cm2<br />

wehave(note:cik=0fork>i): L(B)=L(C)thereexistintegercoecientstj;tj+1;:::;tm2Z,sothatforeveryiwithjim LetjbethemaximalindexwithBj6=Cj.Sincecmm=bmm,wehavej+1m.Since<br />

(3.2) Asinthecaseofthemthcolumnvec<strong>to</strong>rs,weobtainbjj=cjj,sotj=1.Fori=j+1wehave bij=iXk=jtkcik<br />

(3.3) <strong>and</strong>weobtain bj+1;j=tjcj+1;j+tj+1cj+1;j+1=cj+1;j+tj+1cj+1;j+1<br />

SinceBisinHermiteNormalForm,wehavebythechoiceofjthemaximumindexwithBj6=Cj: (3.4) cj+1;j+1jtj+1cj+1;j+1=(bj+1;jcj+1;j)<br />

SinceCisalsoinHermiteNormalForm,wehave0cj+1;j


Notethatthecolumnvec<strong>to</strong>rsofA<strong>and</strong>HNF(A)yieldthesamelattice,namely 34 nXi=1Zai=Zgcd(a1;a2;:::;an)<br />

CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

InProposition3.2.4wewillseethatthisisnotacoincidence.TheEuclideanAlgorithmisthe basisfortheprocedureinChapter??<strong>to</strong>obtaintheHermiteNormalFormofanarbitrarymatrix A2Mm;n(Z). forL=L(b1;b2;:::;bn)arelinearlyindependent.Clearly,whenm


3.2.FUNDAMENTALSANDPROPERTIES 3.2.3Determinant<strong>and</strong>BasicBlock 35<br />

Definition3.2.9(Determinant) ThedeterminantdetLoflatticeL=(b1;b2;:::;bn)Rmisdenedby: Wedenethedeterminan<strong>to</strong>falattice:<br />

Forthescalarproducthu;vi=uTSvwehave:detL=det(BTSB)12,whereB:=[b1;b2;:::;bn]. detL=det[hbi;bji]1i;jn12<br />

Proof.LetB;Bbebasismatricesofthelattice<strong>and</strong>letT2GLn(Z)satisfyB=BT(theproof Proposition3.2.10 Thedeterminan<strong>to</strong>falatticeisindependen<strong>to</strong>fthechoiceofbasisb1;b2;:::;bn2Rm. holdsingeneralforT2GLn(R)).LetS2Mm;m(R)bethesymmetricmatrixwithhu;vi=uTSv. >FromdetT=1wehave: detL=det[hbi;bji]1i;jn12<br />

=detTTBTSBT12 =det(BT)TS(BT)12 =detBTSB12<br />

SinceB=BTitfollowsthat:detL=detBTSB12 =detbi;bj1i;jn12 =detL<br />

withrespect<strong>to</strong>thegivenbasis.Figure3.2.1showsthebasicblockofthelatticedeterminedby a;binR2. Theparallelepipeddeterminedbythebasisvec<strong>to</strong>rsofthelatticeisthebasicblockofthelattice <br />

Definition3.2.11(BasicBlock) Letb1;b2;:::;bnbeabasisofthelatticeL.Theparallelepiped (nXi=1tibi0t1;t2;:::;tn


36 CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

-- - <br />

--<br />

- <br />

---<br />

b ppppppppppppppppppppppppppppppppppppppppppp<br />

0 a -<br />

thedimensionofthelatticeLRmisequal<strong>to</strong>m. isthebasicblockwithrespect<strong>to</strong>basisb1;b2;:::;bn.Wearemostlyinterestedinthecasethat Figure3.2.1:BasicBlockoftheLatticeL(a;b)<br />

Definition3.2.12(FullDimensionalLattice) AlatticeLRmisfulldimensionalwhenrank(L)=m.<br />

Lemma3.2.13 ForeverylatticeL=L(b1;b2;:::;bn)Rm<strong>and</strong>thest<strong>and</strong>ardscalarproductwehave(note: Thefollowinglemmarelatesthest<strong>and</strong>ardscalarproduct<strong>to</strong>thedeterminan<strong>to</strong>falattice:<br />

volumeimplicitlyreliesonthescalarproduct): detL=voln (nXi=1tibi0t1;t2;:::;tn


3.2.FUNDAMENTALSANDPROPERTIES thelemma<strong>to</strong>thefulldimensionallatticeT(L)<strong>and</strong>usethefactthatTpreservesthevolume<strong>and</strong> scalarproduct: 37<br />

detL=detT(L)=det[hT(bi);T(bj)i]1i;jn12=det[hbi;bji]1i;jn12<br />

mensionallatticeswhenconsideringgeometricinvariants.Thisisbecausetheinvariantsremain unchangedbyisometricmappings.Examplesofgeometricinvariantsarevolumes,determinants, AsintheproofofLemma3.2.13wecanalwaysrestric<strong>to</strong>urattention<strong>to</strong>thecaseoffulldi-<br />

<br />

mappingfromspan(L)<strong>to</strong>Rn.Thisisometricmappingdoesnotingeneralmaintaintheintegrality ofvec<strong>to</strong>rs.Combina<strong>to</strong>ric<strong>and</strong>algorithmicinvestigationshouldthusnotberestricted<strong>to</strong>thecase metricconsiderationsdependsonthefactthatforeverylatticeLofranknthereisanisometric scalarproducts,<strong>and</strong>vec<strong>to</strong>rlengths.Theprinciplethatfulldimensionallatticessuceforgeo-<br />

offulldimensionallattices. 3.2.4Sub-<strong>Lattices</strong> Wedeneasub-latticeasasubse<strong>to</strong>flatticepointsthatformalatticeofthesamerank:<br />

Example3.2.15(Sub-Lattice) LetL1;L2belatticesofthesamerankwithL1L2.ThenwesaythatL1isasub-latticeofL2. Definition3.2.14(Sub-Lattice)<br />

Weconsiderasub-latticeofthelatticeZn(whichhasbasismatrixthe22identitymatrixI2).<br />

L(A)Z2isasub-latticeofZ2withrank(L(A))=2(seeFigure3.2.2).Thereisamatrix Let A:=20 02<br />

T2M2;2(Z)withA=I2T: 20 02=10 0120 |{z} 02<br />

detL1=detL2jdetTj. WewillseeinProposition3.2.16thatsuchaTalwaysexists,<strong>and</strong>moreover(Lemma3.2.18),that =:T<br />

Proposition3.2.16 LetL1beasub-latticeofL2Rm. a)Foreverybasisa1;a2;:::;anofL1thereexistsabasisb1;b2;:::;bnforL2with (3.5) foranuppertriangularmatrixT2Mn;n(Z). [a1;a2;:::;an]=[b1;b2;:::;bn]T


38 CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

sss<br />

sss<br />

sss<br />

sss<br />

s(0;0) (0;2) (0;4) (0;6)<br />

s(2;0) (2;2) (2;4) (2;6)<br />

s(4;0) (4;2) (4;4) (4;6)<br />

s(6;0)<br />

(6;2) (6;4) (6;6)<br />

b)Conversely,foreverybasisb1;b2;:::;bnforL2thereisabasisa1;a2;:::;anforL1with Figure3.2.2:Sub-LatticeL(A)ofZ2<br />

Remark3.2.17 >FromProperty(3.5)withuppertriangularmatrixT,wehavethat: property(3.5).<br />

Proof(ofProposition3.2.16).Weshowbothstatements: span(a1;a2;:::;ai)=span(b1;b2;:::;bi) fori=1;2;:::;n<br />

a)Letb1;b2;:::;bnbeanarbitrarybasisforL2.ThensinceL1L2eachelemen<strong>to</strong>fL1can amatrixS2Mn;n(Z)withdetS6=0<strong>and</strong>: beexpressedasanintegerlinearcombinationofthebasiselementsofL2.Thus,thereexists<br />

(US)Tislowertriangular.DeneT:=US(anuppertriangularmatrix)<strong>and</strong>basismatrix (3.6) TheHNF-Proposition3.2.6ensuresthatforSTthereexistsaU2GLn(Z),suchthatSTUT= [a1;a2;:::;an]=b1;b2;:::;bnS<br />

forL2: By(3.6)weobtain: [b1;b2;:::;bn]:=b1;b2;:::;bnU1<br />

[a1;a2;:::;an]=b1;b2;:::;bnS=[b1;b2;:::;bn]US


3.2.FUNDAMENTALSANDPROPERTIES b)Leta1;a2;:::;anbeanarbitrarybasisforL1.ThereisamatrixS2Mn;n(Z)withdetS6=0 <strong>and</strong>: 39<br />

ByCorollary3.2.8thereisaU2GLn(Z)suchthatSUisuppertriangular.Thestatement followsforthebasis [a1;a2;:::;an]=[b1;b2;:::;bn]S<br />

<strong>and</strong>T:=SU. [a1;a2;:::;an]:=[a1;a2;:::;an]U<br />

classes Contintuingourpreviousexample,thefac<strong>to</strong>rgroupZ2=L(A)consistsofthefourequivalence <br />

ThusZ2:L(A)=4.InLemma3.2.18weshowthatingeneraltheindex(i.e.,thenumberof 0+L(A);10+L(A);01+L(A);1+L(A)<br />

Lemma3.2.18 representatives,whichlieinthebasicblock. distinctcosets)isequal<strong>to</strong>jdetTj.Werepresenttheequivalenceclassesthroughtheirrespective<br />

LetL2=L(b1;b2;:::;bn)bealattice<strong>and</strong>L1=L(a1;a2;:::;an)asub-latticeofL2.LetT2<br />

Mn;n(Z)withA=BTforA:=[a1;a2;:::;an]<strong>and</strong>B:=[b1;b2;:::;bn].Then:<br />

Proof.LetS2Mm;m(R)bethesymmetricmatrixwithhu;vi=uTSv.Then: detL1=detL2jdetTj<br />

detL1=detATSA12 =det(BT)TS(BT)12 =detTTBTSBT12 =jdetTjdetBTSB12 =detTTT12detBTSB12 =jdetTjdetL2<br />

Definition3.2.19(IndexofaSub-Lattice) TheintegerjdetTj=detL1 detL2fromLemma3.2.18iscalledtheindexofthesublatticeL1inL2.


additivegroupL2ofindexjdetTj. Theindexisindependen<strong>to</strong>fthechoiceoflatticebasis.Inparticular,L1isasubgroupofthe 40 CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

<strong>and</strong>b1;b2;:::;bnsatisfyequation(3.5)withT=[tij].ThenL2=L1isanitegroup,theindex [L2:L1]ofL1inL2satises:[L2:L1]=detL1<br />

LetL1=L(a1;a2;:::;an)beasub-latticeofL2=L(b1;b2;:::;bn),sothatbasesa1;a2;:::;an<br />

Inparticular,L1=L2if<strong>and</strong>onlyifjdetTj=1.Thisrelationcanbeused<strong>to</strong>provethat a1;a2;:::;anformabasisforL2.LetL1bethelatticedenedbya1;a2;:::;an.Thenwe (3.7) detL2=detT=nYi=1jtiij<br />

Property(3.5).Thus:L2=L(a1;a2;:::;an)() knowdetL1 detL2<strong>and</strong>abasisb1;b2;:::;bnforL2.Further,thesystemofvec<strong>to</strong>rsa1;a2;:::;ansatises<br />

Itfollowsfrom(3.7)that: nYi=1jtiij=detL1 detL2=1<br />

Corollary3.2.20 Leta=(a1;a2;:::;an)T2Znnf0g<strong>and</strong>b2N.IfLa;bistheintegerlattice (whereh;iisthest<strong>and</strong>ardscalarproduct),then La;b=fx2Znjhx;ai0(modb)g<br />

Proof.WerstshowthatLa;bisasub-latticeofZn,thatis,thelatticeLa;bZnhasfullrank. detLa;b= gcd(a1;a2;:::;an;b) b<br />

Letvn=(1;:::;1)T2Zn.Choosen1integervec<strong>to</strong>rsv1;v2;:::;vn1,sothatbv1;bv2;:::;bvn2 Znarelinearlyindependent(forexample,therstn1identityvec<strong>to</strong>rs).Since<br />

sub-latticeofZn.>FromdetZn=1itfollowsfrom(3.7)that: itfollowsthatbv1;bv2;:::;bvn2ZnareinthelatticeLa;b<strong>and</strong>formabasis.ThusLa;bisa hbvi;aibhvi;ai0(modb) fori=1;2;:::;n<br />

Thefac<strong>to</strong>rgroupZn/La;bhasatmostbresidueclasses,namelyR0;R1;:::;Rb1with: (3.8) Ri:=fx2Znjhx;aii(modb)g detLa;b=[Zn:La;b]<br />

(3.9) Weshowthatthefac<strong>to</strong>rgrouphasexactlythesebresidueclasses.Werstconsiderthecase gcd(a1;a2;:::;an;b)=1


3.2.FUNDAMENTALSANDPROPERTIES Thenthereexistintegercoecientst1;t2;:::;tn+12Zwith: nXj=1tjaj+tn+1b=1<br />

41<br />

Forthevec<strong>to</strong>rsci:=it2Znwithi=0;1;:::;b1weobtain: Forthevec<strong>to</strong>rt:=(t1;t2;:::;tn)2Znwehave: ht;ai1tn+1b1(modb)<br />

TheresidueclassesR0;R1;:::;Rb1arethusnon-empty.From(3.9)weobtainby(3.8): hci;aiiht;aii(modb)<br />

Wealsoconsiderthecase detLa;b=[Zn:La;b]=b1= gcd(a1;a2;:::;an;b) b<br />

Thenforallx2Znwehave hx;ai0(modb) d:=gcd(a1;a2;:::;an;b)>1<br />

bjhx;ai<br />

soLa;b=Lad;bd.Sincegcda1 d;a2 d;:::;<strong>and</strong>;bd=1weobtain: () x;ad0(modbd); bdjx;ad<br />

detLa;b=detLad;bd=bd= gcd(a1;a2;:::;an;b) b<br />

3.2.5PrimitiveSystem <br />

abasisforthelattice.Thesecriteriaareusefulfortheconstructionofspeciallatticebases. Definition3.2.21(PrimitiveSystem) InProposition3.2.22wecharacterizesystemsoflatticevec<strong>to</strong>rs,whichcanbecompleted<strong>to</strong>obtain<br />

LetLRnbealattice.Thevec<strong>to</strong>rsb1;b2;:::;bk2LformaprimitivesystemforLif 1.b1;b2;:::;bkarelinearlyindependent. 2.span(b1;b2;:::;bk)\L=L(b1;b2;:::;bk)


whatissignicantisthatspan(b1;b2;:::;bk)\LL(b1;b2;:::;bk).Asinglevec<strong>to</strong>rb2Lforms Sincetheinclusionspan(b1;b2;:::;bk)\LL(b1;b2;:::;bk)forarbitrarylatticevec<strong>to</strong>rsb1;b2;:::;bk, 42 CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

aprimitivesystemwhen1kb62Lforallk2Zwithjkj2,sothegreatestcommondivisorofthe Proposition3.2.22 LetLRnbealatticewithrank(L)=n<strong>and</strong>b1;b2;:::;bk2L.Thevec<strong>to</strong>rsb1;b2;:::;bkform vec<strong>to</strong>rentriesis1.<br />

Proof.Weshowbothdirections: aprimitivesystemforLif<strong>and</strong>onlyiftheycanbecompletedforformabasisforL.<br />

\)"Letb1;b2;:::;bk;bk+1;:::;bnbeabasisofthelatticeL.Thevec<strong>to</strong>rsb1;b2;:::;bkare t1;t2;:::;tn2Z.Wehave, linearlyindependent.Eachvec<strong>to</strong>rb2Lcanbeuniquelydescribedasb=Pni=1tibiwith<br />

\("Letb1;b2;:::;bkbeaprimitivesystem<strong>and</strong>letbetheorthogonalprojection:span(L)! Sospan(b1;b2;:::;bk)\LL(b1;b2;:::;bk). b2span(b1;b2;:::;bk)()tk+1=tk+2==tn=0<br />

Thelattice(L)hasbasis(bk+1);(bk+2);:::;(bn)withbk+1;bk+2;:::;bn2L. WeshowthatL=L(b1;b2;:::;bn).Letb2L.From span(b1;b2;:::;bk)?.BytheproofofProposition3.2.3,(L)isalatticeofdimensionnk.<br />

thereisab2L(bk+1;bk+2;:::;bn)with(b)=(b).Let (b)2L(bk+1);(bk+2);:::;(bn)<br />

Thenbb2span(b1;b2;:::;bk).Sinceb1;b2;:::;bkformbyassumptionaprimitivesystem, (b)=nX i=k+1ti(bi) und b=nX<br />

wehavebb2L(b1;b2;:::;bk).Thusb2L(b1;b2;:::;bn). i=k+1tibi<br />

Thefollowingnotion,ofaMinkowski-reducedbasis,isnotalgorithmicallymotivated<strong>and</strong>ismentionedhereonlyforcompleteness(foraprocedure<strong>to</strong>obtaineaMinkowski-reducedbasissee<br />

B.HelfrichsArbeit[Helfrich85]): Definition3.2.23(Minkowski-Reduced) Letb1;b2;:::;bn2RmbeanorderedbasisforatticeL;thatis,theorderofthevec<strong>to</strong>rsisxed. Theorderedbasisissaid<strong>to</strong>beMinkowski-reducediffori=1;2;:::;n: kbik=minkbkb2L<strong>and</strong>(b1;b2;:::;bi1;b) isaprimitivesystemforL


3.2.FUNDAMENTALSANDPROPERTIES Definition3.2.24(Isometric<strong>Lattices</strong>) TwolatticesLRm<strong>and</strong>LRmareisometric,whenthereisanisometricmappingT:Rm!Rn 43<br />

LetL<strong>and</strong>Lbeisometriclattices<strong>and</strong>letTbeanisometricmappingwithT(L)=Lsothat withT(L)=L. b1;b2;:::;bnisabasisforL.Thenforthebasisbi:=T(bi),i=1;2;:::;n: hbi;bji=bi;bj<br />

isometryclassofalatticebasisB=[b1;b2;:::;bn]ischaracterizedbythematrixBTSB.In Wesaythattwosuchbasesareisometric.Forthescalarproducthu;vi=uTSvwehave:The for1i;jn<br />

particular,twolatticesareisometricif<strong>and</strong>onlyiftheyhaveisometricbases.Forexample,inthe caseofthest<strong>and</strong>ardscalarproduct:<br />

B:=p21=p2 0p3=2 B:=2410 11 01 3 5 BTB=BTB=21 12<br />

3.2.6OrthogonalSystems<br />

Definition3.2.25(OrthogonalSystem,OrthogonalProjectioni) Thegoaloflatticebasisreductionis,givenanarbitrarylatticebasis,obtainabasisofshortest<br />

Letb1;b2;:::;bn2RmbeanorderedlatticebasiswhereL=L(b1;b2;:::;bn)isofrankn.Dene possiblevec<strong>to</strong>rs;thatis,vec<strong>to</strong>rsascloseaspossible<strong>to</strong>mutuallyorthogonal.<br />

<strong>to</strong>betheorthogonalprojection,thatis,forallb2Rm: i:Rm!span(b1;b2;:::;bi1)?<br />

Wewritebi:=i(bi). bi(b)2span(b1;b2;:::;bi1) i(b)2span(b1;b2;:::;bi1)?<br />

Wecanobtainb1;b2;:::;bnbytheGram-Schmidtprocedure:<br />

(3.11) (3.10) b1:=b1 bi:=i(bi)=bii1 Xj=1i;jbj fori=2;3;:::;n


wheretheGram-Schmidtcoecientsi;jaredenedby: 44<br />

i;j:=Dbi;bjE CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

Dbj;bjE=Dbi;bjE<br />

vec<strong>to</strong>rbiintermsoftheorthogonalsystem: Inparticular,i;i=1<strong>and</strong>forj>i,i;j=0.Bydenition(3.10)<strong>and</strong>(3.11)wecanexpressthe kbjk2<br />

(3.12) ThenPkj=1i;jbjistheprojectionofbion<strong>to</strong>span(b1;b2;:::;bk)<strong>and</strong>Pij=k+1i;jbjtheprojection bi=bi+i1 Xj=1i;jbj ofbion<strong>to</strong>span(b1;b2;:::;bk)?.Moreover,fori=1;2;:::;nwehave<br />

Thedescription(3.12)ofthebasisvec<strong>to</strong>rsb1;b2;:::;bncanbeexpressedinmatrixform: span(b1;b2;:::;bi)=spanb1;b2;:::;bi<br />

Notethatifm=nthenbBissquarewithnon-zerodeterminant<strong>and</strong>wehave [b1;b2;:::;bn]=hb1;b2;:::;bni[i;j]T1i;jn<br />

becausedet[i;j]T1i;jn=1(uppertriangularwith1'sonthediagonal).Thelastequalityholds becausethecolumnsofbBaremutuallyorthogonal.Ifm>n,thenconsiderthemappingT: detB=detbBdet[i;j]T1i;jn=detbB=nYi=1kbik<br />

Tiseasilyshown<strong>to</strong>beisisometric,sothematrix span(L)!Rnwith T(bi)=hi;1kb1k;i;2kb2k;:::;i;nkbnkiT<br />

[T(b1);T(b2);:::;T(bn)]= 2 64 kb 1kkb2k... 0 0. 0kbnk 0.<br />

3 75 2 640 1 102;13;1 3;2 1 n;1 n;2 0 . ......n;n1 0 1.<br />

375 isabasisforalatticeinRnthatisisometric<strong>to</strong>L(b1;b2;:::;bn).Wecannowproceedasinthe (3.13) casem=n:Sincedet[i;j]T1i;jn=1,wehavefromtheproofofProposition3.2.10: detL(b1;b2;:::;bn)=nYi=1kbik


3.2.FUNDAMENTALSANDPROPERTIES Proposition3.2.26 LetL=L(b1;b2;:::;bn)<strong>and</strong>letb1;b2;:::;bndenotetheGram-Schmid<strong>to</strong>rthogonalizationof 45<br />

b1;b2;:::;bn.Let(L)denotethelength,usingthe`2norm,oftheshortestnon-zerolattice vec<strong>to</strong>rinL.Then(L)minibi2.<br />

eachbiintermsoftheGram-Schmid<strong>to</strong>rthogonalizationweget: vec<strong>to</strong>rinL.Sincea2Lwecanwritea=Pni=1ibi,wherei2Zfori=1;2;:::;n.Expressing Proof.Intheproof,weletkkdenotethe`2norm.Leta2Lbeaminimumlengthlattice<br />

Letkbethelastindexforwhichk6=0,soj=0forallj>k.Letusdenejfor1jnby a=nXi=1iiXj=1ijbj<br />

(Notethatbychoiceofk,k=k.)Thena=Pnj=1jbj.Sincethebjaremutuallyorthogonal, wehave j=nXi=jiij<br />

Thus,kakjkjbk=jkjbk.Sincek2Znf0gwehavejkj1,<strong>and</strong>theproofiscomplete. kak2=nXj=1(j)2bj2(k)2bk2<br />

mutuallyorthogonal: Wedenetheorthogonalitydefect,ameasureofhowfarthebasisvec<strong>to</strong>rsarefrombeing <br />

Definition3.2.27(OrthogonalityDefect) Theorthogonalitydefec<strong>to</strong>fabasisb1;b2;:::;bnofthelatticeLis:<br />

Sincekbikkbik,theorthogonalitydefectisgreaterthanorequal<strong>to</strong>1.Itis1preciselywhenthe OrthDefect(L):=Qni=1kbik detL;<br />

basisvec<strong>to</strong>rsaremutuallyorthogonal,<strong>and</strong>thusbi=bifori=1;2;:::;n(see(3.13)). 3.2.7QuadraticForms Inthissectionthescalarproductalwaysdenotesthest<strong>and</strong>ardscalarproduct.Abasismatrix B=[b1;b2;:::;bn]yieldsthefollowingquadraticformQFBintherealvariablesx1;x2;:::;xn: QFB(x1;x2;:::;xn):=X 1i;jnhbi;bjixixj


ThisquadraticformQFBispositivedenite,thatis,QFB(x1;x2;:::;xn)0<strong>and</strong>QFB(x1;x2;:::;xn)= 46 0if<strong>and</strong>onlyifx1=x2=:::xn=0.QFBispositivedenitimplies: CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

<strong>and</strong>Pni=1xibiisthenullvec<strong>to</strong>rexactlywhenx1=x2=:::xn=0.Forintegerx1;x2;:::;xnQFB QFB(x1;x2;:::;xn)=nXi=1xibi2<br />

takesthevalueofthesquareofthelengthofthecorrespondinglatticevec<strong>to</strong>rinL(b1;b2;:::;bn). Conversely,foreverypositivedenite,symmetrische,quadraticform<br />

thereisalatticebasisb1;b2;:::;bn,sothathbi;bji=qijfor1i;jn.Theneverypositive QF=X<br />

denit,symmetricmatrix(qij)2Mn;n(R)canbewrittenas(qij)=BTBwithB2Mn;n(R). 1i;jnqijxixj<br />

anisometricmappingthattransformsBin<strong>to</strong>B.Thequadratic,positivedeniteformsexpress formsisinthissenseequivalent.Theoldercontributions,ofJ.L.Lagrange,C.F.Gau,C.Hermite, A.Korkine,G.ZolotareundH.Minkowski,wereexpressedintermsofquadraticforms.Two uniquelytheisometryclassesofthelatticebasis.Thetheoryoflatticebases<strong>and</strong>positivedenite TwolatticebasesB;ByieldthesamequadraticformQFB=QFB,preciselywhenthereis<br />

arecongruent,whentheycanbetransformedin<strong>to</strong>oneanotherthroughunimodulartransformations,thatis,thereexistsamatrixU2GLn(Z)with:<br />

1i;jnqijxixj QFQ=X 1i;jnqijxixj <strong>and</strong> QFQ=X<br />

Forexample,fortwobasesB<strong>and</strong>Bofthesamelattice,thecorrespondingquadraticformsQFB <strong>and</strong>QFBarecongruent. [qij]1i;jn=UTqij1i;jnU<br />

Inthissectionthescalarproductisthest<strong>and</strong>ardscalarproduct.Wedenetheduallattice: 3.2.8Dual<strong>and</strong>WholeLattice<br />

LetLbealattice.Thedual(polar,reciprocal)latticeiddenedas: Definition3.2.28(Dual(polar,reciprocal)Lattice)<br />

Proposition3.2.29 LetLRnbealatticeoffullrankwithbasismatrixB:=[b1;b2;:::;bn].ThenB1Tisabasis L:=fx2span(L)j8b2L:hx;bi2Zg<br />

matrixoftheduallatticeL.Inparticular,(L)=L.


3.2.FUNDAMENTALSANDPROPERTIES Proof.Let B:=[b1;b2;:::;bn]:=B1T 47<br />

WemustshowthatL=L(B).Weshowbothinclusions: L(B)L:FortheidentitymatrixI I=B1B=B1TTB=(B)TB=264hb1;b1ihb1;b2ihb1;bni hb2;b1ihb2;b2ihb2;bni<br />

Thushbi;bji2f0;1gfori;j=1;2;:::;n.Forx=Pni=1tibi2L(B)witht1;t2;:::;tn2Z hbn;b1ihbn;b2ihbn;bni . . ... . 3 75 wehave:<br />

Sincebi2Rn=span(L)fori=1;2;:::;n,itfollowsforallx2L(B),thatx2L. hx;bji=nXi=1tihbi;bji2Z forj=1;2;:::;n<br />

L(B)L:Foreverya2Lwemustshowthata2L(B).Sinceb1;b2;:::;bn2L,we obtainfromthedenitionoftheduallattice: BTa=264b1 bn b2. 3 75a=264hb1;ai hb2;ai<br />

Further, hbn;ai . 3 752Zn<br />

Letti:=hbi;ai2Zfori=1;2;:::;n.Thenacanbeexpressedas a=Ea=BB1Ta=B1TBTa=BBTa |{z} 2Zn<br />

<strong>and</strong>a2L(B). a=nXi=1tibi<br />

Moreover,(L)=L,sincetheoperationsofinverting<strong>and</strong>transpositioncommute. Definition3.2.30(Self-DualLattice) AlatticeLiscalledself-dualwhenL=L. <br />

Forexample,Znisaself-duallattice.Toconlcudethissection,wedeneawholelattice:


AlatticeLiscalledwholewhenha;bi2Zforalla;b2L. Definition3.2.31(WholeLattice) 48 CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

Proposition3.2.32 ForeverywholelatticeL,LL1 Proof.Theproofisleft<strong>to</strong>thereader. detL2L.<br />

3.3 Length-<strong>and</strong>Weight-ReducedLatticeBases <br />

Inthissectionwedenetwoalgorithmicallymotivatedtypesofreducedness:lengthreduced<strong>and</strong> weightreducedlatticebases. Definition3.3.1(LengthReducedBasis) Theorderedbasisb1;b2;:::;bnislengthreducedwhenji;jj12for1jFromEquation(3.12)onPage44wehavefori=1;2;:::;n: fori=1;2;:::;n<br />

bi=bi+i1<br />

Thevec<strong>to</strong>rsb1;b2;:::;bnoftheorthogonalsystemaremutuallyorthogonal,soitfollowsthat: Xj=1i;jbj<br />

kbik2=bbi2+i1<br />

Sincethebasisisbyassumptionlengthreduced,wehaveji;jj12,whencetheProposition Xj=12i;jkbjk2<br />

follows. Toprovecorrectnessofthelengthreductionalgorithmweconsiderthebasisvec<strong>to</strong>rsexpressedin Algorithm3.3.1transformsagivenlatticebasisin<strong>to</strong>alengthreducedbasisforthesamelattice.


3.3.LENGTH-ANDWEIGHT-REDUCEDLATTICEBASES Algorithm3.3.1LengthReduction INPUT:OrderedLatticeBasisb1;b2;:::;bn2Rm 49<br />

1.FORi=2;3;:::;nDO FORj=i1;i2;:::;1DO<br />

ENDfori ENDforj bi:=bidi;jcbj<br />

termsoftheircoordinatesintheorthogonalsystem: OUTPUT:LengthReducedBasisb1;b2;:::;bn<br />

b1 b2 b3 b1 1 b2 0 b3 bn1bn<br />

. 2;1 3;1 . 3;2 1 01 ... 0 Weobservethat: bn1=(n1;1n1;2n1;3 n;1 n;2 n;3 n;n11) 1 0) .)<br />

Thisisaweakformofreduction. Thestepbi:=bidi;jcbjcausesnew Inparticular,new i;j12,<strong>and</strong>thei;remainunchangedfor>j. i;j:=old i;j1old i;j.<br />

Theorderedbasisb1;b2;:::;bnisweightreducedwhen: Definition3.3.3(WeightReducedBasis) a)jhbi;bjij b)kbikkbi+1k kbjk212 for1j


50Propertyb)ofweightreducednesssaysthatthebasisvec<strong>to</strong>rsareinascendingorderoflength:<br />

kb1kkb2kkb3kkbn1kkbnk CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

Propertya)isequivalent<strong>to</strong>kbikkbibjk From kbibjk2=hbibj;bibji=kbik22hbi;bji+kbjk2 for1j12THENbi:=bidrcbj<strong>and</strong>F:=true r:=hbi;bjikbjk2 =reductionstep=<br />

ENDwhile ENDfori ENDforj<br />

OUTPUT:WeightReducedBasisb1;b2;:::;bn Proposition3.3.4 Everylatticehasaweightreducedbasis. TheproofofthePropositionisimmediatefromthecorrectnessofAlgorithm3.3.2,whichtransformsanarbitrarylatticebasisin<strong>to</strong>aweightreducedbasisforthesamelattice.Thecorrectness<br />

ofthealgorithmfollowsfromthetwoobservations:


3.4.EXAMPLES Thealgorithmterminates,sinceeveryreductionstepreducesthelengthofonebasisvec<strong>to</strong>r Itisclearbyinspectionthattheouptutisaweightreducedbasis. 51<br />

thesequenceofthevec<strong>to</strong>rs,<strong>and</strong>inparticulartheshortestbasisvec<strong>to</strong>risthers<strong>to</strong>ne.However, thealgorithmisnotecient.Inthefollowingchapterswewillstudystrongernotionsofreduction: IncontrastwithAlgorithm3.3.1forlengthreduction,theweightreductionprocedureexchanges whileleavingtheremainingbasisvec<strong>to</strong>rsunchanged,<strong>and</strong>everylatticeisdiscrete.<br />

3.4 LLLreductioninchapter6<strong>and</strong>HKZ-<strong>and</strong>-reductioninChapter??.<br />

st<strong>and</strong>ardone.Therstsuccessiveminimum1isthelengthoftheshortestnon-zerolatticevec<strong>to</strong>r Inthissectionwedescribesomelattices<strong>and</strong>theirrespectivebases.Thescalarproductisthe Examples<br />

(aformaldenitionfollowslater). Forthelattice<br />

thefollowingrowvec<strong>to</strong>rsformabasis: An:=((x0;x1;:::;xn)2Zn+1nXi=0xi=0)<br />

264 bn1 b b2. 1<br />

3 75:= 2 64 01+10 . ...... 0 0<br />

Clearly,L(b1;b2;:::;bn)isasub-latticeofAn,sinceb1;b2;:::;bn2An<strong>and</strong>theyarelinearly<br />

0 01+100 . 3 75<br />

i=n;n1;:::;2,subtractthevec<strong>to</strong>rbifromx,untilthe(i+1)thcomponentiszero.Onlythe rsttwoentriesofthevec<strong>to</strong>rx0arenon-zero,<strong>and</strong>itisinAn.Moreover,x0=x01,<strong>and</strong>thevec<strong>to</strong>r isanintegermultipleofb1. independent.Conversely,letx=(x0;x1;:::;xn)2Anbeanarbitraryelemen<strong>to</strong>fAn.For<br />

Thelattice<br />

hasabasiscomprisedofthefollowingrowvec<strong>to</strong>rs: Dn:=((x1;x2;:::;xn)2ZnnXi=0xi0(mod2))<br />

264 b b2 b3. 1<br />

bn1<br />

3 75:= 2 64<br />

+20 0+110 0 . ...... 0 0+1100 0.<br />

3 75


L(b1;b2;:::;bn)isasublatticeofDn,sinceb1;b2;:::;bn2Dn<strong>and</strong>theyarelinearlyindependent. Alternatively,b01=(1;1;0;:::;0).Sinceb1=b01+b2oneobtainsthesamelattice.Clearly 52 CHAPTER3.INTRODUCTIONTOLATTICETHEORY<br />

removefromxthevec<strong>to</strong>rbiuntiltheithcomponentis0.Theonlynon-zeroentryinthevec<strong>to</strong>r Conversely,letx=(x1;x2;:::;xn)2Dnbeanarbitraryvec<strong>to</strong>rinDn.Fori=n;n1;:::;2,<br />

Letn0mod4.Thelattice isthers<strong>to</strong>ne,<strong>and</strong>thevec<strong>to</strong>rliesinDn.Thusx10(mod2),<strong>and</strong>thevec<strong>to</strong>risaninteger multipleofb1.NotethatdetDn=2.Clearlythereisnolatticevec<strong>to</strong>rshorterthanthebasis vec<strong>to</strong>rs,so1(Dn)=p2.<br />

hasabasiscomprisedofthefollowingrowvec<strong>to</strong>rs: En:=(x1;x2;:::;xn)2ZnPni=0xi0(mod4)<strong>and</strong> xjxj+1(mod2)for1j4,v:=(2;2;0;:::;0)isa 8else<br />

x2En,xn1xn(mod2)).<br />

Weshow: Leta=(a1;a2;:::;an)2Znnf0g.Considerthelatticeofintegervec<strong>to</strong>rsorthognal<strong>to</strong>a: La:=span(a)?\Zn=ft2Znjha;ti=0g<br />

ThelatticeLaclearlyhasdimensionn1.Letc2;c3;:::;cnbeabasisforLa.Since detLa= gcd(a1;a2;:::;an) kak<br />

span(La)\Zn=La


existsavec<strong>to</strong>rc12Zwith: 3.4.EXAMPLES thevec<strong>to</strong>rsofthebasisformaprimitivesystemforthelatticeZn<strong>and</strong>byProposition3.2.22there 53<br />

Thepar<strong>to</strong>fc1orthogonal<strong>to</strong>Lais Theentriesofthevec<strong>to</strong>rc1arerelativelyprime,sincec1isaprimitivevec<strong>to</strong>rwithrespect<strong>to</strong>Zn. L(c1;c2;:::;cn)=Zn<br />

<strong>and</strong>haslength hc1;ai<br />

*hc1;ai kak2a<br />

block(\surfaceareadetLatimesheightjhc1;aij Weobtainfromthegeometricinterpretationofthelatticedeterminantasthevolumeofthebasic kak2a;hc1;ai kak2a+12=hc1;ai kak2ha;ai12=hc1;ai<br />

kak"):<br />

thatjhc1;aijistheleastpositivewholenumberintheprincipalidealPni=1Zai=Zgcd(a1;a2;:::;an), forwhichthereexistsac12Znwith detZn=1=detLajhc1;aij kak<br />

thatjkj=1.Wethusobtain <strong>and</strong>c1=kc1forsomek2Z.Sincetheentriesofthevec<strong>to</strong>rc1arerelativelyprime,itfollows hc1;ai=gcd(a1;a2;:::;an)<br />

<strong>and</strong>theclaimfollows. hc1;ai=gcd(a1;a2;:::;an);


54 CHAPTER3.INTRODUCTIONTOLATTICETHEORY


Chapter4<br />

SucessiveMinima,Minkowski's Theorems,<strong>and</strong>theHermite Constant<br />

denotestheEuclideannorm. Inthischapterwedenethesuccessiveminimaofalattice.Wepresenttwowellknowntheorems ofMinkowski.WedenetheHermiteconstantn<strong>and</strong>presentlower<strong>and</strong>upperboundsforthis value.Unlessotherwisestate,thescalarproductisthest<strong>and</strong>ardone:h;i<strong>and</strong>kxk=phx;xi<br />

minimaaredenedby: Inchapter3.4weinformallyintroducedtherstsuccessiveminimum.Ingeneral,thesuccessive 4.1 SuccessiveMinima<strong>and</strong>theFirstMinkowskiTheorem<br />

Letkkbeanarbitrarynorm.ForeverylatticeLRmofranknthesuccessiveminima Definition4.1.1(SuccessiveMinima1;2;:::;n) 1;2;:::;nwithrespect<strong>to</strong>thenormkkaredenedas: i=i(L):=inf80Thereareilinearlyindependent<br />

Thedenitionofsuccessiveminimaisdue<strong>to</strong>H.Minkowski.Figure4.1.1illustratestheconcept withkcjkrforj=1;2;:::;i9=; vec<strong>to</strong>rsc1;c2;:::;ci2L fori=1;2;:::;n<br />

phu;uiaregeometriclatticeinvariants;thatis,thesevaluesremainunchangedunderisometric x.12n.Thesuccessiveminimawithrespect<strong>to</strong>aEuclideannormkuk:= foralatticeL(a;b):Therstsuccessiveminimumwithrespect<strong>to</strong>theEuclideannormisthevec<strong>to</strong>r<br />

55


56 CHAPTER4.SUCC.MINIMA,MINKOWSKI'STHEOREMS,HERMITECONSTANT<br />

x<br />

0 -b<br />

1<br />

a<br />

transformationsofthelattice.Foreverylatticebasisb1;b2;:::;bn,fori=1;2;:::;n: Figure4.1.1:Exampleoftherstsuccessiveminimum1(L(a;b))<br />

whenthevalueskbik Thesuccessiveminimayieldameasureofthereducednessofalatticebasis.Abasisis\reduced", ifori=1;2;:::;nare\small"(\close"<strong>to</strong>1).Thevec<strong>to</strong>rsofareducedbasis j=1;2;:::;ikbjki max<br />

arenearlyorthogonal.Ingeneralthereisnobasisb1;b2;:::;bnwithkbik=ifori=1;2;:::;n. Considerforexamplethelattice<br />

<strong>and</strong>theEuclideannorm.Forn5,1=2==n=1<strong>and</strong>thecanonicalunitvec<strong>to</strong>rs(the L:=Zn+Z12;:::;12 |{z} nT<br />

onlyvec<strong>to</strong>rswithlength1)donotformabasis. cessiveminimumintheEuclideannorm Thesuccessiveminimadependontheunderlyingnorm.Inparticular,considertherstsuc-<br />

<strong>and</strong>inthesup-norm: kLk1:=1;1(L)=minfkbk1:b2Lnf0gg kLk:=1(L)=minfkbk:b2Lnf0gg<br />

ForlatticeLRmitfollowsfromkxk1kxkpnkxk1forallx2Rm,that:<br />

Proposition4.1.2(Minkowski1896) ThequantitykLk1isnotageometricinvariant.However,wehavethefollowingtightbound: 1;1(L)1(L)pn1;1(L)<br />

LetLRmbealatticeofrankn.ThenkLk1(detL)1n.


ThisboundistightsincekZnk1=1=(detZn)1n. 4.1.SUCCESSIVEMINIMAANDTHEFIRSTMINKOWSKITHEOREM Werstgivetheproofsketchappearingin[Lovasz86]forthecasen=m.LetQdenotethe 57<br />

cube ThenQhasn-dimensionalvolumedetL.ConsiderthecubesQ+b,whereb2L.Ifallthese cubeswerepairwisedisjoint,thenthedensityoftheirunionwouldbelessthan1;however,a Q:=nx2Rn:kxk112(detL)1=no<br />

(Q+b1)\(Q+b2)6=;.Lety2Q+b1\Q+b2.Thenkyb1k1;kb2yk112(detL)1=n.It simplecountingargumentshowsthatitisexactlyone.Thus,thereexistb1;b22Lsuchthat followsfromthetriangleinequalitythat<br />

everyfulldimensionallatticeL,thereexistsb2Lwithkbk2pn(detL)1=n.(Thesamefollows Sinceb2b12Lwearedone.Asasimplecorollary,sincekbk2pnkbk1,wehavethatfor kb2b1k1kyb1k1+kb2yk1(detL)1=n<br />

fromProposition4.1.2forthegeneralcasenm.) suchthatQ+b0intersectsQ.Forsimplicity,letusassumedetA=1.IfL=L(a1;a2;:::;an), thenwe\only"need<strong>to</strong>considervec<strong>to</strong>rsb=1a1++nanwhere Followingtheproofsketch,wecanndb2Lwithkbk1(detL)1=nbysearchingforb02L<br />

(TheequalitycomesfromCramer'srule.Theinequalityrequiresproof.Thekeypointsarethat, assumingkbk1


58 yi: sothatthefollowingintersectionisnon-empty<strong>and</strong>wecantherforechoosefromtheintersection CHAPTER4.SUCC.MINIMA,MINKOWSKI'STHEOREMS,HERMITECONSTANT<br />

poin<strong>to</strong>fasubsequencey(i)i2N.Thesequenceb(i)i2NLconverges<strong>to</strong>y.Thesequence b(i)i2NLisbounded<strong>and</strong>runsthroughonlynitelymanylatticepoints.Atleas<strong>to</strong>nelattice SinceQiscompact,thesequence(yi)i2Nhasalimitpointy2Q.Lety2Qbeaboundary yi21+1iQ\1+1iQ+bi i=1;2;:::<br />

is,n=m.WeapplyLemma4.1.3<strong>to</strong>theset Proof(ofProposition4.1.2).WerstconsiderthecaseinwhichLisfulldimensional,that pointb2Lnf0gappearsinnitelyoften.Itfollowsthaty2Q\(Q+b).<br />

Q:=nx2Rm:kxk112(detL)1mo <br />

Qisanm-dimensionalcubewithsidelength(detL)1m.Thenvol(Q)=detL.ByLemma4.1.3 thereexistsab2Lnf0g<strong>and</strong>y2Q\(Q+b).Sincey;yb2Qwehave:<br />

Itfollowsfromthetriangleinequalitythat: kybk112(detL)1m kyk112(detL)1m<br />

inmlet: Thecasen


f:Rm!Rmbedenedasfollows.Forb2Rmthevalueoff(b)isthevec<strong>to</strong>rinRmthat 4.1.SUCCESSIVEMINIMAANDTHEFIRSTMINKOWSKITHEOREM Weconcludetheproofoftheclaimbyshowingthatdet'I(bB)T'I(bB)=det'I(B)T'I(B).Let 59<br />

hf(x);f(y)i=h'I(x);'I(y)i.Thus, coordinatesnotinI.Thenkf(b)k=k'I(b)kforallb2Rm,<strong>and</strong>ingeneral,forallx;y2Rm, agreeswithbonallthecoordinatesintheindexsetI(so'I(f(b))='I(b))<strong>and</strong>iszeroonallthe<br />

<strong>and</strong>itremains<strong>to</strong>provethatdet'I(bB)T'I(bB)=detf(bB)Tf(bB)<br />

det'I(B)T'I(B)=detf(B)Tf(B)<br />

afterright-multiplicationbyT<strong>and</strong>obtainthesameresult).Thus, Now,sincebB=BT,wehavef(bB)=f(BT)=f(B)T(wecanzerooutagivenrowbeforeor detf(bB)Tf(bB)=detf(B)Tf(B)<br />

detf(bB)Tf(bB)=det((f(B)T)Tf(B)T)<br />

<strong>and</strong>theproofoftheclaimiscomplete. =detf(B)Tf(B) =det(TTf(B)Tf(B)T)<br />

jxjj=kxk1.ThenthereexistsachoiceforIsuchthatj2I<strong>and</strong>'I(L)isoffullrank.Forthis Letx2Lhavesup-norm1;1(L).Writex=(x1;x2;:::;xn)T,<strong>and</strong>letjbesuchthat <br />

choiceofI: kLk1=kxk1=k'I(b)k1det'I(L)1n(detL)1n<br />

Wedene: <br />

Definition4.1.5(Convex,Null-SymmetricSet) SRmisconvex,ifwhenx;y2S<strong>and</strong>2[0;1]itisalsothecasethatx+(1)y2S.Sis callednull-symmetric,orsimply,symmetric,ifxisinSwheneverxisinS. Proposition4.1.2withn=misaspecialcaseofthefollowingtheorem,commonlyknownas Proposition4.1.6(Minkowski'sFirstTheorem1893) LetLRmbeafulldimensionallattice<strong>and</strong>SRmaconvex,symmetric,compactsetwith \Minkowski'sTheoremforConvexBodies."<br />

vol(S)2mdetL.ThenjS\Lj3,thatis,Scontainsatleasttwonon-zerovec<strong>to</strong>rsy2L.


60 Proof.Wetake: CHAPTER4.SUCC.MINIMA,MINKOWSKI'STHEOREMS,HERMITECONSTANT<br />

Thenvol(Q)=2mvol(S)detL.ByBlichfeldt'sLemma4.1.3,thereexistsb2Lnf0gwith: Q:=12S=12xx2S 12S\12S+b6=; Rm<br />

f0;bgS\L. Letybeinthisintersection.Theny212S<strong>and</strong>y=x+bwithx212S.Letw=2x<strong>and</strong>z=2y. Thenw;z2S.Moreover,sinceSisconvex,12w+(112)z2S.Thus,b=xy2S.Inparticular,<br />

m,sincevol(S)=2mdetL.AfurtherconsequenceofProposition4.1.6isaproofG.L.Dirichlet's Proposition4.1.6applied<strong>to</strong>S=x2Rm:kxk1(detL)1=m yieldsProposition4.1.2forn= <br />

Chapter1: Proposition4.1.7(Dirichlet1842) [Di1842]theoremregardingsimultaneousapproximationofrealnumbersbyrationalsdiscussedin<br />

Let1;2;:::;nberealnumbers<strong>and</strong>2(0;12).Thenthereexistintegersp1;p2;:::;pn<strong>and</strong>q with0


4.2.HERMITECONSTANTANDCRITICALLATTICE <strong>and</strong>thelatticeL(A)generatedbythecolumnsofA.Everyvec<strong>to</strong>rb2L(A)canbewrittenas b=Ap,wherep=p1;p2;:::;pTn+12Zn+1.Letb2L(A)benon-zerowithkbk2".Then 61<br />

pn+16=0.Assumewithoutlossofgeneralitythatpn+1


62 Inthefollowingweimprovethisupperbound.Wedenoteby CHAPTER4.SUCC.MINIMA,MINKOWSKI'STHEOREMS,HERMITECONSTANT<br />

thendimensionalballwithrRadiusrcenteredattheorigin.Thevolumeofthendimensional ballwithradius1is Sn(r):=fx2Rn:kxkrg<br />

(4.1) TheGammafunctionisdenedby12=p,(n+1)=n!forn2N<strong>and</strong>ingeneralforx2R+: vol(Sn(1))= 1+n2=2n2<br />

n12n<br />

WeobtainfromthisanupperboundfortheHermitconstantn: Proposition4.2.3 (x+1)=x(x)<br />

Proof.LetLRmbeafulldimensionallatticewithn=1(L)2 nvol(Sn(1))2n=41+n22n2n<br />

4 e+O(1)0;2342n+O(1)<br />

Proposition4.1.6<strong>to</strong>theballSn(r).Wechooseradius: r:= 2 (detL)2n<strong>and</strong>detL=1.Weapply<br />

with(4.1): Thenvol(Sn(r))=2n.ByProposition4.1.6thereexistsb2Sn(r)\Lwithb6=0.Weobtain vol(Sn(1))1n<br />

ByStirling'sapproximation[Knuth71,1.2.11.2,Aufgabe5]: (x+1)=p2xxex1+O1x 1(L)kbkr= vol(Sn(1))1n=2 2 p1+n21n<br />

<strong>and</strong>x=n2followsfromthefactthatdetL=1: (4.2)<br />

n=1(L)2 forx2R+;<br />

4pnn2en21+O1n2n (detL)2n<br />

=4n 2e(n)1n1+O1n2n


4.2.HERMITECONSTANTANDCRITICALLATTICE Sincelim n!1n1n=1,thatis,n1n=1+O(1),weobtain: 63<br />

n2n e+O(1)<br />

1=2 <br />

qppppppppppppppppppppppppp ]<br />

q pppppppppppppppppppppppppp<br />

qq<br />

radiusr:=121oneverypoin<strong>to</strong>fthefulldimenionallatticeLRnThisyieldsthegittertige??? TheproofofProposition4.2.3canbeillustratedasfollows(seeFigure4.2.1):Placeaballof Figure4.2.1:IllustrationfortheProofofProposition4.2.3<br />

Definition4.2.4(GitterartigeSpherePacking) spherepackingofL,denedasfollows. LetMRnbeanon-emptydiscreteset<strong>and</strong>r>0.ThespherepackingforM<strong>and</strong>ris:<br />

ballsm1+Sn(r)<strong>and</strong>m2+Sn(r)sharenointeriorpoint. Thespherepackingiscalledgitterartig,whenforeverypairofneighboringpointsm1;m22Mthe fm+Sn(r)jm2Mg<br />

(4.3) ofthelatticeyield<strong>to</strong>getheroneballofradius121.Itfollowsthat: LetusconsiderthegitterartigespherepackingforL<strong>and</strong>121.The2npartialballsinabasicblock<br />

Definition4.2.5(WidthofaLattice) detLvolSn121=n1vol(Sn(1))<br />

ThewidthofafulldimensionallatticeLRnisthewidthofthegitterartigespherepackingfor 2n<br />

L,thatis,thevolume|volumenanteil???oftheballofthegitterartigespherepackinginRn. ThewidthofthespherepackingoflatticeLis: volSn121 detL =n1 detLvol(Sn(1)) 2n


64 oftheHermiteconstant Forxedn,thewidthismaximalwhenthefac<strong>to</strong>rn1 CHAPTER4.SUCC.MINIMA,MINKOWSKI'STHEOREMS,HERMITECONSTANT<br />

n=max1(L0)2 (detL0)2nL0Rnfulldimensionallattice detLisaslargeaspossible.Bythedenition<br />

weobtainthatthewidthofthelatticeLRnismaximalexactlywhen<br />

Wedenethegloballyextreme,orcriticallattice: detL!=(n)n2=max1(L0)n n1 detL0L0Rnfulldimensionallattice<br />

Definition4.2.6(GloballyExtreme(orCritical)Lattice) AfulldimensionallatticeLRniscalledgloballyextreme,orcritical,when<br />

thatis,1(L)2 (detL)2nistheabsolutemaximumforlatticesofrankn. (detL)2n=n; 1(L)2<br />

Alatticeiscriticalexactlywhentheballofradius121aroundthelatticepointsformsthewidest<br />

AfulldimensionallatticeL=L(b1;b2;:::;bn)Rnissaid<strong>to</strong>belocallyextremewhen Definition4.2.7(LocallyExtremeLattice) gitterartigespherepackingofRn.<br />

doesnotincreasewithinnitesimallysmallchanges<strong>to</strong>thebasisvec<strong>to</strong>rs. (detL)2n 1(L)2<br />

treme,buttheconverseisnottrue.DiederBasisb1;b2;:::;bnzugeordneteFormF(x1;x2;:::;xn):= Pni=1hbi;bjixixjnenntm<strong>and</strong>annExtremform???. ThispropertydoesnotdependonthechoiceofbasisforL.Everycriticallatticeislocallyex-<br />

wasimprovedbyH.F.Blichfeldt[Blich14]: Theupperboundn2n e+O(1)fortheHermiteconstantfromProposition4.2.3onpage62<br />

ofthespaceoftheballwithradius121overlaps<strong>and</strong>estimatesthisportionbyp2+o(1)nform (4.4) Thisboundtakesin<strong>to</strong>considerationthatintheproofofProposition4.2.3onlyaverysmallportion n21+n22nn e+o(n):


4.2.HERMITECONSTANTANDCRITICALLATTICE above.Forexample,102(6!)0;22;373.G.A.KabatianskyundV.I.Levenshtein[KaLe78] showedin1978,that: 65<br />

TheseimprovementsshowthattherstMinkowskiTheoremforconvexbodies(Proposition4.1.6, page59)isno<strong>to</strong>ptimalforballs.ForalowerboundfortheHermitconstantnwehave: n1;744 2en+o(n)0;1021n+o(n)<br />

E.Hlawka[Hlawka44]forballsS:=Sn(r).Theproofis,however,notconstructive,itshows ThislowerboundisobtainedbyapplicationofthefollowingtheoremofH.Minkowskiund n11+n2=n 2e+o(n)<br />

Proposition4.2.8(Minkowski,Hlawka) onlytheexistenceofcertainlattices;explicitconstructionsarenotknown. LetSRnhaveJordanvolumelessthan1.ThenthereisafulldimensionallatticeLRnwith Proof.See[GrLek87,Theorem1,Paragraph19,Chapter3]. detL=1<strong>and</strong>(L\S)nf0g6=;.<br />

Taken<strong>to</strong>gether,wehavefortheHermiteconstantthefollowingestimates: 2e+o(n)nn n <br />

Chapter6)thefollowingestimatesappear: Itisconjecturedthatngrowsmono<strong>to</strong>nicallyasafunctionofn.In[GrLek87](Paragraph38, e+o(n)<br />

Itisnotknowniflimn!1nnexists. 2eliminf 1 n!1nnlimsup n!1nn1<br />

TheHermiteconstants2;3;4;5wereobtainedinthesecondhalfofthe1900'sbyA.Korkine e<br />

6;7;8: undG.Zolotare[KoZo1872,KoZo1873,KoZo1877].In1935,H.F.Blichfeldt[Blich35]obtained<br />

n.21+n22n0;9070;8870;9070;8920;9070;9180;949 (n)n n 234 32 4 58 26 63 27 72 28 8<br />

ThelastrowshowstherelationndividedbyBlichfeldt'sestimate(4.4).Blichfeldt'sproofis complicated<strong>and</strong>wasimprovedbyG.L.Watson[Watson66]<strong>and</strong>N.M.Vetchinkin[Vetchin82].For n=9;10onlytheupperbounds(9)929<strong>and</strong>(10)1013210areknown.


66 isometry(withouttherequirement1=1theyareuniquelydeterminedup<strong>to</strong>isometry<strong>and</strong>scaling, Forn=2;3;:::;8thecriticallatticeofranknwith1=1isuniquelydeterminedup<strong>to</strong> CHAPTER4.SUCC.MINIMA,MINKOWSKI'STHEOREMS,HERMITECONSTANT<br />

Weshowthattheselatticeshaveabasisb1;b2;:::;bnwihthepropertythat: hencesimilarity).ThiswasshownbyE.S.Barnes[Barnes59]<strong>and</strong>N.M.Vetchinkin[Vetchin82]. hbi;bji=8>:1fori=j<br />

Thescalarproductshbi;bjideterminethecorrespondinglatticebasisb1;b2;:::;bnup<strong>to</strong>isometry. (4.5) 0forjijj>2 12forjijj2undi6=j Intheisometryclassoflatticebaseswiththescalarproductshbi;bjidescribedabove,thereis exactlyoneuppertriangularmatrix[b1;b2;:::;bn]withpositivediagonalelements.Thecorresdpondinglowertriangularmatrix[b1;b2;:::;bn]Thasthefollowingrowvec<strong>to</strong>rs:<br />

264 b b2 b3 b4 b51<br />

b6 b7 b8 3 75:= 21 12q34 p12q23 1<br />

p12323 1p3 p6 2<br />

0 64 q38 1p2 1p8 1p2 1p8q38 p6 01<br />

0 3p300<br />

LetL(n):=(b1;b2;:::;bn)forn8.Sincekb1k=1wehave11.Letb:=Pni=1tibiwith 0 0 0 0 0q23 1 7<br />

t1;:::;tn2Zbeanarbitrarynon-zerolatticevec<strong>to</strong>rinL(n).Wewish<strong>to</strong>showthatkbk1.From 5<br />

itfollwsfrom(4.5)thathbi;bji20;12;1 kbk2=DXni=1tibi;Xnj=1tjbjE=nXi=1tiDbi;Xnj=1tjbjE=nXi=1tinXj=1hbi;bjitj<br />

kbk2=nXi=1 tihbi;bii+nXj=1 j6=itjhbi;bji!=nXi=1 <strong>and</strong>kbik2=1,sothatweobtain | tihbi;bii+2Xj


4.2.HERMITECONSTANTANDCRITICALLATTICE Definition4.2.9(DeepHole) LetLbealattice.Thepointx2span(L)iscalledadeepholeofthelatticeL,when 67<br />

Definition4.2.10(LaminatedLattice) minfkxyk:y2Lg=max p2span(L)(minfkpyk:y2Lg)<br />

isadeepholeofthelatticeL(b1;b2;:::;bn). ThelatticeL(b1;b2;:::;bn+1)islaminatedwithrespect<strong>to</strong>L(b1;b2;:::;bn),when bn+1n+1(bn+1)2span(b1;b2;:::;bn)<br />

Figure4.2.2showsthisconstructionforthetwolatticesL(2)<strong>and</strong>L(3). L(b1;b2;:::;bi)islaminatedwithrespect<strong>to</strong>L(b1;b2;:::;bi1)<strong>and</strong>kbik=1fori=2;3;:::;n. ToconstructthelatticeL(n),wechoose(1;0;:::;0)2Rn<strong>and</strong>b2;b3;:::;bnsothat,respectively<br />

tiefesLochL(b1) b2qqq<br />

qqqqqqq<br />

tiefeL"ocher L(b1;b2)<br />

foralllatticesLofrankn. ThefollowinginequalityofH.Minkowskisharpenstheinequaltiy21n(detL)2n,<strong>and</strong>holds Figure4.2.2:ConstructionofL(2)<strong>and</strong>L(3)<br />

Proposition4.2.11(Minkowski'sInequality) ForeverylatticeLofrankn, nYi=1i(L)(n)n2detL Remark4.2.12 Sinceforacriticallattice(n)n2detL=n1<strong>and</strong>ingeneralQni=1in1,foracriticallattice Proof(ofProposition4.2.11).Leta1;a2;:::;an2Lbelinearlyindependentvec<strong>to</strong>rssuch that: 1=2==n.<br />

kaik=i fori=1;2;:::;n


68 Proposition3.2.16(page37)yields:thereexistsanuppertriangularmatrixT2Mn;n(Z)with: L1=L(a1;a2;:::;an)isasub-latticeofL.Choosebasisb1;b2;:::;bnforL,sothatforL2:=L CHAPTER4.SUCC.MINIMA,MINKOWSKI'STHEOREMS,HERMITECONSTANT<br />

(4.6) Forallb2L(b1;b2;:::;bn)<strong>and</strong>s=1;2;:::;n: b=2L(b1;b2;:::;bs1)=)kbks [b1;b2;:::;bn]T=[a1;a2;:::;am]<br />

Thus,fromb=2L(b1;b2;:::;bs1)<strong>and</strong>b2Litfollowsthatb=2span(b1;b2;:::;bs1),<strong>and</strong>since<br />

sothata1;a2;:::;as1;barelinearlyindependent.Fori=1;2;:::;nwelet TisuppertriangularwehavefromRemark3.2.17onpage38: span(b1;b2;:::;bs1)=span(a1;a2;:::;as1);<br />

<strong>and</strong>considerthelatticeL:=Lb1;b2;:::;bn.Claim: bi:=iXj=1i;jbj j<br />

(4.7) Letb:=Pni=1tibibeanarbitraryvec<strong>to</strong>rinLnf0g<strong>and</strong>s:=maxifijti6=0g.Then, 1L1<br />

sothatfrom(4.6)<strong>and</strong>ts6=0itfollowsthat: nXi=1tibi2=sXj=1 sXi=jtii;j!2kbjk2 2jsXj=1 sXi=jtii;j!2kbjk2 2s;<br />

FromdetL=detL Qni=1i,inequality(4.7)<strong>and</strong>thedenitionoftheHermiteconstant nXi=1tibi212ssXj=1 sXi=jtii;jkbjk!2=12ssXi=1tibi21<br />

weget: (detL)2nn 21<br />

Byraising<strong>to</strong>thepowern2<strong>and</strong>multiplyingbyQni=1iwegettheassertion: 11L2ndetL2n=n(detL)2n nYi=1i!2n<br />

nYi=1i(L)(n)n2detL


Thelatticedeterminantyieldsalowerboundontheproduc<strong>to</strong>fthesuccessiveminima: 4.3.GAUGEFUNCTIONSANDMINKOWSKI'STHEOREMS 69<br />

Proposition4.2.13(Minkowski'sSecondTheorem) ForeverylatticeLofrankn: nYi=1idetL<br />

(4.8) SinceL(a1;a2;:::;an)isasub-latticeofL: Proof.Leta1;a2;:::;anbelinearlyindependentlatticevec<strong>to</strong>rswithkaik=ifori=1;2;:::;n.<br />

Moreover,byHadamard'sinequality: nYi=1kaikdetL(a1;a2;:::;an)<br />

detL(a1;a2;:::;an)detL<br />

(4.9)<br />

4.3 >Fromtheestimates(4.8)<strong>and</strong>(4.9)wegettheclaimedboundQni=1idetL. GaugeFunctions<strong>and</strong>Minkowski'sTheorems <br />

convexbody. Minkowski'stheoremsforthelatticeZn<strong>and</strong>generalizethem.Werstdenethenotionofa Weintroducetheguagefunction(seeforexample[GrLek87,Siegel89])<strong>and</strong>formulatebothof<br />

Definition4.3.1(ConvexBody) AconvexbodyBRnisabounded,convex,openset. Theset@Bforaconvexbodyisthese<strong>to</strong>fallpointsp2RmnB,sothateveryneighborhoodofp containsapoin<strong>to</strong>utsideofB.<br />

y2@B 0qqq<br />

B<br />

Figure4.3.1:IllustrationoftheGaugeFunction x=y


70 Definition4.3.2(GaugeFunction) LetBRmbeaconvexbodywith02B.Theguagefunctionf:Rn![0;1)isdenedas CHAPTER4.SUCC.MINIMA,MINKOWSKI'STHEOREMS,HERMITECONSTANT<br />

follows: f(x):=8>:0ifx=0<br />

Aguagefunctionfissymmetricifforallx2Rnf(x)=f(x). ifx6=0,x=2@B,x=ywithy2@B<strong>and</strong>>0 1ifx2@B<br />

Weobtainthe`2-normwith <strong>and</strong>thesup-NormwithB:=(x1;x2;:::;xn)2Rn:x21+x2+:::x2n0forx6=0<strong>and</strong>f(0)=0 c)f(x+y)f(x)+f(y).<br />

Proposition4.3.4 Letf:Rn!Rbeafunctionwith: <br />

a)f(x)=f(x)for>0<strong>and</strong>x2Rn<br />

ThenthereexistsaconvexbodyBRnwithguagefunctionf. b)f(x)>0forx6=0 c)f(x+y)f(x)+f(y).<br />

Proof.See[Siegel89,Theorem7]mitB:=fx2Rmjf(x)


Proposition4.3.5 LetfbeaguagefunctionofaconvexbodyB.Then0isthecenterofB,if<strong>and</strong>onlyiffsymmetric. 4.3.GAUGEFUNCTIONSANDMINKOWSKI'STHEOREMS 71<br />

Proof.See[Siegel89,Theorems8<strong>and</strong>9]. Symmetricguagefunctionscorrespond<strong>to</strong>norms.TherstMinkowskitheoremcanbeexpressed intermsofguagefunctionasfollows(seeProposition4.1.6onpage59): <br />

Letf:Rn![0;1)beasymmetricguagefunctionfortheconvexbodyBRn.Ifvol(B)2n, thenthereexistsg2Znnf0gwithf(g)1. Proposition4.3.6(FirstMinkowskiTheorem)<br />

Proof.See[Siegel89,Theorems10<strong>and</strong>11]. Corollary4.3.7 Letf:Rn![0;1)beasymmetricguagefunctionfortheconvexbodyBRn,:=min x2Znnf0gf(x) <br />

Proof.For>0letB:=fx2Rnjf(x)g.Thenvol(B)=nV<strong>and</strong>for00,thatis,thereisan>0 (0)nV2n<br />

g2B0+exists.Thisyieldsacontradiction:<br />

Remark:Itfollowsfromthedenitionof0thatsince


72 of[Siegel89]withproof,inLectureIV. Proof.SeeParagraph9.1inChapter2of[GrLek87].FortheupperboundseealsoTheorem16 CHAPTER4.SUCC.MINIMA,MINKOWSKI'STHEOREMS,HERMITECONSTANT<br />

Let1;2;:::;nbethesuccessiveminimaofthefulldimensionallatticeLRnaccording Proposition4.3.9(SecondMinkowskiTheoremforGeneralLattice) <br />

B:=fx2Rnjf(x)


Chapter5<br />

Gauss'BasisReductionProcedure<br />

InthischapterwepresentGauss'basisreductionprocedurefortwodimensionallattices.The procedureisageneralizationoftheEuclideanAlgorithm.Wewillstudythereductionprocedure<br />

casesuchabasisalwaysexists. forthespecialcaseoftheEuclideannorm<strong>and</strong>thenconsiderthegeneralcaseofanarbitrarynorm. vec<strong>to</strong>rshavethelengths,respectively,ofthesuccessivemimima;however,inthetwodimensional WhileinChapter4.1wesawbyanexamplethatinthegeneralcasethereisnobasiswhose<br />

Weintroduceanotionofreducednessfortwo-vec<strong>to</strong>rbases: 5.1 ReducedBasis<br />

Anorderedlatticebasisa;b2Rnis(Gau)reducedwithrespect<strong>to</strong>normkk,when: Definition5.1.1(GauReducedBasis)<br />

Gram-Schmidtcoecients2;1=ha;bi Weconsiderthecasethatthenormisgivenbythescalarproductkxk=phx;xi.Forthe kakkbkkabkka+bk<br />

2;112() kak2wehave:<br />

Thusthebasisa;bisreducedif<strong>and</strong>onlyif: 2;10()kabkka+bk kbkkabk<br />

b)02;112 a)kakkbk 73


Incontrastwithaweightreducedbasis,werequirethatno<strong>to</strong>nlydoestheabsolutevalueof2;1 74 liebetween0<strong>and</strong>12,but2;1itselfdoes.Thiscanbeensuredbyreplacingbwithb.Figure5.1.1 CHAPTER5.GAUSS'BASISREDUCTIONPROCEDURE<br />

showstheGaussreducednessconditioninthecaseofthest<strong>and</strong>ardscalarproduct.Theangle 2;1=0 Rppppppppppppppppppp<br />

b Y2;1=12<br />

a-) kak=kbk<br />

ppppppppppppppppppp<br />

betweenthetwolatticevec<strong>to</strong>rsofthereducedbasisisbetween60<strong>and</strong>90: Figure5.1.1:ReducedBasisforSt<strong>and</strong>ardScalarProduct<br />

Since02;112<strong>and</strong>kakkbk: cos=ha;bi kakkbk=2;1kak<br />

0cos12 kbk<br />

reduced.Intheremainingcaseswehaveonlythereducedbasisa;b. with2;1=12,thena;abisalsoreduced.Ifa;bisreducedwithkak=kbk,thenb;aisalso Ifa;bisreducedwith2;1=0,thena;bisalsoareducedbasis.Similarly,ifa;bisreduced<br />

Proof.Withoutlossofgenerality,assumekakkbk.Theclaimsays: Proposition5.1.2 Forareducedbasisa;b2Rn,kak<strong>and</strong>kbkarethetwosuccessiveminimaofthelatticeL=Za+Zb. kakkra+sbk kbkkra+sbk 8r2Z<strong>and</strong>s2Znf0g 8(r;s)2Z2nf(0;0)g


5.1.REDUCEDBASIS Theseinequalitiesfollow<strong>to</strong>getherfromthefollowingproperties,whichweprovebelow. 75<br />

(5.1) kakkbk<br />

Wenowproveinequalities(5.1).ConsiderFigure5.1.2:Intheunionofthefour\quadrants" kakkrak kbkka+bk 8r2Znf0g 8,2Rwithjj;jj1<br />

rr<br />

ppppppp<br />

rrrrr<br />

rrrrr rrrr<br />

pppp r<br />

ppp 0 a pppppppp s<br />

-1 pp<br />

ba ba+b<br />

ppp ab a a+b<br />

rr<br />

indicatedbysmalldots,thenormtakesitsminimumonsomesubse<strong>to</strong>fthefourpointsab. Ofthelatticepointslyingonthethickerlines,thenormisminimizedinthemiddlepoints.That Figure5.1.2:ReducedBasisa;b<br />

is,fromtheGaussreducednessconditionsitiseasy(buttedious)<strong>to</strong>verifythat<br />

Bytheconvexityofthenorm,forjj1: kabkkakka+bk kabkkbk kabk<br />

minimizedontheboundary,henceonthethicklines. Thusthepointsabhavetheminimalnormonthethicklines.Byconvexity,thenormis kabkkabkkak kabkkabkkbk<br />

Anorderedlatticebasisa;b2Rniswellorderedreducedaccording<strong>to</strong>normkkwhen: Definition5.1.3(WellOrdered,ReducedBasis) <br />

kakkabkkbk


5.2 76 Algorithms CHAPTER5.GAUSS'BASISREDUCTIONPROCEDURE<br />

Wepresenttwoalgorithmsforobtainingareducedbasisforatwodimensionallattice:onefor generalnorms<strong>and</strong>oneforthespecialcaseoftheEuclideannorm.<br />

Algorithm5.2.1Gau'ReductionProcedurefortheEuclideanNorm 5.2.1ReductionAlgorithmfortheEuclideanNorm<br />

INPUT:LatticeBasisa;b2Rnwithkakkbk 1.WHILEj2;1j>12DO 1.2.IFkak>kbkTHENexchangea<strong>and</strong>b 1.1.[a;b]:=[a;b]d2;1c1 1 0<br />

OUTPUT:ReducedBasisa;b 2.b:=bsign(2;1) ENDwhile =2;10=<br />

correct. Algorithm5.2.1reducesbbyb:=bd2;1ca<strong>and</strong>thenexchangesaundb.Theoutputisclearly Algorithm5.2.1producesareducedbasisaccording<strong>to</strong>theEuclideannorm.Aniterationof<br />

Proposition5.2.1 Oninputa;bwithkakkbk,Algorithm5.2.1terminatesinatmost<br />

iterations. log1+p2kak 2+3<br />

Proof.SeeProposition4.4of[Schnorr94b]. 5.2.2ReductionAlgorithmforArbitraryNorm <br />

[Kaib94],whichdescribesecientimplementationsofStep1.1inthel1-<strong>and</strong>sup-norms. detailedanalysisappearsin[KaSchn96]byM.KaibundC.P.SchnorrinM.Kaib'sDissertation Gau'ReductionProcedurefortheEuclideanNormcanbegeneralized(Algorithm5.2.2).A


5.2.ALGORITHMS 77<br />

Algorithm5.2.2Gau'ReductionProcedureforArbitraryNorm INPUT:LatticeBasisa;b2Rnwithkakkbk 1.WHILEkbk>kabkDO<br />

1.3.exchangeaundb 1.2.IFka+bk


78 CHAPTER5.GAUSS'BASISREDUCTIONPROCEDURE


Chapter6<br />

LLLReducedLatticeBasis<br />

nwasproposedin1982byA.K.Lenstra,H.W.LenstraundL.Lovasz[LLL82].Itusesthe Thefollowingnotionofreducednessfororderedlatticebasisb1;b2;:::;bn2Rmofarbitraryrank Euclideannorm.<br />

particularweconsiderhowwellthelengthoftherstreducedbasisvec<strong>to</strong>rapproximatestherst 6.1 WeintroducethenotionofLLLreducedbasis<strong>and</strong>showpropertiesofanLLLreducedbasis;in Denition<strong>and</strong>Properties<br />

Definition6.1.1(LLLReducedBasis) thebasisb1;b2;:::;bn,<strong>and</strong>leti;j(1i;jn)bethecorrespondingGram-Schmidtcoecients. successiveminimumofthelattice.Letb1;b2;:::;bnbetheGram-Schmid<strong>to</strong>rthogonalizationof<br />

Anorderedlatticebasisb1;b2;:::;bn2RmiscalledLLLreduced(L3-reduced)withparameter, 14


T:span(b1;b2;:::;bn)!Rnbetheisometricmappingwith Ifthebasisconsistsoftwovec<strong>to</strong>rs,thenwhen=1weobtainaGaureducedbasis.Let 80 CHAPTER6.LLLREDUCEDLATTICEBASIS<br />

whereeiistheithunitvec<strong>to</strong>rinRm.Thebasismatrix[T(b1);T(b2);:::;T(bn)]forthelattice T(L),isometric<strong>to</strong>L,isanuppertriangularmatrix: T(bi)=kbikeifori=1;2;:::;n;<br />

[T(b1);:::;T(bn)]= 64 kb 1kkb2k... 0 0. 0kbnk 0.<br />

3 75 2 640 1 012;13;1 3;2 1 n;2 n;1<br />

0 . ......n;n1 0 1.<br />

375<br />

= 2 64 kb 1k ..."kbk1kk;k1kbk1k<br />

0 kbkk # ... kbnk <br />

3 Thebasisb1;b2;:::;bn2RmisLLLreducedwithif<strong>and</strong>onlyif: 75<br />

b)the22matricesonthediagonal a)thebasisb1;b2;:::;bnislengthreduced; "kbk1kk;k1kbk1k 0 kbkk #<br />

for1k


6.1.DEFINITIONANDPROPERTIES Proof.>FromPropertiesa)undb)ofLLLreducednessitfollowsthat: kbik2b) kbi+1k2+2i+1;ikbik2a) 81<br />

<strong>and</strong>thus: 14 kbi+1k2+14kbik2<br />

Theclaimfollowsbyinductionoverji. |{z} =1=kbik2kbi+1k2<br />

Corollary6.1.3 Letb1;b2;:::;bnbeLLLreducedwithparameter.Thenfor=1 14,kb1k2(n1)=21(L). <br />

Proof.ByProposition3.2.26,wehave1(L)minfkb1k;:::;kbnkg.Lettheminimumlength basisvec<strong>to</strong>rbebk.Thenkb1k2k1kbkk2byLemma6.1.2<br />

Bytakingsquarerootsweobtainthestatemen<strong>to</strong>ftheCorollary. n1(1(L))2 n1kbkk2because1<strong>and</strong>kn<br />

ThenextlemmaisageneralizationofProposition3.2.26. <br />

Lemma6.1.4 Letb1;b2;:::;bnbeabasisofthelatticeL.Thenfori=1;2;:::;n:<br />

j=1;2;:::;n.Let Proof.Therearelinearlyindependentvec<strong>to</strong>rsa1;a2;:::;an2L,sothatkajk=j(L)for ji=j;j+1;:::;nkbik<br />

min<br />

Hencethecoecientstikareintegers<strong>and</strong>thetikarereals.Let ak=nXi=1tikbi=nXi=1tikbifork=1;2;:::;n<br />

contrary.Thenbytheassumptionthat(k)Fromthelinear (k):=maxfi:tik6=0g<br />

a1;a2;:::;aj2span(b1;b2;:::;bj1);


sothata1;a2;:::;ajarelinearlydependent,whichyieldsacontradiction.Wethereforeobtain 82 CHAPTER6.LLLREDUCEDLATTICEBASIS<br />

2j2k=kakk2t2(k);kkb(k)k2kb(k)k2i=j;j+1;:::;nkbik2 min<br />

showsthatthelengthskbjkinanLLLreducedbasisare\rough"approximations<strong>to</strong>thesuccessive WhilethelowerboundonjinLemma6.1.4applies<strong>to</strong>arbitrarybases,thefollowingtheorem<br />

Proposition6.1.5(Lenstra,Lenstra,Lovasz1982) minimaj. Letb1;b2;:::;bnbeanLLLreducedlatticebasiswithparameter.Thenfor=1 a)1jkbjk2<br />

forj=1;2;:::;n 14:<br />

b)kbjk2 c)kbkk2j1kbjk2forkj 2jn1forj=1;2;:::;n<br />

Proof.9k,1kj,suchthatjkbkk.Itfollowsthat: 2jkbkk2+14k1<br />

jk+14k1 Xi=1kbik2 Xi=1ji! (bytheLLLreducednessproperties)<br />

kbjk2j1 1k+14k1 Xi=11i! (byLemma6.1.2)<br />

Weshow:<br />

Fork=1theinequalityisobvious.Fork2,since1=1434: 1k+14k1 Xi=11i1<br />

1k+14k1 geom.series34k1+14134k1 |{z} Xi=11i 134 =141 134=1


6.1.DEFINITIONANDPROPERTIES Thus, 2jkbkk2kbjk2j1 83<br />

<strong>and</strong>sowehaveprovedtherst<strong>and</strong>thirdassertions.ByLemma6.1.4thereisakj,sothat jkbkk.ItfollowsfromLemma6.1.2that: 2jkbkk2<br />

n+1kbjk2 k+1kbjk2 k+jkbjk2 (fromLemma6.1.2) (fromAssertion(c)withk=j) (fromLemma6.1.4)<br />

(sincekn<strong>and</strong>1)<br />

Corollary6.1.6 Letb1;b2;:::;bnbeanLLLreducedbasisofthelatticeL.Thenfor=1 14: <br />

a)kb1k2n1 b)nYi=1kbik2(n2)(detL)2 2(detL)2n<br />

Proof.WehaveQni=1kbik2=(detL)2.BythethirdAssertionofTheorem6.1.5:<br />

Itfollowsthat: kb1k2kbik2i1<br />

Thusweobtainpart(a):kb1k2n1 kb1k2n12n1nYi=1kbik2=(n2)(detL)2<br />

Theorem6.1.5. Part(b)followsfromQni=1kbik2=(detL)2<strong>and</strong>kbik2kbik2i1,thethirdAssertionof 2(detL)2n.<br />

Remark6.1.7 TheproofofCorollary6.1.3usesonlythefactthattheji+1;ij12,<strong>and</strong>notthefullpower <br />

ofweak-reducedness.Moreover,sincetheproofofCorollary6.1.3(a)onlyusescasek=1of Proposition6.1.5,thesameistruethere.However,theproofofCorollary6.1.3(b)usesthefull strengthofweakreducedness.Lovaszremarksthat<strong>to</strong>guaranteethatthenumbersoccurringinthe proceduredonotgrow<strong>to</strong>obigthefullpowerofweakreducednessseems<strong>to</strong>benecessary[Lovasz86].


6.1.1Two<strong>Application</strong>sofLLLReducedness 84 CHAPTER6.LLLREDUCEDLATTICEBASIS<br />

obtainsanLLLreducedbasiswith=34(the\`LLLalgorithm,"describedinSection6.2,has ProblemsdescribedinChapter1.Letusassumetheexistenceofanalgorithmthateciently latticedenedbythecolumnsofthematrixA: thisproperty).AttheendofSection4.1wecasttheSDAintermsofndingashortvec<strong>to</strong>rinthe Wereturnbriey<strong>to</strong>theSimultaneousDiophantineApproximation<strong>and</strong>SmallIntegerCombination<br />

A= 2 6400:::1n<br />

01:::02 10:::0 . ... . 1<br />

detL(A)="=Q.LetQ=2n(n+1)=4"n.UsingtheLLLalgorithm,wecanobtainavec<strong>to</strong>r 00:::0"=Q 3 75 b2L(A)suchthat Sinceb2L(A)thereexistintegersp1;p2;:::;pn;qsuchthat kbk22n=4(detL(A))1 n+1=2n=4"Q1 n+1 b= 2 64p2qn<br />

p 1q1<br />

Sincekbk1kbk2itfollowsthat pnqn q"Q . 3 75<br />

<strong>and</strong>moreoverq"Q",orinotherwords,jqjQ.Thusweobtainanapproximatesolution<strong>to</strong> thesimultaneousdiophantineapproximationproblem,inwhichthedenomina<strong>to</strong>rqisatmosta jpiqij"<br />

fac<strong>to</strong>rof2n(n+1)=4greaterthanoptimal. Recallthattheproblemis,givenrationals0;1;2;:::;n,<strong>and</strong>rationals";Q>0,<strong>to</strong>nd q0;q1;q2;:::;qn2Z,notall0,suchthat WemaycasttheSmallIntegerCombinationproblemintermsofalatticeproblemasfollows.<br />

<strong>and</strong>qiQfor1in.Traditionally,0=1,sowewillmakethatassumptionhere.Wedene nXi=0qii"<br />

B=26401:::n 0"=Q::: . ... 0 :::"=Q 0.<br />

3 75


6.2.THELLLREDUCTIONALGORITHM b=Bq,q2Zn+1,satisfykbk2",b6=0.Thenkbk1"<strong>and</strong>so ThendetB=0("=Q)n,or,sinceweareassuming0=1,wehavesimplydetB=("=Q)n.Let 85<br />

<strong>and</strong>qi"Q",orinotherwordsjqijQ. Choosing"=2n=4(detB)1 n+1,i.e.,Q=2(n+1)=4"1=n,wehavebyCorollary6.1.6(b)thatthe jq00++qnnj"<br />

6.2 LLLalgorithmwillndsuchavec<strong>to</strong>rb2L(B)inpolynomialtime.<br />

Givenanarbitrarylatticebasis,wepresentaprocedure<strong>to</strong>obtainanLLLreducedbasisforthe samelattice.Weanalyzetherunningtime<strong>and</strong>sizeofthecoecientsinthecalculations.We TheLLLReductionAlgorithm<br />

generalizetheprocedure<strong>to</strong>systemsErzeugendensysteme???{systemsofgenera<strong>to</strong>rs,inwhichthe vec<strong>to</strong>rsneednotbelinearlyindependent.<br />

Algorithm6.2.1,transforms,foragiven,14


86 Algorithm6.2.1LLLReductionAlgorithm INPUT:.LatticeBasisb1;b2;:::;bn2ZmCHAPTER6.LLLREDUCEDLATTICEBASIS<br />

1.k:=2 .Parameterwith14


6.2.THELLLREDUCTIONALGORITHM <strong>to</strong>bounddEndfrombelow,<strong>to</strong>getherwiththesimpleboundjj. proofofthefollowinglemmaissimilar<strong>to</strong>tha<strong>to</strong>fLemma6.2.1,butweuseMinkowski'sinequality 87<br />

terminatesafteratmost Lemma6.2.2 Letb1;b2;:::;bn2Rmbearealinputbasis<strong>and</strong>M:=maxikbik2.Thenfor


Lemma6.2.3 Theexchangebk1$bkresultsin:=k;k1<strong>and</strong>new:=new 88 CHAPTER6.LLLREDUCEDLATTICEBASIS<br />

a)kbnew k1k2=kbkk2+2kbk1k2 k;k1,suchthat:<br />

b)kbnew<br />

c)new=kbnewk1k2 kk2=kbkk2kbk1k2<br />

kk2<br />

d)new ijTk1i;jk=new1new<br />

k1k21 [i;j]Tk1i;jk01 10<br />

Proof.Weshowthefourclaims<strong>to</strong>gether(notethatbkundbk1aremutuallyorthogonal).<br />

b)Thisfollowsbecausetheproductkbkkkbk1kremainsunchangedbytheexchangebk1$bk. a)Thisfollowsfrombnew k1=bk+k;k1bk1.<br />

c)Since new=Dbnew kbnewk1E k;bnew k1k2=Dbk1;bk+bk1E<br />

weobtain: kbnew<br />

new=Dbk1;bkE+Dbk1;bk1E<br />

kbnew k1k2 =kbk1k2<br />

d)bnew k1=bk+bk1<strong>and</strong>according<strong>to</strong>thebackwardsexchangebk1=bnew kbnew k1k2<br />

thesetwoequalitiesitfollowsthat: hbk1;bki=hbnew k1;bnew kinew1newk+newbnew<br />

k1.From<br />

1


6.2.THELLLREDUCTIONALGORITHM Onethusobtains(onthediagonaldotsare1's;theotherentriesareall0): b1;b2;:::;bk2;bnew k1;bnew k;bk+2;:::;bm 89<br />

=hb1;:::;bmi[i;j]T1i;jn 2 64 . ..01 10 . .. 3 75<br />

=hbnew 1;:::;bnew mi2 6 4 . ..new1new 1 . .. 3 75[i;j]T1i;jn 2 64 . ..01 10<br />

=hbnew 1;:::;bnew mi[new ij]T1i;jn . .. 3 75<br />

Weboundfromabovethenumberofarithmeticoperationsforanexchangeby: Theclaimfollows. <br />

Proposition6.2.4<br />

Proof.TherstclaimfollowsfromLemma6.2.3.Thesecondclaimisfollowsfromthefactthat Anexchangebk1$bkrequiresatmostO(k)arithmeticoperations.Thelengthreductionofbk requiresatmostO(nk)arithmeticoperations. theoperationbk=bkbjalterstheGram-Schmidtcoecientsby: k;i:=k;ij;i fori=1;2;:::;j<br />

cangrowduringthecalculation. numbersgrow?Westudythisquestion<strong>and</strong>obtainanupperboundonhowlargethecoecients WehaveanalyzedthenumberofarithmeticstepsintheLLLAlgorithm.Howlargecanthe <br />

Foranintegerinputbasisb1;b2;:::;bn2Zm: Lemma6.2.5<br />

Moreover,Dj=detL(b1;b2;:::;bj)2=jQ a)Di1bi2Zm b)Dji;j2Z i=1kbik2isaninteger.


Proof.Thesecondclaimwillfollowfromtherst. 90 CHAPTER6.LLLREDUCEDLATTICEBASIS<br />

a)From[b1;b2;:::;bn]=hb1;b2;:::;bni[i;j]Titfollowsfor[ij]:=[i;j]1:<br />

Moreover,[ij]Tlike[i;j]isanuppertriangularmatrixwith1'sonthediagonal.Since Dbi;bjE=0forj=1;2;:::;i1itfollowsfrombi=bi+Pi1 hb1;b2;:::;bni=[b1;b2;:::;bn][ij]T<br />

hbi;bji=i1 Xt=1ithbt;bji forj=1;2;:::;i1 t=1itbt<strong>and</strong>ii=1that:<br />

Thesei1inequalitiesdenei1;i2;:::;i;i1.Thedeterminan<strong>to</strong>fthesystemofequations is: SinceDi16=0,itfollowsfromCramer'srulethat: Di1ij2Z Di1=det[hbj;bki]1j;ki1<br />

b)Bydenition(6.1)ofthedeterminantsDj=Qjs=1kbsk2wehave: Sincebi=bi+Pi1 j=1ijbj<strong>and</strong>b1;b2;:::;bn2Zm,theclaimDi1bi2Zmfollows. forj=1;2;:::;i1<br />

Dji;j=DjDbi;bjE<br />

Fromtherstclaim,bjDj12Zm,itfollowsthatDbi;Dj1bjE2Z.Wethusobtainthe kbjk2=Dj1Dbi;bjE=Dbi;Dj1bjE<br />

claimDji;j2Z.<br />

executionoftheLLLAlgorithm.Therstclaimholds,sinceforeachexchangebk1$bk: Thevaluemaxikbik2doesnotgrow,<strong>and</strong>thevalueminikbik2doesnotshrinkduringthe <br />

b)kbnew a)kbnew kk2kbold k1k2kbold<br />

k1k2<br />

Thesecondclaimfollowsfrom: b)kbnew a)kbnew kk21kbold k1k2kbold<br />

kk2


6.2.THELLLREDUCTIONALGORITHM Wegiveboundsforthecoecientsi;jtha<strong>to</strong>ccurintheLLLAlgorithm.Foranintegerinput basisb1;b2;:::;bn2Zmlet 91<br />

inthefollowing. M:=max i=1;2;:::;nkbik2<br />

Lemma6.2.6 IntheLLLAlgorithm,atthebeginningofStagekwith:=1 a)kbik2i+3<br />

1fori=1;2;:::;n:<br />

Proof.Weshowbothclaims: b)ji;jj2i+3 4Mj1 forj


92 CHAPTER6.LLLREDUCEDLATTICEBASIS Lemma6.2.7 DuringexecutionofStagek,forj=1;2;:::;k1,itisalwaysthecasethat:<br />

Proof.Instagekthelengthreductionstep jk;jj2k+3 4M94k1<br />

yields,forj=1;2;:::;k1: k;j:=k;jdk;ici;j bk:=bkdk;icbi<br />

wecanboundthenewMkfromaboveby: Eachofthek1operations(6.4)changesMk:=maxj=1;2;:::;k1jk;jjsothatfromdk;icMk+12 ji;jj1=2 |{z}<br />

(6.5) ByLemma6.2.6atthebeginningofStagek Mnew kMold<br />

Mkrk+3 k+12Mold k+1232Mold<br />

4Mk1k+14<br />

negligible).Thus: >Fromthebound(6.5)thevalueMkgrowsbyatmostthefac<strong>to</strong>r32k1(thesumm<strong>and</strong>14becomes<br />

duringStagek.Theclaimfollows. jk;jj2322(k1)k+3 4Mk1=k+3 4M94k1<br />

isnolongerstable.Forj>k,fromthetwoclaimsofLemma6.2.6wecanonlygetthebound Atthestar<strong>to</strong>fStagekthevaluesi;jwithj>kcanbeverylarge,inwhichcasetheprocedure <br />

>Fromtheinequality ji;jj2n+3<br />

ji;jj2kbik2 4Mj:<br />

fromDj=Qji=1kbik2,<strong>and</strong>fromLemma6.2.6onPage91wehave: kbjk2;<br />

ji;jj2kbik2Dj1 Djn+3 4MDj1n+3 4Mj


6.2.THELLLREDUCTIONALGORITHM TheLLLAlgorithmwithiterativeorthogonalization(Algorithm6.2.2)avoidsthevaluesi;jwith j>k,<strong>and</strong>calculatesinStagekonlywiththevaluesi;jwhere1j


Proposition6.2.8 Oninputanintegerlatticebasisb1;b2;:::;bn2ZmwithM:=maxi=1;2;:::;nkbik2,Algorithm 94 CHAPTER6.LLLREDUCEDLATTICEBASIS<br />

6.2.2performs<br />

arithmeticstepsonrationalsji;jjqn+3 On2m1+nlog1=M<br />

oftheserationalsareboundedby <strong>and</strong>vec<strong>to</strong>rsbiwithkbik2n+3 4M.Theabsolutevaluesofthesenumera<strong>to</strong>rs<strong>and</strong>denomina<strong>to</strong>rs 4Mj<strong>and</strong>jk;jjqn+3 4M94k1,kbjk2M,<br />

Proof.ByLemma6.2.1: Mn12sn+3 494n<br />

Sincethestagekis2atthebeginningofthealgorithm<strong>and</strong>n+1attheend,itisclearthat #Exchangesn2log1=M=On2log2M<br />

exchange.EveryiterationrequiesatmostO(nm)arithmeticsteps.Theupperboundonthe Everyiterationwithanexchange<strong>and</strong>stagereductionyieldsatmos<strong>to</strong>neiterationwithoutan numberofstepsfollows. #Iterationsn1+2#Exchanges<br />

BythesecondclaimofLemma6.2.6<strong>and</strong>byLemma6.2.7wehave: ByLemma6.2.5thedenomina<strong>to</strong>rsofthenumbersijforj


Chapter7<br />

Babai'sApproximation<strong>to</strong>CVP<br />

InthischapterwepresentBabai'sapplicationoftheLLLalgorithm<strong>to</strong>theproblemofapproximatingtheclosestlatticevec<strong>to</strong>r<strong>to</strong>agivenpointinRn.Weshowanapplication<strong>to</strong>theinhomogeneous<br />

7.1 Diophantineapproximationproblem.Allthematerialinthischapterappearsin[Babai86].<br />

Considerthefollowing\inhomogeneous"versionoftheshortestvec<strong>to</strong>rproblem: ApproximateCVP<br />

ClosestVec<strong>to</strong>rProblem(CVP) Definition7.1.1<br />

Find:x2Rnsuchthatkxbkisminimized. Given:LRnoffullrank<strong>and</strong>x2Rn<br />

vec<strong>to</strong>rproblem,CVPcanbeapproximated<strong>to</strong>withinafac<strong>to</strong>rofninpolynomialtime[Kannan83]. CVPcanbesolvedinO(ncn)arithmeticoperations;moreover,givenanoraclefortheshortest hard[DKS98].Inthischapterweconsideronlythe2-norm. CVPisNP-hardforanynorm;indeed,approximatingCVPwithinafac<strong>to</strong>rof2lg1"nisNP-<br />

ApproximateCVP Definition7.1.2 Given:basis(b1;b2;:::;bn)forlatticeLoffullrank<strong>and</strong>x2Rn Find:w2Lsuchthatkxwkcnkxuk,whereuisanearestneighborofxinL<strong>and</strong> cndependsonlyonn. 95


96Let=34.AnLLL-reducedbasiswith=34satises<br />

CHAPTER7.BABAI'SAPPROXIMATIONTOCVP<br />

1p2kbk1kkbkk,k=2;:::;n jijj12<br />

Babai[Babai86]presentstwosimpleproceduresforsolvingtheapproximateCVPyielding,respectively,cn=2n=2<strong>and</strong>cn=1+2n(9=2)n=2.<br />

w=Pni=1ibi. Solvetheequationx=Pni=1ibi,forreals1;2;:::;n.Fori=1;:::;n,seti=dic.Output Procedure1:RoundingO Procedure2:NearestPlane 1.LetU=Pn1<br />

2.Find(detailsexplainedbelow)v2Lsuchthatthedistancefromx<strong>to</strong>U+visminimal.Let L0=L\U. L0=L\span(b1;b2;:::;bn1)bethe(n1)-dimensionalsublatticeofLcontainedinU: i=1RbidenotethelinearsubspaceofRnequal<strong>to</strong>span(b1;b2;:::;bn1).Let<br />

3.Recursivelyndy2L0nearx0v. 4.Outputw=y+v. x0betheorthogonalprojectionofxon<strong>to</strong>U+v.<br />

1;2;:::;n2Rn.Let=dnc.Letx0=Pn1 ForStep2,writexasareallinearcombinationoftheorthogonalizedvec<strong>to</strong>rsx=Pni=1ibi,where<br />

solution. inLisinU+v.Inthiswaythealgorithm\makesamistake"<strong>and</strong>soweonlygetanapproximate cose<strong>to</strong>fUintersectingLnearest<strong>to</strong>x,itisnotnecessarilythecasethatthenearestneighborofx i=1ibi+bn.NotethatalthoughU+visthe<br />

Proposition7.1.3 IfBisLLL-reducedwith=34,thenProcedureRoundingOndsalatticepointwnearest<strong>to</strong>x<br />

latticeparallelepipeds,whichisofindependentinterest. withinafac<strong>to</strong>rof1+2n(9=2)=2.<br />

Proposition7.1.4 Proof.TheproofofthePropositionusesthefollowingresultregardingtheshapeofLLL-reduced<br />

Letb1;b2;:::;bnbeLLL-reduced.Fork=1;:::;n,letUk=Pj6=kRbj,<strong>and</strong>letkbetheangle betweenbk<strong>and</strong>Uk.Thensink(p2=3)n. Proof.See[Babai86].


Since 7.1.APPROXIMATECVP 97<br />

anequivalentstatemen<strong>to</strong>fProposition7.1.4isthat,forallm2Uk, sink=min m2Ukkmbkk<br />

(7.1) kbkk(9=2)n=2kmbkk kbkk<br />

(7.2) Letdn=(9=2)n=2.Letwbetheoutpu<strong>to</strong>fProcedureRoundingO.Letuswrite<br />

wherefori=1;:::;n,jij12becauseweroundedtherealcoecientsinx=Pibi<strong>to</strong>getw. wx=nXi=1ibi<br />

points,'1;'2;:::;'n2Z. Lemma7.1.5 Letubeanearestlatticepoint<strong>to</strong>x.Writeuw=Pni=1'ibi.Sinceu<strong>and</strong>warebothlattice<br />

Proof.Assumeu6=w.Letksatisfy kuwk2ndnkuxk.<br />

Then k'kbkk=max 1jnfk'jbjkg:<br />

But (7.3) kuwknXj=1k'jbjknk'kbkk<br />

Letting ux=(uw)+(wx)=nXi=1('i+i)bi<br />

wehavem2Uk,<strong>and</strong>wecanwritem='k+kXj6=k('j+j)bj<br />

1<br />

ux=nXi=1('i+i)bi=('k+k)(bkm)


98 'k2Znf0gwehave: Thus,byInequality7.1,<strong>and</strong>thefactthatthei'sareallboundedinabsolutevalueby12,<strong>and</strong> CHAPTER7.BABAI'SAPPROXIMATIONTOCVP<br />

Thus, kuxk=j'k+kjkbkmkj'k+kjkbkk dnj'kj 2dnkbkk<br />

Frominequalities7.3<strong>and</strong>7.4wehave (7.5) (7.4) kuwknj'jkbkkn2dnkuxk kuxk2dnj'kjkbkk<br />

Bythetriangleinequality<strong>and</strong>Lemma7.1.5wehave: ThiscompletestheproofoftheLemma.<br />

ThiscompletestheproofofProposition7.1.3. kxwkkxuk+kuwkkxuk(1+2ndn):<br />

7.2 InhomogeneousDiophantineApproximation <br />

RecallthedenitionoftheinhomogeneousDiophantineapproximationproblemfromChapter1: InhomogeneousDiophantineApproximation Given:1;2;:::;n;1;2;:::;n,";Q>0 Findintegersp1;p2;:::;pn;qsuchthat <strong>and</strong>00thereexistintegersp1;p2;:::;pn;qsuchthatq>0<strong>and</strong><br />

2.Thereexistintegersu1;u2;:::;unsuchthatPni=1uiiisanintegerwhilePni=1uiiisnot. jqipiij":


7.2.INHOMOGENEOUSDIOPHANTINEAPPROXIMATION theoremsaysthatthereexistp1;p2;:::;pn;qsuchthatqipi=ifori=1;:::;i.This Suppose1;2;:::;n<strong>and</strong>1;2;:::;nareallrationals.ThentherstchoiceinKronecker's 99<br />

denomina<strong>to</strong>rqatmost3ntimeslargerthanoptimal,yieldinganerrorofatmost3n"[Babai86]. hard[EmBoas81].Wenextdescribeapoylnomialtimealgorithmdue<strong>to</strong>Babaiforobtaininga polynomialtime(see,e.g.,[Frumkin76]).Tondaleastcommondenomina<strong>to</strong>rq=q(")isNP-<br />

isaspecialkindoflineardiophantineequationwhich,withoutaboundonq,canbesolvedin<br />

Proposition7.2.1 Given1;2;:::;n;1;2;:::;n,<strong>and</strong>">0,allinQ,intimepolynomialinthelengthsofthe inputsonecanndeither 1.integersp1;p2;:::;pn;qs.t. jqipiijcn"<br />

2.aproofthatq=q(")isinnite(nosolutionexists). wherecn=4pn2n=2,or: jqjcnq(")<br />

wecanassumetheseareallpositive). DeneQ<strong>to</strong>betheproduc<strong>to</strong>fthedenomina<strong>to</strong>rsofthei,i=1;2;:::;n(withoutlossofgenerality Proof.Replace"bysuchthat"2(=2;]<strong>and</strong>=2ifori2Z,sothat`()


100 ApproximatesatisesInequality7.6,ifany.Ifnosuchsexiststhenq()isinnite.Moreover, sinceifthereisasolutionwitherroratmost"thenthereisasolutionwitherroratmost>", CHAPTER7.BABAI'SAPPROXIMATIONTOCVP<br />

ifnosuchsexiststhenq(")isinnite. equalities7.6with"replacedby.Then (q() p2;p2q()].Letu=Pni=1pibi+qibi+1,wherep1;p2;:::;pn;qsatisfytheconstraintsinIn-<br />

Assumenowthatq(")isnite.Thenwehavejustarguedthatsoisq().Lets=2i2<br />

kuxk1=max maxf1;jqj ifjqipiij;jqj sg sg<br />

Now,sq() p2<strong>and</strong>jqjq(),so<br />

Thuskuxk1p2,whence jqj sq() q()=p2=p2<br />

TheNearestPlaneprocedureisshownin[Babai86]<strong>to</strong>obtainanapproximation<strong>to</strong>within2d=2of theclosestvec<strong>to</strong>rinlatticesofdimensiond.Thus, kuxk2p2n<br />

(7.7) (7.6) (7.8) kwxk22n+1 =2pn2n=2<br />

2kuxk2<br />

Moreover, 2pn212<br />

(7.9) so kwxk2kwxk1=max ifjqipiij;jqj<br />

jqipiij2pn2n=2


Chapter8<br />

BreakingtheLinearCongruential Genera<strong>to</strong>r<br />

WepresentageneraltechniqueofFrieze,Hastad,Kannan,Lagarias,<strong>and</strong>Shamirforreconstructing truncatedintegervariablessatisfyinggivenlinearcongruences,<strong>and</strong>theapplicationofthistechnique<strong>to</strong>breakingpseudo-r<strong>and</strong>omgenera<strong>to</strong>rsbasedonlinearcongruentialsequences.Thematerial<br />

8.1 forthischapterappearsin[FHKLS88].<br />

sequencegenera<strong>to</strong>r<strong>to</strong>D.H.Lehmerin1948.Thegenera<strong>to</strong>rhasfourparameters: Knuth[Knuth80]attributestheideaofusingalinearcongruentialsequenceasapseudo-r<strong>and</strong>om LinearCongruentialSequences<br />

2.themultipliera0; 1.theseedx00;<br />

Thecorrespondinglinearcongruentialsequenceis: 3.theincrementc0; 4.themodulusM>x0;a;c.<br />

shorter,butthecomputationalcostislower.Otherrestrictionsonthechoiceofparameters arestudiedextensivelyin[Knuth80],wherethereisalsoalong<strong>and</strong>interestingdiscussionofthe Thesequencealwayscycles(wewantcycles<strong>to</strong>belong).Ifc=0thentheperiodisgenerally xn+1=(axn+c)modM<br />

ofpseudo-r<strong>and</strong>omsequencesintermsofpolynomialtimestatisticaltests[Yao82].Analternative \right"denitionofapseudor<strong>and</strong>omsequence.ThisdiscussionmotivatedYao'smoderndenition 101


102 denition,involvingunpredictability,hadpreviouslybeenproposedbyBlum<strong>and</strong>Micali[BluMi]. Forcompleteness,thesetwodenitions,provedequivalentbyYao[Yao82],arestatedinSection8.4. CHAPTER8.BREAKINGTHELINEARCONGRUENTIALGENERATOR<br />

of)theparametersaresecret. predictabilityofthesequencewithhighaccuracyevenforrelativelysmallvaluesofk[Boyar82]. Ifa;c,<strong>and</strong>Mareunknown,butallofthebitsofxi,1ikareknown,thenBoyarshowed Asusual,weassumethatthemethodforgeneratingthesequenceisknown;thevaluesof(some<br />

accordingly.BoyarshowsthatatmostO(logM)disagreementscaneveroccur. extrapolatedsequencediersfromtheactualsequence,thentheguessedvaluesarecorrected Inparticular,shendsba;bc,<strong>and</strong>cM,consistentwiththeavailabledata,<strong>and</strong>extrapolates.Ifthe<br />

O(n222`=k2)steps[Knuth80]. known,1ik,thenKnuthgivesanattackwhichusuallyreconstructsa,c,<strong>and</strong>theseedin Ifcisunknown,M<strong>and</strong>aareknown,<strong>and</strong>thehighorderbitsyiofeachxi,1ik,aregiven Ifa<strong>and</strong>careunknown,M=2nisknown,<strong>and</strong>then`highorderbitsyiofeachxiare<br />

asdata,thenanattackdue<strong>to</strong>Frieze,Hastad,Kannan,Lagarias,<strong>and</strong>Shamir[FHKLS88]yields: apolynomialtimereconstructionorpredictionprocedure(seebelow),proved<strong>to</strong>besuccessful<br />

anearlyalwayspolynomialtimereconstructionprocedureforallMprovided`(yi) onnearlyallproblemsinwhichsucientlymanybitsofdataareknown<strong>to</strong>permitunique reconstructioninformation-theoretically,providedMissquare-free;<br />

Whileuniquereconstructionofx0maybepossibleifc=0,thecasec6=0isdierent.Ifweset givenonlythreesamples(i.e.,k=3). `(xi)13,<br />

x0i=xi+1xi<strong>and</strong>y0i=yi+1yi,thenx0isatisestherecurrence<br />

x0iforanydyetbotharelinearcongruentialsequences;indeed,forsmalldthesequencesfxig<strong>and</strong> <strong>and</strong>y0iisessentiallyatruncatedversionofx0i.Thisisbecausexi<strong>and</strong>xi+dyieldthesamesequence x0i+1ax0i(modM)<br />

fxi+dgwillusuallyhavethesamehighorderbits.Inthiscasefuturevaluesofthegenera<strong>to</strong>r maybepredictedwithgreataccuracy,evenifexactreconstructionisnotpossible. unknown,inthatinpolynomialtimetheyobtainavaluecM<strong>and</strong>aheuristicargumentthatcM decreasesquickly<strong>to</strong>M,<strong>to</strong>getherwithatechniquethat,givenM,ndsa. Joux<strong>and</strong>Stern[JoSt94]extendtheresultsof[FHKLS88]<strong>to</strong>thecaseinwhicha<strong>and</strong>Mare<br />

Friezeetal.giveageneralresult,ofwhichthereconstructionofthecongruentialgenera<strong>to</strong>risa 8.2 simplespecialcase.Thegeneralproblemis: AGeneralReconstructionResult


8.2.AGENERALRECONSTRUCTIONRESULT Definition8.2.1(ReconstructionProblem) Given:aijfor1i`<strong>and</strong>1jk,cifor1i`,M,<strong>and</strong>`modularequations103<br />

kXj=1aijxjcimodM bitsofthexj,specically,blocksofconsecutivebinarydigits: (8.1) where0xj


Thesystem 104 Proposition8.2.2(Frieze,Hastad,Kannan,Lagarias,<strong>and</strong>Shamir) CHAPTER8.BREAKINGTHELINEARCONGRUENTIALGENERATOR<br />

kXj=1aijxjcimodM \smallsolution).Iftheaij,ci,<strong>and</strong>Mareknown,thenthereexistsapolynomialtimealgorithm hasatmos<strong>to</strong>nesolutionx2Zksatisfyingkxk2Mk(L(A;M))12(k=2)1(i.e.,atmos<strong>to</strong>ne 1i`<br />

thateitherndsxorprovesthatnosuchxexists. Remark8.2.3 LetusgivesomeintuitionfortheapplicationofProposition8.2.2.Ifthexiarelargebutwe knowthemostsignicantbits,thenwecanrewritethemodularrelationsinsuchawaythat thenewunknownsare\small."Assumethishasalreadybeendone.IfL(A;M)hasverysmall kthsuccessiveminimumk,thentheclaimedpolynomialtimealgorithmcanndrelativelylarger<br />

Aischosenatr<strong>and</strong>om(asinmostcryp<strong>to</strong>graphicapplications).Muchdependsonthestructureof oftheproofofapplicabilityofthetheoremisinanalyzingtheexpectedvalueofk(L(A;M))when bitsoftheoriginalxi's,sothatwhenwerewritetheequationstheunknownsaresmaller.Thebulk smallerinorderforthealgorithm<strong>to</strong>besure<strong>to</strong>ndthem.Intuitively,thismeansthatweneedmore unknowns(Mkislargerwhenkissmaller).If,however,k(L(A;M))islarge,thentheximustbe<br />

Schnorr[LLS90]: M.See[FHKLS88]fordetails. Proof.(Sketch)Theupperboundkisfoundviathefollowingresul<strong>to</strong>fLenstra,Lenstra,<strong>and</strong><br />

<strong>and</strong>thekthsuccessiveminimumoftheprimalisboundedabovebyk2. Thatis,forfulldimensionallatticesinRk,theproduc<strong>to</strong>ftherstsuccessiveminimumofthedual 1kk2<br />

1.ReducethebasisforL=L(A;M)usingtheLLLalgorithm<strong>to</strong>getmodularrelationswth ThestructureoftheproofofProposition8.2.2isasfollows:<br />

2.Usethe(assumed)sizeconstraintsonthexj's<strong>to</strong>transformthesemodularequations<strong>to</strong> smallcoecients.<br />

Wenowexplainthesestepsinmoredetail. 3.Solvetheseequationsovertheintegers<strong>to</strong>recovertheexactvaluesofx1;x2;:::;xk. equationsovertheintegers.<br />

GLk+`(Z)suchthatVBisinHermiteNormalFormwithnon-zerorowsz1;z2;:::;zk.Thus, Next,deneX2Zk(k+`)suchthatX(VB)=Z. z1;z2;:::;zkisabasisforthelatticeL(A;M).LetZbethekkmatrixwithrowsz1;z2;:::;zk. Fortherststep,startingwiththematrixBdenedabove,inpolynomialtimendV2


obtainaunimodularmatrixU2GLk(Z)<strong>and</strong>thereducedbasisw1;w2;:::;wksuchthat,ifWis 8.2.AGENERALRECONSTRUCTIONRESULT ApplytheLLLalgorithmwith=34<strong>to</strong>thek-dimensionallatticewithbasisz1;z2;:::;zk<strong>to</strong> 105<br />

LetY=UXV.ThenW=YB. thekkmatrixwithrowsw1;w2;:::;wkwehaveUZ=W<strong>and</strong>,bythereducednessofW,<br />

Whathashappened<strong>to</strong>ourinitialse<strong>to</strong>fequations?Wehad: kwik22k=2k 1ik<br />

<strong>and</strong>so Y(Bx)YbCmodM BxbCmodM<br />

<strong>and</strong>hence<br />

LetC0=YbCmodM(sinceY<strong>and</strong>bCareknownwecaneasilyndC0).ThensinceW=YBwe havethemodularequalities (YB)xYbCmodM<br />

(8.3)Letuswritex=(x1;x2;:::;xk)T.Forthesecondstepofthealgorithm(transformation<strong>to</strong>a systemofequationsoverZ)observethat,for1i`, WxC0modM<br />

jc0ij=kXj=1wijxj =jhwi;xij<br />

=M2 kwikkxk


106 Corollary8.2.4 Lets0=logk+k2+12logk+1.Thesystem CHAPTER8.BREAKINGTHELINEARCONGRUENTIALGENERATOR<br />

hasatmos<strong>to</strong>nesolutionxinwhichthes0mostsignicantbitsofeachxjarespecied. kXj=1aijxjcimodM 1i`<br />

Proof.Writexi=x(1) i+x(2) iwherex(1) x(2) iM2s0=M1 iaretheknowns0mostsignicantbitsofxi<strong>and</strong><br />

x(1) SincethisisimpliestheboundonkxkassumedinProposition8.2.2wecansubstituteintheknown i<strong>and</strong>thenapplytheProposition. k2(k=2)1k1=2<br />

Remark8.2.5 WecanactuallyapplythealgorithmofProposition8.2.2(<strong>and</strong>theCorollary)withoutknowingk. <br />

checkifthenumberofbitsknownineachxjisatleast Wefollowthestepsofthealgorithmuntilweobtainthereducedbasisw1;w2;:::;wk.Wethen<br />

Thisprovidesasucientconditionforthealgorithm<strong>to</strong>work,becausemaxilog2kwiklog2k since,bydenitionofthekthsuccessiveminimum,foranybasisb1;b2;:::;bkforL(A;M), max iflog2kwikg+12logk+1<br />

8.3 maxikbik2k.<br />

Forreasonsdiscussedabove,wemayrestrictdiscussion<strong>to</strong>thecaseinwhichc=0.Thusthereare unknownsx1;x2;:::;xkthatsatisfythecongruencesxi+1aximodM,<strong>and</strong>wearegiventhe <strong>Application</strong><strong>to</strong>theLinearCongruentialGenera<strong>to</strong>r<br />

L=L(A;M)<strong>to</strong>bethese<strong>to</strong>fallintegerlinearcombinationsoftherowsofthefollowingmatrix: highorderbitsofthexi's. Itiseasy<strong>to</strong>seebyinductionthatai1x1xi0modM,for2ik.Wedenethelattice 264 ak10 M a2 a 01::: 0 .:::1 0<br />

NotethatdetL=M.<br />

3 75


Forsquare-freeM>c(";k)thereisanexceptionalsetE(M;";k)ofmultipliersofcardinality Proposition8.3.1(Frieze,Hastad,Kannan,Lagarias,<strong>and</strong>Shamir) 8.4.MODERNDEFINITIONSOFPSEUDO-RANDOMNESS 107<br />

where jE(M;";k)jM1"suchthatforanymultipliernotinE(M;";k)thefollowingistrue.Thexiare uniquelydeterminedbyknowledgeofthe(1=k+")logM+c(k)leadingbitsofallfxi:1ikg,<br />

Furthermore,thereisanalgorithmwhichrunsintimepolynomialinlogM+kwhichndsthexi. c(k)=k2+(k1)log3+72logk+2:<br />

Remark8.3.2 2.a=1isclearlyexceptional(althoughextrapolationiseasyinthiscase!). 1.Thenumberofbitsneededforreconstructionisalmos<strong>to</strong>ptimaloninformation-theoretic grounds.<br />

3.Theproofactuallyshowsthat"approaches 4.Thexjaretreatedasindependentunknowns. loglogMasMapproachesinnity. c<br />

5.Forthespecialcasek=3[FHKLS88]areable<strong>to</strong>showthat,forany">0,forallM, knowledgeof(13+")logM+c(k)leadingbitsofx1;x2,<strong>and</strong>x3,allowsrecoveryinpolynomial<br />

8.4timeforallmultipliersaexceptase<strong>to</strong>fcardinalityc(")M1"=2.<br />

<strong>and</strong>Micali[BluMi]<strong>and</strong><strong>to</strong>Yao[Yao82]. Wepresenttwoequivalentdenitionsofpseudor-r<strong>and</strong>omsequences,due,respectively,<strong>to</strong>Blum ModernDenitionsofPseudo-R<strong>and</strong>omness<br />

Definition8.4.1(ensemble) thateachSkisaprobabilitydistributiononk.Ther<strong>and</strong>omensembleR=fRkgisthesequence Letkdenotethese<strong>to</strong>fallbinarystringsoflengthk.AnensembleSisasequencefSkgsuch<br />

Apolynomial-sizefamilyofcircuitsisasequenceofcircuitsC=fCkgsuchthatforsomepositive ofuniformdistributionsi.e.,Rk(x)=2kforallx2k.<br />

integerd,Ckhasatmostkinputs<strong>and</strong>atmostkdgates. Definition8.4.2(polynomial-sizefamilyofcircuits)<br />

Definition8.4.3(predictingcollection) Apredictingcollectionisapolynomial-sizefamilyofcircuitsCsuchthateachCkhasi


108 Definition8.4.4(unpredictablebyC) Fori


Bibliography<br />

[Ajtai96] [Adel83] 412 M.Ajtai,GeneratingHardInstancesofLatticeProblems,Proceedings28th L.Adleman,OnBreakingGeneralizedKnapsackPublicKeyCryp<strong>to</strong>systems,Proceedings15thAnnualACMSymposiumonTheoryofComputing,1983,pp.402{<br />

[Ajtai98] M.Ajtai,TheShortestVec<strong>to</strong>rProbleminL2isNP-HardforR<strong>and</strong>omizedReductions,<strong>to</strong>appear,Proceedings30thAnnualACMSymposiumonTheoryotrier.de/eccc-local/Lists/TR-1996.htmtronicColloquiumonComputationalComplexityTR96-007,http://www.eccc.uni-<br />

AnnualACMSymposiumonTheoryofComputing,1996,pp.99{108Elec-<br />

[AjtDw97] M.Ajtai,C.Dwork,APublic-KeyCryp<strong>to</strong>systemwithAverage-Case/Worst-Case Equivalence,Proceedings29thAnnualACMSymposiumonTheoryofComputing, Computing,1998<br />

[Babai86] L.Babai(1986):OnLovasz'LatticeReduction<strong>and</strong>thenearestLattice 1997;seealsoElectronicColloquiumonComputationalComplexityTR96-065, http://www.eccc.uni-trier.de/eccc-local/Lists/TR-1996.html<br />

[Barnes59] Arithmetica,B<strong>and</strong>5,Seiten205-222. E.S.Barnes(1959):TheContructionofperfect<strong>and</strong>extremeFormsII,Acta PointProblem,Combina<strong>to</strong>rica,B<strong>and</strong>6,Seiten1{13.<br />

[BeWe93] [BaKa84] Approach<strong>to</strong>commutativeAlgebra,GraduateTextsinMathematics,B<strong>and</strong> Th.BeckerundV.Weispfennig(1993):Gr"obnerBases|acomputational rithm,TechnischerReport,Carnegie-Mellon-Universit"at(USA). A.BachemundR.Kannan(1984):<strong>Lattices</strong><strong>and</strong>theBasisReductionAlgo-<br />

[Blich14] 141,Springer-Verlag,Berlin/Heidelberg.<br />

[Blich29] some<strong>Application</strong>s,TransactionoftheAmericanMathematicalSociety,B<strong>and</strong> 15,Seiten227{235. H.F.Blichfeldt(1914):AnewPrincipleintheGeometryofNumberswith H.F.Blichfeldt(1929):TheMinimumValueofquadraticForms<strong>and</strong>the closetPackingofSphere,MathematischeAnnalen,B<strong>and</strong>101,Seiten366-389. 109


110 [Blich35] H.F.Blichfeldt(1935):TheminimumValueofpositiveQuadraticForms insix,seven<strong>and</strong>eightVariables,MathematischeZeitschrift,B<strong>and</strong>39,Seiten BIBLIOGRAPHY<br />

[BluMi] 1{15.<br />

[BoHi89] M.Blum<strong>and</strong>S.Micali(1984):How<strong>to</strong>GenerateCryp<strong>to</strong>graphicallyStrong SequencesofPseudo-R<strong>and</strong>omBits,SIAMJ.Computing,Vol.13,No.4, pp.850{864<br />

[Boyar82] plexityClasses,inR<strong>and</strong>omness<strong>and</strong>Computation,Vol.5,S.MicaliEdi<strong>to</strong>r,Ad-<br />

vancesinComputingResearch,JAIPress,Greenwich,pp.1{26. J.Boyar(1982):InferringSequencesProducedbyPseudo-R<strong>and</strong>omNum-<br />

R.B.Boppana<strong>and</strong>R.Hirschfeld(1989),Pseudor<strong>and</strong>omGenera<strong>to</strong>rs<strong>and</strong>Com-<br />

[Bb65] B.Buchenberger(1965):EinAlgorithmuszumAundenderBasiselemente desRestklassenringsnacheinemnulldimensionalenPolynomideal,Dissertation,FachbereichMathematik,Universit"atInsbruck("Osterreich)ence,pp.153{159berGenera<strong>to</strong>rs,Proc.23rdIEEEConferenceonFoundationsofComputerSci-<br />

[Cassels71] [Cohen93] GraduateTextsinMathematics,B<strong>and</strong>138,Springer-Verlag,Berlin/Heidelberg. J.W.S.Cassels(1971):AnIntroduction<strong>to</strong>theGeometryofNumbers, H.Cohen(1993):ACourseinComputationalAlgebraicNumberTheory,<br />

[Copper] [CoSl88] Springer-Verlag,NewYork.<br />

D.Coppersmith,FindingaSmallRoo<strong>to</strong>faUnivariateModularEquation,Proc. J.H.ConwayundN.J.Sloane(1988):SpherePackings,<strong>Lattices</strong><strong>and</strong>Groups,<br />

[CFJP] [CJLOSS92]M.J.Coster,A.Joux,B.A.LaMacchina,A.M.Odlyzko,C.P.SchnorrundJ.Stern D.Coppersmith,M.Franklin,J.Patarin,<strong>and</strong>M.Reiter,LowExponentRSAwith RelatedMessages,Proc.EUROCRYPT'96<br />

[CR88] Complexity,B<strong>and</strong>2,Seiten111{128. basedonArithmeticinniteFields,IEEETransactionInformationTheory, B.ChorundR.L.Rivest(1988):AKnapsacktypePublicKeyCryp<strong>to</strong>system (1992):Animprovedlow-densitySubsetSumAlgorithm,Computational<br />

[Di1842] G.L.Dirichlet(1842):VerallgemeinerungeinesSatzesausderLehrerevon B<strong>and</strong>IT-34,Seiten901{909.<br />

[Damgard89]I.B.Damgard(1989):ADesignPrincipleforHashFunctions,Advancesin Bericht"uberdiezurBekanntmachunggeeigneterVerh<strong>and</strong>lungenderK"oniglich PreussischenAkademiederWissenschaftenzuBerlin,Seiten93{95. Kettenbr"uchennebsteinigenAnwendungenaufdieTheoriederZahlen,<br />

Cryp<strong>to</strong>logy|ProceedingsEuroCrypt'89,LectureNotesinComputerScience, B<strong>and</strong>435(1990),Springer-Verlag,Berlin/Heidelberg,Seiten416{427.


BIBLIOGRAPHY [Dantzig63] versityPress,Prince<strong>to</strong>n,NewJersey(dt."Ubersetzung"`LineareProgrammierung G.B.Dantzig(1963):LinearProgramming<strong>and</strong>Extensions,Prince<strong>to</strong>nUni-<br />

111<br />

[DKS98] undErweiterungen"'1966imSpringer-Verlag,Berlin/Heidelberg,erschienen).<br />

[DKT87] I.Dinur,G.Kindler,S.Safra,ApproximatingCVP<strong>to</strong>WithinAlmost-<br />

Research,B<strong>and</strong>12,Nr.1(Februar),Seiten50{59. putationusingmoduloDeterminantArithmetic,MathematicsofOperation P.D.Domich,R.KannanundL.E.Trotter(1987):HermitenormalFormCom-<br />

PolynomialFac<strong>to</strong>rsisNP-Hard,manuscript1998.<br />

[EmBoas81]P.vanEmdeBoas(1981):AnotherNP-completePartitionProblem<strong>and</strong>the [Euchner91]M.Euchner(1991):PraktischeAlgorithmenzurGitterreduktionundFak-<br />

ComplexityofComputingshortVec<strong>to</strong>rsinaLattice,TechnischerReport<br />

Universit"at,Frankfurt/Main. <strong>to</strong>risierung,Diplomarbeit,FachbereichInformatikderJohann-Wolfgang-Goethe- 81-04,FachbereichMathematikderUniversit"atAmsterdam.<br />

[Frieze86] [Feller68] A.M.Frieze(1986):OntheLagarias-OdlyzkoAlgorithmfortheSubset SumProblem,SIAMJournalonComputing,B<strong>and</strong>15,Nr.2,Seiten536{539. tion,B<strong>and</strong>I,3.Auage,JohnWiley&Sons,NewYork. W.Feller(1968):AnIntroduction<strong>to</strong>ProbabilityTheory<strong>and</strong>itsApplica-<br />

[Frumkin76]M.A.Frumkin(1976):PolynomialTimeAlgorithmsintheTheoryofLin-<br />

[FHKLS88] SIAMJournalonComputing,Volume17,No.2,pp262{280. structingTruncatedIntegerVariablesSatisfyingLinearCongruences, A.Frieze,J.Hastad,R.Kannan,J.Lagarias,<strong>and</strong>A.Shamir(1988):Recon-<br />

[GaSi78] J.vonzurGathenundM.Sieveking(1978):ABoundonSolutionoflinearIntegerEquations<strong>and</strong>Inequations,ProceedingsoftheAmericanMathematicaski,ed.,LectureNotesinComputerScience56,Springer,Berlin,pp.386{392earDiophantineEquations,FundamentalsofComputationTheory,M.Karpin-<br />

[GaJo79] M.R.GareyundD.S.Johnson(1979):Computer<strong>and</strong>Intractability:AGuide <strong>to</strong>theTheoryofNP-Completness,W.H.Freeman<strong>and</strong>Company,SanFrancisco.<br />

Society,B<strong>and</strong>72,Seiten155{158.<br />

[Gau1801] [GrLek87] C.F.Gau(1801):DisquisitionesArithmeticae,GerhardFleischer,Leipzig.<br />

North-Holl<strong>and</strong>,Amsterdam. Deutsche"Ubersetzung(1889):"`Untersuchung"uberh"ohereArithmetik"', M.GruberundC.G.Lekkerkerker(1987):GeometryofNumbers,2.Auage, Springer-Verlag,Berlin/Heidelberg.<br />

[GLLS88] Verlag,Berlin/Heidelberg. combina<strong>to</strong>rialOptimization,Algorithms<strong>and</strong>Combina<strong>to</strong>rics,B<strong>and</strong>2,Springer- M.Grotschel,L.LovaszundA.Schrijver(1988):GeometricAlgorithms<strong>and</strong>


112 [HaMcC91] cesoverRing,SIAMJournalonComputing,B<strong>and</strong>20,Nr.6,Seiten1068{1083. J.HafnerundK.McCurley(1991):Asymp<strong>to</strong>ticFastTriangulationofMatri-<br />

BIBLIOGRAPHY<br />

[HJLS89] [H] AlgorithmsforFindingIntegerRelationsamongrealNumbers,SIAM J.Hastad,B.Just,J.C.LagariasundC.P.Schnorr(1989):PolynomialTime puting17(2),pp.336{341,1988 J.Hastad,SolvingSimultaneousModularEquationsofLowDegree,SIAMJ.Com-<br />

[Helfrich85] B.Helfrich(1985):Algorithms<strong>to</strong>constructMinkowskireduced<strong>and</strong>HermitereducedLatticeBases,TheoreticalComputerScience,B<strong>and</strong>41,Seiten<br />

125{139. JournalonComputing,B<strong>and</strong>18,Nr.5,Seiten859{881.<br />

[Hermite1850]C.Hermite(1850):ExtraitsdelettresdeM.Ch.HermiteaM.Jacobi [Hlawka44] surdierentsobjetsdelatheoriedesnombres,Deuxiemelettre,Reine Angew<strong>and</strong>teMathematik,B<strong>and</strong>40,Seiten279{290.<br />

[Horner94] 49,Seiten285{312. H.H.H"orner(1994): E.Hlawka(1944):ZurGeometriederZahlen,MathematischeZeitschrift,B<strong>and</strong><br />

lemen,Diplomarbeit,FachbereichMathematikderJohann-Wolfgang-Goethe- amChor-Rivest-Kryp<strong>to</strong>systemundanallgemeinenRucksackprob-<br />

Universit"at,Frankfurt/Main. VerbesserteGitterbasenreduktion;getestet<br />

[John48] [ImpNa96] ditions,inK.O.Friedrichs,O.E.NeugebauerundJ.J.S<strong>to</strong>ker(Ed.):"`Studies<strong>and</strong> F.John(1948):ExtremumProblemswithInequalitiesassubsidiaryCon-<br />

asSubsetSum,J.Cryp<strong>to</strong>logy9,pp.199{216,1996 R.Impagliazzo<strong>and</strong>M.Naor,EcientCryp<strong>to</strong>graphicSchemesProvablyasSecure<br />

[JoSt94] Essayspresented<strong>to</strong>R.Couran<strong>to</strong>nhis60thBirthdayJanuar8,1948"',Interscience Publisher,NewYork,Seiten187{204.<br />

[KaLe78] A.JouxundJ.Stern(1994):LatticeReduction:AToolboxfortheCryptanalyst,TechnischerReport,DGA/CELAR,Bruz(Frankreich).Toappear,Journal<br />

Sphere<strong>and</strong>inSpace,ProblemsofInformationTransmission,B<strong>and</strong>14,Seiten G.A.KabatianskyundV.I.Levenshtein(1978):BoundsforPackingsona 1{17. ofCryp<strong>to</strong>logy.<br />

[Kaib91] [Kaib94] M.Kaib(1991):TheGauLatticeBasisReductionsucceedswithany Norm,ProceedingsofFundamentalsofComputationTheory(FCT'91),Springer<br />

furt/Main. LectureNotesinComputerScience,B<strong>and</strong>591,Seiten275{286. tion,FachbereichMathematikderJohann-Wolfgang-Goethe-Universit"at,Frank-<br />

M.Kaib(1994):Gitterbasenreduktionf"urbeliebigeNormen,Disserta-


BIBLIOGRAPHY [KaSchn96] M.KaibundC.P.Schnorr(1996):TheGeneralizedGaussReductionAlgorithm,JournalofAlgorithms,B<strong>and</strong>21,Nr.3(November),Seiten565{578.<br />

113<br />

[Kannan83] [KaBa79] R.Kannan(1983):ImprovedAlgorithmsforIntegerProgramming<strong>and</strong> theSmith<strong>and</strong>theHermiteNormalFormofanIntegerMatrix,SIAM JournalonComputing,B<strong>and</strong>8,Seiten499{507. R.KannanundA.Bachem(1979):PolynomialAlgorithmforComputing<br />

[Kannan87] gramming,MathematicsofOperationResearch,B<strong>and</strong>12,Nr.3(August),Seiten RelatedLatticeProblems,Proceedingsofthe15thACMSymposiumonTheory<br />

415{440. R.Kannan(1987):Minkowski'sConvexBodyTheorem<strong>and</strong>IntegerPro-<br />

ofComputing,pp.193{206<br />

[Karma84] [Khach79] M.Karmarkar(1984):AnewPolynomial-TimeAlgorithmforLinearProgramming,Combina<strong>to</strong>rica,B<strong>and</strong>4,Seiten373{395.<br />

[Khach80] L.G.Khachiyan(1979):APolynomialAlgorithminLinearProgramming, L.G.Khachiyan(1980):PolynomialAlgorithmsinLinearProgramming, SovietMathmaticsDoklady,B<strong>and</strong>20,Seiten191{194. U.S.S.R.ComputationalMathematics<strong>and</strong>MathematicalPhysics,B<strong>and</strong>20,Seiten<br />

[KoZo1872] [Khin35] 53{72.<br />

quaternaires,MathematischeAnnalen,B<strong>and</strong>5,Seiten366-389. A.KorkineundG.Zolotare(1872):Surlesformesquadratiquepositive A.Khintchine(1935):ContinuedFractions,Moscow-Leningrad<br />

[KoZo1873] [KoZo1877] A.KorkineundG.Zolotare(1873):Surlesformesquadratique,MathematischeAnnalen,B<strong>and</strong>6,Seiten366{389.<br />

[Knuth71] MathematischeAnnalen,B<strong>and</strong>11,Seiten242{292. D.E.Knuth(1971):TheAr<strong>to</strong>fComputerProgramming,FundamentalAlgorithms,B<strong>and</strong>I,Addison-Wesley,Reading.<br />

A.KorkineundG.Zolotare(1877):Surlesformesquadratiquepositive,<br />

[Knuth80] D.E.Knuth(1980):TheAr<strong>to</strong>fComputerProgramming,FundamentalAlgorithms,VolumeII,Addison-Wesley,Reading,MA.<br />

[LARIFARI]M.Kaib,R.Mirwald,C.R"ossner,H.H.H"orner,H.Ritter(1994):Program-<br />

[La1773] l'AcademieRoyaledesSciencesetBelles-Lettres,Berlin,Seiten265{312. J.L.Lagrange(1773):Recherchesd'arithmetique,NouveauxMemoiresde matikundInformatikderJohann-Wolfgang-Goethe-Universit"at,Frankfurt/Main. mieranleitungf"urLARIFARI|Version13.07.1994,FachbereicheMathe-<br />

[Lang93] S.Lang(1993):Algebra,3.Auage,Addison-Wesley,Reading.


114 [LLS90] J.C.Lagarias,H.W.LenstraundC.P.Schnorr(1990):Korkin-ZolotarevBases <strong>and</strong>successiveMinimaofaLattice<strong>and</strong>itsreciprocallattice,Combina<strong>to</strong>rica,B<strong>and</strong>10,Seiten333{348.<br />

BIBLIOGRAPHY<br />

[LaOd85] [LLL82] Problems,JournalofACM,B<strong>and</strong>32,Nr.1,Seiten229{246. A.K.Lenstra,H.W.LenstraundL.Lovasz(1982):Fac<strong>to</strong>ringPolynomialswith J.C.LagariasundA.M.Odlyzko(1985):Solvinglow-densitySubsetSum<br />

[Lenstra83] 534. H.W.Lenstra(1983):IntegerProgramminginaxedNumberofVariables, RationalCoecients,SpringerMathematischeAnnalen,B<strong>and</strong>261,Seiten515{<br />

[Lovasz86] MathematicsofOperationResearch,B<strong>and</strong>8,Nr.4(November),Seiten538{548<br />

[LoSc92] vexity,CBMS-NSFRegionalConferenceSeriesinAppliedMathematics,B<strong>and</strong>50, L.Lovasz(1986):AnalgorithmicTheoryofNumbers,Graphs<strong>and</strong>Con-<br />

MathematicsofOperationResearch,B<strong>and</strong>17,Nr.3(August),Seiten751{764. L.LovaszundH.Scarf(1992):TheGeneralizedBasisReductionAlgorithm, SIAMPublications,Philadelphia.<br />

[MaOd90] [Mink1896] J.E.MazoundA.M.Odlyzko(1990):LatticePointsinhigh-dimensional Sphere,MonatsheftMathematik,B<strong>and</strong>110,Seiten47{61.<br />

[Mink1911] H.Minkowski(1896):GeometriederZahlen,ersteAuage,Teubner-Verlag,<br />

[Mishra93] H.Minkowski(1911):GesammelteAbh<strong>and</strong>lungen,B<strong>and</strong>IundII,Teubner- B.Mishra(1993):AlgorithmicAlgebra,Texts<strong>and</strong>MonographsinComputer Verlag,Leipzig.<br />

[Or<strong>to</strong>n1994]G.Or<strong>to</strong>n(1994):Amultiple-iteratedTrapdoorfordensecompactKnapsacks,AdvancesinCryp<strong>to</strong>logy|ProceedingsEuroCrypt'94,LectureNotesin<br />

Science,Springer-Verlag,New-York.<br />

[PaSchn87] ComputerScience,B<strong>and</strong>950(1995),Springer-Verlag,Berlin/Heidelberg,Seiten 112{130. A.PazundC.P.Schnorr(1987):ApproximatingInteger<strong>Lattices</strong>by<strong>Lattices</strong>withcyclicFac<strong>to</strong>rGroup,14.thInternationalColloquiumonAu<strong>to</strong>mata,<br />

[Rieger78] G.J.Rieger(1978):"uberdiemittlereSchrittzahlbeiDivisionsalgorithmen,MathematischeNachrichten,B<strong>and</strong>82,Seiten157{180.<br />

267,Springer-Verlag,Berlin/Heidelberg,Seiten386{393. Languages<strong>and</strong>Programming(ICALP),LectureNotesinComputerScience,B<strong>and</strong><br />

[Ritter96] <strong>Application</strong>sofCryp<strong>to</strong>graphy|PragoCrypt'96,CTUPublishingHouse,Prag, Seiten480{492. H.Ritter(1996):BreakingKnapsackCryp<strong>to</strong>systemsby`1-normEnumeration,Proceedingsofthe1.stInternationalConferenceontheTheory<strong>and</strong>


BIBLIOGRAPHY [Ritter97] Dissertation,FachbereichMathematikderJohann-Wolfgang-Goethe-Universit"at, H.Ritter(1997):Aufz"ahlungkurzerGittervek<strong>to</strong>reninallgemeinerNorm, 115<br />

[Rogers64] bridge. C.A.Rogers(1964):Packing<strong>and</strong>Covering,CambridgeUniversityPress,Cam-<br />

Frankfurt/Main.<br />

[Schnorr87] [Schnorr88] tionAlgorithms,TheoreticalComputerScience,B<strong>and</strong>53,Seiten201{224. C.P.Schnorr(1987):AHierarchyofpolynomialtimeLatticeBasisReduc-<br />

[Schnorr91a]C.P.Schnorr(1991):GittertheorieundganzzahligeOptimierung,Skriptzur Vorlesung,Johann-Wolfgang-Goethe-Universit"at,Frankfurt/Main. tion,JournalofAlgorithms,B<strong>and</strong>9,Seiten47{62. C.P.Schnorr(1988):AmoreecientAlgorithmforLatticeBasisReduc-<br />

[Schnorr91b]C.P.Schnorr(1991):Fac<strong>to</strong>ringIntegers<strong>and</strong>ComputingDiscreteLoga-<br />

[SchnEu91] ingsEuroCrypt'91,LectureNotesinComputerScience,B<strong>and</strong>547,Springer- Verlag,Berlin/Heidelberg,Seiten171{181. rithmsviaDiophantineApproximation,AdvancesinCryp<strong>to</strong>logy{Proceed-<br />

Springer-Verlag,Berlin/Heidelberg,Seiten68{85. ofComputationTheory(FCT'91),LectureNotesinComputerScience,B<strong>and</strong>591, Algorithms<strong>and</strong>solvingSubsetSumProblems,ProceedingsofFundamentals C.P.SchnorrundM.Euchner(1991):LatticeBasisReduction:improved<br />

[Schnorr93] oreticalComputerScience,B<strong>and</strong>13,Seiten171{182. plexity,Ed.Jim-YiCai,AMSDIMACSSeriesinDiscreteMathematics<strong>and</strong>The-<br />

rithmsviaDiophantineApproximation,AdvancesinComputationalCom-<br />

C.P.Schnorr(1993):Fac<strong>to</strong>ringIntegers<strong>and</strong>ComputingDiscreteLoga-<br />

[Schnorr94b]C.P.Schnorr(1994):GittertheorieundKryp<strong>to</strong>graphie,Ausarbreitung, [Schnorr94a]C.P.Schnorr(1994):BlockReducedLatticeBases<strong>and</strong>SuccessiveMinima, Johann-Wolfgang-Goethe-Universit"at,Frankfurt/Main. Combina<strong>to</strong>rics,Probability<strong>and</strong>Computing,B<strong>and</strong>3,Seiten507{522.<br />

[SchnHo95] <strong>to</strong>systembyimprovedLatticeReduction,AdvancesinCryp<strong>to</strong>logy|Pro-<br />

ceedingsEuroCrypt'95,LectureNotesinComputerScience,B<strong>and</strong>921,Springer- Verlag,Berlin/Heidelberg,Seiten1{12. C.P.SchnorrundH.H.H"orner(1995):AttackingtheChor-RivestCryp-<br />

[Schrijver86]A.Schrijver(1986):TheoryofLinear<strong>and</strong>IntegerProgramming,Wiley- [Seysen93] InterscienceSeriesindiscreteMathematics<strong>and</strong>Optimization,JohnWiley&Son Ltd. M.Seysen(1993):SimultaneousReductionofaLattice<strong>and</strong>itsreciprocal Basis,Combina<strong>to</strong>rica,B<strong>and</strong>13,Seiten363{376.


116 [Shamir82] Cryp<strong>to</strong>system,Proc.23rdAnnualSymposiumonFoundationsofComputerScience,1982,pp.145{152<br />

A.Shamir,APolynomial-TimeAlgorithmforBreakingtheBasicMerkle-Hellman BIBLIOGRAPHY<br />

[Siegel89] Berlin/Heidelberg. C.L.Siegel(1989):LecturesontheGeometryofNumbers,Springer-Verlag,<br />

[SpStr76] [Smith1861]H.J.S.Smith(1861):OnSystemsoflinearindeterminateEquations<strong>and</strong> E.SpeckerundV.Strassen(1976):Komplexit"atvonEntscheidungsproblemem,LectureNotesinComputerScience,B<strong>and</strong>43,Springer-Verlag,<br />

151,Seiten293{326. Congruences,PhilosophicalTransactionoftheRoyalSocietyofLondon,B<strong>and</strong><br />

[Vetchin82] Berlin/Heidelberg.<br />

[Watson66] N.M.Vetchinkin(1982):UniquenessofClassesofpositivequadraticForms onwhichValuesoftheHermiteConstantsareattainedfor6n8,<br />

oftheCambrigdePhilosophicalSociety(Mathematical<strong>and</strong>PhysicalScience),B<strong>and</strong> G.L.Watson(1966):OntheMinimumofapositivQuadraticForminn (n8)Variables(VericationofBlichfeldt'sCalculations),Proceeedings ProceedingsoftheSteklovInstituteofMathematics,Nr.3,Seiten37{95.<br />

[Ye91] 62,Seite719.<br />

[Yao82] MathematicalProgramming,B<strong>and</strong>51,Seiten239{258. ofthe23rdIEEESymposiumonFoundationsofComputerScience Y.Ye(1991):PotentialReductionAlgorithmforLinearProgramming, A.Yao(1982):Theory<strong>and</strong><strong>Application</strong>sofTrapdoorFunctions,Proceedings

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!