ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Randomregulargraphs

NickWormald UniversityofWaterloo

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Randomregulargraphs

•Whatarethey(models) •Typicalandpowerfulresults •Handlesforanalysis •Oneortwogeneraltechniques •Someopenproblems

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Regulargraphs

Avertexhasdifitisincident withdedges.

Ad-regulargraphhasallverticesof degreed.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

sparrow Australia pigeon ChristmasIsland lyrebird lark Ireland spotted Noumea treecreeper PEI dodo Tasmania pegasus TrafficIsland Kentuckyfried chicken Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA Randomregulargraphs Whatarethey?

Faultilyfaultless,icilyregular, splendidlynull,deadperfection; nomore.

-LordAlfredTennyson

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Randompermutation

Chooseapermutationof{1,...,n}uniformlyat random. 1 2 e.g.(1235)(467) 6 7 3

5 4 Thenforgetorientations 1 2

6 7 3

5 4

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Uniformmodel

n,d d G :Probabilityspace,elementsarethe - n regulargraphson vertices. 1 Eachhasthesameprobability: |G n,d |

Butisnotknownexactly.Hardtoanalyse.|G n,d |

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Algorithmicmodels

e.g.Degree-restrictedprocess: addedgestorandomplaces,keepingallvertex degreesatmostd.

1 2 1 2

6 3 6 3

5 4 5 4

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA Superpositionmodels

Random ∈ H6 ∈G6,1 Hamilton

Takeone Throwitaway memberfrom ifamultiple eachoftwo ∈ H6 G6,1 edgeis models,and created. superimpose.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

a.a.s.

ApropertyQholdsasymptoticallyalmost surely(a.a.s.)inarandomgraphmodelif

P n (GhasQ)→1as → ∞

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA Questions-uniformmodel

n,d DographsinG a.a.s.satisfythefollowing? d connected,andmoreover -connected n containaperfectmatching(for even)

hamiltonian(havecyclethroughallvertices)

trivialautomorphismgroup ...andhowarethefollowingdistributed?

subgraphcountschromaticnumber

eigenvalues

independent&dominatingsetsizes

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

n,d SomeanswersforG Manythingsareknown.Someexamples: 3 d n 4 d For :a.a.s. -connectedand hamiltonian(Bollobas,Wormald,Frieze,Robinson, Cooper,Reed,Krivelevich,Sudakov,Vu), trivialautomorphismgroup(B,McKay,W,K,S,V),

Forfixedd:distributionofeigenvalues(McKay), < 2√d 1 + secondeigenvaluea.a.s. (Friedman) Chromaticnumberboundsknown(Achlioptas& Moore,Shi&Wormald)

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Contiguity

n n TwosequencesofmodelsG andF are contiguousifforanysequenceofeventsAn

An n isa.a.s.trueinG ifandonlyif An n isa.a.s.trueinF

n n Notation:G ≈F

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

ContiguityofSuperpositionmodels

d 3 Thm1(~Robinson&W;Janson)For n, d  n,  n, d n G G ≈G ( even).

Thm2(Janson;Molloy,Reed,Robinson&Wormald)

n,  n,  n,  n,  n G G G ≈G ( even). 3 n,  n,  i.e. G ≈G d 3 Thm3(Robalewska)For . n, d  n,  n, d G G ≈G

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Arithmeticofcontiguity

Examplewithneven:

n,  n,  n,  G ≈G G (Thm3) n,  2 n,  ≈G G (Thm3) n,  n,  2 n,  ≈G G G (Thm1) n,  2 n,  2 n,  ≈G G G (Thm1) 3 n,  2 n,  2 n,  ≈ G G G (Thm2) 5 n,  2 n,  = G G

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

1+1isnot2

n, d IngeneralallsuchequationswithG and respectingdegreesumsaretrue(neven):

n, d n, d ... n, d n, d G  G  G k≈G d  + d  + ... + d = d 3. provided k

Thereisonefailure:

n,  n,  n,  G G ≈G

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

n ContiguitywithH n H =randomHamiltoncycleonnvertices Thm4(Frieze,Jerrum,Molloy,Robinson&Wormald)

n, d  n n, d d 3 G H ≈G (fixed ). Thm5(Kim&Wormald) n n n,  H H ≈G n, d n ThusallequationsinvolvingvariousG andH respectingdegreesumsaretrue,provided thetotaldegreeisatleast3.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

n ContiguitywithF

n F =graphformedfromuniformlychosen randompermutationofnvertices(withno loopsormultipleedges).

Thm6(Greenhill,Janson,Kim&Wormald)

n n n,  n, d  n n, d d 3 F F ≈G , G F ≈G ( ).

n, d n n ThusequationsinvolvingvariousG ,F andH respectingdegreesumsaretrue,provided thetotaldegreeisatleast3.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Corollaries n n n,  n, d  n F F ≈G ≈G H n n impliesthatF F isa.a.s.hamiltonian. (AlsoprovedalgorithmicallybyFrieze.)

n, d  n n, d G H ≈G n, d impliesthatG isa.a.s.hamiltonian. d n,  n, d n G ≈G ( even) n, d impliesthatG isa.a.s.decomposableinto dperfectmatchings(soisd-edge-colourable).

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA Othercorollaries Eachmodelisa.a.s.decomposableintod/ 2 d 4 edge-disjointHamiltoncycles(even ). Bipartiteversionofthisisalsotrue(Greenhill, Kim&Wormald).Sothereexist4-regular bipartitegraphswithahamiltonian decompositionandarbitrarilylargegirth (=lengthofshortestcycle). Thisgivesexamplesofcomplexeswithin- coherentfundamentalgroup(McCammond&Wise).

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Othercorollaries

Anythinga.a.s.trueinoneofthemodelsis alsoa.a.s.trueintheothers.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Handlesforanalysis

1.Pairingmodel(esp.forsmalld)

2.Switchings(esp.formoderated)

3.Enumerationresults(esp.forlarged)

Forthistalk:only1indetail.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

n, d Pairingmodel-P Tostandforavertex,takedpointsin a“cell”orbucket.

vertex1vertex2...

Thentakearandompairing(perfect matching)ofallthepoints.Thepairs determinetheedgesofthegraph.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

n, d Pairingmodel-P

2 n=6d=3

1 3

4 6 Rejectifloopsor 5 multipleedges

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

n, d Pairingmodel-P

2 n=6d=3

1 3

4 6 Uniformly uniformlyrandom randompairinggives 5 graph

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Analysisofrandompairings Foranalysis,permitloopsandmultipleedges. Example:distributionofnumberoftriangles.

Forthisweusethemethodofmomentsasin Lecture4.Butnowpairsaredependent. (c.f.G (n , p ) ,whereedgesareindependent.)

CreateanindicatorvariableI jforeachtriple ofpairsthatinduceatriangleinthegraph. Callsuchatripleatriangleofthepairing.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Expectednumberoftriangles

IfXisthetotalnumberoftrianglesinthe randompairingthensinceI jisanindicator EX = E I j = P(I j = 1 ) j j X X

P(I j = 1 ) = M (dn 6) / M (dn ) whereM ( k ) isthenumberof perfectmatchingsofkpoints.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

triangles(cont.)

M ( k ) = ( k 1)(k 3) ... 1 Weeasilyget 3 P(I j = 1 ) (dn ) andthen .Thenumberof waystochooseatriangleinthepairingis

(d(d 1 )) 3 n 3

3 EX (d 1 ) / 6 Thus .

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

ASIDE:Easyexercise

ShowthatifFhasmoreedgesthanvertices thenita.a.s.doesnotoccurasasubgraphof n, d G .

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Distributionofcyclecounts

Highermomentseasilycomputedinasimilar way.Conclusion:

X hasasymptoticallyPoissondistributionwith 3  = (d 1 ) / 6 expectation .

X r -cyclesoflengthr-canbedonesimilarly andagainthedistributionisasymptotically = (d 1 ) r / 2 r Poissonwithexpectation r .

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Jointdistribution Jointmomentsalsocomputedinthesame fashion.Forinstance i j E(X ) (X )    i  j fromwhichwemayconcludeX andX are asymptoticallyjointlyindependentPoisson. Oneimplicationofthisis P(X = X = 0 ) exp(   )   (1 d 2 ) / 4 = e .

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Simplegraphs

Letsimpledenotetheeventthattherandom pairingproducesnoloopsormultipleedges. Thenwehavefound P(simple) e (1 d 2)/ 4 .

JointmomentsofX ,X andotherXr‘s givetheasymptoticdistributionof Xr Xs n,d , ,...,inG .

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

a.a.s.propertiesofsimplegraphs SinceP(simple)isboundedbelow,iffor A P(A) = 1 o(1) anyevent weshowthat n,d P(A)= 1 o(1) n,d inP then alsoinG . Thisisthebasisofattackformanyproblems. Example:“1+1”isnot“2”becausethe n,  probabilityofG havingnooddcycleof oflengthlessthan2gisasymptotically exp(    g  ) , whichtendsto0asg goestoinfinity.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

VarianceandHamiltoncycles NowletY bethetotalnumberofHamilton cyclesin(thegraphof)therandompairing. UseanindicatorvariableI jforeachpossible setofpairsinducingaHamiltoncycletofind E(Y ) var(Y ) d 3 n,d and .For wefindinP var(Y ) / E ( Y ) 2 d /(d 2 ) 1 , apositiveconstant.Bysecondmoment methodthisisanupperboundonP(Y = 0 ).

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA Smallsubgraphconditioning -forprovingcontiguity ImplicitinworkofRobinson&Wormald,dis- tilledbyJanson,alsoMolloy,Robalewska,R&W. Thetechniquemayapplywhenvar(Y ) isof 2 theorderof E(Y ) andthevariability signifiedbythelargevarianceis“induced”by somevariablesdescribinglocalproperties.

OftentheseareX  , X  ,...(shortcyclecounts).

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Thehypotheses

LetY countdecompositionsofagraphofa specifictype.Forexample,Hamiltoncycle+ (d 2 ) -regulargraph.

1.X , X ,...,Xk areasymptotically independentPoissonwithexpectations i.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Thehypotheses(PART2) Y X X X 2.E (  ) j  (  ) j  ( k ) j k k → ji EY ( i (1 + i )) i=  Y foreveryfinitesequencej , j ,...,jk of > 1 non-negativeintegers,whereall i . ∞ 2 2 2 3.EY (EY) exp i i i =  X andthesumconverges.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Smallsubgraphconditioning-conclusion

n n, d R ≈G n whereR isthespaceofrandomregulargraphs eachwithprobabilityproportionaltothenum- berofdecompositionsofthespecifiedtype. Superpositionmodelsrelatetodecompositions! CalculationsintheHamiltoncycleexample d 3 verifytheconditionsfor ,so n, d  n n, d G H ≈G

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

andsoon

Allthecontiguityresultsstatedbefore areprovedbythatmethod.

Ifwehavetime,let’slookatthe differentialequationmethod.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Greedyalgorithms

Problem:whatisthesizeofthelargest independentsetinarandomregular graph?Largestdominatingset?

Greedyalgorithmsoftenachievegooda.a.s. bounds.

Howdoweanalysethem?

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Differentialequationmethod Arandomisedalgorithmisappliedtoa graph.

Whenthealgorithmisappliedtoarandom regulargraph,itsstepsdependonsome variablesthata.a.s.followclosetothe solutionsofsomesystemofdifferential equations.(Justificationbymartingale techniques.)

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

DEmethodforrandompairings

Wewillneedtocomputetheexpected changesinthesevariables,ineachstep.

Considerthealgorithmappliedtoa randompairing.Ineachstepofthe algorithm,onemaygenerateatrandomjust thosepairsinvolvingwhicheverpointsare relevantforthenextstepofthealgorithm.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

example:maxindependentset Anindependentsetisasetofvertices,no twoofwhichareadjacent. (G)denotes thelargestsizeofanindependentsetinG. Fromexpectationarguments, ( G ) < ( d ) n a.a.s.,wheree.g.(McKay) (3)=0.4554, (4)=0.4163. Lowerboundscomefromgreedyalgorithms.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA Greedyalgformaxindependentset Simplealgorithm:selectverticesconsecutively atrandomtobuildanindependentset.Upon selectingavertex,deleteitanditsneighbours. Yi(t):numberofverticesofdegreeiaftert stepsofthealgorithm.Inpairingmodel,find (asymptotically)expectedchangeinYiinone step,asfunctionofYj‘s. Writingtheexpectedchangeasaderivative givesadifferentialequation:

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

d.e.forindependentsetalgorithm

yi0 = f (y  , y  , ... , y d )

whereyi ( x ) approximatesYi ( t)/ n attimet, x = t / n .(Detailsomitted!)

Thed.e.methodincludesgeneralresultsfor showingthata.a.s.theYistayclosetothe scaledsolutionsofthed.e.:

Yi (t) =nyi (t/ n ) + o(n ) a.a.s.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

Conclusionofsimplealgorithm

Letxbethesolutionofyi (x)= 0.Then a.a.s.theprocesslastsforXx n + o(n ) steps. Thus (G)>x n + o(n ) a.a.s. d 3 Wefindfor that x = (1 / 2)(1 (d 1) 2 / (d 2 ))  dx 3 0.3750 4 0.3333

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA Degreegreedyalgorithm

Giveprioritytoverticeswithminimumdegree intheever-shrinkinggraph. dxtheupperboundsagain 30.43270.4554 4 0.39010.4163 (Analysisrequiresextrabellsandwhistles.) Thecased=3alsoobtainedbyFriezeand Suen(analyticallyas6log(1.5)-2)analysing thesamealgorithmanotherway.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA Colouring

( n, ) = 3 Easyexercisethat G a.a.s. (Hints:BrooksThm,andshortoddcycles.) Greedyalgorithm:assigncoloursrandomlyto vertices:a.a.s.requiresd + 1colours.

Betteralgorithm:higherprioritytovertices withmorecoloursalreadyontheirneighbours. AchlioptasandMooreshowedinthiswaythat P( ( n, ) = 3) > c + o(1 ) c > 0 G forsome .

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA Colouring-evenbetteralgorithm Modifiedbetteralgorithm:firstcolourthe shortoddcycles,thenproceedasbefore. Thisshowsabovewithc = 1 (Shi&Wormald). ( n, ) = 3 So G a.a.s. ( n, ) = 3 4 Similarly,weget G or a.a.s., ( n, ) = 4 G a.a.s.,etc. Analysisforupperboundsbyd.e.methodusing bells,whistlesandflashinglights.Lowerbounds provedearlier(Molloy&Reed)usingexpectation.

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

UnsolvedProblems

Conjecturethatarandomd-regulargraph withanevennumberofverticesa.a.s.hasa perfect1-factorisation(d≥3).

Doesarandomd-regulardirectedgrapha.a.s. havededge-disjointHamiltoncycles(d≥3)?

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

MoreunsolvedProblems

Istheuniformrandomd-regulargraph contiguoustothealgorithmicallydefinedmodel (addedgesatrandomsubjecttomaximum degreed)?

Isarandom5-regulargrapha.a.s.3-colourable?

Jan4,2005 LECTURE5 Jan4,2005 ATLANTAMAA SHORTCOURSEONRANDOMGRAPHS ATLANTAMAA

References

“Recentpublications”16,17,27,36,54,66on mywebpage:

http://www.math.uwaterloo.ca/~nwormald/abstracts.ht

Jan4,2005 LECTURE5 Jan4,2005