• Aucun résultat trouvé

Graph Extremities Defined by Search Algorithms

N/A
N/A
Protected

Academic year: 2021

Partager "Graph Extremities Defined by Search Algorithms"

Copied!
18
0
0

Texte intégral

(1)

HAL Id: lirmm-00106696

https://hal-lirmm.ccsd.cnrs.fr/lirmm-00106696

Submitted on 16 Oct 2006

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of

sci-entific research documents, whether they are

pub-lished or not. The documents may come from

teaching and research institutions in France or

L’archive ouverte pluridisciplinaire HAL, est

destinée au dépôt et à la diffusion de documents

scientifiques de niveau recherche, publiés ou non,

émanant des établissements d’enseignement et de

recherche français ou étrangers, des laboratoires

Graph Extremities Defined by Search Algorithms

Anne Berry, Jean Blair, Jean-Paul Bordat, Geneviève Simonet

To cite this version:

Anne Berry, Jean Blair, Jean-Paul Bordat, Geneviève Simonet. Graph Extremities Defined by Search

Algorithms. [Research Report] 05055, LIRMM. 2005, 17 p. �lirmm-00106696�

(2)

A. Berry  J. Blair y J.-P. Bordat z G. Simonet z Abstract

Graph search algorithms have exploited and in some cases identi ed graph extremities, such as the leaves of atree and the simplicial vertices of a chordal graph. Recently, several well-knowngraph search algorithms have beencollectivelyexpressed astwogeneric algorithmscalledMLSand MLSM,eachofwhichinstantiateswithparametersde ningthesetoflabels to be used, apartial orderon labels usedto choose at each stepa vertex ofmaximallabelandthewayalabelisinitializedandincremented. Inthis paper, we investigate the properties of the vertex that is numbered 1 by MLS ona chordal graphand by MLSM onanarbitrary graph. We show thattheminimalseparatorsincludedintheneighborhoodofthisvertexare totally orderedby inclusionandthat thisvertexis indeedanextremity,as itbelongstoamoplexofthegraph. Whentheorderonlabels istotal,this extremitypropertyis stillstronger sinceall thevertices ofthis moplexare numberedconsecutively. WhenMLSisrunonanon-chordalgraphwitha totalorderonlabels,vertexnumber1isaweakerkindextremity,calledan OCF-vertex.

1 Introduction

Variouspropertiesthatidentifyavertexasoutermostorasanextremityofagraph havelongbeen exploited in both graphtheory and the designof eÆcient graph algorithms. Theendpointsof apath, forexample,are itstwoextremities;leaves arethe extremities of atree. Because this simplenotion hasproved veryuseful indealingwithtrees,graphtheoristshaveendeavoredtoextendittowidergraph classes.

For chordalgraphs, extremities were de ned as thesimplicial vertices (a vertex is simplicial if its neighborhood is a clique), concurrently by Dirac [9] and by LekkerkerkerandBoland[19]. ThisconceptledtoeÆcientrecognitionalgorithms forchordalgraphs,basedonthecharacterizationofFulkersonandGross[14],who showed that a graph is chordal if and only if it has a peo (perfect elimination ordering). Apeoisanorderingoftheverticesgivenbythesimplicial elimination scheme,whichrepeatedly ndsasimplicialvertexandremovesitfromthegraph. This was eÆciently implemented by Algorithm LexBFS, introduced by Rose, TarjanandLueker[21],which ndsapeoin asinglelinear-timepassiftheinput graph is chordal, numbering the vertices from n to 1. Thus LexBFS ends on a simplicial vertex. Tarjan and Yannakakis [22] later simpli ed Lex BFS into AlgorithmMCS.



LIMOS,Ensemblescienti quedesCezeaux,F-63177Aubiere,France. berry@isima.fr y

Dept. of EE & CS, United States Military Academy, West Point, NY 10996, USA. jean.blair@usma.edu

z

(3)

andGoss[16]de nedanedge-eliminationforthesubclassofweaklychordalgraphs calledchordalbipartitegraphs. ThiswaslaterextendedbyHayward,Spinradand Sritharan [17] to an elimination scheme involving extremities of weakly chordal graphscalledco-pairs. Dahlhaus,Hammer,Ma rayandOlariu[13]usedMCSto ndadominationeliminationordering onHHD-free graphs. OnAT-freegraphs, Corneil, Olariu and Stewart [11] de ned dominating pairs of vertices, and used LexBFSto ndsuchapaireÆciently[12], asthevertexnumbered1byLexBFS belongstoadominatingpair,andasecondpassofLexBFSwill ndasecondsuch vertex.

Onarbitrary graphs,workwasdoneon computinga minimal triangulation of a graph(a chordalgraphobtainedfrom thisgraphbyaddinganinclusion-minimal setofedges). Ohtsuki,CheungandFujisawa[20]usedwhatiscalledthesimplicial elimination game,which simulates asimplicial eliminationscheme byrepeatedly choosingavertex,addingeveryedgewhoseabsenceviolatesthesimpliciality condi-tionandremovingitfromthecurrentgraph. Thegraphobtainedfromtheoriginal graphGbyaddingalledgesaddedinthisprocessischordal,and[20]showedthat itisaminimal triangulationof Gifand onlyifnoother orderingof thevertices canproducebythesimplicialeliminationgame astrictly inclusion-smallerset of addededges. Theycalledsuchanorderinganmeo(minimaleliminationordering) ofG,andshowedthatanorderingisameoifandonlyifateach stepofthe sim-plicialeliminationgame,aspecialvertex(whichwecallanOCF-vertex)ischosen. ThusOCF-verticescanbeviewedasextremitiesofanarbitrarygraph.

Algorithm LEX M, introduced in the same seminal paper as LexBFS [21] (LexBFS is astreamlinedversionof LEXM), nds anmeoeÆciently, sovertex number1of LEX M is always anOCF-vertex. LEXM was extendedto Algo-rithmMCS-M byBerry, Blair,Heggernesand Peyton[1]using thesamelabel simpli cationasMCStocomputeanmeoeÆciently.

BerryandBordat[2]de ned anewkindofextremitybyre ningthenotionof a simplicialvertex into that of asimplicial moplex. This strengthened the notion of simplicial vertex in a chordal graph, as any vertex of a simplicial moplex is simplicial,but somesimplicialverticesmaynotbelong to anysimplicialmoplex. Inthesameway,thenotionofmoplexstrengthensthenotionofOCF-vertexinan arbitrarygraph,asanyvertexofamoplexisanOCF-vertex,whereassome OCF-vertices may notbelong to anymoplex. Thus emerged the notionsof weak and strong extremities ofa graph: the weak extremities areOCF-vertices, which are exactlythesimplicialverticeswhenthegraphischordal,andthestrongextremities arethe verticesbelonging to amoplex, which are exactlythe verticesbelonging to a simplicialmoplex when the graph is chordal. Blair and Peyton [8] studied AlgorithmMCSonachordalgraphinrelationwiththecliquetreerepresentation ofchordalgraphs. Thoughnotexplicitlystatedin [8], itfollowsfromtheir work that MCSrunonachordalgraphendson amoplex, i.e. numbertheverticesof amoplexlastwiththesmallestconsecutivenumbers. Inanarbitrarygraph,LEX MwasshownbyBerryandBordattoendonamoplex[3],whereasMCSendson anOCFvertexthatdoesnotnecessarilybelongtoamoplex[1]. Anotherrelevant resultisthat LexBFSalwaysendsonamoplex,evenonanon-chordalgraph[2]. Corneiland Krueger [10] introduced MNS (Maximal Neighborhood Search), as analgorithm that encompassesboth LexBFSandMCS, and showed that italso computesapeoifthegraphischordal. Recently,Berry,KruegerandSimonet[7] extended the family of search algorithms by de ning ageneric algorithm MLS (MaximalLabelSearch). AlgorithmMLShastwoinputvariables: agraphanda labelling structure containing in particular aset of labelsand apartial order on

(4)

aspeci clabellingstructure. [7]showedthatthesetoforderingsoftheverticesof agivengraphcomputablebyMLS(withallpossiblelabellingstructures)isequal totheset oforderingscomputablebyMNS, which ensuresthat MLS ndsapeo if the graph is chordal. Its corresponding extension MLSM nds an meo and generalizesbothLEXMandMCS-M.

Inthispaper,weinvestigatethepropertiesofthevertexnumbered1byMLSand MLSM,inordertodetermineinwhat measurethesemoregeneralgraphsearches preservesomeof thestrong extremalproperties exhibitedby MCS,LexBFSand LEXM.Weshowthat indeedtheydo. Figure1summarizestheextremity prop-ertiesofAlgorithmsMLSandMLSM.Itpresentsknownresultsdescribedin this Sectionaswellasnewonesgiveninthis paper.

MLS with total order

LexBFS

MCS

whole moplex

MLS chordal graph

MLS=MNS

MLSM arbitrary graph

MLSM=MNSM

MLSM with total order

LEX M

MCS−M

MLS non−chordal graph

MLS=MNS

MLS with total order

whole moplex

LexBFS

MCS

OCF−vertex

not always an extremity *

OCF−vertex *

vertex of a moplex *

whole moplex *

whole moplex

whole moplex *

vertex of a moplex *

whole moplex *

whole moplex *

Figure1: Extremitypropertiesof AlgorithmsMLS andMLSM.These properties concern the vertex numbered 1 by the algorithms: it is an OCF-vertex (weak extremity),avertexofamoplex(strongextremity)oravertexofamoplexwhose vertices are numbered consecutively, denoted by 'whole moplex'. The equality MLS=MNS (resp. MLSM=MNSM)meansthat bothalgorithms computethe sameorderingson anygraph, andtherefore havethesameextremity properties. 'withtotalorder'means'withtotalorderonlabels'. Newresultspresentedinthis paperareindicatedby*.

Thepaperis organizedas follows: in Section2wegivesomenotationsand de -nitions,aswellassomeusefulpreviousresults. Section3describestheproperties oftheneighborhoodofvertex1. Section4describesthekindofgraphextremities whichvertex1belongstoforMLSandMLSM,anddiscussesthespecialproperties oftherelatedordering.

2 Preliminaries

2.1 Basic de nitions and notations

All graphs in this work are undirected and nite. The readeris referredto the classicalworkof[15]foranyde nitionsnotincludedinthissection. Thesubgraph ofGinducedbythesubsetAof verticesisdenotedG(A). Theneighborhood ofa vertexxinGisdenotedN

G

(x). TheclosedneighborhoodisN G

[x]=fxg[N G

(x). TheneighborhoodofasetofverticesAisN

G (A)=[ x2A N G (x)nA,andN G [A]= A[N G

(A). Wewill omitsubscript G whenthe graphweworkon isclear from thecontext.

(5)

simplicial ifitsneighborhoodisaclique. Amodule isasubsetX ofverticeswhich sharethesameexternalneighborhood: 8x;y2X;N(x)nX=N(y)nX,sothata cliquemoduleisasubsetX ofverticeswhichsharethesameclosedneighborhood: 8x;y2X;N[x]=N[y].

Anordering onthesetV ofverticesisaone-to-onemappingfromf1;2;:::;ngto V. In every gureof thispapershowinganordering onV, everyvertexx will be named by itsnumber

1

(x). Be carefulthat all algorithms considered here numbertheverticesfrom ndown to1, sothe rst numberedvertex is (n) and thelast numberedoneis (1).

A chordal (or triangulated) graph is a graph with no chordless cycle of length greater or equal to 4. The notions of simplicial elimination scheme and game, minimal triangulation, peo and meohave been de ned in Section 1. The graph obtainedbyrunningsimplicialeliminationgameongraphG,choosingthevertices accordingtoordering ,isdenotedG

+ .

Itremainsto de nethegraphextremitieswewill workon, namelyOCF-vertices andmoplexes.

De nition2.1 A vertex x in G = (V;E) is an OCF-vertex of G if, for any non-adjacent distinct neighbors y;z of x, there is a connected component C of G(V nN[x]) suchthaty andz belong toN(C).

AsubsetS ofverticesisaminimal separator i there areat leasttwoconnected components of G(V nS) whose neighborhood in G is S (see [5] for extensive de nitionsonminimalseparation).

De nition2.2 (Berry, Bordat [2])A moplexofagraphGisacliquemodule of Gwhoseneighborhoodisaminimal separator.

2.2 Algorithms MLS and MLSM

Thede nitionsofAlgorithmsMLSandMLSMarebasedonthenotionoflabelling structure.

De nition2.3 A labelling structureisastructure(L;;l 0

;Inc),where:  Lis aset(thesetof labels),

  is apartial order on L (which may be total or not,  denoting the cor-responding strict order), which will be usedto choose a vertex of maximal label,

 l 0

isanelementof Lwhich will beusedtoinitialize thelabels,

 IncisamappingfromLf2;3;:::;ngtoL,whichwillbeusedtoincrement alabel, andsuchthatforany integerifrom2tonandany labelslandl

0 in L

i

,the followingpropertieshold: (ls1)lInc(l;i) (ls2)if ll 0 thenInc(l;i)Inc(l 0 ;i) whereL i

isthe subsetofL de nedby inductionby: - L

n =fl

0

g,andfor anyifrom ndown to2 - L i 1 =L i [fl2L j9l 0 2L i j l=Inc(l 0 ;i)g.

(6)

input :AgraphG=(V;E)andalabellingstructure L=(L;;l 0

;Inc). output:Anmeo onV andaminimaltriangulationH=G

+

ofG. Initializealllabelsasl

0 ;E 0 E;G 0 G; fori=ndownto 1do ChooseavertexxofG 0 ofmaximallabel; (i) x; foreachvertexy of G 0 di erent fromxdo if there isa path from x to y in G

0

suchthat every internalvertex on the pathhas alabelstrictlysmaller thanlabel(y)then

E 0

E 0

[fxyg;

foreachy inV suchthatxy2E 0

do label(y) Inc(label(y);i); Removex fromG 0 ; H (V;E 0 ); Figure2: AlgorithmMLSM.

Algorithm,MLSM,isgiveninFigure2.

AlgorithmMLScanbestreamlinedfromMLSM,exactlyas[21]streamlinedLEX M into LexBFS, by removing the rst foreach loop. Thus variables E

0

and H become useless (E

0

= E and H = G) and the second foreach loop becomes: foreach neighbory ofx inG

0

do. MLScomputesapeoofanychordalgraph. Whenusing labelling structure Lwith MLS (resp. MLSM),wewill referto the algorithmas L-MLS (resp. L-MLSM). LexBFS,MCSandMNS are instancesof algorithmL-MLSandLEXM,MCS-MandMNSM(de nedin[7]fromMNS)are instancesofalgorithmL-MLSM,withthefollowingrespectivelabellingstructure L=(L;;l

0 ;Inc):

LexBFS and LEX M:Listheset oflists ofelementsof f2;3;:::;ngin strictly decreasingorder, is lexicographical order (a total order), l

0

is theempty list, Inc(l;i) is obtainedfrom l by adding i to theend of thelist, provided that i is strictlysmallerthanthelastintegerinl.

MCS and MCS-M: L = f0;1;2;:::;n 1g,  is  (a total order), l 0

= 0, Inc(l;i)=l+1.

MNSand MNSM:Listhepowersetoff2;3;:::;ng,is(notatotalorder), l

0

=;,Inc(l;i)=l[fig.

Wecall MLS (resp. MLSM, L-MLS, L-MLSM) ordering of a graphG=(V;E) anyorderingonV that canbecomputedbythegivenalgorithm. Thereadercan easilyconvincehimself ofthefollowingPropertyprovedin[7].

Property 2.4 ([7 ] ) Forany graphG andany labelling structureL, Algorithms L-MLSM computing ordering on G has the same behavior as L-MLS on G

+ (theygivethesamelabelsandnumberstovertices,providedthattheybreak tiesin the sameway).

Property2.4hastwoconsequencesthatwillbelargelyused inthispaper.

1. AnyL-MLSMordering ofGisaL-MLSorderingofG +

. Thusaproperty of MLS on chordalgraphscan in somecasesbeextended to apropertyof

(7)

MLSM on arbitrary graphssince G

is chordal (Corollaries 3.10and 3.11 andTheorem4.5).

2. MLS andMLSM havethesamebehavioron achordalgraph,sincein that case, G

+

=G. Any execution of MLS on achordal graphcan be seen as an execution of MLSM, and conversely. Thus any property of MLSM on arbitrarygraphsalsoholdsforMLSonchordalgraphs(Corollary4.3).

3 The neighborhood of vertex 1

3.1 Super-components and slices

Inthis Section, we will studysomepropertiesof theneighborhood ofthe vertex numbered1byMLSorMLSM,whichwillleadtoextremitypropertiesinSection4.

Notations3.1 Forany graphG=(V;E)andany ordering onV, -acomponent of(G; )isaconnectedcomponentof G(V nN( (1))),

-asuper-componentof (G; ) istheunion ofallcomponents of(G; )having the sameneighborhood assomecomponentof (G; ),

-pdenotesthenumberofsuper-componentsof(G; )and(D 1 ;D 2 ;:::;D p )denotes the sequenceof the super-components of (G; ) orderedas follows: 8i2[1;p 1], 1 (d i ) > 1 (d i+1 ), where d i

denotes the vertex of D i

with maximum number ( 1 (d i )=maxf 1 (d);d2D i g), -theslicesof(G; )arethesubsetsT

1 ,T 2 ,...,T p ofV partitionningV asfollows: 8i2[1;p],T i =D i [S i ,whereS i =N(D i )n([ 1j<i N(D j ))=N(D i )n([ 1j<i S j ), andt i

denotes the vertexof T i

withmaximum number.

We say that numbers the slices in increasing order if 8i 2 [1;p 1], 8t2 T i , 1 (t)> 1 (t i+1 ).

The partition of V in slices can be re ned into itspartition in thin slices, where thethinslicesof(G; )arede nedfromthecomponentsof(G; )inthesameway asthe slices arede nedfrom thesuper-components of(G; ).

Example3.2 Figure 3 shows the slices de ned by an execution of MCS on a chordal graph. N( (1))=f8;3;2gand the components of (G; ) are f7;5g,f6g, f4g and f1g, with N(f7;5g) = N(f6g) = f8g, N(f4g) = f8;3g and N(f1g) = f8;3;2g,sop=3and -D 1 =f7;5g[f6g=f7;6;5g,d 1 =7,S 1 =f8g,T 1 =f8;7;6;5gandt 1 =8, -D 2 =f4g, d 2 =4,S 2 =f3g, T 2 =f4;3gandt 2 =4, -D 3 =f1g, d 3 =1,S 3 =f2g, T 3 =f2;1gandt 3 =2. Weobserve that: -N(D 1 )N(D 2 )N(D 3 ),

- numberstheslicesinincreasing order(butnotthethinslices, sincesliceT 1

is partitionnedintothe thinslices T

0 1 =f8;7;5gandT 0 2 =f6g).

Wewillseehowthese propertiescanbegeneralizedtoMLS andMLSM orderings.

We will use the following notations and Lemmas, which are proved in the Ap-pendix.

Notations3.3 ForanyexecutionofMLS onanygraphGcomputingordering andanyverticesx andy of Gsuchthat

1 (x) 1 (y), l x

(y)denotes the labelofy justbefore numbering x, NN

x

(8)

4

1

8

2

3

6

5

7

T1

T2

T3

S

Figure3: Theslices de nedbyanexecutionofMCSonachordalgraph.

NN x (y)=fz2N(y)j 1 (z)> 1 (x)g.

Lemma3.4 ForanyMLS ordering ofachordalgraph, 8i2[1::p], NN d i (d i )= NN di ( (1))  N(D i ), NN ti (t i ) = NN ti ( (1)), and if i < p then 1 (t i )  1 (d i )> 1 (t i+1 ).

Lemma3.5 ForanyMLS ordering ofachordal graph, 8i2[1::p 1];8s2S i , 1 (s)> 1 (t i+1 ).

Lemma3.6 ForanyMLSordering ofachordalgraph,ifthe orderonlabelsis totalthen8i2[1::p 1],8d2D i , 1 (d)> 1 (t i+1 ).

WehavethefollowingProperty.

Property 3.7 For any chordal graph G and any MLS ordering of G 8i 2 [1;p 1],N(D

i

)N(D i+1

).

Proof: LetGbeachordalgraph,let beaMLSorderingofGandi2[1;p 1]. Letus showthatN(D

i )N(D i+1 ). Lets2N(D i

),andletji suchthats2S j . ByLemmas3.4and3.5, 1 (s)> 1 (t j+1 )  1 (t i+1 )  1 (d i+1 ). So s 2 NN di+1 ( (1)) and by Lemma 3.4 s2N(D i+1 ). ThusN(D i )N(D i+1 )and,asN(D i )6=N(D i+1 )byde nitionof thesequence(D i ),N(D i )N(D i+1 ). 2

ForanygraphGandanyordering onV,theminimalseparatorsofGincluded in N( (1)) are exactly the neighborhoods of the componentsof (G; ) di erent from f (1)g([4]), which are also theneighborhoods of the super-componentsof (G; )di erentfrom f (1)g. Sowehavethefollowingresult.

Theorem3.8 Foranychordal graphGandany MLSordering ofG, the min-imalseparatorsof Gincludedin N( (1)) aretotally orderedbyinclusion.

(9)

onlabelsis total thenany L-MLS ordering of Gnumbers the slices in increasing order.

ByProperty2.4 any L-MLSM ordering of any graphG is a L-MLS ordering ofG

+

,whichveri estheprecedingneighborhoodpropertiessinceG +

ischordal. Moreover, the neighborhood of (1)is the same in G and G

+ (by de nition of G +

)and since isameoofG,the componentsof(G; )and of(G +

; ) arethe samewith thesameneighborhoods(this followsforinstancefrom Theorem5.12, Invariant4.9andLemma6.2in[5]). Thus thesuper-components,their neighbor-hoodsandtheslicesof(G; )andof(G

+

; )arethesame,andTheorems3.8and 3.9extendto MLSMonanarbitrarygraph.

Corollary3.10 Forany graphG andany MLSM ordering of G, the minimal separatorsofGincludedinN( (1))aretotallyorderedby inclusion.

Corollary3.11 For any graph Gand any labelling structure L, if the order on labels is total then any L-MLSM ordering of G numbers the slices in increasing order.

If the order on labels is not total, an L-MLS ordering does not necessarily numberslicesin increasingorder. Figure4givesacounterexampleforthis.

1

2

3

5

6

4

T1

T2

Figure 4: This MNS ordering doesnot numberslices in increasing order onthis chordalgraph.

3.2 MLS on a non-chordal graph

WewillendthissectionbyafewremarksonrunningMLSonanon-chordalgraph. [2]showedTheorems3.8and3.9foranexecutionofLexBFSonanarbitrarygraph, withtheevenstrongerpropertythatitnumbersthethinslicesinincreasingorder insteadoftheslices.

Property 3.12 ([2 ])ForanygraphG(chordalornot)andanyLexBFSordering of G, the minimal separators of G included in N( (1)) are totally ordered by inclusion,and numbersthe thinslicesinincreasing order.

Upon investigation, it turns out that in a non-chordal graph, MLS does not in generalexhibit theseproperties,evenwhentheorderistotal. Figure5givesa counter-exampleofthis.

(10)

1

3

4

2

8

5

7

6

Figure5: MCS onanon-chordalgraph. Thesuper-componentsof(G; )(which are also itscomponents), are: D

1 = f5;7g, D 2 = f6gand D 3 =f1g; N(D 1 ) = f2;8g, N(D 2 ) = f3;8g and N(D 3

) = f2;3;4;8g. The neighborhoods are not pairwise inclusive, does notnumberthe slices in increasing order,vertex (1) doesnotbelongtoamoplexofG. However, (1)isanOCF-vertexofG.

4 Extremities

4.1 The extremities de ned by MLS and MLSM

ForanyMLSMordering ofagraphG,since isanmeoofG,by[20] (1)isan OCF-vertexof G(weakextremity). Toshowthat (1)isastrongextremity,i.e. avertexofamoplexofG,weneedonlyusethepropertiesoftheneighborhoodof (1)describedin Section3andthefollowingLemmaprovedin theAppendix.

Lemma4.1 For any graph G andany vertex x of G, x isa vertexof amoplex ofGi x isanOCF-vertexof Gandtheset ofminimal separatorsofGincluded inN(x)has agreatestelement S

max

for inclusion. Inthat case,the moplex containing xisN[x]nS

max .

We will also show that if moreover the order on labels is total then the ver-tices of the moplex X containing (1) are numbered consecutively, i.e. X = f (1);::::: (jXj)g. Wesaythat thealgorithmendsonamoplex.

Theorem4.2 For any non-clique graph G and any labelling structure L, L-MLSM on input graph G ends on a vertex of a moplex. If moreover the order onlabelsistotalthenitends onamoplex.

Proof: Since is a meo of G (1) is an OCF-vertex of G [20]. Moreover, by Corollary 3.10 the set of minimal separators of G included in N( (1)) has a greatest element S max for inclusion. If D p 6= f (1)g then S max = N( (1)) elseS max =N(D p 1

)(p>1becauseotherwiseGwouldbereducedtotheclosed neighborhoodofOCF-vertex (1)andwouldbeaclique). ItfollowsbyLemma4.1 that (1)isavertexofamoplexX ofG,withX =N[ (1)]nS

max

. SoXiseither equaltof (1)gortoD

p

. ThusiftheorderonlabelsistotalthenbyCorollary3.11 L-MLSMendsonX. 2

AsbyProperty2.4,anychordalgraphhasthesameL-MLSandL-MLSM order-ings,wehavethefollowingCorollary.

Corollary4.3 Foranynon-cliquechordalgraphGandanylabellingstructureL, L-MLSon inputgraphGends onavertexof amoplex. Ifmoreoverthe orderon labelsistotal thenitendson amoplex.

(11)

MLSde nesapeoofanychordalgraph,andMLSMde nesanmeoofanygraph. Wecanstrengthenthesepropertiestomoplexorderings,introducedbyBerryand Bordat[3]. Wede ne amoplexeliminationschemein achordalgraphby repeat-edlychoosingamoplexof thecurrentgraph, which issimplicialsincethis graph is chordal, and removing it from the current graph, until this graph is aclique. Ifthe graphfails to be chordal, wede ne moplex elimination game in the same wayaswede nedsimplicialeliminationgame: wesimulate amoplexelimination schemebyrepeatedlychoosingamoplexofthecurrentgraph,addingeveryedge whoseabsenceviolatesthesimplicialityof thismoplex,andremovingitfromthe currentgraph, until this graphis aclique. Thusweobtainan ordered partition (X 1 ;X 2 ;:::;X k )where X i

is amoplexof thecurrentgraphat stepiifi<k and X

k

isthe nalclique. Amoplexordering isanordering onV thatiscompatible withsuchanorderedpartition,i.e. suchthat8i2[1;k 1],8x2X

i ,8x 0 2X i+1 , 1 (x)< 1 (x 0 ). Then G +

isexactlythegraphobtainedfrom Gbyaddingall edgesaddedinthemoplexeliminationgame.

Berry and Bordat [3] showed that any moplex ordering of any graph is a meo of this graph. They also showed that any LexBFSordering of a chordal graph, respectivelyanyLEXMorderingofanygraph,isamoplexordering. Weextend theseresultstoMLSandMLSM.FollowingProperty4.4easilyfollowsfrom Corol-lary4.3,andTheorem4.5followsfromProperty2.4and4.4andfromsomeknown resultsaboutminimaltriangulation. Theirproofsaregivenin theAppendix.

Property 4.4 ForanychordalgraphGandanylabellingstructureL,iftheorder onlabelsistotalthenany L-MLS orderingofGis amoplexorderingof G.

Theorem4.5 For any graph G and any labelling structure L, if the order on labelsistotal thenany L-MLSM orderingofGisamoplex orderingof G.

4.3 MLS on a non-chordal graph

Again,wewill discuss the properties of MLS when runon anon-chordal graph. [2] showedthat even onanon-chordalgraph, LexBFSalwaysends onamoplex. Ingeneral,anMLSexecutiondoesnotendonamoplexanddoesnotevenendon avertexofamoplex,eveniftheorderonlabelsistotal. However,weshowthat iftheorderonlabelsistotalthenMLSendsonaweakextremity.

Theorem4.6 Forany graphG,anylabellingstructureLandanyL-MLS order-ing ofG, if theorder onlabelsistotalthen (1)is anOCF-vertexof G.

ThisisthecaseinparticularforMCS(asshownin[1]). TheproofofTheorem4.6 isgivenintheAppendix.

Iftheorderonlabelsisnot totalthenMLSdoesnotnecessarilyendonan OCF-vertex. For instance, on the graphshown in Figure 6, =(1;2;3;4;5;6) is an MNSorderingofGand1isnotanOCF-vertexofGsince2and4arenon-adjacent and 4does notbelong to the neighborhood of the uniqueconnected component f3;5gofG(V nN[1]).

4.4 The MLS-Terminal Vertex Problem on chordal graphs

Wede ne theMLS-Terminal VertexProblem asfollows: given agraphGand a vertexxofG,isxMLS-terminal,i.e. isthereaMLSexecutionwhichgivesxthe

(12)

1

3

2

4

6

5

Figure6: MNSonanon-chordalgraphdoesnotendonanOCF-vertex

number1? Wede ne theL-MLS-Terminal Vertex Problem in thesamewayfor anylabellingstructureL.

Lanlignel[18]showedthatLexBFS-TerminalVertexProblemisNP-completeeven intherestrictiveclassesofweaklychordalgraphsandHDP(Heriditary Dominat-ingPath) graphs,andleftopenitscomplexityonachordalgraph.

WewillgiveasimplecharacterizationofMNS-terminalverticesinachordalgraph.

Characterization 4.7 ForanychordalgraphGandanyvertexxofG,xis MNS-terminali x issimplicial andthe neighborhoods of the connectedcomponents of G(V nN[x]) aretotally orderedbyinclusion.

The MLS orderings of a graph are exactly its MNS orderings [7], so its MLS-terminalverticesareexactlyitsMNS-terminalverticesandwehavethefollowing Corollary.

Corollary4.8 MNS-Terminal (resp. MLS-Terminal)Vertex Problem is polyno-mialin theclass of chordal graphs.

5 Conclusion

Wehave shown that Algorithms MLS on achordalgraph and MLSM onan ar-bitrarygraphde neasnumber1verticeswhicharewell-de nedasextremitiesof thegraph,withspecial properties.

Whenrunonanon-chordalgraph,thepropertiesof MLSare weaker,exceptfor AlgorithmLexBFS,which standsoutashavingproperties in manywayssimilar to the meo-computingalgorithms such as LEXM, MCS-M, and more generally MLSM.Inviewofthis,perhapsitwouldbeinterestingto investigate experimen-tallywhethertheordering producedbyanexecutionofLexBFSonanon-chordal graphde nesatriangulation G

+

which is closeto minimal. MLS run ona non-chordal graph with a non-total order on labels does not always end on a weak extremityofthe graph. It is anopen questionwhether it endsin that caseon a stillweakerkindofextremity.

WealsoleaveopenthecomplexityofLexBFS-TerminalVertexProblemonchordal graphs,aswellas the complexityof MLS-TerminalVertex Problemon arbitrary graphs.

(13)

[1] A. Berry,J.Blair, P.HeggernesandB.Peyton. MaximumCardinality Searchfor computingminimaltriangulationsofgraphs. Algorithmica,2003.

[2] A.Berryand J.-P.Bordat. Separabilitygeneralizes Dirac'stheorem. Discrete Ap-pliedMathematics,84:43{53, 1998.

[3] A.BerryandJ.-P.Bordat. Moplexeliminationorderings. ElectronicNotesin Dis-creteMathematics,Volume8,2001,ProceedingsofFirstCologne-TwenteWorkshop onGraphsandCombinatorialOptimization.

[4] A. Berry, J.-P. Bordat,and P.Heggernes. Recognizing weaklychordal graphsby edgeseparability. NordicJournalComputing,7:3(2000),pp.164{177.

[5] A. Berry,J-P.Bordat,P.Heggernes,G.Simonet,and Y.Villanger. A wide-range algorithm for minimal triangulation fromanarbitrary ordering. Journalof Algo-rithms,toappear(inpress).

[6] A. Berry, P.Heggernes and Yngve Villanger. A vertex incremental approach for dynamicallymaintainingchordalgraphs. Proceedings ISAAC2003 -14th Interna-tional SymposiumonAlgorithms andComputation,Kyoto,Japan,December2003. SpringerVerlag,LectureNotesinComputerScience2906, pages47-57.

[7] A. Berry, R. Krueger, and G.Simonet. Ultimate generalizations of LexBFS and LEXM. ResearchReport***,submitted.

[8] J. R. S. Blairand B. W. Peyton. An introduction to chordal graphs and clique trees. InJ.A.George,J.R.Gilbert,andJ.W.H.Liu,editors,GraphTheoryand Sparse Matrix Computations,pages 1{30.SpringerVerlag, 1993. IMAVolumesin MathematicsanditsApplications,Vol.56.

[9] G.A.Dirac.Onrigidcircuitgraphs. PAbh.Math.Sem.Univ.Hamburg,25(1961), pp.71{76.

[10] D.CorneilandR.Krueger. Auni edviewofgraphsearching. Submitted. [11] D.GCorneil,S.Olariu,andL.Stewart.Asteroidaltriplefreegraphs.SIAMJournal

DiscreteMathematics,10:399{430, 1997.

[12] D.GCorneil,S.Olariu,andL.Stewart.Lineartimealgorithmsfordominatingpairs inasteroidaltriple{freegraphs.Proceedingsof22ndICALPConference,LNCS944 (Springer),pages292{302,1995.

[13] E.Dahlhaus,P.L.Hammer,F.Ma ray,andS.Olariu.OnDominationElimination OrderingsandDominationGraphs. Proceedings ofWG1994,81-92,1994.

[14] D.R. Fulkersonand O.A. Gross. Incidencematrixes and interval graphs. Paci c JournalofMath.,15:835{855, 1965.

[15] M. C.Golumbic. Algorithmic Graph Theory andPerfect Graphs. AcademicPress, NewYork,1980.

[16] GolumbicandGoss. V***

[17] R. Hayward,J. Spinrad, andR. Sritharan. Weaklychordalgraph algorithmsvia handles.InProceedingsoftheEleventhAnnualACM-SIAMSymposiumonDiscrete Algorithms (SODA'2000).

[18] J-M. Lanlignel. Autour de la decomposition en coupes. PhD Thesis, LIRMM, Montpellier,France,2001.

[19] C.LekkerkerkerandJ.Boland. Representationofa nitegraphbyasetofintervals onrealline. Fund.Math.,51(1962),pp.45{64.

[20] T.Ohtsuki,L.K.Cheung,andT.Fujisawa. Minimaltriangulationofagraphand optimal pivoting ordering in a sparse matrix. J. Math. Anal. Appl., 54 (1976), pp.622{633.

[21] D.Rose,R.E.Tarjan,andG.Lueker. Algorithmicaspectsofvertexeliminationon graphs. SIAMJourn.Comput,5:146{160, 1976.

[22] R.E.TarjanandM.Yannakakis.Simplelinear-timealgorithmstotestchordalityof graphs, testacyclicity ofhypergraphs, and selectively reduceacyclic hypergraphs. SIAMJ.Comput.,13:566{579, 1984.

(14)

Thesubscriptinnotationsl x

(y)andNN x

(y)willbeomittedwhentheconsidered stepisclearfromthecontext.

 WegivetheproofsoftheLemmasofSection3andofsomeotherusefulones.

Lemma5.1 At any step of an execution of MLS on a graph G, for any unnumberedvertices u;v of G,

(i)if NN(u)=NN(v)thenl(u)=l(v), (ii) ifNN(u)NN(v)thenl(u)l(v).

Proof: ThisLemmaisprovedin[7]. 2

Lemma 3.4 For any MLS ordering of a chordal graph, 8i 2 [1::p], NN d i (d i ) = NN d i ( (1))  N(D i ), NN t i (t i ) = NN t i ( (1)), and if i < p then 1 (t i ) 1 (d i )> 1 (t i+1 ). Proof: NN d i (d i )NN d i ( (1)) by de nition of d i and NN d i (d i ) 6 NN d i ( (1)) (otherwise by Lemma 5.1 l d i (d i )l d i

( (1)), whichcontradictsthechoiceof d i

at thatstep). Hence NN d i (d i )=NN d i ( (1))N[D i ]\N( (1))=N(D i ). Wesupposethati<p. Byde nitionofd

i andt i , 1 (t i ) 1 (d i ). Letus showthat 1 (d i )> 1 (t i+1 ). Ift i+1 62N( (1))thent i+1 =d i+1 andwe aredonesincebyde nition ofd

i , 1 (d i )> 1 (d i+1 ). Otherwise,assume forcontradiction that

1 (t i+1 )> 1 (d i ). Then t i+1 2NN di ( (1)), with NN di ( (1))N(D i )V nT i+1 ,so t i+1 62T i+1 ,acontradiction. Letus showthat NN

ti (t i )=NN ti ( (1)). Since8ji, 8t2T j , 1 (t i ) 1 (t j ) 1 (t),NN ti (t i )N(t i )\([ j<i T j ). Since8j<i,N(t i )\D j =; (because T i \N(D j )= ;) NN ti (t i ) [ j<i S j  N( (1)). So NN ti (t i )  NN ti

( (1)),andweshowtheequalityasaboveford i

. 2

Lemma5.2 IfGisachordalgraph,achordlesspathinGandzavertex of G not belonging to and seeing both extremities of , thenz sees every vertexof .

Proof: It is a well-known property of chordal graphs which follows for instancefrom resultsfrom[6]. 2

Lemma5.3 Let G be achordal graph, let be a MLS ordering of G, let i2[1::p], s2S i such that 1 (d i )> 1

(s), and let be achordless d i

s-path suchthat every internal vertexof  belongsto D

i

. Then the following property P holds justafter processing d

i

and remains true untilnumbering s.

P : thereisa vertexv of suchthat every vertexof [v;s] isunnumbered, NN( (1))NN(v) and8t2[v;s]nfvg,NN(t)=NN( (1)).

Proof: Let us show that P holds just after processing d i . NN d i (d i ) = NN d i ( (1)) by Lemma 3.4, so NN d i ( (1)) N(d i ). Also NN d i ( (1))  N(s) since, as is a peo of G, N( (1)) is a clique of G. It follows by

(15)

di di NN

di

(t), and as the reverse inclusion holds by de nition of d i and S i , NN di (t) = NN di ( (1)). After processing d i , NN( (1)) is unchangedand P istrue, taking as vertex v the vertexof  next to d

i

(which is theonly vertexofseeingd

i

sinceischordless).

Letusshowthat P ispreservedwhenprocessingsomevertexw, 1 (d i )> 1 (w)> 1 (s).

Firstcase: w62N( (1)). Ifnovertexof[v;s]seesw thenw62[v;s]and P remainstruewiththesamevertexv,otherwiseP remainstrue,takingas newvertexv thelast vertexof[v;s]seeingw.

Secondcase: w2N( (1)). ws isanedgeof G,asN( (1)) isaclique. We getfromedgewsandpath[v;s]awv-pathsuchthateveryinternalvertext isunnumberedandveri esNN(t)=NN( (1))NN(v),sobyLemma5.1 l(t)l(v). ByProperty 2.4,thisexecutionofMLSonchordalgraphGcan beseen asan execution of MLSM, accordingto which wv will bean edge of H with H =G

+

=G. It followsthat w sees v in G. ByLemma 5.2 w seeseveryvertexof[v;s]. Itfollowsthatwhenprocessingw,wisaddedto NN( (1)) aswell as to NN(t)for everyvertex tof [v;s], andP remains truewiththesamevertexv. 2

Lemma3.5ForanyMLSordering ofachordalgraph,8i2[1::p 1];8s2 S i , 1 (s)> 1 (t i+1 ).

Proof: Suppose there is some s 2 S i such that 1 (t i+1 ) > 1 (s). Then by Lemma 3.4 1 (d i ) > 1 (t i+1 ) > 1 (s). Lemma 5.3 ensures the existence of some vertex v such that NN

t i+1 ( (1))  NN t i+1 (v), so by Lemma 3.4 NN t i+1 (t i+1 ) NN t i+1 (v)and by Lemma 5.1 l t i+1 (t i+1 )  l t i+1

(v),which contradictsthechoiceoft i+1

atthat step. 2

Lemma 3.6 For any MLS ordering of a chordal graph, if the order on labelsistotalthen8i2[1::p 1],8d2D

i , 1 (d)> 1 (t i+1 ).

Proof: Suppose there is some d 2 D i such that 1 (t i+1 ) > 1 (d). We can choose d as the rst vertex of D

i numbered after t i+1 . Just be-fore processing t i+1 , l(t i+1

) is maximum (since the order  is total) with l t i+1 (t i+1 )=l t i+1 ( (1))byLemmas3.4and5.1,sol t i+1 (d)l t i+1 ( (1)). As byLemma3.5,everyvertexofN(D

i )isnumberedbeforet i+1 ,l(d)remains equaltol ti+1

(d)untildisnumbered. Wethusonlyhavetoprovethatl( (1)) isincrementedbefore numberingd. This willleadto acontradiction, since (1)isnumberedlast.

If t i+1

2 N( (1)), l( (1)) is incremented when numbering t i+1 and we aredone. If t i+1 = 2 N( (1)), t i+1 = d i+1 . ByProperty 3.7, S i+1 is notempty, so lets 2 S i+1 . 1 (d i+1 )> 1

(s), and by Lemma 5.3, until numberings, there is someunnumberedvertex v such that NN( (1))  NN(v), so by Lemma 5.1 l( (1)) l(v), hence l(d) l(v). This implies that

1 (s) >

1

(d) and asl( (1)) is incremented when processing s, it is incremented beforenumberingd. 2

(16)

moplexof Gi x isanOCF-vertexofGand theset ofminimalseparators ofGincludedinN(x)hasagreatestelementS

max

forinclusion. Inthatcase,themoplexcontainingx isN[x]nS

max .

Proof: Wesupposethat x isavertexofamoplex X ofG. ThenN[x]= N[X]andthereisaconnectedcomponentCofG(Vn N[x])suchthatN(C)= N(X). Since X is a clique module of G, any non-adjacent neighbors of x must belong to N(X) i.e. to N(C), so x is anOCF-vertex of G, and the neighborhood of any connected component of G(V nN[x]) is a subset of N(X)i.e. ofN(C),soN(C)isthegreatestminimalseparatorofGincluded inN(x)forinclusion.

Conversely,wesupposethatxisanOCF-vertexofGandthesetofminimal separatorsofGincludedinN(x)hasagreatestelementS

max

forinclusion. LetX =N[x]nS

max

. Weimmediatelyverifythat X isacliquemodule of G,withN(X)=S

max

. LetC betheconnectedcomponentofG(V nN[x]) such that S

max

=N(C). X and C are distinct connected components of G(V nN(X))such that N(X)=N(C),soN(X)isaminimalseparatorof GandthereforeX isamoplexof G. 2

 WegivetheproofsofProperty4.4andTheorem4.5:

Property4.4ForanychordalgraphGandanylabellingstructureL,ifthe orderonlabelsistotalthenanyL-MLSordering ofGis amoplexordering ofG.

Proof: WehaveonlytopointoutthatbyCorollary4.3anyL-MLSordering of anon-clique chordal graphGends on amoplex X of Gand that the restrictionof tothechordalsubgraphG(V nX)isitselfaL-MLSordering ofthissubgraph,andtheprooffollowsbyinductiononjVj. 2

Theorem4.5ForanygraphGandanylabelling structureL, iftheorder onlabelsis totalthen any L-MLSMordering of Gis amoplexordering of G.

Proof: Let be a L-MLSM ordering of G. Foranyk from 1to n, let G

k

denotethecurrentgraphofthesimplicialelimination gamejust before processing (k),letV

k

beitssetofvertices,i.e. V k =f (k);:::; (n)g,and k be the restrictionof to V k

. To provethat is amoplex ordering of G,itis suÆcienttoprovethatforanyk from1ton suchthat G

k isnota clique, k endsonamoplexofG k

. Letk,1kn,suchthat G k

isnota cliqueandletus showthat

k endsonamoplexofG k . ByProperty 2.4 is aL-MLS orderingof G + ,so k is aL-MLS ordering of G + (V k ), which is equalto (G k ) + k

. As is ameo of G, its restriction

k

is ameoofG k

(otherwise there would existanother ordering k

onV k thatproducessomebetter ll-inthan

k

,andthereforeanordering onV that produces somebetter ll-in than , acontradiction). So,(G

k ) + k is a minimaltriangulationofG k andtherefore,asG k isnotaclique,(G k ) + k is notacliqueeither. ItfollowsfromCorollary4.3that

k endsonamoplex of (G k ) + k

, which is also a moplex of G k

since any moplex of a minimal triangulationofagraphisamoplexofthisgraph[2]. 2

(17)

Lemma5.4 We consider aMLS executiononagraph Gcomputing and vertices u;v of G such that

1 (u) > 1 (v). If l u (v)  l u ( (1)) and every neighbor t of v such that

1 (u) 1 (t)> 1 (v) isa neighbor of (1)thenl v (v)=l v

( (1)) andevery neighbor t of (1) suchthat 1 (u) 1 (t)> 1 (v)isaneighbor ofv.

Proof: Ateverystepbetweennumberinguand numberingv, ifthelabel ofvisincrementedthenthelabelof (1)isincrementedtoo,soby(ls1)and (ls2)theinequalityl(v)l( (1))ispreserved,andsincel(v)ismaximaljust beforenumberingv,l

v (v)=l

v

( (1)). Nowifsomevertextnumberedbefore uandvwasaneighborof (1)butnotofvthentheinequalitywouldbecome l(v)l( (1)),whichwouldbepreserveduntilnumberingv,acontradiction. 2

Theorem4.6For anygraphG, anylabellingstructureLand anyL-MLS ordering ofG,iftheorderonlabelsistotalthen (1)isanOCF-vertexof G.

Proof: Wede nethepartitionofV intosubsetsT 0 i

,1ip 0

,whichisa variantofthepartition ofV intothin slices. Foranyiin [1::p

0 ], -t 0 i = (n)ifi=1,otherwiset 0 i

isthe rst vertexnumberedafter c 0 i 1 not belongingtoN[C 0 i 1 ], - c 0 i =t 0 i ift 0 i 62 N( (1)), otherwisec 0 i

is the rst vertexnumbered after t 0 i notbelongingtoN( (1))(c 0 p 0 = (1)), -C 0 i

isthecomponentof(G; ) containingc 0 i (C 0 p 0 =f (1)g), - T 0 i

is the set of vertices t such that 1 (t 0 i )  1 (t) and (if i < p 0 ) 1 (t) > 1 (t 0 i+1

), so by de nition, the subsets T 0 i

are numbered in in-creasingorder.

Notethatforanyiin[1::p 0 ]T 0 i N( (1))[C 0 i

andthattheremaybedistinct i,j suchthat C 0 i =C 0 j ,sop 0

canbegreaterthanthenumberofcomponents of(G; ).

We rst showthat for anyi in [1::p 0

], l( (1)) is maximumamong labelsof unnumbered vertices just before numbering t

0 i

. The proof goes by induc-tionon i. It obviously holds fori =1. We suppose that itholds for some i < p 0 . l t 0 i (t 0 i+1 )  l t 0 i ( (1)). By Lemma 5.4 with u = t 0 i and v = t 0 i+1 l t 0 i+1 (t 0 i+1 )=l t 0 i+1 ( (1)),andsincel(t 0 i+1

)ismaximumatthatstep(because theorderistotal)l( (1))ismaximumtoo.

Nowlety;z2N( (1))suchthat 1

(y)> 1

(z). Letusshowthat either y2N(z)orthereissomej<p 0 suchthaty;z2N(C 0 j ). Leti2[1::p 0 ]such thaty2T 0 i .

Firstcase: 9j2[i::p 0 1]j z2N(C 0 j ) Letj 0

bethe smallestintegersuchthat ij 0 andC 0 j 0 =C 0 j . If 1 (c 0 j 0 ) 1 (y) then 1 (c 0 j 0 )  1 (y) > 1 (t 0 i+1 )  1 (t 0 j 0 +1 ), so by de ni-tion oft 0 j 0 +1 , y 2N(C 0 j 0 ). Otherwise 1 (t 0 i )  1 (y)> 1 (c 0 j 0 ) andby Lemma 5.4 with u=t 0 i and v =c 0 j 0 , y 2 N(c 0 j 0 ), soy 2 N(C 0 j 0 ). Thus in bothcasesy;z2N(C 0 j ). Secondcase: 8j2[i::p

0 1],z62N(C 0 j ) 1 (t 0 i )  1 (y) > 1 (z) so by Lemma 5.4 with u = t 0 i and v = z,

(18)

Hence (1)isanOCF-vertexoftheinputgraph. 2

 WenowpresenttheproofofCharacterization4.7:

Characterization4.7 ForanygraphGandanyvertexx ofG,x is MNS-terminali x issimplicialand theneighborhoods ofconnectedcomponents ofG(V nN[x]) aretotallyorderedbyinclusion.

Wewillusethefollowingwell-knownpropertyofchordalgraphs:

Property 5.5 (con uence vertex property) Let G = (V;E) be a chordal graph, let C be a connected subset such that N(C) isa clique. Then 9z 2 C j N(C)N(z).

Proof: (ofCharacterization4.7)If xisMNS-terminalthen xis simplicial since MNS computes a peo of G and by Theorem 3.8 neighborhoods are totallyorderedbyinclusion.

Conversely,wesupposethatxissimplicialandneighborhoodsofconnected componentsofG(VnN[x])aretotallyorderedbyinclusion.Let(C

1 ;C 2 ;:::;C q ) beanorderingonthecomponentsofG(V nN(x))suchthat 8j2[1::q 1], N(C j )  N(C j+1 ), with C q = fxg. . For any j 2 [1::q], let S 0 j = N(C j )nN(C j 1 ),letT 0 j =S 0 j [C j andletc j 2C j jN(C j )N(c j )(c j exists byProperty5.5becauseasxissimplicial,C

j

issimplicialtoo). Letusshow thatthereisaMNSexecutiononGnumberingthesetsT

0 j

inincreasingorder. Weprovebyinduction onj thatthere isaMNSexecutiononGnumbering successivelyT 0 1 , T 0 2 , ...,T 0 j

rst. It obviouslyholds for j =0. We suppose that it holds for j 1for some j  q. After numbering T

0 1 , T 0 2 , ..., T 0 j 1 , foreveryunnumberedvertext,l(t)=NN(t)N(C

j 1 )N(C j )N(c j ). Soc j

canbechosenat thatstep. It remainstoshowthat at anystepafter numberingc

j

andsome, but notall, vertices ofT 0 j

, it is possible to choose thenextvertexinT

0 j

. Lety beanunnumberedvertexinT 0 j

,beac j

y-path whoseinternalverticesarein C

j ,lety

0

bethe rstunnumberedvertexof from c

j andz

0

bethe numberedvertexpreceding y 0 on. Thenz 0 2l(y 0 ). Let y 00

be an unnumbered vertex whose label is maximal among those of unnumbered vertices t such that z

0

2 l(t). So the label of y 00

is maximal amongthoseofallunnumberedverticesandy

00 2T 0 j (sincez 0 2C j \N(y 00 )). Thusy 00

canbechosenatthatstep. HenceT 0 1 ,T 0 2 ,...,T 0 q canbenumberedin increasingorder. AstheverticesofT

0 q

form acliquemodule(sincex is sim-plicial)theycanbenumberedinanarbitraryorderandxcan benumbered last. 2

Figure

Figure 1: Extremity properties of Algorithms MLS and MLSM. These properties
Figure 3: The slices dened by an execution of MCS on a chordal graph.
Figure 4: This MNS ordering does not number slices in increasing order on this
Figure 5: MCS on a non-chordal graph. The super-components of (G; ) (which
+2

Références

Documents relatifs

University of Zurich (UZH) Fall 2012 MAT532 – Representation theory..

3.3.3 Unicyclic graphs with exactly three vertices of maximum degree 3 whose exactly two belongs to the

Graph minors play a central role in parameterized complexity, and the edge contraction operation in turn is essential in the study of graph minors: a graph H is a minor of a graph G

As stated in the beginning, the real purpose of these notes is not to simply repeat the standard calculation of the derivative of the logarithm function, but to replace the

I present an efficient algorithm which lists the minimal separators of a 3- connected planar graph in O(n) per separator.. Keywords: 3-connected planar graphs, minimal

We will now prove that the complexity of finding the a, ∗-minimal separators of a planar graph is O(n/separator) and that the complexity of finding all the minimal separators of

However, since the graph is 3-connected planar, anyone of these neighbourhoods is necessarily maximal for inclusion, because if some neighbourhood S was a strict subset of some

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des