scieee Science in your language
[en] (orig)

Rebuilding convex sets in graphs

Abstract

The usual distance between pairs of vertices in a graph naturally gives rise to the notion of an interval between a pair of vertices in a graph. This in turn allows us to extend the notions of convex sets, convex hull, and extreme points in Euclidean space to the vertex set of a graph. The extreme vertices of a graph are known to be precisely the simplicial vertices, i.e., the vertices whose neighborhoods are complete graphs. It is known that the class of graphs with the Minkowski–Krein–Milman property, i.e., the property that every convex set is the convex hull of its extreme points, is precisely the class of chordal graphs without induced 3-fans. We define a vertex to be a contour vertex if the eccentricity of every neighbor is at most as large as that of the vertex. In this paper we show that every convex set of vertices in a graph is the convex hull of the collection of its contour vertices. We characterize those graphs for which every convex set has the property that its contour vertices coincide with its extreme points. A set of vertices in a graph is a geodetic set if the union of the intervals between pairs of vertices in the set, taken over all pairs in the set, is the entire vertex set. We show that the contour vertices in distance hereditary graphs form a geodetic set.

Read accessible full text

Rebuilding convex sets in graphs

Author: Cáceres González, José; Márquez Pérez, Alberto; Oellermann, Ortrud R.; Puertas González, María Luz
Year: 2005
DOI: 10.1016/j.disc.2005.03.020
Source: https://idus.us.es/bitstreams/abcde7f4-2f3c-49a6-89f9-1a1d6a13c331/download
Disc e e Ma hema ics 297 (2005) 26–37
www.else ie .com/loca e/disc
Rebuilding con ex se s in g aphs
José Cáce esa,1, Albe o Má quezb,2, O ud R. Oelle mannc,3,
Ma ía Luz Pue asa,4
aDepa men o S a is ics and Applied Ma hema ics, Uni e si y o Alme ia, Spain
bDepa men o Applied Ma hema ics I, Uni e si y o Se ille, Spain
cDepa men o Ma hema ics and S a is ics, The Uni e si y o Winnipeg, 515 Po age A enue, Winnipeg,
Mani oba, Canada R3B 2E9
Recei ed 17 Decembe 2002; ecei ed in e ised o m 18 June 2004; accep ed 1 Ma ch 2005
A ailable online 11 July 2005
Abs ac
Theusualdis ancebe weenpai so e icesinag aphna u allygi es ise o heno iono anin e al
be weenapai o e icesin ag aph.Thisin u nallowsus oex end heno ionso con exse s,con ex
hull, and ex eme poin s in Euclidean space o he e ex se o a g aph. The ex eme e ices o a
g aph a e known o be p ecisely he simplicial e ices, i.e., he e ices whose neighbo hoods a e
comple e g aphs. I is known ha he class o g aphs wi h he Minkowski–K ein–Milman p ope y,
i.e., he p ope y ha e e y con ex se is he con ex hull o i s ex eme poin s, is p ecisely he class
o cho dal g aphs wi hou induced 3- ans.We define a e ex o be a con ou e ex i he eccen ici y
o e e y neighbo is a mos as la ge as ha o he e ex. In his pape we show ha e e y con ex
se o e ices in a g aph is he con ex hull o he collec ion o i s con ou e ices. We cha ac e ize
hose g aphs o which e e y con ex se has he p ope y ha i s con ou e ices coincide wi h i s
ex eme poin s.A se o e ices in a g aph is a geode ic se i he union o he in e als be ween pai s
o e ices in he se , aken o e all pai s in he se , is he en i e e ex se . We show ha he con ou
e ices in dis ance he edi a y g aphs o m a geode ic se .
© 2005 Else ie B.V. All igh s ese ed.
Keywo ds: Eccen ici y; Con ou e ex; Dis ance he edi a y g aph; Con ex hull; Geode ic se
E-mail add ess: o.oelle [email protected] (O. Oelle mann).
1Resea ch suppo ed in pa by FQM 164 g an .
2Resea ch suppo ed in pa by FQM 305 g an .
3Resea ch suppo ed by an NSERC g an Canada.
4Resea ch suppo ed in pa by BFM2001-2474 g an .
0012-365X/$-see on ma e © 2005 Else ie B.V. All igh s ese ed.
doi:10.1016/j.disc.2005.03.020
J. Cáce es e al./Disc e e Ma hema ics 297 (2005) 26–37 27
1. In oduc ion
The s udy o abs ac con exi y began in he ea ly fi ies wi h he sea ch o an axiom
sys em ha defines a con ex se and in some way gene alizes he classical concep o a
Euclidean con ex se . Nume ous con ibu ions o his opic ha e been made. An ex ensi e
su ey o his subjec can be ound in [20].
Among he wide a ie y o s uc u es ha ha e been s udied unde abs ac con exi y a e
me ic spaces, o de ed se s o la ices and g aphs, he las being he ocus o his pape . We
now gi e a b ie in oduc ion o abs ac con exi y as i pe ains o g aphs. Le Vbe a fini e
se and Ma fini e collec ion o subse s o V.Then Mis an alignmen o Vi and only i M
is closed unde in e sec ion and con ains bo h Vand he emp y se . I Mis an alignmen o
V, hen he elemen s o Ma e called con ex se s and he pai (V, M)is called an aligned
space.I S⊆V, hen he con exhullo S, deno ed byCH(S),is he smalles con exse ha
con ains S. Suppose X∈M. Then, x∈Xis an ex eme poin o Xi X−{x}∈M. The
collec ion o all ex eme poin s o Xis deno ed by ex(X).Acon ex geome y on a fini e se
is an aligned space wi h he addi ional p ope y ha e e y con ex se is he con ex hull o
i s ex eme poin s. This p ope y is e e ed o as he Minkowski–K ein–Milman p ope y.
Se e al abs ac con exi ies associa ed wi h he e ex se o a g aph a e well known (see
[10]).Thei s udy is o in e es in compu a ional geome y and has some di ec applica ions
o o he a eas such as, o example, game heo y (see [4]).
Fo g aph e minology we ollow [14]; excep ha we use e ex ins ead o poin and
edge ins ead o line. All g aphs conside ed he e a e connec ed, fini e, simple, unweigh ed
and undi ec ed. The dis ance be ween a pai o e ices u, o Gis he leng h o a sho es
u– pa h in Gand is deno ed by dG(u, ) o d(u, ) i Gis clea om con ex . The in e al
be ween a pai u, o e ices in a g aph Gis he collec ion o all e ices ha lie on some
sho es u– pa h in Gand is deno ed by IG[u, ]o I[u, ]i Gis unde s ood. In e als
in g aphs ha e been s udied ex ensi ely (see [2,17,18]) and play an impo an ole in he
s udy o se e al classes o g aphs such as he P olemaic g aphs (see [16]) o block g aphs.
A subse So e ices o a g aph is said o be g-con ex i i con ains he in e al be ween
e e y pai o e ices in S. I is no di ficul o see ha he collec ion o all g-con ex se s is
an alignmen o V. We hus e e o he g-con ex se s simply as con ex se s.A e ex in a
g aph is simplicial i i s neighbo hood induces a comple e subg aph. I can eadily be seen
ha pis an ex eme poin o a con ex se Si and only i pis simplicial in he subg aph
induced by S. I is ue, in gene al, ha he con ex hull o he ex eme poin s o a con ex
se Sis con ained in S, bu equali y holds only in special cases. In [10] i is shown ha a
g aph has he Minkowski–K ein–Milman p ope y i and only i i has no induced cycles
o leng h bigge han 3 and has no induced 3- an (see Fig. 1). Fo ano he mo e ecen and
excellen e e ence ex con aining ma e ial on g aph con exi y see [6].
I a g aph Ghas he Minkowski–K ein–Milman p ope y and Sis a con ex se o V (G),
hen we can ebuild he se S om i s ex eme e ices using he con ex hull ope a ion.
Since his canno be done wi h e e y g aph, using only he ex eme e ices o a gi en
con ex se S, i is na u al o ask i i is possible o ex end he se o ex eme e ices o S o
a se ha allows us o ebuild Susing he e ices in his ex ended se and he con ex hull
ope a ion. In Sec ion 2 we answe his ques ion in he a fi ma i e using he collec ion o
‘con ou e ices’o a se . To his end, le Sbe a se o e ices in a g aph Gand ecall ha
28 J. Cáce es e al./Disc e e Ma hema ics 297 (2005) 26–37
c
b
d
a
Fig. 1. A 3- an.
he eccen ici y in So a e ex u∈Sis gi en by eccS(u) =max{d(u, ) : ∈S}and a
e ex ∈S o which d(u, ) =eccS(u) is called an eccen ic e ex o uin S. In case
S=V (G), we deno e eccS(u) by ecc(u). A e ex u∈Sis said o be a con ou e ex o
Si eccS(u)⩾eccS( ) o e e y neighbo o uin S. The se o all con ou e ices o S
is called he con ou se o Sand is deno ed by C (S).I S=V (G), he subg aph induced
by he con ou se o Sis called he con ou o Gand is deno ed by C (G). In Sec ion 3 we
es ablish s uc u al p ope ies o con ou e ices and cha ac e ize hose g aphs ha a e he
con ou o some o he g aph using a cons uc ion simila o he one used in [3].
In o de o find he con ex hull o a se Sone begins by aking he union o he in e als
be ween pai s o e ices o S, aken o e all pai s o e ices in S. We deno e his se by
IG[S]o I[S], i.e., I[S]={u, }⊆SI[u, ]and call i he geode ic closu e o S. One hen
epea s his p ocedu e wi h he new se and con inues un il, o he fi s ime, one eaches a
se T o which he geode ic closu e is he se i sel , i.e., T=I[T]. This hen is he con ex
hull o S. I his p ocedu e only has o be pe o med once, we say ha he se Sis a geode ic
se o i s con ex hull. In gene al a subse So a con ex se Tisageode ic se o Ti
I[S]=T. The no ion o a geode ic se o he e ex se o a g aph was fi s defined in [7].
InSec ion4we ocusongeode icse sin‘dis ancehe edi a yg aphs’.Wefi s discusshe e
how heseg aphsa e ela ed o heg aphswi h he Minkowski–K ein–Milmanp ope y and
how he esul s o Sec ion 4 ex end esul s known o he las class. Howo ka [15] defined
a connec ed g aph G o be dis ance he edi a y i o e e y connec ed induced subg aph
Ho Gand e e y wo e ices u, in H,dH(u, ) =dG(u, ). In he same pape se e al
cha ac e iza ions o his class o g aphs a e gi en. We s a e he e only one o hese which
we will use in his pape .
Theo em 1. A connec ed g aph Gis dis ance he edi a y i and only i e e y cycle in G
o leng h a leas 5has a pai o c ossing cho ds.
Fu he use ul cha ac e iza ions o his class o g aphs we e es ablished in [1,9,13].
Apa omha ingelegan cha ac e iza ions,dis ancehe edi a yg aphspossesso he use ul
p ope ies. I is a class o g aphs o which se e al NP-ha d p oblems ha e polynomial
solu ions. Fo example he S eine p oblem o g aphs, which is known o be NP-ha d
(see [11]), can be sol ed in polynomial ime in dis ance he edi a y g aphs (see [5,8,12]).
Mo eo e , hese g aphs a e S eine dis ance he edi a y as was shown in [9]; i.e., he S eine
dis ance o a se o e ices is he same, in any connec ed induced subg aph ha con ains
i , as i is in he g aph i sel .
J. Cáce es e al./Disc e e Ma hema ics 297 (2005) 26–37 29
The class o dis ance he edi a y g aphs also p ope ly con ains he g aphs ha possess he
Minkowski–K ein–Milman p ope y since a g aph is cho dal wi hou an induced 3- an i
and only i i is a dis ance he edi a y g aph wi hou an induced 4-cycle. I was shown in [10]
ha in a cho dal g aph e e y non-simplicial e ex lies on a cho dless pa h be ween wo
simplicial e ices. I Gis a cho dless g aph wi hou an induced 3- an, hen Gis dis ance
he edi a y and hus e e y induced pa h is necessa ily a sho es pa h. Hence he simplicial
e ices o a con ex se Sin a g aph wi h he Minkowski–K ein–Milman p ope y is a
geode ic se o S. In Sec ion 4 we show ha he con ou e ices o a dis ance he edi a y
g aph o m a geode ic se o he g aph. In [19] i shown ha he con ou e ices can be
used o find minimum S eine geode ic se s o dis ance he edi a y g aphs.
2. The con ou se o a g aph
In his sec ion we will show ha he con ou se o a con ex se So e ices in a g aph
Gcan be used o ebuild he se by finding i s con ex hull, in he same way ha ex eme
e ices a e used in cho dal g aphs in [10]. Mo eo e , we cha ac e ize hose g aphs ha ing
he p ope y ha he ex eme e ices and he con ou e ices o e e y con ex se coincide.
Fi s we show ha he con ou se o Gcon ains all he ex eme e ices.
Lemma 2. Le Gbe a g aph and S⊆V (G).Then C (S) con ains all ex eme e ices
o S.
P oo . Le u∈Sbe an ex eme e ex o S. Then uis a simplicial e ex o S.Wenow
show ha uis a con ou e ex o S. Le be a neighbo o uin Sand e∈San eccen ic
e ex o in S, i.e., d( , e)=eccS( ). Suppose ha d(u, e)=d( , e)−1 and le
Pbe a sho es u– epa h. Then he e ex ollowing uon P, say w,isno . Since uis
simplicial, and wmus be adjacen . Howe e , hen d(u, e)⩾d( , e), a con adic ion.
So eccS(u)⩾d(u, e)⩾d( , e)=eccS( ) and he e o e uis a con ou e ex o S.
The ela ionshipbe weencon ou e icesandex eme e icesise enclose o heclass
o dis ance he edi a y g aphs wi hou induced 4-cycles.The nex esul is a cha ac e iza ion
o con ou e ices in g aphs wi h he Minkowski–K ein–Milman p ope y ha esembles
he cha ac e iza ion o simplicial e ices.
P oposi ion 3. Le Gbe a dis ance he edi a y g aph wi hou induced 4-cycles. A e ex
x∈V (G) is acon ou e ex o Gi and only i eachneighbo o xwhichis on asho es
pa h be ween xand some eccen ic e ex o xsa isfies N[x]⊆N[ ].
P oo . I Gis comple e, he esul is immedia e. Suppose now ha Gis no comple e. Then
no con ou e ex o Gcan ha e eccen ici y 1. So i xis a con ou e ex and xeis an
eccen ic e ex o x, hen d(x,xe)⩾2. Le P:(x=)y0y1...y
k(=xe)be a sho es x–xe
pa h. Suppose u= y1is a neighbo o x. Then uis no on Pand uP canno be a sho es
u–xepa h; o he wise, ecc(u) > ecc(x) which is no possible since xis a con ou e ex.
Since Gis dis ance he edi a y, he subg aph induced by uand he e ices o Pcon ains a
30 J. Cáce es e al./Disc e e Ma hema ics 297 (2005) 26–37
sho es u–xepa h. Hence he e is a cho d be ween uand some e ex on Pwhose dis ance
om xis less han o equal o 2. I uy1is a cho d, hen u∈N(y1)as desi ed. I uis a
neighbo o y2, hen he 4-cycle xy1y2ux mus ha e a cho d. So uy1is an edge and again
u∈N(y1).
Con e sely, suppose ha xhas he p ope y ha each o i s neighbo s which is on a
sho es pa h be ween xand some eccen ic e ex o xsa isfies N[x]⊆N[ ]. Suppose x
has a neighbo ysuch ha ecc(x) < ecc(y). Then xlies on a sho es pa h Pbe ween yand
an eccen ic e ex ye o y.Soyeis also an eccen ic e ex o x. By ou hypo hesis yis
a neighbo o he e ex adjacen o xin P−y. This is no possible as Pis a sho es y–ye
pa h. So ecc(y)⩽ecc(x) and xis a con ou e ex o G.
Rema k 4. The abo e esul does no hold o all cho dal g aphs. Take o example he
3- ano Fig.1. Fo hisg aphbo h heneighbo s, o ei he oneo he wosimplicial e ices,
lieonsomesho es pa h oaneccen ic e exbu hei closedneighbo hoodsa eno equal.
Howe e , he con e se o he abo e esul holds o all connec ed g aphs G, i.e., i a e ex
x∈V (G) has he p ope y ha o each neighbo o xwhichis ona sho es pa h be ween
xand some eccen ic e ex o x,N[x]⊆N[ ], hen xis a con ou e ex.
The ollowing esul shows ha he con ex hull o he con ou se o a con ex se o
e ices in a g aph is he en i e se , wi hou any es ic ion on he g aph. So his esul is
simila o he Minkowski–K ein–Milman p ope y and holds o all g aphs.
Theo em 5. Le Gbe a g aph and Sa con ex subse o e ices. Then S=CH(C (S)).
P oo . Suppose, o hecon a y, ha S= CH(C (S)).SinceSisacon exse ,CH(C (S)) ⊆
S. So, by ou assump ion, S−CH(C (S)) =∅. Le u∈S−CH(C (S)) be such ha
ecc(u)⩾ecc( ) o all ∈S−CH(C (S)). Since u/∈C (S), he e exis s a neighbo
o uin Ssuch ha eccS( ) > eccS(u) and, by ou choice o u, he e ex belongs o
CH(C (S)).
Le e∈Sbe an eccen ic e ex o in S, i.e., d( , e)=eccS( ). No e ha in his
case eccS( e)⩾eccS( ) > eccS(u) and e∈CH(C (S)). The e o e d(u, e)⩽eccS(u) <
eccS( ) =d( , e)and so d(u, e)+1⩽d( , e).
Le Pbe a sho es e–upa h in S.Then P ollowed by he edge u ,isa e– pa h whose
leng h is d(u, e)+1⩽d( , e). So i is a sho es pa h be ween eand ha con ains u.
This con adic s he ac ha u/∈CH(C (S)).
In g aphs wi h he Minkowski–K ein–Milman p ope y, he se o ex eme e ices o a
con ex se Sis minimal in he sense ha any ex eme poin o Sis no in he con ex hull
o a subse o S ha does no con ain i . Un o una ely he con ou se does no sha e his
p ope y in gene al as can be seen in he example o Fig. 1. In his case he con ou is he
se {a, b, c, d}, bu CH({a,b,d})=CH({a, b, c, d}).
Howe e , he e a e examples whe e he con ou e ices a e a minimal se in a simila
way ha ex eme e ices a e. In he g aph o Fig. 2 wi h S=V (G), he con ou se is
C (S) ={a, b, c, d}and he con ex hull o any p ope subse o C (S) is a p ope subse
o S.

J. Cáce es e al./Disc e e Ma hema ics 297 (2005) 26–37 31
acbd
Fig. 2. G aph wi h a minimal con ou .
u
w
x
y
Fig. 3. A da .
We now cha ac e ize hose connec ed g aphs o which e e y con ex se has he p ope y
ha i s con ou e ices coincide wi h i s ex eme poin s.
Theo em 6. A connec ed g aph Ghas he p ope y ha C (S) =ex(S) o all con ex se s
So e ices o Gi and only i Ghas he Minkowski–K ein–Milman p ope y and does no
con ain a da as induced subg aph (see Fig. 3).
P oo . SupposeGhas he p ope y ha C (S)=ex(S) o all con exse sSo e iceso G.
Le Sbe any con ex se o G.Then we know ha Sis he con ex hull o i s con ou e ices.
SinceC (S)=ex(S) i ollows ha Sis also hecon exhull o i sex eme e ices.HenceG
has heMinkowski–K ein–Milmanp ope y.The e o eGischo dalwi hou induced3- ans.
Hence Gis dis ance he edi a y wi hou induced 4-cycles. Suppose Ghas a da as induced
subg aph.Label he e iceso suchada asinFig.3.Le Xbe he e icesinI[u, ]−{u, }.
Then he subg aph Xinduced by Xis comple e; o he wise, Ghas an induced 4-cycle,
con adic ing he ac ha Gis cho dal.Also i x∈X, hen {u, w, y, x, } is a connec ed
subg aph o Gand since Gis dis ance he edi a y i con ains a sho es w– pa h as well as
a sho es y– pa h. Hence wx,yxa e edges o G.Sowand ya e adjacen o e e y e ex
in X. Since d(w,y) =2, i ollows ha I[y,w]−{w, y} is a comple e g aph. Suppose
I[w, y]con ains e ices no in X∪{u}, say u∈I[w, y]−(X ∪{u}). Then d(u, )=2.
Since Gcon ains no induced 4-cyclesI[u, ]−{u, } is comple e. I I[u, ]=X,
hen he e is some e ex such ha u /∈E(G) bu u , ∈E(G). Hence uu is an
induced u– pa h o leng h 3. This con adic s he ac ha Gis dis ance he edi a y. Thus
S=CH({u, , w, y, x})=I[w, y]∪I[u, ]. Hence all e ices o Sexcep hose in Xa e
con ou e ices o S. This con adic s he hypo hesis since uis a con ou e ex o S ha
is no an ex eme poin o S, i.e., uis no simplicial.
32 J. Cáce es e al./Disc e e Ma hema ics 297 (2005) 26–37
Fo he con e se, suppose Ghas he Minkowski–K ein–Milman p ope y and does no
con ain a da as induced subg aph. Suppose Sis a con ex se ha has a con ou e ex
u ha is no simplicial. Then Sis no comple e and uis adjacen wi h a pai w, y o
non-adjacen e ices. Le X=N(u) ∩I[u, ]. Since Ghas he Minkowski–K ein–Milman
p ope y i can be shown ha Xis comple e. So wand ycanno bo h belong o I[u, ].I
∈(N(u)−I[u, ]), hen mus be adjacen oe e y e exinXsinceuis acon ou e ex
o Gand since Gis dis ance he edi a y. So nei he wno ybelongs o X.I uu1u2...u
e=
is a sho es u– pa h in G, hen e⩾2 and nei he wno yis adjacen wi h u2. Hence
{u, w, y, u1,u
2} is isomo phic o a da , con a y o hypo hesis. Thus C (S) =ex(S).
Cha ac e izing g aphs G o which C (G) =ex(G) appea s much mo e di ficul . Any
connec ed g aph His an induced subg aph o a g aph Gwi h his p ope y. To see his,
ake |V(H)|pai wise e ex disjoin , non- i ial cliques and pai o each e ex o Hin a
one- o-one manne wi h one o hese cliques. Now iden i y each e ex o Hwi h exac ly
one e exin heclique ha i hasbeenpai edo wi h and le Gbe he esul ingg aph.Then
Ghas he p ope y ha C (G) =ex(G). I ollows ha hose g aphs o which he con ou
se and he collec ion o ex eme poin s coincide can ha e induced cycles o a bi a ily la ge
o de . Howe e , no e e y g aph wi h his p ope y can be cons uc ed in his manne . Take
o example he g aph ob ained om he 6-cycle 1,
2,...,
6,
1by joining a lea u1 o
1and a lea u4 o 4. Then he esul ing g aph Ghas he p ope y ha C (G) =ex(G) bu
Gis no ob ained by he abo e cons uc ion.
3. G aphs wi h a gi en con ou se
In hissec ionwecha ac e ize hoseg aphswhicha e hecon ou o someo he g aph.The
ob ious ela ionshipbe weencon ou andpe iphe al e icesallowus ouse hecons uc ion
used in [3] o also cha ac e ize hose g aphs ha a e he con ou o some g aph.
Lemma 7. Le Gbe a connec ed g aph and Ca componen o i s con ou . Then all e ices
in Cha e he same eccen ici y.
The ollowing esul ells us which g aphs a e no he con ou o any g aph.
P oposi ion 8. I His a connec ed,non-comple e g aph wi h adius 1, hen His no he
con ou o any g aph.
P oo . Le Hbe a connec ed, non-comple e g aph wi h adius 1. Then some e ex u∈
V(H)is a neighbo o e e y o he e ex in H. Since His no comple e he e a e wo non-
adjacen e ices and .Suppose he eexis sag aphGsuch ha His hesubg aphinduced
by i s con ou .Then Gmus be connec ed, since His connec ed. So, using Lemma 7, e e y
e ex in Hhas he same eccen ici y, say k. No e ha ecc( )⩾2, because dG( , )=2,
so k⩾2.
Le w∈V (G) be such ha ecc(u) =d(u, w) =k. Then w/∈C (G), since k⩾2.
So he e exis s a neighbo w1o wsuch ha ecc(w1)>ecc(w)⩾k. Again w1/∈C (G),
J. Cáce es e al./Disc e e Ma hema ics 297 (2005) 26–37 33
H1
Hi
H
k
2 M
Fig. 4. A disconnec ed con ou se .
because i s eccen ici y is bigge han k. So he e exis s a neighbo w2o w1such ha
ecc(w2)>ecc(w1)>k. This p ocess canno con inue indefini ely since Gis a fini e g aph.
Howe e , hen he las e ex picked should be a con ou e ex. Since i s eccen ici y is
bigge han kwe ha e a con adic ion. 
Suppose ha His a g aph wi h adius g ea e han 1. We now desc ibe a g aph Gsuch
ha i s con ou is H, using he cons uc ion gi en in [3]. Le Gbe he join o Hand K1.
Then e e y e ex o Hhas eccen ici y 2 and he e ex o G−V(H)has eccen ici y 1.
Hence he e ices o Ha e p ecisely he con ou e ices o G.
A sligh ly di e en cons uc ion allows us o ob ain a g aph wi h gi en disconnec ed
con ou se such ha he eccen ici ies o he e icesine e ycomponen a e gi ennumbe s
a leas 2.Mo ep ecisely,le Hbeadisconnec edg aphwi hcomponen s,H1,H
2,...,H
k.
Le n1,n
2,...,n
kbe kna u al numbe s such ha n1=nk=max{n1,n
2,...,n
k}and M=
max{n1,n
2,...,n
k}⩽2min{n1,n
2,...,n
k}=2m. No e ha hese a e na u al es ic ions,
because Mwill be he diame e o he g aph Gand mwill be g ea e han o equal o he
adius. Then he e exis s a connec ed g aph Gsuch ha His he con ou o Gand he
eccen ici y o e e y e ex in each componen Hio His equal o ni. To cons uc such
a g aph Gwe begin wi h he pa h 1 2...
M+1o o de M+1. Now eplace 1by H1
and M+1by Hkso ha all e ices in H1a e neighbo s o 2and all e ices in Hka e
neighbo s o M.
Now, o each i,2⩽i⩽k−1, he eexis sa e ex nion hepa hsuch ha i seccen ici y
is ni−1. We now add Hi o he g aph and join all he e ices o Hi o ni(see Fig. 4).
Then ecc(ui)=ni o all ui∈Hi,andC (G) =H.
4. Con ou se s and geode ic se s in dis ance he edi a y g aphs
In his sec ion we show ha he con ou e ices o a dis ance he edi a y g aph o m a
geode ic se . I is no di ficul o see ha a se So e ices is a con ex se o a dis ance
he edi a y g aph Gi and only i Sinduces a connec ed g aph and is he union o e ices in
blocks o Gminus any collec ion o simplicial e ices om he subg aph induced by hese
blocks. The esul s o his sec ion hus show ha he con ou e ices o all con ex se s in
dis ance he edi a y g aphs a e geode ic se s o such se s.As poin ed ou in he in oduc ion
his may be iewed as an ex ension o he esul which s a es ha he simplicial e ices o
con ex se s in g aphs wi h he Minkowski–K ein–Milman p ope y a e a geode ic se o
he con ex se .
34 J. Cáce es e al./Disc e e Ma hema ics 297 (2005) 26–37
The nex esul shows ha i Gis a dis ance he edi a y g aph, hen e e y e ex has an ec-
cen ic e ex ha isacon ou e ex.Mo eo e ,i Gsa isfies heMinkowski–K ein–Milman
p ope y and i xis a e ex o G, wi h ecc(x)⩾2, hen e e y eccen ic e ex o xmus be
a con ou e ex.
Lemma 9. (1) I Gis adis ance he edi a yg aphandx∈V (G), hen he eis an eccen ic
e ex o x ha is a con ou e ex.
(2) Le Gbe a dis ance he edi a y g aph wi hou induced 4-cycles. I x∈V (G) is such
ha ecc(x)⩾2, hen each eccen ic e ex o xis a con ou e ex o G.
P oo . (1) The esul holds o all dis ance he edi a y g aphs wi h diame e a mos 2.
Suppose hus ha diam(G)⩾3. Among all eccen ic e ices o xle xebe one wi h
maximum eccen ici y. Le P:x= 0 1...
k=xebe a sho es x–xepa h. We show xe
is a con ou e ex. I his is no he case, hen xeis adjacen wi h some e ex uwhose
eccen ici y exceeds ha o xe.Thus ecc(u) > ecc(xe)⩾ecc(x) =k.Soecc(u)⩾3. We may
assume ulies on Pand ha u= k−1. Suppose ueis an eccen ic e ex o u.Then he e is
a sho es u–uepa h Q ha con ains xe. Suppose Q:u=u0u1...u
=uewhe e u0= k−1.
Thenu2isno onP.Also clea ly uu2/∈E(G).Theonly e exon P ha u2may be adjacen
ois k−2. Indeed u2 k−2isan edge; o he wise,ecc(x)⩾d(x,u2)>k.Since ecc( k−1)⩾3,
u2mus be adjacen o a e ex no on P.I u3is adjacen wi h a e ex o Pi can only be
adjacen wi h k−3; o he wise, ei he d( k−1,u
3)= 3o d(x, k)= k. Howe e , i u3 k−3
is an edge, we ha e a 6-cycle k−3 k−2 k−1 ku2u3wi hou c ossing cho ds, which is no
possible in a dis ance he edi a y g aph. So u3 k−3/∈E(G) and d( 0,u
3)=k. By ou choice
o xe= k,3⩽ecc(u3)⩽ecc( k)<ecc( k−1).Soecc( k−1)⩾4 and hence u4is no on P.
As be o e we can a gue ha he only e ex o P ha u4is possibly adjacen o is k−4.I
u4 k−4∈E(G), henasbe o eweob aina6-cycle k−4 k−3 k−2u2u3u4 k−4whichhasno
c ossing cho ds. So u4 k−4/∈E(G). Howe e , hen d(x,u4)=k+1>ecc(x) which is no
possible. So u= k−1. No e uis no adjacen wi h k−i o i⩾3; o he wise, d(x,xe)<k,
whichisno possible.Also k−2u∈E(G);o he wise,we ha ea con adic ion oou choice
o xe= k.I k−1u∈E(G), hen u2is no on P. No e ha in his case u2is adjacen wi h
a mos one e ex on P, namely, k−2.Sou3is no on P.Fo i⩾3, uicanno be on Psince
in his case ei he d(x, k)<k o d(u, ui)<i, bo h o which canno happen. Mo eo e ,
he only e ex on P ha uican be adjacen wi h (i any) is k−i; o he wise, as in he
p e ious case, we ha e a con adic ion. I u3 k−3∈E(G), hen k−3 k−2 k−1 ku2u3 k−3
is a 6-cycle wi hou c ossing cho ds. So u3 k−3/∈E(G). In his case u2 k−3∈E(G) and
d(x,u3)=k. So by ou choice o xe= k,ecc(u3)⩽ecc( k)<ecc(u)⩽ecc(u ).So3< .
I u4 k−4∈E(G), hen k−4 k−3 k−2u2u3u4 k−4is a 6-cycle wi hou c ossing cho ds. So
u4 k−4/∈E(G). Bu hen d(x,u4)=k+1 con adic ing he ac ha ecc(x) =k. Hence we
may assume k−1u/∈E(G).
I u2= k−1, i.e., u2is no on P, hen a leas one o u2 k−2and u2 k−1is an edge o G.
I u2 k−2/∈E(G), we ha e a 5-cycle wi h ou c ossing cho ds. So assume u2 k−2∈E(G).
Fo i⩾3 we may a gue, as be o e, ha uiis no on Pand ha uiis adjacen o a mos
one e ex on P, namely, k−i. Bu hen, as in he p e ious case, we ha e a con adic ion
o he ac ha ecc( ) =k. So we may assume k−1=u2. Clea ly, u3= k−2since
d(u, u3)=3= d(u, k−2). Indeed, o i⩾3, uiis no on Pand uiis adjacen wi h a mos