• No results found

Locally constrained homomorphisms on graphs of bounded treewidth and bounded degree

N/A
N/A
Protected

Academic year: 2021

Share "Locally constrained homomorphisms on graphs of bounded treewidth and bounded degree"

Copied!
10
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Contents lists available atScienceDirect

Theoretical

Computer

Science

www.elsevier.com/locate/tcs

Locally

constrained

homomorphisms

on

graphs

of

bounded

treewidth

and

bounded

degree

Steven Chaplick

a

,

1

,

Jiˇrí Fiala

b

,

2

,

Pim van ’t Hof

c

,

Daniël Paulusma

d

,

,

Marek Tesaˇr

b

aInstitutfürMathematik,TUBerlin,Germany

bDepartmentofAppliedMathematics,CharlesUniversity,Prague,CzechRepublic cDepartmentofInformatics,UniversityofBergen,Norway

dSchoolofEngineeringandComputingSciences,DurhamUniversity,UK

a

r

t

i

c

l

e

i

n

f

o

a

b

s

t

r

a

c

t

Articlehistory:

Received 5 September 2013

Received in revised form 26 October 2014 Accepted 16 January 2015

Available online 21 January 2015 Keywords:

Computational complexity

Locally constrained graph homomorphisms Bounded treewidth

Bounded degree

AhomomorphismfromagraphG toagraphH islocallybijective,surjective,orinjective if its restriction to the neighborhood of every vertex of G is bijective, surjective, or injective,respectively. We prove thatthe problemsof testingwhether agivengraph G

allowsahomomorphismtoagivengraphH thatislocallybijective,surjective,orinjective, respectively,are NP-complete,evenwhenG haspathwidthatmost5,4,or2,respectively, orwhenbothG andH havemaximumdegree 3.Wecomplementthesehardnessresultsby showingthatthethreeproblemsarepolynomial-timesolvableifG hasboundedtreewidth andinadditionG orH hasboundedmaximumdegree.

©2015ElsevierB.V.All rights reserved.

1. Introduction

Allgraphsconsideredinthispaperarefinite,undirected,andhaveneitherself-loopsnormultipleedges.A

graph

homo-morphism from agraph

G

= (

VG

,

E

G

)

toagraph H

= (

VH

,

EH

)

isamapping

ϕ

:

VG

VH thatmapsadjacentverticesof

G

to adjacentverticesof H , i.e.,

ϕ

(

u

)

ϕ

(

v

)

EH whenever

uv

EG.Thenotionofagraphhomomorphismiswell studiedin

theliteratureduetoitsmanypracticalandtheoreticalapplications;werefertothetextbookofHellandNešetˇril[29]fora survey.

Wewrite

G

H to indicatetheexistenceofahomomorphismfrom

G to

H . Wecall

G the guest graph and H the host

graph. We denotetheverticesof H by 1

,

. . . ,

|

H

|

andcall them

colors.

The reasonfordoingthisisthat graph homomor-phismsgeneralizegraphcolorings:thereexistsahomomorphismfromagraph

G to

acompletegraphon

k vertices

ifand onlyif

G is k-colorable.

Theproblemoftestingwhether

G

H for twogivengraphs

G and H is

calledthe Hom problem.If onlytheguestgraphispartoftheinputandthehostgraphis

fixed,

i.e.,notpartoftheinput,thenthisproblemisdenoted

This paper is supported by the Natural Sciences and Engineering Research Council of Canada (NSERC), the Research Council of Norway (197548/F20), EPSRC (EP/G043434/1) and the Royal Society (JP100692). An extended abstract of it appeared in the Proceedings of FCT 2013[10].

*

Corresponding author.

E-mailaddresses:chaplick@math.tu-berlin.de(S. Chaplick), fiala@kam.mff.cuni.cz(J. Fiala), pim.vanthof@ii.uib.no(P. van ’t Hof), daniel.paulusma@durham.ac.uk(D. Paulusma), tesar@kam.mff.cuni.cz(M. Tesaˇr).

1 Supported by the ESF GraDR EUROGIGA grant as project GACR GIG/11/E023 and the NSERC grants of: K. Cameron and C. Hoàng (Wilfrid Laurier University), D. Corneil (University of Toronto), and P. Hell (Simon Fraser University).

2 Supported by MŠMT ˇCR grant LH12095 and GAˇCR grant P202/12/G061.

http://dx.doi.org/10.1016/j.tcs.2015.01.028 0304-3975/©2015 Elsevier B.V. All rights reserved.

(2)

as

H -Hom.

TheclassicalresultinthisareaistheHell–Nešetˇrildichotomytheoremwhichstatesthat H -Hom is solvablein polynomialtimeifH is bipartite,and NP-completeotherwise[27].

We consider so-called locally constrained homomorphisms. The neighborhood of a vertex u in a graph G is denoted

NG

(

u

)

= {

v

VG

|

uv

EG

}

. If for every u

VG the restriction of

ϕ

to the neighborhood of u, i.e., the mapping

ϕ

u

:

NG

(

u

)

NH

(

ϕ

(

u

))

, is injective, bijective, or surjective, then

ϕ

is said to be locally injective, locally bijective, or lo-cally surjective, respectively.Locallybijectivehomomorphismsarealsocalled

graph coverings.

Theyoriginatefromtopological graph theory [4,37] and have applications in distributed computing [2,3,7] and in constructing highly transitive regular graphs [5]. Locally injective homomorphisms are also calledpartial graph coverings. Theyhave applications in models of telecommunication [16] andindistance constrainedlabeling [17].Moreover, they are used asindicatorsofthe existence ofhomomorphisms ofderivative graphs [38]. Locallysurjective homomorphisms are alsocalled

color dominations

[35].In additionthey areknownas

role assignments due

totheir applicationsinsocialscience[13,39,40].Justlikelocallybijective homomorphismstheyalsohaveapplicationsindistributedcomputing[9].

Ifthereexistsahomomorphismfromagraph

G to

agraph

H that

islocallybijective,locallyinjective,orlocallysurjective, respectively,thenwe write G

−→

B H , G

−→

I H , andG

−→

S H , respectively. Wedenotethedecisionproblemsthat aretotest whether

G

−→

B H , G

−→

I H , or

G

−→

S H for twogivengraphs G and H by LBHom, LIHom and LSHom, respectively.Allthree problemsare knowntobe NP-completewhen bothguestandhostgraphsaregivenasinput (seebelowfordetails),and attemptshavebeenmadetoclassifytheir computationalcomplexitywhenonly theguestgraphbelongstotheinputand thehostgraphisfixed.Thecorresponding problemsaredenotedby H -LBHom, H -LIHom, and H -LSHom, respectively.The

H -LSHom problem ispolynomial-timesolvableeitherifH has noedgeorif H is bipartiteandhasatleastoneconnected component isomorphic to an edge; in all other cases H -LSHom is NP-complete, even when the guest graph belongsto the class of bipartite graphs [20]. The complexity classification of H -LBHom and H -LIHom is still open, although many partial results are known forboth problems; we refer to the papers [1,6,16,18,33,34,36] andto the survey by Fiala and Kratochvíl[15]forboth NP-completeandpolynomiallysolvablecases.

Instead of fixing the host graph, another natural restriction is to only take guest graphs from a special graph class. Heggernesetal.[30]provedthat LBHom is GraphIsomorphism-completewhentheguestgraphischordal,and polynomial-time solvablewhen theguestgraph isinterval. Incontrast, LSHom is NP-complete whenthe guestgraphis chordal and polynomial-timesolvablewhen theguest graphisproper interval, whereas LIHom is NP-completeeven forguestgraphs thatareproperinterval[30].Itisalsoknownthattheproblems LBHom and LSHom arepolynomial-timesolvablewhenthe guestgraphisatree[21].

Inthispaperwefocusonthefollowinglineofresearch.The

core of

agraph

G is

asubgraph F of G such that

G

F and

thereisnopropersubgraph Fof F with G

F.Itisknownthatthecoreofagraphisuniqueuptoisomorphism[28]. Dalmau,KolaitisandVardi[12]provedthatthe Hom problemispolynomial-timesolvablewhentheguestgraphbelongsto anyfixedclassofgraphswhosecoreshaveboundedtreewidth.Inparticular,thisresultimpliesan earlierresultthat Hom is polynomial-timesolvablewhen theguestgraph hasboundedtreewidth [11,22].Grohe [25] strengthened theresult of Dalmauetal.[12]by provingthat underacertain complexity assumption,namelyFPT

=

W

[

1

]

,the Hom problem canbe solvedinpolynomialtimeifandonlyifthisconditionholds.

ItisanaturalquestionwhethertheaboveresultsofDalmauetal.[12]andGrohe [25]remaintruewhenweconsider locallyconstrainedhomomorphismsinsteadofgeneralhomomorphisms.Wecanalreadyconcludefromknownresultsthat thisis not the casefor locally surjective homomorphisms. Recallthat H -LSHom is NP-complete even for bipartite guest graphsif H contains atleastoneedgeandiseithernon-bipartiteordoesnotcontain aconnectedcomponentisomorphic to an edge[20].The coreofevery bipartite graph withatleastone edge isan edge, andconsequently, hastreewidth 1. Thismeansthatbipartitegraphsformaclassofgraphswhosecoreshaveboundedtreewidth.Duetothisnegativeanswer, we turnourattentionto classesofgraphs ofboundedtreewidth insteadofclasses ofgraphswhose coreshavebounded treewidth,andposethefollowing(weaker)questioninstead:

Are LBHom, LIHom and LSHom polynomial-time solvable when the guest graph belongs to a class of bounded treewidth?

Thisquestionisfurthermotivatedbytwoknownresults,namelythat LBHom and LSHom canbothbesolvedinpolynomial timeiftheguestgraphisatree,andconsequentlyongraphsoftreewidth1[21].

1.1. Our contribution

InSection 3,we providea negativeanswer tothisquestion by showingthatthe problems LBHom, LSHom and LIHom are NP-complete alreadyinthe restrictedcasewherethe guestgraphhaspathwidthatmost 5,4 or2, respectively.We

also show that the three problems are NP-complete even if both the guest graph and the host graph have maximum

degree 3.Thelatterresultshowsthatlocallyconstrainedhomomorphismsproblemsbehavemorelikeunconstrained homo-morphismsongraphs ofboundeddegreethanongraphsofboundedtreewidth,asitisknownthat,forexample,

C5

-Hom is NP-completeonsubcubicgraphs[23].

Onthepositiveside, inSection 4,weshow thatall threeproblemscanbesolved inpolynomialtime ifwe boundthe treewidth of the guest graphand at the sametime bound the maximum degree of the guestgraph or the hostgraph. Becauseagraphclassofboundedmaximumdegreehasboundedtreewidthifandonlyifithasboundedclique-width[26], all three problemsare alsopolynomial-time solvablewhen we bound theclique-widthand themaximum degree ofthe

(3)

guestgraph.InSection4wealsoshowthat LIHom canbesolvedinpolynomialtimewhentheguestgraphhastreewidth 1, whichisbestpossiblegiventhehardnessresultfor LIHom showninSection3.

InSection5westatesomerelevantopenproblems. 2. Preliminaries

Let

G be

agraph.The

degree of

avertexv in G is denotedby

d

G

(

v

)

= |

NG

(

v

)

|

,and

(

G

)

=

maxvVGdG

(

v

)

denotesthe maximumdegreeof

G.

Let

ϕ

beahomomorphismfrom

G to

agraph

H .

Moreover,let

G

beaninducedsubgraphof

G,

and let

ϕ

 beahomomorphismfrom

G

to H . Wesaythat

ϕ

extends (or, equivalently,isan

extension of) ϕ

if

ϕ

(

v

)

=

ϕ



(

v

)

for every

v

VG.

A

tree decomposition of G is

atree

T

= (

VT

,

E

T

)

,wheretheelementsof

V

T,calledthe

nodes of T ,

aresubsetsof

V

Gsuch

thatthefollowingthreeconditionsaresatisfied:

1. foreachvertex

v

VG,thereisanode X

VT withv

X ,

2. foreachedge

uv

EG,thereisanode X

VT with

{

u

,

v

}

X ,

3. foreachvertex

v

VG,thesetofnodes

{

X

|

v

X

}

inducesaconnectedsubtreeof

T .

The

width of

atreedecomposition T is thesize ofalargestnode X minus one. The

treewidth of G,

denotedby tw

(

G

)

,is theminimumwidthoverallpossibletreedecompositionsof G.A path

decomposition of G is

atreedecomposition T of G

where

T is

apath.The

pathwidth of G is

theminimumwidthoverallpossiblepathdecompositionsof G.Bydefinition,the pathwidthof

G is

atleastashighasitstreewidth.Atreedecomposition

T is nice

[31]if

T is

abinarytree,rootedinaroot

R such thatthenodesof

T belong

tooneofthefollowingfourtypes: 1. a

leaf node X is

aleafof

T ,

2. an

introduce node X has

onechild

Y and

X

=

Y

∪ {

v

}

forsomevertex

v

VG

\

Y ,

3. a

forget node X has

onechild

Y and

X

=

Y

\ {

v

}

forsomevertex

v

Y ,

4. a

join node X has

twochildren

Y

,

Z satisfying

X

=

Y

=

Z .

An

equitable partition of

aconnectedgraph

G is

apartitionofitsvertexsetinblocks B1

,

. . . ,

Bksuchthatanyvertexin Bi hasthesame numbermi,j ofneighborsin Bj.Wecall thematrix M

= (

mi,j

)

corresponding to thecoarsest equitable

partitionof

G (in

whichtheblocksareorderedinsomecanonicalway;cf.[2])the

degree refinement matrix of G,

denotedas drm

(

G

)

.InSection3,wewillmakeuseofthefollowinglemma;aproofofthefirststatementinthislemmacanbefound inthepaperofFialaandKratochvíl[16],whereasthesecondstatementisduetoKristiansenandTelle[35].

Lemma1.

Let G and H be two graphs. Then the following two statements hold:

(i) if

G

I H and drm

(

G

)

=

drm

(

H

)

, then G

B H ;

(ii) if

G

S H and

drm

(

G

)

=

drm

(

H

)

, then G

B H .

3. NP-completenessresults

Forthe NP-hardnessresultsin

Theorem 1

belowweuseareductionfromthe3-Partition problem.Thisproblemtakesas inputamultiset A of 3m integers,denotedinthesequelby

{

a1

,

a2

,

. . . ,

a3m

}

,andapositiveinteger

b,

suchthat b4

<

ai

<

b2

for all i

∈ {

1

,

. . . ,

3m

}

and



1i3mai

=

mb. The taskisto determine whether A can be partitionedinto m disjoint sets A1

,

. . . ,

A

m suchthat



aAia

=

b for all

i

∈ {

1

,

. . . ,

m

}

.Notethat therestrictions onthesizeofeachelementin A implies that eachset Ai inthedesiredpartitionmustcontainexactlythreeelements,whichiswhysuch apartition A1

,

. . . ,

A

m is

called a3-partition of A. The3-Partition problem isstrongly NP-complete [24], i.e.,it remains NP-complete even ifthe problemisencodedinunary.

Theorem1.

The following three statements hold:

(i) LBHom

is

NP-complete on input pairs

(

G

,

H

)

where G has pathwidth at most 5 and H has pathwidth at most 3;

(ii) LSHom

is

NP-complete on input pairs

(

G

,

H

)

where G has pathwidth at most 4 and H has pathwidth at most 3;

(iii) LIHom

is

NP-complete on input pairs

(

G

,

H

)

where G has pathwidth at most 2 and

H has pathwidth at most 2.

Proof. Firstnotethatallthreeproblemsarein NP.Weproveeachstatementseparatelyandstartwithstatement(i). Givenaninstance

(

A

,

b

)

of3-Partition,weconstructtwographs

G and H as

follows;see

Figs. 1 and 2

forsomehelpful illustrations.Theconstruction ofG starts bytaking3m disjointcyclesC1

,

. . . ,

C3m oflength

b,

one foreachelement of A.

Foreach

i

∈ {

1

,

. . . ,

3m

}

,theverticesof

C

i arelabeled

u

i1

,

. . . ,

uib andwe add,foreach j

∈ {

1

,

. . . ,

b

}

,two newvertices pij

and

q

i

j aswellastwonewedges

u

i jp i j and

u

i jq i

(4)

Fig. 1. A

schematic illustration of the graphs

G andH that

are constructed from a given instance

(A,b)of 3-Partition in the proof of statement (i) in

Theorem 1. See also Fig. 2for a more detailed illustration of the “leftmost” part of G and

the “rightmost” part of

H ,

including more labels.

Fig. 2. More detailed illustration of parts of the graphs G and H inFig. 1.

vertices

p

i

1

,

p

i2

,

. . . ,

piai and

q

i

1

,

qi2

,

. . . ,

qiai forevery

i

∈ {

1

,

. . . ,

3m

}

.Finally,thevertex y is madeadjacenttoeveryvertexp i

j

that isnotadjacentto x, andthevertex z is madeadjacentto everyvertex

q

i

j that isnotadjacentto x. Thisfinishesthe

constructionof G.

To construct H , we take m disjoint cycles C1

˜

,

. . . ,

C

˜

m of length b, where the vertices of each cycle C

˜

i are labeled

˜

ui1

,

. . . ,

u

˜

ib. For each i

∈ {

1

,

. . . ,

m

}

and j

∈ {

1

,

. . . ,

b

}

, we add two vertices p

˜

ij and q

˜

ij and make both of them adjacent tou

˜

i

j.Finally,weaddavertexx and

˜

makeitadjacenttoeachoftheverticesp

˜

ijandq

˜

ij.Thisfinishestheconstructionof H .

Wenowshowthatthereexistsalocallybijectivehomomorphismfrom

G to

H if andonlyif

(

A

,

b

)

isayes-instanceof 3-Partition.

Letusfirstassume thatthereexists alocallybijective homomorphism

ϕ

fromG to H . Since

ϕ

isa degree-preserving mapping,wemusthave

ϕ

(

x

)

= ˜

x. Moreover,since

ϕ

islocallybijective,therestrictionof

ϕ

to

N

G

(

x

)

isabijectionfrom

N

G

(

x

)

to NH

(

x

˜

)

.Againusingthedefinitionofalocallybijectivemapping,thistimeconsideringtheneighborhoodsofthevertices

in

N

H

(

˜

x

)

,wededucethatthereisabijectionfromtheset

N

G2

(

x

)

:= {

uij

|

1

i

3m

,

1

j

ai

}

,i.e.,fromthesetofvertices

in

G at

distance2 from

x,

totheset N2

H

(

˜

x

)

:= {˜

ukj

|

1

k

m

,

1

j

b

}

ofverticesthatareatdistance2 fromx in H .

˜

For

every

k

∈ {

1

,

. . . ,

m

}

,wedefineaset Ak

A such that Akcontainselement

a

i

A if andonlyif

ϕ

(

ui1

)

∈ {˜

uk1

,

. . . ,

u

˜

kb

}

.Since

ϕ

isabijectionfrom

N

2

G

(

x

)

to

N

2H

(

x

˜

)

,thesets

A1

,

. . . ,

A

maredisjoint;moreovereachelement

a

i

A is containedinexactly

oneofthem.Observethatthesubgraph of

G induced

by N2

G

(

x

)

isadisjointunionof3m pathsoflengths

a1

,

a2

,

. . . ,

a3m,

respectively,whilethesubgraphof

H induced

by N2H

(

˜

x

)

isadisjointunionof

m cycles

oflength

b each.

Thefactthat

ϕ

is ahomomorphismandthereforenevermapsadjacentverticesof

G to

non-adjacentverticesin H implies that



aA

ia

=

b forall

i

∈ {

1

,

. . . ,

m

}

.Hence A1

,

. . . ,

A

m isa3-partitionof

A.

Forthereversedirection,supposethereexistsa3-partition A1

,

. . . ,

A

mofA. Wedefineamapping

ϕ

asfollows.Wefirst

set

ϕ

(

x

)

=

ϕ

(

y

)

=

ϕ

(

z

)

= ˜

x. Let Ai

= {

ar

,

as

,

at

}

beanysetofthe 3-partition.We mapthe verticesofthecycles Cr

,

Cs

,

Ct

that are at distance 2 from x to the vertices of the cycle C

˜

i in the following way:

ϕ

(

urj

)

= ˜

u i

j foreach j

∈ {

1

,

. . . ,

ar

}

,

ϕ

(

usj

)

= ˜

uia

r+j foreach j

∈ {

1

,

. . . ,

as

}

,and

ϕ

(

u t

j

)

= ˜

u i

ar+as+j foreach j

∈ {

1

,

. . . ,

at

}

. Theverticesof

C

r,

C

s and

C

t thatare atdistancemore than2 from x in G are mappedto verticesof C

˜

i suchthat the verticesof Cr, Cs andCt appearinthe

(5)

and Ct thatare atdistancemorethan 2 from x are mappedto verticesofC

˜

i analogously.Afterthe verticesofthecycles C1

,

. . . ,

C3mhavebeenmappedinthewaydescribedabove,itremainstomapthevertices

p

ijand

q

ijforeach

i

∈ {

1

,

. . . ,

3m

}

and j

∈ {

1

,

. . . ,

b

}

. Let pi

j

,

qij bea pairofvertices inG that are adjacentto x, andlet uij bethe second commonneighborof pij andqij.

Suppose u

˜

k istheimage of

u

i

j,i.e.,supposethat

ϕ

(

uij

)

= ˜

uk.Then wemap pij and

q

ij to p

˜

k andq

˜

k,respectively. Wenow consider the neighborsof y and z in G. Byconstruction, the neighborhood of y consists ofthe 2mb vertices inthe set

{

pij

|

ai+1

j

b

}

,whileNG

(

z

)

= {

qij

|

ai+1

j

b

}

.

Observethatx,

˜

theimageof y and z, isadjacenttotwosetsof

mb vertices:

oneoftheformp

˜

k,theotheroftheformq

˜

k. Hence, we needtomap halftheneighborsof y to verticesofthe formp

˜

k andhalf theneighborsof y to verticesofthe formq

˜

kinordertomake

ϕ

alocallybijectivehomomorphism.Thesameshouldbedonewiththeneighborsof

z.

Forevery vertexu

˜

kin H , wedoasfollows.Byconstruction,exactlythreeverticesofG are mappedtou

˜

k,andexactlytwoofthose vertices, say uij and uhg, are atdistance 2 from y in G. We set

ϕ

(

pij

)

= ˜

pk and

ϕ

(

phg

)

= ˜

qk.We also set

ϕ

(

qij

)

= ˜

qk and

ϕ

(

qgh

)

= ˜

pk.Thiscompletesthedefinitionofthemapping

ϕ

.

Since themapping

ϕ

preservesadjacencies, itclearly isahomomorphism.In ordertoshow that

ϕ

islocallybijective, wefirstobservethatthedegreeofeveryvertexin

G is

equaltothedegreeofitsimageinH ; inparticular,

d

G

(

x

)

=

dG

(

y

)

=

dG

(

z

)

=

dH

(

x

˜

)

=

2mb.Fromtheabovedescriptionof

ϕ

wegetabijectionbetweentheverticesof

N

H

(

˜

x

)

andtheverticesof NG

(

v

)

foreach

v

∈ {

x

,

y

,

z

}

.Foreveryvertex

p

ijthatisadjacentto

x and u

ijin

G,

itsimage p

˜

kisadjacenttotheimagesx of

˜

x and u

˜

kof

u

i

j.Foreveryvertex

p

i

j thatisadjacentto y (respectively z) and

u

i jin

G,

itsimage p

˜

k  orq

˜

k  isadjacenttox of

˜

y (respectively z) andu

˜

k

of

u

ij.Hencetherestrictionof

ϕ

to

N

G

(

pij

)

isbijectiveforevery

i

∈ {

1

,

. . . ,

3m

}

and j

∈ {

1

,

. . . ,

b

}

,

andthesameclearly holdsfortherestrictionof

ϕ

to NG

(

qij

)

.Theverticesofeachcycle

C

i are mappedtotheverticesof

somecycleC

˜

k insuchawaythattheverticesandtheirimagesappearinthesameorderonthecycles.This,togetherwith

thefactthattheimageu

˜

kofeveryvertex

u

i

jisadjacenttotheimages p

˜

k andq

˜

k  oftheneighbors

p

i j and

q

i j of

u

i j,shows

that therestrictionof

ϕ

to NG

(

uij

)

isbijectiveforevery

i

∈ {

1

,

. . . ,

3m

}

and j

∈ {

1

,

. . . ,

b

}

.Weconcludethat

ϕ

isalocally

bijectivehomomorphismfrom

G to H .

Inordertoshowthatthepathwidthof

G is

atmost 5,letusfirstconsiderthesubgraphof

G depicted

ontheleft-hand side of

Fig. 2

; wedenotethissubgraph by L1,andwesaythatthecycle

C1

defines the subgraph L1.The graphL1 thatis obtainedfrom

L1

bydeletingvertices

x

,

y

,

z and

edge

u

1

1u1bisacaterpillar,i.e.,atreeinwhichthereisapathcontainingall

verticesofdegreemorethan1.Sincecaterpillarsarewell-knowntohavepathwidth1,graphL1 hasapathdecomposition

P1 ofwidth 1.Startingwith P1,we cannowobtainapathdecompositionofthegraphL1 bysimplyaddingvertices

x,

y, z and u1

1 toeachnodeof P1;thispathdecompositionhaswidth 5.EverycycleCi in

G defines

asubgraph Li of

G in

the

sameway

C

1 definesthesubgraph

L

1.Supposewehaveconstructedapathdecomposition Pi ofwidth 5 ofthesubgraph Li foreach

i

∈ {

1

,

. . . ,

3m

}

inthewaydescribedabove.Sinceanytwosubgraphs

L

iand

L

jwith

i

=

j have onlythevertices x

,

y

,

z in common, andthesethreeverticesappearinallnodesofeachofthepathdecompositions Pi,wecanarrangethe

3m pathdecompositions P1

,

. . . ,

P

3m insuchawaythatweobtainapathdecomposition P of G of width 5.Hence

G has

pathwidthatmost 5.Similarbuteasierargumentscanbeusedtoshowthat H has pathwidthatmost 3.

The NP-hardness reductionforthelocallybijectivecasecanalsobe usedtoprovethat LIHom and LSHom are NP-hard forinputpairs

(

G

,

H

)

whereG has pathwidthatmost 5 and

H has

pathwidthatmost 3.Thisfollowsfromtheclaimthat

G

−→

B H if andonlyif

G

−→

S H if andonlyif

G

−→

I H for thegadgetgraphs

G and

H displayed in

Fig. 1

.Thisclaimcanbe seenasfollows.Firstsupposethat

G

−→

B H . Then,bydefinition,

G

−→

S H and G

−→

I H . Nowsupposethat

G

−→

I H or G

−→

S H .

Sinceitcaneasilybeverifiedthat

drm

(G

)

=

drm

(H

)

=



0 0 2mb 0 2 2 1 1 0



,

wecanuse

Lemma 1

(i)or(ii),respectively,todeducethat

G

−→

B H . However,wecanstrengthenthehardnessresultsforthe locallysurjectiveandinjectivecasesbyreducingthepathwidthoftheguestgraphtobeatmost4 and 2,respectively,and inthelattercasewecansimultaneouslyreducethepathwidthofthehostgraphtobeatmost 2,asclaimedinstatements (ii)and(iii)of

Theorem 1

.Inordertodoso,wegivethefollowingalternativeconstructionsbelow.

The alternative hardness construction for LSHom is similar to buteasierthan the construction for LBHom;see Fig. 3. Let

(

A

,

b

)

beaninstanceof3-Partition.Weconstructagraph

G

bytaking3m disjointcycles

C1

,

. . . ,

C3m oflength b,and labelingtheverticesofeachcycle

C

iwithlabels

u

i1

,

. . . ,

uiBinthesamewayaswelabeledtheverticesofthecycles

C

iinthe

construction for LBHom (seealso

Fig. 2

).Wethenaddtwovertices

x and

y. Forevery

i

∈ {

1

,

. . . ,

3m

}

,wemake

x adjacent

to eachofthe verticesui1

,

ui2

,

. . . ,

uaii,and y is madeadjacenttoeach ofthevertices

u

i

ai+1

,

. . . ,

u i

B.Graph H isobtained

fromthedisjointunionof

m cycles

C1

˜

,

. . . ,

C

˜

moflength

b by

addingoneuniversalvertexx.

˜

Usingsimilarargumentsasthe

onesusedinthe NP-hardnessproofof LBHom,itcanbeshownthatthereexistsalocallysurjectivehomomorphism

ϕ

from

Gto H ifandonlyif

(

A

,

b

)

isa yes-instanceof3-Partition.Suchahomomorphism

ϕ

 maps

x and

y to x,

˜

asotherwise a neighbor of

x or

y, whichhasdegree 3, mustbe mappedto x which

˜

hasdegreelarger than 3.Moreover,

ϕ

 mapsthe

(6)

Fig. 3. A

schematic illustration of the graphs

Gand Hthat are constructed from a given instance (A,b)of 3-Partition in the proof of statement (ii) in

Theorem 1.

Fig. 4. A

schematic illustration of the graphs

Gand Hthat are constructed from a given instance (A,b)of 3-Partition in the proof of statement (iii) in

Theorem 1.

verticesofcycles

C

1

,

. . . ,

C3mtotheverticesofcyclesC

˜

1

,

. . . ,

C

˜

minexactlythesamewayas

ϕ

mappedtheseverticesinthe

NP-hardnessproofof LBHom.Itisa routineexercisetoshowthat Ghaspathwidthatmost 4 andthat Hhaspathwidth atmost 3.

The reduction for LIHom iseven easier;see Fig. 4. Givenan instance

(

A

,

b

)

of3-Partition, we createa graph G by addingauniversalvertex

x to

thedisjointunionof3m pathson

a1

,

a2

,

. . . ,

a3mvertices,respectively.Graph H isobtained fromthedisjointunionof

m paths

on

b vertices

byaddingauniversalvertex

˜

x. Itiseasytoverifythatthereexistsalocally injectivehomomorphism

ϕ

 from

G

 to H,mapping

x to

x and

˜

allotherverticesof

G

 totheverticesofdegree 2 or3 in

H,ifandonlyif

(

A

,

b

)

isayes-instanceof3-Partition.Theobservationthatboth

G

 andH havepathwidth 2 completes theproofof

Theorem 1

.

2

Wenowconsiderthecasewhereweboundthemaximumdegreeof

G instead

ofthetreewidthofG. Wewillcombine some knownresultsinorderto showthat boundingthemaximumdegreeof G does not yieldtractability foranyofour

threeproblems LBHom, LIHom and LSHom.

Kratochvíl andKˇrivánek [32] showed that K4-LBHom is NP-complete,where K4 denotes the complete graph onfour vertices.Sinceagraph

G allows

alocallybijectivehomomorphismto

K

4onlyif

G is

3-regular,

K

4-LBHom is NP-completeon

3-regulargraphs.Thedegreerefinementmatrixofa3-regulargraphisthe1

×

1 matrixwhoseonlyentryis 3.Consequently, dueto Lemma 1, K4-LBHom is equivalent to K4-LIHom andto K4-LSHom on 3-regular graphs. Thisyields thefollowing result.

Theorem2.

The problems LBHom, LIHom and LSHom are

NP-complete on input pairs

(

G

,

K

4

)

where G has maximum degree 3. Theorem 2is tightinthefollowing sense.Allthree problems LBHom, LIHom and LSHom arepolynomial-timesolvable oninputpairs

(

G

,

H

)

where

G has

maximumdegreeatmost 2.Moreover,thefirsttwoproblemsarealsopolynomial-time solvableoninputgraphs

(

G

,

H

)

whereonly

H has

maximumdegreeatmost 2,as

G is

requiredtobeofmaximumdegree atmost 2 aswellinthesetwocases.Thisdoesnotholdfor LSHom,as

K3

-LSHom is NP-complete[35].

4. Polynomial-timeresults

InSection3,weshowedthat LBHom, LIHom and LSHom are NP-completewheneitherthetreewidthorthemaximum degree oftheguestgraphisbounded.Inthissection,we showthat all threeproblemsbecomepolynomial-timesolvable ifweboundboththetreewidthandthemaximumdegreeof G.Fortheproblems LBHom and LIHom,ourpolynomial-time resultfollows fromreformulating these problemsasconstraint satisfactionproblems andapplyinga result ofDalmauet al.[12].Inordertoexplainthis,weneedsomeadditionalterminology.

(7)

Arelationalstructure

(

A

,

R1

,

. . . ,

R

k

)

isafiniteset A, calledthe

base

set, togetherwithacollectionofrelations

R1

,

. . . ,

R

k.

Thearitiesoftheserelationsdeterminethevocabularyofthestructure.Ahomomorphismbetweentworelationalstructures ofthesamevocabularyisamappingbetweenthebasesetssuchthatalltherelationsarepreserved.

Fiala andKratochvíl [14] observedthat locallyinjective andlocally bijective homomorphisms betweengraphs can be expressedashomomorphismsbetweenrelationalstructuresasfollows.Alocallyinjectivehomomorphism f

:

G

H can be expressedasahomomorphismbetweenrelationalstructures

(

VG

,

E

G

,

EG

)

and

(

VH

,

EH

,

EH

)

,wherethenewbinaryrelation Econsistsofpairsofdistinctverticesthathaveatleastonecommonneighbor.Since f maps distinctneighborsofavertex

v to distinct neighbors of f

(

v

)

,we getthat f is a homomorphismof the associatedrelational structures. Onthe other hand, if

(

VG

,

E

G

,

E

G

)

and

(

VH

,

E

H

,

E

H

)

are constructed fromG and H as described above,andif f is a homomorphism

betweenthem,thentherelations EG andEH guaranteethatnotwoverticeswithacommonneighborin

G are

mappedto the sametarget in H .Inother words, f is alocallyinjectivehomomorphismbetweenthegraphs G and H . Ananalogous constructionworksforlocallybijectivehomomorphisms.Here,weneedtoexpress

G using

twobinaryrelations

E

G andEG

asabove,togetherwith

(

G

)

+

1 unaryrelations.Aunaryrelationcanbeviewedasaset:here,the

i-th

setwillconsistsof allverticesofdegree i

1.Theseunaryrelationsguaranteethatdegreesarepreserved,andconsequentlythattheassociated graphhomomorphismsarelocallybijective.

The

Gaifman graph

G

Aofarelationalstructure

A

= (

A

,

R1

,

. . . ,

Rk

)

isthegraphwithvertexset

A,

whereanytwodistinct

vertices u and v are joinedby an edgeif theyare bound by some relation.Formally u

,

v

EGA if andonlyifforsome relation Riofarity

r and

(

a1

,

. . . ,

ar

)

Riitholdsthat

{

u

,

v

}

⊆ {

a1

, . . . ,

ar

}

.

As a directconsequence of a resultof Dalmauet al. [12], the existence ofa homomorphismbetween two relational structures

A

and

B

can be decided in polynomial time if the treewidth of

G

A is bounded by a constant. Thisleads to

Theorem 3below.

Theorem3.

The problems LBHom and LIHom can be solved in polynomial time when G has bounded treewidth and G or

H has bounded maximum degree.

Proof. First supposethat G has boundedtreewidth andboundedmaximumdegree. Observethatforlocallyinjectiveand locallybijectivehomomorphisms,theGaifmangraph

G

Aisisomorphicto

G

2,whichisthegrapharisingfrom

G by

adding

an edge betweenany two vertices at distance 2. It suffices to observe that tw

(

G2

)

≤ (

G

)(

tw

(

G

)

+

1

)

1, as we can

transformanytreedecomposition

T of G of

widthtw

(

G

)

intoadesiredtreedecompositionofG2 byaddingtoeachnode X of T all theneighborsofevery vertexfrom X . Since G

−→

I H implies that

(

H

)

≥ (

G

)

,the theoremalsoholds ifwe

boundthemaximumdegreeofH instead of G.

2

To our knowledge,locally surjective homomorphisms have not yet beenexpressed ashomomorphisms between rela-tionalstructures.Hence,intheproofof

Theorem 4

below,wepresentapolynomial-timealgorithmfor LSHom when G has

boundedtreewidthandboundedmaximumdegree.Wefirstintroducesomeadditionalterminology.

Let

ϕ

bealocallysurjectivehomomorphismfrom

G to

H . Letv

VG andp

VH.If

ϕ

(

v

)

=

p, i.e.,if

ϕ

mapsvertex v

tocolor p, thenwesaythat

p is assigned to

v. Bydefinition,foreveryvertex

v

VG,thesetofcolorsthatareassignedto

theneighborsof

v in G is

exactlytheneighborhoodof

ϕ

(

v

)

in H . Nowsupposewearegivenahomomorphism

ϕ

froman inducedsubgraph

G

of

G to H .

Foranyvertex

v

VG,wesaythat

v misses a

colorp

VH if p

NH

(

ϕ



(

v

))

\

ϕ



(

NG

(

v

))

,

i.e.,if

ϕ

doesnotassign

p to

anyneighborof

v in G

,butanylocallysurjectivehomomorphism

ϕ

from

G to H that

extends

ϕ

 assigns

p to

someneighborofv in G.

Let T be a nicetreedecomposition ofG rooted in R. Foreverynode X

VT,we define GX to bethe subgraphof G

induced bytheverticesof X together withthe verticesofall thenodesthatare descendantsof X . Inparticular, wehave

GR

=

G.

Definition1.Let X

VT,andlet

c

:

X

VH and

μ

:

X

2VH betwo mappings.The pair

(

c

,

μ

)

is

feasible for G

X ifthere

existsahomomorphism

ϕ

from

G

X to H satisfying thefollowingthreeconditions:

(i) c

(

v

)

=

ϕ

(

v

)

forevery v

X ;

(ii)

μ

(

v

)

=

NH

(

ϕ

(

v

))

\

ϕ

(

NGX

(

v

))

forevery v

X ; (iii)

ϕ

(

NG

(

v

))

=

NH

(

ϕ

(

v

))

forevery

v

VGX

\

X .

Inotherwords,apair

(

c

,

μ

)

consistsofacoloring

c of

theverticesof X , togetherwithacollectionofsets

μ

(

v

)

,onefor each

v

X , consistingofexactlythosecolorsthat v misses. Informallyspeaking,apair

(

c

,

μ

)

isfeasibleforGX ifthereis

a homomorphism

ϕ

:

GX

H such that

ϕ

“agrees”withthecoloring c on theset X , andsuch that noneofthevertices

in VGX

\

X misses anycolor. The idea is that ifa pair

(

c

,

μ

)

is feasible, then such a homomorphism

ϕ

might havean extension

ϕ

∗ that isa locallysurjective homomorphism fromG to H . Afterall, foranyvertex v

X that missesa color whenconsidering

ϕ

,thiscolormightbeassignedby

ϕ

∗toaneighborofv in theset

V

G

\

VGX.

(8)

Theorem4.

The

problem LSHom can be solved in polynomial time when G has bounded treewidth and G or H has bounded maximum degree.

Proof. Let

(

G

,

H

)

beaninstanceof LSHom suchthatthetreewidthoftheguestgraph

G is

bounded.Throughouttheproof,

we assume that the maximum degree of H is bounded, and show that the problemcan be solved in polynomial time

undertheserestrictions.Since

G

−→

S H implies

that

(

G

)

≥ (

H

)

,ourpolynomial-timeresultappliesalsoifweboundthe maximumdegreeof

G instead

of H .

We mayassume withoutloss ofgenerality that both G and H are connected,asotherwise we justconsiderall pairs

(

Gi

,

H

j

)

separately, where Gi is a connected component of G and Hj is a connected component of H . Because G has

bounded treewidth, we can compute a tree decomposition of G of width tw

(

G

)

inlinear time using Bodlaender’s algo-rithm [8]. We transformthistree decompositioninto anice treedecomposition T of G with width tw

(

G

)

withat most 4

|

VG

|

nodesusingthelinear-timealgorithmofKloks[31].Let

R be

therootof

T and

let

k

=

tw

(

G

)

+

1.

Foreachnode X

VT,let FX bethesetofallfeasiblepairs

(

c

,

μ

)

for

G

X.Forevery feasiblepair

(

c

,

μ

)

FX andevery v

X , itholdsthat

μ

(

v

)

isasubsetof

N

H

(

c

(

v

))

.Since

|

X

|

k and

|

NH

(

c

(

v

))

|

≤ (

H

)

k for every

v

X and everymapping c

:

X

VH,thisimpliesthat

|

FX

|

≤ |

VH

|

k2(H)k foreach X

VT.Asweassumedthatboth

k and

(

H

)

areboundedbya

constant,theset FX isofpolynomialsizewithrespectto

|

VH

|

.

The algorithmconsiders the nodesof T in a bottom-up manner,starting withthe leavesof T and processinga node

X

VT only after its children have been processed. For every node X , the algorithm computes the set FX in the way

describedbelow.Wedistinguishbetweenfourdifferentcases.Thecorrectnessofeachofthecaseseasilyfollowsfromthe definitionofalocallysurjectivehomomorphismand

Definition 1

.

1. X isa leaf node of T . We considerallmappings

c

:

X

VH.Foreachmapping

c,

wecheckwhether

c is

ahomomorphism

from

G

X to H . Ifnot,thenwediscard

c,

asitcannot belongtoa feasiblepairduetocondition(i)in

Definition 1

.For

eachmapping

c that

isnotdiscarded,wecompute theuniquemapping

μ

satisfying

μ

(

v

)

=

NH

(

c

(

v

))

\

c

(

NGX

(

v

))

for each

v

X , andweaddthepair

(

c

,

μ

)

to FX.Itfollowsfromcondition(ii)thattheobtainedset FX indeedcontainsall

feasiblepairsfor

G

X.Asthereisnovertexin

V

GX

\

X , everypair

(

c

,

μ

)

triviallysatisfiescondition(iii).Thecomputation of

F

X canbedonein O

(

|

VH

|

kk

((

H

)

+

k

))

timeinthiscase.

2. X is

a forget node.

Let

Y be

thechildofX in T , andlet

{

u

}

=

Y

\

X . Observethat

(

c

,

μ

)

FX ifandonlyifthereexists

afeasiblepair

(

c

,

μ



)

FY suchthat

c

(

v

)

=

c

(

v

)

and

μ

(

v

)

=

μ



(

v

)

forevery

v

X , and

μ



(

u

)

= ∅

.Henceweexamine

each

(

c

,

μ



)

FY andcheckwhether

μ



(

u

)

= ∅

issatisfied.Ifso,we firstrestrict

(

c

,

μ



)

on X to get

(

c

,

μ

)

andthen

weinserttheobtainedfeasiblepairinto FX.Thisprocedureneeds

O

(

|

FY

|

k

(

H

))

timeintotal.

3. X is

an introduce node.

Let Y be thechildof X in T , andlet

{

u

}

=

X

\

Y . Observethat

(

c

,

μ

)

FX ifandonlyifthere

existsafeasiblepair

(

c

,

μ



)

FY such that,forevery

v

Y , itholdsthat

c

(

v

)

=

c

(

v

)

,

μ

(

v

)

=

μ



(

v

)

\

c

(

u

)

if

uv

EG,

and

μ

(

v

)

=

μ



(

v

)

ifuv

/

EG.Hence, foreach

(

c

,

μ



)

FY, we consider all

|

VH

|

mappings c

:

X

VH that extend c. Foreach such extension

c,

we test whetherc is ahomomorphism from GX to H by checkingthe adjacencies of c

(

u

)

in H .Ifnot,thenwemaysafelydiscard

c due

tocondition(i)in

Definition 1

.Otherwise,wecomputetheunique mapping

μ

:

X

2VH satisfying

μ

(v)

=



N H

(c(u))

\

c(NGX

(u))

if v

=

u

μ



(v)

\

c(u) if v

=

u and uv

EG

μ



(v)

if v

=

u and uv

/

EG

,

andwe addthepair

(

c

,

μ

)

to FX;duetocondition(ii),thispair

(

c

,

μ

)

istheuniquefeasiblepaircontaining

c.

Com-putingtheset FX takesatmost O

(

|

FY

||

VH

|

k

(

H

))

timeintotal.

4. X is

a join node.

Let

Y and

Z be thetwochildrenof X in T . Observethat

(

c

,

μ

)

FX ifandonlyifthereexistfeasible

pairs

(

c1

,

μ1

)

FY and

(

c2

,

μ2

)

FZ such that, for every v

X , c

(

v

)

=

c1

(

v

)

=

c2

(

v

)

and

μ

(

v

)

=

μ

1

(

v

)

μ

2

(

v

)

.

Hence the algorithm considers every combinationof

(

c1

,

μ1

)

FY with

(

c2

,

μ2

)

FZ andifthey agree on thefirst

component c,theothercomponent

μ

isdetermineduniquelybytakingtheintersectionof

μ1

(

v

)

and

μ2

(

v

)

forevery

v

X . Thisprocedurecomputestheset FX in O

(

|

FY

||

FZ

|

k

(

H

))

timeintotal.

Finally,observe that a locallysurjective homomorphism from G to H exists ifandonly ifthere exists a feasiblepair

(

c

,

μ

)

for GR such that

μ

(

v

)

= ∅

for all v

R. Since T has at most 4

|

VG

|

nodes, we obtain a total running time of O

(

|

VG

|(|

VH

|

k2(H)k

)

2k

(

H

))

.Asweassumedthatboth

k

=

tw

(

G

)

+

1 and

(

H

)

areboundedbyaconstant,ouralgorithm

runsinpolynomialtime.

2

Theproof of

Theorem 4

suggestan alternativeproof ofTheorem 3thatdoesnotrely onthegeneralresultby Dalmau etal.[12].Toseewhythisisthecase,firstobservethatwecansolve LIHom usingadynamicprogrammingapproachthat stronglyresemblestheonefor LSHom describedintheproofof

Theorem 4

;insteadofkeepingtrackofsets

μ

(

v

)

ofcolors thatavertex v

X is missing,wekeeptrackofsets

α

(

v

)

ofcolorsthathavealreadybeenassignedtotheneighborsofa vertex v

X . Thisisbecausein alocallyinjectivehomomorphism fromG to H , no colormaybeassignedto morethan oneneighborofanyvertex.Inthiswaywecanadjust

Definition 1

insuchawaythatitworksforlocallyinjectiveinstead oflocallysurjectivehomomorphisms.Tosolve LBHom inpolynomialtimewithoutusingtheresultbyDalmauetal.[12],it

(9)

sufficestoobservethat

(

G

,

H

)

isayes-instanceof LBHom ifandonlyifitisayes-instanceforboth LIHom and LSHom and that we cansolve thelattertwo problemsinpolynomial time usingdynamic programmingasexplained above.Weomit furtherdetails,butweexpectthatdynamicprogrammingalgorithmsofthiskindwillhavesmallerhiddenconstantsinthe runningtimeestimatesthanthemoregeneralmethodofDalmauetal.[12].

We concludethissectionwithonemorepolynomial-timeresult. Itisknownthattheproblems LBHom and LSHom are polynomial-timesolvablewhen G is atree[21],andconsequentlywhen

G has

treewidth 1.Weshowthatthesameholds forthe LIHom problem.

Theorem5.

The LIHom problem can be solved in polynomial time when G has treewidth 1, i.e., when G is a forest.

Proof. Let usfirststatesometerminologyandusefulknownresults.TheuniversalcoverTG ofaconnectedgraphG is the

unique tree(whichmayhavean infinitenumberofvertices)such thatthereisalocallybijectivehomomorphismfromTG

to

G.

Onewaytodefinethismappingisasfollows.Consider allfinitewalksin G that startfroman arbitraryfixedvertex in G and thatdonottraverse thesameedgeintwoconsecutivesteps.Eachsuch walkwillcorrespondtoa vertexof TG.

We lettwoverticesofTG beadjacentifandonlyifonecanbeobtainedfromtheother bydeletingthelast vertexofthe

walk. Thenthemapping fG thatmapseverywalktoitslast vertexisalocallybijectivehomomorphismfrom

T

G to G[2].

It isalso knownthat TG

=

G if andonlyif G is a tree[2].Moreover,foranytwo graphs G and H , G

−→

I H implies that TG

−→

I TH [19].

Now let

(

G

,

H

)

be aninstanceof LIHom where

G has

treewidth 1. Weassume,withoutlossofgenerality, thatboth G

and H are connected.In particular, G is atree.We claim that G

−→

I H if andonlyif TG

−→

I TH.The forwardimplication

followsfromabove.Toshowthebackward implication,suppose that TG

−→

I TH.Then G

−→

I TH,because

T

G

=

G. Let f be

a locallyinjectivehomomorphismfrom

G to

TH.Then,because

G

−→

I TH and TH

−→

B H , wehave G

−→

I H . Toexplainthis,

consider themapping f

:

VG

VH definedby f

(

u

)

=

fH

(

x

)

ifandonlyif f

(

u

)

=

x. Notice that f isa locallyinjective

homomorphism from G to H . The desiredresult follows from thisclaim combined with the fact that we can check in polynomialtimewhether

T

G

−→

I TH holdsfortwographs

G and H

[21].

2

5. Conclusion

Theorem 5statesthat LIHom canbesolvedinpolynomialtimewhentheguestgraphhastreewidth 1,while

Theorem 1

impliesthat theproblemis NP-completewhentheguestgraphhastreewidth 2.Thisshowsthatthebound onthe path-width in thethird statement ofTheorem 1 is best possible.We leave it as an open problemto determine whether the boundsonthepathwidthintheothertwostatementsof

Theorem 1

canbereducedfurther.

Weconcludethispaperwithsomeremarksontheparameterizedcomplexityoftheproblems LIHom, LSHom andLBHom. The hardness resultsinthispapershowthat allthree problemsarepara-NP-completewhen parameterizedbyeither the

treewidth of G or the maximum degree of G. Theorems 3 and 4 show that the problems are in XP when

parameter-ized jointly by thetreewidth of G and the maximumdegree of G. A naturalquestion is whetherthe problemsare FPT when parameterized by the treewidth of G and the maximum degree of G, i.e., whether they can be solved in time

f

(

tw

(

G

),

(

G

))

· (|

VG

|

+ |

VH

|)

O(1)forsomefunction f thatdoesnotdependonthesizesof

G and H .

Acknowledgements

We wouldliketothankIsoldeAdlerforposingtheresearch questionsthatwe addressedinourpaperandforhelpful discussions.ThefourthauthoralsothanksJanArneTelleforfruitfuldiscussions.

References

[1]J.Abello,M.R.Fellows,J.C.Stillwell,Onthecomplexityandcombinatoricsofcoveringfinitecomplexes,Australas.J.Combin.4(1991)103–112. [2]D.Angluin,Localandglobalpropertiesinnetworksofprocessors,in:Proc.STOC1980,1980,pp. 82–93.

[3]D.Angluin,A.Gardiner,Finitecommoncoveringsofpairsofregulargraphs,J.Combin.TheorySer.B30(1981)184–187. [4]N.Biggs,AlgebraicGraphTheory,CambridgeUniversityPress,1974.

[5]N.Biggs,Constructing5-arctransitivecubicgraphs,J.Lond.Math.Soc.(2)26(1982)193–200.

[6]O.Bílka,B.Lidický,M.Tesaˇr,Locallyinjectivehomomorphismtothesimpleweightgraphs,in:Proc.TAMC2011,in:LNCS,vol. 6648,2011,pp. 471–482. [7]H.L.Bodlaender,Theclassificationofcoveringsofprocessornetworks,J.ParallelDistrib.Comput.6(1989)166–182.

[8]H.L.Bodlaender,Alinear-timealgorithmforfindingtree-decompositionsofsmalltreewidth,SIAMJ.Comput.25 (6)(1996)1305–1317.

[9]J.Chalopin,Y.Métivier,W.Zielonka,Election,namingandcellularedgelocalcomputations,in:Proc.ICGT2004,in:LNCS,vol. 3256,2004,pp. 242–256. [10]S.Chaplick,J.Fiala,P.van’tHof,D.Paulusma,M.Tesar,Locallyconstrainedhomomorphismsongraphsofboundedtreewidthandboundeddegree,in:

Proc.FCT2013,in:LNCS,vol. 8070,2013,pp. 121–132.

[11]Ch.Chekuri,A.Rajaraman,Conjunctivequerycontainmentrevisited,in:Proc.5thInternationalConferenceonDatabaseTheory,in:LNCS,vol. 1186, 1997,pp. 56–70.

[12]V.Dalmau,P.G.Kolaitis,M.Y.Vardi,Constraintsatisfaction,boundedtreewidth,andfinite-variablelogics,in:Proc.CP2002,in:LNCS,vol. 2470,2002, pp. 223–254.

[13]M.G.Everett,S.Borgatti,Rolecoloringagraph,Math.SocialSci.21(1991)183–188.

[14]J.Fiala,J.Kratochvíl,Locallyinjectivegraphhomomorphism:listsguaranteedichotomy,in:Proc.WG2006,in:LNCS,vol. 4271,2006,pp. 15–26. [15]J.Fiala,J.Kratochvíl,Locallyconstrainedgraphhomomorphisms–structure,complexity,andapplications,Comput.Sci.Rev.2(2008)97–111.

(10)

[16]J.Fiala,J.Kratochvíl,Partialcoversofgraphs,Discuss.Math.GraphTheory22(2002)89–99.

[17]J.Fiala,J.Kratochvíl,T.Kloks,Fixed-parametercomplexityofλ-labelings,DiscreteAppl.Math.113(2001)59–72.

[18]J.Fiala,J.Kratochvíl,A.Pór,Onthecomputationalcomplexityofpartialcoversofthetagraphs,DiscreteAppl.Math.156(2008)1143–1149. [19]J.Fiala,J.Maxová,Cantor–Bernsteintypetheoremforlocallyconstrainedgraphhomomorphisms,EuropeanJ.Combin.27(2006)1111–1116. [20]J.Fiala,D.Paulusma,Acompletecomplexityclassificationoftheroleassignmentproblem,Theoret.Comput.Sci.349(2005)67–81. [21]J.Fiala,D.Paulusma,Comparinguniversalcoversinpolynomialtime,TheoryComput.Syst.46(2010)620–635.

[22]E.C.Freuder,Complexityofk-treestructuredconstraintsatisfactionproblems,in:Proc.8thNationalConferenceonArtificialIntelligence,1990,pp. 4–9. [23]A.Galluccio,P.Hell,J.Nešetˇril,ThecomplexityofH -colouringofboundeddegreegraphs,DiscreteMath.222(2000)101–109.

[24]M.R.Garey,D.S.Johnson,ComputersandIntractability:AGuidetotheTheoryof NP-completeness,W.H.Freeman&Co.,NewYork,1979. [25]M.Grohe,Thecomplexityofhomomorphismandconstraintsatisfactionproblemsseenfromtheotherside,J.ACM54 (1)(2007).

[26]F.Gurski,E.Wanke,Thetree-widthofclique-widthboundedgraphswithoutKn,n,in:Proc.WG2000,in:LNCS,vol. 1928,2000,pp. 196–205. [27]P.Hell,J.Nešetˇril,OnthecomplexityofH -colouring,J.Combin.TheorySer.B48(1990)92–110.

[28]P.Hell,J.Nešetˇril,Thecoreofagraph,DiscreteMath.109(1992)117–126. [29]P.Hell,J.Nešetˇril,GraphsandHomomorphisms,OxfordUniversityPress,2004.

[30]P.Heggernes,P.van’tHof,D.Paulusma,Computingroleassignmentsofproperintervalgraphsinpolynomialtime,J.DiscreteAlgorithms14(2012) 173–188.

[31]T.Kloks,Treewidth,ComputationsandApproximations,LNCS,vol. 842,Springer,1994.

[32]J.Kratochvíl,M.Kˇrivánek,Onthecomputationalcomplexityofcodesingraphs,in:Proc.MFCS1988,in:LNCS,vol. 324,1988,pp. 396–404. [33]J.Kratochvíl,A.Proskurowski,J.A.Telle,Coveringregulargraphs,J.Combin.TheorySer.B71(1997)1–16.

[34]J.Kratochvíl,A.Proskurowski,J.A.Telle,Complexityofgraphcoveringproblems,NordicJ.Comput.5(1998)173–195. [35]P.Kristiansen,J.A.Telle,GeneralizedH -coloringofgraphs,in:Proc.ISAAC2000,in:LNCS,vol. 1969,2000,pp. 456–466.

[36]B.Lidický,M.Tesaˇr,Complexityoflocallyinjectivehomomorphismtothethetagraphs,in:Prov.IWOCA2010,in:LNCS,vol. 6460,2010,pp. 326–336. [37]W.S.Massey,AlgebraicTopology:AnIntroduction,Harcourt,BraceandWorld,1967.

[38]J.Nešetˇril,Homomorphismsofderivativegraphs,DiscreteMath.1(1971)257–268.

[39]A.Pekeˇc,F.S.Roberts,Theroleassignmentmodelnearlyfitsmostsocialnetworks,Math.SocialSci.41(2001)275–293. [40]F.S.Roberts,L.Sheng,Howhardisittodetermineifagraphhasa2-roleassignment?,Networks37(2001)67–73.

Referenties

GERELATEERDE DOCUMENTEN

Les tombes les plus anciennes sont présentement reconnues dans Ie groupe septentrional.. Cette zone s'avère particulièrement endommagée par les pillages

cent syllables, would make it les s easy to allow one and only one syllable identification per presented fragment, whereas any such addition would also distort

Hier en daar vonden we tussen de verbrande beenderen kleine stukjes brons, niets uitzonderlijks aan- gezien dit in zulke beendergraven een gewoon verschijnsel is, dat zijn

Cet éperon protégé naturellement sur trois flancs a été barré par deux levées flanquées chacune d'un fossé.. Localisé au lieu-ditLes Aisances de Bellefontaine

, indien mogelijk met de strafmaatregelen erbij vermeld • Door een groep ons bekende zelfbouwers is zelfs bij de notaris een contract ondertekend • De andere

In het najaar van 1997 maakten we de grond bouwrijp voor de aa nleg van meid oorn - en beu­ kenhagen.. een gerief­ houtbo

Ik denk dat deze soort niet kan concurreren met bet al­ gemene parapluutjesmos (Marcbantia polymorpha) uit dezeIfde groep met ronde broedbekertjes op de

In de academische wereld wordt veel onderzoek gedaan naar de samenhang tussen innovatie, internationale handel en economische groei. Uit eerder onderzoek van CBS werd al duidelijk