Guarding Lines and
2
-Link Polygons is APX-Hard Björn Brodén∗
Mikael Hammar∗
Bengt J. Nilsson†
Abstra tWeprovethattheminimumline overingproblemand
the minimum guard overing problem restri ted to
2
-link polygonsareAPX-hard.keywords: Computational Geometry, Polygon
De- omposition, Art Gallery Theorem, Minimum Guard
Covering,MinimumLineCovering.
1 Introdu tion
Pi ture yourself as the president of the national
mu-seum in your ountry holding invaluable treasures of
historyandart. Howmanysurveillan e ameraswould
you need installed at the museum to make you sleep
tightatnight? Beingas ientistyoumight ometothe
on lusionthatifeverypartofthegalleryisseenbythe
ameras,thiswouldfulll yourneeds. Theansweryou
seekisthesolutiontotheArtGalleryProblem[4,9,10℄:
howmany amerasdoweneedtoguardagivengallery
and how do we de ide where to pla e them? This
problemwasposedbyViktorKleein 1973inresponse
to a questionraisedby Va²ek Chvátal [6℄. The latter
showed in 1975 that
⌊n/3⌋
amerasare su ientand sometimesne essary to guardanygallery representedasatwodimensionalsimplepolygon[3℄.
The algorithmi version of this problem is to nd
a minimum number of guards for a given art gallery.
Thisis alledtheminimumguard overingproblemand
wasshowntobeNP-hardbyAggarwal[1℄andLeeand
Lin[8℄. Thisisa tuallyapolygonde omposition
prob-lem. The visibility polygon from a point, that is the
areaagivenguard ansee,isstarshaped. Thismakes
theminimumguard overageproblemidenti altothat
of nding a minimum star over of a polygon. The
problem is known to be omputationally di ult [5℄
and it is strongly related to the minimum set over
problem.
In1995Joseph S.B. Mit hell raisedthe questionof
orresponden e between link diameter and
guardabil-ityofsimplepolygons.
1
-linkpolygonsare onvexand hen etriviallyguardableinpolynomialtime. Theorig-∗
DepartmentofComputerS ien e,LundUniversity,Box118,
S-22100Lund,Sweden. e-mail:{bjorn, mikael} s.lth.se
†
S hool of Te hnology and So iety, MalmöUniversity
Col-lege, Citadellsvägen 7, 205 06 Malmö, Sweden. e-mail:
inal NP-hardness proofs [1, 8℄ immediately give
NP-hardnessforpolygonswith linkdiameter
≥ 4
. Nilsson provedtheNP-hardnessfor≥ 3
-linkpolygons[9℄. This leavesthe omplexityquestionfortheminimumguardoveringproblemfor
2
-linkpolygonsopen. Notethata large lassof2
-linkpolygonsarestarshapedand there-foreguardableinpolynomialtime. Inthispaperwesetouttoanswerthe omplexity questionfor
2
-link poly-gons. Indoing soweanalyzearelatedproblem alledthe minimumline overingproblem whi h is
interest-ingin itsownright. Wegiveasu in tredu tionfrom
MAX
2
SAT proving that this problem is APX-hard. A similarresult was laimedby Kumaret. al.[7℄ butour onstru tion givesan expli it lowerbound on the
approximationratio.
From theminimumline overingproblemthere is a
straightforwardgap preserving redu tion to the
mini-mumguard overingproblem,afa tthatMit helland
Kone£nýindependentlyobserved.
2 Problem Formulation
Before we prove our laim we give formal
deni-tions of the Minimum Guard Covering Problem, the
MinimumLineCoveringProblem,and
k
-linkpolygons.Minimum GuardCovering Problem (MGCP)
Instan e: A polygon
P
.Solution: A minimum set of points in
P
from whi h theentirepolygon,interiorandboundary anbeseen.Twopoints
x, y ∈ P
seeea h other ifthestraightline segmentxy ⊂ P
.TheMinimumLineCoveringProblem(MLCP)
Instan e: A set
L
ofnon-parallellinesin theplane. Solution: A minimumsetP
of pointssu h that there isatleastonepointinP
onea hlineinL
.DenitionApolygon
P
haslinkdiameterk
ifk
isthe minimum integer value su h that for all pointsx, y ∈
P
there exists a path fromx
toy
that does not ross the boundary ofP
and that onsists ofk
straight line segments. Wesaythat apolygon isk
linkifithaslink diameterk
.The Minimum Guard Covering Problem is trivial for
Covering Problem onstru tsa polygon instan e
hav-ing link diameter
4
[1, 8℄ and this was strengthened to3
-linkpolygonsby Nilsson [9℄. In the nextse tion weprovethattheMinimumLine CoveringProblemisAPX-hard. This,inturn,leadsustotheAPX-hardness
resultfortheMinimumGuardCoveringProblemin
2
-link polygoninstan es, strengthening thepreviousre-sults.
3 The Redu tion
The redu tion is made from a spe ial ase of MAX
SAT where ea h lause ontainstwoliterals and ea h
literalo ursatmosttwi e.
MAX
2
SAT(2
L)Instan e: A set
U
of variables and a olle tionC
of disjun tive lauses withexa tly2
literals. Ea h literal anappear at mosttwi e inC
. A literalis avariable oranegatedvariableinU
.Solution: A truth assignment for
U
that satises as many lauses aspossibleinC
.This problem is APX-hard,a dire t onsequen eof
thefollowinglemma:
Lemma3.0.1(Berman& Karpinski [2℄) For
ev-ery
ǫ > 0
,itishardtoapproximate3
-OCC-MAX2
SAT withinfa tor2012/2011 − ǫ
.Here
3
-OCC-MAX2
SAT is the MAX2
SAT problem where ea h variable an o urat most three times, aspe ial aseofMAX
2
SAT(2
L).The redu tion is divided into two parts. The rst
partredu es anarbitraryMAX
2
SAT(2
L)instan e to theminimumline overingproblem(MLCP).These -ond part redu es the MLCP to the minimum guard
overing problem for
2
-link polygon instan es. Note that the instan es overed by the MLCP all onsistsof non-parallellines. This is important in the se ond
partof the redu tion, sin e we an onstru t a
2
-link polygonfrom su hanarrangement.3.1 The Redu tion to MLCP
Let
(U, C)
beaMAX2
SAT(2
L)instan e. Westartthe redu tionby reating, forea h variablex
i
∈ U
, a setL
i
of eight lines with interse tions asin Fig. 1. Ea h setL
i
forms a onsisten y gadget and they are built so that no two lines are parallel. Furthermore, thereare no interse tions between three lines ex ept those
expli itlydes ribed. Notethat thelinesin Fig.1that
appeartobeparallelreallyarenot. Inea h
L
i
thelinesa
andc
represent theliteralx
i
, and thelinesb, d
the literal¬x
i
. We allthese linesliterallines.A B C D a d
Figure1: The onsisten ygadget.
Forea h lause
(l
i
, l
j
) ∈ C
we reate aline, denoted lauselinethatinterse tsallotherlinesintheonstru -tion.
Furthermore,forea hliteralinea h lausewe reate
anadditionalline. This linepasses throughthe
inter-se tion point between the lause line and the literal
line of the onsisten y gadget representing the
orre-sponding variable; seeFig.2. Note that thereare two
possible literal lines to hoose from and that a literal
in the MAX
2
SAT(2
L) instan e ano ur in at most two lauses. Thus, there isalwaysafreeliteralline tohoosefromatanytimeinthe onstru tion. Notethat
the onstru tion ontainsinterse tionsoftwoandthree
linesonly,andthatnolinesareparallel. The
onstru -tion ontainsatotalof
8|U |
linesin onsisten ygadgets and3|C|
additionallines forthe lauses, foratotalof3|C| + 8|U |
lines. LetL
be the set of lines thus on-stru ted.Aninterestingset
S
forL
isamaximalsetofpoints su hthateverypointx ∈ S
istheinterse tionofthree lines inL
and every line inL
ontains at most one pointfromS
. Notethatonevery lauseline thereare two interse tion points that ould be in luded in aninteresting set and in every onsisten y gadget there
are four interse tion points. No other points an be
in luded.
Lemma3.1.1 The size of a solution to the MLCP
onstru tionis
3|C| + 8|U | − |S|
2
,
where
S
isthemaximuminterestingsetinthesolution.Proof: We an view the solution to the MLCP as
follows. Every point in the maximum interesting set
x
i
x
j
x
i
¬x
i
¬x
i
¬x
j
x
j
x
j
¬x
j
(¬x
i
, x
j
)
x
i
Figure2: Wherethelinesfrom theliteralsin a lause rossthelineforthat lauseweaddanewline.
an overat mosttwolines ea h. Thetotalnumberof
linesinthe onstru tionis
3|C| + 8|U |
sothesolution's sizeis3|C| + 8|U | − 3|S|
2
+ |S| =
3|C| + 8|U | − |S|
2
.
Weusetheinterestingsettodes ribeatruth
assign-ment to the orresponding MAX
2
SAT(2
L) instan e and to ount the number of lauses satised by theassignment. Theinterestingset restri tedto a
onsis-ten y gadgethasasize nolargerthan two;see Fig.1.
Thesemaximalsetsare
{A, C}
and{B, D}
. Whenthey o urinaninterestingsettheyrepresentthevaluefalse(
{A, C}
) or true ({B, D}
) of the orresponding vari-able. Wesaythatthe onsisten ygadgetinthis aseisseteither tofalseortrue(bytheinterestingset). Itis
importantto noti ethat three-wayinterse tionpoints
indierent onsisten ygadgetsareindependentofea h
other in the sense that all onsisten ygadgets anbe
set arbitrarilyto either trueorfalsewithout violating
the denition of interesting sets. However, they will
put restri tions on therest of the points in the
inter-esting set. These points alllie on lauselines and we
usethemto ountthenumberof lausessatisedbythe
truth assignment. Let us assumethat the interesting
set ontainsades riptionofatruthassignment,thatis,
every onsisten y gadgetis set to somevalue. Bythe
denition of interesting set there an be at most one
point in
S
on ea h lause line. This point represents a literal that satises the lause. It an be in ludedin the interestingset ifand only if the orresponding
onsisten ygadgetissettoavaluethat makesthe
lit-eral satisfy the lause; see Fig. 3. This implies that
the truth assignment des ribed by the interesting set
satises
c
lausesof theinstan eif andonly iftheset ontainsc
pointsfrom the lauselines, i.e. the sizeof theinterestingsetis2|U |+c
,stillundertheassumption thatevery onsisten ygadget issettosomevalue.Figure3: Theinterse tionpointonthe lauseline an
bein ludedintheinterestingsetifandonlyifthe
on-sisten ygadgetisset (totruein thisexample).
there is always a maximum interesting set onsistent
withatruthassignmentofthevariables. Ifthisistrue
thenwehaveadire t orresponden ebetweenthesize
of the over and the number of satised lauses.
As-sume that there is no maximum interesting set that
orrespondstoatruthassignment. Considerthe
maxi-muminterestingset,
S
,that ontainsthelargest num-berofpointsfromthe onsisten ygadgets. Thereisatleast one onsisten y gadget
G
that is not set, i.e., it ontainsless than twopointsinS
. Set thegadgetG
by xing oneor two points appropriately. Letp ∈ G
be oneof these points. The reasonforp
not being inS
is that the literal line that ontainsp
also ontains a pointq ∈ S
. If we removeq
fromS
then we an in ludep
inS
,therebygettinganewinterestingsetS
′
with
|S| = |S
′
|
butwithonemorepointfromthe
onsis-ten ygadgets, ontradi tingour hoi eof
S
andhen e ourassumption that there is no maximum interestingset orrespondingtoatruthassignment.
s ription of the lines is polynomial in the size of the
3
-OCC-MAX2
SAT instan e. To see this we give a des ription of how the lines are pla ed in the plane.Firstthe onsisten ygadgetsarepla edevenlyspa ed
asin Figure2. Next werotate therstgadget
45/|U |
degrees, the se ond one twi e as mu h, the third onethree timesasmu h and soforth. Sin etheangle
be-tweentwolinesinthe onsisten ygadgetisnolessthan
45
degreesthismeansthatalllinesinterse t. Wepla e the|C|
lauselinesinsu hawaythat notwolinesare parallel. This an bedone easily. Now, we pla e theextralinesattheinterse tionpointsbetweenthe lause
linesandtherelevantliterallines. This learlygivesa
polynomial des ription of the lines. We have proved
thefollowingtheorem.
Theorem3.1.2 Forevery
δ > 0
,itishardto approx-imateMLCP within fa tor28169/28168 + δ
.Proof: Given an instan e to MAX
2
SAT(2
L), letc
denotethemaximumnumberof lausessatisable. Wean bound the number of variables to
|U | ≤ 2|C|
, sin e ea h lause ontains2
literals. From Lem-mas3.0.1, 3.1.1, and thefollowingdis ussion we inferthattheinapproximabilityratiois
l
3|C|+6|U|−2011c/2012+ǫ
2
m
l
3|C|+6|U|−c
2
m
≥
28169/28168 + δ,
sin e theworst ase o urs when
c = |C|
. Theδ
de-pends onǫ
.3.2 The Redu tion to MGCP
In the se ond part of the redu tion we are given an
arrangementofnon-parallellines. A re tangle
R
su- iently large to ontainall interse tions between linesin
L
is reated. Consideraline segmentinR ∩ L
. At oneendpointweputaspikeinthere tangle;seeFig4.The ones visiblefrom anytripletof spikesshould
in-terse t if and only if the orresponding line segments
ross. If this riterion is met then the interse tions
betweenline segmentswill orrespondto areasin this
polygon. Thus,fromaguardsetinthepolygonweget
apoint set in the MLCP. That is, ifwe an solve the
minimumguard overingproblemthenweimmediately
getasolutiontotheMLCP.
Theorem3.2.1 The minimum guard overing
prob-lemisAPX-hard.
4 Con lusion
Wehaveprovedthattheminimumguard overing
prob-lem restri tedto
2
-linkpolygonsis APX-hard. In ad-dition wegive anexpli itlowerbound ontheapprox-Figure4:Thesetoflinesareembeddedinsidea
re tan-gle.Wheretheresultinglinesegmentsmeetthe
bound-arywe reateanarrowspike.
A knowledgments We would like to thank Piotr
Bermanforprovidingus withvaluableinformationon
the
3
-OCC-MAX2
SATproblem.Referen es
[1℄ A. Aggarwal. The Art Gallery Theorem: Its
Varia-tions,Appli ationsandAlgorithmi Aspe ts. PhD
the-sis,JohnsHopkinsUniversity,1984.
[2℄ P.Bermanand M. Karpinski. Onsome tighter
inap-proximability results. In Pro . of the 26th
Interna-tional Colloquium on Automata, Languages and
Pro-gramming,pages200209,1999.
[3℄ V.Chvátal. A ombinatorialtheoreminplane
geome-try.J.Combin.Theory Ser.B,18:3941, 1975.
[4℄ M. de Berg, M. van Kreveld, M. Overmars, and
O.S hwarzkopf. Computational Geometry. Springer
Verlag,2000.
[5℄ S.Eidenbenz. Inapproximability results for guarding
polygons without holes. In Pro . 9th Annual
Inter-nationalSymposiumonAlgorithms andComputation,
pages427436,1998.
[6℄ R.Honsberger. Mathemati la GemsII. Mathemati al
Asso iationofAmeri a,1976(104110).
[7℄ V.S.A. Kumar, S.Arya,and H.Ramesh. Hardness
ofset overwithinterse tion
1
. InPro .27th Interna-tionalColloquium,ICALP,pages624635,2000.[8℄ D.T.LeeandA.K.Lin. Computational omplexityof
artgalleryproblems. IEEETransa tions on
Informa-tionTheory,IT-32:276282,1986.
[9℄ B.J. Nilsson. Guarding Art Galleries Methods for
MobileGuards. PhDthesis,LundUniversity,1995.
[10℄ J.O'Rourke. Art Gallery Theorems and Algorithms.