You are on page 1of 13

!"!

#$#% '()*+,( -
Creedy lnLerval Colourlng AlgorlLhm
lnLerval Craphs
Craph MlS and graph colourlng
Chordal graphs
1he problem seL
llxed order vs adapuve order greedy
lnLerval colourlng
We wlll now conslder a mlnlmlzauon problem,
namely glven a seL of lnLervals, we wanL Lo colour
all lnLervals so LhaL lnLervals glven Lhe same
colour do noL lnLersecL and Lhe goal ls Lo Lry Lo
mlnlmlze Lhe number of colours used.
We could slmply apply Lhe opumal !-machlne
lS for lncreaslng ! unul we found Lhe smallesL
! LhaL ls sumclenL. (noLe: 1hls ls a slmple
example of a !"#$%"&'(# *&+ ,+-./*"% whlch ls
an essenual concepL when we sLudy n-
compleLeness.)
Creedy lnLerval colourlng
Conslder Lhe "#$ (earllesL sLarung ume) for
lnLerval colourlng. namely, havlng sorLed Lhe
lnLervals by non decreaslng sLarung umes, we
asslgn each lnLerval Lhe smallesL numbered
colour LhaL ls feaslble glven Lhe lnLervals
already coloured. (8ecall LhaL "#$ ls a Lerrlble
algorlLhm for lS.) noLe LhaL Lhls algorlLhm ls
equlvalenL" Lo %&$ (laLesL nlshlng ume rsL).
1heorem: "#$ ls opumal for lnLerval colourlng
Creedy lnLerval colourlng
roof of opumallLy (skeLch)
1he proof Lechnlque we wlll use here ls also
one oen used for provlng approxlmauons.
1he ldea ls Lo nd some bound (or bounds)
LhaL '() +,-./,( musL sausfy and Lhen relaLe
LhaL Lo Lhe algorlLhms soluuon.
ln Lhls case, conslder Lhe maxlmum number of
lnLervals ln Lhe lnpuL seL LhaL lnLersecL aL any
glven polnL. 1he number of colours musL be aL
leasL Lhls large.
lL remalns Lo show LhaL Lhe greedy algorlLhm
wlll never use more Lhan Lhls number of
colours.
Why doesn'L Lhe greedy colourlng
algorlLhm exceed Lhls lnLrlnslc bound?
8ecall LhaL we have sorLed Lhe lnLervals by non-
decreaslng sLarung ume (l.e. earllesL sLarL ume
rsL).
Conslder Lhe rsL ume (say on some lnLerval 0)
LhaL Lhe greedy algorlLhm would have used lLs
maxlmum number 1 of colours. 1hen lL musL be
LhaL Lhere are 123 lnLervals lnLersecung 04 8uL Lhen
Lhese lnLersecung lnLervals musL all lnclude Lhe
sLarL ume + of lnLerval 0 '(5 67(87 Lhere are 1
lnLervals lnLersecung aL +.
lnLerval graphs
1here ls a naLural way Lo vlew Lhe lnLerval
schedullng and colourlng problems as graph
problems.
LeL I be a seL of lnLervals. We can consLrucL
Lhe '%0+,1+/*"% 2,(!3 C(I) = 9:; "< where
: = I and (.;=) ls an edge ln " l Lhe lnLervals
correspondlng Lo . and = lnLersecL. Any graph
LhaL ls Lhe lnLersecuon graph of a seL of
lnLervals ls called an '%0+,4(# 2,(!3.
Craph MlS and colourlng
LeL C = (v,L) be a graph. 1he followlng Lwo problems
are known Lo be ``n- hard Lo approxlmaLe'' for
arblLrary graphs:
Craph MlS: A subseL v' of v ls an lndependenL seL (aka
sLable seL) ln C lf for all u,v ln v', (u,v) ls noL an edge ln
L. 1he maxlmum lndependenL seL problem ls Lo nd a
maxlmum slze lndependenL seL v'.
Craph colourlng: A funcuon c mapplng veruces Lo [1,2,
.,k} ls a valld colourlng of C lf c(u) ls noL equal Lo c(v)
for all (u,v) ln L. 1he graph colourlng problem ls Lo nd
a valld colourlng so as Lo mlnlmlze Lhe number of
colours k.
LmclenL algorlLhms for lnLerval graphs
Clven a seL I of lnLervals, lL ls easy Lo consLrucL lLs
lnLersecuon graph C(I). lL ls noL obvlous buL glven
any graph C, Lhere ls a (llnear ume) algorlLhm Lo declde
lf C ls an lnLerval graph and lf so Lo consLrucL an
lnLerval represenLauon.
1he MlS problem (resp. colourlng problem) for
lnLerval graphs ls Lhe MlS (resp. colourlng) problem for
lLs lnLersecuon graph and hence Lhese problems are
emclenLly solved for lnLerval graphs. ls Lhere a graph
Lheoreuc explanauon? ?LS : lnLerval graphs are chordal
graphs.
1he mlnlmum colourlng number (chromauc number)
of a graph ls always aL leasL Lhe slze of a maxlmum
cllque. 1he greedy lnLerval colourlng proof shows LhaL
for lnLerval graphs Lhe chromauc number = max cllque
slze.
WhaL are chordal graphs?
Many equlvalenL ways Lo dene chordal graphs.
lor our purposes, leLs dene chordal graphs C =
(v,L) as Lhose havlng a !+,5+/0 +#'&'%(*"%
",-+,'%2 LC of Lhe veruces, LhaL ls an orderlng
v(1), v(2), .,v(n) such LhaL for all l, nbhd(v(l))
lnLersecL [v(l+1), .,v(n)} ls a cllque.
noLe LhaL orderlng lnLervals by earllesL nlshlng
umes wlll provlde a LC for Lhe lnLersecuon
graph of lnLervals and hence lnLerval graphs are
chordal.
More on chordal graphs
We can absLracL Lhe argumenLs used for lnLerval
selecuon (resp. colourlng) Lo show Lhe opumallLy
of greedy algorlLhms for any chordal graph uslng
a LC (reverse LC) orderlng.
An equlvalenL (and lnlual) denluon of chordal
graphs are Lhose whlch do noL have any k-cycles
(for k > 3) as lnduced subgraphs.
WhaL are and are noL chordal graphs? lor
example a 4-cycle cannoL be an lnLerval graph.
CommenLs on Lhe problem seL
Cuesuon 1 of Lhe problem seL dlscusses a
greedy algorlLhm for graph colourlng. lL uses a
breadLh rsL search Lo deLermlne Lhe order of
veruces belng coloured. lor arblLrary graphs,
deLermlnlng ls a gfraph can be 3 coloured ls
n hard.
Cuesuon 2 dlscusses a schedullng problem
relaLed Lo Lhe !lS problem as wlll be
explalned. We wlll rsL explaln Lhe noLauon.
llxed order vs adapuve order greedy
1he algorlLhms for lnLerval schedullng and
colourlng choose a (xed) orderlng of Lhe
lnpuL lLems (l.e. lnLervals) and Lhen conslder
Lhem ln LhaL order. 1he algorlLhm sLaLed ln
quesuon 2 of Lhe problem seL has a somewhaL
dlerenL sLrucLure Lhan Lhe algorlLhms for
lnLerval schedullng/colourlng ln LhaL Lhe order
ln whlch lnpuLs (e.g. [obs) are consldered ls
declded adapuvely based on prevlous
declslons.

You might also like