Frédéri Meunier
∗
Deember 1,2007
Abstrat
ThispaperdealswithdiretproofsandombinatorialproofsofthefamousNeklae
theoremofAlon,GoldbergandWest. Thenewresultsareadiretprooffortheaseof
twothievesandthreetypesofbeads,andaneientonstrutiveproofforthegeneral
asewithtwothieves. Thislastproof usesatheoremofKyFanwhihisaversionof
Tuker'slemmaonerningubialomplexesinsteadofsimpliialomplexes.
KeyWords: onstrutiveproof;ubialomplex;splittingneklaes.
1 Introdution
A lassialappliation oftheBorsuk-Ulamtheoremis theSplitting Neklaetheorem (see
the book of Matou²ek [10℄). Provedrst by Goldberg andWest in 1985 [9℄, this theorem
got asimplerproof in 1986,foundbyAlon andWest andusing theBorsuk-Ulamtheorem
[4℄.
Twothieveshavestolenapreiousneklae,whihhas
n
beads. These beads belong tot
dierenttypes. Assume thatthereisanevennumberofbeadsofeahtype,say2a i
beadsof type
i
, for eahi ∈ {1, 2, . . . , t}
, wherea i
is a nonzero integer. Remark that we have2 P t
i=1 a i = n
.Thebeads are xed onan openhainmade of gold. An exampleisgivenin Figure 1.
Thedierenttypesareenodedbythenumbers1,2,3.
Aswedonotknowtheexatvalueofeahtypeofbeads,afairdivisionoftheneklae
onsistsofgivingthesamenumberofbeadsofeahtypetoeahthief. Thenumberofbeads
of eah type is even, hene suh adivision is alwayspossible: utthe hain at the
n − 1
possiblepositions. But thehain is made ofgold! It isa shamethat wedamageit. Isn't
it possibletodo thedivision withlessuts? Theansweris yes, asshownby thefollowing
theorem provedbyGoldbergandWest:
Theorem 1 Afairdivisionof theneklaewith
t
typesofbeads betweentwothievesanbedone with nomorethan
t
uts.In 1987, Alon proved the following generalization, for aneklaehaving
qa i
beads foreahtype
i
,a i
integer,using ageneralizationoftheBorsuk-Ulamtheorem [1℄:Theorem 2 A fair division of the neklae with
t
types of beads betweenq
thieves an bedone with nomorethan
t(q − 1)
uts.∗
LVMT,EoleNationaledesPontsetChaussées,6-8avenueBlaisePasal,CitéDesartesChamps-sur-
Marne,77455Marne-la-Valléeedex2,Frane. E-mail:frederi.meunierenp.fr
Figure1: Anopenneklae,with12beads,of3dierenttypes,tobedividedbetweenAlie
and Bob.
2 1 2 1 3 1 3 1 3 2 3 2
.
.
.
.
.
.
.
.
. .
. . . . . . . . .
.
.
.
.
.
.
.
.
N
^ ?
Alie Bob Alie Bob
Figure2: Afairdivision betweenAlieandBob.
For
q = 2
,thislasttheoremoinideswithTheorem1. Furthermore,thisboundistight, evenwhena 1 = a 2 = . . . = a t = 1
: onsidertheneklaeforwhihallthebeadsofthersttypeome rst,then allbeads oftheseond type,thenallbeads ofthethird type,andso
on.
Exept for partiular ases, all known proofs of this theorem are topologial and use
ontinuous tools. Setion 2is devoted to diret and non-topologial proofs for partiular
ases ofthese theorems: oneonthem isthe rstproof ofthis typeforthe ase
q = 2
andt = 3
.In Setion 3, we give apurely ombinatorial 1
and onstrutive proof for the omplete
ase
q = 2
(althoughthetopologialinspirationremainsnotable).It uses a ubial version of Tuker's lemma found by Ky Fan (Theorem 3), involving
ubial omplexes instead of simpliial omplexes. We give a onstrutive proof of this
theorem (suh aproof wasnot known before thepresent paper; nevertheless, ourproof is
inspiredbytheonstrutiveproofgivenbyPresottandSuforaombinatorialgeneralization
of Tuker'slemmafound byKyFan, [13℄)and showthat it suitsto theneklaeproblem.
It givesaneientalgorithmforthedivisionoftheneklaebetweentwothieves: ontrary
to the known algorithm, our result needs no approximation or rounding method and no
arbitrarilynetriangulationoftheross-polytope. Furthermore,the
d
-ubesofourubialomplexhaveasimpleinterpretation: eah
d
-ubeorrespondstothehoieofd
beads. Wewill also see that our method gives also an almost-fair division of the neklae between
two thieves whenthe numberof beads in any typeis not neessarilyeven(2-splittings in
theterminologyofAlon,Moshkovitzand Safra[3℄).
2 Diret proofs
2.1 A diret proof for
t ≤ 2
and anyq
This proofwasrstgivenin thepaperbyEpping,HohstättlerandOertel[7℄. This paper
deals with dierent aspets of a paint shop problem: one of the ases orresponds to the
Neklae theorem presented as aonjeture (Conjeture 10) by the authors who did not
1
AordingtoZiegler[15℄,a ombinatorialproofinatopologialontextisaproof usingno simpliial
approximation,nohomology,noontinuousmap.
theneklae,see [11℄.
If
t = 1
,i.e.,ifthebeadsareonlyof onetype,thesolutionisstraightforward: splitthe neklaeinq
partsofthesamesize,whihneedsq − 1
uts(andhasomplexityO(q)
).If
t = 2
: the proofworksby indution onq
. Forq = 1
, thereis nothingto prove. Letusnowsupposethat
q ≥ 2
. We(virtually)splittheneklaeintoq
disjointsub-neklaesofa 1 + a 2
onseutivebeads;theneklaeisopen andhorizontal;thebeadsgetthenumbers 1ton
,from lefttoright.If oneofthese
q
sub-neklaesontainsa 1
beads of type 1anda 2
beads of type2, weremovethissub-neklae,andgiveitto oneofthethieves,andbyindution,weknowthat
thereisafairsplittingoftheremainingin
2(q −2)
uts;alltogether,thisgives2 + 2(q −2) = 2(q − 1)
uts.Hene, we an assume that there is no suh sub-neklae among the
q
sub-neklaes.Thus, thereisatleastonesub-neklaewhihontainsstritlymorethan
a 1
beadsoftype1,andanotherwhihhasstritlylessthan
a 1
beadsoftype1. Let'salltherstsub-neklaeM +
andtheseond oneM −
.Ifweslideawindowofsize
a 1 + a 2
overtheneklae,startingatM −
untilwereahM +
using movesofexatlyonebead,thewindowwillontain
a 1
beadsof type1at leastone.The
a 1 + a 2
beadsofthewindowanbegiventooneofthethieves,andthenbyindution,asbefore,
2 + 2(q − 2) = 2(q − 1)
utsaresuient.The time omplexity an be omputed as follows: eah bead reeives the number of
beadsofeahofthetwotypespreedingitontheneklae(thebeaditselfinluded),whih
anbedonein
O(n)
operations: so,whenweonsiderasub-neklae,weknowimmediately howmanybeadsofeahtypeareinthissub-neklae;adihotomyallowsustondasub-neklaehaving
a 1
beadsoftype1anda 2
beadsoftype2inO(log n)
. Thetotalomplexityis thus:
O(n + q log n)
.2.2 A diret proof for the ase
a 1 = a 2 = . . . = a t = 1
, anyt
and anyq
Weonsider the beads oneafter theother; from left to right. Wealwayshavea urrent
thief. Whenonsideringthe
k
thbead,wehekiftheurrentthiefhasabeadofthistypeornot.
Ifhedoes: wehangetheurrentthief andtakeathiefnothavingabeadof thistype,
wegivethebeadtothisnewurrentthief andgotothenextbead.
Ifhedoesnot: wegivethebeadto theurrentthief andgotothenextbead.
Thisproedureensuresadivisionin atmost
t(q − 1)
uts: indeed, weneverhangetheurrent thief when wemeet the rst bead of a given type. A hange of the urrentthief
is preisely aut ofthe neklae. Thenumberof utsis thus less thanorequalto
n
, thenumber of all beads, minus
t
, the number of rst meetings of a type of beads: in total:n − t = qt − t = t(q − 1)
.Moreover,thisalgorithmisgreedyandhasomplexity
O(n)
.2.3 A diret proof for
t = 3
andq = 2
2.3.1 Main steps
The neklae is identied with the interval
]0, n[⊂ R
. The beads are numbered with theintegers
1, 2, . . . , n
,fromlefttoright. Thek
thbeadoupiesuniformlytheinterval]k−1, k[
.For
S ⊆]0, n[
, wedenote byφ i (S)
thequantity ofbeads oftypei
in positions inludedin
S
. Morepreisely,wedenote by1 i (u)
themappingwhihindiates ifthereis abeadoftype
i
at pointu
of]0, n[
:1 i (u) =
1
ifthe⌈u⌉
thbeadisoftypei 0
ifnot.Wehavethen
φ i (S) :=
Z
S
1 i (u)du.
If there is a window of
n/2
beads induing a fair division, there is nothing to prove.Thus, we anassumethat there is no
x
suh thatφ i (]x, x + n/2[) = a i
for alli = 1, 2, 3
.Moreover,weanalsoassumethat
φ 1 (]0, n/2[) > a 1
. Thesetwoassumptionsarealledthe starting assumptions.First, we unite the types 2 and 3 in a single type, alled type 2'. Hene, we dene
φ 2 ′ := φ 2 + φ 3
.Then,webuildagraph
G
,whosevertiesaretriples(c 1 , c 2 , c 3 )
,suhthat• c 1
,c 2
andc 3
areintegers,• 0 ≤ c 1 ≤ c 2 ≤ c 3 ≤ n
,• φ 1 (]0, c 1 [∪]c 2 , c 3 [) = a 1
andφ 2 ′ (]0, c 1 [∪]c 2 , c 3 [) = a 2 + a 3
.Distint triples
(c 1 , c 2 , c 3 )
and(d 1 , d 2 , d 3 )
are neighbors inG
if|c i − d i | ≤ 1
for alli = 1, 2, 3
.Vertiesof
G
orrespondthusto utsofthe neklaegivingthesamenumberof beadsto eah thief,and dividingfairly thebeads oftype1. Withoutlossofgenerality,wedeide
that Aliegetsthepart
]0, c 1 [∪]c 2 , c 3 [
andBob theremaining.Remarkthatthestartingassumptionimpliesthatthereisnovertexoftheform
(0, c 2 , n)
.Inthenextparagraph,thefollowinglemmaisproved:
Lemma 2.1 Thefollowing holdsunder thestarting assumptions:
(i) Theverties
(c 1 , c 2 , c 3 )
of odddegree haveeitherc 1 = 0
orc 3 = n
.(ii) Thenumberofvertiesofodddegreesuhthat
c 1 = 0
idoddandequalsthe numberofvertiesofodd degreesuhthat
c 3 = n
.Theendoftheproofisthenstraightforward. Wesplitthesetofvertiesofodddegrees
of
G
into twodisjoint subsetsA
andB
:A
is the set of vertiesof odddegrees suh thatφ 3 (]0, c 1 [∪]c 2 , c 3 [) < a 3
andB
thesetofvertiesofodddegreessuhthatφ 3 (]0, c 1 [∪]c 2 , c 3 [) >
a 3
(aording tothestartingassumptions,there isnovertex(c 1 , c 2 , c 3 )
of odddegreesuhthat
φ 3 (]0, c 1 [∪]c 2 , c 3 [) = a 3
). Hene|A| = |B|
and this number isodd. This impliesthatthere isa pathin
G
that links a vertex inA
to avertexinB
.φ 1
andφ 2 ′ = φ 2 + φ 3
areonstanton
G
. Moreover,thevalueofφ 3
fortwoadjaentvertiesdiersbyat mostone.Hene,weneessarilyhaveavertexonthis pathsuhthat
φ 1 = a 1
,φ 2 = a 2
andφ 3 = a 3
.2.3.2 Proof ofLemma2.1 (
t = 3
andq = 2
)Intheproof, wedistinguishtheverties
(c 1 , c 2 , c 3 )
aordingtothetypeofbeadsadjaentto thepoint where the neklaeis ut. Below, therst
|
representsc 1
, theseond onec 2
andthethird one
c 3
. Thenumbersonbothsidesofthej
thline|
showthetypeofthebeadwhihisontheleftsideofpoint
c j
andthetypeontherightsideofpointc j
. Forinstane:. . . 1
|{z}
Alie
| 2 ′ . . . 2 ′
| {z }
Bob
| 1 . . . 1
| {z }
Alie
| 2 ′ . . .
| {z }
Bob
.
This vertexis ofdegree0,beausenomoveofa
|
anbeompensatedby anothermoveinorder to maintainthe numberof beads of types1 and2' reeivedby eah thief. Another
exampleisthefollowing(where
c 1 = 0
):|1 . . . 1|1 . . . 1|2 ′ . . . .
Thedegreeofsuhavertexequals3. Indeed,thereare3neighbors:
|1 . . . |11 . . . |12 ′ . . . , 1| . . . 11| . . . 1|2 ′ . . .
and
1| . . . 1|1 . . . |12 ′ . . .
Inanyofthese threeases,themoveofautisompensatedbyanothermove.
Let us ompute the degree of a vertex
(c 1 , c 2 , c 3 )
in general. Three ases have to bedistinguished
1.
c 1 = 0
: the vertexis of theform:|a . . . b|c . . . d|e . . .
witha, b, c, d, e ∈ {1, 2 ′ }
. Foursubaseshavetobedistinguished,thehekingbeingeasy:
•
Ifb 6= e
andc = d
,thedegreeis1ifa 6= c
and3ifnot.•
Ifb 6= e
andc 6= d
,thedegreeis1ifb = c
and3ifnot.•
Ifb = e
andc = d
, the degreeis 4ifa = b = c
, 2 ifa 6= b
, and 0ifa = b
andb 6= c
.•
Ifb = e
andc 6= d
,thedegreeis2.2.
c 3 = n
: thisasean betreatedexatlyastheasebefore.3.
c 1 > 0
andc 3 < n
: we enumerate subases while avoiding subases whih an bededued from theothers bysymmetries ortheexhange of 1and 2': wegiveonlya
fewsubases,forthehekingisalwaysstraightforward:
•
Subase. . . 1|1 . . . 1|1 . . . 1|1 . . .
: thedegreeequals6.•
Subase. . . 1|1 . . . 1|1 . . . 1|2 ′ . . .
: thedegreeequals4.•
Subase. . . 1|1 . . . 1|1 . . . 2 ′ |1 . . .
: thedegreeequals4.•
Subase. . . 1|1 . . . 1|1 . . . 2 ′ |2 ′ . . .
: thedegree equals2. Indeed, there areexatlytwoneighbors:
. . . 11| . . . 11| . . . 2 ′ |2 ′ . . .
and. . . |11 . . . |11 . . . 2 ′ |2 ′ . . .
.•
Subase. . . 1|1 . . . 1|2 ′ . . . 2 ′ |1 . . .
: thedegreeequals2.•
Subase. . . 1|1 . . . 2 ′ |1 . . . 1|2 ′ . . .
: thedegreeequals2.•
Subase. . . 1|1 . . . 1|2 ′ . . . 2 ′ |2 ′ . . .
: thedegreeequals2. Indeed,there areexatlytwoneighbors:
. . . 1|1 . . . 12 ′ | . . . 2 ′ 2 ′ | . . .
et. . . |11 . . . |12 ′ . . . 2 ′ |2 ′ . . .
.•
Subase. . . 1|1 . . . 2 ′ |1 . . . 2 ′ |2 ′ . . .
: thedegreeequals2.•
Subase. . . 1|2 ′ . . . 2 ′ |1 . . . 1|2 ′ . . .
: thedegreeequals0.•
Subase. . . 1|2 ′ . . . 1|2 ′ . . . 2 ′ |1 . . .
: thedegreeequals4.Alldegreesareeven.
It followsthat a vertex of odd degreeis neessarily suh that
c 1 = 0
orc 3 = n
. Thisprovespoint(i) ofLemma 2.1. Notethat avertex hasan odd degreeifand only if
b 6= e
(withthenotationsdenedin therstaseabove).
Weprovenowthat: Thereisan odd numberof vertiesof odd degree suhthat
c 1 = 0
.Let
ǫ > 0
be averysmall real. We rst show that a vertex has an odd degreeif thefollowingholds: oneof theintervals
]c 2 − 1, c 3 − 1 + ǫ[
and]c 2 , c 3 + ǫ[
hasaproportion ofbeads of type1stritly smallerthan
a 1
a 1 +a 2 +a 3
and theother hasaproportionof beads of
type1stritlygreaterthan
a 1
a 1 +a 2 +a 3
. Formally,wearegoingto showthat theoddverties
(0, c 2 , c 3 )
aresuhthatφ 1 (]c 2 − 1, c 3 − 1 + ǫ[)
a 1 + a 2 + a 3 + ǫ − a 1
a 1 + a 2 + a 3
and
φ 1 (]c 2 , c 3 + ǫ[)
a 1 + a 2 + a 3 + ǫ − a 1
a 1 + a 2 + a 3
haveoppositesigns.
Thistwoquantitiesarerespetivelyequalto
a 1 + 1 {b=1} − (1 − ǫ) 1 {d=1}
a 1 + a 2 + a 3 + ǫ − a 1
a 1 + a 2 + a 3
and
a 1 + ǫ1 {e=1}
a 1 + a 2 + a 3 + ǫ − a 1
a 1 + a 2 + a 3
,
where
b
,d
, ande
are beads in positionsspeiedin therstaseabove,and1 P
takesthevalue1(resp. 0)if
P
istrue(resp. false). It iseasy to seethat these twoquantities haveoppositesignsifandonlyif
b 6= e
. Hene theyhaveoppositesignsifandonlyifthevertexhasodddegree.
We show now that there is an odd number of suh verties. With the same kind of
argumentsasin Subsetion2.1,i.e., withtheinterval
]c 2 , c 3 + ǫ[
movingfrom lefttoright,wesee thatthereisanoddnumberofsuh situations,forthesignof
φ 1 (]u, u + a 1 + a 2 + a 3 + ǫ[)
a 1 + a 2 + a 3 + ǫ − a 1
a 1 + a 2 + a 3
hangesstritlywhentheinterval
]u, u + a 1 + a 2 + a 3 + ǫ[
movesfrom]0, a 1 + a 2 + a 3 + ǫ[
to
]a 1 + a 2 + a 3 , n + ǫ[
(this isatrueforǫ = 0
, aordingto thestartingassumptions,and thus forallǫ > 0
smallenough). Hene, thereis anoddnumberof verties ofodddegreesuhthat
c 1 = 0
.By symmetry, there is also an odd number of verties of odd degree having
c 3 = n
.Furthermore,wehavethe followingequivalene, sinethe ondition
b 6= e
is the samefortheverties
(0, c 2 , c 3 )
and(c 2 , c 3 , n)
:deg(0, c 2 , c 3 )
odd⇔ deg(c 2 , c 3 , n)
odd.
Thelemmaisproved.
2.3.3 Final omments
Let'sgiveanillustrationofthisprooffortheexampleofFigure1.
Theorrespondinggraph
G
isillustratedin Figure3. Thereisapath(atually,many) linking vertex(0, 4, 10)
, whih gives more beads of type 3 to Alie, to vertex(4, 10, 12)
,(1, 4, 9) (0, 4, 10) (0, 5, 11) (1, 5, 10)(2, 6, 10) (3, 7, 10) (4, 8, 10) (5, 9, 10) (4, 9, 11)(4, 10, 12) (5, 10, 11)
(5, 11, 12) (5, 8, 9)
(3, 6, 9) (2, 7, 11) (1, 2, 7)
(5, 6, 7)
(3, 4, 7)
Figure3: Graph
G
orrespondingto theexample(witht = 3
andq = 2
)ofFigure1.whih givesmore beads of type3 to Bob. The verties
(1, 5, 10)
and(2, 6, 10)
lie on thispathandprovidefairdivisions.
On the other hand, it is not lear whether it is possible to derive from this proof an
eientalgorithmthatwouldhaveabetteromplexitythan
O(n 3 )
(G
mightbenotexpliitlyomputed).
Furthermore,themethod istedious. Is itpossibletondsomesimpliation?
3 Ky Fan's ubial theorem and onstrutive proof for
two thieves
3.1 Preliminaries
Wegivenow aonstrutiveand ombinatorial proof of Theorem 1. We ould think that
suh aproofould be easilydedued from Tuker'slemma. Indeed,Theorem 1isadiret
appliation of Borsuk-Ulam's theorem, whih has a onstrutive proof through Tuker's
lemma. ThisisdoneforinstanebySimmonsandSuin[14℄,buttheresultisnotompletely
satisfatorybeausethismethodrequiresatriangulation
T
ofthe(t + 1)
-dimensionalross- polytopesuh thatthediameter ofanysimplexinT
is aO(1/(nt))
, andthen itrequiresarounding method.
Ourmethoddoesnotpresentthosediulties. Itbasesitselfonatheorem(Theorem3)
foundbyKyFanin 1960whihisaubialversionofTuker'slemma[8℄. Wegivebelowa
onstrutiveproofofthistheorem. Theubialomplexweneedisompletely natural.
3.2 Notation and denitions
Let
C
beaube,∂C
theboundaryofC
, andV (C)
thesetofvertiesofC
.Twofaetsofad
-ubearesaidto beadjaent iftheirintersetion isa(d − 2)
-ube. Foragivenfaetσ
ofC
,there isonlyonenon-adjaentfaet. Twonon-adjaentfaetsareopposite.Aubialomplex isaolletion
C
ofubesembeddedinaneulidianspaesuh that•
ifτ ∈ C
andifσ
isafaeofτ
thenσ ∈ C
and•
ifσ
andσ ′
areinC
,thenσ ∩ σ ′
isafae ofbothσ
andσ ′
.Aubialomplexissaidto bepureifallmaximalubeswithrespettoinlusionhave
samedimension.
V ( C )
isthesetofvertiesoftheubialomplexC
.A
d
-dimensional ubial pseudo-manifoldM
is apured
-dimensional ubialomplexin whiheah(d − 1)
-dimensionalubeisontainedin atmosttwod
-dimensionalubes. The boundaryofM
,denotedby∂ M
,isthesetofall(d−1)
-ubesσ
suhthatσ
belongstoexatlyone
d
-dimensionalubeofM
.Letus denoteby
C d
theubialomplexwhih istheunionofthe ubed := [−1, 1] d
and allitsfaes(seeFigure6).
Let
k ≥ 0
be an integer. Givenk
distint integersi 1 , i 2 , . . . , i k
, eah of them in{1, 2, . . . , d}
,andk
numbersǫ 1 , ǫ 2 , . . . , ǫ k
,eahofthembelongingto{−1, +1}
,wedenoteF
i 1 i 2 . . . i k
ǫ 1 ǫ 2 . . . ǫ k
:= {(x 1 , x 2 , . . . , x d ) ∈ d : x i j = ǫ j
forj = 1, 2, . . . , k},
whihisa
(d − k)
-faeofd
. Fork = 0
,thisfaeisd
itself.We endow
∂ d
with hemispheres. These hemispheres are denoted+H 0
,−H 0
,+H 1
,−H 1
,. . .
,+H d−1
,−H d−1
.+H i
(resp.−H i
)isthesetofpoints(x 1 , x 2 , . . . , x d )
of∂ d
suhthat
x i+1 ≥ 0
(resp.x i+1 ≤ 0
),and suh that, ifi ≤ d − 2
, thenx j = 0
forj > i + 1
(inFigure 4,thehemispheresof
∂ 3
arehighlighted).Thearrier hemisphere ofaubeis thehemisphereofsmallestpossibledimensionon-
tainingit.
Observation2.1 For
i ∈ {0, 1, . . . , d − 1}
,H i
is homeomorphi tothe ballB i
,(+H i ) ∪ (−H i )
is homeomorphi to thei
-dimensional sphereS i
, ifi ≥ 1
,∂(+H i ) = ∂(−H i ) = (+H i−1 ) ∪ (−H i−1 )
andH d−1 ∪ −H d−1 = ∂ d
.Let
C
andD
be ubial omplexes, and letλ : V ( C ) → V ( D )
bea map.λ
is alled aubial mapif
λ
satisesthefollowingonditions:1. forevery
σ ∈ C
,thereis aτ ∈ D
suhthatλ(V (σ)) ⊆ V (τ)
2.
λ
takesadjaentvertiestoadjaentvertiesorbothto thesamevertex.Byaslightabuseof notation,wewrite
λ : C → D
. Weemphasizethefatthatλ(V (σ))
inthedenition aboveisnotneessarilythevertexsetofaube.
AordingtoLemma18ofthepaperofEhrenborgandHetyei[6℄,if
λ : V ( n ) → V ( n )
is anon-injetiveubialmap andif
ρ
isafaetofn
,then thereare0,2or4faetsσ
ofn
suhthatλ(V (σ)) = V (ρ)
. Itimpliesthefollowinglemma:Lemma 2.2 Let
λ : V ( d ) → V ( m )
be a ubial map (fromC d
intoC m
), withm ≥ d
,and let
ρ
be a(d − 1)
-fae ofm
. Ifλ
is not injetive, there are0
,2
or4
faetsσ
ofd
suhthat
λ(V (σ)) = V (ρ)
. Moreover,ifthereare4
suhfaetsandifσ
isoneofthem,thenthe faetopposite to
σ
isalsoone ofthese4
faets.Proof: Let
λ : V ( d ) → V ( m )
beanon-injetiveubialmapandletρ
bea(d−1)
-faeof
m
. Ifthere is nofaetσ
suh thatλ(V (σ)) = V (ρ)
, there isnothing to prove. Soletus assumethat there is afaet
σ
ofd
suh thatλ(V (σ)) = V (ρ)
. Wewant toapply theresult ofLemma 18of [6℄, whihis validif
λ : V ( d ) → V ( d )
. Let usprovethat weanmakethisassumption,thatis,that thereisa
d
-faeofm
whihontainstheimageofλ
.•
Ifλ(V ( d )) = V (ρ)
,anyd
-faeontainingρ
ontainstheimageofλ
.•
If not, there isv ∈ V ( d ) \ V (σ)
suh thatλ(v) ∈ / V (ρ)
. Letw
be thevertex ofσ
whihisadjaentto
v
,andletw ′
bethesinglevertexofσ
suhthatd(w, w ′ ) = d − 1
,where
d(·, ·)
isthedistaneinthe1-skeletonofaube,ountedinnumberofedges. Itisstraightforwardtohekthat
d(λ(w ′ ), λ(v)) = d
(taketheminimalfae ontainingboth
ρ
andλ(v)
). Letv ′
beanyvertexofd
,wehavethefollowinghainofinequalities:d = d(v, w ′ ) = d(v ′ , v)+d(v ′ , w ′ ) ≥ d(λ(v ′ ), λ(v))+d(λ(v ′ ), λ(w ′ )) ≥ d(λ(v), λ(w ′ )) = d.
Hene,
d(v ′ , v) = d(λ(v ′ ), λ(v))
andd(v ′ , w ′ ) = d(λ(v ′ ), λ(w ′ ))
. It impliesthatλ(v ′ )
,forany
v ′
ind
,isin thed
-ubeofm
ontainingρ
andλ(v)
.Weanthus assumethat
λ : V ( d ) → V ( d )
.Ifthere are4faetssuhthat
λ(V (σ)) = V (ρ)
(inthis aseonehasneessarilyd ≥ 2
),at least two of them, say
σ
andσ ′
, are adjaent. Letσ ′′
be another faet adjaent toσ
and
σ ′
. Letx ∈ V (σ ′ ∩ σ ′′ ) \ V (σ)
, and lety
be the vertex adjaenttox
inσ
. Wehavey ∈ σ ∩ σ ′ ∩ σ ′′
, andλ(x)
adjaent toλ(y)
. Moreover,letz
bethe vertexofσ
suh thatλ(z) = λ(x)
(whihexistsbyinjetivity).z
isadjaenttoy
inσ
andannotbeinσ ′
. Henez
istheonlyvertexofσ
whihisatdistane1ofy
andnotinσ ′
. Itimpliesthatz
isinσ ′′
.As
λ(z) = λ(x)
andasz
andx
aretwodierentvertiesofσ ′′
,λ(V (σ ′′ )) 6= V (ρ)
. A faetσ ′′
adjaenttoσ
andtoσ ′
annotsatisfyλ(V (σ ′′ )) = V (ρ)
.3.3 Ky Fan's ubial formula
Supposethat wehaveaubialmap
λ
ofM
,at
-dimensional ubialpseudo-manifold, intoC m
, theubialomplexmadebym
anditsfaes.Wedenoteby
α λ
i 1 i 2 . . . i m−t
ǫ 1 ǫ 2 . . . ǫ m−t
thenumberofthose
t
-ubesσ
ofM
suhthatthe verties of
σ
are mapped underλ
in a one-to-one way onto the verties of thet
-faeF
i 1 i 2 . . . i m−t
ǫ 1 ǫ 2 . . . ǫ m−t
(if
m = t
,thisfaeism
itself).Wewilllatermakeuseofthefollowingtheorem(forthease
m = t
),whoseonstrutiveproofisgivenbelow.
Theorem 3 Let
M
betheubialpseudo-manifoldwithoutboundaryobtainedbysubdividing∂ C t+1
bymeansof thehyperplanesx i = n j
wherei ∈ {1, 2, . . . , t + 1}
andj ∈ {−n + 1, −n + 2, . . . , −1, 0, 1, . . . , n − 2, n − 1}
. Letλ
beaubial map fromM
intoC m
,withm ≥ t
. Ifλ
is antipodal(i.e. if
λ(−x) = −λ(x)
),then wehave•
ifm > t
,X
ǫ 2 ,...,ǫ m−t =−1,+1
α λ
t + 1 t + 2 . . . m +1 ǫ 2 . . . ǫ m−t
= 1
mod2.
•
ifm = t
,thereisat
-ubeσ
ofM
suhthatλ(V (σ)) = V ( C m )
.For
t = 2
andn = 2
,M
isillustratedinFigure4.3.4 Construtive proof of Theorem 3
Wefollow roughlythe method found by Presott and Su [13℄ for provinga similar result
onerningsimpliialomplexes.
•
Ad
-ubeofM
(with0 ≤ d ≤ t
)issaidtobefullifitsimageunderλ
is oftheformF
d + 1 d + 2 . . . m ǫ 1 ǫ 2 . . . ǫ m−d
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . .
. .
. .
. .
. .
. . .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
.. .. . .. . .. .. . .. .. . .. . .. .. . .. . .. .. . .. . .. ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . .. .. . .. . .. .. . .. . .. .. . .. . .. .. . .. .. . .. . .. .. . .. . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .. . .. .. . .. . .. .. . .. . .. .. . .. .. . .. . .. .. . .. . .. ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.. . .. . .. .. . .. . .. .. . .. .. . .. . .. .. . .. . .. .. . .. .
. . .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
.
. . .
. . .
. .
. . .
. .
. .
. .
. . .
. . .
. . .
. . .
. . .
.
. . .
. .
. .
. .
. . .
. .
. .
. .
. .
. . .
. .
. .
. . .
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.. .. . .. . .. .. . .. .. . .. . .. .. . .. . .. .. . .. . .. ..
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
I
I
?
K
+H 0
−H 0
+H 1
−H 1
+H 2
−H 2
- 6
x 1
x 2
x 3
Figure4: Anillustrationof
M
forn = 2
andt = 2
.•
Ad
-ubeofM
issaidtobealmost-fullifitisnotfullbutoneofitsfaetsisfull.Exeptintheasewhen
d = m = t
,oneandenethesignofafulld
-ubeasthevalueof
ǫ 1
in theexpressionsabove.Lemma 3.1 For analmost-full
d
-ubeτ
,thereareonly threepossibilities:1. eitheritsvertiesaremappedunder
λ
inaone-to-onewayontothevertiesofad
-fae(withthe entries
j
andǫ j−d+1
deleted)F
d . . . ˆ j . . . m
ǫ 1 . . . ˆ ǫ j−d+1 . . . ǫ m−d+1
,
2. or
λ
isnot injetive on the vertexset ofτ
andτ
has exatly twofull faets, eah ofthemhavingthe samesign,
3. or
λ
is notinjetive on the vertex set ofτ
andτ
has exatly four full faets,eah ofthem having the same sign. Moreover, in this ase, if
σ ⊆ τ
is a full faet, then theopposite of
σ
isalsoafullfaet.Proof: Firstremarkthat if
λ
isinjetive,theimage ofτ
is ad
-fae (thiseasyassertionanbeprovedbyindution). Sine
τ
isnotfull, thisimage hasneessarilytheform givenin point1.
Hene,oneanassumethat
λ
isnotinjetive. Taketwofullfaets ofτ
, sayσ
andσ ′
.Sinetheyarefull,
λ
isinjetiveoneahofthem.λ(V (σ))
andλ(V (σ ′ ))
arethevertexsetsof
(d − 1)
-faes.Theyhaveanon-emptyintersetion,otherwiseλ
wouldhavebeeninjetiveon
τ
. Moreover,aordingtothedenitionofafullube,theimageoftwofullubesofthesamedimensionare parallel. But,twoparallel faesof aubehavingsamedimensionand
a nonemptyintersetion areequal. Thus,
λ(V (σ)) = λ(V (σ ′ ))
, and there isa(d − 1)
-faeρ
suh that forall full faetsσ
ofτ
, onehasλ(V (σ)) = V (ρ)
. This settles thestatementaboutthesign. Lemma2.2allowsto onlude.
Wean then extendthe notionof sign to almost-full ubes: thesign of analmost-full
d
-ubeisthesignofanyofitsfullfaets.Aubeissaidto beagreeableifitssignmathesthesignofitsarrierhemisphere.
Wedenethefollowinggraph
G
. Aubeσ
witharrierhemisphere±H d
isanodeofG
ifitisoneofthefollowing:
(1) anagreeablefull
(d − 1)
-ube,(2) anagreeablealmost-full
d
-ube,or(3) afull
d
-ube.Twonodes
σ
andτ
areadjaentinG
ifallthefollowinghold:(a)
σ
isafaetofτ
,(b)
σ
isfull,()
τ
hasthedimensionofitsarrierhemisphere, and(d) thesignofthearrierhemisphereof
τ
mathesthesignofσ
.Welaim that
G
isa graphin whih everyvertex hasdegree1, 2,or 4. Furthermore, a vertex hasdegree 1ifand only ifits arrierhemisphere is±H 0
, orifit is afullt
-ube.Using antipodality, we willthen show that thepoint
H 0
is neessarilylinkedby apathinG
to afullt
-ube.Toseewhy, let'sonsiderthethreekindsofnodesin
G
:(1)An agreeablefull
(d − 1)
-ubeσ
,witharrierhemisphere±H d
,isafaetofexatlytwod
-ubes,eahofwhih mustbeafulloranagreeablealmost-full ubein thesamearrier.Thissatisestheadjaenyonditions(a)(d)with
σ
,heneσ
hasdegree2inG
. Theubeσ
annotbeadjaentinG
toanyofitsfaets,otherwiseitwouldontraditondition().(2) An agreeablealmost-full
d
-ubeσ
, whose arrierhemisphere is±H d
, is adjaent inG
to its twoor four faets that are full
(d − 1)
-ubes see Lemma 3.1. A full faet ofσ
isautomatiallyanodeof
G
,sineifitisnotagreeable,thenitisarriedby∓H d−1
thesignofthearrierofthisfullfaetisthentheoppositeofthesignof
σ
. (Adjaenyondition(d)is satised beause
σ
isagreeable and beause anagreeable almost-fulld
-ube hasalwaysthesamesignasitsfullfaets).
(3)Exeptif
d = 0
,afulld
-ubeσ
witharrier±H d
hasexatlyonefaetτ
thatisfullandwhose signisthesameasthearrierhemisphereof
σ
. Heneσ
isadjaenttoτ
inG
.Ontheotherhand,
σ
isthefaet ofexatlytwo(d + 1)
-ubesin+H d+1 ∪ −H d+1
: onein
+H d+1
andonein−H d+1
, butσ
is adjaentinG
to exatlyoneof them: whih one isdetermined bythesignof
σ
,sineadjaenyondition(d)mustbesatised.Finally,
σ
is of degree2or4inG
unlessd = 0
orσ
isafullt
-ube: ifd = 0
,σ
is thepoint
±H 0
andithasnofae,heneσ
hasdegree1;andifσ
is afullt
-ube,thenσ
isnotthefaetofanyotherube,andisthereforeofdegree1.
Everynodein
G
hasdegree2or4,exeptforthepoints±H 0
andforthefullt
-ubes.Wetransform
G
intoanew graphG ′
: wereplaeeah nodev
ofdegree4inG
by twonodes
v ′
andv ′′
,andwereplaethefouredges,byfournewedges: twolinkingv ′
andapairof oppositefullfaets,andtwolinking
v ′′
andtheotherpairofoppositefullfaets.In
G ′
,allvertiesareofdegree1or2,heneG ′
istheunionofdisjointpathsandyles.Notethattheantipodeofapathin
G ′
isalsoapathinG ′
. Nopathanhaveantipodalendpoints, otherwise it should be its own antipode and have a node oran edge whih is
antipodal to itself. Hene the number of endpoints is a multiple of four. Sine exatly
twoendpointsare
H 0
and−H 0
,therearetwieanoddnumberofendpointswhih arefullt
-ubes. Ifm > t
,half ofthemhaveapositivesign(ifm = t
,afullt
-ubehasnosign).A path of
G ′
starting inH 0
an not terminate in−H 0
, but in afullt
-ube. Thus wehaveanalgorithmwhih allowstondafull
t
-ube. It isillustratedinFigure 5.3.5 Splitting the neklae between Alie and Bob
We an easily dedue Theorem 1 from Theorem 3. Indeed let
m = t
, and letx :=
(x 1 , x 2 , . . . , x t+1 )
be avertex ofM
. We order thex i
in the inreasing order of their ab-solute value. When
|x i | = |x i ′ |
, weputx i
beforex i ′
ifi < i ′
. Thisgivesapermutationπ
suhthat
|x π(1) | ≤ |x π(2) | ≤ . . . ≤ |x π(t+1) | = 1.
Then, wedene
λ = (λ 1 , λ 2 , . . . , λ t ) : M → C t
asfollows: weuttheneklaeat pointsn|x i |
(atmostt
aredierentfrom 1). Byonvention,x π(0) := 0
; wedene:λ i (x) :=
+1
ifP t
j=0
sign(x ǫ(j) )φ i
n|x π(j) |, n|x π(j+1) |
> 0
−1
ifnot,for
i = 1, 2, . . . , t
, whereǫ(j)
is thesmallestintegerk
suh that|x k | ≥ |x π(j+1) |
, and withtheonventionsign
(0) := 0
.This formula anbe understood as follows:
λ i
equals+1
(resp.−1
) if the rst thiefgets stritlymore(resp. stritlyless) beads oftype
i
thantheseondone,whenweutatpoints
n|x j |
oftheneklae,andwhenthesignofx k
showsthethiefwhogetsthe(j + 1)
thsub-neklae,where
k
isthesmallestintegersuhthatn|x k |
isgreaterthanorequaltotheoordinateofanypointofthesub-neklae.
(+1, +1, +1)
(+1, −1, +1) (−1, −1, +1)
(+1, −1, −1) (−1, −1, −1)
(+1, +1, −1) (−1, +1, −1)
Figure 5: Illustration of the onstrutive proof of Theorem 3. The triples indiate the
orrespondinglabelontheubeimage oftheFigure 6.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
. . .. .. . .. .. . .
6
+
-
x 1
x 2
x 3
(−1, −1, −1)
k s ?
-
6 (−1, −1, +1)
(−1, +1, −1) (+1, +1, +1) (−1, +1, +1)
(+1, +1, −1) (+1, −1, −1)
(+1, −1, +1)
Figure6: An illustrationof
C m
form = 3
.Theonlyproblemwiththisdenitionisthat
λ
maynotbeantipodal: ifX t
j=0
sign
(x ǫ(j) )φ i
n|x π(j) |, n|x π(j+1) |
= 0,
then
λ i (x) = λ i (−x) = −1
. This anbe avoided by a slight generi perturbation of the valueofallthebeads: adivisionusingutsbetweenbeadsanthennotbefairwithrespetto anytypeofbead 2
.
λ
isaubialmap. Indeed,therstonditiondeningubialmapsistriviallysatisedandtheseondonefollowsfromthefattheimagesby
λ
oftwoadjaentvertiesofM
have at mostoneoordinatethatdiersevenintheasewhenthereisj
suhthatǫ(j)
diers.Theorem 3 showsthat there is a
t
-ubeσ
whose imagebyλ
ist
. Oneof thevertiesofthis
t
-ubeorrespondsto afairdivision. Indeed,for anyvertexx
of suh aσ
, moving toeahofthe
t
adjaentvertiesofx
hangesthevalueofeahofthet
omponentsofλ(x)
.Theonstrutiveproof given abovefor Theorem3 suitsto theneklae problem. The
ubialomplexisompletelynaturalinthisontext: a
k
-ube(k ≤ t
)orrespondspreiselytothehoieof
k
beads. Moreover,itisstraightforwardtohekifak
-ubeisfullintermsof beads. WehavethusaonstrutiveproofofTheorem1.
3.6 General disrete neklae splitting theorem
In[3℄,Alon,MoshkovitzandSafraprovethefollowingextensionofTheorem2(withows).
Supposethattheneklaehas
n
beads,eahofaertaintypei
,where1 ≤ i ≤ t
. Supposethere are
A i
beadsoftypei
,1 ≤ i ≤ t
,P t
i=1 A i = n
,whereA i
isnotneessarilyamultipleof
q
. Aq
-splittingof theneklaeisapartitionofitintoq
parts,eahonsistingofanitenumber of non-overlapping sub-neklaes of beads whose unionaptures either
⌊A i /q⌋
or⌈A i /q⌉
beads oftypei
,forevery1 ≤ i ≤ t
.Theorem 4 Everyneklaewith
A i
beads oftypei
,1 ≤ i ≤ t
,has aq
-splittingrequiringatmost
t(q − 1)
uts.For
q = 2
, theproofof thepreeding setion showssomethingmore. Theproofis stillvalid,evenifthe
A i
'sarenoteven. IfA i
isodd,the(t − 1)
-faetoft
whosei
thoordinateisequalto1(resp.
−1
). orrespondtodivisiongiving⌈A i /2⌉
beadsoftypei
toAlie(resp.Bob)and
⌊A i /2⌋
beadsoftypei
toBob(resp. Alie). Wehavethusthefollowingtheorem,also withaonstrutiveproof:
Theorem 5 Every neklae with
A i
beads of typei
,1 ≤ i ≤ t
, has a2
-splitting requiringatmost
t
uts. Moreover, for eahtypei
of beadssuhthatA i
isodd,we anhoosewhihthief gets
⌈A i /2⌉
beadsof typei
.NotethatthistheoremanalsobederivedfromtheproofofAlon,MoshkovitzandSafra.
4 Open questions
There arealotof questionsthatremainopen.
1. Isitpossibleto extendthediretproofsofSetion 2.1to otherases?
2
Ifonedoesnotwanttouseaperturbation,thereisanotherwayforavoiding0:hooseapartiularbead
ofeahtypeanddene
λ i ( x ) := +1
(resp.− 1
)iftherstthiefgetsstritlymore(resp. stritlyless)beadsoftype
i
thantheseondthief,oriftherstthiefandtheseondonegetthesamenumberofbeadsoftypei
andtherstonegets(resp. doesnotget)thepartiularbeadoftype
i
.
forinstaneinthepaperofAlon[2℄.
3. Whatistheomplexityofthefollowingproblem?
INPUT: A neklae, with
t
beads,qa i
beads of eah type (a i
is an integer), andq
thieves,
TASK: Findafairdivisionoftheneklaeusingnomorethan
t(q − 1)
uts,This questionis still open(for moreaboutproblems of this type,see Papadimitriou
[12℄). Theorrespondingoptimizationproblem,namelyndingtheminimumnumber
ofutsprovidingafair division,isNP-hard,evenif
q = 2
anda 1 = a 2 = . . . = a t = 1
(see[5℄,[11℄).
4. IsthereananalogueofTheorem 5foranynumber
q
ofthieves?Referenes
[1℄ N. Alon,Splittingneklaes,Advanes inMath.,63(1987),247-253.
[2℄ N.Alon,Non-onstrutiveproofsinombinatoris,Pro. oftheInternational Congress
of Mathematiians,Kyoto 1990,Japan,Springer Verlag,Tokyo, pp.1421-1429,1991.
[3℄ N. Alon, D. Moshkovitz and S. Safra, Algorithmi Constrution of Sets for
k
-Restritions,toappearinThe ACM Transations onAlgorithms.
[4℄ N. Alon and D. West, The Borsuk-Ulam theorem and bisetion of neklaes, Pro.
Amer.Math. So., 98(1986),623-628.
[5℄ P. S. Bonsma, T. Epping and W. Hohstättler, Complexity results on restrited in-
stanesofapaintshopproblemforwords,DisreteAppliedMathematis,to appear.
[6℄ R.EhrenborgandG.Hetyei,GeneralizationsofBaxter'stheoremandubialhomology,
J. CombinatorialTheory,SeriesA, 69(1995),233-287.
[7℄ T.Epping,W.HohstättlerandP.Oertel,Complexityresultsonapaintshopproblem,
DisreteApplied Mathematis,136(2004),217-226.
[8℄ K.Fan,Combinatorialpropertiesofertain simpliialandubialvertexmaps,Arh.
Math., 11(1960),368-377.
[9℄ C.H. GoldbergandD. West,Bisetionofirleolorings,SIAMJ.Algebrai Disrete
Methods,6(1985),93-106.
[10℄ J.Matou²ek,UsingtheBorsuk-Ulamtheorem,Springer-Verlag,Berlin-Heidelberg-New
York,2003.
[11℄ F.MeunierandA.Sebö,Paintshop,oddyleandsplittingneklae,submitted.
[12℄ C.Papadimitriou,Ontheomplexityoftheparityargumentandotherineientproofs
ofexistene,J. ComputerandSystem Sienes,48(1994),498-532.
[13℄ T. Presottand F.E. Su,A onstrutiveproofof KyFan'sgeneralizationof Tuker's
lemma,J. Combin. Theory,Series A,111(2005),257-265.
[14℄ F. W. Simmons and F. E. Su, Consensus-halving via theorems of Borsuk-Ulam and
Tuker,Mathematial soialsienes,45 (2003),15-25.
[15℄ G.M.Ziegler,GeneralizedKneseroloringtheoremswithombinatorialproofs,Inven-
tionesMathematiae,147(2002),671-691.