Dean Corbae, Maxwell B. Stinchcombe & Juraj Zeman: An Introduction To Mathematical Analysis For Economic Theory and Econometrics
Dean Corbae, Maxwell B. Stinchcombe & Juraj Zeman: An Introduction To Mathematical Analysis For Economic Theory and Econometrics
Dean Corbae, Maxwell B. Stinchcombe & Juraj Zeman: An Introduction To Mathematical Analysis For Economic Theory and Econometrics
,thestatementx0isidentiedwiththesetR
={x X:x0}.
+
Inthisway,weidentifyastatementwiththesetofelementsofXforwhichthe
statement is true. There are deep issues in logic and the foundations of mathematics
relatingtothequestionofwhetherornotallsetscanbeidentiedbyproperties.
Fortunately,theseissuesrarelyimpingeonthemathematicsthateconomistsneed.
Chapter2ismoreexplicitabouttheseissues.
We are very often interested in establishing the truth of statements of the formIf
A,thenB.Therearemanyequivalentwaysofwritingsuchastatement:AB,
AimpliesB,AonlyifB,AissufcientforB,orBisnecessaryforA.To
rememberthesufciencyandnecessity,itmayhelptosubvocalizethemasAis
sufcientlystrongtoguaranteeBandBisnecessarilytrueifAistrue.
Thelogicalrelationofimplicationisasubsetrelation.IfA={x X:A(x)}
andB={x X:B(x)},thenABisthesameasAB.
Example1.1.1 LetXbethesetofnumbers,A(x)thestatementx
2
<1,and
B(x)thestatement|x| 1.Now,AB.Intermsofsets,A={x X:A(x)}
isthesetofnumbersstrictlybetween1and+1,B={x X:B(x)} isthesetof
numbersgreaterthanorequalto1andlessthanorequalto+1,andAB.
The statements of interest can be quite complex to write out in their entirety. IfX
is the set of allocations in a modelEof an economy andA(x)is the statement xis a
WalrasianequilibriumallocatedfortheeconomyE,thenacompletespecication
of the statement takes a great deal of work. Presuming some familiarity with
generalequilibriummodels,weofferthefollowing.
Example1.1.2 LetXbethesetofallocationsinamodelEofaneconomy;let
A(x)bethestatementxisaWalrasianequilibriumallocation;andB(x)bethe
statementxisParetoefcientforE.Therstfundamentaltheoremofwelfare
economics is AB. In terms of the denition of subsets, this is expressed as,
EveryWalrasianequilibriumallocationisParetoefcient.
Inothercases,weareinterestedinthetruthofstatementsoftheformAifand
onlyifB,oftenwrittenAiffB.Equivalently,suchastatementcanbewritten:
ABandBA,whichisoftenshortenedtoAB.Otherfrequentlyused
formulationsare:AimpliesBandBimpliesA, Aisnecessaryandsufcient
forB,orAisequivalenttoB.IntermsofthecorrespondingsetsAandB,these
arealldifferentwaysofwritingA=B.
Example 1.1.3 LetXbethesetofnumbers,A(x)thestatement0x1,
and B(x)the statement x
2
x. From high school algebra, AB. In terms
ofsets,A={x X:A(x)} andB={x X:B(x)} areboththesetsofnumbers
greaterthanorequalto0andlessthanorequalto1.
1.2 StatementsandTheirTruthValues 3
1.2
4 Chapter1 Logic
ThestatementAorB,thatis,AB,istrue.Thisissensible:sinceAistrue,
itisthecasethatatleastoneofAandBistrue,correspondingtoxbeinginthe
unionofAandB.
Itisimportanttonotethatweusethewordorinitsnonexclusivesense.When
wedescribesomeoneastallorred-headed, wemeantoallowtallred-headed
people.Wedonotmeanorintheexclusivesensethatthepersoniseithertallor
red-headedbutnotboth.Oneseesthisbyconsideringthelastcolumnsinthetwo
tables,theoneswiththepatternsT T T F and. AorBistrueaslongas
atleastoneofAandBistrue,andwedonotexcludethepossibilitythatbothare
true.Theexclusiveorisdenedby(AB)((AB)),whichhasthetruth
patternFT T F.Intermsofsets,theexclusiveoris(AB)(AB)
c
,which
hasthecorrespondingmembershippattern.
1.2.b Implies/Equivalence as Subset/Equality
Two of the remaining four ways of constructing new statements are: AB,
whichmeansAimpliesBandAB,whichmeansAisequivalenttoB.In
termsofsets,theseareAB and A=B. These are statements about relations
betweensubsetsofX.
Indicatorfunctionsareaveryusefulwaytotalkabouttherelationsbetween
subsets.ForeachxXandAX,denetheindicatorofthesetAby
1 if xA,
1
A
(x):= (1.1)
0 if xA.
Remember, a proposition, A, is a statement about elements xXthat can be
eithertrueor false.When it is true, wewrite A(x).The corresponding set Ais
{xX:A(x)}. The indicator of Atakes on the value 1for exactly those xfor
whichAistrueandtakesonthevalue0forthosexforwhichAisfalse.
Indicatorfunctionsareorderedpointwise;thatis,1
A
1
B
when 1
A
(x)1
B
(x)
for every pointxin the setX. Saying 1
A
1
B
is the same as saying that AB.
ItiseasytogivesetsAandBthatsatisfyneitherABnorBA.Therefore,
unlikepairsofnumbersrands, forwhichitisalwaystruethateitherrsor
sr,pairsofindicatorfunctionsmaynotberankedby.
Example1.2.1 IfXisthethree-pointset{a, b, c},A={a, b},B={b, c},and
C={c},then1
A
1
X
,1
B
1
X
,1
C
1
B
,(1
A
1
B
),and(1
B
1
A
).
Proving statements of the form AB and AB is the essential part of
mathematical reasoning. For the rst, we take the truth of A as given and then
establish logically that the truth ofBfollows.Forthesecond,wetaketheadditional
stepoftakingthetruthofBasgivenandthenestablishlogicallythatthetruthof
Afollows.Intermsofsets,forprovingtherst,wetakeapoint,x,assumeonly
thatxA,andestablishthatthisimpliesthatxB,thusprovingthatAB. For
provingthesecond,wetaketheadditionalstepoftakingapoint,x,assumeonly
thatxB,andestablishthatthisimpliesthatxA.Hereisthetruthtablefor
and,bothforstatementsandforindicatorfunctions.
1.2 StatementsandTheirTruthValues 5
Table1.c
A B A B A B x A x B
T T T T x A x B
T F F F x A x B
F T T F x A x B
F F T T x A x B
Table1.d
1
A
(x)1
B
(x) 1
A
(x)=1
B
(x)
T T
F F
T F
T T
1.2.c The Empty Set and Vacuously True Statements
Wenowcometotheideaofsomethingthatisvacuouslytrue, andasubstantial
proportionofpeoplendthisideatrickyorannoying,orboth.Theideathatwe
areafteristhatstartingfromfalsepremises,onecanestablishanything.InTable
1.c,ifA isfalse,thenthestatementA B istrue,whetherB istrueorfalse.
A statement that is false for all x X corresponds to having an indicator
function with the property that for all x X, 1
A
(x)= 0. In terms of sets, the
notationforthisisA=,whereweread as the emptyset,thatis,thevacuous
set,theonethatcontainsnoelements.NomatterwhatthesetBis,ifA=,then
1
A
(x)1
B
(x)forallx X.
Denition 1.2.2 ThestatementA B isvacuouslytrueifA=.
This denition follows the convention that we use throughout: we showthe term
ortermsbeingdenedinboldfacetype.
Intermsofsets,thisistheobservationthatforallB, B,thatis,thatevery
element ofbelongs toB. What many people nd distasteful is that every element
of belongstoBsuggeststhatthereisanelementof,andsincethereisnosuch
element,thestatementfeelswrongtothem.Thereisnothingtobedoneexceptto
getoverthefeeling.
1.2.d Indicators and Ands/Ors/Nots
Indicatorfunctionscanalsobeusedtocaptureands,ors,andnots.Oftenthismakes
proofssimpler.
Thepointwiseminimumofapairofindicatorfunctions,1
A
and1
B
,iswritten
as1
A
1
B
,andisdenedby(1
A
1
B
)(x)=min{1
A
(x),1
B
(x)}. Now, 1
A
(x)
and 1
B
(x)are equal either to 0 or to 1. Since the minimum of 1and 1is 1, the
minimumof0and1is0,andtheminimumof0and0is0,1
AB
=1
A
1
B
.This
means that the indicator associated with the statement A B is 1
A
1
B
. By
checkingcases, wenotethatforall x X, (1
A
1
B
)(x)=1
A
(x)
.
1
B
(x). As a
result,1
A
1
B
isoftenwrittenas1
A
.
1
B
.
Inasimilarway,thepointwisemaximumofapairofindicatorfunctions,1
A
and
1
B
,iswrittenas1
A
1
B
anddenedby(1
A
1
B
)(x)=max{1
A
(x),1
B
(x)}.
Here,1
AB
=1
A
1
B
,andtheindicatorassociatedwiththestatementA B is
1
A
1
B
.Basicpropertiesofnumberssaythatforallx,(1
A
1
B
)(x)=1
A
(x)+
1
B
(x)1
A
(x)
.
1
B
(x), so 1
A
1
B
couldbedenedas1
A
+1
B
1
A
.
1
B
.
6 Chapter1 Logic
Forcomplements,wedene1tobetheindicatorofX,thatis,thefunction
thatisequalto1everywhereonX,andwenotethat1
A
c =11
A
.
1.3
Logical Quantiers
The last two of our seven ways to construct statements use the two quantiers,
, read as there exists, and , read as for all. More specically, (x
A)[B(x)] means there exists an x in the set Asuch that B(x) and (x
A)[B(x)] means for all x in the set A, B(x). Our discussion of indicator
functions has already used these quantiers; for example, 1
A
1
B
was dened
as (x X)[1
A
(x) 1
B
(x)]. We now formalize the ways in which we use the
quantiers.
Quantiers should be understood as statements about the relations between sets,
andheretheemptyset,,isagainuseful.Intermsofsets,(x A)[B(x)]isthe
statement(A B)=,while(x A)[B(x)]isthestatementAB.
Notation Alert 1.4.A Followingcommonusage,whenthesetAissupposed
tobeclearfromcontext,weoftenwrite(x)[B(x)]for(x A)[B(x)]. If Aisnot
infactclearfromcontext,weruntheriskofleavingtheintendedsetAundened.
Thetwocrucialpropertiesofquantiersarecontainedinthefollowing,which
givestherelationshipamongquantiers,negations,andcomplements.
Theorem1.4.1 ThereisnoxinAsuchthatB(x)iffforallxinA,itisnotthe
casethatB(x),
(x A)[B(x)](x A)[B(x)], (1.9)
anditisnotthecasethatforallxinAwehaveB(x)iffthereissomexinAfor
whichB(x)fails,
(x A)[B(x)](x A)[B(x)]. (1.10)
Proof. Intermsofsets,(1.9)is[A B=][AB
c
].Intermsofindicators,
letting0bethefunctionidenticallyequalto0,itis1
A
.
1
B
=0 iff 1
A
(11
B
).
10 Chapter1 Logic
Intermsofsets,(1.10)is[AB][A B
c
=].Intermsofindicators,
the left-hand side of (1.10) is [1
A
1
B
], which is true iff for some x in X,
1
A
(x) >1
B
(x). This happens iff for some x, 1
A
(x)=1and 1
B
(x)=0, that is,
iffforsomex, 1
A
(x)
.
(11
B
(x))=1,whichistheright-handsideof(1.10).
The second tautology in Theorem 1.4.1 is important since it illustrates the
conceptofacounterexample.Inparticular,(1.10)states:IfitisnottruethatB(x)
forallxinA,thentheremustexistacounterexample(i.e.,anxsatisfyingB(x)),
andviceversa.Counterexamplesareimportanttools,sinceknowingthatx A
andx Bfor hundreds and hundreds ofxsdoesnotprovethatAB,butasingle
counterexampleshowsthat[AB].
Often,onecanprotablyapplytherulesin(1.9)and(1.10)timeaftertime.The
followinganticipatesmaterialfromthetopicsofconvergenceandcontinuitythat
wecoverextensivelylater.
Example1.4.2 Asequenceofnumbersisalist(x
1
, x
2
, x
3
, . . .),onex
n
foreach
countingnumbern N={1,2,3, . . .},wherethe. . .indicateskeepgoingin
thisfashion.LetR
++
denotethesetofstrictlypositivenumbers;foranyN N,
letN
)[A(, x
n
)],
whichismoremuchconveniently,andjustasprecisely,writtenas
( > 0)(N N)(nN)[|x | <].
n
This captures the idea that the numbers in the sequence become and stay arbitrarily
smallaswemovefurtherandfurtheroutinthesequence.Averbalshorthandfor
this is that for all positive (no matter how small), |x | is smaller than for
n
largen.
Applying(1.9)and(1.10)repeatedlyshowsthatthefollowingareallequivalent
tothesequencenotconvergingto0:
( > 0)(N N)(nN)[|x | <],
n
( > 0)(N N)(nN)[|x | <],
n
( > 0)(N N)(nN)[|x | <],
n
( > 0)(N N)(nN)[|x | <], and
n
( > 0)(N N)(nN)[|x | ].
n
Thus,thestatementasequencefailstoconvergeto0isequivalenttoforsome
strictlypositive,forallN(nomatterhowlarge),thereisanevenlargernsuch
that|x | .
n
Oneshouldalsonotethatthecommutativeanddistributivelawswefoundwith
andinthemmaybreakdownwithquantiers.While
(x)[A(x) B(x)](x)[A(x)] (x)[B(x)], (1.11)
(x)[A(x) B(x)](x)[A(x)] (x)[B(x)]. (1.12)
1.5 TaxonomyofProofs 11
Example1.4.3 Toseewhy(1.12)cannotholdasanifandonlyif statement,
suppose xis the set of countries in the world, A(x)is the property that xhas
a gross domestic product strictly above average, and B(x)is the property that
x has a gross domestic product strictly below average. There will be at least
one country above the mean and at least one country below the mean. That is,
(x)[A(x)] (x)[B(x)]istrue,butclearlytherecannotbeacountrythatisboth
aboveandbelowthemean,(x)[A(x) B(x)].
Intermsofsets,(1.11)canberewrittenas[(A B)=][(A=) (B=
)].Thesetformof(1.12)is[A B=][(A=) (B=)].Hopefullythis
formulationmakesthereasonwedonothaveanifandonlyifrelationin(1.12)
evenclearer.
Wecanalsomakeincreasinglycomplexstatementsbyaddingmorevariables.
Forexample,statementsoftheformA(x, y)asxandybothvaryacrossX.One
canalwaysviewthisasastatementaboutapair(x, y)andchangeXtocontain
pairs,butthismaynotmitigatetheadditionalcomplexity.
Example1.4.4 WhenXisthesetofnumbersandA(x, y) statesthatythatis
largerthanx,wherexandyarenumbers,thestatement(x)(y)(x<y)says
foreveryxthereisaythatislargerthanx.Thestatement(y)(x)(x<y)
saysthereisaythatislargerthaneveryx.Theformerstatementistrue,but
thelatterisfalse.
1.5
Taxonomy of Proofs
Wenowdiscussbroadlythemethodologyofproofsyouwillfrequentlyencounter
in economics. The most intuitive is the direct proof in the form of AB,
discussedin(1.6).Theworkistollintheintermediatestepssothat AA
1
,
A
1
A
2
, and . . . A
n
B are all tautologies. In terms of sets, this involves
constructingnsetsA
1
, . . . , A
n
suchthatAA
1
. . .
A
n
B.
Notation Alert 1.5.A The. . .indicatesA
2
throughA
n1
intherstlist.
The
. . .
indicatesthesamesetsinthesecondlist,butwealsomeantoindicate
thatthesubsetrelationholdsforalltheintermediatepairs.
Insomecases,thesetsA
1
, . . . , A
n
arisefromsplittingBintocases.Ifwend
B
1
, B
2
suchthat[B
1
B
2
]BandcanshowthatA[B
1
B
2
], thenweare
done.
In other cases it may be simpler to split A into cases. That is, sometimes it
is easier to nd A
1
and A
2
for which A[A
1
and A
2
]and then to show that
[A
1
B] [A
2
B].
Anotherdirectmethodofproof, calledinduction, worksonlyforthenatural
numbersN={1,2,3, . . .}.Supposewewishtoshowthat(n N)A(n)istrue.
This is equivalent to proving A(1) (n N) (A(n)A(n+1)). This works
sinceA(1)istrueandA(1)A(2)andA(2)A(3)andsoon.InChapter2we
showwhyinductionworks.
Proofsbycontradictionarealsoknownasindirectproofs.Theymay,initially,
seem less natural than direct proofs. To help you on your way to becoming
uent in indirect proofs, we now give the exceedingly simple indirect proof of
12 Chapter1 Logic
the rst fundamental theorem of welfare economics. This is perhaps one of the
most important things you will learn in all of economics, and nding a direct
proof seems rather difcult. Again, we presume some familiarity with general
equilibriummodels.
Denition 1.5.1 An exchange economy model is a triple, E= (I,y
i
,
i
),
where I is a nite set (meant to represent the people in the economy), y
i
R
+
isiendowmentofthegoodsthatareavailableinthemodeloftheeconomy,and
i
isispreferencerelationoverhisorherownconsumption.
Therearetwothingstonotehere:rst,wedidnotsaywhatapreferencerelation
isandwedoit,indetail,inChapter2;andsecond,weassumedthatpreferences
aredenedonlyoverownconsumption,whichisaverystrongassumption,and
wediscussitfurther.
Denition 1.5.2 An allocation is a list of vectors, written (x
i
)
iI
, where
R
)
iI
suchthatallagentspreferx
i
tox
i
.
Denition1.5.4 ApricepisanonzerovectorinR
.Anallocation-pricepair
+
((x
i
)
iI
,p)isaWalrasianequilibriumforEifitisfeasible,andifx
i
ispreferred
byitox
i
,thenicannotaffordx
i
,thatis,
p
k
x
> (1.14)
i,k
p
k
y
i,k
.
k k
Theorem 1.5.5 (First Fundamental Theorem of Welfare Economics)
If((x
i
)
iI
,p)isaWalrasianequilibrium,then(x
i
)
iI
isParetoefcient.
Let A= {(x
i
)
iI
:(p)[((x
i
)
iI
,p)isaWalrasianequilibrium]}, and B=
{(x
i
)
iI
:(x
i
)
iI
isParetoefcient}. In terms of sets, Theorem 1.5.5 states that
AB.
Proof. Aisthestatement((x
i
)
iI
,p)isaWalrasianequilibriumandBisthe
statement(x
i
)
iI
isParetoefcient.AproofbycontradictionassumesA B,
and shows that this leads to a contradiction, C C. In words, suppose that
((x
i
)
iI
,p)isaWalrasianequilibriumbutthat(x
i
)
iI
isnotParetoefcient.We
havetoshowthatthisleadstoacontradiction.
BythedenitionofParetoefciency,failingtobeParetoefcientmeansthat
thereexistsafeasibleallocation,(x
i
)
iI
,thathasthepropertythatallagentsprefer
1.5 TaxonomyofProofs 13
x
i
tox
i
.BythedenitionofWalrasianequlibrium,wecansum(1.14)acrossall
individualstoobtain
> . (1.15)
i k i k
p
k
x
i,k
p
k
y
i,k
Rearrangingthesummationsin(1.15)gives
> equivalently p
k
x
i,k
p
k
y
i,k
,
k i k i
p
k
x
i,k
> p
k
y
i,k
. (1.16)
k i k i
Since (x
i
)
iI
is a feasible allocation, multiplying each term in (1.13) by the
nonnegativenumberp
k
andthensummingyields
p
k
x
i,k
p
k
p
k
y
i,k
. (1.17)
k i k i
Letrbethenumber
k
p
k i
y
i,k
andletsbethenumber
k
p
k i
x
i,k
.
Equation (1.16) is the statement, C, that s >r, whereas (1.17) is the statement
Cthatsr.Wehavederivedthecontradiction[C C],whichweknowtobe
false,fromthesupposition[A B].Fromthis,weconcludethat[AB].
Asonebecomesmoreaccustomedtothepatternsoflogicalarguments,details
oftheargumentsaresuppressed.Hereisashorthand,three-sentenceversionofthe
foregoingproof.
Proof. If(x
i
)
iI
isnotParetoefcient,(x
i
)
iI
feasibleandunanimouslypre-
ferred to (x
i
)
iI
. Summing (1.14) across individuals yields
k i
p
k
x
i,k
>
. Since (x )
iI
is feasible, summing (1.13) over goods, we have
i
p
k
y
i,k
i
k i
p
k
x
i,k
k i
p
k
y
i,k
.
Just as 7x
2
+9x < 3 is a shorter and clearer version of seven times the square
ofanumberplusninetimesthatnumberaddstoanumberlessthanthree, the
shortening of proofs is mostly meant to help. It can, however, feel like a diabolically
designedcode,onemeanttoobfuscateratherthanelucidate.
Somedecodinghints:
1. Looking at the statement of Theorem 1.5.5, we see that it ends in then
(x
i
)
iI
is Pareto efcient. Since the shortened proof starts with the sentence
If(x
i
)
iI
isnotParetoefcient,youshouldconcludethatweareoffering
a proof by contradiction. This means that you should be looking for a
conclusion that is always false. Reaching such a falsity completes the proof.
2. Despitewhatitsays,thesecondsentenceintheshortenedproofdoesmore
than sum (1.14); it rearranges the summation. Your job as a reader is to
14 Chapter1 Logic
look at (1.14) and see that it leads to what is claimed. If the requisite
rearrangement is tricky, then it should be given explicitly. Like beauty,
trickinessisintheeyeofthebeholder.
3. Thethirdsentenceprobablycompressestoomanysteps.Sometimes,this
willhappen.
ThroughoutmostofChapter2,wetrytobeexplicitaboutthestrategyofproof
being used. As we get further and further into the book, we shorten proofs more and
more.Hopefully,theearlypracticewithproofswillhelprenderourshortenings
transparent.
CHAPTER 2
Set Theory
Inthefoundationsofeconomictheory,oneworriesabouttheexistenceofoptima
forsingle-persondecisionproblemsandabouttheexistenceofsimultaneousop-
timaforlinked,multiple-person-decisionproblems.Thesimultaneousoptimaare
calledequilibria.Oftenmoreinterestingthanthestudyofexistencequestionsis
thestudyofthechangesintheseoptimaandequilibriaasaspectsoftheeconomic
environmentchange,whichiscalledcomparativestatics.Sinceachangeinone
persons behavior can result in a change in anothers optimal choices when the
problemsarelinked,thecomparativestaticsofequilibriawilltypicallybeamore
complicatedundertaking.
Theearlysectionsofthischaptercovernotation,productspaces,relations,and
functions. This is sufcient background for the foundational results in rational
choicetheory:conditionsonpreferencesthatguaranteetheexistenceofoptimal
choicesinnitecontexts; representationsoftheoptimalchoicesassolutionsto
utilitymaximizationproblems;andsomeelementarycomparativestaticsresults.
Anintroductiontoweakorders,partialorders,andlatticesprovidessufcient
backgroundforthebasicsofmonotonecomparativestaticsbasedonsupermodu-
larity.ItisalsosufcientbackgroundforTarskisxed-pointtheorem,therstof
thexedpointtheoremswecover.Fixed-pointtheoremsareoftenthetoolused
toshowtheexistenceofequilibria.Tarskistheoremalsogivesinformationuseful
forcomparativestatics,andweapplyittostudytheexistenceandpropertiesof
thesetofstablematchings.
Whether or not the universe is innite or nite but very large seems to be
unanswerable. However, the mathematics of innite sets often turns out to be much,
much easier than nite mathematics. Imagine trying to study planar geometry
under the simplifying assumption that the plane contains 293 million (or so) points.
Attheendofthischapterwedealwiththebasicresultsconcerninginnitesets,
resultsthatweuseextensivelyinourstudyofmodelsofprices, quantities, and
time,allofwhichbegininChapter3.
15
16 Chapter2 SetTheory
2.1
,asaconsumptionset),thesetofpreference
+
relations on the positive orthant, and the set of functions from price-income
pairstothepositiveorthant.Supposewewishtodiscussaresultoftheform,The
demandfunctionsofallsmoothpreferencerelationswithindifferencecurveslying
insidethestrictlypositiveorthantarethemselvessmooth.ThismeansthatXhas
toincludesubsetsofthepositiveorthant(e.g.,thestrictlypositiveorthant),subsets
ofthepreferencerelations,andsubsetsofthepossibledemandfunctions.
The smaller group of objects that we talk about is called a superstructure.
Superstructures are formally dened in 2.13 at the end of this chapter. The
essentialideaisthatonestartswithasetS.Wehavetostartwithsomekindof
primitive, weagreethatSisaset, andweagreethatnoneoftheelementsofS
containsanyelements.Wethenbeginaninductiveprocess,addingtoStheclass
ofallsubsetsofS,thentheclassofallsubsetsofeverythingwehavesofar,and
so on and so on. As we will see, this allows us to construct and work with all
ofthespacesoffunctions,probabilities,preferences,stochasticprocessmodels,
dynamicprogrammingproblems,equilibriummodels,andsoon,thatweneedto
studyeconomics.Italsokeepsussafelyoutoftroublebyavoidingsituationslike
Russells example and allows us to identify our restricted class of sets with the
propertiesthattheyhave.
2.2 NotationandOtherBasics 17
2.2
+
, then the Walrasian budget set, denoted B(p, w), is dened
by B(p, w) = {x X:A(x)}. With more detail about the statement A, this is
B(p, w) ={x R
:p
.
xw}.
+
Denition 2.2.2 For AandBsubsetsofX,wedene:
1. A B,theintersectionofAandB, by A B={x X: [x A] [x
B]},
2. A B,theunionofAandB, by A B={x X: [x A] [x B]},
3. AB,AisasubsetofB, or BcontainsA, if [x A][x B],
4. A=B,AisequaltoB, if [AB] [BA],
5. A=B,AisnotequaltoB, if [A=B],
6. AB,AisapropersubsetofB, if [AB] [A=B],
7. A\ B,thedifferencebetweenAandB, by A\ B={x A:x B},
8. AB,thesymmetricdifferencebetweenAandB, by AB=(A\ B)
(B\ A),
9. A
c
,thecomplementofA, by A
c
={x X:x A},
10. ,theemptyset, by =X
c
,and
11. AandBtobedisjointifA B=.
ThesedenitionscanbevisualizedusingVenndiagramsasinFigure2.2.2.
Example 2.2.3 If X= {1,2, . . . , 10}, the counting numbers between 1and
10,A={evennumbersinX},B={oddnumbersinX},C={powersof2inX},
andD={primesinX},thenA B=,A D={2},A\ C={6,10},CA,
B=C,C D={2,3,4,5,7,8},andBD={2,9}.
Thereisapurposetothenotationalchoicesmadeindeningusing
and dening using . Being in A B requires being in Abeing in
B, beinginA BrequiresbeinginA beinginB.Thedenitionsofunions
and intersections can easily be extended to arbitrary collections of sets. Let I
beanindexset,forexample,I=N={1,2,3, . . .} asinExample1.4.2(p.10),
andletA
i
,i IbesubsetsofX.Then
iI
A
i
={x X:(i I)[x A
i
]} and
iI
A
i
={x X:(i I)[x A
i
]}.
We have seen the following commutative, associative, and distributive prop-
erties before in Theorem 1.3.2 (p. 7), and they are easily checked using Venn
diagrams.
18 Chapter2 SetTheory
A B A B
A B A B
A
B
A
B
B A\B A
A B
A
A
c
A B
A B
A B 0
FIGURE 2.2.2
2.2 NotationandOtherBasics 19
A
B
C
A (B C) (A B) (A C)
A
B
C
A (B C) (A B) (A C)
FIGURE 2.2.4
Theorem 2.2.4 ForsetsA,B,andC,
1. AB=BA,AB=BA;
2. (AB)C=A(BC),(AB)C=A(BC);and
3. A(BC)=(AB)(AC),A(BC)=(AB)(AC).
Exercise 2.2.5 ProveTheorem2.2.4fromTheorem1.3.2.[SeeFigure2.2.4.
The proof amounts to applying the logical connectives and above denitions:
to show AB=BA, it is sufcient to note that xAB(xA)
(xB)(xB)(xA)xBA.]
The following properties are used extensively in probability theory and are
easilycheckedinaVenndiagram.[SeeFigure2.2.6.]
Theorem 2.2.6 (DeMorgans Laws) IfA,B,andCareanysets,then
1. A\(BC)=(A\B)(A\C),and
2. A\(BC)=(A\B)(A\C).
Inparticular,takingA=X,(BC)
c
=B
c
C
c
and(BC)
c
=B
c
C
c
.
Thelasttwoequalitiesarethecomplementofaunionistheintersectionof
the complements and the complement of an intersection is the union of the
complements. When we think of B and Cas statements, (BC)
c
is not B
orC,whichisequivalentto,neitherBnorC,whichisequivalentto,notB
andnotC,andthisisB
c
C
c
.Inthesameway,(BC)
c
isnotbothBandC,
whichisequivalenttoeithernotBornotC,andthisisB
c
C
c
.
Proof. For(1)weshowthatA\(BC)(A\B)(A\C),andA\(BC)
(A\B)(A\C).
()SupposexA\(BC).ThenxAandx(BC).ThusxAand
(xB and xC). This implies xA\B and xA\C. But this is just x
(A\B)(A\C).
() Supposex(A\B)(A\C). Thenx(A\B)andx(A\C). ThusxA
and (xB and xC). This implies xAand x(BC). But this is just
xA\(BC).
Exercise 2.2.7 FinishtheproofofTheorem2.2.6.
20 Chapter2 SetTheory
A
B
C
A\(B C) (A\B) (A\C)
A
B
C
A\(B C) (A\B) (A\C)
FIGURE 2.2.6
Denition2.2.8 For AasubsetofX,thepowersetofA,denotedP(A),isthe
setofallsubsetsofA. A collectionorclassofsetsisasubsetofP(A),thatis,a
setofsets.Afamilyisasetofcollections.
Example 2.2.9 Let X= {a, b, c}. If A= {a, b}, B= {b}, C= {b, c}, then
P(X), C= {A}, D= {A, B}, and E= {A, C,} are collections, whereas F=
{D,E} isafamily.
Toagreatextent,thedistinctionamongsets,collections,andfamiliesdepends
onwhereonestartstheanalysis.Forexample,wedenefunctionsassetsofpairs
(x, f (x)).Weareofteninterestedinthepropertiesofdifferentsetsoffunctions.
Ifthepossiblepairsarethepoints,thenasetoffunctionsisafamily.However,
ifXisthesetofallfunctions,thenasetoffunctionsisjustthat,aset.Wehavethe
set/collection/familyhierarchyinplaceforcasesinwhichwehavetodistinguish
amongseverallevelsinthesamecontext.Thefollowinganticipatesmaterialon
probabilitytheory,whereweassignprobabilitiestoeverysetinaeldofsets.
Example 2.2.10 A eld is a collection, F
, [A
F
][A
c
F
] and [A, B F
][(A B F
) (A B F
(E)denotethefamilyofalleldscontainingE,thatis,
F
(E)={F
:EF
,F
aeld}.TheeldgeneratedbyEisdenedasF
(E)=
{F
:F
(E)}.Thisisasensibledenitionbecausetheintersectionofany
familyofeldsgivesanothereld.InExample2.2.9,F
(D)=F
(E)=P(X).
2.3 Products,Relations,Correspondences,andFunctions 21
Thefollowingaresomeofthemostimportantsetsweencounterinthisbook:
N={1, 2, 3, . . .},thenaturalorcountingnumbers.
Z={. . . , 2, 1, 0, 1, 2, . . .}, theintegers.
Z
+
={0, 1, 2, . . .},thenonnegativeintegers.
Q={
m
:m, n Z, n =0},thequotients,orrationalnumbers.
n
R,thesetofrealnumbers,thatweconstructinChapter3byaddingthe
so-calledirrationalnumberstoQ.
Note thatQcontainsallofthenite-lengthdecimals,forexample,7.96518=
m
n
form=796,518andn =100,000.ThismeansthatQcontainsarepresentation
foreveryphysicalmeasurementthatwecanmakeandeverynumberwewillever
see from a computer. The reason for introducing the extra numbers in R is not
oneofrealism.Rather,weshallseethatQhasholesinit,andeventhoughthe
holesareinnitelysmall,theymakeanalyzingsomekindsofproblemsmiserably
difcult.
Even though we have not yet formally developed the set of numbers R, the
followingexampleisworthseeingearlyandoften.
Example 2.2.11 (The Field of Half-Closed Intervals) Let X =R and
for a, b X, a < b, dene (a, b]= {x X :a < x b}. Set E= {(a, b] : a <
b, a, b R}and let X= F
andn
n iffn =n
.
2.3
0 1 2 3 4
=
0 1 2 3 4
<
0 1 2 3 4
Notethat,<,=,and= aresets.Intermsofthesesets, istheunionofthe
disjointsets,<and=,andthecomplementof= is=.
Relationscanalsobeusedinallkindsofcuteways.
Example 2.3.7 Let A= {Austin, Des Moines, Harrisburg} and B= {Texas,
Iowa,Pennsylvania}.ThentherelationR={(Austin,Texas),(DesMoines,Iowa),
(Harrisburg,Pennsylvania)} expressesisthestatecapitalof.
ArelationbetweenAandBisasubsetofAB.AfunctionfromAtoBis
aspecialkindofrelation,andacorrespondencefromAtoBisawaytoviewa
relationasafunction;thatis,itisanalternatedenitionofarelation.
24 Chapter2 SetTheory
Denition2.3.8 Afunction(ormapping)f ,denotedf:AB,isarelation
betweenAandB(i.e.,fAB)satisfyingthefollowingtwoproperties:
1. forallaA,thereexistsbBsuchthat(a, b)f ,and
2. if(a, b)fand(a, b
)f ,thenb=b
.
ForeachaA,theuniquebsuchthat(a, b)fisdenotedf (a).Afunction
may be written as af (a) (read amaps to f (a)). The set Ais called the
domainof f , sometimes denoted D(f ). The rangeof f , denoted Range(f )or
f (A), is {bB:aAsuch that(a, b)f }. ThegraphoffisGr(f )={(a, b):
(a, b)f }.
Verbally, for eachainA,fassociates a uniqueb,denotedb=f (a). Afunction
f anditsgraphareoneandthesame.Itisoddtodistinguishverballybetweena
functionanditsgraph,butwe(daringly)doitanyway.
3
Example2.3.9 ForA=B={0,1,2,3,4},thefunctionsf (x) =xandg(x)=
4xcanberepresentedby
4 4
3 3
2 2
1 1
0 0
0 1 2 3 4 0 1 2 3 4
f (x) =x g(x)=4x
Probabilitiesareanimportantexampleoffunctions.
Example 2.3.10 [Example 2.2.10 (p. 20)] A probability is a function, P:
F
,m, m
).
Proof. By the denition of cardinality, we know that n is the smallest nat-
ural number with the property that there exists an f:{1, . . . , n} X with
f ({1, . . . , n}) =X.Supposethatforsomem=m
).
Consider the function g :{1, . . . , n 1} Xdened by g(k) = f (k) for k
{1, . . . , m1} andbyg(k) =f (k +1) fork {m, . . . , n 1}.Sinceg({1, . . . ,
n 1}) =X, n wasnotthecardinalityof X, acontradictionthatcompletesthe
proof.
2.4
Equivalence Relations
As the name suggests, equivalence relations are relations of a special kinda
kindthatappearsfrequentlyinthemathematicsthateconomistsuse.Thefamiliar
equivalence classes from intermediate microeconomics are indifference curves,
setsofconsumptionbundlesthatareallindifferentfortheconsumer,andisoprot
lines, setsofinput-outputvectorsthatyieldthesameprotfortheproducer.In
gametheory,oneseesthestrategicequivalenceofstrategiesandtheequivalence
ofgames.
Denition 2.4.1 AnequivalencerelationonasetA isarelation thatis:
1. reexive,a A,(a, a) ,
2. symmetric,a, b A,[(a, b) ][(b, a) ],and
3. transitive,foralla, b, c A,[[(a, b) ][(b, c) ]][(a, c) ].
This is perhaps more intuitive with the aRbnotation: (A A) is an
equivalence relation iff for all a, b, c A, a a, [a b][b a], and [a
b b c][a c].
Example2.4.2 Equality is an equivalence relation onR. If u :XRis a utility
functionrepresentingpreferencesonasetX,thendeningx y byu(x) =u(y)
givestheindifferenceequivalencerelation.
4.Thisisafancywayofsayingthatourdenitionmakessensethatifasetisnite,thenthe
cardinalityofthatsetexists.
2.4 EquivalenceRelations 27
Example2.4.3 Denethecongruencemodulo4relationM
4
onZbyx, y
Z,xM
4
y ifremaindersobtainedbydividingx andy by4areequal.Forexample,
13M
4
65becausedividing13and65by4givesaremainderof1.
Exercise 2.4.4 Showthatcongruencemodulo4isanequivalencerelation.
Denition 2.4.5 Givenanequivalencerelation onasetA andanelement
x A,wedenetheequivalenceclassdeterminedbyx byE
x
={y A :y x}.
Notethatx E sincex x.
x
Example 2.4.6 The equivalence classes of Z for the relation M
4
are deter-
minedbyx {0, 1, 2, 3}, whereE ={z Z:k Z, z =4k +x}, thatis, xis
x
theremainderwhenz isdividedby4.
Equivalenceclasseshavethefollowingproperty.
Theorem2.4.7 TwoequivalenceclassesE andE
areeitherdisjointorequal.
Proof. LetE ={y A :y x} andE
={y A :y x
}. If E E
=,then
E andE
aredisjoint.Ifz E E
, weshowthatE =E
.Therststepisto
demonstratethatE E
,andthesecondstepistoshowthatE
E.
Letw E.Wemustshowthatw E
.Sincew E,w x. As z E E
, we
knowthatz x andz x
.Bytransitivityw z;hencew x
,sothatw E
.
ReversingtherolesofE andE
inthisargumentdemonstratesthatE
E.
LookingatExample2.4.6inlightofTheorem2.4.7,weseethatiftwoelements
areinrelation,theyhavethesameequivalenceclass.SoE
1
=E
5
=E
9
=. . . and
E
2
=E
6
=E
10
=. . . . Moregenerally,foralln, k Z,E
n
=E
4k+n
.
Notation 2.4.8 A/ denotesthecollectionofall-equivalenceclasses.
Mnemonically,dividesA intoacollectionofdisjointsets,sowewriteA/.
Theunionofallthesetsin A/ equalsallofA becauseeveryelementa ofA
belongstoexactlyoneoftheequivalenceclasses.AnotherwaytounderstandA/
isasapartitionofA.
Denition 2.4.9 A partition of a set A is a collection of nonempty disjoint
subsetsofA whoseunionisallofA.
WesawinTheorem2.4.7thatequivalencerelationsgiverisetopartitions.The
reverseisalsotrue.
Exercise2.4.10 Forapartition,C, of A,dene
C
byx
C
y iffx andy belong
tosameelementofC.Showthat
C
isanequivalencerelation.
Example 2.4.11 Theequivalenceclassesof ZinExample2.4.6constitutea
partition;E
0
={. . . , 8, 4, 0, 4, 8, . . .},E
1
={. . . , 7, 3, 1, 5, . . .},E
2
=
{. . . , 6, 2, 2, 6, . . .},andE
3
={. . . , 5, 1, 3, 7, . . .} aredisjointandtheir
unionisallofZ.Generally,Zcanbepartitionedton subsetsviatheequivalence
relationx y iffx, y have the same remainder after division byn. The partitioning
setscontainthosesubsetshavingremainders0, 1, . . . , n 1ton.Anothersimple
exampleisacointossexperimentwherethesamplespaceS ={Heads,Tails} has
mutuallyexclusiveevents(i.e.,Heads Tails=).
28 Chapter2 SetTheory
Example 2.4.12 Considertherelation onRgivenbyxyiffxy Z.
Itcanbeeasilycheckedthatthisisanequivalencerelation.Theequivalenceof
anarbitraryx Rlookslikex+Z={x+n:n Z}.Foralln Z,xandx+n
areinthesameequivalenceclass.Sinceforeachx,thereexistsann Zsuchthat
n x <n +1,xis in the same equivalence class asxn, which we denote by(x),
wherexn [0,1).Thusforeachx,(x)isarepresentationoftheequivalence
classofx.Notethatifx, y [0,1),thenxy Z, so xy.
WhatdoesthequotientspaceR/ lookslike?Thisspaceconsistsofequiva-
lenceclassesof.Bytheaboveargument,wecanmakeeachmemberof[0,1)
correspond to exactly one equivalence class of . That is, we can think of
[0,1)asR/.
Chapter3developstherealnumbers,R,asacollectionofequivalenceclasses
ofsequencesofelementsofQ.
2.5
on
Xshouldhaveinordertorepresentpreferencesthatarerational.Rememberthat
wewritexRyfor(x, y) R.
Denition2.5.1 ArelationRonXiscompleteif for allx, y X,xRyoryRx;
itistransitiveifforallx, y, z X,[[xRy] [yRz]][xRz];anditisrational
ifitisbothcompleteandtransitive.
2.5 OptimalChoiceforFiniteSets 29
Example2.5.2 Oneofthecrucialorderpropertiesofthesetofnumbers,R, is
that and arecompleteandtransitive.
Completeness neither implies nor is implied by transitivity. To see this, the
followingexercisegivesanexampleofarelationthatsatisesbothcompleteness
andtransitivity,givesotherrelationsthatsatisfyoneoftheconditionsbutnotthe
other, and gives a relation that satises neither. When you see a new concept,
youshoulddevelopthetwohabitsthatthisexerciseexemplies:ndingexamples
in which the new concept does and does not hold and nding examples that
demonstratehowthenewconceptinteractswithother,possiblyrelatedconcepts.
Exercise2.5.3 InExample2.3.6(p.23),showthatis complete and transitive,
that< and= aretransitivebutnotcomplete,andthat= isneithertransitivenor
complete.Checkthattherelation
givenlaterinExample2.5.6iscompletebut
nottransitive.
In thinking about preference relations, completeness is the requirement that
anypairofchoicescanbecomparedforthepurposesofmakingachoice.Given
howmucheffortitistomakelifedecisions(jobs,marriage,kids),completeness
is astrongrequirement.When arelation is not complete, there arechoices that
cannotbecomparedandtheremaybetwoormoreoptimalchoicesintheset.For
example, consider the relationon the set of all subsets ofA={1, . . . , 10} except
A itself.Supposewearelookingforthelargestsubset.Theneachofthesubsets
withnineelementsisalargestelementandtheycannotbecomparedwitheach
other. Transitivity is another rationality requirement. If violated, vicious cycles
could arise among three or more optionsany choice would have another that
strictlybeatsit.Tosaystrictlybeatsweneedthefollowing.
Denition 2.5.4 Givenarelation
,denex y by[x
y] [y
x]and
x y by[x
y] [y
x].
When talking about preference relations, x y is read as x is strictly
preferredtoyandx yisreadasx isindifferenttoy.Fromthedenitions,
youcanshowthat[x
isrational.
Example 2.5.6 Supposeyouareatarestaurantandyouhaveachoiceamong
fourmeals, pork, beef, chicken, orsh, allcostingthesame.Supposethatyour
preferences,
,andstrictpreferences,,aregivenby
pork pork
beef beef
sh sh
chic chic
chic sh beef pork chic sh beef pork
Thebasicbehavioralassumptionineconomicsisthatyouchoosetheoption
thatyoulikebest.Herep b fc p.Supposeyoutrytondyourfavorite
30 Chapter2 SetTheory
meal.Startbythinkingabout(say)c,discoveryoulikefbettersoyouswitchyour
decisiontof,butyoulikebbetter,soyouswitchagain,butyoulikepbetterso
youswitchagain,butyoulikecbettersoyouswitchagain,comingbacktowhere
youstarted.Youbecomeconfusedandstarvetodeathbeforeyoumakeupyour
mind.
Exercise2.5.7 Givethegraphicalrepresentation
,,andforthecomplete
transitivepreferencessatisfyingcfbp.
Exercise 2.5.8 Give the relation associated with the preferences given in
Example2.5.6.Is anequivalencerelation?Canitreasonablybeinterpretedas
indifference?
2.5.b Representing Preferences
Denition 2.5.9 u:XR represents
A utility function
if [x y]
[u(x) >u(y)]and[xy][u(x)=u(y)].
Sinceuisafunction,itassignsanumericalvaluetoeverypointinX.Since
we can compare any pair of numbers using , any preference represented by a
utilityfunctioniscomplete.As istransitiveonR,anypreferencerepresented
byautilityfunctionistransitive.
Exercise 2.5.10 Showthaturepresents
y][u(x)u(y)].
iff[x
SinceXisnite,wecanreplaceRbyNorbysomeset{1, . . . , n} inthisresult.
is rational. We must showthat there exists a utility function Proof. Suppose that
u:XNthatrepresents
y};thisisthesetofoptions
thatareweaklyworsethanx.Acandidateutilityfunctionisu(x)=#W(x). By
transitivity, [x
iscomplete
.Wemustshowthat
andtransitive.Forx, y X,eitheru(x)u(y)oru(y)u(x)(orboth).Bythe
denitionofrepresenting,x
yory
x.Supposenowthatx, y, z X,x
y,
andy
z.Wemustshowthatx
z.Weknowthatu(x)u(y)andu(y)u(z).
Thisimpliesthatu(x)u(z),sothatx
z.
ThemappingxW(x) intheproofisyetanotherexampleofacorrespon-
dence,inthiscasefromXtoX.Wenowdenethemaincorrespondenceusedin
rationalchoicetheory.
Denition 2.5.12 AchoiceruleisafunctionC:P(X)P(X),equivalently
a correspondence from P(X)to X, such that C(B)Bfor all B P(X), and
C(B)= ifB=.
TheinterpretationisthatC(B)isthesetofoptionsthatmightbechosenfrom
themenuBofoptions.Thebest-knownclassofchoicerulesismadeupofthose
2.5 OptimalChoiceforFiniteSets 31
oftheformC
(B)=C
(B,
x B:y B,
)={ x
y}.InlightofTheorem
2.5.11,C
(B)isthesetofutility
maximizingelementsofB.
The set of maximizers, the argmax, is a sufciently important construct in
economicsthatithasitsownnotation.
Denition 2.5.13 For a nonempty set X and function f :XR,
argmax
xX
f (x) istheset{x
X:(x X)[f (x
)f (x)]}.
Thebasicexistenceresulttellsusthatthepreference-maximizingchoicerule
yieldsanonemptysetofchoices.
is a rational
preferencerelationonX,thenC
(B)=.
Theorem 2.5.14 If B is a nonempty nite subset of Xand
Proof. DeneS
=C
(B)(andyou
xB
{y B:y
x}.ItisclearthatS
shouldcheckbothdirectionsoftheinclusionifyouarenotusedtowritingproofs).
AllthatisleftistoshowthatS
=.
Letn
B
=#Band pick a functionf:{1, . . . , n
B
} Bsuch thatB=f ({1, . . . ,
n
B
}). This means we order (or count) members of Bas f (1), . . . , f (n
B
). For
m {1, . . . , n
B
},letS
(m)={y B:nm, y
f (n)},sothatS
=S
(n
B
).
Inotherwords,S
,weinductivelypickupthelargestelement
.Nowusingafunctionf
amongf (1), . . . , f (n)foralln.Denef
(1)=f (1).Giventhatf
(m1)has
beendened,dene
f (m) iff (m)
(m1),
f
f (m)= (2.1)
f
(m1) iff
(m1)f (m).
Foreachm {1, . . . , n
B
},S
(m)= becauseitcontainsf
(m),andbytransi-
tivity,f
(n
b
) S
.
TheideaoftheproofwassimplytolabelthemembersofthenitesetBand
checkitsmembersstepbystep.Wesimplyformalizedthisideausinglogicaltools
andthedenitionofniteness.
ForR, SX,wewriteR
Sifx
isarational
preferencerelationonX,then
1. [x, y C
(A)][xy],optimaareindifferent,
2. C
(B)
(A),largersetsareatleastweaklybetter,and
3. [C
(B) C
(A)=][C
(B)C
(A)],alargersetisstrictlybetterif
ithasadisjointsetofoptima.
Proof. The proof of (1) combines two proof strategies: contradiction and splitting
into cases. Suppose that [x, y C
(A), a
contradiction.Ifyx,thenx C
(A),acontradiction.
32 Chapter2 SetTheory
Toprove(2),wemustshowthat[[x C
(B)] [y C
(A)]][x
y].We
againgiveaproofbycontradiction.Supposethat[x C
(B)] [y C
(A)] but
[x
iscomplete, [x
y].Since
(A).
Inwhatisbecomingapattern,wealsoprove(3)bycontradiction.Supposethat
[C
(B) C
(A)= ] but [C
(B) C
[C
(B) C
(A)
,
and x C
(B)such that y
(B,
), which
contradicts[C
(B) C
(A)=].
2.5.c Revealed Preference
We now approach the choice problem starting with a choice rule rather than
with a preference relation. The question is whether there is anything new or
different when we proceed in this direction. The answer is No, provided the choice
rule satises a minimal consistency requirement, and satisfying this minimal
consistencyrequirementrevealsapreferencerelation.
A choice rule C denes a relation,
y] is (B
P(X))[[x,y B] [x C(B)]], equivalently (B P(X))[[x C(B)]
[y B]]. In words, x is revealed preferred to y if there is a choice situation,
B,inwhichbothxandyareavailable,andxbelongstothechoiceset.
Fromthe relation
, as in Denition
2.5.4(p.29).Itisbothausefulexerciseinmanipulatinglogicandagoodwayto
understand a piece of choice theory to explicitly write out two versions of the
meaningofx
y:
(B
x
P(X))[[x, y B
x
] [x C(B
x
)]]
(B P(X))[[y C(B)][x B]], (2.2)
equivalently
(B
x
P(X))[[x, y B
x
] [x C(B
x
)] [y C(B
x
)]]
(B P(X))[[y C(B)][x B]].
Inwords,thelatterofthesesaysthatthereisachoicesituationwherexandyare
bothavailable,xischosenbutyisnot,andifyiseverchosen,thenweknowthat
xwasnotavailable.
A set B P(X)reveals a strict preference of y over x, written y
B
x, if
x, y Bandy C(B)butx C(B).
Denition 2.5.16 Achoicerulesatisestheweakaxiomofrevealedprefer-
enceif[x
y](B)[y
B
x].
This is the minimal consistency requirement. Satisfying this requirement means
thatchoosingxwhenyisavailableinonesituationisnotconsistentwithchoosing
ybutnotxinsomeothersituationwheretheyarebothavailable.
2.6 DirectandInverseImages,Compositions 33
Theorem 2.5.17 If C is a choice rule satisfying the weak axiom, then
(B,
is rational, then
). If
)satisestheweakaxiom,and
. C
(B,
Proof. SupposethatCisachoicerulesatisfyingtheweakaxiom.
Wemustrstshowthat
iscompleteandtransitive.
C({x, y})=,sothatx
yory
x.
Transitivity:Supposethatx
yandy
z.Wemustshowthatx
z. To
dothis,itissufcienttodemonstratethatx C({x, y, z}).SinceC({x, y, z})isa
nonemptysubsetof{x, y, z},weknowthattherearethreecases:x C({x, y, z}),
y C({x, y, z}), andz C({x, y, z}). We must showthat each of these cases leads
totheconclusionthatx C({x, y, z}).
Case1:Thisoneisclear.
Case2:y C({x, y, z}),theweakaxiom,andx
yimpliesthatx C({x,
y, z}).
Case3:z C({x, y, z}),theweakaxiom,andy
zimpliesthaty C({x,
(B,
).PickanarbitraryB
P(X).ItissufcienttoestablishthatC(B)C
(B,
) andC
(B,
)C(B).
Pickanarbitraryx C(B).Bythedenitionof
,forally B,x
y. By
thedenitionofC
(
.
,
.
),thismeansthatx C
(B,
).
Nowpickanarbitraryx C
(B,
).BythedenitionofC
(
.
,
.
),thismeans
thatx
yforally B.Bythedenitionof
,foreachy B,thereisasetB
y
suchthatx, y B
y
andx C(B
y
). As Csatisestheweakaxiom,forally B,
thereisnosetB
y
withthepropertythaty
B
x.SinceC(B)=, if x C(B),
y
thenwewouldhavey
B
xforsomey B,acontradiction.
Exercise 2.5.18 What is left to be proved in Theorem 2.5.17? Provide the
missingstep(s).
ItisimportanttonotethereachandthelimitationofTheorem2.5.17.
Reach:First,wedidnotuseXbeingniteatanypointintheproof,soitapplies
toinnitesets.Second,theproofwouldgothroughsolongasCisdenedonall
two- andthree-pointsets.ThismeansthatwecanreplaceP(X)withafamilyof
setsBthroughout,providedBcontainsalltwo- andthree-pointsets.
Limitation:Inmanyoftheeconomicsituationsofinterest,thetwo- andthree-
pointsetsarenottheonesthatpeoplearechoosingfrom.Forexample,theleading
casehasBastheclassofWalrasianbudgetsets.
2.6
= x
3
.
Clearlyf (R) iscontainedinthepositiverealnumbers.Foreveryr 0,f ( r) =
f ( r) =r, whichimpliesthatf (R) =f (R
+
) =R
+
.Alsog(
3
r) =rshows
thatallrealnumbersappearintherangeofg,thatis,g(R) =R.
Theorem 2.6.4 LetfbeafunctionfromA toBandletE, FA:
1. IfE F,thenf (E) f (F),
2. f (E F) f (E) f (F),
3. f (E F) =f (E) f (F),
4. f (E\F) f (E),and
5. f (EF) f (E)f (F).
Exercise 2.6.5 ProveTheorem2.6.4andgiveexamplesinwhichsubsetrela-
tionsareproper.
2.6 DirectandInverseImages,Compositions 35
f
1
(H)
f
H
[
]
FIGURE 2.6.10
Exercise2.6.6 Find and prove the analogue of Theorem2.6.4 when the function
fisreplacedwithacorrespondenceG,givingexampleswiththesubsetrelations
beingproper.
2.6.b Inverse Relations
Inverserelationssimplyreversetheorderinwhichweconsidertheaxes.
Denition 2.6.7 GivenarelationRbetweenAandB,theinverseofRisthe
relationR
1
betweenBandAdenedbyR
1
={(b, a):(a, b) R}.Imagesof
setsunderR
1
arecalledinverseimages.
The inverse of a function need not be a function, though it will always be a
correspondence.
Example2.6.8 Ingeneral,functionsaremany-to-one.Forexamplef (x) =x
2
fromRtoRmapsboth+ rand rtorwhenr0.Inthiscase,therelation
f
1
,viewedasacorrespondencemapseverynonnegativerto{ r,+ r},and
mapseverynegativerto.
Example2.6.9 LetWbeaniteset(ofworkers)andFaniteset(ofrms).A
functionmappingWtoF Wis amatchingif for allw,[(w) F][(w)=
w].Weinterpret(w)=wastheworkerwbeingself-employedorunemployed.
For f F,
1
(f )Wisthesetofpeoplewhoworkatrmf.
Itiswellworththeefforttobeevenmorespecicforfunctions.
Denition 2.6.10 IffisafunctionfromAtoBandHB,thentheinverse
image of H under f, denoted f
1
(H), is the subset {a A|f (a) H}. [See
Figure2.6.10.]
WhenH={b} isaone-pointset,wewritef
1
(b)insteadoff
1
({b}).
Exercise2.6.11 Just to be sure that the notation is clear, prove the following and
illustratetheresultswithpictures:f
1
(H)=
bH
f
1
(b),proj
B
1
(H)=AH,
andproj
1
(E)=EB.
A
36 Chapter2 SetTheory
Exercise 2.6.12 (Level Sets of Functions) Letf:ABbeafunction.
Denea
f
a
ifbBsuchthata, a
f
1
(b).Theseequivalenceclassesare
calledlevelsetsofthefunctionf.
1. Showthat
f
isanequivalencerelationonA.
2. Showthat[a
f
a
][f (a) =f (a
)].
3. Give an example with f,g being different functions from Ato B but
f
=
g
.
4. Provethattheinverseimagesf
1
(b)andf
1
(b
)aredisjointwhenb=b
.
[Thismeansthatindifferencecurvesneverintersect.]
Wereturntoinverseimagesundercorrespondenceslater.Sincetherearetwo
waystoviewthem, asrelationsfromAtoBandasfunctionsfromAtoP(B),
therearetwoimmediatepossibilitiesforthedenitionofGinverse.Itturnsout
thatthereisalsoathirdpossibility.
Inverse images under functions preserve the set operations, unions, inter-
sections, and differences. As seen in Exercise 2.6.5, images need not have this
property.
Theorem 2.6.13 LetfbeafunctionmappingAtoB,andletG, HB:
1. ifGH,thenf
1
(G)f
1
(H),
2. f
1
(GH) =f
1
(G)f
1
(H),
3. f
1
(GH) =f
1
(G)f
1
(H),and
4. f
1
(G\H) =f
1
(G)\f
1
(H).
Proof. (1)Ifaf
1
(G),thenf (a) GHsoaf
1
(H).
Exercise 2.6.14 FinishtheproofofTheorem2.6.13.
2.6.c Injections, Surjections, and Bijections
Ingeneral,functionsaremany-to-one.InExample2.6.8(p.35),f (x) =x
2
maps
both+ rand rtor.InExample2.6.9(p.35),manyworkersmaybematched
toasinglerm.Whenfunctionsarenotmany-to-onebutone-to-one, theyhave
niceadditionalpropertiesthatowfromtheirinversesalmostbeingfunctions.
Denition2.6.15 f:ABisone-to-oneor aninjectionif[f (a) =f (a
)]
[a=a
].
Since [a=a
][f (a) =f (a
)][a=a
].
Recall the correspondence G(b)=f
1
(b)from B to Aintroduced earlier.
Whenfismany-to-one,thenforsomeb,thecorrespondenceG(b)containsmore
thanonepoint.Whenacorrespondencealwayscontainsexactlyonepoint,itisa
function.Hence,theinverseofaone-to-onefunctionisafunctionfromtherange
offtoA.Thatis,theinverseofaninjectionf:ABfailstobeafunctionfrom
BtoAonlyinthatitmaynotbedenedforallofB.
2.6 DirectandInverseImages,Compositions 37
f(E)
0
1
2
1 1/2 a
F
1
(H)
f(a) 2a
E
H
FIGURE 2.6.18 f:RRgivenbyf (a) =2a.
Example2.6.16 LetE={2,4,6, . . .} bethesetofevennaturalnumbersand
denef (n)=2n.Thenfisone-to-one,f (N)=E,andf
1
isafunctionfromE
toN.
Denition2.6.17 IfRange(f )=B,fmapsAontoBandwecallfasurjec-
tion;f isabijectionifitisone-to-oneandonto,thatis,ifitisbothaninjection
andasurjection,inwhichcasewewritef:AB.
Note that surjectiveness of a map depends on the set into which the map is
dened. For example, if we consider f (x) =x
2
as f:RR
+
then it is onto,
whereasthesamefunctionviewedasf:RRisnotonto.
Tosummarize:
injectionsareone-to-oneandmapAintoBbutmaynotcoverallofB;
surjectionsputAalloverBbutmaynotbeone-to-one;and
bijections from Ato Bare one-to-one onto functions, which means that
theirinversecorrespondencesarefunctionsfromBtoA.
Example2.6.18 LetE=[0, 1]A=R,H=[0, 1]B=R, andf (a) =2a.
Range(f )=Rso thatfisasurjection;theimagesetisf (E) =[0,2];theinverse
imagesetisf
1
(H)=[0,
2
1
];andfisaninjection,hasinversef
1
(b)=
2
1
b,and
asaconsequenceofbeingone-to-oneandontoisabijection.[SeeFigure2.6.18.]
Exercise 2.6.19 Showthatiff:ABisabijectionbetweenAandB,then
thesubsetrelationsinTheorem2.6.4(p.34)holdwithequality.
2.6.d Compositions of Functions
Ifwerstapplyf toana Atogetb=f (a),andthenapplygtob, we have a
new,compositefunction,h(a)=g(f (a)).
38 Chapter2 SetTheory
Denition2.6.20 Letf:ABandg:B
C, withB
Band Range(f )
B
)[f (a) =f (a
)], which
impliesthat(g f )(a)=(g f )(a
),sothatg fisnotone-to-one.
To have g fbeonto,oneneedsg(Range(f ))=C. To have g fbeone-to-
one,inadditiontofbeingone-to-one,gshouldbeinjectiveontheRange(f ), but
notnecessarilyonthewholeB.
Exercise 2.6.27 Giveanexampleoffunctionsf, gwheref isnotontobutg
andg f areonto.Alsogiveanexamplewheregisnotinjective, butg f is
injective.
Denition 2.6.28 TheidentityfunctiononasetAisthefunctionf:AA
denedbyf (a) =aforalla A.
5. Hints: In the case of a surjection, we must showthat for(g f ) (a):=g(f (a)),itisthecasethat
c C,thereexistsa Asuchthat(g f ) (a)=c.Toseethis,letc C.Sincegisasurjection,
b Bsuchthatg(b)=c.Similarly,sincef isasurjection,a Asuchthatf (a) =b.Then
(g f )(a)=g(f (a))=g(b)=c.
2.7 WeakandPartialOrders,Lattices 39
Example2.6.29 LetMbeaniteset(ofmen)andWaniteset(ofwomen).A
matchingisafunctionfromMWtoitselfsuchthatforallmM,[(m)
W][(m)=m], forallwW, [(w)M][(w)=w], andisthe
identityfunction.
2.7
)an ordered
set.Ordersarerelations,butweuseadifferentnamebecauseouremphasisison
orderswithinterpretationsreminiscentoftheusualorder,, on R.Table2.7gives
namestopropertiesthatanorder
mayormaynothave.
Table2.7
Property Name
(x)[x
x] Reexivity
(x, y)[[x
y][y
x]] Symmetry
(x, y)[[[x
y][y
x]][x=y]] Antisymmetry
(x, y, z)[[x
y][y
z]][x
z]] Transitivity
(x, y)[[x
y][y
x] Completeness
(x, y)(u)[x, y
u] Upperbound
(x, y)()[
Lowerbound
x, y]
(x, y)(u)[[x, y
u][[x, y
][u
]] Leastupperbound
(x, y)()[[
x, y][
Greatestlowerbound
x, y][[
]]
Toseethateverycompleterelationisreexive,takex=yinthedenition.
Thethreemainkindsoforderedsetswestudyareinthefollowing.
Denition2.7.1 (X,
(X,
)isapartiallyorderedset(POSET)if
transitive.(X,
)isalatticeifitisaPOSETwiththeleastupperboundandthe
greatestlowerboundproperty.
Thefollowingare,foreconomists,themostfrequentlyusedexamplesofthese
kindsoforderedsets.
6.Earlier, in2.5, weusedrationalinsteadofweakforcompleteandtransitiverelations.
Itwouldhavebeentoohardforus,aseconomists,toaccepttheideathatthefoundationofour
theoryofchoicewasweak;hencethechoiceofagrandername,rational.
40 Chapter2 SetTheory
2
3
y x y
1
x y x
1 3
x (3, 1) x y (1, 1)
y (1, 3) x y (3, 3)
FIGURE 2.7.2
Example 2.7.2 Weak ordering: If X= R
2
and
y iff
is dened by
+
u(x)u(y)forautilityfunctionu,then(X,
)isaweaklyorderedset.Weoften
representthisweakorderingbydrawingrepresentativeindifferencecurves.
Partialordering: If X=R
2
isdenedbytheusualvectororder,that and
+
is,byx
)isa(POSET)thatisnotweakly
yiffx
1
y
1
andx
2
y
2
,then(X,
orderedbecauseitfailscompleteness,forexample,thepair(1,3)and(3,1)are
notcomparable,asneitherisgreaterinthevectororder.
Lattices: R
2
with the vector order is the canonical example of a lattice. For
allx=(x
1
, x
2
),y=(y
1
, y
2
) R
2
,thegreatestlowerbound,denotedx y,isthe
point(min{x
1
, y
1
},min{x
2
, y
2
}),andtheleastupperbound,denotex y,isthe
point(max{x
1
, y
1
},max{x
2
, y
2
}).[SeeFigure2.7.2.]Bythesamelogic, R
isa
latticewiththevectororder.
IfX=AB,A, BR,isarectangularset,thenitisnotonlyaPOSET,but
alsoitisalattice.However,ifXistheline{(x
1
, x
2
) R
2
:x
1
+x
2
=1} withthe
vectorordering,thenitisaPOSETthatisnotalattice.
Notation2.7.3 Ifthegreatestlowerboundofapairx, yinaPOSETXexists,
itiscalledtheinmumanddenotedinf
X
(x, y),inf(x, y), or x y.Iftheleast
upper bound of a pair x, y Xexists, it is called the supremum and denoted
sup
X
(x, y),sup(x, y), or x y.
Example 2.7.4 The relation on R is reexive, antisymmetric, transitive,
complete; sup(x, y)isthemaximumofthetwonumbersxandyandinf(x, y)
istheminimumofthetwo.Thus,(R,)isaweaklyorderedset,aPOSET,anda
lattice.OnR,theequalityrelation,=,isreexive,symmetric,antisymmetric,and
transitive,butfailstheotherproperties.
There is a tight connection with the logical connectives and introduced
earlier.RecallthattheindicatorofasetAXisthatfunction1
A
:X{0,1}
denedby
1 if x A,
1
A
(x):= (2.3)
0 if x A.
Thesetofindicatorfunctionsofsubsetsprovidesaclassicalexampleoflattices.
2.7 WeakandPartialOrders,Lattices 41
Example 2.7.5 The relation on indicator functions is dened by 1
A
1
B
when1
A
(x) 1
B
(x)foreveryx X.Notethat1
A
1
B
iffAB,sothat is
notcomplete.Ontheclassofindicatorfunctions, isreexive, antisymmetric,
transitive; 1
A
1
B
=1
AB
and1
A
1
B
=1
AB
,sotheclassofindicatorsisa
latticewiththerelation.
Exercise 2.7.6 Eventhoughtherelation forvectorsinR
2
isnotrationalbe-
causeitisnotcomplete,dening[(x
1
, x
2
)(y
1
, y
2
)][[(x
1
, x
2
) (y
1
, y
2
)]
[(y
1
, y
2
)(x
1
, x
2
)]]givesanequivalencerelation.Whatistheequivalencerela-
tion?
You may be wondering how any reasonable relation with upper and lower
boundscouldfailtohavetheleastupperorgreatestlowerboundproperty.
Example2.7.7 Therelation<onR(oronQ)isnotreexive,isantisymmetric
because r <s and s <r never happens, is transitive, and has the upper bound
andthelowerboundpropertybutnottheleastupperboundnorthegreatestlower
boundproperty.Thisfailureof<tohavetheleastupperandgreatestlowerbound
hasfar-reachingconsequences.Toseewhythefailureoccurs,letr=s=0and
note that for any integer n, no matter how large, 1/10
n
is an upper bound, but
1/10
n+1
isasmallerupperbound.
Thetheoryofrationalchoicebyindividualsstudiesweaklyorderedsets.
Example2.7.8 Supposethat
isarationalpreferenceorderingonX,thatis,
itiscompleteandtransitivesothat(X,
)isaweaklyorderedset.Completeness
implies that
yand u=yif y
x.
Parallellogicimpliesthatthereexistsatleastonelowerbound.Iftherearetwo
ormoreindifferentpointsinX,thenantisymmetryfails,sothat(X,
)isnota
lattice.
Thesubsetrelationisnottheonlyinterestingoneforsets.
Exercise 2.7.9 LetXbethenonemptysubsetsofY,thatis,X=P(Y)\ {}.
Dene the relation t on Xby At Bif A B=. Mnemonically, At B if A
touchesB.Fory Y,letX(y)={A X:y A}.Which,ifany,oftheproperties
giveninTable2.7(p.39)doestherelationthaveonX? On X(y)?
FromDenition2.5.4(p.29),givenarelation(e.g.,apartialordering)
ona
setX,wedenexyifx
yandy
xandxxifx
yand(y
x).
Exercise 2.7.10 Showthat=(
\ ).
)andthat=(
(y X)[yx].
if
Example 2.7.12 IfX={(x
1
, x
2
) R
2
:x
1
+x
2
=1},theneverypointinXis
undominatedin.
Denition 2.7.13 Let{
:i I} beacollectionofrationalpreferenceorder-
i
ings on a set X. Dene the unanimity order on Xby x
U
y if for all i I,
x
i
y.ApointxisParetooptimalorParetoefcientifxisundominatedin
U
.
Ifx
U
x,thenx
Paretoimprovesonx.
42 Chapter2 SetTheory
Inallbutthesimplestcases,theunanimityorderwillfailcompleteness.
Example 2.7.14 LetX =[0, 1]withxXrepresentingtheproportionofthe
apple pie that person 1 will consume and (1x)the proportion that person
2 will consume. Suppose that preferences are selsh, [x > x
][x
1
x
]and
[x >x
][x
2
x
].Nopairofpointscanberankedbytheunanimityorderand
everypointisParetooptimal.
ThedenitionofParetooptimalityinDenition2.7.13istheusualone.Note
thaty
U
xiff(iI)[y
i
x]and(jI)[y
j
x].Thus,xisParetoefcient
iffthereisnoythateveryonelikesatleastaswellandsomeonestrictlyprefers.
ThisdiffersfromDenition1.5.3(p.12),whichisoftenknownasweak Pareto
efciency:xisweaklyParetoefcientifthereisnoysuchthat(iI)[y
i
x].
Exercise 2.7.15 Show that if x is Pareto efcient, then it is weakly Pareto
efcient. Modify the preferences of person 1in Example 2.7.14 so that person
1gainsnoextrautilityfromanypiebeyondthree-quartersofthepie;thatis,three-
quartersofthepiesatiatesperson1.Withthesepreferences,giveanexampleofan
xthatisweaklyParetoefcientbutnotParetoefcient.
Whenpreferencesarenonsatiable,theallocationcanbecontinuouslyadjusted,
andpreferencesarecontinuous,thenweakParetooptimalityandParetooptimality
arethesame.
2.8
(t ) denotethesolution(s)totheproblemP(t ),
maxf (x, t ), (2.4)
xX
andaskhowx
(t ) dependsontastvariesoverT .
SincetheproblemP(t ) in(2.4)ismeantasanapproximationto,ratherthana
quantitativerepresentationof,behavior,weareafterqualitativeresults.These
areresultsthatareoftheformiftincreases,thenx
(t )willincrease,andthey
should be fairly immune to details of the approximation.
7
If x
(
.
)is differ-
entiable, then we are after a statement of the form dx
/dt0. If x
(
.
)is not
7.Anotherpartoflearningtothinklikeaneconomistinvolvesdevelopinganaestheticsenseof
whatfairlyimmunemeans.Aestheticsarecomplicatedandsubtle,bestgainedbyimmersion
and indoctrination in and by the culture of economists, as typically happens during graduate
school.
2.8 MonotonicChangesinOptima:SupermodularityandLattices 43
differentiable,thenweareafterastatementoftheformthatx
isnondecreasing
int .Wearegoingtogoaftersuchstatementsinthreeways.
2.8.a. The implicit function theorem, using derivative assumptions onfwhen
XandT areone-dimensionalintervals.
2.8.b. The simple univariate Topkis theorem,usingsupermodularityassump-
tions on f when Xand T are linearly ordered sets, for example, one-
dimensionalintervals.
2.8.c. Monotone comparative statics, using supermodularity when X is a
latticeandT ispartiallyordered.
Unlikerankingbyutility,supermodularityisnotimmunetomonotonicrescal-
ing, that is, supermodularity is a cardinal, not an ordinal concept. Quasi-super-
modularityistheordinalversionofsupermodularity, andwedealwithitinthe
lastpartofthissection.
Denition 2.8.1 A partial order that also satises completeness is called a
total(orlinear)ordering,and(X,
)iscalledatotallyorderedset. A chainina
partiallyorderedsetisasubset,X
X,suchthat(X
)istotallyordered.
The classical example of a linearly ordered set is the real line (R,). In a
total ordering, any two elements x and y in Acan be compared, whereas in
a partial ordering, there are noncomparable elements. For example, (P(N),)
is a partially ordered set with many noncomparable elements. However, the set
containing{1},{1,2}, . . . , {1,2, . . . , n}, . . . isachaininP(N).
Exercise 2.8.2 ShowthatifABandBistotallyordered,thenAistotally
ordered.
2.8.a The Implicit Function Approach
AssumethatXandT areintervalsubsetsofRandthatf istwicecontinuously
differentiable.
8
The terms f
x
, f
t
, f
xx
, and f
xt
denote the corresponding partial
derivativesoff . To have f
x
(x, t )=0characterizex
(t ),wemusthavef
xx
<0
(whichisastandardresultaboutconcavityinmicroeconomics).Fromtheimplicit
function theorem, we know that f
xx
= 0 is what is needed for there to exist a
functionx
(t )suchthat
f
x
(x (t ), t )0. (2.5)
Tonddx
/dt ,takethederivativeonbothsideswithrespecttotandnd
f
dx
=0, (2.6)
xx
dt
+f
xt
sothatdx
/dt=f
xt
/f
xx
.Sincef
xx
<0,thismeansthatdx
/dtandf
xt
have
thesamesign.
8.Westudythecontinuityoffunctionsinsomedetaillater.Fornow,weareassumingthatthe
readerhashadacalculus-basedmicroeconomicscourse.
44 Chapter2 SetTheory
ThisoughttobeanintuitiveresultabouttheproblemP(t ) in(2.4):iff
xt
>0,
thenincreasesintincreasef
x
;increasesinf
x
areincreasesinthemarginalreward
ofx;andasthemarginalrewardtoxgoesup,weexpecttheoptimallevelofxto
goup.Inaparallelfashion:iff
xt
<0,thenincreasesintdecreasef
x
;decreases
inf
x
aredecreasesinthemarginalrewardofx;andasthemarginalrewardtox
goesdown,weexpecttheoptimallevelofxtogodown.
Exercise 2.8.3 LetX=T=R
+
andf (x, t ) =x
2
1
(xt )
2
.Findx
(t )and
verifydirectlythatdx
/dt>0.Alsondf
x
,f
xx
,andf
xt
,andverify,usingthe
signtestjustgiven, thatdx
/dt>0.Ifyoucandrawthree-dimensionalgures
(andthisisaskillworthdeveloping), drawf andverifyfromyourpicturethat
f
xt
>0andthatitisthisfactthatmakesdx
/dt>0.Topracticewithwhatgoes
wrongwithderivativeanalysiswhentherearecornersolutions,repeatthisproblem
withX=R
+
,T=R,andg(x, t )=x
2
1
(x+t )
2
.
Example2.8.4 Theamountofapollutantthatcanbeemittedisregulatedtobe
nomorethant0.Thecostfunctionforamonopolistproducingxisc(x, t )with
c
t
<0andc
xt
<0.Thesederivativeconditionsmeanthatincreasesintheallowed
emissionlevellowercostsandlowermarginalcosts,sothatthermwillalways
chooset .Foragivent ,themonopolistsmaximizationproblemistherefore
maxf (x, t ) =xp(x)c(x, t ), (2.7)
x0
where p(x)is the (inverse) demand function. Since f
xt
= c
xt
, we know that
increasesintleadthemonopolisttoproducemore,providedf
xx
<0.
The catch in the previous analysis is thatf
xx
=xp
xx
+p
x
c
xx
, so that it seems
wehavetoknowthatp
xx
<0,orconcavityoftheinversedemand,andc
xx
>0,
orconvexityofthecostfunction, beforewecanreliablyconcludethatf
xx
<0.
Theglobalconcavityoff (
.
, t ) seemstohavelittletodowiththeintuitionthatit
istheloweringofmarginalcoststhatmakesx
dependpositivelyont .However,
globalconcavityoff (
.
, t ) isnotwhatweneedfortheimplicitfunctiontheorem,
ratheronlytheconcavityoff (
.
, t ) intheregionofx
(t ).Withdifferentiability,
thislocalconcavityisan implication ofx
(t )beingastrictlocalmaximumfor
f (
.
, t ).Supermodularitymakesitclearthatthelocalmaximumpropertyisallthat
isbeingassumedandallowsustoworkwithoptimathatarenondifferentiable.
2.8.b The Simple Supermodularity Approach
The simplest case has Xand T being linearly ordered sets. The most common
example has Xand T being intervals in R with the usual less-than-or-equal-to
order.However,nothingrulesoutthesetsXandT beingdiscrete.
Denition 2.8.5 ForlinearlyorderedXandT ,afunctionf:XTRis
supermodularifforallx
xandallt
t ,
f (x
, t
)f (x, t
)f (x
, t ) f (x, t ), (2.8)
2.8 MonotonicChangesinOptima:SupermodularityandLattices 45
equivalently
f (x
, t
) f (x
, t ) f (x, t
) f (x, t ). (2.9)
It is strictly supermodular if the inequalities are strict. For submodularityand
strictlysubmodularity,reversetheinequalities.
At t , the benet of increasing from x to x
is f (x
, t ) f (x, t ), at t
, it
is f (x
, t
) f (x, t
() isthelargestsolutiontomax
xX
f (x, ) forall,then
[t
t ][x
(t
(t )].
andt ,thenx
(t
) x
(t ).
Proof. The idea of the proof is that having x
(t
) x
t butthatx
:=x
(t
) x :=
x
(t ). As x
(t )and x
(t
)are maximizers, f (x
, t
) f (x, t
)and f (x, t )
f (x
, t ).Sincex
isthelargestofthemaximizersatt
andx x
,thatis,x islarger
thanthelargestmaximizeratt
, weknowabitmorethatf (x
, t
) >f (x, t
).
Addingtheinequalities,wegetf (x
, t
) +f (x
, t ), or
f (x, t ) f (x
, t ) >f (x, t
) f (x
, t
),
thatis,strictlydecreasingdifferencesinx andt .
GoingbacktothepollutingmonopolistofExample2.8.4(p.44),weseethat
the supermodularity offreduces to the supermodularity ofc. Thus assumingc
(andhencef )issupermodular,wecanuseTheorem2.8.6toconcludethatx
(t ) is
increasing.Noneofthesecond-derivativeconditionsexceptc
xt
< 0isnecessary,
andthiscanbereplacedbythelooserconditionthatc issupermodular.
CleverchoicesofT sandf scanmakesomeanalysesveryeasy.
Example2.8.7 Supposethattheone-to-onedemandcurveforagoodproduced
by a monopolist is x(p), so that CS(p) =
46 Chapter2 SetTheory
Setf (x, t ) =(x)+t CS(p(x)),whereX=R
+
andT={0,1}.SinceCS(p(x))
isnondecreasing,f (x, t ) issupermodular(andyoushouldcheckthis).Therefore
x
(1)x
(0),sothemonopolistalways(weakly)restrictsoutputrelativetothe
socialoptimum.
Hereistheexternalitiesintuition:increasesinxincreasethewelfareofpeople
themonopolistdoesnotcareabout,aneffectexternaltothemonopolist;themarket
gives the monopolist insufcient incentives to do the right thing. To fully appreciate
howmuchsimplerthesupermodularanalysisis,wehavetoseehowcomplicated
thedifferentiableanalysiswouldbe.
Example 2.8.8 [Example2.8.7(p.45)]Supposethatforeveryt[0,1],the
problem
max(x)+t
.
CS(p(x))
x0
has a unique solution, x
(t )is continuously
differentiable.(Thiscanbeguaranteedifwemaketherightkindsofassumptions
on(
.
)andCS(p(
.
)).)To nd the sign ofdx
(t )/dt ,weassumethattherst-order
conditions,
(x
(t ))+t dCS(p(x
(t )))/dx0,
characterize the optimum. In general, this means that we have to assume that
x(x)+t CS(p(x))isasmooth,concavefunction.Wethentakethederivative
ofbothsideswithrespecttot .Thisinvolvesevaluatingd(
p(x
(t ))
x(r) dr)/dt . In
general,d(
f (t )
x(r) dr)/dt=f
(t )x(f (t )),sothatwhenwetakederivatives
onbothsides,wehave
(x
(t ))(dx
/dt )+dCS(p(x
(t )))/dxp
(x
)(dx
/dt )x(p(x
))=0.
Gatheringtermsyields
[
(x
)p
(x
)x(p(x
))](dx
/dt )+dCS(p(x
(t )))/dx=0. (2.12)
Since we are assuming that we are at an optimum, we know that
(x
)0.
By assumption, p
(x
(t )))/dx>0.Therefore,theonlywaythat
(2.12) can be satised is if dx
(1)x
(0)=
0
1 dx
dt
(r)
dr. The integral of a positive function is positive, so
thisyieldsx
(1)x
(0) > 0.
2.8.c Monotone Comparative Statics
Supposethat(X,
T
)arelattices.Denetheorder
X
)and(T,
XT
onXT
by(x
, t
(x, t )iffx
X
xandt
T
t .(Thisistheunanimityorderagain.)
XT
Lemma 2.8.9 (XT,
)isalattice.
XT
2.8 MonotonicChangesinOptima:SupermodularityandLattices 47
Proof. (x
, t
)(x, t )=(max{x
, x},max{t
, t })XT ,and(x
, t
)(x, t )
=(min{x
, x},min{t
, t })XT .
Denition 2.8.10 Foralattice(L,
),f:LRissupermodularifforall
,
L,
f (
)+f (
)f ()+f (
), (2.13)
equivalently
f (
)f (
)f ()f (
). (2.14)
Taking
=(x
, t ) and=(x, t
),werecoverDenition2.8.5(andyoushould
checkthis).
Exercise2.8.11 Suppose that(L,
=(R
n
, ),xyiffx
i
y
i
,i=1, . . . , n.
)
ShowthatL isalattice.Showthatf:LRissupermodulariffithasincreasing
differencesinx
i
andx
j
foralli=j.Showthatatwicecontinuouslydifferentiable
f:LRissupermodulariff
2
f/x
i
x
j
0foralli=j.
Example 2.8.12 On the lattices (R
n
,), the function xmin{f
i
(x
i
):i=
1, . . . n}is supermodular if each f
i
is nondecreasing, as is the function x
x
1
.
x
2
. . .
x
n1
x
n
. For x, tR
n
, thefunction(x, t )x
.
tissupermodularon
R
n
R
n
.
Exercise 2.8.13 Showthefollowing.
1. Iff:R
n
Ristwicecontinuouslydifferentiable,hasanonnegativegra-
dient,andissupermodular,andg:RRistwicecontinuouslydifferen-
tiableandconvex,theng(f (x))isincreasingandsupermodular.
2. Iff: [1,)[1,)Risdenedbyf (x
1
, x
2
)=x
1
+x
2
andg(r)=
log(r), thenf issupermodularandstrictlyincreasing, whilegisstrictly
increasingandconcave.However,g(f (x))isstrictlysubmodular.
Denition2.8.14 For A, BL,Lalattice,thestrongsetorderisdenedby
A
St rong
Biff (a, b)AB,abAandabB.
Interval subsets ofRare sets of the form(, r),(, r],(r, s),(r, s], [r, s),
[r, s],(r,), or [r,).
Exercise2.8.15 ShowthatforintervalsA, BR,A
St rong
Biffeverypoint
inA\BislessthaneverypointinAB,andeverypointinABislessthan
every point in B\A. Also show that this is true when R is replaced with any
linearlyorderedset.
Thestrongsetorderisnot,ingeneral,reexive.
Example2.8.16 IfA={xR
2
:x
1
+x
2
1},then[A
A]. However,
+
St rong
ifALisitselfalattice,then[A
St rong
A].Inparticular,subsetsof(R,)are
linearlyordered,hencetheyarelattices.
Notation 2.8.17 For SLandtT ,letM(t, S)Sbethesetofsolutions
totheproblemmax
S
f (, t ). For t, t
T ,S, S
L,dene(t
, S
(t, S)if
t
T
tandS
St rong
S.
48 Chapter2 SetTheory
Theorem 2.8.18 If (L,
T
)is a partially ordered set,
L
)is a lattice, (T,
f:LTRissupermodularinforallt ,andhasincreasingdifferencesin
andt ,thenM(t, S)isnondecreasingin(t, S)for(t, S)withM(t, S)=.
Proof. Pick(t
, S
St rong
M(t, S).
(t, S);wemustshowthatM(t
, S
Pick
M(t
, S
)S
andM(t, S)S.Bythedenitionofthestrong
setorder,wehavetoshowthat
M(t
, S
)and
St rong
S,
and
S. As is optimal inSand
S,
weknowthatf (, t )f (
)0.Combiningthesupermodularityoff (
.
, t )
andthislastinequality,wehave
f (
, t ) f (
, t ) f (, t )f (
, t ) 0.
Increasingdifferences,t
T
t ,
,andthislastinequalityyield
f (
, t
)f (
, t
)f (
, t ) f (
, t ) 0.
Since
isoptimalinS
and
,wehavejustdiscoveredthat
isalso
optimalinS
,thatis,
M(t
, S
).
Exercise 2.8.19 CompletetheproofofTheorem2.8.18.
Thefollowingisimmediatefromthelastresult,simplytake(t
, S
)=(t, S).
However,adirectproofmakesclearerhowsubmodularityisworking.
Corollary 2.8.20 If (L,
somepoint(x
1
, x
2
)R
2
++
.
Theproblemisthatsupermodularityisnotimmunetomonotonictransforma-
tions.Thatis,supermodularity,likeexpectedutilitytheory,isacardinalratherthan
anordinaltheory.Hereistheordinalversion.
Denition2.8.22(MilgromandShannon) Afunctionu:XRisquasi-
supermodularonthelatticeXif,x, yX,
[u(x)u(xy)][u(xy)u(y)], and
[u(x) >u(xy)][u(xy) >u(y)].
2.9 TarskisLatticeFixed-PointTheoremandStableMatchings 49
Bywayofcontrast,f:XRissupermodularifx, y X,andf (x y)+
f (x y)f (x) +f (y), which directly implies that it is quasi-supermodular. The
reasonfortheadjectivequasicomesfromintermediateeconomics,whereyou
shouldhavelearnedthatamonotonicallyincreasingtransformationofaconcave
utilityfunctionisquasi-concave.
Lemma 2.8.23 A monotonic increasing transformation of a supermodular
functionisquasi-supermodular.
Exercise 2.8.24 ProveLemma2.8.23.
Recall that a binary relation,
iff[x
y][u(x) u(y)].Forchoicetheoryonnitelatticeswithmonotonic
preferences, quasi-supermodularity and supermodularity of preferences are in-
distinguishable.
Theorem2.8.25(ChambersandEchenique) Abinaryrelationonanite
latticeXhasaweaklyincreasingandquasi-supermodularrepresentationiffithas
aweaklyincreasingandsupermodularrepresentation.
Proof. Since supermodularity implies q-supermodularity, we need only show
that a weakly increasing q-supermodular representation can be monotonically
transformedtobesupermodular.Letubequasi-supermodular,setu(X)={u
1
<
u
2
<
. . .
<u
N
},anddeneg(u )=2
n1
,n=1, . . . , N.Youcanshowthatthe
n
functionv(x)=g(u(x))issupermodular.
Exercise2.8.26 Therearethreeresultsthatyoushouldknowabouttherelation
betweenmonotonicityandquasi-supermodularity:
1. Strongmonotonicityimpliesquasi-supermodularity,hencesupermodular-
ity: Show that if a binary relation
(x
, y
)iff(x, y)(x
, y
).Showthat
the utility function u(x, y)= 0 if x= y= 0 and u(x, y)= 1otherwise
is weakly monotonic, but no monotonic transformation of it is quasi-
supermodular.
3. Supermodularity does not imply monotonicity: Let X= {(0,0), (0,1),
(1,0), (1,1)} with(x, y)
(x
, y
)iff(x, y)(x
, y
meansthatmpreferstobematchedwithwratherthan
w
, and m
m
w means that m would prefer to be single, that is, matched
withhimself,ratherthanbeingmatchedtow.Inexactlythesamefashion,each
w W has strict rational preferences over M {w}. Man mis acceptable to
woman wif she likes him at least as much as staying single, and woman wis
acceptabletomanmifhelikesheratleastasmuchasstayingsingle.
AsinExample2.6.29(p.39),amatchingisaone-to-oneontofunctionfrom
M Wtoitselfsuchthat istheidentity,[(m)=m][(m) W],and
[(w)=w][(w) M].Interestfocusesonpropertiesofthesetofmatchings
that are acceptable to everyone and that have the property that no one can nd
someonethatheorshepreferstohisorherownmatchandwhoprefershimorher
tohisorhermatch.Formally,
Denition 2.9.1 Amatchingisstableifitsatises:
1. individualrationality:m,(m)
m
mandw,(w)
w
w,and
2. stability:(m, w) MWsuchthatw
m
(m)andm
w
(w).
Forthepurposesofalaterproof,thefollowingisworthhavingasaseparate
result,eventhoughitisanimmediateimplicationofthestrictpreferenceswehave
assumed.
Lemma 2.9.2 Ifisindividuallyrational,thenitisstableiff(m, w)such
that[w
m
(m) m
w
(w)]or[m
w
(w) w
m
(m)].
Exercise2.9.3 Amatching
isindicatedbyboldfaceandthe
sinfollowing
table,whichshowspreferences(Ex.2.4,RothandSotomayor1990).
w
2
m
1
w
m
1
w
3
m
1
m
1
m
w
1
m
3
w
1
m
2
w
1
w
1
w
1
m
2
w
m
2
w
2
m
2
m
2
m
w
2
m
1
w
2
m
2
w
2
w
2
w
1
m
3
w
m
3
w
3
m
3
m
3
m
1
w
3
m
3
w
3
m
w
3
w
3
1. Verifythat
isstable.
2. Given a matching , for each m, dene ms possibility set as P
(m)=
{w W:m (w)} {m}.Showthatthematching
maximizeseach
w
ms preferences over P
(w).[Thesimultaneousoptimizationofpreferencesovertheavailable
setofoptionsisapatternweseerepeatedly.]
Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.
Alternative Proxies: