MathematicsStackExchangeisaquestionandanswersiteforpeoplestudyingmathatanylevelandprofessionals

inrelatedfields.It's100%free,noregistrationrequired.

login

I'mhavingtroublefindingagoodexplanationoftheLebesgueintegral.Asperthedefinition,itistheexpectationofarandomvariable.
Thenhowdoesitmodeltheareaunderthecurve?Let'stakeforexampleafunctionf (x) = x .Howdowefindtheintegraloff (x)
under[0, 1] usingtheLebesgueintegral?
(probability ) (integration) (analysis ) (probabilitytheory ) (lebesgueintegral)

editedAug23'14at7:01

askedOct21'10at18:11

MartinSleziak
30.3k

user957

60

151

743

12

22

5 Whatdefinitiongives"theexpectationofarandomvariable"?Whichbookdoesthis?ArturoMagidinOct
21'10at18:19
Canyoupointmetoasourcethatbestexplainsitwithsomeexamples? user957 Oct21'10at18:32
Therandomvariableistheheightofthecurve.QiaochuYuanOct21'10at23:08
Isuspecthe'slearningfromamathematicalfinancebook.Raskolnikov Dec7'10at19:35

9Answers

Ashasbeennoted,theusualdefinitionoftheLebesgueintegralhaslittletodowithprobability
orrandomvariables(thoughthenotionsofmeasuretheoryandtheintegralcanthenbe
appliedtothesettingofprobability,whereundersuitableinterpretationsitwillturnoutthat
the(Lebesgue)integralof(acertain)functionscorrespondstotheexpectationof(acertain)
randomvariable).
ButthisisnottheoriginoftheLebesgueintegral.Hereisanintuitiveideaofwhatthe
Lebesgueintegralis,ascomparedtotheRiemannintegral.
b

RecallfromCalculustheideabehindtheRiemannintegral:theintegral f (x) dx ismeant

torepresentthenetsignedareabetweenthexaxis,thegraphofy = f (x),andthelines
x = a and x = b .Thewayweattempttodothisisbybreakingupthedomain, [a, b] ,into
subintervals[a = x , x ],[x , x ], , [x
,x
= b] .Then,oneachsubinterval [x , x
]we
pickapoint x ,andweestimatetheareaunderthegraphofthefunctionwiththerectangleof
heightf (x ) andbase[x , x ].ThisleadstotheRiemannsums
a

n1

i+1

i+1

n1

f (x )(x i+1 x i )
i

i=0

asestimatesoftheareaunderthegraph.Wethenconsiderfinerandfinerpartitionsof[a, b]
andtakelimitstoestimatethearea.
Lebesgue'sideawasthatinsteadofpartitioningthedomain,wewillpartitiontherangeifthe
functiontakesvaluesbetweenc andd,wecandividetherange[c, d] intosubintervals
[c = y , y ] , [y , y ], , [y
,y
= d] .Then,welet E bethesetofallpointsin [a, b]
whosevalueunderf liesbetween y and y .Thatis,
0

m1

Ei = f

i+1

([yi , yi+1 ]) = {x [a, b] | yi f (x) yi+1 }.

Ifwehaveawayofassigninga"size"to E ,callitits"measure"(E ) ,thentheportionofthe

graphofy = f (x)thatliesbetweenthehorizontallinesy = y andy = y willbeA ,where,
i

i+1

yi (Ei ) A yi+1 (Ei ).

SoLebesguesuggeststoapproximatethetheareabypickinganumber y between y and

y
,andconsideringthesums

i+1

n1

(Ei )y .
i

i=0

tour

help

Takethe2minutetour

Lebesgueintegralbasics

signup

Thenconsiderfinerandfinerpartitionsof[c, d] ,andthisgivesfinerandfinerapproximations
ofoftheareabythesesums.TheLebesgueintegralwillbethelimitofthesesums.(Theanalogy
givenbyMikeSpiveyisveryaptforthedistinctionbetweenpartitioningthedomainand
partitioningtherangetofindthesum.)
Butinorderforthistomakesense,weneedtodevelopawayofmeasuringfairlyintricate
subsetsoftheline,sothatwecancompute(E ) .Sowefirstdevelopawayofdoingthis
turnsoutthatifyouaccepttheAxiomofChoice,thenitisimpossibletocomeupwithawayof
measuringthatwill(i)assigntoaninterval[a, b] the"measure"b a(ii)willbeinvariant
undertranslation,sosothatifF = E + c = {e + c|e E} then(F ) = (E) (iii)willbe
countablyadditive:ifE = E andthe E arepairwisedisjoint,then(E) = (E )
and(iv)everysubsetofthelinewillhaveawelldefined(possiblyinfinite)measure.(Ifyou
don'taccepttheAxiomofChoice,thentherearemodelsoftherealswherewecanachieve
this).Soonedropstherestriction(iv),andconstructsameasureforwhichsomesetswillbe
"tooweird"tohaveameasure.Wethenrestrictattentiontocertainkindsoffunctions(called
themeasurablefunctions),whicharetheonesforwhichthesetswegetintheprocess
describedaboveareallmeasurablesets.AndthenwedefinetheLebesgueintegralforthose
functions,followingtheideadescribedabove(butonedoesnotdefineitexactlythatway
insteadtheusualwayistodescribef asalimitoffunctionsforwhichtheintegraliseasy,and
thencomputetheintegraloff asalimitoftheintegralsthatareeasy).
i

i=1

Foryourfunction,f (x) = x ,thisisfairlyeasy:thevalueallliebetween0 and1 ,sosaythat

webreakuptherangeintosubintervalsoflength1/n,so y = i/n ,i = 0, , n .Then
2

([yi , yi+1 ]) = f

sothenthestimate,picking y

= yi = i/n

([i/n, (i + 1)/n]) = [ i/n, (i + 1)/n],

isjust

(i/n) ( (i + 1)/n i/n) .

i=0

Takethelimitasn ,andyouwillgetthatthelimitis ,asexpected.(Iwillspareyouthe

detailsseetheendofthisanswerforahighpowerwayofgettingtheanswersimilartothe
wayyoudoitwiththeRiemannintegral).
1
3

ItturnsoutthatnoteveryfunctionisLebesgueintegrable,justlikenoteveryfunctionis
Riemannintegrable.ButeveryfunctionthatisRiemannintegrablewillalsobeLebesgue
integrable,andthevalueofitsLebesgueintegralwillbethesameasthevalueofitsRiemann
integral.ButtherearefunctionsthatarenotRiemannintegrablebutareLebesgueintegrable
(forexample,thecharacteristicfunctionoftherationalsisLebesgueintegrable,withintegral0
overanyinterval,butisnotRiemannintegrable).Wealsohavea"FundamentalTheoremof
Calculus"fortheLebesgueIntegral:
Theorem.IfF isadifferentiablefunction,andthederivative F isboundedonthe
interval[a, b] ,then F isLebesgueintegrableon[a, b] and

d = F (x) F (a).

Here,theintegralistheLebesgueintegral.
3

Inparticular,tofinallyanswerthequestionyouaskaboutyourexample,sinceF (x) =
isa
differentiablefunctionwhosederivativeisboundedoveranyfiniteinterval,inparticularover
[0, 1] ,thenfromthistheoremyoucandeducethattheintegralovertheinterval [0, 1] ofthe
derivative F (x) = x isequaltoF (1) F (0) thatis,
x

d =
0

1
) d =

1
=

.
3

IrecommendthebookAGardenofIntegralsbyFrankE.Burk(DolcianiMathematical
Expositions31,MAA,2007,ISBN9780883853375)itdiscussesandcomparestheCauchy
integral,theRiemannintegral,theRiemannStieltjesintegral,theLebesgueintegral,the
LebesgueStieltjesintegral,andtheHenstockKurzweilintegralitalsodiscussestheWiener
andFeynmanintegral.Ijustfinishedreadingitrecently.
editedOct22'10at3:19

answeredOct21'10at19:09
ArturoMagidin
197k

Wellsaid,Arturo.MikeSpivey Oct21'10at19:32

20

379

651

 @Mike:ThanksIliketheanalogyyougive,too.ArturoMagidinOct21'10at19:40
Thanks.IwishIcouldclaimtheanalogywasoriginal.:)MikeSpivey Oct21'10at19:43
+1,verycompletepost.JonasTeuwenOct22'10at12:39
excellentdescription!veryhelpfulthanks! user957 Oct24'10at21:18

Oneofmygraduateschoolprofessors,ErhanCinlar,usedtogivethefollowinganalogyto
explaintheintuitivedifferencebetweentheLebesgueintegralandtheRiemannintegral.
Supposeyouhaveapileofcoinsofdifferentdenominations,andyouwanttoknowhowmuch
moneyyouhave.TheRiemannintegralislikepickingupthecoins,onebyone,andadding
thedenominationofeachtoarunningtotal.TheLebesgueintegralislikesortingthecoinsby
denominationfirst,andthengettingthetotalbymultiplyingeachdenominationbyhowmany
youhaveofthatdenominationandthenaddingupthosenumbers.Themethodsaredifferent,
butyouobtainthesameresultbyeithermethod.
Inthesameway,whenboththeRiemannintegralandtheLebesgueintegralaredefined,they
givethesamevalue.Asothershavesaid,though,therearefunctionsforwhichtheLebesgue
integralisdefinedbuttheRiemannintegralisnot,andsointhatsensetheLebesgueintegralis
moregeneralthantheRiemann.
editedOct22'10at4:12

answeredOct21'10at19:04
MikeSpivey
32.7k

92

182

5 OnewaytoseeyouranalogyinthecontextofintegralsistonoticethatRiemanintegralsapproximatethe
areabysumsofverticalrectangles,whiletheLebesgueintegralinsteaduseshorizontalrectangles...
MarianoSurezAlvarez Oct21'10at21:22
1 @Mariano:Or,youhaveabunchofpilesofcoins.Riemannaddsupeachpileseparately,thenaddsupthe
totals.Lebesguecountshowmanypenniesareinallthepiles,andgetsapartialtotalthencountshow
manynickelsthenhowmanydimesetc.Andthenaddsupthetotals ArturoMagidinOct22'10at4:45
5 ThisanalogywasactuallygivenbyLebesguehimself(accordingtoDunham,TheCalculusGallery).
MichaelGreinecker Jan24'12at7:39

TheLebesgueintegralisageneralizationoftheusualRiemannintegraltaughtinbasic
calculus.IftheRiemannintegralofafunctionoverasetexiststhenitequalstheLebesgue
integral.SotheLebesgueintegralof x over[0, 1] isjusttheold(1/3)1 (1/3)0
2

TheLebesgueintegralhasthebenefitofbeingdefinedformanymorefunctionsthanthe
Riemannintegral.EvenmoreimportantlytheLebesgueintegralhasusefullimitproperties:
http://en.wikipedia.org/wiki/Dominated_convergence_theorem
http://en.wikipedia.org/wiki/Monotone_convergence_theorem
TheexpectationofarandomvariableisaparticularapplicationoftheLebesgueintegralwhere
thefunctiontobeintegratedistherandomvariable(viewedasafunctiononthesamplespace)
andtheintegrationiswithrespecttoaprobabilitymeasure.
Youneedtolookatoneofthemanyprobabilityandmeasurebooksforthedetails.Myown
favouritesare:
Pollard,AUser'sGuidetoMeasureTheoreticProbability
Dudley,RealAnalysisandProbability
TerenceTaohassomeonlinelecturenotes:
http://terrytao.wordpress.com/2010/09/09/245anotes1lebesguemeasure/
http://terrytao.wordpress.com/2010/09/19/245anotes2thelebesgueintegral/
http://terrytao.wordpress.com/2010/01/01/254anotes0areviewofprobabilitytheory/
answeredOct21'10at18:45
JyotirmoyBhattacharya
2,254

14

33

TheRiemannintegralisprettygoodandveryintuitive,howeverthemainreasontoconsider
othertypesofintegralsisthat"thespaceoffunctionsthatareRiemannintegrable",sayR(I )
whereI R iscompact,istosmall(itisalinearspaceinthesenseyoucanaddthemand
multiplybyconstants).
n

Ifyoujustlookatapiecewisecontinuousfunctionthatvanishoutsideaboundedregionand
thenyoucangoonwiththeRiemannintegral.Inmathematicalanalysiswelookatvarious
kindsoflimitsoffunctionsandwewouldlikethelimitfunctionstostayin"thespace"(we
wantthespacetobecomplete).
AboutthebestwecandointheRiemanncaseistolookatuniformlyconvergentsequences f
onacompactintervalI R inthatcasethelimitlim f R(I ) andlim f = lim f .
However,uniformconvergentisveryrare!(ManyFourierseriesarenotcontinuouseven
thoughtherepartialsumsare,etc..).
n

TheLebesgueintegralcanbeconstructedinseveralways(endingupwiththesamespace
though).AfirsttrymightbetostartwithnormingR(I ),f = |f | andthenwewouldgeta
distancebetweenf , g R(I )byf g,thusturningupwithametricspacewhichwemay
completebyaddingallpossiblelimitsthiswillnotworkhoweverbecauseeventhoughR(I )is
smallitistolarge(thereareunboundedfunctionssuchthatf = ).Abetterstartwould
betolookatC (I ) =thespaceofcontinuousfunctionson(thecompactset)I ,(certainlyeach
f R(I ) isapointwiselimitof C (I ) functions)ifwenorm C (I ) inthesamemannerwe
wouldbeanormedspaceandthecompletionofthatspaceis L (I ) .
1

In L youcansuretakelimitsinnormandmoreover,ashasalreadybeenpointedoutinother
answers,youhavemanyotherbetterlimittheoremssuchasLebesguedominatedtheoremor
themonotoneconvergencetheorem.Also,boundedfunctionsofR(I )dobelongto L (I ) .
1

answeredOct21'10at19:33
AD.
6,864

22

46

Youmaywanttoconsiderthefollowingsources:
Lebesgue'stheoryofintegration:itsoriginsanddevelopmentThomasHawkins
Thecalculusgallery:masterpiecesfromNewtontoLebesgueWilliamDunham
TheLebesgueStieltjesintegral:apracticalintroductionMichaelCarter,Bruce
VanBrunt
answeredOct21'10at20:18
Neves
2,075

16

36

Shouldn'tyoupostreferencesascomments?ValJul13'14at13:20

Thedefinitionheredoesn'tmentionprobabilityorexpectationorrandomvariable.Intuitively,
itjustsaysthemeasure(in R )istheareaofthesmallestsetofrectanglesthatwillcoverthe
set.ThenforanareatheLebesgueintegralisjusttheintegralof1overtheset.
2

answeredOct21'10at18:26
RossMillikan
165k

15

98

215

Youmayalsoliketorefer,thesetwobooks:
AradicalapproachtoLebesgueTheoryofIntegration:Bressoud
RealAnalysisbyG.B.Folland.
AsJyotirmoypointedout,LebesgueintegralisthegeneralizationofRiemannIntegral.There
areshortcomingsoftheRiemannIntegral,duetowhichtheLebesgueintegral,wasdiscovered.
ArigorousdefinitionoftheLebesgueintegralneeds,youtoknowwhataSimpleFunctionis,
andyoucanreadmoreonthisathttp://en.wikipedia.org/wiki/Lebesgue_integration
answeredOct21'10at19:05

anonymous

IfirstunderstoodLebesgueasintegraloverrangeratherthanintegraloverdomainbutitnever
seemedplausibletomethatitcanresolvetheproblemofdiscontinuity.Ithinkthatthetrickis
thatmeasureissomethingthatbetter"width"ofthearea(functionvalueisthesame)than
Rieman'sx 0 does.TheLebegueisbasically

f (s)d = f (s)(s)

Youmaythinkthatyouscanoverf(s)allthevaluesthatyoumayhaveinthedomain
S = (s) andsumupallrectanglesofsize height(s) width(s) = f (s)(s) .
Youcansaythatyouintegrateoverrangeofvaluesf(s).However,youseealsoother
differences.WhereasRiemannimpliesthatthewidthofeveryintervalisthesamedx,thatis,if
youhavefunction
f (1) => 1
f (2) => 1
f (3) => 2
f (4) => 1

theintegralwillaccumulatethesumas0+1+1+2+1.So,ifyouhave4coins,youcancompute
thesumbysimplyiteratingthem.Youseethatthespeedofgrowthisproportionalonlytothe
coinvaluef(x)andnottothecoinindex,x.ThisisnotthecaseofLebesgue.ForLebesgue,you
groupthecoinsbyvalues(IamherefromScala),List(1,1,2,1)=>Occurrences(1>3,2>1),and
youseeimmediatelythatyouhavetwoheapsofcoins:onesinthefirstheapanddoublesinthe
second.Thesearetworectangles(coinvaluexheapsize)thatyouneedtoaddtogether.So,
youintegratebyrangeherebutyounotsimplyaddtogetherthemeasures(theamountof
coins)thatexceedcurrentvalue,youaddcoinvaluexcount.ThisisdifferentfromRiemann.
Thisismysecondstageofunderstandingtheintegrals.
Thisenablesintegratingpointchargelikediscontinuities.Considerastepfunction.Itjumps
instantlyatsomepoints.Itisanintegralofsomeinfinitelynarrowspikesandintegralgrows
instantlyatthespikepoint,asifthereisafinitefiniteamountofcharge/massconcentratedin
infinitesimalpointofspace,incontrasttocontinouslydistributedcharge/massinspace.The
amountofthatchargeinthepointdeterminestheheightofthestep.ItseemsthatRiemann
hasdifficultywithintegratingthespikebecauseregardlesshowinfinitecimalyoumakeyour
dx,theyfailtobreaktheinfinitesimalintervalintorectanglesofconstant(andinfinite)height
f(x)tosumthemup.Ontheotherhand,wecansaythatthereischargenatx = x and
whenweintegrateoveraxisofvalues,wesuddenlypassthroughthevaluen,whichis
measured(countablemeasure)tohavewidthofk(coins),regardlessitisconfinedinasingle
realpointofwidth0.
0

Now,howisthisrelatedtoexpectations?Your4coins,3*1+1*2addupto5eurosand,when
youdrawarbitrarycoin,youexpectitsvaluetobe3*1+2*1/4=5/4=1.25euros.Thatis,one
coincontributes1.25euroinaverage.TheLebesgueintegralisequaltoexpectationwhenthe
measureofallcoins(amountofcoins)isone.Thatis,everyofncoinsisnotactuallyacoinbut
1/n thofit.Now,Lebesgueintegralis3/4*1+1/4*2=5/4istheexpectation.Thatisnot
surprisingbecauseintegratingeveryvaluetimesitsprobabilityiswhattheexpectationis.
Ihesitatedwhethertopostornotmyundergraduategarbagereflections.But,thisvideo
persuadedmethatIamontherighttrack.
answeredJul16'14at12:14
Val
1

@ArturoMagidinIthinkyourexaplanation"SoLebesguesuggeststoapproximatethethe

areabypickinganumber y between y and y ,andconsideringthesums"...iswrong.

IMHO,oneshouldtakethewidthoftherectanglelyingbetween y and y andcalculatethe
areaas:

i+1

i+1

n1

(Ei ) (yi+1 yi ).
i=1

Furtheryourworkedoutexampleonf (x) =
Lebesgueintegral.CorrectmeifIamwrong.
answeredMay15at8:41
chandresh
88

iscomputingReimannintergalandNOT