The Evolution of Integration

Author(s): A. Shenitzer and J. Steprans

Source: The American Mathematical Monthly, Vol. 101, No. 1 (Jan., 1994), pp. 66-72
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2325128 .
Accessed: 24/11/2013 21:58

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The American Mathematical Monthly.

http://www.jstor.org

This content downloaded from 129.219.247.33 on Sun, 24 Nov 2013 21:58:53 PM

All use subject to JSTOR Terms and Conditions
 THE EVOLUTION OF...
Edited byAbe Shenitzer
NorthYork,OntarioM3J1P3, Canada
Mathematics,YorkUniversity,

An Englishmajormayor maynotbe a novelist or a poet,butwouldundoubtedly

be expectedto be able to evaluatea novelor a poem.The term"Englishmajor"
impliessomehistorical, philosophical,and evaluativetrainingand competence. It
is sad but truethatthe term"mathematician" does not implycorresponding
training and competence.
Integration of the narrowly mathematical and historical,philosophical and
criticalaspectsofourdiscipline is boundto makeit moremeaningful notonlyto
thosewhoidentify themselves as mathematicians butalso to thosewhohaveno
morethana tangential interestin thesubject.
To promotesuchintegration, and thusencouragean approachto mathematics
thatemphasizesitsmeaningand significance, theMonthly willpublisheverytwo
months an articleof2-5 pagesunderthegenerictitle"The evolution of... " The
coreof suchan articlewillbe an accountof important mainstream mathematics.
The essaythatfollows exemplifies thekindofmaterial, andtheapproach, we have
in mind.
Whilewe preferoriginalarticles, we willalso publishtranslationsor adapta-
tionsofappropriate articlesin thepublicdomain.
Abe Shenitzer

The EvolutionofIntegration

A. Shenitzerand J. Steprians

THE GREEK PERIOD. The Greek problemunderlying integrationis the quadra-

tureproblem:Given a plane figure,constructa square of equal area.
It is easy to solve the quadratureproblemfora polygon,a figurewithrectilinear
boundary.The firstquadratureof a figurewithcurvilinearboundarywas achieved
by Hippocratesin the fifthcenturyB.C. Hippocratesshowed thatthe area of the
lunule in FIGURE 1 (that is, the figurebounded by one-halfof a circleof radius 1
and one-quarterof a circleof radius V2) is equal to the area of the unitsquare B.
Hippocratesmanaged to square two otherlunules.*
In the thirdcenturyB.C. Archimedeseffectedthe quadratureof a parabolic
segment.He showed that its area is 4A , where A is the triangleof maximalarea
inscribedin the parabolic segment.
Archimedeseffecteda numberof other quadratures(and cubatures).Some of
his quadraturesinvolvedinventiveconstructions but most relied on the technique
of wedgingan area between ever closer upper and lower approximatingsums.

*Two morequadrablelunuleswerefoundbyT. Clausenin the19thcentury. In the20thcentury,

twoRussianalgebraists thatthesefivelunulesare theonlyquadrableones.
proved(independently)

66 THE EVOLUTION OF ... [January

This content downloaded from 129.219.247.33 on Sun, 24 Nov 2013 21:58:53 PM

All use subject to JSTOR Terms and Conditions
 B

Figure1

Figure2

Analogs of such sums are a keyelementof thedefinition

of theDarboux integral(a
variantof theRiemannintegralintroducedbyDarboux in the19thcentury)as wellas
of quadratureprogramsfor computers.We illustrateboth of Archimedes' ap-
proachesnext.
ConsiderFIGURE 3. Here the hypotenuse AB of the righttriangleOAB is
tangentto the spiralat A. It thenturnsout thatthe side AB is equal to the

rA

Figure3

1994] THE EVOLUTION OF ... 67

This content downloaded from 129.219.247.33 on Sun, 24 Nov 2013 21:58:53 PM

All use subject to JSTOR Terms and Conditions
circumference ofthecirclewithradiusOA. (Thisis a specialcase ofArchimedes'
of circulararcsbyusingtangents
rectification to spirals.)Sincehe knewthatthe
area of a circleis halftheproductof itscircumferencebyitsradius,we can say
a circleand square its area.
that Archimedesused (a tangentto) a spiralto rectify
Theirbrilliance notwithstanding,suchconstructionshavebeen reducedto histori-
cal footnotesbecausetheyfailedto yieldgeneralmethods.
FIGURE 4 showsa turnofArchimedes' spiralr = aG andtheassociatedcircleof
radius27ra,andthusofarea K = 47-3a2. To computethearea S oftheturnofthe
spiralin FIGURE 4 Archimedes itfrombelowand abovebyunionsof
approximates
circularsectorsindicatedin FIGURE 5.

Figure4 Figure5

are,respectively,
figures
The areasoftheseapproximating

n= 3 [12 + 22 + = - (n 1)(2

and
=4 3 27w3a2(n+ 1)(2n + 1)
n [1 + 2 + +n]=3

to see that
It is notdifficult

S' Sn< 32r3<5"1

<n -wa < St/
3n
as
can be rewritten
forall n. Thisdoubleinequality
K
S'n < - < S5n
3
forall n. Obviously,
SI < S < SI
forall n. To provethatS = (K/3) Archimedes showsthatSn - S= (473a2/3n2)
andis thussmallforlargen. He can nowshowthattheassumption S -K/3 leads
to a contradictionand can concludethatS = (K/3).
WhileArchimedes he relieson the"methodof
makesno explicituse oflimits,
exhaustion," and, in modernterms,the finalpartof the argument in a proof
involving themethodofexhaustion (intheaboveexampleitis disproving S -K/3)
amountsto proving theuniquenessofthelimitofa Cauchysequence.

68 THE EVOLUTION OF ... [Jatiuary

This content downloaded from 129.219.247.33 on Sun, 24 Nov 2013 21:58:53 PM

All use subject to JSTOR Terms and Conditions
CONTINUATION IN THE 17TH CENTURY. Using nonrigorousinfinitesimal
techniques (ratherthanrigorousalgebraicmethods ofthekindusedbyArchimedes)
Cavalieri(1598-1647)managedto compute(whatwe nowwriteas) Jflxkdx for
was the evaluationof lk + * +nk. In about
k = 1,2,... ,9. His chiefdifficulty
1650FermatevaluatedJfxp/q dx bymeansof a brilliant yetsimplecomputation.
Furtherprogress was due to Torricelli, Pascal
Wallace,and Pascal. In particular,
interpreted Cavalieri's"sum of lines" (the equivalentof area) as a sum of
infinitesimalrectangles.
Ifwe combineFermat'sresultwithCavalieri'sunderstanding ofthelinearity
of
thedefinite integral thenwe see thatbythemiddleofthe17th
(our terminology!)
century one could evaluateJfP(x) dx, P(x) a "polynomial" withrationalexpo-
nents.
In 1647Gregory St. Vincentmade a discovery thatlinkedNapier'slogarithm
functionand the area under the hyperbolaxy= 1. This connectionis now
expressedas loge(x) = Jf(dt/t).
NewtonandLeibnizinvented thecalculusandmadeitintoa toolwithcountless
applications butneither gavewhatwe wouldcall a rigorous ofa definite
definition
integral(or sawtheneedforsucha definition). Suchconcerns becamedominant in
the 19thcentury.
FROM CAUCHY TO LEBESGUE. The firstrigorousdefinition of a definite
integral was givenbyCauchyin the1820s.Cauchydealtwithcontinuous functions.
In view of the importanceof Fourierserieswhose coefficients are givenby
integrals it was necessary to definetheintegralformoregeneralfunctions. This
was firstdone byRiemann.The limitations of the Riemannintegral werereme-
died at thebeginning ofthe20thcentury byLebesgue.An explanation follows.
With each theoryof integration thereis associateda theoryof measure.
Specifically,if f is a functionon a set E and f = f+ - f- (recall that f+(x) =
max{f(x),0) and f-(x) = max{-f(x), 0)) then JEf is definedas the difference
JEf'- JEf- ofthemeasuresJEf+ and JEf- oftheordinate setsofthenonnega-
tivefunctions f+ and f- respectively.
The measureunderlying the Riemannintegralis Jordanmeasureand the
measureunderlying the Lebesgueintegralis Lebesguemeasure.How do they
differ? In whatwayis one "better"thantheother?
Considerthe simplecase of the ordinateset M of a bounded,nonnegative
function f on an interval,0 < f(x) < c forx in [a, b]. The Jordanmeasureof M
is the commonvalue,if any,of theouterand innerJordanmeasuresof M. The
outerJordan measureof M is theglboftheareasofthecoverings of M consisting
of finiteunionsofrectangles. The innermeasureof M is thedifference between
thearea C(b - a) oftherectangle S withbase [a, b] and heightC and theouter
measureofthecomplement of M in S. Lebesguereplacedtheword"finite"inthe
Jordandefinition ofthemeasureof a subsetof S by"countable."Thisincreased
greatly thenumberofmeasurablesubsetsof S and led to a theory of integration
farmorecomprehensive and mathematically flexible than Riemann's.
THE HK-INTEGRAL.Surprisingly, Henstock(in 1955) and Kurzweil(in 1957)
cameup witha newversionoftheRiemannintegral-callit theHK-integral (see
[71)-thatis "as goodas" theLebesgueintegral!Its definitionandmaincharacter-
isticsfollow(see [7]):
Definition:A taggeddivisionof[a, b] givenbya finiteorderedset a = x0 < x1 <
... witha collection
< Xn = b ofpoints,together of tagszi suchthatxi-1 < zi <

1994] THE EVOLUTION OF ... 69

This content downloaded from 129.219.247.33 on Sun, 24 Nov 2013 21:58:53 PM

All use subject to JSTOR Terms and Conditions
xi fori = 1,. .., n. We denotea taggeddivision
byD(xi, zi) andthecorresponding
Riemannsumby
n

S(D( xi, zi)) f( zi)( xi -xi-,).i

i=l

A gaugeon [a, b] is a function8 definedon [a, b] suchthat 8(x) > 0 forall

x E [a, b]. An importantexampleof a gaugeis a constantfunction. If 8 is any
gauge on [a, b], we say thata taggeddivisionD(xi, zi) is 8-finein case that
[xi-1, xi] C [zi - 8(zi), zi + 8(zi)]; that is, in case zi - 8(zi) < xi-1 < zi < xi <
zj + 8(zi) for all i = 1,2,..., n. Finally,we say that the number A is an HK-
integralof f if,foreveryE > 0, thereexistsa gauge86 suchthatif D(xi, zi) is any
taggeddivisionof[a, b] thatis 8--fine,
thenwe have
IS(D(xi, zi)) - Al < E.

It turnsout that"the HK-integral is uniquelydefinedwhenit

of a function
ifand onlyifthegauge86 can be
is Riemannintegrable
existsand thata function
chosento be constant."Moreimportantly, function
"everyLebesgueintegrable is
HK-integrablewiththesamevalue."

THERE IS NO PERFECT INTEGRAL. While in the eyes of some mathematicians

the Lebesgueintegralwas the finalanswerto the difficulties associatedwith
integration, therewereotherswhowerenotwillingto giveup the searchforthe
perfectintegral,onewhichwouldmakeall functions BecauseLebesgue's
integrable.
construction had shownthatthekeyto a comprehensive theory ofintegration was
theconstruction ofan appropriate measure,thesearchnowfocussedon finding a
totalmeasureon IR,thatis,one whichassignsa measureto each subsetofthereal
numbers.
Vitali[6] showedthata totalmeasureon therealscannotbe countably additive
and translation invariant. This being so, it is naturalto ask whichof these
propertiesshouldbe retained.This decisionis, of course,somewhatarbitrary.
Whileretaining translation invarianceleads to somefascinating grouptheory and
theBanach-Hausdorff-Tarski Paradox,we willconsiderwhathappensifcountable
additivityis retainedinstead.
In 1930S. Ulam[1] showedthatthereis no suchmeasureon w1,w02or on any
cardinal'whichis the successorof some othercardinal.Ulam's proofwas a
spectacularadvancein thatit did notrelyon anyof the geometric assumptions,
suchas translation invariance, on whichearlierproofsoftheexistence ofnon-mea-
surablesetshad relied.
By Ulam's theorem, the existenceof a countably additivemeasureon R that
measuresall of its subsetsimpliesthat2'o is not the successorof any other
cardinal,thatis,itis a limitcardinal.Byarguing a bitmorecarefully one can show
thattheremustexistsomelimitcardinalA < 2.o whichis nottheunionoffewer
thanA setsof size less thanA. The existenceof sucha cardinalhas a profound
influence on settheory.
In orderto understand thisinfluence,itis necessaryto recall(a consequenceof)
Godel'ssecondincompleteness theorem whichsaysthatset theory can notprove
itsownconsistency. One waytoprovetheconsistency ofa theory is to finda model

thenotationw to represent
tCantorintroduced and it is still
thenextordinalaftertheintegers
favored today.The nextcardinalafterw is denotedw1andso on.
bysettheorists

70 THE EVOLUTION OF ... [January

This content downloaded from 129.219.247.33 on Sun, 24 Nov 2013 21:58:53 PM

All use subject to JSTOR Terms and Conditions
ofthattheory, thatis,a mathematical structure satisfyingall oftheaxiomsofthat
theory.We ask: Whatare the implications forset theoryof the existenceof a
modelofsettheory? Recalltheprocedurefortheconstruction ofthehierarchy of
sets.One beginswiththe emptyset-call it V0-and thendefinesVk to be the
powersetof Vk- 1 foreachintegerk ? 1. Thisis nottheend,though, becauseone
can thendefineV,Oto be theunionofthesetsVk and thendefineV.,, to be the
powerset of V,O. If one continuesthisas faras possibleand takestheunionone
getsa modelof settheory-or,at least,whatwouldbe a modelofsettheory ifit
werea setand nota properclass.
How soon,ifever,does thisconstruction processlead to a modelofsettheory?
It turnsoutthatmanyoftheaxiomsofsettheory are satisfiedat earlystagesofthe
construction. For examplethe axiomof infinity is satisfiedas soon as a single
infinite set is includedand thisis alreadytrueof V.1. The powerset axiomis
satisfied at anylimitstagebecauseanysetwhichoccurs,occursat a stagebefore
thelimitand so all ofitssubsetsare added at theverynextstage.The powerset
itselfis therefore added in no morethantwostagesand,in anycase,beforethe
limit.For similarreasons,the pairingaxiomis also satisfiedat all limitstages.
Well-foundedness and comprehension are also easyto deal with.
The problematic axiomis theaxiomof replacement, whichsaysthattherange
of anyfunction definedbya formula is a set.It has alreadybeen mentioned that
V,,+,,willsatisfy all of theaxiomsof settheory exceptforreplacement. Replace-
mentfailsbecause the mappingwhichtakes2n to GO + n and 2n + 1 to n is
definableby a formulaand its domainis GOwhichbelongsto V,+1 c V,,+,O
However,therangeofthisfunction is co+ cowhichdoes notbelongto V.+.. The
sameargument can be usedto showthatV, is a modelofsettheory ifand onlyif
thefollowing holds:
. if A < a then2A < a
. ifA < a thenanyfunction
F: A -- a (definedusingonlyparameters
fromV,)
has rangeboundedin a.
Anycardinalsatisfying theserequirements is knownas a largeor inaccessible
cardinal.Sincetheexistence ofa largecardinalimpliesthata modelofsettheory
exists,itfollowsfromG6del'sTheoremthatit is impossible to provetheexistence
ofinaccessible cardinals.
Ulam's argument showsthatif thereis a countably additivemeasurewhich
measureseveryset of realsthenthereis a cardinala whichsatisfies the second
requirement ofbeingan inaccessible cardinal.Suchcardinalsare knownas weakly
inaccessible.Anotherof Godel's majorcontributions is the notionof the Con-
structibleUniverse,one of whoseconsequencesis thatanymodelof set theory
containsa submodelwhichsatisfiesthe generalizedcontinuum hypothesis. This
allowsus to concludethatif thereis a weaklyinaccessiblecardinalthen,in the
Constructible Universe, theweaklyinaccessible cardinalis in factan inaccessible
cardinal; this is so because the cardinalarithmetic of thissmallermodelof set
theory easilyimplies the first requirement being largecardinal.
for a
In otherwords,ifthereis a countably additivemeasurewhichmeasuresevery
setof realsthan set theory is This
consistent. and G6del'stheoremshowthatthe
of a
existence perfect integralis notprovable. theotherhand,it is conceivable
On
thatsome day theremaybe a proofthatit is not possibleto have a perfect
integral.The impactof thison set theorywouldbe devastating. It wouldfollow
thatmanyofthelargecardinalswhichexpertsnowconsiderquiteinnocuous, and
whichhaveplayedan important rolein manyimportant independence do
results,

1994] THE EVOLUTION OF ... 71

This content downloaded from 129.219.247.33 on Sun, 24 Nov 2013 21:58:53 PM

All use subject to JSTOR Terms and Conditions
notexist.Whilethiswouldnotshowthatset theory it would
itselfis inconsistent
thatit is.
shakeourfaithin theassumption
severely

We've told our storybut would neverthelesslike to tack on the following

relevant"postscript":
solvetheGreekquadrature
In whatsensedoes theintegral problemandwhatis
answerto thesetwoquestionsfollows.
A telegraphic
itsconceptualsignificance?
The integralprovidesa direct"analytic"solutionof the Greek quadrature
problemforregionsoftheform

f(x)

0 a b
Figure6

Indeed,thearea oftheregionin thefigureis

A =bf(x) dx.

thisas
Ifwe rewrite

A = f(x) dx = (b-a)( ba f(x) dx),

region"hasbeenreplacedbya rectangle
thenitis clearthatour"integral ofequal
area with base b - a and height (1/(b - a))fa'f(x) dx. The quantity(1/(b -
a))Jabf(x)dx is theaverageof thefunctionalvaluesof f on [a, b]. Thisaveraging
of theintegral
ability is thekeyto itsimportance
in countless
applications.

REFERENCES

1. S. Ulam,Zur Masstheorie in derallgemeinen

Mengenleh, Fund.Math.16 (1930),140-150.
2. C. H. Edwards, Development
Jr.TheHistorical 1979.
oftheCalculus,Springer-Verlag,
3. 0. Toeplitz,Thecalculus-A Genetic University
Approach, ofChicagoPress,1963.
4. A. Aaboe,EpisodesfromtheEarlyHistory ofMathematics, theMAA,NML 13.
5. T. Jech,Set Theory,AcademicPress,NewYork,1978.
6. G. Vitali,Sulproblemadellamesuradeigruppidipuntidi unaretta,Bologna,1905.
7. R. G. Bartle,reviewof R. Henstock'sTheGeneralTheory BAMS,v. 29, #1,July
ofIntegration,
1993,pp. 136-139.

and Statistics
ofMathematics
Department
YorkUniversity
NorthYork,Ontario
CanadaM3J1P3.

72 THE EVOLUTION OF ... [January

This content downloaded from 129.219.247.33 on Sun, 24 Nov 2013 21:58:53 PM

All use subject to JSTOR Terms and Conditions