A Note on the History of the Cantor Set and Cantor Function Author(s): Julian F. Fleron Source: Mathematics Magazine, Vol. 67, No. 2 (Apr., 1994), pp. 136-140 Published by: Mathematical Association of America Stable URL: http://www.jstor.org/stable/2690689 . Accessed: 14/10/2013 11:42 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 support@jstor.org. . Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to Mathematics Magazine. http://www.jstor.org This content downloaded from 132.178.2.64 on Mon, 14 Oct 2013 11:42:52 AM All use subject to JSTOR Terms and Conditions MATHEMATICS MAGAZINE 136 A Noteon theHistory oftheCantorSet and CantorFunction JULIAN F. FLERON SUNY at Albany Albany,New York12222 A searchthroughthe primaryand secondaryliteratureon Cantoryieldslittleabout the historyofthe Cantorset and Cantorfunction. In thisnote,we wouldlike to give some of thathistory,a sketchof the ideas underconsiderationat the timeof their discovery,and a hypothesisregardinghow Cantorcame upon them.In particular, Cantorwas not the firstto discover"Cantorsets." Moreover,althoughthe original discoveryof Cantorsets had a decidedlygeometricflavor,Cantor'sdiscoveryof the Cantorset and Cantorfunction was neithermotivatedby geometry nordid it involve geometry, even thoughthisis how theseobjectsare oftenintroduced(see e.g. [1]). In fact,Cantormayhave come upon themthrougha purelyarithmetic program. The systematic studyofpointset topologyon the real line arose duringthe period 1870-1885 as mathematicians investigated twoproblems: could be integrated, and 1) conditionsunderwhicha function 2) uniquenessof trigonometric series. It was withinthe framework of these investigationsthat the two apparently independentdiscoveriesofthe Cantorset were made; each discoverylinkedto one of theseproblems. BernhardRiemann(1826-1866) spentconsiderabletimeon the firstquestion,and suggestedconditionshe thoughtmightprovidean answer.Althoughwe will not discuss the two formshis conditionstook(see [2, pp. 17-18]), we note thatone of these conditionsis importantas it eventuallyled to the developmentof measure theoreticintegration [2, p. 28]. An important step in thisdirectionwas the workof HermannHankel (1839-1873) duringthe early 1870s. Hankel showed,withinthe framework of Riemann,thatthe integrability of a functiondepends on the natureof certainsets of pointsrelatedto the function.In particular,"a function, he [Hankel] wouldbe Riemann-integrable thought, if,and onlyif,it were pointwisediscontinuous thatforeverya > 0 theset ofpointsx at [2, p. 30],"meaning,in modernterminology, which the functionoscillatedby more than a- in every neighborhoodof x was nowheredense. Basic to Hankel's reasoningwas his belief that sets of the form forall nowheredense subsetsofthereal line.Workingunder {1/2'} were prototypes thisassumptionHankelclaimedthatall nowheredense subsetsofthe real line could be enclosedin intervalsof arbitrarily smalltotallength(i.e. had zero outercontent) [2, p. 30]. As we shall see, thisis notthe case. (See also [3].) intothenatureofcertainpointsets wouldbecome AlthoughHankel'sinvestigation extremelyimportant,"as was the case with Dirichlet and Lipschitz,it was the of the possibilitiesof infinitesets-in particular, inadequacyof his understanding nowheredense sets-that led him astray.It was not until it was discoveredthat nowheredense sets can have positiveoutercontentthatthe importanceofnegligible sets in the measure-theoretic sense was recognized[2, p. 32]." The discoveryof such sets, nowheredense sets withpositiveoutercontent,was made by H. J. S. Smith (1826-1883), SavilianProfessorof Geometryat Oxford,in a paper [4] of 1875. After an expositionofthe integration ofdiscontinuous Smithpresenteda method functions, forconstructing nowheredense sets thatwere muchmore"substantial"thanthe set This content downloaded from 132.178.2.64 on Mon, 14 Oct 2013 11:42:52 AM All use subject to JSTOR Terms and Conditions 137 VOL. 67, NO. 2, APRIL 1994 he observedthe following: {1/2'}. Specifically, Let mnbe any givenintegralnumbergreaterthan2. Divide the intervalfrom0 to 1 into mnequal parts;and exemptthe last segmentfromany subsequentdivision.Divide each of the remaining in - 1 segmentsintom equal parts;and exemptthe last segmentsfromanysubsequentsubdivision. we shallobtainan infinite numberofpointsof division If thisoperationbe continuedad infinitum, P upon the line from0 to 1. These pointslie in loose order... [4, p. 147]. Smith's'loose order'is whatwe referto as nowheredense. In modernterminology discussionis the assumptionthatthe exemptedintervals Implicitin Smith'sfurther are open, so the resultingset is closed. Today thisset would be knownas a general Cantorset,and thisseems to be the firstpublishedrecordof such a set. Later in the same paper, Smithshows thatby dividingthe intervalsremaining beforethe nth step into Mn equal partsand exemptingthe last segmentfromeach subdivisionwe obtaina nowheredense set ofpositiveoutercontent.Smithwas well as he states,"the resultobtainedin thelast ofthisdiscovery, awareoftheimportance example deserves attention,because it is opposed to a theoryof discontinuous whichhas receivedthe sanctionof an eminentgeometer,Dr. Hermann functions, in the contemporary Hankel [4, p. 149]." He continuesby explainingthe difficulties examples illuminate. that his theoriesof integration to note thatan editor'sremarkat the conclusionof Smith'spaper It is interesting states"thispaper,thoughit was notread, was offeredto the societyand acceptedin the usual manner."(Emphasis added.)1 In fact,this paper went largelyunnoticed Smith'scrucial on the Europeancontinentand unfortunately amongmathematicians almosta decade later,of similar discoverieslay unknown.It tookthe rediscovery, of contemporary theoriesof integration ideas by Cantorto illuminatethe difficulties integration. and to begin the evolutionofmeasure-theoretic GeorgCantor(1845-1918) came to the studyofpointset topologyaftercompleting a thesison numbertheoryin Berlinin 1867. He began workingwithEduard Heine (1821-1881) at the Universityof Halle on the question of the uniqueness of series.This questioncan be posed as follows: trigonometric If forall x exceptthosein someset P we have 00 2ao+ E (an cos(nx) + bnsin(nx))= 0 n1 mustall thecoefficients an and bnbe zero? "when the convergencewas uniform Heine answeredthe questionin the affirmative in general with respectto the set P, which is thus takento be finite[2, p. 23]," thatdid on anysubinterval thatthe convergencewas uniform meaning,by definition, notcontainanypointsof the finiteset P. withthis problem.In papers [5, 6] of 1870 and Cantorproceeded much further in general"and 1871,he removedtheassumptionthattheconvergencewas "uniform set. In doingso he beganto lookat beganto considerthecase when P was an infinite whatwe now considerthe fundamental pointset topologyofthe real line. In a paper [8] of 1872, Cantorintroducedthe notionof a limitpointof a set thathe definedas we do today,callingthe limitpointsofa set P the derivedset,whichhe denotedby P'. Then P" was thederivedset of P', and so on. Cantorshowedthatiftheset P was IIt is possible that "not read" simplymeant thatthe paper was not presentedat a meetingof the of thisnote,one mustconsiderthat London MathematicalSociety.However,in weighingthe significance in vols. 3-10 of the Proceedingsof the London MathematicalSociety(1871-1879), and perhaps even no otherpaperwas similarlynoted. further, This content downloaded from 132.178.2.64 on Mon, 14 Oct 2013 11:42:52 AM All use subject to JSTOR Terms and Conditions MATHEMATICS MAGAZINE 138 series Ia0 + such that p(n) = 0 for some integer n and the trigonometric = + on then all of the coefficients P, except possibly 0, sin(nx)) cos(nx) bn Eon=(an had to be zero. Cantor'sworkon thisproblemwas "decisive"[9, p. 49], and doubly role in muchofhis upcoming as his derivedsets would play an important important work. In theyears1879-1884 Cantorwrotea seriesofpapersentitled"Uber unendliche, [10-15]," thatcontainedthe firstsystematictreatlinearePunktmannichfaltigkeiten ofthreetermsin mentofthepointset topologyofthereal line.2It is theintroduction of this series Cantor this series thatconcernsus mosthere. In the firstinstallment dense (literally"uiberalldicht"),a defineswhatit meantfora set to be everywhere termwhose usage is still current.He gives a few examples,includingthe set of numbersoftheform22n?+1/211 wheren and m are integers,and continuesby noting the relationshipbetween everywheredense sets and theirderived sets. Namely, dense in (a, ,B) if[and onlyif] P' = (a, ,3) [10, pp. 2-3]. In P c (a, ,3) is everywhere the fifthinstallmentof this series Cantordiscusses the partitionof a set into two of a componentsthathe termsreducibleand perfect[14, p. 575]. His definitionperfectset is also stillcurrent:A set P is perfectprovidedthat P = P'. Cantor states that Afterintroducingthe term perfectin the fifthinstallment, perfectsets need not be everywheredense [14, p. 575]. In the footnoteto this statementCantorintroducesthe set thathas become knownas the Cantor(ternary) set: The set ofreal numbersof the form Cv Cl x=3 + * + cv 31) +* perfectset wherec1 is 0 or 2 foreach integerv. Cantornotesthatthisis an infinite, withthe propertythatit is not everywheredense in any interval,regardlessof how smalltheintervalis takento be. We are givenno indicationofhowCantorcame upon thisset. papers, During the time Cantorwas workingon the 'Punktmannichfaltigkeiten' otherswere workingon extensionsof the FundamentalTheorem of Calculus to functions. Cantoraddressedthisissue in a letter[18] dated November discontinuous 1883, in whichhe definesthe Cantorset,just as it was definedin the paper [14] of 1883 (whichhad actuallybeen writtenin Octoberof 1882). However,in theletterhe It the firstknownappearanceof thisfunction. goes on to definethe Cantorfunction, whose values is firstdefinedon the complementofthe Cantorset to be the function are I (c2 2 1 foranynumberbetween 6 3 3+ * + 3_l-+ 3/1-1 31L and b= 3-l + .+ 31`1 2 + 3/i' whereeach c1V is 0 or 2. Cantorthenconcludesthissectionoftheletterby notingthat on [0,1]. can be extendednaturallyto a continuousincreasingfunction thisfunction to Harnack'sextensionofthe FundamentalTheorem Thatservesas a counterexample which was in vogue at the time (see e.g. of Calculus to discontinuousfunctions, [2, p. 60]). We are givenno indicationof how Cantorcame upon thisfunction. There are two othertopicsthatinterestedCantorthatwe would like to mention and it is witharithmetic constructions because theyare indicativeof Cantor'sfacility (see [16, 17]), 2in addition,thesepaperscontainedmanyothertopicsthathad farreachingimplications of higherorder derived sets thatmarkedthe "beginningsof Cantor's includingCantor'sinvestigation numbers[2, p. 72]." theoryoftransfinite This content downloaded from 132.178.2.64 on Mon, 14 Oct 2013 11:42:52 AM All use subject to JSTOR Terms and Conditions 139 VOL. 67, NO. 2, APRIL 1994 possiblywithinthis settingthat Cantor came upon the Cantor set and Cantor function.First,Cantorspent some time in the mid 1870s consideringthe possible betweena line and a plane,a questionmostof existenceofa bijectivecorrespondence In 1877,in a letterto RichardDedekind as absurd. dismissed had his contemporaries had found such a correspondence.This that he explained (1831-1916), Cantor as follows: can be expressed "correspondence" Let(xl,x2)be a pointintheunitsquare,and let O-X1 lX 2X1,3 ... andOX2 1X2 22X2,3* be decimal Map the point(x1, x2) to the pointon the real line whose expansionsof xl and x2 respectively. is O.X1 1X2,1X1,2X2,2 ... (See e.g.[19,p. 187].) decimalexpansion Dedekind pointedout that therewas a problemwith this approach.The decimal expansionsof rationalsare not unique, so to avoid duplicationwe must not allow expansionsof some type, say expansionsthat contain infinitestringsof zeros. However, by disallowingexpansionswith infinitestringsof zeros, the irrational number0.11010201010201010102... could neverbe obtainedunderCantor'scorrespondence. This reasoningdoes howevergive us an injectionof [0,1] X [0,1] into[0,1]. It is trivialto findan injectionof[0,1] into[0,1] X [0,1]. These twofacts,togetherwiththe Theorem(if thereare injectionsofthe set A intothe set B and Schroeder-Bernstein betweenA and B; see thenthereis a bijectivecorrespondence B intoA respectively, between[0,1] e.g. [20]), allow us to concludethatthereis a bijectivecorrespondence duringtheperiodin question and [0,1] X [0,1]. However,set theorywas in itsinfancy and it would be 20 years beforeE. Schroederand Felix Bernsteinindependently provedthe theoremthatbears theirnames[16, p. 172-173] and occasionallyCantor's name as well (e.g. [21, 22]). So thiswas notan optionforCantor. exhibita bijection.To do thishe modifiedhis Instead,Cantorneeded to explicitly [23]. Denote the continuedfraction previousapproachto use continuedfractions 1 al + a2 where a1, a2,a3,... > 0 are integers. by [a1, a2,a3,...] 1 +a number, if,and onlyif,it representsan irrational Since a continuedfraction is infinite is unique [see e.g. 24], Cantorcould set up the in which case the representation correspondence ([a,,, al2,...], , [a 1, a2,2, ...] [a2,1, a2 1, .. an1, al,2, a2,2,. [an,1an,2 ... ' , an,2) I in (0, 1)n = (0, 1) X (0, 1) X ... x (0, 1) and irrationals betweenn-tuplesofirrationals of the previousapproachand gives a bijective in (0, 1). This avoids the difficulties _ between([0, 1] Q)n and [0,1] - Q. Cantorthentookgreatlengths correspondence between[0,1] and [0,1] - Q. Repeated to provetherewas a bijectivecorrespondence gives a bijective applicationof thisfactcombinedwiththe previouscorrespondence and [0, 1]. between[0, 1]n correspondence Secondly,it is knownthatCantorstudiedbinaryexpansions.In fact: denotedby o, could be representedas Cantorrecognisedthatthe powerof the linearcontinuum, well by [thepowerof]the set of all representations: x= wheref(v) = 2 + + 2v + 0 or 1 [foreach integerv] [19, p. 209]. This content downloaded from 132.178.2.64 on Mon, 14 Oct 2013 11:42:52 AM All use subject to JSTOR Terms and Conditions 140 MATHEMATICS MAGAZINE There is, so it seems,no substantiveevidence about how Cantorcame upon the Cantor set and Cantor function.However, given Cantor's route into point set topology,his arithmetic introduction of the Cantorset and Cantorfunction, and his facilitywitharithmetic methods,as we have just illustrated, it is feasiblethatit is withinthe arithmetic framework of binaryand ternaryexpansionsthatCantorcame upon the Cantorset and Cantorfunction. REFERENCES 1. R. L. Wheeden and A. Zygmund,Measure and Integral:An Introduction to Real Analysis,Marcel Decker,Inc., New York,1977,p. 35. 2. T. Hawkins,Lebesgue's Theoryof Integration: Its Originsand Development, Chelsea PublishingCo., Madison,WI, 1975. 3. H. L. Royden,Real Analysis,MacmillanPublishingCo., New York,1988,p. 64. 4. H. J. S. Smith,On the integration of discontinuous functions, Proc. LondonMath. Soc. (1) 6 (1875), 140-153. 5. G. Cantor,Beweis, daB eine fir jeden reellen Werth von x durch eine trigonometrische Reihe gegebeneFunctionf(x) sich nuraufeine einzigeWeise in dieserFormdarstellenliBt,Part1, Crelle Jl. Math. 72 (1870), 139-142. Reprinted[7, pp. 80-83]. 6. G. Cantor,Beweis, daB eine fir jeden reellen Werth von x durch eine trigonometrische Reihe gegebeneFunctionf(x) sich nuraufeine einzigeWeise in dieserFormdarstellenliBt,Part2, Crelle Jl. Math. 73 (1871), 294-296. Reprinted[7, pp. 84-86]. 7. G. Cantor,GesammelteAbhandlungen mathematischen und philosophischen Inhalts,E. Zermelo(ed.), New York,1980. Springer-Verlag, 8. G. Cantor,Uber die Ausdehnungeines Satzes aus der Theorieder trigonometrischen Reihen,Math. Ann.5 (1872), 123-132. Reprinted[7, pp. 92-102]. 9. W. Purkert,Cantor'sPhilosophicalViews,in The Historyof ModernMathematics, Vol. 1: Ideas and TheirReception,D. E. Rowe and J. McCleary(eds.), AcademicPress,Boston,1989. 10. G. Cantor,Uber unendliche,lineare Punktmannichfaltigkeiten, Part 1, Math. Ann. 15 (1879), 1-7. Reprinted[7, pp. 139-145]. 11. G. Cantor,Uber unendliche,linearePunktmannichfaltigkeiten, Part2, Math.Ann.17 (1880), 355-358. Reprinted[7, pp. 145-148]. 12. G. Cantor,Uber unendliche,linearePunktmannichfaltigkeiten, Part3, Math.Ann.20 (1882), 113-121. Reprinted[7, pp. 149-157]. 13. G. Cantor,Uber unendliche,linearePunktmannichfaltigkeiten, Part4, Math. Ann.21 (1883), 51-58. Reprinted[7, pp. 157-164]. 14. G. Cantor,Uberunendliche,linearePunktmannichfaltigkeiten, Part5, Math.Ann.21 (1883), 545-591. Reprinted[7, pp. 165-209]. 15. G. Cantor,Uberunendliche,linearePunktmannichfaltigkeiten, Part6, Math.Ann.23 (1884), 453-488. Reprinted[7, pp. 210-246]. 16. J.W. Dauben, Georg Cantor:His Mathematicsand Philosophyof the Infinite,PrincetonUniversity Press,Princeton,NJ,1990. 17. W. Purkertand H. J. Ilgauds,GeorgCantor:1845-1918, BirkhauserVerlag,Basel, 1987. 18. G. Cantor,De la puissancedes ensemblesparfaitsde points,Acta Math.4 (1884), 381-392. Reprinted [7, pp. 252-260]. 19. J. W. Dauben, The developmentof Cantorianset theory,in From the Calculus to Set Theory: 1630-1910, I. Grattan-Guinness (ed.), Gerald Duckworth& Co., London,1980. 20. G. F. Simmons,Introduction to Topologyand ModernAnalysis,R. E. KriegerPublishingCo., Malabar, FL, 1983,pp. 28-30. 21. K. Hrbacekand T. Jech,Introduction to Set Theory,Marcel Decker,Inc., New York,1984,p. 72. New York,1982, 22. G. Takeutiand W. M. Zaring,Introduction to AxiomaticSet Theory,Springer-Verlag, p. 86. 23. G. Cantor,Ein Beitragzur Mannigfaltigkeitslehre, CrelleJl. Math.84, (1878), 242-258. Reprinted[7, pp. 119-133]. 24. G. H. Hardyand E. M. Wright,An Introduction to the TheoryofNumbers,ClarendonPress,Oxford, 1989,pp. 129-140. This content downloaded from 132.178.2.64 on Mon, 14 Oct 2013 11:42:52 AM All use subject to JSTOR Terms and Conditions