Who Gave You the Epsilon?

Cauchy and the Origins of Rigorous Calculus

Author(s): Judith V. Grabiner
Source: The American Mathematical Monthly, Vol. 90, No. 3 (Mar., 1983), pp. 185-194
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2975545 .
Accessed: 07/11/2014 00:25

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
The American Mathematical Monthly.

http://www.jstor.org

This content downloaded from 66.241.104.242 on Fri, 7 Nov 2014 00:25:17 AM

All use subject to JSTOR Terms and Conditions
1983] WHO GAVE YOU THE EPSILON? CAUCHY AND THE ORIGINS OF RIGOROUS CALCULUS 185

4. K. Hoffmanand P. Kunze, Linear Algebra,2nd ed., Prentice-Hall,EnglewoodCliffs,NJ, 1971.

5. N. Jacobson,Structureof Rings,reviseded., Amer.Math. Soc., Providence,RI, 1964.
6. M. Kline, Whythe ProfessorCan't Teach, St. Martin'sPress,New York, 1977.
7. Iu. I. Manin, A Course in MathematicalLogic, Springer-Verlag,
1977.
8. G. Polya,How to Solve It, 2nd ed., Doubleday,Garden City,NY, 1957.
9. ___ MathematicalDiscovery,vol. I, Wiley,New York, 1964.
10. W. W. Sawyer,A ConcreteApproachto AbstractAlgebra,Freeman,San Francisco,1959.

WHO GAVE YOU THE EPSILON?

CAUCHY AND THE ORIGINS OF RIGOROUS CALCULUS

JUDITH V. GRABINER
424 West7thStreet,Claremont,
California91711

Student:The car has a speedof 50 milesan hour.Whatdoes thatmean?

S2
Teacher:Givenany? > 0, thereexistsa 8 such thatif jt2 - tlj < 8, then -
-
S1 - 50 < ?.
t2 tl
Student:How in theworlddid anybodyeverthinkof suchan answer?
* * * * * * * * * * * * *

Perhapsthisexchangewill remindus thatthe rigorousbasis forthe calculusis not at all

intuitive-infact,quitethecontrary. The calculusis a subjectdealingwithspeedsand distances,
withtangents and areas-not inequalities. WhenNewtonand Leibnizinventedthecalculusin the
late seventeenthcentury, theydid notuse delta-epsilon proofs.It tooka hundredand fifty years
to developthem.This meansthatit was probablyveryhard,and it is no wonderthata modern
studentfindstherigorous basisof thecalculusdifficult. How, then,did thecalculusgeta rigorous
basis in termsof thealgebraof inequalities?
Delta-epsilonproofsare firstfoundin theworksofAugustin-Louis Cauchy(1789-1867).This
is not alwaysrecognized, since Cauchygave a purelyverbaldefinition of limit,whichat first
glancedoes not resemblemodemdefinitions: "When the successively attributedvalues of the
samevariableindefinitely approacha fixedvalue,so thatfinallytheydiffer fromit by as littleas
desired,thelastis calledthelimitof all theothers"[1].Cauchyalso gavea purelyverbaldefinition
of thederivativeoff(x) as thelimit,whenit exists,of thequotientof differences (f(x + h) -
f(x))/h whenh goesto zero,a statement muchlikethosethathad alreadybeenmadebyNewton,
Leibniz,d'Alembert, Maclaurin,and Euler.Butwhatis significant is thatCauchytranslated such
verbalstatements intothepreciselanguageofinequalities whenhe neededthemin hisproofs.For
instance,forthederivative [2]:
Let 8, e be twoverysmallnumbers;thefirstis chosenso thatforall numerical
(1) [i.e.,absolute]valuesof h less than8, and forany value of x included[in the
interval theratio(f(x + h) - f(x))/h willalwaysbe greater
ofdefinition], than
f'(x) - e and less thanf'(x) + e.

JudithV. Grabinerhas taught ofsciencesince1972at California

thehistory StateUniversity,
DominguezHills,
wheresheis Professor
ofHistory.After
gettinga B. S. in Mathematics ofChicago,shereceived
fromtheUniversity
She is Book ReviewEditorof HistoriaMathematica,Chairman
her M. A. and Ph. D. (1966) at HarvardUniversity.
Sectionof theMathematical
California
of theSouthern ofAmerica,
Association and theauthorof TheOrigins of
(M. I. T. Press,1981).In 1982-1983shewasVisiting
Calculus
Cauchy'sRigorous ofHistory
Professor at U. C. L. A.
Thispaperis a revised
version colloquia,at theSummer
ofa talkgivenat variousmathematics ofthe
Meeting
Association
Mathematical inAnnArbor,
ofAmerica in 1980,andat theNewYorkAcademy
Michigan, ofSciences
in March,1982.Someof theresearchwas supported by theNationalScienceFoundation underGrantNo. SOC
7907844.

This content downloaded from 66.241.104.242 on Fri, 7 Nov 2014 00:25:17 AM

All use subject to JSTOR Terms and Conditions
186 JUDITH V. GRABINER [March

Thisone examplewillbe enoughto indicatehowCauchydid thecalculus,becausethequestion

to be answeredin thepresentpaperis not,"how is a rigorousdelta-epsilon proofconstructed?"
As Cauchy'sintellectualheirswe all knowthis.The centralquestionis,howand whywas Cauchy
able to put thecalculuson a rigorousbasis,whenhispredecessors werenot?The answersto this
historicalquestioncannotbe foundbyreflecting on thelogicalrelationsbetweentheconcepts,but
by lookingin detailat thepast and seeinghowtheexisting in factdevelopedfrom
stateof affairs
thatpast. Thus we will examinethe mathematical situationin the seventeenth and eighteenth
centuries-thebackgroundagainstwhichwe can appreciate Cauchy's innovation.We will
the
describe powerful techniquesof thecalculusof thisearlierperiodand the unimpres-
relatively
siveviewsput forthtojustifythem.We willthendiscusshow a senseof urgency aboutrigorizing
analysisgraduallydevelopedin the eighteenth century.Most important, we will explainthe
development of themathematical techniquesnecessaryforthenew rigorfromtheworkof men
like Euler, d'Alembert,Poisson, and especiallyLagrange.Finally,we will show how these
mathematical thoughoftendevelopedforpurposesfarremovedfromestablishing
results, founda-
tionsforthecalculus,wereusedby Cauchyin constructing his newrigorous analysis.

The PracticeofAnalysis:FromNewtonto Euler. In thelate seventeenth century, Newtonand

Leibniz,almostsimultaneously, independently invented thecalculus.Thisinvention involvedthree
things.First,theyinventedthegeneralconceptsof differential quotientand integral(theseare
Leibniz'sterms;Newtoncalled the concepts"fluxion"and "fluent").Second,theydeviseda
notationfortheseconceptswhichmadethecalculusan algorithm: themethodsnotonlyworked,
but wereeasyto use. Theirnotationshad greatheuristic power,and we stilluse Leibniz'sdy/dx
and fydx, and Newton'sx, today.Third,bothmenrealizedthatthebasic processesof finding
tangentsand areas,thatis, differentiating and integrating, are mutuallyinverse-whatwe now
call theFundamental Theoremof Calculus.
Once thecalculushad been invented, mathematicians possessedan extremely powerfulset of
methods for solving problems in geometry, in physics, and in pure analysis.But whatwas the
natureof thebasic concepts?For Leibniz,thedifferential quotientwas a ratioof infinitesimal
differences,and theintegralwas a sumof infinitesimals. For Newton,thederivative, or fluxion,
was describedas a rateof change;theintegral, or fluent,was itsinverse.In fact,throughout the
eighteenth century, theintegralwas generally thoughtof as theinverseof the differential. One
mightimagineaskingLeibnizexactlywhatan infinitesimal was,or Newtonwhata rateof change
mightbe. Newton'sanswer,thebestof theeighteenth century, Considera ratioof
is instructive.
finitequantities(in modem notation,(f(x + -
h) f(x))/h) as h goes to zero). The ratio
eventually becomes what Newton called an " ultimate ratio."Ultimate ratiosare "limitsto which
theratiosof quantitiesdecreasing withoutlimitdo alwaysconverge, and to whichtheyapproach
nearerthanby any givendifference, but nevergo beyond,nor everreachuntilthe quantities
vanish"[3].Exceptfor"reaching"thelimitwhenthequantities vanish,we can translate Newton's
wordsintoour algebraiclanguage.Newtonhimself, however, did notdo this,nordid mostof his
followers in theeighteenth century. Moreover,"nevergo beyond"does not allow a variableto
oscillateabout its limit.Thus,thoughNewton'sis an intuitively pleasingpicture,as it standsit
was notand couldnotbe usedforproofsabout limits.The definition soundsgood,butit was not
understood in
or applied algebraic terms.
But mosteighteenth-century mathematicians wouldobject,"Whyworryabout foundations?"
In theeighteenth century, thecalculus,intuitively understoodand algorithmically executed,was
appliedto a widerangeof problems.For instance,thepartialdifferential equationforvibrating
stringswas solved; the equationsof motionfor the solar systemwere solved; the Laplace
transform and thecalculusof variationsand thegammafunction wereinventedand applied;all
of mechanicswas workedout in thelanguage of the calculus.These weregreatachievements on
thepartof eighteenth-century mathematicians. Who would be greatly concerned about founda-
tionswhensuchimportant problemscould be successfully treatedby thecalculus?Resultswere
whatcounted.

This content downloaded from 66.241.104.242 on Fri, 7 Nov 2014 00:25:17 AM

All use subject to JSTOR Terms and Conditions
1983] WHO GAVE YOU THE EPSILON? CAUCHY AND THE ORIGINS OF RIGOROUS CALCULUS 187

This point will be betterappreciatedby lookingat an examplewhichillustrates

both the
"uncritical"approachto conceptsof the eighteenth centuryand the immensepowerof eigh-
teenth-centurytechniques,fromtheworkof thegreatmasterof suchtechniques:LeonhardEuler.
The problemis to findthesumof theseries
1/1 + 1/4 + 1/9 + + l/k2 +*
It clearlyhas a finitesumsinceit is boundedaboveby theseries
I + 1/1 *2+ 1/2 3 + 1/3 *4+ -- + l/(k- l)k+- --,
whose sum was knownto be 2; JohannBernoullihad foundthissum by treating1/1 *2 +
1/2 - 3 + 1/3 - 4 + - - as the difference
betweenthe series1/1 + 1/2 + 1/3 + and the
series1/2 + 1/3 + 1/4 + *, and observing thatthisdifferencetelescopes[4].
Euler'ssummation of Ek. II/k2 makesuse of a lemmafromthetheoryof equations:givena
polynomialequationwhoseconstanttermis one,thecoefficient of thelineartermis theproduct
of the reciprocalsof the rootswiththe signschanged.This resultwas both discoveredand
demonstratedbyconsidering theequation(x - a)(x - b) = 0, havingrootsa and b. Multiplying
and thendividingout ab, we obtain
(l/ab)x2 - (1/a + l/b)x + 1 = 0;
theresultis nowobvious,as is theextensionto equationsof higherdegree.
Euler'ssolutionthenconsiderstheequationsinx = 0.
Expandingthisas an infinite series,Eulerobtained
x-x3/3! + x5/5! -O0.
Dividingby x yields
I - x2/3! + x4/5!-.. -0.
x2
Finally,substituting = u produces
1-u/3! + u2/5!-. = 0.
But Eulerthoughtthatpowerseriescould be manipulated just likepolynomials.
Thus,we now
have a polynomialequationin u, whoseconstanttermis one. Applyingthe lemmato it, the
of thelineartermwiththesignchangedis 1/3! = 1/6. The rootsof theequationin u
coefficient
u = x2, namely 2, 4,g2, 9r2 .
are therootsof sinx = 0 withthesubstitution Thus thelemma
implies
1/6 = /1 2 + l/4 r2 + 1/97T2 +
by r 2 yieldsthesumof theoriginalseries[5]:
Multiplying
1/1 + 1/4 + 1/9 + - - + Il/k2 + * = 7r2/6.
Thoughit is easyto criticizeeighteenth-centuryarguments likethisfortheirlack of rigor,it is
also unfair.Foundations, of theconditionsunderwhichsuchmanipulations
precisespecifications
withinfinitesor infinitesimals were admissible,were not veryimportantto men like Euler,
becausewithoutsuchspecifications theymade important new discoveries,whoseresultsin cases
like thiscould readilybe verified.Whenthe foundations of the calculuswerediscussedin the
eighteenth century, theyweretreatedas secondary.Discussionsof foundations appearedin the
introductions to books,in popularizations,and in philosophicalwritings,and werenot-as they
are nowand havebeensinceCauchy'stime-the subjectof articlesin research-oriented journals.
Thus,wherewe once had one questionto answer,we now havetwo.The firstremains,where
do Cauchy'srigoroustechniques comefrom?Second,one mustnowask,whyrigorizethecalculus
in the firstplace? If fewmathematicians wereveryinterested in foundations in the eighteenth
century [6],thenwhen,and why,wereattitudes changed?
Of course,to establishrigor,it is necessary-though not sufficient-tothinkrigoris signifi-

This content downloaded from 66.241.104.242 on Fri, 7 Nov 2014 00:25:17 AM

All use subject to JSTOR Terms and Conditions
188 JUDITH V. GRABINER [March

cant.But moreimportant, to establishrigor,it is necessary(thoughalso notsufficient)

to have a
setoftechniques whicharesuitableforthatpurpose.In particular,
in existence ifthecalculusis to
be maderigorousby beingreducedto thealgebraofinequalities, one musthaveboththealgebra
and factsabouttheconceptsof thecalculusthatcan be expressedin termsof the
of inequalities,
algebraof inequalities.
In theearlynineteenth century,threeconditionsheldforthefirsttime:Rigorwas considered
important; therewas a well-developed algebraofinequalities;and,certainproperties wereknown
about the basic concepts of analysis-limits, convergence, continuity,derivatives,
integrals-properties whichcouldbe expressedin thelanguageof inequalities ifdesired.Cauchy,
followedby Riemannand Weierstrass, gavethecalculusa rigorousbasis,usingthealready-exist-
ing algebra of inequalities,and built a logically-connected structureof theoremsabout the
conceptsof the calculus.It is our task to explainhow thesethreeconditions-thedeveloped
algebraof inequalities,theimportance of rigor,theappropriate of theconceptsof the
properties
calculus-came to be.
The Algebraof Inequalities.Today,the algebraof inequalitiesis studiedin calculuscourses
becauseofitsuse as a basisforthecalculus,butwhyshouldit havebeenstudiedin theeighteenth
centurywhenthis applicationwas unknown?In the eighteenth century, inequalitieswereim-
portantin the studyof a major class of results:approximations. For example,consideran
equationsuchas (x + 1) ' = a, for,unotan integer. Usuallya cannotbe foundexactly, butit can
be approximated by an infiniteseries.In general,givensome numbern of termsof such an
approximating series,eighteenth-century mathematicianssoughtto computean upperboundon
theerrorin theapproximation-that is, thedifferencebetweenthesumof theseriesand thenth
partialsum. This computation was a problemin the algebraof inequalities.Jeand'Alembert
solvedit fortheimportant case of thebinomialseries;giventhenumberof termsof theseriesn,
and assumingimplicitly thatthe seriesconvergesto its sum,he could findthe boundson the
error-thatis, on theremainder of theseriesafterthenthterm-by boundingtheseriesabove
and belowwithconvergent geometricprogressions Joseph-Louis
[7]. Similarly, Lagrangeinvented
a new approximation methodusingcontinuedfractionsand, by extremely intricateinequality-
gavenecessary
calculations, and sufficient fora giveniteration
conditions of theapproximation to
be closer to the resultthan the previousiteration[8]. Lagrangealso derivedthe Lagrange
remainder of theTaylorseries[9],usingan inequalitywhichboundedtheremainder above and
below by the maximumand minimumvalues of the nth derivativeand then applyingthe
intermediate-valuetheoremforcontinuousfunctions. Thusthrough sucheighteenth-century work
[10],therewas bytheendof theeighteenth a developedalgebraofinequalities,
century and people
used to working withit.Givenan n, thesepeopleareusedto finding an error-thatis,an epsilon.
ChangingAttitudestowardRigor.Mathematicians were much more interestedin finding
rigorousfoundations forthecalculusin 1800thantheyhad been a hundredyearsbefore.There
aremanyreasonsforthis:no one enoughbyitself, butapparently sufficient
whenactingtogether.
Of courseone mightthinkthateighteenth-century mathematicians were alwaysmakingerrors
becauseof thelack of an explicitly-formulatedrigorousfoundation. But thisdid notoccur.They
wereusuallyright,and fortworeasons.One is thatifone deals withrealvariables,functions of
one variable,serieswhichare powerseries,and functions arisingfromphysicalproblems,errors
will not occurtoo often.A secondreasonis thatmathematicians likeEuler and Laplace had a
deep insightinto the basic propertiesof the conceptsof the calculus,and wereable to choose
methodsand evade pitfalls.The only"error"theycommitted
fruitful was to use methodsthat
shockedmathematicians of laterageswhohad grownup withtherigorof thenineteenth century.
What then were the reasonsfor the deepenedinterestin rigor?One set of reasonswas
philosophical.In 1734,theBritishphilosopher BishopBerkeleyhad attackedthecalculuson the
groundthat it was not rigorous.In The Analyst,or a DiscourseAddressedto an Infidel
Mathematician, he said thatmathematicians had no businessattackingtheunreasonableness of

This content downloaded from 66.241.104.242 on Fri, 7 Nov 2014 00:25:17 AM

All use subject to JSTOR Terms and Conditions
1983] WHO GAVE YOU THE EPSILON? CAUCHY AND THE ORIGINS OF RIGOROUS CALCULUS 189

religion,giventhewaytheythemselves reasoned.He ridiculedfluxions-"velocitiesofevanescent

increments"-calling theevanescent increments "ghostsof departedquantities"[11].Even more
to the point,he correctly criticizeda numberof specificarguments fromthe writingsof his
mathematical contemporaries. For instance,he attackedtheprocessof findingthe fluxion(our
derivative) by reviewing thestepsof theprocess:if we considery = x2, takingtheratioof the
differences ((x + h)2 - x2)/h, thensimplifying to 2x + h, thenlettingh vanish,we obtain2x.
But is h zero?If it is, we cannotmeaningfully divideby it; if it is notzero,we have no rightto
throwit away.As Berkeleyput it,thequantitywe have calledh "mighthave signified eitheran
increment or nothing.But then,whichof thesesoeveryou make it signify, you mustargue
consistently withsuchitssignification" [12].
Sincean adequateresponseto Berkeley's objectionswouldhave involvedrecognizing thatan
equationinvolving limitsis a shorthand expressionfora sequenceof inequalities-a subtleand
difficultidea-no eighteenth-century analystgavea fullyadequateanswerto Berkeley. However,
manytried.Maclaurin,d'Alembert, Lagrange,Lazare Carnot,and possiblyEuler,all knewabout
Berkeley's work,and all wrotesomething aboutfoundations. So Berkeleydid call attention to the
question.However,exceptforMaclaurin,no leadingmathematician spentmuch timeon the
questionbecauseof Berkeley's work,and evenMaclaurin'sinfluence lay in otherfields.
Anotherfactorcontributing to the new interestin rigorwas thattherewas a limitto the
numberof resultsthatcould be reachedby eighteenth-century methods.Near the end of the
century, someleadingmathematicians had begunto feelthatthislimitwas at hand.D'Alembert
and Lagrangeindicatethisin theircorrespondence, withLagrangecallinghighermathematics
"decadent"[13].The philosopher Diderotwentso faras to claimthatthemathematicians of the
eighteenth century had "erectedthepillarsof Hercules"beyondwhichit was impossibleto go
[14].Thus,therewas a perceivedneed to consolidatethegainsof thepast century.
Another"factor"was Lagrange,who became increasingly interestedin foundations, and,
throughhis activities,interestedothermathematicians. In the eighteenth century,scientific
academiesofferedprizesforsolvingmajor outstanding problems.In 1784, Lagrangeand his
colleaguesposed the problemof foundationsof the calculus as the BerlinAcademy'sprize
problem.Nobodysolvedit to Lagrange'ssatisfaction, but twoof theentriesin thecompetition
werelaterexpandedinto full-length books,the firston the Continent, on foundations:Simon
L'Huilier'sExpositionelemrentaire des principesdes calculssuperieurs, Berlin,1787,and Lazare
Carnot'sReflexions surla metaphysique du calculinfinitesimal,Paris,1797.Thus Lagrangeclearly
helpedreviveinterest in theproblem.
Lagrange'sinterest stemmedin partfromhis respectforthepowerand generality of algebra;
he wantedto gain forthe calculusthe certainty he believedalgebrato possess.But therewas
anotherfactorincreasinginterestin foundations, not only forLagrange,but formanyother
mathematicians by theend of the eighteenth century:the need to teach.Teachingforcesone's
attentionto basic questions.Yet beforethe mideighteenth century, mathematicians had often
made theirlivingby beingattachedto royalcourts.But royalcourtsdeclined;the numberof
mathematicians increased;and mathematics began to look useful.Firstin military schoolsand
lateron at the Ecole Polytechnique in Paris,anotherline of workbecame available:teaching
mathematics to students of scienceand engineering. The Ecole Polytechnique was foundedby the
Frenchrevolutionary government to trainscientists, who,thegovernment believed,mightprove
usefulto a modernstate.And it was as a lecturerin analysisat the Ecole Polytechnique that
Lagrangewrotehis twomajorworkson thecalculuswhichtreatedfoundations; similarly,it was
40 yearsearlier,teachingthecalculusat theMilitaryAcademyat Turin,thatLagrangehad first
setoutto workon theproblemof foundations. Becauseteachingforcesone to ask basic questions
about thenatureof themostimportant concepts,thechangein theeconomiccircumstances of
mathematicians-the needto teach-provideda catalystforthecrystallization of thefoundations
of the calculusout of the historicaland mathematical background.In fact,even well into the
nineteenth century,much of foundationswas born in the teachingsituation;Weierstrass's

This content downloaded from 66.241.104.242 on Fri, 7 Nov 2014 00:25:17 AM

All use subject to JSTOR Terms and Conditions
190 JUDITH V. GRABINER [March

foundationscomefromhislecturesat Berlin;Dedekindfirstthoughtoftheproblemofcontinuity
whileteachingat Zurich;Dini and Landau tumedto foundations whileteachinganalysis;and,
mostimportant forourpresentpurposes,so did Cauchy.Cauchy'sfoundationsof analysisappear
in thebooks based on his at
lectures the Ecole his
Polytechnique; book of 1821 was the first
exampleof thegreatFrenchtradition of Coursd'analyse.

The Conceptsof theCalculus.Arisingfromalgebra,thealgebraof inequalities was now there

forthecalculusto be reducedto; the desireto make thecalculusrigoroushad arisenthrough
consolidation, through philosophy,throughteaching,throughLagrange.Now let us turnto the
mathematical substanceof eighteenth-century analysis,to see whatwas knownabouttheconcepts
of thecalculusbeforeCauchy,and whathe had to workout forhimself, in orderto define,and
provetheorems about,limit,convergence, continuity, and integrals.
derivatives,
First,considertheconceptoflimit.As we havealreadypointedout,sinceNewtonthelimithad
been thought of as a boundwhichcould be approachedcloserand closer,thoughnot surpassed.
By 1800,withtheworkof L'Huilierand Lacroixon alternating series,therestriction thatthelimit
be one-sidedhad been abandoned.Cauchysystematically translatedthisrefinedlimit-concept
intothealgebraof inequalities, and usedit in proofsonceit had beenso translated; thushe gave
realityto theoft-repeated eighteenth-centurystatement thatthecalculuscouldbe based on limits.
For example,considertheconceptof convergence. Maclaurinhad said alreadythatthesumof
a serieswas thelimitofthepartialsums.For Cauchy,thismeantsomething precise.It meantthat,
givenan e, one could findn such that,formorethann terms,thesum of theinfinite seriesis
within- of the nth partialsum. That is the reverseof the error-estimating procedurethat
d'Alembert had used. Fromhis definition of a serieshavinga sum,Cauchycould provethata
geometric progression withradiusless in absolutevaluethan1 converged to its usualsum.As we
have said,d'Alembert had shownthatthebinomialseriesfor,say,(1 + x)P/q could be bounded
above and below by convergent geometricprogressions. Cauchy assumedthat if a seriesof
positivetermsis boundedabove,term-by-term, by a convergent geometric progression,thenit
converges; he then used such comparisons to prove a number of testsfor convergence: the root
test,theratiotest,thelogarithm test.The treatment is quiteelegant[15].Takinga techniqueused
a fewtimesby men like d'Alembertand Lagrangeon an ad hoc basis in approximations, and
usingthedefinition of thesum of a seriesbased on thelimit-concept, Cauchycreatedthefirst
rigoroustheoryof convergence.
Let us nowturnto theconceptof continuity. Cauchygaveessentially themoderndefinition of
continuousfunction, sayingthatthefunction f(x) is continuouson a givenintervalifforeach x
in thatinterval"the numerical[i.e.,absolute]value of thedifference f(x + a) -a(x) decreases
indefinitely a"
with [16]. He used this definition in proving theintermediate-value theoremfor
continuousfunctions [17].The proofproceedsby examininga function f(x) on an interval, say
[b, c], wheref(b) is negative,f(c) positive,and dividingtheinterval[b, c] intom partsof width
h = (c - b)/m. Cauchyconsideredthe sign of the functionat thepointsf(b), f(b + h),...,
f(b + (m - l)h), f(c); unlessone of thevaluesoff is zero,thereare twovaluesofx differing by
h suchthatf is negativeat one,positiveat theother.Repeatingthisprocessfornewintervals of
width(c - b)/m,(c - b)/m2,..., givesan increasingsequenceof valuesof x: b, b1,b2,... for
whichf is negative,and a decreasing sequenceof valuesof x: c, cl, c2,. . . forwhichf is positive,
and such that the difference between bk and Ck goes to zero. Cauchyassertedthatthesetwo
sequencesmusthave a commonlimita. He then arguedthatsincef(x) is continuous,the
sequenceof negativevaluesf(bk) and of positivevaluesf(Ck) bothconvergetowardthecommon
limitf(a), whichmusttherefore be zero.
Cauchy's proof involvesan alreadyexistingtechnique,which Lagrange had applied in
approximating realrootsof polynomialequations.If a polynomialwas negativeforone value of
thevariable,positiveforanother,therewas a rootin between,and thedifference betweenthose
twovaluesof thevariableboundedtheerrormade in takingeitheras an approximation to the

This content downloaded from 66.241.104.242 on Fri, 7 Nov 2014 00:25:17 AM

All use subject to JSTOR Terms and Conditions
1983] WHO GAVE YOU THE EPSILON? CAUCHY AND THE ORIGINS OF RIGOROUS CALCULUS 191

root [18]. Thus again we have the algebraof inequalitiesprovidinga techniquewhichCauchy

transformed froma toolof approximation to a toolof rigor.
It is worthremarking at thispointthatCauchy,in his treatment bothof convergence and of
continuity, assumedvariousformsof thecompleteness
implicitly property forthereal numbers.
For instance, he treatedas obviousthata seriesofpositiveterms, boundedaboveby a convergent
geometric progression,converges: also,hisproofof theintermediate-value theorem assumesthata
boundedmonotonesequencehas a limit.WhileCauchywas the firstsystematically to exploit
inequalityprooftechniques to prove theorems in analysis,he did not identifyall the implicit
assumptionsabout the real numbersthatsuch inequalitytechniquesinvolve.Similarly, as the
readermayhavealreadynoticed,Cauchy'sdefinition of continuous function does notdistinguish
betweenwhatwe nowcall point-wise and uniform continuity;also,in treatingseriesof functions,
Cauchydid notdistinguish betweenpointwiseand uniform convergence. The verbalformulations
like"forall" thatareinvolvedin choosingdeltasdidnotdistinguish between"foranyepsilonand
forall x" and "foranyx, givenanyepsilon"[19].Nor was it at all clearin the1820'showmuch
dependedon thisdistinction, sinceproofsaboutcontinuity and convergence werein themselves so
novel.We shallsee thesameconfusion betweenuniform and point-wise convergence as we turn
now to Cauchy'stheoryof thederivative.
Again we beginwithan approximation. Lagrangegave the followinginequalityabout the
derivative:
(2) f (x + h) = f (x) + hf'(x) + hV,
where V goes to 0 withh. He interpreted this to mean that,givenany D, one can findh
sufficiently is
smallso thatV between - D and + D [20].Clearlythisis equivalentto (1) above,
Cauchy'sdelta-epsilon characterization Buthowdid Lagrangeobtainthisresult?
of thederivative.
The answeris surprising; forLagrange,formula(2) was a consequenceof Taylor'stheorem.
Lagrangebelievedthatany function (thatis, any analyticexpression,whetherfiniteor infinite,
involvingthevariable)had a uniquepower-series expansion(exceptpossiblyat a finitenumberof
isolatedpoints).This is because he believedthattherewas an "algebraof infiniteseries,"an
algebraexemplified byworkofEulersuchas theexamplewe gaveabove.And Lagrangesaid that
thewayto makethecalculusrigorouswas to reduceit to algebra.Althoughthereis no "algebra"
ofinfiniteseriesthatgivespower-series expansionswithoutanyconsideration of convergence and
limits,thisassumption led Lagrangeto definef'(x) withoutreference to limits,as thecoefficient
of thelineartermin h in theTaylorseriesexpansionforf(x + h). FollowingEuler,Lagrange
thensaid that,foranypowerseriesin h, one couldtakeh sufficiently smallso thatanygiventerm
of theseriesexceededthesumof all therestof thetermsfollowing it; thisapproximation, said
Lagrange,is assumedin applicationsof thecalculusto geometry and mechanics[21].Applying
thisapproximation to thelineartermin theTaylorseriesproduces(2), whichI call theLagrange
property of thederivative. (Like Cauchy's(1), theinequality-translationLagrangegivesfor(2)
assumesthat,givenany D, one finds
h small
sufficiently so lVI< D with no mentionwhatever of
x.)
Not onlydid Lagrangestateproperty he used themas a
(2) and theassociatedinequalities,
basisfora numberofproofsaboutderivatives: forinstance,to provethata function
withpositive
derivative on an interval there,to provethemean-valuetheorem
is increasing and
forderivatives,
to obtaintheLagrangeremainder fortheTaylorseries.(Details maybe foundin theworkscited
in [22].)Lagrangealso appliedhis resultsto characterize
theproperties of maximaand minima,
and ordersof contactbetweencurves.
Witha fewmodifications, Lagrange'sproofsare valid-provided thatproperty(2) can be
justified.Cauchyborrowedand simplified whatare in effectLagrange'sinequalityproofsabout
derivatives,witha fewimprovements, basingthemon hisown(1). But Cauchymadetheseproofs
legitimate because Cauchydefinedthe derivativepreciselyto satisfythe relevantinequalities.
Once again,thekeyproperties come froman approximation. was
For Lagrange,thederivative

This content downloaded from 66.241.104.242 on Fri, 7 Nov 2014 00:25:17 AM

All use subject to JSTOR Terms and Conditions
192 JUDITH V. GRABINER [March

exactly-no epsilonsneeded-the coefficient of thelineartermin theTaylorseries;formula(2),

and the corresponding inequalitythatf(x + h) - f(x) lies betweenh(f'(x) ? D), were ap-
proximations. Cauchy broughtLagrange'sinequalitypropertiesand proofstogetherwith a
definition devisedto makethosetechniques
of derivative rigorouslyfounded[22].
The lastof theconceptswe shallconsider, theintegral,followedan analogousdevelopment. In
theeighteenth theintegralwas usuallythought
century, of as theinverseof thedifferential. But
sometimes theinversecouldnotbe computedexactly, so menlikeEulerremarked thattheintegral
could be approximated as closelyas one likedby a sum.Of course,thegeometric pictureof an
area beingapproximated by rectangles, or the Leibniziandefinition of the integralas a sum,
suggeststhisimmediately. Butwhatis important forourpurposesis thatmuchworkwas doneon
approximating thevaluesof definite integrals in theeighteenth century,includingconsiderations
of howsmallthesubintervals usedin thesumsshouldbe whenthefunction oscillatesto a greater
or lesserextent.For instance,Euler treatedsums of the formE'2Iof(xk)(xk? - x,) as ap-
proximations to theintegralfxxnf(x) dx [23].
In 1820, S.-D. Poisson,who was interestedin complexintegrationand therefore more
concernedthanmostpeople about theexistenceand behaviorof integrals, asked thefollowing
question.If theintegral
F is definedas theantiderivative off, and ifb - a = nh,can it be proved
thatF(b) - F(a) = abff(x)dx is thelimitof thesum
S = hf(a) + hf(a + h) + + hf(a + (n - I)h)
as h getssmall?(S is an approximatingsumof theeighteenth-century sort.)Poissoncalled this
result"the fundamental of the theoryof definiteintegrals."He provedit by using
proposition
theTaylorserieswithremainder.
anotherinequality-result: First,he wroteF(b) - F(a) as the
telescopingsum

(3) F(a + h)-F(a) + F(a + 2h)-F(a + h) + + F(b)-F(a + (n- )h).

Then, foreach of the termsof the formF(a + kh) - F(a + (k -1) h), Taylor'sserieswith
remainder F' = f,
gives,sinceby definition
F(a + kh) - F(a + (k - I)h) = hf(a + (k - I)h) + Rkhl+w
wherew > 0, forsomeRk* Thus thetelescoping
sum(3) becomes
hf(a) + hf(a + h) + + hf(a + (n- +
I)h) + (R1+IR)hl?w
So F(b) - F(a) and thesumS differ
by(R1 + + Rn)h' +w. LettingR be themaximum
value
fortheRk,
(RI + + Rn)hl+w < n R(hl+w) = R nh hw= R(b-a)hW
Therefore, small,F(b) - F(a) differs
if h is takensufficiently fromS by less thanany given
quantity[24].
Poisson'swas thefirstattempt to provetheequivalenceof theantiderivative and limit-of-sums
conceptionsof the integral.However,besides the implicitassumptionsof the existenceof
antiderivatives and boundedfirstderivatives
forf on thegiveninterval, theproofassumesthatthe
subintervals on whichthe sum is takenare all equal. Should the resultnot hold forunequal
divisionsalso? Poissonthought so, andjustifiedit by saying,"If theintegral
is representedby the
area of a curve,thisareawillbe thesame,ifwe dividethedifference... intoan infinite
numberof
equal parts,or an infinitenumberof unequalpartsfollowing anylaw" [25].This,however, is an
assertion, not a proof.And Cauchysaw thata proofwas needed.
Cauchydid not like formalistic arguments in supposedlyrigoroussubjects,sayingthatmost
algebraicformulashold "onlyundercertainconditions, and forcertainvaluesof thequantities
theycontain"[26]. In particular, one could not assumethatwhatworkedforfiniteexpressions
automaticallyworkedfor infiniteones. Thus, Cauchy showed that the sum of the series

This content downloaded from 66.241.104.242 on Fri, 7 Nov 2014 00:25:17 AM

All use subject to JSTOR Terms and Conditions
1983] WHO GAVE YOU THE EPSILON? CAUCHY AND THE ORIGINS OF RIGOROUS CALCULUS 193

I/1 + 1/4 + 1/9 + was 7r2/6by actuallycalculatingthedifferencebetweenthenthpartial

sumand 7T2/6and showingthatit was arbitrarily just becausetherewas an
small[27].Similarly,
operationcalled takinga derivativedid not mean that the inverseof thatoperationalways
produceda result.The existenceof thedefiniteintegralhad to be proved.And how does one
proveexistencein the 1820's?One constructs themathematical objectin questionby usingan
approximation
eighteenth-century thatconverges to it. Cauchydefinedtheintegralas thelimitof
sumsY2f(xk)(xk+I
Euler-style - Xk) forXk+I - Xk sufficientlysmall.Assumingexplicitly that
f(x) was continuouson the giveninterval(and implicitly thatit was uniformlycontinuous),
Cauchywas able to showthatall sumsof thatformapproacha fixedvalue,calledby definition
the integralof the functionon thatinterval.This is an extremely hard proof [28]. Finally,
borrowingfromLagrangethemean-valuetheorem CauchyprovedtheFundamental
forintegrals,
Theoremof Calculus[29].
Conclusion.Here are all the pieces of the puzzle we originallyset out to solve. Algebraic
approximations producedthe algebraof inequalities;eighteenth-century approximations in the
calculusproducedtheusefulproperties of theconceptsof analysis:d'Alembert's error-bounds for
series,Lagrange'sinequalitiesaboutderivatives,Euler'sapproximations Therewas a
to integrals.
newinterest in foundations.All thatwas neededwas a sufficiently greatgeniusto buildthenew
foundation.
Two men came close. In 1816, Carl FriedrichGauss gave a rigoroustreatment of the
convergenceof the hypergeometric series,using the techniqueof comparinga series with
convergent geometric progressions;however,Gauss did not givea generalfoundation forall of
analysis.BernhardBolzano,whoseworkwas littleknownuntilthe1860's,echoingLagrange'scall
to reducethecalculusto algebra,gavein 1817a definition of continuousfunction likeCauchy's
and thenproved-by a different techniquefromCauchy's-the intermediate-value theorem[30].
But it was Cauchywho gaverigorousdefinitions and proofsforall thebasic concepts;it was he
whorealizedthefar-reaching powerof theinequality-based limitconcept;and it was he whogave
us-except for a few implicitassumptionsabout uniformity and about completeness-the
modemrigorousapproachto calculus.
Mathematicians are used to takingthe rigorousfoundations of the calculusas a completed
whole.WhatI have triedto do as a historianis to revealwhatwentintomakingup thatgreat
achievement. Thisneedsto be done,becausecompletedwholesby theirnaturedo notrevealthe
separatestrandsthatgo intoweavingthem-especiallywhenthestrandshavebeen considerably
transformed. In Cauchy'swork,though, one traceindeedwas leftoftheoriginofrigorous calculus
in approximations-the letterepsilon.The e correspondsto theinitialletterin theword"erreur"
(or "error"),and Cauchyin factused - for"error"in someof his workon probability [31].It is
both amusingand historically appropriatethatthe "s," once used to designatethe "error"in
approximations, has becometransformed intothecharacteristicsymbolof precisionand rigorin
thecalculus.As Cauchytransformed thealgebraofinequalities froma toolof approximation to a
toolof rigor,so he transformed thecalculusfroma powerful methodof generating resultsto the
rigoroussubjectwe knowtoday.

References
1. A. -L. Cauchy,Coursd'analyse, Paris,1821;in Oeuvres completesd'AugustinCauchy, series2, vol.3, Paris,
1899,p. 19.
Gauthier-Villars,
2. A. -L. Cauchy,Resumedes lesonsdonneesa l'ecoleroyalepolytechnique surle calculinfinitesimal, Paris,
1823;in Oeuvres, series2, vol.4, p. 44. Cauchyusedi fortheincrement; otherwisethenotation is his.
3. IsaacNewton, Mathematical Principles
ofNaturalPhilosophy, 3rded.,1726,tr.A. Motte,revised byFlorian
Cajori,UniversityofCalifornia Press,Berkeley, 1934,Scholium to LemmaXI, p. 39.
4. Johann OperaOmnia,IV, 8; sectionentitled
Bernoulli, "De seriebusvaria,Corollarium III," citedbyD. J.
Struik,A SourceBookin Mathematics, 1200-1800,Harvard,Cambridge, 1969,p. 321.
5. Boyer, HistoryofMathematics, p. 487;Euler'spaperis inComm.Acad.Sci.Petrop., 7, 1734-5,pp. 123-34;

This content downloaded from 66.241.104.242 on Fri, 7 Nov 2014 00:25:17 AM

All use subject to JSTOR Terms and Conditions
194 JUDITH V. GRABINER

in LeonhardEuler,Opera omnia,series 1, vol. 14, pp. 73-86.

6. J. V. Grabiner,The Originsof Cauchy's RigorousCalculus, M. I. T. Press,Cambridgeand London, 1981,
chapter2.
7. J. d'Alembert,Reflexionssur les suiteset sur les racinesimaginaires,in Opuscules mathematiques, vol. 5,
Briasson,Paris, 1768,pp. 171-215; see especiallypp. 175-178.
8. J.-L. Lagrange,Traitede la resolutiondes equationsnumeriquesde tousles degres,2nd ed., Courcier,Paris,
1808; in Oeuvresde Lagrange,Gauthier-Villars, Paris, 1867-1892,vol. 8, pp. 162-163.
9. Lagrange,Theoriedes fonctionsanalytiques,2nd ed., Paris, 1813,in Oeuvres,vol. 9, pp. 80-85; compare
Lagrange,Lecons sur le calcul des fonctions,Paris, 1806,in Oeuvres,vol. 10,pp. 91-95.
10. Grabiner,Origins of Cauchy's Rigorous Calculus, pp. 56-68; compare H. Goldstine, A History of
NumericalAnalysisfromthe 16ththroughthe 19thCentury,Springer-Verlag, New York, Heidelberg,Berlin,1977,
chapters2-4.
11. GeorgeBerkeley,The Analyst,section35.
12. Analyst,section 15. Berkeleyused the functionXn wherewe have used x2, and a Newtoniannotation,
lower-caseo, fortheincrement.
13. LetterfromLagrangeto d'Alembert,24 February1772,in Oeuvresde Lagrange,vol. 13, p. 229.
14. D. Diderot, De l'interpretation de la nature,in Oeuvres philosophiques,ed., P. Verniere,Garnier,Paris,
1961,pp. 180-181.
15. Cauchy,Cours d'analyse,Oeuvres,series2, vol. 3; forreal-valuedseries,see especiallypp. 114-138.
16. Cauchy,op. cit.,p. 43. So did Bolzano; see below,and note 30.
17. Cauchy,op. cit.,pp. 378-380. For an Englishtranslationof thisproof,see Grabiner,Origins,pp. 167-168.
For clarity,I have substitutedb, b1,b2,... and c, c1,c2,... forCauchy's x0, x1, x2,... and X, X', X",... in the
presentversion.
18. Lagrange,Equationsnumeriques,sections2 and 6, in Oeuvres,vol. 8; also in Lagrange,Lecons elementaires
sur les mathematiquesdonnees a l'&cole normaleen 1795, Seances des Ecoles Normales, Paris, 1794-1795; in
Oeuvres,vol. 7, pp. 181-288; thismethodis on pp. 260-261.
19. I. Grattan-Guinness, Developmentof the Foundationsof MathematicalAnalysisfromEuler to Riemann,
M. I. T. Press,Cambridgeand London, 1970,p. 123,puts it well: "Uniformconvergencewas tuckedaway in the
word "always," withno referenceto thevariableat all."
20. Lagrange,Leoons sur le calcul des fonctions,Oeuvres 10, p. 87; compareLagrange,Theoriedes fonctions
analytiques,Oeuvres9, p. 77. I have substitutedh forthe i Lagrangeused fortheincrement.
21. Lagrange,Theoriedes fonctionsanalytiques,Oeuvres9, p. 29. Compare Lesons sur le calcul des fonctions,
Oeuvres10,p. 101. For Euler,see his Institutiones St. Petersburg,1755; in Opera, series1, vol.
calculi differentialis,
10, section122.
22. Grabiner,Originsof Cauchy's RigorousCalculus, chapter5; also J. V. Grabiner,The originsof Cauchy's
theoryof thederivative,Hist. Math.,5, 1978,pp. 379-409.
23. The notationis modernized.For Euler,see Institutiones calculiintegralis,St. Petersburg,1768-1770,3 vols;
in Opera, series 1, vol. 11, p. 184. Eighteenth-century summationsapproximatingintegralsare treatedin A. P.
Iushkevich,0 vozniknoveniya poiyatiyaob opredelennomintegraleKoshi, TrudyInstitutaIstoriiEstestvoznaniya,
AkademiaNauk SSSR, vol. 1, 1947,pp. 373-411.
24. S. D. Poisson, Suite du memoiresur les integralesdefinies,Joum. de l'Ecole polytechnique,Cah. 18, 11,
1820,pp. 295-341, 319-323. I have substitutedh, w forPoisson's a, k, and have used R1 forhis R0.
25. Poisson,op. cit.,pp. 329-330.
26. Cauchy,Cours d'analyse,Introduction, Oeuvres,Series2, vol. 3, p. iii.
27. Cauchy,Cours d'analyse,Note VIII, Oeuvres,series2, vol. 3, pp. 456-457.
28. Cauchy,Calcul infinitesimal, Oeuvres,series2, vol. 4, 122-25; in Grabiner,Originsof Cauchy's Rigorous
Calculus,pp. 171-175 in Englishtranslation.
29. Cauchy,op. cit.,pp. 151-152.
30. B. Bolzano, Rein analytischerBeweis des Lehrsatzesdass zwischenje zwey Werthen,die ein entgegenge-
setztesResultatgewaehren,wenigstenseine reele Wurzelder Gleichungliege,Prague, 1817. Englishversion,S. B.
Russ, A translationof Bolzano's paper on theintermediate value theorem,Hist. Math., 7, 1980,pp. 156-185. The
contentionby Grattan-Guinness, Foundations,p. 54, thatCauchytook his programof rigorizing analysis,definition
of continuity,Cauchy criterion,and proof of the intermediate-value theorem,from Bolzano's paper without
acknowledgement is not, in my opinion,valid; the similaritiesare betterexplainedby commonpriorinfluences,
especiallythatof Lagrange.For a documentedargumentto thiseffect,see J. V. Grabiner,Cauchy and Bolzano:
Traditionand transformation in the historyof mathematics,to appear in E. Mendelsohn,Transformation and
Traditionin the Sciences, CambridgeUniversityPress, Cambridge,forthcoming; see also Grabiner,Originsof
Cauchy's RigorousCalculus,pp. 69-75, 102-105, 94-96, 52-53.
31. Cauchy,Sur la plus grandeerreura craindredans un resultatmoyen,et sur le systemede facteursqui rend
cetteplus grandeerreurun minimum,Comptesrendus37, 1853; in Oeuvres,series 1, vol. 12,pp. 114-124.

This content downloaded from 66.241.104.242 on Fri, 7 Nov 2014 00:25:17 AM

All use subject to JSTOR Terms and Conditions