Bringing Calculus Up-to-Date Author(s): M. E. Munroe Source: The American Mathematical Monthly, Vol. 65, No. 2 (Feb.

, 1958), pp. 81-90 Published by: Mathematical Association of America Stable URL: http://www.jstor.org/stable/2308879 . Accessed: 01/01/2014 09:37
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 128.59.62.83 on Wed, 1 Jan 2014 09:37:05 AM All use subject to JSTOR Terms and Conditions

BRINGING CALCULUS UP-TO-DATE

ofIllinois M. E. MUNROE, University

1. Introduction.There are many conversations,committee meetings,etc., today about the modernizationof the undergraduatecalculus course; but all Calculus too oftenthe attack on the problemfallsshortof being comprehensive. years now. These have been highlyprohas been in cold storage forover fifty ductive years in mathematics,and the result is that more changes are in order than most people would like to admit. to be real funcThe relevant branches of modernmathematicswould seemn geometry,and a survey of these fieldssuggests at tion theoryand differential that point (i) can be gleaned least the fivepoints listed below. It is significant froma study of real functiontheoryand is already coming into vogue in calgeometryand are too culus, while points (ii)-(v) come largelyfromdifferential generallyoverlooked. Calculus is too often erroneouslyclassifiedas a part of real functiontheoryonly. f and its values f(a). (i) It is essential to distinguishbetween a function (ii) The coordinate variables x and y of analytic geometryare themselves mappings; x maps points into theirabscissas; y maps points into theirordinates. in point of view toward coordinate (iii) There is an importantdifference in variables geometry.In analytic geometryx and y analytic and differential are definedover the entireplane, and y =f(x) is a conditionalequation describing a locus. In differential geometrythe locus is preassigned; x and y are reto this stricted locus, and y =f(x) is an identity.A calculus problem starts as geometry.Thus, somewhere in analytic geometryand finishesas differential the meaning. (More details on this point in Section the middle symbolschange 2.) of the geometryhas finallyproduced a definition (iv) Modern differential forpurposesof integration differential that is quite satisfactory and differentiathe concept of differential has been divorced from tion theory.In this definition that of approximateincrement.(Details in Section 3.) formsthe (v) In the modern,advanced study of integrationof differential the algebra of the theoryof the integralis liftedfromreal functiontheory,and is shown to fitinto this framework. forms When this abstract theory differential is specialized to the simplified,concrete cases discussed in calculus, modern differential (and simpler) tool forcertain developtheorybecomes an effective ments in the theoryof the integral. (See Sections 5 and 6.) 2. Functions and variables. To begin with, define the word function to mean a mapping of numbersinto numbers. A later generalizationwill include

81

This content downloaded from 128.59.62.83 on Wed, 1 Jan 2014 09:37:05 AM All use subject to JSTOR Terms and Conditions

82

BRINGING CALCULUS UP-TO-DATE

[February

mappingsof n-tuplesof numbersinto numbers.This is the traditionaluse of the word functionin calculus. One also encountersspaces whose elementsare called points.These take the formof lines, planes, curves, surfaces, 3-space, space-time, etc. An essential part of the machineryis the mapping of these points into numbers. Let these mappings be called variables.Linguisticallyincongruousas it may seem, this is the most commonuse of the word variable in calculus. This definitraditionally tion followsthe principlethat language is established by usage, not by logic. descriptionof what it has been It retainsthe word and gives a moreenlightened used to mean. It should be noted that probabilitytheoryhas already established this usage with the phrase randomvariable. The coordinatevariables x and y are variables in the sense just defined.In plane analytic geometryeach of these symbols stands for a mapping of the entireplane onto the real numbersystem.If x has this meaningand f is a function, thenf(x) is used to denote a composite mapping: x f point-* number anothernumber. The form
(1) y
=

f(x)

equation.Here thereare two symbols,y and f(x), each standis now a conditional ing fora mapping; but in analytic geometry(1) is not regardedas an assertion that these are two names forthe same mapping. Rather, (1) appears (or should appear) only as a noun clause in the phrase, "the locus of y =f(x)." This locus is, of course, the set of all points p in the commondomain of x and y forwhich y(p) =f[x(p) ]. Note that this last "=" means what it should. Suppose, now, that (1) has been given in proper context and its locus has been found and named C. The next step in the logical analysis of a calculus problem is to restrictthe mappings x and y to C. Note that this could not be described above was used to define C. done beforebecause the superstructure mappingswere given new names; say, start with It would help if the restricted X and Y and boil them down to x and y. However, this is probably asking for too much of a change in set habits. In any case, with x and y restrictedto C, (1) becomes an identity-an assertion that two mappings are the same. At this studied in moderndifferential stage one has a special case of the type structure of that disciplineforfurther enlightenand so can turnto the literature geometry ment. What followsis an informalsummaryof the specialization of this work to calculus. For a more completediscussionof the general theorysee, forexample, Chevalley, Theoryof Lie Groups,Princeton,1946. 3. Differentials.A one-dimensional manifoldis a structureconsistingof a variablesdefinedthereon.Precise postulates point set C and a set of coordinate forthe presentdiscussion. can be given, but a very loose descriptionwill suffice The set C is a smoothcurve, and each of the coordinatevariables is continuous

This content downloaded from 128.59.62.83 on Wed, 1 Jan 2014 09:37:05 AM All use subject to JSTOR Terms and Conditions

1958]

BRINGING CALCULUS UP-TO-DATE

83

and locally one-to-one.Since each is locally one-to-one,each pair u, v is related at least locally by an identityof the form v=f(u), and it is convenientto assume that each such connectingfunction f has threederivatives.Note that a locus in the plane with the rectangularcoordinatevariables restricted to it will generally formsuch a structure. Consider threeclasses of variables on C. X: all variables on C, Y: those related to a coordinate by a differentiable function, Z: those related to a coordinate by a thrice-differentiable function. Let D be a mapping of Y into X, and let Du denote the map of u by D. If p is a point on the curve, denote the value of Du at p by (Du),. Such a mapping D is called a derivative operator at p provided (a) [D(au + bv)], = a(Du)p + b(Dv)p, (b) [D(uv)i] = u(p)(Dv)p + v(p)(Du)p. Here u and v are variables and a and b real numbers. Regarding a constant a as a variable, observe that (2) (Da)p = 0, because by (a), [D(au)]p=a(Du)p, and by (b), [D(au)J,=a(Du)p+u(p)(Da)p; and if u is chosen so that u(p) #0, then (2) follows.Observe also that setting u=v in (b) yields (3) [D(u2)]p
=

2u(p)(Du)p.

Now, let u be a coordinate,and let v=f(u). Assumef has threederivatives; then thereis a differentiable function g such that

(4) (5)

v = v(p) +f'[u(p)][u

u(p)] + g(u)[u -

u(p)]2.

Recalling (2) and (3), operate on (4) with D at p to obtain

(Dv)p = f'[u(p)](Du) p.

Each of the other termson the rightvanishes eitherbecause of (2) or because it contains the factoru(p) - u(p). Note that in this step D must be definedon Y because it must operate on g(u). However, (5) is proved only forvEZ because this is required in order to set up (4). Let D. be the operator such that D.u = 1 at each point on the curve. By (5) this determinesD. on v where v=f(u), and indeed

(6)

D.v = f'(u).

The variable D.v is called the derivative of v withrespectto u. The notationf' denotes a purelyfunction theoretic concept; f' is definedfrom f in termsof limits of difference quotients. On the other hand, D. is an operator definedby purely

This content downloaded from 128.59.62.83 on Wed, 1 Jan 2014 09:37:05 AM All use subject to JSTOR Terms and Conditions

84

BRINGING CALCULUS UP-TO-DATE

[February

algebraic conditions.Thus, (6) is not a definition; it is a theoremrelatingthese two basically different notions. Consider now D1 and D2, two derivativeoperatorsat p. If forsome variable u, (Diu),=a(D2u),, then forv=f(u), (Div) = f'[u(p) ] (Dlu)
= af'[u(p) ] (D2u)p
=

a(D2v),.

That is, at p any two derivative operators are constant multiples one of the other. So, the set of derivative operators at p has the algebraic structureof a straightline. Identifyit with T., the tangentline to the curve at p, and define the tangent bundleforthe given curve as the set of all orderedpairs (p, D) where p is a point of the curve and D is a derivativeoperatorat p. Finally, the differential is definedas follows.For u a variable on the curve,du is a variable on the tangent bundle. Its value at (p, D) is denoted by du,(D), and du is definedby (7) dup(D) = (Du)p. Substitute (6) into (5) to obtain
(8) (Dv), = (Duv)v(Du)v.

so (8) may be By (7), (Dv), and (Du)p are values of appropriate differentials; written
dvp(D) = (Duv)vduv(D).

This is foran arbitraryp on the curve and D on Tp; so

(9) dv = Duvdu

over the tangentbundle. An appropriatename for(9) is "Fundamental Theorem on Differentials." not a revolvAt any rate, in the moderntheoryit is a theorem, ing definition that changes meaning with every change of "independent"variable. Multidimensionalcases followthe same pattern. Let C be an n-dimensional manifold-intuitively, a uniformly n-dimensional,smooth set with variables attached. The basic form relating variables will be v =f(u1, * , us). Derivative i and by the n-dimensional operatorsare definedby the same postulates as before, vector generalizationof (4) it is shown that they forma familyof n-dimensional spaces. A derivativeoperator is determinedif specifiedon n variables, and the operatorsd/duidefinedby 0uj Oui

f1 for j= i

tO for j - i

are definedagain by (7), and forma basis in the tangent bundle. Differentials the fundamentaltheoremreads

This content downloaded from 128.59.62.83 on Wed, 1 Jan 2014 09:37:05 AM All use subject to JSTOR Terms and Conditions

1958] (10)

BRINGING CALCULUS UP-TO-DATE

85

dv=

n
E

av

dui.

4. Some immediate advantages. The ultimate advantage of saying things correctlyrather than incorrectly need not be discussed. Let it be understood, then, that the following list-which is quite probably nonrepresentativemerelypoints out a few of the common difficulties that can be cleared up by the modernapproach. withvariables.A variable v on C generatesa differ(a) Differentials identified ential dv on the tangentbundle,and this differential retainsits identitywhether v =f(u) or v= g(w). Note that df and dg are not defined.Understandingof this will help to clear up many notation obscurities. No one seems to know exactly what these ought to (b) Second differentials. be. Function theorists sometimes want d2v-D.vdu2; at other times (when changes of variable are in order) they want d2v=D'vdu2+D.vd2u. In working, say, with parametricequations the student is apt to use the first of these forms as though it had the invariance propertiesof the second. It seems advisable to leave these out of calculus completely,and the development outlined above does this very nicely. There the differential of dv is not definedbecause the domain of dv is the tangent bundle while the differential is defined only for a variable whose domain is C. (c) Chain rules. Given f(x, y, u, v) =g(x, y, u, v) =0, it is common practice to find,say, au/axin termsof other partial derivatives by writingout various expressionsmodeled after (10), making substitutionsand equating coefficients This is an effective of an appropriate differential. technique forderivingchain rules, and in the modern theory it is quite acceptable. However, if (10) is a definition ratherthan a theorem,it must be proved invariant under coordinate formis invariant under the changes. One must oftenknow that the differential very chain rule he is seeking to find! The undergraduateis afraid to use the and approximations. (d) Increments such as differential-in manipulations (c), for example-because of the apthe with now outmoded definition. In the introduced proximation bugaboo is this never fear modern theory introduced,and the approximate increment is classified-under problem properly Taylor's theorem. 5. Line and surface integrals. Let u and v be variables on a curve C. Partition C by points po, pi, * * *, pn.For each i, let qi be the point on the tangent line Tj whose distance frompi is the lengthof the arc Pipi+i on C. Intuitively, "unroll" each incrementof arc onto a tangent line. As noted in Section 3, the with the tangent a "line," to be identified set of derivativeoperatorsat pi forms line T, To obtain the usual geometricinterpretation, specifythat ifs is the arc with the unit tangentvector. In any lengthvariable on C, then D8 is identified

This content downloaded from 128.59.62.83 on Wed, 1 Jan 2014 09:37:05 AM All use subject to JSTOR Terms and Conditions

86

BRINGING CALCULUS UP-TO-DATE

[February

with a derivative operator at case, each point qi constructedabove is identified so the notation has an obvious pi; meaning. Now, formsums dv,i(qi)
n-1

(11)

~~~~~~E u(pj)dvp,(qi)
i=O

Take an appropriatelimit of these, and call the result (12) fudv.

The importantfeatureto this development of the line integral is that the points qi on the tangentlines are determinedby an intrinsic notion (arc length) on C and have no connectionwith the variables u and v. Thus, the substitution theorem,fe udv= is an immediate consequence of the fundamental =fc uDwvdw, theoremon differentials (9). The approximatingsums (11) are invariant under the substitution;hence so is the integral.The familiarfa8'f(x)dx is now a special case of (12)-in which C is an interval on the x-axis. In this development the is clearly the same thingin differential differential and integralcalculus. In a double integral-carefully to be distinguished froman iteratedintegral -exterior multiplication of differentials must be introduced.This is an operation denoted by * and characterizedby the rule (13) du*dv =
-

dv* du.

This stems fromthe fact that du * dv is to be a signed measure on orientedplane regions.Setting u = v in (13) yields the corollarydu * du = 0. Using these rules of multiplication(plus a distributivelaw) and substitutingfromthe two-dimensional case of (10), one has du*dv=

(14)

a8x ay a<x ay auadv auadv - dx * dx + - - dx * dy =axay ax ax aua v au av + - -dy*dx+-dy*dy ay ax ay ay

-I

-dx

+ -

dy

dx+-

dy

Itaav- - au Ov\ dx*dy \axay ay dx

d(u, v)

a(x,y)

dx*dy.

In general,a pair (dup,dvp)formsa set of affine coordinatesin the tangent plane T,. If the pair is orthogonal(and some pairs always are), then multipliers

This content downloaded from 128.59.62.83 on Wed, 1 Jan 2014 09:37:05 AM All use subject to JSTOR Terms and Conditions

1958]

BRINGING CALCULUS UP-TO-DATE

87

a and b can be chosen so that (adu,, bdv,) formsa set of rectangularcoordinates in T,. In general,a and b depend on p; forexample, on the tangent planes to the unit sphere (d+, sin 4 dO) generatesrectangularcoordinates. In any case, if (adu,, bdv,) is a rectangularset on Tp, and if A is a regionin T, positivelyoriented with respect to this coordinate system,then defineab(du * dv),(A) to be the area of A. Since the Jacobian of a rotation is unity, (14) shows that this is consistentforall such rectangularsets. Values of other exteriorproduct operators are then determinedby (14). Now, the substitutionrule formultipleintegralsconsists merelyof making of the surfaceintegral the substitution(14) behind the integralsign. A definition patterned after that given above for the line integral makes this remarkably easy to justify. Take a smooth surface S and partition it in an arbitrary fashion-not necessarilyfollowingcoordinate lines-into subsets Si. At some point pi in Si take a tangent plane. Here a minorcomplicationappears; one cannot "unroll" the Si as he did the arcs and preservemeasure. However, definesurfaceareathere are many tricks for accomplishingthis-and constructon each tangent plane a set having the same area as Si. Call this plane set Ai. The shape of Ai above) (du *dv),i(Ai) depends only on the is immaterialbecause (see definition area and orientationof Ai, not on its shape. Now, formsums

(15)

n
i=l

(du*dv) p,(Ai). w(pi)

Take the usual limit,and call it fwdu * dv.

Since the Ai depend in no way on u and v, the sums (15) are invariantunderthe substitution(14); and thereemergesthe familiarresult
frwdu*dv s
=

w
s

(x, y)

dx*dy.

6. Theorems of Green, Gauss, and Stokes. This view of the multipleintegral to the theoryrelatingan integralover a manifold bringsan elegant unification to one over the boundary. There is a master formula:

(16)

...

udv1*...

*dvk =..

f du*dv*

dVk

where M is an oriented (k+1)-dimensional manifold and B is its boundary. This is proved in the usual way by reducingthe integralon the rightto an iterone integration. ated integraland performing Now, if A is a regionin the plane and C its boundary,then (16) yields

This content downloaded from 128.59.62.83 on Wed, 1 Jan 2014 09:37:05 AM All use subject to JSTOR Terms and Conditions

88

BRINGING CALCULUS UP-TO-DATE

[February

f(udx + vdy)= However, du*dx =

-ax

"A

*dx + dv* dy).

au 1u _ au dx*dy, *dx = -dx++ -ddy

ay _ 49y

-dv av av 1 dv*dy = [ dx +dy *dy=--dx*dy; so the result is Green's theorem.A similarmaneuver with the form f(udx + vdy+ wdz) where C is an appropriate space curve yields the classical theoremof Stokes. The divergencetheoremis also a special case of (16) as is easily shown by direct computation. Minus signs do not appear in the divergencetheorembecause in and a all this work the permutationsof differentials are cyclic permutations, cyclic permutationof threethingsis an even permutation. When presentedin this light, all these theoremsappear as generalizations of the fundamentaltheoremof calculus. For, this latter may be writtenu(q) on the -u(p) = fdu, and this is a specialization of (16) in which the difference boundary consistingof leftis regardedas an integralover the zero-dimensional two points. Subtractionresultsbecause the boundary is consideredoriented. 7. Notation. Roughly, the structurestudied in calculus seems to consist of mappings (variables) "up" frompoint sets into the real numbersystemtogether with other mappings (functions) "across the top." It is well to preserve this in the notation,and the usual use off and g forfunctions and the stratification there are last lettersof the alphabet forvariables is all to the good. However, other distinctionsthat common notational practice fails to make. whose values are appropriate The derivativeof a function is anotherfunction "with limits of difference quotients. Note that a functionis not differentiated respectto" anythingin particular.Thus, an appropriatenotationforthe derivative off isf', and it would be well ifthe primesymbolwerereservedforfunctions only. Derivatives ofvariables are generatedby the derivativeoperatorsintroduced formdy/dxis not merelyan above, and the basic formis Dxy. The differential the fundamental alternativeto this. It means dy . dx, but it-follows at once from = Dxy. theoremon differentials that dy/dx but it is not Hybrid notation such as y' and df/dxabounds in the literature, One step in the modernizationof calculus is to break some really well-defined. well-formed habits in this respect. Frequently,the confusion (too well-formed) stemsfrom the symbolsy and f. As forhabits, writing y =f(x) and thenconfusing

This content downloaded from 128.59.62.83 on Wed, 1 Jan 2014 09:37:05 AM All use subject to JSTOR Terms and Conditions

1958]

BRINGING CALCULUS UP-TO-DATE

89

however,it is sad to relate that the author of a recenttextbookmade essentially the commentjust made here about confusing y and f and then proceeded to do exactly that threepages later. Similar distinctions need to be made in the multidimensional cases. Iff maps n-tuplesof numbersinto numbers,its argumentsare distinguishedby position only. For example,f(a, b) =a-b and f(b, a) =b-a definethe same function f. Thus, fi is appropriate notation for the partial derivative of f with respect to its ith argument.As noted above, a/axis a derivativeoperatorin manifoldtheory and should be reservedforapplication to variables. 8. Existence theorems. Little has been said so far about the role of real calculus. In a sense this is a quesfunctiontheoryin a programof modernizing tion quite independentof the ones raised so farhere,but certainlythe following question is pertinentto the presentdiscussion. The algebraic theoryof derivative operators and differentials outlined above yields formulasand techniques in a veryelegant fashion,but it certainlyis not selfcontained.Is any additional real functiontheory required as a background to tightenup the logic in this development? In integralcalculus one must develop the notion of integral of a function alongside that of integralof a variable (outlinedabove). The two are fairly easily related, and the usual existenceproofappears in the functiontheory. In differential calculus the existence questions are answered a priori in the conditionsare imposed by postulates for a manifold.That is, differentiability fiat on the connectingfunctionsin such a way that the algebraically defined in the function derivativeoperatorsdo represent differentiations theoreticsense. However, to keep fromoperating in a vacuum, one must show that certain thingsare manifolds. For purposes of calculus one gets an adequate supply of manifoldsby taking the rectangularcoordinatesin (n+k)-space and considering the locus of n equations in them. The implicit functiontheorem (a standard item in advanced calculus) gives conditions under which k of these variables become local coordinateson the locus, and the Brouwertheoremon invariance of domain (now appearing in the better advanced calculus texts) guarantees that the locus has the proper dimension.*To verifythe hypothesesof these theorems,one must know the differentiability These may be of the elementaryfunctions. properties establishedin the usual way throughtheorems on differentiability ofsums,products, composites,etc. It is probably fairto say, then, that the introduction of the moderntheory the the the of differential numberor nature of the real funcchanges very little tion theoryexistenceproofsrequired fora logically sound developmentof calculus. This, of course, is subject to the stipulation that calculus is restrictedto
* Addedin proof: A recent noteby Yamabe,thisMONTHLY, vol. 64, 1957,p. 725,givesprethe result cisely needed.

This content downloaded from 128.59.62.83 on Wed, 1 Jan 2014 09:37:05 AM All use subject to JSTOR Terms and Conditions

90

COEFFICIENTS

OF RECIPROCAL POWER SERIES

[February

the well-behaved cases. A general attack on the topological problemsof manifold theoryis much more difficult. 9. Two facets to the problem. Most programsforthe improvementof calculus amount to the injection into the course of more and betterreal function theory.It is the purpose of the presentpaper not to discourage this,but to suggest that a more imperative project is the injection of better-if not moreThis latter is more imperativebecause the garden variety differential geometry. calculus contains more downrighterrorsin differential geometrythan in real functiontheory.On the other hand, it is more difficult forat least two reasons. First, to bringthe differential geometry up to date willchange the appearance of a calculus text; it will modifydefinitions, terminology, notation, procedures. Second, while most calculus teachersare well groundedin real functiontheory, formany of them graduate study came beforesome of the importantdevelopments in differential geometry. Despite thesedifficulties an effort should be made. In failingto bringcalculus to the next generationnot the information up to date we are transmitting availbut that available to our grandfathers. able to our contemporaries,

ON THE COEFFICIENTS OF RECIPROCAL POWER SERIES

JOHN LAMPERTI, California Institute ofTechnology

satisfythe relations
(l)

{un} and ofnumbers 1. Introduction. Supposethatthetwosequences

~~~~~~Un
n-1 =Ui,fn-i i=O

{fn}

+ bon)

*I* n =0,1,

where J1if i=j

bj 0Oif i

js
00

of the sequences, Upon taking the generatingfunctions

00

U(Z)

= n=O

UnZn

and

F(z)

n=l

fnZ,

as relation (1) may be re-expressed (2) U(z) = U(z)F(z) + 1, in a purely are to be interpreted wherethe powerseriesand theirmultiplication formalway. If U(z) and F(z) satisfy(2), then we shall say that F(z) is the re-

This content downloaded from 128.59.62.83 on Wed, 1 Jan 2014 09:37:05 AM All use subject to JSTOR Terms and Conditions