Similarities in Irrationality Proofs for 7r,ln2, C(2),and C(3) Dirk Huylebrouck 1. FOUR REMARKABLENUMBERS. The firsttwo numbers,wrandln2, are familiarto high school graduates.Theirexpressionsas seriesarestandard: ST = . 2i + 1 IiI>O-1)' 1-3 +357 ln2 = 1 -2 + 3-4 2-l) -+ +*= (Leibniz'series) =0.693147.... i>1 Similarexpressionsdefinethe less familiar( (2) and (3): (2) = 1 + 2+ 3 + 32 22 + and ((3) = 1+ I+ 33 23 + i + i =E j2 j2 i>1 + j3i + = 1.64493... = = 1.20205.... j ~~~~i>1 The irrationalityof 7rdominateda good 2000 years of mathematicalhistory,starting with the closely relatedcircle-squaringproblemof the ancientGreeks.In 1761 Lamof 7r(Lindemannwouldcompletethe transcendenceproof bertprovedthe irrationality in 1882 [2, pp. 52 and 172]). The interestin ir's youngerbrother((3) startedonly a few centuriesago, but the numberresisteduntil 1978, when R. Apery presentedhis 'miraculous'proof [14]. Even afterApery'slecture,scepticismremainedgeneral,until Beukers'simplifiedversionconfirmedit [3]. The characterof the (-numbersstill fascinatesthe mathematicalcommunity,andeven veryrecentlyit was upsetby results of TanguyRivoal(communication J. VanGeel, Universityof Ghent). of Q(2)and (3) as inteEssentialin the simplifiedproofsare the representations grals.Since J~dxdy If'AX1 Ixy = If AX =E Zxy'dxdy >E xidx yidy o >E[i+1o[+ E (i + 1)29 we have Q(2)= f I 222 J- dxdy. ?dTHlEMATHEMATICALASSOCIATION OF AMERICA [Monthly108 Similarlyfor <(3): t f =21 x ylnxydxdy =ffLxnxydxdy xiln xdx f =2E xiyInxdxdy yidy i>o?o 2E ([xi+llnx]- =2E i+1 (-i Xi-+1dx) i d f + 1) i+1 -2E = (i + l)3 - 2<(3) Thus, <(3) =I- 1 f inxydxdy. Incidentally,the integralfor <(2) can be evaluatedvery easily. This computationis not reallyneededhere,butit providesan illustrationof a calculationinvolvinga zeta expression.The easiest componentsof Apostol's,Beuker's,andKalman'sresultsare combined,andexceptfor someelementaryknowledgeaboutdoubleintegrals,no other prerequisitesseem needed: see [1], [3], [10]. The combinationprovidesa proof at graduatelevel for Euler's<(2) result;see [4] for a morerigorousandgeneralproof. First, rewritethe differenceof two integralsusing the substitutionX = x2 and Y = y: (I ffl: f' (I 1 11(1 )dxdY= ddxdy -1 - Y) ) )dXdY. (1) dxdy. (2) Next, obtaintheirsum: f' (iixy+ 1 xy1+x - 1 dxdy = 2 yo lxy2 22 Adding(1) and(2) gives 2 ff | dxdy = f1 1 dXdY + 2 2!dxdy, 2 or 1 +2 2<(2) = -2(2) JJ l-X2y2 2 dxdy. Thus, 31 -<(2) = March20011] dxdy. 223 Substitutingx = sin0/cosq5 and y = sinq/cos0 yields a Jacobian equal to 1 - tan20 tan24 andthis is the denominatorof the integrand. (2)=- 3 d0f ? ? 1 2 (sinO doj o tan2p) dy C cosO kCosq0 =j =3 2 (1-tan20 sinS q2 dc~p=~~3 8 =6. This is Euler'swell-knownresult:( (2) = 7r2/6. 2. FOUR PROOFS IN ONE. The Borweinscollectedirrationalityproofsfor these four numbers[5, pp. 353, 366, 369, and 370], in a very rigoroustreatise.In orderto maketheir similaritymore evident,we first summarisethe highlightsof the demonstrationsin generalterms. Supposethe irrationalityof a numbert mustbe shown.In the fourcases we present here,afamily of integrals(j E N) concerningthatnumberis proposed: = Rj +Sj X x'f(x)dx whereRj and Sj E Q;f is an unknownfunction. Now if t were a fractiona/b, this family of integralswould yield rationalexpressions f0' xlf(x)dx = Cj/Dj, where Cj and Dj E Z. Integermultiplesof these integralsandtheirsumswouldagainexhibitthis property.Thus,if Pnj E Z andn E N: n In j E PnjXf(x)dx = EPnj n I x'f(x)dx f E Pnj Dj =0 - F' with againEn and Fn E Z. Applythis propertyto the Legendrepolynomials: Pn(x) = 1 dn = ! dXn (xn(1 -x)n) PnjXj. For example, Po(x) = 0, PI(x) = 1 - 2x, P2(x) = 2 - 12x - 12x2. Note that Pnj E Z. Thus,for the given familyof integrals, /Pn(x)f (x)dx - Bn, with An,BnE Z The choice of Legendrepolynomialsis inspiredby the possibilityof performing integrationsby partseasily. Indeed,since Pn(0) = 0 = Pn(1), and similarlyfor the derivativesof xn(1- x)n up to the ordern, manytermscan be simplified: P (x) 224 f (x)dx (X n(l - x)) f(x)dx THE MATHEMATICAL ASSOCIATION OF AMERICA n [Monthly108 [n! d;n-1 (x (1 -x)) = Oj t n!dXn nl(X 1 1d f 1Xn(-x)nd dx nJ Integratingby partsn timesleadsto f(x)] dx dx. (1X)) X)n t(Xn f(x)df(x dx. dXn-d ~ ~ a In the fourencounteredcases the functionf(x) happensto be suchthat (l-x)n jAxn d x) dx |= fX (g(x))nh(x)dx| wherethe maximumvalueM of g(x) is smallenoughto ensurethat -d0. |B nM fh(x)dx In addition,all the integralsfleP(x)f(x)dx arenon-zero,andthis immediatelyimplies the irrationalityof t: O<IA = Bn]1 Pn(x)f(x)dx < B-M ] h(x)dx O. a is a positive integerso this is impossible; cannotbe Indeed,for any n E , al rational. Of course, the difficultyin the proofs lies in the appropriatechoice of f(x). It must yield a family of non-zerointegralswhose membersare easily expressedas a combinationrationalnumberandt. In addition,it shouldsimultaneouslybe possible to maximizetheir productwith an integerthat becomes largerand larger,and still ensurethe indicatedconvergenceto 0. 3-I. Irrationality of -7rTake f(x) = sin(wx). It is a standardcalculus exercise 1 to show that the membersof family of integralsof the form xJ sin(wx)dx are polynomials in wrof degree at most j, divided by wi. The linear combinations f0oPn(x)sin(wx)dx arenon-zero,andthus,if w were the rationalnumbera/b: 0 < An, = an j = n an Pn(x) sin(7rx)dx J !_xn(1 - X)n (sinrx)dx| < an _Xn(I - x)n ndx since the n-th orderderivativeof sin(wx) is ?wr sin(wx) or ?rn cos(rx). The maximumvalueof x(I - x) is 1/4, attainedat x = 1/2. Thus,the finalexpressionis less March2001] 225 than (n in1 ni whichis arbitrarysmallfor largevaluesof n; see [13]. 3-l1. Irrationalityof 1n2 The choice is f (x) = 1/(1 + x). An Euclidiandivisionof xi by 1 + x allowsus to computethe familyof integrals xj 1 =- x_dx 1+x 1 - ] -1I F ?1 ln2. + If ln2 were a/b, then 0< lAnl = = (bdn)Jo fX( -x)4 < (1 (bdn) (wheredn= LCM{1,2, 3, ...,n}) dx (bdn) IPn(x) [d+(1 I )]dx dx. -x)\' since the n-th orderderivativeof 1/(1 + x) is (-1) ... (-n)(1/(l + X))n+l Now on [0, 1], themaximumvalueof x (1 -x)/( + x) is 3 -2 , achievedatx = -1 ?V+. A roughinequalityfrom numbertheoryis dn = LCM{1, . . , n} c 3n. Finally,since (3(3 - 2V) )n < 1, the irrationalityof ln2 is established;see [5, p. 370]. 3-IIl. Irrationalityof ((2) The choice is f(x) = f l-Y) dy -xy Eachmemberof the familyof integrals [j jl 1 0 (1IY)nd]d I -xy is a sumof integralsof the form Y dydx, with r, f O s EN. -xy Thesecan againbe computedthroughan Euclidiandivisionof xiyk by 1 - xy, which gives a sumof integralsof the form Jl qdydx, o I dYdx, 110-xy _dydx or 1 xy dydx. The latteris ( (2), while the othersaresumsof fractions,whichcanbe computedusing partial integration for the integral f xmln xdx. Thus when r s, 226 ?gTHE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly108 J ff -dydx 1-x4y is a sum of fractionswhose commondenominatoris the squareof least commonmultiple (LCM)of the firstn + 1 integers.Whenr = s, the integralequals i >rji2 +2) =r2-(+ Thus, Pn(x)f(x)dx f - d2A wheredn+1is the LCMof the firstn + 1 naturalnumbersandAn E ZLO Now O< 10 Pn(x)f (x)dx IAnI= dn2+1 = 2 = dn+1 I = d2?1f d 1 Xn(fl Jo ! y(n x .(1-.x).(-1) (X(1 (-n) x)y(xy)) [x (I (dy )f -_dydx Xy)n- (3) - y)/(t -xy) is ((-1 + V5)/2)5, and On [0, 1], the maximum value of x(l -x)y(l is attainedfor x = y = (-1 ? V)/2. Togetherwith dn+1< 3n+1, this showsthatthe expression(3) is less than (3n+1)2f ( dxdy = 9 ( (2) < 1, whichestablishesthe irrationalityof ( (2); see [3]. 3-IV. Irrationalityof C(3) Take 1 f (x) [(y) lnxy dxdy. The membersof the familyof integrals [f |1xJ o 0 Pnf(Y) lnxy dxdy dx 1-xy arecomputedthroughthe derivativeof March2001] 227 J dd Xr+tyS+t 1d-xy withrespectto t, whichis I S+t 1 Xr+t 1 lnxydxdy. -xY If r 0 s, this is a sum of fractions since d(r + t)-m/dt =-m/(r + t)'1, and the LCMof the denominatorsis (dr+t)3. When r = s, the result is d 1 irt d1t(r+ i >r+t -2 t)2 (r+ t)3 i>t ~ i>r+t ~ =-2 ((3)(1 + (r 3 *+ + (r +1l)3 + +(r + t)3) Thus, fPn (x)f (x)dx -Id A 3- 1 whereAn E Z anddn+lis the LCMof the firstn + 1 naturalnumbers. Now IAni = dn+1f Pn(x) = d3+1 = d3 = 3 f fl Xf(l fJl; t } - 1 -(I - XY)z~1 x) Pn(y)(1w- xn (-X)nyn j- WJJ ( The maximum value of x(l (1 dxdydw y)nwn(1- - (1 - xy)W)n+l 1_ )Y(1- Y)W( - x)y(l - y)w(l l-Z ) (integration by parts) O W) - ( (put w = ~X)fPn(Y)YlZfldXdyd dn+1f f = d3 (integrationby parts) Pn(X)(P(Y)_dxdydz f n+1 = ld3~ lnxydy] dx 1PY) [f dxdydw _1-(1 w)/(l - dxdYdw (4) (1 - xy)w) on [0, 11 is (\2 - 1)4, and is attainedfor x = y = -1 + \2 and z = 1/\2. Togetherwith dn+l < 3n1 this shows that the expression (4) is less than (3n+1)3 -fff ( - XY)wdxdydw = 27 (27 (2 - <1, whichestablishesthe irrationalityof ? (3), see [3]. 228 ?) THE MATHEMATICALASSOCIATIONOF AMERICA [Monthly108 4. THE GOLDENSECTION. In ourproofs,the maximumvaluesof certainfunctionsplay an importantrole. Theywere attainedat the points 1/2, (-1 + V5)/2, and -1 + V2. Thereis a coincidentallink since they areall relatedto the goldensection. Classically,thegoldensection0 ariseswhena line segmentof lengthx greaterthan 1 is dividedinto two parts.This couldbe doneby cuttingit into two halves(recallthat 1/2 poppedup in the irrationalityproof of r). Or, if unequalsegmentsare desired, one could look for two pieces of lengths 1 and x - 1, such thatthe ratiox/ 1 equals the ratio I/(x - 1). This equality produces the quadraticequation x2- x - 1 = 0, of which 1.6180 ... = q)is the positivesolution. Moregenerally,the positiveroots of x2 - nx - 1 = 0 yield the family of metallic means,for variousvaluesof n E N. Forn = 2, we get the silvermeanaAg = 1 + V2, for n = 3 the bronzemeanUBr = (3 + VB)/2, etc. The propertiesof these numbers have been describedin numerouspublications;for a comprehensivesurvey,see [6], while [8] and [12] pointedout that some authorsoften had too much enthusiasm.A commonmisconceptionis thata rectangleof width1 andlengthq)wouldbe the "most elegant"one andthusis usedin variousdesigns.However,no reliablestatisticalstudies confirmthis statementaboutthe optimalchoice providedby the goldennumber[7]. A statementthatcomes close is the fact that0 wouldbe "themost irrationalof all irrationalnumbers"becauseits representation as a continuedfractioncontainsonly is: 1 1+ 11+* The silvermeanwouldbe "thesecondmostirrationalnumber"since cAg = [2, 2, ... ], of the goldensectionan etc.; see [6]. Yet,this againdoes not providean interpretation optimalsolution,in the standardmathematicalsense. However,the variousirrationality proofslead to a propertyof these metallicmeans where the expression "optimalsolution"has its common mathematicalmeaning. Table1 illustratessome interestingfacts: TABLE1. Proof 1n Maximum Functionused x(l-x) x= x= x=-1+ In2 (1+X) 3-2V/ ?(2) x(l (( 2 1 2 x)y(l-y) (I- xy) ?(3) x(1 ( -x)y(i )( - y)w(l Y( - w) A substitutionof X = -l/x )(-I (i<-4X + )2 and Y =-l/y ) Name Attainedat x 2- 1 = ) 2-AgI 2 2 x = y = 1+ -1 + 2;J in x (1 -x)y(1 -y)/(l )_ z Z> -xy) A -OAg changes the expressioninto (1 + x)(1 + y)/(xy(xy - 1)). Its extremumis obtainedat X = Y = 4. Similarly,cAg providesthe optimalsolutionto (X - 1)/((X + 1)X). In these cases, the word "optimal"is used in the usual mathematicalway, in contrastto the of these loose termsoftenused in goldensectionpapers.The geometricinterpretation facts is developedin a forthcomingtext [9]. March2001] 229 Thereareotherlinksbetween (2) and (3), andthe metallicmeans.Forexample, in the easy proof for Euler's ( (2) result,hyperbolicsines and cosines (x = sinh0/ cosh p and y = sinh // cosh) can be substitutedinsteadof the similarexpressions with circularsines andcosines. In thatcase, ((2)= (ln ( +d)) + 2] (t - argcosh (sinh t)) dt. Now the silversection1 + vX appears,whilein Table1, ? (2) was alreadylinkedto the goldensection.Moreinvolvedcomputationsrelate( (3) to the goldensection,too [1, p. 156]: ln ) = 10 15-) ) t2coshtdt. These relationsdid not informus aboutthe "optimal"propertiesof the metallicnumbers,andwe give themherefor sakeof completeness. Incidentally,since xy I(4)= dxdydwdv = 1 + - + T + ) a naturalgeneralizationof the functionsused in the studyof the irrationalityof (2) and( (3) wouldbe x(l - x)y(1 - y)w(l - w)v(1 - v)(1 - xy) (1- (1 -xy)w)(l -(1 -xy)v) The maximumof this functionis (5- 13) (-7 + 213) 1) 54(-3+ obtained for z = w = (1 - (xy))/(1 - xy) and x = y = (-3 + VT3)/2. Here it does againa metallicmeanis found,andit is the next one, -aBr-1. Unfortunately, not providea prooffor the irrationalityof ( (4) (andby extensionfor ( (5)) since the membersof the familyof integralsarenotcombinationsof ? (4) withrationalnumbers. The questfor theseproofsremainsopen. REFERENCES 1. Tom M. Apostol, A Proof thatEuler Missed: Evaluating (2) the Easy Way,Math. Intelligencer 5 (1983) 59-60. 2. P. Beckman, A History of Pi, 4th ed. Golem Press, Boulder, Colorado, 1977. 3. F. Beukers, A Note on the irrationalityof ((2) and Q(3),Bull. London Math. Soc. 11 (1979) 268-272. 4. F Beukers, J. A. C. Kolk, and E. Calabi, Sums of generalized harmonicseries and volumes, Nieuw Arch. Wisk.(4) 11 (1993) 217-224. 5. J. M. Borwein and P. B. Borwein, Pi and the AGM: A Study in Analytic Number Theoryand Computational Complexity,Wiley, New York, 1987. 6. Vera W. de Spinadel, From the Golden Mean to Chaos, Nueva LibreriaS.R.L. Buenos Aires, Argentina 1998. 7. ChristopherD. Green, All that Glitters: A Review of Psychological Research on the Aesthetics of the Golden Section, Perception24 (1995) 937-968. 230 ? THE MATHEMATICALASSOCIATIONOF AMERICA [Monthly108 8. Roger Herz-Fischler,A MathematicalHistory of Division in Extreme and Mean Ratio, Wilfrid Laurier University Press, Waterloo,Canada, 1987. 9. Dirk Huylebrouck,the golden section as an optimal solution, to appear. 10. Dan Kalman, Six Ways to Sum a Series, College Math. J. 24 (1993) 402-421. 11. L. Lewin, Polylogarithmsand Associated Functions, North Holland, New York, 1981. 12. George Markowsky,Misconceptions about the Golden Ratio, College Math. J. 23 (1992) 2-19. 13. I. Niven, A Simple Proof that wris Irrational,Bull. Amer.Math. Soc. 53 (1947) 509. 14. A. van der Poorten, A Proof that Euler Missed... Apery's Proof of the Irrationalityof (3), An informal report,Math. Intelligencer 1 (4) (1978/9) 195-203. D. HUYLEBROUCK obtainedhis Ph.D. in linear algebraat the University of Ghent (Belgium). Several years of teaching in Congo came to a sudden end after a conflict between former president Mobutu and Belgium. He taught in Portugal,and later at University of Marylandfor American GIs in Europe. His fascination for the history of mathematicsled him to produce a poster of 100,000 coloured decimals of pi. Another stay in Africa again came to an abruptend aftera coup in Burundi(by someone else), and so he settled at the Sint-Lucas Institutefor Architecturein Brussels, Belgium. He still has an African dream:to let its oldest mathematicalartifactone day reach space. Aartshertogstraat42, 8400 Oostende, Belgium dirk.huylebrouck@pi.be Solution to Sherwood Forest Puzzle on p. 143 of the February issue: The idea of 'singerof a person'ssong to another'can be takenas an operation*, thatis, "x * y = z" denotes"x's song is sungto y by z". Thus: 1) 2) 3) 4) we haveclosureof * (everyone'ssong is sungto everyoneby a singer), we havean 'identity'(the priest), we have 'inverse'for everyone(mates), is the associativityof the operathe (unavoidably)crypticthirdparagraph tion *. Thuswe havea groupstructure.But, 10,201 = 101 x 101 and 01 is a prime, andany groupof orderof squareof a primeis abelian. Thus, it was Marianwho sang Little John'ssong to Robin. For the second question,it suffices that we have a group structuresince we have Marian= Robin* LittleJohn.We shouldrecall Marianand Robin are mates and so we can left multiplythe equationby Robin's inverse to get Marian* Marian= LittleJohn.And so LittleJohnsang to Marianher song. March2001] 231