SUMMATION A!D TABLE OF FI1ITE SUMS by ROBERT DELMER STALLE! A THESIS subnitted to OREGON STATE COLlEGE in partial fulfillment of the requirementh for the degree of MASTER OF ARTS June l94 APPROVED: Professor of Mathematics In Charge of Major Head of Deparent of Mathematics Chairman of School Graduate Committee Dean of the Graduate School ACKOEDGE!'T The writer dshes to eicpreßs his thanks to Dr. W. E. Mime, Head of the Department of Mathenatics, who has been a constant guide and inspiration in the writing of this thesis. TABLE OF CONTENTS I. i Finite calculus analogous to infinitesimal calculus. . .. . .. . . . . . . . . . . . . .. . .. e s 2 Theconstantofsuirrnation......................... 2 3 a Suniming as the inverse of . . . . . . . . . perfornungA............ 31nite calculus as a brancn of niathematics........ Application of finite 5lflITh1tiOfl................... II. III 5 LVELOPMENT OF SULTION FORiRLAS.................... 6 ttethods...........................a..........,.... Three genera]. sum formulas........................ 6 6 S1ThATION FORMULAS DERIVED B OF A Z INVERSION FELkTION....,..................,........... TIlE 7 s urnmation by parts..................15...... Ratlona]. functions................................ 7 Gamma and related functions........,........... 9 ... Ecponential and logarithrnic functions...... . Thigonoretric arÎ hyperbolic functons..........,. Combinations of elementary functions......,..... IV. 4 SUMUATION BY IfTHODS OF APPDXIMATION..............,. Tewton s formula . . . . a a a a S a C . a e a a . a Extensionofpartialsunmation................a... Formulas relating a sum to an ifltegral..a.aaaaaaa. Sumfromeverym'thterm........aa..a..aaa........ a s e a a a a a J-3 14 15 15 15 16 17 V. TABLE OFST.Thß,..,,..,,...,.,,.....,....,,,........... 18 VI. SLThMTION OF A SPECIAL TYPE OF 26 VI I BIBLIOGRAPHY. . a a a a a a a a a a . . POER SERIES.......... a . . a a a a s . . . . . . e . * . a a a . 44 TABlE OF FINITE STThSIAATION AND I To T}OR! STJ OF FINITE SU1ATION the knowledge of' the writer, no table of' finite sum formulas of value as a reference is in existence. For this reason a table of such formulas, analogous to a table of' integral formulas, is presented here. The formulas for this table were selected with their utility in mind. A series. finite sum formula is a Finite suntion is the relation giving the eluation of a sum of a finite finite series. This valuation cari be accomplished by the development and application of general sum formulas. These formulas are of closed form and give the exact or approximate sum, depending on the type of series. The theory of finite sunvation is presented first and is followed by the development of and finite sun formulas, both general special. Finally the table is presented. As we will see, the subject of summation arises as a branch of the calculus of finite differences. finite summation will be Throughout throwgh this paper Uence this treatment of this calculus. numbers enclosed to relationships in the discussion, whereas parentheses refer to formulas in the table. in parentheses refer nwbers not enclosed in 2 Finite calculus analoous to infinitesiiial calculus. There is a striking analogy bet'reen the finite calculus, or calculus of finite differences, and the infinitesimal calculus. Consider the function Ut subjected to these calculi. The two fundanntal operations of the infinitesimal calculus are differentiation, L dt and its inrse, lim h40 %th'1t h The two fundanental operations of integration. ¿ - operation, the finite calculus aro the LUt UUt h and its inverse. The analogy is nt perfect because of the additional Uniting operation in differentiation. Results in finite calculus may be reduced to the analogous results in infinitesimal calculus at any time if we perform this additional limiting operation. In fact by stxiying finite calculus gain a greater perspective onto and insight into infinitesimal calculus. In the finite calculus we may assume, without loss of generality, that the linear subatitution, t = hx, haa been made Lt * bx, L X is unity. Hence ¿ performed on u Now, since nw gives: (1) SuDud.n2 as the inverse of performing ¿ Let us inveitigate the inverse of the ¿\ operatIon, which 3 It is represented by George i3oole cafled difference integration. ¿. the srnbo1 to n, in we u, if A 1 then = from Given exists such that a function A = If we SUYa '-a for values of x obtain: n 'n-i.]. '-uX =VI Xl (2) . m mus i iç term is The the series whose general inay be found we niay sun u. constant of suiiniiation. is interrogative and hence that Note that the operation we have no assurance of the existence or uniqueness of the result u. from perforning it on a function Consider a case in is not guaranteed. of -1 and V+ an expression for results from ¿ exists. which reesent and Hence we let V the uniqueness Now be an expression Now all such expressions. for to Lind thereby the relation between possible we consider the Removing the parentheses we see relation, £v ¿1&) = = Thus the O. expression for (1, p.79-3l). is a general periodic function of period one In finite calculua x assimies successive values differing by unity so we may consider the rrre as a constant. suggestive notation I for Thus Ô1 = v+ c. We use and have; :=v+c. (3) 4 If we consider this sum from m to n have by (2): we n+l n+l n U=Vx+C (4) Finite calculus as a branch of mathematics. A look at finite calculus as a subject is In order. is some There literature about the technical terms used to the close analor with infinitesimal calculus confusion in Due (1, p.&L-E2). many terna are borrowed here. Below is a list of terms from finite calculus paired with the analogous terms of infinitesimal calculus. From this and previous discussion their meaning is held to be self evident. Finite calculus Performing Infinitesimal calculus Differentiating ¿\ Difference integral Indefinite integral Sum Definite integral Difference calculus Differential calculus Sum calculus Integral calculus Suinand Inte grand Infinitesima]. calculus includes evaluation of by other methods than through the indefinite definite integrals integral. it is convenient to include in finite calculus Likewise summation performed by other methods than through the difference integral. Sonatines it is necessary to define and tabulate a new function in order to 5 perform exact evaluation as for X the integral, dx = : and the sum, =f(n) -fm-1). Sometimes approximate evaluation is the beet we can do as in the case of the integral, _____ (lT dx, )Vsinx o and the sum, n arc tan x. i Summation of infinite series is analogous to evaluation of a integral with an infinite limit. is also definite Hence sunmation of infinite series included in the calculus of finite differences. Applications of finite summation. Some of the methods to be discussed in the paper permit the summation of convergent infinite series By definition, the sum of the convergent in closed form (3, p.18-23). infinite series n is fl-4 E a j:;]. a j Thus to sum the infinite serles is to evaluate the sum of the first n terms and then the limit of this sum. Another application of the sums and swmmation theory here developed is in the solution of difference equations. We proceed to develop these methods whereby we nay sum anr finite series. 6 SU!ATIO' T OF DE VEI II ORULA3 Methods. The two nx,st connon methods for derivation of finite sum formulas are (1) the : invorson of iie of the close relation methods inc1e been a difference relation, and (2) a sum and an integral. Other mathematical induction, operational attacks, resolution of difference equations, use of generating functions, use of geometric considerations, and use of the calculus of probability (4, p.109-141). Part III includes formulas, with one exception exact, developed by invarsion of a formulas, A relation. rart iv inclixles approximating sum hich are derivad by the principle of inversion or by using the sum-integral relation. In the table, sums appear 5ii form (4) with the limits substituted, or in the easier written by equivalent form (3). Three enera1 sum formulas. The derivations of formulas 2, 3, and 5 of the table come under a separate heading. In each case tho sums are expanded and the identity of the expansions noted (6, called Dirichlets ' sum-formula. functions u and V p.9l3). Formula 5 is It is useful when one of the can be suimmed exactly, but the other not. 7 III. - by inversion le to find To sum u and apply 31ThATION TORW1JIAS DERIVED BY THE INVERSION OF A A RELATION (3). For example from that cx = c e have fornula 1, ¿ L- relation uay be In fact any e = cx + C. such V inverted to ve a sum. SummatIon by The by parts. general formula for summation by parts, 4 is developed this inversion. ?(e have: V whence by summing each ffrst side from in to n: n n lin in in vixl first sum and (5) Lux Vx+i, + _tn+l x Solving for the = substituting EV= V1 tue we hove form of 5, tni-1 n =ux :vxi L-xx Im in I Mjuctment of the limits and second form. The reapplications of from the second auxmnand n _2ILu:v+l. in in the is developed third form it its right.hand to in the same manner. analogous to integration by parts. first form yLelds the from the sum. The first by r = i fom'th is derived Summation by parts plays a role C tional fwictiona, is defined for positive The factorial (ax + b) (ax for negative a)...(ax + b + b)(ax + b ni: am + a), ni: i (6) (ax + b)(ax + b + a)...(ax + b + ma - a)' and for ni = O as unity. If we invert: + b)(m+1) = and divide through by (ni + l)a(ax + b)(m), + 1), we obtain a(rn foririula 6: (ax Z(ax +b) (7) a(m+l) +C. Evidently ni cannot equal -1. Formula 7 is Aux+ seen to hold if u,cui+ ... + and reduce the resulting, equating like we expand in as: u u A ni-1 X fractions. oowers of x 4)(x) The 4(x) and A 's may be determined by this expansion. The sums resulting from application of 7 may be summed by 6. Formula C is proved by use of the relation: X x\ (a) al for the binomial coefficient. z 'a (a) ai () ' Hence, (a+i) (a + l) +C=ai Evidently a and x must be positive integers such that O<ax. Q Formula 9 used with 6 enables us to sun the power function x11 (4, p.16E3-173). We have as the expansion of x'' by Ne'wton'a formula, (23): (m) () x0 Stirling's numbers of the second kind are defined as: G: arìd hence we have formula 9. relation and ['J, = (10) From (10) we have the recursion initial condition: m-1 m Gn +mGn Gn+l n O = 1, O A brief table of G (11) i G = O for n O. O included later. s is Gamma and related functions. The gamma function r(x) is the accepted generalization of the factorial (x - 1)1 For x>0, r(x) = Integration by parts yields QJ ?1 the recursion xr(x) = r(x + which defines r() may be found in for x almost any 0. dy. (12) relation: 1), (13) The foflowing fundamental relations treatrient of r(): lo r__'- i+th+ [ log f(x + The 3- 2c2 - X - 5l? 571 {log ni + (x + 1) = successive + 24832 derivatis of l)log . ...] - n lor(x) (34) log(x +v)J . (15) are defined as the triauxna, tetragamrna, et cetera functions and denoted diganirna, et cetera, respectively. These functions are fundamental in the theory of sunirnation of rational fractions. If a rational fraction s improper, division 4e1ds a rational inteal function and a proper fraction. Hence we need consider only rational integral functions and proper rational fractions. A by rational integral function formula 9 or direct expression in factorials and use of 6 (5, p.27-23). A proper rational fraction may be expressed as the sun of fractions in in may be summed by X and the denominator is a povrer of a linear function which the numerator is a constant. This is affected by the theory of rational fractions where the constant term in the linear function .s a complex number. 'r a) r 3. (x 2 3-, 2, 3, .., and a comnlex. affected by use of the above Essentially wo Hence our problem gamma and are inverting a sum This sumnation is related functions. relation but attack wi].]. be made on this si.vnrnation problem. of the digainma function: is to From a more direct the definition 11 f(x) =1ogr(x +1), we have by (15): r n+2 = (x) L n - 1 x+rJ (16) . I!enee, n +1) F( '- - = J. L +1' ni or as Y is a dtnînny variable we have formula 10, n =F(n + 1) ni As the trigainina, tetraga!runa, penthgamrna, et cetera functions r(x) we have are defined as successive derivatives of log from (16): 00 F(x)= 's .T 1-= 1 (x +Y) 00 f(x)=-2 -r= i (x 1 f(x)=6 r:;1 (x - - e Proceeding as with +T) +Y)4 ------ f(x) we obtain formulas U, ñ'om the assnnptotic expansion the aSsyìnptotic expansion for foi' r(x + 12, and 13. 1), (34), we obtain 12 f(x) =nx(x +1) + 1(1 1 1 + (x i + 210 (x 5)- s.., (17) + 1) and for f(x),Rx), et cetera, by the derivative relation between these finctions. A table has been prepared which f(x) for real ives Rx),r(x), f(x), <x x in increments of 0.01 for O of 0.1 for 10 <X = <60 = (2, and < i and increments This table used with the p.43-59). following reduction forrnuias, (2, p. XIX-XXII), yields considerable accuracy: n-1 lC(nX) =iog n + f(.c) =((x - f(x -/). 1) + ir ctn inc. n-1 /(rD:) = -. n : rc -i). (18) f(x - f(.x) ein 1). 1cc n-1 -ç- r ç- L (rix) 9(flX) i =7 n-i. : f(x -/). if Tables are now being computed for ana and digaa functions for inaginary arguments (Ref. 7). Formula 14 is proved by reduction of the suimriand to type form: 13 i1 ;i. x2+a22ai i i L(x -i (x+ai -l)+lJ' -ai -l)+]. and application of formula 10. Note that formulas 10 to 13 are the one exception to exact formulas in this chapter. However, they are exact in the sense that the sums are in exact terms of a defined tabulated function. Lote that f(x) plays a role in finite calculus analogous to that of 9n(x + 1) in infinitesimal calculus. Exponential and logarithmic functions. Formula 15 is proved by inversion of (a - 1)aX and This division is by zero or inipossible if a = 1. division by a - 1. n r(x) = Rn x. Formula 16 is proved by inversion of 1'rjgonontrjc and hyperbolic Lunc tions. Formulas 17 to 20 are proved by inversion of the following A- relations, adjustment of the argument, constant coefficient arid division by the e LXcos ax = - 2 sin(ax + gin ax = 2 cos(nx + ) a sin sin ) a (19) Acosh ax = 2 sinh(ax + ¿sinh ax = . 2 cosh(ax + ) sinìi ) sinh a .. a Combinations of elementary functions. Fornulas 2]. previous results. to 32 are derived by summation by parts and Evidently recursion formulas 30 arid 32 must be used as a pair. Formulas 33, 34, arid 35 are proved by the substitutions: [sin(a + b)x + sin(a - b)xJ sin ax coe bx , [cos(a + b)x - cos(a - b) X] sin ax sin bx = , (20) and, cos ax cos bx = cos(a + b)x + cos(a - b)x and application of formulas 17 and Fórmulas 36, 37, and 3 and 35 respectively. are , l. special cases of formulas 33, 34, They are of common application in practical Fourier analysis (4, p.123-129). Formulas 39 and 40 aro proved by the substitutions, cos 2 ax, (21) and, cos2ax = + cos 2 ax, and application of formulas 17 and 16. Note that analogous formulas for hiperbolic functions may be easily derived by substitution through the relations, sin(iz) = i sirth z (22) and, cos(iz) = cosh z. IV. STJMM.TION 13! METHOPS OF APPROXIMkTION - relation or Inversion of a is possible relatively in any few cases as exact evaluation of a for iO analogous of a derivative relation or any exact evaluation of an calculus. infinitesimal Hence as with infinitesLaal evaluation must often be through series exansions. those assymptotic series so their series are sum is sum inrsion integral in calculus In summation apnroximated their convergent part, as in the case of the expansion of gimia and related functions. Of course convergence and the remainder tern by must alvys he investigated. Newton's formula. By Newton's formula and cnn terni by we may tern. It states expand u in a series of factorials that: k (k) L2U° k.t kD For the proof we assn u may be expanded into a series of ol uX = a + a x + a2x(2)+ and (23) . evaluate the coefficients by setting x factorials, a3x ... = O in successive differences of this relation. Extension of partial summation. If wo let r41. in the fourth form of formula 5 w have formula 16 Formulas relating a to an integral, sum over The sum and integral of u related as may he seen by geometry. the curves Hence Y %, a the sum correction terms. Thc integral is th area under is an approximation to this may be expressed as the a siin interval are closely the same area. corresponding integral and If these terms are in the form of differences, the formula is Laplace's type, and if in the form of derivatives, Euler's type (6, p.104). Formula 42 is of Euler's type and is called the formula. It in derived by expanding the operational equivalent of (Q,d/dX...l)_]\, '' B+i which The in powers of d/dx. appear in the coefficients rnnubers and are Euler-MacLaurin nwnbers of The the form are called Bernoulli' s of froquent occurrence in approximate sum formulas. recursion relation: n-1 E (-1)(')B i relation is consistent with the definition, list of Bernoulli's (24) Bernoullian numbers may be used to evaluate the brief = O B2 numbers is included (4, p.233). = for n O later. This This 1. A list is sufficient for most practical purposes because the magnitude of the numbers increases so rapidly that any assymptotic series whose coeffIcients contain them has a small convergent part. Formula 43 is of Laplace's type and is called Gregory'a formula. It is derived by substitution for the derivatives in formula 42 of their difference equivalents (, p.62). 17 Sum from every m'th term. We have a Woolhouse or Lubback type formula according as the correction terma in such a formula are in terms of derivatives or differences respectively. Formula 44, a boThouse formula, is derived from formula 42 by eliminating the integral between formula 42 and a formula derived from m parts. 42 by subdividing each interval into Formula 45, a Lubback formula, is derived by replacing the differences in 44 by their differential equivalents. Formulas to x, 46, 47, ax1 48 1/x, and l/x2 are derived by applying respectively. Formula 49 is discussed in Chapter VI. formula 42 TALE V. i. 2. OF S1Th c=cx+c. 2111 e f(x). x) = f[r(x) ] ±t:(x) 1n+1 n u11 4. f(x) = ini 'ni-1 XIIm X = X z g(x) ± ni-1 ¿ - I nl ± mi-1 n - ni mi-2 ¡ni-1 y xl +2ux-2 u x-1 ¿ i In+3 Z. -' ir i J ni-r ., = u X E V X -u E rn-fr 2\ X z + (1)rL : x n In v -T n u = nl ni-1 3J. xi-2 ¡ n u In x x+r. ni 5. x-r X V. (ni-1) 6. E(x + b)(nl) a(: +1) 3 i- nl ?ero or any integer - 1. 19 i 4(x) 2: .. Ai Z "x1c+r 1 ax + b, integra], function of degree of (x), 4'(x) Ç) 2: = 9 Gm a, Rn Au+i + - i, A's from expaneion ... + z integers such that O <a x L=iJ 1 = 10. + C; A0ux+ ni ; a rational cj)(x) <. z. on page 25. Values of + i)- f(m); f(x) = DQn ["(z + i). n f'(n + 1)+f(m); j'(x) = D f(x). i]. ni (x+i)2 n 12. 4: ni n 1:3. x: ni = (z r(n + i) + f(m); f(x) = +i) tn4a i 14. Im X 15. ZaX a a-i C; a + i. ln+1 __ -2f(x-1+ai)I Im 20 16. 17. 1L =nr(x) 2n(x) sin(ax + b) Zcos(ax + b) + C. Cos(ax + -2 - b a/2) + sin a/2 - sin(ax + b a/2) + c. 2 sin a/2 19. sinh(ax + b) cosh(ax + b sinh 2 20. Zcosh(ax + b) 21. xaX 22. X (ni) a/2 sinh(ax + b 2 sinh a/2 a x a a -1 - - (a aX =x (ni) a-i + 1)2 X (ni) a/2) + C. ma X cos(ax - 8inax= sin ax aJ2) 2 sin a/2 24. a/2) + X = sin 23. - x(m)cos(ax - a/2) ____________ sin a/2 2 m(m - + + 4 sin2a/2 1) nix + (m-1) 4 sin2a/2 X(m_2)sin(ax + 4 sin2a/2 25. x cos ax = X sin(ax - a12) 2 sin a/2 cos ax + 4 sin2a/2 sinax + a). 21 Xs (in) 26. Zx'cos - a/2) 1r« 2 sin a/2 ni(m - 1) in + 4 sin2a/2 Zx(m_2)cos( + a). 2 4 sin a/2 In formulas 27-32: acosb-1 Ka2+1_2acosb ' a cos b a2+1_2a cosb 27. 2. 29. aXsin bx = lÇ1aXsln bx - K2aXcos bx + C. ZaCcos bx = K2aXsjn bx + IçÍaxcos bx + C. xasin bx = (K1- K2)xaX(sin bx + cos bx)+(I- 4)a''sin(bx-th) + 2 K1yXtbcos(hc + b) + 30. x (m)x. aslnbx=K1x -in K1 31. (m)x. asinbx-Kx (m)x acosbx Zz(m_asin(bx xa%os bx = (K1+ x+l a C. jç)X(5j . 4b)+ mK X (m-l) x+1 a cos(bx + b). bx + C05 bx)+(K+ cos(bx + b) + C. K) 22 32. ZxaXco naK2 bx = Kx(m)axsin bx + jç1x(m)axcos bx Zx(m1)a1sin(bx + cos[(a+ID)x 33. sin ax x1a1cos(bx b)- mX cos bx a+bl 34. Zain ax sin bx = - [(a..)x 48m sin [(a-ö)x cos ax cos hx = :35. 36. k L k sin ax1cos 37. z :3e. x: 39. : sin2ax 40. os 2 : bx= in axsin bxi k io sin axcos = 4c ax = 1 -] ] a) + a a-1 + C. 4am a-k1 .' - -j--J sin [(a-bix A = O. sin(2ax-G) + - +C. 4 bx1 = O. x + a.b)x sin +- O; a, h integers - sin(2ax a in 4 + C. a-b 4sin-- 2 _ b). eo8(a-.b)x -j--J2 sin + C. < k, x1= - 1ì + 2d. . -i- 23 2(2 00 v= 41. 3. .+(.)r+1ôr-1 + .'. d3u ' ux = - 5u& C + iu + 2 x + dx L I! dx3 1du = C + - ValueB 2n+1 n dx 12 1x d5u + 30,240 720 +)rn m udx n/rn u= m + I m-1 2 3f2+ ml [un+ UJ n du 2 - i d\ n/rn 45. in 2 24mL (in2- l)( - rn-i n-2 N +A2 - (° in + I r/203 n - [íxu+ ó2u] + i im + n 5 on page 25. (n+]. = 44. 1d3u dx3 lo 2 m-1 n-1 Au (m2-1)(19m2-i) 720m3 o m2_I[A4 +4u} - 3 In-3 o 46. C 47. ZC 4i. + - + (2n-1)x'1 00 n(n - 1)(n - 2) +---2x - r=). [ rn. (1z) 72Ox2' 1 (1)1()(ra+1)fl] 1nk_ki - s-logx 2-1-=c -¿rl LJ. = i ? 25 G Stirling's Nwnbers of the Second Jind, i i i 2 :i 2 3 4 5 6 7 1 1 3 i 3 1 4 i 7 6 1 5 I 15 25 10 1 6 1 31 90 65 15 7 i 63 301 350 140 21 1 127 966 1701 1050 266 2 Bernoulli's Nuribers B0= 1 -11 691 2,730 i 36i7 1 133 =55 x 15 1 17 1 510 43.867 798 174.611 330 137=55 B 21= 854.513 138 26 STJM&TION OF A SPECIAL TYPE POWER SERIES VI. The sirnation o1 a power series in which the coefficient of the k*th term i a rational integral function of k is 8tudied on the following pages. By sunniation iiean here, the replacement of the power series or infn1te sum by a finite expression, identical for values of the variable in the intercral of convergence of the por series. In other rds the problem of approximate evaluation of the infinite series is reduced to exact evaluation by substitution in an identical finite form. We proceed to devalop thin finite expression. Also we develop three recursion relationships 'which are of more practical use and which iield identical resuJts for this type of sination, Finaily e present a triangle, the sutnnation triangle, which, with a simple forirmla, nt oily provides the most rapid method for such summation, but is also easy to remember. The general form of such a series being written: (C0k+ s x) there the Cs C1l+ + c)xk, (p5) are constants and n Is a non-negativa integer, we have immediately the form: S (x ) = C Ol : kxk+ c + : f '. ¿ 1cl Thus the stated problem of summation may be reduced to the problem of evaluat4ng sums of the form: = ithere E k1xk, (26) In other words rather than n is a non-negative integer. studying the original power series (25), we str1y the power serios (26). K(x), The interval of convergence of (-1, + 1), may be readily determined by tho ratio test. Then n = O, we have: K(x) which is a geometric series. 1Ç0(x) We see Çx) Now Hence, = /x/<l. - (27) geometric series. is a particular oceeding from (27) we have: lk= I%(x) = c1 (x) = ° k=l (l-c) (28) 00 JL(x) ¿ = : 2= :K4 (x) = x(1 + x) kl (l-x)3' and in general z 00 nlc xP(x) .1 kx=xç_1(x)= K(x)=. n k=l may be seen to be a polynomial of deee n I by luduction. ')n Note that P2(x) is a polpiomial of degree one and that if is a polrnoinial of degree n of degree n. 1, then ?1(x) P(x) will be a poljnomial 20 Since K(x) satisfies the recursion relation, Ic1(x) = x K (29) (z), and since, X P (x) K(x) (1 ..x) (30) ii+Ï' have, x?,(x) dfxPn(x)_1 X)saJ L __ (1 ' vhence we obtain, by perforning the differentiation and removing X the factor, (1 ..x) P (x) s (nx n-s-1 +l)Pn (x) +x(1 P, The successive polynomials, easily by repeated use of (31), starting P1(x) = x)p1 (x). n P3S ... froia th3 , (31) can be built up known fact tbàt, Thus wo haves L1 P1(x) = i P2 =i +X P3(x)s1+4x+x2 P4(x) s P5(x) s -s- (32) lix + 11x2+ 26 + ¿ 26x+ x4 It may be noted that the coefficiente in these polynomials are smmetrica1, positIve, integral, and that the first arid last coefficiente are unity. If we assune that: :Pn(x) fr(fl)Xr_l, =:;l (33) 29 and substitute in (31) we obtain the recm'sion relation for the coefficients fr(n) as follows: = (n - r Since Ç(ii) -O, ie see that f(n) = i + 1)1' r-1 (n 1)+ r r' we see that f1(n) = i for all n. for all n. - 1). Since (34) f1(n) = O, Hence we see that the first and last coefficients are always unity. The other the coefficients follow easily fron (34). iroperties noted for Their stretry is shown later, (47). Using (34) w build up a triangular array o± coefficients for the polynomials P(x), shown on page 30. Looking at the geometrical neaning of (34) with respect to the triangle we find a de'vice whereby the law of formation may be easily applied and remembered. triangle. This law is similar to that of Pascal's Rather than adding t elements to obtain the one below we add multiples of the elements. by the number of steps we must take to reach the elements. The These multiples are determined frein the sides of the triangle ìethod is evident from (34). uewdoOAej ;o UO?tLUfl UÇ.ZJ1 eI{'4 .zoJ eut ewçi c E)q'4 t -t -i: t 1 i t ri: 99 9 t t-t 9 t t t t ot t t 9t t6tt 6t9't O9't t6tt ot t 6t9'I O6t'9t t O9't t Cr' o 31 The next step is to find a finite expression for 'iVe have by repeated application of (34): (n = (n = (n -r + 1)Ç1(n 1) r + -r + 1)f1(n 1) i» r(n - r)fr 1(n - 2)+ 1)-s. r(ri -r -1)f1(n + r2(n n + r -r)fri(n 3) f(n - i) = 1 when n + - i i<n = r or when i -1. = n - Hence, = r(n -r -i - + 2r-.i( i» r"'Ç,(r), (r - 1), or since f (r) = f r-1 r n-r+l f (n) r = r' (n -r -i + 2)Í'r Now we substitute r = 2 in (35) and note i <n (n - i) (n - i) (35) = i for to obtain, n-]. 2'(n f2(n) = - i). i=]. Summing by parts, formula 4, form 3, in the table, we obtain f2(n), f3(n), f4(n), and f5(n) successively, each time depending on the previous result and noting f1(n) - 2) - 2) +r'(n_r.4+2)Ç_1(n si)+r1fr(n -i); Now since f(r) = I, r2Ç(n 1, to obtain: r. 32 in = 2fl( 3fl_( f3(n) 4fl(fl 5-r 1'5(n) + 1) + 1)n + 1) 2"+ 3n + 1) (36) 2Z 1)n 2n (n (n + l)n(n - li 3Z 4n + 1) (n +l)n 3n (n + l)n(ri - i) 2n 31 (n + + i)n(n L)(n -2) . II '4. is evidently, From this an intelligent conjecture for r F (n) = r (l)fh(fl+l)(r -ni + i)'. inductive proof that The fol1owin () ni-1 m. Fr(fl) consists in = 'r showing, F(n) = + (n - r l)F1(n - 1)+ r Fr(fl - 2 r < n (3g) and, F1(n) = F(ri) = 1 in this order. Hence first prove: ïe (1)fli+l(fl+l)( n ( )(r-m) n-1 1)fl_ - -r (n r +r (-1) n rii+i ( + 1) 1)(r -in+ (1)m+i. 1)1 m=1 We proceed to reduce the express4. un the right to that on the left. Changing the sunmiand of the first terni on the right sud adjusting 33 through the limits of summation, r (n + have the two obtain: (_l)mÇhl2)(r i) + We we r rZ ml + (4)m+1( 1)fl1 )(r -in relations: _m-1 -n +i n I n+l., Ç_i" (39) -n + 2 - n +i n n 'm-l' ,n which follow imnecliatelç from the expression of the binomial coefficients in their factorial form. Substitution from these equations in the above expression ylelds: (n -r + i) r (_1)rn + m + If the r rL ml (_1)r exponents of -1 and (r _ (n -r + 1) r + -i n+l ni r - n nl (_1)fl i to r ,n+l in-1 are adjusted, this becomes: ,n+l, i)r l)(r-'n+l) n-in+2 - in sum may in = + ni-1 1)fl just as well be (n+l)(r-m+i) ( in-i )(r-m+ as the term corresponding to make this change, includo )(r-rn+i) n-i in-1 (l) iii+i limits of siination for the first from ir w + 2 -n + i) rn-_ The in n+l O is zero. for Hence the constant factors inside the swrxrnation, and add sums, we have: s/t r çUpon I (n-r+l)(m-1 (_m1 r(n.n+2) (n+l)(r-n+l1 + (n+1)(rn+l)j (:)(r - in algebraic reduction this becomes: mL and the first To ( -))fl1+l equation of (38) continue, we is ,n+l 1)fl, m-1 proved. evaluate F1(n) by substitutIon in (37) as follows: F1(n) Finally we show (_l)2(1) Fa(n) = 1. To do (40) this it will be convenient to consider an array which is our triangle augmented by the elements F(n - 1), n . F(n - 1) must be defined Hence to satisfy 2, thich by (37) as this defition - =E 1) (_1)m+l = m + 1)nl 1, = (E - which becomes by the binomial formula, (-l)' If we apply the causes it Thus, (33). F(n Now since satisfy the recursion relation of our triangle. E operator n+l zm=l we (_1)in have, (1)(x + n -n + 1)fll (41) 35 Hence we have, n+]. Afl n-1 1)XTI+1 I xD or as the term Lor ra 'x'I ( m1 n , n-1 )I%n-m+1j , m-1 = n + i equals zero, (1)m+1 IxJ ( - = -m (rn1)(n + (42) m=l p.l9-20) (1, If we compare (41) with (42) we have: F(n_1)=xhf n lx Iow an-1 = (n - 1) means n n-1 =0. ¿ X F( n - i) = 0. Ience we have: Now from the proved recursion relationship of (38), F (n) n = F n-i (n - i) then r + n F (n - 1), n or as F (n - 1) = 0, n F (n) n =Fn-J. (n -1). Since n i3 arbitrary, F (n) = F n n 1(n - 1) = F(n - 2) = ... = F1(1). Now since F1(1) = 1, F(n) = i. (43) 36 This completes the proof that, f(ri) =F(n), or that, ()Iìi+l Now with f(n) we have from (30) arid (')(r -m (33): (-i)()(r i (44) m + l)vEJxr_1, (45) [ or in a slightly different form, r1, , knXk k=]. i m+in+i. n Çn_ii(1)] r x I (46) (1 - The application of ali this information about (26) to (25) is Lmnediate. 37 A swmary of the results of th1 of their application is now in order. investigation and a discussion The methods and relationships mentioned at the first of this chapter have now been developed. have the finite expression of the swn (25) by applying (45) or (46). Wo have the recursion relationships (29), (31), and (34). The best method of evaluation of the swn is by use of the stwnation triangle with (30) and (33). In the suirnnation of (25) the value of methods depends on The magnitude of n, the proxidty of ¡xi to unity, the following: deee and the required of accuracy of the evaluation. considerations must be given each series. siaU but ¡xi oi.r Indidual For example, if n is is near unity our nethods are of value, but if n is large and ¡xi is snall the convergence of the series is not seriously slow and our results are of less value practically speaking. The evaluation of (25) for sorne particular value of x may he accomplished in various ways such as direct substitution in a finite niiher of terms or by expression of the polmornial coefficient of xk in factorials followed by sunnation by parts. JJovevr, the methods presented here are of gieatest value for this evaluation. For clarification we give typical examples where these methods are ìarticu1arly suited. First, let us evaluate the infinite power series, A J. + x + 2+ 27x3+ 64x4+ for x = 0.90000 to five decimal accuracy. ... + k3xk + a.., The exact value may be found by Í'orrnulas(30)and(33)and the summation triangle and substitution 38 îor x. Thus we have, s=i+E k3xk (-l)4x = = 2.2 = + Again using + c + x2) 43,69l. Second, evaluate the convergent irif:Lnite S (1 + - (z 3.22 332 2.3 2 2k + 3k2+ series, k4 + 34 + ... forrnulas(30)and(33)and the suimiation triangle, we have: 2k = +3k2+ i/ k()k3 =2 k=l + k=1 L = 172. Finally, infinite series finite expression for the sum k'th teri!l1 is the product of the k'th let us find a whose of an arithmetic progression, b, b + d, b + 2d, and the k'th term of a geometric proession, The series is convergent for r <1. a, ..., ar, b + (k 2, of an term - 1)d, ar. :39 ab (b + kd)ark 1k rk+ ad = ab + ab -a b-br+dr (1 A study of the siation interesting Sorne immediately frani its First, the - r)' triangLe. 'operties of the summation triangle follow law of formation. elents of each ro are syrìetrica]. about the element or elements at the middle of the row. = i n-r+i' r Ç. The proof is inductive with respect to n. n = 1. <n. (47) It is evidently true for If it is true for n1 we have the two relations: (n -r + 2)2 1(n) = (n -r + 2)f(n), 2 i n. = r r I_f That is, r r we add equals to equals in the above relationships, (n -r + 2)21(n) + r f (n) r r f 2 < = r n-'r+l (n)+(n -r = n-r+2 < = n. 1), 2 r e have: + 2)2 Now by (34) we have: 1) + 1. n. n-'r+2 (n), 40 Now since f1(n + 1) = f1(n r <n + 1. Thus the (47) is complete. J. Socond, this relation holds for + 1) = 1, for n symmetry holds + i and the proof of for each line the elements increase from the ends to the middle, that is, f(n) < 1' noven It is sufficient first inequality as the second first by the smetry just proved. proof is inductive with respect to n. Evidently the inequality is true for n = holds for n odd and prove it will Now we 3. n fr(fl)<fr+i(fl) for r inequalities: Hence we multiply the inequalities by add, and use (34) we r for frorii the srnuetry. = this inequality and n odd ç4ves us the <fr+i')' r, r n + 1, and 1 respectively, obtain: n even we use the -i and i er-i 1), r r+i Now ; r_i<r' fr<1r+i( assume holds for n + 1. Repeating for n even completos the roof. If (48) to prove the inequality follows from the The j,n_.l nodd = +(r)> f_(n) relations for r Proceeding as above we <fr+i + r for fr')<fr+i This = i n + latter equation is obtain: (n + l - i 41 Thus the induction is comleted and (48) is proved. An Abel sum. if a series diverges, it may he first combination than the (9, sums have a These sums are p.261-265). definition and not by analogy with cbviously these other some terms as an approximation to the i "sum" of the divergent series up by possible to use different of convergent series. sums meaning than in the case of convergent sanes, but they aro often of practical value. true especiafly when this sum is regular, to obtain the Now we p fl sum a from the that is, the definition of -i in the Abe1 This is methods used convergent series to the ordinary sum. have an example of Abels sum for the obtained by setting x = set oscillating series expression for }Ç1(x). We have sum, A = and (46), 00 : (_l)kkfl = n lini x-p-1 (1 -x)1 r ;;i I ()(r (_l)m+1()(r = ':;i by the method of Abel. =i - (_1)m+1. ;:i - mn ni + i)7J r + l)nJ (_1), (49) 42 K(x) for negative integral To new some gain perspective on our previous results results n = 1, n and O. we and to obtain some will develop a finite expression for K(x) for develop a method whereby K(x) may be approximated for particular -2. 'value of x where n relation (29), evidently true for initial condition, K0(x) e will use recursion any complex number n, in the form: klxk, kflxk = (xD) II with the arid (50) , for this extension. = Hence we have, kxk = (xD) -1 k=l k kl X = (xP) = (dx )(i -x) = -.2n(l -x) + C. Since klxk O for O, X l C klXk = O. This gives, -.Qn(l .ix), (51) The known expression for the sun of this common series. For n in the - -2 we obtain in the same way a form of an integral, an expected non-finite ex'ession result for the sum of such 43 a series. It is convenient to consider the form with limits O and x. The constant E,-2k c X =-\ /o k1 in its definite drops out and we obtain: Pn(1-x) X x2n = (52) f X I 10 Repeating intea1 this process for successive negative inteers, obtain a multiple integral of multiplicity (n - 1) for K(x). ve :re have: E. kxk= kl (x (X /0 I0 From a practical 10 11n x 1x ()fl1 (3) standpoint approximate values of these integration of (53). series may be found by numerical of (53) decreases as ¡n! increases for infinite The utility two reasons, the integration bocomes more tedius and evaluation by other means becoies easier due to the increasing convergency of the series. series converge for x = -1 when n - 1. Note that these 4h BIBLIOGRAPHY VII. 1. 2. Bode, George. A treat1e on the calculus of finite differences. 3d ed. New York, G. E. Stechart arid Co., 172. 336p. British association tables, 72p. for the advancenent of science. vo].. 1. London, The Cambridge Mathematical University press. 3. Fort, Tomlinson. Finite differcnce ai difference equations in the rea]. domain. London, Cc.ford University press, l94. 251p. 4. Jordan, 5. MIme-Thomson, Charles. Calculus of finite differences. York, Chelsea publishin; co., 1947. 652p. L. ì. The calculus of London, Maca11an and 6. Steffenson, J. F. Interpolation. .lkine co., 1927. 7. co., ltd., finite 1933. New 2d ed. differences. 558p. Baltimore, The Williams and 248p. Mathematical tables project. and digaurna function for complex arguients. Now in process of computation. U.S. Bureau of Standards. The ga!nna . Whittaker, E. T. and Robinson, G. Calculus of observations. 1944. 397p. 4th ed. London, Blackie and eon, 9. Widder, David. Advanced calculus. 1947. 432p. ltd., New York, Frentice-HaU inc.,