SUMMATION OF DIVERGENT INFINITE SERIES BY ARITHMETIC, GEOMETRIC, AND HARMONIC MEANS by FREDERICK HARRIS YOUNG A THESIS submitted to OREGON STATE COLLEGE In partial t'u1fi11nnt of the requirements for the degree of MASTER OF ARTS Juno j913 APPROVED: Professor of Department of Mathematics In Charge of Major Head of Department of )4athematics Chairman of School Graduate Comxuitte Dean of Graduate Sohool ACKNOWLEDGMENT The writer wishes to express his deep gratitude to Dr. C. L. Clark for the latter's skillful direction, timely help, and abundant patience. TABLE OF CONTENTS Titlo Chapter I. II III. INTROI5ITCTION. 1 , . . s., . * . ...... , . , . . . .1 TYPE S OF REQULA.R SUNMATI ON À. Cesàr'o Summation.... B. aölder C. Nörilund Mean. Suat ton. . . . , . . . . . . . . . INTEGRAL I1EQJALIT).. . .. . . . . . . . . . . . . . . s s . . . . . . . s a . * ... .3 . . . . . a . . . b . . a .9 EAN HARMONIC-GEOMETRIC-ARITHMETIC INEQUALITY. IV. . . . .10 I. a a . . . . . . . . . . .1) V. SUNATIONBYIIâRMOXUCNEÀL...............ILI VI. WEIGHTEDiiARMONICMEAN....................21 VII. ZUMMATIONBYGEOMETRICKEAN...............22 VIII. Ix. WEIGHPEDGEOMETRICMEAL..................2k X. XIs DIVERGENT INFINITE PRODUCPS...............25 FURTHER EXAMPLES OF HARMONIC MEAN COMPARISON OF PROPERTIES C? CESARO ui HARMONIC MEANS. . . . . . . . . . . . . . . . . . . . . . . . . .29 SUMMATION OF DIVERGENT INFINITE SERIES BY ARITHMETIC, GEOMETRIC, AND HARMONIC MEANS I. INTROL1CTION Prior to the time of Cauchy and Abel, infinite series vere used with no regard for the question of conIf a function could be expressed as an infinite vergence. sortes, well and good; the eones represented the function. This carefree attitude, which led to many discoveries later validated by sounder means, was brought to a sudden halt when Cauchy and Abel showed that the use of series could lead to incorrect results. sweep, divergent divergent Thus, with one series were relegated to the limbo of things tried and found wanting. Not until ne8rly the end of the nineteenth century did these treacherous traps for the unwary begin to regain Frobenius, in 1880, an aura of respectability. extended the idea of convergency by employing the limit of the arithmetic mean of the partial sums of a series instead of the limit of the sum of the first n ternis. The work of Frobenius opened the way for a host of workers. vniue to A way had been found to attach a unique certain nonconvergent series. Further, this method, when applied to convergent series, yielded the ordinary sum. Within a decade Hölder, Cesro, and Borel had successfully generalized the method of Frobenius, 2 and the early twentieth eentuiy wiis a period of intürthe and fruitful activity in suimnation. Certain terms 8tloUld be defined at this point. A "regular" method of summation ta one that will yield the ordinary value when applied to a convergent series. when applied to a nonconvergent yields a unique result series, the result is known as a "conventional" sum. Similar terms apply to If, in addition, this method el; o infinite products. The purpose i. of this paper is es follows: To define two new methods of summation, those of the harmonic mean and of the geometric mean of the partial sums of a series, and to show these methods to be regular. 2. To define a conventional product of an infinite product as the geometric mean of the partial products and to show this method is regular. 3. Po compere these methods of summation with those of Ces.ro, Hölder, and Nörlund. 3 TYPE S OF REGULAR BUMMAPI ON II. Cesro Summation One of the moat useful methods of regular sum- mation that of Cesaro. 1.3 cO n (o) Let u. Consider the series . U. k=1 n (i) and, = in general, let k=1 C 1(r)) (r) (r) Let (r-1) n (r) (o) Let rl(n tri + n - rfJT s] CJ if for some r, then, ; has 00 a defluite limit, the series u1, 1 is snid to be sum- k=1 mable (C,r). Chapman and Knopp S. Cesro's method to the case bility, is not r, (r) (5, in which the order of summe- necessarily integral. + n) denotes n (r+n 2...(r4- (r - +- 11 jr + and r ) (o? = s - (p /) (o) s = fl(r or rCn + 8n1 p.26k), let n n. + i.) 2 For a proof of the (7, Then (r) , where ( p.70) have extended + n + i) i7Çr " (o) 8n-2 regularity of Cesro 1/r + n - 1\ n cO Ak=1 uk=lime n - _! -n:--- Jo sumnii.tion and s 1f Case 1. A n > O such that Given any (L (m il' + < 21' + I there exists an > O, Hence, if n -n vfl = Thus, !T .... + + - + s i n n fr-/E, Case 2. to zero, and ve lin i orltmç=O. therefore, lin O. my apply case I to - A) (s, n-' P, s, of unity, mny converge to !, , Sn' w LA(.Y single lixaitin o a Howevei, the set of points centioid of 2J 2' happen that they do not converse O; ' A. which rcpreent the numbers is the - A) = lim(cr tends That is, it. n If we consider the points where - Then the sequence A n-'nk=1 point. {21 + nl , ... n' 2' each wtth a masa which thsn represents the Cesàro sum. For an illustration of Cesèro's method, lot us consider the series I (o) - I + I - i + (i) =1 32n ... . Then 5 (i) (o) =0 (o) =0 52n (o) 8 2n + I 22n+1n =1 Then C2 Thus, (i) (i) (i) n I C2 and 11m C ri + 2ñ+I n- 2n+1=. the original series is summable (C,i) to the value 1/2. For an illustration of Cesro summation of a series not swnmable (C,i), let us consider the series u = (o) si I(_l)kk. k=1 Then =_i si =_I 82n = = '=-2 (i) 83 O 82n =-2 (i) (o) = O = 2 ......................... (o) (i) $2n_1 82n-1 (i) (o) 82n = S - I -n(n + i) (2) (i) (o) (o) S3 (2) (i) - O -n(n + i) 3 (2) 21(2n - i f2n - I + - í-n' t) 2- i) (n + ) I7T (2) +n ;limC n-1= n -1 2n (2) 212r1 - (_)(n + i) i- 2n (2) limC 2n _-1. (_i)kk is suinmable (C,2) to the Thus, the series U = value k=1 -rn. omehat different, For another illustration, sin la where x is not an even consider the series U = k=i 7. multiple of Then n s sin la = n cos x/2 - k=1 2 - cosjn+í/21x, and - sIr[x7 n n = n s k=1 n cos x/2 n 2In cos (k+1/2)x. I I1iE7 c7 Ic=1 n k=1 n + - - hm sununable O (niJx ± (c,i) ° that I cot x/2. 2iTh/2 = if x is an even niultiple of r, cos x/2 = However, U = sin iIn X7 2sTh x/2 fl n-*oo - sin x n cos x/2 flOE-. Therefore, -sin x + sin (n+i)x. 2sTn x/2 cos (k+1/2)x Now, O + ... and lin noo = O. to the value I cot x/2 if Therefore, U is x is not an even 7, multiple of r 8nd to the value O otherwise. roi L!J Holder Sumnmt ion H6lder's method is siniilar to that of Cesàro, and the two methods may be shown to be equivalent (3, p.85). n where Consider the series (i) + Let h s n value of r, has Uke and, in general, let (r-1) +h. (r) If 11m h L. k=1 +...+a, (r-1)(r-1) h1 4 h2+... (r) s r definite value for some integral - the sertes 2 Uk is said to be suinmable (H,r). Since Hldor's method is equivrlent to that of Cesro, it is necessarily regular. 9 Nórlund Meen Consider the sequence consG1r1ts with C0 O, fc1} real or complex oi' c. and let C For any k.o ri 00 sortes 1 u, s where and k=o k=o cl-j:1 Uk, CS + + - um n + CrìSc let If li S "n On 00 bae a definite value, the series O, - n- - u U said to be k=o summable (N;c). The method of Nrlund's Mean can be shown to be a recular method of summation (k, p.37). It is interesting to note that if all c Nòrlund's iethod Hölder for te :Ls identical with those of first order, are i, Cesro and III. HARMONIC-GEOMEPRIC-ARITBJIETIC MEAN INEQUALITY inequality A fundamentsl which we shall find of great use is the hrmonic-geometric-arithmetic mean in- equality. The (i, p.)25). proof presented here is due to Cauchy First, ve shall prove that the geometric mean of several non-negative numbers A,B,C,D,... is always less than their arithmetic mean. Let n be the number of letters A,B,C,D,... is . It sufficient to prove that in general A+B+C+D+... BCD.Tt (i) n or, what amounts to the same thing, that n -/A+B+C+D+ ABCD... (2) n = (--- - In the f trat place, for n = 2, + B2 ___ - A - 2 - ( 2) ( ) = and one concludes, by taking successively n = 1 and finally n = 2m, J n = k J ' ________ ________ ¿ - , that - ABCflEFGH = 2 A + B B ¿ _// (A+B+c+D+E+F+a+H8, _- 's___ __-) n - -- + ...) ABCD... - (A + B + - C + D 2m -_-- -J term of the In the second case, if n is not (3) , ' 8, geometric progl'e8Siofl 2,l,8,16,..., let us designate by a term of this progression larger than n, and we shell let K A + B + C + D + n Then, returning to formula (3) and suppOsing in the first member of this formula that the last 2m - n factors are equal to K, we twve LJA + B ABCD...K2 +C + J L or, in other terme, ABCD...K' Then, multiplying both members by K' .ABCD... (A+B K , we have C+D which was to be proved. Nov, denote by A(a), O(a), and H(e) the arithmetic, geometric, and harmonie means of a finite number (n) of nonnegative numbers. lin A(a) uni Also, for all a and i/a(e) = 0(1/a). Thus, 1/O(a) 1/il(e). - lin Hut) exist. O(a), ) O, , we have if the8e limita exist. 'e know that A(1/a) = 1/H(a), Therefore, i/o() 1/H(s), or H(a) 0(1/a) lin G(a) li n- n-*oo i(a), A(1/a) = O(a), and lin O(a), if these limits cxist. lin li(a) n- Then, as n-* Therefore, if the limits i'. If the 1imit of A(a), G(), and H(a) do not exist, then the inequality holds true for the upper limite, least upper bounds, etc. of the functions. t) IIEGRAL IV. I1EQ.UAIJITY An interesting extension of the preceding inlet us define, equality is the following: terval O < A(f) over the in- ç X2, f(x)dx, J x2_xix1 i X -(2 H(f) 2 - log f (x ) dx ./ G(f)=eX21 X1 where f(x) has a positive lower , dx Ixl bound. Á(f). G(f) Then H(f) The general proof of this theoreni has been given by Hardy, Littlewood, &ad Póly inequality H(f) (2, p.135). However, the A(f) may be proved by Schwartz's inequality. The inequality is eviaentlj true ir I I (x,. C. -x)2 - ax / I f(x)dx, but by Schwartz's d'-r i î inequality, rfX212 the latter member is not smaller thanf dxj L'j. which is obviously (x2 - x1)2. 3 Hence the inequality is proved. It can e&sily be shown that the general integral inequality still holds a x increases without bound. 1k V. SUMMATION BY HARMONIC MEANS It has been shown that the summed by suma. an uso of arithmetic infinite series may means of be the partial We shall now show that this may aleo be done by harmonic means as well. the method of harmonic Further, means is regular; that is, when applied to a convergent series it will yield the ordinary sum. Consider the series u where again we shell k=1 n let s S = 11m U. = sn If the series is convergent, We must add the restriction that none of the partial sums nor S be zero. um let 1, i. and it has Then, been shown in section II that n- lims1+e2+...+s S. 4CO n Let p The p1 I and P Then lin 1. p= P. may now be considered as the partial sums of some Thus convergent series. ... 11m p1 + p2 + n __ n+ool+r hm 1 s2 + p P, or = ..n Hence, if a series is convergent in the ordinary sense, the limit of the harmonic mean of the partial sums is 15 equal to the ordinary sui». If, That i», the method however, the series to be tested is not ordinarily convergent but the limit may be We series. exists, the conventional sum for the limit of the harmonic mean considered to be a shall denote b,r H the lii» u +.. i s It follows when such exists. the series H. is regular. or + s 2 - 00 , so F1180 will k=1 The most interesting application of this method, then, will to series be whose partial sums oscillate between infinite limits. For an example, let finite or F1 Then + I - I + I - I + us consider the series ... 81=a a + 1 83 = s. a + i s 2 n-1 =a =a+1. 2n-1 H21_ +T. 1 52n Nov, n definition that if from the diverges to u I i a-+1 ¡ I 2n - ___________ a and +1J lia H 2n-1 i + u + n - ¡ (2n - i)Ça)Sa + 2ia-+ 2a(a + 2a+1 i i i), -1 Also, 2a(i. 2n__ n H2 1 ñ+ swnblo Thus, the given serLes is to the value 2e(a From by i). + this result we can see that mean sum of a nonconvergent series t, chancing That linear. the harmonic mean haraonic not necessarily a by an amount b does not hmonic necessarily change the is the mean sum by b. Let us investigate the result of interpolating an infinite number of zeros in the example. Corwider the series series of the previous a+1+O-i+i+O-1+1+OThen s=a+1 83n+1. = = a + n H3n+i = Thus lUn Aleo 11 n+1+ a ! H31 +2 = n 3n+ -i:--- and 1imH32 + 3a+I +I+ - 3aIa + The harmonic mean sum (n 1)ja1a + )na+n+a+1 2n 3a(a - i) 3a'a + 3n 2h Now And j i- i) i) on 2 + 2 3naÏ- 2ii-:i:--I (a)[ n+ + 1 ¿a + 1 ----:i:--i i) is now 3a(a a + i). +-I-- Hence, the (; 17 interpolation of an infinite number of zeros may affect the sum. This is also true of Cesro summation. This interpolation does not always affect the result, for an examination of the series a+1+O-1+O1+... will show the harmonic mean. sum still 2a(a + , as it was without the zeros. The typo of serios most frequently encountered in summation by means is that number of terna, of terms, and O H n .-. = in. which, after a finite (k) is recurrent < n K with a fixed period (a) M, n + and M finite. ni ---+ . ..- ... + (pk1+...+pk+a)=A. ThenHk+ For convenience, let (p1 + + and let fl8 k+na = K+ii k + na + d, where d is an integer such and H that Then I d a, and D = p «. i- k + D, can take on at most a finite (a) number of values. then, Then liuiHk+na=lifllHk+na+da From the last result it is evident that the restriction must be added that A O. Under these con- ditions, then, a series is summable by the harmonic mean method. It may easily be shown. that it is al8o summable under these conditions but not necessarily to the same value, as we have seen in some of the previous (c,i) example s. Suinmability by Cesro means does not necessarily imply suimaability by iarmonic means, nor is the reverse true. For example, consider the sequence N0E: 2' 1' 52n. where the pair 82n - i' 1' ' or 1/10, 19/10. For n = 1, 2, ... 82n' 82n is either 1, (i) i (i) LU + , and = either I or (i) Hence, liza C 1 i. 2m + 1/10 I = _____________________ , Now, Hn - I-:: I +-.; I + n 'A +1 I 82n-1 2 82n, 2n where the Then pair I 82n-1 i.. î. , is either 1, I 1 or 10, 10 82n th aritbmetic mean of n pairs which are 2n either 1, 1 or 10, 10/19. The aritbnietic is 1, and that of IO, 10/19 is 100/19. follows: Construct si, 2' ' as mean of 1, I let the first 19 two terms be 1. 1, Then 1. 1 Now use the pair 1/10. 19/10 repeated enough times so that the reciprocal of the harmonic mean for these terms exceed8 k. Next, use enough of the pairs 1, 1 80 that the reciprocal of the harmonic mean is less than 2. When this process is continued, a series is obtained for which the Cesro sum is 1, but the harmonic mean oscillates, never approaching a limit. Obviously, a similar method could be used to derive a series summable by harmonic neans but not by Cesro a1eeris. It is nov A slight change in notation is necessary. monic means. Let possible to define higher orders of harn : -y then , n We have seen that i1' i 8 1 -p ; . - Then 8. let n -F...+ + 1? Since 1H} 2i S. is a sequence approaching the value S, In general, H_____nrr.______ -- r n . r - and rHn - r - -, S We may now state a definition of order. If r is 20 the smallest integer such that urn ki u is H n exista, then summable (rH) by harmonic means of order r. From the above, it is apparent that if a series is summable (rH) If k1 u it is sumniable is convergent, positive integer. (r + it is 1H) to the same value. sumrnable (rl!) for r 21 TRE WEIGHTED HARMONIC MEAN VI. 00 Cone ider the series . k=1 If the series is convergent, n u, where s urn s u = Let S. O. k=1 I Sn Then P = urn p n = urn I flPco I . Consider n Cn , =a1+cn_1P2+ n C = , and um O. Then en IP2 + + n-Ç k=1 + where s.. w=____ --cand hm I = P I (by Norlund's Mean). flc0 Thus, him = S, and this method is regular. As with n-400 the Närlund mean, if all c the ordinary harmonic mean. = 1, this method reduces to 2 SUMMATION BY GEOMETRIC MEAN VII. fl 2 Again consider where u1 s k=1. n = u k=1 k s - We may define a geometric conventional sum as liai (8152...8)hl'fl. Let G follows: e)h/'1 Then = G, the geometric conventional sum. (2152.. Phis method is regular, for by section III we saw that H G A, and if Uk is ordinari1 I convergent to S, k=1. H A = S. Thus, G also equals 5. exist when S does not. However, G may In that case G is a conventional sum for the series. There are certain restrictions that must be Clearly, in order to placed upon the geometric meen. have a value different from zero, none of the partial Also, sums can be zero. in order to avoid imaginary we must insist that the product of the partial results sums be positive for al]. n where N is a finite N, integer. For an illustration of this method, let us conaider our familiar series, a + 82n = 52n + i - i + i - i + ... a + I a I G2n = [a(a + 1)nj 1/2 I/2n [a(a + i)J O. 2 n+1 I rn + + . = La 1 (a + I )fl]2fl+'l a ri+L ri a 1)2flI 1/2 Then, urn = ' fl-CK Thui, 2n + i the serles te suiuinable by the geometric mean 1/2 to the valuo = [a(a + 1)7 lIa(a + l)J VIII TRE WEIGHTED GEOMETRIC MKAN 00 where Once again consider the series k=í n are greeter then zero. and ali = Consider k=1 i s1)7vhere as defined in section VII. nd 11m in G Therefore, G + . = lin in S if the s + e1 ina . = in S, provided this exists. is convergent to &eries k=1 S, ere Then - in the e1 and C and the method is regular. X. DIVERGENT IN?INITE PRODTJCT In considering infinite series we have worked with conventional sums, regular methods of summation that yielded the ordinary sum when applied to convergent series but that also gave rnrnerical values wten apolied In a similar way we can to certain nonconvergent series. f a method of evaluating infi1te products that is md regular; that is, Lb yields the ordinary product when applied to a convergent infinite product but also may be extended to assign a numerical value to certain noncon- vergent products. Consider the product Let . G , is convergent, e and let p = a1a2...a. Then we have in p1 + in p2 + ... + -n in G If TTa1 in p (in thon the sequence ) converges to lnîíwherelTis the ordtnry value of the infinite product. Hence, by arithmetic mean summation, in G - in iT, does not or urn G exist, but um =TF. does n-ì°o 0tl However, if um exist, then urn p 0ri may be co considered to Since be a conventional product of this method is equivalent IT a, to that of Holder, 26 extend it easily to higher orders by taking the limit of the root of the partial products of the partial product s. Let us apply this method to the infinite product we may (1)(2)(1/2)(2)(1/2)... =1 p1 . Now =2 p2 ........... =2 2n 2n = 1 + ± Obviously, the product does not converge. flowever, 1/2 n = 21/'2, and (i)(2)(1)(2)...(2)] I 02n + Then, uni = lu = [(I)(2)(1)(2)...(2)(1)] a2 + the conventional product. 21/'2. n = Therefore, 21'2 ja r L FURTHER EXAMPLE$ OF TILE &RMOtIC MEhN METHOD Consider the sertes In tht8 SetieE, first ten I I U=1_ I I + r I I - I I + 7 - + I + .pplyirg, tc hernicnic ìieri method to the approximation, we find that The aun of the first ten terms is 0.61456. terms to obtain 0.685. &n cf the infinite series is In 2 or 0.693. Hence, the method of the haxinonic mean is in this case a better approxiiwtion than is the partial sum. Next, consider the series The velue U=2-5-2+7-5-2-i-7-5-2+7for which =2 83 ... . 0 s. 2 fl + I = 2 83n + 2 result in section V concerning series with oscilleting partial sums are satisfied with Here the conditions of the a = 3 and A = (-1/5 expect the result + lin )-3X 1/2 - 1/3) L = - 1/30. )/(-I/30) = -90. Hence, we In order to 28 check this', we find 3n -n + n + I = + 2 tht -90, - n + - 33 __n +2 -n+1-1 , 11m n-'°o -90, anc + 2 = -90. Thus, the series is summb1e (1H) to the value -90. XI. COMPARISON OF CESARO AND HARMONIC SUMMATION Consider the series (s/ M, M vergerit, let its sum be (i) + and C n D = --I W - + k=1 where s k=1 finite. If the series is let 4H con- Let denoted by S. n , and - a, U = I. 3_. n - We may i 3 J- n now state the following properties or characteristics of Cesàro and harmonic summation: () is convergent, it is suxnmeble (c,i) and to the value S. 2. If U ta eummï.ble (C,r), it is eummble (C,r + i), and if it is aummable (rH) it is simmble r + 1H). then both the or - 00 3. If U diverges to + arithmetic and harmonic means diverge to +aO or .00 k. IC U is suinniable (c,i) or (1H), it either converges I. If U or oscillates 5. If U fl (i) (i) C1 between finite = is said to 11m Cr = U fl-3 6. is said to and be bounded (c,i). n1n=i;, n- ' or infinite limits. finite, thon Similarly, if I11n' 'nd$ftnite, then be bounded (1H). The interpolation of a finite number of zeros does 30 not 'T. or harmonic $um. The interpolation of an infinite number of zeros my 8. chníe either the Cesro chance both the Ces&ro and harmonic sums. If U is summs.ble (C,r) and (.H), then rU is sum- mable (C,r) and (rH) to r timos the former velues. BIBLIOGRi.Pift 1. OEvres coniplètes, series 2, Cauchy, Augustin Loui3. 512p. Parie, Gauthier-Villare, 1897. vol. 3. 2. Hardy, G. H., Littlewood, J. E., and P6lya, G. Inequalitiea. London, C8mbridge University press, Jlkp. 193k. 3. . 5. Hobson, E. W. The theory of functions of a real variable and the theory of Fourier's series, London, Cambridge University press, vol. 2. 1926. 78Op. Summable sertes and convergence Moore, Charles N. factors. New York, American Mathematical (American Mathematical lO5p. Society, 1938. Society colloquium publications, vol. 22) Aufgaben und lehrsitze aus Berlin, Julius Springer, 1925. Pólya, G. and Szegó', G. der analysis. JJ8p. History and synopsis of the theory Eugene, of suznmable infinite processes 2:8. 175p. 1925. publication, University of Oregon 6. Sznil, Lloyd L. 7. New York, Widder, David V. Advanced calculus. k32p. Prentice-Hall, Inc., 197.