Professional Documents
Culture Documents
ANDCONDITIONALLY
CONVERGENT
SERIES
WHOSEPRODUCT
IS ABSOLUTELY
CONVERGENT*
BY
FLORIAN CAJOKI
§ 1. Introduction.
It has been shown by Abel that, if the product :
71—«
I(¥»
71=0
+ V«-i + ■•• + M„«o)>
of two conditionally convergent series :
71—0 71— 0
Z(-l)"+13,
7>=i n
* Presented to the Society April 28, 1900. Received for publication April 28, 1900.
fMathematische Annalen, vol. 21 (1883), p. 327 ; vol. 2(5 (1886), p. 157.
%Mathematische Annalen, vol. 24 (1884), p. 42.
§ American Journal of Mathematics, vol. 15 (1893), p. 339 ; vol. 18 (1896), p. 195 ;
Bulletin of the American Mathematical Society, (2) vol. 1 (1895), p. 180.
|| This may be a convenient place to point out a slight and obvious extension of the results
which I have published in the American Journal of Mathematics, vol. 18, p. 201. It
was proved there that the conditionally convergent series :
V(_l)M-lI (0<r5|>).
Sí n
when raised by Cauchy's multiplication rule to a positive integral power g , is convergent whenever
(î — l)/ï <C r ! out the power of the series is divergent, if (q— 1 )¡q > r. The case for which
(q — \)¡q = r was left untouched, but an examination of the formulae appearing in that article
readily yields the result that in this case the series is divergent. Hence the condition for diver-
gency should be written (q — l)jq = r .
25
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
26 F. CAJORI: DIVERGENT AND CONDITIONALLY CONVERGENT [January
convergent series, yielding powers of the series which converge less rapidly the
higher the power and which for a sufficiently high power yield a divergent re-
sult, suggest the following questions : Does the product of two conditionally
convergent series never converge more rapidly than do one or both of the factor-
series ? Can the product of two conditionally convergent series or of a condi-
tionally convergent and a divergent series in no case be absolutely convergent ?
The first doubt of the correctness of a negative reply arose in connection
with the conditionally convergent series :
»=» (_ iy+i
¿=i n
This yields convergent results, no matter to how high a positive integral power it
may be raised.* If the series is squared, the constituents, l/(n — r) (r -\- 1),
which go to make up the nth term of the product, namely,
r=n-l 1 1
£o n — r r + 1 '
are all positive or all negative for that term. Hence the numerical value
of the nth term is the sum of the numerical values of all the products
l/(ra — r)(r + 1) . If the square of the series is indicated by
»1=1
then all the constituents, an_T • l/(r + 1) , which enter into the composition of
the 7¿th term in the cube of the series have all like signs in that term. Simi-
larly for higher powers. Now it has been shown that all the positive integral
powers of the conditionally convergent series :
n=°o (_ 1^"+I
Si n
continue to converge, even though each term of a given power of the series is
numerically the sum of the numerical values of its constituents, and it is evi-
dent that, if the constituents of each term were partly positive and partly neg-
ative, each term would be numerically smaller. Can the signs of the terms of
two conditionally convergent series and the numerical values of its terms be so
chosen that the product of the two series is absolutely convergent ?
In an important article on the multiplication of conditionally convergent
series-j- A. Pringsheim states that he knows no a priori reason why there might
not be special cases in which such a product is absolutely convergent. " Aller-
dings," he continues, " war es mir bisher nicht möglich an irgend einem bestimm-
ten derartigen Falle die unbedingte Convergenz zu constatiren, und es dürfte dies
— wenn es überhaupt solche Fälle giebt, was ja immerhin als fraglich erscheinen
muss — wegen der zusammengesetzten Beschaffenheit der Reihenglieder wv eben
nur sehr schwer gelingen."
_____ 1 1 1 1 1 1 1
a _ 2?i+ 2 2«,+ 2 ~~2w-l 2w- 1 ~ 2n + 2 2n+2~ ~ 2n-l 2n-l;
2/i + l 2n + l'
In the first summation we are adding 4n' -f 4 constituents from one end of the
column or series yielding the (4n + l)-th term of the product ; in the second
summation we are adding 4/j' -f 4 constituents from the other end. In the
middle of the series the sets of constituents will overlap, if n is odd, and will
not touch if n is even. When n is odd, then a represents the sum of the consti-
tuents which have been counted twice ; when n is even, then a stands for the
negative value of the constituent not counted at all. Each of these constituents
is of the degree — 2 with respect to n and as their number is finite, their sum
cannot be of a higher degree with respect to n than the degree — 2.
The (4«,-)- l)-th term, simplified, is
p=n'/ 2 _ 4 \
2Sv47+T (4«-4^ + l)(47i-4p-3)J-a-
For any positive integral value of n the fraction 4/(4re — 4p + l)(4?i — 4p — 3)
has its maximum value when p = n'. Writing n fovp, we obtain a fraction
which is less than 1 /n1. Hence for any value of p,
1 4 11
-<7——-,-
4p + l (4w—4p -f-l)(4/i —4jo—3) 4/>-f 1 n
It is easy to see that
2k 2
2C-< a log (2n),
n=i n
547TÏ<alog(2w)'
and
PSi! 1 4 e1 1 1 alog(2re)
¿4^1 (4n-4p + l)(4n-4p-8) < ¿47+1 ñ2 < n1 '
But a number m can be found, such that, when ?i~> m , we have
a log'(2/i) 1
w2 /i (log w)s '
where Ö > 1 . Therefore, for n > m , the (4/i + l)-th term is numerically less
than the nth term of a series known to be absolutely convergent.
«?/ i_ i i_i
¿¿ \~ An—4p + 4 4p + 1 4n —4p + l 4p + 4
1 _1_ 1_1 \
4/i —4p 4j»-f 1 ~~4w—4/>—3 4p-\-4J
'-7 1_1_ 1_1_
^o\ 4n —4p + 4 4p + 1~ 4n —4p + l 4p + 4
11 11
= 0,
2/1+2 2/1-1 2/1-12/1+2
when w is odd, and
1 1 _1_ _1_
a_ +2/1 + 4-2/1 + 1 2/1+1 2/1 + 4,-'
when n is even. We see that the entire (4n + 2)-th term is identically equal to
zero.
The (4n + 3)-th term of the product-series is
«7 1_1_ 1_1
¿Í\+ 4/i —4p + Í 4/T+Ï —4n —4p + 4 4p + 4
1_1_ 1 1 \
~ 4n —4p + 1 4jp +1 4/1—4/) 4p + 4)
^'/ 1_1 1_1_
+¿ \ 4n —4p + l 4^ + 1 —4n —4p + 4 4p + 4
111
4n —4p-\-\ 4p + 1 4« —4/> 4p + 4
1
i o„
2/1+ 1 2/i + 1^2/i +i 4I O™
2w +i 4,4 ^l 2/1+ 1 2/i + 1 '
when n is even. The reasoning which we applied to the (an + l)-th term will
show that this term, too, when «>m, is numerically less than l/n(logn)e,
0>1.
We have now shown that the 4?i-th term and the three immediately succeed-
ing terms are either identically zero, or, for «. > m, numerically less than
l/n(\ogn)e. Hence the product of the series Sx and S2 converges absolutely.
Again, take ■the conditionally convergent series, obtained by dropping the
parentheses in the series :
7= 7.-1/
p=n—\ J ^
5 V 6n-6p
6> +
+ 3 6^
6>4+ 1 +^6w-
fi„ ß„
6> +i O2 ß™
6j>+i o2 +T6w-6> + 1 6^ + 3
1 _j._ _1_ 1 1 1 \
+ QnT—QJ)
+~3"6>TÏ ~"6n - 6j9+ 2 6>+l! ~ 6n - <op+ 1 67+~3/'
is seen to be identically zero.
The (6ra + l)-th term of the product of Tx and T2 is
f.u_-_-_-
¿SV^6n-6p-|-l 6/j + l 6»-6 6^ - 3 ñp + 2 6w- 6p - 4 6^>+ 3
1
6n —6p —5 6/>+ 1 + 6»i
5-ñ-0
—6p —3 5—r-H
6jj + 2 + 6?i—6p —4 6jo+ 3
p=n' /
11 11 11
M + a--_ T7 5-rr + _-5-6 «—7^ +
Qn- 6p + 1 Qp+ 1 6« - 6p - 3 6>-f 2 ' 6_ - 6> - 4 6> + 3
_1_ _1__1_1_1_ 1 \
6n-6p-6 6/>+ l 6ra-6>-3 6> + 2 6rc-6>-4 6j_+ 3/ _a
_ *=%/ 1 -6 \
~ èi\6>+"ï (6n-6> + 1)(6»-6^35)J _a'
1_1_ 1 1_ 1 1 1 1_ 1 1_
a~ 8w 8» —1 3/^Tl 3» 8» —2 3ra^2 + Wn3»-1 + 3/i-l 3w
1 1
~~ 3/1-2 3w —2'
In explaining the further process of constructing our series we shall start out
with a particular selection of signs, but the process is the same as for any other
selection, similarly made. Let a = 4, 6 = 3, then c = 2 and d = — 1, a
factor of c . Accordingly, the signs of terms in each set of the first series are
+ H-. Let the signs of the b terms in each set of YLv be + -)-.
We have now the two factor-series obtained by dropping the parentheses from
the series on the right hand sides of the formulas :
E«m-E({«3,
p=l
+ r51}+ K + i + /3j-{^ + 2 + ^}).
In the product of __wm an(i __«m7 a set w^l consist of 12 constituents. The
first set in the (12w)-th term will be (since {vX2—9 + ßx} = {v3+ /3,} , and
similarly in other cases)
(I)
+ í«u.-8 + a;} {«„_ 3 + #} + {Mi2n_4 + «¡} {%- 2 + /3'2'}
+ {«_h-6 + aJ {«.2-1 + #} - {ul2n- 6 + a3} {v12 + /.;"}
+ {•«, - 7 + a2} {.u + 1 + /3'2"}_ {«^ - 8 + a,} {vl2 + 2 + ^"}
Here each of the quantities ßx, ß2, ß3 occurs four times. We have broad-
ened our assumptions by attaching a single accent to each letter the first time
it is repeated and a double and triple accent the second and third time it is re-
peated. Accordingly ßx, ß[, ß'[, ß"x , ß2, ß'2, ß'%, ß"2 , ß3, ß'3, ß'3, ß'3" may
in general represent each a different number. Similarly, the letters a{, a2, a3, a4
receive a single accent the first time they are repeated in a set of constituents
and a double accent the second time they are repeated.
We shall limit ourselves to the case where, in any term of the product series,
each set, when multiplied by n, yields a product that converges toward zero
when n = co. Accordingly, if the set I be multiplied by n, we get, for
n = œ (supposing the Greek letters to be positive constants, or variables which
approach positive finite limits), the first of the two following equations :
The second and third equations in IV are derived from the last set in the
(12«+l)-th and (12n + 2)-th term, respectively. The (12zi + 3)-th,
Trans. Am. Math. Soc. 3
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
34 F. CAJORI : DIVERGENT AND CONDITIONALLY CONVERGENT [January
(12/i + 6)-th, and (12n+9)-th terms yield the same equation in IV as the
(12_)-th term; the (I2n + 4)-th, (12n + 7)-th, (12re + 10)-th terms the same
as the (12«4-l)-th term; the (12« + 5)-th, (12_ + 8)-th, (12w+ H)-th the
same as the (12n + 2)-th term.
The equations II, when simplified and by the substitution of vi2 for vXi
made to apply to other sets in the terms of the product, are
If we aim to assign to the Greek letters constant values, then, in every equa-
tion in V each fraction must admit of being associated singly with another fraction
in that equation which is numerically equal to it, but of opposite sign. This
relation is brought about by the following six equations :
Applying relations VIII, one may write the series _^w_ thus, the parentheses
being suppressed :
Applying relations VII, one may write the series £«m thus, the parentheses be-
ing suppressed :
2>„^i:(Np-9+0.}
p=i + Kp 8 + /32}-{%,-7 + £3}
(X)
+ {« -6+/3;} + {«i2i-5 + /32}-{«i2î,-4 + /33}
+ {«12,-9 + ßi} + {«„-S + /32}- {vX2p-l + ß3}
E(+K»-i%,+i6
1 + {«.2„-i2J,+8
+ «,
+ «;}
■'Up 9 + ^l}-{«12„-12P + 9 + S {«12,-8 + ^2}
«.2p-7+^} +K„-12P+7+ «4] «.2,-6 + /3¡}
+ {«12n-12, + 6 + «3} 12,
■6+ Ä1 + K 12n-12, + 5 + a2 «i2,-4 + /3¡}
[u 12n-12, + 4 + aJ 12, 9 + ßx} + {uX2n_x2p+ 9 + a'2} -8 + /32}
12n-12r-K8 + ai ^12, ' -7 + ^3}-{Wl2„_12;) + 7 + a4}
= -12£{« 12}>
9 A} 12n-12, + 16 + «,) 12n—12, + 4 + «1]
,=1
where n = \n when n is even, and «' = 1(« + 1) when n is odd, and where
a is used in the same way as before. In this case a consists of one constituent
when n is even, and of eleven constituents when « is odd. By the very same
reasoning as that applied in the series Sx and S2 it follows that the (12« + l)-th
term of the product of IX and X, for n > m, has a numerical value less than
l/«.(log n)e, where 0 > 1. The two factor-series IX and X differ from the pairs
8X, S2 and Tx, T2 in this—that (suppressing in each series each parenthesis
which encloses a set) corresponding terms in IX and X need not have the same
numerical values. However, the three pairs of factor-series have in common
one property, to which allusion has not yet been made ; viz., in one of the series
of each pair, the sum of the terms in each set is identically zero, so that the
sum of the series is itself zero.
Colorado College, Colorado Springs.