Divergent Series: A Numeristic Approach

Kevin Carmody
1000 University Manor Drive, Apt. 34
Fairfield, Iowa 52556
http://kevincarmody.com

4 March 2012

ABSTRACT

Infinite divergent series can generate some striking results but have been
controversial for centuries. The standard approaches of limits and methods of
summation have drawbacks which do not account for the full range of behavior
of these series. A simpler approach using numeristics is developed, which bet-
ter accounts for divergent series and their sums. Numeristics is introduced in a
separate paper.

[I]n the early years of this century [the 20th] the subject, while
in no way mystical or unrigorous, was regarded as sensational,
and about the present title [Divergent Series], now colourless, there
hung an aroma of paradox and audacity.—J. E. Littlewood in his
preface to [Ha]

Copyright
c 2001–2012 Kevin Carmody
2 KEVIN CARMODY

INTRODUCTION

In this paper, we will see that we can find finite sums for infinite di-
vergent series. Surprisingly, we find that it is controversial among mathe-
maticians as to whether these series have sums. We will examine this topic
in some detail, and we will develop an approach which enables us to bet-
ter appreciate these series and the surprising nature of the infinite which
they show.
A finite geometric series

n
X
ak = am + am+1 + am+2 + am+3 + . . . + an
k=m

can easily be summed by through a recurrence formula. We let x be the

sum:
n
X
x= ak = am + am+1 + am+2 + am+3 + . . . + an ,
k=m

and then we multiply both sides by a:

xa = am+1 + am+2 + am+3 + am+4 + . . . + an+1 ,

and then we observe that xa + am = x + an+1 . We therefore have xa − x =

an+1 − am , or alternatively x − xa = am − an+1 , which yields

an+1 − am am − an+1
x= = ,
a−1 1−a

the well-known result

n
X am − an+1
ak = am + am+1 + am+2 + am+3 + . . . + an = . (1)
k=m
1−a

It may come as a surprise, although it is well known among math-

ematicians, that an infinite geometric series can also be summed in this
way. If we have

∞
X
ak = am + am+1 + am+2 + am+3 + . . . ,
k=m
Divergent series 3

then again we set

∞
X
x= ak = am + am+1 + am+2 + am+3 + . . . ,
k=m

again multiply both sides by a,

xa = am+1 + am+2 + am+3 + am+4 + . . . ,

and observe that xa + am = x, yielding x − xa = am , or

am
x= ,
1−a

and another well-known result

∞
X am
ak = am + am+1 + am+2 + am+3 + . . . = . (2)
k=m
1−a

If m = 0, that is, if the first term is 1, then this result becomes

∞
X 1
ak = 1 + a + a2 + a3 + . . . = . (3)
k=0
1−a

For example, if a = 21 and m = 1, then the infinite geometric series is

1 1/2
2
+ 14 + 81 + 16
1
+ . . .. Equation (2) says that this sum is 1−1/2 = 1, that is

∞
X 1 1 1 1 1
k
= + + + + . . . = 2. (4)
k=1
2 2 4 8 16

Figure 1 shows a visualization of this sum, and the formula seems to agree
with what we observe in such a diagram.
4 KEVIN CARMODY

F IG . 1: Diagram of 1/2 + 1/4 + 1/8 + 1/16 + . . . = 1

It has long been noticed that, when derived this way, equation (2)
seems to be true regardless of the magnitude of a, except possibly a = 1.
For example, if a = 2, we could conclude that

∞
X
2k = 1 + 2 + 4 + 8 + . . . = −1. (5)
k=0

This may come as an even bigger surprise than Equation (4). Natu-
rally, it might be wondered how Equation (5) could be true. The interme-
diate results, 1, 1 + 2 = 3, 1 + 2 + 4 = 7, 1 + 2 + 4 + 8 = 15, etc., known as
partial sums, grow without limit and are always positive, whereas in Equa-
tion (4), the partial sums 1, 1 + 21 = 32 , 1 + 21 + 14 = 47 , 1 + 12 + 41 + 18 = 15
8
, etc.,
get progressively closer to 2 but never exceed it.
Infinite series such as the one in Equation (4), in which the partial sums
approach a fixed number, are known as convergent, while those that do not,
such as the one in Equation (5), are known as divergent.
Among mathematicians, there are two points of view on convergent
and divergent infinite series. The conventional point of view is that diver-
gent series are meaningless and have no sum, and only convergent series
have a sum. In this view, the number that the partial sums converge to,
called the limit, is considered as the sum of the infinite series.
The alternative point of view is that divergent series are not automat-
ically meaningless but may have a sum. Following this point of view, a
theory of divergent series has been developed. This theory is generally
consistent and even has a number of applications. References [Bo, Fo, Ha,
Mo, Sm, Sz] are some of the important standard works of this theory, with
[Ha] being generally regarded as the most important. In this paper, we
will explore some of the results of this theory. We will use a new approach
which we hope will resolve some longstanding problems. This approach is
Divergent series 5

an application of numeristics, a number based foundational theory, which

includes an extension of arithmetic that allow us to include even the value
a = 1 in equation (2).

SOME RESULTS OF EQUATION (2)

∞
X 1
(−1)k = 1 − 1 + 1 − 1 + . . . = . (6)
k=0
2

P ROOF. Straightforward application of Equation (2) with a = −1.

∞
X 1
(−1)k+1 k = 1 − 2 + 3 − 4 + 5 − . . . = . (7)
k=1
4

P ROOF. Let x = 1 − 2 + 3 − 4 + 5 − . . . = 1 − (2 − 3 + 4 − 5 + . . .) =
1 − (1 − 2 + 3 − 4 + . . .) − (1 − 1 + 1 − 1 + . . .) = 1 − x − 12 . Then 2x = 12 , and
x = 41 .
∞
X 1
k = 1 + 2 + 3 + 4 + 5 + ... = − . (8)
k=1
12

P ROOF. Let x = 1 + 2 + 3 + 4 + 5 + . . .. Then −3x = x − 4x = (1 + 2 + 3 + 4 +

5+. . .)−4(1+2+3+4+5+. . .) = (1+2+3+4+5+. . .)−2(2+4+6+8+10+. . .) =
1 − 2 + 3 − 4 + 5 − . . . = 14 , and x = − 12
1
.

0.999 . . . = 1. (9)

P ROOF. This is a convergent series which we mention for complete-

ness. 0.999 . . . = 9(0.1 + 0.01 + 0.001 + 0.0001 . . .) = 9(10−1 + 10−2 + 10−3 +
10−4 +. . .) = 9(0.11 +0.12 +0.13 +0.14 +. . .) = 9 ∞ k 0.1 0.1
P
k=1 .01 == 9( 1−0.1 ) = 9( 0.9 ) =
9( 91 ) = 1.

. . . 999 = −1. (10)

P ROOF. By the notation . . . 999 we mean an infinite series of digits

going out to the left, just as the notation 0.999 . . . means an infinite series
of digits going out to the right. Then . . . 999 = 9(1 + 10 + 100 + 1000 + . . .) =
9(100 + 101 + 102 + 103 + . . .) = 9 ∞ k 1 1
P
k=0 10 = 9( 1−10 ) = 9(− 9 ) = −1.
6 KEVIN CARMODY

∞
X
ak = . . . + am−3 + am−2 + am−1 + am + am+1 + am+2 + am+3 + . . . = 0,
k=−∞

a 6= 1. (11)

P ROOF. We have . . . + am−3 + am−2 + am−1 + am + am+1 + am+2 + am+3 + . . . =

(am + am+1 + am+2 + am+3 + . . .) + (. . . + am−1 + am−2 + am−3 + am−4 + . . .) = (am +
am+1 +am+2 +am+3 +. . .)+(. . .+(a−1 )1−m +(a−1 )2−m +(a−1 )3−m +(a−1 )4−m +. . .) =
(am +am+1 +am+2 +am+3 +. . .)+(. . .+(a−1 )1−m +(a−1 )2−m +(a−1 )3−m +(a−1 )4−m +
am am−1 am a am−1 am am am am
. . .) = 1−a
+ 1−a−1
= 1−a
+ a(1−a−1 )
= 1−a
+ a−1
= 1−a
− 1−a
= 0.
Alternatively, if x = . . . + am−3 + am−2 + am−1 + am + am+1 + am+2 + am+3 + . . .,
then xa = . . . + am−2 + am−1 + am + am m + 1 + am+2 + am+3 + am+4 + . . . = x, so
0 = x − xa = x(1 − a). Then x must be 0, unless a = 1.

. . . 999.999 . . . = 0. (12)

P ROOF. We have . . . 999.999 . . . = 9(. . . + 1000 + 100 + 10 + 1 + 0.1 +

0.01 + 0.001 + . . .) = 9(. . . + 103 + 102 + 101 + 100 + 10−1 + 10−2 + 10−3 + . . .) =
9 ∞ k
P
k=−∞ 10 = 9(0) = 0.

∞
X 1 + i cot x2
ekx = 1 + eix + e2ix + . . . = . (13)
k=0
2
 
1 1 i 2i
P ROOF. enix = (eix )n , so 1 + eix + e2ix + . . . = 1−eix
= 2
+ 2 eix −1
−1 =
1
2
+ 2i cot x2 .
∞
X sec x
(−1)k e2k+1 = eix − e3ix + e5ix − . . . = . (14)
k=0
2
eix 1
P ROOF. eix − e3ix + e5ix − . . . = 1+e2ix
= 2
sec x.
∞
X 1
cos kx = cos x + cos 2x + cos 3x + . . . = − . (15)
k=1
2

P ROOF. From Equation (13), cos x + cos 2x + . . . =

< 1 + eix + e2ix + . . . − 1 = 12 − 1 = − 21 .

∞
X cot x2
sin kx = sin x + sin 2x + sin 3x + . . . = . (16)
k=1
2
Divergent series 7

P ROOF. From Equation (13), sin x + sin 2x + . . . =

= 1 + eix + e2ix + . . . = 12 cot x2 .

∞
X 1
(−1)k+1 cos kx = cos x − cos 2x + cos 3x − . . . = − . (17)
k=1
2

P ROOF. Starting with Equation (15), we replace x with x + π. Then

cos nx remains unchanged when n is even, because we are adding 2mπ
to x, where n = 2m and m is an integer. But when when n is odd, then
n = 2m+1, and we are adding 2mπ+π to x, and so cos nx becomes − cos nx.
This yields − cos x +cos 2x −cos 3x +. . . = − 21 , or cos x −cos 2x +cos 3x −. . . =
− 12 .

∞
X tan x2
(−1)k+1 sin kx = sin x − sin 2x + sin 3x − . . . = . (18)
k=1
2

P ROOF. We start with Equation (16) and use the same substitution,
replacing x with x + π. Then sin nx remains unchanged for n even and
becomes − sin nx for n odd. In addition, cot x2 becomes cot x2 + π2 + =

tan x2 .

∞
X 1 1 1 1
= 1 + + + + . . . = ln 2 (19)
k=1
k 2 3 4
∞
X 1 1 1 1
(−1)k+1 = 1 − + − + . . . = − ln 2 (20)
k=1
k 2 3 4

P ROOF. Starting with Equation (16), integrating each term from 0 to x

gives (cos x − 1) + 21 (cos 2x − 1) + 13 (cos 3x − 1) + . . . = ln sin x2 = (cos x +
1
2
cos 2x + 13 cos 3x + . . .) − (1 + 12 + 31 + . . .).
For x = π, this gives (−1 + 12 − 13 + . . .) − (1 + 12 + 31 + . . .) = 0. Letting
y = 1 + 12 + 13 + . . ., then also y = −1 + 21 − 31 + . . ..
For x = π/2, the integration gives (− 21 + 41 − 16 + . . .) − (1 + 12 + 31 + . . .) =
1
2
(−1 + 12 − 31 + . . .) − (1 + 12 + 13 + . . .) = − 21 ln 2 = 12 y − y, or y = ln 2.

∞
X xk x2 x3 x4
=x+ + + + . . . = ln(1 − x) (21)
k=1
k 2 3 4
∞
X xk x2 x3 x4
(−1)k+1 =x− + − + . . . = ln(1 + x) (22)
k=1
k 2 3 4
8 KEVIN CARMODY

1
P ROOF. From Equation (2), we have 1 + x + x2 + x3 + . . . = 1−x and
2 3 1
1 − x + x − x + . . . = 1+x . Integrating these from 0 to x gives the results,
which are known as Mercator’s Series.

∞
X 2k 2 4 8 16
= + + + + . . . = (2k + 1)πi, k an integer. (23)
k=1
k 1 2 3 4

P ROOF. Substitute x = 2 into Equation (21) and use the fact that
ln(−1) = (2k + 1)πi.

∞
X sec x
(−1)k cos(2k + 1)x = cos x − cos 3x + cos 5x − . . . = . (24)
k=0
2

P ROOF. From Equation (14), cos x − cos 3x + cos 5x − . . . = <(eix − e3ix +

e5ix − . . .) = 21 sec x.

∞
X
(−1)k sin(2k + 1)x = sin x − sin 3x + sin 5x + . . . = 0. (25)
k=0

P ROOF. From Equation (14), sin x − sin 3x + sin 5x + . . . = =(eix − e3ix +

e5ix − . . .) = 0.

∞
X
(−1)k+1 k 2n cos kx = 12n cos x − 22n cos 2x + 32n cos 3x − . . .
k=1

= 0, n = 0, 1, 2, . . . (26)
∞
X
(−1)k+1 k 2n+1 sin kx = 12n+1 sin x − 22n+1 sin 2x + 32n+1 sin 3x − . . .
k=1

= 0, n = 0, 1, 2, . . . (27)

P ROOF. Repeated differentiation, n times, of Equation (17).

Divergent series 9

∞
X
(−1)k+1 k 2n sin kx = 12n sin x − 22n sin 2x + 32n sin 3x − . . .
k=1

tan x2
2n
d

n
= (−1) , n = 0, 1, 2, . . . (28)
dx 2
∞
X
(−1)k+1 k 2n+1 cos kx = 12n+1 cos x − 22n+1 cos 2x + 32n+1 cos 3x − . . .
k=1

tan x2
2n+1
d

n
= (−1) , n = 0, 1, 2, . . . (29)
dx 2

P ROOF. Repeated differentiation, n times, of Equation (18).

∞
X
(−1)k (2k + 1)2n cos(2k + 1)x =
k=0

d 2n sec x
 
2n 2n 2n n
1 cos x − 3 cos 3x + 5 cos 5x − . . . = (−1) ,
dx 2
n = 0, 1, 2, . . . (30)
∞
X
(−1)k (2k + 1)2n+1 sin(2k + 1)x =
k=0

d 2n+1 sec x
 
2n+1 2n+1 2n+1 n
1 sin x − 3 sin 3x + 5 sin 5x − . . . = (−1) ,
dx 2
n = 0, 1, 2, . . . (31)

P ROOF. Repeated differentiation, n times, of Equation (24).

∞
X
(−1)k (2k + 1)2n sin(2k + 1)x =
k=0
2n
1 sin x − 32n sin 3x + 52n sin 5x − . . . = 0, n = 0, 1, 2, . . . (31)
∞
X
(−1)k (2k + 1)2n+1 cos(2k + 1)x =
k=0
2n+1
1 cos x − 32n+1 cos 3x + 52n+1 cos 5x − . . . = 0, n = 0, 1, 2, . . . (33)

P ROOF. Repeated differentiation, n times, of Equation (25).

10 KEVIN CARMODY

∞
X
(−1)k+1 k 2n = 12n − 22n + 32n − . . . = 0, n = 1, 2, 3, . . . . (34)
k=1

P ROOF. Substitute x = 0 into Equation (26).

∞
X
(−1)k (2k+1)2n+1 = 12n+1 −32n+1 +52n+1 −. . . = 0, n = 0, 1, 2, . . . . (35)
k=0

π
P ROOF. Substitute x = 2
into Equation (27).

∞
X
(2k + 1)2n+1 = 12n+1 + 32n+1 + 52n+1 + . . . = 0, n = 0, 1, 2, . . . . (36)
k=0

π
P ROOF. Substitute x = 2
into Equation (32).

∞
X
(−1)k (2k + 1)2n = 12n − 32n + 52n − . . . = 0, n = 0, 1, 2, . . . . (37)
k=0

P ROOF. Substitute x = 0 into Equation (33).

∞
X
(−1)k+1 k 2n+1 = 12n+1 − 22n+1 + 32n+1 + . . .
k=1

22n+2 − 1
= B2n+2 , n = 0, 1, 2, . . . . (38)
2n + 2
P ROOF. Bk stands for the k-th Bernoulli nxumber, which occurs in the
2k 2k
k−1 2 (2 −1)
power series tan x = ∞ B2k x2k−1 .
P
k=1 (−1) (2k)!

To prove the above identity, we evaluate 12n+1 cos x − 22n+1 cos 2x +

 2n+1
d
32n+1 cos 3x − . . . = (−1)n dx 1
2
tan x2 for x = 0. We begin by com-
2k 2k
k−1 2 (2 −1) 2k−1
puting the power series 12 tan x2 = 12 ∞ B2k x22k−1 =
P
k=1 (−1) (2k)!
k−1 22k −1
P∞
k=1 (−1) (2k)!
B2k x2k−1 , and we will continue by differentiating this
series at x = 0.
 n
d
To differentiate this series, we observe that dx
xk = k(k − 1)(k −
2) · · · (k −n+1) xk−n . When x = 0, this product is nonzero only when xk−n is
Divergent series 11
 n
d
constant, i.e. when k = n, at which time dx
xk = k(k −1)(k −2) · · · 1 = k!.
This means that all terms of the power series vanish, except for k = n.
 2n+1
d
Therefore, dx x2k−1 = (2n + 1)! when 2n + 1 = 2k − 1 or k =
 2n+1
d
n + 1, and is zero when k 6= n + 1. We then have (−1)n dx 1
2
tan x2 =
(22n+2 −1) (2n+1)! 22n+2 −1
(−1)n (−1)n+2 (2n+2)!
B2n+2 = 2n+2
B2n+2 .
Hardy and other older authors state this result in a different form,
because they used an older system of indexing the Bernoulli numbers. If
we let Bk∗ represent the old system and Bk the new, then the relation is
B2k = (−1)k−1 Bk∗ . We then obtain the older statement of the result, 12n+1 −
22n+2 −1 ∗
22n+1 − 32n+1 + . . . = (−1)n 2n+2
Bn+1 .

∞
X 1
(−1)k (2k + 1)2n = 12n − 32n + 52n + . . . = E2n , n = 0, 1, 2, . . . . (39)
k=0
2

P ROOF. Ek stands for the k-th Euler number, which occurs in the
power series sec x = ∞ k 1
E2k x2k . To prove our identity, we
P
n=0 (−1) (2k)!
 2n
2n 2n 2n n d 1
evaluate 1 cos x − 3 cos 2x + 5 cos 5x − . . . = (−1) dx 2
sec x for
x = 0.
As before, to differentiate the power series, we use the fact that for
 n
d
x = 0, dx xk = k! when k = n, and is zero when k 6= n. We then have
 2n
d 1
(−1)n dx 2
sec x = (−1)n (−1)n 2(2n)!
(2n)!
E2n = 12 E2n .
As with the Bernoulli numbers, there is also an older system of in-
dexing the Euler numbers, which leads to a different form of the result by
older authors. Letting Ek∗ represent the old system and Ek the new, the
relation is given by E2k = (−1)k Ek∗ . The older statement of the result is then
12n − 32n + 52n + . . . = (−1)n 12 En∗ .
∞
X 1 1 1 π2
= 1 + + + . . . = . (40)
k=0
(2k + 1)2 32 52 8

P ROOF. Integrate the negative of Equation (17) from 0 to x to obtain

sin x − 21 sin 2x + 31 sin 3x − . . . = 12 x. Integrate again from 0 to x to obtain
(1 − cos x) − 212 (1 − cos 2x) + 312 (1 − cos 3x) − . . . = 14 x2 . Evaluate this at x = π.
12 KEVIN CARMODY

∞
X 1 1 1 π2
= 1 + + + . . . = . (41)
k=1
k2 22 32 6
π2
P ROOF. Let y = 1 + 212 + 312 + . . .. Then 8
= 1 + 212 + 312 + . . . = 1 + 212 + 312 +
1 1 1 1 π2
... − 22
− 42
− ... = 1 + 22
+ 32
+ . . . − 14 (1 + 212 + 1
32
+ . . .) = y − 14 y = 34 y = 6
.
∞
X 1 1 1 π2
(−1)k+1
= 1 − + − . . . = . (42)
k=1
k2 22 32 12
   
P ROOF. 1 − 212 + 312 − . . . = 1 + 312 + 512 + . . . − 212 + 1
42
+ 1
82
+ ... =
    2 2 2
1 + 312 + 512 + . . . − 212 1 + 212 + 312 + . . . = π8 − 41 π6 = π12 .

RESULTS INVOLVING INFINITE PRODUCTS

We now consider various infinite products involving prime numbers.

Before this, we will first need to know how to multiply two infinite series,
and how to multiply an infinite number of binomials.

  !
Xm n
X
 aj  bk = (a1 + a2 + a3 + . . . + am )(b1 + b2 + b3 + . . . + bn )
j=1 k=1
m X
X n
= aj bk = a1 b1 + a1 b2 + a1 b3 + . . . + a1 bn
j=1 k=1

+ a2 b1 + a2 b2 + a2 b3 + . . . + a2 bn
+ a3 b1 + a3 b2 + a3 b3 + . . . + a3 bn
..
.
+ am bn + am bn + am b3 + . . . + am bn (43)

P ROOF. This is the product of two series. Both series are finite, and we
simply multiply out each series.

! !
∞
X ∞
X
am bn = (a1 + a2 + a3 + . . .)(b1 + b2 + b3 + . . .)
m=1 n=1
∞ X
X ∞
= am bn = a1 b1 + a1 b2 + a1 b3 + . . .
m=1 n=1
Divergent series 13

+ a2 b1 + a2 b2 + a2 b3 + . . .
+ a3 b1 + a3 b2 + a3 b3 + . . .
..
.
X∞
= am bn . (44)
m,n=1

P ROOF. This is similar to the previous equation, but with each series
now infinite.

n
Y
(1 + ak ) = (1 + a1 )(1 + a2 )(1 + a3 ) . . . (1 + an )
k=1
n
X
=1+ bk = 1 + b1 + b2 + b3 + . . . + bn ,
k=1
where
n
X
b1 = at ,
t=1
n
X
b2 = at au ,
t,u=1
t6=u
n
X
b3 = at au av ,
t,u,v=1
t,u,v distinct

..
.
n
X
bk = at1 at2 at3 . . . atk
t1 ,...,tk =1
t1 ,...,tk distinct
n
X k
Y
= atj . (45)
t1 ,...,tk =1 j=1
t1 ,...,tk distinct

P ROOF. This is the product of a finite number of binomials, each with

a first term of 1. The number of binomials is finite, and we multiply out
their product to obtain 1 plus a sum of b terms.
For the b terms, b1 is the sum of all a, b2 is the sum of the products
of any two distinct a, b3 is the sum of the products of any three distinct a,
and so on. Each bk term consists of k factors from the set of the a terms,
with the index of each a term being distinct.
14 KEVIN CARMODY

∞
Y
(1 + ak ) = (1 + a1 )(1 + a2 )(1 + a3 ) . . .
k=1
∞
X
=1+ bk = 1 + b1 + b2 + b3 + . . . ,
k=1
where
∞
X
b1 = at ,
t=1
∞
X
b2 = at au ,
t,u=1
t6=u
n
X
b3 = at au av ,
t,u,v=1
t,u,v distinct

..
.
∞
X
bk = at1 at2 at3 . . . atk
t1 ,...,tk =1
t1 ,...,tk distinct
∞
X k
Y
= atj . (46)
t1 ,...,tk =1 j=1
t1 ,...,tk distinct

P ROOF. Similar to the previous equation, but with the number of bi-
nomials now infinite.

∞
Y p 2 3 5 7 11
= · · · · ...
p=2
p − 1 1 2 4 6 10
p prime

1 1
= ∞ =
1 − 2 1 − 3 1 − 15 1 − 71 1 −
1 1 1






...

Y 1 11
1−
p=2
p
p prime
∞
X 1 1 1 1
= = 1 + + + + .... (47)
k=1
k 2 3 4
Divergent series 15

P ROOF. The products in the first and second lines are taken over all
prime numbers. The first line becomes the second line through two iden-
p
tities: 1−1 1 = p−1 , and a1 · b1 = ab
1 Q1
or its extended form a
= Q1 a .
p

To see how the second line becomes the third line, we take the first
two primes, 2 and 3. From Equation (2), 1−1 1 = 1 + 12 + 41 + 81 + . . ., and
2
1 1 1 1
1− 13
=1+ 3
+ 9
+ 27
+ . . ..
By Equation (43), when we multiply these two infinite series, we get
1
= 1 + 12 + 13 + 14 + 16 + 18 + 12 + . . ., where the denominators in the
(1− 12 )(1− 13 )
sum all have as their prime factors powers of 2 and 3 only.
As we continue to multiply each side by 1−1 1 = 1 + p1 + p12 + p13 + . . . for
p

successive prime numbers p, the denominators in the sum have as their

prime factors powers of the primes up to p. When all the prime numbers
have been included in the product, all the integers are included in the de-
nominators in the sum.

∞
Y pn 2n 3n 5n 7n 11n
= · · · · ...
p=2
pn − 1 1n 2n 4n 6n 10n
p prime

1 1
= ∞ = 1

1

1

1

1


1− 1− 1− 1− 1− ...

Y 1 2n 3n 5n 7n 11n
1− n
p=2
p
p prime
∞
X 1 1 1 1
= n
= 1 + n + n + n + . . . = ζ(n),
k=1
k 2 3 4
n = . . . , −2, −1, 0, 1, 2, . . . . (48)

P ROOF. The proof is similar to the previous equation, except that now
we use a constant power of the primes instead of the primes themselves.
The last equality is the definition of the Riemann zeta function.

∞
Y
(1 + p) = (1 + 2)(1 + 3)(1 + 5)(1 + 7)(1 + 11) . . .
p=2
p prime
∞
X
= k = 1 + 2 + 3 + 5 + 6 + 7 + 10 + . . . ,
k=1
k squarefree

n = . . . , −2, −1, 0, 1, 2, . . . . (49)

16 KEVIN CARMODY

P ROOF. The product in the first line is taken over all prime numbers.
The sum on the second line is taken over all squarefree integers, which are
those which contain only single powers of prime factors, and so are not
divisible by the square of any integer greater than 1.
By Equation (46), the product on the first line when multiplied out
becomes 1, plus the sum of all p, plus the sum of the products of any two
distinct p, plus the sum of the products of any three distinct p, and so on.
Since each occurrence of p can be used at most once in a product, each
term in the sum is squarefree.

∞        
Y 1 1 1 1 1 1
1+ = 1+ 1+ 1+ 1+ 1+ ...
p=2
p 2 3 5 7 11
p prime
∞
X 1 1 1 1 1 1 1
= =1+ + + + + + + ...,
k=1
k 2 3 5 6 7 10
k squarefree

n = . . . , −2, −1, 0, 1, 2, . . . . (50)

P ROOF. Similar to the above equation, except that p is replaced by p1 .

∞        
Y 1 1 1 1 1 1
1+ n = 1+ n 1+ n 1+ n 1+ n 1 + n ...
p=2
p 2 3 5 7 11
p prime
∞
X 1 1 1 1 1 1 1
= = 1 + + + + + + + ...,
k=1
kn 2n 3n 5n 6n 7n 10n
k squarefree

n = . . . , −2, −1, 0, 1, 2, . . . . (51)

1
P ROOF. Similar to the previous equation, except that p
is replaced by
1
pn
.

∞
Y
(1 − p) = (1 − 2)(1 − 3)(1 − 5)(1 − 7)(1 − 11) . . .
p=2
p prime
∞
X
= ±k = 1 − 2 − 3 − 5 + 6 − 7 + 10 + . . . ,
k=1
k squarefree
Divergent series 17
n+o n even o
where ± = when k has an number of primes,
− odd
n = . . . , −2, −1, 0, 1, 2, . . . . (52)

P ROOF. Similar to Equation (49), except that p is replaced by −p.

∞        
Y 1 1 1 1 1 1
1− = 1− 1− 1− 1− 1− ...
p=2
p 2 3 5 7 11
p prime
∞
X 1 1 1 1 1 1 1
= ± =1− − − + − + + ...,
k=1
k 2 3 5 6 7 10
k squarefree
n+o n even o
where ± = when k has an number of primes,
− odd
n = . . . , −2, −1, 0, 1, 2, . . . . (53)

P ROOF. Similar to the above equation, except that p is replaced by p1 .

∞      
Y 1 1 1 1
1− n = 1− n 1− n 1 − n ...
p=2
p 2 3 5
p prime
∞
X 1 1 1 1 1 1 1 1
= ± = 1 − − − + + + + . . . = ,
k=1
kn 2n 3n 5n 6n 7n 10n ζ(n)
k squarefree
n+o n even o
where ± = when k has an number of primes,
− odd
n = . . . , −2, −1, 0, 1, 2, . . . . (54)

1
P ROOF. Similar to the previous equation, except that p
is replaced by
1
pn
.
The product is the reciprocal of the product in Equation (48), which
equals ζ(n).
18 KEVIN CARMODY

RESULTS INVOLVING DIVERGENT INTEGRALS

We use the above results to evaluate divergent or improper integrals.

Z∞
−1
ax dx = . (55)
0 ln a
Rn Pn R k x
P ROOF. For an integer n, we have 0 ax dx = k=1 k−1 a dx =
1 Pn k k−1

1−1/a P n k
R∞ x
ln a k=1 a − a = ln a k=1 a . For the infinite case, 0
a dx =
1−1/a P∞ k 1−1/a a −1
ln a k=1 a = ln a 1−a = ln a .
Z∞
ax dx = 0. (56)
−∞
R∞ R∞ R∞ R∞
P ROOF. By equation (55), −∞ ax dx = 0 ax dx + 0 a−x dx = 0 ax dx +
1 x
R∞
dx = ln−1a + −−1


0 a ln a
= 0.
Z∞
ex dx = −1. (57)
0

P ROOF. Application of equation (55) for a = e.

Z∞
e−x dx = 1. (58)
0

P ROOF. Application of equation (56) for a = e and equation (57). This

is a convergent integral and the result agrees with the usual approach.

a∞ = 0. (59)
R∞
P ROOF. By integration and equation (55), 0 ax dx = ln1a (a∞ − 1) = ln1a ,
so a∞ = 0. Alternatively, for n = ∞ in equation (1) and by equation (2),
P∞ k am −a∞ am ∞
m a = 1−a = 1−a , again yielding a = 0.
N OTE . The values a = 0 and a = 1 in this and all the equations above
must be carefully examined. This is done below in the section on idempo-
tents.
Divergent series 19

A PRACTICAL APPLICATION

It can be argued that the formula 1 + 2 + 4 + 8 + . . . = −1 has an ap-

plication in the way that computers store negative numbers. Almost uni-
versally, computers store a signed integer using a convention known as
twos-complement arithmetic, which uses the highest-order bit as a sign bit
(0 = non-negative, 1 = negative), and the negative of a positive number
is generated by switching all the bits (ones-complement) and then adding 1
(twos-complement).
Example: In a 8-bit signed integer, the decimal number 20 has the bi-
nary representation 00010100. To get −20, we transform the 0’s into 1’s and
the 1’s into 0’s to get a ones-complement representation of 11101011, and
then we add 1 to get a twos-complement representation of 11101100.
The great advantage of the twos-complement system is that the same
rules can be used for arithmetic operations on both positive and negative
numbers. This is turn helps to make processors faster. Another advantage
is that the highest-order bit can be interpreted either as a sign bit or as the
negative of the highest power of 2, and the arithmetic is the same.
In the same system, 8-bit signed integers, the twos-complement rep-
resentation of −1 is 11111111. No matter how many bits there are, the
twos-complement representation of −1 is always all 1’s. Since the value of
each bit is a power of 2, this is the sum of all the powers of 2 that are avail-
able in the computer representation of the number, except that the sign bit
is the negative of the highest power. If we increase the number of bits, the
sign bit moves to higher powers of 2.
In the theoretical case of an infinite number of bits, the sign bit dis-
appears, and we are left with an infinite number of ones, which rep-
resent the number −1. This is identical to the divergent series formula
1 + 2 + 4 + 8 + . . . = −1.
20 KEVIN CARMODY

THREE APPROACHES

We will consider the following three approaches to divergent series.

1. Limits. In this approach, any infinite series is considered to be the
limit of its partial sums:

∞
X n
X
ak = lim ak .
n→∞
k=1 k=1
If the limit exists, the series is said to be convergent, and otherwise di-
vergent. A divergent series has no limit and therefore, strictly speak-
ing, no sum. More informally we may say that the sum is ∞, but this
symbol is usually not defined as a number.
2. Methods. In this approach, a convergent series is still considered to
be a limit, but a divergent series may be said to have a finite sum
in a restricted sense. It has this sum if some calculational procedure
based on the terms of the series yields a finite sum. The procedure is
called a method of summation. Ideally, a method applied to a convergent
series should yield the same sum as the limit, and when applied to at
least some divergen series may yield a finite sum. Since the sum of a
divergent series may vary from one method to another, the equality of
the series with its sum is said to exist only in the sense of the method.
3. Numeristic. In this approach, any infiite series is considered sim-
ply as a purely infinite arithmetic sum, without regard to a limit. It
regards infinite sums as actual and uses an analytic approach called
equinfinitesimal to perform infinite and infinitesimal arithmetic.
The limits approach was developed in the early nineteenth century
and became nearly universal as direct calculation with the infinite and the
infinitesimal was replaced with limits and set theoretical notions. Today,
most professional mathematicians are unaware of any alternatives.
The methods approach developed in parallel to the limit approach but
remained much less popular. It is today known as the theory of divergent
series. By far the most comprehensive treatment of this approach is [Mo],
now considered the standard reference.
The numeristic approach is the one we develop in this paper. As we
will see later, this is not really a new approach to divergent series, but only
a better understanding of an old approach. P
P [Ha, p. 6–7] defines the notation am = s (P) to mean that the
Hardy
series am has a sum s in the sense of a method P . This symbolism means
that equality holds only in a certain sense, because changing the sense or
the method of determining the sum can change the value s that we define
as the sum. This weakens the meaning of equality, since it is relative to a
Divergent series 21

method of computation, and not, as is normally the case, independent of

it.
The methods approach has several weaknesses:
1. An unworkable conception of weak equality.
2. The failure of all methods, except the Euler continuation method, to
account for Equation (2) or any other significant finding.
3. A faulty understanding of the Euler continuation method, including
a circular definition in [Ha], and failure to realize this method as an
extension of arithmetic.
We further examine these points in later sections and show how the
numeristic approach avoids all of them.

OTHER APPROACHES

A few other approaches to divergent series deserve consideration.

Transformation to convergent series. In this approach, a divergent
series is considered as en encoded form of a convergent series, using the
transformation

1
 
1 1
= − .
1−a a 1 − a1
This approach has the disadvantage of conceptualizing simple arith-
metic statements as something different from what they state, and there-
fore of introducing extra steps of calculation. Another disadvantage is
that an expression which combines convergent and divergent series, such
n
as +∞
P
−∞ a , must be considered as the sum of two separate series.
Pj
Congruence. In this approach, we observe that the partial sums 0 2n
are each congruent to −1 mod 2j+1 . The drawback of this approach is that
it requires us to define an equality through a congruence. It also is not
Pj
immediately clear how to extend it to any other case than 0 2n , since for
integral bases greater than two, the sum is a fraction.
Adjusted series. Another approach is to return to the twos-comple-
ment arithmetic described above, and generalize its sign bit mechanism.
This seems at first to be a combination of the limit and numeristic ap-
proach, but it actually ends up being essentially just the numeristic ap-
proach.
If we define  k
a , k<n
f(k) = k
a
 , k=n
1−a
22 KEVIN CARMODY

then !
n n−1
X X an 1 − an an 1
f(k) = ak + = + = .
k=0 k=0
1−a 1−a 1−a 1−a
This value does not depend on n—for all values of n, the sum is constant.
In the infinite case, the last term does not exist. Hence we can say that

∞ ∞
X X 1
f(k) = ak = .
k=0 k=0
1−a
This satisfies the limits approach, because the infinite case is the limit
of the finite case. But a limits approach alone does not suggest how to
modify the definition of ∞ k
P
k=0 a so that the result is constant. To get the
crucial last term, the numeristic approach is both necessary and sufficient.

POWER SERIES WITH IDEMPOTENTS

The series
∞
X
x= 0k = 0 + 0 + 0 + . . .
k=0
is of fundamental importance, since it is implicit in any infinite series, con-
vergent or divergent, and results whenever we subtract any infinite series
from itself. From an numeristic point of view, x = 0 · ∞, and so the sum is
indeterminate, but according to equation (2), x = 0, because 0x = x.
We resolve this by taking the approach of Musès that 0 is multivalued.
See [Mu65, p. 178–179] and especially [Mu72]. In this view, an expression
such as x = 0 · ∞ is indeterminate because 0 is multivalued, and the ex-
pression does not specify which values of 0 are involved. We may have
two distinct values of 0, which we can call 01 , 02 , etc., and which may have
relations such as by 01 = 4 · 02 .
Ordinary equality does not distinguish these multiple values. The
sense of equality when we write this type of relation must have the sensi-
tivity to distinguish 01 from 02 , so that 01 6= 02 at this level of sensitivity,
whereas at the ordinary level of sensitivity, 01 = 02 , and a + 01 = a + 02 = a
for any real number a.
To denote these levels of sensitivity, we use the notation a = b sen u,
where u is the unit whose multiples we distinguish. In the above example,
we write 01 = 02 sen 1, 01 6= 02 sen 01 , 01 = 4 · 02 sen 01 , and a + 01 = a sen 1.
The default unit is 1, so that a = b means the same as a = b sen 1, unless
otherwise stated.
We now examine a power series in 01 at the 01 level of sensitivity. For
x = 0k1 = 01 + 021 + 031 + . . ., we have x = 01 + 01 x, and 01 = x − 01 x, and
P
Divergent series 23

01 01
x = 1−0 1
, and 1−0 1
= 0 sen 1. Power series in 01 thus come out to 0, because
n
01 is smaller than 01 for n > 1. The sum 01 + 01 + 01 + . . . is not a power
series. The value of 0 · ∞ depends on a more precise knowledge of which
0 and which ∞ is being used.
A similar approach can be used on the series

∞
X
x= 1k = 1 + 1 + 1 + . . . .
k=0

Equation (2) gives the value x = 10 , which is derived from x = x + 1 and

x − x = x · 0 = 1. If we take the view that 0 is multivalued, then 10 is also
multivalued.

OBJECTIONS

Complete rejection of sums of divergent series results in the limits ap-

proach. From the methods approach, the following objections have histor-
ically been raised, without complete resolution.
Hardy [Ha, p. 16] points out several examples of divergent series
which appear to give multiple values. For example,

x + (2x2 − x) + (3x3 − 2x2 ) + (4x4 − 3x3 ) + . . . = 0

and
x + (3x2 − x) + (7x4 − 3x2 ) + (15x8 − 7x4 ) + . . . = 0
give 1 + 1 + 1 + . . . = 0 and 1 + 2 + 4 + . . . = 0. But he does not recognize that

x + (2x2 − x) + (3x3 − 2x2 ) + (4x4 − 3x3 ) + . . .

= (x − x) + (2x2 − 2x2 ) + (3x3 − 3x3 ) + (4x4 − 4x4 ) + . . .
= 0 + 0 + 0 + . . . = 0 × ∞,
which of course is indeterminate, not merely zero.
Using Equation (2), we have shown that we can calculate that 1 − 1 +
1 − 1 + . . . = 21 . It is frequently objected that this cannot be the only correct
result, since we could also have 1 − 1 + 1 − 1 + . . . = (1 − 1) + (1 − 1) + . . . =
0 + 0 + . . . = 0. This objection comes from one or both of two points of view:
(1) an infinite sum is a limit; and/or (2) the sum of an infinite number of
zeros is zero. The first point of view overlooks the important fact that the
type of sum we are considering here is not a limit. The second point of
view overlooks the fact that an infinite sum of zeros is ∞ × 0, which is
indeterminate.
24 KEVIN CARMODY

Euler was the first to systematically use Equation (2) for divergent se-
ries and was the first to state this case of it. Some time afterwards, Callet
challenged this view, citing the series

1 − a2 + a3 − a5 + a6 − a7 + a8 − . . . .
Setting x to this series, we have

x = 1 − a2 + a3 − a5 + a6 − a8 + a9 − . . .
= 1 − a2 + a3 (1 − a2 + a3 − a5 + a6 − . . .)
= 1 − a2 + a3 x.
Then x − a3 x = 1 − a2 , and

1 − a2 1+a
x= 3
= .
1−a 1 + a + a2
When a = 1, we would then have

2
1 − 1 + 1 − 1 + ... = .
3
Lagrange replied that actually

1 − a2
3
= 1 + 0a − a2 + a3 + 0a4 − a5 + a6 + 0a7 − a8 + a9 + . . . ,
1−a
which for a = 1 becomes

2
1 + 0 − 1 + 1 + 0 − 1 + 1 + 0 − 1 + 1 + ... = .
3
This differs from 1 − 1 + 1 − 1 + . . . by the addition of an infinite number
of zeroes, which, as we have seen above, is not zero but indeterminate.
1+a 2 1−a2
Besides this, for a = 1, 1+a+a 2 = 3 , but 1−a3 is indeterminate.

We have shown above that 1 + 2 + 4 + 8 + . . . = −1. Hardy [Ha, p. 15]

challenges this as an exclusive result. Starting from the identity

1 1 2
= + 2 ,
x−1 x+1 x −1
repeated application yields

2 2 4 8 16
= + + + ....
x2 − 1 x2 + 1 x4 + 1 x8 + 1 x16 + 1
Hardy states that when x = 1, this becomes 1 + 2 + 4 + 8 + . . . = ∞, but
he does not seem to notice that the original identity is not valid for x = 1.
This is true even in equinfinitesimal arithmetic. It depends on the identity
Divergent series 25

−1 2 2
−1
x + 1 = xx−1 , but in equinfinitesimal arithmetic 11−1 x = 1 may be interde-
terminate or assume any single value, because 1 − 1 is indeterminate at the
infinitesimal level.
Hardy [Ha, p. 16] also states that 1 + 1 + 1 + . . . = − 12 , because the
well-known Riemann zeta function, which he states as
∞
X
ζ(x) = kx ,
k=1

has the value ζ(0) = − 12 . The problem with this value is that it is calculated
from the definition Z ∞ x−1
1 u
ζ(x) = du,
Γ(x) 0 eu − 1
which is equivalent to the first expression for x > 1, but not elsewhere.

THE THEORY OF DIVERGENT SERIES

For several centuries there been a theory of divergent series, which

attempts to show how and why we can find a sum for many types of di-
vergent series. It is generally agreed that the theory of divergent series
started with Euler, whose findings on the subject are probably best repre-
sented in [Eu]. This theory has attracted other very well known names,
including Poisson, Abel, and Hardy. Hardy’s posthumous book [Ha] of
1949 is generally considered a standard work.
Euler and other mathmaticians of the 18th century generally either
took a basically numeristic approach to divergent series, or rejected such
series, or vacillated. In the 19th century, the limits approach developed
and gradually came to dominate the approach to infinite arithmetic. In
the 20th century, the methods approach developed and attracted some at-
tention, but it remained poorly known and never came to dominate over
the limits approach.
In recent years, the theory seems to be once again attracting some
attention. One example of this transition is the Encyclopedic Dictionary
of Mathematics, a standard general mathematics reference work. In the
first edition [En77], published in 1977, the article Summability describes
the methods approach to divergent series, while the article Infinite series
gives only the conventional limits approach and does not even reference
the summability article, and the index does not reference the summability
article under series, only under summability. In the second edition [En87] of
1987, the summability material has been merged into the it Infinite series
article. Both [En77] and [En87], in their succinct but formalistic way, define
26 KEVIN CARMODY

a method as a linear transformation and therefore omit the Euler contin-

uation method. They also use weak equality, the assertion that equality
established through a method is relative to the method.
Hardy does not define method but does define and use weak equality.
Hardy additionally admits the Euler continuation method, as described
below. This method is best suited to power series, which is the key to the
whole theory.

METHODS OF SUMMATION

Hardy [Ha] defines several dozen methods of summation. We will

now examine a significant cross section of these methods, and a few that
have been developed since [Ha] was published. This includes all the meth-
ods commonly encountered in the literature and many other more obscure
methods. We will apply each one on the series

∞
X
ak , |a| > 1. (60)
k=0
Surprisingly, we will find that only one of these methods is capable of sum-
ming this series. That method is the Euler continuation method, which we
examine in detail in the next section.
A key criterion for assessing the validity of any method is whether P
Pis regular. APmethod is P is regular if, for any convergent series am ,
it
(P) am = lim am , i.e. if the method sums any convergent series to its
ordinary value as a limit. Most, but not all, methods defined in [Ha] are
regular.
For each of these methods:
• We give a page reference to [Ha], except for two methods which are
not in [Ha].
• We indicate whether the method is regular.
• We give a definition of the method when it is fairly simple. Otherwise,
we express our conclusions in the same notation that [Ha] uses.
• We apply each method to Equation (60).
• We denote partial sums as sn := n ak .
P
Euler method (E), which we call the Euler continuation method:
am
[Ha, p. 7]. See below for a full discussion. Regular. Sums (60) to 1−a .
P
Cesàro mean (C, 1): [Ha, p. 7]. Defined as (C,1) an =
limn→∞ s0 +s1 +s
n+1
2 +...+sn
. Regular. Diverges for (60).
Abel sum (A): [Ha, p. 7]. Defined as (A) an = limx→1− an xn if
P P
n
P
P an xn is Pconvergent
n n
Pfor 0n≤ x < 1. Regular. Does not apply to (60) since
an x = a x = (ax) is not convergent for all x ∈ [0, 1).
Divergent series 27

Euler’s polynomial method (E, 1): [Ha, p. 7]. Regular. Diverges for
(60), since bn = (a + 1)n > 2n .
Hutton’s method (Hu, k): [Ha, p. 21]. Regular. A limit of positive
terms for (60), and thus diverges.
Borel integral method (B0 ): [Ha, p. 83]. Regular. Sums (60) to 1−a
1
only
when <a < 1.
1
Borel exponential method (B): [Ha, p. 80]. Regular. Sums (60) to 1−a
only when <a < 1.
Nörlund means (N, pn ): [Ha, p. 64]. Scheme requiring choice of {pn }.
Regular for some {pn }. The partial quotients for (60) tm are always posi-
1
tive, and thus cannot yield 1−a , which is negative.
Abel means (A, λn ): [Ha, p. 71]. Scheme requiring choice of {λn }.
Regular. For (60), yields only positive terms, so the method yields a posi-
tive limit or no limit.
Lindelöf method (L): [Ha, p. 77]. Regular. Same as (A, n ln n) for (60),
which fails in all cases.
Mittag-Leffler method (M): [Ha, p. 79]. Regularity not stated in [Ha].
Similar to Lindelöf method. Hardy shows that L and M methods take (60)
1
to 1−a , but only on a region ∆ in the complex plane, called the Mittag-Leffler
star of an for a ∈ C. This region does not include any point in (1, ∞).
Riemann method (R, k): [Ha, p. 89]. Regular for k > 1. For (60), an
 k
eventually overwhelms sinnhnh , so the limit diverges.
Euler’s general polynomial method (E, q): [Ha, p. 178]. Regular. Re-
1
quires q > 0 and sums (60) to 1−a only within a circle with center at −q and
radius q + 1, which excludes a > 1.
General Cesàro means (C, k): [Ha, p. 96]. Regular for k > 0. Limit of
the quotient of two positive terms for (60), diverges.
Hölder means (H, k): [Ha, p. 94]. Regularity not stated in [Ha]. Limit
of positive terms for (60), diverges.
Ingham’s method (I): [Ha, p. 376]. Not regular. Limit of the sum of
positive terms for (60), diverges.
Second Nörlund method (N, pn ): [Ha, p. 57]. Regular. Yields a pos-
itive fraction for (60) and any set of positive {pn }, and thus cannot yield
1
1−a
, which is negative.
De la Vallée-Poussin’s method (VP): [Ha, p. 88]. Regular. Limit of
the sum of positive terms for (60), which is never negative.
P
Bernoulli summability (Be): Not in [Ha]. Defined by (Be) an =
limN→∞ N N
n N−n
, where sn = n ak and 0 < p < 1. Al-
P P
n=0 sn n p (1 − p)
ways yields a positive result for (60) and thus cannot assume a negative
value.
28 KEVIN CARMODY
P
Dirichlet summability (D): Not in [Ha]. Defined by (D) an =
limx→1+ ∞
P sn Pn
n=1 ns , where sn = ak . Always yields a positive result for
(60) and thus cannot assume a negative value.

THE EULER CONTINUATION METHOD OF SUMMATION

We now examine in detail the Euler continuation method of summa-

tion, which Hardy denotes as E. We have seen how every other method
commonly used in the current theory of divergent series fails to derive
P n 1
a = 1−a for |a| ≥ 1. Strangely, even though this is the only such method
that can derive this sum, it is usually omitted from works on divergent
series. For instance, [En87, p. 1415] defines a method as a linear transfor-
mation and thus defines away this method.
We will see that this method is poorly understood. We will develop a
better understanding of it and see how this yields a satisfactory theory of
divergent series. We start with Hardy’s definition [Ha, p. 7]:

If an xn is convergent for small x, and defines a function f(x)

P
of the complex variable x, one-valued and regular in an open and
connected region containing the originPand the point x = 1; and
f(1) = s; then we call s the E sum of an . The value of s may
naturally depend on the region chosen.

Unfortunately, this definition is circular. If the series defines f, then

n
P
f(x) = P an x within some region, and if this region contains x = 1, then
f(1) = an by definition. This does not extend the definition of the series
from the convergent case to the divergent case, but instead assumes that
we already have a definition of the divergent case.
Hardy also states [Ha, p. 10]:

In fact [the Euler continuation method]P

is not even regular, since
f(x) need not be regular at x = 1 when an converges.

It is even harder to tell what Hardy meant by this statement, since he

does not define what it means for a function to be regular.
It may be that Hardy meant something more by “defining” the func-
tion than simply constructing it from the series. Hardy states [Ha, p. 8]
that he is attempting to bring more rigor to Euler’s original method, since
the level of rigor in Euler’s time was generally less than today. Euler sum-
marized his method in a phrase that Hardy quotes [Ha, p. 8] and that is
well known in this field. Here is Euler’s original Latin and a translation:
Divergent series 29

Summa cujusque seriei est valor expressionis illius finitae, ex cujus evo-
lutione illa series oritur.

The sum of every series is the value of an expression which is de-

fined by the process from which that series arises.

This definition is quite different from Hardy’s. While it is simpler, it

unfortunately leaves us a bit in the dark about what kind of “process”
might give rise to a series, and whether the sum will be independent of
the process.
However, if we think in terms of radius of convergence, i.e. that the
1
has a series an which is valid within this radius, where the
P
function 1−a
series is convergent, then what we are looking for is a way of extending
the validity of the equality of the series and the function outside of the
radius of convergence. In this light, perhaps Euler’s criterion means that
we simply assume that this equality is valid. By “process,” Euler may have
1
meant the algebraic steps of transforming an to 1−a
P
.
We will take this as the basis of our interpretation of the Euler contin-
uation method. Here is a more expanded version of this interpretation:

If a function is continuous and has continous derivatives within

a region of the complex plane, and if a power series converges to
the function within a second region that is contained within the
first region, then we assume that the power series represents the
function everywhere within the first region.

In other words, we do not reject an algebraic process simply because

it deals with a divergent series. If we have found an algebraic process that
enables us to compute equality in the convergent case, then we assume
that that process is valid in the divergent case also. This method is clearly
regular, since it does not involve any transformation of a convergent series.
1
This method enables us to compute an = 1−a
P
for all a, except pos-
sibly a = 1. It is simple and natural, since it involves no limits and no
transformation of terms or partial sums. It can calculate the sum of se-
ries that no other method can, and there is no known series that any other
method can sum that this method cannot. Therefore, we take this as the as
the sole basis of the arithmatic approach. There is no need for a weakened
form of equality, since we assume that a divergent series is strictly equal
to its arithmetic sum. The results of this approach can be applied with-
out ambiguity, since the equalities it establishes are as stong as all other
equalities.
30 KEVIN CARMODY

In this approach, if we wish to use other methods of summation, we

denote them as modified summation
P rather than weakened equality. Szász
[Sz, p. 2] uses the notation P ( am ) = s. This leaves equality strong,
P but
unfortunately this implies that we first compute the value of am and
then apply P after we compute the sum, whereas the real situation is that
a method P operates first on the terms of the series, and then either sums
the modified terms or performs some other operation on them. This means
that P is really a modified form of summation. We will use the notation
P
(P) am = s to denote that a method P sums a series with terms am to the
value s.

SUMMARY OF THE NUMERISTIC APPROACH TO DIVERGENT SE-

RIES

The numeristic approach to divergent series developed in this paper

can be summarized as follows.
1. Use only strict equality, not any form of weakened equality.
2. Use algebraic transformations to compute the equality of a series with
its sum, instead of merely defining the sum as a limit.
3. Denote methods of summation other than the Euler continuation
method as modified forms of summation, instead of modified forms
of equality.
4. Use equinfinitesimals to determine the sum of an infinite number of
zeros, instead of assuming that the sum is always zero.

NUMERISTICS IN GENERAL

A deeper aspect of the numeristic approach is the view that numbers

and arithmetic stand on their own and do not require definitions to exist.
In Although our discussions require that we define our terms, definitions
are not a substitute for mathematical truth. The truth of mathematical
expressions must ultimately be verified by observation of nature, just as
in any other science. In other words, mathematical truth cannot simply be
defined into existence.
In this light, we examine Hardy’s remarks in [Mo, p. 5–6] concerning
the role of definitions serve in the methods approach to divergent series,
which, in the numeristic view, contain a mixture of both valid and exces-
sive claims about definitions.

[I]t does not occur to a modern mathematician that a collection of

mathematical symbols should have a ‘meaning’ until one has been
Divergent series 31

assigned by definition. It was not a triviality even to the greatest

mathematicians of the eighteenth century. They had not the habit
of definition; it was not natural to them to say, in so many words,
‘by X we mean Y. There are reservations to be made, . . . but it is
broadly true to say that mathematicians before Cauchy asked not
‘How shall we define 1 − 1 + 1 − . . .?’ but ‘What is 1 − 1 + 1 − . . .?’,
and that this habit of mind led them into unnessary perplexities
and controversies which were often really verbal. [Emphasis his.]

While Hardy further acknowledges that the value of 12 seems “natu-

ral,” he says that assigning this as the sole value of the series actually has
problems, which he attributes to lack of proper definition. His solution is
to use methods of summation, with its concept of weak equality. However,
his definition of the Euler continuation method is quite different from Eu-
ler’s, and it is circular. He seems to miss its essence as a simple extension
of arithmetic.
The numeristic approach gains support from the recent development
of Maharishi Vedic Mathematics, a formulation of ancient Vedic philoso-
phies and technologies of consciousness in mathematical terms [CaI, My]
by Maharishi Mahesh Yogi, the founder of the Transcendental Meditation
program. In this approach, pure consciousness is the basic experience and
governing intelligence of life. The technologies of Maharishi Vedic Math-
ematics, including Transcendental Meditation, give the experience of pure
consciousness. Pure consciousness is equated with zero, which is charac-
terized as the point of infinity and the Absolute Number, the support of all
number systems, and thereby all of natural law. Number thereby becomes
a natural experience of self-referral available to everyone.
In the author’s experience, Maharishi Vedic Mathematics gives both
experience and understanding that makes one at home with the infinite.
The infinite sums of the numeristic approach to divergent series are an
example of the surprising and beautiful nature of the infinite.
32 KEVIN CARMODY

APPENDIX: MATHEMATICIANS’ OPINIONS ABOUT DIVERGENT SE-

RIES

The theory of divergent series makes startling claims about elemen-

tary facts of arithmetic which it often does not prove in a convincing
way. It therefore elicits controversy even from well known mathemati-
cians. Here is a sampling.

Lorsque s(n), sans tendre vers une limite, admet une valeur moyenne
s finie et déterminée, nous dirons que la série a(0) + a(1) + a(2) + . . .
est simplement indéterminée, et nous conviendrons de dire que s est la
somme de la série.

When s(n), without tending towards a limit, admits an average

value s which is finite and determinate, we will say that the series
a(0) + a(1) + a(2) + . . . is simply indeterminate, and we will agree
to say that s is the sum of the series.

Cesàro, quoted in [Ha, p. 8]

Il résulte de là une classification des séries indéterminées, qui est sans
doute incomplte et pas assez naturelle.

This results in a classification of the indeterminate series, which is

undoubtedly incomplete and not natural enough.

Cesàro, quoted in [Ha, p. 8]

[1 + 2 + 4 + 8 + . . . = −1] has an air of paradox, since it does not seem

natural to attribute a negative sum to a series of positive terms.

Hardy, [Ha, p. 10]

−1 and ∞ are the only “natural” sums [of 1 + 2 + 4 + 8 + . . .].

Hardy, [Ha, p. 19]

Divergent series 33

Porro hoc argumentandi genus, etsi Metaphysicum magis quam Mathe-

maticum videatur, tamen firmum est: et alioqui Canonum Verae Meta-
physicae major est usus in Mathesi, In Analysi, in ipsa Geometria, quam
vulgo putatur.

Again, this kind of argument, although it appears more metaphys-

ically magical than mathematical, nevertheless is well founded;
and besides, the true canon of the metaphysics of our forefathers
is used in mathematics, in analysis, in its geometry, for ordinary
reckoning.

Leibnitz, quoted in [Ha, p. 14]

Les géomtres doivent savoir gré au cit. Callet d’avoir appelé leur at-
tention sur l’espce de paradoxe que présentent les séries dont il s’agit,
et d’avoir cherché à les prémunir contre l’application des raisonnements
métaphysiques aux questions qui, n’étant que de pure analyse, ne peu-
vent être décidées que par les premeiers principes et les rgles fondamen-
tales du calcul.

The geometricians must agree with the cited text. Callet has
drawn their attention to the species of paradox that present the
series as it really is, and to have sought to secure them against the
application of metaphysical reasoning on questions which, not be-
ing that of pure analysis, can be decided only by first principles
and the fundamental rules of calculation.

Lagrange, quoted in [Ha, p. 14]

Darüber hat er zwar kein Exempel gegeben, ich glaube aber gewiß zu
sein, daß nimmer eben dieselbe series aus der Evolution zweier wirklich
verschiedenen expressionum finitorum entstehen könne.

He has given no example about it; however I believe it is certain

that the same series can never develop from the evolution of two
really different finite expressions.

Euler, quoted in [Ha, p. 14]

Per rationes metaphysicas . . . quibus in analysi acquiescere queamus.

34 KEVIN CARMODY

By metaphysical reasoning . . . which is able to submit to analysis.

Euler, quoted in [Ha, p. 14]

Ich glaube, daß jede series einen bestimmten Wert haben müsse. Um aber
allen Schwierigkeiten, welche dagegen gemacht worden, zu begegnen, so
sollte dieser Wert nicht mit dem Namen der Summe belegt werden, weil
man mit deisem Wort gemeiniglich einen solchen Begriff zu verknüpfen
pflegt, als wenn die Summe durch eine wirkliche Summierung heraus-
gebracht würde: welche Idee bei den seriebus divergentibus nicht stat-
tfindet.

I believe that every series must have a certain value. However, in

order to meet all difficulties which could be made against it, then
this value should not be assigned the name of sum, because one
tends to commonly link such a term with this word, as if the sum
was brought about by a real summation, which idea does not take
place with divergent series.

Euler, quoted in [Ha, p. 15]

Quamadmodum autem iste dissensus realis videatur, tamen neutra pars

ab altera ullius erroris argui potest, quoties in analysi hujusmodi se-
rierum usus occurrit: quod gravi argumento esse debet, neutram partem
in errore versari, sed totum dissidium in solis verbis esse positum.

However, whatever that disagreement seems to give rise to, still

neither party can prove any error by the other, as often in analy-
sis this kind of series resists being used; because serious evidence
must exist, neither party is in error, but everyone disagrees only
in words of expression.

Euler, quoted in [Ha, p. 15]

Cette série n’est ni convergente ni divergente et ce n’est qu’en la con-

sidérant ainsi que nous la faisons comme la limite d’une série conver-
gente, qu’elle peut avoir une valeur déterminée. . . . Nous admettrons
avec Euler que les sommes de ces séries considérées en elles-même n’ont
pas de valueurs déterminées; mais nous ajouterons que chacune d’elles a
une valeur unique et qu’on peut les employer dans l’analyse, lorsqu’on
les regarde comme les limites des séries convergentes, c’est à dire quand
Divergent series 35

on suppose implicitement leurs termes successifs multipliés par les puis-

sances d’une fraction infiniment peu différent de l’unité.

This series is neither convergent nor divergent, and it is not that

by considering it thus that we make it like the limit of a convergent
series, that it can have a given value. . . . We will admit with Eu-
ler that the sums of these series considered in themselves do not
have definite values; but we will add that each one of them has
a single value and that one can employ them in analysis, when
one looks upon them like the limits of convergent series, this be-
ing said when one implicitly supposes their successive terms mul-
tiplied by the powers of a fraction differing infinitesimally from
unity.

Poisson, quoted in [Ha, p. 17]

Pour moi j’avoue que touts les raisonnements et les calculs fondés sur des
séries que ne sont pas convergents . . . me paratront toujours trs suspects,
même quand les résultats de ces raisonnements s’accorderaient avec des
vérités connues d’ailleurs.

For my part, I acknowledge that all the reasoning and the calcula-
tions based on series that are not convergent . . . will always appear
very suspect to me, even when the results of this reasoning would
agree with truths known elsewhere.

D’Alembert, quoted in [Ha, p. 17]

Je mets encore au rang des illusions l’application que Leibniz et Dan.

Bernoulli ont faite du calcul des probabilités.

I still put at the level of illusion the application that Leibniz and
Dan. Bernoulli have made of the theory of probability.

Laplace, quoted in [Ha, p. 17]

[Divergent series is] the only subject yet remaining, of an elemen-

tary character, on which serious schism exists among mathemati-
cians as to absolute correctness or incorrectlness of results. . . . The
moderns seem to me to have made a similar confusion in regard
to their rejection of divergent series; meaning sometimes that they
36 KEVIN CARMODY

cannot safely be used under existing ideas as to their meaning and

origin, sometimes that the mere idea of anyone applying them at
all, under any circumstances, is an absurdity. We must admit that
many series are such as we cannot safely use, except as means
of discovery, the results of which are to be subsequently verified.
. . . But to say that what we cannot use no others ever can . . . seems
to me a departure from al rules of prudence.

De Morgan, quoted in [Ha, p. 19]

Would analysis ever have developed as it has done if Euler and

√
others had refused to use −1?

Hardy, [Ha, p. 19], immediately after the above De Morgan quote

The divergent series are the invention of the devil, and it is a

shame to base on them any demonstration whatsoever.

Abel, quoted in [Ga, p. 170–171]

I must say a few words on the subject of . . . divergent series. . . . It

is not easy to get up any enthusiasm after it has been artificially
cooled by the wet blankets of rigorists. . . . I have stated the growth
of my views about divergent series. . . . I have avoided defining the
meaning of equivalence. The definitions will make themselves in
time. . . . My first notion of a series was that to have a finite value
it must be convergent. . . . A divergent series also, of course, has
an infinite value. Solutions of physical problems must always be
in finite terms or convergent series, otherwise nonsense is made.
. . . Then came a partial removal of ignorant blindness. In some
physical problems divergent series are actually used, notably by
Stokes, referring to the divergent formula for the oscillating func-
tion Jn (x). He showed that the error was less than the last term in-
cluded. . . . Equivalence does not mean identity. . . . But the numer-
ical meaning of divergent series still remains obscure. . . . There
will have to be a theory of divergent series, or say a larger theory
of functions than the present, including convergent and divergent
series in one harmonious whole.

Heaviside, quoted in [Ha, p. 36]

Divergent series 37

ACKNOWLEDGMENT

The author wishes to thank the organizers, participants, and supporters of the
Invincible America Assembly at Maharishi University of Management in Fair-
field, Iowa for creating one of the best research environments in the world.

REFERENCES

Bo É. Borel, Leçons sur les series divergentes, Gauthier-Villars, 1901.

CaE K. Carmody, Equinfinitesimal analysis: An extension of numeris-
tics, http://kevincarmody.com/math/equinfinitesimal.pdf.
CaI K. Carmody, Is consciousness a number? How Maharishi Vedic
Mathematics resolves problems in the foundations and philoso-
phy of mathematics, http://kevincarmody.com/math/mvmath.
pdf.
CaN K. Carmody, Numeristics: a number-based foundational theory of
mathematics, http://kevincarmody.com/math/numeristics.pdf.
En77 Encyclopedic Dictionary of Mathematics (first edition), MIT Press,
1977.
En87 Encyclopedic Dictionary of Mathematics (second edition), MIT Press,
1987.
Eu L. Euler, De seriebus divergentibus, orig. pub. 1760, repro. Opera
Omnia I:14, p. 585–617, Teubner, 1925, or http://math.dartmouth.
edu/∼euler/docs/originals/E247.pdf.
Fo W. Ford, Studies on divergent series, Chelsea, 1960.
Ga M. Gardner, The Sixth Book of Mathematical Games from Scientific
American, University of Chicago Press, 1984.
Ha G. H. Hardy, Divergent series (second ed.), Chelsea, 1991 (first edi-
tion published at Oxford in 1949).
Ma Maharishi Mahesh Yogi, Maharishi’s Absolute Theory of Defence,
Age of Enlightenment Publications, 1996.
Mo C. Moore, Summable series and convergence factors, Dover, 1966.
Mu65 C. Musès, Systemic stability and cybernetic control: An introduc-
tion to the cybernetics of higher integrated behavior, in Cybernet-
ics of neural processes: course held at the International School of Physics,
sponsored by NATO at the Istituto di fisica teorica, Università di Napoli,
April 26–May 13, 1962: proceedings, ed. E. R. Caianiello, Roma,
Consiglio nazionale delle ricerche, 1965, pp. 149–212.
Mu72 C. Musès, Anatomy of zero and infinity, Journal for Study of Con-
sciousness 6 (1972) 308–312.
38 KEVIN CARMODY

Sm L. L. Smail, Some generalizations in the theory of summable divergent

series, Press of the New Era, 1913.
Sz O. Szász, Introduction to the theory of divergent series, Univ. of
Cincinnati Math. Dept., 1952.