Abel's Theorem: A. Eremenko October 24, 2020

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Abel’s theorem

A. Eremenko
October 24, 2020

Abel’s Theorem. Let ∞


X
an = A (1)
n=0

be a convergent series with sum A. Define



an z n ,
X
f (z) = |z| < 1. (2)
n=0

Then
lim f (x) = A, (3)
x→1−

where x is real.
Condition that the series is convergent implies that the radius of conver-
gence of the series in (2) is at least 1, so definition (2) makes sense.
Proof of the Theorem. Replacing a0 by a0 − A we may assume wlog that
A = 0. Consider the partial sums sk = a0 + . . . + an ; and notice the equality

sn z n .
X
f (z) = (1 − z) (4)
n=0

Since the series (1) is convergent its partial sums are bounded, so the series
in (4) has radius of convergence at least 1. So the terms in the RHS can be
rearranged:

s0 + s1 z + s2 z 2 + . . . − (zs0 + z 2 s1 + z 3 s2 + . . .
= s0 + (s1 − s0 )z + (s2 − s1 )z 2 + . . . = a0 + a1 z + a2 z 2 + . . . .

1
Now to prove that f (x) → 0 for real x → 0−, we choose arbitrary  > 0 and
find an integer N such that

|sn | ≤  for n ≥ N.

Then ∞


n |1 − z|
|z|n = 
X X
(1 − z) s z ≤ |1 − z| .

n
1 − |z|


n=N n=N

If z is on [0, 1), the last fraction equals 1, so the tail of the series has absolute
value at most  and this holds for all z ∈ (0, 1). Now for the part up to N − 1
of our series we evidently have
N −1
sn z n → 0 as z → 1,
X
(1 − z)
n=0

since for fixed N the sum is bounded. Therefore, |f (z)| < 2 when z ∈ (0, 1)
is sufficiently close to 1, and this proves the Theorem.
Remark. We did not fully use that z is positive in (3). It is sufficient that
|1 − z|/(1 − |z|) stays bounded as z → 1, and this is called a non-tangential
limit. This means that z stays in some sector of opening < π with the vertex
at 1 bisected by the interval (0, 1). Such a sector is sometimes called a Stolz
angle.
Example of application. Find the sum

X (−1)n−1
S= = 1 − 1/2 + 1/3 − 1/4 + . . . .
n=1 n
This series is convergent by the alternating series criterion. So we can apply
Abel’s theorem to the function

X (−1)n−1 z n
f (z) = .
1 n
This function can be found explicitly: just differentiate and obtain
∞ ∞
1
f 0 (z) = (−1)n−1 z n−1 = (−1)n z n =
X X
,
n=1 n=0 1+z
So
f (z) = Log (1 + z),

2
where we used that f (0) = 0, to choose the correct branch. This function is
continuous at z = 1, so we can plug z = 1 and obtain that S = Log 2.
Another example. Find the sum of the Leibniz series:

X (−1)n
S= = 1 − 1/3 + 1/4 − 1/5 + . . . .
n=0 2n + 1

Let ∞
z 2n+1
(−1)n
X
f (z) = .
n=0 2n + 1
Then ∞
1
f 0 (z) = (−1)n z 2n =
X
,
n=0 1 + z2
so, since f (0) = 0, Z z
f (z) − f 0 (t)dt = Arctan z,
0

and plugging z = 1 we obtain the answer S = π/4.


It is said that Leibnitz decided to quit being a lawyer and a diplomat in
order to pursue mathematics because of this discovery.
Abel’s theorem permits to prescribe sums to some divergent series, this
is called the summation in the sense of Abel. If we have an analytic function
f in the unit disk and the limit in (3) exists, then we call this limit the sum
of the series (2) in the sense of Abel.
Abel’s theorem ensures that this is indeed a generalization of convergence
in the ordinary sense: a convergent series is Abel-summable and its sum in
the sense of Abel is the same as its ordinary sum.
Example. The divergent series

(−1)n−1 n = 1 − 2 + 3 − 4 + . . .
X

n=1

has the Abel sum 1/4. Indeed, since


∞ ∞ ∞
n−1 n n n−1 d X
(−1)n−1 z n
X X
f (z) = (−1) nz = z (−1) nz =z
n=1 n=1 dz n=1
d z z2
= = .
dz 1 + z (1 + z)2

3
Plugging z = 1 we obtain that the sum equals 1/4.
Exercise. Show that Abel’s sum of

1
(−1)n = 1 − 1 + 1 − 1 + 1 − 1 + . . . = .
X

n=0 2

From this example we see that grouping the terms does not preserve Abel’s
sum.
It may happen that the function

an z n
X
f (z) =
n=0

which is used in Abel’s summation has a (single valued) analytic continuation.


Then some values of z with |z| greater than the radius of convergence can be
unambigously plugged to it. This is called Euler’s summation. For example,
in Euler’s sense:
∞ ∞
!
1 1

n n n
X X
(−2) = (−2) z = = ,
n=0 n=0 z=1
1 + 2z z=1 3

and similarly

1

n
X
2 = = −1.
n=0 1 − 2n z=1
(The radius of convergence of both series is 1/2.)
This becomes ambiguous if the analytic continuation of our function is
not single valued, thus in our first example, it is not clear how to plug for
example z = −2 into Log (1 + z).

You might also like