Math 133 Absolute Convergence Stewart §11.

6/II

Series with positive terms. So far, we have mostly considered positive series ∞
P
PN n=1 an
with an ≥ 0, whose partial sums sN = n=1 an = a1 + a2 + · · · + aN can only increase as
we add more positive terms. As N → ∞, these can behave in one of two ways:
∞
• Convergence: partial sums level off beneath a ceiling value:∗ lim sN =
P
an = L.
N →∞ n=1
∞
P
• Divergence to infinity: partial sums increase without bound: lim sN = an = ∞.
N →∞ n=1
We can picture the sequence {sn }∞
n=1 as a line graph connecting the points (n, sn ):

Series with positive and negative terms. In the more general case where an can be
positive or negative, the partial sums can osciallate up and down depending on the sign
of each term added.
• Convergence (oscillating): partial sums wiggle above and below the horizontal asymp-
P∞
tote which is their limiting value: lim sN = an = L.
N →∞ n=1
• Divergence to infinity (oscillating): partial sums have more ups than downs, making
P∞
an overall increase without bound: lim sN = an = ∞ ; or more downs than
N →∞ n=1
ups, so the limit is −∞.
• Divergence (indecisive oscillation): partial sums do not consistently go up or down
∞
P
or approach a horizontal asymptote, so lim sN = an does not exist at all.
N →∞ n=1

Notes by Peter Magyar [email protected]

∗
In fact if the increasing partial sums have an upper bound, sn ≤ B for all n, then the completeness
axiom of real analysis states that the least upper bound lim sn exists.
n→∞
An example of indecisive oscillation is an = (−1)n , for which:

1 for n odd
sn = 1 − 1 + 1 − 1 + · · · ± 1 =
0 for n even.

Absolute convergence. We say that a seriesP ∞

P
n=1 an is absolutely convergent whenever
∞
the series of absolute values is convergent: n=1 |an | = M . A series is conditionally
convergent if it is convergent, ∞
P P∞
a
PNn
n=1 = L, but n=1 |an | = ∞.
In terms of the graph of sN = n=1 an , absolute convergence means the total length
of ups and downs is a finite number M . Equivalently, if we change all down steps PNan < 0
to up steps |an | > 0, we obtain the graph of a convergent positive series tN = n=1 |an |
converging to the ceiling M :

P∞
Absolute Convergence Theorem:
P∞ If a series is absolutely convergent with n=1 |an | = M,
then it is convergent with n=1 an = L.
Proof: LetP
P bn = |an |, and p(n) = ±1 be the sign of an , so that an = p(n) bn . By hypothesis,
|an | = bn is convergent, hence so are the sums of only the positive an and only the
negative an :
X∞ X∞
bn = L1 and bn = L2 .
n=1 n=1
p(n)=+1 p(n)=−1

Now:
P∞ N
P (∗) N
P N
P
n=1 an = lim an = lim bn − bn
N →∞ n=1 N →∞ n=1 n=1
p(n)=+1 p(n)=−1

(∗∗) N
P N
P
= lim bn − lim bn = L1 − L2
N →∞ n=1 N →∞ n=1
p(n)=+1 p(n)=−1

Here the equality (∗) follows from rearranging a finite sum of terms, and (∗∗) follows from
the Limit Sum Law from Calculus I §1.6.

Series with alternating signs. We say that a series is alternating when successive
terms an are of opposite sign; i.e. an = (−1)n bn or an = (−1)n−1 bn with bn ≥ 0.
 Alternating Series Test: If an is an alternating series withPbn = |an | decreasing,
meaning bn ≥ bn+1 for all n, and limn→∞ bn = 0, then ∞ n=1 an converges to
some L. Also, the error of a partial sum is bounded by the next term:
N
X
L− an ≤ |aN +1 |.
n=1

Proof:
PN Assuming an = (−1)n−1 bn where b1 ≥ b2 ≥ b3 ≥ · · · ≥ 0, and setting sN =
n=1 an = b1 − b2 + b3 − b4 + · · · ± bN , we see that:

s3 = b1 − b2 + b3 = b1 − (b2 − b3 ) < b1 = s1 ,

and similarly:
s2 ≤ s4 ≤ s6 ≤ · · · ≤ s5 ≤ s3 ≤ s1 ,
so the even values of sN form an increasing subsequence, and the odd values form a
decreasing subsequence. Furthermore, we have limn→∞ |sn+1 − sn | = limn→∞ bn = 0, so
the even and odd subsequences become arbitrarily close, clearly zeroing in on a finite limit
L. Error estimate: for N even, sN ≤ L ≤ sN +1 = sN + bN +1 ; similarly for N odd. Q.E.D.

Absolutely convergent series have several nice properties which conditionally convergent
series lack. For example, if we rearrange the order of terms in an absolutely convergent
series, the limit does not change, but this is not true for a conditionally convergent series.
example: Consider ∞ n−1 1 = 1 − 1 + 1 − 1 + · · · , which is convergent by the
P
n=1 (−1) n 2 3 4
Alternating Series Test. We easily see that the series of positive terms ∞
† 1
P
P∞ n=1 2n−1 = ∞
1
and the series of negative terms n=1 (− 2n ) = −∞ are both divergent, so the conditionally
convergent sum of the alternating series involves competing infinities. If we rearrange to
give the positive terms a head start, so that a large number of positive terms outrun each
negative term, then the positive infinity will win. In a sum like:
1
1+ 3 + 15 − 21 + 1
7 + 1
9 + ··· + 1 1
21 − 4 + 1
23 + 1
25 + ··· + 1 1
101 − 6 + ··· ,

all the terms an = (−1)n−1 n1 eventually appear, but the partial sums tend to ∞, not to
the finite value of the original alternating series.
† 1 1 1
In fact, we will see later that 1 − 2
+ 3
− 4
+ · · · = ln(2).