Chapter 6

Sequences and Series of Real

Numbers

We often use sequences and series of numbers without thinking about it. A decimal
representation of a number is an example of a series, the bracketing of a real number
by closer and closer rational numbers gives us an example of a sequence. We want
to study these objects more closely because this conceptual framework will be used
later when we look at functions and sequences and series of functions. First, we will
take on numbers.
Sequences have an ancient history dating back at least as far as Archimedes who
used sequences and series in his “Method of Exhaustion” to compute better values of
π and areas of geometric figures.

6.1 The Symbols +∞ and −∞

We often use the symbols +∞ and −∞ in mathematics, including courses in high
school. We need to come to some agreement about these symbols. We will often write
∞ for +∞ when it should not be confusing.
First of all, they are not real numbers and do not necessarily adhere to the rules
of arithmetic for real numbers. There are times that we “act” as if they do, so we
need to be careful.
We adjoin +∞ and −∞ to R and extend the usual ordering to the set R ∪
{+∞, −∞}. Explicitly, we will agree that −∞ < a < +∞ for every real number
a ∈ R ∪ {+∞, −∞}. This gives the extended set with an ordering that satisfies our
usual properties:

1) If a, b ∈ R ∪ {+∞, −∞}, then a ≤ b or b ≤ a.

2) If a ≤ b and b ≤ a, then a = b.
3) If a ≤ b and b ≤ c, then a ≤ c.
We will not extend the usual algebraic structure of the reals to R ∪ {+∞, −∞}.

73
74 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

Instead, when we have to we will discuss the algebra that might involve +∞ and/or
−∞. Do not apply any theorem that is stated for the real numbers to the symbols
+∞ or −∞.
The symbols make it convenient to extend our notation about intervals to the real
line.

[a, ∞) = {x ∈ R | a ≤ x} (a, ∞) = {x ∈ R | a < x}

(−∞, b] = {x ∈ R | x ≤ b} (−∞, b) = {x ∈ R | x < b}.

Occasionally, you will see R = (−∞, ∞).

6.2 Sequences
Sequences are, basically, countably many numbers arranged in an ordered set that
may or may not exhibit certain patterns.

Definition 6.1 A sequence of real numbers is a function whose domain is a set of

the form {n ∈ Z | n ≥ m} where m is usually 0 or 1. Thus, a sequence is a function
f : N → R. Thus a sequence can be denoted by f (m), f (m + 1), f (m + 2), . . . .
Usually, we will denote such a sequence by {ai }∞i=m or {am , am+1 , am+2 , . . .}, where
ai = f (i). If m = 1, we may use the notation {an }n∈N .

Example 6.1 The sequence {1, 1/2, 1/3, 1/4, 1/5, . . . } is written as {1/i}∞
i=1 . Keep
in mind that this sequence can be thought of as an ordinary function. In this case
f (n) = 1/n.

Example 6.2 Consider the sequence given by an = (−1)n for n ≥ 0. This time we
have started the sequence with 1 and the terms look like, {1, −1, 1, −1, 1, −1, . . .}.
Note that this time the function has domain N but the range is {−1, 1}.
¡ ¢
Example 6.3 Consider the sequence an = cos πn 3
, n ∈ N. The first term in the
π ◦ 1
sequence is cos 3 = cos 60 = 2 and the sequence looks like

1 1 1 1 1 1 1 1 1 1
, − , −1, − , , 1, , − , −1, − , , 1, , − , −1, . . .}.
2 2 2 2 2 2 2 2 2 2
Note that likes its predecessor, the function takes on only a finite number of values,
but the sequence has an infinite number of elements.

Example 6.4 If an = n1/n , n ∈ N, the sequence is

√
1, 2, 31/3 , 41/4 , . . .

MATH 6101-090 Fall 2006

6.2. SEQUENCES 75

We might use an approximation for each of these and, arbitrarily choosing 5 decimal
places, the sequence would look like

1, 1.41421, 1.44225, 1.41421, 1.37973, 1.34801, 1.32047, 1.29684, 1.27652, 1.25893, . . .

We would find that a100 = 1.04713 and a10000 = 1.00092.

Example 6.5 Consider the sequence bn = (1 + n1 )n , n ∈ N. This is the sequence

µ ¶2 µ ¶3 µ ¶4
3 4 5
2, , , ,...
2 3 4

or by approximation

2, 2.25, 2.37037, 2.44141, 2.48832, 2.52163, 2.54650, 2.56578, 2.58117, 2.59374 . . . .

Again a100 = 2.74081 and a10000 = 2.71815.

In looking at these examples we might think that some of them are giving us a
pattern of numbers that are “getting close” to some other real number. Others may
not give us that indication. We are interested in what the long-term behavior of the
sequence is. What happens for larger and larger values of n? Does the sequence
approach a real number? Could it approach more than one real number?

Definition 6.2 A sequence of real numbers is said to converge to a real number L

if for every ² > 0 there is an integer N > 0 such that if k > N then |ak − L| < ².
The number L is called the limit of the sequence.

If {ak } converges to L we will write lim ak = L or simply

k→∞
ak → L. If a sequence does not converge, then we say that it n ≈ n/2n
diverges. 1 0.5
Note that the N in the definition depends on the ² that we 2 0.5
were given. If you change the value of ² then you may have to 3 0.375
“recalculate” N . 4 0.25
n
Consider the sequence an = n , n ∈ N. Now, if we look at 5 0.15625
2 6 0.09375
the values that the sequence takes
7 0.0546875
1 2 3 4 8 0.03125
, , , ,...
2 22 23 24 9 0.0175781
10 0.00976562
we might think that the terms are getting smaller and smaller
so maybe the limit of this sequence would be 0. Let’s take a
look and compare how N would vary as ² varies. Let’s start
with some simple small numbers and let ² be 0.1, 0.01, 0.001, and 0.0001, and 0.00001.

MATH 6101-090 Fall 2006

76 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

For ² = 0.1, we need to find an integer N so that

¯ ¯
¯N ¯
¯ ¯
¯ 2N − 0¯ < 0.1

Look in the table of values here and we see that for N = 6 we have satisfied the above
condition. Following this we get the following by using a calculator or a computer
algebra system:

¯ ¯
¯N ¯
N > 0 implies ¯ − 0¯¯ < 1
¯ 2N
¯ ¯
¯N ¯
N > 5 implies ¯ − 0¯¯ < 0.1
¯ 2N
¯ ¯
¯N ¯
N > 9 implies ¯ − 0¯¯ < 0.01
¯ 2N
¯ ¯
¯N ¯
N > 14 implies ¯ − 0¯¯ < 0.001
¯ 2N
¯ ¯
¯N ¯
N > 18 implies ¯ − 0¯¯ < 0.0001
¯ 2N
¯ ¯
¯N ¯
N > 22 implies ¯ − 0¯¯ < 0.00001
¯ 2N

We are going to establish several properties of convergent sequences. Many proofs

will use a proof much like this next result. While this type of argument may not easy
to get used to, it will appear again and again, so you should try to get as familiar
with it as you can.

Theorem 6.1 (Convergent sequences are bounded) Let {an }, n ∈ N be a con-

vergent sequence. Then the sequence is bounded, and the limit is unique.

Proof: The easier property to show is that the limit is unique, so let’s do that first.
Suppose the sequence has two limits, L and K. Take any ² > 0. Then there is an
integer N such that
²
|ak − L| < if k > N.
2
0
Also, there is another integer N such that
²
|ak − K| < if k > N 0 .
2
Then, by the triangle inequality:
² ²
|L − K| < |ak − L| + |ak − K| < + = ² if k > max{N, N 0 }.
2 2

MATH 6101-090 Fall 2006

6.3. THE ALGEBRA OF CONVERGENT SEQUENCES 77

Therefore |L − K| < ² for any ² > 0. But the only way that that can happen is for
L = K, so that the limit is indeed unique.
Next, we need to prove boundedness. Since the sequence converges, we can take
any ² we wish, and tradition shows us to take ² = 1. Then there is an integer N so
that
|ak − L| < 1 if k > N.
Fix that integer N . Then we have that
|an | ≤ |an − L| + |L| < 1 + |L| = P for all n > N.
Now, define M = max{{|ak |, k = 1, . . . , N }, P }. Then |an | < M for all n, which
makes the sequence bounded.

6.3 The Algebra of Convergent Sequences

This section proves some basic results that do not come as a surprise to the student.
Theorem 6.2 If the sequence {an } converges to L and c ∈ R, then the sequence
{can } converges to cL; i.e., lim can = c lim an .
n→∞ n→∞

Proof: Let’s assume that c 6= 0, since the result is trivial if c = 0. Let ² > 0. Since
{an } converges to L, we know that there is an N ∈ N so that if n > N
²
|an − L| < .
|c|
Thus, for n > N we then have that
²
|can − cL| = |c||an − L| < |c| = ².
|c|
which is what we needed to prove.
Theorem 6.3 If the sequence {an } converges to L and {bn } converges to M , then
the sequence {an + bn } converges to L + M ; i.e., lim (an + bn ) = lim an + lim bn .
n→∞ n→∞ n→∞

Proof: Let ² > 0. We need to find an N ∈ N so that if n > N

|(an + bn ) − (L + M )| < ².
Since {an } and {bn } are convergent, for the given ² there are integers N1 , N2 ∈ N so
that
² ²
If n > N1 then |an − L| < and if n > N2 then |bn − M | <
2 2
Thus, if n > max{N1 , N2 } then
² ²
|(an + bn ) − (L + M )| ≤ |an − L| + |bn − M | < + = ².
2 2

MATH 6101-090 Fall 2006

78 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

Theorem 6.4 If the sequence {an } converges to L and {bn } converges to M , then
the sequence {an · bn } converges to L · M ; i.e., lim (an · bn ) = lim an · lim bn .
n→∞ n→∞ n→∞

The trick with the inequalities here is to look at the inequality

|an bn − LM | = |an bn − an M + an M − LM |
≤ |an bn − an M | + |an M − LM |
= |an ||bn − M | + |M ||an − L|.
So for large values of n, |bn − M | and |an − L| are small and |M | is constant. Now,
by Theorem 6.1 shows that |an | is bounded, so that we will be able to show that
|an bn − LM | is small.
Proof: Let ² > 0. By Theorem 6.1 there is a constant K > 0 such that |an | ≤ K
for all n. Since {bn } is convergent, for the given ² there is an integer N1 ∈ N so that
²
If n > N1 then |bn − M | < .
2K
Also, since {an } is convergent there is an integer N2 ∈ N so that
²
If n > N2 then |an − L| < .
2(|M | + 1)
Thus if N = max{N1 , N2 } then if n > N
|an bn − LM | ≤ |an ||bn − M | + |M ||an − L|
² ² ² ²
≤ K· + |M | < + = ².
2K 2(|M | + 1) 2 2

Lemma 6.1 If the sequence {an } converges to L and {bn } converges to M and if
an ≤ bn for all n ≥ m, then L ≤ M .
Lemma 6.2 If the sequence {an } converges to L, if an 6= 0 for all n ∈ N, and if
L 6= 0, then glb{|an | | n ∈ N} > 0.
Proof: Let ² = 12 |L| > 0. Since {an } converges to L, there is an N ∈ N so that if
n > N then |an − L| < |L|/2.
Now if n > N we must have that |an | ≥ |L|/2. If not then the triangle inequality
would imply
|M | |M |
|L| = |L − an + an | ≤ |L − an | + |an | < + = |M |.
2 2
Now we set ½ ¾
|L|
m = min , |a1 |, |a2 |, . . . , |aN | .
2
Then clearly m > 0 and |an | ≥ m for all n ∈ N.

MATH 6101-090 Fall 2006

6.3. THE ALGEBRA OF CONVERGENT SEQUENCES 79

Theorem 6.5 If the sequence {an } converges to L, if an 6= 0 for all n ∈ N, and if

L 6= 0, then the sequence {1/an } converges to 1/L; i.e., lim 1/an = 1/ lim an .
n→∞ n→∞

Proof: Let ² > 0. By Lemma 6.2 there is an m > 0 such that |an | ≥ m for all n.
Since {an } is convergent there is an integer N ∈ N so that if n > N |L−an | < ²·m|L|.
Then for n > N ¯ ¯
¯1 ¯
¯ − ¯ = |an − L| ≤ |an − L| < ².
1
¯ an L ¯ |an L| m|L|

Theorem 6.6 Suppose that the sequence {bn } converges to M and if {an } converges
to L. If bn 6= 0 for all n ∈ N, and if M 6= 0, then the sequence {an /bn } converges to
an lim an
L/M ; i.e., lim = n→∞ .
n→∞ bn lim bn
n→∞

Proof: We use two of the previous theorems to prove this. By Theorem 6.5 {1/bn }
converges to 1/M , so

an 1 1 L
lim = lim an · =L· = .
n→∞ bn n→∞ bn M M

1
Example 6.6 Let p > 0 then lim = 0.
n→∞ np
1 1/p 1 1
Let ² > 0 and let N = n > N implies that np >
( ² ) . Then ²
and hence ² > np
.
Since np > 0, this shows that n > N implies | n1p − 0| < ².
1

Example 6.7 Let |a| < 1, then lim an = 0.

n→∞
Suppose a 6= 0, because lim an = 0 is clear for a = 0. Since |a| < 1, we can write
1
|a| = where b > 0. By the binomial theorem
1 + |b|

n(n − 1) 2
(1 + b)n = 1 + nb + b + · · · + bn ≥ 1 + nb > nb,
2
so
1 1
|an − 0| = |an | = n
< .
(1 + b) nb
1 1
Now consider ² > 0 and let N = ²b
. Then if n > N , we have n > ²b
and hence
1
|an − 0| < nb < ².

MATH 6101-090 Fall 2006

80 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

Example 6.8 lim (n1/n ) = 1.

n→∞
Let an = (n1/n ) − 1 and note that an ≥ 0 for all n. By Theorem 6.3 it is sufficient
for us to show that limn→∞ an = 0.
Since 1 + an = n1/n , we have n = (1 + an )n . For n ≥ 2 the binomial theorem gives
us
1 1
n = (1 + an )n ≥ 1 + nan + n(n − 1)a2n > n(n − 1)a2n .
2 2
q
2
Thus, n > 12 n(n − 1)a2n , so a2n < n−1 . Thus, we have shown that an < 2
n+1
for
n ≥ 2. Thus, lim an = 0.

Example 6.9 lim (a1/n ) = 1 for a > 0.

n→∞
Suppose a ≥ 1. Then for n ≥ a we have 1 ≤ a1/n ≤ n1/n . Since lim n1/n = 1,
it easily follows that lim a1/n = 1. Suppose that 0 < a < 1. Then a1 > 1, so that
lim( a1 )1/n = 1. Thus,
µ ¶1/n
1
lim = 1
a
11/n
lim 1/n = 1
a
1
= 1
lim a1/n
lim a1/n = 1

Definition 6.3 For a sequence {an } we write lim an = +∞ provided for each M > 0
there is a number N such that n > N implies that an > M .

In this case we will say that {an } diverges to +∞.

We can make a similar definition for lim an = −∞.
Of course, we cannot use the previous theorems when dealing with infinite limits.

Theorem 6.7 Let {an } and {bn } be sequences such that lim an = +∞ and lim bn > 0.
Then lim an bn = +∞.

Proof: Let M > 0. Choose a real number m so that 0 < m < lim bn . Whether
lim bn = +∞ or not, there exists N1 so that if n > N1 then bn > m. Since lim an =
+∞ there is an N2 so that if n > N2 then
M
an > .
m
M
Setting N = max{N1 , N2 } means that for n > N an bn > m
· m = M.

MATH 6101-090 Fall 2006

6.4. MONOTONICITY AND CAUCHY SEQUENCES 81

Theorem 6.8 For a sequence {an } of positive real numbers lim an = +∞ if and only
if lim a1n = 0.

Proof: Let {an } be a sequence of positive numbers. We need to show

1
If lim an = +∞ then lim =0 (6.1)
an
and
1
If lim = 0 then lim an = +∞. (6.2)
an
To prove 6.1 we will suppose that lim an = +∞. Let ² > 0 and let M = 1/². Since
{an } diverges to +∞, there is an N so that if n > N then an > M = 1/². Therefore,
if N > n then ² > 1/an > 0, so
¯ ¯
¯1 ¯
if n > N then ¯¯ − 0¯¯ < ².
an

Thus, lim 1/an = 0. This proves 6.1.

To prove 6.2 suppose that lim 1/an = 0 and
¯ let ¯M > 0. Let ² = 1/M . Then since
¯ ¯
² > 0 there is an N so that if n > N then ¯ a1n − 0¯ < ² = M1 . Since an > 0 we then
know that
1 1
if n > N then 0 < <
an M
and hence
if n > N then an > M.
This means that lim an = +∞ and 6.2 holds.

6.4 Monotonicity and Cauchy Sequences

The previous section showed us how to work with convergent sequences, but does not
tell us how to determine (quickly) if a sequence does converge. We need such a tool.

Definition 6.4 A sequence {an } of real numbers is called monotonically increas-

ing1 if an ≤ an+1 for all n, and {an } is called a monotonically decreasing2 if
an ≥ an+1 for all n.

If a sequence is monotonically increasing or monotonically decreasing, we will call

it a monotonic sequence or a monotone sequence.

Theorem 6.9 All bounded monotone sequences converge.

1
Sometimes called nondecreasing
2
Sometimes called nonincreasing.

MATH 6101-090 Fall 2006

82 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

Proof: Let {an } be a bounded monotonically increasing sequence and let S = {an
mod n ∈ N}. Since the sequence is bounded, an < M for some real number M and
for all n ∈ N. This means that the set S is bounded, and thus it has a least upper
bound. Let u = lub S. Let ² > 0. Since u = lub S and ² > 0, u − ² is not an upper
bound for S. This means that there must be some N so that aN > u − ². Since
{an } is monotonically increasing we have that for all n > N an ≥ aN and hence for
all n > N it follows that u − ² < an ≤ u. Thus, |an − u| < ² for all n > N . Thus,
lim an = u = lub S.
The proof for bounded monotonically decreasing sequences is the same with the
greatest lower bound playing the role of the least upper bound.

Now, we can also handle unbounded monotone sequences.

Theorem 6.10 Let {an } be a sequence of real numbers.
(i) If {an } is an unbounded monotonically increasing sequence, then lim an = +∞.

(ii) If {an } is an unbounded monotonically decreasing sequence, then lim an = −∞.

Let {an } be a bounded sequence of real numbers. While it may converge or
may not converge, the limiting behavior of {an } depends only on the “tails” of the
sequence, or sets of the form {an | n > N }. This leads us to a concept that we can
discuss without knowing a priori if a given sequence converges or diverges.
Let uN = glb{an | n > N } = inf{an | n > N } and let vn = lub{an | n > N } =
sup{an | n > N }. We have seen that if lim an exists, then it must lie in the interval
[uN , vN ]. As N increases, the sets {an | n > N } get smaller, so we have

u1 ≤ u2 ≤ u3 ≤ . . . and v1 ≥ v2 ≥ v3 ≥ . . . .

By the above theorem the limits u = limN →∞ uN and v = limN →∞ vN both exist
and u ≤ v since uN ≤ vN for all N . If the limit exists, then uN ≤ lim an ≤ vN so
u ≤ lim an ≤ v. These numbers u and v turn out to be useful whether lim an exists
or not.

Definition 6.5 Let {an } be a sequence of real numbers. We define

lim sup an = lim lub{sn | n > N }

N →∞

and
lim inf an = lim glb{sn | n > N }.
N →∞

Note that we do not require that {an } be bounded. We will take some precautions
and adopt the following conventions. If {an } is not bounded above, lub{sn | n >
N } = +∞ for all N and we define lim sup an = +∞. Likewise, if {an } is not bounded
below, glb{sn | n > N } = −∞ and we define lim inf an = −∞ in this case.

MATH 6101-090 Fall 2006

6.4. MONOTONICITY AND CAUCHY SEQUENCES 83

Now, is it true that lim sup an = lub{an | n > N }? Not necessarily, because while
it is true that lim sup an ≤ lub{an | n > N }, some of the values an may be much
larger than lim sup an . Note that lim sup an is the largest value that infinitely many
an ’s can get close to.

Theorem 6.11 Let {an } be a sequence of real numbers.

(i) If lim an is defined [as a real number, +∞ or −∞], then lim inf an = lim an =
lim sup an .

(ii) If lim inf an = lim sup an , then lim an is defined and lim an = lim inf an =
lim sup an .

Proof: Let us use the notation from above; i.e.,Let uN = glb{an | n > N }, vn =
lub{an | n > N }, u = lim uN = lim inf an and v = lim vN = lim sup an .

(i) Suppose lim an = +∞. Let M > 0 be a positive number. Then there is N ∈ N
so that if n > N then an > M . Then uN = glb{an | n > N } ≥ M . It follows
that m > N implies that um ≥ M . Thus, the sequence {uN } satisfies the
condition that lim uN = +∞ or lim inf an = +∞. Likewise, we can show that
lim sup an = +∞. We do the case that lim an = −∞ similarly.
Suppose that lim an = L ∈ R. Let ² > 0. There exists an N ∈ N so that
|an − L| < ² for n > N . Thus an < L + ² for n > N . This means that

vN = lub{an | n > N } ≤ L + ².

Also, if m > N then vm ≤ L + ² for all ² > 0, no matter how small. This means
that lim sup an ≤ L = lim an . Similarly, we can show that lim an ≤ lim inf an .
Since lim inf an ≤ lim sup an , these inequalities give us that

lim inf an = lim an = lim sup an .

(ii) If lim inf an = lim sup an = ±∞ it is easy to show that lim an = ±∞. Suppose
that lim inf an = lim sup an = L ∈ R. We need to show that lim an = L. Let
² > 0. Since L = lim vN there is an N0 ∈ N so that

|L − lub{an | n > N0 }| < ².

Thus, lub{an | n > N0 } < L + ², so

an < L + ² for all n > N0 .

Similarly, since L = lim uN there is an N1 ∈ N so that

|L − glb{an | n > N1 }| < ².

MATH 6101-090 Fall 2006

84 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

Thus, glb{an | n > N1 } > L − ², so

an > L − ² for all n > N1 .

These two conditions tell us that

L − ² < an < L + ² for n > max{N0 , N1 },

or, equivalently,
|an − L| < ² for n > max{N0 , N1 }.
This proves that lim an = L as needed.

This tells us that if {an } converges, then lim inf an = lim sup an , so for large N
the numbers lub{an | n > N } and glb{an | n > N } must be close together. This
means that all of the numbers in the set {an | n > N } must be close together. This
leads to the following definition.

Definition 6.6 A sequence {an } of real numbers is called a Cauchy sequence if

for each ² > 0 there is a number N ∈ N so that if m, n > N then |an − am | < ².

Lemma 6.3 Convergent sequences are Cauchy sequences.

Proof: Suppose that lim an = L. Note that

|an − am | = |an − L + L − aM | ≤ |an − L| + |am − L|.

Thus, given any ² > 0 there is an N ∈ N so that if k > N then |ak − L| < 2² . Thus,
if m, n > N we have
² ²
|an − am | ≤ |an − L| + |am − L| < + = ².
2 2

Thus, {an } is a Cauchy sequence.

Lemma 6.4 Cauchy sequences are bounded.

This leads us to the following theorem.

Theorem 6.12 A sequence is a convergent sequence if and only if it is a Cauchy

sequence.

MATH 6101-090 Fall 2006

6.5. SUBSEQUENCES 85

Proof: We have just proven half of this above. That means that we are left to show
that any Cauchy sequence must converge. To see this let {an } be a Cauchy sequence.
From the above lemma we know that it is bounded. That means then that we only
need to show that lim inf an = lim sup an .
Let ² > 0. Since {an } is a Cauchy sequence, there is an N ∈ N so that if m, n > N
then |an − am | < ². In particular, an < am + ² for all m, n > N . This shows that
am + ² is an upper bound for {an | n > N }. Thus vN = lub{an | n > N } ≤ am + ²
for m > N . Now, this shows that vN − ² is a lower bound for {am | m > N }, so that
vN − ² ≤ glb{am | m > N } = uN . Therefore

lim sup an ≤ vN ≤ uN + ² ≤ lim inf an + ².

Since this holds for all ² > 0, we have that lim sup an ≤ lim inf an and this is enough
to give us that the two quantities are equal.

6.5 Subsequences
So far we have learned the basic definitions of a sequence (a function from the nat-
ural numbers to the reals), the concept of convergence, and we have extended that
concept to one which does not pre-suppose the unknown limit of a sequence (Cauchy
sequence). Unfortunately, however, not all sequences converge. We will now intro-
duce some techniques for dealing with those sequences. The first is to change the
sequence into a convergent one (extract subsequences) and the second is to modify
our concept of limit as we did with lim sup and lim inf.

Definition 6.7 Let {an } be a sequence. When we extract from this sequence only
certain elements and drop the remaining ones we obtain a new sequences consisting
of an infinite subset of the original sequence. That sequence is called a subsequence
and denoted by {ank }.

One can extract infinitely many subsequences from any given sequence.

Example 6.10 Take the sequence {(−1)n }, which we know does not converge. Ex-
tract every other member, starting with the first. Does this sequence converge? What
if we extract every other member, starting with the second. What do you get in this
case?

Example 6.11 Take the sequence {1/n}. Extract three different subsequences of
your choice. Do these subsequences converge? Is so, to what limit?

The last example is an indication of a general result.

Theorem 6.13 (i) If {an } is a convergent sequence, then every subsequence of that
sequence converges to the same limit.

MATH 6101-090 Fall 2006

86 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

(ii) If is a sequence such that every possible subsequence extracted from that sequences
converges to the same limit, then the original sequence also converges to that
limit.

The next statement is probably one on the most fundamental results of basic real
analysis, and generalizes the above proposition. It also explains why subsequences
can be useful, even if the original sequence does not converge.

Theorem 6.14 (Bolzano-Weierstrass) Let {an } be a sequence of real numbers

that is bounded. Then there exists a subsequence {ank } that converges.

Proof: Since the sequence is bounded, there exists a number M such that |an | < M
for all n. Then either [−M, 0] or [0, M ] contains infinitely many elements of the
sequence. Say that [0, M ] does. Choose one of them, and call it an1 . Now, either
[0, M/2] or [M/2, M ] contains infinitely many elements of the (original) sequence. Say
it is [0, M/2]. Choose one of those elements, and call it an2 . Again, either [0, M/4] or
[M/4, M/2] contains infinitely many elements of the (original) sequence. This time,
say it is [M/4, M/2]. Pick one of those elements and call it an3 .
Keep on going in this way, halving each interval from the previous step at the
next step, and choosing one element from that new interval. Here is what we get:

• |an1 − an2 | < M , because both are in [0, M ]

• |an2 − an3 | < M/2, because both are in [0, M/2]

• |an3 − an4 | < M/4, because both are in [M/2, M/4]

and in general, we see that

|ank − ank+1 | < M/2k−1 ,

because both are in an interval of length M/2k−1 .

So, this proves that consecutive elements of this subsequence are close together.
That is not enough, however, to say that the sequence is Cauchy, since for that not
only consecutive elements must be close together, but all elements must get close to
each other eventually.
So take any ² > 0, and pick an integer N such that for any k, m > N (with m > k)

MATH 6101-090 Fall 2006

6.6. SERIES 87

we have:

|ank − anm | = |(ank − ank+1 ) + (ank+1 − ank+2 ) + · · · + (anm−1 − anm )|

≤ |ank − ank+1 | + |ank+1 − ank+2 | + · · · + |anm−1 − anm |
µ ¶
1 1 1 1
= M + + + · · · + m−2
2k−1 2k 2k+1 2
µ ¶
M 1 1 1
= k−1 1 + + 2 + · · · + m−k−1
2 2 2 2
µ ¶
M 1 1
≤ k−1 1 + + 2 + . . .
2 2 2
∞ µ ¶j
M X 1 2M M
= k−1 = k−1 = k−2
2 j=1
2 2 2

In order for this to be less than ² for m, k > N we would need to take N so that

M
< ²
2N −2
M
2N −2 >
²
(N − 2) ln 2 > ln M − ln ²
ln M − ln ²
N > 2+ .
ln 2

This proves what we wanted.

6.6 Series
Now we will investigate what may happen when we add all terms of a sequence
together to form what will be called an infinite series. The old Greeks already won-
dered about this, and actually did not have the tools to quite understand it. This is
illustrated by the old tale of Achilles and the Tortoise.

Zeno’s Paradox (Achilles and the Tortoise)

Achilles, a fast runner, was asked to race against a tortoise. Achilles can run 10
yards per second, the tortoise only 5 yards per second. The track is 100 yards long.
Achilles, being a fair sportsman, gives the tortoise 10 yard advantage. Who will win?
Both start running, with the tortoise being 10 yards ahead. After one second,
Achilles has reached the spot where the tortoise started. The tortoise, in turn, has
run 5 yards. Achilles runs again and reaches the spot the tortoise has just been. The
tortoise, in turn, has run 2.5 yards. Achilles runs again to the spot where the tortoise
has just been. The tortoise, in turn, has run another 1.25 yards ahead.

MATH 6101-090 Fall 2006

88 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

This continuous for a while, but whenever Achilles manages to reach the spot
where the tortoise has just been a second ago, the tortoise has again covered a little
bit of distance, and is still ahead of Achilles. Hence, as hard as he tries, Achilles only
manages to cut the remaining distance in half each time, implying, of course, that
Achilles can actually never reach the tortoise. So, the tortoise wins the race, which
does not make Achilles very happy at all. What is wrong with this line of thinking?
Let us look at the difference between Achilles and the tortoise:

Time Difference
t=0 10 yards
t=1 5 = 10/2 yards
t = 1 + 21 2.5 = 10/4 yards
t = 1 + 12 + 14 1.25 = 10/8 yards
t = 1 + 12 + 14 + 1
8
0.625 = 10/16 yards

and so on. In general we have:

Time Difference
10
t = 1 + 12 + 1
22
+ 1
23
+ ··· + 1
2n
yards
2n
Now we want to take the limit as n goes to infinity to find out when the distance
between Achilles and the tortoise is zero. But that involves adding infinitely many
numbers in the above expression for the time, and we (the Greeks and Zeno) don’t
know how to do that. However, if we define
1 1 1 1
sn = 1 + + 2 + 3 + ··· + n
2 2 2 2
then, dividing by 2 and subtracting the two expressions:
1 1
sn − sn = 1 − n+1
2 2
or equivalently, solving for sn :
µ ¶
1
sn = 2 1 − .
2n+1

Now sn is a simple sequence, for which we know how to take limits. In fact, from the
last expression it is clear that lim sn = 2 as n approaches infinity.
Hence, we have - mathematically correctly - computed that Achilles reaches the
tortoise after exactly 2 seconds, and then, of course passes it and wins the race. A
much simpler calculation not involving infinitely many numbers gives the same result:

• Achilles runs 10 yards per second, so he covers 20 yards in 2 seconds.

MATH 6101-090 Fall 2006

6.6. SERIES 89

• The tortoise runs 5 yards per second, and has an advantage of 10 yards. So, it
also reaches the 20 yard mark after 2 seconds.

• Therefore, both are even after 2 seconds.

Of course, Achilles will finish the race after 10 seconds, while the tortoise needs 18
seconds to finish, and Achilles will clearly win.
The problem with Zeno’s paradox is that Zeno was uncomfortable with adding
infinitely many numbers. In fact, his basic argument was that if you add infinitely
many numbers, then — no matter what those numbers are — you must get infinity. If
that was true, it would take Achilles infinitely long to reach the tortoise, and he would
loose the race. However, reducing the infinite addition to the limit of a sequence, we
have seen that this argument is false.
One reason for looking so carefully at sequences is that it allows us to to quickly
obtain the properties of infinite series.
We know (at least theoretically) how to deal with finite sums of real numbers.
n
X
ak = am + am+1 + . . . an .
k=m

More interest in mathematics though tends to lie in the area of infinite series:
∞
X
ak = am + am+1 + am+2 + . . . .
k=m

P∞
What do we mean by this infinite series, k=m ak ? Define the nth partial sum,
Sn by
n
X
Sn = am + am+1 + . . . an = ak .
k=m

∞
This now
P∞gives us a sequence, the sequence of partial sums, {Sk }k=m . The infinite
series k=m ak is said to converge providedPthe sequence of partial sums converges
to a real number S. In this case we define ∞
n=m an = S. Thus

∞
à n
!
X X
an = S means lim Sn = S or lim ak ) = S.
n→∞ n→∞
n=m k=m

If a series does not converge we say that it diverges. We can then say that a series
diverges to +∞ if lim sn = +∞ Por that it diverges to −∞ if lim sn = −∞. Some texts
will indicate that the symbol ∞ Pa∞
n=m n has no meaning unless the series converges
or diverges to +∞ or −∞. Thus, n=0 (−1)n will have no meaning.

MATH 6101-090 Fall 2006

90 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

X∞
1 1 1 1 1
Example 6.12 n
=1+ + + + + . . . is an infinite series. The sequence
n=0
2 2 4 8 16
of partial sums looks like:

3 7 15
S0 = 1, S1 = , S2 = , S3 = , . . .
2 4 8
We saw above that this sequence converges to 2, so

X∞
1
= lim Sn = 2.
n=0
2n

Example 6.13 The harmonic series is

X∞
1 1 1 1
= 1 + + + + ....
n=1
n 2 3 4

The first few terms in the sequence of partial sums are:

3 11 25 137
S1 = 1, S2 = , S3 = , S4 = , S5 = ,
2 6 12 60
49 363 761 7129 7381
S 6 = , S7 = , S8 = , S9 = , S10 =
20 140 280 2520 2520

This series diverges to +∞. To prove this we need to estimate the nth term in the
sequence of partial sums. The nth partial sum for this series is

1 1 1 1
SN = 1 + + + + ··· + .
2 3 4 N
Now consider the following subsequence extracted from the sequence of partial sums:

S1 = 1
1
S2 = 1 +
2 µ ¶
1 1 1
S4 = 1 + + +
2 3 4
µ ¶
1 1 1 1 1 2
≥ 1+ + + =1+ + =1+
2 4 4 2 2 2
µ ¶ µ ¶
1 1 1 1 1 1 1
S8 = 1 + + + + + + +
2 3 4 5 6 7 8
µ ¶ µ ¶
1 1 1 1 1 1 1 1 1 1 3
≥ 1+ + + + + + + =1+ + + =1+
2 4 4 8 8 8 8 2 2 2 2

MATH 6101-090 Fall 2006

6.6. SERIES 91

In general, by induction we have that that

k
S2k ≥ 1 +
2
for all k. Hence, the subsequence {S2k } extracted from the sequence of partial sums
{SN } is unbounded. But then the sequence {SN } cannot converge either, and must,
in fact, diverge to infinity.
P
If the terms an of an infinite series an are all nonnegative, then P the partial
sums {Sn } form a nondecreasing sequence, soP by Theorems 6.9 and 6.10 an either
converges or diverges to +∞.PIn particular, |an | is meaningful P
for any sequences
{an } whatsoever. The series an is said to converge absolutely if |an | converges.
P
Example 6.14 A series of the form ∞ n
n=0 ar for constants a and r is called a geo-
metric series. For r 6= 1 the partial sums are given by
N
X
k 1 − rn+1
SN = ar = a .
k=0
1−r

Taking the limit as N goes to infinity, gives us that

∞
( a
X if |r| < 1
arn = 1 − r
n=0 +∞ if a 6= 0 and |r >≥ 1

Example 6.15 [p-Series]For a positive number p

X∞
1
converges if and only if p > 1.
n=1
np

The exact value of this series for p > 1 is extremely difficult to determine. A few are
known. The first of these below is due to Euler.
X∞
1 π2
=
n=1
n2 6
X∞
1 π4
=
n=1
n4 90

If p > 1 then the sum of the series is ζ(p), i.e., the Riemann zeta function evaluated
at p. There If p is an even integer then there are formulas like the above, but there
are no elegant formulas for p an odd integer.

A series converges conditionally, if it converges, but not absolutely.

MATH 6101-090 Fall 2006

92 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

P
Example 6.16 1. Does the series ∞ n
n=0 (−1) converge absolutely, conditionally,
or not at all?
P
2. Does the series ∞ 1 n
n=0 ( 2 ) converge absolutely, conditionally, or not at all?
∞
X (−1)n+1
3. Does the series converge absolutely, conditionally, or not at all (this
n=1
n
series is called alternating harmonic series)?
Conditionally convergent sequences are rather difficult with which to work. Several
operations that one would expect to be true do not hold for such series. The perhaps
most striking example is the associative law. Since a + b = b + a for any two real
numbers a and b, positive or negative, one would expect also that changing the order
of summation in a series should have little effect on the outcome. Not true.
P
Theorem 6.15 (Order of Summation) (i) Let an be an absolutely convergent
series. Then any rearrangement of terms in that series results in a new series
that is also absolutely convergent to the same limit.
(ii) Let be a conditionally convergent series. Then, for any real number c there is a
rearrangement of the series such that the new resulting series will converge to
c.

This will be proved later. One sees, however, that conditionally convergent series
probably contain a few surprises. Absolutely convergent series, however, behave just
as one would expect.
P P
Theorem 6.16 (Algebra on Series) Let an and bn be two absolutely conver-
gent series. Then
(i) The sum of the two series is again absolutely convergent. Its limit is the sum of
the limit of the two series.
(ii) The difference of the two series is again absolutely convergent. Its limit is the
difference of the limit of the two series.
(iii) The product of the two series is again absolutely convergent. Its limit is the
product of the limit of the two series.
P P
The Cauchy product of two series an and bn of real is defined as follows. The
Cauchy product is
à ∞ ! à ∞ ! à ∞ ! n
X X X X
an · bn = cn where cn = ak bn−k
n=m n=m n=m k=0

for n = 0, 1, 2, . . . .

MATH 6101-090 Fall 2006

6.7. CONVERGENCE TESTS 93

6.7 Convergence Tests

P
Definition 6.8 We say that a series an satisfies the Cauchy criterion if its se-
quence of partial sums is a Cauchy sequence.

This means that for each ² > 0 there exists a number N such that if m, n > N
then |Sn − Sm | < ². Nothing is lost in this definition if we impose the restriction
n ≥ m. Moreover, it is only a notational matter to work with m − 1 where m ≤ n
instead of m where m < n. That means that the definition is equivalent to for each
² > 0 there exists a number N such that if n P≥ m > N then |Sn − Sm−1 | < ².
n
The reason for doing this is that Sn − Sm−1 = k=m ak . Then this condition can be
rewritten
Pn as for each ² > 0 there exists a number N such that if n ≥ m > N then
| k=m | < ².

Theorem 6.17 A series converges if and only if it satisfies the Cauchy criterion.
P
Corollary 6.1 If a series an converges, then lim an = 0.

It is often easier to prove that a limit exists or that a series converges than it is
to determine its exact value. As an example consider the following.
P
Theorem 6.18 (Comparison Test) Let an be a series where an ≥ 0 for all n.
P P
(i) If an converges and |bn | ≤ an for all n, then bn converges.
P P
(ii) If an = +∞ and bn ≥ an for all n, then bn = +∞.

Proof: (i) For n ≥ m we have

¯ n ¯
¯X ¯ X n n
X
¯ ¯
¯ bk¯ ≤ |bk | ≤ ak
¯ ¯
k=m k=m k=m
P
where the first inequality follows from the Triangle Inequality. Since an
converges,
P it satisfies the Cauchy criterion. It then follows from the above that
bn also satisfies the Cauchy criterion and hence it also converges.
P P
(ii) Let {Sn } and {Tn } be the sequences of partial sums for an and bn , respec-
tively. Since bn ≥ an for all n we then clearly have that
P Tn ≥ Sn for all n. Since
lim Sn = +∞, we conclude that lim Tn = +∞, and bn = +∞.

Corollary 6.2 Absolutely convergent series are convergent.

P P
Proof: Suppose that an is absolutely convergent. This Pmeans that bn con-
verges where bn = |an | for all n. Then |an | ≤ bn , so that an converges by the
Comparison Test.

MATH 6101-090 Fall 2006

94 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

P P
Theorem 6.19 (Limit Comparison Test) Suppose an and bn are twoP
infi-
nite series. Suppose also that r = Plim |an /bn | exists, and 0 < r < +∞. Then an
converges absolutely if and only if bn converges absolutely.

Proof: Since r = lim |an /bn | exists, and r is between 0 and +∞, there exist con-
stants c and C, 0 < c < C < +∞ such that for some N > 1 we have that if n > N
¯ ¯
¯ an ¯
c < ¯¯ ¯¯ < C.
bn
P
AssumeP that an converges absolutely. For n > N we have that c|bn | < |an |.
Therefore, bn converges
P absolutely by the Comparison Test.
Now assume that bn converges absolutely. From
Pthe above inequality we have
that |an | < C|bn | for n > N . But since the series P
C bn also converges absolutely,
we can use again the Comparison Test to see that an must converge absolutely.

Theorem 6.20 (Cauchy Condensation Test) P Suppose {an } is a decreasing se-

quence of positive terms. Then the series
P an converges if and only if the series
k
2 a2k converges.
∞
X
Proof: Assume that an converges. Since {an } is a decreasing sequence, we have
n=1
that
2k−1 a2k = a2k + a2k + a2k + ... + a2k
2 k
X
≤ a2k−1 +1 + a2k−1 +2 + · · · + a2k−1 +2k−2 + a2k = am .
m=2k−1 +1

Therefore, we have that

 
N
X N
X 2k
X 2
X
N

k−1
2 a2k ≤  am  = am .
k=1 k=1 m=2k−1 +1 m=2

Now the partial sums on the right are bounded, by assumption. Hence the partial
sums on the left are also bounded. Since all terms are positive, the partial sums
now form an increasing sequence that is bounded above, hence it must converge.
Multiplying
PN k the left sequence by 2 will not change convergence, and hence the series
k=1 2 a2k converges. P
Now, assume that N k
k=1 2 a2k converges: We have

2 N
X
am = a2k−1 +1 + a2k−1 +2 + · · · + a2k−1 +2k−2 + a2k
m=2k−1 +1

≤ a2k−1 + a2k−1 + a2k−1 + ... + a2k−1 = 2k−1 a2k−1

MATH 6101-090 Fall 2006

6.7. CONVERGENCE TESTS 95

Therefore, similar to above, we get:

 
2N
X XN 2
X
k N
X
am =  am  ≤ 2k−1 a2k .
m=2 k=1 m=2k−1 +1 k=1

Now the sequence of partial sums on the right is bounded, by assumption. There-
fore, the left side forms an increasing sequence that is bounded above, and therefore
must converge.
Corollary 6.3 For a positive number p
X∞
1
p
converges if and only if p > 1.
n=1
n
½ ¾
1
Proof: If p < 0 then the sequence diverges to infinity. Hence, the series
np
diverges by the Divergence Test.
If p > 0 then consider the series
∞
X ∞
X X ∞
1
2n a2n = 2n n p
= (21−p )n .
n=1 n=1
(2 ) n=1

This right-hand side is a geometric series. Thus, we know that

• if 0 < p ≤ 1 then 21−p ≥ 1, hence the right-hand series diverges;
• if p > 1 then 21−p < 1 and the right-hand series converges,
Now the result follows from the Cauchy Condensation Test .
P
TheoremP 6.21 (Root Test) Let an be a series and let α = lim sup |an |1/n . The
series an
(i) converges absolutely if α < 1,
(ii) diverges if α > 1.
(iii) Otherwise α = 1 and the test gives no information.
Although this Root Test is more difficult to apply, it is better than the Ratio Test
in the following sense. There are series for which the Ratio Test give no information,
yet the Root Test will be conclusive. We will use the Root Test to prove the Ratio
Test, but you cannot use the Ratio Test to prove the Root Test. It is important to
remember that when the Root Test gives 1 as the answer for the lim sup, then no
conclusion at all is possible.
The use of the lim sup rather than the regular limit has the advantage that we
do not have to be concerned with the existence of a limit. On the other hand, if
the regular limit exists, it is the same as the lim sup, so that we are not giving up
anything using the lim sup.

MATH 6101-090 Fall 2006

96 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

Proof:

(i) Suppose that α < 1. Then choose an ² > 0 so that α + ² < 1. Then by the
definition of the limit superior there is a natural number N such that

α − ² < lub{|an |1/n | n > N }, α + ².

In particular, we have |an |1/n < α + ² for n > N , so

|an | < (α + ²)n for n > N.

P∞ n
Since 0 < α + ² < 1, the geometric
P∞ series n=N +1 (α + ²) converges. P
Thus, the
Comparison Test shows that n=N +1 an converges. This means that an also
converges.
(ii) If α > 1, then there is a subsequence of |an |1/n that has limit α > 1. That means
that |an | > 1 for infinitely many choices
P of n. In particular, the sequence {an }
cannot converge to 0, so the series an cannot converge.
P1 P 1
(iii) For the series n
and for the series P n2
, α turns out to be 1. Since the
1
harmonic series diverges and the series n2
converges, the equality α = 1
cannot guarantee either convergence or divergence of the series.

P
Theorem 6.22 (Ratio Test) A series an of nonzero series
(i) converges absolutely if lim sup |an+1 /an | < 1,
(ii) diverges is lim inf |an+1 /an | > 1.
(iii) Otherwise lim inf |an+1 /an | ≤ 1 ≤ lim sup |an+1 /an | and the test gives no infor-
mation.

Proof:
(i) Suppose that lim sup |an+1 /an | < 1. Then choose an ² > 0 so that |an+1 /an |+² < 1.
Then by the definition of the limit superior there is a natural number N such that
for n > N ¯ ¯
¯ an+1 ¯
¯ ¯
¯ an ¯ < 1 − ².
Multiplying both sides by |an | we get

|an+1 | < (1 − ²)|an | for n > N.

Therefore, we also have

|an+2 | < (1 − ²)|an+1 | < (1 − ²)2 |an | for n > N.

MATH 6101-090 Fall 2006

6.7. CONVERGENCE TESTS 97

Repeating this procedure, we get that

|ak | < (1 − ²)k−N |aN | for k > N.

ThesePterms form a convergent geometric series. Thus, the Comparison Test shows
that an also converges.
The other two parts are proven in much the same fashion as the previous theorem.

The next lemma is used to prove Abel’s Convergence Test. It is computational in

nature.

Lemma 6.5 (Summation by Parts) Consider the two sequences {an } and {bn }.
n
X
Let Sn = an be the n-th partial sum. Then for any 0 ≤ m ≤ n we have
k=1

n
X n+1
X
aj bj = [Sn bn − Sm−1 bm ] + Sj (bj − bj+1 ).
j=m j=m

Proof: Just be careful with the subscripts:

n
X n
X
aj bj = (Sj − Sj−1 )bj
j=m j=m
Xn n
X
= Sj b j − Sj−1 bj
j=m j=m
n
X n−1
X
= Sj b j − Sj bj+1
j=m j=m−1
n−1
X
= Sj (bj − bj+1 + (Sn bn − Sm−1 bm )
j=m

P
Theorem 6.23 (Abel’s Test) Consider the series an bn . Suppose that
P
(i) the partial sums SN = N
n=1 an form a bounded sequence,

(ii) the sequence {bn } is decreasing,

(iii) lim bn = 0.
P
Then the series an bn converges.

MATH 6101-090 Fall 2006

98 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

This test is rather sophisticated. Its main application is to prove the Alternating
Series test, but one can sometimes use it for other series as well, if the more obvious
tests do not work.

Proof: First, let’s assume that the partial sums SN are bounded by K. Next,
since the sequence {bn } converges to zero, we can choose an integer N such that
|bn | < ²/2K. Using the Summation by Parts lemma, we then have:
¯ n ¯ ¯ ¯
¯X ¯ ¯ n+1
X ¯
¯ ¯ ¯ ¯
¯ aj bj ¯ = ¯[Sn bn − Sm−1 bm ] + Sj (bj − bj+1 )¯
¯ ¯ ¯ ¯
j=m j=m
n−1
X
≤ K|bn | + K|bm | + K |bj − bj+1 |.
j=m

But the sequence {bn } is decreasing to zero, so in particular, all terms must be
positive, and all absolute values inside the summation above are superfluous. But
then the sum is a telescoping sum. Therefore, all that remains is the first and last
term, and we have:
¯ ¯
¯Xn ¯
¯ ¯
¯ aj bj ¯ ≤ K(bm + bn + bm − bn ) = 2Kbm .
¯ ¯
j=m

But by our choice of N, this is less than ² if we choose n and m larger than the
predetermined N .

Theorem 6.24 (Alternating Series Test)P If a1 ≥ a2 ≥ · · · ≥ an ≥ . . . 0 and {an }

converges to zero, then the alternating series (−1)n an converges.

This test does not prove absolute convergence. In fact, when checking for absolute
convergence the term ’alternating series’ is meaningless. It is important that the
series truly alternates, that is each positive term is followed by a negative one, and
visa versa. If that is not the case, the alternating series test does not apply (while
Abel’s Test may still work).
P
Proof: Let an = (−1)n . Then the formal sum ∞ n=1 an has bounded partial sums
although the sum does not converge. Then, with the given choice of {bn } Abel’s test
applies directly, showing that the series converges.

6.7.1 Riemann Series Theorem

This is Theorem 6.15. We will prove it here. Let me restate it in two parts.

MATH 6101-090 Fall 2006

6.7. CONVERGENCE TESTS 99

We are used to using the law of commutativity when adding real numbers. It does
not matter in which order we calculate a sum
n
X
ak
k=1

we always end up with the same number. We can express this in a more formal way
by introducing the notion of rearrangement. A rearrangement is a one-to-one, onto
function
π : {1, . . . , n} → {1, . . . , n}
As a consequence every number k ∈ {1, . . . , n} can be written as k = π(`) for some ` ∈
{1, . . . , n}. Often, such a function is called a permutation of the numbers {1, . . . , n}.
For example take the set {1, 2, 3} and consider the map

π1 (1) = 2, π1 (2) = 3, π1 (3) = 1.

This is a rearrangement or permutation of the the three numbers.

Question: In how many ways can we rearrange the numbers {1, . . . , n}?
Returning to the series we can express the fact that it does not matter in which
order we add the numbers by the equation
n
X n
X
aπ(k) = ak
k=1 k=1

What do we mean by a rearrangement of a series? If we denote the natural

numbers by N, we have to be precise what we mean by a rearrangement of N. We
have to specify the rearrangement π for infinitely many numbers. Again, we must
have that π is one-to-one and onto. This means that every element k ∈ N can be
written as π(`) for some ` ∈ N.
With the help of such a map π we can rearrange series simply by writing
X
aπ(k) .
k

It is natural to expect, that for any convergent series it doe not really matter how we
sum up this series, i.e., X X
aπ(k) = ak .
k k

The Reimann Series Theorem tells us first that it is not true in general, but it is
true under certain reasonable conditions.
P
Theorem 6.25 (Riemann Series Theorem) Let kP ak be an absolutely
P conver-
gent series. then for any rearrangement π we have that k aπ(k) = k ak .

MATH 6101-090 Fall 2006

100 CHAPTER 6. SEQUENCES AND SERIES OF REAL NUMBERS

Thus, absolutely convergent series really do behave like finite sums when it comes to
changing the order of summation.

Proof: We will do this by looking at the partial sums. The partial sum for the
rearranged series we will call
Xn
Rn = aπ(k)
k=1

whereas n
X
Sn = ak .
k=1

Since we are only interested in absolute convergence, we may assume that the numbers
ak > 0.
Define the number
D(n) = max |π(k) − k|.
1≤k≤n

This number describes the range of the rearrangement, in particular we have that

π(k) ≤ k + D(n)

for all 1 ≤ k ≤ n. Now we have that

n n+D(n)
X X
Rn = aπ(k) ≤ ak .
k=1 k=1

In the first expression the n indices π(k) for k = 1, . . . , n are distinct numbers, each
somewhere between 1 and n + D(n). Here is one place that we use the fact that
the function π is one-to-one. These indices are also included on the right side of the
equation but there may be more and hence the inequality sign.
In particular we realize that
n+D(n)
X
Rn ≤ Sn + ak .
k=n+1
P
Since the series k ak is absolutely convergent we have that
∞
X
Rn ≤ Sn + ak ,
k=n+1
P
and as n → ∞ Sn has a limit L and ∞ n+1 ak → 0. Hence Rn is a bounded sequence
and since it is increasing it has a limit R which satisfies R ≤ L.
Now we have to show that the converse is true. Notice that we have not used the
assumption that π is onto, which means that we do not miss any terms. Pick any n

MATH 6101-090 Fall 2006

6.7. CONVERGENCE TESTS 101

and notice that there exists a number E(n) so that the numbers a1 , . . . , an are among
the numbers aπ(1) , . . . , aπ(E(n)) . Thus

E(n)
X
Sn ≤ aπ(k) .
k=1

As n → ∞ E(n) → ∞ and since both sides converge, we have L ≤ R and our theorem
is proven.

The following is one of the most scandalous results in series.

P
Theorem 6.26 Let k ak be a conditionally convergent series.PFor any given num-
ber L there exists a rearrangement π of N, so that the series k aπ(k) converges to
L.

Thus we see that our expectations fail in a spectacular way and conditionally conver-
gent sequences continue to be promiscuous.

Proof: We assume that L is positive. We have to find the desired rearrangement.

The procedure goes as follows. Make two boxes, in one we put all the positive numbers
ak in an ordered fashion and in the other we put all the negative numbers ak also
in an ordered fashion. Since the series is conditionally convergent we have that the
series consisting of the numbers in the positive box diverges to +∞ whereas the series
consisting of the numbers in the negative box has to diverge towards −∞.
Now start picking numbers from the positive box until the sum overshoots L for
the first time. Notice that this is possible since the series in the positive box diverges.
Then we start adding the numbers from the negative box until we undershoot the
number L for the first time. Notice again, that this must happen since the series
from the negative box diverges. Now we keep doing this, and produce in this fashion
a sequence of partial sums of a rearranged series that oscillates about the value L. it
remains to see that this sequence converges to L, but this follows immediately from
the fact that the numbers ak → 0 as k → ∞.

MATH 6101-090 Fall 2006