2 Sequences and Series

In this chapter we will study two related questions. Given an innite collec-
tion X of numbers, which can be taken to be rational, real or complex, the
rst question is to know if there is a limit point to which the elements of X
congregate (converge). The second question is to know if one can add up all
the numbers in X.
Even though we will be mainly thinking of Q, R or C, many things will
go through for numbers in any eld K with an absolute value | . | satisfying
|0| = 0, |x| > 0 if x = 0, |xy| = |x| |y|, and the triangle inequality |x + y|
|x| +|y|.
A necessary thing to hold for convergence is the Cauchy condition, which
is also sucient if we work with R or C, but not Q. What we mean by this
remark is that a Cauchy sequence of rational numbers will have a limit L
in R, but L need not be rational. One can enlarge Q to have this criterion
work, i.e., adjoin to Q limits of Cauchy sequences of rational numbers, in
which case one gets yet another construction of R. This aspect is discussed
in section 2.3, which one can skip at a rst reading.
The last three sections deal with the second question, and we will derive
various tests for absolute convergence. Sometimes, however, a series might
converge though not absolutely. A basic example of this phenomenon is given
by the Leibniz series
1
1
3
+
1
5

1
7
+
1
9

1
11
+ . . . =

4
.
But the series on the left is not absolutely convergent, i.e., the series 1 +
1
3
+
1
5
+ . . . diverges.
2.1 Convergence of sequences
Let K denote either Q or R or C. When we say a number, or a scalar, we
will mean an element of K. But by a positive number we will mean a positive
real number. (One could restrict to positive rational numbers instead, and it
will work just as well to use this subset.)
By a sequence, or more properly an innite sequence, we will mean a
collection of numbers
a
1
, a
2
, a
3
, . . . , a
n
, a
n+1
, . . . ,
1
which is indexed by the set N of natural numbers. We will often denote it
simply as {a
n
}.
A simple example to keep in mind is given by a
n
=
1
n
, which appears to
decrease towards zero as n gets larger and larger. In this case we would like
to have 0 declared as the limit of the sequence.
It should be noted that if we are given a collection of numbers indexed by
any countably innite set, we can reindex it by N and apply all our results
below to it. For example, given M > 0 and {c
n
| n > M}, we can renumber
it by putting a
n
= c
n+M
and get a sequence indexed by N.
We will say that a sequence {a
n
} converges to a limit L in K i for
every positive number there exists an index N such that
(2.1.1) n > N = |L a
n
| < .
It is very important to note that we have to check the condition above for
every positive . Clearly, if (2.2.1) holds for N, it also holds for any N

> N.
When {a
n
} does converge to a limit L, we will write
a
n
L or lim
n
a
n
= L.
When we write n , we do not necessarily mean that there is a number
called , which we have certainly not dened; we mean only that n becomes
arbitrarily large. In other words we want to know what happens to a
n
when
n is larger than any number one can think of.
Let us quickly prove that lim
n
1
n
is 0, as expected. For any positive number
, we can nd a rational number N such that 0 <
1
N
< . Indeed, by the
property R4 of R (see section 1.0), we can nd a rational number q with
0 < q < , and since q s a positive rational number, we can nd an N > 0
such that
1
N
< q. So for every n > N, we have
|0
1
n
| =
1
n
<
1
N
< ,
proving that
1
n
does converge to 0.
When a sequence does not converge to any number L, we will say that it
diverges. A simple example is given by the sequence {1, 1, 1, 1, 1, 1, . . .}.
Given any < 1, we will be unable to nd an N for which (2.1.1) holds for
any L. This is so because for any N {a
N+1
, a
N+2
} = {1, 1}, and (for any
2
L) at least one of them will be at a distance > 1 from L; to see this use the
triangle inequality |L a
N+1
| |L a
N+2
| + |a
N+2
a
N+1
|, together with
|a
N+2
a
N+1
| = 2.
Any sequence which converges (to some limit) will be called a convergent
sequence, and one that does not converge will be called a divergent sequence.
Proposition 2.1.2
(i) If {a
n
} is a convergent sequence with limit L, then for any scalar c, the
sequence {ca
n
} is convergent with limit cL;
(ii) If {a
n
}, {b
n
} are convergent sequences with respective limits L
1
, L
2
,
then their sum {a
n
+b
n
} and their product {a
n
b
n
} are convergent with
respective limits L
1
+ L
2
and L
1
L
2
.
The proof is a simple application of the properties of absolute values. To
illustrate, let us prove the assertion about the sum sequence {a
n
+b
n
}. Pick
any > 0. Since a
n
L
1
and b
n
L
2
, we can nd N > 0 such that
n > N = |L
1
a
n
| <

2
, |L
2
b
n
| <

2
.
Using the triangle inequality we then have, for all n > N,
|(L
1
+ L
2
) (a
n
+ b
n
)| |L
1
a
n
| +|L
2
b
n
| <

2
+

2
= .
Hence a
n
+ b
n
L
1
+ L
2
.
Corollary 2.1.3 Let {z
n
= x
n
+ iy
n
} be a sequence of complex numbers,
with x
n
, y
n
real for each n. Then {z
n
} converges i the real sequences {x
n
}
and {y
n
} are both convergent.
Proof. Suppose {x
n
}, {y
n
} are both convergent, with respective limits
u, v. We claim that {z
n
} then converges to w = u + iv. Indeed, by the
Proposition above, {iy
n
} is convergent with limit iv, and so is {x
n
+ iy
n
},
with limit w. Conversely, suppose that {z
n
} converges, say to w. We may
write w as u + iv, with u, v real. For any complex number z = x + iy, |x|
and |y| are both bounded above by

x
2
+ y
2
, i.e., by |z|. Since w z
n
=
(u x
n
) + i(v y
n
), we get
|u x
n
| |w z
n
| and |v y
n
| |w z
n
|.
3
For any > 0, pick N > 0 such that for all n > N, |w z
n
| is < . Then
we also have |u x
n
| < and |v y
n
| is < for all n > N, establishing the
convergence of {x
n
} and {y
n
} with respective limits u and v.
Thanks to this Corollary, the study of sequences of complex numbers can
be reduced to that of real numbers.
A sequence {a
n
} is bounded i there is a positive number M such that
|a
n
| M for all n.
Lemma 2.1.4 If a sequence {a
n
} converges, then it must be bounded.
Proof Let L be the limit of {a
n
}. Fix any positive number . Then by
denition, there exists an N > 0 such that for all n > N, |La
n
| is < . By
the triangle inequality,
|a
n
| |a
n
L| +|L|.
So if we set
M = max{a
1
, a
2
, . . . , a
N
, L + },
then every |a
n
| is bounded by M.
The converse of this Lemma is not true: Remember that the sequence
{1, 1, 1, 1, . . .} has not limit, but it is bounded in absolute value by 1.
For any real number x and consider the power sequence {x
n
}. When x = 1
(resp. 0), we get the stationary sequence {1, 1, 1, . . .} (resp. {0, 0, 0, . . .}),
which is obviously convergent with limit 1 (resp. 0). When x = 1, we get
{1, 1, 1, . . .}, which is divergent, as we saw above. Applying the following
Lemma we see that this sequence is unbounded when |x| > 1, hence divergent,
and that it converges to zero when |x| < 1.
Lemma 2.1.5 Let y be a positive real number. Then {y
n
} is unbounded if
y > 1, while
lim
n
y
n
= 0 if y < 1.
Proof of Lemma. Suppose y > 1. Write y = 1 + t with t > 0. Then by
the binomial theorem (which can be proved by induction),
y
n
= (1 + t)
n
=
n

k=1
(
n
k
)
t
k
,
4
which is 1 + nt. Since 1 + nt is unbounded, i.e, larger than any number
for a big enough n, y
n
is also unbounded.
Now let y < 1. Then y
1
is > 1 and hence {y
n
} is unbounded. This
implies that, for any > 0, y
n
is < for large enough n. Hence the sequence
y
n
converges to 0.
A sequence of real (or rational) numbers a
n
is said to be monotone in-
creasing, resp. monotone decreasing, i for all n, a
n
a
n+1
, resp. a
n
a
n+1
;
it is said to be simply monotone if it is either monotone increasing or mono-
tone decreasing.
Theorem A If {a
n
} is a bounded, monotone sequence of real numbers,
then it converges in R, in fact to the supremum, resp. inmum, when {a
n
}
is increasing, resp. decreasing.
Proof. Suppose {a
n
} is monotone increasing. Since R is a complete
ordered eld, every bounded subset has a least upper bound. Since {a
n
} is
bounded, we may look at its supremum M, say. We claim that a
n
M.
Suppose not. Then there exists an > 0 such that for every N > 0, there
exists an n > N such that |M a
n
| . Since the sequence is monotone
increasing, we then have
|a
n
| < M

2
, n 1.
Then M

2
is an upper bound for {a
n
}, and so M could not be the least
upper bound. Contradiction! So M must be the limit of the sequence.
Now suppose {a
n
} is monotone decreasing. Then the sequence {a
n
} is
monotone increasing, and it is bounded because {a
n
} is. So {a
n
} converges;
then so does {a
n
}.
Here is an example. Dene a sequence {a
n
} by putting
a
n
= 1 +
1
1!
+
1
2!
+ . . . +
1
(n 1)!
.
It is not hard to see that this sequence is bounded. Try to give a proof.
(In fact one can show that it is bounded by 3, but we do not need the best
possible bound at this point.) Clearly, a
n+1
> a
n
, so the sequence is also
monotone increasing. So we may apply the Theorem and conclude that it
5
converges to a limit e, say, in R. But it should be remarked that one can
show with more work that e is irrational. So there is a valuable lesson to be
learned here. Even though {a
n
} is a bounded, monotone sequence of rational
numbers, there is no limit in Q; one has to go to the enlarged number system
R.
A subsequence of a sequence {a
n
} is a subcollection indexed by a countably
innite subset J of N. For example, {a
2n
} and {a
2n+1
} are subsequences.
Theorem B Suppose {a
n
} is a bounded sequence of real numbers. Then it
has a subsequence which converges.
Proof. Put
J = {m N| a
n
a
m
, n m}.
Suppose J is an innite set. Write its elements as m
1
, m
2
, m
3
, . . .. By de-
nition, we have
a
m
1
a
m
2
a
m
3
. . . .
So the subsequence {a
m
| m J} (of {a
n
}) is a bounded, monotone decreas-
ing sequence, hence convergent by Theorem A. We are done in this case.
Now suppose J is nite. Pick an n
1
which is larger than any element
of J, which is possible. Then we can nd an n
2
> n
1
such that a
n
1
< a
n
2
.
Again we can nd an n
3
such that a
n
2
< a
n
3
, and so on. This way we get a
subsequence {a
n
1
, a
n
2
, a
n
3
, . . .} which is bounded and monotone increasing,
hence convergent by Theorem A.
2.2 Cauchys criterion
The main problem with the denition of convergence of a sequence is that
it is hard to verify it. For instance we need to have a candidate for the
limit to verify (2.1.2). One wants a better way to check for convergence. As
seen above, boundedness is a necessary, but not sucient, condition, unless
the sequence is also monotone (and real). Is there a necessary and sucent
condition? A nineteenth century French mathematician named Cauchy gave
an armative answer for real sequences. His criterion also works for complex
sequences.
6
A sequence {a
n
} is said to be a Cauchy sequence i we can nd, every
positive , an N > 0 such that
(2.2.1) |a
m
a
n
| < whenever n, m > N.
This is nice because it does not mention any limit.
Lemma 2.2.2 Every convergent sequence is Cauchy.
Proof. Suppose a
n
L. Pick any > 0. Then by denition, we can
nd an N > 0 such that, for all n > N, we have
|L a
n
| <

2
Then for n, m > N, the triangle inequality gives
|a
m
a
n
| |a
m
L| +|L a
n
| <

2
+

2
= .
Lemma 2.2.3 A complex sequence {z
n
= x
n
+ iy
n
}, with x
n
, y
n
real, is
Cauchy i {x
n
} and {y
n
} are Cauchy.
Proof. Suppose {x
n
}, {y
n
} are both Cauchy. Pick any > 0. Then we
can nd an N > 0 such that for all m, n > N,
|x
m
x
n
| <

2
, and |y
m
y
n
| <

2
.
Since z
m
z
n
is (x
m
x
n
) + i(y
m
y
n
) and |i| = 1, , we get by the triangle
inequality,
|z
m
z
n
| |x
m
x
n
| +|y
m
y
n
| < .
Conversely, suppose that {z
n
} is Cauchy. Recall that for any complex number
z = x + iy, |x| and |y| are both bounded by

x
2
+ y
2
, i.e., by |z|. This
immediately implies that {x
n
} and {y
n
} are Cauchy.
Here comes the beautiful result of Cauchy:
Theorem C Every Cauchy sequence of real (resp. complex) numbers con-
verges in R (resp. C).
Proof. By Corollary 2.1.3, a sequence {z
n
= x
n
+ iy
n
} of complex
numbers is convergent i the real sequences {x
n
} and {y
n
} are convergent.
7
By Lemma 2.2.3, {z
n
} is Cauchy i {x
n
}, {y
n
} are. So it suces to prove
Theorem C for real Cauchy sequences {a
n
}.
First note that such a sequence must be bounded. Indeed, x an > 0 and
nd an N > 0 for which (2.2.1) holds. Then for all n > N, a
n
lies between
a
N+1
and a
N+1
+ . Then the set {a
n
| n > N} is bounded. Then so is
the entire sequence because {a
1
, . . . , a
N
} is nite.
Now since {a
n
}is a bounded, real sequence, we can apply Theorem B and
nd a subsequence {a
m
1
, a
m
2
, . . .} which converges, say to the limit L. We
claim that the original sequence {a
n
} also converges to L. Indeed, let be
an arbitrary positive number. Then we can nd an N > 0 such that for all
m
k
, n > N,
|L a
m
k
| <

2
and |a
m
k
a
n
| <

2
.
Applying the triangle inequality we get
|L a
n
| < , n > N,
proving the claim.
2.3 Construction of Real Numbers revisited
Given any Cauchy sequence {a
n
} of rational numbers, we may of course
view it as a real sequence and apply Theorem C to deduce that it converges
to a limit L, say. But this L need not be rational. This leads us to wonder
if all the real numbers could be constructed by adjoining limits of Cauchy
sequences of rational numbers. This is essentially correct, and this approach
is quite dierent from that given in chapter 1. Now let us explain how it is
done.
Let us begin by pretending that we do not know what real numbers are,
but that we know everything about rational numbers.
The rst thing to note is that dierent Cauchy sequences of rational
numbers might have the same limit. For example, the sequences {0, 0, 0, . . .}
and {1,
1
2
,
1
3
, . . .} have 0 as their limit. So we dene an equivalence between
any pair of Cauchy sequences {a
n
} and {b
n
} with a
n
, b
n
Q are as follows:
(2.3.1) {a
n
} {b
n
} lim
n
(a
n
b
n
) = 0.
8
It will be left as a good exercise for the students to check that is indeed
an equivalence relation in the sense of chapter 0.
Denote the equivalence class of any {a
n
} by [a
n
]. It is not hard to check
that given two classes [a
n
] and [b
n
], they are either the same or disjoint, i.e.,
no rational Cauchy sequence {c
n
} can belong to two distinct classes.
Dene a real number to be an equivalence class [a
n
] as above. Denote
by R the collection of all real numbers in this sense. We then have a natural,
mapping
Q R, q [q, q, q, . . .],
where [q, q, q, . . .] is the class of the stationary sequence {q, q, q, . . .} (which
is evidently Cauchy). This mapping is one-to-one because if q, q

are unequal
rational numbers, then {q, q, q, . . .} cannot be equivalent to {q

, q

, q

, . . .}.
Consequently, we have 0 and 1 in R.
We know that the sum and product of Cauchy sequences are again Cauchy.
So we may set
[a
n
] [b
n
] = [a
n
b
n
], and [a
n
] [b
n
] = [a
n
b
n
].
It is immediate that one has commutativity, associativity and distributivity.
It remains to dene a multiplicative inverse of a non-zero element of R.
Let {a
n
} be a Cauchy sequence belonging to . Since is non-zero, all but
a nite number of the a
n
must be non-zero. Let {a
m
n
} be the subsequence
consisting of all the non-zero terms of {a
n
}. Then this subsequence also
represents , and we may set

1
= [a
1
m
n
].
Check that this gives a well dened multiplicative inverse.
Now it is more or less obvious that R is a eld under these operations,
containing Q as a subeld.
Next comes the notion of positivity. Dene a real number to be positive
i for some Cauchy sequence {a
n
} representing , we can nd an N > 0 such
that a
n
is positive for all n > N. Check that this intuitive notion is the right
one.
Armed with this powerful notion, one sees that R becomes an ordered
eld. By denition, the ordering on R is compatible with the usual one on
Q.
9
Finally, every Cauchy sequence converges in R. This is a nice consequence
of our construction. When is represented by a Cauchy sequence {a
n
}, the
sequence is forced to converge to .
Now we will end this section by discussing how one can mathematically
dene real numbers via their decimal expansions, which everyone intu-
itively understands. Here is what one does.
Dene a real number to be a pair (m, {a
n
}), where m is an integer and
{a
n
} is a sequence of non-negative integers, with the understanding that a
n
is not all 9 after some index N. One often writes
(m, {a
n
}) = m +

n=1
a
n
10
n
.
If we have a rational number of the form m+
r

n=1
a
n
, we will identify it with
(m, {a
n
}) with a
n
= 0 for all n > r.
Dene an ordering by setting
(m, {a
n
}) < (k, {b
n
}) m < k or m = k, N s.t. a
N
< b
N
, a
n
= b
n
n < N.
This leads quickly to the least upper bound property, so we can take the
suprema of bounded sets.
Given any real number = (m, {a
n
}), we will dene, for each r 1, its
truncation at stage r to be

r
= m +
r

n=1
a
n
Q.
Evidently,
= sup{
r
| r 1}.
So it is natural to set, for real numbers ,

= sup{
r

r
}
and

= sup{
k

k
}.
The denition of the multiplicative inverse of a non-zero is left as an exer-
cise.
10
2.4 Innite series
Given a sequence {a
n
}, we will often want to know if we can sum the terms
of the sequence, denoted
(2.4.1) S =

n=1
a
n
,
and get something meaningful. Such an innite sum is called an innite
series.
A natural way to proceed is as follows. We know how to sum, for any
n 1, the rst n terms of the sequence, namely
(2.4.2) s
n
=
n

m=1
a
n
.
This way we get another sequence {s
n
}, called the sequence of partial sums.
(One calls s
n
the nth partial sum.) We will say that the original sequence
{a
n
} is summable, or that the innite series (2.4.1) is convergent, i the
sequence of partial sums converges.
Let us rst look at the example of a geometric series. Fix any real
number x with |x| < 1, and put
a
n
= x
n1
( n 1),
and
(2.4.3) S =

n=1
x
n1
= 1 + x + x
2
+ x
3
+ . . .
We claim that
s
n
=
1 x
n
1 x
.
Indeed, we have
(1x)s
n
= (1x)(1+x+x
2
+. . .+x
n1
) = (1x)+(xx
2
)+. . .+(x
n1
x
n
) = 1x
n
,
proving the claim.
11
Now by Lemma 2.1.5 (and the discussion preceding it), we have
x
n
0 since |x| < 1.
Consequently,
(2.4.4) S = lim
n
s
n
=
1
1 x
.
Check that the series is divergent if |x| 1.
Lemma 2.4.5 If {a
n
}, {b
n
} are summable sequences, then so is {c
1
a
n
+c
2
b
n
}
for any pair of scalars c
1
, c
2
.
The proof is straight-forward and will be left as an exercise. The reader
should take note, however, that nothing is claimed here about the product
of two sequences.
We will say that a series

n1
a
n
is absolutely convergent if the series

n1
|a
n
|
is convergent. In this case one also says that the sequence {a
n
} is absolutely
summable. This notion is important and makes us want to understand well
the summability of positive sequences, which will be our focus in the next
section. It should be noted however (see section 2.6) that there are summable
sequences, which are not absolutely summable.
One of the most important reasons for studying absolutely summable
sequences is that it allows one to study the product of two sequences. This
is encapsulated in the following Proposition, which we will quote without
proof.
Proposition 2.4.6 Suppose {a
n
}, {b
n
} are two absolutely summable se-
quences. Consider the sequence {c
n
} with
c
n
=

i+j=n
a
i
b
j
.
Then {c
n
} is absolutely summable and moreover,

n=1
c
n
=

n=1
a
n

n=1
b
n
.
12
2.5 Tests for Convergence
We will concentrate mainly on the summability of sequences with non-negative
terms.
The rst thing one notices about such a sequence {a
n
} with each a
n
non-negative, is that
s
1
s
2
. . . s
n
. . . ,
i.e., {s
n
} is monotone increasing. In view of Theorem A, we then get the
following
Lemma 2.5.1 A sequence of non-negative numbers is summable i the
associated sequence {s
n
} of partial sums is bounded.
The rst test for convergence is given by the following
Proposition 2.5.2 (Comparison test) Let {a
n
}, {b
n
} be non-negative
sequences of real numbers, such that
(i) a
n
b
n
for all n 1; and
(ii) The sequence {b
n
} is summable.
Then {a
n
} is summable.
Proof. Let s
n
, t
n
denote the n-th partial sums of {a
n
}, {b
n
} respectively.
Then we have (for all n)
s
n
t
n
.
Since by hypothesis the sequence {b
n
} is summable, the sequence {t
n
} con-
verges. Then the t
n
, and hence the s
n
, must be bounded. So the sequence
{a
n
} is summable by Lemma 2.5.1.
Here comes the second test.
Proposition 2.5.4 (Ratio test)) Let {a
n
} be a sequence of non-negative
terms. Suppose the sequence {a
n+1
/a
n
} of ratios converges to a real number
r.
(i) If r < 1, then {a
n
} is summable;
(ii) If r > 1, then {a
n
} is not summable, i.e., the series

n1
a
n
diverges.
13
It is important to note that this test says nothing when r = 1.
Proof. Suppose r is less than 1. Pick any > 0 with r + < 1. Write
t = r + , so that r < t < 1. Then as {a
n+1
/a
n
} converges to r, we can nd
an N > 0 such that for all n > N, |r a
n+1
/a
n
| < . This implies that
n > N
a
n+1
a
n
< t.
In other words,
a
N+2
< a
N+1
t, a
N+3
< a
N+2
t < a
N+1
t
2
, . . .
We can rewrite it as saying that for all j 1, we have
a
N+j
< a
N+1
t
j1
,
so that

j=1
a
N+j
< a
N+1

j1
t
j1
.
This converges by the comparison test, because the geometric series

j1
t
j1
occurring on the right is convergent as t is positive and < 1 (with limit
1
1t
).
Consequently, the (partial) series

n>N
a
n
, and hence the whole series

n1
a
n
,
converges.
When r > 1, one sees easily that the sequence of partial sums is not
bounded, hence divergent.
Given a positive real number x and a positive integer n, there is a unique
positive real number, denoted x
1/n
, which is an n-th root of x, i.e., (x
1/n
)
n
=
x. We will also set 0
1/n
= 0.
Proposition 2.5.5 (Root test) Let {a
n
} be a sequence of non-negative
real numbers. Suppose the associated root sequence {a
1/n
n
} is convergent with
limit R, Then the following hold:
(i) If R < 1, then the series

n1
a
n
converges;
(ii) If R > 1, then the series

n1
a
n
diverges.
14
Just as in the Ratio test, this test says nothing when R = 1.
Proof. Suppose R < 1. Pick any > 0 with R+ < 1. Write t = R+,
so that R < t < 1. Then as {a
1/n
n
} converges to R, we can nd an N > 0
such that for all n > N, |R a
1/n
n
| < . This implies that
n > N a
n
< t
n
.
Hence

n>N
a
n
<

n>N
t
n
<

n1
t
n
.
The (partial) series on the left now converges by the comparison test, because
the geometric series

n1
t
n
occurring on the right is convergent (as t is
positive and < 1). Consequently, the complete series

n1
a
n
also converges.
When R > 1, the sequence of partial sums is not bounded, hence diver-
gent.
2.6 Alternating series
One of the most beautiful formulas in Mathematics is the formula of Leibniz
for /4 given at the beginning of this chapter. It is one of the glorious
examples of what are known as alternating series, i.e., whose terms alternate
in sign, which converge but not absolutely. Some people use the terminology
conditional convergence for this phenomenon. Leibniz was amazing in that
he not only found a slew of such formulas, he also discovered a theorem which
gives precise conditions on an alternating series to be convergent. Let us now
state and prove it.
Theorem D (Leibniz ) Let {b
n
} be a sequence of non-negative, mono-
tone decreasing sequence such that
lim
n
b
n
= 0.
Then the alternating series
S =

n=1
(1)
n+1
b
n
= b
1
b
2
+ b
3
b
4
+ . . .
15
is convergent.
Proof. Put a
n
= (1)
n+1
b
n
and
s
n
= a
1
+ a
2
+ . . . + a
n
= b
1
b
2
+ . . . + (1)
n+1
b
n
.
We have to show that the sequence {s
n
} of partial sums converges. First
note that for any n 1,
(2.6.1) s
2n
= s
2n1
b
2n
s
2n1
,
since the b
j
are non-negative. We also have (for all n 2)
(2.6.2) s
2n
= s
2n2
+ b
2n1
b
2n
s
2n2
,
and
(2.6.3) s
2n1
= s
2n3
b
2n2
+ b
2n1
s
2n3
,
both since the b
j
are monotone decreasing.
Now let i, j be natural numbers such that i is even and j odd. Then we
may pick an n > 0 such that 2n i and 2n1 j. Applying (2.6.1), (2.6.2)
and (2.6.3), we deduce that
(2.6.4) s
i
s
2n
s
2n1
s
j
.
In particular, s
i
is bounded from above, for any even index i, by s
1
= b
1
.
Similarly, for any odd index j, s
j
is bounded from below by s
2
= b
1
b
2
.
The inequalities given by (2.6.2) and (2.6.3) show that the subsequence
{s
2n
} is monotone increasing and {s
2n1
} is monotone decreasing. Since the
former is bounded from above, and the latter from below, they both converge,
say to respective limits and L, i.e., we have
lim
n
s
2n
= L = lim
n
s
2n1
.
The assertion L follows from (2.6.4).
Finally, note that by denition,
s
2n
s
2n1
= b
2n
.
Since the b
n
converge to zero as n , we see that the subsequences {s
2n
}
and s
2n1
must both converge to the same limit, i.e., = L. Consequently,
lim
n
s
n
= L.
16
Here are some examples of alternating series:
(2.6.5) S
1
= 1
1
2
+
1
3

1
4
+ . . .
(2.6.6) S
2
= 1
1
3
+
1
5

1
7
+ . . .
(2.6.7) S
3
=

2
3!
+

5
5!


7
7!
+ . . .
Clearly, {
1
n
} and {
1
2n1
} are non-negative, decreasing sequences converging
to zero. So we may apply Leibnizs theorem above and conclude that S
1
and
S
2
converge. Even though we will not evaluate these limits here, let us note
that
(2.6.8) S
1
= log 2 and S
2
=

4
.
Let us now turn to (2.6.7). We need to know that
(2.6.9)

n
n!


n1
(n 1)!
and that
(2.6.10)

n
n!
0 as n .
Since is known to be < 4, (2.6.9) is satised for all n 4, and when
we study exponential functions later in the course, we will see the truth of
(2.6.10). So by Leibnizs theorem, the series starting with n = 4 converges.
So S
3
itself converges. The amazing things is that one can show:
S
3
= 0.
17