You are on page 1of 19

Department of Mathematics

Indian Institute of Technology Delhi


MTL 100: Calculus
Lecture Notes

Sequences and Series

You
may
want
to
read about
the life &
works
of
Cauchy.

Augustin-Louis Cauchy
(1789 1857)

Sequences

Let X be a nonempty set. A sequence in X is a list of elements in X, not necessarily distinct,


with one chosen for each positive integer. In other words, there corresponds a unique element
xn X corresponding to each n N. We shall use the notation {xn }n1 to denote a sequence
consisting of the elements x1 , x2 , x3 , . . .. The element xn is called the n th term of the sequence.
More formally, a sequence in X can be thought of as a list of terms f (1), f (2), f (3), . . . in order
and f : N X is any function. Occasionally, we allow the domain of the function f to include 0,
so that the corresponding sequence begins with x0 instead of x1 . Throughout these notes, we will
only consider real sequences: X = R.
The main topic of study in regard to sequences is whether or not is eventually approaches a real
number. When the terms of a sequences approach a real number and stay within an arbitrarily
small interval around this real number, the sequence is said to converge to this limit. If this is
not the case, the sequence is said to diverge.
Definition 1. A sequence s of real numbers is a listing, in order, of values of any function
f : N R.
We typically denote a sequence in these notes by the letter s. If we relate the sequence s to
a function s : N R, then the elements or terms of the sequence are s(1), s(2), s(3), . . ., with
the typical nth term being s(n). More traditionally, we denote the terms by using the positive
integers as a subscript and enclosing the terms within brackets:
s = {s1 , s2 , s3 , . . . , sn , . . .}.
Definition 2. A sequence {sn } of real numbers is said to converge to a real number ` provided
that, for any given  > 0, there exists a positive integer N such that |sn `| <  whenever n > N .
1

Elements
of any set
that is in
bijection
with
the
set
of
natural
numbers is
a sequence.

We use one of the equivalent notations


lim sn = `,

{sn } `

to denote that the sequence {xn } converges to `. A sequence that does not converge to a real
number is said to diverge.
Example 1.
(i) Every constant sequence converges. In fact, if sn = c, then {sn } c.

{sn } `
means
sN +1 , sN +2 ,
sN +3 , . . .
(`, `+)
 > 0
and some
N = N ().

(ii) {sn } 0 if and only if {|sn |} 0. If ` 6= 0, {sn } ` implies {|sn |} |`| but {|sn |} |`|
does not imply {sn } `.
(iii) The sequence

n o
1
n

converges to 0.

(iv) The sequence {(1)n } does not converge, whereas the sequence
n

(v) The sequence (1 + n1 )n


n

(vi) The sequence (1 +

1
2

(1)n
n

converges to 0.

e and
are two of
the most
important
constants
in
Mathematics.
You
may
want
to
read more
about
these.

converges.1
o

+ + n1 ) ln n converges.2

You
may
want
to
read about
the life &
works
of
Euler.

Leonard Euler
(1707 1783)
Theorem 1. Every convergent sequence of real numbers has a unique limit.
Proof. Suppose the real sequence {sn } converges to both `1 and `2 . If `1 6= `2 , we may suppose
without loss of generality that `1 > `2 . We will show that `1 `2 <  for every choice of  > 0,
thereby proving that `1 `2 = 0. This will prove that {sn } converges to a unique real number.
Let  > 0 be any real number. Since {sn } `1 , there exists N1 such that |sn `1 | <  whenever
n > N1 . And since {sn } `2 , there exists N2 such that |sn `2 | <  whenever n > N2 . So if
N = max{N1 , N2 }, both |sn `1 | <  and |sn `2 | <  whenever n > N . But then
`1 `2 = |`1 `2 | = |(sn `1 ) (sn `2 )| |sn `1 | + |sn `2 | <  +  = 2.

Theorem 1 allows us to say that ` is the limit of the sequence {sn }.
1
The limit of this sequence is denoted by the letter e. It turns out that e is not only an irrational number, but
also transcendental, and e 2.7182818284 to ten decimal places.
2
The limit of this sequence is denoted by the Greek letter . It turns out that is not only an irrational number,
but also transcendental, and 0.5772156649 to ten decimal places.

Proposition 1. (The Squeeze Principle)


Let {sn }, {xn }, {yn } be a sequences of real numbers, and n1 , n2 be positive integers such that
xn sn for n > n1 and sn yn for n > n2 . If {xn } ` and {yn } `, then {sn } `.
Proof. Let  be any positive real number. Since {xn } and {yn } converge to `, there exist positive
integers N1 and N2 such that |xn `| <  whenever n > N1 and |yn `| <  whenever n > N2 .
Thus with N = max{n1 , n2 , N1 , N2 } we have
x n s n yn ,

|xn `| < ,

|yn `| < 

whenever n > N . Therefore `  < xn sn yn < ` +  whenever n > N , so that {sn } `.

Definition 3. We say the sequence {sn } of real numbers is bounded if there is a real number M
such that |sn | M for each n 1.
All sequences in Example 1 are bounded. The sequence {1, 2, 3, . . .} is not bounded. The following
theorem shows that the class of all bounded sequences properly contains the class of all convergent
sequences.
Theorem 2. Every convergent sequence of real numbers is bounded. However, not every bounded
sequence is convergent.
Proof. Let {sn } be a sequence of real numbers converging to ` R. Then there exists N such
that |sn `| < 1 whenever n > N ; in other words, sn (` 1, ` + 1) whenever n > N . On the other
hand, the first N terms of the sequence lie in the interval [m, M ], where m = min{s1 , . . . , sN } and
M = max{s1 , . . . , sN }. Thus the entire sequence lies in [m, M ] (` 1, ` + 1), and is therefore
bounded.
The sequence {+1, 1, +1, 1, . . .} is bounded (by 1, for instance) but not convergent (can you see
why?). Thus we have an example of a bounded and divergent sequence.

Proposition 2. (Properties of the Limits of Sequences)
Suppose {xn }, {yn } are convergent sequences of real numbers, with {xn } X and {yn } Y .
(i) For any c R, {cxn } cX.
(ii) {xn yn } X Y .
(iii) {xn yn } XY .
(iv) If yn 6= 0 for all sufficiently large n and Y 6= 0, then {1/yn } 1/Y .
(v) If yn 6= 0 for all sufficiently large n and Y 6= 0, then {xn /yn } X/Y .
Proof. Let  be any positive real number. Then there exists N1 and N2 such that
|xn X| <  for n > N1

and

|yn Y | <  for n > N2 .

(1)

By setting N = max{N1 , N2 }, we note that both inequalities in (1) hold for n > N .
(i) From (1),
|cxn cX| = |c| |xn X| < |c| for n > N1 .
Hence {cxn } cX.
(ii) From (1),
|(xn + yn ) (X + Y )| = |(xn X) + (yn Y )| |xn X| + |yn Y | <  +  = 2 for n > N.
Hence {xn + yn } X + Y .
Putting c = 1 in part (i), we obtain {yn } Y . Hence {xn yn } = {xn +(yn )} XY .
3

Convergent
sequence
are
bounded,
but
bounded
sequences
need
not
be convergent.

(iii) We know that the convergent sequence {xn } is bounded by Theorem 2; let |xn | M for
n 1. From (1),
|xn yn XY | = |xn (yn Y )+(xn X)Y | |xn | |yn Y |+|xn X| |Y | < M +|Y | = (M +|Y |).
Hence {xn yn } XY .
Note that if yn = c for each n 1, then Y = c and we have the result of part (i).
(iv) Suppose yn 6= 0 whenever n > n0 . We choose  = |Y |/2 in (1). Then there exists N0 such
that yn (Y , Y + ) whenever n > N0 . In particular, m := inf{|yn | : n > n0 } > 0. Hence
from (1),



1

1 yn Y

y Y = y Y < m |Y | .
n
n
Therefore {1/yn } 1/Y .
(v) From (1), and parts (iii), (iv),

xn
yn

= xn

1
yn

1
Y

.


Proposition 3. (Some Important Limits)


(i) {1/np } 0 for p > 0.
(ii) {an } 0 for |a| < 1.

(iii) { n n} 1.

(iv) { n a} 1 for a > 0.


Proof.
(i) Fix p > 0, and let  be any positive real number. Then for n > N =

q
p

1
,

we have np >

1






and so n1p 0 < . Thus {1/np } 0.

(ii) If a = 0, the sequence is a constant 0 and so the limit is also 0. If |a| < 1, a 6= 0, we may
write |a| = 1/(1 + b) with b > 0. Then
0 < |an 0| = |a|n =

1
1
<
,
n
(1 + b)
nb

the last inequality from the binomial expansion (1 + b)n > 1 + nb > nb since all missing terms
are positive. The result now follows from Example 1, part (ii) and Proposition 1.

n
n = 1 + sn with sn > 0 for n > 1. Thus
(iii) Since n n > 1 for n > 1, we may write
q

n 2
2
n
n = (1 + sn ) > 2 sn , so that 0 < sn < n1 for n > 1. Thus {sn } 0 by Proposition 1,

and { n n} 1.

(iv) If a 1, then for n a we have 1 n a n n. Now { n a} 1, by part (iii) and


Proposition 1.

If 0 < a < 1, then applying the above result to b = 1/a gives { n b} 1. Proposition 2, part

(v) now implies { n a} 1.




Definition 4. A sequence {sn } of real numbers is nondecreasing (increasing) if sn sn+1


(sn < sn+1 ) whenever n 1, and nonincreasing (decreasing) if sn sn+1 (sn > sn+1 ) whenever n 1.
A sequence that is one of nondecreasing, increasing, nonincreasing, decreasing is said to be monotonic.
It is often convenient to indicate by {sn } that the sequence is nondecreasing or increasing, and
by {sn } that the sequence is nonincreasing or decreasing.
Theorem 3. Every nondecreasing sequence that is bounded above and every nonincreasing sequence
that is bounded below converges.

Bounded
monotonic
sequences
converge.

Proof. We prove the theorem only for bounded nondecreasing sequences; the same proof holds
for increasing sequences and an analogous proof holds for nonincreasing and decreasing sequences.
Let {sn } be a bounded nondecreasing sequence of real numbers, and let S = {sn : n N}. Since
{sn } is bounded, sup S R. We show that {sn } sup S. Let  be any positive real number.
From the definition of supremum, we have that sup S  is not an upper bound for S; consequently
sn > sup S  for at least one n, say N . But then sn sN > sup S  for all n > N , because {sn }
is nondecreasing. Hence
sup S  < sn sup S for n > N,
and this proves that |sn sup S| <  whenever n > N . Therefore {sn } is convergent.

Definition 5. We say that a sequence {sn } of real numbers diverges to +, and write lim sn =
+, when to each real number R there corresponds a positive integer N such that sn > R whenever
n > N.
We say that a sequence {sn } of real numbers diverges to , and write lim sn = , when to
each real number r there corresponds a positive integer N such that xn < r whenever n > N .
Proposition 4. Suppose {sn } is an unbounded sequence of real numbers. If {sn } is nondecreasing,
then lim sn = +; if {sn } is nonincreasing, then lim sn = .
Proof. The first term of a nondecreasing sequence serves as its lower bound (in fact, its infimum),
while the first term of a nonincreasing sequence serves as its upper bound (in fact, its supremum).
So since {xn } is unbounded, for any real number R, we must have sn R for all sufficiently large
n in the first case, and sn R for all sufficiently large n in the second case. This proves the
proposition.

Remark 1. Providing a proof for the convergence of a given sequence typically requires the application of Definition 2. This often requires playing round with algebraic inequalities involving
the absolute value function x 7 |x|. The behaviour of many sequences is not apparent from the
definition we do not know whether the sequence is convergent or divergent even after computing
several terms. Many sequences that we encounter are related to the ones given in Example 1 and in
Proposition 3. In general, a standard method to prove the convergence of a sequence without having
to go through the definition is to use either Theorem 3 or Corollary 1.
n

Example 2. We show that the sequence 1 +


between 2 and 3. This is Example 1, part (v).

1
n

n o

is convergent, and converges to a real number

You
may
want
to
read
about the
constant e.

Let an = (1 + n1 )n for n 1. Then


n
X

an =

k=0
n
X

n 1
k nk

n (k 1)
1 n n1

k! n
n
n
k=0


n
X
1

1
=
1
k!
n
k=0

n
X
1



2
k1
1
1
n
n


1
n+1



1
1
1
k!
n+1
k=0
= an+1 .



<

k=0
n+1
X

<

k!

(2)

2
k1
1
n+1
n+1


2
k1
1
n+1
n+1

Thus the sequence {an } is increasing.


Using the fact that k! = (2 3 k) 2k1 for k 2, (2) gives
2=1+

n
X
n
1
< an < 1 +
= 1 + 2 = 3.
k1
n
2
k=1

Thus the sequence {an } is bounded above by 3, and hence converges (to a real number in between 2
and 3), by Theorem 3. The limit is denoted by e, was discovered by Jacob Bernoulli in 1683, but
given this notation by Euler after himself.


n

Example 3. We show that the sequence


1 + 21 + + n1 ln n is convergent, and converges
to a real number between 0 and 1. This is Example 1, part (vi).
We use a little bit of Calculus to prove this. The function f : [0, 1) R given by f (x) = x+ln(1x)
is decreasing, since f 0 (x) = 1 1/(1 x) = x/(1 x) < 0 in (0, 1). Thus f (x) < f (0) = 0 in
(0, 1). Hence
for each positive
integer n, x = 1/n gives f (1/n) = 1/n + ln(1 1/n) < 0.


1
1
Let bn = 1 + 2 + + n ln n for n 1. Then

1
1
1
+ ln(n 1) ln n = + ln 1
n
n
n


bn bn1 =

< 0.

Hence the sequence {bn } is decreasing.


If we plot the graph of y = 1/x from x = 1 to x = n + 1, and estimate the area under the curve
y = 1/x between x = 1 and x = n with the areas obtained by approximating by rectangle of unit
width, we find that
n
X
1
k=1

1=

n
X
1
k=2

<

Z n
1
1

dx = ln n <

n1
X

n
X
1
1
<
.
k k=1 k
k=1

Thus the sequence {bn } is bounded below by 0, and hence converges (to a real number in between 0
and 1), by Theorem 3. The limit is denoted by , and called the Euler-Mascheroni constant, was
discovered by Euler in 1734. However, Euler used the symbol C to denote the limit.
Remark 2. More can be said about how closely the sequence {bn } in Example 3 approaches . For
instance,
1
1
< bn <
.
2(n + 1)
2n

You
may
want
to
read
about the
constant .

However, a small change in the sequence {bn } leads to a sequence that approximates much better:


cn = 1 +

1
1
+ +
2
n

ln n +

1
2

satisfies
1
1
< cn <
.
2
24(n + 1)
24n2
Let {sn } be a sequence of real numbers. For each positive integer N , let
uN = inf{sn : n > N }

vN = sup{sn : n > N }.

and

(3)

Since the sets SN := {sn : n > N } get smaller (SN +1 SN for each N 1) as N increases,
u1 u2 u3 . . . un . . .

v1 v2 v3 . . . vn . . . .

and

Since the sequence {un } , Theorem 3 would imply convergence if it was bounded and Proposition 4
would imply divergence to + if unbounded. The sequence {vn } , and the same conclusions may
be drawn for the sequence {vn } except that it would diverge to if unbounded. The sequence
{un } and {vn } play a major role in the theory of sequences.
Definition 6. Let {sn } be a sequence of real numbers. We define the limit superior and the
limit inferior of the sequence {sn } by
lim sup {sn } = lim

sup{sn : n > N }

lim inf {sn } = lim

and

inf{sn : n > N } .

If {sn } is not bounded above, then we define lim sup {sn } = +. If {sn } is not bounded below,
then we define lim inf {sn } = .
Proposition 5. (Properties of lim sup & lim inf)
Let {sn } be a sequence of real numbers.

limsup and
liminf
of
sequences
always exist, either
as a real
number, or
as + or
.

(i) lim inf {sn } lim sup {sn }.


(ii) lim inf {sn } = lim sup {sn }.
(iii) If {sn } is bounded, and if sn tn whenever n > n0 for some positive integer n0 , then
lim inf {sn } lim inf {tn },

lim sup {sn } lim sup {tn }.

(iv) If {sn } and {tn } are bounded sequences of real numbers, then
lim inf {sn + tn } lim inf {sn } + lim inf {tn },

lim sup {sn + tn } lim sup {sn } + lim sup {tn }.

(v) If {sn } is bounded, then lim inf {sn } = m if and only if, for any positive real number , (a)
sn > m  for all sufficiently large values of n, and (b) sn < m +  for infinitely many values
of n.
(vi) If {sn } is bounded, then lim sup {sn } = M if and only if, for any positive real number , (A)
sn < M +  for all sufficiently large values of n, and (B) sn > M  for infinitely many
values of n.
Proof. Let {sn } be a sequence of real numbers. We use the notation given by (3) for the sequence
{sn }, and UN and VN for the corresponding terms for the sequence {tn } throughout this proof.

Basic properties
of
limsup and
liminf
of
sequences.

(i) If {sn } is bounded, then uN vN for each N N. Hence lim inf {sn } = lim uN lim vN =
lim sup {sn }.
If {sn } is unbounded, the either lim sup {sn } = +, or lim inf {sn } = . The inequality
now follows from the definition < x < + for all x R.
(ii) For any nonempty set X of real numbers, we claim that
inf X = sup(X).
Note that X = {x : x X}. By the definition of supremum, x sup(X), and for
each positive real number , there exists x (X) such that x > sup(X) . This is
the same as x sup(X), and and for each positive real number , there exists x X
such that x < sup(X) + . Thus sup(X) is the greatest lower bound for X, that is,
sup(X) = inf X, as claimed.
Therefore
lim sup {sn } = lim

sup{sn : n > N } = lim sup{sn : n > N }


N

= lim

inf{sn : n > N } = lim inf {sn }.

(iii) Since sn tn whenever n > n0 , we have


uN UN and vN VN .
whenever N > n0 . Taking limits as n on both sides of these inequalities, we have the
desired inequalities.
(iv) For each N N, sn uN and tn UN for n > N by definition of uN and UN . Hence
sn + tn uN + UN for n > N , so that inf{sn + tn : n > N } uN + UN by the definition of
infimum. Therefore
lim inf {sn + tn } =

lim

lim

inf{sn + tn : n > N }

uN + UN

lim uN + lim UN

= lim inf {sn } + lim inf {tn }.


For each N N, sn vN and tn VN for n > N by definition of vN and VN . Hence
sn + tn vN + VN for n > N , so that sup{sn + tn : n > N } vN + VN by the definition of
supremum. Therefore
lim sup {sn + tn } =

lim

lim

sup{sn + tn : n > N }

vN + V N

lim vN + lim VN

= lim sup {sn } + lim sup {tn }.


(v) Suppose lim inf {sn } = m. We prove both conditions (a) and (b) hold, by contradiction. If
(a) was false, there would exist  > 0 such that sn m  for infinitely many values of n.
But then, for every N N, uN = inf{sn : n > N } m . Taking limits as n , we get
lim inf {sn } m , which contradicts our hypothesis. Thus (a) is true.

If (b) was false, there would exist  > 0 such that sn < m+ for only finitely many values of n.
So there must exist a positive integer N for which sn m +  whenever n > N . Applying the
result of part (ii), now leads to lim inf {sn } m + , which again contradicts our hypothesis.
Thus (b) is true.
Conversely suppose both conditions (a) and (b) hold for the sequence {sn }. Let  be any
positive real number. By condition (a), there exists a positive integer N such that sn >
m  for n > N . Hence uN = inf{sn : n > N } m , and since the sequence {un } is
nondecreasing, un m  for n > N . Taking limits as n , we get lim inf {sn } m .
Since this is true for any  > 0, lim inf {sn } m.
On the other hand, for each positive integer N , condition (b) implies sn < m +  for at least
one n > N . Hence uN = inf{sn : n > N } m + . Taking limits as N , we get
lim inf {sn } m + . Since this is true for any  > 0, lim inf {sn } m.
Combining the two inequalities we see that lim inf {sn } = m.
(vi) Suppose lim sup {sn } = M . We prove both conditions (A) and (B) hold, by contradiction. If
(A) was false, there would exist  > 0 such that sn M +  for infinitely many values of n.
But then, for every N N, vN = sup{sn : n > N } M + . Taking limits as n , we get
lim sup {sn } M + , which contradicts our hypothesis. Thus (A) is true.
If (B) was false, there would exist  > 0 such that sn > M  for only finitely many values
of n. So there must exist a positive integer N for which sn M  whenever n > N .
Applying the result of part (ii), now leads to lim sup {sn } M , which again contradicts
our hypothesis. Thus (B) is true.
Conversely suppose both conditions (A) and (B) hold for the sequence {sn }. Let  be any
positive real number. By condition (A), there exists a positive integer N such that sn < M +
for n > N . Hence vN = sup{sn : n > N } M + , and since the sequence {vn } is
nonincreasing, vn M +  for n > N . Taking limits as n , we get lim sup {sn } M + .
Since this is true for any  > 0, lim sup {sn } M .
On the other hand, for each positive integer N , condition (B) implies sn > M  for at least
one n > N . Hence vN = sup{sn : n > N } M . Taking limits as N , we get
lim sup {sn } M . Since this is true for any  > 0, lim sup {sn } M .
Combining the two inequalities we see that lim sup {sn } = M .

Exercise 1. If {sn } is a sequence of real numbers, and if n = (s1 + + sn )/n for n N, prove
that
lim inf {n } lim inf {sn }, lim sup {n } lim sup {sn }.
Theorem 4. Let {sn } be a sequence of real numbers, and let ` R {}.
(i) If lim {sn } = `, then lim sup {sn } = lim inf {sn } = `.
(ii) If lim sup {sn } = lim inf {sn } = `, then lim {sn } = `.
Proof. Let {sn } be a sequence of real numbers. We use the notation given by (3) throughout
this proof. Thus
lim inf {sn } = lim uN = U

and

lim sup {sn } = lim vN = V.

(i) Suppose lim {sn } = +, and let R be any positive real number. Then there exists a positive
integer N such that sn > R whenever n > N . Thus uN = inf{sn : n > N } R and

Convergent
sequences
are
precisely those
sequences
whose
limsup and
liminf are
equal.

vN = sup{sn : n > N } R, which imply that un R and vn R whenever n N . Hence


lim inf {sn } = lim {uN } = + and lim sup {sn } = lim {vN } = +.
Suppose lim {sn } = , and let r be any negative real number. Then there exists a positive
integer N such that sn < r whenever n > N . Thus uN = inf{sn : n > N } r and
vN = sup{sn : n > N } r, which imply that un r and vn r whenever n N . Hence
lim inf {sn } = lim {uN } = and lim sup {sn } = lim {vN } = .
Suppose lim {sn } = ` R, and let  be any positive real number. Then there exists a positive
integer N such that sn (` , ` + ) whenever n > N . Thus uN = inf{sn : n > N } ` 
and vN = sup{sn : n > N } ` + , which imply that un `  and vn ` +  whenever
n N . Since both inequalities hold for every  > 0, we see that lim inf {sn } = lim {uN } `
and lim sup {sn } = lim {vN } `. Since lim inf {sn } lim sup {sn }, we have ` lim inf {sn }
lim sup {sn } `. Thus lim inf {sn } = lim sup {sn } = `.
(ii) Suppose lim sup {sn } = lim inf {sn } = +, and let R be any positive real number. Thus
{uN } +, and so there exists N1 such that uN > R whenever N > N1 . Since uN =
inf{sn : n > N }, we have sn > R whenever n > N1 . Hence lim {sn } = +.
Suppose lim sup {sn } = lim inf {sn } = , and let r be any negative real number. Thus
{vN } , and so there exists N2 such that vN < r whenever N > N2 . Since vN =
sup{sn : n > N }, we have xn < r whenever n > N2 . Hence lim {sn } = .
Suppose lim sup {sn } = lim inf {sn } = ` R, and let  be any positive real number. Since
{uN } ` and {vN } `, there exists N1 and N2 such that |uN `| <  whenever N > N1
and |vN `| <  whenever N > N2 . Set N0 = max{N1 , N2 }. Then both uN and vN
lie in the interval (` , ` + ) whenever N > N0 . Since uN = inf{sn : n > N } and
vN = sup{sn : n > N }, this implies both sn > `  and sn < ` +  whenever n > N0 .
Therefore |sn `| <  whenever n > N0 , so that {sn } `.

Definition 7. A sequence {sn } is said to be a Cauchy sequence if, to each  > 0, there exists a
positive integer N such that |sm sn | <  whenever m > N and n > N .
Theorem 5. A sequence {sn } of real numbers converges to a real number if and only if it is a
Cauchy sequence.
Proof. Suppose {sn } is a convergent sequence with limit `. Let  be any positive real number.
Then there exists a positive integer N such that |sn `| <  whenever n > N . Thus
|sm sn | = |(sm `) (sn `)| |sm `| + |sn `| <  +  = 2.
Thus the sequence {sn } is Cauchy.
Conversely suppose {sn } is a Cauchy sequence. Let  be any positive real number. Choose a
positive integer N such that |sm sn | <  whenever m, n > N . This implies, in particular,
that sn < sm +  for all m, n > N . Hence sm +  is an upper bound for {sn : n > N }, and
so vN = sup{sn : n > N } sm +  whenever m > N . But now vN  is a lower bound for
{sn : n > N }, and so vN  inf{sn : n > N } = uN . Therefore
lim sup {sn } vN uN +  lim inf {sn } + .
Since this holds for all  > 0, we have lim sup {sn } lim inf {sn }. Since lim sup {sn } lim inf {sn }
always holds, we lim sup {sn } = lim inf {sn }. Now Theorem 4 implies the sequence {sn } is convergent.


10

Convergent
sequences
and
Cauchy
sequences
are
the
same
for
real
sequences.
They are
not
for
rational
sequences!

Exercise 2. Consider the sequence of rational numbers


1 3 7 17 41 99
, , , , , ,....
1 2 5 12 29 70
(i) Determine the next three terms of the sequence. What formula connects the nth term to the
previous two?
(ii) Prove that the sequence is Cauchy.
(iii) Prove that the sequence converges to the irrational number

2.

Remark 3. Theorem 5 holds for real sequences, but not for rational sequences. Convergent sequences are always Cauchy, but the converse does not hold true for rational sequences; see Exercise 2.
Definition 8. A subsequence of a sequence {sn } of real numbers is a part of the sequence that
takes the form {snk }, where {nk } be an increasing sequence of positive integers.
Theorem 6. If {sn } converges to `, then every subsequence of {sn } also converges to `.
Proof. Let {sn } `, and let {snk } be a subsequence of {sn }. Let  be any positive real number.
Then there exists a positive integer N such that |sn `| <  whenever n > N . Since {snk } is a
subsequence, nk k. So if k > N , then nk > N and |snk `| < . Thus {snk } `.


An
example
of
a
nonconvergent
Cauchy
sequence
of rational
numbers.
A
subsequence is
obtained
from a sequence by
selecting
some
or
all of its
terms, in
the order in
which they
appear.

Theorem 7. Every sequence has a monotonic subsequence.


Proof. Let {sn } be a sequence of real numbers. For this proof, we call the term sn dominant if
sn > sm whenever m > n.
Suppose there are infinitely many dominant terms in the sequence. Then any subsequence {snk }
consisting only of dominant terms is a decreasing sequence, since snk > snk+1 for each k 1.
Otherwise, there are only finitely many dominant terms in the sequence. If sN is the last dominant
term, let n1 = N + 1. Since sn1 is not a dominant term, there exists n2 > n1 such that sn2 sn1 .
We select the smallest such n2 , and note that n1 < n2 while sn1 sn2 . Suppose n1 , n2 , . . . , nk have
been selected such that
n1 < n2 < < nk

and

sn1 sn2 snk .

Since nk > n1 > N , snk is not a dominant term. Among all terms in the sequence that succeed
snk and are at least as large as snk , select the first such term. Call this term snk+1 , and note that
nk+1 > nk and snk+1 snk . Thus this selection can be continued indefinitely, and {snk } provides
us a nondecreasing subsequence of {sn }.


Bernard Placidus Johann Gonzal


Nepomuk Bolzano
(1781 1848)

Karl Theodor Wilhelm Weierstrass


(1815 1897)

11

You
may
want
to
read about
the life &
works
of
Bolzano
and
of
Weierstrass.

Theorem 8. (Bolzano-Weierstrass, 1817)


Every bounded sequence of real numbers has a convergent subsequence.
Proof. Let {sn } be a bounded sequence of real numbers. Then {sn } has a monotonic subsequence

{snk } by Theorem 7, which must converge (since it is also bounded) by Theorem 3.
Definition 9. Let {sn } be a real sequence. A subsequential limit is any real or extended real
number that is the limit of some subsequence of {sn }.
Theorem 9. Let {sn } be a sequence of real numbers, and let S denote the set of subsequential
limits of {sn }. Then S contains both lim sup {sn } and lim inf {sn }. In particular, {sn } converges
if and only if S = {lim {sn }}.
Proof. We use the notation given by (3) for the sequence {sn }. If lim sup {sn } = , then the
sequence itself converges to its limit superior.
We follow the proof of Theorem 7. If there are infinitely many dominant terms in the sequence,
the subsequence {snk } consisting only of dominant terms satisfies snk = sup{sn : n nk } = vnk 1 .
Hence {snk } converges to lim sup {sn }, since {vN } converges to lim sup {sn } by definition.
Now assume there are infinitely many dominant terms in the sequence, and let the sequence {tn }
be given by tn = lim sup {sn } 1/n if lim sup {sn } is finite, and let tn = n if lim sup {sn } = +.
Since tn < lim sup {sn } vN = sup{sn : n > N } for each positive integer N , the subsequence in
the proof of Theorem 7 may be selected to further satisfy snk+1 > tnk for all k 1. Since we always
have sn sup{sm : m n} = vn1 , we conclude that
tnk < snk+1 vnk+1 1 for all k 1.
Since both {tnk } and {vnk+1 1 } are subsequences of sequences that converge to lim sup {sn }, we
have {tnk } lim sup {sn } and {vnk+1 1 } lim sup {sn }. Therefore {snk+1 } is a subsequence of
{sn } that converges to lim sup {sn }. Thus lim sup {sn } S .
To prove that lim inf {sn } S , choose a monotonic subsequence {tnk } of {sn } converging to
lim sup {sn }. Then {tnk } is a monotonic subsequence of {sn }, and converges to lim inf {sn } by
Proposition 5, part (ii). Thus lim inf {sn } S .
The particular case now follows immediately from Theorem 4.

Corollary 1. Suppose {sn } is a sequence of real numbers such that {s2n+1 } converges to `1 and
{s2n } converges to `2 . Then {sn } converges if and only if `1 = `2 .
Proof. Although this is a direct consequence of Theorem 9, we also provide a direct proof.
Suppose {s2n+1 } ` and {s2n } `, and let  be any positive real number. Then there exist
positive integers N1 and N2 such that |s2n+1 `| <  whenever n > N1 and |s2n `| <  whenever
n > N2 . For N = max{N1 , N2 } and n > N , we therefore have both inequalities, and consequently
|sn `| <  whenever n > N .

Theorem 10. For any sequence {sn } of nonzero real numbers,
lim inf




q

q


sn+1
sn+1
n
n

lim inf


|s
|

lim
sup
|s
|

lim
sup
n
n
s
s .
n

Proof. Let us denote the four limits in order by , , , . Observe the second inequality follows
from the definition. There is nothing to prove in the first inequality if = , and in the third
inequality if = +.

12

To see the third inequality, let R be any real number greater than . By the definition of limit
superior, there exists a positive integer N such that



sn+1


: n N < R.
sup
s
n

Hence |sn+1 /sn | < R for n N , and for n > N we have








sn sn1


sn =

sN +1 sN < RnN sN = c Rn ,
s
s
s

n1
n2
N

where c = |sN |/RN is fixed since R and N are fixed. Thus


q

n
sn < n c R for n > N,

and since limn

c = 1 by Proposition 3, part (iv), we conclude that R. Therefore . 

Corollary 2. Let {sn } be a sequence of nonzero real numbers for which the sequence
converges to `. Then the sequence

np

o
n
sn+1
sn

|sn | also converges to `.

Proof. Suppose the sequence sn+1


sn converges to `. Then by Theorem 4, = = ` in Theorem
10. Thus all four limits in Theorem 10 are equal to `, and Theorem 4 shows that

np
n

|sn | `. 

Series

The terms sequence and series are often used interchangeably in daily usage. They are, in fact,
mathematically speaking quite different. A series of real numbers is obtained by adding in order
the terms of a sequence. When there are only finitely many terms to add in the series, there is
no question about the existence of the sum. However, there is no guarantee that an infinite series
would have a finite sum.
Definition 10. A series of real numbers is an expression of the form a1 + a2 + a3 + + an + .
The series is said to be finite if there are finitely many terms in the sum, and infinite if there
are infinitely many terms in the sum.
Definition 11. The sequence of partial sums corresponding to the series a1 +a2 +a3 + +an +
is the sequence {Sn }n1 , where
Sn := a1 + a2 + a3 + + an =

n
X

ai .

i=1

The series a1 + a2 + a3 + + an + is said to converge or diverge according to the convergence


or divergence of the sequence {Sn } of partial sums. The sum of a convergent series is defined to be
limn Sn .
Theorem 11. (Cauchys Criterion for Convergence)
P
A series
an converges if and only if, given any positive real number , there exists a positive
integer N such that


n
X



ak <  whenever n m > N.



k=m

13

(4)

Proof. By definition, the series an converges if and only if the sequence {Sn } of partial sums
converges, and the sequence {Sn } converges if and only if it is a Cauchy sequence by Theorem 5.
The condition on being Cauchy is as follows: given any  > 0, there exists a positive integer N
such that |Sn Sm | <  whenever m, n > N . Note that this inequality is valid when n = m. If
P
P
Pn
n > m, then |Sn Sm | = | nk=1 ak m
k=1 ak | = |
k=m+1 ak |; if n = m, |Sn Sm | = 0. Hence
the condition in (4).

P

Corollary 3. If a series

an converges, then lim an = 0.

Proof. Suppose an converges. By Theorem 11, given any positive real number , there exists a
P
positive integer N such that | nk=n ak | <  whenever n > N . Thus lim |an | = 0, and so lim an = 0
by Example 1, part (ii).

Remark 4. The converse to Corollary 3 is false: lim an = 0 does not imply that
as the example an = 1/n shows.
Proposition 6. The geometric series

Proof. If |r| 1, |a an | = |a| |r|n does not converge to 0. Hence a an diverges, by Corollary 3.
P
If |r| < 1, then {arn } 0. Thus Sn = a ni=1 ri = ar(1 rn )/(1 r) ar/(1 r), proving that
P
an converges.

P

Theorem 12. (The Comparison Tests)


Let N0 be a positive integer, and let an 0 for all n > N0 .
P

an converges and |bn | an for all n > N1 , then

(ii) If

an = + and bn an for all n > N2 , then

bn converges.

bn = +.

Proof.
(i) Let N = max{N0 , N1 }. Then for n N and any k 0, by triangle inequality,


+k
+k
N
+k
NX
NX
X


bn
|bn |
an .



n=N

n=N

(5)

n=N

Since an converges, the sum on the right is less than any prescribed  > 0 whenever N is
P
sufficiently large by Theorem 4. Hence the same holds true for bn by (5), so that the series
converges, again by Theorem 11.
n
n
0
(ii) If Sn =
0 , N2 }. Since
i=1 ai and Tn =
i=1 bi , then Tn Sn for n N = max{NP
P
an = +, we have {Sn } +. Hence {Tn } +, which is the same as bn = +.


Theorem 13. (The Limit Comparison Tests)
Let {an } and {bn } be sequence of positive real numbers, and let lim an /bn = `.
n

(i) If 0 < ` < , then the series


(ii) If ` = 0 and the series

(iii) If ` = and the series

an and

bn both converge or both diverge.

bn converges, then the series

bn diverges, then the series


14

an also converges.

an also diverges.

Convergence
implies
an 0.
So an 6 0
implies
divergence,
but divergence does
not imply
an 6 0.

an converges,

arn converges if and only if |r| < 1.

(i) If

Cauchys
criterion
for convergence.

arn
converges
iff |r| < 1.

Series
converges
if
it
is
dominated
term-byterm by a
convergent
series,
and
diverges if it
dominates
term-byterm
a
divergent
series.

Proof.

(i) Let  be any positive real number. Then there exists a positive integer N such that abnn ` < 
whenever n > N . Hence `  < abnn < ` + , and so (` )bn < an < (` + )bn whenever
P
P
P
n > N . Therefore if
bn converges, then so does
an ; and if
bn diverges, then so does
P
an , by Theorem 12.

Limit
Comparison tests
follow from
the Comparison
tests.

(ii) Let  be any


real number less than 1. Then there exists a positive integer N such
positive



that abnn = abnn 0 <  whenever n > N . Hence an < bn < bn whenever n > N , and so the
P
series an converges by Theorem 12.
(iii) Let R be any positive real number greater than 1. Then there exists a positive integer N
such that abnn > R whenever n > N . Hence an > Rbn > bn whenever n > N , and so the series
P
an diverges by Theorem 12.

Definition 12. The series
verges.

an is said to converge absolutely provided the series

|an | con-

Corollary 4. If a series is absolutely convergent, then it is convergent.


P

Definition 13. The series an is said to converge conditionally provided the series converges
but does not converges absolutely.
Theorem 14. (The Root Test) (Cauchy, 1821)
P
A series an
np
o

converges absolutely
if lim sup n |an | < 1;
diverges

if lim sup

The test gives no information when lim sup


Proof. We use the notation = lim sup

np
n

np
n

np
n

|an | > 1.
P

|an | = 1.
o

|an | , as in Theorem 10.

Suppose < 1. Then we may choose  > 0 such that 0 < +  < 1. By the definition of limit
superior, there exists a positive integer N such that
0 <  < sup

q
n

|an | : n > N

< +  < 1.

In particular, we have 0 |an | < ( + )n whenever n > N . Since the geometric series ( + )n
P
P
converges by Proposition 6, Theorem 5pimplies n>N |an | converges. Hence |an | also converges.
Suppose > 1. Then the sequence { n |an |} has a subsequential limit > 1 by Theorem 9, and
consequently |an | > 1 for infinitely many choices of n. Thus {an } does not converge to 0, and so
cannot converge by Corollary 3.
For an = 1/n and bn = 1/n2 , Proposition 3, part (iii) shows = 1 in both cases, and we know
P
P
that an diverges whereas bn converges. Thus the test fails to provide a definite answer when
= 1.

P

15

an converges absolutely if
< 1 &
diverges if
> 1.

Jean le Rond dAlembert


(1717 1783)

You
may
want
to
read about
the life &
works
of
dAlembert.

Theorem 15. (The Ratio Test) (dAlembert, 1768)


P
A series an of nonzero terms
n

converges absolutely

if lim sup an+1


an < 1;

diverges

if lim inf

o
n
an+1
an > 1.

o
o
n
n


an+1
The test gives no information when lim inf an+1

lim
sup


an
an .
o
o
n
n


an+1
and
Proof. We use the notation = lim inf an+1

=
lim
sup


an
an , as in Theorem 10.

Recall that by the same theorem. So if < 1, then < 1 and the series converges
absolutely by Theorem 14; and if > 1, then > 1 and the series diverges by Theorem 14.
P
For an = 1/n and bn = 1/n2 , = 1 = in both cases, and we know that an diverges whereas
P
bn converges. Thus the test fails to provide a definite answer when 1 .

Exercise 3. Give a direct proof of Theorem 15.
Remark 5. The Ratio test may be derived as a consequence of the Root test, as was shown in the
proof of Theorem 15. In fact, the Root test is stronger than the Ratio test in the sense that there are
series which fail the Ratio test but whose convergence or divergence can be determined by applying
the Root test.
Observe that
1/n

lim sup |an |


n


1/n


an

an
an1
a1
an1
a1 1/n



= lim sup

a0
= lim sup

,
an1 an2
a0
an1 an2
a0
n
n
1/n

the second equality since limn a0 = 1. Since this is the geometric mean of the first n consecutive
ratios of the series, the Ratio test depends on the behavior in the limit of each consecutive ratio, the
Root test only considers the average behavior of these ratios. Clearly, if all the consecutive ratios
get small then their average value will get small as well. The converse is false, which is why the
Root test is stronger.
For instance, consider the following rearrangement of the geometric series with first term 1 and
ratio 1/2:
1 1
1
1
1
+1+ + +
+
+ .
2
8 4 32 16
16

an converges absolutely if
< 1 &
diverges if
> 1.

Root test
is stronger
than Ratio
test,
but
often
harder to
apply.

The Ratio test fails since consecutive ratios alternate between 2 and 1/8, which implies = 1/8
and = 2. However, the geometric mean of the first 2n consecutive ratios is


n 1 1/2n
1

2
= .

8n
2

The Root test shows the series converges.


Exercise 4. Let {an } be a sequence of positive real numbers. Prove that, if
1
n n
lim

then the series

an
an1
a1
+
+ +
an1 an2
a0

< 1,

an converges.

Gottfried Wilhelm Leibniz


(1646 1716)
Theorem 16. (The Alternating Series Test) (Leibniz)
P
An alternating series (1)n1 an converges if {an } is a nonincreasing sequence of positive real
numbers that tends to 0. Moreover, if the series converges to S, then
|S Sn | < an+1
for each n 1 and Sn =

Pn

i1 a .
i
i=1 (1)

Proof.
Consider the sequences S2n+1 of partial sums with an odd number of terms and the
sequences S2n of partial sums with an even number of terms. Observe that
S2n+1 S2n1 = a2n+1 a2n 0,

S2n+2 S2n = a2n+1 a2n+2 0.

Since S2n+1 S2n = a2n+1 0, we have


a1 a2 = S2 S2n S2n+1 S1 = a1 .
Thus {S2n+1 } and are bounded below by a1 a2 whereas {S2n } and are bounded above by
a1 . By Theorem 3, both sequences must converge; say, {S2n+1 } S 0 and {S2n } S 00 . Thus
S 0 S 00 = lim(S2n+1 S2n ) = lim a2n+1 = 0, so that S 0 = S 00 . This proves the convergence of the
sequence {Sn } by Corollary 1.
If S is the common limit of the sequences {S2n+1 } and {S2n }, we also S2n S S2n+1 for each
n 1. Hence |S S2n+1 | = S2n+1 S < S2n+1 S2n+2 = a2n+2 and |S S2n | = S S2n <
S2n+1 S2n = a2n+1 . Thus |S Sn | < an+1 for each n 1.


17

You
may
want
to
read about
the life &
works
of
Leibniz.

For convergence of an
alternating
series a1
a2 + a3
a4 + ,
you need
{an } 0.

Colin Maclaurin
(1698 1746)
Theorem 17. (The Integral Test) (Maclaurin-Cauchy)
Let N be a positive integer, and let f be a continuous, decreasing and nonnegative real-valued
function defined on [N, ). Then the series

You
may
want
to
read about
the life &
works
of
Maclaurin.

Z R

f (n) converges if and only if

n=N

lim

R N

f (x) dx exists.

Moreover, if the series converges then


Z R

lim

R N

f (x) dx

f (n) f (N ) + lim

R N

n=N

Z R

and
R

f (x) dx.

Proof. Since f is a decreasing function, for each integer n N , we have


Z n+1

f (x) dx

Z n+1

Z n

f (n) dx = f (n) =
n

f (n) dx

Z n

n1

f (x) dx.

(6)

n1

Summing all three terms in (6) from n = N + 1 to n = R, and adding f (N ) to each term, we get
Z R+1

f (N ) +

f (x) dx

N +1

Z N +1

Since
N

f (x) dx

R
X

f (n) f (N ) +

Z R

f (x) dx.

(7)

n=N

Z N +1

Z R+1

f (N ) dx = f (N ), the first term in (7) may be replaced by


N

f (x) dx.
N

The condition on convergence of the series, as well as the bounds on the sum when convergent, now
follow on taking limits as R .

Example 4. (The p-series Test)
P
p
We show that the series
> 1. Observe that both the Ratio Test
n=1 1/n converges

if and only if p p
an+1
and the Root Test fail because limn an = 1 = limn n |an |.
For p > 0, the function f (x) = 1/xp is continuous, decreasing and positive on [1, ). If p 6= 1,
then
Z R

lim

R 1

xp dx =


1
lim R1p 1
1 p R
(

18

1
p1

if p > 1;

if 0 < p < 1.

f (n)

f (x) dx
behave
alike if f :
[N, )
R+
is
continuous
and .

By Theorem 17,
the series converges when p > 1 and diverges when 0 < p < 1.
Z
R

If p = 1, lim

R 1

x1 dx = lim ln R = +. Hence the series diverges when p = 1, by Theorem


R

17.
If p 0, limn np 6= 0. Hence the series diverges when p 0, by Corollary 3.

References
Richard R. Goldberg, Methods of Real Analysis, Second Edition, John Wiley & Sons, 1976.
Kenneth A. Ross, Elementary Analysis: The Theory of Calculus, Third Edition, McGraw Hill
International Editions, 1976.

19

You might also like