Lim Inf and Sup PDF

Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

Chapter 3

Sequences

In this chapter, we discuss sequences. We say what it means for a sequence to


converge, and define the limit of a convergent sequence. We begin with some
preliminary results about the absolute value, which can be used to define a distance
function, or metric, on R. In turn, convergence is defined in terms of this metric.

3.1. The absolute value


Definition 3.1. The absolute value of x R is defined by
(
x
if x 0,
|x| =
x if x < 0.
Some basic properties of the absolute value are the following.
Proposition 3.2. For all x, y R:
(a) |x| 0 and |x| = 0 if and only if x = 0;

(b) | x| = |x|;

(c) |xy| = |x| |y|;

(d) |x + y| |x| + |y| (triangle inequality).


Proof. Parts (a), (b) follow immediately from the definition. Part (c) remains
valid if we change x to x or y to y, so we can assume that x, y 0 without loss
of generality. Then xy 0 and |xy| = xy = |x||y|. Part (d) remains valid if we
change the signs of both x and y or exchange x and y. Therefore we can assume
that x 0 and |x| |y| without loss of generality, in which case x + y 0. If
y 0, corresponding to the case when x and y have the same sign, then
|x + y| = x + y = |x| + |y|.
If y < 0, corresponding to the case when x and y have opposite signs, then
|x + y| = x + y = |x| |y| < |x| + |y|,
29

30

3. Sequences

which proves (c).

One useful consequence of the triangle inequality is the following reverse triangle inequality.
Proposition 3.3. If x, y R, then

||x| |y|| |x y|.

Proof. By the triangle inequality,


|x| = |x y + y| |x y| + |y|

so |x| |y| |x y|. Similarly, exchanging x and y, we get |y| |x| |x y|, which
proves the result.

We can give an equivalent condition for the boundedness of a set by using the
absolute value instead of upper and lower bounds as in Definition 2.9.
Proposition 3.4. A set A R is bounded if and only if there exists a real number
M 0 such that
|x| M for every x A.
Proof. If the condition in the proposition holds, then M is an upper bound of
A and M is a lower bound, so A is bounded. Conversely, if A is bounded from
above by M and from below by m , then |x| M for every x A where M =
max{|m |, |M |}.

A third way to say that a set is bounded is in terms of its diameter.
Definition 3.5. Let A R. The diameter of A is

diam A = sup {|x y| : x, y A} .

Then a set is bounded if and only if its diameter is finite.


Example 3.6. If A = (a, a), then diam A = 2a, and A is bounded. If A =
(, a), then diam A = , and A is unbounded.

3.2. Sequences
A sequence (xn ) of real numbers is an ordered list of numbers xn R, called the
terms of the sequence, indexed by the natural numbers n N. We often indicate a
sequence by listing the first few terms, especially if they have an obvious pattern.
Of course, no finite list of terms is, on its own, sufficient to define a sequence.
Example 3.7. Here are some sequences:
1, 8, 27, 64, . . . ,
1 1 1
1, , , , . . .
2 3 4
1, 1, 1, 1, . . .
2 
3

1
1
, 1+
, ...
(1 + 1) , 1 +
2
3

xn = n3 ,
1
xn = ;
n
xn = (1)n+1 ,
n

1
xn = 1 +
.
n

31

3.2. Sequences

3
2.9
2.8
2.7

xn

2.6
2.5
2.4
2.3
2.2
2.1
2

10

15

20
n

25

30

35

40

Figure 1. A plot of the first 40 terms in the sequence xn = (1 + 1/n)n ,


illustrating that it is monotone increasing and converges to e 2.718, whose
value is indicated by the dashed line.

Note that unlike sets, where elements are not repeated, the terms in a sequence
may be repeated.
The formal definition of a sequence is as a function on N, which is equivalent
to its definition as a list.
Definition 3.8. A sequence (xn ) of real numbers is a function f : N R, where
xn = f (n).
We can consider sequences of many different types of objects (for example,
sequences of functions) but for now we only consider sequences of real numbers,
and we will refer to them as sequences for short. A useful way to visualize a
sequence (xn ) is to plot the graph of xn R versus n N. (See Figure 1 for an
example.)
If we want to indicate the range of the index n explicitly, we write the sequence
as (xn )
n=1 . Sometimes it is convenient to start numbering a sequence from a
different integer, such as n = 0 instead of n = 1. In that case, a sequence (xn )
n=0
is a function f : N0 R where xn = f (n) and N0 = {0, 1, 2, 3, . . . }, and similarly
for other starting points.
Every function f : N R defines a sequence, corresponding to an arbitrary
choice of a real number xn R for each n N. Some sequences can be defined
explicitly by giving an expression for the nth terms, as in Example 3.7; others can

32

3. Sequences

be defined recursively. That is, we specify the value of the initial term (or terms) in
the sequence, and define xn as a function of the previous terms (x1 , x2 , . . . , xn1 ).
A well-known example of a recursive sequence is the Fibonacci sequence (Fn )
1, 1, 2, 3, 5, 8, 13, . . . ,
which is defined by F1 = F2 = 1 and
Fn = Fn1 + Fn2

for n 3.

That is, we add the two preceding terms to get the next term. In general, we cannot
expect to solve a recursion relation (especially if it is nonlinear) and get an explicit
expression for the nth term. The recursion relation for the Fibonacci sequence is
linear with constant coefficients, so it can be solved to give explicit expression for
Fn , called the Euler-Binet formula.
Proposition 3.9 (Euler-Binet formula). The nth term in the Fibonacci sequence
is given by



n 
1
1+ 5
1
n
,
=
.
Fn =

2
5
Proof. The terms in the Fibonacci sequence are uniquely determined by the linear
difference equation
Fn Fn1 Fn2 = 0,
n 3,
with the initial conditions
F1 = 1,
F2 = 1.
We see that Fn = rn is a solution of the difference equation if r satisfies
r2 r 1 = 0,
which gives

1
1+ 5
r = or ,
=
1.61803.

2
By linearity, the general solution of the difference equation is
n

1
n
Fn = A + B

where A, B are arbitrary constants. Using the initial conditions F1 = F2 = 1 to


evaluate A, B, we get the result.

Alternatively, once we know the answer, we can prove Proposition 3.9 by induction. The details are left as an exercise. Note that
although the right-hand side
of the equation for Fn involves the irrational number 5, its value is an integer for
every n N.

The number appearing in Proposition 3.9 is the golden ratio. It has the
property that subtracting 1 from it gives its reciprocal, or
1
1= .

Geometrically, this property means that removing a square from a rectangle whose
sides are in the ratio leaves a rectangle whose sides are in the same ratio. As a

33

3.3. Convergence and limits

result, the Ancient Greeks considered such rectangles to be the most aesthetically
pleasing.

3.3. Convergence and limits


Roughly speaking, a sequence (xn ) converges to a limit x if its terms xn get arbitrarily close to x for all sufficiently large n.
Definition 3.10. A sequence (xn ) of real numbers converges to a limit x R,
written
x = lim xn , or xn x as n ,
n

if for every > 0 there exists N N such that


|xn x| <

for all n > N .

A sequence converges if it converges to some limit x R, otherwise it diverges.


Although we dont show it explicitly in the definition, N is allowed to depend
on . Typically, the smaller we choose , the larger we have to make N . One way
to view a proof of convergence is as a game: If I give you an > 0 you have to
come up with an N that works. Also note that xn x as n means the
same thing as |xn x| 0 as n .
It may appear obvious that a limit is unique if one exists, but this fact requires
proof.
Proposition 3.11. If a sequence converges, then its limit is unique.
Proof. Suppose that (xn ) is a sequence such that xn x and xn x as n .
Let > 0. Then there exist N, N N such that

|xn x| <
for all n > N ,
2

for all n > N .


|xn x | <
2
Choose any n > max{N, N }. Then, by the triangle inequality,

|x x | |x xn | + |xn x | < + < .


2 2
Since this inequality holds for all > 0, we must have |x x | = 0 (otherwise it
would be false for = |x x |/2 > 0), so x = x .

The following notation for sequences that diverge to infinity is convenient.
Definition 3.12. If (xn ) is a sequence then
lim xn = ,

or xn as n , if for every M R there exists N R such that


xn > M

for all n > N .

Also
lim xn = ,

34

3. Sequences

or xn as n , if for every M R there exists N R such that


xn < M

for all n > N .

That is, xn as n means the terms of the sequence (xn ) get arbitrarily
large and positive for all sufficiently large n, while xn as n means
that the terms get arbitrarily large and negative for all sufficiently large n. The
notation xn does not mean that the sequence converges.
To illustrate these definitions, we discuss the convergence of the sequences in
Example 3.7.
Example 3.13. The terms in the sequence
1, 8, 27, 64, . . . ,

xn = n3

eventually exceed any real number, so n3 as n and this sequence


does not converge. Explicitly, let M R be given, and choose N N such that
N > M 1/3 . (If < M < 1, we can choose N = 1.) Then for all n > N , we have
n3 > N 3 > M , which proves the result.
Example 3.14. The terms in the sequence
1
1 1 1
xn =
1, , , , . . .
2 3 4
n
get closer to 0 as n gets larger, and the sequence converges to 0:
1
lim
= 0.
n n
To prove this limit, let > 0 be given. Choose N N such that N > 1/. (Such a
number exists by the Archimedean property of R stated in Theorem 2.18.) Then
for all n > N


1

0 = 1 < 1 < ,
n
n
N
which proves that 1/n 0 as n .
Example 3.15. The terms in the sequence
1, 1, 1, 1, . . .

xn = (1)n+1 ,

oscillate back and forth infinitely often between 1 and 1, but they do not approach
any fixed limit, so the sequence does not converge. To show this explicitly, note
that for every x R we have either |x 1| 1 or |x + 1| 1. It follows that there
is no N N such that |xn x| < 1 for all n > N . Thus, Definition 3.10 fails if
= 1 however we choose x R, and the sequence does not converge.
Example 3.16. The convergence of the sequence

2 
3
n

1
1
1
(1 + 1) , 1 +
, 1+
,...
xn = 1 +
,
2
3
n
illustrated in Figure 1, is less obvious. Its terms are given by
1
xn = ann ,
an = 1 + .
n
As n increases, we take larger powers of numbers that get closer to one. If a > 1
is any fixed real number, then an as n so the sequence (an ) does not

35

3.3. Convergence and limits

converge (see Proposition 3.31 below for a detailed proof). On the other hand, if
a = 1, then 1n = 1 for all n N so the sequence (1n ) converges to 1. Thus, there
are two competing factors in the sequence with increasing n: an 1 but n .
It is not immediately obvious which of these factors wins.
In fact, they are in balance. As we prove in Proposition 3.32 below, the sequence
converges with
n

1
= e,
lim 1 +
n
n
where 2 < e < 3. This limit can be taken as the definition of e 2.71828.
For comparison, one can also show that
n2
n


1
1
= .
= 1,
lim 1 +
lim 1 + 2
n
n
n
n

In the first case, the rapid approach of an = 1+1/n2 to 1 beats the slower growth
in the exponent n, while in the second case, the rapid growth in the exponent n2
beats the slower approach of an = 1 + 1/n to 1.
An important property of a sequence is whether or not it is bounded.
Definition 3.17. A sequence (xn ) of real numbers is bounded from above if there
exists M R such that xn M for all n N, and bounded from below if there
exists m R such that xn m for all n N. A sequence is bounded if it is
bounded from above and below, otherwise it is unbounded.
An equivalent condition for a sequence (xn ) to be bounded is that there exists
M 0 such that
|xn | M for all n N.
Example 3.18. The sequence (n3 ) is bounded
from below but not from above,

while the sequences (1/n) and (1)n+1 are bounded. The sequence
1, 2, 3, 4, 5, 6, . . .

xn = (1)n+1 n

is not bounded from below or above.


We then have the following property of convergent sequences.
Proposition 3.19. A convergent sequence is bounded.
Proof. Let (xn ) be a convergent sequence with limit x. There exists N N such
that
|xn x| < 1
for all n > N .

The triangle inequality implies that

|xn | |xn x| + |x| < 1 + |x|

for all n > N .

Defining
M = max {|x1 |, |x2 |, . . . , |xN |, 1 + |x|} ,

we see that |xn | M for all n N, so (xn ) is bounded.

36

3. Sequences

Thus, boundedness is a necessary condition for convergence, and every unbounded sequence diverges; for example, the unbounded sequence in Example 3.13
diverges. On the other hand, boundedness is not a sufficient condition for convergence; for example, the bounded sequence in Example 3.15 diverges.
The boundedness, or convergence, of a sequence (xn )
n=1 depends only on the
behavior of the infinite tails (xn )
n=N of the sequence, where N is arbitrarily

large. Equivalently, the sequence (xn )


n=1 and the shifted sequences (xn+N )n=1
have the same convergence properties and limits for every N N. As a result,
changing a finite number of terms in a sequence doesnt alter its boundedness or
convergence, nor does it alter the limit of a convergent sequence. In particular, the
existence of a limit gives no information about how quickly a sequence converges
to its limit.
Example 3.20. Changing the first hundred terms of the sequence (1/n) from 1/n
to n, we get the sequence
1
1
1
,
,
, ...,
101 102 103
which is still bounded (although by 100 instead of by 1) and still convergent to
0. Similarly, changing the first billion terms in the sequence doesnt change its
boundedness or convergence.
1, 2, 3, . . . , 99, 100,

We introduce some convenient terminology to describe the behavior of tails


of a sequence,
Definition 3.21. Let P (x) denote a property of real numbers x R. If (xn ) is a
real sequence, then P (xn ) holds eventually if there exists N N such that P (xn )
holds for all n > N ; and P (xn ) holds infinitely often if for every N N there exists
n > N such that P (xn ) holds.
For example, (xn ) is bounded if there exists M 0 such that |xn | M
eventually; and (xn ) does not converge to x R if there exists 0 > 0 such that
|xn x| 0 infinitely often.
Note that if a property P holds infinitely often according to Definition 3.21,
then it does indeed hold infinitely often: If N = 1, then there exists n1 > 1 such
that P (xn1 ) holds; if N = n1 , then there exists n2 > n1 such that P (xn2 ) holds;
then there exists n3 > n2 such that P (xn3 ) holds, and so on.

3.4. Properties of limits


In this section, we prove some order and algebraic properties of limits of sequences.
3.4.1. Monotonicity. Limits of convergent sequences preserve (non-strict) inequalities.
Theorem 3.22. If (xn ) and (yn ) are convergent sequences and xn yn for all
n N, then
lim xn lim yn .
n

37

3.4. Properties of limits

Proof. Suppose that xn x and yn y as n . Then, for every > 0 there


exists P, Q N such that

for all n > P ,


|x xn | <
2

|y yn | <
for all n > Q,
2
Choosing n > max{P, Q}, we have

x = xn + x xn < yn + = y + yn y + < y + .
2
2
Since this holds for every > 0, it follows that x y.

This result, of course, remains valid if the inequality xn yn holds only for
all sufficiently large n > N . It follows immediately that if (xn ) is a convergent
sequence with m xn M for all n N, then
m lim xn M.
n

Limits need not preserve strict inequalities. For example, 1/n > 0 for all n N but
limn 1/n = 0.
The following squeeze or sandwich theorem is often useful in proving the
convergence of a sequence by bounding it between two simpler convergent sequences
with equal limits.
Theorem 3.23. Suppose that (xn ) and (yn ) are convergent sequences of real numbers with the same limit L. If (zn ) is a real sequence such that
xn zn yn

then (zn ) also converges to L.

for all n N.

Proof. Let > 0 be given, and choose P, Q N such that


|xn L| < for all n > P ,

If N = max{P, Q}, then for all n > N

|yn L| < for all n > Q.

< xn L < zn L yn L < ,

which implies that |zn L| < . This prove the result.

It is essential here that (xn ) and (yn ) have the same limit.
Example 3.24. If xn = 1, yn = 1, and zn = (1)n+1 , then xn zn yn for all
n N, the sequence (xn ) converges to 1 and (yn ) converges 1, but (zn ) does not
converge.
As once consequence, we show that we can take limits inside absolute values.
Corollary 3.25. If xn x as n , then |xn | |x| as n .
Proof. By the reverse triangle inequality,
0 | |xn | |x| | |xn x|,

and the result follows from Theorem 3.23.

38

3. Sequences

3.4.2. Linearity. Limits respect addition and multiplication. In proving the


following theorem, we need to show that the sequences converge, not just get an
expressions for their limits.
Theorem 3.26. Suppose that (xn ) and (yn ) are convergent real sequences and
c R. Then the sequences (cxn ), (xn + yn ), and (xn yn ) converge, and
lim cxn = c lim xn ,

lim (xn + yn ) = lim xn + lim yn ,


n
n
n



lim yn .
lim (xn yn ) = lim xn
n

Proof. We let

x = lim xn ,
n

y = lim yn .
n

The first statement is immediate if c = 0. Otherwise, let > 0 be given, and choose
N N such that

|xn x| <
for all n > N .
|c|
Then
|cxn cx| <
for all n > N ,
which proves that (cxn ) converges to cx.
For the second statement, let > 0 be given, and choose P, Q N such that

for all n > P ,


|yn y| <
for all n > Q.
|xn x| <
2
2
Let N = max{P, Q}. Then for all n > N , we have
|(xn + yn ) (x + y)| |xn x| + |yn y| < ,
which proves that (xn + yn ) converges to x + y.
For the third statement, note that since (xn ) and (yn ) converge, they are
bounded and there exists M > 0 such that
|xn |, |yn | M

for all n N

and |x|, |y| M . Given > 0, choose P, Q N such that

|xn x| <
for all n > P ,
|yn y| <
2M
2M
and let N = max{P, Q}. Then for all n > N ,

for all n > Q,

|xn yn xy| = |(xn x)yn + x(yn y)|

|xn x| |yn | + |x| |yn y|


M (|xn x| + |yn y|)
< ,

which proves that (xn yn ) converges to xy.

Note that the convergence of (xn + yn ) does not imply the convergence of (xn )
and (yn ) separately; for example, take xn = n and yn = n. If, however, (xn )
converges then (yn ) converges if and only if (xn + yn ) converges.

39

3.5. Monotone sequences

3.5. Monotone sequences


Monotone sequences have particularly simple convergence properties.
Definition 3.27. A sequence (xn ) of real numbers is increasing if xn+1 xn for
all n N, decreasing if xn+1 xn for all n N, and monotone if it is increasing
or decreasing. A sequence is strictly increasing if xn+1 > xn , strictly decreasing if
xn+1 < xn , and strictly monotone if it is strictly increasing or strictly decreasing.
We dont require a monotone sequence to be strictly monotone, but this usage isnt universal. In some places, increasing or decreasing is used to mean
strictly increasing or strictly decreasing. In that case, what we call an increasing sequence is called a nondecreasing sequence and a decreasing sequence is called
nonincreasing sequence. Well use the more easily understood direct terminology.
Example 3.28. The sequence
1, 2, 2, 3, 3, 3, 4, 4, 4, 4, 5, . . .
is monotone increasing but not strictly monotone increasing; the sequence (n3 ) is
strictly monotone increasing; the sequence (1/n) is strictly monotone decreasing;
and the sequence ((1)n+1 ) is not monotone.
Bounded monotone sequences always converge, and unbounded monotone sequences diverge to .
Theorem 3.29. A monotone sequence of real numbers converges if and only if it
is bounded. If (xn ) is monotone increasing and bounded, then
lim xn = sup{xn : n N},

and if (xn ) is monotone decreasing and bounded, then


lim xn = inf{xn : n N}.

Furthermore, if (xn ) is monotone increasing and unbounded, then


lim xn = ,

and if (xn ) is monotone decreasing and unbounded, then


lim xn = .

Proof. If the sequence converges, then by Proposition 3.19 it is bounded.


Conversely, suppose that (xn ) is a bounded, monotone increasing sequence.
The set of terms {xn : n N} is bounded from above, so by Axiom 2.17 it has a
supremum
x = sup{xn : n N}.
Let > 0. From the definition of the supremum, there exists an N N such that
xN > x . Since the sequence is increasing, we have xn xN for all n > N , and
therefore x < xn x. It follows that
|xn x| <

which proves that xn x as n .

for all n > N ,

40

3. Sequences

If (xn ) is an unbounded monotone increasing sequence, then it is not bounded


from above, since it is bounded from below by its first term x1 . Hence, for every
M R there exists N N such that xN > M . Since the sequence is increasing, we
have xn xN > M for all n > N , which proves that xn as n .
The result for a monotone decreasing sequence (xn ) follows similarly, or by
applying the previous result to the monotone increasing sequence (xn ).

The fact that every bounded monotone sequence has a limit is another way to
express the completeness of R. For example, this
is not true in Q: an increasing
sequence of rational numbers that converges to 2 is bounded from above in Q (for
example, by 2) but has no limit in Q.
We sometimes use the notation xn x to indicate that (xn ) is a monotone
increasing sequence that converges to x, and xn x to indicate that (xn ) is a
monotone decreasing sequence that converges to x, with a similar notation for
monotone sequences that diverge to . For example, 1/n 0 and n3 as
n .
The following propositions give some examples of a monotone sequences. In
the proofs, we use the binomial theorem, which we state without proof.
Theorem 3.30 (Binomial). If x, y R and n N, then
 
n  
X
n nk k
n
n!
.
(x + y)n =
x
y ,
=
k!(n k)!
k
k
k=0

Here, n! = 1 2 3 n and, by convention, we define 0! = 1. The binomial


coefficients
 
n (n 1) (n 2) (n k + 1)
n
=
,
k
1 2 3k
read n choose k, give the number of ways of choosing k objects from n objects,
order not counting. For example,
(x + y)2 = x2 + 2xy + y 2 ,
(x + y)3 = x3 + 3x2 y + 3xy 2 + y 3 ,
(x + y)4 = x4 + 4x3 y + 6x2 y 2 + 4xy 3 + y 4 .
We also recall the sum of a geometric series: if a 6= 1, then
n
X
1 an+1
ak =
.
1a
k=0

Proposition 3.31. Let a R. The geometric sequence (an )


n=0 ,
1, a, a2 , a3 , . . . ,

is strictly monotone decreasing if 0 < a < 1, with


lim an = 0,

and strictly monotone increasing if 1 < a < , with


lim an = .

41

3.5. Monotone sequences

Proof. If 0 < a < 1, then 0 < an+1 = a an < an , so the sequence (an ) is strictly
monotone decreasing and bounded from below by zero. Therefore by Theorem 3.29
it has a limit x R. Theorem 3.26 implies that
x = lim an+1 = lim a an = a lim an = ax.
n

Since a 6= 1, it follows that x = 0.


If a > 1, then an+1 = a an > an , so (an ) is strictly increasing. Let a = 1 +
where > 0. By the binomial theorem, we have
an = (1 + )n
n  
X
n k
=

k
k=0

1
= 1 + n + n(n 1) 2 + + n
2
> 1 + n.

Given M 0, choose N N such that N > M/. Then for all n > N , we have
an > 1 + n > 1 + N > M,

so an as n .

The next proposition proves the existence of the limit for e in Example 3.16.
Proposition 3.32. The sequence (xn ) with
n

1
xn = 1 +
n

is strictly monotone increasing and converges to a limit 2 < e < 3.


Proof. By the binomial theorem,

n X
n  
1
n 1
1+
=
n
k nk
k=0

1
n(n 1)(n 2) 1
n(n 1) 1
+
2+
3
n
2!
n
3!
n
n(n 1)(n 2) . . . 2 1 1
n
+ +


 n!  
n
1
1
1
2
1
1
+
1
1
=2+
2!
n
3!
n
n



1
1
2
2 1
+ +
1
1
... .
n!
n
n
n n
=1+n

Each of the terms in the sum on the right hand side is a positive increasing function
of n, and the number of terms increases with n. Therefore (xn ) is a strictly increasing sequence, and xn > 2 for every n 2. Moreover, since 0 (1 k/n) < 1, we
have
n

1
1
1
1
< 2 + + + + .
1+
n
2! 3!
n!

42

3. Sequences

Since n! 2n1 for n 1, it follows that


n



1
1
1 1 (1/2)n1
1
1
<3
< 2 + + 2 + + n1 = 2 +
1+
n
2 2
2
2
1 1/2
so (xn ) is monotone increasing and bounded from above by a number strictly less
than 3. By Theorem 3.29 the sequence converges to a limit 2 < e < 3.


3.6. The lim sup and lim inf


The lim sup and lim inf allow us to reduce questions about the convergence and limits of general real sequences to ones about monotone sequences. They are somewhat
subtle concepts, and after defining them we will consider a number of examples.
Unlike the limit, the lim sup and lim inf of every bounded sequence of real
numbers exist. A sequence converges if and only if its lim sup and lim inf are equal,
in which case its limit is their common value. Furthermore, a sequence is unbounded
if and only if at least one of its lim sup or lim inf diverges to , and it diverges
to if and only if both its lim sup and lim inf diverge to .
In order to define the lim sup and lim inf of a sequence (xn ) of real numbers,
introduce the sequences (yn ) and (zn ) of extended real numbers obtained by taking
the supremum and infimum, respectively, of the tails of (xn ):
yn = sup {xk : k n} ,

zn = inf {xk : k n} .

As n increases, the supremum and infimum are taken over smaller sets, so (yn ) is
monotone decreasing and (zn ) is monotone increasing. The limits of these sequences
are the lim sup and lim inf, respectively, of the original sequence.
Definition 3.33. Suppose that (xn ) is a sequence of real numbers. Then
lim sup xn = lim yn ,
n

lim inf xn = lim zn ,


n

yn = sup {xk : k n} ,
zn = inf {xk : k n} .

We may also write


lim sup xn = inf
n

nN

sup xk ,

kn

lim inf xn = sup


n

nN

inf xk .

kn

Note that lim sup xn exists and is finite provided that yn is finite and (yn ) is bounded
from below; similarly, lim inf xn exists and is finite provided that zn is finite and
(zn ) is bounded from above.
As for the limit of monotone sequences, it is convenient to allow the lim inf or
lim sup to be , and we say what this means. We have < yn for every
n N, since the supremum of a non-empty set cannot be , but it is possible
that yn ; similarly, zn < , but it is possible that zn . This leads
to the following explicit definition.

3.6. The lim sup and lim inf

43

Definition 3.34. Suppose that (xn ) is a sequence of real numbers. Then


lim sup xn =

if yn = for every n N,

lim sup xn =

if yn as n ,

lim inf xn =

if zn = for every n N,

lim inf xn =
n

if zn as n .

In all of these cases, we have zn yn for every n N, with the usual ordering
conventions for , and taking the limit as n , we get that
lim inf xn lim sup xn .
n

We illustrate the definition of the lim sup and lim inf with a number of examples.
Example 3.35. Consider the bounded, increasing sequence
1
1 2 3
xn = 1 .
0, , , , . . .
2 3 4
n
Defining yn and zn as above, we have




1
1
1
yn = sup 1 : k n = 1, zn = inf 1 : k n = 1 ,
k
k
n
and both yn 1 and zn 1 converge monotonically to the limit 1 of the original
sequence. Thus,
lim sup xn = lim inf xn = lim xn = 1.
n

Example 3.36. Consider the bounded, non-monotone sequence


1, 1, 1, 1, . . .

xn = (1)n+1 .

Then


yn = sup (1)k+1 : k n = 1,



zn = inf (1)k+1 : k n = 1,

and yn 1, zn 1 converge to different limits. Thus,


lim sup xn = 1,

lim inf xn = 1.
n

The original sequence does not converge, and lim xn is undefined.


Example 3.37. The bounded, non-monotone sequence


3 4
1
5
n+1
2, , , , . . .
1+
xn = (1)
2 3
4
n
is shown in Figure 2. We have

yn = sup {xk : k n} =
zn = inf {xk : k n} =

1 + 1/n
if n is odd,
1 + 1/(n + 1) if n is even,

[1 + 1/(n + 1)] if n is odd,


[1 + 1/n]
if n is even,

44

3. Sequences

2
1.5
1

0.5
0
0.5
1
1.5
2

10

15

20
n

25

30

35

40

Figure 2. A plot of the first 40 terms in the sequence xn = (1)n+1 (1 + 1/n)


in Example 3.37. The dashed lines show lim sup xn = 1 and lim inf xn = 1.

and it follows that


lim sup xn = 1,
n

lim inf xn = 1.
n

The limit of the sequence does not exist. Note that infinitely many terms of the
sequence are strictly greater than lim sup xn , so lim sup xn does not bound any
tail of the sequence from above. However, every number strictly greater than
lim sup xn eventually bounds the sequence from above.
Example 3.38. Consider the unbounded, increasing sequence
1, 2, 3, 4, 5, . . .

xn = n.

We have
yn = sup{xk : k n} = ,

zn = inf{xk : k n} = n,

so the lim sup, lim inf and lim all diverge to ,

lim sup xn = lim inf xn = lim xn = .


n

Example 3.39. Consider the unbounded, non-monotone sequence


(
n
if n is odd,
1, 2, 3, 4, 5, . . .
xn =
n if n is even.
We have yn = and zn = for every n N, and
lim sup xn = ,
n

lim inf xn = .
n

3.6. The lim sup and lim inf

45

The sequence oscillates and does not diverge to either or , so lim xn is


undefined even as an extended real number.
Example 3.40. Consider the unbounded non-monotone sequence
(
n
if n is odd.
1
1
1, , 3, , 5, . . .
xn =
2
4
1/n if n is even,
Then yn = and
zn =

1/n
if n even,
1/(n + 1) if n odd.

Therefore
lim sup xn = ,
n

lim inf xn = 0.
n

As these examples illustrate, the lim sup of a sequence neednt bound any tail
of the sequence. Nevertheless, the sequence is eventually bounded from above
by every number that is strictly greater than the lim sup, and the sequence is
greater infinitely often than every number that is strictly less than the lim sup.
This property gives another characterization of the lim sup, which is one that we
often use in practice.
Theorem 3.41. Let (xn ) be a real sequence. Then
y = lim sup xn
n

if and only if y satisfies one of the following conditions.


(1) < y < and for every > 0: (a) there exists N N such that xn < y +
for all n > N ; (b) for every N N there exists n > N such that xn > y .

(2) y = and for every M R, there exist infinitely many n N such that
xn > M , i.e., (xn ) is not bounded from above.
(3) y = and for every m R there exists N N such that xn < m for all
n > N , i.e., xn as n .

Similarly,
z = lim inf xn
n

if and only if z satisfies one of the following conditions.


(1) < z < and for every > 0: (a) there exists N N such that xn > z
for all n > N ; (b) for every N N there exists n > N such that xn < z + .
(2) z = and for every m R, there exist infinitely many n N such that
xn < m, i.e., (xn ) is not bounded from below.
(3) z = and for every M R there exists N N such that xn > M for all
n > N , i.e., xn as n .
Proof. We prove the result for lim sup. The result for lim inf follows by applying
this result to the sequence (xn ).
First, suppose that y = lim sup xn and < y < . Then (xn ) is bounded
from above and
yn = sup {xk : k n} y
as n .

46

3. Sequences

Therefore, for every > 0 there exists N N such that yN < y + . Since xn yN
for all n > N , this proves (1a). To prove (1b), let > 0 and suppose that N N is
arbitrary. Since yN y is the supremum of {xn : n N }, there exists n N such
that xn > yN y , which proves (1b).
Conversely, suppose that < y < satisfies condition (1) for the lim sup.
Then, given any > 0, (1a) implies that there exists N N such that
yn = sup {xk : k n} y +

for all n > N ,

and (1b) implies that yn > y for all n N. Hence, |yn y| < for all n > N ,
so yn y as n , which means that y = lim sup xn .
We leave the verification of the equivalence for y = as an exercise.

Next we give a necessary and sufficient condition for the convergence of a sequence in terms of its lim inf and lim sup.
Theorem 3.42. A sequence (xn ) of real numbers converges if and only if
lim inf xn = lim sup xn = x
n

are finite and equal, in which case


lim xn = x.

Furthermore, the sequence diverges to if and only if

lim inf xn = lim sup xn =


n

and diverges to if and only if

lim inf xn = lim sup xn =


n

Proof. First suppose that


lim inf xn = lim sup xn = x
n

for some x R. Then yn x and zn x as n where


yn = sup {xk : k n} ,

zn = inf {xk : k n} .

Since zn xn yn , the sandwich theorem implies that


lim xn = x.

Conversely, suppose that the sequence (xn ) converges to a limit x R. Then


for every > 0, there exists N N such that
It follows that

x < xn < x +

for all n > N .

x zn yn x +
for all n > N .
Therefore yn , zn x as n , so lim sup xn = lim inf xn = x.

The sequence (xn ) diverges to if and only if lim inf xn = , and then
lim sup xn = , since lim inf xn lim sup xn . Similarly, (xn ) diverges to if
and only if lim sup xn = , and then lim inf xn = .


47

3.7. Cauchy sequences

If lim inf xn 6= lim sup xn , then we say that the sequence (xn ) oscillates. The
difference
lim sup xn lim inf xn

provides a measure of the size of the oscillations in the sequence as n .


Every sequence has a finite or infinite lim sup, but not every sequence has a
limit (even if we include sequences that diverge to ). The following corollary
gives a convenient way to prove the convergence of a sequence without having to
refer to the limit before it is known to exist.
Corollary 3.43. Let (xn ) be a sequence of real numbers. Then (xn ) converges
with limn xn = x if and only if lim supn |xn x| = 0.
Proof. If limn xn = x, then




lim |xn x| = lim xn x = 0,

so lim supn |xn x| = limn |xn x| = 0.

Conversely, if lim supn |xn x| = 0, then

0 lim inf |xn x| lim sup |xn x| = 0,


n

so lim inf n xn |xn x| = lim supn xn |xn x| = 0. Theorem 3.42 implies that
limn |xn x| = 0, or limn xn = x.

Note that the condition lim inf n |xn x| = 0 doesnt tell us anything about
the convergence of (xn ).
Example 3.44. Let xn = 1 + (1)n . Then (xn ) oscillates between 0 and 2, and
lim inf xn = 0,
n

lim sup xn = 2.
n

The sequence is non-negative and its lim inf is 0, but the sequence does not converge.

3.7. Cauchy sequences


Cauchy has become unbearable. Every Monday, broadcasting the known
facts he has learned over the week as a discovery. I believe there is no
historical precedent for such a talent writing so much awful rubbish. This
is why I have relegated him to the rank below us. (Jacobi in a letter to
Dirichlet, 1841)
The Cauchy condition is a necessary and sufficient condition for the convergence of a real sequence that depends only on the terms of the sequence and not on
its limit. Furthermore, the completeness of R can be defined by the convergence of
Cauchy sequences, instead of by the existence of suprema. This defines completeness in terms of the distance properties of R rather than its order properties and
generalizes to other metric spaces that dont have a natural ordering.
Roughly speaking, a Cauchy sequence is a sequence whose terms eventually get
arbitrarily close together.

48

3. Sequences

Definition 3.45. A sequence (xn ) of real numbers is a Cauchy sequence if for


every > 0 there exists N N such that
|xm xn | <

for all m, n > N .

Theorem 3.46. A sequence of real numbers converges if and only if it is a Cauchy


sequence.
Proof. First suppose that (xn ) converges to a limit x R. Then for every > 0
there exists N N such that

for all n > N .


|xn x| <
2
It follows that if m, n > N , then
|xm xn | |xm x| + |x xn | < ,
which implies that (xn ) is Cauchy. (This direction doesnt use the completeness of
R; for example, it holds equally well for sequence of rational numbers that converge
in Q.)
Conversely, suppose that (xn ) is Cauchy. Then there is N1 N such that
|xm xn | < 1

for all m, n > N1 .

It follows that if n > N1 , then


|xn | |xn xN1 +1 | + |xN1 +1 | 1 + |xN1 +1 |.
Hence the sequence is bounded with
|xn | max {|x1 |, |x2 |, . . . , |xN1 |, 1 + |xN1 +1 |} .
Since the sequence is bounded, its lim sup and lim inf exist. We claim they are
equal. Given > 0, choose N N such that the Cauchy condition in Definition 3.45
holds. Then
xn < xm < xn +

for all m n > N .

It follows that for all n > N we have


xn inf {xm : m n} ,

sup {xm : m n} xn + ,

which implies that


sup {xm : m n} inf {xm : m n} + .
Taking the limit as n , we get that
lim sup xn lim inf xn + ,
n

and since > 0 is arbitrary, we have


lim sup xn lim inf xn .
n

It follows that lim sup xn = lim inf xn , so the sequence converges.

3.8. The Bolzano-Weierstrass theorem

49

3.8. The Bolzano-Weierstrass theorem


The Bolzano-Weierstrass theorem is a fundamental compactness result. It allows
us to deduce the convergence of a subsequence from the boundedness of a sequence
without having to know anything specific about the limit. In this respect, it is analogous to the result that a monotone increasing sequence converges if it is bounded
from above, and it also provides another way of expressing the completeness of R.
Before stating the theorem, we define and discuss subsequences. A subsequence
of a sequence (xn )
x1 , x2 , x3 , . . . , xn , . . .
is a sequence (xnk ) of the form
xn1 , xn2 , xn3 , . . . , xnk , . . .
where n1 < n2 < n3 < < nk < . . . .
Example 3.47. A subsequence of the sequence (1/n),
1 1 1 1
1, , , , , . . . .
2 3 4 5
is the sequence (1/k 2 )
1
1 1 1
,
, ....
1, , ,
4 9 16 25
Here, nk = k 2 . On the other hand, the sequences
1 1 1 1
1
1 1 1
1, 1, , , , , . . . ,
, 1, , , , . . .
2 3 4 5
2
3 4 5
arent subsequences of (1/n) since nk is not a strictly increasing function of k in
either case.
The standard short-hand notation for subsequences used above is convenient
but not entirely consistent, and the notion of a subsequence is a bit more involved
than it might appear at first sight. To explain it in more detail, we give the formal
definition of a subsequence as a function on N.
Definition 3.48. Let (xn ) be a sequence, where xn = f (n) and f : N R. A
sequence (yk ), where yk = g(k) and g : N R, is a subsequence of (xn ) if there is
a strictly increasing function : N N such that g = f . In that case, we write
(k) = nk and yk = xnk .
Example 3.49. In Example 3.47, the sequence (1/n) corresponds to the function
f (n) = 1/n, the subsequence (1/k 2 ) corresponds to g(k) = 1/k 2 , and (k) = k 2 .
Note that since the indices in a subsequence form a strictly increasing sequence
of integers (nk ), we have nk as k .
Proposition 3.50. Every subsequence of a convergent sequence converges to the
limit of the sequence.
Proof. Suppose that (xn ) is a convergent sequence with limn xn = x and (xnk )
is a subsequence. Let > 0. There exists N N such that |xn x| < for all
n > N . Since nk as k , there exists K N such that nk > N if k > K.
Then k > K implies that |xnk x| < , so limk xnk = x.


50

3. Sequences

A useful criterion for the divergence of a sequence follows immediately from


this result and the uniqueness of limits.
Corollary 3.51. If a sequence has subsequences that converge to different limits,
then the sequence diverges.
Example 3.52. The sequence ((1)n+1 ),
1, 1, 1, 1, 1, . . . ,

has subsequences (1) and (1) that converge to different limits, so it diverges.
In general, we define the limit set of a sequence to be the set of limits of its
convergent subsequences.
Definition 3.53. The limit set of a sequence (xn ) is the set
S = {x R : there is a subsequence (xnk ) such that xnk x as k }

of limits of all of its convergent subsequences.

The limit set of a convergent sequence consists of a single point, namely its
limit.
Example 3.54. The limit set of the divergent sequence ((1)n+1 ),
1, 1, 1, 1, 1, . . . ,

contains two points, and is {1, 1}.

Example 3.55. Let {rn : n N} be an enumeration of the rational numbers


in [0, 1]. Every x [0, 1] is a limit of a subsequence (rnk ). To obtain such a
subsequence recursively, choose n1 = 1, and for each k 2 choose a rational
number rnk such that |x rnk | < 1/k and nk > nk1 . This is always possible since
the rational numbers are dense in [0, 1] and every interval contains infinitely many
terms of the sequence. Conversely, if rnk x, then 0 x 1 since 0 rnk 1.
Thus, the limit set of (rn ) is the interval [0, 1].
Finally, we state a characterization of the lim sup and lim inf of a sequence in
terms of of its limit set. We leave the proof as an exercise.
Theorem 3.56. Suppose that (xn ) sequence of real numbers with limit set S.
Then
lim sup xn = sup S,
lim inf xn = inf S,
n

where we use the usual conventions about .


We now come to the main result in this section.
Theorem 3.57 (Bolzano-Weierstrass). Every bounded sequence of real numbers
has a convergent subsequence.
Proof. Suppose that (xn ) is a bounded sequence of real numbers. Let
M = sup xn ,
nN

and define the closed interval I0 = [m, M ].

m = inf xn ,
nN

51

3.8. The Bolzano-Weierstrass theorem

Divide I0 = L0 R0 in half into two closed intervals, where


L0 = [m, (m + M )/2],

R0 = [(m + M )/2, M ].

At least one of the intervals L0 , R0 contains infinitely many terms of the sequence,
meaning that xn L0 or xn R0 for infinitely many n N (even if the terms
themselves are repeated).
Choose I1 to be one of the intervals L0 , R0 that contains infinitely many terms
and choose n1 N such that xn1 I1 . Divide I1 = L1 R1 in half into two
closed intervals. One or both of the intervals L1 , R1 contains infinitely many terms
of the sequence. Choose I2 to be one of these intervals and choose n2 > n1 such
that xn2 I2 . This is always possible because I2 contains infinitely many terms
of the sequence. Divide I2 in half, pick a closed half-interval I3 that contains
infinitely many terms, and choose n3 > n2 such that xn3 I3 . Continuing in this
way, we get a nested sequence of intervals I1 I2 I3 . . . Ik . . . of length
|Ik | = 2k (M m), together with a subsequence (xnk ) such that xnk Ik .
Let > 0 be given. Since |Ik | 0 as k , there exists K N such
that |Ik | < for all k > K. Furthermore, since xnk IK for all k > K we have
|xnj xnk | < for all j, k > K. This proves that (xnk ) is a Cauchy sequence, and
therefore it converges by Theorem 3.46.

The subsequence obtained in the proof of this theorem is not unique. In particular, if the sequence does not converge, then for some k N the left and right
intervals Lk and Rk both contain infinitely many terms of the sequence. In that
case, we can obtain convergent subsequences with different limits, depending on
our choice of Lk or Rk . This loss of uniqueness is a typical feature of compactness
proofs.
We can, however, use the Bolzano-Weierstrass theorem to give a criterion for
the convergence of a sequence in terms of the convergence of its subsequences. It
states that if every convergent subsequence of a bounded sequence has the same
limit, then the entire sequence converges to that limit.
Theorem 3.58. If (xn ) is a bounded sequence of real numbers such that every
convergent subsequence has the same limit x, then (xn ) converges to x.
Proof. We will prove that if a bounded sequence (xn ) does not converge to x, then
it has a convergent subsequence whose limit is not equal to x.
If (xn ) does not converges to x then there exists 0 > 0 such that |xn x| 0
for infinitely many n N. We can therefore find a subsequence (xnk ) such that
|xnk x| 0

for every k N. The subsequence (xnk ) is bounded, since (xn ) is bounded, so by


the Bolzano-Weierstrass theorem, it has a convergent subsequence (xnkj ). If
lim xnkj = y,

then it follows that |x y| 0 , so x 6= y. This contradicts our original assumption


and proves the result.


You might also like