CM221A ANALYSIS I NOTES ON WEEK 3

UPPER AND LOWER LIMITS

One says that + is an accumulation point of a sequence {a
n
} if there is a subse-
quence of {a
n
} which converges to +. Similarly, is an accumulation point
of {a
n
} if there is a subsequence which converges to . This convention makes
sense, even though are not proper numbers. Note that an unbounded sequence
must have a subsequence which converges either to +, or to . Therefore the
BolzanoWeierstrass theorem implies the following
Corollary. Every sequence has a subsequence which converges either to a nite
limit, or to .
Denition. The largest accumulation point of a sequence {a
n
} is called the upper
limit of {a
n
} and is denoted limsup a
n
. The smallest accumulation point of {a
n
}
is called the lower limit of {a
n
} and is denoted liminf a
n
.
In view of the above corollary, the upper and lower limits always exist, even if
the sequence does not converge. If the sequence is bounded, both upper and
lower limits are nite numbers. If the sequence is not bounded from above then
limsup a
n
= +. If the sequence is not bounded from below then liminf a
n
= .
Example. If a
n
= n then limsup a
n
= liminf a
n
= +. If a
n
= n then
limsup a
n
= liminf a
n
= .
Theorem. A sequence {a
n
} converges to a number c if and only if limsup a
n
=
liminf a
n
= c.
Proof. If a
n
c then c is the unique accumulation point of {a
n
} (see Week 2).
Thus it is the smallest and the largest accumulation point at the same time, so
that limsup a
n
= liminf a
n
= c.
On the other hand, if limsup a
n
= liminf a
n
= c then c is the only accumulation
point of {a
n
}. This implies that there are only nitely many a
n
outside any interval
of the form (c , c + ) (otherwise these elements would form a sequence lying
outside (c , c + ), which would have another accumulation point). This means
that a
n
c (see Week 2).
Exercise. Show that the above theorem remains valid for c = + and c =
Warning. It may well happen that limsup(a
n
+ b
n
) = limsup a
n
+ limsup b
n
and/or liminf(a
n
+ b
n
) = liminf a
n
+ liminf b
n
. For instance, if a
n
= (1)
n
and
b
n
= (1)
n+1
then limsup a
n
= limsup b
n
= 1, liminf a
n
= liminf b
n
= 1 but
a
n
+ b
n
= 0 for all n.
Lemma. If r > 0 then limsup(r a
n
) = r limsup a
n
and liminf(r a
n
) = r liminf a
n
.
If r < 0 then limsup(r a
n
) = r liminf a
n
and liminf(r a
n
) = r limsup a
n
.
1
Proof. Let r = 0. From the algebraic rules for limits it follows that a subsequence
{a
n
k
} converges to a limit a if and only if {r a
n
k
} converges to the limit r a. There-
fore c is an accumulation point of {a
n
} if and only if r c is an accumulation point
of the sequence {r a
n
}. In other words, the accumulation points of the sequence
{r a
n
} are obtained from the accumulation points of {a
n
} by multiplying them
by the number r. If r > 0 then the multiplication maps the largest and smallest
accumulation points of {a
n
} into the largest and smallest accumulation points of
{r a
n
}. If r < 0 then the multiplication transfers the largest accumulation point of
{a
n
} into the smallest accumulation point of {r a
n
}, and the other way round.
Theorem. Let s
k
= sup
nk
a
n
. Then the sequence {s
k
} converges (to a nite
number or ) and lims
k
= limsup a
n
.
Proof. If the sequence {a
n
} is not bounded from above then limsup a
n
= + and
s
k
= + for all k.
Assume that {a
n
} is bounded from above. By denition, s
k
is the least upper bound
for the set {a
k
, a
k+1
, a
k+2
, . . .}. Obviously, it is also an upper bound for the smaller
set {a
k+1
, a
k+2
, . . .}. Since s
k+1
is the least upper bound for {a
k+1
, a
k+2
, . . .}, it
follows that s
k+1
s
k
. Thus {s
k
} is a nonincreasing sequence. By the monotone
convergence theorem, it converge either to a nite number or to .
If s
k
then a
n
because a
k
s
k
. In this case limsup a
n
= lima
n
=
= lims
k
.
If s
k
converge to a nite number c, it must be an accumulation point. Indeed, if c
is not an accumulation point then there is an interval (c , c +) which contains
only nitely many elements a
n
. This implies that the elements a
k
, a
k+1
, a
k+2
, . . .
lie outside this interval for all suciently large k. It follows that s
k
(c , c +)
and, consequently, s
k
cannot converge to c.
Finally, if c is not the largest accumulation point then the sequence {a
n
} has
another accumulation point b such that b > c. Let > be small enough, so that
c < b . Since b is an accumulation point, there are innitely many elements a
n
in the interval (b , b + ). But then, for every k, the set {a
k
, a
k+1
, a
k+2
, . . .} has
innitely many elements in the interval (b , b +). It follows that its least upper
bound s
k
cannot be smaller than b . Consequently, s
k
does not converge to c.
The obtained contradiction shows that c is the largest accumulation point.
In a similar way one can prove
Theorem. Let r
k
= inf
nk
a
n
. Then the sequence r
k
converges (to a nite number
or ) and limr
k
= liminf a
n
.
SERIES
Series is just an innite sum of the form

n=1
a
n
or a
1
+ a
2
+ a
3
+ . . . , where
a
n
R.
Denition. We say that the series converges and write

n=1
a
n
= c if the se-
quence of nite partial sums s
k
=

k
n=1
a
n
converges to c. Equivalently
> 0 k

: k k

n=1
a
n
c

< .
2
We say that a series is divergent if it is not convergent.
Example. The series 1 1 + 1 1 + 1 . . . does not converge because its partial
sums oscillate between -1 and 1.
Example. An innite decimal fraction is an example of a convergent series. In-
deed, when we write a = 0.a
1
a
2
a
3
. . . we actually mean that a =

n=1
a
n
10
n
.
Remark. The series

n=1
a
n
converges if and only if the series

n=m
a
n
con-
verges. In other words the rst few terms do not aect convergence, even though
they change the value of the sum. Indeed, if s
k
are the partial sums of the original
series then the partial sums
k
of the series

n=m
a
n
are equal to s
k
a where
a =

m1
n=1
a
n
. From the denition of convergence for sequences, it follows that the
sequence {s
k
} converges if and only if the sequence {
k
} converges.
Example (geometric progression). If |b| < 1 then the series

n=0
b
n
converges
and

n=0
b
n
= (1 b)
1
. Indeed, by induction in m we obtain
m

n=0
b
n
= (1 b)
1
b
m+1
(1 b)
1
.
If |b| < 1 then b
m+1
0 and, consequently,
m

n=0
b
n
(1 b)
1
as m .
Theorem. If the series

n=1
a
n
converges then lim
n
a
n
= 0.
Proof. Since the series converges, its partial sums s
k
form a convergent sequence
{s
k
}. By Cauchys theorem, {s
k
} is a Cauchy sequence, that is, |s
m
s
k
| 0 as
k, m . Taking k = m1, we see that a
m
= s
m
s
m1
0 as m .
Remark. The converse is FALSE: it may well happen that lim
n
a
n
= 0 but
the series

n=1
a
n
diverges. For instance, the series

n=1
1/n diverges.
Theorem. If

n=1
a
n
and

n=1
b
n
are convergent series then

n=1
(a
n
+ b
n
) =

n=1
a
n
+

n=1
b
n
and

n=1
c a
n
= c

n=1
a
n
for all c R.
Proof. Let s
m
=

m
n=1
a
n
and
m
=

m
n=1
b
n
. Then from the algebraic rules for
limits it follows that
lim
m
m

n=1
(a
n
+ b
n
) = lim
m
(s
m
+
m
) = lim
m
s
m
+ lim
m

m
=

n=1
a
n
+

n=1
b
n
,
which means that

n=1
(a
n
+ b
n
) =

n=1
a
n
+

n=1
b
n
. Similarly,

n=1
c a
n
= lim
m
m

n=1
c a
n
= lim
m
c s
n
= c lim
m
s
n
= c

n=1
a
n
.
3
ABSOLUTELY CONVERGENT SERIES
Denition. A series

n=1
a
n
is said to be absolutely convergent if the positive
series

n=1
|a
n
| converges.
Instead of saying that the positive series is absolutely convergent, one often writes

n=1
|a
n
| < . Obviously, a positive series

n=1
a
n
is convergent if and only if
it is absolutely convergent (in this case a
n
= |a
n
|).
Absolute Convergence Theorem. Every absolutely convergent series is con-
vergent. Or: if

n=1
|a
n
| < then

n=1
a
n
converges.
Proof. Let s
k
=

k
n=1
a
n
and
k
=

k
n=1
|a
n
|. If m k then, using the triangle
inequality for the modules, we obtain
|s
m
s
k
| =

n=k+1
a
n

n=k+1
|a
n
| =
m

k
.
Since the series

n=1
|a
n
| converges, the sequence {
k
} converges. Therefore it is
a Cauchy sequence, that is,
m

k
0 as m, k . The above inequality
implies that |s
m
s
k
| 0 as m, k , which means that {s
k
} is also a Cauchy
sequence. Finally, by Cauchys theorem, the sequence {s
k
} converges.
Comparison Theorem for Series. If |a
n
| b
n
for all suciently large n and
the series

n=1
b
n
is convergent then the series

n=1
a
n
is absolutely convergent.
Proof. The words for all suciently large n mean that |a
n
| b
n
for all n n
0
,
where n
0
is some xed positive integer.
Denote
k
=

k
n=1
|a
n
| and

k
=

k
n=1
b
n
. Then 0 (
m

k
) (

k
)
whenever m k n
0
. The right hand side of this inequality converges to 0 as
m, k because the sequence {

k
} is convergent (Cauchys theorem). There-
fore |
m

k
| 0 as m, k and, consequently, {
k
} is a Cauchy sequence.
By Cauchys theorem, the sequence {
k
} converges, which means that the series

n=1
a
n
is absolutely convergent.
Denition. We say that the integral

1
f(x) dx converges and

1
f(x) dx = c
if lim
k

k
1
f(x) dx

= c, where c is a nite number.

Integral Comparison Theorem. If f(x) is a positive nonincreasing function on
[1, ) then the sum

n=1
f(n) converges if and only if the integral

1
f(x) dx
converges. (This is not stating that

n=1
f(n) =

1
f(x) dx.)
Proof. Denote c
n
=

n+1
n
f(x) dx . Obviously,

k
0
f(x) dx =
k1

n=1

n+1
n
f(x) dx =
k1

n=1
c
n
.
It follows that the integral

0
f(x) dx converges if and only if the series

k
n=1
c
n
converges.
4
Since f is nonincreasing, f(n + 1) f(x) f(n) for all x [n, n + 1]. Therefore
we obviously have
f(n + 1) =

n+1
n
f(n + 1) dx

n+1
n
f(x) dx
and
f(n) =

n+1
n
f(n) dx

n+1
n
f(x) dx ,
that is, f(n + 1) c
n
f(n).
If the series

n=1
f(n) converges then, by the comparison theorem, the series

k
n=1
c
n
is also convergent. On the other hand, the series

k
n=1
c
n
converges then,
by the comparison theorem, the series

n=1
f(n + 1) converges. The latter is
equivalent to the convergence of the series

n=1
f(n) because

n=1
f(n + 1) = f(2) + f(3) + f(4) + . . . =

n=2
f(n).
Remark. The obvious estimates and equalities used in the proof will be dis-
cussed in the end of the course.
Example. The series

n=1
1/n

converges if > 1 and diverges if 1.

5