11 Uniform convergence, Taylor series, Com-

plex series
In this chapter we will lay down the basic results underlying the convergence
pointwise and uniform of innite series of functions. No proofs will be
given, but the student should feel free to make use of them, of course taking
care to apply them only when the relevant hypotheses are satised. We
will use this theory to study the ever important Taylor series expansions
of various functions, including the exponential, trigonometric and simple
rational functions, as well as the logarithm. We will at the end extend these
functions to the domain of complex numbers to the extent possible.
11.1 Innite series of functions, convergence
Let {f
n
} be a sequence of functions on a subset X of R or C. We will say
that this sequence is pointwise convergent with limit f on X i for every
x X, the sequence {f
n
(x)} of numbers converges to f(x). In other words,
for every > 0, there is a positive number N(x) such that
(11.1.1) n > N(x) = |f(x) f
n
(x)| < .
It is natural to wonder if N(x) can be taken to be a number N, say, which
is independent of x. In such a case, we will say that f
n
is uniformly
convergent with limit f on X.
Uniform convergence is a very important concept for Calculus. To elab-
orate, we have the following useful result:
Theorem 11.1.2 Let {f
n
} be a sequence of functions on a nite interval I
in R, and let f be a function on the same interval I. Then
(a) (Integration) If f
n
f uniformly on I, and if f
n
, f are integrable
on I, then

I
f(x)dx = lim
n

I
f
n
(x)dx.
(b) (Continuity) If f
n
f uniformly on I, and if the f
n
are continuous,
then f is also continuous.
1
(c) (Dierentiability) Suppose {f
n
} converge pointwise to f, where each
f
n
is dierentiable with f

n
integrable. Further assume that the sequence
{f

n
} converges uniformly to a continuous function on I. Then f is
dierentiable and
f

(x) = lim
n
f

n
(x),
for all x I.
Now consider any innite sum of functions on a subset X of R or C,
given as
(11.1.3)

n=0
f
n
.
To know if this makes sense, we need to look at the partial sums s
N
dened
by
(11.1.4) s
N
=
N

n=0
f
n
.
We will say that the sum converges uniformly, resp. pointwise, to a function
S on X i the sequence {s
N
} converges to S uniformly, resp. pointwise, on
X.
Theorem 11.1.2 has a natural analog for innite sums.
Theorem 11.1.5 Suppose we are given

n=0
f
n
, S on an interval J in R.
(a) (Integrability) If

n=0
f
n
converges uniformly to S in J, and if the
f
n
, f are integrable there, then

J
S(x)dx =

n=1

J
f
n
(x)dx.
(b) (Continuity) If

n=0
f
n
converges uniformly to S in J, and if the f
n
are continuous, the so is S.
2
(c) (Dierentiability) Suppose

n=1
f
n
converges pointwise to S, where
each f
n
is dierentiable with f

n
integrable. Further assume that the sum

n=1
f

n
converges uniformly to a continuous function T on J. Then S
is dierentiable and
S

(x) =

n=1
f

n
(x),
for all x I.
Here is a very helpful test to determine if a given innite series converges
uniformly or not.
Theorem 11.1.6 (Weierstrasss test) Let {f
n
} be a sequence of functions
on an interval J in R, and let {M
n
} be a sequence of positive numbers such
that
(i) |f
n
(x)| M
n
for all n, and for all x; and
(ii) the innite series of numbers

n=1
M
n
converges.
Then
(a) For all x J, the series

n=1
f
n
(x) converges absolutely; and
(b) the sum

n=1
f
n
converges uniformly on J to the function
S(x) =

n=1
f
n
(x).
Here is an example illustrating the usefulness of this test. For any integer
k 2, consider the series
(11.1.7)

n=1
sin nx
n
k
.
3
Put
M
n
=
1
n
k
.
Then, since sin nx is always bounded between 1 and 1,

sin nx
n
k

M
n
n 1.
Moreover, since k > 1,

n=1
M
n
converges. So we may apply Weierstrasss test and conclude that the series
(11.1.7) converges.
Note that if we take the above series, but with k = 1, then the test does
not apply as the sum of 1/n diverges. But that does not mean that the series

n1
sin nxn
1
is divergent for all x. It only means that the method does not
apply. In fact one can show, with some trouble, that this series is convergent
for x in [t, t], for any t > 0. Of course it is problematic when x is a
multiple of .
11.2 Taylor series
We studied in chapter 10 Taylor polynomials p
n
(f(x); a) of nice functions f
at x = a. If f is innitely dierentiable at x = a, we might want to consider
the limit of p
n
(f(x); a) as n tends to . The rst problem is that this series,
called the Taylor series of f at x = a, may not converge. The second is that
even when it converges, it might not represent the function f near x = a.
Finally, assuming the rst two problems have armative solutions, which is
equivalent to requiring that the sequence of remainders R
n
(f(x); a) converges
to 0 as n , we need to ascertain when we can dierentiate (or integrate)
the Taylor series term by term to get the derivative (or the integral) of f.
The key to understanding these aairs is the notion of uniform convergence
which we introduced in the previous section.
A series of the form
(11.2.1)

n=0
a
n
x
n
4
is called a power series.
Since the power functions x x
n
are integrable and dierentiable, in
fact any number of times, we may specialize Theorem 11.1.5 to the case of
power series and obtain the following:
Theorem 11.2.2 For any sequence {a
n
} of real numbers, put
S(x) =

n=0
a
n
x
n
(i) If the series is uniformly convergent in an interval J, then S is inte-
grable on [a, b] and

J
S(x)dx = lim
n
a
n

J
x
n
dx.
(ii) Suppose the series converges pointwise, and more importantly, its series
of derivatives, namely
T(x) =

n=1
na
n
x
n1
converges uniformly in [a, b]. Then S is dierentiable with derivative
T.
A particular kind of power series, which is of great utility, is the so called
Taylor series (at 0), which is of the form
(11.2.3)

n=0
f
(n)
(0)
n!
x
n
where f is an innitely dierentiable function.
Theorem 11.2.2 says that if a power series is uniformly convergent we can
integrate it term by term to get its integral, and if the series of derivatives
is also uniformly convergent, then we can dierentiate it term by term to
get the derivative. So, roughly speaking, when we have uniform convergence
of the appropriate power series we are allowed to integrate and dierentiate
term by term. What is left to do is to nd a criterion to know when a power
5
series is uniformly convergent. We also want this criterion to tell us when
the series and the formally derived series both converge uniformly. Here is
the completely satisfactory result giving us what we need:
Theorem 11.2.4 Let {a
n
} be a sequence of real numbers. Suppose there is
a positive real number c such that

n=0
a
n
c
n
converges. Pick any positive number b < c. Then each of the power series
S(x) =

n=0
a
n
x
n
and
T(x) =

n=1
na
n
x
n1
converges absolutely and uniformly in [b, b]. Moreover,
S

(x) = T(x) x (c, c).

The proof of this result, which we will skip, makes clever use of the
Weierstrass test and Theorem 11.2.2.
Let us try to understand the exponential function using this. We have
seen that
e
x
= 1 + x +
x
2
2!
+ . . . +
x
n
n!
+ R
n
(e
x
; 0),
with
|R
n
(e
x
; 0)| e
x
|x|
n+1
(n + 1)!
.
Lemma 11.2.5 Let t be any positive real number. Then we have
lim
n
t
n
n!
= 0.
Granting this Lemma (for the moment) we see for any positive c, the
remainder term R
n
(e
c
; 0) goes to zero as n goes to innity. This gives us the
convergent innite series expression
e
c
= 1 + c +
c
2
2!
+
c
3
3!
+ . . .
6
Applying Theorem 11.2.4 we see then that the Taylor series
(11.2.6) e
x
= 1 + x +
x
2
2!
+
x
3
3!
+ . . .
converges absolutely and uniformly on [c, c] for any c > 0. Moreover, its
derivative is given by dierentiating the series term by term, which gives
back e
x
, as expected.
One proves Lemma 11.2.5 by appealing to a famous limit formula in
Mathematics involving the factorial function, namely the following:
Stirlings formula One has
lim
n
n!
n
n+1/2
e
n
=

2.
For a proof of this formula see the posting on the class website (under
notes).
We can write
t
n
n!
=

n
n+1/2
e
n
n!

u
n
n
n+1/2

where u = et. So, thanks to Stirling and the easy fact that
1
n
1/2
goes to 0 as
n , Lemma 11.2.5 will follow from knowing that
u
n
n
n
< 1
for n suciently large. (If u < 1, u
n
goes to zero as n , so we may
assume, if necessary, that u is greater than 1.) Applying the (one-to-one
function) logarithm to both sides, we see that we need to prove that
nlog u nlog n < 0,
i.e., that log u < log n, which holds as soon as n > u. Done.
By a similar argument, we also get the Taylor series expressions for the
sine and cosine functions, valid for all x:
(11.2.8 i) sin x = x
x
3
3!
+
x
5
5!

x
7
7!
+ . . .
7
(11.2.8 ii) cos x = 1
x
2
2!
+
x
4
4!

x
6
6!
+ . . .
We can also apply Theorem 11.2.4 and dierentiate the expression for sin x
term by term, and as expected, one gets the series expression for cos x.
The series
(11.2.9) arctan x = x
x
3
3
+
x
5
5
. . .
is, as we saw before, convergent at x = 1. Thus by Theorem 11.2.4, it
converges absolutely in 1 < x < 1 and uniformly in [b, b] for any positive
b < 1.
We claim that this Taylor series for arctan x does not converge at any
x > 1. Indeed, if it did, then by Theorem 11.2.4, it would converge absolutely
at x = 1 which it does not. So the number 1, which some call the radius
of convergence of the Taylor series (11.2.9), is the boundary beyond which
there is no convergence.
Taking the derivative of the series (11.2.9) term by term, and applying
Theorem 11.2..4, we get (as expected)
(11.2.10)
1
1 + x
2
= 1 x
2
+ x
4
x
6
+ . . . ,
which is valid in {|x| < 1}. At x = 1 the series makes no sense, but the
left hand side makes sense there and equals
1
2
.
Incidentally, the Taylor series for arctan x converges at x = 1, with
arctan(1) = arctan 1 =

4
.
It should be noted, despite what the arctan function may suggest, that in
Theorem 11.2.4, the absolute convergence of S(x) is asserted only for |x| < c,
and one can say nothing in general about the convergence, or the lack of it,
at x = c. Sometimes it does happen that S(x) is divergent at c. The
simplest example illustrating this is the Taylor series for log(1 +x) at x = 0,
which we claim to be
(11.2.11) log(1 + x) = x
x
2
2
+
x
3
3

x
4
4
+ . . .
absolutely convergent in 1 < x < 1 and uniformly convergent in [b, b] for
any positive b < 1. One way to see this is the following:
8
Since log(1 + x) is (dened and) dierentiable in x > 1 with deriva-
tive
1
1+x
, we could nd the Taylor series expansion of
1
1+x
rst and then
dierentiate.
The identity (11.2.10) implies that
(11.2.12)
1
1 + x
= 1 x + x
3
x
4
+ . . .
which is absolutely convergent in 0 x < 1. Then it is also absolutely
convergent in (1, 1). Pick any positive b < 1. Then since the series (11.2.12)
converges at c for any c with b < c < 1. So we may apply Theorem 11.2.4
and conclude that (11.2.12) converges uniformly in [b, b]. We can also
dierentiate term by term, by this Theorem, and obtain the desired expansion
(11.2.11), which is valid in 1 x 1.
At x = 1, the series (11.2.11) is alternating and converges to log 2. But
it is important to note, however, that it is divergent at x = 1.
A consequence of (11.2.11) is the series for log(1 x), given by
(11.2.13) log(1 x) = log

1
1 x

= x +
x
2
2
+
x
3
3
+ . . .
A very useful thing to remember, when computing the Taylor series of
various functions is the following
Lemma 11.2.14 For j = 1, 2, suppose the power series
S
j
(x) =

n=0
a
n,j
x
n
are both uniformly convergent in some [a, b]. Then for all scalars , , the
power series

n=0
(a
n,1
+ a
n,2
)x
n
converges uniformly in [a, b] to S
1
(x) + S
2
(x).
In particular, the sum or dierence of two uniformly convergent power
series is again uniformly convergent.
9
11.3 Complex power series
It is very important to understand the power series
(11.3.1)

n=0
a
n
z
n
where the coecients a
n
are complex numbers and z is in C.
We will say that such a series is absolutely convergent if the associated
(non-negative) real series
(11.3.2)

n=0
|a
n
| |z|
n
converges. For example, the complex exponential function
(11.3.3) e
z
= 1 + z +
z
2
2!
+
z
3
3!
+ . . .
is absolutely convergent for all z because the corresponding series of absolute
values, namely
1 +|z| +
|z|
2
2!
+
|z|
3
3!
+ . . .
is convergent, it being the Taylor series of the real exponential e
|z|
.
Here is the main result of this section.
Theorem 11.3.4 Suppose the complex power series
S(z) =

n=0
a
n
z
n
converges at some non-zero complex number z
0
. Then either this series is
absolutely convergent for all z in C, or there is a positive real number R > 0
such that
S(z) is absolutely convergent in {|z| < R}; and
S(z) diverges for any z with |z| > R.
10
Moreover, S(z) is dierentiable at any z with |z| < R with derivative
S

(z) =

n=1
na
n
z
n1
.
R is called the radius of convergence of the power series. Note that the
Theorem says nothing about the convergence at the points z with |z| = R.
Some people would say that the radius of convergence is innite when S(z)
is absolutely convergent everywhere in C.
One can dene the complex extensions of the various functions we have
encountered like the sine and cosine functions in terms of their power series.
To be specic we have
(11.3.5) sin z = z
z
3
3!
+
z
5
5!

z
7
7!
+ . . .
(11.3.6) cos z = 1
z
2
2!
+
z
4
4!

z
6
6!
+ . . .
Both these power series have innite radii of convergence.
It is crucial to note, by comparing the power series expansions, that
(11.3.7) e
iz
= cos z + i sin z.
This leads to the identities
(11.3.8) sin z =
e
iz
e
iz
2i
and
(11.3.9) cos z =
e
iz
+ e
iz
2
.
We will be remiss if we forget the complex logarithm
(11.3.10) log(1 + z) = z
z
2
2
+
z
3
3

z
4
4
+ . . .
which is absolutely convergent in {|z| < 1}. We have already seen that it
converges at z = 1 and diverges at z = 1.
11
What we now have a is a very interesting situation for log(1 + z). On
the one hand, it makes sense at any complex number z with |z| < 1. On the
other, it makes sense whenever 1 + z is a positive real number, though it is
not given by its Taylor series (11.3.10) if |z| > 1. Thus log(1 + z) is dened
on the following strange-shaped subset of C:
(11.3.11) {z C| |z| < 1 or z (1, ) R}.
The natural question which arises immediately is whether this function can
be given sense at all points of C, or at least on a larger subset. The quick
answer is that it can be dened on the complement in C of the half-line
(, 1] R. On all of C it can only be dened as a multi-valued function.
To understand all these intriguing matters better, one will need to understand
a fundamental mathematical concept called analytic continuation, and for
that one will need to take a course in Complex Variables.
12