Analysis III: Integration

Easter 2007
Professor Terry Lyons
F19, Mathematical Institute
tlyons@maths.ox.ac.uk
March 27, 2008
1
In these lectures we dene a simple integral and study its properties; prove
the Mean Value Theorem for Integrals and the Fundamental Theorem of Cal-
culus. This gives us the tools to justify term by term dierentiation of power
series and deduce the elementary properties of the trigonometric functions.
0.1 Synopsis
Step functions, their integral, basic properties. The application of uniform con-
tinuity to approximate continuous functions above and below by step functions.
The integral of a continuous function on a closed bounded interval. Elemen-
tary properties of the integral of a continuous function: positivity, linearity,
subdivision of the interval.
The Mean Value Theorem for Integrals. The Fundamental Theorem of Cal-
culus; linearity of the integral, integration by parts and substitution.
The interchange of integral and limit for a uniform limit of continuous func-
tions on a bounded interval. Term by term integration and dierentiation of a
(real) power series (interchanging limit and derivative for a series of functions
where the derivatives converge uniformly); examples to include the derivation
of the main relationships between exponential, trigonometric functions and hy-
perbolic functions.
0.2 Reading
T. Lyons Lecture Notes.
J. Roe, Integration, Mathematical Institute Notes (1994).
H. A. Priestley, Introduction to Integration, Oxford Science Publications
(1997), Chapters 1-8. (These chapters commence with a useful summary
of background cont and di and go on to cover not only the integration
but also the material on power series.
Robert G. Bartle, Donald R. Sherbert, Introduction to Real Analysis,
Third Edition (2000), Wiley, Chapter 8.
2
1 Integrating continuous functions
The space of functions mapping an interval into the reals
f : J R
is huge! We are obviously familiar with those such as x x
2
which are given
by a formula but in some sense these are a minority. Even simple functions such
as
(x) = [x] = max {n | n x}
(which is always dened because of the standard axioms for the real numbers)
generated considerable controversy when they were rst introduced.
Many functions cannot be graphed very easily if at all. For example
f
_
p
q
_
=
1
q
, (p, q) = 1, p, q Z, q > 0
f (x) = 0, x R\Q
f (0) = 0.
Our goal in this course is to identify a sensible class of functions that can be
integrated. Having done this, we must make it happen, make sense of
_
b
a
f (x) dx
and prove that it has all the obvious properties, such as linearity. I hope that
you will be surprised by the way that we are able to move relatively smoothly
from rather primitive notions to quite sophisticated information about the basic
functions, such as the exponential, logarithm, and the trigonometric functions
sin, cos etc..
Integration will provide one of the rst serious contexts you will meet where
we examine, in a rigorous way, functions on functions!
In order to dened the integral we are going to consider two basic classes of
functions
1. Step Functions: The functions f dened on an interval I, for which there
is a nite partition P
f
I so that f is constant on each remaining (open)
interval.
2. The continuous functions on a closed interval J.
If we x J then both of these spaces of functions are real vector spaces.
1.1 Step functions and intervals
Denition 1.1 A non-empty subset I of R is an interval if, whenever x I,
y I, and
x < z < y, z R
then z I.
3
Denition 1.2 The endpoints of interval I are dened to be b = sup I and
a = inf I. It is allowed and can happen that b = + or a = .
Because an interval I is always nonempty we may choose x I and, recalling
the denition of inf I as the greatest lower bound, concluded that a x and
similarly that x b. It follows that a b. With similar eort (exercise) and
using the denition of inmum and the basic property of an interval, one may
prove that (a, b) I.
Lemma 1.3 Any interval is of the form
[a, b) , (a, b] , [a, b] , (a, b)
[a, ) , (a, ) , (, b] , (, b) , R
for some a b R. In any case where one of these sets is non-empty, and in
particular when a < b, any such set is an interval.
Exercise 1.4 If I and J are intervals, and J contains no endpoint of I then
either J I or J and I are disjoint.
Exercise 1.5 The intersection of two intervals is an interval or it is empty.
Denition 1.6 A partition P of an interval I with endpoints a and b (possibly
) is a nite sequence (a
i
)
k
i=0
so and that
a
0
= a < a
1
< a
2
< < a
k
= b.
The compliment of a partition P in the interval I , and written I\P, is the
set
I\P =
_
x | x
k
j=1
(a
j1
, a
j
)
_
,
and is a nite union of disjoint open intervals.
Denition 1.7 A function dened on an interval I is a step function if there
is a partition P = (a
i
)
k
i=0
of the interval so that is constant on each interval
(a
j1
, a
j
), j k. Any particular choice P = (a
i
)
k
i=0
of partition for the interval
I with this property is called a witness.
If is a step function then there are, in general, innitely many partitions
witnessing this fact. The class of Roe step functions and the class we have
dened here correspond.
Remark 1.8 By considering the intervals [a
j
, a
j
] as well as the (a
j1
, a
j
) one
sees that a function dened on an interval I is a step function if and only if
one can decompose the interval into nitely many disjoint subintervals so that
is constant on each of them. The decomposition is not unique.
Lemma 1.9 Any step function dened on an interval I can be extended to a
step function

dened on the interval R by making it zero o I.
4
Proof. We treat the case of a bounded interval; the other cases are similar. It
is enough to add the points to a partition
P = (a
0
= a < a
1
< a
2
< < a
k
= b)
witnessing the fact that is a step function on I. By denition is zero on
(b, +) and (, a) and so is constant there.

P = ( < a
0
= a < a
1
< a
2
< < a
k
= b < )
is a partition demonstrating that

is a step function.
Denition 1.10 If P is a partition of an interval I and

P is a nite set of
points satisfying P

P I then we call

P renement of P.
Proposition 1.11 If is a step function on I and P is a partition that gives
witness to this, and if

P is a renement of P, then

P is also a witness to the
fact that is a step function.
Suppose is a step function on I witnessed by the partition P, and that
a second partition

P is a renement of P so that P

P. Then J\

P J\P.
Consider one of the disjoint open intervals

N which together comprise J\

P.
It must intersect one of the disjoint open intervals N that make up J\P. By
construction

N does not contain any end point of any such an interval. So using
exercise 1.4 it must be properly contained in N since it is not disjoint. But is
constant on N and so constant on

N.
Denition 1.12 If A R is a set, then the indicator function
A
of A is the
function

A
(x) = 1, x A

A
(x) = 0, x = A.
If I is an interval then it is clear that its indicator function
I
(x) is a step
function.
If f is a step function on J and R then it is obvious that f is also a
step function on J.
Proposition 1.13 The sum of two step functions f and g on an interval I is
a step function on I.
Proof. Let P
f
and P
g
be nite partitions witnessing that f and g are step
functions. Now let P = P
f
P
g
. Then P is nite and a renement of P
f
and
P
g
. In view of Proposition 1.11 P is witness to the fact that f and g are both
step functions. We can order the points of P: (a
0
< a
1
< a
2
< < a
k
) . Since
f and g are both constant on (a
j1
, a
j
) their sum is as well. As this holds for
all j k the function f + g is constant on every interval in I\P and so a step
function with witness P.
5
Exercise 1.14 Prove that if I J are intervals then the restriction of a step
function on J to I is also a step function.
Exercise 1.15 Any step function can be written in the form of a nite sum
=
k

j=1

Ij
where I
r
are intervals. Conversely any such sum is a step function on R.
A Priestley step function is any function that can be expressed in the form

k
j=1

Ij
(x) where I
j
are bounded intervals.
Exercise 1.16 The Priestley step functions are the step functions on the
interval R that are zero on the initial and nal intervals
(, a
1
) , (a
k1
, )
of any partition that is witness to being a step function.
Remark 1.17 In this way, we have reconciled the dierent notations used in
dierent publications from the Institute. We will stick to our version, which is
equivalent to the denition used in Roe. However, the dierences between the
classes of functions described by these denitions are minor, and the key issues
lay elsewhere!
1.2 The Integral of a Step Function
Suppose that is a step function on a xed closed interval J = [a, b] R, and
that P = (a
0
= a < a
1
< a
k
= b) is a partition of J that witnesses this; then,
for i k, there are constants c
i
R so that
(x) = c
i
, x (a
i1
, a
i
) .
We might dene I (, P) =

k
i=1
c
i
(a
i
a
i1
), and perhaps this makes a good
starting point, as the denition for the integral of . However, it might be
the case that a second partition

P witnessing as a step function could give a
dierent integral.
Lemma 1.18 Suppose that is a step function on J witnessed by the partition
P, and that a second partition

P is a renement of P, then

P also witnesses
that is a step function, and
I (, P) = I
_
,

P
_
.
6
Proof. Let |P| = n and

= n. We proceed by induction. Suppose the

proposition is false, then among all counterexamples, we can choose a one: ,
J, P,

P with
I (, P) = I
_
,

P
_
.
which minimises n and among these choose an example that maximises n. If
n < n 1, then, by dropping one point from

P\P we can nd a partition

P
with

= n 1 that is a renement of P and for which

P is a renement. By
induction I (, P) = I
_
,

P
_
and so I
_
,

P
_
= I
_
,

P
_
, but

= n 1 >
|P| which contradicts the assumption that our (counter) example was maximal
among all those with

= n.
Hence, we may assume that |P| =

1. Let
P = (a
0
= a < a
1
< a
n
= b)
and let a be the unique element of

P\P. Then a J\P so there is a j n so
that
a (a
j1
, a
j
) .
Since is a step function witnessed by P there are c
i
so that (x) = c
j
for
x (a
j1
, a
j
), j n and
I (, P) =
n

i=1
c
i
(a
i
a
i1
)
while, using the denition again we see that
I
_
,

P
_
=
j1

i=1
c
i
(a
i
a
i1
)
+c
j
( a a
j1
)
+c
j
(a
j
a)
+
n

i=j
c
i
(a
i
a
i1
)
and because
c
j
(a
j
a
j1
) = c
j
( a a
j1
) + c
j
(a
j
a)
We have I
_
,

P
_
= I
_
,

P
_
. This contradicts the existence of a counterexam-
ple as required.
Corollary 1.19 The integral I (, P) of a step function does not depend on the
choice of partition.
7
Proof. Let P and P

be two partitions witnessing that is a step function.

Let

P = P P

then

P is a renement of both of these partitions and
I (, P) = I
_
,

P
_
= I (, P

) .
Denition 1.20 We dene the integral
_
b
a
of a step function to be I (, P)
for any partition witnessing that is a step function.
Theorem 1.21 (Linearity and Positivity on step functions) Let and
be two step functions on the interval J = [a, b] and let and be real numbers;
then the integrals of the step functions + , and satisfy
_
b
a
+ =
_
b
a
+
_
b
a
.
If on the interval J = [a, b] then
_
b
a

_
b
a
.
If, for some non-empty open interval (c, d) [a, b] one has (a) > (a), a
(c, d), then
_
b
a
>
_
b
a
.
Proof. Fix partitions of the interval J = [a, b] witnessing that and are
both step functions on J and, if neccessary rene the partition to include {c, d},
by taking the union of these two partitions, choose a single partition P =
(a
0
= a < a
1
< a
n1
= b) that is a renement of both. Then, by assumption,
there are constants c
i
and d
i
so that
(x) = c
i
, x (a
i1
, a
i
)
(x) = d
i
, x (a
i1
, a
i
)
and for each i
( + ) (x) =
i
c
i
+ d
i
, x (a
i1
, a
i
) .
So, from the denition of I and the usual distributive laws of arithmetic
I (( + ), P) =
n

i=1
(
i
c
i
+ d
i
) (a
i
a
i1
)
=
n

i=1

i
c
i
(a
i
a
i1
)
+
n

i=1
d
i
(a
i
a
i1
)
= I (, P) + I (, P) .
8
The case of inequality can be proved by following a virtually identical argu-
ment. The inequality between the functions becomes an inequality between the
terms in a sum and noting that if there is a strict inequality on one of the terms
in the sum there is a strict inequality on the sums.
Alternatively, one can use the linearity established in the rst part of this
proof to reduce the problem to a special case: if on the interval J = [a, b],
then ( ) 0 on the interval J = [a, b] .and is also a step function. The
integral of a positive step function is clearly (from the denition) positive. So
_
b
a
( ) 0
and so
_
b
a

_
b
a
0
_
b
a

_
b
a

For strict inequality between the integrals of and it is enough that the step
function ( ) > 0 on a non-empty open interval.
1.3 The Integral of a Continuous Function - details
We have dened step functions, shown that they are a vector space, dened
an integral on this space, and observed that the integral is linear and positive.
Although these results simple and extremely intuitive they are not quite obvious,
one must validate our model for the real numbers by developing these concepts
within this context without fuss or complexity. In this section our goal is to
integrate continuous functions on bounded intervals that contain their endpoints.
These are exactly the intervals on which continuous functions are bounded and
uniformly continuous. Both of these properties will play an important role in
our discussion. First was set up some notation. The space of step functions on
[a, b] will be denoted by L
step
[a, b] and the space of continuous functions will be
denoted by C [a, b].
Denition 1.22 If f is a real valued function on [a, b] then we may dene
_
b
a
f = sup
_
_
b
a

L
step
[a, b] , f
_
_
b
a
f = inf
_
| L
step
[a, b] , f
_
Recall the following a standard result from last term:
Lemma 1.23 If f C [a, b] then there exists m, M (, ) so that for all
x [a, b] m f (x) M.
9
From this and the positivity of the integral on L
step
[a, b] we see that
Lemma 1.24 If f C [a, b] then
< m(b a) <
_
b
a
f
_
b
a
f < M (b a) <
where M = sup
[a,b]
f and m=inf
[a,b]
f.
Proof. Fix a continuous function f C [a, b] Then m and M are nite numbers.
So there is always a step function m
[a,b]
less than f on the interval [a, b].
similarly, there were always be a step function M
[a,b]
dominating f on the
interval [a, b] .hence
_
b
a
f < M (b a)
<
and
< m(b a)
<
_
b
a
f.
on the other hand if L
step
[a, b] and f then
L
step
[a, b] , f

and so the positivity of the integral on step functions gives
_
b
a

_
b
a

sup
_
_
b
a

L
step
[a, b] , f
_

_
b
a

_
b
a
f
_
b
a

and since was arbitrarily chosen from { L

step
[a, b] , f}we can take
the inmum of the right-hand side over this class of step functions to get
_
b
a
f
_
b
a
f.
We now come to the most fundamental theorem in the course.
10
Theorem 1.25 If f is a continuous function on the closed and bounded interval
[a, b] then
_
b
a
f =
_
b
a
f
and we call the result the integral of f.
The proof depends absolutely crucially on:
Theorem 1.26 (uniform continuity - from last term or books) Every con-
tinuous function on a closed bounded interval [a, b] is uniformly continuous.
That is to say, if f C [a, b] then for each > 0 there exists a > 0 so that if
|x y| and x, y [a, b] then |f (x) f (y)| < .
Proof. The result is obvious if b = a as in this case both sides are zero.Fix
f and > 0. Choose > 0 so that if |x y| then |f (x) f (y)| <

ba
.
Choose a partition
P = (a
0
= a < a
1
< a
n
= b)
so that |a
r
a
r1
| for every r n.
Dene
c
r
= inf {f (x) | x [a
r1
, a
r
]}
d
r
= sup {f (x) | x [a
r1
, a
r
]}
and dene , by:
= c
r
on (a
r1
, a
r
)
= d
r
on (a
r1
, a
r
)
and
(a
r
) = (a
r
) = f (a
r
) .
Then and are in L
step
[a, b] .Clearly
f
and our choice of ensures that
+

b a
.
Thus, and in view of Lemma 1.24,it follows that
_
b
a

_
b
a
f
_
b
a
f
_
b
a
.
11
and
_
b
a

_
b
a
_
+

b a
_
(1)
=
_
b
a
+
_
b
a

b a
(2)
=
_
b
a
+ (3)

_
b
a
f + (4)
so
_
b
a
f
_
b
a
f
_
b
a
f +
and as is arbitrary
_
b
a
f =
_
b
a
f.
We can extract the following remark out of the above proof:
Corollary 1.27 Suppose that f C [a, b] is a continuous function on a closed
bounded interval. Then for each > 0 there is a step function L
step
[a, b]
such that
f < f + .
This will often be convenient.
12
2 The Fundamental Theorem of Calculus and
the link with dierentiation
We now wish to develop some basic consequences that follow from our ability to
integrate continuous functions. We start by proving that the integral is positive
and linear.
Theorem 2.1 (Strict positivity of the integral) Suppose that a < b, f
[a, b], f (x) 0 for all x [a, b], and that there is an x [a, b] so that f (x) > 0.
Then
_
b
a
f > 0.
Proof. Suppose that x [a, b] and f (x) > 0. Choose (0, f (x) /2) ;the
interval is nonempty by hypothesis. Since f is continuous there is a > 0 so
that if y (x , x + ) [a, b] then |f (y) f (x)| < . In particular f (y) >
. Suppose that = on (x , x + ) [a, b] and zero o it. Then
L
step
[a, b] .Since a < b and x [a, b] it follows that there is a

> 0 so that one

of the following holds:
_
x

, x
_
(x , x + ) [a, b]
_
x, x +

_
(x , x + ) [a, b]
Then, one of

(x

,x)

(x,x+

)

and in either case, 0 <

_
b
a
. By denition
_
b
a

_
b
a
f
_
b
a
f.
Theorem 2.2 (Linearity of the integral) Suppose that f, g C [a, b] and
that , R then
_
b
a
f + g =
_
b
a
f +
_
b
a
g.
Proof. It is enough to treat the cases
_
b
a
f =
_
b
a
f
and
_
b
a
(f + g) =
_
b
a
f +
_
b
a
g.
13
separately. If 0 then is a step function less than f if is a step function
less than f.From the linearity of the integral on step functions one can conclude

_
b
a

_
b
a
af
and taking the supremum over the step functions less than f one has

_
b
a
f
_
b
a
af.
and similarly
_
b
a
af a
_
b
a
f
and recalling that, because f is continuous, f is integrable and
a
_
b
a
f
_
b
a
af.
_
b
a
af a
_
b
a
f
and the result follows. The case where 0 is equally simple.
If b = a we are nished as all the integrals will be zero. Fix > 0, applying
corollary 1.27 identify and in L
step
[a, b] so that
f +

2 (b a)
g +

2 (b a)
+ .f + g + +

(b a)
Then
_
b
a
(f + g)
_
b
a
_
+ +

(b a)
_
=
_
b
a
+
_
b
a
+

_
b
a
f +
_
b
a
g + .
Because was arbitrary, we have
_
b
a
(f + g)
_
b
a
f +
_
b
a
g. Applying this to
f and g using the relation in the rst part of the proof we conclude that
_
b
a
(f + g) =
_
b
a
f +
_
b
a
g.
Theorem 2.3 (additivity on disjoint intervals) Suppose that b [a, c] and
that f C [a, c] then
_
c
a
f =
_
b
a
f +
_
c
b
f.
14
Proof. The result is obvious for step functions;
=
[a,b)
+
[b,c]

so that
_
c
a
=
_
c
a

[a,b)
+
_
c
a

[b,c]

but, adding the point b to any partition P witnessing that

[a,b)
is a step
function on [a, c] one quickly sees that
_
b
a
=
_
c
a

[a,b)

_
c
b
=
_
c
a

[b,c]

and yields the required additivity.

The result for continuous functions then follows from corollary 1.27. Choose
so that
f +

(c a)
then
_
c
a
f
_
c
a
_
+

(c a)
_
=
_
c
a
+
=
_
b
a
+
_
c
b
+

_
b
a
f +
_
c
b
f +
and was arbitrary so
_
c
a
f
_
b
a
f +
_
c
b
f ; using this inequality with f as well
we deduce there is equality.
Corollary 2.4 If f and g are in C [a, b] and f g then
_
b
a
f
_
b
a
g.
The proof is an excercise! using the previous results.
2.1 The main theorems
Theorem 2.5 (The rst mean value theorem (for integrals) FMVT) Suppose
that f C [a, b] then for some c [a, b]
_
b
a
f = f (c) (b a) .
15
Proof. Let
M = sup {f (x) | x [a, b]}
m = inf {f (x) | x [a, b]} .
Then f attains the values m and M. Fix e, E [a, b] so that
f (e) = m
f (E) = M.
Of course m
[a,b]
f M
[a,b]
so that using the denition of integral,
m
1
b a
_
b
a
f M
(assuming b > a) and by the intermediate value theorem there exists c [e, E]
[a, b] so that
1
b a
_
b
a
f = f (c) .
If b = a the result is immediate with c = b.
Exercise 2.6 Suppose f, and g are continuous functions on the closed bounded
interval [a, b] and that
_
b
a
f =
_
b
a
g then there exists a point c [a, b] such that
f (c) = g (c).
Note that the rst mean and value theorem (FMVT) follows from this remark
by taking g (x)
1
ba
_
b
a
f .
Denition 2.7 If b < a then that we dene
_
b
a
f to be -
_
a
b
f.
We can immediately,and easily, extend theorem 2.3
Theorem 2.8 (additivity) suppose f is a continuous function on some inter-
val J and that a, b, c J then
_
c
a
f =
_
b
a
f +
_
c
b
f
Proof. We only treat the new case a < c < b. Using theorem2.3 we see that
_
b
a
f =
_
c
a
f +
_
b
c
f
and so
_
c
a
f =
_
b
a
f
_
b
c
f
16
and by denition this
=
_
b
a
f +
_
c
b
f.
Suppose again that f is a continuous function on some interval J and that
a is in J.
Denition 2.9 We say F(x) is the (an) indenite integral of f on J if there
is a constant C R so that for all x J
F (x) =
_
x
a
f + C.
Theorem 2.10 (the fundamental theorem of calculus) If f C [a, b] where
a < b, and for some xed C
F (x) =
_
x
a
f + C
then F is dierentiable on (a, b) with the left derivative at b and a right derivative
at a and
F

(x) = f (x) , x [a, b] .

Proof. If we prove that
lim
h0

h>0
F (x + h) F (x)
h
= f (x) .
for all x [a, b). Then a similar argument (with the additivity of the integral
on disjoint intervals) will prove the existence of a right derivative for x (a, b] .
Taken together these give the theorem.
Fix > 0, because f is continuous, we may choose > 0 so that
|f (x) f (x + h)| <
whenever h (0, ) and by reducing we can always ensure that in addition
x + [a, b]. Now there is a C so that F (x + h) is dened by
F (x + h) =
_
x+h
a
f + C
=
_
x+h
x
f +
_
x
a
f + C
=
_
x+h
x
f + F (x) .
and applying the rst mean value theorem there exists [x, x + h] so that
F (x + h) F (x)
h
= f () .
17
If h (0, ) then |x | < and so

F (x + h) F (x)
h
f (x)

< .
which completes the proof.
This allows us to construct an enormous number of dierentiable functions!
There are many versions of the mean value theorem - and we set the following
as an easty excercise:
Theorem 2.11 (the second mean value theorem for integrals) Suppose
f, g C [a, b], and that g 0 then there exists c [a, b] so that
_
b
a
fg = f (c)
_
b
a
g
2.1.1 Applications
Example 2.12 (Logarithm) For y R
>0
dene
log y =
_
y
1
1
x
then log y this dierentiable and its derivative is
1
y
.
The following corrolary is also frequently referred to as the Fundamental
Theorem of Calculus
Corollary 2.13 If F is dierentiable on the interval J with derivative f and
if f is continuous then there exists a constant C R so that for all x J
F (x) =
_
x
a
f + C.
Proof. Consider H (x) dened by
H (x) = F (x)
_
x
a
f.
then we have just proved that H is dierentiable and H

0 So applying the
mean value theorem for functions, H is a constant and the result follows.
Example 2.14 We will prove that if f is continuous then
I
n
:=
_
1
0
nf (x)
1 + n
2
x
2
dx
1
2
f (0)
Put
g (x) =
n
1 + n
2
x
2
18
then g is the derivative of tan
1
nx. Hence
_
1
0
g (x) dx =
_
tan
1
nx

1
0

1
2

The graph of
3
1+3
2
x
2
We know there is a choice of c
n
so that
_
1
0
nf(x)
1+n
2
x
2
dx = f (c
n
)
_
1
0
n
1+n
2
x
2
dx.If
we can show that c
n
0 we will have nished. For this we need to be to be
cleverer.
_
1
0
nf (x)
1 + n
2
x
2
dx =
_
1/

n
0
nf (x)
1 + n
2
x
2
dx +
_
1
1/

n
nf (x)
1 + n
2
x
2
dx
= f (c
n
)
_
tan
1
nx

1/

n
0
+
_
1
1/

n
nf (x)
1 + n
2
x
2
dx
where c
n
[0, 1/

n]. Now

_
1
1/

n
nf (x)
1 + n
2
x
2
dx

M
_
1
1/

n
n
1 + n
2
x
2
dx
= M
_
tan
1
n tan
1

n
_
M
_

2


2
_
= 0
as n .Meanwhile,
_
tan
1
nx

1/

n
0
= tan
1
(

n)

2
, and f (c
n
) f (0)
as |c
n
|

n.
3 Integration by parts, substitution and meth-
ods of integration
The methods we have introduced are practically useful for explicitly calculating
integrals and derivatives. Unfortunately, any subject that has existed for 400
years in some form develops multiple notations and slight variants of denitions.
19
Denition 3.1 A function F is the primitive (or anti-derivative) of f on [a, b]
if F

= f on [a, b].
In view of Cor. 2.13 F is the indeninte integral of f if the latter is contin-
uous.
Remark 3.2 There are good and bad notations; one often adds a dx to indicate
the argument in the function:
_
xdx
and sometimes wrongly writes:
_
xdx =
1
2
x
2
+ C
which is not really correct. The following is much better:
_
x
tdt =
1
2
x
2
+ C
or even more informatively
_
x
a
tdt =
1
2
_
x
2
a
2
_
We can (easily) use our methods to do integration of some complicated
algebraic functions:
Example 3.3 the function
x
2
+ 1
(x + 2)
2
=
(x + 2)
2
4 (x + 2) + 5
(x + 2)
2
= 1
4
x + 2
+
5
(x + 2)
2
so has primitive x 4 log (x + 2)
5
x+2
and so
_
x
0
t
2
+ 1
(t + 2)
2
dt = x 4 log (x + 2)
5
x + 2
+ 4 log 2 +
5
2
As you probably learnt at school - anti-derivatives/primitives allow us to
translate known properties of dierentiation to properties of integration. The
following is one of the most important:
Proposition 3.4 (Integration by parts) If u, v are dierentiable on [a, b]
and u

, v

C [a, b] then
_
b
a
(u

v) +
_
b
a
(uv

) = u(b) v (b) u(a) v (a)

= [uv]
b
a
20
Proof. The function uv is dierentable, its derivative is (u

v) + (uv

)
Substitution
Proposition 3.5 (Substitution) Suppose that g C [c, d] is monotone in-
creasing, and continously dierentiable so g

C [c, d] , g

> 0 and suppose that

g (c) = a, g (d) = b.Suppose that f C [a, b] . Then
_
d
c
f gg

=
_
b
a
f
Proof. We can apply the fundamental theorem of calculus. Let F (x) :=
_
x
a
f
and dene G(u) = F (g (u)) then
G(c) = F (g (c))
= F (a)
= 0
_
b
a
f = G(d)
= G(d) G(c)
and providing G is continuously dierentiable one has
G(d) G(c) =
_
d
c
G

so recalling that (FTC again) that F

= f and applying the chain rule

G

(u) = f g (u) g

(u)
and hence
_
b
a
f (x) dx =
_
d
c
f g (u) g

(u) du
Example 3.6 The integral of log :
_
x
1
log t =
_
x
1
(t

) log t
=
_
x
1
(t)
1
t
+ [t log t]
x
1
= xlog x x + 1
Example 3.7 The integral
I =
_
2
0
1
1 + sin x
dx
and use substitution t = tan (x/2)
21
http://math.berkeley.edu/sassaf/math1b/fa03/tech.pdf
22
4 Power series and limits - The trancendental
mathematical functions exp, sin etc.
4.1 The Interchange of Limits and Integration
In general, Real numbers (even interesting ones such as ) lack descriptions
in terms of simple formulae, and instead are dened in terms of limits. In an
analogous way, most of the interesting mathematical functions are constructed
as limits; for example, as an innite series. Operations such as dierentiation
and integration are often well understood when applied to the the intermediate
functions used in the construction and we need understand how to inherit some
of these properties for the limit.
For example, it is a simple calculation to see that
u(x, t) = a
0
+
N

1
e
n
2
t
(a
n
cos nx + b
n
sin nx)
is a solution to the heat equation

t
u =

2

2
x
u
and most of you will know that it is possible to write a reasonably a general
function F as the sum of an innite Fourier series
F (x) = a
0
+

1
(a
n
cos nx + b
n
sin nx) .
With this in mind it would certainly make sense to ask about the convergence
of
a
0
+

1
e
n
2
t
(a
n
cos nx + b
n
sin nx)
and particularly to ask whether the limit is a solution to the same dierential
equation. If it were, then one would have constructed a solution to the heat
equation on the unit interval having periodic boundary conditions at the end of
the intervals, and having as its initial condition a temperature F.
Problem 4.1 It is a simple calculation and an application of the FTC to see
that
1 +
_
x
0
N

0
t
n
n!
dt =
N+1

0
x
n
n!
Can we deduce from this that:
1 +
_
x
0
e
t
dt = e
x
23
and hence conclude from a second application of the FTC that
d
dx
e
x
= e
x
Remark 4.2 In the previous two examples, we hoped (and we shall see our
hope fullled in this context) that the derivative of the limit (or the integral of
the limit) the limit of the derivatives (integrals) of the approximating sequences -
which can be computed computed term by term. Our wish is not always fullled!
There are even really important examples where it works in our favour that the
derivative of the limit is not the limit of the derivatives. So it is important
to understand the sort of hypotheses involved in proving the conditions we give
below.
Example 4.3 Consider the function dened for x > 0 by
f (x) = n
2
x, x [0, 1/n]
f (x) = 2n n
2
x, x [1/n, 2/n]
f (x) = 0, [2/n, ]
then
_
2
0
f
n
= 1 for all n . But for each xed x, f
n
(x) 0.
Recall the following denition:
Denition 4.4 Let f
n
, f be functions on the set [a, b]. we say that f
n
f
uniformly on [a, b] if
lim
n
sup
x[a,b]
|f
n
(x) f (x)| = 0.
and one of the basic theorems about uniform convergence:
Theorem 4.5 Suppose that f
n
f uniformly on [a, b], where f
n
C [a, b].
Then f C [a, b].
Remark 4.6 If you have too show that a sequence of functions converges uni-
formly then one should always be careful. It often happens that the location
where
|f
n
(x) f (x)|
takes its maximum will frequently depend on n and so moves. It pays to use
calculus to determine the critical points of f
n
(x) f (x).
Exercise 4.7 The functions
f
n
(x) = n

xe

n
2
n
2
2
, x > 0
f
n
(x) = 0, x 0
converge to 0 uniformly i < 1. (Hint - compute f (x) = limf
n
(x). Then
take logs of |f
n
(x) f (x)| and dierentiate to identify the maximum.)
24
Theorem 4.8 Suppose that f
n
f uniformly on [a, b], where f
n
C [a, b] ,(and
so f C [a, b]) then
_
b
a
f
n

_
b
a
f.
Proof. If b = a then both sides of the expression are zero and the argument is
complete. So x > 0 and use the uniform convergence of the f
n
to choose N
so that |f
n
(x) f (x)| < / (b a) for all n > N. Then

_
b
a
f
n

_
b
a
f

_
b
a
(f
n
f)

_
b
a
|(f
n
f)|
< .
Hence
_
b
a
f
n

_
b
a
f.
Corollary 4.9 If u
r
C [a, b] and

0
u
r
converges uniformly then
_
b
a

0
u
r
=

0
_
b
a
u
r
.
The Wierstrass M-test from last term (look it up!) is a convenient way to
see that

0
u
r
converges uniformly.
Example 4.10 Consider
f (x) = 1
x
2
3!
+
x
4
5!
+
=

n=0
(1)
n
x
2n
(2n + 1)!
for x [0, 1]. (this is
sin x
x
if x = 0) then, for t [0, x]

(1)
n
t
2n
(2n + 1)!

t
2n
(2n + 1)!

x
2n
(2n + 1)!

= M
n
and

n=0
M
n
< so the series for f converges uniformly on [0, x]. As a
result, we may integrate term by term:
_
x
0
f =

n=0
(1)
n
_
x
0
t
2n
(2n + 1)!
=

n=0
(1)
n
x
2n+1
(2n + 1)! (2n + 1)
25
4.2 Power series
Now recall form last term that a power series

n=0
a
n
x
n
always has a radius
of convergence R, and that if |x| < R then the series is absolutely convergent
with

n=0
|a
n
| |x|
n
< and moreover, for |z| |x| one has
|a
n
z
n
| |a
n
| |x|
n
and applying the M-test, one sees that

n=0
a
n
z
n
converges uniformly for |z| |x|.
Lemma 4.11 For any a
n
R, the power series

n=0
a
n
x
n
,

n=0
na
n
x
n1
and

n=0
a
n
x
n+1
n+1
have the same radius of convergence.
Proof. Let R

be the radius of convergence of

n=1
na
n
t
n1
. Then if |x| < R

one has for n 1

|a
n
x
n
| R

a
n
x
n1

and so

n=0
a
n
x
n
converge absolutely by comparison with
|a
0
x| + R

n=0
n

a
n
x
n1

.
Hence the radius of convergence R of

n=0
a
n
t
n
satises R

R. On the other
hand if |x| < R then there is a > 1 so that |x| < R. Then for x = 0
n

a
n
x
n1

n
|x|
n
|a
n
x
n

n
|
Now, since > 1 the quantity
C (x, ) = sup
n
n
|x|
n
is strictly nite for each and x. So, by comparison with C (x, ) |a
n
x
n

n
| one
deduces that

n=0
na
n
x
n1
is also absolutely convergent. So R

R.
The other comparison can be achieved by simply relabelling the co-ecients
- put
b
0
= 0
b
n+1
=
a
n
n + 1
26
then

n=0
b
n
x
n
,

n=0
nb
n
x
n1
have the same radius of convergence so that

n=1
b
n
x
n
and

n=1
nb
n
x
n1
have the same radius of convergence. Substi-
tuting in the denitions of the b
n
we get

n=1
b
n
x
n
=

n=1
a
n1
n
x
n
=

n=1
a
n1
n
x
n
=

n=0
a
n
n + 1
x
n+1
.

n=1
nb
n
x
n1
=

n=1
a
n1
x
n1
=

n=0
a
n
x
n
Theorem 4.12 (term by term integration of power series) Suppose that
f (t) =

n=0
a
n
t
n
has a radius of convergence R. Then f is continuous on
(R, R) and
_
x
0
f (t) dt =

n=0
a
n
x
n+1
n + 1
for any x with |x| < R.
Proof. If |x| < R then we know the series

n=0
a
n
t
n
converges uniformly on
[|x| , |x|]. By the Theorem 4.8 the integral
_
x
0
f (t) dt = lim
N
_
x
0
N

n=0
a
n
t
n
dt
and by linearity this
= lim
N
N

n=0
a
n
_
x
0
t
n
dt
= lim
N
N

n=0
a
n
x
n+1
n + 1
=

n=0
a
n
x
n+1
n + 1
.
27
Example 4.13 The series
1
1 x
=

0
x
n
, |x| < 1
leads one to
log (1 x) =

n=0
x
n+1
n + 1
=

n=1
x
n
n
for x (1, 1) .Note: although it is true that log 2 =

n=1
(1)
n
n
it does not
follow from the above arguments and is much more delicate.
Theorem 4.14 Suppose that f (t) =

n=0
a
n
t
n
has a radius of convergence R
then f can be dierentiated term by term on (R, R) and f

(t) =

n=0
a
n1
t
n1
for t (R, R).
Proof. Put g (t) =

n=0
a
n1
t
n1
for t (R, R) and
G(x) =
_
x
0
g (t) dt, x (R, R) .
Applying term by term integration one has that G(x) + a
0
= f (x). Since g is
continuous we may apply the fundamental theorem of calculus to deduce that
f is dierentiable and its derivative is g.
4.3 Those famous functions
Rudin starts his (famous) text-book on Real and Complex Analysis with the
statement that exp is the most important function in mathematics!
We dene it by
exp (z) =

n=0
x
n
n!
and this series converges absolutely for every (complex) number z and, among
other things, is a homomorphism from the abelian group (R, +) to the abelien
group (R
>0
, ) . There are a number of important functions that are very closely
related to it.
Denition 4.15 We dene sin and cos to be the series
sin z =

n=0
(1)
n
z
2n+1
(2n + 1)!
cos z =

n=0
(1)
n
z
2n
(2n)!
28
and so z C
exp iz = cos z + i sin z
and if z R then
cos z = R(exp iz)
sin z = I (exp iz)
Denition 4.16 Similarly we dene
sinh z =
1
2
(exp z exp (z))
cosh z =
1
2
(exp z + exp (z))
tanh z =
sinh z
cosh z
.
It is not (at all) obvious that these functions satisfy our intuitive axioms.
The logical sequence of events is that we started with axioms for real numbers
and slowly but eectively we have created a rigorous machine in which these
functions can be dened and studied. To justify this approach we need to prove
that these objects do correpond to our intuitive picture.
One basic property of sin and cos is that
Proposition 4.17 If t R then
(sin t)
2
+ (cos t)
2
= 1
Proof. Using term by term dierentiation one observes that
d
dt
sin t = cos t and
d
dt
cos t = sin t so applying the product rule the derivative of
I (t) := (sint)
2
+ (cos t)
2
is identically 0. By the mean value theorem the function I is constant and we
have I (0) = 1 so I 1.
It is not quite obvious that sin and cos are periodic! What is
Denition 4.18 If x is the smallest positive solution of cos x = 0 then we
dene = 2x.
Proposition 4.19 exists and is strictly between 2

2 and 4.
Proof. Since cos 0 = 1 and cos is continous, there exists a > 0 so that cos>
1
2
on [0, ] so if x exists then x > 0. But we can be more precise! Recall the
basic estimate of alternating series: if a
n
are positive numbers decreasing to
zero and if
S
N
=
N

n=0
(1)
n
a
n
29
then the series converges to some limit l and in fact one can be much more
precise
S
0
S
2
S
2n
l S
2n+1
S
3
S
1
|S
2n
S
2n+1
| a
2n+1
and we can apply this to the series for cos .
Lemma 4.20 If z > 0 then the terms
a
n
:=
z
2n
(2n)!
are monotone decreasing for all n such that 0 z <
_
(2n + 2) (2n + 1) in
particular if z 3 then a
1
> a
2
> a
3
etc.
Proof. Note that
a
n+1
a
n
=
z
2
(2n + 2) (2n + 1)
.
Corollary 4.21 For 0 x 2 the function cos x satises
1
x
2
2
+
x
4
4!
cos x
and if x [0, 3] then
cos x > 1
x
2
2
+
_
x
4
4!

x
6
6!
_
and in particular cos 2
3
8
while for x
_
0,

2
_
the function cos x > 0.
Corollary 4.22 The number exists and satises 2

2 < 4.
Proof. Now cos 0 = 1 and cos 2 < 0 so that the continuity of cos,together
with the intermediate value theorem ensure that the set of x (0, 2) for which
cos x = 0 is closed and nonempty. It is therefore also compact and has a smallest
element which we can (and do) call . By using higher approximations we can
of course get arbirarily good numerical approximations - but at the moment we
are happy that is exists.
Corollary 4.23 One also has sin
_

2
_
= 1
Proof. The derivative of sin is cos and cos 0 on [0, /2] so that sin is
increasing on [0, /2]. Since (sint)
2
+(cos t)
2
= 1 and cos
_

2
_
= 0 one concludes
that sin
_

2
_
= 1 Putting the two remarks together gives the result.
30
4.3.1 The periodicity of sin and cos
We assume that you proved the basic fact that exp(a) exp (b) = exp (a + b)
Lemma 4.24 The addition formulae
cos (A + B) = cos Acos B sin Asin B
sin (A + B) = cos Asin B + sin Acos B
Proof. Using exp(a) exp (b) = exp (a + b) and exp iz = cos z +i sin z the we see
that
cos (A + B) + i sin (A + B) = (cos A + i sin A) (cos B + i sin B)
= (cos Acos B sin Asin B)
+i (cos Asin B + sin Acos B)
and if Aand B are real, collecting real and imaginary parts the result follows.
Corollary 4.25 One has for all A that
cos (A + /2) = sin A
sin (A + /2) = cos A
and
cos (A + ) = cos A
sin (A + ) = sin A
and nally
cos (A + 2) = cos A
sin (A + 2) = sin A
4.3.2 Fourier series
The existence of and orthogonality of integrals of trignometric functions follows
easily from our estimates above. Functions such as sin (2mt) sin (2nt) are, by
what we have proved, continuous and indeed innitely dierentiable, so we can
integrate by parts and use the periodicity to deal with the end values in order
to evaluate the integrals:
_
1
0
sin (2mt) sin (2nt) dt = 0, m = n
_
1
0
sin (2mt) cos (2nt) dt = 0, m, n
_
1
0
cos (2mt) cos (2nt) dt = 0, m = n
31
Suppose that f (t) =

N
0
a
n
cos 2nt +

N
1
b
n
sin 2nt then the following inte-
grals make sense and
a
0
=
_
1
0
f (t) dt, and for n 1
a
n
= 2
_
1
0
cos (2nt) f (t) dt
b
n
= 2
_
1
0
sin (2nt) f (t) dt
Denition 4.26 If f C [0, 1], with f (0) = f (1) then we dene the fourier
series of f to be
a
0
=
_
1
0
f (t) dt, and for n 1
a
n
= 2
_
1
0
cos (2nt) f (t) dt
b
n
= 2
_
1
0
sin (2nt) f (t) dt
Which leads to a basic question - can two dierent continuous functions
have the same fourier co-ecients. The answer is no - the proof is only a little
more dicult than one of the earlier excercises. Indeed we could prove a much
stronger theorem and explain how to recover f:
Theorem 4.27 Suppose that f C [a, b] with f (a) = f (b) and with fourier
coecients a
n
, b
n
. Dene
S
N
=
N

0
a
n
cos 2nt +
N

1
b
n
sin 2nt,
then the average C
N
=
1
N+1

N
n=0
S
n
of the S
n
converges uniformly to f on
[0, 1]!
Unfortunately time runs out. However the situation is delicate. In general,
the S
N
themselves do not converge uniformly or even pointwise.
32