Advanced Calculus: MATH 410

Professor David Levermore

12 October 2006
1. Real Number System
Numbers are at the heart of mathematics. By now you must be fairly
familiar with them. Basic sets of numbers are:

natural numbers, N = 0, 1, 2, ;

integers (die Zahlen), Z = , 2, 1, 0, 1, 2, ;
 
p
rational numbers (quotients), Q = : p, q Z, q 6= 0 ;
q

real numbers, R = (, ) ;

complex numbers, C = x + iy : x, y R .

Each of these sets is endowed with natural algebraic operations (like

addition and multiplication) and order relations (like less than) by
which their elements are manipulated and compared. It is fairly clear
how N, Z, and Q are related through an increasingly richer algebraic
structure. It is also fairly clear that R and C bear a similar relationship.
What is less clear is the relationship between Q and R. In particular,
what are the properties that allow R and not Q to be identified with a
line ? In this section we address some of these issues.

1.1. Fields. The sets Q, R, and C endowed with their natural alge-
braic operations are each an example of a general algebraic structure
known as a field.
Definition 1.1. A field is a set X equipped with two distinguished bi-
nary operations, called addition and multiplication, that satisfy the ad-
dition, multiplication, and distributive axioms presented below. Taken
together, these axioms constitute the so-called field axioms.

Addition axioms. Addition maps any two x, y X to their sum

x + y X such that:
A1: x + y = y + x x, y X, commutativity;
A2: (x + y) + z = x + (y + z) x, y, z X, associativity;
A3: 0 X such that x + 0 = x x X, identity;
1
2

A4: x X x X, such that x + (x) = 0, inverse.

Multiplication axioms. Multiplication maps any two x, y X to
their product xy X such that:
M1: xy = yx x, y X, commutativity;
M2: (xy)z = x(yz) x, y, z X, associativity;
M3: 1 X such that 1 6= 0 and x1 = x x X, identity;
M4: x X such that x 6= 0 x X such that xx = 1,
1 1

inverse.
Distributive axiom. Addition and multiplication are related by:
D: x(y + z) = xy + xz x, y, z X, distributivity.
Examples: When addition and multiplication have their usual mean-
ing, the field axioms clearly hold in Q, R, and C, but not in N or Z.
They also hold in Zn Z/(nZ) (the integers mod n) when n is prime.
If you do not know this last example, do not worry. It is not critical in
this course. You will see it in a basic algebra course.
All of the usual rules for algebraic manipulations involving addition,
subtraction, multiplication, and division can be developed from the
field axioms. This is not as easy as it sounds!

1.1.1. Consequences of the Addition Axioms. We begin by isolating the

addition axioms.
Definition 1.2. A set X equipped with a distinguished binary opera-
tion that satisfies the addition axioms is called an Abelian group or a
commutative group.

Remark: When working with Abelian groups, it is both convenient

and common to write
xy, x+y+z, 2x , 3x , ,
rather than
x + (y) , x + (y + z) , x +x, x +x+x, .
Examples: When addition has its usual meaning, the axioms for an
Abelian groups clearly hold in Z, Q, R, and C, but not in N. (As
defined here, N satisfies all these axioms but A4.) They also hold in
Zn for every positive integer n.
The addition axioms immediately imply the following.
Proposition 1.1. Let X be an Abelian group.
(a) If x, y, z X and x + y = x + z then y = z.
 3

(b) If x, y X and x + y = x then y = 0.

(c) If x, y X and x + y = 0 then y = x.
(d) If x, y X then (x + y) = (x) + (y).
(e) If x X then (x) = x.

Proof: Exercise.
Assertion (a) states that addition enjoys a so-called cancellation law.
Assertion (b) states that there is a unique additive identity of the type
assumed in A3. This unique additive identity is called zero. All other
elements of X are said to be nonzero. Assertion (c) states that for
every x X there is a unique additive inverse of the type assumed in
A4. This unique additive inverse is called the negative of x. The map
defined for every x X by x 7 x is called negation. Assertion (d)
states that the negative of a sum is the sum of the negatives. Assertion
(e) states that for every x X the negative of the negative of x is
again x.

1.1.2. Consequences of the Multiplcation Axioms. The only connection

of the multiplication axioms to addition is through the references to
zero in M3 and M4. An immediate consequence of M3 is that every
field must have at least two elements 0 and 1. It is also clear
that the nonzero elements of a field considered with the operation of
multiplication form an Abelian group.
Remark: When working with fields, it is both convenient and common
to write
x/y , xyz , x 2 , x3 , ,
rather than
xy 1 , x(yz) , xx , xxx , .

Examples: When addition and multiplication have their usual mean-

ing, the addition and multiplication axioms clearly hold in Q, R, and
C, but not in N or Z. They also hold in Zn when n is prime.
The multiplication axioms immediately imply the following.
Proposition 1.2. Let X be a field.
(a) If x, y, z X, x 6= 0, and xy = xz then y = z.
(b) If x, y X, x 6= 0, and xy = x then y = 1.
(c) If x, y X, x 6= 0, and xy = 1 then y = x1 .
(d) If x, y X, x 6= 0 and y 6= 0 then xy 6= 0 and (xy)1 = x1 y 1 .
(e) If x X and x 6= 0 then (x1 )1 = x.
4

Proof: Exercise.
Assertion (a) states that multiplication enjoys a so-called cancellation
law. Assertion (b) states that there is a unique multiplicative identity of
the type assumed in M3. This unique multiplicative identity is called
one. Assertion (c) states that for every nonzero x X there is a
unique multiplicative inverse of the type assumed in M4. This unique
multiplicative inverse is called the reciprocal of x. The map defined for
every nonzero x X by x 7 x1 is called reciprocation. Assertion (d)
states that the reciprocal of a product is the product of the reciprocals.
Assertion (e) states that for every nonzero x X the reciprocal of the
reciprocal of x is again x.
1.1.3. Consequences of the Distributive Axiom. The distributive axiom
gives the key connection between addition and multiplication. Taken
together, the field axioms imply the following.
Proposition 1.3. Let X be a field.
(a) If x X then x0 = 0.
(b) If x, y X and xy = 0 then x = 0 or y = 0.
(c) If x, y X then (x)y = (xy) = x(y).
(d) If x X and x 6= 0 then (x)1 = x1 .
Proof: Exercise.
Assertion (a) states that the product of anything with zero is zero.
In particular, it shows that zero cannot have a multiplicative inverse.
Hence, an element has a multiplicative inverse if and only if it is
nonzero. Assertion (b) states that if a product is zero, at least one of
its factors must be zero. This should be compared with (d) of Propo-
sition 2.2. Assertions (c) and (d) state how negation, multiplication,
and reciprocation relate.

1.2. Ordered Sets. The sets N, Z, Q, and R endowed with their

natural order relation are each an example of a general structure known
as an ordered set.
Definition 1.3. An ordered set (X, <) is a set X equipped with a
distinguished binary relation <, called an order, that satisfies the
order axioms presented below.

Order axioms. A binary relation < on a set X is called an order

whenever:
O1: if x, y, z X then x < y and y < z implies x < z,
transitivity;
 5

O2: if x, y X then exactly one of x < y, x = y, or y < x is true,

trichotomy.
Remark: When working with ordered sets, it is both convenient and
common to use the notation
x > y, x y, x y,
to mean
y < x, x < y or x = y , y < x or x = y .
Another common notational shorthand is
x <y < z, x < y z, ,
to mean
x < y and y < z , y < x and x y , .
Examples: When < has its usual meaning of less than, the order
axioms clearly hold in N, Z, Q, and R.

1.2.1. Bounds. Ordered sets have associated notions of boundedness.

Definition 1.4. Let (X, <) be an ordered set. A point x X is an
upper bound (a lower bound ) of a set S X whenever y x (x y)
for every y S. If S X has an upper bound (a lower bound) then
S is said to be bounded above (bounded below ). A set S X that is
both bounded above and bounded below is said to be bounded.
Definition 1.5. Let (X, <) be an ordered set, and let S X be
bounded above. A point x X is a least upper bound or supremum of
S whenever:
(i) x is an upper bound of S;
(ii) if y X is also an upper bound of S then x y.
We similarly define a greatest lower bound or infimum of S.

If a supremum or infimum of S exists then it must be unique. The

supremum of S is denoted sup{S} or sup{z : z S}, while the infi-
mum is denoted inf{S} or inf{z : z S}.
The notions of supremum and infimum should not be confused with
those of maximum and minimum.
Definition 1.6. Let (X, <) be an ordered set, and let S X. A point
x S is a maximum (minimum ) of S whenever x is an upper (lower)
bound of S.
6

If a maximum or minimum of S exists then it must be unique. The

maximum of S is denoted max{S} or max{z : z S}, while the
minimum is denoted min{S} or min{z : z S}. Moreover, if a
maximum (minimum) of S exists then


sup{S} = max{S} inf{S} = min{S} .
Examples: Any bounded open interval (a, b) in R has no maximum
or minimum, yet sup{(a, b)} = b and inf{(a, b)} = a. For any bounded
closed interval [a, b] in R one has max{[a, b]} = b and min{[a, b]} = a.
The same is true if these intervals are restricted to elements of Q.

1.2.2. Least Upper Bound Property. What distinguishs R from Q is the

following property.
Definition 1.7. Let (X, <) be an ordered set. Then X is said to have
the least upper bound property whenever every nonempty subset of X
with an upper bound has a least upper bound.

Remark: It may seem we should also define a greatest lower bound

property, but the next proposition shows that this is unnecessary
because it is exactly the same property.
Proposition 1.4. Let (X, <) be an ordered set. Let X have the least
upper bound property. Then every nonempty subset of X with a lower
bound has a greatest lower bound.

Proof: Let S X be a nonempty set with a lower bound. Let L X

be the set of all lower bounds of S. It is nonempty and bounded above
by any element of S. Therefore sup{L} exists. It is easy to check that
sup{L} = inf{S}. 
Examples: When < has its usual meaning of less than, the sets
N and Z have the least upper bound property. However, as we will
show in the next proposition, the set Q does not.
Proposition 1.5. The set Q does not have the least upper bound prop-
erty.

Proof: Consider the sets

S = r Q : r > 0 , r2 < 2 ,


S = r Q : r > 0 , r 2 > 2 .


These sets are clearly nonempty because 1 S and 2 S. One can

show that every point in S is an upper bound for S. In order to show
that S has no least upper bound, one first shows that there is no r Q
 7

such that r 2 = 2. It follows (by trichotomy) that if p is a least upper

bound of S then either p S or p S. We will show that neither can
be the case. More specifically, we will show that if p S then p is not
an upper bound of S, and that if p S then p is not a least upper
bound of S.
Let p S. We will construct a q S such that p < q, thereby
showing that p is not an upper bound of S. There are many ways to
construct such a q. We are seeking a rational approximation of 2 from
below that is better than p. This can be done by taking one iteration
of Newtons method applied to f (x) = 1 2/x2 = 0. Set
f (p) p2 2 6 p2
q =p = p p = p.
f 0 (p) 4 4
Because x 7 f (x) is increasing and concave over x > 0, a picture alone
should convince you this is a suitable q. Indeed, it is clear from the
above formula that 0 < p < q. A skeptic only needs to check that
q 2 < 2. We confirm this fact by the calculation
36 12p2 + p4 2 (2 p2 )2 (8 p2 )
2 q2 = 2 p = > 0.
16 16
Now let p S. We will construct a q S such that q < p, thereby
showing that p is not a least upper bound of S. Once again, there
are many ways to construct such a q. This time we are seeking a
rational approximation of 2 from above that is better than p. This
can be done by taking one iteration of Newtons method and applied
to f (x) = x2 2 = 0. Set
f (p) p2 2 p2 + 2
q =p = p = .
f 0 (p) 2p 2p
Because x 7 f (x) is increasing and convex over x > 0, a picture alone
should convince you this is a suitable p. Indeed, it is clear from the
above formula that q < p and that q > 0. A skeptic only needs to
check that q 2 > 2. We confirm this fact by the calculation
2
p4 + 4p2 + 4 p4 4p2 + 4
 2
2 p 2
q 2= 2= = > 0.
4p2 4p2 2p

Remark. An alternative construction that one could use for both
cases in the above argument is
p2 2 2p + 2
q =p = .
p+2 p+2
8

Then
4p2 + 8p + 4 p2 2
q2 2 = 2 = 2 .
p2 + 4p + 4 (p + 2)2
While this construction yields a much slicker proof, it is not clear (to
me) how one might have come up with it.

1.3. Ordered Fields. The sets Q and R endowed with their natural
algebraic operations and order relation are each an example of a general
algebraic structure known as an ordered field.
Definition 1.8. A set X that is both a field and an ordered set is called
an ordered field whenever:
OF1: if x, y, z X then x < y implies x + z < y + z;
OF2: if x, y X then 0 < x and 0 < y implies 0 < xy.
If x > 0 (x < 0, x 0, x 0) then we say x is positive (negative,
nonnegative, nonpositive). The set of all positive (negative) elements
of X is denoted X+ (X ).

Examples: When addition, multiplication, and < have their usual

meanings, the sets Q and R are ordered fields.
Proposition 1.6. Let X be an ordered field.
(a) If x > 0 then x < 0, and vice versa.
(b) If x > 0 and y < z then y < x + z and xy < xz.
(c) If x < 0 and y < z then x + y < z and xy > xz.
(d) If x 6= 0 then x2 > 0.
(e) If 0 < x < y then 0 < y 1 < x1 .

Proof: Exercise.
The above proposition shows that X+ satisfies the following.
P1: If x, y X+ then x + y X+ and xy X+ .
P2: For every x X exactly one of x X+ , x X+ , or x = 0 is
true.
These so-called positivity properties alone characterize the order re-
lation on the field X.
Proposition 1.7. Let X be a field. Let X+ X satisfy the above
positivity properties (P1-P2). Define the binary relation < on X by
x<y means y x X+ .
Then (X, <) is an ordered field.
 9

Proof: Exercise.
Proposition 1.7 implies that we could have defined an ordered field
as a field X that has a subset X+ satisfying the positivity properties.
In this case the positivity properties become the positivity axioms, and
the order axioms become order properties. This is the approach taken
in Fitzpatricks book.

Definition 1.9. Let X be an ordered field. The absolute value function

on X is defined by
x
if x > 0
|x| = 0 if x = 0 .

x if x < 0

Proposition 1.8. Let X be an ordered field. Then for every x, y X

(a) |x| 0,
(b) |x| = 0 if and only if x = 0,
(c) |x + y| |x| + |y|,
(d) |xy|
 |x| |y|,
(e) |x| |y| |x y|.


Proof: Exercise.
We can characterize bounded sets with the absolute value function.
Proposition 1.9. Let X be an ordered field. Then S X is bounded
if and only if there exists an m X+ such that
xS = |x| m .

Proof: Exercise.
1.4. Real Numbers. The main theorem of this section, which we
state without proof, is the following.
Theorem 1.1. There exists a unique (up to an isomorphism) ordered
field with the least upper bound property that contains Q (up to an
isomorphism) as a subfield.

Proof: Proofs of this theorem are quite long and technical. You
can find a proof of all but the uniqueness in the book by W. Rudin,
Principles of Mathematical Analysis, McGraw-Hill, New York, 1976.
That proof is based on a construction due to Dedekind using so-called
Dedekind cuts. Another proof is based on a construction due to Cantor
using Cauchy sequences. Both Dedekind and Cantor published their
constructions in 1872. 
10

Definition 1.10. The real numbers are defined to be the unique ordered
field with the least upper bound property whose existence is guaranteed
by Theorem 1.1. This field is denoted R.

Remark: The least upper bound property that sets R apart from Q.
As we will see, it is why R can be identified with a line.
The following important properties relate the reals R with the posi-
tive integers Z+ , the integers Z, and the rationals Q.
Proposition 1.10. The following hold.
If x, y R and x > 0 then there exists n Z+ such that
nx > y .
If x R then there exists a unique m Z such that


m (x 1, x] or equivalently x [m, m + 1) .
If x, y R and x < y then there exists a q Q such that
x < q < y.

Remark. The first assertion above is called the Archimedean property

of R, the second is a statement about the uniform distribution of the
integers, while the third asserts that Q is dense in R i.e. that
between any two reals lies a rational.
Proof. Suppose the first assertion is false. Then y is an upper bound
for the set S = {nx : n N}. By the least upper bound property S has
a supremum. Let z = sup{S}. Because x > 0 one has that z x < z.
Hence, z x is not an upper bound for S because z = sup{S}. This
implies there exists some n N such that z x < nx. But then
z < (n + 1)x, which contradicts the fact z is an upper bound of S.
Therefore the first assertion holds.
To prove the second assertion, by the first assertion there exists
k, l Z+ such that x < k and x < l. It follows that k < x < l.
Because
l1
[
x (k, l) [m.m + 1) ,
m=k

there exists some m Z such that

k m < l and m x < m +1.
It then follows that m (x 1, x]. To prove uniqueness, suppose that
m, n Z and m, n (x 1, x]. Without loss of generality we may
 11

suppose that m n otherwise simply exchange m and n. Because

x 1 < m n x, we see that n m Z satisfies
0 n m < x (x 1) = 1 .
It follows that n m = 0, thereby establishing the uniqueness in the
second assertion.
To prove the third assertion, because y x > 0, by the first assertion
there exists n Z+ such that n(y x) > 1. Then by the second
assertion there exists a unique m (nx, nx + 1]. Combining these
facts yields
nx < m nx + 1 < nx + n(y x) = ny .
Because n is positive, we conclude that
m
x< < y.
n
Therefore the third assertion holds with q = m/n. 
Recall that we showed that y 2 = 2 had no solution within Q. One of
the most important consequences of the fact R has least upper bound
property is the existence of solutions to such equations.
Proposition 1.11. For every x R+ and every n Z+ there exists a
unique y R+ such that y n = x.

Proof: The uniqueness of such a y is clear because if y < z then

y n < z n . So we only have to show such a y exists. Consider the sets
S = r R+ : r n < x , S = r R+ : r n > x .
 

The set S is nonempty because s = x/(1 + x) < 1 implies sn < s < x,

whereby s S. The set S is nonempty because 1 + x S. One can
show that every point in S is an upper bound for S. Let y = sup{S}.
Then (by trichotomy) either y S, y S, or y n = x. We will show
that the first two cases cannot occur, which will thereby prove the
theorem. More specifically, we will show that no point in S is an upper
bound of S, and that no point in S is the least upper bound of S.
Let p S. We can construct a q S such that p < q, thereby
showing that p is not an upper bound of S. This can be done by taking
one iteration of Newtons method applied to f (r) = 1 x/r n = 0. The
details are left as an exercise.
Now suppose p S. We can construct a q S such that q < p,
thereby showing that p is not a least upper bound of S. This can
be done by taking one iteration of Newtons method and applied to
f (r) = r n x = 0. The details are left as an exercise. 
12
1
The number y asserted in Proposition 1.11 is written x n . One can
1 1 1
easily show that for every x, y R+ one has (xy) n = x n y n .

1.5. Extended Real Numbers. It is often convenient to extend the

real numbers R by appending two elements designated and .
This enlarged set is called the extended real numbers and is denoted
by Rex .
The order < on R is extended to Rex by defining
< x < for every x R .
Interval notation thereby extends naturally to Rex . In particular, one
has Rex = [, ]. The ordered set (Rex , <) has the property that
() is an upper (lower) bound for every subset of Rex . It also has
the least upper bound property. Indeed, every nonempty S Rex has
a supremum given by


if S R has no upper bound in R or S ,
sup{S} = if S = {} ,

sup{S R} otherwise .

In particular, every nonempty S R that has no upper bound in R

(and therefore no supremum in R) has sup{S} = in Rex . Similar
statements hold for lower bounds and infimums.
The operations of addition and multiplication on R cannot be ex-
tended so as to make Rex into a field. It is however natural to extend
addition by defining for every x R
x+ = +x = , x = + x = ,
and by defining
+ = , = ,
while leaving and + undefined. Similarly, it is natural
to extend multiplication by defining for every nonzero x R
(
if x > 0
x = x =
if x < 0 ,
(
if x > 0
x () = () x =
if x < 0 ,
and by defining
= () () = , () = () = ,
while leaving 0 , 0, 0 (), and () 0 undefined.
 13

2. Sequences of Real Numbers

2.1. Sequences and Subsequences. Sequences play a central role in
analysis. We introduce them here in the context of an arbitrary set X
before specializing to sets of real numbers.
Definition 2.1. A sequence in a set X is a map from N into X, often
denoted {xk } or {xk }kN , where k 7 xk maps the index k N to the
point xk X.

Remark. Any countable ordered set may be used as the index set
instead of N.
When X is an ordered set, sequences that either preserve or reverse
order are special.
Definition 2.2. Let (X, <) be an ordered set. A sequence {xk }kN in
X is called
increasing whenever xl > xk for every k, l N with l > k,
nondecreasing whenever xl xk for every k, l N with l > k,
decreasing whenever xl < xk for every k, l N with l > k,
nonincreasing whenever xl xk for every k, l N with l > k.
It is called monotonic if it is either nondecreasing or nonincreasing.

When dealing with sequences, it is convenient to introduce the con-

cepts of ultimately and frequently.
Definition 2.3. Let A(x) be any assertion about any x X. (For
example, A(x) could be the assertion x S for a given S X.) Let
{xk } be a sequence in X. Then one says:
A(xk ) ultimately as k when m N such that k m A(xk ) ;
A(xk ) frequently as k when m N k m such that A(xk ) .
When there is no possible confusion as to the index set, one says simply
A(xk ) ultimately or A(xk ) frequently, dropping the as k .

Exercise. Show that 2k < .001 ultimately.

Exercise. Let {xk } be a sequence in X. Let A(x) be any assertion
about any x X and let A(x) be its negation. Show that the
negation of A(xk ) ultimately is A(xk ) frequently.
Exercise. Show that cos(k) > .5 frequently, but not ultimately.
Another useful concept is that of a subsequence.
14

Definition 2.4. A subsequence {xnk }kN of a sequence {xk }kN in a

set X is a map from N into X given by k 7 xnk , where {nk }kN is an
increasing sequence in N.

Example. If {xk } is a sequence in a set X, then {x2k } is the subse-

quence with indices that are even, while {xk3 } is the subsequence with
indices that are cubes.
Exercise. Consider the sequence {2k }. Write out the first three terms
(i.e. k = 0, 1, 2) in the subsequences {23k } and {22k+1 }.
In an ordered set, subsequences of monotonic sequences are again
monotonic.
Proposition 2.1. Let (X, <) be an ordered set. Let {xk }kN be a se-
quence in X that is increasing (nondecreasing, decreasing, nonincreas-
ing). Then every subsequence of {xk }kN is also increasing (nonde-
creasing, decreasing, nonincreasing).

Proof: Exercise.
You should test your understanding of the concepts in this section
by proving the following.
Proposition 2.2. Let X be a set. Let A(x) be any assertion about
any x X. Let {xk }kN be a sequence in X. Then A(xk ) frequently
as k if and only if there exists a subsequence {xnk }kN such that
A(xnk ) ultimately as k .

Proof: Exercise.
2.2. Convergence and Divergence. The most important concept
related to sequences is that of convergence. Here we see it in the
context of real sequences.
Definition 2.5. A sequence {ak }kN R is said to converge or is said
to be convergent whenever there exists a point a R such that for every
 > 0 one has that
(1) |ak a| <  ultimately as k .
This is denoted as
ak a as k ,
or as
lim ak = a .
k
One then says that the sequence converges to a. A sequence that does
not converge is said to diverge or is said to be divergent.
 15

An immediate consequence of this definition and ideas from the pre-

vious section is the following proposition.
Proposition 2.3. A sequence {ak }kN R diverges if and only if for
every a R there exists an a > 0 such that
|ak a| a frequently as k .

Proof: Exercise.
Definition 2.5 does not assert that there is a unique number a that
satisfies (1). The following proposition establishes this and more.
Proposition 2.4. If a sequence {ak }kN R converges, there is a
unique point in R to which it converges. Moreover, the set {ak } R
is bounded.

Proof: Here we prove only the boundedness assertion. The proof of

the uniqueness assertion is left as an exercise.
Let {ak }kN converge to a R. Then m N such that k m
|ak a| < 1. In particular, for every k m one has that a 1 < ak <
a + 1. Then for every k N we have

|ak | < 1 + max |a0 |, |a1 |, . . . , |am1 |, |a| .
The sequence {ak }kN is therefore bounded. 
Definition 2.6. The unique point to which a convergent sequence in
R converges is called the limit of the sequence.

An important characterization of the limit of a convergent sequence

is given by the following.
Proposition 2.5. Let {bk }kN be a convergent sequence in R. Let
b R. Then
lim bk = b ,
k
if and only if
a<b<c = a < bk < c ultimately .

Proof: Exercise.
It is fairly easy to check that subsequences of a convergent sequence
are also convergent, and have the same limit.
Proposition 2.6. If a sequence {ak }kN R converges to a limit
a R then every subsequence of {ak }kN also converges to a.
16

Proof: Exercise.
The main theorem regarding algebraic operations, order, and limits
is the following.
Proposition 2.7. Let {ak } and {bk } be convergent sequences in R with
ak a and bk b as k . Then
(i) (ak + bk ) (a + b) ,
(ii) ak a ,
(iii) ak bk ab ,
(iv) 1/ak 1/a provided no division by zero occurs .
Moreover, if ak bk frequently then a b.

Proof: The proof of assertion (iii) was given in class. The remainder
of the proof is an exercise. 
When working with real sequences, it is useful to distinguish two of
the many ways in which a sequence might diverge namely, the cases
where the sequence approaches either or .
Definition 2.7. A sequence {ak }kN R is said to diverge to (to
) if for every b R one has that
ak > b ultimately (ak < b ultimately) as k .
This is denoted as
ak (ak ) as k ,
or as  
lim ak = lim ak = .
k k
One then says that the sequence approaches (approaches ).
2.3. Monotonic Sequences. For monotonic sequences the least up-
per bound property can be employed to show the existence of limits.
Proposition 2.8. (Monotonic Sequence Theorem) If the sequence
{ak }kN R is nondecreasing (nonincreasing) then:
(i) it converges if and only if it is bounded above (bounded below);
(ii) if it converges,
 
lim ak = sup{ak } lim ak = inf{ak } ,
k k

while if it diverges,
 
lim ak = lim ak = .
k k
 17

Proof: We give the proof for the nondecreasing case only; the nonin-
creasing case goes similarly. The divergence case in (ii) is left as an
exercise.
() This follows immediately from Proposition 2.4, which states that
every convergent sequence is bounded.
() By the least upper bound property, let a = sup{ak ; k N}. We
claim that ak a as k . Let  > 0 be arbitrary. There exists
some m N such that 0 a am < . (For if not, it would mean
that ak a  for every k N, which would contradict the definition
of a.) But then for every k > m one has that 0 a ak a am < ,
which establishes the claim. 
For monotonic sequences it is enough to know what happens to a
single subsequence.
Proposition 2.9. Let {ak } be a monotonic sequence in R. Then {ak }
is convergent if it has a convergent subsequence.

Proof: Exercise.

2.4. Limit Superior and Limit Inferior. The power of the Mono-
tonic Sequence Theorem (Proposition 2.8) lies in the fact that from ev-
ery real sequence {ak }kN that is bounded above (bounded below), one
can construct a nonincreasing (nondecreasing) from its tails. Specif-
ically, the sequence {ak } ({ak }) with elements defined by
 

ak = sup al : l k ak = inf al : l k
is nonincreasing (nondecreasing). The convergence of such sequences
is characterized by Proposition 3.4, which motivates the following def-
inition.
Definition 2.8. For every sequence {ak }kN in R, define its limit su-
perior and limit inferior by
(
lim ak if sup{ak } < ,
lim sup ak k
k otherwise ;

(
lim ak if inf{ak } > ,
lim inf ak k
k otherwise .
These are called simply the lim sup and lim inf for short.
18

Remark: By the Monotonic Sequence Theorem (Proposition 2.8) we

have that
(
inf{ak } if sup{ak } < ,
lim sup ak
k otherwise ;

(
sup{ak } if inf{ak } > ,
lim inf ak
k otherwise .
Example: Consider the sequence {ak } given by
k+1
ak = (1)k for k Z+ .
k
The first seven terms of the sequences {ak }, {ak }, and {ak } are
ak ak ak

3
2 2
2
3 3
2 2
43
34 5
4
43
5 5
4 4
65
56 7
6
65
7 7
6 6
87
78 9
8
87
Notice that {ak } diverges while {ak } and {ak } are both monotonic and
converge to 1 and 1 respectively. Therefore
lim sup ak = 1 , lim inf ak = 1 .
k k

Remark: Notice that, unlike the limit, the lim sup and lim inf are
defined for every real sequence, taking values in Rex , and that in general
lim inf ak lim sup ak .
k k

Example: The sequence {(1)k k} is neither bounded above nor bounded

below. Therefore
lim sup(1)k k = , lim inf (1)k k = .
k k
 19

The key to learning how to use lim sup and lim inf is an understand-
ing of the following characterizations. These should be compared with
the characterization of the limit of a convergent sequence given by
Proposition 2.5
Proposition 2.10. Let {bk } be a sequence in R. Let b R. Then
 
(2) lim sup bk = b lim inf bk = b ,
k k

if and only if
(3) b<c = bk < c ultimately (frequently) ,
and
(4) a<b = a < bk frequently (ultimately) .
Proof: We give the proof of the lim sup assertion. The lim inf assertion
is proved similarly.
() Suppose that (2) holds. Let a < b and c > b. Then because
b = lim sup bk = lim bk ,
k k

where bk = sup{bl : l k}, it follows that

a < bk < c ultimately .
Because bk bk for every k, we see directly that bk < c ultimately,
whereby (3) holds. Moreover, a < bk implies that for some l k
one has a < bl . (Otherwise a would be an upper bound for the set
{bl : l k}, which contradicts the fact b is the least upper bound of
this set.) Hence, a < bk frequently, whereby (4) holds.
() Suppose that (3) and (4) hold. Let a < b and c > b be arbitrary.
Then (3) implies that
bk = sup{bl : l k} c ultimately ,
while (4) implies that
a < bk = sup{bl : l k} ultimately .
(If for some k one had bk a then bl a for every l k, which
contradicts (4).) It follows that
a inf{bk } c .
But a < b and c > b were arbitrary, so that
lim sup bk = inf{bk } = b ,
k

whereby (2) holds. 

20

The convergence of a sequence may be characterized in terms of the

lim sup and lim inf as follows.
Proposition 2.11. Let {ak } be a sequence in R. Then {ak } converges
if and only if
< lim inf ak = lim sup ak < ,
k k

in which case
lim ak = lim inf ak = lim sup ak .
k k k

Proof: One proof was given in class. Can you find a proof that uses
Proposition 2.10?
When adding and comparing sequences, lim sup and lim inf generally
behave as follows.
Proposition 2.12. Let {ak } and {bk } be sequences in R. Then
lim sup(ak + bk ) lim sup ak + lim sup bk ,
k k k
lim inf (ak + bk ) lim inf ak + lim inf bk ,
k k k

whenever the sum on the right-hand side is defined. Moreover, if ak

bk ultimately then
lim sup ak lim sup bk , lim inf ak lim inf bk .
k k k k

Proof: Exercise.
When multiplying sequences, lim sup and lim inf generally behave as
follows.
Proposition 2.13. Let {ak } and {bk } be sequences in R. If {ak } is
convergent with
lim ak = a > 0 ,
k

then
lim sup ak bk = a lim sup bk , lim inf ak bk = a lim inf bk .
k k k k

Proof: Exercise.
When they converge, the lim sup and lim inf of a sequence are actu-
ally limits of some of its subsequences.
 21

Proposition 2.14. Let {ak } be a sequence in R. If {ank } is any sub-

sequence of {ak } then
lim inf ak lim inf ank lim sup ank lim sup ak .
k k k k

Moreover, there exist subsequences {ank } and {amk } such that

lim ank = lim sup ak , lim amk = lim inf ak .
k k k k

Proof: We outlined a proof in class that used Proposition 2.10.

In particular, the following famous theorem of Weierstrass is an imme-
diate consequence.
Proposition 2.15. (Weierstrass Theorem) Every bounded sequence
in R has a convergent subsequence.
Proof: Let {bk } be a bounded sequence in R. This implies that there
exists [a, c] R such that {bk } [a, c]. Then
< a lim inf bk lim sup bk c < .
k k
The result then follows by Proposition 2.14. 

2.5. Cauchy Criterion. When a sequence is monotonic, just know-

ing that it is bounded tells you that it is convergent. When a sequence
is not monotonic, determining whether it is convergence or divergent
is generally much harder. For example, to establish convergence di-
rectly from Definition 2.5 you must first know the limit of the sequence.
Cauchy introduced a criterion for convergence that does not require
knowledge of the limit.
Definition 2.9. A sequence {ak }kN R is said to be Cauchy when-
ever for every  > 0 there exists N N such that
(5) k, l N = |ak al | <  .
In other words, a sequence is Cauchy if for every  > 0 one can find a
tail of the sequence such that any two terms in the tail are within  of
each other. Roughly speaking, a Cauchy sequence is one whose terms
generally get closer together.
The main result of this section is the so-called Cauchy criterion for
convergence namely, that a sequence in R is convergent if and only
if it is Cauchy. The easier half of this criterion is established by the
following.
Proposition 2.16. A convergent sequence in R is Cauchy.
22

Proof: Let {ak } be a convergent sequence in R with limit a. Let  > 0.

Then by the definition of convergence there exists N N such that

k N = |ak a| < .
2
It follows from the triangle inequality that if k, l N then
 
|ak al | = (ak a) + (a al ) |ak a| + |al a|
 
< + = .
2 2
Hence, the sequence {ak } is Cauchy. 
We now take the first step towards establishing the harder half of
the Cauchy criterion.
Proposition 2.17. A Cauchy sequence in R is bounded.
Proof: The proof is very similar to the proof that a convergent se-
quence is bounded. It is left as an exercise. 
We are now ready to establish the Cauchy Criterion.
Proposition 2.18. (Cauchy Criterion Theorem) A sequence in R
is convergent if and only if it is Cauchy.
Proof: Proposition 2.16 established that convergent sequences are
Cauchy. We only need to establish the other direction.
Let {ak } be a Cauchy sequence in R. By Proposition 2.17 the se-
quence {ak } is bounded. By the Weierstrass Theorem (Proposition
2.15) it has a convergent subsequence {ank }. Let a be the limit of
this convergent subsequence. We will use the fact {ak } is a Cauchy
sequence to show that it converges to a.
Let  > 0. Because the subsequence {ank } converges to a while the
sequence {ak } is Cauchy, there exists an N N such that

k N = |ank a| < ,
2
and

k, l N = |ak al | < .
2
Because k N implies that nk N , the line above implies that

k N = |ak ank | < .
2
It follows from the triangle inequality that if k N then
 
|ak a| |ak ank | + |ank a| < + =  .
2 2
The sequence {ak } therefore converges to a. 
 23

2.6. Closures and Closed Sets.

Definition 2.10. Given any A R its closure is given by
n o
Ac = a R : a is the limit of a sequence in A .

Remark. It is clear that A Ac for every A R. Indeed, every

a A is the limit of the constant sequence {ak } with ak = a for every
k N.

Proposition 2.19. For every A, B R one has that

(i) AB = Ac B c ,
(ii) (A B)c = Ac B c ,
(iii) (A B)c Ac B c .

Proof: Exercise.
An important fact is that the closure of Q is R. In other words,
every real is the limit of a sequence of rationals.
Proposition 2.20. Qc = R.

Proof: Let a R. Consider the sequence of intervals {Ik }kN where

each Ik is given by
Ik = a 21k , a + 21k .

For each k N the third assertion of Proposition 1.10 implies there

exists a ak Ik Q. The step of showing that ak a as k is
left as an exercise. It then follows that a Qc , whereby the assertion
follows. 
Definition 2.11. A subset A of R is said to be closed when A = Ac .

Consistency of our terminology would seem to demand that closures

be closed. This is indeed the case.
Proposition 2.21. Let A R. Then Ac is closed (i.e. (Ac )c = Ac ).

Proof: Let a (Ac )c . We must show that a Ac . Because a (Ac )c

there exists a sequence {bi }iN in Ac such that bi a as i . If
bi = a for some i N then a = bi Ac .
On the other hand, if bi 6= a for every i N then because bi Ac for
each i N there exists a sequence {b(i,j) }jN in A such that b(i,j) bi
24

as j . The picture is
b(0,0) , b(0,1) , b(0,2) , b(0,j) , b0
b(1,0) , b(1,1) , b(1,2) , b(1,j) , b1
b(2,0) , b(2,1) , b(2,2) , b(2,j) , b2
.. .. .. .. .. .. ..
. . . . . . .
b(i,0) , b(i,1) , b(i,2) , b(i,j) , bi
.. .. .. .. .. .. ..
. . . . . . .

a,
Because for each i N we have b(i,j) bi as j and |bi a| > 0,
there exists a ji N such that
 
b(i,j ) bi  < |bi a| .
i

Set ai = b(i,ji ) for each i N. It is clear that the sequence {ai }iN
lies within A. The step of showing that ai a as i is left as an
exercise. It then follows that a Ac , whereby the assertion follows. 
Examples. The empty set is closed. Intervals of the form [a, b],
[a, ), (, b], and R = (, ) are closed, while intervals of the
form (a, b), (a, b], [a, b), (a, ), and (, b) are not. Moreover,
(a, b)c = (a, b]c = [a, b)c = [a, b] ,
(a, )c = [a, ) , (, b)c = (, b] .
You should be able to prove these facts.

Proposition 2.22. If A and B are closed subsets of R then A B and

A B are closed.
If {Ak }kN is a sequence of closed subsets of R then
\
Ak is closed .
kN

Proof: Exercise.
Remark. By repeated application of the first assertion above, we see
that the union and intersection of any finite collection of closed sets
is again closed. The second assertion states that the intersection of
any countable collection of closed sets is again closed. The analogous
statment for unions is generally false. Indeed, consider the countable
collection of closed intervals {Ik }kN where each Ik is given by
Ik = 1 + 21k , 1 21k .
 
 25

You can easily show that

[
Ik = (1, 1) ,
kN
which is not closed.
Finally, we have the concept of a set being dense in a larger one.
Definition 2.12. Let A B R. Then A is said to be dense in B if
B Ac .
Examples. Proposition 2.20 states the Q is dense in R. In a similar
manner one can show that (a, b) Q is dense in [a, b], that (a, ) Q
is dense in [a, ), and (, b) Q is dense in (, b].
Proposition 2.23. If A B C D R and A is dense in D then
B is dense in C.
Proof. Exercise.
2.7. Sequential Compactness.
Definition 2.13. A set A R is said to be sequentially compact if
every sequence in A has a subsequence that converges to a limit in A.
Sequentially compact subsets of R are characterized by the following.
Proposition 2.24. A set A R is sequentially compact if and only if
A is closed and bounded.
Proof: () Let A R be closed and bounded. Let {ak } be an ar-
bitrary sequence in A. Because A is bounded, the sequence {ak } is
bounded. By the Weierstrass Theorem, {ak } has a converging sub-
sequence {ank }. Let a be the limit of this subsequence. Because A is
closed and {ank } is in A, the limit a must also be in A. By the arbitrari-
ness of {ak }, we conclude that every sequence in A has a subsequence
that converges to a limit in A. Therefore A is sequentially compact.
() Suppose that A is either not bounded or not closed. Here we
give the proof for the case when A is not closed. The proof for the case
when A is not bounded is left as an exercise.
Because A is not closed there exists a sequence {ak } in A and a
point a / A such that ak a as k . By Proposition 2.6 every
subsequence of {ak } also converges to a, which is not in A. Therefore
A is not sequentially compact. 
26

3. Series of Real Numbers

3.1. Infinite Series. Any finite set of real numbers can be summed.
Here we study one way to make sense of the sum of an infinite sequence
of real numbers.
Definition 3.1. Given any real sequence {ak }
k=0 , for every m, n N
with m n define the sigma notation:
n
X
ak am + am+1 + + an1 + an .
k=m

Associate with the sequence of terms {ak } the so-called sequence of

partial sums {sn } defined by
n
X
sn ak .
k=0

It is convenient to encode {sn } with the formal infinite series

X
ak .
k=0

If the sequence {sn } converges to a limit s then we say that the series
converges, and that s is the sum of the series. In that case we write

X
ak = s .
k=0

If the sequence {sn } diverges then we say that the series diverges.

Remark: It is clear that changing, adding, or removing a finite number

of terms in a series does not affect whether the series converges or
diverges, but if it converges, the sum would almost always be affected.
For example,

X
X
ak converges ak converges ,
k=0 k=5

but when they do converge the sums will generally differ namely,

X
X
in general ak 6= ak .
k=0 k=5

Specifically, these sums will be equal only when a0 +a1 +a2 +a3 +a4 = 0.
 27

Example: Consider the infinite series

X 1
.
k=1
k(k + 1)

The nth partial sum is given by

n n  
X 1 X 1 1
sn = =
k=1
k(k + 1) k=1 k k + 1
     
1 1 1 1 1
= 1 + ++
2 2 3 n n+1
   
1 1 1 1 1
=1+ + ++ +
2 2 n n n+1
1
=1 .
n+1
One then sees that the series is convergent with

X 1
= lim sn = 1 .
k(k + 1) n
k=1

The previous example is a special case of series that can be explicitly

put into telescoping form.
Definition 3.2. A formal inifinite series with terms {ak }
k=m is said to
be in telescoping form if ak = ck1 ck for some sequence {ck }k=m1 ,
so that the series is expressed as
X
(6) (ck1 ck ) .
k=m

If a series can be put into telescoping form with a sequence {ck } that
is known explicitly then the convergence or divergence of the series can
be easily determined. Moreover, if it converges then its sum can be
easily determined. This is because the sequence {ck } is simply related
to the sequence of partial sums {sn }. Indeed, for every n m one has
that
X n Xn
sn = ak = (ck1 ck )
k=m k=m
= (cm1 cm ) + (cm cm+1 ) + + (cn2 cn1 ) + (cn1 cn )
= cm1 + (cm + cm ) + + (cn1 + cn1 ) cn
= cm1 cn .
28

It immediately follows that the sequence {sn } converges if and only if

the sequence {ck } converges, and that if these sequences converge then
lim sn = cm1 lim ck .
n k

Hence, the following proposition holds.

Proposition 3.1. Let {ck }
k=m1 be a real sequence. Then

X
(ck1 ck ) converges {ck }
k=m1 converges .
k=m

Moreover, when these are convergent one has

X
(ck1 ck ) = cm1 lim ck .
k
k=m

Remark: The above considerations show that if a series is in the

telescoping form (6) then there is a c R such that ck = c sk
for every k m, where {sk } k=m is the sequence of partial sums. This
means that finding an explicit telescoping form for a series is equivalent
to finding an explicit expression for its partial sums. It should be clear
that one can do this in only the rarest of cases.
Given a general infinite series, it is usually impossible to evaluate
the limit of the sequence of partial sums. However, one can commonly
determine whether a series is convergent or divergent without explicitly
evaluating this limit. The following proposition gives the simplest test
for divergence.
Proposition 3.2. (Divergence Test) Let {ak } be a real sequence.

X
If the series ak converges then lim ak = 0.
k
k=0

X
Equivalently, if lim ak 6= 0 then the series ak diverges.
k
k=0

Proof: The proof is based on the fact that the k th term in a formal
infinite series can be expressed as ak = sk sk1 , where s1 = 0 and
{sk }kN is the sequence of partial sums. If the series converges then
one knows that
lim sk = s , lim sk1 = s ,
k k

where s is the sum of the series. It thereby follows that

lim ak = lim sk lim sk1 = s s = 0 . 
k k k
 29

Remark: You can easily find examples of a series whose terms con-
verge to zero, yet the series is divergent. One such example is the
harmonic series:

X 1
.
k
k=1

Clearly 1/k 0 as k . However, we will soon show that this

series diverges.

3.2. Geometric Series. An important example is that of geometric

series.
Definition 3.3. A formal infinite series of the form

X
ar k
k=0

for some nonzero a and some r R is called a geometric series.

The convergence or divergence of a geometric series is easy to de-
termine because it is one of those rare series where one can find an
explicit experssion for its partial sums. For every n N let sn denote
the partial sum given by
n
X
sn = ar k .
k=0

It is clear that if r = 1 then sn = (n + 1)a and the series will diverge.

So suppose that r 6= 1. One checks that
n
X n
X
k
sn rsn = ar ar k+1
k=0 k=0
n
X n+1
X
= ar k ar k = a ar n+1 ,
k=0 k=1

whereby the partial sum sn is found to be

a ar n+1
sn = .
1r
By letting n tend to in this expression we find that

( a
X if |r| < 1 ,
ar k = 1 r
k=0 diverges otherwise .
30

Remark: The fact the geometric series diverges when |r| 1 can also
be seen easily from the Divergence Test. Indeed, in that case one sees
that
lim ar k diverges (and hence is nonzero) ,
k
whereby the Divergence Test (Proposition 3.2) shows that the geomet-
ric series diverges. Of course, the Divergence Test does not show the
geometric series converges when |r| < 1.
Exercise. Consider a formal infinite series of the form
X
kr k
k=1
for some r R. Find all the values of r for which this series converges
and evaluate the sum. (Hint: Find an explicit expression for the partial
sums and evaluate the limit. The explicit expression may be derived
from the analogous expression for a geometric series.)

3.3. Series with Nonnegative Terms. If the terms of an infinite

series are nonnegative then the associated sequence of partial sums
will be nondecreasing. Hence, the least upper bound property can be
employed in the guise of the Monotonic Sequence Theorem (Proposition
2.8) to show the convergence or divergence of the series. Specifically,
one has the following proposition, which lies at the heart of most proofs
about the convergence or divergence of series with nonnegative terms.
Proposition 3.3. (Series with Nonnegative Terms Theorem)
Let {ak }
k=m be a nonnegative sequence. Then

X
ak converges {sk }
k=m is bounded above ,
k=m

where {sk }k=m is the sequence of partial sums associated with the formal
infinite series.
Proof: One first shows that the sequence {sk }
k=m is nondecreasing.
One then applies the Monotonic Sequence Theorem. The details are
left as an exercise. 
One way to establish whether or not a sequence of partial sums is
bounded above is to compare it with a sequence for which the answer is
known. This is often done with one of the following comparison tests.
Proposition 3.4. (Comparison Tests for Series with Nonnega-
tive Terms) Let {ak } and {bk } be nonnegative sequences that satisfy
one of the following comparison conditions:
 31

(i) the direct comparison

M R+ such that a k M bk ultimately ;
(ii) the limit comparison (if each bk is positive)
ak
lim sup < ;
k bk

(iii) the ratio comparison (if each ak and bk is positive)

ak+1 bk+1
< ultimately .
ak bk
Then
X X
(7) bk converges = ak converges ,
k=0 k=0

!
X X
ak diverges = bk diverges .
k=0 k=0

Proof:
P First, condition (i) implies (7) because it and the fact that
bk converges yields the upper bound
n
X Xn X
ak M bk M bk < .
k=0 k=0 k=0
P
Proposition 3.3 therefore implies that ak converges. Next, condition
(ii) implies condition (i) (and hence (7)) upon observing that
ak
lim sup < M < = ak M bk ultimately .
k bk

Finally, condition (iii) implies condition (ii) (and hence (7)) upon ob-
serving that
ak+1 bk+1
< ultimately
ak bk
ak+1 ak
= ultimately
bk+1 bk
 
ak
= is nonincreasing ultimately
bk
ak
= lim < . 
k bk

Exercise: The proof argues that the direct comparison test applies
whenever the limit comparison test applies, and that the limit compar-
ison test applies whenever the ratio comparison test applies. Can you
find an examples where (a) the direct comparison test applies but the
32

limit comparison test fails, or (b) the limit comparison test applies but
the ratio comparison test fails?
Example: We can apply the direct comparison test to show the har-
monic series diverges. Consider the comparison
1 1,
1 1
2
2
,
1 1
4
3
,
1 1
4
4
,
1 1
8
5
,
1 1
8
6
,
1 1
8
7
,
1 1
8
8
,
..
.

Summing both sides, one sees that
2n
n X1
1+ .
2 k
k=1

The partial sums clearly diverge.

3.4. Series with Nonincreasing Positive Terms. The harmonic

series is a special case of the so-called p-series, which is formally given
by

X 1
(8) .
k=1
kp

Because the terms of this series are nonincreasing and positive, the
following two convergence tests can be applied. Proposition 3.3 plays
a central role in their proofs.
Proposition 3.5. (Cauchy 2k Test) Let {ak } be a nonincreasing,
positive sequence. Then

X
X
ak converges 2k a 2k converges .
k=0 k=0
 33

Proof: The result is a consequence of the direct comparisons:

a2 a2 a1 ,
a4 a3 a2 ,
a4 a4 a2 ,
a8 a5 a4 ,
a8 a6 a4 ,
a8 a7 a4 ,
a8 a8 a4 ,
..
.

These can be written in the general form
a2j ak a2j1 for 2j1 < k 2j .
The details are left as an exercise. 
p
Example: Because {1/k } is a nonincreasing, positive sequence, Propo-
sition 3.5 implies that the p-series (8) converges or diverges as the series

k 1
X X
k
2 kp = 21p .
k=0
2 k=0
But this is a geometric series that clearly converges for p > 1 and
diverges for p 1.
Remark: The proof of the Cauchy 2k test suggested above is an exten-
sion of the argument we used to show that the harmonic series diverges.
Indeed, the harmonic series is just the p-series for p = 1.
Remark: The Cauchy 2k test can be generalized to subsequences of
{ak } of the form {ank } where there exist constants m and m such that
nk+1 nk
0< m m < .
nk nk1
For example, one can choose nk = mk for some m N with m > 1.
This satisfies
nk+1 nk m1
= = m,
nk nk1 1 m1
whereby m = m = m. This leads to a mk test The statement and
proof is left to you, but it is not on the homework.
34

The second convergence test of this section requires the use of inte-
grals in fact, the use of improper integrals. These will be developed
rigorously later in the course. However, here we will assume you have
some familiarity with them from your elementary calculus courses.
Proposition 3.6. (Integral Test) Let f be a nonincreasing, positive,
locally integrable (continuous, for example) function over [0, ). Then
X Z
f (k) converges f (x) dx converges ,
k=0 0

where the integral is understood in the sense of an improper integral.

Proof: The key fact we need from integration theory is that the im-
proper integral
Z
f (x) dx converges
0

whenever the sequence {Sn } converges, where each Sn is defined by

Z n
Sn = f (x) dx .
0

Because {Sn } is an increasing sequence, showing convergence reduces

to showing it is bounded above.
The result will then be a consequence of the fact that
Xn Z k
Sn = f (x) dx for every n Z+ ,
k=1 k1

and the direct comparisons

Z k
f (k) f (x) dx f (k 1) , for every k Z+ .
k1

These facts should be clear to you based on your knowledge of definite

integrals from elementary calculus. If not, a picture should help clarify
things. We will establish them rigorously later. Here we will assume
they are true and complete the proof.
By summing the above direct comparisons, one obtains
n
X n
X n1
X
f (k) Sn f (k 1) = f (k) .
k=1 k=1 k=0

The remaining details are left as an exercise. 

 35

Example: Because {1/k p } is a nonincreasing, positive sequence, Propo-

sition 3.6 implies that the p-series (8) converges or diverges as the im-
proper integral Z
1
dx .
1 xp
But for p 6= 1 one can easily check that
Z k  
1 1 1
Sk = dx = 1 p1 ,
1 x
p p1 k
while for p = 1 one has
Z k
1
Sk = dx = log(k) .
1 x
(Here log( ) denotes the natural logarithm.) One then sees that the
sequence {Sk } converges for p > 1 and diverges for p 1. The same is
therefore true for the p-series.

3.5. Alternating Series. Until now we have only studied convergence

tests for nonnegative series. The underlying tool has been the Mono-
tonic Sequence Theorem (Proposition 2.8), which was used to prove
Proposition 3.3. Here we use the Monotonic Sequence Theorem to ob-
tain the following characterization of convergence for a special class of
series with alternating sign.
Proposition 3.7. (Alternating Series Test) Let {ak } be a positive,
nonincreasing sequence in R. Then
X
(1)k ak converges lim ak = 0 .
k
k=0

Proof: The direction = is just Proposition 3.2 (Divergence Test).

To prove the other direction, let
X n
sn = (1)k ak .
k=0
First, the picture is that
{s2k }kN is nonincreasing , while {s2k+1 }kN is nondecreasing ,
and that
s2k > s2j+1 for every j, k N .
Indeed, the first two assertions follow because
s2k+2 s2k = a2k+2 a2k+1 0 ,
s2k+3 s2k+1 = a2k+3 + a2k+2 0 .
36

Next, because s2k > s2k+1 for every k N, for any j k one has
s2k > s2k+1 s2j+1 , s2j s2k > s2k+1 .
The result follows by exchanging j and k in the last inequality. The
monotonic subsequences {s2k } and {s2k+1 } are thereby bounded below
and above respectively. By the Monotonic Sequence Theorem they
therefore converge. Let
s = lim s2k , s = lim s2k+1 .
k k
Then

s s = lim s2k s2k+1 = lim a2k+1 = 0 ,
k k
whereby s = s. The last step is to show that this fact implies that {sk }
converges. This is left as an exercise. 
Examples:

X (1)k X (1)k
converges for p > 0 , converges .
k=1
kp k=2
log(k)

3.6. Absolute Convergence. The Monotonic Sequence Theorem has

been the tool underlying all the convergence tests we have studied so
far. We now use the Cauchy criterion to establish a test that does not
require the series to be nonnegative.
Proposition 3.8. (Absolute Convergence Test) Let {ak } be a real
sequence. Then

X
X
|ak | converges = ak converges .
k=0 k=0

Proof: Let {pn } and {qn } be the sequences of partial sums given by
X n Xn
pn = |ak | , qn = ak .
k=0 k=0

By hypotheses {pn } is convergent, and thereby Cauchy. The idea of

the proof is to show that {qn } is Cauchy, and thereby convergent.
The key to doing so is the fact that for every m, n N one has the
inequality
|qn qm | |pn pm | .
This is trivially true when m = n. When n > m the triangle inequality
yields  
 X n  Xn
|qn qm | =  ak  |ak | = |pn pm | .
 
 
k=m+1 k=m+1
 37

The case n < m goes similarly.

Let  > 0. Because {pn } is Cauchy there exists an N N such that
m, n N = |pn pm | <  .
Because |qn qm | |pn pm |, one immediately sees that
m, n N = |qn qm | <  .
Hence, {qn } is Cauchy, and thereby convergent. 
Proposition 3.8 motivates the following definition.
Definition 3.4. If {ak } is a real sequence such that

X
|ak | converges ,
k=0

then one says

converges absolutely ,
X
ak or
k=0 is absolutely convergent .

Example: Consider the alternating p-series

X (1)n
p
.
n=1
n
This converges for p > 0 by the alternating series test, but it con-
verges absolutely only for p > 1. This example shows that not every
convergent series is absolutely convergent. In other words, absolute
convergence is a stronger property than convergence.
When the definition of absolute convergence is combined with the
Comparison Tests for series with nonnegative terms (Proposition 3.4),
we get an array of new comparison tests for absolute convergence that
can be applied to general series.
Proposition 3.9. (Absolute Comparison Tests) Let {ak } and {bk }
be real sequences that satisfy one of the following comparison condi-
tions:
(i) the direct comparison
M R+ such that |ak | M bk ultimately ;
(ii) the limit comparison (if each bk is positive)
|ak |
lim sup < ;
k bk
38

(iii) the ratio comparison (if each |ak | and bk is positive)

|ak+1 | bk+1
< ultimately .
|ak | bk
Then

X
X
(9) bk converges = ak converges absolutely .
k=0 k=0

Proof: Exercise.
Example: Because | cos(kx)| 1 for every x R and k Z+ , direct
comparison with the p-series shows that the series

X cos(kx)
converges absolutely for p > 1 .
k=1
kp

3.7. Root and Ratio Tests. The root and ratio tests both draw their
conclusions about the convergence of a series based on absolute com-
parisons with a geometric series.
Proposition 3.10. (Root Test) Let {ak } be a real sequence. Let
p
= lim sup k |ak | .
k
Then

X
<1 = ak converges absolutely ,
k=0
X
>1 = ak diverges .
k=0
If = 1 the series may either converge or diverge.
Proof: The convergence conclusion when < 1 follows by a direct
comparison of the series with a convergent geometric series. Specifi-
cally, by Proposition 2.10 one has that
p
lim sup k |ak | < r < 1 = |ak | < r k ultimately .
k
The absolute convergence follows from the direct comparison test of
Proposition 3.9.
The divergence conclusion when > 1 follows by showing that
lim sup |ak | > 0. Specifically,
p
lim sup k |ak | > r > 1 = |ak | > r k frequently . 
k
 39

This implies there exists a subsequence {ank }k of {ak }k such that

|ank | > r nk ultimately .
Then lim sup |ak | lim sup |ank | lim r nk = . But lim sup |ak | > 0
implies the sequence {ak } does not converge to zero, which by the
Divergence Test (Proposition 3.2) implies the associated series diverges.
We leave as an exercise them problem of finding examples of both a
convergent and a divergent series with = 1. 

Proposition 3.11. (Ratio Test) Let {ak } be a nonzero real sequence.

Then

|ak+1 | X
lim sup < 1 = ak converges absolutely ,
k |ak | k=0

|ak+1 | X
1 ultimately = ak diverges .
|ak | k=0

Proof: As with the proof of the root test, the convergence conclusion
follows by a direct comparison of the series with a convergent geo-
metric series, while the divergence conclusion follows by showing that
lim sup |ak | > 0. Specifically, by Proposition 2.10 one has that
|ak+1 | |ak+1 |
lim sup < r < 1 = < r ultimately .
k |ak | |ak |
An induction argument can then be used to show that for some m N
one has
|ak | |am |r km for every k m .
Because the geometric series

X |am | k
r converges ,
k=m
rm
the comparison theorem implies
X
ak converges .
k=0

The proof of the divergence assertion is left as an exercise. 

Remark: Some books give the divergence criterion of the ratio test as

|ak+1 | X
lim inf > 1 = ak diverges .
k |ak |
k=0

This is clearly weaker than the one we give.

40

Remark: The root test is harder to apply, but as the following indi-
cates, its convergence assertion can be sharper.
Proposition 3.12. Let {ak } be a positive sequence. Then
ak+1 ak+1
lim inf lim inf k ak lim sup k ak lim sup .
k ak k k k ak
Proof: Exercise. (The middle inequality is obvious, so just prove the
other two.) 
Remark: Because both the root and ratio tests draw their conclusion
about the convergence of a series based on comparison with a geometric
series, they should only be used when such a comparison makes sense.
For example, these tests can be used to assert the absolute convergence
of series like

X
4 k
X
l2 l
X (m!)2
k 2 , e 4 , (3)m ,
m=0
(2m)!
k=1 l=0

but will not yield any information about the convergence of series like
 1
X log(k) X 3l + 2 2 X (1)m
, , .
k=2
k2 l=0
l4 + 2 m=2
m(log(m))2

Example: Find the least upper bound and greatest lower bound of
the set
(
)
X (3n)! (2n)! n
S= xR : x converges .
n=0
n! (4n)!
This can be easily done employing the ratio test. Indeed, because
(3n + 3)! (2n + 2)! n+1
|x|
|an+1 | (n + 1)! (4n + 4)!
=
|an | (3n)! (2n)! n
|x|
n! (4n)!
(3n + 3)(3n + 2)(3n + 1)(2n + 2)(2n + 1)
= |x| .
(n + 1)(4n + 4)(4n + 3)(4n + 2)(4n + 1)
one finds that
|an+1 | 33 2 2 33
lim = |x| = |x| .
n |an | 44 43
It follows from the ratio test that the series converges when |x| < (4/3)3
and diverges when |x| > (4/3)3 . The least upper bound of S is therefore
(4/3)3 while the greatest lower bound is (4/3)3 .
 41

3.8. Dirichlet Test. We now apply the Cauchy criterion to establish

a test that, like the Alternating Series Test, can be applied to series
that do not converge absolutely.
Proposition 3.13. (Dirichlet Test) Let {ak }kN be a positive, non-
increasing sequence in R such that
lim ak = 0 .
k

Let {bk }kN be a sequence in R for which there exists M such that
 n 
X 
(10) 
 bk  M for every n N .
k=0

Then

X
a k bk converges .
k=0

Remark: The Dirichlet Test implies the convergence conclusion of the

Alternating Series Test. Indeed, if we set bk = (1)k then
n n
(
X X 1 for n even ,
bk = (1)k =
k=0 k=0
0 for n odd .
Hence, bound (10) holds with M = 1. The Dirichlet Test then tells us
that for every positive, nonincreasing real sequence {ak }kN such that
ak 0 as k the series
X
(1)k ak converges .
k=0

As the following example illustrates, the Dirichlet Test is far more

powerful than the Alternating Series Test.
Example: Consider the problem of determining all x, p R for which
the Fourier p-series

X cos(kx)
converges .
kp
k=1

Because | cos(kx)| 1 for every k Z+ , direct comparison with the

regular p-series shows that this series converges absolutely for p > 1.
However this argument says nothing about what happens when p 1.
First observe that when x {2m : m Z} one has cos(kx) = 1
for every k Z+ . In this case the Fourier p-series reduces to a regular
p-series, which diverges for every p 1.
42

Next observe that when x {(2m+1) : m Z} one has cos(kx) =

(1)k for every k Z+ . In this case the Fourier p-series reduces to an
alternating p-series, which (by the Alternating Series Test) converges
for every p > 0 and diverges for every p 0.
We now use the Dirichlet Test to analyze the more general case
when x / {2m : m Z}. Let ak = 1/k p and bk = cos(kx). Clearly
the sequence {ak } is positive, decreasing, and vanishes as k . The
hard step is to show that the partial sums associated with the sequence
{bk } satisfy (10). To do this we use the trigonometric identity

2 sin( 21 x) cos(kx) = sin((k + 21 )x) sin((k 21 )x) ,

and the fact sin( 21 x) =

6 0 when x
/ {2m : m Z} to obtain (by a
telescoping sum) the formula
n n n
X X X sin((k + 21 )x) sin((k 21 )x)
bk = cos(kx) =
k=1 k=1 k=1
2 sin( 12 x)
sin((n + 21 )x) sin( 21 x)
= .
2 sin( 12 x)
It is clear from this formula that
 n   sin((n + 12 )x) sin( 21 x)  | sin((n + 12 )x)| + | sin( 21 x)|
X   
 bk  =  

k=1
2 sin( 1 x) 2
 2 | sin( 1 x)| 2
1
.
| sin( 12 x)|

Hence, bound (10) holds with M = 1/| sin( 21 x)|. The Dirichlet Test
then implies that when x / {2m : m Z} the Fourier p-series
converges for every p > 0.
Finally, one can use the Divergence Test to show that the Fourier
p-series diverges for every p 0. We leave the details as an exercise.
You may use the fact that lim supk cos(kx) > 0. 

Remark: When applying the Dirichlet test to a given series, one must
identify the sequences {ak } and {bk }, and check that all the hypotheses
on them are satisfied. The hypotheses on {ak } are easy to check, so
do that first: the sequence {ak } must be positive, nonincreasing, and
vanish as k . The hypothesis on {bk } is typically much harder
to check: the associated partial sums must satisfy (10). The key to
checking this in the above example was to write bk = ck+1 ck (by using
a trigonometric identity) for some bounded sequence {ck }, whereby the
 43

partial sums telescoped as

X n Xn
bk = (ck+1 ck ) = cn+1 c0 .
k=0 k=0

This telescoping approach can be taken for a variety of other {bk } too.

Our proof of the Dirichlet Test uses an identity that is a discrete

analog of the integration-by-parts formula from calculus. Because it
has many other applications, this identity gets its own proposition.
Proposition 3.14. (Summation-by-Parts Identity) Let {ak }kN
and {bk }kN be sequences in R. Let B1 = 0 and
Xn
Bn = bk for every n N .
k=0

Then for every m, n N with m n one has identity

n
X n1
X
ak bk = an Bn am Bm1 + (ak ak+1 )Bk ,
k=m k=m

with the understanding that the last sum is zero when m = n.

Remark: This is called the summation-by-parts identity because it is

a discrete analog of the integration-by-parts formula
Z n n Z n
a(x)b(x) dx = a(x)B(x) a0 (x)B(x) dx ,

m m m
where B (x) = b(x).
0

Proof: Because bk = Bk Bk1 we have

Xn n
X
a k bk = ak (Bk Bk1 )
k=m k=m
Xn n
X
= ak Bk ak Bk1
k=m k=m
n
X n1
X
= ak Bk ak+1 Bk
k=m k=m1
n1
X
= an Bn am Bm1 + (ak ak+1 )Bk .
k=m

To get from the second to the third line above we re-indexed the last
sum. All the other steps are straightforward algebra. 
44

We now turn to the proof of the Dirichlet Test.

Proof of Dirichlet Test: Let
Xn
sn = a k bk .
k=0

We will show the sequence {sk }kN is Cauchy, and therefore convergent.
Let  > 0. We seek N N such that
m, n N = |sn sm | <  .
For m = n this is always true. Suppose m < n. (For the case n < m
simply reverse the roles of m and n.) Then
 n 
 X 
|sn sm | = 
 ak bk 
k=m+1
 n1
X

 
= an Bn am+1 Bm + (ak ak+1 )Bk 
k=m+1
n1
X
an |Bn | + am+1 |Bm | + (ak ak+1 )|Bk |
k=m+1
n1
X
an M + am+1 M + (ak ak+1 )M
k=m+1
= 2am+1 M .
Here we have used the summation-by-parts identity in the second line,
the triangle inequality and the fact that {ak }kN is positive and non-
increasing in the third line, the bound |Bk | M in the fourth line,
and evaluated the telescoping sum in the last line. Because ak 0 as
k , we can choose N so that m N implies 2am+1 M < . Hence,
for every n > m N the above inequalities imply |sn sm | < .