This action might not be possible to undo. Are you sure you want to continue?
MATH1115, 1999
John Hutchinson
Author address:
Department of Mathematics, School of Mathematical Sciences,
Australian National University
Email address: John.Hutchinson@anu.edu.au
CHAPTER 1
Introduction
This year I will take a slightly diﬀerent approach to the material. The main
reference will be the 1998 Notes, but adjusted as indicated in the current Notes.
The main diﬀerence this year is that sequences will be introduced earlier and
will play a more central role. One still gets the same results in the end, but I think
the main ideas will be easier to understand in this manner.
i
CHAPTER 2
The Real Number System
Sections 2.1–3
As before.
Section 2.4
Omit this section.
Thus at this stage we assume the real number system satisﬁes Axioms 1–13.
All the usual properies of addition, multiplication, subtraction, division, and of
inequalities, are consequences
As noted in Remark 2.6(98), the set of rational numbers also satiﬁes Axioms
1–13, and hence have all these same properties. (However, certain existence prop
erties, such as the existence of a number x for which x
2
= 2, are not true of the
rationals, see Section 5.)
We will introduce the Completeness Axiom after we discuss sequences. It will
be in a diﬀerent, but equivalent, form to that in the 98 Notes. The Completeness
Axiom is true for the set of real numbers but the analogous statement is not true
for the set of rational numbers.
Section 2.5
In the summary, omit everything except the ﬁrst sentence “We prove that
√
2
is irrational”.
Omit everything in Section 2.5 from Theorem 2.12 onwards, leaving only The
orem 2.11 which shows that
√
2 is irrational. More precisely, what Theorem 2.11
proves is that if there is a number whose square is 2 then that number cannot be
rational.
( One cannot prove from Axioms 1–13 that there is a positive number whose
square is 2. The reason is that Axioms 1–13 also hold for the rational numbers, and
so if we could prove this fact then it would also be true in the rational numbers,
i.e. there would be a a positive rational number whose square is 2. But this would
contradict Theorem 2.11. )
Sections 2.6–10
As before. Note that the last three sections are marked , i.e. extra non
examinable material to put the other material in a broader context.
CHAPTER 3
LIMITS
Omit this chapter. We will return to it later, after ﬁrst treating sequences.
1
CHAPTER 4
Sequences
This replaces the chapter from the 98 Notes.
The reference for this chapter is [Adams, Section 10.1], but we do considerably more
material than this. Another reference is Chapter 2 of Fundamentals of Analysis by Michael
Reed; this book will also be the text for the second year course “Analysis and its applica
tions” beginning next year.
4.1. Examples of sequences
We introduce the idea of a sequence and give a few examples.
• A sequence is an inﬁnite list of numbers with a ﬁrst, but no last, element. Simple
examples are
1, 2, 1, 3, 1, 4, . . .
1,
1
2
,
1
3
, . . .
1, −1, 1, −1, 1, . . .
• A sequence can be written in the form
a
1
, a
2
, a
3
, . . . , a
n
, . . . .
• More precisely, a sequence is a function f whose domain is the set of natural
numbers, where in the above example f(n) = a
n
. We often just write (a
n
) or (a
n
)
n≥1
to
represent the sequence.
• If the pattern is clear, we may just write the ﬁrst few terms, as in the ﬁrst three
examples above.
• The general term a
n
may instead be given by a formula, such as
a
n
=
_
1 +
1
n
_
n
,
which gives the sequence
1 + 1,
_
1 +
1
2
_
2
,
_
1 +
1
3
_
3
, . . . .
• A sequence may be given by a method for calculating each element of the sequence
in terms of preceding elements. One example is the Fibonacci sequence
a
1
= 1, a
2
= 1, a
n
= a
n−1
+ a
n−2
if n ≥ 3.
Hence the sequence is
1, 1, 2, 3, 5, 8, 13, 21, . . . .
• Sometimes it is convenient to write a sequence in the form
a
k
, a
k+1
, a
k+2
, . . .
where k is some other integer than 1 (e.g. 0).
• One can represent a sequence by its graph. For example the sequence
_
(−1)
n+1
/n
2
_
,
i.e. (1, −1/4, 1/9, −1/16, . . . ) has graph
2
.
.
.
.
.
.
.
1 2 3
4
5 6 7
0
1 1
a
3
a
6
a
1
a
5
a
4
a
2
4.2. LIMIT OF SEQUENCES 3
where the vertical scale is somewhat distorted. However, this is not usually useful. It is
often more helpful to think of a sequence geometrically in terms of points on the real line.
4.2. Limit of sequences
We discuss the idea of the limit of a sequence and give the precise deﬁnition.
We give some examples.
We are interested in the behaviour of sequences (a
n
)
n≥1
for large n. The idea is that
“the sequence (a
n
)
n≥1
converges to the limit a” if the terms of the sequence get closer
and closer to a as n gets larger and larger. This is not exactly what we want, as the terms
of the sequence
a
1
= 1, a
2
= 1
1
2
, a
3
= 1
2
3
, a
4
= 1
3
4
, . . .
get closer and closer to 2, but also to 3 (for example), as n gets larger and larger. Also,
the sequences
1, 1 +
1
2
, 1 −10
−3
, 1 +
1
4
, 1 −10
−5
, 1 +
1
6
, 1 −10
−7
, . . . ,
1, 1 +
1
2
, 1 + 10
−3
, 1 +
1
4
, 1 + 10
−5
, 1 +
1
6
, 1 + 10
−7
, . . . ,
both converge to 1, but the distance a
n
−1 between a
n
and 1 is not a decreasing function
in either case.
A more precise version of what we mean by “the sequence (a
n
)
n≥1
converges to the
limit a” is that beyond some term in the sequence, all terms are within .1 of a; beyond a
further point in the sequence all terms are within .01 of a; beyond a further point in the
sequence all terms are within .001 of a; etc.
For example, we may have
a
n
−a ≤ .1 if n ≥ 50,
a
n
−a ≤ .01 if n ≥ 300,
a
n
−a ≤ .001 if n ≥ 780,
etc.
More precisely:
no matter which positive number is chosen (let’s call it ε), there is always
an integer (let’s call it N, or N(ε) to emphasise that it may depend
on ε), such that the Nth and all later members of the sequence are
within ε of a.
In the above situation we could choose N as follows:
4.2. LIMIT OF SEQUENCES 4
ε N
.1 50
.01 300
.001 780
For example, according to this criterion, the sequence 1, −1/4, 1/9, −1/16, . . . con
verges to 0, the sequence 1,
1
2
,
1
3
, . . . converges to 0, the sequence 1, −1, 1, −1, 1, . . . does
not converge (it “oscillates back and forth between ±1”), the sequence 1, 2, 1, 3, 1, 4, . . .
does not converge and the sequence 1, 4, 9, 16, . . . also does not converge (we sometimes
say that it “diverges to +∞”).
Often it is not clear whether or not a sequence converges. For example, it is not
immediately clear if the sequences
_
(1 +
1
n
)
n
_
n≥1
,
_
nsin
1
n
_
n≥1
, or the sequence given by
a
1
= 1 and a
n+1
=
1
2
a
n
+ 2 for n ≥ 1, will converge. For this and more complicated
examples we need a precise deﬁnition of convergence. Also, in order to prove general
theorems about convergent sequences, we need to have a precise deﬁnition.
The following deﬁnition makes our previous idea very precise.
Definition 4.1. We say that the sequence (a
n
) converges to a limit a, and write
lima
n
= a, or a
n
→ a,
if for every positive number ε there exists an integer N such that
n ≥ N implies a
n
−a ≤ ε. (4.1)
Note that a
n
− a ≤ ε is equivalent to a
n
∈ [a − ε, a − ε]. Thus another way of
expressing the deﬁnition is “for every interval [a − ε, a + ε] (provided ε > 0) there is an
integer N such that all members of the sequence from the Nth onwards belong to this
interval”.
We sometimes say a
n
converges to a as n approaches ∞ and write
lim
n→∞
a
n
= a or a
n
→ a as n → ∞.
(Note that ∞ is not a number, and the symbol ∞ by itself here has no meaning, just as
→ has no meaning by itself.)
Example 4.2. Show that the sequence given by a
n
= 1+
1
n
2
converges to 1 according
to the deﬁnition.
Solution. Let ε > 0 be given.
We want to ﬁnd N such that (4.1) is true with a = 1. But
a
n
−1 =
1
n
2
.
Since
1
n
2
≤ ε if n
2
≥
1
ε
,
i.e.
if n ≥
1
√
ε
,
we can take N to be any integer ≥ 1/
√
ε, e.g. take
N =
_
1
√
ε
_
+ 1,
where [ ] denotes “the integer part of”.
Thus if ε = .1 we can take any integer N ≥ 1/
√
.1, for example N = 4 (or anything
larger). If ε = .01 we can take N = 10 (or anything larger). If ε = .001 we can take
N = 32 (or anything larger). But the above proof works of course for any ε > 0.
4.2. LIMIT OF SEQUENCES 5
Example 4.3. We previously mentioned the sequence given by a
1
= 1, and a
n+1
=
1
2
a
n
+ 2 for n ≥ 1.
The ﬁrst few terms are
1, 2.5, 3.25, 3.625, 3.8125, 3.90625, 3.953125, 3.9765625, . . . .
It seems reasonable that the sequence is converging to 4. One way to prove this is as
follows.
Proof. Let ε > 0 be given.
We want to ﬁnd N such that
1
n ≥ N ⇒ a
n
−4 ≤ ε. (4.2)
We have a formula for a
n+1
in terms of a
n
, and we ﬁrst use this to get a formula for
a
n+1
−4 in terms of a
n
−4. Thus
a
n+1
−4 =
¸
¸
¸
1
2
a
n
+ 2 −4
¸
¸
¸ =
¸
¸
¸
1
2
a
n
−2
¸
¸
¸ =
¸
¸
¸
1
2
(a
n
−4)
¸
¸
¸ =
1
2
a
n
−4.
Thus a
1
− 4 = 3, a
2
− 4 = 3/2, a
3
− 4 = 3/2
2
, a
4
− 4 = 3/2
3
, . . . . In general
2
a
n
−4 = 3/2
n−1
.
It follows that
a
n
−4 ≤ ε for those n such that
3
2
n−1
≤ ε.
This last inequality is equivalent to 2
n−1
/3 ≥ 1/ε, i.e. 2
n−1
≥ 3/ε, i.e. (n − 1) ln 2 ≥
ln(3/ε), i.e. n ≥ 1 + ln(3/ε)/ ln 2.
Hence (4.2) is true for
N = 1 +
_
ln
3
ε
ln 2
_
.
You may object that we used ln, the natural logarithm, in the previous example, but
we have not yet shown how to deﬁne logarithms and establish their properties from the
axioms. This is a valid criticism. But in order to have interesting examples, we will often
do this sort of thing.
However, we will not do it when we are establishing the underlying theory. In partic
ular, the development of the theory will not depend on the examples.
Example 4.4. Show that lim
n→∞
c
n
p
= 0 for any real number c and any p > 0.
Solution. (See [Adams, Example 4, page 522]). Let ε > 0 be given. Then
¸
¸
¸
c
n
p
¸
¸
¸ ≤ ε if n
p
≥
c
ε
, i.e. if n ≥
_
c
ε
_
1/p
.
Thus we can take any integer N ≥
_
c
ε
_
1/p
, and it then follows that
¸
¸
¸
c
n
p
¸
¸
¸ ≤ ε if n ≥ N.
This implies the required limit exists and equals zero.
Remark 4.5. Have a look again at (4.1) and compare the four statements
n ≥ N implies a
n
−L ≤ ε,
n ≥ N implies a
n
−L < ε,
n > N implies a
n
−L ≤ ε,
n > N implies a
n
−L < ε.
1
We will often write “⇒” for “implies”.
2
This could easily be proved by induction, but it is not necessary to do so.
4.3. THE ARCHIMEDEAN AXIOM 6
These statements are certainly not equivalent. However, the following four statements are
equivalent!
for every positive number ε there exists an integer N such that
n ≥ N implies a
n
−a ≤ ε,
for every positive number ε there exists an integer N such that
n ≥ N implies a
n
−a < ε,
for every positive number ε there exists an integer N such that
n > N implies a
n
−a ≤ ε,
for every positive number ε there exists an integer N such that
n > N implies a
n
−a < ε.
Do you see why? I will discuss this in class. Which version does Adams use?
If the sequence (a
n
) does not converge, we say that it diverges. This may happen in
various ways. There are three examples in the ﬁrst paragraph on page 4.
There is a special case that is important. This is as in the following deﬁnition where
for each real number K, a
n
≥ K for all suﬃciently large n. Note that of all the examples
in the ﬁrst paragraph on page 4, this only occurs for the last of these examples.
Definition 4.6. The sequence (a
n
) diverges to +∞ if for each real number K there
is an integer N such that
a
n
≥ K for all n ≥ N.
We write a
n
→ ∞ or lima
n
= ∞.
Similarly, (a
n
) diverges to −∞ if for each real number K there is an integer N such
that
a
n
≤ K for all n ≤ N.
We write a
n
→ −∞ or lima
n
= −∞.
(In the case a
n
→ ∞, think of K as large and positive. In the case a
n
→ −∞, think
of K as large and negative.)
Note, by the way, that we never say “(a
n
) converges to ∞” or “(a
n
) converges to −∞”.
Example 4.7. Show from the deﬁnition that 2
n
→ ∞.
Proof. Let K be any (positive) real number.
We want to show that for all suﬃciently large n, 2
n
≥ K. But this inequality
is equivalent to nln 2 ≥ ln K, which is equivalent to n ≥ ln K/ ln 2. Hence if we let
N = 1 +
_
ln K
ln 2
¸
, this last inequality is true whenever n ≥ N. Hence
n ≥ N ⇒ 2
n
≥ K.
This means 2
n
→ ∞.
Note that if a
n
→ ∞ then 1/a
n
→ 0 (assuming a
n
= 0 in order that 1/a
n
is deﬁned).
This follows from the deﬁnitions, because 1/a
n
 ≤ ε is equivalent to a
n
 ≥ 1/ε. But the
latter is true for all suﬃciently large n since a
n
→ ∞, and so the former is also true for
all suﬃciently large n.
4.3. The Archimedean Axiom
If you look at the 1998 Calculus Notes, Section 2.4, you will see that there is another
axiom, the Completeness Axiom, as well as the Algebraic and Order axioms for the ral
numbers.
In these Notes, we take a diﬀerent approach, and instead of the Completeness Axiom
we introduce two axioms:
• The Archimedean Axiom,
• The Cauchy Completeness Axiom.
4.3. THE ARCHIMEDEAN AXIOM 7
The second will be discussed in a later section and the ﬁrst will be discussed here.
It can be shown that no more axioms are necessary. More precisely, it can be shown
that any two models of all the axioms are essentially the same (more precisely, they are
“isomorphic”). If you are adventurous, this is discussed in the last Chapter of the book
Calculus by Michael Spivak.
It can also be proved that these two axioms together are equivalent to the Complete
ness Axiom used in the 1998 Notes. More precisely, using just the algebraic and order
axioms, one can prove that
1. (Archimedean Axiom + Cauchy Completeness Axiom) ⇒ Completeness Axiom,
2. Completeness Axiom ⇒ (Archimedean Axiom + Cauchy Completeness Axiom).
If you are very adventurous, see the book Kripke, Introduction to Analysis, p38 Q’s 9 &10,
where the equivalence is essentially set as an (advanced) exercise.
Here then is the Archimedean Axiom:
Axiom (Archimedean Axiom). For every real number x there is a natural number
n such that x ≤ n.
Here is a simple consequence. If x ≤ n, then by properties of inequalities, x < n+1,
x < n + 2, etc. In particular, we can ﬁnd natural numbers strictly greater than x.
While the axiom may seem obvious, and indeed it is clearly true for the model of the
real numbers which we have in our mind, it does not actually follow from the algebraic
and order axioms, as I discuss below in a starred comment.
One diﬀerence between the Archimedean Axiom and the Algebraic and Order Axioms
is as follows. The Algebraic and Order Axioms are either of the form
1. for all real numbers, some elementary property (such as a+b = b+a, or a+0 = a,
or a < b ⇒ a + c < b + c) is true, where “elementary” means that the property
does not involve any further “quantiﬁers” (i.e. expressions of the form “for all” or
“there exists”); or are one of the
2. additive or multiplicative inverse axioms, in which case the real number which is
asserted to exist is in fact unique.
On the other hand, the natural number n asserted to exist by the Archimedean Axiom
for each real number x is certainly not unique. If x ≤ n is true, then n may be replaced
by any larger natural number. Moreover, the set N of natural numbers (which is needed
in the statement of the Archimedean Axiom) also involves a certain level of complexity in
its (precise) deﬁnition; see the last exercise in Chapter 2 of Spivak.
There is one immediate consequence of the Archimedean Axiom which is quite im
portant.
Corollary 4.8. For every real number ε > 0 there is a natural number N such
that 1/N ≤ ε.
Proof. Suppose ε > 0. From the Archimedean Axiom there is a natural number N
such that N ≥ 1/ε. From properties of inequalities it follows that 1/N ≤ ε.
Consider the ε and N as in the Corollary. If n ≥ N then 1/n ≤ 1/N (by properties
of inequalities). Since 1/N ≤ ε it follows that
n ≥ N ⇒ 1/n ≤ ε.
Since ε > 0 was an arbitrary positive number, this is the same as asserting 1/n → 0. Thus
we have used the Archimedean Axiom to prove 1/n → 0.
_
By essentially arguing in the reverse direction we can also prove the converse;
namely that the statement 1/n → 0 implies the Archimedean Axiom. Thus if we assume
just the algebraic and order axioms, the statement 1/n → 0 is in fact equivalent to the
Archimedean Axiom.
_
Fattened up copy of R





0
a
b
c
d

e
d < e < c < 0 < 1 < a < b
A "number'' on any line is less than any number to the right ,
and less than any any number on any higher line.
Between any two lines there is an infinite number of other
lines.

1
4.4. PROPERTIES OF LIMITS 8
So now I am forced to confess that I cheated in the previous section.
3
In each of Ex
amples 4.2–4.4, 4.7 there was a hidden application of the Archimedean Axiom somewhere
in the last three lines of the solution. Find the hidden application for Easter.
The fact that the Archimedean Axiom is necessary, in the sense that it does not
follow from the previous axioms, is a consequence of the fact that there are models of
the algebraic and order axioms which do not satisfy the Archimedean Axiom. They are
sometimes called the Hyperreals!
Part of any such model looks like a “fattened up” copy of R, in the sense that it
contains a copy of R together with “inﬁnitesimals” squeezed between each real a and all
reals greater than a. This part is followed and preceded by inﬁnitely many “copies” of
itself, and between any two copies there are inﬁnitely many other copies. See the following
crude diagram.
Remark 4.9. The main point to remember from this section is that in speciﬁc ex
amples where we have to show that for any real number of a given form (such as
1
√
ε
in
Example 4.2) there is always an integer N at least as large (N will depend of course on
the particular real number
1
√
ε
), the Archimedean Axiom is probably needed. You should be
aware of this, but after a few more examples we will adopt our previous cavalier attitude
and not explicitly note when the axiom is needed.
4.4. Properties of limits
We prove that limits of sequences behave as we expect under addition, sub
traction, multiplication and division, and we prove the Squeeze Theorem.
It is not usually not very eﬃcient to use the deﬁnition of a limit in order to prove that
a sequence converges. Instead, we prove a number of theorems which will make things
much easier.
The ﬁrst theorem shows that if two sequences converge, then so does their sum,
and moreover the limit of the new sequence is just the sum of the limits of the original
sequences. The theorem may be written more brieﬂy as:
if a
n
→ a and b
n
→ b then a
n
+ b
n
→ a + b .
Note that the theorem has two claims; ﬁrst that (a
n
+b
n
) is convergent, and second that
the actual limit is a + b.
The result is not very surprising, since if a
n
is getting close to a and b
n
is getting
close to b then we expect that a
n
+ b
n
is getting close to a + b. So you may consider
3
I promise not to do it again, and I only cheated in the examples, not in deﬁnitions or
theorems.
4.4. PROPERTIES OF LIMITS 9
the theorem as a partial justiﬁcation that Deﬁnition 4.1 does indeed capture the informal
notion of a limit.
Notice in the (rather subtle and elegant) proof how the deﬁnition of a limit is used
three times; once to get information from the fact a
n
→ a, once to get information from
the fact b
n
→ b, and ﬁnally to deduce that a
n
+ b
n
→ a + b.
By the way, why do we use ε/2 in (4.3) and (4.4), and why is this justiﬁable by
Deﬁnition 4.1?
Theorem 4.10. Suppose (a
n
) and (b
n
) are convergent sequences with limits a and b
respectively. Then (a
n
+ b
n
) is a convergent sequence, and its limit is a + b.
Proof. Let ε > 0 be given.
Since a
n
→ a there exists an integer N
1
(by Deﬁnition 4.1) such that
n ≥ N
1
implies a
n
−a ≤ ε/2. (4.3)
Since b
n
→ b there exists an integer N
2
(again by Deﬁnition 4.1) such that
n ≥ N
1
implies b
n
−b ≤ ε/2. (4.4)
It follows that if n ≥ max{N
1
, N
2
} then
(a
n
+ b
n
) −(a + b) = (a
n
−a) + (b
n
−b)
≤ a
n
−a +b
n
−b by the triangle inequality
≤
ε
2
+
ε
2
by (4.3) and (4.4)
= ε.
It follows from Deﬁnition 4.1, with N = max{N
1
, N
2
}, that (a
n
+ b
n
) converges and
the limit is a + b.
The next easy result is useful in a number of situations. It is certainly not true in
general, consider for example the sequence
1, 2, 3, 4, . . . .
Theorem 4.11. Suppose a
n
→ a. Then the sequence is bounded; i.e. there is a real
number M such that a
n
 ≤ M for all n.
Proof. From the deﬁnition of convergence, taking ε = 1, there is an integer N such
that
a −1 ≤ a
n
≤ a + 1 for all n ≥ N. (4.5)
Fix this N. Since the set of terms
a
1
, a
2
, . . . , a
N−1
is ﬁnite, it follows that there exist real numbers M
1
and M
2
such that
M
1
≤ a
n
≤ M
2
for all n < N. (4.6)
(Just take M
1
= min{a
1
, a
2
, . . . , a
N−1
} and M
2
= max{a
1
, a
2
, . . . , a
N−1
}.)
From (4.5) and (4.6),
M
∗
1
≤ a
n
≤ M
∗
2
for all n,
where M
∗
1
= min{a −1, M
1
}, M
∗
2
= max{a + 1, M
2
}.
The required result follows by taking M = max{M
∗
1
, M
∗
2
}.
The standard properties about products, quotients etc. of convergent sequences can
all be similarly established. We state them together in the following theorem (which
includes Theorem 4.10). The proofs are material and are at the end of this section.
But you should try to understand them, and I will discuss them in class.
[ ]  
a b
] [
ε
ε ε
ε
ε = ab/3
4.4. PROPERTIES OF LIMITS 10
Theorem 4.12. Suppose
lima
n
= a, limb
n
= b,
and c is a real number. Then the following limits exist and have the given values.
lima
n
±b
n
= a ±b,
limca
n
= ca,
lima
n
b
n
= ab,
lim
a
n
b
n
=
a
b
, assuming b = 0 and b
n
= 0 for each n.
Example 4.13. Let a
n
=
_
1 +
1
√
n
_
2
−(1 + 2
−n
).
We can prove directly from the deﬁnition of convergence that
1
√
n
→ 0 and 2
−n
→ 0.
It then follows from the previous theorem that 1+
1
√
n
→ 1 (since we can think of 1+
1
√
n
as
obtained by adding the term 1 from the constant sequence (1) to the term
1
√
n
). Applying
the theorem again,
_
1 +
1
√
n
_
2
→ 1. Similarly, 1 + 2
−n
→ 1.
Hence (again from the theorem) a
n
→ 0.
Example 4.14. Let a
n
=
2n
2
−1
3n
2
−7n+1
.
Write
2n
2
−1
3n
2
−7n + 1
=
2 −
1
n
2
3 −
7
n
+
1
n
2
.
Since the numerator and denominator converge to 2 and 3 respectively, it follows a
n
→ 0.
The next theorem says that a sequence cannot have two distinct limits. It is not
surprising of course, but note how it does follow from the actual deﬁnition of a limit.
Theorem 4.15. Suppose a
n
→ a and a
n
→ b. Then a = b.
Proof. Assume (in order to obtain a contradiction) that a = b. Take ε = a−b/3 in
the deﬁnition of a limit, Deﬁnition 4.1. (For motivation, look at the following diagram).
Since a
n
→ a, it follows that
a
n
∈ [a −ε, a + ε] (4.7)
for all suﬃciently large n, say for n ≥ N
1
.
Since a
n
→ b, it follows that
a
n
∈ [b −ε, b + ε] (4.8)
for all suﬃciently large n, say for n ≥ N
2
.
But this implies
a
n
∈ [a −ε, a + ε] and a
n
∈ [b −ε, b + ε]
for all n ≥ max{N
1
, N
2
}, which is impossible as ε = a −b/3.
Thus the assumption a = b led to a contradiction and so a = b.
The following theorem says that if a sequence is “squeezed” between two sequences
which both converge to the same limit, then the original sequence also converges, and it
converges to the same limit.
Theorem 4.16. Suppose a
n
≤ b
n
≤ c
n
for all n (or at least for all n ≥ N for some
N). Suppose a
n
→ L and c
n
→ L as n → ∞. Then b
n
→ L as n → ∞.

[ ]
Lε L
L+ε
  
a
n
b
n
c
n
4.4. PROPERTIES OF LIMITS 11
Proof. Let ε > 0 be given. (For motivation, look at the following diagram).
Since a
n
→ L there is some integer N
1
such that
n ≥ N
1
⇒ a
n
∈ [L −ε, L + ε]. (4.9)
Since c
n
→ L there is some integer N
2
such that
n ≥ N
2
⇒ c
n
∈ [L −ε, L + ε]. (4.10)
Let N be the larger of N
1
and N
2
, i.e. let N = max{N
1
, N
2
}. Then since a
n
≤ b
n
≤ c
n
it follows from (4.9) and (4.10) that
n ≥ N ⇒ b
n
∈ [L −ε, L + ε].
But ε was an arbitrary positive number, and so it follows that b
n
→ L.
Example 4.17. Consider the sequence 3 + (sin cos n)/n.
Since −1 ≤ sin x ≤ 1, it follows that 3 − 1/n ≤ 3 + (sin cos n)/n ≤ 3 + 1/n. But
3 −1/n → 3 and 3 + 1/n → 3. Hence 3 + (sin cos n)/n → 3.
We ﬁnish this section with the promised proofs of the algebraic properties of limits.
Try to understand the ideas, although the material is (well, sort of
1
2
).
Proof of Theorem 4.12.
• We ﬁrst establish the result for ca
n
.
Let ε > 0 be any positive number. We want to show
ca
n
−ca ≤ ε
for all suﬃciently large n.
Since a
n
→ a there exists an integer N such that
a
n
−a ≤ ε/c for all n ≥ N.
(This assumes c = 0. But if c = 0, then the sequence (ca
n
) is the sequence all of whose
terms are 0, and this sequence certainly converges to ca = 0.) Multiplying both sides of
the inequality by c we see
ca
n
−ca ≤ ε for all n ≥ N,
and so ca
n
→ ca by the deﬁnition of convergence.
• The result for a
n
−b
n
now follows easily. Just note that
a
n
−b
n
= a
n
+ (−1)b
n
.
But (−1)b
n
→ (−1)b by the previous result with c = −1, and so the result now follows
from Theorem 4.10 about the sum of two sequences.
• The result for a
n
b
n
uses Theorem 4.11 in the proof.
As usual, let ε > 0 be any positive number. We want to show that
a
n
b
n
−ab ≤ ab
for all n ≥ some N.
To see how to choose N, write
a
n
b
n
−ab = a
n
b
n
−a
n
b + a
n
b −ab
= a
n
(b
n
−b) + b(a
n
−a)
≤ a
n
(b
n
−b) +b(a
n
−a)
= a
n
 b
n
−b +b a
n
−a.
(4.11)
(This trick of adding and subtracting the same term, here it is a
n
b, is often very useful.)
We will show that both terms are ≤ ε/2 for all suﬃciently large n.
4.4. PROPERTIES OF LIMITS 12
For the second term b a
n
−a, the result is certainly true if b = 0, since the term is
then 0. If b = 0, since a
n
→ a, we can choose N
1
such that
a
n
−a ≤
ε
2b
for all n ≥ N
1
,
and so
b a
n
−a ≤
ε
2
for all n ≥ N
1
. (4.12)
For the ﬁrst term a
n
 b
n
− b, we use Theorem 4.11 to deduce for some M that
a
n
 ≤ M for all n. By increasing M if necessary take M = 0. By the same argument as
for the second term, we can choose N
2
such that
M b
n
−b ≤
ε
2
for all n ≥ N
2
,
and so
a
n
 b
n
−b ≤
ε
2
for all n ≥ N
2
. (4.13)
Putting (4.11), (4.12) and (4.13) together, it follows that if n ≥ N, where N =
max{N
1
, N
2
}, then
a
n
b
n
−ab ≤
ε
2
+
ε
2
= ε.
Since ε > 0 was arbitrary, this proves a
n
b
n
→ ab.
• We can prove that a
n
/b
n
→ a/b by ﬁrst showing 1/b
n
→ 1/b and then using the
previous result about products of sequences.
We ﬁrst prove the Claim: there is some number K > 0 such that b
n
 > K for all n.
The proof is similar to that in Theorem (4.11), and goes as follows:
Since b = 0 we can choose ε = b/2 (> 0) in the deﬁnition of convergence and deduce
that for some integer N
1
,
n ≥ N
1
⇒ b
n
−b ≤ b/2,
and so in particular
n ≥ N
1
⇒ b
n
 ≥ b/2.
Next let
c = min{b
1
, . . . , b
N−1
}.
Then
n < N
1
⇒ b
n
 ≥ c,
where c > 0 since b
1
, . . . , b
N−1
 > 0.
4
Putting the previous results together, and letting K = min{c, b/2} > 0, we see
b
n
 ≥ K (4.14)
for all n. This establishes the Claim.
We now proceed with the proof that 1/b
n
→ 1/b. For this let ε > 0 be any positive
number.
In order to see how to choose N in the deﬁnition of convergence, we compute
¸
¸
¸
1
b
n
−
1
b
¸
¸
¸ =
b −b
n

b
n
 b
≤
b −b
n

Kb
, (4.15)
where for the inequality we have replaced b
n
 by the smaller (but still positive) number
K in (4.14).
Since b
n
→ b we can ﬁnd an integer N such that for all n ≥ N,
b −b
n
 ≤ Kb ε.
4
This is an important point. If we have an inﬁnite set of numbers, such as
{1, 1/2, 1/3, . . . , 1/n, . . . }, all of which are > 0, then there may not be a minimum member of
the set. In fact for this example, if c ≤ 1/n for all n, we have to take c = 0 (or c < 0), not c > 0.
4.5. CAUCHY SEQUENCES 13
It follows from (4.15) that if n ≥ N then
¸
¸
¸
1
b
n
−
1
b
¸
¸
¸ ≤ ε.
Since ε > 0 was arbitrary, it follows that 1/b
n
→ 1/b.
Since a
n
→ a, it now follows from the result for products that a
n
/b
n
→ a/b.
4.5. Cauchy sequences
A sequence converges iﬀ it is Cauchy. This gives a criterion for convergence
which does not require knowledge of the limit. We deduce that a bounded
monotone sequence is convergent.
Is there a method (or criterion) for telling if a sequence (a
n
) converges, if we do not
know the actual limit? Deﬁnition 4.1 does not help, as it involves the limit a. But there
is indeed such a criterion, and it is due to Cauchy,
Loosely speaking, the Cauchy criterion says that if we go far enough out in the
sequence then we can make the members of the sequence as close to each other as we like.
The previous sentence is vague and ambiguous, and you should never ever say or write
anything like that in a mathematical argument. What I mean is that for each ε > 0 there
is an N such that any two members of the sequence from the Nth onwards are within ε
of each another. More precisely:
Definition 4.18. A sequence (a
n
) is a Cauchy sequence if for each number ε > 0
there exists an integer N such that
a
m
−a
n
 ≤ ε whenever m ≥ N and n ≥ N.
If the condition in the above deﬁnition is true, that is if the sequence (a
n
) is Cauchy,
we sometimes write:
a
m
−a
n
 → 0 as m, n → ∞.
Example 4.19. Show that the sequence a
n
=
2n−1
n+3
is Cauchy.
Solution. Let ε > 0 be given.
In order to ﬁnd N as in the deﬁnition, we compute
a
m
−a
n
 =
¸
¸
¸
2m−1
m + 3
−
2n −1
n + 3
¸
¸
¸
=
¸
¸
¸
¸
7(m−n)
(m + 3)(n + 3)
¸
¸
¸
¸
≤
¸
¸
¸
¸
7m
(m + 3)(n + 3)
¸
¸
¸
¸
+
¸
¸
¸
¸
7n
(m + 3)(n + 3)
¸
¸
¸
¸
≤
7
n + 3
+
7
m + 3
since
m
m + 3
≤ 1 and
n
n + 3
≤ 1
≤
7
n
+
7
m
.
Next we note that
7
n
≤
ε
2
provided n ≥
14
ε
, i.e. provided n ≥ N where N =
_
14
ε
¸
+1.
Similarly,
7
m
≤
ε
2
provided m ≥ N.
It follows that
a
m
−a
n
 ≤ ε whenever m, n ≥ N.
Hence the sequence is Cauchy.
It is important to realise that the deﬁnition requires more than just that consecutive
members of the sequence from the Nth onwards are within ε of each other.
For example, it is not too hard to see from the “graph” of the function
√
x that the
following statement is true:
for each number ε > 0 there exists N such that
n ≥ N implies 
√
n + 1 −
√
n ≤ ε.
4.5. CAUCHY SEQUENCES 14
(The main point is that the slope of the graph is 1/
_
2
√
x
_
, and so is very small when x
is very large.) The algebraic proof is in footnote
5
.
On the other hand, it is clear that the next statement is false:
for each number ε > 0 there exists N such that
m, n ≥ N implies 
√
m−
√
n ≤ ε.
The algebraic proof is in footnote
6
.
The major fact in this section is:
a sequence converges iﬀ it is Cauchy. (4.16)
One direction of (4.16), the fact that a convergent sequence is Cauchy, is easy to
prove.
Theorem 4.20. If a sequence converges then it is Cauchy.
Proof. Suppose a
n
→ a. Let ε > 0 be given.
In order to ﬁnd N as in the deﬁnition of a Cauchy sequence, we write
a
m
−a
n
 = (a
m
−a) + (a −a
n
) ≤ a
m
−a +a
n
−a. (4.17)
Since a
n
→ a there exists an integer N such that a
n
− a ≤ ε/2 whenever n ≥ N (and
what is the same thing, a
m
−a ≤ ε/2 whenever m ≥ N). Hence
a
m
−a
n
 ≤
ε
2
+
ε
2
= ε whenever m, n ≥ N.
This proves that (a
n
) is Cauchy.
It turns out that the other direction of (4.16), that every Cauchy sequence is conver
gent, does not follow from the axioms so far, and so must be assumed as an extra (and
ﬁnal!) axiom.
Axiom (Cauchy Completeness Axiom). If a sequence is Cauchy then it is con
vergent.
The Cauchy Completeness Axiom says, informally, that any sequence which is “trying
to converge”, really does have something to converge to, and so is indeed convergent in
the sense of Deﬁnition 4.1. That is, there are no “gaps” in the real numbers.
The analogous statement is not true for the rational numbers. That is, if a sequence
of rational numbers is Cauchy, then it is not necessarily true that the sequence will converge
to a rational number. (Of course, it will converge to a real number — this is just what
the axiom says — but the limit may be irrational.)
5
To show that
¸
¸
√
n + 1 −
√
n
¸
¸
≤ ε for all suﬃciently large n, we compute
√
n + 1 −
√
n =
√
n + 1 +
√
n
√
n + 1 +
√
n
_
√
n + 1 −
√
n
_
=
(n + 1) −n
√
n + 1 +
√
n
=
1
√
n + 1 +
√
n
≤
1
√
n
.
It follows that
√
n + 1 −
√
n ≤ ε if
1
√
n
≤ ε, i.e. if n ≥
1
ε
2
.
6
It is very important to realise that the statement is false means there is some (“bad”) ε > 0
for which there is no N such that
m, n ≥ N implies 
√
m−
√
n ≤ ε.
Thus we have to ﬁnd just one “bad” ε.
But ε = 1 is bad since there is certainly no N such that
m, n ≥ N implies 
√
m−
√
n ≤ 1.
Thus we have shown the required statement is false! (In fact, in this case any ε > 0 will be “bad”.)
.. . . . . . . .
I
1
I
2
I
3
I
4
I
5
.
a
1
a
3
a
5
a
6
a
8
.
M
4.5. CAUCHY SEQUENCES 15
For example, take any sequence (a
n
) of rational numbers which converges to
√
2 (we
saw in Section 2.5 that
√
2 cannot be rational). One example of such a sequence is obtained
from the decimal expansion of
√
2 = 1.414213562 . . . ; thus we can take the sequence
1, 1.4, 1.41, 1.414, 1.4142, 1.41421, . . . . (4.18)
This sequence of rational numbers converges, and so is Cauchy by Theorem 4.20. But it
does not converge to a rational number. Thus the analogue of the Cauchy Completeness
Axiom is not true in the rational numbers.
In particular, we now see that the Cauchy Completeness Axiom cannot follow from
the other algebraic, order and Archimedean axioms. The analogues of all these other
axioms hold for the rational numbers. But the analogue of the Cauchy Completeness
Axiom does not hold for the rational numbers.
We have now come to the end of the list of axioms. They are the algebraic, order,
Archimedean and Cauchy Completeness axioms. (To compare the approach here with that
in the 1998 Calculus Notes, go back and reread the ﬁrst four paragraphs in Section 4.3.)
Throughout the rest of the course, we will rarely indicate when we are using the
algebraic, order or Archimedean axioms, but we will usually remark if we are using the
Cauchy Completeness axiom.
We next show that if a sequence is bounded and increasing (or decreasing) then it is
convergent. We do this by proving that such a sequence is Cauchy, and then we use the
Cauchy Completeness Axiom to show that the sequence is convergent.
First we give a deﬁnition.
Definition 4.21. A sequence (a
n
)
n≥1
is monotone increasing if a
n
≤ a
n+1
for all
n ≥ 1. It is is monotone decreasing if a
n
≥ a
n+1
for all n ≥ 1.
Thus the sequences 1, 1, 2, 2, 3, 3, . . . and 1/2, 2/3, 3/4, 4/5, . . . are both monotone
increasing.
Theorem 4.22. If a sequence is monotone increasing (or decreasing) and bounded,
then it is convergent.
Proof. Suppose the sequence (a
n
) is monotone increasing and bounded. (A similar
proof will apply if it is decreasing.) In particular, suppose a
n
≤ M for all n.
To motivate the argument look at the following diagram:
Let I be the interval [a
1
, M].
Divide I into two closed intervals of equal length. Either all members of the sequence
are in the left interval, or at least one member of the sequence is in the right interval (and
in the second case, since the sequence is monotone increasing, all later members of the
sequence are also in the right interval). In either case, all members of the sequence after
some term are in one of the two subintervals. Choose this subinterval and denote it by
I
1
. Thus
a
n
∈ I
1
for all n ≥ N
1
(say).
(In the diagram, N
1
= 1 and I
1
is the left interval; this implies there are no members of
the sequence beyond the right endpoint of I
1
. There is not enough information just in the
diagram to tell us this fact. )
Similarly divide I
1
into two closed intervals of equal length and choose I
2
to be that
subinterval which eventually contains all members of the sequence, i.e.
a
n
∈ I
2
for all n ≥ N
2
(say).
(In the diagram, N
2
= 4 and I
2
is the right interval.)
4.6. SUBSEQUENCES AND THE BOLZANOWEIERSTRASS THEOREM 16
Similarly divide I
2
into two closed intervals of equal length and choose I
3
to be that
subinterval which eventually contains all members of the sequence, i.e.
a
n
∈ I
3
for all n ≥ N
3
(say).
(In the diagram, N
3
= 7 and I
3
is the right interval.)
etc.
In order to show that (a
n
) is Cauchy, let ε > 0 be given.
Since the length
7
of I
k
is (M − a)/2
k
, it follows that there is an integer k such that
the length of I
k
is ≤ ε
8
. But
a
m
, a
n
∈ I
k
if m, n ≥ N
k
,
and so
a
m
−a
n
 ≤ ε if m, n ≥ N
k
.
It follows that the sequence (a
n
) is Cauchy (take N in Deﬁnition 4.18 to equal N
k
).
Hence (a
n
) converges by the Cauchy Completeness Axiom.
Example 4.23. Show the sequence
n
√
n is monotone decreasing and bounded below,
and hence is convergent.
Solution. As remarked before, in order to have interesting examples we will occa
sionally use material that has not been rigorously established. We do that here.
So for x > 0, let
9
f(x) = x
1/x
= e
1
x
ln x
.
Then f(n) = n
1/n
(i.e.
n
√
n). Moreover,
f
(x) =
_
−
1
x
2
ln x +
1
x
2
_
e
1
x
ln x
,
which is negative for x > e.
It follows that the sequence
n
√
n is decreasing if n ≥ 3.
It is clear that the sequence is bounded below by 1, i.e. 1 ≤
n
√
n for all n.
By the previous theorem, the sequence must converge.
In one of the assignment problems we will see that
n
√
n → 1.
The analogue of Theorem 4.22 is not true for the rational numbers. In other words,
a bounded increasing (or decreasing) sequence of rational numbers need not converge to
a rational number. (Of course it will converge to a real number — this is just what the
theorem says.)
The sequence in (4.18) is an example of a bounded increasing sequence of rational
numbers which does not converge to a rational number; instead it converges to the irra
tional number
√
2.
4.6. Subsequences and the BolzanoWeierstrass Theorem
A sequence need not of course converge, even if it is bounded. But by
the BolzanoWeierstrass theorem, every bounded sequence has a convergent
subsequence.
Suppose (a
n
) is a sequence of real numbers. A subsequence is just a sequence obtained
by skipping terms. For example, the following are subsequences:
a
2
, a
4
, a
6
, . . . ,
a
1
, a
27
, a
31
, a
44
, a
101
, . . . .
We usually write a subsequence of (a
n
) as
a
n
1
, a
n
2
, a
n
3
, a
n
4
, . . . ,
7
The length of the interval [a, b] is deﬁned to be b −a.
8
This actually requires the Archimedean Axiom. We have (M − a)/2
k−1
≤ (M − a)/k.
Now use the Archimedean Axiom to choose k so that k ≥ (M −a)/ε.
9
Recall that if x > 0 then x
a
= e
ln x
a
= e
a ln x
.
. . . . . . . .
I
1
I
2
I
3
I
4
I
5
. .
1 1 a
5
a
3
a
7
a
1
a
4
a
6
a
8
a
2
4.6. SUBSEQUENCES AND THE BOLZANOWEIERSTRASS THEOREM 17
or just as (a
n
k
). Thus in the above two examples, we have
n
1
= 2, n
2
= 4, n
3
= 6, . . . ,
n
1
= 1, n
2
= 27, n
3
= 31, n
4
= 44, n
5
= 101, . . . ,
respectively.
A bounded sequence certainly need not be convergent. For example, the sequences
1, −1, 1, −1, 1, −1, 1, −1, . . . ,
1 +
1
2
, −1 +
1
3
, 1 +
1
4
, −1 +
1
5
, 1 +
1
6
, −1 +
1
7
, . . . , (4.19)
do not converge.
But there are subsequences of each sequence which converge to 1. For example,
1, 1, 1, 1, 1, 1, . . . ,
1, −1, 1, −1, 1, −1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, . . . ,
are two diﬀerent subsequences of the ﬁrst sequence which converge to 1. In fact there are
inﬁnitely many diﬀerent subsequences converging to 1.
Similarly, there are subsequences converging to −1. There are also subsequences of
the second sequence converging to 1 and subsequences converging to −1.
The following theorem turns out to have many important applications.
Theorem 4.24 (BolzanoWeierstrass Theorem). If a sequence is bounded then it has
a convergent subsequence.
Proof. Let (a
n
) be bounded. We will construct a convergent subsequence
a
n
1
, a
n
2
, a
n
3
, a
n
4
, a
n
5
, a
n
6
, . . . .
Since the sequence (a
n
) is bounded, there is a closed bounded interval I
1
which
contains all terms from the sequence. Choose one such term and denote it by a
n
1
(n
1
= 1
will do).
(The following diagram, not to scale, corresponds to the sequence (4.19).)
Divide I
1
into two closed intervals of equal length (and having only the midpoint of
I
1
in common). At least one of these intervals contains an inﬁnite number of diﬀerent
terms
10
from (a
n
); call this interval I
2
. Choose one such term from (a
n
) and denote it by
a
n
2
, but with the condition n
1
< n
2
. (In the diagram, we could take n
2
= 2.)
Divide I
2
into two closed intervals of equal length (and having only the midpoint of
I
2
in common). At least one of these intervals contains an inﬁnite number of diﬀerent
terms from (a
n
); call this interval I
3
. Choose one such term from (a
n
) and denote it by
a
n
3
, but with the condition n
2
< n
3
. (In the diagram, we could take n
3
= 4.)
Divide I
3
into two closed intervals of equal length (and having only the midpoint of
I
3
in common). At least one of these intervals contains an inﬁnite number of terms from
(a
n
); call this interval I
4
. Choose one such term from (a
n
) and denote it by a
n
4
, but with
the condition n
3
< n
4
. (In the diagram, we could take n
4
= 8.)
etc.
The sequence
a
n
1
, a
n
2
, a
n
3
, a
n
4
, . . . ,
is a subsequence of (a
n
).
Moreover, it is Cauchy.
10
For example, a
3
and a
4
are always considered to be diﬀerent terms, even if they have the
same value.
  
a b c
4.6. SUBSEQUENCES AND THE BOLZANOWEIERSTRASS THEOREM 18
To see this, ﬁrst note that for each N,
a
n
N
, a
n
N+1
, a
n
N+2
, a
n
N+3
, · · · ∈ I
N
.
Now let ε > 0 be a given positive number. The length of I
N
is I
N
 =
I
1

2
N−1
. Choose
N so
I
1

2
N−1
≤ ε
11
. It follows that
a
n
p
−a
n
q
 ≤ ε whenever p, q ≥ N.
This means the sequence (a
n
k
) is Cauchy. Hence it converges. This proves the theorem.
Remark 4.25. If all members of the original sequence (a
n
) belong to a closed bounded
interval [a, b], then so does the limit of any subsequence.
Proof. To simplify notation, denote the subsequence by (x
k
). Suppose in order to
obtain a contradiction that x
k
→ c and x
k
≤ b for all k, but c > b (see the next diagram).
Let ε = (c − b)/2, or any number less than the distance from b to c. Eventually all
members of the sequence are within ε of c, and so in particular are > b. This contradicts
the fact that x
k
≤ b for all k.
Similarly if x
k
≥ a for all k and x
k
→ c, then c ≥ a.
Example 4.26. We have seen examples of bounded sequences which have sub
sequences converging to 1 and subsequences converging to −1. Given m distinct real
numbers x
1
, . . . , x
m
it is easy to see that there is a sequence which has a subsequence
converging to x
1
, another converging to x
2
, . . . , and another converging to x
m
.
What is more surprising is that there is a sequence, which for each real number
x ∈ [0, 1], has a subsequence converging to x.
The sequence can be constructed as follows. First enumerate all decimal expansions
of length one, then all of length two, then all of length three, then all of length four, etc.
Thus the sequences is
·0, ·1, ·2, ·3, . . . , ·9, ·00, ·01, ·02, ·03, . . . , ·99, ·000, ·001, ·002, ·003,
. . . , ·999, ·0000, ·0001, ·0002, ·0003, . . . , ·9999, ·0000, ·0001, . . . .
Note that various values are repeated, e.g. ·2, ·20, ·, 200, . . . etc., but this does not matter.
By using decimal expansions, we see that for any number x there is a subsequence
converging to x. For example, the subsequence converging to
√
2/2 = ·7071067810 . . . is
·7, ·70, ·707, ·7071, ·70710, ·707106, . . . .
It is possible to change this construction a little and obtain a sequence which for every
real number x has a subsequence converging to x. (The sequence cannot be bounded, since
it must contain arbitrarily large real numbers.)
11
This needs the Archimedean Axiom. It is equivalent to choosing N such that 2
N−1
≥
I
1
/ε. But since 2
N−1
≥ N, for example by induction, it is suﬃcient to use the Archimedean
Axiom to ﬁnd a natural number N ≥ I
1
/ε.
°
graph of f
1
CHAPTER 5
Continuous Functions
5.1. Deﬁnition and examples
The notion of continuity of a function is deﬁned in terms of sequences.
Some examples of continuous and discontinuous functions are given. (We
also show that between any two numbers there are an inﬁnite number of
rationals and an inﬁnite number of irrationals.)
Recall that the domain of a function f, denoted by D(f), is the set of numbers x such
that f(x) is deﬁned. We will usually be interested in functions whoxe domains are intervals
[a, b], (a, b), (a, ∞)
1
, etc. But it is possible for the domain to be a more complicated set
of real numbers.
We will deﬁne the notion of continuity in terms of convergence of sequences. The
informal idea of continuity of a function f at a point c is that “as x appraoches c then
f(x) approaches f(c)”.
More accurately, we have the following natural deﬁnition.
Definition 5.1. A function f is continuous at a point c ∈ D(f) if for every sequence
(x
n
) such that x
n
∈ D(f) and x
n
→ c, we have f(x
n
) → f(c).
We say f is continuous (on its domain) if f is continuous at every point in its domain.
In other words,
x
n
∈ D(f) and x
n
→ c ⇒ f(x
n
) → f(c).
We will often not write x
n
∈ D(f), although this is always understood in order that f(x
n
)
be deﬁned.
Thus in order to show f is continuous at c, we have to show that for every sequence
x
n
→ c one has f(x
n
) → f(c).
In order to show f is not continuous at c, we only have to show there is one (“bad”)
sequence x
n
→ c with f(x
n
) → f(c).
2
Example 5.2. Suppose
f(x) =
_
x 0 ≤ x < 1
1
2
x
2
1 ≤ x ≤
3
2
The domain of f is [0,
3
2
]. The following is an attempt to sketch the graph of f.
1
I emphasise that ∞ is not a numbr, and that for us the symbol ∞ has no neaning by itself.
The interval (a, ∞) is just the set of real numbers strictly greater than a.
2
If there is one, there will in fact be many such “bad” sequences — we can always change
the ﬁrst million or so terms — but the point is that to show continuity fails it is suﬃcient to just
prove there is one “bad” sequence.
19
a b 0
> 1/n
1/n 2/n 3/n m/n
5.1. DEFINITION AND EXAMPLES 20
It is clear that f is not continuous at 1. For example, take the sequence x
n
= 1 −
1
n
.
Then x
n
→ 1 but f(x
n
) (= 1 −
1
n
) → f(1) since f(1) =
1
2
.
On the other hand, if c = 1 and c ∈ D(f) then
x
n
→ c ⇒ f(x
n
) → f(c).
To see this, ﬁrst suppose x
n
→ c and 1 < c ≤
3
2
. Then x
n
≥ 1 for all suﬃciently
large n, and so f(x
n
) =
1
2
x
2
n
for all suﬃciently large n. From properties of sequences if
x
n
→ c then x
2
n
→ c
2
and so
1
2
x
2
n
→
1
2
c
2
. But f(x
n
) =
1
2
x
2
n
for all suﬃciently large n,
and so limf(x
n
) = lim
1
2
x
2
n
=
1
2
c
2
.
In particular, f is not continuous on its domain.
The case 0 ≤ c < 1 is similar, and easier.
If we vary this example a little, and deﬁne
g(x) =
_
x 0 ≤ x < 1
1
2
x
2
1 < x ≤
3
2
,
then the domain of g is [0, 1) ∪ (1,
3
2
]. The function g is continuous at each c ∈ D(g), and
so is continuous on its domain.
Example 5.3. The absolute value function f (given by f(x) = x) is continuous.
We ﬁrst show continuity at 0. For this, suppose x
n
→ 0. Then x
n
 → 0 (this is
immediate from the deﬁnition of convergence, since x
n
− 0 ≤ ε iﬀ  x
n
 − 0  ≤ ε), i.e.
f(x
n
) → f(0).
To prove continuity at c = 0 is similar to the previous example.
The following result is established directly from the properties of convergent se
quences.
Proposition 5.4. Every polynomial function is continuous.
Proof. Let
f(x) = a
0
+ a
1
x + a
2
x
2
+· · · + a
k
x
k
.
To show that f is continuous at some point c, suppose x
n
→ c.
Then x
2
n
→ c
2
, x
3
n
→ c
3
, etc., by the theorem about products of convergent sequences.
It follows that a
1
x
n
→ a
1
c, a
2
x
2
n
→ a
2
c
2
, a
3
x
3
n
→ a
3
c
3
, etc., by the theorem about
multiplying a convergent sequence by a constant. Finally,
a
0
+ a
1
x + a
2
x
2
+· · · + a
k
x
k
→ a
0
+ a
1
c + a
2
c
2
+· · · + a
k
c
k
by repeated applications of the theorem about sums of convergent sequences (a
0
is here
regarded as a constant sequence).
Example 5.5. Here is a surprising example.
Let
f(x) =
_
x x rational
x −x irrational.
The following diagram is misleading, since between any two real numbers there is both a
rational and an irrational number (in fact an inﬁnite number of each).
3
3
To see this, ﬁrst suppose 0 < a < b. Choose an integer n such that
1
n
< b − a (from the
Archimedean axiom!). Then at least one member
m
n
of the sequence
1
n
,
2
n
,
3
n
,
4
n
,
5
n
, . . .
will lie between a and b. This is very plausible from the diagram. To show it rigorously requires
showing there is a largest k such that
k
n
< a, which in turn requires a slightly more careful
deﬁnition of the natural numbers. Then take m = k + 1.
Since we can then obtain another rational between
m
n
and b, etc., etc., there is in fact an inﬁnite
number of rationals between a and b.
5.2. PROPERTIES OF CONTINUOUS FUNCTIONS 21
The function f is continuous at 0. To see this, suppose x
n
→ 0. Then x
n
 → 0 (this
follows from the deﬁnition of a limit). Since −x
n
 ≤ f(x
n
) ≤ x
n
, it follows from the
Squeeze theorem that f(x
n
) → 0, i.e. f(x
n
) → f(0).
On the other hand, f is not continuous at c if c = 0. For example if c is irrational
then we can choose a sequence of rationals x
n
such that x
n
→ c (by repeated applications
of the remark above in italics). It follows that f(x
n
) = x
n
→ c = f(c). Similarly if c is
irrational.
We will later deﬁne the exponential, logarithm, and trigonometric functions, and show
they are continuous. Meanwhile, we will use them in examples (but not in the development
of the theory).
5.2. Properties of continuous functions
The basic properties of continuous functions follow easily from the analo
gous properties of sequences.
Theorem 5.6. Let f and g be continuous functions and let D = D(f) ∩D(g).
4
Then
1. f + g is continuous on D,
2. fg is continuous on D,
3. αf is continuous on D(f) (α any real number)
4. f/g is continuous at any point c ∈ D such that g(c) = 0.
Proof. Suppose c ∈ D. Let (x
n
) be any sequence such that x
n
→ c (and as usual,
x
n
∈ D).
Then f(x
n
) → f(c) and g(x
n
) → g(c), since f and g are continuous at c. It follows
f(x
n
) + g(x
n
) → f(c) + g(c)
by Theorem 4.12 about sums of convergent sequences. That is,
(f + g)(x
n
) → (f + g)(c).
It follows that f + g is continuous at c.
The proof in the other cases is similar. Just note for the case f/g that if x
n
→ c and
g(c) = 0, then g(x
n
) = 0 for all suﬃciently large n
5
.
If a < 0, a similar argument works with the sequence
−
1
n
, −
2
n
, −
3
n
, −
4
n
, −
5
n
, . . .
Finally, choosing n so
√
2
n
< b −a and applying a similar argument to the sequence
√
2
n
,
2
√
2
n
,
3
√
2
n
,
4
√
2
n
,
5
√
2
n
, . . .
gives the result for irrational numbers.
4
If A and B are sets, then their intersection A ∩ B is the set of numbers in both A and B.
Their union A∪ B is the set of numbers in at least one of A and B.
5
If g(c) > 0, by continuity of g at c and the deﬁnition of convergence of a sequence, g(x
n
) ∈
[
1
2
g(c),
3
2
g(c)] for all suﬃciently large n and so it is positive. The argument in case g(c) < 0 is
similar.
5.2. PROPERTIES OF CONTINUOUS FUNCTIONS 22
The composition of two continuous functions is continuous. (See Adams page 35 for
a discussion about the composition of functions, or the 1998 Calculus Notes page 16.)
Theorem 5.7. Suppose f and g are continuous. Then f ◦ g is continuous.
Proof. The domain D of f ◦ g is the set of numbers x such that both x ∈ D(g) and
g(x) ∈ D(f).
Suppose c ∈ D. Let x
n
→ c and x
n
∈ D. It follows that g(x
n
) → g(c) since g is
continuous at c. It then follows that f(g(x
n
)) → f(g(c)) since f is continuous at g(c) (note
that g(x
n
) ∈ D(f)). In other words, (f ◦ g)(x
n
) → (f ◦ g)(c), and so f ◦ g is continuous
at c.
It follows from our results so far that rational functions (quotients of polynomials)
and in general functions deﬁned from other continuous functions by means of algebraic
operations and composition, will be continuous on their domain.
Example 5.8. The function
f
1
(x) = sin
1
x
is the composition of the two continuous functions sin(x) and 1/x
6
and so is continuous.
The domain of f
1
is the set of real numbers x such that x = 0. That is, D(f
1
) = { x  x =
0 }.
Similarly, the function
f
2
(x) = xsin
1
x
is continuous on its domain, which is the same domain as for f
1
.
However, there is an interesting diﬀerence between f
1
and f
2
. In the case of the latter
we can deﬁne a new function g
2
by
g
2
(x) =
_
xsin
1
x
x = 0
0 x = 0.
Then D(g
2
) = R and g
2
(x) = f
2
(x) if x = 0, i.e. if x ∈ D(f
2
). Moreover, g
2
is continuous
on its domain R.
To show continuity of g
2
at x = 0, take any sequence x
n
→ x. For all suﬃciently
large n, x
n
∈ D(f
2
), and so g
2
(x
n
) = f
2
(x
n
). It follows that g
2
(x
n
) → g
2
(x) since
f
2
(x
n
) → f(x) by the continuity of f. This means g
2
is continuous at x if x = 0.
To show continuity of g
2
at x = 0, take any sequence x
n
→ 0. Then
−x
n
 ≤ g
2
(x
n
) ≤ x
n
,
6
The notation may seem a bit confusing. You may ask “is it the same x in both cases”? But
this is not the right way to look at it. By the function sin x, is meant the function which assigns
to each real number x (say) the real number sin x. If we said the function sin y, or just sin, we
would mean the same thing.
Similarly, the function 1/x, or 1/y, or “the reciprocal function”, all mean the same thing.
5.3. SUPREMUM OF A SET 23
and so g
2
(x
n
) → 0 (= g
2
(0)) by the Squeeze Theorem. (We need to be a bit careful since
some of the x
n
may equal zero.) This means g
2
is continuous at 0.
In the case of f
1
there is no way of extending the function to a continuous function g
1
deﬁned on all of R. This is essentially because there is no number y such that f
1
(x
n
) → y
for every sequence x
n
→ 0 (with x
n
= 0.)
We sometimes say that f
2
has a removable singularity at 0, and that the singularity
of f
1
at 0 is not removable.
5.3. Supremum of a set
A set which is bounded above need not have a maximum member, but there
will always be a supremum. If the supremum belongs to the set then it is
just the maximum, if the supremum does not belong to the set then there is
no maximum. There is always a sequence of elements from the set which
converges to the supremum.
For the important results in the next section, and elsewhere, we need the idea of the
supremum of a set.
If we consider sets such as [1, 2] or (1, 2] or the set A = {2, 1, 0, −1, . . . } of integers
less than or equal to 2, we see that each of these sets contains a largest (or maximum)
member, namely 2.
On the other hand, the set (1, 2), or the set B of irrrationals less than 2, does not
contain a largest member. But we say that 2 is the supremum of the set. In the ﬁrst three
examples we also say that 2 is the supremum of the set.
The precise deﬁnitions are a follows:
Definition 5.9. Let S be a set of real numbers.
We say a is the maximum member of S if
1. x ≤ a for every x ∈ S,
2. a ∈ S.
We write a = max S.
We say a is the supremum (or sup) of S if
1. x ≤ a for every x ∈ S,
2. for every ε > 0 there is at least one member of S in the interval [a −ε, a].
We write a = sup S.
First note that by the previous deﬁnition, 2 is the maximum element of [1, 2], (1, 2]
and A. Next note by the deﬁnition that 2 is the supremum of these three sets as well as
of the sets (1, 2) and B.
Note that if S has a maximum member a, then a is also the supremum of S. This is
immediate, since property 1 is the same in each case, and if a ∈ S then we can just take
a as the member of S in [a −ε, a].
Note that if the supremum a of S exists and is also a member of S, then it is the
maximum member of S (this is immediate from the deﬁnition of “maximum member”).
Note that there can be at most one maximum member of S. For if a
1
and a
2
were
both maximum elements
7
, then we would have from property 1 that a
2
≤ a
1
and a
1
≤ a
2
,
which implies a
1
= a
2
. Similarly, there can be at most one supremum of S, again by
property 1.
Theorem 5.10. Suppose S is a set of real numbers and a = sup S. Then there exists
a sequence (x
n
) of elements of S such that x
n
→ a.
Proof. From property 2 of the deﬁnition of supremum, for each ε > 0 there is a
member of S in the interval [a −ε, a].
For each natural number n there is thus a member of S in [a −
1
n
, a]. For each n,
choose one such member and denote it by x
n
. Then x
n
→ a, and we are ﬁnished.
7
A member of an element of a set means the same thing.
I
I
1
I
2
I
3
I
4
° °
S
  
c b
n
x ε S
5.3. SUPREMUM OF A SET 24
Example 5.11. Going back to the ﬁve examples at the beginning of this section, we
can take x
n
= 2−
1
n
for each of the sets [1, 2], (1, 2] and (1, 2). For the sets [1, 2] and (1, 2]
we could also take the constant sequence x
n
= 2, and this sequence also works for the set
A. For the set B we could take x
n
= 2 −
√
2
n
.
Although a set need not have a maximum member, even if it is bounded above
8
, the
following theorem shows that every set which is bounded above does have a supremum.
(although it may or may not be a member of S).
Theorem 5.12. Let S be set of real numbers which is bounded above. Then S has a
supremum.
Proof. We use a bisection argument.
Since S is bounded above we can ﬁnd an interval I = [a, b] such that x ≤ b for every
x ∈ S and such that there is at least one member of S in I.
Divide I into two closed bounded intervals of equal length, meeting only at the mid
point of I. If the right interval contains at least one member of S, then we take I
1
to be
this interval. Otherwise (if the right interval contains no members of S) we take I
1
to be
the left interval. Let I
1
= [a
1
, b
1
]. In either case it follows that x ≤ b
1
for every x ∈ S and
there is at least one member of S in I
1
.
Divide I
1
into two closed bounded intervals of equal length, meeting only at the
midpoint of I
1
. If the right interval contains at least one member of S, then we take I
2
to
be this interval. Otherwise, (if the right interval contains no members of S) we take I
2
to
be the left interval. Let I
2
= [a
2
, b
2
]. In either case it follows that x ≤ b
2
for every x ∈ S
and there is at least one member of S in I
2
.
Divide I
2
into two closed bounded intervals of equal length, meeting only at the
midpoint of I
2
. If the right interval contains at least one member of S, then we take I
3
to
be this interval. Otherwise, (if the right interval contains no members of S) we take I
3
to
be the left interval. Let I
3
= [a
3
, b
3
]. In either case it follows that x ≤ b
3
for every x ∈ S
and there is at least one member of S in I
3
.
Etc.
In this way we construct a sequence of intervals (I
n
), each containing the next. That
is
I
1
⊃ I
2
⊃ I
3
⊃ · · · .
Moreover,
a
1
≤ a
2
≤ a
3
≤ · · · . . . · · · ≥ b
3
≥ b
2
≥ b
1
.
Since (a
n
) is increasing and bounded it follows that a
n
→ a
∗
, say. Similarly b
n
→ b
∗
,
say.
It follows that a
∗
= b
∗
. To see this, note that for every n we have a
n
≤ a
∗
≤ b
n
and
a
n
≤ b
∗
≤ b
n
, and so a
∗
− b
∗
 ≤ b
n
− a
n
. But b
n
− a
n
= (b − a)/2
n
and so a
∗
= b
∗
,
as otherwise we would have a contradiction for suﬃciently large n (by the Archimedean
Axiom).
We now deﬁne c = a
∗
= b
∗
and show it is the supremum of S.
To check property 1, suppose x ∈ S. Then x ≤ b
n
for every n, and so x ≤ limb
n
(see
the following proposition). . That is, x ≤ c for all x ∈ S.
8
Recall that a set S (of real numbers) is bounded above if there is a number K such that
x ≤ K for all x ∈ S.

c

b
n

x ε S

a
n

c  ε
c x x  ε x + ε
 
 
5.4. THREE BIG THEOREMS 25
To check property 2, let ε > 0 be given. Since a
n
→ c and a
n
is increasing, we can
choose n so a
n
∈ [c −ε, c]. . From before
we have a
n
≤ c ≤ b
n
. There is at least one member x ∈ S in I
n
= [a
n
, b
n
] by the way we
constructed the I
n
, and every member of S is ≤ c by property 1. Hence this member of
S is in fact in [a
n
, c] and hence in [c −ε, c]. This proves property 2.
In the previous theorem we used the following easy result, which is also useful else
where.
Proposition 5.13. Suppose (x
n
) is a convergent sequence and x
n
≤ c (x
n
≥ c) for
all n. Then limx
n
≤ c (limx
n
≥ c ).
Proof. Let x = limx
n
and suppose x
n
≤ c for all n. Suppose, in order to gain a
contradiction, that x > c.
Let ε be any positive number less than x −c . Then for all
suﬁciently large n, x
n
∈ [x −ε, x + ε]. This contradicts x
n
≤ c.
Hence we must have x ≤ c.
A similar proof applies for the “≤” result.
Note that if x
n
< c for all n, then it does not necessarily follow limx
n
< c; we can
still only deduce in general that limx
n
≤ c. For example, −
1
n
< 0 for all n, but lim
1
n
= 0.
Remark 5.14. We can also deﬁne the minimum member of a set and the inﬁmum of
a set, and prove that if a set of real numbers is bounded below, then it has an inﬁmum.
5.4. Three big theorems
A continuous function deﬁned on a closed bounded interval is bounded above
and below, and takes a maximum and a minimum value. If a continuous
function deﬁned on an interval takes two values, then it takes all values in
between.
The theorems in this section are global, in that they refer to properties of continuous
functions over their entire domain, or assert the existence of a point in the domain with a
particular property. The require the Bolzano Weierstrass Theorem in their proof, which
in turn uses the Cauchy Completeness Axiom.
The properties of continuous functions in Section 5.2 followed from the deﬁnitions in
a relatively straightforward way. They are local properties in that they essentially refer
to properties of continuous function near a prescribed point (although the point may be
an arbitrary one in the domain).
We say a function f deﬁned on a set E is bounded on E iﬀ there exists a real number
M such that
f(x) ≤ M for all x ∈ S.
Theorem 5.15. Suppose f is a continuous function deﬁned on a closed bounded in
terval. Then f is bounded.
Proof. Suppose (in order to obtain a contradiction) that f is a continuous function
deﬁned on an interval [a, b] but f is not bounded. This means that for each real number
M there must be some x ∈ [a, b] such that f(x) > M.
(In the following diagram, f is of course not continuous. The diagram is just to give
an indication of the argument used to obtain a contradiction.)
a
x
1
x
n
x
2
x
3
x
4
b
f(x
1
)
f(x
2
)
f(x
3
)
f(x
4
)
f(x
n
)
c

graph of f
5.4. THREE BIG THEOREMS 26
In particular, for each n choose x
n
∈ [a, b] such that f(x
n
) > n. (Note that f(x
n
)
diverges to inﬁnity.)
By the Bolzano Weierstrass theorem (and the remark which follows it) there is a
subsequence (x
n
k
)
k≥1
which converges to some c ∈ [a, b]. By continuity of f, f(x
n
k
) →
f(c). But this contradicts the fact that for each n, f(x
n
) > n.
Hence f is bounded.
Remark 5.16.
1. The corresponding result is not true for open intervals or unbounded intervals.
For example, if f(x) = 1/x for x ∈ (0., 1], then f is continuous on (0, 1] but is not
bounded on (0, 1]. Also, if f(x) = x for x ∈ [0, ∞) then f is continuous on [0, ∞)
but is not bounded on [0, ∞).
2. The result is not as obvious as might ﬁrst appear; it is possible to construct some
pretty wild continuous functions.
For example, there are continuous functions which are nowhere diﬀeren
tiable. See Section 5.3 p 55 of the 1998 Calculus Notes
3. The analogous result is not true for the rationals. For example, let f(x) =
1/(1 − x
2
) for 1 ≤ x ≤ 2. This is continuous at every point other than x =
√
2
(and in particular is continuous at every rational point) and it takes rational values
at rational points. However, it is not bounded.
This explains why we need the Cauchy Completeness Theorem in the proof of
the previous theorem (it was used to prove the Bolzano Weierstrass Theorem) —
if we only required the other axioms in the proof then because these axioms are
true for the rationals it would follow that the analogous result would also be true
for the rationals.
We say a function f deﬁned on a set E takes its maximum value at c ∈ E and c is
a maximum point iﬀ f(c) ≥ f(x) for all x ∈ [a, b]. We say f takes its minimum value at
d ∈ E and d is a minimum point iﬀ f(d) ≤ f(x) for all x ∈ [a, b].
A function can take its maximum or minimum value at more than one point (a
constant function is a simple example).
A function can be bounded and not take a maximum or a minimum value. For
example, if
f(x) =
_
x −1 ≤ x ≤ 1, x = 0
1 x = 0,
then f is bounded on [−1, 1] but does not take a minimum value.
Theorem 5.17. Suppose f is a continuous function deﬁned on a closed bounded in
terval. Then f takes a maximum value and a minimum value.
Proof. Suppose f is a continuous function deﬁned on the closed bounded inter
val [a, b].
Let S be the set of all values taken by f. This means S = { f(x)  x ∈ [a, b] }.
The function f is bounded by the previous theorem, and another way of expressing
this is to say that the set S is bounded. From Theorem 5.12 it follows that S has a
supremum α, say.
α
y
n
y
2
y
1
y
3
x
1
x
3
x
n
x
2
a b c
S
graph of f
a b c
graph of f
γ
A
f(b)
f(a)
5.4. THREE BIG THEOREMS 27
We claim that α is a member of S, in other words that α = f(c) for some c ∈ [a, b].
This will imply that f(x) ≤ f(c) for all x ∈ [a, b] and so c is a maximum point.
Proof of claim. There exists a sequence (y
n
) of elements of S such that y
n
→ α (by
Theorem 5.10).
For each y
n
there exists (at least one) x
n
∈ [a, b] such that f(x
n
) = y
n
. By the
Bolzano Weierstrass Theorem there exists a subsequence (x
n
k
) of (x
n
) which converges
to some c ∈ [a, b]. By continuity of f, f(x
n
k
) → f(c), i.e. y
n
k
→ f(c). On the other hand,
(y
n
k
) is a subsequence of (y
n
) and so y
n
k
→ α.
9
Because a sequence can have only one limit (Theorem 4.15) it follows that f(c) = α.
This proves the claim, and hence that c is a maximum point.
The proof that there is a minimum point is similar.
Remark 5.18. While it may seem clear that there is a maximum point c, particu
larly in the case of the diagram, there are some rather wild continuous functions as noted
before. Just from the deﬁnition of supS we can ﬁnd for each ε > 0 a number x ∈ [a, b] such
that f(x) is within ε of α, even if f is not continuous but is merely bounded. However,
to show there is some c such that one actually has f(c) = α requires the continuity of f
and uses the Cauchy Completeness Axiom.
The last result implies that if a continuous function deﬁned on an interval I (not
necessarily closed or bounded) takes two particular values, then it must take all values
between. In other words, for any two points a, b ∈ I and any γ between f(a) and f(b)
then f(c) = γ for some c ∈ [a, b].
Theorem 5.19. Suppose f is continuous on [a, b]. Then for any γ between f(a) and
f(b) there exists c ∈ [a, b] such that then f(c) = γ.
Proof. Suppose f(a) < γ < f(b) (the case f(a) > γ > f(b) is similar).
In order to prove there is some c ∈ [a, b] such that f(c) = γ, let
A = {x ∈ [a, b]  f(x) ≤ γ }.
(That is, A is the set of all x ∈ [a, b] such that f(x) ≤ γ.)
Since A is a bounded set it follows that it has a supremum c, say.
We want to show that f(c) = γ.
9
It is clear, and not hard to prove, that if a sequence converges then any subsequence con
verges to the same limit.
5.4. THREE BIG THEOREMS 28
There is a sequence x
n
∈ A such that x
n
→ c (Theorem 5.10). By continuity,
f(x
n
) → f(c). Since f(x
n
) ≤ γ for all n, it follows f(c) ≤ γ (Proposition 5.13). (So, in
particular, c ∈ A.)
Since c = b (because we now know that f(c) ≤ γ, while f(b) > γ), there is a sequence
x
n
→ c and c < x
n
< b. But f(x
n
) > γ (since otherwise x
n
∈ A, which contradicts
c = sup A) and so f(c) ≥ γ (Proposition 5.13 again).
Because f(c) ≤ γ and f(c) ≥ γ it follows f(c) = γ.
.
CHAPTER 1
Introduction
This year I will take a slightly diﬀerent approach to the material. The main reference will be the 1998 Notes, but adjusted as indicated in the current Notes. The main diﬀerence this year is that sequences will be introduced earlier and will play a more central role. One still gets the same results in the end, but I think the main ideas will be easier to understand in this manner.
i
what Theorem 2.11 which shows that 2 is irrational. extra nonexaminable material to put the other material in a broader context. certain existence properties.) We will introduce the Completeness Axiom after we discuss sequences. division. The reason is that Axioms 1–13 also hold for the rational numbers. All the usual properies of addition. Thus at this stage we assume the real number system satisﬁes Axioms 1–13. CHAPTER 3 LIMITS Omit this chapter. The Completeness Axiom is true for the set of real numbers but the analogous statement is not true for the set of rational numbers. and so if we could prove this fact then it would also be true in the rational numbers. there would be a a positive rational number whose square is 2. such as the existence of a number x for which x2 = 2. and of inequalities.e. But this would contradict Theorem 2.11 proves is that if there is a number whose square is 2 then that number cannot be rational.5 √ In the summary.5 from Theorem 2.12 onwards. multiplication. i. It will be in a diﬀerent. leaving only The√ orem 2. but equivalent. We will return to it later. More precisely. ( One cannot prove from Axioms 1–13 that there is a positive number whose square is 2. omit everything except the ﬁrst sentence “We prove that 2 is irrational”.11. ) Sections 2. are not true of the rationals. see Section 5. subtraction.1–3 As before. Section 2. 1 .CHAPTER 2 The Real Number System Sections 2. (However. Section 2.e. i. are consequences As noted in Remark 2. Omit everything in Section 2. after ﬁrst treating sequences. form to that in the 98 Notes.4 Omit this section. Note that the last three sections are marked . the set of rational numbers also satiﬁes Axioms 1–13.6–10 As before. and hence have all these same properties.6(98).
1]. −1. we may just write the ﬁrst few terms. 1 1 1. . 13. a sequence is a function f whose domain is the set of natural numbers. . Examples of sequences We introduce the idea of a sequence and give a few examples. but we do considerably more material than this.CHAPTER 4 Sequences This replaces the chapter from the 98 Notes.. • One can represent a sequence by its graph. element. 1. . −1/4. ) has graph 2 . . 2. ak+1 . . a3 . a2 = 1. . . Section 10. For example the sequence (−1)n+1 /n2 . • More precisely. where k is some other integer than 1 (e. . a n . −1/16. 3. 1. . 1/9. ak+2 . (1. Hence the sequence is 1. but no last. . 21. as in the ﬁrst three examples above. such as an = 1 + which gives the sequence 1 + 1. . . . . where in the above example f (n) = an . . • A sequence can be written in the form a1 . an = an−1 + an−2 if n ≥ 3. a2 . 5. 2 3 1. 4. Simple examples are 1. . • A sequence may be given by a method for calculating each element of the sequence in terms of preceding elements. 2. 0). 1. • The general term an may instead be given by a formula.e.. . Another reference is Chapter 2 of Fundamentals of Analysis by Michael Reed.. . 3. 4. . . The reference for this chapter is [Adams. i. . . this book will also be the text for the second year course “Analysis and its applications” beginning next year. 1. . . . 1+ 1 3 3 .1. . . We often just write (an ) or (an )n≥1 to represent the sequence. • A sequence is an inﬁnite list of numbers with a ﬁrst. One example is the Fibonacci sequence a1 = 1. 8. 1 + 1 2 2 1 n n ..g. . −1. 1. • If the pattern is clear. . • Sometimes it is convenient to write a sequence in the form ak . .
. 1 + .01 of a. 6 both converge to 1. as the terms of the sequence 1 2 3 a1 = 1.1 of a. a3 = 1 . as n gets larger and larger. In the above situation we could choose N as follows: . this is not usually useful. . 1 + 10 . 5 6 7 4 where the vertical scale is somewhat distorted. but the distance an − 1 between an and 1 is not a decreasing function in either case. but also to 3 (for example). . a2 = 1 . all terms are within . Also.01 if n ≥ 300. This is not exactly what we want. or N (ε) to emphasise that it may depend on ε). 1 − 10−7 . etc. an − a ≤ . However.4.1 if n ≥ 50. . . A more precise version of what we mean by “the sequence (an )n≥1 converges to the limit a” is that beyond some term in the sequence. . LIMIT OF SEQUENCES 3 .001 of a. . . . 2 3 4 get closer and closer to 2. a4 = 1 . 1 + 2 4 1 1 −3 1.2. 1 + . we may have an − a ≤ . 1 − 10−3 . there is always an integer (let’s call it N . a6 a5 a 2 a4 1 0 a3 a1 1 4. . The idea is that “the sequence (an )n≥1 converges to the limit a” if the terms of the sequence get closer and closer to a as n gets larger and larger. We are interested in the behaviour of sequences (an )n≥1 for large n. 1 − 10−5 . etc. . More precisely: no matter which positive number is chosen (let’s call it ε). the sequences 1.2. an − a ≤ . We give some examples. 3 . 1 + 10−7 . Limit of sequences We discuss the idea of the limit of a sequence and give the precise deﬁnition. . . 1 + 10−5 . beyond a further point in the sequence all terms are within . For example. It is often more helpful to think of a sequence geometrically in terms of points on the real line.001 if n ≥ 780. . 1 + 2 4 1 . 1 + . beyond a further point in the sequence all terms are within . . 1 + 1 1 . 6 1 . such that the N th and all later members of the sequence are within ε of a. 1 2 .
We want to ﬁnd N such that (4. 1. LIMIT OF SEQUENCES 4 ε . 1 √ + 1. 3. a + ε] (provided ε > 0) there is an integer N such that all members of the sequence from the N th onwards belong to this interval”. ε 1 . Solution. But the above proof works of course for any ε > 0. ε if n2 ≥ 1 . −1/16. We say that the sequence (an ) converges to a limit a. take N= where [ ] denotes “the integer part of”. Let ε > 0 be given. . an − a ≤ ε. if 1 n≥ √ . 9. 2.g. . . Also. we need to have a precise deﬁnition. also does not converge (we sometimes say that it “diverges to +∞”). 1. does not converge and the sequence 1. . the sequence 1. But an − 1 = Since 1 ≤ε n2 i. e.1. does 2 3 not converge (it “oscillates back and forth between ±1”). . . .2. for example N = 4 (or anything larger). the sequence 1. 1. the sequence 1. or the sequence given by a1 = 1 and an+1 = 1 an + 2 for n ≥ 1. n sin n n≥1 . . the sequence 1. 16. . will converge. if for every positive number ε there exists an integer N such that (4. 1 . Show that the sequence given by an = 1 + n2 converges to 1 according to the deﬁnition.1 .001 N 50 300 780 For example. converges to 0. and the symbol ∞ by itself here has no meaning. . For this and more complicated 2 examples we need a precise deﬁnition of convergence. it is not 1 1 immediately clear if the sequences (1 + n )n n≥1 .01 we can take N = 10 (or anything larger). −1/4. n2 √ we can take N to be any integer ≥ 1/ ε. −1. . converges to 0. in order to prove general theorems about convergent sequences. or an → a. . . .1) is true with a = 1.4. We sometimes say an converges to a as n approaches ∞ and write n→∞ lim an = a or an → a as n → ∞. If ε = .e. . 4. 1 .1 we can take any integer N ≥ 1/ . a − ε]. according to this criterion. The following deﬁnition makes our previous idea very precise. For example. .01 .) 1 Example 4.1.1) n≥N implies Note that an − a ≤ ε is equivalent to an ∈ [a − ε. 1/9. 4. just as → has no meaning by itself. 1. Definition 4. If ε = . (Note that ∞ is not a number. and write lim an = a. ε √ Thus if ε = .001 we can take N = 32 (or anything larger). Thus another way of expressing the deﬁnition is “for every interval [a − ε.2. −1. Often it is not clear whether or not a sequence converges.
8125. a3 − 4 = 3/22 . Remark 4. Let ε > 0 be given. a4 − 4 = 3/23 . 3. in the previous example. . Hence (4. Example 4. np This implies the required limit exists and equals zero. Example 4. Let ε > 0 be given.4. We want to ﬁnd N such that1 (4. Thus an+1 − 4 = 1 1 1 1 an + 2 − 4 = an − 2 = (an − 4) = an − 4. and an+1 = + 2 for n ≥ 1.9765625.2. However. ε and it then follows that c ≤ ε if n ≥ N. 2n−1 ≥ 3/ε. page 522]).625. an − L ≤ ε. Have a look again at (4. . . and we ﬁrst use this to get a formula for an+1 − 4 in terms of an − 4. 3. . In particular.953125. i. One way to prove this is as follows. . Then c ≤ε np if np ≥ c .e. (n − 1) ln 2 ≥ ln(3/ε).2) is true for N =1+ ln 3 ε . Solution. . Thus we can take any integer N ≥ c 1/p . 3.e. We previously mentioned the sequence given by a1 = 1. We have a formula for an+1 in terms of an . we will often do this sort of thing. i.90625. . . 3. (See [Adams.5.1) and compare the four statements n≥N n≥N n>N n>N 1 We implies implies implies implies an − L ≤ ε. an − L < ε. i. Show that limn→∞ c np = 0 for any real number c and any p > 0.e.5. 2 2 2 2 Thus a1 − 4 = 3. In general2 an − 4 = 3/2n−1 . The ﬁrst few terms are 1.3. if n≥ c ε 1/p . 3. but we have not yet shown how to deﬁne logarithms and establish their properties from the axioms. 2 This will often write “⇒” for “implies”. but it is not necessary to do so. ln 2 You may object that we used ln. n ≥ 1 + ln(3/ε)/ ln 2. 2n−1 This last inequality is equivalent to 2n−1 /3 ≥ 1/ε.2) n≥N ⇒ an − 4 ≤ ε. It seems reasonable that the sequence is converging to 4. 2. This is a valid criticism.4. the development of the theory will not depend on the examples. Proof. LIMIT OF SEQUENCES 5 1 a 2 n Example 4. could easily be proved by induction. we will not do it when we are establishing the underlying theory.e. ε i. 3. . a2 − 4 = 3/2.25. an − L < ε. the natural logarithm. It follows that 3 an − 4 ≤ ε for those n such that ≤ ε. But in order to have interesting examples.
Let K be any (positive) real number. Example 4. Proof. the Completeness Axiom. which is equivalent to n ≥ ln K/ ln 2. • The Cauchy Completeness Axiom. an − a < ε. because 1/an  ≤ ε is equivalent to an  ≥ 1/ε.3. There is a special case that is important. (an ) diverges to −∞ if for each real number K there is an integer N such that an ≤ K We write an → −∞ or lim an = −∞. Section 2. This may happen in various ways. Similarly. Note that if an → ∞ then 1/an → 0 (assuming an = 0 in order that 1/an is deﬁned). The sequence (an ) diverges to +∞ if for each real number K there is an integer N such that an ≥ K for all n ≥ N.4. There are three examples in the ﬁrst paragraph on page 4. THE ARCHIMEDEAN AXIOM 6 These statements are certainly not equivalent.3.4.) Note. we say that it diverges. as well as the Algebraic and Order axioms for the ral numbers. But this inequality is equivalent to n ln 2 ≥ ln K. for every positive number ε there exists an integer N such that for every positive number ε there exists an integer N such that for every positive number ε there exists an integer N such that Do you see why? I will discuss this in class. But the latter is true for all suﬃciently large n since an → ∞. This is as in the following deﬁnition where for each real number K. the following four statements are equivalent! for every positive number ε there exists an integer N such that n≥N n≥N n>N n>N implies implies implies implies an − a ≤ ε. Definition 4. you will see that there is another axiom. this only occurs for the last of these examples. We write an → ∞ or lim an = ∞. an − a ≤ ε. by the way. Show from the deﬁnition that 2n → ∞. and so the former is also true for all suﬃciently large n. We want to show that for all suﬃciently large n. an ≥ K for all suﬃciently large n. and instead of the Completeness Axiom we introduce two axioms: • The Archimedean Axiom. for all n ≤ N. This means 2n → ∞. This follows from the deﬁnitions. this last inequality is true whenever n ≥ N . . In the case an → −∞. Hence if we let N = 1 + ln K . However. an − a < ε. Note that of all the examples in the ﬁrst paragraph on page 4.6. 4. that we never say “(an ) converges to ∞” or “(an ) converges to −∞”. The Archimedean Axiom If you look at the 1998 Calculus Notes. we take a diﬀerent approach. 2n ≥ K. Which version does Adams use? If the sequence (an ) does not converge. (In the case an → ∞. In these Notes. think of K as large and negative. Hence ln 2 n ≥ N ⇒ 2n ≥ K. think of K as large and positive.7.
More precisely. expressions of the form “for all” or “there exists”). Corollary 4. some elementary property (such as a + b = b + a.4. for all real numbers. as I discuss below in a starred comment. Thus we have used the Archimedean Axiom to prove 1/n → 0. where the equivalence is essentially set as an (advanced) exercise. Here is a simple consequence. using just the algebraic and order axioms. or are one of the 2. THE ARCHIMEDEAN AXIOM 7 The second will be discussed in a later section and the ﬁrst will be discussed here. in which case the real number which is asserted to exist is in fact unique. they are “isomorphic”). or a + 0 = a. Proof. this is discussed in the last Chapter of the book Calculus by Michael Spivak. see the last exercise in Chapter 2 of Spivak. additive or multiplicative inverse axioms. One diﬀerence between the Archimedean Axiom and the Algebraic and Order Axioms is as follows. On the other hand. If n ≥ N then 1/n ≤ 1/N (by properties of inequalities). From the Archimedean Axiom there is a natural number N such that N ≥ 1/ε. this is the same as asserting 1/n → 0. Introduction to Analysis. The Algebraic and Order Axioms are either of the form 1. where “elementary” means that the property does not involve any further “quantiﬁers” (i. Consider the ε and N as in the Corollary. If you are very adventurous. While the axiom may seem obvious. If x ≤ n. From properties of inequalities it follows that 1/N ≤ ε. then n may be replaced by any larger natural number. 2. the statement 1/n → 0 is in fact equivalent to the Archimedean Axiom. and indeed it is clearly true for the model of the real numbers which we have in our mind. Thus if we assume just the algebraic and order axioms. Here then is the Archimedean Axiom: Axiom (Archimedean Axiom). see the book Kripke. namely that the statement 1/n → 0 implies the Archimedean Axiom.3. (Archimedean Axiom + Cauchy Completeness Axiom) ⇒ Completeness Axiom. For every real number x there is a natural number n such that x ≤ n. it can be shown that any two models of all the axioms are essentially the same (more precisely. It can also be proved that these two axioms together are equivalent to the Completeness Axiom used in the 1998 Notes.e. If you are adventurous. Completeness Axiom ⇒ (Archimedean Axiom + Cauchy Completeness Axiom). If x ≤ n is true. Since 1/N ≤ ε it follows that n ≥ N ⇒ 1/n ≤ ε. x < n + 2. Moreover. It can be shown that no more axioms are necessary. etc. p38 Q’s 9 &10. it does not actually follow from the algebraic and order axioms. More precisely. By essentially arguing in the reverse direction we can also prove the converse.8. or a < b ⇒ a + c < b + c) is true. we can ﬁnd natural numbers strictly greater than x. In particular. x < n+1. one can prove that 1. . For every real number ε > 0 there is a natural number N such that 1/N ≤ ε. Since ε > 0 was an arbitrary positive number. Suppose ε > 0. the natural number n asserted to exist by the Archimedean Axiom for each real number x is certainly not unique. the set N of natural numbers (which is needed in the statement of the Archimedean Axiom) also involves a certain level of complexity in its (precise) deﬁnition. then by properties of inequalities. There is one immediate consequence of the Archimedean Axiom which is quite important.
The main point to remember from this section is that in speciﬁc ex1 amples where we have to show that for any real number of a given form (such as √ε in Example 4.4.4. b a 0 c     1  d   e d<e<c<0< 1<a<b Fattened up copy of R A "number'' on any line is less than any number to the right . They are sometimes called the Hyperreals! Part of any such model looks like a “fattened up” copy of R. and between any two copies there are inﬁnitely many other copies.7 there was a hidden application of the Archimedean Axiom somewhere in the last three lines of the solution. then so does their sum. . You should be aware of this. Properties of limits We prove that limits of sequences behave as we expect under addition.2) there is always an integer N at least as large (N will depend of course on 1 the particular real number √ε ). ﬁrst that (an + bn ) is convergent. It is not usually not very eﬃcient to use the deﬁnition of a limit in order to prove that a sequence converges. not in deﬁnitions or theorems. and I only cheated in the examples. we prove a number of theorems which will make things much easier. Remark 4. in the sense that it contains a copy of R together with “inﬁnitesimals” squeezed between each real a and all reals greater than a. The result is not very surprising. since if an is getting close to a and bn is getting close to b then we expect that an + bn is getting close to a + b. See the following crude diagram. and we prove the Squeeze Theorem. PROPERTIES OF LIMITS 8 So now I am forced to confess that I cheated in the previous section. is a consequence of the fact that there are models of the algebraic and order axioms which do not satisfy the Archimedean Axiom. the Archimedean Axiom is probably needed.2–4. in the sense that it does not follow from the previous axioms. 4. and moreover the limit of the new sequence is just the sum of the limits of the original sequences. Between any two lines there is an infinite number of other lines. The fact that the Archimedean Axiom is necessary. and second that the actual limit is a + b. The ﬁrst theorem shows that if two sequences converge. The theorem may be written more brieﬂy as: if an → a and bn → b then an + bn → a + b . This part is followed and preceded by inﬁnitely many “copies” of itself. but after a few more examples we will adopt our previous cavalier attitude and not explicitly note when the axiom is needed.9.4. 4.3 In each of Examples 4. So you may consider 3 I promise not to do it again. Instead. multiplication and division. Note that the theorem has two claims. subtraction.4. and less than any any number on any higher line. Find the hidden application for Easter.
e.10. and I will discuss them in class. . i. Fix this N . it follows that there exist real numbers M1 and M2 such that (4. But you should try to understand them. It follows from Deﬁnition 4. . . a2 . once to get information from the fact bn → b. PROPERTIES OF LIMITS 9 the theorem as a partial justiﬁcation that Deﬁnition 4. a2 . taking ε = 1. of convergent sequences can all be similarly established. From the deﬁnition of convergence. . M2 }. It follows that if n ≥ max{N1 . .3) and (4. .6) M1 ≤ an ≤ M2 for all n < N.5) and (4.6). . . Then (an + bn ) is a convergent sequence.1? Theorem 4.) From (4. M2 }.1 does indeed capture the informal notion of a limit. . once to get information from the fact an → a. quotients etc. 2. that (an + bn ) converges and the limit is a + b. ∗ ∗ The required result follows by taking M = max{M1 . Then the sequence is bounded. N2 } then (an + bn ) − (a + b) = (an − a) + (bn − b) ≤ an − a + bn − b by the triangle inequality ε ε ≤ + by (4.5) a − 1 ≤ an ≤ a + 1 for all n ≥ N. there is a real number M such that an  ≤ M for all n. why do we use ε/2 in (4.3) n ≥ N1 implies an − a ≤ ε/2.1. Theorem 4. It is certainly not true in general. M2 = max{a + 1.4) n ≥ N1 implies bn − b ≤ ε/2. . . Since an → a there exists an integer N1 (by Deﬁnition 4.4. aN −1 is ﬁnite. M1 }. The next easy result is useful in a number of situations.4. Suppose (an ) and (bn ) are convergent sequences with limits a and b respectively.4). with N = max{N1 .3) and (4. aN −1 } and M2 = max{a1 . and its limit is a + b. . aN −1 }. We state them together in the following theorem (which includes Theorem 4. and ﬁnally to deduce that an + bn → a + b.10). . consider for example the sequence 1.11. . and why is this justiﬁable by Deﬁnition 4. Suppose an → a.1) such that (4. Let ε > 0 be given. The proofs are material and are at the end of this section. Proof. (Just take M1 = min{a1 . The standard properties about products. . . 3.4) 2 2 = ε. there is an integer N such that (4. 4.1) such that (4. Notice in the (rather subtle and elegant) proof how the deﬁnition of a limit is used three times. N2 }. Since the set of terms a1 . Since bn → b there exists an integer N2 (again by Deﬁnition 4. ∗ ∗ M1 ≤ an ≤ M2 for all n. Proof. By the way. a2 . ∗ ∗ where M1 = min{a − 1. .
= 7 1 3n2 − 7n + 1 3 − n + n2 Since the numerator and denominator converge to 2 and 3 respectively. PROPERTIES OF LIMITS 10 Theorem 4. Proof. But this implies an ∈ [a − ε. a + ε] for all suﬃciently large n. (For motivation. 3n2 −7n+1 Hence (again from the theorem) an → 0. Let an = 1 + − (1 + 2−n ).14. and it converges to the same limit.16. Similarly. b + ε] for all n ≥ max{N1 .8) an ∈ [b − ε. Deﬁnition 4. N2 }. lim an bn = ab. Thus the assumption a = b led to a contradiction and so a = b. 1 1 It then follows from the previous theorem that 1+ √n → 1 (since we can think of 1+ √n as 1 obtained by adding the term 1 from the constant sequence (1) to the term √n ). bn b assuming b = 0 and bn = 0 for each n. 1 √ n 2 Example 4. Then a = b. Then the following limits exist and have the given values. Example 4. Suppose an → a and an → b. say for n ≥ N2 . lim can = ca. an a lim = . Then bn → L as n → ∞. say for n ≥ N1 . b + ε] for all suﬃciently large n. Suppose an → L and cn → L as n → ∞.12.4. Suppose an ≤ bn ≤ cn for all n (or at least for all n ≥ N for some N ). then the original sequence also converges. Theorem 4. a + ε] and an ∈ [b − ε. it follows that (4. lim an ± bn = a ± b. Since an → b.4. 1+ 1 √ n 2 → 1.15.7) an ∈ [a − ε. The next theorem says that a sequence cannot have two distinct limits. It is not surprising of course. Theorem 4. and c is a real number. it follows that (4. look at the following diagram). but note how it does follow from the actual deﬁnition of a limit. Suppose lim an = a. 2n2 −1 . Assume (in order to obtain a contradiction) that a = b. 1 + 2−n → 1. . Applying the theorem again. Let an = Write 1 2 − n2 2n2 − 1 .13. ε ε ε ε [  a ] [  b ] ε = ab/3 Since an → a.1. lim bn = b. The following theorem says that if a sequence is “squeezed” between two sequences which both converge to the same limit. it follows an → 0. which is impossible as ε = a − b/3. 1 We can prove directly from the deﬁnition of convergence that √n → 0 and 2−n → 0. Take ε = a − b/3 in the deﬁnition of a limit.
9) n ≥ N1 ⇒ an ∈ [L − ε. But ε was an arbitrary positive number. let N = max{N1 . Since −1 ≤ sin x ≤ 1. L + ε]. look at the following diagram). and so the result now follows from Theorem 4. i.4. Then since an ≤ bn ≤ cn it follows from (4.17. (This assumes c = 0. Since an → a there exists an integer N such that an − a ≤ ε/c for all n ≥ N. Lε [  an L   bn  cn L+ε ] Since an → L there is some integer N1 such that (4. n ≥ N2 ⇒ cn ∈ [L − ε. it follows that 3 − 1/n ≤ 3 + (sin cos n)/n ≤ 3 + 1/n.12. PROPERTIES OF LIMITS 11 Proof.10) that n ≥ N ⇒ bn ∈ [L − ε. • The result for an − bn now follows easily. But (−1)bn → (−1)b by the previous result with c = −1.4.11 in the proof. We want to show can − ca ≤ ε for all suﬃciently large n. and so can → ca by the deﬁnition of convergence. We ﬁrst establish the result for can . let ε > 0 be any positive number.) We will show that both terms are ≤ ε/2 for all suﬃciently large n. N2 }. and this sequence certainly converges to ca = 0.11) = an (bn − b) + b(an − a) ≤ an (bn − b) + b(an − a) = an  bn − b + b an − a. (For motivation. L + ε].e. is often very useful. Try to understand the ideas. Hence 3 + (sin cos n)/n → 3. sort of 1 ). (This trick of adding and subtracting the same term.10) Let N be the larger of N1 and N2 . Consider the sequence 3 + (sin cos n)/n. Since cn → L there is some integer N2 such that (4. Example 4. • The result for an bn uses Theorem 4. We ﬁnish this section with the promised proofs of the algebraic properties of limits.10 about the sum of two sequences. Just note that an − bn = an + (−1)bn . To see how to choose N . L + ε].9) and (4. 2 • Proof of Theorem 4. here it is an b. But if c = 0. and so it follows that bn → L. Let ε > 0 be given. write an bn − ab = an bn − an b + an b − ab (4. then the sequence (can ) is the sequence all of whose terms are 0. We want to show that an bn − ab ≤ ab for all n ≥ some N . But 3 − 1/n → 3 and 3 + 1/n → 3. Let ε > 0 be any positive number.) Multiplying both sides of the inequality by c we see can − ca ≤ ε for all n ≥ N. . As usual. although the material is (well.
This establishes the Claim.15) b − bn  b − bn  1 1 − = ≤ . The proof is similar to that in Theorem (4. .11). By the same argument as for the second term. since an → a. where N = max{N1 . and so in particular n ≥ N1 ⇒ bn  ≥ b/2. . 2 2 Since ε > 0 was arbitrary. N2 }. b − bn  ≤ K b ε. we have to take c = 0 (or c < 0). For this let ε > 0 be any positive number.13) together. 4 This . we see (4. bN −1  > 0. PROPERTIES OF LIMITS 12 For the second term b an − a. Next let c = min{b1 .12) and (4. then there may not be a minimum member of the set. bn b bn  b K b bn  ≥ K where for the inequality we have replaced bn  by the smaller (but still positive) number K in (4. where c > 0 since b1 . . (4. Since bn → b we can ﬁnd an integer N such that for all n ≥ N . . the result is certainly true if b = 0. We now proceed with the proof that 1/bn → 1/b. n ≥ N1 ⇒ bn − b ≤ b/2. since the term is then 0. and letting K = min{c. . . (4. }. In order to see how to choose N in the deﬁnition of convergence.4. we can choose N2 such that ε M bn − b ≤ for all n ≥ N2 . If b = 0.14). . Then n < N1 ⇒ bn  ≥ c. 1/n. 1/2. bN −1 }. and goes as follows: Since b = 0 we can choose ε = b/2 (> 0) in the deﬁnition of convergence and deduce that for some integer N1 .11 to deduce for some M that an  ≤ M for all n. . By increasing M if necessary take M = 0. not c > 0. . . . all of which are > 0. b/2} > 0. . we use Theorem 4.14) for all n. 1/3.11). 2 Putting (4. this proves an bn → ab. is an important point. such as {1. . we compute (4. it follows that if n ≥ N . if c ≤ 1/n for all n. 2 For the ﬁrst term an  bn − b.12) b an − a ≤ • We can prove that an /bn → a/b by ﬁrst showing 1/bn → 1/b and then using the previous result about products of sequences.4.4 Putting the previous results together. then ε ε an bn − ab ≤ + = ε. We ﬁrst prove the Claim: there is some number K > 0 such that bn  > K for all n. we can choose N1 such that ε an − a ≤ for all n ≥ N1 . 2 and so ε (4. . 2b and so ε for all n ≥ N1 . If we have an inﬁnite set of numbers. In fact for this example. .13) an  bn − b ≤ for all n ≥ N2 .
it is not too hard to see from the “graph” of the function x that the following statement is true: for each number ε > 0 there exists N such that √ √ n ≥ N implies  n + 1 − n ≤ ε. n → ∞. m ≤ 2 provided m ≥ N . Loosely speaking. Since an → a. that is if the sequence (an ) is Cauchy. if we do not know the actual limit? Deﬁnition 4. we compute am − an  = 2m − 1 2n − 1 − m+3 n+3 7(m − n) = (m + 3)(n + 3) ≤ 7m 7n + (m + 3)(n + 3) (m + 3)(n + 3) 7 7 m n ≤ + since ≤ 1 and ≤1 n+3 m+3 m+3 n+3 7 7 ≤ + .5. as it involves the limit a. If the condition in the above deﬁnition is true. . In order to ﬁnd N as in the deﬁnition. A sequence (an ) is a Cauchy sequence if for each number ε > 0 there exists an integer N such that am − an  ≤ ε whenever m ≥ N and n ≥ N.18.19. CAUCHY SEQUENCES 13 It follows from (4.1 does not help. and it is due to Cauchy. n ≥ N. What I mean is that for each ε > 0 there is an N such that any two members of the sequence from the N th onwards are within ε of each another. But there is indeed such a criterion. Solution. 2n−1 n+3 Example 4. Let ε > 0 be given. we sometimes write: am − an  → 0 as m. It is important to realise that the deﬁnition requires more than just that consecutive members of the sequence from the N th onwards are within ε of each other. Is there a method (or criterion) for telling if a sequence (an ) converges. We deduce that a bounded monotone sequence is convergent. The previous sentence is vague and ambiguous. √ For example. This gives a criterion for convergence which does not require knowledge of the limit. am − an  ≤ ε Hence the sequence is Cauchy.15) that if n ≥ N then 1 1 − ≤ ε. n m 7 ε Next we note that n ≤ 2 provided n ≥ 7 ε Similarly. Cauchy sequences A sequence converges iﬀ it is Cauchy. the Cauchy criterion says that if we go far enough out in the sequence then we can make the members of the sequence as close to each other as we like. 4. ε i. Show that the sequence an = is Cauchy. bn b Since ε > 0 was arbitrary. it follows that 1/bn → 1/b.e. More precisely: Definition 4. and you should never ever say or write anything like that in a mathematical argument.5.4. whenever m. it now follows from the result for products that an /bn → a/b. provided n ≥ N where N = 14 ε + 1. It follows that 14 .
Let ε > 0 be given.16). that every Cauchy sequence is convergent. if a sequence of rational numbers is Cauchy. am − a ≤ ε/2 whenever m ≥ N ). (Of course. really does have something to converge to. it will converge to a real number — this is just what the axiom says — but the limit may be irrational.1. there are no “gaps” in the real numbers. Thus we have shown the required statement is false! (In fact. It turns out that the other direction of (4. Theorem 4. and so is very small when x is very large. On the other hand. If a sequence converges then it is Cauchy.) . That is. is easy to prove. then it is not necessarily true that the sequence will converge to a rational number. 6 It is very important to realise that the statement is false means there is some (“bad”) ε > 0 for which there is no N such that √ √ m. and so is indeed convergent in the sense of Deﬁnition 4. informally. Proof.) n ≤ ε for all suﬃciently large n. The analogous statement is not true for the rational numbers. in this case any ε > 0 will be “bad”. But ε = 1 is bad since there is certainly no N such that √ √ m. 2 2 This proves that (an ) is Cauchy. That is. In order to ﬁnd N as in the deﬁnition of a Cauchy sequence. One direction of (4. it is clear that the next statement is false: for each number ε > 0 there exists N such that √ √ m. If a sequence is Cauchy then it is convergent. Axiom (Cauchy Completeness Axiom). CAUCHY SEQUENCES 14 √ (The main point is that the slope of the graph is 1/ 2 x . the fact that a convergent sequence is Cauchy. does not follow from the axioms so far. n ≥ N implies  m − n ≤ ε. and so must be assumed as an extra (and ﬁnal!) axiom. n ≥ N implies  m − n ≤ 1. Hence ε ε am − an  ≤ + = ε whenever m. we write (4. n ≥ N implies  m − n ≤ ε.) The algebraic proof is in footnote5 . i. if n ≥ ε2 .17) am − an  = (am − a) + (a − an ) ≤ am − a + an − a. that any sequence which is “trying to converge”. The major fact in this section is: (4.5. we compute √ √ √ √ n+1+ n √ n+1− n= √ n+1− n √ n+1+ n (n + 1) − n = √ √ n+1+ n 1 = √ √ n+1+ n 1 ≤ √ .16). Since an → a there exists an integer N such that an − a ≤ ε/2 whenever n ≥ N (and what is the same thing.16) a sequence converges iﬀ it is Cauchy. The Cauchy Completeness Axiom says. n √ √ 1 1 It follows that n + 1 − n ≤ ε if √n ≤ ε. n ≥ N.e.20.4. The algebraic proof is in footnote6 . Suppose an → a. 5 To show that √ n+1− √ √ Thus we have to ﬁnd just one “bad” ε.
an ∈ I2 for all n ≥ N2 (say). Divide I into two closed intervals of equal length. this implies there are no members of the sequence beyond the right endpoint of I1 . go back and reread the ﬁrst four paragraphs in Section 4. suppose an ≤ M for all n. 2.21. Let I be the interval [a1 .) In particular. The analogues of all these other axioms hold for the rational numbers. . 2. and then we use the Cauchy Completeness Axiom to show that the sequence is convergent. .4142. To motivate the argument look at the following diagram: a1 I1 . .) . (A similar proof will apply if it is decreasing. Thus an ∈ I1 for all n ≥ N1 (say). . . 2/3. (In the diagram. In particular. . and 1/2. since the sequence is monotone increasing. But the analogue of the Cauchy Completeness Axiom does not hold for the rational numbers. . and so is Cauchy by Theorem 4. M ]. 3. . We next show that if a sequence is bounded and increasing (or decreasing) then it is convergent. . In either case. 1. all members of the sequence after some term are in one of the two subintervals.414.. A sequence (an )n≥1 is monotone increasing if an ≤ an+1 for all n ≥ 1. or at least one member of the sequence is in the right interval (and in the second case. 3. 4/5. order and Archimedean axioms.3. .. Theorem 4. 1. One example of such a sequence is obtained √ from the decimal expansion of 2 = 1. . a3 a5 a6 a8 I3 I5 I2 I4 M . N2 = 4 and I2 is the right interval. Either all members of the sequence are in the left interval. then it is convergent.5 that 2 cannot be rational). It is is monotone decreasing if an ≥ an+1 for all n ≥ 1.4. Archimedean and Cauchy Completeness axioms. Suppose the sequence (an ) is monotone increasing and bounded.4.. . .22. thus we can take the sequence (4. We have now come to the end of the list of axioms.e. all later members of the sequence are also in the right interval).20.) Throughout the rest of the course. This sequence of rational numbers converges. Choose this subinterval and denote it by I1 .5.. take √ sequence (an ) of rational numbers which converges to 2 (we any saw in Section 2. we will rarely indicate when we are using the algebraic.41421. Thus the analogue of the Cauchy Completeness Axiom is not true in the rational numbers. are both monotone increasing. 1.414213562 . 1. Thus the sequences 1. (To compare the approach here with that in the 1998 Calculus Notes. order or Archimedean axioms.41. If a sequence is monotone increasing (or decreasing) and bounded. . Definition 4.18) 1. we now see that the Cauchy Completeness Axiom cannot follow from the other algebraic. They are the algebraic. i. (In the diagram. We do this by proving that such a sequence is Cauchy. 1. . . ) Similarly divide I1 into two closed intervals of equal length and choose I2 to be that subinterval which eventually contains all members of the sequence. N1 = 1 and I1 is the left interval. . First we give a deﬁnition. 3/4. But it does not converge to a rational number. There is not enough information just in the diagram to tell us this fact. but we will usually remark if we are using the Cauchy Completeness axiom. order. Proof. 1. CAUCHY SEQUENCES 15 √ For example.
7 The 8 . In other words. n ≥ Nk .e. . which is negative for x > e. For example. it follows that there is an integer k such that the length of Ik is ≤ ε 8 .4. b] is deﬁned to be b − a. Suppose (an ) is a sequence of real numbers. instead it converges to the irrational number 2. 1 1 if m.22 is not true for the rational numbers. a6 . an3 . Hence (an ) converges by the Cauchy Completeness Axiom. f (x) = − 1 1 ln x + 2 x2 x e x ln x . a101 . √ It is clear that the sequence is bounded below by 1. . (Of course it will converge to a real number — this is just what the theorem says. As remarked before. in order to have interesting examples we will occasionally use material that has not been rigorously established. 1 ≤ n n for all n. i. a4 . In order to show that (an ) is Cauchy. . even if it is bounded.23. Moreover. Since the length7 of Ik is (M − a)/2k . √ In one of the assignment problems we will see that n n → 1. i. 4. Subsequences and the BolzanoWeierstrass Theorem A sequence need not of course converge.18) is an example of a bounded increasing sequence of rational numbers which√ does not converge to a rational number.6. a n ∈ Ik and so am − an  ≤ ε if m. n). We have (M − a)/2k−1 ≤ (M − a)/k. So for x > 0.6. . (In the diagram. the sequence must converge. By the previous theorem. N3 = 7 and I3 is the right interval. 9 Recall that if x > 0 then xa = eln xa = ea ln x . a31 . Solution. This actually requires the Archimedean Axiom.) The sequence in (4. the following are subsequences: a2 . an2 . √ It follows that the sequence n n is decreasing if n ≥ 3. A subsequence is just a sequence obtained by skipping terms.) etc.e. But a m . a bounded increasing (or decreasing) sequence of rational numbers need not converge to a rational number.e. n ≥ Nk . every bounded sequence has a convergent subsequence. . a44 . SUBSEQUENCES AND THE BOLZANOWEIERSTRASS THEOREM 16 Similarly divide I2 into two closed intervals of equal length and choose I3 to be that subinterval which eventually contains all members of the sequence.18 to equal Nk ). The analogue of Theorem 4. a1 . let ε > 0 be given. a27 . . an ∈ I3 for all n ≥ N3 (say). Show the sequence n n is monotone decreasing and bounded below. . But by the BolzanoWeierstrass theorem. length of the interval [a. √ Example 4. Then f (n) = n1/n (i. . Now use the Archimedean Axiom to choose k so that k ≥ (M − a)/ε. It follows that the sequence (an ) is Cauchy (take N in Deﬁnition 4. and hence is convergent. . We usually write a subsequence of (an ) as an1 . We do that here. . . . an4 . let9 √ n f (x) = x1/x = e x ln x .
n4 = 44. but with the condition n3 < n4 . n3 = 31. 1. . . we have n1 = 2. call this interval I3 . 1. At least one of these intervals contains an inﬁnite number of terms from (an ). . . Choose one such term and denote it by an1 (n1 = 1 will do). We will construct a convergent subsequence an1 .) etc. . . . Choose one such term from (an ) and denote it by an2 . n2 = 27. .6.24 (BolzanoWeierstrass Theorem). an6 . . 1. . SUBSEQUENCES AND THE BOLZANOWEIERSTRASS THEOREM 17 or just as (ank ). 1. −1. the sequences 1. Choose one such term from (an ) and denote it by an3 .) Divide I3 into two closed intervals of equal length (and having only the midpoint of I3 in common). 1. . The sequence an1 . an3 . an2 . . 1. 10 For example. (In the diagram. 1. −1. 2 3 4 5 6 7 (4. 1. . not to scale. n1 = 1. 1. we could take n2 = 2. n2 = 4. 1. Let (an ) be bounded.) Divide I2 into two closed intervals of equal length (and having only the midpoint of I2 in common). corresponds to the sequence (4. −1 + . . a and a are always considered to be diﬀerent terms. . . n5 = 101. . . a7 a5 a3 a1 Divide I1 into two closed intervals of equal length (and having only the midpoint of I1 in common). are two diﬀerent subsequences of the ﬁrst sequence which converge to 1. Proof. Since the sequence (an ) is bounded. there is a closed bounded interval I1 which contains all terms from the sequence. even if they have the 3 4 same value. call this interval I4 . 1 + . 1. respectively. . The following theorem turns out to have many important applications. . −1. . 1. (In the diagram. . 1.19). . 1. Moreover. . an5 . . For example. For example. an3 . . −1. . . there are subsequences converging to −1. . Theorem 4.. 1. Thus in the above two examples. . .) 1 I1 I3 I5 . . Choose one such term from (an ) and denote it by an4 . At least one of these intervals contains an inﬁnite number of diﬀerent terms from (an ).. (In the diagram. 1. There are also subsequences of the second sequence converging to 1 and subsequences converging to −1. 1. we could take n3 = 4.19) do not converge. 1. . If a sequence is bounded then it has a convergent subsequence. −1. . 1. (The following diagram. 1. A bounded sequence certainly need not be convergent. Similarly. we could take n4 = 8. an4 . 1. 1 1 1 1 1 1 1 + . is a subsequence of (an ). . −1 + . −1 + . call this interval I2 . it is Cauchy. but with the condition n2 < n3 . 1.4. an4 . I4 a8 a6 a4 a2 1 I2 . . But there are subsequences of each sequence which converge to 1. 1. . −1. 1 + . −1. . 1. At least one of these intervals contains an inﬁnite number of diﬀerent terms10 from (an ). an2 . but with the condition n1 < n2 . . n3 = 6. In fact there are inﬁnitely many diﬀerent subsequences converging to 1.
·. It follows that N −1 anp − anq  ≤ ε whenever p. This proves the theorem.g. ·03.26. 200. etc. ·707106. . anN +2 . anN +3 . ·0003. SUBSEQUENCES AND THE BOLZANOWEIERSTRASS THEOREM 18 To see this. . then all of length four. has a subsequence converging to x. xm it is easy to see that there is a sequence which has a subsequence converging to x1 . I1  . Now let ε > 0 be a given positive number. . ·9. . e. we see that for any number x there is a subsequence √ converging to x. ·707. . ·0000. ·1. · · · ∈ IN . This contradicts the fact that xk ≤ b for all k. and so in particular are > b. it is suﬃcient to use the Archimedean Axiom to ﬁnd a natural number N ≥ I1 /ε. . . By using decimal expansions. . for example by induction. the subsequence converging to 2/2 = ·7071067810 . . but this does not matter. . is ·7. First enumerate all decimal expansions of length one. etc. For example. . ·20. . . then all of length three. Example 4. (The sequence cannot be bounded. anN +1 . Note that various values are repeated. ·7071. since it must contain arbitrarily large real numbers. ·000. 2N −1 Choose This means the sequence (ank ) is Cauchy. denote the subsequence by (xk ). ·001. . . Eventually all members of the sequence are within ε of c. . a  b  c  Let ε = (c − b)/2. . . then so does the limit of any subsequence. It is possible to change this construction a little and obtain a sequence which for every real number x has a subsequence converging to x. . . b]. . ·0002. ·00. Remark 4. . . The sequence can be constructed as follows. q ≥ N.6. We have seen examples of bounded sequences which have subsequences converging to 1 and subsequences converging to −1. . 1]. . Thus the sequences is ·0. Given m distinct real numbers x1 . . . The length of IN is IN  = N so 2I1  ≤ ε 11 .) This needs the Archimedean Axiom.4. . . ·01. ·3. What is more surprising is that there is a sequence. then all of length two. another converging to x2 . ﬁrst note that for each N . It is equivalent to choosing N such that 2N −1 ≥ I1 /ε. Hence it converges. then c ≥ a. To simplify notation. If all members of the original sequence (an ) belong to a closed bounded interval [a. ·70710. . . But since 2N −1 ≥ N . ·2. or any number less than the distance from b to c. ·70. . 11 . ·99.. anN . ·02. which for each real number x ∈ [0. . ·0001. ·2. . ·0001. . Proof. ·002. ·9999. but c > b (see the next diagram). ·0000. ·999. and another converging to xm .25. Similarly if xk ≥ a for all k and xk → c. Suppose in order to obtain a contradiction that xk → c and xk ≤ b for all k. . ·003. .
In other words. 19 . is the set of numbers x such that f (x) is deﬁned. The interval (a.2. We say f is continuous (on its domain) if f is continuous at every point in its domain.2 Example 5. 2 graph of f ° 1 1 I emphasise that ∞ is not a numbr. We will often not write xn ∈ D(f ). although this is always understood in order that f (xn ) be deﬁned.) Recall that the domain of a function f . Thus in order to show f is continuous at c. Suppose f (x) = x 1 2 x 2 0≤x<1 1≤x≤ 3 2 The domain of f is [0. Deﬁnition and examples The notion of continuity of a function is deﬁned in terms of sequences. (a. ∞) is just the set of real numbers strictly greater than a. and that for us the symbol ∞ has no neaning by itself. ∞)1 . there will in fact be many such “bad” sequences — we can always change the ﬁrst million or so terms — but the point is that to show continuity fails it is suﬃcient to just prove there is one “bad” sequence. (a. 2 If there is one. The informal idea of continuity of a function f at a point c is that “as x appraoches c then f (x) approaches f (c)”. denoted by D(f ). We will usually be interested in functions whoxe domains are intervals [a. we have f (xn ) → f (c).1.CHAPTER 5 Continuous Functions 5. (We also show that between any two numbers there are an inﬁnite number of rationals and an inﬁnite number of irrationals. etc. Definition 5. b]. Some examples of continuous and discontinuous functions are given. But it is possible for the domain to be a more complicated set of real numbers. b). More accurately. we have the following natural deﬁnition. In order to show f is not continuous at c. A function f is continuous at a point c ∈ D(f ) if for every sequence (xn ) such that xn ∈ D(f ) and xn → c. We will deﬁne the notion of continuity in terms of convergence of sequences. we have to show that for every sequence xn → c one has f (xn ) → f (c). The following is an attempt to sketch the graph of f .1. 3 ]. we only have to show there is one (“bad”) sequence xn → c with f (xn ) → f (c). xn ∈ D(f ) and xn → c ⇒ f (xn ) → f (c).
. Then x2 → c2 . and deﬁne g(x) = x 1 2 x 2 0≤x<1 1 < x ≤ 3. Here is a surprising example. . a3 x3 → a3 c3 . n n n n n will lie between a and b.. For example.3.. m n and b. suppose xn → c. 2 On the other hand. n n It follows that a1 xn → a1 c. since xn − 0 ≤ ε iﬀ  xn  − 0  ≤ ε). 2 n 2 In particular. by the theorem about n n multiplying a convergent sequence by a constant. f (xn ) → f (0). n 2 n 2 2 n and so lim f (xn ) = lim 1 x2 = 1 c2 . x3 → c3 .3 3 To see this. Let f (x) = a0 + a1 x + a2 x2 + · · · + ak xk . 2 then the domain of g is [0. etc. The function g is continuous at each c ∈ D(g). To show that f is continuous at some point c. etc. a0 + a1 x + a2 x2 + · · · + ak xk → a0 + a1 c + a2 c2 + · · · + ak ck by repeated applications of the theorem about sums of convergent sequences (a0 is here regarded as a constant sequence). The absolute value function f (given by f (x) = x) is continuous. Then xn  → 0 (this is immediate from the deﬁnition of convergence. Choose an integer n such that Archimedean axiom!).. Proof. Let f (x) = x x x rational −x irrational. ﬁrst suppose 0 < a < b. n It is clear that f is not continuous at 1. Example 5. Example 5.e. 1) ∪ (1. there is in fact an inﬁnite .5. . Then take m = k + 1. To see this. Finally.. and so f (xn ) = 1 x2 for all suﬃciently large n. by the theorem about products of convergent sequences. which in turn requires a slightly more careful deﬁnition of the natural numbers.1.5. From properties of sequences if 2 n xn → c then x2 → c2 and so 1 x2 → 1 c2 . since between any two real numbers there is both a rational and an irrational number (in fact an inﬁnite number of each). and 2 so is continuous on its domain. suppose xn → 0. For this. ﬁrst suppose xn → c and 1 < c ≤ 3 . We ﬁrst show continuity at 0. To prove continuity at c = 0 is similar to the previous example. f is not continuous on its domain. The following diagram is misleading. Then xn ≥ 1 for all suﬃciently 2 large n. Proposition 5. . DEFINITION AND EXAMPLES 20 1 . and easier. a2 x2 → a2 c2 . . Then at least one member m of the sequence n 1 n < b − a (from the 1 2 3 4 5 . This is very plausible from the diagram. take the sequence xn = 1 − 1 Then xn → 1 but f (xn ) (= 1 − n ) → f (1) since f (1) = 1 . etc. But f (xn ) = 1 x2 for all suﬃciently large n. To show it rigorously requires k showing there is a largest k such that n < a. The following result is established directly from the properties of convergent sequences.. The case 0 ≤ c < 1 is similar. if c = 1 and c ∈ D(f ) then xn → c ⇒ f (xn ) → f (c).. Every polynomial function is continuous. i. 0 1/n 2/n 3/n a m/n > 1/n b Since we can then obtain another rational between number of rationals between a and b. 3 ]. If we vary this example a little.4. etc.
That is.. Then xn  → 0 (this follows from the deﬁnition of a limit). by continuity of g at c and the deﬁnition of convergence of a sequence. (f + g)(xn ) → (f + g)(c). It follows f (xn ) + g(xn ) → f (c) + g(c) by Theorem 4. f is not continuous at c if c = 0. it follows from the Squeeze theorem that f (xn ) → 0. αf is continuous on D(f ) (α any real number) 4.12 about sums of convergent sequences.. Meanwhile. Finally. It follows that f + g is continuous at c. f + g is continuous on D. 3. 4 If A and B are sets.4 Then 1. f g is continuous on D. .6. Properties of continuous functions The basic properties of continuous functions follow easily from the analogous properties of sequences. If a < 0. suppose xn → 0. and show they are continuous. g(x ) ∈ n [ 1 g(c). 3 g(c)] for all suﬃciently large n and so it is positive. then g(xn ) = 0 for all suﬃciently large n5 . Suppose c ∈ D. choosing n so √ 2 n . Proof. Let (xn ) be any sequence such that xn → c (and as usual. For example if c is irrational then we can choose a sequence of rationals xn such that xn → c (by repeated applications of the remark above in italics). f /g is continuous at any point c ∈ D such that g(c) = 0.2. then their intersection A ∩ B is the set of numbers in both A and B. Their union A ∪ B is the set of numbers in at least one of A and B. It follows that f (xn ) = xn → c = f (c). f (xn ) → f (0). xn ∈ D). a similar argument works with the sequence 2 3 4 5 1 − . The argument in case g(c) < 0 is 2 2 similar.. 2.− . 5 If g(c) > 0. To see this. The proof in the other cases is similar.− .. We will later deﬁne the exponential. . since f and g are continuous at c. ..2. Then f (xn ) → f (c) and g(xn ) → g(c)..5. Just note for the case f /g that if xn → c and g(c) = 0.− . and trigonometric functions. On the other hand. Since −xn  ≤ f (xn ) ≤ xn . PROPERTIES OF CONTINUOUS FUNCTIONS 21 The function f is continuous at 0.e. Similarly if c is irrational. Let f and g be continuous functions and let D = D(f ) ∩ D(g). . Theorem 5. 5.− . we will use them in examples (but not in the development of the theory). i. logarithm. n n n n n < b − a and applying a similar argument to the sequence √ √ √ √ √ 2 2 2 3 2 4 2 5 2 . n n n n n gives the result for irrational numbers.
The domain D of f ◦ g is the set of numbers x such that both x ∈ D(g) and g(x) ∈ D(f ). It then follows that f (g(xn )) → f (g(c)) since f is continuous at g(c) (note that g(xn ) ∈ D(f )). there is an interesting diﬀerence between f1 and f2 . In other words. . Example 5. take any sequence xn → x. It follows that g2 (xn ) → g2 (x) since f2 (xn ) → f (x) by the continuity of f . the function 1/x. or the 1998 Calculus Notes page 16. will be continuous on their domain. (f ◦ g)(xn ) → (f ◦ g)(c). If we said the function sin y. D(f1 ) = { x  x = 0 }. i. Then f ◦ g is continuous.8. 6 The notation may seem a bit confusing. For all suﬃciently large n. Moreover. It follows that g(xn ) → g(c) since g is continuous at c. The domain of f1 is the set of real numbers x such that x = 0. take any sequence xn → 0. which is the same domain as for f1 .e. or just sin. The function f1 (x) = sin 1 x is the composition of the two continuous functions sin(x) and 1/x 6 and so is continuous. Similarly. That is. Suppose c ∈ D.7. is meant the function which assigns to each real number x (say) the real number sin x. In the case of the latter we can deﬁne a new function g2 by g2 (x) = x sin 0 1 x x=0 x = 0. f2 (x) = x sin However. You may ask “is it the same x in both cases”? But this is not the right way to look at it. or “the reciprocal function”. It follows from our results so far that rational functions (quotients of polynomials) and in general functions deﬁned from other continuous functions by means of algebraic operations and composition. all mean the same thing. the function 1 x is continuous on its domain. To show continuity of g2 at x = 0. Similarly. PROPERTIES OF CONTINUOUS FUNCTIONS 22 The composition of two continuous functions is continuous. or 1/y. This means g2 is continuous at x if x = 0. g2 is continuous on its domain R.5. if x ∈ D(f2 ). Suppose f and g are continuous. and so f ◦ g is continuous at c. we would mean the same thing. Let xn → c and xn ∈ D. xn ∈ D(f2 ). Then D(g2 ) = R and g2 (x) = f2 (x) if x = 0. By the function sin x. To show continuity of g2 at x = 0.2. Then −xn  ≤ g2 (xn ) ≤ xn . Proof. (See Adams page 35 for a discussion about the composition of functions.) Theorem 5. and so g2 (xn ) = f2 (xn ).
we need the idea of the supremum of a set. There is always a sequence of elements from the set which converges to the supremum.10. In the ﬁrst three examples we also say that 2 is the supremum of the set. 1. 2. But we say that 2 is the supremum of the set. 2]. If we consider sets such as [1. for each ε > 0 there is a member of S in the interval [a − ε. x ≤ a for every x ∈ S. and elsewhere. 2 is the maximum element of [1. namely 2. The precise deﬁnitions are a follows: Definition 5.) We sometimes say that f2 has a removable singularity at 0. then a is also the supremum of S. 7A member of an element of a set means the same thing. which implies a1 = a2 . a]. Note that there can be at most one maximum member of S. If the supremum belongs to the set then it is just the maximum. Suppose S is a set of real numbers and a = sup S. 2) and B.9. if the supremum does not belong to the set then there is no maximum. . This is immediate.3. or the set B of irrrationals less than 2. We say a is the supremum (or sup) of S if 1. For if a1 and a2 were both maximum elements7 . and if a ∈ S then we can just take a as the member of S in [a − ε. SUPREMUM OF A SET 23 and so g2 (xn ) → 0 (= g2 (0)) by the Squeeze Theorem.5. 1 For each natural number n there is thus a member of S in [a − n . Similarly. Proof. . then it is the maximum member of S (this is immediate from the deﬁnition of “maximum member”). For the important results in the next section. a].) This means g2 is continuous at 0. 2). since property 1 is the same in each case. and that the singularity of f1 at 0 is not removable. Then xn → a. the set (1. . a]. a]. This is essentially because there is no number y such that f1 (xn ) → y for every sequence xn → 0 (with xn = 0. For each n. 2] or (1. On the other hand. Note that if S has a maximum member a. choose one such member and denote it by xn . In the case of f1 there is no way of extending the function to a continuous function g1 deﬁned on all of R. there can be at most one supremum of S. −1. (We need to be a bit careful since some of the xn may equal zero. then we would have from property 1 that a2 ≤ a1 and a1 ≤ a2 . 2] or the set A = {2. We say a is the maximum member of S if 1. (1. for every ε > 0 there is at least one member of S in the interval [a − ε. Next note by the deﬁnition that 2 is the supremum of these three sets as well as of the sets (1. 2] and A. x ≤ a for every x ∈ S. Note that if the supremum a of S exists and is also a member of S. 5. and we are ﬁnished.3. First note that by the previous deﬁnition. we see that each of these sets contains a largest (or maximum) member. a ∈ S. does not contain a largest member. We write a = max S. From property 2 of the deﬁnition of supremum. Let S be a set of real numbers. Then there exists a sequence (xn ) of elements of S such that xn → a. 2. Theorem 5. Supremum of a set A set which is bounded above need not have a maximum member. We write a = sup S. again by property 1. } of integers less than or equal to 2. . 0. but there will always be a supremum.
note that for every n we have an ≤ a∗ ≤ bn and an ≤ b∗ ≤ bn . suppose x ∈ S. meeting only at the midpoint of I2 . x ≤ c for all x ∈ S. Theorem 5. 8 Recall that a set S (of real numbers) is bounded above if there is a number K such that x ≤ K for all x ∈ S. we 1 can take xn = 2 − n for each of the sets [1. Although a set need not have a maximum member. (although it may or may not be a member of S). the following theorem shows that every set which is bounded above does have a supremum. · · · ≥ b3 ≥ b2 ≥ b1 . Similarly bn → b∗ . If the right interval contains at least one member of S. That is. But bn − an = (b − a)/2n and so a∗ = b∗ . even if it is bounded above8 . say. If the right interval contains at least one member of S. Otherwise. b] such that x ≤ b for every x ∈ S and such that there is at least one member of S in I. then we take I3 to be this interval. meeting only at the midpoint of I. then we take I2 to be this interval. Since S is bounded above we can ﬁnd an interval I = [a.  x εS c  bn  say. then we take I1 to be this interval. . 2]. . S I ° I1 I3 ° I4 I2 Divide I into two closed bounded intervals of equal length. In either case it follows that x ≤ b1 for every x ∈ S and there is at least one member of S in I1 .5. . That is I1 ⊃ I2 ⊃ I3 ⊃ · · · . SUPREMUM OF A SET 24 Example 5. 2] and (1. meeting only at the midpoint of I1 . If the right interval contains at least one member of S. Divide I1 into two closed bounded intervals of equal length. b1 ]. a1 ≤ a2 ≤ a3 ≤ · · · .11. We use a bisection argument. Going back to the ﬁve examples at the beginning of this section. b3 ]. Proof. b2 ]. and so a∗ − b∗  ≤ bn − an . Since (an ) is increasing and bounded it follows that an → a∗ . (if the right interval contains no members of S) we take I3 to be the left interval. 2] and (1. We now deﬁne c = a∗ = b∗ and show it is the supremum of S. In this way we construct a sequence of intervals (In ). Otherwise. (if the right interval contains no members of S) we take I2 to be the left interval.3. For the sets [1. (1. Let I2 = [a2 . 2] we could also take the constant sequence xn = 2. . as otherwise we would have a contradiction for suﬃciently large n (by the Archimedean Axiom). Then S has a supremum.12. Then x ≤ bn for every n. In either case it follows that x ≤ b3 for every x ∈ S and there is at least one member of S in I3 . To check property 1. and so x ≤ lim bn (see the following proposition). Let S be set of real numbers which is bounded above. Otherwise (if the right interval contains no members of S) we take I1 to be the left interval. Divide I2 into two closed bounded intervals of equal length. and this sequence also works for the set √ A. For the set B we could take xn = 2 − n2 . To see this. It follows that a∗ = b∗ . each containing the next. 2). Let I1 = [a1 . In either case it follows that x ≤ b2 for every x ∈ S and there is at least one member of S in I2 . Moreover. Let I3 = [a3 . Etc.
c] and hence in [c − ε. c]. Remark 5. Proposition 5.5. In the previous theorem we used the following easy result. cε  an  c   x εS bn  .4. − n < 0 for all n. then it has an inﬁmum. This contradicts xn ≤ c. Then lim xn ≤ c (lim xn ≥ c ). and prove that if a set of real numbers is bounded below. We say a function f deﬁned on a set E is bounded on E iﬀ there exists a real number M such that f (x) ≤ M for all x ∈ S. and takes a maximum and a minimum value. Let ε be any positive number less than x − c c  xε  x  x+ε  . which in turn uses the Cauchy Completeness Axiom. A similar proof applies for the “≤” result. Hence we must have x ≤ c. c].) . that x > c. The require the Bolzano Weierstrass Theorem in their proof.13. then it takes all values in between. or assert the existence of a point in the domain with a particular property. From before we have an ≤ c ≤ bn . we can choose n so an ∈ [c − ε.2 followed from the deﬁnitions in a relatively straightforward way. in order to gain a contradiction. Proof. f is of course not continuous. Then for all suﬁciently large n. Then f is bounded. Since an → c and an is increasing. let ε > 0 be given. xn ∈ [x − ε. The properties of continuous functions in Section 5. Proof. There is at least one member x ∈ S in In = [an . Suppose (xn ) is a convergent sequence and xn ≤ c (xn ≥ c) for all n. but lim n = 0.15. Suppose f is a continuous function deﬁned on a closed bounded interval. Theorem 5. Note that if xn < c for all n. b] such that f (x) > M . x + ε]. and every member of S is ≤ c by property 1. The diagram is just to give an indication of the argument used to obtain a contradiction. which is also useful elsewhere. 5. Let x = lim xn and suppose xn ≤ c for all n.4. we can 1 1 still only deduce in general that lim xn ≤ c. Suppose. bn ] by the way we constructed the In . The theorems in this section are global. then it does not necessarily follow lim xn < c. THREE BIG THEOREMS 25 To check property 2. We can also deﬁne the minimum member of a set and the inﬁmum of a set. in that they refer to properties of continuous functions over their entire domain. Three big theorems A continuous function deﬁned on a closed bounded interval is bounded above and below. b] but f is not bounded. This means that for each real number M there must be some x ∈ [a. Hence this member of S is in fact in [an . For example. This proves property 2. If a continuous function deﬁned on an interval takes two values. Suppose (in order to obtain a contradiction) that f is a continuous function deﬁned on an interval [a. They are local properties in that they essentially refer to properties of continuous function near a prescribed point (although the point may be an arbitrary one in the domain).14. (In the following diagram.
3 p 55 of the 1998 Calculus Notes 3. This means S = { f (x)  x ∈ [a.. b] }. f (xn ) > n. b]. 1] but is not bounded on (0. 1]. if f (x) = 1/x for x ∈ (0. For example. Suppose f is a continuous function deﬁned on the closed bounded interval [a. f (xnk ) → f (c). if f (x) = x for x ∈ [0. (Note that f (xn ) diverges to inﬁnity. b]. it is not bounded. The analogous result is not true for the rationals. Hence f is bounded. there are continuous functions which are nowhere diﬀerentiable. However. then f is bounded on [−1. it is possible to construct some pretty wild continuous functions. let f (x)√ = 1/(1 − x2 ) for 1 ≤ x ≤ 2. 2. We say f takes its minimum value at d ∈ E and d is a minimum point iﬀ f (d) ≤ f (x) for all x ∈ [a. b].4. b]. THREE BIG THEOREMS 26 f(xn) graph of f f(x4) f(x3) f(x 2) f(x 1) a x1  xn c x2 x3 x4 b In particular. This is continuous at every point other than x = 2 (and in particular is continuous at every rational point) and it takes rational values at rational points. . For example. Remark 5. For example. then f is continuous on (0. b] such that f (xn ) > n. The function f is bounded by the previous theorem. The corresponding result is not true for open intervals or unbounded intervals. and another way of expressing this is to say that the set S is bounded. 1. 1]. The result is not as obvious as might ﬁrst appear. See Section 5. Suppose f is a continuous function deﬁned on a closed bounded interval.5.17. By continuity of f . Also. ∞) then f is continuous on [0. ∞). Let S be the set of all values taken by f . Theorem 5.) By the Bolzano Weierstrass theorem (and the remark which follows it) there is a subsequence (xnk )k≥1 which converges to some c ∈ [a. x = 0 x = 0. A function can be bounded and not take a maximum or a minimum value. ∞) but is not bounded on [0. This explains why we need the Cauchy Completeness Theorem in the proof of the previous theorem (it was used to prove the Bolzano Weierstrass Theorem) — if we only required the other axioms in the proof then because these axioms are true for the rationals it would follow that the analogous result would also be true for the rationals. From Theorem 5. for each n choose xn ∈ [a. if f (x) = x 1 −1 ≤ x ≤ 1. We say a function f deﬁned on a set E takes its maximum value at c ∈ E and c is a maximum point iﬀ f (c) ≥ f (x) for all x ∈ [a. Then f takes a maximum value and a minimum value. say. A function can take its maximum or minimum value at more than one point (a constant function is a simple example). For example. Proof.16.12 it follows that S has a supremum α. 1] but does not take a minimum value. But this contradicts the fact that for each n.
b] such that f (c) = γ. i. in other words that α = f (c) for some c ∈ [a. (That is. b] such that f (x) is within ε of α. Proof. b]  f (x) ≤ γ }. that if a sequence converges then any subsequence converges to the same limit. b]. THREE BIG THEOREMS 27 We claim that α is a member of S. Theorem 5. b]. By the Bolzano Weierstrass Theorem there exists a subsequence (xnk ) of (xn ) which converges to some c ∈ [a. In order to prove there is some c ∈ [a.9 Because a sequence can have only one limit (Theorem 4. A is the set of all x ∈ [a. 9 It is clear.e.5. Just from the deﬁnition of sup S we can ﬁnd for each ε > 0 a number x ∈ [a. This proves the claim. While it may seem clear that there is a maximum point c. There exists a sequence (yn ) of elements of S such that yn → α (by Theorem 5.) Since A is a bounded set it follows that it has a supremum c. For each yn there exists (at least one) xn ∈ [a. Then for any γ between f (a) and f (b) there exists c ∈ [a. f (xnk ) → f (c). say. b] such that f (xn ) = yn . b]. This will imply that f (x) ≤ f (c) for all x ∈ [a. and not hard to prove. b]. In other words. On the other hand. b] such that f (x) ≤ γ. b] such that then f (c) = γ. particularly in the case of the diagram. The last result implies that if a continuous function deﬁned on an interval I (not necessarily closed or bounded) takes two particular values. ynk → f (c). Suppose f is continuous on [a. b] and so c is a maximum point. . The proof that there is a minimum point is similar. even if f is not continuous but is merely bounded. α yn y3 y2 y1 S a x2 graph of f x3 x1 c xn b Proof of claim. However. Remark 5. then it must take all values between. for any two points a.4. f(b) γ f(a) graph of f a A c b We want to show that f (c) = γ.18.15) it follows that f (c) = α. (ynk ) is a subsequence of (yn ) and so ynk → α. b ∈ I and any γ between f (a) and f (b) then f (c) = γ for some c ∈ [a. By continuity of f .10). Suppose f (a) < γ < f (b) (the case f (a) > γ > f (b) is similar). there are some rather wild continuous functions as noted before. let A = {x ∈ [a. to show there is some c such that one actually has f (c) = α requires the continuity of f and uses the Cauchy Completeness Axiom. and hence that c is a maximum point.19.
THREE BIG THEOREMS 28 There is a sequence xn ∈ A such that xn → c (Theorem 5.13). . which contradicts c = sup A) and so f (c) ≥ γ (Proposition 5. there is a sequence xn → c and c < xn < b.13 again). (So. .4. Because f (c) ≤ γ and f (c) ≥ γ it follows f (c) = γ. By continuity.5. f (xn ) → f (c). it follows f (c) ≤ γ (Proposition 5.10). c ∈ A. in particular.) Since c = b (because we now know that f (c) ≤ γ. Since f (xn ) ≤ γ for all n. while f (b) > γ). But f (xn ) > γ (since otherwise xn ∈ A.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.