Академический Документы
Профессиональный Документы
Культура Документы
1
0
0
1
0
1 f (b) f (a)
0f 0 (c) =
1
0
1
0
1 ba
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1 0
1
0
1
0
1
0
1 0
1
0
1 0
1
0
1 0
1
0
1
0
1
0
1 0
1
0
1 0
1
0
1
0
1
1111111111111111111111
0000000000000000000000
a c b
Vesna Perisic
MATH2039 ANALYSIS
PERISIC
MATH2039: Analysis 1
Introductory Remarks 2
List of Symbols 3
Greek letters 4
1. Revision of standard notation and terminology 5
2. Properties of real numbers 9
3. Bounded sets 15
4. Limit of Sequence 19
5. Completeness properties of the reals 35
6. Continuity 39
7. Differentiability 47
8. The Cauchy mean value theorem and lHopitals rule 51
9. Integration 55
10. Series 66
11. Sequences and series of functions 81
12. Power series 84
13. Taylor series 88
14. The exponential function 92
15. The -function 97
Appendix A: Solutions to selected exercises 100
Appendix B: Examination Papers 165
Appendix C: Learning Outcomes 170
Appendix D: Further reading 171
Index 172
Contents
1
2 PERISIC
Introductory Remarks
About the authorship of the booklet: My name on the cover of this booklet means
only that I am currently responsible for teaching MATH2039 Analysis and that
I have edited already existing text for this purpose. The lecture notes are writ-
ten by my predecessors, my colleagues J. Anderson, J.Brodzki, N.Wright and
B. Nucinkis. Special thanks go to them.
The material is organizes in sections. The sections are kept short and followed
by relevant exercises. The booklet starts with a list of the most common mathe-
matical symbols used in analysis including the letters from the Greek alphabet.
After stating the notation, it also gives a revision of standard terminology intro-
duced earlier in the first year modules. These empresses importance of correct
notation and terminology.
These lecture notes work best together with the lectures. Reading the notes in
advance will help you following and engaging with the lectures. What you learn
in the lectures will help you to understand the notes even better.
All the teaching material is only good as the people who learn from it. If there
are bits of these notes you like, or dont like, please tell me. In particular tell
me if you find mistakes. The notes are suppose to be mathematically correct.
Vesna Perisic
September 2017
MATH2039 ANALYSIS 3
List of Symbols
Quantifiers
for all
there exists
Sets
N natural numbers {1, 2, 3, . . . }
Z integers {. . . , 3, 2, 1, 0, 1, 2, 3, . . . }
Q rational numbers e.g. 1, 12 , 13 etc.
R real numbers
(rational numbers and irrational numbers e.g. = 3.14159 . . . , 2 = 1.41421 . . . )
C complex numbers a + ib where a, b are real and i = 1
the empty set {}
is an element of (is in e.g. 1 Z, 2 R)
Greek letters
H I K M
Eta Theta Iota Kappa Lambda Mu
N O P
o
Nu Xi Omicron Pi Rho Sigma
T Y X
or
Tau Upsilon Phi Chi Psi Omega
MATH2039 ANALYSIS 5
By a set we shall mean any collection of objects, which are called elements of
the set. A set with three elements a, b, c will be denoted {a, b, c}. If we want to
give it a name we can write
S = {a, b, c}
To say that a is an element of A, or which is the same, a belongs to A, we write
a A. To indicate that a is not an element of A we write a 6 A. Sets can be
described by providing a list of their elements:
S = {1, 2, 3}
which works best for finite sets. For infinite sets we may use a similar notation,
as in the case of the natural numbers
N = {1, 2, 3, 4, . . . }
where the dots are meant to suggest that the list continues indefinitely. Other
sets of numbers which we shall use frequently are:
Integers Z = {. . . , 3, 2, 1, 0, 1, 2, 3, . . . }
Rational numbers (or fractions), denoted by Q, contain all elements of
the form p/q, where p and q are integers
p
Q = { | p, q Z }.
q
You might want to ask yourself what this set really is.
Similarly we could take the union of the sets [0, 1/2], [0, 2/3], . . . , [0, n/(n +
1)], . . . . This is
[ n
n
0, = x R | n N for which x 0, .
n+1 n+1
nN
Again you might want to ask yourself what this set is. Note that defining these
sets did not involve limits or letting n tend to infinity, although answering
the questions posed above will involve limits.
A function f from a set A to a set B, denoted f : A B, is a rule that assigns
to each element a of the set A a unique element of the set B. The value of f on
an element a is denoted f (a), the set A is called the domain of f. The set of all
elements b of B that are the image of some element a in A under the function f
is called the range or the image of f and is denoted f (A). Note that the image
of f need not be the whole set B.
We say that a function f : A B is an injection if and only if any element of B
can be the image of at most one element of A. In other words, we require that
if f (x) = f (y) then this must imply that x = y. The function f is a surjection,
or onto if every element of B is the image of some element in A, i.e. when
the image of f is all of B: f (A) = B. Finally, f is a bijection if and only if
it is both an injection and surjection. In this case we say that f establishes a
one-to-one correspondence between the sets A and B.
When describing properties of sets, we shall need to use the quantifiers. These
useful symbols allow one to state that a property is to hold for all or for some
element of a given set. The symbol is read for all, and a typical example of
its usage is the statement
x R, x2 0
which should be read: For all real numbers x, x squared is greater than or equal
to zero.
When we want to say that there exists an element satisfying a certain property
we use the symbol , which is read: there exists, as in the statement:
x R, x 1 = 0.
This should be read: There exists a real number x such that x 1 equals
zero.
We can put logical statements together to obtain new ones. If p and q are
propositions, then the statement p or q, also denoted p q is true if and only
if p or q or both are true. The statement p and q, which is sometimes denoted
p q, is true if and only if both p and q are true. The implication p = q is
true in the following cases: p is true and q is true; p is false and q is true; p is
false and q is false. The implication is false when p is true and q is false. The
equivalence p q is true if and only if both p and q are true or both p and
q are false.
If p is a proposition in the above sense, then the negation p of p is a proposition
that is true if and only if p is false. We note the following relations:
(p q) p q;
(p q) p q;
(p = q) p q.
The converse of the implication p = q is the statement q = p. We note
also that an implication p = q is equivalent to the following contrapositive
statement q = p. For instance, the statement
9 divides n = 3 divides n
is equivalent to the statement
3 does not divide n = 9 does not divide n
The negation of the quantifier is the quantifier ; the negation of is as in
the following examples. The negation of the statement
x R x2 0
is the statement
x R x2 < 0
In other words, to form the negation of the first statement, we replace the
quantifier with the quantifier (note that the x R remains the same in
both cases) and replace the statement following the quantifier with its negation.
(The negation of x2 0 is x2 < 0.)
Now the negation of the statement
x R x2 + 1 = 0
is the statement:
x R x2 + 1 6= 0
We leave it to you to figure out which proposition in each pair is true and which
is false.
A word about proofs. While it is reasonably easy to establish if a simple
statement is true or false, more complicated propositions will require a proof,
which is simply a sequence of logical deductions starting from a proposition
whose truthfulness is established and which ends with the proposition that we
want to prove. A proof often does more than just establishing if a statement is
true, it frequently provides insight into its meaning and context. For instance,
we have said that rational numbers are fractions with integer numerator and
8 PERISIC
denominator. For sure you have heard that 2 is irrational, which means that it
is impossible to represent it as a fraction. But how do we know that? It is clearly
impossible to form allpossible fractions p/q, with p, q integer, and to show that
none of these equals 2. It is only by providing a proof of this proposition that
we begin to believe in the existence of irrational numbers.
MATH2039 ANALYSIS 9
First well show that for a F , we have (a) = a. What do we know about
(a)? The additive inverse axiom says that if we add this to (a) then we
get 0, thus
(a) + ((a)) = 0.
Let us add a to both sides to get a+((a)+((a))) = a. Now by associativity
of addition we can move the parentheses to get
(a + (a)) + ((a)) = a.
As (a) is the additive inverse to a we have a + (a) = 0, so we get
0 + ((a)) = a.
Finally as 0 is the additive identity we get
(a) = a.
Now well show that for a and b in F we have (a + b) = (a) + (b). Again,
what do we know? We have
a + (a) = 0 and b + (b) = 0
MATH2039 ANALYSIS 11
Exercise 2.5. Prove that each of the following statements holds in a field F ,
using only the axioms of a field.
(1) a (b) = (a) b = (a b) for all a, b F ;
(2) (1) a = a for all a F .
Example 2.6. Let p be a prime number, and consider the integers modulo
p, usually denoted Zp or Z/pZ. This is a field.
The integers modulo p means that we only allow numbers 0, 1, . . . p1 and then
we return to 0. For example if we add 1 to p 1 we get 0, and if we multiply
p 1 by 2 we get p 2 (youd expect 2p 2, but the extra p loops back to zero).
Here are the multiplication and addition tables for Z5 .
+ 0 1 2 3 4 0 1 2 3 4
0 0 1 2 3 4 0 0 0 0 0 0
1 1 2 3 4 0 1 0 1 2 3 4
2 2 3 4 0 1 2 0 2 4 1 3
3 3 4 0 1 2 3 0 3 1 4 2
4 4 0 1 2 3 4 0 4 3 2 1
Exercise 2.7. Let n 4 be an integer that is not prime. Show that the
integers modulo n, Zn , is not a field (which axiom, or axioms fail?).
Definition 2.8. An order on a set X is a relation <, called less than, on X
satisfying the following conditions:
(trichotomy) for all a, b in X, exactly one of the following holds
(i) a < b
(ii) a = b
(iii) b < a
(transitivity) for any three elements a, b, and c of X, if a < b and
b < c, then a < c.
12 PERISIC
Of course we use the notation a > b (a greater than b) to mean b < a, and
a b (and b a) to mean that either a < b or a = b.
An ordered field is a field F with an order < that is well behaved with respect
to the operations of addition and multiplication, namely
for any three elements a, b, and c of F , if a < b, then a + c < b + c;
for any three elements a, b, and c of F with c > 0, if a < b, then
a c < b c.
Example 2.9. The real numbers R and the rational numbers Q with the usual
notion of < are ordered fields.
Exercise 2.10. Let F be an ordered field. Prove that if a in F and a 6= 0 then
a a > 0. Hence deduce that 0 < 1 in an ordered field.
Example 2.11. The field Zp of integers modulo a prime p do not form an
ordered field.
In analysis we will make a lot of use of inequalities (i.e. ordering). We will now
discuss some standard inequalities involving absolute values.
The absolute value, |x|, of a real number x is defined by
x,
if 0 < x
|x| = 0, if 0 = x
x, if 0 > x
Exercise 2.14. Show that |x| is always greater than or equal to zero.
MATH2039 ANALYSIS 13
Let us note some consequences of the definition. First, |x| x |x| (in fact
x always equals either |x| or |x| (or both in the case of 0)). This implies in
particular that if |x| = 0 then 0 x 0, so x = 0.
The following are key properties of the absolute value and will be used fre-
quently.
Proposition 2.15. For F an ordered field and x, y in F we have
(1) |xy| = |x||y|;
(2) |x + y| |x| + |y| (The triangle inequality);
(3) ||x| |y|| |x y| (The reverse triangle inequality).
Finally, when one of the two, say x, is negative and the other one, y is positive,
then xy < 0 so
|xy| = (xy) = (x)y = |x||y|
(2) The following argument is very useful. We noted that x |x| and y |y|,
hence x + y |x| + |y|. We also have that x |x| and y |y| so rearranging
we have x |x| and y |y| adding these together we get (x+y) |x|+|y|.
Since either |x + y| = x + y or |x + y| = (x + y) and both of these are less than
or equal to |x| + |y| we have |x + y| |x| + |y| as claimed.
(3) This is a similar argument to 2. We have
|x| = |(x y) + y| |x y| + |y|
by 2. Hence rearranging we have |x| |y| |x y|. We can argue the same
way to get
|y| |x y| + |x|
which implies |y| |x| |x y|. Thus |x y| is greater than or equal to both
|x| |y| and |y| |x|. Since ||x| |y|| is equal to one of these two numbers we
conclude that
||x| |y|| |x y|
Proof. We shall only prove part (1). The other two parts are left as an exercise.
: Suppose, for a contradiction, that x < y + for all > 0 yet x > y. Set
0 = x y > 0. Then y + 0 = x and hence, by trichotomy it cannot be greater
than x. This contradicts the hypothesis for = 0 . Hence x y.
: Suppose x y. Take an arbitrary > 0. By the trichotomy, we have
either x < y or x = y and we consider both cases separately. If x < y, then
x + 0 < y + 0 < y + by the transitive and additive properties. If x = y, we
have x + 0 = y + 0 < y + by the additive property. Hence, in both cases, for
all > 0, we have x < y + .
Definition 2.17. A subset A of an ordered field F has maximum s if s A
and x A we have s x. A subset A of an ordered field F has minimum t
if t A and x A we have t x.
Proposition 2.18. If A is a finite nonempty subset of an ordered field F then
it has a minimum and a maximum.
Another key property of the real numbers and rational numbers is the following
fact called the Archimedean property: Given any x, y with x > 0 and y > 0
there exists n N such that nx > y. This says that x cannot be infinitesimally
small and y cannot be infinitely large.
Proposition 2.19. The rational numbers Q satisfy the Archimedean property.
3. Bounded sets
In words this says that s is an upper bound for A, and if u is any upper
bound for A, then s is smaller than or equal to u. The supremum is therefore
sometimes called the least upper bound of A, as it is the smallest of all
possible upper bounds for A. Any set of real numbers that is bounded above
has a supremum.
Definition 3.5. Let A be a set of real numbers. The infimum of A, denoted
inf(A), is a number t R satisfying two properties:
(1) x A t x;
(2) if w R is a number such that x A w x, then w t.
In words this says that t is a lower bound for A, and if w is any lower bound for
A, then t is greater than or equal to w. The infimum is therefore sometimes
called the greatest lower bound of A, as it is the largest of all possible
lower bounds for A. Any set of real numbers that is bounded below has an
infimum.
Example 3.6. Let S = [0, 1] We shall show that sup(S) = 1 : By the definition
of the interval, we have that 1 x for all x S, hence 1 is an upper bound.
Now we have to show that 1 is a least upper bound. Suppose u x for all
x S, i.e u is an upper bound of S. Since 1 S, it follows that u 1 as
required.
Remark 3.7. If a set S has an upper bound u, then it has infinitely many
upper bounds, as any number M > u is also an upper bound of S.
Exercise 3.8. Let S = [0, 1). Show that sup(S) = 1.
16 PERISIC
Lemma 3.9. If a set S has a supremum (infimum) then this supremum (infi-
mum) is unique.
Proof. We prove this statement for the supremum. Suppose s1 and s2 are
both suprema of S. Then, by the definition of supremum, they are both upper
bounds of S, and hence, by the second part of the definition applied to s1 and
s2 separately, we have s1 s2 and s2 s1 . Hence by trichotomy, we have
s1 = s2 .
We can use this approximation property to find the supremum s or the infimum
t of a set S when s
/ S or t
/ S as the following example illustrates. But first, let
us mention that we shall make one further assumption on the set of real numbers
R. We assume that every non-empty subset of R, which is bounded above, has
a supremum. This property is often referred to as the Completeness Axiom.
We shall use a different definition of completeness later on, but it will turn
out that the two are equivalent. We shall also see later that completeness for
suprema is equivalent to completeness for infima, i.e. a non-empty subset of R,
which is bounded below has an infimum.
Example 3.11. For the subset S = n1 | n N of R, determine whether S is
S is bounded below by 0 (since 0 < n1 for all n N), and so has an infimum.
Since 0 is a lower bound and since there are elements of S arbitrarily close to 0
(given > 0, we can find n so that n1 < by taking n to be any integer greater
than 1 ), the infimum inf(S) = 0. In this case, inf(S) 6 S.
MATH2039 ANALYSIS 17
The approximation property is also used to show the following property of sets
of integers, which, for example doesnt hold in Q as we shall see later.
Theorem 3.12. Let S Z be a set of integers with a supremum s. Then
s S.
(5) S = (1)n 1 + n1 | n N ;
Proof. We only show (1), the second part is left as an exercise. Since A B,
we have for all a A that a sup(B) and sup(B) is an upper bound of
A. Now, the second property for supremum applied to sup(A) implies that
sup(A) sup(B) as required.
Exercise 3.16. In this question, A and B are subsets of R. Show that each
of the following holds. (You can use that a subset, which is bounded above
(below) has a supremum (infimum).
(1) if A B 6= , then sup(A B) min(sup(A), sup(B));
(2) if A B 6= , then inf(A B) max(inf(A), inf(B));
(3) if t is an lower bound for A and if t A, then t = inf(A);
(4) if inf(A) exists, then inf(A) = sup{y | y is a lower bound of A};
(5) if sup(A) exists, then sup(A) = inf{y | y is a upper bound of A};
(6) sup(A) is unique if it exists;
(7) inf(A) is unique if it exists.
Exercise 3.17. In this question, A is a subset of R. Define A = {a| a A}.
Show that each of the following holds.
(1) if sup(A) exists, then inf(A ) exists and inf(A ) = sup(A);
(2) if inf(A) exists, then sup(A ) exists and sup(A ) = inf(A).
Exercise 3.18. For each of the following, either give an example of a subset S
of R satisfying the stated property, or prove that no such set exists.
(1) S has a rational lower bound and inf(S) is irrational;
(2) S has a rational lower bound and inf(S) is rational;
(3) S has an irrational lower bound and inf(S) is rational;
(4) S has an irrational lower bound and inf(S) is irrational.
MATH2039 ANALYSIS 19
4. Limit of Sequence
The indexing set of a sequence is usually, but not always, taken to be the
natural numbers N. The terms of a sequence are more commonly denoted
by a1 , a2 , a3 , . . . or something similar, rather than by f (1), f (2), f (3), . . . . A
sequence a1 , a2 , a3 , . . . can be defined by giving a formula for its n-th term
an .
Example 4.2. The functions
an = n, n N
and
1
bn = , n2
ln(n)
are two examples of sequences.
and
(ck )kN , ck = ak+1 = k + 1, k N
are examples of subsequences.
Definition 4.6. Let {an : n I} be a sequence in F and let L F . Then, the
sequence {an : n I} converges to L if, for every > 0, there exists n0 I
such that
|an L| < for every n I with n > n0 .
Symbolically, the sentence an converges to L is denoted either by an L as
n or by limn an = L. We also say L is the limit of the sequence
{an : n I}0 .
Example 4.7. We show that n1 0 as n . Let > 0 be arbitrary. Then
there exists an n0 N such that n0 > 1 . (The existence of n0 follows from the
Archimedean Property for R which we will show later, and the existence of a
well defined floor-function will be seen later as well.) Hence, for all n > n0 we
have n1 < n10 < . Since, for all n N, we have n1 > 0, we conclude that
1 1
| | = | 0| <
n n
for all n > n0 as required.
Example 4.8. To show that 2n+1 2n 1 as n we need to consider | 2n
2n+1
1 1
1| = |1 + 2n 1| = | 2n |. Let > 0 be arbitrary. As above, we can find an
1 1
n0 N such that n0 > 2 . Hence, for all n > n0 we have 2n < 2n1 0 < . Since,
1
for all n N, we have 2n > 0, we conclude that
2n + 1 1
| 1| = | | <
2n 2n
for all n > n0 as required.
Definition 4.9. A sequence that does not converge, diverges .
Example 4.10. The sequence {an = (1)n : n N} diverges. Suppose that
for some a R, (1)n a as n . Hence, there is an n0 N such that
for all n > n0 we have |(1)n a| < for all > 0. Now suppose = 1. For
n odd this gives | 1 a| = |1 + a| < 1 and for n even this gives |1 a| < 1.
Furthermore, by the triangle inequality,
2 = |1 + 1| = |(1 a) + (1 + a)| |1 a| + |1 + a| < 1 + 1 = 2
implying 2 < 2, a contradiction.
MATH2039 ANALYSIS 21
and therefore
|an | max{|a1 |, |a2 |, . . . , |aN 1 |, 1 + |L|}, for all n N.
Here, clearly R = max{|a1 |, |a2 |, . . . , |aN 1 |, 1 + |L|}.
Theorem 4.16. ( Arithmetic of sequences/limits) Let (an )nI be a se-
quence converging to a R, limn an = a, and let (bn )nI be a sequence
converging to b R, limn bn = b, a, b R. Then the following holds:
(1) sum: the sequence (an + bn )nI converges to a + b as n ;
(2) difference: the sequence (an bn )nI converges to a b as n ;
(3) product: the sequence (an bn )nI converges to ab as n ;
(4) reciprocal: if a 6= 0, then the sequence ( a1n )nI converges to 1
a as
n ;
(5) quotient: if b 6= 0, then the sequence ( abnn )nI converges to a
b as n
.
Proof.
(1) We need to show that for any > 0, there exists n0 I so that
|(an + bn ) (a + b)| < for n I, n > n0 . So, start by simplifying:
|(an + bn ) (a + b)| = |an a + bn b| |an a| + |bn b|.
Since limn an = a, there exists n1 so that |an a| < 21 for n > n1 .
Since limn bn = b, there exists n2 so that |bn b| < 12 for n > n2 .
So, for n > n0 = max{n1 , n} ), we have that both |an a| and |bn b|
are less than 12 , and so
1 1
|(an + bn ) (a + b)| |an a| + |bn b| < + = .
2 2
(2) We need to show that for any > 0, there exists n0 I so that
|(an bn ) (a b)| < for n I, n > n0 . So, start by simplifying:
|(an bn ) (a b)| = |an a bn + b| |an a| + | bn + b| = |an a| + |bn b|.
Since limn an = a, there exists n1 so that |an a| < 21 for n > n1 .
Since limn bn = b, there exists n2 so that |bn b| < 12 for n > n2 .
So, for n > n0 = max{n1 , n2 }, we have that both |an a| and |bn b|
are less than 12 , and so
1 1
|(an bn ) (a b)| |an a| + |bn b| < + = .
2 2
(3) We need to show that for any > 0, there exists n0 so that |an bn ab| <
for n > n0 . We follow the same general pattern as for the previous
two, though the algebra is a bit more intricate. As before, we begin by
simplifying |an bn a b|:
|an bn a b| = |an bn an b + an b a b| |an bn an b| + |an b a b|.
We take these two terms one at a time.
MATH2039 ANALYSIS 23
(4) we need to show that for any > 0, there exists n0 so that | a1n a1 | <
for n > n0 . So, as with all the others, we start by simplifying:
1 1 |a an |
an a = |a an | .
|a|2
Since limn an = a, we can choose n2 so that |an a| < 2 for
n > n2 . Then, for n > n0 = max{n1 , n2 }, we have that
2 |a|2
1 1 |a an |
= < 2 = ,
an a |a an | |a| 2
as desired.
24 PERISIC
(5) One way to do this would be to repeat the style of argument just given
for reciprocals, which would work but which is somewhat involved. An-
other approach, and the one we take here, is to note that since an con-
verges to a and since bn converges to b 6= 0, we have that b1n converges
to 1b , by what we just did with reciprocals, and hence abnn converges to
a
b , by taking products.
Theorem 4.17. (Squeeze rule) Let (an ), (bn ), and (cn ) be sequences. If
an bn cn for all n and if (an ) and (cn ) both converge with limn an =
limn cn = L, then (bn ) converges and limn bn = L.
Proof. We are given that an bn cn for all n and that an and cn both con-
verge with limn an = limn cn = L. We want to show that limn bn =
L, which is to say, for every > 0, there exists n0 so that |bn L| < for
n > n0 .
So, for > 0, choose n1 so that |an bn | < 21 for all n > n1 , and choose n2 so
that |an L| < 21 for n > n2 . Then, for n > n0 = max{n1 , n2 }), we have that
1 1
|bn L| = |bn an + an L| |bn an | + |an L| < + = ,
2 2
as desired.
We will now define the real numbers. As mentioned in the previous section, we
require one more property of the real numbers, completeness. In the previous
section we mentioned that one can define completeness in terms of suprema,
but we will use the following definition. We shall see later, Theorems 5.1 and
5.3, that the two definitions are equivalent.
Definition 4.19. The real numbers, denoted R, are an ordered field satisfy-
ing the following additional axiom.
We can use the following result to prove that in fact (1/n) converges to zero
(as we might expect it to!).
Lemma 4.22. Let (an ) be a sequence and c R. If (an ) converges to L as
n then the sequence (bn ), bn = can converges to cL as n .
Proof. We know from the above example that the sequence (an ) converges to
1
some limit L. Let bn = a2n = 2n . The sequence (bn ) is a subsequence of
(an ), so it also converges to L. On the other hand bn = 12 n1 = 12 an , so by
the previous lemma (bn ) converges to 12 L. Thus 12 L = L so L = 0. This uses
Lemma 4.11.
Proof. Since (an ) converges, write limn an = L, so that for every > 0 there
exists n0 so that |an L| < 21 for n > n0 . Consider the difference |ap aq | for
p, q > n0 :
1 1
|ap aq | = |ap L + L aq | |ap L| + |L aq | < + = ,
2 2
as desired.
Remark 4.26. The Cauchy criterion is a very useful test for divergence. If
a sequence converges then it is a Cauchy sequence. Thus if a sequence is not
a Cauchy sequence then it cannot converge i.e. it diverges.
If (an ) is a sequence, and there exists > 0 so that for every n0 , there are p,
q > n0 with |ap aq | , then the Cauchy criterion fails for (an ), hence (an )
diverges.
Example 4.27. Let an = (1)n . Then for every n0 we can pick p, q > n0 such
that |ap aq | = 2. Specifically, for any n0 if we take p = n0 + 1 and q = n0 + 2
then one of p, q is even, and the other is odd, hence ap aq = 2. Hence the
Cauchy criterion fails for = 2, so an diverges.
Example 4.28. Let an = n. Then for every n0 we can pick p, q > n0 such
that |ap aq | = 1. Specifically, for any n0 if we take p = n0 + 1 and q = n0 + 2
then |ap aq | = |p q| = 1. Hence the Cauchy criterion fails for = 1, so an
diverges.
This means that there are no infinitely large or infinitesimally small real num-
bers.
MATH2039 ANALYSIS 27
Proof. Suppose this is not the case, that is for some x, y > 0 we have nx y
for all n N. We define a sequence an = nx. Since x > 0 this is an increasing
sequence: an+1 = nx + x > nx = an . It is also a bounded sequence. It is
bounded below by 0, since x is positive, and by assumption it is bounded above
by y. By the completeness axiom the sequence therefore converges.
On the other hand for every n0 we can pick p, q > n0 such that |ap aq | = x.
Specifically, for any n0 if we take p = n0 + 1 and q = n0 + 2 then |ap aq | =
|px qx| = x. Hence the Cauchy criterion fails for = x, so an diverges, which
is a contradiction. We thus conclude that there are no numbers x, y > 0 such
that nx y for all n N.
A sequence can diverge in a number of different ways. There are two particularly
useful ways that this can happen. A sequence (an )nI diverges to infinity
if, for every R > 0 there exists n0 in I such that an > R for every n in I with
n > n0 .
Similarly, a divergent sequence (an )nI diverges to minus infinity if, for
every R > 0 there exists n0 in I such that an < R for every n in I with
n > n0 .
Proof. We must show that for all R R with R > 0 there exists n0 N such
that n > n0 implies n > R. Given R > 0 apply the Archimedean property to
the numbers 1, R. Then there exists n0 N such that n0 1 > R. Now for any
n > n0 we have an = n > n0 > R.
Definition 4.32. The integer part of a real number x is the greatest integer
n such that n x. This is denoted bxc, and is sometimes called the floor
function.
Proposition 4.33. The floor function is well defined, i.e. for every x R
there exists a greatest integer n such that n x.
28 PERISIC
So, suppose we can find a function f (x) so that an = f (n) and so that f 0 (x) 0.
Then, the sequence an is monotonically non-decreasing. (If f 0 (x) > 0, then an
is monotonically increasing.) Similarly, if f 0 (x) 0, then an is monotonically
non-increasing. (If f 0 (x) < 0, then an is monotonically decreasing.)
MATH2039 ANALYSIS 29
1 1
Exercise 4.38. Prove that n+1 < ln(n + 1) ln(n) < n for all n N.
Pn 1
Now, consider the sequence given by an = i=1 i ln(n). Prove that an
is a decreasing sequence and that each an is positive. Conclude that the limit
= limn an exists. [This number is known as Eulers constant, and
little is known about it. For instance, it is not known whether is rational or
irrational.]
Exercise 4.39. Explain exactly what is meant by the following statements:
(1) limn 32n1 = ;
(2) limn (1 2n) = ;
(3) limn en = 0;
Theorem 4.40. Inequalities of sequences: Let (an )nI be a sequence con-
verging to a and let (bn )nI be a sequence converging to b. If an bn for all n,
then a b.
Corollary 4.41. Let (an ) be a convergent sequence with an in the interval [a, b]
for all n I. Then limn an [a, b].
Remark 4.42. Standard tricks for evaluating limits: Some of these are
facts that we will prove later in the course but which are nonetheless very useful
to know now.
(1) exponents: for c > 0, if an converges to a, then can converges to ca ;
(2) trigonometric functions: if an converges to a, then cos an converges
to cos a, and sin an converges to sin a.
We shall now derive some basic results on limits of various useful sequences.
We begin with the following fact, due to Bernoulli.
Theorem 4.43. (The Bernoulli inequality) For all a 0 and all n N:
(1 + a)n 1 + na
30 PERISIC
We shall use Bernoullis inequality quite a lot in what follows. For now we shall
derive some simple facts about limits.
1
Example 4.44. If limn an = then limn an = 0.
Proof. Assume that limn an = . This means that for any > 0 there
exists an n0 such that for all n > n0 , an > 1/ > 0. But this implies that
1
0 < 1/an < , for all n > n0 , i.e. |1/an | < and this implies that lim = 0.
n an
n
The converse of this statement is false. Consider the sequence an = (1)
n .
Then lim an = 0 but the sequence 1/an = (1)n n does not diverge to either
n
or (though it does diverge).
Example 4.45. For all q > 0
lim nq =
n
Proof. Take any R > 0. Then for n > n0 = bR/qc we have n > R/q so nq > R
and the result follows.
Example 4.46. If a > 1, then lim an = .
n
1
Proof. First we let 0 < q < 1. Then 1/q > 1 and lim ( )n = and so
n q
lim q n = 0.
n
n
Example 4.48. For all a > 0, lim a = 1.
n
Proof. Assume first that a > 1. We shall show that the sequence un = n a is
decreasing. As a > 1, we have that an < an+1 . Now take the n(n + 1) root of
both sides of this inequality:
n(n+1)
n(n+1)
an < an+1
to get
n+1
a< na
hence un+1 < un . Now assume that inf{un : n N} = 1 + h, where h 0.
This is a reasonable assumption as un > 1 for all n. But then 1 + h is a lower
bound of the set {un } so
1 + h n a = (1 + h)n a
Using the Bernoulli inequality again we have that
1 + nh (1 + h)n < a n
If h > 0 then nh , which is impossible, as the left hand side has to be
smaller than a fixed number a. So we must have that h = 0 and so inf{un } = 1,
hence lim un = 1.
n
Now assume 0 < a 1, so 1/a 1. Then by the first part of this proof,
lim (1/a)1/n = 1, or in other words:
n
1
n
lim = 1 = lim a=1
n a1/n n
Example 4.49. If |q| < 1 then
1
lim (1 + q + q 2 + + q n ) =
n 1q
Proof. We have
1 + q + q 2 + + q n = q(1 + q + + q n1 ) + 1
so
(1 + q + q 2 + + q n1 )(1 q) = 1 q n
which gives
1 qn
1 + q + q 2 + + q n1 =
1q
n
Now use the fact that q 0 as n for |q| < 1 to get the result.
Example 4.50. For the two sequences given below, do the following:
Determine whether the sequence converges or diverges;
if the sequence converges, determine its limit;
if the sequence diverges, determine whether the sequence diverges to
, diverges to or neither.
32 PERISIC
(1) an = n+5 n;
8k
diverges: for nk = 8k, k N, ank =
sin 4 = sin(2k) = 0, and
(8k+2)
for nk = 8k + 2, k N, ank = sin 4 = sin 2k + 2 = 1.
Hence, |a8k a8k+2 | = 1 for all k N, and so the sequence (an ) fails
the Cauchy criterion, and hence (an ) diverges.
Exercise 4.51. The sequence scavenger hunt: for each of the following
sequences (an ), do the following:
Determine whether the sequence converges or diverges;
if the sequence converges, determine its limit;
if the sequence diverges, determine whether the sequence diverges to
, diverges to or neither.
(1) an = (n + 2)1/n ;
n2 +3n+2
(2) an = 6n3 +5 ;
(3) an = (1 + n1 )n ;
sin(n)
(4) an = 3n ;
(5) an = 2n + 3 n + 1;
(6) an = cos n
4 ;
(7) an = (1 + n1 )1/n ;
(8) an = ln(n);
(9) an = en ;
ln(n)
(10) an = ;
n
2 n
(11) an = 1 n2 ;
n3
(12) an = 10n2 +1 ;
1n2
(16) an = 2+3n2 ;
n3 n+7
(17) an = 2n3 +n2 ;
9 n
(18) an = 1 + 10 ;
n
(19) an = 2 12 ;
n
(20) an = 1 + (1) ;
1+(1)n
(21) an = n ;
1+(1)n n
(22) an = n ;
( 32 )
sin2 (n)
(23) an =
n
;
q
2+cos(n)
(24) an = n ;
(25) an = n sin(n);
(26) an = n cos(n);
(27) an = sin(n)/n ;
(28) an = 2cos(n) ;
ln(2n)
(29) an = ln(3n) ;
ln2 (n)
(30) an = n ;
1
(31) an = n sin n ;
arctan(n)
(32) an = n ;
n3
(33) an = en/10
;
2n +1
(34) an = en ;
sinh(n)
(35) an = cosh(n) ;
(38) an = (0.001)1/n ;
(39) an = 2(n+1)/n ;
3/n
(40) an = n2 ;
So, for n > n1 , we have that | |xn | 4| = | xn 4| = |xn + 4|, and we have
been given that for any > 0, there is n2 so that | |xn | 4| = | xn 4| =
|xn + 4| < for n > n2 . So, for any > 0, take n0 to be the larger of n1
(chosen so that xn < 0 for n > n1 ) and n2 (which comes from the definition
that limn xn = 4). Then, for n > n0 , we have that | |xn | 4| < , and we
are done.
Exercise 4.54. Prove that each of the following statements is true, using the
definition of limit.
p
(1) if xn 4 as n , then |xn | 2 as n ;
(2) if xn 4 as n , then x2n 16 as n ;
xn
(3) if xn 4 as n , then 3 34 as n ;
Exercise 4.55. Let (an ) be a sequence converging to a. Show that the following
holds:
(1) square roots: if a > 0, then an converges to a;
(2) |an | converges to |a|;
(3) If a 6= 0, then (1)n an diverges;
x1 ++xn
Exercise 4.56. Prove that if xn x as n , then n x as n .
MATH2039 ANALYSIS 35
Proof. We will prove the first statement. The second is proved similarly (or
by noting that A = {a|a A} is bounded above so has a supremum, and
inf A = sup(A)).
The proof uses a bisection argument. As A is nonempty, there exists some
a R which is not an upper bound for A (e.g. take a = x 1 for any x A).
As A is bounded above there exists an upper bound b for A. Clearly a < b.
Now let a0 = a and b0 = b. Let c0 = a0 +b 2 . Either c0 is an upper bound for A
0
Proof. Set a = limn an and apply the definition of limit of a sequence with
= 1. Then there exists n0 I so that |an a| < 1 for all n > n0 . In
particular, for n > n0 , we have that an lies in the interval (a 1, a + 1). The set
{a1 , . . . , an0 , a + 1} is finite, so it has a maximum s. We have an < a + 1 s
for all n > n0 while for n n0 , an lies in the set so an s. Thus {an : n I}
is bounded above by s.
Similarly, let t be the minimum of the finite set {a1 , . . . , an0 , a 1}, and note
that t an for all n, so that {an : n I} is bounded below by t.
Proof. Let (an )nI be a bounded monotone sequence. There are two cases:
either (an ) is monotone non-increasing (which includes the case of monotone
decreasing) or is monotone non-decreasing (which includes the case of monotone
increasing). Suppose to start that an is monotone non-decreasing. Let A =
{an |n I}, and let s = sup(A).
Since s = sup(A), we know that an s for all an A and also that for each
> 0, there exists some n0 so that |s an0 | = s an0 < . (Because, if there
is some 0 > 0 so that s an 0 for all n, then s 0 is an upper bound for
A which is smaller than s, contradicting the choice of s = sup(A).)
Proof. The proof uses a bisection argument. We will bisect the interval [a, b] in
such a way that each subinterval [ak , bk ] that we construct contains infinitely
many of the terms of the original sequence xn .
Let a0 = a and let b0 = b. For each k = 1, 2, . . . , we suppose that we have
defined ak1 , bk1 such that [ak1 , bk1 ] contains xn for infinitely many val-
ues of n. Then either [ak1 , ak1 +b 2
k1
] contains xn for infinitely many n, or
ak1 +bk1
[ 2 , bk1 ] contains xn for infinitely many n (or both). If [ak1 , ak1 +b
2
k1
]
ak1 +bk1
contains xn for infinitely many n then we define ak = ak1 , bk = 2 ,
otherwise we define ak = ak1 +b 2
k1
, bk = bk1 . Hence we have defined ak , bk
such that [ak , bk ] contains xn for infinitely many n.
Note that at each stage we have ak1 ak1 +b
2
k1
bk1 . We define ak = ak1
ak1 +bk1 ak1 +bk1
or ak = 2 so a k ak1 , and we define bk = 2 or bk = bk1
so bk bk1 . The sequences ak , bk are bounded (as they lie in [a, b]) and
monotonic, so they converge. Moreover, at each stage bk ak = bk1 a 2
k1
,
so their limits are equal: limk ak = limk bk = L for some L. Note that
L [a, b] by inequality of sequences.
Now for each k we can pick nk such that nk > nk1 and xnk [ak , bk ]. By
assumption there are infinitely many possible choices of nk , so to be specific,
let nk be the least integer satisfying these conditions. By construction ak
xnk bk , so as (ak ), (bk ) converge to L, so does (xnk ) by the squeeze rule. We
have thus constructed a subsequence (xnk ) of (xn ), such that (xnk ) converges
to a limit L in [a, b] as required.
Definition 5.6. A set S R is sequentially compact if for every sequence
(xn ), xn S, there is a subsequence (xnk ) which converges to a limit in S.
We know that if a sequence converges then it is Cauchy. We will now prove the
converse of this.
Theorem 5.8. (General principle of convergence) Suppose that (an )nI
is a Cauchy sequence in R. Then (an ) converges in R.
6. Continuity
Remark 6.2. Note that the definition of limxa f (x) = L includes the re-
quirement that 0 < |x a|, and so the limit does not depend on the value of
f (a). For example, limxo sinx x = 1 but sinx x is even not defined at x = 0.
Note that the left and right-hand limits simply restrict the domain of f to either
{x S : x < a} or {x S : x > a}.
The basic properties of limits of sequences also hold for limits of functions. The
proofs are simple modifications of the corresponding results for sequences.
Theorem 6.4. Let f, g, h : S R be functions with common domain S.
If limxa f (x) = L and limxa g(x) = M then
1 1 1
(iv) reciprocal: if L 6= 0, then limxa f (x) = limxa f (x) = L;
Note that this definition implies the following: The function f is defined at a,
the limit limxa f (x) exists, and limxa f (x) = f (a).
MATH2039 ANALYSIS 41
Since continuous functions are defined in terms of limits, the rules of arithmetic
for limits of functions, as given in Theorem 6.4, extend immediately to rules of
arithmetic for continuous functions.
Theorem 6.8. Let f, g : S R be functions that are continuous at a S.
Then the following holds:
(1) the sum f +g, defined by setting (f +g)(x) = f (x)+g(x), is continuous
at a;
(2) the difference f g, defined by setting (f g)(x) = f (x) g(x), is
continuous at a;
(3) the product f g, defined by setting (f g)(x) = f (x)g(x), is continuous
at a;
(4) if g(a) 6= 0, then the quotient f /g, defined by setting (f /g)(x) =
f (x)/g(x), is continuous at a.
Proposition 6.9. If f is continuous at a and if g is continuous at f (a), then
the composition (g f )(x) = g(f (x)) is continuous at a.
Proof. We need to show that for all > 0 there exists > 0 such that if
|x a| < , then |g(f (x)) g(f (a))| < . Since g is continuous at f (a), we
know that for each > 0, there exists > 0 so that if |z f (a)| < , then
|g(z) g(f (a))| < . [Here z is used as a variable so that there arent too many
xs running around.]
We also know that f is continuous at a, so that for each > 0, there exists
> 0 so that if |x a| < , then |f (x) f (a)| < . Take the that is output
by the definition of continuity of g at f (a) and input it into the definition of
continuity of f at a: since |f (x) f (a)| < , we can apply the definition of
continuity of g at f (a) to get that if |x a| < , then |f (x) f (a)| < , and so
|g(f (x)) g(f (a))| < as desired.
Proposition 6.10. Suppose that f is continuous and that the sequence an
converges to a. Then, the sequence f (an ) converges to f (a).
42 PERISIC
Proof. Since f is continuous at a, for every > 0, there exists some > 0 so
that if |x a| < , then |f (x) f (a)| < . Since an converges to a, for each
> 0, there exists some n0 so that if n > n0 , then |an a| < . So, suppose
we are given some > 0, and take = , where comes from our choice of
in the definition of continuity at a and where is the input in the definition of
an converging to a. Then, for n > n0 , we have that |an a| < = , and hence
that |f (an ) f (a)| < , which is precisely the definition of f (an ) converges to
f (a), as desired. [This proof should convince you, if you have not already been
convinced, of the power of appropriate definition.]
Exercise 6.11. Suppose that f is continuous and that the sequence c, f (c),
f (f (c)), f (f (f (c))), . . . converges to a. Prove that f (a) = a.
Definition 6.12. A function f : S R is uniformly continuous if for
each > 0, there exists > 0 so that for all a, x S, if |x a| < , then
|f (x) f (a)| < .
Note that this definition is very similar to the definition of continuity, except in
one aspect: in the definition of continuity, the value of depends on both and
on the point at which continuity is being checked, while for uniform continuity,
the value of depends only on and not on the point at which the definition
is being checked. In symbols the definitions are as follows:
(f continuous)
a S > 0 a > 0 such that x S, |x a| < a = |f (x) f (a)| <
(f uniformly continuous)
> 0 > 0 such that a, x S, |x a| < = |f (x) f (a)| <
where in the first formula we write a to emphasise the fact that a can depend
on a.
To see that the two definitions are in fact different, consider the following
example.
Example 6.13. The function f : R R given by f (x) = x2 is NOT uniformly
continuous. Note however that since f is a polynomial, it is continuous.
We will show this for = 1 (in fact it works for any > 0). Fix > 0. We
need to find x, a with |x a| < but with |x2 a2 | = 1. To ensure that
|x a| < let us choose x to be a + 2 . We consider what happens as a .
We have |x2 a2 | = |x a| |x + a| = 2 (a + 2 + a) a. For a 1/ we then
have a = 1. Thus we see that for any fixed , if a is large enough, that is,
for a 1/ we can find a value of x with |x a| < , but with |x2 a2 | 1.
Exercise 6.14. Show that the function f : [u, v] R given by f (x) = x2 is
uniformly continuous, for any fixed u, v R and u < v.
MATH2039 ANALYSIS 43
Exercise 6.15. Show that the function f : (0, ) (0, ) given by f (x) = x1
is not uniformly continuous. (Note that it is continuous on the given domain.)
In Example 6.13, the reason why the function is not uniformly continuous is
that the domain is unbounded.Exercise 6.14 shows that over a bounded interval
[u, v] it is uniformly continuous. In Exercise 6.15 the issue is slightly different.
Restricting 1/x to a bounded domain such as (0, 1], it is still not uniformly
continuous: the problem is that the function diverges to as we approach
zero. If however we take a closed interval [u, v] for our domain, with 0 < u < v,
then the function is uniformly continuous.
The following two theorem shows that this holds in general. It is proved as
applications of the Bolzano-Weierstrass theorem.
Theorem 6.16. (Uniform continuity) Let f be a function defined on a
closed interval [a, b]. Then f is continuous on [a, b] if and only if f is uniformly
continuous on [a, b].
The above result shows that continuous functions behave particularly well on
closed intervals. We now prove some further important results about continuous
functions on closed intervals. The first of these is another application of the
Bolzano-Weierstrass theorem.
Theorem 6.17. (Maximum Value Theorem) Let f be a function that is
continuous on the closed interval [a, b]. Then f achieves its maximum on [a, b];
that is, there exists some xmax in [a, b] so that f (xmax ) f (x) for all x [a, b].
44 PERISIC
Proof. First we prove that f is bounded above, that is, there exists R in R
such that f (x) R for all x. Suppose not. Then for each n N there exists
a point xn in [a, b] such that f (xn ) n. By the Bolzano-Weierstrass theorem
there exists a subsequence xnk of xn , with xnk L for some L [a, b]. As f is
continuous, f (xnk ) converges to f (L) as k . However f (xnk ) > nk which
diverges to infinity. This contradiction proves that f is bounded above.
Let A = {f (x) : x [a, b]}. We have shown that this is bounded above, hence
it has a supremum s. As s is the least upper bound, given any > 0, we
have that s is not an upper bound of A. In particular, for each n, there
exists an A with an > s 1/n. As an A, it follows that an = f (yn ) for
some yn in [a, b]. Applying the Bolzano-Weierstrass theorem again, there is a
subsequence ynk of yn such that with ynk xmax for some limit xmax . Now
f (ynk ) = ank > s 1/nk , while on the other hand f (ynk ) s as s is an upper
bound. Thus by the squeeze rule f (ynk ) s as k . As f is continuous we
deduce that f (xmax ) = s f (x) for all x in [a, b].
Exercise 6.18. Prove, if f is continuous and if limx (f (x + 1) f (x)) = 0,
then limx f (x)/x = 0.
Exercise 6.19. The Minimum Value Theorem states that, if f is contin-
uous on [a, b], then f achieves its minimum on [a, b]; that is, there exists some
y0 in [a, b] so that f (y0 ) f (x) for all x [a, b]. Use the Maximum Value
Theorem to prove the Minimum Value Theorem.
Proof. We will suppose that f (a) < y < f (b), the case f (a) > y > f (b) being
similar. Define
A = {x [a, b] : f (x) y}.
Note that a A, so in particular A is non-empty, and A [a, b] so that A is
bounded. Hence A has a supremum s in [a, b]. As in the proof of the Maximum
Value Theorem, there exists an in A with an > s 1/n, and since an in A we
have an s. Thus by the squeeze rule, an s as n , so as f is continuous,
f (an ) f (s). We deduce that f (s) y, since f (an ) < y for all n.
As y < f (b), and f (s) y we must have s 6= b, so s < b. It follows that
s+1/n [a, b] for n sufficiently large. As s+1/n > s we know that s+1/n / A,
so f (s + 1/n) > y. By continuity of f and inequality of sequences, it follows
that f (s) = lim f (s + 1/n) y. Since f (s) y we conclude that in fact
n
f (s) = y.
Example 6.21. For the function f (x) which is continuous on the closed interval
[a, b] and satisfies a < f (x) < b for all x [a, b], use the Intermediate value
MATH2039 ANALYSIS 45
Now break [0, 1] in half: the value of g at 0.5 is g(0.5) = (0.5)2 cos(0.5) =
0.6276..., and so there is a solution to g(x) = 0 in [0.5, 1].
46 PERISIC
We could continue, but lifes too short. However this is an easy method to
teach a computer, since it involves evaluating a function, comparing numbers,
and diving by 2. It is also possible to make this method a bit more intelligent:
there is no reason to divide the intervals in the middle. For instance, in the
last step done above, it would make sense to break the interval [0.75, 1] closer
to 0.75 than to 1, since the value of g at 0.75 is closer to 0 than the value of g
at 1.
MATH2039 ANALYSIS 47
7. Differentiability
Proof. The proof of this is the evaluation of a single limit. Recall that f is
continuous at a if limxa f (x) = f (a), or equivalently, if limxa (f (x)f (a)) =
0. So,
f (x) f (a) f (x) f (a)
lim (f (x) f (a)) = lim (x a) = lim lim (x a) = 0
xa xa xa xa xa xa
f (x)f (a)
as desired, since limxa xa = f 0 (a) and limxa (x a) = 0.
Remark 7.3. The converse is not true: f (x) = |x| is a continuous function
but it is not differentiable at x = 0.
Theorem 7.4. (Rolles theorem) Suppose that the function f is continuous
on the closed interval [a, b] and differentiable on the open interval (a, b), and
that f (a) = f (b). Then, there exists a number c in the interval (a, b) so that
f 0 (c) = 0.
If f does not achieve its maximum at some point in (a, b), then it must achieve
its maximum at the endpoints, and so f (x) f (a) for all x [a, b]. We
can make the same argument at the point c in (a,b) at which f achieves its
minimum, making use of the minimum value property for f , and again argue
that if f achieves its minimum at a point c in (a, b), then f 0 (c) = 0.
The only remaining alternative is that f achieves both its maximum and its
minimum at the endpoints of [a, b], in which case it must be that f is constant
on [a, b]. In this case, we can easily calculate that f 0 (c) = 0 at every c in (a, b).
This completes the proof of Rolles theorem.
Theorem 7.5. (Mean Value Theorem (MVT)) Suppose that the func-
tion f is continuous on the closed interval [a, b] and differentiable on the open
interval (a, b). Then c (a, b) such that
f (b) f (a)
f 0 (c) = .
ba
Example 7.8. Use the mean value theorem to prove that if f 0 (x) is constant
on R, then f (x) is a linear function; that is, there exist constants a and b so
that f (x) = ax + b.
Since f 0 (x) is constant on R, there exists some a R so that f 0 (x) = a for all
x R. Consider the function g(x) = f (x) ax. Since both f and the linear
polynomial ax are differentiable on all of R, and hence continuous on all of R,
we have that g is differentiable on all of R, and hence is also continuous on all
of R. In order to apply the mean value theorem, we need to work on closed
intervals.
So, for x0 > 0, consider the interval [0, x0 ]. Since g is continuous on [0, x0 ] and
differentiable on (0, x0 ), the mean value theorem states that there exists some
c in (0, x0 ) so that g(x0 ) g(0) = g 0 (c)(x0 0). However, g 0 (c) = f 0 (c) a =
a a = 0, and so g(x0 ) g(0) = 0, and so g(x0 ) = g(0) for all x0 > 0. To
show that g(x0 ) = g(0) for all x0 < 0 as well, work with the interval [x0 , 0] and
repeat the argument just given.
So, g(x) is constant, that is, there is b R so that g(x) = b for all x R.
Substituting in g(x) = f (x) ax, this yields that f (x) ax = b for all x R,
or that f (x) = ax + b for all x R, where a and b are constants, as desired.
Exercise 7.9. Use the mean value theorem to prove each of the following
statements.
(1) If g 0 (x) is a polynomial of degree n 1, then g(x) is a polynomial of
degree n;
(2) x/(x + 1) < ln(1 + x) < x for 1 < x < 0 and for x > 0;
(3) sin(x) < x for x > 0;
Example 7.10. For the function g(x) = x2 cos(x), the same as in Exercise
6.22, use Rolles theorem or the mean value theorem to determine whether the
solutions described in Exercise 6.22 to the equation g(x) = 0 are the only ones.
In Exercise 6.22, we saw that there exist at least two solutions c1 and c2 to
this equation, where 0 < c1 < 2 and 2 < c2 < 0. Suppose there were a third
solution c3 to g(x) = 0. Then, since there are three points c1 , c2 , and c3 at
which g(x) = 0, by Rolles theorem there would exist two points e1 and e2 at
which g 0 (x) = 0. (For instance, if c3 < c2 , then e1 would lie between c3 and c2 ,
and e2 would lie between c2 and c1 .) Note that there is already one point at
which g 0 (x) = 0, namely x = 0.
However, by the same sort of argument used in the solution to parts 3 and 4 of
Exercise 7.9, we have that g 0 (x) satisfies g 0 (x) = 2x + sin(x) 6= 0 for all x 6= 0.
(Specifically, we have that g 0 (0) = 0, and that g 00 (x) = 2 + cos(x) > 0 for all
x R, since 1 cos(x) 1 for all x R. Hence, g 0 (x) < 0 for all x < 0
and g 0 (x) > 0 for all x > 0.) Hence, by Rolles theorem, there are only the two
solutions to g(x) = 0 that we had already found.
50 PERISIC
Exercise 7.11. For each of the following functions, the same as in Exercise
6.22, use Rolles theorem or the mean value theorem to determine whether the
solutions described in Exercise 6.22 are the only ones.
(1) f (x) = 0 on the interval [a, a], where a is an arbitrary positive real
number and f (x) = x1995 + 7654x123 + x;
(2) tan(x) = ex for x in [1, 1];
(3) 3 sin2 (x) = 2 cos3 (x) for x > 0;
(4) 3 + x5 1001x2 = 0 for x > 0;
Exercise 7.12. For each of the following functions described below, determine
whether there is a solution to the given equation in the specified set.
(1) g 0 (a) = 0 = g 0 (b), where a < b are real numbers and g(x) = x3
12x2 + 44 2 x 48 3 + cos(x) 1;
(2) f 0 (a) = 0, where f (x) = x4 3 x sin(x) and a R;
(3) g 0 (x) = 0 for at least k 1 distinct real numbers, where g(x) is a
differentiable function on R which vanishes at at least k distinct real
numbers.
(4) x3 + px + q = 0 has exactly one real root for p > 0;
MATH2039 ANALYSIS 51
Theorem 8.1. (Cauchy Mean Value Theorem) Let f and g be two func-
tions that are both continuous on [a, b] and differentiable on (a, b). Suppose
further that g 0 (x) is never zero on (a, b). Then, there exists some c in (a, b) so
that
f (b) f (a) f 0 (c)
= 0 .
g(b) g(a) g (c)
The main use of the Cauchy mean value theorem for us is to give a proof of
lHopitals rule.
Theorem 8.2. (lHopitals rule) Suppose that f and g are differentiable on
the union I = (a , a) (a, a + ) for some > 0, and that g 0 (x) is non-zero
on I. Suppose also that
lim f (x) = lim g(x) = 0.
xa xa
Then
f (x) f 0 (x)
lim = lim 0 ,
xa g(x) xa g (x)
[a, x] (if x > a, or on the interval [x, a] if x < a). Hence, regardless of which
case were in, there exists some number z between a and x so that
f (x) f (x) f (a) f 0 (z)
= = 0 .
g(x) g(x) g(a) g (z)
Since z lies between a and x, it must be that z a as x a, and so
f 0 (z) f 0 (z)
lim 0
= lim 0 .
za g (z) xa g (z)
Since
f (x) f 0 (z)
= 0 ,
g(x) g (z)
we also have that
f (x) f 0 (z)
lim = lim 0 ,
xa g(x) xa g (z)
There are other indeterminate forms that can be handled by lHopitals rule
and some algebraic manipulations:
indeterminate form 0 : limxa f (x)g(x), where limxa f (x) = 0
and limxa g(x) = . In this case, rewrite
f (x)
lim f (x)g(x) = lim
xa xa 1/g(x)
to get indeterminate form 00 .
We close by noting that lHopitals rule also holds for limits of the form
limx fg(x)
(x)
which have one of these indeterminate forms. Remember
that it is necessary to check that the limit has an indeterminate form
before applying lHopitals rule.
54 PERISIC
9. Integration
In this section we will define the Riemann integral. We note that the integral
should have the following properties:
Z b Z b Z b
(1) (f (x) + g(x)) dx = f (x) dx + g(x) dx.
a a a
(2) If f, g are real valued functions and f (x) g(x) for all x [a, b] then
Z b Z b
f (x) dx g(x) dx.
a a
Z b
The first condition says that f 7 f (x) dx is a linear map, so this is
a
an algebraic property. The second condition involves inequalities, so it is an
analytic property. The integral is a first example of functional analysis, that
is, analysis involving linear algebra.
Our aim is to define the Riemann integral of a function f over an interval [a, b],
Rb Rb
written as a f . When the function f is non-negative the integral a f gives the
area under the graph of f over the interval [a, b]. It is easy to calculate the area
of an rectangle. The definition of the Riemann integral is based on calculation
of areas of rectangles. In this section we will explore that idea.
We start by defining a partition of an interval [a, b].
Definition 9.1. Let [a, b] R be a finite, bounded interval. A partition
of the interval [a, b] is a finite set of points {x0 , x1 , . . . , xn } such that
a = x0 < x1 < . . . < xn = b.
The points xi , for i = 0, 1, . . . , n, are called partition points of , and for
i = 1, . . . , n the interval [xi1 , xi ] is called the i-th subinterval of . Let
xi = xi xi1 be the length of the i-th interval, i = 1, . . . , n. The norm of
the partition , ||, is the length of its longest subinterval, formally defined
as
|| = max{x1 , x2 , . . . , xn }.
Example 9.2. For example = {0, 81 , 14 , 12 , 1} is a partition of [0, 1] with the
partition points
1 1 1
x0 = 0, x1 = , x2 = , x3 = , x4 = 1.
8 4 2
This partition consists of five points which divide the interval [0, 1] into four
subintervals. The length of subintervals are
1 1 1 1 1
x1 = x1 x0 = 0 = , x2 = x2 x1 = = ,
8 8 4 8 8
1 1 1 1 1
x3 = x3 x2 = = , x4 = x4 x3 = 1 = .
2 4 4 2 2
The norm of the partition is the length of the longest subinterval, that is
1 1 1 1
|| = max{ , , } = .
8 4 2 2
56 PERISIC
By adding points into a partition we form finer partitions. We define the notion
of a refinement of a partition.
Definition 9.3. Given a partition of an interval [a, b], a partition 0 is called
a refinement of provided every partition point of is a partition point of
0 , 0 . If 0 \ contains k points, we say that 0 is a k-point refinement
of . Note that is a 0-point refinement of itself.
Example 9.4. Take the partition = {0, 18 , 14 , 12 , 1} of [0, 1] studied in Example
9.2 and define a new partition 0 by adding points,
3 5 7
0 = { , , , }.
8 8 8
The partition 0 : 0 < 1
8 < 1
4 < , < 7
8 < 1 is a refinement of . The partition
points of 0 are
i
x0i = ; i = 0, 1, 2, . . . , 8
8
and the norm | 0 | = 18 .
(Note that these numbers exist since f is bounded.) The upper Darboux
sum associate with the function f and the partition is defined by
n
X
U (f, ) = Mi xi ,
i=1
and the lower Darboux sum associate with the function f and the partition
is defined by
n
X
L(f, ) = mi xi .
i=1
We will calculate the lower and upper Darboux sums in a simple example.
Example 9.8. Let f (x) = x, [a, b] = [0, 1], n N. Consider the partition
with uniform subintervals
1 2 n
: 0 < < < ... < = 1
n n n
Hence, the partition points are xi = ni , i = 0, 1, . . . , n. For i = 0, 1, . . . , n, we
calculate
i
Mi = sup{f (x)|xi1 x < xi } = xi = ,
n
i1
mi = inf{f (x)|xi1 x < xi } = xi1 = ,
n
and the length of the subintervals
i i1 1
xi = xi xi1 = = .
n n n
Now we can calculate the Darboux sums
n n n
X X i 1 1 X 1 n(n + 1) 1 1
U (f, ) = Mi xi = = 2 i= 2 = + ,
i=1 i=1
n n n i=1 n 2 2 2n
n n n
X X i1 1 1 X 1 (n 1)n 1 1
L(f, ) = mi xi = = 2 (i 1) = 2 = .
i=1 i=1
n n n i=1 n 2 2 2n
Upper and lower Darboux sums have a useful monotonicity property stated in
the following lemma.
Lemma 9.10. (Refinement lemma) Let f : [a, b] R be a bounded function.
Let be a partition of [a, b] and let 0 be a refinement of . Then
L(f, ) L(f, 0 ) U (f, 0 ) U (f, ).
58 PERISIC
Proof. By the definition of the upper and lower Darboux sums L(f, 0 )
U (f, 0 ) . To prove
U (f, 0 ) U (f, )
it suffices to prove the inequalities for the case when 0 is 1-point refinement
of . We consider the refinement 0 = {c} where xk1 < c < xk for some
k {0, 1, 2, . . . , n}.
Because the most intervals are unchanged, the sum U (f, 0 ) is obtained from
U (f, ) by replacing only the term Mk xk with
(1) (2)
Mk (c xk1 ) + Mk (xk c)
where
(1)
Mk = sup{f (x)|xk1 x < c}, and
(2)
Mk = sup{f (x)|c x < xk }.
(1) (2)
Since Mk , Mk Mk , it follows
(1) (2)
Mk (cxk1 )+Mk (xk c) Mk (cxk1 )+Mk (xk c) = Mk (xk xk1 ) = Mk xk
and hence U (f, 0 ) U (f, ).
Similarly we prove L(f, ) L(f, 0 ). This completes the proof of the theorem.
The Refinement lemma states that when we add more points into a partition,
the upper sums decrease and the lower sums increase. Our next goal is to
compare L(f, ) and U (f, 0 ) for any two partitions and 0 of [a, b]. We know
that for two partitions and 0 of [a, b], 0 is again a partition of [a, b]
and it is the smallest partition that is a common refinement for both partitions
and 0 . With this in mind, we are equipped to prove so called Comparison
lemma.
Lemma 9.11. (Comparison lemma) Let f : [a, b] R be a bounded function
and let , 0 be two partitions of [a, b]. Then
L(f, 0 ) U (f, ).
Proof. Let , 0 be any two partitions of [a, b]. Since the partition 0 is the
refinement for both partitions, and 0 , by the Refinement Lemma we have
L(f, 0 ) L(f, 0 ) U (f, 0 ) U (f, ).
The number
I = sup L(f, ) = inf U (f, )
Exercise 9.16. Show that the indicator function of the rational numbers Q
(
1 xQ
Q (x) =
0 x /Q
is not Riemann integrable over [0, 1].
Let : x0 = 0 < x1 < . . . < xn = 1 be a partition of [0, 1]. For each subinterval
[xk , xk+1] we have
mk = inf{Q (x)|xk1 x < xk } = 0 and Mk = sup{Q (x)|xk1 x < xk } = 1
and therefore the Darboux sums are
n
X n
X
U (Q , ) = 1 xk = 1 and L(Q , ) = 0 xk = 0
k=1 k=1
which is Riemann integrable and we have seen that the function which has
discontinuity at each rational number is not integrable over [0, 1] or any other
finite interval [a, b], a < b. A function does not need to be continuous in order
to be integrable. This is illustrated on the next example.
Example 9.17. The function f : [0, 2] R defined by
(
1 0x<1
f (x) =
2 1 x < 2
is a Riemann integrable function over [0, 2].
To show that the function f is Riemann integrable it is enough to find a partition
such that U (f, ) = L(f, ) since
L(f, ) sup L(f, ) inf U (f, ) U (f, ).
Consider the partition 0 < 1 < 2 where the partition points are x0 = 0 <
x1 = 1 < x2 = 2. Then m1 = M1 = 1 and m2 = M2 = 2 giving
L(f, ) = U (f, ) = M1 1 + M2 1 = 1 + 2 = 3.
Example 9.18. Using the Riemann integrability criterion, we will show that
the function f (x) = x is Riemann integrable over [0, 1].
1
Proof. Let > 0. By the Archimedean property n0 N such that 0 < n0 < .
Now let be the partition with the partition points
2
i
xi = , i = 0, 1, . . . , n0 ,
n0
that is, the partition
2 2 2
1 2 n0
: 0 = x0 < < < ... < = 1.
n0 n0 n0
Observe that f (x) = x is an increasing function. Hence for i = 1, 2, . . . , n0
we have s
2
i i
Mi = f (xi ) = = ,
n0 n0
s 2
i1 i1
mi = f (xi1 ) = = .
n0 n0
Now,
n0 n0
X 1 X 1 1 1
U (f )L(f, ) = (Mi mi )xi = xi = (xn0 x0 ) = (10) = <
i=1
n 0 i=1 n 0 n 0 n 0
Proof. To prove this, for a given > 0 we need to find a partition of [a, b]
such that
U (f, ) L(f, ) < .
Since f is a continuous function on [a, b], f is uniformly continuous on [a, b].
For a given > 0, we can find > 0 such that
x, y [a, b], with |x y| < = |f (x) f (y)| < .
ba
Now pick any partition : a = x0 < x1 < x2 < < xn = b with || < . For
k {1, 2, . . . , n} By the min/max value theorem
Mk = max{f (x)|x [xk1 , xk ]}
mk = min{f (x)|x [xk1 , xk ]}
i.e. x , x [xk1 , xk ] such that mk = f (x0 ) and Mk = f (x00 ). Now
0 00
Mk mk = f (x00 ) f (x0 ) < ,
ba
and we get
n n
X X
U (f, ) L(f, ) = (Mk mk )xk < xk = (xn x0 ) = .
ba ba
k=1 k=1
Exercise 9.22. Suppose f is integrable on [a, b]. For [u, v] [a, b], show that
the restriction of f to [u, v] is integrable.
The Riemann integral is defined for bounded functions on closed bounded in-
tervals. In order to integrate a function that is not bounded, or is defined on
an unbounded interval, we use limits. These are called improper integrals.
There are many kinds of improper integrals. We do not give a precise definition
that covers all cases.
Definition 9.24. An improper integral is an integral in which one (or more)
of the following occurs:
the interval of integration is not a closed interval, but instead is one of
(, a], [a, ), or (, ), or
the integrand has an infinite discontinuity at some point c [a, b],
namely limxc f (x) = .
R Rc
integrals f (x)dx and f (x)dx both converge. This is different from
c
RM
assuming that limM M f (x)dx exists, as we will see later.
Rb
An improper integral of the form f (x)dx (where f is continuous on (a, b]
a Rb
and limxa+ f (x) = ) converges if the limit limca+ c f (x)dx exists (as
Rb
a finite number). If the improper integral a f (x)dx exists, we set
Z b Z b
f (x)dx = lim f (x)dx.
a ca+ c
Similarly for improper integrals of this form where f is continuous on [a, b) and
limxb f (x) = .
Rb
An improper integral of the form a
f (x)dx (where f is continuous on [a, b] ex-
Rc Rb
cept at a single point c in (a, b), and both the integrals a f (x)dx and c f (x)dx
Rc
are of the previous form) converges if both the improper integrals a f (x)dx
Rb
and c f (x)dx converge.
because the interval of integration is [4, ) and hence is not a closed interval.
So, we attempt to evaluate this integral by formulating it as a limit of integrals
over closed intervals, namely
Z Z M
1 1
3/2
dx = lim 3/2
dx
4 x M 4 x
Z M
= lim x3/2 dx
M 4
2 2
= lim + = 1.
M M 4
Hence, this improper integral converges to 1.
Exercise 9.26. Determine whether the following improper integrals converge
or diverge, and evaluate those which converge.
R4 dx
(1) 0 x3/2
;
R dx
(2) 1 (x+1) ;
R dx
(3) 5 (x1)3/2
;
R9 dx
(4) 0 (9x)3/2
;
MATH2039 ANALYSIS 65
R 2 dx
(5) (x+1)3
;
R8 dx
(6) 1 x1/3
;
R dx
(7) 2 (x1)1/3
;
R xdx
(8) (x2 +4)
;
R1
e x dx
(9) 0 x
;
R dx
(10) 1 x ln(x) .
10. Series
Definition
P10.1. A series (or infinite series) is a mathematical construct of
the form n=0 an . A series is essentially the sum of the elements in a sequence.
P P
Consider a series n=0 an . We define what it means for n=0 an to converge
or diverge by reducing to what weP have done before, namely sequences. Define
the k-th partial sum of the series n=0 an as
k
X
Sk = an
n=0
and consider the sequence (Sk )Pk=0 . The sequence (Sk ) is called the sequence
of partial sums of the series n=0 an .
P
Definition 10.2. The series Pan converges if the sequence of partial
n=0
sums (Sk ) converges. The series n=0 an diverges if the sequence of Ppartial
sums (Sk ) diverges. If (Sk ) converges to S, we say that the series n=0 an
converges to S as well and we write
X
an = S = lim Sk .
k
n=0
P
Example 10.3. Consider the series n=0 an , where an = 1 for all n 0. This
series diverges. To see this, consider the partial sums: the k th partial sum Sk
is
Xk Xk
Sk = an = 1 = k + 1,
n=0 n=0
and the sequence Sk = k + 1 diverges.
P P
Fact 10.4. Let n=0 an and n=0 bn be two infinite series, and suppose there
exists P N so that an = bn for all n > P . (That
Pis, assume the terms of the
two
P series are equal after some point.) Then, n=0 an converges if and only
if n=0 bn converges. That is, the convergence or divergence of a series is not
affected by mucking about with finitely many terms of the series.
This is a powerful and highly useful fact that Ill make repeated use of, often
without being explicit about it, so keep a sharp eye out.
For s = 1, this series is called the harmonic series, and we can prove directly
that it diverges. Note that 31 + 14 > 12 , that 15 + + 81 > 4 81 = 12 , and in general
that
1 1 1 1 1
+ + + k > 2k1 k = .
2k1 + 1 2k1 + 2 2 2 2
Hence, the (2k )th partial sum S2k satisfies S2k > 1 + k 12 . Since the terms in
the harmonic series are all positive, the sequence of partial sums is monoton-
ically increasing, and by the calculation done the sequence of partial sums is
unbounded, and so the sequence of partial sums diverges. Hence, the harmonic
series diverges.
P 1
Exercise 10.8. Prove that n=1 ns diverges for s < 1, by estimating its
partial sums.
P P
Theorem 10.9. (Arithmetic
P of series)
P Let n=0 an and n=0 bn be con-
vergent series with n=0 an = A and n=0 bn = B.
P P P
(1) sum: n=0 (an + bn ) = n=0 an + n=0 bn = A + B;
P P P
(2) difference: n=0 (an bn ) = n=0 an n=0 bn = A B;
(3) P
multiplication Pby a constant: For a constant c,
n=0 c a n = c n=0 an = cA.
68 PERISIC
Pk Pk
Proof. Let Sk = n=0 an and Vk = n=0 bn be the partial sums of the two
series. Since the series are both convergent, we have that limk Sk = A and
limk Vk = B.
(1) This follows immediately from the definition of convergence
P of a series
in terms of partial sums: the partial sums of the series n=0 (an + bn )
are
Xk k
X Xk
Tk = (an + bn ) = an + bn = Sk + Vk
n=0 n=0 n=0
(since the sums are finite). Since limk Sk = A and limk Vk = B,
we have that
lim Tk = lim (Sk + Vk ) = lim Sk + lim Vk = A + B.
k k k k
P
So, not only does the series of sums n=0 (an + bn ) converge (since
its sequence of partial sums converges), but it converges to A + B, as
expected.
(2) Much as the rule for sums just done, this follows immediately from the
definition of convergence
P of a series in terms of partial sums: the partial
sums of the series n=0 (an bn ) are
k
X k
X k
X
Wk = (an bn ) = an bn = Sk Vk
n=0 n=0 n=0
P P
(2) Show that, if n=0 an diverges and if c 6= 0, then the series n=0 c an
diverges.
P
Example
P 10.11. Construct an example of convergent
P series n=0 an and
n=0 bn with positive terms for which the series n=0 an bn diverges, or prove
that no such example exists.
P P
No such example exists: since n=0 an and n=0 bn both converge, the se-
Pk Pk
quences of partial sums Sk = n=0 an and Vk = n=0 bn both converge. Note
that the k th partial sum Wk of the series of products satisfies
k k
! k !
X X X
Wk = a n bn an bn = Sk Vk .
n=0 n=0 n=0
P P
Since Sk A = n=0 an and Vk B = n=0 bn for all k, we have that
Wk AB for all k. Since Wk is a monotonically increasing sequence (as an
and bn are positive for all n) and since Wk is bounded
P (by AB), we have that
Wk converges, and hence that the series of products n=0 an bn converges.
Exercise 10.12. Unlike sequences, the convergence of series whose terms are
products and quotients of convergent series does not necessarily follow. Ex-
ploring this phenomenon is the purpose of this example. Construct examples
of each of the following, or prove that no such example exists:
P P
(1) convergent series n=0 Pa
n and n=0 bn with positive terms for which
the series of products n=0 an bn converges;
P P
(2) divergent series n=0P an and n=0 bn with positive terms for which
the series of products n=0 an bn diverges;
P P
(3) divergent series n=0P an and n=0 bn with positive terms for which
the series of products n=0 an bn converges;
P P
(4) convergent series n=0Pan and n=0 bn with positive terms for which
an
the series of quotients n=0 bn diverges;
P P
(5) convergent series n=0Pan and n=0 bn with positive terms for which
an
the series of quotients n=0 bn converges;
P P
(6) divergent series n=0 P an and n=0 bn with positive terms for which
the series of quotients n=0 abnn diverges;
P P
(7) divergent series n=0 P an and n=0 bn with positive terms for which
the series of quotients n=0 abnn converges;
(If all the terms in the series are non-negative, then the sequence of partial
sums is monotonically non-decreasing.) This fact makes an appearance in the
proofs of several of the following tests for convergence or divergence of series.
Theorem 10.14. (Series convergence tests) Be careful when reading the
hypotheses, as not all these tests have the same hypotheses. In particular, some
only apply to series with non-negative terms, while others apply to all series.
nth term test for divergence: If limn an 6= 0P (so that either an
diverges, or an converges to a 6= 0), then the series n=1 an diverges.
P P
Comparison test: If Pn=0 an and n=0 bn areP series with 0 an
bn for all n 0, and if n=0 bn converges, then n=0 an converges.
P P
Limit comparison test: If n=0 an and n=0 bn are series with
non-negative terms and if the limit limn abnn = L exists with 0 <
P P
L < , then n=0 an converges if and only if n=0 bn converges.
P
Integral test: If n=0 an is a series of positive terms and if there
exists
R a decreasing continuous function f (x) for which f (n)
P= an , then
0
f (x)dx converges (that is, is finite) if and only if n=0 an con-
verges. [This integral is an improper integral, as described earlier.]
P
Ratio test: Let n=0 an be a series ofPpositive terms and suppose
limn aan+1 = L exists. If L < 1, then n=0 na converges. If L > 1,
P n
then n=0 an diverges. If L = 1, this test gives no information.
P
Root test: Let n=0 an be a series of positive terms and suppose
1/n
P
limn (an )P = L exists. If L < 1, then n=0 an converges. If
L > 1, then n=0 an diverges. If L = 1, this test gives no information.
P
Alternating series test: Consider a series of the form n=0 (1)n an ,
where an > 0 for all n 0. If an+1 an for all n 0 and limn an =
0, then the series converges.
P
Suppose now that n=0 an converges. For each n 1, let Wn be
the rectangle over the interval [n 1, n] with height f (n 1) = an1 .
The part of the graph of f over [0, M ] is contained in the union of the
rectangles W0 WM 1 , and so comparing areas, we see that
Z M XM M
X
f (x)dx (area of Wn ) = an1 .
0 n=1 n=1
RM
As above, the sequence 0 f (x)dx is monotonically increasing and
P RM
bounded (by n=0 an ), and so limM 0 f (x)dx exists (and is fi-
nite), as desired.
Ratio test: [note: the proofs of the ratio and root tests are similar
to each other, but different from the proofs already given, in that they
dont use partial sums, but instead use comparison to an appropriately
chosen geometric series.]
Suppose now that L > 1, and essentially repeat the argument. Choose
some so that 1 < < L; applying the definition of limit with = L,
there exists M so that aan+1
n
> for n M . (Note the change from the
usual n > M to n M , made here purely for notational convenience.)
So, aM +1 > aM , and aM +2 > aM +1 > 2 aM , and in general, we
have that aM +k > k aM for k 0. (Were using here that the an are
all positive, so that among other things, the inequalities dont change
direction
P k when we multiply through by an .) Since the geometric series
k=0 diverges P(since > 1), the comparison test yields that the
truncated
P series n=M an diverges, and hence that the original series
n=0 an diverges.
[The reason this proof does not work when L = 1 is that we cannot
find a number between L and 1, as we did in both of the parts of the
proof just given.]
Root test: The proof here is very similar to the proof just given (and
fails when L = 1 for the same reason). When L < 1, again choose
satisfying L < < 1, and then apply the definition of limit to find M
so that (an )1/n < for n M . Then, taking the nth power of both
MATH2039 ANALYSIS 73
sides, we get that an < n for all n M , and so again we can use the
P
second comparison testPwith the convergent geometric series n=M n
to get convergence of n=0 an .
When L > 1, choose satisfying 1 < < L, and apply the definition
of limit to get M so that (an )1/n > for n M , so that an > n for
P
n
M . By the comparison
n
P test with the divergent geometric series
n=M , we get that n=0 an diverges.
2p+1
X
S2p+1 = an = a0 (a1 a2 ) (a3 a2p ) a2p+1 ,
n=0
and again using that the parenthetical terms are non-negative, we see
that S2p+1 a0 for all p, and so the odd partial sums form a bounded
monotone sequence. Let S = limp S2p+1 .
We need to show now that the even partial sums S2p converge to the
same limit. However, since S2p = S2p1 +a2p and since limp a2p = 0,
we have that
and soPthe sequence Sk of all partial sums converges to S. That is, the
series n=0 (1)n an converges.
So, consider the function f (x) = x1s = xs , so that n1s = f (n). Since s > 1,
f 0 (x) = s xs+1
1
< 0 for all x > 0, and so f (x) is decreasing. Further,
Z Z M
f (x)dx = lim xs dx
1 M 1
1
xs+1 M
= lim 1
M s + 1
1 1 1
= lim 1 = .
M s + 1 M s1 s1
Since the limit converges, the series converges, as desired.
P
Remark 10.17. It is known that for s an even positive integer, that n=1 n1s
is a rational multiple of s . Moreover, there is an explicit formula for the sum
of this series.
For s an odd positive integer, we have already seen that this series
P diverges for
s = 1 (as this is the harmonic series). Further, it is known that n=1 n13 is an
P
irrational number, but it is not known that n=1 n13 is a rational multiple of
P 1
. Nothing is known about n=1 ns for s an odd positive integer s 5, other
than it is a convergent series.
The general philosophy with convergence of series is that a series will converge
provided that the terms tend to zero fast enough. As a general principal, always
start by applying the nth term test for divergence (whether you write out the
details or just apply the test mentally). If the terms dont go to zero then
youre done: the series diverges. Otherwise the question is just how fast do the
terms tend to zero. Is the series similar to one that you already know about?
Then perhaps you can use the comparison, or limit comparison tests. There
are also some rules of thumb for applying the ratio and root tests. If theres an
exponential or factorial then the ratio test is usually a good idea. If you have
something raised to the power n then the root test is useful.
In the series problems which follow, there are often many correct way to do
any problem. There are no hard and fast rules about which test to use, and
sometimes it may be easiest to use more than one: just keep applying different
tests until you get an answer, or a series that you already understand.
P
Example
P 10.18. For a convergent series n=1 an with positive terms, prove
that n=1 ann converges.
Pk th
P
Let Sk = n=1 an be the kP partial sum of P n=1 an . Consider the k th partial
an an
sum Tk of the new series n=1 n , Tk = n=1 n , and compare TP k to Sk :
an
since n an for all n 1, we have that Tk Sk for all n 1. Since n=1 an
is a convergent series with positive terms, its sequence of partial sums Sk is a
monotonically increasing sequence that converges to S. In particular, Sk S
for all k 1. Since Tk Sk S, we see that P Tk is a bounded monotonically
increasing sequence, and hence converges. So, n=1 ann is a convergent series.
MATH2039 ANALYSIS 75
P
Exercise 10.19. In each of the following, n=1 an is a convergent series with
positive terms.
(1) Prove that,
Pif cn is a sequence of positive terms satisfying limn cn =
0, then n=1 an cn converges;
(2) Prove that, P
if cn is a sequence of positive terms satisfying limn cn =
c 6= 0, then n=1 an cn converges.
In general, a series whose terms are positive is much easier to handle, particu-
larly in terms of determining convergence and divergence. One way to handle
a general series, that is one without the restriction that the terms are positive,
is to compare it to a series with positive terms.
P P
Definition 10.20. Let n=0 an be a series. The series P n=0 an converges
absolutely if the associated series of absolute values n=0 |an | converges.
Note that absolute convergence and convergence are the same for a series with
positive terms.
P P
Proof. Let n=0 an be a series that converges absolutely,
P so that n=0 |an |
converges. By the arithmetic of series, the series n=0 2|an | then also con-
verges.
P
We wish to understand whether or not the original series n=0 an converges.
Note that 0 an + |an | 2|an |, P and so by the comparison test, the series
P
(a +
n=0 n P n |a |) converges. Since |an | converges, by assumption, their
n=0P
P
difference n=0 (an +|an |) n=0 |an | = n=0 an converges, by the arithmetic
of series, and we are done.
In Theorem 10.14, we stated the ratio and root tests for series with positive
terms. Combining Theorem 10.14 with Proposition 10.21, we obtain the ratio
and root tests for series with non-zero terms, as tests to determine whether the
series converges absolutely or diverges.
P
Proposition 10.22. Ratio and root tests for general series: Let n=0 an
be a series with non-zero terms, so that an 6= 0 for all n.
Ratio test: Suppose that limn aan+1 = L exists. If L < 1, then
n
P P
a
n=0 n converges absolutely. If L > 1, then n=0 an diverges. If
L = 1, this test gives no information.
76 PERISIC
1/n
Root
P test: Suppose that limn (|an |) =LPexists. If L < 1, then
n=0 an converges absolutely. If L > 1, then n=0 an diverges. If
L = 1, this test gives no information.
Definition 10.23. Proposition 10.21 gives us that a series that converges ab-
solutely then necessarily converges. The converse however is not true: there
are series that converge but do not converge absolutely.
P
n
(9) n=1 n2 +n ;
P 2
(10) n=2 ln(n) ;
P sin2 (n)
(11) n=1 n2 +1 ;
n+2n
P
(12) n=1 n+3n ;
P 1
(13) n=2 n2 ln(n) ;
n3 +1
P
(14) n=1 n4 +2 ;
P 1
(15) n=1 n+n3/2 ;
10n2
P
(16) n=1 n4 +1 ;
n2 n
P
(17) n=2 n4 +2 ;
P
(18) 1 ;
n=1 n2 +1
P 1
(19) n=1 3+5n ;
P 1
(20) n=2 nln(n) ;
P cos2 (n)
(21) n=1 3n ;
P 1
(22) n=1 2n +3n ;
P 1
(23) n=1 n(1+ n) ;
P
(24) n=1 1/(2n (n + 1));
P
(25) n=1n!/(n2 en );
P n
(26) n=2 n/(3 ln(n));
P 3
(27) n=2 (2n)!/(n!) ;
P n 4
(28) n=1 (1 (1) )/n ;
P
(29) n=1 (2 + cos(n))/(n + ln(n));
P p
(30) n=3 1/(n ln(n) ln(ln(n)));
P n n
(31) n=1 n /( n!);
P n+1 n
(32) n=1 2 /n ;
P n1
(33) n=1 (1) / n;
P
(34) n=1 cos(n)/((n + 1) ln(n + 1));
P n 2 2
(35) n=1 (1) (n 1)/(n + 1);
P n n
(36) n=1 (1) /(n );
P n 2 3 2
(37) n=1 (1) (20n n 1)/(n + n + 33);
78 PERISIC
P
(38) n=1 n!/(100)n ;
P
(39) n=31/(n ln(n)(ln(ln(n)))2 );
P n
(40) n=1 (1 + (1) )/ n;
P n 2 n
(41) n=1 e cos (n)/(1 + );
P 4
(42) n=2 n /n!;
P n
(43) n=1 (2n)!6 /(3n)!;
P 100 n
(44) n=1 n 2 / n!;
P
(45) n=3 (1 + n!)/(1 + n)!;
P 2n 2
(46) n=1 2 (n!) /(2n)!;
P n 2
(47) n=1 (1) /(n + ln(n));
P 2n n
(48) n=1 (1) /2 ;
P n
(49) n=1 (2) /n!;
P 2
(50) n=0 n/(n + 1);
P
(51) n=1 100 cos(n)/(2n + 3);
P
(52) n=10 sin((n + 1/2))/ ln(ln(n));
P 2n 2
(53) n=1 (2n)!/(2 (n!) );
P n2
(54) n=1 (n/(n + 1)) ;
P
(55) n=1 1/(1 + 2 + + n);
P 3
(56) n=1 ln(n)/(2n 1);
P 2
(57) n=1 sin(n)/n ;
P n
(58) n=1 (1) (n 1)/n;
P n 3n n
(59) n=1 (1) 2 /7 ;
P 4
(60) n=1 cos(n)/n ;
P n n n
(61) n=1 (1) 3 /(n(2 + 1));
P n1
(62) n=1 (1) n/(n2 + 1);
P
(63) n=2 (1)
n1
/(n ln2 (n));
P n1 n
(64) n=1 (1) 2 /n2 ;
P n
3/2
(65) n=1 (1) sin( n)/n ;
P 4 n 2
(66) n=1 n e ;
P
(67) n=1 sin(n/2)/n;
P 8
(68) n=2 1/(ln(n)) ;
MATH2039 ANALYSIS 79
P
(69) n=13 1/(n ln(n)(ln(ln(n)))p ), where p > 0 is an arbitrary positive real
number.
P
Exercise 10.27. Let n=1 anPbe a convergent series of positive terms. Show
that for each s 1, the series n=1 asn is also convergent.
Example 10.28. Rearranging conditionally convergent series: There
is a rather strange fact, that illustrates the difference between an absolutely
convergent and a conditionally convergent series. First, we note that for an
absolutely convergent series, rearranging the terms does not affect the sum of
the series.
P
It remains only to check that n=0 bn converges to S, but this follows immedi-
ately from the construction of this new series, and the fact that limn pn =
limn qn = 0.
MATH2039 ANALYSIS 81
As with the difference between continuity and uniform continuity, the difference
between these two definitions is on one level small, merely the placement of a
quantifier, but it has major effects. To see this, if we rewrite the definition of
pointwise convergence, we get: A sequence of functions fn : S R converges
pointwise to a function f : S R if for each > 0 and for each a S, there
exists M so that if n > M , then |fn (a) f (a)| < . Namely, the difference
is in the placement of quantifier for each a S. We demonstrate that these
two definitions are different in two steps, one on a theorem and the other on an
example.
Theorem 11.3. Suppose that fn is a sequence of functions fn : S R, where
each fn is continuous on S. Suppose further that fn converges uniformly to f
on S. Then f is continuous on S.
since we know that each sum converges, and we have fn (x) |fn (x)| Mn for
each n. Similarly
X X
X
sk (x) s(x) = fn (x) |fn (x)| Mn <
n=k+1 n=k+1 n=k+1
since fn (x) |fn (x)| Mn for each n. We deduce that for all k > k0 , and
for all x S we have |sk (x) s(x)| < . Thus sk converges uniformly to s.
P
Corollary 11.7. Let n=0 fn be a series of continuous functions satisfying
the hypotheses of the Weierstrass M -test, and let s be the function this series
converges to. Then s is continuous.
MATH2039 ANALYSIS 83
P n
If r is the radius of convergence of n=0 an (x a) then by definition the
power series converges for x in
[
[a s, a + s] = (a r, a + r).
s<r
P
Theorem 12.3. For a real power series n=0 an (x a)n with radius of con-
vergence r, the series converges for all x R with |x a| < r and diverges for
all x R with |x a| > r.
P
Example 12.4. Consider the power series n=0 xn /(n + 1), which is a power
series centred at a = 0. We always begin the same way with a power series, by
using the ratio test. The ratio test asks us to calculate
n+1
x /((n + 1) + 1) n+1
lim
n xn /(n + 1) = |x| n
lim
n+2
= |x|.
By Proposition 10.22, this series converges absolutely for |x| < 1 and diverges
for |x| > 1. So, the radius of convergence is 1. Proposition 10.22 yields that the
open interval (1, 1) lies in the interval of convergence. In order to determine
the interval of convergence, we need to check the behavior of the series at
the two endpoints of Pthis interval, namely x = 1 and x = 1. At x = 1,
the series becomes n=0 1/(n + 1), which P is the harmonic series and hence
diverges. At x = 1, this series becomes n=0 (1)n /(n + 1), which is the
alternating harmonic series, and hence converges conditionally. So, the interval
of convergence is the half-open interval [1, 1).
Exercise 12.5. The power series scavenger hunt: for each of the power
series given below, determine the radius and interval of convergence.
P (1)n n
(1) n=0 n! x ;
5n
P
(2) n=1 n2 xn ;
P 1
(3) n=1 n(n+1) xn ;
P (1)n
(4) n=1
n
xn ;
P (1)n
(5) n=0 (2n+1)! x2n+1 ;
3n
P
(6) n=0 n! xn ;
P 1
(7) n=0 1+n2 xn ;
P (1)n+1
(8) n=1 n (x + 1)n ;
3n
P
(9) n=0 4n (x + 5)n ;
P (1)n
(10) n=1 n2 +4 (x + 1)2n+1 ;
n
P
(11) n=0 (2n+1)! (x 1)2n ;
P 1
(12) n=2 (ln n)n xn ;
86 PERISIC
P
(13) n=0 3 n xn ;
P n!
(14) n=0 2n xn ;
P (2)n
(15) n=1 n+1 xn+1 ;
P (1)n
(16) n=1 (2n)! x2n ;
P (1)n
(17) n=1 n3/2 x3n ;
P (1)n+1
(18) n=2 n ln2 (n) xn ;
P 1
(19) n=0 2n (x 3)n ;
P (1)n
(20) n=1 (n+1)2 (x 4)n ;
P (2n+1)!
(21) n=0 n3 (x 2)n ;
P ln(n)
(22) n=1 n (x 3)n ;
P 1
(23) n=0 42n (2x 3)n ;
P 1
(24) n=2 bn (x a)n , where b > 0 is arbitrary.
P (n+p)!
(25) n=0 n!(n+q)! xn , where p, q N;
P 1
(26) n=1 n3n xn1 ;
P (1)n1
(27) n=1 (2n1)! x2n1 ;
P
(28) n=1 n!(x a)n , where a R is arbitrary;
P n
(29) n=1 2n (3n1) (x 1)n .
1/n
Exercise 12.6. Prove,Pif an is an sequence satisfying limn1 |an | = L 6= 0,
then the power series n=0 an x has radius of convergence L .
Note that, if we use the ratio test to determine the radius of convergence of a
power series, we cannot then use the ratio test to determine whether the series
converges or diverges at the endpoints of the interval of convergence. This is
MATH2039 ANALYSIS 87
because the limit is equal to 1 at the endpoints of the interval, and when the
limit is 1 is precisely when the ratio test gives no information.
Exercise 12.7. For each of the following series, determine the values of x for
which the series converges.
P n
1 x+2
(1) n=1 2n1 x1 ;
P 1
(2) n=1 (x+n)(x+n1) .
P
Exercise 12.8. Suppose n=0 an (x a)n P is a power series with a radius of
convergence r. Show that the power series n=1 nan (x a)n1 also has the
radius of convergence r.
Theorem 12.9.
P (Arithmetic of powerPseries)
Let f (x) = n=0 an (x a)n and g(x) = n=0 bn (x a)n be two power series
with radii of convergence r1 and r2 respectively. Let r be the minimum of r1 , r2 ;
i.e. r = min{r1 , r2 }. Then, on the open interval (a r, a + r) the following
holds:
(i) The sum (f + g)(x) is given by the power series
X
(f + g)(x) = (an + bn )(x a)n .
n=0
(iv) The derivative of f (x) is given by differentiating the power series term
by term:
X
f 0 (x) = n an (x a)n1 .
n=1
Exercise 13.2. For each of the given functions, calculate its Taylor series about
the given point; also, determine the radius and interval of convergence of the
resulting power series wherever possible.
(1) f (x) = x3 + 6x2 + 5x 7 about a = 6;
(2) f (x) = e3x about a = 2;
(3) f (x) = cosh(x) about a = 1.
Lemma P 13.3. (UniquenessPof series representations) Consider two power
series n=0 an (x a)n and n=0 bn (x a)n that both converge P absolutely nand
uniformly
P on the open interval (a, a+) for some > 0. If n=0 an (xa) =
n
n=0 bn (x a) for all x in (a , a + ), then an = b n for all n 0.
We now present a test to determine when a function is equal to its Taylor series,
followed by an example of a function NOT equal to its Taylor series.
MATH2039 ANALYSIS 89
Let f be a function which has derivatives of all orders in the interval (ar, a+r),
for some r > 0. Then by definition, the series
X f (n) (a)
(x a)n
n=0
n!
converges to f (x) if and only if
Rk (x) = f (x) sk (x)
converges to zero as k .
Theorem 13.4. (Taylors theorem)
Suppose f is (k + 1)-times differentiable on (a r, a + r) for some r > 0. Let
k
X f (n) (a)
sk (x) = (x a)n .
n=0
n!
Then for each x (a r, a + r) with x 6= a, there exists c between a and x such
that
f (k+1) (c)
f (x) = sk (x) + (x a)k+1
(k + 1)!
The theorem says that for each x, k there exists c such that
f (k+1) (c)
Rk (x) = (x a)k+1 .
(k + 1)!
Thus we have the following corollary.
Corollary 13.5. Let f be a function which has derivatives of all orders in the
(k+1)
interval (ar, a+r). Suppose that x (ar, a+r), x 6= a, and f (k+1)!(c) (xa)k+1
converges to zero uniformly in c as k (for x, a fixed). Then the series
X f (n) (a)
(x a)n
n=0
n!
converges to f (x).
Example 13.6. Consider the Taylor series for the function f (x) = ex centred
at a = 0, namely
X f (n) (0) n X 1 n
x = x ,
n=0
n! n=0
n!
since f (n) (0) = e0 = 1 for all n 0. In order to show that
X 1 n
ex = x
n=0
n!
for all x in R, we need to show for each x that Rk (x) 0. We do this by
estimating Rk (x). By Taylors Theorem we have
ec
Rk (x) = xk+1 0,
(k + 1)!
where c lies between 0 and x and depends on x and k.
90 PERISIC
First take the case that x > 0. Let M = ex . We have that 1 < ec ex = M
for all c between 0 and x, since ex is an increasing function. We know that
1 k
limk k! x = 0 and limk M k
k! x = 0, for fixed x, the squeeze law for limits
ec k+1
yields that limk (k+1)! x = 0 for each x, as desired.
In the case that x 0, we note that 0 < ex 1, and the same argument applies
here as well. Since limk Rk (x) = 0 for each x, we have that f (x) = ex is
equal to its Taylor series.
Example 13.7. WePcan pass between functions and series in the following
n
way: a power series n=0 aP n (x a) defines a function on its interval of con-
vergence I, namely f (x) = n=0 an (x a)n for every x I (since the interval
of convergence is precisely the set of values of x for which the power series
converges).
We can then take the Taylor series for f at any point a in the interval of
convergence I. When we do this, we get back the power series we started with,
since power series representations of functions are unique.
Calculating one finds that f (n) (0) = 0 for all n 0, and so the Taylor series
for this function centred at 0 is
X 1 (n) X
f (0)xn = 0 xn ,
n=0
n! n=0
One way would be to calculate the Taylor series for f (x) centred at a = 0,
but this gets complicated, as the derivatives of f (x) get complicated. Another
way is to use the arithmetic of power series. We start P by deriving a series
1
representation for 1/(x + 2), using the fact that 1r = n=0 rn for |r| < 1.
MATH2039 ANALYSIS 91
In this section we will apply what we have learned to understanding the function
ex . To begin with we need to define the function ex .
Let c be any constant greater than 1. Then we can define a function cx . If
x Z this is defined algebraically:
c c c(x times)
x>0
x
c = 1 x=0
1
ccc (x times) x<0
hence ak (c 1)/k. On the other hand since 1/k > 0 we know that c1/k > 1,
so we also know that ak 0. It follows from the squeeze rule for sequences
that ak 0, so c1/k = 1 + ak 1 as k .
Why does this definition make sense? We use the following lemma.
MATH2039 ANALYSIS 93
Proof. First we must show that cxn converges. Since xn is increasing, we know
that cxn must also be an increasing sequence. If s is any rational number greater
than x, then s > xn for all n, so cxn cs for all n. The sequence cxn is bounded
and monotonic, therefore it converges.
To justify this we will show that if we picked any other sequence of rational
numbers yn converging to x then the limit of cyn would be the same as the limit
of cxn , so cx is in fact the limit of cyn for any rational sequence tending to x.
The proof is as follows.
We must now show that for any sequence yn with yn x as n , we
have limn cxn = limn cyn . The difference yn xn converges to 0, so by
definition, for any k we can find M such that if n > M then |yn xn | < 1/k.
Thus we have c1/k < cyn xn < c1/k for n > M . However by the previous
lemma we know that c1/k 1 and so c1/k = 1/c1/k also tends to 1. It
follows from the squeeze rule for sequences, that cyn /cxn = cyn xn 1. Hence
limn cyn = limn cxn .
We must show for all x0 , that given > 0, there exists > 0 such that
|x x0 | < implies |cx cx0 | < . Write
cx cx0 = (cxx0 1)cx0 .
We know that the sequences c1/k and c1/k tend to 1 as k , therefore there
exists k such that |c1/k 1| and |c1/k 1| are both less than cx0 .
Take = 1/k. If |x x0 | < , then 1/k < x x0 < 1/k, so c1/k < cxx0 <
c1/k . This implies
x0 < c1/k 1 < cxx0 1 < c1/k 1 < x0 .
c c
Multiplying through by cx0 we get
< (cxx0 1)cx0 = cx cx0
as required.
We will now show that the function cx is differentiable. We must show that
x+h x
the limit of c hc as h tends to 0 exists. A first simplification is to write this
h h
as cx c h1 . It is therefore sufficient to show that c h1 converges to some limit
L as h tends to 0, in other words that the function is differentiable at 0. Note
d
that once we have shown this we will have shown that dx (cx ) = Lcx where the
ch 1
constant L = limh0 h is the derivative at 0.
ch 1
Consider the two sides of the limit separately limh0+ h and
ch 1 c|h| 1
We know that c|h| tends to c0 = 1 as h 0, so limh0 h = limh0 |h| .
h
But this is the same as limh0+ c h1 . In other words, if we can show that
h h
limh0+ c h1 exists, then so does limh0 c h1 , and the two are equal as re-
quired.
ch 1
Lemma 14.6. The limit limh0+ h exists.
Proof. We have h positive, and are considering the limit as h tends to zero. We
can therefore assume that h is less than 1, in which case we can find k such
1
that k+1 h k1 . Then
ch 1
k(c1/(k+1) 1) (k + 1)(c1/k 1).
h
Let ak = c1/k 1 as above, and let bk = kak . The sequences (k + 1)(c1/k 1)
is equal to k+1k bk , so this converges if and only if bk converges. Similarly
1/(k+1) k
k(c 1) = k+1 bk+1 , which converges if and only if bk+1 converges, or
equivalently bk converges.
Pk k n Pk
Write the sum in the form n=0 a = n=0 bk,n (kak )n where the coeffi-
n k
cients bk,n are given by the formula
n
k 1 k! k(k 1)(k 2) . . . (k (n 1))
bk,n = = =
n k n!(k n)!k n n!k k k . . . k
1 1 2 n1
= 1 1 ... 1 .
n! k k k
Note that for each n, as k increases bk,n increases, and tends to 1/n! as k .
Pk
The sum n=0 bk,n (kak )n is a constant (for each k it adds up to c), the number
of terms increases, and for each n the coefficient of (kak )n increases. Therefore,
so that the sum can remain constant, kak must be decreasing. As it is bounded
below it converges, which completes the proof.
We conclude that for any c > 1 the function cx is well defined for all real values
x, and is differentiable with derivative Lcx where L = limk kak equals the
derivative of cx at 0. Well now pick a special value of c, namely c = e. This
value is suppose to have the property that ex has derivative ex . Equivalently
limk k(e1/k 1) = 1.
(Note: Right away we see that this has something to do with the above com-
putation of derivatives of cx .)
We noted, in the proof of the previous lemma, that the terms bk,n are increasing
k
and are bounded
P 1 above by 1/n!. Hence (1 + 1/k) is increasing and bounded
above by n=0 n! . This sum is a finite number: we know the series converges
by the ratio test. Therefore the sequence (1 + 1/k)k converges.
P 1
The above argument shows us more: in fact the sequence converges to n=0 n! .
P 1 Pp 1
P 1
Given > 0, for p sufficiently large, n=0 n! n=0 n! = n=p+1 n! < /2.
On the other
Pp hand for P each n, the term bk,n 1/n! as k , so the fi-
p 1
nite sum n=0 bk,n n=0 n! . Hence for k sufficiently large (and at least
Pp 1
Pp
p) we have n=0 n! n=0 bk,n < /2. Putting all of this together we find that
96 PERISIC
k p
X 1 X X 1 X
bk,n bk,n
n=0
n! n=0 n=0
n! n=0
p p p
! !
X 1 X 1 X 1 X
= + bk,n < .
n=0
n! n=0 n! n=0
n! n=0
Pk P 1
Therefore (1 + 1/k)k = n=0 bk,n converges to n=0 n! . In other words
X 1
e = lim (1 + 1/k)k = .
k
n=0
n!
d x
Proof. We have already seen that dx e must be of the form Lex where L =
1/k
limk kak and ak = e 1. We will show that this limit is 1.
Recall we have the identity
k
X
k
e = (1 + ak ) = bk,n (kak )n
n=0
k 1 n
where bk,n = . On the other hand, by definition e is the limit of the
n k
Pk
increasing sequence (1 + 1/k)k = n=0 bk,n as k tends to infinity. We have
k
X k
X k
X
bk,n ((kak )n 1) = bk,n (kak )n bk,n = e (1 + 1/k)k
n=0 n=0 n=0
k
and we know that e (1 + 1/k) is a positive, decreasing, and tends to 0 as
k . Thus kak > 1 for all k, so each term of the sum is positive. Moreover
as the sum tends to zero, each individual term bk,n ((kak )n
1) must also tend
k
to 0 as k . In particular setting n = 1, we have bk,1 = (1/k)1 = 1, for
1
each k, so kak 1 = bk,1 ((kak )1 1) 0.
Hence we have shown that for ak = e1/k 1 the limit of kak as k is 1, so
the derivative of ex is ex .
MATH2039 ANALYSIS 97
Proof. Since
n
t
X t
e =
n=0
n!
tm
we have that et > m! for every m > 0. Let > 0 and choose m > . Then
et tm t
>
t m!
Thus limt et /t = .
tx+1
and use that et 0 as t we have that
tx1 1
t
< 2
e t
for all large enough t. Thus
Z Z
1
tx1 et <dt
1 1 t2
Given that the integral on the right of the above inequality converges, the
comparison test implies that the same is true for the integral on the left hand
side of this inequality.
We conclude that the integral defining (x) converges for all x > 0.
We shall now prove that the function generalises the factorial in the following
sense.
Theorem 15.2. For all natural n,
(n) = (n 1)!
det
Z
(x) = tx1 dt
0 dt
Z
= [tx1 et ]
0 + (x 1) tx2 et dt
0
We note here that the first of these two terms is to be understood as follows:
[tx1 et ]
0 = ( lim t
x1 t
e lim tx1 et ) = 0
t t0
R x2 t
We also note that 0 t e dt = (x 1) and so our calculation shows that
(x) = (x 1)(x 1)
Our theorem will follow by induction once we establish that (1) = 0! = 1
which can be seen by a simple calculation:
Z
(1) = et dt = ( lim et 1) = 1 = 0!
0 t
Solution 2.12: We begin by showing that the natural numbers N = {0, 1, 2, 3...}
lie in our ordered field F : We have seen in the lectures that 0 < 1 and hence
1 + 1 F and also 1 = 0 + 1 < 1 + 1. Therefore we can conclude that 1 6= 1 + 1.
Also, by transitivity, 0 < 1 + 1 and hence 0 6= 1 + 1. Continuing this process
we obtain elements 1 + 1 + ... + 1 F , such that for all non-negative inte-
| {z }
n
gers n 6= m we have that 1 + 1 + ... + 1 6= 1 + 1 + ... + 1 . Now rename these
| {z } | {z }
n m
elements as follows: 1 + 1 + ... + 1 =: n and we have shown that
| {z }
n
{0, 1, 2, ...} F.
MATH2039 ANALYSIS 101
Now, by the field axioms it follows that for each n F , also n = (1 + ... + 1) =
| {z }
n
(1) + ...(1) F and by the ordering axioms n < 0 and an argument
| {z }
n
similar to above shows that for n 6= m also n 6= m. Furthermore, for all
n, m N\{0} we have n 6= m as n < 0 and m > 0. Hence
Z = {..., 2, 1, 0, 1, 2, ...} F.
n
Each element in Q is of the form m with n, m Z. By the field axioms, we
have m1 F and hence n m1 = m n
also lies in F . Hence we can conclude
that Q F.
Solution 2.13: Suppose there were such an order on C, and denote it by <.
Compare 0 and i. Since 0 6= i, it must be that either 0 < i or i < 0.
Suppose that 0 < i. Multiplying both sides by i and remembering that 0 < i,
we see that 0 i < i i, which simplifies to 0 < 1. Adding 1 to both sides, we
see that 1 < 0. Again multiplying both sides by i and remembering that 0 < i,
we see that 1 i < 0 i, which simplifies to i < 0. Hence, if 0 < i, then i < 0,
contradicting the second condition in the definition of an order.
Suppose now that i < 0. Adding the additive inverse i of i to both sides, we get
that 0 < i. Multiplying both sides by i, we get that 0 (i) < (i) (i), and
so 0 < 1. Multiplying both sides by i again, we get that 0 < (1) (i) = i.
Hence, if i < 0, then then 0 > i, again contradicting the second condition in
the definition of an order.
Hence, since we have that neither 0 < i nor i < 0, we see that there cannot
exist an order on C that makes C into an ordered field.
102 PERISIC
Solution 3.13:
(1) Bounded above by 1 (since for n Z {0}, either n 1 in which case
1 1
n 1, or n 1, in which case n 0), and so has a supremum. Again
making use of Exercise 3.16, since 1 is an upper bound for S and since
1 S, 1 = sup(S). In this case, sup(S) S.
Since 2x > x for positive integers x, given any C > 0 we can find an
x so that 2x > C, and so there is no upper bound. That is, S is not
bounded above.
Since x and 2y are both positive for x, y N, we have that 2xy > 0 for all
x, y N. Therefore, S is bounded below by 0, and so has an infimum.
Considering the subset of S in which x = 1, we have that S contains
1
2y for all y N. In particular, for each > 0, we can find y N so
that 21y < , namely take log2 of both sides to get y < log2 (), or
equivalently y > log2 (). Hence, there is no positive lower bound, and
so 0 = inf(S). Since 2xy is never 0 for x, y > 0, in this case inf(S) 6 S.
MATH2039 ANALYSIS 103
(5) Break S up into two subsets, one of the positive terms (when n is even)
and the negative terms (when n is odd). So, S = {2, 34 , 65 , . . .}
{ 32 , 45 , 67 , . . .}.
The positive terms are all of the form 1 + n1 where n is even. Since n1
decreases as n increases, the largest positive term is 1 + 21 = 32 , and so
S is bounded above and hence has a supremum. Since S is bounded
above by 32 and since 32 S, sup(S) = 32 , and in this case sup(S) S.
The negative terms are all of the form 1 + n1 where n is odd. Since
1 1
n decreases as n increases, n increases as n increases, and so the
11
smallest negative term is 1 = 2, and so S is bounded below and
hence has an infimum. Since S is bounded below by 2 and since
2 S, inf(S) = 2, and in this case inf(S) S.
Solution 3.16:
(1) The easiest way to do this is to begin with an intermediate fact: if
A B and if sup(B) exists, then sup(A) exists and sup(A) sup(B).
The proof uses the definition of supremum: since sup(B) exists, we
have that b sup(B) for all b B and that if u is an upper bound for
B, then sup(B) u. Since b sup(B) for all b B and since A B,
we have that a sup(B) for all a A. In particular, A is bounded
above, and so sup(A) exists. To see the second statement, note that
since sup(B) is an upper bound for A, we have that sup(A) sup(B)
by definition.
that b inf(B) for all b B and that if t is a lower bound for B, then
inf(B) t. Since b inf(B) for all b B and since A B, we have
that a inf(B) for all a A. In particular, A is bounded below, and
so inf(A) exists. To see the second statement, note that since inf(B) is
a lower bound for A, we have that inf(A) inf(B) by definition.
Solution 3.17:
(1) Since sup(A) exists, the set A is bounded above. Let u be any upper
bound for A, so that a u for all a A. Multiplying through by
1, this becomes a u for all a A. Since a ranges over all of
A as a ranges over A, this yields that u is a lower bound for A ,
and so inf(A ) exists. In particular, taking u = sup(A), we have that
sup(A) is a lower bound for A .
To see that there is no lower bound for A that is greater than sup(A),
note that t is a lower bound for A if and only if t is an upper
bound for A. Therefore, a lower bound for A greater than sup(A)
exists if and only if an upper bound for A less than sup(A) exists,
but by the definition of supremum no such upper bound can exist.
Hence, sup(A) is the greatest lower bound for A , or in other words,
sup(A) = inf(A ), as desired.
(2) Since inf(A) exists, the set A is bounded below. Let t be any lower
bound for A, so that a t for all a A. Multiplying through by 1,
this becomes a t for all a A. Since a ranges over all of A as
a ranges over A, this yields that t is an upper bound for A , and so
sup(A ) exists. In particular, taking t = inf(A), we have that inf(A)
is an upper bound for A .
To see that there is no upper bound for A that is less than inf(A),
note that u is an upper bound for A if and only if u is a lower bound
for A. Therefore, an upper bound for A less than inf(A) exists if
and only if a lower bound for A greater than inf(A) exists, but by the
definition of infimum no such lower bound can exist. Hence, inf(A) is
the least upper bound for A , or in other words, inf(A) = sup(A ),
as desired.
Solution 3.18: [Note that each of these exercises has many, many possible
solutions. And yes, it is a very silly question.]
(1) Take S = {x R | x > 2}, so that inf(S) = 2, which is irrational,
and S is also bounded below by 0, which is rational. (In fact, any set of
real numbers that is bounded below has both infinitely many rational
lower bounds and infinitely many irrational lower bounds.)
(2) Take S = (0, ), so that inf(S) = 0, which is rational, and S is bounded
below by 1, which is also rational.
(3) Take S = (2, 4), so that inf(S) = 2, which is rational, and S is also
bounded below by , which is irrational.
(4) Take S = ( 3, ), so that
inf(S) = 3, which is irrational, and S is
also bounded below by 2, which is also irrational.
106 PERISIC
Solution 4.20: We know that 2 = 1.414213562... is irrational. Look at the
sequence a0 = 1, a1 = 1.4, a2 = 1.41, a3 = 1.414, a4 = 1.4142 etc. Each term
in the sequence lies in Q. The sequence is bounded above (by 1.5 for example)
and is increasing. However
it does not tend to a limit in Q. (It does tend to a
limit in R, namely 2, but this does not lie in Q.)
Solution 4.35:
3(1)1 2
u1 = 4(1)5 = 1 = 2;
3(5)1 14
u5 = 4(5)5 = 15 0.9333;
3(10)1 29
u10 = 4(10)5 = 35 0.8286;
3(100)1 299
u100 = 4(100)5 = 395 .7570;
3(1000)1 2999
u1000 = 4(1000)5 = 3995 0.7507;
3(10000)1 29999
u10000 = 4(10000)5 = 39995 0.7501;
3(100000)1 299999
u100000 = 4(100000)5 = 399995 0.7500;
So, it seems that a reasonable guess would be that L = limn un exists and
equals 0.75 = 43 . To verify this, we use the definition: we need to show that for
any choice of > 0, we can find n0 so that |un L| < for all n > n0 .
To find the value of n0 so that |un L| < for n > n0 , we start by solving for
11
n: since 4(4n5) < , we have that 11 11 5
4 < 4n 5, and so 16 + 4 < n. That is,
11 5 11+20
for a specified value of , we can take n0 = b 16 + 4 c = b 16 c. Then, for
any choice of > 0, we set n0 = b 11+20
16 c, and then if we take n > n0 , working
backwards we have that |un L| < .
n
1+210 2
Solution 4.37: Set an = 5+310 n and L = 3 . For each choice of > 0, we
need to show that there exists n0 so that |an L| < for all n > n0 .
7
Hence, for a given value of > 0, we want to find n0 so that 15+910n < for
7
n > n0 . So, we solve for n in terms of . First, note that 15+910n > 0 for all
7
positive integers n. So, we need only solve 15+910n < for n.
To get a specific value of n0 so that |an L| < 103 for n > M , we substitute
3
= 103 into the above equation to get that n > log10 15109103
+7
2.8899.
So, we can take n0 = 2.
1
Solution 4.38: We start with the first part of the inequality, that n+1 <
n+1 x+1 1
ln(n + 1) ln(n) = ln n . Set f (x) = ln x x+1 and bn = f (n). We
want to show that f (x) > 0 for all x 1. Calculating, we see that f 0 (x) =
1
x(x+1) 2 < 0 for all x > 0. This implies that f (x) is decreasing, and hence
1 n+1
Since n > ln n = ln(n + 1) ln(n) for all n, we have that
n
! n
!
X 1 X
an = ln(n) > ln(i + 1) ln(i) ln(n) = ln(n + 1) ln(n) > 0
i=1
i i=1
Since an is bounded above (an < a1 for all n, since it is a monotonically decreas-
ing sequence) and bounded below, it is bounded. Since it is also monotonic, we
have that an converges.
Solution 4.39: (This is an exercise in writing out the definition of the conver-
gence or divergence of a sequence for a triple of specific examples. Note that
we are not asked to determine whether the given statements are true or false,
or to prove them if they are true, but just to write them down.)
for every > 0, there exists n0 so that 32n1 > for all n > n0 .
for every > 0, there exists n0 so that 1 2n < for all n > n0 .
for every > 0, there exists n0 so that |en 0| < for all n > n0 .
Solution 4.51: [Please note that some solutions require knowledge from Cal-
culus, which we havent proved yet. So, please use with caution.]
(1) converges: whenever we are evaluating a limit in which the variable
(in this case n) appears in both the base and the exponent, we follow
the same basic procedure. First use the identity x = exp(ln(x)) to
rewrite the term. Here,
1/n ln(n + 2)
an = (n + 2) = exp .
n
Next, we check to see whether we are dealing with an indeterminate
form. Since the limit limn ln(n+2)
n has the indeterminate form
,
we may use lHopitals rule to evaluate
ln(n + 2) 1
lim = lim = 0.
n n n n + 2
Hence, an converges to e0 = 1.
(2) converges: there is a standard way of evaluating the limit as n
of a rational function in n (where a rational function is the quotient
of two polynomials). First, locate the highest power of n that appears
in either the numerator or the denominator, and then multiply both
numerator and denominator by its reciprocal. Here, the highest power
of n that appears is n3 , and so we calculate
1 1
n2 + 3n + 2 n2 + 3n + 2 n3 n + n32 + 2
n3
an = = 1 = .
6n3 + 5 6n3 + 5 n3 6 + n53
We then use several properties of limits: that the limit of a quotient
is the quotient of the limits, that the limit of a sum is the sum of the
limits, and that limn n1 = 0. Here,
1
n + n32 + 2
n3 0
lim an = lim = = 0.
n n 6 + n53 6
Hence, an converges to 0.
MATH2039 ANALYSIS 109
Hence, an converges to e1 = e.
(4) converges: here we use the squeeze law. Since 1 sin(n) 1 for
all n, we have that 31n sin(n)
3n 31n . Since limn 31n = 0, we have
1
that limn 3n = 0 as well, and so an converges to 0.
(5) diverges: write
2n + 3 + n + 1 n+2
an = ( 2n + 3 n + 1) = .
2n + 3 + n + 1 2n + 3 + n + 1
We now massage algebraically, in order to simplify:
n + 32 + 12 n + 32
r
n+2 n+2 1 3
= q > q = n+ .
2n + 3 + n + 1 2 2n + 3 2 2(n + 3 ) 2 2(n + 3 ) 2 2 2
2 2
q
3
Since limn n+ 2 = , we see that limn an = , and so an
diverges.
8k
(6) diverges: for
n = 8k, a8k = cos 4 = 1, while for n = 8k + 1,
(8k+1)
a8k+1 = cos 4 = 2 . In particular, |a8k a8k+1 | = 12 , and so
1
1
lim ln(1 + ) = 0,
n n
1
ln(1+ n )
we have that limn n = 0 (by the squeeze law for instance,
1
ln(1+ n ) 1
since 0 n ln(1 + n ) for n 1). Hence,
1
ln(1 + n)
lim exp = e0 = 1,
n n
and so an converges to 1.
(8) diverges: given > 0, we show that there exists M so that an >
for n > M . Since an = ln(n), this becomes ln(n) > for n > M .
Exponentiating both sides of ln(n) > , we get that n > e (and vice
110 PERISIC
(15) converges: using the standard trick for rational functions, write
1
2n 2n n 2
an = = 1 = 3 .
5n 3 5n 3 n 5 n
1
As n , n 0 and so an converges to 25 .
(16) converges: using the standard trick for rational functions, write
1 1
1 n2 1 n2 n2 n2 1
an = = 1 = 2 .
2 + 3n2 2 + 3n2 n2 n2 +3
1
As n , n2 0 and so an converges to 13 .
(17) converges: using the standard trick for rational functions, write
1
n3 n + 7 n3 n + 7 n3 1 n12 + 7
n3
an = = 1 = .
2n3 + n2 2n3 + n2 n3 2 + n1
1 1
As n , both n2 0 and n 0, and so an converges to 21 .
9 n
(18) converges: by a previous part of this exercise, we know that ( 10 )
9 9 n
converges to 0, since | 10 | < 1, and so limn (1 + ( 10 ) ) = 1 +
9 n
limn ( 10 ) = 1.
(19) converges: by a previous part of this exercise, we know that ( 21 )n
converges to 0, since | 12 | < 1, and so limn (2 ( 21 )n ) = 2
limn ( 12 )n = 2.
(20) diverges: for n even, an = 2, while for n odd, an = 0. In particular,
|an an+1 | = 2 for all n, and so the sequence fails the Cauchy criterion
and hence diverges.
(21) converges: note that 0 1 + (1)n 2 for all n, and so the squeeze
law yields that since limn n2 = 0, we have that limn an = 0.
(22) converges: we begin by noting that
1 + (1)n n 2 n
0 ,
( 32 )n ( 32 )n
2 n
and so well concentrate on evaluating limn ( 32 )n
and hope to be able
2 n
to apply the squeeze law. Since limn ( 3 )n has the indeterminate
2
form
, we may use lHopitals rule to evaluate
1
2 n n 1
lim 3 n = lim = lim =0
n ( ) n ln( 3 ) exp(n ln( 3 )) n ln( 3 ) n( 3 )n
2 2 2 2 2
p
(24) converges: since 1 2 + cos(n) 3 for all n and since n1 0 as
q
n , the squeeze law yields that 2+cos(n)
n 0 as n . That
is, an converges to 0.
(25) converges: since sin(n) = 0 for all integers n, this sequence is the
constant sequence an = n 0 = 0 for all n. In particular, an converges
to 0.
(26) diverges: since cos(n) = (1)n , this sequence can be rewritten as
an = (1)n n. For n 1, |an+1 an | 2, and so the sequences fails
the Cauchy criterion, and so diverges.
(27) converges: since 1 sin(n) 1 for all n, we have that n1
sin(n)
n n1 for all n, and so sin(n)
n converges to 0. Hence, an converges
0
to = 1.
(28) diverges: for n even, cos(n) = 1 and for n odd, cos(n) = 1. In
particular, |an+1 an | = |21 21 | = 32 for all n, and so this sequences
fails the Cauchy criterion, and hence an diverges.
(29) converges: we could use lHopitals rule, since limn ln(2n)
ln(3n) has the
indeterminate form , but we proceed in a more low tech way. Use
the laws of logarithms and a variant of the standard trick for rational
functions, we rewrite
1 ln(2)
ln(2n) ln(2) + ln(n) ln(2) + ln(n) ln(n) 1+ ln(n)
an = = = 1 = ln(3)
.
ln(3n) ln(3) + ln(n) ln(3) + ln(n) ln(n) 1+ ln(n)
ln(2) ln(3)
Since ln(n) as n , we have that both ln(n) and ln(n) go to 0
as n , and so limn an = 1.
ln2 (n)
(30) converges: since limn n has the indeterminate form , we can
use lHopitals rule:
ln2 (n) 2 ln(n) n1 2 ln(n)
lim = lim = lim .
n n n 1 n n
This limit still has the indeterminate form , and we can apply lHopitals
rule again to get
2
2 ln(n)
lim = lim n = 0.
n n n 1
Hence, an converges to 0.
(31) converges: write
sin( n1 )
1
an = n sin = 1 .
n n
Since limn an has the indeterminate form 00 , we can apply lHopitals
rule to get
sin n1 cos n1 n12
1
lim 1 = lim 1 = lim cos = cos(0) = 1.
n
n
n n2 n n
MATH2039 ANALYSIS 113
(37) converges: as with all limits in which the variable appears in both
the base and the exponent, we begin by rewriting using the identity
m = exp(ln(m)) to get
n n n
n1 n+12 2 2
an = = = 1 = exp n ln 1 .
n+1 n+1 n+1 n+1
Since the exponent has the indeterminate form 0 as n , we
rewrite it as
2
ln(1 n+1 )
2
n ln 1 = 1 ,
n+1 n
2
Hence, an converges to e .
(38) converges: since n1 0 as n , we see that an converges to
(0.001)0 = 1.
n+1 1
(39) converges: as n , n = 1+ n 1, and so an converges to
21 = 2.
(40) converges: one way to evaluate this limit is to write an = ( n2 )3/n =
23/n
n3/n
and to evaluate the limits of the numerator and denominator sep-
arately. To evaluate limn 23/n , all we need note is that limn n3 =
0, and so 23/n converges to 20 = 1.
(41) diverges: begin by ignoring the (1)n and worrying about what hap-
pens to the rest of the term. Using the standard trick, massage to get
2 2
(n2 + 1)1/n = exp( ln(nn+1) ). Since limn ln(nn+1) has the indetermi-
nate form , we may use lHopitals rule to evaluate
2n
ln(n2 + 1) n2 +1
lim = lim = 0,
n n n 1
MATH2039 ANALYSIS 115
and so
ln(n2 + 1)
lim exp = e0 = 1.
n n
So, putting the (1)n back into the picture, we see that an fails the
n2 +1
Cauchy criterion: specifically,
2 since
n converges to 1, for any > 0,
there exists M so that n 1 < for n > M . Choose = 21 , and
n +1
note that for n > M , we get that |an an+1 | > 1, since one of an ,
an+1 is within 12 of 1 and the other is within 12 of 1 (remember the
alternating signs). So, an diverges.
(42) converges: we perform a bit of algebraic massage: note that
2 n 2 n n
3 3 20
an = 1 n 9 n
< 9 n =
.
2 + 10 10
27
n
Since 20
27 0 as n (since 20
27 < 1), the squeeze rule yields that
an converges to 0 as well.
Solution 4.52: Suppose that qn converges and set x = limn qn . Now, note
that
an an1 + an2 an2 1
qn = = =1+ =1+ .
an1 an1 an1 qn1
Hence,
1 1 1
x = lim qn = lim 1 + =1+ =1+ ,
n n qn1 limn qn1 x
since limn qn1 = x as well. Therefore, x = 1 + x1 , and so (multiplying
2
x x 1 = 0.
through by x and simplifying) x satisfies the quadratic equation
1
By the quadratic formula, this yields that x = 2 1 5 . However, since
qn 0 for all n, it must be that x 0 as well, and so x = 21 1 + 5 .
Solution 4.54: In all three of these statements, we start with the same piece of
information, namely that limn xn = 4. That is, for each > 0, there exists
M (which depends on ) so that |xn (4)| = |xn + 4| < for n > M .
p
(1) we need to show that limn |xn | = 2, which is phrased mathemat-
ically
p as needing to show that for each > 0, therepexists P so that
| |xn | 2| < for n > P . We start by rewriting | |xn | 2|, using
the standard trick for handling differences of square roots, namely
p
p p | |xn | + 2| | |xn | 4| | |xn | 4|
| |xn | 2| = | |xn | 2| p = p .
| |xn | + 2| | |xn | + 2| 2
p
(The last inequality follows from the fact that | |xn | + 2| 2 for all
possible values of xn .) Since for any > 0, there exists M so that
| |xn | 4| < 2 (by using the definition of limn |xn | = 4) for n > M ,
we have that
p | |xn | 4| 2
| |xn | 2| < =
2 2
for n > M , and so we are done.
116 PERISIC
(2) we need to show that limn x2n = 16, which is phrased mathematically
as needing to show that for each > 0, there exists P so that |x2n 16| <
for n > P . We start by rewriting |x2n 16|, using that it is the
difference of two squares:
|x2n 16| = |(xn 4)(xn + 4)| = |xn 4| |xn + 4|.
Now apply the definition of limn xn = 4 with = 1, so that there
exists M so that if n > M , then |xn (4)| < 1. In particular, if n > M ,
then 5 < xn < 3, and so |xn | < 5, and so |xn 4| |xn | + 4 < 9.
Solution 4.55:
(1) since a > 0, we can apply the definition of limn an = a with = 12 a
to see that there exists P so that an > 0 for n > P (since the interval
of radius 12 a centred at a contains only positive numbers), and so for
n > P , an makes sense.
We need to get our hands on | an a|, which we do with our usual
trick for handling differences of square roots:
| an + a| |a a|
| an a| = | an a| = n .
| an + a| an + a
(Here were using
that both an> 0 and
a > 0 to say that | an +
a| = an + a.) Since an + a > a for n > P , we have that
|an a| |a a|
| an a| = < n
an + a a
for n > P . Since an converges
to a, for every > 0, we can choose
M > P so that |an a| < a for n > M . For this choice of M , we
have that
|an a| |a a| a
| an a| = < n < = ,
an + a a a
and so an converges to a.
MATH2039 ANALYSIS 117
(2) this one, we break into three cases. If a > 0, then (applying the def-
inition of limn an = a with = a) there exists M0 so that an > 0
for n > M0 . In this case, we have |an | = an for n > M0 and |a| = a,
and so ||an | |a|| = |an a|. Since there is M1 so that |an a| < for
n > M1 , we have that ||an | |a|| < for n > M = max(M0 , M1 ), and
so limn |an | = |a|.
Now consider the sequence (1)n an . For n > M and n even, (1)n an =
an lies in the interval centred at a with radius 12 |a|. For n > M and
n odd, (1)n an = an lies in the interval centred at a with radius
1
2 |a|. In particular, we have, regardless of whether n is odd or even,
that |(1)n an (1)n+1 an+1 | > |a| for n > M , since (1)n an and
(1)n+1 an+1 lie on opposite sides of 0 and are both distance at least
1 n
2 |a| from the origin. Hence, (1) an violates the Cauchy criterion (see
Theorem 4.25), and so diverges.
Solution 4.56: Since limn xn = x, we have that for each > 0, there exists
M so that |xn x| < 13 for n > M . For any m > 0 and n > M , we now have
that
|xn+1 + + xn+m mx| = |xn+1 x + + xn+m x|
|xn+1 x| + + |xn+m x|
1
m .
3
Dividing by n + m, we obtain that
1 m m 1 1
n + m (xn+1 + + xn+m ) n + m x n + m 3 < 3
m
(since n+m < 1). Viewing n as fixed for the moment, choose m so that both
| n+m x x| < 13 (which we can do since limm n+m
m m
= 1 for n fixed) and
1 1
n+m |x 1 + x 2 + + x n | < 3 (which we can do since x 1 + x2 + + xn is a
118 PERISIC
for all m > 0. Since this is true for all n > M and all m > 0, we have that
1
p (x1 + + xp ) x < for all p > M , as desired.
MATH2039 ANALYSIS 119
Solution 5.4:
(1) an = (1)n , bounded above by 1 and bounded below by 1, hence
bounded. This sequence fails the Cauchy criterion, since |an an+1 | = 2
for all n, and so diverges.
(2) sin(n), bounded above by 1 and bounded below by 1, hence bounded.
Though it seems fairly clear why this sequence diverges, the actual
proof is a bit subtle, and we do not give it here. If you are intrigued,
ask me after class, or come to my office hours.
(3) 0, 1, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 1, . . ., bounded above by 1 and bounded
below by 0, hence bounded. Arbitrarily far out in the sequence, there
are consecutive terms taking the values 0 and 1, so the sequence fails
the Cauchy criterion and hence diverges.
(4) an = the nth digit of , bounded above by 9 and bounded below by
0, hence bounded. Does not converge, because the only way for a
sequence of integers to converge is for it to be eventually constant,
that is, constant past some index, which in this case would then imply
that is a repeating decimal, hence a rational number, which it isnt.
(In fact, fixing an irrational number x and taking an to be the nth digit
of the decimal expansion of x gives a sequence that is bounded but not
convergent, by the same argument.)
(5) an = the nth digit of the rational number 17 = .142857, using the same
argument as above (which works for rational numbers, as long as the
length of the repeating section in the decimal expansion is longer than
one digit).
120 PERISIC
Solution 6.7:
(1) To show that hn (x) is continuous at a R, we need to show that for
each > 0, there exists > 0 so that if |xa| < , then |hn (x)hn (a)| <
. Since hn (x) = xn , this is the same as showing that for each > 0,
there exists > 0 so that if |x a| < , then |xn an | < . Lets break
the proof into cases.
Consider now the case that n 2 and a > 0, and factor |xn an | to get
|xn an | = |(x a)(xn1 + axn2 + + an2 x + an1 )|. Recall that
we have a great deal of choice in how we choose , so we may restrict
our attention to the interval |x a| < 21 a, so that 12 a < x < 32 a, by
requiring that < 21 a (which makes sense, since a > 0). Calculating,
we see that
|xn an | = |(x a)(xn1 + axn2 + + an2 x + an1 )|
|x a|(xn1 + axn2 + + an2 x + an1 )
n1 n2 !
3 3 n2 3 n1
< |x a| a +a a + + a a+a
2 2 2
n1
X 3 k
= |x a|an1
2
k=0
1 (3/2)n
= |x a|an1 = C|x a|,
1 (3/2)
n
where C = an1 1(3/2)
1(3/2) > 0 depends on both a > 0 and n 2. So,
take to be the smaller of C1 and 12 a. Then, for |x a| < , we have
that |xn an | < C|x a| as desired. (The first inequality follows
from the calculation above and the fact that |x a| < < 12 a, while
the second inequality follows from < C1 .)
Solution 6.11: First, for the sake of notational clarity, define the n-fold compo-
sition of f with itself by f n , so that f n = f f (n1) . The hypothesis can then
be restated as saying that the sequence {f n (c)} converges to a. Now, apply f
to both sides. Since f is continuous, the sequence {f (f n (c))} converges to f (a),
by Proposition 6.10. However, since f (f n (c)) = f f n (c) = f (n+1) (c), the
sequence {f (f n (c))} is the same as the sequence {f n (c)} with the first term
removed, and so {f (f n (c))} converges to a as well. Hence, since {f (f n (c))}
converges to both a and f (a), we have that a = f (a).
Solution 6.18: First, since limx (f (x + 1) f (x)) = 0, for any > 0, there
exists x0 (which we can take to be positive) so that |f (x + 1) f (x)| < 21 for
x > x0 . Now, using the maximum value property, Theorem 6.17, there exists a
maximum value M of |f (x)| on the interval [x0 , x0 + 1].
The first claim is that for any k 0, we have that |f (x)| 12 k + M for x in
the interval [x0 + k, x0 + k + 1]. To see this, let K be the maximum value of
|f (x)| on [x0 + k, x0 + k + 1], occurring at y. Then, x0 + k y x0 + k + 1, and
so x0 y k x0 + 1. We now engage in some algebraic manipulation:
|f (y)| = |f (y) f (y k) + f (y k)|
|f (y) f (y k)| + |f (y k)|
|f (y) f (y 1) f (y k + 1) f (y k)| + |f (y k)|
1 1 1
+ + + + M
2 2 2
k
+ M.
2
Now, choose x1 > x0 so that xM1 < 12 . Then, for all y > x1 , we have that
f (y) |f (y)| 1 M 1 1
y = y < 2 + y < 2 + 2
f (x)
for all y > x1 . In particular, we have that the definition of limx x = 0 is
satisfied, as desired.
Hence, f (x0 ) f (x) for all x in [a, b], and so f (x0 ) f (x) for all x in [a, b].
That is, f satisfies the minimum value property.
Solution 6.22:
(1) Missing
(2) Again missing :(
(3) for f (x) = x1995 + 7654x123 + x on the closed interval [a, a], start
by verifying continuity; actually, f is continuous on all of R being a
polynomial, and hence is continuous on [a, a]. Now, check the sign of
f on the endpoints of the given interval: f (a) = a1995 +7654a123 +a > 0
(since a > 0) and f (a) = (a)1995 +7654(a)123 +(a) = f (a) < 0,
and so the intermediate value property implies that there exists some c
in (a, a) with f (c) = 0. (And actually, casual inspection reveals that
f (0) = 0.)
(4) for tan(x) = ex for x in [1, 1], start by defining g(x) = tan(x) ex ,
so that tan(c) = ec if and only if g(c) = 0, as was done above. Note
that g is continuous on [1, 1], since ex is continuous on all of R
and tan(x) is continuous as long as its denominator cos(x) is non-zero,
which holds true on [1, 1]. Since we are working on the closed interval
[1, 1], check the values of g on the endpoints: g(1) = tan(1) e1 =
1.1895... > 0 and g(1) = 4.2757... < 0, and so there exists some c in
(1, 1) with g(c) = 0, and hence with tan(c) = ec .
(5) as above, f (x) = x3 + 2x5 + (1 + x2 )2 is continuous on [1, 1], as it
is the sum of a polynomial and a rational function whose denominator
is non-zero on [1, 1]. As always, check the endpoints of the interval
first: f (1) = 13 11
4 and f (1) = 4 , and so by the intermediate value
property, there is some c in (1, 1) at which f (c) = 0.
(6) consider f (x) = 3 sin2 (x) 2 cos3 (x). Since both sin(x) and cos(x) are
continuous on all of R, we have that f is continuous on all of R. Since no
specific closed interval is given, we need to find an appropriate interval
on which to apply the intermediate value property for f , if in fact such
an interval exists. Fortunately, we remember that sin(k) = 0 for all
integers k, and so we may consider the interval [k, (k + 1)] for any
integer k 1, so that the interval lies in (0, ). At the endpoints of
this interval, f (k) = 2 cos3 (k) and f ((k + 1)) = 2 cos3 ((k + 1)).
Since cos(k) and cos((k+1)) are equal to 1 and have opposite signs,
f (k) and f ((k+1)) are both non-zero and have opposite signs, and so
by the intermediate value property, there is a point ck in (k, (k + 1))
at which f (ck ) = 0, that is, at which 3 sin2 (ck ) = 2 cos3 (ck ), as desired.
(7) first, note that f (x) = 3 + x5 1001x2 is a polynomial and so is contin-
uous on all of R, and in particular is continuous for x > 0. As above,
we need to choose a closed interval on which to apply the intermedi-
ate value property. Lets start by evaluating f at some of the natural
numbers: f (1) = 997; f (2) = 3969; f (10) = 90097; f (11) = 880.
MATH2039 ANALYSIS 123
Solution 7.7: First, note that f is continuous on [1, 4], as it is the composi-
tion of two continuous functions, namely absolute value and a linear polyno-
mial. However, f is not differentiable at x = 2 (since absolute value is not
differentiable at 0), and so the hypotheses of the mean value theorem are not
satisfied.
To see that f does not satisfy the conclusion of the mean value theorem, we
calculate: f (4) f (1) = |4 2| |1 2| = 2 1 = 1 and 4 1 = 3. However,
for x > 2, we have that f 0 (x) = 1 and for x < 2 we have that f 0 (x) = 1, and
so there cannot be a point c in (1, 4) at which f 0 (c) = (f (4) f (1))/(4 1) =
1/3.
Solution 7.9:
(1) This proof follows the same general outline as the proof just given.
Suppose that g 0 (x) = an1 xn1 +an2 xn2 + +a1 x+a0 , and consider
the new function h(x) = n1 an1 xn + n1
1
an2 xn1 + + 12 a1 x2 +a0 x
g(x). Note that since g and polynomials are differentiable, and hence
continuous, on all of R, we have that h is differentiable, and hence
continuous, on all of R. Also, h0 (x) = an1 xn1 + an2 xn2 + +
a1 x + a0 g 0 (x) = 0 for all x R.
For x0 > 0, apply the mean value theorem to h on the interval [0, x0 ].
Since h is continuous on [0, x0 ] and differentiable on (0, x0 ), the mean
value theorem yields that there exists some c in (0, x0 ) so that h(x0 )
h(0) = h0 (c)(x0 0) = 0, since h0 (c) = 0. That is, h(x0 ) = h(0) for
all x0 > 0. As above, we also get that h(x0 ) = h(0) for all x0 < 0 by
applying the mean value theorem to h on the interval [x0 , 0].
Set h(x) = x ln(x + 1), and note that h is differentiable, and hence
continuous, on (1, ). The two cases, of 1 < x < 0 and of x > 0, are
handled in the same fashion, and we write out the details only for the
case x > 0. Apply the mean value theorem to h on any closed interval
in [0, ). Note that h(0) = 0 ln(1) = 0. If there were another point
x0 > 0 at which h(x0 ) = 0, then by applying either Rolles theorem or
the mean value theorem to h on the interval [0, x0 ], there would exist a
point c in (0, x0 ) at which h0 (c) = 0. However, h0 (c) = 1 c+1
1
, which
is non-zero for c 6= 0. Hence, h(x) 6= 0 for all x (0, ). By the
MATH2039 ANALYSIS 125
intermediate value theorem, this forces either h(x) > 0 for all x > 0 or
h(x) < 0 for all x > 0 (because if there are points a and b in (0, ) at
which h(a) > 0 and h(b) < 0, then there is a point c between a and b at
which h(c) = 0). Since h(1) = 1 ln(2) = 0.3069... > 0, we have that
h(x) > 0 on (0, ), that is, that x > ln(x + 1) for all x > 0, as desired.
(As noted above, the argument to show that h(x) > 0 for 1 < x < 0,
or equivalently that x > ln(x + 1) for 1 < x < 0, is similar, and is left
for you to write out.)
x
For the other inequality, set g(x) = ln(x + 1) x+1 , and note that g
is differentiable, and hence continuous, for x > 1. (As above, we give
the details in the case that x > 0, and leave the case of 1 < x < 0 to
you the reader.) Note that g 0 (x) = (x+1) x
2 > 0 for x > 0. In particular,
Solution 7.11:
(1) we know that there is one solution to f (x) = 0 in [a, a], namely
x = 0 (which can be found with using the intermediate value theorem
or by inspection). To see that there are no others, we again use Rolles
theorem: if there were b in [a, a], b 6= 0, with f (b) = 0, then there
would exist some point c between b and 0 with f 0 (c) = 0. However,
f 0 (x) = 1995x1994 + 941442x122 + 1 and so f 0 (c) 1 > 0 for all c R.
Hence, by Rolles theorem, there is no second solution to f (x) = 0.
(2) again working with g(x) = tan(x) ex , we saw earlier that there is
a solution to g(x) = 0 in the interval [1, 1]. However, since g 0 (x) =
sec2 (x) + ex > 0 for all x (1, 1), Rolles theorem implies that there
can be no second solution to g(x) = 0 in the interval [1, 1]. (It is the
same reasoning as before: if there were two solutions to g(x) = 0, then
there would exist a point c between them at which g 0 (c) = 0; however,
the calculation above shows that g 0 (c) 6= 0 for all c in (1, 1))
(3) we dont have enough information to decide whether weve found all the
solutions to f (x) = 0. With f (x) = 3 sin2 (x) 2 cos3 (x), we have that
f 0 (x) = 6 sin(x) cos(x) + 6 cos2 (x) sin(x) = 6 sin(x) cos(x)(1 + cos(x)) =
0 when x = k for k N (since sin(k) = 0) and when x = (k + 12 )
(since cos((k + 21 )) = 0 for k N). Note that f (k) = 2 cos3 (k) =
(1)k+1 2 6= 0 and that f ((k + 21 )) = 3 sin2 ((k + 21 )) = 3 6= 0. So, for
any m N, consider the interval (m, (m + 2)).
So, there exist three points in this interval at which f 0 (x) = 0, namely
at (m + 12 ), (m + 1), and (m + 32 ), and our earlier analysis using
the intermediate value theorem found only two points in this interval at
which f (x) = 0. However, while Rolles theorem yields that two points
126 PERISIC
To see that there is a root, we note that since h has odd degree (and
since the coefficient of the highest degree term is positive), we have
that limx h(x) = and limx h(x) = . Hence, we can find
a point a at which h(a) > 0 and a point b at which h(b) < 0, and the
intermediate value property then implies that there is a point between
a and b at which h(x) = 0.
128 PERISIC
Solution 8.6: [Note that for some of these limits, we do not need to use
as heavy a piece of machinery as lHopitals rule, just some clever simplify-
ing.]
(1) since this limit has the indeterminate form 00 (since both limx2 (1
cos(x)) = 0 and limx2 sin2 (x) = 0), we may use lHopitals rule:
1 cos(x) sin(x) 1 1
lim 2 = lim = lim = .
x2 sin (x) x2 2 sin(x) cos(x) x2 2 cos(x) 2
(Note that we may also evaluate this limit without lHopitals rule,
using the trigonometric identity sin2 () + cos2 () = 1, as follows:
1 cos(x) 1 cos(x) 1 1
lim 2 = lim 2
= lim = .)
x2 sin (x) x2 1 cos (x) x2 1 + cos(x) 2
(2) again, here we have the choice of factoring or using lHopitals rule. I
feel like factoring:
x7 + 1 (x + 1)(x6 x5 + x4 x3 + x2 x + 1)
lim = lim
x1 x3 + 1 x1 (x + 1)(x2 x + 1)
x6 x5 + x4 x3 + x2 x + 1 7
= lim = .
x1 x2 x + 1 3
(4) as this has the indeterminate form 00 , and since there seems to be no
easy simplification possible, we use lHopitals rule:
1 x + ln(x) 1 + x1
lim = lim .
x1 1 + cos(x) x1 sin(x)
Since this limit still has the indeterminate form 00 , we may use lHopitals
rule again:
1 + x1 x12 1
lim = lim = 2.
x1 sin(x) x1 2 cos(x)
The exponent has the indeterminate form , and so we may use
lHopitals rule:
1 1
ln(ln(x)) ln(x) x
lim = lim = 0.
x x x 1
Hence, we see that
lim (ln(x))1/x = elimx ln(ln(x))/x = e0 = 1.
x
(7) as this limit has the indeterminate form 00 , we may use lHopitals rule:
x + sin(2x) 1 + 2 cos(2x) 1+2
lim = lim = = 3.
x0 x sin(2x) x0 1 2 cos(2x) 12
(8) since this limit has the indeterminate form , we may apply lHopitals
rule:
ex 1 ex ex
lim 2
= lim = lim = .
x0 x x0 2x x0 2
(The second equality follows from applying lHopitals rule a second
time, which is valid since the limit still has the indeterminate form .)
(9) in this limit, though we need to check at each stage, we will apply
lHopitals rule four times, as the original limit has the indeterminate
form 00 , and each of the first three applications of lHopitals rule results
in a limit still in the indeterminate form 00 .
ex + ex x2 2 ex ex 2x ex ex 2x
lim = lim = lim
x0 sin2 (x) x2 x0 2 sin(x) cos(x) 2x x0 sin(2x) 2x
ex + ex 2
= lim
x0 2 cos(2x) 2
ex ex
= lim
x0 4 sin(2x)
ex + ex 1
= lim = .
x0 8 cos(2x) 4
(10) this limit has the indeterminate form , and so we apply lHopitals
rule:
1
ln(x)
lim = lim x = 0.
x x x 1
(11) here, we first attempt to evaluate the limit by factoring, a sensible first
step for limits of rational functions:
x3 x2 x 2 (x 2)(x2 + x + 1) x2 + x + 1 7
lim 3 2
= lim 2
= lim 2
= .
x2 x 3x + 3x 2 x2 (x 2)(x x + 1) x2 x x + 1 3
130 PERISIC
Solution 9.26:
(1) this is an improper integral because 1/x3/2 is continuous on (0, 4] and
limx0+ 1/x3/2 = . So, we evaluate:
Z 4 Z 4
1 1
3/2
dx = lim 3/2
dx
0 x c0+ c x
Z 4
= lim x3/2 dx
c0+ c
2 2
= lim +
c0+ 4 c
1
= 1 + 2 lim = ,
c0+ c
and so this improper integral diverges.
(2) this is an improper integral because the interval of integration is [1, ),
which is not a closed interval. So, we evaluate:
Z Z M
1 1
dx = lim dx
1 x+1 M 1 x+1
1
= lim ln(M + 1) ln = ,
M 2
and so this improper integral diverges.
(3) this is an improper integral, as the interval of integration is [5, ),
which is not a closed interval. So, we evaluate:
Z Z M
1 1
3/2
dx = lim dx
5 (x 1) M 5 (x 1)3/2
Z M
= lim (x 1)3/2 dx
M 5
2
= lim + 1 = 1,
M M 1
and so this improper integral converges to 1.
(4) this is an improper integral because 1/(9 x)3/2 is continuous on [0, 9)
and limx9 1/(9 x)3/2 = . So, we evaluate:
Z 9 Z c
1 1
3/2
dx = lim dx
0 (9 x) c9 (9 x)3/2
Z0 c
= lim (9 x)3/2 dx
c9 0
2 2
= lim + = ,
c9 3 9c
and so this improper integral diverges.
132 PERISIC
Since one of these two improper integrals diverges, weR dont need to
0
evaluate the other one, as the original improper integral xdx/(x2 +
4) necessarily diverges.
and so this
R improper integral diverges, and so the original improper
integral 1 dx/x ln(x) necessarily diverges.
R
Solution 9.27: We first need to write
(1 + x)dx/(1 + x2 ) as the sum of
two improper integrals, for instance
Z Z 0 Z
1+x 1+x 1+x
dx = dx + dx,
1 + x2 1 + x2 0 1 + x2
134 PERISIC
and then evaluate the two resulting improper integrals separately. So,
Z Z M
1+x 1+x
2
dx = lim dx
0 1 + x M 0 1 + x2
"Z #
M Z M
1 x
= lim dx + dx
M 0 1 + x2 0 1 + x2
1 2 1
= lim (arctan(M ) arctan(0)) + ln(1 + M ) = ,
M 2 2
R
since limM ln(1 + M 2 ) = , and so the original improper integral (1 +
x)dx/(1 + x2 ) diverges.
Rt
However, when we evaluate limt t (1 + x)dx/(1 + x2 ), we get
Z t Z t Z t
1+x 1 x
lim dx = lim dx + dx
t t 1 + x2 t t 1 + x
2
t 1 + x
2
= lim 2 arctan(t) = 2 = ,
t 2
Solution 10.6:
(1) Before hitting the ground the first time, the ball travels distance a.
Between hitting the ground the first and second times, the ball travels
distance 2ra (distance ra up from the ground, and then distance ra
back to down to earth again). Between hitting the ground the second
and third times, the ball travels distance 2r2 a (distance r2 a up from the
ground, and then distance r2 a back to down to earth again). Between
hitting the ground the nth and the (n + 1)st times, the ball travels
distance 2rn a (distance rn a up from the ground, and then distance rn a
back to down to earth again). Hence, the total distance traveled is
X
X
a + 2ra + 2r2 a + . . . = a+ 2rn a = a + 2ra rn1
n=1 n=1
X 2ra a + ra
= a + 2ra rk = a + = .
1r 1r
k=0
(2) One way to do this problem is to actually write out the appropriate
geometric series and summing it. The easier way is to note that the
cars will crash exactly one hour after the fly leaves the front of Jacks
car, and in that hour (given the assumption that the fly loses no time
in changing direction) the fly flies exactly 257 miles.
Solution 10.8: Note that for s < 1, we have that ns < n (even for s = 0
or s negative), and hence that n1s > n1 . Hence, if we let Sk be the k th partial
P
sum of the harmonic series n=1 n1 , and Tk be the k th partial sum of the series
P 1 1
n=1 ns under consideration, then Tk > Sk . Since ns > 0 for all n, we have
that Tk is an unbounded monotonically increasing sequence, unbounded since
Sk is unbounded byP the argument given in Example 10.7, and so Tk diverges.
So, by definition, n=1 n1s diverges.
Solution 10.10:
P
(1) we argue
P by contradiction: suppose that n=0 (an + bn ) converges.
Since n=0 an converges by assumption, the arithmetic of series, The-
orem 10.9, yields
P that their difference
P also converges. However, their
difference is n=0 (an +bn an ) = n=0 bn , which diverges by assump-
tion, yielding the desired contradiction.
(2) again we argue by contradiction: suppose that the series of multiples
P Pk
n=0 can converges. Then, the sequence P Tk = n=0 can of partial
k Pk
sums converges. Note though that Tk = n=0 Pcan = c n=0 an =
cSk , where Sk is the k th partial sum of the series n=0 an . Since Tk
converges, the sequence 1c Tk = Sk also converges, by the arithmetic of
sequences (since the constant sequence 1c converges), and so the original
series converges, a contradiction.
136 PERISIC
Solution 10.12:
(1) by what has just been done, all we need are two convergent series.
For instance,
Ptake an =P (0.5)n and bn = (0.3)n for all n 0. Then,
P
n=0 an , n=0 bn , and n=0 an bn are all convergent geometric series.
P
(2) takeP an = 1 for all n 0 and bn = 1 for all n 0. Then, both n=0 an
P
and n=0 bn are both divergent geometric series, as is n=0 an bn (since
an bn = 1 for all n 0).
1
P
Pone, lets take an = bn = n for all n 1. Then, both n=1 an
(3) for this
and n=1 bn are the harmonic
P series, P
and hence divergent. However,
1
the series of products a
n=1 n nb = n=1 n2 is convergent, by the
discussion in Example 10.7.
P
(4) take any convergent series, for example n=0 P (0.5)n , andPset an = bn =
an
(0, 5)n . Then, the series of quotients is n=0 bn = n=0 1, which
diverges.
(5) here, we can take an = n12 and bn = n14 for n 1. Then, both
P P 1 P P 1
n=1 an = n=1 n2 and n=1 bn = n=1 n4 converge by Exercise
10.7, as does the series of quotients, as abnn = n12 .
P P n
(6) lets
P use geometric
P nseries again: both of n=0 an = n=0 6 and
n=0 bn =
P n=0 2 are
P divergent geometric series, and the series of
quotients n=0 abnn = n=0 3n is also a divergent geometric series.
P P P P 2
(7) n=1 an = n=1 1 and n=1 bn =P n=1 n both P diverge, but the
an
corresponding sequence of quotients n=1 bn = n=1 n12 converges.
Pk P
Solution 10.19: Let Sk = n=1 an be the k th partial sum of n=1 an .
(1) Since limn cn = 0 and cn > 0 for all n, there exists M > 0 so that
0 < cn < 1 for n > M . Let M = max(1, c1 , c2 , . . . , cM ), and note that
th
n M for all n 1. In particular, the k
cP partial sum of the series
n=1 an cn satisfies
k
X k
X
an cn an M = M Sk .
n=1 n=1
P
Hence, the sequence of partial sums of the series n=1 an cn forms
a monotonic
P (since all the a n and cn are positive), bounded (above
by
P M a
n=1 n , below by 0) sequence, and so converges. That is,
a c
n=1 n n converges.
(2) The proof in the case that limn cn = c 6= 0 is very similar to the
proof in the case that limn cn = 0. Since limn cn = c 6= 0, there
exists M > 0 so that cn < c + 1 for n > M . Let M = max(c +
1, c1 , c2 , . . . , cM ), and note that cn M for all n 1. In particular,
MATH2039 ANALYSIS 137
P
the k th partial sum of the series n=1 an cn satisfies
k
X k
X
an cn an M = M Sk .
n=1 n=1
P
Hence, the sequence of partial sums of the series n=1 an cn forms
a monotonic
P (since all the a n and c n are positive), bounded (above
by
P M n=1 an , below by 0) sequence, and so converges. That is,
n=1 an cn converges.
Solution 10.26: we make implicit use of the fact that convergence and absolute
convergence are the same for series with positive terms.
(1) converges absolutely: we could apply the ratio test, but we do not
need to use such heavy machinery. Instead, we note that
n
X 2n1 1 X 2n 1X 2 1 1 3
n
= n
= = = ,
n=0
3 2 n=0 3 2 n=0 3 2 (1 2/3) 2
P 2n
since n=0 3n is a convergent geometric series.
(2) diverges: this is a geometric series, and since 1.01 > 1, it is a divergent
geometric series.
(3) converges absolutely: this is a convergent geometric series, since
e
10 < 1, and it converges to
X e n X e n 1 10 10 e e
= 1= 1= = .
n=1
10 n=0
10 1 e/10 10 e 10 e 10 e
(4) converges absolutely: we use the comparison test: since n2 + Pn+ 1 >
1
n2 for all n 1, we have that n2 +n+1 < n12 for all n 1. Since n=1 n12
P 1
converges, we have that n=1 n2 +n+1 converges.
(5) diverges: note that for n 1, we have that n n, and so n P + n
2n. Therefore, n+ n 2n for n 1. Since the harmonic series n=1 n1
1 1
P 1
diverges, its multiple
P 2n diverges, and hence by the comparison
n=1
test the series n=1 1/(n + n) diverges.
(6) converges absolutely: since 1 + 3n > 3n for all n 1, we have that
1 n
1 1
P 1 P
1+3n < 3n for all n 1. Since n=1
P 3n = n=1 3 converges, the
second convergence test yields that n=1 1/(1 + 3n ) converges.
(7) diverges: well use the limit comparison test: for large values of n, it
2
seems that n10n 1
3 1 behaves like a constant multiple of n , and in fact
(8) converges absolutely: again well use the limit comparison test: for
large values of n, it seems that 1/ 37n3 + 3 behaves like 1/n3/2 , and
in fact
1/ 37n3 + 3 n3/2 1 1
lim = lim = lim p = = L.
n 1/n3/2 n 3
37n + 3 n 37 + 3/n 3 37
Since the limit exists and 0 < L = 137 < , and since n=1 n3/21
P
con-
P
3
verges, the limit comparison test yields that n=1 1/ 37n + 3 con-
verges.
(9) converges absolutely: we start this one with a bit of algebra, namely
n n 1
< 2 = 3/2 .
n2 + n n n
P
From Example 10.7,P we know that n=1 1/n3/2 converges, and so by
2
the comparison test, n=1 n/(n + n) converges.
1
(10) diverges: since ln(n) < n for all n 2, we have that ln(n) > n1
P
for all n 2, and so n=2 2/ ln(n)P diverges by the comparison test,
comparing it to the harmonic series n=1 n1 .
(11) converges absolutely: since 0 < sin2 (n) 1 for all n 1, we have
that
sin2 (n) 1 1
0< 2 2 < 2
n +1 n +1 n
n 1. Since we are dealing with a series with positive
for allP P terms and
since n=1 n12 converges by Example 10.7, we have that n=1 sin2 (n)/(n2 +
1) converges by the comparison test.
(12) converges absolutely: for this series, we start with a bit of algebraic
massage:
n
n + 2n n + 2n 2n + 2n 2
< < =2 .
n + 3n 3n 3n 3
So,
Pthe comparison
n test, comparing with the convergent geometric series
P
2 n=0 32 yields that n=1 (n + 2n )/(n + 3n ) converges.
(13) converges absolutely: since 1/(n2 ln(n)) < 1/n2 for n 3, since
P 1 2for n 3, we have by the second comparison test that
ln(n)
n=2 1/(n ln(n)) converges.
(14) diverges: for large values of n, it seems that the nth in the series is
approximately n1 , and so we might guess that the series diverges by the
limit comparison test. To check this guess, we need to evaluate
(n3 + 1)/(n4 + 2) n4 + n
lim = lim 4 = 1 = L.
n 1/n n n + 2
Plimit exists and since 0 < L =P1< , and since the harmonic
Since the
series n=1 n1 diverges, we have that n=1 (n3 + 1)/(n4 + 2) diverges
by the limit comparison test.
MATH2039 ANALYSIS 139
P 1
(22) converges absolutely: since 1/(2n + 3n ) < 1/2n and
Psince nn=0 2nn
converges, the second comparison test yields that n=1 1/(2 + 3 )
converges.
(23) converges
absolutely: since 1 + n 2 for n 1, we have that
n1+ nP n2 for n 1, and so 1/n(1+ n) 1/n2 for n 1. Hence,
since n=1 n12 converges by Example 10.7, we have by the comparison
P
test that n=1 1/n(1+ n) converges.
(24) converges absolutely: since 2n (n + 1) > P 2n for n 1, we have
that 1/(2 (n + 1)) < 1/2 for n 1. Since n=1 21nP
n n
is a convergent
geometric series, we have by the comparison test that n=1 1/(2n (n +
1)) converges.
(25) diverges: since factorials are involved, we first see whether the ratio
test gives us any information, and so we evaluate
(n + 1)!/((n + 1)2 en+1 ) (n + 1)!n2 en n2 n+1
lim = lim = lim = ,
n n!/(n2 en ) n n!(n + 1)2 en+1 n (n + 1)2 e
P
and since > 1, the ratio test implies that n=1 n!/(n2 en ) diverges.
[Though its not obvious how, we could also have applied the nth term
test for divergence, since for large values of n we have
n! (n 1)(n 2)! n1n2 2 1 n12 1 1
= = 2 > > 3.
n2 en nen n e ee n e e2 e
We simplified by noting that the middle terms n2 3
e , . . . , e are all greater
than 1 and that n1 1 n!
n > 2 for n large. Hence, limn n2 en 6= 0.]
(27) converges absolutely: since there are factorials involved, we first try
the ratio test:
(2(n + 1))!/((n + 1)!)3 (2n + 2)(2n + 1)
lim 3
= lim = 0 < 1,
n (2n)!/(n!) n (n + 1)3
P
and so the ratio test yields that n=2 (2n)!/(n!)3 converges.
(28) converges absolutely: note that the numerator of each term is either
0 or 2, and so this is a series with non-negative
P terms. Also, (1
(1)n )/n4 < 2/n4 for all n 1 and n=1 n14 converges by Example
P
10.7, and so by the comparison test n=1 (1 (1)n )/n4 converges.
MATH2039 ANALYSIS 141
(32) converges absolutely: since both the numerator and the denominator
are raised to (essentially) the same power, we try the root test, and so
142 PERISIC
1/n
2n+1
need to calculate: limn nn = limn 21/n n2 = L = 0 (since
limn 21/n = 20 =P1). Since the limit exists and since L < 1, the
root test yields that n=1 2n+1 /nn converges.
(33) converges conditionally: P we first test for
absolute convergence,
by
P
considering the related series n=1 |(1)n1 / n| = n=1 1/ n, which
diverges by Example 10.7.
Hence, this series converges but does not converge absolutely. That is,
the series converges conditionally.
(34) converges conditionally: we first check Pfor absolute convergence,
that is,Pconvergence of the associated series n=1 | cos(n)/((n+1) ln(n+
1))| = n=1 1/((n + 1) ln(n + 1)). For this series, we apply the integral
test, with f (x) = 1/((x + 1) ln(x + 1)). Since
1
ln(x + 1) + (x + 1) x+1 (ln(x + 1) + 1)
f 0 (x) = 2 2
= <0
(x + 1) (ln(x + 1)) (x + 1)2 (ln(x + 1))2
for x 1, the integral test yields that the series converges if and only
R RM
if 1 f (x)dx = limM 1 f (x)dx exists, so we calculate:
Z M
1
lim ln(ln(x + 1)) M
lim dx = 1
M 1 (x + 1) ln(x + 1) M
which diverges (very very slowly). So, the series does not converge
absolutely.
Hence, this series converges but does not converge absolutely. That is,
the series converges conditionally.
MATH2039 ANALYSIS 143
Since the limit exists and 0 < L < , the series being considered here
diverges, since the harmonic series diverges. So, the original series does
not converge absolutely.
P
We now test for convergence. The series n=1 (1)n (20n2 n1)/(n3 +
2
n1
n2 + 33) = n=1 (1)n an is an alternating series, since 20n
P
n3 +n2 +33 > 0
for n 1, and so lets check whether it satisfies the conditions of the
alternating series test. Since (20n2 n 1)/(n3 + n2 + 33) is a rational
function and the denominator has higher degree than the numerator,
we have that limn (20n2 n1)/(n3 +n2 +33) = 0. All that remains
to check is whether the an are monotonically decreasing. For this, let
f (x) = (20x2 x 1)/(x3 + x2 + 33), so that f (n) = an , and check
that its decreasing, which involves calculating f 0 (x):
20x4 + 2x3 + 4x2 + 1322x 33
f 0 (x) = <0
(x3 + x2 + 33)2
for all x greater than any of the roots of the numerator. So, the alter-
nating series test applies, and yields that this series converges.
Hence, this series converges but does not converge absolutely. That is,
the series converges conditionally.
(38) diverges: note that, for n 101, we have
n! n(n 1) 1 n n1 101 100 99 1 99 1
n
= n
= > ,
100 100 100 100 100 100 100 100 100 100
and so limn n!/(100)n does not exist. Hence, by the nth term test
for divergence, the series diverges.
(39) converges absolutely: we apply the integral test, with the function
1
f (x) = x ln(x)(ln(ln(x))) 2 . First, we check to see that f (x) is decreasing,
144 PERISIC
Since the limit exists and since 0 = L < 1, the ratio test yields that the
series converges.
MATH2039 ANALYSIS 145
(44) converges absolutely: and yet again, since there are factorials in-
volved, our first attempt should be with the ratio test:
p 100
(n + 1)100 2(n+1) / (n + 1)!
n+1 2
lim = lim = 0 = L.
n 100 n
n 2 / n! n n n+1
Since the limit exists and since L < 1, the ratio test yields that this
series converges.
(45) diverges: since there are factorials involved, we first try the ratio test:
(1 + (n + 1)!)/(1 + (n + 1))! 1 + (n + 1)!
lim = lim
n (1 + n!)/(1 + n)! (1 + n!)(n + 2)
n
1/n! + n + 1
= lim =1
n (1/n! + 1)(n + 2)
and so the ratio test gives no information. (This discussion was put in
to remind you that the ratio test doesnt always work with factorials.)
So, lets explicitly try the nth term test for divergence. We start with
a bit of algebraic massage, namely:
22n (n!)2 = (2n n!)2 = ((2n) (2n 2) (2n 4) 4 2)2 ,
and so
22n (n!)2 (2n) (2n) (2n 2) (2n 2) 2 2
=
(2n)! (2n) (2n 1) (2n 2) (2n 3) 2 1
(2n) (2n 2) 2
= > 1.
(2n 1) (2n 3) 1
2n 2
In particular, the limit limn
P2 (n!) /(2n)! cannot be zero, and so
the n term test yields that n=1 2 (n!)2 /(2n)! diverges.
th 2n
P n 2
That is, the original series n=1 (1) /(n + ln(n)) converges abso-
lutely.
(48) converges absolutely: we begin with a bit of algebraic massage, not-
ing that
n
X (1)2n X ((1)2 )n X 1 X 1
n
= n
= n
= .
n=1
2 n=1
2 n=1
2 n=1
2
This is a convergent geometric series, converging to
1
1 = 1.
1 12
(The subtraction of 1 arises from the fact that the starting index in this
series is not 0, so that
0 X
X 1 X 1 1 1
n
= n
= n
1 = 2 1 = 1.)
n=1
2 n=0
2 2 n=0
2
Since
P this limit exists and satisfies L < 1, the ratio test Pyields that
n n
n=1 2 /n! converges, and hence that the original series n=1 (2) /n!
converges absolutely.
(50) diverges: first, note that this is not an alternating series, but is a
series with all non-positive terms. Hence, for this series, convergence
and absolute convergence are equivalent, as they are for series with
non-negative terms.
since this limit exists and satisfies 0 < L = 50 < , and since
the
P harmonic series diverges, the limit comparison test yields that
n=1 100/(2n + 3) diverges.
Hence, this series converges but does not converge absolutely. That is,
the series converges conditionally.
Hence, this series converges but does not converge absolutely. That is,
the series converges conditionally.
Now, notice that (3/2)n > n for n 3 (since (3/2)3 > 3 and the
derivative of (3/2)n n is positive for n 3), and so
n
3n 3 1 1
> >
n(2n + 1) 2 2n 2
for n 3. (So, not exactly large for large values of n, but big enough
to do the trick.) Hence, the limit limn (1)n 3n /(n(2n + 1)) does not
exist (as it oscillates positive and negative and
P never settles down to 0),
and so by the nth term test for divergence, n=1 (1)n 3n /(n(2n + 1))
diverges.
Hence, this series converges but does not converge absolutely. That is,
the series converges conditionally.
We first
P test for absolute convergence,
P namely the convergence of the
series k=0 |(1)k /(2k+1)| = k=0 1/(2k+1).
P However, since 1/(2k+
1) > 1/(2k + 2) =P1/2(k + 1) and since k=0 1/(k + 1) is the harmonic
series, the series k=0 1/(2k + 1) diverges by the first comparison test,
and hence the original series does not converge absolutely.
Hence, this series converges but does not converge absolutely. That is,
the series converges conditionally.
(68) diverges: for this series, we first note that ln(x) < x1/8 for x large
(x > e32 works), as follows: consider the function f (x) = x1/8 ln(x),
and note that
f (e8k ) = (e8k )1/8 ln(e8k ) = ek 8k,
and so f (e32 ) = e4 32 = 22.5982... > 0.
for x > 13, since both ln(x) > 0 and ln(ln(x)) > 0 for x > 13 and since
p > 0 by assumption.
For p 6= 1, we get:
Z M
1 1 1 M
limM p
= lim p1
13
13 x ln(x)(ln(ln(x)) M p + 1 ln(ln(x))
1
lim ln(ln(M ))p+1 ln(ln(13))p+1 ,
=
p + 1 M
which converges for p > 1 (since p + 1P< 0)and diverges for p < 1
p
(since p + 1 > 0). Hence, the series n=13 1/(n ln(n)(ln(ln(n))) )
converges if and only if p > 1. (Note that this is really just Example
10.7 in a bit of disguise.)
th
Solution 10.27: PBy the contrapositive to the n term test for divergence,
since the series n=1 an converges, we have that limn an = 0. In particular,
taking = 1 and remembering that each an > 0, there exists M so that
0 < an < 1 for all n > M . Since 0 < an < 1 for n > M and since s 1,
s
Pthat ans < an for n > M , and so byPthe
we have comparison test,
Pwe have
that n=M +1 an converges by comparison to n=M +1 an . Since n=M +1 asn
P
converges, we see that n=0 asn converges, as desired.
MATH2039 ANALYSIS 153
Solution 12.5:
(1) radius of convergence is , interval of convergence is R: Apply
the ratio test and calculate:
(1)n+1 xn+1 /(n + 1)!
= |x| lim 1
lim
n n
= 0.
n (1) x /n! n n+1
Hence, this series converges absolutely for all values of x (since this
limit is 0 for every value of x).
(2) radius of convergence is 51 , interval of convergence is 15 , 15 :
Hence, this series converges absolutely for 5|x| < 1, that is for |x| < 15 ,
and so the radius of convergence is 15 . We now need to check the
endpoints of the interval ( 15 , 15 ):
P P
At x = 51 , the series becomes n=1 5n (1/5)n /n2 = n=1 (1)
n
/n2 ,
which converges absolutely.
P P
At x = 51 , the series becomes n=1 5n (1/5)n /n2 = n=1 1/n2 , which
converges absolutely.
Hence, this series converges absolutely for |x| < 1, and so the radius of
convergence is 1. We now need to check the endpoints of the interval
(1, 1):
P n
At x = 1, the series becomes n=1 (1) /(n(n + 1)), which converges
absolutely.
P
At x = 1, the series becomes n=1 1/(n(n + 1)), which converges ab-
solutely.
So, the series converges absolutely for all x in the closed interval [1, 1],
and diverges elsewhere.
154 PERISIC
So, the series converges absolutely for all x in the open interval (1, 1),
converges conditionally at x = 1, and diverges elsewhere.
(5) radius of convergence is , interval of convergence is R: Apply
the ratio test and calculate:
(1)n+1 x2(n+1)+1 /(2(n + 1) + 1)!
lim = |x|2 lim (2n + 1)!
n (1)n x2n+1 /(2n + 1)! n (2n + 3)!
1
= |x|2 lim = 0.
n (2n + 3)(2n + 2)
Hence, this series converges absolutely for all values of x (since this
limit is 0 for every value of x).
(6) radius of convergence is , interval of convergence is R: Apply
the ratio test and calculate:
n+1 n+1
3 x /(n + 1)! 3
lim = |x| n
lim = 0.
n 3n xn /n! n+1
Hence, this series converges absolutely for all values of x (since this
limit is 0 for every value of x).
(7) radius of convergence is 1, interval of convergence is [1, 1]:
Apply the ratio test and calculate:
n+1
/(1 + (n + 1)2 ) 1 + n2
x
lim
= |x| lim = |x|.
n xn /(1 + n2 ) n 2 + 2n + n2
Hence, this series converges absolutely for |x| < 1, and so the radius of
convergence is 1. We now need to check the endpoints of the interval
(1, 1):
P n
At x = 1, the series becomes n=0 (1) /(1 + n2 ), which converges
absolutely.
P
At x = 1, the series becomes n=0 1/(1 + n2 ), which converges abso-
lutely.
MATH2039 ANALYSIS 155
So, the series converges absolutely for all x in the closed interval [1, 1],
and diverges elsewhere.
(8) radius of convergence is 1, interval of convergence is (2, 0]:
Apply the ratio test and calculate:
(1)(n+1)+1 (x + 1)n+1 /(n + 1)
lim
= |x + 1| lim n = |x + 1|.
n (1) n+1 n
(x + 1) /n n n + 1
Hence, this series converges absolutely for |x + 1| < 1, and so the radius
of convergence is 1. We now need to check the endpoints of the interval
(2, 0):
P P
At x = 2, the series becomes n=1 (1)n+1 (1)n /n = n=1 1/n,
which diverges, being a constant multiple of the harmonic series.
P
At x = 0, the series becomes n=1 (1)n+1 /n, which converges condi-
tionally, as it is the alternating harmonic series.
So, the series converges absolutely for all x in the open interval (2, 0),
converges conditionally at x = 0, and diverges elsewhere.
(9) radius of convergence is 43 , interval of convergence is ( 19 11
3 , 3 ):
Apply the ratio test and calculate:
n+1
(x + 5)n+1 /4n+1 3
3
lim
n 3n (x + 5)n /4n = 4 |x + 5|.
At x = 19
3 , the series becomes
n
X 3n 19
3 +5
X
n
= (1)n ,
n=0
4 n=0
At x = 11
3 , the series becomes
n
X 3n 11
3 +5
X
n
= 1,
n=0
4 n=0
So, the series converges absolutely for all x in the open interval ( 19 11
3 , 3 ),
and diverges elsewhere.
(10) radius of convergence is 1, interval of convergence is [2, 0]:
Apply the ratio test and calculate:
(1)n+1 (x + 1)2(n+1)+1 /((n + 1)2 + 4) n2 + 4
lim
= |x+1|2 lim = |x+1|2 .
n (1)n (x + 1)2n+1 /(n2 + 4) n n2 + 2n + 5
156 PERISIC
Hence, this series converges absolutely for |x + 1|2 < 1, that is for
|x + 1| < 1, and so the radius of convergence is 1. We now need to
check the endpoints of the interval (2, 0).
So, the series converges absolutely for all x in the closed interval [2, 0],
and diverges elsewhere.
(11) radius of convergence is , interval of convergence is R: Apply
the ratio test and calculate:
n+1
(x 1)2(n+1) /(2(n + 1) + 1)!
lim
n 2n
= |x 1|2 lim = 0.
n (x 1) /(2n + 1)! n (2n + 2)(2n + 3)
Hence, this series converges absolutely for all values of x (since this
limit is 0 for every value of x).
(12) radius of convergence is , interval of convergence is R: This
time, since the coefficients are nth powers, we apply the root test and
calculate:
xn 1/n
1
lim = |x| lim = 0.
n (ln(n))n n ln(n)
Hence, this series converges absolutely for all values of x (since this
limit is 0 for every value of x).
(13) radius of convergence is 13 , interval of convergence is ( 31 , 13 ):
Apply the ratio test and calculate:
n+1 n+1
3 x
lim = 3|x|.
n 3n xn
Hence, this series converges absolutely for 3|x| < 1, that is |x| < 13 , and
so the radius of convergence is 13 . We now need to check the endpoints
of the interval ( 13 , 13 ).
So, the series converges absolutely for all x in the open interval ( 31 , 13 ),
and diverges elsewhere.
(14) radius of convergence is 0, interval of convergence is {0}: Apply
the ratio test and calculate:
(n + 1)!xn+1 /2n+1
lim = |x| lim n + 1 = .
n n!xn /2n n 2
Hence, this series converges only for x = 0 and diverges elsewhere.
(15) radius of convergence is 21 , interval of convergence is ( 21 , 12 ]:
Apply the ratio test and calculate:
(2)n+1 x(n+1)+1 /((n + 1) + 1)
lim = |x| 2(n + 1) = 2|x|.
n (2)n xn+1 /(n + 1) n+2
Hence, this series converges absolutely for 2|x| < 1, that is |x| < 21 , and
so the radius of convergence is 12 . We now need to check the endpoints
of the interval ( 21 , 12 ).
So, the series converges absolutely for all x in the open interval ( 21 , 12 ),
converges conditionally at x = 21 , and diverges elsewhere.
(16) radius of convergence is , interval of convergence is R: Apply
the ratio test and calculate:
(1)n+1 x2(n+1) /(2(n + 1))!
= |x|2 lim 1
lim = 0.
n (1)n x2n /(2n)! n (2n + 2)(2n + 1)
Hence, this series converges absolutely for all values of x (since this
limit is 0 for every value of x).
(17) radius of convergence is 1, interval of convergence is [1, 1]:
Apply the ratio test and calculate:
(1)n+1 x3(n+1) /(n + 1)3/2 n3/2
lim
= |x|3 lim = |x|3 .
n (1)n x3n /n3/2 n (n + 1)3/2
158 PERISIC
Hence, this series converges absolutely for |x|3 < 1, that is |x| < 1, and
so the radius of convergence is 1. We now need to check the endpoints
of the interval (1, 1).
P P
At x = 1, the series becomes n=1 (1)n (1)n /n3/2 = n=1 1/n3/2 ,
which converges, by Example 10.7.
P n
At x = 1, the series becomes n=1 (1) /n3/2 , which converges abso-
lutely, by Example 10.7.
So, the series converges absolutely for all x in the closed interval [1, 1],
and diverges elsewhere.
(18) radius of convergence is 1, interval of convergence is [1, 1]:
Apply the ratio test and calculate:
(1)(n+1)+1 xn+1 /((n + 1) ln2 (n + 1)) n ln2 (n)
lim
= |x| lim = |x|.
n (1)n+1 xn /(n ln2 (n)) n (n + 1) ln2 (n + 1)
Hence, this series converges absolutely for |x| < 1, and so the radius of
convergence is 1. We now need to check the endpoints of the interval
(1, 1).
Px = 1,n+1
At the series becomes P
n=2 (1) (1)n /(n ln2 (n)) = n=2 1/(n ln2 (n)), which converges
by the integral test: take f (x) = 1/(x ln2 (x)). Then,
(ln2 (x) + 2 ln(x))
f 0 (x) = <0
x2 ln4 (x)
for x 2, and so f (x) is decreasing. Then, we evaluate
Z Z M
1
f (x)dx = lim 2 dx
2 M 2 x ln (x)
1 M
= lim 2
M ln(x)
1 1 1
= lim + = ,
M ln(M ) ln(2) ln(2)
which converges. Hence, by the integral test, the series converges.
P
At x = 1, the series becomes n=2 (1)n+1 /(n ln2 (n)), which converges
absolutely by the argument just given.
So, the series converges absolutely for all x in the closed interval [1, 1],
and diverges elsewhere.
(19) radius of convergence is 2, interval of convergence is (1, 5):
Apply the ratio test and calculate:
(x 3)n+1 /2n+1 1
lim = |x 3|.
n (x 3)n /2n 2
MATH2039 ANALYSIS 159
So, the series converges absolutely for all x in the open interval (1, 5),
and diverges elsewhere.
(20) radius of convergence is 1, interval of convergence is [3, 5]: Ap-
ply the ratio test and calculate:
(1)n+1 (x 4)n+1 /((n + 1) + 1)2 2
lim
= |x 4| lim (n + 1) = |x 4|.
n (1)n (x 4)n /(n + 1)2 n (n + 2)2
Hence, this series converges absolutely for |x 4| < 1, and so the radius
of convergence is 1. We now need to check the endpoints of the interval
(3, 5).
P P
At x = 3, the series becomes n=1 (1)n (1)n /(n+1)2 = n=1 1/(n+
1)2 , which converges by Example 10.7.
P
At x = 5, the series becomes n=1 (1)n /(n + 1)2 , which converges
absolutely, again by Example 10.7.
So, the series converges absolutely for all x in the closed interval [3, 5],
and diverges elsewhere.
(21) radius of convergence is 0, interval of convergence is {2}: Apply
the ratio test and calculate:
(2(n + 1) + 1)! (x 2)n+1 /(n + 1)3 (2n + 3)! n3
lim
= |x2| lim =
n (2n + 1)! (x 2)n /n3 n (2n + 1)! (n + 1)3
P
At x = 4, the series becomes n=1 ln(n)/n, which diverges by the
comparison
P test, since ln(n)/n > 1/n for n 3 and the harmonic series
n=1 1/n diverges.
So, the series converges absolutely for all x in the open interval (2, 4),
converges conditionally at x = 2, and diverges elsewhere.
(23) radius of convergence is 8, interval of convergence is ( 13 19
2 , 2 ):
Apply the ratio test and calculate:
(2x 3)n+1 /42(n+1)
1 1 3
lim = |2x 3| = x .
n (2x 3)n /42n 16 8 2
1
3
Hence, this series converges absolutely for 8 x 2 < 1, that is for
x 3 < 8, and so the radius of convergence is 8. We now need to
2
check the endpoints of the interval ( 13 19
2 , 2 ).
P P
At x = 13
2 , the series becomes
n
n=0 (2(13/2)3) /4
2n
= n=0 (1)
n
,
which diverges.
P P
At x = 192 , the series becomes n=0 (2(19/2) 3)n /42n = n=0 1,
which diverges.
So, the series converges absolutely for all x in the open interval ( 13 19
2 , 2 ),
and diverges elsewhere.
(24) radius of convergence is b, interval of convergence is (ab, a+b):
Apply the ratio test and calculate:
(x a)n+1 /bn+1 1
lim = |x a|.
n (x a)n /bn b
Hence, this series converges absolutely for 1b |x a| < 1, that is for
|x a| < b, and so the radius of convergence is b. We now need to check
the endpoints of the interval (a b, a + b).
P P
At x = a b, the series becomes n=2 (a b a)n /bn = n=2 (1)
n
,
which diverges.
P n n
P
At x = a + b, the series becomes n=2 (a + b a) /b = n=2 1, which
diverges.
So, the series converges absolutely for all x in the open interval (a
b, a + b), and diverges elsewhere. (Note that the previous series is a
specific example of this general phenomenon, with a = 23 and b = 8.)
(25) radius of convergence is , interval of convergence is R: Apply
the ratio test and calculate:
((n + 1) + p)!xn+1 /((n + 1)!((n + 1) + q)!)
limn
(n + p)!xn /(n!(n + q)!)
n+1+p
= |x| lim = 0.
n (n + 1)(n + 1 + q)
MATH2039 ANALYSIS 161
Hence, this series converges absolutely for all values of x (since this
limit is 0 for every value of x).
(26) radius of convergence is 3, interval of convergence is [3, 3):
Apply the ratio test and calculate:
(n+1)1
/((n + 1)3n+1 )
x n 1
lim
n1 n
= |x| lim = |x|.
n x /(n3 ) n 3(n + 1) 3
Hence, this series converges absolutely for 13 |x| < 1, that is for |x| <
3, and so the radius of convergence is 3. We now need to check the
endpoints of the interval (3, 3).
P P n1
At x = 3, the series becomes n=1 (3)n1 /(n3n ) = 31 n=1 (1)n ,
which converges conditionally, as it is a constant multiple of the alter-
nating harmonic series.
P P
At x = 3, the series becomes n=1 3n1 /(n3n ) = 13 n=1 n1 , which
diverges, as it is a constant multiple of the harmonic series.
So, the series converges absolutely for all x in the open interval (3, 3),
converges conditionally at x = 3, and diverges elsewhere.
(27) radius of convergence is , interval of convergence is R: Apply
the ratio test and calculate:
(1)(n+1)1 x2(n+1)1 /(2(n + 1) 1)!
= |x|2 lim 1
lim
n1 2n1
= 0.
n (1) x /(2n 1)! n 2n(2n + 1)
Hence, this series converges absolutely for all values of x (since this
limit is 0 for every value of x).
(28) radius of convergence is 0, interval of convergence is {a}: Apply
the ratio test and calculate:
(n + 1)!(x a)n+1
lim = |x a| lim (n + 1) =
n n!(x a)n n
P P
At x = 3, the series becomes n=1 n(31)n /(2n (3n1)) = n=1 n/(3n
1), which again diverges by the nth term test for divergence.
So, the series converges absolutely for all x in the open interval (1, 3),
and diverges elsewhere.
Solution 12.6: The condition that the an satisfy is similar to the condition
P
of the root test, and so we apply the root test to the power series n=0 an xn .
Namely, we calculate
1/n 1/n
lim |an xn | = |x| lim |an | = L|x|.
n n
Hence, the series converges absolutely for L|x| < 1, that is |x| < L1 , and di-
verges for L|x| > 1, and so the radius of convergence of this series is L1 , as
desired.
Solution 12.7: We can use the same techniques that we have developed for
power series for other series, that are not strictly speaking power series. For
instance, we can apply the ratio test to the series, for all the values of x for
which the terms are defined.
(1) first, we note that this series is not defined at x = 1, but is defined for
all other values of x. Applying the ratio test, we calculate:
((x + 2)/(x 1))n+1 /(2(n + 1) 1) |x + 2|
2n 1 |x + 2|
lim
n
= lim = .
n ((x + 2)/(x 1)) /(2n 1) |x 1| n 2n + 1 |x 1|
|x+2|
Hence, this series converges absolutely for |x1| < 1, that is for |x+2| <
|x+2|
|x 1|, which is the open ray (, 12 ), and diverges for |x1| > 1,
which is the union ( 12 , 1) (1, ).
P 1
converges for all , , by limit comparison to P the series n=1 n2 .
Hence, taking = x and = x 1, we have that n=1 1/((x + n)(x +
n 1)) converges at every value of x for which it is defined, namely the
union
(, 0) (0, 1) (1, 2) (2, 3) = R W.
164 PERISIC
Solution 13.2:
(1) we start by calculating the derivatives of f at a = 6:
f (0) (6) = f (6) = 455; f (1) (6) = f 0 (6) = 185; f (2) (6) = 48;
f (3) (6) = 6; f (n) (6) = 0 for n 4.
Hence, the Taylor series for f centred at a = 6 is
X 1 (n) 1 1
f (6)(x 6)n = 455 + 185(x 6) + 48(x 6)2 + 6(x 6)3 .
n=0
n! 2 6
The radius of convergence of this series is (using the root test, for
instance), and so the interval of convergence is R.
(2) we start by calculating that f (n) (x) = 3n e3x for n 0, and so f (n) (2) =
3n e6 . Hence, the Taylor series for f centred at a = 2 is
X 1 (n) X 3n
f (2)(x + 2)n = e6 (x + 2)n .
n=0
n! n=0
n!
The radius of convergence of this series is (using the ratio test, for
instance), and so the interval of convergence is R.
(3) we start here by recalling that
(n) cosh(x) for x even, and
f (x) =
sinh(x) for x odd.
So, we have that f (n) (1) = cosh(1) = 12 (e+ 1e ) for n even, and f (n) (1) =
sinh(1) = 12 (e 1e ) for n odd. Hence, the Taylor series for f centred at
a = 1 is
X 1 (n) X 1 (k) X 1 (k)
f (1)(x 1)n = f (1)(x 1)k + f (1)(x 1)k
n=0
n! (k)! (k)!
k even k odd
X 1 X 1
= f (2k) (1)(x 1)2k + f (2k+1) (1)(x 1)2k+1
(2k)! (2k + 1)!
k=0 k=0
e2 + 1 X 1 e2 1 X 1
= (x 1)2k + (x 1)2k+1 .
2e (2k)! 2e (2k + 1)!
k=0 k=0
The radius of convergence of this series is (using the ratio test, for
instance), and so the interval of convergence is R.
MATH2039 ANALYSIS 165
You MUST write the question number at the top of each page of the answer
booklet.
P (1)n
(ii) Determine whether the series n=1
n
converges absolutely,
converges conditionally or diverges.
[5 marks]
P
(iii) Show that if the series n=1 an converges absolutely then it con-
verges.
[5 marks]
(iv) Determine the radius and interval of convergence for the power
series
X n(x 1)n
.
n=1
2n (3n 1)
[9 marks]
END OF PAPER
1. (i) State the definition of the limit L of a sequence (an )nN , symbol-
ically written as limn an = L.
[2 marks]
(ii) State what it means for a sequence (an )nN to diverge to infinity,
symbolically written as limn an = .
[2 marks]
(iii) Show, if the sequence (an )nN diverges to infinity, then the se-
quence of reciprocals (bn )nN , bn = a1n , converges to 0.
[4 marks]
(iv) Determine whether the sequence (an )nN , an = n + 2 n,
converges or diverges. If it converges, determine the limit and
using the definition of convergence, show that your limit is correct.
[5 marks]
(v) State what it means for a sequence (an )nN to be a Cauchy se-
quence.
[2 marks]
(vi) Show that, if the sequence (an )nN converges, then (an )nN is a
Cauchy sequence.
[5 marks]
(vii) Show that the sequence (an )nN , an = cos( n
2 ), diverges.
[5 marks]
168 PERISIC
[6 marks]
4. (i) Let a > 0 be a real number. Show that the following improper
integrals converge and determine their limits:
R
(a) [4 marks] 0 eax dx;
R
(b) [7 marks] 0 xeax dx.
MATH2039 ANALYSIS 169
[8 marks]
END OF PAPER
Appendix C: Learning Outcomes
170
Appendix D: Further reading
There are many excellent analysis books around. Here is a selection of books
that you may find helpful:
If you like comics, you may want to read the worlds first calculus comic book:
171
Index
-function, 97 divergence(of a series), 66
k-point refinement, 56 domain, 6
sequence, 19
sequence (monotone or monotonic),
24
sequence of functions, 81
sequence of functions - pointwise
convergence, 81
sequence of functions - uniform
convergence, 81
sequence of partial sums, 66
sequentially compact set, 37
Vesna Perisic received her mathematics de-
grees from the University of Zagreb, Croa-
tia. After spending several years at the
University of Saarbrucken, Germany, in De-
cember 2000 she moved with her family to
UK. Vesna has held a teaching position at
the University of Southampton since 2001.